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
1202.0386	2	1202	2012-10-15T12:41:56	On rings each of whose finitely generated modules is a direct sum of cyclic modules	[ "math.RA", "math.AC" ]	In this paper we study (non-commutative) rings $R$ over which every finitely generated left module is a direct sum of cyclic modules (called left FGC-rings). The commutative case was a well-known problem studied and solved in 1970s by various authors. It is shown that a Noetherian local left FGC-ring is either an Artinian principal left ideal ring, or an Artinian principal right ideal ring, or a prime ring over which every two-sided ideal is principal as a left and a right ideal. In particular, it is shown that a Noetherian local duo-ring $R$ is a left FGC-ring if and only if $R$ is a right FGC-ring, if and only if, $R$ is a principal ideal ring. Moreover, we obtain that if $R=\Pi_{i=1}^n R_i$ is a finite product of Noetherian duo-rings $R_i$ where each $R_i$ is prime or local, then $R$ is a left FGC-ring if and only if $R$ is a principal ideal ring.each $R_i$ is prime or local, then $R$ is a left FGC-ring if and only if $R$ is a principal ideal ring.	math.RA	math	On rings each of whose finitely generated modules is a direct sum of cyclic modules ∗†‡ M. Behboodia,b §and G. Behboodi Eskandaria aDepartment of Mathematical Sciences, Isfahan University of Technology P.O.Box: 84156-83111, Isfahan, Iran bSchool of Mathematics, Institute for Research in Fundamental Sciences (IPM) P.O.Box: 19395-5746, Tehran, Iran [email protected], [email protected] Abstract In this paper we study (non-commutative) rings R over which every finitely generated left module is a direct sum of cyclic modules (called left FGC-rings). The commutative case was a well-known problem studied and solved in 1970s by various authors. It is shown that a Noetherian local left FGC-ring is either an Artinian principal left ideal ring, or an Artinian principal right ideal ring, or a prime ring over which every two-sided ideal is principal as a left and a right ideal. In particular, it is shown that a Noetherian local duo-ring R is a left FGC-ring if and only if R is a right FGC-ring, if and only if, R is a principal ideal ring. Moreover, we obtain that if R = Πn Ri is a finite i=1 product of Noetherian duo-rings Ri where each Ri is prime or local, then R is a left FGC-ring if and only if R is a principal ideal ring. 1. Introduction The question of which commutative rings have the property that every finitely gen- erated module is a direct sum of cyclic modules has been around for many years. We will call these rings FGC-rings. The problem originated in I. Kaplanskys papers ∗The research of the second author was in part supported by a grant from IPM (No. 91130413). †Key Words: Cyclic modules; FGC-rings; duo-rings; principal ideal rings. ‡ 2010 Mathematics Subject Classification.Primary 16D10, 16D70, 16P20, Secondary 16N60. §Corresponding author; 1 [13] and [14], in which it was shown that a local domain is FGC if and only if it is an almost maximal valuation ring. For several years, this is one of the ma jor open problems in the theory. R. S. Pierce [19] showed that the only commutative FGC- rings among the commutative (von Neumann) regular rings are the finite products of fields. A deep and difficult study was made by Brandal [3], Shores-R. Wiegand [22], S. Wiegand [24], Brandal-R. Wiegand [4] and V´amos [23], leading to a com- plete solution of the problem in the commutative case. To show that a commutative FGC-ring cannot have an infinite number of minimal prime ideals required the study of topological properties (so-called Zariski and patch topologies). For complete and more leisurely treatment of this sub ject, see Brandal [2]. It gives a clear and detailed exposition for the reader wanting to study the sub ject. The main result reads as follows: A commutative ring R is an FGC-ring exactly if it is a finite direct sum of commutative rings of the following kinds: (i) maximal valuation rings; (ii) almost maximal B´ezout domains; (iii) so-called torch rings (see [2] or [8] for more details on the torch rings). The corresponding problem in the non-commutative case is still open; see [21, Appendix B. Dniester Notebook: Unsolved Problems in the Theory of Rings and Modules. Pages 461-516] in which the following problem is considered. [21, Problem 2.45] (I. Kaplansky, reported by A. A. Tuganbaev): Describe the rings in which every one-sided ideal is two-sided and over which every finitely gen- erated module can be decomposed as a direct sum of cyclic modules. Through this paper, all rings have identity elements and all modules are unital. A left FGC-ring is a ring R such that each finitely generated left R-module is a direct sum of cyclic submodules. A right FGC-ring is defined similarly, by replacing the word left with right above. A ring R is called a FGC-ring if it is a both left and right FGC-ring. Also, a ring R is called duo-ring if each one-sided ideal of R is two-sided. Therefore, the Kaplansky problem is: Describe the FGC-duo-rings. In this paper, we study left FGC-rings and, among other results, we will present a partial solution to the above problem of Kaplansky. 2. On Left FGC-Rings A ring R is local in case R has a unique left maximal ideal. An Artinian (resp. Noetherian) ring is a ring which is both left and right Artinian (resp. Noetherian). A principal ideal ring is a ring which is both left and right principal ideal ring. Also, for a subset S of RM , we denote by l.AnnR (S ), the left annihilator of S in R. A left R-module M which has a composition series is called a module of finite length. The 2 length of a composition series of RM is said to be the length of RM and denoted by length(RM ). Note that in the sequel, if we have proved certain results for rings or modules “on the left,” then we shall use such results freely also “on the right,” provided that these results can indeed be proved by the same arguments applied “to the other side.” We begin with the following lemma which is an associative, non-commutative version of Brandal [2, Proposition 4.3] for local rings (R, M) with M2 = (0). Also, the proof is based on a slight modification of the proof of [1, Theorem 3.1]. Lemma 2.1. Let (R, M) be a local ring with M2 = (0) and RM = Ry1 ⊕ . . . ⊕ Ryt such that t ≥ 2 and each Ryi is a minimal left ideal of R. If there exist 0 6= x1 , x2 ∈ M such that x1R ∩ x2R = (0), then the left R-module (R ⊕ R)/R(x1 , x2) is not a direct sum of cyclic modules. Proof. Since RM = Ry1 ⊕ . . . ⊕ Ryt and each Ryi is a minimal left ideal of R, we conclude that R is of finite composition length and length(RR) = t + 1. We put RG = (R ⊕ R)/R(x1 , x2). Since x1 , x2 ∈ M and M2 = (0), we conclude that l.AnnR (R(x1 , x2)) = M. Thus R(x1 , x2 ) is simple and hence length(RG) = 2 × length(RR) − length(RR(x1 , x2 )) = 2(t + 1) − 1. We claim that every non-zero cyclic submodule Rz of G has length 1 or t + 1. If Mz = 0, then length(Rz) = 1 since Rz ≃ R/M. Suppose that Mz 6= 0, then there exist c1 , c2 ∈ R such that z = (c1 , c2) + R(x1 , x2). If c1 , c2 ∈ M, then Mz = 0, since M2 = 0. Thus without loss of generality, we can assume that z = (1, c2) + R(x1 , x2 ) (since if c1 6∈ M, then c1 is unit). Now let r ∈ l.AnnR (z), then r(1, c2) = t(x1 , x2 ) for some t ∈ R. It follows that r = tx1 and rc2 = tx2 . Thus tx2 = tx1 c2 . If t /∈ M, then t is unit and so x2 = x1 c2 that it is contradiction (since x1R ∩ x2R = (0)). Thus t ∈ M and so r = tx1 = 0. Therefore, l.AnnR (z) = 0 and so Rz ∼= R. It follows that length(Rz) = t + 1. Now suppose the assertion of the lemma is false. Then RG is a direct sum of cyclic modules and since RG is of finite length, we have G = Rw1 ⊕ . . . ⊕ Rwk ⊕ Rv1 ⊕ . . . ⊕ Rvl , where l, k ≥ 0, and each Rwi is of length t + 1 and each Rvj is of length 1. Clearly M ⊕ M is not a simple left R-module. Since R(x1 , x2) is simple, MG = (M ⊕ M)/R(x1 , x2) 6= 0. It follows that k ≥ 1. Also, length(RG) = 2(t+1)−1 = k(t+1)+l and this implies that k = 1 and l = t. Since Mvi = 0 for each i, MG = Mw1 and hence 3 G/MG ≃ Rw1/Mw1 ⊕ Rv1 ⊕ . . . ⊕ Rvt . It follows that length(RG/MG) = 1 + t. On the other hand, we have G/MG ∼= R/M ⊕ R/M and so length(RG/MG) = 2 and so t = 1, a contradiction. Thus the left R-module (R ⊕ R)/R(x1 , x2) is not a direct sum of cyclic modules. (cid:3) We recall that the socle soc(RM ) of a left module M over a ring R is defined to be the sum of all simple submodules of M . Theorem 2.4. Let (R, M) be a local ring such that RM and MR are finitely generated. If every left R-module with two generators is a direct sum of cyclic modules, then either M is a principal left ideal or M is a principal right ideal. Proof. We can assume that M is not a principal left ideal of R. One can easily see that MR is generated by {x1 , · · · , xn} if and only if M/M2 is generated by the set {x1 + M2 , · · · , xn + M2} as a right ideal of R/M2 . Thus it suffices to show that M/M2 is a principal right ideal of R/M2 . Since every left R-module with two generators is a direct sum of cyclic modules, we conclude that every left R/M2 -module with two generators is also a direct sum of cyclic modules. Therefore, without loss of generality we can assume that M2 = (0). It follows that soc(RR) = soc(RR ) = M. Since RM is finitely generated, RM = Ry1 ⊕ . . . ⊕ Ryt such that t ≥ 2 and each Ryi is a minimal left ideal of R. We claim that MR = xR, for if not, then we can assume that MR = ⊕i∈I xiR where I ≥ 2 and each xiR is a minimal right ideal of R. We can assume that {1, 2} ⊆ I and so 0 6= x1 , x2 ∈ M and x1R ∩ x2R = (0). Now by Lemma 2.1, the left R-module (R ⊕ R)/R(x1 , x2 ) is not a direct sum of cyclic modules, a contradiction. Thus M is principal as a right ideal of R. (cid:3) A ring whose lattice of left ideals is linearly ordered under inclusion, is called a left uniserial ring. A uniserial ring is a ring which is both left and right uniserial. Note that left and right uniserial rings are in particular local rings and commutative uniserial rings are also known as valuation rings. Next, we need the following lemma from [18]. Lemma 2.5. (See Nicholson and S´anchez-Campos [18, Theorem 9]) For any ring R, the fol lowing statements are equivalent: (1) R is local, J (R) = Rx for some x ∈ R and xk = 0 for some k ∈ N. (2) There exist x ∈ R and k ∈ N such that xk−1 6= 0 and R ⊃ Rx ⊃ . . . ⊃ Rxk = (0) are the only left ideals of R. 4 (3) R is left uniserial of finite composition length. Theorem 2.6. Let (R, M) be a local ring such that RM and MR are finitely generated and Mk = (0) for some k ∈ N. If every left R-module with two generators is a direct sum of cyclic modules, then either R is a left Artinian principal left ideal ring or R is a right Artinian principal right ideal ring. Proof. Assume that every left R-module with two generators is a direct sum of cyclic modules. Then by Theorem 2.4, either M is a principal left ideal or M is a principal right ideal. If M is a principal left ideal, then by Lemma 2.5, R is a left Artinian principal left ideal ring. Thus we can assume that M is a principal right ideal. Then by using Lemma 2.5 to the right side, R is a right Artinian principal right ideal ring. (cid:3) Next, we need the following lemma from Mohamed H. Fahmy-Susan Fahmy[9]. We note that their definition of a local ring is slightly different than ours; they defined a local ring (resp. scalar local ring) as a ring R such that it contains a unique maximal ideal M and R/M is an Artinian ring (resp. division ring). Thus our definition of a local ring and the scalar local ring coincide. Lemma 2.7. (See [9, Theorem 3.2] Let (R, M) be non-Artinian Noetherian local ring. Then the fol lowing conditions are equivalent: (1) M is principal as a right ideal. (2) M is principal as a left ideal. (3) Every two-sided ideal of R is principal as a left ideal. (4) Every two-sided ideal of R is principal as a right ideal. Moreover, R is a prime ring. Now we are in a position to prove the main theorem of this section. Theorem 2.8. Let (R, M) be a Noetherian local ring. If every left R-module with two generators is a direct sum of cyclic modules, then one of the fol lowing holds: (a) R is an Artinian principal left ideal ring. (b) R is an Artinian principal right ideal ring. (c) R is a prime ring and every two-sided ideal of R is principal as both left and right ideals. Proof. First we assume that R is an Artinian ring. Thus by Theorem 2.6, either R is an Artinian principal left ideal ring or R is an Artinian principal right ideal ring. Now we assume that R is not an Artinian ring. By Theorem 2.4, either M is a principal left ideal or M is a principal right ideal. Thus by Lemma 2.7, R is a 5 prime ring and every two-sided ideal of R is principal as both left and right ideals. (cid:3) 4. A Partial Solution of Kaplansky’s Problem on Duo-Rings A ring R is said to be left (resp. right) hereditary if every left (resp. right) ideal of R is pro jective as a left (resp. right) R-module. If R is both left and right hereditary, we say that R is hereditary. Recall that a PID is a domain R in which any left and any right ideal of R is principal. Clearly, any PID is hereditary. Let R be an hereditary prime ring with quotient ring Q and A be a left R- module. Following Levy [17], we say that a ∈ A is a torsion element if there is a regular element r ∈ R such that ra = 0. Since, by Goldie’s theorem, R satisfies the Ore condition, the set of torsion elements of A is a submodule t(A) ⊆ A. A/t(A) is evidently torsion free (has no torsion elements). Lemma 3.1. (Eisenbud-Robson [6, Theorem 2.1]) Let R be an hereditary Noethe- rian prime ring, and let A be a finitely generated left R-module. Then A/t(A) is projective and A ∼= t(A) ⊕ A/t(A). A Dedekind prime ring [20] is an hereditary Noetherian prime ring with no proper idempotent two-sided ideals (see [7]). Clearly if a duo-ring R is a PID, then R is a Dedekind prime ring. Lemma 3.2. (Eisenbud-Robson [6, Theorem 3.11]) Let R be a Dedekind prime ring. Then every finitely generated torsion left R-module A is a direct sum of cyclic modules. Lemma 3.3. (Eisenbud-Robson [6, Theorem 2.4]) Let R be a Dedekind prime ring, and let A be a projective left R-module. Then: (i) If A is finitely generated, then A ∼= F ⊕ I where F is a finitely generated free module and I is a left ideal of R. (ii) If A is not finitely generated, then A is free. Proposition 3.4. Let R be a Dedekind prime ring. If R is a left principal ideal ring, then R is a left FGC-ring. Proof. Suppose that A is a finitely generated left R-module. Since R is a Dedekind prime ring, R is Noetherian and so A is also a Noetherian left R-module. Thus by Lemma 3.1, A/t(A) is pro jective and A ∼= t(A) ⊕ A/t(A). By Lemma 3.2, t(A) is a direct sum of cyclic modules. Also by Lemma 3.3, A/t(A) ∼= F ⊕ I where F is a free module and I is a left ideal of R. Since R is a principal left ideal ring, I is a cyclic left R-module, i.e., A/t(A) is a direct sum of cyclic modules. Thus, A ∼= t(A) ⊕ A/t(A) is a direct sum of cyclic modules. Therefore, R is a left FGC-ring. (cid:3) 6 The following proposition is an answer to the question: “What is the class of FGC Noetherian prime duo-rings?” Proposition 3.5. (See also Jacobson [11, Page 44, Theorems 18 and 19]) Let R be a Noetherian prime duo-ring (i.e., R is a Noetherian duo-domain). Then the fol lowing statements are equivalents: (1) R is an FGC-ring. (2) R is a left FGC-ring. (3) R is a principal ideal ring. The same characterizations also apply for right R-modules. Proof. (1) ⇒ (2) is clear. (2) ⇒ (3). Suppose that I is an ideal of R. Since I is a direct sum of principal ideals of R and R is a domain, we conclude that I is principal. Thus, R is a principle ideal ring. (3) ⇒ (1) is by Proposition 3.4. (cid:3) A left (resp., right) Kothe ring is a ring R such that each left (resp., right) R-module is a direct sum of cyclic submodules. A ring R is called a Kothe ring if it is a both left and right Kothe ring. In [16] Kothe proved that an Artinian principal ideal ring is a Kothe ring. Furthermore, a commutative ring R is a Kothe ring if and only if R is an Artinian principal ideal ring (see Cohen and Kaplansky [5]). The corresponding problem in the non-commutative case is still open (see [21, Appendix B, Problem 2.48] or Jain-Srivastava [12, Page 40, Problem 1]. Recently, a generalization of the Kothe-Cohen-Kaplansky theorem is given in [1]. In fact: in [1, Corollary 3.3.], it is shown that if R is a ring in which all idempotents are central, then R is a Kothe ring if and only if R is an Artinian principal ideal ring. Next, the following theorem is an answer to the question: “What is the class of FGC Noetherian local duo-rings?” Theorem 3.6. Let (R, M) be a Noetherian local duo-ring. Then the fol lowing statements are equivalent: (1) R is an FGC-ring. (2) R is a left FGC-ring. (3) Every left R-module with two generators is a direct sum of cyclic modules. (4) Either R is an Artinian principal ideal ring or R is a principal ideal domain. (5) R is a principal ideal ring. The same characterizations also apply for right R-modules. Proof. (1) ⇒ (2) ⇒ (3) is clear. 7 (3) ⇒ (4). Suppose that every left R-module with two generators is a direct sum of cyclic modules. Thus by Theorem 2.4, M is principal as both left and right ideals. If R is Artinian, then by Theorem 2.6, R is an Artinian principal ideal ring. If R is not Artinian, then by Lemma 2.7, R is a principal ideal domain. (4) ⇒ (1). If R is an Artinian principal ideal ring, then by the Kothe result, each left, and each right R-module is a direct sum of cyclic modules. Thus R is an FGC- ring. Now assume that R is a principal ideal domain. Then by Proposition 3.5, R is an FGC-ring. (4) ⇒ (5) is clear. (5) ⇒ (4). Assume that R is a principal ideal ring. Then M is principal as both left and right ideals. If R is Artinian, then Lemma 2.5, R is an Artinian principal ideal ring. If R is not Artinian, then by Lemma 2.7, R is a principal ideal domain. (cid:3) Let R = Πn i=1Ri be a finite product of rings Ri . Clearly R is a principal ideal ring if and only if each Ri is a principal ideal ring. On the other hand if R is a left FGC-ring, then each Ri is also a left FGC-ring. Thus as a corollary of Proposition 3.5 and 3.6, we have the following result. Corollary 3.7. Let R = Πn i=1Ri be a finite product of Noetherian duo-rings Ri such that each Ri is a domain or a local ring. Then the fol lowing statements are equivalent: (1) R is an FGC-ring. (2) R is a left FGC-ring. (3) R is a principal ideal ring. The same characterizations also apply for right R-modules. Next, we need the following lemma from [10] about Artinian duo-rings (its proof is worthwhile even in the commutative case (see [10, Corollary 4] or [15, Lemma 4.2]) Lemma 3.8. Let R be an Artinian duo-ring. Then R is a finite direct product of Artinian local duo rings. Next, we give the following characterizations of an Artinian FGC duo-ring. In fact, on Artinian duo-rings, the notions “FGC” and “Kothe” coincide. Theorem 3.9. Let R be an Artinian Duo-ring. Then the fol lowing statements are equivalent: (1) R is a left FGC-ring (2) R is an FGC-ring (3) Every left R-module with two generators is a direct sum of cyclic modules. 8 (4) R is a left Kothe-ring (5) R is a Kothe-ring (6) R is a principal ideal ring. The same characterizations also apply for right R-modules. Proof. Since R is an Artinian duo-ring, by Lemma 3.8, R = Πn i=1Ri such that each Ri is an Artinian local duo-ring. Thus by the Kothe result and Corollary 3.7, the proof is complete. (cid:3) References [1] M. Behboodi, A. Ghorbani, A. Moradzadeh-Dehkordi and S. H. Sho jaee, On left Kothe rings and a generalization of the Kothe-Cohen-Kaplansky theorem, Proc. Amer. Math. Soc. (to appear). [2] W. Brandal, Commutative Rings Whose Finitely Generated Modules decompose, Lecture Notes in Mathematics, Vol. 723 (Springer, 1979). [3] W. Brandal, Almost maximal integral domains and finitely generated modules. Trans. Amer. Math. Soc. 183 (1973), 203-222. [4] W. Brandal and R. Wiegand, Reduced rings whose finitely generated modules decompose. Comm. Algebra 6(2) (1978), 195-201. [5] I. S. Cohen and I. Kaplansky, Rings for which every module is a direct sum of cyclic modules, Math. Z. 54 (1951), 97-101. [6] D. Eisenbud and J. C. Robson, Modules over Dedekind prime rings, J. Algebra 16 (1970), 67-85. [7] D. Eisenbud and J. C. Robson, Hereditary Noetherian prime rings, J. Algebra 16 (1970), 86-104. [8] A. Facchini, On the structure of torch rings. Rocky Mountain J. Math. 13(3) (1983), 423-428. [9] M. H. Fahmy and S. Fahmy, On non-commutative noetherian local rings, non- commutative geometry and particle physics, Chaos, Solitons and Fractals 14 (2002), 1353-1359. [10] J. M. Habeb, A note on zero commutative and duo rings, Math. J. Okayama Univ 32 (1990), 73-76. [11] N. Jacobson, The Theorey of Rings, Amerian Mathematical Society Mathemat- ical Syrveys, vol. I, American Mathematical Society, New York, 1943. 9 Srivastava, and Ashish K. Jain [12] S. K. by right ideals: cyclic modules and their http://euler.slu.edu/∼srivastava/articles.html ). Rings characterized A survey-I.(See [13] I. Kaplansky, Elementary divisors and modules. Trans. Amer. Math. Soc. 66, (1949). 464-491. [14] I. Kaplansky, Modules over Dedekind rings and valuation rings. Trans. Amer. Math. Soc. 72 (1952), 327-340. [15] N. S. Karamzadeh and O. A. S. Karamzadeh, On artinian modules over Duo rings, Comm. Algebra 38 (2010), 3521-3531. [16] G. Kothe, Verallgemeinerte abelsche gruppen mit hyperkomplexem operatoren- ring, Math. Z. 39 (1935), 31-44. [17] L. Levy, Torsion-Free and divisible modules over non-integral domains, Canad. J. Math. 15 (1963), 132-151. [18] W. K. Nicholson and E. S´anchez-Campos, Rings with the dual of the isomor- phism theorem, J. Algebra 271(1) (2004), 391-406. [19] R. S. Pierce, Modules over commutative regular rings. Mem. Amer. Math. Soc. 70 (1967). [20] J. C. Robson, Non-commutative Dedekind rings, J. Algebra 9 (1968), 249-265. [21] L. Sabinin, L. Sbitneva and I. Shestakov, Non-associative algebra and its appli- cations, Lecture Notes in Pure and Applied Mathematics 246. Chapman and Hall/CRC 2006. [22] T. S. Shores and R. Wiegand, Rings whose finitely generated modules are direct sums of cyclics. J. Algebra 32 (1974), 152-172. [23] P. V´amos, The decomposition of finitely generated modules and fractionally self-injective rings. J. London Math. Soc. 16 (1977), 209-220. [24] R. Wiegand and S. Wiegand, Commutative rings whose finitely generated mod- ules are direct sums of cyclics. Abelian group theory (Proc. Second New Mexico State Univ. Conf., Las Cruces, N.M., 1976), pp. 406423. Lecture Notes in Math., Vol. 616, Springer, Berlin, 1977. 10
1612.00069	2	1612	2017-10-03T09:29:16	D-groups and the Dixmier-Moeglin equivalence	[ "math.RA" ]	A differential-algebraic geometric analogue of the Dixmier-Moeglin equivalence is articulated, and proven to hold for $D$-groups over the constants. The model theory of differentially closed fields of characteristic zero, in particular the notion of analysability in the constants, plays a central role. As an application it is shown that if $R$ is a commutative affine Hopf algebra over a field of characteristic zero, and $A$ is an Ore extension to which the Hopf algebra structure extends, then $A$ satisfies the classical Dixmier-Moeglin equivalence. Along the way it is shown that all such $A$ are Hopf Ore extensions	math.RA	math	D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA Abstract. A differential-algebraic geometric analogue of the Dixmier-Moeglin equivalence is articulated, and proven to hold for D-groups over the constants. The model theory of differentially closed fields of characteristic zero, in par- ticular the notion of analysability in the constants, plays a central role. As an application it is shown that if R is a commutative affine Hopf algebra over a field of characteristic zero, and A is an Ore extension to which the Hopf algebra structure extends, then A satisfies the classical Dixmier-Moeglin equivalence. Along the way it is shown that all such A are Hopf Ore extensions in the sense of [Brown et al., "Connected Hopf algebras and iterated Ore extensions", Jour- nal of Pure and Applied Algebra, 219(6), 2015]. Contents Introduction 1. 2. The δ-DME for D-groups over the constants 3. Twisting by a group-like element 4. The DME for Ore extensions of commutative Hopf agebras References 1 4 16 25 33 1. Introduction This article is about an analogue of the Dixmier-Moeglin equivalence for differential- algebraic geometry. (The immediate motivation is an application to the classical noncommutative Dixmier-Moeglin problem, which we will describe later in this introduction.) The main objects of study here are D-varieties. An introduction to this category is given in §2.1, but let us at least recall here that a D-variety (over the constants) is an algebraic variety V over a field k of characteristic zero, equipped with a regular section to the tangent bundle s : V → T V . A D-subvariety is an algebraic subvariety W for which the restriction s↾W is a section to the tangent bundle of W . There are natural notions of D-morphism and D-rational map. For convenience, let us assume that k is algebraically closed. We are interested in the following properties of an irreducible D-subvariety W ⊆ V over k: Date: October 4, 2017. 2010 Mathematics Subject Classification: 03C98, 12H05, 16T05, 16S36. Keywords: D-groups, model theory of differentially closed fields, Dixmier-Moeglin equivalence, Hopf Ore extensions. Acknowledgements: J. Bell and R. Moosa were partially supported by their respective NSERC Discovery Grants. 1 2 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA • δ-primitivity. There is a k-point of W that is not contained in any proper D-subvariety of W over k. • δ-local-closedness. There is a maximum proper D-subvariety of W over k. • δ-rationality. There is no nonconstant rational map from (W, s) to (A1, 0) over k, where here 0 denotes the zero section to the tangent bundle of the affine line. The question is, for which ambient D-varieties (V, s) are these three properties equivalent for all D-subvarieties? It is not hard to see, and is spelled out in the proof of Corollary 2.14 below, that in general δ-local-closedness =⇒ δ-primitivity =⇒ δ-rationality. In earlier work [1], together with St´ephane Launois, we used the model theory of the Manin kernel to produce (in any dimension ≥ 3) a D-variety which is itself δ-rational but not δ-locally-closed. Here we focus on positive results. The main one, which appears as Corollary 2.17 below, is the following: Theorem A. Suppose (G, s) is a D-group over the constants – that is, G is an algebraic group and s : G → T G is a homomorphism of algebraic groups. Then for any D-subvariety of (G, s), δ-rationality implies δ-local-closedness. In particular, for every D-subvariety of (G, s), δ-rationality, δ-primitivity, and δ-local-closedness are equivalent properties. The proof of Theorem A relies on the model theory of differentially closed fields. In model-theoretic parlance, the point is that δ-rationality of (V, s) is equivalent to the generic type of the corresponding Kolchin closed set being weakly orthogonal to the constants, while δ-local-closedness means that the type is isolated. One context in which one can prove, using model-theoretic binding groups, for example, that weak orthogonality to the constants implies isolation, is when the type in question is analysable in the constants. We give a geometric explanation of analysability in §2.5 in terms of what we call compound isotriviality of D-varieties. The reader can look there for a precise definition, but suffice it to say that a compound isotrivial D-variety is one that admits a finite sequence of fibrations where at each stage the fibres are isomorphic (possibly over a differential field extension of the base) to D- varieties where the section is the zero section. We show that for compound isotriv- ial D-varieties δ-rationality implies δ-local-closedness (Proposition 2.13). Then we show, using known results about the structure of differential-algebraic groups, that every D-subvariety of a D-group over the constants is compound isotrivial (Propo- sition 2.16). Theorem A follows. It turns out that for our intended application, namely Theorem B2 appearing later in this introduction, we need Theorem A to work for D-varieties that are slightly more general than D-groups. Given an affine algebraic group G, we may as well assume that G ⊆ GLn, so that a regular section to the tangent bundle is then of the form s = (id, s) where s : G → Matn. It is not hard to check from how the algebraic group structure is defined on the tangent bundle, that s : G → T G being a homomorphism is equivalent to the following identity: s(gh) = s(g)h + gs(h), where g, h ∈ G are matrices and all addition and multiplication here is matrix addition and multiplication. Now suppose we are given a homomorphism to the multiplicative group, a : G → Gm. By an a-twisted D-group we mean a D-variety (G, s) where G is an affine algebraic group and s = (id, s) satisfies the identity: s(gh) = s(g)h + a(g)gs(h). So an a-twisted D-group is a D-group exactly D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 3 when a = 1. We are able to show that D-subvarieties of a-twisted D-groups are also compound isotrivial. This yields the following generalisation of Theorem A: For any D-subvariety of an a-twisted D-group over the constants, δ-rationality, δ-primitivity, and δ-local-closedness are equivalent properties. The passage from D-groups to a-twisted D-groups turns out to be technically quite difficult, and is done in Section 3. It is worth pointing out that we have been intentionally ambigious about the field of definitions in the statements of Theorem A and its a-twisted generalisation. The reason for this is that the results actually hold true for D-subvarieties of (G, s) that are defined over differential field extensions of the base field k. To make this precise one has to give a more general definition of D-variety using prolongations rather than tangent bundles, and we have decided to delay this to the main body of the article. While the final conclusion we are interested in is about D-subvarieties over k, this possibility of passing to base extensions is an important part of the inductive arguments involved. Now for the application to noncommutative algebra, to which Section 4 is dedi- cated. The classical Dixmier-Moeglin equivalence (DME) is about prime ideals in a noetherian associative algebra over a field of characteristic zero; it asserts the equiv- alence between three properties of such prime ideals: primitivity (a representation- theoretic property), local-closedness (a geometric property), and rationality (an algebraic property). Precise definitions are given at the beginning of §4. We are interested in the question of when the DME holds for skew polynomial rings R[x; δ] over finitely generated commutative integral differential k-algebras (R, δ). Recall that R[x; δ] is the noncommutative polynomial ring in x over R where xr = rx+δ(r). This question is not vacuous since examples of such skew polynomial rings failing the DME were given in [1]; indeed, these were the first counterexamples to the DME of finite Gelfand-Kirillov dimension. The connection to D-varieties should be clear: R = k[V ] for some irreducible algebraic variety V , and the k-linear derivation δ induces a regular section s : V → T V . So the study of such (R, δ) is precisely the same thing as the study of D-varieties. We are able to prove (this is Proposition 4.6 below) that R[x; δ] will satisfy the DME if δ-rationality implies δ-locally-closedness for all D-subvarieties of the D-variety (V, s) associated to (R, δ). Theorem A there- fore answers our question in the special case of differential Hopf algebras. Theorem B1. If (R, δ) is a finitely generated commutative integral differential Hopf k-algebra then R[x; δ] satisfies the DME. Being a differential Hopf algebra means that R has the structure of a Hopf algebra and that δ commutes with the coproduct – this is equivalent to saying that (R, δ) comes from an affine D-group (G, s). More generally than skew polynomial rings, we consider Ore extensions: Sup- pose R is a finitely generated commutative integral k-algebra, σ is a k-algebra automorphism of R, and δ is a k-linear σ-derivation of R – meaning that δ(rs) = σ(r)δ(s) + δ(r)s. Recall that the Ore extension R[x; σ, δ] is the noncommutative polynomial ring in the variable x over R where xr = σ(r)x + δ(r). So when σ = id we are in the skew polynomial case discussed above. What about the DME for R[x; σ, δ]? 4 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA Theorem B2. Suppose R is a finitely generated commutative integral Hopf k- algebra. If an Ore extension R[x; σ, δ] admits a Hopf algebra structure extending that on R, then R[x; σ, δ] satisfies the DME. That Theorem B1 is a special case of Theorem B2 uses the (known) fact that one can always extend the Hopf structure on a differential algebra R to the skew polynomial ring extension R[x; δ], namely by the coproduct induced by ∆(x) = x ⊗ 1 + 1 ⊗ x. Theorem B2 appears as Theorem 4.1 below. Its proof goes via a reduction to the case when σ = id and then an application of the stronger a-twisted version of Theorem A discussed above. Both of these steps use the work of Brown et al. [5] on Hopf Ore extensions. One obstacle is that while their results hold for much more general R than we are considering, they are conditional on the coproduct of the variable x in the Ore extension taking the special form ∆(x) = a ⊗ x + x ⊗ b + v(x ⊗ x) + w where a, b ∈ R and v, w ∈ R ⊗k R. This is part of their definition of a Hopf Ore extension, though they speculate about its necessity. We prove that when k is algebraically closed, after a linear change of variable, ∆(x) always has the above form. This is Theorem 4.2 below, and may be of independent interest: Theorem C. Suppose k is algebraically closed and R is a finitely generated commu- tative integral Hopf k-algebra. If an Ore extension R[x; σ, δ] admits a Hopf algebra structure extending that of R then, after a linear change of the variable x, ∆(x) = a ⊗ x + x ⊗ b + w for some a, b ∈ R, each of which is either 0 or group-like, and some w ∈ R ⊗k R. In particular, R[x; σ, δ] is a Hopf Ore extension of R. It has been conjectured [2] that all finitely generated complex noetherian Hopf algebras of finite Gelfand-Kirillov dimension satisfy the DME. Theorem B2 verifies a special case. To make more significant progress on this conjecture one would like to pass from Hopf Ore extensions to iterated Hopf Ore extensions. As of now, this appears to be beyond the scope of the techniques used here. Throughout this paper, by an affine k-algebra we mean a finitely generated commutative k-algebra that is an integral domain. Acknowledgements. We are grateful to an anonymous referee for a very thorough reading which lead to the discovery of an error in an initial version of this paper. 2. The δ-DME for D-groups over the constants In this chapter we prove Theorem A of the introduction. After some preliminar- ies, we articulate in §2.3 the differential-algebraic geometric analogue of the DME suggested in the introduction, and call it the δ-DME. A sufficient condition for this to hold in terms of the model-theoretic notion of analysability to the constants is given in §2.5, and then applied to show that D-groups over the constants sat- isfy the δ-DME in §2.6. In a final section we reformulate δ-DME algebraically, as a statement about commutative differential Hopf algebras, thereby preparing the stage for the application to the classical DME in chapter 4. D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 5 2.1. Preliminaries on D-varieties. Suppose k is a field of characteristic zero equipped with a derivation δ. In this section we review the notion of a D-variety over k. Several more detailed expositions can be found in the literature, for instance Buium [6] who introduced the notion, and also [16, §2]. We first need to recall what prolongations are. If V ⊆ An is an affine algebraic variety over k, then by the δ-prolongation of V is meant the algebraic variety τ V ⊆ A2n over k whose defining equations are P δ(X1, . . . , Xn) + P (X1, . . . , Xn) = 0 (X1, . . . , Xn) · Xi = 0 ∂P ∂Xi nXi=1 for each P ∈ I(V ) ⊂ k[X1, . . . , Xn]. Here P δ denotes the polynomial obtained by applying δ to all the coefficients of P . The projection onto the first n coordinates gives us a surjective morphism π : τ V → V . Note that if K is any δ-field extension of k, and a ∈ V (K), then ∇(a) := (a, δa) ∈ τ V (K). If V is defined over the constant field of (k, δ) then τ V is nothing other than the tangent bundle T V . In general, τ V will be a torsor for the tangent bundle; for each a ∈ V the fibre τaV is an affine translate of the tangent space TaV . In particular, if V is smooth and irreducible then so is τ V . Taking prolongations is a functor which acts on morphisms f : V → W by acting on their graphs. It preserves the following properties of a morphism: ´etale- ness, being a closed embedding, and being smooth. The functor τ acts naturally on rational maps also; this is because for U a Zariski open subset of an irreducible variety V , τ V ↾U = τ (U ) is Zariski open in τ (V ). Moreover, prolongations commute with base extension to δ-field extensions. We have restricted our attention here to the affine case merely for concreteness. The prolongation construction extends to abstract varieties by patching over an affine cover in a natural and canonical way. A D-variety over k is an algebraic variety V over k equipped with a regular section s : V → τ V over k. An example is when V is defined over the constants and s : V → T V is the zero section. If V is affine then a D-variety structure on V is nothing other than an extension Indeed, if s : V → τ V is given by s(X) = of δ to the co-ordinate ring k[V ]. k[X] by Xj 7→ sj(X), and this will induce a derivation on k[V ]. Conversely, given (cid:0)X, s1(X), . . . , sn(X)(cid:1) in variables X = (X1, . . . , Xn), then we can extend δ to an extension of δ to k[V ], and choosing sj(X) to be such that δ(cid:0)Xj + I(V )(cid:1) = sj(X) + I(V ), we get that s :=(cid:0) id, s1, . . . , sn(cid:1) : V → τ V is a regular section. A D-subvariety of (V, s) is a closed algebraic subvariety W ⊆ V , over a possibly larger δ-field K, such that s(W ) ⊆ τ W . In principle one should talk about the base extension of V to K before talking about subvarieties over K, but as prolongations commute with base extension, and following standard model-theoretic practices, we allow D-subvarieties to be defined over arbitrary δ-field extensions unless explicitly stated otherwise. A D-variety (V, s) over k is said to be k-irreducible if V is k-irreducible as an algebraic variety. In this case s induces on k(V ) the structure of a δ-field extend- In ing k. A D-variety (V, s) is called irreducible if V is absolutely irreducible. 6 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA general, every irreducible component of V is a D-subvariety over kalg and these are called the irreducible components of (V, s). A morphism of D-varieties (V, s) → (W, t) is a morphism of algebraic varieties f : V → W such that τ V s V τ f / τ W f t / W commutes. It is not hard to verify that the pull-back of a D-variety, and the Zariski closure of the image of a D-variety, under a D-morphism, are again D-varieties. In the same way, we can talk about rational maps between D-varieties. A useful fact is that if U is a nonempty Zariski open subset of V , then the prolongation of U is the restriction of τ V to U , and so (U, s↾U ) is a D-variety in its own right. So a rational map on V is a D-rational map if it is a D-morphism when restricted to the Zariski open subset on which it is defined. A D-constant rational function on a D-variety (V, s) over k is a rational map over k from (V, s) to (A1, 0) where 0 denotes the zero section to the tangent bundle of the affine line. In the case when (V, s) is k-irreducible, they correspond precisely to the δ-constants of(cid:0)k(V ), δ(cid:1). 2.2. Differentially closed fields and the Kolchin topology. Underlying our approach to the study of D-varieties is the model theory of existentially closed δ- fields (of characteristic zero). These are δ-fields K with the property that any finite sequence of δ-polynomial equations and inequations over K which have a solution in some δ-field extension, already have a solution in K. The class of existentially closed δ-fields of characteristic zero is axiomatisable in first-order logic, and its theory is denoted by DCF0. We will work in a fixed model of this theory, an existentially closed δ-field K. In particular, K is algebraically closed. We let K δ denote the field of constants of K; it is an algebraically closed field that is pure in the sense that the structure induced on it by DCF0 is simply that given by the language of rings. Suppose (V, s) is a D-variety over K. Let x ∈ V (K). Note that {x} is a D- subvariety if and only if ∇(x) = s(x) where recall that ∇ : V (K) → τ V (K) is given by x 7→ (x, δx). We call such points D-points, and denote the set of all D-points in V (K) by (V, s)♯(K). It is an example, the main example we will encounter, of a Kolchin closed subset of V (K). In general a Kolchin closed subset of the K-points of an algebraic variety is one that is defined Zariski-locally by the vanishing of δ- polynomials. Note that when (V, s) is defined over the constants and s is the zero section, (V, s)♯(K) = V (K δ). One of the main consequences of working in an existential closed δ-field is that (V, s)♯(K) is Zariski-dense in V (K). In particular, for any subvariety W ⊆ V , we have that W is a D-subvariety if and only if W ∩ (V, s)♯(K) is Zariski dense in W (K). Suppose (V, s) is k-irreducible for some δ-subfield k. If we allow ourselves to pass to a larger existentially closed δ-field, then we can always find a k-generic D-point of (V, s), that is, a D-point x ∈ (V, s)♯(K) that is Zariski-generic over k in V . Note that such a point is also Kolchin-generic in (V, s)♯(K) over k in the sense that it is not contained in any proper Kolchin closed subset defined over k. / O O / O O D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 7 In order to ensure the existence of generic D-points without having to pass to larger δ-fields, it is convenient to assume that K is already sufficiently large, namely saturated. This means that if k is a δ-subfield of strictly smaller cardinality than K, and F is a collection of Kolchin constructible sets over k every finite subcollection of which has a nonempty intersection, then F has a nonempty intersection. 2.3. An analogue of the DME for D-varieties. We fix from now on a satu- rated existentially closed δ-field K of sufficiently great cardinality. So K serves as a universal domain for δ-algebraic geometry (and hence, in particular, algebraic geometry). We also fix a small δ-subfield k ⊂ K that will serve as the field of coefficients. Definition 2.1 (δ-DME for D-varieties). Suppose (V, s) is a D-variety over k. We say that (V, s) satisfies the δ-DME over k, if for every k-irreducible D-subvariety W ⊆ V , the following are equivalent: (i) (W, s) is δ-primitive: there exists a point p ∈ W (kalg) that is not contained in any proper D-subvariety of W over k. (ii) (W, s) is δ-locally-closed: it has a maximum proper D-subvariety over k. (iii) (W, s) is δ-rational: k(W )δ ⊆ kalg. Remark 2.2. As the model-theorist will notice, and as we will prove in the next sec- tion, W being δ-locally closed means that the Kolchin generic type p of (W, s)♯(K) over k is isolated. The model-theoretic meaning of δ-rationality is that p is weakly orthogonal to the constants. On the other hand, it is not clear how to express a priori the δ-primitivity of W as a model-theoretic property of p. Without additional assumptions on k there is no hope for the δ-DME being satisfied. For example, there are positive-dimensional δ-rational D-varieties over any k, but if k is differentially closed then the only δ-locally closed D-varieties over k are zero-dimensional. This is because every D-variety over a differentially closed field k will have a Zariski dense set of D-points over k, and so a D-subvariety over k containing all of them could not be proper. We are interested, however, in the case when k is very much not differentially closed; namely, when δ is trivial on k. Proposition 2.3. For any k-irreducible D-variety, δ-local-closedness implies δ- primitivity. Moreover, if k ⊆ K δ then δ-primitivity implies δ-rationality. Proof. Let (W, s) be a k-irreducible D-variety. Suppose (W, s) is δ-locally-closed, and denote by A the maximum proper D- subvariety of W over k. Then p ∈ W (kalg) \ A(kalg) witnesses δ-primitivity. This proves the first assertion. Now suppose that k ⊆ K δ, (W, s) is δ-primitive, and p ∈ W (kalg) is not contained in any proper D-subvariety over k. Suppose f ∈ k(W ) is a δ-constant. We want to show that f ∈ kalg. We view it as a rational map of D-varieties, f : (W, s) → (A1, 0), and suppose for now that it is defined at p. So f (p) ∈ A1(kalg). Because of our additional assumption that k ⊆ K δ, and hence kalg ⊆ K δ, we have that f (p) is a D-point of (A1, 0). Now, let Λ be the orbit of f (p) under the action of the absolute Galois group of k. Then Λ is a finite D-subvariety of (A1, 0) over k. Hence the Zariski closure of f −1(Λ) is a D-subvariety of W over k that contains p. It follows that f −1(Λ) = W . So f is kalg-valued on all of W . We have shown that every element of k(W )δ that is defined at p is in kalg. 8 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA We have still to deal with the possibility that f is not defined at p. In that case, writing f = a b with a, b ∈ k[W ], we must have b(p) = 0. The fact that δf = 0 b = δa implies by the quotient rule that bδa = aδb. So either δa = δb = 0 or f = a δb . Since a and b are defined at p, if δa = δb = 0 then a, b ∈ kalg by the previous paragraph, and hence f ∈ kalg. If, on the other hand, f = δa δb , then we iterate the argument with (δa, δb) in place of (a, b). What we get in the end is that either f ∈ kalg or f = δℓa δℓb for all ℓ ≥ 0. We claim the latter is impossible. Indeed, it would imply that δℓb(p) = 0 for all ℓ, and so p is contained in the D-subvariety V (I) where I is the δ-ideal of k[W ] generated by b. But the assumption on p would then imply that V (I) = W , contradicting the fact that b 6= 0. So f ∈ kalg, as desired. (cid:3) So the question becomes: Question 2.4. Under the assumption that δ is trivial on k, for which D-varieties does δ-rationality imply δ-local-closedness? Question 2.4 should be, we think, of general interest in differential-algebraic geometry. In [1] it was pointed out that Manin kernels can be used to construct, in all Krull dimensions at least three, examples that were δ-rational but not δ-locally closed. Let us point out that in dimension ≤ 2 the answer is affirmative: Proposition 2.5. If k ⊆ K δ then every D-variety over k of dimension ≤ 2 satisfies the δ-DME over k. Proof. Suppose (V, s) is a D-variety over k of dimension at most 2. By Proposi- tion 2.3 it suffices to show that if W ⊆ V is a k-irreducible δ-rational D-subvariety over k then it has a maximum proper D-subvariety over k. We may assume that dim W > 0. Now, it is a known fact that δ-rationality implies the existence of only finitely many D-subavrieties of codimension one over k. Indeed, this is an unpub- lished theorem of Hrushovski [10, Proposition 2,3]; see [1, Theorem 6.1] and [8, The- orem 4.2] for published generalisations. So it remains to consider the 0-dimensional D-subvarieties of W over k. But as k ⊆ K δ, the union of these is contained in the Zariski closure X of (W, s)♯(K) ∩ W (K δ). Note that s restricts to the zero section on X, and hence X is a D-subvariety of W over k that must be proper by δ-rationality of (W, s). So the union of X and the finitely many codimension one D-subvarieties of W form the maximal proper D-subvariety over k. (cid:3) We will give a sufficient condition for δ-rationality to imply δ-local-closedness, and hence for the δ-DME, having to do with analysability to the constants in the model theory of differentially closed fields. We will then use this condition to prove the δ-DME for D-groups over the constants. 2.4. Maximum D-subvarieties. Here we look closer at which D-varieties over k have a proper D-subvariety over k that contains all other proper D-subvarieties over k. This is something that never happens in the pure algebraic geometry setting: every variety over k has a Zariski dense set of kalg-points, and each kalg- point is contained in a finite subvariety defined over k. So a k-irreducible variety cannot have a maximum proper subvariety over k. In the enriched context of D- varieties there will be many D-points over a differential closure of k, sayek, but a ek-point need not live in a proper D-subvariety defined over k. So D-points are not D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 9 an obstacle to the existence of a maximum proper D-subvariety. In fact, as the following lemma points out, the existence of a maximum proper D-subvariety is a natural property to consider from both the Kolchin topological and model-theoretic points of view. We continue to work in our sufficiently saturated differentially closed field (K, δ), and fix a δ-subfield k of coefficients. Lemma 2.6. Suppose (V, s) is a k-irreducible D-variety. The following are equiv- alent. (i) (V, s) is δ-locally closed. (ii) (V, s) has finitely many maximal proper k-irreducible D-subvarieties. (iii) Λ := (V, s)♯(K) \S{(W, s)♯(K) : W ( V D-subvariety over k} is Kolchin (iv) The Kolchin generic type of (V, s)♯(K) over k is an isolated type. constructible. Proof. (i) =⇒ (ii). Let W be the maximum proper D-subvariety over k. The k-irreducible components of W are D-subvarieties of V , see for example [14, Theo- rem 2.1]. Every proper k-irreducible D-subvariety of V is contained in one of these components. So the maximal proper k-irreducible D-subvarieties of V are precisely the k-irreducible components of W . (ii) =⇒ (iii). Let W1, . . . , Wℓ be the maximal proper k-irreducible D-subvarieties of V . Then [{(W, s)♯(K) : W ( V D-subvariety over k} = (W1, s)♯(K) ∪ · · · ∪ (Wℓ, s)♯(K). (iii) =⇒ (iv). The Kolchin generic type of (V, s)♯(K) over k is the complete type p(X) in DCF0 axiomatised by the formulas saying that "X ∈ (V, s)♯(K)", and, for each proper Kolchin closed subset A of (V, s)♯(K) over k, the formula "X 6∈ A". Note that as (V, s)♯(K) is defined by ∇(X) = s(X), the occurrences of each δX in the defining equations of A can be replaced by polynomials, so that A = AZar ∩ (V, s)♯(K), where AZar denotes the Zariski closure of A in V . When A is over k, we have that AZar is a D-subvariety of V over k. It follows that the set of realisations of p is precisely Λ, so that Λ being Kolchin constructible implies that p is axiomatised by a single formula, that is, it is isolated. (iv) =⇒ (i). Let Λ be as in statement (iii). As we have seen, this is the set of re- alisations of the Kolchin generic type of (V, s)♯(K) over k. The latter being isolated implies, by quantifier elimination, that Λ is Kolchin constructible. By saturation, this in turn implies that A := S{(W, s)♯(K) : W ( V D-subvariety over k} is a finite union, and hence is itself a proper Kolchin closed subset over k. Then AZar is the maximum proper D-subvariety over k. (cid:3) The following lemma will be useful in showing that certain D-varieties satisfy the equivalent conditions of Lemma 2.6. Lemma 2.7. Suppose f : (V, s) → (W, t) is a dominant D-rational map of k- irreducible D-varieties over k. The following are equivalent (i) (V, s) has a maximum proper D-subvariety over k. (ii) (W, t) has a maximum proper D-subvariety over k, and for some (equiv- alently every) k-generic D-point η of W , the fibre Vη := f −1(η)Zar has a maximum proper D-subvariety over k(η). 10 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA Proof. We show how this follows easily from basic properties of isolated types, leaving it to the reader to make the straightforward, but rather unwieldy, translation it into an algebro-geometric argument if desired. (i) =⇒ (ii). Suppose a ∈ (V, s)♯(K) is a k-generic D-point. Since f is a dominant D-rational map, f (a) ∈ (W, T )♯(K) is also k-generic. By characterisation (iv) of the previous lemma, tp(a/k) is isolated. As f (a) is in the definable closure of a over k, it follows that tp(f (a)/k) is isolated. Hence, (W, t) has a maximum proper D-subvariety over k. Now fix η ∈ (W, t)♯(K) a k-generic D-point, and let a be a k(η)-generic D-point It of the fibre Vη. Then a is k-generic in (V, t), and hence tp(a/k) is isolated. follows that the extension tp(a/k(η)) is also isolated. So Vη has a maximum proper D-subvariety over k(η). (ii) =⇒ (i). Fix η ∈ (W, t)♯(K) a k-generic D-point such that Vη has a max- imum proper D-subvariety over k(η). Let a be a k(η)-generic D-point of Vη. So tp(a/k(η)) and tp(η/k) are both isolated, implying that tp(a/k) is isolated. Since a ∈ (V, s)♯(K) is k-generic, condition (i) follows. (cid:3) 2.5. Compound isotriviality. Our sufficient condition for δ-rationality to imply δ-local-closedness will come from looking at isotrivial D-varieties. Definition 2.8. An irreducible D-variety (V, s) over k is said to be isotrivial if there is some δ-field extension F ⊇ k such that (V, s) is D-birationally equivalent over F to a D-variety of the form (W, 0) where W is defined over the constants F δ and 0 is the zero section. We will say that a possibly reducible D-variety is isotrivial if every irreducible component is. This definition comes from model theory, it is a geometric translation of the state- ment that the Kolchin generic type of (V, s)♯(K) over k is K δ-internal. Note that there is some tension, but no inconsistency, between isotriviality and δ-rationality; for example, (W, 0) is far from being δ-rational, instead of there being no new δ- constants in the rational function field we have that δ is trivial on all of k(W ). The reasons these notions are not inconsistent is that the isotrivial (V, s) is only of the form (W, 0) after base change – that is, over additional parameters – and that makes all the difference. Fact 2.9. A k-irreducible D-variety that is at once both δ-rational and isotrivial must be δ-locally closed. Proof. Suppose (V, s) is a k-irreducible isotrivial D-variety with k(V )δ ⊆ kalg. We want to show that V has a maximum proper D-subvariety over k. The proof we give makes essential use of model theory. We will show how the statement translates to the well-known fact that a type internal to the constants but weakly orthogonal to the constants is isolated. Let p be the Kolchin generic type of (V, s)♯(K) over k. By Lemma 2.6, it suffices to show that p is isolated. That in turn reduces to showing that every extension of p to kalg is isolated. Fix q an extension of p to kalg. So q is the Kolchin generic type of (bV , s)♯(K) over kalg, for some irreducible component bV of V . Isotriviality of (V, s) implies isotriviality of (bV , s), and this means that q is internal to the constants K δ, see for example [16, Fact 2.6]. By stability, this implies that the binding group G = Aut(q/kalg(K δ)) is type-definable over kalg, see for instance [11, Appendix B]. D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 11 In fact, G is definable: G lives in the constants and by ω-stability of the induced structure on the constants, every type-definable group in K δ is a definable group. So we have a definable group acting definably on the set of realisations of q. On the other hand, for all a = q we have that kalg(a)δ ⊆ (k(a)alg)δ = (k(a)δ)alg ⊆ kalg, where the last containment uses our assumption on k(V ) = k(a). This shows that q is weakly orthogonal to K δ. So the action of G on the set of realisations of q is transitive. As G is definable, the set of realisations of q must be definable – that is, q is isolated. (cid:3) Using Lemma 2.7 we can extend Fact 2.9 to the case of D-varieties that are built up by a finite sequence of fibrations by isotrivial D-subvarieties. Definition 2.10. An irreducible D-variety (V, s) over k is said to be compound isotrivial if there exists a sequence of irreducible D-varieties (Vi, si) over k, for i = 0, . . . , ℓ, with dominant D-rational maps over k V = V0 f0 / V1 f1 / · · · fℓ−1 / Vℓ−1 / Vℓ = 0 where 0 denotes an irreducible zero-dimensional D-variety, and such that the generic fibres of each fi are isotrivial. That is, for each i = 0, . . . , ℓ − 1, if η is a k-generic D-point in Vi+1, then f −1 (η)Zar, which is a k(η)-irreducible D-subvariety of (Vi, si), is isotrivial. We say (V, s) is compound isotrivial in ℓ steps. i While isotriviality is equivalent to the Kolchin generic type being internal to the constants, compound isotriviality corresponds to that type being analysable in the constants. As this is a less familiar notion, even among model theorists, we spell out the equivalence here. Lemma 2.11. Suppose (V, s) is an irreducible D-variety over k, and a ∈ (V, s)♯(K) is a k-generic D-point. Then (V, s) is compound isotrivial if and only if the type of a over k in DCF0 is analysable in K δ. Proof. Analysability in K δ means that there are tuples a = a0, a1, . . . , aℓ, such that (i) ai is in the δ-field generated by ai−1 over k, for i = 1, . . . , ℓ − 1, aℓ ∈ k, and (ii) the type of ai over the algebraic closure of the δ-field generated by k(ai+1) is internal to K δ. If (V, s) is compound isotrivial one simply takes ai = fi−1(ai−1) for i = 1, . . . , ℓ. Condition (i) is clear – in fact with "δ-field generated by" replaced by "field gen- erated by" – and condition (ii) follows from the fact that ai will be a k(ai+1)alg- generic D-point of one of the irreducible components of f −1 (ai+1)Zar, all of which are isotrivial. For the converse, given a = a0, a1, . . . , aℓ satisfying condition (i) and (ii), one first replaces ai with (ai, δ(ai), . . . , δn(ai)) for some sufficiently large n so that ai is a k-generic D-point of an irreducible D-variety (Vi, si) over k. This sequence of D-varieties will witness the compound isotriviality, using the fact that the irreducible components of f −1 (ai+1)Zar are all conjugate over k(ai+1) and hence the isotriviality of one implies the isotriviality of them all. (cid:3) i i Remark 2.12 (Stability under base change). The definition of compound isotrivial- ity seems to be sensitive to parameters; the D-varieties Vi and the D-rational maps fi need also be defined over k. In fact the notion is stable under base change: if / / / / 12 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA an irreducible D-variety (V, s) over k is compound isotrivial when viewed as a D- variety over some δ-field extension F ⊇ k then it was already compound isotrivial over k. A model-theoretic re-statement of this is the well-known fact that a station- ary type with a nonforking extension that is analysable in the constants is already analysable in the constants. We leave it to the reader to formulate a geometric argument. Note also that (compound) isotriviality is preserved by D-birational maps. Proposition 2.13. For an irreducible compound isotrivial D-variety over k, δ-rationality implies δ-local-closedness. Proof. Suppose (V, s) is an irreducible compound isotrivial D-variety over k with k(V )δ ⊆ kalg. We need to show that V has a maximum proper D-subvariety over k. We proceed by induction on the number of steps witnessing the compound isotriviality. The case ℓ = 0 is vacuous. Suppose we have a compound isotrivial (V, s) witnessed by V = V0 f0 / V1 f1 / · · · fℓ−1 / Vℓ−1 / Vℓ = 0 with ℓ ≥ 1. Then V1 is compound isotrivial in ℓ − 1 steps, and as k(V1) is a δ- subfield of k(V ) by dominance of f0, the induction hypothesis applies to give us a maximum proper D-subvariety of V1 over k. On the other hand, the generic fibre Vη := f −1 0 (η)Zar is an isotrivial k(η)- irreducible D-subvariety of V ; where η is a k-generic D-point of V1. Moreover, as k(η)(Vη) = k(V ), Vη is δ-rational and therefore Fact 2.9 applies to Vη and we obtain a maximum proper D-subvariety over k(η). Now Lemma 2.7 implies that V has a maximum proper D-subvariety over k. (cid:3) Corollary 2.14. Suppose k ⊆ K δ and (V, s) is a D-variety over k with the property that every irreducible D-subvariety of V over kalg is compound isotrivial. Then (V, s) satisfies δ-DME. Proof. By Proposition 2.3, it suffices to show that every δ-rational k-irreducible D-subvariety (W, s) is δ-locally closed. Note that if k = kalg then (W, s) is abso- lutely irreducible, and compound isotrivial by assumption, so that δ-local-closedness follows by Proposition 2.13. In general, let (W0, s) be an absolutely irreducible component of (W, s). It is over kalg. The δ-rationality of (W, s) over k implies the δ-rationality of (W0, s) over kalg – see for example the last paragraph of the proof of Fact 2.9. By assumption (W0, s) is compound isotrivial, and so by Proposition 2.13 it is δ-locally closed over kalg. We have shown that every irreducible component of (W, s) is δ-locally closed over kalg, and it is not hard to see, by taking the union of the maximum proper D-subvarieties of these components, for example, that this implies that (W, s) is δ-locally closed, as desired. (cid:3) 2.6. D-groups over the constants. A D-group is a D-variety (G, s) over k whose underlying variety G is an algebraic group, and such that the section s : G → τ G is a morphism of algebraic groups. (Note that there is a unique algebraic group structure on τ (G) which makes the embedding ∇ : G(K) → τ G(K) a homomor- phism.) The notions of D-subgroup and homomorphism of D-groups are the natural ones, with the caveat that, unless stated otherwise, parameters may come from a / / / / D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 13 larger δ-field. The quotient of a D-group by a normal D-subgroup admits a natural D-group structure. The terms connected and connected component of identity when applied to D-groups refer just to the underlying algebraic group, though note that the connected component of identity of a D-group over k is a D-subgroup. In the context of D-groups isotriviality is better behaved. A connected D-group (G, s) is isotrivial if and only if it is isomorphic as a D-group to one of the form (H, 0) where H is an algebraic group over the constants and 0 is the zero section. So one remains in the category of D-groups, and D-birational equivalence is replaced by D-isomorphism. See the discussion around Fact 2.6 of [16] for a proof of this. In particular, every D-subvariety of an isotrivial D-group is itself isotrivial. Quotients of isotrivial D-groups are also isotrivial. Moreover, by [22, Corollary 3.10], if a D-group (G, s) has a finite normal D-subgroup H such that G/H is isotrivial, then (G, s) must have been isotrivial to start with. We also note that, as (compound) isotriviality is preserved under D-birational maps, when working inside a D-group (compound) isotriviality is preserved under translation by D-points of G (as these translations will in fact be D-automorphisms of G). The following fact is mostly a matter of putting together various results in the literature on D-groups. As we will see, it will imply that every D-subvariety of a D-group over the constants is compound isotrivial in at most 3 steps. At this point it is worth noting that the set of D-points of a D-group is a subgroup definable in DCF0 of finite Morley rank. Moreover, the ♯ functor is an equivalence between the categories of D-groups over k and finite Morley rank groups in DCF0 definable over k (see [16, Fact 2.6]). Fact 2.15. Suppose (G, s) is a connected D-group over the constants. (a) The centre Z(G) is a normal D-subgroup of G over the constants, and the quotient G/Z(G) is an isotrivial D-group. (b) Let H be the algebraic subgroup of points in Z(G) where s agrees with the zero section. Then Z(G)/H is an isotrivial D-group. Proof. For a proof that Z(G) is a D-subgroup see [16, 2.7(iii)]. That G/Z(G) is isotrivial was originally proved by Buium [6] in the centerless case, and then generalised by Kowalski and Pillay in [16, 2.10]. and H ◦ := Z ◦ ∩ H. Let Z :=(cid:0)Z ◦, s(cid:1)♯ For part (b), note first of all that H is a D-subgroup of Z(G) by definition; the zero section does map to the tangent bundle of H. Now, it suffices to show that Z ◦/H ◦ is isotrivial where Z ◦ is the connected component of identity of Z(G) (K), the subgroup of D-points of Z. Then (H ◦, s)♯(K) = Z(K δ), the δ-constant points of Z. These are now commutative δ-algebraic groups. As the ♯ functor is an equivalence of categories, isotriviality of Z ◦/H ◦ will follow once we show that Z/Z(K δ) is definably isomorphic (over some parameters) to (K δ)n for some n. Because Z ◦ is a connected commutative algebraic group over the constants, there exists a δ-algebraic group homomorphism ℓd : Z ◦ → L(Z ◦) over k0, where L(Z ◦) is the Lie algebra of Z ◦, the tangent space at the identity. This homomorphism is called the logarithmic derivative and is defined as ℓd(X) = ∇(X) · (s(X))−1 where the operations occur in T Z ◦. One can check that ℓd is surjective with kernel Z ◦(K δ) (a proof appears in [17, §3], see also [15, §V.22]). So Z/Z(K δ) is definably isomorphic to a δ-algebraic subgroup, F , of L(Z ◦). Since L(Z ◦) is a vector group, 14 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA F is a finite-dimensional K δ-vector subspace (see, for example, [21, Fact 1.3]), and hence definably isomorphic over a basis to some (K δ)n. (cid:3) Suppose (G, s) is a D-group over a δ-field k and (H, s) is a D-subgroup over k Even when H is not normal, it makes sense to consider the quotient space G/H as an algebraic variety, and s will induce on G/H the natural structure of a D- variety (G/H, ¯s) over k, in such a way that the quotient map π : G → G/H is a D-morphism. See [16, Fact 2.7(ii)], for details. Now if α is a D-point of (G/H, ¯s), then the fibre π−1(α) is a D-subvariety over k(α); and for β a D-point of this fibre we have π−1(α) = β + H. So each fibre (cid:0)π−1(α), s) is isomorphic to (H, s) over k(β). One could develop in this context the notion of "D-homogeneous spaces". Using Fact 2.15 we obtain the following highly restrictive property on the struc- ture of D-subvarieties of D-groups over the constants. Proposition 2.16. Suppose (G, s) is a connected D-group over k0 ⊆ K δ. If k is any δ-field extension of k0 and W is any irreducible D-subvariety of G over k, then W is compound isotrivial in at most 3 steps. In particular, if W is δ-rational then it is δ-locally closed. Proof. Consider the normal sequence of D-subgroups G ⊲ Z(G) ⊲ H ⊲ 0 where Z(G) is the centre of G and H is the algebraic subgroup of points in Z(G) where s agrees with the zero section. Consider the corresponding sequence of irre- ducible D-varieties and D-morphisms over k0 π0 G / G/H π1 / G/Z(G) π2 / 0 . Since G/Z(G), Z(G)/H, and H are isotrivial – the first two by Fact 2.15 and the last as sH is the zero section – this exhibits G as compound isotrivial in three steps. We can then obtain the same result for any irreducible D-subvariety of G by using the fact that any element of (G, s)♯(K) is a product of two generic elements. Alternatively we can argue as follows, keeping in mind that every D-subvariety of an isotrivial D-group is itself isotrivial. If W ⊆ G is an irreducible D-variety over k, then we get a sequence of dominant D-morphisms W f0 / W1 f1 / W2 f2 / 0 where W1 ⊆ G/H is the Zariski closure of π0(W ), and W2 ⊆ G/Z(G) is the Zariski closure of π1(W1), and the fi are the appropriate restrictions of the πi. Then W2 is isotrivial as it is a D-subvariety of G/Z(G). If α is a D-point of W2 then f −1 1 (α) is a D-subvariety of π−1 1 (α) which is isomorphic as a D-variety to Z(G)/H. So the fibres of f1 over D-points are all isotrivial D-subvarieties of W1. If β is a D-point of W1 then f −1 0 (β) which is isomorphic as a D-variety to H. So the fibres of f0 over D-points are all isotrivial. Hence W is compound isotrivial in 3 steps. 0 (β) is a D-subvariety of π−1 The "in particular" clause is by Proposition 2.13. (cid:3) We have now proved Theorem A of the introduction: Corollary 2.17. If k ⊆ K δ then every D-group over k satisfies δ-DME. / / / / / / D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 15 Proof. Suppose (G, s) is a D-group over k and W is an irreducible D-subvariety of G over kalg. Then, over kalg, it is isomorphic to an irreducible D-subvariety of the connected component of identity, G0. Applying Proposition 2.16 to G0, we have that W is compound isotrivial. So every irreducible D-subvariety of G over kalg is compound isotrivial. The δ-DME now follows from Corollary 2.14. (cid:3) 2.7. Differential Hopf algebras. In this section we give equivalent algebraic for- mulations of the δ-DME and our results so far. This will help us make the con- nection to the classical DME, which is about noncommutative associative algebras and as such does not have a direct geometric formulation. We restrict our attention to the case when δ is trivial on the base field k. As explained in §2.1, the standard geometry-algebra duality which assigns to a variety its co-ordinate ring, induces an equivalence between the category of k- irreducible affine D-varieties (V, s) and that of differential rings (R, δ) where R is an affine k-algebra and δ is a k-linear derivation. This equivalence associates to a k-irreducible D-subvariety of V a prime δ-ideal of R. Using this dictionary, we can easily translate the geometric Definition 2.1, in the case when k ⊆ K δ, into the following algebraic counterpart. Definition 2.18 (δ-DME for affine differential algebras). Suppose R is an affine k-algebra equipped with a k-linear derivation δ. We say that (R, δ) satisfies the δ-DME if for every prime δ-ideal P of R, the following conditions are equivalent: (i) P is δ-primitive: There exists a maximal ideal m of R such that P is maximal among the prime δ-ideals contained in m. (ii) P is δ-locally-closed: The intersection of all the prime δ-ideals of R that properly contain P is a proper extension of P . (iii) P is δ-rational: Frac(R/P )δ is contained in kalg. The algebraic counterpart of an affine algebraic group G over k is the commuta- tive Hopf k-algebra R = k[G], where the group law G×G → G induces a co-product ∆ : R → R ⊗k R. So what is the algebraic counterpart of a D-group (G, s) over k? The following lemma says that it is a differential Hopf k-algebra, a commutative Hopf k-algebra R equipped with a k-linear derivation δ that commutes with the coproduct, where δ acts on R ⊗k R by δ(r1 ⊗ r2) = δr1 ⊗ r2 + r1 ⊗ δr2. Lemma 2.19. Suppose k ⊆ K δ and let (G, s) be a D-variety defined over k such that G is a connected affine algebraic group. Let δ on R = k[G] be the corresponding k-linear derivation. Then s : G → T G is a group homomorphism if and only if δ commutes with the coproduct. Proof. Unravelling the fact that s induces the derivation δ on k[G] and that the group operation m : G × G → G induces the coproduct ∆ on k[G], we have that for all f ∈ k[G], (2.1) ∆(δf ) = δ∆(f ) ⇐⇒ df (s(m(y, z))) = d(f ◦ m)(s(y), s(z)) where (y, z) are coordinates for G × G. But d(f ◦ m)(s(y), s(z)) = df ◦ dm(s(y), s(z)) = df (s(y) ∗ s(z)), where ∗ denotes the group operation dm : T G × T G → T G. And so the right hand side of (2.1) is equivalent to df (s(m(y, z))) = df (s(y) ∗ s(z)). 16 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA But this, asserted for all f ∈ k[G], is equivalent to s(m(y, z)) = s(y) ∗ s(z), i.e., that s is a group homomorphism. (cid:3) In other words, the study of connected affine D-groups over the constants is the same thing as the study of affine differential Hopf k-algebras. So our Theorem A becomes: Theorem 2.20. Every commutative affine differential Hopf algebra over a field of characteristic zero satisfies δ-DME. Proof. By Lemma 2.19 our differential Hopf algebra is of the form k[G] for some connected affine D-group (G, s) with k ⊆ K δ. By Corollary 2.17 (G, s) satisfies the δ-DME. So (k[G], δ) satisfies δ-DME. (cid:3) 3. Twisting by a group-like element As it turns out, the application to the classical Dixmier-Moeglin problem that we have in mind, and that will be treated in §4, requires a generalisation of Theo- rem 2.20. Instead of working with differential Hopf algebras, we need to consider Hopf algebras equipped with derivations that do not quite commute with the co- product. Suppose R is a commutative affine Hopf k-algebra. We will use Sweedler notation1 and write ∆(r) =P r1 ⊗ r2 for any r ∈ R. Now, for a k-linear derivation δ to commute with ∆ on R means that for all r ∈ R, We wish to weaken this condition by asking instead simply that there exists some a ∈ R satisfying ∆(a) = a ⊗ a – that is, a is a group-like element of R – such that for all r ∈ R, (3.1) ∆(δr) =X δr1 ⊗ r2 + r1 ⊗ δr2. ∆(δr) =X δr1 ⊗ r2 + ar1 ⊗ δr2. That is, we ask δ to be what Panov [20] calls an a-coderivation. We wish to prove: Theorem 3.1. Suppose k is a field of characteristic zero, R is a commutative affine Hopf k-algebra, and δ is a k-linear derivation on R that is an a-coderivation for some group-like a ∈ R. Then (R, δ) satisfies the δ-DME. When a = 1 this is just the case of affine differential Hopf k-algebras, and hence is dealt with by Theorem 2.20. The general case requires some work. Throughout this section k is a fixed field of characteristic zero. Let us begin with a geometric explanation of what this twisting by a group-like element means. First of all, we have R = k[G] for some connected affine algebraic group G over k, with the coproduct ∆ on R induced by the group operation on G, and the derivation δ on R induced by a D-variety structure s : G → T G. Note that δ being a k-derivation implies that k ⊆ K δ and so τ G = T G. Now, as G is an affine algebraic group, we may assume it is an algebraic subgroup of GLn, so that T G ⊆ G × Matn. Writing s = (id, s) where s : G → Matn, we want to express as a property of s what it means for δ to be an a-coderivation. That a ∈ R is group-like means that a : G → Gm is a homomorphism of algebraic groups. 1Recall that in Sweedler notation P r1 ⊗ r2 is used to denote an expression of the form Pd j=1 rj,1 ⊗ rj,2. We will use Sweedler notation throughout, hopefully without confusion. D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 17 Lemma 3.2. Suppose G ⊆ GLn is a connected affine algebraic group over k, a : G → Gm is a homomorphism, and s = (id, s) : G → T G ⊆ G × Matn is a D-variety structure on G over k. Then the corresponding k-linear derivation δ on k[G] is an a-coderivation if and only if (3.2) s(gh) = s(g)h + a(g)gs(h) for all g, h ∈ G, where all addition and multiplication is in the sense of matrices. Proof. Note that for r ∈ k[G], ∆(δr) ∈ k[G × G] is given by ∆(δr)(g, h) = dghr(cid:0)s(gh)(cid:1) for all g, h ∈ G, where dr : T G → A2 is the differential of r : G → A1. On the other hand, writing ∆(r) =P r1 ⊗ r2 we have X(δr1 ⊗ r2 + ar1 ⊗ δr2)(g, h) = X dgr1(s(g)) r2(h) + a(g)r1(g) dhr2(s(h)) = d(g,h)(cid:16)X r1 ⊗ r2(cid:17) (s(g), a(g)s(h)) = d(g,h)(∆r)(s(g), a(g)s(h)) where the second equality uses the fact that a(g) is a scalar. Now, as an element of k[G × G], ∆(r) = r ◦ m where m : G × G → G is the restriction of matrix multiplication on GLn. Note that when we differentiate matrix multiplication we get d(g,h)m(A, B) = Ah + gB, for all g, h ∈ GLn and A, B ∈ Matn. Hence, X(δr1 ⊗ r2 + ar1 ⊗ δr2)(g, h) = d(g,h)(∆r)(s(g), a(g)s(h)) = dghr(cid:0)s(g)h + a(g)gs(h)(cid:1). = dghr ◦ d(g,h)m(s(g), a(g)s(h)) = dghr(s(g)h + ga(g)s(h)) Hence, δ being an a-coderivation, that is equation (3.1), is equivalent to for all r ∈ k[G]. But this implies dghr(cid:0)s(gh)(cid:1) = dghr(cid:0)s(g)h + a(g)gs(h)(cid:1) s(gh) = s(g)h + a(g)gs(h) as desired. (cid:3) Definition 3.3. When G is an affine algebraic group and (G, s) is a D-variety structure such that (3.2) holds, we will say that (G, s) is an a-twisted D-group. The following family of examples of 2-dimensional twisted D-groups will play an important role in the proof. Example 3.4. Let c ∈ k be a parameter. Let R = k[x, 1 x , y] with δ the k-linear derivation induced by δ(x) = xy and δ(y) = y2 2 + c(1 − x2). Note that R is the co-ordinate ring of the algebraic subgroup E ≤ GL2 made up of matrices of the form (cid:18) x y 1 (cid:19) , 0 18 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA and hence is a commutative affine Hopf k-algebra. We denote by (E, tc) the D- variety structure on E induced by δ. Writing tc = (id, tc), we have b2 2 + c(1 − a2) (cid:19) . 0 0 0 1 (cid:19) =(cid:18) ab tc(cid:18) a b 1 (cid:19)(cid:19) = a2a′b′ + aa′b 1 (cid:19)(cid:18) a′ = tc(cid:18) a b b′ 0 0 0 Now a straightforward computation shows that tc(cid:18)(cid:18) a b 1 (cid:19)(cid:18) a′ 0 0 a2(b′)2 2 + abb′ + b2 2 + c(1 − (aa′)2) 0 b′ 1 (cid:19) + a(cid:18) a b 1 (cid:19) tc(cid:18) a′ 0 0 ! 1 (cid:19) . b′ That is, (E, tc) is an x-twisted D-group. (Note that x ∈ R is group-like.) Note that since (E, tc) is not a D-group, we cannot use Theorem 2.20 to deduce the δ-DME. However, since the Krull dimension is two, (E, tc) does satisfy the δ-DME (see Proposition 2.5). Our strategy for proving Theorem 3.1 is to show that every a-twisted D-group over the constants admits the example described above as an image, with each fibre having the property that every D-subvariety is compound isotrivial. From the δ-DME for (E, tc), together with our earlier work around compound isotriviality and maximum proper D-subvarieties, we will then be able to conclude that every a-twisted D-group satisfies the δ-DME. To relate an arbitrary a-twisted D-group to one of those considered in Exam- ple 3.4, we will require the following proposition, whose proof is rather technical, and for which it would be nice to give a conceptual explanation. Proposition 3.5. Suppose R is a commutative affine Hopf k-algebra, and δ is a k-linear derivation on R that is an a-coderivation for some group-like a ∈ R. Then for some c ∈ k we have aδ2a = 3 2 (δa)2 + c(a2 − a4). We delay the proof of this proposition until we have established the preliminary Lemmas 3.6 and 3.7 below, for which we fix a commutative affine Hopf k-algebra, R, equipped with a k-linear derivation, δ, such that δ is also an a-coderivation for some group-like a ∈ R. As a is group-like it is invertible in R. We fix the following sequence of elements in R: u0 u1 := a := δa a u2 := δu1 − u2 1 2 um := δ(um−1) for m ≥ 3. Note that the desired identity aδ2a = 3 c(1 − a2); this is our eventual aim. 2 (δa)2 + c(a2 − a4) is equivalent to u2 = Lemma 3.6. For all m ≥ 1, we have ∆(um) = um ⊗ 1 + am ⊗ um + m−1Xj=2 cj,majum−j ⊗ uj +X fi ⊗ gi D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 19 where the cj,m are positive (nonzero) integers, the fi ∈ (u1, . . . , um−1)2k[u0, . . . , um−1], and the gi ∈ k[u0, . . . , um−1]. Proof. We can compute the coproducts of the elements u0, u1, . . . using the fact that a = u0 is group-like and δ is an a-coderivation: ∆(u0) = a ⊗ a ∆(u1) = ( 1 a ⊗ 1 a )(δa ⊗ a + a2 ⊗ δa) = u1 ⊗ 1 + a ⊗ u1 ∆(u2) = δu1 ⊗ 1 + δa ⊗ u1 + a2 ⊗ δu1 − u2 1 2 ⊗ 1 − au1 ⊗ u1 − a2 ⊗ u2 1 2 = u2 ⊗ 1 + a2 ⊗ u2 Then for m = 1, 2, the conclusion of the statement of the lemma follows from the above computations with fi = gi = 0 and the middle sum being empty. Now one computes ∆(um+1) = ∆(δum) for m ≥ 2, using the inductively given expression for ∆(um) and the fact that δ is an a-coderivation. The rest is a straightforward brute force computation that we leave to the reader. (cid:3) Lemma 3.7. There exist n ≥ 1, a polynomial P ∈ k[u0, . . . , un−1] and some r ≥ 0 such that un = P (u0, . . . , un−1) . ur 0 Proof. Since R is finitely generated as a k-algebra, this sequence (um) cannot be algebraically independent over k. Choose n minimal such that (u0, . . . , un) is alge- braically dependent over k. Note that if n = 0 then a = u0 is a constant and so u1 = u2 = 0 by definition. So we will assume that n > 0. So there is some d ≥ 1 such that (3.3) Ai(u0, . . . , un−1)ui n = 0, ud n +Xi<d with A0, . . . , Ad−1 rational functions over k. We may assume that d is minimal. Our first step is to show that d = 1. Since R = k[G] for some connected affine algebraic group G over k, we have that R ⊗ R is a domain. Indeed, R ⊗ R = k[G × G] and G × G is a connected affine algebraic group. We can thus work inside the fraction field of R ⊗ R. Let F be the subfield which is the fraction field of k[u0, . . . , un−1] ⊗k k[u0, . . . , un−1]. Note that by the minimality of d, {1, un, . . . , ud−1 n } is linearly independent over k(u0, . . . , un−1), from which it follows that {ui n ⊗ uj n : 0 ≤ i, j < d} is linearly independent over F . Applying ∆ to both sides of Equation (3.3), Lemma 3.6 gives us that (un ⊗ 1 + an ⊗ un)d ∈Xi<d On the other hand, ud It follows that n ⊗ 1 and and ⊗ ud F · (un ⊗ 1 + an ⊗ un)i ⊆ Xi+j<d n are also inPi+j<d F · (ui n ∈ Xi+j<d n ⊗ ud−i n ⊗ uj F · (ui n). d−1Xi=1(cid:18)d i(cid:19)an(d−i)ui F · (ui n ⊗ uj n). n ⊗ uj n) by (3.3). 20 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA If d > 1 then ud−1 n ⊗ un appears with a nonzero coefficient on the left-hand side but with zero coefficient on the right-hand side. This contradicts the F -linear independence of (ui n : 0 ≤ i, j < d). n ⊗ uj So d = 1, and we have that (3.4) un = P (u0, . . . , un−1) Q(u0, . . . , un−1) , for some relatively prime polynomials P and Q over k. We aim to show that Q is a monomial in u0. First we argue that Q⊗Q divides ∆(Q) in S := k[u0, . . . , un−1]⊗kk[u0, . . . , un−1], which note is the polynomial ring over k in the variables ui ⊗ 1, 1 ⊗ uj, and hence is a UFD. Indeed, ∆(P ) = ∆(Q)∆(un) by applying ∆ to both sides of Equation (3.4) = ∆(Q)(un ⊗ 1 + an ⊗ un + y) = ∆(Q) ((P/Q) ⊗ 1 + an ⊗ (P/Q) + y) by Lemma 3.6, for some y ∈ S We can then multiply both sides by 1⊗Q to see that ∆(Q)((P/Q)⊗Q) ∈ S. Hence, multiplying by Q⊗1, we see that Q⊗1 divides ∆(Q)(P ⊗Q) = ∆(Q)(P ⊗1)(1⊗Q). Since P and Q are relatively prime, Q ⊗ 1 divides ∆(Q). A similar argument shows that 1 ⊗ Q divides ∆(Q). Since we are working in a UFD and 1 ⊗ Q and Q ⊗ 1 are relatively prime, we see that Q ⊗ Q divides ∆(Q), as desired. Let i ≤ n − 1 be the largest index for which ui appears in Q. Then we can write i Qj(u0, u1, . . . , ui−1) with M > 0 and QM nonzero. So j=0 uj Q =PM Q ⊗ Q = (uM i ⊗ uM (uj i ⊗ uk i )(Qj ⊗ Qk) i )(QM ⊗ QM ) + Xj,k<M while, if i ≥ 1, then ∆(Q) = MXj=0 ∆(ui)jQj(∆(u0), . . . , ∆(ui−1)) = Xℓ+m≤M fℓ,m(uℓ i ⊗ um i ). where fℓ,m ∈ k[u0, . . . , ui−1]⊗kk[u0, . . . , ui−1], by Lemma 3.6. (Note that Lemma 3.6 fails for u0, so we are using that i ≥ 1 in the above calculation.) But this contra- dicts Q ⊗ Q dividing ∆(Q), since uM i appears in the former while in the latter no uℓ i appears with ℓ, m ≥ M . So it must be that i = 0 and we have shown that Q is a polynomial in u0. i ⊗ uM i ⊗ um Multiplying by a nonzero scalar if necessary, we may assume that Q is in fact a polynomial in u0 with leading coefficient 1. Let M denote the degree of Q. Then Q(u0) ⊗ Q(u0) divides ∆(Q) = Q(u0 ⊗ u0) in S, recalling that u0 = a is group-like. Since both Q(u0) ⊗ Q(u0) and Q(u0 ⊗ u0) are polynomials of total degree 2M in the variables u0 ⊗ 1 and 1 ⊗ u0 and since they both have leading coefficient 1, we see that they must be the same. In particular, Q(u0) ⊗ Q(u0) is a polynomial in u0 ⊗ u0 with leading coefficient 1, which implies that Q(u0) is of the form ur 0. (cid:3) Proof of Proposition 3.5. Let (R, δ), a, and the ui be as above. We need to show that u2 = c(1 − a2) for some c ∈ k. By Lemma 3.7 we have that there is some n ≥ 1, some r ≥ 0 and some polynomial P ∈ k[u0, . . . , un−1] such that (3.5) un = P (u0, . . . , un−1) ur 0 , D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 21 for some polynomial P over k. Our first step is to show that n ≤ 2. Let S := k[u0, . . . , un−1] ⊗k k[u0, . . . , un−1]. Let I := (u1, . . . , un−1)2k[u0, . . . , un−1] and consider the ideal of S given by J := I ⊗ k[u0, . . . , un−1] + k[u0, . . . , un−1] ⊗ I. So these are the elements of S in which each monomial has degree at least 2 in either the variables u1 ⊗ 1, . . . , un−1 ⊗ 1, or in the variables 1 ⊗ u1, . . . , 1 ⊗ un−1. Using Lemma 3.6 we can compute that for 1 ≤ i, j, ℓ ≤ n − 1 (3.6) ∆(uiujuℓ) ∈ J. Moreover, for 1 ≤ i, j ≤ n − 1, Lemma 3.6 gives (3.7) ∆(uiuj) = uj 0ui ⊗ uj + ui 0uj ⊗ ui mod J. Now, write the polynomial P of equation (3.5) as P = P0(u0) + n−1Xi=1 Pi(u0)ui + X1≤i≤j≤n−1 Pi,j (u0)uiuj + H, where H is of degree at least three in u1, . . . , un−1. Applying ∆ to both sides of equation (3.5) we get (u0 ⊗ u0)r∆(un) = ∆(P ). We therefore have: (u0 ⊗ u0)r∆(un) = P0(u0 ⊗ u0) + Pi(u0 ⊗ u0)∆(ui) + n−1Xi=1 (3.8) Xi≤j Pi,j(u0 ⊗ u0)∆(uiuj) + ∆(H). We claim that this forces Pi = 0 for all i = 1, . . . , n − 1. To prove this, note that by Lemma 3.6 and equation (3.5), both sides of equation (3.8) are elements of the polynomial ring k(u0 ⊗ 1, 1 ⊗ u0)[u1 ⊗ 1, . . . , un−1 ⊗ 1, 1 ⊗ u1, . . . , 1 ⊗ un−1]. We first compute, for both sides of (3.8), the coefficient of ui ⊗ 1. On the right-hand side, using equations (3.6) and (3.7), the only term that contributes is Pi(u0 ⊗ u0)∆(ui). By Lemma 3.6, that contribution is Pi(u0 ⊗ u0). On the left-hand side, using Lemma 3.6 and equation (3.5), the coefficient of ui ⊗ 1 is (u0 ⊗ u0)r(cid:18) Pi(u0) ⊗ 1(cid:19). 0 = Pi(u0 ⊗ u0). This forces Pi = dur So Pi(u0) ⊗ ur 0 for some d ∈ k. On the other hand, comparing the coefficient of 1 ⊗ ui on both sides of equation (3.8) we have that ur+n 0 we get that d(ur+n 0 ⊗ Pi(u0) = Pi(u0 ⊗ u0)(ui 0) = d(ur+i 0). As i < n, this forces d = 0 and hence Pi = 0. 0 ⊗ 1). Plugging in Pi = dur u−r 0 0 ⊗ ur Equation (3.8) therefore becomes 0 ⊗ ur (u0⊗u0)r∆(un) = P0(u0⊗u0)+ X1≤i≤j≤n−1 Pi,j(u0⊗u0)(uj 0ui⊗uj+ui 0uj⊗ui) mod J. Assume towards a contradiction that n ≥ 3. Then by Lemma 3.6 we must have u1 ⊗ un−1 appearing in ∆(un) on the left with a nonzero coefficient. So P1,n−1 6= 0. But then P1,n−1(u0 ⊗ u0)(u0un−1 ⊗ u1) appears on the right, while it does not appear on the left since un−1 ⊗ u1 does not appear in ∆(un) by Lemma 3.6. 22 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA This contradiction proves that n ≤ 2. Suppose n = 1. Then equation (3.5) says u1 = P (u0) ur 0 . Applying ∆ to both sides yields P (u0) ⊗ ur 0 + ur+1 0 ⊗ P (u0) = P (u0 ⊗ u0) which is only possible if P0 = 0. Hence u1 = u2 = 0, as desired. So we are left to consider the case when n = 2. Equation (3.5) becomes u2 = 1 ur 0 MXj=0 Pj(u0)uj 1 with M ≥ 1, the Pj are polynomials over k, and PM is nonzero. Multiplying by ar (recall that u0 = a) and applying ∆ gives (ar ⊗ ar)(cid:0)u2 ⊗ 1 + a2 ⊗ u2(cid:1) = which we can write as Pj(a ⊗ a)(u1 ⊗ 1 + a ⊗ u1)j  Pj (a)uj 1 ⊗ ar + MXj=0 MXj=0 ar+2 ⊗ Pj (a)uj Pj(a ⊗ a)(u1 ⊗ 1 + a ⊗ u1)j. MXj=0 Notice that if M > 1 then the right-hand side involves terms with ui 1 with i, j ≥ 1 while the left-hand side does not, and so we cannot have equality. Thus M = 1. Writing out the above equation with this in mind we get that P0(a) ⊗ ar + P1(a)u1 ⊗ ar + ar+2 ⊗ P0(a) + ar+2 ⊗ P1(a)u1 1 ⊗ uj is equal to P0(a ⊗ a) + P1(a ⊗ a)(u1 ⊗ 1 + a ⊗ u1). We look at this as an equation in k[a ⊗ 1, 1 ⊗ a][u1 ⊗ 1, 1 ⊗ u1]. Then taking the coefficient of u1 ⊗ 1 gives that P1(a) ⊗ ar = P1(a ⊗ a), which can only occur if P1 = dar for some d ∈ k. Then computing the coefficient of 1 ⊗ u1 gives ar+2 ⊗ dar = d(ar+1 ⊗ ar), and so d = 0. Hence P1 = 0. Now taking the constant coefficient (regarding constants as being in k[a ⊗ 1, 1 ⊗ a]) gives that Now write P0(t) =PL LXj=0 P0(a) ⊗ ar + ar+2 ⊗ P0(a) = P0(a ⊗ a). j=0 pjtj. Then we have pj(aj ⊗ ar + ar+2 ⊗ aj − aj ⊗ aj) = 0. Notice that if j 6∈ {r, r + 2} then we have that the coefficient of aj ⊗ aj on the left-hand side is equal to pj whereas the right-hand side is zero and so pj = 0. It follows that P0(t) = prtr + pr+2tr+2. Then 0 = pj(aj ⊗ ar + ar+2 ⊗ aj − aj ⊗ aj) = prar+2 ⊗ ar + pr+2ar+2 ⊗ ar. MXj=0 1 = LXj=0 This forces pr = pr+2 and so we see P0(t) = c(tr − tr+2) for some constant c ∈ k. (cid:3) 0) = c(1 − a2), as desired. So u2 = 1 ur 0(cid:0)P0(u0) + P1(u0)u1(cid:1) = c(1 − u2 D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 23 The following is a geometric interpretation of the proposition. Proposition 3.8. Suppose k ⊆ K δ and (G, s) is an affine connected a-twisted D-group over k where a ∈ k[G] is group-like. Then g 7−→ a(g) 0 δa(g) a(g) 1 ! defines a homomorphism π : G → E where E ≤ GL2 is the algebraic subgroup made up of matrices of the form(cid:18) x y 1 (cid:19). Moreover, there exists some c ∈ k such that if (E, tc) is the a-twisted D-group from Example 3.4, then π : (G, s) → (E, tc) is a D-morphism. 0 Proof. Recall that as a ∈ k[G] is group-like, a : G → Gm is a homomorphism of algebraic groups. It follows immediately that π is well-defined and does indeed map G to E. We check that it is a group homomorphism: given g, h ∈ G, note first of all that as ∆(a) = a ⊗ a and δ is an a-coderivation we have ∆(δa) = δa ⊗ a + a2 ⊗ δa and so (3.9) δa(gh) = ∆(δa)(g, h) = δa(g)a(h) + a(g)2δa(h). We can therefore compute π(g)π(h) = a(g) 0 δa(g) a(g) 1 ! a(h) 0 a(h) + δa(g) a(g) a(g)2δa(h)+δa(g)a(h) 1 0 = a(gh) a(g) δa(h) = a(gh) = a(gh) 1 ! δa(gh) a(gh) 0 0 a(gh) 1 δa(h) a(h) 1 ! ! ! by (3.9) = π(gh) where we have used repeatedly that a(gh) = a(g)a(h). We note that we have not up until this point used the parameter c ∈ k; the reason for this is that the groups Ec are isomorphic as algebraic groups. It remains to show that π is a D-morphism from (G, s) to some (E, tc). Let c be as given by Proposition 3.5. It suffices to show that π takes D-points to D-points. That is, if g ∈ (G, s)♯(K) then a(g) 0 δa(g) a(g) 1 ! should be a D-point of (E, tc). 24 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA Writing tc = (id, tc) we have that tc a(g) 0 δa(g) a(g) 0 1 ! = δa(g) = δa(g) = δ a(g) δa(g)2 2a(g)2 + c(1 − a(g)2) 0 ! by Example 3.4 + c(1 − a(g)2) a(g)δ2a(g)−δa(g)2 −c(a(g)2−a(g)4) a(g)2 0 δa(g) a(g) 1 ! ! 0 0 where the penultimate step follows from Proposition 3.5 telling us that aδ2a = 3 2 (δa)2 + c(a2 − a4), and in the final equality we are using the fact that as g is a D- point of G, δ(r(g)) = (δr)(g) for all r ∈ k[G]. This shows that π(g) ∈ (E, tc)♯(K), as desired. (cid:3) We can now complete the proof of the theorem. Proof of Theorem 3.1. We have already established that (R, δ) is the co-ordinate ring of an affine connected a-twisted D-group (G, s) where a ∈ k[G] is group- like. Here recall that k ⊆ K δ. By Proposition 2.3, it suffices to show that every irreducible δ-rational D-subvariety of G over k is δ-locally-closed. Let π : (G, s) → (E, tc) be the D-morphism from Proposition 3.8. We first show that every fibre of this map has the property that all its D-subvarieties, over arbitrary δ-field extensions, are compound isotrivial. Let us start with the fibre above the identity, that is, H = ker(π). Since π is a D-morphism, (H, s) is a D-subvariety of (G, s). Here, by abuse of notation, we write (H, s) instead of (H, s ↾H). Since π is an algebraic group homomorphism H is an algebraic subgroup of G. It follows that (H, s) is an a↾H -twisted D-group also. On the other hand, a↾H = 1 by the definition of π. So (H, s) is an actual D-group. By Proposition 2.16, every irreducible D-subvariety of H, over any δ-field extension of k, is compound isotrivial. What about other fibres of π over D-points of (E, tc)? Any such fibre is a D- subvariety of (G, s) of the form Hg, for some g ∈ (G, s)♯(K). Since (G, s) is not necessarily a D-group, the multiplication-by-g-on-the-right map, ρg : G → G, is not necessarily a D-automorphism. Nevertheless, when we restrict this map to H we do get a D-isomorphism between H and Hg. To see this we need only check that ρg takes D-points of H to D-points of Hg. Letting h ∈ (H, s)♯(K) we compute s(hg) = s(h)g + a(h)hs(g) = s(h)g + hs(g) = δ(h)g + hδ(g) by (3.2) as a↾H = 1 as h and g are D-points = δ(hg) as ∇ : G → T G is a group homomorphism as desired. So H and Hg are D-isomorphic over k(g). It follows that every fibre of π above a D-point has the property that all its D-subvarieties, over arbitrary δ-field extensions, are compound isotrivial. Now suppose that V ⊆ G is an irreducible δ-rational D-subvariety over k. We need to prove that it has a maximum proper D-subvariety over k. Let W ⊆ E be the D-subvariety obtained by taking the Zariski closure of the image of V under π, and consider the dominant D-morphism π↾V : (V, s) → (W, tc). Since k(W ) ⊆ k(V ), D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 25 W is also δ-rational. Since (E, tc) is of dimension two, it satisfies the δ-DME by Proposition 2.5. Hence W has a maximum proper D-subvariety over k. Next, let η be a k-generic D-point of W and consider the fibre Vη. Note that Vη is δ-rational since k(η)(Vη) = k(V ). But Vη is a D-subvariety of the fibre of π : (G, s) → (E, tc) above the D-point η, and hence as we have argued above, is compound isotrivial. So, by Proposition 2.13, Vη has a maximum proper D-suvariety over k(η). We have shown that both the image and the generic fibre have maximum proper D- subvarieties, and so by Lemma 2.6, (V, s) has a maximum proper D-subvariety over k, as desired. (cid:3) Remark 3.9. In the end of above proof we could also have used the fact that (E, tc), while not in general isotrivial, is compound isotrivial in two steps. This was observed by Ruizhang Jin, in whose PhD thesis this example will be worked out. In any case, using the compound isotriviality of (E, tc) the above arguments actually give that every D-subvariety of (G, s) is compound isotrivial (in at most five steps) from which it follows by Corollary 2.14 that (G, s) satisfies the δ-DME. 4. The DME for Ore extensions of commutative Hopf agebras We will now apply the results of the previous sections to the classical study of certain (noncommutative) Hopf agebras. Recall that if A is a noetherian associative algebra over a field k of characteristic zero, then we say that the Dixmier-Moeglin equivalence (DME) holds for A if for every (two-sided) prime ideal P of A, the following are equivalent: (i) P is primitive: it is the annhilator of a simple left A-module. (ii) P is locally closed: the intersection of all the prime ideals of A that properly contain P is a proper extension of P . (iii) P is rational: the centre of the Goldie quotient ring2 Frac(A/P ) is an algebraic field extension of k. Of course, for commutative algebras the DME always holds as the notions of prim- itive, locally closed and rational all coincide with maximal. It is known that in any algebra that satisfies the Nullstellensatz locally closed implies primitive and primitive implies rational, see [4, II.7.16]. Thus, the central question is: when does rational imply locally closed? Certainly this is not always the case; even in finite Gelfand-Kirillov dimension a counterexample was found in [1]. In [2] the DME was conjectured specifically about all Hopf algebras of finite Gelfand-Kirillov dimension. We will show here that the DME holds for Hopf algebras that arise as certain twisted polynomial rings over commutative Hopf algebras. Recall that if R is a k- algebra equipped with an automorphism σ then a k-linear σ-derivation is a k-linear map δ satisfying the twisted Leibniz rule: δ(rs) = σ(r)δ(s) + δ(r)s. Given σ and δ, the Ore extension of R, denoted by R[x; σ, δ] is the ring extension of R with the property that it is a free left R-module with basis {xn : n ≥ 0} and such 2The Goldie quotient is an artinian ring of quotients for any prime noetherian ring that imitates the field of fractions construction for integral domains in the commutative case. See [18, Chapter 2] for details. 26 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA that xr = σ(r)x + δ(r) for all r ∈ R. We aim to prove the DME for Hopf algebras that arise as the Ore extensions of commutative Hopf algebras. More precisely, Theorem 4.1. Suppose k is a field of characteristic zero and R is a commutative affine Hopf k-algebra equipped with a k-algebra automorphism σ and a k-linear σ- derivation δ. Assume that the Ore extension A := R[x; σ, δ] admits a Hopf algebra structure extending that of R. Then A satisfies the DME. This is Theorem B2 of the introduction. preliminaries. Its proof is preceded by a number of 4.1. Hopf Ore extensions. In this section we prove a result (Corollary 4.4, below) that severely restricts what (R, ∆, σ, δ) can be if A = R[x; σ, δ] is to admit a Hopf algebra structure extending that on R. Actually this was already done by Brown et al. in [5], answering a question of Panov [20], in a more general context where R is not necessarily commutative, but under the additional assumption on A that (4.1) ∆(x) = a ⊗ x + x ⊗ b + v(x ⊗ x) + w for some a, b ∈ R and v, w ∈ R ⊗k R. When (4.1) holds, possibly after a change of the variable x, Brown et al. call R[x; σ, δ] a Hopf Ore extension. They ask if every Ore extension admitting a Hopf algebra structure extending that on R is a Hopf Ore extension. We prove that this is the case, in a strong way, when R is commutative and affine (this is Theorem C of the introduction): Theorem 4.2. Suppose k is an algebraically closed field of characteristic zero and R is a commutative affine Hopf k-algebra equipped with a k-algebra automorphism σ and a k-linear σ-derivation δ. If R[x; σ, δ] admits a Hopf algebra structure extending that of R then, after a linear change of the variable x, ∆(x) = a ⊗ x + x ⊗ b + w for some a, b ∈ R, each of which is either 0 or group-like, and some w ∈ R ⊗k R. Proof. Our starting point is [5, §2.2, Lemma 1] which says that if R ⊗k R is a domain (which is true here as R is a commutative domain and k is algebraically closed) then (4.2) ∆(x) = s(1 ⊗ x) + t(x ⊗ 1) + v(x ⊗ x) + w where s, t, v, w ∈ R ⊗k R. We let A = R[x; σ, δ] and let S denote the antipode of A. By making a substitution x 7→ x − λ for some λ ∈ k, we may assume that ǫ(x) = 0 where ǫ : A → k is the counit. This substitution does not change the form of ∆(x) given in (4.2). Our first goal is to show that v = 0. Recall that since R is commutative it is of the form R = k[G] for a connected affine algebraic group G. We can therefore view v ∈ R ⊗k R as a regular function on G × G. We show first that v(g−1, g) = 0 for all g ∈ G, and then that in fact v = 0. We now consider the antipode S. By [26, Corollary 1], S is bijective on A and its restriction to R is bijective on R. Thus we can write S(x) = a0 + a1x + · · · + adxd for some d ≥ 1 and a0, . . . , ad ∈ R with ad 6= 0. Writing m : A ⊗k A → A for the homomorphism induced by multiplication, we have the identity m ◦ (S ⊗ id) ◦ ∆(x) = ǫ(x). D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 27 So, as ǫ(x) = 0, we may let µ = m ◦ (S ⊗ id) and use (4.2) to write 0 = µ(s)x + S(x)µ(t) + S(x)µ(v)x + µ(w). Notice that m ◦ (S ⊗ id)(R ⊗ R) ⊆ R and so if we look at the coefficient of xd+1 on the right-hand side, we see that it is adσd(µ(v)). Since R ⊗ R is a domain and ad is nonzero and σ is an automorphism, we see that µ(v) = m ◦ (S ⊗ id)(v) = 0. Geometrically, this means precisely that v(g−1, g) = 0 for all g ∈ G. Next we apply coassociativity, which tells us that (∆⊗id)(∆(x)) = (id ⊗∆)(∆(x)) in R ⊗k R ⊗k R = k[G × G × G]. Writing this out using (4.2), and equating the coefficients of x ⊗ x ⊗ x, yields (∆ ⊗ id)(v) · (v ⊗ 1) = (id ⊗∆)(v) · (1 ⊗ v). Evaluating at (g, h−1, h) for any fixed g, h ∈ G we get v(gh−1, h)v(g, h−1) = v(g, 1G)v(h−1, h) = 0 where the final equality uses what we proved in the previous paragraph. Now, if v 6= 0 then for a Zariski dense set of (g, h) ∈ G × G, v(g, h−1) 6= 0. But then for each such (g, h) the above equation implies that v(gh−1, h) = 0. Hence, in fact, v(gh−1, h) = 0 for all (g, h) ∈ G × G. As every element of G × G can be written in the form (gh−1, h), we have shown that v = 0. We have thus proven that (4.3) ∆(x) = s(1 ⊗ x) + t(x ⊗ 1) + w for some s, t, w ∈ R ⊗k R. We claim now that either t = 0 or t = 1 ⊗ b for some group-like b ∈ R. We again apply coassociativity to x, this time using (4.3) and equating the coefficients of x ⊗ 1 ⊗ 1, to get (∆ ⊗ id)(t) · (t ⊗ 1) = (id ⊗∆)(t) The geometric interpretation is that (4.4) t(f g, h)t(f, g) = t(f, gh) for all f, g, h ∈ G. Suppose t(1G, g0) = 0 for some g0 ∈ G. We show in this case that t = 0. Indeed, for all h ∈ G we have 0 = t(g0, h)t(1G, g0) = t(1G, g0h), by (4.4) with f = 1G. Hence, t(1G, h) = 0 for all h ∈ G. But then, by (4.4) with f = g−1, we get 0 = t(1G, h) = t(g−1g, h)t(g−1, g) = t(g−1, gh) for all g, h ∈ G. As every element of G × G is of the form (g−1, gh) for some g, h ∈ G, we have t = 0, as desired. Suppose on the contrary that t(1G, g) 6= 0 for every g ∈ G. Then t(g, h) = is a never vanishing regular function on G × G, and hence t = λt′ where : G × G → Gm is an algebraic group homomorphism (see [23, t(1G,gh) t(1G,g) λ ∈ k∗ and t′ Theorem 3]). So t′ = b′ ⊗ b where b′, b ∈ R are group-like. But then we have λb′(g)b(h) = t(g, h) = t(1G, gh) t(1G, g) = λb(gh) λb(g) = λb(g)b(h) λb(g) for all g, h ∈ G. It follows that b′ = λ = 1 and t = 1 ⊗ b, as desired. A similar argument shows that in (4.3) either s = 0 or s = a ⊗ 1 for some (cid:3) group-like a ∈ R. This proves the theorem. 28 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA Remark 4.3. It may be worth pointing out that our proof of Theorem 4.2 made no use of δ. We used only the properties of a Hopf algebra extension and the fact that σ is injective, as well as the fact that every element of A can be written as a left polynomial in x over R. Corollary 4.4. Suppose k is an algebraically closed field of characteristic zero and R is a commutative affine Hopf k-algebra equipped with a k-algebra automorphism σ and a k-linear σ-derivation δ. If R[x; σ, δ] admits a Hopf algebra structure extending that of R then (R, ∆, σ, δ) must satisfy the following two conditions. (1) There exists w ∈ R ⊗k R and a group-like a ∈ R such that, for all r ∈ R, ∆(δ(r)) =X (δ(r1) ⊗ r2 + ar1 ⊗ δ(r2)) + w(cid:0)∆(r) − ∆(σ(r))(cid:1) (2) There is a character χ : R → k such that for all r ∈ R, σ(r) =X χ(r1)r2 =X r1χ(r2). In the above we are using Sweedler notation, writing ∆(r) =P r1 ⊗r2 for all r ∈ R. Proof. Statements (1) and (2) are proven for Hopf Ore extensions in [5]. Indeed, remembering that in our case R is commutative, statement (2) is just part (i)(c) of the main theorem of [5] (see also Theorem 2.4(d) of [5]), and statement (1) is the identity labelled (21) in [5] which is asserted in part (i)(d) of the main theorem. So to prove the corollary it suffices to show that A = R[x; σ, δ] is a Hopf Ore extension, that is, after a change of variable ∆(x) has the form (4.1) discussed above. But Theorem 4.2 gives us an even stronger form for ∆(x). (cid:3) Remark 4.5. The main theorem of [5] also includes a converse; namely, assuming that (R, ∆, σ, δ) satisfies (1) and (2), with w ∈ R⊗k R satisfying two other identities, one can always extend in a natural way the Hopf algebra structure from R to R[x; σ, δ]. This gives many examples to which our Theorem 4.1 will apply. 4.2. The case when σ is the identity. When σ = id note that a σ-derivation is just a derivation. In this case we write the Ore extension as R[x; δ]; it is the skew polynomial ring in x over R where xr = rx + δ(r) for all r ∈ R. Statement (1) of Corollary 4.4 now says that if R[x; δ] admits a Hopf algebra structure extending that on R, then δ must have been an a-coderivation on R. So Theorem 3.1 applies and we have that (R, δ) satisifes the δ-DME. The following proposition relates the δ-DME for (R, δ) to the DME for R[x; δ]. Proposition 4.6. Suppose k is a field of characteristic zero and R is a commutative affine k-algebra equipped with a k-linear derivation δ. If the δ-rational prime δ- ideals of (R, δ) are δ-locally-closed, then the rational prime ideals of R[x; δ] are locally closed. In particular if the δ-DME holds for (R, δ) then the DME holds for R[x; δ]. Proof. Suppose P is a rational prime ideal of R[x; δ]. Let I := P ∩ R. Then I is a prime ideal of R (see [7, Corollary to Lemma 2]). Moreover, I is a δ-ideal since if a ∈ P ∩ R then δ(a) = [x, a] ∈ P ∩ R. It follows easily that J := IR[x; δ] is an ideal of R[x; δ] that is contained in P and that R[x; δ]/J ∼= (R/I)[x; δ] where we use δ to denote the induced derivation on S := R/I. Let F = Frac(S) be the field of fractions of S and extend δ to F . We claim that the δ-constants of D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 29 F are all algebraic over k. Indeed, note that if f ∈ F with δ(f ) = 0 then f is a central element of F [x; δ]. We let P denote the prime ideal in S[x; δ] corresponding to P under the isomorphism R[x; δ]/J ∼= S[x; δ]. As P ∩ R = I, we have that P ∩ S = 0, so that P lifts to a prime ideal P0 of F [x; δ]. The image of f in B := F [x; δ]/P0 is again a central element of B. By construction B is a localization of S[x; δ]/ P ∼= R[x; δ]/P and thus passing to the full localization gives that f is a central element of Frac(R[x; δ]/P ). As P is rational f must be algebraic over k. We have shown that the prime δ-ideal I is δ-rational. By assumption it is therefore δ-locally-closed. Consequently, there is some g ∈ R \ I such that every prime δ-ideal of R properly containing I must contain g. Q ) P is prime then Q ∩ R ) P ∩ R = I. In order to prove that P is locally closed it now suffices to show that whenever Indeed, if this is the case, then we have that g ∈\{Q ∩ R : Q ) P prime}. Since g /∈ P , we have in particular that \{Q : Q ) P prime} 6= P . That is, P is locally closed. Towards a contradiction therefore, let us assume that there exists a prime ideal Q ) P with Q ∩ R = P ∩ R = I. It follows that F [x; δ] is not simple: under the isomorphism R[x; δ]/J ∼= S[x; δ], Q corresponds to a nonzero prime ideal Q in S[x; δ] whose intersection with S = R/I is trivial, so that Q lifts to a nonzero prime ideal Q0 in F [x; δ]. On the other hand, it is well-known that, as F is a field of characteristic zero, if δ is nontrivial on F then F [x; δ] is a simple ring (indeed this is a consequence of the fact that F [x; δ] is a left and right PID, see [25, §2.1]). Thus, δ is trivial on F and so F [x; δ] = F [x] is a PID. So P0, the lift of P from S[x; δ] to F [x; δ], as it is properly contained in Q0, must be 0. That is, F [x; δ] rationality of P . is a localisation of R[x; δ]/P . Hence, Frac(cid:0)R[x; δ]/P(cid:1) = F (x), contradicting the For the "in particular" clause, note that R[x; δ] satisfies the Nullstellensatz – by [12, Theorem 2] for example – and hence we already know that local-closedness implies primitivity and primitivity implies rationality. (cid:3) Corollary 4.7. Suppose k is an algebraically closed field of characteristic zero and R is a commutative affine Hopf k-algebra equipped with a k-linear derivation δ that is also an a-coderivation for some group-like a ∈ R. Then R[x; δ] satisfies the DME. Proof. Theorem 3.1 together with Proposition 4.6. (cid:3) A special case of Corollary 4.7 is when R is a differential Hopf k-algebra – this yields Theorem B1 of the introduction. But the DME for R[x; δ] in that case is easier: one uses only Theorem 2.20 and the material in Section 3 is not necessary. 4.3. The case when δ is inner. If σ is an automorphism of R, and a ∈ R, then the map r 7→ a(r − σ(r)) is a σ-derivation on R. Such σ-derivations are called inner. Here is a sufficient criterion for a σ-derivation δ being inner.3 Lemma 4.8. Suppose R is a commutative ring with an automorphism σ and a σ-derivation δ. Suppose there exists an element f ∈ R such that f − σ(f ) is a unit. Then δ is inner. Proof. It is easy to see, using the commutativity of R, that a := δ(f ) the inner-ness of (R, σ, δ). f −σ(f ) witnesses (cid:3) 3For a more general statement in the noncommutative case, see [9, Lemma 2.4(b)]. 30 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA When δ is inner the Dixmier-Moeglin equivalence for R[x; σ, δ] follows easily from known results. It makes use however of one more notion: Definition 4.9. Let A be a finitely generated algebra over a field k. We say that a k-vector subspace V of A is a frame for A if V is finite-dimensional, contains 1A, and generates A as a k-algebra. Lemma 4.10. Suppose R is a commutative affine Hopf algebra over a field k of characteristic zero and σ is a k-algebra automorphism of R satisfying statement (2) of Corollary 4.4. Then there there is a frame for R such that σ(V ) = V . Proof. Suppose R = k[G] where G is an affine algebraic group. Then G is linear and hence we may embed G into GLn. This gives us a frame V of R spanned by the restriction to G of 1, the coordinate functions xi,j , and 1 xk,j and ∆( 1 then implies that σ(V ) ⊆ V , and hence by finite-dimensionality σ(V ) = V .4 det . So ∆(V ) ⊆ V ⊗ V . Statement (2) of Corollary 4.4 (cid:3) det . Now, ∆(xi,j ) =P xi,k⊗ det ) = 1 det ⊗ 1 Proposition 4.11. Suppose k is an uncountable algebraically closed field of char- acteristic zero, R is a finitely generated commutative k-algebra, σ is a k-algebra automorphism of R that preserves a frame, and δ is an inner σ-derivation on R. Then R[x; σ, δ] satisfies the DME. Proof. When δ = 0 this is [3, Theorem 1.6]; while the theorem there is stated for k = C it holds for any uncountable algebraically closed field. But if a ∈ R is such that δ(r) = a(r − σ(r)) for all r ∈ R, then R[x; σ, δ] = R[t; σ, 0] where t := x − a. Indeed, tr = (x − a)r = σ(r)x + δ(r) − ar = σ(r)x − aσ(r) = σ(r)t for all r ∈ R, and {tn : n ≥ 0} can be seen to be another left R-basis for R[x; σ, δ] using the fact that, for any polynomial P , P (t) is equal to P (x) plus terms of strictly lower degree. So the inner case reduces to the case when δ = 0. (cid:3) 4.4. The general case. Let us fix from now on a field k of characteristic zero. Our proof of Theorem 4.1 will go via reducing either to the case when σ = id or when δ is inner. But it will require some preparatory lemmas. First, let us point out that statement (2) of Corollary 4.4 forces (R, σ) to be of a very restricted form. Lemma 4.12. Let G be a connected affine algebraic group over k and τ : G → G an automorphism of G over k. Let R = k[G], and σ = τ ∗ the corresponding k- algebra automorphism of R. If (R, σ) satisfies statement (2) of Corollary 4.4 then τ : G → G is translation by some central element of G(k). Proof. Since χ : R → k is a homomorphism there is some c ∈ G(k) such that χ(f ) = f (c) for all f ∈ R. If we write ∆(f ) =P f1⊗f2, then by the definition of the coproduct on R we have f (ab) =P f1(a)f2(b) for all a, b ∈ G. Now, property (2) 4As a referee pointed out to us, the existence of a frame V with ∆(V ) ⊆ V ⊗ V can be deduced for arbitrary finitely generated Hopf algebras by starting with any frame W and extending it to a finite-dimensional subcoalgebra V by the Finiteness Theorem for Coalgebras [19, Theorem 5.1.1]. D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 31 gives us that for all a ∈ G, σ(f )(a) =P χ(f1)f2(a) =P f1(c)f2(a) = f (ca). The other half of the equality in (2) gives σ(f )(a) = f (ac). So f (ca) = f (ac) for all f ∈ R, and hence c is central in G. On the other hand, f (ca) = σ(f )(a) = f (τ a) for all f ∈ R, so τ is translation by c. (cid:3) We will make use of the following notion. Definition 4.13. Suppose σ is an automorphism of a commutative ring R. By a σ-prime ideal is meant a σ-ideal I such that whenever J and K are σ-ideals with JK ⊆ I then either J ⊆ I or K ⊆ I. Note that a σ-prime ideal need not be prime. But, at least in the case when R is a commutative noetherian ring, a σ-prime ideal is radical; this follows from the fact that the nilpotent radical of I is a σ-ideal and some power of it is contained in I. We will sometimes need to quotient out by σ-prime ideals that we do not know are prime, which means we will have to work with reduced difference rings that are not necessarily integral domains. The following lemma about such difference rings will be very useful. Lemma 4.14. Suppose R is a commutative ring endowed with an automorphism σ such that (0) is σ-prime. If 0 6= f ∈ R satisfies σ(f ) ∈ Rf then f is not a zero divisor in R. Proof. Let J = Rf . Then J is a σ-ideal of R. It follows that K := {r ∈ R : rf = 0} is also a σ-ideal of R. Then by construction JK = (0). Since (0) is σ-prime and J is nonzero, we see that K = (0) and so we obtain the desired result. (cid:3) Finally, we will make use of the following fundamental result on Ore extensions of commutative noetherian rings. Fact 4.15 (Goodearl [9]). Suppose R is a commutative noetherian ring, σ is an automorphism of R, and δ is a σ-derivation. Suppose P is a prime ideal of the Ore extension R[x; σ, δ], and let I = P ∩ R. Then one of the following three statements must hold: I. R[x; σ, δ]/P is commutative. II. I is a (σ, δ)-ideal of R – that is, I is preserved under σ and δ – and there is a prime ideal I ′ of R containing I such that σ(r) − r ∈ I ′ for all r ∈ R. III. I is a σ-prime (σ, δ)-ideal of R and IR[x; σ, δ] is a prime ideal of R[x; σ, δ]. We can now prove the theorem. Proof of Theorem 4.1. We have that R is a commutative affine Hopf k-algebra equipped with a k-algebra automorphism σ and a k-linear σ-derivation δ, and that the Ore extension A := R[x; σ, δ] admits a Hopf algebra structure extending that of R. We wish to show that A satisfies the DME. By [12, Theorem 2] we have that A satisfies the Nullstellensatz, and so it suffices to prove that if P is a rational prime ideal of A then P is locally closed. We first reduce to the case when k is algebraically closed. Since R = k[G] where G is an affine algebraic group, and hence smooth, R is integrally closed. Let F denote the field of fractions of R and let F0 := kalg ∩ F . Since R is integrally closed, F0 ⊆ R. Since F is a finitely generated extension of k, F0 is a finite extension of k. Let R′ := R ⊗F0 kalg. Since F0 is relatively algebraically closed in F , we see that R′ is again an integral domain. Thus R′ is a commutative affine Hopf kalg-algebra to 32 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA which we extend σ and δ by kalg-linearity. Suppose we have proven the DME for R′[x; σ, δ]. Then Irving-Small reduction techniques (see Irving-Small [13] and also Rowen [24, Theorem 8.4.27]) give that A = R[x; σ, δ] satisfies the DME over F0. But since F0 is a finite extension of k, we get the DME over k also. Next we reduce to the case when k is uncountable (in order to be able to use Proposition 4.11). Let L be an uncountable algebraically closed extension of k. Then since k is algebraically closed we see that R ⊗k L is a commutative affine Hopf L-algebra to which we extend σ and δ by L-linearity, and B := A ⊗k L ∼= (R ⊗k L)[x; σ, δ]. Assume the DME holds for B. Let P be a rational prime ideal of A and let Q = P ⊗k L. Since k is algebraically closed, Q is a prime ideal of B. Since P is rational and B/Q = (A/P ) ⊗k L, we see that Q is a rational. Hence Q is locally closed. Since the primes in A containing P lift to primes in B containing Q, it follows that P is locally closed in A. So A satisfies the DME. We may therefore assume that k is uncountable and algebraically closed. If σ = id then δ is a k-linear derivation on R and statement (1) of Corollary 4.4 tells us that it is also an a-coderivation on for some group-like a ∈ R. It follows by Corollary 4.7 that A = R[x; δ] satisfies the DME. So we may assume σ 6= id. We may also assume that A/P is not commutative. Indeed, if it were, as P is rational, we would have that Frac(A/P ) ⊆ k, so that P is a maximal ideal and hence locally closed. Write R = k[G] where G is a connected affine algebraic group over k. By Lemma 4.12 we know that σ = τ ∗ where τ : G → G is translation by a central (non-identity) element c ∈ G(k). Our next goal is to reduce to the case that P ∩R = (0), though in order to obtain this we will have to give up on R being an integral domain. Let I = R∩P . We have already ruled out case (I) of Fact 4.15. On the other hand, case (II) cannot hold: σ would induce the identity map on R/I ′, implying that τ is the identity on V (I ′), which contradicts the fact that it is translation on G by a non-identity element. Hence case (III) holds; I is a σ-prime (σ, δ)-ideal of R and J := IA is a prime ideal of A. Consider now the reduced quotient ring R := R/I with the induced automorphism which we continue to denote by σ, and the induced σ-derivation which we continue to denote by δ. Let A = A/J ∼= R[x; σ, δ] and P the image of P in A. Since J is contained in P , P is rational in A and it suffices to show that P is locally closed in A. Note that we have achieved P ∩ R = (0). Next, we claim that there is some non zero-divisor f ∈ R such that (σ, δ) extends to eR := R[1/f ] and δ is inner on eR. To see this, consider the frame V for R, given by Lemma 4.10, that is preserved by σ. The image V of V in R is then a frame for R that is also preserved by σ. Using k = kalg, let f be an eigenvector for the action of σ on V , say σ(f ) = λf for some λ ∈ k∗. By Lemma 4.14, f is not a zero divisor. Moreover, the multiplicatively closed subset {1, f, f 2, . . .} of R is preserved by σ, and hence by [9, Lemma 1.3], (σ, δ) extends uniquely to the localisation at this set, namely to eR := R[1/f ]. It remains to show that f can be chosen so that δ is inner on eR. If the eigenvalue λ is not equal to 1, then f − σ(f ) = (1 − λ)f is a unit in eR, and we get δ inner by Lemma 4.8. So suppose that 1 is the only eigenvalue for σ on V . Note that σ is not the identity operator on R because τ is not the identity on V (I). Since V generates R as a k-algebra, σ is not the identity on V either. Hence there must be some Jordan block that is of size greater than one, but with eigenvalue 1. So we can choose the eigenvector f in such a way that there exists D-GROUPS AND THE DIXMIER-MOEGLIN EQUIVALENCE 33 nonzero g ∈ V with σ(g) = g + f . Hence g − σ(g) is a unit in eR = R[1/f ], and so by Lemma 4.8 again, δ is inner on eR. To prove that P is locally closed let us consider the following partition of the set of prime ideals of A that properly extend P . S1 S2 S3 := {Q ) P : Q prime, and no power of f is in Q} := {Q ) P : Q prime, not in S1, and Q ∩ R is a σ-prime (σ, δ)-ideal} := {Q ) P : Q prime, and not in S1 or S2} It suffices to show that for each of i = 1, 2, 3,T Si 6= P . Q ∈ S2, but f /∈ P as P ∩ R = (0). For i = 2, note that as σ-prime implies radical, we have that f ∈ Q for all For i = 3, applying Fact 4.15 to Q ∈ S3, we have that either A/Q is commutative or there is in R = R/I a prime ideal I := I ′/I extending Q ∩ R, and such that σ is the identity on R/I = R/I ′. The latter case is impossible using again that σ = τ ∗ and τ is translation on G by a non-identity element. So A/Q is commutative for all Q ∈ S3. As A/P = A/P is not commutative there exist a, b ∈ A such that g := [a, b] /∈ P . But g ∈ Q for all Q ∈ S3. It remains therefore to consider S1. Let eA = eR[x; σ, δ] = R[ 1 f ][x; σ, δ]. equivalence. As P ∩ R = (0), we know that no power of f is in P , and hence As δ is inner on eR, Proposition 4.11 tells us that eA satisfies the Dixmier-Moeglin eP := PeA is a prime ideal. As eA is a localisation of A we have that eP is rational, and hence locally closed. If Q ∈ S1 then QeA is a prime ideal properly extending eP . So there is α ∈ eA \ eP such that α ∈ QeA for all Q ∈ S1. For some n ≥ 0, f nα ∈ A. So f nα ∈ QeA ∩ A = Q. But f nα /∈ P . SoT S1 6= P , as desired. (cid:3) References [1] J. Bell, S. Launois, O. Le´on S´anchez, and R. Moosa. Poisson algebras via model theory and differential-algebraic geometry. J. Eur. Math. Soc. (JEMS), 19(7):2019–2049, 2017. [2] J. Bell and W. Leung. The Dixmier-Moeglin equivalence for cocommutative Hopf algebras of finite Gelfand-Kirillov dimension. Algebras and Representation Theory, 17(6):1843–1852, 2014. [3] J. Bell, K. Wu, and S. Wu. The Dixmier-Moeglin equivalence for extensions of scalars and Ore extensions. To appear in Contemp. Math. [4] K. A. Brown and K. Goodearl. Lectures on algebraic quantum groups. Advanced Courses in Mathematics. CRM Barcelona. Birkhauser Verlag, Basel, 2002. [5] K.A. Brown, S. O'Hagan, J.J. Zhang, and G. Zhuang. Connected Hopf algebras and iterated Ore extensions. Journal of Pure and Applied Algebra, 219(6):2405–2433, 2015. [6] A. Buium. Differential algebraic groups of finite dimension, volume 1506 of Lecture Note in Mathematics. Springer-Verlag, 1992. [7] J. R. Fisher. A Goldie theorem for differentiably prime rings. Pacific J. Math., 58(1):71–77, 1975. [8] J. Freitag and R. Moosa. Finiteness theorems on hypersurfaces in partial differential-algebraic geometry. To appear in Advances in Mathematics. [9] K. Goodearl. Prime ideals in skew polynomial rings and quantized Weyl algebras. Journal of Algebra, 150(2):324–377, 1992. [10] E. Hrushovski. A generalization of a theorem of Jouanolou's. Unpublished, 1995. 34 JASON BELL, OMAR LE ´ON S ´ANCHEZ, AND RAHIM MOOSA [11] E. Hrushovski. Computing the galois group of a linear differential equation. In Differential Galois theory, volume 58, pages 97–138. Banach Center Publications, Polish Academy of Sciences, 2002. [12] R. Irving. Generic flatness and the Nullstellensatz for Ore extensions. Comm. Algebra, 7(3):259–277, 1979. [13] R. Irving and L. Small. On the characterization of primitive ideals in enveloping algebras. Math. Z., 173(3):217–221, 1980. [14] I. Kaplansky. An introduction to differential algebra. Hermann, Paris, second edition, 1976. [15] E. R. Kolchin. Differential algebra and algebraic groups. Academic Press, New York-London, 1973. Pure and Applied Mathematics, Vol. 54. [16] P. Kowalski and A. Pillay. Quantifier elimination for algebraic D-groups. Trans. Amer. Math. Soc., 358(1):167–181, 2006. [17] D. Marker. Manin kernels. In Connections between Model Theory and Algebraic and Analytic Geometry, volume 6 of Quaderni di Matematica, pages 1–21. Dipartimento di Matematica Seconda Universit`a di Napoli, 2000. [18] J. C. McConnell and J. C. Robson. Noncommutative Noetherian rings, volume 30 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, revised edition, 2001. With the cooperation of L. W. Small. [19] S. Montgomery. Hopf algebras and their actions on rings, volume 82 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, 1993. [20] A. N. Panov. Ore extensions of Hopf algebras. Mat. Zametki, 74(3):425–434, 2003. [21] A. Pillay. Differential algebraic groups and the number of countable differentially closed fields. In Model Theory of Fields, volume 5 of Lecture Notes in Logic, pages 114–134. Springer, 1996. [22] A. Pillay. Remarks on algebraic D-varieties and the model theory of differential fields. In Logic in Tehran, volume 26 of Lect. Notes Log., pages 256–269. Assoc. Symbol. Logic, La Jolla, CA, 2006. [23] M. Rosenlicht. Toroidal algebraic groups. Proc. Amer. Math. Soc., 12:984–988, 1961. [24] L. Rowen. Ring theory. Vol. II, volume 128 of Pure and Applied Mathematics. Academic Press, Inc., Boston, MA, 1988. [25] M. Singer and M. van der Put. Galois theory of linear differential equations, volume 328. Grundlehren der mathematischen Wissenschaften, Springer, 2003. [26] S. Skryabin. New results on the bijectivity of antipode of a Hopf algebra. J. Algebra, 306(2):622–633, 2006. Department of Pure Mathematics, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada N2L 3G1 E-mail address: [email protected] School of Mathematics, University of Manchester, Oxford Road, Manchester, United Kingdom M13 9PL E-mail address: [email protected] Department of Pure Mathematics, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada N2L 3G1 E-mail address: [email protected]
1307.8419	1	1307	2013-07-31T18:45:00	A note on extensions of nilpotent algebras of Type 2	[ "math.RA" ]	We propose the study and description of the structure of complex Lie algebras with nilradical a nilpotent Lie algebra of type 2 by using sl2(C)-representation theory. Our results will be applied to review the classification given in [1] (J. Geometry and Physics, 2011) of the Lie algebras with nilradical the quasiclassical algebra L5;3. A non-Lie algebra has been erroneously included in this classification. The 5-dimensional Lie algebra L5;3 is a free nilpotent algebra of type 2 and it is one of two free nilpotent algebras admitting an invariant metric. According to [, Ok98] quasiclassical algebras let construct consistent Yang-Mills gauge theories.	math.RA	math	A note on extensions of nilpotent Lie algebras of type 2 Pilar Benitoa, Daniel de-la-Concepci´ona aDpto. Matem´aticas y Computaci´on, Universidad de La Rioja, 26004, Logrono, Spain Abstract We propose the study and description of the structure of complex Lie algebras with nilradical a nilpotent Lie algebra of type 2 by using sl2(C)-representation theory. Our results will be applied to review the classification given in [1] (J. Ge- ometry and Physics, 2011) of the Lie algebras with nilradical the quasiclassical algebra L5,3. A non-Lie algebra has been erroneously included in this classifi- cation. The 5-dimensional Lie algebra L5,3 is a free nilpotent algebra of type 2 and it is one of two free nilpotent algebras admitting an invariant metric. Ac- cording to [13], quasiclassical algebras let construct consistent Yang-Mills gauge theories. Keywords: Lie algebra, quasiclassical algebra, Levi subalgebra, nilpotent algebra, free nilpotent algebra, derivation, representation. 2010 MSC: 17B10, 17B30 1. Introduction From Levi Theorem, any finite-dimensional complex Lie algebra g (product denoted by [x, y]) with (solvable) radical r is of the form g = s ⊕ r, where s is a semisimple subalgebra named Levi subalgebra (also termed Levi factor ). The Lie algebra g is called faithful if the adjoint representation of the subalgebra s on r is faithful (equivalently, g contains no nonzero semisimple ideals). Note that any Lie algebra with radical r can be decomposed into the direct sum (as ideals) of a semisimple Lie algebra and a faithful Lie algebra with the same radical. Let now consider the nilradical n of g, the largest nilpotent ideal inside g. The nilradical is contained in the solvable radical and they satisfy the nice product relation: 3 1 0 2 l u J 1 3 ] . A R h t a m [ 1 v 9 1 4 8 . 7 0 3 1 : v i X r a (1) [g, r] ⊆ n. r n In other words, the action of g on is trivial. Corcerning the problem of classification of Lie algebras of a given radical, in 1944, Malcev [10] (see [15, Theorem 4.4, section 4]) stablished the following structure result: Email addresses: [email protected] (Pilar Benito), [email protected] (Daniel de-la-Concepci´on) Preprint submitted to Elsevier June 3, 2021 Theorem 1.1. Any faithful Lie algebra g with radical r is isomorphic to a Lie algebra of the form s ⊕id r where s is a semisimple subalgebra of a Levi subalgebra s0 of the derivation algebra of r, named Der r. Moreover, given s1 and s2 semisimple subalgebras of s0, the algebras s1 ⊕id r and s2 ⊕id r are isomorphic if and only if s2 = As1A−1, where A is an authomorphism of r. (cid:3) Previous theorem reduces the classification problem of Lie algebras with a given radical r, to the analysis of derivations and automorphisms of the solvable Lie algebra r. The same argument can be apply if we consider the problem of classifying Lie algebras with a given nilradical. According to [11], solvable Lie algebras can be classify through nilpotent ones and, from [17], any nilpotent Lie algebra can be obtained as a quotient of a free nilpotent. So, it seems to be quite natural to start classifying nilpotent Lie algebras, their derivations and automorphisms and then to extend nilpotent algebras through suitable derivations sets to solvable and nonsolvable Lie algebras. This is the starting point in [2] where Malcev decompositions of Lie algebras are studied, or in [14] where the problem of classifying Lie algebras whose radical is just the nilradical of a parabolic subalgebra of a semisimple Lie algebra is treated. In this paper we will deal with free nilpotent Lie algebras of type 2 (i.e.: the codimension of the derived algebra is 2) and their extensions; from [17], this class of algebras encodes the classification of filiforms and (type 2) quasi-filiform Lie algebras. The Levi subalgebra of the derivation algebra of any free nilpotent Lie algebra of type 2 is just a 3-dimensional split simple Lie algebra sl2(C). Our main tactic to get extensions is based on the well known representation theory of this simple Lie algebra. This technic has been previously used in [18] to get the classification of the 9-dimensional nonsolvable indecomposable Lie algebras. The paper splits into four Sections apart from the Introduction. In Section 2, some basic facts on free nilpotent Lie algebras n2,t, type 2 and arbitrary nilindex t, are given; this section also includes a complete description of their derivation Lie algebras Der n2,t. As a enlighten application, we present the classification of nilpotent Lie algebras of type 2 and nilindex t ≤ 4 in a nested way. Section 3 deals with the structure of solvable Lie algebras with nilradical n2,t. It turns out that the algebras in this class appear as extensions through vector subspaces of Der n2,t of dimension at most two. Section 4 is devoted to non-solvable Lie algebras with nilradical n2,t; this type of extensions have dimension at most one and are determined by derivations that centralize the Levi subalgebra of Der n2,t. The results in Sections 3 and 4 will be used in the final section 5 to review and present in a unify way the classifications of Lie algebras with nilradical n2,2 and n2,3 given in [16] (only solvable extensions are considered) and [1]. The Lie algebra n2,2 is the 3-dimensional Heisenberg algebra and n2,3 is a 5-dimensional quasi-classical Lie algebra denoted in [1] as L5,3. Here quasi-classical means endowed with a symmetric non-degenerate invariant form according to [12, Definition 2.1]. In [3], the authors stablished that n2,3 and n3,2 (free nilpotent of type 3) are the unique quasi-classical free nilpotent Le algebras and, following [13], quasi-classical Lie algebras let construct consistent Yang-Mills gauge theories. Hence, the results in this paper could have general 2 interest in theorical physics. Along the paper we deal with complex Lie algebras (complex field is denoted by C), although most part of the results are valid over arbitrary fields of charac- teristic zero except those included on Section 3, where algebraically closed field features are used. Basics definitions and known facts on Lie algebras follows from [7] and [8]. 2. Lie algebras of type 2 (The results in this section are partially included in [5, Section 3] where universal free nilpotent Lie algebras of characteristic zero and of arbitrary type are treated. So, the results in this section hold over any field of characteristic zero.) A nilpotent Lie algebra n is said to be t-nilpotent in case nt 6= 0 and nt+1 = 0; the type of n is given by the natural number dim n − dim n2. So type 2 implies dim n = 2+dim n2. This is the first condition that a filiform Lie algebra satisfies; in fact from [17, Proposition 4] the class of free nilpotent Lie algebras of type 2 encodes the classification of the filiform class. From the 2-element set {x1, x2} and the (2-dimensional) vector space m = spanhx1, x2i, the universal nilpotent Lie algebra n2,t is defined as the quotient: n2,t = FL2 FLt+1 2 (2) of the free Lie algebra FL2 over 2-elements through the ideal, FLt+1 2 = Xj≥t+1 spanh[. . . [xi1 xi2 ] . . . xij ] : xis = x1, x2i. For the Lie algebra in (2) we have (cf. [17, Proposition 4, Proposition 2] and [6, Proposition 1.4]): Proposition 2.1. The algebra n2,t is a quasicyclic t-nilpotent Lie algebra of type 2 and any other t-nilpotent Lie algebra of type 2 is an homomorphic image of n2,t. Moreover, the derivation Lie algebra of n2,t is: Der n2,t = {bδ : δ ∈ Hom(m, m)} ⊕ {bδ : δ ∈ Hom(m, n2 2,t)}, where bδ([xi1 , xi2 , . . . , xij , . . . xis ]) = sXj=1 [xi1 , xi2 , . . . , δ(xij ), . . . xis ], (3) δ ∈ Hom(m, m) or δ ∈ Hom(m, n2 [. . . [[xi1 xi2 ] . . .]xis ]. 2,t) and for s ≥ 2, [xi1 , xi2 , . . . , xij , . . . xis ] = (cid:3) According to [9], when a nilpotent Lie algebra n has a subspace u that complements n2 and u lets n be decomposed as a (finite) direct sum of subspaces 3 uk = [u, uk−1], n is called quasicyclic. So, from Proposition 2.1, we can assume w.l.o.g.: n2,t = m ⊕ m2 ⊕ . . . ⊕ mt, (4) where m = spanhx1, x2i and for k ≥ 2, mk = spanh[. . . [xi1 xi2 ] . . . xik ] : xij = x1, x2i. The direct sum in (4) provides a graded decomposition and, the di- mension of each homogeneous component ms, is given by the expression (µ the Moebius function): µ(d)2 s d . (5) dim ms =Xds Denote by gl(m) the general linear Lie algebra of C-linear maps δ : m → m and, for j ≥ 1, Derj n2,t = {bδ : δ ∈ Hom(m, mj)}. So Der1 n2,t = {bδ : δ ∈ gl(m)}, is a Lie subalgebra isomorphic to gl(m) = sl(m) ⊕ C · idm (the 3- dimensional Lie algebra of traceless maps sl(m) plus the identity map) and Der n2,t = tMj=1 Derj n2,t. (6) Previous decomposition follows the multiplication rule [di, dk] ∈ Deri+k−1 n2,t, ds ∈ Ders n2,t. Moreover, the derived algebra Der0 1 n2,t = [Der1 n2,t, Der1 n2,t] is just Der0 split 3-dimensional Lie algebra sl2(C). In this way, 1 n2,t = {bδ : δ ∈ sl(m)}, a Levi subalgebra of Der n2,t isomorphic to the Der1 n2,t = Der0 1 n2,t ⊕ C · id2,t (7) where id2,t = didm, i.e. the extended derivation of idm. The solvable radical R2,t and the nilradical N2,t of the derivation algebra are given by: N2,t = Derj n2,t tMj≥2 R2,t = C · id2,t ⊕ N2,t, (8) (9) We also note that the derivations inside N2,t are nilpotent maps and [id2,t,bδ] = (k − 1)bδ for any bδ ∈ Derk n2,t (k ≥ 1). Concerning the classification problem of Lie algebras having n2,t as nilradical we have (cf. of [5, Proposition 3.2] and Theorem 1.1): Theorem 2.2. Up to isomorphisms, g2,t = Der0 1 n2,t ⊕id n2,t is the unique faithful complex Lie algebra with radical the free nilpotent Lie algebra n2,t. In particular, apart from g2,t, any nonsolvable complex Lie algebra with radical n2,t is a direct sum as ideals of the form either s ⊕ n2,t or s ⊕ g2,t, where s is an arbitrary semisimple Lie algebra. (cid:3) The product in g2,t is given by considering the product in Der0 1 n2,t as subal- gebra of the linear algebra of derivations (so, [d1, d2] = d1d2 − d2d1), the usual 4 product in n2,t and [d, a] = d(a) in case d ∈ Der0 1 n2,t, a ∈ n2,t. By using rep- resentation theory of sl2(C) (see [7, section II.7]) we can get explicit basis for g2,t through the natural action of Der0 1 n2,t on m and the induced action on the j-tensor vector space ⊗jm (for a computational approach see [4]). The next result shows the way in which this method works: Proposition 2.3. Up to isomorphisms, the nonsolvable complex Lie algebras with solvable radical a universal nilpotent algebra of type 2 and nilindex t ≤ 4 are: a) The trivial extensions s ⊕ n2,t (direct sum as ideals), s any arbitrary semisimple Lie algebra. b) The direct sum as ideals of any semisimple Lie algebra s and one of the following faithful Lie algebras: i) The 5-dimensional algebra g2,1 with basis {e, f, h, v0, v1} and nonzero products [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, [h, v1] = −v1, [e, v1] = v0 and [f, v0] = v1. In this case, n2,1 = Cv0⊕, Cv1 is a V (1)-module of the Levi subalgebra Ce ⊕ Cf ⊕ Ch (∼= Der0 1 n2,1). ii) The 6-dimensional algebra g2,2 with basis {e, f, h, v0, v1, w0} and non- zero products [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1 and [v0, v1] = w0. In this case, n2,2 = Cv0 ⊕ Cv1 ⊕ Cw0 is a V (1) ⊕ V (0)-module of the Levi subalgebra Ce ⊕ Cf ⊕ Ch (∼= Der0 1 n2,2). iii) The 8-dimensional algebra g2,3 with basis {e, f, h, v0, v1, w0, z0, z1} and nonzero products [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1, [h, z0] = z1, [h, z0] = −z0, [e, z1] = z0, [f, z0] = z1, [v0, v1] = w0, [v0, w0] = z0 and [v1, w0] = z1. In this case, n2,3 = Cv0 ⊕ Cv1 ⊕ Cw0 ⊕ Cz0 ⊕ Cz1 is a V (1) ⊕ V (0) ⊕ V (1)-module of the Levi subalgebra Ce ⊕ Cf ⊕ Ch (∼= Der0 1 n2,3). iv) The 11-dimensional algebra g2,4 with basis {e, f, h, v0, v1, w0, z0, z1, x0, x1, x2} and nonzero products [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1, [h, z0] = z1, [h, z0] = −z1, [e, z1] = z0, [f, z0] = z1, [h, x0] = 2x0, [h, x2] = −2x2, [e, x1] = 2x0, [e, x2] = x1, [f, x0] = x1, [f, x1] = 2x2, [v0, v1] = w0, [v0, w0] = z0, [v1, w0] = z1, [v0, z0] = x0, [v0, z1] = [v1, z0] = 1 2 x1 and [v1, z1] = x2. In this case, n2,4 = Cv0 ⊕ Cv1 ⊕ Cw0 ⊕ Cz0 ⊕ Cz1 ⊕ Cx0 ⊕ Cx1 ⊕ Cx2 is a V (1) ⊕ V (0) ⊕ V (1) ⊕ V (2)-module of the Levi subalgebra Ce ⊕ Cf ⊕ Ch (∼= Der0 1 n2,4). Proof. Let L = s ⊕ n2,t be a such algebra, s a Levi subalgebra. By using the adjoint (restricted) representation ρ = adn2,t of L, the radical n2,t is viewed as a s-module. In case ρ is trivial, a) follows; otherwise, ρ(s) is a semisimple subalgebra of the Levi subalgebra of Der n2,t. So, s = Ker ρ ⊕ s1 where s1 is a simple split 3-dimensional ideal of s and s1 ⊕ n2,t is a faithful Lie algebra. From Theorem 2.2, up to isomorphisms, we can assume s1 ⊕ n2,t = Der0 1 n2,t ⊕id n2,t. From now on, we fixed a standard basis {e, f, h} of the split simple 3-dimensional Der0 1 n2,t- action, m is a module of type V (1) and we can also take a standard basis 1 n2,t so: [e, f ] = h, [h, e] = 2e, [h, f ] = −2f . Through the natural Der0 5 {v0, v1} of m as in [7]. In case t = 1, b) − i) follows immediately. For s ≥ 2 the homogeneous components ms of given in (4) are Der0 1 n2,t-submodules inside the tensor product m ⊗ ms−1 (the Lie product [·, ·] : m ⊗ ms−1 → ms is an onto Der0 1 n2,t-module homomorphism). So, from Clebch-Gordan formula an a counting dimension argument based on (5) we have: • m2 ⊆ m ⊗ m = V (1) ⊗ V (1) = V (2) ⊕ V (0); so m2 = V (0), is a trivial module. Thus m2 = spanh[v0, v1]i which leads to case b) − ii) for t = 2 by defining w0 = [v0, v1]. • m3 ⊆ m ⊗ m2 = V (1) ⊗ V (0) = V (1); so m3 = V (1), is a 2-irreducible module. Since m3 = [m, m2] = spanh[v0, w0] = z0, [v1, w0] = z1i, b) − iii) is obtained in case t = 3. • m4 ⊆ m ⊗ m3 = V (1) ⊗ V (1) = V (2) ⊕ V (0); so m4 = V (2), is a 3- irreducible module. From m4 = [m, m3] and [v1, z0] = [v0, z1], the set {[v0, z0] = x0, 2[v1, z0] = x1, [v1, z0] = x2} turns out a standar basis of V (2) inside m4. So, b) − iv) follows. Now, from Proposition 2.3, [6, Proposition 1.5] and [5, Theorem 3.5] we get the whole list of nilpotent Lie algebras of type 2 and nilindex ≤ 4: Corollary 2.4. Up to isomorphisms, the nilpotent Lie algebras of type 2 and nilindex t ≤ 4 are: a) n2,t for t = 1, 2, 3, 4; b) the quotient Lie algebra n3 2,3 = spanhz0, z1i; c) the quotient Lie algebra n2,3 I n2,4 I , where I is any 1-dimensional subspace of where I is one of the following ideals: i) any 1 or 2-dimensional subspace of n4 ii) I = spanhz1 + αx0, x1, x2i, α ∈ C or iii) I = spanhz0 + αz1 + βx2, 2x0 + αx1, x1 + 2αx2i, α, β ∈ C. 2,3 = spanhx0, x1, x2i, Moreover, the algebras in the previous list that admit a nontrivial Levi ex- tension are those given in item a). Proof. From [6, Proposition 1.5], these algebras are of the form ideal I is contained in n2 a straightforward computation. 2,t and such that nt 2,t 6⊆ I. Now, the result for t ≤ 4 is n2,t I where the Remark 1. Corollary 2.4 provides the classification of filiform Lie algebras of nilindex ≤ 4. Apart from n2,1 and n2,2 the algebras in this class are given in items b), c)-ii), c)-iii). By considering 2-subspaces I, item b)-i) and n2,3 gives us all quasifiliform algebras of type 2 and nilindex ≤ 4. 6 3. Solvable extensions of n2,t Now we study the solvable Lie algebras r having a 2-free nilpotent algebra n2,t as nilradical. The basic idea is to get r by extending n2,t through commuting vector spaces of derivations of the Lie algebra Der1 n2,t = Der0 1 n2,t ⊕ C · id2,t described in (7) which is isomorphic to gl2(C). In this section we will show that, the codimension n2,t on its extended solvable radical r is at most 2. Lemma 3.1. Let r be a complex solvable Lie algebra with nilradical n2,t. Then, r = n2,t ⊕ t where t is a complex vector space equidimensional to the projec- tion p1(t) = projDer1 n2,t adn2,t t which is an abelian subalgebra of Der1 n2,t of dimension at most 2 and the derivation set adn2,t t contains no nilpotent ele- ments. Moreover, in case t = C · x + C · y, p1(t) = CDer1 n2,t (δ) for some δ ∈ Der1 n2,t\C · id2,t and the braket derivation [adn2,t x, adn2,t y] is an inner derivation of n2,t. Proof. Let t be any arbitrary complement of n2,t in r. From (1), we have the relation t2 ⊆ r2 ⊆ n2,t. Note that, since the nilradical of m is n2,t, the vector linear space adn2,t t is a subspace of derivations in n2,t with no nilpotent linear maps. Consider now the adjoint representation adn2,t : r → Der n2,t and the projection homomorphism π0 : Der n2,t → Der1 n2,t. Then, Ker π0 ◦ adn2,t = is an abelian Lie algebra {a ∈ r : adn2,t a ∈ N2,t} = n2,t. Thus the quotient equidimensional to t and isomorphic to Im π0 ◦ adn2,t . Since Der1 n2,t ∼= gl2(C), up to isomorphisms, Im π0 ◦adn2,t is a set of commutative linear transformations over a 2-dimensional vector space. For any α · idm 6= f ∈ Im π0 ◦ adn2,t , the centralizer of f , Cgl2(C)(f ) is a 2-dimensional subspace which contains Im π0 ◦ adn2,t . Note that for 2-dimensional extensions, Im π0 ◦adn2,t is just a centralizer. The final assertion follows from the fact adn2,t [x, y] = [adn2,t x, adn2,t y] and [x, y] ∈ n2,t. n2,t r 3.1. Derivations and automorphisms From Proposition 2.1, the set of derivations of any Lie algebra n2,t is com- pletely determined by Hom(m, m) and Hom(m, n2 d,t) = ⊕2≤j≤tHom(m, mj). Moreover, any automorphism appears as extension of a map of the general linear group GL(m) according to [17, Proposition 3]. Along this section, the deriva- tion algebras and automorphisms groups of free nilpotent Lie algebras will be represented through matrices (αij ) relative to the basis given in Proposition 2.3 (m = C · v0 ⊕ C · v1, only nonzero products are displayed): n2,1 : Abelian 2-dimensional. Der n2,1 = gl2(C) = Der1n2,1 : u =(cid:18) α1 + β α3 Dβ −α1 + β (cid:19) , u = (α1, α2, α3); α2 extended derivation of the identity map I2 is I2,1 = I2; 7 N2,1 = Inner n2,1 = 0; Aut n2,1 = GL2(C) : Φv =(cid:18) α1 α2 α3 α4 (cid:19) , v = (α1, . . . , α4) and ǫ = α1α4 − α2α3 6= 0. n2,2 : Heisenberg 3-dimensional, [v0, v1] = w0. Der n2,2 : Dβ Der1 n2,2 : Dβ α3 α4 α1 + β u =  (α1,α2,α3,0,0) =  −α1 + β α2 α5 0 0 2β  , u = (α1, . . . , α5); α1 + β α2 α3 0 −α1 + β 0 0 0 2β   ; extended derivation of I2 is the 3 × 3 matrix I2,2 = D1 0; N2,2 = Inner n2,2 : D0 (0,0,0,α4,α5) =  0 0 0 0 α4 α5 0 0 0   . Aut n2,2 : Φv =  α1 α2 α3 α4 α5 α6 0 0 ǫ   , v = (α1, . . . , α5). n2,3 : denoted as L3,5 in [1], [v0, v1] = w0, [vi, w0] = zi, i = 0, 1. Der n2,3 : Dβ u =   α1 + β α2 α3 α4 α6 α8 −α1 + β α5 α7 α9 0 0 2β α5 −α4 0 0 0 α1 + 3β 0 0 0 α2 α3 −α1 + 3β   , u = (α1, . . . , α9); Der1n2,3 : Dβ (α1,α2 ,α3,0,0,0,0,0,0) =   α1 + β α2 α3 0 0 0 −α1 + β 0 0 0 0 0 2β 0 0 0 0 0 α1 + 3β 0 0 0 α2 α3 −α1 + 3β   ; 8 N2,3 : D0 (0,0,0,α4,...,α9) = Inner n2,3 : D0 (0,0,0,α4,α5,α6,0,0,0) =   0 0 0 0 0 0 0 α4 α5 α6 α7 α5 α8 α9 −α4 0 0 0 0 0 0 0 0 0 0   ;   0 0 0 0 0 0 0 α4 α5 0 α6 α5 α6 −α4 0 0 0 0 0 0 0 0 0 0 0   ; extended derivation of I2 is the 5 × 5 matrix I2,3 = D1 0; Aut n2,3 : Φv =   0 0 ǫ α2 α4 α6 α8 α1 α3 α5 α7 α1α6 − α2α5 α9 α10 α3α6 − α4α5 0 0 0 ǫα1 ǫα3 0 0 0 ǫα2 ǫα4   , v = (α1, . . . , α10). 3.2. Solvable 1-extensions of n2,t for t = 1, 2, 3 Solvable 1-extensions t = k · x of n2,t follows from no nilpotent derivations adn2,t x = D + O where D ∈ Der1n2,t and O ∈ N2,t. Moreover, the isomorphism problem between two different one dimensional extensions is solved in an easy way: n2,t ⊕ k · x ∼= n2,t ⊕ k · x′ if and only if there exists a Φ ∈ Aut n2,t such that Φ · (D + O) · Φ−1 = ad a + α · (D′ + O′) for some a ∈ n2,t, α ∈ C. Since Φ·Inner n2,t ·Φ−1 = Inner n2,t, we can consider w.l.o.g. extensions by derivations with zero projection on Inner n2,t. 3.2.1. 1-extensions in case n2,1 : From Inner n2,1 = 0, we have that solvable 1-extensions are given by Jordan forms of 2 × 2 matrices. So, up to isomorphisms, we can extend by one of the following derivations: 1. D1 2. D 2 0 (0,0,1) =(cid:18) 1 =(cid:18) 1 1 (cid:19) . 0 α (cid:19), α ∈ C; 2 ,0,0) ( 1−α 1+α 0 1 3.2.2. 1-extensions in case n2,2 : Since Inner n2,2 = N2,2, solvable 1-extensions are also given by Jordan forms of 2 × 2 matrices. As in the previous case, we have two possibilities: 1. D1 2. D 1 1 0 (0,0,1,0,0) =  =  2 ,0,0,0,0) ( 1−α 1+α 2 0 0 1 0 0 2 1 0 0 α 0 0   . 0 0 1 + α  , α ∈ C; 9 3.2.3. 1-extensions in case n2,3 : Note that derivations with no projection on Inner n2,3 are of the form:   0m,n zero matrix m × n, A and M = (cid:18) α6 α7   0 (cid:19) 2 × 2 matrices. According A 01,2 M 02,1 A + 2βI2 02,2 02,1 02,2 01,2 01,1 01,2 M 02,1 02,2 02,1 tr A 02,1 A + 2βI2   =    +  A 01,2 02,2 02,1 tr A 02,2 01,2 02,2 01,2 α8 , to (6), the second summand in previous matrix decomposition corresponds to a derivation δ : n2,3 → Z(n2,3) = n3 2,3 which belongs to Der3n2,3, so δ has the matrix general form:   02,2 01,2 α6 α7 α8 α9 02,1 02,2 01,1 01,2 02,1 02,2   We note that, Der3n2,3 is an abelian subalgebra which is invariant by conjuga- tion. This subalgebra contains a 1-dimensional subspace of inner derivations. Then, up to isomorphisms, the extensions we are looking for are given by one of the following type of derivations: (a) The derivation matrix D1 u where u = (0, 0, 1, 0, 0, α6, α7, α8, 0), i.e.: D1 u =   0 0 0 0 2 0 0 3 0 1 0 1 1 1 0 0 α6 α7 0 α8 4 (2α6 + α7), α7 0 0 0 0 3   . Φ−1 v · D1 u · Φv = D1 u1   1 1 0 0 0 0 0 1 0 0 2 0 0 0 0 0 0 0 0 0 0 3 0 1 3   . D1 u1 = (b) The derivation matrix D u 2 1+α Denoting v = (1, 0, 0, 1, 0, 0, 1 automorphism Φv transforms previous derivation by conjugation into: 4 − α7 4 + α8 2 , − α6 2 , − α7 4 ), the where u1 = (0, 0, 1, 0, 0, 0, 0, 0, 0). Then, we get: with u = ( 1−α 2 , 0, 0, 0, 0, α6, α7, α8, 0), i.e.: 1+α D 2 u =   0 1 0 α 0 0 α6 α7 0 α8 0 0 1 + α 0 0 0 0 0 2 + α 0 0 0 0 0 1 + 2α   . Now three subcases can be considered: 10 (b.1) α = −1: By conjugation through the automorphism Φv, with v = (1, 0, 0, 1, 0, 0, 0, α7 into: 2 , 0), the derivation D u = D0 u becomes 2 , α8 1+α 2 Φ−1 v · D0 u · Φv = D0 w, where w = (1, 0, 0, 0, 0, α6, 0, 0, 0). v′ = ( 1√α6 , 0, 0, 0, 0, 0, 0), we get: 1√α6 , 0, 0, If α6 6= 0, by using Φv′ with Φ−1 v′ · D0 w · Φv′ = D0 u3 , u3 = (1, 0, 0, 0, 0, 1, 0, 0, 0). So, up to isomorphisms, two new possible derivations appear (u2 = (1, 0, 0, 0, 0, 0, 0, 0, 0) is related to α6 = 0): D0 u2 =   1 0 0 −1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 −1   and D0 u3 =   1 0 0 −1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 −1   . (b.2) α = 0: Using v = (1, 0, 0, 1, 0, 0, α6, α7 Φv, transforms D u = D u into: 1+α 1 2 2 2 , 0, 0), the automorphism Φ−1 v · D 1 2 u · Φv = D 1 2 w, where w = ( 1 v′ = ( 1√α8 , 0, 0, 2 , 0, 0, 0, 0, 0, 0, α8, 0). Now, in case α8 6= 0, taking 1√α8 , 0, 0, 0, 0, 0, 0) we have: Φ−1 v′ · D 1 2 w · Φv′ = D 1 2 u5 , u5 = ( 1 derivations appear (u4 = ( 1 2 , 0, 0, 0, 0, 0, 0, 1, 0). So, up to isomorphisms, two additional 2 , 0, 0, 0, 0, 0, 0, 0, 0) is related to α8 = 0): 1 2 u4 = D   1 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 2 0 0 0 0 0 1   and D 1 2 u5 =   1 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 0 0 2 0 0 0 0 0 1   . (b.3) α 6= 0, −1: Taking v = (1, 0, 0, 1, 0, 0, α6 1+α , α7 2 , α8 2α , 0), we have: Φ−1 v · D 1+α 2 u 1+α · Φv = D 2 u6 , u6 = ( 1−α derivation: 2 , 0, 0, 0, 0, 0, 0, 0, 0). So, up to isomorphisms, we get the 1+α D 2 u6 =   1 0 0 α 0 0 0 0 0 0 0 0 1 + α 0 0 0 0 0 2 + α 0 0 0 0 0 1 + 2α   . 11 Summarizing: 1-extensions of n2,3 are given by one of the following derivations: 1. D1 2. D0 3. D u1 where u1 = (0, 0, 1, 0, 0, 0, 0, 0, 0); u3, where u3 = (1, 0, 0, 0, 0, 1, 0, 0, 0); u5, where u5 = ( 1 2 , 0, 0, 0, 0, 0, 0, 1, 0); 1 2 1+α 4. D 2 u6 , where u6 = ( 1−α 2 , 0, 0, 0, 0, 0, 0, 0, 0) and α ∈ C. 3.3. Solvable 2-extensions Solvable 2-extensions t = C · x ⊕ C · y are related to 2-dimensional centralizer subalgebras: CDer1n2,t (D) = C · I2,t ⊕ C · D, (10) inside Der1n2,t. According to Lemma 3.1, adn2,t x = I2,t + O1 and adn2,t y = D + O2 where Oi ∈ N2,t; in fact, we can assume Oi have no projection on Inner n2,t. Then, we can get a 2-extension in case the set spanhI2,t + O1, D + O2i has no nilpotent derivations and: [I2,t + O1, D + O2] = [I2,t, O2] + [O1, D] + [O1, O2] ∈ Inner n2,t. (11) 3.3.1. 2-extensions in case n2,1 : Since N2,1 = 0, solvable extensions are just given by C = CDer1n2,1 (D), where D is a 2 × 2 matrix (I2,1 = I2). If D is a non-semisimple (non nilpotent) matrix, up to isomorphisms and rescaling if necessary, we can assume that D = D1 (0,0,1); but then, D − I2 ∈ C is a nilpotent derivation, which is not possible. Hence, D is a semisimple matrix with two different eigenvalues, so we can assume w.l.o.g. D = D 1+α 2 ( 1−α 2 ,0,0) , α 6= 1 and therefore: Cn2,1 (D) = C · I2 ⊕ C ·(cid:18) 1 0 −1 (cid:19) . 0 Now, the extension t is given by the derivations adn2,1 x = I2,1 = I2 and adn2,1 y = D0 (1,0,0). Note that, t2 = C · [x, y] ⊆ Z(n2,1) = n2,1. Thus, we can consider the (unique) extension n2,1 ⊕ t1 with t1 = C · x ⊕ C · (y − [x, y]) an abelian subalgebra (x acts as the identity on n2,1). 3.3.2. 2-extensions in case n2,2 : In this case, I2,2 = D1 0 and N2,2 = Inner n2,2. From the last condition, solvable extensions are also given by C = CDer1n2,2 (D), D = Dβ (α1,α2,α3,0,0). As in the previous case, D must be semisimple. Otherwise, we can assume D = D1 (0,0,1,0,0), but then we get that D − I2,2 ∈ C is a nilpotent derivation. Hence, with α 6= 1 and therefore: up to isomorphisms, we can assume D = D 1+α 2 ( 1−α 2 ,0,0,0,0) n2,1(D) = C · I2,2 ⊕ C ·  12 0 1 0 0 −1 0 0 0 0   Now, the extension t is given by the derivations adn2,2 x = I2,2 and adn2,2 y = D0 (1,0,0,0,0). Since the extension follows from an abelian subalgebra of deriva- tions, we must observe that in the extended algebra n2,2 ⊕ t, [x, y] ∈ Z(n2,2) = C·w0. Then, we can consider that the (unique) extension is given by the abelian subalgebra t1 = C · x ⊕ C · (y − 1 2 [x, y]) (in this case, x acts as 2 · id on C · w0.) 3.3.3. 2-extensions in case n2,3 : We look for centralizer subalgebras C = CDer1n2,3 (D), D = Dβ u where u = (α1, α2, α3, 0, 0, 0, 0, 0, 0). Since derivations inside N2,3 without projection on Inner n2,3 belongs to Der3n2,3, which is a subspace invariant by conjugation through automorphisms, and taking in account that spanhI2,t + O1, D + O2i has no nilpotent derivations, in a similar vein as in previous cases, we get that I2,t + O1 = D1 u and D + O2 = D1 w where u = (0, 0, 0, 0, 0, α6, α7, α8, 0) and w = (1, 0, 0, 0, 0, β6, β7, β8, 0; 0), i.e.: D1 u =   0 1 1 0 0 0 α6 α7 0 α8 0 0 2 0 0 0 0 0 0 0 0 3 0 0 3 and D1 w =   0 1 0 0 −1 0 0 0 0 β6 0 β8 0 β7 0 0 0 0 0 0 0 1 0 0 −1   .   0 0 0 2β6 Now, [D1 u, D1 w] =   2(α8 + β8) 0 0 0 0 0 0 0 0 0 2(β7 − α7) 0 0 0 0 0 0 0 0 0 0   is an inner derivation if and only if β6 = 0, β7 = α7 and β8 = −α8. Consid- ering the automorphism: Φv where v = (1, 0, 0, 1, 0, 0, − α6 2 , 0), by conjugation, we get the derivations: 2 , − α7 2 , − α8 Φ−1 v · D1 u · Φv = Φ−1 v · D1 w · Φv = 1 0 0 0 0 0 0 1 0 0 2 0 0 0 0 0 0 0 0 0 0 3 0 0 3   0 0 1 0 −1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 −1     = I2,3 and   = D0 u1 , where u1 = (1, 0, 0, 0, 0, 0, 0, 0, 0). Hence, up to isomorphisms, there is a unique way to extend through derivations which is given by declaring adn2,2 x = I2,3 and adn2,2 y = D0 u1. Finally, we must observe that in the corresponding extension 13 n2,3 ⊕ t, [x, y] ∈ Z(n2,3) = C · z0 ⊕ C · z1. Since x acts as 3 · id on Z(n2,3), we can consider the (unique) extension n2,3 ⊕ t1 where t1 = spanhx, y − 1 3 [x, y]i is an abelian subalgebra. 4. Non-solvable extensions of n2,t From Theorem 2.2 and [18, Theorem 2.1 and 2.2] (see also [5, Proposition 2.2]), the faithful nonsolvable Lie algebras with nilradical n2,t are of the form, g = g2,t ⊕ t0 (12) where g2,t = Der0 1 n2,t ⊕id n2,t is as described in Theorem 2.2, r = n2,t ⊕ t0 is the solvable radical of g and t0 is a Der0 1 n2,t-trivial module with no ad-nilpotent elements. So, in the Lie algebra g, we have [Der0 1 n2,t, t0] = 0. Denote by (n2,t)0 the sum of all trivial modules in n2,t and note that this vector space is a subalgebra of n2,t. Since t2 0 ⊆ (n2,t)0, the direct sum: r0 = t0 ⊕ (n2,t)0 (13) is a subalgebra of r that contains the (solvable) subalgebra generated through t0. But from the Jacobi identity in the Lie algebra (12), we have [Der0 1 n2,t, adn2,t t0] = 0, so the set of derivations adn2,t t0 centralizes Der0 1 n2,t in Der n2,t. All these basic ideas yield to the following result: Lemma 4.1. Let g be a nonsolvable Lie algebra with nilradical n2,t. Then g is one of the following Lie algebras: a) g = s ⊕ r is a direct sum as ideals of a semisimple algebra s and a solvable Lie algebra r with nilradical n2,t; b) g = s ⊕ g2,t, a direct sum as ideals where s is either the null algebra or any arbitrary semisimple algebra or c) g = s ⊕ g(δ), a direct sum as ideals where s is either the null algebra or any arbitrary semisimple algebra and g(δ) is the Lie algebra g(δ) = g2,t ⊕ C · x, with [Der0 1 n2,t, x] = 0 and adn2,t x = id2,t + δ, δ ∈ N2,t where δ ∈ CDer n2,t (Der0 1 n2,t). Proof. Let g = s ⊕ r a Levi decomposition of g. In case [s, r] = 0 or t0 = 0 we get items a) and b). Otherwise, we have t0 6= 0 and [s, n2,t] 6= 0 and, from Theorem 2.2 and preliminary comments in this section, g decomposes as a direct sum of ideals, g = s1 ⊕ (Der0 1 n2,t ⊕id n2,t ⊕ t0), s1 being either semisimple or s1 = 0. Let r0 the subalgebra defined in (13) and consider the (restricted) adjoint representa- tion adn2,t : r0 → Der n2,t, which is both an homomorphism of Lie algebras and a Der0 1 n2,t-module homomorphism. Note that Ker adn2,t = Cr0(n2,t) = Z(n2,t)0, i.e. the trivial module living inside the center of the nilradical n2,t. Decom- pose (n2,t)0 as a direct sum of subspaces, (n2,t)0 = Z(n2,t)0 ⊕ v0. The quotient 14 Lie algebra (∼= v0 ⊕ t0) is isomorphic to adn2,t (r0), a subalgebra of r0 Z(n2,t)0 Der n2,t contained in CDer n2,t (Der0 1 n2,t) = k · id2,t ⊕ (N2,t)0, the last sumand is the sum of trivial Der0 1 n2,t-modules inside N2,t via the adjoint representation [δ, µ] = δµ − µδ. But adn2 t0 has no nilpotent derivations (the nilradical of g is just n2,t), so none of the elements inside adn2,t (t0) belongs to N2,t. Let x, y ∈ t0 be two nonzero elements. Rescaling, we can assume ad x = id2,t ⊕ δ and ad y = id2,t ⊕ µ, where δ, µ ∈ (N2,t)0. Then ad (x − y) ∈ (N2,t)0 and therefore, x − y is a nilpotent element in t0; the only possibility is x = y and part c) follows. It is immediate to check that items a), b) and c) provides Lie algebras. 2,t For free nilpotent of low index, the centralizer CDer n2,t (Der0 1 n2,t) is easy to be computed. In fact, we have: • t = 1, 2: CDer n2,t (Der0 1 n2,t) = C · id2,t; • t = 3, 4: CDer n2,t (Der0 1 n2,t) = C · id2,t ⊕ C · ad ω0, where ω0 = [v0, v1] according to Proposition 2.3. The above centralizer computation and Lemma 4.1 allow us to stablish: Theorem 4.2. For t = 1, 2, 3, 4 and up to isomorphisms, the faithful non- solvable complex Lie algebras with nilradical n2,t are g2,t = Der0 1 n2,t ⊕id n2,t described in Theorem 2.2 and those of the form g = g2,t ⊕ k · x where adg x acs trivially on Der0 1 n2,t and as the id2,t on n2,t. Proof. The result follows trivially from Lemma 4.1 in cases t = 1, 2, and for t = 2, 3 we note that the elements inside CDer n2,t (Der0 1 n2,t) are of the form α · id2,t ⊕ β · ad ω0. But ad ω0 is an inner derivation of n2,t and therefore it can be removed. 5. Conclusion: Lie algebras with nilradical n2,t for t = 1, 2, 3 From previous results and comments, the classification of Lie algebras with nilradical n2,t, boils down to the determination of suitable derivation sets con- taining no nilpotent elements as described in Lemmas 3.1 and 4.1. The sets of derivations has been completely determined for t = 1, 2, 3 in section 3 (solvable extensions) by using explicit matrix descriptions of derivations, inner deriva- tions and automorphisms of n2,t and in section 4 (nonsolvable extensions) by computing centralizers inside Der n2,t. In this final section we summarized the results in a unique theorem, where the whole list of Lie algebras with nilradical n2,1, n2,2 or n2,3 is given through their multiplication tables and the nested basis descriptions obtained in Proposition 2.3: Theorem 5.1. Up tp isomorphisms, the Lie algebras with nilradical a free nilpo- tent Lie algebra of type 2 and nilindex t ≤ 3 are (only nonzero products involving elements not in the nilradical are given): 15 i) The abelian 2-dimensional n2,1; ii) r1 2,1 = n2,1 ⊕ C · x : [x, v0] = v0 + v1, [x, v1] = v1; iii) r1,α 2,1 = n2,1 ⊕ C · x : [x, v0] = v0 and [x, v1] = αv1, α ∈ C; iv) r2 2,1 = n2,1 ⊕ C · x ⊕ C · y : [x, v0] = [y, v0] = v0, [x, v1] = −[y, v1] = v1; v) g2,1 = sl2(C) ⊕ n2,1 : [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1; 2,1 = sl2(C) ⊕ r1,1 2,1 : [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1, [x, v0] = v0, [x, v1] = v1; [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, vi) g1 vii) The Heisenberg 3-dimensional n2,2: [v0, v1] = w0, viii) r1 2,2 = n2,2 ⊕ C · x : [x, v0] = v0 + v1, [x, v1] = v1, [x, w0] = 2w0; ix) r1,α 2,2 = n2,2 ⊕ C · x : [x, v0] = v0, [x, v1] = αv1, [x, w0] = (1 + α)w0, α ∈ C; x) r2 2,2 = n2,2 ⊕ C · x ⊕ C · y : [x, w0] = 2w0; [x, v0] = [y, v0] = v0, [x, v1] = −[y, v1] = v1, xi) g2,2 = sl2(C) ⊕ n2,2 : [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, xii) g1 [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1; 2,2 = sl2(C) ⊕ r1,1 [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, 2,2 : [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1, [x, v0] = v0, [x, v1] = v1; [x, w0] = 2w0. xiii) The 5-dimensional n2,3: [v0, v1] = w0, [vi, w0] = zi, i = 0, 1, xiv) r1 2,3 = n2,3 ⊕ C · x : [x, z0] = 3z0 + z1, [x, z1] = 3z1; [x, v0] = v0 + v1, [x, v1] = v1, [x, w0] = 2w0, xv) r1,α 2,3 = n2,3 ⊕ C · x : [x, z0] = (2 + α)z0,[x, z1] = (1 + 2α)z1, α ∈ C; [x, v0] = v0, [x, v1] = αv1,[x, w0] = (1 + α)w0, xvi) r2 2,3 = n2,3 ⊕ C · x : z1; [x, v0] = v0 + z1, [x, w0] = w0, [x, z0] = 2z0, [x, z1] = xvii) r3 2,3 = n2,3 ⊕ C · x : [x, z1] = −z1; [x, v0] = v0 + z0, [x, v1] = −v1, [x, z0] = z0, xviii) r4 2,3 = n2,3 ⊕ C · x ⊕ C · y : [x, w0] = 2w0, [x, z0] = 3z0, [x, z1] = 3z1, [y, z0] = z0, [y, z1] = −z1; [x, v0] = [y, v0] = v0, [x, v1] = −[y, v1] = v1, xix) g2,3 = sl2(C) ⊕ n2,3 : [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1, [h, z0] = z0, [h, z1] = −z1,[e, z1] = z0, [f, z0] = z1, [v0, v1] = w0, [v0, w0] = z0 and [v1, w0] = z1; 16 xx) g1 2,3 = sl2(C) ⊕ r1,1 2,3 : [e, f ] = h, [h, e] = 2e, [h, f ] = −2f , [h, v0] = v0, [h, v1] = −v1, [e, v1] = v0, [f, v0] = v1, [h, z0] = z0, [h, z1] = −z1,[e, z1] = z0, [f, z0] = z1, [v0, v1] = w0, [v0, w0] = z0, [v1, w0] = z1, [x, v0] = v0, [x, v1] = v1, [x, w0] = 2w0,[x, z0] = 3z0, [x, z1] = 3z1; xxi) the direct sum as ideals s ⊕ g, where s is a semisimple Lie algebra and g one of the Lie algebras given in any of the previous items i) − xx). Remark 2. Theorem 5.1 provides the complete classification of complex Lie algebras with nilradical the 3-dimensional Heisenberg algebra n2,2; the classifi- cation problem in case of solvable extensions was solved in [16] even for the real field. We note that the technics we are used here can also be extended to provide the analogous classification problem over the real field for n2,t, t = 1, 2, 3, 4. 5,3, g6,2 5,3, g6,3 5,3 and g6,4 Remark 3. Extensions of n2,3 are treated in [1, Section 3, subsection 3.2]. In this paper, n2,3 is denoted as L5,3 and introduced in Theorem 1, subsection 2.2, through the multiplication table [X0, X1] = X2, [X0, X2] = X3, [X1, X2] = X4; we note that the basis {X0, X1, X2, X3, X4} is just our basis {v0, v1, w0, z1, z2}. The extended Lie algebras appear listed in Propositions 2, 3 and 4. Proposition 2 provides the 1-solvable extensions g6,1 5,3. The first algebra and the third one are just r1,α 2,3 by rescal- ing the basis of g6,2 5,3 given [1] in the folowing way: {−Y ; X0, X1, X2, X3, −X4}. The multiplication table for g6,4 5,3 has a misprint, [Y, X4] = 3X4 must be de- clared instead of 4X4 (otherwise, ad Y is not a derivation, in other words, Ja- cobi identity J(a, b, c) = [[a, b], c] + [[a, b], c] + [[a, b], c] = 0 fails for the terna (a, b, c) = (Y, X1, X2)); by doing the correctio, g6,4 2,3. Proposition 3 provides the 2-sovable extension g7,1 2,3 if we consider the new basis {Y + Z, Y − Z; X0, X1, X2, X3, X4}. 5,3 is just r1 5,3 which is equal to our r4 2,3. The second algebra is equal to r3 2,3 and r2 In Proposition 4, the authors give the nonsolvable extensions g8,1 5,3 and g9,1 5,3. The former algebra is just our g2,t taking in account that {Y, Z, X} is a stan- dard basis of the split 3-dimensional simple Lie algebra ({h, e, f } with [h, e] = 2e, [h, f ] = −2f, [e, f ] = h). But g9,1 5,3 has been erroneously included in the classification, this is not a Lie algebra: J(Y − Y ′, Z, X1) = −X0 and J(Y − Y ′, Z, X4) = −X3. In this case we have not a misprint but an structural error: s = spanhY − Y ′, Z, Xi is just a 3-split simple Lie algebra and L5,3 is ad s- invariant, so L5,3 decomposes as a finite sum of s-irreducible modules; but the Cartan subalgebra C · (Y − Y ′) acts with 5 different eigenvalues on L5,3 (ex- actly 1, −2, −1, 0, −3) with corresponding eigenvectors X0, X1, X2, X3, X4; this fails in any s-module decomposition of L5,3. The final algebra that left in the classification is our Lie algebra g1 2,3 described in Theorem 5.1, item xx). Acknowledgements The authors would like to thank Spanish Government project MTM 2010- 18370-C04-03. Daniel de-la-Concepci´on also thanks support from Spanish FPU grant 12/03224. 17 References References [1] J.M. Ancochea-Berm´udez, R. Campoamor-Stursberg, L. Garc´ıa Vergnolle: Classification of Lie algebras with naturally graded quasi-filiform nilradi- cals, J. Geometry Phys. 61 (2011), 2168-2186. [2] L. Auslander, J. Brezin: Almost algebraic Lie algebras, J. Algebra 8 (1968), 259-313. [3] V.J. Del Barco, G.P. Ovando: Free nilpotent Lie algebras admitting ad- invariant metrics, J. Algebra 366 (2012), 2015-216. [4] P. Benito, D. de-la-Concepci´on: Sage computations of sl2(k)-Levi exten- sions, Tibilisi Math. Journal 5 (2) (2012), 3-17. [5] P. Benito, D. de-la-Concepci´on: On Levi extensions of nilpotent Lie alge- bras, Linear Alg. Appl. 439 (5) (2013), 1441-1457. [6] M.A. Gauger: On the classification of metabelian Lie algebras, Trans. Amer. Math. Soc. 179 (1973) ,293 -- 329. [7] J.E. Humphreys: Introduction to Lie algebras and representation theory, vol. 9, Springer-Verlag, New York, 1972 . [8] N. Jacobson: Lie algebras, Dover Publications, Inc. New York, 1962. [9] G. Leger: Derivations of Lie algebras III, Duke Math. J., 30 (1963), 637- 645. [10] A.I. Malcev: On semisimple subgroups of Lie groups, Izv. Akad. Nauk SSSR Ser. Mat. 8 (1944), 143-174. [11] A.I. Malcev: Solvable Lie algebras, Izv. Akad. Nauk SSSR Ser. Mat. 9 (1945), 329-352; English transl., Amer. Math. Soc. Transl. (1) 9 (1962), 228-262. MR 9, 173. [12] S. Okubo: A generalization of Hurwitz theorem and flexible Lie-admissible algebras, Hadronic Journal 3 (1979), 1-52. [13] S. Okubo: Gauge theory based upon solvable Lie algebras, J. Phys. A: Math. Gen. 31 (1998), 7603-7609. [14] A.L. Onishchick, Y.B. Khakimdzhanov: On semidirect sums of Lie al- gebras, Mat. Zametki 18 (1975), 31-40. English transl.: Math. Notes 18 (1976), 600-604. [15] A.L. Onishchick, E.B. Vinberg: Lie groups and Lie algebras III, Ency- clopaedia of Mathematical Sciences, vol 41, Springer-Verlag, 1994. 18 [16] J.L. Rubin, P. Winternitz: Solvable Lie algebras with Heisenberg ideals, J. Phys. A: Math. Gen. 26 (5) (1993) 1123-1138. [17] T. Sato: The derivations of the Lie algebras, Tohoku. Math. Journ., 23 (1971) 21-36. [18] P. Turkowski: Structure of real Lie algebras, Linear Alg. Applic. 171 (1992), 197-212. 19
1202.3126	1	1202	2012-02-14T20:01:09	Sheaf Structures On a Class of Noncommutative Spectra	[ "math.RA" ]	We introduce a class of noncommutative spectra and give the sheaf structure on the class of noncommutative spectra.	math.RA	math	Sheaf Structures On a Class of Noncommutative Spectra Keqin Liu Department of Mathematics The University of British Columbia Vancouver, BC Canada, V6T 1Z2 September, 2011 Abstract We introduce a class of noncommutative spectra and give the sheaf structure on the class of noncommutative spectra. How to select a class of noncommutative rings such that we can rewrite algebraic geometry in the context of the class of noncommutative rings? The purpose of this paper is to present our answer to this interesting problem. It is our belief that if a ring R is in the right class of non-commutative rings which can be used to rewrite algebraic geometry over commutative rings, then R should have the following Partially Commutative Property: R is a graded ring, and for all homoge- neous elements x and y of R, there exist homogeneous elements x′ and y′ of R such that either xy = yx′ or xy = y′x. In this paper, we introduce a class of noncommutative rings which are called Hu-Liu trirings by us. The trivial extension of a ring by a bimodule over the ring has been used in different mathematical areas for a long time ([3] and [6]). Roughly speaking, a Hu-Liu triring is a trivial extension of a ring by a bimodule such that the bimodule carries an extra commutative ring structure. Many results in commutative algebra and algebraic geometry have the satisfac- tory counterparts in Hu-Liu trirings. The main result of this paper is the sheaf structures on the class of noncommutative spectra based on Hu-Liu trirings. In section 1, we give the basic properties of Hu-Liu trirings. The most important example of Hu-Liu trirings is triquaternions which is defined in section 1. Tri- quaternions, which are regarded as a kind of new numbers by us, can be used to replace complex numbers to develop the counterpart of complex algebraic geometry. In section 2, we introduce prime triideals and prove some basic facts about prime triideals. In section 3, we use prime triideals to characterize the trinilradical. In section 4, we make the class of noncommutative spectra into a 1 topological space by introducing extended Zariski topology. In section 5, we de- fine localization of Hu-Liu trirings. In the last section of this paper, we explain how to define the sheaf structures on the class of noncommutative spectra. Throughout this paper, the word "ring" means an associative ring with an identity. A ring R is also denoted by (R, +, ·) to indicate that + is the addition and · is the multiplication in the ring R. 1 Basic Definitions Let A and B be two subsets of a ring ( R, +, · ). We shall use A + B and AB to denote the following subsets of R A + B := { a + b a ∈ A, b ∈ B }, AB := { ab a ∈ A, b ∈ B }. We now introduce a class of noncommutative rings in the following Definition 1.1 A ring R with a multiplication · is called a Hu-Liu triring if the following three properties hold. (i) There exist a commutative subring R0 of the ring (R, +, ·) and a subgroups R1 of the additive group (R, +), called the even part and odd part of R respectively, such that R = R0 ⊕ R1 (as Abelian groups) and R0R0 ⊆ R0, R0R1 + R1R0 ⊆ R1, R1R1 = 0; (1) (ii) There exists a binary operation ♯ on the odd part R1 such that ( R1, +, ♯ ) is a commutative ring and the two associative products · and ♯ satisfy the triassociative law: x(α ♯ β) = (xα) ♯ β, (α ♯ β)x = α ♯ (βx), where x ∈ R and α, β ∈ R1; (iii) For each x0 ∈ R0, we have R1x0 = x0R1. (2) (3) (4) A Hu-Liu triring R = R0 ⊕ R1, which clearly has the partially commutative property, is sometimes denoted by ( R = R0⊕R1, +, ·, ♯ ), where the associative product ♯ on the odd part R1 is called the local product, and the identity 1♯ of the ring ( R1, +, ♯ ) is called the local identity of the triring R. If ( R = R0 ⊕ R1, +, ·, ♯ ) is a Hu-Liu triring, then R is the trivial extension of a ring R0 by a R0-bimodule R1, and the triassociative law interweaves the 2 noncommutative ring structure on the trivial extension with the commutative ring structure on the odd part. Since a commutative ring is a Hu-Liu triring with zero odd part, the con- cept of Hu-Liu trirings naturally generalizes the concept of commutative rings. The first and the most important example of Hu-Liu trirings which is not a commutative ring is triquaternions whose definition is given in the following example. Example Let Q = R 1 ⊕ R i ⊕ R j ⊕ R k be a 4-dimensional real vector space, where R is the field of real numbers. Then ( Q = Q0 ⊕ Q1, +, ·, ♯ ) is a Hu-liu triring, where Q0 = R 1 ⊕ R i is the even part, Q1 = R j ⊕ R k is the odd part, the ring multiplication · and the local product ♯ are defined by the following multiplication tables: · 1 1 1 i j k k i i j j k k i −1 k −j j −k 0 0 0 0 j ♯ j k j j k k k −j The Hu-Liu triring Q is called the triquaternions. Definition 1.2 Let ( R = R0⊕R1, +, ·, ♯ ) be a Hu-Liu triring with the identity 1, and let I be a subgroup of the additive group of R. (i) I is called a triideal of R if IR + RI ⊆ I, I = (R0 ∩ I) ⊕ (I ∩ R1), and I ∩ R1 is an ideal of the ring ( R1, +, ♯ ). (ii) I is called a subtriring of R if 1 ∈ I, II ⊆ I, I = (R0 ∩ I) ⊕ (I ∩ R1) and I ∩ R1 is a subring of the ring ( R1, +, ♯ ). = (cid:18) R Let I be a triideal of a Hu-Liu triring ( R = R0 ⊕ R1, +, ·, ♯ ). It is clear I (cid:19)0 ⊕(cid:18) R I (cid:19)1 I(cid:19)i with (cid:18) R for i = 0 and 1. We now I(cid:19)1 define a local product on (cid:18) R Ri + I that R I by := I (α + I) ♯ (β + I) := α ♯ β + I for α, β ∈ R1. (5) Then the local product defined by (5) is well-defined, and the triassociative law holds. Therefore, Hu-Liu triring of R with respect to the triideal I. becomes a Hu-Liu triring, which is called the quotient R I 3 Definition 1.3 Let R = R0 ⊕ R1 and R = R0 ⊕ R1 be Hu-Liu trirings. A map φ : R → R is called a triring homomorphism if φ(x + y) = φ(x) + φ(y), φ(xy) = φ(x)φ(y), φ(1R) = 1R, φ(R0) ⊆ R0, φ(R1) ⊆ R1, φ(α ♯ β) = φ(α) ♯ φ(β), φ(1♯) = ¯1♯ where x, y ∈ R, α, β ∈ R1, 1R and 1R are the identities of R and R respectively, and 1♯ and ¯1♯ are the local identities of R and R respectively. A bijective triring homomorphism is called a triring isomorphism. We shall use R ≃ R to indicate that there exists a triring isomorphism from R to R. Let φ be a triring homomorphism from a Hu-Liu triring R to a Hu-Liu triring R. the kernel Kerφ and the image Imφ of φ are defined by Kerφ := { x a ∈ R and φ(x) = 0 } and Clearly, Kerφ is a triideal of R, Imφ is a subtriring of R and Imφ := { φ(x) x ∈ R}. for x ∈ R is a triring isomorphism from the quotient Hu-Liu triring φ : x + Kerφ → φ(x) Imφ of R. If I is triideal of R, then R Kerφ to the subtriring ν : x 7→ x + I for x ∈ R is a surjective triring homomorphism from R the quotient tiring I. The map ν is called the natural triring homomorphism. R I with kernel Proposition 1.1 Let φ be a surjective homomorphism from a Hu-Liu triring R = R0 ⊕ R1 to a Hu-Liu triring R = R0 ⊕ R1. (i) φ(Ri) = Ri for i = 0 and 1. (ii) Let and The map S := { I I is a triideal of R and I ⊇ Kerφ} S := { I I is a triideal of R}. Ψ : I 7→ φ(I) := { φ(x) x ∈ I } is a bijection from S to S, and the inverse map Ψ−1 : S → S is given by Ψ−1 : I 7→ φ−1(I) := { x x ∈ R and φ(x) ∈ I }. 4 (iii) If I is a triideal of R containing Kerφ, then the map for x ∈ R R is a triring isomorphism from the quotient triring I x + I 7→ φ(x) + φ(I) onto the quotient triring R φ(I) . Proof A routine check. We now prove a basic property for Hu-Liu trirings. Proposition 1.2 Let ( R = R0 ⊕ R1, +, ·, ♯ ) be a Hu-Liu triring. If xi ∈ Ri for i ∈ { 0, 1 }, then both Rx0 = R0x0 ⊕ R1x0 and R1 ♯ x1 are triideals of R. Proof Since (R1, +, ♯) is a commutative ring, R1 ♯ x1 is clearly a triideal of R. Using the triassociative law and (4), we have (cid:16)Ri(R0x0 + R1x0)(cid:17)[(cid:16)R0x0 + R1x0)Ri(cid:17) ⊆ R0x0 + R1x0 for i = 0, 1. By (6), R0x0 ⊕ R1x0 is an ideal of the ring ( R, +, · ). Also, R1x0 = R1 ♯ (1♯x0) is obviously an ideal of the commutative ring ( R1, +, ♯ ). Thus, R0x0 ⊕ R1x0 is a triideal of R. (6) The next position gives some operations about triideals in a Hu-Liu triring. Proposition 1.3 Let I, J, Iλ with λ ∈ Λ be triideals of a Hu-Liu triring R. (i) The intersection I ∩ J and the sum Xλ∈Λ Iλ!i have (I ∩ J)i = Ii ∩ Ji and Xλ∈Λ Iλ are triideals of R.Moreover, we (Iλ)i for i = 0, 1. = Xλ∈Λ ♯ J)0 ⊕ (I · · ♯ J := (I · ♯ J)1 = I1 ♯ J1. · ♯ J)1 of I and J is a triideal, (ii) The mixed product I where (I · ♯ J)0 = I0J0 and (I Proof (i) It is clear. (ii) By triassociative law, we have · R0(I · (I ♯ J) ⊆ R0I0J0 + R0(I1 ♯ J1) ⊆ I0J0 + (R0I1) ♯ J1 ⊆ I ♯ J)R0 ⊆ I0J0R0 + (I1 ♯ J1)R0 ⊆ I0J0 + I1 ♯ (J1R0) ⊆ I · ♯ J)R1 ⊆ R1I0J0 + I0J0R1 ⊆ I1J0 + I0J1 ⊆ I ♯ J) + (I · ♯ J. This proves that (ii) holds. · R1(I and R1 ♯ (I1 ♯ J1) ⊆ R1 ♯ I1 ♯ J1 ⊆ I · ♯ J, · ♯ J, · ♯ J 5 2 Prime Triideals We begin this section by introducing the notion of prime triideals. Definition 2.1 Let ( R = R0 ⊕ R1, +, · , ♯ ) be a Hu-Liu triring. An triideal P = P0 ⊕ P1 of R is called a prime triideal if P 6= R and x0y0 ∈ P0 ⇒ x0 ∈ P0 or y0 ∈ P0, x0y1 ∈ P1 ⇒ x0 ∈ P0 or y1 ∈ P1, x1y0 ∈ P1 ⇒ x1 ∈ P1 or y0 ∈ P0, x1 ♯ y1 ∈ P1 ⇒ x1 ∈ P1 or y1 ∈ P1, (9) (10) (7) (8) where xi, yi ∈ Ri for i = 0 and 1. Let ( R = R0⊕ R1, +, · ♯ ) be a Hu-Liu triring. The set of all prime triideals of R is called the trispectrum of R and denoted by Spec♯R. It is clear that Spec♯R = Spec♯ 0R ∪ Spec♯ 1R and Spec♯ 0R ∩ Spec♯ 1R = ∅, where Spec♯ 0R := { P P ∈ Spec♯R and P ⊇ R1 } is called the even trispectrum of R and Spec♯ 1R := { P P ∈ Spec♯R and P 6⊇ R1 } is called the odd trispectrum of R. Let P = P0⊕ P1 be a triideal of a Hu-Liu triring R. Then P ∈ Spec♯ 0R if and only if P1 = R1 and P0 is a prime ideal of the commutative ring ( R0, +, · ). It is also obvious that P ∈ Spec♯ 1R if and only if P0 is a prime ideal of the commu- tative ring ( R0, +, · ), P1 is a prime ideal of the commutative ring ( R1, +, ♯ ), and the (cid:18) R0 P0(cid:19)-bimodule is faithful as both left module and right mod- R P R0 P0 , ule, where the left -module action on is defined by R0 P0 R P (x0 + P0)(y + P ) := x0y + P for x0 ∈ P0 and y ∈ R R0 P0 -module action on is defined by R P and the right (y + P )(x0 + P0) := yx0 + P for x0 ∈ P0 and y ∈ R. Clearly, the even trispectrum Spec♯ basic property of Hu-Liu trirings is that the odd trispectrum Spec♯ not empty provided R1 6= 0. This basic fact is a corollary of the following 0R of a Hu-Liu triring R is not empty. A 1R is always 6 Proposition 2.1 Let ( R = R0 ⊕ R1, +, · ♯ ) be a Hu-Liu triring with R1 6= 0. If P1 is a prime ideal of the commutative ring ( R1, +, ♯ ), then there exists prime ideal P0 of the commutative ring ( R0, +, · ) such that P := P0 ⊕ P1 is a prime triideal of R, and P0 contains every ideal I0 of the ring ( R0, +, · ) which has the property: R1I0 ⊆ P1. Proof Consider the set Ω defined by Ω := { I0 I0 is an ideal of ( R0, +, · ) and R1I0 ⊆ P1 }. Clearly, 1 6∈ I0 if I0 ∈ Ω. Since 0 ∈ Ω, Ω is not empty. The relation of inclusion, ⊆, is a partial order on Ω. Let ∆ be a non-empty totally ordered subset of Ω. Let J0 := [I0∈∆ I0. Then J0 ∈ Ω. Thus J0 is an upper bound for ∆ in Ω. By Zorn's Lemma, the partial order set (Ω, ⊆) has a maximal element P0. We are going to prove that P := P0 ⊕ P1 is a prime triideal of R. Clearly, P = P0 ⊕ P1 is a triideal satisfying (10). Let xi, yi ∈ Ri for i = 0, 1. If x1y0 ∈ P1 and x1 6∈ P1, then x1 ♯ (1♯y0) = x1y0 ∈ P1, which implies that 1♯y0 ∈ P1. Hence, we have R1(P0 + R0y0) ⊆ R1P0 + R1R0y0 ⊆ P1 + R1y0 = P1 + R1 ♯ (1♯y0) ⊆ P1 + R1 ♯ P1 ⊆ P1. (11) Using (11) and the fact that P0 + R0y0 is an ideal of R0, we get P0 + R0y0 ∈ Ω. Since P0 + R0y0 ⊇ P0, we have to have P0 + R0y0 = P0, which implies that y0 ∈ P0. This proves that (9) holds. If x0y1 ∈ P1 and y1 6∈ P1, then (x01♯) ♯ y1 = x0y1 ∈ P1, which implies that x01♯ ∈ P1. It follows from this fact and (4) that R1(P0 + R0x0) ⊆ R1P0 + R1R0x0 ⊆ P1 + R1x0 = P1 + x0R1 = P1 + (x01♯) ♯ R1 ⊆ P1 + P1 ♯ R1 ⊆ P1. (12) Using (12) and the fact that P0 + R0x0 is an ideal of R0, we get P0 + R0x0 ∈ Ω. Since P0 + R0x0 ⊇ P0, we have to have P0 + R0x0 = P0, which implies that x0 ∈ P0. This proves that (8) holds. Finally, if x0y0 ∈ P0, then (1♯x0) ♯ (1♯y0) = 1♯x0y0 ∈ P1. Thus, (1♯x0) ∈ P1 or (1♯y0) ∈ P1, which implies that either x0 ∈ P0 or y0 ∈ P0 by the argument above. This proves that (7) holds. Summarizing what we have proved, P = P0 ⊕ P1 is a prime triideal of R. The following proposition gives another characterization of prime triideals. Proposition 2.2 Let ( R = R0 ⊕ R1, +, ·, ♯ ) be a Hu-Liu triring. The follow- ing are equivalent. 7 (i) P is a prime triideal. (ii) For two triideals I, J of R, I · ♯ J ⊆ P implies that I ⊆ P or J ⊆ P . · Proof (i) ⇒ (ii): Assume that I I1 6⊆ P1. Let yi ∈ Ji with i = 0, 1. ♯ J ⊆ P and I 6⊆ P . Then either I0 6⊆ P0 or If I0 6⊆ P0, then there exists x0 ∈ P0 \ I0. Clearly, x0yi ∈ I ♯ J ⊆ P for i = 0, 1. Since P is a prime triideal, we get yi ∈ P for i = 0, 1 by (7) and (8). Thus J ⊆ P in this case. If I1 6⊆ P1, then there exists x1 ∈ P1 \ I1. Since x1y0 ∈ I ♯ J ⊆ P and ♯ J ⊆ P , we get yi ∈ P for i = 0, 1 by (9) and (10). Thus we also get · · · x1 ♯ y1 ∈ J ⊆ P in this case. (ii) ⇒ (i): Let xi, yi ∈ Pi with for i = 0, 1. By Proposition 1.2, both R0x0 ⊕ R1x0 and R0y0 ⊕ R1y0 are triideals. By the definition of the mixed product of two triideals in Proposition 1.3, we have (R0x0 ⊕ R1x0) · ♯ (R0y0 ⊕ R1y0) = R0x0R0y0 + (R1x0) ♯ (R1y0). Since (R1x0) ♯ (R1y0) = R1x0y0, we get · (R0x0 ⊕ R1x0) ♯ (R0y0 ⊕ R1y0) ⊆ R0x0y0 + R1x0y0. If x0y0 ∈ P0 and x0 6∈ P0, then R0x0 ⊕ R1x0 6⊆ P. (13) (14) It follows from (13) and (14) that R0y0⊕ R1y0 ⊆ P , which implies that y0 ∈ P0. Thus, (7) holds. If x0y1 ∈ P1 and x0 6∈ P0, then (14) holds. By (4), 1♯x0 = x0z1 for some z1 ∈ R1. Hence, we have · (R0x0 ⊕ R1x0) ♯ (R1 ♯ y1) = (R1x0) ♯ (R1 ♯ y1) ⊆ R1 ♯ (1♯x0) ♯ y1 ⊆ R1 ♯ (x0z1) ♯ y1 ⊆ R1 ♯ (x01♯) ♯ z1 ♯ y1 ⊆ R1 ♯ z1 ♯ (x01♯) ♯ y1 ⊆ R1 ♯(cid:0)x0(1♯ ♯ y1)(cid:1) ⊆ R1 ♯ (x0y1). It follows from (13) and (15) that R1 ♯ y1 ⊆ P , which implies that y1 ∈ P . Thus, (8) holds. (15) If x1y0 ∈ P1 and x1 6∈ P1, then R1 ♯ x1 6⊆ P. 8 (16) It follows from x1y0 ∈ P1 that (R1 ♯ x1) · ♯ (R0y0 + R1y0) = (R1 ♯ x1) ♯ (R1y0) = R1 ♯ x1 ♯ R1 ♯ (1♯y0) = R1 ♯ x1 ♯ (1♯y0) = R1 ♯ (x1 ♯ 1♯)y0 = R1 ♯ (x1y0) ⊆ P. (17) By (16) and (17), we get R0y0 + R1y0 ⊆ P , which implies that y0 ∈ P . Thus, (9) holds. If x1 ♯ y1 ∈ P1 and x1 6∈ P1, then (16) holds and (R1 ♯ x1) · ♯ (R1 ♯ y1) = R1 ♯ x1 ♯ y1 ⊆ P. (18) By (16) and (18), R1 ♯ y1 ⊆ P , which implies that y1 ∈ P . Thus, (10) holds. This proves that P is a prime triideal. 3 Trinilradicals Let R = R0 ⊕ R1 be a Hu-Liu triring with the local identity 1♯. For α ∈ R1, the local nth power α♯n is defined by: if n = 0; if n is a positive integer. , The products (xm)(α♯n) and (α♯n)(xm) will be denoted by xmα♯n and α♯nxm respectively, where x ∈ R and α ∈ R1. Proposition 3.1 Let ( R = R0⊕ R1, +, ·, ♯ ) be a Hu-Liu triring with the local identity 1♯. If x, y ∈ R, α, β ∈ R1 and m ∈ Z>0, then (xα) ♯ (yβ) = (xy)(α ♯ β), (αx) ♯ (βy) = (α ♯ β)xy (19) and (xα)♯m = xmα♯m, (αx)♯m = α♯mxm. (20) Proof By the triassociative law, we have (xα) ♯ (yβ) = x (α ♯ (yβ)) = x ((yβ) ♯ α) = (xy)(β ♯ α) = (xy)(α ♯ β) and (αx) ♯ (βy) = ((αx) ♯ β)) y = ((β) ♯ (αx)) y = (β ♯ α)xy = (α ♯ β)xy. Hence, (19) holds. Clearly, (20) follows from (19). 9 α♯n :=  1♯, α ♯ α ♯ ··· ♯ α } {z n Definition 3.1 Let ( R = R0 ⊕ R1, +, ·, ♯ ) be a Hu-Liu triring. (i) An element x of R is said to be trinilpotent if 0 = 0 and x ♯ n xm 1 = 0 for some m, n ∈ Z>0, (21) where x0 and x1 are the even component and the old component of x respectively. (ii) The set of all graded nilpotent elements of R is called the trinilradical of R and denoted by nilrad♯(R) or ♯√0. The ordinary nilradical of a ring ( A, +, · ) is denoted by nilrad(A) or nilrad(A, +,·); that is, nilrad(A) := { x xm = 0 for some m ∈ Z>0 }. If ( R = R0 ⊕ R1, +, ·, ♯ ) is a Hu-Liu triring, then nilrad(R, +,·) = nilrad(R0, +,·) ⊕ R1 and nilrad♯R = nilrad(R0, +,·) ⊕ nilrad(R1, +, ♯ ). (22) Hence, nilrad♯R ⊆ nilrad(R, +,·), i.e., the trinilradical of a Hu-Liu triring R is smaller than the ordinary nilradical of the ring (R, +, ·). Proposition 3.2 Let ( R = R0 ⊕ R1, +, ·, ♯ ) be a Hu-Liu triring. (i) The trinilradical nilrad♯(R) is a triideal of R. (ii) nilrad♯(cid:18) R nilrad♯(R)(cid:19) = 0. Proof (i) By the definition of trinilradicals, we have (nilrad♯R) ∩ R0 = nilrad(R0, +,·) and (nilrad♯R) ∩ R1 = nilrad(R1, +, ♯ ). Using the fact above and (22), we need only to prove {xa, ax} ⊆ nilrad♯R for x ∈ R and a ∈ nilrad♯R. (23) Let x = x0 + x1 ∈ R and a = a0 + a1 ∈ nilrad♯R, where xi, ai ∈ Ri for i = 0, 1. Then we have am 0 = a ♯ m 1 = 0 for some m ∈ Z>0. Since xa = (x0 + x1)(a0 + a1) = x0a0 + (x0a1 + x1a0), (24) (x0a0)m = xm (x0a1)♯m = xm (x1a0)♯m = (x1a0)♯m = x♯m 0 = xm 1 = xm 0 0 = 0, 0 0 = 0, 1 am 0 am 0 a♯m 0 = x♯m 1 0 = 0, 10 we get x0a0 ∈ nilrad(R0, +, ·) and x0a1 + x1a0 ∈ nilrad♯(R1, +, ♯). It follows from (24) and (25) that xa ∈ nilrad♯R. Similarly, we have ax ∈ (25) nilrad♯R. This proves (i). (ii) If x + nilrad♯R ∈ such that R nilrad♯R , then there exist positive integers m and n and xm 0 + nilrad♯R =(cid:0)x0 + nilrad♯R(cid:1)m 1 + nilrad♯R =(cid:0)(x1 + nilrad♯R)1(cid:1) ♯ n x ♯ n where x = x0 + x1 and xi ∈ Ri for i = 0, 1. Hence, we get 1 ∈ nilrad♯R, 0 ∈ nilrad♯R and x ♯ n xm = nilrad♯R = nilrad♯R, which imply that xmu 0 = (xm 0 )u = 0 and x ♯ (nv) 1 = (x ♯ n 1 ) ♯ v = 0 for some u, v ∈ Z>0. Thus, x = x0 + x1 ∈ nilrad♯R or x + nilrad♯R is the zero element of . This proves (ii). R nilrad♯R If I is a triideal of a Hu-Liu triring R = R0 ⊕ R1, then the trinilradical ♯√I of I is defined by ♯√I := { x ∈ R xm 0 ∈ I0 and x ♯ n 1 ∈ I1 for some m, n ∈ Z>0 }, where x = x0 + x1, xi ∈ Ri and Ii = I ∩ Ri for i = 0, 1. Since nilrad♯(cid:18) R I (cid:19) = ♯√I I , ♯√I is a triideal of R. A triideal I of a Hu-Liu triring R is called a radical triideal if ♯√I = I. We now characterize the trinilradical of a Hu-Liu triring by using prime triideals. Proposition 3.3 Let ( R = R0 ⊕ R1, +, · , ♯ ) be a Hu-Liu triring. The trinil- radical of R is the intersection of the prime triideals of R. Proof Let x = x0 + x1 be any element of nilrad♯(R), where x0 ∈ R0 and x1 ∈ R1. Then xm 1 = 0 for some m, n ∈ Z>0. Let P = P0 ⊕ P1 be any prime triideal of R. Since xm 0 = 0 and x ♯ n 0 = 0 ∈ P0, we have x0 ∈ P0 11 (26) by (7). If P 6⊇ R1, then P1 is a prime ideal of the commutative ring ( R1, +, ♯ ). Using this fact and x ♯ n 1 = 0, we get x1 ∈ P1. (27) If P ⊇ R1, then (27) is obviously true. x = x0 + x1 ∈ P . This proves that It follows from (26) and (27) that nilrad♯(R) ⊆ \P ∈Spec♯R P. Conversely, we prove that z 6∈ nilrad♯(R) ⇒ z 6∈ \P ∈Spec♯R P. (28) (29) Case 1: zm 6= 0 for all m ∈ Z>0, in which case, zm 6∈ R1 for all m ∈ Z>0. ; that is, Hence, z + R1 is not a nilpotent element of the commutative ring R R1 R1(cid:19) = z + R1 6= nilrad(cid:18) R R1(cid:19) is the ordinary spectrum of the commutative ring(cid:18) R \I R1(cid:1) ∈Spec(cid:0) R R1(cid:19) , (cid:18) I R1 R1 where Spec(cid:18) R , +, ·(cid:19). such that R R1 Hence, there exists a prime ideal z 6∈ I. Since I is a prime triideal of R, (29) holds in this case. of the commutative ring I R1 Case 2: zm = 0 for some m ∈ Z>0. Let z = z0 + z1 with z0 ∈ R0 and 0 = 0, 6= 0 for all n ∈ Z>0 in this case. We now consider the r1 for some r1 ∈ R1 by (4). Thus, zm z1 ∈ R1. Then 0 = zm = zm which implies that z ♯ n following set 0 + zm−1 1 0 T :=n J (cid:12)(cid:12)(cid:12) J is a triideal of R and z ♯ n 1 6∈ J for all n ∈ Z>0o . Since {0} ∈ T , T is nonempty. Clearly, ( T, ⊆ ) is a partially order set, where ⊆ is the relation of set inclusion. If { Jλ λ ∈ Λ } is a nonempty totally ordered subset of T , then Sλ∈Λ Jλ is an upper bound of { Jλ λ ∈ Λ } in T . By Zorn's Lemma, the partially ordered set ( T, ⊆ ) has a maximal element P . We are going to prove that P is a prime triideal of R. Let x = x0 + x1 and y = y0 + y1 be two elements of R, where xi, yi ∈ Ri for i = 0, 1. First, if x0 6∈ P0 and y0 6∈ P0, then P ⊂ P + Rx0 and P ⊂ P + Ry0. (30) 12 Since both P + Rx0 and P + Ry0 are triideals of R by Proposition 1.2, (30) implies that z ♯ u 1 ∈ P + Rx0 and z ♯ v 1 ∈ P + Ry0 or z ♯ u 1 ∈ P1 + R1x0 and z ♯ v 1 ∈ P1 + R1y0 for some u, v ∈ Z>0. Thus, we have z ♯ (u+v) 1 = z ♯ u 1 ♯ z ♯ v 1 ∈ (P1 + R1x0) ♯ (P1 + R1y0) ⊆ P1 ♯ P1 + P1 ♯ (R1y0) + (R1x0) ♯ P1 } This is a subset of P {z ⊆ P + R1x0y0, +(R1x0) ♯ (R1y0) which implies that x0y0 6∈ P . This proves that x0 6∈ P0 and y0 6∈ P0 ⇒ x0y0 6∈ P0. Next, if x0 6∈ P0 and y1 6∈ P1, then P ⊂ P + Rx0 and P ⊂ P + R1 ♯ y1. (31) (32) Since both P + Rx0 and P + R1 ♯ y1 are triideals of R by Proposition 1.2, (32) implies that there exist some positive integers s and t such that z ♯ s 1 ∈ P + Rx0 and z ♯ t 1 ∈ P + R1 ♯ y1 z ♯ s 1 ∈ P1 + R1x0 and z ♯ t 1 ∈ P1 + R1 ♯ y1. (33) or It follows that z ♯ (s+t) 1 = z ♯ s 1 ♯ z ♯ t 1 ∈ (P1 + R1x0) ♯ (P1 + R1 ♯ y1) ⊆ P1 ♯ P1 + P1 ♯ R1 ♯ y1 + (R1x0) ♯ P1 } {z ⊆ P1 + R1 ♯ (x0y1), This is a subset of P1 +(R1x0) ♯ R1 ♯ y1 which implies that x0y1 6∈ P1. This proves that x0 6∈ P0 and y1 6∈ P1 ⇒ x0y1 6∈ P1. Similarly, we have y1 6∈ P1 and x0 6∈ P0 ⇒ y1x0 6∈ P1 13 (34) (35) and x1 6∈ P1 and y1 6∈ P1 ⇒ x1y1 6∈ P0. (36) By (31), (34), (35) and (36), P is a prime triideal. Since z1 6∈ P , (29) also holds in Case 2. It follows from (28) and (29) that Proposition 3.3 is true. The next proposition is a corollary of Proposition 3.3. Proposition 3.4 If I is a triideal of a Hu-Liu triring R and I 6= R, then ♯√I = \ P ∈ Spec♯R and P ⊇ I P. Proof By Proposition 3.3, we have x ∈ ♯√I ⇔ x + I ∈ nilrad♯(cid:18) R \ ⇔ x ∈ I (cid:19) = P ∈ Spec♯R and P ⊇ I P I \P I ∈ Spec♯(cid:0) R I(cid:1) P. 4 Extended Zariski Topology Let ( R = R0 ⊕ R1, +, ·, ♯ ) be a Hu-Liu triring. For a triideal I of R, we define a subset V(I) of Spec♯R by V ♯(I) := { P P ∈ Spec♯R and P ⊇ I }. (37) Proposition 4.1 Let R be a Hu-Liu triring. (i) V ♯(0) = spec♯R and V ♯(R) = ∅. (ii) V ♯(I) ∪ V ♯(J) = V ♯(I ∩ J) = V ♯(I of R. · ♯ J), where I and J are two triideals (iii) \λ∈Λ of R. V ♯(cid:0)I(λ)(cid:1) = V ♯ Xλ∈Λ I(λ)!, where (cid:8) I(λ)(cid:12)(cid:12) λ ∈ Λ(cid:9) is a set of triideals 14 Proof Since (i) and (iii) are clear, we need only to prove (ii). · Since I we have V ♯(I ♯ J ⊆ I ∩ J ⊆ I, we get V ♯(I · ♯ J) ⊇ V ♯(I ∩ J) ⊇ V ♯(I). Similarly, · ♯ J) ⊇ V ♯(I ∩ J) ⊇ V ♯(J). Thus, we get V ♯(I) ∪ V ♯(J) ⊆ V ♯(I ∩ J) ⊆ V ♯(I · ♯ J). (38) Conversely, if P ∈ V ♯(I · ♯ J), then I · I ⊆ P or J ⊆ P . Hence, P ∈ V ♯(I) ∪ V ♯(J). This proves that ♯ J ⊆ P . By Proposition 2.2, we get V ♯(I · ♯ J) ⊆ V ♯(I) ∪ V ♯(J). (39) It follows from (38) and (39) that (ii) is true. Let ( R = R0 ⊕ R1, +, · , ♯ ) be a Hu-Liu triring. By Proposition 4.1, the collection V ♯ := { V ♯(I) I is a triideal of R } of subsets of Spec♯R satisfies the axioms for closed sets in a topological space. The topology on Spec♯R having the elements of V ♯ as closed sets is called the extended Zariski topology. The collection D♯ := { D♯(I) I is a triideal of R } consists of the open sets of the extended Zariski topology on Spec♯R, where D♯(I) := Spec♯R \ V ♯(I) = { P P ∈ Spec♯R and P 6⊇ I }. For xi ∈ Ri with i ∈ {0, 1}, both Rx0 and R1 ♯ x1 are triideals of R by Proposition 1.2. Let D♯(x0) := D♯(Rx0), D♯(x1) := D♯(R1 ♯ x1). If I0 and I1 are the even part and odd part of an triideal I, then D♯(I) = [xi∈Ii, i=0,1 D♯(xi). Thus, { D♯(xi) xi ∈ Ri with i = 0, 1 } forms an open base for the extended Zariski topology on Spec♯R. Each D♯(xi) is called a basic open subset of Spec♯R. Clearly, D♯(0) = ∅, D♯(1) = Spec♯R, D♯(1♯) = Spec♯ 1R, and D♯(x1) ⊆ Spec♯ Proposition 4.2 Let I and J be triideals of a Hu-Liu triring. 1R for x1 ∈ R1. 15 (i) V ♯(I) ⊆ V ♯(J) if and only if (ii) V ♯(I) = V ♯( ♯√I). ♯√J ⊆ ♯√I. Proof (i) If V ♯(I) ⊆ V ♯(J), then P ⊇ I implies that P ⊇ J for P ∈ Spec♯R. Thus, we have { P P ∈ Spec♯R and P ⊇ J } By Proposition 3.4, we get = { P P ∈ Spec♯R and P ⊇ I }[{ P P ∈ Spec♯R, P ⊇ J and P 6⊇ I }. P! ♯√J = P = P!\ P ⊇ J P ⊇ I \P ∈ spec♯R P ⊇ J, P 6⊇ I P = ♯√J . \P ∈ spec♯R \P ∈ spec♯R \P ∈ spec♯R P ⊇ I ♯√J ⊆ ♯√I, then for any Q ∈ V ♯(I), we have Q ⊇ I and \P ∈ Spec♯R \P ∈ Spec♯R P = ♯√I ⊇ ♯√J = P ⊇ J, P ⊇ I P ⊇ J ⊆ Conversely, if Q ⊇ which proves that Q ∈ V ♯(J). Thus, we get V ♯(I) ⊆ V ♯(J). (ii) Since ♯p ♯√I = ♯√I, (ii) follows from (i). Definition 4.1 Let X be a topological space. (i) A closed subset F of X is reducible if F = F(1) ∪ F(2) for proper closed subsets F(1), F(2) of X. We call a closed subset F irreducible if it is not reducible. (ii) X is quasicompact if given an arbitrary open covering { U(i) i ∈ I } of X, there exists a finite subcovering of X, i.e., there exist finitely many members U(i1), . . ., U(in) of { U(i) i ∈ I } such that X = U(i1)∪···∪ U(in). Using the topological concepts above, we have the following Proposition 4.3 Let ( R = R0 ⊕ R1, +, ·, ♯ ) be a Hu-Liu triring. 16 (i) The trispectrum Spec♯R is quasicompact. (ii) Both Spec♯ 0R and Spec♯ 1R are quasicompact subsets of Spec♯R. (iii) If I is a triideal of R, then the closed subset V ♯(I) of Spec♯R is irreducible if and only if ♯√I is a prime triideal. Proof (i) Let { D♯(I(i)) i ∈ ∆} be an open covering of Spec♯R, where I(i) = I(i)0⊕I(i)1 is a triideal with the even part I(i)0 and the odd part I(i)1 for each i ∈ ∆. Thus, Spec♯R =Si∈∆ D♯(I(i)) = D♯(Pi∈∆ I(i)). Hence, V ♯(Pi∈∆ I(i)) = ∅. If 1 6∈ (Pi∈∆ I(i))0 = Pi∈∆ I(i)0, then (Pi∈∆ I(i))0 is a proper ideal of the commutative ring (R0, +, ·). Hence, there exists a maximal ideal M0 of the ring (R0, +, ·) such that (Pi∈∆ I(i))0 ⊆ M0. Since M0 ⊕ R1 is a prime triideal of R and Pi∈∆ I(i) ⊆ M0 ⊕ R1, we get that M0 ⊕ R1 ∈ V ♯(Pi∈∆ I(i)) = ∅, which is impossible. Therefore, 1 ∈ (Pi∈∆ I(i))0 = Pi∈∆ I(i)0, which implies that x(i1)0 +x(i2)0 +···+x(in)0 = 1 for some positive integer n and x(ik)0 ∈ I(ik)0 with ik ∈ ∆ and n ≥ k ≥ 1. It follows that D♯(x(ik )0) ⊆ D♯(I(ik)) and n n Spec♯R = [i∈∆ [k=1 D♯(I(i)) ⊇ D♯(I(ik )) ⊇ D♯(Rx(ik)0) = D♯ n Rx(ik)0! = D♯(R) = Spec♯R, Xk=1 [k=1 D♯(x(ik)0) [k=1 n = which implies that Spec♯R =Sn k=1 D♯(I(ik )). (ii) Note that a closed subset of a quasicompact topological space is a quasi- 0R = V ♯(R1) is a closed subset of the quasicompact compact subset. Since Spec♯ topological space Spec♯R, Spec♯ 0R is a quasicompact subset. Since Spec♯ It is well-known that a subset C of a topological space X is a quasicompact subset of X if and only if every covering of C by open subsets of X has a finite subcovering. Hence, in order to prove that Spec♯ 1R is a quasicompact subsets 1R =Sj∈Γ D♯(J(j)) for triideals J(j) of Spec♯R, it suffices to prove that if Spec♯ of R, then there exists a positive integer m such that Spec♯ for some j1, ···, jk ∈ Γ. 1R = Sj∈Γ D♯(J(j)) = D♯(Pj∈Γ J(j)), we have V ♯(Pj∈Γ J(j)) = 0R. If (Pj∈Γ J(j))1 6= R1, then there exists a Spec♯R \ D♯(Pj∈Γ J(j)) = Spec♯ maximal ideal N1 of the commutative ring (R1, +, ♯) such that (Pj∈Γ J(j))1 ⊆ (R0, +, ·) such that N0 ⊇ (Pj∈Γ J(j))0 and N0 ⊕ N1 is a prime triideal of R. Thus, N0 ⊕ N1 ∈ V ♯(Pj∈Γ J(j)) = Spec♯ N1 6= R1. This proves that (Pj∈Γ J(j))1 = R1. Hence, we have y(j1)1 + y(j2)1 + N1. By Proposition 2.1, there exists an ideal N0 of the commutative ring 0R, which is impossible because 1R =Sm k=1 D♯(J(jk)) 17 ··· + y(jm)1 = 1♯ for some positive integer m and y(jk)1 ∈ J(jk)1 with jk ∈ Γ and m ≥ k ≥ 1. It follows that D♯(y(jk)1) ⊆ D♯(J(jk)) and D♯(J(j)) ⊇ D♯(y(jk)1) = m [k=1 D♯(R1 ♯ y(jk)1) m [k=1 Spec♯ 1R = [j∈Γ = D♯ m Xk=1 which implies that Spec♯ R1 ♯ y(jk)1! = D♯(R1) = Spec♯ 1R =Sm k=1 D♯(y(jk)1). 1R, (iii) By Proposition 4.2, we may assume I = ♯√I in the following proof. First, we prove that if V ♯(I) is irreducible , then I is a prime triideal. Suppose that a0b0 ∈ I for some a0, b0 ∈ R0. Let J(1) = I + Ra0 = (I0 + R0a0) ⊕ (I1 + R1a0) and K(1) = I + Rb0 = (I0 + R0b0) ⊕ (I1 + R1b0). Then both J(1) and K(1) are triideals of R and J(1) · ♯ K(1) = (I0 + R0a0)(I0 + R0b0) + (I1 + R1a0) ♯ (I1 + R1b0) · ⊆ I + R0a0R0b0 + (R1a0) ♯ (R1b0) ⊆ I + R0a0b0 + R1a0b0 ⊆ I. ♯ K(1)) ⊇ V ♯(I) by Proposition 4.1 Hence, we get V ♯(J(1))∪V ♯(K(1)) = V ♯(J(1) (ii). It is clear that V ♯(J(1)) ⊆ V ♯(I) and V ♯(K(1)) ⊆ V ♯(I). Hence, we get that V ♯(J(1)) ∪ V ♯(K(1)) ⊆ V ♯(I). Thus we have V ♯(J(1)) ∪ V ♯(K(1)) = V ♯(I). Since V ♯(I) is irreducible, V ♯(J(1)) = V ♯(I) or V ♯(K(1)) = V ♯(I), which imply that I = ♯pJ(1) ⊇ J(1) ∋ a0 or I = ♯pK(1) ⊇ K(1) ∋ b0. This proves that a0b0 ∈ I =⇒ a0 ∈ I or b0 ∈ I for a0, b0 ∈ R0. (40) Suppose that a0b1 ∈ I for some a0 ∈ R0 and b1 ∈ R1. Using the triideals J(2) = I + Ra0 = (I0 + R0a0) ⊕ (I1 + R1a0) K(2) = I + R1 ♯ b1 = I0 ⊕ (I1 + R1 ♯ b1), and we have J(2) · ♯ K(2) = (I0 + R0a0)I0 + (I1 + R1a0) ♯ (I1 + R1 ♯ b1) ⊆ I + (R1a0) ♯ R1 ♯ b1 ⊆ I + (a0R1) ♯ b1 ♯ R1 ⊆ I + a0(R1 ♯ b1) ♯ R1 ⊆ I + a0(b1 ♯ R1) ♯ R1 ⊆ I + (a0b1) ♯ R1 ♯ R1 ⊆ I, 18 which implies that I = ♯pJ(2) ⊇ J(2) ∋ a0 or I = ♯pK(2) ⊇ K(2) ∋ b1. This proves that a0b1 ∈ I =⇒ a0 ∈ I or b1 ∈ I for a0 ∈ R0 and b1 ∈ R1. Similarly, we have a1b0 ∈ I =⇒ a1 ∈ I or b0 ∈ I for a1 ∈ R1 and b0 ∈ R0 and a1 ♯ b1 ∈ I =⇒ a1 ∈ I or b1 ∈ I for a1, b1 ∈ R1. By (40), (41), (42) and (43), I is a prime triideal. (41) (42) (43) Next, we prove that if I is a prime triideal, then V ♯(I) is irreducible. Sup- pose that V ♯(I) = V ♯(J) ∪ V ♯(K), where J and K are triideals of R. Using Proposition 4.2 (ii), we can assume that ♯√J = J and ♯√K = K. In this case, we have and V ♯(J) ⊆ V ♯(I) =⇒ I = ♯√I ⊆ ♯√J = J V ♯(K) ⊆ V ♯(I) =⇒ I = ♯√I ⊆ ♯√K = K. By Proposition 4.1 (ii), we have V ♯(I) = V ♯(J) ∪ V ♯(K) = V ♯(J ♯ K ⊆ ♯√I = I. Since I is a prime triideal, fact and Proposition 4.2 (i) give J we get J ⊆ I or K ⊆ I by Proposition 2.2. Hence, I = J or I = K. Thus, V ♯(I) = V ♯(J) or V ♯(I) = V ♯(K). This proves that V ♯(I) is irreducible. · · ♯ K). This 5 Localization of Hu-Liu Trirings Let ( R = R0 ⊕ R1, +, ·, ♯ ) be a Hu-Liu triring with the identity 1 and the local identity 1♯. A subset S of R is called a multiplicative subset if 1♯ ∈ S1 := S ∩ R1, S = S0 ∪ S1 1 ∈ S0 := S ∩ R0, 0 6∈ S, and s0t0 ∈ S0, s0s1 ∈ S1, s1s0 ∈ S1, s1 ♯ t1 ∈ S1, s0S1 = S1s0, (44) where si, ti ∈ Si and i = 0, 1. Given a multiplicative subset S of a Hu-Liu triring R, we define a relation in the Cartesian product (R0 × S0) × (R1 × S1) as follows: ((a0, s0), (a1, s1)) ∼ ((b0, t0), (b1, t1)) ⇔ u0(a0t0 − b0s0) = 0 and u1 ♯ (a1 ♯ t1 − b1 ♯ s1) = 0 (45) 19 for some ui ∈ Si, where ai, bi ∈ Ri, si, ti ∈ Si and i = 0, 1. It is clear that ∼ is an equivalence relation. Let(cid:18) a0 a1 class containing ((a0, s0), (a1, s1)); that is, , s0 (a′ 1, s′ i, s′ u0(a0s′ u1 ♯ (a1 ♯ s′ s1(cid:19) be the equivalence   ((a0, s0), (a1, s1)) ∈ (R0 × S0) × (R1 × S1)(cid:27) i) ∈ Ri × Si, 0 − s0a′ 0) = 0, 1 − s1 ♯ a′ for some ui ∈ Si with i = 0, 1. 1)) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 1) = 0 . (cid:18) a0 s0 Let S , a1 s1(cid:19) :=  −1R :=(cid:26)(cid:18) a0 s0 ♯ ((a′ 0, s′ 0), (a′ , a1 s1(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) ♯ ♯ ♯ −1R)0 ⊕ (S −1R(cid:17)1 −1R)1 is a are given ♯ −1R = (S be the set of all equivalence classes. Then S ♯ ♯ by Hu-Liu triring, where the even part(cid:16)S −1R(cid:17)0 and odd part(cid:16)S 1♯(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) (a0, s0) ∈ R0 × S0(cid:27) , s1(cid:19)(cid:12)(cid:12)(cid:12)(cid:12) (a1, s1) ∈ R1 × S1(cid:27) (cid:16)S −1R(cid:17)0 −1R(cid:17)1 (cid:16)S :=(cid:26)(cid:18) a0 :=(cid:26)(cid:18) 0 a1 s0 1 0 , , ♯ and the addition +, the ring multiplication · and the local product ♯ are given by , b1 t1(cid:19) :=(cid:18) a0t0 + s0b0 s0t0 , a1 ♯ t1 + s1 ♯ b1 s1 ♯ t1 (cid:19) , , (a0b1) ♯ (s1t0) + (s0t1) ♯ (a1b0) (s0t1) ♯ (s1t0) (cid:19) , , t0 s0 a1 a1 s1(cid:19) +(cid:18) b0 (cid:18) a0 (cid:18) a0 s1(cid:19) ·(cid:18) b0 (cid:18) 0 s0 b1 t0 1 , , s0t0 t1(cid:19) :=(cid:18) a0b0 s1(cid:19) ♯ (cid:18) 0 a1 1 , b1 t1(cid:19) :=(cid:18) 0 1♯(cid:19) is the identity of the ring S a1 ♯ b1 s1 ♯ t1(cid:19) . −1R, and (cid:18) 0 0 1 1 , ♯ , 1♯ 1♯(cid:19) is the local and Clearly, (cid:18) 1 identity. 1 , Let i♯S R : R → S ♯ −1R be the map defined by i♯S R : a = a0 + a1 7→(cid:16) a0 1 , a1 1♯(cid:17) , where ai ∈ Ri for i = 0 and 1. Then i♯S −1R. S ♯ R is a triring homomorphism from R to 20 ♯ respect to S, and i♯S −1R, i♯S The pair(cid:16)S R(cid:17) constructed above is called the localization of R with R is called the canonical triring homomorphism from R −1R. The next proposition gives the universal mapping characterization of to S ♯ ♯ −1R, i♯S R(cid:17). the localization(cid:16)S Proposition 5.1 Let R = R0 ⊕ R1 be a Hu-Liu triring. If S = S0 ∪ S1 is a R(cid:17) of R with respect multiplicative subset of R, then the localization (cid:16)S to S has the following two properties. (i) For any si ∈ Si with i = 0 and 1, i♯S R (s0) is invertible in the commutative −1R, i♯S ♯ R (s1) is invertible in the commutative ring , +,·(cid:19) and i♯S ♯ ring (cid:18)(cid:16)S (cid:18)(cid:16)S −1R(cid:17)1 −1R(cid:17)0 , +, ♯(cid:19). ♯ (ii) If ψ : R → R is a triring homomorphism from the Hu-Liu triring R to a Hu-Liu triring R such that ψ(s0) is invertible in the commutative ring (R0, +,·) and ψ(s1) is invertible in the commutative ring (R1, +, ♯ ) for any si ∈ Si with i = 0 and 1, then there exists a unique triring homomor- phism ψ : S −1R → R such that ψ = ψ i♯S R . ♯ Proof A direct computation. tions of R are very useful in the study of Hu-Liu trirings: Let R = R0 ⊕ R1 be a Hu-Liu triring. The following three types of localiza- Type 1. If P = P0⊕P1 is a triideal of R with P1 6= R1, then (R0\P0)S(R1\P1) is a multiplicative subset of R. The localization of R with respect to (R0 \ P0) ∪ (R1 \ P1) is called the localization of R at P and is denoted by R♯ P . Type 2. If f0 ∈ R0 and D♯(f0)T spec♯ 0 n ∈ Z≥0 } and 1R 6= ∅, then T (f0) = T0(f0) ∪ T1(f0) is a multiplicative subset of R, where T0(f0) := { f n 0 n ∈ Z≥0 and z1 ∈ R1 \ P for all P ∈ D♯(f0)T spec♯ T1(f0) := { z1f n 1R }. The localization of R with respect to T (f0) = T0(f0)∪ T1(f0) is called the localization of R at f0 and is denoted by R♯ f0 0 (cid:12)(cid:12)(cid:12)(cid:12) 0 ∈ T1(f0)(cid:27) . m, n ≥ 0, ai ∈ Ri for i = 0, 1 and z1f m =(cid:26) a0 . Clearly, we have z1f m R♯ f0 f n 0 a1 + 21 Type 3. If f1 ∈ R1 and f1 is not trinilpotent, then { 1}S{ f ♯ m plication subset of R. The localization of R with respect to { 1}S{ f ♯ m is called the localization of R at f1 and is denoted by R♯ f1 have }m≥0 is multi- }m≥0 . Thus, we 1 1 R♯ f1 =( a0 + a1 f ♯ m 1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) m, n ≥ 0 and ai ∈ Ri for i = 0, 1) . 6 Sheaf Structures on Trispectra We begin this section with the following Definition 6.1 Let X be a topological space. A presheaf P of Hu-Liu trirings on X assigns a Hu-Liu triring P(U ) to each open set U of X and a triring homomorphism: ρU V : P(U ) → P(V ), called the restriction map, to each inclusion of open sets V ⊆ U , subject to the conditions: (i) P(0) = ∅; (ii) ρU U : P(U ) → P(U ) is the identity map; (iii) If W ⊆ V ⊆ U are three open subsets, then ρUW = ρV W ρU V . If P is a presheaf of Hu-Liu trirings on a topological space X, then an element s ∈ P(U ) is called a section of P over the open set U . Definition 6.2 A presheaf P of Hu-Liu trirings on a topological space X is called a sheaf of Hu-Liu trirings if it satisfies the following two conditions for an arbitrary open subset U of X and an arbitrary open covering { U(i) i ∈ ∆} of U : (i) If s ∈ P(U ) and ρU U(i) (ii) If we have elements s(i) ∈ P(U(i)) for each i ∈ ∆ having the property (s(j)) for i, j ∈ ∆, then there is an (s) = 0 for all i ∈ ∆, then s = 0; (s(i)) = ρU(j) ,U(i) ∩U(j) that ρU(i) ,U(i) ∩U(j) element s ∈ P(U ) such that ρU U(i) (s) = s(i) for each i ∈ ∆. Let ( R = R0 ⊕ R1, +, ·, ♯ ) be a Hu-Liu triring with the identity 1 and the local identity 1♯. If f0 ∈ R0 and f1 ∈ R1, then we define O(D♯(f0)) :=  (R0)f0 ⊕ R0 f0 ⊕ R♯ R♯ 0 1♯f0 if f0 6∈ ♯√0 and D♯(f0)T spec♯ if f0 6∈ ♯√0 and D♯(f0)T spec♯ if f0 ∈ ♯√0 1R = ∅, 1R 6= ∅, 22 and O(D♯(f1)) :=( R♯ 0 f1 if f1 6∈ ♯√0, if f1 ∈ ♯√0 , where (R0)f0 is the ordinary localization of the commutative ring (R0, +, ·) with respect to the multiplication set { f n 0 n ∈ Z≥0 }, and (R0)f0 ⊕ R0 is a Hu-Liu triring with zero odd part. We then have the following result. Proposition 6.1 Let ( R = R0⊕R1, +, ·, ♯ ) be a Hu-Liu triring. If fi, gi ∈ Ri for i = 0, 1 and D♯(gi) ⊆ D♯(fj) for i, j ∈ { 0, 1 }, then there exist a triring homomorphism ρfj ,gi : O(D♯(fj)) → O(D♯(gi)) such that (i) ρfj ,fj = id D♯ (fj ) , provided D♯(hk) ⊆ D♯(gi) ⊆ D♯(fj), = ρgi,hk (ii) ρfj ,hk ρfj ,gi where i, j, k ∈ { 0, 1 }. Proof If gi ∈ ♯√0, then we define ρfj ,gi homomorphism, i.e., : O(D♯(fj)) → O(D♯(gi)) to be the zero ρfj ,gi (x) := 0 for x ∈ O(D♯(fj)) and gi ∈ ♯√0. (46) We now define ρfj ,gi cases. : O(D♯(fj)) → O(D♯(gi)) for gi 6∈ ♯√0 and fj 6∈ ♯√0 by Case 1: i = 0, j = 0, in which case, we have 0 = r0f0 D♯(Rg0) = D♯(g0) ⊆ D♯(f0) = D♯(Rf0) =⇒ V ♯(Rg0) ⊇ V ♯(Rf0) =⇒ ♯pRg0 ⊆ ♯pRf0 =⇒ gu for some u ∈ Z>0 and r0 ∈ R0. Since D♯(g0)T spec♯ 1R ⊆ D♯(f0)T spec♯ Case 1(i): D♯(g0)T spec♯ In this case, using (47) and universal property of the ordinary localization (R0)f0 , we get a triring homomorphism 1R = ∅ and D♯(f0)T spec♯ 1R = ∅. (by Proposition 4.2 (i)) (47) 1R, we have three subcases. ρf0 ,g0 such that : O(D♯(f0)) = (R0)f0 ⊕ R0 → (R0)g0 ⊕ R0 = O(D♯(g0)) ρf0 ,g0(cid:18) a0 , b0(cid:19) =(cid:18) a0rn for a0, b0 ∈ R0. , b0(cid:19) 0 gnu 0 f n 0 (48) Case 1(ii): D♯(g0)T spec♯ have a triring homomorphism : O(D♯(f0)) = R♯ 1R = ∅ and D♯(f0)T spec♯ f0 ⊕ R♯ ρf0 ,g0 1♯f0 → (R0)g0 ⊕ R0 = O(D♯(g0)) 1R 6= ∅. In this case, we 23 such that where z1f m + a1 f n 0 , b0 + z1f m 0 b1 1♯f k ρf0,g0(cid:18)(cid:18) a0 0(cid:19)(cid:19) =(cid:18) a0rn , b0(cid:19) , 0 ∈ T1(f0), n, m, k ∈ Z≥0, ai, bi ∈ Ri and i = 0, 1. 1R 6= ∅ and D♯(f0)T spec♯ and R♯ 0 gnu 0 and universal property of the localization R♯ f0 homomorphism φ : R♯ g0 and ψ : R♯ 1R 6= ∅. Using (47) , we get two triring (49) (50) (51) (52) (53) Case 1(iii): D♯(g0)T spec♯ f0 → R♯ φ(cid:18) a0 ψ(cid:18)b0 + f n 0 + and 1♯f0 such that 1♯g0 1♯f0 → R♯ a0rn 0 gnu 0 + a1rm 0 z1gmu 0 a1 z1f m 0 (cid:19) = 0(cid:19) = b0 + b1 1♯f k b1rk 0 1♯gku 0 , where z1f m 0 ∈ T1(f0), n, m, k ∈ Z≥0, ai, bi ∈ Ri and i = 0, 1. By (50) and (51), we get a triring homomorphism ρf0 ,g0 such that : O(D♯(f0)) = R♯ f0 ⊕ R♯ 1♯f0 → R♯ g0 ⊕ R♯ 1♯g0 = O(D♯(g0)) , b0 + ρf0,g0(cid:18)(cid:18) a0 f n 0 a1 z1f m 0 + a1 f n 0 = (cid:18)φ(cid:18) a0 = (cid:18) a0rn 0 gnu 0 + 0 (cid:19) , ψ(cid:18)b0 + 0 (cid:19) , b1rk 0 1♯gku , b0 + z1f m a1rm 0 z1gmu 0 + b1 1♯f k b1 1♯f k 0(cid:19)(cid:19) 0(cid:19)(cid:19) where z1f m 0 ∈ T1(f0), n, m, k ∈ Z≥0, ai, bi ∈ Ri and i = 0, 1. Case 2: i = 0 and j = 1, in which case, we have D♯(g0) ⊆ D♯(f1) =⇒ g0 ∈ ♯√0. Hence, ρf1 ,g0 = 0 in this case according to the definition given at the beginning. Case 3: i = 1 and j = 0, in which case, D♯(g1) ⊆ D♯(f0) 1 = r1f0 and D♯(f0)T spec♯ phism ρf0 ,g1 for some v ∈ Z>0 and r1 ∈ R1 =⇒ g♯v 1R ⊇ D♯(g1) 6= ∅ It follows that we have a triring homomor- : O(D♯(f0)) = R♯ g1 = O(D♯(g1)) 1♯f0 → R♯ f0 ⊕ R♯ (54) 24 such that a1 ρf0 ,g0(cid:18)(cid:18) a0 0 ∈ T1(f0), n, m, k ∈ Z≥0, ai, bi ∈ Ri and i = 0, 1. 0(cid:19)(cid:19) = b0 + b1 1♯f k b1 ♯ r♯k 1 g♯kv 1 z1f m 0 , b0 + + f n 0 where z1f m , (55) Case 4: i = 1 and j = 1, in which case, D♯(g1) ⊆ D♯(f1) 1 = t1 ♯ f1 =⇒ g♯θ for some θ ∈ Z>0 and t1 ∈ R1. ρf1 ,g1 such that It follows that we have a triring homomorphism f1 → R♯ 1 ! = b0 + : O(D♯(f1)) = R♯ ρf1,g1 b0 + b1 f ♯n where n ∈ Z≥0, bi ∈ Ri and i = 0, 1. g1 = O(D♯(g1)) b1 ♯ t♯n 1 g♯θn 1 , (56) (57) It is clear that the triring homomorphisms ρfj ,gi defined by (46), (52), (55) and (57) has the property (i). Note that if hk ∈ ♯√0, then the property (ii) holds because ρfj ,hk If hk 6∈ ♯√0, then we have k ≥ i ≥ j by (53). Therefore, there are only four = ρgi ,hk = 0. cases: (k, i, j) = (0, 0, 0) or (1, 0, 0) or (1, 1, 0) or (1, 1, 1). One can check that the property (ii) also holds for each of the four cases above. Let U be a nonempty open subset of Spec♯R, and let SU := { D♯(f(α))∅ 6= D♯(f(α)) ⊆ U and α ∈ ΛU } be the set of the nonempty basic open subsets contained in U . The set ΛU is a partial order set with respect to the following partial order: α ≥ β if and only if D♯(f(α)) ⊇ D♯(f(β)). (cid:16)O(D♯(f(α))), ρf(α) ,f(β)(cid:17) is an inverse system on the set ΛU , where the triring homomorphism ρf(α) ,f(β) : O(D♯(f(α))) → O(D♯(f(β))) for α ≥ β is defined by Proposition 6.1. We define O(U ) := lim ←−ΛU O(D♯(f(α))), (58) 25 where lim ←−ΛU O(D♯(f(α))) is the inverse limit of the inverse system lim (x(α)) . (59) The even part and odd part of O(D♯(f(α))), which is given by x(α) ∈ O(D♯(f(α))) and x(β) = ρf(α) ,f(β) whenever α ≥ β ←−ΛU O(D♯(f(α))) are given by lim (cid:16)O(D♯(f(α))), ρf(α) ,f(β)(cid:17). The inverse limit is a subtriring of the direct product Yα∈ΛU ←−ΛU O(D♯(f(α))) = (x(α))α∈ΛU (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)  = (x(α)i)α∈ΛU (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)  (α)(cid:0)(x(γ))γ∈ΛU(cid:1) := x(α).   x(α)i ∈(cid:0)O(D♯(f(α)))(cid:1)i and x(β)i = ρf(α) ,f(β) whenever α ≥ β (cid:18) lim ←−ΛU O(D♯(f(α)))(cid:19)i where i = 0, 1. For each α ∈ ΛU , let pU be the triring homomorphism defined by (60) ←−ΛU O(D♯(f(α))) → O(D♯(f(α))) lim Clearly, we have (x(α)i) ,   (α) : pU (61) pU (β) = ρf(α) ,f(β) pU (α) if α ≥ β and α, β ∈ ΛU . (62) The pair(cid:18) lim erty: ←−ΛU O(D♯(f(α))), { pU (α) α ∈ ΛU }(cid:19) has the following universal prop- Let ( X, { q(α) α ∈ ΛU }) be a pair consisting of a triring X and a family of triring homomorphism { q(α) : X → O(D♯(f(α))) α ∈ ΛU }. q(α) whenever α ≥ β, then there exists a unique ←−ΛU O(D♯(f(α))) such that q(α) = If q(β) = ρf(α) ,f(β) triring homomorphism σ : X → lim pU (α)σ for each α ∈ ΛU . If V ⊆ U , then SV ⊆ SU and ΛV ⊆ ΛU . Consider the pair (α) α ∈ ΛV }(cid:19) . (cid:18) lim ←−ΛU O(D♯(f(α))), { pU Note that (62) is clearly true for α, β ∈ ΛV ⊆ ΛU with α ≥ β. By the universal property of the following pair (cid:18) lim ←−ΛV O(D♯(f(α))), { pU (α) α ∈ ΛV }(cid:19) , 26 we have a unique triring homomorphism ρU,V : O(U ) = lim ←−ΛU O(D♯(f(α))) → lim ←−ΛV O(D♯(f(α))) = O(V ) such that (α) = pV pU It follows from (63) and (64) that (α)ρU,V for α ∈ ΛV . ρU,U = idO(U) for each open subset U of Spec♯ R and (63) (64) (65) ρU,W = ρV,W ρU,V (66) By (65) and (66), O is a presheaf with the restriction map ρU,V given by for three open subsets W ⊆ V ⊆ U . (63). This presheaf is called the structure presheaf on Spec♯R. Proposition 6.2 If ( R = R0⊕R1, +, ·, ♯ ) is a Hu-Liu triring, then the struc- ture presheaf on Spec♯R is a sheaf of Hu-Liu trirings. Proof According to the definition of the structure presheaf on Spec♯R, it suffices to prove that the properties (i) and (ii) in Definition 6.2 holds for an arbitrary basic open set and an arbitrary open covering which consists of basic open sets. Let D♯(fi) be an arbitrary basic open set of Spec♯R. Consider the following arbitrary open covering of D♯(fi) D♯(fi) =(cid:18) [α∈Λ0 D♯(g(α)0)(cid:19)[(cid:18) [β∈Λ1 D♯(g(β)1)(cid:19), (67) where f0, g(α)0 ∈ R0 for α ∈ Λ0, f1, g(β)1 ∈ R1 for β ∈ Λ1, Λ0 and Λ1 are two index sets. To prove the properties (i) in Definition 6.2, we need to prove that if s ∈ (68) O(D♯(fi)) satisfies (s) = 0 ρfi ,g(α)0 for α ∈ Λ0 and then s = 0. ρfi ,g(β)1 (s) = 0 for β ∈ Λ1, (69) (70) Case 1: i = 0, in which case, Λ0 6= ∅. Since s ∈ O(D♯(f0)), we have s =(cid:18) a0 f n 0 for some n, m, k ∈ Z≥0, z1f m D♯(g(α)0) ⊆ D♯(f0), there exist some r(α)0 ∈ R0 and u(α) ∈ Z>0 such that 0 ∈ T1(f0), ai, bi ∈ Ri and i = 0, 1. Since 0(cid:19) b1 1♯f k z1f m 0 , b0 + a1 + g u(α) (α)0 = r(α)0f0 for each α ∈ Λ0. (71) 27 1R 6= ∅, then for each α ∈ Λ0, we have • If D♯(f0)T Spec♯ 1R 6= ∅ Using (52), (49), (68) and (71), we have either D♯(g(α)0)\ Spec♯ 1R = ∅. a1 , b0 + z1f m 0 b1 1♯f k or D♯(g(α)0)\ Spec♯ 0(cid:19)(cid:19) (cid:19) if D♯(g(α)0)T Spec♯ if D♯(g(α)0)T Spec♯ 0 = ρf0 ,g(α)0 (s) = ρf0 ,g(α)0(cid:18)(cid:18) a0 + f n 0 b1rk + = a1rm g g , b0 + (α)0 nu(α) 0 (α)0 nu(α) 0 (α)0 ku(α) 1♯g 0 (α)0 mu(α) z1g 0 (cid:18) a0rn (cid:18) a0rn  , b0(cid:19)  1R = ∅, then we have D♯(g(α)0)T Spec♯ • If D♯(f0)T Spec♯ α ∈ Λ0. By (49), (68) and (71), we have (s) = ρf0,g(α)0 (cid:18)(cid:18) a0 , b0(cid:19) for each α ∈ Λ0 = (cid:18) a0rn 0 = ρf0 ,g(α)0 (α)0 nu(α) 0 z1f m 0 , b0 + a1 + f n 0 g 1R 6= ∅ 1R = ∅ (72) 1R = ∅ for each 0(cid:19)(cid:19) b1 1♯f k (73) Note that D♯(g(α)0)\ Spec♯ 1R = ∅ =⇒ 1♯gb (α)0 = 0 for some b ∈ Z≥0 (74) It follows from (72), (73), (74) and Proposition 4.2 (i) that there exist t, d, δαi ∈ Z>0, αi ∈ Λ0 and h(αi)0 ∈ R0 such that f t 0 = d Xi=1 h(αi)0 g δαi (αi)0 and By (76), a0 f n 0 + 0a0 = (1♯f t f t a1 = 0 and b0 + 1♯f m 0 0) ♯ a1 = b0 = (1♯f t 0) ♯ b1 = 0. b1 1♯f k 0 = 0. This proves s = 0. (75) (76) for some Case 2: i = 1, in which case, s ∈ O(D♯(f1)). Thus s = a0 + ai ∈ Ri and m ∈ Z≥0. By (53), D♯(g(α)0)) = ∅ for α ∈ Λ0 in this case. Thus, (67) becomes (77) D♯(g(β)1). a1 f ♯m 1 D♯(f1) = [β∈Λ1 Since D♯(g(β)1) ⊆ D♯(f1), there exist some r(β)1 ∈ R1 and θ(β) ∈ Z>0 such that (78) θ(β) (β)1 = r(β)1 ♯ f1 g for each β ∈ Λ1. 28 Using (57) and (78), we have 0 = ρf1 ,g(β)1 (s) = ρf1 ,g(β)1 a0 + 1 ! = a0 + a1 f ♯m a1 ♯ r♯m (β)1 ♯mθ(β) (β)1 g , which implies that there exist t′, d′, δ′ that βi ∈ Z>0, βi ∈ Λ1 and h(βi)0 ∈ R0 such f t′ 1 = h(βi)1 g δ′ βi (βi)1 d′ Xi=1 a0 = f ♯t′ 1 ♯ a1 = 0. and Hence, s = 0 by (80). (79) (80) (81) This proves that the properties (i) in Definition 6.2 holds for the basic open set D♯(fi) and the open covering (67). In order to prove that the properties (ii) in Definition 6.2 holds for the basic open set D♯(fi) and its open covering { D♯(g(α)0) α ∈ Λ0 }[{ D♯(g(β)1) β ∈ Λ1 }, it is sufficient to prove that the properties (ii) in Definition 6.2 holds for the basic open set D♯(fi) and a finite subcovering of the open covering given by (81). A direct computation proves that this fact is indeed true. This completes the proof of Proposition 6.2. References [1] M. F. Atiyah & I. G. MacDonald, Introduction to Commutative Algebra, Addison-Wesley Publishing Company, 1969. [2] David Eisenbud & Joe Harris, The Geometry of Schemes, Graduate Texts in Mathematics 52, Springer-Verlag, 1977. [3] Robert M. Fossum, Phillip A. Grififth & Idun Reiten, Trivial Extensions of Abelian Categories, Lecture Notes in Mathematics 456, Springer-Verlag, 1975. [4] Robin Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics 197, Springer-Verlag, 2000. [5] Nathan Jacobson, Basic Algebras II, W. H. Freeman and Company, 1989. 29 [6] Hideyuki Matsumura, Commutative Algebra, Second Edition, Ben- jamin/Commings Publishing Company, INC, 1980. [7] Masayoshi Miyanishi, Algebraic Geometry, Translations of Mathematical Monographs, Vol. 136, American Mathematical Society, 1994. [8] I. R. Shafarevish, Basic Algebraic Geometry, Springer-Verlag, 1974. 30
0807.4848	3	0807	2011-03-28T14:01:28	Groupoid sheaves as quantale sheaves	[ "math.RA", "math.AT", "math.CT" ]	Several notions of sheaf on various types of quantale have been proposed and studied in the last twenty five years. It is fairly standard that for an involutive quantale Q satisfying mild algebraic properties the sheaves on Q can be defined to be the idempotent self-adjoint Q-valued matrices. These can be thought of as Q-valued equivalence relations, and, accordingly, the morphisms of sheaves are the Q-valued functional relations. Few concrete examples of such sheaves are known, however, and in this paper we provide a new one by showing that the category of equivariant sheaves on a localic etale groupoid G (the classifying topos of G) is equivalent to the category of sheaves on its involutive quantale O(G). As a means towards this end we begin by replacing the category of matrix sheaves on Q by an equivalent category of complete Hilbert Q-modules, and we approach the envisaged example where Q is an inverse quantal frame O(G) by placing it in the wider context of stably supported quantales, on one hand, and in the wider context of a module theoretic description of arbitrary actions of \'etale groupoids, both of which may be interesting in their own right.	math.RA	math	Groupoid sheaves as quantale sheaves∗ Pedro Resende Abstract Several notions of sheaf on various types of quantale have been pro- posed and studied in the last twenty five years. It is fairly standard that for an involutive quantale Q satisfying mild algebraic properties the sheaves on Q can be defined to be the idempotent self-adjoint Q-valued matrices. These can be thought of as Q-valued equiva- lence relations, and, accordingly, the morphisms of sheaves are the Q-valued functional relations. Few concrete examples of such sheaves are known, however, and in this paper we provide a new one by show- ing that the category of equivariant sheaves on a localic ´etale groupoid G (the classifying topos of G) is equivalent to the category of sheaves on its involutive quantale O(G). As a means towards this end we be- gin by replacing the category of matrix sheaves on Q by an equivalent category of complete Hilbert Q-modules, and we approach the envis- aged example where Q is an inverse quantal frame O(G) by placing it in the wider context of stably supported quantales, on one hand, and in the wider context of a module theoretic description of arbitrary actions of ´etale groupoids, both of which may be interesting in their own right. Keywords: Hilbert modules, sheaves, involutive quantales, ´etale group- oids, groupoid actions, ´etendues. 2010 Mathematics Subject Classification: 06D22, 06F07, 18B25, 18F20, 22A22, 54B40. ∗Research supported in part by Funda¸cao para a Ciencia e a Tecnologia through FEDER. 1 Contents 1 Introduction 2 Background 2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Supported quantales . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Locale sheaves as modules . . . . . . . . . . . . . . . . . . . . . . . 3 Groupoid actions as quantale modules 3.1 Actions of open groupoids . . . . . . . . . . . . . . . . . . . . . . . 3.2 Actions of ´etale groupoids . . . . . . . . . . . . . . . . . . . . . . . 3.3 Actions on sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Groupoid sheaves as Hilbert modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Hilbert bases 4.2 Quantale-valued sets . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Supported modules . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Groupoid sheaves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Concluding remarks 6 Appendix 1 Introduction 2 7 7 12 16 19 19 21 26 29 29 34 39 45 50 51 By an involutive quantaloid is meant a sup-lattice enriched category (or sometimes a non-unital generalization of this) equipped with an order two sup-lattice enriched automorphism (−)∗ : Qop → Q. Involutive quantales are the involutive quantaloids with a single object. These structures can be regarded as generalized topological spaces in their own right, and sev- eral papers devoted to notions of sheaf on them have been published. In most of these the sheaves are, one way or the other, defined to be quantale- valued or quantaloid-valued matrices in a way that generalizes the notion of frame-valued set of [17, 31] (see also [6, Section 2; 35, pp. 502 -- 513]). The appropriateness of such definitions of sheaf is emphasized by the fact that every Grothendieck topos is, up to equivalence, the category of sheaves on an involutive quantaloid of "binary relations" that is obtained directly from the topos or from a site [14, 18, 60, 77]. In essence, this fact is the source of all the concrete examples known so far of toposes of (set valued) sheaves on involutive quantaloids. 2 In this paper we give another example, based on the correspondence be- tween ´etale groupoids and quantales [65], showing that the topos of equiv- ariant sheaves on a localic ´etale groupoid G is equivalent to the category of sheaves on the involutive quantale O(G) of the groupoid. This provides a way, via the groupoid representation of toposes [35, §C5; 36], in which ´etendues arise as categories of sheaves on involutive quantales. We shall begin with a somewhat detailed survey on quantaloid sheaves, meant both to provide the reader with a reasonable view of the literature and also to help trace the origins of the notion of sheaf that we shall consider in this paper. Sheaves and Grothendieck toposes. Locales are a point-free substitute for topological spaces [33,34], in particular providing a generalization of sober topological spaces, such as Hausdorff spaces. However, in many ways this is only a mild generalization. A much broader one is given by sites, where suit- able families of morphisms generalize the role of open covers. A Grothendieck topos is by definition the category (up to equivalence) of sheaves on a site [6,35,43]. Whereas locales are determined up to isomorphism by their toposes of sheaves, it is the case that very different sites may yield equivalent toposes, and it is the toposes, rather than the sites, that provide the right notion of generalized space. In particular, the category of locales Loc is equivalent to a reflective full subcategory of the category of Grothendieck toposes GTop, and the role played by toposes as spaces in their own right is emphasized by the existence of invariants such as homotopy and cohomology for them [32]. Indeed, the fact that toposes provide good grounds on which to study cohomology is precisely one the reasons that led Grothendieck to considering them [3]. A "measure" of how far toposes really differ from locales is provided by the representation theorem of Joyal and Tierney [35, §C5; 36]: a Grothendieck topos is, up to equivalence, the category of equivariant sheaves on an open groupoid G in Loc, in other words the sheaves on the locale of objects G0 that in addition are equipped with an action of G. We shall refer to these simply as G-sheaves. In particular, every ´etendue can be represented by an ´etale groupoid [36, Th. VIII.3.3] (see also [3, §IV.9.8.2; 37]). Similarly to the sit- uation with sites, groupoids themselves should be regarded as presentations of spaces rather than spaces themselves, since two groupoids yield equivalent toposes if and only if they are Morita equivalent [45, 46]. For suitable open groupoids (the ´etale-complete ones, in particular the ´etale groupoids), the topos BG of equivariant sheaves on a groupoid G is called the classifying topos of G because it classifies principal G-bundles [13, 45]. It can also be 3 thought of as the space of orbits of G, since it arises as a colimit, in the appropriate bicategorical sense, of the nerve of G regarded as a simplicial topos [45]. A different way to represent Grothendieck toposes (and also more gen- eral categories) stems from work on "categories of relations" by various au- thors. For instance, Freyd's work on allegories [18; 35, §A3] dates back to the seventies and already contains many ingredients of later theories of sheaves on involutive quantaloids. One is the equivalence of categories E ≃ Map(Rel(E)), where Rel(E) is the involutive quantaloid of binary rela- tions on a Grothendieck topos E and Map(Rel(E)) its category of "maps", which are the arrows f in Rel(E) satisfying f f ∗ ≤ 1 and f ∗f ≥ 1. Another idea is the completion [18, §2.226] that replaces a quantaloid by a suitable category of matrices that play the role of sheaves. A different characterization of categories of relations is that of Carboni and Walters [14], based on which Pitts [60] has obtained an adjunction be- tween the 2-category of Grothendieck toposes and the 2-category of quan- taloids known as distributive categories of relations (dcr's for short). The construction of the category of sheaves on a dcr by matrices is analogous to completion for allegories in [18, §2.226]. This also parallels the represen- tation of toposes by localic groupoids: dcr's are another generalization of locales because a dcr is a sup-lattice enriched category A equipped with a suitable product A ⊗ A → A that turns it into a "locale with many objects" (a fancy name would be "localoid") in the sense that if A has only one object then it is a locale and the product is ∧ (we also note that involution and in- tersection are not primitive operations as in an allegory); every Grothendieck topos arises as the category of sheaves on a dcr. Still another representation of Grothendieck toposes by involutive quan- taloids can be found in the work of Walters [77], which shows how to associate an involutive quantaloid B to any site in such a way that the topos is equiv- alent to the category of sheaves on B. Here the notion of "sheaf" is that of skeletal Cauchy-complete B-enriched category. Similarly to the previous examples, such sheaves can again be regarded as matrices and again they generalize frame-valued sets, but there are some technical differences, in par- ticular the fact that they must lie above identity matrices. This difference can be quickly described in the case of sheaves on a locale A: whereas A- valued sets directly model sheaves on A, Walters' approach uses categories enriched in the quantaloid B that arises as the split idempotent completion of A [76]. 4 Quantales and noncommutative topology. Quantales surface in a wide range of mathematical subjects, including algebra, analysis, geometry, topol- ogy, logic, physics, computer science, etc. Technically they are just sup- lattice ordered semigroups, but the name "quantale" (coined by Mulvey in the eighties) is associated with the idea of "quantizing" the point free spaces of locale theory, in analogy with the generalization of Hausdorff topology via noncommutative C-algebras [1, 2, 15, 21, 22]. In particular, Mulvey's original idea was precisely to define the "non-commutative spectrum" of a C-algebra to be a suitable quantale [47]. Several variants of this idea have been pursued for C-algebras [10,39,40,52,53,72] and arbitrary rings [8,12,69,70], often in- volving classes of quantales that satisfy specific properties, along with purely algebraic investigations of the spatial aspects of quantales [38, 59]. More recent examples include the space of Penrose tilings [55] and groupoids [65]. The idea that quantales can be regarded as spaces leads naturally to the question of how such spaces can be studied, for instance via which coho- mology theories, etc. While there has never been a systematic pursuit in this direction, there has nevertheless been some effort aimed at finding good notions of sheaf on quantales or, more generally, on quantaloids. This ef- fort is justified by a number of reasons. One, of course, is the possibility of using sheaves in order to define topological invariants for quantales and for the objects they are associated with. Another motivation for looking at quantale or quantaloid sheaves is the role they may play as structure sheaves when studying spectra of noncommutative rings or C-algebras [8, 12, 75]; in particular, there is interest in understanding more about the sheaves on the quantale Max A of a C-algebra A due to the hope that a suitable notion of structure sheaf may do away with the excess of quantale homomorphisms that make Max a somewhat ill-behaved functor [40]. Also interesting is the possibility of obtaining useful extensions of the K-theory sheaves of [16] in the context of the program of classification of C-algebras. Despite a few exceptions [4, 7, 11], most of the papers on sheaves for quan- tales or quantaloids are to a greater or lesser extent based on the definition of sheaves as matrices (frame-valued sets) introduced originally for sheaves on locales. Many technicalities depend, of course, on the specific types of quantales or quantaloids under consideration. For instance, there have been direct generalizations of frame-valued sets for right-sided or left-sided quan- tales [5, 23, 50, 56], which are those that model quantales of right-sided (resp. left-sided) ideals of rings. There have also been proposals for definitions of sheaf on general quantaloids [24, 25, 71, 75] that carry a generalization of frame-valued sets, but it has been shown by Borceux and Cruciani [9] that such definitions applied to general non-involutive quantales should at most give us a notion of ordered sheaf, rather than a discrete one. Subsequent 5 work by Stubbe [73, 74] adheres to this point of view while also generalizing the bicategorical enrichment approach of Walters [76]. For (discrete) sheaves one needs the base quantale or quantaloid Q to be equipped with an involution in order to be able to define self-adjoint Q-valued matrices (equivalently, in order to define a "symmetric" notion of Q-valued equality). For instance, the quantaloids that appear in the early works on categories of relations mentioned above are involutive, and the matrices that correspond to sheaves on them are self-adjoint. For more general involu- tive quantales and quantaloids a fairly stable theory of sheaves seems to be emerging, with many basic definitions being close in spirit, if not in form. For instance, sheaves on involutive quantaloids have been studied by Gylys [26]. Slightly later Garraway [19, 20] extended the theory of sheaves on the dcr's of [60] to a rather general class of non-unital quantaloids, while Mulvey and Ramos [54, 63] have produced a theory for involutive quantales directly inspired by the axiomatic approach of [17, 50]. More recently, Heymans [27] has provided a study of sheaves based on quantaloid enriched categories in the style of Walters [76, 77], making the connections to [26] explicit and leading to a representation theorem for Grothendieck toposes by so-called Grothendieck quantales [28]. While the theory of sheaves on involutive quantaloids appears to be thriv- ing, a negative aspect should nevertheless be mentioned, namely that so far the increase in generality of the theory has not been accompanied by a cor- responding rise in the number of known examples. Overview of the paper. Each ´etale groupoid G has an associated involu- tive quantale O(G) [65] (for a topological groupoid this quantale is just the topology equipped with pointwise multiplication of groupoid arrows), and the natural question of how the G-sheaves relate to notions of sheaf on O(G) arises. The main aim of this paper is to provide an answer to this question. As we shall see, the conclusion is that the "standard" category of matrix sheaves on an involutive quantale such as O(G) (by which we shall mean the category of O(G)-sets as, say, in [20]) and the classifying topos BG are isomorphic. Hence, ´etale groupoid sheaves yield a new example of matrix sheaves on involutive quantales. We shall proceed in three steps: Step 1: We show, in section 3, how the actions of an open or ´etale groupoid G can be described in terms of modules on the quantale O(G). The main result of this section is theorem 3.21, which proves that the cate- gory G-Loc of actions of an ´etale groupoid G is equivalent to a suitable 6 category of modules on O(G), whose algebraic description is quite sim- ple. As a restriction of this we obtain a definition of sheaf on O(G) in terms of quantale modules, and two categories of quantale modules, O(G)-LH and O(G)-Sh, which are isomorphic to BG. Step 2: We shall recall the basics of the theory of quantale-valued sets and show that these can be replaced by the theory of Hilbert modules [57,58] equipped with Hilbert bases in a way that generalizes the work of [68] for sheaves on locales. This is the contents of section 4.2, whose main result is theorem 4.29, from which it follows, for an arbitrary involutive quantale Q, that the category Set(Q) of Q-sets is equivalent to the category of maps of the involutive quantaloid Q-HMB of Hilbert Q- modules with Hilbert bases. Step 3: Finally, in section 4.4 we show that the objects of the categories O(G)-LH and O(G)-Sh coincide with the Hilbert O(G)-modules with Hilbert bases (theorems 4.47 and 4.55). Hence, in particular, O(G)-Sh coincides with the category of maps of O(G)-HMB (lemma 4.61), and therefore BG is equivalent to O(G)-Set (theorem 4.62). We hope these results provide further evidence of what should be consid- ered a "good" notion of sheaf for involutive quantales and quantaloids, and we provide a brief discussion of this at the end of section 5. Besides the main results, the paper contains subsidiary aspects of inde- pendent interest: we provide a comparison, in section 2.2, between supported quantales and modular quantales; in section 3.2, a corollary of our results is a proof of the multiplicativity of inverse quantal frames that is simpler than the original one in [65] -- in particular not using the representation of inverse quantal frames by inverse semigroups; and, in order to convey a sense of the robustness of the notion of quantale sheaf that we assume in this paper we include, as an appendix (section 6), a brief survey of some of the variants of quantale-valued set that can be found in the literature along with the relation between complete quantale-valued sets and Hilbert modules. 2 Background 2.1 Preliminaries The purpose of this section is mostly to recall some definitions and exam- ples concerning locales, localic groupoids and involutive quantales and quan- taloids, and to set up notation and terminology. 7 Locales. We shall adopt the same conventions regarding notation and ter- minology for locales that are used in [65]. In particular, following [33], we shall denote the category of frames and frame homomorphisms by Frm. We shall adopt the terminology locale instead of frame when referring to objects of the dual category Frmop, which we denote by Loc and whose arrows we refer to as continuous maps, or simply maps, of locales. If X is a locale we shall usually write O(X) for the same locale regarded as an object of Frm. If f : X → Y is a map of locales we shall refer to the frame homomorphism f ∗ : O(Y ) → O(X) that defines it as its inverse image. If f is an open map, the left adjoint to f ∗ is referred to as the direct image of f and it is denoted by f! : O(X) → O(Y ). The product of X and Y in Loc is denoted by X × Y . It coincides with the coproduct of O(X) and O(Y ) in Frm, which is the tensor product in the category of sup-lattices SL [36, §I.5]. Hence, we write O(X × Y ) = O(X) ⊗ O(Y ). Groupoids. A groupoid in a category C with enough pullbacks is an inter- nal groupoid in C. We denote the locales of objects and arrows of a groupoid G respectively by G0 and G1, and we adopt the following notation for the structure maps G = G2 m i / G1 r u d / G0 , where G2 is the pullback of the domain and range maps: G2 π2 π1 G1 r . G1 / G0 d We remark that, since G is a groupoid rather than just an internal category, the multiplication map m is the pullback of d along itself: G2 m π1 G1 d . G1 / G0 d The following are examples: • A topological groupoid is an internal groupoid in the category of topo- logical spaces and continuous maps. 8 / / / / o o / / / / / / • A Lie groupoid is an internal groupoid in the category of smooth man- ifolds such that d is a submersion (this condition ensures that the pull- back G2 exists). • The category Loc has pullbacks and a localic groupoid is an internal groupoid in Loc. A localic groupoid G is said to be open if d is an open map. Hence, if G is open the multiplication map m is also an open map. An ´etale groupoid is an open groupoid such that d is a local homeomorphism, in which case all the structure maps are local homeomorphisms and, hence, G0 is isomorphic to an open sublocale of G1. Conversely, any open groupoid for which u is an open map is necessarily ´etale [65, Corollary 5.12]. Similar conventions and remarks apply to topological groupoids. Involutive quantales. By an involutive quantale is meant an involutive semigroup in the monoidal category SL of sup-lattices. We shall adopt the following terminology and notation: • The product of two elements a and b of an involutive quantale Q is denoted by ab, the involute of a is denoted by a∗, the join of a subset S ⊂ Q by W S, the top element by 1Q or simply 1, and the bottom element by 0Q or simply 0. The elements such that a∗ = a are self- adjoint. The idempotent self-adjoint elements are the projections. • The involutive quantale Q is unital if there is a unit for the multi- plication, which is denoted by eQ or simply e. (This is necessarily a projection.) • By a homomorphism of involutive quantales h : Q → R is meant a homomorphism of involutive semigroups in SL. If Q and R are unital, the homomorphism h is unital if h(eQ) = eR. Similarly, given an involutive quantale Q, by a (left) Q-module will be meant a sup-lattice M equipped with an associative left action Q ⊗ M → M in SL (the involution of Q plays no role). The action will be assumed to be unital whenever Q is. The notations for joins, top, bottom, are similar to those of quantales themselves, and the action of an element a ∈ Q on x ∈ M is denoted by ax. By a homomorphism of left Q-modules h : M → N is meant a Q-equivariant homomorphism of sup-lattices. Example 2.1 The following are examples of involutive quantales: 9 1. Any frame L is a unital involutive quantale with e = 1, trivial involution and multiplication ab = a ∧ b. 2. Let G be a topological groupoid. If G is open the topology Ω(G1) is an involutive quantale with the product of two open sets U and V being given by the pointwise multiplication of groupoid arrows: UV = m(U ×G0 V ) = {m(x, y) x ∈ U, y ∈ V, r(x) = d(y)} ; and the involute of an open set U is its pointwise inverse U ∗ = i(U). The quantale is unital if and only if G is an ´etale groupoid, in which case the unit e coincides with the set of unit arrows u(G0). Analogous facts apply to localic groupoids (see section 2.2). 3. In particular, the topology of any topological group is an involutive quantale, and the quantale is unital if and only if the group is discrete. 4. Another particular example is ℘( X × X), the quantale of binary rela- tions on the set X [51], which is the discrete topology of the greatest equivalence relation on X, sometimes referred to as the "pair groupoid" on X. 5. Let A be a C-algebra. The set of closed (under the norm topology) linear subspaces of A is an involutive quantale Max A [48, 49, 52]. The multiplication of two closed linear subspaces is the closure of the linear span of their pointwise product. The involute of a closed linear subspace if its pointwise involute. The quantale is unital if A has a unit 1, in which case e is the linear span C1. Involutive quantaloids. Quantaloids are the many objects generalization of quantales. However, although at odds with our terminology for involutive quantales, we shall not need to consider quantaloids without units: Definition 2.2 1. By a quantaloid is meant a sup-lattice enriched cate- gory. 2. An involutive quantaloid is a quantaloid Q equipped with a contravari- ant sup-lattice enriched isomorphism (−)∗ : Qop → Q which is both the identity on objects and its own inverse. 3. An involutive quantaloid Q is modular if any arrows a, b, c ∈ Q satisfy the modularity axiom of Freyd whenever the compositions are defined: (2.3) ab ∧ c ≤ a(b ∧ a∗c) 10 The category of sets with relations as morphisms, Rel, is the prototypical example of a modular quantaloid: the morphisms R : X → Y are the rela- tions R ⊂ Y × X; and the (total) functions f : X → Y can be identified with the relations R : X → Y such that RR∗ ⊂ ∆Y and R∗R ⊃ ∆X ; that is, R∗ is right adjoint to R (equivalently, R has a right adjoint, which is necessarily R∗). This justifies the following notation and terminology: Definition 2.4 Let Q be an involutive quantaloid. 1. A map is a morphism f : x → y which is left adjoint to f ∗; that is, f f ∗ ≤ idy and f ∗f ≥ idx. The map f is injective if it further satisfies f ∗f = idx, and surjective if it satisfies f f ∗ = idy. If the map f is both injective and surjective we say that it is unitary. 2. The subcategory of Q containing the same objects as Q and the maps as morphisms is denoted by Map(Q). Remark 2.5 It is more or less standard, in the case of non-involutive quan- taloids, to use the terminology map for a morphism that has a right adjoint. This does not coincide, in general, with the above definition in the case of an involutive quantaloid, but for modular quantaloids the definitions coincide [20, Th. 2.2]. Definition 2.6 Two involutive quantaloids Q and R are equivalent if there exist two sup-lattice enriched and involution preserving functors (in other words, two homomorphisms of involutive quantaloids) Q F G ) R such that G ◦ F and F ◦ G are naturally isomorphic to idQ and idR, respec- tively, via unitary natural isomorphisms (i.e., natural isomorphisms whose components are unitary maps). It is an easy exercise to show that an adjoint equivalence (of categories) F ⊣ G between the involutive quantaloids Q and R is an equivalence in the stronger sense just defined if G is a homomorphism of involutive quantaloids and the unit of the adjunction is unitary (equivalently, F is a homomorphism and the counit is unitary). 11 ) i i 2.2 Supported quantales The quantales which are associated to ´etale groupoids are the inverse quantal frames. They are instances of the more general and algebraically well behaved class of stable quantal frames, which in turn is included in the equally well behaved class of stably supported quantales. We begin by recalling some properties of these quantales, following [65], and we study their relation to modularity. Groupoid quantales. Let us recall a few aspects of the correspondence between localic groupoids and quantales. Let G be an open localic groupoid. Since the multiplication map m is open, there is a sup-lattice homomorphism defined as the following composition (in SL): O(G1) ⊗ (G1) / O(G2) m! / O(G1) . This defines an associative multiplication on O(G1) which together with the isomorphism O(G1) i!→ O(G1) makes O(G1) an involutive quantale. This quantale is denoted by O(G) -- it is the "opens of G". The involutive quantale O(G) of an open groupoid G is unital if and only if G is ´etale [65, Corollary 5.12], in which case the unit is e = u!(1) and u! defines an order-isomorphism u! : O(G0) ∼=−→ ↓(e) = {a ∈ O(G) a ≤ e} . Hence, in particular, ↓(e) is a frame (cf. 2.14). Stably supported quantales. Let Q be a unital involutive quantale. We recall that by a support on Q is meant a sup-lattice homomorphism ς : Q → Q satisfying the following conditions for all a ∈ Q: (2.7) (2.8) (2.9) ς(a) ≤ e ς(a) ≤ aa∗ a ≤ ς(a)a . The support is said to be stable, and the quantale is stably supported, if in addition we have, for all a, b ∈ Q: (2.10) ς(ab) = ς(aς(b)) . 12 / / / / Example 2.11 The quantale O(G) of an ´etale groupoid G is stably sup- ported, and the support is given by ς = u! ◦ d! : O(G) → O(G) (cf. proof of [65, Theorem 5.11]). For any quantale Q with a support, the following equalities hold for all a, b ∈ Q [65, Lemma 3.3(12)], (2.12) (2.13) ς(a)1 = a1 , ς(b) = b if b ≤ e , and the unital involutive subquantale ↓(e) = {a ∈ Q a ≤ e} is a base locale in the following sense [65, Lemma 3.3]: Definition 2.14 Let Q be a unital involutive quantale and let B = ↓(e). We say that B is a base locale for Q if b = b∗ and bc = b ∧ c for all b, c ∈ B. (B is necessarily a frame, by [36, §III.1].) We further recall [65, Lemma 3.4 and Theorem 3.8] that any stably sup- ported quantale Q admits a unique support, which is given by the following formulas, (2.15) (2.16) ς(a) = a1 ∧ e , ς(a) = aa∗ ∧ e , and, moreover, a support is stable if and only if (2.17) for all a ∈ Q. ς(a1) ≤ ς(a) It has also been proved [65, Lemma 3.4-5] that, if Q is a stably supported quantale, then (2.18) ba = b1 ∧ a for all b ∈ B and a ∈ Q, from which the following useful property follows: Lemma 2.19 Let Q be a stably supported quantale, and let a, b ∈ Q with b ≤ e. Then ba ∧ e = b ∧ a. Proof. ba ∧ e = (b1 ∧ a) ∧ e = (b1 ∧ e) ∧ a = be ∧ a = b ∧ a. Furthermore, every stable support is B-equivariant, because for all a ∈ Q and b ∈ B we have, by (2.13), ς(ba) = ς(bς(a)) = bς(a), and in fact equivariance is equivalent to stability (cf. 4.41): 13 Theorem 2.20 Let Q be a supported quantale. The support of Q is stable if and only if it is a homomorphism of B-modules. Remark 2.21 All the above properties of supports and stable supports still hold if the definition of support is weakened by requiring supports to be only monotone instead of join-preserving. In particular, 2.20 still holds because if ς : Q → B is a monotone B-equivariant map satisfying (2.8) -- (2.9) then it is left adjoint to the assignment (−)1 : B → Q, and hence it preserves joins. Hence, the exact definition of support is irrelevant as far as stable supports are concerned. Inverse quantal frames. By a stable quantal frame is meant a stably supported quantale which is also a frame. The following condition holds for all stable quantal frames [65, Lemma 4.17(28)] and will be used in the proof of 4.47: (2.22) cover condition y ∧ z . (a ∧ e)1 ≥ _yz ∗≤a _ I(Q) = 1 , An inverse quantal frame is a stable quantal frame Q that satisfies the where I(Q) = {s ∈ Q ss∗ ∨ s∗s ≤ e} is the set of partial units of Q. This set, equipped with the multiplication of Q, has the structure of a complete and infinitely distributive inverse monoid whose inverses are given by the involution of the quantale [65, Corollary 3.26]. We remark that we have (2.23) ς(s) = ss∗ for all s ∈ I(Q) . The inverse quantal frames Q are precisely the quantales of the form Q ∼= O(G) for a localic ´etale groupoid G [65, Theorem 4.19 and Theorem 5.11]. Example 2.24 For the sake of illustration, let us describe this correspon- dence in the case of a topological ´etale groupoid G. The topology Ω(G1) is an inverse quantal frame (cf. 2.1) whose support is given by ς(U) = u(d(U)) for all open sets U of G1. By a local bisection of G is meant a continuous local section s of d such that r ◦ s is an open embedding of the domain of s into G0, and the partial units are precisely the images of the local bisec- tions. Equivalently, a partial unit is the same as an open set U ∈ Ω(G1) such that the restrictions dU and rU are injective. In particular, the partial units of the quantale of binary relations ℘( X × X) on a set X are the partial bijections on X. 14 Modular quantales. The notion of modularity of Freyd (cf. 2.2) is crucial in his characterization of abstract quantaloids of binary relations. Similarly, the existence of stable supports provides us with a definition of what may be meant by an abstract quantale of binary relations, as in [44]. We are thus provided with two natural ways of abstracting quantales of binary relations, and it is worth comparing them. In addition, the fact that inverse quantal frames are modular (cf. 2.29) will play a role at the end of section 4.4. As a first step we see that stably supported quantales are more general than modular quantales: Theorem 2.25 Every modular quantale is stably supported. Proof. An involutive quantale Q is modular (cf. 2.2) if for all a, b, c ∈ Q we have (2.26) a(a∗b ∧ c) ≥ b ∧ ac or, equivalently, for all a, b, c ∈ Q we have (2.27) (c ∧ ba∗)a ≥ ca ∧ b . Let then Q be modular, and define the operation ς : Q → B by ς(a) = aa∗ ∧ e . This operation is monotone, and in order to see that it is a stable support we check that it satisfies the required three laws, namely (2.8) -- (2.9) and stability. Whereas (2.8) holds almost by definition, (2.9) follows from a direct application of (2.27): ς(a)a = (aa∗ ∧ e)a ≥ a ∧ ea = a . And we obtain stability by a direct application of (2.26): ς(a1) = a1a∗ ∧ e = a1a∗ ∧ e ∧ e ≤ a(1a∗ ∧ a∗) ∧ e = aa∗ ∧ e = ς(a) . The two notions do not coincide, however, as the following example due to Jeff Egger shows: 15 Example 2.28 Let Q be the 8-element boolean algebra with atoms a, b, c, equipped with the trivial involution and the following multiplication: 0 a b c x y z 1 0 0 0 0 0 0 0 0 0 a 0 a b c x y z 1 b 0 b 1 1 1 1 1 1 c 0 c 1 1 1 1 1 1 x 0 x 1 1 1 1 1 1 y 0 y 1 1 1 1 1 1 z 0 z 1 1 1 1 1 1 1 0 1 1 1 1 1 1 1 Defining ς(q) = a for all q 6= 0 we obtain a stable support, but Q is not modular because bc ∧ a = 1 ∧ a = a and b(c ∧ ba) = b(c ∧ b) = b0 = 0. In this example the quantale is also a frame. Hence, modularity is stronger than being stably supported even for quantal frames. In turn, every in- verse quantal frame is necessarily modular, as has been mentioned in [30] by taking into account the representation of inverse quantal frames by inverse semigroups of [65]. A direct proof is the following: Theorem 2.29 Every inverse quantal frame is modular, but not every mod- ular quantal frame is an inverse quantal frame. Proof. Let Q be an inverse quantal frame, and let a, b, c ∈ Q. Let a =Wi si, b =Wj tj and c =Wk uk, where si, tj and uk are partial units for all i, j, k. We have si(tj ∧ s∗ si(tj ∧ s∗ sitj ∧ sis∗ i uk i uk) = _i,j,k a(b ∧ a∗c) = _i,j,k,ℓ and ℓ uk) ≥ _i,j,k ab ∧ c = _i,j,k sitj ∧ uk , and modularity follows from the equality sitj ∧sis∗ i uk = sitj ∧uk. An example showing that not every modular quantal frame is an inverse quantal frame is the four-element quantale R of 4.12. 2.3 Locale sheaves as modules Sheaves on locales can be described as quantale modules (on locales) in more than one way. With the exception of 2.35, all the statements that follow are recalled from [68], whose terminology we follow. 16 Maps as modules. If p : X → B is a map of locales then O(X) is an O(B)-module by change of "base ring" along p∗; that is, the action is given by bx = p∗(b) ∧ x for all b ∈ O(B) and x ∈ O(X). This makes X an O(B)-locale, by which is meant that O(X) is equipped with a structure of O(B)-module satisfying the condition b1 ∧ x = bx for all b ∈ O(B) and x ∈ O(X). We remark that the map p can be recovered from the module structure by the condition p∗(b) = b1. By a map of O(B)-locales is meant a map of locales f whose inverse image f ∗ is a homomorphism of O(B)-modules. The resulting category of O(B)- locales is denoted by O(B)-Loc, and it is isomorphic to the slice category Loc/B [68, Theorem 1]. If p : X → B is an open map the unit of the adjunction p! ⊣ p∗ Open maps. gives us p!(x)x = x for all x ∈ O(X). Conversely, if X is a locale for which O(X) is an O(B)-module equipped with a homomorphism ς : O(X) → O(B) of O(B)-modules such that ς(x)x = x for all x ∈ O(X) then X is an O(B)- locale and the corresponding map of locales p : X → B is open with p! = ς [68, Theorem 3]. Such an O(B)-locale is called open. For each x ∈ O(X) the element ς(x) is referred to as the support of x, and ς itself is called the support of X, in imitation of the terminology for supported quantales (cf. section 2.2). Sheaves. Now let p : X → B be a local homeomorphism. The images of the local sections of p can be identified [68, §2.3] with the elements s ∈ O(X) such that (2.30) ∀x∈O(X) x ≤ s ⇒ x = ς(x)s . Henceforth we shall refer to the elements that satisfy (2.30) simply as local sections, and we shall denote the set of all the local sections by ΓX . Of course, we haveW ΓX = 1 ('the local sections cover X'). Any open O(B)-locale X which is thus covered by the local sections is called an ´etale O(B)-locale and the full subcategory of O(B)-Loc whose objects are the ´etale O(B)-locales is denoted by O(B)-LH. Of course, O(B)- LH is equivalent to LH/B, the full subcategory of Loc/B whose objects are the local homeomorphisms into B, which in turn is equivalent to Sh(B), the category of sheaves on B and natural transformations between them. Hence, from here on we adopt the following shorter terminology: 17 Definition 2.31 Let A be a frame. By an A-sheaf locale. is meant an ´etale A- If X and Y are O(B)-sheaves, by a sheaf homomorphism h : O(X) → O(Y ) is meant a homomorphism of O(B)-modules which preserves supports and local sections; that is, (2.32) (2.33) ς(h(x)) = ς(x) for all x ∈ O(X) h(ΓX) ⊂ ΓY . The sheaf homomorphisms are the direct images f! : O(X) → O(Y ) of the maps f : X → Y [68, Theorem 5]. The category whose objects are the O(B)-sheaves and whose arrows are the sheaf homomorphisms between them is isomorphic to O(B)-LH and it is denoted by O(B)-Sh. There is an alternative way of describing the sheaves on B, in terms of Hilbert modules on O(B) (cf. section 4.1): the O(B)-sheaves are precisely the same as the Hilbert O(B)-modules which are equipped with Hilbert bases (cf. 4.19). The Hilbert module inner product of an O(B)-sheaf X is given by (2.34) hx, yi = ς(x ∧ y) , and the adjoint ϕ† of a sheaf homomorphism ϕ = f! : O(X) → O(Y ), which is defined by the condition hϕ(x), yi = hx, ϕ†(y)i, coincides with the inverse image homomorphism f ∗ [68, Theorem 11]. We conclude this brief exposition on locale sheaves with a useful fact not mentioned in [68]: Lemma 2.35 Let B be a locale, let X be an O(B)-sheaf, and let s ∈ O(X). Then s is a local section if and only if (2.36) ∀x∈O(X) hx, sis ≤ x . Proof. The equivalence is easily proved: • If s satisfies (2.30) and x ∈ O(X) then x ∧ s ≤ s and, hence, we have hx, sis = ς(x ∧ s)s = x ∧ s ≤ x. • Conversely, if s satisfies (2.36) and x ≤ s then x = ς(x)x = ς(x ∧ x)x ≤ ς(x ∧ s)s = hx, sis ≤ x. 18 3 Groupoid actions as quantale modules In this section we show that the assignment from open groupoids to quan- tales has a one-sided generalization whereby actions of open groupoids define quantale modules. We shall begin by addressing the more general situation, for open groupoids, after which ´etale groupoids will be considered along with actions on open maps and local homeomorphisms. A module-theoretic formu- lation of the actions of ´etale groupoids will be obtained, and a first description of groupoid sheaves in terms of quantale modules will be achieved. 3.1 Actions of open groupoids Preliminaries on groupoid actions. Let G be a localic groupoid. By a locale over G0, or simply a G0-locale, will be meant a locale X together with a map p : X → G0 called the projection into G0. The category of G0- locales is the slice category Loc/G0. A (left) action of G on the G0-locale (X, p) is a map of locales a : G1 ×0 X → X such that the following diagrams commute, where G1 ×0 X, G2 ×0 X and G1 ×0(G1 ×0 X) are pullbacks in Loc respectively of r and p, r ◦ π2 and p, and r and d ◦ π1: (3.1) (3.2) (3.3) G1 ×0 X a X π1 p G1 d / G0 G1 ×0(G1 ×0 X) 1×a / G1 ×0 X ∼= G2 ×0 X m×1 G1 ×0 X a (Associativity) a / X G1 ×0 X hu◦p,1i uuuuuuuuu X a $IIIIIIIII (Unitarity) X The G0-locale (X, p) together with the action a will be referred to as a (left) G-locale and we shall denote it by (X, p, a), or simply by X when no confusion will arise. The following simple fact will be useful a few times later on: 19 / / / / / $ : : Lemma 3.4 Let p : X → G0 be a map of locales and let G1 ×0 X be the pullback of r and p. Then the projection π1 : G1 ×0 X → G1 coincides with the map m ◦ (1 × (u ◦ p)). In particular, (3.1) is equivalent to the equation p ◦ a = d ◦ m ◦ (1 × (u ◦ p)). Proof. This follows from the commutativity of the following diagram, whose left triangle is obviously commutative and whose right triangle is commuta- tive due to one of the unit laws of G: G1 ×0 X 1×p G1 ×0 G0 1×u G1 ×0 G1 )SSSSSSSSSSSSSSSS π1 ukkkkkkkkkkkkkkkk m π1∼= G1 From actions to modules. It is easy to show that the diagram (3.1) is a pullback (briefly, because the action can be reversed due to the inversion operation i of the groupoid), and thus if G is an open groupoid the action map a is necessarily open. Hence, in this case, taking into account that G1 ×0 X is, in Frm, a quotient G1 ⊗0 X of the tensor product G1 ⊗ X, we obtain a sup-lattice homomorphism by composing with the direct image of the action: G1 ⊗ X / G1 ⊗0 X a! / X Showing that this defines an action of O(G) on X (a left quantale module) is straightforward and essentially the same as the proofs of associativity and unit laws for the quantale O(G) (cf. [65, Theorem 5.2 and Lemma 5.8]). Definition 3.5 Let G be an open groupoid. We shall denote by O(X) the left O(G)-module which is obtained from a G-locale X. Equivariant maps. Let X and Y be G-locales with actions a and b, re- spectively. An equivariant map from X to Y is a map f : X → Y in Loc/G0 that commutes with the actions; that is, such that the following diagram commutes: G1 ×0 X a X 1×f f G1 ×0 Y b / Y We shall refer to the category of G-locales and equivariant maps between them as G-Loc. It is simple to see that, since G is a groupoid rather than just a category, the above diagram is actually a pullback. Hence, if G is an 20 / / ) / / u / / / / / / / open groupoid, in which case as we have seen the actions are open maps, the following diagram in SL also commutes [36, Proposition V.4.1]: G1 ⊗0 X o a! X o 1⊗f ∗ f ∗ G1 ⊗0 Y b! Y This implies that the locale homomorphism f ∗ commutes with the actions of O(G) on O(X) and O(Y ), and thus it is a homomorphism of O(G)-modules. Hence, we obtain: Lemma 3.6 The assignments X 7→ O(X) and f 7→ f ∗ define a faithful functor O : G-Loc → O(G)-Modop. Comparing this with [65, Theorem 5.14 and Example 5.15] we see that the assignment from groupoid actions to modules has better functorial properties than the assignment from groupoids to quantales. This functor is not full, but we make the following observation: Lemma 3.7 Let G be an open groupoid and let f : X → Y be a map of locales such that f ∗ is a homomorphism of O(G)-modules. Denoting the actions of X and Y by a and b, respectively, we have f ◦ a ≥ b ◦ (1 × f ). Proof. Let us prove the inverse image version of the inequality, that is a∗ ◦ f ∗ ≥ (1 ⊗ f ∗) ◦ b∗ , using the equality f ∗◦b! = a!◦(1⊗f ∗) that corresponds to the Q-equivariance of f ∗: a∗ ◦f ∗ ≥ a∗ ◦f ∗ ◦ b! ◦ b∗ = a∗ ◦ a! ◦(1⊗f ∗)◦ b∗ ≥ (1⊗f ∗)◦ b∗ = (1×f )∗ ◦ b∗ . 3.2 Actions of ´etale groupoids Now we study actions of localic ´etale groupoids. As we shall see, the existence of a base locale (cf. 2.14) for the quantale of such a groupoid enables us to extend to groupoid actions the module language of locale sheaves (cf. 2.3). We remark that more could have been said along these lines for open groupoids, too, since there is a (more general) notion of base locale for the quantales of these [61, 62], but for the purposes of this paper that is not needed. 21 o o Q-locales. For any localic ´etale groupoid G, if X is a G-locale with projec- tion p : X → G0 then O(X) is an O(G0)-module by change of "ring" along the inverse image homomorphism p∗ : O(G0) → O(X). Letting B denote the base locale of O(G), the same action of O(G0) on O(X) can be obtained through the isomorphism O(G0) ∼= B by restricting the action of Q: Lemma 3.8 Let G be an ´etale groupoid and let X be a G-locale with projec- tion p : X → G0. For all b ∈ O(G0) and x ∈ O(X) we have u!(b)x = p∗(b)∧x. In particular, O(X) is a unital Q-module and the action uniquely defines p by the equation p∗(b) = u!(b)1. Proof. Axiom (3.3) of G-locales is a ◦ hu ◦ p, 1i = 1, which we can rewrite as a ◦ (u × 1) ◦ hp, 1i = 1, where the pairing hp, 1i : X → G0 ×0 X is an isomorphism and thus a ◦ (u × 1) = hp, 1i−1. Hence, we have and the required equation follows: a! ◦ (u! ⊗ 1) = [p∗, 1] u!(b)x = a!(u!(b) ⊗ x) = (a! ◦ (u! ⊗ 1))(b ⊗ x) = [p∗, 1](b ⊗ x) = p∗(b) ∧ x . Hence, the faithful functor O : G-Loc → Q-Modop of 3.6 restricts to a functor to the following category Q-Loc: Definition 3.9 Let Q be an inverse quantal frame with base locale B. By a Q-locale will be meant a locale X such that O(X) is a (unital) left Q-module whose action satisfies the condition bx = b1 ∧ x for all b ∈ B and x ∈ O(X). The category of Q-locales, Q-Loc, is that whose objects are the Q-locales and whose morphisms f : X → Y are the maps of locales such that f ∗ is a homomorphism of Q-modules. Example 3.10 Any inverse quantal frame Q itself defines a Q-locale, since (G, d, m) is a G-locale: the equality ba = b1∧a holds for all b ∈ B and a ∈ Q, and, due to the involution, ab = 1b ∧ a also holds (corresponding to the right G-locale structure of G with projection r). Example 3.11 Let Q = O(G) be an inverse quantal frame with base locale B. If X is a B-locale then Q ⊗B O(X) is a frame whose natural left Q-action defines a Q-locale: b(a ⊗ x) = ba ⊗ x = (b1 ∧ a) ⊗ x = b(1 ⊗ 1) ∧ (a ⊗ x) . If X corresponds to a G0-locale p : X → G0 then the Q-locale Q ⊗B O(X) corresponds to a G-locale G1 ×0 X whose projection d ◦ π1 (where π1 is the pullback of p along r) is an open map (resp. a local homeomorphism) if p is. 22 Example 3.12 If the inverse quantal frame Q coincides with its base locale B (i.e., the corresponding groupoid G is just the locale G1 = G0 with identity structure maps) the category B-Loc is that of section 2.3. Multiplicativity. Let G be an ´etale groupoid. Any left O(G)-module M (not necessarily an O(G)-locale, or even a locale) is also a left B-module due to the inclusion of its base locale B ⊂ O(G). Hence, we can form the tensor product O(G) ⊗B M. The associativity of the action O(G) ⊗ M → M implies that it factors through the quotient O(G) ⊗ M → O(G) ⊗B M and a sup-lattice homomorphism α : O(G) ⊗B M → M, whose right adjoint α∗ is given by α∗(x) = _{a ⊗ y ∈ O(G) ⊗B M α(a ⊗ y) ≤ x} = _{a ⊗ y ∈ O(G) ⊗B M ay ≤ x} . But the fact that O(G) is an inverse quantal frame provides us with a more useful formula for α∗: Lemma 3.15 Let Q be an inverse quantal frame with base locale B and let M be a left Q-module with action α : Q ⊗B M → M. The right adjoint α∗ is given by, for all x ∈ M, (3.13) (3.14) (3.16) α∗(x) = _s∈I(Q) s ⊗ s∗x . It follows that α∗ preserves arbitrary joins (besides arbitrary meets). Proof. Since I(Q) is join-dense in Q and joins distribute over tensors we can equivalently replace a in (3.14) by s ∈ I(Q) and thus obtain α∗(x) = _sy≤x = _s∗sy≤s∗x ≤ _s∈I(Q) s ⊗ y ≤ _s∗sy≤s∗x s ⊗ y = _s∗sy≤s∗x ss∗s ⊗ y s ⊗ s∗sy [because s∗s ∈ B -- cf. (2.23)] s ⊗ s∗x ≤ α∗(x) , where the last inequality is a consequence of the fact that for each s ∈ I(Q) we have ss∗x ≤ x and thus s ⊗ s∗x ≤ α∗(x). Hence, all the above inequalities are in fact equalities. The fact that α∗ preserves joins is an immediate consequence, for if Y ⊂ M then α∗(cid:16)_ Y(cid:17) = _s∈I(Q) s ⊗ s∗_ Y = _x∈Y _s∈I(Q) s ⊗ s∗x =_ α∗(Y ) . 23 Remark 3.17 This result holds under more general assumptions, namely it suffices that Q be a unital involutive quantale containing a join-dense sub- involutive-semigroup S ⊂ Q such that ss∗ ≤ e and s ≤ ss∗s (hence, s = ss∗s) for all s ∈ S (notice that B = ↓(e) is always a unital involutive subquantale of Q and the same remarks about the tensor product Q ⊗B M apply). In this more general situation we obtain α∗(x) =_s∈S s ⊗ s∗x . Examples of such quantales are the inverse quantales of [65] -- the set I(Q) of partial units of an inverse quantale Q is a join-dense complete inverse monoid whose locale of idempotents coincides with B. Such a quantale is of the form O(G) for an ´etale groupoid G if and only if it is also a frame [65]. As a corollary of this we conclude that the multiplication µ : Q ⊗B Q → Q of an inverse quantale Q necessarily has a join preserving right adjoint given by (3.18) µ∗(a) = _s∈I(Q) s ⊗ s∗a In particular, we obtain in this way a new and simpler proof of the fact that every inverse quantal frame is multiplicative. Equivalence between G-locales and Q-locales. Now we shall see that the categories of G-locales and of O(G)-locales, for any ´etale groupoid G, amount to the same thing. Lemma 3.19 Let G be an ´etale groupoid. The assignment X 7→ O(X) from G-locales to O(G)-locales is a (strict) bijection. Proof. Let Q = O(G) and let X be a Q-locale. The inclusion of the base locale B ⊂ Q makes O(X) a B-locale and thus we have a map p : X → G0 defined by p∗(b) = u!(b)1 (cf. 3.12). Since the pullback G1 ×0 X of r and p is, in the category of frames, the quotient of the frame coproduct O(G1)⊗O(X) generated by the equalities (3.20) π∗ 1(r∗(b)) = π∗ 2(p∗(b)) , the Q-locale conditions p∗(b) ∧ x = u!(b)x and a ∧ r∗(b) = au!(b) (cf. 3.10) show, if we stabilize (3.20) under finite meets, that G1 ×0 X coincides with the sup-lattice quotient generated by the equalities au!(b) ⊗ x = a ⊗ u!(b)x, 24 in other words it is the tensor product of B-modules Q ⊗B O(X). Since the right adjoint α∗ of the module action α : Q ⊗B O(X) → O(X) preserves joins (see 3.15), we define a groupoid action a : G1 ×0 X → X by a∗ = α∗ and in order to see that we have obtained a G-locale all we need is to verify that the three axioms (3.1) -- (3.3) are satisfied. Of course, once this is done our proof will be finished because it is clear that the construction of the G-locale structure from the Q-locale thus obtained is the inverse of the assignment Y 7→ O(Y ). Axiom (3.2) (the associativity of a) follows in a straightforward manner from the associativity of α because α = a!. (This is completely analogous to the way in which the associativity of the multiplication of an open groupoid follows from the associativity of the multiplication of its quantale, cf. [65, Theorem 4.8].) Proving the two other axioms is less easy because p is not necessarily an open map and thus we do not have straightforward direct image versions of the axioms we want to prove. Let us start with axiom (3.1). By 3.4, this is equivalent to the equation p ◦ a = d ◦ m ◦ (1 × (u ◦ p)), which we can verify directly in terms of inverse images using the formulas (3.16) and (3.18) for a∗ and m∗: on one hand we have and, on the other, a∗(p∗(b)) = _s∈I(Q) m∗(d∗(b)) = _s∈I(Q) s ⊗ s∗u!(b)1X s ⊗ s∗u!(b)1Q . The inverse image of 1 × (u ◦ p) is given by (1 ⊗ (p∗ ◦ u∗))(a ⊗ c) = a ⊗ ((c ∧ e)1X ) and, combining these formulas, we obtain 1 ⊗ (p∗ ◦ u∗)(m∗(d∗(b))) = _s∈I(Q) s ⊗ (s∗u!(b)1Q ∧ e)1X = a∗(p∗(b)) , where the last step follows from the following three facts: (i) s∗u!(b) belongs to I(Q); (ii) for all t ∈ I(Q) we have t1Q ∧ e = ς(t) = tt∗; (iii) for all t ∈ I(Q) we have tt∗1X ≤ t1X = tt∗t1X ≤ tt∗1X, and thus (s∗u!(b)1Q∧e)1X = s∗u!(b)1X . 25 Now let us verify axiom (3.3). The inverse image of a ◦ hu ◦ p, 1i is given by [p∗ ◦ u∗, 1](a∗(x)) = _s∈I(Q) p∗(u∗(s)) ∧ s∗x = _s∈I(Q) (s ∧ e)1X ∧ s∗x . Since X is a Q-locale we have (s ∧ e)1X ∧ s∗x = (s ∧ e)s∗x and, since s is in the inverse monoid I(Q), we also have (s ∧ e)s∗ = s ∧ e. Hence, _s∈I(Q) (s ∧ e)1X ∧ s∗x = _s∈I(Q) (s ∧ e)x =(cid:16)_ I(Q) ∧ e(cid:17) x = ex = x and we conclude that a ◦ hu ◦ p, 1i = 1 as required. Theorem 3.21 Let G be an ´etale groupoid. The categories G-Loc and O(G)-Loc are isomorphic. Proof. Let Q = O(G). All we need to do is show that the functor O : G- Loc → Q-Loc is full. Let X and Y be G-locales, let f : X → Y be a map of locales such that f ∗ is a homomorphism of Q-modules, and let the actions of G on X and Y be a and b, respectively. By 3.7, in order to prove that the functor is full we only have to prove, for all y ∈ O(Y ), the inequality (3.22) a∗(f ∗(y)) ≤ (1 ⊗ f ∗)(b∗(y)) . From 3.15 and the fact that f ∗ is Q-equivariant we have a∗(f ∗(y)) = _s∈I(Q) s ⊗ s∗(f ∗(y)) = _s∈I(Q) s ⊗ f ∗(s∗y) . The expression s ⊗ f ∗(s∗y) on the right equals (1 ⊗ f ∗)(s ⊗ (s∗y)), and we have s ⊗ (s∗y) ≤ b∗(y) because b!(s ⊗ (s∗y)) = ss∗y ≤ y. This proves the inequality (3.22). 3.3 Actions on sheaves Actions on open maps. For any groupoid G, by an open G-locale will be meant a G-locale whose projection is an open map. Similarly, for an ´etale groupoid the corresponding O(G)-locales will be called open. Their descrip- tion is very simple and does not even require the O(G)-locale condition: 26 Lemma 3.23 Let Q be an inverse quantal frame with base locale B, and let X be a locale such that O(X) is a Q-module. Then X is an open Q-locale if and only if there exists a (necessarily unique) homomorphism of B-modules such that ς(x)x = x for all x ∈ O(X). ς : O(X) → B Proof. This follows immediately from the description of open maps of locales p : X → B in terms of O(B)-modules (cf. section 2.3): if p is open, the homomorphism ς equals u! ◦ p!. If Q is an inverse quantal frame and X is an open Q-locale with x ∈ O(X), we shall refer to ς(x) as the support of x, and ς itself will be said to be the support of O(X), thus extending the terminology of section 2.3. Example 3.24 Let Q be an inverse quantal frame with base locale B. If X is an open B-locale then Q ⊗B O(X) defines an open Q-locale (cf. 3.11). Its support is defined by ς(a ⊗ x) = ς(aς(x)). The following are useful properties of open Q-locales: Theorem 3.25 Let Q be an inverse quantal frame and let X be an open Q-locale. 1. ς(ax) = ς(aς(x)) for all a ∈ Q and x ∈ O(X). 2. ς(ax) ≤ ς(a) for all a ∈ Q and x ∈ O(X). 3. ς(sx) = sς(x)s∗ for all s ∈ I(Q) and x ∈ O(X). Proof. Denoting by p and a the projection and the action of the correspond- ing G-locale and using the equality p ◦ a = d ◦ m ◦ (1 × (u ◦ p)) of 3.4 we prove 1: ς(ax) = (u ◦ p ◦ a)!(a ⊗ x) = (u ◦ d ◦ m ◦ (1 × (u ◦ p)))!(a ⊗ x) = ς(aς(x)) . Then 2 follows immediately: ς(ax) = ς(aς(x)) ≤ ς(ae) = ς(a); and 3 is a consequence of the inequalities sς(x)s∗ ≤ ss∗ ≤ e and ς(sx) = ς(sς(x)) ≤ (sς(x))(sς(x))∗ = sς(x)s∗ = ς(sς(x)s∗) ≤ ς(sς(x)) = ς(sx) . 27 Actions on local homeomorphisms. Let G be an ´etale groupoid. A G-sheaf is a G-locale whose projection is a local homeomorphism. The full subcategory of G-Loc whose objects are the G-sheaves (the classifying topos of G) is usually denoted by BG and the isomorphism G-Loc ∼= O(G)-Loc yields, by restriction, a corresponding full subcategory: Definition 3.26 Let Q be an inverse quantal frame with base locale B. By an ´etale Q-locale, or simply Q-sheaf, is meant a (necessarily open) Q-locale X such that the induced action of B on O(X) defines a B-sheaf. The category of ´etale Q-locales, denoted by Q-LH, is the full subcategory of Q-Loc whose objects are the ´etale Q-locales. This is the natural definition, for an inverse quantal frame Q, of a "Q- equivariant" sheaf on B. We borrow the following terminology from the B-sheaves of [68]: Definition 3.27 Let Q be an inverse quantal frame with base locale B, and let X be a Q-sheaf. The local sections of X are the local sections of X regarded as a B-sheaf; that is, a local section is an element s ∈ O(X) satisfying the equivalent conditions (2.30) and (2.36). The set of local sections of X is denoted by ΓX. Example 3.28 Any inverse quantal frame Q itself defines a Q-sheaf and we have I(Q) ⊂ ΓQ. Example 3.29 If Q is an inverse quantal frame and X is a Q-sheaf then Q ⊗B O(X) defines a Q-sheaf (cf. 3.24). An alternative notion of morphism of Q-sheaves, which maps local sec- tions to local sections in the same way that a natural transformation between sheaves does, is the following (cf. the sheaf homomorphisms of section 2.3): Definition 3.30 Let Q be an inverse quantal frame and let X and Y be Q-sheaves. A sheaf homomorphism h : O(X) → O(Y ) is a homomorphism of left Q-modules that preserves supports and local sections; that is, (3.31) (3.32) ς(h(x)) = ς(x) for all x ∈ O(X) h(ΓX) ⊂ ΓY . The category of Q-sheaves and sheaf homomorphisms between them is de- noted by Q-Sh. 28 Theorem 3.33 Let G be a localic ´etale groupoid and let Q = O(G). The categories BG, Q-LH and Q-Sh are isomorphic. Proof. BG and Q-LH are isomorphic by definition, so let us see that Q- LH and Q-Sh are isomorphic. Let X and Y be G-sheaves with actions a and b, respectively. If f : X → Y is a map of G-sheaves then f is a local homeomorphism and f! is necessarily a sheaf homomorphism of B-sheaves (cf. 2.3). From the equivariance condition (3.34) f ◦ a = b ◦ (1 × f ) we obtain, passing to direct images, the condition (3.35) f! ◦ a! = b! ◦ (1 ⊗ f!) and thus f! is also a homomorphism of Q-modules. Therefore the assignment f 7→ f! defines a faithful functor F : Q-LH → Q-Sh which is the identity on objects. Now let h : O(X) → O(Y ) be an arbitrary sheaf homomorphism of Q- sheaves. This is also a sheaf homomorphism of B-sheaves and thus it is the direct image f! of a locale map f : X → Y . The Q-equivariance of h is therefore the condition (3.35). We obtain the inverse image homomorphism version of (3.34) by taking right adjoints, and thus we conclude that F is full. 4 Groupoid sheaves as Hilbert modules In this section we begin by studying the notion of complete Hilbert mod- ule, by which is meant a Hilbert quantale module equipped with a "basis", following which we establish an equivalence of quantaloids, for a given involu- tive quantale Q, between the quantaloid of Q-valued sets and the quantaloid of complete Hilbert Q-modules (cf. 4.29). Then we specialize the theory of complete Hilbert modules to supported quantales and finally we prove, for an ´etale groupoid G, that the O(G)-sheaves can be identified with the complete Hilbert O(G)-modules, which leads to the envisaged equivalence between the classifying topos BG and the category of O(G)-valued sets. 4.1 Hilbert bases The terminology "Hilbert module" was introduced by Paseka [57] as an adap- tation to the context of involutive quantales of the notion of Hilbert C- module (see [42]), partly with the goal of relating aspects of the theory of 29 operator algebras to quantales (see, e.g., [58]). The notion of Hilbert basis appeared subsequently as a means of describing sheaves on involutive quan- tales and locales [66, 68], moreover in a way that meanwhile [30] has been related in a precise way to (ordered) sheaves on non-involutive quantales via the notion of principally generated module of [29]. Basic definitions and examples. Let us begin by recalling the notion of Hilbert module: Definition 4.1 ([57]) Let Q be an involutive quantale. By a pre-Hilbert Q- module will be meant a left Q-module X equipped with a binary operation h−, −i : X × X → Q , called the inner product, which for all x, xi, y ∈ X and a ∈ Q satisfies the following axioms: (4.2) (4.3) (4.4) hax, yi = ahx, yi _i xi, y+ = _i hx, yi = hy, xi∗ . hxi, yi By a Hilbert Q-module will be meant a pre-Hilbert Q-module whose inner product is "non-degenerate": (4.5) hx, −i = hy, −i ⇒ x = y . We remark that, in particular, inner products are "sesquilinear forms": (4.6) Dx,_ aiyiE =_ hx, yiia∗ i . ha, bi = ab∗ . Example 4.7 Q itself is a pre-Hilbert Q-module [57] with the inner product defined by The inner product is non-degenerate if Q is unital. More generally, if I is a set then the set QI of maps v : I → Q is a left Q-module with the usual function module structure given by pointwise joins and multiplication on the left, and it is a pre-Hilbert module with the inner product hv, wi = v · w given by the standard dot product formula (We adopt, for functions in QI and their values, the same notation as for vectors and their components in linear algebra -- cf. section 4.2.) v · w =_α∈I vαw∗ α . 30 Example 4.8 If p : X → B is a local homeomorphism of locales then O(X) is a Hilbert O(B)-module whose inner product is defined by hx, yi = p!(x∧y) -- cf. (2.34). Hilbert sections. Let us pursue the analogy between Hilbert modules and sheaves suggested by 4.8 and define what should be meant in general by a "section" of a Hilbert module, using 2.35 as motivation: Definition 4.9 Let Q be an involutive quantale and let X be a pre-Hilbert Q-module. By a Hilbert section of X is meant an element s ∈ X such that hx, sis ≤ x for all x ∈ X. The set of all the Hilbert sections of X is denoted by ΓX . We say that the Hilbert module X is complete, or that it has enough sections, if for all x ∈ X we have the equality x = _s∈ΓX hx, sis . Any set Γ ⊂ X such that x =Ws∈Γ hx, sis for all x ∈ X is called a Hilbert basis (in particular, we have Γ ⊂ ΓX and Γ is a set of Q-module generators for X). The name "Hilbert basis" is suggested by the obvious formal resemblance with the properties of a Hilbert basis of a Hilbert space. Of course, a Hilbert basis in our sense is not an actual basis as in linear algebra because there is no freeness, but for the sake of simplicity and following [68] we retain this terminology. Example 4.10 The Hilbert O(B)-module determined by a local homeomor- phism of locales p : X → B (cf. 4.8) is complete with ΓX as a Hilbert basis [68]. Furthermore, ΓX is an actual basis of X in the sense of locale theory (the analogue for locales of a basis of a topological space). Example 4.11 If Q is a unital involutive quantale, then Q itself, regarded as a Hilbert Q-module with ha, bi = ab∗ as in 4.7, has a set of Hilbert sections ΓQ = {s ∈ Q s∗s ≤ e} . This set is a Hilbert basis, and so is the singleton Γ = {e}. Example 4.12 The condition W ΓX = 1 of 4.10 does not necessarily hold over more general quantales. In order to see this let R be the unital involutive 31 quantale whose involution is trivial and whose order and multiplication table are the following (cf. [65, Example 4.21]): e >>>>>>> 1 0 ???????? a 0 e a 1 0 0 0 0 0 e 0 e a 1 a 0 a 1 1 1 0 1 1 1 If we regard R as a Hilbert R-module with hx, yi = xy∗ then R has enough sections (because it is a unital quantale) but ΓR = {0, e}. The existence of a Hilbert basis has useful consequences. In particular the inner product is necessarily non-degenerate: Lemma 4.13 Let Q be an involutive quantale, let X be a pre-Hilbert Q- module, and let Γ ⊂ X. If Γ is a Hilbert basis then the following properties hold, for all x, y ∈ X. 1. If hx, si = hy, si for all s ∈ Γ then x = y. (Hence, X is a Hilbert module.) 2. hx, yi =Ws∈Γ hx, sihs, yi =Ws∈Γ hx, sihy, si∗. ("Parseval's identity".) Conversely, Γ is a Hilbert basis if h−, −i is non-degenerate and 2 holds. Proof. Assume that Γ is a Hilbert basis. The two properties are proved as follows. 1. If hx, si = hy, si for all s ∈ Γ then x =Ws∈Γ hx, sis =Ws∈Γ hy, sis = y. 2. hx, yi =(cid:10)Ws∈Γ hx, sis, y(cid:11) =Ws∈Γ hx, sihs, yi. For the converse assume that h−, −i is non-degenerate and that 2 holds. Then for all x, y ∈ X we have _s∈Γ hx, sis, y+ = _s∈Γ hx, sihs, yi = hx, yi , and by the non-degeneracy we obtainWs∈Γ hx, sis = x. 32 Adjointable maps. Similarly to Hilbert C-modules, the module homo- morphisms which have "operator adjoints" play a special role: Definition 4.14 ([57]) Let Q be an involutive quantale and let X and Y be pre-Hilbert Q-modules. A function ϕ : X → Y is adjointable if there is another function ϕ† : Y → X such that for all x ∈ X and y ∈ Y we have hϕ(x), yi = hx, ϕ†(y)i . (The notation for ϕ† in [57] is ϕ∗, but we want to avoid confusion with the notation for inverse image homomorphisms of locale maps.) We note that if ϕ is adjointable and Y is a Hilbert Q-module (i.e., the bilinear form of Y is non-degenerate) then ϕ is necessarily a homomorphism of Q-modules [57]: we have Dϕ(cid:16)_ aixi(cid:17) , yE = D_ aixi, ϕ†(y)E =_ aihxi, ϕ†(y)i = _ aihϕ(xi), yi =D_ aiϕ(xi), yE ϕ(cid:16)_ aixi(cid:17) =_ aiϕ(xi) . and thus by the non-degeneracy of h−, −iY we conclude that Conversely, and similarly to the situation in [68] where Q was a locale, the homomorphisms of complete Hilbert Q-modules are necessarily adjointable. In order to prove this only the domain module need have enough sections: Theorem 4.15 Let Q be an involutive quantale and let X and Y be pre- Hilbert Q-modules such that X has a Hilbert basis Γ (hence, X is a Hilbert module), and let ϕ : X → Y be a homomorphism of Q-modules. Then ϕ is adjointable with a unique adjoint ϕ†, which is given by (4.16) ϕ†(y) =_t∈Γ hy, ϕ(t)it . 33 Proof. Let x ∈ X, y ∈ Y , and let us compute hx, ϕ†(y)i using (4.16): hy, ϕ(t)it+ hx, sis,_t∈Γ hx, sihs, tihy, ϕ(t)i∗ hx, ϕ†(y)i = _s∈Γ = _s,t∈Γ = _t∈Γ = _t∈Γ = ϕ _t∈Γ = hϕ(x), yi . hx, tihϕ(t), yi hx, tiϕ(t), y+ hx, tit! , y+ This shows that ϕ† is adjoint to ϕ, and the uniqueness is a consequence of the non-degeneracy of the inner product of X. Definition 4.17 Let Q be an involutive quantale. The category of complete Hilbert Q-modules, denoted by Q-HMB (standing for 'Hilbert Modules with Basis'), is the category whose objects are the complete Hilbert Q-modules and whose arrows are the homomorphisms of Q-modules (equivalently, the adjointable maps). Corollary 4.18 For any involutive quantale Q, Q-HMB is an involutive quantaloid whose involution is the strong self-duality (−)† : (Q-HMB)op → Q-HMB. Example 4.19 If B is a frame, the category B-Sh (cf. section 2.3) coincides with Map(B-HMB). Moreover, if f : X → Y is a map of B-sheaves we have f! = (f ∗)† [68, Theorem 8]. 4.2 Quantale-valued sets Most of the definitions of sheaf for involutive quantales in the literature are based on generalizations of the notion of frame-valued set. In this section we take one such definition and show, for an arbitrary involutive quantale Q, that the category of Q-valued sets is equivalent to the category of maps of the quantaloid of complete Hilbert Q-modules. 34 Basic definitions. Let Q be an involutive quantale, and let I and J be sets. By a Q-valued matrix of type I × J will be meant a mapping A : I × J → Q . Terminology and notation are analogous to those of linear algebra: • aαβ or (A)αβ denotes the value A(α, β), and we refer to α and β as the row and column indices, respectively; • if I is a singleton we say that A is a row matrix, and if J is a singleton we say that A is a column matrix ; • the transpose of A is the matrix AT defined by (AT )αβ = aβα, and the adjoint of A is the matrix A∗ defined by (A∗)αβ = a∗ βα; • if B : J × K → Q is another matrix, the product AB is defined by (AB)αγ = _β∈J aαβbβγ . In accordance with these conventions, we shall often think of a mapping v : I → Q as being a "vector" (cf. 4.7), in particular writing vα instead of v(α) and adopting the following notation and terminology with respect to a matrix A : I × J → Q: • vA : J → Q is the mapping whose components are defined by (vA)β =_α∈I vαaαβ ; that is, v is always regarded as being a row matrix {∗} × I → Q; • if α ∈ I, the α-row of A is the mapping α : I → Q defined by αβ = aαβ; • if β ∈ J, the β-column of A is the mapping β : J → Q defined by βα = aαβ. The following definition stems from some of the early works on categories of relations [18, §2.226; 60, Prop. 2.6]. We adapt it from Garraway [20], who applies it to non-unital involutive quantaloids rather than just involutive quantales. (The same definition has been used earlier by Gylys [26] for unital involutive quantaloids.) 35 Definition 4.20 Let Q be an involutive quantale. By a Q-set is meant a set I together with a matrix A : I × I → Q which is both self-adjoint (A = A∗) and idempotent (AA = A). The matrix entry aαβ of a Q-set (I, A) can be regarded as the generalized truth-value of the equality α = β, and the diagonal entry aαα is regarded as the extent to which the element α exists. The fact that A is required to be a projection matrix reflects the fact that equality should be a partial equivalence relation. These ideas lead naturally to the following definition of a Q-valued relation between Q-sets (another common name for this is distributor or bimodule): Definition 4.21 Let Q be an involutive quantale and let X = (I, A) and Y = (J, B) be Q-sets. A relation R : X−→7 Y is a matrix R : J × I → Q such that the following equations hold: (4.22) BR = R = RA . We shall denote by Rel(Q) the quantaloid whose objects are the Q-sets and whose morphisms are the relations between them: composition is given by matrix multiplication and the identity relation on a Q-set (I, A) is A. From this a notion of map immediately follows (cf. 2.4): Definition 4.23 Let Q be an involutive quantale. The category of Q-sets Set(Q) is defined to be Map(Rel(Q)). Hence, explicitly, a map of Q-sets F : (I, A) → (J, B) is a relation such that the following two additional conditions hold: (4.24) (4.25) F F ∗ ≤ B A ≤ F ∗F . (Note: we write −→7 for general relations and → for maps.) 36 Matrices versus modules. Every Hilbert Q-module X with a Hilbert basis Γ has an associated matrix A : Γ ×Γ → Q defined by ast = hs, ti (this is analogous to the metric of an Euclidian space with respect to a chosen basis). This defines a Q-set (Γ , A) because we have A = A∗ by definition of the inner product, and A = A2 by "Parseval's identity" (cf. 4.13-2). Conversely, we have: Lemma 4.26 Let Q be an involutive quantale. For any Q-set (I, A) the subset of QI defined by QIA = {vA v ∈ QI} is a Hilbert Q-module whose inner product is the dot product of QI (cf. 4.7), hv, wi = v · w = vw∗ =_α∈I vαw∗ α , it has a Hilbert basis Γ consisting of all the rows of A, and for all α, β ∈ I and v ∈ QI we have (4.27) (4.28) Dv, βE = (vA)β , D α, βE = aαβ . Proof. The assignment j : v 7→ vA is a Q-module endomorphism of QI, and QI A is its image, hence a submodule of QI. Next note that Γ is a subset of QI A because for each α ∈ I we have α = αA ∈ QI A: Now we prove (4.27): αβ = aαβ = (A2)αβ =_γ∈I aαγaγβ =_γ∈I Dv, βE = v · β =_γ βγ =_γ γ =_γ vγ β∗ vγa∗ αγaγβ = ( αA)β . vγaγβ = (vA)β . In particular, if v ∈ QI A we have vA = v, henceDv, βE = vβ, and (4.28) is an immediate consequence: Finally, Γ is a Hilbert basis because for all v ∈ QIA we have _β Dv, βE β! α = _β vβ vβaβα = (vA)α = vα . Dα, βE = αβ = aαβ . β! βα =_β =_β vβ α 37 hψ†(s), uiY hu, ϕ(t)iY =_u (G(ψ))su(G(ϕ))ut = (G(ψ)G(ϕ))st . hs, ψ(u)iZhu, ϕ(t)iY = _u∈ΓY = _u Theorem 4.29 Let Q be an involutive quantale. The involutive quantaloids Rel(Q) and Q-HMB are equivalent. Proof. For each complete Hilbert Q-module X let G(X) = (ΓX, AX ) be the Q-set defined by (AX)st = hs, ti, and for each homomorphism ϕ : X → Y in Q-HMB let G(ϕ) : ΓY × ΓX → Q be the matrix defined by (G(ϕ))st = hs, ϕ(t)iY . It is straightforward to see that these assignments define a faithful homomorphism of quantaloids G : Q-HMB → Rel(Q) (in particular faith- fulness is a consequence of the non-degeneracy of the inner products), and we prove only that G preserves composition: for all homomorphisms ϕ : X → Y and ψ : Y → Z we have (G(ψ ◦ ϕ))st = hs, ψ(ϕ(t))iZ = hψ†(s), ϕ(t)iY We remark that this also shows that G(ϕ) is a morphism in Rel(Q): AY G(ϕ) = G(idY )G(ϕ) = G(idY ◦ ϕ) = G(ϕ) = G(ϕ)G(idX) = G(ϕ)AX . Next we show, for an arbitrary Q-set (I, A), that (I, A) ∼= G(QIA). Writ- ing X for QIA, let R : ΓX ×I → Q be the matrix given by rσα = hσ, αi = σα. This defines a relation R : (I, A)−→7 G(X): (AXR)σα = Wτ ∈ΓX hσ, τ iτα = (Wτ hσ, τ iτ )α = σα = rσα , (RA)σα = Wβ∈I hσ, βiaβα =Wβ hσ, βi βα =(cid:16)Wβ hσ, βi β(cid:17)α = σα = rσα . The above two lines are justified, respectively, because ΓX and { α α ∈ I} are Hilbert bases of X. The same facts show that R is a unitary map: (RR∗)στ = Wα∈I hσ, αihτ , αi∗ =Wα hσ, αih α, τ i = hσ, τ i = (AX)στ , (R∗R)αβ = Wσ∈ΓX h α, σihσ, βi = h α, βi = aαβ . Now let us prove that (R, X) is a universal arrow from (I, A) to G. Let Y be a complete Hilbert Q-module and let H : (I, A)−→7 G(Y ). Define a mapping ϕ : X → Y as follows: (4.30) ϕ(v) = _s ∈ I t ∈ ΓY vsh∗ tst . 38 This is a homomorphism of left Q-modules because vA = v and, by (4.27), vs = hv, si. In order to prove the universal property we only need to show (because R is an isomorphism and G is faithful) that G(ϕ)R = H: for all α ∈ ΓY and all β ∈ I we have hϕ†(α), σihσ, βi hα, βsh∗ tsti = hϕ†(α), βi = hα, ϕ( β)i (G(ϕ)R)αβ = _σ∈ΓX = α, _s ∈ I = _s, t hα, ϕ(σ)ihσ, βi = _σ∈ΓX tst+ = _s ∈ I s = _s, t t ∈ ΓY hα, tihts β∗ βsh∗ = (AY HA)αβ = hαβ . t ∈ ΓY hα, tihtshs, βi We thus conclude that G has a left adjoint and that the unit of the adjunction is a unitary map. Therefore, in order to establish the desired equivalence we only need to prove that the co-unit is an isomorphism (cf. comments after 2.6). So let X be an arbitrary complete Hilbert Q-module, and, for simplicity, let us write G(X) = (Γ , A) instead of (ΓX, AX). Let ϕ : QΓ → X be the Q-module quotient defined by ϕ(v) =Ws∈Γ vss. By 4.13 we have, for all v, w ∈ QΓ , the following equivalences: ϕ(v) = ϕ(w) ⇐⇒ ∀t∈Γ hϕ(v), ti = hϕ(w), ti ⇐⇒ ∀t∈Γ D_s∈Γ ⇐⇒ ∀t∈Γ _s vss, tE =D_s vshs, ti =_s ⇐⇒ ∀t∈Γ (vA)t = (wA)t ⇐⇒ vA = wA . wshs, ti wss, tE Hence, ϕ factors uniquely through the quotient v 7→ vA : QΓ → QΓ A and ∼=→ X, which is the X-component of the an isomorphism of Q-modules QΓ A co-unit of the adjunction. Corollary 4.31 For any involutive quantale Q, the category Set(Q) is equiv- alent to Map(Q-HMB) (cf. 2.4). 4.3 Supported modules Let us provide an independent study of Hilbert modules on supported quan- tales. This has two purposes: one is to achieve a better understanding of how 39 the various axioms interact with each other; and the other is that by doing so one is paving the way for obtaining possible extensions of the theory devel- oped in this paper in a way that may be applicable to theories of sheaves on supported quantales that are more general than inverse quantal frames. As an example, we mention the Lindenbaum quantales for propositional normal modal logic of [44], which are stably supported and whose sheaves may pro- vide good semantic grounds for interpreting non-propositional modal logic. Modules on supported quantales. We begin with a simple but useful property of arbitrary modules on supported quantales: Lemma 4.32 Let Q be a supported quantale and let X be a left Q-module. Then for all a ∈ Q we have a1X = ς(a)1X = aa∗1X = aa∗a1X = aa∗aa∗1X . Proof. Let a ∈ Q. The axioms of supported quantales give us a1X ≤ ς(a)a1X ≤ ς(a)1X ≤ aa∗1X ≤ ς(a)aa∗1X ≤ aa∗aa∗1X ≤ aa∗a1X ≤ a1X . Let us introduce a notion of support for modules that is formally similar to that of quantales if we replace ab∗ by ha, bi (however, we require supports to be only monotone instead of sup-preserving -- cf. 2.21): Definition 4.33 Let Q be a supported quantale with base locale B (cf. 2.14). By a supported Q-module is meant a pre-Hilbert Q-module X equipped with a monotone map ς : X → B , called the support of X, such that the following properties hold for all x ∈ X: ς(x) ≤ hx, xi x ≤ ς(x)x . Example 4.34 Any supported quantale Q defines a supported module over itself, with ha, bi = ab∗. The existence of the base locale B enables us to define a notion of local section in analogy to that of local homeomorphisms regarded as B-modules (cf. sections 2.3 and 3.3), since the action of Q restricts to an action of B making X a supported B-module: 40 Definition 4.35 Let Q be a supported quantale and let X be a supported Q-module. By a local section of X is meant an element s ∈ X such that ς(x)s = x for all x ≤ s. The set of local sections of X is denoted by Γ ℓ X . But, as we shall see, in the cases of interest later in the paper the local sections coincide with the Hilbert sections due to the following proposition: Lemma 4.36 Let Q be a supported quantale and let X be a supported Q- module. 1. Γ ℓ X = {s ∈ X ς(x ∧ s)s ≤ x for all x ∈ X}. 2. Γ ℓ X is downwards closed. 3. ΓX ⊂ Γ ℓ X. 4. ΓX = Γ ℓ sections. X if and only if every local section s is a join s =Wi ti of Hilbert Proof. 1 is proved in the same way as the equivalence of (2.30) and (2.36) in 2.35: if s is a local section and x ∈ X then x ∧ s ≤ s and thus we have ς(x ∧ s)s = x ∧ s ≤ x; conversely, if s satisfies ς(x ∧ s)s ≤ x for all x ∈ X then if x ≤ s we have x = ς(x)x ≤ ς(x)s = ς(x ∧ x)s ≤ ς(x ∧ s)s ≤ x. Now 2 is an immediate consequence because if s is a local section and t ≤ s we have ς(x ∧ t)t ≤ ς(x ∧ s)s ≤ x. Similarly, 3 follows from the inequality ς(x ∧ s) ≤ hx, si. In order to prove the nontrivial inclusion in 4 let s be a local section For all x ∈ X we have ς(x∧s)s ≤ x, and thus also ς(x∧t)u ≤ x. In particular, making x = t we obtain ς(t)u ≤ t. The conclusion that s is a Hilbert section and let I ⊂ ΓX be such that s =W I. Let t and u be arbitrary elements of I. follows immediately, since for all x ∈ X we have hx, sis =Wt,u∈I hx, tiu and hx, tiu = hx, ς(t)tiu = hx, tiς(t)u ≤ hx, tit ≤ x. Stably supported modules. The notion of stable support for quantales has an equally useful analogue for modules: Definition 4.37 Let Q be a supported quantale and let X be a supported Q-module. The support is called stable, and the module is said to be stably supported, if in addition the support is B-equivariant; that is, the following condition holds for all b ∈ B and x ∈ X: ς(bx) = b ∧ ς(x) . Example 4.38 Any stably supported quantale Q is itself a stably supported Q-module, with ha, bi = ab∗. 41 Lemma 4.39 Let Q be a supported quantale and let X be a stably supported Q-module. The map b 7→ b1X from B to X is right adjoint to ς : X → B. (In particular, ς preserves joins.) Proof. The unit of the adjunction follows from x ≤ ς(x)x ≤ ς(x)1X , and the co-unit follows from the B-equivariance: ς(b1X ) = b ∧ ς(1X) ≤ b. There are several alternative definitions of stability: Lemma 4.40 Let Q be a supported quantale and let X be a supported Q- module. The following conditions are equivalent: 1. ς(ax) = ς(aς(x)) for all a ∈ Q and x ∈ X; 2. ς(ax) ≤ ς(a) for all a ∈ Q and x ∈ X; 3. ς(a1X ) ≤ ς(a) for all a ∈ Q; 4. The support of X is stable. Proof. 2 and 3 are of course equivalent. Let us prove the equivalence of 1 and 2. First assume 1. Then for all a ∈ Q and x ∈ X we have ς(ax) = ς(aς(x)) ≤ ς(ae) = ς(a) . Now assume 2. Then we have ς(ax) ≤ ς(aς(x)x) ≤ ς(aς(x)) ≤ ς(ahx, xi) = ς(hax, xi) ≤ ς(hς(ax)ax, xi) = ς(ς(ax)hax, xi) ≤ ς(ς(ax)) = ς(ax) , and thus 1 holds. Now assume again 1, and let b ∈ B. Then ς(bx) = ς(bς(x)) = ς(b ∧ ς(x)) = b ∧ ς(x) , and thus we see that 1 implies 4. Finally, assume that 4 holds. Then for all a ∈ Q we have, using 4.32, ς(a1X) = ς(ς(a)1X) = ς(a) ∧ ς(1X) ≤ ς(a) , and thus 4 implies 3. Remark 4.41 Since a stably supported quantale is a stably supported mod- ule over itself and 4.40-1 translates to the definition of stability for supported quantales [cf. (2.10], it follows that a supported quantale Q is stably sup- ported if and only if its support is B-equivariant (cf. 2.20). This fact is not mentioned in [65]. 42 Supported modules on stably supported quantales. In order for some of the good properties of stable supports of quantales to carry over to supports of modules we need the quantale Q itself to be stably supported. First we observe the following: Lemma 4.42 Let Q be a stably supported quantale. Any supported Q-module is necessarily stably supported. Proof. For all a ∈ Q and x ∈ X we necessarily have ς(ax) ≤ ς(a), as the following sequence of (in)equalities shows, ς(ax) ≤ hax, axi ∧ e = ahx, axi ∧ e ≤ a1Q ∧ e = ς(a) , where the latter equality is (2.15). We obtain similar properties to those of stable supports of quantales regarding uniqueness of supports, and analogous formulas for ς when we formally replace ab∗ by ha, bi: Lemma 4.43 Let Q be a stably supported quantale and let X be a (neces- sarily stably) supported Q-module. The following properties hold: 1. For all x, y ∈ X we have ς(hx, yi) ≤ ς(x) = ς(hx, xi) = ς(hx, 1i). 2. For all x ∈ X and a ∈ Q we have ς(x)a = hx, 1i ∧ a. 3. For all x ∈ X we have ς(x) = hx, 1i ∧ e. 4. For all x ∈ X we have ς(x) = hx, xi ∧ e. 5. X does not admit any other support. Proof. First we prove 1. Let x, y ∈ X. Using the stability of the support of Q we have ς(hx, yi) ≤ ς(hς(x)x, yi) = ς(ς(x)hx, yi) ≤ ς(ς(x)) = ς(x) . On the other hand, using the inequality just proved we obtain ς(x) = ς(ς(x)) ≤ ς(hx, xi) ≤ ς(hx, 1i) ≤ ς(x) , thus proving 1. Now let us prove 2. We have ς(x)a ≤ hx, 1i because ς(x)a ≤ hx, xi1 ≤ hx, 1i1 = hx, 1∗1i = hx, 1i . 43 Since we also have ς(x)a ≤ a we obtain the inequality ς(x)a ≤ hx, 1i ∧ a , and the converse inequality is proved as follows, using 1: hx, 1i ∧ a ≤ ς(hx, 1i ∧ a)(hx, 1i ∧ a) ≤ ς(hx, 1i)a = ς(x)a . Making a = e we obtain 3 (which immediately implies that the support of X is unique), and for 4 it suffices to prove the inequality hx, xi ∧ e ≤ ς(x), again using 1: hx, xi ∧ e = ς(hx, xi ∧ e) ≤ ς(hx, xi) ∧ ς(e) = ς(x) ∧ e = ς(x) . One consequence of this is that the existence of a support (when Q is stably supported) is a property of a pre-Hilbert Q-module rather than extra structure. In fact this uniqueness is even "pointwise", in the following sense: Lemma 4.44 Let Q be a stably supported quantale, let X be a (necessarily stably) supported Q-module, and let x ∈ X and b ∈ B be such that b ≤ hx, xi , x ≤ bx . Then we necessarily have b = ς(x). Proof. Using the B-equivariance of the support of X we obtain ς(x) ≤ ς(bx) = b ∧ ς(x) ≤ b , and, conversely, b = ς(b) ≤ ς(hx, xi) = ς(x). Finally, if there exists a Hilbert basis we obtain: Theorem 4.45 Let Q be a stably supported quantale. Any complete Hilbert Q-module is a (necessarily stably) supported Q-module. Proof. Define ς(x) = hx, xi ∧ e for all x ∈ X, and let Γ be a Hilbert basis of X. By definition, in order to verify that ς is a support it only remains to be seen that x ≤ ς(x)x for all x ∈ X. Let then s ∈ Γ . We have, using the properties of the stable support of Q, hx, si = ς(hx, si)hx, si = (hx, sihx, si∗ ∧ e)hx, si ≤ (hx, xi ∧ e)hx, si = ς(x)hx, si = hς(x)x, si . Hence, we have x ≤ ς(x)x due to 4.13, and by 4.42 we conclude that X is stably supported. 44 4.4 Groupoid sheaves Now we achieve the main aim of this paper, which is to show that if G is an ´etale groupoid then the classifying topos BG is equivalent to Set(O(G)) or, equivalently, to Map(O(G)-HMB). We shall do this by showing that the O(G)-sheaves of section 3.3 coincide with the complete Hilbert O(G)- modules. From complete Hilbert Q-modules to Q-sheaves. We begin by recall- ing an observation from [30, Example 4.7-3]: Lemma 4.46 Let Q be an inverse quantal frame. If X is a complete Hilbert Q-module thenW ΓX = 1. In order to bypass differences in notation and make this paper more self- contained we include the proof here: Proof. First, if s ∈ I(Q) and t ∈ ΓX then st ∈ ΓX because for all x ∈ X we have hx, stist = hx, tis∗st ≤ hx, tit ≤ x . Hence, for all t ∈ ΓX we have 1Qt =Ws∈I(Q) st ≤W ΓX , and thus 1X = _t∈ΓX h1X, tit ≤ _t∈ΓX 1Qt ≤_ ΓX . Theorem 4.47 Let Q be an inverse quantal frame. Every complete Hilbert Q-module is a Q-sheaf. Proof. Let X be an arbitrary but fixed complete Hilbert Q-module, and let us just write Γ for the Hilbert basis ΓX. The action restricts to an action of the base locale B ⊂ Q and we shall prove that this defines a B-sheaf by showing that it has a Hilbert B-module structure with respect to which Γ is a Hilbert basis. First, by the local inner product will be meant the operation h−, −iℓ : X × X → B defined by (4.48) hx, yiℓ = hx, yi ∧ e . The operation h−, −iℓ is of course symmetric, and it preserves joins in the left variable because h−, −i does and Q is a locale: xi, y+ℓ _i =_i xi, y+∧e = _i hxi, yi!∧e =_i hxi, yi∧e =_i hxi, yiℓ . 45 We show that h−, −iℓ is B-equivariant in the left variable, using 2.19: hbx, yiℓ = hbx, yi ∧ e = bhx, yi ∧ e = b ∧ hx, yi = b ∧ e ∧ hx, yi = b ∧ hx, yiℓ . Hence, X together with the local inner product is a pre-Hilbert B-module. In order to see that this is a B-sheaf let us prove that Γ is itself a Hilbert basis for the pre-Hilbert B-module structure; that is, we shall prove that for all x ∈ X we have x =Ws∈Γ hx, siℓs. One inequality is trivial: hx, siℓs . hx, sis ≥ _s∈Γ (hx, si ∧ e)s = _s∈Γ x = _s∈Γ In order to prove the other inequality, first we apply (2.18) with b = hx, siℓ and a = hs, ti: hx, siℓhs, ti = _s∈Γ (hx, si ∧ e)1 ∧ hs, ti . hx, siℓ1 ∧ hs, ti hx, siℓs, t+ = _s∈Γ = _s∈Γ (a ∧ e)1 ≥ _yz ∗≤a y ∧ z . _s∈Γ (4.49) (4.50) Now recall the inequality (2.22): Applying this to the right hand side of (4.50) we obtain (4.51) _s∈Γ (hx, si ∧ e)1 ∧ hs, ti ≥ _s∈Γ _yz ∗≤hx,si y ∧ z ∧ hs, ti . A particular choice of y and z for which yz∗ ≤ hx, si is to take y = hx, ti and z = hs, ti, and thus with these values of y and z the right hand side of (4.51) is greater than or equal to _s∈Γ y ∧ z ∧ hs, ti = _s∈Γ hx, ti ∧ hs, ti hx, ti ∧ hs, ti ∧ hs, ti = _s∈Γ hs, ti = hx, ti ∧_s∈Γ = hx, ti ∧_s∈Γ = hx, ti ∧ h1, ti = hx, ti , s, t+ where the transition to the last line follows from 4.46. Hence, we have con- cluded thatDWs∈Γ hx, siℓs, tE ≥ hx, ti for all t ∈ Γ , which finally gives us: hx, siℓs ≥ x . _s∈Γ 46 From Q-sheaves to complete Hilbert Q-modules. We begin with a simple technical lemma: Lemma 4.52 Let Q be an inverse quantal frame and let X be a Q-sheaf. The action of Q on O(X) restricts to a monoid action of I(Q) on the set of local sections ΓX. Proof. Let t ∈ ΓX and s ∈ I(Q). We want to prove that st ∈ ΓX. So let x ≤ st and let us show that x = ς(x)st. First we note that s∗x ≤ s∗st ≤ t, and thus (4.53) ς(s∗x)t = s∗x because t is a section. Secondly, we have ς(x) ≤ ς(st) ≤ ς(s) = ss∗, and thus (4.54) ss∗x = x . Hence, since s = ss∗s and both ς(x) and ss∗ belong to the base locale B ⊂ Q, applying (4.53) -- (4.54) and the equality ς(s∗x) = s∗ς(x)s from 3.25-2 we obtain ς(x)st = ς(x)ss∗st = ss∗ς(x)st = sς(s∗x)t = ss∗x = x . Theorem 4.55 Let Q be an inverse quantal frame. Every Q-sheaf is a com- plete Hilbert Q-module. Proof. Let X be an arbitrary but fixed Q-sheaf, and let us denote the base locale of Q by B. The proof has two parts: first, we construct a Q-set (I, M); then, we prove that the Hilbert Q-module QIM (cf. 4.26) is isomorphic to O(X). For the set I we take the set ΓX of local sections of X. And the matrix M : I × I → Q is defined by I(Q)st = {a ∈ I(Q) ς(a∗) ≤ ς(t) and at ≤ s} , (4.56) (4.57) mst = _ I(Q)st . First we remark that the condition at ≤ s in (4.56) is equivalent to (4.58) ς(at)s = at because, by 4.52, at is a local section [cf. (2.30)]. Next we mention that the two conditions ς(a∗) ≤ ς(t) and at ≤ s in (4.56) also imply ς(a) ≤ ς(s), since (4.59) ς(a) = ς(aς(a∗)) ≤ ς(aς(t)) = ς(at) ≤ ς(s) . 47 In addition, for all a ∈ I(Q)st we have a∗s ≤ t because from (4.59) we obtain ς(a) ≤ ς(at), and from at ≤ s and (4.58) we obtain aa∗s = ς(a)s ≤ ς(at)s = at, whence a∗s = a∗aa∗s ≤ a∗at ≤ t . It follows that a∗ ∈ I(Q)ts, and we conclude that M ∗ = M. Furthermore, for all a ∈ I(Q)st and b ∈ I(Q)tu we have ab ∈ I(Q)su: ς((ab)∗) = ς(b∗a∗) ≤ ς(b∗) ≤ u abu ≤ at ≤ s . Hence, mstmtu ≤ msu. This shows that we have M 2 ≤ M, which is equivalent to M 2 = M because Q is a supported quantale (cf. 6.4 in the appendix), and thus (I, M) is a Q-set. Therefore, by 4.26, we have a Hilbert Q-module QIM, whose inner prod- uct is the dot product of QI , with a Hilbert basis Γ consisting of the rows of M. As usual (cf. section 4.2), for each s ∈ I we denote the s-row of M by s. The local inner product h−, −iℓ of QIM, as defined by (4.48), satisfies, for all s, t ∈ I, ℓ hs, ti = s · t ∧ e = mst ∧ e = _{a ∧ e a ∈ I(Q) and ς(a∗) ≤ ς(t) and at ≤ s} ≤ _{a ∧ e a ∈ I(Q) and ς((a ∧ e)∗) ≤ ς(t) and (a ∧ e)t ≤ s} = _{b ∈ B ς(b∗) ≤ ς(t) and bt ≤ s} = _{b ∈ B ς(b∗) ≤ ς(t) and bt ≤ s} ∧ e ≤ _{a ∈ I(Q) ς(a∗) ≤ ς(t) and at ≤ s} ∧ e = mst ∧ e = hs, tiℓ . Therefore all the expressions in the above derivation are equal, and we obtain hs, tiℓ = _{b ∈ B ς(b∗) ≤ ς(t) and bt ≤ s} = _{b ∈ B b ≤ ς(t) and bt ≤ s} = ς(s ∧ t) . Since ς(s ∧ t) is the B-valued inner product of O(X) regarded as a B-sheaf, we conclude, by 4.29 (or [68, Lemma 5]), that O(X) ∼= BIM ℓ, where M ℓ is the matrix defined by (M ℓ)st = ς(s ∧ t). Since this matrix is also that of the local inner product of QIM, it follows that BIM ℓ ∼= QIM, and thus O(X) is a complete Hilbert Q-module. 48 Remark 4.60 We can provide a more geometric interpretation of the for- mula (4.57) in terms of an ´etale groupoid G rather than its quantale O(G), whereby mst is the union of all the local bisections a that satisfy the following three conditions (cf. 2.24): • the domain of a is contained in the domain of the local section s; • the image of r ◦ a is contained in the domain of the local section t; • a acts on t yielding a subsection of s. Hence, the logical interpretation of G-sheaves via O(G)-sets is a generaliza- tion of that of frame-valued sets, with the truth values now defining "equal- ity" of local sections up to local translations rather than just restriction. The classifying topos of an ´etale groupoid. From [30, Theorem 4.1], which applies to modular quantal frames, and hence also to inverse quantal frames (cf. 2.29), it follows that if Q is an inverse quantal frame then a left Q-module is a complete Hilbert Q-module with respect to at most one inner product. Hence, being such a module is a property rather than extra structure. Similarly, being a Q-sheaf is a property, and the results that we have just obtained show that the two properties coincide. Hence, in order to conclude our comparison of the classifying topos BG of an ´etale groupoid G and Set(O(G)) we only need to see how the morphisms of O(G)-Sh and Map(O(G)-HMB) relate to each other: Lemma 4.61 Let Q be an inverse quantal frame. The sheaf homomorphisms of Q-sheaves (cf. 3.30) coincide with the arrows of Map(Q-HMB). Proof. Let X and Y be Q-sheaves and let ϕ : O(X) → O(Y ) be a sheaf homomorphism of Q-sheaves. This is also a sheaf homomorphism of B- sheaves, where B ⊂ Q is the base locale of Q, and thus we have ϕ = f! for a map of B-sheaves f : X → Y (cf. 4.19 and section 2.3). The inner products of X and Y as B-sheaves are defined by the formula (4.48), which we recall: hz, z′iℓ = hz, z′i ∧ e . Hence, the adjoint of ϕ, which is given by hϕ(x), yi = hx, ϕ†(y)i, also satisfies hϕ(x), yiℓ = hx, ϕ†(y)iℓ, showing that ϕ† is the adjoint of ϕ also with respect to the B-sheaf structures of X and Y . Therefore we have f ∗ = ϕ†, showing that ϕ† is right adjoint to ϕ; that is, ϕ is a morphism in Map(Q-HMB). Conversely, let ϕ : O(X) → O(Y ) be a morphism in Map(Q-HMB). By the same reasoning as above, this is also a morphism in Map(B-HMB), and thus ϕ is a sheaf homomorphism of B-sheaves. Since it is also a homomor- phism of Q-modules, it is a sheaf homomorphism of Q-sheaves. 49 Our main result follows immediately: Theorem 4.62 For any ´etale groupoid G, the category Set(O(G)) (equiv- alently, Map(O(G)-HMB)) is a Grothendieck topos and it is equivalent to the classifying topos BG (hence it is an ´etendue). Corollary 4.63 Any ´etendue is equivalent to Set(Q) (equivalently, Map(Q- HMB)) for some inverse quantal frame Q. 5 Concluding remarks Our results add credibility to the notion of sheaf for involutive quantales that is prevalent in the literature, due to the following two reasons: • for an ´etale groupoid G the O(G)-sets give us the classifying topos of G; • for involutive quantales with base locales another natural notion of "equivariant sheaf" is obtained by generalizing the definition of Q- sheaf of 3.26 and, as we have seen, for an inverse quantal frame Q the resulting category Q-Sh is equivalent to Set(Q). Another natural criterion for assessing the value of any notion of quantale sheaf is that of whether the categories of sheaves are toposes. This topic has been very recently addressed in [28, Proposition 4.4.12] by showing that the category of Q-sheaves is a Grothendieck topos if and only if Q is a so- called Grothendieck quantale, by which is meant a modular quantal frame that satisfies an additional simple algebraic condition. (In particular, inverse quantal frames are Grothendieck quantales.) This criterion leaves out stable quantal frames which, as we have seen (cf. 2.28), are more general than modular quantal frames but nevertheless form an algebraically nice class of quantales. While it is not inconceivable that a slightly more restricted notion of sheaf might be appropriate for stable quantal frames, especially if this yields toposes of sheaves, presently not much motivation seems to exist in order to pursue this question due to the unavailability of natural examples of stable quantal frames beyond those that are inverse quantal frames (despite which there may be reasons for addressing even more general stably supported quantales, as hinted at in the beginning of section 4.3). Nevertheless we remark that any stable quantal frame Q has a base locale B, and that in those particular cases where Q is an inverse quantal frame every complete Hilbert Q-module X is also a B-sheaf, as has been proved in Theorem 4.47. This is a rather natural property, but it 50 depends crucially on the fact that the Hilbert sections of X cover X, which is false in general for Hilbert modules on arbitrary stable quantal frames, as example 4.12 shows. Hence, if Q is a stable quantal frame, we regard this cover condition as an example of an axiom that may make sense adding to the definition of sheaf for Q. 6 Appendix We provide a brief survey of the variants of the notion of quantale-valued set that exist in the literature in order to make clear that they are all equivalent in the cases that interest us in this paper, namely inverse quantal frames, and even more generally in the case of stably Gelfand quantales, which encompass the known examples of quantales-as-spaces. In this way we intend to convey an idea of robustness of the definitions that surround the notion of sheaf for involutive quantales and that such sheaves can indeed be taken to be quantale-valued sets, or, equivalently, complete Hilbert modules. We shall also describe completeness of quantale-valued sets and compare it with the natural notion of completeness that arises from complete Hilbert modules, concluding that the two notions coincide. Stably Gelfand quantales. Mulvey [48, 51] has noticed that the notion of Gelfand quantale (referred to as "localic quantale" in [48]) plays a relevant role in the study of quantales of C-algebras. A Gelfand quantale is an involutive quantale whose right sided elements (that is, those elements a such that a1 ≤ a) satisfy the regularity (or "Gelfandness") condition aa∗a = a. Later, he and Ramos [54, 63] have put forward the stronger notion of locally Gelfand quantale, which is important in connection with their study of quantal sets and sheaves on involutive quantales; in a locally Gelfand quantale, all the elements a such that a ≤ p and ap ≤ a for some projection p are required to be regular. An even stronger notion is the following: Definition 6.1 By a stably Gelfand quantale is meant an involutive quantale Q such that, for all a ∈ Q, if aa∗a ≤ a then aa∗a = a (in other words, if a is "stable" under the operation a 7→ aa∗a then it is a "Gelfand element", hence the terminology). The main examples of "quantales-as-spaces" are stably Gelfand: Example 6.2 1. Any involutive quantale Q that satisfies the condition a ≤ aa∗a for all a ∈ Q is stably Gelfand. This includes supported quantales and, hence, modular quantales. It also includes the involutive quantales associated to localic or topological open groupoids. 51 2. The involutive quantale Max A of a C-algebra A is stably Gelfand be- cause it satisfies the condition V V ∗V V ∗ ≤ V V ∗ ⇒ V V ∗V = V for all V ∈ Max A. To some extent this example enables us to generalize to ar- bitrary sub-C-algebras the constructions, due to Kumjian and Renault [41, 64], of ´etale groupoids from suitable commutative sub-C-algebras of C-algebras ("diagonals"), via a construction that associates a lo- calic ´etale groupoid to each projection of a stably Gelfand quantale. See [67] for this construction and a proof that Max A is stably Gelfand. The first appearance of stably Gelfand quantales in writing seems to be in [20], where these quantales (in fact quantaloids) are called pseudo-rightsided and play an important role in unifying variants of the notion of quantale- valued set, as we shall see. Quantale-valued sets. Let Q be an involutive quantale. As we have men- tioned in section 4.2, there is a logical interpretation of Q-sets (cf. com- ments after 4.20). This point of view is emphasized by Mulvey and Ramos [54, 63], who adopt the style of notation introduced earlier for frame-valued sets [6, Section 2; 17; 31; 35, pp. 502 -- 513] (subsequently applied to right- sided quantales) and define a quantal set over Q to be a set I equipped with mappings E : I → Q [[· = ·]] : I × I → Q , referred to respectively as extent and equality, satisfying the following con- ditions for all α, β, γ ∈ I: 1. E(α) = [[α = α]] 2. [[α = β]]∗ = [[β = α]] 3. [[α = β]][[β = γ]] ≤ [[α = γ]] 4. [[α = β]] ≤ E(α)[[α = β]] 5. [[α = β]] ≤ [[α = β]]E(β). We remark that E is used for convenience only, since it is derived from equal- ity. The third condition (transitivity of equality) ensures that the matrix A defined by aαβ = [[α = β]] satisfies AA ≤ A, and the converse inequality holds due to the first and fourth (or fifth) conditions. Hence, since the mapping E is redundant, a quantal set is the same as a strict Q-set in sense of Garraway [20] and Gylys [26]: 52 Definition 6.3 Let Q be an involutive quantale. A Q-set is strict if for all α, β ∈ I we have aααaαβ = aαβ (and thus also aβαaαα = aβα). Mulvey and Ramos further define a Gelfand quantal set to be a quantal set satisfying the following condition for all α and β: [[α = β]] ≤ [[α = β]][[α = β]]∗[[α = β]] . Theorem 6.4 If Q is stably Gelfand all the above variants of Q-valued set, namely Q-sets, quantal sets (= strict Q-sets), and Gelfand quantal sets, coin- cide. In particular, if Q is a frame we obtain the usual notion of frame-valued set. Proof. Every Q-set (I, A) satisfies AA∗A ≤ A and thus aαβa∗ all α, β ∈ I. Hence, if Q is stably Gelfand we have aαβa∗ α, β ∈ I and thus the Q-set is strict [20, Lemma 4.1]: αβaαβ ≤ aαβ for αβaαβ = aαβ for all aαβ = aαβa∗ αβaαβ = aαβaβαaαβ ≤ aααaαβ ≤ aαβ . If Ω is a frame, in matrix language its frame-valued sets are the Ω-valued matrices A that satisfy AA ≤ A = AT , and thus they coincide with Ω-sets according to 4.20 because frames, seen as quantales with trivial involution, are stably Gelfand. Maps. Similarly to quantale-valued sets, there are strict notions of map: Definition 6.5 Let Q be an involutive quantale, and let F : (I, A) → (J, B) be a map of Q-sets. The map F is said to be strict if it satisfies the conditions (6.6) (6.7) fβα = fβαaαα fβα = bββfβα for all α ∈ I and β ∈ J. Lemma 6.8 Let F : (I, A) → (J, B) be a map of Q-sets. If (I, A) is strict then F satisfies (6.6). If (J, B) is strict then F satisfies (6.7). Proof. Assume that (I, A) is strict. Then for all α ∈ I and β ∈ J we have fβα = (F A)βα =_α′ fβα′aα′α =_α′ fβα′aα′αaαα = (F A)βαaαα = fβαaαα . The strictness condition (6.7) follows from the strictness of (J, B) is a similar way using the equality F = BF . 53 Mulvey and Ramos [54, 63] have proposed a notion of map which is for- mally equivalent to the above notion of strict map, and thus also equivalent to a general map because their quantal sets are strict Q-sets. They have also defined a map F to be Gelfand if the condition fβα ≤ fβαf ∗ βαfβα holds for all α and β, and they have proved that, if Q is locally Gelfand, the composition of two Gelfand maps is a Gelfand map. In this way a category of Gelfand quantal sets and Gelfand maps is defined. But, again, if Q is stably Gelfand (rather than just locally Gelfand) there are additional simplifications, since any map is necessarily strict and Gelfand: Theorem 6.9 If Q is stably Gelfand all the above variants of map of Q-set coincide. In particular, if Q is a frame we obtain the usual notion of map of frame-valued sets. Proof. Any map F : (I, A) → (J, B) satisfies F F ∗F ≤ BF = F , and thus fβαf ∗ βαfβα ≤ fβα for all α and β. Hence, if Q is stably Gelfand we obtain fβαf ∗ βαfβα = fβα, showing that F is a Gelfand map. And F is strict due to 6.4 and 6.8. Finally, in matrix language, for a frame Ω a map of Ω-valued sets F : (I, A) → (J, B) is [35, pp. 502 -- 513] an Ω-valued matrix F : J × I → Ω satisfying fβα ≤ aαα ∧ bββ fβα ∧ aαα′ ∧ bββ ′ ≤ fβ ′α′ fβα ∧ fβ ′α ≤ bββ ′ aαα ≤ _β fβα for all α ∈ I and β ∈ J, and it is easy to see that this is just the definition of strict map in frame language: strictness is the first condition; together with the second one it gives us BF A = F ; the third one is F F ∗ ≤ B; and the fourth one is aαα ≤ (F ∗F )αα, which together with the strictness of (I, A) gives us A ≤ F ∗F . Complete quantale-valued sets. In order to describe completeness of quantale-valued sets it will be convenient to depart from our usual convention and identify mappings S : I → Q with column matrices instead of row matrices -- with the adjoint S∗ being regarded as a row matrix. The following definition is adapted from [20, 26]. 54 Definition 6.10 Let Q be an involutive quantale, and let (I, A) be a Q-set. By a singleton map (or simply a singleton) of (I, A) is meant a mapping S : I → Q for which there exists a projection q ∈ Q such that S, regarded as a column matrix I × {∗} → Q, defines a map of Q-sets S : [q] → (I, A) where [q] is the Q-set defined by the {∗} × {∗} matrix with single entry q. In other words, S : I → Q is a singleton if and only if it satisfies, for some projection q, the following conditions for all α, β ∈ I: (6.11) (6.12) (6.13) (6.14) s∗ γsγ sα = sαq q ≤ S∗S =_γ∈I sα = _γ∈I aαγsγ sαs∗ β ≤ aαβ . Again there is a logical interpretation, namely S can be regarded as a "sub- set" of I with sα being the truth value of the assertion that α belongs to S. In particular, (6.13) implies (6.15) aαβsβ ≤ sα , which can be read "if α equals β and β is in S then so is α", and (6.14) can be read "if α is in S and β is in S then α equals β"; the latter expresses the idea that S is a "singleton subset". We remark that if (I, A) is a strict Q-set then, multiplying both sides of (6.13) on the left by aαα, we obtain a strictness condition for singletons: (6.16) aααsα = sα . We also remark that the conjunction of (6.15) and (6.16) is equivalent to (6.13). As before, simplifications arise if Q is stably Gelfand: Theorem 6.17 Let Q be a stably Gelfand quantale, and let S : I → Q be a mapping. Then S is a singleton if and only if it satisfies (6.14) and (6.15). Furthermore, if S is a singleton we have (6.18) for all α ∈ I. sαs∗ αsα = sα 55 Proof. If S is a singleton it satisfies (6.14) by definition, and (6.15) follows from (6.13), as we have already mentioned above. Conversely, assume that S satisfies (6.14) and (6.15), and let q = S∗S. This projection trivially satisfies (6.12). It also satisfies (6.11) because, by 6.9, S is necessarily a Gelfand map: (6.19) (6.20) sαq = sα_γ sαq = _γ sαs∗ γsγ ≤_γ s∗ γsγ ≥ sαs∗ αsα ≥ sα ; aαγsγ ≤ sα . In addition, we conclude that all the above inequalities are in fact equalities, and thus from (6.20) we obtain (6.13). Hence, S is a singleton. Similarly, the conclusion that singletons satisfy (6.18) follows from (6.19). Mulvey and Ramos [54, 63] define singletons in a different way. According to their definition a singleton is, in matrix language, a mapping S : I → Q that satisfies (6.14) and (6.15) together with the condition sα ≤ sαs∗ αsα. Hence, if Q is a stably Gelfand quantale their notion of singleton is equivalent to the one we have been using. In particular, if Q is a locale this coincides with the standard notion of singleton for frame-valued sets [6, Section 2; 17; 31; 35, pp. 502 -- 513]. Let (I, A) be a Q-set, again with Q stably Gelfand. It is immediate that every column of A is a singleton. A Q-set is said to be complete if each of its singletons arises in this way from a unique column of A, and there is a notion of completion of (I, A), which is the complete Q-set that consists of the set of all the singletons of (I, A) equipped with the matrix bA defined by a dot product: aST = S∗T =_α∈I s∗ αtα . If B is a frame, the difference between arbitrary B-sets and the complete ones is that the latter are more canonical in the sense that there is a functor from Sh(B) to Set(B) that to each sheaf assigns a B-set that is complete. However, it is well known that the category Set(B) is equivalent to its full subcategory of complete B-sets (and equivalent to Sh(B)). Analogously, for a stably Gelfand quantale Q the category Set(Q) is equivalent to its full subcategory of complete Q-sets. This fact is proved by Garraway [20, Section 4] and it shows that, to a large extent, for stably Gelfand quantales the completeness of Q-sets is irrelevant. Completeness via Hilbert modules. Now let us compare, for a stably Gelfand quantale Q, the notion of complete Hilbert Q-module with the notion 56 (6.21) (bA)ST = hS∗, T ∗i . of complete Q-set, namely seeing that the former is more canonical than Q- sets exactly for the same reason that complete Q-sets are: the completion of a Q-set (I, A) coincides (up to renaming of matrix indices) with the Q-set associated to the Hilbert Q-module QIA. Theorem 6.21 Let Q be a stably Gelfand quantale, let (I, A) be a Q-set, and let S : I → Q be an arbitrary mapping. The following statements are equivalent: 1. S is a singleton of (I, A); 2. S∗ is a Hilbert section of the Hilbert Q-module QIA. Moreover, letting (bI, bA) be the completion of (I, A) we have, for all S, T ∈bI, the following equality: Proof. First assume that S is a singleton; that is, S satisfies the following two conditions: AS = S , SS∗ ≤ A . Hence, S∗ is in QIA, for S∗A = S∗A∗ = (AS)∗ = S. Moreover, for any other mapping v ∈ QI A (regarded as a row matrix) we have hv, S∗iS∗ = vSS∗ ≤ vA = v , and thus S∗ is a Hilbert section. Now assume that S∗ is a Hilbert section of QIA (regarded as a row ma- trix). The condition AS = S follows from S∗A = S∗, and the Hilbert section condition gives us sSS∗ ≤ s for all s ∈ I: sSS∗ = hs, S∗iS∗ ≤ s . Hence, we obtain ASS∗ ≤ A, i.e., SS∗ ≤ A, and thus S is a singleton. Finally, (6.21) is immediate: (bA)ST = S∗T = S∗T ∗∗ = hS∗, T ∗i. 57 References [1] C. A. Akemann, Left ideal structure of C ∗-algebras, J. Functional Analysis 6 (1970), 305 -- 317. MR0275177 (43 #934) [2] , A Gelfand representation theory for C ∗-algebras, Pacific J. Math. 39 (1971), 1 -- 11. MR0328608 (48 #6950) [3] Th´eorie des topos et cohomologie ´etale des sch´emas. Tome 1: Th´eorie des topos, Lecture Notes in Mathematics, Vol. 269, Springer-Verlag, Berlin, 1972 (French). S´eminaire de G´eom´etrie Alg´ebrique du Bois-Marie 1963 -- 1964 (SGA 4); Dirig´e par M. Artin, A. Grothendieck, et J. L. Verdier. Avec la collaboration de N. Bourbaki, P. Deligne et B. Saint-Donat. MR0354652 (50 #7130) [4] U. Berni-Canani, F. Borceux, R. Succi Cruciani, and G. Van den Bossche, ´Etale maps of quantales, Bull. Soc. Math. Belg. S´er. A 41 (1989), no. 2, 195 -- 218. Actes du Colloque en l'Honneur du Soixanti`eme Anniversaire de Ren´e Lavendhomme (Louvain- la-Neuve, 1989). MR1031749 (90m:06017) [5] U. Berni-Canani, F. Borceux, and R. Succi Cruciani, A theory of quantal sets, J. Pure Appl. Algebra 62 (1989), no. 2, 123 -- 136. MR1027752 (91c:06027) [6] F. Borceux, Handbook of Categorical Algebra. 3, Encyclopedia of Mathematics and its Applications, vol. 52, Cambridge University Press, Cambridge, 1994. Categories of sheaves. MR1315049 (96g:18001c) [7] F. Borceux and R. Cruciani, Sheaves on a quantale, Cahiers Topologie G´eom. Diff´erentielle Cat´eg. 34 (1993), no. 3, 209 -- 228 (English, with French summary). MR1239469 (94h:06017) [8] [9] , A generic representation theorem for non-commutative rings, J. Algebra 167 (1994), no. 2, 291 -- 308. MR1283288 (95f:18013) , Skew Ω-sets coincide with Ω-posets, Cahiers Topologie G´eom. Diff´erentielle Cat´eg. 39 (1998), no. 3, 205 -- 220 (English, with French summary). MR1641850 (2000b:18003) [10] F. Borceux, J. Rosick´y, and G. Van den Bossche, Quantales and C ∗-algebras, J. London Math. Soc. (2) 40 (1989), no. 3, 398 -- 404. MR1053610 (91d:46075) [11] F. Borceux and G. Van den Bossche, Quantales and their sheaves, Order 3 (1986), no. 1, 61 -- 87. MR850399 (87k:18012) [12] , A generic sheaf representation for rings, Category theory (Como, 1990), Lecture Notes in Math., vol. 1488, Springer, Berlin, 1991, pp. 30 -- 42. MR1173003 (93f:18010) [13] M. Bunge, An application of descent Math. Proc. Cambridge Philos. Soc. 107 (1990), no. 10.1017/S0305004100068365. MR1021873 (90k:18002) to a classification theorem for toposes, 59 -- 79, DOI 1, [14] A. Carboni and R. F. C. Walters, Cartesian bicategories. I, J. Pure Appl. Algebra 49 (1987), no. 1-2, 11 -- 32. MR920513 (88k:18009) [15] A. Connes, Noncommutative Geometry, Academic Press Inc., San Diego, CA, 1994. MR1303779 (95j:46063) 58 [16] M. Dadarlat and G. A. Elliott, One-parameter continuous fields of Kirchberg algebras, Comm. Math. Phys. 274 (2007), no. 3, 795 -- 819. MR2328913 [17] M. P. Fourman and D. S. Scott, Sheaves and logic, Applications of sheaves (Proc. Res. Sympos. Appl. Sheaf Theory to Logic, Algebra and Anal., Univ. Durham, Durham, 1977), Lecture Notes in Math., vol. 753, Springer, Berlin, 1979, pp. 302 -- 401. MR555551 (82d:03061) [18] P. J. Freyd and A. Scedrov, Categories, allegories, North-Holland Mathematical Library, vol. 39, North-Holland Publishing Co., Amsterdam, 1990. MR1071176 (93c:18001) [19] W. D. Garraway, Generalized supremum enriched categories and their sheaves, PhD thesis, Dalhousie Univ., 2002. [20] , Sheaves for an involutive quantaloid, Cah. Topol. G´eom. Diff´er. Cat´eg. 46 (2005), no. 4, 243 -- 274 (English, with French summary). MR2248095 (2007d:18020) [21] R. Giles, Foundations for quantum mechanics, J. Mathematical Phys. 11 (1970), 2139 -- 2160. MR0272275 (42 #7156) [22] R. Giles and H. Kummer, A non-commutative generalization of topology, Indiana Univ. Math. J. 21 (1971/72), 91 -- 102. MR0293408 (45 #2485) [23] R. P. Gylys, Quantal sets and sheaves over quantales, Liet. Mat. Rink. 34 (1994), no. 1, 9 -- 31, DOI 10.1007/BF02335387 (English, with Lithuanian summary); English transl., Lithuanian Math. J. 34 (1994), no. 1, 8 -- 29. MR1324216 (96d:06003) [24] [25] [26] , On quantaloids and quantal categories, Liet. Mat. Rink. 35 (1995), no. 3, 266 -- 296 (English, with Lithuanian summary); English transl., Lithuanian Math. J. 35 (1995), no. 3, 210 -- 233. MR1407431 (97h:18013) , Sheaves on quantaloids, Liet. Mat. Rink. 40 (2000), no. 2, 133 -- 171 (English, with English and Lithuanian summaries); English transl., Lithuanian Math. J. 40 (2000), no. 2, 105 -- 134. MR1806339 (2002e:18014) , Sheaves on involutive quantaloids, Liet. Mat. Rink. 41 (2001), no. 1, 44 -- 69 (English, with English and Lithuanian summaries); English transl., Lithuanian Math. J. 41 (2001), no. 1, 35 -- 53. MR1849807 (2002j:18009) [27] H. Heymans, Q-∗-categories, Appl. Categ. Structures 17 (2009), 1 -- 28. [28] , Sheaves on Quantales as Generalized Metric Spaces, PhD thesis, Univ. Antwerp, 2010. [29] H. Heymans and I. Stubbe, On principally generated quantaloid-modules in general, and skew local homeomorphisms in particular, Ann. Pure Appl. Logic 161 (2009), no. 1, 43 -- 65, DOI 10.1016/j.apal.2009.05.001. MR2567926 (2011c:06026) [30] , Modules on involutive quantales: canonical Hilbert structure, applications to sheaf theory, Order 26 (2009), no. 2, 177 -- 196, DOI 10.1007/s11083-009-9116-x. MR2525366 [31] D. Higgs, Injectivity in the topos of complete Heyting algebra valued sets, Canad. J. Math. 36 (1984), no. 3, 550 -- 568. MR752984 (85m:18003) 59 [32] P. T. Johnstone, Topos theory, Academic Press [Harcourt Brace Jovanovich Publish- ers], London, 1977. London Mathematical Society Monographs, Vol. 10. MR0470019 (57 #9791) [33] [34] [35] , Stone Spaces, Cambridge Studies in Advanced Mathematics, vol. 3, Cam- bridge University Press, Cambridge, 1986. Reprint of the 1982 edition. MR861951 (87m:54001) , The point of pointless topology, Bull. Amer. Math. Soc. (N.S.) 8 (1983), no. 1, 41 -- 53. MR682820 (84f:01043) , Sketches of an Elephant: A Topos Theory Compendium. Vol. 2, Oxford Logic Guides, vol. 44, The Clarendon Press Oxford University Press, Oxford, 2002. MR2063092 (2005g:18007) [36] A. Joyal and M. Tierney, An extension of the Galois theory of Grothendieck, Mem. Amer. Math. Soc. 51 (1984), no. 309, vii+71. MR756176 (86d:18002) [37] A. Kock and I. Moerdijk, Presentations of ´etendues, Cahiers Topologie G´eom. Diff´erentielle Cat´eg. 32 (1991), no. 2, 145 -- 164 (English, with French summary). MR1142688 (92m:18007) [38] D. Kruml, Spatial quantales, Appl. Categ. Structures 10 (2002), no. 1, 49 -- 62. MR1883084 (2002m:06010) [39] D. Kruml, J. W. Pelletier, P. Resende, and J. Rosick´y, On quantales and spectra of C ∗-algebras, Appl. Categ. Structures 11 (2003), no. 6, 543 -- 560. MR2017650 (2004i:46107) [40] D. Kruml and P. Resende, On quantales that classify C ∗-algebras, Cah. Topol. G´eom. Diff´er. Cat´eg. 45 (2004), no. 4, 287 -- 296 (English, with French summary). MR2108195 (2006b:46096) [41] A. Kumjian, On C ∗-diagonals, Canad. J. Math. 38 (1986), no. 4, 969 -- 1008. MR854149 (88a:46060) [42] E. C. Lance, Hilbert C ∗-modules, London Mathematical Society Lecture Note Se- ries, vol. 210, Cambridge University Press, Cambridge, 1995. A toolkit for operator algebraists. MR1325694 (96k:46100) [43] S. Mac Lane and I. Moerdijk, Sheaves in geometry and logic, Universitext, Springer- Verlag, New York, 1994. A first introduction to topos theory; Corrected reprint of the 1992 edition. MR1300636 (96c:03119) [44] S. Marcelino and P. Resende, An algebraic generalization of Kripke structures, Math. Proc. Cambridge Philos. Soc. 145 (2008), no. 3, 549 -- 577. MR2464775 [45] I. Moerdijk, The classifying topos of a continuous groupoid. I, Trans. Amer. Math. Soc. 310 (1988), no. 2, 629 -- 668. MR973173 (90a:18005) [46] , The classifying topos of a continuous groupoid. II, Cahiers Topologie G´eom. Diff´erentielle Cat´eg. 31 (1990), no. 2, 137 -- 168 (English, with French summary). MR1080241 (92c:18003) [47] C. J. Mulvey, &, Rend. Circ. Mat. Palermo (2) Suppl. 12 (1986), 99 -- 104. Second topology conference (Taormina, 1984). MR853151 (87j:81017) 60 [48] [49] , Quantales. Invited talk at the Summer Conference on Locales and Topological Groups (Cura¸cao, 1989). , Quantales, The Encyclopaedia of Mathematics, third supplement (M. Hazewinkel, ed.), Kluwer Acad. Publ., 2002, pp. 312 -- 314. [50] C. J. Mulvey and M. Nawaz, Quantales: quantal sets, Non-classical logics and their applications to fuzzy subsets (Linz, 1992), Theory Decis. Lib. Ser. B Math. Statist. Methods, vol. 32, Kluwer Acad. Publ., Dordrecht, 1995, pp. 159 -- 217. MR1345644 (96k:03146) [51] C. J. Mulvey and J. W. Pelletier, A quantisation of the calculus of relations, Category theory 1991 (Montreal, PQ, 1991), CMS Conf. Proc., vol. 13, Amer. Math. Soc., Providence, RI, 1992, pp. 345 -- 360. MR1192157 (94c:18016) [52] [53] , On the quantisation of points, J. Pure Appl. Algebra 159 (2001), no. 2-3, 231 -- 295. MR1828940 (2002g:46126) , On the quantisation of spaces, J. Pure Appl. Algebra 175 (2002), no. 1- 3, 289 -- 325. Special volume celebrating the 70th birthday of Professor Max Kelly. MR1935983 (2003m:06014) [54] C. J. Mulvey and J. Z. Ramos, Quantal Sets over Involutive Quantales, Workshop on Logic from Quantales (Univ. Oxford, Oxford, UK, January 21, 2005), available at http://www.maths.sussex.ac.uk/Staff/CJM/research/CJMResearchTalks.htm. [55] C. J. Mulvey and P. Resende, A noncommutative theory of Penrose tilings, Internat. J. Theoret. Phys. 44 (2005), no. 6, 655 -- 689. MR2150184 (2006a:58011) [56] M. Nawaz, Quantales, quantal sets, PhD thesis, Univ. Sussex, 1985. [57] J. Paseka, Hilbert Q-modules and nuclear ideals in the category of W-semilattices with a duality, CTCS '99: Conference on Category Theory and Computer Science (Edinburgh), Electron. Notes Theor. Comput. Sci., vol. 29, Elsevier, Amsterdam, 1999, pp. Paper No. 29019, 19 pp. (electronic). MR1782793 (2001k:03143) [58] , Hermitian kernels, Hilbert Q-modules and Ando dilation, Contributions to general algebra, 12 (Vienna, 1999), Heyn, Klagenfurt, 2000, pp. 317 -- 335. MR1777672 (2001h:46110) [59] J. W. Pelletier and J. Rosick´y, Simple involutive quantales, J. Algebra 195 (1997), no. 2, 367 -- 386. MR1469630 (98m:06007) [60] A. M. Pitts, Applications of sup-lattice enriched category theory to sheaf theory, Proc. London Math. Soc. (3) 57 (1988), no. 3, 433 -- 480. MR960096 (89m:18002) [61] M. C. Protin, Quantales of open groupoids, PhD thesis, Univ. T´ecnica de Lisboa, 2008. [62] M. C. Protin and P. Resende, Quantales of open groupoids, 2009, arXiv:0811.4539v2. [63] J. Z. Ramos, Involutive quantales, PhD thesis, Univ. Sussex, 2006. [64] J. Renault, Cartan subalgebras in C ∗-algebras, Irish Math. Soc. Bull. 61 (2008), 29 -- 63. MR2460017 (2009k:46135) [65] P. Resende, ´Etale groupoids and their quantales, Adv. Math. 208 (2007), no. 1, 147 -- 209. MR2304314 (2008c:22002) 61 [66] [67] , Quantal sets, quantale modules, and groupoid actions. Talk at the Internat. Conf. Category Theory 2007 (Carvoeiro, Portugal, June 17 -- 23, 2007). Gelfand and Cartan quantales, Stably 5th Workshop on Categories, Logic, (Imperial College, London, UK, August sub-C*- , of algebras, Physics at http://www.math.ist.utl.pt/~pmr/slides-IC-aug-2009.pdf. Paper in prepara- tion. groupoids and the Foundations available 2009), 6, [68] P. Resende and E. Rodrigues, Sheaves as modules, Appl. Categ. Structures 18 (2010), no. 2, 199 -- 217, DOI 10.1007/s10485-008-9131-x (on-line 2008). MR2601963 [69] K. I. Rosenthal, Quantales and Their Applications, Pitman Research Notes in Math- ematics Series, vol. 234, Longman Scientific & Technical, Harlow, 1990. MR1088258 (92e:06028) [70] [71] , A general approach to Gabriel filters on quantales, Comm. Algebra 20 (1992), no. 11, 3393 -- 3409. MR1186715 (93k:06012) , The Theory of Quantaloids, Pitman Research Notes in Mathematics Series, vol. 348, Longman, Harlow, 1996. MR1427263 (97j:18009) [72] J. Rosick´y, Multiplicative lattices and C ∗-algebras, Cahiers Topologie G´eom. Diff´erentielle Cat´eg. 30 (1989), no. 2, 95 -- 110 (English, with French summary). MR1004734 (91e:46079) [73] I. Stubbe, Categorical structures enriched in a quantaloid: categories and semicate- gories, PhD thesis, Univ. Catholique de Louvain, 2003. [74] , Categorical structures enriched in a quantaloid: orders and ideals over a base quantaloid, Appl. Categ. Structures 13 (2005), no. 3, 235 -- 255. MR2167792 (2006e:18006) [75] G. Van den Bossche, Quantaloids and non-commutative ring representations, Appl. Categ. Structures 3 (1995), no. 4, 305 -- 320. MR1364011 (96h:18009) [76] R. F. C. Walters, Sheaves and Cauchy-complete categories, Cahiers Topologie G´eom. Diff´erentielle 22 (1981), no. 3, 283 -- 286. Third Colloquium on Categories, Part IV (Amiens, 1980). MR649076 (83f:18011) [77] , Sheaves on sites as Cauchy-complete categories, J. Pure Appl. Algebra 24 (1982), no. 1, 95 -- 102. MR647583 (83e:18006) Centro de An´alise Matem´atica, Geometria e Sistemas Dinamicos Departamento de Matem´atica, Instituto Superior T´ecnico Universidade T´ecnica de Lisboa Av. Rovisco Pais 1, 1049-001 Lisboa, Portugal E-mail: [email protected] 62
1212.6368	1	1212	2012-12-27T13:35:14	Lie bialgebra structures on the deformative Schr\"{o}dinger-Virasoro algebras	[ "math.RA" ]	In this paper we investigate Lie bialgebra structures on the deformative Schr\"{o}dinger-Virasoro algebras mainly using the techniques introduced recently by Liu, Pei and Zhu, which indicate that all cases considered in this paper except one behave different from their centerless ones.	math.RA	math	Lie bialgebra structures on the deformative Schrodinger-Virasoro algebras Huanxia Fa1,2) 1)Wu Wen-Tsun Key Laboratory of Mathematics and School of Mathematical Sciences, University of Science and Technology of China, Hefei 230026, China 2)School of Mathematics and Statistics, Changshu Institute of Technology, Changshu 215500, China E-mail: sd [email protected] Abstract. In this paper we investigate Lie bialgebra structures on the deformative Schrodinger- Virasoro algebras mainly using the techniques introduced recently by Liu, Pei and Zhu, which indicate that all cases considered in this paper except one behave different from their centerless ones. Key words: deformative Schrodinger-Virasoro algebras, Lie bialgebras, Yang-Baxter equations. Mathematics Subject Classification (2010): 17B05, 17B37, 17B62, 17B65, 17B68 2 1 0 2 c e D 7 2 ] . A R h t a m [ 1 v 8 6 3 6 . 2 1 2 1 : v i X r a §1. Introduction Both original and twisted Schrodinger-Virasoro Lie algebras were introduced by [7] in the context of non-equilibrium statistical physics and further investigated in [8, 9], whose deformations were introduced in [22]. All of them are closely related to the Schrodinger Lie algebra and the Virasoro Lie algebra. The deformative Schrodinger-Virasoro Lie algebras Ls λ (λ ∈ C) considered in this paper, possess the C-basis {Ln, Mn, Ys+n, c n ∈ Z, s = 0 or 1 2 } with the following non- vanishing Lie brackets: [ Ln, Lm] = (m − n)Lm+n + m3−m [Ln, Ys+m] = (s + m − λ+1 2 n)Ys+m+n, 12 δm+n,0c, [Ln, Mm ] = (m − λn)Mm+n, [ Ys+n, Ys+m ] = (m − n)M2s+m+n. (1.1) It is easy to see that Ls w = spanC{Ln n ∈ Z} as their subalgebras and hs = {Mn, Ys+n, c n ∈ Z} as their ideals. λ contain the Virasoro algebra v = spanC{Ln, c n ∈ Z} and Witt algebra Due to their mathematical and physical backgrounds, interests and importance, a series of papers on Lie algebras of these types appeared. The Lie bialgebra structures, modules of inter- mediate series, non-trivial vertex algebra representations and Whittaker modules over the original Schrodinger-Virasoro Lie algebra were respectively investigated in [6, 15, 26, 35]. The derivations, central extensions and automorphism group of the extended Schrodinger-Virasoro Lie algebra were determined in [5]. The Verma modules and derivations over generalized Schrodinger-Virasoro alge- bras were respectively investigated in [25] and [27]. The second cohomology group of both original and twisted deformative Schrodinger-Virasoro algebras were completely determined by [11, 14, 22]. Automorphisms and derivations of twisted and original deformative Schrodinger-Virasoro Lie al- gebras were respectively determined in [28] and [10]. The notion of Lie bialgebras was introduced in 1983 by [1] to find solutions of the Yang-Baxter Initially, the equation. Many types of Lie bialgebras were considered over different algebras. Witt and Virasoro Lie bialgebras were investigated in [24] and classified in [21]. Lie bialgebras of generalized Witt types, generalized Virasoro-like type, generalized Weyl type, Hamiltonian type and Block type were considered respectively in [23, 29, 34, 32, 17]. Lie superbialgebra structures on Supported by the NSF grant 11101056 of China 1 the Ramond N = 2 super Virasoro algebra were considered in [33]. The Lie bialgebra structures on 1 2 0 (with c = 0) and a twisted case L0 the original case L by [6] and [3]. In this paper we shall investigate Lie bialgebra structures on Ls techniques introduced recently in [13]. −3 (with c = 0) were respectively investigated λ mainly using the §2. Preliminaries and main results For convenience, some basic concepts about Lie bialgebras are presented firstly. Respectively denote ξ and τ the cyclic map of L ⊗ L ⊗ L and the twist map of L ⊗ L for any given vector space L, which imply ξ(x1 ⊗ x2 ⊗ x3) = x2 ⊗ x3 ⊗ x1 and τ (x1 ⊗ x2) = x2 ⊗ x1, ∀ x1, x2, x3 ∈ L. For a vector space L and two bilinear maps δ : L ⊗ L → L and ∆ : L → L ⊗ L, the pair (L, δ) becomes a Lie algebra if the following two conditions hold: Ker(1 − τ ) ⊂ Ker δ, δ · (1 ⊗ δ) · (1 + ξ + ξ2) = 0, and the pair (L, ∆) becomes a Lie coalgebra if the following two conditions hold: Im ∆ ⊂ Im(1 − τ ), (1 + ξ + ξ2) · (1 ⊗ ∆) · ∆ = 0. The triple (L, δ, ∆) will become a Lie bialgebra if the Lie algebra (L, δ) and Lie coalgebra (L, ∆) satisfy the following compatible condition: ∆δ(x ⊗ y) = x · ∆y − y · ∆x, ∀ x, y ∈ L, (2.1) where the symbol "·" is defined to be the diagonal adjoint action: x · (Pi ai ⊗ bi) = Pi ([x, ai] ⊗ bi + ai ⊗ [x, bi]). Denote U the universal enveloping algebra of L and 1 the identity element of U. For any r = Pi ai ⊗ bi ∈ L ⊗ L, we denote r12 = Pi ai ⊗ bi ⊗ 1, r13 = Pi ai ⊗ 1 ⊗ bi, r23 = Pi 1 ⊗ ai ⊗ bi. Define c(r) to be elements of U ⊗ U ⊗ U by c(r) = [r12, r13] + [r12, r23] + [r13, r23] = Pi,j [ai, aj] ⊗ bi ⊗ bj + Pi,j ai ⊗ [bi, aj] ⊗ bj + Pi,j ai ⊗ aj ⊗ [bi, bj]. Definition 2.1 (1) A coboundary Lie bialgebra is a 4-tuple (L, δ, ∆, r), where (L, δ, ∆) is a Lie bialgebra and r ∈ Im(1 − τ ) ⊂ L ⊗ L such that ∆ = ∆r is a coboundary of r, i.e., ∆r(x) = x · r for x ∈ L. (2.2) (2) A coboundary Lie bialgebra (L, δ, ∆, r) is called triangular if it satisfies the following classical Yang-Baxter Equation (3) r ∈ Im(1 − τ ) ⊂ L ⊗ L is said to satisfy the modified Yang-Baxter equation if c(r) = 0. x · c(r) = 0, ∀ x ∈ L. 2 (2.3) (2.4) Regard V s λ = Ls λ as an Ls λ) the set of derivations Ds λ ⊗ Ls λ-module under the adjoint diagonal action. Denote by λ : Ls λ is a linear map satisfying λ, namely, Ds λ → V s Der(Ls λ, V s Ds λ([x, y]) = x · Ds λ(y) − y · Ds λ(x), (2.5) λ) the set consisting of the derivations vinn, v ∈ V s λ, V s and Inn(Ls defined by vinn : x 7→ x · v. Then H 1(Ls λ, V s first cohomology group of the Lie algebra Ls λ) ∼= Der(Ls λ, V s λ with coefficients in the Ls λ)/Inn(Ls λ, V s λ), where H 1(Ls λ-module V s λ. λ, where vinn is the inner derivation λ) is the λ, V s Lemmas 2.2 and 2.3 can be found in [10], [16] and [28]. Lemma 2.2 H 1(L 1 2 λ , L 1 2 λ ) ∼= 2 ⊕ CD0 3 CD1 ⊕ CD0 CD1 ⊕ CD−1 2 CD1 ⊕ CD−2 2 CD1 if λ = 0, if λ = −1, if λ = −2, otherwise,   where the corresponding non-vanishing components are given as follows: D1(Yn+ 1 2 ) = Yn+ 1 2 , D1(Mn) = 2Mn, D0 2(Ln) = Mn, D0 3(Ln) = nMn, D−1 2 (Ln) = (n2 − n)Mn, D−2 2 (Ln) = n3Mn, for all n ∈ Z. Lemma 2.3 H 1(L0 λ, L0 λ) ∼= 2 ⊕ Cd0 3 2 ⊕ Cd−1 3 Cd1 ⊕ Cd0 Cd1 ⊕ Cd−1 Cd1 ⊕ Cd−2 2 Cd1 ⊕ Cd1 2 Cd1 if λ = 0, if λ = −1, if λ = −2, if λ = 1, otherwise,   where the corresponding non-vanishing components are given as follows: d1(Yn) = Yn, d1(Mn) = 2Mn, d0 2(Ln) = Mn, d0 3(Ln) = nMn, d−1 2 (Ln) = n2Mn, d−1 3 (Yn) = nMn, d−2 2 (Ln) = n3Mn, d1 2(Yn) = Mn, for all n ∈ Z. 1 The Lie bialgebra structures on the Schrodinger-Virasoro Lie algebras L0 0 have already been considered respectively in [3] and [6] (with c = 0). We just need to consider the following cases: −3 and L 2 1 1 1 L 2 −2, L 2 −1, L L0 −2, L0 −1, L0 2 λ (cid:0) ∀ λ /∈ S 1 2(cid:1), 0, L0 1, L0 λ (cid:0) ∀ λ /∈ S0(cid:1), (2.6) (2.7) where S 1 2 = {0, −1, −2} and S0 = {0, ±1, −2, −3}. Lemmas 2.2 and 2.3 can be partially rewritten as follows. 3 Lemma 2.4 H 1(L 1 2 λ , L 1 2 λ ) ∼= D1 D2 D3   if λ = −1, if λ = −2, if λ /∈ S 1 2 , where Di consist of the following derivations σi respectively: σ1(Ln) = α1(n2 − n)Mn, σ1(Yn+ 1 2 ) = β1Yn+ 1 2 , σ1(Mn) = 2β1Mn, σ2(Ln) = α2n3Mn, σ2(Yn+ 1 2 ) = β2Yn+ 1 2 , σ2(Mn) = 2β2Mn, σ3(Yn+ 1 2 ) = β3Yn+ 1 2 , σ3(Mn) = 2β3Mn, for all n ∈ Z and any α1, α2, βi ∈ C with i = 1, 2, 3. Lemma 2.5 H 1(L0 λ,µ, L0 λ,µ) ∼= d1 d2 d3 d4 d5   if λ = 0, if λ = −1, if λ = −2, if λ = 1, if λ /∈ S0, where di consist of the following derivations i respectively: 1(Ln) = (µ1n + ν1)Mn, 1(Yn) = γ1Yn, 1(Mn) = 2γ1Mn, 2(Ln) = µ2n2Mn, 2(Yn) = γ2Yn + ζ2 nMn, 2(Mn) = 2γ2Mn, 3(Ln) = µ3n3Mn, 3(Yn) = γ3Yn, 3(Mn) = 2γ3Mn, 4(Yn) = γ4Yn + ζ4Mn, 4(Mn) = 2γ4Mn, 5(Yn) = γ5Yn, 5(Mn) = 2γ5Mn, for all n ∈ Z and any µ1, µ2, µ3, ν1, ζ2, ζ4, γi ∈ C with i = 1, 2, 3, 4, 5. A derivation Ds all p ∈ Zs. Denote Der(Ls λ ∈ Der(Ls λ, V s λ, V s λ)α = {Ds λ, V s Let Ds λ be an element of Der(Ls as follows: For any µ ∈ (Ls (Ds λ)α(µ) = µq+α. Obviously, (Ds λ) is homogeneous of degree α ∈ Zs if Ds λ(cid:0)(Ls λ = α} for α ∈ Zs. λ, V s λ) deg Ds λ ∈ Der(Ls λ). For any α ∈ Zs, define the linear map (Ds λ(µ) = Pp∈Zs µp with µp ∈ (V s λ)q with q ∈ Zs, write Ds λ)α ∈ Der(Ls λ, V s λ)α and we have λ)p(cid:1) ⊂ (V s λ)α+p for λ)α : Ls λ → V s λ λ)p, then we set Ds λ = Pα∈Zs (Ds λ)α, (2.8) which holds in the sense that for every u ∈ Ls Pα∈Zs (Ds λ)α(u) (we call such a sum in (2.8) summable). λ, only finitely many (Ds λ)α(u) 6= 0, and Ds λ(u) = 4 Lemma 2.6 One can find the following elements of Der(L σ♮ 2 for the case λ = −2, σ♮ respectively: 3 for the case λ /∈ S 1 2 1 1 2 2 λ , L 1 for the case λ = −1, , which are defined by the following relations λ ⊗ L 1 2 λ ): σ♮ σ♮ 1(Ln) = (n2 − n)(α1z1 ⊗ Mn + α† σ♮ 1(Yn+ 1 2 ) = β1w1 ⊗ Yn+ 1 2 + β† 1Yn+ 1 2 1Mn ⊗ z† 1), σ♮ 1(Mn) = 2(β1w1 ⊗ Mn + β† ⊗ w† 1, 1Mn ⊗ w† 1), σ♮ 2(Ln) = n3(α2z2 ⊗ Mn + α† 2Mn ⊗ z† 2), σ♮ 2(Yn+ 1 2 ) = β2w2 ⊗ Yn+ 1 2 σ♮ 2(Mn) = 2(β2w2 ⊗ Mn + β† 2Mn ⊗ w† 2), σ♮ 3(Yn+ 1 2 ) = β3w3 ⊗ Yn+ 1 2 + β† 2Yn+ 1 2 + β† 3Yn+ 1 2 ⊗ w† 2, ⊗ w† 3, σ♮ 3(Mn) = 2(β3w3 ⊗ Mn + β† 3Mn ⊗ w† 3), σ♮ 3(Ln) = σ♮ 1(c) = σ♮ 2(c) = σ♮ 3(c) = 0, for all n ∈ Z and some α1, α2, α† 1, α† 2, βi ∈ C and wi, zj, w† i , z† j ∈ C with i = 1, 2, 3, j = 1, 2. Lemma 2.7 One can find the following elements of Der(L0 for the case λ = −1, ♮ by the following relations respectively: 3 for the case λ = −2, ♮ λ ⊗ L0 4 for the case λ = 1, ♮ λ, L0 λ): ♮ 1 for the case λ = 0, ♮ 2 5 for the case λ /∈ S0, defined ♮ 1(Ln) = (µ1n + ν1)z1 ⊗ Mn + (µ† 1n + ν † ♮ 1(Yn) = γ1w1 ⊗ Yn + γ† 1)Mn ⊗ z† 1, ♮ 2(Ln) = n2(µ2z2 ⊗ Mn + µ† 2Mn ⊗ z† 2), 1Yn ⊗ w† 1, ♮ 1(Mn) = 2(γ1w1 ⊗ Mn + γ† ♮ 2(Yn) = γ2w2 ⊗ Yn + γ† 1Mn ⊗ w† 1), 2Yn ⊗ w† 2 + n(ζ2v2 ⊗ Mn + ζ † 2Mn ⊗ v† 2), ♮ 2(Mn) = 2(γ2w2 ⊗ Mn + γ† ♮ 3(Yn) = γ3w3 ⊗ Yn + γ† ♮ 3(Ln) = n3(µ3z3 ⊗ Mn + µ† 2Mn ⊗ w† 2), ♮ 3(Mn) = 2(γ3w3 ⊗ Mn + γ† 3Yn ⊗ w† 3, 3Mn ⊗ w† 3), 3Mn ⊗ z† 3), ♮ 4(Yn) = γ4w4 ⊗ Yn + γ† 4Yn ⊗ w† 4 + ζ4v4 ⊗ Mn + ζ † 4Mn ⊗ v† 4, ♮ 4(Mn) = 2(γ4w4 ⊗ Mn + γ† 4Mn ⊗ w† 4), ♮ 5(Mn) = 2(γ5w5 ⊗ Mn + γ† 5Mn ⊗ w† 5), ♮ 5(Yn) = γ5w5 ⊗ Yn + γ† ♮ 4(Ln) = ♮ 5(Ln) = ♮ k(c) = 0, 5Yn ⊗ w† 5, for all n ∈ Z, some µi, µ† 1 ∈ C0, v2, v† z1, z† 1, w1, w† i , ν1, ν † 2, v4, v† 1, ζ2, ζ4, ζ † 4, zk, z† 2, ζ † k, wk, w† 4, γj, γ† k ∈ C with k = 2, 3, 4, 5. j ∈ C with i = 1, 2, 3, j = 1, 2, 3, 4, 5 and Respectively, denote the vector spaces spanned by σ♮ i and ♮ j as D♮ i and d♮ j for i = 1, 2, 3 and j = 1, 2, 3, 4, 5. The main results of this paper can be formulated as follows. Theorem 2.8 (i) Every Lie bialgebra (Ls Ls λ given in (2.6) and (2.7) with c = 0 and (λ, s) 6= (0, 0). λ, [·, ·], ∆) is triangular coboundary for the Lie algebras (ii) Every Lie bialgebra (L0 (iii) No Lie bialgebra (Ls 0, [·, ·], ∆) is not triangular coboundary with c = 0 and ♮ 1 6= 0. λ, [·, ·], ∆) with c 6= 0 is triangular coboundary if the derivations given in Lemmas 2.6 and 2.7 are not equal to zero. 5 §3 Proof of the main result Throughout the paper we denote by Z∗ the set of all nonnegative integers and C∗ the set of all nonnegative complex numbers. The following lemma can be found in Lemma 2.2 of [13]. Lemma 3.1 Suppose that g = ⊕n∈Zgn is a Z-graded Lie algebra with a finite-dimensional center Cg, and g0 is generated by {gn, n 6= 0}. Then H 1(g, Cg ⊗ g + g ⊗ Cg)0 = Cg ⊗ H 1(g, g)0 + H 1(g, g)0 ⊗ Cg. Denote Zs = Z ∪ {s + Z} and Z∗ s = Zs\{0}, i.e., Z0 = Z, Z∗ the above lemma can be generalized immediately as follows. Lemma 3.2 Suppose that g = ⊕n∈Zsgn is a Zs-graded Lie algebra with a finite-dimensional center Cg, and g0 is generated by {gn, n 6= 0}. Then 0 = Z∗, Z 1 Z∗. Then Z and Z∗ = 1 2 = 1 2 2 1 2 H 1(g, Cg ⊗ g + g ⊗ Cg)0 = Cg ⊗ H 1(g, g)0 + H 1(g, g)0 ⊗ Cg. The following lemma is one of the main results given in [21]. Lemma 3.3 Every Lie bialgebra on the Witt algebra w and Virasoro algebra v is triangular coboundary and H 1(w, w ⊗ w) = H 1(v, v ⊗ v) = 0. Denote Ls λ ⊗n the tensor product of n copies of Ls λ-module under the adjoint diagonal action of Ls λ. The first item of the following lemma can be obtained by using the similar arguments as those given in the known references and the left two can be found in the [1, 2, 21, 29]). For convenience, we introduce the following notations: C = Cc, references (e.g. C0 = spanC{c, M0} and Cs λ = C for all the other cases referred in (2.6) and (2.7). λ = C0 for the case (s, λ) = (0, 0) and Cs λ and regard it as an Ls λ, where Cs Lemma 3.4 (i) If x · r = 0 for some r ∈ Ls λ ⊗n and all x ∈ Ls λ, then r ∈ Cs λ ⊗n. (ii) The r satisfies (2.3) if and only if it satisfies (2.4). (iii) Let L be a Lie algebra and r ∈ Im(1 − τ ) ⊂ L ⊗ L, then (1 + ξ + ξ2) · (1 ⊗ ∆r) · ∆r(x) = x · c(r), ∀ x ∈ L, and the triple (L, [·, ·], ∆r) is a Lie bialgebra if and only if r satisfies (2.3). Proposition 3.5 D♮ 1 D♮ 2 D♮ 3 d♮ 1 d♮ 2 d♮ 3 d♮ 4 d♮ 5   if s = 1 if s = 1 if s = 1 2 , λ = −1, 2 , λ = −2, 2 , λ /∈ S 1 , if s = 0, λ = 0, 2 if s = 0, λ = −1, if s = 0, λ = −2, if s = 0, λ = 1, if s = 0, λ /∈ S0. H 1(Ls λ, V s λ) = Der(Ls λ, V s λ)/Inn(Ls λ, V s λ) ∼= 6 Proof of Proposition 3.5 This proposition follows from a series of claims. Denote Hs λ = Ls λ ⊗ hs + hs ⊗ Ls λ. Then Hs λ-module V s λ/Hs λ is an Ls λ-submodule of V s λ, since hs is an ideal of Ls λ λ, on which hs acts trivially. The exact sequence λ as Qs and denote the quotient Ls 0 → Hs λ → V s λ → V s λ/Hs λ → 0 induces the following long exact sequence −→ H 0(Ls λ, Qs λ) −→ H 1(Ls λ, Hs λ) −→ H 1(Ls λ, V s λ) −→ H 1(Ls λ, Qs λ) −→ of Zs-graded vector spaces, where all coefficients of the tensor products are in C. It is easy to see that H 0(Ls λ) if we can prove H 1(Ls λ · x = 0 } = 0. Then H 1(Ls Ls λ = {x ∈ Qs λ) ∼= H 1(Ls λ, Hs λ, V s λ Ls Denote LC s λ → Hs 0 → LC λ + Cs λ ⊗ Cs s λ → 0 induces the following long exact sequence λ/LC λ-submodule of Hs λ. Then LC s λ is an Ls λ ⊗ Ls λ. The exact sequence λ) = Qs λ, Qs λ λ, Qs λ) = 0. s λ = Ls λ → Hs −→ H 0(Ls λ, Hs λ/LC s λ) −→ H 1(Ls λ, LC s λ) −→ H 1(Ls λ, Hs λ) −→ H 1(Ls λ, Hs λ/LC s λ) −→ . It is easy to see that H 0(Ls H 1(Ls λ) ∼= H 1(Ls s λ) = (Hs λ/LC λ) if we can prove H 1(Ls λ, Hs λ, Hs λ/LC λ, LC s s λ)Ls λ, Hs λ = {x ∈ Hs λ/LC s λ) = 0. λ/LC s λ Ls λ · x = 0 } = 0. Then Claim 1 If α ∈ Z∗ s, then (Ds λ)α ∈ Inn(Ls λ, V s λ). For α 6= 0, denote γ = α−1(Ds λ)α(L0) ∈ (V s λ)s+n, λ)α to [L0, Ln] = nLn, [L0, Mn] = nMn and [L0, Ys+n] = (s + n)Ys+n, we obtain λ)α(Ln), (Ds λ)α(Ln) = λ)α(Ys+n) = (α + s + λ)n+α and (Ds λ)α(Mn) = (α + n)(Ds λ)α(Ys+n) ∈ (V s λ)α(Mn) and L0 · (Ds λ)s+n+α, i.e., L0 · (Ds λ)α. For any Ln, Mn ∈ (Ls λ)n and Ys+n ∈ (Ls λ)α(Mn) ∈ (V s λ)α(Ln), L0 · (Ds applying (Ds (cid:0)recalling (Ds (α + n)(Ds n)(Ds λ)α(Ys+n)(cid:1) λ)α(Ln) − Ln · (Ds λ)α(Mn) − Mn · (Ds λ)α(L0) = n(Ds λ)α(L0) = n(Ds λ)α(Ln), λ)α(Mn), (α + n)(Ds (α + n)(Ds (α + s + n)(Ds λ)α(Ys+n) − Ys+n · (Ds λ)α(L0) = (s + n)(Ds λ)α(Ys+n). Then Claim 1 follows. Claim 2 (Ds λ)0(L0) ≡ 0 (mod Cs λ ⊗ Cs λ). For any Ln, Mn ∈ (Ls λ)n and Ys+n ∈ (Ls nMn and [L0, Ys+n] = (s + n)Ys+n, we obtain λ)s+n, applying (Ds λ)0 to [L0, Ln] = nLn, [L0, Mn] = Ln · (Ds λ)0(L0) = Mn · (Ds λ)0(L0) = Ys+n · (Ds λ)0(L0) ≡ 0 (mod Cs λ ⊗ Cs λ), which forces (Ds λ)0(L0) ≡ 0 (mod Cs λ ⊗ Cs λ) according to Lemma 3.4. Claim 3 For any (Ds λ)α ∈ Der(Ls λ, V s λ), (2.8) is a finite sum. For any α ∈ Z∗ 0} is an infinite set, then Ds infinite sum, which is not in V s Then Claim 3 follows. s, one can suppose (Ds λ(L0) = (Ds λ)α = (vα)inn for some vα ∈ (V s λ)0(L0) + Pα∈∆s L0 · vα = (Ds λ)α. If ∆s = {α ∈ Z∗ s vα 6= λ)0(L0) + Pα∈∆s α vα is an λ to V s λ. λ is a derivation from Ls λ, contradicting the fact that Ds 7 Claim 4 H 1(Ls λ, Qs λ) = 0. The exact sequence 0 → hs → Ls the Hochschild-Serre spectral sequence λ → Ls λ/hs → 0 induces an exact sequence of low degree in 0 −→ H 1(Ls λ/hs, Qs λ hs ) −→ H 1(Ls λ, Qs λ) −→ H 1(hs, Qs λ) Ls λ/hs . Lemma 3.3 forces H 1(Ls can be embedded into HomU (w)(hs, w ⊗ w), which can be easily proved to be zero. Then Claim 4 follows. ) = 0. Since Ls λ/hs ∼= w and Qs ∼= w ⊗ w, H 1(hs, Qs λ/hs, Qs λ hs λ λ/hs λ)Ls Claim 5 H 1(Ls λ, Hs λ/LC s λ) = 0. Denote the subalgebra spanned by {Ln, Mn, c n ∈ Z} of Ls s λ = W s WC immediately from the following Subclaim 1 and Subclaim 2. λ ⊗ W s λ ⊗ Cs λ + Cs λ and Ys = {Yn+s n ∈ Z}. This claim of the case s = 1 λ as W s λ, J s λ = W s λ ⊗ hs + hs ⊗ W s λ, 2 follows Subclaim 1 H 1(W s λ, J s λ /WC s λ) = 0. This subclaim can be found in Remark 1 of [13]. Subclaim 2 H 1(W s λ, Ys ⊗ Ys) = 0 for the case s = 1 2 . This subclaim can be proved similar to the proof of Theorem 4.5 (i) of [13]. According to λ(Mn) = 0 for any n ∈ Z and ϕs λ ∈ ) = 0. Then Claim 5 holds for the case s = 1 2 . Subclaim 1 and Subclaim 2, we know that ϕs 2 , which implies ϕs H 1(Ls λ(Ln) = ϕs λ(Y 1 s λ) when s = 1 λ, Hs λ/LC 2 Similar to the proof of Claim 3 of [3], we can prove this claim also holds for the case s = 0. Claim 6 H 1(Ls λ, LC s λ) ∼= D♮ 1 D♮ 2 D♮ 3 d♮ 1 d♮ 2 d♮ 3 d♮ 4 d♮ 5   if s = 1 if s = 1 if s = 1 2 , λ = −1, 2 , λ = −2, 2 , λ /∈ S 1 , if s = 0, λ = 0, 2 if s = 0, λ = −1, if s = 0, λ = −2, if s = 0, λ = 1, if s = 0, λ /∈ S0. This claim follows from Lemmas 3.2, 2.6, 2.7 and Claims 4, 5. By now we have proved Proposition 3.5. (cid:3) The following lemma is still true for Ls λ by employing the technique of Lemma 2.5 in [6]. Lemma 3.6 Suppose v ∈ V s u ∈ Im(1 − τ ) such that v − u ∈ Cs ⊗ Cs. λ such that x · v ∈ Im(1 − τ ) for all x ∈ Ls λ. Then there exists some Proof of Theorem 2.8 This theorem follows from Lemmas 2.6, 2.7, 3.6 and Proposition 3.5.(cid:3) 8 References [1] V.G. Drinfeld, Constant quasiclassical solutions of the Yang-Baxter quantum equation, Soviet Math. Dokl. 28(3) (1983), 667 -- 671. [2] V.G. Drinfeld, Quantum groups, in: Proceeding of the International Congress of Mathe- maticians, Vol. 1, 2, Berkeley, Calif. 1986, Amer. Math. Soc., Providence, RI, 1987, 798 -- 820. [3] H. Fa, Y. Li, J. Li, The Schrodinger-Virasoro type Lie bialgebra: a twisted case, Front. Math. China 6 (4) (2011) 641 -- 657. [4] H. Fa, J. Li, B. Xin, Lie superbialgebra structures on the centerless twisted N = 2 super- conformal algebra, Algebra Colloq. 18 (3) (2011) 361 -- 372. [5] S. Gao, C. Jiang, Y. Pei, Structure of the extended Schrodinger-Virasoro Lie algebra, Alg. Colloq. 16 (2009), 549 -- 566. [6] J. Han, J. Li, Y. Su, Lie bialgebra structures on the Schrodinger-Virasoro Lie algebra, J. Math. Phys. 50 (2009), 083504. [7] M. Henkel, Schrodinger invariance and strongly anisotropic critical systems, J. Stat. Phys. 75 (1994), 1023 -- 1029. [8] M. Henkel, Phenomenology of local scale invariance: from conformal invariance to dynam- ical scaling, Nucl. Phys. B 641 (2002), 405 -- 410. [9] M. Henkel, J. Unterberger, Schrodinger invariance and space-time symmetries, Nucl. Phys. B 660 (2003), 407 -- 412. [10] Q. Jiang, S. Wang, Derivations and automorphism groups of the original deformative Schrodinger-Virasoro algebras, arXiv:1209.3164v1. [11] J. Li, 2-cocycles of twisted deformative Schrodinger-Virasoro algebra, Comm. Algebra 40 (2012), 1933 -- 1950. [12] D. Liu, L. Chen, L. Zhu, Lie superbialgebra structures on the N = 2 superconformal Neveu- Schwarz algebra, J. Geometry and Phys. 62 (2012), 3826 -- 831. [13] D. Liu, Y. Pei, L. Zhu, Lie bialgebra structures on the twisted Heisenberg-Virasoro algebra, J. Alg. 359 (2012) 35 -- 48. [14] J. Li, Y. Su, L. Zhu, 2-cocycles of original deformative Schrodinger-Virasoro algebras, Science in China Series A 51 (2008), 1989 -- 1999. [15] J. Li, Y. Su, Representations of the Schrodinger-Virasoro algebras, J. Math. Phys. 49 (2008), 053512. [16] J. Li, Y. Su, The derivation algebra and automorphism group of the twisted Schrodinger- Virasoro algebra, arXiv:0801.2207v1, (2008). [17] J. Li, Y. Su, B, Xin, Lie bialgebras of a family of Block type, Chinese Annals of Math. (Series B) 29 (2008), 487 -- 500. [18] W. Michaelis, A class of infinite-dimensional Lie bialgebras containing the Virasoro algebras, Adv. Math. 107 (1994), 365 -- 392. [19] W. Michaelis, Lie coalgebras, Adv. Math. 38 (1980), 1 -- 54. [20] W. Michaelis, The dual Poincare-Birkhoff-Witt theorem, Adv.Math. 57 (1985), 93 -- 162. 9 [21] S.H. Ng, E.J. Taft, Classification of the Lie bialgebra structures on the Witt and Virasoro algebras, J. Pure Appl. Alg. 151 (2000), 67 -- 88. [22] C. Roger, J. Unterberger, The Schrodinger-Virasoro Lie group and algebra: representation theory and cohomological study, Ann. Henri Poincar´e 7 (2006), 1477 -- 1529. [23] G. Song, Y. Su, Lie bialgebras of generalized Witt type, Science in China: Series A 49 (2006), 533 -- 544. [24] E.J. Taft, Witt and Virasoro algebras as Lie bialgebras, J. Pure Appl. Alg. 87 (1993), 301 -- 312. [25] S. Tan, X. Zhang, Automorphisms and Verma modules for generalized Schrodinger-Virasoro algebras, J. Algebra 322 (2009), 1379 -- 1394. [26] J. Unterberger, On vertex algebra representations of the Schrodinger-Virasoro Lie algebra, Nuclear Physics B 823 (2009), 320 -- 371. [27] W. Wang, J. Li, B. Xin, Central extensions and derivations of generalized Schrodinger- Virasoro algebras, Alg. Colloq. 19 (2012), 735 -- 744. [28] W. Wang, J. Li, Y. Xu, Derivations and automorphisms of twisted deformative Schrodinger- Virasoro Lie algebras, Comm. Algebra 40 (2012), 3365 -- 3388. [29] Y. Wu, G. Song, Y. Su, Lie bialgebras of generalized Virasoro-like type, Acta Mathematica Sinica, English Series 22 (2006), 1915 -- 1922. [30] Y. Wu, G. Song, Y. Su, Lie bialgebras of generalized Witt type. II, Comm. Algebra 35 (6) (2007), 1992 -- 2007. [31] Y. Xu, J. Li, Lie bialgebra structures on the extended affine Lie algebra ^ sl2(Cq), J. Pure and Applied Alg. 217 (2013), 364 -- 376 [32] B. Xin, G. Song, Y. Su, Hamiltonian type Lie bialgebras, Science in China A 50 (2007), 1267 -- 1279. [33] H. Yang, Y. Su, Lie bialgebras structures on the Ramond N = 2 super-Virasoro algebras, Chaos, Solitons and Fractals 40 (2009), 661 -- 671. [34] X. Yue, Y. Su, Lie bialgebra structures on Lie algebras of generalized Weyl type, Comm. Algebra 36(4)(2008), 1537 -- 1549. [35] X. Zhang, S. Tan, H. Lian, Whittaker modules for the Schrodinger-Witt algebra, J. Math. Phys. 51 (2010), 083524. 10
1709.03910	3	1709	2018-11-14T17:42:31	Cohomology for partial actions of Hopf algebras	[ "math.RA", "math.QA" ]	In this work, the cohomology theory for partial actions of co-commutative Hopf algebras over commutative algebras is formulated. This theory generalizes the cohomology theory for Hopf algebras introduced by Sweedler and the cohomology theory for partial group actions, introduced by Dokuchaev and Khrypchenko. Some nontrivial examples, not coming from groups are constructed. Given a partial action of a co-commutative Hopf algebra $H$ over a commutative algebra $A$, we prove that there exists a new Hopf algebra $\widetilde{A}$, over a commutative ring $E(A)$, upon which $H$ still acts partially and which gives rise to the same cochain complex as the original algebra $A$. We also study the partially cleft extensions of commutative algebras by partial actions of cocommutative Hopf algebras and prove that these partially cleft extensions can be viewed as a cleft extensions by Hopf algebroids.	math.RA	math	COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS ELIEZER BATISTA, ALDA D.M. MORTARI, AND MATEUS M. TEIXEIRA Abstract. In this work, the cohomology theory for partial actions of co-commu- tative Hopf algebras over commutative algebras is formulated. This theory ge- neralizes the cohomology theory for Hopf algebras introduced by Sweedler and the cohomology theory for partial group actions, introduced by Dokuchaev and Khrypchenko. Some nontrivial examples, not coming from groups are constructed. Given a partial action of a co-commutative Hopf algebra H over a commutative algebra A, we prove that there exists a new Hopf algebra eA, over a commutative ring E(A), upon which H still acts partially and which gives rise to the same cohomologies as the original algebra A. We also study the partially cleft extensions of commutative algebras by partial actions of cocommutative Hopf algebras and prove that these partially cleft extensions can be viewed as a cleft extensions by Hopf algebroids. 1. Introduction The history of Hopf algebras began within the context of algebraic topology with the seminal paper by Heinz Hopf, published in 1941, describing the algebraic proper- ties of the cohomology ring of a group manifold [21]. The subject of group cohomology soon became increasingly more independent of its topological background assuming a more algebraic formulation [1, 27]. The first formulation of a cohomology theory of co-commutative Hopf algebras acting over commutative algebras was done by Moss E. Sweedler in 1968 [26], which, in certain sense, became paradigmatic for further developments in this area. Partial group actions, in its turn, had its beginning with the work of Ruy Exel in the classification of certain class of C*-algebras which had an action of the unit circle but which cannot be described as a usual crossed product [19]. A more algebraic formulation for partial actions was done by Mikhailo Dokuchaev and Ruy Exel in [14] and then, partial actions drew the attention of algebraists and allowed further developments in several directions. One of the developments particularly relevant for our discussion here is the notion of twisted partial actions of groups [15] and its globalization [16]. There, one can see the definition of partial 2-cocycles in order to define twisted actions and partial crossed products, this suggested the existence of a general cohomology theory in which these partial 2-cocycles could be placed. This cohomological theory for partial group actions was achieved by Mikhailo Dokuchaev and Mykola Khrypchenko in [17]. This theory is constructed upon partial actions of groups over commutative algebras. As the very notion of a partial action of a group is deeply related with actions of inverse semigroups [20], the same authors, 2000 Mathematics Subject Classification: Primary 16W30; Secondary 16S40, 16S35, 58E40. Key words and phrases: partial Hopf actions, cohomology for partial Hopf actions, cleft extensions. 1 2 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA [18], could place their cohomology theory for partial group actions within the context of cohomology of inverse semigroups, developed by Hans Lausch [23]. Partial actions came into the Hopf algebra context by the work of Stefaan Caenepeel and Kris Janssen in [13]. This work allowed the generalization of classical results in Hopf algebra theory as well several ideas developed for partial group actions, as glo- balization theorem [2], Morita equivalence between the partial smash product and the invariant subalgebra [3], duality for partial actions [4], partial representations [7], etc. For a more detailed account on recent developments of partial actions of groups and Hopf algebras, see [8] and references therein. Of particular interest for the present article are the notions of twisted partial actions of Hopf algebras, partial crossed products and partially cleft extensions of algebras by Hopf algebras [5]. The aim of this paper is exactly to f ormulate a cohomology theory for partial actions of Hopf algebras, in the same spirit of [17], such that the partial 2-cocycles defined in [5] can be placed properly. This cohomological theory is obtained for the case of partial actions of a co-commutative Hopf algebra H acting partially over a commu- tative algebra A. Moreover, one can, without loss of generality, replace the original E(A) ⊆ A and still obtain the same cohomology theory. algebra A by a commutative and co-commutative Hopf algebra eA over a base algebra As we have already learned in [7], the theory of partial actions of Hopf algebras in fact is deeply related to the theory of representations of Hopf algebroids. This opens a totally new landscape to be explored, for example, in this work we prove that partially cleft extensions can be understood as cleft extensions by Hopf algebroids in the sense of Gabriella Böhm and Tomasz Brzezinski [11] and then one can raise new questions on how to put this cohomological theory for partial actions in the context of cohomology for Hopf algebroids [12, 22]. This article is organized in the following way. In Section 2, we give some mathe- matical preliminaries, recalling the notion of a partial action of a Hopf algebra over a unital algebra and giving some examples of such partial actions. Special attention is required for examples 2.5 and 2.6, which will serve as basis for our specific examples of cohomologies given in Subsection 3.4. Section 3 is dedicated to the construction of our cohomological theory for partial actions of a co-commutative Hopf algebra H over a commutative algebra A. We start with the study of a system of idempotents in the commutative convolution algebras Homk(H ⊗n, A), for a natural n. In subsection 3.2 we define the cochain complex, C • par(H, A) associated for the partial action of H upon A. The ideas involved follows the classical construction due to M.E. Sweedler [26] but in order to overcome the complexities coming from partial actions we define some auxiliary operators which help us to prove that the coboudary operator is a morphism of abelian groups (Theorem 3.9) and that it is nilpotent in this context (Theorem 3.11). Example 3.14 considers the case of H = kG, for G a given group, in this case, we recover the cohomology theory for partial group actions developed by M. Dokuchaev and M. Khrypchenko in [17]. More specific examples, namely, the cohomology theory for partial group actions and partial group gradings over the base field are given in Subsection 3.4. More specifically, in Example 3.17 we calculate the first cohomology groups of a partial grading of the base field by the Klein four group, this extends extends what has been done in [6]. In Section 4, given a partial action of H on A, we first define the reduced cochain par(H, A)/A× this new cochain complex produces the same par(H, A) ∼= C n complex eC n COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 3 cohomology groups as the original cochain complex C n par. Next, we consider the alge- par(H, A). This defines a commutative and co-commutative Hopf algebra over the commutative algebra E(A), which is the subalgebra of A generated bra eA, which is a quotient of the free commutative algebra generated by the images of all cochains f ∈ eC n by elements of the form h · 1A. One proves that eA generates the same cohomolo- par(H, eA) ∼= H n over a commutative and co-commutative Hopf algebra eA. Section 5 is devoted to gies as the original algebra A, that is, for any n ∈ N , we have the isomorphisms H n par(H, A). Then, without loss of generality, we can consider only the cohomological theory for partial actions of a co-commutative Hopf algebra H analise twisted partial actions and partial crossed products [5]. For the case of a co-commutative Hopf algebra H and a commutative algebra A, all twisted partial actions are, in fact partial actions. Nonetheless, we still can get nontrivial partial crossed products by means of chosing a partial 2-cocycle. In fact, the partial crossed products are classified by the second partial cohomology group H 2 par(H, A). The new feature which appears in the context of partial actions is that the crossed product eA#ωH has a structure of a Hopf algebroid over the same base algebra E(A) (Theorem 5.10). This suggests that the cohomology theory for partial actions can be viewed as a cohomological theory for Hopf algebroids. Section 6 is devoted to study partially cleft extensions and then we have Theorem 6.6, which we consider to be the main result of this paper. Basically, we proved that given a partially cleft extension B of a commutative co-commutative Hopf algebra subalgebra E(A), namely, the partial smash product E(A)#H, such that B is a eA by a co-commutative Hopf algebra H, there exists a Hopf algebroid over the base E(A)#H-cleft extension of eA in the sense of G. Böhm and T. Brzezinski [11]. Finally, For sake of a concise exposition, we omit several proofs which consists in basic manipulation of properties of partial actions. The reader interested in more details can find them in the apendice of this paper. in Section 7 we give some perspectives for future developments. 2. Mathematical preliminaries Given a field k, k-bialgebra or a Hopf algebra H and a k-algebra A, for each n ≥ 0 we have the convolution algebras Homk(H ⊗n, A), with convolution product given by f ∗ g(h1 ⊗ · · · ⊗ hn) = f (h1 (1) ⊗ · · · ⊗ hn (1))g(h1 (2) ⊗ · · · ⊗ hn (2)), and unit 1(h1 ⊗ · · · ⊗ hn) = ǫH (h1) . . . ǫH (hn)1A. In particular, for n = 0, we have Homk(H ⊗0, A) = Homk(k, A) ∼= A. The following result can be easily obtained, we leave the details of the proof to the reader. Proposition 2.1. Let H be a cocommutative bialgebra, or Hopf algebra and A be a commutative algebra. Then, for every n ≥ 0 the convolution algebras Homk(H ⊗n, A) are commutative. 4 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA Henceforth, for sake of simplicity, we will denote f(h1⊗· · ·⊗hn) just by f(h1, . . . , hn). The main concept underlying this work is the notion of a partial action of a Hopf algebra H over an algebra A. Definition 2.2. [13] A partial action of a Hopf algebra H over an algebra A is a linear map · : H ⊗ A → A, such that, for every a, b ∈ A and h, l ∈ H, we have (PA1) 1H · a = a; (PA2) h · (ab) = (h(1) · a)(h(2) · b); (PA3) h · (l · a) = (h(1) · 1A)(h(2)l · a). We say that the partial action is symmetric if, in addition, we have (PA3') h · (l · a) = (h(1)l · a)(h(2) · 1A) . The algebra A is said to be a partial H-module algebra. Note that, if H is a cocommutative Hopf algebra and A is a commutative algebra, then every partial action of H on A is automatically symmetric. Example 2.3. [2] Let G be a group, recall that a partial action of G over an algebra A is a pair ({Ag}g∈G, {αg : Ag−1 → Ag}g∈G), where Ag is an ideal of A for each g ∈ G and αg is an isomorphism of (not necessarily unital) algebras. We say that the partial action of G on A is unital if, for each g ∈ G, Ag = 1gA, where 1g is a central idempotent in A and αg is a unital isomorphism between Ag−1 and Ag. For the case where H = kG, the group algebra of G, symmetric partial actions of kG are in one to one correspondence with unital partial actions of G. This correspondence can be easily seen: Given a unital partial action ({Ag = 1gA}g∈G, {αg : Ag−1 → Ag}g∈G), one defines · : kG ⊗ A → A, by δg · a = αg(1g−1 a). Conversely, given a symmetric partial action of kG over A, define, for each g ∈ G, the idempotents 1g = δg · 1A, by them, construct the ideals Ag = 1gA and the isomorphisms αg = δg · Ag−1 . Example 2.4. [2] Given a Hopf algebra H, a left H-modules algebra B and a central idempotent e ∈ B, one can define a partial action of H on A = eB. Denoting by ⊲ the action of H over B, the induced partial action is given by h · ea = e(h ⊲ (ea)), for every a ∈ B and h ∈ H. The next two examples will be explored with more details throughout this paper for giving examples of cohomologies. Example 2.5. Consider a group G, let us see the partial actions of the Hopf algebra H = kG over A = k, the base field. A partial action· : kG ⊗ k → k, associates to each : K → k, this is the same as defining a linear g ∈ G the linear transformation δg · functional λ : kG → k. Denoting λ(δg) simply by λg, one can write δg · a = λga, for every a ∈ k. Using the functional λ, The axiom (PA1) says that λe = 1, where e is the neutral element of the group G. Axiom (PA2), in its turn, implies that λg = λgλg, for every g ∈ G, and consequently λg = 1 or λg = 0. Define H = {g ∈ G λg = 1}, it is clear that e ∈ G. Axiom (PA3) says that λgλh = λgλgh, this implies that for g, h ∈ H, we have gh ∈ H. Finally, putting h = g−1 in the previous identity, we conclude that g ∈ H implies that g−1 ∈ H, therefore H is a subgroup of G. It is easy to see that the action is global if, and only if H = G. Then, we can label the partial actions of kG over k by the subgroups of G. Xg∈G λpg =Xh∈G λpg = 1; λphλph−1g ; λpg λph = λpgh−1 λph = λpgh−1 λpg . Pg∈G Defining L = {g ∈ G λpg 6= 0}, one can see that L is a subgroup of G: First, as λpg = 1 then there exists some g ∈ G such that λpg 6= 0, and therefore L 6= ∅. Moreover, given g, h ∈ L, the third equation says that 0 6= λpg λph = λpgh−1 λpg , which implies that λpgh−1 6= 0, and therefore gh−1 ∈ L. In order to analyze the possible values of λpg , for g ∈ L, take g = h in the third equation, then λpg λpg = λpgg−1 λpg = λpeλpg . This implies that, λpg = λpe, ∀ g ∈ L. Finally, from the first equation, 1 =Xg∈G λpg =Xg∈L λpe = Lλpe, COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 5 Example 2.6. Let G be a finite abelian group and consider H = (kG)∗ = hpg g ∈ Gi, the dual of the group algebra, with bialgebra structure given by pgph = δg,hpg, ph ⊗ ph−1g, ǫ(pg) = δg,e. 1 =Xg∈G pg, ∆(pg) =Xg∈G As in the previous example, partial actions of (kG)∗ over k are associated to a linear functional λ : (kG)∗ → k defined by λ(pg) = λpg = pg · 1. In this case, the axioms for a partial action (PA1), (PA2) and (PA3) become, respectively and therefore λpg = λpe = 1 that the action is global if, and only if, L = {e}. L , for all g ∈ L. We leave to the reader the verification We conclude that the partial actions of (kG)∗ over the base field k are classified by subgroups of G and given by λpg =(cid:26) 1 L 0 , g ∈ L , c.c. 3. Cohomology for partial actions In his 1968's seminal article, M.E. Sweeder presented a cohomology theory for com- mutative algebras which are modules over a given cocommutative Hopf algebra. Ba- sically, the cochain complexes C n(H, A) are defined as the abelian groups consisting of the invertible elements of the commutative convolution algebras Homk(H ⊗n, A). The main difference between the cohomology theory of global and partial actions is that in the partial case one needs to find appropriated unital ideals in the convo- lution algebras in order to define correctly the cochain complexes. These ideals are constructed upon a system of idempotents for the convolution algebras. Henceforth, H is always a cocommutative Hopf algebra acting partially on a com- mutative algebra A with partial action · : H ⊗ A → A. 3.1. A system of idempotents for the convolution algebras. We start introdu- cing some idempotent elements of Homk(H ⊗n, A) which are very important through- out this paper. As the convolution algebras are commutative, for each n ≥ 0, an idempotent is automatically a central idempotent. Moreover, the convolution pro- duct of a finite number of idempotents is also an idempotent. In what follows, we introduce a nested system of idempotents in the convolution algebra related to the partial action. 6 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA Proposition 3.1. (1) For each n ≥ 1, the linear maps H ⊗n (h1 ⊗ · · · ⊗ hn) −→ A 7→ (h1 . . . hn) · 1A een : are idempotent in the corresponding convolution algebras Homk(H ⊗n, A). (2) Let n < m and e ∈ Homk(H ⊗n, A) an idempotent, then ∈ Homk(H ⊗m, A) en,m = e ⊗ ǫ ⊗ · · · ⊗ ǫ is an idempotent in Homk(H ⊗m, A). m−n {z } Proof: The proof of item (1) follows from basically from item (PA2) of definition of a partial action. Item (2) is straightforward. (cid:4) Definition 3.2. For n ≥ 1 and 1 ≤ l ≤ n arbitrary, we define: eel,n :=eel ⊗ ǫ ⊗ · · · ⊗ ǫ {z } Note that, for l = n in the definition aboveeen,n =een. Definition 3.3. For n ≥ 1, we define n−l . en :=ee1,n ∗ee2,n ∗ · · · ∗een ∈ Homk(H ⊗n, A). Homk(H ⊗n, A). Proposition 3.4. For any (h1 ⊗ . . . ⊗ hn) ∈ H ⊗n we have that The next proposition gives us a useful characterization of the idempotents en ∈ en(h1, . . . , hn) = h1 · (h2 · (· · · · (hn · 1A) · · · )). Proof: The proof of this result consists basically in elementary manipulations of itens (PA2) and (PA3) of the definition of a partial action. (cid:4) 3.2. Cochain complexes. Based on the idempotents defined in the previous sub- sectin, one can define the cochain complexes for the partial action of H on A. For each n > 0 define the following ideals of Homk(H ⊗n, A): I(H ⊗n, A) = en ∗ Homk(H ⊗n, A) = {en ∗ g : g ∈ Homk(H ⊗n, A)}. As en is a central idempotent, the ideal I(H ⊗n, A) can be considered as a unital algebra with unit en. An element f ∈ I(H ⊗n, A) is said to be (convolution) invertible in this ideal if there is another element g ∈ I(H ⊗n, A) such that f ∗ g = g ∗ f = en. Definition 3.5. Let H be a cocommutative bialgebra and A be a partial H-modules algebra with partial action · : H ⊗ A → A. For n > 0 , a "partial" n-cochain of H taking values in A is an invertible element in the ideal I(H ⊗n, A). Denoting by par(H, A) = (I(H ⊗n, A))×. For C n n = 0 we say that a 0-cochain is an invertible element in the agebra A, that is C 0 par(H, A) the set of n cochains, we have that C n par(H, A) is an abelian group with relation to the convolution product, par(H, A) = A×, is an abelian group with the ordinary multiplication in A par(H, A) = A×. Note that C n while C 0 and the unit e0 = 1A. COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 7 Definition 3.6. For an arbitrary f ∈ C n define the "partial" coboundary operator (1) · f (h2 (δnf )(h1, . . . , hn+1) = (h1 (1), . . . , hn+1 (1) )) ∗ par(H, A), (h1 ⊗ · · · ⊗ hn+1) ∈ H ⊗n+1, we f (−1)i (h1 (i+1), . . . , hi (i+1)hi+1 (i+1), . . . , hn+1 (i+1)) ∗ f (−1)n+1 (h1 (n+2), . . . , hn (n+2)) ∗ nYi=1 If n = 0 and a ∈ A×, we have (δ0a)(h) = (h · a)a−1. The challenge is to prove that the coboundary operators are well defined, that is, par (H, A). Moreover, one needs par(H, A), the map δnf is indeed in C n+1 for every f ∈ C n to prove that the sequence C 0 par(H, A) δ0→ C 1 par(H, A) δ1→ · · · δn−1→ C n par(H, A) δn→ C n+1 par (H, A) δn+1→ · · · par(H, A) and C n+1 is a cochain complex, that is, each δn is a homomorphism of abelian groups between C n par(H, A). For this purpose, we introduce some auxiliary operators which will help us to describe the coboundary operators in a more intrinsic way and whose properties will lead us to the desired results. par (H, A) satisfying δn+1 ◦ δn(f ) = en+2, for each f ∈ C n Definition 3.7. (1) For each n ≥ 0, define the map En : C n par(H, A) −→ Homk(H ⊗n+1, A) f 7→ En(f ) , given by En(f )(h1, . . . , hn+1) := h1 · f (h2, . . . , hn+1). (2) For n < m, define the map, in,m : C n par(H, A) −→ Homk(H ⊗m, A) f 7→ in,m(f ) given by in,m(f )(h1, . . . , hm) := f (h1, . . . , hn)ǫ(hn+1) . . . ǫ(hm). (3) For i ∈ {1, . . . , n}, define the map µi : H ⊗n+1 (h1 ⊗ · · · ⊗ hn+1) −→ H ⊗n 7→ (h1 ⊗ · · · ⊗ hihi+1 ⊗ · · · ⊗ hn+1). With these auxiliary operators, the coboundary operator can be rewritten as δn : C n par(H, A) −→ C n+1 par (H, A) f 7→ δn(f ) := En(f )∗ f (−1)i ◦ µi ∗ in,n+1(f (−1)n+1 ) and the properties of δn are based upon the properties of these operators. nQi=1 par(H, A), we have En(f ∗ g) = En(f ) ∗ En(g). par(H, A), we have in,m(f ∗ g) = in,m(f ) ∗ in,m(g). (i) For f, g ∈ C n Lemma 3.8. (ii) En(en) = en+1. (iii) For n < m and f, g ∈ C n (iv) in,m(en) ∗ em = em. (v) For f, g ∈ C n (vi) (en ◦ µn) ∗ in,n+1(en) = en+1. (vii) (en ◦ µi) ∗ en+1 = en+1, ∀ i ∈ {1, · · · , n − 1}. par(H, A), we have (f ∗ g) ◦ µi = (f ◦ µi) ∗ (g ◦ µi), ∀ i ∈ {1, . . . , n}. 8 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA Proof: The proof of these items is done by a long, but straightforward computation evaluating on elements of the corresponding tensor product. (cid:4) Theorem 3.9. For any n ≥ 0, f ∈ C n is an element of C n+1 a morphism of abelian groups. par (H, A). Moreover, the map δn : C n par(H, A), the linear map δn(f ) : H ⊗n+1 → A par (H, A) is par(H, A) → C n+1 Proof: If f ∈ C n the expression of δn(f ), par(H, A), thenf = f ∗en and it is invertible by convolution. Consider δn(f ) := En(f )∗ f (−1)i ◦ µi ∗ in,n+1(f (−1)n+1 ). nYi=1 Items (i), (iii) and (v) of Lemma 3.8 and using the fact that the convolution algebra Homk(H ⊗n+1, A) is commutative, we conclude that δn(f ∗ g) = δn(f ) ∗ δn(g), in par- ticula rδn(f ) = δn(f ∗ en) = δn(f ) ∗ δn(en). By item (ii), we know that En(en) = en+1 and by items (iv), (vi) and (vii) we see that the unit en+1 absorbs the other factors, leading to δn(en) = en+1. Then we have δn(f ) = δn(f ) ∗ en+1. A straightforward cal- culation leads us to δn(f −1) = (δn(f ))−1 Therefore, we prove that δn is well defined and it is a morphism of abelian groups. (cid:4) par(H, A), we need the In order to prove that δn+1 ◦ δn(f ) = en+2, for every f ∈ C n following lemma. Lemma 3.10. Let f ∈ C n−1 par (H, A), then (i) En(in−1,n(f )) = in,n+1(En−1(f )). (ii) (en−1 ◦ µi ◦ µi+1) ∗ en+1 = en+1, ∀ i ∈ {1, · · · , n − 1}. (iii) (en−1 ◦ µi ◦ µi+j) ∗ en+1 = en+1, ∀ i ∈ {1, · · · , n − 1}, j ∈ {2, · · · , n − i}. (iv) En(f ◦ µi) = En−1(f ) ◦ µi+1, ∀ i ∈ {1, . . . , n − 1}. (v) En ◦ En−1(f ) = i1,n+1(ee1) ∗ (En−1(f ) ◦ µ1). (vi) in,n+1(f ◦ µi) ∗ in−1,n(f −1) ◦ µi = in,n+1(en−1 ◦ µi), ∀ i ∈ {1, . . . , n − 1}. (vii) (in−1,n(f ) ◦ µn) ∗ in−1,n+1(f −1) = in−1,n+1(en−1). (viii) (f ◦ µi ◦ µi) ∗ (f −1 ◦ µi ◦ µi+1) = en−1 ◦ µi ◦ µi, ∀ i ∈ {1, . . . , n − 1}. (ix) (f ◦ µi ◦ µi+j) ∗ (f −1 ◦ µi+j−1 ◦ µi) = en−1 ◦ µi ◦ µi+j, ∀ i ∈ {1, . . . , n − 2}, ∀ j ∈ {2, . . . , n − i}. Proof: The proofs of these items come after a long but straightforward calculation. (cid:4) With this lemma, one can prove the following result. Theorem 3.11. For any f ∈ C n par(H, A), we have that δn+1 ◦ δn(f ) = en+2. Proof: For f ∈ C n par(H, A), we have δn+1(δn(f )) = En+1(δn(f ))∗ n+1Qi=1 = En+1(En(f )∗ f (−1)j ◦ µj ∗ in,n+1(f (−1)n+1 )) (δn(f ))(−1)i ◦ µi ∗ in+1,n+2((δn(f ))(−1)n+2 ) nQj=1 (En(f (−1)i )∗ ∗ n+1Qi=1 ∗in+1,n+2(En(f (−1)n+2 nQj=1 f (−1)i+j ◦ µj ∗ in,n+1(f (−1)n+i+1 )) ◦ µi f (−1)n+j+2 ◦ µj ∗ in,n+1(f (−1)2n+3 )). )∗ nQj=1 COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 9 expression is equal to en+2. Then, using the items of Lemma 3.8 and Lemma 3.10, one can show that this (cid:4) par(H, A), δn)n∈N, which allows Therefore, we ended up with a cochain complex (C n us to define a cohomology theory. 3.3. Cohomologies. Definition 3.12. Let H be a cocommutative Hopf algebra acting partially over a commutative algebra A and consider the cochain complex (C n par(H, A), δn)n∈N defined in the previous sections. For n > 0, define the abelian groups Z n(H, A) = ker δn, Bn(H, A) = Im δn−1 e H n(H, A) = ker δn/ Im δn−1, respectively, as the groups of partial n-cocycles, partial n-coboundaries and partial n-cohomologies of H taking values in A. n ≥ 1. For n = 0, define H 0(H, A) = Z 0(H, A) = ker δ0. Let us characterize the partial cocycles and the partial coboundaries for n = 0, 1 and 2. For n = 0, we have by definition H 0(H, A) = Z 0(H, A) = {a ∈ A×h · a = (h · 1A)a, ∀h ∈ H}. Thus the partial 0-cocycles are the elements of A invariant under the partial action as defined in [3]. For n = 1, the partial 1-coboundaries are B1(H, A) = Imδ0 = {f ∈ C 1 par(H, A) ∃a ∈ A×, f (h) = δ0(a)(h)}, this means B1(H, A) = {f ∈ C 1 par(H, A) ∃a ∈ A×, f (h) = (h · a)a−1}. Also, for f ∈ C 1 par(H, A), we have δ1(f )(h, l) = E2(f )(h(1), l(1))f −1(h(2)l(2))f (h(3))ǫ(l(3)) = (h(1) · f (l(1)))f −1(h(2)l(2))f (h(3))ǫ(l(3)). Then, forall h, l ∈ H, the partial 1-cocycles are Z 1(H, A) = {f ∈ C 1 par(H, A) δ1(f )(h, l) = e2(h, l)} = {f ∈ C 1 = {f ∈ C 1 par(H, A) (h(1) · f (l(1)))f −1(h(2)l(2))f (h(3)) = h · (l · 1A)} par(H, A) (h(1) · f (l(1)))f (h(3)) = (h(1) · (l(1) · 1A))f (h(2)l(2))}. Due to the fact that for a 1-cocycle f we have f = e1 ∗ f , then the condition of 1-cocycle can also be rewritten as (h(1) · f (l(1)))f (h(3)) = (h(1) · 1A)f (h(2)l(2)), For n = 2, we have the partial 2-coboundaries B2(H, A) = {g ∈ C 2 par(H, A) δ1(f )(h, l) = g(h, l)} = {g ∈ C 2 par(H, A) g(h, l) = (h(1) · f (l(1)))f −1(h(2), l(2))f (h(3))}. Also, for f ∈ C 2 par(H, A), we have δ2(f )(h, l, m) = E2(f )∗ f (−1)i ◦ µi ∗ i2,3(f (−1)3 )(h, l, m) = (h(1) · f (l(1), m(1)))f −1(h(2)l(2), m(2))f (h(3), l(3)m(3))f −1(h(4), l(4))ǫ(m(4)). Then, the partial 2 cocycles are 2Qi=1 10 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA = = Z 2(H, A) = {f ∈ C 2 {f ∈ C 2 par(H, A) δ2(f )(h, l, m) = e3(h, l, m), ∀ h, l, m ∈ H} par(H, A) (h(1) · f (l(1), m(1)))f −1(h(2)l(2), m(2))f (h(3), l(3)m(3)) f −1(h(4), l(4)) = h · (l · (m · 1A)), ∀ h, l, m ∈ H} {f ∈ C 2 par(H, A)(h(1) · f (l(1), m(1)))f (h(2), l(2)m(2)) = (h(1) · (l(1) · (m(1) · 1A)))f (h(2)l(2), m(2))f (h(3), l(3)), ∀ h, l, m ∈ H}. Again by absorption of units, one can rewrite the condition of 2-cocycle as (h(1) · f (l(1), m(1)))f (h(2), l(2)m(2)) = (h(1) · 1A)f (h(2)l(2), m(2))f (h(3), l(3)), which is the form presented in [5]. Example 3.13. In the case of a global action of H over A, which is equivalent to say that h · 1A = ǫ(h)1A, ∀h ∈ H, the cochain complexes are simply given by C n(H, A) = Homk(H ⊗n, A)×. Then we recover exactly the cohomology theory ob- tained by Sweedler in [26]. Example 3.14. Let G be a group and H = kG, the group algebra of G. Using the canonical basis {δg ∈ kG g ∈ G}, the axioms (PA1), (PA2) and (PA3) of partial action read (PA1) δe · a = a, for every a ∈ A; (PA2) δg · (ab) = (δg · a)(δg · b), for every g ∈ G and a, b ∈ A; (PA3) δg · (δh · a) = (δg · 1A)(δgh · a), for every g, h ∈ G and a ∈ A. In order to calculate the partial n-cocycles, partial n-coboundaries and partial n-cohomologies, we denote the coboundary operator by ∂n instead of δn to avoid confusion with the elements δg ∈ kG. For n = 0, we have H 0(kG, A) = Z 0(kG, A) = {a ∈ A×(δg · a)a−1 = (δg · 1A), ∀δg ∈ kG}, For n = 1, the 1-coboundaries are B1(kG, A) = {f ∈ C 1 = {f ∈ C 1 par(kG, A) ∃a ∈ A×, par(kG, A) f (δg) = (δg · a)a−1}. f (δg) = ∂0(δg)(a)} Also, we have, for every f ∈ C 1 par(kG, A), ∂1(f )(δg, δh) = (δg · f (δh))f −1(δgh)f (δg). Then, for all δg, δh ∈ kG, we obtain the 1-cocycles Z 1(kG, A) = {f ∈ C 1 = {f ∈ C 1 par(kG, A) ∂1(f )(δg, δh) = δg · (δh · 1A)} par(kG, A) (δg · f (δh))f (δg) = (δg · 1A)f (δgh)}. For n = 2, the 2 coboundaries are B2(kG, A) = {i ∈ C 2 = {i ∈ C 2 par(kG, A) ∂1(f )(δg, δh) = i(δg, δh)} par(kG, A) i(δg, δh) = (δg · f (δh))f −1(δgh)f (δg)}. Moreover, for f ∈ C 2 par(kG, A) ∂2(f )(δg, δh, δl) = (δg · f (δh, δl))f −1(δgh, δl)f (δg, δhl)f −1(δg, δh). COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 11 Then, for all δg, δh, δl ∈ kG, the partial 2-cocycles are Z 2(kG, A) = {f ∈ C 2 par(kG, A) ∂2(f )(δg, δh, δl) = e3(δg, δh, δl)} = {f ∈ C 2 par(kG, A) (δg · f (δh, δl))f (δg, δhl) = (δg · 1A)f (δgh, δl)f (δg, δh)}. This cohomology for partial actions of the group algebra kG corresponds to the cohomology for partial group actions described in [17]. Recall from Example 2.3 that there is a one-to-one correspondence between partial actions f kG and unital partial Ag−1 . actions of the group G, given by Ag = 1gA in which 1g = δg ·1A, and αg = δg · For elements x1, . . . xn ∈ G we define the ideals A(x1,...,xn) := Ax1Ax1x2 . . . Ax1...xn, in which Axi = 1xiA. This expression for the ideals are natural, considering the units en(δx1, . . . , δxn) = δx1 · (δx2 · (· · · (δxn · 1A))) = 1x11x1x2 . . . 1x1...xn. The set of these ideals forms a semilattice, because the product of two ideals of this type is also an ideal of this type, this product is commutative and each ideal is idempotent, that is A(x1,...,xn) = A(x1,...,xn)A(x1,...,xn). This can be viewed easily by the properties of the system of idempotents presented before. The correspondence between the cochain complexes presented here and those pre- sented in [17] can be viewed more exactly by the identification of the convolution algebra Homk(kG⊗n, A) with the algebra of functions F un(Gn, A), moreover, the functions f : Gn → A can also be viewed as a collection of elements of A indexed by n-tuples in G, that is f (g1, ..., gn) = fg1,...gn ∈ A. As the canonical basis elements δg, for g ∈ G are group-like, the convolution product is in fact the pointwise product, that is, for f 1, f 2 ∈ F un(Gn, A) and g1, . . . , gn ∈ G, we have f 1 ∗ f 2(g1, ..., gn) = f 1(g1, ..., gn)f 2(g1, ..., gn) = f 1 Therefore, the n-cochains C n The partial n-cocycles, partial n-coboundaries and partial n-cohomologies in the par(G, A). g1,...,gn. par(kG, A) coincide with the n cochains C n g1,...,gnf 2 group setting are written as. H 0(G, A) = Z 0(G, A) = {a ∈ A×(αg(1g−1a))a−1 = 1g, ∀g ∈ G}, For n = 1 the partial 1-coboudaries are B1(G, A) = {f ∈ C 1 = {f ∈ C 1 par(G, A) ∃a ∈ A×, par(G, A) ∃a ∈ A×, f (g) = ∂0(g)(a)} f (g) = (αg(1g−1a))a−1}. Moreover, for f ∈ C 1 par(G, A) we have ∂1(f )(g, h) = (g · (1g−1f (h)))f −1(gh)f (g). Then, the partial 1-cocycles are Z 1(G, A) = {f ∈ C 1 = {f ∈ C 1 par(G, A) ∂1(f )(g, h) = e2(g, h), ∀ g, h ∈ G} par(G, A) (αg(1g−1f (h)))f (g) = 1gf (gh), ∀ g, h ∈ G}. Note that δg · (δh · 1A) = (δg · 1A)(δgh · 1A) = 1g1gh, and 1gh is absorbed by f (gh). For n = 2, the partial 2-coboundaries are B2(G, A) = {i ∈ C 2 = {i ∈ C 2 par(G, A) ∂1(f )(g, h) = i(g, h)} par(G, A) i(g, h) = (αg(1g−1f (h)))f −1(gh)f (g)}. 12 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA For f ∈ C 2 par(G, A), we have, ∂2(f )(g, h, l) = (g · (1g−1f (h, l)))f −1(gh, l)f (g, hl)f −1(g, h), then, Z 2(G, A) = {f ∈ C 2 = {f ∈ C 2 par(G, A) ∂2(f )(g, h, l) = e3(g, h, l), ∀ g, h, l ∈ G} par(G, A) (αg(1g−1 f (h, l)))f (g, hl) = 1gf (gh, l)f (g, h), ∀ g, h, l ∈ G}. Again, the appearance only of 1g in the right hand side of the 2-cocycle condition is due to absorption of units. Therefore, the cohomology obtained is the same as [17]. In the next subsection we will give more specific examples of cohomologies for partial actions in which the algebra A is the base field k. 3.4. Cohomology for partial actions on the base field. Example 3.15. (Partial group actions over the base field) Let G be a group. We have already seen in Example 2.5 that partial actions kG over k are in correspon- dence with subgroups L ≤ G by means of the linear functional λ : kG −→ k δg 7→ λg = λ(δg) =(cid:26) 1 , g ∈ L 0 , c.c. . Fix the subgroup L of G which defines the partial action. Let us now calculate the par(kG, k) (we use the symbol ∂n for the coboundary map to avoid cohomologies H n confusion with the basis elements δg ∈ kG): • For n = 0, C 0 par(kG, k) = k× = k\{0}. Let a ∈ C 0, then, (∂0a)(δg) = (δg · a)a−1 = λgaa−1 = λg =(cid:26) 1, g ∈ L 0, g 6∈ L Therefore, H 0 par = Z 0 • For n = 1, let f ∈ Z 1 par = k×. par = C 0 par(kG, k). Then, (∂1f )(δg, δh) = λgf (δh)f −1(δgh)f (δg) = λgλh Denote f (δg) simply by f (g) using the identification between the convolution algebra and the algebras of functions f : G → k. If g, h ∈ L, then the 1 cocycle condition can be rewritten as f (gh) = f (g)f (h), which means that fL : kG → k× is a character of the subgroup L. If g /∈ L then it is easy to see that for a 1 cocycle f , we have f (g) = 0. As the partial 1-coboundaries are given by λ : G → k such that λ(g) = 1 for any g ∈ L, we have that H 1 = Z 1/B1 are given by the nontrivial 1-dimensional representations of the subgroup L which determines the partial action. • For n = 2, let ω ∈ Z 2 par(kG, k). Then, denoting ω(δg, δh) simply by ωg,h, we have (∂ω)(δg, δh, δl) = λgωh,lω−1 gh,lωg,hlω−1 g,h = λgλhλl COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 13 It is easy to see from the identity above that, if (g, h) /∈ L×L then ωg,h = 0. Then, defining ω : G × G → k by ω(g, h) =(cid:26) 0, ω(δg, δh), (g, h) 6∈ L × L (g, h) ∈ L × L , we have that the partial 2-cocycles relative to G are in fact usual 2-cocycles of the subgroup L [1, 27], in other words Z 2 par(kG, k) = Z 2(L, k). Example 3.16. (Partial group gradings over the base field) Let G be a finite abelian group. In Example 2.6 we saw that the partial actions of the Hopf algebra H = (kG)∗ = < pg g ∈ G > over the base field k are in one-to-one correspondence with subgroups L ≤ G, namely λpg =(cid:26) 1 L 0 , g ∈ L , c.c. Let us now calculate explicitly the partial cohomologies for (kG)∗. For n = 0, recalling that δa(h) = (h·a)a−1, for every a ∈ k×, we have δa(pg) = λpg , and this leads to Z 0((kG)∗, k) = C 0((kG)∗, k) = H 0((kG)∗, k) = k×. Moreover, the 1-coboundaries are basically given by the functional λ. For n = 1, let ω : (kG)∗ → k be a partial 1-cocycle, then λpg λph = δ(ω)(pg, ph) = Xl,m,i∈G = Xl,m∈G LXh∈L λphω(ph−1g) ⇒ (e1 ∗ ω)(pg) =Xh∈G 1 This means that (pml−1 · ω(phi−1))ω(plpi)ω(pm−1g) λml−1ω(phl−1))ω(pl)ω(pm−1g). ω(pgh−1) = ω(pg) = 1 LXh∈L ω(ph−1g). Recalling that e1 ∗ ω = ω = ω ∗ e1 e que e1(pg) = λpg , we have ω(pg) = ω(phg) = 1 LXh∈L ω(pgh). 1 LXh∈L LPh∈L 1 For any g ∈ L, we have ω(pg) = ω(ph), which is an invariance by translations in the subgroup. Furthermore, using the normalization condition, ω(1) = 1, we have that Lω(pe) =Xg∈L ω(pg) = 1 ⇒ ω(pg) = , ∀ g ∈ L. 1 L We don't have, a priori any further constraint for the values of ω(pg), for g /∈ L. If we impose that, ω(pg) = 0, for g /∈ L, then the only possible choice is the linear functional λ : (kG)∗ → k, which defines the partial action. Therefore, Z 1 B1 H 1 par((kG)∗, k) = {ω : kg → k, ω(pg) = 1 L , par((kG)∗, k) = {λ} par((kG)∗, k) = {ω : kg → k, ω(pg) = 1 g ∈ L}, L , g ∈ L, ω(pg) 6= 0, g /∈ L} 14 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA For n = 2, first recall that e2 =ee2,1 ∗ee2,2, in which e2(h, l) = h · (l · 1A),ee2,1(h, l) = (h · 1A)ǫ(l) andee2,2(h, l) = hl · 1A. Then, for g, h ∈ G, we have ω(pg, ph) =ee2,1 ∗ ω(pg, ph) =Xl∈G LXl∈L This leads to an invariance by translation on the left slot, that is ω(plg, ph) = λplω(pl−1g, ph) = ω(pl−1g, ph). 1 ω(pg, ph), for any g, h ∈ G and l ∈ L. On the other hand, ω(pg, ph) = ee2,2 ∗ ω(pg, ph) = Pl,m∈L ω(pl−1g, pl−1h) = = 1 LPl∈L λplpmω(pl−1g, pm−1h) ω(pg, pl−1h) 1 LPl∈L This gives an invariance by translation on the right slot, that is, ω(pg, plh) = ω(pg, ph), for any g, h ∈ G and l ∈ L. As these invariances are independent, we have finally ω(plg, pmh) = ω(pg, ph), for any g, h ∈ G and l, m ∈ L [6]. This translation invariance is a useful tool for searching solutions of partial 2-cocycles in specific cases. Besides the translation invariance, we have the normalization constraint, given by ω(1, pg) = ω(pg, 1) = λpg ⇒ 1 L =Xh∈L ω(pg, ph) =Xh∈L Finally, we have the cocycle condition. For g, h, i ∈ L, we have ω(ph, pg), ∀ g ∈ L. 1 L3 = λpg λphλpi = δω(pg, ph, pi) λplω(pr, px)ω( pl−1mpr−1s , px−1y)ω(pm−1n, ps−1tpy−1i )ω(pn−1g, pt−1h) ⇒ l−1m=r−1s ⇒ l=rms−1 } {z ⇒s−1t=y−1i ⇒ t−1=i−1ys−1 {z } ω(pr, px)ω(pr−1s, px−1y)ω(pm−1n, py−1i)ω(pn−1g, pi−1ys−1h) 1 1 = = x,y∈L t,x,y∈L = Xl,m,n,r,s, L Xm,n,r,s, L Xm,n,s,y∈L L2 Xn,s,y∈L Xm∈L L Xn,s,y∈L L Xn,y∈L L Xn,x∈L = = = 1 1 1 1 (∗) = λpsλpy ω(pm−1n, py−1i)ω(pn−1g, pi−1ys−1h) ω(pm−1n, py−1i)! ω(pn−1g, pi−1ys−1h) 1 L ω(pn, py−1i)ω(pn−1g, ps−1yi−1h) 1 L ω(pn, py−1i)ω(pn−1g, pyi−1h) ω(pn, px)ω(pn−1g, px−1h), in which (∗) is taken putting y−1i = x and i−1y = yi−1 = x−1. Therefore, COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 15 1 L2 = Xn,x∈L ω(pn, px)ω(pn−1g, px−1h) The next example is a specific case of a partial grading of the base field for a fixed group G and a fixed subgroup L ≤ G defining the partial action. Example 3.17. Fix G = ha, b a2 = b2 = ei = {e, a, b, ab} and L = hai, then L = 2. Let us calculate the partial 1-cocycles in this case. The invariance by translations gives us ω(pe) = ω(pa) = x and ω(pb) = ω(pab) = y, ω(pe) = ω(pa) = x e ω(pb) = ω(pab) = y. By the normalization constraint Pg∈G ω(pg) = 1 = Pg∈G ω(pg), we have (1) and (2) x + y = 1 2 x + y = 1 2 Moreover, the condition ω ∗ ω = e, which can be written as ω(ph)ω(ph−1g) =(cid:26) 1 Xh∈G L 0 g ∈ L g /∈ L , gives us two equations, (3) (4) Finally, the cocycle condition, xx + yy = 1 4 , xy + yx = 0 λpg λph = Xm∈G ω(pm)ω(pm−1h)ω(pmh−1g) gives us • For g = h = e, ω(pe)ω(pe)ω(pe) + ω(pa)ω(pa)ω(pa) + ω(pb)ω(pb)ω(pb) + ω(pab)ω(pab)ω(pab) = 1 4 xy=−yx ⇒ 2x2x + 2y2y = 1 ⇒ x2x + xyy + yyx + y2y = 1 8 ⇒ 1 8 = x2x + xyy + yyx + y2y = (x + y)(xx + yy) 4 ⇒ x2x + y2y = 1 8 ⇒ x2x + xyy − xyy + y2y = 1 8 This is the product of equations (1) and (3), therefore, no new information is added. The same occurs for g = e and h = a, g = a and h = e, and g = h = a. • For g = e e h = b, ω(pe)ω(pb)ω(pb) + ω(pa)ω(pab)ω(pab) + ω(pb)ω(pe)ω(pe) + ω(pab)ω(pa)ω(pa) = 0 ⇒ xyy + xyy + xyx + xyx = 0 ⇒ xyy + yxx = 0. As x + y = 1 2 , then we have xy = 0. The same condition is obtained for the cases g = e and h = ab, g = a and h = b, g = a and h = ab, g = b and h = e, g = b and h = a, g = ab and h = e, and g = ab and h = a. 16 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA • For g = b e h = b, ω(pe)ω(pb)ω(pe) + ω(pa)ω(pab)ω(pa) + ω(pb)ω(pe)ω(pb) + ω(pab)ω(pa)ω(pab) = 0 ⇒ ⇒ xyx=−xyy ⇒ ⇒ x2y + x2y + y2x + y2x = 0 ⇒ x2y + y2x = 0 x2y + xyy − xyy + y2x = 0 x2y + xyy + yxx + y2x = 0 0 = x2y + xyy + yxx + y2x = (x + y)(xy + yx) This equation is the product of (1) and (4), therefore, no new information is added. The same occurs if we take g = b and h = ab, g = ab and h = b, and g = h = ab. Resuming, we have the following equations: 1 2 1 2 1 4 x + y = , x + y = , xx + yy = , xy + yx = 0, xy = 0, whose unique possible solution is ω(pg) = λpg =(cid:26) 1 L 0 , g ∈ L , g 6∈ L . For n = 2, first note that ω(pg, ph) = ω(pag, ph) = ω(pg, pah) = ω(pag, pah), for every g, h ∈ G, then, • ω(pe, pe) = ω(pa, pe) = ω(pe, pa) = ω(pa, pa); • ω(pb, pe) = ω(pab, pe) = ω(pb, pa) = ω(pab, pa); • ω(pe, pb) = ω(pa, pb) = ω(pe, pab) = ω(pa, pab); • ω(pb, pb) = ω(pab, pb) = ω(pb, pab) = ω(pab, pab). The normalization constraint gives us Xh∈G ω(pg, ph) = λpg =Xh∈G ω(ph, pg). Applying the above normalization constraint respectively for g = e, a, b, ab, we have • For g = e (λpe = 1/2), ω(pe, pe) + ω(pe, pa) + ω(pe, pb) + ω(pe, pab) = ω(pe, pe) + ω(pa, pe) + ω(pb, pe) + ω(pab, pe) = 1 2 1 2 , ⇒ ω(pe, pe) + ω(pe, pb) = 1 4 and ω(pe, pb) = ω(pb, pe) The same is obtained for g = a (λpa = 1 2 ). • For g = b (λpb = 0), ω(pb, pe) + ω(pb, pa) + ω(pb, pb) + ω(pb, pab) = 0 ω(pe, pb) + ω(pa, pb) + ω(pb, pb) + ω(pab, pb) = 0, ⇒ ω(pb, pb) = −ω(pe, pb) = −ω(pb, pe) The same is obtained for g = ab (λpab = 0). Therefore, the only remaining independent components are ω(pe, pe) and ω(pe, pb). Moreover ω(pe, pe) + ω(pe, pb) = 1 4 . COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 17 The cocycle condition h(1) · ω(l(1), m(1))ω(h(2), l(2)m(2)) = (h(1) · 1)ω(h(2), l(1))ω(h(3)l(2), m), can be written, in our case, as ω(pg, ps)ω(phs−1, pls−1) =Xs∈G Xs∈G ω(pgs−1, phs−1)ω(ps, pl). This condition gives us 64 equations which are, in fact redundant, that is, every 2-cochain in this case is a 2-cocycle Finally, from ω ∗ ω = e2, taking x = ω(pe, pe) and y = ω(pe, pe), we obtain the equation [6] 16xy − 3(x + y) + 1 2 = 0. 4. The associated Hopf algebra of a partial action In [17], for a partial action of a group G on a commutative algebra A, the authors introduced an inverse semigroup eA, given by all the invertible elements of all ideals of the form 1x1 . . . 1xnA, for x1, . . . , xn ∈ G and n ∈ N. Once showed that θx(1x−1eA) = 1xeA, in other words, θ restricted to eA defines a partial action eθ of G on eA such that This construction brings advantages because eA possesses a richer structure than A and then one can study, for example, extension theory by partial group actions from a wider perspective, namely, the theory of extensions of inverse semigroups [18]. their cohomologies are the same, that is, H n par(G, eA). par(G, A) ∼= H n In our context, we can have also very similar constructions, allowing us to trade partial actions of a co-commutative Hopf algebra H on a commutative algebra A by a the same cohomology. partial action of H on a commutative and co-commutative Hopf algebra eA generating In order to proceed the construction of this new Hopf algebra, one has a tech- nical obstruction concerning the invertible elements of the algebra A. Indeed, the multiplicative abelian group A× embeds into the abelian group C n par(H, A) for each n ∈ N by means of the group monomorphisms φn : A× → C n par(H, A), given by φn(a) = aen. These morphisms φn are coherent with the coboundary morphisms, that is, for each n ∈ N, we have δn ◦ φn(a) = δn(aen) = aen+1 = φn+1(a). Therefore, one can construct a new cochain complex which gets rid of these invertible elements and yet defining the same cohomology. C n par(H,A) A× . Definition 4.1. Let H be a co-commutative Hopf algebra and A be a commutative partial H-module algebra. We define, for n ∈ N the n-th reduced partial cochain par(H, A) as the quotient abelian group mology groups isomorphic to those relative to the cochain complex C • group eC n Proposition 4.2. The reduced partial cochain complex eC • eH n Proof: Indeed, denote, for each n ∈ N, the n-th reduced cohomology group by par(H, A) by ψn([f ]) 7→ [f A×] par(H, A) and define the map ψn : H n One can easily see that this map is well defined, surjective and it is a morphism of abelian groups. The injectivity comes from the fact that given a partial n-cochain par(H, A) and a ∈ A×, we have af ∗ f −1 = aen = δn−1(aen−1), then f and af f ∈ C n par(H, A) → eH n par(H, A) generates coho- par(H, A). 18 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA are cohomologous. Therefore the cohomology groups H n isomorphic. par(H, A) and eH n par(H, A) are (cid:4) Remark 4.3. We will denote the reduced n-cochains again by f instead of f A× in order to make the notation cleaner. It is clear also that at level zero we have tative unital algebra par(H, A) = {1A}. eC 0 Now, define the algebra eA as the quotient eA = bA The set of variables runs over the distinct f ∈ eC n bA = k[X1A, Xf (h1,...,hn) n ≥ 1, h1, . . . , hn ∈ H, f ∈ eC n par(H, A), that is, if f and g are two partial n-cochains such that f = g, then, for every h1 ⊗ · · · ⊗ hn ∈ H ⊗n we have Xf (h1,...hn) = Xg(h1,...hn). The ideal I is taken exactly to recover certain properties from the original algebra A and from the partial partial action of H. This ideal is generated by elements of the type I , in which bA is the free commu- par(H, A)]. (5) X1A − 1; (6) Xf (h1,...,Pi λihj λiXf (h1,...,hj par(H, A), ∀n > 0; i ,...,hn) −Xi Xi i ,...hn), for each f ∈ eC n λiXen1 (h1,1,...,h1,n1 ) . . . X (hki ,1,...,h ki,nki ) , enki (7) (8) for each zero combinationPi λien1(h1,1, . . . , h1,n1) . . . enki (9) X(h·(f1(h1,1,...,h1,n1)+...+fm(lm,1,...,lm,nm)))−(cid:16)X(h·f1(h1,1,...,h1,n1 ))+. . .+X(h·fm(lm,1,...,lm,nm ))(cid:17) ; (2)) − X(f ∗g)(h1 ,...hn); (hki,1, . . . , hki,nki ) = 0 ∈ A; (1))Xg(h1 (1),...,hn (2),...,hn Xf (h1 (10) X(1H ·f (h1,...,hn)) − Xf (h1,...,hn); (11) X(h·(f1(h1,1,...,h1,n1 )...fm(lm,1,...,lm,nm)))−X(h(1)·f1(h1,1,...,h1,n1 )) . . . X(h(m)·fm(lm,1,...,lm,nm)); and (12) Remark 4.4. X(h·(k·f (h1,...,hn))) − X(h(1)·1A)X(h(2)k·f (h1,...,hn)). one element h · (f1(h1,1, . . . , h1,n1 ) . . . fm(lm,1, . . . , lm,nm)) can be written as (1) Note that I is indeed an ideal of the algebra bA, for example, according to Definition 3.7. En1+···nm(cid:0)in1,n1+···+nm(f1) ∗ (ǫ⊗n1 ⊗ in2,n2+···+nm(f2)) ∗ · · · · · · ∗ (ǫ⊗(n1+···+nm−1) ⊗ fm)(cid:17) (h, h1,1, . . . , h1,n1 , . . . , lm,1, . . . , lm,nm), the unit of the algebra eA. (2) The condition (5) means that the unit of the algebra A will play the role of COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 19 the units of the cochain groups eC • (3) The condition (7) refers to every linear combination of monomials involving par(H, A) which vanish in the algebra A. Of course, some of these relations are in fact present among elements of the form (6), but there are other vanishing linear combinations in A involving partial actions of elements of H upon the unit 1A which needed to be ruled out in order to remember the structure of A. (4) Casting out elements of the form (6) is needed to remember that the genera- tors of eA are linear maps between H ⊗n and A. In particular, in the quotient we have identities of the type Xf (h1,...,hi,...,hn) = Xf (h1,...,hi ǫH(hi all n > 0. ,...,hn), for each i ∈ {1, ..., n}, for each f ∈ eC n (5) Casting out elements of the form (8) is needed in order to make the relations (2)) = par(H, A) for (1),...,hn)ǫH(hi (1))Xf (h1,...,hi (2) coming from the convolution product between cochains still valid in eA. (6) Finally, we need to mod out elements of the form (9), (10), (11) and (12) in order to recover the linearity of the partial action of H on A and the identities coming from axioms (PA1), (PA2) and (PA3). After taking the quotient, as far as it doesn't lead to a misunderstanding, we are This map is well defined, because among the generators of the ideal I which defines Moreover, the unit map is injective. This can be easily seen considering the evaluation involving the units of the cochain complex. Also, by construction it is an algebra Define also the subalgebra of A, E(A) = hh · 1A h ∈ Hi and the unit map going to denote the classes Xf (h1,...,hn) + I ∈ eA simply by Xf (h1,...,hn) η : E(A) → eA given by η((h1 · 1A) . . . (hn · 1A)) = Xe1(h1) . . . Xe1(hn). the algebra eA there are all linear combinations representing null combinations in A morphism (note that η(1A) = X1A = 1 eA ∈ eA, consequently, eA is a E(A) algebra. map bev : bA → A which simply associate to each element ofba ∈ bA its value at bev(ba) ∈ A. It is easy to see that bev(I) = 0, then one can define a linear map ev : eA → A with the same content. Therefore, if η(a) = 0 in eA, then a = ev(η(a)) = 0. By the injectivity of the unit map, one can identify E(A) with its image in η(E(A)) ⊆ eA. of variables Xe, in which is the image of an idempotent in eC n en ∈ eC n Every element in the image of E(A) different of 1 eA can be written as a combination par(H, A) for some n > 0. In fact, what we are going to prove is that one can rewrite an element of the form (h1 · 1A) . . . (hn · 1A) as a linear combination of images of the idempotent par(H, A). Let us make induction on the number n of factors h · 1A involved. Now, suppose that the result is valid for r ∈ N, 1 ≤ r < n, that is, For n = 1, we have h · 1A = e1(h). (h1 · 1A)(h2 · 1A) . . . (hr · 1A) = er(l1 i , . . . , lr i ) sXi=1 for some elements lj then i ∈ H, for i ∈ {1, . . . , s} and j ∈ {1, . . . r}. Take h1, . . . , hn ∈ H, (h1 · 1A)(h2 · 1A) . . . (hn · 1A) = (h1 · 1A)[(h2 · 1A) . . . (hn · 1A)] · (ln−1 i , . . . , ln−1 i i · (l2 i · (. . . · 1A) . . . ))) i = Pi(h1 · 1A)en−1(l1 ) =Pi(h1 · 1A)(l1 20 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA = Pi(h1 = Pi h1 = Pi en(h1 (1) · 1A)(h1 (1) · (S(h1 (2)S(h1 i · (l2 (2))l1 (1), S(h(2))l1 i , l2 i · (l2 (3))l1 i · (. . . i , . . . , ln−1 i i ). i · (. . . · (ln−1 · (ln−1 i · 1A) . . . ))) · 1A) . . . ))) This proves our claim. Moreover, for each n > 0, we have en(h1, . . . , hn) ∈ E(A). Indeed, en(h1, . . . , hn) = h1 · (h2 · (. . . · (hn · 1A) . . . )) = (h1 = (h1 = (h1 (1) · 1A)(h1 (1) · 1A)(h1 (1) · 1A)(h1 (2)h2 · (. . . (2)h2 (2)h2 · (hn · 1A) . . . )) (2)h3 · (. . . (n)h2 (3)h2 (1) · 1A)(h1 (1) · 1A) . . . (h1 · (hn · 1A) . . . )) (n−1) . . . hn−1 (2) hn · 1A) ∈ E(A). advantage of getting the same cohomology theory as the original algebra A. Theorem 4.5. Let H be a co-commutative Hopf algebra H and A be commutative Hopf algebra which is also a partial H-module algebra such that for any n ∈ N the n-cohomology group H n Our construction will enable us to see a richer structure on the algebra eA with the partial H-module algebra A. Then the algebra eA is a commutative and co-commutative par(H, eA) is isomorphic to the n-cohomology H n Proof: We have already shown that eA is a commutative algebra over E(A). For the coalgebra structure, define the map b∆ : bA → bA ⊗E(A) bA, given by, b∆(Xf1(h1,1,...,h1,n1 ) . . . Xfm(lm,1,...,lm,nm )) = (1) ) . . . Xfm(lm,1 (2) ) . . . Xfm(lm,1 ) ⊗ Xf1(h1,1 = Xf1(h1,1 (2) ,...,lm,nm (1) ,...,lm,nm par(H, A). (2),...,h (1),...,h 1,n1 1,n1 ), (1) (2) 1 (h1,1,...,h1,n1 ) . . . Xf −1 verify that I is a Hopf ideal. Most of the verifications are long, but straightforward. for f1 ∈ eC n1(H, A), . . . , f m ∈ eC nm(H, A). And the mapbǫ : bA → E(A), given by bǫ(Xf1(h1,1,...,h1,n1 ) . . . Xfm(lm,1,...,lm,nm )) = en1(h1,1, . . . , h1,n1) . . . enm(lm,1, . . . , lm,nm). Finally, we define the antipode bS : bA → bA as bS(Xf1(h1,1,...,h1,n1 ) . . . Xfm(lm,1,...,lm,nm )) = Xf −1 for f1 ∈ eC n1(H, A), . . . , f m ∈ eC nm(H, A). One needs first to show that these maps can be well defined in eA that is, we must Basically, for bǫ, as its image lies in E(A) ⊆ A, where the relations are valid, then bǫ(I) = 0. For bS, it is also easy to see that bS(I) ⊆ I. Therefore, one can define algebra maps ǫ : eA → E(A) and S : eA → eA, (S is an algebra map because eA is commutative) The most involved ones are the verifications for b∆. For this task, it is convenient to divide the process into two steps. First, we consider the ideal J E bA generated only by elements of the form (5), (6) and (7). For elements of the form (5) and (6), it is quite straightforward, now take an element of the form (7) that is, a linear combination with the same form. m (lm,1,...,lm,nm ), x =Xi λiXen1 (h1,1,...,h1,n1 ) . . . X enki (hki ,1,...,h ki,nki ) ∈ J such that bev(x) = 0. Then, we have COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 21 ) (1) (h (2) X 2,n2 1,n1 2,n2 2,n2 1,n1 2,n2 ⊗ 1 X ) . . .X ,...,h ) enki (h enki (h (1),...,h (2),...,h (1),...,h 1,n1 ,...,h 1,n1 ,...,h (1),...,h (h ki,1 (2) ,...,h (1),...,h (2),...,h (2),...,h enki (h enki (h enki (h enki (h enki (h (h ki,1 (1) ki,nki (2) ki,nki (1) ) ki,nki (1) ) ki,nki (2) ) ki,nki (2) ki,nki (2) ki,nki (1) ki,1 (1) ,...,h ki ,1 (2) ,...,h ki ,1 (1) ,...,h ki ,1 (2) ,...,h ki ,1 (2) ,...,h ki ,1 (1) ,...,h Xen2 (h2,1 1,n1 (1) ). . .X 1,n1 (2) ) . . .X ⊗Xen1(h1,1 (2) ) . . . X λiXen1(h1,1 enki ki ,1 (2) ,...,h 1,n1 (1) )Xen1 (h1,1 (1) )Xen2 (h2,1 ) ⊗1 ) ⊗1 ) (2),...,h (2) ) . . . X enki ki,1 (1) ,...,h 1,n1 ki,nki (1) Xen1 (h1,1 λiXen1(h1,1 λiXen1 (h1,1 (2),...,h (2) ). . .X X ki,nki (1) (1) )Xen2 (h2,1 ki,nki (2) (1),...,h (1) ). . .X enki (1) )Xen1 (h1,1 ⊗1+. . . +Xen1 (h1,1,...,h1,n1 )Xen2 (h2,1,...,h2,n2 ). . . ) (2) ) − Xen1 (h1,1,...,h1,n1 )(cid:19) b∆(x)=Pi =Pi =Pi λi(cid:18)(cid:18)Xen1 (h1,1 =Pi +Xen1 (h1,1,...,h1,n1 )(cid:18)Xen2 (h2,1 . . . X +Pi Therefore, b∆(x) ∈ bA ⊗ J + J ⊗ bA. Then, one can define a new linear map ∆ : bA/J →(cid:16)bA/J(cid:17) ⊗E(A)(cid:16)bA/J(cid:17) ∼=(cid:16)bA ⊗E(A) bA(cid:17) /(cid:16)bA ⊗E(A) J + J ⊗E(A) bA(cid:17) with the same form. Recall that in bA/J we have identities of the form Xf (h1,...,hn) = Now define the ideal I′ E bA/J generated by the elements of the form (8), (9), (10), (2) )− Xen2 (h2,1,...,h2,n2 )(cid:19) . . . ki,nki )! ⊗ 1! (11) and (12). Take an element of the form (8), x = Xf (h1 X(f ∗g)(h1 ,...hn) ∈ I′, then, enki λiXen1 (h1,1,...,h1,n1 ) . . . X (2)) . . . ǫH(hn (1))ǫH(h1 (1))Xg(h1 ki ,1 (2) ,...,h ki,1 (1) ,...,h (2)) − (1),...,hn (hki,1,...,h (hki,1,...,h Xf (h1 ki,nki (1) ki,nki (2) ki,nki ) enki (h ,...,hn (1) ,...,hn (2) (2)). X ) − X ) enki ⊗ 1. (h enki (1),...,hn (1))Xg(h1 (3),...,hn (3)) ⊗ Xf (h1 (2),...,hn (2))Xg(h1 (4),...,hn (4)) ∆(x) = Xf (h1 −X(f ∗g)(h1 = Xf (h1 (1) ,...,hn −X(f ∗g)(h1 +X(f ∗g)(h1 −X(f ∗g)(h1 (1) (1) (1),...hn (1))Xg(h1 ,...hn ,...hn (1)) ⊗ X(f ∗g)(h1 (2) ,...hn (2)) ⊗ Xf (h1 (2)) (2) (3) (2) (3) ,...,hn ,...,hn (1)) ⊗ Xf (h1 (1)) ⊗ Xf (h1 (1)) ⊗ X(f ∗g)(h1 ,...,hn (2))Xg(h1 (2))Xg(h1 ,...hn (2)) ⊗ Xf (h1 (2)) . . . ǫH(hn (2),...,hn (1))ǫH(h1 ,...,hn (2) (2) (3) (2)) (3),...,hn (2)) ⊗ Xf (h1 (3))Xg(h1 ,...,hn (4)) (4) (3))Xg(h1 ,...,hn (3)) ,...,hn (3)) = Xf (h1 (1) (1),...,hn −X(f ∗g)(h1 ,...hn (1))Xg(h1 (1),...hn +X(f ∗g)(h1 (1),...hn =(cid:16)Xf (h1 ⊗Xf (h1 (1) ,...,hn ,...,hn (3) +X(f ∗g)(h1 (1) (1))Xg(h1 (3))Xg(h1 (2) ,...,hn (1)) ⊗(cid:16)Xf (h1 (1)) ⊗(cid:16)Xf (h1 ,...,hn (4) ,...hn (4),...,hn (4)) (4),...,hn (3))Xg(h1 (3),...,hn (3)) − X(f ∗g)(h1 (1))ǫH(h1 (2),...hn (2)) . . . ǫH(hn (3),...,hn ,...hn (2),...,hn (2))Xg(h1 (2)) − X(f ∗g)(h1 (4)) (1) ,...,hn (2))Xg(h1 (3) ,...,hn (3)) − X(f ∗g)(h1 (2) (2) ,...hn (4)) (2))(cid:17) (2))(cid:17) (2))(cid:17). 22 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA the same for elements of the form (9), (10), (11) and (12). Therefore, there exists a Therefore ∆(x) ∈ I′ ⊗(cid:16)bA/J(cid:17) +(cid:16)bA/J(cid:17) ⊗ I′. With similar strategies, one can prove well defined algebra map ∆ : eA → eA ⊗E(A) eA with the same form on generators. It is easy to see that (eA, µ, η, ∆, ǫ) gives a commutative and co-commutative bial- eC n1(H, A), . . . f m ∈ eC nm(H, A) we have Let us verify the antipode axioms, (I ∗ S) = (S ∗ I) = η ◦ ǫ. (S ∗ I)(Xf1(h1,1,...,h1,n1 ) . . . Xfm(lm,1,...,lm,nm )) gebra over the base algebra E(A). =µ(S ⊗ I) ◦ ∆(Xf1(h1,1,...,h1,n1 ) . . . Xfm(lm,1,...,lm,nm )) =S(Xf1(h1,1 (1) ). . . Xfm(lm,1 1,n1 =Xf −1 (1) ) . . . Xf −1 m (lm,1 1 (h1,1 =X(f −1 1 ∗f1)(h1,1,...,h1,n1 ) . . . X(f −1 =Xen1 (h1,1,...,h1,n1 ) . . . Xenm (lm,1,...,lm,nm ) = η ◦ ǫ(Xf1(h1,1,...,h1,n1 ) . . . Xfm(lm,1,...,lm,nm )) (2) ). . . Xfm(lm,1 (2) ) . . . Xfm(lm,1 (2) ,...,lm,nm (2) ,...,lm,nm )Xf1(h1,1 )Xf1(h1,1 Indeed, for f1 ∈ (1) (1) ,...,lm,nm (1) ,...,lm,nm (1) m ∗fm)(lm,1,...,lm,nm ) ,...,h ,...,h ,...,h 1,n1 ,...,h 1,n1 1,n1 (2) ) ) (2) (2) (2) (1) (1) and co-commutative Hopf algebra over E(A). Analogously, we have the equality (I ∗ S) = η ◦ ǫ. Therefore, eA is a commutative One can define a partial action of H on eA, • : H ⊗ eA → eA. First, define a linear map ◮: H ⊗ bA → bA given by h◮(Xf1(h1,1,...,h1,n1) . . . Xfm(lm,1,...,lm,nm))=X(h(1)·f1(h1,1,...,h1,n1)) . . . X(h(m)·fm(lm,1,...,lm,nm)). For each h ∈ H, one can prove that h ◮ I ⊆ I. For example, taking an element x = Xf (h1 (1),...,hn (1))Xg(h1 (2),...,hn (2)) − X(f ∗g)(h1,...hn), we have h ◮ x = Xh(1)·f (h1 = Xh(1)·f (h1 (1) (1) ,...,hn ,...,hn (1))Xh(2)·g(h1 (1))Xh(2)·g(h1 (2) (2) ,...,hn ,...,hn (2)) − Xh·(f ∗g)(h1,...hn) (2)) − Xh·(f (h1 ,...,hn (1) (1))g(h1 (2) ,...,hn (2))) ∈ I from the fact that · is a partial action of H on A. h•(Xf1(h1,1,...,h1,n1 ) . . . Xfm(lm,1,...,lm,nm))=X(h(1)·f1(h1,1,...,h1,n1 )) . . . X(h(m)·fm(lm,1,...,lm,nm )). Then, there is a well defined map • : H ⊗ eA → eA, again, given by. It is straightforward to show that • is a partial action of H on eA. This follows directly Finally, it remains to verify that A and eA generate the same cohomology groups, that is, for any n ∈ N we have H n to prove is that H n par(H, A), which implies our result. First note that, for any n ∈ N and h1 ⊗ · · · ⊗ hn ∈ H ⊗n we have Xen(h1,...,hn) = Xh1·(···(hn·1A)··· ) = h1 • (· · · (hn • X1A) · · · ) = h1 • (· · · (hn • 1 eA) · · · ). For each n ∈ N, a reduced partial n-cochain f ∈ eC n par(H, A) generates a partial n- par(H, eA) given by ef (h1, . . . , hn) = Xf (h1,...,hn). On the other hand, cochain ef ∈ C n par(H, eA), in order to be convolution invertible, must be of the form g(h1, . . . hn) = Xg(h1,...,hn), for some g ∈ eC n par(H, A), for each h1 ⊗ · · · ⊗ hn ∈ H ⊗n. Therefore, one can define, for each n ∈ N, two mutually inverses well defined par(H, eA) ∼= eH n par(H, eA) ∼= H n par(H, A). In fact, what we are going each n-cochain g ∈ C n COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 23 and Ψ : H n par(H, A) given by Ψ([g]) 7→ [g]. par(H, A) → H n par(H, eA) → eH n These maps produced the isomorphism between the cohomology groups H n par(H, A), and consequently between H n par(H, A). (cid:4) morphisms of abelian groups Φ : eH n and eH n eA, and in this case, as h · 1A = ǫH(h)1A, the base subalgebra E(A) coincides with the base field. Therefore, the Hopf algebra eA is a commutative and co-commutative par(H, eA) given by Φ([f ]) 7→ [ef ] par(H, eA) Hopf algebra over k which gives the same classical cohomological theory as A. The properties of this Hopf algebra and its role in the classical cohomology theory is still an interesting topic to be explored Remark 4.6. For the classical case of a global action of a co-commutative Hopf algebra H on a commutative algebra A, [26], one can still construct this Hopf algebra par(H, eA) and H n 5. Twisted partial actions and crossed products In reference [5], the authors introduced the notion of a twisted partial action of a Hopf algebra H over an algebra A and described the construction of the crossed product by a 2 cocycle. In particular two crossed products are isomorphic if their associated cocycles are related by a linear map which has properties similar to a convolution invertible 2-coboundary. Nevertheless, we still did not have a cohomology theory underlying those crossed products. In what follows, we shall see that in the case of co-commutative Hopf algebras acting partially over commutative algebras the crossed products are indeed classified by the second cohomology group defined before. Definition 5.1. [5] Let H be a Hopf algebra and A be a unital algebra (with unit 1A). Let · : H ⊗ A → A and ω : H ⊗ H → A two linear maps. The pair (·, ω) is called a twisted partial action of H over A if, (TPA1) 1H · a = a, for every a ∈ A. (TPA2) h · (ab) = (h(1) · a)(h(2) · b), for every h ∈ H and a, b ∈ A. (TPA3) (h(1) · (l(1) · a))ω(h(2), l(2)) = ω(h(1), l(1))(h(2)l(2) · a), for every h, l ∈ H and a ∈ A. (TPA4) ω(h, l) = ω(h(1), l(1))(h(2)l(2) · 1A), for every h, l ∈ H. In this case, we say that (A, ·, ω) is a twisted partial H-module algebra.. Definition 5.2. [5] Let H be a Hopf algebra and (A, ·, ω) be a twisted partial H- module algebra as above. como acima. Define over A ⊗ H a multiplication given by (a ⊗ h)(b ⊗ l) =X a(h(1) · b)ω(h(2), l(1)) ⊗ h(3)l(2) for every a, b ∈ A e h, l ∈ H. We define the partial crossed product as A#ωH = (A ⊗ H)(1A ⊗ 1H ). Proposition 5.3. [5] Given a Hopf algebra H and (A, ·, ω) a twisted partial H- modules algebra, the partial crossed product A#ωH is unital if, and only if, (13) ω(h, 1H ) = ω(1H , h) = h · 1A, ∀h ∈ H. Moreover, the crossed product is associative if, and only if (14) (h(1) · ω(l(1), m(1)))ω(h(2), l(2)m(2)) = ω(h(1), l(1))ω(h(2)l(2), m), ∀h, l, m ∈ H. 24 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA A linear map ω : H ⊗2 → A satisfying (13) and (14) of the above Proposition is called a normalized cocycle. We denote by a#h the element (a ⊗ h)(1A ⊗ 1H) = a(h(1) · 1A) ⊗ h(2) ∈ A#ωH. One can easily deduce that a#h = a(h(1) · 1A)#h(2). There is an injective algebra morphism i : A → A#ωH, given by i(a) = a#1, this enhances the crossed product A#ωH with a left A-module structure. Also one can show that the linear map ρ : A#ωH → A#ωH ⊗ H a#h 7→ a#h(1) ⊗ h(2) defines a right H-comodule algebra structure on A#ωH. This left A-module and right H-comodule structures on the partial crossed product will be important in order to relate crossed products with extensions of A by H. Definition 5.4. [5] Let A = (A, ·, ω) be a twisted partial H-module algebra. We say that the twisted partial action is symmetric if (i) The linear maps e1,2, e2 : H ⊗ H → A, given by e1,2(h, l) = (h · 1A)ǫ(l) and e2(h, l) = hl · 1A are central idempotents in the convolution algebra Homk(H ⊗ H, A); (ii) The map ω satisfies the cocycle condition (14) and it is an invertible element in the ideal he1,2 ∗ e2i ⊂ Hom(H ⊗ H, A). (iii) For any h, l ∈ H, we have e2(h, l) = (h · (l · 1A)) =P(h(1) · 1A)(h(2)l · 1A) = The algebra A is called, in this case, a symmetric twisted partial H-module algebra. (e1,2 ∗ e2)(h.l). For the case of a co-commutative Hopf algebra H and a commutative algebra A, every symmetric twisted partial action of H over A is in fact a partial action. Proposition 5.5. Let H be a co-commutative Hopf algebra and A be a commutative symmetric twisted partial H-module algebra, then A is a partial H-module algebra. Proof: Indeed, from axiom (TPA3) of Definition 5.1, we conclude that X(h(1) · (l(1) · a))ω(h(2), l(2)) =X ω(h(1), l(1))(h(2)l(2) · a), P(h · (l · a)) = P ω(h(1), l(1))(h(2)l(2) · a)ω−1(h(3), l(3)) = P ω(h(1), l(1))ω−1(h(2), l(2))(h(3)l(3) · a) = P(h(1) · (l(1) · 1A))(h(2)l(2) · a) = P(h(1) · 1A)(h(2)l(1) · 1A)(h(2)l(2) · a) = P(h(1) · 1A)(h(2)l · a) By the commutativity of the convolution algebra Hom(H ⊗H, A), we also conclude that h · (l · a) = (h(1)l · a)(h(2) · 1A). Therefore, A is a partial H-module algebra. (cid:4) In the case of H being a co-commutative Hopf algebra and A being a partial H module algebra, we still can define a partial crossed product for each 2-cocycle ω ∈ Z 2 par(H, A). In fact, all possible partial crossed products which can be constructed in this case are classified by the second cohomology H 2 par(H, A) COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 25 5.1. Partial crossed products and H 2 par(H, A). Theorem 4.1 of reference [5] gives a necessary and sufficient condition on two different symmetric partial actions of a Hopf algebra H over an algebra A in order to get the associated crossed products isomorphic. Theorem 5.6. [5] Let A be a unital algebra and H a Hopf algebra with two symmetric twisted partial actions, h ⊗ a 7→ h · a and h ⊗ a 7→ h • a, with cocycles ω and σ, respectively. Suppose that there is an isomorphism Φ : A#ωH → A#σH which is also a left A-module and a right H-comodule map. Then there exists linear maps u, v ∈ Homk(H, A) such that for all h, k ∈ H, a ∈ A (i) u ∗ v(h) = h · 1A; (ii) u(h) = u(h(1))(h(2) · 1A) = (h(1) · 1A)u(h(2)); (iii) h • a = v(h(1))(h(2) · a)u(h(3)) (iv) σ(h, k) = v(h(1))(h(2) · v(h(2)))ω(h(3), k(2))u(h(4)k(3)); (v) Φ(a#ωh) = au(h(1))#σh(2). Conversely, given maps u, v ∈ Homk(H, A), satisfying (i),(ii), (iii) and (iv) and, in addition u(1H ) = v(1H ) = 1A, then the map Φ, as presented in (v), is an isomorphism of algebras. For the case of a cocommutative Hopf algebra H and a commutative algebra A, itens (i) and (ii) imply that u is the convolution inverse of v in the ideal e1 ∗ Homk(H, A). Item (iii), in its turn, implies that the two partial actions • and · are equal. At last, but not least, item (iv) can be rewritten as σ ∗ ω−1(h, k) = (h(1) · v(k(1)))u(h(2)k(2))v(h(3)) = δ1(v)(h, k). Therefore, one can rewrite Theorem 5.6 as: Theorem 5.7. Let H be a co-commutative Hopf algebra and A be a partial H module algebra. Then, given two partial 2-cocycles ω, σ ∈ Z 2 par(H, A), the associated partial crossed products A#ωH and A#σH are isomorphic if, and only if, ω and σ are coho- mologous, that is, they belong to the same class in the cohomology group H 2 par(H, A). In order to conclude that the second partial cohomology fully classify all the iso- morphism classes of partial crossed products, it remains to check that every class in H 2 par(H, A) contains a normalized two cocycle. Proposition 5.8. Given a partial 2-cocycle ω ∈ Z 2 par(H, A), there exists a normalized 2-cocycle eω ∈ Z 2(H, A), which is cohomologous to ω. Proof: Indeed, take ω a 2-cocycle, then ω satisfies (h(1) · ω(k(1), l(1)))ω(h(2), k(2)l(2)) = ω(h(1), k(1))ω(h(2)k(2), l). Putting h = 1H in the expression above , we have (1H · ω(k(1), l(1)))ω(1H , k(2)l(2)) = ω(1H , k(1))ω(1H k(2), l) ⇒ ω(k(1), l(1))ω(1H , k(2)l(2)) = ω(1H , k(1))ω(k(2), l) ⇒ ω(k(1), l(1))ω(k(2), l(2))ω(1H , k(3)l(3)) = ω(1H , k(1))ω(k(2), l(1))ω(k(3), l(2)) ⇒ k(1) · (l(1) · 1A)ω(1H , k(2)l(2)) = ω(1H, k(1))(k(2) · (l(1) · 1A)) k=1H⇒ (l(1) · 1A)ω(1H , l(2)) = ω(1H, 1H ) · (l · 1A) 26 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA As ω−1(1H , 1H )ω(1H , 1H ) = 1H · 1A = 1A, we conclude that ω(1H , 1H ) ∈ A×. (l · 1A). Then, one can defineeω(h, k) = ω(h, k)(ω(1H , 1H ))−1. It is easy to see thateω(1H , l) = On the other hand, putting l = 1H in the 2-cocycle condition, we have (h(1) · ω(k(1), 1H ))ω(h(2), k(2)) = ω(h(1), k(1))ω(h(2)k(2), 1H ) ⇒ h(1) · ω(k(1), 1H ))ω(h(2), k(2))ω(h(3), k(3)) = ω(h(1), k(1))ω(h(2), k(2))ω(h(3)k(3), 1H ) ⇒ h(1) · ω(k(1), 1H ))(h(2) · (k(2) · 1A)) = (h(1) · (k(1) · 1A))ω(h(2)k(2), 1H ) ⇒ h(1) · ω(k(1), 1H )(k(2) · 1A)) = (h(1) · 1A)(h(2)k(1) · 1A)ω(h(3)k(2), 1H ) ⇒ h · ω(k, 1H ) = (h(1) · 1A)ω(h(2)k, 1H ) k=1H⇒ h · ω(1H , 1H ) = (h(1) · 1A)ω(h(2), 1H ) ⇒ h · ω(1H , 1H ) = ω(h, 1H ). C 1 Finally, Therefore, h ·eω(1H , 1H ) = h · (1H · 1A) = h · 1A =eω(h, 1H ). let us verify that eω is cohomologous to ω, that is, there exists φ ∈ par(H, A) such that eω ∗ ω−1 = δ1φ. Indeed, on one hand, note that eω(h(1), k(1))ω−1(h(2), k(2)) = (h · (k · 1A))(ω(1H , 1H ))−1. On the other hand, δφ(h, k) = (h(1) · φ(k(1)))φ−1(h(2)k(2))φ(h(3)). Then, if we define φ(k) = (k · 1A)(ω(1H , 1H ))−1, we have δφ(h, k) = h(1) · ((k(1) · 1A)(ω(1H , 1H ))−1)(h(2)k(2) · 1A)(ω(1H , 1H )) (h(3) · 1A)(ω(1H , 1H ))−1 = (h(1) · (k(1) · 1A))(ω(1H , 1H ))−1(h(2)k(2) · 1A)(h(3) · 1A) = (h(1) · (k(1) · 1A))(ω(1H , 1H ))−1 This concludes our proof. (cid:4) 5.2. The Hopf algebroid structure of the partial crossed product. We saw that the cohomology theory for a co-commutative Hopf algebra H acting partially over a commutative algebra A is equivalent to the cohomology theory of the same Hopf algebra H acting on a commutative and co-commutative Hopf algebra eA whose base ring is the commutative algebra E(A). This replacement gives us a deeper understanding about the structure of crossed products. In fact, we shall see that the crossed product has a structure of a Hopf algebroid over the base algebra E(A). Let us recall briefly the definition of a Hopf algebroid, for a detailed presentation, se the reference [10]. Definition 5.9. [10] Given a k algebra A, a left (resp. right) bialgebroid over A is given by the data (H, A, sl, tl, ∆l, ǫl) (resp. (H, A, sr, tr, ∆r, ǫr)) such that: (1) H is a k algebra. (2) The map sl (resp. sr) is a morphism of algebras between A and H, and the map tl (resp. tr) is an anti-morphism of algebras between A and H. Their images commute, that is, for every a, b ∈ A we have sl(a)tl(b) = tl(b)sl(a) (resp. sr(a)tr(b) = tr(b)sr(a)). By the maps sl, tl (resp. sr, tr) the algebra H inherits a structure of A bimodule given by a ⊲ h ⊳ b = sl(a)tl(b)h (resp. a ◮ h ◭ b = hsr(b)tr(a)). COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 27 (3) The triple (H, ∆l, ǫl) (resp. (H, ∆r, ǫr)) is an A coring relative to the structure of A bimodule defined by sl and tl (resp. sr, and tr). (4) The image of ∆l (resp. ∆r) lies in the Takeuchi subalgebra HA × H = {Xi H ×A H = {Xi hi ⊗ ki ∈ H ⊗A,⊲ ⊳ H Xi hi ⊗ ki ∈ H ⊗A,◮ ◭ H Xi respectively, and it is an algebra morphism. hitl(a) ⊗ ki =Xi sr(a)hi ⊗ ki =Xi hi ⊗ kisl(a) ∀a ∈ A}, hi ⊗ tr(a)ki ∀a ∈ A}, (5) For every h, k ∈ H, we have ǫl(hk) = ǫl(hsl(ǫl(k))) = ǫl(htl(ǫl(k))), respec- tively, ǫr(hk) = ǫr(sr(ǫr(h))k) = ǫr(tr(ǫr(h))k). Given two anti-isomorphic algebras Al and Ar (ie, Al r ), a left Al bialgebroid (H, Al, sl, tl, ∆l, ǫl) and a right Ar bialgebroid (H, Ar, sr, tr, ∆r, ǫr), a Hopf algebroid structure on H is given by an antipode, that is, an algebra anti-homomorphism S : H → H such that ∼= Aop (i) sl ◦ ǫl ◦ tr = tr, tl ◦ ǫl ◦ sr = sr, sr ◦ ǫr ◦ tl = tl and tr ◦ ǫr ◦ sl = sl; (ii) (∆l ⊗Ar I) ◦ ∆r = (I ⊗Al ∆r) ◦ ∆l and (I ⊗Ar ∆l) ◦ ∆r = (∆r ⊗Al I) ◦ ∆l; (iii) S(tl(a)htr(b′)) = sr(b′)S(h)sl(a), for all a ∈ Al, b′ ∈ Ar and h ∈ H; (iv) µH ◦ (S ⊗Al I) ◦ ∆l = sr ◦ ǫr and µH ◦ (I ⊗Ar S) ◦ ∆r = s ◦ ǫl. In our case, both algebras, Al and Ar, coincide with the commutative algebra the previous definition. Theorem 5.10. Let H be a co-commutative Hopf algebra and A be a commutative partial H-module algebra. Consider the commutative and co-commutative Hopf al- E(A) and the crossed product eA#ωH will play the role of the Hopf algebroid H of gebra eA, constructed in Theorem 4.5, over the commutative algebra E(A), which is also a partial H-module algebra. Then, the crossed product eA#ωH, in which ω is par(H, eA), is a Hopf algebroid over the base a partial 2-cocycle in H 2 algebra E(A). Proof: The source and target maps, both left and right, are defined by the restriction par(H, A) = H 2 to E(A) of the canonical inclusion of eA into the crossed product . sl, tl, sr, tr : E(A) −→ eA#ωH 7→ a#1H a We have already seen that this inclusion is an algebra map and by the commu- Note that, even though the left and right sources and targets are equal, their asso- ciated bimodule structures are different nonetheless. Indeed, for a, a′ ∈ E(A) and tativity of eA, the images of source and target maps commute among themselves. b#h ∈ eA#ωH we have a ⊲ (b#h) ⊳ a′ = (a#1H )(a′#1H )(b#h) = aa′b#h, and a ◮ (b#h) ◭ a′ = (b#h)(a#1H )(a′#1H) = b(h(1) · aa′)#h(2), 28 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA The left and right comultiplication maps are defined, respectively, as and ∆l : ∆r : 7→ a(1)#h(1) ⊗E(A),⊲ ⊳ a(2)#h(2) a#h eA#ωH −→ eA#ωH ⊗ eA#ωH eA#H −→ eA#H ⊗ eA#H a#h 7→ a(1)#h(1) ⊗E(A),◮ ◭ a(2)#h(2) , in which the tensor product ⊗E(A),⊲ ⊳ (resp. ⊗E(A),◮ ◭) is balanced with respect to the E(A)-bimodule structure implemented by sl, tl (resp. sr, tr). It is easy to see that (∆l ⊗E(A),◮ ◭ I) ◦ ∆r = (I ⊗E(A),⊲ ⊳ ∆r) ◦ ∆l (I ⊗E(A),◮ ◭ ∆l) ◦ ∆r = (∆r ⊗E(A),⊲ ⊳ I) ◦ ∆l. One can see also that, for any a ∈ E(A), ∆l(sl(a)) = sl(a) ⊗ (1A#1H), and ∆l(tl(a)) = (1A#1H) ⊗ tl(a). This is because that any element of E(A) is a linear combination of monomials of the form (h1 · 1A) . . . (hn · 1A), for h1, . . . , hn ∈ H, and then ∆l(sl((h1 · 1A) . . . (hn · 1A))) (1) · 1A) . . . (hn (1) · 1A) . . . (hn (1) · 1A) . . . (hn (1) · 1A) . . . (hn (1) · 1A) . . . (hn (2) · 1A) . . . (hn (2) · 1A) . . . (hn (1) · 1A) . . . (hn (1) · 1A)#1H ⊗ (h1 (1) · 1A)#1H ⊗ ((h1 (1) · 1A)#1H ⊗ sl((h1 (1) · 1A)#1H ⊗ ((h1 (1) · 1A)#1H ) ⊳ ((h1 (2) · 1A))((h1 (2) · 1A)#1H )((h1 (1) · 1A)(h1 = (h1 = (h1 = (h1 = (h1 = ((h1 = tl((h1 = ((h1 = ((h1 = (h1 · 1A) . . . (hn · 1A)#1H ⊗ 1A#1H = sl((h1 · 1A) . . . (hn · 1A)) ⊗ 1A#1H. (2) · 1A) . . . (hn (2) · 1A) . . . (hn (2) · 1A)#1H (2) · 1A)#1H )(1A#1H ) (2) · 1A) . . . (hn (2) · 1A) . . . (hn (2) · 1A) . . . (hn (2) · 1A))(1A#1H ) (2) · 1A)) ⊲ (1A#1H) (2) · 1A)) ⊗ 1A#1H (1) · 1A)#1H ) ⊗ 1A#1H (1) · 1A) . . . (hn (1) · 1A) . . . (hn (1) · 1A)#1H ) ⊗ 1A#1H (2) · 1A) . . . (hn (2) · 1A)#1H ) ⊗ 1A#1H From this we deduce that ∆l is a morphism of E(A)-bimodules with the bimodule structure given by ⊲ and ⊳. The same occurs for ∆r, that is, for any a ∈ E(A), we have ∆r(tr(a)) = tr(a) ⊗ (1A#1H ), and ∆r(sr(a)) = (1A#1H) ⊗ sr(a), and then ∆r is a morphism of E(A)-bimodules, with the bimodule structure given by ◮ and ◭. COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 29 The image of the left comultiplication ∆l lies in the left Takeuchi product eA#ωH E(A) × eA#ωH. Indeed, for a#h ∈ eA#ωH and b ∈ E(A), we have (a(1)#h(1))tl(b) ⊗ a(2)#h(2) = (a(1)#h(1))(b#1H ) ⊗ a(2)#h(2) = a(1)(h(1) · b)ω(h(2), 1H )#h(3) ⊗ a(2)#h(4) = a(1)(h(1) · b)(h(2) · 1A)#h(3) ⊗ a(2)#h(4) = (h(1) · b)a(1)#h(2) ⊗ a(2)#h(3) = ((h(1) · b)#1H )(a(1)#h(2)) ⊗ a(2)#h(3) = tl(h(1) · b)(a(1)#h(2)) ⊗ a(2)#h(3) = a(1)#h(2) ⊗ sl(h(1) · b)(a(2)#h(3)) = a(1)#h(2) ⊗ ((h(1) · b)#1H )(a(2)#h(3)) = a(1)#h(1) ⊗ a(2)(h(2) · b)#h(3) = a(1)#h(1) ⊗ (a(2)#h(2))(b#1H ) = a(1)#h(1) ⊗ (a(2)#h(2))sl(b). Moreover, ∆l is a morphism of algebras. Take any a#h, b#l ∈ eA#ωH, then ∆l((a#h)(b#l)) = ∆l(a(h(1) · b)ω(h(2), l(1))#h(3)l(2)) = (a(h(1) · b)ω(h(2), l(1)))(1)#(h(3)l(2))(1) ⊗ (a(h(1) · b)ω(h(2), l(1)))(2)#(h(3)l(2))(2) = a(1)(h(1) · b(1))ω(h(3), l(1))#h(5)l(3) ⊗ a(2)(h(2) · b(2))ω(h(4), l(2))#h(6)l(4) On the other hand, ∆l(a#h)∆l(b#l) = (a(1)#h(1) ⊗ a(2)#h(2))(b(1)#l(1) ⊗ b(2)#l(2)) = (a(1)#h(1))(b(1)#l(1)) ⊗ (a(2)#h(2))(b(2)#l(2)) = a(1)(h(1) · b(1))ω(h(2), l(1))#h(3)l(2) ⊗ a(2)(h(4) · b(2))ω(h(5), l(3))#h(6)l(4). The equality follows from the co-commutativity of H. Analogously, one can prove that the image of the right comultiplication lies in the right Takeuchi product eA#ωH ×E(A) eA#ωH, and it is an algebra morphism. The left and right counits are defined, respectively, as 7→ ǫl(a#h) := ǫ eA(a)(h · 1A) 7→ ǫr(a#h) := S(h) · ǫ eA(a) . ǫl : ǫr : and a#h eA#ωH −→ E(A) eA#ωH −→ E(A) Take a, a′ ∈ E(A) and b#h ∈ eA#ωH, then a#h and First, both are morphisms of E(A)-bimodules, with their respectives structures. ǫl(a ⊲ (b#h) ⊳ a′) = ǫl((aa′b#h)) = ǫ eA(aa′)ǫ eA(b)(h · 1A) = aǫ eA(b)(h · 1A)a′ = aǫl(b#h)a′, ǫl(a ◮ (b#h) ◭ a′) = ǫl(b(h(1) · aa′)#h(2)) = S(h(2)) · (ǫ eA(b)ǫ eA(h(1) · aa′)) = (S(h(3)) · ǫ eA(b))(S(h(2)) · (h(1) · aa′)) = (S(h(3)) · ǫ eA(b))(S(h(1))h(2) · aa′) = (S(h) · b)aa′ = aǫr(b#h)a′. One can easily verify the compatibility relations with the left and right counits with the left and right source and targets, that is, sl ◦ ǫl ◦ tr = tr, tl ◦ ǫl ◦ sr = sr, sr ◦ ǫr ◦ tl = tl and tr ◦ ǫr ◦ sl = sl. their respectives bimodule structures. we have already seen that ∆l, ǫl, ∆r and ǫr Let us verify that (eA#ωH, ∆l, ǫl) and (eA#ωH, ∆r, ǫr) are corings over E(A) with 30 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA are morphisms of E(A) bimodules. It is easy to see that the left and right comultipli- cations are coassociative. It remains to verify the counit axiom for both structures. Take a#h ∈ eA#ωH ,then (ǫl ⊗E(A),⊲⊳ I) ◦ ∆l(a#h) = ǫl(a(1)#h(1)) ⊲ (a(2)#h(2)) = (ǫ eA(a(1))(h(1) · 1A)#1H )(a(2)#h(2)) = ǫ eA(a(1))(a(2)(h(1) · 1A)#h(2)) = a(h(1) · 1A)#h(2)) = a#h. Using the cocommutativity of H and the commutativity of A it is also easy to see that (I ⊗E(A),⊲⊳ ǫl) ◦ ∆l(a#h) = a#h. Therefore, (eA#ωH, ∆l, ǫl) is a coring over E(A). For the right structure, for a#h ∈ eA#ωH, we have (ǫr ⊗E(A),◮◭ I) ◦ ∆r(a#h) = ǫr(a(1)#h(1)) ◮ (a(2)#h(2)) = (a(2)#h(2))((S(h(1)) · ǫ eA(a(1)))#1H ) = (a(2)#h(1))((S(h(2)) · ǫ eA(a(1)))#1H ) = a(2)(h(1) · (S(h(4)) · ǫ eA(a(1))))ω(h(2), 1H )#h(3) = a(2)(h(1) · (S(h(4))· ǫ eA(a(1))))(h(2) · 1A)#h(3) = a(2)(h(1)S(h(4)) · ǫ eA(a(1)))(h(2) · 1A)#h(3)=a(2)(h(1)S(h(2))· ǫ eA(a(1)))(h(3) · 1A)#h(4) = a(2)ǫ eA(a(1))(h(1) · 1A)#h(2) = a(h(1) · 1A)#h(2)) = a#h. Analogously, one can prove that (I ⊗E(A),◮◭ ǫr) ◦ ∆r(a#h) = a#h. Therefore, (eA#ωH, ∆r, ǫr) is a coring over E(A) Let us verify now that ǫl((a#h)(b#k)) = ǫl((a#h)sl(ǫl(b#k))) = ǫl((a#h)tl(ǫl(b#k))), and ǫr((a#h)(b#k)) = ǫr(sr(ǫr(a#h))(b#k)) = ǫr(tr(ǫr(a#h))(b#k)), for any a#h, b#k ∈ eA#ωH. For the left counit, on one hand, ǫl((a#h)(b#k)) = ǫl(a(h(1) · b)ω(h(2), k(1))#h(3)k(2)) = ǫ eA(a)ǫ eA(h(1) · b)ǫ eA(ω(h(2), k(1)))(h(3)k(2) · 1A) = ǫ eA(a)(h(1) · ǫ eA(b))(h(2) · (k(1) · 1A))(h(3)k(2) · 1A) = ǫ eA(a)(h(1) · ǫ eA(b))(h(2) · (k · 1A)) = ǫ eA(a)(h · (ǫ eA(b)(k · 1A))). On the other hand ǫl((a#h)sl(ǫl(b#k))) = ǫl((a#h)sl(ǫ eA(b)(k · 1A))) =ǫl((a#h)(ǫ eA(b)(k · 1A)#1H )) = ǫl(a(h(1) · (ǫ eA(b)(k · 1A)))ω(h(2), 1H )#h(3)) =ǫl(a(h(1) · (ǫ eA(b)(k · 1A)))(h(2) · 1H )#h(3)) = ǫ eA(a(h(1) · (ǫ eA(b)(k · 1A))))(h(2) · 1A) =ǫ eA(a)(h(1) · (ǫ eA(b)(k · 1A)))(h(2) · 1A) = ǫ eA(a)(h · (ǫ eA(b)(k · 1A))). COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 31 Therefore, ǫl((a#h)(b#k)) = ǫl((a#h)sl(ǫl(b#k))) = ǫl((a#h)tl(ǫl(b#k))). For the right counit, on one hand ǫr((a#h)(b#k)) = ǫr(a(h(1) · b)ω(h(2), k(1))#h(3)k(2)) = S(h(3)k(2)) · ǫ eA(a(h(1) · b)ω(h(2), k(1))) = S(h(3)k(2)) · (ǫ eA(a)(h(1) · ǫ eA(b))(h(2) · (k(1) · 1A)) = (S(h(5)k(4)) · ǫ eA(a))(S(h(4)k(3)) · (h(1) · ǫ eA(b)))(S(h(3)k(2)) · (h(2) · (k(1) · 1A))) = (S(h(5)k(4)) · ǫ eA(a))(S(h(4)k(3))h(1) · ǫ eA(b))(S(h(3)k(2))h(2) · (k(1) · 1A)) = (S(k(4))S(h(5)) · ǫ eA(a))(S(k(3))S(h(4))h(1) · ǫ eA(b))(S(k(2))S(h(3))h(2) · (k(1) · 1A)) = (S(k(4))S(h(5)) · ǫ eA(a))(S(k(3))S(h(3))h(4) · ǫ eA(b))(S(k(2))S(h(1))h(2) · (k(1) · 1A)) = (S(k(4))S(h) · ǫ eA(a))(S(k(3)) · ǫ eA(b))(S(k(2)) · (k(1) · 1A)) = (S(k(4))S(h) · ǫ eA(a))(S(k(3)) · ǫ eA(b))(S(k(2))k(1) · 1A)) = (S(k(4))S(h) · ǫ eA(a))(S(k(3)) · ǫ eA(b))(S(k(1))k(2) · 1A)) = (S(k(2))S(h) · ǫ eA(a))(S(k(1)) · ǫ eA(b)) = (S(k(2)) · (S(h) · ǫ eA(a)))(S(k(1)) · ǫ eA(b)) = S(k) · ((S(h) · ǫ eA(a))ǫ eA(b)). On the other hand, ǫr(sr(ǫr(a#h))(b#k)) = ǫr((S(h) · ǫ eA(a))#1H )(b#k)) = ǫr((S(h) · ǫ eA(a))b#k) = S(k) · ((S(h) · ǫ eA(a))ǫ eA(b)). Therefore, ǫr((a#h)(b#k)) = ǫr(sr(ǫr(a#h))(b#k)) = ǫr(tr(ǫr(a#h))(b#k)). Finally, we define the antipode as S : eA#ωH −→ eA#ωH a#h 7→ (SH (h(3)) · S eA(a))ω−1(SH(h(2)), h(4))#SH(h(1)) Take b, c ∈ E(A) and a#h ∈ eA#ωH, then, one can prove that S(t(b)(a#h)t(c)) = s(c)S(a#h)s(b). Indeed, S(tl(b)(a#h)tr(c)) = S((b#1H )(a#h)(c#1H )) = S((ba#h)(c#1H )) = S(ba(h(1) · c)ω(h(2), 1H )#h(3)) = S(ba(h(1) · c)#h(2)) = (S(h(4)) · S eA(ba(h(1) · c)))ω−1(S(h(3)), h(5))#S(h(2)) = (S(h(4)) · ((h(1) · c)S eA(a)b))ω−1(S(h(3)), h(5))#S(h(2)) (∗) in which the equality (∗) is valid because S eA restricted to the base algebra E(A) is equal to the identity. On the other hand. 32 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA sr(c)S(a#h)sl(b) = = (c#1H )((S(h(3)) · S eA(a))ω−1(S(h(2)), h(4))#S(h(1)))(b#1H ) = (c(S(h(4)) · S eA(a))ω−1(S(h(3)), h(5))ω(1H , S(h(2)))#S(h(1)))(b#1H ) = (c(S(h(3)) · S eA(a))ω−1(S(h(2)), h(4))#S(h(1)))(b#1H ) = c(S(h(4)) · S eA(a))ω−1(S(h(3)), h(5))(S(h(2)) · b)#S(h(1)) = (S(h(6)) · b)(S(h(5)) · S eA(a))(S(h(3))h(4) · c)ω−1(S(h(2)), h(7))#S(h(1)) = (S(h(6)) · b)(S(h(5)) · S eA(a))(S(h(3)) · (h(4) · c))ω−1(S(h(2)), h(7))#S(h(1)) = (S(h(6)) · b)(S(h(5)) · S eA(a))(S(h(4)) · (h(1) · c))ω−1(S(h(3)), h(7))#S(h(2)) = (S(h(4)) · ((h(1) · c)S eA(a)b))ω−1(S(h(3)), h(5))#S(h(2)). It remains to check that and µ(S ⊗E(A),⊲⊳ Id) ◦ ∆l = sr ◦ ǫr µ(Id ⊗E(A),◮◭ S) ◦ ∆r = sl ◦ ǫl. µ(S ⊗E(A),⊲⊳ Id) ◦ ∆l(a#h) = S(a(1)#h(1))(a(2)#h(2)) Take a#h ∈ eA#ωH, then = (S(h(3)) · S eA(a(1))ω−1(S(h(2)), h(4))#S(h(1)))(a(2)#h(5)) = (S(h(5)) · S eA(a(1)))ω−1(S(h(4)), h(6))(S(h(3)) · a(2))ω(S(h(2)), h(7))#S(h(1))h(8) = (S(h(4)) · S eA(a(1)))(S(h(3)) · a(2))ω−1(S(h(2)), h(5))ω(S(h(1)), h(6))#S(h(7))h(8) = (S(h(3)) · S eA(a(1)))(S(h(2)) · a(2))(S(h(1)) · (h(4) · 1A))#1H = (S(h(3)) · S eA(a(1)))(S(h(2)) · a(2))(S(h(1))h(4) · 1A)#1H = (S(h(2)) · S eA(a(1)))(S(h(1)) · a(2))#1H = (S(h) · (S eA(a(1))a(2)))#1H = sr(S(h) · ǫ eA(a)) = s ◦ ǫr(a#h). and µ(Id ⊗E(A),◮◭ S) ◦ ∆r(a#h) = (a(1)#h(1))S(a(2)#h(2)) =(a(1)#h(1))(S(h(4)) · S eA(a(2))ω−1(S(h(3)), h(5))#S(h(2))) =(a(1)(h(1) · (S(h(7)) · S eA(a(2))ω−1(S(h(6)), h(8))))ω(h(2), S(h(5)))#h(3)S(h(4)) =(a(1)(h(1) · (S(h(6)) · S eA(a(2))))(h(2) · ω−1(S(h(5)), h(7)))ω(h(3), S(h(4)))#1H =(a(1)(h(1) · (S(h(2)) · S eA(a(2))))(h(3) · ω−1(S(h(6)), h(7)))ω(h(4), S(h(5)))#1H =(a(1)(h(1)S(h(2)) · S eA(a(2))))(h(3) · ω−1(S(h(6)), h(7)))ω(h(4), S(h(5)))#1H =a(1)S eA(a(2))(h(1) · ω−1(S(h(4)), h(5)))ω(h(2), S(h(3)))#1H COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 33 (∗) =ǫ eA(a)ω−1(h(1)S(h(8)), h(9))ω(h(2), S(h(7))h(10))ω−1(h(3), S(h(6)))ω(h(4), S(h(5)))#1H =ǫ eA(a)ω−1(h(1)S(h(6)), h(7))ω(h(2), S(h(5))h(8))(h(3) · (S(h(4)) · 1A))#1H =ǫ eA(a)ω−1(h(1)S(h(5)), h(6))ω(h(2), S(h(4))h(7))(h(3) · 1A)#1H =ǫ eA(a)ω−1(h(1)S(h(2)), h(7))ω(h(3), S(h(5))h(6))(h(4) · 1A)#1H =ǫ eA(a)ω−1(1H , h(3))ω(h(1), 1H )(h(2) · 1A)#1H =ǫ eA(a)(h · 1A)#1H = sl ◦ ǫl(a#h). The equality (∗) we used that, for any h, k, l ∈ H, we have h · ω−1(k, l) = ω−1(h(1)k(1), l(1))ω(h(2), k(2)l(2))ω−1(h(3), k(3)). This folllows easily from h · (ω(k(1), l(1))ω−1(k(2), l(2)) = h · (k · (l · 1A)) and h · ω(k, l) = ω(h(1), k(1))ω−1(h(2), k(2)l(1))ω(h(3)k(3), l(2)), which, in its turn is an immediate consequence of the 2-cocycle identity. over E(A). Therefore, (eA#ωH, sl, tl, sr, tr, ∆l, ∆r, ǫl, ǫr, S) is a strucutre of a Hopf algebroid (cid:4) 6. Partially cleft extensions and cleft extensions by Hopf algebroids In [5], the authors introduced the notion of a partially cleft extension of an algebra A by e Hopf algebra H. Using the Hopf algebroid structure of the crossed product, one can rethink partially cleft extensions of commutative algebras by co-commutative Hopf algebras in a broader scenario, namely, the theory of cleft extensions of algebras by Hopf algebroids developed by G. Böhm and T. Brzezinski [11]. Definition 6.1. [5] Let H be a Hopf algebra, and A ⊂ B be an H-extension, that is B is a right H-comodule algebra and A = BcoH. The extension A ⊂ B is partially cleft if there exists a pair of linear maps γ, γ : H → B such that: (i) γ(1H ) = 1B; (ii) The following diagrams are commutative γ H ∆ B ρ γ H ∆cop B ρ H ⊗ H / B ⊗ H γ⊗IdH H ⊗ H / B ⊗ H γ⊗S (iii) (γ ∗ γ) ◦ µ is a central element in the convolution algebra Homk(H ⊗ H, B), in which µ : H ⊗ H → H is the multiplication in H and (γ ∗ γ)(h) commutes with every element of A, for each h ∈ H, then: and, for all b ∈ B, h, l ∈ H, if we write eh = (γ ∗ γ)(h) and eeh = (γ ∗ γ)(h), (iv) P b(0)γ(b(1))γ(b(2)) = b; (v) γ(h)el =P eh(1)lγ(h(2)); (vi) γ(l)eeh =Peehl(1)γ(l(2)); (vii) P γ(hl(1))eel(2) =P eh(1)γ(h(2)l). / / / / / / 34 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA Partially cleft extensions are related to partial crossed products, as one can see in the next two results. Proposition 6.2. [5] If (A, ·, (ω, ω−1)) is a symmetric twisted partial H- module algebra with a 2-cocycle ω, then,A ⊂ A#ωH is a partially cleft H-extension. For the crossed product A#ωH, the cleaving maps γ, γ : H → A#ωH are given by (15) γ(h) = 1A#h, and γ(h) = ω−1(S(h(2)), h(3))#S(h(1)). Theorem 6.3. [5] Let B be H-comodule algebra and A = BcoH. Then the H- extension A ⊂ B is partially cleft if, and only if there is a symmetric twisted partial action · : H ⊗ A → A with a 2-cocycle ω : H ⊗ H → A such that B is isomorphic to the partial crossed product A#ωH. In the case of a co-commutative Hopf algebra H acting partially over a commutative algebra A, we have already seen that we can replace A by a commutative and co- note that the H-comodule structure on both crossed products is the same, namely it is interesting to see whether one can replace the crossed product A#ωH by the commutative Hopf algebra eA with the same cohomology theory, moreover, the crossed product eA#ωH has a structure of Hopf algebroid over the base algebra E(A). Then crossed product eA#ωH in the analysis of cleft extensions by the Hopf algebra H. First ρ(a#h) = a#h(1) ⊗ h(2). Furthermore (A#ωH)coH ∼= A and (eA#ωH)coH ∼= eA, then in eA#ωH. crossed product eA#ωH can be viewed as a cleft extension of eA by H. both crossed products are H-extensions of their respectives algebras of coinvariants. Finally, the cleaving maps γ, γ : H → A#ωH, given by (15), take their values actually In what follows, we shall see that there exists a Hopf algebroid H such that the Definition 6.4. [11] Let (H, L, R, sL, tL, sR, tR, ∆L, ∆R, ǫL, ǫR, S) be a Hopf alge- broid and A be a right H-comodule algebra. Denote by ηR(r) = r · 1A = 1A · r the unit map of the corresponding R-ring structure of A. Let B be the subalgebra of HR-coinvariants in A. The extension B ⊂ A is called H-cleft if (a) A is an L-ring (with unit ηL : L → A) and B is an L-subring of A; (b) there exists a convolution invertible left L-linear right H-colinear morphism γ : H → A. A map γ satisfying condition (b) is called a cleaving map. Remark 6.5. Some small remarks have to be made about this definition. (1) First is that the structure of right H-comodule algebra on A is relative to the base ring R, that is ρ : A → A ⊗R H is a right R − R-bilinear map in the sense that ρ(ηR(r)aηR(s)) = a(0) ⊗R sR(r)a(1)sR(s), for every a ∈ A and r ∈ R. (2) The map γ : H → A being H-colinear implies that it is right R-linear in the sense that γ(hsR(r)) = γ(h)ηR(r) and left R-linear in the sense that γ(sR(r)h) = ηR(r)γ(h), for any a ∈ A and r ∈ R. (3) The notion of a convolution invertible map γ : H → A, in which A is both L-ring and R-ring, means that there is a unique γ : H → A such that [11] µA ◦ (γ ⊗R γ) ◦ ∆R = ηL ◦ ǫL µA ◦ (γ ⊗L γ) ◦ ∆L = ηR ◦ ǫR. COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 35 (4) The left L-linearity of the map γ in condition (b) of Definition 6.4 means, that the cleaving map satisfies γ(sL(l)h) = ηL(l)γ(h), for any a ∈ A and l ∈ L. In our case, for a co-commutative Hopf algebra H acting partially upon a commu- the following theorem. left and right structures will coalesce. The Hopf algebroid is given by the partial smash product E(A)#H. For a proof that this partial smash product is in fact a Hopf algebroid over E(A), see reference [9], Theorem 3.5. The extension to be con- tative and co-commutative Hopf algebra eA, the base algebras L and R will coincide with the commutative subalgebra E(A) of eA and then many distinctions about the sidered is the previous defined partially H-cleft extension eA ⊂ eA#ωH. Then we have mutative and co-commutative Hopf algebra eA and let ω be a partial 2 cocycle in par(H, eA). Then the partial crossed product eA#ωH is a right H = E(A)#H-module algebra with eA ∼= (eA#ωH)coH. Moreover, the extension eA ⊂ eA#ωH is H-cleft in the Theorem 6.6. Let H be a co-commutative Hopf algebra acting partially on a com- sense of Definition 6.4. H 2 Proof: First, define the linear map a#h eA#ωH −→ eA#ωH ⊗E(A),◮◭ E(A)#H eρ : Note that the expression of eρ(a#h) can also be written as eρ(a#h) = a#h(1) ⊗ 1E(A)#h(2) = a#h(1) ⊗ 1A#h(2), 7→ a#h(1) ⊗ (h(2) · 1A)#h(3) that is because a#h(1) ⊗ (h(2) · 1A)#h(3) = a#h(1) ⊗ (1A#h(2))(1A#1H) = a#h(1) ⊗ (1A#h(2)) ◭ 1A = a#h(1) ⊗ (1A#h(2)). right comultiplication in H, given by It is easy to see that (eρ ⊗E(A),◮◭ Id) ◦eρ = (Id ⊗E(A),◮◭e∆r) ◦eρ, in which e∆r is the Also, one can check that the map eρ is E(A)-bilinear. Indeed, for a#h ∈ eA#ωH e∆r(r#h) = r#h(1) ⊗ 1A#h(2), eρ((r#1H )(a#h)(s#1H )) =eρ(ra(h(1) · s)#h(2)) ∀r ∈ E(A), ∀h ∈ H. and r, s ∈ E(A) then = ra(h(1) · s)#h(2) ⊗ 1A#h(3) = (ra#h(1)) ◭ s ⊗ 1A#h(2) = ra#h(1) ⊗ s ◮ (1A#h(2)) = ra#h(1) ⊗ (1A#h(2))(s#1H) = a(h(1)S(h(2)) · r)#h(3) ⊗ (1A#h(4))(s#1H ) = a(h(1) · (S(h(3)) · r))#h(2) ⊗ (1A#h(4))(s#1H) = (a#h(1)) ◭ (S(h(2)) · r) ⊗ (1A#h(3))(s#1H ) = a#h(1) ⊗ (S(h(2)) · r) ◮ (1A#h(3))(s#1H ) = a#h(1) ⊗ (1A#h(2))((S(h(3)) · r)#1H )(s#1H) = a#h(1) ⊗ ((h(2) · (S(h(4)) · r))#h(3))(s#1H ) = a#h(1) ⊗ ((h(2)S(h(4)) · r)#h(3))(s#1H ) = a#h(1) ⊗ (r#h(2))(s#1H ) = a#h(1) ⊗ (r#1H )(1A#h(2))(s#1H ) 36 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA Denote by eǫr : E(A)#H → E(A) the right counit of the partial smash product E(A)#H, given byeǫr(a#h) = S(h) · 1A. Then, we have (Id ⊗E(A),◮◭eǫr) ◦eρ(a#h) = (a#h(1)) ◭eǫr(1A#h(2)) = (a#h(1))((S(h(2)) · 1A)#1H ) = a(h(1) · (S(h(3)) · 1A))#h(2) = a(h(1)S(h(2)) · 1A)#h(3) = a#h. Finally, for a#h, b#k ∈ eA#ωH, we have eρ((a#h)(b#k)) = eρ(a(h(1) · b)ω(h(2), k(1))#h(3)k(2)) = a(h(1) · b)ω(h(2), k(1))#h(3)k(2) ⊗ 1A#h(4)k(3), on the other hand, eρ(a#h)eρ(b#k) = (a#h(1))(b#k(1)) ⊗ (1A#h(2))(1A#k(2)) = a(h(1) · b)ω(h(2), k(1))#h(3)k(2) ⊗ (h(4) · 1A)#h(5)k(3) = a(h(1) · b)ω(h(2), k(1))#h(3)k(2) ⊗ (h(4)k(3)S(k(4)) · 1A)#h(5)k(5) = a(h(1) · b)ω(h(2), k(1))#h(3)k(2) ⊗ (h(4)k(3) · (S(k(4)) · 1A))#h(5)k(4) = a(h(1) · b)ω(h(2), k(1))#h(3)k(2) ⊗ (S(k(4)) · 1A) ◮ (1A#h(4)k(3)) = (a(h(1) · b)ω(h(2), k(1))#h(3)k(2)) ◭ (S(k(4)) · 1A) ⊗ 1A#h(4)k(3) = a(h(1) · b)ω(h(2), k(1))(h(3)k(2) · (S(k(5)) · 1A))#h(4)k(3) ⊗ 1A#h(5)k(4) = (a(h(1) · b)ω(h(2), k(1))(h(3)k(2)S(k(3)) · 1A)#h(4)k(4)) ⊗ (1A#h(5)k(5) = a(h(1) · b)ω(h(2), k(1))(h(3) · 1A)#h(4)k(2) ⊗ 1A#h(5)k(3) = a(h(1) · b)ω(h(2), k(1))#h(3)k(2) ⊗ 1A#h(4)k(3) Applying Id ⊗ ǫH ⊗ Id ⊗ Id to this ai#hi(1) ⊗ 1A#hi(2) =Xi ai#hi ⊗ 1A#1H ∈ eA ⊗ H ⊗E(A) E(A) ⊗ H. Therefore, the crossed product eA#ωH is a right H-comodule algebra. It is obvious that i(eA) ⊆ (eA#ωH)coH, now takePi ai#hi ∈ (eA#ωH)coH, then Xi eA ⊗E(A) E(A) ∼= eA, we obtain Xi Therefore i(eA) = (eA#ωH)coH. In order to see that the H-extension is cleft, one needs only to define the cleaving map, once item (a) of Definition 6.4 is automatically satisfied, since L = R = E(A). Define the maps ai#hi =Xi identity and doing the identification aiǫH(hi)#1H . and eγ : E(A)#H → eA#ωH 7→ r#h r#h eγ : E(A)#H → eA#ωH r#h 7→ rω−1(S(h(2)), h(3))#S(h(1)) the image because the first is in the partial smash product, while the second lies in Note that a#h in the domain of eγ means something quite different from a#h in COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 37 r ∈ E(A) and a#h inE(A)#H, we have the partial crossed product. It is easy to see that eγ is left E(A)-linear. Indeed, for eγ(sl(r)(a#h)) = eγ((r#1H )(a#h)) =eγ(ra#h) = ra#h = (r#1H)(a#h). Also, the cleaving map is a morphism of right H-comodules. Consider a#h ∈ E(A)#H, then the sense that Consider a#h ∈ E(A)#H, then eρ ◦eγ(a#h) = eρ(a#h) = a#h(1) ⊗ 1A#h(2) Finally, let us check that the mapseγ andeγ are mutually inverse by convolution in = eγ(a#h(1) ⊗ 1A#h(2) = (eγ ⊗E(A),◮◭ Id) ◦eρ(a#h). µ ◦ (eγ ⊗E(A),◮◭eγ) ◦e∆r = i ◦eǫl µ ◦ (eγ ⊗E(A),⊲⊳eγ) ◦e∆l = i ◦eǫr. µ ◦ (γ ⊗E(A),◮◭ γ) ◦e∆r(a#h) =eγ(a#h(1))eγ(1A#h(2)) = (a#h(1))(ω−1(S(h(3)), h(4))#S(h(2))) = a(h(1) · ω−1(S(h(6)), h(7)))ω(h(2), S(h(5)))#h(3)S(h(4)) = a(h(1) · ω−1(S(h(4)), h(5)))ω(h(2), S(h(3)))#1H = aω−1(h(1)S(h(9)), h(10))ω(h(2), S(h(8))h(11))ω−1(h(4), S(h(7)))ω(h(5), S(h(6)))#1H = aω−1(h(1)S(h(6)), h(7))ω(h(2), S(h(5))h(8))(h(3) · (S(h(4)) · 1A))#1H = aω−1(h(1)S(h(2)), h(7))ω(h(3), S(h(5))h(6))(h(4) · 1A)#1H = a(h · 1A)#1H = i(eǫl(a#h)). Also, we have µ ◦ (eγ ⊗E(A),⊲⊳eγ) ◦e∆l(a#h) = µ ◦ (eγ ⊗E(A),⊲⊳eγ)(a#h(1) ⊗ 1A#h(2)) = µ ◦ (eγ ⊗E(A),⊲⊳eγ)((1A#h(1)) ⊳ a ⊗ 1A#h(2)) = µ ◦ (eγ ⊗E(A),⊲⊳eγ)(1A#h(1) ⊗ a ⊲ (1A#h(2))) = µ ◦ (eγ ⊗E(A),⊲⊳eγ)(1A#h(1) ⊗ a#h(2)) = eγ(1A#h(1))eγ(a#h(2)) = (ω−1(S(h(2)), h(3))#S(h(1)))(a#h(4)) = aω−1(S(h(4)), h(5))(S(h(3)) · a)ω(S(h(2)), h(6))#S(h(1))h(7) = aω−1(S(h(5)), h(6))ω(S(h(4)), h(7))(S(h(3)) · a)#S(h(1))h(2) = (S(h(2)) · (h(3) · 1A))(S(h(1)) · a)#1H = (S(h) · a)#1H = i(eǫr(a#h)). Therefore, eA ⊂ eA#ωH is a E(A)#H-cleft extension. (cid:4) 38 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA 7. Conclusions and outlook - All the cohomology theory done in this paper was done over co-commutative Hopf algebras acting partially over commutative algebras. This can be ge- neralized for co-commutative Hopf algebra objects and commutative algebra objects in braided monoidal categories. - One topic of interest is to relate this cohomology for partial actions and the cohomology for its globalization, then constructing a bridge between this theory and the classical Sweedler's theory. - The last theorem placed the notion of a partially cleft extension within the context of cleft extensions for Hopf algebroids. This suggests, perhaps, that this entire cohomological theory can be understood properly as a cohomolog- ical theory of Hopf algebroids. - Another topic to be explored in further research can be the obstruction theory for the existence of partially cleft extensions and its relation with the third cohomology group in the same spirit of [25]. Acknowledgements The authors would like to thank Mikhailo Dokuchaev and Mykola Khrypchenko for their precious suggestions and comments on this work. We also thank Joost Vercruysse for pointing us some problems in the construction of the Hopf algebra eA. References [1] A. Adem, R. Milgram: "Cohomology of finite groups", Springer Verlag (1994). [2] M.M.S. Alves, E. Batista: "Enveloping actions for partial Hopf actions", Comm. Algebra 38 (2010), 2872-2902. [3] M. M. S. Alves, E. Batista: "Partial Hopf actions, partial invariants and a Morita context", J. Algebra Discrete Math. 3 (2009), 1-19. [4] M. M. S. Alves, E. Batista: "Globalization theorems for partial Hopf (co)actions and some of their applications", Contemp. Math. 537 (2011), 13-30. [5] M.M.S. Alves, E. Batista, M. Dokuchaev, A. Paques: "Twisted partial actions of Hopf Algebras", Israel Journal of Mathematics, 197 (2013) 263-308 [6] M.M.S. Alves, E. Batista, M. Dokuchaev, A. Paques: "Globalization of Twisted partial Hopf actions", Journal of the Australian Mathematical Society 101 (2016) 1-28. [7] M.M.S. Alves, E. Batista, J. Vercruysse: "Partial representations of Hopf algebras", Journal of algebra 426 (2015) 137-187. [8] E. Batista: "Partial actions: what they are and why we care", Bulletin of the Belgian Mathe- matical Society Simon Stevin 24 (2017) 35-71. [9] E. Batista, J. Vercruysse: "Dual constructions for partial actions of Hopf algebras", J. Pure Appl. Algebra 220.2 (2016) 518-559. [10] G. Böhm: "Hopf algebroids", Handbook of Algebra Vol 6, ed. by M. Hazewinkel, Elsevier (2009) 173-236. arXiv:0805.3806 [11] G. Böhm, T. Brzezinski: "Cleft extensions of Hopf algebroids", Appl. Categorical Structures 14.5-6 (2006) 431-469. [12] G. Böhm, D. Stefan: "(Co) cyclic (co) homology of bialgebroids: an approach via (co) monads", Communications in Mathematical Physics 282.1 (2008) 239-286. [13] S. Caenepeel, K. Janssen: "Partial (co)actions of Hopf algebras and partial Hopf-Galois theory", Communications in Algebra 36, (2008) 2923-2946. [14] M. Dokuchaev, R. Exel:"Associativity of Crossed Products by Partial Actions, Enveloping Ac- tions and Partial Representations", Trans. Amer. Math. Soc. 357 (5) (2005) 1931-1952. [15] M. Dokuchaev, R. Exel, J. J. Simon: "Crossed products by twisted partial actions and graded algebras", J. Algebra 320 No. 8 (2008) 3278-3310. COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 39 [16] M. Dokuchaev, R. Exel, J. J. Simon: "Globalization of twisted partial actions", Trans. Amer. Math. Soc. 362 No. 8 (2010) 4137-4160. [17] M. Dokuchaev, M. Khrypchenko: "Partial cohomology of groups", Journal of Algebra 427 (2015) 142-182. [18] M. Dokuchaev, M. Khrypchenko: "Twisted partial actions and extensions of semilattices of groups by groups" preprint arXiv:1602.02424 (2016). [19] R. Exel: "Circle actions on C ∗-algebras, partial automorphisms and generalized Pimsner- Voiculescu exact sequences", J. Funct. Anal. 122.3 (1994) 361-401. [20] R. Exel: "Partial actions of groups and actions of inverse semigroups", Proc. Amer. Math. Soc. 126.12 (1998) 3481-3494. [21] H. Hopf: "Über die topologie der Gruppen-Mannigfaltigkeiten und ihrer verallgemeinerungen", in Collected Papers - Gesammelte Abhandlungen (Springer Collected Works in Mathematics), Springer-Verlag (2013). [22] N. Kowalzig, H. Posthuma: "The cyclic theory of Hopf algebroids", Journal of Noncommutative Geometry 5.3 (2011) 423-476. [23] H. Lausch: "Cohomology of inverse semigroups", J. Algebra 35 (1975), 273-303. [24] M. Lawson: "Inverse Semigroups", World Scientific (1998). [25] P. Schauenburg: "Cohomological obstructions to cleft extensions over cocommutative Hopf algebras", K-Theory. 24 (2001) 227-242. [26] M.E. Sweedler: "Cohomology of algebras over Hopf algebras", Transactions of the American Mathematical Society 133.1 (1968) 205-239. [27] E. Weiss: "Cohomology of groups", Academic Press (1969). Appendix A. Some additional proofs In this appendix, we provide with details the proofs of Section 3 that were omitted in the final journal version of the paper. We leave them here for convenience of the interested reader. Proof of Proposition 3.1: In fact, consider n ≥ 1 and (h1 ⊗ · · · ⊗ hn) ∈ H ⊗n, then (1)een(h1 ⊗ . . . ⊗ hn) = (h1 . . . hn) · 1A is an idempotent in Homk(H ⊗n, A). (2), . . . , hn (2)) (1))een(h1 = = (P A2) = (1) . . . hn (2) . . . hn (1)) · 1A)((h1 (1), . . . , hn ((h1 (2)) · 1A) ((h1 . . . hn)(1) · 1A)((h1 . . . hn)(2) · 1A) een ∗een(h1, . . . , hn) =een(h1 (h1 . . . hn) · 1A =een(h1, . . . , hn). {z } m−n Take any n < m and (h1 ⊗ . . . ⊗ hm) ∈ H ⊗m, then This proves our statement. (2) en,m = e ⊗ ǫ ⊗ . . . ⊗ ǫ is an idempotent in Homk(H ⊗m, A). = e(h1 = e(h1 (1), . . . , hm (1))en,m(h1 (1), . . . , hn (1), . . . , hn en,m ∗ en,m(h1, . . . , hm) = en,m(h1 (2), . . . , hm (2)) (2) ) . . . ǫ(hm (2) ) . . . ǫ(hm (∗) = e ∗ e(h1, . . . , hn)ǫ(hn+1) . . . ǫ(hm) = e(h1, . . . , hn)ǫ(hn+1) . . . ǫ(hm) = en,m(h1, . . . , hm). (2), . . . , hn (1) ) . . . ǫ(hm (1) ) . . . ǫ(hm (2), . . . , hn (2))ǫ(hn+1 (1))ǫ(hn+1 (1))e(h1 (2))ǫ(hn+1 (1))ǫ(hn+1 (1))e(h1 (2)) (2)) 40 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA The identity (∗) follows from ǫ(h(1))ǫ(h(2)) = ǫ(h(1)ǫ(h(2))) = ǫ(h), and therefore (cid:4) en,m is idempotent. Proof of Proposition 3.4: en(h1, . . . , hn) = h1 · (h2 · (. . . Indeed, take (h1 ⊗ . . . ⊗ hn) ∈ H ⊗n, then · (hn · 1A) . . . )). (1), . . . , hn (1))ǫ(h2 en(h1, . . . , hn) =ee1,n ∗ee2,n ∗ · · · ∗een(h1, . . . , hn) = ee1,n(h1 (2)) . . .een(h1 = ee1(h1 . . .een−1(h1 (n−1))een(h1 (1))ee2,n(h1 (1))ee2(h1 (1)) . . . ǫ(hn (n−1), . . . , hn−1 (1)) . . . ǫ(hn (2), . . . , hn (2), h2 (2))ǫ(h3 (2)) . . . ǫ(hn (n), . . . , hn (n), . . . , hn (2)) (n)) (2)) . . . ǫ(hn (1) ·1A)(h1 (n−1))ǫ(hn (1))(h1 (1)·1A) . . . (h1 (1) · 1A) . . . ((h1 (2)h2 (n−1)h2 (2) · 1A)ǫ(h3 (n−2) . . . hn−1 (2)) . . . (h1 (n)h2 (n−2) . . . hn−1)(1) · 1A) (n−1)h2 = (h1 = (h1 (n)) (1) · 1A)ǫ(h2 (1)·1A)(h1 (1) · 1A)(h1 (n−1)h2 (2)h2 (2)h2 = (h1 ((h1 (n−2) . . . hn−1)(2)hn · 1A) (n) . . . hn (n−1) . . . hn−1 (n) · 1A) (2) hn· 1A) (P A3) = (h1 (1) · 1A)(h1 (2)h2 (1) · 1A) . . . ((h1 (n−1) . . . hn−2 (2) hn−1 · (hn · 1A)) = (h1 = (h1 ((h1 (n−2) . . . hn−2 (1)· 1A) . . . (h1 (1) · 1A) . . . ((h1 (1)· 1A)(h1 (2)h2 (1) · 1A)(h1 (2)h2 (n−2) . . . hn−3 (2) hn−2)(2)hn−1 · (hn · 1A)) (n−1) . . . hn−3 (2)h2 (1) · 1A) . . . (h1 (1) · 1A)(h1 (n−2) . . . hn−3 (P A3) = (h1 (1) · 1A)(h1 (n−1) . . . hn−2 (2) hn−1 · (hn · 1A)) (1) hn−2)(1) · 1A) (1) hn−2 · (hn−1 · (hn · 1A))) = . . . = h1 · (h2 · (. . . · (hn−1 · (hn · 1A)) . . . )). in which (· · · ) between the last two equalities means applying repeatedly the process using (PA3) until we obtain the result. (cid:4) Proof of Lemma 3.8: Take (h1 ⊗ · · · ⊗ hn+1) ∈ H ⊗n+1, then, (i) En(f ∗ g) = En(f ) ∗ En(g), for f, g ∈ C n par(H, A). En(f ∗ g)(h1, . . . , hn+1) = h1 · (f ∗ g(h2, . . . , hn+1)) = h1 · (f (h2 (1), . . . , hn+1 (1) )g(h2 = (h1 (1) · (f (h2 = En(f )(h1 (1), . . . , hn+1 (1), . . . , hn+1 (1) ))(h1 (1) )En(g)(h1 (2), . . . , hn+1 (2) · (g(h2 (2) )) (2), . . . , hn+1 (2) ))) (2), . . . , hn+1 (2) ). (ii) En(en) = en+1. En(en)(h1, . . . , hn+1) = h1 · en(h2, . . . , hn+1) = h1 · (h2 · (· · · · (hn+1 · 1A) . . . )) = en+1(h1, . . . , hn+1) COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 41 (iii) in,m(f ∗ g) = in,m(f ) ∗ in,m(g), for n < m and f, g ∈ C n par(H, A). in,n+1(f ∗ g)(h1, . . . , hn+1) = f ∗ g(h1, . . . , hn)ǫ(hn+1) = f (h1 (1), . . . , hn = f (h1 (1), . . . , hn = in,n+1(f )(h1 = in,n+1(f ) ∗ in,n+1(g)(h1, . . . , hn+1) (2))ǫ(ǫ(hn+1 (2), . . . , hn (1) )g(h1 (2), . . . , hn (1) )in,n+1(g)(h1 (1))g(h1 (1))ǫ(hn+1 (1), . . . , hn+1 (1) )hn+1 (2) ) (2))ǫ(hn+1 (2) ) (2), . . . , hn+1 (2) ) (iv) in,m(en) ∗ em = em. Take h1 ⊗ . . . ⊗ hm ∈ H ⊗m, then, in,m(en) ∗ em(h1, . . . , hm) = in,m(en)(h1 (1), . . . , hm (1))em(h1 (2), . . . , hm (2)) = en(h1 = (h1 (1), . . . , hn (1) · (. . . · (hn (1) ) . . . ε(hm (1))ε(hn+1 (1) · 1A) . . . ))(h1 (1))(h1 (2) · (. . . · (hn (2)·(hn+1 (2) ·(. . . ·(hm (2)·1A). . .))). . .)) (2) · (hn+1 · (. . . · (hm · 1A) . . . ))) . . . )) (2)·(. . . ·(hn (P A2) = h1·[(h2 (1)·(. . . ·(hn (1)·1A) . . . ))(h2 (2)·(. . . ·(hn (2)·(hn+1·(. . . ·(hm·1A) . . . ))) . . . ))] (P A2) · (hn · (1A(hn+1 · (. . . = h1 · (h2 · (h3 · (. . . = h1 · (h2 · (. . . = em(h1, . . . , hm) (v) (f ∗ g) ◦ µi = (f ◦ µi) ∗ (g ◦ µi), for f, g ∈ C n · (hn · (hn+1 · (. . . · (hm · 1A) . . . )))) . . . ))) · (hm · 1A) . . . ))) . . . )) par(H, A) and ∀ i ∈ {1, . . . , n}. In fact, for all i ∈ {1, . . . , n} and for f, g ∈ C n par(H, A) (f ∗ g) ◦ µi(h1, . . . , hi, hi+1, . . . , hn+1) = (f ∗ g)(h1, . . . , hihi+1, . . . , hn+1) = f (h1 (1) )g(h1 (1), . . . , (hihi+1)(1), . . . , hn+1 (1), . . . , hi (1)hi+1 = f (h1 (2), . . . , hi (1) )g ◦ µi(h1 = f ◦ µi(h1 = (f ◦ µi) ∗ (g ◦ µi)(h1, . . . , hi, hi+1, . . . , hn+1) (1) )g(h1 (1) , . . . , hn+1 (1) , . . . , hn+1 (1), hi+1 (1), . . . , hi (2), . . . , (hihi+1)(2), . . . , hn+1 (2) ) (2)hi+1 (2), . . . , hi (2) , . . . , hn+1 (2) ) (2) , . . . , hn+1 (2), hi+1 (2) ) (vi) (en ◦ µn) ∗ in,n+1(en) = en+1. (en ◦ µn) ∗ in,n+1(en)(h1, . . . , hn+1) = en ◦ µn(h1 en(h1 (h1 (1), . . . , hn (1)hn+1 (1) · (... · (hn (1), hn+1 (1) )en(h1 (1)hn+1 · 1A)...)))(h1 (1) )in,n+1(en)(h1 (2), . . . , hn (1), . . . , hn (1) · (h2 (2))ε(hn+1 (2) ) (2) · (h2 (2), . . . , hn (2), hn+1 (2) ) (2) · (... · (hn (2) · 1A)...))) (1) · (... · (hn h1 · [(h2 h1 · (h2 · (... · (hn−1 · [(hn (1)hn+1 · 1A)...))(h2 (2) · (... · (hn (2) · 1A)...))] (1)hn+1 · 1A)(hn (2) · 1A)])...)) h1 · (h2 · (... · (hn−1 · (hn · (hn+1 · 1A)))...)) en+1(h1, . . . , hn+1) = = = (P A2) = = (P A3) = = 42 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA (vii) (en ◦ µi) ∗ en+1 = en+1, ∀ i ∈ {1, · · · , n − 1}. (en ◦ µi) ∗ en+1(h1, . . . , hi, hi+1, . . . , hn+1) = (1) )en+1(h1 (2) · (... · (hi (1), hi+1 (1) , . . . , hn+1 (1) , . . . , hn+1 (1) )(h1 (1), . . . , hi (1), . . . , hi (2), . . . , hi (2) · (hi+1 = en ◦ µi(h1 = en(h1 = (h1 (1)·(...·(hi (1)hi+1 (2) , . . . , hn+1 (2) ) · 1A)...)))...)) (2) (2) ·1A)...)))...)) (2), hi+1 (2) · (... · (hn+1 (2) ·(...·(hn+1 (2)·(hi+1 (2) ·(...·(hn+1 (2)·(hi+1 (2) ·1A)...)))...))] (2) ·1A)...)))])...) (1) ·1A)...))...))(h1 (1) ·1A)...))...))(h2 (1) ·(...·(hn+1 (2)·(...·(hi (2)·(...·(hi (1) ·1A)...)))(hi (2)·(hi+1 (2) ·(...·(hn+1 (1)hi+1 (1) ·(...·(hn+1 (1)hi+1 (1)hi+1 (1) ·(...·(hn+1 (1) ·(hi+2 (1)hi+1 (2) · (... · (hn+1 (1)hi+1·[(hi+2 (2) (P A2) = h1·[(h2 (1)·(...·(hi = h1·(...·(hi−1·[(hi (P A3) = h1 · (... · (hi−1 · [(hi (hi (2)hi+1 (2) · (hi+2 (P A2) = h1·(...·(hi−1·[(hi (P A2) = h1 · (... · (hi−1 · [(hi (1) · (hi+2 (1) · (... · (hn+1 · 1A)...)))(hi (1) (3) · 1A)])...) · 1A)...))) (1) ·(...·(hn+1 (1) ·1A)...))(hi+2 (2) ·(...·(hn+1 (1)hi+1 · [hi+2 · (... · (hn+1 · 1A)...)])(hi (2) · 1A)])...) (2) ·1A)...))])(hi (2)·1A)])...) (P A3) = h1 · (... · (hi−1 · (hi · (hi+1 · (hi+2 · (... · (hn+1 · 1A)...)))))...) = en+1(h1, . . . , hn+1) Proof of Lemma 3.10: Let f ∈ C n−1 (i) En(in−1,n(f )) = in,n+1(En−1(f )). par (H, A) and h1 ⊗ . . . ⊗ hn+1 ∈ H ⊗n+1. En(in−1,n(f ))(h1, . . . , hn+1) = h1 · (in−1,n(f )(h2, . . . , hn+1)) = h1 · (f (h2, . . . , hn)ǫ(hn+1)) = (h1 · f (h2, . . . , hn))ǫ(hn+1) = (En−1(f ))(h1, . . . , hn)ǫ(hn+1) = in,n+1(En−1(f ))(h1, . . . , hn+1) (ii) (en−1 ◦ µi ◦ µi+1) ∗ en+1 = en+1, ∀ i ∈ {1, · · · , n − 1}. COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 43 For every i ∈ {1, . . . , n − 1}, (en−1 ◦ µi ◦ µi+1) ∗ en+1(h1, . . . , hn+1) = (1), . . . , hi = en−1(h1 = h1 (1) · (... · (hi (h1 (2) · (... · (hi (1) )en+1(h1 · 1A)...))...) (1)hi+1 (1) hi+2 (1)hi+1 (2) · (hi+1 (1) , . . . , hn+1 (1) hi+2 (1) · (... · (hn+1 (2) · (hi+2 (1) hi+2 (1) · (... · (hn+1 (2) · (... · (hn+1 (1) (2) (1) · 1A)...)) · 1A)...))))...) (2), . . . , h(n+1) (2) ) (P A2) (P A3) (hi (2) · (hi+1 = h1 · (h2 · (... · [(hi (2) · (hi+2 = h1 · (h2 · (... · [(hi (2) · (hi+2 = h1 · (h2 · (... · [(hi (2)hi+1 (hi (1)hi+1 (2) · (... · (hn+1 (1)hi+1 (2) (1) hi+2 (2) · (... · (hn+1 (1) hi+2 (1)hi+1 (2) (P A3) · 1A)...))))]...)) (1) · (... · (hn+1 · 1A)...)))(hi (1) · 1A)...)) (3) · 1A)]...)) (hi (2)hi+1 (2) hi+2 (2) · (... · (hn+1 (2) (P A3) = h1 · (h2 · (... · [(hi (1)hi+1 (1) · (... · (hn+1 (1) (3)hi+1 · 1A)...))(hi · 1A)...)) (3) · 1A)(hi (1) hi+2 · (... · (hn+1 · 1A)...))(hi (4) · 1A)]...)) (2)hi+1 (2) · 1A)(hi (3) · 1A)]...)) (P A3) (1)hi+1 · (hi+2 · (... · (hn+1 · 1A)...)))(hi = h1 · (h2 · (... · [(hi = h1 · (h2 · (... · (hi · (hi+1 · (hi+2 · (... · (hn+1 · 1A)...))))...)) = en+1(h1, . . . , hn+1). (2) · 1A)]...)) (iii) (en−1 ◦ µi ◦ µi+j) ∗ en+1 = en+1, ∀ i ∈ {1, · · · , n − 1}, j ∈ {2, · · · , n − i}. For i ∈ {1, · · · , n − 1}, j ∈ {2, · · · , n − i}, we have (en−1 ◦ µi ◦ µi+j) ∗ en+1(h1, . . . , hn+1) = (1)hi+1 (1) hi+j+1 (1) , . . . , hi+j (2), hi+1 (2) , . . . , hi+j (2) , hi+j+1 (1) , . . . , hn+1 (1) ) = en−1(h1 (1), . . . , hi = h1 en+1(h1 (1) · (... · (hi h1 (2) · (... · (hi (2), . . . , hi (1)hi+1 (2) · (hi+1 (P A2) = h1 · [(h2 (1) · (... · (hi (h2 (2) · (... · (hi (2) · (hi+1 (1) · (... · (hi+j (1) hi+j+1 (1) (2) (2) · (... · (hi+j (1)hi+1 (2) · (... · (hi+j (2) · (hi+j+1 (1) hi+j+1 (2) · (hi+j+1 (1) · (... · (hi+j (1) (2) , . . . , hn+1 (2) ) · 1A)...))...))...) (2) · (... · (hn+1 (1) · (... · (hn+1 (2) · 1A)...)))...)))...) · (... · (hn+1 (1) · (... · (hn+1 (2) · 1A)...))...))...)) · 1A)...)))...)))...))] 44 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA = h1 · (... · (hi−1 · [(hi (hi (2) · (hi+1 (2) · (... · (hi+j (P A3) = h1 · (... · (hi−1 · [(hi (1) (1) hi+j+1 · (... · (hn+1 (2) (1) · (... · (hi+j (1)hi+1 (2) · (hi+j+1 (1) · (hi+2 (1)hi+1 (2) (1) · (... · (hi+j · (... · (hn+1 (1) · 1A)...))...)) · 1A)...)))...)))])...) (1) hi+j+1 (1) · (... · (hn+1 (2) · (... · (hn+1 (1) · 1A)...))...))) · 1A)...)))...)))(hi (3) · 1A)])...) (hi (2)hi+1 (2) · (hi+2 (2) · (... · (hi+j (2) · (hi+j+1 (2) (1)hi+1 · [(hi+2 (P A2) = h1 · (... · (hi−1 · [(hi (hi+2 (2) · (... · (hi+j = h1 ·(...·(hi−1 ·[(hi (P A2) (1) · (... · (hi+j (1) hi+j+1 (1) · (...·(hn+1 (1) ·1A)...))...)) (2) · (hi+j+1 (2) · (... · (hn+1 (2) · 1A)...)))...))])(hi (2) · 1A)])...) (hi+j (2) · (hi+j+1 (2) (1)hi+1 ·[(hi+2 ·(...·(hi+j−1 ·[(hi+j · (... · (hn+1 (2) · 1A)...)))])...))])(hi (1) hi+j+1 (2) · 1A)])...) (1) ·(... · (hn+1 (1) · 1A)...)) (P A3) = h1 ·(...·(hi−1 ·[(hi (1)hi+1 ·[...·(hi+j−1 ·[(hi+j ·(hi+j+2 (1) ·(...·(hn+1 (1) ·1A)...))) (hi+j (2) hi+j+1 (2) · (hi+j+2 (2) · (... · (hn+1 (2) (3) · 1A)])...])(hi (2) · 1A)])...) (1) hi+j+1 · 1A)...)))(hi+j (1) (P A2) = h1 ·(...·(hi−1 ·[(hi (1)hi+1 ·[...·(hi+j−1 ·[(hi+j (hi+j+2 (2) · (... · (hn+1 (2) · 1A)...))])(hi+j (2) · 1A)])...])(hi (1) hi+j+1 ·[(hi+j+2 (2) · 1A)])...) (1) ·(...·(hn+1 (1) ·1A)...)) (P A2) = h1 ·(...·(hi−1 ·[(hi (2) · 1A)]...])(hi (hi+j (1)hi+1 ·[...·(hi+j−1 ·[(hi+j (2) · 1A)])...) (1) hi+j+1 ·[(hi+j+2 ·(...·(hn+1 ·1A)...))]) (P A3) = h1·(...·(hi−1·[(hi (1)hi+1·[...·[(hi+j·(hi+j+1·(hi+j+2·(...·(hn+1·1A)...))))]...])(hi (2)·1A)])...) (P A3) = h1·(...·(hi−1 ·[(hi·(hi+1 ·(...·(hi+j ·(hi+j+1·(hi+j+2·(...·(hn+1 ·1A)...)))))...)))])...) = en+1(h1, . . . , hn+1). (iv) En(f ◦ µi) = En−1(f ) ◦ µi+1, ∀ i ∈ {1, . . . , n − 1}. In fact, for every i ∈ {1, . . . , n − 1}, we have En(f ◦ µi)(h1, . . . , hn+1) = h1·((f ◦ µi)(h2, . . . , hi, = h1 · (f (h2, . . . , hi, hi+1hi+2, . . . , hn+1)) On the other hand, , hi+2, . . . , hn+1)) i−esima coord. hi+1 {z} En−1(f ) ◦ µi+1(h1, . . . , hn+1) = En−1(f )(h1, . . . , hi, hi+1hi+2, . . . , hn+1) = h1 · (f (h2, . . . , hi, hi+1hi+2, . . . , hn+1)) So, we prove the equality. (v) En ◦ En−1(f ) = i1,n+1(ee1) ∗ (En−1(f ) ◦ µ1). i1,n+1(ee1) ∗ (En−1(f ) ◦ µ1)(h1, . . . , hn+1) = = i1,n+1(ee1)(h1 (1) )En−1(f ) ◦ µ1(h1 (1), . . . , hn+1 (2), . . . , hn+1 (2) ) COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 45 = ee1(h1 = (h1 = (h1 (1))ε(h2 (1)) . . . ε(hn+1 (1) )En−1(f )(h1 (2)h2 (2), h3 (1) · 1A)ε(h2 (1) · 1A)ε(h2 (1) · 1A)(h1 (1) )(h1 (1) )(h1 (1)) . . . ε(hn+1 (1)) . . . ε(hn+1 (2)h2 · (f (h3, . . . , hn+1))) (2)h2 (2)h2 (2) · (f (h3 (2) · (f (h3 (2), . . . , hn+1 (2) ) (2) ))) (2) ))) (2), . . . , hn+1 (2), . . . , hn+1 = (h1 = h1 · (h2 · (f (h3, . . . , hn+1))) = h1 · (En−1(f )(h2, h3, . . . , hn+1)) = En(En−1(f ))(h1, . . . , hn+1) = En ◦ En−1(f )(h1, . . . , hn+1) (vi) in,n+1(f ◦ µi) ∗ in−1,n(f −1) ◦ µi = in,n+1(en−1 ◦ µi), ∀ i ∈ {1, . . . , n − 1}. For i ∈ {1, . . . , n − 1}, we have in,n+1(f ◦ µi) ∗ in−1,n(f −1) ◦ µi(h1, . . . , hn+1) = (1) )in−1,n(f −1) ◦ µi(h1 (1) )in−1,n(f −1)(h1 = in,n+1(f ◦ µi)(h1 = f ◦ µi(h1 (1), . . . , hn (1), . . . , hn+1 (1))ǫ(hn+1 (2), . . . , hn+1 (2) ) (2), . . . , hi (2)hi+1 (2) , . . . , hn+1 (2) ) (2), . . . , hi (1) )f −1(h1 (1))f −1 ◦ µi(h1 (2), . . . , hi (2)hi+1 (2) , . . . , hn (2), hi+1 (2) , . . . , hn (2))ǫ(hn+1 (2) ) (2))ǫ(hn+1) n−th coord {z} (1)hi+1 (1), . . . , hi (1), . . . , hi (1))ǫ(hn+1 (1) , . . . , hn (1) , . . . , hn (1), hi+1 = f (h1 = f ◦ µi(h1 = (f ◦ µi) ∗ (f −1 ◦ µi)(h1, . . . , hn)ǫ(hn+1) = (f ∗ f −1) ◦ µi(h1, . . . , hn)ǫ(hn+1) = en−1 ◦ µi(h1, . . . , hn)ǫ(hn+1) = in,n+1(en ◦ µi)(h1, . . . , hn+1). (vii) (in−1,n(f ) ◦ µn) ∗ in−1,n+1(f −1) = in−1,n+1(en−1). (1), . . . , hn+1 = (in−1,n(f ) ◦ µn)(h1 = in−1,n(f )(h1 = f (h1 (in−1,n(f ) ◦ µn) ∗ in−1,n+1(f −1)(h1, . . . , hn+1) = (1) )in−1,n+1(f −1)(h1 (2), . . . , hn−1 (2), . . . , hn−1 (1)hn+1 (1) )f −1(h1 (1) )f −1(h1 (2), . . . , hn−1 = (f ∗ f −1)(h1, . . . , hn−1)ε(hn)ε(hn+1) = (en−1)(h1, . . . , hn−1)ε(hn)ε(hn+1) = in−1,n+1(en−1)(h1, . . . , hn+1) (1), . . . , hn (1) )ε(hn (1) )f −1(h1 (1), . . . , hn−1 (1), . . . , hn−1 (1)hn+1 (1)hn+1 (2) )ε(hn = f (h1 (2), . . . , hn+1 (2) ) (2))ε(hn+1 (2) ) (2))ε(hn+1 (2) ) (2))ε(hn+1 (2) ) (2) )ε(hn (2) )ε(hn (1) )ε(hn (viii) (f ◦ µi ◦ µi) ∗ (f −1 ◦ µi ◦ µi+1) = en−1 ◦ µi ◦ µi, ∀ i ∈ {1, . . . , n − 1}. For i ∈ {1, . . . , n − 1}, we have 46 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA (f ◦ µi ◦ µi) ∗ (f −1 ◦ µi ◦ µi+1)(h1, . . . , hn+1) = =f ◦ µi ◦ µi(h1 =f ◦ µi(h1 =f (h1 (1), . . . , hi (1), . . . , hn+1 (1)hi+1 (1) hi+2 (1) )f −1 ◦ µi ◦ µi+1(h1 (1) , hi+2 (1) , . . . , hn+1 (1) , . . . , hn+1 (1) )f −1(h1 (1) )f −1 ◦ µi(h1 (2), . . . , hi (1)hi+1 (1), . . . , hi (2), . . . , hn+1 (2) ) (2), . . . , hi (2) hi+2 (2)hi+1 (2) , . . . , hn+1 (2) hi+2 (2), hi+1 (2) ) (2) , . . . , hn+1 (2) ) =(f ∗ f −1)(h1, . . . , hihi+1hi+2, . . . , hn+1) =en−1(h1, . . . , hihi+1hi+2, . . . , hn+1) =en−1 ◦ µi ◦ µi(h1, . . . , hn+1) (ix) (f ◦ µi ◦ µi+j) ∗ (f −1 ◦ µi+j−1 ◦ µi) = en−1 ◦ µi ◦ µi+j, ∀ i ∈ {1, . . . , n − 2}, ∀ j ∈ {2, . . . , n − i}. For i ∈ {1, . . . , n − 2} e j ∈ {2, . . . , n − i}, we have (f ◦ µi ◦ µi+j) ∗ (f −1 ◦ µi+j−1 ◦ µi)(h1, . . . , hn+1) = (1) )f −1 ◦ µi+j−1 ◦ µi(h1 = f ◦ µi ◦ µi+j(h1 = f ◦ µi(h1 = f (h1 (1), . . . , hi (1), . . . , hi+j (1)hi+1 (1) (1), . . . , hn+1 (1) hi+j+1 (1) , . . . , hi+j (2), . . . , hi , . . . , hn+1 (1) hi+j+1 (1) (2)hi+1 (2) f −1 ◦ µi+j−1(h1 , . . . , hn+1 (1) ) , hi+2 (2) , . . . , (1) )f −1 ◦ µi+j−1(h1 (2)hi+1 (2) , . . . , hn+1 (2) ) (2), . . . , hn+1 (2) ) (2), . . . , hi i−th coord {z } , . . . , hn+1 (2) ) (i+j−1)−th coord hi+j (2) {z} = f (h1 (1), . . . , hi (1)hi+1 f −1(h1 (2), . . . , hi (1) hi+j+1 (1) , . . . , hi+j (2)hi+1 (2) , . . . , hi+j (2) hi+j+1 (2) (1) , . . . , hn+1 (1) ) , . . . , hn+1 (2) ) = (f ∗ f −1)(h1, . . . , hihi+1, . . . , hi+jhi+j+1, . . . , hn+1) = en−1(h1, . . . , hihi+1, . . . , hi+jhi+j+1, . . . , hn+1) = en−1 ◦ µi ◦ µi+j(h1, . . . , hn+1). Proof of Theorem 3.11: For any f ∈ C n Indeed, take any f ∈ C n par(H, A), then (cid:4) par(H, A), we have that δn+1◦δn(f ) = en+2. δn+1(δn(f )) = En+1(δn(f ))∗ (δn(f ))(−1)i ◦ µi ∗ in+1,n+2((δn(f ))(−1)n+2 ) n+1Yi=1 δn(f (−1)i ) ◦ µi ∗ in+1,n+2(δn(f (−1)n+2 )) =En+1(δn(f ))∗ =En+1(En(f )∗ n+1Yi=1 nYj=1 f (−1)j ◦ µj ∗ in,n+1(f (−1)n+1 ))∗ (En(f (−1)i )∗ f (−1)i+j ◦ µj nYj=1 ∗in,n+1(f (−1)n+i+1 )) ◦ µi ∗ in+1,n+2(En(f (−1)n+2 )∗ f (−1)n+j+2 ◦µj ∗ in,n+1(f (−1)2n+3 )) n+1Yi=1 nYj=1 COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 47 = En+1(En(f )) ∗ En+1(f (−1)j ◦ µj) ∗ En+1(in,n+1(f (−1)n+1 ∗ )) En(f (−1)i ) ◦ µi Lemma 3.10 (v) Lemma 3.10 (iv) Lemma 3.10 (i) f (−1)i+j ◦ µj ◦ µi∗ in,n+1(f (−1)n+i+1 ) ◦ µi ∗ in+1,n+2(En(f (−1)n+2 )) } {z {z n+1Yi=1 n+1Yi=1 } ∗in+1,n+2( f (−1)n+j+2 ◦ µj) ∗ in+1,n+2(in,n+1(f (−1)2n+3 )) ∗ {z n+1Yi=1 nYj=1 nYj=1 } nYj=1 ∗ nYj=1 = i1,n+2(ee1) ∗ En(f ) ◦ µ1∗ n+1Yi=1 n+1Yi=1 nYj=1 nYj=1 ∗in+1,n+2(En(f (−1)n+2 En(f (−1)i ) ◦ µi∗ )∗ En(f (−1)j ) ◦ µj+1 ∗ in+1,n+2(En(f (−1)n+1 )) f (−1)i+j ◦ µj ◦ µi∗ n+1Yi=1 in,n+1(f (−1)n+i+1 ) ◦ µi in+1,n+2(f (−1)n+j+2 ◦ µj) ∗ in,n+2(f (−1)2n+3 )) En(f (−1)j+1 ) ◦ µj∗ En(f (−1)i ) ◦ µi ∗ ∗in+1,n+2(En(f (−1)n+1 )) ∗in+1,n+2(En(f (−1)n+2 f (−1)i+j ◦ µj ◦ µi∗ in+1,n+2(f (−1)n+j+2 ◦ µj) nYj=1 in,n+1(f (−1)n+i+1 ) ◦ µi ∗ in,n+1(f (−1)2n+2 ) ◦ µn+1 ∗ in,n+2(f (−1)2n+3 )) n+1Yi=1 )∗ n+1Yi=1 nYj=1 En(f (−1)j+1 ∗ f (−1)j ) ◦ µj ∗ in+1,n+2(En(f (−1)n+1 ∗ f (−1)n+2 )) f (−1)i+j ◦ µj ◦ µi ∗ in+1,n+2(f (−1)n+j+2 ◦ µj)∗ in,n+1(f (−1)n+i+1 ) ◦ µi n+1Yj=1 = i1,n+2(ee1)∗ ∗ nYi=1 n+1Yj=1 = i1,n+2(ee1)∗ n+1Yi=1 nYj=1 ∗ ∗ = i1,n+2(ee1)∗ nYj=1 = i1,n+2(ee1)∗ nYj=1 ∗ nYj=1 nYi=1 {z Lemma 3.10 (vi) } } )∗ n+1Yi=1 nYj=1 f (−1)i+j ◦ µj ◦ µi ∗in,n+1(f (−1)2n+2 ) ◦ µn+1 ∗ in,n+2(f (−1)2n+3 )) Lemma 3.10 (vii) {z {z } En(en) Lema 3.8 (i) n+1Yj=1 ◦ µj ∗ in+1,n+2( En(en) Lemma 3.8 (i) {z } in+1,n+2(en ◦ µj) ∗ in,n+2(en) en+1 ◦ µj ∗ in+1,n+2(en+1)∗ n+1Yj=1 n+1Yi=1 nYj=1 f (−1)i+j ◦ µj ◦ µi in+1,n+2(en ◦ µj) ∗ in+1,n+2(in,n+1(en)) 48 E. BATISTA, A.D.M. MORTARI, AND M.M. TEIXEIRA en+1 ◦ µj ∗ in+1,n+2(en+1)∗ f (−1)i+j ◦ µj ◦ µi in+1,n+2(en ◦ µj) ∗ in+1,n+2(en ◦ µn) ∗ in+1,n+2(in,n+1(en)) en+1 ◦ µj ∗ in+1,n+2(en+1)∗ f (−1)i+j ◦ µj ◦ µi n+1Yi=1 nYj=1 {z n+1Yi=1 nYj=1 Lemma 3.8 (vi) } in+1,n+2(en ◦ µj) ∗ in+1,n+2(en+1) Lemma 3.8 (vii) {z en+1 ◦ µj ∗ in+1,n+2(en+1)∗ f (−1)i+j ◦ µj ◦ µi ∗ in+1,n+2(en+1) en+1 ◦ µj ∗ en+1 ◦ µn+1 ∗ in+1,n+2(en+1) ∗ en+1 ◦ µj ∗ en+2 ∗ f (−1)i+j ◦ µj ◦ µi n+1Yi=1 nYj=1 f (−1)i+j ◦ µj ◦ µi } } nYj=1 n+1Yi=1 {z n+1Yi=1 nYj=1 Lemma 3.8 (vi) n+1Yj=1 ∗ ∗ n+1Yj=1 = i1,n+2(ee1)∗ n−1Yj=1 = i1,n+2(ee1)∗ n−1Yj=1 = i1,n+2(ee1)∗ = i1,n+2(ee1)∗ = i1,n+2(ee1) ∗ = i1,n+2(ee1) ∗ en+2∗ n+1Yj=1 nYj=1 nYj=1 = en+2∗ n+1Yi=1 nYj=1 nYi=1 n−i+1Yj=2 n−1Yi=1 nYi=1 ∗ = en+2 = en+2 ∗ f (−1)2i ◦ µi ◦ µi∗ f (−1)2i+1 ◦ µi ◦ µi+1 ∗ } Lemma 3.8 (vii) {z n+1Yi=1 nYj=1 f (−1)i+j ◦ µj ◦ µi f (−1)i+j ◦ µj ◦ µi f (−1)2i+j ◦ µi ◦ µi+j∗ nYi=1 {z Lemma 3.10 (viii) n−i+1Yj=2 n−1Yi=1 {z Lemma 3.10 (ix) n−i+1Yj=2 n−1Yi=1 = en+2∗ en ◦ µi ◦ µi∗ en ◦ µi ◦ µi+j } f (−1)2i+j−1 ◦ µi+j−1 ◦ µi } COHOMOLOGY FOR PARTIAL ACTIONS OF HOPF ALGEBRAS 49 Once en+2 absorbs en ◦ µi ◦ µi for all i ∈ {1, . . . , n} and en ◦ µi ◦ µi+j, for all i ∈ {1, . . . , n − 1}, j ∈ {2, . . . , n − i − 1}, by Lemma 3.10 (ii) and (iii) respectively, we conclude that δn+1 ◦ δn(f ) = en+2 as we wanted to proof. (cid:4) Departamento de Matemática, Universidade Federal de Santa Catarina, Brazil E-mail address: [email protected] Departamento de Matemática, Universidade Federal de Santa Catarina, Brazil E-mail address: [email protected] Istituto Federal de Santa Catarina, Brazil E-mail address: [email protected]
1503.02341	1	1503	2015-03-08T23:25:30	Terwilliger algebras of wreath products by 3-equivalent schemes	[ "math.RA" ]	Recently G. Bhattacharyya, S.Y. Song and R. Tanaka began to study Terwilliger algebras of wreath products of one-class association schemes. K. Kim determined the structure of Terwilliger algebras of wreath products by one-class association schemes or quasi-thin schemes. In this paper, we study Terwilliger algebras of wreath products by $3$-equivalenced schemes.	math.RA	math	TERWILLIGER ALGEBRAS OF WREATH PRODUCTS BY 3-EQUIVALENCED SCHEMES KIJUNG KIM Abstract. Recently G. Bhattacharyya, S.Y. Song and R. Tanaka be- gan to study Terwilliger algebras of wreath products of one-class asso- ciation schemes. K. Kim determined the structure of Terwilliger alge- bras of wreath products by one-class association schemes or quasi-thin schemes. In this paper, we study Terwilliger algebras of wreath products by 3-equivalenced schemes. Key words: Terwilliger algebra; Wreath product; 3-equivalenced. 1. Introduction The Terwilliger algebra of an association scheme is an algebraic tool for the study of association schemes. In general, Terwilliger algebras are non- commutative, finite dimensional, and semisimple C-algebras. Also they are more combinatorial and complicated than Bose-Mesner algebras. Recently G. Bhattacharyya, S.Y. Song and R. Tanaka began to study Terwilliger al- gebras of wreath products of one-class association schemes in [1]. General- izations and related works were developed by the authors of [8, 4, 6, 9, 13]. In particular, the structure of Terwilliger algebras of wreath products by one-class association schemes or quasi-thin schemes was determined in [6]. In this way, the attempt of generalization seems to be far from a success. However, we propose studying Terwilliger algebras of wreath products by k-equivalenced schemes. For a given scheme, its Terwilliger algebra is contained in one point ex- tension of the scheme. For a positive integer k, an association scheme with more then one point is called k-equivalenced if each non-diaginal basic rela- tion has valency k. A coherent configuration (X, S) is called semiregular if xs ≤ 1 for x ∈ X, s ∈ S, where xs := {y ∈ X (x, y) ∈ s}. In [3, Lemma 5.13], the following lemma was proved. Lemma 1.1. Let (X, S) be an imprimitive k-equivalenced scheme. Then (X \ {x}, (Sx)X\{x}) is semiregular, where Sx is the basic relation set of the one point extension of (X, S). If the following problems for a given k ≥ 4 are solved, then it seems to be close to determine the structure of Terwilliger algebras of wreath products by k-equivalenced schemes. Date: May 31, 2021. 1991 Mathematics Subject Classification. 05E15; 05E30. 0This research was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Education (2013R1A1A2005349). 1 (i) When Terwilliger algebras of k-equivalenced schemes coincide with their one point extensions ? (ii) When the imprimitivity condition of Lemma 1.1 can be removed ? In this paper, we determine the structure of Terwilliger algebras of wreath products by 3-equivalenced schemes. Furthermore, it supports the impor- tance of the above two problems and provides another motivation to study k-equivalenced schemes. This paper is organized as follows. In Section 2, we review notations and basic results on coherent configurations and Terwilliger algebras as well as recent results on 3-equivalenced schemes. In Section 3, based on the fact that the adjacency algebras of the one point extensions of 3-equivalenced schemes coincide with their Terwilliger algebras, we determine all central primitive idempotents of Terwilliger algebras of wreath products by 3-equivalenced schemes. In Section 4, we state our main theorem. 2. Preliminaries In this section, we use the notations and terminology given in [2, 4]. 2.1. Coherent configurations and adjacency algebras. Let X be a finite set and S a partition of X × X. Put by S∪ the set of all unions of the elements of S. A pair C = (X, S) is called a coherent configuration on X if the following conditions hold: (i) 1X = {(x, x) x ∈ X} ∈ S∪; (ii) For s ∈ S, s∗ = {(y, x) (x, y) ∈ s} ∈ S; (iii) For all s, t, u ∈ S and all x, y ∈ X, pu st = {z ∈ X (x, z) ∈ s, (z, y) ∈ t} is constant whenever (x, y) ∈ u. The elements of X, S and S∪ are called the points, the basic relations and the relations, respectively. The numbers X and S are called the degree and rank, respectively. Any set ∆ ⊆ X for which 1∆ := {(x, x) x ∈ ∆} ∈ S is called the fiber. The set of all fibers is denoted by F ib(C). For s, t ∈ S, the product st is defined by {u ∈ S pu st 6= 0}. The coherent configuration C is called homogeneous or a scheme if 1X ∈ S. If Y is a union of fibers, then the restriction of C to Y is defined to be a coherent configuration CY = (Y, SY ), where SY is the set of all non-empty relations s ∩ (Y × Y ) with s ∈ S. Two coherent configurations (X, S) and (Y, T ) are called isomorphic if there exists a bijection τ : X ∪ S → Y ∪ T such that X τ = Y , Sτ = T and (ws)τ = (wτ )sτ for all s ∈ S and w ∈ X with ws 6= ∅. We denote it by (X, S) ≃ (Y, T ). For s ∈ S, let σs denote the matrix in M atX(C) that has entries (σs)xy = (cid:26) 1 if (x, y) ∈ s; 0 otherwise. We call σs the adjacency matrix of s ∈ S. Then Ls∈S subalgebra of M atX(C). We call Ls∈S Cσs becomes a Cσs the adjacency algebra of S, and denote it by A(S). 2 Let B be the set of primitive idempotents of A(S) with respect to the Hadamard multiplication. Then B is a linear basis of A consisting of {0, 1}- matrices such that σs = JX and σs ∈ B ⇔ σt s ∈ B. Xs∈B Remark 2.1. There are bijections between the sets of coherent configurations and adjacency algebras as follows: where C(A) = (X, S′) with S′ = {s ∈ X × X σs ∈ B}. S 7→ A(S) and A 7→ C(A), In a scheme (X, S), for each s ∈ S, ns := p1X ss∗ is called valency of s. Lemma 2.1 ([14]). Let (X, S) be a scheme. For u, v, w ∈ S, we have the following: (i) pv uwnv = pw u∗vnw = pu uvns. vw∗nu; (ii) nunv = Ps∈S ps Let (X, S) be a scheme and u, v ∈ S. By Lemma 2.1(i), we have pv u∗v = 0. Thus, given x ∈ X, the cardinality of the set if and only if pw uw = 0 {w ∩ (xu × xv) w ∈ S, pv uw 6= 0} equals the number u∗v. 2.2. Terwilliger algebras and one point extensions. Let (X, S) be a scheme. For U ⊆ X, we denote by εU the diagonal matrix in M atX (C) with entries (εU )xx = (cid:26) 1 if x ∈ U ; 0 otherwise. Note that for U, V ⊆ X. JU,V := εU JX εV and JU := JU,U The Terwilliger algebra of (X, S) with respect to x0 ∈ X is defined as a subalgebra of M atX(C) generated by {σs s ∈ S} ∪ {εx0s s ∈ S} (see [10]). The Terwilliger algebra will be denoted by T (X, S, x0) or T (S). Since A(S) and T (S) are closed under transposed conjugate, they are semisim- ple C-algebras. The set of irreducible characters of T (S) and A(S) will be denoted by Irr(T (S)) and Irr(A(S)), respectively. The trivial character 1A(S) of A(S) is a map σs 7→ p1X ss∗ and the corresponding central primitive idempotent is X−1JX . The trivial character 1T (S) of T (S) corresponds to the central primitive idempotent n−1 s εx0sJX εx0s Xs∈S of T (S). For χ ∈ Irr(A(S)) or Irr(T (S)), eχ will be the corresponding central primitive idempotent of A(S) or T (S). For convenience, we denote Irr(A(S)) \ {1A(S)} and Irr(T (S)) \ {1T (S)} by Irr(T (S))× and Irr(A(S))×, respectively. 3 Let C = (X, S) be a coherent configuration and x ∈ X. Denote by Sx the set of basic relations of the smallest coherent configuration on X such that 1x ∈ Sx and S ⊂ S∪ x . Then the coherent configuration Cx = (X, Sx) is called a one point extension of C. It is easy to see that given s, t, u ∈ S, the set xs is a union of fibers of Cx and the relation uxs,xt is a union of basic relations of Cx. Remark 2.2. The adjacency algebra of the one point extension Cx of a scheme C is related to T (X, S, x). In fact, T (X, S, x) ⊆ A(Sx). 2.3. Wreath products. Let (X, S) and (Y, T ) be schemes. For s ∈ S, set s = {((x, y), (x′, y)) (x, x′) ∈ s, y ∈ Y }. For t ∈ T , set ¯t = {((x, y), (x′, y′)) x, x′ ∈ X, (y, y′) ∈ t}. Also set S ≀ T = {s s ∈ S} ∪ {¯t t ∈ T \ {1Y }}. Then (X × Y, S ≀ T ) is a scheme called the wreath product of (X, S) by (Y, T ). For the adjacency matrices, we have σs = σs ⊗ IY , σ¯t = JX ⊗ σt. Note that and (x0, y0)s = (x0s, y0) = {(x, y0) x ∈ x0s}, (x0, y0)¯t = (X, y0t) = {(x, y) x ∈ X, y ∈ y0t} ε(x0,y0)s = εx0s ⊗ εy01Y , ε(x0,y0)¯t = Ps∈S εx0s ⊗ εy0t = IX ⊗ εy0t. 2.4. 3-equivalenced schemes. A scheme C = (Y, T ) is called 3-equivalenced if T = {1Y } ∪ T3, where T3 is the set of basic relations with valency 3. Lemma 2.2 ([5], Lemma 3.1). Let C = (Y, T ) be a 3-equivalenced scheme. For all u, v ∈ T × := T \ {1Y } with u 6= v∗, one of the following holds: (i) σuσv = σw1 + σw2 + σw3 for some distinct w1, w2, w3 ∈ T ×; (ii) σuσv = σw1 + 2σw2 for some distinct w1, w2 ∈ T ×. Lemma 2.3 ([5], Lemma 3.2). Let C = (Y, T ) be a 3-equivalenced scheme. For all u ∈ T ×, one of the following holds: (i) σuσu∗ = 3σ1Y + σw + σw∗ for some w ∈ T with w 6= w∗; (ii) σuσu∗ = 3σ1Y + 2σu and u = u∗. Lemma 2.4 ([5], Lemma 3.14). Let C = (Y, T ) be a 3-equivalenced scheme with T > 2 and y0 ∈ Y . Then the coherent configuration (Y \{y0}, (Ty0 )Y \{y0}) is semiregular. Theorem 2.5. Let C = (Y, T ) be a 3-equivalenced scheme and y0 ∈ Y . Then A(Ty0) = T (Y, T, y0). Proof. It follows from [5, Theorem 1.1] and [7, Section 8.5] that A(Ty0) = T (Y, T, y0). (cid:3) 4 3. Wreath products by 3-equivalenced schemes Let (X, S) and (Y, T ) be schemes. Fix x0 ∈ X and y0 ∈ Y , and consider (X × Y, S ≀ T ) and T (X × Y, S ≀ T, (x0, y0)). In the rest of this section, we assume that (Y, T ) is a 3-equivalenced scheme with T > 2. 3.1. The restriction of T (S ≀ T ) to X × (Y \ {y0}). The contents of this subsection are given in [6]. The Terwilliger algebra T (S ≀ T ) is generated by {JX ⊗ σt, IX ⊗ εy0t t ∈ T \ {1Y }} ∪ {σs ⊗ IY , εx0s ⊗ ε{y0} s ∈ S}. Since Ps∈S εx0s ⊗ ε{y0} = IX ⊗ ε{y0} and Ps∈S σs ⊗ IY = JX ⊗ IY , we consider a subalgebra U generated by {JX ⊗ σt, IX ⊗ εy0t t ∈ T }. It is easy to see that U is generated by {X−1JX ⊗ εy0t1 σtεy0t2 t1, t2 ∈ T } and isomorphic to T (T ). So by Theorem 2.5, a basis B(U ) of U can be determined by C(T (T )), i.e. B(U ) = {JX ⊗ σc c ∈ R}, where R is the set of basic relations of C(T (T )). We consider (εX ⊗ εY \{y0})T (S ≀ T )(εX ⊗ εY \{y0}). Since T (S ≀ T ) is generated by B(U ) ∪ {σs ⊗ IY , εx0s ⊗ ε{y0} s ∈ S}, (εX ⊗ εY \{y0})T (S ≀ T )(εX ⊗ εY \{y0}) is generated by {JX ⊗ σc c ∈ RY \{y0}} ∪ {σs ⊗ IY \{y0} s ∈ S}. Thus, we can determine a basis of (εX ⊗ εY \{y0})T (S ≀ T )(εX ⊗ εY \{y0}) with respect to the set of basic relations of C(T (T ))Y \{y0}. 3.2. A basis of A(Ty0) = T (Y, T, y0). By Theorem 2.5, we have A(Cy0) = T (Y, T, y0). Let W be {r ∈ Ty0 y0r = 3}. In this paper, we shall say that the point set of Ty0 is well-ordering if for all r, r′ ∈ W and t ∈ Ty0, the adjacency matrix of t ∩ (y0r × y0r′) 6= ∅ is one of the followings: (1) 1 0 0 0 1 0 0 0 1     ;   0 1 0 0 0 1 1 0 0   ;   0 0 1 1 0 0 0 1 0   , where the adjacency matrix of t ∩ (y0r × y0r′) 6= ∅ means a 3 × 3 matrix whose rows and columns are indexed by y0r and y0r′, respectively. Lemma 3.1. Let (Y, T ) be a 3-equivalenced scheme, y0 ∈ Y and u, v, w ∈ T3. Assume w ∩ (y0u × y0v) = 3. Then there exist a permutation τ on Y and a scheme (Y τ , T τ ) isomorphic to (Y, T ) such that τ Y \y0v = id and the adjacency matrix of (w ∩ (y0u × y0v))τ is the identity matrix, where id is the identity permutation. 5 Proof. If the adjacency matrix of w ∩ (y0u × y0v) is the identity matrix, then we set τ = id. Assume that the adjacency matrix of w ∩ (y0u × y0v) is not the identity matrix. Then we can choose a permutation τ on Y such that τ Y \y0v = id and the adjacency matrix of (w ∩ (y0u × y0v))τ is the identity matrix. It follows from the definition that (Y τ , T τ ) is a scheme isomorphic to (Y, T ). (cid:3) Proposition 3.2. For a given 3-equivalenced scheme (Y, T ), there exists a scheme (Y ′, T ′) isomorphic to (Y, T ) such that the point set of T ′ y0 is well- ordering. 3 Proof. Since u∗v = 2 or 3 for u, v ∈ T ×, there exists at least one basic relation t ∈ T such that t ∩ (y0u × y0v) = 3. We denote all elements of T × by t1, t2, . . . , tl, where l = Y −1 . First, we take an element t ∈ T × such that t∩(y0t1×y0t2) = 3. Replacing u, v and w by t1, t2 and t, we apply Lemma 3.1 for them. Then we can obtain a scheme (Y τ1 , T τ1) isomorphic to (Y, T ). Again we denote all elements of (T τ1)× by tτ1 2 × y0tτ1 1 , tτ1 l . We take an 3 ) = 3. Replacing u, v and 3 and t′, we apply Lemma 3.1 for them. Then we can obtain a element t′ ∈ (T τ1)× such that t′ ∩ (y0tτ1 w by tτ1 scheme (Y τ1τ2 , T τ1τ2) isomorphic to (Y τ1 , T τ1). 2 , . . . , tτ1 2 , tτ1 mi ∈ T ′ such that the adjacency matrix of t′ Repeating the above argument, we can obtain a scheme (Y ′, T ′) ≃ (Y, T ) with the following property: For each 1 ≤ i ≤ l − 1, there exists the only basic relation t′ i+1) is the identity matrix, where (T ′)× = {t′ i × y0t′ i+1) by t′ l − 1}. We can check that t′ mi The adjacency matrix of t′ t′ mi+1 · · · t′ Lemma 2.4 imply that the point set of T ′ l} and 1 ≤ mi ≤ l. y0 contains {t′ mi 1 ≤ i ≤ y0 for 1 ≤ i ≤ j ≤ l − 1. mj is the identity matrix. This and (cid:3) 2, . . . , t′ 1, t′ mi. Then T ′ mj ∈ T ′ We denote t′ t′ mi+1 · · · t′ mi ∩ (y0t′ mi ∩ (y0t′ i × y0t′ mi y0 is well-ordering. By Proposition 3.2, from now on, we assume that the point set of Ty0 is well-ordering. 3.3. Central primitive idempotents of T (X × Y, S ≀ T, (x0, y0)). Set F (t) = (X, y0t) and U (t) = (S ≀ T )F (t) for t ∈ T . For χ ∈ Irr(T (U (1Y ))) \ {1T (U (1Y ))}, define eχ = eχ ⊗ ε{y0} ∈ T (S ≀ T ). For t ∈ T3 and ψ ∈ Irr(A(S)) \ {1A(S)}, define eψ = eψ ⊗ εy0t ∈ T (S ≀ T ). It is easy to see that they are idempotents of T (S ≀ T ). Lemma 3.3 ([4], Lemma 4.2 and 4.4). For χ ∈ Irr(T (U (1Y ))) \ {1T (U (1Y ))}, eχ is a central primitive idempotent of T (S ≀ T ). 6 Lemma 3.4. For t ∈ T3 and ψ ∈ Irr(A(S)) \ {1A(S)}, eψ is a central primitive idempotent of T (S ≀ T ). Proof. First, we show that eψ commutes with σs ⊗ IY (s ∈ S), JX ⊗ σu (u ∈ T \ {1Y }), εx0s ⊗ ε{y0} (s ∈ S), and IX ⊗ εy0u (u ∈ T \ {1Y }). For s ∈ S, we have eψ(σs ⊗ IY ) = Xu∈T eψ(σs ⊗ εy0u) = (eψ ⊗ εy0t)(σs ⊗ εy0t). Since eψ commutes with σs, we have eψ(σs ⊗ IY ) = (σs ⊗ IY )eψ. Since t 6= 1Y , we have eψ(εx0s ⊗ ε{y0}) = (εx0s ⊗ ε{y0})eψ = 0. Since e1A(S) = X−1JX and e1A(S)eψ = eψe1A(S) = 0, we have eψ(JX ⊗ σu) = (JX ⊗ σu)eψ = 0. Also, eψ(IX ⊗ εy0u) = (IX ⊗ εy0u)eψ is trivial. Next, we show that eψ is primitive. The map π : T (S ≀ T ) → eψT (S ≀ T ) is a projection. Actually, eψT (S ≀ T ) is naturally isomorphic to eψA(S). Since eψ is a central primitive idempotent of A(S), eψ is primitive. (cid:3) Now we define the other central primitive idempotents of T (S ≀ T ). Put (i) ω is a cube root of unity, (ii) y0t = {yt(1), yt(2), yt(3)} and (iii) y0t′ = {yt′(1), yt′(2), yt′(3)} such that the adjacency matrix of {(yt(1), yt′(1)), (yt(2), yt′(2)), (yt(3), yt′(3))} is the identity matrix. For all t, t′ ∈ T3, define 1 Gy0t,y0t′ = 3X JX ⊗ (εy0t,y0t′ + ωεy0t,y0t′ + ω2εy0t,y0t′) and where and G′ y0t,y0t′ = 1 3X JX ⊗ (εy0t,y0t′ + ω2εy0t,y0t′ + ωεy0t,y0t′), εy0t,y0t′ := J{yt(1)},{yt′(1)} + J{yt(2)},{yt′(2)} + J{yt(3)},{yt′(3)}, εy0t,y0t′ := J{yt(1)},{yt′(2)} + J{yt(2)},{yt′(3)} + J{yt(3)},{yt′(1)}, εy0t,y0t′ := J{yt(1)},{yt′(3)} + J{yt(2)},{yt′(1)} + J{yt(3)},{yt′(2)}. It is easy to see that {Gy0t,y0t′ t, t′ ∈ T3} and {G′ y0t,y0t′ t, t′ ∈ T3} are linearly independent subsets of T (S ≀ T ). 7 Lemma 3.5. h{Gy0t,y0t′ t, t′ ∈ T3}i is an ideal and isomorphic to M atT3(C). Proof. First, we prove that σuGy0t,y0t′ , Gy0t,y0t′ σu ∈ h{Gy0t,y0t′ t, t′ ∈ T3}i for u ∈ S ≀ T . If σuGy0t,y0t′ 6= 0, then (σu)X×y0h,X×y0t 6= 0 for some h ∈ T3. This means that σu = JX ⊗ σt1 for some t1 ∈ T3 and t1 ∈ h∗t. By Lemma 2.2, h∗t = 2 or 3. This means that t1 ∩ (y0h × y0t) = 3 or 6. Note that the adjacency matrix of t1 ∩ (y0h × y0t) is a sum of at most two matrices given in (1). So, σuGy0t,y0t′ is either (JX ⊗ε1)Gy0t,y0t′ or (JX ⊗(ε1 +ε2))Gy0t,y0t′, where ε1, ε2 ∈ {εy0h,y0t, εy0h,y0t, εy0h,y0t} are distinct. By calculation, we have the followings: (JX ⊗ ε1)Gy0t,y0t′ = (JX ⊗ ε1)( 1 3X JX ⊗ (εy0t,y0t′ + ωεy0t,y0t′ + ω2εy0t,y0t′)) = 1 3 JX ⊗ ωi(εy0h,y0t′ + ωεy0h,y0t′ + ω2εy0h,y0t′) for some 0 ≤ i ≤ 2; (JX ⊗ (ε1 + ε2))Gy0t,y0t′ = (JX ⊗ (ε1 + ε2))( 1 3X JX ⊗ (εy0t,y0t′ + ωεy0t,y0t′ + ω2εy0t,y0t′)) = 1 3 JX ⊗ (ωi + ωj)(εy0h,y0t′ + ωεy0h,y0t′ + ω2εy0h,y0t′) for some distinct 0 ≤ i, j ≤ 2. Thus, σuGy0t,y0t′ is either XωiGy0h,y0t′ or X(ωi + ωj)Gy0h,y0t′ for some distinct 0 ≤ i, j ≤ 2. Whichever the case may be, it implies σuGy0t,y0t′ ∈ h{Gy0t,y0t′ t, t′ ∈ T3}i. Similarly, we can show Gy0t,y0t′ σu ∈ h{Gy0t,y0t′ t, t′ ∈ T3}i. For u ∈ S ≀ T and t, t′ ∈ T3, clearly ε(x0,y0)uGy0t,y0t′ = δ(x0,y0)u(X×y0t)Gy0t,y0t′ ∈ h{Gy0t,y0t′ t, t′ ∈ T3}i and Gy0t,y0t′ ε(x0,y0)u ∈ h{Gy0t,y0t′ t, t′ ∈ T3}i. Thus, h{Gy0t,y0t′ t, t′ ∈ T3}i is an ideal. Next, we prove that Gy0t,y0t′ Gy0t′′′,y0t′′ = δt′t′′′ Gy0t,y0t′′. It suffices to show that Gy0t,y0t′ Gy0t′,y0t′′ = Gy0t,y0t′′. By calculation, we have Gy0t,y0t′ Gy0t′,y0t′′ = ( 1 3X JX ⊗ (εy0t,y0t′ + ωεy0t,y0t′ + ω2εy0t,y0t′)) ( 1 3X JX ⊗ (εy0t′,y0t′′ + ωεy0t′,y0t′′ + ω2εy0t′,y0t′′ )) JX ⊗ 3(εy0t,y0t′′ + ωεy0t,y0t′′ + ω2εy0t,y0t′′) = Gy0t,y0t′′ . = 1 9X Finally, we prove that h{Gy0t,y0t′ t, t′ ∈ T3}i ∼= M atT3(C). For t, t′ ∈ T3, let ett′ be a T3 × T3 matrix whose (t, t′)-entry is 1 and whose other entries are all zero. Then the linear map ϕ : h{Gy0t,y0t′ t, t′ ∈ T3}i → M atT3(C) defined by ϕ(Gy0t,y0t′) = ett′ is an isomorphism. (cid:3) 8 Similarly to the proof of Lemma 3.5, we can prove the following lemma. Lemma 3.6. h{G′ y0t,y0t′ t, t′ ∈ T3}i is an ideal and isomorphic to M atT3(C). Define and where and eη1 = Xt∈T3 1 3X JX ⊗ (εy0t + ωεy0t + ω2εy0t) eη2 = Xt∈T3 1 3X JX ⊗ (εy0t + ω2εy0t + ωεy0t), εy0t := J{yt(1)},{yt(2)} + J{yt(2)},{yt(3)} + J{yt(3)},{yt(1)} εy0t := J{yt(1)},{yt(3)} + J{yt(2)},{yt(1)} + J{yt(3)},{yt(2)}. Then, by Lemma 3.5 and 3.6, eη1 and eη2 are central primitive idempotents of T (S ≀ T ). Lemma 3.7. The sum of e1T (S≀T ), eχ's, eψ's, eη1 and eη2 is the identity element. Proof. It is easy to see that e1T (S≀T ) = e1 T (U (1Y )) ⊗ ε{y0} + Xt∈T3 1 3X εF (t)JX×Y εF (t), Xχ∈Irr(T (U (1Y )))× eχ = εF (1Y )IX×Y εF (1Y ) − e1 T (U (1Y )) ⊗ ε{y0}, Xψ∈Irr(A(S))× eψ = εF (t)IX×Y εF (t) − 1 X (JX×yt(1) + JX×yt(2) + JX×yt(3)) for each t ∈ T3, and eη1 = Xt∈T3 1 3X JX ⊗ (εy0t + ωεy0t + ω2εy0t) eη2 = Xt∈T3 1 3X JX ⊗ (εy0t + ω2εy0t + ωεy0t). Thus, we have e1T (S≀T ) + Xχ∈Irr(T (U (1Y )))× eχ + Xt∈T3 Xψ∈Irr(A(S))× eψ + eη1 + eη2 = IX×Y . 9 (cid:3) 4. Main result In conclusion, we have determined all central primitive idempotents of Terwilliger algebras of wreath products by 3-equivalenced schemes. Com- bining Section 3 and [4, Theorem 4.1] gives the following theorem. Theorem 4.1. Let (X, S) be a scheme and (Y, T ) a 3-equivalenced scheme. Fix x0 ∈ X and y0 ∈ Y , and consider the wreath product (X × Y, S ≀ T ). Then (i) If T = 2, then {e1T (S≀T )} ∪ {eχ χ ∈ Irr(T (U (1Y )))×} {eψ ψ ∈ Irr(A(S))×} ∪ [t∈T3 is the set of all central primitive idempotents of T (X×Y, S≀T, (x0, y0)). (ii) If T > 2, then {e1T (S≀T )} ∪ {eχ χ ∈ Irr(T (U (1Y )))×} {eη1 , eη2 } ∪ [t∈T3 {eψ ψ ∈ Irr(A(S))×} is the set of all central primitive idempotents of T (X×Y, S≀T, (x0, y0)). References [1] G. Bhattacharyya, S.Y. Song, R. Tanaka, Terwilliger algebras of wreath products of one-class association schemes, J. Algebr. Comb. 31 (2010) 455 -- 466. [2] S. Evdokimov, I. Ponomarenko, Permutation group approach to association schemes, European J. Combin. 30 (2009) 1456 -- 1476. [3] , On primitive cellular algebras, Zap. Nauchn. Sem. S.-Peterburg. Otdel. Mat. Inst. Steklov. (POMI) 256 (1999) 38 -- 68, translation in J. Math. Sci. (New York) 107 (2001), 4172-4191. [4] A. Hanaki, K. Kim, Y. Maekawa, Terwilliger algebras of direct and wreath products of association schemes, J. Algebra 343 (2011) 195 -- 200. [5] M. Hirasaka, K.-T. Kim, J.R. Park, Every 3-equivalenced association scheme is Frobe- nius, J. Algebr. Comb. (in press). [6] K. Kim, Terwilliger algebras of wreath products by quasi-thin schemes, Linear Alge- bra Appl. 437 (2012) 2773 -- 2780. [7] M. Muzychuk, I. Ponomarenko, On pseudocyclic association schemes, Ars Math. Contemp. 5 (2012) 1 -- 25. [8] R. Tanaka, Classification of commutative association schemes with almost commuta- tive Terwilliger algebras, J. Algebr. Comb. 33 (2011) 1 -- 10. [9] R. Tanaka, P.-H. Zieschang, On a class of wreath products of hypergroups and asso- ciation schemes, J. Algebr. Comb. 37 (2013) 601 -- 619. [10] P. Terwilliger, The subconstituent algebra of an association scheme I, J. Algebr. Comb. 1 (1992) 363 -- 388. [11] [12] , The subconstituent algebra of an association scheme II, J. Algebr. Comb. 2 (1993) 73 -- 103. , The subconstituent algebra of an association scheme III, J. Algebr. Comb. 2 (1993) 177 -- 210. [13] B. Xu, K. Zuo, On semisimple varietal Terwilliger algebras whose non-primary ideals are 1-dimensional, J. Algebra 397 (2014) 426 -- 442. [14] P.-H. Zieschang, Theory of association schemes, Springer monographs in mathemat- ics, Springer, Berlin, 2005. 10 Department of Mathematics, Pusan National University, Busan 609-735, Re- public of Korea E-mail address: [email protected] 11
1212.0615	2	1212	2013-07-31T18:38:36	Alternative algebras admitting derivations with invertible values and invertible derivations	[ "math.RA" ]	We descibed all alternative algebras with invertible derivations (the analogue of Bergen-Herstein-Lanski's Theorem) and proved the analogue of Moens's Theorem.	math.RA	math	Alternative algebras admitting derivations with invertible values and invertible derivations. 1 Ivan Kaygorodova,b, Yury Popovb,c a Instituto de Matem´atica e Estat´ıstica, Universidade de Sao Paulo, Brasil, b Sobolev Institute of Mathematics, Novosibirsk, Russia, c Novosibirsk State University, Russia. Contents Introduction. 1. 2. Basic definitions and identities. 3. Alternative algebras with derivations with invertible values. 4. A characterization of nilpotent alternative algebras by invertible Leibniz-derivations. 5. Alternative and Jordan algebras with QDer = End. References 1 1 3 7 9 10 1. Introduction. The notion of derivation with invertible values as a derivation of a ring with unity that takes only multiplicatively invertible or zero values appeared in [1]. Bergen, Herstein and Lanski determined the structure of associative rings that admit derivations with invertible values. Later, the results of this paper were generalized in [2]–[6]. Another interesting type of derivations are invertible derivations. The definition of an invertible derivation as an invertible mapping first arose in [7], where the nilpotency of a Lie algebra admitting an invertible derivation was proved. The research on that topic was then continued in [8, 9]. Nowadays, a great interest is shown in the study of nearly associative algebras and superalgebras with derivations. For example, works [10, 11] determine the structure of differentiably simple alternative and Jordan algebras, and papers [12]– [19] give the description of generalizations of derivations of simple and semisimple alternative, Jordan and structurable (super)algebras. Nevertheless, the problem of specification of algebras from classical non-associative varieties (such as alternative, Jordan, structurable, etc.), admitting derivations with invertible values and invertible derivations, remains unconsidered. The present work is to make up this gap. 2. Basic definitions and identities. We are using standard notation: (x, y, z) := (xy)z − x(yz) — the associator of elements x, y, z, [x, y] := xy − yx — the commutator of elements x, y, x ◦ y := xy + yx — the Jordan product of elements x, y. An algebra A is called alternative (see [20] for more information on alternative algebras), if A satisfies the following identities: It’s easy to check that in any alternative algebra the associator is a skew-symmetric function of its arguments, and the flexible identity x(yx) = (xy)x holds. It’s also well known [20, p.35] that every alternative algebra satisfies the middle Moufang identity: (xy)(zx) = x(yz)x. A commutative algebra J is called Jordan if it satisfies the Jordan identity: (x, x, y) = 0, (x, y, y) = 0. (x2, y, x) = 0. 1The authors were supported by RFBR 12-01-31016, 12-01-33031, by RF President Grant council for support of young scientists and leading scientific schools (project MK-330.2013.1), and FAPESP (Grant 2011/51132-9). 1 2 It is widely known that if A is an alternative algebra, then vector space A with new multiplication a ◦ b is a Jordan algebra which we will denote by A(+). The nucleus of an algebra A is the set N (A) = {n ∈ A (n, A, A) = (A, n, A) = (A, A, n) = (0)} , the commutative center of A is the set K(A) = {k ∈ A [k, A] = [A, k] = (0)} , and the center of A is Z(A) = N (A) ∩ K(A). Derivation d is called inner if it lies in the smallest subspace of the space of all linear operators on A containing all right and left multiplications by elements of A and closed under commutation. Otherwise d is called outer. In studying the structure of alternative algebras, one class is of great importance: Cayley–Dickson algebras. The definition and properties of Cayley–Dickson algebras and the Cayley–Dickson process can be found, for instance, in [20]. It’s known that every Cayley–Dickson algebra C over field F is 8–dimensional, non-associative, alternative, simple and has an unit element. Also, C is quadratic over F , that is, for every x ∈ C the following relation holds: x2 − t(x)x + n(x) = 0, (1) where t(x), n(x) ∈ F, t(x) is a F -linear mapping, and n(x) is a strictly nondegenerate quadratic form satisfying n(xy) = n(x)n(y) for all x, y ∈ C. A Cayley–Dickson algebra is also equipped with a symmetric bilinear nondegenerate form f (x, y) = n(x + y) − n(x) − n(y). For a subset M ⊆ C, by M ⊥ we will denote the orthogonal complement to M with respect to f. A Cayley–Dickson algebra containing zero divisors is called split. It’s known [20, p.43] that element x of a split Cayley–Dickson algebra is invertible if and only if n(x) 6= 0. It’s also known [20, p.46] that every split Cayley–Dickson algebra over field F is isomorphic to a Cayley–Dickson matrix algebra C(F ), comprising matrices of the form a = (cid:18)α u v β(cid:19), where α, β ∈ F , u, v ∈ F 3. Addition and scalar multiplication of elements of the algebra C(F ) will then correspond to the usual addition and scalar multiplication of matrices. However, multiplication of elements of the algebra C(F ) will correspond to the following matrix multiplication: (cid:18)α u v β(cid:19) ·(cid:18)γ t w δ(cid:19) = (cid:18) αβ + (u, w) γv + βw + u × t αt + δu − v × w βγ + (v, t) (cid:19), where for vectors x = (x1, x2, x3), y = (y1, y2, y3) ∈ F 3, by (x, y) = x1y1 + x2y2 + x3y3 we denote their dot product, and by x × y = (x2y3 − x3y2, x3y1 − x1y3, x1y2 − x2y1) their cross product. Under given representation t(a) = α + β, n(a) = αβ − (u, v). In the case when char(F ) 6= 2, C can be obtained from F by applying the Cayley–Dickson process thrice to F with the identical involution and parameters α, β, γ ∈ F . We will not go into the full details here and will only provide the formula that defines multiplication in algebra B = A + vA obtained by the Cayley–Dickson process from algebra A with involution ¯ : (a1 + vb1)(a2 + vb2) = (a1a2 + γb2b1) + v(a1b2 + a2b1), where ai, bi ∈ A, v2 = γ ∈ F . We will also need the following statement, which describes simple alternative non-associative algebras. Theorem 1. Let A be a simple non-associative alternative algebra. Then the center of the algebra A is a field and A is a Cayley–Dickson algebra over its center. 3. Alternative algebras with derivations with invertible values. 3 Let A be an algebra with unit element 1 over field F . We will denote by U the set of invertible elements of A. In this section we will only consider derivations with invertible values, by which we understand such non–zero derivations d that for every x ∈ A, d(x) ∈ U or d(x) = 0 holds. In 1983, Bergen, Herstein and Lanski initated the study whose purpose is to relate the structure of a ring to the special behavior of one of its derivations. Namely, in their article [1] they described associative rings admitting derivations with invertible values. They proved that such ring must be either a division ring, or the ring of 2 × 2 matrices over a division ring, or a factor of a polynomial ring over a division ring of characteristic 2. They also characterized those division rings such that a 2 × 2 matrix ring over them has an inner derivation with invertible values. Further, associative rings with derivations with invertible values (and also their generalizations) were discussed in variety of works (see, for instance, [2]–[6]). So, in [2], semiprime associative rings with involution, allowing a derivation with invertible values on the set of symmetric elements, were given an examination. In work [3] Bergen and Carini determined the associative rings admitting a derivation with invertible values on some non–central Lie ideal. Also, in papers [4] and [5] the structure of associative rings that admit α-derivations with invertible values and their natural generalizations — (σ, τ )-derivations with invertible values — was described. And in paper [6] Komatsu and Nakajima described associative rings that allow generalized derivations with invertible values. The purpose of this section is to generalize the results of Bergen, Herstein and Lanski to the alternative case. In this part, A is an alternative algebra with unit element 1 and derivation with invertible values d. The following lemmas were proved in [1] for associative algebras and can be easily generalized to the alternative case with minor differences, but in order to ensure the complteness of the narration we shall provide their proofs. Lemma 2. If d(x) = 0, then either x = 0, or x is invertible. Proof. Let’s notice [20, p.204], that in every alternative algebra the following identity holds: (a−1, a, b) = 0. (2) It’s then easy to see that in an arbitrary alternative algebra the product of two invertible elements is also invertible. Using identity (2), for invertible a and b we find (b−1a−1)(ab) = a−1((ab)b−1) − (a−1(ab))b−1 + (b−1a−1)(ab) = −(a−1, ab, b−1) + (b−1a−1)(ab) = −(b−1, a−1, ab) + (b−1a−1)(ab) = b−1(a−1(ab)) = 1. Assume that x 6= 0. Since d 6= 0, there exists y ∈ A such that d(y) ∈ U . Hence d(yx) = d(y)x ∈ U and d(y)−1d(yx) = x. In view of d(y) and d(yx) being invertible, x is also invertible. The lemma is proved. Now we shall study the ideal structure of A: Lemma 3. a) If L 6= 0 is a one–sided ideal in A then d(L) 6= (0). b) If I is a proper one–sided ideal of A, then I is both minimal and maximal. c) If I is a proper ideal of A then I 2 = (0). d) If char(A) 6= 2 then A is simple. Proof. a) Since the statement is obvious when L = A, we should only consider the case when L 6= A. If 0 6= a ∈ L, then, by lemma 2, d(a) 6= 0, since a is not invertible. b) It suffices to show that every proper one–sided ideal in A is maximal. Let I ⊂ J be a proper one–sided ideal in A. It’s easy to see that d(I) ∩ I = (0) and I ⊕ d(I) is also an one–sided ideal in A. By lemma 3(a), d(I) 6= (0), hence d(I) contains invertible elements, in consequence of which I ⊕ d(I) = A. For arbitrary j ∈ J we have j = a + d(b), a, b ∈ I. Consequently, d(b) = j − a ∈ J ∩ d(I) = (0); thus j = a ∈ I. c) If I 6= A is an ideal of A, then d(I 2) ⊂ d(I)I + Id(I) ⊂ I, consequently, by lemma 3(a), I 2 = (0), since the product of two ideals in an alternative algebra is also an ideal [20, p.115] and I does not contain any invertible elements. d) Let 2A 6= 0 and I 6= (0). Then, by lemma 3(a), there exists b ∈ I such that d(b) ∈ U . Since b2 = 0, 0 = d2(b2) = d2(b)b + 2d(b)2 + bd2(b), 4 and consequently 2d(b)2 ∈ I. Now, since d(b) is invertible, d(b)2 is also invertible and 2d(b)2 = 0, therefore 2 = 0. We have obtained a contradiction which proves the lemma. By Der(A) we will denote the set of all derivations of algebra A. Let us fix some subset D ⊆ Der(A). The ideal I is called a D–ideal, if for all ∂ ∈ D, x ∈ I we have ∂(x) ∈ I. Algebra A is called D–simple if A2 6= 0 and A contains no proper D–ideals (for more detailed information on D–simple algebras see [10, 11] and their references). As an immediate consequence of lemma 3(a) we have Lemma 4. If alternative algebra A admits a derivation with invertible values d, then A is d–simple. Now, if char(A) 6= 2 we can apply lemma 3(d) and theorem 1 and conclude that A is either an associative or Cayley– Dickson algebra over its center. We will now consider the non–simple non–associative case, which is examined in Lemma 5. If A is not simple and not associative, then A = C[x]/(x2), where C is a Cayley–Dickson algebra over its center Z(C), C is a division algebra, char(C) = 2, d(C) = 0, d(x) = 1 + ax for some a ∈ Z(C), and d is an outer derivation. Proof. Combining lemma 3(b) and (d), we have char(A) = 2, I 2 = (0) for any proper ideal I in A and all proper one–sided ideals in A are both minimal and maximal. Consequently, we can easily deduce that A contains a unique (left, right, two–sided) ideal M and M 2 = 0. Therefore, as in the proof of lemma 3(b), we have A = M ⊕ d(M ), particularly, for any a ∈ A there exist m, n ∈ M such that d(a) = m + d(n). Hence m = d(a − n) ∈ M ∩ d(A) = (0) and so, denoting C = ker(d), we have A = C + M . By lemma 2, C is a division algebra, therefore A = C ⊕ M . We define linear mappings λ : M → C and µ : M → M by d(m) = λ(m) + µ(m) for any m ∈ M . It’s easy to notice that for any a ∈ C, b ∈ M the following holds: aµ(b) + aλ(b) = ad(b) = d(ab) = µ(ab) + λ(ab), where aµ(b), µ(ab) ∈ M and consequently aλ(b) = λ(ab) ∈ λ(M ); similarly λ(ba) = λ(b)a ∈ λ(M ). This implies that λ(M ) is an ideal in C. Since C is simple and λ(M ) 6= (0) we derive that C is isomorphic to M as a left C–module. Putting x = λ−1(1), we have A = C ⊕ Cx. Using the fact that λ is a module isomorphism, it’s easy to see that [x, C] = 0. Considering the identity 3(k, x, y) = 3(y, k, x) = 3(x, y, k) = [xy, k] − x[y, k] − [x, k]y = 0, satisfied for any k ∈ K(B), x, y ∈ B in arbitrary alternative algebra B [20, p.136], and taking into account the structure of A we deduce that x ∈ Z(A). Therefore we have A ∼= C[x]/(x2). Now nonassociativity of A and theorem 1 imply that C is a Cayley–Dickson algebra over its center Z(C). We can write µ(x) = ax for some a ∈ C. Now, since x ∈ Z(A) and char(A) = 2, for arbitrary c ∈ C we have: 0 = d(cx + xc) = c(1 + ax) + (1 + ax)c = cax + axc = (ca + ac)x. Since C is a division algebra, we obtain ca + ac = 0, thus a ∈ Z(C). Finally, since every ideal of A is invariant under the action of any inner derivation, x ∈ M , and d(x) /∈ M , it is clear that d is not inner. The lemma is proved. Theorem 6. Let A be an alternative algebra with unit element 1, admitting derivation with invertible values d. Then: 1) A is an associative algebra and one of the following conditions holds: a) A is a division algebra D; b) A is a 2 × 2 matrix algebra M2(D) over division algebra D; c) A is a factor–algebra of polynomial algebra D[x]/(x2) over division algebra D; furthermore, char(D) = 2, d(D) = 0 and d(x) = 1 + ax for some a in the center of D, and d is an outer derivation; 2) A is a non–associative alternative algebra and one of the following conditions holds: a) A is a Cayley–Dickson algebra over its center Z(A); b) A is a factor–algebra of polynomial algebra C[x]/(x2) over a Cayley–Dickson division algebra; furthermore, char(C) = 2, d(C) = 0 and d(x) = 1 + ax for some a in the center of C, and d is an outer derivation. Proof. The associative case follows from [1], and the non-associative case follows from theorem 1, lemmas 3 and 5. Now, to complete the characterization of alternative algebras allowing derivations with invertible values we only have to describe split Cayley–Dickson algebras with derivations with invertible values, which is done in the following. 5 Lemma 7. An algebra C, which is a split Cayley–Dickson algebra over its center Z, admits a derivation with invertible values d if and only if one of the following conditions holds: I) C is obtained by means of the Cayley–Dickson process from its associative division subalgebra B: C = B + vB, v2 = γ ∈ Z, γ 6= 0, where B = ker(d) and dimZB = 4. Furthermore, in this case an arbitrary derivation with invertible values d is of the form d(a + vb) = v(bu), where a, b ∈ B and u ∈ B is a fixed element with t(u) = 0. II) C can be represented as a direct sum: C = B + xB, where t(x) = 0, B = ker(d), B is a subfield of C, B = B⊥ and dimZB = 4. Furthermore, in this case an arbitrary derivation with invertible values d is of the form d(a + xb) = b, where a, b ∈ B. Proof. It’s generally known (see, for example, [21]) that every derivation of C is inner. It’s easy to see then that Z ⊆ ker(d) and d is a Z–linear mapping. Therefore we will consider C as a Z–algebra. Suppose that C allows a derivation with invertible values d. Take a subspace V ⊂ C such that dimZV = 4 and V does not contain invertible elements. For example (taking into account that C ∼= C(F ) — the Cayley–Dickson matrix algebra over F ), we can take V = (cid:26)(cid:18)α u 0(cid:19) α ∈ Z, u ∈ Z 3 (cid:27) . 0 From lemma 2 it follows that dimZd(V ) = 4 and V ∩ d(V ) = (0), hence C = V ⊕ d(V ). In particular, for any x ∈ C there exist u, v ∈ V such that d(x) = u + d(v). Consequently, u = d(x − v) ∈ V ∩ d(A) = (0), and, denoting B = ker(d), we have C = B + V . By lemma 2, B is a division algebra, thus C = B ⊕ V and dimZB = 4. Combining the facts that B is simple, Z(C) ⊆ Z(B) and applying theorem 1 we have that B is an associative subalgebra in C. By [20, p.39], in C the following relation is valid: Putting b = d(a), we obtain Applying d on (1), we have a ◦ b − t(a)b − t(b)a − f (a, b) = 0. a ◦ d(a) − t(a)d(a) − t(d(a))a − f (a, d(a)) = 0. a ◦ d(a) − t(a)d(a) = 0. (3) (4) (5) Subtracting (4) from (5), we obtain t(d(a))a + f (a, d(a)) = 0. If a and 1 are linearly independent over Z, then we have f (a, d(a)) = 0. (6) In the case when a ∈ Z, then a ∈ ker(d) and relation (6) is then obvious. Linearizing (6), we obtain f (a, d(b))+f (d(a), b) = 0. Consequently, since B = ker(d), for arbitrary a ∈ C we have f (d(a), B) = −f (a, d(B)) = 0, and so d(C) ⊆ B⊥. We now have to study two cases: (I) If the restriction of the form f on B is nondegenerate, then, by [20, p.32, Th.1], C can be obtained from B by means of the Cayley–Dickson process, that is, C = B + vB, v2 = γ 6= 0, B⊥ = vB. Particularly, d(v) = vu for some u ∈ B, and therefore for arbitrary a, b ∈ B we have d(a + vb) = d(v)b = (vu)b = v(bu). By [20, p.26], for any x, y, w ∈ C we have And for x = v, y = 1, w = u, using (6), we obtain n(x)f (y, w) = f (xy, xw). 0 = f (v, vu) = n(v)t(u). Since v2 = γ ∈ Z, γ 6= 0, then n(v) 6= 0 and t(u) = 0. (II) Now, let the restriction of the form f on B be degenerate. Hence there exists 0 6= b ∈ B such that f (b, B) = 0. Therefore 0 = f (b, b) = 2n(b). Since b is invertible, n(b) 6= 0 and we must have char(C) = 2. By [20, p.26], in C the following relation holds: f (x, z)f (y, w) = f (xy, zw) + f (xw, yz). (7) Putting into (7) x = b, z = a, y = b−1c, w = 1, where a, c ∈ B, we have and so by arbitrariness of a, c we conclude that f (B, B) = 0, that is, B ⊆ B⊥. 0 = f (b, a)f (b−1c, 1) = f (c, a) + f (b, ab−1c), 6 Now we shall show that opposite inclusion also takes place: Suppose for a moment that there exists x ∈ B⊥, x /∈ B. By (2) and skew–symmetry of the associator, dimZxB = 4 and A = B ⊕ xB. Using (7), we have f (a, xc) = f (a · 1, xc) = −f (ac, x) + f (a, x)f (1, c) = 0 for any a, c ∈ B. Consequently, xB ⊂ B⊥ and C = B⊥, which contradicts the nondegeneracy of the form f . We put x = d−1(1). It’s obvious that x /∈ B and C = B ⊕ xB. Relation (6) implies that 0 = f (x, 1) = t(x). Now we only have to prove that B is a field. By the definition of f and the fact that f (B, B) = 0, for any a, c we have 0 = f (a, c) = n(a + c) − n(a) − n(c), which means that n is a ring homomorphism from B to Z. In view of B being simple, together with n(1) = 1, ker(n) = 0, and we conclude that B is a subfield of Z. Conversely, suppose that condition (I) holds, that is C is obtained from B by means of the Cayley–Dickson process. Let 0 6= u be an element of B such that t(u) = 0. Consider the mapping d : a + vb 7→ v(bu), where a, b ∈ B. We are to show that d is a derivation. Indeed, for any a1, b1, a2, b2 ∈ B we have d(a1 + vb1)(a2 + vb2) + (a1 + vb1)d(a2 + vb2) = γ(b2(u + u)b1) + v((a2b1 + a1b2)u) = γ(b2t(u)b1) + v((a2b1 + a1b2)u) = v((a2b1 + a1b2)u) = d((a1a2 + γb2b1) + v(a1b2 + a2b1)) = d((a1 + vb1)(a2 + vb2)). Also, n(d(a + vb)) = n(v(bu)) = n(v)n(b)n(u) = −γn(b)n(u) 6= 0 if b 6= 0, since B is a division algebra. Hence d(a + vb) is invertible for any a ∈ B, 0 6= b ∈ B, so d takes invertible values. Now assume that condition (II) holds. Consider the mapping d : a + xb 7→ b. We are to show that d is a derivation with invertible values. Since we have B = B⊥, then for any a ∈ B, t(a) = 0 holds. Combining (3) and char(C) = 2, we obtain (8) particularly, d([x, a]) = 0. Substituting x in (1), we deduce that x2 ∈ Z. Using (8), it’s easy to check that for a, c ∈ B the following identity holds: [x, a] = x ◦ a = t(a)x + t(x)a + f (a, x) = f (a, x) ∈ Z, (a, c, x) = af (c, x) + f (a, x)c + f (x, ac), (9) and consequently, d((a, c, x)) = 0. Now we will prove that d is a derivation. For arbitrary a, b, c, h ∈ B we have d((ax + b)(cx + h)) = d((ax)(cx) + (ax)h + b(cx) + bh) = d((ax)(cx)) + d((ax)h) + d(b(cx)). Consider the last two summands: d((ax)h) = d((xa)h) = d(x(ah)) = ah, d(b(cx)) = d((bc)x) = bc. On the other hand, d(ax + b)(cx + h) + (ax + b)d(cx + h) = a(cx + h) + (ax + b)c = a(cx) + ah + (ax)c + bc. Therefore we need to show that d((ax)(cx)) = a(cx) + (ax)c. Transforming the corresponding expressions, we have: a(cx) + (ax)c = (ac)x + (a, c, x) + a(xc) + (a, x, c) = (ac)x + a(cx + f (c, x)) = (a, c, x) + af (c, x). Using the middle Moufang identity, we obtain d((ax)(cx)) = d((xa + f (a, x))cx) = d((xa)(cx)) + f (a, x)d(cx) = d(x(ac)x) + f (a, x)c = d(x(x(ac) + f (x, ac))) + f (a, x)c = d(x2(ac)) + f (x, ac) + f (a, x)c = f (x, ac) + f (a, x)c, since x2 ∈ Z and d(x2(ac)) = n(x)d(ac) = 0. Equating the expressions, we will arrive at the relation (9), which, as was shown earlier, holds identically. Therefore d is a derivation of C. Since d takes values in B, which is a field, it’s obvious that d is a derivation with invertible values. The lemma is proved. Example. In work [22] an example of a split Cayley–Dickson algebra C which has a subfield of dimension 4 was provided. Let’s consider an imperfect field F of characteristic 2 and elements α, β ∈ F such that α, β, αβ are linearly independent over F 2. Then subalgebra B of matrix Cayley–Dickson algebra C(F ), generated by elements is a subfield of C, and dimF B = 4. (cid:18) 0 (1, 0, 0) (α, 0, 0) 0 (cid:19), (cid:18) 0 (0, 1, 0) (0, β, 0) 0 (cid:19), 4. A characterization of nilpotent alternative algebras by invertible Leibniz-derivations. 7 In 1955, Jacobson [7] proved that a Lie algebra over a field of characteristic zero admitting a non–singular (invertible) derivation is nilpotent. The problem of whether the inverse of this statement is correct remained open until work [23], where an example of nilpotent Lie algebra, whose derivations are nilpotent (and hence, singular), was constructed. Such types of Lie algebras are called characteristically nilpotent Lie algebras. The study of derivations of Lie algebras leads to the appearance of the notion of their natural generalization — a pre- derivation of a Lie algebra, which is a derivation of a Lie triple system induced by that algebra. In [8] it was proved that Jacobson’s result is also true in terms of pre-derivations. Several examples of nilpotent Lie algebras whose pre-derivations are nilpotent were presented in [8], [24]. In paper [9] a generalization of derivations and pre-derivations of Lie algebras is defined as a Leibniz-derivation of order k. Moens proved that a Lie algebra over a field of characterisic zero is nilpotent if and only if it admits an invertible Leibniz-derivation. After that, Fialowski, Khudoyberdiyev and Omirov [25] showed that with the definition of Leibniz-derivations from [9] the similar result for non-Lie Leibniz algebras is not true. Namely, they gave an example of a non–nilpotent Leibniz algebra which admits an invertible Leibniz-derivation. In order to extend the results of paper [9] for Leibniz algebras they introduced a definition of Leibniz–derivations of Leibniz algebras which agrees with the case of Leibniz-derivations of Lie algebras, and proved that a Leibniz algebra is nilpotent if and only if it admits an invertible Leibniz-derivation. It should be noted that there exist non-nilpotent Filippov (n-Lie) algebras with invertible derivations (see [26]). Also, in [27] a generalization of pre-derivations of associative algebras was considered. The main purpose of this section is to prove the analogue of Moens’s theorem for alternative algebras. Throughout the section all spaces of algebras are assumed finite-dimensional over a field of characteristic zero. Definition. A Leibniz-derivation (by Moens) of order n for an algebra A is an endomorphism φ of that algebra satisfying the identity φ((. . . (x1x2) . . .)xn) = n Xi=1 (. . . (. . . (x1x2) . . . φ(xi)) . . .)xn. Theorem 8. An alternative algebra over a field of characteristic zero is nilpotent if and only if it has an invertible Leibniz-derivation. Proof. Let A be a finite–dimensional alternative algebra with an invertible Leibniz-derivation φ of order n and β(A) be the nilpotent radical of A (it’s also widely known that in the finite-dimensional case it coincides with rad(A), the solvable radical of A). Using [20], we can establish that A/β(A) can be represented as finite sum of its minimal ideals, where each of them is either a full matrix algebra over some division ring or a Cayley–Dickson algebra over its center. Therefore, algebra A/β(A) possesses unit element 1. We will regard A as a direct sum: A = As + β(A), where As is a semisimple alternative algebra isomorphic to A/β(A) (Wedderburn–Malcev decomposition). Using the idea of the proof from [9] we shall prove that φ(β(A)) ⊆ β(A). We will remark that in the case when φ is a derivation it was proved for all algebras with locally nilpotent radical in [28]. Step 1. We define on vector space A the structure of n-ary algebra An with multiplication Hence φ is a derivation of n-ary algebra An. We shall show that solvable radicals rad(An) and rad(A) of algebras An and A coincide. It’s clear that rad(A) ⊆ rad(An). Consider the natural projection π : A → As. It’s easy to see that π(rad(An)) is a solvable ideal in As: applying π to the both sides of relation [a1, a2, . . . , an]n = (. . . (a1a2) . . .)an. and using the fact that As has a unit, we have [A, . . . , rad(An), . . . , A]n ⊆ rad(An), π(rad(An))As + Asπ(rad(An)) ⊆ π(rad(An)). Consequently, since As is semisimple, we have π(rad(An)) = 0. Step 2. We will now show that φ(β(A)) ⊆ β(A). Let β(A) = τ = τ1 = rad(An) and τt+1 = [τt, τt, . . . , τt]n. Then we have τ = τ1 ) τ2 ) . . . ) τp ) τp+1 = 0. Since the product of two ideals in an alternative algebra is also an ideal, then τt is an ideal in An for any t. 8 We need to show that φi(τt) ⊆ τ holds for any i. We use induction on t. The induction base is trivial for t = p + 1. Now let’s suppose that φi(τt+1) ⊆ τ for arbitrary i. We need to prove that the inclusion φi(τt) ⊆ τ holds for any i. The set τ + φ(τt) is a solvable ideal of An, since [A, . . . , A, τ + φ(τt), A, . . . , A]n ⊆ τ + τt + φ(τt) = τ + φ(τt) and [τ + φ(τt), . . . , τ + φ(τt)]n ⊆ τ + [φ(τt), . . . , φ(τt)]n ⊆ τ + φn(τt+1) ⊆ τ. Now we are to show that τ + φk(τt) is a solvable ideal of An for any k. Suppose that φi(τt) ⊆ τ for each 1 6 i < k. Using the induction hypothesis, we have [A, . . . , A, τ + φk(τt), A, . . . , A]n ⊆ τ + φ([A, . . . , A, φk−1(τt), A, . . . , A]n) +X[A, . . . , φ(A), . . . , A, φk−1(τt), A, . . . , A]n) ⊆ . . . ⊆ τ + Xa0+...+an−1=k,ai≥0 φa0 ([φa1 (A), . . . , φal−1 (A), τt, φal (A), . . . , φan−1 (A)]n) + φk([A, . . . , A, τt, A, . . . , A]n) ⊆ τ + k−1 Xi=0 φi(τt) + φk(τt) = τ + τt + φk(τt) = τ + φk(τt) and [τ + φk(τt), . . . , τ + φk(τt)]n ⊆ τ + [φk(τt), . . . , φk(τt)]n ⊆ τ + φkn(τt+1) ⊆ τ. Therefore, φi(τt) ⊆ τ and φ(τ ) ⊆ τ. Step 3. Considering the fact that φ is an invertible mapping, that is, it has a trivial kernel, we conclude that φ(A/β(A)) = A/β(A), which contradicts the unitality of A/β(A), since φ(1) = nφ(1) and dim(φ(A/β(A))) < dim(A/β(A)). This contradiction implies that A = β(A), that is, A is nilpotent. Step 4. The converse also takes place: in order to see that nilpotent alternative algebra A with nilpotency index s 2 ] + 1 it suffices to consider the sum of vector spaces A = W + An and has an invertible Leibniz–derivation of order n = [ s linear mapping φ, defined this way: It is easy to see that φ is a Leibniz-derivation of A of order n. The theorem is proved. Further, it’s easy to check that φ(x) = (cid:26) x, if x ∈ W, nx, if x ∈ An. Remark 9. Over a field of positive characteristic there exist a nilpotent alternative algebra possessing only singular derivations. Proof. Non–associative alternative nilpotent algebras of dimension not greater than 7 were classified in [29]. Using this classification, over a field of characteristic 3 we define a 7–dimensional algebra A with basis {e1, e2, e3, u1, u2, v, w} by this mulptiplication table (all other products are zero): 1 = u1, e2 e2 e1u1 = u1e1 = v, e2u2 = u2e2 = w, e1v = ve1 = u2 2 = u2, e2e3 = e3e2 = −v, e3e1 = u2, e1e3 = u2, 1 = w. It’s easy to notice that A2 = hu1, u2, v, wi, A3 = hv, wi, A4 = hwi. Then for any derivation D of A we have D(v) = D(e1e1e1) ∈ A4 and D(w) = D(e1e1e1e1) ∈ A4, then there exist x 6= 0 such that D(x) = 0. We also note that the theorem only holds for characteristic zero: Remark 10. For an arbitrary alternative algebra over a field of positive characteristic p the identity map is a Leibniz- derivation of order p + 1. Remark 11. The free alternative algebra with n generators admits an invertible derivation, but it is not nilpotent. 9 Proof. It suffices to consider a derivation which acts identically on the generators of the algebra. Remark 12. Theorem 8, Remarks 10 and 11 take place in case of associative algebras too. Remark 13. Following the article [28] and using methods provided in the proof of Theorem 8 we can show that a finite-dimensional Jordan algebra admitting an invertible derivation is nilpotent. 5. Alternative and Jordan algebras with QDer = End. Following Leger and Luks [30], we call an additive mapping f a quasiderivation if there exists a linear map Q such that Q(xy) = f (x)y + xf (y). Leger and Luks described all finite–dimensional Lie algebras in which an arbitrary endomorphism is a quasiderivation. They found out that such an algebra is either an abelian Lie algebra, a two–dimensional solvable Lie algebra or a three–dimensional simple Lie algebra. Later on, this result was generalized for Lie superalgebras in [31]. Also, quasiderivations and generalized derivations were studied in [32]–[42] and in other works. Here we shall describe all finite-dimensional Jordan and alternative algebras over arbitrary fields of characteristic not equal to 2, in which every endomorphism is a quasiderivation. Theorem 14. Let J be a finite–dimensional Jordan algebra over a field of characteristic not equal to 2 and such that QDer(J) = End(J). Then either J is a field or J has zero multiplication. Proof. Let f be an endomorphism of algebra J. Hence there exists Qf such that Qf (xy) = f (x)y + xf (y). Suppose that 0 6= x ∈ Ann(J) = {a ∈ Ja · J = 0}. Then for arbitrary y ∈ J we have f (x)y = Qf (xy) − xf (y) = Qf (xy) = 0. Since f (x) can be any element of algebra J, then either Ann(J) = 0 or J has zero multiplication. Suppose for a moment that there exists x such that x2 = 0. Then for any f ∈ End(J) we have 2f (x)x = Q(x2) = 0 and by the above argument J has zero multiplication. It’s now easy to notice, if we suppose that J has a nonzero nilpotent radical, then considering that J is a power–associative algebra we obtain that there exists a nilpotent element of index 2. Therefore we conclude that J has a trivial nilpotent radical and consequently J can be represented as a direct sum of simple Jordan algebras. Notice (see, for example, [15]) that every direct summand is f –invariant, thus from now on J will be regarded as a simple unital Jordan algebra. Description of quasiderivations of simple finite-dimensional Jordan algebras [15] implies that f can be represented as a sum of a scalar mapping and a derivation of algebra J. We notice that if we denote by Ra the operator of right multiplication by element a ∈ J, then RJ + Der(J) is a subspace in End(J). Furthermore, RJ ∩ Der(J) = 0, since for every unital algebra 1Ra = a and dim(RJ ) = dim(J). But dim(Der(J)) 6= dim(J)2 − 1, if dim(J) 6= 1. That is, we can conclude that J is a field. The theorem is now proved. Theorem 15. Let A be a finite-dimensional alternative algebra over a field of characteristic not equal to 2 and such that QDer(A) = End(A). Then either A is a field or A has zero multiplication. Proof. It suffices to notice that if f is a quasiderivation of algebra A then f is a quasiderivation of algebra A(+). Furthermore, if A is an alternative algebra then A(+) is a Jordan algebra. Consequently, if A satisfies the condition QDer(A) = End(A), then A(+) satisfies this condition too, therefore it is either a field or a zero multiplication algebra. Therefore we can conclude that A is either a field or an anticommutative algebra. It’s generally known that an alternative anticommutative algebra is nilpotent, that is, it has nontrivial annihilator Ann(A). So, if we have 0 6= x ∈ Ann(A) then for any y ∈ A we have f (x)y = 0 and by the above argument A has zero multiplication. The theorem is proved. Remark 16. Theorem 15 for associative algebras can be proved as consequence of the result of Leger and Luks [30]. It suffices to notice that a quasiderivation f of algebra A is also a quasiderivation of algebra A(−), and to consider associative commutative algebras and associative algebras A such that A(−) is either a two–dimensional solvable Lie algebra or a three–dimensional simple Lie algebra. Acknowledgements. The authors are grateful to Prof. Ivan Shestakov (IME-USP, Brazil) and Prof. Alexandre Pozhidaev (Sobolev Inst. of Math., Russia) for interest and constructive comments, and Mark Gannon (IME-USP, Brazil) for the translation. References 10 [1] Bergen J., Herstein I. N., Lanski C., Derivations with invertible values, Canad. J. Math., 35 (1983), 2, 300–310. [2] Giambruno A., Misso P., Milies P. C., Derivations with invertible values in rings with involution, Pac. J. Math., 123 (1986), 1, 47–54. [3] Bergen J., Carini L., Derivations with invertible values on a Lie ideal, Canad. Math. Bull., 31 (1988), 1, 103–110. [4] Chang J. C., α-derivations with invertible values, Bull. Inst. Math. Acad. Sinica, 13 (1985), 4, 323–333. [5] Hongan M., Komatsu H., (σ, τ )–derivations with invertible values, Bull. Inst. Math. Acad. Sinica, 15 (1987), 4, 411–415. [6] Komatsu H., Nakajima A., Generalized derivations with invertible values, Comm. Algebra, 32 (2004), 5, 1937–1944. [7] Jacobson N., A note on automorphisms and derivations of Lie algebras, Proc. Amer. Math. Soc., 6 (1955), 281–283. [8] Bajo I., Lie algebras admitting non-singular pre-derivations, Indag. Math. (N.S.), 8 (1997), 4, 433–437. [9] Moens W. A., A characterisation of nilpotent Lie algebras by invertible Leibniz-derivations, Comm. Alg., (2013), - . [10] Popov A., Differentiably simple alternative algebras, Algebra and Logic, 49 (2010), 5, 456–469. [11] Popov A., Differentiably simple Jordan algebras, Siberian Math. J., 54 (2013), , [12] Filippov V. T., On δ-derivations of prime alternative and Malcev algebras, Algebra and Logic, 39 (2000), 5, 354–358. [13] Kaygorodov I., On δ-derivations of simple finite-dimensional Jordan superalgebras, Algebra and Logic, 46 (2007), 5, 318–329. [14] Kaygorodov I., δ-superderivations of simple finite-dimensional Jordan and Lie superalgebras, Algebra and Logic, 49 (2010), 2, 130–144. [15] Shestakov A., Ternary derivations of separable associative and Jordan algebras, Siberian Math. J., 53 (2012), 5, 943–956. [16] Shestakov A., Ternary derivations of simple Jordan superalgebras, Algebra and Logic, 2014, ??. [17] Kaygorodov I., Zhelyabin V., On δ-superderivations of simple superalgebras of Jordan brackets, St.-Peterburg Math. J., 23 (2012), 4, 40–58. [18] Kaygorodov I., On δ-superderivations of semisimple finite-dimensional Jordan superalgebras, Mathematical Notes, 91 (2012), 2, 187–197. [19] Kaygorodov I., Okhapkina E., On δ-derivations of semisimple finite-dimensional structurable algebras, J. of algebra and its applications (to appear), arXiv:1212.0618 [20] Zhevlakov K. A., Slinko A. M., Shestakov I. P., Shirshov A. I., Rings that are nearly associative, Pure and Applied Mathematics, 104, Academic Press, Inc., New York-London, 1982. [21] Elduque A., On triality and automorphisms and derivations of compositional algebras, Linear Algebra and its Applications, 314 (2000), 49–74. [22] Gagola S.M, Maximal subalgebras of the octonions, J. of Pure and Appl. Algebra, 217 (2013), 1, 20–21. [23] Dixmier J., Lister W. G., Derivations of nilpotent Lie algebras, Proc. Amer. Math. Soc., 8 (1957), 155–158. [24] Burde D., Lie Algebra prederivations and strongly nilpotent Lie algebras, Comm. Algebra, 30 (2002), 7, 3157–3175. [25] Fialowski A., Khudoyberdiyev A. Kh., Omirov B. A., A characterization of nilpotent Leibniz algebras, Algebras and Repesentation Theory, doi: 10.1007/s10468-012-9373-z [26] Williams M. P., Nilpotent n-Lie algebras, Comm. Algebra, 37 (2009), 6, 1843–1849. [27] Kaygorodov I., Jordan δ-derivations of associative algebras, Journal of Mathematical Sciences (New York), (2013), , . [28] Slinko A. M., A remark on radicals and derivations of rings, Siberian Math. J., 13 (1972), 6, 984–986. [29] Badalov M. I., Nilpotent alternative algebras, Algebra and Logic, 23 (1984), 3, 167–181. [30] Leger G., Luks E., Generalized derivations of Lie algebras, J. Algebra, 228 (2000), 1, 165–203. [31] Zhang R., Zhang Y., Generalized derivations of Lie superalgebras, Comm. Alg. 38 (2010), 10, 3737–3751. [32] Chen L., Ma Y., Ni L., Generalized Derivations of Lie Color Algebras, Results in Mathematics, 63 (2013), 3-4, 923–936. [33] Kaygorodov I., (n + 1)-Ary derivations of simple n-ary algebras, Algebra and Logic, 50 (2011), 5, 477–478. [34] Kaygorodov I., (n + 1)-Ary derivations of simple n-ary Malcev algebras, St.-Peterburg Math. J., 25 (2014), 4, . [35] Kaygorodov I., (n + 1)-Ary derivations of semisimple Filippov algebras, arXiv:1110.0926 [36] Zusmanovich P., On δ-derivations of Lie algebras and superalgebras, J. of Algebra, 324 (2010), 12, 3470–3486. [37] Burde D., Dekimpe K., Post-Lie algebra structures and generalized derivations of semisimple Lie algebras, Moscow Math. Journal, 13 (2013), 1, 1–18. [38] Filippov V. T., On δ-derivations of Lie algebras, Siberian Math. J., 39 (1998), 6, 1218–1230. [39] Filippov V. T., δ-derivations of prime Lie algebras, Siberian Math. J., 40 (1999), 1, 174–184. [40] Kaygorodov I., δ-derivations of algebras and superalgebras, Proceedings of the International Conference on Algebra 2010, World Sci. Publ., Hackensack, NJ (2012), 374–380. [41] Kaygorodov I., On δ-derivations of classical Lie superalgebras, Siberian Math. J., 50 (2009), 3, 434–449. [42] Kaygorodov I., δ-derivations of n-ary algebras, Izvestiya: Mathematics, 76 (2012), 6, 1150–1162.
1710.03083	3	1710	2018-05-14T10:18:10	The equation solvability problem over nilpotent Mal'cev algebras	[ "math.RA" ]	By a result of Horv\'ath the equation solvability problem over finite nilpotent groups and rings is in P. We generalize his result, showing that the equation solvability over every finite supernilpotent Mal'cev algebra is in P. We also give an example of a nilpotent, but not supernilpotent Mal'cev algebra, whose identity checking problem is coNP-complete.	math.RA	math	THE EQUATION SOLVABILITY PROBLEM OVER SUPERNILPOTENT ALGEBRAS WITH MAL'CEV TERM MICHAEL KOMPATSCHER Abstract. In 2011 Horv´ath gave a new proof that the equation solvability problem over finite nilpotent groups and rings is in P. In the same paper he asked whether his proof can be lifted to nilpotent algebras in general. We show that this is in fact possible for su- pernilpotent algebras with a Mal'cev term. However, we also describe a class of nilpotent, but not supernilpotent algebras with Mal'cev term that have coNP-complete identity check- ing problems and NP-complete equation solvability problems. This proves that the answer to Horv´ath's question is negative in general (assuming P6=NP). 1. Introduction One of the oldest problems in algebra is to decide whether an equation over a given algebraic structure has a solution. In the last decades this problem has received increasing attention from a computational complexity point of view; in particular for finite algebras the aim is to identify conditions that either imply tractability or hardness of the corresponding equation solvability problem. Many results are known for finite groups and rings, of which several are based on commutator theory. For rings a complexity dichotomy holds: In [3] it was shown that the equation solvability problem over non-nilpotent rings is NP-complete, by [8] the problem is in P for nilpotent rings. Nilpotency is also a source of tractability in the group case: It was proven in [7] (and reproven in [8]) that the equation solvability over nilpotent groups is in P. By [12] non- solvable groups induce NP-complete problems. Furthermore it was shown in [14] that every solvable, non-nilpotent group has a polynomial extension, whose equation solvability problem is NP-complete. However it is still open whether a complexity dichotomy like over rings holds. In particular nilpotency does not demark the border between problems in P and NP-complete: By [15] the equation solvability over the non-nilpotent group A4 is in P but its extension by the commutator [·, ·] has an NP-complete equation solvability problem. More general, meta- abelian groups [10] and semipattern groups [4] induce equation solvability problems that are in P, while not necessarily being nilpotent. Congruence permutable varieties generalize both the varieties of groups and rings and are well-studied in the context of commutator theory. It is hence natural to ask, whether the above dichotomy results can be generalized to all congruence permutable varieties. It was already observed in [1] that the identity checking problem for supernilpotent such algebras is in P. Also our results indicate that not nilpotency, but supernilpotency is the right notion to work with: 2010 Mathematics Subject Classification. 08A70, 08B10, 08A50, 68Q17. Key words and phrases. equation solvability; supernilpotency; Mal'cev term; congruence permutable vari- ety; identity checking. Supported by Charles University Research Centre program No.PRIMUS/SCI/12 and No.UNCE/SCI/022 as well as grant 18-20123S of the Czech Grant Agency (GA CR). ORCID iD: 0000-0002-0163-6604. 1 2 MICHAEL KOMPATSCHER In Section 2 we show that the equation solvability problems over finite supernilpotent algebras with Mal'cev term are in P. As a corollary of our proof we obtain a new characterization of supernilpotent algebras in congruence permutable varieties. In Section 3 we give examples of nilpotent, but not supernilpotent algebras (of infinite type) that have a Mal'cev term and induce coNP-complete identity checking problems and NP-complete equation solvability problems. The following two subsections provide some necessary preliminary definitions and facts about nilpotent and supernilpotent algebras. 1.1. Equation solvability and identity checking. We are going to denote algebras by bold characters and their domain by the corresponding non-bold character (e.g. A is an algebra on the set A). A polynomial over an algebra A is a term that is built from variables and elements of A using the operation symbols of A. We write Pol(A) for the set of all polynomials over A and Poln(A) for the set of polynomials of arity n. We say two algebras on the same domain A are polynomially equivalent if their polynomials induce the same operations on A. The (polynomial) equation solvability problem over an algebra A, short pEq(A), asks whether or not two polynomials f (¯x), g(¯x) over A can attain the same value for some substitu- tion over A. In other words, for input f (¯x), g(¯x) the question is whether A = ∃¯xf (¯x) = g(¯x). The (polynomial) identity checking problem over A, short pId(A), asks whether a given equation is satisfied under all substitution of the variables by elements of A. So, for two input polynomials f (¯x), g(¯x) the question is whether A = ∀¯xf (¯x) = g(¯x). In literature the identity checking problem is sometimes also referred to as equivalence problem. When phrasing pEq(A) and pId(A) as actual computational problems, it is not obvious how to encode the input. However, when we are studying finite algebras of finite type, then an input term can just be encoded by the string defining it. Hence the size of an input polynomial p(¯x) is proportional to its length l(p(¯x)), i.e. the total number of functions, constants and variable symbols it contains. We remark that this encoding might not be optimal, in the sense that it does not take into account that some expressions might be repeatedly used in the definition of a term. An alternative would be to describe terms by algebraic circuits. This approach was suggested by Ross Willard; not much is known for this encoding. Also, for some algebras it makes sense to restrict the input to terms of a certain canonical form (for instance [18], [9] and [11] study the sum of monomials over rings); this is also something we are not considering here. For discussions on the size of term representation in supernilpotent algebras, see also [2]. 1.2. Nilpotent algebras with Mal'cev term. A ternary term m over an algebra A is called a Mal'cev term if it satisfies m(y, x, x) = m(x, x, y) = y for all x, y ∈ A. It is well- known that an algebra has a Mal'cev term if and only if it is from a congruence permutable variety. In this section we provide some results on nilpotent and supernilpotent algebras that have a Mal'cev term. For background on commutator theory we refer to [6], for a survey on higher commutators we refer to [1]. Let A be an algebra and α1, . . . , αn be congruence relations of A. Then δ = [α1, . . . , αn] denotes the smallest congruence relation of A such that for all polynomials f (¯x1, . . . , ¯xn) ¯bi we have that, f (¯a1, ¯x2, . . . , ¯xn) ≡δ and for all tuples ¯a1, ¯b1, . . . , ¯an, ¯bn in A with ¯ai ≡αi f (¯b1, ¯x2 . . . , ¯xn) for all (x2, . . . , xn) ∈Qn f (¯b1, ¯b2 . . . , ¯bn). i=2{ai, bi}\{(b2, . . . , bn)}, implies that f (¯a1, ¯b2, . . . , ¯bn) ≡δ n {z ] = 0A. } n+1 {z } THE EQUATION SOLVABILITY PROBLEM OVER SUPERNILPOTENT ALGEBRAS WITH MAL'CEV TERM3 Definition 1.1. Let A be an algebra and let 1A denote the total equivalence relation and 0A the identity on A. Then: • A is called nilpotent (of degree n) if [1A, [1A, . . . , [1A, [1A, 1A]] . . .]] ] = 0A. • A is called supernilpotent (of degree n) if [1A, 1A, . . . , 1A We remark that, for algebras with Mal'cev term, supernilpotency of degree n implies nilpo- tency of degree n. For groups and rings the two notions are equivalent, this is however not true in general. Every nilpotent algebra with Mal'cev term gives rise to loop operations, where a loop is defined as follows: Definition 1.2. An algebra L = (L, ·, \, /, 0) is called a loop if for all x, y ∈ L: (1) x\(x · y) = y and (y · x)/x = y (2) x · (x\y) = y and (y/x) · x = y (3) 0 · x = x · 0 = x Then the following holds: Theorem 1.3 (Chapter 7 of [6]). Let A be a nilpotent algebra with a Mal'cev term m(x, y, z). For each 0 ∈ A the operation defined by x · y = m(x, 0, y) is a loop multiplication with neutral element 0. Also the left and right inverse operations \ and / can be defined as polynomials over A. (cid:3) In other words every nilpotent algebra A = (A, F ) with Mal'cev term is polynomially equivalent to a nilpotent loop (A, ·, \, /) expanded by additional operations F . We denote this expansion by (A, ·, \, /). In order to further give a characterization of supernilpotent algebras we introduce the following notation: Definition 1.4. Let f ∈ Polm(A). We say f (x1, . . . , xm) absorbs (a1, . . . , am) to a if f (b1, . . . , bm) = a, whenever bi = ai for some i. We say f (x1, . . . , xm) is a-absorbing if f absorbs (a, a, . . . , a) to a. Then the following holds: Theorem 1.5 (Proposition 6.16. in [1]). Let A be an algebra with Mal'cev term and let 0 ∈ A. Then A is supernilpotent of degree n if and only if every 0-absorbing c ∈ Poln A is equivalent to 0. (cid:3) Theorems 1.5 and 1.3 were used in [19] to give a canonical representation of polynomials in supernilpotent algebras with Mal'cev term. In the proof of our main result in the next section we will recapitulate Wires' proof and slightly refine it. 2. Equation solvability in supernilpotent algebras with Mal'cev term In this section we show that the polynomial equation solvability problem pEq(A) is in P for supernilpotent A with Mal'cev term. The main ingredient for this is Lemma 2.3, which states that computing the range of a polynomial expression p(¯x) over A requires only to check substitutions of ¯x for which the number of non 0 entries is bounded by some constant d. In fact, we can show that this interpolation property is equivalent to supernilpotency for finite algebras with Mal'cev term. 4 MICHAEL KOMPATSCHER We start with some basic observations. By Theorem 1.3 we know that A has polynomials that define loop operations ·, \, / and 0. Clearly pEq(A) reduces to the equation solvability problem over the expanded algebra (A, ·, \, /, 0). Hence without loss of generality we can assume that A contains the loop operations given by Theorem 1.3. (Note however, that we do not know a priori, whether (A, ·, \, /, 0) and A have the same complexity up to polynomial time.) In every algebra A with loop operations, f (¯x) = g(¯x) is equivalent to f (¯x)/g(¯x) = 0. Hence for both for the equation solvability problem and the identity checking problem we can restrict our input to equations of the form f (x1, . . . , xm) = 0. Also note that f (x1, . . . , xm) = 0 holds for all x1, . . . , xm if and only if none of the equations f (x1, . . . , xm)/a = 0 for 0 6= a ∈ A has a solution. Hence if A is finite, the complement of It is however open if this holds for finite pId(A) reduces to pEq(A) in polynomial time. algebras in general, see also Problem 1 in [13]. It was already observed in [1] that the polynomial identity checking problem is in P for supernilpotent algebras with Mal'cev term, hence our result can be seen as a strengthening of that. We are going to stick to the following notation: Let us writeQn left associated product (· · · ((x1 · x2) · x3) · · · xn) and let xn =Qn [n] = {1, . . . , n} and (cid:0)[n] or elements of A and S ⊆ [n], let us write ¯xS for the n-tuple, where the i-th entry is equal to xi if i ∈ S and 0 otherwise, and let us write ¯x↾S for the S-tuple (xi)i∈S. If for instance ¯x = (x1, x2, x3, x4, x5) and S = {1, 2, 5} then ¯xS = (x1, x2, 0, 0, x5) and ¯x↾S = (x1, x2, x5). k(cid:1) = {S ⊆ [n] : S = k}. For a tuple ¯x = (x1, . . . , xn) of variables i=1 xi or x1 · x2 · · · xn for the i=1 x. For n ∈ N let us write Our proof is going to rely on a representation of polynomials f (x1, . . . , xm) ∈ Pol(A) as the product of S-ary 0-absorbing terms tS(¯x↾S) for all subsets S ⊆ [m]. That such representations exist was already known (see for instance Theorem 3.8 of [19]), but we are going to give a slightly more restrictive version that depends on a given enumeration of the elements of A and a tuple ¯b ∈ Am: Lemma 2.1. Let A be a finite nilpotent algebra with Mal'cev term, let 0 ∈ A and f (x1, . . . , xm) ∈ Polm(A). Furthermore let a1, a2, . . . , aN be an enumeration of the elements of A and ¯b ∈ Am. i=0 ri(x1, . . . , xm), where the Then f (x1, . . . , xm) is equivalent to a polynomial of the form Qm terms ri(x1, . . . , xm) are given by the recursion r0(x1, . . . , xm) = f (0, 0, . . . , 0) and (1) (2) ri(¯xS)! \f (¯xS) for S ∈(cid:0) [m] k+1(cid:1), tS(¯x↾S) = k Yi=0 Yi=0 YS∈( [m] N k+1) tS (¯b↾S )=ai rk+1(x1, . . . , xm) = tS(¯x↾S), for 0 ≤ k < m. For all S ⊆ [m] the S-ary terms tS(¯x↾S) are 0-absorbing. If A is moreover supernilpotent of degree n, then all terms rk(x1, . . . , xm) for k ≥ n are equivalent to 0. Before we give the proof of Lemma 2.1 we would like to point out that this representation is not unique: depending on the ordering of the sets S ∈ (cid:0) [m] might get different representations of f (x1, . . . , xm). However the main reason for us to show Lemma 2.1 is not to represent f in a canonical way, but to show that for every T ⊆ [m], f (¯bT ) k+1(cid:1) with tS(¯b↾S) = ai in (2) we THE EQUATION SOLVABILITY PROBLEM OVER SUPERNILPOTENT ALGEBRAS WITH MAL'CEV TERM5 is evaluated to a product of powers of the elements of A f (¯bT ) = n−1 Yi=1 aβi,1 1 · aβi,2 2 · · · a βi,N N , where n is the degree of supernilpotency of A. This fact will be essential in the proof of Lemma 2.3. Proof of Lemma 2.1. We are going to prove the following: For every 0 ≤ k ≤ m and every S ∈(cid:0)[m] (3) k (cid:1) we have that tS(¯x↾S) is 0-absorbing and Yi=0 f (¯xS) = k ri(¯xS). We prove the claim by induction on k. For k = 0 we have r0(x1, . . . , xm) = f (0, . . . , 0) and t∅ = 0, which clearly satisfies the claim. So let us consider the induction step k → k + 1. We first show that for every S ∈(cid:0) [m] term tS(¯x↾S) is 0-absorbing: For that, let S′ be a proper subset of S. Then, by the definition of tS(¯x↾S) in (1) and the induction hypothesis (3) for S′ we have k+1(cid:1) the tS((¯x↾S)S ′) = k Yi=0 ri(¯xS ′)! \f (¯xS ′) = f (¯xS ′)\f (¯xS ′) = 0. In order to show (3) for S note that if we evaluate rk+1 at ¯xS, all factors in (2) except for tS(¯x↾S) are equivalent to 0, since they are 0-absorbing. Therefore Hence tS is 0-absorbing for every S ∈ (cid:0) [m] k+1(cid:1). ri(¯x↾S)! · tS(¯x↾S) = k Yi=0 ri(¯xS) = k Yi=0 Yi=0 k+1 ri(¯xS )! · k Yi=0 ri(¯xS)! \f (¯xS)! = f (¯xS), which proves the claim for k + 1. Thus we proved our claim. The first part of the Lemma follows from (3) for m = k. If A is moreover supernilpotent of degree n all the terms tS(¯x↾S) for S ≥ n are equivalent to 0, since they are 0-absorbing (cf. Theorem 1.5). Therefore also all terms rk(x1, . . . , xm) are equivalent to 0 for k ≥ n. (cid:3) We are going to use Lemma 2.1 together with the following iterated version of Ramsey's theorem to prove Lemma 2.3. Theorem 2.2 (Ramsey's theorem). Let n, k and l be positive integers. Then there exists a positive integer d = d(n, k, l), such that for all sets S with S ≥ d and for all k-colorings γ of the ≤ n-elements subsets of S, there exists H ⊆ S with H = l such that all subsets of H of the same size have the same color. Lemma 2.3. Let A be a finite algebra with Mal'cev term that is supernilpotent of degree n and let 0 ∈ A. Then there exists a positive integer d = d(A) such that for every m ≥ d, for every polynomial f ∈ Polm(A) and for every ¯b = (b1, . . . , bm) ∈ Am there exists a set T ∈(cid:0)[m] d (cid:1) with f (¯bT ) = f (¯b). 6 MICHAEL KOMPATSCHER Proof. We follow the proof steps of the analogous result for nilpotent groups in [8, Lemma the smallest positive integer such that xe = 0 for 3.1]. Let e be the exponent of A, i.e. all x ∈ A and let l = e · (n − 1)! and k = en·A. We then claim that the Ramsey number d = d(n − 1, k, l) given by Theorem 2.2 satisfies the Lemma. First recall the representation result in Lemma 2.1. For a given enumeration a1, . . . aN of the elements of A it gives us n−1 f (¯b) = Yi=1 where αi,j is the number of sets S ∈ (cid:0)[m] subset of [m] and H c = [m] \ H. Since all of the terms tS(¯x↾S) are 0-absorbing, we have the same representation for f (¯bH c), i.e. i (cid:1), such that tS(¯b↾S) = aj. Let H be an arbitrary αi,1 a 1 · a αi,2 2 · · · · · a αi,N N , n−1 aβi,1 1 2 βi,N N , · · · · · a · aβi,2 f (¯bH c) = Yi=1 i (cid:1), S ⊆ H c, such that tS(¯b↾S) = aj. where βi,j is the number of sets S ∈(cid:0)[m] We claim that there is a non-empty set H such that the corresponding exponents βi,j are If H c ≤ d, we equal to αi,j modulo e. set T = H c and are done. Otherwise we can find such T by iterating the procedure for the H c-ary polynomial defined by f (¯xH c). Hence it only remains to prove this claim. If this is the case, we have that f (¯b) = f (¯bH c). For every set I ⊆ [m] let γi,j(I) denote the number of all set S ∈(cid:0)[m] and I ⊆ S. By the inclusion-exclusion principle we have for a given H that i (cid:1) such that tS(¯b↾S) = aj αi,j − βi,j = − X∅6=I⊆H (−1)Iγi,j(I) = − X∅6=I⊆H I<n I≤i (−1)Iγi,j(I). are divisible by e. The function γ(I) := (γi,j(I))i∈[n],j∈[N ] is a coloring of subsets of [m] with k colors. By Theorem 2.2 there is an H with H = l such that γ is monochromatic on all We show that there is an H such that all summands of the form PI⊆H,I=s γi,j(I) for s < n s(cid:1) divides PI⊆H,I=s γi,j(I) for every subsets of size at most n − 1 of H. This implies that (cid:0)l s(cid:1) is divisible by e for every s < n. Hence also s < n. Since l = e · (n − 1)!, we know that (cid:0)l αi,j − βi,j is divisible by e, which concludes the proof. (cid:3) We remark that Lemma 2.3 gives us a characterization of finite supernilpotent algebra with Mal'cev term: Corollary 2.4. Let A be a finite algebra with Mal'cev term and 0 ∈ A. Then A is supernilpo- tent if and only if there is a positive integer d such that for every polynomial f (x1, . . . , xm) ∈ Pol(A) and every tuple ¯r ∈ Am there is an index set T ⊆ [m] with T ≤ d and f (¯r) = f (¯rT ). Proof. It only remains to show that if A is not supernipotent, it does not have the inter- polation property described above. By Theorem 1.5 for every d ∈ N there is a 0-absorbing polynomial f (x1, . . . , xd, xd+1) and a tuple ¯r ∈ Ad+1 such that f (¯r) 6= 0. But as f is 0- absorbing, f (¯rT ) = 0 holds for every T ⊆ [d + 1], T 6= [d + 1]. (cid:3) Lemma 2.3 now implies our main result: THE EQUATION SOLVABILITY PROBLEM OVER SUPERNILPOTENT ALGEBRAS WITH MAL'CEV TERM7 Theorem 2.5. Let A be a finite supernilpotent algebra with Mal'cev term. Then the equation solvability problem for A can be decided in polynomial time. Proof. By the discussion at the beginning of this section we only have to consider equations of the form f (x1, . . . , xm) = 0 as input. By Lemma 2.3, there is a solution ¯b with f (¯b) = 0 if and only if f (¯bT ) = 0 for some T with T = min(d, m), where d is a constant only depending on A. For m ≥ d there are Ad ·(cid:0)m f on all those tuples and checking whether the result is 0 takes polynomial time O(l(f (¯x))d) and yields whether f (x1, . . . , xm) = 0 is solvable. (cid:3) d(cid:1) = O(md) many tuples of the form ¯bT . Hence evaluating We remark that due to the use of Ramsey's theorem the value of d in Lemma 2.3 might be very large; we can only obtain upper bounds that are superexponential in A. The best known algorithm for nilpotent groups G runs in polynomial with exponent 1 2 G2 log(G) and is due to Foldv´ari [5]. This indicates that our algorithm in Theorem 2.5 might be far from being optimal. 3. Nilpotent algebras with hard equation solvability and identity checking problems For every prime p let Ap = (Z p2, +, 0, −, (fn)n∈N) be the cyclic group of order p2, together with the n-ary operations fn(x1, . . . , xn) = p · x1 · x2 · · · xn for every n ∈ N. In this section we are going to show that pEq(Ap) is NP-complete and pId(Ap) is co-NP-complete for every p > 2. It is easy to see that Ap is nilpotent of degree 2 for every p; the equivalence classes of [1Ap, 1Ap] are exactly the cosets of Zp. Furthermore it follows straightforward from Theo- rem 1.5 that Ap is not supernilpotent, since every function fn(x1, . . . , xn) is 0-absorbing but not equivalent to 0. However we remark that every restriction of Ap to finitely many of its operators gives us a supernilpotent algebra. It is a priori not clear how to encode the input when phrasing pId(Ap) as computational problem, since Ap is of infinite type. However, in every arity there is exactly one operation fn so one can still find a reasonable such encoding of terms, i.e. one where the size of an input polynomial f (¯x) is linear in its length l(f (¯x)). With respect to such an encoding the following holds: p2, +, 0, −, (fn)n∈N) with fn(x1, . . . , xn) = Theorem 3.1. Let p be an odd prime and let Ap = (Z p·x1·x2 · · · xn. Then the equation solvability problem pEq(Ap) is NP-complete and the identity checking problem pId(Ap) is co-NP-complete. Proof. We prove that pEq(Ap) is NP-complete by reducing the graph p-colorability problem to it. To do so, for every instance of a graph G = (V, E) we define the term: tG((xv)v∈V ) = f(p−1)·E (xv1 − xv2)(v1,v2)∈E i∈[p−1] ! = p · Y(v1,v2)∈E (xv1 − xv2)p−1 Note that the order of edges is irrelevant and for a tuple (rv)v∈V in Z p2 the value of tG((rv)v∈V ) only depends on the cosets of rv with respect to Zp. Moreover tG((rv)v∈V ) = 0 holds if and only if there is an edge (v1, v2) ∈ E such that rv1 and rv2 are in the same coset of Zp; otherwise tG((rv)v∈V ) = p. Thus, if the equation tG((xv)v∈V ) = p has a solution (rv)v∈V , then the coloring that assigns to each vertex v the color rv Zp is a proper coloring of the graph G with p colors. Conversely 8 MICHAEL KOMPATSCHER every coloring of G with p many colors induces a solution of the equation (by assigning to every color a unique coset of Zp). Thus p-colorability reduces to pEq(Ap), which consequently is NP-complete. Analogously a graph is not p-colorable if and only if tG((xv)v∈V ) = 0 holds for all values (cid:3) of (xv)v∈V . Thus pId(Ap) is coNP-complete. We conclude with the question, whether this hardness result fits into a bigger context. By [14] and [12] every non-nilpotent group has a polynomial extension, whose identity checking problem is co-NP-complete. By [3] also for rings this statement is true. Therefore we ask: Question 3.2. Does every non-supernilpotent finite algebra with Mal'cev term have a poly- nomial extension, whose • identity checking problem is co-NP-complete? • equation solvability problem is NP-complete? A first step in answering Question 3.2 would be to study the question for nilpotent, but not supernilpotent algebras. In this case we have much structural information to work with, due to Theorem 1.3 and Theorem 1.5. Note that Question 3.2 might have different answers, depending on the encoding of the input (see also the discussion in Section 1.1), and also de- pending on whether we restrict ourselves to algebras of finite type or not, as in our example. Recent progress: After the submission of this article it came to the authors attention that Idziak and Krzaczkowski independently proved Theorem 2.5 in their paper [16], where they studied the equation solvability problem and the identity checking problem in the more general setting of algebras from congruence modular varieties. Moreover they proved several hardness results that partially answer Question 3.2: By their work, every non-solvable algebra A with a Mal'cev term has polynomial extensions with hard pEq and pId problems. Furthermore every solvable, but non-nilpotent algebra A with Mal'cev term has a quotient for which Question 3.2 has a positive answer. However in the nilpotent, but not supernilpotent case, the situation seems to be more complicated [17]: There are 2-nilpotent, but non-supernilpotent algebras of finite type such that every extension of it by finitely many polynomials has tractable equation solvability and identity checking problem (even if the input polynomials are encoded by circuits). However an extension of these algebras by infinitely many polynomials induced hardness of both problems as in the example of Theorem 3.1. The author would like to thank Jakub Oprsal for introducing him to (higher) commutators and giving several other helpful remarks. Acknowledgments References [1] E. Aichinger and N. Mudrinski. Some applications of higher commutators in Mal'cev algebras. Algebra universalis, 63(4):367 -- 403, 2010. [2] E. Aichinger, N. Mudrinski, and J. Oprsal. Complexity of term representations of finitary functions. arXiv preprint arXiv:1709.01759, 2017. [3] S. Burris and J. Lawrence. The equivalence problem for finite rings. Journal of Symbolic Computation, 15(1):67 -- 71, 1993. [4] A. Foldv´ari. The complexity of the equation solvability problem over semipattern groups. International Journal of Algebra and Computation, 27(02):259 -- 272, 2017. THE EQUATION SOLVABILITY PROBLEM OVER SUPERNILPOTENT ALGEBRAS WITH MAL'CEV TERM9 [5] A. Foldv´ari. The complexity of the equation solvability problem over nilpotent groups. Journal of Algebra, 495:289 -- 303, 2018. [6] R. Freese and R. McKenzie. Commutator theory for congruence modular varieties, volume 125. CUP Archive, 1987. [7] M. Goldmann and A. Russell. The complexity of solving equations over finite groups. Information and Computation, 178(1):253 -- 262, 2002. [8] G. Horv´ath. The complexity of the equivalence and equation solvability problems over nilpotent rings and groups. Algebra universalis, 66(4):391 -- 403, 2011. [9] G. Horvath. The complexity of the equivalence problem over finite rings. Glasgow Mathematical Journal, 54(1):193 -- 199, 2012. [10] G. Horv´ath. The complexity of the equivalence and equation solvability problems over meta-abelian groups. Journal of Algebra, 433:208 -- 230, 2015. [11] G. Horv´ath, J. Lawrence, and R. Willard. The complexity of the equation solvability problem over finite rings. preprint on http://real.mtak.hu/28210/, 2015. [12] G. Horv´ath, L. M´erai, C. Szab´o, and J. Lawrence. The complexity of the equivalence problem for non- solvable groups. Bulletin of the London Mathematical Society, 39(3):433 -- 438, 2007. [13] G. Horv´ath and C. Szab´o. The complexity of checking identities over finite groups. International Journal of Algebra and Computation, 16(5):931 -- 939, 2006. [14] G. Horv´ath and C. Szab´o. The extended equivalence and equation solvability problems for groups. Discrete Mathematics & Theoretical Computer Science, 13(4):23 -- 32, 2011. [15] G. Horv´ath and C. Szab´o. Equivalence and equation solvability problems for the alternating group A4. J. Pure Appl. Algebra, 2012. [16] P. M. Idziak and J. Krzaczkowski. Satisfiability in multi-valued circuits. To appear in the proceedings of LICS 2018; arXiv preprint arXiv:1710.08163, 2017. [17] P. M. Idziak, J. Krzaczkowski, M. Kompatscher, and P. Kawa lek. Private communication. 2018. [18] C. Szab´o and V. Vertesi. The equivalence problem over finite rings. International Journal of Algebra and Computation, 21(03):449 -- 457, 2011. [19] A. Wires. On supernilpotent algebras. preprint arXiv:1701.08949, 2017. Department of Algebra, MFF UK,, Sokolovska 83, 186 00 Praha 8, Czech Republic, E-mail address, [email protected]:,
1905.11812	1	1905	2019-05-27T17:35:26	On completion of a linearly independent set to a basis with shifts of a fixed vector	[ "math.RA", "math.AG" ]	Let $\mathbb{F}$ be an infinite field. Let $n$ be a positive integer and let $1\leq d\leq n$. Let $\vec{f}_1, \vec{f}_2, \ldots, \vec{f}_{d-1} \in \mathbb{F}^{n}$ be $d-1$ linearly independent vectors. Let $\vec{x}=(x_1,x_2,\ldots,x_{d},0,0,\ldots,0)\in\mathbb{F}^{n}$, with $n-d$ zeros at the end. Let $\vec{R}: \mathbb{F}^n \to\mathbb{F}^n$ be the cyclic shift operator to the right, e.g. $\vec{R}\,\vec{x} = (0,x_1,x_2,\ldots,x_{d},0,0,\ldots,0)$. Is there a vector $\vec{x} \in \mathbb{F}^n$, such that the $n-d+1$ vectors $\vec{x},\vec{R}\vec{x}, \ldots ,\vec{R}^{n-d}\vec{x}$ complete the set $\{\vec{f}_j\}_{j=1}^{d-1}$ to a basis of $\mathbb{F}^n$? The answer is in the affirmative for every linearly independent set of $\vec{f}_j$, $j=1,2,\ldots,d-1$. In order to prove this fact, we prove that the $(n-d+1)\times(n-d+1)$ minors of the $(n-d+1)\times(n-d+1)$ circulant matrix. $\begin{bmatrix} \vec{x}, \vec{R} \vec{x}, \ldots, \vec{R}^{n-d} \vec{x} \end{bmatrix}^\intercal$ form a Gr\"obner basis with respect to the graded reverse lexicographic order (grevlex).	math.RA	math	ON COMPLETION OF A LINEARLY INDEPENDENT SET TO A BASIS WITH SHIFTS OF A FIXED VECTOR MAREK RYCHLIK UNIVERSITY OF ARIZONA DEPARTMENT OF MATHEMATICS, 617 N SANTA RITA RD, P.O. BOX 210089 TUCSON, AZ 85721-0089, USA 9 1 0 2 y a M 7 2 ] . A R h t a m [ 1 v 2 1 8 1 1 . 5 0 9 1 : v i X r a Abstract. Let F be an infinite field. Let n be a positive integer and let 1 ≤ d ≤ n. Let f1, f2, . . . , fd−1 ∈ Fn be d − 1 linearly independent vectors. Let x = (x1, x2, . . . , xd, 0, 0, . . . , 0) ∈ Fn, with n − d zeros at the end. Let R : Fn → Fn be the cyclic shift operator to the right, e.g. Is there a vector x ∈ Fn, such that the n − d + 1 vectors R x = (0, x1, x2, . . . , xd, 0, 0, . . . , 0). x, Rx, . . . , Rn−dx complete the set {fj }d−1 j=1 to a basis of Fn? The answer is in the affirmative for every linearly independent set of fj , j = 1, 2, . . . , d − 1. In order to prove this fact, we prove that the (n − d + 1) × (n − d + 1) minors of the (n − d + 1) × (n − d + 1) circulant matrix (cid:2)x, Rx, . . . , Rn−dx(cid:3)⊺ form a Gröbner basis with respect to the graded reverse lexicographic order (grevlex). 1. Background Let F be an infinite field. Therefore, every polynomial f ∈ F[x1, x2, . . . , xd] is either 0 or is not identically 0 as a function. Let n be a positive integer and let 1 ≤ d ≤ n. Let F = {f1, f2, . . . , fd−1} ⊆ Fn be a subset of d − 1 linearly independent vectors. In many problems we need to complete this set with n − d + 1 linearly independent vectors to a basis of Fn. Furthermore, we may need the completing vectors to satisfy some constraints. For instance, we can always pick the completing vectors from the standard basis of Fn. In the current paper we constrain the completing set to shifts of a fixed vector with a contiguous block of n − d zeros. More precisely, let (1) x = (x1, x2, . . . , xd , 0, 0, . . . , 0 ) ∈ Fn. d entries n − d zeros {z } {z } Let R : Fn → Fn be the cyclic shift operator to the right. Thus for k = 0, 1, . . . , n − d: Rk x = (0, 0, . . . , 0 , x1, . . . , xd , 0, 0, . . . , 0 ). k zeros d entries n − d − k zeros {z } {z } {z } The main objective of the current paper is in the theorem below: Theorem 1. For every integer n and d, 1 ≤ d ≤ n and a linearly independent set there is a vector x ∈ Fn of the form (1) such that the set of n − d + 1 vectors F = {fj}d−1 j=1 ⊆ Fn nx, Rx, . . . , Rn−dxo Date: May 29, 2019. 2010 Mathematics Subject Classification. 11C08, 11C20 . 1 completes the set F to a basis of Fn. Proof. Split into a number of lemmas that follow. (cid:3) The pursuit of the proof led us to considerations in ideal theory in the ring of multivariate polynomials. In particular, we find a Gröbner basis for an ideal generated by the minors of a circulant matrix. We proceed to briefly outline the transition to ideal theory which allows us to prove Theorem 1. We will only need a minimum background from Gröbner basis theory which can be found in any general reference on the subject (e.g. [2, 1]). The main result can be formulated in matrix form by considering the square n × n matrix of this special form: M =   x1 0 . . . 0 0 f1,1 f2,1 . . . x2 x1 . . . 0 0 f1,2 f2,2 . . . fd−1,1 fd−1,2 . . . xd x2 . . . 0 . . . xd . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 0 . . . . . . . . . . . . . . . . . . . . . 0 0 0 xd 0 . . . 0 . . . . . . . . . . . . . . . . . . . . . fd−1,n−1 fd−1,n xd−1 f1,n−1 f2,n−1 xd f1,n f2,n . . . . . . = (cid:20)X F(cid:21)   where X is a (n − d + 1) × n matrix and F is a (d − 1) × n matrix. If for every F of rank d − 1 we can find x such that M is non-singular then the proof of Theorem 1 follows. We make an observation that X is a special Toeplitz, circulant matrix [5]. Example 1. We explicitly write down the matrix M for n = 6 and d = 4: M =   We consider the matrix x2 x1 0 x3 x2 x1 x4 x3 x2 0 x4 x3 x1 0 0 0 0 x4 f1,1 f1,2 f1,3 f1,4 f1,5 f1,6 f2,1 f2,2 f2,3 f2,4 f2,5 f2,6 f3,1 f3,2 f3,3 f3,4 f3,5 f3,6 .   obtained by putting x and its shifts in the rows of the matrix X, i.e. the (n − d + 1) × n circulant matrix X = (cid:2)x, Rx, . . . , Rn−dx(cid:3)⊺ x1 x2 . . . xd 0 . . . xd . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 0 0 . . . 0 . . . . . . 0 . . . xd−1 xd xd     X = 0 . . . 0 0 x1 x2 . . . . . . 0 0 0 0 The symbols x1, x2, . . . , xd are variables with range F. Thus, the minors of the matrix X are polynomials in the ring of polynomials F[x1, x2, . . . , xd]. 2 Lemma 1. Let P n q denote the family of all subsets H ⊆ {1, 2, . . . , n} with q elements. The deter- minant of the special matrix M is a homogeneous polynomial in variables x1, x2, . . . , xd of degree n − d and (2) where εH = ±1 and det M = XH ∈P n n−d+1 εH det XH det FH ′. (1) XH is the submatrix of X obtained by taking a subset of n − d + 1 columns of X with indices in the subset H; (2) FH ′ is the submatrix of F obtained from F by taking d − 1 columns with indices in the complement H ′ of H. Proof. We use Laplace's expansion of the determinant by complementary minors [3]. We use the expansion along the first n − d + 1 rows. (cid:3) Lemma 2. Let fj ∈ Fn, j = 1, 2, . . . , d − 1 be an arbitrary linearly independent set. Let the set (3) {det XH}H ∈P n n−d+1 be a linearly independent set in Fn−d+1[x1, x2, . . . , xd], the vector space of homogenous polynomials of degree n−d+1. Then there exists a vector x ∈ Fn with n−d zeros at the end such that det M 6= 0. Proof. By way of contradition, we assume that det M = 0 for all x ∈ Fn whose last n−d coordinates are 0. This implies that det M ∈ F[x1, x2, . . . , xd] is a polynomial identically equal to 0 as a function. As F is an infinite field, this polynomial is a zero polynomial. Therefore, linear independence of the set (3) and equation (2) imply that det FH ′ = 0 for all H ∈ P n n−d+1. Since H ′ traverses all sets of cardinality d − 1, this in turn implies that the rank of the matrix F is strictly less than d − 1, i.e. vectors fj, j = 1, 2, . . . , d − 1, are linearly dependent. This contradicts the assumptions. (cid:3) Corollary 1. Theorem 1 will follow once we prove the independence of the set (3). The objective of this section is to prove 2. Ideal theory considerations Theorem 2. The (n − d + 1) × (n − d + 1) minors of the circulant matrix X form a Gröbner basis with respect to the graded reverse lexicographic order (grevlex). Let X be the (n − d + 1) × (n + 1) matrix x1 x2 . . . xd 0 . . . xd . . . . . . . . . . . . . . . . . . . . . . . . . . . 0 0 0 . . . 0 . . . . . . 0 . . . xd−1 xd xd   X =   0 . . . 0 0 x1 x2 . . . . . . 0 0 0 0 Let us recall that P n n−d+1 is the family of all subsets H ⊂ {1, 2, . . . , n} of cardinality n − d + 1. Also, XH is the submatrix of X with columns whose indices are in H. The set of minors is a family of homogeneous polynomials of degree n − d + 1 in variables x1, x2, . . . , xd. X = (cid:8)det XH : H ∈ P n n−d+1(cid:9) We will use some basic techniques from computational algebraic geometry. One is that of a monomial order or term order. The only monomial order relevant to this paper is the graded 3 reverse lexicographic order also abbreviated as grevlex. We assume that every polynomial has its terms ordered in descending (grevlex) order. The largest term (monomial) is called the leading term (monomial) of a polynomial, and will be denoted LM (f ) for a polynomial f . The family LM (X ) is the set of all leading monomials LM (f ) of all elements f ∈ X . Lemma 3. The set LM (X ) is the set of all monomials of degree n−d+1 in variables x1, x1, x2, . . . , xd. In particular, both LM (X ) and X itself are both bases of the linear vector space Fn−d+1[x1, x2, . . . , xd] of homogeneous polynomials of degree n − d + 1. Proof. Let H = {c1, c2, . . . , cn−d+1} be the set of distinct n − d + 1 column indices (1 ≤ cj ≤ n), where the sequence cj is non necessarily increasing. Using Leibniz formula for determinants [4], we know that the minor is a sum of products, up to the sign: xk1 1 xk2 2 · · · xkd d where for j = 1, 2, . . . , d: kj = #{i : 1 ≤ i ≤ n − d + 1, ci − i = j − 1}. We require that for i = 1, 2, . . . , n − d + 1. (4) 0 ≤ ci − i ≤ d − 1. Essentially, kj represent the frequency table of the sequence ci − i + 1, i = 1, 2 . . . , n − d + 1. Besides kj ≥ 0, we also must have (5) d Xj=1 kj = n − d + 1. Moreover, j = ci − i + 1 represents the horizontal distance from the diagonal, of the entry at the intersection of row i with column ci. In addition, if we reorder the sequence ci to a sequence c′ i = cρ(i), where ρ : {1, 2, . . . , n − d + 1} → {1, 2, . . . , n − d + 1} is a permutation, while preserving condition (4), then the new sequence k′ j represents another monomial in the same minor det XH. The question arises: which ρ yields the monomial maximal in the grevlex order? There are two objectives to achieve by reordering the sequence cj: (1) Let kj > 0 be the last j with this property, i.e. kj+1, kj+2, . . . , kd are all zero; we make j the lowest possible. We recall that grevlex considers variables to be ordered from the highest, i.e. xd > xd−1 > . . . > x1; therefore, a monomial which has variable xj but no variables xj+1, xj+2, . . . , xd, is greater than all polynomials which have one of those variables. (2) Once we achieved the first goal, we minimize kj i.e. the power of xj; this is also the count of i such that ci − i = j − 1. The grevlex order considers the term with the lower power kj greater. We claim that the optimal ordering is achieved when the sequence ci is strictly increasing. The remainder of the current proof is devoted to the proof of this claim. We note that ci is strictly increasing iff the sequence ji = ci − i + 1 is non-decreasing. Indeed ji+1 − ji = (ci+1 − (i + 1) + 1) − (ci − i + 1) = ci+1 − ci − 1 ≥ 0. of the non-decreasing sequence ji. Obviously, knowing kj for j = 1, 2, . . . , d such that Pd The numbers kj are simply the counts of the number of times i such that ji = j, i.e. level counts j=1 kj = n − d + 1 allows us to reconstruct the non-decreasing sequence ji, i = 1, 2, . . . , n − d + 1 uniquely. Then the equation ji = ci − i + 1 allows us to reconstruct ci. Moreover, ci+1 − ci = ji+1 − ji + 1, so ci is strictly increasing if ji is non-decreasing. Thus, we established a 1:1 correspondence between sequences (kj )d with the specified properties. and (ci)n−d+1 j=1, (ji)n−d+1 i=1 i=1 4 Therefore j′ ℓ = c′ ℓ − ℓ + 1 is given by: cℓ ci+1 ci if ℓ /∈ {i, i + 1}, if ℓ = i, if ℓ = i + 1. if ℓ /∈ {i, i + 1}, jℓ ji+1 + 1 if ℓ = i, ji − 1 if ℓ = i + 1. c′ ℓ =   ℓ =   j′ It remains to be shown that the monomial obtained from an unsorted sequence ci is strictly smaller in grevlex order than that obtained from the sorted sequence. Thus, let us assume that the sequence ji = ci − i + 1 is not a non-decreasing sequence; hence, for some i, 1 ≤ i < n − d + 1 ji+1 < ji or ji+1 ≤ ji − 1. Equivalently ci+1 − (i + 1) + 1 ≤ (ci − i + 1) − 1. Hence ci+1 ≤ ci. Thus ci+1 < ci and ji+1 ≤ ji − 2. Let us consider the sequence c′ ℓ, ℓ = 1, 2, . . . , n − d + 1, obtained by swapping ci and ci+1. We will prove that by doing so we increased the monomial xk1 1 xk2 2 · · · xkd d . Thus Indeed, and j′ i = c′ i − i + 1 = ci+1 − (i + 1) + 2 = ji+1 + 1 j′ i+1 = c′ i+1 − (i + 1) + 1 = (ci − i + 1) − 1 = ji − 1. As ji+1 + 2 ≤ ji and ji − 1 ≥ ji+1 + 1, all values remain in the range from 1 to d, and the sequence c′ ℓ is valid (satisfies 1 ≤ c′ ℓ − ℓ + 1 ≤ d). The value of kj does not change unless j ∈ J, where As ji+1 + 1 ≤ ji − 1, we have J = {ji+1, ji+1 + 1, ji − 1, ji}. ji+1 < ji+1 + 1 ≤ ji − 1 < ji the set J consists of either 3 or 4 values. It is easy to see how values kj change for j ∈ J. There are 2 cases. Case ji − 1 = ji+1 + 1: when J consists of 3 elements. Then k′ ji+1 = kji+1 − 1, k′ ji−1 = kji−1 + 2, ji+1+1 = kji+1+1 + 2) (k′ k′ ji = kji − 1. The parenthesized third equation is a copy of the second equation. Case ji+1 + 1 < ji − 1: when J consists of 4 elements. Then k′ ji+1 = kji+1 − 1, k′ ji−1 = kji−1 + 1, k′ ji+1+1 = kji+1+1 + 1, k′ ji = kji − 1. Since the power of the highest-indexed impacted variable xji drops, the new sequence k′ ℓ, ℓ = 1, 2, . . . , n − d + 1 corresponds to a monomial which is strictly greater in grevlex order. This proves that the maximum monomial is obtained when cℓ is sorted in ascending order. This completes the proof of the main claim, and the entire lemma. (cid:3) 5 Remark 1. It follows that LM (det XH) with respect to the grevlex order is the product of the diagonal entries of XH. Remark 2. The set X is a Gröbner basis of the ideal generated by X . This follows from Lemma 3 and the Buchberger criterion. This ideal is also mn−d+1 where m = hx1, x2, . . . , xdi is the maximal ideal generated by x1, x2, . . . , xd and is the ideal of homogeneous polynomials vanishing at 0. Example 2. We give a counterexample to the conjecture that the grevlex order could be substituted with the (graded) lexicographic order (lex or grlex). Let the matrix be: Thus d = 3 and n = 4. We find: X = (cid:20)x1 x2 x3 x1 x2 x3(cid:21) . 0 0 We also find that with grlex or lex used instead of grevlex: 1, x1 x2, x1 x3, x2 X = (cid:8)x2 LMlex(X ) = (cid:8)x2 3(cid:9) . 2 − x1 x3, x2 x3, x2 3(cid:9) . 1, x1 x2, x1 x3, x2 x3, x2 2 < x1 x3 with respect to the (graded) lexicographic order, but Notably, x2 x2 2 > x1 x3 with respect to grevlex. Therefore, 2 is missing. We note that x2 LMgrlex(cid:0)x2 LMgrevlex(cid:0)x2 2 − x1 x3(cid:1) = x1 x3, 2 − x1 x3(cid:1) = x2 2. Hence, consistent with the statement of Lemma 3, LMgrevlex(X ) = (cid:8)x2 1, x1 x2, x1 x3, x2 3(cid:9) . 2, x2 x3, x2 3. Generalizations The assumption that F is an infinite field can be relaxed. If we assume that F is an infinite integral domain then the only identically vanishing polynomial is the zero polynomial. This is sufficient to prove Theorem 1. Theorem 2 and Lemma 3 are valid over any ring with unity. I thank Johnatan Ashbrock for the question that led to Theorem 1. 4. Acknowledgments References [1] Thomas Becker and Volker Weispfenning. Gröbner Bases. Springer, 1993. [2] David A. Cox, John Little, and Donal O'Shea. Springer, 2012. [3] Wikipedia. determinant expansion Laplace of a by complementary minors. Laplace expansion of a determinant by complementary minors, May 2019. [4] Wikipedia. Leibniz formula for determinants. Leibniz formula for determinants, May 2019. [5] Wikipedia. Toeplitz matrix. Toeplitz Matrix, May 2019. 6
1307.0993	1	1307	2013-07-03T12:47:07	Few remarks on evolution algebras	[ "math.RA" ]	In the present paper we study some algebraic properties of evolution algebras. Moreover, we reduce the study of evolution algebras of permutations to two special types of evolution algebras, idempotents and absolute nilpotent elements of the algebra. We study three-dimensional evolution algebras whose each element of evolution basis has infinite period. In addition, for an evolution algebra with some properties we describe its associative enveloping algebra.	math.RA	math	FEW REMARKS ON EVOLUTION ALGEBRAS ABROR KH. KHUDOYBERDIYEV, BAKHROM A. OMIROV, IZZAT QARALLEH Abstract. In the present paper we study some algebraic properties of evolution algebras. Moreover, we reduce the study of evolution algebras of permutations to two special types of evolution algebras, idempotents and absolute nilpotent elements of the algebra. We study three-dimensional evolution algebras whose each element of evolution basis has infinite period. In addition, for an evolution algebra with some properties we describe its associative enveloping algebra. Mathematics Subject Classification 2010 : 17D92, 17D99. Key Words and Phrases: evolution algebra, algebra of permutations, absolute nilpotent element, idempotent, algebra of multiplications, associative enveloping algebra. 1. Introduction. In 20s and 30s of the last century the new object was introduced to mathematics, which was the product of interactions between Mendelian genetics and mathematics. One of the first scientist who give an algebraic interpretation of the " × " sign, which indicated sexual reproduction was Serebrowsky. It is known that there exists an intrinsic and general mathematical structure behind the neutral Wright-Fisher models in population genetics, the reproduction of bacteria involved by bacteriophages, asexual reproduction or generally non-Mendelian inheritance and Markov chains. In [10] a new type of algebras was associated with it -- the evolution algebras. Although an evolution algebra is an abstract system, it gives an insight for the study of non- Mendelian genetics. For instance, an evolution algebra can be applied to the inheritance of organelle genes, one can predict, in particular, all possible mechanisms to establish the homoplasmy of cell populations. The general genetic algebras developed into a field of independent mathematical interest, because these algebras are in general non-associative and do not belong to any of the well-known classes of non-associative algebras such as Lie algebras, alternative algebras, or Jordan algebras. Until 1980s, the most comprehensive reference in this area was Worz-Busekros's book [11]. More recent results, such as genetic evolution in genetic algebras, can be found in Lyubich's book [7]. A good survey is Reed's article [8]. In Tian's book [10] a foundation of the framework of the theory of evolution algebras is established and some applications of evolution algebras in the theory of stochastic processes and genetics are discussed. Recently, Rozikov and Tian [9] studied algebraic structures evolution algebras associated with Gibbs measures defined on some graphs. In [2], [5], [6] derivations, some properties of chain of evolution algebras and dibaricity of evolution algebras were studied. In [1], [3], [4] certain algebraic properties of evolution algebras (like right nilpotency, nilpotency and solvability etc.) in terms of matrix of structural constants have been investigated. In fact, nilpotency, right nilpotency and solvability might be interpreted in a biological way as a various types of vanishing ("deaths") populations. The present paper is organized as follows: In Section 2 we give some definitions and preliminary results. In Section 3 we reduce the study of arbitrary evolution algebra of permutations into two spe- cial evolution algebras. Section 4 is devoted to the description of n-dimensional associative enveloping algebras of n-dimensional evolution algebras with some restrictions on rank of the matrix A of struc- tural constants. Moreover, associative enveloping algebras for 2-dimensional evolution algebras are described, as well. In Section 5 we establish some properties of three-dimensional evolution algebras whose each basis element has infinite period. Throughout the paper we consider finite-dimensional evolution algebras over a field of zero charac- teristic. Moreover, in the multiplication table of an evolution algebra the omitted products are assumed to be zero. Let us define the main object of this work - evolution algebra. 1 2. Preliminaries. 2 ABROR KH. KHUDOYBERDIYEV, BAKHROM A. OMIROV, IZZAT QARALLEH Definition 2.1. [10] Let (E, ·) be an algebra over a field F. If it admits a basis {e1, e2, . . . } such that ei · ej = 0, f or i 6= j, then this algebra is called evolution algebra. ei · ei =Xk ai,kek, f or any i, It is remarkable that this type of algebra depends on evolution basis {e1, e2, . . . }. In the following theorem we present the list (up to isomorphism) of 2-dimensional complex evolution algebras. Theorem 2.2. [4] Any 2-dimensional non abelian complex evolution algebra E is isomorphic to one of the following pairwise non isomorphic algebras: (1) dim E2 = 1 • E1 : • E2 : • E3 : • E4 : e1e1 = e1, e1e1 = e1, e2e2 = e1, e1e1 = e1 + e2, e2e2 = −e1 − e2, e1e1 = e2. (2) dim E2 = 2 • E5 : • E6 : cos 2πk e1e1 = e1 + a2e2, e2e2 = a3e1 + e2, 1 − a2a3 6= 0, where E5(a2, a3) ∼= E ′ e1e1 = e2, 3 + i sin 2πk e2e2 = e1 + a4e2, where for a4 6= 0, E6(a4) ∼= E6(a′ for some k = 0, 1, 2. 3 5(a3, a2), = 4) ⇔ a′ 4 a4 Further we shall show the role of idempotents and absolute nilpotent elements of an evolution algebra. Definition 2.3. An element x of an evolution algebra E is called idempotent, if xx = x. An element y of an evolution algebra E is called absolute nilpotent if yy = 0. Consider a complex evolution algebra En,π(a1, a2, . . . , an) with a basis {e1, e2, . . . , en} and the table of multiplications given by ( ei · ei = aieπ(i), ei · ej = 0, 1 ≤ i ≤ n, i 6= j, where π is an element of the group of permutations Sn. An evolution algebra En,π(a1, a2, . . . , an) is said to be evolution algebra of permutations. In what follows, by a cycle permutation we mean a permutation in which a part of symbols {l1, l2, . . . , lt} ⊆ {1, 2, . . . , n} are cyclic permutated and the rest ones are stationary, i.e., l1 → l2 → · · · → lt → l1, and we denote π = (l1, l2, . . . , lt). It is known that any permutation is up to order uniquely decomposed into product of independent cycles. For permutations of the form π = (l1, l2, . . . , lr)(m1, m2, . . . , ms) . . . (p1, p2, . . . , pt) it is known the following result. Proposition 2.4. Two permutations are conjugated in Sn if and only if the corresponding sets {r, s, . . . , t} are coincided. Definition 2.5. [10] An evolution algebra E with a table of multiplications ai,kek, aiej = 0, i 6= j ei · ei =Xk ai,k = 1. is called Markov evolution algebra if Pk For a given element x of an evolution algebra E we consider the right multiplication operator Rx : E → E defined by Rx(y) = yx, y ∈ E. Note that operators of right and left multiplications are coincided, since evolution algebras are commutative. For an evolution algebra E, by M (E) we denote an associative enveloping algebra which is generated by the set R(E) = {Rx x ∈ E}. It is clear that M (E) is a subalgebra of End(E). For an element x ∈ E we define plenary powers as follows: x[k+1] = x[k] · x[k], x[1] = x, k ≥ 1. FEW REMARKS ON EVOLUTION ALGEBRAS 3 Definition 2.6. Let ej be a generator of an evolution algebra E, the period d of ej is defined to be the greatest common divisor of the set {log2 m ej < e[m] j }. That is d = g.c.d.{log2 m ej < e[m] j }. Let us first present two important examples of evolution algebra of permutations. 3. Evolution algebra of permutations Example 3.1. Consider the following evolution algebra: ei · ei = ei+1, 1 ≤ i ≤ n − 1, en · en = e1, ei · ej = 0, i 6= j. ei · ei = ei+1, 1 ≤ i ≤ n − 1, en · en = 0, ei · ej = 0, i 6= j, En :   ENn :   Evidently, the algebra En is evolution algebra of permutations of the form En,π(1, 1, . . . , 1), with π = (1, 2, 3, . . . , n). Example 3.2. Evolution algebra defined as follows: is the algebra of permutations of the form En,π(1, 1, . . . , 1, 0) with π = (1, 2, 3, . . . , n). Note that En, ENn are single-generated simple and nilpotent evolution algebras, respectively. More- over, algebras E1, EN1 define one-dimensional evolution algebras, whose basis elements are idempotent and absolute nilpotent elements, respectively. Now we shall consider some properties of evolution algebra of permutations. Proposition 3.3. Let En,π(a1, a2, . . . , an) be an evolution algebra of permutations with the following conditions: (i) ai 6= 0 for all i (1 ≤ i ≤ n), (ii) π = π1 ◦ π2 ◦ · · · ◦ πr, where π1 = (l1, l2, . . . , lk1 ), π2 = (m1, m2, . . . , mk2), . . . , πr = (p1, p2, . . . , pkr ) are independent cycles and k1 + k2 + · · · + kr = n. Then En,π(a1, a2, . . . , an) ∼= Ek1,π1(b1, b2, . . . , bk1 ) ⊕ Ek2,π2 (c1, c2, . . . , ck2) ⊕ · · · ⊕ Ekr ,πr (d1, d2, . . . , dkr ). Proof. The isomorphism is provided by the following change of basis: ei,1 = eli , 1 ≤ i ≤ k1, ei,2 = emi , 1 ≤ i ≤ k2, . . . , ei,r = epi , 1 ≤ i ≤ kr. Thus, we have the evolution algebra Eks,πs(∗, ∗, . . . , ∗) with the basis ei,s, 1 ≤ i ≤ ks, 1 ≤ s ≤ r and En,π(a1, a2, . . . , an) ∼= Ek1,π1(b1, b2, . . . , bk1) ⊕ Ek2,π2(c1, c2, . . . , ck2 ) ⊕ · · · ⊕ Ekr ,πr (d1, d2, . . . , dkr ) for some non-zero values of bi, cj, . . . , ds. (cid:3) In the following proposition we specify more details on the terms of direct sum in the statement of Proposition 3.3. Proposition 3.4. Any evolution algebra of permutation En,τ (a1, a2, . . . , an) with τ = (l1, l2, . . . , ln) and condition ai 6= 0 for all i (1 ≤ i ≤ n) is isomorphic to the algebra En,π(a1, aπ(1), . . . , aπn−1(1)) with π = (1, 2, . . . , n). Proof. The isomorphism is established by basis permutation: e′ 1 = e1, e′ i = eπi−1(1), 2 ≤ i ≤ n. (cid:3) Theorem 3.5. Any evolution algebra of permutation En,τ (a1, a2, . . . , an) with τ = (l1, l2, . . . , ln) and condition ai 6= 0 for all i (1 ≤ i ≤ n) is isomorphic to the algebra En. 4 ABROR KH. KHUDOYBERDIYEV, BAKHROM A. OMIROV, IZZAT QARALLEH Proof. Taking into account Proposition 3.4 it is sufficient to establish isomorphism between evolution algebra En,π(a1, a2, . . . , an) with π = (1, 2, . . . , n) and non zero values of ai, 1 ≤ i ≤ n and evolution algebra En. The application of the following scaling of basis: e′ i = Aiei, 1 ≤ i ≤ n with Ai = 2n−1s 1 a2n−2 i+1 . . . a2i−1 n a2n−1 i a2i−2 1 a2i−3 2 . . . ai−1 , deduces products i · e′ e′ e′ n · e′ e′ i · e′ i = e′ i+1, 1 ≤ i ≤ n − 1, n = e′ 1, j = 0, i 6= j.   (cid:3) Now we consider the case of ai = 0 for some i ∈ {1, 2, . . . , n}. Proposition 3.6. Any evolution algebra of permutation En,π(a1, a2, . . . , an) with π = (l1, l2, . . . , ln) and condition ai = 0 for some i ∈ {1, 2, . . . , n} is isomorphic to the algebra ENk1 ⊕ ENk2 ⊕ · · ·⊕ ENkr . Proof. Similarly to the proof of Proposition 3.4 taking the change e′ 1 = e1, e′ i = eπi−1(1), 2 ≤ i ≤ n, we can suppose π = (1, 2, . . . , n). Let ai1 = ai2 = · · · = air = 0 for i1 < i2 < · · · < ir and the rest are non-zero. If ir = n, then similarly as above we can assume that all ai = 1 for 1 ≤ i ≤ n − 1 and hence, En,π(a1, a2, . . . , an) with π = (1, 2, . . . , n) is isomorphic to the algebra En. If ir < n, then taking the following change of basis: e1 s = eir +s, 1 ≤ s ≤ n − ir, e1 n−ir+s = es, 1 ≤ s ≤ i1, e2 s = ei1+s, 1 ≤ s ≤ i2 − i1, . . . . . . . . . er s = eir−1+s, 1 ≤ s ≤ ir − ir−1. we obtain that En,π(a1, a2, . . . , an) with π = (1, 2, . . . , n) isomorphic to the algebra ENk1(a1, . . . , ak1−1) ⊕ ENk2 (b1, . . . , bk2−1) ⊕ · · · ⊕ ENkr (c1, . . . , ckr−1), where each of the algebra ENks(∗, ∗, . . . , ∗), 1 ≤ s ≤ r has the model of the following evolution algebra: is ENk(a1, . . . , ak−1) : ( ei · ei = aiei+1, 1 ≤ i ≤ k − 1, ai 6= 0, ei · ej = 0, i 6= j. Taking the basis transformation in the algebra ENk(a1, . . . , ak−1) : k = a2k−2 1a2e3, . . . , e′ 2 = a1e2, e′ e′ 1 = e1, e′ 3 = a2 a2k−3 2 . . . ak−1ek 1 we have that ENk(a1, . . . , ak−1) is isomorphic to the algebra ENk, which complete the proof of propo- sition. (cid:3) Below we establish the isomorphism of evolution algebras of permutations with given conjugated permutations. Theorem 3.7. If permutations π1, π2 ∈ Sn are conjugated, then evolution algebras En,π1 (a1, . . . , an) and En,π2(a1, . . . , an) are isomorphic. Proof. Let π1, π2 ∈ Sn are conjugated, then there exists g ∈ Sn such that gπ1 = π2g. The map f : En,π1 → En,π2 defined by f (ei) = eg(i) is isomorphism. Indeed, aiegπ1(i) = f (aieπ1(i)) = f (ei · ei) = f (ei) · f (ei) = aieg(i) ∗ eg(i) = aieπ2g(i). (cid:3) FEW REMARKS ON EVOLUTION ALGEBRAS 5 Thus, for an algebra En,π(a1, . . . , an) we can always assume that π = (1, 2, . . . , n) and table of multiplication is where ai ∈ {0; 1}. ei · ei = aiei+1, 1 ≤ i ≤ n − 1, en · en = ane1, ei · ej = 0, i 6= j,   Proposition 3.8. An arbitrary evolution algebra En,π(a1, a2, . . . , an) with π = (1, 2, . . . , n) is isomor- phic to the algebra En or the direct sum of evolution algebras ENk1 ⊕ ENk2 ⊕ · · · ⊕ ENkr . Proof. If all ai 6= 0, then due to Theorem 3.5 we have that algebra En,π(a1, a2, . . . , an) is isomorphic to the En. Let ak = 0 for some k and ai 6= 0 for i 6= k. Taking the basis transformation in the following form: e′ 1 = A1ek+1, e′ 2 = A2ek+2, . . . , e′ n−k = An−ken, e′ n−k+1 = An−k+1e1, e′ n−k+2 = An−k+2e2, . . . , e′ n = Anek, where A1 = 1, A2 = ak+1, A3 = a2 k+1ak+2, . . . , An−k+1 = a2n−k−1 k+1 a2n−k−2 k+2 . . . an, An−k+2 = a2n−k k+1 a2n−k−1 . . . ak−1, we derive isomorphism between algebra En,π(a1, . . . , ak−1, 0, ak+1, . . . , an) and algebra ENn. na1, k+2 . . . , An = a2n−2 k+1 a2n−3 k+2 . . . a2k−1 a2k−3 2 a2k−2 1 . . . a2 n Applying similar arguments, we can establish that for (r + 1)-times of parameters ai are equal to zero an algebra En(a1, a2, . . . , an) is isomorphic to ENk1 ⊕ ENk2 ⊕ · · · ⊕ ENkr . (cid:3) We resume the above results in the main theorem of this section. Theorem 3.9. An arbitrary evolution algebra of permutations En,π(a1, a2, . . . , an) is isomorphic to a direct sum of algebras Ep1 , Ep2 , . . . , Eps , ENk1 , ENk2 , . . . , ENkr , i.e., En,π(a1, a2, . . . , an) ∼= Ep1 ⊕ Ep2 ⊕ · · · ⊕ Eps ⊕ ENk1 ⊕ ENk2 ⊕ · · · ⊕ ENkr . In the description of evolution algebras of permutations from above theorem we get the importance of algebras En, ENn, idempotents and absolute nilpotent elements. 4. Associative enveloping algebras of some evolution algebras For a complex two-dimensional evolution algebra E of the list of Theorem 2.2 we describe its associative enveloping algebra M (E): 1 0!+ ∼= xx = x, yx = x; −1 −1!+ ∼= xx = x, xy = −x, yx = y, yy = −y; 0 1 0 0 0 0 M (E1) = algRe1 = 1 M (E2) = algRe1 = 1 M (E3) = algRe1 = 1 M (E4) = algRe1 = 0 M (E5(0, 0)) = algRe1 = 1 0 M (E5(a2, a3))(a2 = 0 or a3 = 0) ∼=(P ∈ M2(C) P = b1 0!+ ∼= xx = x; 0! , Re2 = 0 0 0! , Re2 = 0 0!+ ∼= xx = 0; 0 0! , Re2 = 0 0 1 0 0 0 0 1!+ ∼= xx = x, yy = y; b2 b3!) ; 6 ABROR KH. KHUDOYBERDIYEV, BAKHROM A. OMIROV, IZZAT QARALLEH M (E5(a2, a3))(a2a3 6= 0) = algRe1 = 1 a2 0 0! , Re2 = 0 M (E6(a4)) = alg*Re1 = 0 1 0! , Re2 = 0 1 a4!+ ∼= M2(C). a3 0 0 0 1!+ ∼= M2(C); n Take the element x = xiei. Let x · x = 0, then we have Pi=1 Xi=1 x · x =(cid:0) n From this we have n Xi=1 xiei(cid:1) = x2 i n Xi=1 n Xj=1 ai,jej = n n Xj=1(cid:0) Xi=1 x2 i ai,j(cid:1)ej. a1,1x2 a1,2x2 1 + a2,1x2 1 + a2,2x2 2 + · · · + an,1x2 2 + · · · + an,2x2 n = 0 n = 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . a1,nx2 1 + a2,nx2 2 + · · · + an,nx2 n = 0 xiei(cid:1)(cid:0)   This system has a non-trivial solution if and only if detAt = detA = 0. Note that for the case of Markov evolution algebra, by summing above equalities, we conclude 1 + x2 x2 Thus we proved the following proposition. n = 0. If Markov evolution algebra is real, then xi = 0, 1 ≤ i ≤ n, that is, x = 0. 2 + · · · + x2 Proposition 4.1. A complex evolution algebra has non-trivial absolute nilpotent elements if and only if detA = 0. Moreover, if real evolution algebra is Markov, then it has only trivial absolute nilpotent elements. For an n-dimensional evolution algebra E we consider an associative enveloping alge- bra M (E). Due to linearity the right multiplication operator one gets M (E) = alg < Rei ei is a basis element of E >. n Pk=1 Using the equalities Rei (ei) = as follows: where ei,k are matrix units. One can get ai,kek and Rei(ej) = 0 for i 6= j, we obtain a matrix form of Rei Rei = n Xk=1 ai,kei,k, n Rei Rej = ai,j aj,kei,k. (4.1) Xk=1 From (4.1) we conclude that dimM (E) = r1 + r2 + · · · + rn with ri = rank ai,1a1,1 ai,2a2,1 ai,1a1,2 ai,2a2,2 ... ... ai,nan,1 ai,nan,2   . . . . . . ... . . . ai,1a1,n ai,2a2,n ... ai,nan,n .   Further we shall consider some cases for n-dimensional evolution algebra E with structural constant matrix A and satisfying the condition: dimM (E) = n. Proposition 4.2. Let rankA = 1. Then associative enveloping algebra M (E) is isomorphic to one of the following algebras: M s : xixj = xi, 1 ≤ i ≤ s, 1 ≤ j ≤ n. Proof. Due to conditions of proposition we have the table of multiplication of algebra E : ei · ei = ti akek. n Xk=1 FEW REMARKS ON EVOLUTION ALGEBRAS 7 The condition dimM (E) = n implies ti 6= 0 for any i (1 ≤ i ≤ n) and basis of M (E) can be chosen akei,k. Therefore, the table of multiplication of the algebra as {Re1 , Re2 , . . . , Ren }, where Rei = ti M (E) has the form: n Pk=1 Rei Rej = tjajRei , 1 ≤ i, j ≤ n. By appropriate shifting of basis elements, without loss of generality, one can assume ai 6= 0 for 1 ≤ i ≤ s, s ≤ n and aj = 0 for s + 1 ≤ j ≤ n. The scaling the basis elements ei, 1 ≤ i ≤ s reduces our study to the case of tiai = 1. Thus, we obtain the tables of multiplication of associative enveloping algebras M s. (cid:3) Below we present the description of n-dimensional associative enveloping algebras M (E) for evolu- tion algebra E with rankA = n. Proposition 4.3. Let rankA = n. Then associative enveloping algebra M (E) is isomorphic to the algebra M1 : xixi = xi, 1 ≤ i ≤ n. Proof. Taking into account in the equalities n Rei Rej = ai,j aj,kei,k = β1Re1 + β2Re2 + · · · + βnRen Xk=1 that {Re1 , Re2 , . . . , Ren } are linear independent (they are forms a basis of M (E)), we conclude βk = 0 for k 6= i. Therefore, From (4.2) we derive ai,j n Xk=1 aj,kei,k = βiRei = βi ai,kei,k. n Xk=1 ai,jaj,k = βiai,k, 1 ≤ k ≤ n. (4.2) (4.3) The condition rankA = n, implies rank ai,1 . . . Using the arbitrariness k in the equality (4.3) we obtain ai,2 aj,1 aj,2 . . . ai,n aj,n! = 2 for any i 6= j. Therefore, ai,j = 0, i 6= j ⇒ ai,i 6= 0. Rei Rei = ai,iRei , 1 ≤ i ≤ n. By scaling the basis elements, we can suppose ai,i = 1 and the algebra M1 is obtained. (cid:3) The list of n-dimensional algebras M (E) for an evolution algebra, satisfying the condition rankA = n − 1, is presented in the following theorem. Theorem 4.4. Let rankA = n − 1. Then associative enveloping algebra M (E) is isomorphic to one of the following algebras: M2 : xixi = xi, 1 ≤ i ≤ n, x1xn = x1, xnx1 = xn, M4 : xixi = xi, 1 ≤ i ≤ n − 1, M3 : xixi = xi, 1 ≤ i ≤ n − 1, xnx1 = xn, x1xn = xn, x1x2 = xn, xnx2 = xn. Proof. Without loss of generality, one can assume A =   a1,1 a2,1 ... a1,2 a2,2 ... an−1,1 an−1,2 . . . . . . ... . . . n−1 Pi=1 αiai,1 n−1 Pi=1 αiai,2 . . . a1,n a2,n ... an−1,n αiai,n n−1 Pi=1 .   8 ABROR KH. KHUDOYBERDIYEV, BAKHROM A. OMIROV, IZZAT QARALLEH Case 1. Let Ren = n Pk=1(cid:0) n−1 Pi=1 Similarly as in the proof of Proposition 4.3, from the condition αiai,k(cid:1)en,k 6= 0. Then {Re1 , Re1 , . . . , Ren−1 , Ren } is a basis of M (E). rank ai,1 aj,n! = 2, ai,2 aj,1 aj,2 ai,n . . . . . . for any 1 ≤ i, j ≤ n − 1, i 6= j, we deduce Consequently, and ai,j = 0, 1 ≤ i, j ≤ n − 1, i 6= j. Rei = ai,iei,i + ai,nei,n, 1 ≤ i ≤ n − 1 n−1 n−1 Ren = αkak,ken,k + ( αsas,n)en,n. Xk=1 Xs=1 If αkak,k 6= 0 for some k (1 ≤ k ≤ n − 1), then without loss of generality we can suppose α1a1,1 6= 0. Consider Ren Re1 = α1a2 1,1en,1 + α1a1,1a1,nen,n. Since the product Ren Rei should be expressed by Ren , we conclude αkak,k = αkak,n = 0, 2 ≤ k ≤ n−1, which yield αk = 0, 2 ≤ k ≤ n − 1. If αkak,k = 0 for any k (1 ≤ k ≤ n − 1), then from the condition Ren 6= 0 we have the existence some k0 such that αk0 6= 0. Hence, ak0,k0 = 0. Since rankA = n − 1, then ak,k 6= 0 for any k 6= k0, consequently, αk = 0 for any k 6= k0. Without loss of generality, one can assume k0 = 1. Thus, in Case 1 we obtain αk = 0, 2 ≤ k ≤ n − 1. Therefore, Consider Ren Rei = 0, 2 ≤ i ≤ n − 1, Re1 Ren = a1,nRe1 . Rei Ren = (ai,iei,i + ai,nei,n)(α1a1,1en,1 + α1a1,nen,n) = ai,n(α1a1,1ei,1 + α1ai,nei,n). Since Rei Ren should belong to < Rei >, we get ai,n = 0, Thus, the table of multiplication of the algebra M (E) has the following form: 2 ≤ i ≤ n − 1. Rei Rei = ai,iRei 1 ≤ i ≤ n − 1, Re1 Ren = a1,nRe1 , Ren Re1 = a1,1Ren , Ren Ren = a1,nRen , where (a1,1, a1,n) 6= (0, 0) and ai,i 6= 0, 2 ≤ i ≤ n. Considering the possible cases: a1,1a1,n 6= 0 and a1,1a1,n = 0, one finds the algebras M2, M3. n Case 2. Let Ren = 0. Since a non-zero products of the form Rei Rej = ai,j aj,kei,k are linear independent, we obtain the existence of a unique non-zero coefficient ai0,j0 , 1 ≤ i0, j0 ≤ n − 1, i0 6= j0. Without loss of generality, we can suppose i0 = 1, j0 = 2, i.e. a1,2 6= 0. Therefore, ai,j = 0, 1 ≤ i, j ≤ n − 1, i 6= j, (i, j) 6= (1, 2). ai,i 6= 0, 1 ≤ i ≤ n − 1, Putting xi = Rei , 1 ≤ i ≤ n − 1 and xn = a2,2e1,2 + a2,ne1,n, we obtain the table of multiplications of the algebra M (E) in the form: xixi = ai,ixi, 1 ≤ i ≤ n − 1, x1x2 = a1,2xn, x1xn = a1,1xn, xnx2 = a2,2xn. Taking the change we get the algebra M4. x′ 1 = 1 ai,i xi, 1 ≤ i ≤ n − 1, x′ n = a1,2 a1,1a2,2 xn (cid:3) Pk=1 FEW REMARKS ON EVOLUTION ALGEBRAS 9 5. Three-dimensional evolution algebras whose generators have infinite period. In this section we study three-dimensional evolution algebras whose generators have infinite period. Let E be a three-dimensional evolution algebra E with table of multiplications: e1 · e1 = a1e1 + a2e2 + a3e3, e2 · e2 = b1e1 + b2e2 + b3e3, e3 · e3 = c1e1 + c2e2 + c3e3. (5.1) Since the periods of generators are infinite, we get a1 = b2 = c3 = 0. Consider e[3] i and e[4] i for 1 ≤ i ≤ 3 e[3] 1 = (a2 e[3] 2 = b2 e[3] 3 = c2 2b1 + a2 3c1e1 + (b2 2b1e1 + c2 3c1)e1 + a2 1a2 + b2 1a2e2 + (c2 3c2e2 + a2 3c2)e2 + b2 1a3 + c2 1a3e3, 2b3)e3, 2b3e3,   e[4] 3c2 2b2 3c1)e1 + a4 3c2e2 + a4 3c2 1 = (a4 2b2 2b1 + a4 e[4] 3c2)e2 + b4 3c2 3c1e1 + (b4 1a2 3c2 1a2 1a2 + b4 2 = b4 e[4] 1a2 2b2 3 = c4 1a2 2b1e1 + c4 1a2e2 + (c4 2b2 1a3 + c4 2b3)e3. Taking account the condition on periods of generators, we derive 1a3e3, 2b3e3,   a2 2b1 + a2 3c1 = 0, b2 1a2 + b2 3c2 = 0, c2 1a3 + c2 2b3 = 0, a4 3c2 2b1 + a4 2b2 3c1 = 0, b4 3c2 1a2 + b4 1a2 3c2 = 0, c4 2b2 1a3 + c4 1a2 2b3 = 0. (5.2) (5.3) Theorem 5.1. Let E be a three-dimensional evolution algebra with the table of multiplication (5.1). Let any basis element has an infinite period and a2a3b1b3c1c2 = 0. Then E is isomorphic to the following evolution algebra: E1 : e1 · e1 = a2e2 + a3e3, e2 · e2 = b3e3. Proof. Let a2a3b1b3c1c2 = 0, then, without loss of generality, we can assume b1 = 0. The equalities (5.2) and (5.3) imply a3c1 = 0, b3c2 = 0, a2b3c1 = 0. (5.4) Case 1. Let a3 = b3 = 0. Then we obtain products e1 · e1 = a2e2, e3 · e3 = c1e1 + c2e2. Taking the change e′ 1 = e3, e′ 2 = e1, e′ 3 = e2, we derive that this algebra is isomorphic to the algebra E1. Case 2. Let a3 = 0 and b3 6= 0. Then from (5.4) we have c2 = a2c1 = 0. If a2 6= 0, then c1 = 0 and we obtain evolution algebra with multiplications: e1 · e1 = a2e2, e2 · e2 = b3e3. If a2 = 0, then the table of multiplications of the algebra E is as follows: e2 · e2 = b3e3, e3 · e3 = c1e1. 2 = e3, e′ 1 = e2, e′ Putting e′ Case 3. Let a3 6= 0. Then restrictions (5.4) imply c1 = b3c2 = 0. If b3 6= 0, then c2 = 0 and the algebra E1 is obtained. If b3 = 0, then by taking basis transformation as follows: 3 = e1, we derive that this algebra is isomorphic to the algebra E1. e′ 1 = e1, e′ 2 = e3, e′ 3 = e2 we get E1. (cid:3) Theorem 5.2. Let E be a three-dimensional evolution algebra with the table of multiplication (5.1) and let a2a3b1b3c1c2 6= 0. Then period of each basis elements of the algebra E is infinite if and only if equations (5.2) hold true. 10 ABROR KH. KHUDOYBERDIYEV, BAKHROM A. OMIROV, IZZAT QARALLEH Proof. It is sufficient to proof the part Only if. From equalities (5.2) we get Putting this restrictions to the equality a2 0 = a2 2b1 + a2 3c1 = a2 = − . , a3 = − c2 2b3 c2 1 b2 3c2 b2 1 2b1 + a2 3c1 = 0 we obtain 1c2 3c2 b2 2) 3c3 1 + b3 2(b2 1c3 b3 1 , which implies Similarly, we derive b2 3c3 1 + b3 1c2 2 = 0. b1 = − a2 3c1 a2 2 , b3 = − c2 1a3 c2 2 b2 1a2 b2 3 Applying induction we will prove the following: a2 2b1 a2 3 c1 = − c2 = − , ⇒ a2 3c3 2 + a3 2c2 1 = 0, ⇒ a2 2b3 3 + a3 3b2 1 = 0. (5.5) (5.6) (5.7) e[k] 1 = Ak,2e2 + Ak,3e3, e[k] 2 = Bk,1e1 + Bk,3e3, e[k] 3 = Ck,1e1 + Ck,2e2, k ≥ 3 (5.8) with recurrence expressions k−1,2b1 + A2 k−1,1a2 + B2 k−1,1a3 + C2 where A2,2 = a2, A2,3 = a3, B2,1 = b1, B2,3 = b3, C2,1 = c1, C2,2 = c2. k−1,3c2, Ak,3 = A2 k−1,3c1, Bk,3 = B2 k−1,2b1, Ck,2 = C2 k−1,2b3, A2 k−1,1a3, B2 k−1,1a2, C2 Ak,2 = A2 Bk,1 = B2 Ck,1 = C2 k−1,3c1 = 0, k−1,3c2 = 0, k−1,3b3 = 0, (5.9) In fact, the correctness of expressions (5.8) is equivalent to that each basis element of evolution algebra E has infinite period. From decompositions of e[3] i , 1 ≤ i ≤ 3 it is easy to see the correctness of (5.9) for k = 3, i.e. A3,2 = a2 B3,1 = b2 C3,1 = c2 3c2 = A2 3c1 = B2 2b1 = C2 2,3c2, A3,3 = a2 2,3c1, B3,3 = b2 2,2b1, C3,2 = c2 2b3 = A2 1a3 = B2 1a2 = C2 2,2b3, A2 2,1a3, B2 2,1a2, C2 2,2b1 + A2 2,1a2 + B2 2,1a3 + C2 2,3c1 = a2 2,3c2 = b2 2,2b3 = c2 2b1 + a2 1a2 + b2 1a3 + c2 3c1 = 0, 3c2 = 0, 2b3 = 0. Suppose that (5.8) and (5.9) are true for k. We will prove it for k + 1. The chain of equalities e[k+1] 1 = e[k] 1 · e[k] 1 = (Ak,2e2 + Ak,3e3) · (Ak,2e2 + Ak,3e3) = (A2 k,2b1 + A2 k,3c1)e1 + A2 k,3c2e2 + A2 k,2b3e3 deduces Ak+1,2 = A2 k,3c2, Ak+1,3 = A2 k,2b3. Applying induction assumption, that is, Ak,2 = A2 k−1,3c2, Ak,3 = A2 k−1,2b3, A2 k−1,2b1 + A2 k−1,3c1 = 0, we get A2 k,2b1 + A2 k,3c1 = A4 k−1,3c2 2b1 + A4 k−1,2b2 3c1 = A4 k−1,2b2 1 c2 1 2b1 + A4 c2 k−1,2b2 3c1 = A4 k−1,2 c2 1 (b3 1c2 2 + b2 3c3 1) The equality (5.5) implies A2 Similarly, one finds k,2b1 + A2 k,3c1 = 0. B2 k,1a2 + B2 k,3c2 = 0, C2 k,1a3 + C2 k,2b3 = 0. This works is supported by the Grant No.0251/GF3 of Education and Science Ministry of Republic of Kazakhstan and the Grant (RGA) No:11-018 RG/Math/AS I -- UNESCO FR: 3240262715. Acknowledgements (cid:3) FEW REMARKS ON EVOLUTION ALGEBRAS 11 References [1] Camacho L.M., G´omez J.R., Omirov B.A., Turdibaev R.M. Some properties of evolution algebras, Bulletin of Korean Mathematical Society (to appear), arXiv:1004.1987v1. [2] Camacho L.M., G´omez J.R., Omirov B.A., Turdibaev R.M. The derivations of some evolution algebras, Linear and Multilinear Algebra. doi: 10.1080/03081087.2012.678342. [3] Casas J.M., Ladra M., Omirov B.A., Rozikov U.A. On nilpotent index and dibaricity of evolution algebras, Linear Algebra Appl., vol. 439, 2013, p. 90 -- 105. [4] Casas J.M., Ladra M., Omirov B.A., Rozikov U.A. On evolution algebras, Algebra Colloquium (to apperar), arXiv:1004.1050v1. [5] Casas J. M., Ladra M., Rozikov U. A. Chain of evolution algebras, Linear Algebra Appl., vol. 435, 2011, no. 4, pp. 852 -- 870. [6] Ladra M., Omirov B.A., Rozikov U.A. On dibaric and evolution algebras, arXiv:1104.2578v1. [7] Lyubich Y.I. Mathematical structures in population genetics, Springer-Verlag, Berlin, 1992. [8] Reed M.L. Algebraic structure of genetic inheritance, Bull. Amer. Math. Soc. (N.S.) 34(2) (1997), 107130. [9] Rozikov U.A., Tian J.P. Evolution algebras generated by Gibbs measures, Lobachevskii Jour. Math. 32 (4) (2011) 270 -- 277. [10] Tian J.P. Evolution algebras and their applications, Lecture Notes in Math., 1921. Springer, Berlin, 2008. [11] Worz-Busekros A. Algebras in genetics, Lecture Notes in Biomathematics 36, Springer-Verlag, Berlin-New York, 1980. [A. Kh. Khudoyberdiyev and B. A. Omirov] Institute of Mathematics, National University of Uzbekistan, Tashkent, 100125, Uzbekistan. E-mail address: [email protected], [email protected] [Izzat Qaralleh] I Department of Computational & Theoretical Sciences, Faculty of Science, Interna- tional Islamic University Malaysia, P.O. Box, 141, 25710, Kuantan, Pahang, Malaysia E-mail address: izzat [email protected]
1001.4240	1	1001	2010-01-24T12:14:06	Paradigm of Nonassociative Hom-algebras and Hom-superalgebras	[ "math.RA", "math.RT" ]	The aim of this paper is to give a survey of nonassociative Hom-algebra and Hom-superalgebra structures. The main feature of these algebras is that the identities defining the structures are twisted by homomorphisms. We discuss Hom-associative algebras, Hom-Flexible algebras, Hom-Lie algebras, $G$-hom-associative algebras, Hom-Poisson algebras, Hom-alternative algebras and Hom-Jordan algebras and $\mathbb{Z}_2$-graded versions. We give an overview of the development of Hom-algebras structures which have been intensively investigated recently.	math.RA	math	PARADIGM OF NONASSOCIATIVE HOM-ALGEBRAS AND HOM-SUPERALGEBRAS ABDENACER MAKHLOUF Abstract The aim of this paper is to give a survey of nonassociative Hom-algebra and Hom-superalgebra structures. The main feature of these algebras is that the iden- tities defining the structures are twisted by homomorphisms. We discuss Hom- associative algebras, Hom-Flexible algebras, Hom-Lie algebras, G-hom-associative algebras, Hom-Poisson algebras, Hom-alternative algebras and Hom-Jordan alge- bras and Z2-graded versions. We give an overview of the development of Hom- algebras structures which has been intensively investigated recently. Keywords: Hom-associative algebra, Hom-Lie algebra, Hom-Lie superalge- bra, Hom-Poisson algebra, Hom-Lie admissible algebra, Hom-Lie admissible su- peralgebra, Hom-alternative algebra, Hom-Jordan algebra. 1 Introduction We mean by Hom-algebra or Hom-superalgebra a triple consisting of a vector space, respectively Z2-graded vector space, a multiplication and a homomorphism. The main feature of these algebras is that the identities defining the structures are twisted by homomorphisms. Such algebras appeared in the ninetieth in examples of q-deformations of the Witt and the Virasoro algebras. The study of nonassociative algebras was originally motivated by certain prob- lems in physics and other branches of mathematics. The first motivation to study nonassociative Hom-algebras comes from quasi-deformations of Lie alge- bras of vector fields, in particular q-deformations of Witt and Virasoro algebras [2, 7, 8, 9, 10, 11, 13, 25, 30, 24]. The deformed algebras arising in connection with σ-derivation are no longer Lie algebras. It was observed in the pioneering works that in these examples a twisted Jacobi identity holds. Motivated by these ex- amples and their generalization, Hartwig, Larsson and Silvestrov introduced and studied in [22, 27, 28, 29] the classes of quasi-Lie, quasi-Hom-Lie and Hom-Lie algebras. In the class of Hom-Lie algebras skew-symmetry is untwisted, whereas the Jacobi identity is twisted by a homomorphism and contains three terms as in 1 Lie algebras, reducing to ordinary Lie algebras when the twisting linear map is the identity map. They showed that Hom-Lie algebras are closely related to discrete and deformed vector fields and differential calculus and that some q-deformations of the Witt and the Virasoro algebras have the structure of Hom-Lie algebra. The Hom-associative algebras plays the role of associative algebras in the Hom- Lie setting. They were introduced by Makhlouf and Silvestrov in [32], where it is shown that the commutator bracket of a Hom-associative algebra gives rise to a Hom-Lie algebra and where a classification of Hom-Lie admissible algebras is established. Given a Hom-Lie algebra, a universal enveloping Hom-associative algebra was constructed by Yau in [44]. See also [15, 17, 18, 19, 34, 45] for other works on Hom-associative algebras. In a similar way Yau proposed in [44] a notion of Hom-dialgebra which gives rise to Hom-Leibniz algebra. While Hom-associative superalgebras and Hom-Lie superalgebras were studied in [4]. The Hom-alternative algebras and Hom-Jordan algebras which are twisted version of the ordinary alternative algebras and Jordan algebras were introduced by the author in [31]. Their properties are discussed and construction procedures using ordinary alternative algebras or Jordan algebras are provided. Also, it is shown that a plus Hom-algebra of a Hom-associative algebra leads to Hom-Jordan algebra. Beyond the binary algebras, a generalization of n-ary algebras to Hom situa- tion was studied in [5], n-ary Hom-algebras of Lie type and associative type are dis- cussed. Dualization of Hom-associative algebras leads to Hom-coassociative coal- gebras. The Hom-coassociative coalgebras, Hom-bialgebras and Hom-Hopf alge- bras were introduced in [33, 35]. A twisted version of Yang-Baxter equation, quasi- triangular bialgebras and quantum groups were discussed in [47, 48, 49, 50, 51]. A study from the monoidal category point of view is given in [15]. The main purpose of this paper is to summarize nonassociative Hom-algebra structures extending the ordinary nonassociative algebras to Hom-algebra setting. We set the definitions and properties, and provide some examples. In Section 2, we fix the notations and some general setting of Hom-algebras. The third Section con- cerns Hom-associative algebras, Hom-associative superalgebras, Hom-flexible al- gebras and Hom-dialgebras. Section 4 deals with Hom-Lie algebras, Hom-Leibniz, Hom-Poisson algebras and Hom-Lie superalgebras. It is shown in particular that there is an adjoint pair of functors between the category of Hom-associative alge- bras (resp. Hom-dialgebras) and the category of Hom-Lie algebras (resp. Hom- Leibniz algebras). On the other hand the supercommutator of a Hom-associative superalgebra determines a Hom-Lie superalgebra. The Hom-Poisson algebra struc- ture which emerges naturally from deformation theory of Hom-associative algebras (see [34]) is also given. We also present the definition of quasi-Lie algebras which offer to treat within the same framework the well known generalizations of Lie alge- bras, that is Lie superalgebras and color Lie algebras as well as their corresponding versions in Hom-superalgebra setting. Section 5 is dedicated to Hom-Lie admissi- ble algebras and Hom-Lie admissible superalgebras. Their classifications lead to G-Hom-associative algebras and G-Hom-associative superalgebras, which include in particular Hom-algebra versions of left-symmetric algebras (Vinberg algebras) 2 or their opposite algebras, that is right symmetric algebras (pre-Lie algebras). In Section 6, we discuss Hom-alternative algebras and study their properties. In par- ticular, we show that the twisted version of the associator is an alternating function of its arguments. We show that an ordinary alternative algebra and one of its al- gebra endomorphisms lead to a Hom-alternative algebra where the twisting map is actually the algebra endomorphism. This process was introduced in [45] for Lie and associative algebras and more generally G-associative algebras (see [32] for this class of algebras) and generalized to coalgebras in [35], [33] and to n-ary algebras of Lie and associative types in [5]. We derive examples of Hom-alternative alge- bras from 4-dimensional alternative algebras which are not associative and from algebra of octonions. In the last Section we introduce a notion of Hom-Jordan algebras and show that it fits with the Hom-associative structure, that is the plus Hom-associative algebra leads to Hom-Jordan algebra. ❧ 2 Preliminaries Throughout this paper K is a field of characteristic 0, V is a K-linear space or, when talking about superalgebras, V is a superspace. Let V be a superspace over K that is a Z2-graded K-linear space with a direct sum V = V0 ⊕ V1. The element of Vj, j = {0, 1}, are said to be homogenous and of parity j. The parity of a homogeneous element x is denoted by x. One may consider that K is any commutative ring and V a K-module. We mean by a Hom-algebra, respectively Hom-superalgebra, a triple (V, µ, α) where µ : V × V → V is a K-bilinear map and α : V → V is a K-linear map, respectively an even K-linear map. The type of the Hom-algebra or the Hom- superalgebra is fixed by the identity satisfied by the elements. Let (V, µ, α) and (V ′, µ′, α′) be two Hom-algebras (resp. Hom-superalgebras) of the same type, a morphism f : (V, µ, α) → (V ′, µ′, α′) is a linear map (resp. even linear map) f : V → V ′ such that f ◦ µ = µ′ ◦ (f × f ) and f ◦ α = α′ ◦ f . In some statements the multiplicativity of α is required, that is α ◦ µ = µ ◦ (α × α). We call such Hom-algebras, multiplicative Hom-algebras. ij , where µ(ei, ej) = Pn k=0 C k by n2 structure constants aij , where α(ei) = Pn ij , aij) to an algebraic variety imbedded in Kn3+n2 Let V be a n-dimensional K-linear space and {e1, · · · , en} be a basis of V . A Hom-algebra structure on V with product µ is determined by n3 structure constants C k ij ek and homomorphism α which is given j=0 aijej. If we require that the Hom-algebra is of a given type then this limits the set of structure constants (C k . The polynomials defining the algebraic variety being derived from the identities. A point in such an algebraic variety represents an n-dimensional Hom-algebra, along with a particular choice of basis. A change of basis may give rise to a different point of the algebraic variety. The group GL(n, K) acts on the algebraic varieties of Hom-structures by the so-called "transport of structure" action defined as follows: 3 Let A = (V, µ, α) be a Hom-algebra. Given f ∈ GL(n, K), the action f · A transports the structures for x, y ∈ V by (1) (2) f · µ(x, y) = f −1µ(f (x), f (y)) f · α(x) = f −1α(f (x)) The orbit of the Hom-algebra A is given by ϑ (A) = {A′ = f · A, f ∈ GLn (K)} . The orbits are in 1-1-correspondence with the isomorphism classes of n-dimensional algebras. The stabilizer subgroup of A (stab (A) = {f ∈ GLn (K) : A = f · A}) is exactly Aut (A) , the automorphism group of A. A subalgebra of a Hom-algebra (V, µ, α) is a triple (W, µ, α) where W is a subspace of V closed under µ and α. A subspace I of V is a two-sided ideal if µ(I, V ) ⊂ I and µ(V, I) ⊂ I. To any Hom-algebra A = (V, µ, α), we associate a minus Hom-algebra A− = 2 (µ(x, y) − µ(y, x)) and a plus Hom-algebra A+ = 2 (µ(x, y) + µ(y, x)). (V, µ−, α) where µ−(x, y) = 1 (V, µ+, α) where µ+(x, y) = 1 We call Hom-associator associated to a Hom-algebra (V, µ, α) the trilinear map asα defined for any x, y, z ∈ V by (3) asα(x, y, z) = µ(α(x), µ(y, z)) − µ(µ(x, y), α(z)). 3 Hom-associative algebras, Hom-Flexible algebras, Hom- dialgebras 3.1 Hom-associative algebras and superalgebras The notion of Hom-associative algebra was introduced in [32], while the Hom- associative algebras were discussed slightly in [4]. Definition 3.1 ([32]). A Hom-associative algebra is a triple (V, µ, α) consisting of a vector space V , a bilinear map µ : V × V → V and a linear map α : V → V satisfying for all x, y, z ∈ V (4) µ(α(x), µ(y, z)) = µ(µ(x, y), α(z)). In terms of Hom-associator (3), the identity (4) writes asα(x, y, z) = 0. Example 3.2. Let {e1, e2, e3} be a basis of a 3-dimensional linear space V over K. The following multiplication µ and linear map α on V define Hom-associative algebras over K3: µ(e1, e1) = a e1, µ(e1, e2) = µ(e2, e1) = a e2, µ(e1, e3) = µ(e3, x1) = b e3, µ(e2, e2) = a e2, µ(e2, e3) = b e3, µ(e3, e2) = µ(e3, e3) = 0, α(e1) = a e1, α(e2) = a e2, α(e3) = b e3, 4 where a, b are parameters in K. The algebras are not associative when a 6= b and b 6= 0, since µ(µ(e1, e1), e3)) − µ(e1, µ(e1, e3)) = (a − b)be3. Example 3.3 (Polynomial Hom-associative algebra [45]). Consider the polyno- mial algebra A = K[x1, · · · xn] in n variables. Let α be an algebra endomorphism of 1 , · · · xrn A which is uniquely determined by the n polynomials α(xi) = P λi;r1,··· ,rnxr1 n for 1 ≤ i ≤ n. Define µ by (5) µ(f, g) = f (α(x1), · · · α(xn))g(α(x1), · · · α(xn)) for f, g in A. Then, (A, µ, α) is a Hom-associative algebra. Example 3.4 ([46]). Let A = (V, µ, α) be a Hom-associative algebra. Then (Mn(A), µ′, α′), where Mn(A) is the vector space of n × n matrix with entries in V , is also a Hom-associative algebra in which the multiplication µ′ is given by matrix multiplication and µ and α′ is given by α in each entry. We define now Hom-associative superalgebras. Definition 3.5 ([4]). A Hom-associative superalgebra is a triple (V, µ, α) con- sisting of a superspace V , an even bilinear map µ : V × V → V and an even homomorphism α : V → V satisfying for all x, y, z ∈ V (6) µ(α(x), µ(y, z)) = µ(µ(x, y), α(z)) Remark 3.6. Some properties of Hom-associative algebras and unital Hom-asso- ciative algebras were discussed in [4, 5, 17, 18, 19, 34, 45]. A study from the point of view of monoidal category was provided in [15] and where a new kind of unital Hom-associative algebras is introduced. 3.2 Hom-Flexible algebras The flexible algebras was initiated by Albert [1] and investigated by number of authors Myung, Okubo, Laufer, Tomber and Santilli, see for example [38]. The notion of Hom-flexible algebra was introduced in [32]. We summarize in the following the definition and some characterizations. Definition 3.7 ([32]). A Hom-algebra A = (V, µ, α) is called flexible if and only if for any x, y in V (7) µ(α(x), µ(y, x))) = µ(µ(x, y), α(x)) The condition (7) may be written using the Hom-associator by asα(x, y, x) = 0. We recover the classical dialgebra when α is the identity map. Lemma 3.8. Let A = (V, µ, α) be a Hom-algebra. The following assertions are equivalent (1) A is flexible. (2) For any x, y in V , asα(x, y, x) = 0. (3) For any x, y, z in V , asα(x, y, z) = −asα(z, y, x) 5 Proof. The first equivalence follows from the definition. To prove the last equiva- lence, one writes asα(x − z, y, x − z) = 0 which is equivalent, using the linearity, to asα(x, y, z) + asα(z, y, x) = 0. (cid:3) Corollary 3.9. Any Hom-associative algebra is flexible. Let A = (V, µ, α) be a Hom-algebra. Let A− = (V, µ−, α) (resp. A+ = (V, µ+, α)) be the plus Hom-algebra (resp. minus Hom-algebra) with multiplica- tion defined for x, y ∈ V by µ−(x, y) = 1 2 (µ(x, y) − µ(y, x)) (resp. µ+(x, y) = 1 2 (µ(x, y) + µ(y, x))). We have the following characterization of Hom-flexible alge- bras. Proposition 3.10 ([32]). A Hom-algebra A = (V, µ, α) is Hom-flexible if and only if (8) µ−(α(x), µ+(y, z)) = µ+(µ−(x, y), α(z)) + µ+(α(y), µ−(x, z)) Proof. Let A be a Hom-flexible algebra. Then by lemma 3.8 it is equivalent to asα(x, y, z) + asα(z, y, x) = 0, for any x, y, z in V , where asα is the Hom-associator associated to A. This implies (9) asα(x, y, z) + asα(z, y, x) + asα(x, z, y) +asα(y, z, x) − asα(y, x, z) − asα(z, x, y) = 0 By expansion, the previous relation is equivalent to (8). Conversely, assume that we have the condition (8), by setting x = z in the equation (9), one gets asα(x, y, x) = 0, Therefore A is a Hom-flexible algebra. (cid:3) 3.3 Hom-dialgebras The Hom-dialgebra structure introduced by Yau extends to Hom-algebra set- ting the classical dialgebra structure introduced by Loday. Definition 3.11 ([44]). A Hom-dialgebra is a tuple (V, ⊣, ⊢, α), where ⊣, ⊢: V × V → V are bilinear maps and α : V → V is a linear map such that the following five identities are satisfied for x, y, z ∈ V (10) (11) (12) α(x) ⊣ (y ⊣ z) = (x ⊣ y) ⊣ α(z) = α(x) ⊣ (y ⊢ z) α(x) ⊢ (y ⊢ z) = (x ⊢ y) ⊢ α(z) = α(x) ⊢ (y ⊣ z) α(x) ⊢ (y ⊣ z) = (x ⊢ y) ⊣ α(z) We recover the classical dialgebra when α is the identity map. A morphism of Hom-dialgebras is a linear map that is compatible with α and the two multiplica- tions ⊢ and ⊣. Note that (V, ⊣, α) and (V, ⊢, α) are Hom-associative algebras. In the classical case, Loday showed that the commutator defined for x, y ∈ V by [x, y] = x ⊣ y − y ⊢ x defines a Leibniz algebra on V . This result is extended to Hom-algebra setting in the next Section.❧ 6 4 Hom-Leibniz algebras and Hom-Lie algebras 4.1 Hom-Leibniz algebras A class of quasi Leibniz algebras was introduced in [28] in connection to general quasi-Lie algebras following the standard Loday's conventions for Leibniz algebras (i.e. right Loday algebras). Definition 4.1 ([28]). A Hom-Leibniz algebra is a triple (V, [·, ·], α) consisting of a linear space V , bilinear map [·, ·] : V × V → V and a linear map α : V → V satisfying (13) [[x, y], α(z)] = [[x, z], α(y)] + [α(x), [y, z]] In terms of the (right) adjoint homomorphisms Ady : V → V defined by Ady(x) = [x, y], the identity (13) can be written as (14) Adα(z)([x, y]) = [Adz(x), α(y)] + [α(x), Adz (y)] or in pure operator form (15) Adα(z) ◦ Ady = Adα(y) ◦ Adz + AdAdz(y) ◦ α 4.2 Hom-Lie algebras The Hom-Lie algebras were initially introduced by Hartwig, Larson and Sil- vestrov in [22] motivated initially by examples of deformed Lie algebras coming from twisted discretizations of vector fields. Definition 4.2 ([22]). A Hom-Lie algebra is a triple (V, [·, ·], α) consisting of a linear space V , bilinear map [·, ·] : V × V → V and a linear map α : V → V satisfying (16) (17) [x, y] = −[y, x] (skewsymmetry) (cid:9)x,y,z [α(x), [y, z]] = 0 (Hom-Jacobi identity) for all x, y, z from V , where (cid:9)x,y,z denotes summation over the cyclic permutation on x, y, z. Using the skew-symmetry, one may write the Hom-Jacobi identity in the form (14). Hence, if a Hom-Leibniz algebra is skewsymmetric then it is a Hom-Lie algebra. Example 4.3. Let {e1, e2, e3} be a basis of a 3-dimensional linear space V over K. The following bracket and linear map α on V define a Hom-Lie algebra over K3: [e1, e2] = ae1 + be3 [e1, e3] = [e2, e3] = de1 + 2ae3, ce2 α(e1) = e1 α(e2) = 2e2 α(e3) = 2e3 7 with [e2, e1], [e3, e1] and [e3, e2] defined via skewsymmetry. It is not a Lie algebra if and only if a 6= 0 and c 6= 0, since [e1, [e2, e3]] + [e3, [e1, e2]] + [e2, [e3, e1]] = ace2. Example 4.4 (Jackson sl2(K)). In this example, we will consider the Hom-Lie algebra Jackson sl2(K) which is a Hom-Lie deformation of the classical Lie algebra sl2(K) defined by [e1, e2] = 2e2, [e1, e3] = −2e3, [e2, e3] = e1. This family of Hom- Lie algebras was constructed in [29] using a quasi-deformation scheme based on discretizing by means of Jackson q-derivations a representation of sl2(K) by one- dimensional vector fields (first order ordinary differential operators) and using the twisted commutator bracket defined in [22]. The Hom-Lie algebra Jackson sl2(K) is a 3-dimensional vector space over K with the vector space bases {e1, e2, e3} and with the bilinear bracket multiplication defined on the basis by [e1, e2]t = 2e2, [e1, e3]t = −2e3 − 2te3, [e2, e3]t = e1 + t 2 e1, and by skew-symmetry for [e2, e1]t, [e3, e1]t and [e3, e2]t. The linear map αt is defined by αt(e1) = e1, αt(e2) = 2 + t 2(1 + t) e2 = e2 + ∞X k=0 (−1)k 2 tk e2, αt(e3) = e3 + t 2 e3. Thus Jackson sl2(K) algebra is a Hom-Lie algebra deformation of sl2(K). See [34] for other examples of Hom-Lie algebra deformation of sl2(K). 4.3 Hom-Lie and Hom-Leibniz Functors We provide in the following a different way for constructing Hom-Lie alge- bras by extending the fundamental construction of Lie algebras from associative algebras via commutator bracket. Theorem 4.5 ([32]). Let (V, µ, α) be a Hom-associative algebra. The Hom-algebra (V, [·, ·], α), where the bracket is defined for all x, y ∈ V by is a Hom-Lie algebra. [x, y] = µ(x, y) − µ(y, x) Proof. The bracket is obviously skewsymmetric and with a direct computation we have [α(x), [y, z]] − [[x, y], α(z)] − [α(y), [y, z]] = µ(α(x), µ(y, z)) − µ(α(x), µ(z, y)) − µ(µ(y, z), α(x)) + µ(µ(z, y), α(x)) −µ(µ(x, y), α(z)) + µ(µ(y, x), α(z)) + µ(α(z), µ(x, y)) − µ(α(z), µ(y, x)) −µ(α(y), µ(x, z)) + µ(α(y), µ(z, x)) + µ(µ(x, z), α(y)) − µ(µ(z, x), α(y)) = 0 (cid:3) A similar construction is obtained for Hom-dialgebra. 8 Theorem 4.6 ([44]). Let (V, ⊣, ⊢, α) be a Hom-dialgebra (V, µ, α). The Hom- algebra (V, [·, ·], α), where the bracket is defined for all x, y ∈ V by is a Hom-Leibniz algebra. [x, y] = x ⊣ y − y ⊢ x Therefore, we have a functor HLie (resp. HLeib) from the category of Hom- associative algebras HomAs (resp. category of Hom-dialgebras HomDi) to the category of Hom-Lie algebras HomLie (resp. category of Hom-Leibniz algebras HomLeib). Conversely, an enveloping Hom-associative algebra UHLie(g) (resp. enveloping Hom-dialgebra UHLeib(L)) of a Hom-Lie algebra g (resp. Hom-Leibniz algebra L) are constructed in [44]. Hence, UHLie is the left adjoint functor of HLie and UHLeib is the left adjoint functor of HLeib. 4.4 Hom-Poisson algebras We introduce in the following the notion of Hom-Poisson structure which emerges naturally in deformation theory of Hom-associative algebras, see [34]. Definition 4.7 ([34]). A Hom-Poisson algebra is a quadruple (V, µ, {·, ·}, α) con- sisting of a vector space V , bilinear maps µ : V × V → V and {·, ·} : V × V → V , and a linear map α : V → V satisfying (1) (V, µ, α) is a commutative Hom-associative algebra, (2) (V, {·, ·}, α) is a Hom-Lie algebra, (3) for all x, y, z in V , (18) {α(x), µ(y, z)} = µ(α(y), {x, z}) + µ(α(z), {x, y}). Condition (18) expresses the compatibility between the multiplication and the Poisson bracket. It can be reformulated equivalently, for all x, y, z in V , as (19) {µ(x, y), α(z)} = µ({x, z}, α(y)) + µ(α(x), {y, z}) Note that in this form it means that adz(·) = {·, z} is a sort of generalization of a derivation of an associative algebra, and also it resembles the identity (13) in the definition of Leibniz algebra. We recover the classical Leibniz identity when α is the identity map. Example 4.8. Let {e1, e2, e3} be a basis of a 3-dimensional vector space V over K. The following multiplication µ, skew-symmetric bracket and linear map α on V define a Hom-Poisson algebra over K3: µ(e1, e1) = e1, µ(e1, e2) = µ(e2, e1) = e3, {e1, e2} = ae2 + be3, {e1, e3} = ce2 + de3, α(e1) = λ1e2 + λ2e3, α(e2) = λ3e2 + λ4e3, α(e3) = λ5e2 + λ6e3 where a, b, c, d, λ1, λ2, λ3, λ4, λ5, λ6 are parameters in K. 9 4.5 Hom-Lie Superalgebras Hom-Lie Superalgebras is a subclass of quasi-Lie algebras introduced in [28]. They were studied in [4], where construction procedures are provided. Definition 4.9. A Hom-Lie superalgebra is a triple (V, [·, ·], α) consisting of a superspace V , an even bilinear map [·, ·] : V × V → V and an even superspace homomorphism α : V → V satisfying (20) [x, y] = −(−1)xy[y, x] (21) (−1)xz[α(x), [y, z]] + (−1)zy[α(z), [x, y]] + (−1)yx[α(y), [z, x]] = 0 for all homogeneous element x, y, z in V . The identity (21) is called Hom-superJacobi identity, while the identity (21) expresses the usual supersymmetry of the bracket. Remark 4.10. We recover the classical Lie superalgebra when α = id. The Hom- Lie algebras algebras are obtained when the part of parity one is trivial. Example 4.11 (2-dimensional abelian Hom-Lie superalgebra). Every bilinear map µ on a 2- dimensional linear superspace V = V0 ⊕ V1, where V0 is gener- ated by x and V1 is generated by y and such that [x, y] = 0 defines a Hom-Lie superalgebra for any homomorphism α of superalgebra. Indeed, the graded Hom- Jacobi identity is satisfied for any triple (x, x, y). Example 4.12 (Affine Hom-Lie superalgebra ). Let V = V0⊕V1 be a 3-dimensional superspace where V0 is generated by {e1, e2} and V1 is generated by e3. The triple (V, [·, ·], α) is a Hom-Lie superalgebra defined by [e1, e2] = e1, [e1, e3] = [e2, e3] = [e3, e3] = 0 and α is any homomorphism. Example 4.13 (A q-deformed Witt superalgebra, [4]). We provide an example of infinite dimensional Hom-Lie superalgebra which is given by a realization of the q-deformed Witt superalgebra constructed in [4]. It corresponds to a superspace V generated by the elements {Xn}n∈N of parity 0 and the elements {Gn}n∈N of parity 1. Let q ∈ C\{0, 1} and n ∈ N, we set {n} = 1−qn 1−q , a q-number. The q-numbers have the following properties {n + 1} = 1 + q{n} = {n} + qn and {n + m} = {n} + qn{m}. Let [−, −]σ be a bracket on the superspace V defined by [Xn, Xm]σ = ({m} − {n})Xn+m [Xn, Gm]σ = (qn{m + 1} − qm+1{n})Gn+m The others brackets are obtained by supersymmetry or are 0. Let α be an even linear map on V defined on the generators by α(Xn) = (1 + qn)Xn α(Gn) = (1 + qn+1)Gn Then the triple (V, [−, −]σ, α) is a Hom-Lie superalgebra. 10 In the following, we show that the supercommutator bracket defined using the multiplication in a Hom-associative superalgebra leads naturally to Hom-Lie superalgebra. Theorem 4.14 ([4]). Let (V, µ, α) be a Hom-associative superalgebra. The Hom- superalgebra (V, [·, ·], α), where the bracket (super bracket) is defined for all homo- geneous elements x, y ∈ V by [x, y] = µ(x, y) − (−1)xyµ(y, x) and extended by linearity to all elements, is a Hom-Lie superalgebra. Proof. The bracket is obviously supersymmetric and with a direct computation we have (−1)xz[α(x), [y, z]] + (−1)zy[α(z), [x, y]] + (−1)yx[α(y), [z, x]] = (−1)xzµ(α(x), µ(y, z)) − (−1)xz+yzµ(α(x), µ(z, y)) −(−1)xyµ(µ(y, z), α(x)) + (−1)xy+yzµ(µ(z, y), α(x)) +(−1)yxµ(α(y), µ(z, x)) − (−1)xy+zxµ(α(y), µ(x, z)) −(−1)yzµ(µ(z, x), α(y)) + (−1)yz+zxµ(µ(x, z), α(y)) +(−1)zyµ(α(z), µ(x, y)) − (−1)yz+xyµ(α(z), µ(y, x)) −(−1)zxµ(µ(x, y), α(z)) + (−1)zx+xyµ(µ(y, x), α(z)) = 0 (cid:3) The following theorem gives a way to construct Hom-Lie superalgebras, start- ing from a regular Lie superalgebra and an even superalgebra endomorphism. Theorem 4.15 ([4]). Let (V, [·, ·]) be a Lie superalgebra and α : V → V be an even Lie superalgebra endomorphism. Then (V, [·, ·]α, α), where [x, y]α = α([x, y]), is a Hom-Lie superalgebra. Moreover, suppose that (V ′, [·, ·]′) is another Lie superalgebra and α′ : V ′ → V ′ is a Lie superalgebra endomorphism. If f : V → V ′ is a Lie superalgebra morphism that satisfies f ◦ α = α′ ◦ f then is a morphism of Hom-Lie superalgebras. f : (V, [·, ·]α, α) −→ (V ′, [·, ·]′, α′) Example 4.16 ([4]). We construct an example of Hom-Lie superalgebra, which is not a Lie superalgebra starting from the orthosymplectic Lie superalgebra. We consider in the sequel the matrix realization of this Lie superalgebra. Let osp(1, 2) = V0 ⊕ V1 be the Lie superalgebra where V0 is generated by: 0 0 0 0 0 0   , 0 0 0 0 1 0 0 0 0   , X =     , G =   0 0 0 0 1 0 0 1 0 0 0 1   , Y =     . 1 0 0 −1 0 0 0 0 0 H =   1 0 0 0 0 0 0 0 −1 and V1 is generated by: F =   11 The defining relations (we give only the ones with non zero values in the right hand side) are [H, X] = 2X, [H, Y ] = −2Y, [X, Y ] = H, [Y, G] = F, [X, F ] = G, [H, F ] = −F, [H, G] = G, [G, F ] = H, [G, G] = −2X, [F, F ] = 2Y. Let λ ∈ R∗, we consider the linear map αλ : osp(1, 2) → osp(1, 2) defined by: αλ(X) = λ2X, αλ(Y ) = 1 λ2 Y, αλ(H) = H, αλ(F ) = 1 λ F, αλ(G) = λG. We provide a family of Hom-Lie superalgebras osp(1, 2)λ = (osp(1, 2), [·, ·]αλ , αλ) where 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. These Hom-Lie superalgebras are not Lie superalgebras for λ 6= 1. Indeed, the left hand side of the superJacobi identity (21), for α = id, leads to [X, [Y, H] − [H, [X, Y ]] + [Y, [H, X]] = and also [H, [F, F ] − [F, [H, F ]] + [F, [F, H]] = Then, they do not vanish for λ 6= 1. 2(1 − λ4) λ2 Y, 4(λ − 1) λ4 Y. 4.6 Quasi-Lie algebras The class of quasi-Lie algebras where introduced by Larsson and Silvestrov in order to treat within the same framework such a well known generalizations of Lie algebras as color and Lie superalgebras, as well as Hom-Lie algebras (see [27, 28]). Let LK(V ) be the set of linear maps of the linear space L over the field K. Definition 4.17 ( [28]). A quasi-Lie algebra is a tuple (V, [·, ·], α, β, ω, θ) consisting of • V is a linear space over K, • [·, ·] : V × V → V is a bilinear map called a product or bracket in V ; • α, β : V → V , are linear maps, • ω : Dω → LK(V ) and θ : Dθ → LK(V ) are maps with domains of definition Dω, Dθ ⊆ V × V , such that the following conditions hold: 12 • (ω-symmetry) The product satisfies a generalized skew-symmetry condition [x, y] = ω(x, y)[y, x], for all (x, y) ∈ Dω; • (quasi-Jacobi identity) The bracket satisfies a generalized Jacobi identity (cid:9)x,y,z (cid:8) θ(z, x)(cid:0)[α(x), [y, z]] + β[x, [y, z]](cid:1)(cid:9) = 0, for all (z, x), (x, y), (y, z) ∈ Dθ. Note that (ω(x, y)ω(y, x) − id)[x, y] = 0, if (x, y), (y, x) ∈ Dω, which follows from the computation [x, y] = ω(x, y)[y, x] = ω(x, y)ω(y, x)[x, y]. The class of Quasi-Lie algebras incorporates as special cases hom-Lie algebras and more general quasi-hom-Lie algebras (qhl-algebras) which appear naturally in the algebraic study of σ-derivations (see [22]) and related deformations of infinite- dimensional and finite-dimensional Lie algebras. To get the class of qhl-algebras one specifies θ = ω and restricts attention to maps α and β satisfying the twisting condition [α(x), α(y)] = β ◦ α[x, y]. Specifying this further by taking Dω = V × V , β = id and ω = −id, one gets the class of Hom-Lie algebras including Lie algebras when α = id. The class of quasi-Lie algebras contains also color Lie algebras and in particular Lie superalgebras.❧ 5 Hom-Lie admissible algebras and G-Hom-associative alge- bras The Lie-admissible algebras was introduced by A. A. Albert in 1948. Physi- cists attempted to introduce this structure instead of Lie algebras. For instance, the validity of Lie-Admissible algebras for free particles is well known. These alge- bras arise also in classical quantum mechanics as a generalization of conventional mechanics (see [1, 38]). 5.1 Hom-Lie admissible algebras In this section, we discuss the concept of Hom-Lie-Admissible algebra, extend- ing to Hom-algebra setting, the classical notion of Lie-admissible algebra. Definition 5.1 ([32]). Let A = (V, µ, α) be a Hom-algebra structure. Then A is said to be Hom-Lie-admissible algebra if the bracket defined for all x, y ∈ V by satisfies the Hom-Jacobi identity (17). [x, y] = µ(x, y) − µ(y, x) Remark 5.2. Since the bracket is also skewsymmetric then it defines a Hom-Lie algebra. Remark 5.3. Note that any Hom-associative and Hom-Lie algebras are Hom-Lie- admissible. 13 We aim now to give another characterization of Hom-Lie admissible algebras. Let A = (V, µ, α) be a Hom-algebra. We denote by brackets the multiplication of plus Hom-algebra A+, that is [x, y] = µ(x, y) − µ(y, x) and by S the trilinear map defined for x, y, z ∈ V by S(x, y, z) := asα(x, y, z) + asα(y, z, x) + asα(z, x, y) We have the following properties Lemma 5.4. S(x, y, z) = [µ(x, y), α(z)] + [µ(y, z), α(x)] + [µ(z, x), α(y)] Proof. [µ(x, y), α(z)] + [µ(y, z), α(x)] + [µ(z, x), α(y)] = µ(µ(x, y), α(z)) − µ(α(z), µ(x, y)) + µ(µ(y, z), α(x)) − µ(α(x), µ(y, z))+ µ(µ(z, x), α(y)) − µ(α(y), µ(z, x)) = S(x, y, z) (cid:3) Proposition 5.5. A Hom-algebra A is Hom-Lie-admissible if and only if it sat- isfies for any x, y, z in V Proof. S(x, y, z) = S(x, z, y) S(x, y, z) − S(x, z, y) = [µ(x, y), α(z)] + [µ(y, z), α(x)] + [µ(z, x), α(y)] −[µ(x, z), α(y)] − [µ(z, y), α(x)] − [µ(y, x), α(z)] = − (cid:9)x,y,z [α(x), [y, z]] (cid:3) 5.2 G-Hom-associative algebras In the following, we explore some other Hom-Lie-Admissible algebras, extend- ing to Hom-algebra setting the results obtained in [21]. Definition 5.6 ([32]). Let G be a subgroup of the permutations group S3, a Hom- algebra (V, µ, α) is called G-Hom-associative if for any x, y, z ∈ V , we have (−1)ε(σ)µ(µ(xσ(1), xσ(2)), α(xσ(3))) − µ(α(xσ(1)), µ(xσ(2), xσ(3))) = 0 (22) X σ∈G (23) where xi are in V and (−1)ε(σ) is the signature of the permutation σ. The condition (22) may be written in terms of Hom-associator by X σ∈G (−1)ε(σ)asα ◦ σ = 0 where σ is the extension of the permutation, still denoted by the same notation, to a trilinear map defined by σ(x1, x2, x3) = (xσ(1), xσ(2), xσ(3)). 14 Remark 5.7. If µ is the multiplication of a Hom-Lie-admissible Lie algebra then the condition (22) is equivalent to the property that the bracket defined by [x, y] = µ(x, y) − µ(y, x) satisfies the Hom-Jacobi condition or equivalently to (−1)ε(σ)µ(µ(xσ(1), xσ(2)), α(xσ(3))) − µ(α(xσ(1)), µ(xσ(2), xσ(3))) = 0 (24) X σ∈S3 which may be written as (25) X σ∈S3 (−1)ε(σ)asα ◦ σ = 0. Theorem 5.8 ([32]). Let G be a subgroup of the permutations group S3. Then any G-Hom-associative algebra is a Hom-Lie-admissible algebra. Proof. The skewsymmetry follows straightaway from the definition. We have a subgroup G in S3. Take the set of conjugacy class {gG}g∈I where I ⊆ G, and for any σ1, σ2 ∈ I, σ1 6= σ2 ⇒ σ1GT σ1G = ∅. Then X σ∈S3 (−1)ε(σ)asα ◦ σ = X X σ1∈I σ2∈σ1G (−1)ε(σ2)asα ◦ σ2 = 0. (cid:3) The result says that for any subgroup of S3 corresponds a class of G-Hom- associative algebra. Since the subgroups are G1 = {Id}, G2 = {Id, σ12}, G3 = {Id, σ23}, G4 = {Id, σ13}, G5 = A3, G6 = S3, where A3 is the alternating group and where σij is the transposition between i and j, then we obtain the following type of Hom-Lie-admissible algebras. • The G1-Hom-associative algebras are the Hom-associative algebras defined above. • The G2-Hom-associative algebras satisfy the condition µ(α(x), µ(y, z)) − µ(α(y), µ(x, z)) = µ(µ(x, y), α(z)) − µ(µ(y, x), α(z)) When α is the identity the algebra is called Vinberg algebra or left sym- metric algebra. • The G3-Hom-associative algebras satisfy the condition µ(α(x), µ(y, z)) − µ(α(x), µ(z, y)) = µ(µ(x, y), α(z)) − µ(µ(x, z), α(y)) When α is the identity the algebra is called pre-Lie algebra or right sym- metric algebra. • The G4-Hom-associative algebras satisfy the condition µ(α(x), µ(y, z)) − µ(α(z), µ(y, x)) = µ(µ(x, y), α(z)) − µ(µ(z, y), α(x)) • The G5-Hom-associative algebras satisfy the condition µ(α(x), µ(y, z)) + µ(α(y), µ(z, x) + µ(α(z), µ(x, y)) = µ(µ(x, y), α(z)) + µ(µ(y, z), α(x)) + µ(µ(z, x), α(y)) 15 If the product µ is skewsymmetric then the previous condition is exactly the Hom-Jacobi identity. • The G6-Hom-associative algebras are the Hom-Lie-admissible algebras. Special cases of G-Hom-associative algebras include generalization of Vinberg and pre-Lie algebras. Definition 5.9 ([32]). A Hom-Vinberg algebra (Hom-left-symmetric algebra) is a triple (V, µ, α) consisting of a linear space V , a bilinear map µ : V × V → V and a homomorphism α satisfying (26) µ(α(x), µ(y, z)) − µ(α(y), µ(x, z)) = µ(µ(x, y), α(z)) − µ(µ(y, x), α(z)) Definition 5.10 ([32]). A Hom-pre-Lie algebra (Hom-right-symmetric algebra) is a triple (V, µ, α) consisting of a linear space V , a bilinear map µ : V × V → V and a homomorphism α satisfying (27) µ(α(x), µ(y, z)) − µ(α(x), µ(z, y)) = µ(µ(x, y), α(z)) − µ(µ(x, z), α(y)) Remark 5.11. A Hom-pre-Lie algebra is the opposite algebra of a Hom-Vinberg algebra. Remark 5.12. The multiplicative Hom-Novikov algebras which are multiplicative Hom-Vinberg algebra with the additional identity µ(µ(x, y), α(z)) = µ(µ(x, z), α(y)), were studied in [52]. The following theorem states that G-associative algebras deform into G-Hom- associative algebras along any algebra endomorphism. Therefore, it provides a construction procedure. Theorem 5.13 ([45]). Let (V, µ) be a G-associative algebra and α : V → V be an algebra endomorphism. Then (V, µα, α), where µα = α◦ µ, is a G-Hom-associative algebra. Moreover, suppose that (V ′, µ′) is another G-associative algebra and α′ : V ′ → If f : V → V ′ is an algebras morphism that V ′ is an algebra endomorphism. satisfies f ◦ α = α′ ◦ f then is a morphism of G-Hom-associative algebras. f : (V, µα, α) −→ (V ′, µ′ α′ , α′) 5.3 Hom-Lie-Admissible Superalgebras We discuss in this section the concept of Hom-Lie-Admissible superalgebras studied in [4]. This study borders also an extension to graded case of the Lie- admissible algebras discussed in [21]. Let A = (V, µ, α) be a Hom-superalgebra, that is a superspace V with an even bilinear map µ and an even linear map α satisfying eventually identities. Let [x, y] = µ(x, y) − (−1)xyµ(y, x), for all homogeneous element x, y ∈ V , be the associated supercommutator. The bracket is extended to all elements by linearity. 16 Definition 5.14 ([4]). Let A = (V, µ, α) be a Hom-superalgebra on V defined by an even multiplication µ and an even homomorphism α. Then A is said to be Hom-Lie admissible superalgebra if the bracket defined for all homogeneous element x, y ∈ V by (28) [x, y] = µ(x, y) − (−1)xyµ(y, x) satisfies the Hom-superJacobi identity (21). Remark 5.15. Since the supercommutator bracket (28) is always supersymmetric, this makes any Hom-Lie admissible superalgebra into a Hom-Lie superalgebra. Remark 5.16. As mentioned in in the proposition (4.14), any associative super- algebra is a Hom-Lie admissible superalgebra. Lemma 5.17. Let A = (V, µ, α) be a Hom-superalgebra and [·, ·] be the associated supercommutator then (29) (cid:9)x,y,z (−1)xz[α(x), [y, z]] = (−1)xzasα(x, y, z) + (−1)yxasα(y, z, x) +(−1)zyasα(z, x, y) − (−1)xz+yzasα(x, z, y) −(−1)xy+yzasα(z, y, x) − (−1)xy+xzasα(y, x, z) Proof. By straightforward calculation. (cid:3) Remark 5.18. If α = id, then we obtain a formula expressing the left hand side of the classical superJacobi identity in terms of classical associator. In the following we aim to characterize the Hom-Lie admissible superalgebras in terms of Hom-associator. We introduce a trilinear map eS defined for homoge- neous elements x, y, z ∈ V by eS(x, y, z) := (−1)xzasα(x, y, z) + (−1)yxasα(y, z, x) + (−1)zyasα(z, x, y). Then, we have the following properties. Proposition 5.19 ([4]). Let A = (V, µ, α) be a Hom-superalgebra, then A is a Hom-Lie admissible superalgebra if and only if it satisfies (30) eS(x, y, z) = (−1)xy+xz+yzeS(x, z, y) for any homogenous elements x, y, z ∈ V. 17 Proof. We have eS(x, y, z) − (−1)xy+xz+yzS(x, z, y) = (−1)xzasα(x, y, z) + (−1)yxasα(y, z, x) + (−1)zyasα(z, x, y) −(−1)xy+xz+yz((−1)xyasα(x, z, y) + (−1)zxasα(z, y, x) +(−1)yzasα(y, x, z)) = (−1)xzasα(x, y, z) + (−1)yxasα(y, z, x) + (−1)zyasα(z, x, y) −(−1)xz+yzasα(x, z, y) − (−1)xy+yzasα(z, y, x) −(−1)xy+xzasα(y, x, z) =(cid:9)x,y,z (−1)xz[x, [y, z]] (lemma 5.17) Then the Hom-superJacobi identity (21) is satisfied if and only if the condition (30) holds. (cid:3) 5.4 G-Hom-associative superalgebras A classification of Hom-Lie admissible superalgebras using the symmetric group S3 is provided in [4]. We extended to Z2-graded case the notion of G-Hom- associative algebras and in particular G-associative algebras which was introduced in the classical ungraded Lie case in ([21]) and developed for the Hom-Lie case in ([32]). Let S3 be the permutation group generated by the transpositions σ1, σ2. We extend a permutation τ ∈ S3 to a map τ : V ×3 → V ×3 defined for x1, x2, x3 ∈ V by τ (x1, x2, x3) = (xτ (1), xτ (2), xτ (3)). We keep for simplicity the same notation. In particular, σ1(x1, x2, x3) = (x2, x1, x3) and σ2(x1, x2, x3) = (x1, x3, x2). We introduce a notion of a parity of transposition σi where i ∈ {1, 2}, by setting σi(x1, x2, x3) = xixi+1. We assume that the parity of the identity is 0 and for the composition σiσj , it is defined by σiσj(x1, x2, x3) = σj (x1, x2, x3) + σi(σj(x1, x2, x3)) = σj (x1, x2, x3) + σi(xσj (1), xσj (2), xσj (3)) We define by induction the parity for any composition. For the elements id, σ1, σ2, σ1σ2, σ2σ1, σ2σ1σ2 of S3, we obtain id(x1, x2, x3) = 0, σ1(x1, x2, x3) = x1x2, σ2(x1, x2, x3) = x2x3, σ1σ2(x1, x2, x3) = x2x3 + x1x3, σ2σ1(x1, x2, x3) = x1x2 + x1x3, σ2σ1σ2(x1, x2, x3) = x2x3 + x1x3 + x1x2. 18 Now, we express the condition of Hom-Lie admissibility of a Hom-superalgebra using permutations. Lemma 5.20 ([4]). A Hom-superalgebra A = (V, µ, α) is a Hom-Lie admissible superalgebra if the following condition holds (31) X τ ∈S3 (−1)ε(τ )(−1)τ (x1,x2,x3)asα ◦ τ (x1, x2, x3) = 0 where xi are in V , (−1)ε(τ ) is the signature of the permutation τ and τ (x1, x2, x3) is the parity of τ . Proof. By straightforward calculation, the associated supercommutator bracket satisfies (cid:9)x1,x2,x3 (−1)x1x3[α(x1), [x2, x3]] = (−1)x1x3 X τ ∈S3 (−1)ε(τ )(−1)τ (x1,x2,x3)asα ◦ τ (x1, x2, x3). (cid:3) It turns out that, for the associated supercommutator of a Hom-superalgebra, the Hom-superJacobi identity (21) is equivalent to X τ ∈S3 (−1)ε(τ )(−1)τ (x1,x2,x3)asα ◦ τ (x1, x2, x3) = 0. We introduce now the notion of G-Hom-associative superalgebra, where G is a subgroup of the permutations group S3. Definition 5.21 ([4]). Let G be a subgroup of the permutations group S3, a Hom- superalgebra on V defined by the multiplication µ and a homomorphism α is said to be G-Hom-associative superalgebra if (32) X τ ∈G (−1)ε(τ )(−1)τ (x1,x2,x3)asα ◦ τ (x1, x2, x3) = 0. where xi are in V , (−1)ε(τ ) is the signature of the permutation and τ (x1, x2, x3) is the parity of τ defined above. In particular, we call G-associative superalgebra a G-Hom-associative super- algebra where α is the identity map. The following result is a graded version of the results obtained in ([21, 32]). Theorem 5.22 ([4]). Let G be a subgroup of the permutations group S3. Then any G-Hom-associative superalgebra is a Hom-Lie admissible superalgebra. Proof. The supersymmetry follows straightaway from the definition. We have a subgroup G in S3. Take the set of conjugacy class {gG}g∈I where I ⊆ G, and for any τ1, τ2 ∈ I, σ1 6= τ2 ⇒ τ1GT τ2G = ∅. Then 19 X X τ ∈S3 τ1∈I (−1)ε(τ )(−1)τ (x1,x2,x3)asα ◦ τ (x1, x2, x3) = X (−1)ε(τ2)(−1)τ2(x1,x2,x3)asα ◦ τ2(x1, x2, x3) = 0 τ2∈τ1G where (x1, x2, x3) ∈ V , with V the underlaying superspace of the G-Hom-associative superalgebra. (cid:3) It follows that in particular, we have: Corollary 5.23. Let G be a subgroup of the permutations group S3. Then any G-associative superalgebra is a Lie admissible superalgebra. Now, we provide a classification of the Hom-Lie admissible superalgebras through G-Hom-associative superalgebras. The subgroups of S3, which are G1 = {Id}, G2 = {Id, σ1}, G3 = {Id, σ2}, G4 = {Id, σ2σ1σ2}, G5 = A3, G6 = S3, where A3 is the alternating group, lead to the following six classes of Hom-Lie admissible superalgebras. • The G1-Hom-associative superalgebras are the Hom-associative superal- gebras defined above. • The G2-Hom-associative superalgebras satisfy the identity µ(α(x), µ(y, z)) − µ(µ(x, y), α(z)) = (−1)xy(µ(α(y), µ(x, z)) − µ(µ(y, x), α(z))) • The G3-Hom-associative superalgebras satisfy the identity µ(α(x), µ(y, z)) − µ(µ(x, y), α(z)) = (−1)yz(µ(α(x), µ(z, y)) − µ(µ(x, z), α(y))) • The G4-Hom-associative superalgebras satisfy the identity µ(α(x), µ(y, z)) − µ(µ(x, y), α(z)) = (−1)xy+xz+yz(µ(α(z), µ(y, x)) − µ(µ(z, y), α(x))) • The G5-Hom-associative superalgebras satisfy the identity −(−1)xz+yz(µ(α(z), µ(x, y)) + µ(µ(z, x), α(y))) = µ(α(x), µ(y, z)) − µ(µ(x, y), α(z)) +(−1)xy+xz(µ(α(y), µ(z, x) + µ(µ(y, z), α(x))) • The G6-Hom-associative superalgebras are the Hom-Lie admissible super- algebras. Remark 5.24. Moreover, if in the previous identities we consider α = id, then we obtain a classification of Lie-admissible superalgebras. 20 Remark 5.25. A G2-Hom-associative (resp. G2-associative) superalgebras might be called Hom-Vinberg superalgebras (resp. Vinberg superalgebras) or Hom-left- symmetric superalgebras (resp. left-symmetric superalgebras). Similarly, A G3-Hom-associative (resp. G3-associative) superalgebras might be called Hom-pre-Lie superalgebras (resp. pre-Lie superalgebras) or Hom-right- symmetric superalgebras (resp. right-symmetric superalgebras). Notice that a Hom-pre-Lie superalgebra is the opposite algebra of a Hom- Vinberg superalgebra. Therefore, they actually form a same class. The following result generalizes the theorem 4.15 to any G-Hom-associative superalgebra. Theorem 5.26. Let (V, [·, ·]) be a G-associative superalgebra and α : V → V be an even G-Hom-associative superalgebra endomorphism. Then (V, [·, ·]α, α), where [x, y]α = α([x, y]), is a G-Hom-associative superalgebra. Moreover, suppose that (V ′, [·, ·]′) is another G-associative superalgebra and α′ : V ′ → V ′ is a G-associative superalgebra endomorphism. If f : V → V ′ is a G-associative superalgebra morphism that satisfies f ◦ α = α′ ◦ f then is a morphism of G-Hom-associative superalgebras. f : (V, [·, ·]α, α) −→ (V ′, [·, ·]′, α′) Proof. Similar to proof of theorem 4.15. (cid:3) 6 Hom-alternative algebras Now, we introduce and discuss the notion of Hom-alternative algebra, following [31]. Definition 6.1 ([31]). A left Hom-alternative algebra (resp. right Hom-alternative algebra) is a triple (V, µ, α) consisting of a K-linear space V , a linear map α : V → V and a multiplication µ : V ⊗ V → V satisfying, for any x, y in V , the left Hom- alternative identity, that is (33) µ(α(x), µ(x, y)) = µ(µ(x, x), α(y)), respectively, right Hom-alternative identity, that is (34) µ(α(x), µ(y, y)) = µ(µ(x, y), α(y)). A Hom-alternative algebra is one which is both left and right Hom-alternative algebra. Remark 6.2. Any Hom associative algebra is a Hom-alternative algebra. Using the Hom-associator (3), the condition (33) (resp. (34)) may be written using Hom-associator respectively asα(x, x, y) = 0, asα(y, x, x) = 0. 21 ❧ By linearization, we have the following equivalent definition of left and right Hom-alternative algebras. Proposition 6.3. A triple (V, µ, α) is a left Hom-alternative algebra (resp. right alternative algebra) if and only if the identity (35) µ(α(x), µ(y, z)) − µ(µ(x, y), α(z)) + µ(α(y), µ(x, z)) − µ(µ(y, x), α(z)) = 0. respectively, (36) µ(α(x), µ(y, z)) − µ(µ(x, y), α(z)) + µ(α(x), µ(z, y)) − µ(µ(x, z), α(y)) = 0. holds. Proof. We assume that, for any x, y, z ∈ V , asα(x, x, z) = 0 (left alternativity), then we expand asα(x + y, x + y, z) = 0. The proof for right Hom-alternativity is obtained by expanding asα(x, y + z, y + z) = 0. Conversely, we set x = y in (35), respectively y = z in (36). (cid:3) Remark 6.4. The multiplication could be considered as a linear map µ : V ⊗ V → V , then the condition (35) (resp. (36)) writes (37) respectively (38) µ ◦ (α ⊗ µ − µ ⊗ α) ◦ (id⊗3 + σ1) = 0, µ ◦ (α ⊗ µ − µ ⊗ α) ◦ (id⊗3 + σ2) = 0. where id stands for the identity map and σ1 and σ2 stands for trilinear maps defined for any x1, x2, x3 ∈ V by σ1(x1 ⊗ x2 ⊗ x3) = x2 ⊗ x1 ⊗ x3, σ2(x1 ⊗ x2 ⊗ x3) = x1 ⊗ x3 ⊗ x2. In terms of associators, the identities (35) (resp. (36)) are equivalent respec- tively to (39) asα + asα ◦ σ1 = 0 and asα + asα ◦ σ2 = 0. Hence, for any x, y, z ∈ V , we have (40) asα(x, y, z) = −asα(y, x, z) and asα(x, y, z) = −asα(x, z, y). We have also the following property. Lemma 6.5. Let (V, µ, α) be an Hom-alternative algebra. Then (41) asα(x, y, z) = −asα(z, y, x). Proof. Using (39), we have asα(x, y, z) + asα(z, y, x) = −asα(y, x, z) − asα(y, z, x) = 0. Remark 6.6. The identities (39),(40) lead to the fact that an algebra is Hom- alternative if and only if the Hom-associator asα(x, y, z) is an alternating function of its arguments, that is asα(x, y, z) = −asα(y, x, z) = −asα(x, z, y) = −asα(z, y, x). (cid:3) 22 Proposition 6.7. A Hom-alternative algebra is Hom-flexible. Proof. Using lemma 6.5, we have asα(x, y, x) = −asα(x, y, x). Therefore, asα(x, y, x) = 0. (cid:3) Proposition 6.8. Let (V, µ, α) be a Hom-alternative algebra and x, y, z ∈ V . If x and y anticommute, that is µ(x, y) = −µ(y, x), then we have (42) and (43) µ(α(x), µ(y, z)) = −µ(α(y), µ(x, z), µ(µ(z, x), α(y)) = −µ(µ(z, y), α(x)). Proof. The left alternativity leads to (44) µ(α(x), µ(y, z)) − µ(µ(x, y), α(z)) + µ(α(y), µ(x, z)) − µ(µ(y, x), α(z)) = 0. Since µ(x, y) = −µ(y, x), then the previous identity becomes (45) µ(α(x), µ(y, z)) + µ(α(y), µ(x, z)) = 0. Similarly, using the right alternativity and the assumption of anticommutativity, we get the second identity. (cid:3) The following theorem provides a way to construct Hom-alternative algebras starting from an alternative algebra and an algebra endomorphism. This procedure was applied to associative algebras, G-associative algebras and Lie algebra in [45]. It was extended to coalgebras in [35] and to n-ary algebras of Lie type respectively associative type in [5]. Theorem 6.9 ([31]). Let (V, µ) be a left alternative algebra (resp. a right alter- native algebra ) and α : V → V be an algebra endomorphism. Then (V, µα, α), where µα = α ◦ µ, is a left Hom-alternative algebra (resp. right Hom-alternative algebra). Moreover, suppose that (V ′, µ′) is another left alternative algebra (resp. a right alternative algebra ) and α′ : V ′ → V ′ is an algebra endomorphism. If f : V → V ′ is an algebras morphism that satisfies f ◦ α = α′ ◦ f then f : (V, µα, α) −→ (V ′, µ′ α′ , α′) is a morphism of left Hom-alternative algebras (resp. right Hom-alternative alge- bras). Remark 6.10. Let (V, µ, α) be a Hom-alternative algebra, one may ask whether was addressed and discussed for Hom-associative algebras in [18, 19]. this Hom-alternative algebra is induced by an ordinary alternative algebra (V,eµ), that is α is an algebra endomorphism with respect to eµ and µ = α◦eµ. This question First observation, if α is an algebra endomorphism with respect to eµ then α is also an algebra endomorphism with respect to µ. Indeed, µ(α(x), α(y)) = α ◦ eµ(α(x), α(y)) = α ◦ α ◦ eµ(x, y) = α ◦ µ(x, y). Second observation, if α is bijective then α−1 is also an algebra automorphism. Therefore one may use an untwist operation on the Hom-alternative algebra in order to recover the alternative algebra (eµ = α−1 ◦ µ). 23 6.1 Examples of Hom-Alternative algebras We construct examples of Hom-alternative using theorem (6.9). We use to this end the classification of 4-dimensional alternative algebras which are not associa- tive (see [20]) and the algebra of octonions (see [14]). For each algebra, algebra endomorphisms are provided. Therefore, Hom-alternative algebras are attached according to theorem (6.9). Example 6.11 (Hom-alternative algebras of dimension 4). According to [20], p 144, there are exactly two alternative but not associative algebras of dimension 4 over any field. With respect to a basis {e0, e1, e2, e3}, one algebra is given by the following multiplication (the unspecified products are zeros) µ1(e0, e0) = e0, µ1(e0, e1) = e1, µ1(e2, e0) = e2, µ1(e2, e3) = e1, µ1(e3, e0) = e3, µ1(e3, e2) = −e1. The other algebra is given by µ2(e0, e0) = e0, µ2(e0, e2) = e2, µ2(e0, e3) = e3, µ2(e1, e0) = e1, µ2(e2, e3) = e1, µ2(e3, e2) = −e1. These two alternative algebras are anti-isomorphic, that is the first one is isomor- phic to the opposite of the second one. The algebra endomorphisms of µ1 and µ2 are exactly the same. We provide two examples of algebra endomorphisms for these algebras. (1) The algebra endomorphism α1 with respect to the same basis is defined by α1(e0) = e0 + a1 e1 + a2 e2 + a3 e3, α1(e1) = 0, a5a3 a2 e3, α1(e3) = a5 e2 + α1(e2) = a4 e2 + a4a3 a2 e3, with a1, · · · , a5 ∈ K and a2 6= 0. (2) The algebra endomorphism α2 with respect to the same basis is defined by α2(e0) = e0 + a1 e1 + a2 e2 + a3 e3, α2(e1) = a4 e1, α2(e2) = − a4a2 a5 e2 − a4a3 a5 e3, α2(e3) = a5 e1 + a6 e2 + a6a3 − a5 a2 e3, with a1, · · · , a6 ∈ K and a2, a5 6= 0. According to theorem (6.9), the linear map α1 an the following multiplications • µ1 1(e0, e0) = e0 + a1 e1 + a2 e2 + a3 e3, µ1 µ1 1(e2, e0) = a4 e2 + 1(e2, e3) = 0, e3, µ1 1(e0, e1) = 0, µ1 1(e3, e0) = a5 e2 + e3, µ1(e3, e2) = 0. a4a3 a2 a5a3 a2 24 • µ1 2(e0, e0) = e0 + a1 e1 + a2 e2 + a3 e3, µ1 µ1 2(e0, e3) = a5 e2 + a5a3 a2 e3, µ1 2(e1, e0) = 0, µ1 2(e0, e2) = a4 e2 + a4a3 a2 2(e2, e3) = 0, µ1 2(e3, e2) = 0. e3, determine 4-dimensional Hom-alternative algebras. The linear map α2 leads to the following multiplications • µ2 1(e0, e0) = e0 + a1 e1 + a2 e2 + a3 e3, µ2 µ2 1(e2, e0) = − e2 − e3, a4a2 a5 a4a3 a5 1(e0, e1) = a4 e1, µ2 1(e2, e3) = a4 e1, µ2 1(e3, e0) = e3, µ2 1(e3, e2) = −a4 e1. • µ2 2(e0, e0) = e0 + a1 e1 + a2 e2 + a3 e3, µ2 2(e0, e2) = − a4a2 a5 e2 − a4a3 a5 e3, µ2 2(e0, e3) = a5 e1 + a6 e2 + a6a3 − a5 a2 e3, µ2 2(e1, e0) = a4 e1, µ2(e2, e3) = a4 e1, µ2 2(e3, e2) = −a4 e1. Example 6.12 (Octonions). Octonions are typical example of alternative alge- bra. They were discovered in 1843 by John T. Graves who called them Octaves and independently by Arthur Cayley in 1845. See [14] for the role of the oc- tonions in algebra, geometry and topology and see also [3] where octonions are viewed as a quasialgebra. The octonions algebra which is also called Cayley Octaves or Cayley algebra is an 8-dimensional defined with respect to a basis {u, e1, e2, e3, e4, e5, e6, e7}, where u is the identity for the multiplication, by the following multiplication table. The table describes multiplying the ith row ele- ments by the jth column elements. e6 e6 e4 e3 e2 e1 u e3 e4 e2 u e1 e7 −e2 e1 −u e4 e2 −e4 −u e5 e3 −e7 −e5 −u e4 e5 −e6 e6 e7 e7 e5 e5 e7 e6 −e5 −e3 e7 −e6 e1 e3 −e5 e4 e1 e4 −e3 −e1 −u e2 e5 −e4 −e2 −u e5 −e7 e3 e1 −e3 e6 e2 −e1 −e6 −u e2 −e4 e7 e3 −e2 −e7 −u e6 −e1 u e1 e2 e3 e4 e5 e6 e7 The diagonal algebra endomorphisms of octonions are give by maps α defined with respect to the basis {u, e1, e2, e3, e4, e5, e6, e7} by α(u) = u, α(e1) = a e1, α(e2) = b e2, α(e3) = c e3, α(e4) = ab e4, α(e5) = bc e5, α(e6) = abc e6, α(e7) = ac e7, 25 where a, b, c are any parameter in K. The associated Hom-alternative algebra to the octonions algebra according to theorem (6.9) is described by the map α and the multiplication defined by the following table. The table describes multiplying the ith row elements by the jth column elements. e2 be2 abe4 −u e3 ce3 ace7 bce5 −u e1 ae1 −u u u ae1 be2 ce3 abe4 bce5 −abce6 u e1 e2 e3 e4 e5 e6 abce6 ace7 e7 be2 −abe4 −ace7 −bce5 −ae1 −abce6 ce3 −be2 −ace7 abe4 −ae1 bce5 ce3 −ace7 abce6 e6 e7 ace7 −ce3 ace7 −abce6 −abe4 ae1 e4 abe4 −be2 ae1 abce6 −u e5 bce5 abce6 abce6 −bce5 −ce3 be2 ce3 ace7 −u ae1 −u −ce3 −ae1 bce5 −abe4 −be2 −bce5 abe4 be2 −u Notice that the new algebra is no longer unital, neither an alternative algebra since µ(u, µ(u, e1)) − µ(µ(u, u), e1) = (a2 − a)e1, which is different from 0 when a 6= 0, 1. 7 Hom-Jordan algebras In this section, we consider, following [31], a generalization of Jordan algebra by twisting the usual Jordan identity (x · y) · x2 = x · (y · x2). We show that this generalization fits with Hom-associative algebras. Also, we provide a procedure to construct examples starting from an ordinary Jordan algebras. Definition 7.1 ([31]). A Hom-Jordan algebra is a triple (V, µ, α) consisting of a linear space V , a bilinear map µ : V × V → V which is commutative and a homomorphism α : V → V satisfying (46) µ(α2(x), µ(y, µ(x, x))) = µ(µ(α(x), y), α(µ(x, x))) where α2 = α ◦ α. Remark 7.2. Since the multiplication is commutative, one may write the identity (46) as (47) µ(µ(y, µ(x, x)), α2(x)) = µ(µ(y, α(x)), α(µ(x, x))). When the twisting map α is the identity map, we recover the classical notion of Jordan algebra. The identity (46) is motivated by the following functor which associates to a Hom-associative algebra a Hom-Jordan algebra by considering the plus Hom- algebra. 26 ❧ Theorem 7.3 ([31]). Let (V, m, α) be a Hom-associative algebra. Then the Hom- algebra (V, µ, α), where the multiplication µ is defined for all x, y ∈ V by µ(x, y) = 1 2 (m(x, y) + m(y, x)). is a Hom-Jordan algebra. Proof. The commutativity of µ is obvious. We compute the difference D = µ(α2(x), µ(y, µ(x, x))) − µ(µ(α(x), y), α(µ(x, x))) A straightforward computation gives D = m(α2(x), m(y, m(x, x))) + m(m(y, m(x, x)), α2(x)) +m(α2(x), m(m(x, x), y)) + m(m(m(x, x), y), α2(x)) −m(m(α(x), y), α(m(x, x))) − m(α(m(x, x)), m(α(x), y)) −m(m(y, α(x)), α(m(x, x))) − m(α(m(x, x)), m(y, α(x))). We have by Hom-associativity m(α2(x), m(y, m(x, x))) − m(m(α(x), y), α(m(x, x))) = 0 m(m(m(x, x), y), α2(x)) − m(α(m(x, x)), m(y, α(x))) = 0. Therefore D = m(m(y, m(x, x)), α2(x)) + m(α2(x), m(m(x, x), y)) −m(α(m(x, x)), m(α(x), y)) − m(m(y, α(x)), α(m(x, x))). One may show that for any Hom-associative algebra we have m(α(m(x, x)), m(α(x), y)) = m(m(m(x, x), α(x)), α(y)) = m(m(α(x), m(x, x)), α(y)) = m(α2(x), m(m(x, x), y)), and similarly Thus m(m(y, α(x)), α(m(x, x))) = m(m(y, m(x, x)), α2(x)). D = m(m(y, m(x, x)), α2(x)) + m(α2(x), m(m(x, x), y)) −m(α2(x), m(m(x, x), y)) − m(m(y, m(x, x)), α2(x)) = 0. Remark 7.4. The definition of Hom-Jordan algebra seems to be non natural one expects that the identity should be of the form (48) or (49) µ(α(x), µ(y, µ(x, x)))) = µ(µ(x, y), α(µ(x, x))) µ(α(x), µ(y, µ(x, x)))) = µ(µ(x, y), µ(x, α(x))). It turns out that these identities do not fit with the previous proposition. (cid:3) 27 Notice also that in general a Hom-alternative algebra doesn't lead to a Hom- Jordan algebra. The following theorem gives a procedure to construct Hom-Jordan algebras using ordinary Jordan algebras and their algebra endomorphisms. Theorem 7.5 ([31]). Let (V, µ) be a Jordan algebra and α : V → V be an algebra endomorphism. Then (V, µα, α), where µα = α ◦ µ, is a Hom-Jordan algebra. Moreover, suppose that (V ′, µ′) is another Jordan algebra and α′ : V ′ → V ′ is an algebra endomorphism. If f : V → V ′ is an algebras morphism that satisfies f ◦ α = α′ ◦ f then is a morphism of Hom-Jordan algebras. f : (V, µα, α) −→ (V ′, µ′ α′ , α′) Remark 7.6. We may give here similar observations as in the remark (6.10) concerning Hom-Jordan algebra induced by an ordinary Jordan algebra. We provide in the sequel an example of Hom-Jordan algebras. Example 7.7. We consider Hom-Jordan algebras associated to Hom-associative algebras described in example (3.2). Let {e1, e2, e3} be a basis of a 3-dimensional linear space V over K. The following multiplication µ and linear map α on V define Hom-Jordan algebras over K3: eµ(e1, e1) = a e1, eµ(e1, e2) = eµ(e2, e1) = a e2, eµ(e1, e3) = eµ(e3, x1) = b e3, eµ(e2, e2) = a e2, eµ(e2, e3) = 1 eµ(e3, e2) = eµ(e3, e3) = 0, 2 b e3, α(e1) = a e1, α(e2) = a e2, α(e3) = b e3 where a, b are parameters in K. It turns out that the multiplication of this Hom-Jordan algebra defines a Jordan algebra. Remark 7.8. We may define the noncommutative Jordan algebras as triples (V, µ, α) satisfying the identity (46) and the flexibility condition, which is a gen- eralization of the commutativity. Eventually, we may consider the Hom-flexibilty defined by the identity µ(α(x), µ(y, x)) = µ(µ(x, y), α(x)). Remark 7.9. A Z2-garded version of Hom-alternative algebras and Hom-Jordan algebras might be defined in a natural way. Acknowledgments Dedicated to Amine Kaidi on his 60th birthday. 28 ❧ References [1] ALBERT, A.A.,(1948). Power associative rings, Trans. Amer. math. Soc. 64, 552-597. [2] AIZAWA, N., SATO, H.,(1991). q-deformation of the Virasoro algebra with central exten- sion, Physics Letters B, Phys. Lett. B 256, no. 1, 185 -- 190. [3] ALBUQUERQUE, H., MAJID, S.,(1999). Quasialgebra Structure of the Octonions, J. Al- gebra 220 , 188-224. [4] AMMAR, F., MAKHLOUF, A.,(2009). Hom-Lie Superalgebras and Hom-Lie admissible Superalgebras, arXiv:0906.1668v2 [math.RA]. [5] ATAGUEMA, H., MAKHLOUF, A., SILVESTROV, S.,(2009). Generalization of n-ary Nambu algebras and beyond, J. Math. Phys., 50, 1. [6] BREMNER, M.R., MURAKAMI, L.I., SHESTAKOV, I.P, Nonassociative algebras, Preprint. [7] CHAICHIAN, M. , ELLINAS, D., POPOWICZ, Z., (1990). Quantum conformal algebra with central extension, Phys. Lett. B 248, no. 1-2, 95 -- 99. [8] CHAICHIAN, M., ISAEV, A. P. , LUKIERSKI, J., POPOWICZ, Z., PRESNAJDER, P., (1991). q-deformations of Virasoro algebra and conformal dimensions, Phys. Lett. B 262 (1), 32 -- 38. [9] CHAICHIAN, M., KULISH, P., LUKIERSKI, J., (1990). q-deformed Jacobi identity, q- oscillators and q-deformed infinite-dimensional algebras, Phys. Lett. B 237 , no. 3-4, 401 -- 406. [10] CHAICHIAN, M., POPOWICZ, Z. , PRESNAJDER, P., (1990). q-Virasoro algebra and its relation to the q-deformed KdV system, Phys. Lett. B 249, no. 1, 63 -- 65. [11] CURTRIGHT, T. L., ZACHOS, C. K., (1990). Deforming maps for quantum algebras, Phys. Lett. B 243, no. 3, 237 -- 244. [12] DAMASKINSKY, E. V., KULISH, P. P., (1992). Deformed oscillators and their applica- tions, Zap. Nauch. Semin. LOMI 189, 37 -- 74 (1991); Engl. transl. in J. Soviet Math. 62 (5), 2963 -- 2986. [13] DASKALOYANNIS, C., (1992). Generalized deformed Virasoro algebras, Modern Phys. Lett. A 7 no. 9, 809 -- 816. [14] BAEZ, J.C., (2001). T he octonions, Bull. of the Amer. Math. Soc., 39 2, 145 -- 2005. [15] CAENEPEEL, S., GOYVAERTS, I., (2009). Hom-Hopf algebras, arXiv:0907.0187v1 [math.RA]. [16] ELHAMDADI, M., MAKHLOUF, A., (2009). Cohomology and Formal Deformations of Left Alternative algebras, arXiv:0907.1548v1 [Math.RA]. [17] FREGIER, Y., GOHR, A., (2009). Lie Type Hom-algebras, arXiv:0903.3393v2 [Math.RA]. [18] FREGIER, Y., GOHR, A., (2009). On unitality conditions for Hom-associative algebras, arXiv:0904.4874v2 [Math.RA]. [19] GOHR, A., (2009). On Hom-algebras with surjective twisting, arXiv:0906.3270v3 [Math.RA]. [20] GOODAIRE, E., (2000). Alternative rings of small order and the hunt for Moufang circle loops, Nonassociative algebra and its applications (Sao Paulo, 1998), 137 -- 146, Lecture Notes in Pure and Appl. Math., 211, Marcel Dekker, New York. [21] GOZE, M., REMM, E., (2004). Lie-admissible algebras and operads, J. Algebra 273, 129 -- 152. [22] HARTWIG J. T. LARSSON, D., SILVESTROV, S. D., (2006). Deformations of Lie algebras using σ-derivations, J. of Algebra 295, 314-361. [23] HELLSTR OM, L., SILVESTROV S. D., (2000). Commuting elements in q-deformed Heisen- berg algebras, World Scientific. [24] HU, N., (1999). q-Witt algebras, q-Lie algebras, q-holomorph structure and representations, Algebra Colloq. 6 , no. 1, 51 -- 70. [25] KASSEL, C., (1992). Cyclic homology of differential operators, the Virasoro algebra and a q-analogue, Commun. Math. Phys. 146 , 343-351. [26] KERDMAN, F. S., (1979). Analytic Moufang loops in the large, Algebra i Logica 18, 523 -- 555. [27] LARSSON, D., SILVESTROV, S. D., (2005). Quasi- Hom-Lie algebras, Central Extensions and 2-cocycle-like identities, J. of Algebra 288, 321 -- 344. 29 [28] LARSSON, D., SILVESTROV, S. D., (2005). Quasi-Lie algebras, in "Noncommutative Ge- ometry and Representation Theory in Mathematical Physics", Contemp. Math., 391, Amer. Math. Soc., Providence, RI , 241 -- 248. [29] LARSSON, D., SILVESTROV, S. D., (2007). Quasi-deformations of sl2(F) using twisted derivations, Comm. in Algebra 35, 4303 -- 4318 . [30] LIU, K. Q., (1992). Characterizations of the quantum Witt algebra, Lett. Math. Phys. 24 , no. 4, 257 -- 265. [31] MAKHLOUF, A., (2009). Hom-Alternative algebras and Hom-Jordan algebras, arXiv:0909.0326 [math.RA]. [32] MAKHLOUF, A., SILVESTROV, S. D., (2008). Hom-algebra structures, J. Gen. Lie Theory Appl. 2 (2) , 51 -- 64. [33] MAKHLOUF, A., SILVESTROV, S. D., (2008). Hom-Lie admissible Hom-coalgebras and Hom-Hopf algebras, Published as Chapter 17, pp 189-206, S. Silvestrov, E. Paal, V. Abramov, A. Stolin, (Eds.), Generalized Lie theory in Mathematics, Physics and Beyond, Springer-Verlag, Berlin, Heidelberg. [34] MAKHLOUF, A., SILVESTROV, S. D., (2007). Notes on Formal deformations of Hom- Associative and Hom-Lie algebras, Forum Mathematicum, to appear. arXiv:0712.3130v1 [math.RA]. [35] MAKHLOUF, A., SILVESTROV, S. D., (2008). Hom-Algebras and Hom-Colgebras, J. Al- gebra and its App., to appear. (arXiv:0811.0400 [math.RA]). [36] MALTSEV, A. I., (1955). Analytical loops, Matem. Sbornik. 36, 569-576 (in Russian). [37] McCRIMMON, Alternative available algebras, K., in http://www.mathstat.uottawa.ca/∼neher/Papers/alternative/ [38] MYUNG, H. C., (1982) Lie algebras and Flexible Lie-admissible algebras, Hadronic Press INC, Hadronic Press Monographs in Mathematics, 1, Massachusetts. [39] PAAL, E., (2004). Note on analytic Moufang loops, Comment. Math. Univ. Carolinae, 45 (2), 349 -- 354. [40] PAAL, E., (2008). Moufang loops and generalized Lie-Cartan theorem, Journal of Gen. Lie Theory and Applications, 2, 45 -- 49. [41] SANDU, N. I., (2008). About the embedding of Moufang loops in alternative algebras II, arXiv:0804.2049v1 [math.RA]. [42] SHESTAKOV, I.P., (2003). Moufang Loops and alternative algebras, Proc. of Amer. Math. Soc., 132 (2), 313 -- 316. [43] YAMAGUTI, K., (1963). On the theory of Malcev algebras Kumamoto J. Sci. Ser. A 6 1963 9 -- 45. [44] YAU, D., (2008). Enveloping algebra of Hom-Lie algebras, J. Gen. Lie Theory Appl. 2 (2), 95 -- 108. [45] YAU, D., (2007). Hom-algebras and homology, arXiv:0712.3515v2. [46] YAU, D., (2008). Hom-bialgebras and comodule algebras, arXiv:0810.4866. [47] YAU, D., (2009). Hom-Yang-Baxter equation, Hom-Lie algebras, and quasi-triangular bial- gebras, J. Phys. A 42, 165 -- 202. [48] YAU, D., (2009)., The Hom-Yang-Baxter equation and Hom-Lie algebras, arXiv:0905.1887v2. [49] YAU, D., (2009). The classical Hom-Yang-Baxter equation and Hom-Lie bialgebras, arXiv:0905.1890v1. [50] YAU, D., (2009). Hom-quantum groups I:quasi-triangular Hom-bialgebras, arXiv:0906.4128v1. [51] YAU, D., (2009). Hom-quantum groups II:cobraided Hom-bialgebras and Hom-quantum ge- ometry, arXiv:0907.1880v1. [52] YAU, D., (2009). Hom-Novikov algebras, arXiv:0909.0726v1. Laboratoire de Math´ematiques, Informatique et Applications, Universit´e de Haute- Alsace, France E-mail address: [email protected] 30
1911.05138	1	1911	2019-11-12T20:55:38	Varieties corresponding to classes of complemented posets	[ "math.RA" ]	As algebraic semantics of the logic of quantum mechanics there are usually used orthomodular posets, i.e. bounded posets with a complementation which is an antitone involution and where the join of orthogonal elements exists and the orthomodular law is satisfied. When we omit the condition that the complementation is an antitone involution, then we obtain skew-orthomodular posets. To each such poset we can assign a bounded lambda-lattice in a non-unique way. Bounded lambda-lattices are lattice-like algebras whose operations are not necessarily associative. We prove that any of the following properties for bounded posets with a unary operation can be characterized by certain identities of an arbitrary assigned lambda-lattice: complementarity, orthogonality, almost skew-orthomodularity and skew-orthomodularity. It is shown that these identities are independent. Finally, we show that the variety of skew-orthomodular lambda-lattices is congruence permutable as well as congruence regular.	math.RA	math	Varieties corresponding to classes of complemented posets Ivan Chajda, Miroslav Kolar´ık and Helmut Langer Abstract As algebraic semantics of the logic of quantum mechanics there are usually used orthomodular posets, i.e. bounded posets with a complementation which is an antitone involution and where the join of orthogonal elements exists and the ortho- modular law is satisfied. When we omit the condition that the complementation is an antitone involution, then we obtain skew-orthomodular posets. To each such poset we can assign a bounded λ-lattice in a non-unique way. Bounded λ-lattices are lattice-like algebras whose operations are not necessarily associative. We prove that any of the following properties for bounded posets with a unary operation can be characterized by certain identities of an arbitrary assigned λ-lattice: comple- mentarity, orthogonality, almost skew-orthomodularity and skew-orthomodularity. It is shown that these identities are independent. Finally, we show that the variety of skew-orthomodular λ-lattices is congruence permutable as well as congruence regular. AMS Subject Classification: 06A11, 06B75, 06C15, 03G12, 08B10 Keywords: Bounded poset, complemented poset, orthogonal poset, skew-orthomodular poset, λ-lattice, variety, congruence distributive, congruence regular It is well-known that an algebraic semantics of the logic of quantum mechanics is provided by means of orthomodular lattices as shown by G. Birkhoff and J. von Neumann ([2]) or, independently, by K. Husimi ([10]). The details of this construction can be found e.g. in the monograph by L. Beran ([1]). However, it was shown later that in the logic of quantum mechanics the connective disjunction represented by the lattice operation ∨ need not exist for elements that are not orthogonal. Hence the concept of an orthomodular poset was introduced as follows: A bounded poset P = (P, ≤, ′, 0, 1) with a unary operation is called orthomodular (see e.g. [4]) if ′ is an antitone involution on (P, ≤) which is a complementation, i.e. x ≤ y implies y ′ ≤ x′, x′′ = x, sup(x, x′) exists for all x ∈ P and it is equal to 1, and inf(x, x′) exists for all x ∈ P and it is equal to 0; moreover, sup(x, y) must exist in case x ≤ y ′; finally, for all x, y ∈ P with x ≤ y there exists x ∨ (x′ ∧ y) and it is equal to y. Due to De Morgan's laws, also dually, y ∧ (y ′ ∨ x) exists for all x, y ∈ P with x ≤ y and it is equal to x. The last property is called the orthomodular law and can be expressed in the case of lattices alternatively in the form of the equivalent identities x ∨ (x′ ∧ (x ∨ y)) ≈ x ∨ y, y ∧ (y ′ ∨ (x ∧ y)) ≈ x ∧ y. 1Support of the research of the first and third author by OAD, project CZ 02/2019, and support of the research of the first author by IGA, project PrF 2019 015, is gratefully acknowledged. 1 It was shown by V. Sn´asel ([11]) that every bounded poset can be organized into a so-called bounded λ-lattice. Bounded λ-lattices can be considered as bounded lattices whose binary operations are not necessarily associative. More precisely, a bounded λ- lattice is a bounded lattice if and only if its binary operations are associative, see [7] for details. The notion of a λ-lattice was successfully used by the first two authors for constructing a variety of λ-lattices which corresponds to the class of orthomodular posets. This variety turns out to be congruence permutable and congruence regular. Of course, it is of advantage to work with varieties of algebras instead of classes of posets since for varieties the well-known methods of Universal Algebra can be applied. Back to the logic of quantum mechanics, we take as an appropriate structure for the algebraic semantics complemented posets in which the join of two orthogonal elements exists and which satisfy the orthomodular law. We do not ask this complementation to be an antitone involution. The concept of complemented lattices which satisfy the orthomodular law, but whose complementation need not be an antitone involution was introduced in [3] and studied in [9]. This concept was generalized by the first and third author to posets in [8]. In the present paper we show that similarly to the case of orthomodular posets, for the posets described above the method of considering assigned λ-lattices can be successfully applied. In fact, it turns out that important properties of certain posets can be characterized by identities of assigned λ-lattices. Let P = (P, ≤, ′, 0, 1) be a bounded poset with a unary operation and a, b ∈ P . We define L(a, b) := {x ∈ P x ≤ a, b}, U(a, b) := {x ∈ P a, b ≤ x}. If there exists sup(a, b) or inf(a, b) then we will denote these elements by a ∨ b or a ∧ b, respectively. We call P complemented if it satisfies the identities x ∨ x′ ≈ 1 and x ∧ x′ ≈ 0. In this case the operation ′ is called a complementation. We say that a, b are orthogonal elements of P , shortly a ⊥ b, if a ≤ b′. We call a complemented poset P orthogonal if x ∨ y exists for arbitrary orthogonal elements x, y of P . We call an orthogonal poset P almost skew-orthomodular if x ∨ (x′ ∧ y) exists for all x, y ∈ P with x ≤ y. We call an almost skew-orthomodular poset P skew-orthomodular if x ∨ (x′ ∧ y) = y for all x, y ∈ P with x ≤ y. Example 1. The poset shown in Fig. 1 i ✉ ✉ ❅ g j ✉ ❅ ❅ ✟✟✟✟✟✟✟✟ ❍❍❍❍❍❍❍❍ ❛❛❛❛❛❛❛❛❛❛ ❍❍❍❍❍❍❍❍ ❍❍❍❍❍❍❍❍ ✦✦✦✦✦✦✦✦✦✦ d ✉ ✑ ◗ ✉ ✑ ✉ ❅ ✑ ✑ ◗ e ❅ ✉ f a b ✑ ✑ ✉ ❅ ◗ ◗ ◗ ◗ ✑ ✑ 1 ✑ ◗ ✉ ✁ ❆ ✑ ◗ ✁ ❆ ✁ ❆ ✁ ❆ h ✉ ❅ ❅ c ✉ ❆ ✁ ❆ ◗ ✁ ❆ ✁ ◗ ✑ ❆ ◗ ✁ ✑ ✉ 0 ◗ Fig. 1 2 with x 0 a b x′ 1 j j c d e f j j 1 i h f f e c 0 g h i i belongs to Ps, but is not a lattice. Moreover, ′ is antitone, but not an involution. In the following let Pc, Po, Pa and Ps denote the class of all complemented, orthogonal, almost skew-orthomodular and skew-orthomodular posets, respectively. We are going to show that these classes do not coincide, i.e. the inclusions are proper. Theorem 2. We have Ps $ Pa $ Po $ Pc. Proof. The poset shown in Fig. 2 ❅ ❅ d a ✉ ❅ ✉ ❅ ❅ ❅ ❅ ✉ ✉ f c 1 ✉ ❅ ✉ ❅ ✉ e b ❅ ✉ 0 ❅ ❅ with Fig. 2 x 0 a b x′ 1 c c d e f 1 c d f f a 0 is not a lattice and belongs to Pa \ Ps since a ≤ d, but a ∨ (a′ ∧ d) = a ∨ (c ∧ d) = a ∨ 0 = a 6= d. Moreover, ′ is neither antitone nor an involution. The poset shown in Fig. 3 ✁ ✁ ✁ e ✉ ❅ 1 ✉ ❅ ❆ ✁ ❅ ❆ ❆ ❆ ❅ ❅ ❅ ✉ ❆ ✁ ❆ ✁ ❆ ✁ ◗ ◗ ❆ ✁ ✉ 0 a ◗ ✉ ◗ b ◗ ◗ ❅ ❅ f ✉ c ✉ ❅ ❅ ✉ d with Fig. 3 x 0 a b c d e f 1 x′ 1 f d d a d d 0 3 belongs to Po \ Pa since a ≤ e, but a′ ∧ e = f ∧ e does not exist. Moreover, ′ is neither antitone nor an involution. The poset shown in Fig. 4 ✑ ✑ ✑ e ❅ d ✑ ✉ ❅ ❅ ❅ f ✉ b ✉ ❅ ❅ ✉ c 1 ✑ ✉ ❅ ❆ ✁ ✑ ❅ ❆ ✁ ❆ ✁ ❆ ✁ ✉ ❅ ❅ ❅ ❅ ✁ ✁ ✁ ✁ ❆ ✉ 0 ❅ ❅ a ✉ ❆ ❆ ❆ with Fig. 4 x 0 a b x′ 1 c d a c c d e f 1 c 0 c belongs to Pc \ Po since a ≤ d = b′, but a ∨ b does not exist. Moreover, ′ is neither antitone nor an involution. Now we introduce the concept of a bounded λ-lattice taken from [11]. A bounded λ-lattice is an algebra (L, ⊔, ⊓, 0, 1) of type (2, 2, 0, 0) satisfying the identities x ⊔ y ≈ y ⊔ x, x ⊓ y ≈ y ⊓ x, x ⊔ ((x ⊔ y) ⊔ z) ≈ (x ⊔ y) ⊔ z, x ⊓ ((x ⊓ y) ⊓ z) ≈ (x ⊓ y) ⊓ z, x ⊔ (x ⊓ y) ≈ x, x ⊓ (x ⊔ y) ≈ x, x ⊔ 0 ≈ x, x ⊔ 1 ≈ 1. Hence the class of bounded λ-lattices forms a variety. Notice that every bounded λ-lattice satisfies the identities x ⊓ 0 ≈ 0 and x ⊓ 1 ≈ x. Recall from [7] that every variety of bounded λ-lattices is congruence distributive. It is well-known that in every bounded λ-lattice x ⊔ y = y is equivalent to x ⊓ y = x. Let P = (P, ≤, ′, 0, 1) be a bounded poset with a unary operation. We introduce binary operations ⊔ and ⊓ on P as follows (x, y ∈ P ): If x∨y exists then x⊔y := x∨y. Otherwise x ⊔ y = y ⊔ x is an arbitrary element of U(x, y). If x ∧ y exists then x ⊓ y := x ∧ y. Otherwise x ⊓ y = y ⊓ x is an arbitrary element of L(x, y). Then (P, ⊔, ⊓, ′, 0, 1) is a bounded λ-lattice with a unary operation which we call λ-lattice assigned to the bounded poset P. Let A(P) denote the set of all λ-lattices assigned to P. To every bounded λ-lattice L = (L, ⊔, ⊓, ′, 0, 1) with a unary operation we assign a bounded poset P(L) = (L, ≤, ′, 0, 1) as follows: (x, y ∈ L). It was shown in [11] that (L, ≤, 0, 1) is a bounded poset and x ≤ y if and only if x ⊔ y = y x ≤ y if and only if x ⊓ y = x. 4 Moreover, using the absorption laws, we easily derive the identities x ⊔ x ≈ x and x ⊓ x ≈ x. For i ∈ {c, o, a, s} let Li denote the class of all bounded λ-lattices L with a unary operation satisfying P(L) ∈ Pi. Hence Li can be considered as a representation of Pi. This means that the properties of Li may be considered as properties of Pi. In the following we will characterize the above mentioned properties of bounded posets with a unary operation by means of identities of assigned λ-lattices. Surprisingly, this works despite the fact that this assignment is not unique. Hence, classes of comple- mented, orthogonal, almost skew-orthomodular and skew-orthomodular posets will be characterized by means of varieties of bounded λ-lattices. We start with complemented posets. Theorem 3. Let P = (P, ≤, ′, 0, 1) be a bounded poset with a unary operation and L = (P, ⊔, ⊓, ′, 0, 1) ∈ A(P). Then P is complemented if and only if L satisfies the identities (x ⊔ y) ⊔ (x′ ⊔ y) ≈ 1, (x ⊓ y) ⊓ (x′ ⊓ y) ≈ 0. (1) (2) Hence Lc is a variety. Proof. Let a, b ∈ P . First assume P ∈ Pc. Then a ≤ a ⊔ b ≤ (a ⊔ b) ⊔ (a′ ⊔ b), a′ ≤ a′ ⊔ b ≤ (a ⊔ b) ⊔ (a′ ⊔ b) and hence (a ⊔ b) ⊔ (a′ ⊔ b) ∈ U(a, a′) = {1}, i.e., (a ⊔ b) ⊔ (a′ ⊔ b) = 1. Dually, (a ⊓ b) ⊓ (a′ ⊓ b) ≤ a ⊓ b ≤ a, (a ⊓ b) ⊓ (a′ ⊓ b) ≤ a′ ⊓ b ≤ a′ and hence (a ⊓ b) ⊓ (a′ ⊓ b) ∈ L(a, a′) = {0}, i.e., (a ⊓ b) ⊓ (a′ ⊓ b) = 0. Conversely, suppose L to satisfy identities (1) and (2). If a, a′ ≤ b then a ⊔ b = a′ ⊔ b = b and hence b = b ⊔ b = (a ⊔ b) ⊔ (a′ ⊔ b) = 1 showing a ∨ a′ = 1. Similarly, b ≤ a, a′ implies a ⊓ b = a′ ⊓ b = b and therefore b = b ⊓ b = (a ⊓ b) ⊓ (a′ ⊓ b) = 0 showing a ∧ a′ = 0. Hence P ∈ Pc. The identities x ⊔ x′ ≈ 1, x ⊓ x′ ≈ 0. are necessary, but not sufficient for a bounded λ-lattice to be complemented. Orthogonal posets can be characterized by an identity that is a bit more complicated than the previous ones (1) and (2). 5 Theorem 4. Let P = (P, ≤, ′, 0, 1) ∈ Pc and L = (P, ⊔, ⊓, ′, 0, 1) ∈ A(P). Then P is orthogonal if and only if L satisfies the identity (((x ⊓ y ′) ⊔ z) ⊔ (y ⊔ z)) ⊓ ((x ⊓ y ′) ⊔ y) ≈ (x ⊓ y ′) ⊔ y. (3) Hence Lo is a variety. Proof. Let a, b, c ∈ P . First assume P ∈ Po. Then (a ⊓ b′) ∨ b exists. Now a ⊓ b′ ≤ (a ⊓ b′) ⊔ c ≤ ((a ⊓ b′) ⊔ c) ⊔ (b ⊔ c), b ≤ b ⊔ c ≤ ((a ⊓ b′) ⊔ c) ⊔ (b ⊔ c) and hence ((a ⊓ b′) ⊔ c) ⊔ (b ⊔ c) ∈ U(a ⊓ b′, b) which yields (a ⊓ b′) ⊔ b ≤ ((a ⊓ b′) ⊔ c) ⊔ (b ⊔ c) which is equivalent to identity (3). Conversely, suppose L to satisfy identity (3). Assume a ⊥ b. Then a ⊓ b′ = a. If a, b ≤ c then a ⊔ c = b ⊔ c = c and hence a ⊔ b = (a ⊓ b′) ⊔ b = (((a ⊓ b′) ⊔ c) ⊔ (b ⊔ c)) ⊓ ((a ⊓ b′) ⊔ b) ≤ ((a ⊓ b′) ⊔ c) ⊔ (b ⊔ c) = = (a ⊔ c) ⊔ c = c ⊔ c = c showing a ⊔ b = a ∨ b, i.e. a ∨ b exists. Hence P ∈ Po. Similarly as above we can characterize almost skew-orthomodular posets. Theorem 5. Let P = (P, ≤, ′, 0, 1) ∈ Po and L = (P, ⊔, ⊓, ′, 0, 1) ∈ A(P). Then P is almost skew-orthomodular if and only if L satisfies the identity (((x ⊓ y)′ ⊓ z) ⊓ (y ⊓ z)) ⊔ ((x ⊓ y)′ ⊓ y) ≈ (x ⊓ y)′ ⊓ y. (4) Hence La is a variety. Proof. Let a, b, c ∈ P . First assume P ∈ Pa. Then (a ⊓ b)′ ∧ b exists. Now ((a ⊓ b)′ ⊓ c) ⊓ (b ⊓ c) ≤ (a ⊓ b)′ ⊓ c ≤ (a ⊓ b)′, ((a ⊓ b)′ ⊓ c) ⊓ (b ⊓ c) ≤ b ⊓ c ≤ b and hence ((a ⊓ b)′ ⊓ c) ⊓ (b ⊓ c) ∈ L((a ⊓ b)′, b) which yields ((a ⊓ b)′ ⊓ c) ⊓ (b ⊓ c) ≤ (a ⊓ b)′ ⊓ b which is equivalent to identity (4). Conversely, suppose L to satisfy identity (4). Assume a ≤ b. Then a ⊓ b = a. If c ≤ a′, b then a′ ⊓ c = b ⊓ c = c and hence c = c ⊓ c = (a′ ⊓ c) ⊓ c = ((a ⊓ b)′ ⊓ c) ⊓ (b ⊓ c) ≤ ≤ (((a ⊓ b)′ ⊓ c) ⊓ (b ⊓ c)) ⊔ ((a ⊓ b)′ ⊓ b) = (a ⊓ b)′ ⊓ b = a′ ⊓ b showing a′ ⊓ b = a′ ∧ b, i.e. a′ ∧ b exists. Moreover, a ⊔ b = b. a ⊔ c = (a′ ⊓ b) ⊔ c = c and hence If a, a′ ⊓ b ≤ c then a ⊔ (a′ ⊓ b) = (b ⊓ a′) ⊔ a = (((b ⊓ a′) ⊔ c) ⊔ (a ⊔ c)) ⊓ ((b ⊓ a′) ⊔ a) ≤ ≤ ((b ⊓ a′) ⊔ c) ⊔ (a ⊔ c) = ((a′ ⊓ b) ⊔ c) ⊔ c = c ⊔ c = c showing a ⊔ (a′ ⊓ b) = a ∨ (a′ ∧ b), i.e. a ∨ (a′ ∧ b) exists. Hence P ∈ Pa. 6 Example 6. The poset shown in Fig. 5 1 ✉ ❅ ❅ ❅ ❅ ❅ ✉ 0 e a ✉ ❅ f ✉ ❍❍❍❍❍❍❍❍ ❅ ❅ ❅ ❅ g ✉ ❅ ❅ ❍❍❍❍❍❍❍❍ ✟✟✟✟✟✟✟✟ c ✉ ❅ ✉ ❅ ✉ b ❅ ❅ ✉ d with Fig. 5 x 0 a b c d e f g 1 x′ 1 g g 1 1 g g 1 g satisfies identity (4), but does not belong to Pc since c ∧ c′ = c ∧ 1 = c 6= 0, and is not a lattice. Moreover, ′ is antitone, but not an involution. Next we characterize skew-orthomodular posets by identities of assigned λ-lattices. Since in almost skew-orthomodular λ-lattices we have x ⊔ (x′ ⊓ (x ⊔ y)) = x ∨ (x′ ∧ (x ∨ y)), we only need to add a single identity. Let us note that the poset shown in Fig. 1 is almost skew-orthomodular, but not skew-orthomodular. For skew-orthomodularity we have the following result. The proof is evident. Corollary 7. Let P = (P, ≤, ′, 0, 1) be a bounded poset with a unary operation and L = (P, ⊔, ⊓, ′, 0, 1) ∈ A(P). Then P is skew-orthomodular if and only if L satisfies the identities (1) -- (5) where x ⊔ (x′ ⊓ (x ⊔ y)) ≈ x ⊔ y. (5) Hence Ls is a variety. In the following we show some important congruence properties of the variety Ls. Let V be a variety. The variety V is called congruence permutable if Θ ◦ Φ = Φ ◦ Θ for all A ∈ V and all Θ, Φ ∈ Con A. The variety V is called congruence regular if for each A = (A, F ) ∈ V, a ∈ A and Θ, Φ ∈ Con A with [a]Θ = [a]Φ we have Θ = Φ. It is well-known (cf. [5]) that V is congruence permutable if and only if there exists a so-called Malcev term, i.e. a ternary term p satisfying p(x, x, y) ≈ p(y, x, x) ≈ y and it is regular if and only if there exists a positive integer n and ternary terms t1, . . . , tn such that t1(x, y, z) = · · · = tn(x, y, z) = z if and only if x = y. 7 Theorem 8. Let V be a variety of bounded λ-lattices (L, ⊔, ⊓, ′, 0, 1) with a unary oper- ation satisfying the identities x ⊓ x′ ≈ 0 and (5). Then V is congruence permutable. In particular, Ls is congruence permutable. Proof. The term p(x, y, z) := (x ⊔ (y ′ ⊓ (y ⊔ z))) ⊓ (z ⊔ (y ′ ⊓ (y ⊔ x))) is a Malcev term since p(x, x, z) ≈ (x ⊔ (x′ ⊓ (x ⊔ z))) ⊓ (z ⊔ (x′ ⊓ (x ⊔ x))) ≈ (x ⊔ z) ⊓ (z ⊔ (x′ ⊓ x)) ≈ ≈ (x ⊔ z) ⊓ (z ⊔ 0) ≈ (x ⊔ z) ⊓ z ≈ z, p(x, z, z) ≈ (x ⊔ (z ′ ⊓ (z ⊔ z))) ⊓ (z ⊔ (z ′ ⊓ (z ⊔ x))) ≈ (x ⊔ (z ′ ⊓ z)) ⊓ (z ⊔ x) ≈ ≈ (x ⊔ 0) ⊓ (z ⊔ x) ≈ x ⊓ (z ⊔ x) ≈ x. We are going to show also congruence regularity of the variety Ls. Theorem 9. Let V be a variety of bounded λ-lattices (L, ⊔, ⊓, ′, 0, 1) with a unary oper- ation satisfying the identities 0′ ≈ 1, x ⊓ x′ ≈ 0 and (5). Then V is congruence regular. In particular, the variety Ls is congruence regular. Proof. Put Then t(x, y) := (x′ ⊓ (x ⊔ y)) ⊔ (y ′ ⊓ (x ⊔ y)). t(x, x) ≈ (x′ ⊓ (x ⊔ x)) ⊔ (x′ ⊓ (x ⊔ x)) ≈ (x′ ⊓ x) ⊔ (x′ ⊓ x) ≈ 0 ⊔ 0 ≈ 0, and if t(x, y) = 0 then x′ ⊓ (x ⊔ y) = y ′ ⊓ (x ⊔ y) = 0 and hence x = x ⊔ 0 = x ⊔ (x′ ⊓ (x ⊔ y)) = x ⊔ y = y ⊔ x = y ⊔ (y ′ ⊓ (x ⊔ y)) = y ⊔ 0 = y. If we put then t1(x, y, z) := t(x, y) ⊔ z, t2(x, y, z) := (t(x, y))′ ⊓ z t1(x, x, z) ≈ t(x, x) ⊔ z ≈ 0 ⊔ z ≈ z, t2(x, x, z) ≈ (t(x, x))′ ⊓ z ≈ 0′ ⊓ z ≈ 1 ⊓ z ≈ z, and if t1(x, y, z) = t2(x, y, z) = z then t(x, y) ≤ z ≤ (t(x, y))′ and hence t(x, y) = t(x, y) ⊓ (t(x, y))′ = 0 whence x = y. (Observe that in Lc we have 0′ ≈ 0 ⊔ 0′ ≈ 1.) Finally, we are going to show the independence of identities (1) -- (4). Theorem 10. (i) Identities (1) -- (4) are independent, 8 (ii) identities (1) -- (4) do not imply identity (5). Proof. (i) The λ-lattice ({0, 1}, ⊔, ⊓, 0, 1) with x 0 1 x′ 0 0 satisfies (2), (3) and (4), but not (1) since (0 ⊔ 0) ⊔ (0′ ⊔ 0) = 0 ⊔ (0 ⊔ 0) = 0 6= 1. The λ-lattice ({0, 1}, ⊔, ⊓, 0, 1) with x 0 1 x′ 1 1 satisfies (1), (3) and (4), but not (2) since (1 ⊓ 1) ⊓ (1′ ⊓ 1) = 1 ⊓ (1 ⊓ 1) = 1 6= 0. The λ-lattice shown in Fig. 6 e c a ✉ ✉ ✉ ❅ 1 ✉ ❆ ✉ ❆ ❆ f ❅ ❅ ✉ b ✁ ✁ ❅ ✁ ✉ 0 ❆ ❆ ❆ ✁ ✁ ✁ ❆ ❆ ✉ ✁ ✁ d with a ⊔ b = 1, b ⊔ c = f , e ⊓ f = c and Fig. 6 x 0 a b c d e f 1 x′ 1 d e d a b d 0 satisfies (1), (2) and (4), but not (3) since (((a ⊓ b′) ⊔ f ) ⊔ (b ⊔ f )) ⊓ ((a ⊓ b′) ⊔ b) = (((a ⊓ e) ⊔ f ) ⊔ f ) ⊓ ((a ⊓ e) ⊔ b) = = ((a ⊔ f ) ⊔ f ) ⊓ (a ⊔ b) = (f ⊔ f ) ⊓ 1 = = f 6= 1 = a ⊔ b = (a ⊓ e) ⊔ b = = (a ⊓ b′) ⊔ b. 9 The λ-lattice shown in Fig. 7 a e ✁ ✁ ✁ ✑ ✑ ✁ ◗ ✉ ◗ ◗ ✉ ❆ ◗ ❆ ✑ ✑ ❆ b ◗ ◗ 1 ✉ ❅ ❅ ❅ ❅ g ✑ ✉ ❆ ✁ ✑ ✁ ❆ ◗ ◗ ✁ ◗ ✁ ◗ c ✉ ✁ ✑ ✑ ✑ ✑ ◗ ✑ f ◗ ✉ ◗ ✑ ◗ ✑ ✑ ◗ ❆ ✑ ✉ ❆ ❆ ✁ ❆ ✁ ◗ ✑ ❆ ◗ ✁ ✑ ✉ 0 ❆ ❆ ◗ ✑ ✉ d with a ⊔ b = e, a ⊔ c = e, a ⊔ d = f , b ⊔ c = 1, b ⊔ d = g, c ⊔ d = g, e ⊓ f = a, e ⊓ g = c, f ⊓ g = d and Fig. 7 x 0 a b x′ 1 g f f c d e f g 1 e d b a 0 satisfies (1), (2) and (3), but not (4) since (((d ⊓ g)′ ⊓ b) ⊓ (g ⊓ b)) ⊔ ((d ⊓ g)′ ⊓ g) = ((d′ ⊓ b) ⊓ b) ⊔ (d′ ⊓ g) = = ((e ⊓ b) ⊓ b) ⊔ (e ⊓ g) = (b ⊓ b) ⊔ c = = b ⊔ c = 1 6= c = e ⊓ g = d′ ⊓ g = = (d ⊓ g)′ ⊓ g. (ii) The λ-lattice shown in Fig. 8 c a ✉ ✉ ❅ 1 ✉ ❏ ❅ ❅ ✡ ❅ ✉ 0 ❏ ❏ ✡ ✡ ❏ ✡ ❏ ✡ ❏ ✡ ✉ b with Fig. 8 x 0 a b c 1 x′ 1 b a b 0 satisfies (1) -- (4), but not (5) since a ⊔ (a′ ⊓ (a ⊔ c)) = a ⊔ (b ⊓ c) = a ⊔ 0 = a 6= c = a ⊔ c. 10 References [1] L. Beran, Orthomodular Lattices. Algebraic Approach. D. Reidel, Dordrecht 1985. ISBN 90-277-1715-X. [2] G. Birkhoff and J. von Neumann, The logic of quantum mechanics. Ann. of Math. 37 (1936), 823 -- 843. [3] S. Bonzio and I. Chajda, A note on orthomodular lattices. Internat. J. Theoret. Phys. 56 (2017), 3740 -- 3743. [4] G. Bruns and J. Harding, Algebraic aspects of orthomodular lattices. Fund. Theories Phys. 111 (2000), 37 -- 65. [5] I. Chajda, G. Eigenthaler and H. Langer, Congruence Classes in Universal Algebra, Heldermann, Lemgo 2012. ISBN 3-88538-226-1. [6] I. Chajda and M. Kolar´ık, Variety of orthomodular posets. Miskolc Math. Notes 15 (2014), 361 -- 371. [7] I. Chajda and H. Langer, Directoids. An Algebraic Approach to Ordered Sets. Hel- dermann, Lemgo 2011. ISBN 978-3-88538-232-4. [8] I. Chajda and H. Langer, Weakly orthomodular and dually weakly orthomodular posets. Asian-Eur. J. Math. 11 (2018), 1850093 (18 pages). [9] I. Chajda and H. Langer, Weakly orthomodular and dually weakly orthomodular lattices. Order 35 (2018), 541 -- 555. [10] K. Husimi, Studies on the foundation of quantum mechanics. I. Proc. Phys.-Math. Soc. Japan 19 (1937), 766 -- 789. [11] V. Sn´asel, λ-lattices. Math. Bohem. 122 (1997), 267 -- 272. 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] Miroslav Kolar´ık Palack´y University Olomouc Faculty of Science Department of Computer Science 17. listopadu 12 771 46 Olomouc Czech Republic [email protected] 11 Helmut Langer TU Wien Faculty of Mathematics and Geoinformation Institute of Discrete Mathematics and Geometry Wiedner Hauptstrasse 8-10 1040 Vienna Austria, and Palack´y University Olomouc Faculty of Science Department of Algebra and Geometry 17. listopadu 12 771 46 Olomouc Czech Republic [email protected] 12
1503.03981	2	1503	2015-03-17T15:27:31	Finite Abelian algebras are fully dualizable	[ "math.RA" ]	We show that every finite Abelian algebra A from congruence-permutable varieties admits a full duality. In the process, we prove that A also allows a strong duality, and that the duality may be induced by a dualizing structure of finite type. We give an explicit bound on the arities of the partial and total operations appearing in the dualizing structure. In addition, we show that the enriched partial hom-clone of A is finitely generated as a clone.	math.RA	math	FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE WOLFRAM BENTZ, PIERRE GILLIBERT, AND LU´IS SEQUEIRA Abstract. We show that every finite Abelian algebra A from congruence- permutable varieties admits a full duality. In the process, we prove that A also allows a strong duality, and that the duality may be induced by a dualizing of finite type. We give an explicit bound on the arities of the partial structure A ∼ and total operations appearing in A . In addition, we show that the enriched partial hom-clone of A is finitely generated as a clone. ∼ 1. Introduction A full duality represents elements of abstract algebraic structures by using func- tions on a topological space that is often enriched with a relational and/or oper- ational structure, and vice versa. This representation allows us to solve algebraic questions by the way of the additional structure. For example in Stone duality, Boolean algebras are dual to Boolean spaces. Under this correspondence, the fa- miliar Cantor space is dual to the denumerable free Boolean algebra, with many of the universal properties of the Cantor space being dual counterparts to the natural universal properties of being a free algebra (the universal mapping property for example). In a natural full duality, the representation is constructed in a certain systematic way, using a generating algebra A and a corresponding topological structure A , ∼ called an "alter ego" of A. We say that A is fully dualizable, if there exists an alter ego A such that every algebra from the quasi-variety generated by A and ∼ every topological structure from the topological quasivariety generated by A has a ∼ representation. We remark that in case of a full duality, the correspondence can be extended to homomorphisms and continuous structure preserving maps, yielding a category-theoretic dual equivalence between the corresponding categories. A full duality is the symmetrized concept of a duality. The definitions of duality and dualizability differ from that of full duality and full dualizability by requir- ing that only the algebras in the quasivariety generated by A have duals, while the topological quasivariety generated by A might contain structures without a ∼ representation. Despite a growing understanding of duality theory, dualizability and full dual- izability of an algebra continue to be mysterious properties. For some classes of algebras (such as algebras generating congruence-distributive varieties) there exists a well-behaved dividing line between the dualizable and non-dualizable algebras. In other cases, the partial results available seem to defy any discernable pattern. Date: September 25, 2018. 2010 Mathematics Subject Classification. 08C20, 08B10 . Key words and phrases. natural duality, strong duality, dualizable algebra, Abelian algebra. 1 2 W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA This latter case includes classes of algebras that are otherwise considered to be well-understood, such as Malcev algebras (or even extensions of groups). Abelian algebras are among the most well-behaved classes of algebras, being polynomially equivalent to modules. Surprisingly, they have until very recently resisted any attempts to obtain results concerning their dualizability. At the 4th Novi Sad Algebra Conference in 2013, Kearnes and Szendrei presented a proof that showed the dualizability of all finite modules [7]. In an unpublished result, Bentz and Peter Mayr extended their argument to finite modules with all constants, which is equivalent to showing dualizability for all finite abelian algebras with all constants. Finally, Gillibert proved the dualizability of all finite Abelian algebras [4], answering a question from [1]. The same result was independently shown by Kearnes and Szendrei [8]. In this article, we will complete the remaining dualizability question for abelian algebras by showing the following Theorem. Theorem 1.1. Let A be an Abelian algebra generating a congruence-modular va- riety. Then A is fully dualizable. In fact we will show slightly more. Firstly, we show that a full duality can of finite type, and we give an explicit bound on be obtained by an alter ego A ∼ the arity of the (partial) functions and relations in A . Secondly, we establish full ∼ dualizability by showing that every finite Abelian algebra satisfies the stronger property of (adequately named) strong dualizability. Additionally, we obtain a structural result in clone theory by showing that the clone of all partial functions compatible with an Abelian algebra A is finitely gen- erated as a clone (Corollary 5.4). The proof of our main theorem relies on a technical condition from [5] (Theorem 2.9), that requires us to find a suitable factorization for each partial A-compatible function through a bounded set of partial functions. Our article is structured around this requirement as follows: In Section 2 we define basic terms and estab- lish several results about Abelian algebras. Section 3 provides a technical result about the factorization of projections on partial domains in the quasivariety gen- erated by A. This result will allow us to concentrate our further considerations on partial homomorphisms without proper extensions. In Section 4, we prove a crucial theorem about those partial homomorphisms: namely, a partial homomor- phism that cannot be extended must have a large domain. This result is then used in Section 5 to prove a factorization property for all partial homomorphisms, and to prove our main theorem. Section 6 and Section 7 contain an example calculation and a list of problems motivated by our research. Moreover, we have included an appendix that gives explicit bounds on the number of various algebraic objects. While the results of the appendix are used in our arguments, they are only necessary in establishing an explicit bound on the arities used in a fully dualizing alter ego. A reader without an interest in such an explicit bound may ignore the appendix and instead check the simple fact that all quantities in our argument are finite. For simplicity, we will not provide definitions of the various types of dualities in this article, relying on established technical results to prove our claims. For definitions and a general background on duality theory, we refer to the standard work [2]. FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 3 2. Basic concepts Given an algebra A, we denote by A its underlying set, and by Sub(A) the set of subalgebras of A. The variety Var(A) (respectively, the quasivariety QVar(A)) generated by A is the smallest classes of algebras, with the signature of A, that is closed under taking products, subalgebras, and homomorphic images (respectively, products, subalgebras, and isomorphic algebras). Given X ⊆ A we denote by hXi the subalgebra of A generated by X. For an arbitrary set X and a variety V, we denote by FV (X) the algebra freely generated by X in V. A subproduct algebra A ≤ ΠiAi is called a subdirect product if πi(A) = Ai for each projection πi. An algebra is subdirectly irreducible if whenever it is isomorphic to a subdirect product, it is already isomorphic to one of its factors. An algebra A is affine if there exists an Abelian group structure hA; +, 0, −i such that t(x, y, z) = x − y + z is both a term function of A and a homomorphism from A3 to A. A class of algebras C is affine if all of its algebras are. In the case of an affine variety V, it is easy to see that we may choose one term t that witnesses the affinity simultaneously for all members of V (e.g. we could take the term witnessing the affinity of FV (ω)). An algebra A is Abelian if [1A, 1A] = 0A, where 1A and 0A are the universal and trivial relations on A, and [·, ·] denotes the binary commutator on the con- gruences of A (we refer to [3] for the definition of the commutator). A class of algebras is Abelian if all of its members are. As usual, when dealing with commu- tator theoretic conditions, we restrict to algebras that generate congruence-modular varieties. With this condition, Abelian algebras and varieties coincide with affine algebras and varieties [3, Corollary 5.9], and we will use the two notations inter- changeably throughout the paper. Our results will rely exclusively on the defining property of affine algebras. We repeat several results about congruences of Abelian algebras from [4]. Definition 2.1 ([4], Definition 3.1). Let A be an Abelian algebra and B ∈ Sub(A). The congruence generated by B, denoted by ΘB is the smallest congruence of A containing B2. We remark that not every congruence of A can be written in the form ΘB for some subalgebra B, and that we might have ΘB = ΘC with B 6= C. Lemma 2.2 ([4], Lemma 3.3). Let A be an Abelian algebra, let B ∈ Sub(A), and let t be a term witnessing the affinity of A. Then ΘB = {(x, y) ∈ A2 ∃b ∈ B, t(x, y, b) ∈ B} . Note that this result implies that B is a congruence class of ΘB. Lemma 2.3 ([4], Corollary 3.7). Let A be an Abelian algebra and let B ∈ Sub(A) such that B is meet irreducible in the semilattice hSub(A); ∩i. Then A/ΘB is subdirectly irreducible. We will next establish several results about varieties of Abelian algebras and their relationship with varieties of modules. We remark that our result can be obtained from more general results in commutator theory, see in particular [3, Chapter 9]. As the bound estimates in our results require explicit descriptions of 4 W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA the constructions involved in the proofs of these auxiliary lemmas, we have decided to provide self-contained arguments. It is well known that any Abelian variety is polynomially equivalent to a variety of modules. The following lemma gives a corresponding result, stating that a variety of Abelian algebras with at least one nullary symbol is term-equivalent to a variety of modules extended with constants. Here and throughout, we will always consider all rings to be rings with unit. Lemma 2.4. Let V be a variety of Abelian algebras (over a language L ). Assume that c ∈ L is a nullary operation symbol. Then there exists a language L ′ ∪ C and an interpretation of L ′ ∪ C on the underlying sets of algebras in V with the properties listed below. Here, for A ∈ V, A′ denotes the algebra obtained from A by considering the interpretation of L ′ ∪ C , and V ′ will denote the collection of these algebras. (1) C contains only nullary symbols. (2) For every operation symbol f ∈ L ′ ∪ C there exists a term sf over L , such that f A′ = sA f for all A ∈ V. (3) For every f ∈ L there exists a term sf over L ′ ∪ C , such that f A = sA′ f for all A ∈ V. (4) L ′ may be considered as a module language over a suitable ring R. (5) The reduct of V ′ to L ′ is a variety of R-modules. (6) For every A ∈ V, cA denotes the neutral element of the module A′. Proof. Set C = {d d constant symbol in L } ∪ {f (c, . . . , c) f an operation symbol in L } , and set dV ′ = dV for all d ∈ C . Let +, −, 0, be binary, unary, and nullary operation symbols, respectively. We set and 0V ′ = cV. x +V ′ −V ′ y = tV (x, cV , y) , y = tV (cV , y, cV ) , Given an operation symbol f ∈ L of arity n ≥ 1, we consider the L -terms f1, . . . , fn given by fi(x) = f (c, . . . , c, x, c, . . . , c) − f (c, . . . , c) , (2.1) where the x appears at position i. Let R be the closure under +, −, and ◦ of all the fi and the term x modulo the equational theory of V. As t acts as a homomorphism of V, it is easy to check that the induced actions of +, −, 0, and ◦ on R are well- defined and give to R the structure of a ring with unit [x]. Moreover, each element of R is a set of V-equivalent unary terms in L . Hence we may interpret any element of R on V ′ in the same way as any of its member terms interprets in V. Set L ′ = R ∪ {+, −, 0}. Note that L ′ can be considered as an R-module language, and it is easy to check that the interpretations of L ′ induce an R-module structure on V. By construction, each operation in L ′ ∪ C interprets as some term in L . Conversely, all constant operations of L are in C , and so interpret as L ′ ∪ C - terms. Moreover, if f is an n-ary operation symbol in L , then f V(x1, . . . , xn) = [f1]V ′ (x2)+V ′ · · ·+V ′ (x1)+V ′ [f2]V ′ [fn]V ′ (xn)+V ′ (f (c, c, . . . , c))V ′ , FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 5 which is the interpretation of a term in L ′ ∪ C . The last assertion is obvious. (cid:3) The following lemma will be used to expand varieties by adding a constant operation. The structure of the resulting variety is similar to the original variety. Combining this lemma with Lemma 2.4 shows us that varieties of Abelian algebras are almost varieties of modules. Lemma 2.5. Let L be a similarity type. Pick a constant operation symbol c which is not in L . Given an L -algebra A and x ∈ A, we consider Ax to be the expansion of A to the language L ∪ {c}, where the new constant element is interpreted as x, that is cAx = x. The following statements hold. (1) If f : A → B is a morphism of L -algebras and x ∈ A, then f : Ax → Bf (x) is a morphism of L ∪ {c}-algebras. (2) If V is variety of L -algebras, then Vc = {Ax A ∈ V and x ∈ A} is a variety of L ∪ {c}-algebras. (3) If V is affine, then Vc is also affine. (4) If F is freely generated by x1, . . . , xn, y over V, then F y is freely generated by x1, . . . , xn over Vc. (5) If V is locally finite, then Vc is also locally finite. Proof. Let V be a variety of L -algebras. (1) Clearly, f preserves the structure of L -algebras. Moreover f (cAx) = f (x) = cBf (x) , therefore f is a morphism of L ∪ {c}-algebras. (2) Denote by W the variety of L ∪ {c}-algebras satisfying the equational theory of V. Then Vc ⊆ W. Conversely, let A ∈ W, denote by A(L ) the reduct of A to L . Note that A = A(L )cA . Therefore Vc = W is a variety of algebras. (3) Let V be affine, as witnessed by the L -term t. Then t is a term in L ∪ {c} that witnesses the affinity of Vc. (4) First note that F y is generated by x1, . . . , xn. Pick an algebra in Vc. It can be written as Av where A ∈ V and v ∈ A. Let u1, . . . , un ∈ A. As x1, . . . , xn, y freely generate A, there is a morphism f : F → A such that f (xi) = ui for all 1 ≤ i ≤ n, and f (y) = v. Hence f : F y → Av is a morphism. (5) Follows immediately from (4). (cid:3) Corollary 2.6. Suppose that V is generated by an Abelian algebra A of type L, where A = pα1 for distinct primes pi. Let Vc be the variety constructed in Lemma 2.5 and R be the ring constructed in Lemma 2.4 with regard to the variety Vc. Then R satisfies R = pr1 1 . . . pαk i for i = 1, . . . , k. k 1 . . . prk k , where ri ≤ α2 Proof. All elements of R are represented by unary terms λ in L ∪ {c} that satisfy λ(c) ≈ c in Vc, as this identity holds for the generating set (2.1). We call any such term λ a ring term. For each ring term λ with variable x, let ¯λ be the binary L-term over {x, y} obtained by replacing every occurrence of c with y. By our proof of Lemma 2.5 (2), the equational theory ¯λ is generated by the equational theory of V. It follows that any Vc-valid identity involving c is obtained from a V-valid identity by replacing a variable with c. Hence, as Vc = λ(c) ≈ c , V = ¯λ(y, y) ≈ y, and so all terms of the form ¯λ are idempotent in V. Let λ, µ be ring terms of Vc, a ∈ A, and assume that Aa = λ(x) ≈ µ(x). We claim that Vc = λ(x) ≈ µ(x). 6 W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA As Vc is generated by (Ay y ∈ A), it suffices to show that for every b ∈ A, Ab = λ(x) ≈ µ(x). Set ¯ν(x, y) = ¯λ(x, y) − ¯µ(x, y) + x, and note that this is an L-term. Let ν(x) = ¯λ(x, c) − ¯µ(x, c) + c be the corresponding L ∪ {c}-term. Then as Aa = λ(x) ≈ µ(x), it follows that Aa = ν(x) ≈ c. Hence ¯νA(x, a) = a for all x. Moreover, as ¯λ and ¯µ are idempotent ¯νA(a, a) = a = ¯νA(x, a) for all x ∈ A. As A is Abelian, we have that ¯νA(a, b) = ¯νA(x, b) for all x ∈ A. Hence νAb(a) = νAb(x) for all x ∈ A, and so ν is constant on Ab. As νAb(b) = b, we have that Ab = ν(x) ≈ c. This implies that Ab = λ(x) ≈ ν(x). As this holds for all b ∈ A, we have Vc = λ(x) ≈ µ(x), as required. It follows that if ν and µ represent distinct elements of R, then λAa 6= νAa. Conversely, any term function λAa with λAa (a) = a determines an element of R. Let ¯R be the set of these term functions, so that ¯R = R. As term functions of Aa, the elements of ¯R are compatible with the functions tAa and by construction, they are compatible with cAa. Hence every λAa ∈ ¯R is an endomorphisms of the algebra hA; tAa , cAa i. As this algebra has an underlying Abelian group structure, the result follows from Lemma 8.1. (cid:3) Given sets A, B we denote by F (A, B) the set of all maps A → B. Let A be a set. Given n ∈ N we consider the set of n-ary partial operations defined by: The set of all partial operations over A is Cn(A) =[{F (X, A) X ⊆ An , X 6= ∅} . C(A) = {∅} ∪ Gn∈N Cn(A) . Note that alternative definitions distinguish empty functions of different arity; the difference is immaterial for our results. Denote by πn i : An → A, ~x 7→ xi the canonical projection for all positive integers n and all 1 ≤ i ≤ n. A partial clone over a A is a set F ⊆ C(A), containing all projections and closed under composition of partial functions. Let F be a partial clone over A. A domain of arity n of F is a subset D of An such that there exists f ∈ F such that dom f = D. Lemma 2.7. Let F be a partial clone over a set A. Let n be a positive integer. Let C, D be domains of arity n of F . The following statements hold. (1) For all 1 ≤ i ≤ n, πn (2) The set C ∩ D is a domain in F . (3) Let p ∈ F of arity n. If D ⊆ dom p, then p ↾ D ∈ F . (4) If p = (p1, . . . , pn) : Ak → An are in F , then p−1(D) is a domain of F . i ↾ D ∈ F . Proof. Take f : C → A and g : D → A in F . (1) is a special case of (3), shown below. (2) π2 1(f (~x), g(~x)) is defined if an only if ~x ∈ C ∩ D, thus C ∩ D is a domain in F . (3) π2 1(p(~x), g(~x)) is defined if and only if ~x ∈ dom p ∩ dom g = D. Moreover p(~x) = π2 1(p(~x), g(~x)), for all ~x ∈ D, therefore p ↾ D ∈ V. (4) The domain of g ◦ p is p−1(D), therefore p−1(D) is a domain of F . (cid:3) We will consider partial functions whose domains are subalgebras and which are homomorphisms. For algebras A ≤ B and C and a homomorphism f from A to C, FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 7 we say that f has a proper extension if there is an algebra D with A < D ≤ B and a homomorphism f ′ from D to B that extends f . As mentioned in the introduction, we will avoid giving a detailed definition of full and strong dualizability. Instead we will utilize the following results from [2] and [5]. Definition 2.8 ([5]). A finite algebra A has enough total algebraic operations, if there exists ϕ : ω → ω such that for all B ≤ C ≤ An and every h ∈ hom(B, A), which has an extension to C, there exists X ⊆ hom(An, A) such that (1) X ≤ ϕ(B), (2) There is a homomorphism k from C/ ∩ {ker(f C ) f ∈ X} to A such that k ◦ α = h, where α is the natural map from B to C/ ∩ {ker(f C ) f ∈ X}. Theorem 2.9 ([5], Theorem 4.3). A finite dualizable algebra that has enough total algebraic operations is strongly dualizable. Definition 2.10 ([2], pg. 73). Let A be an algebra. The enriched partial hom- clone of A consists of all homomorphisms from B to A, for all subalgebras B of An, and all positive integers n. Theorem 2.11 ([2], Brute Force Strong Duality Theorem 3.2.2). Let A be a finite = hA, P, τ i, algebra. If some alter ego A ∼ yields a strong duality on A, where P is the enriched partial hom-clone of A and τ is the discrete topology on A. ′ yields a strong duality on A, then A ∼ Our next result is a special application of the M-shift strong duality Lemma from [2] to the alter ego hA, P, τ i. [2], Lemma 3.2.3). Let A, P, and τ be as in the Theorem Lemma 2.12 (cf. 2.11. Let P ′ ⊆ P be a generating set of P, that is, every h ∈ P is a composition of elements of P ′ and projections. If hA, P, τ i yields a strong duality on A, then hA, P ′, τ i yields a strong duality on A. 3. A generating set for domains of partial functions In order to show our main result, we want to establish that every Abelian algebra satisfies the conditions of Definition 2.8, so that we may use Theorem 2.9. The set X appearing in the definition can actually be taken as a set of coordinate projections. Hence to establish a necessary bound on X, we need to be able to show that partial compatible functions on A (i.e. homomorphisms from subpowers of A to A), factor though partial compatible functions of bounded arity. As a first step towards our result, in this section we show that we can generate all possible domains of such functions from a finite set. The following definitions are from [4]. Given Abelian algebras A and S and a homomorphism k : A → S, let Hk(A2, S) consist of all homomorphisms f : A2 → S that satisfy f (x, x) = k(x). We set ¯k ∈ Hk(A2, S) as ¯k(x, y) = k(y). In [4, Lemma 5.4], it is shown that hHk(A2, S); +¯ki is an Abelian group (where +¯k is defined as in Lemma 2.4), and that the isomorphism type of hHk(A2, S); +¯ki does not depend on k. We let hH(A2, S); +i stand for this isomorphism type. The following lemma, proved in [4, Lemma 5.7], expresses that (total) homomor- phisms f : An → S can be factored through a small power of A, which does not depend on n but depends only on hH(A2, S); +i. 8 W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA Lemma 3.1. Let A, S be algebras in a variety of Abelian algebras. Let N be a positive integer such that hH(A2, S); +i has a family of generators with N − 1 elements. Let f : An → S be a homomorphism. Then there exists a homomorphism p : An → AN that is a term in t, and a homomorphism q : AN → S, such that f = q ◦ p. . . . pαkβk k k 1 . . . pβk 1 . . . pαk k , S = pβ1 , and hH(A2, S); +i has a generating set of size max1≤i≤k(αiβi). Corollary 3.2. [4, Corollary 5.6] Let A and S be Abelian algebras such that A = pα1 for distinct primes pi. Then H(A2, S) divides pα1β1 1 Corollary 3.3. Let A be a finite Abelian algebra. Let pα1 decomposition of A. Let N = 1 + max1≤i≤k(α3 over A generated by t and all πN for all positive integer n and all subalgebras D of An. be the prime i ). Denote by F the partial clone 1 ↾ D ∈ F 1 ↾ C for C a subalgebra of AN . Then πn 1 pα2 2 . . . pαk k Proof. Let n be a positive integer, let D be a completely meet-irreducible subalge- bra of An. Set S = An/ΘD, and denote by f : An → S the canonical projection. By Lemma 2.2, {D} is the underlying set of a (one-element) subalgebra of S, moreover x ∈ D ⇐⇒ f (x) = D. That is f −1({D}) = D. As D is completely meet-irreducible, by Lemma 2.3, S is subdirectly irreducible. k . By Corollary 3.2 the group H(A2, S) has i ) = N − 1 elements. By Theorem 8.5, S divides pα2 1 . . . pα2 a generating family with max1≤i≤k(α3 Therefore, by Lemma 3.1, there exists a homomorphism p : An → AN , which is k 1 a term in t, and a homomorphism q : AN → S such that q ◦ p = f . Set C = q−1({D}) = {x ∈ AN q(x) = D}. As {D} is the underlying set of a subalgebra of S and q is a homomorphism, it follows that C is the underlying set of a subalgebra of AN , hence C is a domain of F . Moreover p is a term of t, thus it follows from Lemma 2.7(4) that p−1(C) is a domain of F . However p−1(C) = p−1(q−1({D})) = f −1({D}) = D, so D is a domain of F . Let B be an arbitrary subalgebra of An. We can write B as the intersection of finitely many underlying sets of completely subdirectly irreducible subalgebras of An. Since each of these sets is a domain of F , it follows from Lemma 2.7(2) that B is a domain of F . Therefore, by Lemma 2.7(3), πn (cid:3) 1 ↾ B is in F . 4. Extensions of Partial homomorphisms The results of Corollary 3.3 imply that we may generate a partial homomorphism from its extension to a larger domain and a bounded number of partial projections. Thus, the goal of this section is to extend partial homomorphisms of Abelian al- gebras (in a finitely generated variety of Abelian algebras). We will show that if the domain of a partial homomorphism is small enough, then the partial homomor- phism has a proper extension (cf. Lemma 4.5). We will first establish this result for modules before generalizing to Abelian algebras. Lemma 4.1. Let B, C be submodules of a module A. Let E be a module. Let f : B → E and g : C → E be homomorphisms. If f ↾ B ∩ C = g ↾ B ∩ C, then there exists a homomorphism h : B + C → E that is a common extension of f and g. Proof. Let b, b′ ∈ B and c, c′ ∈ C. Assume that b + c = b′ + c′. Then b − b′ = c′ − c belongs to B ∩ C, hence f (b) − f (b′) = f (b − b′) = g(b − b′) = g(c′ − c) = g(c′) − g(c) . FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 9 Therefore f (b) + g(c) = f (b′) + g(c′). It follows that the map h : B + C → E b + c 7→ f (b) + g(c) , is well-defined, and is a homomorphism of modules. Moreover h(b) = h(b + 0) = f (b) + g(0) = f (b) for all b ∈ B. Similarly h extends g. (cid:3) Lemma 4.2. Let V be a locally finite variety of algebras. Let A ∈ V. Let n be a positive integer. If each finitely generated subalgebra of A is generated by n elements, then A is finite. Proof. Assuming that A is infinite, there is an infinite sequence (xi)i∈N of distinct elements of A. Let k = F V(n). Denote by B the subalgebra of A generated by {x0, x1, x2, . . . , xk}. Note that B ≥ k + 1, but B is finitely generated, so is generated by n elements, hence B ≤ FV (n) = k, a contradiction. (cid:3) Lemma 4.3. Let A be an Abelian group such that A = pα1 k where p1, . . . , pk are distinct primes. Set N = 1 + α1 + · · · + αk. Let a1, . . . , aN ∈ A. Then there 1 . . . pαk are integers u1, . . . , ui−1 for some i with 1 ≤ i ≤ N , such that ai =Pi−1 Proof. Set A0 = {0}. Given 1 ≤ i ≤ N , denote by Ai the subgroup of A generated by {a1, . . . , ai}. Note that Ai is a subgroup of Aj for 0 ≤ i ≤ j ≤ N . j=1 ujaj. As a maximal chain of subgroups of A has size at most N , it follows that there is 1 ≤ i ≤ N such that Ai = Ai−1. Therefore ai ∈ Ai−1, so there are integers (cid:3) u1, . . . , ui−1 such that ai =Pi−1 j=1 ujaj. Lemma 4.4. Let V be a locally finite variety of R-modules, and E ∈ V finite. Then there exists a positive integer N such that, given modules B ⊆ C in V and a homomorphism f : B → E, if C/B is not generated by N elements, then f has a proper extension. Moreover, let R = pr1 k , where the pi are distinct primes, and assume that R as an R-module has ℓ strict non-trivial submodules. k , and E = pβ1 1 . . . pβk 1 . . . prk Then we can pick N = ℓ ×(cid:16)Pk i=1 riβi − 1(cid:17). Proof. Denote by F the R-module freely generated by {u}, so that F ∼= R as R-modules. Given a strict submodule G of F , denote by NG a positive integer with the following property: given ϕ1, . . . , ϕNG ∈ Hom(G, E), there are integers u1, . . . , ui−1 j=1 ujϕj. As G is a strict sub-module strictly, of F , it follows from Lemma 8.1 that Hom(G, E) divides pr1β1 . . . prkβk 1 k i=1 riβi. Set for some i with 1 ≤ i ≤ NG, such that ϕi =Pi−1 hence by Lemma 4.3, we may choose NG =Pk Note that N ≤ ℓ × (Pk i=1 riβi − 1). generated by N elements. N =X(NG − 1 G is a nontrivial strict submodule of F ) . (4.1) Let B ⊆ C in V and f : B → E be a homomorphism. Assume that C/B is not First note that if all finitely generated submodules of C/B are generated by N elements, then it follows from Lemma 4.2 that C/B is generated by N elements, 10 W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA which contradicts the assumption. Therefore there is a finitely generated submodule Q of C/B, whose minimal number of generators is k ≥ N + 1. Let P the submodule of C containing B such that P /B = Q. Note that {x1 +B, x2 +B, . . . , xk +B} generates Q if and only if B ∪{x1, . . . , xk} generates P . We say that x1, . . . , xk generate P over B. Given x ∈ C we denote by ϕx : F → C the unique homomorphism that maps u to x. Note that ϕx + ϕy = ϕx+y for all x, y ∈ C. Pick x1, . . . , xk ∈ P \ B, generating P over B, such that (ϕ−1 maximal. That is, if y1, . . . , yk generate P over B and ϕ−1 1 ≤ i ≤ k, then ϕ−1 sequence follows from the finiteness of Sub F . xi (B))1≤i≤k is xi (B) for all xi (B) for all 1 ≤ i ≤ k. The existence of such a yi (B) = ϕ−1 yi (B) ⊇ ϕ−1 Set Si = ϕ−1 xi (B). That is, Si is the largest submodule of F such that ϕxi (Si) ⊆ B, for each 1 ≤ i ≤ k. Note that ϕxi (F ) ∩ B = ϕxi (Si). Assume that Si = {0} for some 1 ≤ i ≤ k. Then ϕxi (F ) ∩ B = {0}. Let g : ϕxi(F ) → E, x 7→ 0. Note that dom f ∩ dom g = {0}, hence it follows from Lemma 4.1 that there exist a morphism h : B + ϕxi (F ) that extends both f and g. As xi 6∈ B, it follows that h is a proper extension of f . We now assume that Si 6= {0} for each 1 ≤ i ≤ k. Claim 1. The Si are strict submodules of F , for all 1 ≤ i ≤ n. Proof of Claim. Assume we have i such that Si = F . Hence u, the generator of F , belongs to Si, so xi = ϕxi(u) ∈ ϕxi(Si) ⊆ B, contradicting that xi /∈ B. (cid:3) Claim 1. As k ≥ N + 1 > N and the Si are strict nontrivial submodules of F , it follows that there are a submodule G of F , G strict and non-trivial, and I ⊆ {1, . . . , k}, such that I = NG and Si = G, for all i ∈ I. Set ψi = f ◦ ϕxi ↾ G. As I = NG, there is i ∈ I and a family of integers (uj)j∈J (where J = I \ {i}) such that ujψj. ψi =Xj∈J Let y = xi −Pj∈J ujxj . Note that x1, . . . , xi−1, y, xi+1, . . . , xk generates P over B. Moreover ϕy(s) = ϕxi−Pj∈J uj xj (s) = ϕxi (s) −Xj∈J ujϕxj (s) ∈ B , for all s ∈ G y (B) ⊇ G = Si. thus ϕ−1 ϕ−1 y (B) = Si = G. It follows from the maximality of (S1, . . . , Sk) that FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 11 Let z ∈ ϕy(F ) ∩ B, and take s ∈ G such that ϕy(s) = z. The following equalities hold f (z) = f (ϕy(s)) ujϕxj (s)  ujf (ϕxj (s)) ujψj(s) = f ϕxi (s) −Xj∈J = f (ϕxi (s)) −Xj∈J = ψi(s) −Xj∈J = 0 Denote by g : ϕy(F ) → E the constant 0 morphism. We have Set D = B + ϕy(F ). extension h : D → E. f ↾ ϕy(F ) ∩ B = 0 = g ↾ ϕy(F ) ∩ B . It follows from Lemma 4.1 that f and g have a common Note that y = ϕy(u) ∈ D and y 6∈ B, so h is a strict extension of f . (cid:3) Lemma 4.5. Let V be a locally finite variety of affine algebras. Let E be a finite algebra in V. Then there exists a positive integer N such that, given algebras B ⊆ C in V and a homomorphism f : B → E, if C/ΘB is not generated by N elements, then f has a proper extension. Proof. Denote by L the similarity type of V. Let c be a constant symbol that is not in L . We consider Vc, as defined in Lemma 2.5. Note that Vc is a locally finite variety of affine algebras, moreover there is a constant operation. It follows from Lemma 2.4 that there is a similarity type L ′ ∪ C satisfying (1)-(5) of Lemma 2.4. We denote by W the class of all reducts of algebras in Vc, to the type L ′. So W is a variety of modules. Let E be a finite algebra in V. Given y ∈ E, we consider its reduct Ey(L ′) ∈ W and take Ny as in (4.1), with regard to the variety W. Set N = 1+max{Ny y ∈ E}. Let B ⊆ C in V. Let f : B → E be a homomorphism. Assume that C/ΘB is not generated by N elements. Pick x ∈ B. Note that (C/ΘB)[x]ΘB (L ∪ {c}) is not generated by N − 1 elements. It follows that Cx(L ′)/Bx(L ′) = (C/B)x/B(L ′) is not generated by N − 1 elements. The map f : Bx → Ef (x) is a homomorphism of L ∪ {c} algebras, and so is a homomorphism of L ′ ∪ C -algebras. However, since N − 1 ≥ Nf (x), there exists h : D(L ′) → Ef (x)(L ′) where D(L ′) is a subalgebra of Cx(L ′) properly contain- ing B, and so h is a homomorphism of L ′-algebras that strictly extends f . Since B contains all the constants operations in C , D(L ′) contains the constants. Hence we can consider D(L ′ ∪ C ); moreover, as h extends f , it follows that h is a morphism of L ′ ∪ C -algebras, and so is a homomorphism of L -algebras. Therefore f has a proper extension (as homomorphism of L -algebras). (cid:3) The following corollary is an immediate consequence of Lemma 4.5. Informally the domain of a non-extensible homomorphism of Abelian algebras is large. 12 W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA Corollary 4.6. Let V be a locally finite variety of affine algebras. Let E be a finite algebra in V. Then there exists a positive integer N such that, given algebras B ⊆ C in V and f : B → E, if f has no proper extension, then C/ΘB is generated by N elements, and so is finite. 5. Factoring partial homomorphisms The main goal of this section is to factorize a partial homomorphism f : C → E (where C ⊆ An) through a smaller power D ⊆ AN , where N only depends on A and E. This will allow us to use Theorem 2.9 and to prove our main result. First note that Abelian algebras have the congruence extension property. To be more precise we give the following description of extensions of congruences. Lemma 5.1. Let A be a subalgebra of an Abelian algebra B. Let α be a congruence of A. Then there exists a smallest extension of α to B. It is the unique congruence β of B satisfying the following conditions. (1) For all (x, y) ∈ β, if y ∈ A, then x ∈ A. (2) β ∩ A2 = α. Proof. We can assume that the neutral element of + belongs to A. Indeed we pick 0 ∈ A and set x + y = t(x, 0, y). It follows that A is stable for +. Also note that x − y = t(x, y, 0) and t(x, y, z) = x − y + z. Let (x, x′) ∈ α, (y, y′) ∈ α. As α is compatible with t, it follows that (x + y, x′ + y′) = (t(x, 0, y), t(x′, 0, y′)) ∈ α. Therefore α is compatible with + and −. Define β = {(x, y) ∈ B2 (∃a ∈ B)((x − a, y − a) ∈ α)}. We will leave it to the reader to check that β is a congruence that satisfies conditions (1) and (2). Let γ be a congruence of B containing α. Let x, y, a ∈ B such that (x−a, y−a) ∈ α, so (x − a, y − a) ∈ γ, hence (x, y) = (x − a + a, y − a + a) ∈ γ. Therefore γ contains β, hence β is the smallest extension of α to B. Let γ be a congruence satisfying (1) and (2). First note that γ contains α, so γ contains β. Conversely, let (x, y) ∈ γ. Note that (x − y, 0) = (x − y, y − y) ∈ γ and 0 ∈ A, so x − y ∈ A, and so (x − y, y − y) ∈ γ ∩ A2 = α, that is (x, y) ∈ β. Therefore γ = β. (cid:3) Lemma 5.2. Let A ⊆ B be Abelian algebras. Let α be a congruence of A. Let β be the minimal extension of α to B. Then B/β = B/ΘA × A/α. Proof. First note that ΘA ⊇ α, hence ΘA ⊇ β. Also note that A/β is a subalgebra of B/β. Let (x/β, y/β) ∈ ΘA/β. There is u ∈ A/β such that y/β − x/β + u ∈ A/β, hence there is c ∈ A such that (y − x + c)/β ∈ A/β. That is (y − x + c)/β = a/β for some a ∈ A. It follows from Lemma 5.1(1) that (y − x + c) ∈ A. Therefore (x, y) ∈ ΘA, hence (x/β, y/β) ∈ ΘA/β. Conversely, let (x/β, y/β) ∈ ΘA/β. That is (x, y) ∈ ΘA. There is c ∈ A such that x − y + c ∈ A, so x/β − y/β + c/β ∈ A/β. As c/β ∈ A/β, it follows that (x/β, y/β) ∈ ΘA/β. This proves that ΘA/β = ΘA/β. Note that B/ΘA ∼= (B/β)/(ΘA/β) = (B/β)/(ΘA/β). Lemma 8.6 implies that B/ΘA = B/β/A/β. Now β ∩ A2 = α, so A/β ∼= A/α, and therefore B/β = B/ΘA × A/β. (cid:3) FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 13 Theorem 5.3. Let A and E be finite in a locally finite variety of Abelian algebras. Then there exists a positive integer ℓ such that, given a positive integer n, a sub- algebra C of An and a homomorphism h : C → E, there exists a homomorphism p : An → Aℓ and a homomorphism k : p(C) → E such that k ◦ p ↾ C = h. Moreover we can choose p to be a term in t. Proof. Denote by V a locally finite variety of Abelian algebras containing A and E. Note that most of the homomorphisms and commuting relations used in this proof are illustrated in Figure 1. Let N be as in Corollary 4.6. Set N ′ = FV (N ) × A. Given S ∈ V, such that S divides N ′, we pick ℓS such that hH(A2, S); +i has a generating family with ℓS elements. Set ℓ = 1 + max{ℓS S ∈ V and S divides N ′} . Let n be a positive integer, C ∈ Sub(An), and h : C → E a homomorphism. Let h′ : D → E be a maximal extension of h. Denote by η1 : C → D the inclusion homomorphism. As h′ extends h, we have (5.1) It follows from Corollary 4.6 that An/D is generated by N elements, hence h′ ◦ η1 = h . An/D divides FV (N ). Denote by ε1 : D → An the inclusion morphism. Denote by α the kernel of h′ and by β the minimal extension of α to An. Set R = D/α and S = An/β. Let ε2 : R → S be the canonical embedding. Let π : D → R and π′ : An → S be the canonical projections. Note that Note that h′ factors through π, so there is an embedding σ : R → A such that π′ ◦ ε1 = ε2 ◦ π . (5.2) h′ = σ ◦ π . (5.3) Hence R = D/α divides A. It follows from Lemma 5.2 that S = An/β = An/ΘD × D/α divides N ′ , hence 1+ℓS ≤ ℓ. That is hH(A2, S); +i has a generating family with ℓ−1 elements. From Lemma 3.1, we have morphisms p : An → Aℓ and q : Aℓ → S such that (5.4) As D is a subalgebra of An, it follows that p(D) is a subalgebra of Aℓ. Denote q ◦ p = π′ . by ε3 : p(D) → Aℓ the canonical embedding. Note that ε3 ◦ p ↾ D = p ◦ ε1 . Similarly we denote by η2 : p(C) → p(D) the inclusion morphism, so η2 ◦ p ↾ C = p ↾ D ◦ η1 . The following equalities are direct consequences of (5.2), (5.4), and (5.5) q ◦ ε3 ◦ p ↾ D = q ◦ p ◦ ε1 = π′ ◦ ε1 = ε2 ◦ π . (5.5) (5.6) (5.7) So q(ε3(p(D))) = ε2(π(D)) = ε2(R). However, ε2 is an embedding, so q(ε3(p(D))) corresponds to a subalgebra of R; hence, there is a homomorphism u : p(D) → R such that q ◦ ε3 = ε2 ◦ u . (5.8) 14 E σ R W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA h π h′ (5.3) l❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳❳ !❉❉❉❉❉❉❉❉❉❉❉❉❉❉❉ xrrrrrrrrrrrrrrrrrrrr (5.1) (5.2) π′ ε2 ε1 C η1 D p↾C (5.6) ✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻✻ ☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛☛ η2 S An p(C) (5.4) d❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍❍ (5.8) q u p Aℓ (5.5) p↾D g❖❖❖❖❖❖❖❖❖❖❖❖❖❖❖❖❖❖ ε3 p(D) Figure 1. Homomorphisms factor through sub-power of small dimension. It follows from (5.8) and (5.7) that ε2 ◦ u ◦ p ↾ D = q ◦ ε3 ◦ p ↾ D = ε2 ◦ π. As ε2 is an embedding it follows that u ◦ p ↾ D = π . (5.9) The following equalities hold σ ◦ u ◦ η2 ◦ p ↾ C = σ ◦ u ◦ p ↾ D ◦ η1 = σ ◦ π ◦ η1 = h′ ◦ η1 = h by (5.6). by (5.9). by (5.3). by (5.1). Therefore, denoting k = σ ◦ u ◦ η2 : p(C) → E, we have k ◦ p ↾ C = h. (cid:3) Corollary 5.4. Let A be a finite Abelian algebra. The partial clone of partial operations on A compatible with A is finitely generated. It is generated by partial operations of arity at most max{3, ℓ, N }, where N is the bound from Corollary 3.3, and ℓ is the bound from Theorem 5.3 in the case that E = A. Proof. Given integers 1 ≤ i ≤ n , we denote by πn tion on the i-th coordinate. i : An → A the canonical projec- Let ℓ be as in Theorem 5.3 for E = A. Let N be as in Corollary 3.3. Set X = {f : D → A D is a subalgebra of Aℓ}. Set Y = {πN 1 ↾C C is a subalgebra of AN }. Denote by F the partial clone generated by {t} ∪ X ∪ Y . o o _ O O p ! l o o K k x o o S s d 4 T g R R FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 15 Let n be a positive integer, let C be a subalgebra of An, let h : C → A be a homomorphism. By Theorem 5.3, there is a term p : An → Aℓ in t, and a homomorphism k : p(C) → A such that h = k ◦ p ↾ C. Note that k ∈ F , as it is a partial homomorphism of arity ℓ, so k ◦ p belongs to 1 C ∈ F , therefore, by Lemma 2.7(3), h = k ◦ p ↾ C F . Moreover, by Corollary 3.3, πn belongs to F . Therefore F is the set of all partial operations on A, compatible with A. More- over F is, by construction, finitely generated. The arity bound follows by construc- tion as well. (cid:3) We remark in passing that our corollary provide an additional proof that every finite Abelian algebra A is dualizable (even though it is unnecessarily complicated compared to the arguments in [4]). Dualizability of A follows as by the Duality Compactness Theorem [9, 10], it suffices to show that the enriched partial hom- clone of A is finitely generated. Our arguments are however not independent, as we rely on Lemmas 2.2, 2.3, and 3.1 from [4]. We are now ready to prove our final result about the strong dualizability of Abelian algebras. Lemma 5.5. An Abelian algebra has enough total algebraic operations. Proof. Let A be an Abelian algebra, let ℓ be as in Theorem 5.3 for A = E. We consider ϕ : ω → ω the constant map equal to ℓ. Let n be a positive integer, and let B ⊆ C be subalgebras of An. Denote by ι : B → C the inclusion homomorphism. Let h : B → A, let h′ : C → A be an extension of h, that is h′ ◦ ι = h. By Theorem 5.3 there exist a homomorphism p : An → Aℓ and a homomorphism k : p(C) → A such that h′ = k ◦ p ↾ C. We obtain k ◦ p ↾ C ◦ ι = h′ ◦ ι = h. The result now follows with X = {πi ◦ p 1 ≤ i ≤ ℓ}. (cid:3) Theorem 5.6. Finite Abelian algebras are strongly dualizable. Proof. Let A be a finite Abelian algebra. By Lemma 5.5, A has enough total alge- braic operations, moreover A is dualizable ([4], see also the remark after Corollary 5.4). By [5, Theorem 4.3], A is strongly dualizable. (cid:3) Our main Theorem 1.1 now follows from the well known fact that any strongly dualizable algebra is fully dualizable (see for example [2], Theorem 3.2.4). In our final result, we provide an explicit bound on the partial functions in the strongly dualizing alter ego. Theorem 5.7. Let A be a finite Abelian algebra with A = pα1 k . Then A is strongly dualized by hA; P, τ i, where τ is the discrete topology on A and P is the set of all algebraic partial operations on A of arity at most 1 . . . pαk i + 2α2 i(cid:1) , where N = 1 + pα4 1 . . . pα4 1 1 + max 1≤i≤k(cid:0)N α3 i − 1(cid:17). i=1 α3 k k ×(cid:16)Pk Proof. We may assume that A is non-trivial. As A is strongly dualizable, it is in particular strongly dualized by the strong brute force alter ego hA; P, τ i of Theorem 2.11. Moreover, by Lemma 2.12 we may replace P with a generating set, and by 16 W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA Corollary 5.4, P is finitely generated. Hence A is finitely strongly dualizable. It remains to establish the bound. Corollary 5.4 gives an arity bound of max{3, ℓ, ¯N}, where ¯N is the bound from Corollary 3.3 (there called N ), and ℓ is the bound from Theorem 5.3 in the case that E = A. By Corollary 3.3, ¯N = 1 + maxk i ). Thus ¯N and 3 are clearly smaller than the bound from the statement of the theorem, as we assumed that A was non-trivial. 1≤i≤k(α3 It remains to bound the quantity ℓ. With the notation of Theorem 5.3, if E = A, then V = Var A. In the theorem, ℓ is given as ℓ = 1 + max{ℓS S ∈ V and S divides N ′} . (5.10) where N ′ = FV (N ) × A for suitable N , and ℓS is a positive integer such that hH(A2, S); +i has a generating family with ℓS elements, for each S ∈ V with S dividing N ′. By Lemma 8.7, FV (N ) divides pN α2 1+2α1 . . . pN α2 . . . pN α2 k k+2αk 1 k+2αk Hence N ′ divides pN α2 S divides pN α2 from Corollary 3.2 that hH(A2, S); +i has a generating family with . Let S such that S divides N ′, then k , therefore, it follows . Moreover A = pα1 1 . . . pαk 1 1+2α1 k 1 1+α1 . . . pN α2 k k+αk . elements. Hence we can pick 1 + max i + 2α2 1≤i≤k(cid:0)N α3 1≤i≤k(cid:0)N α3 i(cid:1) i(cid:1) . ℓ = 1 + max i + 2α2 already bounded the Ny in the paragraph after (4.1) as Ny ≤ m × (Pk In Lemma 4.5, N was given as 1 + max{Ny y ∈ A}, where Ny is as in (4.1). We i=1 riαi − 1). Here, ri is the exponent of pi in the prime factorization of R where R is the ring of the module-variety Vc. Moreover, m is the number of strict, non-trivial submodules of R when considered as an R-module. By Corollary 2.6, ri ≤ α2 i . With Lemma 8.1, we see that m ≤ pα4 1 . . . pα4 k . k 1 Therefore N ≤ 1 + pα4 1 . . . pα4 1 k k × k Xi=1 α3 i − 1! . Hence ℓ is smaller than the estimate in the statement of the Theorem. The result follows. (cid:3) 6. Example We apply our results to an algebra whose examination was a crucial in developing the proofs of the previous sections. Consider the 8-element ring R = F2[x, y]/I, where I is the ideal generated by {x2, y2, xy, yx}. Let A be the module that is obtained by considering R as a module over itself. By [4], A is dualizable by an alter ego that includes all compatible relation of size 28. By our main result, A is strongly dualizable. A direct application of Theorem FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 17 5.7 will result in a very large bound of 702 · 281 + 46 on the arity of the partial operations in the alter ego. By adopting the results of the previous sections to this specific example, we can show that a lower bound suffices. As A is a module, we may use the value N (for E = A) from (4.1), instead of Ny from Lemma 4.5. Moreover, instead of the bound on N from the statement of Lemma 4.4, we can calculate N directly by (4.1), and the NG according to the definition preceding it, obtaining N = 14. As A generates a variety of modules, we see that N ′ in Theorem 5.3 takes the value 814 · 8 = 245. By Corollary 3.2, we may choose ℓ = 1 + 3 · 45 in Theorem 5.3. Hence we may obtain a strong duality by using an alter ego with "only" the compatible partial operations of arity 136. We close with several problems motivated by our results. 7. Problems Problem 1. Which Abelian algebras that do not generate congruence-modular varieties are dualizable? Which are fully and strongly dualizable? Problem 2. Are nilpotent dualizable algebras (from congruence-modular varieties) always fully dualizable? Are they strongly dualizable? We remark that in many well-behaved classes of algebras, dualizabilty, full du- alizability and strong dualizability coincide. Among nilpotent algebras, the results of [1] show that in the subclass of supernilpotent algebras, all non-abelian algebras are non-dualizable (and by [1] supernilpotence may be replaced by a slightly weaker condition). Problem 3. Can the arity bound in our main theorem be improved upon? We conjecture that a bound of the form (log2 A)n suffices, for some fixed integer n. Problem 4. Which Abelian algebras are strongly dualized by some alter ego that is a total structure? 8. Appendix: Counting homomorphisms and algebras Lemma 8.1. Let E, F be Abelian groups. Assume that E = pα1 F = pβ1 hold. and k , where the pi are distinct primes. Then the following statements 1 . . . pβk 1 . . . pαk k (1) The number of subgroups of E is at most pα2 (2) Hom(F , E) divides pα1β1 . . . pαkβk . 1 1 . . . pα2 k . k 1 k Proof. Note that Abelian subgroups of E are determined by their pi-Sylow sub- groups. Denote by Ei the pi-Sylow subgroup of E, such that Ei = pαi . Each i subgroup of Ei has a family of generators with αi elements. Therefore Ei has at most pα2 subgroups. It follows that (1) holds. i i We refer to [4, Lemma 4.1] for (2). (cid:3) Lemma 8.2. Let E, F be Abelian algebras. Assume that E = pα1 F = pβ1 hold. and k , where the pi are distinct primes. Then the following statements 1 . . . pαk 1 . . . pβk k 18 W. BENTZ, P. GILLIBERT, AND L. SEQUEIRA (1) The number of subalgebras of E is at most p1+α2 (2) Hom(F , E) divides p(α1+1)β1 . . . p(αk+1)βk . 1 1 1 k . . . p1+α2 k k . Proof. For each c ∈ E, x +c y = t(x, c, y) induces an Abelian group structure on E. There are E such structures. Let A be a subalgebra of E, then for c ∈ A, hA; +ci is a subgroup of hE; +ci. With Lemma 8.1, the number of subalgebras of E is at most E × pα2 . . . p1+α2 1 . . . pα2 k = p1+α2 k 1 . k k 1 1 We refer to [4, Lemma 4.1] for (2). (cid:3) Lemma 8.3 ([4], Lemma 4.2). Let E be an Abelian algebra such that E = 1 . . . pαk pα1 k , where the pi are distinct primes, and let N = 1 + max1≤i≤k(αi). Then E has a generating set with N elements. 1 . . . pαk Lemma 8.4. Let E be an Abelian algebra, and k a positive integer. Assume that E = pα1 k , where the pi are distinct primes. Then the number of distinct k-array term functions on E divides pkα2 Proof. Let t be the ternary function witnessing the Abelianess of E. As t is a homomorphism from E3 to E, every k-array term function of E is a homomorphism from hEk; ti to hE; ti. The result now follows with Lemma 8.2, (2). (cid:3) . . . pkα2 k+αk 1+α1 . k 1 The following theorem is a particular case of Kearnes result in [6], also see [4, Corollary 4.4]. Theorem 8.5. Let A be finite Abelian algebra. Let pα1 be the prime de- composition of A. Let S be a subdirectely irreducible algebra in Var A. Then S divides pα2 Lemma 8.6. Let A ⊆ B be Abelian algebras, then B/ΘA = B/A. 1 . . . pα2 k . 1 . . . pαk k k 1 Proof. Abelian algebras are well known to have congruence classes of equal cardi- nality (see for example [3], Corollary 7.5). As B is a congruence class of ΘB, the result follows. (cid:3) The following result may be easily shown using basic module theory. Lemma 8.7. Let R be a (unital) ring. Let V be a variety of R-modules with constants, let N be a positive integer. Suppose that there exists M ∈ V, a ∈ M such that the action r 7→ ra from R to M is injective. Then F V (N ) = RN ×F V (0), where R is considered as a module over itself. Corollary 8.8. Let A be an Abelian algebra, with A = pα1 are disjoint primes. Set V = Var A. Let N ∈ N. Then . . . pN α2 F V (N ) divides pN α2 k where the pi 1 . . . pαk k+αk 1+α1 . 1 k Proof. Assume that F is freely generated by x1, . . . , xN −1, y in V, then by Lemma 2.5, F y is freely generated by x1, . . . , xn−1 in Vc. As Vc is term-equivalent to a variety of modules with constants, by Lemma 8.7, F y = RN −1 × FVc(0), where 1 . . . pα2 R is the ring constructed in lemma 2.4. By Corollary 2.6, R divides pα2 k . It remains to estimate the cardinality of F = F Vc(0). If y = cF c(0), then {y} freely generates the reduct F ′ of F in V, once again by Lemma 2.5. As A generates V, F ′ corresponds to all unary term functions on A. By Lemma 8.4, F ′ divides pα2 , and the result follows. . . . pα2 k+αk 1+α1 (cid:3) k 1 1 k FINITE ABELIAN ALGEBRAS ARE FULLY DUALIZABLE 19 Acknowledgements The first author has received funding from the European Union Seventh Frame- work Programme (FP7/2007-2013) under grant agreement no. PCOFUND-GA- 2009-246542 and from the Foundation for Science and Technology of Portugal under PCOFUND-GA-2009-246542, SFRH/BCC/52684/2014, and through the CAUL / CEMAT project. References [1] W. Bentz and P. Mayr, Supernilpotence Prevents Dualizability, Journal of the Australian Mathematical Society 96 (2014), 1 -- 24. [2] D. M. Clark and B. A. Davey, Natural Dualities for the Working Algebraist. Cambridge Studies in Advanced Mathematics, 57. Cambridge University Press, Cambridge, 1998. [3] R. Freese and R. McKenzie, Commutator theory for congruence modular varieties. London Mathematical Society Lecture Note Series, 125. Cambridge University Press, Cambridge, 1987. [4] P. Gillibert, Abelian algebras are dualizable, preprint, http://arxiv.org/abs/1503.02651 [5] W. A. Lampe, G. F. McNulty and R. Willard, Full duality among graph algebras and flat graph algebras, Algebra Universalis 45 (2001), 311 -- 334. [6] K. A. Kearnes, Residual bounds for varieties of modules, Algebra Universalis 28 (1991), 448 -- 452. [7] K. Kearnes and ´A. Szendrei, Dualizable Algebras, Novi Sad Algebra Conference, June 5 -- 9, 2013. [8] K. Kearnes and ´A. Szendrei, Dualizable algebras with parallelogram terms, preprint, http://arxiv.org/abs/1502.02192 [9] R. Willard. New tools for proofing dualizability. In: Dualities, Interpretability and Ordered Structures (J. Vaz de Carvalho and I. Ferreirim, eds), Centro de Algebra da Universidade de Lisboa: 69 -- 74, 1999. [10] L. Zadori. Natural Duality via a finite set of relations. Bull. Aust. Math. Soc. 51:469 -- 478, 1995. (W. Bentz) Center of Algebra/Center for Computational and Stochastic Mathemat- ics, Universidade de Lisboa, Lisbon, Portugal E-mail address, W. Bentz: [email protected] (P. Gillibert) Pontificia Universidad Cat´olica de Valpara´ıso, Instituto de Matem´aticas, Valpara´ıso, Chile E-mail address, P. Gillibert: [email protected], [email protected] (L. Sequeira) Departamento de Matem´atica, Faculdade de Ciencias da Universidade de Lisboa, Lisbon, Portugal E-mail address, L. Sequeira: [email protected]
1901.10222	1	1901	2019-01-29T11:14:58	A note on the structure of underlying Lie algebras	[ "math.RA" ]	Every Lie algebra over a field $E$ gives rise to new Lie algebras over any subfield $F \subseteq E$ by restricting the scalar multiplication. This paper studies the structure of these underlying Lie algebra in relation to the structure of the original Lie algebra, in particular the question how much of the original Lie algebra can be recovered from its underlying Lie algebra over subfields $F$. By introducing the conjugate of a Lie algebra we show that in some specific cases the Lie algebra is completely determined by its underlying Lie algebra. Furthermore we construct examples showing that these assumptions are necessary. As an application, we give for every positive $n$ an example of a real $2$-step nilpotent Lie algebra which has exactly $n$ different bi-invariant complex structures. This answers an open question by Di Scala, Lauret and Vezzoni motivated by their work on quasi-K\"ahler Chern-flat manifolds in differential geometry. The methods we develop work for general Lie algebras and for general Galois extensions $F \subseteq E$, in contrast to the original question which only considered nilpotent Lie algebras of nilpotency class $2$ and the field extension $\mathbb{R} \subseteq \mathbb{C}$. We demonstrate this increased generality by characterizing the complex Lie algebras of dimension $\leq 4$ which are defined over $\mathbb{R}$ and over $\mathbb{Q}$.	math.RA	math	A note on the structure of underlying Lie algebras Jonas Der´e∗ January 30, 2019 Abstract Every Lie algebra over a field E gives rise to new Lie algebras over any subfield F ⊆ E by restricting the scalar multiplication. This paper studies the structure of these underlying Lie algebra in relation to the structure of the original Lie algebra, in particular the question how much of the original Lie algebra can be recovered from its underlying Lie algebra over subfields F . By introducing the conjugate of a Lie algebra we show that in some specific cases the Lie algebra is completely determined by its underlying Lie algebra. Furthermore we construct examples showing that these assumptions are necessary. As an application, we give for every positive n an example of a real 2-step nilpotent Lie algebra which has exactly n different bi-invariant complex structures. This answers an open question by Di Scala, Lauret and Vezzoni motivated by their work on quasi-Kahler Chern-flat manifolds in differential geometry. The methods we develop work for general Lie algebras and for general Galois extensions F ⊆ E, in contrast to the original question which only considered nilpotent Lie algebras of nilpotency class 2 and the field extension R ⊆ C. We demonstrate this increased generality by characterizing the complex Lie algebras of dimension ≤ 4 which are defined over R and over Q. 1 Introduction Given a field extension F ⊆ E and a Lie algebra g over the field E, we define the underlying Lie algebra gF by regarding g as a Lie algebra over only the subfield F . Many properties of the underlying Lie algebra correspond to properties of the original Lie algebra, e.g. the Lie algebra g is abelian if and only if the same holds for the Lie algebra gF . Similarly, it easily follows that being nilpotent or solvable is equivalent for the Lie algebras g and gF , where moreover the nilpotency class and the derived length correspond. In this paper we investigate more deeply the algebraic structure of the underlying Lie algebra gF in relation to the Lie algebra g. The main question we explore is under which conditions we can completely recover the algebraic structure of the Lie algebra g from the underlying Lie algebra gF . The motivation comes from a geometric problem of [4] in the special case of the field extension R ⊆ C, where the algebraic counterpart goes as follows. Question 1. Do there exist two non-isomorphic complex 2-step nilpotent Lie algebras g and h such that gR ≈ hR? The background of this question lies in the study of compact quasi-Kahler Chern-flat manifolds. These manifolds are obtained as the quotient of a 2-step nilpotent Lie group by a cocompact lattice, where the Lie algebra corresponding to the Lie group has a anti-bi-invariant almost complex structure. In [4] it is shown that these Lie algebras are exactly the underlying real Lie algebras of 2- step nilpotent complex Lie algebras with the anti-bi-invariant almost complex structure determined by the bi-invariant complex structure. ∗The author was supported by a postdoctoral fellowship of the Research Foundation -- Flanders (FWO). 1 Underlying Lie algebras occur in many other places as a tool to construct new Lie algebras with specific properties and this for arbitrary field extensions F ⊆ E of finite degree. For example, in [1] the authors use underlying Lie algebras over field extensions of finite degree to construct almost inner derivations which are not inner, related to the construction of isospectral nilmanifolds which are not isometric [7]. Another instance comes from the construction of Anosov Lie algebras, where the underlying rational Lie algebra of the Heisenberg Lie algebra h3(cid:16)Q(√d)(cid:17) with d ∈ N a square-free number was the first example of an Anosov Lie algebra which is not abelian, see [12]. These applications indicate that we should not study Question 1 for 2-step nilpotent Lie al- gebras over the complex numbers C only. We suggest the following generalized question for field extensions of finite degree. Question 2. Let F ⊆ E be a field extension of finite degree and g, h be Lie algebras over the field E. Under which conditions on the Lie algebras g and h does gF ≈ hF imply that g ≈ h? We assume that the field extension F ⊆ E has finite degree to ensure that the underlying Lie algebra gF has finite dimension if the original Lie algebra g is finite-dimensional. To study Question 2 we introduce the σ-conjugate gσ of a Lie algebra g over E for every σ ∈ Aut(E, F ). One way of defining the Lie algebra gσ is taking a basis for the Lie algebra g and applying the map σ to the structure constants for this basis, see Example 3.8. Our first main result states that two conjugate Lie algebras have the same underlying Lie algebra over F . Theorem 3.13. Let F ⊆ E be a field extension and σ ∈ Gal(E, F ). If we denote by gσ the σ-conjugate of the Lie algebra g over the field F , then (gσ)F ≈ gF . So in particular, to produce a negative answer on Question 1, it suffices to construct a complex Lie algebra g such that g 6≈ g, where g denotes the complex conjugate of the Lie algebra g. We give 2-step nilpotent examples in Section 5 and even give for every n ∈ N a real 2-step nilpotent Lie algebra which is the real underlying Lie algebra of exactly n different complex Lie algebras. This demonstrates the necessity of extra conditions on g and h in Question 2. The main theorem for studying Question 2 gives the structure of the underlying Lie algebra if we again extend the scalars on gF to the original field E for Galois extensions F ⊆ E. Theorem 4.1. Let F ⊆ E be a Galois extension and let g be a Lie algebra over the field E. Let gF be the underlying Lie algebra over F , then there is an isomorphism gF ⊗F E ≈ Mσ∈Gal(E,F ) gσ where gσ is the σ-conjugate of the Lie algebra g. We will show that if g is defined over F , then every conjugate Lie algebra over F is isomorphic to g, i.e. for every σ ∈ Aut(E, F ) we have gσ ≈ g. This fact also serves as a tool to characterize the Lie algebras which are defined over the field F and we illustrate this for Lie algebras of dimension ≤ 4 in Section 5.2. By using the techniques in [6] about indecomposable Lie algebras, we achieve the following solution to Question 2. Corollary 4.2. Let F ⊆ E be a Galois extension and g a Lie algebra over E which is defined over the subfield F . If h is another Lie algebra over E such that gF ≈ hF , then the original Lie algebras g ≈ h are isomorphic as well if either 1. the Lie algebra h is defined over F , or 2. the Lie algebra g is indecomposable. 2 As another consequence, we show that there are up to isomorphism only finitely many Lie algebras having the same underlying Lie algebra, including a way to describe them in Theorem 4.3. This paper is structured as follows. First we give some preliminaries about field extensions, restricting and extending the scalars of Lie algebras and how to decompose them into indecompos- able ideals in Section 2. Next, Section 3 introduces the conjugate of a Lie algebra and describes its properties, including a proof of Theorem 3.13. The proof and applications of Theorem 4.1 follow in Section 4 and we give some examples, both over the complex numbers as over other fields in Section 5, including the answer to Question 1. Finally we discuss some open questions about the existence of real forms and conjugate Lie algebras in Section 6. I would like to express my gratitude to J. Lauret for pointing out the open Acknowledgements question in his work [4] and K. Dekimpe for introducing me to the results about decomposing Lie algebras in [6]. I want to thank B. Verbeke for showing me the preliminary results of [1] about almost inner derivations. 2 Preliminaries In this section, we introduce the necessary background for the main results and fix notations for the remainder of this paper. First we present the main concepts from Galois theory, afterwards we demonstrate how to construct from a given Lie algebra over a field new Lie algebras over subfields and field extensions by restricting and extending the scalar multiplication respectively. Finally we recall some results from [6] about decomposing Lie algebras into indecomposable ideals. Galois extensions Let F ⊆ E be a field extension. The automorphisms Aut(E) of the field E form a group under the composition which contains the subgroup Aut(E, F ) = {σ ∈ Aut(E) ∀x ∈ F : σ(x) = x} of automorphisms fixing the subfield F . If Q ⊆ E is a field of characteristic 0, then Aut(E, Q) is equal to automorphism group Aut(E). Let F ⊆ E be a field extension such that E is algebraically closed, then every automorphism σ : F → F extends to an automorphism σ : E → E. In particular this property holds for every automorphism σ : F → F of a subfield F ⊆ C. The field E always forms a vector space over the field F and we define the degree [E : F ] of the field extension by [E : F ] = dimF (E). We say that F ⊆ E is a field extension of finite degree if [E : F ] < ∞ and in this case the order of the group Aut(E, F ) satisfies Aut(E, F ) ≤ [E : F ]. If equality holds in the latter inequality, we say that E is a Galois extension of F , calling Gal(E, F ) = Aut(E, F ) the Galois group of the field extension F ⊆ E. Under this assumption the field F is equal to the fixed field of the Galois group, i.e. x ∈ F if and only if σ(x) = x for all σ ∈ Gal(E, F ). In fact, the latter statement is equivalent to being a Galois extension for finite degree field extensions. If F ⊆ E is a field extension of finite degree in characteristic 0, then there always exists a field extension E ⊆ E ′ of finite degree such that F ⊆ E ′ is a Galois extension. Hereinafter we usually assume that field extensions are Galois extensions, although the main results could also be stated for general finite seperable field extensions, albeit more technically. For fixing notations, we introduce the easiest example of a Galois extension, namely R ⊆ C. : C → C Example 2.1. The Galois group Gal(C, R) = {1C, } consists of two elements, where is the complex conjugation map. Although this example is important for answering Question 1, we will also apply our main results about underlying Lie algebras to Galois extensions Q ⊆ E in Section 5. 3 Restricting and extending scalar multiplication Given a Lie algebra g over the field E, there are two different ways in which we can construct a new Lie algebra over a different field, namely by either restricting or extending the scalar multiplication. The first method, which we already discussed in the introduction, starts by taking a subfield F ⊆ E to construct the underlying Lie algebra gF . If V is a vector space over the field E, then the scalar multiplication is given by a map E × V → V which satisfies certain conditions for being a vector space. The underlying vector space VF is defined by the inclusion of the set F × V in E × V and thus defining a new scalar multiplication F × V ֒→ E × V → V . In this case we say that we restricted the scalar multiplication to the subfield F . Note that if V is finite-dimensional and if F ⊆ E is a field extension of finite degree, then VF is also finite-dimensional with dimension dimF (VF ) = [E : F ] dimE(V ). If g is a Lie algebra over E, then the Lie bracket will in particular be F -linear, hence gF forms a Lie algebra over F of dimension dimF (gF ) = [E : F ] dimE(g), which we call the underlying Lie algebra over F . In the special case of R ⊆ C we say that g is given by a bi-invariant complex structure on gR. The second method is to take a field extension E ⊆ F ′ and extend the scalar multiplication of the Lie algebra g by considering the tensor product g⊗E F ′. In some papers, this Lie algebra is denoted by a superindex F ′, but this notation is avoided because of its similarity to the notation for the underlying Lie algebra. In this case, dimF ′ (g ⊗E F ′) = dimE (g) and we say that g is an E-form of the Lie algebra g⊗E F ′. If a Lie algebra h over F ′ is isomorphic to g ⊗E F ′ for some Lie algebra g over E, we say that h is defined over E. In some texts the real underlying Lie algebra gR of a complex Lie algebra g is called the realification of g. We avoid this name because of its resemblance to the complexification of a real Lie algebra, which means extending the field to C, in contrast to restricting the field to R. The underlying real Lie algebra has been studied for its importance in studying simple Lie algebras over R, see for example [10, Page 34]. Although this is clear from the definitions, we want to emphasize that for a non-trivial field extension F ⊆ E and a Lie algebra g over E, it does not hold that gF ⊗F E is isomorphic to g, which can already be seen by comparing the dimensions. Up till now, it is not described what the relation is between these two methods of changing the field. The results of this paper can also be regarded as to which Lie algebra you get by first restricting and then again extending the scalar multiplication, see Theorem 4.1. In this part we recall some results of [6] about how to decompose Indecomposable Lie algebras a Lie algebra into indecomposable ideals. Recall that a Lie algebra g is called decomposable if there exists two proper ideals g1, g2 ≤ g such that g = g1 ⊕ g2. Note that in this case [g1, g2] = 0 since both subspaces are ideals and g1 ∩ g2 = 0. A Lie algebra is called indecomposable if it is not decomposable into proper ideals. A decomposition of a Lie algebra into indecomposable ideals is a finite set of ideals g1, . . . , gk such that g = gi k Mi=1 and every ideal gi is indecomposable. Although [6, Theorem 3.3] is only formulated for real Lie algebras, the proof works for Lie algebras over any field. Theorem 2.2. If g is a Lie algebra over the field E with two decompositions g = k Mi=1 gi = hi l Mj=1 4 into indecomposable ideals gi and hj, then the number of ideals is equal, i.e. k = l, and moreover up to renumbering the ideals it holds that gi ≈ hi for all 1 ≤ i ≤ k. The statement in [6] is a stronger version which also describes other uniqueness properties of the decomposition, but for our purposes the formulation of Theorem 2.2 suffices. Note that uniqueness of the decomposition only holds up to isomorphism, as we illustrate for clarity in the following example. Example 2.3. Let h3(E) be the Heisenberg Lie algebra of dimension 3 over the field E, i.e. the Lie algebra with basis X, Y, Z and only non-trivial bracket given by [X, Y ] = Z. It is easy to see that this Lie algebra is indecomposable since every proper ideal is abelian. Consider the direct sum g = h3(E) ⊕ h3(E) with basis Xi, Yi, Zi for the i-th component of the direct sum. On the one hand, we can decompose g as we defined it, namely as the direct sum of the ideals hX1, Y1, Z1i and hX2, Y2, Z2i. On the other hand, the ideals hX1, Y1 + Z2, Z1i and hX2, Y2, Z2i give a second decomposition of g. Although each of these ideals is isomorphic to h3(E), the de- compositions are not identical. This example illustrates that the isomorphisms we derive from Theorem 2.2 are not canonical, hence the main results of this paper should only be interpreted as isomorphic. The main application of Theorem 2.2 for our purposes is to recognize a Lie algebra as a component in a direct sum. The following two consequences will play an important role in the remainder of this paper. Since the proofs are elementary, we ommit the details. Proposition 2.4. Suppose that g and h are finite dimensional Lie algebras over a field E such that g ⊕ . . . ⊕ g } {z n times ≈ h ⊕ . . . ⊕ h } {z n times for some n > 0, then g ≈ h. Proposition 2.5. Let g1, g2, h be a finite dimensional Lie algebra over a field E. Suppose that g1 ⊕ h ≈ g2 ⊕ h, then g1 ≈ g2. Proof. Both propositions follow directly from Theorem 2.2 and writing either g, h or g1, g2, h as a direct sum of indecomposable ideals. 3 Conjugate Lie algebras One of the main ingredients to realize the results in this paper is the notion of conjugate Lie algebras. In this section we introduce this construction and prove the properties needed for the remaining part of this paper. In Theorem 3.13 we show that the underlying Lie algebras of conjugate Lie algebras are isomorphic, although the Lie algebras itself might not be isomorphic, hence leading to examples for Question 1. The concrete examples are only given in Section 5, after we develop some more machinery in Section 4 for computing the exact number of Lie algebras which have isomorphic underlying Lie algebra. σ-linear maps: From now on, we fix a field E over which we will consider vector spaces and Lie algebras. In order to introduce conjugate Lie algebras, we need to consider maps between vector spaces and Lie algebras which are not linear, but behave well under the action of a field automorphism σ ∈ Aut(E). 5 Definition 3.1. Let V and W be vector spaces over a field E and σ : E → E a field automorphism. We say that a map ϕ : V → W is σ-linear if ϕ (λv + µw) = σ(λ)v + σ(µ)w for all v, w ∈ V and λ, µ ∈ E. If ϕ is moreover bijective, we call it a σ-isomorphism between the vector spaces V and W . In the special case where V = g and W = h are Lie algebras, we say that ϕ is a σ-morphism if it is both σ-linear and if it preserves the Lie bracket, i.e. for all X, Y ∈ g. If ϕ is bijective we call it an σ-isomorphism between the Lie algebras g and h. ϕ(cid:0) [X, Y ](cid:1) = [ϕ (X) , ϕ (Y )] The following two lemmas are elementary, so we ommit the proof. Lemma 3.2. Let E be any field and U, V, W be vector spaces over the field E. For all field automorphisms σ, τ : E → E such that ϕ : U → V is σ-linear and ψ : V → W is τ -linear, the composition ψ ◦ ϕ : U → W is τ ◦ σ-linear. If ϕ and ψ are moreover σ- and τ -isomorphisms respectively, then ψ ◦ ϕ is a τ ◦ σ-isomorphism. If U, V and W are Lie algebras and ϕ, ψ are σ- and τ -(iso)morphisms between Lie algebras respectively, then ψ ◦ ϕ is a τ ◦ σ-(iso)morphism between Lie algebras. If ϕ is a σ-isomorphism, then ϕ−1 is a σ−1-isomorphism. Lemma 3.3. Let F ⊆ E be a field extension, σ ∈ Aut(E, F ) and V a vector space over E. Every σ-linear map ϕ : V → W induces a linear map ϕF : VF → WF between the underlying vector spaces over F . If V = g and W = h are Lie algebras and ϕ is a σ-morphism, then the map ϕF is a Lie algebra morphism between the underlying Lie algebras ϕF : gF → hF . If ϕ is a σ-isomorphism then ϕF is an isomorphism between the underlying Lie algebras. Although σ-morphisms between Lie algebras are not linear, they do share many properties with morphisms. Although this is not important for the remainder of this paper, one can show that the image ϕ(g) of a σ-morphism ϕ : g → h is a subalgebra of h and the kernel forms an ideal of g. For us, the following fact about linearly independent vectors is the most important one. Lemma 3.4. Let ϕ : V → W be a σ-isomorphism between vector spaces. linearly independent in V , then the images ϕ(X1), . . . , ϕ(Xn) are linearly independent in W . If X1, . . . , Xn are Proof. Assume that i=1 σ−1(λi)Xi = 0. Hence we obtain that σ−1(λi) = 0 since the Xi are linearly independent and consequently λi = 0 for all i. We conclude that the vectors ϕ(X1), . . . , ϕ(Xn) are linearly independent. for some λi ∈ E. By applying the map ϕ−1 we get that Pn In particular if ϕ : V → W is a σ-isomorphism between vector spaces, then dimE(V ) = dimE(W ). This is a property which we will use further on. λiϕ(Xi) = 0 n Xi=1 6 Construction of conjugate Lie algebras Given a Lie algebra g over the field E and σ ∈ Aut(E) an automorphism, we construct a Lie algebra gσ and a σ-isomorphism ϕσ : g → gσ between Lie algebras. First, let us describe the map between the vector spaces. Note that the automorphism σ : E → E induces a σ-isomorphism ϕσ : En → En between vector spaces by applying σ to every entry, so ϕσ(e1, . . . , en) = (σ(e1), . . . , σ(en)) for all (e1, . . . , en) ∈ En. After fixing a vector space basis X1, . . . , Xn for the Lie algebra g, we can identify g with En and hence also find a σ-isomorphism ϕσ : g → g on the vector space g. To make ϕσ into a σ-morphism, we need to adapt the Lie bracket on g accordingly to define a new Lie algebra gσ. As a vector space, gσ is just the Lie algebra g, but the Lie bracket [·, ·]σ : gσ × gσ → gσ is given by [X, Y ]σ = ϕσ(cid:16)h(ϕσ)−1 (X), (ϕσ)−1 (Y )i(cid:17) . The map [·, ·]σ : gσ × gσ → gσ is linear over E, since [λX, Y ]σ = ϕσ(cid:16)h(ϕσ)−1 (λX), (ϕσ)−1 (Y )i(cid:17) = ϕσ(cid:16)hσ−1(λ) (ϕσ)−1 (X), (ϕσ)−1 (Y )i(cid:17) = λϕσ(cid:16)h(ϕσ)−1 (X), (ϕσ)−1 (Y )i(cid:17) = λ[X, Y ]σ. It is immediate that [X, X]σ = 0 for all X ∈ gσ. Moreover the Jacobi identity on gσ also follows from the Jacobi identity on g since [X, [Y, Z]σ]σ = ϕσ(cid:16)h(ϕσ)−1 (X),h(ϕσ)−1 (Y ), (ϕσ)−1 (Z)ii(cid:17) . We conclude that [·, ·]σ is a well-defined Lie bracket on gσ and that the map ϕσ : g → gσ is a σ-isomorphism by construction. Note that the definition of the conjugate Lie algebra gσ and the σ-isomorphism ϕσ depends on the initial choice of basis X1, . . . , Xn, which determines the action of σ ∈ Aut(E) on the Lie algebra g. In the next proposition we show that by changing the basis X1, . . . , Xn, we get a Lie algebra which is isomorphic. Proposition 3.5. Let ϕ1 : g → h1 and ϕ2 : g → h2 be two σ-isomorphisms between Lie algebras over the field E. There exists a unique isomorphism ψ : h1 → h2 such that ψ ◦ ϕ1 = ϕ2. Proof. Consider the composition ψ = ϕ2 ◦ ϕ−1 : h1 → h2, then Lemma 3.2 implies that this is 1 a linear map over E. Since the condition also implies that it is bijective and preserves the Lie bracket, we conclude that ψ is in fact an isomorphism and ϕ2 = ψ ◦ϕ1 by construction. Uniqueness follows immediately, since the maps ϕ1 and ϕ2 are bijections. Corollary 3.6. The σ-conjugate Lie algebra gσ and the σ-isomorphism ϕσ : g → gσ are defined up to isomorphism and do not depend on the choice of basis X1, . . . , Xn for g. Hence from now on, we will just talk about the σ-conjugate of a Lie algebra over E without mentioning the choice of basis X1, . . . , Xn. We will denote the Lie bracket on gσ as [·, ·] for simplicity. Definition 3.7. Let g be a Lie algebra over the field E. If σ ∈ Aut(E) is an automorphism of the field E, we say that gσ is the σ-conjugate Lie algebra of g if there exists a σ-isomorphism ϕσ : g → gσ. We call g and h conjugate over a subfield F ⊆ E if there exists a σ ∈ Aut(E, F ) such that h is the σ-conjugate of g. 7 If g is a Lie algebra over E and X1, . . . , Xn are a basis for g, then we define the structure constants of g as the elements ck ij ∈ E such that Xk=1 [Xi, Xj] = n ck ij Xk. Given the structure constants of the Lie algebra g for the basis vectors X1, . . . , Xn, there is an easy way to compute the structure constants of gσ for the basis ϕσ(X1), . . . , ϕσ(Xn). Example 3.8. Let X1, . . . , Xn be a basis for g with the structure constants ck ij . For every σ- isomorphism ϕσ : g → gσ the vectors ϕσ(Xi) form a basis for gσ by Lemma 3.4. If we compute the structure constants for the Lie bracket on gσ, we get [ϕσ(Xi), ϕσ(Xj)] = ϕσ ([Xi, Xj]) ck ij Xk! = ϕσ n Xk=1 Xk=1 σ(ck = n ij )ϕσ (Xk) . Hence the structure constants of gσ for the basis ϕσ(X1), . . . , ϕσ(Xn) are equal to σ(ck ij ). Alternatively, we could have defined the σ-conjugate Lie algebra gσ from the action of σ on the structure constants as in Example 3.8. However, since the σ-isomorphism ϕσ : g → gσ plays such an important role in the remaining part of this paper, we defined conjugate Lie algebras by the existence of the map ϕσ as in Definition 3.7. Properties of conjugate Lie algebras: conjugate Lie algebras which we will use in the following parts. In this part, we introduce some properties about The first lemma shows that being conjugate over F is a transitive relation. Lemma 3.9. Let g1, g2 and g3 Lie algebra over the field E and F ⊆ E a subfield. Let σ, τ ∈ Aut(E, F ) be such that g1 is σ-conjugate to g2 and g2 is τ -conjugate to g3, then g1 is τ ◦σ-conjugate to g3. Proof. By definition, there exist ϕσ : g1 → g2 and ϕτ : g2 → g3 which are a σ- and τ -isomorphism between Lie algebras, respectively. By Lemma 3.2 the composition is a τ ◦ σ-isomorphism and hence the lemma holds. ϕτ ◦ ϕσ : g1 → g3 The next result we will use is that the conjugates of an indecomposable Lie algebra are again indecomposable. To prove this, we first study the conjugates of direct sums. Lemma 3.10. Let g and h be Lie algebras over the field E and σ : E → E an automorphism, then there is an isomorphism between the Lie algebras (g ⊕ h)σ ≈ gσ ⊕ hσ. Proof. Let ϕσ : g → gσ and ψσ : h → hσ be σ-isomorphisms, which exist by the definition of σ-conjugate Lie algebra. The natural map ϕσ ⊕ ψσ : g ⊕ h → gσ ⊕ hσ is again a σ-isomorphism, so in particular Definition 3.7 implies that the lemma holds. Corollary 3.11. Let g and h be Lie algebras over the field E which are conjugate, then g is indecomposable if and only if h is indecomposable. 8 Proof. This follows immediately from Lemma 3.10, since if g is isomorphic to a non-trivial direct sum of Lie algebras, then so is h. The next proposition on Lie algebras which are defined over a subfield F ⊆ E is crucial for positive answers on Question 2. Proposition 3.12. Let g be a Lie algebra over E which is defined over a subfield F ⊆ E, then every Lie algebra h which is conjugate over F to g is isomorphic to g. Proof. Since g is defined over F , there exists a basis X1, . . . , Xn for g such that the structure constants ck ij ∈ F . Take σ ∈ Aut(E, F ) such that there exists a σ-isomorphism ϕσ : g → h. By Example 3.8 the structure constants for the basis ϕσ(Xi) are equal to and hence h is isomorphic to g. σ(cid:0)ck ij(cid:1) = ck ij ∈ F Although Proposition 3.12 is elementary, we can use it to show that certain Lie algebras are not defined over a field F by showing that there exists σ ∈ Aut(E, F ) such that gσ 6≈ g, as we illustrate in Proposition 5.7. The following theorem indicates why we are interested in conjugate Lie algebras for studying underlying Lie algebras, in particular for answering Question 1. Theorem 3.13. Let g and h be Lie algebras over E which are conjugate over a subfield F ⊆ E, then the underlying Lie algebras over F are isomorphic, so gF ≈ hF . Proof. By definition, there exists σ ∈ Aut(E, F ) and a σ-isomorphism ϕσ : g → h. By Lemma 3.3 ϕσ induces an isomorphism between the vector spaces gF and hF . Since by definition it also preserves the Lie bracket, we conclude that gF ≈ hF . Thereom 3.13 suffices to give counterexamples for Question 1 and we will do so in Example 5.1 in Section 5. Before giving this example, we first investigate the algebraic structure of the underlying Lie algebra gF , allowing us to give a positive answer to Question 1 in some specific cases and describing the number of bi-invariant complex structures on a given real Lie algebra. 4 Embedding the underlying Lie algebra In this section we investigate the algebraic structure of the underlying Lie algebra gF depending on the Lie algebra g. The main goal is to give an answer to Question 2 and determine which Lie algebras have isomorphic underlying Lie algebras. To do this we embed the underlying Lie algebra gF as a F -form in a direct sum of conjugate Lie algebras over F , see Theorem 4.1 underneath. By applying this result in combination with the decomposition into indecomposable ideals from Theorem 2.2, we give a charactization for Lie algebras to have isomorphic underlying Lie algebra. Let g be a Lie algebra over any field E. From now on, we restrict our attention to Galois extensions F ⊆ E of finite degree. The results could also be formulated for general seperable extensions of finite degree, but this would make the statements more technical. This seems un- neccesary since every finite degree field extension in characteristic 0 lies in a Galois extension of finite degree, hence offering a wide range of applications in Section 5. Theorem 4.1. Let F ⊆ E be a Galois extension and let g be a Lie algebra over the field E. Let gF be the underlying Lie algebra over F , then there is an isomorphism where gσ denotes the σ-conjugate of the Lie algebra g as in Definition 3.7. gF ⊗F E ≈ Mσ∈Gal(E,F ) gσ 9 The techniques in this proof are similar to the ones in [3, Lemma 2.1] for constructing rational forms in Lie algebras. Proof. Consider the maps ϕσ : g → gσ for every σ ∈ Gal(E, F ). Define the natural map which is given by ϕσ in every component. The induced injective map ϕ : g → Mσ∈Gal(E,F ) gσ X 7→(cid:16)ϕσ(X)(cid:17)σ∈Gal(E,F ) gσ F  Mσ∈Gal(E,F ) ϕF : gF → is linear over F and we will show that the image of ϕF is an F -form of (cid:0)⊕σ∈Gal(E,F )gσ(cid:1) as a Lie algebra over E. Note that dimF (ϕF (gF )) = dimF (gF ) = [E : F ] dimE(g) = dimE(cid:0)⊕σ∈Gal(E,F )gσ(cid:1) , so it suffices to show that the image under ϕF of linearly independent vectors in gF over F are linearly independent in Lσ∈Gal(E,F ) gσ over E. To prove this, assume for a contradiction that this is not the case. Take X1, . . . , Xk ∈ gF which are linearly independent over F and such that there exists λ1, . . . , λk ∈ E with at least one λi non-zero and 0 = k Xi=1 λiϕ(Xi) = k Xi=1 λiϕσ(Xi)!σ∈Gal(E,F ) ∈ Mσ∈Gal(E,F ) gσ. (1) Without loss of generality we can assume that k is the minimal number for which such an example exists and hence it holds that all λi 6= 0. By multiplying by λ−1 if necessary we can assume that λ1 = 1. The minimality of k moreover implies that the λi are unique under the latter assumption. 1 Equation (1) is equivalent, by considering every component seperately, to λiϕσ(Xi) = 0 k Xi=1 (2) for every σ ∈ Gal(E, F ). We now claim that for every σ, τ ∈ Gal(E, F ) the equality τ (λi)ϕσ(Xi) = 0 k Xi=1 to Equation (2) for the automorphism ◦σ(cid:17)−1 τ −1 ◦ σ and get holds. To show this, we apply the map ϕσ ◦(cid:16)ϕτ −1 ◦σ(cid:17)−1 k Xi=1 ϕσ(cid:0)σ−1 (τ (λi)) Xi(cid:1) 0 = ϕσ ◦(cid:16)ϕτ −1 = k λiϕτ −1 ◦σ(Xi)! Xi=1 Xi=1 k = τ (λi)ϕσ(Xi). 10 Fixing τ ∈ Gal(E, F ) and using that the claim holds for all σ ∈ Gal(E, F ), we conclude that τ (λi) = λi by the fact that τ (λ1) = τ (1) = 1 and the uniqueness of the λi. Because this holds for every τ ∈ Gal(E, F ), we conclude that λi ∈ F , leading to a contradiction to the fact that X1, . . . , Xk are linearly dependent over F . As an application of this result, we get a positive answer on Question 1 under some specific assumptions for the Lie algebras g and h. Corollary 4.2. Let F ⊆ E be a Galois extension and g a Lie algebra over E which is defined over the subfield F . If h is another Lie algebra over E such that gF ≈ hF , then the original Lie algebras g ≈ h are isomorphic as well if either 1. the Lie algebra h is defined over F , or 2. the Lie algebra g is indecomposable. Proof. Since g is defined over F , every conjugate Lie algebra gσ is isomorphic to g, see Proposition 3.12. By applying Theorem 4.1, the Lie algebra gF is an F -form of g ⊕ . . . ⊕ g . We assume that } h is a Lie algebra with gF ≈ hF . . Consequently, by extending the field to E, Proposition 2.4 implies that g ≈ h. of h ⊕ . . . ⊕ h } If h is defined over F , then Theorem 4.1 implies in exactly the same way that hF is a F -form If g is indecomposable, we find that {z {z [E:F ] times [E:F ] times ≈ Mσ∈Gal(E,F ) hσ. g ⊕ . . . ⊕ g } {z [E:F ] times Since g is indecomposable and the number of components correspond, we find that the Lie algebras hσ are also indecomposable. In particular, since h is a indecomposable ideal of the right hand side, it is isomorphic to an indecomposable ideal on the left hand side, so isomorphic to g. In particular, in the second case of Corollary 4.2 we conclude that h is defined over F as well. As we will demonstrate in Example 5.3, the conditions on g or h are necessary for this corollary. More generally, given two Lie algebras g and h over E, we will fully characterize when the underlying Lie algebras gF ≈ hF are isomorphic. Observe that (g ⊕ h)F = gF ⊕ hF for all Lie algebras g and h over E. Together with Theorem 3.13 this implies that for all k ∈ N, if σi ∈ Gal(E, F ) and gi are Lie algebras over E for 1 ≤ i ≤ k, we find that k Mi=1 gi!F ≈ k Mi=1 gσi i !F . Our next theorem shows that the converse holds as well if the gi are indecomposable. This is hence the best possible result in this direction. Theorem 4.3. Let F ⊆ E be a Galois extension and g and h be Lie algebras over E such that gF ≈ hF . If we decompose into indecomposable ideals, then there exists σi ∈ Gal(E, F ) such that g ≈ gi k Mi=1 h ≈ gσi i . k Mi=1 11 j=1 hj with the ideals gi, hj indecomposable. Since gF ≈ hF Proof. Write g =Lk and hence also gF ⊗F E ≈ hF ⊗F E, applying Theorem 4.1 gives us j  . i=1 gi and h =Ll   Mσ∈Gal(E,F )   Mσ∈Gal(E,F ) i  ≈ Mj=1 Mi=1 hσ gσ k l Since gi and hj are indecomposable, we have that for every σ ∈ Gal(E, F ), the Lie algebras gσ i and hσ j are indecomposable by Corollary 3.11. By comparing the number of indecomposable ideals on the left hand side and right hand side of the isomorphism in (3) we get that k = l. We prove the theorem by induction on the number k of indecomposable ideals of g. First assume that k = 1 or thus that g is indecomposable. Since h is an indecomposable component of the right hand side in (3), Theorem 2.2 implies that h ≈ gσ for some σ ∈ Gal(E, F ), which proves the theorem in this case. For the induction step, we take k > 1. Since h1 is an indecomposable component of the right hand side of (3), there exists a 1 ≤ j ≤ k and a σ1 ∈ Gal(E, F ) such that h1 ≈ gσ1 j . Up to renumbering the ideals of g, we can assume that j = 1. For every σ ∈ Gal(E, F ), we find that 1 ≈ (gσ1 hσ 1 )σ ≈ gσ◦σ1 by Lemma 3.9. Hence 1 (3) (4) and by applying Proposition 2.5 we thus get that hσ Mσ∈Gal(E,F )   Mσ∈Gal(E,F ) Mi=2 k 1 ≈ Mσ∈Gal(E,F ) gσ◦σ1 1 gσ 1 ≈ Mσ∈Gal(E,F ) j  .   Mσ∈Gal(E,F ) hσ gσ i  ≈ l Mj=2 By applying the induction hypothesis to the isomorphism in (4), we get the statement of the theorem. Note that Theorem 4.3 implies that if two Lie algebras have isomorphic underlying Lie algebras, they have the same number of indecomposable ideals in a decomposition. As another consequence, we conclude that there are only a finite number of Lie algebras over E which can have isomorphic underlying Lie algebras. Corollary 4.4. Let F ⊆ E be a Galois extension. Given a Lie algebra h over the field F , there exist only finitely many (possibly none) Lie algebras g over the field E such that gF ≈ h. Proof. This follows from Theorem 4.3, since there are only a finite number of possibilities for the σi ∈ Gal(E, F ) and a finite number of indecomposable ideals. So in particular, a real Lie algebra admits only a finite number of bi-invariant complex structures up to isomorphism. We will give a method to compute the exact number in the following section. 5 Applications This section demonstrates some applications of the main results. We focus on R ⊆ C and Galois extensions Q ⊆ E because of the importance for applications, but the methods work for general Galois extensions. Recall from Example 2.1 that we denote the non-trivial element in Gal(C, R), namely the complex conjugation map, as : C → C. In the first subsection we construct negative examples for Question 1, where the strategy is to apply Theorem 3.13 on a complex 2-step Lie algebra g such that g 6≈ g, where g is the complex conjugate Lie algebra of g. To show that the complex conjugate Lie algebra is not isomorphic to the original one, we use invariants constructed from the Pfaffian form. The second subsection shows how to apply our main results as a tool to determine which Lie algebras in dimension ≤ 4 are defined over Q and R. 12 5.1 Bi-invariant complex structures on 2-step nilpotent Lie algebras A necessary condition for complex Lie algebras g with g 6≈ g is that g has no real forms. It is easy to construct such examples, for example as a semi-direct product C2 ⋊ C, see [11], but unfortunately this is harder to achieve for nilpotent Lie algebras. We start by recalling the approach from [4] for constructing such examples. The essence of this method is to associate to every 2-step nilpotent Lie algebra n a certain polynomial called the Pfaffian form. Using the properties of the Pfaffian form we construct an invariant cn ∈ C such that for the complex conjugate Lie algebra n the invariant is equal to cn = cn. Our example will be a Lie algebra for which cn ∈ C \ R. Let n be a 2-step nilpotent Lie algebra of dimension n over the field E. Write q = dimE ([n, n]) for the dimension of the commutator subalgebra and p = n − q, then we say in short that the If p is even, then we can associate to n, with a given basis, a Lie algebra n is of type (p, q). homogeneous polynomial fn(x1, . . . , xq) of degree p 2 in q variables with coefficients in E which we call the Pfaffian form of n. Hereinafter we will denote the vector space of homogeneous polynomials of degree p 2 in q variables as Pq, p The Pfaffian form depends on the choice of basis for n, hence it is not unique, but if two Lie algebras n1, n2 are isomorphic, then the Pfaffian forms are projectively equivalent, meaning that there exists a matrix A ∈ GL(q, E) and a constant k ∈ E ∗ such that 2 , so in particular fn ∈ Pq, p 2 . fn1(x1, . . . , xq) = kfn2 (A(x1, . . . , xq)). For more details about the construction of the Pfaffian form and the proof of this statement we refer to [8]. We now construct invariants for a complex 2-step nilpotent Lie algebra n from the Pfaffian form fn. Suppose that S, T : Pq, p 2 → C are homogeneous polynomials of the same degree which are SL(2, C)-invariant. It is an exercise to check that in this case if f1, f2 ∈ Pq, p are projectively equivalent then 2 S(f1) T (f1) = S(f2) T (f2) . Hence for a Lie algebra n of type (p, q), we can define an invariant S(fn) does not depend on the choice of a basis. T (fn) ∈ C which in particular We demonstrate this technique for the specific example of type (8, 2), where we use two different homogeneous polynomials S, T : P2,4 → C. The first one of degree 2 is defined as S(ax4 + bx3y + cx2y2 + dxy3 + ey4) = ae − 4bd + 3c2 whereas the second one of degree 3 is given by T (ax4 + bx3y + cx2y2 + dxy3 + ey4) = ace − ad2 + 2bcd − b2e − c3. On [5, Page 150] it is shown that these are indeed invariant under SL(2, C), leading to the invariant cn = (S(fn))3 (T (fn))2 for the Lie algebras n of type (8, 2). Note that the coefficients of the polynomials S and T are real numbers, even integers. Now we are ready to show that the following example of [4, Example 4.4.] has a complex conjugate which is not isomorphic. This example will be the building block for many other types of examples. 13 Example 5.1. Let gλ be the Lie algebra of dimension 10 with basis X1, . . . , X8, Z1, Z2 and Lie bracket [·,·] defined as [X1, X5] = Z1 [X4, X8] = Z1 [X4, X7] = Z2 [X2, X6] = Z1 [X2, X5] = Z2 [X1, X8] = −Z2 [X3, X7] = Z1 [X3, X6] = Z2 [X2, X7] = −λZ2. Using the definition of the Pfaffian form in [8], a computation shows that fλ(x, y) := fgλ(x, y) = x4 + λx2y2 + y4. Note that the Lie algebra gλ is indecomposable. By Example 3.8 the complex conjugate Lie algebra of gλ satisfies gλ ≈ gλ. The Lie algebra of Example 5.1 was already used in [4] for giving complex Lie algebras which are not defined over R. We apply the same techniques to find Lie algebras for which the complex conjugate Lie algebra is not isomorphic to the original one. Proposition 5.2. There exists λ ∈ C such that the complex Lie algebras gλ and gλ of Example 5.1 are not isomorphic. Proof. Take the polynomials S, T as introduced just above Example 5.1. By evaluating in the polynomial fλ we get S(fλ) = 3λ2 + 1 and T (fλ) = λ− λ3 and thus the invariant introduced above Example 5.1 is equal to c(λ) := cgλ = S3(fλ) T 2(fλ) = (cid:0)3λ2 + 1(cid:1)3 (λ − λ3)2 the proposition. Now pick any λ ∈ C such that c(λ) /∈ R, then c(cid:0)λ(cid:1) = c (λ) 6= c (λ) and hence gλ 6≈ gλ which proves There exist λ ∈ C \ R such that c(λ) ∈ R, for example c(i) = 2, which make the invariant c(λ) inapplicable for investigating whether or not gi and g−i are isomorphic. Note that Proposition 5.2 is true for every Lie algebra n of type (8, 2) for which the quotient S 3(fn) T 2(fn) does not lie in R, in particular also for Lie algebras as in [4, Example 4.5.]. In Corollary 4.2 we gave some conditions under which we recover the original Lie algebra from the underlying Lie algebra. Starting from Proposition 5.2 we construct 2-step nilpotent examples illustrating the necessity of these conditions. Example 5.3. Take λ ∈ C such that gλ 6≈ gλ with gλ the Lie algebra as in Example 5.1. Define the Lie algebra g = gλ ⊕ gλ, which is defined over R since it contains (gλ)R as a real form by Theorem 4.1. Note that the Lie algebra h = gλ ⊕ gλ satisfies gR ≈ hR by Theorem 3.13, but that the Lie algebras g and h are not isomorpic since g has gλ as an indecomposable component and h on the contrary does not. Comparing to Corollary 4.2, we point out that the Lie algebra h is not defined over R and is not indecomposable. Using the indecomposable Lie algebra of Example 5.1 we are able to construct examples with exactly n different bi-invariant complex structures. Example 5.4. Let gλ be the Lie algebra as in Example 5.1 for any λ ∈ C as in Proposition 5.2. Consider the complex Lie algebra then we show that the underlying Lie algebra gR has exactly k + 1 different bi-invariant complex structures. , g = gλ ⊕ . . . ⊕ gλ } k components {z 14 First of all, consider the Lie algebras hj = gλ ⊕ . . . ⊕ gλ } j components {z ⊕ gλ ⊕ . . . ⊕ gλ } k−j components {z for 0 ≤ j ≤ k, so in particular hk = g. The Lie algebras hj are all non-isomorphic because of Theorem 2.2 and Proposition 5.2. Moreover, Theorem 3.13 implies that the underlying real form (hj )R is isomorphic to gR, leading to at least k + 1 different bi-invariant complex structures on gR. Next we show that if h is a complex Lie algebra such that hR ≈ gR, then h ≈ hj for some 0 ≤ j ≤ k. In fact, this follows directly from Theorem 4.3 and using that every conjugate of gλ over R is either isomorphic to gλ or gλ. This shows that indeed gR has exactly k + 1 different bi-invariant complex structures. Example 5.3 works for every indecomposable complex Lie algebra g such that g 6≈ g. As an immediate consequence of this example we have the following statement. Corollary 5.5. For every n ∈ N, there exists a real 2-step Lie algebra gR such that gR has exactly n different bi-invariant complex structures. More generally, the method of Example 5.4 leads to the following result which counts the number of bi-invariant complex structures of an underlying real Lie algebra, giving an explicit form of Corollary 4.4. Theorem 5.6. Let g be a complex Lie algebra with decomposition g = Ll that ki i=1 gi into ideals such gi = gij Mj=1 is a direct sum of indecomposable ideals with either gi1 ≈ gij or gi1 ≈ gij for all 1 ≤ j ≤ ki. Moreover we assume that if gi1 ≈ gi′1 or gi1 ≈ gi′1, then i = i′ and that there exists 1 ≤ l0 ≤ l such that gi1 6≈ gi1 if and only i ≤ l0. There exists up to isomorphism exactly (ki + 1) l0 Yi=1 complex Lie algebras h such that gR ≈ hR. The proof is exactly the same as the argument in Example 5.4, by on the one hand constructing Lie algebras to show the lower bound and using Theorem 4.3 to show that we achieved all possible Lie algebras. We leave the details to the reader to check. 5.2 Real and rational forms in low dimensions Another application of Theorem 3.13 in combination with Proposition 3.12 is a method to verify which Lie algebras over E are defined over the subfield F ⊆ E. We demonstrate this by describing the complex Lie algebras up to dimension 4 which are definied over R and over Q. The complex Lie algebras up to dimension 4 have been classified and a list can be found for example in [2]. We will use the notations from the latter paper, in particular the Lie algebras as defined in Lemma 2 and 3. The importance of determining whether a Lie algebra is defined over Q or over R lies in geo- metric applications. If a Lie algebra is defined over R, then the Lie algebra is the complexification of a Lie algebra corresponding to a Lie group. Furthermore, in the nilpotent case Mal'cev showed in [9] that a 1-connected nilpotent Lie group admits a cocompact lattice if and only if the corre- sponding Lie algebra is defined over Q. Hence the fields Q and R are the most important ones for studying lattices in nilpotent Lie groups. 15 Note that most Lie algebras on [2, Page 4] are already given by a basis with rational structure constants, so there is only a limited number of examples we have to check. More specifically, the only Lie algebras which are a priori not defined over Q are the ones depending on a parameter λ, α, β ∈ C, namely r3,λ(C), r3,λ(C) ⊕ C, g1(α), g2(α, β), g3(α), g8(α). We will first deal with the last four types and treat with the first two afterwards. If a complex Lie algebra g is defined over R, then the complex conjugate g must be isomorphic to g by Proposition 5. Consider for every α ∈ C∗ the Lie algebra g1(α) of dimension 4 for which the Lie bracket of the basis vectors X1, X2, X3, X4 is determined by [X1, X2] = X2 [X1, X3] = X3 [X1, X4] = αX4. Note that g1(α) ≈ g1(β) if and only if α = β. Since g1(α) ≈ g1(α) by Example 3.8, we see that g1(α) is defined over R if and only if α ∈ R∗. A similar argument deals with the Lie algebras g2(α, β), g3(α) and g8(α) of [2], showing that these Lie algebras are defined over R if and only if the defining parameters are real. To check for which parameters α ∈ C∗ the Lie algebra g1(α) is defined over the rationals, we can make a similar analysis, by replacing the field extension R ⊆ C by the extension Q ⊆ Q(α) ⊆ C. If we assume that α /∈ Q, then either α is algebraic over Q, meaning that [Q(α) : Q] < ∞, In the first case there exists a or α is transcendental over Q, meaning that [Q(α) : Q] = ∞. Galois extension E of Q with α ∈ E, in the second case we take E = Q(α). both leading to an automorphism σ ∈ Aut(E) with σ(α) 6= α. By taking an extension σ : C → C of σ to C, we get that g1(α)σ ≈ g1(σ(α)) = g1(σ(α)) is not isomorphic to g1(α). This shows that the Lie algebra g1(α) and similarly the Lie algebras g2(α, β), g3(α) and g8(α) of [2] are defined over Q if and only if the defining parameters are rational. The final cases which remain are the 3-dimensional Lie algebra r3,λ(C) and the 4-dimensional Lie algebra r3,λ(C) ⊕ C for λ ∈ C∗. The Lie algebra r3,λ(C) has basis X1, X2, X3 and Lie bracket given by [X1, X2] = X2 [X1, X3] = λX3. The Lie algebras r3,λ(C) ≈ r3,µ(C) are isomorphic if and only if λ = µ±1. By Proposition 2.5, this also implies that the Lie algebras r3,λ(C)⊕ C ≈ r3,µ(C)⊕ C are isomorphic if and only if λ = µ±1. We answer the question over which fields Q ⊆ F ⊆ C of characteristic 0 these Lie algebras are defined in full generality. Proposition 5.7. The complex Lie algebra r3,λ(C) or r3,λ(C) ⊕ C with λ ∈ C∗ is defined over F if and only if λ2 + aλ + 1 = 0 (5) for some a ∈ F . Proof. We only give the proof for r3,λ(C) since the other case is identical. To show that Equation (5) is a necessary condition, we assume that r3,λ(C) is defined over F and we consider the field E = F (λ). First we show that λ has to be algebraic over F . If not, then there would exist an automorphism σ ∈ Aut(E, F ) such that σ(λ) = 2λ. By extending σ to an automorphism σ : C → C, Proposition 3.12 and Example 3.8 imply that (r3,λ(C))σ ≈ r3,σ(λ)(C) = r3,2λ(C) and λ 6= 2λ 6= λ−1, we get a contradiction, so λ must be algebraic over F . Similarly as the previous argument, it follows that for every conjugate µ of λ over F that r3,λ(C) ≈ r3,µ(C) by Proposition 5. Since this is only possible for λ = µ±1, we conclude that λ can have at most two conjugates. If λ has only one conjugate, then λ ∈ F and hence Equation (5) is satisfied for a = − 1+λ2 λ ∈ F . If λ has two conjugates λ and λ−1, then the minimal polynomial 16 of λ over F has degree 2 and constant term λλ−1 = 1, so Equation (5) is satisfied for the minimal polynomial over F . Next, we show that Equation (5) is a sufficient condition. Assume that λ satisfies Equation (5) for some a ∈ F . If λ ∈ F , then of course the Lie algebra r3,λ(C) is defined over F , so we can assume that λ /∈ F . Note that a = 2 implies that λ = −1 ∈ F and in particular we get a 6= 2. Write E = F (λ) ⊇ F , which is a Galois extension of degree 2 and take 1 6= σ ∈ Gal(E, F ) the non-trivial element of the Galois group. Note that σ(λ) = λ−1, so in particular r3,λ(E) ≈ r3,σ(λ)(E) ≈ (r3,λ(E))σ, which is a necessary condition for being defined over F . To show that r3,λ(C) is defined over F , we give an explicit Lie algebra g over C with structure constants in F which is isomorphic to r3,λ(C). The Lie algebra g is given by the basis Y1, Y2, Y3 and Lie bracket [Y1, Y2] = Y3, [Y1, Y3] = (a − 2) Y2 + (2 − a) Y3, [Y2, Y3] = 0. Consider the basis X1, X2, X3 given by X1 = σ(λ) + 1 a − 2 X2 = −(σ(λ) + 1)Y2 + Y3 X3 = −(λ + 1)Y2 + Y3. A computation, where we use λ + σ(λ) = −a, shows that Y1 = − 1 λ + 1 Y1 [X1, X2] = 1 λ + 1 ((a − 2)Y2 + (2 − a)Y3 − (σ(λ) + 1)Y3) = −(σ(λ) + 1)Y2 + 1 − a − σ(λ) λ + 1 Y3 = X2 [X1, X3] = 1 λ + 1 ((a − 2)Y2 + (2 − a)Y3 − (λ + 1)Y3) Y3 = σ(λ)X3 1 − a − λ = −(σ(λ) + 1)Y2 + λ + 1 [X2, X3] = 0 and thus g ≈ r3,σ(λ)(C) ≈ r3,λ(C). This ends the proof of the proposition. For example the Lie algebra r3,i(C) is defined over R, although the defining parameter does not lie in R. We summarize Proposition 5.7 for the special case F = R as follows. Proposition 5.8. Let g be a complex Lie algebra of dimension ≤ 4, then g is defined over R if and only if g ≈ g. We conjecture that this is the case for all complex Lie algebras, see Section 6. The methods of this section can be applied for every field E and any classification of Lie algebras, for example also for nilpotent Lie algebras up to dimension 7. 6 Open questions The results of this paper open new directions for further research on Lie algebras over different fields. This sections gathers these open questions and relates them to other problems. Proposition 3.12 states that if a Lie algebra is defined over a field F , then every conjugate Lie algebra over the field F is isomorphic. In the special case of the field extension R ⊆ C, this is stating that if the complex Lie algebra is defined over R, then g ≈ g. We conjecture that the converse holds as well. 17 Conjecture 1. If g is a complex Lie algebra such that g ≈ g, then g is defined over R. For Lie algebras of dimension ≤ 4 the conjecture holds, see Proposition 5.8. We do not know of any method which shows that g is not defined over R which does not imply that in fact g 6≈ g as in Proposition 5.2. We now state an equivalent conjecture about the existence of real forms in direct sums. If the complex Lie algebra g is defined over R, then also the direct sum g ⊕ g is defined over R. The conjecture whether the converse of this statement always holds is equivalent to Conjecture 1. Conjecture 2. Let g be a complex Lie algebra such that g ⊕ g is defined over R, then also g is defined over R. Proof of equivalence between Conjecture 1 and 2: First assume that Conjecture 1 is true and as- sume that g is a complex Lie algebra with g ⊕ g defined over R. By Proposition 3.12 we get that g ⊕ g ≈ g ⊕ g or hence that g ⊕ g ≈ g ⊕ g by Lemma 3.10. By Proposition 2.4 we conclude that g ≈ g or thus that g has a real form by assumption. Next we assume that Conjecture 2 holds and that g is a complex Lie algebra with g ≈ g. By Theorem 4.1 we get that gR is a real form of g ⊕ g ≈ g ⊕ g. Hence by assuming Conjecture 2 we find that g is defined over R. The generalized version of Conjecture 1 for arbitrary field extensions is widely open, since we only dealt with fields E of degree 2 over Q before. Question 3. Let F ⊆ E be a Galois extension and g a Lie algebra over E. Assume that gσ ≈ g for all σ ∈ Gal(E, F ), does it hold that g is defined over F ? A positive answer to Question 3 would relate the existence of F -forms to studying conjugate Lie algebras, leading to new directions for studying Lie algebras over different fields. References [1] Dietrich Burde, Karel Dekimpe, and Bert Verbeke. Almost inner derivations of classes of lie algebras. (preprint), 2018. [2] Dietrich Burde and Christine Steinhoff. Classification of orbit closures of 4-dimensional com- plex Lie algebras. J. Algebra, 214(2):729 -- 739, 1999. [3] Jonas Der´e. A new method for constructing Anosov Lie algebras. Trans. Amer. Math. Soc., 368(2):1497 -- 1516, 2016. [4] Antonio J. Di Scala, Jorge Lauret, and Luigi Vezzoni. Quasi-Kahler Chern-flat manifolds and complex 2-step nilpotent Lie algebras. Ann. Sc. Norm. Super. Pisa Cl. Sci. (5), 11(1):41 -- 60, 2012. [5] Igor Dolgachev. Lectures on invariant theory, volume 296 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 2003. [6] David J. Fisher, Robert J. Gray, and Peter E. Hydon. Automorphisms of real Lie algebras of dimension five or less. J. Phys. A, 46(22):225204, 18, 2013. [7] Carolyn S. Gordon and Edward N. Wilson. Isospectral deformations of compact solvmanifolds. J. Differential Geom., 19(1):241 -- 256, 1984. [8] Jorge Lauret. Rational forms of nilpotent Lie algebras and Anosov diffeomorphisms. Monatsh. Math., 155(1):15 -- 30, 2008. 18 [9] Anatoliı I. Mal'cev. Nilpotent torsion -- free groups. Izvestiya Aka. Nauk SSSR., Ser. Mat., (13):pp. 201 -- 212, 1949. [10] A.L. Onishchik and E.B. Vinberg. Lie Groups and Lie Algebras III, volume 41 of Encyclopedia of Mathematics. Springer Verlag, Berlin Heidelberg New York, 1994. [11] Hans Samelson. Notes on Lie algebras. Van Nostrand Reinhold Mathematical Studies, No. 23. Van Nostrand Reinhold Co., New York- London-Melbourne, 1969. [12] S. Smale. Differentiable dynamical systems. Bull. Amer. Math. Soc., 73,:pp. 747 -- 817, 1967. 19
1904.10643	1	1904	2019-04-24T05:10:25	Heun algebras of Lie type	[ "math.RA" ]	We introduce Heun algebras of Lie type. They are obtained from bispectral pairs associated to simple or solvable Lie algebras of dimension three or four. For $\mathfrak{su}(2)$, this leads to the Heun-Krawtchouk algebra. The corresponding Heun-Krawtchouk operator is identified as the Hamiltonian of the quantum analogue of the Zhukovski-Voltera gyrostat. For $\mathfrak{su}(1,1)$, one obtains the Heun algebras attached to the Meixner, Meixner-Pollaczek and Laguerre polynomials. These Heun algebras are shown to be isomorphic the the Hahn algebra. Focusing on the harmonic oscillator algebra $\mathfrak{ho}$ leads to the Heun-Charlier algebra. The connections to orthogonal polynomials are achieved through realizations of the underlying Lie algebras in terms of difference and differential operators. In the $\mathfrak{su}(1,1)$ cases, it is observed that the Heun operator can be transformed into the Hahn, Continuous Hahn and Confluent Heun operators respectively.	math.RA	math	HEUN ALGEBRAS OF LIE TYPE NICOLAS CRAMP´E†,∗, LUC VINET∗, AND ALEXEI ZHEDANOV‡ Abstract. We introduce Heun algebras of Lie type. They are obtained from bispectral pairs associated to simple or solvable Lie algebras of dimension three or four. For su(2), this leads to the Heun-Krawtchouk algebra. The corresponding Heun-Krawtchouk operator is identified as the Hamiltonian of the quantum analogue of the Zhukovski-Voltera gyrostat. For su(1, 1), one obtains the Heun algebras attached to the Meixner, Meixner-Pollaczek and Laguerre polynomials. These Heun algebras are shown to be isomorphic the the Hahn algebra. Focusing on the harmonic oscillator algebra ho leads to the Heun-Charlier algebra. The connections to orthogonal polynomials are achieved through realizations of the under- lying Lie algebras in terms of difference and differential operators. In the su(1, 1) cases, it is observed that the Heun operator can be transformed into the Hahn, Continuous Hahn and Confluent Heun operators respectively. 1. Introduction This paper elaborates on Heun algebras. This subject is associated to the notion of algebraic Heun operator recently introduced [12]. Let us recall the main points. It is known that the properties of the orthogonal polynomials of the Askey scheme can be encoded algebraically. Indeed the recurrence operator X and Y, the one of which the polynomials are eigenfunctions, form a bispectral pair that generate a quadratic algebra. For the Askey- Wilson (AW) polynomials sitting at the top of the q-tableau, the defining relations of the AW algebra read [25]: (1.1) (1.2) [X, [X, Y ]] = ρXY X + a1X 2 + a2{X, Y } + a3X + a4Y + a5 [Y, [Y, X]] = ρY XY + a1{X, Y } + a2Y 2 + a3Y + a6X + a7 where ρ = q2 + q−2 − 2 and ai, i = 1, ...7 are real parameters. When q = 1 and hence ρ = 0, the cubic terms in (1.1) and (1.2) drop and the reduced relations define the Racah algebra [9], [7]. To all families of hypergeometric polynomials belonging to the Askey classification, there corresponds an algebra of that type which is a special case or a limit of the AW algebra. The algebraic Heun operator is the generic bilinear element constructed from the generators X, Y (1.3) W = r1[X, Y ] + r2{X, Y } + r3X + r4Y + r5 where ri, i = 1, 2, . . . , 5 are arbitrary constants. By Heun algebras, we mean the ones generated by the pairs (X, W ) or (Y, W ). The kernel of W can be viewed as solutions of an ordinary eigenvalue problem if the parameters ri, i = 1, . . . , 4 are fixed and r5 is the eigenvalue, or of a generalized problem if W is regarded as a multiparameter linear pencil. When X is multiplication by the variable and Y the hypergeometric operator, the algebra that is realized is the Jacobi one where ρ = a1 = a6 = a7 = 0. It has been shown [11] that in this case W coincides with the differential Heun operator and the equation W ψ = 0 amounts to the standard Heun equation with four regular Fuchsian singularities. The name 1 2 N. CRAMP´E, L. VINET, AND A. ZHEDANOV algebraic Heun operator has been coined as a result of this observation; as already explained this concept applies to all bispectral problems and associates an operator of Heun type to the polynomials of the Askey scheme. One would then refer to the Heun operator say of Racah type and correspondingly to the Heun-Racah algebra for example. The exploration of these structures has been initiated recently with a focus on the "higher" polynomials. Attention was first paid to the Hahn polynomials [23]; this led naturally to a finite difference version of the Heun equation on the uniform lattice. Examining the little and big q-Jacobi polynomials from this angle allowed [2] to give context to q-Heun operators that had been identified [21] in connection with integrable models. Last, the Heun-Askey-Wilson algebra was thoroughly examined [3]. We here wish to look in a similar way at the "lower" polynomials and study the Heun structures attached to bispectral pairs that generate a Lie algebra. This will lead to a description of the Heun algebras associated to the Krawtchouk, Meixner, Meixner-Pollaczek, Laguerre and Charlier polynomials while establishing and clarifying the general foundations of the subject. The presentation will unfold as follows. We shall begin in Section 2 with a definition of the Heun algebra of Lie type. It will be shown to be generically isomorphic to the Hahn algebra [8, 7, 6] which is obtained from the AW algebra by setting ρ = 0 as well as a2 = 0 in (1.1) and (1.2). We shall pursue by introducing the Lie algebra for a bispectral pair X and Y with the so-called "ladder" property, this will amount to dropping the nonlinear terms in the AW relations. The corresponding W will then be seen to generate together with X (or Y ), the Heun algebra of Lie type. The definitions of the Lie algebras su(2), su(1, 1) and ho - the harmonic oscillator one, will be recorded at the end of the section. Section 3 will focus on the case where the Lie algebra for the bispectral pair is su(2). It will specify the associated Heun algebra and will indicate that it cannot be mapped to the Hahn algebra over R. It will explain in addition that the Heun operator can be viewed as as the Hamiltonian of the quantum analog of the Zhukovski-Voltera gyrostat. The case of su(1, 1) will be treated in Section 4. Three situations will be distinguished depending on whether the generator X is of elliptic, hyperbolic or parabolic type. The Heun algebras will again be characterized and seen in these instances to be isomorphic to the Hahn algebra. Section 5 will be dedicated to the harmonic oscillator algebra ho; the associated Heun algebra will be determined and observed not to be equivalent to the Hahn algebra. By recalling in Section 6 certain realizations of su(2), su(1, 1) and ho in terms of difference and differential operators, we shall recognize in each case that X and Y become the bispectral operators of the Krawtchouk, Meixner, Meixner- Pollaczek, Laguerre and Charlier polynomials confirming that the Heun algebras of the Lie type identified are to be associated to each of these families of orthogonal polynomials. In view of the fact that the Heun-Meixner, Heun-Meixner-Pollaczek and Heun-Laguerre algebras are isomorphic to the Hahn algebra, it is further shown that the corresponding Heun operators can be transformed into the Hahn, Continuous Hahn and Confluent Heun operators respectively. The paper will end with concluding remarks in Section 7. 2. Heun algebra of the Lie type Definition 2.1. The Heun algebra H of the Lie type is generated by X and W with the following defining relations (2.1) (2.2) [[X, W ], X] = x0 + x1X + x2X 2 + x3W , [W, [X, W ]] = y0 + y1X + y2X 2 + y3X 3 + x1W + x2{X, W }, HEUN ALGEBRAS OF LIE TYPE 3 where xi and yi (i = 0, 1, 2, 3) are free parameters. In the Heun algebra H, the following element (2.3) Ω = z1X + z2W + z3{X, W } + z4XW X + z5X 2 + z6W 2 + z7([X, W ])2 + z8X 3 + z9X 4 is central if the parameters zi are given by (2.4) z3 = x1 , z6 = x3 , z9 = y3/2 . (2.5) (2.6) z1 = 2y0 − x3y2/3 , z4 = 2x2 , z7 = 1 , z2 = −x2x3 + 2x0 , z5 = y1 − x3y3/2 , z8 = 2y2/3 , Proposition 2.1. Generically, the Heun algebra of Lie type is isomorphic to the Hahn algebra. To show that, we introduce the following invertible map between the pairs (X, W ) and (X, W ) where (2.7) with µ and ν solutions of W = W + µX + νX 2 (2.8) 2ν2x3 − 4νx2 + y3 = 0 and y2 − 3µx2 − 3νx1 + 3µνx3 = 0 . It follows that X and W satisfy the Hahn algebra (2.9) (2.10) where (2.11) (2.12) [[X, W ], X] = x0 + x1X + x2X 2 + x3W [W , [X, W ]] = y0 + y1X + x1W + x2{X, W } x0 = x0 , y0 = y0 − µx0 , x1 = x1 − µx3 , x2 = x2 − νx3 , y1 = y1 − 2µx1 − 2νx0 + µ2x3 x3 = x3 Remark 1. The word generically is used in the statement of Proposition 2.1 to indicate that it assumes freeness of the parameters. We shall observe in the specific cases that will be discussed in the following that there are instances where the parameters need to belong to C or where there are no solutions if the parameters are not constrained. In such situations, the algebraic equivalence would not prevail. Remark 2. It is interesting to point out that the truncated reflection algebra attached to the Yangian of sl(2) has been shown in [4] to be isomorphic to the Hahn algebra defined above. We bring at this point the generic Lie algebra A for a bispectral pair X and Y . Start with the linear relations: (2.13) (2.14) (2.15) [X, Y ] = Z [Z, X] = aX + c2Y + d2 [Y, Z] = bY + c1X + d1, where a, b, c1, c2, d1 and d2 are real parameters. With the presence of the constants d2 and d1 in the right-hand sides of (2.14) and (2.15), we are de facto assuming in general the presence of an additional central element I which we will omit writing. It is readily seen that one must have a = b for the Jacobi identity [X, [Y, Z]] + [Y, [Z, X]] + [Z, [X, Y ]] = 0 to be verified. Now assume that X has a nondegenerate discrete real spectrum {λp, p ∈ Z} in a representation space with up the corresponding eigenvectors. It is easy to see that for Y 4 N. CRAMP´E, L. VINET, AND A. ZHEDANOV to act tridiagonally on that space: Y up = fp+1up+1 + gpup + fpup−1, in view of (2.14), one must have (2.16) − (λp+1 − λp)2 = c2. This implies the condition c2 < 0. To have (Leonard) duality requires that X and Y be on the same footing. To that end, one would thus ask that there be reciprocally a basis in which Y has also a real discrete spectrum and X is tridiagonal. This would necessitate in turn that c1 < 0. In the following, we shall want however to cover more general bispectral situations where the spectrum of Y could be pure imaginary or continuous. We shall hence not impose restrictions on c1. As a consequence, the Heun algebras to be defined will relate to bispectral problems that are not only the purely discrete ones. Definition 2.2. The generic Lie algebra A over R for the bispectral pair X and Y is defined by (2.14) and (2.15) with (2.17) a = b, c2 < 0. As already indicated this corresponds to eliminating the non-linear terms in (1.1) and (1.2). The following element (2.18) C = 2d1X + 2d2Y + b{X, Y } + c1X 2 + c2Y 2 + Z 2 is central in A. Remark 3. It will be useful as we proceed to take note of the following two obvious facts. First, like the equations (1.1) and (1.2) of which they are a special case, the equations (2.14) and (2.15) preserve their form under affine transformations of X and Y . Second, it is also clear that the elements X = V XV −1 and Y = V Y V −1 obtained from performing on X and Y the same similarity operation will satisfy the same relations as X and Y . Proposition 2.2. Let X and Y be the generators of A. The Heun operator (2.19) W = r1[X, Y ] + r2{X, Y } + r3X + r4Y + r5 together with X satisfy the relations of the Heun algebra H of the Lie type with (2.20) (2.21) (2.22) (2.23) x0 = d2r4 − c2r5 y0 = −d2r4r3 − 2d2r5r2 + c2r5r3 + d1r2 y1 = −4d2r3r2 + c2r2 3 + 8d1r4r2 + c1r2 y2 = 12d1r2 2 + 6c1r4r2 − 6br3r2 4 + (d1c2 − bd2)(r2 4 + (c1c2 − b2)(r2 y3 = 8c1r2 2. x1 = 2d2r2 − c2r3 + br4 x2 = 2br2 x3 = c2 1 − r2 2) − br5r4 − 2r2r4C 1 − r2 2) − 2br4r3 − 4br5r2 − 4r2 2C Remark 4. The defining relations of A are invariant under the exchange of X and Y provided one performs: c1 ↔ c2, d1 ↔ d2. Taking into account the definition of W (2.19), we see that the algebra realized by Y and W is the same as the one stemming from the pair X, Y with the coefficients obtained from the ones given in Proposition 2.2 by doing also the changes r1 → −r1, r3 ↔ r4. Remark 5. The central element Ω of H with the parameters given above is (2.24) Ω = (r2 4 + c2(r2 1 − r2 2))C − 2c2r5r2b − 2c2r2 5 + 2r4c2r2d1 + 2d2r4r5 + d2 2(r2 1 − r2 2) . HEUN ALGEBRAS OF LIE TYPE 5 In the next three sections, we shall identify the specific Heun algebras that are obtained when the algebra A is isomorphic to su(2), su(1, 1) or the oscillator algebra ho. Let us record here the definitions of these algebras. The Lie algebras su(2) and su(1, 1) are generated by J1, J2 and J3 subject to (2.25) [J1, J2] = ±iJ3, [J2, J3] = iJ1, [J3, J1] = iJ2. The upper (resp. lower) sign is associated to su(2) (resp. su(1, 1)). The quadratic Casimir of these algebras is (2.26) c = J 2 1 + J 2 2 ± J 2 3 . The harmonic oscillator algebra ho is solvable and has four generators N, A, A† and I that satisfy the commutation relations [20]: (2.27) [N, A] = −A, [N, A†] = A†, [A, A†] = I, [I, N] = [I, A] = [I, A†] = 0. It is familiar that k = N − A†A is central in ho. 3. The case of the Lie algebra su(2) We shall focus here on su(2) and assume therefore in this section that the generators (J1, J2, J3) obey the commutation relations of this algebra, that is those corresponding to the upper signs in (2.25) and (2.26). 3.1. Specialization. Take X and Y to be (3.1) X = αJ3 + βJ1 , Y = J3 with the coefficients α = cos θ and β = − sin θ. Let us first explain that there is no loss of generality with this choice keeping in mind Remark 3. We can always pick Y in a preferred direction and allowing for scaling have it as above. Assume then that X is initially of the generic form l1J1 + l2J2 + l3J3. The adjoint representation will transform this su(2) element while preserving l2 = l2 3. Under a rotation about the 3-axis, Y does not change and the other one can be transformed into αJ3 + βJ1 with α2 + β2 = l2. Scaling finally by 1/l we arrive at the expression for X in (3.1). It is readily checked that (X, Y ) satisfy the relations (2.14) and (2.15) of the algebra A 2 + l2 1 + l2 with (3.2) d1 = d2 = 0 , c1 = −1 , b = α and c2 = −(α2 + β2) = −1 . The fact that d1 = d2 = 0 indicates that there is no center and that the algebra is three- dimensional. The Casimir C of A is related to c as follows: (3.3) C = α{X, Y } − X 2 + c2Y 2 + ([X, Y ])2 = −β2c. Observe that X is obtained from Y through an automorphism: (3.4) X = U(θ)Y U(θ)−1 with U(θ) = eiθJ2. Let W be given by (2.19). From the results of the previous section, we can see that the algebras generated by the pairs (X, W ) and (Y, W ) will be similar We shall therefore only focus on the Heun algebra generated by X and W . 6 N. CRAMP´E, L. VINET, AND A. ZHEDANOV Proposition 3.1. The elements X and W associated to su(2) as per (3.1) satisfy the rela- tions of the Heun algebra of Lie type with (3.5) (3.6) (3.7) (3.8) x1 = r4α − c2r3 , x0 = −c2r5 , y0 = r5(c2r3 − αr4) − 2r2r4C , y1 = (r2 y2 = −6r2(r4 + αr3) , 1)(c2 + α2) − 4r2r5α + c2r2 2 , y3 = −8r2 2 − r2 x2 = 2r2α , x3 = c2 , 3 − 2αr3r4 − r2 4 − 4r2 2C , with α = cos θ and c2 = −(α2 + β2) = −1. Let us remark that the central element Ω of H with the parameters given above is (3.9) Ω = (c2(r2 1 − r2 2) + r2 4)C − c2r5(2r2α + r5) . Proposition 3.2. The Heun algebra of su(2) type is not isomorphic to the Hahn algebra over R. This is seen by observing that relation (2.8) in Proposition 2.1 for the parameters µ and ν has necessarily complex solutions in this su(2) case, namely (3.10) ν = −2r2 exp(±iθ) , µ = −r3 − exp(±iθ)r4 . This is indicative of the fact that the dual Hahn polynomials cannot be obtained by tridi- agonalization of the Krawtchouk difference operator (see later). 3.2. su(2) Heun operators and Hamiltonians for generalized tops. We have already noted (see Remark 3) that the bispectral pair of operators X, Y can be chosen in various equivalent ways from the algebraic perspective. Different choices will however lead to modi- fied expressions for the corresponding algebraic Heun operator. To point out the connection between Heun operators of su(2) type and quantum Hamiltonians for tops, instead of (3.1), it will be convenient to rather adopt for X and Y (3.11) X = J3 + βJ1, Y = J3 − βJ1 , with β 6= 0 an arbitrary parameter. One easily convinces oneself that this choice is equivalent to the preceding one. Start from X and Y as in (3.1) and conjugate these operators by U(−θ/2) with U(θ) given in (3.4). Scaling by sec(θ/2) then gives the X and Y of (3.11). It is easy to see that these two operators also satisfy the relations (2.14)-(2.15) with (3.12) d1 = d2 = 0, a = b = (1 − β2) and c1 = c2 = −(1 + β2). One finds that the algebraic Heun operator (2.19) is in this case of the form with σ, mi, i = 1, 2, 3, 4, arbitrary parameters. The first term σ (J 2 1 ) is equivalent (up to an affine transformation) to the Hamiltonian of the quantum Euler top (see [17, 22]). The complete operator (3.13) corresponds to the Hamiltonian of a Euler top with additional "magnetic" interactions accounted for by the linear terms in (3.13). In classical mechanics this is the Zhukovsky-Volterra gyrostat [1], [16]. Note moreover that a similar Hamiltonian was exploited to describe spin systems with anisotropy [24]. We thus see that the Heun operator pencil on the su(2) algebra is equivalent to the Hamiltonian of a generalized Euler top (or quantum Zhukovskii-Volterra gyrostat). 3 − β2J 2 (3.13) W = σ(cid:0)J 2 3 − β2J 2 1(cid:1) + m1J1 + m2J2 + m3J3 + m4 HEUN ALGEBRAS OF LIE TYPE 7 4. The case of the Lie algebra su(1, 1) 4.1. Specializations. The situation where su(1, 1) is the underlying algebra will have, not surprisingly, close similarities with the picture in the su(2) case; the main differences will come from the richer orbit structure. Let us indeed explain how the choices for the bispectral pair (X, Y ) can be standardized. Throughout this section, the elements Jκ with κ = 1, 2, 3 obey the relations corresponding to the lower signs in (2.25) and (2.26). So as to relate to the unitary representations of su(1, 1) and to have one discrete spectrum in play, we shall again take Y to be the compact generator J3. Following a reasoning similar to the one given in the last section, we observe that a generic element l1J1 + l2J2 + l3J3 can be transformed into αJ3 + βJ1 under a rotation while not changing Y . The difference here is that the adjoint action of su(1, 1) preserves the non-definite form l2 = l2 3. This will hence require that β2 − α2 = l2. There will therefore be three distinct cases according to whether l2 is negative, positive or zero; elements in these classes are respectively said to be elliptic, hyperbolic and parabolic. 2 −l2 1 + l2 The upshot of this is that the bispectral operators X and Y in the su(1, 1) case will have exactly the same form as those of the su(2) case, namely (4.1) X = αJ3 + βJ1 , Y = J3 but will fall into the three categories where allowing for scaling, α and β will be parametrized as follows: (4.2) (4.3) (4.4) elliptic hyperbolic parabolic with 0 < φ < π. l2 < 0 l2 > 0 l2 = 0 α = cosh θ α = − cos φ α = 1 β = − sinh θ β = 1 β = 1 Owing to the fact that X and Y have the same general expressions in both the su(2) and su(1, 1) cases many formulas will be almost identical to those for su(2) as we determine the Heun algebras associated to su(1, 1). X and Y will again obey the relations (2.14) and (2.15) of the algebra A with the only change with respect to (3.2) being (4.5) c2 = β2 − α2. The Casimir operator C of A will be related exactly as in (3.3) to the Casimir element c of su(1, 1) given by the lower part of (2.26). We may point out that the relations (3.4) remain true for the elliptic case. Always keeping the definition (2.19) for W , Proposition 3.1 readily translates to the su(1, 1) case. Proposition 4.1. The elements X and W associated to su(1, 1) as per (4.1) satisfy the relations of the Heun algebra of Lie type with the coefficients xs and ys , s = 0, 1, 2, 3 as in (3.6) - (3.8) and with α given by (4.2), (4.3), (4.4) for each of the possible classes and c2 given by (4.5). The central element Ω of the corresponding Heun algebras is again given by (3.9) with the appropriate α and c2. 4.2. Connections to the Hahn algebra . Having found for su(1, 1), three Heun algebras corresponding to whether Y is of elliptic, hyperbolic or parabolic type, we now observe 8 N. CRAMP´E, L. VINET, AND A. ZHEDANOV that these algebras are isomorphic to the Hahn algebra. Indeed, the relation (2.8) for the parameters µ and ν has real solutions that read in light of (4.5): (4.6) ν = − 2r2 α ± β and µ = − r4 α ± β − r3. For the parabolic case α = β = 1, only the solution with the upper sign is permitted. The operators X and W therefore satisfy the Hahn algebra with the parameters (4.7) (4.8) x0 = (α2 − β2)r5 , y0 = ∓βr4r5 − 2r2r4C, x1 = ±βr4 , x2 = ±2βr2 , y1 = β2(r2 2 − r2 1) ∓ 4βr2r5 − 4r2 x3 = β2 − α2 , 2C . 5. The case of the harmonic oscillator algebra We shall focus in this section on the situation where the algebra A of Definition 2.2 is isomorphic to the harmonic oscillator algebra ho (2.27). Consider for the bispectral pair (X, Y ): (5.1) X = N + χ(A + A†) + χ2 I , Y = N , where χ is a real constant. Here again, it is straightforward to check that the relations (2.14) and (2.15) of the algebra A are satisfied with (5.2) b = 1 c1 = c2 = −1 d1 = d2 = χ2 and one readily observes that X, Y, Z and I also form a basis for ho. The Casimir C of A becomes: (5.3) C = 2χ2(X + Y ) + {X, Y } − X 2 − Y 2 + Z 2 = χ2[4k + χ2 − 2] with k = N − A†A. We now bring on the associated algebraic Heun operator W given in (2.19) as the generic bilinear expression in these X and Y . Anew, it is seen that the generator Y is obtained from X by an automorphism: (5.4) Y = U(χ)XU(χ)−1 with U(χ) = eχ(A−A†). Recalling from the argument given in Section 2 that the Heun algebras generated by the pairs (X, W ) and (Y, W ) are isomorphic, we shall only concentrate on the former with the following result directly obtained from Proposition 2.2. Proposition 5.1. The elements X and W associated to ho as per (5.1) and (2.19) satisfy the relations of the Heun algebra of Lie type with (5.5) (5.6) (5.7) (5.8) x0 = χ2r4 + r5 , y0 = χ2(2r2 y1 = 4χ2r2(2r4 − r3) − (r3 + r4)2 − 4r2r5 − 4r2 y2 = 12χ2r2 x1 = 2χ2r2 + r3 + r4 , 4 − r3r4 + 2r2r5) − (r3 + r4)r5 − 2r2r4C , 2 − 6r2(r3 + r4) , y3 = −8r2 1 − 2r2 2 − r2 x2 = 2r2 , 2C , 2 . x3 = −1 , With these parameters, the central element Ω of H is (5.9) Ω = (r2 4 + r2 2 − r2 1)C + χ4(r2 1 − r2 2) − 2χ2r4(r2 − r5) + 2r2r5 + r2 5 . In this case, looking at the relations (2.8) for µ and ν one finds that there are no solutions. The Heun algebra associated to the oscillator algebra is therefore not isomorphic to the Hahn algebra. HEUN ALGEBRAS OF LIE TYPE 9 6. Realizations in terms of difference operators It is well known that various families of orthogonal polynomials are related to the Lie algebras that we have considered so far [10, 5]. We shall review these results in this section by considering different models for su(2), su(1, 1) and ho and by showing that the associated operators Y are realized within the corresponding representations as the operators of which the polynomials are eigenfunctions. Since X will always be multiplication by the variable, this will allow us to say that the Heun algebras of Lie type that have been identified are the Heun algebras of Krawtchouk, Meixner, Meixner-Pollaczek, Laguerre and Charlier type. 6.1. su(2). There is a model of su(2) in terms of the difference operators T ± (6.1) T ±f (x) = f (x ± 1). It has the generators given as follows: J3 = sin2(θ/2) (x − N)T + + cos(θ) (x − N/2) − cos2(θ/2) xT − , J− = J+ = sin θ 2 (cid:0)(x − N)T + − 2x + N + xT −(cid:1) , 2 (cid:0)tan2(θ/2)(N − x)T + − 2x + N − cot2(θ/2)xT −(cid:1) , sin θ with J1 = 1 operators X and Y given by 2(J+ + J−) and J2 = − i 2(J+ − J−). As per Section 3.1, we have the bispectral (6.2) (6.3) (6.4) (6.6) (6.5) X = cos(θ)J3 − sin(θ)J1 = x − N 2 and Y = J3 . One readily recognizes in view of (6.2), that Y becomes the difference operator of the Krawtchouk polynomials Kn(x; sin2(θ/2), N) [15]: Y Kn(x; sin2(θ/2), N) =(cid:18)n − N 2(cid:19) Kn(x; sin2(θ/2), N) . The operators J± = J1 ± iJ2 are moreover lowering and raising operators for these polyno- mials: (6.7) (6.8) J−Kn(x; sin2(θ/2), N) = n cot(θ/2)Kn−1(x; sin2(θ/2), N) , J+Kn(x; sin2(θ/2), N) = (N − n) tan(θ/2)Kn+1(x; sin2(θ/2), N) . The associated operator W W = sin2( θ 2 )(N − x)(Nr2 + r1 − r2 − r4 − 2r2x)T + + 2r2 cos(θ)x2 + ρ4x + ρ5 (6.9) + cos2( θ 2 )x(Nr2 − r1 + r2 − r4 − 2r2x)T − is hence the Heun-Krawtchouk operator. The parameters ρ4 and ρ5 are given by (6.10) ρ4 = cos(θ)(r4 − 2Nr2) + r3 , ρ5 = 1 2 cos(θ)N(Nr2 − r4) − Nr3 2 + r5 . The algebra given in Proposition 3.1 can thus appropriately be called the Heun-Krawtchouk algebra since its generators can be realized using the bispectral operators of the Krawtchouk polynomials. 10 N. CRAMP´E, L. VINET, AND A. ZHEDANOV 6.2. su(1, 1). Three models of the positive discrete series representation will be presented in correspondance with the situations where Y is of elliptic, hyperbolic and parabolic type. The orthogonal polynomials of Meixner, Meixner-Pollaczek and Laguerre will arise corre- spondingly [18, 14]. Furthermore, owing to the fact that the associated Heun algebras are isomorphic to the Hahn one, it will be recorded that the respective Heun operators W can be transformed into the Hahn, Continuous Hahn and Jacobi operators. 6.2.1. Elliptic case. The generators of su(1, 1) can be realized as follows: (6.11) (6.12) J− = (6.13) J+ = J3 = −(x + κ) sinh2(θ/2)T + + (x + κ/2) cosh(θ) − cosh2(θ/2)xT − − sinh θ 2 − sinh θ 2 (cid:0)(x + κ)T + − 2x − κ + xT −(cid:1) (cid:0)tanh2(θ/2)(x + κ)T + − 2x − κ + coth2(θ/2)xT −(cid:1) . in terms of the shift operators (6.1). As per Section 4.1, we introduce the following bispectral operators (6.14) X = cosh(θ)J3 − sinh(θ)J1 = x + κ 2 and Y = J3 . Y is then identified as the difference operator of the Meixner polynomials Mn(x; κ, tanh2(θ/2)) [15]: (6.15) Y Mn(x; κ, tanh2(θ/2)) =(cid:16) κ 2 + n(cid:17) Mn(x; κ, tanh2(θ/2)) and the operators J+ and J− are ladder operators for these polynomials: J−Mn(x; κ, tanh2(θ/2)) = n coth(θ/2)Mn−1(x; κ, tanh2(θ/2)) J+Mn(x; κ, tanh2(θ/2)) = (κ + n) tanh(θ/2)Mn+1(x; κ, tanh2(θ/2)) . (6.16) (6.17) The associated operator W W = −(x + κ)(2r2x + r2κ − r1 + r2 + r4) sinh2(θ/2)T + + 2r2 cosh(θ)x2 + ρ4x + ρ5 (6.18) +x(r2x + r2κ + r1 − r2 + r4) cosh2(θ/2)T − is hence the Heun-Meixner operator. We have introduced (6.19) ρ4 = cosh(θ)(2r2κ + r4) + r3 , ρ5 = cosh(θ)κ 2 (r2κ + r4) + r3κ 2 + r5 . The algebra given in Proposition 4.1 may thus be called the Heun-Meixner algebra since it can be realized using the bispectral operators of the Meixner polynomials. We know that this Heun-Meixner algebra is isomorphic to the Hahn one. This can be underscored within the present su(1, 1) representation by recovering the Hahn operator from W recalling 1 tanhx(θ/2) W tanhx(θ/2) which reads as follows the observations made in Section 4.2. Consider to simplify, the conjugated operator fW = fW = W tanhx(θ/2) − (r4e−θ + r3)X − 2r2e−θX 2 tanhx(θ/2) r2κ + r1 − r2 + r4 r2κ − r1 + r2 + r4 1 )T − + (κ + x)(x + )T + r2 = −r2 sinh(θ)hx(x + (6.20) −(2x + κ)(x + 2r2 r2κ + r4 2r2 )i + r5 HEUN ALGEBRAS OF LIE TYPE 11 We readily recognize the difference operator of the Hahn polynomials in the square bracket of the second line of (6.20). 6.2.2. Hyperbolic case. The operators (6.21) J3 = (6.22) J± = 1 2i sin φ(cid:0)eiφ(λ − ix)T + + 2ix cos(φ) − e−iφ(λ + ix)T −(cid:1) 2i sin φ(cid:0)e±iφ(λ − ix)T + + 2ix − e∓iφ(λ + ix)T −(cid:1) eiφ satisfy the su(1, 1) commutation relations with the shift operators T ± defined by T ±f (x) = f (x ± i). With X and Y given according to Section 4.1 by (6.23) X = − cos(φ)J3 + J1 = x sin(φ) , Y = J3, we see that Y is realized in this instance as the difference operator of the Meixner-Pollaczek polynomials P (λ) n (x; φ) [15]: (6.24) n (x; φ) = (n + λ)P (λ) while J+ and J− act as raising and lowering operators: Y P (λ) n (x; φ) , (6.25) J+P (λ) n (x; φ) = (n + 1)P (λ) n+1(x; φ) , J−P (λ) n (x; φ) = (2λ + n − 1)P (λ) n−1(x; φ) . The associated operator W W = (ix − λ)(cid:18)ir2x + r1 − r2 2 ir4 2 sin(φ)(cid:19) T + + 2r2 cos(φ)x2 + ρ4x + r5 + (6.26) +e−iφ(λ + ix)(cid:18)ir2x − r1 − r2 2 + ir4 2 sin(φ)(cid:19) T − is hence the Heun-Meixner-Pollaczek operator. Here ρ4 = sin(φ)r3+cot(φ)r4. In this case the algebra of Proposition 4.1 is really the Heun-Meixner-Pollaczek algebra. In this realization, upon scaling and conjugating W according to fW = 1 r2 ir4 fW = (λ − ix)(cid:18) r2 − r1 +(λ + ix)(cid:18) r2 − r1 2r2 2r2 − + 2r2 sin(φ) ir4 2r2 sin(φ) − ix(cid:19) T + − 2x2 − + ix(cid:19) T −. (6.27) eiφxW e−iφx one finds xr4 r2 sin(φ) + r5 r2 This operator is the difference operator that is diagonalized by the continuous Hahn polyno- mials (−1)ixpn(x; λ, r2−r1 2r2 sin(φ) ) with eigenvalues −n2 − (2n + 1)λ + 2r2 (λ+n)r1+r5 2r2 sin(φ) , λ, r2−r1 − ir4 + ir4 2r2 . r2 6.2.3. Parabolic case. A model of su(1, 1) in terms of differential operator is given by (6.28) (6.29) (6.30) d dx + 1 + a 2 J3 = −x J− = x J+ = x d2 dx2 − (1 + a − x) d2 dx2 + (1 + a) d2 dx2 + (1 + a − 2x) d dx d dx + x − 1 − a 12 N. CRAMP´E, L. VINET, AND A. ZHEDANOV where a is a free parameter associated to this realization. For the parabolic case, the bispectal operators X and Y were taken to be (6.31) X = J3 + J1 = 1 2 x and Y = J3 in Section 4.1. It follows that Y can here be identified with the difference operator of the Laguerre polynomials L(a) n (x) [15]: Y L(a) n (x) = (n + (a + 1)/2)L(a) n (x) . (6.32) One also gets (6.33) J+L(a) n (x) = −(n + 1)L(a) n+1(x) , J−L(a) n (x) = −(a + n)L(a) n−1(x) . The associated operator W (6.34) W = −x(r2x + r4) d2 dx2 + (r2x2 + (r1 − 2r2 − r2a + r4)x − r4(1 + a)) d dx + ρ4x + ρ5 is hence the Heun-Laguerre operator. We have defined (6.35) ρ4 = 1 2 (r3 − r1 + (a + 2)r2) and ρ5 = r5 + 1 2 (r1 − r2 + r4)(a + 1) . In this case, the algebra given in Proposition 4.1 should hence be referred to as the Heun- Laguerre algebra since we have a realization of it based on the bispectral operators of the Laguerre polynomials. It can be also showed that the eigenvalue equation W f (x) = λf (x) becomes the confluent Heun equation [13], [19]. Let us remark that this connection between the confluent Heun equation and the Heun operators associated to the Laguerre differential equation had already been pointed out in [11]. Finally note that the conjugated operator fW = e−x/2W ex/2 fW = e−x/2W ex/2 − (r4/2 + r3)X − r2X 2 = −x(r2x + r4) (6.36) d2 dx2 + ((r1 − 2r2a − r2)x − r4(1 + a)) d dx + 1 2 (r1 − r2)(a + 1) + r5 is recognized to be the Jacobi differential operator which together with X generates the Hahn algebra. with eigenvalues (r1 − r2)(n + 1 Indeed, fW is diagonalized by the Jacobi polynomial P (a,−r1/r2) 2 (a + 1)) − r2n(n + a) + r5. (cid:16) 2r2x + 1(cid:17) r4 n 6.3. Harmonic oscillator algebra ho. The oscillator algebra can also be realized as follows in terms of the shift operators (6.1): (6.37) N = −xT − + x + χ2 − χ2T + , A† = −χ + 1 χ xT − , A = χ(T + − 1) . As in Section 5, the bispectral pair (X, Y ) is X = N + χ(A + A†) + χ2 = x , (6.38) The operator Y is the difference operator of the Charlier polynomial Cn(x, χ2) [15]: (6.39) Y Cn(x, χ2) = nCn(x, χ2) , Y = N . and A† and A are their raising and lowering operators: (6.40) A†Cn(x, χ2) = −χCn+1(x, χ2) , ACn(x, χ2) = − n χ Cn−1(x, χ2) . HEUN ALGEBRAS OF LIE TYPE 13 The associated operator W (6.41) W = x(r2 −r4 −r1 −2r2x)T − +χ2(r1 −r2 −r4 −2r2x)T + +2r2x2 +(2r2χ2 +r3 +r4)x+r4χ2 +r5 is hence the Heun-Charlier operator. The algebra given in Proposition 5.1 should therefore be called the Heun-Charlier algebra since it is realized with X = x and W constructed from x and the Charlier operator. 7. Concluding remarks The results presented here provide a comprehensive picture of the Heun operators and algebras associated to orthogonal polynomials that admit a Lie theoretical interpretation. This complements the previous studies of the Jacobi and Hahn polynomials from this Heun angle which led respectively to the description of the standard Heun differential operator and its discrete version within the framework of bispectral problems. Missing is the parallel treatment of the Racah polynomials especially since the q → 1 limit of the Askey-Wilson case is known to be delicate. It would also be quite pertinent to carry on with the q-Askey tableau and to work out the quantum algebraic analog of the study offered in the present article. It is known that Heun operators have applications in time and band limiting problems, it would be of interest to look at applications in this direction in particular. We hope to report on these questions in the near future. Acknowledgments: N.C. is gratefully holding a CRM -- Simons professorship. The re- search of L.V. is supported in part by a Natural Science and Engineering Council (NSERC) of Canada discovery grant and that of A.Z. by the National Science Foundation of China (Grant No. 11711015). References [1] I. Basak, Explicit solution of the Zhukovski-Volterra gyrostat, Regular and Chaotic Dynamics, 14, 223 -- 236. [2] P. Baseilhac, L. Vinet and A. Zhedanov, The q-Heun operator of big q-Jacobi type and the q-Heun algebra, arXiv:1808.06695. [3] P. Baseilhac, S. Tsujimoto, L. Vinet and A. Zhedanov, The Heun-Askey-Wilson algebra and the Heun operator of Askey-Wilson type, arXiv:1811.11407. [4] N. Crampe, E. Ragoucy, L. Vinet, and A. S. Zhedanov, Truncation of the reflection algebra and the Hahn algebra, arXiv:1903.05674 [5] R. Floreanini, J. LeTourneux, L. Vinet, Quantum Mechanics and Polynomials of a Discrete Variable Ann. Phys. 226 [1993], 331-349 [6] L. Frappat, J. Gaboriaud, L. Vinet, S. Vinet, and A.S. Zhedanov, The Higgs and Hahn algebras from a Howe duality perspective, arXiv:1811.09359. [7] V. X. Genest, L. Vinet, and A. S. Zhedanov, The Racah algebra and superintegrable models, J. Phys.: Conf. Ser. 512 (2014) 012011 and arXiv:1309.3540. [8] Ya. A. Granovskii, I. M. Lutzenko, and A .S. Zhedanov, Mutual integrability, quadratic algebras, and dynamical symmetry, Ann. Phys. 217 (1992) 1. [9] Ya. A. Granovskii, and A. S. Zhedanov, Nature of the symmetry group of the 6j-symbol, Zh. Eksp. Teor. Fiz. 94 (1988) 49. [10] Y. I. Granovskii, A. S. Zhedanov, Orthogonal Polynomials in the Lie Algebras Sov. Phys. J. 29, (1986) 387 [11] F. A. Grunbaum, L. Vinet, and A.S. Zhedanov, Tridiagonalization and the Heun equation, J. Math. Phys. 58 (2017) 031703. 14 N. CRAMP´E, L. VINET, AND A. ZHEDANOV [12] F. A. Grunbaum, L. Vinet and A. Zhedanov, Algebraic Heun operator and band-time limiting, Comm.Math.Phys.(to appear), arXiv:1711.07862. [13] G. Kristensson, Second Order Differential Equations, (2010), Springer, New York, NY [14] H. T. Koelink, J. Van Der Jeugt, Convolutions for Orthogonal Polynomials from Lie and Quantum Algebra Representations, SIAM J. Math. Anal. 29 (1998) 794-822 [15] R. Koekoek, P.A. Lesky, and R.F. Swarttouw, Hypergeometric orthogonal polynomials and their q- analogues, Springer, 1-st edition, 2010. [16] A.M. Levin, A. Olshanetsky, A.V. Zotov, Painlev´e VI, Rigid Tops and Reflection Equation, Comm. Math. Phys., 2006, 268, Issue 1, 67 -- 103. [17] J. Patera, P. Winternitz, A new basis for the representations of the rotation group. Lam´e and Heun polynomials, J.Math.Phys. 14, (1973) 1130 -- 1139. [18] D. R. Masson, J. Repka, Spectral Theory of Jacobi Matrices in l2(Z) and the su(1, 1) Lie Algebra, SIAM J. Math. Anal. 22 (1991), 1131-1146. [19] A. Ronveaux (Ed.), Heun's Differential Equations, Oxford University Press, Oxford, 1995. [20] R. F. Streater, The representations of the oscillator group, Commun.Math. Phys. (1967) 4: 217. [21] K. Takemura On q-deformations of Heun equation, arXiv:1712.09564. [22] A. V. Turbiner, The Heun operator as a Hamiltonian J.Phys. A:Math. Theor. 49 (2016), 26LT01 [23] L. Vinet and A. Zhedanov, The Heun operator of Hahn type, arXiv:1808.00153. [24] O. B. Zaslavskii and V. V. Ul'yanov, Periodic effective potentials for spin systems and new exact solutions of the one-dimensional Schrodinger equation for the energy bands, Theor.Math.Phys., 1987, 71, 520 -- 528. [25] A. S. Zhedanov, Hidden symmetry of the Askey-Wilson polynomials, Theoret. Math. Phys. 89 (1991) 1146. † Institut Denis-Poisson CNRS/UMR 7013 - Universit´e de Tours - Universit´e d'Orl´eans; Parc de Grammont, 37200 Tours, France. E-mail address: [email protected] ∗ Centre de recherches, math´ematiques, Universit´e de Montr´eal; P.O. Box 6128, Centre- ville Station; Montr´eal (Qu´ebec), H3C 3J7, Canada. E-mail address: [email protected] ‡ Department of Mathematics, School of Information; Renmin University of China; Bei- jing 100872, China. E-mail address: [email protected]
1712.00508	2	1712	2019-06-05T06:01:50	Normal basises of algebras and Exponential Diophantine equations in rings of positive characteristic	[ "math.RA" ]	In this paper we discourse basises of representable algebras. This question lead to arithmetic problems. We prove algorithmical solvability of exponential-Diophantine equations in rings represented by matrices over fields of positive characteristic. Consider the system of exponential-Diophantine equations $$ \sum\limits_{i=1}^s P_{ij}(n_1,\dots,n_t) b_{ij0} a_{ij1}^{n_1} b_{ij1} \dots a_{ijt}^{n_t}b_{ijt}=0 $$ where $b_{ijk},a_{ijk}$ are constants from matrix ring of characteristic $p$, $n_i$ are indeterminates. For any solution $(n_1,\dots,n_t)$ of the system we construct a word (over an alphabet containing $p^t$ symbols) ${\overline \alpha_0},\dots,{\overline \alpha_q}$ where ${\overline \alpha_i}$ is a $t$-tuple $\langle n_1^{(i)},\dots,n_t^{(i)}\rangle$, $n^{(i)}$ is the $i$-th digit in the $p$-adic representation of $n$. The main result of this paper is as follows: the set of words corresponding in this sense to solutions of a system of exponential-Diophantine equations is a regular language (i.e. recognizable by a finite automaton). There exists an effective algorithm which calculates this language. This algorithm is constructed in the paper.	math.RA	math	Normal basises of algebras and Exponential Diophantine equations in rings of positive characteristic A. A. Chilikov∗, A. Ya. Belov† UDC 512.5+511 Keywords: finite automata, regular languages, P I-algebra, Shirshov theorem on height, word combinatorics, n-divisibility, Burnside-type problems. Abstract In this paper we discourse basises of representable algebras. This question lead to arithmetic problems. We prove algorithmical solvability of exponential-Diophantine equations in rings represented by matrices over fields of positive characteristic. Con- sider the system of exponential-Diophantine equations s Xi=1 Pij(n1, . . . , nt)bij0an1 ij1bij1 . . . ant ijtbijt = 0 where bijk, aijk are constants from matrix ring of characteristic p, ni are indeter- minates. For any solution (n1, . . . , nt) of the system we construct a word (over an (i) t i, n(i) alphabet containing pt symbols) α0, . . . , αq where αi is a t-tuple hn is the i-th digit in the p-adic representation of n. The main result of this paper is as follows: the set of words corresponding in this sense to solutions of a system of exponential-Diophantine equations is a regular language (i.e. recognizable by a fi- nite automaton). There exists an effective algorithm which calculates this language. This algorithm is constructed in the paper. (i) 1 , . . . , n The research was supported by Russian Science Foundation, Grant No 17-11-01377. 1 Introduction Systems of exponential Diophantine equations (EDE) s Xi=1 Pij(n1, . . . , nt)cni ij1 · · · cnt ijt = 0, (1) ∗MIPT, BIU †College of Math. and Stat., Shenzhen University 1 where Pij are some polynomials arise in various areas of moderm mathematics, and in gen- eral case, as J. Robinson has shown, they are algorithmically undecidable. Yu. V. Matiya- sevich has proved algorithmical undecidability for purely Diophantine equations P (n1, . . . , nt) = 0. A number of problems reduces to undecidability of some EDE. However it turns out that if ci belong to a field of positive characteristic (and even to a matrix ring), the problem of finding the set of solutions is algorithmically decidable. Questions arising in this context occur to be related with formal languages. Investigation of bases of algebras is an inspiration for research of such equations. Suppose a1 ≺ . . . ≺ at is an ordered set of generators for an algebra A. The order ≺ on this set induces lexicographical ordering on the set of words in {ai}. A basis M of A as of a vector space is called normal if it is generated by non-decreasable (that is, not representable by a linear combination of lesser words) elements. If A is a P I-algebra (in particular, if A is representable) then due to Shirshov theorem on height there exist h = ht(A) and a finite tuple v1, . . . , vs such that M consists of elements of the form vk1 i1 · · · vkt it where t ≤ h. In this connection the question arises on the structure of the set consisting of degree vectors hk1, . . . , kti, in particular for the representable case. If the algebra A is repre- sentable and monomial (that is, defining relations are of the form uj = 0 where uj are some words) then the problem of normal basis permits in some sense complete answer. We have the following Theorem (test for representability of a monomial algebra). A monomial algebra A is representable iff A has bounded height over some finite set of words v1, . . . , vt, the set of defining relations can be divided into a finite number of series vk1 1 · · · vkt t = 0 where Pij(k1, . . . , kt)ck1 ij1 · · · ckt ijt = 0 Xi and each series corresponds to a specific system of EDE. This theorem implies, in particular, existence of representable algebras whose Hilbert series is transcendental as well as algorithmic undecidability of isomorphism problem for a pair of subalgebras in a matrix algebra over a polynomial ring. Nevertheless for positive characteristic the situation is much simpler. Although Dio- phantine problems arise more often in this case, their solution is simpler. The set of solutions of an EDE admits effective description in terms of p-adic decomposition of inde- terminates n1, . . . , nt. Since values of Pij(n1, . . . , nt) are periodical with period p in each ni, it suffices to investigate equations of the form s Xi=1 cn1 1 · · · cnt t = 0. Consider some solution of an EDE: hn1, . . . , nti. To each its component ni attach its i . Thus to each solution we attach the sequence of tuples of t i. Interprete these tuples as letters, and our sequences p-adic decomposition nk figures hnk t i,. . . , hn0 1, . . . , n0 1, . . . , nk i . . . n0 2 as words over the alphabet consisting of tuples. The set of all words corresponding to solutions of EDE forms a language over a finite alphabet. We are ready to formulate the main result of this paper. Theorem 1. The set of words corresponding to a system of EDE is a regular language. In other words, there exists an oriented graph with arrows marked by letters corresponding to finite tuples of figures (the number of letters is pt). Some vertex is declared initial, and some other vertices are declared final. There exists 1-1 correspondence between solutions of our system and words which may be read along a path consisting of arrows and going from the initial vertex to a final one. These paths are allowed to have arbitrary length, and each vertex (including initial and final ones) may be passed arbitrarily many times. There exists an effective algorithm for constructing such a graph. Below we present its description. 2 Bases of Representable and P I-algebras The earliest purely combinatorial result of this kind occurred to be A. I. Shirshov height theorem. Let A be a finitely generated PI-algebra. Then there exists a finite set of elements Y and an integer H ∈ N such that A is linearly represented by (that is, is generated by linear combinations of ) the set of elements of the form vk1 1 vk2 2 . . . vkh h where h ≤ H, vi ∈ Y . For Y we may take the set of words of degree ≤ m. Such an Y is called a Shirshov basis of the algebra A. The above theorem implies positive solution of Kurosh problem and of other Burnside- type problems for PI-rings. In fact, if Y is a Shirshov basis consisting of algebraic elements then the algebra A is finite-dimensional. Thus Shirshov theorem explicitly determines the set of elements whose algebraicity implies algebraicity of the whole algebra. We also have Corollary 2.1 If A is a PI-algebra of degree m and all words in its generators of degree ≤ m are algebraic then A is locally finite. Height theorem also implies Corollary 2.2 (Berele) Let A be a finitely generated PI-algebra. Then GKdim(A) < ∞. GKdim(A) is the Gelfand -- Kirillov dimension of the algebra A, that is, GKdim(A) = lim n→∞ ln VA(n)/ ln(n) where VA(n) is the growth function of A, that is, the dimension of the vector space gen- erated by words of degree ≤ n in generators of A. To prove the corollary, it suffices to observe that the number of solutions of inequality k1v1 + · · · + khvh ≤ n with h ≤ H does not exceed N H , and so GKdim(A) ≤ h(A). Thus we obtain various consequences from Height theorem. A little later we discuss questions concerning conversion of these implications. To begin with, we introduce some notions and notation. 3 The number m = deg(A) will denote the degree of the algebra, that is, the minimal degree of an inequality satisfied by it; n = PIdeg(A) is the complexity of A, that is, the maximal k such that Mk, the algebra of matrices of size k, belongs to the variety Var(A) generated by A. It is convenient to replace the notion of height by a close notion of essential height. Definition 2.3 An algebra A has essential height h over a finite set Y which is called an s-basis if there exists a finite set D ⊂ A such that A is linearly representable by elements of the form t1 · . . . · tl where l ≤ 2h + 1 and ∀i(ti ∈ D ∨ ti = yki i ; yi ∈ Y ), and the set of i having ti 6∈ D contains ≤ h elements. Loosely speaking, each long word is a product of periodical parts and of "layers" having bounded length. Essential height is the number of these periodical pieces, and ordinary height depends also upon "layers". Height theorem gives rise to following questions: 1. To what classes of rings Height theorem may be extended? 2. For which Y the algebra A has bounded height? Since now, we consider the associative case. 3. How to evaluate height? 4. What does the degree vector (k1, . . . , kh) look like? First of all, which sets of its components are essential, that is, which sets of ki can be simultaneously unbounded? What is the value of essential height? 5. A question regarding finer structure of the set of degree vectors: does it have any regularity properties? At last, the range of questions forming the subject of this paper. 6. Which sets of words can be chosen for {vi}? Now we proceed to discuss the above questions. Non-associative generalizations. Height theorem has been extended to certain classes of rings close to associative rings. S. V. Pchelintsev has proved it for alternative and (−1, 1) cases, S. P. Mishchenko has obtained an analogue of Height theorem for Lie algebras with a sparse identity. The author has proved Height theorem for a certain class of rings asymptotically close to associative rings and in particular including alternative and Jordan PI-algebras. Shirshov bases. Let A be a PI-algebra, and suppose a subset M ⊆ A is its s-basis. Then if all elements of M are algebraic over K then A is finite-dimensional (Kurosh problem). Boundedness of essential height over Y implies "positive solution of Kurosh problem over Y ". The converse is much less trivial. Theorem 2.4 (A. Ya. Belov) Suppose A is a graded PI-algebra, Y is a finite set of homogeneous elements. Then if the algebra A/Y (n) is nilpotent for each n then Y is an s-basis for A. If in this situation Y generates A as an algebra then Y is a Shirshov basis for A. 4 (Y (n) denotes the ideal generated by nth powers of elements from Y .) The following example demonstrates that the straightforward converse of Kurosh prob- lem for non-graded case does not have positive solution. Suppose A = Q[x, 1/x]. Each projection π such that π(x) is algebraic has finite-dimensional image. Nevertheless the set {x} is not an s-basis for the algebra Q[x, 1/x]. Thus the definition of Kurosh set is chosen as follows: Definition 2.5 A set M ⊂ A is called a Kurosh set if each projection π : A ⊗ K[X] → A′ having image π(M) integral over π(K[X]) is finite-dimensional over π(K[X]). We proceed to formulate a generalization of this theorem for non-homogeneous case. Theorem 2.6 (A. Ya. Belov) Let A be a PI-algebra, M ⊆ A a Kurosh subset in A. Then M is an s-basis for A. The following proposition shows that Theorem 2.6 is a generalization of Theorem 2.4: Proposition 2.7 Let A be a graded algebra, Y a set of homogeneous elements. Then if the algebra A/Y (n) is locally nilpotent for all n then Y is a Kurosh set. Thus boundedness of essential height is a non-commutative generalization of integrity. Remarks. a) Note that in the case of Lie PI-algebras, Kurosh problem has positive solution but Height theorem fails. b) The theorem extends to some class of rings asymptotically close to associative rings (with bounded l-length, finitely generated algebra of left multiplications, and associative powers). Estimates of height. The original A. I. Shirshov's proof was purely combinatorial (based on elimination technique developed by him for Lie algebras, in particular in the proof of Freedom theorem), however it did not provide any reasonable estimates for height. Later A. T. Kolotov obtained an estimate for ht(A) ≤ ssm (m = deg(A), s is the number of generators). Subsequently, E. I. Zel'manov [5] raised the question on existing of an exponential estimate which was obtained later on by the Belov. Shirshov height theorem. Suppose A is an l-generated PI-algebra of degree m. Then the height of A over the set of words having degree ≤ m is bounded by a function H(m, l) where H(m, l) < 2mlm+1. Essential height. Clearly, essential heght is an estimate for Gelfand -- Kirillov di- mension and an s-basis is a Shirshov basis iff it generates A as an algebra. In the representable case the converse is true. Theorem 2.8 (A. Ya. Belov [4]) Suppose A is a finitely generated representable alge- bra and HEssY (A) < ∞. Then HEssY (A) = GKdim(A). Corollary 2.9 (V. T. Markov) The Gelfand -- Kirillov dimension of a finitely gener- ated representable algebra is an integer. 5 Corollary 2.10 If HEssY (A) < ∞ amd an algebra A is representable then HEssY (A) is independent of the s-basis Y . Due to local representability of relatively free algebras, the Gelfand -- Kirillov dimen- sion in this case also equals the essential height. Structure of degree vectors. Thus in the representable case both Gelfand -- Kirillov dimension and essential height behave well. Nevertheless even in this case the set of degree vectors can have bad structure, namely, it can be the complement for the set of solutions for some system of exponential-polynomial Diophantine equations. Consequently, there exists an example of a representable algebra having transcendent Hilbert series. However in the case of relatively free algebra the Hilbert series is rational. Shirshov bases consisting of words. Their description is given by the following theorem: Theorem 2.11 (A. Ya. Belov) A set of words Y is a Shirshov basis of an algebra A iff for each word u having length ≤ m = PIdeg(A), the complexity of A, the set Y contains some word which is cyclically conjugate to some degree of u. A. I. Shirshov himself has shown that for a Shirshov basis we may take the set of words having degree at most deg(A). I. V. Lvov has proved boundedness of height over the set of words having length at most deg(A) − 1. S. Amitsur and I. P. Shestakov conjectured that if all words having length not exceeding the complexity PIdeg(A) are algebraic then the algebra is finite-dimensional. I. V. Lvov reduced this statement to the following: Theorem 2.12 Let A be a finite-dimensional subalgebra in the matrix algebra of order n, and let a1, . . . , as be its generators. Then if all words in a1, . . . , as having degree ≤ n are nilpotent then A itself is nilpotent. Note that n is the precise estimate. Shestakov's conjecture was proved by V. A. Ufnarovsky and by G. P. Chekanu. 1. Later the author [6] showed that for {vi}, we may take the set of words from Shestakov's conjecture. This result also was announced by G. P. Chekanu. Later on, another proof of this fact was obtained by V. Drensky. In the sequel, we focus on the range of problems concerned to relations between Height theorem and Independence theorem. Independence theorem may be formulated, in particular, as follows Theorem 2.13 (Independence theorem) Suppose the following is true: 1. a word W = ai1 . . . ain is the minimal word in the left lexicographical ordering on the set of all nonzero products having length ≤ n; 2. the extreme parts of W are nilpotent. Then initial subwords of W are linearly independent. 1From a private letter by the latter: "we worked in the same area . . . We both have stood this friendly and creative concurrence (we did begin this deliberately, with agreement that we work in different languages)". The proofs were based on "the spirit of independence". Subsequent papers of these authors contained various specifications and generalizations of these theorems [10]. 6 To deduce I. P. Shestakov's conjecture (or, equivalently, I. V. L'vov's statement) from this theorem, it suffices to consider a faithful representation of A by operators on n- dimensional space V . Let v1, . . . , vn be a basis of this space, then for some vi we have miW 6= 0. Consider the auxiliary algebra generated by V and A. Suppose V ·V = A·V = 0 and the action of V A coincides with module multiplication. Reorder the generators as follows: v1 ≻ · · · ≻ vn ≻ a1 ≻ · · · ≻ as and apply Independence theorem. ✷ Original proofs of Independence theorem were rather complicated. Application of symbolic dynamics technique involving infinite words or superwords allowed to clarify them. Technique of superwords occurred to be rather close to the lines of structure theory. Its role does not reduce to proving statements like Independence theorem. Using superwords allows to prove Height theorem, nilpotence of the Lie algebra generated by sandwiches [8], coincidence of nilradical and Jacobson radical in monomial algebras, to describe bases of algebras with extremal growth function VA(n) = n(n+3) , and also to describe weakly Noetherian, semisimple and semiprimary monomial algebras [4] and to obtain some other combinatorial results in the theories of semigroups and rings. 2 Many properties of algebras are defined by monomial relations. For example, such are the conditions of Shestakov's conjecture, namely, nilpotence of words whose degree does not exceed complexity. This conjecture is related to the structure of the matrix algebra. Multiplication of ma- trix units Eij is almost monomial, and the language of representations of matrix algebras clarifies "matrix" properties of semisimple components. It is no coincidence that many au- thors dealing with independence actively used a similar technique of matrix constructions for other problems concerning local finiteness [11]. 3 Preliminries Recall now some facts from the theory of formal languages. Definitions. A finite automaton is a finite oriented graph some vertex of which is declared initial, some vertices are declared final, and each edge is marked by a symbol of some finite alphabet. A regular language is a set of words which may be read at edges of some finite au- tomaton along a path from the initial vertex to a final one. We say that this automaton represents the given language. A concatenation vu of two words u and v is obtained by adding v to u (in our case from the left). A concatenation of two languages L1 and L2 is the language L = {uv u ∈ L1, v ∈ L2}. The closure L∗ of a language L is the set of all powers of words from L. An atomary language consists of a single word consisting of a single letter. One of the simplest instances of regular languages is the set of all words not including subwords from a fixed finite list. A description of regular languages in terms of operations over languages is provided by the following Theorem (Cleenee). A language is regular iff it can be obtained from atomary languages by finite number of operations of joint, meet, comcatenation and closure. For more detail and for the proof of Cleenee's theorem see Salomaa [1, p. 24-37]. 7 4 Basic notation and constructions 1, Σ∗ 2 are written from the right to the left. 1 is the set of all finite words over Σ1. 2 is the set of all finite words over Σ2. In the sequel, we use following notation. ϑ = (ϑ1, . . . , ϑr) is a tuple of variables. n − (n1, . . . , nt) is a tuple of indeterminates. N is the set of natural numbers. Σ1 is the alphabet consisting of tuples of figures 0, 1, . . . , p − 1 having length t. Σ2 is the alphabet consisting of tuples of figures 0, 1, . . . , p − 1 having length r. Σ∗ Σ∗ Words from Σ∗ λ is the empty word. l(u) is the length of the word u. R = Zp[ϑ1, . . . , ϑr] is the ring of polynomials over Zp. F is the quotient field for R. A is the algebraic closure for F. Since to each sequence of figures with radix p there corresponds a number from N, the following maps are well-defined: φ : Σ∗ with radix p, ψ : Σ∗ with radix p. φ(i) : Σ∗ ψ(i) : Σ∗ f = (f1, . . . , ft) is a tuple of polynomials. In the sequel, we also denote by f p and we denote by ϑ Products of the form f φ(1)(u) ϑ 1 → N is ith component of φ. 2 → N is ith component of ψ. the tuple consisting of pth powers of ϑi: (ϑp the tuple consisting of pth powers of fi: (f p 2 to a tuple consisting of r numbers written 1 to a tuple consisting of t numbers written 1 → N t which maps any word from Σ∗ 2 → N r which maps any word from Σ∗ and ϑψ(1)(v) · · · ϑψ(r)(v) r 1 , . . . , f p t ), will be denoted f and φ(u) ψ(v) accordingly. 1, . . . , ϑp r). p · · · f φ(t)(u) t 1 1 5 Equations over a ring of polynomials Let K be a matrix ring having positive characteristic p. We prove now an auxiliary statement which allows to reduce the class of considered equations. Proposition. If for each equation of the form s Xi=1 bi0an1 i1 bi1 · · · ant it bit = 0 (2) having coefficients from K the set of words corresponding to its solutions (as it was defined in Introduction) is a regular language then the set of words corresponding to solutions of any system of equations of the form s Xi=1 Pij(n1, . . . , nt)bij0an1 ij1bij1 · · · ant ijtbijt = 0 (3) 8 over K is a regular language. Proof. First note that the set of solutions of a system is the meet of sets of solutions for equations of the system. So by virtue of Cleenee's theorem, regularity of languages corresponding to single equations of the form (3) implies regularity of languages corre- sponding to systems of such equations. Consider now an equation of the form (3). Let hn1, . . . , nti be a tuple of numbers 1, . . . , n′ ti i is the last digit with radix p in the number ni. For fixed hn′ i where n0 i +pn′ ni = n0 we have ′ P (n ′ 1, . . . , n t) = P (n1, . . . , nt), so hn1, . . . , nti is a solution of an equation of the form (3) iff hn′ ti is a solution of an equation of the form (2). Regularity of the set of all hn′ ti obviously implies regularity of the set of all hn′ ti since the corresponding words are obtained by adding the tuple hn0 t i. Finally, the complete set of solutions of the original equation is the joint of sets of solutions corresponding to distinct tuples hn0 t i. Again by Cleene's theorem we have regularity of languages corresponding to any equations of the form (3) over K. The proof is complete. 1, . . . , n0 1, . . . , n0 1, . . . , n′ 1, . . . , n′ 1, . . . , n′ Thus we have reduced investigation of solutions for some system of EDE to the case of a single equation which furthermore has no polynomial (in n) parts. Consider an EDE over R: s Xi=1 Qi(ϑ)[P n1 i1 ](ϑ) · · · [P nt it ](ϑ) = 0. (4) Its solution is a tuple of numbers n = (n1, . . . , nt), ni ∈ N. Definition. A word-solution of the EDE (4) is u ∈ Σ∗ 1 such that φ(u) = n where n is a solution for (4). Now we may write the equation in u s Xi=1 Qi(ϑ)[Pi φ(u) ](ϑ) = 0. (5) In the sequel, word-solutions will also be called solutions simply. The main result of this Section may be stated in the form of the following Theorem 1. L ⊂ Σ∗ 1, the set of word-solutions for the equation (5), is a regular language. Before presenting the proof, we describe its basic idea. Let Q(x) be a polynomial having coefficients from Zp. Let us investigate the result of removing brackets in Qn(x). Write n with radix p: n = n0 + n1p + . . . + nkpk + nk+1pk+1 + . . . + nsps. Put Qk(x) = Qk where k = 0, 1, . . . , p − 1. Clearly Q0 = 1, Q1 = Q. Then Q(x)n = Qn0(x)Qn1(xp) · · · Qnk (xpk )Qnk+1(xpk+1 ) · · · Qns(xps ) (∗) since Q(x)pk = Q(xpk ). Consider a section of the product Rk = Qn0(x) · · · Qnk(xpk ). (∗∗) 9 Collect terms with the same remainder α of the degree of x modulo pk+1. In other words, represent the section as sum pk+1−1 Xα=0 xαRα(xpk+1 ). (∗ ∗ ∗) Now note the following. 1. Multiplication by the rest of the product (∗) (that is, by Qnk+1(xpk+1) · · · Qns(xps)) does not lead to cancellation of terms in (∗ ∗ ∗) having distinct α. 2. The degree of Rk does not exceed (deg Q)(p − 1)pk. Hence the degree of Rα does not exceed (deg Q)(p − 1). Furthermore, since we work over a finite field, the number of distinct types of Rα (small types) is bounded (and does not exceed p(deg Q)(p−1)). 3. Due to observation 1, we need not all the information on the sum (∗ ∗ ∗) but only the following: which polynomials Rα do exist for given k. The set of all existing polynomials will be called a large type. Clearly the number of large types is finite (and does not exceed 2p(deg Q)(p−1)). It is also clear that the large type for a given k uniquely determines the large type for k + 1. This implies finiteness of the space of large types for products (∗∗). If we use several monomials then we have to define the small type of the sum S1Qn1 1 + . . . + SlQnl l as the tuple consisting of small types of summands, and the large type as the tuple con- sisting of involved small types. Now if we consider polynomials in several variables and products of the form then we have to take tuples of remainders modulo pk+1 and to collect variables having corresponding degrees. In this case we have: SiQn1 i1 · · · Qnt it xα1 1 · · · xαr r Rα(xpk+1 1 , . . . , xpk+1 r ) Xα1,... ,αr for each monomial. Then the small type of the monomial is Rα, the small type of the system is the tuple consisting of small types of monomials with given α, and finally the large type of the system is the tuple consisting of involved small types. It is easily seen that writing a new figure to the left from variables ni corresponds to change of large types (depending of the written figure), and vanishing of the expression X RiQn1 i1 · · · Qnt it depends only on its large types (since vanishing of its components obtained by grouping terms described above depends only on its small types). Thus we obtain a finite graph of states. Its vertices are large types, and arrows marked by tuples of figures 0, . . . , p − 1 are transformations of large types. The initial vertex corresponds to the large type of zero, and final vertices correspond to those large types which provide cancellation of all summands. Clearly there is a correspondence between words which may be read at arrows of the graph along a path from the initial vertex to a final, and solutions of the EDE X RiQn1 i1 · · · Qnt it = 0. 10 Now we turn to formal details. First we introduce some important constructions. Let f be a polynomial from R. Then f (ϑ) = Xy∈Σ2 p fy(ϑ ψ(y) )ϑ , and this decomposition is unique. Definition. The weeding of a polynomial f by a symbol y is the polynomial εy(f ) = fy(ϑ). The weeding of a polynomial f by a word v = yk · · · y0 is the polynomial εv(f ) = εyk(· · · (εy0(f )) · · · ). Remark. Weeding is a way to collect, as it was mentioned above, polynomials in which degrees of variables coincide modulo pk. It is easy to see that εy(f + g) = εy(f ) + εy(g) and deg εy(f ) ≤ 1 p Lemma 1. εy(f (ϑ)g(ϑ )) = εy(f (ϑ))g(ϑ). p deg f. In other words,in weeding polynomials in ϑ ϑ Proof. Represent f in the form Py∈Σ2 p ψ(y) are factored out loosing degree p. fy(ϑ ). Then p p f (ϑ)g(ϑ ) = Xy∈Σ2 ψ(y) ϑ fy(ϑ p p )g(ϑ ) = Xy∈Σ2 ψ(y) ϑ p (fy(ϑ p )g(ϑ )). By definition of weeding we immediately obtain p εy(f (ϑ)g(ϑ )) = εy(f (ϑ))g(ϑ). Lemma 2. Let c be a positive integer. Then a polynomial f (ϑ) vanishes iff every its weeding by a word of length c vanishes. Proof. Use induction in c. a) Base of induction. Suppose c = 1. Then f = 0 obviously implies εy(f ) = 0. Conversely, suppose εy(f ) = 0 for any y ∈ Σ2. Then f = Py∈Σ2 induction is proved. ψ(y) εy(f )ϑ = 0. The base of b) The inductive step. Suppose the statement is valid for c = k. Then f = 0 means that εy(f ) = 0 for any y ∈ Σ2. This in turn is equivalent to εy(εv′(f )) = 0 for any y ∈ Σ2 and any v′ ∈ Σ∗ 2 of length k + 1. The inductive step is proved. 2 of length k. Hence εv(f ) = 0 for any v ∈ Σ∗ Special operators of an equation. Return to the original equation (5) s Xi=1 Qi(ϑ)[Pi φ(u) ](ϑ) = 0. Definition. Suppose i is an integer in the interval from 1 to s, u and v are two words of equal length over alphabets Σ1 and Σ2 accordingly. Then define the special operator of the equation (5) S(i) u,v(f ) as εv(f [Pi ]). φ(u) Lemma 3. If length of words u1, v1 is equal, and similarly for u2, v2, then S(i) u1u2,v1v2(f ) = S(i) u1,v1S(i) u2,v2(f ), 11 that is, the special operator corresponding to concatenation is the composition of special operators corresponding to its factors. Proof. Denote by k the length of u1 (equal to the length of v1). Then the composition ] in the weeding. We have u2,v2(f ) equals εv1(εv2(f [Pi ]). Include [Pi ])[Pi φ(u2) φ(u1) φ(u1) S(i) u1,v1S(i) εv1(εv2(f [Pi φ(u2)+φ(u1)pk ])) = εv1v2(f [Pi φ(u1u2) ]) = S(i) u1u2,v1v2(f ). Lemma is proved. 1, v ∈ Σ∗ u,v(f ) ≤ N ′. Lemma 4 (on decreasing the degree). There exists N0 such that for any N ′ ≥ length deg f ≤ N ′ implies N0, 1 ≤ i ≤ s, and for any u ∈ Σ∗ deg S(i) In other words, rather large degrees can only decrease under the action of special operators. in a special operator, multiplying by fixed polynomials increases the degree of the original polynomial not more by a constant, and after that weeding decreases its degree not less than p times. We proceed to formalize this argument. Proof. The idea of proof is as follows: 2 of equal Denote max deg Pik by M. Then the required N0 equals prM p−1 . Indeed, N ′ = N0 + K. Then if deg f ≤ N ′, then for all x ∈ Σ1 we have deg Pi φ(x) = r Xk=1 (deg Pik)φ(k)(x) ≤ r Xk=1 Mp = Mpr. Then deg S(i) The assertion of Lemma now follows. x,y(f ) = deg εy(f Pi ) ≤ (cid:16) deg f +prM φ(x) p (cid:17) ≤ rM + N0+K p = prM p−1 + K p ≤ N ′. Types and their extensions. Definitions. A small type T = (f1, . . . , fs) is a string of polynomials from R having degree not exceeding N1 = max{max deg (Qi), N0}. A large type T is an arbitrary set of small types. The extension π(u, v)τ of a small type τ = (f1, . . . , fs) by a pair of words u ∈ Σ∗ having equal length is the small type τ ′ = (f ′ The extension Π(u)T of a large type T by a word u ∈ Σ∗ τ ∈ T, w ∈ Σ∗ s) where f ′ i = S(i) 1, . . . , f ′ u,v(fi). 2, l(w) = l(u)}. 1 is the large type T ′ = {π(u, w)τ 1, v ∈ Σ∗ 2 types. Indeed, if deg f1 ≤ N1 where N1 ≥ N0 then deg f ′ Remark. It is easy to observe that the operation of extension is defined for all small i ≤ N1, so τ ′ is also a small type. Moreover small types are strings of polynomials of bounded degree in r variables over a finite field, so their number is finite. The number of large types is finite as well since they are subsets of a finite set. Definitions. The small type of a pair of words u ∈ Σ∗ 1, v ∈ Σ∗ 2 of equal length is τ (u, v) = π(u, v)τ (λ, λ) where τ (λ, λ) = (Q1, . . . , Qs). Define the large type of a word u ∈ Σ∗ 1 as T (u) = {τ (u, w); l(w) = l(u)}. Lemma 5. T (u) = Π(u)T (λ), that is, the large type of a word u may be obtained as an extension by u of the large type of the empty word. 12 Proof. By definition T (λ) = {τ (λ, λ)}. Denote by T ′ the extension of the type T (λ) by u. Then T ′ is the set of various extensions π(u, v)τ (λ, λ) of the type τ (λ, λ) by pairs of words u, v where v is an arbitrary word of the same length as u. Since the small type τ (u, v) is by definition π(u, v)τ (λ, λ) then T ′ is the set of all τ (u, v) where v is an arbitrary word of the same length as u, and so it coincides with T (u). Lemma is proved. s Definitions. A small type τ = (f1, . . . , fs) is good if A large type T is good if all τ ∈ T are good. fi = 0. Pi=1 We proceed to prove the following Theorem 2. A large type T (u) is good iff u is a solution of the equation (5). Proof. Denote the length of u by c. A large type T (u) is good iff for all v ∈ Σ∗ 2 having length c the small type π(u, v)τ (λ, λ) is good. This in turn means that for all such v we have s or equivalently Si u,v(Qi(ϑ)) = 0 Xi=1 s Xi=1 εv(Qi[Pi φ(u) ]) = 0. Furthermore due to linearity of weeding we have and by lemma 2 εv( s Xi=1 φ(u) Qi[Pi ]) = 0 s Xi=1 Qi(ϑ)[Pi φ(u) ](ϑ) = 0. But this means that u is a solution of (5). Lemma 6. a) π(u1u2, v1v2)τ = π(u1, v1)π(u2, v2)τ. b) Π(u1u2)T = Π(u1)Π(u2)T. In other words, an extension of a type by a concatenation is a composition of extensions by factors. 1, . . . , f ′ 1 , . . . , f ′′ i = S(i) i ). Hence f ′′ i = S(i) u1,v1S(i) u2,v2(fi) = S(i) i = Su1,v1(i)(f ′ Proof. a) Suppose τ = (f1, . . . , fs). Then let τ ′ = (f ′ s) denote the extension of s ) denote the extension of τ ′ by a pair u2,v2(fi) τ by a pair of words u2, v2, and let τ ′′ = (f ′′ of words u1, v1. We proceed to prove that π(u1u2, v1v2)τ = τ ′′. Note that f ′ and f ′′ Thus π(u1u2, v1v2)τ = τ ′′ = π(u1, v1)π(u2, v2)τ. First assertion of Lemma is proved. b) Denote Π(u2)T by T ′, and Π(u1)T ′ by T ′′. We shall prove that Π(u1)Π(u2)T = T ′′ = Π(u1u2)T. By definition, T ′ consists of various extensions of types from T by pairs of words u2, v2 having equal length. Also by definition, T ′′ is the set of extensions of types from T ′ by pairs of words u1, v1 having equal length. Hence T ′′ includes all small types of the form π(u1, v1)π(u2, v2)τ. Using the first assertion of Lemma, we obtain that T ′′ consists of types π(u1u2, v1v2)τ where τ ∈ T. Putting v = v1v2 we obtain: T ′′ = {π(u1u2, v)τ τ ∈ T, l(u1u2) = l(v)}. This set is (again by definition) Π(u1u2)T. So Lemma is completely proved. u1u2,v1v2(fi). 13 We proceed to return to the assertion formulated at the beginning of this Section and to prove it. Theorem 1. L ⊂ Σ∗ 1, the set of words-solutions for the equation (5), is a regular language. Proof. Consider the following finite automaton G. Its vertices are various large types. An arrow goes from T to T ′ and is marked by the symbol x iff T ′ = Π(x)T. The initial vertex is T (λ), and final vertices are various good large types. Some u = xk · · · x0 is a solution iff T (u) is a good type. Hence Π(u)T (λ) = Π(xk) · · · Π(x0)T (λ) is a good type, and this in turn is equivalent to the assertion that T (u) is a final vertex and the end of the path xk · · · x0. Thus u is a solution iff u belongs to the language represented by the finite automaton G. This implies the assertion of Lemma. 6 Equations over a matrix ring Any element of the ring Mn(R) nay be interpreted in two ways: as a polynomial in ϑ1, . . . , ϑr with coefficients from Mn(Zp), and as a matrix with entries from R. Suppose f (B) is a polynomial in a matrix B with coefficients from R (the matrix itself belongs to Mn(R)). Denote the ring consisting of such polynomials by R[B]. Let deg B, B ∈ Mn(R), be the sum of powers of B as a polynomial in ϑ1, . . . , ϑr with matrix coefficients. Definition. A matrix B ∈ Mn(F) is rational of standard form if it has the form 0 0 · · · 1 0 · · · 0 1 · · · ... . . . 0 0 · · · ...   f0 0 f1 0 f2 0 ... ... 1 fn−1   fiξi is irreducible and separable over F. where the polynomial ξn − In fact this the matrix of multiplication by ξ in the algebraic extension of the field F by a root of the above polynomial in the basis of extension consisting of powers of this root (see [2, . 429-455]). Definition. A matrix B′ ∈ Mn(R) is entire of standard form if it has the form n−1 Pi=0 n−1 Pi=0 0 0 · · · 0 0 · · · 1 0 · · · 2 ... ... . . . 0 0 · · · n−1 0 0 0 ... ...     i ξi is irreducible and separable over F. where the polynomial ξn − Remark. If a matrix B′ is entire of standard form then there exists a unique matrix B, rational of standard form, such that B′ = B, ∈ R. Let F[B] be the ring of polynomials having the following form: fk(ϑ)Bk(ϑ), fk ∈ F, B ∈ Mn(F). We have the following m Pk=0 14 Lemma 7 (on simplifying the form of an equation). The following assertions are equivalent: a) The set of solutions for an EDE over Mn(A) is a regular language. b) The set of solutions for an EDE over A is a regular language. c) The set of solutions for an EDE over a finite algebraic extension F is a regular language. d) The set of solutions for an EDE over a finite separable algebraic extension F is a regular language. e) The set of solutions for an EDE over F[B] where B is a rational matrix of standard form is a regular language. f ) The set of solutions for an EDE over R[B′] where B′ is an entire matrix of standard form is a regular language. Proof. Observe obvious implications: a) ⇒ b) ⇒ c) ⇒ d). We proceed to prove c) ⇒ b). Consider an EDE over A. It involves only a finite number of coefficients, and all of them are algebraic over F. Hence they belong to a finite algebraic extension of F, and the original equation is an EDE over this extension. Now we shall prove d) ⇒ c). Consider an EDE over a finite algebraic extension of F: s Xi=1 bian1 i1 · · · ant it = 0. For some M all of bpM i , apM ik are separable over F. Consider the equation s Xi=1 (bi)pM (ai1)pM n1 · · · (ait)pM nt = ( s Xi=1 bian1 i1 · · · ant it )pM = 0. This equation is equivalent to the original one and is an EDE over a finite separable algebraic extension of F. Now we postpone the proof for the most difficult implication b) ⇒ a) and proceed to show that d) and e) are equivalent. d) ⇒ e). Since B is rational of standerd form, F[B] is a finite separable algebraic exten- sion for F. e) ⇒ d). By the primitive element theorem, every finite separable algebraic extension may be represented in the form F[ξ] where ξ is a root of some separable over F polynomial zn − n−1 Pi=0 fizi. Consider the matrix B: 0 0 · · · 1 0 · · · 0 1 · · · ... . . . 0 0 · · · ...   0 f0 0 f1 0 f2 ... ... 1 fn−1 .   It is rational of standard form, and F[B] is isomorphic to F[ξ]. Hence every equation from d) is an equation from e). We proceed to prove that e) and f) are equivalent. 15 e) ⇒ f). Consider an EDE over R[B′]: s Xi=1 Qi(B′)[Pi φ(u) ](B′) = 0. Suppose f is an arbitrary polynomial over R. Since B′ = B, ∈ R, B is rational of standard form, we have f (B′) = f (B) ∈ F [B]), hence R[B′] ⊂ F[B]. Thus our equation is an EDE over F[B]. f) ⇒ e) Consider an EDE over F[B] : s Xi=1 Qi(B)[Pi φ(u) ](B) = 0. Find the common denominator for all Qi, Pik and put Qi = Q′ are polynomials over R. We have i σ , Pik = P ′ ik σ where Qi, Pik, σ (cid:18) 1 σ1+φ(1)(u)+···+φ(t)(u)(cid:19) s Xi=1 Q′ i(B)[P ′ i φ(u) ](B) = 0. This equation is equivalent to the following: s Xi=1 Q′ i(B)[P ′ i φ(u) ](B) = 0. B′. Hence for any polynomial f over R we have f (B) = f ′(B ′) Now use the fact that B = 1 where f ′ also is a polynomial over R, m = deg f. Then the above equation may be written in the form m (cid:18) 1 m(1+φ(1)(u)+···+φ(i)(u))(cid:19) s Xi=1 Q′′ i (B′)[P ′′ i φ(u) ](B′) = 0. Multipying by m(1+φ(1)(u)+···+φ(i)(u)), we obtain an EDE over R[B′] equivalent to the orig- inal one. Finally we prove the last implication. b) ⇒ a). Consider an EDE over Mn(A): s Xi=1 Bi0An1 i1 Bi1 · · · Ant it Bit = 0 where Aik, Bil ∈ Mn(A). Since A is algebraically closed, Aik is representable in the form CikAJikC −1 ik where AJik is a Jordan matrix. Then AJik = Dik + Rik where Dik is a diag- onal matrix and Rik is nilpotent. Hence there exists M such that for any i, k : RpM ik = 0 and so ApM Jik Represent all ni in the form n′ i where all n′ new notation, the original equation takeas the form = (Dik + Rik)pM = DpM ik . i ≤ pM . Denote DpM ik by D′ i + pM n∗ ik. In the s Xi=1 (Bi0Ci1An′ Ji1 1 )D′n∗ i1 (C −1 i1 Bi1Ci2An′ Ji2 1 2 16 )D′n∗ i2 · · · D′n∗ it (C −1 2 t it Bit) = 0 or for fixed n′ 1, . . . , n′ t: s Xi=1 B′ i0D′n∗ i1 B′ 1 i1 · · · D′n∗ it B′ t it = 0. Denote the kl-th entry of B′ λij,k ∈ A. Let σ(x, y, n∗) denote the expression ij by βij,kl ∈ A, and k-th entry of the diagonal matrix D′ βi0,xz1βi1,z1z2 · · · βit,ztyλn∗ ij by i1,z1 · · · λn∗ 1 t it,zt 1≤z1,... ,zt≤n,1≤i≤s P for various values of x, y from 1 to n. Then the above equation is equivalent to the following system (depending on n′ − hn′ 1, . . . , n′ ni) : This is a system of n2 EDE over A and so by b) σ(x, y, n∗) = 0. L(n′, x, y) = {u ∈ Σ∗ 1 σ(x, y, φ(u)) = 0} is a regular language. Then L(n′) = \1≤x,y≤n L(n′, x, y) also is a regular language, and hence the set of solutions for the original EDE L = [0≤n′ 1,... ,n′ t<pM {n′} ∗ L(n′) is a regular language (here we apply Cleenee's theorem). Lemma is completely proved. Arguing for the ring of polynomials over a field, we have essentially applied the identity p {f (ϑ)}p = f (ϑ ). For the ring of polynomials over a non-commutative ring, in particular over a matrix ring, this identity fails. But it turns that our constructions can by extended to the case of the ring R[B] considered in assertion f) of the above lemma, by means of the following statement: Lemma 8 (on conjugation). )C −1(ϑ) where C ∈ Mn(F). a) Suppose B(ϑ) is a rational matrix of standard form. Then Bp(ϑ) = p C(ϑ)B(ϑ b) Suppose B′(ϑ) is an entire matrix of standard form. Then B′p(ϑ) = p C ′(ϑ)B′(ϑ Proof. ) )C ′−1(ϑ) where C ′ ∈ Mn(R). B = 0 0 · · · 1 0 · · · 0 1 · · · ... . . . 0 0 · · · ...   0 f0 0 f1 0 f2 ... ... 1 fn−1 .   17 n−1 Pi=0 Suppose ξn − fiξi = 0. Then B(ϑ) is the matrix of the operator x 7→ ξx in the basis 1, ξ, . . . , ξn−1. The matrix Bp(ϑ) corresponds to the operator x 7→ ξpx in the same basis. ) corresponds to the operator x 7→ ξpx in the basis 1, ξp, . . . , ξp(n−1). Due The matrix B(ϑ to well-known theorem of linear algebra, these matrices are conjugate. p p rational of standard form, ∈ R. Then B′(ϑ b) The matrix B′ is entire of standard form. It is known that B′ = B where B is )C −1(ϑ) C(ϑ) = C ′(ϑ), C ′ ∈ Mn(R), σ(ϑ) is the greatest common divisor for denominators of all entries p ) = pBp(ϑ) = pC(ϑ)B(ϑ 1 σ(ϑ) p of C(ϑ). Then C −1(ϑ)= σ(ϑ)C ′−1(ϑ), and so B′p(ϑ)=pC ′(ϑ)B(ϑ )C ′−1(ϑ)=C ′(ϑ)B(ϑ)C ′−1(ϑ). Consider an EDE over R[B] where B is an entire matrix of standard form: s Xi=1 Qi(B)[Pi φ(u) ](B) = 0. (6) Remark. Suppose f (ξ) is an arbitrary polynomial from R[ξ]. Then p f p(B(ϑ)) = C(ϑ)f (B(ϑ ))C −1(ϑ). Now put u = xk · · · x0. Transforming the right side of (6), we subsequently have 0 = s Xi=1 Qi(B(ϑ))[Pi φ(x0) ](B(ϑ))[Pi φ(x1)p )](B(ϑ)) · · · [Pi φ(xk)pk ](B(ϑ)) = = s Xi=1 Qi(B(ϑ))[Pi φ(x0) ](B(ϑ))C(ϑ)[Pi φ(x1) ](B(ϑ p p ))C −1(ϑ) · · · (C(ϑ)C(ϑ ) · · · pk · · · C(ϑ ))[Pi φ(xk) pk ](B(ϑ ))(C −1(ϑ pk ) · · · C −1(ϑ)) = Qi(B(ϑ))[Pi φ(x0) ](B(ϑ))C(ϑ) · · · [Pi φ(xk) pk ](B(ϑ pk ))C(ϑ pk )C −1(ϑ ) · · · C −1(ϑ) = s Xi=1 Now multiply the expression on the right side by an invertible matrix p C(ϑ)C(ϑ ). The resulting equation is equivalent to the original one: ) · · · C(ϑ pk Qi(B)([Pi φ(x0) ](B(ϑ))C(ϑ)) · · · ([Pi φ(xk) pk ](B(ϑ pk ))C(ϑ )) = 0. (7) s Xi=1 We proceed to generalize constructions of the first step to the matrix case. Suppose f ∈ Mn(R). Then as before f (ϑ) = Py∈Σ2 p fy(ϑ ψ(y) )ϑ . Definitions. a)The weeding by a symbol y ∈ Σ2 is εy(f ) = fy(ϑ). b)The weeding by a word v = yk · · · y0 is εv(f ) = εyk(· · · (εy0(f )) · · · ), that is, the compo- sition of weedings by letters of the word. 18 Lemma 9 (properties of the weeding operator). p deg f. )) = εy(f (ϑ))g(ϑ). a) εy(f + g) = εy(f ) + εy(g). b) deg εy(f ) ≤ 1 p c) εy(f (ϑ)g(ϑ d) If c is a constant then f = 0 iff for any v ∈ Σ∗ The proofs of these properties are similar to those given at the first step. Special operators are defined for the matrix case as follows: 2 having length c we have εv(f ) = 0. Definition. a) Suppose x ∈ Σ1, y ∈ Σ2. The special operator S(i) the matrix from Lemma 8. b) Suppose u = xk · · · x0, v = yk · · · y0 are words of equal length from Σ∗ ingly. Then S(i) x,y(f ) = εy(f [Pi u,v(f ) = S(i) xk,yk · · · S(i) x0,y0(f ). φ(x) ](B)C) where C is 1 and Σ∗ 2 accord- Lemma 10 (on decreasing the degree). There exists N0 such that for any N ′ ≥ N0, 1 ≤ i ≤ s, x ∈ Σ1, y ∈ Σ2 we have deg f ≤ N ′ ⇒ deg S(i) x,y(f ) ≤ N ′. In other words, rather high degrees of polynomials can be only decreased by special operators. Proof. Denote max deg Pik by M, and prM +deg C to prove this. Suppose N ′ = N1 + K. Then deg f ≤ N ′ implies by N1. Then N1 is the desired N0. We proceed p−1 φ(x) deg f Pi C ≤ N ′ + Mpr + deg C = (prM + deg C)(cid:18) p p − 1(cid:19) + K. Furthermore deg S(i) x,y(f ) ≤ deg f Pi φ(x) C ≤ 1 p prM + deg C p − 1 + K p ≤ N ′. Lemma is proved. Definitions. a)A small type T = (f1, . . . , fs) is a string of matrices from Mn(R) such that deg fi ≤ N2, N2 = max{max deg (Qi), N0}. b)Let x, y be symbols from alphabets Σ1 and Σ2 accordingly. The extension π(x, y)τ of a small type τ = (f1, . . . , fs) by these symbols is the small type τ ′ = (f ′ i = S(i) x,y(fi). c)Suppose u = xk · · · x0, v = yk · · · y0 are words of length from Σ∗ 2 accordingly. Then the extension π(u, v)τ of a small type τ by this pair of words is the composition of its extensions by pairs of symbols π(xk, yk) · · · π(x0, y0)τ. 1 and Σ∗ 1, . . . , f ′ s), f ′ Remark. If τ = (f1, . . . , fs) then π(u, v)τ = (f ′ follows immediately from definitions of π(u, v) and Si 1, . . . , f ′ u,v. s) where f ′ i = Si u,v(fi). This Definition. a)A large type T is an arbitrary set of small types. b) Suppose u ∈ Σ∗ {π(u, v)τ τ ∈ T, v ∈ Σ∗ 2, l(v) = l(u)}. 1. The extension of a type T by the word u is the large type Π(u)T = Lemma 11. Π(u1u2)T = Π(u1)Π(u2)T, that is, an extension of a large type by a concatenation of two words is composition of extensions of this type by the given words. 19 Proof. Denote Π(u2)T by T ′, and Π(u1)T ′ by T ′′. By definition of extension of a large type, T ′ = {π(u2, v2)τ τ ∈ T, l(v2) = l(u2)}. In turn, T ′′ is (again by definition) {π(u1, v1)τ ′ τ ′ ∈ T ′, l(v1) = l(u1)} and, by virtue of formula for T ′, equals {π(u1, v1)π(u2, v2)τ τ ∈ T, l(v1) = l(u1), l(v2) = l(u2)}, or {π(u1u2, v)τ τ ∈ T, l(v) = l(u1u2)}. This precisely coincides with Π(u1u2)T. Lemma is proved. Definition. Let u, v be words of equal length from Σ∗ 1 and Σ∗ 2 accordingly. a)The small type of the pair of words τ (u, v) is π(u, v)τ (λ, λ) where τ (λ, λ) = (Q1(B), . . . , Qs(B)). b)The large type of the word T (u) is {τ (u, w) l(w) = l(u)}. Lemma 12. T (u) = Π(u)T (λ), that is, the large type of the word u is the extension by this word of the large type of the empty word. Proof. By definition, T (λ) = {τ (λ, λ)}. So, using only definitions for extensions of large and small types, we easily obtain Π(u)T (λ) = {π(u, v)τ (λ, λ) l(v) = l(u)} = {τ (u, v) l(v) = l(u)} = T (u). Lemma is proved. Definition. a) A small type τ = (f1, . . . , fs) is good if b) A large type T is good if all τ ∈ T are good. fi = 0. s Pi=1 Theorem 3. Suppose u is an arbitrary word from Σ∗ 1. Then T (u) is a good type iff u is a solution for the original EDE. Proof. Define the length of u by c. By definition of T (u), it is a good type if for any 2 of length c the type τ (u, v) = π(u, v)τ (λ, λ) is good. This means in turn that for v ∈ Σ∗ any v ∈ Σ∗ 2 of length c we have s Xi=1 S(i) u,v(Qi(B(ϑ))) = 0. This implies that for all v ∈ Σ∗ 2 of length c we have εv(Qi(B)[Pi φ(x0) ](B(ϑ))C(ϑ) · · · [Pi φ(xc) pc ](B(ϑ pc ))C(ϑ )) = 0, s Xi=1 or for any v ∈ Σ∗ 2 of length c εv( s Xi=1 Qi(B)[Pi φ(x0) ](B(ϑ))C(ϑ) · · · [Pi φ(xc) pc ](B(ϑ pc ))C(ϑ )) = 0. Using ae property of weeding (assertion d of Lemma 9 ), we have now Qi(B)[Pi φ(x0) ](B(ϑ))C(ϑ) · · · [Pi φ(xc) pc ](B(ϑ pc ))C(ϑ ) = 0, s Xi=1 20 that is, u is a solution of (7), and so u is a solution of (6). Theorem is proved. Now it remains to construct the desired fiite automaton. Remark. The number of large and small types is finite. Small types are matrices of order n, their entries are polynomials of bounded degree in r variables over Zp. Large types are subsets of some finite sets. Theorem 4. The set of solutions for an EDE over the ring R[B] where B is an entire matrix of standard form, is a regular language. Construction of the finite automaton is completely similar to the case of the ring of polynomials. Again vertices are large types, and an arrow marked by x goes from T1 to T2 iff T2 = Π(x)T1. The initial vertex is T (λ), and final vertices are all of good large types. The proof is similar to the one given in the preceding section. Theorem 4 and Lemma 7 immediately imply the following Theorem 5. Suppose F is a field, char F = p, then the set of solutions for an EDE over Mn(F) is a regular language. Proof. An EDE includes a finite number of matrix entries, so all of them belong to some finite extension of Zp. Any such extension may be included in A if r is its transcendence degree. Hence the original equation is an EDE over Mn(A). By Lemma 7 and Theorem 4 we obtain that the set of its solutions is a regular language. Corollary. If R is a ring representable by matrices over a field F, char F = p, then the set of solutions for an EDE over R is a regular language. References [1] Salomaa A. "Jewels of formal language theory". Computer Science Press, Rockville, 1981. [2] Lang, S. "Algebra". Addison-Wesley, Reading, MA, 1965. [3] Koblitz N. "p-adic numbers, p-analysis, and zeta-functions", 2nd edition. Berlin, Springer-Verlag, 1984. [4] A. J. Belov, V. V. Borisenko, V. N. Latyshev "Monomial algebras". NY, Plenum, Vol 26. [5] Dnestrovskaya tetrad: Unsolved problems in Ring theory, [6] Belov A.J. On a Shirshov basis of relativelly free algebras of P I-degree n. Mat. Sb, 1988, vol 135, No 31, pages 373 -- 384. [7] Pchelintsev S.V. A theorem on height for alternative algebras. Mat. Sb. (N.S.), 1984, Volume 124(166), Number 4(8), Pages 557567 [8] V. A. Ufnarovskii Combinatorial and asymptotic methods in algebra // Itogi Nauki i Tekhniki. Ser. Sovrem. Probl. Mat. Fund. Napr., 1990, Volume 57, Pages 5177 21 [9] Chekanu, G. P. Local finiteness of algebras. (Russian) Mat. Issled. No. 105, Moduli, Algebry, Topol. (1988), 153171, 198. [10] Chekanu, G. P. Independence and quasiregularity in algebras. (Russian) Dokl. Akad. Nauk 337 (1994), no. 3, 316 -- 319; translation in Russian Acad. Sci. Dokl. Math. 50 (1995), no. 1, 8489 [11] Chekanu, G. P. Local Finite algebras. Ph.D. thesis -- Kishinev, 1982, [12] Mishchenko, S. P. A variant of a theorem on height for Lie algebras. (Russian) Mat. Zametki 47 (1990), no. 4, 83 -- 89; translation in Math. Notes 47 (1990), no. 3-4, 368372 22
1807.00597	1	1807	2018-07-02T10:59:53	$\mathbb Z_2$-graded codimensions of unital algebras	[ "math.RA" ]	We study polynomial identities of nonassociative algebras constructed by using infinite binary words and their combinatorial properties. Infinite periodic and Sturmian words were first applied for constructing examples of algebras with arbitrary real PI-exponent greater than one. Later we used these algebras for confirmation of the conjecture that PI-exponent increases precisely by one after adjoining an external unit to a given algebra. Here we prove the same result for these algebras for graded identities and graded PI-exponent, provided that the grading group is cyclic of order two.	math.RA	math	Z2-GRADED CODIMENSIONS OF UNITAL ALGEBRAS DUSAN D. REPOVS AND MIKHAIL V. ZAICEV Abstract. We study polynomial identities of nonassociative algebras con- structed by using infinite binary words and their combinatorial properties. Infinite periodic and Sturmian words were first applied for constructing ex- amples of algebras with arbitrary real PI-exponent greater than one. Later we used these algebras for confirmation of the conjecture that PI-exponent increases precisely by one after adjoining an external unit to a given algebra. Here we prove the same result for these algebras for graded identities and graded PI-exponent, provided that the grading group is cyclic of order two. 1. Introduction exp(A) = lim n→∞ npcn(A) We study numerical invariants of polynomial identities of algebras over a field of characteristic zero. One of the most important characteristics of identities of an algebra A is its codimension sequence {cn(A)}. In many cases this sequence is exponentially bounded and one can ask whether the limit exists. The answer is in general negative [22]. Nevertheless, there is a wide class of algebras where exp(A) exists, for example, associative PI-algebras [11], [12], finite dimensional Lie algebras [9], [8], [20], finite dimensional Jordan and alternative algebras [10], and many others. Given an algebra A, one can consider an extension A# of A, obtained from A by adjoining the external unit element. Then some natural questions arise: is the codimension sequence cn(A#) exponentially bounded, does exp(A#) exist, does there exist a relationship between exp(A) and exp(A#)? It was first mentioned in [14] that exp(A#) exists and is equal to either exp(A) or exp(A) + 1 for any associative PI-algebra A. One of the first examples of a non-associative algebra A with exp(A#) = exp(A) + 1 was found in [21]. In the same paper it was conjectured that for any algebra A either exp(A#) = exp(A) or exp(A#) = exp(A) + 1. In [19] this conjecture was confirmed for a wide class of algebras associated with infinite Sturmian words. Similar results for certain kinds of Poisson algebras were found in [18]. Note also that exp(A#) = exp(A), whenever A is a unital algebra [2]. If A is equipped with a group grading then one can also consider its graded identities and graded codimensions {cgr In the present paper we begin to study connections between asymptotics of {cn(A)} and {cgr n (A)}. We prove that for the class of algebras introduced in [6] and associated with Sturmian words, n (A)}. Date: July 3, 2018. 2010 Mathematics Subject Classification. Primary 17B01, 17B70; Secondary 15A30, 16R10. Key words and phrases. Polynomial identities; graded algebras; codimensions; exponential growth. 1 2 D.D. REPOVS AND M.V. ZAICEV graded PI-exponents exist, expgr(A) = exp(A), and expgr(A#) = expgr(A) + 1 for the most natural Z2-grading. For all details concerning the polynomial identities and their numerical invariants we refer to [3], [13]. 2. Preliminaries and main constructions Let A be an algebra over a field F of characteristic zero and let F {X} be the absolutely free algebra over F with an infinite set of generators X. A polynomial f = f (x1, . . . , xn), x1, . . . , xn ∈ X, is called an identity of A if f (a1, . . . , am) = 0, whenever a1, . . . , an ∈ A. The set Id(A) of all identities of A forms an ideal of F {X}. Denote by Pn the subspace of all multilinear polynomials on x1, . . . , xn. Then Pn ∩ Id(A) is the set of all multilinear identities of A of degree n. It is well known that all identities of A are completely defined by the family of subspaces {Pn ∩ Id(A)}, n = 1, 2, . . . . An important numerical invariant of identical relations of the algebra A is the sequence of codimensions cn(A) = dim Pn Pn ∩ Id(A) . In the case of exponentially bounded growth of {cn(A)}, one can define the lower and the upper PI-exponents by setting exp(A) = lim inf n→∞ and the ordinary PI-exponent npcn(A), exp(A) = lim sup n→∞ npcn(A) exp(A) = lim n→∞ npcn(A), provided that exp(A) = exp(A). A powerful tool for studying asymptotics of codimensions is the representation theory of the symmetric group Sn. The group Sn acts naturally on the space Pn of multilinear polynomials σ f (x1, . . . , xn) = f (xσ(1), . . . , xσ(n)). Under this action, the subspaces Pn, Pn ∩ Id(A) and the quotient Pn(A) = Pn Pn ∩ Id(A) become Sn-modules. Consider the nth cocharacter of A, that is the character of Pn(A), χn(A) = χ(Pn(A)), and its decomposition into irreducible components where χλ denotes the irreducible Sn-character, corresponding to the partition λ of n, and the integer mλ denotes its multiplicity in χn(A). Denote by dλ = deg χλ the dimension of the irreducible Sn-module with the character χλ. It follows from (1) that (1) (2) χn(A) =Xλ⊢n mλχλ, cn(A) =Xλ⊢n mλdλ. Another important numerical characteristic of Id(A) is its nth colength ln(A) =Xλ⊢n mλ. Z2-GRADED CODIMENSIONS OF UNITAL ALGEBRAS 3 In many cases the sequence ln(A) is polynomially bounded while dλ's in (2) are exponentially large. This means that the asymptotics of cn(A) is actually defined by the maximal value of dλ with mλ 6= 0. For group graded algebras, identical relations and corresponding numerical in- variants can also be considered. We restrict ourselves to the case of Z2-gradings. Consider the free algebra F {X, Y } with two independent sets of generators X and Y . We can endow F {X, Y } with a Z2-grading, by setting deg x = 0, deg y = 1, for all x ∈ X, y ∈ Y , and extending this grading to all monomials on X ∪ Y . If A = A0 ⊕A1 is a Z2-graded algebra over F then a polynomial f (x1, . . . , xk, y1, . . . , ym) ∈ F {X, Y, } is a graded identity of A if f (a1, . . . , ak, b1, . . . , bm) = 0, for all a1, . . . , ak ∈ A0, b1, . . . , bm ∈ A1. The set of all graded identities of A forms a homogeneous in Z2-grading ideal Idgr of F {X, Y }. The intersection Pk,m ∩ Idgr(A) consists of all multilinear graded identities of degree k on even variables and of degree m on odd variables, where Pk,m is the subspace of all polynomials multilinear on x1, . . . , xk, y1, . . . , ym. As before, the symmetric groups Sk, Sm act independently on even and odd variables and both Pk,m and Pk,m ∩ Idgr(A), and also Pk,m(A) = Pk,m Pk,m ∩ Idgr(A) are Sk × Sm-modules. One can decompose Pk,m(A) into irreducible components and write χ(Pk,m(A)) = Xλ⊢k µ⊢m mλ,µχλ,µ, where χλ,µ is the irreducible Sk × Sm-character and mλ,µ is its multiplicity. It is well known that χλ,µ = χλ ⊗ χµ and that deg χλ,µ = deg χλ deg χµ = dλdµ. Partial codimensions and colengths are defined as follows: ck,m(A) = deg χ(Pk,m(A)) = dim Pk,m(A), lk,m(A) = Xλ⊢k µ⊢m mλ,µ. Finally, the graded nth codimension and the colength of A are equal to expgr(A) = lim inf n→∞ expgr(A) = lim sup n→∞ exp(A)gr = lim n→∞ n (A). n (A), n (A), nqcgr nqcgr nqcgr cgr n (A) = lgr n (A) = (cid:18)n k(cid:19)ck,n−k(A) nXk=0 nXk=0 lk,n−k(A), and respectively. Graded PI-exponents are defined similarly, 4 D.D. REPOVS AND M.V. ZAICEV Generalizing (2), we get (3) ck,n−k(A) = Xλ⊢k µ⊢n−k mλ,µdλdµ. Graded and ordinary codimensions satisfy the relation (4) (see [7] or [1]). cn(A) ≤ cgr n (A) We will use the following auxiliary function for computing codimensions. Let x1, . . . , xd be non-negative real numbers such that x1 + · · · + xd = 1, d ≤ 2. Then If d = 2 then we write Φ(x1, . . . , xd) = 1 1 · · · xxd xx1 d . Φ(x1, x2) = Φ(x) = 1 xx(1 − x)1−x instead of Φ(x1, x2), where 0 ≤ x ≤ 1. 3. Sturmian words and Sturmian algebras In this section we recall the construction of algebras based on infinite binary words and their combinatorial properties. First, let K = k1k2 . . . be an infinite word with integers ki ≥ 2, i = 1, 2, . . .. Denote by A(K) a non-associative algebra with the basis {a, b, z(i) j 1 ≤ j ≤ ki, i ≥ 1} and with the multiplication table given by ki−1a = z(i) z(i) 1 a = z(i) 2 , . . . , z(i) , z(i) ki b = z(i+1) 1 , i = 1, 2, . . . . ki All other products are zero. Note that A(K) is 2-step left nilpotent, that is x1(x2x3) ≡ 0 is an identity of A(K). It allows us to omit brackets in all prod- ucts and write x1x2x3 · · · xn instead of (· · · ((x1x2)x3) · · · )xn, keeping in mind that all non-left normed products are zero. Algebras of this type are used intensively in the study of numerical invariants of polynomial identities. For instance, in [6] the first examples of algebras with an arbitrary exponential growth αn, 1 ≤ α ∈ R, were presented. Examples of algebras with an intermediate growth nnβ , 0 < β < 1, were constructed in [5]. Recently, examples of commutative algebras with polyno- mial codimension growth nα, 3 < α < 4, were presented in [4]. Other important examples of abnormal codimension growth were constructed in [16], [22]. In the present paper we study identities on algebras A(K) of special kind. Let m ≥ 2 be an integer and let w = w1w2 . . . be an infinite word in the alphabet {0; 1}. We denote by A(m, w) the algebra A(K), where K is constructed as follows: ki = m + wi, i = 1, 2, . . . . Earlier, the algebras A(m, w) have already been used for constructing a continuous family of unitary algebras with non-integer PI-exponents and for confirmation of the conjecture that exp(A#) = exp(A) + 1 (see [19]). We recall some well known facts from the combinatorial theory of infinite words (see, for example, [17]). Given a binary word w = w1w2 . . ., the complexity Compw of w is the function Compw : N → N, where Compw(n) is the number of distinct subwords of w of length n. It is easy to see that for a periodic word w with period Z2-GRADED CODIMENSIONS OF UNITAL ALGEBRAS 5 T , the complexity function is bounded, Compw(n) ≤ T . Moreover, it is well known that Compw(n) ≥ n + 1 for any aperiodic w. An infinite word w is called Sturmian if Compw(n) = n + 1 for all n ≥ 1. For a finite word x = x1 . . . xn in the alphabet {0; 1}, the height h(x) and the length x are defined as h(x) = x1 + · · · + xn and x = n, respectively. Then the slope π(x) is defined by π(x) = h(x) x . One can extend this notion to certain infinite binary words. Namely, if the limit π(w) = lim n→∞ h(w1 . . . wn) n exists then π(w) is called the slope of w. Clearly, the limit does not exist in general. Nevertheless, for periodic and Sturmian words, the slope is well defined. In the next proposition we recall the basic properties which we will need in the sequel. Proposition 1. ([17, Section 2.2]). Let w be a Sturmian or periodic word. Then there exists a constant C such that 1. h(x) − h(y) ≤ C, for any finite subwords x, y of w with x = y; 2. the slope π(w) of w exists; 3. for any non-empty finite subword u of w, π(u) − π(w) ≤ C u ; and 4. for any real α ∈ (0; 1), there exists a word w with π(w) = α and w is Sturmian or periodic, according to whether α is irrational or rational, re- spectively. We will use the following results. Theorem 1. ([6, Theorem 5.1] Let w be a Sturmian or periodic word with the slope 0 < α < 1. If m ≤ 2 then for the algebra A = A(m, w) the PI-exponent exists and exp(A) = Φ(β), where β = 1 m+α . (cid:3) Theorem 2. ([24, Theorem 1]) Let A = A(m, w), where w is an infinite Sturmian or periodic word, and m ≥ 2. Let A# be the algebra obtained from A by adjoining an external unit. Then PI-exponent of A# exists and exp(A#) = exp(A) + 1. (cid:3) 4. Gradings on Sturmian algebras The algebra A = A(m, w) can be equipped by a Z2-grading in different ways. We begin our study with the most natural case when generators of A are homogeneous. The algebra A is generated by the three elements z(1) 1 , a, b. Each generator can be even or odd, so we have eight options. Clearly, if deg z(1) 1 = deg a = deg b = 0 then the grading is trivial and all identities and codimensions are the same as in the non-graded case. In the present paper we consider one of non-trivial cases when z(1) and a are even, whereas b is odd. At the end we will discuss the difference 1 between distinct gradings. 6 D.D. REPOVS AND M.V. ZAICEV Throughout this section, let A = A(m, w) be the algebra defined in the previous section, where m ≥ 2 is an integer and w is an infinite periodic or Sturmian word. Then a Z2-grading A = A0 ⊕ A1 on A is uniquely defined by setting deg z(1) 1 = deg a = 0, deg b = 1. First, we will give an upper bound for the graded codimension cgr n (A). Lemma 1. Let ck,n−k(A) be the partial graded codimension of A. Then ck,n−k(A) ≤ 2n2 for all large enough n. Proof. Consider a left-normed monomial M = M (x1, . . . , xk, y1, . . . , yn−k) on even x1, . . . , xk and odd y1, . . . , yn−k. Then M = xiu1 · · · un−1 or M = yiu1 · · · un−1, where u1, . . . , un−1 are some xj's, yj's. Let for example, M = xk · · · xi1 · · · xik−1 · · · , {i1, . . . , ik−1} = {1, . . . , k − 1}. Then M ≡ M0 modulo the graded ideal Idgr(A), where M0 = xk · · · x1 · · · xk−1 · · · , since any non-zero evaluation ϕ of M and M0 in A can be obtained only if ϕ(xk) = z(i) j , ϕ(x1) = · · · = ϕ(xk−1) = a. Moreover, ϕ(M ) 6= 0 if and only if the positions of xi1 , . . . , xik−1 in M are in 1-1 correspondence with the positions of symbol 0 in the subword ¯w = wt+1 · · · wt+n−1 of length n − 1, where the integer t can be computed from the condition z(i) 1 u1 · · · ut for proper u1, . . . , ut ∈ {a, b}. Similarly, y1, . . . , yn−k in M can be ordered naturally. Since Compw(n − 1) = n for Sturmian word and Compw is bounded in the periodic case, we conclude that the number of subwords ¯w corresponding to monomials that do not vanish on A does not exceed kn ≤ n2 for sufficiently large n. The same upper bound takes place for monomials of the type yiu1 · · · un−1, and we have completed the proof. j = z(1) (cid:3) Lemma 2. For any real number ε > 0 there exists an integer n0 such that condi- tions n ≥ n0, Pk,n−k(A) 6= 0 imply the inequalities (5) β − ε ≤ n − k n ≤ β + ε, where β = 1 A(m, w). m+α and α = π(w) is the slope of the infinite word w defining A = Proof. Any non-zero product of n basis elements of A has the form (6) z(i) j a · · · a {z }s0 b a · · · a {z }s1 b · · · b a · · · a {z }sr b a · · · a {z } sr+1 = z(i+r+1) 1+sr+1 , where 0 ≤ s0, sr+1 ≤ m, (7) s1 = m + wi+1 − 1, . . . , sr = m + wi+r − 1, n = s0 + sr+1 + 2 + mr + wi+1 + · · · + wi+r. The number of factors b in this product is equal to r + 1. Moreover, (6) is the value of monomials from Pk,n−k with n − k = r + 1. Hence n − k r + 1 (8) = n s0 + sr+1 + 2 + mr + wi+1 + · · · + wi+r = 1 + 1 r m + s0+sr+2+2 r + wi+1+···+wi+r r . Z2-GRADED CODIMENSIONS OF UNITAL ALGEBRAS 7 Since s0 + sr+1 + 2 ≤ 2m + 2 and wi+1 + · · · + wi+r ≤ r, it follows that r ≥ n m + 1 − 2. In particular, r → ∞ if n → ∞. Moreover, the limit of 1 r (wi+1 + · · · + wi+r), as r → ∞, is equal to α, by Proposition 1. It follows that the right hand side of (8) goes to β = 1 m+α as n → ∞ and hence (5) holds. Lemmas 1 and 2 give an upper bound for graded codimensions of A. Lemma 3. For any 0 < ε ≤ 1 m+α there exists n0 such that 2 − 1 cgr n (A) ≤ 2n3Φ(β + ε)n for all n ≥ n0, where β = 1 m+α . In particular, expgr(A) ≤ Φ(β). Proof. By Lemmas 1 and 2 we have cgr n (A) ≤ Xβ−ε≤ n−k n ≤β+ε (cid:18)n k(cid:19)ck,n−k(A) ≤ 2n2 Xβ−ε≤ n−k n ≤β+ε (cid:18)n k(cid:19). By Stirling's formula for factorials we have (cid:18)n k(cid:19) ≤ n nn kk(n − k)n−k = nΦ( m+α < 1 2 and n − k n )n. Since m ≥ 2 and 0 < α < 1, we have β = 1 max β−ε≤ n−k n Φ( n − k n ) ≤ Φ(β + ε) as soon as β + ε < 1 2 and n is sufficiently large. Hence cgr n (A) ≤ 2n3Φ(β + ε)n, and we are done. Now we are ready to prove main result of this section. (cid:3) (cid:3) Theorem 3. Let A = A(m, w) be the algebra defined by an integer m ≥ 2 and by an infinite periodic or Sturmian word w with the slope π(w) = α. Suppose that the decomposition A = A0 ⊕ A1 is a Z2-grading of A such that a, z(1) 1 ∈ A0, b ∈ A1. Then the graded PI-exponent expgr(A) exists and expgr(A) = exp(A) = Φ( 1 m + α ). Proof. According to Lemma 3, it is enough to show that expgr(A) ≥ Φ(β), where β = 1 m+α . Since A is not nilpotent, there exists for any n, a non-zero product of the type (6). In particular, given n, there exists 0 ≤ k ≤ n such that Pk,n−k 6= 0. Then n−k n ≥ β − ε asymptotically for any fixed ε > 0 by Lemma 2, and by Stirling's formula we have n (A) ≥(cid:18)n k(cid:19)ck,n−k(A) ≥(cid:18)n k(cid:19) ≥ cgr 1 n2 nn kk(n − k)n−k ≥ 1 n2 Φ(β − ε)n. It follows that expgr(A) ≥ Φ(β), and thus the proof has been completed. (cid:3) 8 D.D. REPOVS AND M.V. ZAICEV 5. Algebras with adjoint unit In this section we study codimensions of algebras with an external unit. Given an algebra B, we denote by B# the algebra obtained by adjoining the external unit to B. Note that if C = ⊕g∈GCg is a G-graded algebra with unit 1 then 1 is a homogeneous element and 1 ∈ Ce, where e ∈ G is the identity element of the group G. Therefore in the case of a Z2-graded algebra B, its extension B# = B ⊕ 1 has a unique Z2-grading B# = B# 1 , where B is a homogeneous subalgebra of B#, namely, B# 0 ⊕ B# 0 = B0 ⊕ 1, B# 1 = B1. First, let A0 ⊕ A1 be an arbitrary Z2-graded algebra. Denote by R{X, Y } the relatively free Z2-graded algebra of the variety vargr(A) of graded algebras gener- ated by A with two infinite sets X and Y of even and odd generators, respectively. That is, R{X, Y } = F {X, Y }/Idgr(A). Consider a partial (k, n − k)-cocharacter of A, (9) χk,n−k(A) = χ(Pk,n−k(A)) = Xλ⊢k µ⊢n−k mλ,µχλ,µ. In order to bound the multiplicities mλ,µ in (9) we denote by Rk,n−k d0,d1 (A) the subspace of polynomials on Xd0 = {x1, . . . , xd0}, Yd1 = {y1, . . . , yd1} in R{X, Y } of total degree k on Xd0 and total degree n − k on Yd1. The same argument as in [23] gives us the next lemma. Recall that the height h(λ) of a partition λ = (λ1, . . . , λt) is the number t of its parts. Lemma 4. Let mλ,µ, λ ⊢ k, µ ⊢ n − k, be the multiplicity from (9) with h(λ) ≤ d0, h(µ) ≤ d1. Then mλ,µ ≤ dim Rk,n−k d0,d1 (A). (cid:3) Now, let A = A(m, w) be the algebra defined by an infinite binary word w = w1w2 . . . . The following lemma holds for any w, not necessarily Sturmian or periodic. Lemma 5. Suppose that A is Z2-graded and that deg z(1) Then 1 = deg a = 0, deg b = 1. dim Rk,n−k d0,d1 (A) ≤ d0m2kCompw(n − k). Proof. Denote by W the span of all monomials (10) x0u1 · · · un−1 in R{X, Y }, where u1, . . . , un−1 ∈ Xd0 ∪ Yd1, ui1, . . . , uik−1 ∈ Xd0 for some i1, . . . , ik−1 ∈ {1, . . . , n − 1}, while uj ∈ Yd1, provided that j 6= i1, . . . , ik−1. Clearly, dim Rk,n−k d0,d1 (A) ≤ d0 dim W . Let f = f (x0, . . . , xd0 , y1, . . . , yd1) ∈ F {X, Y } be a linear combination of mono- mials of the same type as (10). Then f ≡ 0 is an identity of A if and only if σ(f ) = 0 for any homomorphism σ : F {X, Y } → A such that σ(x0) = z(i) j , σ(xs) = a, σ(ys) = b. Z2-GRADED CODIMENSIONS OF UNITAL ALGEBRAS 9 Hence dim W does not exceed the codimension of the intersection of all Ker σ in is a subspace of F {X, Y } defined similarly as Rk,n−k F k,n−k (A). d0,d1 d0,d1 Consider the family of graded homomorphisms ϕij : F {X, Y } → A such that , where F k,n−k d0,d1 ϕij (x0) = z(i) j , ϕij (xs) = a, ϕij (ys) = b, for all xs ∈ X, ys ∈ Y . Then either ϕij(x0u1 · · · un−1) = 0 or ϕij (x0u1 · · · un−1) = z(i+r+1) 1+sr+1 , the element from (6). The latter equality takes place if and only if s0 = m − 1 − j + wi, 0 ≤ sr+1 ≤ m − 1 + wi+r+1, n = s0 + sr+1 + 2 + mr + wi+1 + · · · + wi+r, relations (7) hold and all x1, . . . , xd0 stay on "correct" posi- tions among u1, . . . , un−1, according to the word w. In particular, codim Ker ϕij in F k,n−k is less than or equal to one. Moreover, Ker ϕij = Ker ϕi′j if the sub- d0,d1 words wi+1 · · · wi+r+1, wi′+1 · · · wi′+r+1 coincide. It follows that the codimension is at most m2Compw(r + 1) = m2Compw(n − k). Since of T Ker ϕij in F k,n−k dim W is equal to codim T Ker ϕij, we have completed the proof of the lemma. (cid:3) Next, we will find an upper bound for dim Rk,n−k d0,d1 d0,d1 (B#) in terms of dim Rk,n−k d0,d1 (B) if B is a Z2-graded algebra. Lemma 6. Given a Z2-graded algebra B, suppose that dim Rk,n−k d0,d1 for all 0 ≤ k ≤ n and for some constant θ. Then (B) ≤ θk(n−k)T dim Rk,n−k d0,d1 (B#) ≤ θ(k + 1)d0+2(n − k + 1)T +d1. Proof. Note that a multihomogeneous polynomial f (x1, . . . , xd0 , y1, . . . , yd1) is a graded identity of B# if and only if all multihomogeneous on x1, . . . , xd0, y1, . . . , yd1 components of f (1 + x1, . . . , 1 + xd0 , y1, . . . , yd1) ∈ F {X, Y }# are identities of B. The total number of such components does not exceed (k+1)d0(n−k+1)d1 , provided that the degree on {x1, . . . , xd0} is at most k and the degree on {y1, . . . , yd1} is equal to n − k. Let f1, . . . , fN ∈ F k,n−k d0,d1 . Consider the linear combination f = λ1f1 + · · · + λN fN with unknown coefficients λ1, . . . , λN . Any multihomogeneous component g = g(x1, . . . , xd0, y1, . . . , yd1) of f (1 + x1, . . . , 1 + xd0 , y1, . . . , yd1) gives us at most dim Rj,n−k d0,d1 (B) ≤ θj(n − k)T linear equations on λ1, . . . , λN , provided that g ≡ 0 is an identity of B and the degree of g on x1, . . . , xd0 is equal to j. Hence f ≡ 0 is an identity of B# if λ1, . . . , λN satisfy no more than eN linear equations, where kXj=0 j. eN = (k + 1)d0(n − k)d1θ(n − k)T eN ≤ θ(k + 1)d0+2(n − k + 1)T +d1. Note that (11) Therefore if N is greater than the right hand side of (11) then f1, . . . , fN are linearly dependent modulo Idgr(B#) and we have completed the proof. (cid:3) Now we are ready to get an upper bound for graded colength of A#. Lemma 7. Let A = A(m, w), where m ≥ 2 is an integer and w is an infinite periodic or Sturmian word. Then lk,n−k(A#) ≤ 3m2(k + 1)8(n − k + 1)6 10 and D.D. REPOVS AND M.V. ZAICEV n (A#) ≤ 3m2(n + 1)15. lgr Proof. Consider a partial cocharacter of A# (12) χk,n−k(A#) = Xλ⊢k µ⊢n−k mλ,µχλ,µ. The linear subspace I = Span{z(i) zero multiplication and j i, j ≥ 1} forms a homogeneous ideal of A# with dim(cid:0)A#/I(cid:1)0 = 2, dim(cid:0)A#/I(cid:1)1 = 1. Hence any multilinear polynomial alternating on 4 even variables or on 3 odd vari- ables is an identity of A#. Standard argument implies that mλ,µ 6= 0 in (12) only if h(λ) ≤ 3, h(µ) ≤ 2. By Lemma 5 we have dim Rk,n−k Then by Lemmas 4 and 6, 3,2 (A) ≤ 3m2k(n − k + 1) ≤ 3m2k(n − k)2. mλ,µ ≤ dim Rk,n−k 3,2 (A#) ≤ 3m2k(k + 1)5(n − k + 1)4. The number of summands on the right hand side of (12) is not greater than k3(n − k)2, hence lk,n−k(A#) ≤ 3m2k(k + 1)8(n − k + 1)6 and lgr n (A#) ≤ 3m2(n + 1)15. Now we specify necessary conditions for inequality mλ,µ 6= 0 in (12). (cid:3) Lemma 8. Let A = A(m, w) and suppose that w is a Sturmian or periodic word with the slope α. Suppose that mλ,µ 6= 0 in (12), where λ ⊢ k, µ ⊢ n − k. Then λ and µ satisfy the following conditions: 1. λ = (λ1, λ2, λ3) with λ3 ≤ 1; 2. µ = (µ1, µ2) with µ2 ≤ 1; 3. λ1 + λ2 + µ1 = n or n − 1; and 4. for any 0 < ε < 1 2 − β there exists an integer n0 such that µ1 ≤ β + ε 1 − β − ε λ1 for all n ≥ n0, where β = 1 m+α . Proof. Any multilinear polynomial f containing an alternating set of order 4 on even variables vanishes on A#. If A# contains two alternating sets of order 3 on even variables then also f ∈ Idgr(A#). From the structure of essential idempotents of group ring F Sk it follows that λ4 = 0, λ3 ≤ 1. This proves 1. Similar argument gives us 2. Let λ = (λ1, λ2, 1), µ = (µ1, 1). If M is an irreducible Sk × Sn−k-submodule of Pk,n−k ⊂ F {X, Y } with the character χ(M ) = χλ,µ then M is generated by a polynomial f (x1, . . . , xk, y1, . . . , yn−k) alternating on x1, x2, x3 and on y1, y2. If we evaluate x1, x2, x3 on {1, a} and y1, y2 on {b} then we get zero. Otherwise, x1, x2, x3 should be equal to 1, a, z(1) s . In this case the value is also zero. Hence λ3 + µ2 ≤ 1, and we obtain 3. , and y1, y2 should be equal to b, z(r) j Let us prove 4. If mλ,µ 6= 0 then there exists a polynomial f = f (x1, . . . , xk, y1, . . . yn−k) Z2-GRADED CODIMENSIONS OF UNITAL ALGEBRAS 11 which generates an irreducible Sk × Sn−k-module with the character χλ,µ and an evaluation ϕ : X ∪ Y → A# such that ϕ(f ) 6= 0. Moreover, f contains λ2 disjoint alternating sets of x′s of order 2. The set of values {ϕ(x1), . . . , ϕ(xk)} contains λ1 ≥ p ≥ λ2 − 1 elements a, at most one element z(i) j and k − p or k − p − 1 units. The set {ϕ(y1), . . . , ϕ(yn−k)} contains at most one odd z(i) and q = µ1 − 1 or µ1 j elements b. Furthermore, there exists a non-zero product g = z(i) j a · · · aba · · · ab · · · ba · · · a which is equal (up to a scalar factor) to ϕ(f ). Denote by p = dega g, q = degb g the numbers of entries of a and b in g, respectively. If the total degree N = deg g = 1 + p + q increases then there exists a corellation between the growth of p and q (provided that g 6= 0). Namely, hence lim n→∞ q q + p = β = 1 m + α , lim n→∞ q p = β 1 − β . It follows that there exists r such that for any q ≥ r + 1 (and for corresponding p) we have q p ≤ β + ε/2 1 − (β + ε/2) and 1 p + β + ε/2 1 − (β + ε/2) ≤ β + ε 1 − (β + ε) . Since µ1 − 1 ≤ q and p ≤ λ1, we get and µ1 − 1 µ1 ≤ q p ≤ β + ε/2 1 − (β + ε/2) µ1 λ1 ≤ q p + 1 λ1 ≤ q p + 1 p ≤ β + ε 1 − (β + ε) , provided that µ1 ≥ r. On the other hand, if µ1 < r then n µ1 > n r and λ1 µ1 > n 2r − 1 since 2λ1 + µ1 ≥ λ1 + λ2 + µ1 ≥ n − 1. Denote for short γ = β+ε n ≥ 2(γ+1) r we have γ 1−(β+ε) . Then for all n 2r ≥ 1 γ + 1, λ1 µ1 > 1 γ . This proves 4. (cid:3) In order to get an upper bound for graded codimensions we need some properties of the function Φ(x1, . . . , xd) introduced in Section 2. Recall that Φ(x1, x2, x3) = x−x1 1 x−x2 Lemma 9. Let x3 = γx2 for a fixed coefficient γ. Then , where 0 ≤ x1, x2, x3 ∈ R, x1 + x2 + x3 = 1. 2 x−x3 3 max Φ(x1, x2, x3) = 1 + γ γγ/(γ+1) + 1. 12 D.D. REPOVS AND M.V. ZAICEV Proof. Denote x = x1. Then the relations x1 + x2 + x3 = 1, x3 = γx2 imply x2 = 1 − x 1 + γ , x3 = γ 1 + γ (1 − x). Denote also Φ(x1, x2, x3) = f (x). Then f −1(x) = xx(cid:18) 1 − x 1 + γ(cid:19) 1−x 1+γ (cid:18) γ(1 − x) 1 + γ (cid:19) γ(1−x) 1+γ and Hence g(x) = ln f −1(x) = x ln x + 1 − x 1 + γ ln 1 − x 1 + γ + γ(1 − x) 1 + γ ln γ(1 − x) 1 + γ . g′(x) = ln x ( 1−x 1+γ ) 1 1+γ ( γ(1−x) 1+γ ) γ 1+γ ex =(cid:18) 1 −ex 1 + γ(cid:19) 1 1+γ (cid:18) γ(1 −ex) 1 + γ (cid:19) γ 1+γ γ γ 1+γ = (1 −ex)ρ, and g′(ex) = 0 only if where That is, Since ρ = ex = g′(x) = ln γ 1+γ γ 1 + γ ρ 1 + ρ x 1 − x . . + const on the interval (0; 1), we see that f −1(x) has a local minimum in ex. Direct compu- tations show that f −1(ex) = ex and max Φ = f (ex) = (cid:3) Now we are ready to compute the required upper bound for the upper PI- = 1 + 1 + γ ex + 1. 1 θ = 1+γ γ 1 γ exponent. Remark 1. If we denote give us γ γ+1 by θ then γ = θ 1−θ . In this case direct computations θθ(1 − θ)1−θ = Φ(θ). Moreover, if γ1 < γ2 ≤ 1 then θ1 < θ2 and Φ(θ1) < Φ(θ2). = 1+γ γ γ 1 + γ 1 Lemma 10. Proof. By (3) (13) expgr(A#) ≤ expgr(A) + 1. cgr n (A#) ≤ lgr n (A#)Xk Xλ⊢k,µ⊢n−k mλ,µ6=0 (cid:18)n k(cid:19)dλdµ. Z2-GRADED CODIMENSIONS OF UNITAL ALGEBRAS 13 First estimate a fixed summand (cid:0)n k(cid:1)dλdµ provided that mλ,µ 6= 0. By Lemma 8, we have λ = (λ1, λ2, λ3), λ3 ≤ 1, µ = (µ1, µ2), µ2 ≤ 1. By the Hook formula for degree of an irreducible representation, dλ ≤ k! λ1!λ2! , dµ ≤ n − k − 1 ≤ n. Since n − k = µ1 or µ1 − 1, we have n − k + 1 ≥ µ1 and n(n − k)! ≥ µ1!. Also, n − 2 ≤ λ1 + λ2 + µ1 ≤ n, that is n! ≤ (λ1 + λ2 + µ1)!(n + 2)2. Therefore (14) (cid:18)n k(cid:19)dλdµ ≤ n! k!(n − k)! · k! λ1!λ2! · n ≤ (n + 2)4 (λ1 + λ2 + µ1)! λ1!λ2!µ1! . By the Stirling's formula (λ1 + λ2 + µ1)! λ1!λ2!µ1! ≤ n (15) where (λ1 + λ2 + µ1)λ1+λ2+µ1 λλ1 1 λλ2 2 µµ1 1 ≤ nΦ(x1, x2, x3)n, x1 = λ2 λ1 + λ2 + µ1 , x2 = λ1 λ1 + λ2 + µ1 , x3 = µ1 λ1 + λ2 + µ1 . Denote µ1/λ1 = γ. Then x3 = γx2, and by Lemma 9 and Remark 1, where θ = γ γ+1 . Fix an arbitrary small ε > 0. We can assume that Φ(x1, x2, x3) ≤ Φ(θ) + 1, β + ε 1 − β − ε < 1. Then by Lemma 8 we get that γ < 1, θ is an increasing function of γ on interval (0; 1) and θ ≤ 1 2 . Hence Φ(θ) is also an increasing function of γ and Φ(x1, x2, x3) ≤ 1 + Φ(β + ε) for all sufficiently large n. Applying (13), (14), (15) and Lemma 7 and taking into account that the number of partitions λ ⊢ k with h(λ) ≤ 3, λ3 ≤ 1 is not greater than n, we obtain cgr n (A#) ≤ 6m2(n + 2)22(1 + Φ(β + ε))n, from which it follows that expgr(A#) ≤ 1 + Φ(β) = 1 + expgr(A). (cid:3) Theorem 4. Let A = A(m, w) = A0 ⊕ A1 be the algebra defined by an integer m ≥ 2 and by Sturmian or periodic word w equipped with a Z2-grading, where generators z(1) and a are even whereas b is odd. Let A# be obtained from A by adjoining the external unit. Then its graded PI-exponent exists and 1 expgr(A#) = 1 + expgr(A). Proof. By [19, Theorem 1], exp(A#) = exp(A)+1. Hence expgr(A#) ≥ exp(A#) = exp(A) + 1 = expgr(A) + 1 by (4) and Theorem 3. Now our statement follows from Lemma 10. (cid:3) In conclusion, we discuss other Z2-gradings on A = A(m, w). In the proof of 1 = 0. Hence the same Theorems 3 and 4 we have never used the fact that deg z(1) results hold for graded codimensions if deg z(1) 1 = deg b = 1, deg a = 0. 14 D.D. REPOVS AND M.V. ZAICEV By slightly modifying arguments, one can prove Theorems 3 and 4, provided that deg a = 1, deg b = 0. Finally, if deg a = deg b = 1 then the argument is similar to that of [6], [19]. Therefore we can generalize Theorems 3 and 4 as follows. Theorem 5. Let A = A(m, w) be the algebra defined by an integer m ≥ 2 and by an infinite periodic or Sturmian word w with the slope π(w) = α. Suppose that the decomposition A = A0 ⊕ A1 is a Z2-grading such that the generators a, b, z(1) are 1 homogeneous. Then the graded PI-exponent expgr(A) exists and where expgr(A) = exp(A) = Φ( 1 m + α ), Φ(x) = 1 xx(1 − x)(1−x) . Moreover, if A# is obtained from A by adjoining an external unit with the induced Z2-grading then expgr(A#) also exists and expgr(A#) = expgr(A) + 1. Acknowledgements (cid:3) We express our sincere thanks to the referee for the numerous comments and suggestions. The first author was supported by the Slovenian Research Agency grants BI-RU/16-18-002 and P1-0292. The second author was supported by the Russian Science Foundation grant 16-11-10013. References [1] Yu. Bahturin, V. Drensky, Graded polynomial identities of matrices, Linear Algebra Appl. 357 (2002) 15-34. [2] O. E. Bezushchak, A. A. Beljaev, M. V. Zaitsev. Exponents of identities of algebras with adjoint unit, Vestnik Kievskogo Nats. Univ. Ser. pfiz.-mat. nauk 3 (2012) 7-9. [3] V. Drensky,Free algebras and PI-algebras. Graduate course in algebra (Springer-Verlag Sin- gapore, 2000). [4] A. Giambruno, S. P. Mishchenko, A. Valenti, M. V. Zaicev, Polynomial codimension growth and the Specht problem, J. Algebra 469 (2017), 421436. [5] A. Giambruno, S. P. Mishchenko, M. V. Zaicev, Algebras with intermediate growth of the codimensions, Adv. in Appl. Math. 37(3) (2006) 360-377. [6] A. Giambruno, S. Mishchenko, M. Zaicev, Codimensions of algebras and growth functions, Adv. Math. 217(3) (2008) 1027-1052. [7] A. Giambruno, A. Regev, Wreath products and P.I. algebras, J. Pure Appl. Algebra 35(2) (1985) 133-149. [8] A. Giambruno, A. Regev, M. Zaicev, On the codimension growth of finite-dimensional Lie algebras, J. Algebra 220(2) (1999) 466-474. [9] A. Giambruno, A. Regev, M. Zaicev, Simple and semisimple Lie algebras and codimension growth, Trans. Amer. Math. Soc. 352(4) (2000) 1935-1946. [10] A. Giambruno, I. Shestakov, M. Zaicev, Finite-dimensional non-associative algebras and codi- mension growth, Adv. in Appl. Math. 47 (2011) 125-139. [11] A. Giambruno, M. Zaicev, On codimension growth of finitely generated associative algebras, Adv. Math. 140(2) (1998) 145-155. [12] A. Giambruno, M. Zaicev, Exponential codimension growth of P.I. algebras: an exact esti- mate, Adv. Math. 142(2) (1999) 221-243. [13] A. Giambruno, M. Zaicev, Polynomial Identities and Asymptotic Methods Mathematical Surveys and Monographs, 122, Amer. Math. Soc. (Providence R.I., 2005). Z2-GRADED CODIMENSIONS OF UNITAL ALGEBRAS 15 [14] A. Giambruno, M. Zaicev, Proper identities, Lie identities and exponential codimension growth. J. Algebra 320(5) (2008) 1933-1962. [15] A. Giambruno, M. Zaicev, On codimension growth of finite dimensional Lie superalgebras, J. London Math. Soc. 95 (2012) 534-548. [16] A. Giambruno, M. Zaicev, Anomalies on codimension growth of algebras, Forum Math. 28(4) (2016) 649-656. [17] M. Lothaire, Algebraic Combinatorics on Words, Encyclopedia Math. Appl., vol. 90 (Cam- bridge University Press, Cambridge, 2002). [18] S. M. Ratseev, Correlation of Poisson algebras and Lie algebras in the language of identities, Translation of Mat. Zametki 96(4) (2014) 567-577. Math. Notes 96(3-4) (2014) 538-547. [19] D. Repovs, M. Zaicev, Numerical invariants of identities of unital algebras, Comm. Algebra 43(9) (2015) 3823-3839. [20] M. Zaitsev, Integrality of exponents of growth of identities of finite-dimensional Lie algebras, (Russian) Izv. Ross. Akad. Nauk Ser. Mat. 66 (2002)(3) 23-48; translation in Izv. Math. 66(3) (2002) 463-487. [21] M. V. Zaitsev, Identities of finite-dimensional unitary algebras,(Russian) Algebra Logika 50(5) (2011) 563-594, 693, 695; translation in Algebra Logic 50(5) (2011) 381-404. [22] M. Zaicev, On existence of PI-exponents of codimension growth, Electron. Res. Announc. Math. Sci. 21 (2014) 113-119. [23] M. V. Zaicev, Graded identities in finite-dimensional algebras of codimensions of identities in associative algebras, Translation of Vestnik Moskov. Univ. Ser. I Mat. Mekh. (5) 2015 54-57. Moscow Univ. Math. Bull. 70(5) (2015) 234-236. [24] M. V. Zaicev, D. Repovs, Exponential growth of codimensions of identities of algebras with unity, (Russian) Mat. Sb. 206(10) (2015) 103-126; translation in Sb. Math. 206(10) (2015) 1440-1462. Dusan D. Repovs, Faculty of Education, and Faculty of Mathematics and Physics, University of Ljubljana, Ljubljana, 1000, Slovenia E-mail address: [email protected] Mikhail V. Zaicev, Department of Algebra, Faculty of Mathematics and Mechanics, Moscow State University, Moscow,119992, Russia E-mail address: [email protected]
1507.02326	1	1507	2015-07-08T22:18:48	Algebras of Jordan brackets and Generalized Poisson algebras	[ "math.RA" ]	We construct a basis of free unital generalized Poisson superalgebras and a basis of free unital superalgebras of Jordan brackets. Also, we prove the analogue of Farkas' Theorem for PI unital generalized Poisson algebras and PI unital algebras of Jordan brackets. Relations beetwen generic Poisson superalgebras and superalgebras of Jordan brackets will be studied.	math.RA	math	Algebras of Jordan brackets and Generalized Poisson algebras Ivan Kaygorodova,b1 a Universidade Federal do ABC, Santo Andre, Brazil. b Sobolev Institute of Mathematics, Novosibirsk, Russia. e-mail: [email protected] Abstract. We construct a basis of free unital generalized Poisson superalgebras and a basis of free unital super- algebras of Jordan brackets. Also, we prove the analogue of Farkas' Theorem for PI unital generalized Poisson algebras and PI unital algebras of Jordan brackets. Relations beetwen generic Poisson superal- gebras and superalgebras of Jordan brackets will be studied. MSC2010: 17C70, 17B01, 17B63 1. Introduction. Many interesting and important results have been obtained about the structure of polynomial algebras, free associative algebras, and free Lie algebras. Although free Poisson algebras are closely connected with these algebras, only a few results on their structure are known. Free Poisson superalgebras were introduced in [1]. For example, the free Poisson algebra of rank three and Poisson brackets were used recently in [2, 3] to prove that the Nagata automorphism of the polynomial algebra F [x, y, z] over a field F of characteristic 0 is wild. Systematic study of free Poisson algebras was initiated in [4], and subsequently continued in [5]-[8]. Generic Poisson algebras were studied in [9]. Other generalizations of Poisson superalgebras are generalized Poisson superalgebras and superalgebras of Jordan brackets. As follows from [10, 11], every Poisson superalgebra is a superalgebra of Jordan brackets. In [12, 13] all identities whose define unital superalgebras of Jordan brackets was described, and in [14] analogue for non-unital case was obtained. Superalgebras of Jordan brackets play important role in classification of simple finite-dimensional Jordan superalgebras [15]. The Kantor construction gives interesting relations between Novikov-Poisson algebras and Jordan superalgebras [16]. Recall the notion of Kantor's Double [11]. Let A = A0 ⊕ A1 be an associative supercommutative superalgebra and { , } : A × A → A -- super-anticommutative multiplication, which we shall call a bracket. Using superalgebra A and bracket {, } we can construct superalgebra K(A). Consider the following direct sum of spaces K(A) = A ⊕ Ax, where Ax is an isomorphic copy of the space A. Let a and b are homogeneous elements in A. Then multiplication ∗ on J(A, {, }) is defined by the relations a ∗ b = ab, a ∗ bx = (ab)x, ax ∗ b = (−1)b(ab)x, ax ∗ bx = (−1)b{a, b}. We set K(A)0 = A0 ⊕ A1x, K(A)1 = A1 ⊕ A0x, so K(A) is a Z2-graded algebra. That is, superalgebra JB is called a superalgebra of Jordan brackets, if the Kantor Double is a Jordan superalgebra. As follows from [12, 13], every superalgebra JB with associative-supercommutative multiplication · and superanticommutative multiplication {, } satisfies {a, bc} = {a, b}c + (−1)abb{a, c} − D(a)bc, {a, {b, c}} = {{a, b}, c} + (−1)ab{b, {a, c}}+ D(a){b, c} + (−1)abcD(b){c, a} + (−1)cabD(c){a, b}, (1) (2) where D(a) = {a, 1}, is a superalgebra of Jordan brackets. Note that D is an even derivation of superal- gebra (JB, ·). If D = 0, then superalgebra (JB, ·, {, }) is a Poisson superalgebra [10]. 1The author was supported by RFBR 15-31-21169 and by FAPESP 11/51132-9 and 14/24519-8. 1 2 V. Kac and N. Cantarini studied linearly-compact simple superalgebras of Jordan brackets [17], I. Kaygorodov and V. Zhelyabin studied δ-superderivations of simple superalgebras of Jordan brackets [18], the relation between Jordan brackets and Poisson brackets was studied in [19] by E. Zelmanov, I. Shestakov and C. Martinez, and in [18, 20] special superalgebras of Jordan brackets were considered. As follows from [17], we can consider new multiplication {a, b}D = {a, b} + 1 2 D, and we obtain new superalgebra with following identities 2 (aD(b) − D(a)b) and DD = 1 {a, bc}D = {a, b}Dc + (−1)abb{a, c}D − DD(a)bc, {a, {b, c}D}D = {{a, b}D, c}D + (−1)ab{b, {a, c}D}D. (3) (4) It is a generalized Poisson superalgebra. All vector spaces are considered over fields with characteristic zero. 2. A basis of free Generalized Poisson superalgebra. The description of a basis of a free algebraic structure is one of the main problems in modern alge- bra. Researchers constructed bases for free Lie algebras [21, 22], free Lie superalgebras [23], partially commutative Lie algebras [24] and so on. Let us now consider free unital generalized Poisson superalgebras. Let DX be a free non-associative super-dialgebra over field k with set of even and odd generators X = X0 ∪ X1 = {1, x1, . . . , xn}, and two multiplications · and {, }. We consider ideal I of super-dialgebra DX , generated by (cid:8) {a, b} + (−1)ab{b, a}, a · b − (−1)abb · a, 1 · a − a, {a, b} · c + (−1)abb · {a, c} − D(a) · b · c − {a, b · c}, {{a, b}, c} + (−1)ab{b, {a, c}} − {a, {b, c}} a, b, c ∈ DX0 ∪ DX1 (cid:9), where D(a) = {a, 1}. We say, that super-dialgebra DX /I is a free unital generalized Poisson superalgebra. Let GenP be a free unital generalized Poisson superalgebra with a set of generators X. We assume, that X is a linearly ordered set and 1 < x1 < . . . < xn. All elements from X are good words. Let we defined all good words with length less than n. The word w with length n in alphabet X is a good word, if: 1) w = {u, v}, where u, v are good words and u > v; 2) if w = {{u1, u2}, v}, then u2 ≤ v. We define the order on good words with length less or equal n such that it be consistent with order to correct for existing words, and satisfy the conditions if word w is a good and w = {u, v}, then w > u, v. Define the set M as M =(cid:8)u, {v, v}u, v are good words in alphabet X, v ∈ GenP1(cid:9). We define a linear order on the set M . Let u > v, if deg(u) > deg(v); if deg(u) = deg(v), then for u = {u1, u2}, v = {v1, v2} (where u1 > u2, v1 > v2) we say u > v if u1 > v1, and if u1 = v1 we say u > v if u2 > v2. The elements of set M we denote as ei, where ei < ej if i < j. Let LX be a free non-associative superalgebra over field k with set of generators X and multiplication {, }L. Consider the ideal J of superalgebra LX , generated by Superalgebra LX /J is a free Lie superalgebra [25]. As proved in [25], the set M is a basis of the free Lie superalgebra LX/J. {{a, b}, c} + (−1)ab{b, {a, c}} − {a, {b, c}} a, b, c ∈ LX0 ∪ LX1 (cid:9). (cid:8) {a, b} + (−1)ab{b, a}, Let K be an ideal in DX generated by elements with type {x · yx, y 6= 1}. It is easy to see, that (DX /I)/K ≅ LX/J. Let U = {ek1 i1 . . . ekn in ei1 6= 1, eim ∈ M, i1 < . . . < in} ∪ {1}. The set U is closed under multiplication ·. The multiplication · is a supercommutative, and we can consider elements of M as ek1 , where ki 6= 0, and if eij ∈ GenP1, then kj = 1. We say that element i1 . . . ekn in 3 ek1 i1 . . . ekn in generators X. The degree of element f is the number of generator elements in word f . has degree dege = P ki. For element ei ∈ M we define degL as the length of word ei with The main goal of this chapter is a description of the basis of free unital generalized Poisson superalge- bras. We will prove that set U is a basis of free unital generalized Poisson superalgebras. Lemma 1. The set U is a basis of free superalgebra GenP as a vector space. Proof. We consider word w with degree 2. It is easy to see that w has type xixj, or {xi, xj }, or {1, xi}, and w ∈ U . We prove that every word with degree n is a linear combination of elements from U . That is every element of superalgebra GenP is a linear combination of elements from U . We will use induction. Let every word with degree less than n be a linear combination of elements from U. Then w = w1w2 or w = {w1, w2}, where w1, w2 are words with degree less than n, that is, w1 and w2 are linear combinations of elements from U . U, αi ∈ k. First case: w1 = Pi α1im1i and w2 = Pi α2im2i, where m1i, m2i ∈ U, α1i, α2i ∈ k. That is, w = Pi,j α1iα2jm1im2j and using closed U under multiplication ·, we can say that w is a Pi αimi, mi ∈ Second case: w1 = Pi α1im1i and w2 = Pi α2im2i, where m1i, m2i ∈ U, α1i, α2i ∈ k. That is, w = Pi,j αiαj {mi, mj}. We consider arbitrary summand {mi, mj}. n1n2, where ni ∈ U, ni 6= 1, then, we can say that mj has type n1n2. Using the relation If we can consider mi or mj as {mi, n1n2} = {mi, n1}n2 + (−1)min1n1{mi, n2} − D(mi)n1n2, and note that degrees of elements {mi, n1}, {mi, n2}, D(mi) less than n and n1, n2 ∈ U, we can say that elements {mi, n1}, {mi, n2}, D(mi) are linear combinations of elements from U , and w is a linear combination of elements from U . If mi and mj is not of type n1n2, where n1, n2 ∈ U and n1, n2 6= 1, then mi, mj ∈ M. That is, if mi = mj ∈ GenP1, then {mi, mj} ∈ M. In the other case, mi 6= mj and we can say that mi > mj and if mi = {m1i, m2i}, m1i ≥ m2i then for m2i ≤ mj it follows that {mi, mj} ∈ M, for m2i > mj it follows that {mi, mj} = {{m1i, mj}, m2i} + (−1)m1imj{m1i, {m2i, mj}}, where {{m1i, mj}, m2i} ∈ M. If {m1i, {m2i, mj}} ∈ M , then we obtained that {mi, mj} is a linear combination of elements from U , or we can consider m1i as {, }-multiplication of elements m11i and m21i. After that, we can continue the process, using identity (4). Note that, degree of mi is greater than degree of m1i, degree of m1i is greater than degree of m11i, and so on. This process will be finite. That is, the lemma is proved. For element b = Πn It is easy to see that k=1etk k we say b em = et1 1 · · · etm−1 m−1 etm−1 m etm+1 m+1 · · · etn n . {a, etk k } = {a, etk−1 }ek + (−1)ae tk −1 k etk−1 k {a, ek} − D(a)etk k = {a, etk−2 k }e2 k + (−1)ae etk−2 k {a, ek}ek + (−1)ae tk −1 k etk−1 x {a, ek} − 2D(a)etk k = k tk −2 k Using this notations, we can conclude = tk{a, ek}etk−1 k . . . − (tk − 1)D(a)etk k . {a, b} = {a, Πn k=1etk k } =(cid:26)a, b etn n (cid:27) etn n + (−1) a (cid:12) (cid:12) (cid:12) (cid:12) b tn n e (cid:12) (cid:12) (cid:12) (cid:12) b etn (a, b etn−1 n−1 etn n ) etn−1 n−1 etn n + (−1) a (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) b tn−1 n−1 e tn n e (cid:12) (cid:12) (cid:12) (cid:12) (cid:12) n (cid:8)a, etn n na, etn−1 n (cid:9) − D(a)b = n−1o etn n + b etn−1 n−1 etn a (−1) (cid:12) (cid:12) (cid:12) (cid:12) b tn n e (cid:12) (cid:12) (cid:12) (cid:12) b etn n (cid:8)a, etn n (cid:9) − 2D(a)b = . . . (−1)et1 1 ···e tk−1 k−1 e tk k tk{a, ek} = n Xk=1 4 b ek n − ( Xk=1 tk − 1)D(a)b. We define linear mapping ∗ : U → U as: 1) if w ∈ X, then w∗ = w; 2) if w ∈ M , then w∗ = w; k , b = Πn 3) if a = Πn k , then k=1eak k=1ebk (ab)∗ = (−1)Pn i=1 eai i e bi+1 i+1 ···ebn n Πn k=1eak+bk k . 4) if we defined mapping ∗ for all words with length less than n, then word w = {w1, w2}, where w1, w2 ∈ M , we can consider as element of free Lie superalgebra (M, {, }) and can define w∗ as a linear combination of basis elements of superalgebra (M, {, }), that is w be a linear combination of elements of M . 5) if we defined mapping ∗ for all words with length less than m, then for word w = {w1, w2} with k , then length m, where w1 or w2 is an element of U \ M , we can say that w2 ∈ U \ M and w2 = Πn k=1etk ({w1, w2})∗ = n Xk=1 (−1)et1 1 ···e tk−1 k−1 e tk ek(cid:19)∗ k tk(cid:18){w1, ek}∗ w2 − n Xk=1 tk − 1! (D(w1)∗w2)∗, where in the right part of expression elements with type {x, y}∗ was defined above. Theorem 2. The set U is a basis of free superalgebra GenP. Proof. We consider set U as a basis of a vector space and define two new multiplications ∗ and {, }∗ by: u ∗ v = (uv)∗ and {u, v}∗ = ({u, v})∗. It is a superalgebra (U, ∗, {, }∗). We want to prove that superalgebra (U, ∗, {, }∗) is a generalized Poisson superalgebra. After that we can say that U is a basis of free generalized Poisson superalgebra GenP. It is easy to see that multiplication ∗ is associative and supercommutative, and multiplication {, }∗ is superanticommutative. We want to prove that superalgebra (U, ∗, {, }∗) satisfies {a, b ∗ c}∗ = {a, b}∗ ∗ c + (−1)abb ∗ {a, c}∗ − D(a)∗ ∗ b ∗ c. (5) We say that a, b, c ∈ U and will prove by induction on dege(a) + dege(b) + dege(c). Note that {1, ei ∗ ej}∗ = {1, ei}∗ ∗ ej + ei ∗ {1, ej}∗, {ek, ei ∗ ej}∗ = {ek, ei}∗ ∗ ej + (−1)ekeiei ∗ {ek, ej}∗ − D(ek)∗ ∗ ei ∗ ej by the definition of mapping ∗. That is, if dege(a) + dege(b) + dege(c) < 4, then identity (5) is true. Let dege(a) + dege(b) + dege(c) > 3. We can say that b = Πn k=1ebk k , c = Πn k=1eck k and define new notations: That is, γ(b, c) = n Xi=1 We can see that i eci+1 ebi i+1 · · · ecn n and µ(b, c, k) = eb1+c1 1 · · · ebk−1+ck−1 k−1 ebk+ck k . D(a) ∗ b ∗ c = (−1)γ(b,c)D(a) ∗ Πn k=1ebk+ck k (−1)γ(b,c){a, b ∗ c}∗ = . (6) (7) (−1)µ(b,c,k)(bk + ck){a, ek}∗ ∗ Πn m=1ebm+cm m ek 5 − n (bk + ck) − 1! D(a)∗ ∗ Πn Xk=1 m=1ebm+cm m . n Xk=1 It is easy to see that {a, b}∗ ∗ c = (−1)eb1 1 ···e bk−1 k−1 e bk k bk{a, ek}∗ ∗ n Xk=1 b ek ∗ c − n Xk=1 bk − 1! D(a)∗ ∗ b ∗ c = (−1)eb1 1 ···e bk−1 k−1 e bk k +Pn i=1 ebi i e ci+1 i+1 ···ecn n −ekec1 1 ···e ck−1 k−1 bk{a, ek}∗ ∗ Πn m=1ebm+cm m ek −(−1)Pn i=1 e bi i e ci+1 i+1 ···ecn n n Xk=1 bk − 1! D(a)∗ ∗ Πn m=1ebm+cm m . n Xk=1 That is, (−1)γ(b,c) n Xk=1 Note that, (−1)µ(b,c,k)bk{a, ek}∗ ∗ {a, b}∗ ∗ c = Πn m=1ebm+cm m ek − n Xk=1 bk − 1! D(a)∗ ∗ Πn m=1ebm+cm m (8) ! . (−1)cb{a, c}∗ ∗ b = (−1)cb (−1)ec1 1 ···e ck−1 k−1 e ck k ck{a, ek}∗ ∗ n Xk=1 c ek ∗ b − n Xk=1 ck − 1! D(a)∗ ∗ b ∗ c = n Xk=1 That is, (−1)e ck k b+ekec1 1 ···e ck−1 k−1 +Pn i=1 ebi i e ci+1 i+1 ···ecn n −eke bk+1 k+1 ···ebn n bk{a, ek}∗ ∗ Πn m=1ebm+cm m ek −(−1)Pn i=1 e bi i e ci+1 i+1 ···ecn n n Xk=1 ck − 1! D(a)∗ ∗ Πn m=1ebm+cm m . (−1)γ(b,c) n Xk=1 (−1)µ(b,c,k)ck{a, ek}∗ ∗ (−1)cb{a, c}∗ ∗ b = Πn k=1ebm+cm m ek − n Xk=1 ck − 1! D(a)∗ ∗ Πn m=1ebm+cm m (9) ! . In the end, summing (6) and (7) and subtracting (8) and (9), we can obtain that superalgebra (U, ∗, {, }∗) satisfies (5). We want to prove that the superalgebra satisfies the identity {a, {b, c}∗}∗ = {{a, b}∗, c}∗ + (−1)ab{b, {a, c}∗}∗. (10) If a, b, c ∈ M , then identity (10) is true, because (M, {, }∗) is a free Lie superalgebra. Let a = a1 ∗ a2, where a1, a2 6= 1 and a1, a2 ∈ U , then using (5), we can obtain {{a1 ∗ a2, b}∗, c}∗ + (−1)ab{b, {a1 ∗ a2, c}∗}∗ = −(−1)ba{{b, a1}∗ ∗ a2 + (−1)ba1a1 ∗ {b, a2}∗ − D(b)∗a, c}∗− (−1)a(b+c){b, {c, a1}∗ ∗ a2 + (−1)ca1a1 ∗ {c, a2}∗ − D(c)∗ ∗ a1 ∗ a2}∗ = 6 (−1)ba+c(a+b)({c, {b, a1}∗}∗ ∗ a2 + (−1)(b+a1)c{b, a1}∗ ∗ {c, a2}∗ − D(c)∗ ∗ {b, a1}∗ ∗ a2)+ (−1)ba+c(b+a)+ba1({c, a1}∗ ∗ {b, a2}∗ + (−1)ca1a1 ∗ {c, {b, a2}∗}∗ − D(c)∗ ∗ a1 ∗ {b, a2}∗)+ (−1)ba+c(a+b)({c, D(b)∗}∗ ∗ a + (−1)cbD(b)∗ ∗ {c, a}∗ − D(c)∗ ∗ D(b)∗ ∗ a)− (−1)a(b+c)({b, {c, a1}∗}∗ ∗ a2 + (−1)b(c+a1){c, a1}∗ ∗ {b, a2}∗ − D(b)∗ ∗ {c, a1}∗ ∗ a2)− (−1)a(b+c)+ca({b, a1}∗ ∗ {c, a2}∗ + (−1)ba1a1 ∗ {b, {c, a2}∗}∗ − D(b)∗ ∗ a1 ∗ {c, a2}∗)− (−1)a(b+c)({b, D(c)∗}∗ ∗ a + (−1)bcD(c)∗ ∗ {b, a}∗ − D(b)∗ ∗ D(c)∗ ∗ a) = −(−1)a(b+c)({{b, a1}∗, c}∗ + (−1)ba1{b, {c, a1}∗}∗) ∗ a2 +(−1)(b+c)a1a1 ∗ ({{b, a2}∗, c}∗ + (−1)ba2{b, {c, a2}∗}∗) − {D(b)∗, c}∗ ∗ a − {b, D(c)∗}∗ ∗ a = −(−1)abc({{b, c}∗, a1}∗ ∗ a2 + (−1)bca1a1 ∗ {{b, c}∗, a2}∗ − D({b, c}∗)∗ ∗ a = {a1 ∗ a2, {b, c}∗}∗. If a ∈ M , b ∈ U \ M or c ∈ U \ M , then designating a, b, c, we can use the given calculations. That is, the superalgebra satisfies (10) and it follows that the theorem is proved. Let GenP (n) be a space of multi-homogeneous components of free generalized Poisson algebra GenP with n + 1 generators 1, x1, . . . , xn. Then Theorem 3. dim(GenP (n)) = n · n!. Proof. Note that, we consider the vector space of free algebra GenP [1, x1, . . . , xn] as a subspace in the free Poisson algebra P [1, x1, . . . , xn] with n + 1 generators. It is well-known that the dimension of the space of multi-homogeneous components of the free Poisson algebra with n + 1 generators is a (n + 1)!. Using the description of basis of free generalized Poisson algebra (Theorem 2), we can define the structure of the basis of the space of multi-homogeneous components. That is, the basis of the space of multi-homogeneous components of the free generalized Poisson algebra with generators 1, x1, . . . , xn and the basis of the space of multi-homogeneous components of the free Poisson algebra with generators 1, x1, . . . , xn, has distinction only for elements 1 · y, where y is an arbitrary element from the basis of the free Poisson algebra with generators x1, . . . , xn. That is, dim(GenP (n)) = (n + 1)! − n! = n · n!. The Theorem is proved. 3. A basis of free superalgebra of Jordan brackets. Let us now consider free superalgebras of Jordan brackets. Let DX be a free non-associative super- dialgebra over field k with set of even and odd generators X = X0 ∪ X1 = {1, x1, . . . , xn}, and two multiplications · and {, }. We consider ideal I of super-dialgebra DX , generated by (cid:8) {a, b} + (−1)ab{b, a}, a · b − (−1)abb · a, 1 · a − a, {a, b} · c + (−1)abb · {a, c} − D(a) · b · c − {a, b · c}, {{a, b}, c} + (−1)ab{b, {a, c}} − {a, {b, c}} − D(a){b, c} − (−1)abcD(b){c, a} − (−1)cabD(c){a, b} where D(a) = {a, 1}. We say, that super-dialgebra DX /I is a free unital superalgebra of Jordan brackets. Let JB be a free unital superalgebra of Jordan brackets with set of generators X. We assume, that X is a linearly ordered set and 1 < x1 < . . . < xn. All elements from X are good words. Let we defined all good words with length less than n. The word w with length n in alphabet X is a good word, if: a, b, c ∈ DX0 ∪ DX1 (cid:9), 1) w = {u, v}, where u, v are good words and u > v; 2) if w = {{u1, u2}, v}, then u2 ≤ v. The order on good words with length less or equal n, we define as that it be consistent with order to correct for existing words, and satisfy the conditions if word w is a good and w = {u, v}, then w > u, v. Define the set M as M =(cid:8)u, {v, v}u, v are good words in alphabet X, v ∈ JB1(cid:9). We define a linear order on the set M . Let u > v, if deg(u) > deg(v); if deg(u) = deg(v), then for u = {u1, u2}, v = {v1, v2} (where u1 > u2, v1 > v2) we say u > v if u1 > v1, and if u1 = v1 we say u > v if u2 > v2. The elements of set M we denote as ei, where ei < ej if i < j. Let LX be a free non-associative superalgebra over field k with set of generators X and multiplication {, }L. Consider ideal J of superalgebra LX , generated by 7 Superalgebra LX /J is a free Lie superalgebra [25]. As proved in [25], the set M is a basis of a free Lie superalgebra LX/J. {{a, b}, c} + (−1)ab{b, {a, c}} − {a, {b, c}} a, b, c ∈ LX0 ∪ LX1 (cid:9). (cid:8) {a, b} + (−1)ab{b, a}, Let K be an ideal in DX generated by elements with type {x · yx, y 6= 1}. Easy to see, that (DX /I)/K ≅ LX/J. Let U = {ek1 i1 . . . ekn in ei1 6= 1, eim ∈ M, i1 < . . . < in} ∪ {1}. The set U is closed under multiplication ·. The multiplication · is a supercommutative, and we can consider elements of M as ek1 , where ki 6= 0, and if eij ∈ JB1, then kj = 1. We say that element ek1 i1 . . . ekn in X. The degree of element f is a number of generator elements in word f . has degree dege =P ki. For element ei ∈ M we define degL as length of word with generators The main goal of this chapter is a description of basis of free superalgebra of Jordan brackets. We will i1 . . . ekn in prove that set U is a basis of free superalgebra of Jordan brackets. Lemma 4. The set U is a basis of free superalgebra JB as a vector space. Proof. Similar to proof of Lemma 1. We define linear mapping ∗ : U → U as: 1) if w ∈ X, then w∗ = w; 2) if w ∈ M , then w∗ = w; 3) if a = Πn k , b = Πn k=1eak k=1ebk k , then (ab)∗ = (−1)Pn i=1 eai i e bi+1 i+1 ···ebn n Πn k=1eak+bk k . 4) if we defined mapping ∗ for all words with length less than n and for all elements u with u < w. Then word w = {w1, w2} with w1 = {w11, w12}, where w1, w2, w11, w12 ∈ M and {w11, w12} = w1, w11, w12 > w2, we can consider as element of superalgebra (M, {, }) and using (2) we define w∗ as a linear combination of elements from W =(cid:8){{w11, w2}, w12}, {w12, w2}, w11}, D(w2) · {w11, w12}, D(w11) · {w2, w12}, D(w12) · {w2, w11}(cid:9), where either u ∈ W and u < w or u ∈ W and u = u1 · u2, where u1, u2 < w. It means, that for every u ∈ W we can define u∗. 5) if we defined mapping ∗ for all words with length less than m, then for word w = {w1, w2} with k , then length m, where w1 or w2 is an element of U \ M , we can say that w2 ∈ U \ M and w2 = Πn k=1etk ({w1, w2})∗ = n Xk=1 (−1)et1 1 ···e tk−1 k−1 e tk ek(cid:19)∗ k tk(cid:18){w1, ek}∗ w2 − n Xk=1 tk − 1! (D(w1)∗w2)∗ , where in the right part of expression elements with type {x, y}∗ was defined above. Theorem 5. The set U is a basis of free superalgebra JB. Proof. Similar to proof of Theorem 2. Let JB(n) be a space of multi-homogeneous components of free unital algebra of Jordan brackets JB with n + 1 generators 1, x1, . . . , xn. Then Theorem 6. dim(JB(n)) = n · n!. 8 Proof. Similar to proof of Theorema 3. Using Theorem 2 and 5 and [17], we can conclude Theorem 7. The unital superalgebra of Jordan brackets (A, ·, {, }) is a free (or a simple) if and only if the unital generalized Poisson superalgebra (A, ·, {, }D) is a free (or a simple). 4. Analogues of Farkas' Theorem for PI algebras. The celebrated Amitsur-Levitsky Theorem says that Mk(R), the k × k matrices over a commutative ring R, satisfies the identity s2k = 0. The analogue of Amitsur-Levitsky Theorem for matrix superalgebras was proved in [26]. Furthermore, by a well-known theorem of Amitsur, any associative PI algebra satisfies the condition that some l-th power of some k-th standard polynomial sk is zero: (sk)l = 0. Poisson PI algebras were studied in [27, 28, 29, 30, 31]. D. Farkas defined cσ{xσ(1), xσ(2)} . . . {xσ(2i−1), xσ(2i)} g = Xσ∈Sm as customary polynomials and proved that every Poisson PI algebra satisfies some customary identity [27]. Note that, in [9] was proved the Theorem of Farkas for generic Poisson algebras. For generalized Poisson algebras and algebras of Jordan brackets, the analogue of the customary identities is g∗ = [m/2] Xi=0 Xσ∈Sm where cσ,ihxσ(1), xσ(2)i . . . hxσ(2i−1), xσ(2i)iD(xσ(2i+1)) . . . D(xσ(m)), (11) It is easy to see that algebra (JB, ·, h, i) satisfies identities hx, yi := {x, y} − (D(x)y − xD(y)). hx, yi = −hy, xi, hx, y · zi = hx, yi · z + y · hx, zi, hhx, yi, zi + hhy, zi, xi + hhz, xi, yi = D(x){y, z} + D(y){z, x} + D(z){x, y}. That is, (JB, ·, h, i) is not a Poisson algebra, but it is a generic Poisson algebra (see [?]). In [20] it was proved that if Kantor's double of superalgebra (JB, ·, {, }) is a special Jordan superalgebra, then (JB, ·, h, i) satisfies identity hha, bi, ci = D(a)hb, ci + (−1)abD(b)ha, ci. Using ideas from [27], we will prove an analogue of Farkas' Theorem. Namely, that every PI unital generalized Poisson algebra and PI unital algebra of Jordan brackets satisfies some special identity like (11). For homogeneous Lie word f with letters x1, x2, . . . , xn we define degree as associative monomial degx1 (f ) x 1 . . . xdegxn (f ) n . Introduce additional useful notation {x1, x2, . . . , xn} := {. . . {x1, x2}, . . . , xn}, Following [22, 23], it is easy to see that: Lemma 8. Let L be a free Lie algebra with generators 1, y, x1, . . . , xn. The basis of the space with components with degree 1kyx1 . . . xn is a set of elements with type where yi = xi, 0 < i < n + 1, yi = 1, n < i < n + k + 1. {y, yσ(1), . . . , yσ(n+k)}, Lemma 9. With notation as above, {yz, w1, . . . , wn} = X ±{y, wi1(1), . . . , wi1(k1)}{z, wi2(1), . . . , wi2(k2)}{1, wi3(1), . . . , wi3(k3)} . . . {1, wil(1), . . . , wil(kl)}, summed over all possibilities for which each subscript appears exactly once in each summand and all se- quences {i1(1), . . . , i1(k1)}, . . . , {il(1), . . . , il(kl)} are monotone increasing, all elements ij(t) are different and 1 ≤ ij(t) ≤ n, also k1 + . . . + kl = n. (In the case of p = 0, the singleton {x} is identified as x.) 9 Proof. Easy to see by induction. Lemma 10. With notation as above, {w1, . . . , wr, yz, wr+1, . . . , wr+n} = X( ± {w0, y, wi1(1), . . . , wi1(k1)}{z, wi2(1), . . . , wi2(k2)}{1, wi3(1), . . . , wi3(k3)} . . . {1, wil(1), . . . , wil(kl)} ± {y, wi1(1), . . . , wi1(k1)}{w0, z, wi2(1), . . . , wi2(k2)}{1, wi3(1), . . . , wi3(k3)} . . . {1, wil(1), . . . , wil(kl)} ± {y, wi1(1), . . . , wi1(k1)}{z, wi2(1), . . . , wi2(k2)}{D(w0), wi3(1), . . . , wi3(k3)} . . . {1, wil(1), . . . , wil(kl)}), where w0 = {w1, . . . , wr}. Proof. If r = 0 this lemma follows immediately from the previous one. So we assume that r ≥ 1 and for β = {yz, w0, wr+1, . . . , wr+n} we apply the previous lemma formally to this new expression. Easy calculations give the proof of the lemma. Lemma 11. Assume f is a multi-linear Lie polynomial in the letters x, w1, . . . , wm, and f is an asso- ciative algebraic derivation in x when considered as a Poisson polynomial. Then deg(f ) = 2. Proof. Let f have a degree 1kxw1 . . . wm. Set wi = 1, m < i < m + k + 1 and x = wm+k+1. By Lemma 8, we may write where σ runs over permutations of 2, . . . , m + k + 1 and cσ is a scalar. We compute the derivation difference of h with respect to wm+k+1 → yz. h =X cσ{w1, wσ(2), . . . , wσ(m+k+1)}, One of the many summands will be the product of cσ and {w1, wσ(2), . . . , wσ(t−1), y}{z, wσ(t+1), . . . , wσ(m+k+1)}. And one of the many others summands will be product of cσ and yzD{w1, wσ(2), . . . , wσ(m+k)}. But this expressions determines σ; it does not appear in the derivation difference for any other permutation. If we specify an ordering of the variables then these special expressions are linearly independent for deg(f ) > 2 by the application following Lemma 8. The Lemma is proved. y < z < w1 < . . . < wn, For polynomial f (x, . . .) we define ∆x yz(f ) := f (yz, . . .) − yf (z, . . .) − zf (y, . . .). ∆x It is easy to see, that if ∆x yz(f ). For Poisson polynomial f we define x-height(f ) as maximal length of {, }-subword with x. yz(f ) 6= 0 and algebra A satisfies identity f , then A also satisfies identity Lemma 12. If multi-linear Poisson polynomial f is an associative derivation in any letter x, then x- height(f ) < 3. γ , where the degree of polynomial f k Proof. Let f = Pγ fγγ, where fγ be a Lie polynomials with degree 1kxt for any t. We consider fγ =Pk f k γ we choose a polynomials with maximum deg(t) and after that polynomial with the maximum k. Let it is f ∗ with degree 1kxx1 . . . xn. Continuing the argument similar to that of Lemma 11, we obtain that if deg(f ∗) > 2, then ∆x yz(f ) 6= 0. Following, deg(fγ) < 3 and x-height(f ) < 3. The Lemma is proved. γ is 1kxt. On polynomials f k 10 The main result of this chapter is following theorem. Theorem 13. If unital generalized Poisson algebra A satisfies polynomial identity g0, then A satisfies polynomial identity g∗ of type (11). Proof. Using ideas from [27], we consider three steps. Step 1. Let x-height(g0) > 2. Then algebra A satisfies identity g1 = ∆x yz(g0). By Lemma 12, g1 6= 0. Note that x-height(g0) > y-height(g1), z-height(g1). If polynomial g1 has letter w with w-height(g1) > 2, then we will continue the process. That is, in the end we can obtain that algebra A satisfies some identity f , where arbitrary letter v has v-height(f ) < 3. Step 2. Let algebra A satisfy identity f and x-height(f ) < 3 for arbitrary letter x from f . Assume that f is not an associative derivation for any letter x. Then We consider f = xT + D(x)T0 +X{x, xi}Ti. yz(f ) = −yzT + yzX D(xi)Ti 6= 0. ∆x It is easy to see that unital algebra A will satisfy identity f ∗ = −T +P D(xi)Ti, with less letters than identity f . That is, if identity f ∗ is not a derivation for any letter t, then we can continue this process. The process is finite, and algebra A satisfies some identity g with x-height(g) < 3, and g is an associative derivation on every letter. Step 3. For polynomial g with letters x1, . . . , xn, every associative subword with type {x, y} we can consider as hx, yi+ (D(x)y − xD(y)). It follows that polynomial g we can consider as g∗ + D(x1)T1 + x1R1, where g∗ has type (11). It is easy to see that g∗ is an associative derivation on every letter, then algebra A will satisfy identity R1. It follows that A satisfies identity g∗ + D(x1)T1. We will consider D(x1)T1 as D(x2)T2 + x2R2. That is, algebra A satisfies an identity g∗ + D(x1)D(x2)T1,2. Continuing this process, we find that algebra A satisfies identity g∗ of the form (11). The Theorem is proved. As follows, we can obtain Corollary 14. Every PI generalized Poisson algebra satisfies an identity of the following type [m/2] Xi=0 Xσ∈Sm cσ,i f∗ = where [u1; u2; w1; w2] = i m−2i [xσ(2k−1); xσ(2k); z2k−1; z2k] · Yk=1 Yk=1 {xσ(2i+k); z2i+2k−1; z2i+2k} · z2m−2i+k, (12) 2i Yk=1 {u1, u2}w1w2 + {u1, w1w2}u2 + u1{w1w2, u2} + Xσ1,σ2∈S2 {uσ1(1), wσ2(1)}uσ1(2)wσ2(2), {t1; t2; t3} = {t2t3, t1} − {t2, t1}t3 − {t3, t1}t2. Proof. By theorem 13, every PI generalized Poisson algebra satisfies an identity g∗ of type (11). Multiplying g∗ and Π2m k=1zk, note that It is easy to see the corollary is proved. w1w2hu1, u2i = [u1; u2; w1; w2] and t2t3D(t1) = {t1; t2; t3}. Also, we can prove an analogue of Farkas' Theorem for PI algebras of Jordan brackets: Theorem 15. If unital algebra of Jordan brackets A satisfies polynomial identity g0, then A satisfies polynomial identities of types (11) and (12). Proof. We consider generalized Poisson algebra AD with vector space A and multiplications · and {, }D. It is easy to see, that algebra AD is a PI algebra. Using Theorem 13, we can obtain that algebra AD satisfies a polynomial identity g of type (11). Noted that, 2DD = D and {x, y}D + xDD(y) − DD(x)y = {x, y} + xD(y) − D(x)y, we conclude that algebra A satisfies a polynomial identity of type (11). And using proof of Corollary 14, we find that algebra A satisfies polynomial identity of type (12). The Theorem is proved. 11 5. Kantor's Double for generic Poisson superalgebras. Let us give the definition of a generic Poisson superalgebra [9]. We consider a vector space GP with associative supercommutative multiplication x · y and superanticommutative multiplication {x, y}. If superalgebra (GP, ·, {, }) satisfies the Leibniz identity, i.e., {x, y · z} = {x, y} · z + (−1)xyy · {x, z}, (13) then GP is a generic Poisson superalgebra (GP superalgebra). It is easy to see that if (GP, {, }) is a Lie superalgebra, then (GP, ·, {, }) is a Poisson superalgebra. It is well-known that Kantor's Double of a Poisson superalgebra is a Jordan superalgebra [10]. There naturally arises a Question. When is Kantor's double of a generic Poisson superalgebra a Jordan superalgebra? We gave the answer on this question in following Theorem 16. Let A be a GP superalgebra over field F with characteristic 6= 2, 3, then K(A) is a Jordan superalgebra if and only if A satisfies special identity ({{a, b}, c} − (−1)bc{{a, c}, b} − {a, {b, c}})d = 0 (14) for homogeneous elements a, b, c, d ∈ GP0 ∪ GP1. Proof. In [14] it was proved that if (B, ·) is an associative supercommutative superalgebra and (B, {, }) is a superanticommutative superalgebra, then K(B) is a Jordan superalgebra if and only if for homogeneous elements superalgebra B satisfies following identities (−1)(k+j)i({hkkl, gj}fi − hkkl{gj, fi}) = (−1)(l+j)k({klfi, gj}hk − klfi{gj, hk}), (−1)(i+j)l({fihkgj, kl} − fihk{gj, kl}) = (−1)(k+j)i({hkkl, gj}fi − {hkkl, gjfi})+ (−1)(l+j)k({klfi, gj}hk − {klfi, gjhk}), (−1)(i+j)l{{fi, hk}gj, kl} + (−1)(k+j)i{{hk, kl}gj, fi}+ (−1)(l+j)k{{kl, fi}gj, hk} = (−1)(i+j)l{fi, hk}{gj, kl}+ (−1)(k+j)i{hk, kl}{gj, fi} + (−1)(l+j)k{kl, fi}{gj, hk}, (15) (16) (17) where ki, hi, gi, fi ∈ Bi. It is easy to see that if an algebra satisfies identity (13), then it satisfies identities (15, 16). Using identity (13) for (17), we get (14). The Theorem is proved. Corollary 17. If A is a unital GP (non-Poisson) superalgebra, then K(A) is not a Jordan superalgebra. Corollary 18. If A is a GP (non-Poisson) superalgebra and (A, ·) is an algebra with trivial annihilator (for example, prime algebra), then K(A) is not Jordan superalgebra. Example 19. Let A be a non-Lie anticommutative algebra with multiplication {, } (for example, a Malcev or a binary-Lie algebra). We define new multiplication x · y = 0. Then, algebra (A, ·, {, }) is a GP (non- Poisson) algebra, and satisfies identity (14), and K(A) is a Jordan superalgebra. Acknowledgments: I am grateful to Prof. Dr. Ivan Shestakov (IME-USP, Brazil) for the idea of this work and some interesting comments. References 12 [1] Shestakov I., Quantization of Poisson superalgebras and the specialty of Jordan superalgebras of Poisson type, Algebra and Logic, 32, 1993, no. 5, 309-317. [2] Shestakov I., Umirbaev U., Poisson brackets and two-generated subalgebras of rings of polynomials, J. Amer. Math. Soc., 17, 2004, no. 1, 181-196. [3] Shestakov I., Umirbaev U., The tame and the wild automorphisms of polynomial rings in three variables, J. Amer. Math. Soc., 17, 2004, no. 1, 197-227. [4] Makar-Limanov L., Umirbaev U., Centralizers in free Poisson algebras, Proc. Amer. Math. Soc., 135, 2007, no. 7, 1969-1975 [5] Makar-Limanov L., Turusbekova U., Umirbaev U., Automorphisms and derivations of free Poisson algebras in two variables, J. Algebra, 322, 2009, no. 9, 3318-3330. [6] Makar-Limanov L., Shestakov I., Polynomial and Poisson dependence in free Poisson algebras and free Poisson fields, J. Algebra, 349, 2012, 372-379. [7] Makar-Limanov L., Umirbaev U., The Freiheitssatz for Poisson algebras, J. Algebra, 328, 2011, 495-503. [8] Umirbaev U., Universal enveloping algebras and universal derivations of Poisson algebras, J. Algebra, 354, 2012, 77-94. [9] Kolesnikov P., Makar-Limanov L., Shestakov I., The Freiheitssatz for generic Poisson algebras, Symmetry Integrability and Geometry-Methods and Applications, 10, 2014, 115. [10] Kantor I. L., Connection between Poisson brackets and Jordan and Lie superalgebras, Lie Theory, Differential Equations and Representation Theory (Montreal, 1989), Univ. Montreal, Montreal, QC, 1990, 213-225. [11] Kantor I. L., Jordan and Lie superalgebras determined by a Poisson algebra, Amer. Math. Soc. Transl., 151, 1992, 55-80. [12] King D., McCrimmon K., The Kantor construction of Jordan superalgebras, Comm. Algebra, 20, 1992, no. 1, 109-126. [13] King D., McCrimmon K., The Kantor doubling process revisited, Comm. Algebra, 23, 1995, no. 1, 357-372. [14] Kaygorodov I., Generalized Kantor Double, Vestnik of Samara State Univ., 78, 2010, no. 4, 42-50. [15] Martinez C., Zelmanov E., Simple finite-dimensional Jordan superalgebras of prime characteristic, J. Algebra, 236, 2001, no. 2, 575-629. [16] Zakharov A. Novikov-Poisson algebras and superalgebras of Jordan brackets, Comm. Algebra, 42, 2014, no. 5, 2285- 2298. [17] Cantarini N., Kac V., Classification of linearly compact simple Jordan and generalized Poisson superalgebras, J. Algebra, 313, 2007, 100-124. [18] Kaygorodov I., Zhelyabin V., On δ-superderivations of simple superalgebras of Jordan brackets, St. Peterburg Math. J., 23, 2012, no. 4, 665-677. [19] Martinez C., Shestakov I., Zelmanov E., Jordan superalgebras defined by brackets, J. London Math. Soc. (2), 64, 2001, 357-368. [20] Shestakov I., On speciality of Jordan brackets, Algebra and Discrete Math., 2009, no. 3, 94-101. [21] Shirshov A., On the bases of a free Lie algebra (Russian), Algebra i Logika Sem., 1, 1962, no. 1, 14-19. [22] Chibrikov E., The right-normed basis of a free Lie superalgebra and Lyndon-Shirshov words, Algebra and Logic, 45, 2006, no. 4, 261-276. [23] Chibrikov E., A right normed basis for free Lie algebras and Lyndon-Shirshov words, J. Algebra, 302, 2006, no. 2, 593-612. [24] Poroshenko E., Bases for partially commutative Lie algebras, Algebra and Logic, 50, 2011, no. 5, 405-417. [25] Shtern A., Free Lie superalgebras, Siberian Math. J., 27, 1986, no. 1, 136-140. [26] Samoilov L., An analogue of the Amitsur-Levitzki theorem for matrix superalgebras, Sib. Math. J., 51, 2010, no. 3, 491-495. [27] Farkas D., Poisson polynomial identities, Comm. Algebra, 26, 1998, no. 2, 401-416. [28] Farkas D., Poisson polynomial identities. II, Arch. Math. (Basel), 72, 1999, no. 4, 252-260. [29] Mishchenko S., Petrogradsky V., Regev A., Poisson PI algebras, Trans. Amer. Math. Soc., 359, 2007, no. 10, 4669-4694. [30] Ratseev S., Poisson algebras of polynomial growth, Sib. Math. J., 54, 2013, no. 3, 555-565 [31] Pozhidaev A., Poisson and Filippov superalgebras, Sib. Math. J., 56, 2015, no. 3, 505-515
1908.09179	1	1908	2019-08-24T18:12:08	Commutators, matrices and an identity of Copeland	[ "math.RA", "math.CO" ]	Given two elements $a$ and $b$ of a noncommutative ring, we express $\left( ba\right)^n$ as a "row vector times matrix times column vector" product, where the matrix is the $n$-th power of a matrix with entries $\dbinom{i}{j}\operatorname{ad}_a^{i-j}\left( b\right)$. This generalizes a formula by Tom Copeland used in the study of Pascal-style matrices.	math.RA	math	Commutators, matrices and an identity of Copeland 9 1 0 2 Darij Grinberg August 27, 2019 g u A 4 2 ] . A R h t a m [ 1 v 9 7 1 9 0 . 8 0 9 1 : v i X r a Abstract. Given two elements a and b of a noncommutative ring, we express (ba)n as a "row vector times matrix times column vector" product, where the matrix is the n-th power of a matrix with entries (b). This generalizes a formula by Tom Copeland used in the j(cid:19) adi−j (cid:18)i a study of Pascal-style matrices. Contents 1. Introduction 2. The general formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1. Standing notations . . . . . 2.2. Conventions about matrices . . . 2.3. Conventions about infinite matrices . . 2.4. The matrices S and Ub and the vectors Hc and ej . . . . . . . . 2.4.1. 2.4.2. m and a . . . . . . 2.4.3. The matrix S . . . . . . 2.4.4. The matrix Ub . . . . . 2.4.5. The column vector Hc . 2.4.6. The column vector ej . . . . Iverson brackets (truth values) . . . . . . 2.5. The general formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. The proof . . . . 3.1. The idea . . . 3.2. A lemma about ada . 3.3. Formulas for eT . 3.4. Proving eT i A . u SHc = eT . . . u Hac for u + 1 < m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 2 3 3 4 4 7 7 7 7 8 9 9 10 11 11 12 14 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Commutators, matrices and an identity of Copeland page 2 3.5. Proving Ub Hc = Hbc 3.6. The ≡ relations . . k . . . . . 3.7. Proof of Theorem 2.7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. A Weyl-algebraic application . . . . . . . . . 4.1. The claim . . . 4.2. How derivatives appear in commutators . 4.3. Proofs of Proposition 4.3 and Theorem 4.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 18 21 23 23 25 28 1. Introduction In [MO337766], Tom Copeland stated a formula for the n-th power of a differential operator. Our goal in this note is to prove a more general version of this formula, in which differential operators are replaced by arbitrary elements of a noncommu- tative ring. In a nutshell, this general result (Theorem 2.7) can be stated as follows: If n ∈ N and m ∈ N ∪ {∞} satisfy n < m, and if a and b are two elements of a (noncommu- tative) ring L, then (ba)n = eT 0 (UbS)n H1, where the column vectors e0 and H1 of size m are defined by e0 =   1 0 0 ... 0   and H1 = a0 a1 a2 ... am−1   ,   Commutators, matrices and an identity of Copeland page 3 and where the m × m-matrices S and Ub are defined by 0 0 0 ... 0   and ... ... 0 1 0 · · · 0 0 1 · · · 0 0 0 · · · ... . . . 0 0 0 · · ·    0≤i<m, 0≤j<m 0 0 b ... · · · · · · · · · . . . (b) · · · 0 0 0 ... b   (b) (cid:18)m − 1 2 (cid:19) adm−3 a S = ([j = i + 1])0≤i<m, 0≤j<m = j(cid:19) adi−j (cid:18)i a 0, (b) , b if i ≥ j; if i < j 0 b   Ub =   = ada (b) ad2 a (b) ... adm−1 a (b)  2 ada (b) ... (m − 1) adm−2 a (using the standard Lie-algebraic notation ada for the operator L → L, c 7→ ac − ca). (We shall introduce all these notations in more detail below.) Acknowledgments DG thanks the Mathematisches Forschungsinstitut Oberwolfach for its hospitality during part of the writing process. 2. The general formula 2.1. Standing notations Let us start by introducing notations that will remain in place for the rest of this note: • Let N denote the set {0, 1, 2, . . .}. • "Ring" will always mean "associative ring with unity". Commutativity is not required. • Fix a ring L. • For any two elements a and b of L, we define an element [a, b] of L by This element [a, b] is called the commutator of a and b. [a, b] = ab − ba. Commutators, matrices and an identity of Copeland page 4 • For any a ∈ L, we define a map ada : L → L by (ada (b) = [a, b] for all b ∈ L) . Clearly, this map ada is Z-linear. 2.2. Conventions about matrices In the following, we will use matrices. We shall use a slightly nonstandard conven- tion for labeling the rows and the columns of our matrices: Namely, the rows and the columns of our matrices will always be indexed starting with 0. That is, a k × ℓ- matrix (for k ∈ N and ℓ ∈ N) will always have its rows numbered 0, 1, . . . , k − 1 and its columns numbered 0, 1, . . . , ℓ − 1. In other words, a k × ℓ-matrix is a fam- ily (cid:0)ai,j(cid:1)0≤i<k, 0≤j<ℓ indexed by pairs (i, j) of integers satisfying 0 ≤ i < k and 0 ≤ j < ℓ. We let Lk×ℓ denote the set of all k × ℓ-matrices with entries in L. If A is any k × ℓ-matrix (where k and ℓ belong to N), and if i and j are any two integers satisfying 0 ≤ i < k and 0 ≤ j < ℓ, then we let Ai,j denote the (i, j)-th entry of A. Thus, any k × ℓ-matrix A satisfies A =  A0,0 A1,0 ... A0,1 A1,1 ... Ak−1,0 Ak−1,1 A0,ℓ−1 A1,ℓ−1 · · · · · · . . . · · · Ak−1,ℓ−1 ... .   If k ∈ N, then a column vector of size k means a k × 1-matrix. Thus, a column vector of size k has the form  a0 a1 ... ak−1  0≤i<k, 0≤j<1 . Row vectors are defined similarly. As usual, we shall equate 1 × 1-matrices A ∈ L1×1 with elements of L (namely, by equating each 1 × 1-matrix A ∈ L1×1 with its unique entry A0,0). Thus, if v and w are any two column vectors of size k, then wTv ∈ L. 2.3. Conventions about infinite matrices Furthermore, we shall allow our matrices to be infinite (i.e., have infinitely many rows or columns or both). This will be an optional feature of our results; we will state our claims in a way that allows the matrices to be infinite, but if the reader is only interested in finite matrices, they can ignore this possibility and skip Subsection 2.3 entirely. First of all, let us say a few words about how we will use ∞ in this note. As usual, "∞" is just a symbol which we subject to the following rules: We have n < ∞ and ∞ + n = ∞ − n = ∞ for each n ∈ N. Moreover, we shall use the somewhat strange Commutators, matrices and an identity of Copeland page 5 convention that {0, 1, . . . , ∞} denotes the set N (so it does not contain ∞). This has the consequence that {0, 1, . . . , ∞ − n} = N for each n ∈ N (since ∞ − n = ∞). We will use the following kinds of infinite matrices: • A k × ∞-matrix (where k ∈ N) has k rows (indexed by 0, 1, . . . , k − 1) and infinitely many columns (indexed by 0, 1, 2, . . .). Such a matrix will usually be written as • A ∞ × ℓ-matrix (where ℓ ∈ N) has infinitely many rows (indexed by 0, 1, 2, . . .) and ℓ columns (indexed by 0, 1, . . . , ℓ − 1). Such a matrix will usually be written as a0,0 a1,0 ... a0,1 a1,1 ... a0,2 a1,2 ... ak−1,0 ak−1,1 ak−1,2 · · · · · · ... · · ·   =(cid:0)ai,j(cid:1)0≤i<k, 0≤j<∞ . a0,0 a1,0 a2,0 ... a0,1 a1,1 a2,1 ... · · · · · · · · · ... a0,ℓ−1 a1,ℓ−1 a2,ℓ−1 ...   =(cid:0)ai,j(cid:1)0≤i<∞, 0≤j<ℓ .       • A ∞ × ∞-matrix has infinitely many rows (indexed by 0, 1, 2, . . .) and infinitely many columns (indexed by 0, 1, 2, . . .). Such a matrix will usually be written as a0,0 a1,0 a2,0 ... a0,1 a1,1 a2,1 ... a0,2 a1,2 a2,2 ... · · · · · · · · · . . .   =(cid:0)ai,j(cid:1)0≤i<∞, 0≤j<∞ . Matrices of these three kinds (that is, k × ∞-matrices, ∞ × ℓ-matrices and ∞ × ∞- matrices) will be called infinite matrices. In contrast, k × ℓ-matrices with k, ℓ ∈ N will be called finite matrices. We have previously introduced the notation Ai,j for the (i, j)-th entry of A when- ever A is a k × ℓ-matrix. The same notation will apply when A is an infinite matrix (i.e., when one or both of k and ℓ is ∞). If u, v, w are three elements of N, and if A is a u × v-matrix, and if B is a v × w- matrix, then the product AB is a u × w-matrix, and its entries are given by (AB)i,k = v−1 ∑ j=0 Ai,jBj,k (1) for all i ∈ {0, 1, . . . , u − 1} and k ∈ {0, 1, . . . , w − 1} . The same formula can be used to define AB when some of u, v, w are ∞ (keeping in mind that {0, 1, . . . , ∞ − 1} = N), but in this case it may fail to provide a well- Indeed, if v = ∞, then the sum on the right hand side of (1) defined result. Commutators, matrices and an identity of Copeland page 6 is infinite and thus may fail to be well-defined. Worse yet, even when products of infinite matrices are well-defined, they can fail the associativity law (AB) C = A (BC). We shall not dwell on these perversions, but rather restrict ourselves to a subclass of infinite matrices which avoids them: Definition 2.1. Let u, v ∈ N ∪ {∞}. Let A be a u × v-matrix. Let k ∈ Z. We say that the matrix A is k-lower-triangular if and only if we have (cid:0)Ai,j = 0 for all (i, j) satisfying i < j + k(cid:1) . Definition 2.2. A matrix A is said to be quasi-lower-triangular if and only if there exists a k ∈ Z such that A is k-lower-triangular. Note that we did not require our matrix A to be square in these two defini- tions. Unlike the standard kind of triangularity, our concept of quasi-triangularity is meant to be a tameness condition, meant to guarantee the well-definedness of an infinite sum; in particular, all finite matrices are quasi-lower-triangular. Better yet, the following holds:1 Proposition 2.3. Let k ∈ N ∪ {∞} and ℓ ∈ N. Then, any k × ℓ-matrix is quasi- lower-triangular. More concretely: Any k × ℓ-matrix is (ℓ − 1)-lower-triangular. Proposition 2.4. Let A be a matrix (finite or infinite) such that all but finitely many entries of A are 0. Then, A is quasi-lower-triangular. Quasi-lower-triangular matrices can be multiplied, as the following proposition shows: Proposition 2.5. Let u, v, w ∈ N ∪ {∞}. Let A be a quasi-lower-triangular u × v- matrix, and let B be a quasi-lower-triangular v × w-matrix. Then, the product AB is well-defined (i.e. the infinite sum on the right hand side of (1) is well-defined even if v = ∞) and is a quasi-lower-triangular u × w-matrix. More concretely: If k, ℓ ∈ Z are such that A is k-lower-triangular and B is ℓ-lower-triangular, then AB is (k + ℓ)-lower-triangular. Finally, multiplication of quasi-lower-triangular matrices is associative: Proposition 2.6. Let u, v, w, x ∈ N ∪ {∞}. Let A be a quasi-lower-triangular u × v-matrix; let B be a quasi-lower-triangular v × w-matrix; let C be a quasi- lower-triangular w × x-matrix. Then, (AB) C = A (BC). This proposition entails that we can calculate with quasi-lower-triangular ma- trices just as we can calculate with finite matrices. In particular, the quasi-lower- triangular ∞ × ∞-matrices form a ring. Thus, a quasi-lower-triangular ∞ × ∞- matrix has a well-defined n-th power for each n ∈ N. 1The proofs of all propositions stated in Subsection 2.3 are left to the reader as easy exercises. Commutators, matrices and an identity of Copeland page 7 2.4. The matrices S and Ub and the vectors Hc and ej Let us now introduce several more players into the drama. 2.4.1. Iverson brackets (truth values) We shall use the Iverson bracket notation: If A is any logical statement, then [A] will denote the integer (1, 0, if A is true; if A is false ∈ {0, 1}. This integer [A] is called the truth value of A. 2.4.2. m and a We now return to our ring L. For the rest of this note, we fix an m ∈ N ∪ {∞} and an element a ∈ L. 2.4.3. The matrix S We define an m × m-matrix S ∈ Lm×m by S = ([j = i + 1])0≤i<m, 0≤j<m . (2) This matrix S looks as follows: • If m ∈ N, then • If m = ∞, then S = S = 0 1 0 · · · 0 0 1 · · · 0 0 0 · · · ... . . . 0 0 0 · · · ... ... 0 0 0 ... 0 0 1 0 0 · · · 0 0 1 0 · · · 0 0 0 1 · · · 0 0 0 0 · · · ... . . . ... ... ...     . .     The matrix S (or, rather, the L-linear map from Lm to Lm it represents2) is often called the shift operator. Note that the matrix S is quasi-lower-triangular3 (and, in 2When m = ∞, you can read Lm both as the direct sum Li∈N L and as the direct product ∏ i∈N These are two different options, but either has an L-linear map represented by the matrix S. L. 3See Subsection 2.3 for the meaning of this word (and ignore it if you don't care about the case of m = ∞). n times {z Ub =  } j(cid:19) adi−j (cid:18)i a 0,   (b) , if i ≥ j; if i < j  0≤i<m, 0≤j<m . (3) Commutators, matrices and an identity of Copeland page 8 fact, (−1)-lower-triangular4), but of course not lower-triangular (unless L = 0 or m ≤ 1). 2.4.4. The matrix Ub If n is a nonnegative integer, T is a set and f : T → T is any map, then f n will mean the composition f ◦ f ◦ · · · ◦ f ; this is again a map from T to T. For any b ∈ L, we define an m × m-matrix Ub ∈ Lm×m by (Here, of course, adn a means (ada)n whenever n ∈ N.) This matrix Ub looks as follows: • If b ∈ L and m ∈ N, then Ub =   b ada (b) ad2 a (b) ... adm−1 a (b) 0 b 2 ada (b) ... (m − 1) adm−2 a 0 0 b ... (b) (cid:18)m − 1 2 (cid:19) adm−3 a • If b ∈ L and m = ∞, then Ub =   b 0 b 0 · · · 0 · · · ada (b) ad2 0 · · · a (b) 2 ada (b) a (b) 3 ad2 ad3 a (b) 3 ada (b) b · · · ... ... . . . 0 0 b ... ... · · · · · · · · · . . . (b) · · · 0 0 0 ... b .   .   Note that the matrix Ub is always lower-triangular and thus quasi-lower-triangular5. 4See Subsection 2.3 for the meaning of this word (and ignore it if you don't care about the case of m = ∞). 5See Subsection 2.3 for the meaning of this word (and ignore it if you don't care about the case of m = ∞). Commutators, matrices and an identity of Copeland page 9 2.4.5. The column vector Hc Furthermore, for each c ∈ L, we define an m × 1-matrix Hc ∈ Lm×1 by Hc =(cid:16)aic(cid:17)0≤i<m, 0≤j<1 . (4) Thus, Hc is an m × 1-matrix, i.e., a column vector of size m. It looks as follows: • If c ∈ L and m ∈ N, then • If c ∈ L and m = ∞, then am−1c Hc =  Hc =  a0c a1c ... a0c a1c a2c ... .     . Clearly, the matrix Hc is quasi-lower-triangular6, since it has only one column. 2.4.6. The column vector ej For each integer j with 0 ≤ j < m, we let ej ∈ Lm×1 be the m × 1-matrix defined by ej = ([p = j])0≤p<m, 0≤q<1 . (5) In other words, ej is the column vector (of size m) whose j-th entry is 1 and whose all other entries are 0. This column vector ej is commonly known as the j-th standard basis vector of Lm×1. Thus, in particular, e0 is a column vector with a 1 in its topmost position and 0's everywhere else. It looks as follows: • If m ∈ N, then e0 = 1 0 0 ... 0   .   6See Subsection 2.3 for the meaning of this word (and ignore it if you don't care about the case of m = ∞). Commutators, matrices and an identity of Copeland page 10 • If m = ∞, then e0 = 1 0 0 0 ...   .   Thus, eT 0 is a row vector with a 1 in its leftmost position and 0's everywhere else. This shows that the matrix eT 0 is quasi-lower-triangular7. 2.5. The general formula We are now ready to state our main claim: Theorem 2.7. Let n ∈ N be such that n < m. Let b ∈ L. Then, (ba)n = eT 0 (UbS)n H1. (The right hand side of this equality is a 1 × 1-matrix, while the left hand side is an element of L. The equality thus makes sense because we are equating 1 × 1- matrices with elements of L.) Example 2.8. Let us set m = 3 and n = 2 in Theorem 2.7. Then, Theorem 2.7 claims that (ba)2 = eT 0 (UbS)2 H1. Let us check this: We have Ub =  so that b 0 0 a (b) 2 ada (b) b ada (b) ad2 0 b and S =  0 1 0 0 0 1 0 0 0     0 b   , 0 b   and therefore b ada (b) ad2 UbS =  (UbS)2 =  =  m = ∞). 0 b 0 0 a (b) 2 ada (b) b   0 1 0 0 0 1 0 0 0   =  b 0 0 ada (b) 0 ad2 a (b) 2 ada (b) b 0 0 ada (b) 0 ad2 a (b) 2 ada (b) 2   b ada (b) 0 0 (ada (b))2 + b ad2 a (b) 0 3 ada (b) ad2 a (b) b2 3b ada (b) 4 (ada (b))2 + b ad2 a (b)   . 7See Subsection 2.3 for the meaning of this word (and ignore it if you don't care about the case of Commutators, matrices and an identity of Copeland page 11 b2 3b ada (b) 4 (ada (b))2 + b ad2 a (b)   a0 a1 a2  , we obtain = 0a0 + b ada (b) a1 + b2a2 Multiplying eT 0 =(cid:0) 1 0 0 (cid:1) by this equality, we find eT 0 (UbS)2 =(cid:0) 1 0 0 (cid:1)  b ada (b) 0 0 (ada (b))2 + b ad2 a (b) 0 3 ada (b) ad2 a (b) =(cid:16) 0 b ada (b) b2 (cid:17) . Multiplying this equality by H1 =  0 (UbS)2 H1 =(cid:16) 0 b ada (b) b2 (cid:17)  a01 a11 a21 eT = b   a0 a1 a2 =    a + b2a2 = b [a, b] a + b2a2 ada (b) =[a,b] {z } (by the definition of ada ) =ab−ba {z} = b (ab − ba) a + b2a2 = baba − bbaa + bbaa = baba = (ba)2 . This confirms the claim that (ba)2 = eT 0 (UbS)2 H1. 3. The proof 3.1. The idea Proving Theorem 2.7 is not hard, but it will take us some preparation due to the bookkeeping required. The main idea manifests itself in its cleanest form when m = ∞; indeed, it is not hard to prove the following two facts:8 Proposition 3.1. Assume that m = ∞. Let c ∈ L. Then, SHc = Hac. Proposition 3.2. Let b ∈ L and c ∈ L. Then, UbHc = Hbc. If m = ∞, then we can use Proposition 3.1 and Proposition 3.2 to conclude that (UbS) Hc = Hbac for each b ∈ L and c ∈ L. Thus, by induction, we can conclude that (UbS)n Hc = H(ba)nc for each n ∈ N, b ∈ L and c ∈ L (as long as m = ∞). Applying this to c = 1 and multiplying the resulting equality by eT 0 on both sides, 0 H(ba)n1 = (ba)n (the last equality sign is easy). we then obtain eT This proves Theorem 2.7 in the case when m = ∞. 0 (UbS)n H1 = eT 8We shall prove these two facts later. Commutators, matrices and an identity of Copeland page 12 Unfortunately, this argument breaks down if m ∈ N. In fact, Proposition 3.1 is true only for m = ∞; otherwise, the vectors SHc and Hac differ in their last entry. This "corruption" then spreads further to earlier and earlier entries as we inductively multiply by Ub and by S. What saves us is that it only spreads one entry at a time when we multiply by S, and does not spread at all when we multiply by Ub; thus it does not reach the first (i.e., 0-th) entry as long as we multiply by UbS only n times. But this needs to be formalized and proved. This is what we shall be doing further below. 3.2. A lemma about ada Before we come to this, however, we need a basic lemma about commutators: Lemma 3.3. Let b ∈ L and i ∈ N. Then, aib = i ∑ j=0(cid:18)i j(cid:19) adi−j a (b) · aj. It is not hard to prove Lemma 3.3 by induction on i. However, there is a slicker proof. It relies on the following well-known fact: Proposition 3.4. Let A be a ring. Let x and y be two elements of A such that xy = yx. Then, (x + y)n = n ∑ k=0(cid:18)n k(cid:19)xkyn−k for every n ∈ N. Proposition 3.4 is a straightforward generalization of the binomial formula to two commuting elements of an arbitrary ring. Proof of Lemma 3.3. Let End L denote the endomorphism ring of the Z-module L. Thus, the elements of End L are the Z-linear maps from L to L. Define the map La : L → L by (La (c) = ac for all c ∈ L) . Clearly, this map La is Z-linear; thus, it belongs to End L. Define the map Ra : L → L by (Ra (c) = ca for all c ∈ L) . Clearly, this map Ra is Z-linear; thus, it belongs to End L. Commutators, matrices and an identity of Copeland page 13 We have ada = La − Ra 9. Hence, ada belongs to End L (since La and Ra belong to End L). Also, Ra + ada = La (since ada = La − Ra). Furthermore, the elements La and Ra of End L satisfy Ra ◦ La = La ◦ Ra 10. But End L is a ring with multiplication ◦; thus, in particular, the operation ◦ is distributive (over +) on End L. Since La, Ra and ada belong to End L, we thus have = Ra ◦ (La − Ra) = Ra ◦ La =La◦Ra Ra ◦ ada =La−Ra {z} = La ◦ Ra − Ra ◦ Ra = (La − Ra) ◦Ra = ada ◦Ra. −Ra ◦ Ra =ada {z } {z } j(cid:19)Rj j=0(cid:18)i a = i ∑ Hence, Proposition 3.4 (applied to A = End L, x = Ra, y = ada and n = i) yields (Ra + ada)i = i ∑ k=0(cid:18) i k(cid:19)Rk a ◦ adi−k a ◦ adi−j a (here, we have renamed the index k as j in the sum). In view of Ra + ada = La, this rewrites as Li a = But each k ∈ N satisfies i ∑ j=0(cid:18)i j(cid:19)Rj a ◦ adi−j a . a (c) = akc Lk for each c ∈ L. (6) (7) 9Proof. Let c ∈ L. Then, La (c) = ac (by the definition of La) and Ra (c) = ca (by the definition of Ra). Hence, Comparing this with (La − Ra) (c) = La (c) {z}=ac = ac − ca. − Ra (c) {z }=ca ada (c) = [a, c] (by the definition of ada) = ac − ca (by the definition of [a, c]) , we obtain ada (c) = (La − Ra) (c). Now, forget that we fixed c. We thus have shown that ada (c) = (La − Ra) (c) for each c ∈ L. In other words, ada = La − Ra. Qed. 10Proof. Let c ∈ L. The definition of La yields La (c) = ac and La (Ra (c)) = a · Ra (c). The definition of Ra yields Ra (c) = ca and Ra (La (c)) = La (c) · a. Now, comparing (La ◦ Ra) (c) = La (Ra (c)) = a · Ra (c) = a · ca = aca with (Ra ◦ La) (c) = Ra (La (c)) = La (c) ·a = ac · a = aca, we obtain (Ra ◦ La) (c) = (La ◦ Ra) (c). In other words, Ra ◦ La = La ◦ Ra. Forget that we fixed c. We thus have proven that (Ra ◦ La) (c) = (La ◦ Ra) (c) for each c ∈ L. {z }=ca {z}=ac Commutators, matrices and an identity of Copeland page 14 [Proof of (7): It is straightforward to prove (7) by induction on k.] Furthermore, each k ∈ N satisfies a (c) = cak Rk for each c ∈ L. (8) [Proof of (8): It is straightforward to prove (8) by induction on k.] Now, applying both sides of the equality (6) to b, we obtain Li a (b) = i ∑ j=0(cid:18)i j(cid:19)Rj a ◦ adi−j a ! (b) = i ∑ j=0(cid:18)i j(cid:19) (cid:16)Rj a (cid:17) (b) a ◦ adi−j {z } (b)(cid:17) a(cid:16)adi−j a (b)·aj =ad i−j a =Rj (by (8), applied i−j a to k=j and c=ad (b)) = i ∑ j=0(cid:18)i j(cid:19) adi−j a (b) · aj. Comparing this with a (b) = aib Li (by (7), applied to k = i and c = b) , we obtain aib = This proves Lemma 3.3. 3.3. Formulas for eT i A i ∑ j=0(cid:18)i j(cid:19) adi−j a (b) · aj. We next recall a simple property of the vectors ei: Lemma 3.5. Let ℓ ∈ N ∪ {∞} and i ∈ N be such that 0 ≤ i < m. Let A be an m × ℓ-matrix. Then, eT i A = (the i-th row of A) . Note that the product eT i A on the left hand side of Lemma 3.5 is always well- defined, even when ℓ and m are ∞. (This stems from the fact that the row vector eT i has only one nonzero entry.) Lemma 3.5 says that the i-th row of A can be extracted by multiplying A from the left by the row vector eT the i-th position). This is a known fact from linear algebra and is easy to prove. The next lemma is a slight restatement of Lemma 3.5 in the case when ℓ = m: 0 (cid:1) (here, the 1 is at i = (cid:0) 0 0 · · · 0 1 0 0 · · · Commutators, matrices and an identity of Copeland page 15 Lemma 3.6. Let i ∈ N be such that 0 ≤ i < m. Let A be an m × m-matrix. Then, eT i A = m−1 ∑ j=0 Ai,jeT j . Proof of Lemma 3.6. For each j ∈ {0, 1, . . . , m − 1}, we have ej = ([p = j])0≤p<m, 0≤q<1 (by (5)) and thus eT j = ([q = j])0≤p<1, 0≤q<m (by the definition of the transpose of a matrix) . Hence, m−1 ∑ j=0 Ai,j eT j =([q=j])0≤p<1, 0≤q<m {z} = m−1 ∑ j=0 Ai,j ([q = j])0≤p<1, 0≤q<m = m−1 ∑ j=0 Ai,j [q = j]! . (9) 0≤p<1, 0≤q<m But for each q ∈ {0, 1, . . . , m − 1}, we have m−1 ∑ j=0 Ai,j [q = j] = Ai,q [q = q] + ∑ Ai,j [q = j] j6=q (since q=q) j∈{0,1,...,m−1}; {z }=1 from the sum (since q ∈ {0, 1, . . . , m − 1} ) (cid:19) (cid:18) here, we have split off the addend for j = q {z }=0 (because j6=q) = Ai,q + ∑ Ai,j0 = Ai,q. j∈{0,1,...,m−1}; j6=q Hence, m−1 ∑ j=0 Ai,j [q = j]! 0≤p<1, 0≤q<m =0 } {z =(cid:0)Ai,q(cid:1)0≤p<1, 0≤q<m = (the i-th row of A) (since A =(cid:0)Ai,j(cid:1)0≤i<m, 0≤j<m). Hence, (9) becomes m−1 ∑ j=0 Ai,jeT j = m−1 ∑ j=0 Ai,j [q = j]! = (the i-th row of A) = eT i A 0≤p<1, 0≤q<m (since Lemma 3.5 yields eT i A = (the i-th row of A)). This proves Lemma 3.6. Commutators, matrices and an identity of Copeland page 16 3.4. Proving eT u SHc = eT u Hac for u + 1 < m We can now prove a generalization of Proposition 3.1 to the case of arbitrary m: Proposition 3.7. Let u ∈ N be such that u + 1 < m. Then: (a) We have eT u+1. (b) Let c ∈ L. Then, eT u S = eT u SHc = eT u Hac. Proof of Proposition 3.7. (a) Lemma 3.5 (applied to ℓ = m, A = S and i = u) yields eT u S = (the u-th row of S) = ([q = u + 1])0≤p<1, 0≤q<m (10) (by (2)). But (5) (applied to j = u + 1) yields eu+1 = ([p = u + 1])0≤p<m, 0≤q<1 . Thus, by the definition of the transpose of a matrix, we obtain eT u+1 = ([q = u + 1])0≤p<1, 0≤q<m . Comparing this with (10), we obtain eT u S = eT u+1. This proves Proposition 3.7 (a). (b) Lemma 3.5 (applied to ℓ = 1, A = Hac and i = u) yields eT u Hac = (the u-th row of Hac) = (the u-th entry of Hac) (since Hac is a column vector) = aua c {z}=au+1 = au+1c. Comparing this with (cid:18)since (4) yields Hac =(cid:16)aiac(cid:17)0≤i<m, 0≤j<1(cid:19) Hc eT u S {z}=eT u+1 (by Proposition 3.7 (a)) = eT u+1Hc = (the (u + 1) -th row of Hc) (by Lemma 3.5, applied to ℓ = 1, A = Hc and i = u + 1) = (the (u + 1) -th entry of Hc) (since Hc is a column vector) = au+1c (cid:18)since (4) yields Hc =(cid:16)aic(cid:17)0≤i<m, 0≤j<1(cid:19) , we obtain eT u SHc = eT u Hac. This proves Proposition 3.7 (b). It is now easy to derive Proposition 3.1 from Proposition 3.7 (b): Proof of Proposition 3.1 (sketched). We have m = ∞; thus, every u ∈ N satisfies u + 1 < m. Hence, Proposition 3.7 (b) yields that eT u Hac for every u ∈ N. From this, it is easy to conclude that SHc = Hac (using Lemma 3.5). We leave the details to the reader, since we will not use Proposition 3.1. u SHc = eT Commutators, matrices and an identity of Copeland page 17 3.5. Proving Ub Hc = Hbc Next, we shall prove Proposition 3.2. For convenience, let us recall its statement: Proposition 3.8. Let b ∈ L and c ∈ L. Then, UbHc = Hbc. Proof of Proposition 3.8. Let u ∈ {0, 1, . . . , m − 1}. Hence, 0 ≤ u ≤ m − 1. (Keep in mind that {0, 1, . . . , ∞ − 1} = N, so u cannot be ∞ even when m = ∞.) From (3), we see that j(cid:19) adi−j (cid:18)i a 0, (b) , if i ≥ j; if i < j (11) (Ub)i,j =  for each i ∈ {0, 1, . . . , m − 1} and j ∈ {0, 1, . . . , m − 1}. From (4), we obtain (12) for each i ∈ {0, 1, . . . , m − 1}. The same argument (applied to bc instead of c) yields (Hc)i,0 = aic for each i ∈ {0, 1, . . . , m − 1}. Now, (1) (applied to m, m, 1, Ub, Hc, u and 0 instead of u, v, w, A, B, i and k) (Hbc)i,0 = aibc (13) (Ub)u,j {z } (b) , a (Hc)j,0 (by (12), applied to i=j) {z }=ajc if u ≥ j; if u < j (by (11), applied to i=u) m−1 ∑ j=0 yields (UbHc)u,0 = = m−1 ∑ j=0 = u ∑ j=0 a 0, j(cid:19) adu−j (cid:18)u =   j(cid:19) adu−j (cid:18)u   j(cid:19) adu−j (cid:18)u  {z j(cid:19) ad =(cid:18)u 0, 0, a (b) , (b) , u−j a (b) } (since u≥j (because j≤u)) if u ≥ j; · ajc if u < j if u ≥ j; if u < j ·ajc + m−1 ∑ j=u+1   a 0, j(cid:19) adu−j (cid:18)u {z =0 (b) , if u ≥ j; ·ajc if u < j } (since u<j (because j≥u+1>u)) (here, we have split the sum at j = u, since 0 ≤ u ≤ m − 1) = u ∑ j=0(cid:18)u j(cid:19) adu−j a (b) · ajc + m−1 ∑ j=u+1 0 · ajc = =0 {z } u ∑ j=0(cid:18)u j(cid:19) adu−j a (b) · ajc. Commutators, matrices and an identity of Copeland page 18 Comparing this with (Hbc)u,0 = c (by (13), applied to i = u) (b)·aj u−j a aub = u ∑ {z} j(cid:19) ad j=0(cid:18)u j(cid:19) adu−j j=0(cid:18)u ∑ a (by Lemma 3.3, applied to i=u) = u (b) · aj! c = u ∑ j=0(cid:18)u j(cid:19) adu−j a (b) · ajc, we obtain (UbHc)u,0 = (Hbc)u,0. Now, recall that Ub Hc is a column vector. Hence, (the u-th entry of UbHc) = (Ub Hc)u,0 = (Hbc)u,0 . (14) But Hbc is also a column vector. Thus, (the u-th entry of Hbc) = (Hbc)u,0 . Comparing this with (14), we obtain (the u-th entry of UbHc) = (the u-th entry of Hbc) . Now, forget that we fixed u. We thus have shown that (the u-th entry of Ub Hc) = (the u-th entry of Hbc) for each u ∈ {0, 1, . . . , m − 1}. In other words, each entry of UbHc equals the corresponding entry of Hbc. Thus, the two column vectors Ub Hc and Hbc are identical. In other words, Ub Hc = Hbc. This proves Proposition 3.8. 3.6. The ≡ k relations Now, we introduce a notation for saying that two m × ℓ-matrices are equal in their first m − k + 1 rows: Definition 3.9. Let ℓ ∈ N ∪ {∞}. Let A ∈ Lm×ℓ and B ∈ Lm×ℓ be two m × ℓ- matrices. Let k be a positive integer. We shall say that A ≡ B if and only if we k have u A = eT u B (cid:16)eT for all u ∈ {0, 1, . . . , m − k}(cid:17) . (Recall again that {0, 1, . . . , ∞} means N; thus, "u ∈ {0, 1, . . . , m − k}" means "u ∈ N" in the case when m = ∞. Note that {0, 1, . . . , g} means the empty set ∅ when g < 0.) Note that the condition "eT u B" in Definition 3.9 can be restated as "the u-th row of A equals the u-th row of B", because of Lemma 3.5. But we will find it easier to use it in the form "eT u A = eT u A = eT u B". The following lemma is easy: Commutators, matrices and an identity of Copeland page 19 Lemma 3.10. Let ℓ ∈ N ∪ {∞}. Let A ∈ Lm×ℓ and B ∈ Lm×ℓ be two m × ℓ- B. Let b ∈ L. Then, Ub A ≡ matrices. Let k be a positive integer such that A ≡ k k UbB. All that is needed of the matrix Ub for Lemma 3.10 to hold is that Ub is lower- triangular; we stated it for Ub just for convenience reasons. Proof of Lemma 3.10. We have A ≡ k B. In other words, we have u A = eT u B (cid:16)eT for all u ∈ {0, 1, . . . , m − k}(cid:17) (15) (by the definition of "A ≡ k B"). Let u ∈ {0, 1, . . . , m − k}. Thus, 0 ≤ u ≤ m − k. Also, u ∈ {0, 1, . . . , m − k} ⊆ ≤ m − 1). Hence, 0 ≤ u ≤ m − 1 < m. For each {0, 1, . . . , m − 1} (since m − k j ∈ {0, 1, . . . , u}, we have j ∈ {0, 1, . . . , u} ⊆ {0, 1, . . . , m − k} (since u ≤ m − k) and thus (16) (17) {z}≥1 eT j A = eT j B (by (15), applied to j instead of u). From (3), we see that j(cid:19) adi−j (cid:18)i a 0, (b) , if i ≥ j; if i < j (Ub)i,j =  for each i ∈ {0, 1, . . . , m − 1} and j ∈ {0, 1, . . . , m − 1}. For each j ∈ {u + 1, u + 2, . . . , m − 1}, we have j ≥ u + 1 and (Ub)u,j =  = 0 j(cid:19) adu−j (cid:18)u a 0, (b) , if u ≥ j; if u < j (by (17), applied to i = u) (since u < j (because j ≥ u + 1 > u)) . (18) Now, Lemma 3.6 (applied to i = u and A = Ub) yields eT u Ub = m−1 ∑ j=0 (Ub)u,j eT j = u ∑ j=0 (Ub)u,j eT j + m−1 ∑ j=u+1 (Ub)u,j eT j {z }=0 (by (18)) (here, we have split the sum at j = u, since 0 ≤ u ≤ m − 1) = u ∑ j=0 (Ub)u,j eT j + m−1 ∑ j=u+1 0eT j = u ∑ j=0 (Ub)u,j eT j . =0 {z } Commutators, matrices and an identity of Copeland page 20 Hence, = eT u Ub {z} u ∑ j=0 (Ub)u,jeT j A = u ∑ j=0 (Ub)u,j eT j! A = u ∑ j=0 u ∑ j=0 (Ub)u,j eT j B. (Ub)u,j eT j A = {z}=eT j B (by (16)) Comparing this with B = u ∑ j=0 (Ub)u,j eT j! B = u ∑ j=0 (Ub)u,j eT j B, eT u Ub {z} (Ub)u,jeT j = u ∑ j=0 we obtain eT u Ub A = eT u UbB. Forget that we fixed u. We thus have shown that u Ub A = eT u UbB (cid:16)eT for all u ∈ {0, 1, . . . , m − k}(cid:17) . UbB (by the definition of "Ub A ≡ k UbB"). This proves In other words, Ub A ≡ k Lemma 3.10. The analogue of Lemma 3.10 for S is even simpler: Lemma 3.11. Let ℓ ∈ N ∪ {∞}. Let A ∈ Lm×ℓ and B ∈ Lm×ℓ be two m × ℓ- matrices. Let k be a positive integer such that A ≡ k B. Then, SA ≡ k+1 SB. Proof of Lemma 3.11. We have A ≡ k B. In other words, we have u A = eT u B (cid:16)eT for all u ∈ {0, 1, . . . , m − k}(cid:17) (19) (by the definition of "A ≡ k B"). Let u ∈ {0, 1, . . . , m − (k + 1)}. Then, u ≤ m − (k + 1) = m − k − 1, so that u + 1 ≤ m − k. Combining this with u + 1 ∈ N (since u ∈ {0, 1, . . . , m − (k + 1)} ⊆ N), we obtain u + 1 ∈ {0, 1, . . . , m − k}. Hence, (19) (applied to u + 1 instead of u) yields eT < m. Hence, Proposition 3.7 (a) yields eT u S = eT u+1. Hence, u+1A = eT u+1B. But u + 1 ≤ m − k {z}>0 A = eT u+1A = eT u+1B. Comparing this with eT u S B = eT u+1B, we obtain eT u SA = eT u S {z}=eT u+1 eT u SB. Forget that we fixed u. We thus have shown that {z}=eT u+1 u SA = eT u SB (cid:16)eT for all u ∈ {0, 1, . . . , m − (k + 1)}(cid:17) . Commutators, matrices and an identity of Copeland page 21 In other words, SA ≡ k+1 SB (by the definition of "SA ≡ k+1 SB"). This proves Lemma 3.11. Now, we can prove the following lemma, which is as close as we can get to Proposition 3.1 without requiring m = ∞: Lemma 3.12. Let A ∈ Lm×1 and c ∈ L. Let k be a positive integer such that A ≡ k Hc. Then, SA ≡ k+1 Hac. Proof of Lemma 3.12. Lemma 3.11 (applied to ℓ = 1 and B = Hc) yields SA ≡ k+1 SHc. In other words, we have u SA = eT u SHc for all u ∈ {0, 1, . . . , m − (k + 1)}(cid:17) (20) SHc"). (by the definition of "SA ≡ k+1 Now, let u ∈ {0, 1, . . . , m − (k + 1)}. Thus, u ≤ m − (k + 1) = m − k − 1, so that u Hac. But (20) < m. Thus, Proposition 3.7 (b) yields eT u SHc = eT u + 1 ≤ m − k (cid:16)eT {z}>0 yields 3.12. Now, forget that we fixed u. We thus have shown that eT u SA = eT u SHc = eT u Hac. u SA = eT u Hac (cid:16)eT for all u ∈ {0, 1, . . . , m − (k + 1)}(cid:17) . In other words, SA ≡ k+1 Hac (by the definition of "SA ≡ k+1 Hac"). This proves Lemma 3.7. Proof of Theorem 2.7 Our last stop before Theorem 2.7 is the following lemma, which by now is an easy induction: Lemma 3.13. Let b ∈ L. Let n ∈ N. Then, (UbS)n H1 ≡ n+1 H(ba)n. Proof of Lemma 3.13. We shall prove Lemma 3.13 by induction on n: Induction base: It is easy to see that H1 ≡ 0+1 H1 11. 11Proof. Clearly, u H1 = eT u H1 (cid:16)eT for all u ∈ {0, 1, . . . , m − (0 + 1)}(cid:17) . In other words, H1 ≡ 0+1 H1 (by the definition of "H1 ≡ 0+1 H1"). Commutators, matrices and an identity of Copeland page 22 But (UbS)0 H1 = ImH1 = H1 and H(ba)0 = H1 (since (ba)0 = 1). In view of these {z }=Im two equalities, we can rewrite H1 ≡ 0+1 Lemma 3.13 holds for n = 0. This completes the induction base. H1 as (UbS)0 H1 ≡ 0+1 H(ba)0. In other words, Induction step: Let k be a positive integer. Assume that Lemma 3.13 holds for n = k − 1. We must prove that Lemma 3.13 holds for n = k. We have assumed that Lemma 3.13 holds for n = k − 1. In other words, we have (UbS)k−1 H1 ≡ k H(ba)k−1. Hence, Lemma 3.12 (applied to A = (UbS)k−1 H1 and c = (ba)k−1) yields S (UbS)k−1 H1 ≡ k+1 Ha(ba)k−1. Thus, Lemma 3.10 (applied to 1, k + 1, S (UbS)k−1 H1 and Ha(ba)k−1 instead of ℓ, k, A and B) yields UbS (UbS)k−1 H1 ≡ k+1 Ub Ha(ba)k−1. UbS (UbS)k−1 = (UbS) (UbS)k−1 = (UbS)k In view of and Ub Ha(ba)k−1 = Hba(ba)k−1 = H(ba)k this rewrites as (cid:16)by Proposition 3.8, applied to c = a (ba)k−1(cid:17) (cid:16)since ba (ba)k−1 = (ba) (ba)k−1 = (ba)k(cid:17) , (UbS)k H1 ≡ k+1 H(ba)k. In other words, Lemma 3.13 holds for n = k. This completes the induction step. Thus, Lemma 3.13 is proven by induction. We can now easily prove Theorem 2.7: Proof of Theorem 2.7. We have n < m, thus n + 1 ≤ m (since n and m are integers), hence m − (n + 1) ≥ 0. Thus, 0 ∈ {0, 1, . . . , m − (n + 1)}. Also, 0 ≤ 0 < m (since 0 ≤ n < m). Lemma 3.13 shows that In other words, we have (UbS)n H1 ≡ n+1 H(ba)n. u (UbS)n H1 = eT u H(ba)n (cid:16)eT for all u ∈ {0, 1, . . . , m − (n + 1)}(cid:17) Commutators, matrices and an identity of Copeland page 23 (by the definition of "(UbS)n H1 ≡ n+1 0 ∈ {0, 1, . . . , m − (n + 1)}), and thus obtain H(ba)n"). We can apply this to u = 0 (since 0 (UbS)n H1 = eT eT 0 H(ba)n =(cid:16)the 0-th row of H(ba)n(cid:17) (cid:16)by Lemma 3.5, applied to ℓ = 1, i = 0 and A = H(ba)n(cid:17) =(cid:16)the 0-th entry of H(ba)n(cid:17) yields H(ba)n =(cid:0)ai (ba)n(cid:1)0≤i<m, 0≤j<1 ! {z}=1 (cid:16)since H(ba)n is a column vector(cid:17) since (4) (applied to c = (ba)n ) = a0 (ba)n = (ba)n . This proves Theorem 2.7. 4. A Weyl-algebraic application 4.1. The claim We shall now restrict ourselves to a more special situation. Namely, we let K be a commutative ring, and we assume that the ring L is a K-algebra. Consider the polynomial ring K [t] in one variable t over K. For each polynomial g ∈ K [t] and each n ∈ N, we let g(n) be the n-th derivative of g; that is, g(n) = dn dtn g. Thus, in particular, g(0) = g and g(1) = g′ (where g′ denotes the derivative g). Recall that we fixed a ∈ L. Furthermore, let h ∈ L and x ∈ L be such that [a, x] = h and [h, a] = 0 and [h, x] = 0. (21) d dt g of This situation is actually fairly common: Example 4.1. Let D be the differentiation operator K [t] → K [t] , g 7→ d dt g. Let T be the "multiplication by t" operator K [t] → K [t] , g 7→ tg. Commutators, matrices and an identity of Copeland page 24 Then, the three operators D, T and idK[t] belong to the K-algebra EndK (K [t]) of all endomorphisms of the K-module K [t]. These three operators satisfy [D, T] = idK[t], hidK[t], Di = 0 and hidK[t], Ti = 0. Hence, we can obtain an example of the situation we are considering by setting L = EndK (K [t]), a = D, x = T and h = idK[t]. Further examples can be obtained by varying this one. For example, if K is a field, then K [t] can be replaced by the field of rational functions K (t). Alternatively, if K = R, then K [t] can be replaced by the algebra of C∞-functions R → R. Other examples appear in the theory of Weyl algebras and of 2-step nilpotent Lie algebras. Now, we return to the generality of K, L, a, h, x and m satisfying [a, x] = h and [h, a] = 0 and [h, x] = 0. For any polynomial g ∈ K [t], we define an m × m-matrix Vg ∈ Lm×m by Vg =    j(cid:19)g(i−j) (x) · hi−j, (cid:18)i 0, if i ≥ j; if i < j  0≤i<m, 0≤j<m . This matrix Vg looks as follows: • If g ∈ K [t] and m ∈ N, then g(0)(x) g(1)(x)·h g(2)(x)·h2 0 g(0)(x) 2g(1)(x)·h ... ... 0 0 g(0)(x) ... ··· ··· ··· ... 0 0 0 ... g(m−1)(x)·hm−1 (m−1)g(m−2)(x)·hm−2 (m−1 2 )g(m−3)(x)·hm−3 ··· g(0)(x) • If g ∈ K [t] and m = ∞, then (22) .    .  Vg =   Vg =  g(0) (x) 0 g(0) (x) g(1) (x) · h g(2) (x) · h2 g(3) (x) · h3 3g(2) (x) · h2 3g(1) (x) · h g(0) (x) 2g(1) (x) · h g(0) (x) ... ... ... ... 0 0 0 · · · · · · · · · · · · . . . 0 0 Now, Tom Copeland has found the following identity [MO337766]: Commutators, matrices and an identity of Copeland page 25 Theorem 4.2. Let n ∈ N be such that n < m. Let g ∈ K [t]. Then, (g (x) · a)n = eT 0 (cid:0)VgS(cid:1)n H1. This identity will easily follow from Theorem 2.7 (applied to b = g (x)), once we can show the following: Proposition 4.3. Let g ∈ K [t]. Then, Ug(x) = Vg. We shall thus mostly focus on proving Proposition 4.3. 4.2. How derivatives appear in commutators The main idea of our proof will be the following proposition, which relates deriva- tives in K [t] to commutators in L: Proposition 4.4. (a) We have axi = xia + ixi−1h for each positive integer i. (b) We have a · g (x) = g (x) · a + g′ (x) · h for each g ∈ K [t]. Here, of course, g′ means the derivative d dt g of the polynomial g. Proof of Proposition 4.4. The definition of [a, x] yields [a, x] = ax − xa. Hence, ax − xa = [a, x] = h. Thus, ax = xa + h. From [h, x] = 0, we obtain 0 = [h, x] = hx − xh (by the definition of [h, x]). In other words, hx = xh. (a) We shall prove Proposition 4.4 (a) by induction on i: = ax = xa + h with x1 Induction base: Comparing a x1 a + 1 x1−1 h = xa + h, {z}=x {z}=x {z}=x0=1 we find ax1 = x1a + 1x1−1h. In other words, Proposition 4.4 (a) holds for i = 1. This completes the induction base. Induction step: Let n be a positive integer. Assume that Proposition 4.4 (a) holds for i = n. We must prove that Proposition 4.4 (a) holds for i = n + 1. We have assumed that Proposition 4.4 (a) holds for i = n. In other words, we have axn = xna + nxn−1h. Now, a xn+1 {z}=xnx = =xna+nxn−1h = xn (xa + h) =xnxa+xnh axn {z} {z h = xnx +n xn−1x x =(cid:16)xna + nxn−1h(cid:17) x = xn {z }=xn {z} {z}=xn+1 =x(n+1)−1 xn } a + xnh + nxnh =(n+1)xnh ax {z}=xa+h {z } = xn+1a + (n + 1) h = xn+1a + (n + 1) x(n+1)−1h. +nxn−1 hx {z}=xh Commutators, matrices and an identity of Copeland page 26 In other words, Proposition 4.4 (a) holds for i = n + 1. This completes the induction step. Thus, Proposition 4.4 (a) is proven by induction. (b) Let g ∈ K [t]. Write the polynomial g in the form g = giti for some k ∈ N and some g0, g1, . . . , gk ∈ K. Thus, the definition of the derivative g′ yields g′ = igiti−1. Substituting x for t in this equality, we find k ∑ i=0 k ∑ i=1 g′ (x) = k ∑ i=1 igi {z}=gii xi−1 = k ∑ i=1 giixi−1. (23) k ∑ i=0 giti, we obtain g (x) = k ∑ i=0 gixi. Hence, Substituting x for t in the equality g = a · g (x) = a · {z} k ∑ i=0 gixi = k ∑ i=0 gixi = k ∑ i=0 + k ∑ i=1 gi giaxi = g0a x0 {z}=1 =xia+ixi−1h (by Proposition 4.4 (a)) axi {z} (here, we have split off the addend for i = 0 from the sum) = g0a + k ∑ i=1 = gi(cid:16)xia + ixi−1h(cid:17) } {z giixi−1h gixia+ k ∑ i=1 k ∑ i=1 Comparing this with g (x) ·a + g′ (x) ·h = gixi k ∑ i=0 {z} = k ∑ i=0 = giixi−1 k ∑ i=1 (by (23)) {z} gixi! · a + k ∑ i=1 giixi−1! · h = = g0a + k ∑ i=1 gixia + k ∑ i=1 giixi−1h. + k ∑ i=1 giixi−1h k ∑ i=0 gi xia =g0x0a+ {z } k ∑ i=1 gixia (here, we have split off the addend for i=0 from the sum) a + k ∑ i=1 gixia + k ∑ i=1 giixi−1h = g0a + k ∑ i=1 gixia + k ∑ i=1 giixi−1h, = g0 x0 {z}=1 we obtain a · g (x) = g (x) · a + g′ (x) · h. This proves Proposition 4.4 (b). Note that we have not used the condition [h, a] = 0 in Proposition 4.4; but we will use it now: Commutators, matrices and an identity of Copeland page 27 Proposition 4.5. Let b ∈ L. Then, ada(cid:0)bhi(cid:1) = ada (b) · hi for each i ∈ N. Proof of Proposition 4.5. From [h, a] = 0, we obtain 0 = [h, a] = ha − ah (by the definition of [h, a]). In other words, ha = ah. Hence, hia = ahi for each i ∈ N. (24) [Proof of (24): This follows by induction on i, using ha = ah in the induction step.] Now, let i ∈ N. Then, the definition of ada yields ada (b) = [a, b] = ab − ba (by the definition of [a, b]) . But the definition of ada also yields ada(cid:16)bhi(cid:17) =ha, bhii = a(cid:16)bhi(cid:17) −(cid:16)bhi(cid:17) a = abhi − b hia = abhi − bahi = (ab − ba) (cid:16)by the definition of ha, bhii(cid:17) {z } hi =ada(b) {z}=ahi (by (24)) = ada (b) · hi. This proves Proposition 4.5. Corollary 4.6. Let g ∈ K [t]. Let p ∈ N. Then, adp a (g (x)) = g(p) (x) · hp. Proof of Corollary 4.6. We shall prove Corollary 4.6 by induction on p: {z}=id Induction base: Comparing ad0 a (g (x)) = id (g (x)) = g (x) with g(0) (x) · h0 = {z}=g {z}=1 g (x), we obtain ad0 p = 0. This completes the induction base. a (g (x)) = g(0) (x) · h0. In other words, Corollary 4.6 holds for Induction step: Let n ∈ N. Assume that Corollary 4.6 holds for p = n. We must prove that Corollary 4.6 holds for p = n + 1. We have assumed that Corollary 4.6 holds for p = n. In other words, we have adn a (g (x)) = g(n) (x) · hn. Now, a adn+1 =ada ◦ adn a {z } (g (x)) = (ada ◦ adn a ) (g (x)) = ada    } = ada(cid:16)g(n) (x) · hn(cid:17) = ada(cid:16)g(n) (x)(cid:17) · hn a (g (x)) =g(n)(x)·hn {z adn (25) Commutators, matrices and an identity of Copeland page 28 (by Proposition 4.5, applied to b = g(n) (x) and i = n). But Proposition 4.4 (b) (applied to g(n) instead of g) yields a · g(n) (x) = g(n) (x) · a +(cid:16)g(n)(cid:17)′ (x) · h. = g(n+1) (this follows from the definitions of g(n) and g(n+1)), In view of (cid:16)g(n)(cid:17)′ we can rewrite this as a · g(n) (x) = g(n) (x) · a + g(n+1) (x) · h. (26) Now, the definition of ada yields ada(cid:16)g(n) (x)(cid:17) =ha, g(n) (x)i = a · g(n) (x) − g(n) (x) · a (cid:16)by the definition of ha, g(n) (x)i(cid:17) = g(n+1) (x) · h (by (26)) . Hence, (25) becomes adn+1 a (g (x)) = ada(cid:16)g(n) (x)(cid:17) } =g(n+1)(x)·h {z ·hn = g(n+1) (x) · h · hn {z}=hn+1 = g(n+1) (x) · hn+1. In other words, Corollary 4.6 holds for p = n + 1. This completes the induction step. Thus, Corollary 4.6 is proven by induction. 4.3. Proofs of Proposition 4.3 and Theorem 4.2 Proof of Proposition 4.3. If b ∈ L is arbitrary, then j(cid:19) adi−j (cid:18)i a 0, (b) , if i ≥ j; if i < j j(cid:19)g(i−j) (x) · hi−j, (cid:18)i 0, if i ≥ j; if i < j (Ub)i,j =  (cid:0)Vg(cid:1)i,j =  (27) (28) for each i ∈ {0, 1, . . . , m − 1} and j ∈ {0, 1, . . . , m − 1} (by (3)). On the other hand, (22) shows that for each i ∈ {0, 1, . . . , m − 1} and j ∈ {0, 1, . . . , m − 1}. Now, let us fix i ∈ {0, 1, . . . , m − 1} and j ∈ {0, 1, . . . , m − 1}. We shall prove that (cid:16)Ug(x)(cid:17)i,j =(cid:0)Vg(cid:1)i,j. Commutators, matrices and an identity of Copeland page 29 Indeed, we are in one of the following two cases: Case 1: We have i ≥ j. Case 2: We have i < j. Let us first consider Case 1. In this case, we have i ≥ j. Hence, i − j ∈ N. Thus, Corollary 4.6 (applied to p = i − j) yields adi−j a (g (x)) = g(i−j) (x) · hi−j. Now, (27) (applied to b = g (x)) yields (g (x)) (since i ≥ j) =(cid:18)i a j(cid:19) adi−j {z =g(i−j)(x)·hi−j } Let us next consider Case 2. In this case, we have i < j. Hence, (28) becomes (cid:16)Ug(x)(cid:17)i,j Comparing this with a 0, if i < j if i ≥ j; (g (x)) , = j(cid:19) adi−j (cid:18)i  j(cid:19)g(i−j) (x) · hi−j. =(cid:18)i (cid:0)Vg(cid:1)i,j = j(cid:19)g(i−j) (x) · hi−j, (cid:18)i  j(cid:19)g(i−j) (x) · hi−j =(cid:18)i =(cid:0)Vg(cid:1)i,j. Thus,(cid:16)Ug(x)(cid:17)i,j j(cid:19)g(i−j) (x) · hi−j, (cid:18)i 0, 0, if i < j if i ≥ j; But (27) (applied to b = g (x)) yields we obtain(cid:16)Ug(x)(cid:17)i,j (cid:0)Vg(cid:1)i,j =  =  Comparing these two equalities, we find (cid:16)Ug(x)(cid:17)i,j (cid:0)Vg(cid:1)i,j is proven in Case 2. We have now proven the equality (cid:16)Ug(x)(cid:17)i,j j(cid:19) adi−j (cid:18)i (cid:16)Ug(x)(cid:17)i,j Hence, this equality always holds. (g (x)) , if i ≥ j; if i < j 0, a = 0 = 0 if i ≥ j; if i < j (by (28)) (since i ≥ j) , =(cid:0)Vg(cid:1)i,j is proven in Case 1. (since i < j) . (since i < j) . = = (cid:0)Vg(cid:1)i,j. Thus, (cid:16)Ug(x)(cid:17)i,j = (cid:0)Vg(cid:1)i,j in both Cases 1 and 2. Commutators, matrices and an identity of Copeland page 30 Now, forget that we fixed i and j. We thus have shown that (cid:16)Ug(x)(cid:17)i,j for all i ∈ {0, 1, . . . , m − 1} and j ∈ {0, 1, . . . , m − 1}. In other words, each entry of the m × m-matrix Ug(x) equals the corresponding entry of the m × m-matrix Vg. Hence, Ug(x) = Vg. This proves Proposition 4.3. = (cid:0)Vg(cid:1)i,j Proof of Theorem 4.2. Theorem 2.7 (applied to b = g (x)) yields   n   Ug(x) S (by Proposition 4.3) {z}=Vg H1 = eT 0 (cid:0)VgS(cid:1)n H1. (g (x) · a)n = eT 0 This proves Theorem 4.2. References [MO337766] Tom Copeland, MathOverflow question #337766 ("Expansions of iterated, or nested, derivatives, or vectors -- conjectured matrix computation"). https://mathoverflow.net/questions/337766/
1303.0651	1	1303	2013-03-04T09:35:21	Fine gradings and gradings by root systems on simple Lie algebras	[ "math.RA", "math.RT" ]	Given a fine abelian group grading on a finite dimensional simple Lie algebra over an algebraically closed field of characteristic zero, with universal grading group $G$, it is shown that the induced grading by the free group $G/\tor(G)$ is a grading by a (not necessarily reduced) root system. Some consequences for the classification of fine gradings on the exceptional simple Lie algebras are drawn.	math.RA	math	FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS ON SIMPLE LIE ALGEBRAS ALBERTO ELDUQUE⋆ Lg of a Abstract. Given a fine abelian group grading Γ : L = Lg∈G finite dimensional simple Lie algebra over an algebraically closed field of characteristic zero, with G being the universal grading group, it is shown that the induced grading by the free group G/ tor(G) on L is a grading by a (not necessarily reduced) root system. Some consequences for the classification of fine gradings on the ex- ceptional simple Lie algebras are drawn. 1. Introduction Gradings by abelian groups on simple Lie algebras appear in many in- stances. A systematic study of these gradings was started in [PZ89]. For the classical simple Lie algebra over an algebraically closed field of charac- teristic 0, the fine gradings were classified in [Eld10]. For the exceptional simple algebras they were classified in [DM06] and [BT09] for G2, in [DM09] for F4 and in [DV] for E6. On the other hand, gradings by root systems were introduced by Berman and Moody in [BM92], who used them as tools to study some classes of infinite-dimensional Lie algebras. The goal of this paper is to relate both types of gradings. It will be shown that any fine grading with infinite universal grading group on a simple finite- dimensional Lie algebra over an algebraically closed field of characteristic 0 induces a grading by a (possibly not reduced) root system. Some conse- quences for the classification of fine gradings in the exceptional cases will be derived too. The first two sections will review the gradings by abelian groups and grad- ings by root systems respectively. The main result connecting fine gradings and gradings by root systems will be proved in the next two sections. This result shows that any fine grading is determined by a grading by a root system and a special grading on the coordinate algebra of the root grad- ing. This grading on the coordinate algebra is studied in Section 5. The last section is devoted to draw consequences for the classification of the fine gradings on the simple exceptional simple Lie algebras. 2. Gradings Let A be an algebra (not necessarily associative) over a field F and let G be an abelian group (written additively). ⋆ Supported by the Spanish Ministerio de Econom´ıa y Competitividad and FEDER (MTM2010-18370-C04-02) and by the Diputaci´on General de Arag´on -- Fondo Social Eu- ropeo (Grupo de Investigaci´on de ´Algebra). 1 2 ALBERTO ELDUQUE Definition 2.1. A G-grading on A is a vector space decomposition such that Γ : A =Mg∈G Ag AgAh ⊂ Ag+h for all g, h ∈ G. If such a decomposition is fixed, we will refer to A as a G-graded algebra. The nonzero elements a ∈ Ag are said to be homogeneous of degree g; we will write deg a = g. The support of Γ is the set Supp Γ := {g ∈ G Ag 6= 0}. Let Γ : A =Mg∈G Ag and Γ′ : B = Mh∈H Bh be two gradings on algebras, with supports S and T , respectively. Definition 2.2. We say that Γ and Γ′ are equivalent if there exists an isomorphism of algebras ψ : A → B and a bijection α : S → T such that ψ(As) = Bα(s) for all s ∈ S. Any such ψ will be called an equivalence of Γ and Γ′ (or of A and B if the gradings are clear from the context). Given a group grading Γ on an algebra A, there are many groups G such that Γ, regarded as a decomposition into a direct sum of subspaces such that the product of any two of them lies in a third one, can be realized as a G-grading, but there is one distinguished group among them [PZ89]. Definition 2.3. Suppose that Γ admits a realization as a U -grading for some abelian group U . We will say that U is a universal group of Γ if, for any other realization of Γ as a G-grading, there exists a unique homomorphism U → G that restricts to identity on Supp Γ. One shows that the universal group, which we denote by U (Γ), exists and depends, up to isomorphism, only on the equivalence class of Γ. Indeed, U (Γ) is generated by S = Supp Γ with defining relations s1 + s2 = s3 whenever 0 6= As1As2 ⊂ As3 (si ∈ S). Given a G-grading Γ : A =Lg∈G Ag and a group homomorphism α : G → H, we obtain the induced H-grading αΓ : A = Lh∈H A′ Lg∈α−1(h) Ag. Definition 2.4. Given gradings Γ : A =Lg∈G Ag and Γ′ : A =Lh∈H A′ h, we say that Γ′ is a coarsening of Γ, or that Γ is a refinement of Γ′, if for any g ∈ G there exists h ∈ H such that Ag ⊂ A′ h. The coarsening (or refinement) is said to be proper if the inclusion is proper for some g ∈ Supp Γ. (In particular, αΓ is a coarsening of Γ, which is not necessarily proper.) A grading Γ is said to be fine if it does not admit a proper refinement. h by setting A′ h = Any G-grading on a finite-dimensional algebra A is induced from some fine grading Γ by a homomorphism α : U (Γ) → G. Over algebraically closed fields of characteristic zero, the classification of fine gradings on A up to equivalence is the same as the classification of maximal diagonalizable subgroups (i.e., maximal quasitori) of Aut(A) up to conjugation (see e.g. [PZ89]). More precisely, given a grading Γ on the algebra A with universal group G, let G be its group of characters FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 3 (homomorphisms G → F×). Any χ ∈ G acts as an automorphism of A by means of χ.x = χ(g)x for any g ∈ G and x ∈ Ag. This allows us to identify G with a quasitorus (the direct product of a torus and a finite subgroup) of the algebraic group Aut(A). Conversely, given a quasitorus Q of Aut(A), Q = G for G the group of homomorphisms (as algebraic groups) Q → F×. Then Q induces a G-grading of A, where Ag = {x ∈ A : χ(x) = g(χ)x} for any g ∈ G. In this way [PZ89], the fine gradings on A, up to equivalence, correspond to the conjugacy classes in Aut(A) of the maximal quasitori (or maximal abelian diagonalizable subgroups) of Aut(A). Fine gradings on simple Lie algebras belonging to the series A, B, C and D (including D4) were classified in [Eld10]. The fine gradings on the simple Lie algebra of type G2 were classified in [DM06, BT09], for type F4 in [DM09] (see also [EK12]), and for type E6 in [DV]. Definition 2.5. Let Γ : A =Lg∈G Ag be a grading on the algebra A. • A subspace B of A is said to be graded if B = Lg∈G(B ∩ Ag). (Equivalently, B is graded by G with Bg = B ∩ Ag for any g ∈ G.) • The type of Γ is the r-tuple (n1, . . . , nr), where r = max{dim Ag : g ∈ G} and ni is the number of homogeneous components of dimension i, for i = 1, . . . , r. Hence dim A =Pr From now on, the ground field F will be assumed to be algebraically closed i=1 ini. of characteristic zero. 3. Gradings by root systems Berman and Moody [BM92] started the study of Lie algebras graded by root systems Φ. (See [ABG02] and the references therein.) Definition 3.1. A Lie algebra L over F is graded by the reduced root system Φ, or Φ-graded, if: (i) L contains as a subalgebra a finite-dimensional simple Lie algebra subalgebra h = g0; g = h ⊕ (cid:0)Lα∈Φ gα(cid:1) whose root system is Φ relative to a Cartan (ii) L =Lα∈Φ∪{0} L(α), where L(α) = {X ∈ L : [H, X] = α(H)X for all H ∈ h}; and (iii) L(0) =Pα∈Φ[L(α), L(−α)]. The subalgebra g is said to be a grading subalgebra of L. Berman and Moody [BM92] studied the simply laced case (types Ar, Dr and Er), and Benkart and Zelmanov [BZ96] considered the remaining cases. Under the adjoint action of g, a Φ-graded Lie algebra L decomposes as a sum of finite-dimensional irreducible g-modules whose highest weights are the highest long root, highest short root, or 0. By collecting isomorphic summands into "isotypic components", we may assume that there are F- vector spaces A, B and D such that (3.1) where the grading subalgebra g is identified with g ⊗ 1 for a distinguished element 1 ∈ A; W is 0 if g is of type Ar (r ≥ 1), Dr (r ≥ 4), or Er L = (g ⊗ A) ⊕ (W ⊗ B) ⊕ D, 4 ALBERTO ELDUQUE (r = 6, 7, 8), while W is the irreducible g-module whose highest weight is the highest short root if g is of type Br (r ≥ 2), Cr (r ≥ 3), F4 or G2; and D is the centralizer of g ≃ g ⊗ 1, and hence it is a subalgebra of L. The problem of classifying the Φ-graded Lie algebras reduces to one of determining the possibilities for A, B and D, and of finding the multiplica- tion. The bracket in L is invariant under the adjoint action of g and this gives the sum a = A ⊕ B the structure of a unital algebra. Besides, D acts as derivations on a, with A and B being invariant under this action. The type of the algebra a depends on the root system Φ. This algebra a is called the coordinate algebra of L. For instance (see [BZ96]), assume that Φ is the root system of type G2. Then g is the Lie algebra of type G2, which can be identified with the Lie algebra of derivations of the Cayley (or octonion algebra) O, and W can be identified with the subspace of trace zero octonions O0. The Cayley algebra is endowed with a nondegenerate quadratic form n (the norm) such that any element w satisfies w2 − t(w)w + n(w)1 = 0, where t(w) = n(w, 1) := n(w + 1) − n(w) − 1. Moreover, one has the following properties: (1) Homg(g ⊗ g, g) is spanned by the bracket, (2) Homg(g ⊗ g, F) is spanned by the Killing form κ, which is a scalar multiple of the trace form relative to the representation provided by W. Dw1,w2(w) = [[w1, w2], w] + 3((w1w)w2 − w1(ww2)), [w1, w2] = w1w2 − w2w1. (3) Homg(g⊗W, W) is spanned by the action of g on W (X⊗W 7→ X.W ), (4) Homg(W ⊗ W, g) is spanned by the map w1 ⊗ w2 7→ Dw1,w2, where (5) Homg(W ⊗ W, W) is spanned by the bracket (inside O) w1 ⊗ w2 7→ (6) Homg(W ⊗ W, F) is spanned by the trace form w1 ⊗ w2 7→ t(w1w2). (7) Homg(g ⊗ g, W), Homg(g ⊗ W, g) and Homg(g ⊗ W, F) are trivial. Therefore, the bracket in L is given by: [D ⊗ a, D′ ⊗ a′] = [D, D′] ⊗ a · a′ + κ(D, D′)haa′i, [D ⊗ a, w ⊗ b] = D(w) ⊗ a · b, [w ⊗ b, w′ ⊗ b′] = Dw,w′ ⊗ (bb′) + [w, w′] ⊗ b ◦ b′ + 2t(w1w2)hbb′i, for any D, D′ ∈ g, w, w′ ∈ W, a, a′ ∈ A, b, b′ ∈ B and d, d′ ∈ D, and for linear maps • A ⊗ A → A : a ⊗ a′ 7→ a · a′, which is symmetric, • A ⊗ A → D : a ⊗ a′ 7→ haa′i, which is skew-symmetric, • B ⊗ B → A : b ⊗ b′ 7→ (bb′), which is symmetric, • B ⊗ B → B : b ⊗ b′ 7→ b ◦ b′, which is symmetric, • B ⊗ B → D : b ⊗ b′ 7→ hbb′i, which is skew-symmetric, • A ⊗ B → B : a ⊗ b 7→ a · b. These linear maps satisfy the following properties: (1) A is a unital commutative algebra with the product a · a′, [d, D ⊗ a] = D ⊗ d(a), [d, w ⊗ b] = w ⊗ d(b), (3.2) FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 5 (2) a = A ⊕ B with the multiplication given by (a + b) · (a′ + b′) =(cid:0)a · a′ + (bb′)(cid:1) +(cid:0)a · b′ + a′ · b + b ◦ b′(cid:1), for a, a′ ∈ A and b, b′ ∈ B is a Jordan algebra over A with normalized trace given by trace(a + b) = a, which satisfies the Cayley-Hamilton equation of degree 3. (3) The action of D on a = A⊕ B is an action by derivations. Moreover, haa′i(a) = 0 = hbb′i(A) and hb′b′′i(b) = b′ · (b′′ · b) − b′′ · (b′ · b), for a, a′ ∈ A and b, b′, b′′ ∈ B. (4) D = hAAi +hBBi. (This is imposed by condition (iii) in Definition 3.1) Therefore, in this case, the coordinate algebra a is a Jordan algebra "of degree 3" over the unital commutative associative algebra A (see [BZ96]). Note that hAAi is a central ideal of L, so if L is simple, then this is trivial, and hence D = hBBi. Gradings by nonreduced root systems (type BCr) will also appear at- tached to fine gradings. Following [ABG02] we recall the next definition: Definition 3.2. Let Φ be the nonreduced root system BCr (r ≥ 1). A Lie algebra L over F is graded by Φ, or Φ-graded, if: (i) L contains as a subalgebra a finite-dimensional simple Lie algebra gebra h = g0 is the reduced subsystem of type Br, Cr or Dr contained in Φ; g = h⊕(cid:0)Lα∈Φ′ gα(cid:1) whose root system Φ′ relative to a Cartan subal- (ii) L =Lα∈Φ∪{0} L(α), where L(α) = {X ∈ L : [H, X] = α(H)X for all (iii) L(0) =Pα∈Φ[L(α), L(−α)]. Again, the subalgebra g is said to be a grading subalgebra of L, and L is said to be BCr-graded with grading subalgebra of type Xr, where Xr is the type of g. H ∈ h}; and Only BCr-graded subalgebras of type Br will show up related to fine gradings on simple Lie algebras. For r ≥ 3, let W be the natural module for the Lie algebra g of type Br. Thus W is endowed with a symmetric nondegenerate bilinear form (..), and g = {x ∈ EndF(W) : (xuv) = −(uxv) for all u, v ∈ W}, s = {s ∈ EndF(W) : (suv) = (usv) for all u, v ∈ W and trace(s) = 0}. In this case, a BCr-graded subalgebra of type Br can be described, up to isomorphism, as follows (see [ABG02, (1.30)]): L = (g ⊗ A) ⊕ (s ⊗ B) ⊕ (W ⊗ C) ⊕ D, (3.3) The bracket in L gives b = A ⊕ B ⊕ C the structure of an algebra, which is termed the coordinate algebra of L. Moreover (see [ABG02] for details), for r ≥ 3 we have: • The sum a = A ⊕ B is a unital associative algebra (multiplication denoted by α · α′), with 1 ∈ A (the subalgebra g is identified with g⊗ 1), with involution η whose subspace of symmetric elements is A and whose subspace of skew-symmetric elements is B. 6 ALBERTO ELDUQUE • The space C is an associative left a-module (action denoted by α · c, and it is equipped with a hermitian form ξ relative to η, such that the multiplication in b is given by: (α + c) · (α′ + c′) =(cid:0)α · α′ + ξ(c, c′)(cid:1) +(cid:0)α · c′ + α′η · c(cid:1). For r = 2, the grading subalgebra b = A ⊕ B ⊕ C is a bit more involved, and can be described in terms of structurable algebras. (See [ABG02] for details.) For r = 1, a BC1-graded subalgebra of type B1 can be described, up to isomorphism, as follows: L = (g ⊗ A) ⊕ (s ⊗ B) ⊕ D. (3.4) Here the natural module W for the simple Lie algebra g (isomorphic to sl2(F)) of type B1 is three-dimensional, and hence isomorphic to the adjoint module g, and the subspace of symmetric trace zero endomorphisms s is the five-dimensional irreducible module for g. In this case, results of Allison [All79] give that the coordinate algebra a = A ⊕ B is a structurable algebra whose involution is given by (a + b)η = a − b (so A is the subspace of symmetric elements and B the subspace of skew-symmetric elements), and the quotient of L by its center Z(L) is the Tits-Kantor-Koecher Lie algebra constructed from the structurable algebra (a, η). (See [BS03, Theorem 2.6].) The arguments used in the proof of [EO08, Theorem 7.5] give a more precise picture in this situation. The Lie bracket in L, which is invariant under the action of the subalgebra g ≃ g ⊗ 1, is given by: for any A, B ∈ g, X, Y ∈ h, a, b ∈ A, x, y ∈ B, and d ∈ D, where for any a ∈ A, [1, a] = 0 for any a ∈ A, x ∈ B, x ∈ B, • A× A → A: (a, b) 7→ a◦ b is a symmetric bilinear map with 1◦ a = a • A × A → B: (a, b) 7→ [a, b] is a skew symmetric bilinear map with • A × B → A: (a, x) 7→ [a, x] is a bilinear map with [1, x] = 0 for any • A × B → B: (a, x) 7→ a ◦ x is a bilinear map with 1 ◦ x = x for any • B × B → A: (x, y) 7→ x ◦ y is a symmetric bilinear map, • B × B → B: (x, y) 7→ [x, y] is a skew symmetric bilinear map, • A × A → D: (a, b) 7→ habi is a skew symmetric bilinear map, • B × B → D: (x, y) 7→ hxyi is a skew symmetric bilinear map, 1 • D is a subalgebra of L, • [A⊗a, B⊗b] = [A, B]⊗a◦b −(cid:0)AB +BA− 2 2 trace(AB)habi, • [A ⊗ a, X ⊗ x] = −(AX + XA) ⊗ 1 • [X ⊗ x, Y ⊗ y] = [X, Y ] ⊗ x ◦ y −(cid:0)XY + Y X − 2 • [d, A ⊗ a] = A ⊗ d(a), • [d, X ⊗ x] = X ⊗ d(x), 2 trace(XY )hxyi, 3 trace(AB)I3(cid:1)⊗ 1 2 [a, x] + [A, X] ⊗ a ◦ x, 1 2 [x, y] + 1 2 [a, b] + 3 trace(XY )I3(cid:1) ⊗ FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 7 • the bilinear maps D × A → A: (d, a) 7→ d(a) and D × B → B: (d, x) 7→ d(x), give two representations of the Lie algebra D. Define x ◦ a = a ◦ x and [x, a] = −[a, x] for any a ∈ A and x ∈ B, and define on the vector space a = A ⊕ B a multiplication by means of u · v = u ◦ v + 1 2 [u, v] for any u, v ∈ A∪ B, so u◦ v = 1 2 (u· v + v · u) and [u, v] = u· v − v · u. Define too a linear map − : a → a such that a + x = a− x for any a ∈ A and x ∈ B. Then ([EO08, Theorem 7.5]) the subspace a, with this multiplication and involution, is a structurable algebra. Besides, condition (iii) in Definition 3.2 shows D = hAAi + hBBi, and a straightforward application of the Jacobi identity gives habi(u) = Da,b(u), hxyi(u) = Dx,y(u), for any a, b ∈ A, x, y ∈ B and u ∈ A∪ B, where Du,v is the derivation of the structurable algebra a defined in [All78, Equation (15)]: Du,v(w) = 1 3 [[u, v] + [¯u, ¯v], w] + (w, v, u) − (w, ¯u, ¯v), (3.5) for u, v, w ∈ a, where (w, v, u) = (w · v) · u − w · (v · u) is the associator of the elements w, v, u. 4. Fine gradings on semisimple Lie algebras The aim of this section is to show that any fine grading on a finite di- mensional semisimple Lie algebra, with the property that the free rank of its universal group is > 0, determines in a natural way a (possibly non reduced) root system. This root system is irreducible if the Lie algebra is simple. The first two items of the next Proposition have been proved in [DM09] over the field of complex numbers. Given a finitely generated abelian group G, let tor(G) denote its torsion subgroup and let ¯G be the quotient G/ tor(G), which is free. Its rank is called the free rank of G. Proposition 4.1. Let L be a finite dimensional semisimple Lie algebra and let Γ : L = Lg∈G Lg be a fine grading. Assume that G is the universal group of Γ. (Since the dimension of L is finite, G is a finitely generated abelian group.) Then the following conditions hold: (i) The neutral homogeneous component L0 is a toral subalgebra of L (i.e., adL0 consists of commuting diagonalizable operators in L). (ii) The dimension of L0 coincides with the free rank of G. (iii) Let tor(G) be the torsion subgroup of G. The induced grading ¯Γ : L = L¯g∈G/ tor(G) L¯g is the weight space decomposition relative to L0. Proof. The Killing form of L satisfies κ(Lg, Lh) = 0 unless g + h = 0, and hence the restriction of κ to L0 is nondegenerate. This shows that L0 is reductive (see [Bou98, Chapter I, §6.4, Proposition 5]). Moreover, for any X ∈ Z(L0) (the center of L0), the semisimple and nilpotent parts of X belong to Z(L0) too and κ(Xn, L0) = 0 since adXn is nilpotent, so we get 8 ALBERTO ELDUQUE Xn = 0. Therefore, the elements of Z(L0) are semisimple and L0 is reductive in L (see [Bou98, Chapter 1, §6.4,6.5]). Let h be a Cartan subalgebra of L0. Hence Z(L0) is contained in h and h is maximal among the toral subalgebras of L contained in L0. For any g ∈ G, Lg is invariant under the adjoint action of L0. Therefore, Γ can be refined by means of the weight space decomposition relative to the toral subalgebra h. Since Γ is fine, for any g ∈ G there exists a linear form α ∈ h∗ such [H, X] = that Lg is contained in the weight space L(α) := {X ∈ L : α(H)X for all H ∈ h}. In particular, L0 = L(0) ∩ L0 = h is a toral sub- algebra. This proves the first part. (Note that 0 denotes both the neutral component of G and the trivial linear form, but this should cause no confu- sion.) Therefore, Γ is a refinement of the grading given by the weight space decomposition relative to the toral subalgebra h = L0: Γ : L =Lα∈h∗ L(α). Denote by Φ the set of nonzero weights in this decomposition: (4.1) Then ZΦ is a free abelian subgroup of h∗ and we may look at Γ as a grading by the group ZΦ. Φ := {α ∈ h∗ \ 0 : L(α) 6= 0}. Since G is the universal group of Γ and Γ is a coarsening of Γ, there is a surjective homomorphism π : G → ZΦ (4.2) such that π(g) = α if Lg ⊆ L(α). And since ZΦ is torsion free, π induces a surjective homomorphism ¯π : ¯G := G/ tor(G) → ZΦ. In particular, the rank of the free group ¯G is greater than or equal to the rank of ZΦ. But FΦ is the whole dual vector space h∗, as otherwise there would exist an element 0 6= X ∈ h such that α(X) = 0 for any α ∈ Φ, and then X would belong to the center of L, and this is trivial since L is semisimple. In particular, this shows that the rank of the free abelian group ZΦ is greater than or equal to the dimension of the vector space FΦ = h∗. Hence we obtain rank(ZΦ) ≥ dim h, and thus rank ¯G ≥ dim h. Since the universal group G is generated by the support of Γ, so is ¯G generated by the support of ¯Γ. But ¯G is a finitely generated free abelian group, so there are elements ¯g1, . . . , ¯gm ∈ Supp ¯Γ such that ¯G = Z¯g1 ⊕··· ⊕ Z¯gm (here ¯g denotes the class of g modulo tor(G)). The Lie algebra L is semisimple, and hence any derivation is inner. In particular, for any i = 1, . . . , m, there is a unique element Hi ∈ L such that [Hi, X] = niX for any X ∈ Ln1¯g1+···+nm¯gm. Moreover, we may replace Hi by its component in L0 = h for any i so, by uniqueness, we obtain H1, . . . , Hm ∈ L0. Since the sum L¯g1 ⊕ ··· ⊕ L¯gm is direct, the elements H1, . . . , Hm are linearly independent, and hence we get m = rank ¯G ≤ dim h. This proves the second part: rank ¯G = dim h. The argument above shows that h = FH1 ⊕ ··· ⊕ FHm, and for any ¯g = n1¯g1 + ··· + nm¯gm we have L¯g = L(α), where α is the linear form on h such that α(Hi) = ni for any i. This proves the last part. Remark 4.2. The neutral component of the grading ¯Γ in Proposition 4.1 is (cid:3) L¯0 =Lg∈tor(G) Lg = L(0), and this is the centralizer CentL(L0). FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 9 Using the arguments in the proof above, L(0) is shown to be reductive In particu- in L, so L(0) = Z(L(0)) ⊕ [L(0), L(0)] and L0 ⊆ Z(L(0)). lar, the neutral component of the restriction of Γ to [L(0), L(0)] is trivial: [L(0), L(0)]0 = 0. Therefore, Γ induces a grading on [L(0), L(0)] by the finite group tor(G) whose homogeneous component of degree 0 is trivial. These gradings are called special. (See [Hes82] for properties of these gradings.) Remark 4.3. Condition (ii) in Proposition 4.1 does not suffice to ensure that the grading Γ is fine. As an example, let V be a five-dimensional vector space with a basis {e1, e2, e3, e4, e5}, endowed with the symmetric bilinear form b such that b(ei, ej) = δij for any i, j. Consider the associated orthogonal Lie algebra so(V, b) of skew symmetric endomorphisms relative 2, with deg e1 = (¯1, ¯0, ¯0), deg e2 = to b. The vector space V is graded by Z3 (¯0, ¯1, ¯0), deg e3 = (¯0, ¯0, ¯1), deg e4 = (¯1, ¯1, ¯1) and deg e5 = 0. This induces a grading by Z3 2 on so(V, b) of type (4, 3), with the basic elements Eij − Eji, i 6= j, being homogeneous of degree deg ei + deg ej. Here Eij denotes the endomorphism that takes ej to ei and annihilates the other basic elements. Then sl(V, b)0 = 0, the free rank of the finite grading group is also 0, but this grading is not fine, as it can be refined to get a grading of type (10) 2 with deg e1 = (¯1, ¯0, ¯0, ¯0), deg e2 = (¯0, ¯1, ¯0, ¯0), deg e3 = (¯0, ¯0, ¯1, ¯0), by Z4 deg e4 = (¯0, ¯0, ¯0, ¯1) and deg e5 = 0. Theorem 4.4. Let L be a finite dimensional semisimple Lie algebra and let Γ : L = Lg∈G Lg be a fine grading. Assume that G is the universal group of Γ. Let Φ be as in (4.1). Then, Φ is a (possibly non reduced) root system in the euclidean vector space E = R ⊗Q QΦ. If L is simple, then Φ is an irreducible root system. Proof. Several steps will be followed: 1. Because of Proposition 4.1, the set of weights Φ is precisely π(Supp Γ \ tor(G)), with π in (4.2). Hence, for any g ∈ Supp Γ \ tor(G), let α = π(g) and take an element 0 6= X ∈ Lg ⊆ L(α). Then L−g is contained in L(−α). Since α is not 0, adX is nilpotent. By the Jacobson -- Morozov Theorem [Jac79, Chapter III, Theorem 17], there are elements H, Y ∈ L such that [H, X] = 2X, [H, Y ] = −2Y and [X, Y ] = H (i.e.; X, H, Y form an sl2-triple). We have H = Ph∈G Hh, Y = Ph∈G Yh for homogeneous elements Hh, Yh ∈ Lh, h ∈ G. Then [H, X] = 2X implies [H0, X] = 2X, so α(H0) = 2, and hence [H0, Y−g] = −2Y−g. Also, from [X, Y ] = H we get [X, Y−g] = H0. Therefore, we may take H ∈ L0 = h and Y ∈ L−g. 2. The restriction of the Killing form κ to h = L0 is nondegenerate, so it induces a nondegenerate symmetric bilinear form (. .) on h∗ = FΦ. For any α ∈ Φ, take an element g ∈ G with π(g) = α, and an sl2-triple X ∈ Lg, H ∈ L0, Y ∈ L−g as above. For any β ∈ Φ, the sum Li∈Z L(β + iα) is a module for the subalgebra s = span {X, H, Y } (isomorphic to sl2(F)). With standard arguments we obtain β(H) = r− q ∈ Z and β − β(H)α ∈ Φ, where q = max{n ∈ Z : β + nα ∈ Φ}, r = max{n ∈ Z : β − nα ∈ Φ}. In particular, Hα := H does not depend on g or X, only on α. Also, we get κ(Hα, Hα) = Xβ∈Φ(cid:0)dim L(β)(cid:1)β(Hα)2 ∈ Z>0. 10 ALBERTO ELDUQUE 3. For any α ∈ h∗ there is a unique Tα ∈ h such that α(H) = κ(Tα, H) for any H ∈ h. If the element T ∈ h = L0 satisfies α(T )(= κ(Tα, T )) = 0, then for any β ∈ Φ we have trace(cid:16)(adHαadT )Li∈Z Lβ+iα(cid:17) = β(T )trace(cid:16)adHαLi∈Z Lβ+iα(cid:17) = 0. κ(Tα,Tα) Tα = 2 Hence κ(Hα, T ) = 0 too, and hence Hα ∈ FTα. Since α(Hα) = 2, we get Hα = (αα) = β(Hα). Therefore we have for any α, β ∈ Φ that (αα) Tα. Define, as usual, hβαi := 2(βα) 2 hβαi ∈ Z and β − hβαiα ∈ Φ. (αα)2 κ(Tα, Tα) = 4 Also we have κ(Hα, Hα) = 4 positive rational number. 4. Take a basis {α1, . . . , αm} of h∗ contained in Φ, and let g1, . . . , gm be elements in G with π(gi) = αi for any i = 1, . . . , m. For any γ ∈ QΦ (⊆ h∗), there are rational numbers r1, . . . , rm such that γ = r1α1 + ··· + rmαm, and we get: (αα) , so (αα) = κ(Hα,Hα) is a 4 (γγ) = κ(Tγ, Tγ) = Xβ∈Φ(cid:0)dim L(β)(cid:1)β(Tγ)2 riβ(Tαi!2 ri(βαi)!2 ri(αiαi) = Xβ∈Φ(cid:0)dim L(β)(cid:1) m Xi=1 = Xβ∈Φ(cid:0)dim L(β)(cid:1) m Xi=1 = Xβ∈Φ(cid:0)dim L(β)(cid:1) m Xi=1 2 hβαii!2 ∈ Q>0. Hence E = R ⊗Q QΦ is a euclidean vector space with inner product deter- mined by (. .), Φ is a finite subset of E not containing 0, that spans E and such that hαβi = 2(αβ) (ββ) ∈ Z and β − hβαiα ∈ Φ, for any α, β ∈ Φ. If L is simple, then Φ must be irreducible, as otherwise Φ would split as Therefore, Φ is a root system. 5. a disjoint union Φ = Φ1 ∪Φ2, with (Φ1Φ2) = 0. But then (cid:0)Lα∈Φ1 (cid:0)Pα∈Φ1[L(α), L(−α)](cid:1) would be a proper ideal of L. 5. The main result L(α)(cid:1) ⊕ (cid:3) With the same hypotheses as in the previous section, take a system of simple roots ∆ of the root system Φ in (4.1). Hence ∆ is a basis of h∗ contained in Φ and Φ = Φ+ ∪Φ−, with Φ+ ⊆Pα∈∆ Z≥0α, Φ− = −Φ+. For any α ∈ ∆ choose gα ∈ G such that π(gα) = α. Since G is generated by Supp Γ, we have G =(cid:0)Lα∈∆ Zgα(cid:1) ⊕ tor(G). Let G :=Lα∈∆ Zgα and let (5.1) Lg. g :=Mg∈ G The arguments in the proof of Proposition 4.1 show that g is a reductive subalgebra in L. Also, any 0 6= X ∈ gg, g 6= 0, is contained in a sl2-triple, FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 11 so the center Z(g) is contained in L0 = h. But the dimension of h coincides with the rank of ZΦ, so we conclude that Z(g) = 0 and g is semisimple. Also, any weight of h on g belongs to ±(cid:0)Lα∈∆ Z≥0α(cid:1), so ∆ is a system of simple roots for g relative to its Cartan subalgebra h. We conclude that g is, up to isomorphism, the semisimple Lie algebra with ∆ as a system of simple roots. Now the main result of the paper, relating fine gradings and gradings by root systems, follows easily: Theorem 5.1. Let L be a finite dimensional simple Lie algebra and let Γ : L = Lg∈G Lg be a fine grading. Assume that G is the universal group of Γ. Let Φ be as in (4.1). Then L is graded by the irreducible (possibly nonreduced) root system Φ with grading subalgebra g in (5.1). Moreover, if Φ is nonreduced (type BCr), then g is simple of type Br. Proof. The Lie algebra L contains the semisimple subalgebra g with Cartan subalgebra h and system of simple roots ∆. Since L is simple, Φ (or ∆) is irreducible, and the ideal(cid:0)Lα∈Φ L(α)(cid:1)⊕(cid:0)Pα∈Φ[L(α), L(−α)](cid:1) is the whole L. Hence L is graded by the root system Φ with g as a grading subalgebra. Moreover, any root in Φ is a sum of roots in g. Hence for Φ of type BCn, g is of type Bn. (cid:3) 6. Grading on the coordinate algebra Let Γ : L = Lg∈G Lg be a fine grading on a finite-dimensional simple Lie algebra, with G being the universal group of Γ. As in the proof of Proposition 4.1, let Φ be the set of weights of the adjoint action of L0, and let π : G → ZΦ be the surjective group homomorphism with π(g) = α if Lg ⊆ L(α). Then π induces an isomorphism ¯π : ¯G = G/ tor(G) → ZΦ by item (iii) of Proposition 4.1. Let g be the grading subalgebra in Theorem 5.1, obtained after fixing a system of simple roots ∆ and preimages gα under π of the elements in ∆. Also, consider the free abelian group G =Lα∈∆ Zgα, such that G = G ⊕ tor(G). The restriction of π to G is bijective. If Φ is reduced, then we have a decomposition as in equation (3.1). Then: • g = g ⊗ 1 is, by its own construction, a graded subalgebra of L, and Besides, D is contained in L(0) = Lg∈tor(G) Lg, and hence D is graded by tor(G). Moreover, D0 ⊆ L0 = g0, so D0 = 0. Therefore, the grading of D by tor(G) is a special grading. Remark 4.2.) hence so is its centralizer D = CentL(g). (See • Let λ be the highest root of g (relative to ∆), then λ is not a weight of W, and hence L(λ) = gλ ⊗ A. On the other hand, if gλ is the preimage in G of λ, then L(λ) =M{Lg : π(g) = λ} = Mg∈tor(G) Lgλ+g, so the vector space A is graded by tor(G), where Ah is defined by means of: for any h ∈ tor(G). Lgλ+h = gλ ⊗ Ah, (6.1) 12 ALBERTO ELDUQUE • Since g ⊗ A is the g-submodule of L generated by gλ ⊗ A (λ is the highest root of g), it follows that g ⊗ A is a graded subspace of L and for any g ∈ G and h ∈ tor(G) we have (g ⊗ A)g+h = gπ(g) ⊗ Ah. • By invariance under the adjoint action of g, the subspace W⊗B turns out to be the orthogonal complement of (cid:0)g ⊗ A(cid:1) ⊕ D relative to the Killing form of L. Since this latter subspace is a graded subspace of L, so is W ⊗ B. Let µ be the highest weight of the g-module W relative to ∆ (µ is the highest short root). Let gµ be the preimage by π in G of µ. Then, as for A, we also get that B is graded by tor(G) if we define Bh by means of (W ⊗ B)gµ+h = Wµ ⊗ Bh, for any h ∈ tor(G). And since W is generated, as a module for g, by Wµ (= {w ∈ W : H.w = µ(H)w for all H ∈ h = g0}), it follows that the subspace W ⊗ B is a graded subspace of L and for any g ∈ G and h ∈ tor(G) we have (W ⊗ B)g+h = Wπ(g) ⊗ Bh. On the other hand, if Φ is nonreduced of type BC1, then we have a decomposition as in Equation (3.4), and the same arguments show that D inherits a special grading by tor(G), that if µ is the highest weight then L(µ) = Wµ ⊗ B, and this shows that B is graded by tor(G) as above. Finally, g ⊗ A is the orthogonal complement to (W ⊗ B) ⊕ D relative to the Killing form, and we conclude that A is graded too by tor(G) as above. Finally, if Φ is nonreduced of type BCr, r ≥ 2, then we have a decompo- sition as in Equation (3.3) and one checks as before that D inherits a special grading by tor(G), that if µ is the highest weight of s, then L(µ) = sµ ⊗ B, and hence it follows that B is tor(G)-graded. Then (g ⊗ A) ⊕ (W ⊗ C) is the orthogonal complement, so it is a graded subspace too. Here, if λ is the highest root, then (cid:16)(g ⊗ A) ⊕ (W ⊗ C)(cid:17) ∩ L(λ) = gλ ⊗ A, so again we conclude that A is tor(G)-graded, and from here we deduce that so is C. These arguments prove most of the next result: Proposition 6.1. Under the conditions above, with Φ being an irreducible root system, the coordinate algebra a = A ⊕ B (in the reduced case or for BC1) or b = a⊕C (in the BCr-case, r ≥ 2) inherits a fine grading by tor(G), where A and B, and C in the BCr-case, r ≥ 2, are graded subspaces. Moreover, A0 = F1 while B0 = 0, and also C0 = 0 (in the BCr-case, r ≥ 2), tor(G) is the universal group, and this grading on a, or b, induces a special grading on D by tor(G). Proof. The fact that a inherits a grading by tor(G) is clear from the earlier arguments. Also L0 = g0 = g0 ⊗ 1, so A0 = F1 and B0 = 0 (and C0 = 0 too in the BCr case, r ≥ 2). Hence a0 = F1. Besides, any refinement of this grading on a would give a refinement of Γ, as the grading by tor(G) of D is determined by the grading on a, because of condition (iii) in Definition FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 13 3.1. The last part is a direct consequence of G being the universal group of Γ. (cid:3) 7. Applications The results in the previous sections will be used to classify the fine grad- ings on the simple exceptional Lie algebras whose universal group have free rank > 2. Quick proofs of the classification of fine gradings, up to equiva- lence, on the simple Lie algebras of type G2 and F4 will be given too. Table 25 in [Dyn52] gives a list of the simple subalgebras of rank > 1 of the exceptional simple Lie algebras, together with the decomposition of any such simple Lie algebra as a sum of irreducible modules for the simple subalgebra. This immediately gives the different possibilities, up to conju- gation, of grading an exceptional simple Lie algebra by an irreducible (not necessarily reduced) root system of rank ≥ 2. The different possibilities are summarized in Table 1, where g, s, W, A, B, C and D are as in Equations (3.1) or (3.3). In many cases, this corresponds (see [BZ96, Tit66]) to the well-known Tits construction T(C, J), for a unital composition algebra C and a degree three simple Jordan algebra J, which we recall now (see also [EO08]): Let H be a unital composition algebra (or Hurwitz algebra) with norm n and trace t. The unital composition algebras are, up to isomorphism, F, K = F ⊕ F, H = Mat2(F) (quaternion algebra), and the Cayley algebra O (recall that the ground field F is assumed to be algebraically closed). Let J be a unital simple Jordan algebra of degree 3, so that J is the Jordan algebra H3(H′) of hermitian 3× 3 matrices over another unital composition algebra H′. Denote by H0 and J0 the subspaces of trace zero elements in H and J. For a, b ∈ H, the linear map Da,b : H → H defined by Da,b(c) = [[a, b], c]+ 3(a, c, b), where [a, b] = ab−ba is the commutator, and (a, c, b) = (ac)b−a(cb) the associator, is a derivation of H. These derivations span the Lie algebra Der(H). In the same vein, for x, y ∈ J, the linear map dx,y : J → J defined by is the inner derivation of J determined by the elements x and y. These derivations span the Lie algebra of derivations Der(J) Given H and J as before, consider the space with the anticommutative multiplication [., .] specified by: T(H, J) = Der(H) ⊕(cid:0)H0 ⊗ J0(cid:1) ⊕ Der(J), (7.2) • Der(H) and Der(J) are Lie subalgebras, and [Der(H), Der(J)] = 0, • [D, a ⊗ x] = D(a) ⊗ x, [d, a ⊗ x] = a ⊗ d(x), • [a ⊗ x, b ⊗ y] = trace(xy)Da,b +(cid:0)[a, b] ⊗ x ∗ y(cid:1) + 2t(ab)dx,y, for all D ∈ Der(H), d ∈ Der(J), a, b ∈ H0, and x, y ∈ J0, where x ∗ y = xy − 1 Looking at Equation (7.2) from the left, in case H is the Cayley algebra O (i.e., dim H = 8), then Der(O) is the simple Lie algebra of type G2 and (7.2) gives a decomposition as in Equation (3.1), thus proving that T(O, J) is graded by the root system of type G2 with coordinate algebra J = F1⊕ J0. 3 trace(xy)1. dx,y(z) = x(yz) − y(xz), (7.1) 14 ALBERTO ELDUQUE Looking from the right, we obtain: • If J is the Albert algebra A (i.e.; J is the algebra of hermitian 3 × 3- matrices over the Cayley algebra), then Der(A) is the simple Lie algebra of type F4, and (7.2) proves that T(O, A) is graded by the root system of type F4 with coordinate algebra O = F1 ⊕ O0. • If J is the Jordan algebra H3(H), then Der(J) is the simple Lie algebra of type C3, and T(O, J) is graded by the root system of type C3 with coordinate algebra O. • Also, if J is the Jordan algebra Mat3(F)+ = H3(K), then Der(J) is the simple Lie algebra of type A2, and then T(O, J) is graded by the root system of type A2 with coordinate algebra O. Lie Root algebra system dim A dim B dim C dim D model G2 F4 F4 E6 E6 E6 E6 E6 E7 E7 E7 E7 E7 E8 E8 E8 E8 G2 G2 F4 A2 BC2 G2 F4 E6 BC2 G2 C3 F4 E7 BC2 G2 F4 E8 1 1 1 8 5 1 1 1 7 1 1 1 1 11 1 1 1 0 5 0 -- 1 8 1 0 1 14 7 3 0 1 26 7 0 -- -- -- -- 2 -- -- -- 8 -- -- -- -- 20 -- -- -- 0 3 0 14 4 8 0 0 9 21 14 3 0 24 52 14 0 coordinate algebra F T(O, H3(F)) H3(F) T(O, H3(K)) F O T(O, H3(K)) Mat3(F)+ T(K, A) T(O, H3(H)) T(O, H3(H)) T(H, A) T(O, A) T(O, A) K F H3(H) O H F A O F Table 1. Gradings by root systems of rank ≥ 2 of the ex- ceptional simple Lie algebras. Zr with r the rank of the algebra. Theorem 7.1. The fine gradings, up to equivalence, of the exceptional sim- ple Lie algebras whose universal group has free rank ≥ 3 are the following: • The Cartan gradings of F4, E6, E7 and E8. The universal group is • The gradings of Er, r = 6, 7, 8 induced by gradings by the root system of type F4. The universal groups are Z4 × Zr−5 , r = 6, 7, 8, and the respective types are (72, 1, 0, 1), (120, 0, 3, 1) and (216, 0, 0, 8). • A grading of E7 induced by a grading by the root system of type C3. 2 The universal group is Z3 × Z3 2 and its type is (102, 0, 1, 7). Proof. The only gradings by root systems of rank ≥ 3 in Table 1 are the Cartan gradings, the gradings by the root system of type F4 of Er, r = 6, 7, 8, FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 15 2 and Z3 and the grading by the root system of type C3 of E7. In the second case, the coordinate algebra is H = K, H or O respectively, and the only grading on these algebras with neutral component equal to F1 are the gradings obtained by the Cayley-Dickson doubling process (see [Eld98] or [EK12]), whose universal groups are Z2, Z2 2 respectively. The computation of the types is straightforward using the model T(H, A). Finally, these gradings are fine as the neutral component is the Cartan subalgebra of the subalgebra Der(A) of type F4, and the grading induced in this subalgebra is the Cartan grading, which is fine. Hence, if any of these gradings could be refined, the refinement would be attached to a grading by a root system of rank ≥ 4, which is impossible. Finally, the coordinate algebra for the grading by the root system of type C3 of E7 is O. The only grading of O whose neutral component is F1 is its 2-grading. The resulting grading by Z3 × Z3 Z3 2 of E7 is fine and its type is easily computed using the model T(O, H3(H)). (cid:3) We finish with the promised short proofs of the classification of fine grad- ings for G2 and F4. For G2 it was proved independently in [DM06] and [BT09], and for F4 in [DM09] (see also [Dra12] and [EK12]). The arguments here are very different in nature. Theorem 7.2. Up to equivalence, the simple Lie algebra of type G2 is en- the Cartan grading by Z2, and a dowed with two different fine gradings: special grading by Z3 2 in which the seven nonzero homogeneous spaces are all Cartan subalgebras. Proof. Let Γ : L = Lg∈G Lg be a fine grading of the simple Lie algebra L of type G2, with G its universal group. By Theorem 5.1 and Table 1, either Γ is the Cartan grading, or the free rank of G is one, or G is a finite group. If the free rank is one, then L is graded by the root system BC1 (this includes gradings by A1) and hence L is given by the Tits-Kantor-Koecher Lie algebra constructed from a structurable algebra and Γ is obtained by combining the Z-grading given by the rank one root system, and a grading of the coordinate algebra as in Proposition 6.1. A look at the possibilities in [All79, §8] shows that the coordinate algebra is the structurable algebra a = Mat2(F) with multiplication given by: γ and involution αβ′ + βδ′ + 2γγ′ δδ′ + 3γβ′ (cid:19) , (cid:18)α β δ(cid:19)(cid:18)α′ β′ γ′ γα′ + δγ′ + 2ββ′ δ′(cid:19) =(cid:18) αα′ + 3βγ′ δ(cid:19) =(cid:18)δ β γ α(cid:19) . (cid:18)α β γ 0 1 ) , e = ( 0 1 0 0 ) , f = ( 0 0 Consider the basis (cid:8)1 = ( 1 0 0 −1(cid:1)(cid:9) of a, so that A = span{1, e, f} and B = Fs. Since s2 = 1 and B0 = 0, s ∈ ag with 2g = 0. The subspace Fe + Ff = {x ∈ a : sx + xs = 0} is graded. But for any nonzero homogeneous element αe + βf , the ele- ments s(αe + βf ) = −αe + βf , (αe + βf )2 = 2(β2e + α2f ) + 3αβ1 and (αe + βf )(−αe + βf ) = 2(β2e + α2f ) + 3αβs are homogeneous too, and this forces the nonzero element β2e + α2f to be homogeneous of degree 0 and g at the same time, a contradiction. 1 0 ) , s =(cid:0) 1 0 16 ALBERTO ELDUQUE Finally, if G is finite, consider the finite quasitorus Q of the algebraic group Aut(L) which is the image of the character group G (isomorphic to G). Since Γ is fine, Q is a maximal quasitorus. Also, since L is of type G2, Aut(L) is a connected and simply connected semisimple algebraic group. For any χ ∈ Q, χ is semisimple and Aut(L) is connected and semisimple, so its centralizer CentAut(L)(χ) is reductive [Hum95, Theorem 2.2], and since Aut(L) is simply connected, CentAut(L)(χ) is connected [Hum95, Theorem 2.11]. Hence the solvable radical coincides with the connected component of by χ is semisimple. its center: Z(cid:0)CentAut(L)(χ)(cid:1)◦, and this is a torus [Hum75, Lemma 19.5]. But Z(cid:0)CentAut(L)(χ)(cid:1)◦ is contained in any maximal quasitorus of CentAut(L)(χ), so it is contained in Q. But Q is finite, so Z(cid:0)CentAut(L)(χ)(cid:1)◦ = 0, and hence CentAut(L)(χ)(cid:1) is semisimple, so that the subalgebra of L of elements fixed The automorphisms of finite order of the simple Lie algebras are well- known (see [Kac90, Chapter 8]). They are determined, up to conjugation, by a subset of nodes of the extended Dynkin diagram and some coefficients. Those automorphisms of finite order whose subalgebra of fixed elements is semisimple, correspond to the automorphisms attached to a single node. For G2, the extended Dynkin diagram (with coefficients) is: ❡ 1 ❡ 2 ❡ 3i Therefore, the order of a nontrivial finite order automorphism of L whose subalgebra of fixed elements is semisimple is restricted to 2 or 3. Thus Q is a maximal nontoral elementary p-subgroup of Aut(L), with p = 2 or 3. According to [Gri91], there is just one possibility, up to conjugation, where Q, and hence G, isomorphic to Z3 2. (cid:3) Theorem 7.3. Up to equivalence, the simple Lie algebra L of type F4 is endowed with four different fine gradings, whose universal groups and types are as follows: 2, of type (31, 0, 7), • the Cartan grading by Z4, of type (48, 0, 0, 1), • a grading by Z × Z3 • a grading by Z5 • a grading by Z3 2 of type (24, 0, 0, 7), and 3 of type (0, 26), such that for any 0 6= α ∈ Z3 3, Proof. Let Γ : L = Lg∈G Lg be a fine grading of the simple Lie algebra L of type F4, with G its universal group. By Theorem 5.1 and Table 1, either Γ is the Cartan grading, or the free rank of G is two and Γ is associated to a grading by the root system of type G2, or the free rank of G is one, or G is a finite group. Lα ⊕ L−α is a Cartan subalgebra of L. If the free rank of G is 2, the the coordinate algebra (see Table 1) is the Jordan algebra H3(F) of symmetric 3 × 3-matrices. But the results of [BSZ05] show that the neutral component of any grading on H3(F) by any group has dimension at least 3, and this contradicts Proposition 6.1. If the free rank of G is one, then L is graded by the root system BC1 (this includes gradings by A1) and hence L is given by the Tits-Kantor-Koecher Lie algebra constructed from a structurable algebra and Γ is obtained by FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 17 combining the Z-grading given by the rank one root system, and a grading of the coordinate algebra as in Proposition 6.1. A look at the possibilities in [All79, §8] shows that the coordinate algebra is either the Cayley algebra O, with its standard involution, or a structurable algebra defined on the vector space of matrices ( α a For the Cayley algebra, there is a unique grading, up to equivalence, 2, thus obtaining the b β ), with α, β ∈ F and a, b ∈ H3(F). whose neutral component is F1, with universal group Z3 grading by Z × Z3 2. In the second case, the coordinate algebra a = A ⊕ B has dimension 14, with dim B = 1. Moreover, B = Fs for an element s with s2 = 1, and hence B = Bg for an element 0 6= g ∈ tor(G) with 2g = 0. The Lie algebra L decomposes as in (3.4), and the neutral component of the associated grading by the root system of type BC1 decomposes as: L(0) ≃ A ⊕ B ⊕ D ≃ a ⊕ D. (7.3) This is a reductive Lie algebra with one-dimensional center (corresponding to F1) and derived subalgebra simple of type C3. (Actually L(0) is isomorphic to the structure Lie algebra Str(a,−), see [All79, §1].) On the other hand, D is isomorphic to the Lie algebra of derivations of a, which is simple of type A2. The results in [Eld10] show that the simple Lie algebra of type C3 is endowed with a unique grading with trivial neutral component, with universal group Z4 2 and type (12, 0, 3). On the other hand, the simple Lie algebra of type A2 has a unique grading, up to equivalence, with trivial neutral component and whose universal group is 2-elementary. Its type is (6, 1). It turns out that tor(G) is 2-elementary and that at least two of the three homogeneous components of [L(0), L(0)] of dimension 3 intersect the graded subspace a in (7.3) with dimension ≥ 2. We conclude that there is an element 0 6= h ∈ tor(G) such that dim Ah ≥ 2, and h 6= g (recall B = Bg). Since a is a simple structurable algebra, the form hx, yi = trace(Lx¯y+y ¯x) is nondegenerate [AS89]. But hag1, ag2i = 0 unless g1 +g2 = 0. Therefore, the restriction of this form to Ah is nondegenerate. Now, for any two elements x, y ∈ Ah, xy ∈ a0 = F1, so xy = α1 = xy = ¯y ¯x = yx and hx, yi = trace(L2α1) = 2α dim a. We may then find elements x, y ∈ Ah with x2 = 0 = y2 and xy = 1. But then the derivation Dx,y in (3.5) satisfies Dx,y(x) 6= 0, so 0 6= Dx,y ∈ D0, a contradiction with D0 = 0. 7.2, we consider the extended Dynkin diagram: We are left with the case in which G is finite. As in the proof of Theorem ❡ 1 ❡ 2 ❡ 3 i ❡ 4 ❡ 2 and check that either G is an elementary 2-group or 3-group, or the associ- ated quasitorus Q(≃ G) contains an automorphism χ of order 4. In the latter case, the subalgebra of elements fixed by χ is isomorphic to sl(V )⊕ sl(W ) with dim V = 4, dim W = 2 (see [Kac90, Chapter 8]) and the other eigenspaces of χ are, as modules for sl(V )⊕ sl(W ), isomorphic to V ⊗ W , ∧2V ⊗ S2W and V ∗ ⊗ W , with respective eigenvalues √−1,−1,−√−1. The action of any automorphism in the connected subgroup CentAut(L)(χ) is determined by its restriction to V ⊗ W . It is not difficult to check now that 18 ALBERTO ELDUQUE CentAut(L)(χ) is isomorphic to SL(V ) × SL(W )/h±(IV , IW )i (IX denotes the identity map on the vector space X). For f ∈ SL(V ) and g ∈ SL(W ), denote by ψf,g the automorphisms of L such that ψ(f,g)V ⊗W = f ⊗ g. More- over, Γ induces gradings on sl(V ) and on sl(W ) with trivial neutral com- ponents, induced by the projections πV : SL(V ) × SL(W )/{±(IV , IW )} → P SL(V ) = SL(V )/h√−1IV i (contained, up to isomorphism, in Aut(sl(V ))), and πW : SL(V ) × SL(W )/{±(IV , IW )} → P SL(W ) = SL(W )/{±IW}. There is [Eld10], up to equivalence, only one one possibility for such grading on sl(W ), where πW (Q) = h¯g1, ¯g2i, with g1, g2 ∈ SL(W ) or order 2, g1g2 = −g2g1 and ¯gi denotes the class of gi in P SL(W ). With g0 = IW , g3 = g1g2, and Qi . V ,gi Since Q is abelian, ψf,gψf,g′ = ψf ′,g′ψf,g, and it follows from g1g2 = −g2g1, that the elements of Qi V for 1 ≤ i 6= j ≤ 3, and that the elements of Q0 V commute with the elements in any Qi V . Now, there are [Eld10], up to equivalence, only two possibilities of grad- ings on sl(V ) whose associated quasitorus is contained in P SL(V ) and whose In the first of this possibilities, πV (Q) = neutral component is trivial. V = {f ∈ SL(V ) : ψf,gi ∈ Q} we have Q = ∪3 V anticommute with the elements of Qj i=0ψQi h ¯f1, ¯f2i with f1f2 = √−1f2f1 but since any two elements of πV (Q) must either commute or anticommute by the above, this is not possible. In the other possibility πV (Q) = h ¯f1, ¯f2, ¯f ′ 1, f ′ 2i, with f1, f2, f ′ 2 order two ele- 1 and fif ′ 2f ′ 2 = −f ′ 1f ′ ments of SL(V ) such that f1f2 = −f2f1, f ′ jfi for any i, j = 1, 2. We may assume, scaling the elements if necessary, that f1 ∈ Q1 2 must belong to Q0 V , since they commute with both f1 and f2. This is a contradiction, since 1 and f ′ f ′ We conclude that, if G is finite, the maximal quasitorus cannot contain automorphisms of order 4, and hence G is an elementary 2 or 3-group, and the results in [Gri91] prove that either G ∼= Z5 3. The description of the gradings (with the exception of the Cartan grading) and their types appear, for instance, in [Eld09]. (cid:3) V and f2 ∈ Q2 2 anticommute. V . But then, up to scalars, f ′ 2 or G ∼= Z3 1, ¯f ′ j = f ′ 1 and f ′ References [All78] [All79] 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, Comm. Algebra 7 (1979), no. 17, 1835 -- 1875. [ABG02] B. Allison, G. Benkart, and Y. Gao, Lie algebras graded by the root systems [AS89] BCr, r ≥ 2, Mem. Amer. Math. Soc. 751 (2002), x+158 pp. B.N. Allison, R.D. Schafer, Trace form for structurable algebras, J. Algebra 121 (1989), no. 1, 68 -- 80. [BSZ05] Y.A. Bahturin, I.P. Shestakov, M. Zaicev, Gradings on simple Jordan and Lie algebras, J. Algebra 283 (2005), no. 2, 849 -- 868. [BT09] Y.A. Bahturin and M. Tvalavadze, Group gradings on G2 Comm. Algebra 37 (2009), no. 3, 885 -- 893. [BS03] G. Benkart and O. Smirnov, Lie algebras graded by the root system BC1, J. Lie Theory 13 (2003), no. 1, 91 -- 132. [BZ96] G. Benkart and E. Zelmanov, Lie algebras graded by finite root systems and intersection matrix algebras, Invent. Math. 126 (1996), 1 -- 45. FINE GRADINGS AND GRADINGS BY ROOT SYSTEMS 19 [BM92] S. Berman and R.V. Moody, Lie algebras graded by finite root systems and the intersection matrix algebras of Slodowy, Invent. Math. 108 (1992), 323 -- 347. [Bou98] N. Bourbaki, Lie groups and Lie algebras. Chapters 1 -- 3, Elements of Mathemat- ics (Berlin), Springer-Verlag, Berlin, 1998, Translated from the French, Reprint of the 1989 English translation. [Dra12] C. Draper, A non-computational approach to the gradings on f4, Rev. Mat. Iberoam. 28 (2012), no. 1, 273 -- 296. [DM06] C. Draper and C. Mart´ın-Gonz´alez, Gradings on g2, Linear Algebra Appl. 418 (2006), no. 1, 85 -- 111. [DM09] C. Draper and C. Mart´ın-Gonz´alez, Gradings on the Albert algebra and on f4, Rev. Mat. Iberoam. 25 (2009), no. 3, 841 -- 908. C. Draper and A. Viruel, Fine gradings on e6, arXiv:1207.6690v1 [math.RA]. [DV] [Dyn52] E.B. Dynkin, Semisimple subalgebras of semisimple Lie algebras (Russian) Mat. Sbornik N.S. 30(72) (1952), no. 2, 349 -- 462. Amer. Math. Translations Ser. 2 6 (1957), 111 -- 244. [Eld98] A. Elduque, Gradings on octonions, J. Algebra 207 (1998), no. 1, 342 -- 354. [Eld09] Alberto Elduque, Gradings on symmetric composition algebras, J. Algebra 322 (2009), no. 10, 3542 -- 3579. [Eld10] A. Elduque, Fine gradings on simple classical Lie algebras, J. Algebra 324 (2010), no. 12, 3532 -- 3571. [EK12] A. Elduque and M. Kochetov, Gradings on the exceptional Lie algebras F4 and G2 revisited, Rev. Mat. Iberoam. 28 (2012), no. 3, 773 -- 813. [EO08] A. Elduque and S. Okubo, S4-symmetry on the Tits construction of exceptional Lie algebras and superalgebras, Publ. Mat. 52 (2008), no. 2, 315 -- 346. [Gri91] R.L. Griess Jr., Elementary abelian p-subgroups of algebraic groups, Geom. Ded- icata 39 (1991), no. 3, 253 -- 305. [Hes82] W.H. Hesselink, Special and pure gradings of Lie algebras, Math. Z. 179 (1982), no. 1, 135 -- 149. [Hum75] J.E. Humphreys, Linear algebraic groups, Graduate Texts in Mathematics 21. Springer-Verlag, New York-Heidelberg, 1975, xiv+247 pp. [Hum95] J.E. Humphreys, Conjugacy classes in semisimple algebraic groups, Mathemat- ical Surveys and Monographs, 43. American Mathematical Society, Providence, RI, 1995, xviii+196 pp. [Jac79] N. Jacobson, Lie algebras, Dover Publications Inc., New York, 1979, Republica- tion of the 1962 original. [Kac90] V.G. Kac, Infinite-dimensional Lie algebras, Third edition. Cambridge Univer- [PZ89] [Tit66] sity Press, Cambridge, 1990. xxii+400 pp. J. Patera and H. Zassenhaus, On Lie gradings. I, Linear Algebra Appl. 112 (1989), 87 -- 159. J. Tits, Alg`ebres alternatives, alg`ebres de Jordan et alg`ebres de Lie exception- nelles. I. Construction, Nederl. Akad. Wetensch. Proc. Ser. A 69 = Indag. Math. 28 (1966), 223 -- 237. Departamento de Matem´aticas e Instituto Universitario de Matem´aticas y Aplicaciones, Universidad de Zaragoza, 50009 Zaragoza, Spain E-mail address: [email protected]
1002.0015	2	1002	2011-05-23T19:05:11	Filtrations and Distortion in Infinite-Dimensional Algebras	[ "math.RA" ]	A tame filtration of an algebra is defined by the growth of its terms, which has to be majorated by an exponential function. A particular case is the degree filtration used in the definition of the growth of finitely generated algebras. The notion of tame filtration is useful in the study of possible distortion of degrees of elements when one algebra is embedded as a subalgebra in another. A geometric analogue is the distortion of the (Riemannian) metric of a (Lie) subgroup when compared to the metric induced from the ambient (Lie) group. The distortion of a subalgebra in an algebra also reflects the degree of complexity of the membership problem for the elements of this algebra in this subalgebra. One of our goals here is to investigate, mostly in the case of associative or Lie algebras, if a tame filtration of an algebra can be induced from the degree filtration of a larger algebra.	math.RA	math	Filtrations and Distortion in Infinite-Dimensional Algebras Yuri Bahturin a,c,1 and Alexander Olshanskii b,c,2 aDepartment of Mathematics and Statistics, Memorial University of Newfoundland, St. John's, NL, A1C 5S7, Canada bDepartment of Mathematics, 1326 Stevenson Center, Vanderbilt University, Nashville, TN 37240, USA cDepartment of Algebra, Faculty of Mathematics and Mechanics, 119899 Moscow, Russia Abstract A tame filtration of an algebra is defined by the growth of its terms, which has to be majorated by an exponential function. A particular case is the degree filtration used in the definition of the growth of finitely generated algebras. The notion of tame filtration is useful in the study of possible distortion of degrees of elements when one algebra is embedded as a subalgebra in another. A geometric analogue is the distortion of the (Riemannian) metric of a (Lie) subgroup when compared to the metric induced from the ambient (Lie) group. The distortion of a subalgebra in an algebra also reflects the degree of complexity of the membership problem for the elements of this algebra in this subalgebra. One of our goals here is to investigate, mostly in the case of associative or Lie algebras, if a tame filtration of an algebra can be induced from the degree filtration of a larger algebra. Key words: Contents 0 1 Introduction General properties of distortion 1.1 Filtrations on universal algebras 1.2 Some basic facts about distortion 2 7 7 9 1 Partially supported by NSERC grant # 227060-09 2 Partially supported by NSF grant, DMS 0700811, and Russian Fund for Funda- mental Research, grant 08-01-00573 Preprint submitted to Elsevier 1 November 2018 1.3 Distortion and Membership Problem 1.4 Multitude of tame filtrations on algebras 2 Distortion in classical algebras 2.1 Commutative associative algebras 2.2 Free associative algebras and free group algebras 2.3 Free Lie algebras 2.4 Universal enveloping algebras 3 Realizing tame filtrations as degree filtrations under embeddings 3.1 Composition lemmas for associative and Lie algebras 3.2 Tame filtrations in associative algebras 3.3 Tame filtrations in Lie algebras 3.4 Undistorted embeddings in simple algebras 4 Undistorted embeddings in finitely presented algebras 4.1 Monoids 4.2 Unital associative algebras 4.3 More Examples: Distortion of "cyclic" subalgebras References 0 Introduction 13 16 20 20 22 25 26 31 31 34 39 43 48 49 50 58 62 Let A be a linear algebra over a field F . An ascending filtration α = {An} on A is a sequence of subspaces A0 ⊂ A1 ⊂ . . . ⊂ An ⊂ . . . such that A = ∪∞ n=0An and AkAl ⊂ Ak+l for all k, l = 0, 1, 2, . . . Given a ∈ A, the α-degree of a, denoted by degα a, is defined as the least n such that a ∈ An. If B is a subalgebra of A with a filtration α, as above, then we call the filtration β = {B ∩ An} the restriction of α to B and we write β = α ∩ B. In the case of monoids (that is, semigroups with 1) the terms of filtrations are simply subsets, with all other conditions being the same. In the case of groups the terms of filtrations must be closed under inverses. 2 In this paper we will consider finitary filtrations, that is, filtrations all of whose terms are finite-dimensional. If we deal with monoids or groups, the terms of finitary filtrations are simply finite. A generic example is as follows. Let A be a unital associative algebra generated by a finite set X. Then a finitary filtration α = {An} arises on A, if one sets A0 = Span {1} and An = An−1 + Span {X n}, for any n > 1. The α-degree of a ∈ A in this case is an "ordinary" degree with respect to the generating set X, that is the least degree of a polynomial in X equal a. We write degα a = degX a. Such filtration α is called the degree filtration defined by the generating set X. If B is a subalgebra of an algebra A such that A is generated by a finite set X and B by a finite set Y , α = {An} and β = {Bn} are respective degree filtrations then there is t such that Y ⊂ At and then it immediately follows that for any n ≥ 0 we have Bn ⊂ Atn. If β′ = α ∩ B then β is majorated by β′ in the sense of the following definition. Definition 1 Given two filtrations β = {Bn} and β′ = {B′ n} on the same algebra B, we say that β is majorated by β′ if there is an integer t > 0 such that Bn ⊂ B′ tn, for all n ≥ 0. We then write β (cid:22) β′. If β (cid:22) β′ and β′ (cid:22) β then we say that β and β′ are equivalent and write β ∼ β′. Setting B = A in the argument just before the definition, we can say that any two degree filtration on the same algebra are equivalent. In terms of degrees, β (cid:22) β′ if and only if, there is natural t such that degβ ′ b ≤ t degβ b and β ∼ β′ if both degβ ′ b ≤ t degβ b and degβ b ≤ t degβ ′ b, for any b ∈ B. If B is a subalgebra of A, β defined by a finite generating set Y of B, α by a finite generating set X of a, β′ = β ∩ B then inequality β (cid:22) β′ is equivalent to the existence of t ∈ N such that degX b ≤ t degY b, for any b ∈ B. In a similar way one can speak about the degree filtrations and their equiva- lence in other classes of algebras, say, Lie, Jordan, etc. In the case of monoids the degree filtration on a monoid M with generating set X is defined by setting M0 = {1} and, for n > 1, Mn = {x1 · · · xm xi ∈ X, m ≤ n}. But if G is a group with generating set X, then one sets G0 = {1} and Gn = {y1 · · · ym yi ∈ X ∪ X −1, m ≤ n}. In the case of general universal algebras see Subsection 1.4. The rate of growth of a degree filtration defined by a finite generating set on an algebra with a finite set of operations or a linear algebra with a finite set of multilinear operations is bounded from above by an exponential function. In other words, for such filtration {An}, 3 (D) there is c > 0 such that dim An < cn, for all n = 1, 2, . . . For example, in the case of associative algebras, if #X = d < ∞ then dim An ≤ 1 + d + ... + dn ≤ (d + 1)n. The same estimate works in the case of monoids and similar estimates are true for groups and linear algebras with one binary operation, such as Lie, Jordan, alternative. etc. For general case see Subsection 1.4. These observations lead us to the main definition of the paper. Definition 2 A filtration α = {An} on an algebra A satisfying (D) is called a tame filtration. As noted above, any degree filtration is a tame filtration. Moreover, if B is a subalgebra of an algebra A and α is a tame filtration of A then the restriction α∩B is a tame filtration of B. In particular, the restriction of a degree filtration of an algebra A to a subalgebra B is always a tame filtration of B. This simple fact will be used without further comments in several theorems of this paper designed to deal with the following natural questions. (1) Is it true that every tame filtration of an algebra B is equivalent (or equal) to a filtration restricted from the degree filtration of a finitely generated algebra A where B is embedded as a subalgebra? (2) If the answer to the previous question is "yes", can one choose A finitely presented? If not, indicate conditions ensuring that the answer is still "yes". It follows from our examples in Section 1.4, that the set of pairwise inequiva- lent tame filtrations on any infinite-dimensional algebra B is uncountable. At the same time, the set of finitely generated subalgebras of finitely presented algebras over a fixed countable field is countable. Thus, the answer to the second question is generally "no". (Similar argument works in the case of in- finite groups, monoids, etc.) However, there is an extensive class of so called constructive tame filtrations, see definition in Section 4, for which the answer is "yes". Now our answers to the above questions are as follows. Note that an algebra A is called unital if A has identity element 1; a subalgebra B of such A is called unital if 1 ∈ B. (i) In the case of associative and Lie algebras, we prove that a filtration β on an algebra B is a tame filtration if and only if β ∼ α ∩ B where α is a degree filtration on a 2-generator algebra A, in which B is embedded as a subalgebra. 4 (ii) Under the same condition as in (i), we prove that a filtration β on an algebra B is a tame filtration if and only if β = α ∩ B where α is a degree filtration on a finitely generated algebra A, in which B is embedded as a subalgebra. (iii) In the case of associative algebras over any field, which is finitely gener- ated over prime subfield, a constructive filtration β on a (unital) countable algebra B is a tame filtration if and only if β ∼ α ∩ B where α is a de- gree filtration on a (unital) finitely presented algebra A, in which B is embedded as a (unital) subalgebra. The results of (i) and (iii) have their analogues in Group Theory, see [16] and [17], where the author answers questions asked by M. Gromov [9]. If B is a finitely generated subalgebra of a finitely generated algebra A then we say then B is embedded in A without distortion (or that B is an undistorted subalgebra of A) if a degree filtration of B is equivalent to the restriction to B of a degree filtration of A. As will be shown in Section 1.2, then any degree filtration of B is equivalent to the restriction of any degree filtration of A. Thus the property of being undistorted does not depend on the particular choice of finite generating sets in A and B. All this is obviously true in the case of monoids or groups. Some results about distorted and undistorted subalgebras are as follows. (iv) In the case of commutative associative algebras, all finitely generated subalgebras of finitely generated algebras are undistorted. (v) In the case of noncommutative associative algebras, free associative al- gebras of finite rank and also free algebras of finite rank in certain proper varieties of associative algebras have distorted finitely generated subalge- bras. (vi) All finitely generated subalgebras in free Lie algebras of finite rank are undistorted. (vii) Any finitely generated associative, respectively, Lie algebra can be embed- ded without distortion in a simple 2-generator associative, respectively, Lie algebra. An important tool while comparing two finitary filtrations on the same algebra is given by the following. Definition 3 Let α = {An} and α′ = {A′ same algebra A. For each n ≥ 0 we set n} be two finitary filtrations on the distα′ α (n) = min{m An ⊂ A′ m}. 5 The function distα′ respect to α. α thus obtained is called the distortion function of α′ with In terms of distortion functions, two finitary filtrations α and α′ of an algebra A are equivalent if and only if both distα′ α′ are bounded from above by a linear function. The notion of undistorted embedding in terms of distortion functions (see Proposition 2) is equivalent to the following: given two degree filtrations, α in A and β in B, the function distβ α∩B is bounded from above by a linear function. Again, all the above is true in the case of monoids, groups or even more general universal algebras. α and distα Also the following result will be proven in a far greater generality than we state here (see Subsection 1.4). (vii) Let g : N → N be any function. On any finitely generated infinite- dimensional associative or Lie algebra over a field there exist tame fil- trations α and α′ such that distα′ α (n) > g(n), for any n ≥ 1. Given two functions f, g : N → N we say that f is majorated by g and write f (cid:22) g if there is natural t such that f (n) ≤ tg(tn), for all n ∈ N. If both f (cid:22) g and g (cid:22) f then we say that f is equivalent to g and write f ∼ g. The equivalence classes acquire partially ordering if one sets [f ] ≤ [g] once f (cid:22) g. α ∼ distβ ′ Now suppose we are given two pairs of equivalent filtrations: α ∼ β and α′ ∼ β′. Then it is easy to show (see below Claim (5) of Proposition 1) that distα′ β . This allows us, given an embedding ϕ : B → A, where A and B are finitely generated algebras, to define the distortion dist(ϕ) of ϕ as follows. Take any two degree filtrations α in A, β in B and consider the image ϕ(β) under ϕ. Then dist(ϕ) is defined as the equivalence class of distϕ(β) α∩ϕ(B) under the above equivalence relation. In particular, if B is a subalgebra of A, we define distB A = dist(ϕ) where ϕ is the natural embedding of B in A. The definition of this asymptotic invariant is analogous to the definition introduced by Gromov [9] in order to measure the distortion of the lengths of elements of groups that may occur when one group is embedded in another. G = distF [H] In our simple Proposition 3 we note that distH F [G] , where H is a finitely generated subgroup of a finitely generated group G, and F [H], F [G] are the group algebras of H and G, with natural embedding of F [H] in F [G]. Similar is the situation in the case of semigroups. In the case of Lie algebras and their universal enveloping algebras the situation is somewhat different. For a subalgebra M of a Lie algebra L, let U(M) be the associative subalgebra generated by M in the universal enveloping algebra U(L) for L. It is well- known that U(M) is isomorphic to the universal enveloping algebra for M. Given a function f : N → N we define its superadditive closure as the least function f : N → N such that f (n) ≤ f (n), for all n ∈ N, and satisfying 6 f (m) + f (n) ≤ f (m + n). (ix) If M is a finitely generated subalgebra in a finitely generated Lie algebra L L then distU (M ) then distM U (L) is represented by f . There are examples of pairs M ⊂ L for which the above inequality is strict. U (L) . More precisely, if f represents distM L ≤ distU (M ) As we mentioned earlier, any filtration α on an algebra A allows one to define the degree function degα : A → N. If α is a tame filtration, we call degα tame. Tame degrees have all "natural" properties of ordinary degrees with respect to a finite generating set of A. Still, in Subsection 4.3 we show that a particular tame degree can be a pretty wild function, even in the case of very basic associative or Lie algebras. 1 General properties of distortion In this section we consider tame filtrations on universal algebras. In Subsec- tion 1.1 we discuss filtrations on universal algebras. Using this generality, in Subsection 1.2 we establish some basic facts about distortion functions and embeddings with or without distortion. In Subsection 1.3 we discuss connec- tions with some algorithmic decidability problems. Finally, in Subsection 1.4 we provide with a construction that exhibits uncountably many pairwise in- equivalent tame filtrations on virtually any infinite (-dimensional) algebra. 1.1 Filtrations on universal algebras Given a signature Ω, we will be considering general universal algebras with operations in Ω, we called them Ω-algebras, and linear Ω-algebras, which are vector spaces over a field F , with multilinear operations in Ω. Let us indicate some problems which arise when we try to generalize the notion of a filtration {An}, A0 ⊂ A1 ⊂ . . . , on a universal algebra A, which may have nullary or unary operations. Suppose, following the pattern exhibited in Introduction, we require that Ai1 . . . Ainω ⊂ Ai1+...+in, where ω is an arbitrary operation of arbitrary arity n. Then the condition A1λ ⊂ A1, for a unary operation λ can quickly make A1 infinite (or infinite-dimensional) even if A is finitely-generated and λ is the only operation on A. Another kind of confusion may arise if we require A0 . . . A0ω ⊂ A0, for any n-ary operation ω. These problems can be avoided if we give the following. 7 Definition 4 If A is an Ω-algebra, respectively a linear Ω-algebra, then an ascending chain of subsets, respectively, subspaces, α = {An}, n = 0, 1, 2, . . ., is a filtration provided that A = ∪∞ n=0An and for any m-ary operation ω and any a1 ∈ Ai1, . . . , am ∈ Aim one has a1 . . . amω ∈ Ai1+...+im+1. As we can see, in this approach we count the number of operations rather than the number of elements involved in these operations. In the case of rings, semigroups or groups (in the latter case each term T of a filtration is usually assumed symmetric: T −1 = T ), a filtration {An} in the sense of Definition 4 produces a filtration {Bn} in the "common" sense if one sets Bn+1 = An, for all n = 0, 1, . . .. Two filtrations are equivalent in the "new" sense if and only if they are equivalent in the "old" sense. Speaking about the notions of the equivalence of two filtrations or the distortion functions, they are the same as given in Definitions 1 and 3. The notion of the tame filtration remains the same as in Introduction, except that in the case where we do not have the structure of vector space on A (like in groups, monoids, etc.) we simply have to require the existence of an integer c such that #An ≤ cn, for all n ≥ 1. Here and in what follows, given a set M, we denote the cardinality of M by #M. Let us now define the degree filtration of an Ω-algebra, with finite signature Ω as follows. First we write Ω = Ω0 ∪ Ω1 . . . ∪ Ωq where the operations in Ωm have arity m. We identify Ω0 with the subset of elements of A selected by these operations. If X is a generating set for A then the degree filtration {An}, n = 0, 1, 2, . . ., is defined as follows: A0 = X ∪ Ω0 and if n > 0 then An = q[m=1 [ω∈Ωm [i1+...+im+1≤n Ai1...Aimω. In the case of linear Ω-algebras one has to define A0 as the linear span of X ∪Ω0 and replace all unions in the previous formula by vector space sums. It follows from this definition that the degree filtration is a filtration and if X is finite then all sets Ai are finite (-dimensional). Any two degree filtrations α = {An} and α′ = {A′ n} on a finitely generated algebra A are equivalent. Indeed, if α, respectively, α′ is defined by a finite generating set X, respectively, X ′ then there is k such that X ′ ⊂ Ak. Let us set t = kq + 1. Then, using induction, for each ω ∈ Ωm, m ≤ q, and n = i1 + . . . + im + 1, we will have A′ i1 . . . A′ imω ⊂ Amax{ti1,k} . . . Amax{tim,k}ω ⊂ At(i1+···+im)+qk+1 ⊂ Atn. 8 1.2 Some basic facts about distortion We start by listing few general properties of distortion functions. We denote by id the identity function on N and f ◦ g the usual composition of two functions f, g : N → N. Given t ∈ N and f : N → N we denote by t · f the scalar multiple of f by t. We write f ≤ g if f (n) ≤ g(n), for all n ∈ N. Also, given a homomorphism of algebras ϕ : A → B and a filtration α = {An} on A, we denote by ϕ(α) the image-filtration {ϕ(An)} on ϕ(A). Proposition 1 The following facts are true for the distortion functions of filtrations. (1) If µ, ν, π are three filtrations on the same algebra A then (1) (2) If µ, ν are two filtrations on an algebra A and B is a subalgebra of A µ ≤ distπ ν ◦ distν µ; distπ then (3) If C ⊂ B ⊂ A are algebras with filtrations α on A, β on B and γ on C then distν∩B µ∩B ≤ distν µ; distγ α∩C ≤ distγ β∩C ◦ distβ α∩B; (4) If µ, ν are two filtrations on an algebra A, and ϕ : A → B is a homo- morphism of algebras then distϕ(ν) ϕ(µ) ≤ distν µ. (5) If µ ∼ µ′ and ν ∼ ν′ are two pairs of pairwise equivalent filtrations on an algebra A then distν µ ∼ distν ′ µ′. Proof. (1) Let µ = {Un}, ν = {Vn}, π = {Wn}, f = distπ ν and h = distν µ. Then, by definition of the distortion function, for any n we have Un ⊂ Vh(n) ⊂ Wg(h(n)). Since f (n) = min{m Un ⊂ Wm}, it follows that f (n) ≤ g(h(n)), proving our first claim. µ, g = distπ (2) Let µ = {Un}, ν = {Vn}, g = distν∩B µ. Then, for any n, we have Un ⊂ Vh(n). Taking intersections of both sides of the latter containment with B, we obtain Un ∩ B ⊂ Vh(n) ∩ B. By definition of the distortion function, g(n) ≤ h(n), proving our second claim. µ∩B and h = distν (3) This follows by applying (1) and (2). Indeed, let us apply (1) to the fol- lowing three filtrations on C: µ = α ∩ C, ν = β ∩ C and π = γ. We will have distγ (α∩B)∩C . According to (2), distβ∩C α∩B. Since the distortion functions are α∩C. Now one can rewrite distβ∩C (α∩B)∩C ≤ distβ α∩C = distβ∩C β∩C ◦ distβ∩C α∩C ≤ distγ 9 non-decreasing, it follows that distγ as claimed. α∩B, α∩C ≤ distγ β∩C ◦ distβ∩C α∩C ≤ distγ β∩C ◦ distβ (4) Let µ = {Un}, ν = {Vn}, g = distϕ(ν) µ. Then, for any n, we have Un ⊂ Vh(n). Taking the images of the latter containment under ϕ, we obtain ϕ(Un) ⊂ ϕ(Vh(n)). It follows that g(n) ≤ h(n). ϕ(µ) and h = distν (5) Using (1), we can write distβ α ≤ distβ β ′ ◦ distβ ′ α′ ◦ distα′ α . We may assume that there is t such that both distβ It then follows that distβ The converse inequality follows in the same manner. β ′ ≤ t · id and distα′ α′(tn), for all n ∈ N and so distβ α(n) ≤ t· distβ ′ α ≤ t · id. α (cid:22) distβ ′ α′. ✷ The just proved proposition turns out to be instrumental in proving general properties of distortion in algebras. Proposition 2 Let A be an algebra, B and C subalgebras of A such that C ⊂ B. (i) If A is a finitely generated algebra then any two degree filtrations on A are equivalent. (ii) If α and α′ are two filtrations of A and α ∼ α′ then α ∩ B ∼ α′ ∩ B. (iii) If A is finitely generated then all tame filtrations of B obtained by restric- tion from the degree filtrations of A are pairwise equivalent. (iv) Suppose that A and B are finitely generated and α and β are two degree filtrations on A and B, respectively. Then β (cid:22) α ∩ B. Thus B is undis- torted in A if and only if there are two degree filtrations α in A and β in B such that α ∩ B (cid:22) β. (v) Let all A, B and C be finitely generated. If C is undistorted in B and B undistorted in A, then C is undistorted in A. (vi) Let A, B, C be as in (v). If C is undistorted in A then C is undistorted in B. (vii) Let ϕ be a homomorphism of A which is injective when restricted to B. If A and B are finitely generated and ϕ(B) is undistorted in ϕ(A) then B is undistorted in A. In particular, a retract of an algebra is always an undistorted subalgebra. (viii) Suppose A ∈ Q, where Q is a (quasi)variety Q of algebras, ϕ : B → A the natural embedding of B in A, µ : F (X) → A, ν : F (Y ) → B epimorphisms of free algebras of Q, generated by finite sets X and Y . Suppose B is given by some set T (Y ) ⊂ F (Y )×F (Y ) of defining relations. Let π : F (Y ) → F (X) be any homomorphism such that µ ◦ π = ϕ ◦ ν and T (X) = (π × π)(T (Y )). If B is undistorted in A then B is an undistorted 10 subalgebra in an algebra A′ defined in terms of generators X and relations T (X). Proof. (i) The proof of this was given in Subsection 1.1. (ii) By definition, α (cid:22) α′ if and only if there is a linear function f such that distα′ α ≤ f . Thus α ∩ B (cid:22) α′ ∩ B. By symmetry, α′ ∩ B (cid:22) α ∩ B, and our claim follows. α ≤ f . Applying Claim (2) of Proposition 1, we obtain distα′∩B α∩B ≤ distα′ (iii) Follows from (i) and (ii). (iv) Set γ = α ∩ B. Choose a generating set X for A which includes the generating set Y for B. Let α′ and β′ be respective degree filtrations and γ′ = B ∩ α′. Then obviously distγ ′ β ′ ≤ id. Now α ∼ α′, β ∼ β′ as pairs of degree filtrations on the same algebras and γ ∼ γ′ by (ii). Thus there is integral t such that distβ ′ γ ′ ≤ t · id. Applying Claim (1) of Proposition 1, we obtain β , distγ distγ β ≤ distγ γ ′ ◦ distγ ′ β ′ ◦ distβ ′ β ≤ (t · id) ◦ id ◦ (t · id) = t2 · id. So β (cid:22) γ = α ∩ B, as claimed. Thus if we only assume α ∩ B (cid:22) β, we will have α ∩ B ∼ β, proving that B is an undistorted subalgebra in A. Now if B is undistorted in A then α ∩ B ∼ β, and in particular, β (cid:22) α ∩ B. (v) If α, β and γ are degree filtrations in A, B and C, respectively, then by previous claim we only need to establish that distγ α∩C ≤ u·id, for some integral u. By our assumption, we have such constant t for both distγ α∩B. Using Claim (3) of Proposition 1, we obtain β∩C and distβ distγ α∩C ≤ distγ β∩C ◦ distβ α∩B ≤ (t · id) ◦ (t · id) = t2 · id. Selecting u = t2 completes the proof of this claim. (vi) We apply Claim (1) of Proposition 1 to filtrations µ = β ∩ C, ν = α ∩ C and π = γ on C. We will then have distγ β∩C . Using Claim (2) of Proposition 1, we obtain distα∩C . By (iv) there is t ∈ N such that distα∩B β ≤ t · id. Since C is undistorted in A, we may assume that also distγ α∩C ≤ t · id. Since the distortion functions are non-decreasing, we finally get distγ β∩C ≤ (t · id) ◦ (t · id) = t2 · id. Thus β ∩ C (cid:22) γ and by (iv), C is undistorted in B, as claimed. β∩C ≤ distα∩B α∩C ◦ distα∩C β∩C ≤ distγ β (vii) Let α = {An}, β = {Bn} be degree filtrations in A, respectively, B. Then ϕ(α) and ϕ(β) are degree filtrations in ϕ(A), respectively, ϕ(B). By (iv) we only need to show α ∩ B (cid:22) β. Since ϕ(α) ∩ ϕ(B) (cid:22) ϕ(β), there is t ∈ N, such that for any n ∈ N and any b ∈ B ∩ An we have ϕ(b) ∈ ϕ(Btn). Since the 11 restriction of ϕ to B is injective, we have b ∈ Btn. Thus B ∩ An ⊂ Btn and so B is undistorted in A. (viii) From µ ◦ π = ϕ ◦ ν it follows that (µ × µ)(T (X)) = (µ × µ)(π × π)(T (Y )) = (ϕ × ϕ)(ν × ν)(T (Y )) ⊂ (ϕ × ϕ)(diag(B × B)) ⊂ diag(A × A). Thus A satisfies all relations T (X) and so if A′ is defined by generators X and relations T (X) and µ′ : F (X) → A′ is the respective canonical epimor- phism then there is a natural epimorphism ε : A′ → A such that µ = ε ◦ µ′. Finally setting ϕ′(b) = µ′ ◦ π ◦ ν−1(b) correctly defines the desired undistorted embedding of B in A′. Indeed, if f1, f2 ∈ F (Y ) are such that ν(f1) = ν(f2) then f1 = f2 is a relation in B hence π(f1) = π(f2) is a relation in A′ and so (µ′ ◦ π)(f2). This shows that ϕ′ : B → A′ is well defined. Now, for any b ∈ B, ◦ π)(f1) = (µ′ ε ◦ ϕ′(b) = ε ◦ µ′ ◦ π ◦ ν−1(b) = µ ◦ π ◦ ν−1(b) = ϕ ◦ ν ◦ ν−1(b) = ϕ(b). Thus, ϕ = ε ◦ ϕ′ and it follows from the injectivity of ϕ that ϕ′ is also injective. F (Y ) ❅ ν π ❅ ❅ ϕ′ ✲ ✛ A′ µ′ B ✠ ❅ ❅❘ F (X) ❅ ϕ ❅ ε µ ❅❘ ✠ ❄ A Finally, using our previous Claim (vii), we conclude that ϕ′ is an undistorted embedding, provided that ϕ is undistorted. The proof is now complete. ✷ If G is an Ω-algebra then the linear space A = F [G] with basis G is a linear Ω-algebra. Given a filtration α = {Gn} on G, we denote by F [α] the filtration on A whose terms An are linear spans of the respective terms of α. If α is a finitary filtration then also F [α] is finitary. If α is the degree filtration of G defined by a finite set X then F [α] is the degree filtration of A defined by the same set X. Finally, dim F [G]n = #Gn. Now suppose H is a subalgebra of 12 G. Then the linear span F [H] is a subalgebra of F [G] and F [G]n ∩ F [H] = Span {Gn ∩ H}. It now follows that distβ F [α]∩F [H]. This argument works, in particular, in the case of semigroups and semigroup algebras. In the case of groups and group algebras, one can restrict oneself to degree filtrations α defined by symmetric generating sets X, that is X −1 = X. α∩H = distF [β] Proposition 3 Let H be a finitely generated subgroup of a finitely generated group G, F [G] the group algebra of G over a field F , F [H] the group algebra of H naturally embedded in F [G]. Then the distortion of the embedding of H in G is the same as the distortion of the embedding of F [H] in F [G]. The same claim holds valid in the case of semigroups/monoids and their semigroup/monoid algebras. ✷ This latter result leads to a variety of distortions in associative algebras ob- tained with the help of known results on distortion in groups (see [9] in the case of Examples (i) and (ii) and [18] in the case of Example (iii)). We exhibit here just three. Example 1 Let H be cyclic subgroup with generator c in the Heisenberg group G with presentation G = ha, b, c [a, b] = c, [a, c] = [b, c] = 1i. Then the distortion of F [H] in F [G] is quadratic. Example 2 Let H be cyclic subgroup with generator b in the Baumslag - Solitar group G with presentation G = ha, b aba−1 = b2i. Then the distortion of F [H] in F [G] is exponential. Example 3 Let g(k) be the "exponential tower with basis 2 of height k", that is, the function defined by g(1) = 2 and g(k) = 2g(k−1), for k > 1. Let H be the subgroup generated by b in the 1-relator Baumslag group G with presentation G = ha, b (aba−1)b(aba−1)−1 = b2i. Then the distortion of F [H] in F [G] is represented by a function f such that f (n) = g([log2 n]). 1.3 Distortion and Membership Problem In this subsection we elaborate on the following observation: knowing that the distortion of a subalgebra B in an algebra A is not too "ugly" can help one to build an algorithm deciding whether an element a ∈ A is actually an element of B. Our main result holds valid for any finitely generated Ω-algebra. However, additional details arise in the case of linear Ω-algebras and so we will start with this kind of algebras. As before, the term "algebra" will mean linear "Ω-algebra". In this subsection we will be considering linear Ω-algebras over constructive fields. A field F is called constructive if there is an enumeration of the elements 13 of F : a1, a2, . . . such that aiaj = af (i,j) and aiaj = ag(i,j) where f, g : N → N are recursive functions. Let A be a linear Ω-algebra over a constructive field F , with finite generating set X. We say that the Intersection Problem of Finite-dimensional Subspaces (Intersection Problem, for short) is decidable in A if there is an algorithm that allows, given two finite subsets R and S of elements of A, written as "polynomials" in X, with linear spans Span {R} and Span {S} over F , to produce a finite subset T ⊂ A, again written as "polynomials" in X, which is a linear basis of Span {R}∩Span {S}. An equivalent problem is that of decid- ability of linear dependence of finite sets. Further, given a finitely generated subalgebra B of a finitely generated algebra A, we say that the Generalized Membership Problem in B is decidable in A if there is an effective procedure that allows one, for any finite subset S ⊂ A, to effectively produce a finite subset T ⊂ A, such that T is a linear basis of B ∩ Span {S}. Again, S and T are written as "polynomials" in X or, more precisely, the elements of the free linear Ω-algebra F (X) generated by X. Theorem 1 Let A be a finitely generated Ω-algebra where Ω is a finite set of operations. If A is a linear algebra then we additionally assume that the ground field of coefficients F is constructive and that the Intersection Problem is algorithmically decidable in A. Then the following conditions on a finitely generated subalgebra B are equivalent: (1) The (Generalized) Membership Problem for B in A is algorithmically decidable; (2) Any distortion function for B in A is recursive; (3) A distortion function for B in A is bounded from above by a recursive function. Proof. Let α = {An} be the degree filtrations of A and β = {Bn} of B. We restrict ourselves to the proof in the case of linear Ω-algebras, with few remarks about the "non-linear" case at the end of the proof. To prove Implication (1)⇒(2), assume that A is a linear Ω-algebra and B a subalgebra with Generalized Membership Problem effectively decidable. Given any natural n, let us effectively select a finite basis Pn of An, the nth term of α. Using our hypothesis, we can effectively write the finite set Qn which is a linear basis of Kn = B ∩ An, the nth term of the restriction filtration α ∩ B. After this, starting from m = n, we effectively write the linear bases Rn,m of the subspaces Kn ∩ Bm where each Bm is a finite - dimensional space, the mth term of β. The least value of m such that #Rn,m = #Qn equals the nth value of the distortion function distβ α∩B. So this function is recursive, proving our first implication. 14 The implication (2)⇒(3) being trivial, let us prove (3)⇒(1). We assume that distβ α∩B is bounded from above by a recursive function f : N → N, for a fixed choice of degree filtrations α in A and β in B. Let us prove that the Generalized Membership Problem in B is algorithmically decidable. Indeed, let S be a finite subset of A and n ∈ N be a number such that Span {S} ⊂ Am. By our hypoth- esis, then B ∩Span {S} ⊂ Bf (n). Then also B ∩Span {S} ⊂ Bf (n) ∩Span {S}. Since f (n) is computable and Bf (n) has an effectively computable basis Qf (n), using the decidability of the Intersection Problem in A, we effectively find a basis in Bf (n) ∩ Span {S}, hence in B ∩ Span {S}. Thus (3)⇒(1) has been also established, proving our theorem in the case of linear Ω-algebras. In the case of Ω-algebras which are not linear, the logic of the proof is exactly the same except that we do not need to consider any spans of finite sets of monomials but rather the finite sets themselves, and their intersections, which by necessity are constructive. ✷ As an application, let us consider a (quasi)variety Q of linear algebras where there are finitely presented algebras with algorithmically undecidable equality problem (see [11]). Additionally we assume that the Intersection Problem is algorithmically decidable in free algebras of finite rank of Q. As an example, one can consider the varieties of all associative or all Lie algebras, etc. We will now follow the pattern of a well-known group-theoretic construction of Mikhailova [14], whose relevance to the distortion in groups was first men- tioned by Gromov [9]. Example 4 Let B be a finitely presented linear algebra in a (quasi)variety Q, with undecidable equality problem. We write B = F (X)/I where F (X) is a free algebra with free generating set X and I an ideal in F (X) generated by a finite set R. Let us consider the direct product G = F (X) × F (X). Suppose H is a subalgebra in G generated by the elements of the form (r, 0), where r ∈ R, and also by the set of "diagonal" elements {(x, x) x ∈ X}. Clearly, H contains (I, 0). Conversely, if (f, 0) ∈ H, then f must be an element of (I, 0) which follows because (I, 0) is the kernel of the projection of H onto the right factor F (X) in G. As a result, (f, 0) ∈ H if and only if f ∈ I. This also tells us that we cannot effectively write the basis of the intersection Span {(f, 0)} ∩ H. Hence the (Generalized) Membership Problem for H in G is algorithmically undecidable. According to the above Theorem 1, the em- bedding of H in F (X) × F (X) is distorted, moreover the distortion function is not bounded by any recursive function. Replacing F (X)×F (X) by F (X) is not possible in many important (quasi)va- rieties. In other words, free algebras of finite rank in many (quasi)varieties have no finitely generated distorted subalgebras. This is either known (as in the case of groups), or will be shown later in this paper (as in the case of Lie algebras). 15 1.4 Multitude of tame filtrations on algebras As in Introduction, any degree filtration of a finitely generated (linear) Ω- algebra is a tame filtration. Indeed, suppose #X = r, #Ω = p. If an element a ∈ A is in the nth term An of the degree filtration defined by X on an Ω- algebra A then by induction it easily follows that a can be represented by a word in an (r + p)-letter alphabet X ∪ Ω whose length is at most nq + 1, where q is the maximal arity of operations in Ω. Thus #An ≤ C((r + p)q)n, for a certain constant C. Therefore, the growth of the sequence of the numbers #An is at most exponential, proving that any degree filtration is a tame filtration. As it turns out, the degree filtrations are just a tip of an iceberg in the vast array of all tame filtrations. Theorem 2 Let Ω be a finite signature. On any countable Ω-algebra or count- ably - dimensional linear Ω-algebra there is at least continuum of pairwise non- equivalent tame filtrations. Given any function g(n), one can always choose two tame filtrations α = {An} and α′ = {A′ n} so that for the distortion func- tion f = distα′ α one has f (n) > g(n), for all n > 0. Proof. In our proof, we restrict ourselves to the case of Ω-algebras, the case of linear Ω-algebras being quite analogous. Thus, let A be an infinite finitely generated algebra of finite signature Ω, with finite generating set X, #X = r. Let {An} be the degree filtration of A determined by X. For any a ∈ A we denote by d(a) its degree with respect to {An}, that is, d(a) = n, if a ∈ An \ An−1. As noted previously, d(a) is the minimal number of operations necessary to obtain a, starting from X ∪ Ω0. To simplify the notation, we will assume in the future that Ω0 ⊂ X. Now let us fix a real number λ such that 0 < λ < 1. We choose an infinite set W λ of elements w ∈ A with pairwise different values of [d(w)λ], for any w ∈ W λ. Here [r] is the integral part of a real number r. Such set has to exist because A is infinite and each An is finite. We are going to define a new filtration, {Aλ n}, in the following way. We choose a set Y ∪ V λ so that there is bijection ϕ : Y ∪ V λ → X ∪ W λ with ϕ(Y ) = X and ϕ(V λ) = W λ. We define a function dα on absolutely free algebra F of signature Ω with free generating set Y ∪ V λ (notice that if Ω is the group signature then, being nonassociative, an absolutely free Ω-algebra is a much "larger' object than the free group!). The elements of F , we call them words, appear by induction on the number of applications of operations in Ω starting from the elements of Y ∪ V λ. We set dλ(y) = 0, for all y ∈ Y , and if v ∈ V λ, then ϕ(v) = w, for a unique w ∈ W λ, and we set dλ(v) = [d(w)λ + 1]. Using induction, let us assume that we have already assigned the values of dλ to some words u1, . . . , um ∈ F . If ω ∈ Ωm, 1 ≤ m ≤ q, and u = u 1 . . . umω then 16 we set dλ(u) = dλ(u 1) + · · · + dλ(um) + 1. Now, for any n > 0, we define the nth term of a filtration of F by setting F λ n = {u dλ(u) ≤ n}. Clearly, {F λ n } is an ascending filtration in F . Suppose we have shown that this is also a tame filtration. Consider a unique homomorphism ϕ : F → A extending the above bijection ϕ : Y ∪ V λ → X ∪ W λ and set Aλ n is a tame filtration in A. To provide ourselves with continuum of pairwise inequivalent tame filtrations on A, we will later need to prove that {Aλ n ). Then Aλ n = ϕ(F λ n} if λ 6= µ. n} 6∼ {Aµ Let us denote by Sλ u with dλ(u) = n. Let V λ Sλ 0 = X and, using induction on n ≥ 1, Sλ with (S hypothesis about V λ, #Sλ n the sphere of radius n in F , that is, the set of all words n be the set of letters v in V λ with dα(v) = n. Then n)ω imω, for each ω ∈ Ωm (m ≥ 1). By our n)ω = Si1+···+im+1=n Sλ n = S(Sλ n where Sλ i1 . . . Sλ n + 1. n ≤ #Sλ λ n = Sλ n ∪ V λ It suffices to prove by induction on n that #Sλ n ≤ cn, for big enough constant c. However, for our argument using induction on n, we need to prove a stronger inequality #Sλ (n + 1)2 , where the constant c is big enough. The choice of c is dependent on the cardinalities of Ω and X. Since the sphere of radius 1 is finite, let c be such that the above inequality holds for n = 1. n ≤ cn To handle the case of n > 1, we will need the following. Lemma 1 For any m ≥ 1 there is a constant C(m) > 0, such that Xi1+···+im=n mYk=1 1 (ik + 1)2 ≤ C(m) (n + 2)2 , for any n. Proof. Use induction by s and the convergence of the series 1 + ✷ 1 22 + 1 32 + · · · . Assuming our upper bound for the number of words u with dλ(u) ≤ k true for any k < n, we will now estimate the number of words of the form u1 . . . umω where u1 ∈ Fi1, . . . , um ∈ Fim, ω an operation in Ωm and i1 + · · · + im + 1 = n. Using Lemma 1 and the induction hypothesis, this number does not exceed Xi1+···+im+1=n mYk=1 cik (ik + 1)2 ≤ ci1+...+im C(m) (n + 1)2 = cn−1 C(m) (n + 1)2 . 17 tions of letters from V λ. This brings us to 1 +(cid:18)K c (cid:19) To obtain the upper bound for #Sλ n we have to sum all the expressions just obtained over ω ∈ Ω \ Ω0 and add 1 in order to take care of possible contribu- (n + 1)2 . Here K is the sum of all C(m) (with possible repetitions) over all ω ∈ Ω, of arity m ≥ 1. It is quite obvious now that c can be chosen so that the following is satisfied for all n: cn 1 +(cid:18)K c (cid:19) cn (n + 1)2 ≤ cn (n + 1)2 , for all n > 1. Thus we have shown that #Sλ n ≤ also have #Sλ function, as claimed. n ≤ cn, hence the growth of {#F λ cn (n + 1)2 , for all m ≥ 1. For this c we will n } is bounded by an exponential Notice that, for any word u, one has dλ(u) ≥ d(u)λ. This is true for u ∈ Y ∪V λ. Now if u = u1 . . . umω, then, considering λ < 1 and using induction, we obtain dλ(u) =P dλ(ui)+1 ≥P d(ui)λ+1 ≥ (P d(ui))λ +1 = (d(u)−1)λ+1 ≥ d(u)λ. In terms of filtrations, if u ∈ F is such that d(u)λ > n then u 6∈ Fn. Since An = ϕ(Fn), we have that no elements a ∈ A with d(a)λ > n can be contained in An. Now we can prove that the filtrations obtained for different values of parameter λ are not equivalent. Indeed, suppose 0 < λ < µ < 1. Assume that there is natural number t such that the nth term of the filtration defined by λ is a subset of the (tn)th term of the filtration defined by µ, for all n ≥ 1. Let us find a ∈ W λ such that [d(a)λ + 1] = n, for some n such that (n − 1)µ/λ > tn, n. Since d(a) ≥ (n−1)1/λ, which is possible because λ < µ. In particular, a ∈ Aλ it follows that d(a)µ ≥ ((n − 1)1/λ)µ > tn. Hence, a /∈ Aµ tn, a contradiction. This completes the proof of the first claim of our theorem in the case where A is finitely generated. To complete the proof of the first claim in the general case, let us assume that A is not finitely generated but has a countable set of generators {a0, a1, ...} such that for each i ≥ 1, we have ai 6∈ alg {a1, . . . , ai−1}. Again we introduce free Ω- algebra F , now with free generating set {y0, y1, . . .} and define an epimorphism ϕ : F → A extending the bijection ϕ(yi) = ai, for i = 0, 1, . . .. Now let us choose a real number λ ≥ 1 and define a degree function on F by setting dλ(yi) = [ iλ], for all i ≥ 0. As earlier, dλ extends to the whole of F and induces a function on A denoted by the same symbol. Using this function, one defines a filtration {Aλ n} on A such that A0 consists of the values 6∈ alg {a0, . . . , ai−1}, it follows of 0-ary operations and element a0. Since ai that dλ(ai) = dλ(yi) = [ iλ], for i = 0, 1, . . . 18 n ≤ n + 1, because, for n fixed, there is at most one i such that dλ(yi) = n. n} is a tame filtration. Finally, if λ < µ then Now the proof of Lemma 1 goes without any changes, including the fact #Sλ #Sλ As before, it follows that {Aλ {Aλ n}. This follows because dλ(ai)/dµ(ai) → 0 when i → ∞. n} 6∼ {Aµ Now let us fix an increasing function g : N → N. Again, we start with the case where A is finitely generated. We choose an infinite set W ′ of elements w1, . . . , wn, . . . ∈ A so that d(wn) > g(n). Then we consider an absolutely free Ω-algebra F with free generating set Y ∪ V ′ which is in bijection ϕ with X ∪ W ′, as before. Let ϕ be an epimorphism extending the above bijection. In the same manner, as previously, we define the filtration F ′ n, except that now dλ should be replaced by d ′, such that d ′(y) = 0, for all y ∈ Y , and d ′(vn) = n, for all n = 1, 2, . . ., where vn ∈ V ′ is such that wn = ϕ(vn). The proof that {F ′ n) and consider the distortion function f = distα ′ n} with respect to the degree filtration α = {An}. By definition, A ′ n ⊂ Af (n). Since wn ∈ A ′ n \ Ag(n), it follows that f (n) > g(n), for all n, as claimed. This completes the proof of the second claim of our theorem in the case where A is finitely generated. n} is a tame filtration is the same, as before. We now set A ′ α of a tame filtration α ′ = {A ′ n = ϕ(F ′ To complete the proof of the second claim in the general case, we may now assume that A is not finitely generated but has a countable set of gener- ators a0, a1, ... such that for each i ≥ 1, we have ai 6∈ alg {a1, . . . , ai−1}. Again we introduce free Ω-algebra F with free generating set y0, y1, . . . and define an epimorphism ϕ : F → A extending the bijection ϕ(yi) = ai, for i = 0, 1, . . .. Given a function g as above, we choose d(yi) arbitrarily with d(yi) > max{g(k) k ≤ i}. We also set d(ai) = d(yi). Now let us consider two filtrations: α = {Aλ n} defined by d, that is, A′ ✷ n = {a d(a) ≤ n}. In this case, ai ∈ Aλ n} with λ = 1 and α′ = {A′ i \ A′ g(i). Thus, distα′ α > g(n). Corollary 1 Let A be a countable (semi)group or countably - dimensional as- sociative or Lie algebra over a field. Then A has at least continuum of pairwise inequivalent tame filtrations. Given any function g : N → N one can always find two tame filtrations α and α′ in A such that distα′ α (n) > g(n), for all n ≥ 1. ✷ Remark 1 We can be more precise about the cardinality of the set of pairwise inequivalent tame filtrations. First, in the case of countable algebras this cardinality is continuum simply because the cardinality of the countable Cartesian power of finite subsets of a countable is continuum. Second, in the case of countably-dimensional linear Ω-algebras the cardinality of this set can be even greater than continuum, namely as large as the cardi- 19 nality of the ground field F . In Subsection 4.3 we will produce such examples already on the polynomial algebra F [x] in one variable. 2 Distortion in classical algebras In the previous section we have seen that tame filtrations are abound on virtu- ally any infinite(-dimensional) algebra. As announced in Introduction, we will show in Section 3 that in the classical situation of associative and Lie algebras, every tame filtration of a countably-dimensional algebra B is the restriction of a degree filtration of a suitable finitely generated algebra A. But there are important classes of algebras such that if A is a finitely generated algebra of this class and if a tame filtration of a finitely generated subalgebra B is the restriction of a degree filtration of A, then it is equivalent to a degree filtration of B. In other words, all subalgebras of such algebras A are undistorted. As examples, we show that, among others, this property holds for commutative associative algebras and free Lie algebras. Oddly enough, free noncommutative associative algebras may and do have distorted subalgebras. One more topic we study in this section is the connection between the distortion of the embed- ding of Lie algebras M ⊂ L and their associative counterparts U(M) ⊂ U(L). This is interesting when compared with a similar situation with the embedding of groups and their group algebras, see Proposition 3. 2.1 Commutative associative algebras For the proof of our main result in this subsection we use simple facts of the theory of Groebner bases in the commutative algebras. One of the many possible references is [12]. Theorem 3 The embedding of a finitely generated subalgebra in a finitely generated commutative associative algebra has no distortion. Proof. Let A be a finitely generated commutative algebra with generators a1, . . . , ak, B a subalgebra generated by b1, . . . , bℓ, α = {An} and β = {Bn} respective degree filtrations in A and B. We write A as a factor algebra of the polynomial algebra in k + ℓ variables F [x1, . . . , xk, y1, . . . , yℓ]/I, where ai = xi + I, bj = yj + I, for i = 1, . . . , k and j = 1, . . . , ℓ. We introduce an order on the monomials in x1, . . . , xk, y1, . . . , yℓ in the follow- ing way. Let us assume that any xi is greater than any monomial in y1, . . . , yℓ, but xiyj > xr, for any indexes i, j, r. Two monomials in x1, . . . , xk are or- dered by Short-lex, and the same is true for two monomials in y1, . . . , yℓ. In 20 other words, we start comparing two monomials in x1, . . . , xk, y1, . . . , yℓ first by degX, where X = {x1, . . . , xk}. If these degrees are the same then compare by degY , where Y = {y1, . . . , yℓ}. If even these degrees are the same, we begin comparing lexicographically, considering xi > yj for any i, j. We keep in mind that in the theory of Groebner bases it is allowed to introduce the order in many different ways provided that the set of words becomes a totally ordered monoid with minimum 1. Let us consider the Groebner basis of monic polynomials g1, . . . , gs in I with respect to this order. We also choose C > 0 so that degY (gi) ≤ C, for all i = 1, . . . , s. Being Groebner basis in an ideal I means that g1, . . . , gs ∈ I and that the lead- ing monomials lm(g1), . . . , lm(gs) of g1, . . . , gs generate an ideal ℓ(I) spanned by the leading monomials of all elements in I. A linear basis of F [x1, . . . , xk, y1, . . . , yℓ] modulo I is formed by all monomials which are not divisible by any of lm(g1), . . . , lm(gs). Any polynomial g = g(x1, . . . , yℓ) can be reduced modulo I to a unique linear combination of such monomials using transformations of the following kind. If g contains a monomial u = u′u′′ , where u′ = lm(gi), for some i, then we replace u by (u′ − gi)u′′, followed by reduction of like terms. Notice that under these transformations the weighted "degree" C degX(g) + degY (g) of g cannot grow. Indeed, the "degree" C degX + degY of any mono- mial in the expression of gi is at most the "degree" of lm(gi), which follows from the definition of the order and the inequality degY (gi) ≤ C. In particu- lar, if we start with a polynomial g(x1, . . . , xk) of degree d and end up with a polynomial f (y1 . . . , yℓ) of degree m then m ≤ Cd. Now we can show that distβ α∩B(n) ≤ Cn, for any n = 1, 2, . . .. Indeed, let b ∈ An ∩ B. In this case b = g(a1, . . . , ak) where g(x1, . . . , xk) is a polyno- mial of degree degX g = d ≤ n. Let us pass from g(x1, . . . , xk) to a polyno- mial h(x1, . . . , yℓ), which is reduced modulo I. By the above, C degX(h) + degY (h) ≤ C degX g ≤ Cn. Now since b ∈ B, b = f (b1, . . . , bℓ), for a polyno- mial f (y1, . . . , yℓ) of some degree m = degY f . Let us assume that f (y1, . . . , yℓ) is of minimal possible degree with respect to Y , representing b and depending only on Y . We can pass from f (y1, . . . , yℓ) to h(x1, . . . , yℓ) by transformations of the form indicated above. It follows from the definition of the order on the monomials in X ∪ Y , that the leading monomial lm(gi) for an element gi of the Groebner basis for I is a monomial in Y only if gi is a polynomial in Y and the degree of lm(gi) is the greatest possible among all the other monomials in gi. By the minimality of the degree of f (y1, . . . , yℓ), it then follows that h is a polynomial in Y and its degree is the same as the degree of f . Then by the above, m ≤ Cn, proving our claim. ✷ 21 Example 5 One of possible ways to generalize this result is to consider al- gebras satisfying nontrivial polynomial identity (PI-algebras) (for some basic facts about algebras with polynomial identities, see [3]). Unfortunately, a re- sult by Umirbaev [19] shows that a free algebra FV(Y ), freely generated by a finite set Y , in the variety V of associative algebras over computable fields, defined by the following identity [x1, x2][x3, x4][x5, x6][x7, x8] ≡ 0, (2) contains a finitely generated subalgebra B with undecidable Membership Pro- blem. Also, the Intersection Problem is decidable in FV. Indeed, if A(Y ) is the free associative algebra with free generating set Y (see Subsection 2.2 for formal definition) then one may write a vector space decomposition A(Y ) = FV(Y ) ⊕ V (Y ), where V (Y ) is the set of all consequences of (2) in the A(Y ). Since (2) is multilinear, one can write the same equation for each term of the degree filtration of A(Y ): A(V )n = FV(Y )n ⊕V (Y )n. This latter is an equality for finite-dimensional spaces. The space V (Y ) is spanned by finitely many monomials u[v1, v2][v3, v4][v5, v6][v7, v8]w, where the sum of lengths of all words u, v1, . . . , w is at most n. Since A(Y ) is the free associative algebra, we can effectively write the basis of V (Y )n and hence of FV(Y )n. Now, given two finite subsets S and R in FV(Y ), we effectively find n such that S∪R ⊂ FV(Y )n, find the basis of FV(Y )n, as described above, and solve the Intersection Problem for Span {S} ∩ Span {R}. Thus Theorem 1 applies, showing that none of the distortion functions for B in FV(Y ) is bounded by a recursive function, certainly not by any linear function. Therefore, the embedding of B in FV(Y ) is necessarily distorted. It is easy to see that the above argument providing us with an algorithm for the solution of the Intersection Problem works in the case of every variety defined by not only by multilinear but also homogeneous identities within a variety of all algebras, or all associative algebras, or all Lie algebras, etc. In the case of homogeneous rather than multilinear identities one has to complement the set of identities by their partial linearizations. 2.2 Free associative algebras and free group algebras Let X be a nonempty alphabet and W(X) the set of all words (including the empty word 1) in X. One calls W(X) the free monoid with basis (free generat- ing set) X. Let us fix the field F of coefficients and consider the free associative algebra A(X) over F . The elements of A(X) are linear combinations of words in W(X) with coefficients in F . The elements of A(X) are often called (noncommutative) polynomials in X. 22 The degree n = deg f of f ∈ A(X) is the length of the longest word that enters the decomposition of f through W(X) with nonzero coefficient. Thus the degree of a monomial λw, 0 6= λ ∈ F , w ∈ W(X) is just the length of w. If we replace W(X) by the free group F (X) then the resulting algebra F (X) is called a free group algebra. We start by giving a quick proof to an example by U. Umirbaev [21]. Let A = A(x, y, z) be a free associative algebra with free generators x, y, z and C = A(x, y) a (free) subalgebra generated by x, y. Let I be the ideal of C generated by some elements f1, . . . , fk. We denote by B the subalgebra of A generated by the elements zf1, . . . , zfk, x, y, [z, x] = zx − xz and [z, y] = zy − yz. Lemma 2 Let f be an arbitrary element of C. Then f is an element of I if and only if zf is an element of B. Proof. First we assume f ∈ I. Then f is the sum of expressions of the form ufiv, where u, v are monomials in x, y. Let us prove zf ∈ B. It is sufficient to show zufi ∈ B, because v ∈ B. We proceed by induction on deg u with trivial basis when deg u = 0. If u = xu′ then zufi = zxu′fi = [z, x](u′fi) + x(zu′fi). Here [z, x] ∈ B, u′fi ∈ B (because this depends only on x, y), while the second summand is in B by the induction hypothesis. If u = yu′ then we argue similarly but use [z, y] ∈ B. Now we assume that zf ∈ B. Let us prove f ∈ I. For this we notice that zf is homogeneous of degree 1 in z, hence zf is a linear combination of expressions of one of the forms u[z, x]v, u′[z, y]v′ and u′′(zfi)v′′, where u, v, u′, v′, u′′, v′′ ∈ C. Here the left factors u, u′, u′′ can be removed because in the product zf all monomials start with z. In the expressions z(fiv′′) each factor fiv′′ is in I whereas z(fiv′′) itself is in B. Therefore, without loss of generality, we can remove these expressions if we replace f by f − fiw. After doing so, we arrive at the equation [z, x]v +[z, y]v′ = zf , where v, v′ depend on the generators x, y only. A part of this sum, −xzv − yzv′ equals zero because all the monomials should start with z. It follows that v = v′ = 0, which follows because the right C-module generated by xz and yz is free. As a result, [z, x]v+[z, y]v′ = zf = 0, hence f = 0. Considering that this "final" f was obtained by a number of transformations of the form f → f − fiw, we conclude that, for the "original" f we must have f ∈ I. ✷ It is well-known that over any field there exist 2-generator associative algebras with undecidable equality problem [11]. As a result, by Lemma 2, we have an example of a finitely generated subalgebra B in a 3-generator free algebra A with undecidable membership problem. It is very easy to obtain an example of a subalgebra in a 2-generator free associative algebra, with undecidable membership problem. For this, we simply need to effectively embed A in a 23 free associative algebras A′ = A(u, v) with free generators u, v, for instance, by setting x = u2, y = uv and z = v2. Then obviously, there is no algorithm to decide the membership problem for the image of B in A′. It is now very easy to notice that also in any free group algebra F (X) with #X ≥ 2 there is a subalgebra with undecidable membership problem. Indeed, A(X) ⊂ F (X), and there is an obvious algorithm to decide whether an el- ement of F (X) is an element of A(X). Therefore, if B is a subalgebra with undecidable membership problem in A(X), it will also be is a subalgebra with undecidable membership problem in F (X). Using the above in conjunction with Theorem 1 allows us to make the following conclusion. Theorem 4 (Examples) Let A be a free associative algebra A(X) or a free group algebra F (X) with #X ≥ 2 over a field F . Then one can choose a finitely generated distorted subalgebra B in A in such a way that none of the distortion functions for B in A is bounded from above by any recursive function. Proof. The argument in the case of free group algebras being identical to that in the case of free associative algebras, we restrict ourselves to these latter ones. Since the above example of a finitely generated subalgebra with undecidable membership problem in a free associative algebra or F (X) of rank at least 2 does not depend on the field, we first consider a free associative algebra A0 of rank at least 2 over the prime subfield F0 of F . Then F0 is constructive and by Theorem 1 we have a distorted subalgebra B0 in A0. Now let us consider the F -algebra A = F ⊗F0 A0 and its F -subalgebra B = F ⊗F0 B0. Suppose that β0 = {(B0)n} and α0 = {(A0)n} are degree filtrations in B0 and A0, respectively, defined by the generating sets Y and X, respectively, such that the distortion function distβ0 α0∩B is not bounded by a linear function. Then setting β = {F ⊗F0 (B0)n} and α = {F ⊗F0 (A0)n} gives rise to the degree filtrations on B and A, respectively, defined by the generating sets 1 ⊗ Y and 1 ⊗ X, with distβ α0∩B. Therefore, B is a distorted subalgebra of the free associative algebra A over an arbitrary field F . ✷ α∩B = distβ0 Example 6 In Umirbaev's paper [20] there are examples of subalgebras in free Jordan algebras similar to those associative in Theorem 1. As we noted at the end of Subsection 2.1, there is no problem with the Intersection Problem in the case of free Jordan algebras of finite rank, since the variety of Jordan algebras is defined within the variety of all linear algebras by homogeneous identical relations. Thus we conclude that free Jordan algebras of finite rank contain finitely generated distorted subalgebras. 24 Example 7 It is well-known (and follows from Schreier rewriting) that any finitely generated subgroup of a free group of finite rank is undistorted. The same is true in the case of any semigroup A with finite generating set X and balanced defining relations, that is relations of the form u = v where u, v ∈ W(X) have equal lengths: ℓ(u) = ℓ(v). In such a semigroup the length ℓ(g) of any element g is well-defined and for any two elements g, h ∈ A one has ℓ(gh) = ℓ(g) + ℓ(h). If B is a subsemigroup of A generated by a subset Y and g1g2 · · · gn is an element of B then its length with respect to X will be at least n, proving that B is undistorted in A. Thus it follows from Theo- rem 4 that there exist (semi)group algebras of (semi)groups without distorted sub(semi)groups, which have subalgebras whose distortion function is not even bounded by any recursive function. As shown in Proposition 3, this cannot happen for commutative (semi)groups. 2.3 Free Lie algebras Let L = L(X) be a free Lie algebra over a field F with free generating set X = {x1, . . . , xm}. One can view L as a subalgebra in free associative algebra A(X) generated by X, with respect to the bracket operation [a, b] = ab − ba. Thus the degree degX f of an element f ∈ L(X) is its degree as an element of A(X). Each element f has its leading part Lp (f ), which is the homogeneous component of maximal degree. Theorem 5 The embedding of a finitely generated subalgebra in a finitely generated free Lie algebra has no distortion. Proof. By Shirshov - Witt's Theorem [1, Chapter 3] any subalgebra B of a free Lie algebra L is itself free. Moreover, one can choose a free generating set Y in B so that the set Y = {Lp (y) y ∈ Y } is the free generating set of the subalgebra it generates in L. Now let us assume that B is a finitely generated subalgebra of L(X) and that Y = {f1, . . . , fn} is the set of free generators of B. Let the degrees of Lp (f1) = f ′ n be d1, . . . , dn, respectively. 1, . . . , Lp (fn) = f ′ Let us consider another free Lie algebra L(Z) ⊂ A(Z), where Z = {z1, .., zn}. In addition to the ordinary degree filtration defined by Z, with degree func- tion degZ, we can endow L(Z) with another filtration, with degree function dg , defined on the words in Z by induction on the length as follows. We set dg z1 = d1, . . . , dg zk = dk, for the letters of the alphabet Z. If w = uv is a word of degree > 1 then by induction dg u and dg v are already de- fined and we set dg uv = dg u + dg v. Given an element h(z1, . . . , zn) ∈ A(Z) which is a linear combination of words u1, . . . , us, with nonzero coefficients, we set dg h(z1, .., zn) = max{dg u1, . . . , dg us}. Finally, we denote by Lp′(h) the 25 "leading part" of h(z1, . . . , zn) with respect to the new degree dg , that is, the homogeneous component of the above decomposition of degree dg h(z1, .., zn). Now, for any nonzero Lie polynomial h(z1, ..., zn) we consider the equality g(x1, . . . , xm) = h(f1, . . . , fn). Then the leading part of g(x1, . . . , xm) will be Lp′(h)(f ′ n} is a free generating set. n)). This polynomial in X is nonzero since {f ′ 1, . . . , f ′ 1, . . . , f ′ It follows that degX g = dg h ≥ degZ h. Let α be the degree filtration of L(X) defined by X and β the degree filtation of B defined by Y = {f1, . . . , fn}. We just showed that distβ α∩B(n) ≤ n. Thus B is an undistorted subalgebra in L(X). ✷ Remark 2 Since the universal enveloping algebra U(L(X)) is isomorphic to the free associative algebra A(X) (see, e.g. [1, Chapter 3]) and by Theorem 4 free associative algebras of rank ≥ 2 have distorted subalgebras, it follows that universal enveloping algebras of Lie algebras without distorted subalgebras can have distorted subalgebras. This is quite similar to the case of (semi)groups and their (semi)group algebras (see Remark 7 above). However, in both cases just mentioned, the situation is much better when we consider the connection of the distortion of H in G for a finitely generated subgroup H of a finitely generated group G or a finitely generated subalgebra H in finitely generated Lie algebra G with the distortion of a respective subalgebra F [H] in the group algebra F [G] or U(H) in the universal enveloping algebra U(G). In the case of groups this is described in our simple Proposition 3 above, where it was shown that distH F [G] . The case of Lie algebras requires greater effort, and this will be done in the next subsection of the paper. G = distF [H] 2.4 Universal enveloping algebras If L is a Lie algebra with a finitary filtration α = {Ln} then its universal enveloping algebra U(L) is naturally endowed with a finitary filtration U(α) = {U(L)n} where U(L)n = Xn1+···+nm≤n Ln1 · · · Lnm. If L is finitely generated and M a finitely generated subalgebra of L, then one can speak about the distortions distM U (L) . In the case of groups vs. group rings, as shown in Proposition 3, similarly defined distortions are simply equal. In this subsection we will show first that we always have distM L ≤ distU (M ) U (L) and later that this inequality can be strict. For sharper results we need the following. L and distU (M ) 26 Definition 5 Given a function f : N → N, we say that f is superadditive if for any natural n and m we have f (n) + f (m) ≤ f (n + m). For any function f : N → N we set f (n) = maxn1+···+nt=n{f (n1) + · · · + f (nt)} and call f the superadditive closure of f . Clearly, f is the least superadditive function that majorates f . It is easy to check that if f (cid:22) g then f (cid:22) g. This allows us to correctly define [f ] by setting [f ] = [f ]. If B is a finitely generated subalgebra of a finitely generated algebra A then the distortion distB A is called superadditive if distB A = distB A . The main result of this subsection is as follows. Theorem 6 Let M be a finitely generated subalgebra of a finitely generated Lie algebra L. Suppose U(M) is the universal enveloping algebra of M natu- rally embedded in the universal enveloping algebra of L. Then distU (M ) U (L) = distM L . Proof. Let L be a finitely generated Lie algebra with linearly independent generating set S = {a1, . . . , an}, M a finitely generated subalgebra, with a finite generating set T . One can always assume that T ⊂ S. We denote by α the degree filtration of L defined by S, β the degree filtration of M defined by T and f = distβ α∩M . By Claim (5) in Proposition 1 and Claim (i) in Proposition 2 this particular choice of degree filtrations does not affect our conclusion about the connection between distU (M ) U (L) and distU (M ) U (L) . We have Span {T } ⊂ Span {S}, for the linear spans of T and S. Let us select linear bases in the terms of α and β, as follows. Suppose that α has this form: Span {S} = L1 ⊂ L2 ⊂ . . . One can choose the pairs of sets Cn ⊂ Bn such that Bn is a basis of Ln, and Cn a basis of M ∩ Ln, for all n = 1, 2, . . ., by induction on n. Then B = ∪∞ i=1Bi is a basis of L, which contains a basis C = ∪∞ i=1Ci of M as a subset. If we totally order B in some way then 1 and the monomials b1 · · · bs with b1 ≤ . . . ≤ bs, called PBW-monomials, form a basis of U(L), called PBW-basis (PBW stands for Poincar´e - Birkhoff - Witt). Those of the above PBW-monomials in which all b1, . . . , bs are in C form a PBW-basis for U(M) (see the details in [1, Chapter 1]). Let us assign weight 0 to 1 and weight m = m1 +· · ·+ms to each PBW-monomial u = b1 · · · bs, where bi ∈ Bmi, i = 1, . . . , s. Having done so, one can assign weight to each w ∈ U(L) as the maximum of weights of PBW-monomials in its unique expression with nonzero coefficients through PBW-basis. Let us now consider a product of degree s in U(L) of the form u = b1 · · · bs in U(L) where bi ∈ Bmi, for all i = 1, . . . , s. We claim that one can write u as the linear combination of PBW-monomials of weight at most m = m1 + · · · + ms with respect to B. This is true by definition if b1 ≤ · · · ≤ bs. But if bj > bj+1, 27 for some j, then, since {Ln} is a filtration, [bj, bj+1] = Pt i , where all i ∈ B has weight at most mj + mj+1. We ξi ∈ F , for all i, and each element b ′ then can write i=1 ξib ′ b1 · · · bjbj+1 · · · bs = b1 · · · bj+1bj · · · bs + b1 · · · [bj, bj+1] · · · bs = b1 · · · bj+1bj · · · bs + ξib1 · · · b ′ i · · · bs. tXi=1 All the monomials on the right hand side of this equation have weight at most m1 + · · · + ms. If all of them are PBW then we are done. Otherwise, we apply the same transformation to those monomials which are not PBW. After finite number of steps, the process terminates and our claim follows. Let us now consider an associative word w = y1 · · · yd of length d and its value u = s1 · · · sd, where each factor is an element of the generating set S of L, S = B1. Then, by the above, u can be written as a linear combination of PBW-monomials in B, each having weight at most d. Finally, if g(y1, . . . , yn) is an associative polynomial of degree d then it follows that g(s1, . . . , sn) can also be written as a linear combination of PBW-monomials in B each of weight at most d. Now suppose g(s1, . . . , sn) ∈ U(M), where g(y1, . . . , yn) is an associative poly- nomial of degree d. In this case all PBW-monomials in the expression of g(s1, . . . , sn) through PBW-basis B of U(L) are actually PBW-monomials in C. Suppose, c1 · · · ct is one of these monomials. Then c1 ≤ · · · ≤ ct and the sum of the weights of c1, . . . , ct is at most d. If ck has weight wk then we must have ck ∈ Lwk ⊂ Mf (wk) where f is the distortion function for M in L. Being a Lie polynomial in the generating set T of M, ck is an associative polynomial in the generating set T of U(M), of degree ≤ f (wk). Therefore, c1 · · · ct is the value of a polynomial of degree at most f (w1) + · · · + f (wt) ≤ f (d) with respect to the generating set T of U(M). We have proved that if the degree of an element of U(M) with respect to the generating set S of U(L) is n then its degree with respect to the generating set T of U(M) is at most f (n). Now we need to prove that for each natural n there is an element in U(M) whose degree with respect to S is at most n and with respect to T is at least f (n). For a fixed n > 0, let us consider an arbitrary partition of n as the sum of positive integral summands n = n1 + · · · + nm. In each Lni we choose an element ci ∈ C, which has degree f (ni) with respect to T . Such element exists by the definition of the distortion function f . It is allowed to take cj = ci if nj = ni. We reorder the elements c1, . . . , cm to make sure ci ≤ cj if i < j. Let us consider the PBW-monomial c1 · · · cm. By the choice of ci ∈ Lni, its degree in U(L) with respect to S is at most n = n1 + · · · + nm. Proving by 28 contradiction, let us assume that the degree of this monomial with respect to T is less than d = f (n1) + · · · + f (nm), that is, c1 · · · cs is a linear combination of monomials h(s1, . . . , st) in T such that deg h(y1, . . . , yt) < d. Let us write h(s1, . . . , st) as a linear combination of PBW-monomials in C. As before, the weight of each monomial after rewriting in terms of PBW-basis does not in- crease, if compared with the sum of the weights of the factors in h(s1, . . . , st). Therefore, the resulting weight will be less than d because the original weight was less than d. But the final result of our reductions is our PBW-monomial c1 · · · cm, whose weight is f (n1) + ... + f (nm) = d, a contradiction. Thus, for a given n we have found an element in U(M) of degree at most n with respect to S whose degree with respect to T is at least f (n1) + · · · + f (nk). Because the partition of n was arbitrary, we have obtained that the distortion function of U(M) in U(L) is at least f (n), as claimed. ✷ 2.4.1 Examples of subalgebras with non-superadditive distortion In this excerpt we would like to complement Theorem 6 by exhibiting an L 6= distU (M ) example of a Lie algebra L and its subalgebra M such that distM U (L) . By Theorem 6 we only need to provide an example of a subalgebra M of a Lie algebra L such that the distortion distM L is not superadditive. Actually, such examples can be given not only in the case of Lie algebras but also in the case of associative or Jordan algebras, etc. Let ϕ, ψ be two maps on the set S = {v(0), v(1), . . .}, given by ϕ(v(i)) = v(i + 1), for all i, and ψ(v(j − 1)) = v(j2) if j = 22i, for i = 0, 1 . . ., otherwise, ψ(v(j − 1)) = v(0). These settings define the action of the free monoid W(a, b) on S: a acts as ϕ and b as ψ. Let Sn ⊂ S be the ball of radius n with center v(0) for this action, that is, the set of all elements w(v(0)), where w is of length at most n in W(a, b). Let Qn be the ball of radius n with the same center, for the action of the free submonoid W(a). The distortion function for the action of W(a) with respect to the action of W(a, b) by definition equals f (n) = min{m Sn ⊂ Qm}. Now S22i contains ψ(v(22i − 1)) = v(22i+1), and v(22i+1) ∈ Q22i+1 \Q22i+1 −1. It follows that f (n) ≥ n2 for n = 22i. But if m ∈ [n, n2 − 1], and v(m) results from v(0) without transformations of the form v(22i − 1) → v(22i+1), then it is easy to estimate that we will need to apply ϕ and ψ more times compared to the case where this transformation has been applied. (Using this transformation saves 22i+1 − 22i applications of ϕ and all other possible applications of ψ save less than 22i. Here we take into account that all v(i) where i ∈ [n, n2 − 1) are mapped by ψ to v(0).) It follows that if n = 22i and m ∈ [n, n2 − 1] then the vector with the greatest label in Sm is v(f (n) + (m − n)) and so f (m) = f (n) + m − n ≤ f (n) + m. Now let us assume that g : N → N is a superadditive function and f is equivalent to g, that is, f (n) ≤ tg(tn) and g(n) ≤ tf (tn) for some positive 29 integer t. We choose n = 22i > t4(t2 + 1). If m = t2(t2 + 1)n then n ≤ m < n2, and we can write (t2 + 1)f (n) ≤ t(t2 + 1)g(tn) ≤ tg((t2 + 1)tn) ≤ t2f (t2(t2 + 1)n) = t2f (m) ≤ t2(f (n) + m) and so f (n) ≤ t2m = t4(t2 + 1)n. Since f (n) ≥ n2, we immediately arrive at a contradictory inequality n ≤ t4(t2 + 1). Thus the distortion function f (n) is not equivalent to a superadditive function. Example 8 In each of the cases of associative, Lie or Jordan algebras, there exists a finitely generated algebra L with a finitely generated subalgebra M such that distM L is not superadditive. Proof. Let us consider a linear space V over a field F with basis S from the above example and extend by linearity the maps ϕ and ψ to linear transforma- tions of V . Then V naturally becomes a left module over the free associative algebra A = A(x, y) over F with x acting as ϕ and y as ψ. The linear space T = A ⊕ V becomes an associative algebra if we keep the operations of A, the left action of A on V and additionally set v1v2 = 0 and vg = 0, for all and v1, v2, v ∈ V and g ∈ A, with free term zero. One can then make T into a Lie algebra, respectively, Jordan algebra, if one sets [a, b] = ab − ba, respectively, a ◦ b = ab + ba. Since the argument that follows is similar for all three cases, let us consider Lie algebras. Let L = L(x, y) ⊕ V be a Lie algebra generated by x, y, v(0) and M = Span {x} ⊕ V a (Lie) subalgebra of L generated by x, v(0) (it is natural to call Lie algebras like M cyclic, see also Subsection 4.3). The mth term of the degree filtration β of M defined by {x, v(0)} is Span {x} ⊕ Span {Qm}. The intersection of the nth term of the degree filtra- tion α of L defined by {x, y, v(0)} with M is Span {x}⊕Span {Sn}. Therefore, distβ α∩M = f , the function from our previous example. We know that f is not equivalent to any superadditive function. ✷ This result and Theorem 6 allow one to produce the following. Example 9 There exist a finitely generated Lie algebra L and its finitely gen- erated subalgebra M such that distU (M ) U (L) 6= distM L . Proof. This follows because one can choose L and M as in Example 8 so that distM is always superadditive. ✷ L is not superadditive and at the same time, by Theorem 6, distU (M ) U (L) 30 3 Realizing tame filtrations as degree filtrations under embeddings In this section we obtain our central results concerning tame filtrations in both associative and Lie algebras. We show that any tame filtration of a countably dimensional algebra B can be obtained by restriction from a degree filtration of a finitely generated algebra A, where B is embedded as a subalgebra. Then we prove that actually A can be chosen simple. One general remark about the statements and proofs of our theorems in this and the next section refers to "if and only if" claims. As we noted in Intro- duction, the restriction of a tame filtration of an algebra to a subalgebra is always a tame filtration. Therefore, when we claim that a filtration β of an algebra B is a tame filtration if and only if β is equivalent (or equal) to the restriction α ∩ B of a degree filtration of an algebra A where B is embedded as a subalgebra we actually do not need to proof the "if" part. So in each such theorem we will be proving the "only if" part, without further comment. 3.1 Composition lemmas for associative and Lie algebras Let A(X) be the free associative algebra with free generating set X. Given a nonzero polynomial f ∈ A(X), the highest word, with respect to Shortlex, which enters f with nonzero coefficient is called the leading term or the leading word of f and is denoted by f . We call f monic if the coefficient of f in f is 1. In the definitions and lemma that follow, the total order on W(X) does not need to be Shortlex, however it must be a semigroup order, that is, u ≤ v must imply uw ≤ vw and wu ≤ wv, for any word u, v, w ∈ W(X). The remainder of this subsection can be traced back to [6] (see also [8]). Definition A Let W(X) be given a semigroup total ordering, f and g two monic polynomials in A(X), with leading words f and g . Then, there are two kinds of compositions of f and g: (i) If w is a word such that w = f a = bg for some a, b ∈ W(X) with deg(f ) + deg(g) > deg(w), then the polynomial (f, g)w = f a − bg, is called the intersection composition of f and g, with respect to w. (ii) If w = f = agb for some a, b ∈ W(X), then the polynomial (f, g)w = f − agb is called the inclusion composition of f and g, with respect to w. Notice that the leading word of (f, g)w is strictly less than the leading word of f a or gb in the first case and of f in the second. 31 Definition B A set of monic polynomials S is closed under compositions or is a Groebner-Shirshov set if any composition (f, g)w of elements f, g ∈ S can be written in A(X) as (f, g)w = P αiaifibi, where fi ∈ S, αi ∈ F , and the leading word aif ibi of each aifibi is strictly less than w. Composition Lemma For Associative Algebras Let S be a subset of the free associative algebra A(X) over a field F , consisting of monic poly- nomials, I a two-sided ideal generated by S, A = A(X)/I. Then the following conditions are equivalent: (i) S is a Groebner-Shirshov set. (ii) for any element f ∈ I we have f = asb for some s ∈ S and a, b ∈ W(X). (iii) The set of cosets represented by reduced words {u + I u ∈ W(X), u 6= ✷ asb, s ∈ S, a, b ∈ W(X)} is a basis of A. In the case of Lie algebras the notion of a set of Lie polynomials closed under composition is defined as follows. We always view a free Lie algebra L(X) as a Lie subalgebra in A(X) generated by X under the bracket operation [a, b] = ab − ba. The elements of L(X) are called Lie polynomials in X. The leading word f of a Lie polynomial f ∈ L(X) is defined because f is an element of A(X). Also we call a Lie polynomial monic if it is monic as an associative polynomial. The elements of A(X) resulting from X by repeated application of the bracket operation only are often called (higher) commutators. Forgetting brackets on a commutator c produces a unique word w, which is called the (associative) carrier of c. If X is totally ordered then there is so called Shirshov order (cid:22) on W(X) which is lexicographic unless we compare a word with its proper prefix; in that case the prefix is greater than the word. This order induces an order on the set of commutators: [u] (cid:22) [v] if and only if u (cid:22) v. With each word w ∈ W(X) one can associate the set C(w) of its cyclic permutations. If w is greater than any other word in C(w), we call w regular. If f ∈ L(X) then f is always regular. In particular, if c is a commutator then c is regular. A special set of commutators forms a basis of the vector space L(X) over F . Its elements, called basic commutators, are constructed using induction by degree. The elements of X are basic commutators of degree 1. Suppose we have defined basic commutators of degree < n. Let c, d be any basic commutators of degrees k, n − k < n, satisfying (i) d ≺ c (ii) if c = [c1, c2] then c2 (cid:22) d. In this case, we declare [c, d] a basic commutator of degree n. For the basic 32 commutators, the leading word and the associative carrier are the same thing, and the map c → c is the bijection between the set of basic commutators and the set of regular words. In particular, if [c, d] is a basic commutator then [c, d] = cd. The existence of inverse to the above map means that on any regular word one can set brackets so that we obtain a basic commutator. To formulate an important Special Bracketing Lemma by Shirshov (see [6, Lemma 3.10]), we need the following notation. Let u and v be regular associa- tive words u = avb, a, b ∈ W(X). Suppose we have turned u in a basic com- mutator by setting brackets in such a way that one pair of matching brackets embraces a subword vc of u where b = cd. Symbolically, we write [u] = [a[vc]d]. Let us write c = c1 · · · cm where all ci are regular and c1 (cid:22) . . . (cid:22) cm. After- wards, we set brackets in a unique way on v and each ci to obtain basic commutators [v], [c1], . . . , [cm]. Next we form the (left-normed) commutator w = [. . . [[v], [c1]], . . . , [cm]] and replace [vc] by w in [u]. The resulting commu- tator will be denoted by κ(a, v, b). If now g is a monic Lie polynomial with leading word v included as above in a regular associative word u then we pro- ceed exactly as before but when we form w we replace [v] by g. The resulting Lie polynomial will be denoted by σ(a, g, b). If we use this notation then the following is true. Special Bracketing Lemma Let u and v be two regular associative words such that u = avb, a, b ∈ W(X). (i) In the unique setting of brackets on u which makes it into a basic com- mutator two pairs of matching brackets are set as follows: [u] = [a[vc]d], where b = cd, c, d ∈ W(X). (ii) If we use the bracketing of (i) to form the commutator σ(a, [v], b), as ✷ described before this lemma, then σ(a, [v], b) = u. We will use the above notation in the definitions and lemma that follow. Definition C Let f and g be two monic Lie polynomials in L(X) ⊂ A(X). Then there are two kinds of Lie compositions: (i) If w = f = agb for some a, b ∈ W(X), then the polynomial hf, giw = f − σ(a, g, b) is called the composition of inclusion of f and g with respect to w. (ii) If w is a word such that w = f b = ag (then w has to be regular!), for some a, b ∈ W(X), with deg(f ) + deg(g) > deg(w), then the polynomial hf, giw = σ(1, f, b) − σ(a, g, 1) is called the composition of intersection of f and g with respect to w. Notice that the leading word of hf, giw is strictly less than that of f in the first case and each of f b or ag in the second. 33 Definition D A nonempty set S ⊂ L(X) of monic Lie polynomials is closed under composition or is a Groebner-Shirshov set if any composition h = hf, giw of f, g ∈ S with respect to w can be written as h = Pi αiσ(ai, si, bi) where si ∈ S, αi ∈ F , ai, bi ∈ W(X) and aisibi < w, for all i. Proposition GS A set S ⊂ L(X) ⊂ A(X) is Groebner-Shirshov in L(X) if and only if it is such in A(X). Composition Lemma For Lie Algebras Let S ⊂ L(X) ⊂ A(X) be a nonempty set of Lie polynomials in X. Let I be the Lie ideal generated by S in L(X) and J the two-sided associative ideal generated by S in A(X). Then the following conditions are equivalent. (i) S is a Groebner-Shirshov set in L(X). (ii) for any elements f ∈ J we have f = asb for some s ∈ S and a, b ∈ W(X). (iii) The cosets [u] + I, such that u is a regular word in W(X), without sub- words s where s ∈ S, form a basis in factor-algebra L(X)/I. (iv) The cosets u + J, u a word in W(X) without subwords s where s ∈ S, ✷ form a basis in factor-algebra A(X)/J. 3.2 Tame filtrations in associative algebras Let B be a subalgebra in a free associative algebra A = A(X), #X > 1, generated by a set M of words such that no nonempty suffix of any word in M is a prefix of another word in M and also none of the words in M is a subword of another word in M. We call this condition "non-overlapping". Such sets exist and, moreover, one can choose M so that the growth of the number of elements in the set Mn of words of degree ≤ n in M is a function that majorates an exponential function cn, with c > 1, for all sufficiently large n ("exponential sets"). As an example of such set in the case where there are only two variables, one can consider the set of all words x3ywxy3, where ywx has no subwords x3 or y3. But if we do not restrict the number of variables, then, given any natural c, we can produce a non-overlapping set M with #Mn ≥ cn, for all n ≥ 1 if we proceed in the following way. We choose X with #X = c + 2, select two letter x and y and consider the set M of all words xwy such that w does not depend on x, y. For this set M, we would have #Mn ≥ cn, for any n > 2. But since we also want to have #M1 ≥ c and #M2 ≥ c2 we set d = c2, and add to X new variables z1, . . . , zc, z(1) 1 , . . . , z(1) d , u(2) 1 , . . . , u(2) d and to M new words of length one: z1, . . . , zc and of length two: z(1) 1 z(2) 1 , . . . , z(1) d z(2) d . Notice that from the "non-overlapping" property of M, it is immediate that the set U of products of words in M (including the empty word 1) is a free 34 submonoid U ∼= W(M) and the linear span B of U is the free unital associative algebra B ∼= A(M). Two easy properties of U are as follows. If u ∈ U and u = avb, where also v ∈ U, then a, b ∈ U. Also if u, v, w ∈ W(X), w 6= 1 and vw, wu ∈ U then u, v, w ∈ U. Another remark is that any total order on W(X) induces a total order on U. Now notice that given a total semigroup ordering of the words in a free asso- ciative algebra, any ideal I has a Groebner - Shirshov basis. This is simply any basis S of I consisting of monic polynomials and such that the leading words of different elements of the basis are different. So if I is an ideal in B = A(M), with induced order from W, then there is Groebner - Shirshov basis S of I, as an ideal of B. Lemma 3 Let M be a non-overlapping set in the free associative algebra A(X) of rank ≥ 2, B = alg M, I an ideal of B, S a vector space basis of I consisting of monic polynomials such that the leading words of different elements are different. Consider the ideal J = idAI of A generated by I. Then the leading word of each element of J contains the leading word of an element in S, as a subword. Proof. We would easily derive this by Composition Lemma for Associative Algebras (Subsection 3.1) provided that we have checked that the set S is closed under composition. For the intersection composition, we need to take two polynomials f, g ∈ S and assume that there are words a, b ∈ W(X) such that w = f a = bg. Since f , g ∈ U, by one of the above mentioned properties of M, it follows that a, b ∈ U ⊂ B. Then both f a and bg are in I. Each can be written as a linear combination of elements of S with leading word w in both cases. As a result, (f, g)w = P αifi where each fi is in S and fi < w, for all i. Therefore, S is closed under the intersection composition. For the inclusion composition, suppose that w = f = agb, for a, b ∈ W(X). Then, as before, a, b ∈ U ⊂ B and so both f and agb are in I. The same argument, as just before, shows that S is closed also under the inclusion composition (f, g)w = f − agb. Now the conditions of Composition Lemma for Associative Algebras are satisfied. As a result, the leading word f of every polynomial f ∈ J contains as a subword a leading word g of a polynomial g ∈ S ⊂ I. ✷ Adopting the notation of the previous lemma, let us assume that f ∈ J ∩ B. In this case, f is the product of words in M and also by this lemma, f = a¯gb, g ∈ S. Since g is also a product of words in M, it follows that a and b are products of words in M. Finally, agb ∈ I and then f − agb ∈ J ∩ B with a lesser leading word. Using induction by leading word shows that f ∈ I. We now can conclude that J ∩ B = I. Actually, this result is a particular case (when f ∈ B ∩ J) of the following. Lemma 4 Let f be a polynomial in B. Then in the coset f + I there is a 35 polynomial f0, such that f0 ≤ h, for any polynomial h in the coset f + J. In particular, J ∩ B = I. Proof. Let f0 be a polynomial in f + I with leading word minimal possible in A. Proving by contradiction, let us assume that there is h ∈ f + J, such that h < f0. Let us apply Lemma 3 and Composition Lemma for Associative Algebras to f0 − h, which is a nonzero element of J. We have f0 − h = agb, where g ∈ I. But f0 − h = f0 ∈ B, hence, as earlier, a, b are products of elements in M and then agb ∈ I. It follows that f0 can be replaced by a difference f0 − agb whose leading word is lower, in contradiction with the choice of f0. ✷ One of the simplest non-overlapping sets of words in the variables {x, y, z} is the set M = {xyiz i = 0, 1, 2, . . .}. This set can be used to prove that any countably-dimensional algebra B can be embedded in a finitely generated algebra (A.I.Malcev's result, see [13]). Indeed, let b0, b1, b2, . . . be a subset of B that generates B. Consider B = alg M ⊂ A(x, y, z). As noted above, B is a free associative algebra with free generating set M. Then the map xyiz 7→ bi, i = 0, 1, 2, . . ., extends to an epimorphism ν : B → B. Let I be the kernel of ν, J the two-sided ideal of A generated by I, A = A/J and µ the natural epimorphism µ : A → A/J. By Lemma 4, we have J ∩ B = I and so the well- defined map ϕ : B → A given by ϕ(b) = µ(ν−1(b)) is the desired embedding of B in a 3-generator algebra A. For our argument in Section 4, we need a modification of this general result, as follows. Proposition 4 Suppose B is an arbitrary unital countable F -algebra with a finite generating set S and b1, b2, . . . is an arbitrary enumeration of all of its elements (each element may occur infinitely many times) such that degS bi ≤ i, for all i = 1, 2, . . . Then there exists a unital finitely generated F -algebra A containing B as an undistorted subalgebra and elements a, b, c ∈ A, such that 1 = ac and bi = abic (i = 1, 2, . . .). Proof. The argument preceding the statement of this proposition applies, with b0 = 1. If we select in A = A/J three elements a = µ(x), b = µ(y), c = µ(z) and identify B with its image under emebedding ϕ then we will have ac = 1 and abic = bi ∈ A, for all i = 1, 2, . . ., as needed. To prove the unistortedness of this embedding, we recall the generating set S in B and choose the generating set T = {a, b, c} for A. Let β = {Bn} be the degree filtration of B defined by S and α = {An} the degree filtration of A defined by T . We select u ∈ B and assume u ∈ An. In this case, there is f (x, y, z) ∈ A of degree n such that u = f (a, b, c). By Lemma 4 then there is f0(x, y, z) ∈ B whose degree with respect to {x, y, z} is at most n and such that f0(x, y, z) + I = u. In this case we can rewrite f0(x, y, z) in terms 36 of the free generating set M of B: f0(x, y, z) = g(xz, xyz, . . . , xykz). Each monomial of g has the form xyi1z · · · xyimz and i1 + · · · + im + 2m ≤ n. Each such monomial will map under ϕ to bi1 · · · bim and this element has degree at most i1 + · · · + im ≤ n with respect to the generating set S of B. Hence u = ϕ(g) ∈ Bn. Thus, distβ α∩B(n) ≤ n, and the embedding has no distortion. ✷ We now prove the main result of this subsection. Theorem 7 Let B be a unital associative algebra over a field F . (1) A filtration β on B is tame if and only if β ∼ α ∩ B where α is a degree filtration on a unital 2-generator associative algebra A where B is embedded as a subalgebra. (2) A filtration β on B is tame if and only if β = α ∩ B where α is a degree filtration on a unital finitely generated associative algebra A where B is embedded as a subalgebra. Proof. If B is a subalgebra in a finitely generated algebra A, and β ∼ α ∩ B where α is a degree filtration of A then, as noted in Introduction just after Definition 2, β is tame. This proves the "if" parts in Claims (1) and (2). Let us now prove the "only if" part in Claim (1). Suppose that β = {Bn} is a tame filtration on a unital associative algebra B. We will use the notation of the first two paragraphs of the current subsection. Notice that the subalgebra B defined earlier is a free graded subalgebra in a graded algebra A and that m=0 A(m), where A(m) is the linear span of words of length m in X. Since B is generated by m=0 B(m) where B(m) = B ∩ A(m). These gradings induce m=0 A(m) on A and a tame m=0 B(m) on B. We also set M(m) = M ∩ A(m). M is a free graded generating set of B. We write A = L∞ words, we have B =L∞ filtrations: the degree filtration with terms An =Ln filtration with terms Bn =Ln Since the growth of the sequence {#M(n)} majorates an exponential function, there is a positive integral constant C such that dim Bn ≤ #M(Cn) ≤ dim BCn, for any n. This allows us to define an epimorphism ν : B → B so that for each n = 1, 2, . . . , Span nν(M(Cn))o = Bn. We also may assume that ν(M(k)) = 0, for the values of k not divisible by C. Since M(Cn) ⊂ BCn we have Bn ⊂ ν(BCn). The converse inclusion is also easy. Indeed, 37 ν(BCn) = Span  Xl1+...+ls=Cn = Span  Xk1+...+kt=n ⊂ Xk1+...+kt=n ν(M(l1)) · · · ν(M(ls)) ν(M(Ck1)) · · · ν(M(Ckt)) Bk1 · · · Bkt ⊂ Bn. So we have Bn = ν(BCn). Now let I be the kernel of ν. If J is the ideal of A generated by I then by Lemma 4, B ∩ J = I. Let us set A = A/J and suppose that µ is the canonical epimorphism from A to A. The degree filtration of A induces the degree filtration on A given by An = (An + J)/J, for n = 0, 1, 2, . . .. The natural embedding ϕ : B → A is given by ϕ(u) = µ(ν−1(u)). We identify B with its image in A, using ϕ. To complete the proof of our theorem, it is sufficient to show that Bn = B ∩ ACn. Since BCn ⊂ ACn and for any u ∈ Bn we have u = ν(b) = b + I, where b ∈ BCn ⊂ ACn, we have ϕ(u) = eν(ν−1(u) = µ(b) = b + J ∈ ACn. Thus Bn ⊂ B ∩ ACn. To prove the inverse inclusion, we pick µ(a) ∈ B ∩ ACn. Then a can be chosen in ACn and there is u ∈ B such that µ(a) = ϕ(u) = µ(ν−1(u). If u = ν(b), for some b ∈ B, then a + J = b + J. Using Lemma 4, we can find b0 ∈ b + I such that b0 ∈ BCn. Then µ(a) = µ(b0) = ν(b0) ∈ ν(BCn) ⊂ Bn, and so B ∩ ACn ⊂ Bn, as claimed. Now the proof of Claim (1) is complete. To prove the "only if" part in Claim (2), notice that the previous proof works for any number of elements in the set X, #X ≥ 2. But if #X is big enough, as specified in the first paragraph of this subsection, then the growth of the set M can be made faster than any exponential function. Consequently, the constant C that was used to cover Bn by BCn can be take equal 1. In this case we get Bn = B ∩ An and so the original tame filtration on B is simply the restriction of a degree filtration on an appropriate finitely generated algebra A. This complete the proof of Claim (2), hence of the whole theorem. ✷ An argument very similar to the one used in the proof Theorem 7, allows one to derive the following result about the monoids. Theorem 8 Let N be a countable monoid. (i) There exist a 3-generator monoid M where N is embedded as a sub- monoid. (ii) If N is finitely generated then the embedding of N in a 3-generator monoid M can be done without distortion. 38 (iii) A filtration β on N is a tame filtration if and only if there is a finitely generated monoid M with a degree filtration α such that β ∼ α ∩ N. Proof. All claims follow if we replace the occurrences of the word "monoid" by the words "monoid algebra". Notice that free associative algebra A = F [W(X)], its free subalgebra B = F [W(M)] are monoid algebras. To prove Claim (i) we have to use the argument preceding Proposition 3. If the elements {b0, b1, . . .} are the elements in N then ν : B → B to W(M) is an epimorphism of monoids W(M) to N. The kernel I of ν is then generated by the elements of the form u − v where u, v are some elements of M. The same elements generate J and thus A = A/J is a monoid algebra of the 3-generator monoid M = µ(W(X)). The restriction of ψ to N is an embedding of monoids from N to M, proving Claim (i). To prove Claim (ii) we have to apply Proposition 4 in conjunction with Propo- sition 3, which relates the distortion of the embedding of monoids to that of respective monoid algebras. To prove the Claim (iii), we only need to note that tame filtrations on N come from a tame filtrations on B = F [N] and the degree filtration on M comes from the degree filtration on W(X) defined by X. Thus, these filtrations are merely the restrictions of the tame filtration of B to N and the degree filtration of A to M, and the number of elements in the nth term of each of these filtrations for N or M is the dimension of the the respective filtrations for B and A. As a results, our claim about tame filtration follows from Theorem 7. ✷ 3.3 Tame filtrations in Lie algebras Let X be a totally ordered finite set and A(X) the free associative algebra with X as the set of free generators. Let x, y ∈ X and assume that X is ordered in such a way that x is greater than any other letter in X. This ordering expands to one of the total orderings on W(X): Shortlex or Shirshov ordering described in Subsection 3.1. As in the previous subsection, we will be using here an exponentially growing non-overlapping set M in one of two forms: x3ywxy3, where ywx has no subwords x3 or y3, and also, provided that #X > 3, the set of all word xuy, where u ∈ W(X) is a word without letters x and y. Notice that in either case M consists of regular words in the sense of Subsection 3.1. Some of the properties of the words in the free semigroup W(M) have been indicated in the first paragraphs of the previous subsection. Now let N be the set of basic commutators obtained by setting brackets on the elements of M, B the subalgebra in the free associative algebra A generated by M, and C the Lie subalgebra in the free Lie algebra L = L(X) generated 39 by N . As in the previous subsection, we have free monoid U ∼= W(M), the free associative algebra B ∼= A(M) and now the free Lie algebra C ∼= L(N ). Lemma 5 Let u ∈ W(M) be a regular associative word in W(X). If we set brackets on u in a unique way so that the resulting word is a basic commutator [u] in L(X) then [u] ∈ L(N ). Proof. Let us write w = w1 · · · wm, where each wi is an element of M. In the case where M is the set of the first kind, the argument is as follows. We apply Shirshov's method of bracketing (see [6, Lemma 3.10]). It consists of subsequent iterations during each of which we find the smallest word v among those already bracketed (as noted before Special Bracketing Lemma in Sub- section 3.1, the words of length 1 are already bracketed!). For each already bracketed subword u 6= v we look at the maximal number l of bracketed sub- words equal v, which follow u, and introduce the bracketed subwords of the next level by setting brackets on uvl in a left-normed way to obtain bracket- ing [[u, v], v, . . . v ]. Suppose that after t iterations of the process the brackets l {z } l {z } respect the decomposition w = w1 · · · wm. Let us assume to the contrary that after the next iteration, the brackets have been set to give the commutator [[u, v], v, . . . v ], whose associative carrier contains a suffix of wk and a prefix of wk+1. Since the maximal letter x cannot be the last letter of the regular word v, the subword v · · · v must contain the prefix x3 of wk+1. Since uv · · · v is regular, it follows that x3 is a prefix of u. Now since v · · · v contains x3 but does not end by x, it follows that v itself must contain x3. But v is regular and so x3 must be the prefix of v. As a result, u = wk and v is a prefix of wk+1. But Shirshov's bracketing assumes that v is the smallest of the words already bracketed in course of the first t iterations (but before the (t + 1)st begins!). This means that after t steps, our word is of the form [v1] · · · [vt], where each vi is not less than v, hence x3 is the prefix of vi. But then v1 = w1, v2 = w2, . . ., which is what we want. In the case of M of the second kind the argument goes through if we replace x3 by x. ✷ This Lemma and its proof allow one to modify Special Bracketing Lemma (Subsection 3.1). Corollary 2 Let u, v ∈ W(M) be regular associative words, u = avb where a, b ∈ W(X). Suppose g ∈ L(N ) has leading word v. Then σ(a, g, b) is an element of the Lie ideal of L(N ) generated by g. Proof. First of all, by the properties of the set M we have that a, b ∈ M. Using the same method of choosing the smallest letter, etc., as in the pre- vious lemma, we can see that throughout the process of forming κ(a, v, b) = 40 [a[[v], [c1], . . . , [cn]]d] all commutators arising are in L(N ) and then when we replace [[v], [c1], . . . , [cn]] by g we obtain the desired property of σ(a, g, b). ✷ We now denote by I an ideal of Lie algebra C, J the two-sided (associative) ideal of A generated by I, and K the ideal of Lie algebra L, generated by I. Using Zorn's Lemma, one can choose a linear basis S of I in such a way that different elements of S have different leading words. Each element in S is a linear combination of basic commutators and so these leading words are regular associative words. Lemma 6 Let I, J and K be ideals in C, A and L, respectively, as just defined, and f ∈ J or f ∈ K, f 6= 0. Then the leading word f has a subword equal to the leading word of an element in S. Proof. Since K ⊂ J, we only need to consider the case where f ∈ J. Then, by Composition Lemma for Associative Algebras (Subsection 3.1), we need to check that S is closed under associative compositions. In view of Proposition GS (Subsection 3.1), it is sufficient to check this claim for Lie compositions of two elements in S. First, let us consider the inclusion composition hf, giw for two elements f, g ∈ S where w = f = ugv, for some u, v ∈ W(X). Since f ∈ W(M), it follows that in this case also u, v ∈ W(M). By Corollary 2, σ(u, g, v) is an element of I. As such, σ(u, g, v) can be written as a linear combination of elements of S, with a leading basic commutator c. The leading basic word of σ(u, g, v) is then c and hence by Special Bracketing Lemma (Subsection 3.1), c = ugv = f . It follows then that the leading basic commutator in the expression of f is also c and then hf, giw = σ(u, g, v) − f is a linear combination of elements in S, strictly less than f . Second, let us consider the composition of intersection hf, giw for two elements f, g ∈ I where w = f u = vg, for some u, v ∈ W(X). Again, using the same argument, u, v ∈ W(M). By Corollary 2, σ(1, f, u) is in the ideal of L(N ) generated by f and σ(v, g, 1) is in the ideal of L(N ) generated by g. Each of those ideals is in I. As in the previous case, then each of σ(1, f, u), σ(v, g, 1) is a linear combination of elements of S. By Special Bracketing Lemma (Subsection 3.1) the leading words of both Lie polynomials are the same. When we form the composition hf, giw = σ(1, f, u)−σ(v, g, 1), we observe that this is a linear combination of elements of S whose leading words are strictly smaller than w. To comply with Definition D (Subsection 3.1), let us notice that for any s ∈ S we always have σ(1, s, 1) = s. In that case, for both types of composition, we have hf, giw =X αisi =X αiσ(1, si, 1) 41 where each si ∈ S satisfies si < s. In this case also 1 · si · 1 < s and by Definition D, we have that S is closed under composition. Now we conclude that S is a Groebner-Shirshov basis for J and by Com- position Lemma for Associative Algebras (Subsection 3.1) our claim follows. ✷ The next lemma is an analogue of Lemma 4. We will need it to prove a Lie analogue of Theorem 7. Lemma 7 Let f ∈ C and f0 be an element of minimal degree in f + I. Then deg f0 ≤ deg h, for any h ∈ f + J, hence deg f0 ≤ deg g for any g ∈ f + K. Proof. Arguing by contradiction, we consider f0 − h ∈ J. By Lemma 6, f0 = f0 − h contains as a subword the leading word f of an element f ∈ I. We have f0 = uf v. Since f0 ∈ C, we have that f ∈ W(M). As before, then also u, v ∈ W(M). In this case again by Corollary 2, the special bracketing on w = uf v produces an element σ(u, f, v) ∈ I whose leading word is the same as that of f0. Subtracting σ(u, f, v) from f0 we find an element of f0 + I whose leading word is strictly smaller than that of f0, which is a contradiction. ✷ As in the case of associative algebras, an immediate corollary is the following. Recall that I is an ideal of C = L(N ), K the Lie ideal of L(X) generated by I and J the two-sided associative ideal of A(X) generated by I. Lemma 8 J ∩ C = I = K ∩ C. ✷ The next result and its proof are Lie algebra analogues of Theorem 7. This is the best result about general tame filtrations on Lie algebras in this paper. Notice that if a Lie algebra H is a Lie subalgebra in an associative algebra R with tame filtration {Rn} then {H ∩ Rn} is always a tame filtration in H. Theorem 9 (1) A filtration χ on a Lie algebra H is tame if and only if χ ∼ γ ∩ H where γ is the degree filtration on a 2-generator Lie algebra G where H is embedded as a subalgebra, if and only if χ ∼ ρ ∩ H where ρ is the degree filtration on a 2-generator associative algebra R where H is embedded as a Lie subalgebra. (2) A filtration χ on a Lie algebra H is tame if and only if χ = γ ∩ H where γ is the degree filtration on a finitely generated Lie algebra G where H is embedded as a subalgebra, if and only if χ = ρ ∩ H where ρ is the degree filtration on a finitely generated associative algebra R where H is embedded as a Lie subalgebra. Proof. As in all similar theorems (see notes after Definition 2 and just before this theorem), it suffices to prove the "only if" claim in both (1) and (2). Let us start with Claim (1). Suppose χ = {Hn} is a tame filtration of H. Similarly 42 to the approach of Theorem 7, we consider the free associative algebra A(X), #X = 2, the non-overlapping set M whose growth majorates an exponential function, the set N of basic commutators obtained by setting brackets of words from M, a free associative algebra B ∼= A(M) and a free Lie algebra C ∼= L(N ). The natural degree filtrations {An} of A(X) and {Ln} of L(X) by restriction induce tame filtrations {Bn} on B and {Cn} on C. Using the same constant C and the argument that follows in Theorem 7, we define an epimorphism θ : C → H such that Hn = θ(CCn). Let I be the kernel of this homomorphism, K the Lie ideal of L(X) generated by I and J the two- sided associative ideal generated by I. We set G = L(X)/K and R = A(X)/J. Both these algebras have degree filtrations γ = {Gn} and ρ = {Rn} defined by the natural images of X. By Lemma 8, H naturally embeds in both G and R and using Lemmas 7 and 8 in place of Lemma 4 allows us to conclude that χ ∼ γ ∩ H and χ ∼ ρ ∩ H. This takes care of Claim (1). Now since constant C in our present proof is the same as in the proof of Theorem 7, we can easily pass from present Claim (1) to present Claim (2) in the same manner as we did in the case of associative algebras. ✷ 3.4 Undistorted embeddings in simple algebras In this subsection we will discuss the possibility of undistorted embedding of an algebra as a subalgebra in a simple algebra. Note the many important results about embedding of algebras in simple algebras have been obtained by L.A.Bokut', starting with [5], and his coauthors. One of the most up-to-date sources, which also contains an extensive list of references, is [7] (the behavior of filtrations is not among the questions studied in those papers). Theorem 10 Any finitely generated associative, respectively, Lie algebra can be embedded without distortion in a 2-generator simple associative, respec- tively, Lie algebra. Proof. First let B be a finitely generated associative algebra with generators a1, . . . , am and β = {Bn} the respective degree filtration. Let us add to the above generators two more: x, y, and consider an algebra A given by the set of generators a1, . . . , am, x, y, and a set of defining relations which is the union of the set S of ALL relations of B and a set R of additional relations which we are going to define next. By α = {An} we denote the degree filtration of A defined by the generating set {a1, . . . , am, x, y} To start with, we consider an auxiliary set M in alphabet {x, y} satisfying the same conditions as the set M in Section 3.2: no word in M is a proper subword of another word in M and no proper prefix of a word in M can be 43 a proper suffix of another word in M. Next we introduce a well-order on the words in free monoid W(a1, . . . , am, x, y) as follows. First we set a1 < a2 < . . . < am < x < y. For words of arbitrary length, we write u < u′ if either the length of u is less than the length of u′ and in the case of equality, if u is lesser than u′ lexicographically (ShortLex). Now let {f1, f2, . . .} be the list of all monomials in a1, . . . , am, x, y. We will introduce defining relations, two at a time, for each i = 1, 2, . . ., provided that fi has no subwords equal to the leading words of relations in S and previously introduced relations of the set R under construction. If this condition is met, 6= vi are we first introduce a new relation uifivi = 1, where the "wings"ui arbitrary words in M such that the length of each of the ui and vi is at least twice the degree of fi and of any of previously introduced relations, with the exception of the relations from S. Then we introduce u′ i = 0 with the same conditions on the new "wings" u′ i 6= v′ i and additionally we will require that the length of each of the new wings u′ i, v′ i is at least two lengths of each of the old ones. ifiv′ Notice that S ∪ R is complete under composition (see Definition A). Indeed, the set of "old" relations S was complete from start, as the set of ALL relations of B. Any "new" relation, that is, from R, has the form uf v − g, where g is one of polynomials 0 or 1. The leading word of such relation is uf v, where u, v are monomials in x, y. Therefore no partial overlapping with the leading words of the "old" relations is possible. It could be possible that a leading word of an "old" relation is a subword in f . But we deliberately excluded such monomials f in the process of construction of R. Finally, the leading words of "new" relations cannot overlap or be subwords of each other, thanks to the choice of the set M of words in the variables x, y. Thus, the closure of S ∪ R under composition is itself, as needed. To proceed further, first notice that A is nonzero. Indeed, since 1 does not contain any leading words of the relations in S ∪ R, by Composition Lemma For Associative Algebras (Subsection 3.1) it follows that 1 6= 0 and so A 6= {0}. To show that A is simple, it is sufficient to find, for each nonzero in A polynomial f , two monomials u, v such that uf v = 1. Let us write f as a linear combination of monomials f (i) not containing leading words of relations in S ∪ R. If the number of summands in this expression is 1 then by our construction, there are u, v ∈ U such that S ∪ R contains uf v = 1. Then the ideal I of A containing f must contain 1 and thus be equal to the whole of A. Let us now assume that f is a linear combination of monomials f (1), f (2), . . ., f (n) of the form f = λ1f (1) + λ2f (2) + · · · + λnf (n) where all coefficients λ1, . . . λn are nonzero and f (1) > f (2) > . . . > f (n), in the sense of the ordering introduced in the second paragraph of the present proof. One of the relations introduced by us was u(1)f (1)v(1) = 0. Then u(1)f v(1) will be the 44 linear combination u(1)f v(1) = λ2u(1)f (2)v(1) + · · · + λnu(1)f (n)v(1) ∈ I of less than n summands. We need to show that none of the monomials u(1)f (i)v(1), i = 2, . . . , n, in the previous equation contains a leading word of a relation in S ∪ R, as a subword. Proving by contradiction, let us assume, say, that u(1)f (2)v(1) contains such leading word w. If w has no letters x, y then w is a subword in f (2), which is impossible. Otherwise, w = ufiv, for some i, where the "wings" u and v are in the set of non-overlapping words M. As just above, we cannot have w completely inside f (2) and so either u has overlapping with u(1) (hence, equal to u(1)) or v has overlapping v(1)(hence, equal to u(1)), or both. In either case, by the nature of defining relations in R, we must have fi = f (1) and w = u(1)f (1)v(1). Since this is a subword of u(1)f (2)v(1), the length of f (2) must be at least the length of f (1). Since f (1) > f (2), f (2) cannot be longer than f (1). So their lengths are the same and hence f (1) = f (2), a contradiction. As a result, our ideal I generated by f contains a linear combination of n − 1 nonzero monomials with nonzero coefficients which allows us to apply induc- tion and conclude that I = A. To prove that the embedding of B in A is undistorted it is sufficient to show that if f (a1, . . . , am) ∈ Bk \ Bk−1 for some k and f (a1, . . . , am) = g(a1, . . . , am, x, y) then g(a1, . . . , am, x, y) 6∈ Ak−1. In our proof by contradic- tion, we will additionally assume without loss of generality that f (a1, . . . , am) is different in B from a polynomial with lesser leading word. Since the differ- ence h = g(a1, . . . , x, y) − f (a1, . . . , am) equals 0 in A, we can apply Composi- tion Lemma For Associative Algebras (Subsection 3.1), and then the leading word of h must contain as a subword the leading word of one of defining rela- tions. Now by our assumption, g(a1, . . . , am, x, y) ∈ Ak−1. Hence the leading word of h equals the leading word of f and then f equals in B to a polynomial with lesser leading word, which is a contradiction. Similar construction applies also in the case of Lie algebras. Again, given a Lie algebra M with generators a1, . . . , am, m ≥ 2, and ALL defining relations S, we add two new variables x, y and define a Lie algebra L by generators a1, . . . , am, x, y and a set of defining relations S ∪ R, where R is defined as follows. We call here a1, . . . , am, x, y suitable commutators; by induction [u, v] is called suitable, if u, v are suitable and ¯u > ¯v. Notice that all basic com- mutators are suitable. We enumerate all suitable commutators : g1, g2, . . .. The associative support of each suitable commutator is an associative regular word. Then we consider the set M = {xt(xy)ty2 t = 2, 3, 4, . . .}, similar to the language P in [2, Theorem 1], satisfying the non-overlapping condition. Next 45 we totally order the free monoid W(a1, . . . , am, x, y) so that x is the largest variable. We can easily observe that M consists of regular words. Afterwards, we impose brackets on (regular) words of the set M to produce the set N of basic commutators [u], for u ∈ M. , v(j) i Now for each gi such that gi has no subwords equal to the leading words of relations in S and already introduced relations of R, we add to R the total number of m+3 relations as follows. We consider 2(m+3) words u(j) i ∈ M, j = 0, 1, . . . , m + 2, satisfying the following conditions. The length of each of u(0) is at least twice the length of gi or of any of previously introduced i relations except those in S. Also the length of each of u(j) is at least twice the length of each of u(j−1) , for j = 1, . . . , m + 2. Finally, we require that gi (cid:22) v(j) , for all j. With these conditions in place, all the commutators [u(j) ] obtained by setting brackets on these 2(m + 3) words are suitable. Also, by our definition of suitable commutators, each commutator [[[u(j) ]] is suitable. Now the relations of R added on the ith step are [[[u(0) ]] = 0, [[[u(1) i ≺ u(j) ], [v(j) ]] = a1, . . ., [[[u(m+2) ], gi], [v(j) ], gi], [v(0) i ], gi], [v(m+2) or v(j−1) ], gi], [v(1) i and v(j) i ]] = y. i i i i i i i i i , v(0) i i i i Notice that S ∪ R is closed under Lie composition (see Definition D). Indeed, the set of "old" relations S was complete from start, as the set of ALL relations of M. Any "new" relation, that is from R, has the form [[u, g], v] − h, where h is one of polynomials 0 or a1, . . . , y. The leading word of this relation is ugv, where u, v ∈ M. Therefore no overlapping with the leading words of the "old" relations is possible. It could be possible that a leading word of an "old" relation were a subword in g. But we deliberately excluded such commutators g in the process of construction of R. Finally, the leading words of "new" relations cannot overlap or be subwords of each other, thanks to the choice of the set M of words in the variables x, y. Thus, the closure of S ∪ R under composition is itself, as needed. To proceed further, first notice that L is nonzero. Indeed, since x does not contain any leading words of the relations in S ∪ R, by Composition Lemma for Lie Algebras (Subsection 3.1), it follows that x 6= 0 and so L 6= {0}. To show that L is simple, it is sufficient, for each nonzero Lie polynomial f , to find m + 2 monomials u(1), v(1), . . . , u(m+2), v(m+2) ∈ M such that [[[u(1)], f ], v(1)] = a1, . . . , [[[u(m+2)], f ], v(m+2)] = y. Notice that thanks to Composition Lemma for Lie Algebras (Subsection 3.1), each elements of L is a linear combination of basic commutators g(i) such that g(i) has no subwords equal to the leading words of relations in S ∪ R. If f is a suitable commutator itself, then by our construction, such u(1), v(1), . . . , u(m+2), v(m+2) ∈ M exist and then the ideal I of L containing f must contain all generators of L, hence be equal to the whole of L. Any element f ∈ L can be written as a reduced linear combination f = λ1g(1)+λ2g(2)+· · ·+λng(n) of suitable (even basic) commutators g(1), . . . , g(n) 46 with nonzero coefficients which additionally satisfies g(1) > g(2) > . . . > g(n) and such that none of g(1), g(2), . . . , g(n) contains a leading word of a relation in S ∪ R, as a subword. Any reduced combination of length ≥ 1 is different from 0 in L according to the Composition Lemma for Lie Algebras. We need to prove that if a reduced linear combination as above is an element of an ideal I of L then I = L. Let us use induction by n. We already handled the case n = 1. Now suppose we already sorted out the case of linear combinations of length at most n − 1 and we now deal with a linear combination f of length n > 1, as before. Now one of the relations introduced by us was [[[u(1)], g(1)], [v(1)]] = 0, for appropriate u(1), v(1) ∈ M. Let us set f ′ = [[[u(1)], f ], [v(1)]]. Then f ′ = λ2[[[u(1)], g(2)], [v(1)]] + · · · + λn[[[u(1)], g(n)], [v(1)]] ∈ I. Since by construction, g(2) < g(1) (cid:22) v(1) ≺ u(1), all commutators on the right hand side of the latter equation are suitable. The leading words of these commutators are regular words u(1)g(2)v(1), . . . , u(1)g(n)v(1). As in the case of associative algebras, none of these leading words contains the leading word of a defining relation in S ∪ R, as a subword. We also have [[[u(1)], g(2)], [v(1)]] > . . . > [[[u(1)], g(n)], [v(1)]]. As a result, any ideal I con- taining f contains also f ′ which is a reduced linear combination of length n−1, and so f ′ 6= 0. This allows us to apply induction hypothesis and conclude that I = L. The undistortedness of the embedding of M in L follows by exactly the same argument, as in the "associative portion", if we apply Composition Lemma for Lie Algebras (Subsection 3.1) in place of its associative counterpart. ✷ In combination with our previous results, the theorem just proved yields the following. Corollary 3 A filtration β on a unital associative, respectively, Lie algebra B over a field F is tame if and only if β ∼ α ∩ B where α is a degree filtration on a finitely generated simple unital associative, respectively, Lie algebra A over F . Proof. As usual, no need to prove the "if" claim of the theorem. Using Theorem 7 in the case of associative algebras and Theorem 9 in the case of Lie algebras, we embed B in a finitely generated (associative, Lie) algebra C with a degree filtration γ = {Cn} so that β ∼ γ ∩ B. This would mean that distβ γ∩B ≤ t · id, for some integer t. Afterwards, using Theorem 10, we embed C in a simple (associative, Lie) algebra A with degree filtration α so that γ ∼ α ∩ C. Then distγ α∩C ≤ u · id, for some other integer u. It remains to apply Claim (3) of Proposition 1 to see that distβ α∩B ≤ (tu) · id. Thus β ∼ α ∩ B, as claimed. ✷ 47 Remark 3 A quick look at our proof of Theorem 10 reveals that we can replace the word "simple" in the statement by "divisible" which we understand as follows. A unital algebra A is "divisible" if for any nonzero a there exist p, q ∈ A such that paq = 1. Such algebras form a subclass of the class of simple algebras defined within first order logic. It is known that simple algebras do not form such a class. Note that the embeddings suggested in [5] also enjoy this property. 4 Undistorted embeddings in finitely presented algebras As we noted in Introduction, not every tame filtration β of an infinite - di- mensional finitely generated algebra B is equivalent to the intersection α ∩ B where α is a degree filtration of a finitely presented algebra A where B is em- bedded as a subalgebra. This explains the necessity of restrictions we impose on algebras and their filtrations in this section. For example, if a finitely generated algebra B is embedded as a subalgebra in a finitely presented algebra A, both over a constructive field F , then it is well-known that B can be defined by a recursively enumerable set of defining relations. So to cover all important cases we have to impose the condition of recursive enumerability of defining relations of B. Then, by the above remark, there are two many tame filtration on a given infinite(-dimensional) algebra, so we have to select a reasonably narrow countable subset of filtrations β for which we could expect the desired results true (see constructive filtrations below). Finally, a very strong practical reason is that, in general, the embedding the- orems of algebras in finitely presented algebras are known only in few cases, like groups (G. Higman's theorem [10]), semigroups (Murskii's theorem [15]) and associative algebras (Belyaev's theorem [4]). Unfortunately, Lie algebras are not on this list (see this story in [11]). In the case of an algebra B without the structure of a vector space (such as semigroups or groups, etc) the constructivity of a finitary filtration β = {Bn} means that there exists an algorithm that lists out the pairs (bi, degβ bi), containing all elements of B. In the case of linear algebras, this list must contain all pairs (bi, degβ bi) where bi's are all elements of a β-basis B. Here by β-basis B of B we mean a union B = ∪nBn where Bn is a basis of Bn containing Bn−1. We have the following "constructive" modifications of our results in the pre- vious section. They serve as a tool in the proof of Theorems 12 and 14, the latter being our central result in this section. 48 Lemma 9 Let B be an associative algebra over a field F with a recursively enumerable set of defining relations. A constructive filtration β on B is a tame filtration if and only if β = α ∩ B where α is a degree filtration on a finitely generated associative algebra A with recursively enumerable set of defining relations, where B is embedded as a subalgebra. Proof. We simply have to adapt the proof of Theorem 7 but notice that the homomorphism ϕ in the proof follows the algorithm of enumerating the ele- ments of the β-basis B (with their β-degrees). Now the defining relations of A are the defining relations of B in which the elements of the free generating set M of B are replaced by their expressions as words in X. Since by construc- tion we have an algorithm for enumeration of the preimages under ϕ of the elements of B, and an algorithm for enumeration of the defining relations for B, we have an algorithm for enumeration of defining relations for A. ✷ Similarly, in the case of monoids, the adaptation of the proof of Theorem 8, gives the following. Lemma 10 Let N be a recursively presented monoid. A constructive filtration β on N is a tame filtration if and only β ∼ α∩N where α is the degree filtration in a finitely generated monoid M with recursively enumerable set of defining relations, where B is embedded as a submonoid. ✷ We start our treatment of the possibility of writing tame filtration in terms of degree filtrations of finitely presented algebras with the case of monoids. 4.1 Monoids Suppose a monoid M is given by a set of generators X and a recursively enumerable set of relations. In the paper [15] it is proven that M embeds in a finitely presented monoid M ′ with a set of generators X ′ that includes X. Actually, a more precise result is true. Theorem 11 Any finitely generated monoid M with a recursively enumerable set of defining relations can be embedded in a finitely presented monoid M ′ as an undistorted submonoid. Proof. We will use the argument from [15]. However, since we need additional information about the lengths X ′ of the words in monoids M and M ′ arising in Murskii's proof with respect to their generating sets X and X ′, respectively, we need few comments about the course of Murskii's proof. X and The alphabet X ′ includes, in particular, a copy fX of alphabet X, hence each word p in alphabet X has a copy ep. The next claim is contained in [15, Lemmas 49 3.3 and 3.1]. In monoid M ′, if a word p in alphabet X is equal to a word w in alphabet X ′ then w = p0u1p1u2 · · · pl so that: (1) each pi is a word in the alphabet X; (2) after deleting from each ui some subwords we obtain a word u′ i, containing a subword of the form eqi, where qi is a word in alphabet X such that the (3) The word p0q1p1 · · · qlpl equals p in monoid M. following holds: Using (1) - (3), one can compare the length pX of an element of M, repre- sented by a word p and the length wX ′, where w represents the same element in M ′. We have: pX ≤ p0X + q1X + . . . + plX ≤ p0X ′ + q1X ′ + ... + plX ′ ≤ p0X ′ + u′ 1X ′ + · · · + plX ′ ≤ p0X ′ + u1X ′ + · · · + plX ′ = wX ′. Since w is an arbitrary word equal p in monoid M ′, we have an inequality pX ≤ pX ′, that is, what we wanted to achieve. ✷ If we combine Theorem 11 with Lemma 10, and use Claim (3) of Proposition 1, then we obtain the following. Theorem 12 A constructive filtration β on a monoid N is tame if and only if β ∼ α ∩ N where α is a degree filtration on a finitely presented monoid M where N is embedded as a submonoid. ✷ We now switch to to unital associative algebras. 4.2 Unital associative algebras In this subsection we consider unital associative algebras with recursively enu- merable set of defining relations over fields finitely generated over prime sub- field. It was proven in a paper by Belyaev [4] that every associative algebra A with a recursively enumerable set of defining relations, over a unital commutative and associative ring K or a field K that is finitely generated over its prime subfield, can be isomorphically embedded in a finitely presented algebra B over K. Belyaev's proof does not guarantee that the embedding is undistorted, and neither is it unital. In what follows, we base on Belyaev's proof to produce a new proof ensuring that both properties are satisfied. Theorem 13 Let B be an arbitrary finitely generated unital associative alge- 50 bra with a recursively enumerable set of defining relations, over a field F which is finitely generated over prime subfield. Then there exists a finitely presented unital associative F -algebra A in which B is contained as an undistorted unital subalgebra. To prove this theorem we will need to review and modify several Belyaev's preliminary results. Lemma 11 Let X be an alphabet that includes letters a1, . . . , an, b1, . . . , bn. Suppose I is a two-sided ideal of a free unital associative algebra A(X) over a field F with free generating set X. Let ϕ be a homomorphism of unital algebras ϕ : alg {a1, . . . , an} → alg {b1, . . . , bn} such that ϕ(ai) = bi, for i = 1, . . . , n, and ϕ(I ∩ alg {a1, . . . , an}) ⊂ I. Consider X1 = X ∪ {x, y, z, β1, . . . , βn} and naturally embed A(X) in A(X1). Then there is a two-sided ideal I1 in A(X1) containing all elements (i) xzy − 1; (ii) xaizy − bi (i = 1, . . . , n); (iii) aiz − zβi (i = 1, . . . , n); (iv) xzβiβj − bixzβj (i, j = 1, . . . , n), such that I1 ∩ A(X) = I. In addition, if f ∈ A(X) and f0 ∈ f + I is such that degX f0 ≤ degX g, for all g ∈ f + I, then degX1 f0 ≤ degX1 h, for all h ∈ f + I1. Proof. It will be convenient, for our further argument, to extend ϕ to a lin- ear transformation of the vector space A(X). Given a word in β in the free monoid W(β1, . . . , βn), we denote by β(a) the result of replacing each βi by ai (i = 1, . . . , n) in β. We define I1 in A(X1) as the ideal generated by I, all elements in (iii) and (iv) in the statement of Lemma and also all elements xwzβy − ϕ(wβ(a)) where w, respectively, β run through all words in W(X), respectively, W(β1, . . . , βn). Thus, the element of the latter kind include (i) and (ii), for obvious choices of w and β. We prove our lemma arguing by contradiction. Let us choose f ∈ A(X) and f0 ∈ f + I of the least degree d possible in f + I. Assume that there is an element h ∈ A(X1) of strictly lesser degree representing f0+I1. Let us consider the difference v = f0 − h. Then v ∈ I1 \ {0} and so v can be written as a linear combination, with coefficients in F , of the elements of the form v1uv2, w1(xwzβy − ϕ(wβ(a)))w2, w1(aiz − zβi)w2, w1(xzβiβj − bixzbj)w2 (3) where u ∈ I, v1, v2, w1, w2 ∈ W(X1). We may assume that v1 has no suffix, and that v2 has no prefix which is a letter of X. 51 Let us view x, respectively, y as a left, respectively, right parenthesis. In a word w ∈ W(X1) with "properly" arranged parentheses, these are naturally divided into pairs (x, y), a left parenthesis x and its corresponding right parenthesis y. The depth of a fixed pair (x, y) is the difference between the number of occurrences of x and y to the left of x in (x, y). If w has a pair of parentheses with depth s but no pair with depth s+1 , then the number s is called the rank of w. If w has no parentheses, its rank is zero. It is easy to observe that if in the expression for v we group together the monomials with "properly" arranged parentheses, then we again obtain a linear combination of the elements (3). Since in the expression for v = f0 − h there are no monomials of degree ≥ d with improperly arranged brackets, if we write v as a linear combination of elements of the form (3), then all monomials of degree ≥ d with such "parentheses" cancel, while the terms of degree < d can be included in h (remember that we proceed in our proof "by contradiction"!), which changes v. As a result, from the very beginning, we may assume that in our linear combination of the elements (3), we do not have monomials with "improperly" arranged parentheses. Let s be the largest number such that the expression for v involves words of rank s. If s = 0 then an argument similar to the one just given allows us to assume that z is not present in our expression (those monomials of degree ≥ d with z cancel out and those of degree < d can be included in h). Thus v is a linear combination of the elements v1uv2, where v1, v2 ∈ W(X ∪ {β1, . . . , βn}) (this case was omitted in [4]). However, all v1uv2 where v1v2 6∈ W(X) of degree d must cancel because we do not have such monomials in v = f0 − h. Those with degree < d can be moved to h. Now we have v ∈ I and then h = f0 − v = f0 mod I. Since deg h < d, we obtain a contradiction with the choice of f0. Now suppose s > 0. We will show that v has a presentation of the same form, in which all words have rank less than s. The proof of our lemma will then be complete by induction on s. A word w ∈ W(X1) with properly arranged parentheses is called good if its rank is either less than s, or else equal to s but for any pair (x, y) of depth s − 1 the subword starting at x and ending at y, for these x, y, equals xuzβy with u ∈ W(X) and β ∈ W(β1, . . . , βn). Otherwise we call w bad. If we examine the elements of the last three types in (3), we quickly observe that, in each type, both summands are good or bad words at the same time, which makes some of elements in (3) (including the first type) good and some bad. Since f0 ∈ A(X), the linear combination b of all bad summands has degree strictly less than d, so replacing h by h − b removes all bad summands from our expression for v. So in our argument by contradiction we can assume that we do not have bad summands in the expression for v through (3). 52 In each word w ∈ W(X1) of rank s and degree ≥ d in this expression of v for each pair (x, y) of depth s − 1 we replace the subwords xuzβy for these x, y by ϕ(uβ(a)). Since these subwords cancel while reducing to f0 − h, they should cancel also if we replace the respective subwords xuzβy by any symbol ? not in X1. Therefore, the linear combination v will not change if we replace everywhere these subwords by ϕ(uβ(a)). (Again, those w with degree < d will be moved to h.) We again obtain an expression for v and now we are going to show that this, as before, is a linear combination of elements (3). Since the ranks of words in the new expression are less than s, this will complete the proof of the lemma. Consider a summand v1uv2 where u ∈ I. Suppose u = Pi αiwi where αi ∈ F and wi ∈ W(X). Clearly, the ranks of all of the words v1wiv2 are the same. If their common rank is equal to s, then under the replacement described above the words wi ∈ W(X) are affected only when v1 = v′ 2. Now after the replacement, we have 1x, v2 = zβyv′ ′′ ′′ αiv 1 ϕ(wiβ(a))v 2 = v ′′ 1 Xi αiϕ(wiβ(a))! v ′′ 2 . Xi But Pi αiwiβ(a) ∈ I, and by our hypothesis about ϕ, Pi αiϕ(wiβ(a)) ∈ I. Consider a summand w1(xwzβy − ϕ(wβ(a)))w2. If in the word w1ϕ(wβ(a))w2 the subword ϕ(wβ(a)) occurred within a pair of depth s − 1, then clearly in the word w1xwzβyw2 the pair (x, y) would have depth s, which is impossible. Therefore, obviously, after the replacement the expression under consideration either vanishes or keeps the same form. Consider a summand w1(aiz − zβi)w2. It suffices to look at the case where w1 = w′ 2 and this pair (x, y) has depth s − 1. If this is the case, after the replacement, we arrive at 1xw, w2 = βyw′ w ′′ 1 ϕ(waiβ(a))w ′′ 2 − w ′′ 1 ϕ(wai(βiβ)(a))w ′′ 2 . Since (βiβ)(a) = aiβ(a), this expression vanishes. Finally, consider a summand w1(xzβiβj − bixzβj)w2. Again, it suffices to look at the case where w2 = βyw′ 2 and the pair considered (x, y) has depth s − 1. After the replacement we obtain w ′′ 1 ϕ((βiβjβ)(a))w ′′ 2 − w ′′ 1 biϕ((βjβ)(a))w ′′ 2 . This expression also vanishes, since by the choice of ϕ(w) we have 53 ϕ((βiβjβ)(a)) = bibjbi1 · · · bik = biϕ((βjβ)(a)), if β = βi1 · · · βil. Now the proof is complete. ✷ It will be useful to restate Lemma 11 without involving free algebras. Lemma 12 (1) Let A be an algebra with two subalgebras alg {a1, . . . , an} and alg {b1, . . . , bn} such that there is an algebra homomorphism ϕ : alg {a1, . . . , an} → alg {b1, . . . , bn} satisfying ϕ(ai) = bi, i = 1 . . . , n. Then A can be embedded without distortion in an algebra A1, whose generators are those of A and some x, y, z, β1, . . . , βn, and whose defining relations of are those of A and a finite number of relations (i) through (iv) from Lemma 11. (2) If A1 is any algebra with elements a1, . . . , bn, x, y, z, . . . , βn satisfying (i) through (iv) from Lemma 11, and f (a1, . . . , an) = 0 is a relation in A1 then also f (b1, ..., bn) = 0 is a relation of A1. Proof. Claim (1) is just a restatement of Lemma 11. To prove Claim (2), it is sufficient to check that the mapping ϕ : alg {a1, . . . , an} → alg {b1, . . . , bn} given by ϕ(a) = xayz is a homomorphism of algebras. Since ϕ is linear, we only need to check the claim when a = ai1 · · · aik : ϕ(ai1ai2 · · · aik ) = xai1ai2 · · · aikzy = xzβi1βi2 · · · βiky = bi1bi2 · · · bik−1xaik zy = bi1bi2 · · · bik = xai1zyxai2zy · · · xaik zy = ϕ(ai1)ϕ(ai2) · · · ϕ(aik ) : Also, applying relation xzy = 1, we easily derive ϕ(1) = 1. ✷ The next result is a slight modification of Belyaev's lemma devoted to the de- velopment of the technique which allows one to switch from "additive" defining relations of an associative algebra to "multiplicative" ones. Lemma 13 Let f (i, j) be a recursive function defined for i, j = 1, 2, . . . , i 6= j such that f (i, j) = f (j, i). Suppose that Y ⊆ N2 is a recursively enumerable set and Q a unital algebra over a field F with generators x, y, z and defining relations {xyiz+xyjz = xyf (i,j)z i 6= j; i, j = 1, 2, . . .}∪{xyiz = xyjz (i, j) ∈ Y }∪{xz = 1}. Then there exists an algebra Q1 over a field F with the following properties: 54 (1) Q1 has a finite number of generators and a recursively enumerable set of defining relations, one of which has the form α + β = γ and the others are equalities of words in an appropriate alphabet; (2) Q is a subalgebra of Q1 and the degree of any element in Q, with respect to the generating system of Q, cannot decrease when we consider this element as an element of Q1, with respect to a generating system of Q1. Proof. Let us define an increasing sequence of natural numbers n2(1), n3(1), n3(2), n4(1), n4(2), n4(3), n5(1), n5(2), n5(3), n5(4), . . . by setting n2(1) = f (1, 2) and if nj(i) immediately follows after nk(l) then we set nj(i) = max{nk(l) + 1, f (i, j)}. We also define nj(i) for j ≤ i by setting nj(i) = ni+1(j). Now for i < j we set s(i, j) = s(j, i) = nj(i). Thus, for any n ∈ N there exists at most one pair (i, j), 1 ≤ i < j, such that n = s(i, j) = s(j, i). With this choice we would automatically have ni(j) ≥ i − 1, for all j. Let us define the desired algebra Q1 by the set of generators {x, y, z, u, α, β, γ} and the set of defining relations {xyiz = xyjz (i, j) ∈ Y } ∪ {xyiz = xuni(j)εijz i, j = 1, 2, . . .} ∪ {xyf (i,j)z = xus(i,j)γz i 6= j = 1, 2, . . .} ∪ {xz = 1} ∪ {α + β = γ}. Here εij equals α if i + j is even and β if i + j is odd. First we note that the defining relations of Q1 imply those of Q. Indeed, if i < j, then it follows from the relations of Q1 that xyiz + xyjz = xuni(j−1)εi,j−1z + xunj (i)εijz = xus(i,j)(εij + εi,j−1)z = xus(i,j)γz = xyf (i,j)z. Now consider the ideal I1 of the free F -algebra A(x, y, z, u, α, β, γ) generated by the relations of Q1. Any element of this ideal can be written as a linear combination of the elements of the form w1(xyiz − xyjz)w2 ((i, j) ∈ Y ), w1(xyiz − xuni(j)εijz)w2, w1(xuf (i,j)z − xys(i,j)γz)w2 (i 6= j), w1(xz − 1)w2, w1(α + β − γ)w2, (4) where w1, w2 ∈ W(x, y, z, u, α, β, γ). 55 By a reduction of a word w ∈ W(x, y, z, u, α, β, γ) we will mean the simulta- neous replacement in w all subwords of the following forms: xunαz by xyiz, where i is such that n = ni(j) and i + j is even or xunβz by xyiz, where i is such that n = ni(j) and i + j is odd or xunγz by xyf (i,j)z, where i 6= j is such that n = s(i, j) or xz by 1. It should be stressed, that the kind of replacement we apply is fully defined, that is, knowing the word uniquely defines by what word it has to be replaced. For instance, xunαz should be replaced using the reduction in the first line, etc. As in Lemma 11, we proceed by contradiction. Again, suppose we have v = f0− h, with the same conditions on f0 and h. The reductions of the four kinds just described should be applied to the monomials of degree ≥ d. The monomials of the form (4) of degree < d should be attributed to h. It is important here that when we apply reduction, the degree of monomials does not increase. This follows from the choice of numbers s(i, j) = n ≥ max{f (i, j), i − 1}. Having completed all reductions, we arrive at v which is the linear combina- tions of the words of the form w1(α + β − γ)w2. It follows that the leading word of the polynomial f (which does not depend on α, β, γ) cannot cancel, which is a contradiction. Thus the proof is complete. ✷ Now we proceed to the proof of Theorem 13. Proof. Suppose B is an arbitrary unital countable F -algebra, S a finite gen- erating set for B and b1, b2, . . . an arbitrary enumeration of all of its elements (each element occurs at least twice). It is easy to see that after a possible renu- meration we can have degS bi ≤ i, for all i = 1, 2, . . .. By Corollary 4, there exists a unital F -algebra C with three generators a, b, c, with a recursively enumerable set of defining relations, such that B ⊂ C, 1 = ac, bi = abic, for i = 1, 2, . . ., and this embedding has no distortion. Let Y = {(i, j) abic = abjc}. Then Y ⊂ N2 is a recursively enumerable set. Let f (i, j) be a recursive function, defined for all i, j ∈ N with i 6= j, such that f (i, j) = f (j, i) and abic + abjc = abf (i,j)c in C. Consider the F -algebra Q with generators x, y, z and defining relations {xyiz + xyjz = xyf (i,j)z i 6= j; i, j = 1, 2, . . .} ∪ {xyiz = xyjz (i, j) ∈ Y }. 56 By Lemma 13, Q is contained as an undistorted subalgebra in an F -algebra Q1 which has a recursively enumerable set of defining relations Σ(x, y, z, u, α, β, γ) ∪ {α + β = γ}, (5) where Σ contains only word equalities in x, y, z, u, α, β, γ. Next we consider the tensor product Q1 ⊗ C, where Q1 is embedded as the set of elements {q1 ⊗ 1 q1 ∈ Q1} and C as the set of elements {1 ⊗ c c ∈ C}. Both embeddings are easily seen to be undistorted. There exists a homomorphism of algebras ϕ : Q → C. By Claim (1) of Lemma 11, Q1 ⊗ C can be embedded as undistorted subalgebra in an F -algebra Q2 with additional finite set of defining relations. Now let us consider the monoid G with generators x′, y′, z′, u′, α′, β′, γ′ and set of defining relations Σ(x′, y′, z′, u′, α′, β′, γ′). By Theorem 11, there exists a finitely generated monoid G1 containing G as an undistorted submonoid, with a finite set of defining relations Σ1(x′, y′, z′, u′, α′, β′, γ′). (6) Let F [G] denote the monoid algebra G. By Proposition 3, F [G] is a unital undistorted subalgebra of F [G1]. By Claim (1) of Lemma 11, Q2 ⊗ F [G1] can be embedded as an undistorted subalgebra in an F -algebra Q3 with additional finite set of defining relations. At this stage of the proof, we have obtained a chain of undistorted embeddings B → C → C ⊗ Q1 → Q2 → Q2 ⊗ F [G1] → Q3. We know that all additive relations of B, denoted by R(B, ad), follow from the set of all additive relations abic + abjc = abf (i,j)c (the addition table) of C, denoted by R(C, ad, left). Next we identify C with C ⊗ 1 in C ⊗ Q1, and denote by R(C, ad, right) the relations of the set R(C, ad, left) written in terms of generators x, y, z. The relations in R(C, ad, right) follow from from some set Σ ∪ {r} of relations of Q1 identified with 1 ⊗ Q1,where Σ consists only of some equalities of words while {r} is the set of just one singular relation. By Claim (2) of Lemma 11 all relations in R(C, ad, left) follow from relations of Q2 of the form Σ ∪ R2 where R2 is a finite set. Next, all relations Σ(x′, . . .) of F [G] follow from a finite number of relations of F [G1] and hence, again by Claim (2) of Lemma 11, all relations Σ(x, . . .) 57 of Q2 follow from a finite number of relations of Q3. As a result, all relations in R(B, ad) follow from a finite number of relations of Q3. A similar, actually, even simpler, argument works for the multiplication table R(B, mul) of B. Indeed, let H be a monoid given by relations ¯a¯bi¯c¯a¯bj ¯c = ¯a¯bg(i,j)¯c, where g is a recursive function, whose existence follows from the recursive enumerability of the set of relations bibj = bk in B. We embed Q3 in Q3 ⊗ F [H]. By Theorem 11 and Proposition 3 F [H] is embedded without distortion in a finitely presented monoid algebra F [H1], and Q3 is undistorted in Q3 ⊗ F [H1]. By Claims (1) and (2) of Lemma 11 this tensor product can be embedded without distortion in an algebra Q4 in such a way that all relations in R(B, mul) follow from a finite number of relations of Q4. Finally, let us notice that all relations of B follow from its addition and multi- plication tables and finitely many relations of the form αx = x′, where α is a generator of F , x a generator of B (and x′ ∈ B). Thus, all relations of B follow from a finite number of relations of an algebra Q4, where B is contained as subalgebra. By Claim (viii) of Proposition 2, B is an undistorted subalgebra in an algebra A given by these relations of algebra Q4. ✷ This theorem allows us to prove our final result characterizing tame filtrations in associative algebras. Theorem 14 Let R be a unital associative algebra over a field F , which is finitely generated over prime subfield. A constructive filtration ρ on R is a tame filtration if and only if β ∼ α ∩ R where α is a degree filtration on a finitely presented unital algebra A in which R is embedded as a unital subalgebra. Proof. As in several cases before, we can restrict ourselves to the proof of "only if" portion of the statement. Using Theorem 9, we can embed R without dis- tortion in a finitely generated algebra B presented by a recursively enumerable set of relations in such a way that ρ ∼ β ∩ R where β is a degree filtration on B. By definition then distρ β∩R ≤ t · id, for some t ∈ N. Then we use Theorem 13 and embed B without distortion in a finitely presented algebra A with a degree filtration α. This allows us to write distβ α∩B ≤ t · id. Now by Claim α∩B ◦ distρ α∩R ≤ distβ (3) in Proposition 1, we can write distρ β∩R ≤ t2 · id or that α ∩ R (cid:22) ρ. Using Claim (iv) in Proposition 2, we obtain ρ ∼ α ∩ R. The proof is now complete. ✷ 4.3 More Examples: Distortion of "cyclic" subalgebras We start this subsection by exhibiting the variety of tame degrees on asso- ciative and Lie algebras of particularly simple form. Using the results on the distortion in groups [16], one can show that for any real number θ such that 58 0 < θ < 1 there is a tame degree on F [x], which is equivalent to nθ (where n is the ordinary degree of a polynomial). As it turns out, these degrees come from the tame degrees on the free monoid W(x). In the remainder of the subsection, we show that actually, there are tame filtrations (hence tame degrees) on F [x] that are not induced by tame filtrations on W(X). Let B = F [x] be the polynomial algebra in one variable. Consider any function ϕ on {0, 1, . . .} which is positive on {1, 2, . . .}. Let us define on B a "degree- like" function dϕ(f ) = ϕ(deg(f )). Setting Bϕ n = {f ∈ B dϕ(f ) ≤ n} defines a filtration on B provided that ϕ is subadditive, that is, ϕ(a + b) ≤ ϕ(a) + ϕ(b). If the ratio ϕ(n) log n cannot be separated from 0 when n grows indefinitely then for any k > 0 there is n = n(k) such that Bn contains all polynomials f with deg f > exp(kn); then {Bn} is not a tame filtration. But if log n = O(ϕ(n)), for a subadditive function ϕ, as above, then {Bn} is a tame filtration. By Theorem 7, B can be embedded in a 2-generator algebra Aϕ with a de- gree filtration {Aϕ n }. In this case, the degree of a polynomial f ∈ F [x] with respect to the generators of A will be a function equivalent to ϕ(deg(f )). For example, we may choose ϕ = (log n)µnν, where 0 < ν < 1. n} so that {B ∩ Aϕ n} ∼ {Bϕ If ϕ is computable then by Theorem 14 such degree-like functions can be achieved in finitely presented algebras. For example, we can get the degree function for F [x] in the form (log n)µnν, where 0 < ν < 1 in a finitely presented algebra provided that µ and ν are constructive real numbers (say, π − e, etc.). A similar example can be obtained in the case of Lie algebras. Let M be an abelian Lie algebra with basis {a0, a1, a2, . . .}. We form a semidirect product L = F x ⊕ M of this algebra with a one dimensional algebra F x so that [x, ai] = ai+1 (similar algebras have been introduced in subsubsection 2.4.1). However, in this case, using Theorem 9, we can only guarantee an embedding in a finitely generated Lie algebra, which does not need to be finitely presented. It is obvious that any subadditive "degree-like" function dϕ defines filtration not only on F [x] but also on monoid W(x) and, conversely, is defined by its restriction to this monoid. But one should not think that even in a "simple" case like F [x], all tame filtrations are equivalent to "degree-like" filtrations, that is, filtrations extending the filtrations of W(x). Proposition 5 There exist a family of pairwise inequivalent tame filtrations on F [x], labeled by nonzero elements of F , none being an extension of a tame filtration on W(x), naturally embedded in F [x]. 59 Proof. We start with an infinite increasing sequence of natural numbers d1 = dn n → ∞. There is n0 such that max( dm 1, d2, d3, . . . such that , 1) < dn+1 dn m for all n ≥ n0 and all m < n. Let us consider a polynomial ring F [y1, . . . , ys, . . .] in the set {y1, . . . , ys, . . .} of weighted variables such that the weight of each ys is s. This assignment of the weight to the variables leads to a well-defined assignment of the weight to each polynomial G(y1, . . . , ys) ∈ F [y1, . . . , ys, . . .]. Let us fix λ ∈ F and consider n (x) = xdn + λxdn−1, for all n = 1, 2, . . .. Since f1 = x + λ, any polynomials f λ polynomial f = f (x) ∈ F [x] can be written as f (x) = G(f λ s (x)), where G(y1, . . . , ys) ∈ F [y1, . . . , ys, . . .]. Let D(f ) be the minimum of the weights of all polynomials G(y1, . . . , ys) such that f (x) = G(f λ s (x)). Clearly, setting Bλ n} on B = F [x]. Let us denote by pn the number of monomials of weight n in F [y1, . . . , ys, . . .]. The set of such monomials is the disjoint union S1 ∪ . . . ∪ Sn where the monomials in S1 contain y1, the monomials in S2 do not contain y1 but contain y2, etc. Thus, pn ≤ pn−1 + pn−2 + · · · + p0 and using induction by n, we easily obtain pn ≤ 2n. It follows that βλ is a tame filtration of B = F [x]. 1 (x), . . . , f λ n = {f D(f ) ≤ n} defines a filtration βλ = {Bλ 1 (x), . . . , f λ Now let us estimate from below the βλ-degree (we will be simply saying λ- degree and write degλ) of f = f µ n (= xdn + µxdn−1), where n > n0 and µ 6= λ. Suppose that f = G(f λ s (x)) where the weight of G(y1, . . . , ys) is minimal possible, as above. We need to consider three cases. 1 (x), . . . , f λ Case 1: s < n. When we replace in G(y1, . . . , ys) each ym by f λ tain a polynomial whose degree does not exceed max(d1 s) times the m, we ob- ds , . . . , 1 weight of G(y1, . . . , ys). Since dm m ≤ dn−1 n − 1 , we conclude that the weight of G(y1, . . . , ys) cannot be less than cn = . It follows that the λ- dn−1/(n − 1) dn dn dn−1 → ∞, we have cn n → ∞, for degree of f µ n cannot be less than cn. Since such n. ngt + yt−1 Case 2: s = n. In this case, G(y1, . . . , yn) = yt n gt−1 + . . ., where t > 0, each gj is a polynomial in y1, . . . , yn−1 and gt is nonzero. Then, as in the previous case, we replace all y1, . . . , yn by f λ n . If gt is not a constant or if t > 1 then the leading term (as a polynomial in x) resulting from the first summand, has degree greater than dn. This term can get canceled with a term arising in another gm(f λ n−1) only when the degree of gm(f λ n−1) is at least dn and then, as in the previous case, the weight of gm and G cannot be less than cn. Now if t = 1 and g1 is a constant c 6= 0, then the first summand produces cf λ n on the n and in order to obtain f µ n and obtain f µ 1 , . . . , f λ 1 , . . . , f λ 1 , . . . , f λ 60 left hand side, the weight of g0 (hence of G(y1, . . . , yn)) must be greater or equal to = cn − o(1). dn − 1 dn−1/(n − 1) Case 3: s > n. Then G = yt s gt−1 + . . ., where t > 0, each gj is a polynomial in y1, . . . , ys−1 and gt is nonzero. In this case, the same argument cn . as in Case 2 shows that the weight of G(y1, . . . , ys) is at least cs and n sgt + yt−1 cs s > As a result, in any case, degλ f µ degλ f µ n → ∞), while degµ f µ n grows faster than n (in the sense that n ≤ n. It follows that βλ 6∼ βµ if λ 6= µ, as n claimed. In a particular case where µ = 0, we also have the following. If we restrict any βλ with λ 6= 0 to W(x), we obtain a filtration β whose extension to F [x], viewed as monoid algebra, produces a filtration β′ = F [β], for which degβ ′ f λ i . Since the restriction of β′ to W(x) coincides with β, it follows that βλ is not equivalent to any filtration, which is extended from W(x). In other words, βλ is not a "degree-like" filtration, for any λ 6= 0. ✷ i grows faster than degλ f λ i ≥ degβ ′ xdi = degβ f 0 i = degλ f 0 Our last remark is as follows. Remark 4 By Theorem 7, if we fix λ ∈ F then every filtration βλ on B = F [x] is a restriction of a degree filtration αλ of certain finitely generated al- gebra Aλ where B is contained as a subalgebra. However, none of Aλ can be chosen commutative. Indeed, by Theorem 3 about the embeddings in the commutative case, the embedding B ⊂ Aλ is undistorted. If β is the standard degree filtration on B = F [x] then by Claim (i) of Proposition 2, αλ ∩ B ∼ β. It follows that βλ ∼ β, which was proven impossible because β is the extension of the standard degree filtration of W(x) while degλ f λ n = dn n ≤ n and deg f λ where dn n → ∞ as n → ∞. Acknowledgment The authors would like to thank the following mathematicians with whom they discussed various matters related to this paper: L. A. Bokut', V. Drensky, R. I. Grigorchuk, M. V. Sapir, U. U. Umirbaev. We also thank Qiuhui Mo who pointed out an error in the Lie algebra portion of the original proof of Theorem 10. 61 References [1] Bahturin, Yuri, Identical relations in Lie algebras, VNU Press, 1987. [2] Bahturin, Yuri; Olshanskii, Alexander, Actions of Maximal Growth, Proc. London Math. Soc., doi:10.1112/plms/pdp047. [3] Bahturin, Yu.A.; Olshanskii, A.Yu., Identities. In: Algebra II , Encyclopedia Math. Sci. , 18 , Springer (1991), 107 - 221 [4] Belyaev, V. Ya., Subrings of finitely presented associative rings, Algebra i Logika, 17(1978), 627 - 638. Engl. transl.: Algebra and Logic, 17(1978), 407 - 414. [5] Bokut', L.A., Embedding in simple associative algebras, Algebra i Logika, 15(1976), 117-142 (Russian). Engl. Transl.: Algebra and Logic 15(1976), 73- 90. [6] Bokut', L.A.; Chen, Yuqun, Groebner-Shirshov Bases for Lie Algebras: after A.I. Shirshov, Southeast Asian Bull. Math. 31 (2007), 1057 - 1076 [7] Bokut', L.A.; Chen, Yuqun; Qiuhui, Mo, Groebner-Shirshov bases and embeddings of algebras, arXiv:0908.1992 [8] Bokut', L.A.; Kukin, G.P., Algorithmic and combinatorial algebra, Kluwer AP, 1994, xvi+384p. [9] Gromov, M., Asymptotic invariants of infinite groups, London Math. Soc. Lect. Notes Ser. , 182(1993), Geometric group theory, V.2, P. 1 - 295. [10] Higman, G., Subgroups of finitely presented groups, Proc. Royal Soc. London Ser. A262, 455 - 475, 1961. [11] Kharlampovich, O.; Sapir, M., Algorithmic problems in varieties, a survey, International Journal of Algebra and Computation, 12(1995), 379 - 602. [12] Kreuzer, M.; Robbiano, L., Computational Commutative Algebra, vol. 1. Springer-Verlag, Berlin, 2000. [13] Malcev, A.I., On a representation of nonassociative rings, Uspekhi. Mat. Nauk,(1955), 181 - 185. [14] Mikhailova, K.A., The occurrence problem for direct products of groups, Mat. Sbornik 70(112)(1966), 241 - 251. [15] Murskii, V.L. , Isomorphic imbeddability of a semigroup with an enumerable set of defining relations into a finitely presented semigroup (Russian), Mat. Zametki 1(1967), 217 - 224. [16] Ol'shanskii, A.Yu., Distortion functions for subgroups, Geometric group theory down under (Canberra, 1996), 281 - 291, DeGruyter, Berlin, 1999. 62 [17] Olshanskii, A.Yu., On subgroup distortion in finitely presented groups, Mat. Sbornik 188(1997), 51 - 98. Engl. Transl.: Math. USSR. Sbornik, 188(1997), 1617 - 1664. [18] Platonov, A.N., An isoperimetric function of the Baumslag - Gersten group, Vestnik Moskov. Univ. Ser. 1, Mat. Mech 2004, No. 3, 12-17 (Russian); Engl. transl.: Moscow Univ. Math. Bull., 59(2004), 12-17 (2005). [19] Umirbaev, U.U., Algorithmic problems in associative algebras, Algebra and Logic, 32(1993), 244 - 255. [20] Umirbaev, U.U., The occurrence problem in free associative and free Jordan algebras. In.: Algorithmic problems in Algebra and Model Theory, Almaty, 1995, 11 - 15. [21] Umirbaev, U.U., Universal derivations and subalgebras of free algebras. In Proc. 3rd Internat. Conf. Algebra, Krasnoyarsk, Russia, Walter de Gruyter, Berlin, 1996, 255 - 271. 63
1606.01296	1	1606	2016-06-03T22:43:04	Discriminants and Automorphism Groups of Veronese subrings of skew polynomial rings	[ "math.RA" ]	We study important invariants and properties of the Veronese subalgebras of $q$-skew polynomial rings, including their discriminant, center and automorphism group, as well as cancellation property and the Tits alternative.	math.RA	math	DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS OF SKEW POLYNOMIAL RINGS K. CHAN, A.A. YOUNG, AND J.J. ZHANG Abstract. We study important invariants and properties of the Veronese sub- algebras of q-skew polynomial rings, including their discriminant, center and automorphism group, as well as cancellation property and the Tits alternative. Introduction The determination of the full automorphism group of an algebra is a fundamen- tal problem in mathematics. This is generally extremely difficult, for example, even for the polynomial ring in three variables its automorphism group is not well un- derstood. Aside from a remarkable result of Shestakov-Umirbaev [SU] which shows that the Nagata automorphism is a wild automorphism, the general structure of this automorphism group eludes our grasp. Since the 1990s, researchers have successfully computed the full automorphism group of several interesting families of noncommutative algebras of finite Gelfand- Kirillov dimension, including certain quantum groups, generalized quantum Weyl algebras, skew polynomial rings -- see [AlC, AlD, AnD, BJ, GTK, GY, LL, SAV]. A few years ago, by using a rigidity theorem for quantum tori, Yakimov proved the Andruskiewitsch-Dumas conjecture [Y1] and the Launois-Lenagan conjecture [Y2], each of which determines the full automorphism group of an important class of quantized algebras. Recently, Ceken-Palmieri-Wang and the third-named author introduced a discriminant method to control the automorphism group of certain classes of algebras [CPWZ1, CPWZ2] and then were able to compute the auto- morphism group of several more families of Artin-Schelter regular algebras that satisfy a polynomial identity. The wisdom behind much of this progress is that noncommutative algebras are more rigid, so that more techniques are available for detecting their symmetries. In most of the results mentioned above, the algebras are deformations (in some weak sense) of the commutative polynomial rings. In this paper we apply the discriminant method to certain noncommutative algebras that are not deformations of polynomial rings. We are mainly interested in the automorphism problem, but will briefly touch upon the cancellation problem and the Tits alternative. Throughout the introduction let k denote our base field and k× be the its group of units. For a k-algebra A, let Aut(A) denote the group of k-algebra automorphisms of A. 2000 Mathematics Subject Classification. Primary 16W20. Key words and phrases. skew polynomial ring, Veronese subring, discriminant, automorphism group, cancellation problem, Tits alternative. 1 2 K. CHAN, A.A. YOUNG, AND J.J. ZHANG Fix a q ∈ k×. Let kq[x1, · · · , xn] denote the skew polynomial ring generated by x1, · · · , xn subject to the relations (E0.0.1) xjxi = qxixj, ∀ 1 ≤ i < j ≤ n. In this paper we assume that q is a nontrivial root of unity. First we consider the case when q = −1(6= 1). By [CPWZ1, Theorem 4.10(1)], if n is even, then (E0.0.2) Aut(k−1[x1, · · · , xn]) = Sn ⋉ (k×)n which is virtually abelian [Definition 0.6(1)]. On the other hand, if n is odd, then [CPWZ3, Theorem 2] says that Aut(k−1[x1, · · · , xn]) contains a free group on two generators. In the case of q = −1, these two results present a dichotomy depending on the parity of n. This is a version of the Tits alternative [Definition 0.6(3)]. For any Z-graded algebra A = Li∈Z Ai and for any positive integer v, the vth Veronese subring of A is defined to be A(v) :=Mi∈Z Avi. The following theorem generalizes the result [CPWZ1, Theorem 4.10(1)]. Theorem 0.1. Suppose that char k 6= 2. Let A be k−1[x1, · · · , xn](v) where v is a positive integer. If n and v have different parity, then Aut(A) ∼= Sn ⋉ (k×)n. Considering elements in (k×)n as (a1, · · · , an), the Sn-action on (k×)n in Theo- rem 0.1 does not follow the standard rule (E0.1.1) σ : (a1, · · · , an) 7−→ (aσ−1(1), · · · , aσ−1(n)) for all σ ∈ Sn, due to asymmetry of the automorphisms corresponding to (k×)n, see Lemma 6.2 for some details. On the other hand, the Sn-action appearing in (E0.0.2) does follow the standard rule (E0.1.1). If n and v have the same parity, we are unable to determine the automorphism group of A, but we conjecture that it contains a free subgroup of rank 2. Also in this case we are unable to decide whether or not A is cancellative [Definition 0.3], -- see Theorem 0.4 and Question 0.5 below for related results and questions. We can generalize the above theorem to the case when q is arbitrary of finite order. Let m be the order of q and assume that m is bigger than 2 (or equivalently, q 6= ±1). We have two different hypotheses dependent on the parity of n in the following theorem. Theorem 0.2. Let A be kq[x1, · · · , xn](v) where v is a positive integer and m > 2. Suppose that one of the following is true. (a) n is even and m does not divide v. (b) n is odd and gcd(m, v) 6= 1. Then the following statements hold. (1) If qv is either 1 or −1, then Aut(A) ∼= Z/(n) ⋉ (k×)n. (2) If qv 6= ±1, then Aut(A) ∼= (k×)n. It is very difficult to describe the group Aut(A) if n ≥ 3 and (n, m, v) does not satisfy Theorem 0.2(a,b). Hypotheses (a) and (b) have other significant conse- quences. Furthermore, for any tensor product of algebras in above two theorems, the automorphism group is also computable, see Remark 7.8(1). DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 3 The proofs of the first two theorems are based on calculations of the discriminant of the algebra A over its center. Further, the discriminant method can also be used to answer the cancellation problem which is closely related to the automorphism problem. We recall a definition. Definition 0.3. An algebra A is called cancellative if A[t] ∼= B[t] for any algebra B implies that A ∼= B. One famous open problem in affine algebraic geometry is the Zariski Cancellation Problem which asks if the polynomial ring k[x1, · · · , xn], for n ≥ 3, is cancellative. It is well-known that k[x] and k[x1, x2] are cancellative for any field k. In 2013, Gupta [Gu1, Gu2] settled the Zariski Cancellation Problem negatively in positive characteristic for n ≥ 3. The Zariski Cancellation Problem in characteristic zero remains open for n ≥ 3, see [BZ, Gu3] for more details and relevant references. Our methods of using the discriminant can be applied to show that certain Veronese subalgebras of the skew polynomial rings are cancellative. Theorem 0.4. Let A be kq[x1, · · · , xn](v) where v is a positive integer and let m be the order of q. Suppose that one of the following is true. (a) n is even and m does not divide v. (b) n is odd and gcd(m, v) 6= 1. Then A is cancellative. This says that all the algebras appearing in the first two theorems are cancella- tive, see Remark 7.8(2) for a more general result. As mentioned above, we can not decide whether or not kq[x1, · · · , xn](v) is cancellative if it does not fit into Theorem 0.4. We formally ask Question 0.5. Let A be kq[x1, · · · , xn](v) where v is a positive integer and let 2 ≤ m < ∞ be the order of q. Suppose that one of the following is true. (a) n is even and m divides v. (b) n is odd and gcd(m, v) = 1. Is then A cancellative? The Zariski Cancellation Problem is connected to several other open problems in affine algebraic geometry -- see [BZ, Gu3]. In the noncommutative setting, it is also related to certain properties of the Nakayama automorphism [LMZ] and the Makar-Limanov invariant [BZ]. The last result in this paper concerns the Tits alternative for automorphism groups of the Veronese subalgebras of skew polynomial rings. In 1972, Tits proved a remarkable and surprising dichotomy [Ti]: for any subgroup G of the general linear group GL(C⊕n), either G is virtually solvable, or G contains a free group of rank 2. Since then, similar dichotomy results have generally been referred as the Tits alternative. The original Tits alternative and its variations have many applications in dynamical systems, geometric group theory, Diophantine geometry, topology and so on. There is a version of the Tits alternative for the class of the automorphism groups of skew polynomial rings following [CPWZ3, Theorem 2]. In general it would be very interesting to prove that some classes of algebraic objects must satisfy certain non-obvious dichotomy such as the Tits alternative. To state our result we recall some definitions. Definition 0.6. Let G be a group. 4 K. CHAN, A.A. YOUNG, AND J.J. ZHANG (1) G is called virtually abelian if there is a normal abelian subgroup N ⊆ G such that G/N is finite. (2) G is called virtually solvable if there is a normal solvable subgroup N ⊆ G such that G/N is finite. (3) Let C be a class of groups. We say C satisfies the Tits Alternative if the following dichotomy holds: any G ∈ C is either virtually solvable or it contains a free subgroup of rank 2. For any fixed n ≥ 2, let Cn consist of groups Aut(A) where A = kq[x1, · · · , xn](v) for all q ∈ k× being a root of unity and all v ∈ N. Theorem 0.7. Retain the above notation. (1) If n is odd, the Tits alternative holds for Cn. (2) The Tits alternative holds for C2. This theorem leaves the following question. Question 0.8. Does the Tits alternative hold for Cn for even integer n ≥ 4? In principle, the discriminant method introduced in [CPWZ1, CPWZ2] can be applied to any algebras, though in applications (and examples) given there most algebras are Artin-Schelter regular. In this paper we consider a class of algebras that are not Artin-Schelter regular and show that the discriminant method is still very effective in solving several classical problems. The paper is organized as follows. We provide background material and recall the definition of the discriminant in the noncommutative setting in Section 1. In Section 2, we study some basic properties of the discriminant. In Section 3, we provide some information about the center and Veronese subrings of the q-skew polynomial rings. Detailed discriminant computations are given in Section 4 (when n is odd) and Section 5 (when n is even). Main theorems (Theorems 0.1 and 0.2) are proved in Section 6. In Section 7 we deal with the cancellation problem and prove Theorem 0.4. The Tits alternative is discussed in Section 8 where Theorem 0.7 is proved. Acknowledgments. A.A. Young was partly supported by the US National Science Foundation (NSF Postdoctoral Research Fellowship, No. DMS-1203744) and J.J. Zhang by the US National Science Foundation (No. DMS-1402863). 1. Definitions Throughout the paper let k be a commutative domain, and sometimes we further assume that k is a field. Modules, vector spaces, algebras, tensor products, and morphisms are over k. All algebras are associative with unit. We will recall some definitions given in [CPWZ1, CPWZ2] and introduce some new definitions. In particular, we will introduce a new variant of the discriminant in this section. Let B = Mw(R) be the w × w-matrix algebra over a commutative domain R. We have the internal trace trint : B → R, (bij )w×w 7→ bii w Xi=1 DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 5 which is the usual matrix trace. Now let B be a general R-algebra and F be a localization of R such that that BF := B ⊗R F is finitely generated and free over F . Then the left multiplication defines a natural embedding of R-algebras (E1.0.1) lm : B → BF → EndF (BF ) ∼= Mw(F ), where w is the rank rkF (BF ). We define the regular trace map by composing (E1.0.2) trreg : B lm−−→ Mw(F ) trint−−−→ F ⊆ Q(R) where Q(R) is the field of fractions of R. Note that trreg is independent of the choices of F . In this paper, a trace (function) means the regular trace unless otherwise stated. In computation, we also need to assume that the image of trreg is in R. Let R× denote the set of invertible elements in R. If f, g ∈ R and f = cg for some c ∈ R×, then we write f =R× g. Let A be a domain. We say a normal element x ∈ A divides y ∈ A if y = xz for some z ∈ A. If D is a set of elements in A, a normal element x ∈ A is called a common divisor of D if x divides d for all d ∈ D. We say a normal element x ∈ A is the greatest common divisor or gcd of D, denoted by gcdA D, if (1) x is a common divisor of D, and (2) any common divisor y of D divides x. It follows from part (2) that the gcd of any subset D ⊆ A (if it exists) is unique up to a scalar in A×. In practice, we often choose a domain A such that R ⊆ A ⊆ B. Note that given D ⊆ R, the elements gcdR D, gcdA D, gcdB D may not all exist. Even when they exist, they may not be equal. Definition 1.1. Let R be a commutative domain and B be an R-algebra. Suppose that the image of tr := trreg in (E1.0.2) is in R. Let (r, p) be a pair of positive integers. Let A be a fixed domain between R and B in part (3). (1) [CPWZ2, Definition 1.2(1)] Let Z = {zi}r i=1 be two r- element subsets of B. The discriminant of the pair (Z, Z ′) is defined to be i=1 and Z ′ = {z′ i}r dr(Z, Z ′) = det(tr(ziz′ j)r×r) ∈ R. (2) The p-power discriminant ideal of rank r, denoted by D[p] r (B/R), is the ideal of R generated by the set of elements of the form (E1.1.1) dr(Z1, Z2)dr(Z3, Z4) · · · dr(Z2p−1, Z2p) for all possible r-element subsets Z1, Z2, · · · , Z2p ⊆ B. (3) The p-power discriminant of rank r, denoted by d[p] r (B/R), is defined to be the gcd in A of the elements of the form (E1.1.1). Equivalently, the p-power discriminant d[p] r (B/R) of rank r is the gcd in A of the elements in D[p] r (B/R). The notation d[p] r (B/R) suppresses the dependence of the p-power discriminant of rank r on the choice of an intermediate domain A. In applications, this choice of A will be clearly specified. Allowing different choices of A, as well as different p and r, increases the probability for the existence of the gcd of elements in (E1.1.1). When p = 1, the above definition agrees with [CPWZ2, Definition 1.2]. If d[p] r (B/R) 6 K. CHAN, A.A. YOUNG, AND J.J. ZHANG exists, then the ideal of A generated by d[p] A which is generated by a normal element and contains D[p] r (B/R). r (B/R) is the smallest principal ideal of The following lemma is [CPWZ1, Proposition 1.4(3)]. Lemma 1.2. Suppose B is finitely generated and free over R of rank r. Then d[p] r (B/R) =A× (dr(B/R))p and D[p] r (B/R) is the principal ideal of R generated by (dr(B/R))p. Proof. Let X = {x1, · · · , xr} be a basis of B over R. By [CPWZ1, Definition 1.3(3)], dr(B/R) =R× dr(X, X). For any r-element subset Z := {z1, · · · , zr} ⊂ B, we can write zi =Pj rij xj for an r × r-matrix (rij ). Similarly for Z ′. Then dr(Z, Z ′) = dr(X, X) det(rij ) det(r′ ij ) = dr(B/R) det(rij ) det(r′ ij ). Then the assertion follows from the definition. (cid:3) Lemma 1.3. Let Ψ be a subset of B that generates B as an R-module. (1) D[p] r (B/R) is the ideal of R generated by the set (E1.3.1) {dr(X1, X2) · · · dr(X2p−1, X2p) Xi ⊆ Ψ, ∀ i}. (2) d[p] r (B/R) is the gcd in A of elements in set (E1.3.1). Proof. Every element z ∈ B is an R-linear combination of φi ∈ Ψ. By bilinearity of tr(zz′) and multi-linearity of det, every dr(Z, Z ′) is an R-linear combination of dr(X, X ′) where X, X ′ are r-element subsets of Ψ. Therefore every element of the form (E1.1.1) is an R-linear combination of elements in (E1.3.1). The assertions follow. (cid:3) In this paper we will see that some discriminants satisfy the following. Definition 1.4. Retain the notation as in Definition 1.1. The p-power r-rank discriminant d[p] r (B/R) is called stable if d[ip] r (B/R) =A× (d[p] r (B/R))i for all positive integers i. Under the hypotheses of Lemma 1.2, d[p] r (B/R) is always stable for every p. 2. Properties of the discriminant In this section we list of elementary properties of d[p] r (B/R). The following lemma is similar to [CPWZ2, Lemma 1.4]. Lemma 2.1. Suppose that the image of the regular trace tr is in R. Let g be an automorphism of B such that g and g−1 preserve R. (1) The p-power r-rank discriminant ideal D[p] (2) The p-power r-rank discriminant d[p] r (B/R) is g-invariant. r (B/R) (if exists) is g-invariant up to a unit in A. DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 7 (3) Suppose r1 ≤ r2 and p1 ≤ p2 are positive integers. Then D[p2] r2 (B/R) ⊆ D[p1] r1 (B/R). If both d[p2] r2 (B/R) and d[p1] r1 (B/R) exist, then r1 (B/R) d[p2] d[p1] r2 (B/R). As a consequence, the quotient d[p2] a unit in A. r2 (B/R)/d[p1] r1 (B/R) is g-invariant up to Proof. (1) By [CPWZ1, Lemma 1.8(3)], g(dr(Z, Z ′)) = dr(g(Z), g(Z ′)). Then g maps an element of the form (E1.1.1) to another element of the same form. Simi- larly, this holds for g−1. Hence, g (and g−1) preserves D[p] r (B/R). (2) This follows from part (1) and the fact that the gcd is well-defined up to a unit. (3) When p1 = p2 = 1, this is [CPWZ2, Lemma 1.4(5)]. For general p1 ≤ p2, (cid:3) the proof is similar to the proof of [CPWZ2, Lemma 1.4(5)], so it is omitted. We recall some definitions from [CPWZ2, p.766]. Let C be a domain such that k ⊆ C and that C/k is k-flat. We say that A ⊗ C is A-closed if, for every 0 6= f ∈ A and x, y ∈ A ⊗ C, the equation xy = f implies that x, y ∈ A up to units of A ⊗ C. For example, if C is connected graded and A ⊗ C is a domain, then A ⊗ C is A-closed. The next lemma is similar to [CPWZ2, Lemma 1.12]. Lemma 2.2. Retain the hypotheses as above. Assume that B ⊗ C is a domain. r (B ⊗ C/R ⊗ C) = D[p] (1) D[p] (2) Suppose A ⊗ C is A-closed. If d[p] r (B/R) ⊗ C. r (B/R) exists, then d[p] r (B ⊗ C/R ⊗ C) exists and equals d[p] r (B/R). Proof. (1) First of all, the regular trace tr of B ⊗ C over R ⊗ C is equal to the regular trace tr of B over R when restricted to elements in B. Let Ψ be a subset of B such that B is generated by Ψ as an R-module. Then r (B ⊗C/R⊗C) r (B/R) ⊗ C by B ⊗C is generated by Ψ as an R⊗C-module. By Lemma 1.3(1), D[p] is the ideal of R ⊗ C generated by the set (E1.3.1), which is just D[p] Lemma 1.3(1). (2) Suppose d := d[p] by definition. By part (1), d[p] set D[p] r (B/R) exists. Then it is the gcd in A of the set D[p] r (B/R) r (B ⊗ C/R ⊗ C) (if exists) is the gcd in A ⊗ C of the r (B/R) ⊗ C, which is the gcd in A ⊗ C of the set D[p] r (B/R). Let d′ ∈ A ⊗ C be a common divisor in A ⊗ C of the set D[p] r (B/R). By A- closedness of A ⊗ C, we may assume that d′ is in A (up to a unit). This implies that d′ divides d. It is clear that d is a common divisor in A⊗C of the set D[p] r (B/R). Therefore d is the gcd in A ⊗ C of the set D[p] r (B/R). The assertion follows. (cid:3) Let F be a localization of a commutative domain R (and R is the center Z(B) in most of applications) and F may not be the fraction field of R. We assume that BF := B ⊗R F is finitely generated and free over F . We recall a definition from [CPWZ2]. Definition 2.3. [CPWZ2, Definition 1.10] Retain the above notations. 8 K. CHAN, A.A. YOUNG, AND J.J. ZHANG (1) A subset b = {b1, · · · , bw} ⊆ B is called a semi-basis of B if it is an F -basis of BF , where bi is viewed as bi ⊗R 1 ∈ BF . In this case w is the rank of B over R. (2) Let b be a semi-basis of B and T be a subset of B containing b which generates B as an R-module. We call such a set T an R-generating set of B. Then b is called a quasi-basis (with respect to T ) of B if every t ∈ T can be written as t = cb for some b ∈ b. We denote c by (t : b). We continue to introduce some notation. Again let w be the rank of B over R. Let Z := {z1, · · · , zw} be a subset of B. If b is a semi-basis, then for each i, w zi = aijbj Xj=1 for some aij ∈ F . In this case, the w × w-matrix (aij)w×w is denoted by (Z : b). Let T be as in Definition 2.3(2). Let T /b denote the subset of F consisting of nonzero scalars of the form det(Z : b) for all Z ⊆ T with Z = w. Let (E2.3.1) and D(T /b) = {dw(b, b)f f ′ f, f ′ ∈ T /b}, Dw(T ) = {dw(Z, Z ′) Z, Z ′ ⊆ T, Z = Z ′ = w}. (E2.3.2) Note that if Z and Z ′ are w-element subsets of T , then dw(Z, Z ′) ∈ D(T /b) by [CPWZ2, (1.10.1)]. In fact, we have D(T /b) = Dw(T ). If b = {b1. · · · , bw} is a quasi-basis with respect to an R-generating set T . Then for each i, let (E2.3.3) Ci = {(t : bi) t ∈ T }\{0}. It is easy to see that every element in T /b is of the form c1c2 · · · cw, where ci ∈ Ci for each i. Let (E2.3.4) Dc(T /b) =(dw(b, b) ci, c′ i ∈ Ci) . (cic′ w Yi=1 i)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) If b is a quasi-basis with respect to T , then D(T /b) = Dc(T /b). Let S be a subset of an algebra R. Let Sp denote the subset of R consisting of s1s2 · · · sp for all si ∈ S. The following lemma is similar to [CPWZ2, Lemma 1.11]. Lemma 2.4. Let T be a set of generators of B as an R-module and w = rank(B/R). Let p be a positive integer. w (B/R) is generated by the set Dw(T )p. w (B/R) =A× gcd Dw(T )p. (1) D[p] (2) d[p] (3) If b is a semi-basis of B, then d[p] (4) If b is a quasi-basis of B, then d[p] w (B/R) = gcd D(T /b)p. w (B/R) = gcd(Dc(T /b))p. Proof. (1) This follows from Lemma 1.3(1). (2), (3) and (4) follow from the definition, the above discussion and part (1). (cid:3) In the rest of the section we assume that B1 and B2 are two algebras that are k-flat. If X 1 ⊆ B1 and X 2 ⊆ B2 are two subsets, then X 1 ⊗ X 2 denotes the set {x ⊗ y x ∈ X 1, y ∈ X 2}. We say that the pair (X 1, X 2) is hereditary if for x ∈ X 1 and y ∈ X 2, every divisor of x ⊗ y is of the form x′ ⊗ y′ (up to a unit in B1 ⊗ B2). The following lemma is easy. DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 9 Lemma 2.5. Let X 1 ⊆ B1 and X 2 ⊆ B2 be two subsets such that (a) gcd X 1 and gcd X 2 exist. (b) (X 1, X 2) is hereditary. Then gcd(X 1 ⊗ X2) =(B1⊗B2)× gcd X 1 ⊗ gcd X 2. (cid:3) Lemma 2.6. Let B1 and B2 be two k-algebras containing central subalgebra do- mains R1 and R2 respectively. Let wi = rankRi (Bi) for i = 1, 2. Assume that (a) R := R1 ⊗ R2 is a domain. (b) bi is a quasi-basis of Bi over Ri with corresponding Ri-module generating set T i (and bi ⊆ T i). Then the following hold. (1) b := b1 ⊗ b2 is a quasi-basis of B := B1 ⊗ B2 over R with corresponding R-generating set being T := T 1 ⊗ T 2. (2) Let w := w1w2. Then Dc(T /b) = Dc(T 1/b1)w2 ⊗ Dc(T 2/b2)w1 . (3) Suppose that (Dc(T 1/b1)w2 , Dc(T 2/b2)w1 ) is hereditary. Let p be an inte- w2 (B2/R2) are stable discriminants, then so is w1 (B1/R1) and d[p] ger. If d[p] d[p] w (B/R). Further, d[p] w (B/R) =B× d[p] w1 (B1/R1)w2 ⊗ d[p] w2 (B2/R2)w1. Proof. (1) Let i be either 1 or 2. Let bi = {bi j}j∈J i . By definition, it is routine to check that b1 ⊗ b2 is a quasi-basis of B1 ⊗ B2 over R with corresponding R-generating set being T := T 1 ⊗ T 2. wi} and T i = {ti 2, · · · , bi 1, bi (2) Let 1 ≤ j ≤ w1 and 1 ≤ k ≤ w2. Following (E2.3.3), define C1 of elements c in B1 such that cb1 of elements of the form cj ⊗ dk where cj ∈ C1 elements of the form t(b1 k)−1 for all t ∈ T . j ⊗ b2 j ∈ T 1. Similarly, we define C2 j and dk ∈ C2 By linear algebra, dw(b, b) = dw1 (b1, b1)w2 ⊗ dw2(b2, b2)w1 . And we have the following computation, for all (cj ⊗ dk), (c′ k) ∈ Cj,k for different (j, k), j be the set k. Let Cj,k consist k . Then Cj,k consist of j ⊗ d′ k)] ∈ X w2 ⊗ Y w1 [(cj ⊗ dk)(c′ j ⊗ d′ Yj,k j)) cj, c′ assertion follows from (E2.3.4). where X = {(Qj(cjc′ j ∈ C1 j } and Y = {(Qk(dkd′ (3) This follows from the definition, Lemma 2.4(4), part (2) and Lemma 2.5. (cid:3) k)) dk, d′ k ∈ C2 k }. Now 3. Center and Veronese subrings of q-polynomial rings From now on we fix two integers m, n ≥ 2 and a primitive mth root of unity, say q, in k. The q-skew polynomial ring is generated by x1, · · · , xn and subject to the relations (E3.0.1) xj xi = qxixj, ∀ 1 ≤ i < j ≤ n. and is denoted by kq[x1, · · · , xn], or simply by kq[x]. We will adopt the following notation for monomials xs := xs1 n where s = (s1, ..., sn) ∈ Nn is its degree vector. We will also denote by ei the standard basis vector, with 1 in its ith component and 0 elsewhere. For any 0 ≤ k ≤ m, define 1 · · · xsn (E3.0.2) yk := q−⌊n/2⌋k(k+1)/2 x(k,m−k,k,m−k,...), 10 K. CHAN, A.A. YOUNG, AND J.J. ZHANG in particular, y0 = ym = (−1)⌊n/2⌋(m+1)xm 4 · · · xm 1 xm xm 2 xm 2⌊n/2⌋, 3 · · · xm 2⌈n/2⌉−1. Note that both y0 and ym are in the central subalgebra generated by {xm One can easily check that the yis satisfy the following relations (E3.0.3) yiyj = q−⌊n/2⌋(i+j)(i+j+1)/2 x(i+j,2m−i−j,i+j,2m−i−j,...), As a consequence, ∀ 0 ≤ i, j ≤ m. 1 , ..., xm n }. (E3.0.4) yiyj = ykyℓ, ∀ i + j = k + ℓ. Equations (E3.0.3)-(E3.0.4) also imply that yiyj =(cid:26) y0yi+j i + j ≤ m, ymyi+j−m i + j > m. The following is a consequence of [CYZ, Lemma 4.1]. Let Z(A) denote the center (E3.0.5) of an algebra A. Lemma 3.1. 1 , ..., xm xm n . (1) If n is even, then Z(kq[x]) is a polynomial ring generated by (2) If n is odd, then Z(kq[x]) is generated by xm 1 , ..., xm n , y1, ..., ym−1. Proof. (1) This is [CPWZ2, Example 2.4(2)]. (2) One can check it directly or use [CYZ, Lemma 4.1]. We use some of the notation in [CYZ, Section 4]. Let Y be the skew symmetric n × n-matrix with 1/m in all entries above the diagonal. Let t ∈ Nn. By [CYZ, Lemma 4.1] the monomial xt is in the center Z(kq[x]) if and only if Y t ∈ Zn. Let S = mY . Then xt ∈ Z(kq[x]) if and only if St ∈ mZn. Let ¯S be the endomorphism of (Z/mZ)n represented by the matrix S. Then St ∈ mZn if and only if t is a lift of an element in ker( ¯S). Since n is odd, rank( ¯S ⊗ Fp) = n − 1 for all primes p. It is easy to check that ker( ¯S) is generated by (i, −i, i, . . . , −i, i) ∈ (Z/mZ)m for i = 0, . . . , m. Lifting these to Zn gives (i, m− i, i, . . . , m− i, i) for i = 1, . . . , m− 1 and mei for i = 1, . . . , n. (cid:3) When n is even, the center Z(kq[x]) is easy to understand, namely Z(kq[x]) = k[xm 1 , ..., xm n ]. If n is odd, every element of Z(kq[x]) can be expressed as a linear combination of terms of the form xma or xmayb, with a ∈ Nn and 0 < b < m. Each such term can be rewritten as follows, xmayb = xa1m 1 · · · xanm n yb = q−⌊n/2⌋b(b+1)/2xa1m+b 1 x(a2+1)m−b 2 xa3m+b 3 · · · xanm+b n . Since the above polynomials form a k-linear basis of Z(kq[x]), we have Z(kq[x]) ∼= ∼=     k[xm y0 − xm n , y0, ..., ym] 4 · · · xm n−1, ym − (−1)(m+1)(n−1)/2xm 1 xm 1 , ..., xm 2 xm 3 · · · xm n ,   yiyj − ykyℓ, ∀ i + j = k + ℓ. k[xm ∀i + j < m, 1 , ..., xm yiyj − xm n , y1, ..., ym−1] 4 · · · xm 2 xm n−1yi+j, ∀i + j = m, ∀i + j > m, yiyj − (−1)(m+1)(n−1)/2xm 1 xm yiyj − (−1)(m+1)(n−1)/2xm 1 xm 3 · · · xm 2 · · · xm n , n yi+j−m. .   DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 11 For example, if n = 3, Z(kq[x]) ∼= and if m = 2, k[xm 1 , xm 2 , 2 , xm 3 , y0, ..., ym] ym − (−1)m+1xm (cid:18) y0 − xm yiyj − ykyℓ, 1 xm 3 , ∀i + j = k + ℓ. (cid:19) , Z(kq[x]) ∼= 2 · · · x2 Hopefully this gives some idea on what the center should be. 1 − (−1)(n−1)/2x2 (cid:0) y2 . n. (cid:1) k[x2 1, ..., x2 n, y1] 1x2 For any v ∈ N, the vth Veronese subalgebra of kq[x], denoted by kq[x](v), is the subalgebra generated by elements of total degree v. As before we fix positive integers m, n, v. Let (E3.1.1) g := gcd(v, m). Let kq[x±1] be the localization of kq[x] by inverting all xis. We extend the notation xs for all s ∈ Zn in a natural way. Let F be the center of kq[x±1](v) which is a localization of Z := Z(kq[x](v)), and let kq[x](v) F = kq[x](v) ⊗Z F . Since F is a Zn-graded field, we have (1) kq[x](v) (2) kq[x](v) Since each xmv denote i F = kq[x±1](v) which is a Zn-graded skew field. F is free over F . ∈ F , we have that kq[x](v) F is finite dimensional over F , and we (E3.1.2) w := dimF kq[x](v) F . Let Hv = {s ∈ Zn Pn (respectively, kq[x±1](v)) is the span of xH+ i=1 si ∈ vZ}, and let H + v = Hv ∩ Nn, so that kq[x](v) v (respectively, xHv ). Lemma 3.2. Retain the above notation. Suppose that n is odd. (1) Z(kq[x](v)) = Z(kq[x]) ∩ kq[x](v) = khxm i , yji ∩ kq[x](v). (2) The center Z(kq[x±1](v)) is spanned by xM where M = mZn + gZ As a consequence, (−1)i−1ei! ∩ Hv. n Xi=1 Z(kq[x](v)) = khxm i , yjgi ∩ kq[x](v). Proof. Let xs ∈ Z(kq[x±1](v)) for some s ∈ Zn. Since xixvm−1 i+1 ∈ kq[x](v), xsxixmv−1 i+1 = xixmv−1 i+1 xs = q−(si+si+1)xsxixmv−1 i+1 . Hence, si + si+1 ∈ mZ for all i. Then, for each i, (E3.2.1) si =(aim + b aim + (m − b) i is odd, i is even. for some a1, ..., an ∈ Z and 0 ≤ b ≤ m − 1. This part of the proof works for both even and odd n. 12 K. CHAN, A.A. YOUNG, AND J.J. ZHANG (1) When xs ∈ Z(kq[x](v)) for s ∈ Nn. We obtain that, if b > 0, then ai ≥ 0 for all ai in (E3.2.1) and if b = 0, ai ≥ 0 for odd i and ai ≥ −1 for even i. This is equivalent to 1 xma1 1 for some ai ≥ 0. The assertion follows. xs =k× (xma1 (2) Recall that n is odd. Note that, if xma1 1 n · · · xman · · · xman n b = 0, b 6= 0, yb b + m n − 1 2 + n Xi=1 · · · xman n yb ∈ Z(kq[x±1](v)), then ai! ∈ vZ, and hence, b ∈ gZ. This means that if xs ∈ Z, then s ∈ M . Conversely, it is straightforward to check that if s ∈ M , then xs ∈ Z. (cid:3) We are interested the discriminant of kq[x](v) over its center. We examine sepa- rately the case when n is odd, and the case when n is even. We conclude this section with the hereditary property (as mentioned before Lemma 2.5) for monomials in kq[x](v). Lemma 3.3. Let A1, · · · , As be algebras of type kq[x](v). For each i, let X i ⊆ Ai be a set of monomials. Then, for any f i ∈ X i, every divisor of f 1 ⊗ f 2 ⊗ · · · ⊗ f s is of the form g1 ⊗ g2 ⊗ · · · ⊗ gs where each gi is a divisor of f i. Proof. Consider Ai = kqi [x](vi) as an Nni-graded algebra for all i. Let n =Ps i=1 ni. Then A1 ⊗ · · · ⊗ As is an Nn-graded algebra. Since each f i is Nni -homogeneous, F := f 1 ⊗ · · · ⊗ f s is Nn-homogeneous. Note that Nn is an ordered semigroup. Then any divisor G of F is Nn-homogeneous. Equivalently, G = g1 ⊗ · · · ⊗ gs where each gi is a divisor of f i. (cid:3) 4. Discriminant computation: when n is odd We will freely use the notation introduced in the last section, and further assume that n is odd in a large part of this section. Recall from Lemma 3.2 that, if n is odd, then (E4.0.1) n M = mZn + gZ( Xi=1 (−1)i−1ei)! ∩ Hv. Then M is a subgroup of Hv. We can partition Hv into cosets mod M . It is easy to see the total number of these cosets is equal to w (E3.1.2). Lemma 4.1. Assume n is odd. (1) For each coset of M in Hv, there is a unique representative p := (p1, ..., pn) such that (a) 0 ≤ p1 < g, (b) for each 1 < i < n, we have 0 ≤ pi < m, and (c) 0 ≤ pn < vm/g. Moreover, the above remains true with indices (1, n) replaced by any (µ, ν) with µ 6= ν. (2) w = mn−1. (3) w 6= 0 in k. DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 13 Proof. (1) Pick an arbitrary coset M ′ of M , and let p = (p1, ..., pn) ∈ M ′. Since g = gcd(m, v), there exists c ∈ Z such that cm ≡ g mod v. Hence (g, −g, g, −g, ..., −g, g− cm) ∈ M , and we can translate p by some multiple of this vector to obtain 0 ≤ p1 < g. Furthermore, if t ∈ M then t1 ∈ gZ, so there is no vector in M ′ whose first component is any other 0 ≤ r′ < g. For each 1 < i < n, we have m(ei − en) ∈ M , so we can apply the translation trick above and assume that 0 ≤ pi < m. Furthermore, if t ∈ M and t1 = 0, then each other ti ∈ mZ. This implies that there is no other set of possible values of p1, ..., pn−1 subject to the conditions 0 ≤ p1 < g and 0 ≤ pi < m for every 1 < i < n. Finally, (vm/g)en ∈ M , so there exists a representative p ∈ M ′ subject to constraints (a)-(c) of the lemma. If cen ∈ M , then c ∈ mZ ∩ vZ = (vm/g)Z, so this representative is unique. The last statement is clear since the above calculations do not depend on the ordering of the indices 1, ...n. This finishes the proof of part (1). (2) The value w can be determined by counting the cosets by their represen- tatives. For every sequence of integers p1, ..., pn−1 such that 0 ≤ p1 < g and 0 ≤ pi < m for all 1 < i < n, there are m/g possible values of pn such that 0 ≤ pn < vm/g and (p1, ..., pn) ∈ Hv. Therefore, w = g · mn−2 · m/g = mn−1. (3) Since q ∈ k and o(q) = m, the characteristic of k cannot divide m. Or m 6= 0 (cid:3) and w 6= 0 in k. In this paper we mainly consider the case when B = kq[x](v) for both even and odd n. Using [CPWZ1, Definition 1.10] and notation in Section 2, if B := {b1, ..., bw} ⊆ H + v is a set of representatives of each coset of M , then b := xB is a quasi-basis of kq[x](v) with respect to T := xH+ v , and (E4.1.1) Dc(T /b) =k× dw(b, b)( w Yi=1 xsi−bi xs′ The following lemmas hold for both even and odd n. si, s′ i ∈ Nn ∩ (M + bi)) . i−bi(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Lemma 4.2. Retain the above notation. Let s ∈ H + v such that xs is not central. Then tr(xs) = 0. As a consequence, the trace map tr sends kq[x](v) to Z(kq[x](v)). Proof. Since A := kq[x](v) is Zn-graded, so is the center Z := Z(A). Let F be the graded field of fractions of Z. Then A is a free module over F with F -basis B. Then, for all i, j, there is a unique k such that bibj = ck ij ∈ F . ij = 0. Every (cid:3) If bi is not in the center, then j 6= k. Therefore tr(bi) = Pj=k ck element xs is of the form cbi for some i and c ∈ F . The assertion follows. ij bk for some 0 6= ck Lemma 4.3. Retain the above notation. Suppose that w is invertible. Then Dc(T /b) =k× ( w Yi=1 xsi! w Yi=1 xs′ si, s′ i ∈ Nn ∩ (M + bi)) . Proof. For each bi ∈ B, let b∗ i ∈ M . For any s ∈ H + v , if s /∈ M , then xs is not central, and tr(xs) = 0 by Lemma 4.2. If s ∈ M , then xs is central, and tr(xs) = wxs =k× xs, where the last equation follows from the hypothesis that w is invertible. Therefore, in the matrix (tr(xbi xbj ))w×w, the only i ∈ B be such that bi + b∗ i!(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 14 K. CHAN, A.A. YOUNG, AND J.J. ZHANG nonzero terms appear where bj = b∗ i , and w dw(b, b) =k× det(tr(xbi xbj ))w×w =k× xbi xb∗ Yi=1 i =k× w Yi=1 xbi!2 . (cid:3) The assertion follows by the above formula and equation (E4.1.1). Recall that m is the order of q, the rank of kq[x](v) over its center is w = mn−1 and g = gcd(v, m). Theorem 4.4. Let B = kq[x](v) when n is odd. Suppose that m is invertible in k. Let R be the center of B. Assume that v divides wp(g − 1). Then d[p] w (B/R) =k× (x1x2 · · · xn)wp(g−1) =k× (xv 1xv 2 · · · xv n) wp(g−1) v . As a consequence, d[p] w (B/R) is stable. Proof. By Lemmas 2.4(4) and 4.3, we have d[p] w (B/R) = gcd Λ2p where si ∈ Nn ∩ (M + bi)) . Λ :=( w Yi=1 xsi(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) For each 1 ≤ s ≤ n, let fs ∈ N be maximal such that xfs s divides all elements of Λ. This gives x2pf as the gcd of Λ2p in the over-algebra kq[x] ⊇ B where v , then it is the gcd of Λ2p in kq[x](v) as well, but, f = (f1, · · · , fn). If 2pf ∈ H + otherwise, this is not true. We first calculate f1 by summing the lowest powers of x1 in each coset of M (or more precisely, in each Nn ∩ (M + bi) for different i). These lowest powers can be found by using the representatives outlined in Lemma 4.1(1), which also shows that this power cannot exceed g − 1. For each 0 ≤ k ≤ g − 1, there are mn−1/g cosets with lowest power xk 1 . Therefore, the sum is f1 = mn−1 g(g − 1) g 2 = w(g − 1) 2 . For fi with i 6= 1, we can use the last assertion of Lemma 4.1(1) to relabel indices, so the above calculation remains valid for i 6= 1 and we conclude that f1 = f2 = · · · = fn. Now 2pf = wp(g − 1)(1, 1, · · · , 1) ∈ H + follows from the last paragraph, and stability of d[p] assertion. v as v divides wp(g − 1). The assertion w (B/R) follows from the main (cid:3) 5. Discriminant computation: when n is even In this section we assume that n is even. The following is similar to Lemma 3.2. Lemma 5.1. Suppose that n is even. (1) Z(kq[x](v)) = khxm i , yjm/gi ∩ kq[x](v). (2) The center Z(kq[x±1](v)) is spanned by xM where M := mZn +(cid:18) m g (cid:19) Z n Xi=1 (−1)i−1ei! ∩ Hv. DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 15 Proof. (1) We copy the first part of the proof of Lemma 3.2. Let xs ∈ Z(kq[x±1](v)) for some s ∈ Zn. Since xixvm−1 i+1 ∈ kq[x](v), we have xsxixmv−1 i+1 = xixmv−1 i+1 xs = q−(si+si+1)xsxixmv−1 i+1 . Hence, si + si+1 ∈ mZ for all i. Then, for each i, (E5.1.1) si =(aim + b aim + (m − b) i is odd, i is even. for some a1, ..., an ∈ Z and 0 ≤ b ≤ m − 1. Considering xs ∈ Z(kq[x](v)) for s ∈ Nn. We obtain that, if b > 0, then ai ≥ 0 for all ai in (E5.1.1) and if b = 0, ai ≥ 0 for odd i and ai ≥ −1 for even i. This is equivalent to xs =k× (xma1 1 xma1 1 n · · · xman · · · xman n b = 0, b 6= 0, yb for some ai ≥ 0. Next we need to determine the values of b such that yb ∈ Z(kq[x](v)). Note that xv 1yb = qvbybxv 1. Hence vb ∈ mZ, or equivalently, b is a multiple of m/g. The assertion follows. (2) By the proof of part (1), every monomial xs ∈ Z(kq[x±1](v)) is in xM . Conversely, it is straightforward to check that every element in xM is also in Z(kq[x±1](v)) (cid:3) Much of the work of last section can be reapplied. When n is even we define M as in Lemma 5.1(2): (E5.1.2) M = mZn +(cid:18) m n g (cid:19) Z( Xi=1 (−1)i−1ei)! ∩ Hv. Lemma 5.2. Suppose that n is even. (1) For each coset of M in Hv, there is a unique representative p := (p1, · · · , pn) such that (a) 0 ≤ p1 < m/g, (b) for each 1 < i < n, 0 ≤ pi < m, and (c) 0 ≤ pn < vm/g. (2) w = mn/g2. (3) w 6= 0 in k. Proof. The following proof is similar to the proof of Lemma 4.1. (1) Pick an arbitrary coset M ′ of M , and let p = (p1, ..., pn) ∈ M ′. Since (m/g, −m/g, m/g, −m/g, ..., m/g, −m/g) ∈ M, we can replace p1 by r where 0 ≤ r ≤ g and r ≡ p1 mod m/g within the coset M ′. Therefore we can assume, without loss of generality, that 0 ≤ p1 < m/g. Furthermore, if t ∈ M then t1 ∈ (m/g)Z, so there is no vector in M ′ whose first component is any other 0 ≤ r′ < m/g. For each 1 < i < n, m(ei −en) ∈ M , so we can assume, without loss of generality, that 0 ≤ pi < m. Furthermore, if t ∈ M and t1 = 0, then each other ti ∈ mZ. This implies that there is no other set of possible values of p1, ..., pn−1 subject to the conditions 0 ≤ p1 < m/g and 0 ≤ pi < m for every 1 < i < n. 16 K. CHAN, A.A. YOUNG, AND J.J. ZHANG Finally, (vm/g)en ∈ M , so there exists a representative p ∈ M ′ subject to constraints (a)-(c) of the lemma. If cen ∈ M , then c ∈ mZ ∩ vZ = (vm/g)Z, so this representative is unique. This finishes the proof of part (1). (2) The value w can be determined by counting the cosets by their represen- tatives. For every sequence of integers p1, ..., pn−1 such that 0 ≤ p1 < m/g and 0 ≤ pi < m for all 1 < i < n, there are m/g possible values of pn such that 0 ≤ pn < vm/g and (p1, ..., pn) ∈ Hv. Therefore, w = (m/g) · mn−2 · (m/g) = mn/g2. (3) Since q ∈ k and o(q) = m, then the characteristic of k cannot divide m. (cid:3) Consequently m 6= 0 and w 6= 0 in k. Theorem 5.3. Let B = kq[x](v) when n is even and let R be the center of B. Suppose that m is invertible in k and that v divides wp( m g − 1). Then v ( m 2 · · · xv n) wp g −1). w (B/R) =k× (x1x2 · · · xn)wp( m d[p] g −1) =k× (xv 1xv As a consequence, d[p] w (B/R) is stable. Proof. This proof is similar to the proof of Theorem 4.4. w (B/R) = gcd Λ2p where By Lemmas 2.4(4) and 4.3, d[p] si ∈ Nn ∩ (M + bi)) . Λ :=( w Yi=1 xsi(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) For each 1 ≤ s ≤ n, let fs ∈ N be maximal such that xfs s divides all elements of Λ. This gives x2pf as the gcd of Λ2p in the over-algebra kq[x] ⊇ B where f = (f1, · · · , fn). If 2pf ∈ H + v , then it is the gcd of Λ2p in kq[x](v) as well. By symmetry (see the proof of Theorem 4.4 for a similar argument), f1 = f2 = · · · = fn, and we will only work out f1. We calculate f1 by summing the lowest powers of x1 in each coset of M (or more precisely, in each Nn ∩ (M + bi) for different i). These lowest powers can be found by using the representatives outlined in Lemma 5.2, which also shows that this power cannot exceed m/g − 1. For each 0 ≤ k ≤ m/g − 1, there are mn−1/g cosets with lowest power xk 1. Therefore, the sum is mn−1 (m/g)(m/g − 1) f1 = g 2 = w 2 (cid:18) m g − 1(cid:19) . Now 2pf = wp(m/g − 1)(1, 1, · · · , 1) ∈ H + follows from the last paragraph, and stability of d[p] main assertion. v as v divides wp(m/g − 1). The assertion w (B/R) follows clearly from the (cid:3) 6. Application I: automorphism group For any algebra A, let Aut(A) denote the group of all algebra automorphisms of A. When A is N-graded, let Autgr(A) denote the group of all graded algebra automorphisms of A. In this section we only consider the algebra A := kq[x](v) and use g for an algebra automorphism of A. First we consider an algebra automorphism g satisfying (E6.0.1) g((xv 1 · · · xv n)a) =k× (xv 1 · · · xv n)a, for some positive integer a. The first few lemmas discuss some easy properties of g satisfying (E6.0.1). There is a natural Nn-grading on the skew polynomial ring kq[x] with deg xi = ei for i = 1, · · · , n. We consider kq[x](v) as an Nn-graded subalgebra of kq[x]. Both DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 17 kq[x] and kq[x](v) are also N-graded by considering the total degree. We will use both gradings in this section. For any permutation π of {1, ..., n}, we denote the linear function π : Zn → Zn determined by π : ei 7→ eπ(i). For a permutation π ∈ Sn, we have (E6.0.2) and denote (E6.0.3) sπ−1 (1) xπ(s) = x 1 sπ−1 (n) · · · x n π := xs1 xs π(1) · · · xsn π(n). It is clear that xπ(s) =k× xs π. Lemma 6.1. Let g ∈ Aut(kq[x](v)) satisfying (E6.0.1) in parts (1)-(4). (1) The image of every monomial through g is a k×-multiple of a monomial. (2) deg g(f ) = deg f for any monomial f . As a consequence, g is a graded algebra automorphism. (3) The image of each xv (4) There exists a permutation πg of {1, ..., n} such that each monomial xs is i is a k×-multiple of some xv j . mapped to a k×-multiple of xπg (s). (5) g satisfies (E6.0.1) if and only if, for each i, there is a j such that g(xv i ) =k× xv j . Proof. (1) By (E6.0.1), g((x1x2 · · · xn)vaN ) =k× (x1x2 · · · xn)vaN for all N > 0. Let f be any monomial in kq[x](v). Then f is a factor of (x1x2 · · · xn)vaN for some N > 0. Let f ′ be a monomial such that f f ′ =k× (x1x2 · · · xn)vaN . Then g(f )g(f ′) = g((x1x2 · · · xn)vaN ) =k× (x1x2 · · · xn)vaN . Since Nn is an ordered semigroup and kq[x](v) is an Nn-graded domain, both g(f ) and g(f ′) are Nn-homogeneous. Every Nn-homogeneous element is a k×-multiple of a monomial. The assertion follows. (2) Note that the lowest total degree of a non-scalar element in kq[x](v) is v. Ap- plying g to the monomials f of (total) degree v, we have that deg g(f ) ≥ v = deg f . Since every monomial in kq[x](v) is a product of monomials of degree v, deg g(f ) ≥ deg f for all monomials. By symmetry, deg g−1(f ) ≥ deg f . The assertion follows. (3) If f is a degree v monomial of kq[x](v), f 2 can be decomposed as f 2 =k× f1f2 where f1, f2 are degree v monomials. The decomposition is unique if and only if f = xv i for some i. This property is invariant under g. (4) We choose π so that, for each i, we have g(xv π(i) by part (3). For i ) =k× xv any monomial xs, we have (E6.1.1) g(xs)v = g(xvs) =k× xvs1 π(1) · · · xvsn π(n) =k× (xπg (s))v, which implies that g(xs) =k× xπg(s). (5) One implication is part (3) and the other implication is clear. (cid:3) Next we wish to understand the coefficients of the image of g. The next lemma deals with the case when πg is the identity. For any automorphism g of kq[x](v), we say πg = 1 if g(xv i ) =k× xv i 18 K. CHAN, A.A. YOUNG, AND J.J. ZHANG for all i = 1, · · · , n. Let Aut1(kq[x](v)) be the subgroup of Aut(kq[x](v)) consisting of automorphisms g with πg = 1. It is clear that Aut1(kq[x](v)) ⊆ Autgr(kq[x](v)). Lemma 6.2. Retain the above notation. (1) Let g ∈ Aut1(kq[x](v)). Then there exist (c, k2, ..., kn) ∈ (k×)n such that for each monomial xs of degree N v, (E6.2.1) g(xs) = cN ks2 2 · · · ksn n xs. (2) Conversely, given (c, k2, ..., kn) ∈ (k×)n, then (E6.2.1) defines a unique algebra automorphism g ∈ Aut1(kq[x](v)). As a consequence, Aut1(kq[x](v)) ∼= (k×)n. Proof. (1) Let c be such that g(xv 1) = cxv be such that g(xv−1 1 xi) = ckixv−1 means that it is in kq[x](v)), there exists a scalar r ∈ k× such that 1, and let k1 = 1. For each i 6= 1, let ki 1 xi. For any monomial xs of degree N v (which xN v(v−1) 1 xs = r(xv−1 1 x1)s1 · · · (xv−1 1 xn)sn Therefore cN (v−1)xN v(v−1) 1 which implies that xs) 1 g(xs) = g(xN v(v−1) = r(ck1xv−1 = cN vks1 1 x1)s1 · · · (cknxv−1 1 xn)sn 1 · · · ksn n xN v(v−1) 1 xs, g(xs) = cN ks2 (2) This is easy and the proof is omitted. 2 · · · ksn n xs. (cid:3) = 1, and For any g1, g2 ∈ Aut(kq[x](v)) such that πg1 = πg2 , we have πg−1 1 ◦g2 there exist c, k1 = 1, k2, ..., kn ∈ k× such that for any monomial xs of degree N v, (E6.2.2) g2(xs) = cN ks1 1 · · · ksn n g1(xs). The automorphism group can therefore be fully determined by determining the possible values of πg and producing an example automorphism for each. We discuss possible πg in the next lemma. Lemma 6.3. Let g denote an automorphism of kq[x](v) satisfying (E6.0.1). (1) If q = ±1, for every permutation π of {1, ..., n}, there exists g such that πg = π, and for each xs, g(xs) = xs πg . (2) If qv 6= ±1, then, for any g, we have πg = 1. (3) If qv = ±1, then, for each m ∈ Z, there exists g, such that πg is addition by m modulo n, and g(xs) = xs πg . (4) If qv = ±1 and q 6= ±1, then, for any g, there exists m ∈ Z such that πg is addition by m modulo n. Proof. (1) The relations of kq[x] are simply xixj = qxj xi for all i 6= j. Therefore any permutation π of the generators x1, ..., xn extends to an automorphism g of kq[x], and g restricts to an automorphism of kq[x](v). (2) For any distinct i, j, we have (E6.3.1) (xv−1 i xj )xv i = ri,jxv i (xv−1 i xj ) DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 19 where ri,j =(cid:26) q−v qv j < i, i < j. We apply g to both sides of (E6.3.1). Then Lemma 6.1(5) shows that ri,j = rπ(i),π(j), where π = πg. Since qv 6= q−v, we have that i < j implies π(i) < π(j). Therefore π is the identity. (3) It suffices to prove the assertion in the case m = 1. Let s = (s1, · · · , sn) and π(i) ≡ i + 1 mod n. Then π = xs1 xs For all s and t, 2 · · · xsn−1 n 1 = qPn−1 xsn i=1 snsi xsn 1 xs1 2 · · · xsn−1 n = qPn−1 i=1 snsi xπ(s). Let α(s) = s2 n and define xsxt = qPi<j sj ti xs+t. g : xs 7→ qα(s)xs π for all monomials xs in kq[x]. Note that g cannot extend to an automorphism of kq[x]. But we show next that g extends to an automorphism of kq[x](v). To show this, it suffices to show that g(xs)g(xt) = g(xsxt)(= qPi<j sj ti g(xs+t)) for all xs and xt in kq[x](v). Using the above computation, we have g(xsxt) = qPi<j sj tig(xs+t) = qα(s+t)qPi<j sj ti qPn−1 i=1 (sn+tn)(si+ti)xπ(s+t), and g(xs)g(xt) = qα(s)+α(t)qPn−1 = qα(s)+α(t)qPn−1 i=1 snsi qPn−1 i=1 snsi qPn−1 i=1 tnti xπ(s)xπ(t) i=1 tntiqPi<j π(s)jπ(t)i xπ(s+t). By direct calculation, the difference between the q-powers in the expressions of i=1 ti and qv = ±1, we i=1 ti) = (±1)2 = 1. Therefore g(xsxt) = g(xs)g(xt) so g is an algebra g(xsxt) and g(xs)g(xt) is 2sn(Pn i=1 ti). Since v divides Pn have q2sn(Pn automorphism. (4) For distinct i, j, let yi,j = xv−1 i xj. For any distinct i, j, k, we have where r =(cid:26) q−1 q yi,jyi,k = ryi,kyi,j, if i < j < k or j < k < i or k < i < j, if i < k < j or k < j < i or j < i < k. Recall q 6= q−1. For any i, j, the number of values of k that yield r = q is equal to j − i − 1 mod n. Since this is true for all i 6= j, we have πg(j) − πg(i) − 1 ≡ j − i − 1 mod n. Therefore πg(j) − πg(i) ≡ j − i mod n, and the assertion follows by letting m = πg(n). (cid:3) We are now ready to prove Theorems 0.1 and 0.2. Proof of Theorem 0.1. For each σ ∈ Sn, let Fσ be the algebra automorphism of k−1[x] induced by sending xi to xσ(i) for all i. This automorphism restricts to an algebra automorphism of k−1[x](v), which is still denoted by Fσ -- see Lemma 6.3(1). Then the subgroup generated by all {Fσ σ ∈ Sn} is isomorphic to Sn. 20 K. CHAN, A.A. YOUNG, AND J.J. ZHANG Now assume that n and v have different parity and that g is an algebra auto- morphism of k−1[x](v). Recall that m = 2. If n is odd, gcd(m, v) = 2 and we can apply Theorem 4.4. If n is even, gcd(m, v) = 1, so we can apply Theorem 5.3. In both cases, by Theorem 4.4 or 5.3, the v-power discriminant d[v] w (k−1[x](v)/R) is of the form (xv n)N for some N > 0. By Lemma 2.1(1), this discriminant is g-invariant. This means that g satisfies (E6.0.1). Let πg be the permutation defined in Lemma 6.1(4). It is easy to see that the map φ : g → Fπg is a surjective group homomorphism from Aut(k−1[x](v)) to Sn with kernel being Aut1(k−1[x](v)). By Lemma 6.2, we have Aut1(k−1[x](v)) ∼= (k×)n. Therefore 1 · · · xv Aut(k−1[x](v)) ∼= Sn ⋉ Aut1(k−1[x](v)) ∼= Sn ⋉ (k×)n. (cid:3) The proof of Theorem 0.2 is similar. Proof of Theorem 0.2. The proofs of (1) and (2) are similar, we only provide the proof of (2) here. (2) Under hypotheses (a) or (b), we use Theorem 4.4 or 5.3 to conclude that the v-power discriminant d[v] n)N for some N > 0. By Lemma 2.1(1), this discriminant is g-invariant. This means that g satisfies (E6.0.1). By Lemma 6.3(2), we have πg = 1, or equivalently, g ∈ Aut1(kq[x](v)). Therefore w (kq[x](v)/R) is of the form (xv 1 · · · xv Aut(kq[x](v)) = Aut1(kq[x](v)) ∼= (k×)n. (cid:3) 7. Application II: cancellation problem The second application of the discriminant method is the cancellation problem. We need to recall some definitions and results from [BZ]. Definition 7.1. [BZ, Definition 1.1] Let A be an algebra. (1) We call A cancellative if A[t] ∼= B[t] for some algebra B implies that A ∼= B. (2) We call A strongly cancellative if, for any d ≥ 1, A[t1, · · · , td] ∼= B[t1, · · · , td] for some algebra B implies that A ∼= B. (3) We call A universally cancellative if, for any k-flat finitely generated commu- tative domain R such that R/I = k for some ideal I ⊂ R and any k-algebra B, any algebra isomorphism A ⊗ R ∼= B ⊗ R implies that A ∼= B. The first result is Lemma 7.2. [BZ, Proposition 1.3] Let k be a field and A be an algebra with center C(A) = k. Then A is universally cancellative, hence, strongly cancellative. We only need the following definition for connected graded domains. generated by A1 = Lr Definition 7.3. [CPWZ1, Definition 2.1(2)] Let A be a connected graded domain kxi. An element f ∈ A is called dominating if, for every testing N-filtered PI algebra T with grF T being a connected graded domain, and for every testing subset {y1, · · · , yr} ⊆ T that is linearly independent in the quotient k-module T /F0T , there is a presentation of f of the form f (x1, · · · , xr) in the free algebra khx1, · · · , xri such that the following hold: either f (y1, · · · , yr) = 0, or i=1 (a) deg f (y1, · · · , yr) ≥ deg f , and DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 21 (b) deg f (y1, · · · , yr) > deg f , further, deg yi0 > 1 for some i0. Lemma 7.4. [BZ, Theorem 4.6] Let A be a connected graded PI domain generated in degree 1, of finite Gelfand-Kirillov dimension. Suppose that the discriminant power (d[p] w (A/C))a is dominating for some p, w and a. Then A is strongly can- cellative. Proof. The original [BZ, Theorem 4.6] was proved for discriminant dw(A/C). But the proof works for this more general setting when [BZ, Lemma 4.5(2)] is replaced by Lemma 2.2. So we are not going to repeat the rest of the proof. (cid:3) The following lemma is easy. Lemma 7.5. Let A be the algebra kq[x](v) for some n, q, v. Let f be an element of the form (x1 · · · xn)N for some N > 0. Then there is an integer a > 0 such that f ab is dominating for all integer b > 0. Proof. Let Φ := {xd1 s=1 ds = v} be the set of monomials of degree v, which is a k-basis of the degree 1 component of A after regrading. Let P be the product of elements in Φ. Then P =k× (x1 · · · xn)a for some a > 0. Then f ab =k× P bN . It suffices to show that P bN is dominating. But this is [CPWZ1, Lemma 2.2(1)]. (cid:3) n ds ≥ 0,Pn 1 · · · xdn Now we are ready to prove Theorem 0.4. In fact we prove that the algebras are strongly cancellative. Theorem 7.6. Let A be kq[x1, · · · , xn](v) where v is a positive integer and let m ≥ 2 be the order of q. Suppose that one of the following is true. (a) n is even and m does not divide v. (b) n is odd and gcd(m, v) 6= 1. Then A is strongly cancellative. Proof. Under the hypotheses (a) or (b), by Theorems 4.4 and 5.3, there is some p and w such that d[p] w (A/C) is of the form (x1 · · · xn)N for some N > 0. By Lemma 7.5, the element (d[p] w (A/C))ab is dominating for some a > 0 and all b > 0. The assertion follows from Lemma 7.4. (cid:3) We make some comments and remarks for the rest of this section. Lemma 7.7. Let {A1, · · · , As} be a set of algebras as in Theorem 7.6(a,b) with possible repetition. Let A be the tensor product A1 ⊗ · · · ⊗ As. Then some p-power discriminant of A over its center is dominating. Proof. Each algebra Ai has some p-power discriminant (over its center) that is dominating by Theorems 4.4 and 5.3. The assertion follows from Lemma 2.6(3) together with induction. Some of the hypotheses in Lemma 2.6 can be verified by using Lemma 3.3. (cid:3) Remark 7.8. Let A be as in Lemma 7.7. (1) By using the discriminant method [CPWZ1, CPWZ2], we obtain that every automorphism of A is graded. Therefore it is a linear algebra problem to determine the full automorphism group of A. In many case (when the Gelfand-Kirillov dimension of A is small), one can explicitly work out the full automorphism group of A. (2) By Lemma 7.4 and 7.7, A is strongly cancellative. 22 K. CHAN, A.A. YOUNG, AND J.J. ZHANG 8. Tits alternative Recall that, in the last few sections, we are only considering the case when q 6= 1, which implies that (E8.0.1) k 6= Z/(2). In this section, if q = 1, we will further assume that k 6= Z/(2). Note that (E8.0.1) is one of the hypotheses in [CPWZ3, Proposition 2.5]. Firstly we consider the case when n = 2s + 1 is odd and g := gcd(m, v) = 1 where m ≥ 2 is the order of q. Since gcd(m, v) = 1, there are two positive integers α and β such that (E8.0.2) (α + s)m − βv = 1. Lemma 8.1. Retain the above hypotheses. (1) The following are locally nilpotent derivations of kq[x] of degree βv. (a) (b) ∂1 : xi −→(xαm 0 2 ∂3 : xi −→(xαm 0 2 (x2xm−1 3 x4xm−1 5 · · · x2sxm−1 2s+1) (xm−1 1 x2x4xm−1 5 · · · x2sxm−1 2s+1) i = 1, i 6= 1. i = 3, i 6= 3. (2) Let g1 = exp(∂1) and g3 = exp(∂3). Then g1 and g3 are two automorphisms of kq[x] that generate a free subgroup of Aut(kq[x]). (3) Both g1 and g2 send a homogeneous element f of degree h to a linear com- bination of homogeneous elements of degrees h + βvN. As a consequence, both g1 and g2 restrict to algebra automorphisms of kq[x](v). Proof. (1) (a) The degree of ∂1 is αm + sm − 1 = βv. To check that ∂1 is a derivation we just verify that ∂1(xj xi − qxixj) = 0 for all 1 ≤ i < j ≤ n, which is straightforward by the choice of ∂1(xi). It is clear that ∂2 1 is locally nilpotent. The proof of (1)(b) is similar. 1 (xi) = 0. Then ∂2 (2) By definition, for any power d, we have gd 2 2 xi xi 1(xi) =(x1 + d xαm =(x1 + d x3[xαm 3(xi) =(x3 + d xαm =(x1 + d x1[xαm xi xi 2 2 and gd (x2xm−1 3 x4xm−1 5 · · · x2sxm−1 2s+1) i = 1, i 6= 1, (x2xm−2 3 x4xm−1 5 · · · x2sxm−1 2s+1)] i = 1, i 6= 1, (xm−1 1 x2x4xm−1 5 · · · x2sxm−1 2s+1) i = 1, i 6= 1, (xm−2 1 x2x4xm−1 5 · · · x2sxm−1 2s+1)] i = 1, i 6= 1. Let R be the subalgebra of kq[x] generated by x2, x4, x5, · · · , xn. Then g1 satisfies [CPWZ3, (E2.1.1)] with a0 = 0 and a1 = xαm 2s+1) and g3 satisfies [CPWZ3, (E2.1.2)] (when x2 is replaced by x3) with b0 = 0 and b1 = · · · x2sxm−1 (x2xm−2 x4xm−1 3 5 2 DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 23 (xm−2 1 5 x2x4xm−1 · · · x2sxm−1 xαm It is clear that a1b1 is transcendental over k. 2 Also R + Rx1 + Rx3 is a free R-module of rank 3. Thus we have checked all hypotheses of [CPWZ3, Proposition 2.5]. By [CPWZ3, Proposition 2.5(2)] and its proof, the subgroup generated by g1 and g3 is free. 2s+1). (3) Since ∂1 has degree βv, the first assertion follows because g1 = exp(∂1). In particular, g1 maps a homogeneous element of degree v to a linear combination of homogeneous elements of degrees in v + βvN. Thus g1 restricts to an automorphism of kq[x](v). The same statement holds for g3. (cid:3) Lemma 8.2. Let A be a connected graded domain generated in degree one. Let g be an automorphism of kq[x] such that (a) g(x) = x + higher degree terms for all x of degree 1, and (b) g and g−1 send a homogeneous element of degree v to a linear combination of homogeneous elements of degrees in vN. Then g restricts to an automorphism g′ of A(v). Further g is the identity if and only of g′ is. Proof. The first assertion is easy to show. Now we assume that g′ is the identity. Then g′(xv) = xv for all x ∈ A of degree 1. This implies that v deg g(x) = deg g(x)v = deg g(xv) = deg g′(xv) = deg xv = v. Hence deg g(x) = 1 and g(x) = x by hypothesis (a). (cid:3) Now we are ready to prove the first Tits alternative theorem. Theorem 8.3. Suppose that n is odd. (1) If gcd(m, v) > 1, then Aut(kq[x](v)) is virtually abelian. (2) If gcd(m, v) = 1, then Aut(kq[x](v)) contains a free subgroup of rank 2. Proof. (1) This follows from Theorems 0.1 and 0.2. (2) Let g1 and g3 be the automorphisms of kq[x] given in Lemma 8.1. By Lemma 8.1(2), the elements g1 and g3 generate a free subgroup of rank 2, by Lemma 8.1(3), they restrict to automorphisms g′ 1 and g′ 3 generates a free subgroup of Aut(kq[x](v)). Let Id 6= g ∈ hg1, g3i ⊆ Aut(kq[x]) and 3i ⊆ Aut(kq[x](v)). By Lemma 8.2, the 1, g′ let g′ be the corresponding element in hg′ element g′ is not the identity. Therefore hg′ 1, g′ (cid:3) 3 of kq[x](v). We claim that g′ 3i is a free group of rank 2. 1 and g′ Secondly we consider the case when n is even and write n = 2s. As before let m be the order of q. We consider the case where m divides v and write v = mγ. Lemma 8.4. Let n = 2 and v = mγ and A = kq[x](v). Then Aut(A) contains a free subgroup of rank 2. Proof. By direct computation or equations similar to (E3.0.4)-(E3.0.5), if n = 2, A is isomorphic to the commutative ring k[x1, x2](v). So we identify these two algebras. Consider two derivations and ∂1 : x1 → xv+1 2 , x2 → 0 ∂2 : x1 → 0, x2 → xv+1 1 . 24 K. CHAN, A.A. YOUNG, AND J.J. ZHANG Let g1 = exp(∂1) and g2 = exp(∂2). Then, by [CPWZ3, Proposition 2.5], g1 and g2 generate a free subgroup of rank 2. Since the degree of ∂i is v, we see that g1 and g2 restrict to automorphisms g′ 2 of k[x1, x2](v). By Lemma 8.2, the subgroup of Aut(A) generated g′ 1 and g′ (cid:3) 1 and g′ 2 is free of rank 2. Proof of Theorem 0.7. When n is odd, this follows from Theorem 8.3. When n = 2, this follows from Theorem 0.2 and Lemma 8.4. (cid:3) References [AlC] J. Alev and M. Chamarie, D´erivations et Automorphismes de Quelques Alg´ebres Quan- tiques, Comm. Algebra 20(6) (1992), 1787 -- 1802. [AlD] J. Alev and F. Dumas, Rigidit´e des plongements des quotients primitifs minimaux de Uq(sl(2)) dans l'alg´ebre quantique de Weyl-Hayashi, Nagoya Math. J. 143 (1996), 119 -- 146. q (g), Quantum groups, [AnD] N. Andruskiewitsch and F. Dumas, On the automorphisms of U + 107 -- 133, IRMA Lect. Math. Theor. Phys., 12, Eur. Math. Soc., Zurich, 2008. [BJ] V.V. Bavula and D.A. Jordan, Isomorphism problems and groups of automorphisms for generalized Weyl algebras. Trans. Amer. Math. Soc. 353 (2001), no. 2, 769 -- 794. [BZ] J. Bell and J.J. Zhang, Zariski Cancellation problem for noncommutative algebras, preprint (2016), arXiv:1601.04625. [CPWZ1] S. Ceken, J. Palmieri, Y.-H. Wang and J.J. Zhang, The discriminant controls automor- phism groups of noncommutative algebras, Adv. Math., 269 (2015), 551-584. [CPWZ2] S. Ceken, J. Palmieri, Y.-H. Wang and J.J. Zhang, The discriminant criterion and automorphism groups of quantized algebras, Adv. Math., 285 (2016), 754 -- 801. [CPWZ3] S. Ceken, J. Palmieri, Y.-H. Wang and J.J. Zhang, Invariant theory for quantum Weyl algebras under finite group action, Proceedings of Symposia in Pure Mathematics (to appear), preprint (2015), arXiv:1501.07881. [CYZ] K. Chan, A.A. Young and J.J. Zhang, Discriminant formulas and applications, Algebra & Number Theory, (in press, 2016), arXiv:1503.06327. [GTK] J. G´omez-Torrecillas and L. El Kaoutit, The group of automorphisms of the coordinate ring of quantum symplectic space, Beitrage Algebra Geom. 43 (2002), no. 2, 597 -- 601. [GY] K.R. Goodearl and M.T. Yakimov, Unipotent and Nakayama automorphisms of quantum nilpotent algebras, preprint, arXiv:1311.0278, 2013. [Gu1] N. Gupta, On the Cancellation Problem for the Affine Space A3 in characteristic p, Inven- tiones Math., 195 (2014), no. 1, 279 -- 288. [Gu2] N. Gupta, On Zariski's Cancellation Problem in positive characteristic, Adv. Math. 264 (2014), 296 -- 307. [Gu3] N. Gupta, A survey on Zariski Cancellation Problem, Indian J. Pure Appl. Math., 46(6) (2015), 865 -- 877. [LL] S. Launois and T.H. Lenagan, Automorphisms of quantum matrices, Glasg. Math. J. 55, (2013), no. A, 89 -- 100. [LMZ] J.-F. Lu, X.-F. Mao and J.J. Zhang, Nakayama automorphism and applications, Trans. Amer. Math. Soc. (2016) (to appear). [SU] I. Shestakov and U. Umirbaev, The tame and the wild automorphisms of polynomial rings in three variables, J. Amer. Math. Soc. 17 (1) (2004) 197 -- 227. [SAV] M. Su´arez-Alvarez and Q. Vivas, Automorphisms and isomorphism of quantum generalized Weyl algebras, J. Algebra 424 (2015), 540 -- 552. [Ti] J. Tits, Free subgroups in linear groups, J. Algebra 20 (1972), 250 -- 270. [Y1] M. Yakimov, The Andruskiewitsch-Dumas conjecture, Selecta Math. (N.S.) 20 (2014), no. 2, 421 -- 464. [Y2] M. Yakimov, The Launois-Lenagan conjecture, J. Algebra 392 (2013), 1 -- 9. (Chan) Department of Mathematics, Box 354350, University of Washington, Seattle, Washington 98195, USA E-mail address: [email protected] DISCRIMINANTS AND AUTOMORPHISM GROUPS OF VERONESE SUBRINGS 25 (Young) Department of Mathematics, DigiPen Institute of Technology, Redmond, WA 98052, USA E-mail address: [email protected] (Zhang) Department of Mathematics, Box 354350, University of Washington, Seat- tle, Washington 98195, USA E-mail address: [email protected]
1807.08020	2	1807	2019-05-21T18:52:28	More on a Curious Nucleus	[ "math.RA", "math.GN" ]	Harold Simmons introduced a pre-nucleus and its associated nucleus that measure the subfitness of a frame in Simmons [2010]. In this paper we continue the study if this pre-nucleus. We answer the questions posed in Simmons [2010].	math.RA	math	More on a Curious Nucleus Levon Haykazyan May 23, 2019 Abstract Harold Simmons in ["A curious nucleus," Journal of Pure and Ap- plied Algebra, 2010] introduced a pre-nucleus and its associated nucleus that measure the subfitness of a frame. Here we continue the study of this pre-nucleus. We answer the questions posed by Simmons. 1 Introduction A frame A is a complete lattice (our lattices will always be bounded) satis- fying the distributivity law a ∧ _ B = _{a ∧ b : b ∈ B}, for arbitrary a ∈ A and B ⊆ A. Given a topological space S, the lattice OS of open subsets of S is a frame. The study of frames as an algebraic approach to topology has been initiated by Ehresmann in 1950s. Frames later acquired independent existence as a generalisation of topology that is applicable to wider settings. In an arbitrary distributive lattice D we can consider the preorder a (cid:22) b iff (∀c ∈ D)[a ∨ c = ⊤ =⇒ b ∨ c = ⊤] that refines the order ≤ of the lattice. A frame is called subfit if (cid:22) agrees with ≤. Subfitness was introduces by Isbell [1972] to show that compact subfit frames are spatial (isomorphic to the lattice of open sets of a topological space). It was rediscovered by Simmons [1978] as a weak separation property. 1 Prompted by Coquand [2003], Simmons proposed in Simmons [2010] to study the map ξ : A → A ξ(a) = _{b ∈ A : b (cid:22) a} for an arbitrary frame A. In this way, the frame A is subfit if and only if ξ is the identity. As any algebraic structure, a frame can be studied through its quotients. There is an elegant way of handling quotients of a frame: a quotient is completely determined by the map taking each element to the join of all elements with the same image. Such maps are called nuclei. We can compare two nuclei (or more generally arbitrary maps) f, g : A → A pointwise, i.e. f ≤ g iff for every a ∈ A we have f (a) ≤ g(a). This makes the set of all nuclei into a poset, which turns out to be a frame too. Meets in this frame are computed pointwise, however describing joins is more complicated. For this reason, one often looks at a larger class of maps called pre-nuclei (see the definition below). The map ξ is a pre-nucleus. Simmons [2010] poses a number of questions about the pre-nucleus ξ and its relationship with other nuclei. We answer these questions in this paper. In section 2 we study the pre-nucleus on a general frame. We show that ξ is equal to the join of all (pre-)nuclei that only admit the top element ⊤ (Theorem 2.2). Next we construct an example to show that the pre-nucleus ξ (and consequently the join of all (pre-)nuclei admitting only the top element) need not be a nucleus (Example 2.3). To obtain a nucleus from ξ, in general, we therefore need to iterate it. We show that there is no a priori bound on the closure ordinal of ξ (Theorem 2.5). In section 3 we study ξ on the frame ID of ideals of a distributive lattice D where ξ is idempotent and hence a nucleus. (This is the setting in Coquand [2003] which motivated the definition of ξ in Simmons [2010]. However the nucleus ξ in this setting has already appeared in Johnstone [1984]. This ref- erence was apparently missed by both Coquand [2003] and Simmons [2010].) Actually ID is the frame of open subsets of the spectrum SpecD of D. So a nucleus picks out the subspace of prime ideals fixed by it. The subspace determined by ξ is characterised in Johnstone [1984] as the soberification of the space of maximal ideals of D. This characterisation immediately implies some equivalent conditions for ID to be subfit (Proposition 3.2). We call such lattices Jacobson. Following Simmons [2010] we compare ξ to another nucleus χ : ID → ID defined by χ(I) = Sc∈I{a ∈ D : a (cid:22) c} (note that (cid:22) here is computed in 2 D rather than ID). Actually (cid:22) preserves (binary) meets and joins and therefore determines a congruence ≡ on D. From this it follows that the subspace of SpecD determined by χ is homeomorphic to Spec(D/≡). We characterise it as the least spectral subspace of SpecD containing maximal ideals (Theorem 3.8). Lastly we show that χ = ξ if and only if D/≡ is a Jacobson lattice (Theorem 3.10). Acknowledgement The author is grateful to James Parson and the anonymous referee for careful reading, comments and suggestions for improving the paper. 2 The Pre-Nucleus ξ We assume some familiarity with the theory of frames. The necessary back- ground can be found e.g. in Johnstone [1982] or Picado and Pultr [2012]. Let A be a frame. Definition 2.1. • An inflator is a map f : A → A that is inflationary and monotone, i.e. a ≤ f (a) and a ≤ b =⇒ f (a) ≤ f (b) for all a, b ∈ A. • A pre-nucleus is an inflator that satisfies f (a) ∧ f (b) ≤ f (a ∧ b) for all a, b ∈ A. (The inequality is actually an equality by monotonic- ity.) • A nucleus is an idempotent pre-nucleus, i.e. it satisfies f 2 = f ◦ f = f. Before proceeding further, let us stress that in some literature (e.g. in Picado and Pultr [2012] and in Banaschewski [1988], where it is introduced) the term pre-nucleus is used for an inflator that satisfies the weaker property f (a) ∧ b ≤ f (a ∧ b). We follow Simmons [2010] in our choice of terminology. 3 Given any family F of functions from A to A we can consider their point- wise join W F : A → A defined by (_ F )(a) = _{f (a) : f ∈ F }. The pointwise join of a nonempty directed family of pre-nuclei is a pre- nucleus. For a pre-nucleus f : A → A we say that f admits a ∈ A if f (a) = ⊤. Let Fn and Fp be the set of all nuclei and pre-nuclei respectively that only admit ⊤: that is the set of all respective f such that f (a) = ⊤ =⇒ a = ⊤. Since Fn ⊆ Fp we have that W Fn ≤ W Fp. Note that Fp is closed under composition and is therefore directed. It follows that W Fp is a pre-nucleus. We will later see that, in contrast, W Fn need not be a nucleus. Now let f ∈ Fp and a, b ∈ A. Since f is an inflator we have that f (a)∨b ≤ f (a ∨ b). Therefore we have f (a) ∨ b = ⊤ =⇒ f (a ∨ b) = ⊤ =⇒ a ∨ b = ⊤. We conclude that f (a) (cid:22) a and therefore f ≤ ξ. It follows that W Fp ≤ ξ. We now show that it is in fact an equality, answering question (5) of Simmons [2010]. Theorem 2.2. For an arbitrary frame A we have ξ ≤ W Fn. Therefore W Fn = W Fp = ξ. Proof. Fix a ∈ A and let b (cid:22) a. We construct a nucleus g ∈ Fn such that b ≤ g(a). Let G(c) = {x ∈ A : x ≤ b ∨ c and x ∧ a ≤ c} and define g(c) = W G(c). Since c ∈ G(c) we conclude that c ≤ g(c). Note also that c ≤ c′ =⇒ G(c) ⊆ G(c′) =⇒ g(c) ≤ g(c′). Therefore g is an inflator. Also by frame distributivity, G(c) is closed under arbitrary joins and in particular g(c) ∈ G(c). Now given c, c′ ∈ A we have g(c) ∧ g(c′) ≤ (b ∨ c) ∧ (b∨c′) = b∨(c∧c′) and (g(c)∧g(c′))∧a ≤ c∧c′. Hence g(c)∧g(c′) ∈ G(c∧c′) and therefore g(c) ∧ g(c′) ≤ g(c ∧ c′). Thus g is a pre-nucleus. The map g is actually a nucleus. To see this we need to use the fact that g(c) ∈ G(c). So let d ∈ G(g(c)). Then d ≤ b ∨ g(c) ≤ b ∨ (b ∨ c) = b ∨ c. Also 4 d ∧ a ≤ g(c). But we also have d ∧ a ≤ a. Therefore d ∧ a ≤ g(c) ∧ a ≤ c. We conclude that d ∈ G(c), which implies that g(g(c)) = g(c). We claim that g ∈ Fn. Indeed assume that g(c) = ⊤. Then ⊤ ∈ G(c) and therefore ⊤ ≤ b ∨ c and ⊤ ∧ a ≤ c. Which is to say b ∨ c = ⊤ and a ≤ c. But since b (cid:22) a, the equality b ∨ c = ⊤ implies that a ∨ c = ⊤ and therefore c = ⊤. Finally we have b ≤ b ∨ a and b ∧ a ≤ a. Therefore b ∈ G(a) and so b ≤ g(a) ≤ (W Fn)(a). Thus ξ(a) = W{b ∈ A : b (cid:22) a} ≤ (W Fn)(a). Next we give an example of a frame to show that ξ (and therefore W Fp) need not be a nucleus. This answers the question (1) (and also question (3)) of Simmons [2010]. Our example is actually spatial (i.e. the frame of opens of a topological space). Given a topological space S consider the frame OS of open subsets. In this case for W, U ∈ OS we have U (cid:22) W iff ∀u ∈ (U \ W )[u− 6⊆ U], where u− denotes the topological closure of {u} (see Simmons [2010]). Example 2.3. Let ω = {0, 1, ...} be the first infinite ordinal and ω+ = ω ∪ {∗}, where ∗ is a new element. The universe of our topological space is ω × ω+. The basis of the topology is given by sets of the following form U × {y}, where U ⊆ ω is downward closed and y ∈ ω and ω × V ∪ U × {∗}, where U ⊆ ω is downward closed and V ⊆ ω is cofinite. We claim that ξ(∅) = ω ×ω. Indeed given (x, y) ∈ ω ×ω consider its open neighbourhood U = [0, x + 1) × {y}. The point (x + 1, y), which is outside of U is in the closure of every point of U. This shows that U (cid:22) ∅ and therefore (x, y) ∈ ξ(∅). Now consider a point (x, ∗). Each basic open neighbourhood W of it contains a subset of the form ω × {y}, with y ∈ ω, which is closed. Therefore W 6(cid:22) ∅. To see that ξ is not a nucleus note that ξ 2(∅) = ξ(ω × ω) = ω × ω+. Indeed given a point (x, ∗), consider its open neighbourhood U = ω × ω ∪ [0, x + 1) × {∗}. Now the point (x + 1, ∗), which is not in U, is in the closure of every point of U \ ω × ω. This shows that U (cid:22) ω × ω. 5 So in general ξ is not a nucleus. However for any pre-nucleus there is a least nucleus above it. To obtain it we need to iterate the pre-nucleus as follows. For an ordinal α define ξα by • ξ 0 is the identity function; • ξα+1 = ξ ◦ ξα; • ξα = Wβ<α ξβ if α is a limit ordinal. Then for each frame A (by cardinality considerations) there is an ordinal α such that ξα+1 = ξα, which will then be a nucleus. This ordinal α in general depends on A. We next show that there is no a priori bound, answering question (2) (and question (4)) of Simmons [2010]. For that we need to analyse the above example in more details. We recall two constructions on frames. The first construction is the Cartesian product (or just product) of frames. Suppose we have a collection (Ai : i ∈ I) of frames. Their Cartesian product is the frame with underlying set Πi∈IA where the operations of meet and join are performed coordinatewise. We denote the Cartesian product of two frames A1 and A2 by A1 × A2. It follows that (ai)i∈I (cid:22) (bi)i∈I if and only if ai (cid:22) bi for every i ∈ I. Therefore ξ((ai)i∈I) = (ξ(ai))i∈I and more generally ξα((ai)i∈I) = (ξα(ai))i∈I. Note that there is a slight abuse of notation here, since we use the same letter ξ for nuclei on all frames. The second construction is the tensor product (or coproduct) of frames. We only need binary tensor products, so we introduce only binary ones. Let A and B be frames. Their tensor product A ⊗ B is the frame presented by generators a ⊗ b, where a ∈ A and b ∈ B, subject to the following relations • (a1 ⊗ b1) ∧ (a2 ⊗ b2) = (a1 ∧ a2) ⊗ (b1 ∧ b2), so meets are computed coordinatewise; • Wi(ai ⊗ b) = (Wi ai) ⊗ b, in particular for a nullary join we have ⊥ = ⊥ ⊗ b; • Wi(a ⊗ bi) = a ⊗ (Wi bi), and thus ⊥ = a ⊗ ⊥. The general element of A ⊗ B has the form Wi ai ⊗ bi, though this represen- tation is not unique. Now let A and B be frames and let P(ω) denote the frame of all subsets of ω. Denote C = B ⊗ P(ω) and consider the subframe D of C × A such 6 that (c, a) ∈ D iff either a = ⊥A or there is a cofinite subset Y ⊆ ω such that ⊤B ⊗ Y ≤ c. It is easy to check that D is closed under finite meets and arbitrary joins. For notational purposes below we write ξ(c, a) for ξ((c, a)) and Y c for the complement ω \ Y of a subset Y ⊆ ω. Lemma 2.4. In the frame D the following hold: 1. if ⊤B 6= b ∈ B, then ξ(b ⊗ ω, ⊥A) = (ξ(b) ⊗ ω, ⊥A); 2. ξ(⊤C, a) = (⊤C, ξ(a)). Proof. 1. Let (c, a) (cid:22) (b ⊗ ω, ⊥A) in D. First we show that a = ⊥A. Assume otherwise; then by the definition of D, there is a cofinite subset Y ⊆ ω such that ⊤B ⊗ Y ≤ c. Let j ∈ Y . Consider the element (⊤B ⊗ {j}c, ⊤A). It's join with (c, a) is ⊤D, however the join with (b ⊗ ω, ⊥A) is not (since b 6= ⊤B). This is a contradiction showing that a = ⊥A. Now let c = Wi bi ⊗Xi and Xi 6= ∅. We claim that (c, ⊥A) (cid:22) (b⊗ω, ⊥A) implies that bi (cid:22) b. It is enough to assume that c = b0 ⊗X0 and X0 6= ∅. We need to show that b0 (cid:22) b. Let b′ ∈ B be such that b0 ∨ b′ = ⊤B. Pick j ∈ X0 and consider ((⊤B ⊗ {j}c) ∨ (b′ ⊗ {j}), ⊤A) ∈ D. Its join with (c, ⊥A) is ⊤D and therefore so is its join with (b ⊗ ω, ⊥A). But this implies that b ∨ b′ = ⊤B, which proves that b0 (cid:22) b. From this we deduce that ξ(b ⊗ ω, ⊥A) ≤ (ξ(b) ⊗ ω, ⊥A). To see the converse, we show that if b0 (cid:22) b, then (b0 ⊗ ω, ⊥A) (cid:22) (b ⊗ ω, ⊥A). Let (c′, a′) ∈ D be such that ((b0 ⊗ ω) ∨ c′, a′) = ⊤D. Then a′ = ⊤A and therefore there is a cofinite subset Y = ω \ {i0, ..., in−1} such that ⊤B ⊗ Y ≤ c′. Thus c′ = ⊤B ⊗ Y ∨ Wn−1 j=0 bij ⊗ {ij} for some bij ∈ B. It follows that for each j = 0, ..., n − 1 we have bij ∨ b0 = ⊤B. Thus by the assumption that b0 (cid:22) b we get that bij ∨ b = ⊤B. From this we deduce that ((b ⊗ ω) ∨ c′, a′) = ⊤D. 2. This follows from the following easy observation (c′, a′) (cid:22) (⊤C, a) if and only if a′ (cid:22) a. 7 Theorem 2.5. For every ordinal α > 0 there is a frame Aα such that α is the least ordinal with ξα(⊥A) = ⊤A. Proof. By induction on α. • For α = 1 we can pick e.g. A1 to be the frame of downward closed subsets of ω. • Assume the hypothesis for α. Let Aα+1 be the subframe of (Aα ⊗ P(ω)) × A1 described in the previous lemma. Then, by that lemma, α is the least ordinal such that ξα(⊥Aα+1) = (ξα(⊥Aα) ⊗ ω, ⊥A1) = (⊤Aα ⊗ ω, ⊥A1). Hence ξα+1(⊥Aα+1) = (⊤Aα ⊗ ω, ξ(⊥A1)) = ⊤Aα+1 and α + 1 is the least such. • Let α be a limit ordinal and assume the hypothesis for all β < α. Take Aα = Qβ<α Aβ. 3 The Nuclei ξ and χ on a Frame of Ideals Let D be a distributive lattice. Then the lattice ID of all ideals of D is a frame. It is the frame of open sets of the spectrum SpecD of D. Recall that SpecD is the space of prime ideals of D in the topology generated by basic open sets of the form {I ∈ SpecD : a 6∈ I} for a ∈ D. Since we have two lattices now, we will use lowercase letters for elements of D and uppercase letters for elements of ID. Let I ∈ ID and a ∈ D. Denote by ↓a the principal ideal {b ∈ D : b ≤ a}. We then have ↓a (cid:22) I iff for every ideal J ∈ ID such that ↓a ∨ J = D we have I ∨ J = D. It is enough to check for principal ideals J = ↓b. Thus we can reformulate this as ↓a (cid:22) I iff for every b ∈ D we have a ∨ b = ⊤D implies there is c ∈ I such that c ∨ b = ⊤D. Finally observing that a ∈ ξ(I) iff ↓a (cid:22) I we get a ∈ ξ(I) ⇐⇒ (∀b ∈ D)(cid:2)a ∨ b = ⊤D =⇒ (∃c ∈ I)[c ∨ b = ⊤D](cid:3). 8 From this characterisation it is easy to check that ξ 2(I) = ξ(I) and therefore ξ is a nucleus on ID. The nucleus ξ picks out the subspace (SpecD)ξ = {I ∈ SpecD : ξ(I) = I} of SpecD whose frame of opens is the lattice (ID)ξ = {I ∈ ID : ξ(I) = I}. Part of question (7) of Simmons [2010] asks to characterise the space (SpecD)ξ. In fact such a characterisation had already appeared in Johnstone [1984]. Fact 3.1 (Johnstone [1984]). For every ideal I ∈ ID, the ideal ξ(I) is the intersection of maximal ideals containing I. The space (SpecD)ξ is the soberification of the subspace of maximal ideals of D. It is compact and Jacobson. (Recall that a space is Jacobson if every closed set is the closure of its closed points). We have the following corollary. Proposition 3.2. The following conditions are equivalent on a distributive lattice D. 1. Every prime ideal is an intersection of maximal ideals. 2. Every ideal is an intersection of maximal ideals. 3. The frame ID is subfit, i.e. ξ is the identity on ID. 4. The spectrum SpecD is a Jacobson space. Proof. 1 =⇒ 2 holds since every ideal of a distributive lattice is the inter- section of prime ideals containing it. 2 =⇒ 3. Assuming 2, by Fact 3.1 we get that the nucleus ξ is the identity on ID. Therefore ID is subfit. 3 =⇒ 4 again follows from Fact 3.1. 4 =⇒ 1. Assume that SpecD is a Jacobson space. Let I ∈ SpecD and consider its closure I −. By the assumption, it is the closure of its closed points {Jl : l ∈ L}. Then each Jl is a maximal ideal containing I. We claim that I = Tl∈L Jl. Indeed suppose that a ∈ D \ I. Then I is in the basic open set D(a) = {J ∈ SpecD : a 6∈ J}. Since I is in the closure of {Jl : l ∈ L} there is an l ∈ L such that Jl ∈ D(a). But then a 6∈ Jl. We are not aware of studies of such lattices in the literature. However, given the obvious analogy with rings, there is only one name we can give to them. 9 Definition 3.3. A distributive lattice satisfying the equivalent conditions of Proposition 3.2 is called a Jacobson lattice. Simmons [2010] compares ξ to another nucleus χ : ID → ID defined by χ(I) = [ {a ∈ D : a (cid:22) c}. c∈I The map χ is indeed a nucleus and it is easy to check that χ ≤ ξ. In the rest of this section we will characterise the space (SpecD)χ = {I ∈ SpecD : χ(I) = I} as well as characterise the lattices D for which χ = ξ. For this we need to look at the preorder (cid:22) in more details. We would like to make an order out of (cid:22). For that we need to quotient D by the equivalence relation ≡ defined by a ≡ b if a (cid:22) b and b (cid:22) a. Lemma 3.4. The relation ≡ is the largest lattice congruence on D such that a ≡ ⊤ implies a = ⊤. Proof. It is easy to check that ≡ is a congruence. If a ≡ ⊤, then ⊤ (cid:22) a. But since ⊤ ∨ ⊥ = ⊤ the definition of (cid:22) gives a = a ∨ ⊥ = ⊤. Now suppose ≡′ is a congruence with this property. Let a, b ∈ D be such that a ≡′ b. Fix c ∈ D with property a ∨ c = ⊤. Then we have ⊤ ≡′ a ∨ c ≡′ b ∨ c and hence b ∨ c = ⊤. This shows that a (cid:22) b. Since ≡′ is symmetric we also get b (cid:22) a and so a ≡ b. We now consider the lattice D/≡ (with ordering (cid:22)) and the associated surjection from D. In general, if D′ is another distributive lattice and f : D → D′ a homomorphism, then the inverse image map f −1 : SpecD′ → SpecD is a spectral map (preimages of compact open sets are compact open). Conversely any spectral map from a spectral space to SpecD induces a lattice homomorphism. This is the well known Stone representation theorem for distributive lattices. The surjective lattice homomorphisms are characterised as follows. Fact 3.5 (Dickmann et al. [2019]). A lattice homomorphism f : D → D′ is surjective if and only if the spectral map f −1 : SpecD′ → SpecD is a homeomorphism onto its image. Thus lattice congruences of D are in one-to-one correspondence between spectral subspaces of SpecD in the following sense. 10 Definition 3.6. Let X be a spectral space and Y ⊆ X be a subspace that is also spectral. We say that Y is a spectral subspace of X if the inclusion map is spectral. In other words if Z ⊆ X is compact and open, then Z ∩ Y is also compact. Remark 3.7. Let ≡′ be a congruence on D, then the image of the canonical map SpecD/≡′ → SpecD can be characterised as those prime ideals I of D that are stable under ≡′, i.e. a ∈ I and a ≡′ b implies b ∈ I. Indeed, preimages of ideals of D/≡′ are clearly stable under ≡′. Conversely any ideal I ∈ SpecD that is stable under ≡′ is the preimage of the ideal {a/≡′ : a ∈ I} ∈ SpecD/≡′. Now we can characterise the space (SpecD)χ. This, together with Fact 3.1, answers question (7) of Simmons [2010]. Theorem 3.8. The space (SpecD)χ is the least spectral subspace of SpecD containing the set of maximal ideals. Proof. The space (SpecD)χ is a spectral subspace by Lemma 3.5. Further since χ fixes all maximal ideals, they are contained in (SpecD)χ. Now let X ⊆ SpecD be a spectral space containing all maximal ideals. Let D′ be the lattice of its compact open sets. Then by Lemma 3.5, there is a surjective homomorphism f : D → D′. This map f induces a congruence a ≡′ b iff f (a) = f (b) on D. Note that if a 6= ⊤, then there is a maximal ideal I of D containing a. By the assumption, I ∈ X and so is stably under ≡′. It follows that a 6≡′ ⊤. Thus by Lemma 3.4 ≡′⊆≡ and so all ideals that are stable under ≡ are stable under ≡′. Hence by Remark 3.7 (SpecD)χ ⊆ X. Remark 3.9. James Parson has pointed out to me that the quotient lattice D/ ≡ has been studied in Coquand et al. [2006] under the name Heitmann lattice of D. The above characterisation of (SpecD)χ also appears there. We can now answer question (6) of Simmons [2010] by characterising the lattices for which ξ and χ agree. Theorem 3.10. Let D be a distributive lattice. Then χ = ξ if and only if D/≡ is a Jacobson lattice. Proof. Assume that χ = ξ. Then Spec(D/≡) is homeomorphic to (SpecD)χ = (SpecD)ξ which is Jacobson by Fact 3.1. Hence D/≡ is a Jacobson lattice. 11 Conversely assume that D/≡ is a Jacobson lattice. Let I ∈ ID be an ideal such that I = χ(I). We show that ξ(I) = I. Let a 6∈ I. It is enough to find a maximal ideal extending I and not containing a. Consider the ideal I/≡ of D/≡. Since the latter is a Jacobson lattice, there is a maximal ideal J ′ of D/≡ extending I/≡ and not containing a/≡. Let J = {b ∈ D : b/≡ ∈ J ′} be the preimage of J ′ under the canonical map. Then I ⊆ J and a 6∈ J. But if c 6∈ J, then there is b ∈ J such that b ∨ c ≡ ⊤D (since J ′ = J/≡ is maximal). Then by Lemma 3.4 we get that b ∨ c = ⊤D showing that J is a maximal ideal. As a final remark we highlight some connections with model theory. The passage from D to D/≡ (and therefore the nucleus χ) has a manifestation in model theory: it corresponds to the passage from a theory to its largest companion. This can most elegantly be expressed in coherent logic (also called positive model theory). However it can already be seen in the classical Robinson model theory, which is the setting we adopt here. Suppose T is an inductive theory (i.e. ∀∃-axiomatisable) and Πn is the set of universal formulas in n variables ¯x = (x1, ..., xn) (we identify two formulas φ(¯x) and ψ(¯x) if T = ∀¯x(φ(¯x) ↔ ψ(¯x))). Then Πn is naturally a distributive lattice with the order φ(¯x) ≤ ψ(¯x) iff T = ∀¯x(φ(¯x) → ψ(¯x)). Then it can be shown that φ (cid:22) ψ in Πn if and only if ∀¯x(ψ(¯x) → φ(¯x)) is true in every existentially closed model of T . Let TK denote the ∀∃-theory of all existentially closed models of T . Then TK and T are companions meaning that they imply the same Πn formulas. The fact that TK is the largest companion of T is precisely because of Lemma 3.4. We can further see that T will have a model companion (i.e. every model of TK is existentially closed) if and only if for every n the lattice Πn/≡ is a boolean algebra (and therefore also Jacobson). References Bernhard Banaschewski. Another look at the localic Tychonoff theorem. Commentationes Mathematicae Universitatis Carolinae, 29(4):647 -- 656, 1988. Thierry Coquand. Compact spaces and distributive lattices. Journal of Pure and Applied Algebra, 184(1):1 -- 6, 2003. 12 Thierry Coquand, Henri Lombardi, and Claude Quitt´e. Dimension de Heit- mann des treillis distributifs et des anneaux commutatifs. Publications Math´ematiques de Besan¸con, Alg`ebre et Th´eorie des Nombres, 2006:57 -- 100, 2006. Max Dickmann, Niels Schwartz, and Marcus Tressl. Spectral Spaces. Cam- bridge University Press, 2019. John R. Isbell. Atomless parts of spaces. Mathematica Scandinavica, 31: 5 -- 32, 1972. Peter Johnstone. Stone Spaces. Cambridge University Press, 1982. Peter Johnstone. Almost maximal ideals. Fundamenta Mathematicae, 123: 197 -- 209, 1984. Jorge Picado and Ales Pultr. Frames and Locales. Springer, 2012. Harold Simmons. The lattice theoretical part of topological separation prop- erties. Proceedings of the Edinburgh Mathematical Society, 21:41 -- 48, 1978. Harold Simmons. A curious nucleus. Journal of Pure and Applied Algebra, 214(11):2063 -- 2073, 2010. 13
1511.00679	1	1511	2015-11-01T15:57:47	On intra-regular and left regular and left duo ordered $\Gamma$-semigroups	[ "math.RA" ]	For an intra-regular or a left regular and left duo ordered $\Gamma$-semigroup $M$, we describe the principal filter of $M$ which plays an essential role in the structure of this type of $po$-$\Gamma$-semigroups. We also prove that an ordered $\Gamma$-semigroup $M$ is intra-regular if and only if the ideals of $M$ are semiprime and it is left (right) regular and left (right) duo if and only if the left (right) ideals of $M$ are semiprime.	math.RA	math	On intra-regular and left regular and left duo ordered Γ-semigroups Niovi Kehayopulu, Michael Tsingelis Abstract. For an intra-regular or a left regular and left duo ordered Γ-semigroup M , we describe the principal filter of M which plays an essential role in the structure of this type of po-Γ-semigroups. We also prove that an ordered Γ-semigroup M is intra-regular if and only if the ideals of M are semiprime and it is left (right) regular and left (right) duo if and only if the left (right) ideals of M are semiprime. AMS Subject Classification: 20M99 (06F99) Keywords: ordered Γ-semigroup; filter; intra-regular; left regular 1 Introduction and prerequisites Croisot, who used the term "inversive" instead of "regular", connects the matter of decomposition of a semigroup with the regularity and semiprime conditions [1]. A semigroup S is said to be left (resp. right) regular if for every a ∈ S there exists x ∈ S such that a = xa2 (resp. a = a2x). That is, if a ∈ Sa2 (resp. a ∈ a2S) for every a ∈ S which is equivalent to saying that A ⊆ A2S (resp. A ⊆ SA2) for every A ⊆ S. A semigroup S is said to be intra-regular if for every a ∈ S there exist x, y ∈ S such that a = xa2y. In other words, if a ∈ Sa2S for every a ∈ S or A ⊆ SA2S for every A ⊆ S. For decompositions of an intra-regular, left regular or both left regular and right regular semigroup we refer to [2, 10]. The concepts of intra-regular ordered semigroup and of right regular ordered semigroups have been introduced in [3, 4] in which the decomposition of an intra-regular ordered semigroup into simple components and the decomposition of a right regular and right duo ordered semigroup into right simple components has been studied. The principal filter of S has a very simple form, both for ordered and non-ordered case of semigroups, and it plays an essential role in the decompositions. For two nonempty sets M and Γ, we denote by AΓB the set containing the elements of the form aγb where a ∈ A, γ ∈ Γ and b ∈ B. That is, we define AΓB := {aγb a ∈ A, b ∈ B, γ ∈ Γ}. 1 Then M is called a Γ-semigroup if the following assertions are satisfied: (1) MΓM ⊆ M; (2) aγ(bµc) = (aγb)µc for all a, b, c ∈ M and all γ, µ ∈ Γ; (3) if a, b, c, d ∈ M and γ, µ ∈ Γ such that a = c, γ = µ and b = d, then aγb = cµd. An ordered Γ-semigroup (shortly, po-Γ-semigroup) is a Γ-semigroup M with an order relation "≤" on M such that a ≤ b implies ac ≤ bc and ca ≤ cb for every c ∈ M. A nonempty subset A of M is called a subsemigroup of M if, for every a, b ∈ A and every γ ∈ Γ, we have aγb ∈ A. A subsemigroup F of M is called a filter of M if (1) for every a, b ∈ F and every γ ∈ Γ such that aγb ∈ F , we have a ∈ F and b ∈ F and (2) if a ∈ F and M ∋ b ≥ a, then b ∈ F . For an element x of M, we denote by N(x) the filter of M generated by x (that is, the least with respect to the inclusion relation filter of M containing x). A nonempty subset A of M is called a left (resp. right) ideal of M if (1) MΓA ⊆ A (resp. AΓM ⊆ A) and (2) if a ∈ A and M ∋ b ≤ a, then b ∈ A. It is called an ideal or (two-sided ideal) of M if it is both a left and right ideal of M. A po-Γ-semigroup M is called left (resp. right) duo if the left (resp. right) ideals of M are two-sided. A subset T of M is called semiprime if for every x ∈ M and every γ ∈ Γ such that xγx ∈ T , we have x ∈ T . For a subset H of M we denote by (H] the subset of M defined by (H] = {t ∈ M t ≤ a for some t ∈ H}. We clearly have M = (M], and for any subsets A, B, C, D of M, we have A ⊆ (A] = ((A]]; if A ⊆ B, then (A] ⊆ (B]; if A ⊆ B and C ⊆ D, then (AΓC] ⊆ (BΓD]; (A]Γ(B] ⊆ (AΓB]; and ((A]Γ(B]] = ((A]ΓB] = (AΓ(B]] = (AΓB]. As we know, some results on semigroups (ordered semigroups) can be transferred into Γ-semigroups (po-Γ-semigroups) just putting a Gamma in the appropriate place, while for some other results the transfer is not easy. A Γ-semigroup M is called intra-regular if a ∈ MΓaΓaΓM for every a ∈ M, equivalently if A ⊆ MΓAΓAΓM for every A ⊆ M. It is called left (resp. right) regular if a ∈ MΓaΓa (resp. a ∈ ΓaΓM) for every a ∈ M or A ⊆ MΓAΓA (resp. A ⊆ ΓAΓM) for every A ⊆ M. An ordered Γ-semigroup M is called intra-regular if for every a ∈ M we have a ∈ (MΓaΓaΓM], equivalently if for every A ⊆ M we have A ⊆ (MΓAΓAΓM]. An ordered Γ-semigroup M is called left (resp. right) 2 regular if a ∈ (MΓaΓa] (resp. (a ∈ aΓaΓM]) for every a ∈ M, equivalently if A ⊆ (MΓAΓA] (resp. A ⊆ (AΓAΓM]) for every A ⊆ M. Although some interesting results on Γ-semigroups are obtained with these definitions, these definitions fail to describe the principal filter of intra-regular, left regular and right regular Γ-semigroups (ordered Γ-semigroups) which play an essential role in the investigation. To overcome this difficulty, in [8] a new definition of intra- regular and of left regular Γ-semigroups has been introduced. The intra-regular Γ-semigroup has been defined as a Γ-semigroup such that a ∈ MΓaγaΓM for each a ∈ M and each γ ∈ Γ and the left (resp. right) regular Γ-semigroup as a Γ-semigroup in which a ∈ MΓaγa (resp. a ∈ aγaΓM) for each a ∈ M and each γ ∈ Γ and it is proved that a Γ-semigroup M is left regular (in that new sense) if and only if it is a union of a family of left simple subsemigroups on M. And in [9] we gave some further structure theorems of this type of Γ-semigroups using that new definition and the form of principal filters. But what happens in case of intra-regular and left or right regular po-Γ-semigroups? Can we describe the form of the principal filters using some new definitions like in the non-ordered case? The present paper gives the related answer. For more information on Γ (or po-Γ)-semigroups cf., for example, the papers in [5 -- 7] of the References, and the papers in which these papers refer. Examples on Γ-semigroups are also given in these papers. 2 On intra-regular ordered po-Γ-semigroups We characterize here the intra-regular po-Γ-semigroups in terms of filters, and we prove that a po-Γ-semigroup M is intra-regular if and only if the ideals of M are semiprime. Definition 1. An ordered Γ-semigroup M is called intra-regular if x ∈ (MΓxγxΓM] for every x ∈ M and every γ ∈ Γ. Theorem 2. An ordered Γ-semigroup M is intra-regular if and only if, for every x ∈ M, we have N(x) = {y ∈ M x ∈ (MΓyΓM]}. 3 Proof. =⇒. Let x ∈ M and T := {y ∈ M x ∈ (MΓyΓM]}. Then we have the following: (1) T is a nonempty subset of M. Indeed: Take an element γ ∈ Γ (Γ 6= ∅). Since M is intra-regular, we have x ∈ (MΓxγxΓM] = (cid:16)(MΓx)γxΓMi ⊆ (cid:16)(MΓM)ΓxΓMi ⊆ (MΓxΓM], so x ∈ T . (2) Let a, b ∈ T and γ ∈ Γ. Then aγb ∈ T . Indeed: Since a ∈ T , we have x ∈ (MΓaΓM]. Since b ∈ T , we have x ∈ (MΓbΓM]. Since M is intra-regular, x ∈ M and γ ∈ Γ, we have x ∈ (MΓxγxΓM]. Then we have x ∈ (MΓxγxΓM] ⊆ (cid:16)MΓ(MΓbΓM]γ(MΓaΓM]ΓMi = (cid:16)MΓ(MΓbΓM)γ(MΓaΓM)ΓMi = (cid:16)(MΓM)Γ(bΓM γMΓa)Γ(MΓM)i ⊆ (cid:16)MΓ(bΓM γMΓa)ΓMi. We prove that bΓM γMΓa ⊆ (cid:16)MΓ(aγb)ΓMi. Then we have x ∈ (cid:16)MΓ(cid:16)MΓ(aγb)ΓMiΓMi = (cid:16)MΓ(cid:16)MΓ(aγb)ΓM(cid:17)ΓMi = (cid:16)(MΓM)Γ(aγb)Γ(MΓM)i ⊆ (cid:16)MΓ(aγb)ΓMi, so aγb ∈ T . Let now bλuγvδa ∈ bΓM γMΓa for some u, v ∈ M, λ, δ ∈ Γ. Since M is intra-regular, for the elements bλuγvδa ∈ M and γ ∈ Γ, we have bλuγvδa ∈ (cid:16)MΓ(bλuγvδa)γ(bλuγvδa)ΓMi = (cid:16)(MΓbλuγv)δ(aγb)λ(uγvδaΓM)i ⊆ (cid:16)MΓ(aγb)ΓMi. (3) Let a, b ∈ M and γ ∈ Γ such that aγb ∈ T . Then a, b ∈ T . Indeed: Since aγb ∈ T , we have x ∈ (cid:16)MΓ(aγb)ΓMi ⊆ (cid:16)MΓaγ(MΓM)i ⊆ (MΓaΓM], so a ∈ T . Since x ∈ (cid:16)MΓ(aγb)ΓMi ⊆ (cid:16)(MΓM)γbΓMi ⊆ (MΓbΓM], we have b ∈ T . 4 (4) Let a ∈ T and M ∋ b ≥ a. Then b ∈ T . Indeed: Since a ∈ T , we have x ∈ (MΓaΓM]. Since a ≤ b, we have (MΓaΓM] ⊆ (MΓbΓM]. Then we have x ∈ (MΓbΓM], and b ∈ T . (5) Let F be a filter of M such that x ∈ F . Then T ⊆ F . Indeed: Let a ∈ T . Then x ∈ (MΓaΓM], so F ∋ x ≤ uλ(aµv) for some u, v ∈ M, λ, µ ∈ Γ. Since F is a filter of M, x ∈ F and M ∋ uλ(aµv) ≥ x, we have uλ(aµv) ∈ F . Since F is a filter of M, u, aµv ∈ M, λ ∈ Γ and uλ(aµv) ∈ F , we have aµv ∈ F , again since F is a filter of M, a, v ∈ M and µ ∈ Γ, we have a ∈ F . ⇐=. Let x ∈ M and γ ∈ Γ. Then x ∈ (MΓxγxΓM]. Indeed: Since N(x) is a subsemigroup of M, x ∈ N(x) and γ ∈ Γ, we have xγx ∈ N(x). By hypothesis, we get x ∈ (cid:16)MΓ(xγx)ΓMi = (MΓxγxΓM], thus M is intra-regular. ✷ Theorem 3. An ordered Γ-semigroup M is intra-regular if and only if the ideals of M are semiprime. Proof. =⇒. Let A be an ideal of M, x ∈ M and γ ∈ Γ such that xγx ∈ A. Since M is intra-regular, we have x ∈ (cid:16)MΓ(xγx)ΓMi ⊆ (cid:16)(MΓA)ΓMi ⊆ (AΓM] ⊆ (A] = A, then x ∈ A, and A is semiprime. ⇐=. Let x ∈ M and γ ∈ Γ. Then x ∈ (MΓxγxΓM]. In fact: The set (MΓxγxΓM] is an ideal of M. This is because it is a nonempty subset of M, MΓ(MΓxγxΓM] ⊆ (cid:16)MΓ(MΓxγxΓM]i = (cid:16)MΓ(MΓxγxΓM)i ⊆ (MΓxγxΓM], (MΓxγxΓM]ΓM ⊆ (MΓxγxΓM], and (cid:16)(MΓxγxΓM]i = (MΓxγxΓM]. Since (MΓxγxΓM] is semiprime and (xγx)γ(xγx) = xγ(xγx)γx ∈ MΓxγxΓM ⊆ (MΓxγxΓM], we have xγx ∈ (MΓxγxΓM]. Again since (MΓxγxΓM] is semiprime, we have x ∈ (MΓxγxΓM], so M is intra-regular. ✷ 3 On left regular and left duo po-Γ-semigroups First we notice that the left (and the right) po-Γ-semigroups are intra-regular. Then we characterize the po-Γ-semigroups which are both left regular and left duo in terms of filters and we prove that a po-Γ-semigroup M is left (resp. right) regular if and only if the left (resp. right) ideals of M are semiprime. Definition 4. An ordered Γ-semigroup M is called left regular (resp. right 5 regular) if x ∈ (MΓxγx] (resp. x ∈ (xγxΓM]) for every x ∈ M and every γ ∈ Γ. Proposition 5. Let M be an ordered Γ-semigroup. If M is left (resp. right) regular, then M is intra-regular. Proof. Let M be left regular, x ∈ M and γ ∈ Γ. Then we have x ∈ (MΓxγx] ⊆ (cid:16)MΓ(MΓxγx]γxi = (cid:16)MΓ(MΓxγx)γxi ⊆ (cid:16)(MΓM)Γ(xγx)ΓMi ⊆ (cid:16)MΓxγxΓMi, thus M is intra-regular. ✷ Theorem 6. An ordered Γ-semigroup M is left regular and left duo if and only if, for every x ∈ M, we have N(x) = {y ∈ M x ∈ (MΓy]}. Proof. =⇒. Let x ∈ M and T := {y ∈ M x ∈ (MΓy]}. Since M is left regular, we have x ∈ (MΓxγx] ⊆ (cid:16)(MΓM)Γxi ⊆ (MΓx], so x ∈ T , and T is a nonempty subset of M. Let a, b ∈ T and γ ∈ Γ. Since x ∈ (MΓa], x ∈ (MΓb] and M is left regular, we have x ∈ (MΓxγx] ⊆ (cid:16)MΓ(MΓb]γ(MΓa]i = (cid:16)MΓ(MΓb)γ(MΓa)i ⊆ (cid:16)MΓ(bγMΓa)i. In addition, bγMΓa ⊆ (MΓaγb]. Indeed: Let bγuµa ∈ bγMΓa, where u ∈ M and µ ∈ Γ. Since M is left regular, we have bγuµa ∈ (cid:16)MΓ(bγuµa)γ(bγuµa)i ⊆ (cid:16)MΓ(aγb)ΓMi = (cid:16)(MΓaγb]ΓMi. Since (MΓaγb] is a left ideal, it is a right ideal of M as well, so (MΓaγb]ΓM ⊆ (MΓaγb], then bγuµa ∈ (cid:16)(MΓaγb]i = (MΓaγb]. Hence we obtain x ∈ (cid:16)MΓ(MΓaγb]i = (cid:16)MΓ(MΓaγb)i ⊆ (cid:16)MΓ(aγb)i, from which aγb ∈ T . Let a, b ∈ M and γ ∈ Γ such that aγb ∈ T . Since x ∈ (MΓaγb] ⊆ (MΓb], 6 we have b ∈ T . Besides, x ∈ (MΓaγb] ⊆ (cid:16)(MΓa]ΓMi. The set (MΓa] as a left ideal, is a right ideal of M as well, so (MΓa]ΓM ⊆ (MΓa]. Thus we have x ∈ (cid:16)(MΓa]i = (MΓa], and a ∈ T . Let a ∈ T and M ∋ b ≥ a. Since M is left regular, we have x ∈ (MΓaγa] ⊆ (MΓbγb] ⊆ (cid:16)(MΓb]ΓMi. (MΓb] as a left ideal is a right ideal of M, so (MΓb]ΓM ⊆ (MΓb]. Hence we have x ∈ (cid:16)(MΓb]i = (MΓb], and b ∈ T . Let F be a filter of M such that x ∈ F and let a ∈ T . Since x ∈ (MΓa], we have F ∋ x ≤ uµa for some u ∈ M, µ ∈ Γ. Since F is a filter of M, we have uµa ∈ F , and a ∈ F . ⇐=. Let x ∈ M and γ ∈ Γ. Since x ∈ N(x) and N(x) is a subsemigroup of M, we have xγx ∈ N(x). By hypothesis, we get x ∈ (MΓxγx], so M is left regular. Let now A be a left ideal of M, a ∈ A, γ ∈ Γ and u ∈ M. Since aγu ∈ N(aγu) and N(aγu) is a filter of M, we have a ∈ N(aγu). By hypothesis, we have aγu ∈ (MΓa] ⊆ (MΓA] ⊆ (A] = A. Thus A is right ideal of M. ✷ The right analogue of Theorem 6 also holds, and we have Theorem 7. An ordered Γ-semigroup M is right regular and right duo if and only if, for every x ∈ M, we have N(x) = {y ∈ M x ∈ (yΓM]}. Theorem 8. An ordered Γ-semigroup M is left (resp. right) regular if and only if the left (resp. right) ideals of M are semiprime. Proof. =⇒. Let M be left regular, A a left ideal of M, x ∈ M and γ ∈ Γ such that xγx ∈ A. Then we have x ∈ (cid:16)MΓ(xγx)i ⊆ (MΓA] ⊆ (A] = A, so M is semiprime. ⇐=. Suppose the left ideals of M are semiprime and let x ∈ M and γ ∈ Γ. Since (MΓxγx] is a left ideal of M and (xγx)γ(xγx) ∈ (MΓxγx], we have xγx ∈ (MΓxγx], and x ∈ (MΓxγx], so M is left regular. ✷ 7 References [1] R. Croisot, Demi-groupes inversifs et demi-groupes reunions de demi-groupes sim- ples, Ann. Sci. Ecole Norm. Sup. 3, No. 70 (1953), 205 -- 208. [2] A.H. Clifford, G.B. Preston, The Algebraic Theory of Semigroups, Vol. I, Amer. Math. Soc. Math. Surveys 7, Providence, Rhode Island, 1961. [3] N. Kehayopulu, On right regular and right duo ordered semigroups, Math. Japon. 36, No. 2 (1991), 201 -- 206. [4] N. Kehayopulu, On intra-regular ordered semigroups, Semigroup Forum 46 (1993), 271 -- 278. [5] N. Kehayopulu, On prime, weakly prime ideals in po-Γ-semigroups, Lobachevskii J. Math. 30, No. 4 (2009), 257 -- 262. [6] N. Kehayopulu, On ordered Γ-semigroups, Sci. Math. Jpn. 71, No. 2 (2010), 179 -- 185. [7] N. Kehayopulu, On regular duo po-Γ-semigroups, Math. Slovaca 61, no. 6 (2011), 871 -- 884. [8] N. Kehayopulu, On left regular Γ-semigroups, Int. J. Algebra 8, no. 8 (2014), 389 -- 394. [9] N. Kehayopulu, M. Tsingelis, On intra-regular and some left regular Γ-semigroups, Quasigroups and Related Systems, 23 no. 2 (2015), to appear. [10] M. Petrich, Introduction to Semigroups, Charles E. Merrill Publ. Comp., A Bell & Howell Comp. Columbus, Ohio 1973. 8
1604.08337	4	1604	2017-10-12T16:17:27	Decomposition of integer-valued polynomial algebras	[ "math.RA" ]	Let $D$ be a commutative domain with field of fractions $K$, let $A$ be a torsion-free $D$-algebra, and let $B$ be the extension of $A$ to a $K$-algebra. The set of integer-valued polynomials on $A$ is ${\rm Int}(A) = \{f \in B[X] \mid f(A) \subseteq A\}$, and the intersection of ${\rm Int}(A)$ with $K[X]$ is ${\rm Int}_K(A)$, which is a commutative subring of $K[X]$. The set ${\rm Int}(A)$ may or may not be a ring, but it always has the structure of a left ${\rm Int}_K(A)$-module. A $D$-algebra $A$ which is free as a $D$-module and of finite rank is called ${\rm Int}_K$-decomposable if a $D$-module basis for $A$ is also an ${\rm Int}_K(A)$-module basis for ${\rm Int}(A)$; in other words, if ${\rm Int}(A)$ can be generated by ${\rm Int}_K(A)$ and $A$. A classification of such algebras has been given when $D$ is a Dedekind domain with finite residue rings. In the present article, we modify the definition of ${\rm Int}_K$-decomposable so that it can be applied to $D$-algebras that are not necessarily free by defining $A$ to be ${\rm Int}_K$-decomposable when ${\rm Int}(A) \cong {\rm Int}_K(A) \otimes_D A$. We then provide multiple characterizations of such algebras in the case where $D$ is a discrete valuation ring or a Dedekind domain with finite residue rings. In particular, if $D$ is the ring of integers of a number field $K$, we show that ${\rm Int}_K$-decomposable algebras $A$ correspond to maximal $D$-orders in a separable $K$-algebra $B$, whose simple components have as center the same finite unramified Galois extension $F$ of $K$ and are unramified at each finite place of $F$. Finally, when both $D$ and $A$ are rings of integers in number fields, we show that ${\rm Int}_K$-decomposable algebras correspond to unramified Galois extensions of $K$.	math.RA	math	Decomposition of Integer-valued Polynomial Algebras Department of Mathematics, University of Padova, Via Trieste, 63 35121 Padova, Italy Giulio Peruginelli Nicholas J. Werner Department of Mathematics, Computer and Information Science, SUNY College at Old Westbury, P.O. Box 210, Old Westbury, NY 11568 7 1 0 2 t c O 2 1 ] . A R h t a m [ 4 v 7 3 3 8 0 . 4 0 6 1 : v i X r a Abstract Let D be a commutative domain with field of fractions K, let A be a torsion-free D-algebra, and let B be the extension of A to a K-algebra. The set of integer-valued polynomials on A is Int(A) = {f ∈ B[X] f (A) ⊆ A}, and the intersection of Int(A) with K[X] is IntK(A), which is a commutative subring of K[X]. The set Int(A) may or may not be a ring, but it always has the structure of a left IntK(A)-module. A D-algebra A which is free as a D-module and of finite rank is called IntK-decomposable if a D-module basis for A is also an IntK(A)-module basis for Int(A); in other words, if Int(A) can be generated by IntK(A) and A. A classification of such algebras has been given when D is a Dedekind domain with finite residue rings. In the present article, we modify the definition of IntK-decomposable so that it can be applied to D-algebras that are not necessarily free by defining A to be IntK-decomposable when Int(A) is isomorphic to IntK(A)⊗D A. We then provide multiple characterizations of such algebras in the case where D is a discrete valuation ring or a Dedekind domain with finite residue rings. In particular, if D is the ring of integers of a number field K, we show that an IntK-decomposable algebra A must be a maximal D-order in a separable K-algebra B, whose simple components have as center the same finite unramified Galois extension F of K and are unramified at each finite place of F . Finally, when both D and A are rings of integers in number fields, we prove that IntK-decomposable algebras correspond to unramified Galois extensions of K. Keywords: Integer-valued polynomial, algebra, Int-decomposable, Maximal order, Finite unramified Galois extension MSC Primary 13F20 Secondary 16H10, 11C99 1. Introduction Let D be a commutative integral domain with field of fractions K. The ring of integer-valued polynomials over D is defined to be Int(D) := {f ∈ K[X] f (D) ⊆ D}. The ring Int(D), its Email addresses: [email protected] (Giulio Peruginelli), [email protected] (Nicholas J. Werner) Preprint submitted to Journal of Pure and Applied Algebra July 15, 2018 elements, and its properties have been popular objects of study over the past several decades and continue to be so today. The book [4] is the standard reference on the topic. Beginning around 2010, attention turned to polynomials that are evaluated on D-algebras rather than on D itself. This can be seen in the work of Evrard, Fares and Johnson [6, 7], Frisch [8, 9, 10, 11], Loper [16], Peruginelli [5, 13, 21, 22, 23, 24], Werner [31, 33, 34], and Naghipour, Rismanchian, and Sedighi Hafshejani [18]. A good example of these new rings of integer-valued polynomials comes from considering the polynomials in K[X] that map each element of the matrix algebra Mn(D) back to Mn(D). Example 1.1. Associate K with the scalar matrices in Mn(K). Then, for any polynomial f (X) = Pt i=0 qiX i ∈ K[X] and any matrix a ∈ Mn(D), we can evaluate f at a to produce the matrix f (a) = Pt i=0 qiai. If f (a) ∈ Mn(D) for each a ∈ Mn(D), then f is said to be integer-valued on Mn(D). The set of all such polynomials is denoted by IntK(Mn(D)) := {f ∈ K[X] f (Mn(D)) ⊆ Mn(D)}, and it is easy to verify that IntK(Mn(D)) is a subring of K[X]. We can form a larger collection of polynomials that are integer-valued on Mn(D) by considering polynomials whose coefficients come from Mn(K) rather than from K. That is, we form the set Int(Mn(D)) := {f ∈ Mn(K)[X] f (Mn(D)) ⊆ Mn(D)}. Since Mn(K) is noncommutative, we follow standard conventions regarding polynomials with non- commuting coefficients, as in [14, §16]. In Mn(K)[X], we assume that the indeterminate X com- mutes with each element of Mn(K), and we define evaluation to occur when the indeterminate is to the right of any coefficients. So, given f (X) = Pt i=0 qiX i ∈ Mn(K)[X], we consider f (X) to be equal to Pt i=0 X iqi as an element of Mn(K)[X], but to evaluate f (X) at a matrix a ∈ Mn(D), we must first write f (X) in the form f (X) = Pt i=0 qiai. A consequence of this is that evaluation is no longer a multiplicative homomorphism; that is, if f (X) = g(X)h(X) in Mn(K)[X], then it may not be true that f (a) equals g(a)h(a). Because of this difficulty, it is not clear whether Int(Mn(D)) is closed under multiplication. Despite the complications associated with evaluation of polynomials in this setting, one may prove that Int(Mn(D)) is a (noncommutative) subring of Mn(K)[X] [32, Thm. 1.2]. Thus, we are able to construct a noncommutative ring of integer-valued polynomials. i=0 qiX i, and then f (a) = Pt We can actually say more. In [8, Thm. 7.2], Sophie Frisch proved that Int(Mn(D)) is itself a matrix ring. Specifically, Int(Mn(D)) ∼= Mn(IntK(Mn(D))), where the isomorphism is given by associating a polynomial with matrix coefficients to a matrix with polynomial entries. (This isomor- phism is the restriction of the classical isomorphism between the polynomial ring Mn(K)[X] and the matrix ring Mn(K[X])). Because of Frisch's theorem, many questions about Int(Mn(D)) can be reduced to questions about IntK(Mn(D)), and the latter ring-being commutative-is usually easier to work with. Broadly speaking, the point of this paper is to study the relationship between a commutative ring of integer-valued polynomials such as IntK(Mn(D)) and its extension Int(Mn(D)). In particular, we wish to determine when and how Frisch's theorem [8, Thm. 7.2] can be generalized to algebras other than matrix rings. While matrix rings will be prominent in our work, the majority of our theorems deal with general algebras. However, our basic definitions are inspired by the situation described in Example 1.1. 2 We begin by giving notation and conventions for working with polynomials over algebras. As before, let D be a commutative integral domain with field of fractions K. Let A be a torsion-free D-algebra and take B = K ⊗D A to be the extension of A to a K-algebra. We associate K and A with their canonical images in B via the maps k 7→ k ⊗ 1 and a 7→ 1 ⊗ a. Much of our work will involve polynomials in B[X]. The algebra B may be noncommutative, but we will assume that X commutes with all elements of B. Moreover, we define evaluation of polynomials in B[X] at elements of A just as we did in Example 1.1 where A = Mn(D) and B = Mn(K). Given f (X) = Pt i=0 ciX i ∈ B[X] and b ∈ B, we define f (b) := tX i=0 cibi. Note that the map B[X] → B given by evaluation at b is not a multiplicative homomorphism unless b lies in the center of B. Finally, we define and Int(A) := {f ∈ B[X] f (A) ⊆ A} IntK(A) := Int(A) ∩ K[X] = {f ∈ K[X] f (A) ⊆ A}. We will also require that A ∩ K = D; this assumption is equivalent to the containment IntK(A) ⊆ Int(D). Definition 1.2. When A is a torsion-free D-algebra such that A ∩ K = D, we say that A is a D-algebra with standard assumptions. When A is finitely generated as a D-module, we say that A is of finite type. With these definitions, it is clear that IntK(A) is always a subring of the commutative ring K[X]. The algebraic structure of Int(A) is more difficult to analyze. It is straightforward to verify that Int(A) is closed under addition, and in fact has the structure of a left IntK(A)-module. However, because B[X] may contain polynomials with non-commuting coefficients, there is no guarantee that Int(A) is closed under multiplication. Indeed, let g, h ∈ Int(A) and let f = gh be the product of g and h in B[X]. Then, we have g(a), h(a) ∈ A for all a ∈ A, but because f (a) need not equal g(a)h(a), it is not clear whether or not f (a) is in A. Thus, we arrive at an important question: is Int(A) a ring when B is noncommutative? There are cases where Int(A) has been proved to be closed under multiplication, and thus has a ring structure under the usual operations inherited from B[X]. For instance, this will be true if A itself is a commutative ring. More generally, if each element of A is a sum of units and central elements, then Int(A) is a ring [32, Thm. 1.2]. In particular, this theorem applies when A = Mn(D), the algebra of n × n matrices with entries in D, because Mn(D) has a D-module basis consisting of invertible matrices. The condition in [32, Thm. 1.2] is sufficient for Int(A) to be a ring, but is not necessary; counterexamples may be found in [34, Ex. 3.8] and in [11]. To date, no example has been given of a noncommutative D-algebra A for which Int(A) is not a ring. As mentioned in Example 1.1, Frisch proved in [8, Thm. 7.2] that the rings Int(Mn(D)) and Mn(IntK(Mn(D))) are isomorphic. This result led the second author to search for other algebras with a similar property [31]. To do this, the problem was recast in the following way. Assume that A, as a D-module, is free of finite rank, with D-basis α1, . . . , αt. Then, when does α1, . . . , αt form 3 a basis for Int(A) as an IntK(A)-module? With this formulation, Frisch's theorem shows that Int(Mn(D)) = M 1≤i,j≤n IntK(Mn(D))Eij (1.3) where the Eij are the standard matrix units that form a D-basis of Mn(D). An algebra A = Li Dαi such that Int(A) = M i IntK(A)αi (1.4) is called IntK-decomposable with respect to {αi}i. By [31, Prop. 1.4], this property is independent of the D-basis chosen for A. Thus, a free D-algebra A such that (1.4) holds is called simply IntK- decomposable. We will use the adjective Int-decomposable (with no subscript K) when we wish to speak of these algebras collectively, without reference to a specific domain D or base field K. i=1 Mn(Fq). The main theorem of [31] proved that there is a close connection between Int-decomposable algebras and direct sums of matrix algebras. In [31, Thm. 6.1] it is shown that for D a Dedekind domain with finite residue rings and A a free D-algebra of finite rank, A is IntK-decomposable if and only if for each nonzero prime P of D, there exist n, t ∈ N and a finite field Fq such that A/P A ∼= Lt The work in [31] depended crucially on the presence of a D-module basis for A, and it was desirable to know if Int-decomposable algebras could be defined and studied without assuming that A was free. This is indeed possible, and doing so is the focus of the current paper. The key insight was to notice that an IntK-decomposable algebra is one for which Int(A) can be generated (as a subring of B[X]) by A and IntK(A), and this property can be precisely expressed in terms of tensor products of D-algebras. We say that A is IntK-decomposable if and only if Int(A) ∼= IntK(A)⊗D A (see Definition 2.3); the only limitations we impose on A are our standard assumptions that A is torsion-free and A ∩ K = D. Note that for the case of the full matrix algebra A = Mn(D) considered initially by Frisch, the matrix ring Mn(IntK(Mn(D))) is canonically isomorphic to the ring IntK(Mn(D)) ⊗D Mn(D). If D is Dedekind with finite residue rings and A is finitely generated as a D-module, then we are able to extend the classification given by [31, Thm. 6.1] (Theorem 2.10). Hence, IntK-decomposable algebras are those which are residually a direct sum of copies of a matrix ring over a finite field. Moreover, we are able to obtain two alternate characterizations of Int-decomposability, one in terms of the completions of A at primes of D (Theorem 3.6) and the other in terms of the extended K- algebra B = K ⊗D A (Theorem 4.10). We are also able to describe when IntK(A) = Int(D) (Theorem 2.11), which answers a question raised in [10]. In Section 2, we state the more general definition of IntK-decomposable and prove several of the theorems mentioned above. Section 3 discusses the classification of IntK-decomposable algebras in terms of completions. As our work will show, IntK-decomposable algebras are related to matrix algebras via their residue rings and completions, but the two types of algebras are not the same. Theorem 3.11 clarifies this situation by presenting various counterexamples. We close the paper by studying the consequences of Theorems 2.10 and 3.6. An easy corollary of our classification theorems is that an IntK-decomposable algebra A must be a maximal D-order in the K-algebra B, and B must be a semisimple K-algebra. Along these lines, in Section 4, we use the theory of maximal orders (as presented in [29]) to establish the last part of our classification. In Theorem 4.10 we prove that if D is the ring of integers of a number field K, then A is IntK- decomposable if and only if the following four conditions are satisfied: B is a separable K-algebra 4 with simple components which have the same center F ; F is a finite unramified Galois extension of K; the simple components of B are unramified at each finite place of F ; and A is a maximal D-order in B. From this general theorem, we obtain two relevant corollaries. First, if A is the ring of integers of a finite extension L of K, then A is IntK-decomposable if and only if L is an unramified Galois extension of K (Corollary 4.11). Second, if D = Z, then A is IntQ-decomposable if and only if for some n, A is isomorphic to a finite direct sum of copies of Mn(Z) (Corollary 4.12). This last result implies that if IntQ(A) = Int(Z), then A is isomorphic to a finite direct sum of copies of Z. 2. Int-decomposable Algebras We begin by recalling the definition of IntK-decomposability in the case of free D-algebras which was given in [31]. Definition 2.1. ([31, Def. 1.2]) Let A be a D-algebra that, as a D-module, is free of finite rank, i=1 Dαi for some D-module basis {α1, . . . , αt}. We say that A is IntK-decomposable i=1 IntK(A)αi as an IntK(A)-module. so that A = Lt with respect to {αi}t i=1 if Int(A) = Lt It is shown in [31, Prop. 1.4] that the IntK-indecomposability of A does not depend on the D-module basis {α1, . . . , αt}. That is, A is IntK-decomposable with respect to one basis if and only if it is IntK-decomposable with respect to every basis. Thus, we can-and will-say algebras are IntK-decomposable without referring to a specific basis. A useful way to interpret Definition 2.1 is the following. Assume A = Li Dαi. Then, we have B = Li Kαi and it follows that any f ∈ B[X] can be expressed (uniquely) in the form f = Pi fiαi, where each fi ∈ K[X]. If f ∈ Int(A) and A is IntK-decomposable then we may conclude that each fi ∈ IntK(A). This property can sometimes be used to quickly show that an algebra is not Int-decomposable. For example, let D = Z and A = Z[i], the Gaussian integers. Then, (1+i)(X 2−X) /∈ IntQ(A); hence, A = Z[i] is not IntQ-decomposable. As we shall see in Corollary 4.11, this is related to the fact that 2 is a ramified prime of Z[i]. The most prominent examples of IntK-decomposable algebras are the matrix rings Mn(D). As mentioned in the introduction, Frisch proved in [8, Thm. 7.2] that when A = Mn(D), there is a D-algebra isomorphism between Mn(IntK(A)) and Int(A), as in (1.3). Some rings of algebraic integers and certain quaternion algebras can also be Int-decomposable [31, Sec. 6]. 2 ∈ Int(A), but (X 2−X) 2 The main theorem of [31] shows that IntK-decomposable algebras can be recognized by the structure of their residue rings A/P A, where P runs through the primes of D. Theorem 2.2. ([31, Thm. 6.1]) Let D be a Dedekind domain with finite residue rings. Let A be a free D-algebra of finite rank with standard assumptions. Then, A is IntK-decomposable if and only if for each nonzero prime P of D, there exist n, t ∈ N and a finite field Fq such that A/P A ∼= Lt i=1 Mn(Fq). Both Definition 2.1 and the proof of Theorem 2.2 depended on the presence of a D-basis for A. Our first goal in this paper is to generalize the definition of IntK-decomposable so that it applies to algebras that are not necessarily free. We will then go on to show (Theorem 2.10) that Theorem 2.2 still holds under this more general definition. 5 Definition 2.3. Let D be an integral domain and A a torsion-free D-algebra. Consider the following D-bilinear map: IntK(A) × A → Int(A) (f (X), a) 7→ f (X) · a By the universal property of the tensor product, there exists a unique D-module homomorphism Φ : IntK(A) ⊗D A → Int(A) (2.4) which maps every elementary tensor product f (X) ⊗ a to f (X) · a. We say that A is IntK- decomposable if Φ is an isomorphism of D-modules, so that Int(A) ∼= IntK(A) ⊗D A. Recall that the tensor product IntK(A) ⊗D A has a natural D-algebra structure with multipli- cation given by (f1(X) ⊗ a1)(f2(X) ⊗ a2) = (f1(X)f2(X)) ⊗ (a1a2) (see [2, Chapt. III, §4, n. 1]) , for all fi ∈ IntK(A), ai ∈ A, i = 1, 2. Moreover, since the elements of K are central in B, the map Φ above induces a D-algebra structure on Int(A) (and so, Φ becomes a D-algebra homomorphism). Therefore, when A is IntK-decomposable, Φ is an isomorphism of D-algebras. This definition has a number of immediate consequences, among them that Definition 2.3 reduces to the original Definition 2.1 when A is free. Proposition 2.5. (1) If A is IntK-decomposable, then Int(A) is a ring. (2) The following are equivalent: (i) A is IntK-decomposable. (ii) the D-module Int(A) satisfies the universal property of the tensor product of IntK(A) and A. (iii) Int(A) is equal to the subring of B[X] generated by IntK(A) and A. (3) Assume A = Lt i=1 Dαi is free of finite rank as a D-module. Then, A is IntK-decomposable in the sense of Definition 2.1 if and only if A is IntK-decomposable in the sense of Definition 2.3. Proof. (1) This is a generalization of [31, Prop. 2.2]. When A is IntK-decomposable, it is actually isomorphic as a ring to the D-algebra IntK(A) ⊗D A. (2) The equivalence of (i) and (ii) is clear. For (iii), note that for any A (IntK-decomposable or not), the module Int(A) contains all products of the form f (X) · a, where f ∈ IntK(A) and a ∈ A. Since Int(A) is closed under addition, it also contains the subring of B[X] generated by IntK(A) and A. Given the definition of Φ, A being IntK-decomposable is equivalent to Int(A) equaling this subring. (3) Since A = Li Dαi, we have the chain of equalities IntK(A) ⊗D A = IntK(A) ⊗D (M Dαi) = M (IntK(A) ⊗D Dαi) = M i i IntK(A)αi i from which the equivalence of the definitions is clear. Remark 2.6. 6 • While A being IntK-decomposable implies that Int(A) is a ring, the converse is not true. There are numerous examples of D-algebras A which are not IntK-decomposable but still Int(A) is a ring. For instance, when G is a finite group and A is the group algebra DG, Int(A) is a ring by [32, Thm. 1.2]. However, whenever the characteristic of D/P divides G, the group ring A/P A ∼= (D/P )G is not semisimple [27, Cor. 3.4.8], and hence cannot satisfy Theorem 2.2. Thus, the group algebra DG will not be IntK-decomposable in such cases. Also, if p is an odd prime of Z, D = Z(p), and A is a free D-algebra of finite rank, then for each k > 0 the residue ring A/pkA has odd order. It then follows from [34, Thms. 2.4, 3.7] that Int(A) is a ring; but certainly A can be chosen so that A is not IntQ-decomposable. Finally, let Tn(D) be the D-algebra of n × n upper triangular matrices. Frisch has recently shown [11] that Int(Tn(D)) is a ring. But, Tn(D) does not satisfy the condition of Theorem 2.2, so Tn(D) is not IntK-decomposable. • According to Proposition 2.5, Frisch's result (1.3) can be restated as follows: Int(Mn(D)) = IntK(Mn(D)) ⊗D Mn(D). When D is a Dedekind domain, we can prove that the map Φ in Definition 2.3 is always injective. Lemma 2.7. Let D be a Dedekind domain and A a D-algebra of finite type with standard assump- tions. Then, Φ : IntK(A) ⊗D A → Int(A) is injective. Proof. Define Ψ : K[X] ⊗D A → B[X] by Ψ(f (X) ⊗D a) = f (X)a. Then, Ψ is an isomorphism of K-algebras. Since D is Dedekind and A is torsion-free, it follows that A is a projective D-module, hence flat. Therefore, the containment IntK(A) ⊆ K[X] implies that IntK(A) ⊗D A ⊆ K[X]⊗D A. Thus, the map Φ is the restriction of Ψ to IntK(A) ⊗D A, and since Ψ is injective so is Φ. In this way, for D Dedekind, we may identify IntK(A)⊗D A with the subring of B[X] generated by IntK(A) and A. We use this fact in proving the next proposition. Proposition 2.8. Let D be a Dedekind domain and A a D-algebra of finite type with standard assumptions. If IntK(A) = Int(D) then A is IntK-decomposable. Proof. Since D is Dedekind, Int(A) contains IntK(A) ⊗D A via the map Φ by Lemma 2.7. For the other containment, assume IntK(A) = Int(D) and let Int(D, A) be the set Int(D, A) := {f ∈ B[X] f (D) ⊆ A}. Clearly, Int(A) ⊆ Int(D, A). By [4, Prop. IV.3.3] we have Int(D, A) = Int(D) ⊗D A, so that Int(A) ⊆ Int(D, A) = Int(D) ⊗D A = IntK(A) ⊗D A, as required. We can also generalize [31, Thm. 3.3] and prove that IntK-decomposability is a local property when A is finitely generated. Note that this will hold without the assumption that D is Dedekind. Proposition 2.9. Let D be a domain and let A be a D-algebra of finite type with standard as- sumptions. Then A is IntK-decomposable if and only if AP is IntK-decomposable for each prime P of D. 7 Proof. By definition, A is IntK-decomposable if and only if the map Φ in (2.4) is an isomorphism. By [1, Prop. 3.9], this holds if and only the D-modules Int(A) and IntK(A) ⊗D A are isomorphic locally at each prime ideal P of D, that is, the induced maps ΦP : (IntK(A) ⊗D A) ⊗D DP → Int(A) ⊗D DP are isomorphisms for each prime P of D. Recall that for a D-module M and a multiplicative set S ⊂ D, we have S−1M ∼= M ⊗D S−1D. Thus, we always have AP ∼= A ⊗D DP and IntK(A)P ∼= IntK(A) ⊗D DP . Since A is finitely generated, by [31, Prop. 3.2] we have IntK(A)P = IntK(AP ) and Int(AP ) = Int(A)P . Hence, IntK(A) ⊗D DP ∼= IntK(AP ) and Int(A) ⊗D DP ∼= Int(AP ). Using this and other standard properties of tensor products (as in [2, Chap. II, §5]), we have (IntK(A) ⊗D A) ⊗D DP ∼= IntK(A) ⊗D (A ⊗D DP ) ∼= IntK(A) ⊗D AP ∼= (IntK(A) ⊗D DP ) ⊗DP AP ∼= IntK(AP ) ⊗DP AP . Hence, the induced map ΦP is an isomorphism if and only if Int(AP ) ∼= (IntK(A) ⊗D A) ⊗D DP ∼= IntK(AP ) ⊗DP AP , which means that AP is IntK-decomposable. We can now extend the classification of Int-decomposable algebras given in Theorem 2.2. Theorem 2.10. Let D be a Dedekind domain with finite residue rings. Let A be a D-algebra of finite type with standard assumptions. Then, A is IntK-decomposable if and only if for each nonzero prime P of D, there exist n, t ∈ N and a finite field Fq such that A/P A ∼= Lt exists a finite field Fq such that A/P A ∼= Lt i=1 Fq for some t ∈ N. In particular, if A is commutative, then A is IntK-decomposable if and only if for each P there i=1 Mn(Fq). Proof. Since D is Dedekind and A is finitely generated and torsion-free, A is a projective D-module [19, Cor. p. 30]. Hence, for each prime P , AP is free as a DP -module, and AP has finite rank because A is finitely generated. Applying Theorem 2.2 to AP , we see that AP is IntK-decomposable if and only if AP /P AP ∼= Li Mn(Fq) for some n and some Fq. Using Proposition 2.9 and the fact that A/P A ∼= AP /P AP , we obtain the stated theorem. Under the same hypotheses on D, we can describe those A for which IntK(A) = Int(D). This generalizes [31, Thm. 4.6], which dealt with the case where A was free. Theorem 2.11. Let D be a Dedekind domain with finite residue rings. Let A be a D-algebra of finite type with standard assumptions. Then, IntK(A) = Int(D) if and only if for each nonzero prime P of D, A/P A ∼= Lt Proof. We know that Int(D) = TP Int(D)P (where the intersection is over nonzero primes P of D) i=1 D/P for some t ∈ N. and that Int(D)P = Int(DP ) for each P . The analogous equalities for IntK(A) are shown in [31, Props. 3.1, 3.2]. Thus, IntK(A) = Int(D) if and only if IntK(AP ) = Int(DP ) for each P . But, as in Theorem 2.10, each AP is a free DP -module of finite rank. By [31, Thm. 4.6], IntK(AP ) = Int(DP ) if and only if AP /P AP ∼= Li DP /P DP . The result now follows because AP /P AP ∼= A/P A and DP /P DP ∼= D/P . 8 3. Int-decomposable Algebras via Completions In this section, we provide an alternate characterization of Int-decomposable algebras. As in Section 2, we assume that D is a Dedekind domain with finite residue fields and that A is a D- algebra of finite type with standard assumptions. Theorem 2.10 asserts that A is IntK-decomposable precisely when for each nonzero prime P of D, A/P A is isomorphic to a direct sum of copies of a matrix ring with entries in a finite field. Instead of focusing on A/P A, we can work with the P -adic completion bAP = lim ←− is the P -adic completion of D) because A is of finite type. In Theorems 3.6 and 3.10, we prove that both Int-decomposability and the equality IntK(A) = Int(D) can be characterized in terms of the A/P kA of A, which in this case is isomorphic to bDP ⊗D A (where bDP completions bAP . These results complement Theorems 2.10 and 2.11, respectively. Recall first the following definition. Definition 3.1. For a ring R and an R-algebra A, the null ideal of A with respect to R, denoted NR(A), is the set of polynomials in R[X] that kill A. That is, NR(A) = {f ∈ R[X] f (A) = 0}. When R is commutative, NR(A) is easily seen to be an ideal of R[X]. If R = A, we set NR(A) = N (A). Null ideals are a useful tool for dealing with integer-valued polynomials because there is a correspondence between the elements of IntK(A) and the null ideals ND/dD(A/dA), where d ∈ D. Specifically, let f (X) = g(X)/d ∈ K[X], where g(X) ∈ D[X] and d ∈ D. Then, f ∈ IntK(A) if and only if the residue of g in (D/dD)[X] is in ND/dD(A/dA). Int-decomposability can be expressed in terms of null ideals (this was the main strategy employed in [31]; see [31, Def. 4.3, Thm. 4.4]). To do this, we need a notion of "decomposability" for N (A/P kA). This is accomplished in the next definition, which is the analog of Definition 2.3. Definition 3.2. Let P be a nonzero prime of D and let k > 0. We say that A/P kA is ND/P k - decomposable if the canonical ring isomorphism (D/P k)[X] ⊗D/P k A/P kA ∼= (A/P kA)[X] that maps each elementary tensor product f (X) ⊗ a to f (X) · a induces the following isomorphism of D/P k modules: ND/P k (A/P kA) ⊗D/P k A/P kA ∼= N (A/P kA). Lemma 3.3. Let D be a Dedekind domain with finite residue rings. Let A be a D-algebra of finite type with standard assumptions. Then, A is IntK-decomposable if and only if A/P kA is ND/P k -decomposable for each nonzero prime P of D and each k > 0. Proof. By Proposition 2.9 we may localize at a prime P and assume that D is a discrete valuation ring, and hence that A is free. Furthermore, each A/P kA is free as a D/P k-module, so we can always find α1, . . . , αt ∈ A/P kA such that A/P kA = Lt i=1 D/P kαi. By Proposition 2.5 (3), A is IntK-decomposable if and only if A is IntK-decomposable in the sense of Definition 2.1; and by [31, Thm. 4.4, Def. 4.3], this is equivalent to having N (A/P kA) = Lt i=1 ND/P k (A/P kA)αi for each k > 0. Proceeding as in Proposition 2.5 (3), one may show this last condition is equivalent to having ND/P k (A/P kA) ⊗D/P k A/P kA ∼= N (A/P kA). By Proposition 2.9, IntK-decomposability is a local property. So, in this section we will often reduce to the local case, namely that D is a discrete valuation ring (DVR) with finite residue field Fq0 and maximal ideal P = πD. Moreover, notice that when P is a maximal ideal of D, we have A/P A ∼= AP /P AP , so that ND/P (A/P A) = NDP /P DP (AP /P AP ). 9 We will use this fact freely in our subsequent work. In order to ease the notation, we set Ak = A/P kA and Dk = D/P k, for each k ∈ N. Note that Dk ⊆ Ak and that Ak is a torsion-free Dk-algebra, which is finitely generated as a Dk-module. Lemma 3.4. ([31, Thm. 5.10]) Let k ∈ N. Then, Ak is NDk -decomposable if and only if Ak ∼= Lt i=1 Mn(Tk) for some n, t ∈ N and a finite commutative local ring Tk with principal maximal ideal mk which is generated by the same uniformizer π of D, so that mk = πTk. Recall that a commutative ring is chain ring if its set of ideals is totally ordered by inclusion. In particular, a ring Tk as described in Lemma 3.4 is a chain ring, as in [17]. Note that T1 is equal to a finite field Fq, which contains the residue field of D at P . In Lemma 3.4, the Ak's form an inverse system with respect to the natural projection maps Ak → Ak/πk−1Ak ∼= Ak−1, and these maps are compatible with finite direct product and matrix rings. In particular, we have Tk → Tk/πk−1Tk ∼= Tk−1 so the Tk's also form an inverse system of chain rings. Moreover, since the nilpotency of π in Ak is k, it follows that the nilpotency of π in Tk is also k and the residue field of Tk is T1 = Fq. Given that the P -adic completion bAP is equal to the inverse limit lim ←− Ak, it is natural to consider the inverse limit of the chain rings Tk. It is well known that the completion of any DVR V with maximal ideal m is realized as the inverse limit of the chain rings V /mk, k ∈ N. The next lemma shows that, under certain mild assumptions, the converse is also true (it is probable that this lemma is a known result, but a proof was not found in the available literature, so one is provided for the sake of the reader). Lemma 3.5. Let {Tk}k∈N be an inverse system of chain rings with maximal ideals mk = πkTk such that k is the nilpotency of πk, and the transition maps θk : Tk → Tk−1 are all surjective (so, without loss of generality, we may assume that πk 7→ πk−1). We assume that T1 is the common residue field of the rings Tk. Tk is a complete DVR with residue field isomorphic to T1. Then the inverse limit bT = lim ←− Proof. We identify the inverse limit bT with the subset of coherent sequences of the direct product of the Tk's: bT = lim ←− Tk = {(ak) ∈ Y k≥1 Tk ak+1 7→ ak,∀k ≥ 1}. Let bm = {(ak) ∈ bT ak ∈ mk,∀k ≥ 1}. Note that, for (ak) ∈ bT , (ak) is in bm if and only if for some k we have ak ∈ mk. Clearly, bm is an ideal of bT . We claim that every element of bT \ bm is invertible, which shows that bT is a local ring with maximal ideal bm. Indeed, let (ak) ∈ bT \ bm. Then for each k ∈ N, ak is invertible in Tk, and it is easy to see that (a−1 Moreover, bm = πbT , where π = (πk) ∈ bT . In fact, by definition π ∈ bm. Conversely, if (ak) ∈ bm, then, for all k ∈ N, we have ak = πkbk, for some bk ∈ Tk. One may verify that (bk) is a coherent sequence, so (ak) = π(bk) ∈ πbT . Note also that π is not a nilpotent element of bT , because if e ∈ N is such that πe = (πe Now, we clearly have Tk≥1 bmk = (0), so by [3, Chap. VI, §1, n. 4, Prop. 2] bT is a DVR. It k ) is a coherent sequence and is the inverse of (ak) in bT . k) = 0, then we have e ≥ k, for all k ∈ N, a contradiction. remains to show that bT is complete with respect to the bm-adic topology. 10 Since each transition map θk : Tk → Tk−1 is surjective, each projection ψk : bT → Tk is also surjective. Furthermore, the kernel of ψk is πk bT , since the maximal ideal of Tk has nilpotency k by assumption. So, we may identify each chain ring Tk with the residue ring bT /πk bT = bT /bmk. Hence, bT = lim ←− bT /bmk, which shows that the topology on bT as the inverse limit of the chain rings {Tk}k∈N coincides with the bm-adic topology and that bT is complete with respect to the bm-adic topology [3, Chap. III, §2, n. 6]. The next theorem gives the promised characterization of IntK-decomposable algebras in terms of the completions bAP . Given a prime ideal P of D, we denote by bDP the P -adic completion of D, and denote by bKP the P -adic completion of K (which is also the fraction field of bDP ). Theorem 3.6. Let D be a Dedekind domain with finite residue rings. Let A be a D-algebra of finite type with standard assumptions. Then, A is IntK-decomposable if and only if, for each nonzero prime ideal P ⊂ D, there exist n, t ∈ N such that tM (3.7) bAP ∼= i=1 Mn(bTP ) where bTP is a complete DVR with finite residue field and quotient field which is a finite unramified extension of bKP . prime ideal P ⊂ D, bAP ∼= Lt In particular, if A is commutative then A is IntK-decomposable if and only if, for each nonzero i=1 bTP , where bTP is as above. Proof. Without loss of generality, we may suppose that D is a DVR with maximal ideal P = πD. We retain the notation introduced at the beginning of this section. Suppose first that A is IntK-decomposable. Then, the P -adic completion bAP of A is equal to the inverse limit of the rings Ak [3, Chap. III, §2, n. 6]. For each k ∈ N, let Tk be as in Lemma 3.4, and let bTP be their inverse limit, which is a complete DVR with maximal ideal bmP by Lemma 3.5. Since the formation of inverse limit commutes with finite direct sums, by Lemma 3.4 we have Dk ⊆ bTP = lim ←− Now, Dk ⊆ Ak ∼= Li Mn(Tk), so by [31, Lem. 5.1] each matrix ring Mn(Tk) contains a (central) copy of Dk, which is contained in the set of scalar matrices Tk. Since the maximal ideals of Dk and Tk have the same generator π, Dk ⊆ Tk is an unramified extension of chain rings [17, p. 281]. So, by Lemma 3.5, bDP = lim Tk is an unramified extension of complete DVRs since π ←− generates the maximal ideals of both bDP and bTP . Let bFP be the quotient field of bTP , let Fq be the residue field of bTP , and let Fq0 be the residue field of bDP . Then, bFP / bKP is an unramified field extension of finite degree [bFP : bKP ] = [Fq : Fq0 ]. This establishes that if A is IntK-decomposable, then we have the desired decomposition (3.7) for bAP . Conversely, suppose (3.7) holds, where bTP is a complete DVR which is a finite unramified extension of bDP . In particular, bP · bTP = bm, the maximal ideal of bTP . As above, let Fq be the (finite) residue field of bTP . Then, bAP = lim ←−k≥1 tM Ak = lim ←−k≥1 i=1 Mn(Tk) = tM i=1 Mn(bTP ). (3.8) tM i=1 Mn(Fq) bA/bP bA ∼= A/P A ∼= 11 so by Theorem 2.10 A is IntK-decomposable. Theorem 3.6 is the analog of Theorem 2.10. There is also an analogous form of Theorem 2.11, the proof of which requires the next lemma. Lemma 3.9. Let D be a DVR with maximal ideal P = πD. Let A be a D-algebra with standard assumptions, and let bA be the P -adic completion of A. Then, IntK( bA) = IntK(A). Proof. The containment IntK( bA) ⊆ IntK(A) is clear, since A embeds in bA. Conversely, let f ∈ IntK(A) and α ∈ bA. Suppose f (X) = g(X)/πk, where g ∈ D[X] and k ∈ N. f ∈ D[X] ⊆ IntK( bA), so assume that k > 1. Via the canonical projection bA → A/πkA, we see that there exists a ∈ A such that α ≡ a (mod πk bA). Since the coefficients of g are central in A, we get g(α) ≡ g(a) (mod πk bA). Thus, f (α) = f (a) + λ/πk, where λ ∈ πk bA, so that f (α) ∈ bA. Hence, f ∈ IntK( bA) and IntK( bA) = If k = 0, then IntK(A). Theorem 3.10. Let D be a Dedekind domain with finite residue rings. Let A be a D-algebra of finite type with standard assumptions. Then, IntK(A) = Int(D) if and only if, for each nonzero i=1 bDP , for some t ∈ N. prime ideal P of D, bAP ∼= Lt Proof. Note that if t is the rank of A, defined as the dimension of B = K ⊗D A over K, then, for each prime ideal P of D, the rank of AP = DP ⊗D A ⊂ B is equal to t, so that t does not depend on the particular prime ideal P . As we have already remarked, AP is free as a DP -module, so that AP ∼= Lt i=1 DP (as DP -modules). We may work locally since bAP = (cAP )P DP . So, we will assume that D is a DVR and we will omit the suffix P . If IntK(A) = Int(D) then A is IntK-decomposable, by Proposition 2.8. Hence, by Theorem 2.11, we have A/P A ∼= Lt i=1 D/P . By Theorem 3.6, the completion bA decomposes as a finite direct sum of matrix rings over a complete DVR bT , which is a finite unramified extension of bD (so that bP · bT is equal to the maximal ideal bm of bT ). Since bA/bP bA ∼= A/P A, formula (3.7) becomes bA ∼= Lt i=1 bD, that is, n = 1 and bD = bT , because the residue field bT /bm of bT must be isomorphic to D/P , the residue field of bD. Conversely, if bA is isomorphic to a finite direct sum of copies of bD, then by Lemma 3.9 we have IntK(A) = IntK( bA) = IntK(M i bD) = IntK(bD) = Int(D) as desired. when D = Z, A is IntQ-decomposable if and only if there exist n, t ∈ N such that A ∼= Lt At this point, it is apparent that Int-decomposable algebras are related to matrix algebras via their residue rings and completions. At the close of this paper, we will prove (Corollary 4.12) that i=1 Mn(Z) (which implies that if A is IntQ-decomposable, then IntQ(A) = IntQ(Mn(Z)) for some n). However, in general Int-decomposable algebras need not be direct sums of matrix algebras. We end this section with a theorem that considers some of the conditions that link matrix algebras with Int- decomposable algebras, and examines the implications among these conditions. 12 Theorem 3.11. Let D be a Dedekind domain with finite residue rings and A a D-algebra of finite type with standard assumptions. Consider the following four conditions: (i) there exists n ∈ N such that A ∼= Mn(D). (ii) there exists n ∈ N such that bAP ∼= Mn(bDP ), for all primes P of D. (iii) A is IntK-decomposable. (iv) there exists n ∈ N such that IntK(A) = IntK(Mn(D)). Then, the following implications hold: (i) ⇒ (ii), (ii) ⇒ (iii), and (ii) ⇒ (iv); but for none of these three implications does the converse hold. Finally, (iii) 6⇒ (iv) and (iv) 6⇒ (iii). Proof. The implication (i) ⇒ (ii) is clear, and (ii) ⇒ (iii) is Theorem 2.10. ((ii) ⇒ (iv)) Clearly, it is sufficient to prove the statement locally at each maximal ideal P of D. Thus, we suppose that D is a DVR. By Lemma 3.9 we have IntK(A) = IntK( bA) and IntK(Mn(D)) = IntK(Mn(bD)), and since bA ∼= Mn(bD) the statement holds. We now show by counterexamples that the other stated implications do not hold. ((ii) 6⇒ (i)) Let p be an odd prime of Z, let D = Z(p), and let A be the standard quaternion algebra A = D ⊕ Di ⊕ Dj ⊕ Dk (so that i2 = j2 = −1 and ij = k = −ji). Then, it is well known (cf. [12, Exer. 3A]) that A/pkA ∼= M2(D/pkD) ∼= M2(Z/pkZ) for all k > 0, so that bA ∼= M2(Zp). However, A 6∼= M2(D), because (among other reasons) A contains no nonzero nilpotent elements. ((iii) 6⇒ (ii)) Take A = Mn(D)⊕ Mn(D). Then, A is IntK-decomposable by Theorem 2.10, but A/P A ∼= Mn(D/P ) ⊕ Mn(D/P ), so (ii) does not hold. ((iv) 6⇒ (ii)) Again take A = Mn(D) ⊕ Mn(D). Then, (ii) does not hold, but IntK(A) = IntK(Mn(D)) by [31, Thm. 2.3]. ((iii) 6⇒ (iv)) Let K $ L be an unramified Galois extension of number fields. Let D = OK and take A = OL. Then-as we will show in Corollary 4.11-A is IntK-decomposable. How- ever, we argue that for all n ∈ N we have IntK(OL) 6= IntK(Mn(D)). First, if n = 1, then IntK(OL) = IntK(D) = Int(OK ); but, by Corollary 4.11, this is impossible because L 6= K. So, assume that n > 1. Then, by [24, Prop. 7] IntK(OL) is integrally closed (this also follows from [16, Thm. 3.7]). But, IntK(Mn(D)) is never integrally closed when n > 1 [25, Cor. 3.4], so IntK(OL) 6= IntK(Mn(D)) for n > 1. ((iv) 6⇒ (iii)) Let n > m ≥ 1 and take A = Mn(D) ⊕ Mm(D). Then, IntK(Mn(D)) ⊆ IntK(Mm(D)), so IntK(A) = IntK(Mn(D)) by [31, Thm. 2.3]. But, for any prime P of D, we have A/P A ∼= Mn(D/P ) ⊕ Mm(D/P ), which does not satisfy Theorem 2.10. 4. Extended Algebras and Maximal Orders In this final section, we examine the consequences of Theorems 2.10 and 3.6, which allows us to give a global characterization of Int-decomposable algebras. The descriptions given in Theorems 13 2.10 and 3.6 show that an Int-decomposable algebra can be described-either residually or in terms of its completions-in terms of matrix rings. In the case where D is the ring of integers of a number field K, we are able to completely classify IntK-decomposable algebras A as those which are the maximal orders of the extended K-algebra B; B is a separable K-algebra whose simple components share a common center F ; F is a finite unramified Galois extension of K; and each simple component of B is unramified at each finite place of F . The work in this section relies heavily on the theory of maximal orders, as presented in [29]. We begin by recalling several definitions from [29] and [30]. As in earlier sections, D denotes a Dedekind domain with fraction field K. Definitions 4.1. Let B be a finite dimensional K-algebra. • By the Wedderburn Structure Theorem [28, Thm. 3.5], if B is semisimple, then we have B = Lr i=1 Mni(Di), for some uniquely determined r, ni ∈ N and division rings Di; the Bi = Mni(Di) are the simple components of B. We denote by Z(Di) the center of Di, which is a finite field extension of K. We say that B is separable if B is a finite dimensional semisimple K-algebra, such that the center of each simple component of B is a separable field extension of K [29, p. 99]. • A D-order in B is a subring A of B such that A is a finitely generated D-submodule of B and K · A = B. A maximal D-order in B is a D-order that is not properly contained in any other D-order of B (see [29, p. 108, 110]). Note that in the setting of this paper, a D-algebra A of finite type is a D-order in the extended K-algebra B = K ⊗D A (and vice versa, a D-order A in a K-algebra B is a D-algebra of finite type). • We say that a field extension F/K is a splitting field of the K-algebra B if the extended F - algebra B⊗K F is a direct sum of full matrix algebras over F , that is, B⊗K F ∼= Ls i=1 Mni (F ) [30, Def. 18.30]. It is easy to see that if a finite dimensional K-algebra B admits a splitting field F , then B is semisimple, since the extended F -algebra B ⊗K F is semisimple (cf. [30, p. 151]). • If B = Mn(D) is a K-central simple algebra, where D is a division algebra, we denote by deg(B) = p[B : K] the degree of B. If D is a Dedekind domain and P ⊂ D a maximal ideal, then bKP is a splitting field of B if and only if B is unramified at P in the sense of [29, Chap. 8, §32], that is, bBP = B ⊗K bKP ∼= Mnd( bKP ), where d is the degree of D. When D is the ring of integers of a number field, we will demonstrate that IntK-decomposable algebras can be completely classified in terms of these definitions. We first consider the local case where D is a DVR, and then globalize this to the general case. 4.1. Local case In this subsection, D is a DVR with maximal ideal P and finite residue field. As in Section 3, we denote by bD and bA the P -adic completions of D and A (and in general all completions are with respect to the P -adic topology). Theorem 4.2. Let A be a D-algebra of finite type with standard assumptions and let B = A⊗D K be the extended K-algebra. Then A is IntK-decomposable if and only if A is a maximal order in B, B is a separable K-algebra with simple components B1, . . . , Br and there exists a finite unramified extension bF of bK and n ∈ N which satisfy these conditions for each i = 1, . . . , r: 14 (i) Fi ⊗K bK ∼= Qki (ii) Bi ⊗Fi bF = Mn(bF ). j=1 bF for some ki ∈ N. Note that, by [19, Chap. 6, Prop. 6.1] the above condition (i) is equivalent to the following: for each i = 1, . . . , r, all the prime ideals in the integral closure DFi of D in Fi (which necessarily lie above P ) are unramified and have the same residue field degree, equal to [bF : bK] (which is independent of i). In particular, bF is the completion of Fi at each prime ideal of DFi. The second condition says that bF is a splitting field of each simple component Bi, that is, Bi is unramified at each finite place of its center Fi and the degree of each simple component Bi as a Fi-central simple algebra is constant, independent of i. So we can say that P is unramified in each Bi. Proof. (⇐) Assume the above conditions on A, B are satisfied. Since B is semisimple, we have B = rM i=1 Mni(Di) (4.3) for some r, ni ∈ N and division rings Di. We denote by Bi the simple component Mni(Di) of B and by Fi the center of Di, for i = 1, . . . , r. Note that for each i = 1, . . . , r by condition (i) we have Bi ⊗K bK = (Bi ⊗Fi Fi) ⊗K bK = Bi ⊗Fi (Fi ⊗K bK) kiY kiY j=1 bF ) = = Bi ⊗Fi ( j=1 (Bi ⊗Fi bF ) (4.4) is: so by condition (ii) we conclude that Bi ⊗K bK = Qki j=1 Mn(bF ). Hence, the P -adic completion of B kiM Mn(bF ) = (Bi ⊗K bK) = Mn(bF ). bB ∼= rM rM (4.5) i=1 i=1 j=1 tM h=1 Finally, since B is a separable K-algebra and A is a maximal D-order in B, it follows by [29, Thm. 11.5] that bA is a maximal bD-order in bB. Moreover, by [29, Thm. 10.5], A decomposes as A = Lr i=1 Ai, where Ai is a maximal D-order in Bi, for all i = 1, . . . , r. Similarly, bA decomposes as bA = Lt i=1 bAi, where each bAi is a maximal bD-order in the simple component Mn(bF ) of bB. By [29, Thm. 17.3], each bAi is conjugated by a unit of Mn(bF ) to the maximal bD-order Mn(bT ), where bT is the DVR of the local field bF . In particular, each maximal bD-order bAi is isomorphic to Mn(bT ), (⇒) Assume that A is IntK-decomposable. By Theorem 3.6 we have that bA = A ⊗D bD ∼= Lt i=1 Mn(bT ) for some n, t ∈ N and a finite unramified extension bT of bD. Then, Mn(bF ) bB = B ⊗K bK = (A ⊗D K) ⊗D bD = bA ⊗D K ∼= so by Theorem 3.6 A is IntK-decomposable. tM (4.6) h=1 (note that bT ⊗D K = bF ). Therefore, bB is a bK-semisimple algebra with center equal to Z(bB) ∼= Lt i=1 bF . Since an unramified extension of a local field is separable [19, Thm 5.26], bB is a separable 15 bK-algebra. Moreover, each component Mn(bT ) of bA is a maximal bD-order in the respective simple component bBi = Mn(bF ) of bB (see [29, Thm. 8.7]), so, by [29, Thm. 10.5], bA is a maximal bD-order in bB. By [29, Thm. 11.5], A is a maximal D-order in B. Now, by [28, Prop. 10.6b] B is K-separable, hence semisimple. Therefore, B decomposes as a finite direct sum of matrix algebras Bi = Mni(Di) as in (4.3) for some r, ni ∈ N and division rings Di whose centers Fi = Z(Di) are finite separable field extensions of K. j=1 bFij , where bFij is a finite separable extension of bK for all i, j (the bFij are the completions of Fi at the different prime ideals of DFi which lie above P ). Moreover, Bi ⊗Fi bFij is a central simple algebra over bFij , say equal to Mmij (bDij ), where bDij is a central division algebra over bFij [29, Cor. 7.8]. As in (4.4) we get Mmij (bDij ). By [19, Prop. 6.1], for each i = 1, . . . , r, Fi ⊗K bK = Qki (Bi ⊗Fi bFij ) = (Bi ⊗K bK) = bB ∼= kiM rM rM (4.7) i=1 i=1 j=1 rM kiM i=1 j=1 By comparing (4.6) and (4.7) and applying the Wedderburn Structure Theorem, we deduce that mij = n, bDij = bF for all i and j, and Pr i=1 ki = t. This forces bFij = bF for all i and j, so that condition (i) is satisfied. Also, Bi ⊗Fi bFij = Bi ⊗Fi bF = Mn(bF ), so that condition (ii) is satisfied, too. Remark 4.8. Assume A is IntK-decomposable, so B decomposes as in (4.3). For each i = 1, . . . , r, let DFi be the integral closure of D in the center Fi of the simple component Bi of B and consider it as a (commutative) D-algebra. Then the theorem shows (via condition (ii)) that the DFi are IntK-decomposable (according to Theorem 2.10). It also shows that each component Ai of A is IntK-decomposable and also IntFi -decomposable as well. Note that while the degree of the Bi's as Fi-central simple algebras is the same for all i, it is not necessarily true that the dimension of the Bi's over K is the same for all i. The point is that the centers Fi's may be different from each other (and, in particular, have different degree over K). For example, let D = Z(p) where p is an odd prime, let A1 be the standard quaternion algebra A1 = D ⊕ Di ⊕ Dj ⊕ Dk (so that i2 = j2 = −1 and ij = k = −ji). Then, B1 = Q ⊕ Qi ⊕ Qj ⊕ Qk, so n1 = 1 and m1 = 2. Let F/Q be a quadratic field extension in which p splits completely and let DF,p be the integral closure of Z(p) in F (so DF,p/pDF,p ∼= Z/pZ × Z/pZ). Let A2 = M2(OF,p), so B2 = M2(F ), n2 = 2, and m2 = 1. Then A = A1 ⊕ A2 is IntK-decomposable. Note that n1m1 = n2m2 = 2 and B1 and B2 have different dimension over Q: [B1 : Q] = 4 but [B2 : Q] = 8. In the global case where D is the ring of integers of a number field, which will be treated in the next subsection, we will see that if condition (ii) of Theorem 4.2 holds at each maximal ideal of D, then the simple components of the separable K-algebra B have all the same center F , which is an unramified Galois extension of K. Remark 4.9. Let A be an IntK-decomposable algebra, as in the statement of Theorem 4.2. Since the finite unramified extension bF of bK of condition (ii) of the statement is a Galois extension, from (4.5) it is easy to see that bF is a splitting field of the K-algebra B. However, in general the converse does not hold, that is, a finite unramified extension bF of bK can be a splitting field of B without A being IntK-decomposable. For example, let F be a finite field extension of K such that the maximal ideal P of D is unramified in DF and the prime ideals above P have different residue field degree. Then, by Theorem 2.10 the D-algebra A = DF is not IntK-decomposable. Let bF be 16 a finite unramified extension of bK containing all the completions of A at the different prime ideals above P . Then F ⊗K bF = (F ⊗K bK) ⊗ bK bF = Qj bFj ⊗ bK bF = Qj bF (notice that bFj ⊗ bK bF is equal to a direct product of copies of bF ), so bF is a splitting field of B = F . 4.2. Global case We now establish a global variant of Theorem 4.2. In this final subsection, we assume that D is the ring of integers of a number field K. This enables us to use some of the powerful tools of algebraic number theory, such as the Tchebotarev Density Theorem and the Hasse-Brauer-Noether-Albert Theorem. As usual, A is a D-algebra of finite type with standard assumptions. Theorem 4.10. Let K be a number field with ring of integers D. Let A be a D-algebra of finite type with standard assumptions and let B = A ⊗D K be the extended K-algebra. Then, A is IntK- decomposable if and only if A is a maximal order in B and B is a separable K-algebra with simple components B1, . . . , Br such that the following hold: (i) the Bi share a common center F . (ii) F is a finite unramified Galois field extension of K. (iii) for each i, Bi is unramified at each finite place of F . (iv) the degree of Bi as an F -central simple algebra is the same for each i. Proof. (⇐) Assume the above conditions on A and B are satisfied. Let P be a fixed maximal ideal of D. Then, by conditions (ii) and (iii) and [19, Prop. 6.1] we have F ⊗K bK = Qk j=1 bFP , for some k ∈ N, where bFP is a finite unramified extension of bKP ; note that bFP is the completion of F at any prime ideal which lies above P . Therefore, by (4.4) and conditions (i) and (iv) we have bBP = B ⊗K bKP = rM i=1 (Bi ⊗K bKP ) = rM kM i=1 j=1 (Bi ⊗F bFP ) = tM h=1 Mn(bFP ) D, by Theorem 3.6 it follows that A is IntK-decomposable. Since A is a maximal D-order in B, by [29, Cor. 11.2 & Thm. 11.5] bAP is a maximal bDP -order in bBP , so by [29, Thm. 10.5] bAP decomposes as bAP = Lt h=1 bAh, where each bAh is a maximal bDP -order in Mn(bFP ). If bTP is the valuation ring of bFP , then by [29, Thm. 17.3] bAh is a conjugate of (hence isomorphic to) Mn(bTP ), for each h = 1, . . . , t. Since P was an arbitrary maximal ideal of (⇒) Assume now that A is IntK-decomposable. Since IntK-decomposability is a local property (Proposition 2.9), for each maximal ideal P of D, AP = A ⊗D DP is IntK-decomposable, so we may apply Theorem 4.2 to AP . Thus, AP is a maximal DP -order in B, and B is a separable K-algebra and decomposes as B = Lr i=1 Bi with simple components Bi with centers Fi which are finite separable field extensions of K, for i = 1, . . . , r. Moreover, there exists n ∈ N such that [Bi : Fi] = n2, for each i = 1, . . . , r, and hence (iv) holds. Note that r is independent of the particular maximal ideal P of D. Moreover, since A is locally a maximal DP -order in B, A is a maximal D-order in B ([29, Cor. 11.2]). For each prime ideal P of D, by condition (i) of Theorem 4.2 there exists a finite unramified extension bFP of bKP such that, for each i = 1, . . . , r, bFP is the completion of Fi at any prime ideal 17 Q of the ring of integers DFi which lies over P . Furthermore, by condition (ii) of Theorem 4.2 bFP is a splitting field of the Fi-central simple algebra Bi. These facts imply that all the field extensions F1, . . . , Fr are unramified over K and by the Tchebotarev Density Theorem they also are Galois extensions (see [20, Cor. VII.13.8]). Moreover, since a finite Galois extension F of K is completely determined by the set of prime ideals of K which split completely in F (again by the same theorem of Tchebotarev, see [20, Cor. VII.13.10]), it follows that F1, . . . , Fr are all equal to the same finite unramified Galois extension F . All of this proves conditions (i), (ii), and (iii). We close this paper with two corollaries. In the first one we specialize Theorems 2.10 and 2.11 to the case where D and A are rings of integers in number fields, which results in a very clean description of IntK-decomposable algebras. In the second corollary, we show that over Q, an IntQ- decomposable algebra A must be isomorphic to a finite direct sum of copies of Mn(Z), for some n ∈ N. This corollary also demonstrates that-with our usual assumptions in place-a matrix algebra over Z can be recognized by its residues and completions. Corollary 4.11. Let K ⊆ L be number fields with rings of integers OK and OL, respectively. Consider OL as an OK-algebra. Then (1) OL is IntK-decomposable if and only if L/K is an unramified Galois extension. (2) IntK(OL) = Int(OK ) if and only if L = K. Proof. (1) This follows from Theorem 4.10. (2) Clearly, L = K implies that IntK(OL) = Int(OK ). So, assume that IntK(OL) = Int(OK ). By Theorem 2.11, OL/P OL ∼= Li OK/P for each nonzero prime P of OK . From this, we see that OL is IntK-decomposable as an OK-algebra, and moreover that f (QP ) = 1 for each P and each Q above P (note that the same conclusion can be obtained from Theorem 3.10). By (1), L is an unramified Galois extension of K with all inertial degrees equal to 1, and so L = K by [20, Thm. VI.3.8]. Corollary 4.12. Let A be a Z-algebra of finite type with standard assumptions. The following are equivalent. (1) A is IntQ-decomposable i=1 Mn(Z) i=1 Mn(Zp) i=1 Mn(Z/pZ) (2) There exist n, r ∈ N such that A ∼= Lr (3) For all primes p, there exist n, r ∈ N such that A/pA ∼= Lr (4) For all primes p, there exist n, r ∈ N such that bAp ∼= Lr In particular, if A is IntQ-decomposable, then IntQ(A) = IntQ(Mn(Z)) for some n; and if IntQ(A) = Int(Z), then A ∼= Lr Proof. It suffices to prove (1) ⇔ (2). The other equivalences follow by Theorems 2.10 and 3.6. By Theorem 4.10, if A is isomorphic to a finite direct sum of copies of Mn(Z), then A is IntQ- decomposable (the same conclusion follows by the application of either Theorem 2.10 or Theorem 3.6). Conversely, if A is IntQ-decomposable then by the same theorem A is a maximal Z-order in B and B is a separable Q-algebra, say B = Lr i=1 Bi, where the Bi's are the simple components i=1 Z. 18 of B. Since there is no proper unramified extension of Q, the center of each Bi is equal to Q, by condition (ii) of Theorem 4.10. Now, for each i = 1, . . . , r, by the Hasse-Brauer-Noether-Albert Theorem [29, Thm. 32.11] either Bi ∼= Mn(Q) or there are at least two primes of Q (finite or infinite) which ramify in Bi (see [29, Chap 8, §32.1, Exer. 1]). The latter situation cannot occur, because Bi is unramified at each finite place of its center Q by condition (iii) of Theorem 4.10. Since this holds for each i = 1, . . . , r, we must have B = Lr i=1 Mn(Q) (the fact that n is the same for each simple component of B is a consequence of condition (iv) of Theorem 4.10). Finally, since Z is a PID, each maximal Z-order of Mn(Q) is conjugate (hence isomorphic) to Mn(Z), so A ∼= Lr i=1 Mn(Z), as desired. The implication that IntQ(A) = IntQ(Mn(Z)) is clear, and the last claim follows from Proposition 2.8 and Theorem 2.11. The following example shows that the conclusion of Corollary 4.12 may fail if we work over a number field which is a proper extension of Q. Example 4.13. Let K = Q(√5). Consider the standard quaternion algebra B over K, which is B = K ⊕ Ki ⊕ Kj ⊕ Kk, where i2 = j2 = −1 and ij = k = −ji. This algebra is unramified at each finite prime of the ring of integers D of K but is ramified at the two infinite real primes. Hence, any maximal D-order A in B is IntK-decomposable, by Theorem 4.10, but A cannot be isomorphic to a direct sum of matrix rings because B is a division ring. Acknowledgments This work has been supported by the grant "Assegni Senior" of the University of Padova. The authors wish to thank Daniel Smertnig for pointing out the results which led to Corollary 4.12. References [1] M. F. Atiyah, I. G. MacDonald. Introduction to commutative algebra. Addison-Wesley Pub- lishing Co., 1969. [2] N. Bourbaki. Algebra I Chapters 1 – 3. Springer-Verlag, Berlin, 1989. [3] N. Bourbaki. Commutative Algebra. Springer-Verlag, Berlin, 1989. [4] J.-P. Cahen, J.-L. Chabert. Integer-valued Polynomials. Amer. Math. Soc. Surveys and Mono- graphs, Vol. 48, Providence, 1997. [5] J.-L. Chabert, G. Peruginelli. Polynomial overrings of Int(Z). J. Commut. Algebra 8 (2016), no. 1, 1–28. [6] S. Evrard, Y. Fares, K. Johnson. Integer valued polynomials on lower triangular matrices. Monatsh. Math. 170 (2013), no. 2, 147–160. [7] S. Evrard, K. Johnson. The ring of integer valued polynomials on 2× 2 matrices and its integral closure. J. Algebra 441 (2015), 660–677. [8] S. Frisch. Integer-valued polynomials on algebras. J. Algebra 373 (2013), 414–425. [9] S. Frisch. Corrigendum to "Integer-valued polynomials on algebras" [J. Algebra 373 (2013) 414–425]. J. Algebra 412 (2014), 282. 19 [10] S. Frisch. Integer-valued polynomials on algebras: a survey. Actes du CIRM 2 (2) (2010), 27–32. [11] S. Frisch. Polynomial functions on upper triangular matrix algebras, to appear in Monatsh. Math. (2017), DOI:http://dx.doi.org/10.1007/s00605-016-1013-y. [12] K. Goodearl, R. Warfield Jr. An introduction to noncommutative noetherian rings. Lond. Math. Soc. Stud. Texts, vol. 16, Cambridge University Press, Cambridge, 2004. [13] B. Heidaryan, M. Longo, G. Peruginelli. Galois structure on integral valued polynomials. J. Number Theory 171 (2017), 198–212. [14] T. Y. Lam. A first course in noncommutative rings. Second edition. Graduate Texts in Math- ematics, 131. Springer-Verlag, New York, 2001. [15] T. Y. Lam. Lectures on Modules and Rings. Graduate Texts in Mathematics, 189. Springer, New York, 1999. [16] K. A. Loper, N. J. Werner. Generalized rings of integer-valued polynomials. J. Number Theory 132 (2012), no. 11, 2481–2490. [17] B. R. McDonald. Finite Rings with Identity. Pure Appl. Math., vol. 28, Marcel Dekker, New York, 1974. [18] A. R. Naghipour, M. R. Rismanchian, J. Sedighi Hafshejani. Some results on the integer-valued polynomials over matrix rings. Comm. Algebra 45 (2017), no. 4, 1675–1686. [19] W. Narkiewicz. Elementary and analytic theory of algebraic numbers. Third edition. Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2004. [20] J. Neukirch. Algebraic number theory. Springer-Verlag, Berlin, 1999. [21] G. Peruginelli. Integral-valued polynomials over sets of algebraic integers of bounded degree. J. Number Theory 137 (2014), 241–255. [22] G. Peruginelli. Integer-valued polynomials over matrices and divided differences. Monatsh. Math. 173 (2014), no. 4, 559–571. [23] G. Peruginelli. The ring of polynomials integral-valued over a finite set of integral elements. J. Commut. Algebra 8 (2016), no. 1, 113–141. [24] G. Peruginelli, N. J. Werner. Integral closure of rings of integer-valued polynomials on algebras. Commutative Algebra, 293–305, Springer, New York, 2014. [25] G. Peruginelli, N. J. Werner. Properly integral polynomials over the ring of integer-valued polynomials on a matrix ring, J. Algebra 460 (2016) 320–339. [26] G. Peruginelli, N. J. Werner. Non-triviality Conditions for Integer-valued Polynomials on Al- gebras, Monatsh. Math. 183 (2017), no. 1, 177–189. [27] C. Polcino Milies, S. Sudarshan. An introduction to group rings. Algebras and Applications, 1. Kluwer Academic Publishers, Dordrecht, 2002. 20 [28] R. S. Pierce. Associative algebras. Graduate Texts in Mathematics, 88. Springer-Verlag, New York-Berlin, 1982. [29] I. Reiner. Maximal orders. Corrected reprint of the 1975 original. London Mathematical Society Monographs. New Series, 28. The Clarendon Press, Oxford University Press, Oxford, 2003. [30] L. H. Rowen, Graduate algebra: noncommutative view. Graduate Studies in Mathematics, 91. American Mathematical Society, Providence, RI, 2008. [31] N. J. Werner. Int-decomposable algebras. J. Pure Appl. Algebra 218 (2014), no. 10, 1806–1819. [32] N. J. Werner. Integer-valued polynomials over matrix rings. Comm. Algebra 40 (2012), no. 12, 4717–4726. [33] N. J. Werner. Integer-valued polynomials over quaternion rings. J. Algebra 324 (2010), no. 7, 1754–1769. [34] N. J. Werner. Polynomials that kill each element of a finite ring. J. Algebra Appl. 13 (2014), no. 3, 1350111, 12 pp. 21
1903.09812	1	1903	2019-03-23T12:15:43	The quaternion core inverse and its generalizations	[ "math.RA" ]	In this paper we extend notions of the core inverse, core EP inverse, DMP inverse, and CMP inverse over the quaternion skew-field ${\mathbb{H}}$ and get their determinantal representations within the framework of the theory of column-row determinants previously introduced by the author. Since the Moore-Penrose inverse and the Drazin inverse are necessary tools to represent these generalized inverses, we use their determinantal representations previously obtained by using row-column determinants. As the special case, we give their determinantal representations for matrices with complex entries as well. A numerical example to illustrate the main result is given.	math.RA	math	The quaternion core inverse and its generalizations. Ivan I. Kyrchei ∗ Abstract In this paper we extend notions of the core inverse, core EP inverse, DMP inverse, and CMP inverse over the quaternion skew-field H and get their determinantal representations within the framework of the theory of column-row determinants previously introduced by the author. Since the Moore-Penrose inverse and the Drazin inverse are necessary tools to represent these generalized inverses, we use their determinantal represen- tations previously obtained by using row-column determinants. As the special case, we give their determinantal representations for matrices with complex entries as well. A numerical example to illustrate the main result is given. AMS Classification: 15A09; 15A15; 15B33 Keywords: Core inverse; Core EP inverse; DMP inverse; CMP inverse; Moore-Penrose inverse; Drazin inverse; quaternion matrix; noncommutative de- terminant. 1 Introduction Let R and C be the real and complex number fields, respectively. Throughout the paper, we denote the set of all m × n matrices over the quaternion skew field H = {a0 + a1i + a2j + a3k i2 = j2 = k2 = ijk = −1, a1, a2, a3 ∈ R}, its subset of matrices with a rank r. For A ∈ Hm×n, by Hm×n, and by Hm×n the symbols A∗ and rk(A) will denote the conjugate transpose and the rank of A, respectively. A matrix A ∈ Hn×n is Hermitian if A∗ = A. r The Moore-Penrose inverse of A ∈ Hm×n, denoted by A†, is the unique matrix X ∈ Hn×m satisfying the following equations, AXA =A; XAX =X; (AX)∗ =AX; (XA)∗ =XA. (1) (2) (3) (4) ∗Pidstrygach Institute for Applied Problems of Mechanics and Mathematics, NAS of Ukraine, Lviv, Ukraine, E-mail address: [email protected] 1 For A ∈ Hn×n with k = Ind A the smallest positive number such that rk(Ak+1) = rk(Ak) the Drazin inverse of A, denoted by Ad, is defined to be the unique ma- trix X that satisfying Eq. (2) and the following equations, AX =XA; (5) XAk+1 = Ak (6a); or Ak+1X = Ak (6b). In particular, when Ind A = 1, then the matrix X is called the group inverse and is denoted by X = A#. If Ind A = 0, then A is nonsingular, and Ad ≡ A−1. A matrix A satisfying the conditions (i), (j), . . . is called an {i, j, . . .}-inverse of A, and is denoted by A(i,j,...). The set of matrices A(i,j,...) is denoted In particular, A(1) is an inner inverse, A(2) is an outer inverse, A{i, j, . . .}. and A(1,2) is an reflexive inverse, A(1,2,3,4) is the Moore-Penrose inverse, etc. PA := AA† and QA := A†A are the orthogonal projectors onto the range of A and the range of A∗, respectively. For A ∈ Cn×m, the symbols N (A), and R(A) will denote the kernel and the range space of A, respectively. The core inverse was introduced by Baksalary and Trenkler in [2]. Later, it was investigated by S. Malik in [30] and S.Z. Xu et al. in [45], among others. Definition 1.1. [2] A matrix X ∈ Cn×n is called the core inverse of A ∈ Cn×n if it satisfies the conditions AX = PA, and R(X) = R(A). When such matrix X exists, it is denoted A #(cid:13). In 2014, the core inverse was extended to the core-EP inverse defined by K. Manjunatha Prasad and K.S. Mohana [34]. Other generalizations of the core inverse were recently introduced for n×n complex matrices, namely BT inverses [3], DMP inverses [30], and CMP inverses [31], etc. The characterizations, computing methods, some applications of the core inverse and its generalizations were recently investigated in complex matrices and rings (see, e.g., [6, 8 -- 11, 29, 32, 33, 35, 37, 46]). The determinantal representation of the usual inverse is the matrix with cofactors in entries that suggests a direct method of finding the inverse of a matrix. The same is desirable for the generalized inverses. But, there are various expressions of determinantal representations of generalized inverses even for matrices with complex or real entries, (see, e.g. [4, 5, 12 -- 14, 38, 39]). Because of the non-commutativity of the quaternion algebra, difficulties arise already in determining of the quaternion determinant (see, e.g. [1,7]). The prob- lem of the determinantal representation of generalized inverses only now can be solved thanks to the theory of row-column determinants introduced in [15, 16]. Within the framework of the theory of column-row determinants, determinan- tal representations of various generalized quaternion inverses and generalized inverse solutions to quaternion matrix equations have been derived by the au- thor (see, e.g. [17 -- 28]) and by other researchers (see, e.g. [40 -- 44]). In this paper we distribute notions of the core inverse, core EP inverse, DMP inverse, and CMP inverse over the quaternion skew-field H and get their determinantal representations. As the special cases, we give their determinantal representations for complex matrices as well. We note that a determinantal formula for the core EP generalized inverse in complex matrices has been derived 2 in [34] based on the determinantal representation of an reflexive inverse obtained in [4, 5]. In this paper we propose a new determinantal representation of the core EP inverse in complex matrices as well. Due to noncommutativity of quaternions, the ring of quaternion matri- ces M (n, H) has evidently some differences from the ring of complex matrices M (n, C). So, quaternion generalized core inverses could have some features that will be discussed below. The paper is organized as follows. In Section 2, we start with preliminary introduction of row-column determinants, determinantal representations of the Moore-Penrose inverse and the Drazin inverse previously obtained within the framework of the theory of row-column determinants, and some provisions of quaternion vector spaces. In Section 3, we give determinantal representations of the core, core EP, DMP, and CMP inverses over the quaternion skew-field, namely the right and left core inverses are established in Subsection 3.1, the core EP inverses in Subsection 3.2, the core DMP inverse and its dual in Subsection 3.3, and finally the CMP inverse in Subsection 3.4. A numerical example to illustrate the main results is considered in Section 4. Finally, in Section 5, the conclusions are drawn. 2 Preliminaries. Elements of the theory of row- column determinants. Suppose Sn is the symmetric group on the set In = {1, . . . , n}. Let A ∈ Hn×n. Row determinants of A along its each rows can be defined as follows. Definition 2.1. [15] The ith row determinant of A = (aij ) ∈ Hn×n is defined for all i = 1, . . . , n by putting (−1)n−r (ai ik1 aik1 ik1 +1 . . . aik1 +l1 i) . . . (aikr ikr +1 . . . aikr +lr ikr ), rdetiA = Xσ∈Sn σ = (i ik1ik1+1 . . . ik1+l1) (ik2 ik2+1 . . . ik2+l2 ) . . . (ikr ikr +1 . . . ikr+lr ) , where σ is the left-ordered permutation. It means that its first cycle from the left starts with i, other cycles start from the left with the minimal of all the integers which are contained in it, ikt < ikt+s for all t = 2, . . . , r, s = 1, . . . , lt, and the order of disjoint cycles (except for the first one) is strictly conditioned by increase from left to right of their first elements, ik2 < ik3 < · · · < ikr . Similarly, for a column determinant along an arbitrary column, we have the following definition. Definition 2.2. [15] The jth column determinant of A = (aij) ∈ Hn×n is defined for all j = 1, . . . , n by putting cdetj A = Xτ ∈Sn (−1)n−r(ajkr jkr +lr . . . ajkr +1jkr ) . . . (ajjk1 +l1 . . . ajk1 +1jk1 ajk1 j), τ = (jkr+lr . . . jkr +1jkr ) . . . (jk2+l2 . . . jk2+1jk2 ) (jk1+l1 . . . jk1+1jk1 j) , 3 where τ is the right-ordered permutation. It means that its first cycle from the right starts with j, other cycles start from the right with the minimal of all the integers which are contained in it, jkt < jkt+s for all t = 2, . . . , r, s = 1, . . . , lt, and the order of disjoint cycles (except for the first one) is strictly conditioned by increase from right to left of their first elements, jk2 < jk3 < · · · < jkr . The row and column determinants have the following linear properties. Lemma 2.1. [15] If the ith row of A ∈ Hn×n is a left linear combination of other row vectors, i.e. ai. = α1b1 + · · · + αkbk, where αl ∈ H and bl ∈ H1×n for all l = 1, . . . , k and i = 1, . . . , n, then rdeti Ai. (α1b1 + · · · + αkbk) =Xl αlrdeti Ai. (bl) . Lemma 2.2. [15] If the jth column of A ∈ Hm×n is a right linear combination of other column vectors, i.e. a.j = c1α1 + · · · + ckαk, where αl ∈ H and cl ∈ Hn×1 for all l = 1, . . . , k and j = 1, . . . , n, then cdetj A.j (c1α1 + · · · + ckαk) =Xl cdetj A.j (cl) αl. So, an arbitrary n × n quaternion matrix inducts a set from n row deter- minants and n column determinants that are different in general. Only for Hermitian A, we have [15], rdet1A = · · · = rdetnA = cdet1A = · · · = cdetnA ∈ R, that enables to define the determinant of a Hermitian matrix by putting det A := rdeti A = cdeti A for all i = 1, . . . , n. Properties of the determinant of a Hermitian matrix are similar to the prop- erties of an usual (commutative) determinant and they have been completely explored by row-column determinants in [16]. In particular, for any matrix A ∈ Hm×n it is proved [16] that det(A∗A) = 0 iff some column of A is a right linear combination of others, or some row of A∗ is a left linear combination of others. From this follows the definition of the determinantal rank of a quater- nion matrix A as the largest possible size of a nonzero principal minor of its corresponding Hermitian matrix A∗A. It is shown that the row rank of a quater- nion matrix A ∈ Hm×n (that is a number of its left-linearly independent rows), the column rank (that is a number of its right-linearly independent columns) and its determinantal rank are equivalent herewith rank (A∗A) = rank (AA∗). For introducing determinantal representations of generalized inverses, the following notations will be used. Let α := {α1, . . . , αk} ⊆ {1, . . . , m} and β := {β1, . . . , βk} ⊆ {1, . . . , n} be subsets of the order 1 ≤ k ≤ min {m, n}. Aα β denotes a submatrix of A whose rows are indexed by α and the columns indexed by β. So, Aα α denotes a principal submatrix of A with rows and columns indexed by α. If A is Hermitian, then Aα α denotes the corresponding principal minor of det A. Suppose Lk,n := { α : α = (α1, . . . , αk) , 1 ≤ α1 < · · · < αk ≤ n} denotes the collection of strictly increasing sequences of 1 ≤ k ≤ n integers chosen from 4 {1, . . . , n}. Then, for fixed i ∈ α and j ∈ β, the collection of sequences of row indexes that contain the index i is denoted by Ir, m{i} := { α : α ∈ Lr,m, i ∈ α}, similarly, the collection of sequences of column indexes that contain the index j is denote by Jr, n{j} := { β : β ∈ Lr,n, j ∈ β}. Let a.j be the jth column and ai. be the ith row of A. Suppose A.j (b) denotes the matrix obtained from A by replacing its jth column with the column b, and Ai. (b) denotes the matrix obtained from A by replacing its ith row with the row b. Denote by a∗ i. the jth column and the ith row of A∗, respectively. .j and a∗ Theorem 2.3. [17] If A ∈ Hm×n ij(cid:17) ∈ Hn×m have the following determinantal representations, (cid:16)a† r , then the Moore-Penrose inverse A† = ij = Pβ∈Jr,n{i} a† ij = Pα∈Ir,m{j} a† .j(cid:1)(cid:1)β cdeti(cid:0)(A∗A).i(cid:0)a∗ Pβ∈Jr,n A∗Aβ β β rdetj ((AA∗)j.(a∗ i.))α α AA∗α α Pα∈Ir,m , . (6) (7) and Remark 2.4. For an arbitrary full-rank matrix A ∈ Hm×n H1×m, and a column-vector c ∈ Hn×1 we put, respectively, r , a row-vector b ∈ • if r = m, then rdeti ((AA∗)i. (b)) = Xα∈Im,m{i} det (AA∗) = Xα∈Im,m AA∗α α rdeti ((AA∗)i. (b))α α, for all i = 1, . . . , m; • if r = n, then cdetj(cid:16)(A∗A).j (c)(cid:17) = Xβ∈Jn,n{j} det (A∗A) = Xβ∈Jn,n A∗Aβ β cdetj(cid:16)(A∗A).j (c)(cid:17)β β , for all j = 1, . . . , n. Corollary 2.1. If A ∈ Hm×n inverse A† =(cid:16)a† r is Hermitian, then the then the Moore-Penrose ij(cid:17) ∈ Hn×m have the following determinantal representations, ij = Pβ∈Jr,n{i} a† cdeti(cid:0)(cid:0)A2(cid:1).i (a .j)(cid:1)β Pβ∈Jr,n A2β β β 5 , (8) and ij = Pα∈Ir,m{j} a† rdetj(cid:0)(A2)j.(a i.)(cid:1)α Pα∈Ir,m A2α α α . (9) Corollary 2.2. If A ∈ Hm×n (qij )n×n has the determinantal representation r , then the projection matrix A†A =: QA = qij = Pβ∈Jr,n{i} cdeti ((A∗A).i ( a.j))β β A∗Aβ β , (10) Pβ∈Jr,n Pα∈Ir,m where a.j is the jth column of A∗A ∈ Hn×n. Corollary 2.3. If A ∈ Hm×n (pij )m×m has the determinantal representation r , then the projection matrix AA† =: PA = pij = Pα∈Ir,m{j} rdetj((AA∗)j.(ai.))α α AA∗α α , (11) where ai. is the ith row of AA∗ ∈ Hm×m. The following corollary gives determinantal representations of the Moore- Penrose inverse and of both projectors in complex matrices. . Then the following determinantal repre- Corollary 2.4. [12] Let A ∈ Cm×n sentations are obtained r (i) for the Moore-Penrose inverse A† =(cid:16)a† ij = Pβ∈Jr,n{i}(cid:12)(cid:12)(A∗A).i(cid:0)a∗ .j(cid:1)(cid:12)(cid:12)β a† β A∗Aβ β Pβ∈Jr,n , ij(cid:17)n×m = Pα∈Ir,m{j} Pα∈Ir,m (AA∗)j.(a∗ i.)α α AA∗α α ; (ii) for the projector QA = (qij)n×n, qij = Pβ∈Jr,n{i} Pβ∈Jr,n (A∗A).i ( a.j)β β A∗Aβ β where a.j is the jth column of A∗A; (iii) for the projector PA = (pij )m×m, pij = Pα∈Ir,m{j} Pα∈Ir,m (AA∗)j.(ai.)α α AA∗α α where ai. is the ith row of AA∗. 6 , , Lemma 2.3. [18] If A ∈ Hn×n with Ind A = k and then the Drazin inverse Ad =(cid:0)ad (i) if A is an arbitrary, then a(k) ad ij = nPt=1 it Pβ∈Jr, n{t} ij(cid:1) ∈ Hn×n possess the determinantal representations cdett(cid:16)(cid:16)(cid:0)A2k+1(cid:1)∗ A2k+1(cid:17). t A2k+1(cid:12)(cid:12)β Pβ∈Jr,n(cid:12)(cid:12)(A2k+1)∗ rdets(cid:16)(cid:16)A2k+1(cid:0)A2k+1(cid:1)∗(cid:17) .s Pα∈Ir, n(cid:12)(cid:12)A2k+1 (A2k+1)∗(cid:12)(cid:12)α nPs=1 Pα∈Ir, n{s} ad ij = α β and (a. j)(cid:17)β β α! a(k) sj (ai .)(cid:17) α where rk Ak+1 = rk Ak = r, a.j is the jth column of (A2k+1)∗Ak =: A = (aij) ∈ Hn×n, and ai. the tth row of Ak(A2k+1)∗ =: A = (aij ) ∈ Hn×n; (12) (13) (ii) if A is Hermitian, then ij = Pβ∈Jr,n{i} ad ij = Pα∈Ir,n{j} ad or β Ak+1β β .j (cid:17)(cid:17)β cdeti(cid:16)(cid:0)Ak+1(cid:1).i(cid:16)a(k) Pβ∈Jr,n i. (cid:17)(cid:17)α rdetj(cid:16)(Ak+1)j.(cid:16)a(k) Pα∈Ir,n Ak+1α α α , , (14) (15) where rk Ak+1 = rk Ak = r, a(k) row of Ak, respectively. .j and a(k) i. are the jth column and the ith Corollary 2.5. If A ∈ Hn×n with Ind A = 1, then the group inverse A# possess the determinantal representations (i) if A is an arbitrary, then nPt=1 a# ij = ait Pβ∈Jr,n{t} and nPs=1 Pα∈Ir,n{s} a# ij = A3(cid:17). t (a.j)(cid:17)β β α! asj (ai.)(cid:17)α β cdett(cid:16)(cid:16)(cid:0)A3(cid:1)∗ Pβ∈Jr,n(cid:12)(cid:12)(A3)∗ A3(cid:12)(cid:12)β rdets(cid:16)(cid:16)A3(cid:0)A3(cid:1)∗(cid:17) .s Pα∈Ir, n(cid:12)(cid:12)A3 (A3)∗(cid:12)(cid:12)α α (16) (17) where rk A2 = rk A = r, a.j is the jth column of (A3)∗A =: A = (aij) ∈ Hn×n and ai. the ith row of A(A3)∗ =: A = (aij) ∈ Hn×n; 7 Pβ∈Jr,n Pα∈Ir,n (ii) if A is Hermitian, then ij = Pβ∈Jr,n{i} a# ij = Pα∈Ir,n{j} a# or where rk A2 = rk A = r. β A2β β cdeti(cid:0)(cid:0)A2(cid:1).i (a.j)(cid:1)β Pβ∈Jr,n rdetj(cid:0)(A2)j. (ai.)(cid:1)α Pα∈Ir,n A2α α α , . (18) (19) The following corollary gives determinantal representations of the Drazin inverse in complex matrices. representations Corollary 2.6. [12] Let A ∈ Cn×n with Ind A = k and rk Ak+1 = rk Ak = r. ij(cid:1) ∈ Cn×n has the following determinantal Then the Drazin inverse Ad = (cid:0)ad i. (cid:17)(cid:12)(cid:12)(cid:12) = Pα∈Ir,n{j}(cid:12)(cid:12)(cid:12)(Ak+1)j.(cid:16)a(k) .j (cid:17)(cid:12)(cid:12)(cid:12) ij = Pβ∈Jr,n{i}(cid:12)(cid:12)(cid:12)(cid:0)Ak+1(cid:1).i(cid:16)a(k) Ak+1α α Ak+1β β (20) ad β β α α , where a(k) .j and a(k) i. are the jth column and the ith row of Ak, respectively. It is clear that the determinantal representations of the group inverse in complex matrices have been obtained from (20) by putting k = 1. Due to quaternion-scalar multiplying on the right, quaternion column-vectors form a right vector H-space, and, by quaternion-scalar multiplying on the left, quaternion row-vectors form a left vector H-space denoted by Hr and Hl, respec- tively. It can be shown that Hr and Hl possess corresponding H-valued inner products by putting hx, yir = y1x1 + · · · + ynxn for x = (xi)n i=1 ∈ Hr, and hx, yil = x1y1 + · · · + xnyn for x, y ∈ Hl that satisfy the inner product relations, namely, conjugate symmetry, linearity, and positive-definiteness but with specialties i=1 , y = (yi)n hxα + yβ, zi = hx, ziα + hy, ziβ when x, y, z ∈ Hr, hαx + βy, zi = αhx, zi + βhy, zi when x, y, z ∈ Hl, for any α, β ∈ H. So, an arbitrary quaternion matrix induct vector spaces that introduced by the following definition. Definition 2.5. For an arbitrary matrix over the quaternion skew field, A ∈ Hm×n, we denote by • Rr(A) = {y ∈ Hm×1 : y = Ax, x ∈ Hn×1}, the right column space of A, • Nr(A) = {x ∈ Hn×1 : Ax = 0}, the right null space of A, • Rl(A) = {y ∈ H1×n : y = xA, x ∈ H1×m}, the left row space of A, • Nl(A) = {x ∈ H1×m : xA = 0}, the left null space of A. 8 3 Determinantal representations of the core in- verse and its generalizations 3.1 Determinantal representations of the core inverses Due to quaternion noncommutativity, Definition 1.1 of the core inverse can be expand to matrices over H as follows. Definition 3.1. A matrix X ∈ Hn×n is called the right core inverse of A ∈ Hn×n if it satisfies the conditions AX = PA, and Rr(X) = Rr(A). When such matrix X exists, it is denoted A #(cid:13). Definition 3.2. A matrix X ∈ Hn×n is called the left core inverse of A ∈ Hn×n if it satisfies the conditions XA = QA, and Rl(X) = Rl(A). When such matrix X exists, it is denoted A #(cid:13). Remark 3.3. A definition similar to Definition 3.2 has been introduced for A ∈ = A†A = QA, and Rl(A) = Cn×n in [36] given that PA∗ = A∗(A∗)† =(cid:0)A†A(cid:1)∗ Rr(A∗). In [36], A #(cid:13) is called the dual core inverse of A. Due to [2], we introduce the following sets of quaternion matrices HCM n ={A ∈ Hn×n : rk A2 = rk A}, HEP n ={A ∈ Hn×n : A†A = AA†} = {Rr(A) = Rr(A∗)}. n are called group matrices or core matrices. If A ∈ HEP The matrices from HCM n , then clearly A† = A#. Similarly as for complex matrices, the core inverses of a square quaternion matrix A ∈ Hn×n exist if and only if A ∈ HCM or Ind A = 1. Moreover, if A is non-singular, Ind A = 0, then its core inverses are the usual inverse. n Due to [2], the following representations of right and left core inverses can be extended to quaternion matrices. Lemma 3.1. Let A ∈ HCM n . Then A #(cid:13) =A#AA†, A #(cid:13) =A†AA# (21) (22) Lemma 3.2. Let A ∈ HCM n . Then (i) A #(cid:13), A #(cid:13) ∈ HEP n , (ii) (A #(cid:13))† = APA, (A #(cid:13))† = QAA, (iii) (A #(cid:13))# = APA, (A #(cid:13))# = QAA, (iv) A #(cid:13), A #(cid:13) ∈ A{1, 2} 9 (v) (A #(cid:13))2 A = A#, A (A #(cid:13))2 = A#, (vi) (A #(cid:13))m = (Am) #(cid:13) , A (A #(cid:13))m = A (Am) #(cid:13) , (vii) (A #(cid:13)) A = A#A, A (A #(cid:13)) = AA#. n . Since A ∈ HCM Remark 3.4. In Theorems 3.5 and 3.6, we will suppose that A ∈ HCM but A /∈ HEP n (in particular, A is Hermitian), then it follows from Lemma 3.1 and the definitions of the Moore-Penrose inverse and group inverse that A #(cid:13) = A #(cid:13) = A# = A†. and A ∈ HEP n n Theorem 3.5. Let A ∈ HCM A #(cid:13) =(cid:0)a #(cid:13),r ij (cid:1) has the following determinantal representation rdetj((AA∗)j.(ui.))α α n , rk A2 = rk A = s. Then its right core inverse a #(cid:13),r ij = Pα∈Is,n{j} Pα∈Is,n(cid:12)(cid:12)(cid:0)A3 (A3)∗(cid:1)α α(cid:12)(cid:12) Pα∈Is,n rdetf(cid:18)(cid:16)A3(cid:0)A3(cid:1)∗(cid:17).f uif = Xα∈Is,n{f } where eui. is the ith row of eU := UA2A∗, and U = (uif ) ∈ Hn×n such that (ai .)(cid:19)α α , , (23) AA∗α α where ai . is the ith row of A = A(A3)∗. Proof. By (21), a #(cid:13),r ij = a# il plj. nXl=1 Using (16) for the determinantal representation of A# and (11) for the deter- minantal representation of PA = AA†, we obtain (ai .)(cid:19)α α! af l × a #(cid:13),r ij = nXl=1 nPf =1 Pα∈Is,n{f } α rdetj((AA∗)j.(al.))α rdetf(cid:18)(cid:16)A3(cid:0)A3(cid:1)∗(cid:17).f Pα∈Is,n(cid:12)(cid:12)A3 (A3)∗(cid:12)(cid:12)α nPf =1 Pα∈Is,n{f } Pα∈Is,n{j} Pα∈Is,n rdetf(cid:18)(cid:16)A3(cid:0)A3(cid:1)∗(cid:17).f (ai .)(cid:19)α α Pα∈Is,n{j} α(cid:12)(cid:12) Pα∈Is,n Pα∈Is,n(cid:12)(cid:12)(cid:0)A3 (A3)∗(cid:1)α AA∗α α = α AA∗α α Denote by uif := Xα∈Is,n{f } rdetf(cid:18)(cid:16)A3(cid:0)A3(cid:1)∗(cid:17).f (ai .)(cid:19)α α . 10 rdetj((AA∗)j.(af.))α α , where ai . is the ith row of A = A(A3)∗ and af. is the f th row of A := A2A∗. Construct the matrix U = (uif ) ∈ Hn×n and denote U = U A = UA2A∗. It follows that Xf uif Xα∈Is,n{j} rdetj(cid:16)(AA∗)j. (af.)(cid:17)α α = Xα∈Is,n{j} rdetj(cid:16)(AA∗)j. (ui.)(cid:17)α α , where ui. is the ith row of U. Thus we have (23). Taking into account (22), the following theorem on determinantal represen- tations of the left core inverse can be proved similarly. Theorem 3.6. Let A ∈ HCM n , rk A2 = rk A = s. Then its left core inverse where v.j is the jth column of V := A∗A2V, and V = (vf j ) ∈ Hn×n such that A #(cid:13) =(cid:16)a #(cid:13),l a #(cid:13),l ij = cdeti((A∗A).i(v.j ))β β ij (cid:17) has the following determinantal representation β Pβ∈Js,n(cid:12)(cid:12)(A3)∗ A3(cid:12)(cid:12)α (a.j)(cid:19)β Pβ∈Js,n{i} Pβ∈Js,n A∗Aβ A3(cid:17).f α β , , vf j = Xβ∈Js,n{f } cdetf(cid:18)(cid:16)(cid:0)A3(cid:1)∗ where ba.j is the jth column of bA := (A3)∗A. core inverses for complex matrices. (24) The next corollary gives determinantal representations of the right and left Corollary 3.1. Let A ∈ CCM inverse has the determinantal representations n and rk A2 = rk A = s. Then its right core where , A2β β a #(cid:13),r β = A2β β AA∗α (AA∗)j.(vi.)α α α Pβ∈Js,n ij = Pα∈Is,n{j} Pα∈Is,n vi. = Xβ∈Js,n{i}(cid:12)(cid:12)(cid:0)A2(cid:1).i (a.f )(cid:12)(cid:12)β v.j = Xα∈Is,n{j} AA∗α Pβ∈Js,n{i}(cid:12)(cid:12)(cid:0)A2(cid:1). i (v.j )(cid:12)(cid:12)β Pα∈Is,n α Pβ∈Js,n β ∈ H1×n, f = 1, . . . , n α ∈ Hn×1, l = 1, . . . , n, (AA∗)j.(al.)α are the row-vector and the column-vector, respectively; a.f and al. are the f th column and lth row of A := A2A∗. And, its left core inverse has the determinantal representations a #(cid:13),l ij = Pα∈Is,n{j}(cid:12)(cid:12)(A2)j.(ui.)(cid:12)(cid:12)α Pα∈Is,n α Pβ∈Js,n A2α α A∗Aβ β 11 (A∗A). i (u.j)β β = Pβ∈Js,n{i} Pα∈Is,n A2α α Pβ∈Js,n , A∗Aβ β where (A∗A).i (¯a.f )β ui. = Xβ∈Js,n{i} u.j = Xα∈Is,n{j}(cid:12)(cid:12)(A2)j.(¯al.)(cid:12)(cid:12)α β ∈ H1×n, f = 1, . . . , n α ∈ Hn×1, l = 1, . . . , n, are the row-vector and the column-vector, respectively; ¯a.f and ¯al. are the f th column and lth row of ¯A := A∗A2. 3.2 Determinantal representations of the core EP inverses Similar as in [34], we introduce two core EP inverses. Definition 3.7. A matrix X ∈ Hn×n is called the right core EP inverse of A ∈ Hn×n if it satisfies the conditions XAX = A, and Rr(X) = Rr(X∗) = Rr(Ad). It is denoted A †(cid:13). Definition 3.8. A matrix X ∈ Hn×n is called the left core EP inverse of A ∈ Hn×n if it satisfies the conditions XAX = A, and Rl(X) = Rl(X∗) = Rl((A)d). It is denoted A †(cid:13). Remark 3.9. Since Rr((A∗)d) = Rl((A)d), then the left core inverse A †(cid:13) of A ∈ Cn×n is similar to the left ∗core inverse introduced in [34], and the dual core EP inverse introduced in [36]. Due to [34], we have the following representations the core EP inverses of A ∈ Hn×n , A †(cid:13) =A{2,3,6a} and Rr(A †(cid:13)) ⊆ Rr(Ak), A †(cid:13) =A{2,4,6b} and Rl(A †(cid:13)) ⊆ Rl(Ak). Thanks to [36], the following representations of the core EP inverses will be used for their determinantal representations. Lemma 3.3. Let A ∈ Hn×n and IndA = k. Then A †(cid:13) =Ak(Ak+1)†, A †(cid:13) =(Ak+1)†Ak. (25) (26) Moreover, if Ind A = 1, then we have the following representations of the right and left core inverses A #(cid:13) =A(A2)†, A #(cid:13) =(A2)†A. 12 (27) (28) Theorem 3.10. Suppose A ∈ Hn×n A †(cid:13). Then A †(cid:13) = (cid:0)a †(cid:13),r representations, respectively, ij (cid:1) and A †(cid:13) = (cid:16)a †(cid:13),l , Ind A = k, and there exist A †(cid:13) and ij (cid:17) have the following determinantal s (29) (30) where ai . is the ith row of A = Ak(Ak+1)∗ and a.j is the jth column of A = (Ak+1)∗Ak. , , β β α α a †(cid:13),l a †(cid:13),r (a.j)(cid:17)β (ai .)(cid:19)α ij = Pα∈Is,n{j} ij = Pβ∈Js,n{i} rdetj(cid:18)(cid:16)Ak+1(cid:0)Ak+1(cid:1)∗(cid:17)j. Pα∈Is,n(cid:12)(cid:12)Ak+1 (Ak+1)∗(cid:12)(cid:12)α Ak+1(cid:17).i cdeti(cid:16)(cid:16)(cid:0)Ak+1(cid:1)∗ Ak+1(cid:12)(cid:12)β Pβ∈Js,n(cid:12)(cid:12)(Ak+1)∗ ij (cid:17). By (25), (cid:17)† and Ak =(cid:16)a(k) it (cid:16)a(k+1) (cid:17)† nXt=1 Using (7) for the determinantal representation(cid:0)Ak+1(cid:1)† rdetj(cid:16)(cid:0)Ak+1(Ak+1)∗(cid:1)j. (a(k+1,∗) Pα∈Ir,m(cid:12)(cid:12)Ak+1 (Ak+1)∗(cid:12)(cid:12)α Proof. Let(cid:0)Ak+1(cid:1)† it Pα∈Is,n{j} =(cid:16)a(k+1) nXt=1 a †(cid:13),r ij = a †(cid:13),r ij = a(k) a(k) tj ij α i. . is the ith row of (Ak+1)∗. Since , we obtain where a(k+1,∗) i. we have (29). )(cid:17)α α , it a(k+1,∗) a(k) i. ) = ai ., finally, nPt=1 The determinantal representation (30) is obtained similarly by using (6) for the determinantal representation(cid:0)Ak+1(cid:1)† in (26). Taking into account the representations (27)-(28), we evidently obtain de- terminantal representations of the right and left core inverses which have more simpler expressions than (23)-(24). Corollary 3.2. Let A ∈ Hn×n , IndA = 1, and there exist A #(cid:13) and A #(cid:13). Then A #(cid:13) = (cid:0)a #(cid:13),r tions, respectively, s ij (cid:1) and A #(cid:13) = (cid:16)a #(cid:13),l ij = Pα∈Is,n{j} ij = Pβ∈Js,n{i} a #(cid:13),r a #(cid:13),l (ai .)(cid:19)α ij (cid:17) have the following determinantal representa- rdetj(cid:18)(cid:16)A2(cid:0)A2(cid:1)∗(cid:17)j. Pα∈Is,n(cid:12)(cid:12)A2 (A2)∗(cid:12)(cid:12)α cdeti(cid:16)(cid:16)(cid:0)A2(cid:1)∗ A2(cid:17).i Pβ∈Js,n(cid:12)(cid:12)(A2)∗ A2(cid:12)(cid:12)β (a.j )(cid:17)β (31) (32) α α β β , , 13 where ai . is the ith row of A = A(A2)∗ and a.j is the jth column of A = (A2)∗A. The following corollary gives determinantal representations of the right and left core EP inverses and the right and left core inverses for complex matrices. Corollary 3.3. Suppose A ∈ Cn×n s and A †(cid:13) =(cid:16)a †(cid:13),l respectively, α β α β β , α , a †(cid:13),l a †(cid:13),r , IndA = k, and there exist A †(cid:13) =(cid:0)a †(cid:13),r ij (cid:1) ij (cid:17). Then they have the following determinantal representations, (ai .)(cid:12)(cid:12)(cid:12)(cid:12) ij = Pα∈Is,n{j}(cid:12)(cid:12)(cid:12)(cid:12)(cid:16)Ak+1(cid:0)Ak+1(cid:1)∗(cid:17)j. Pα∈Is,n(cid:12)(cid:12)Ak+1 (Ak+1)∗(cid:12)(cid:12)α ij = Pβ∈Js,n{i}(cid:12)(cid:12)(cid:12)(cid:16)(cid:0)Ak+1(cid:1)∗ (a.j)(cid:12)(cid:12)(cid:12) Ak+1(cid:17).i Ak+1(cid:12)(cid:12)β Pβ∈Js,n(cid:12)(cid:12)(Ak+1)∗ ij (cid:17) have the following ij (cid:1) and A #(cid:13) = (cid:16)a #(cid:13),l (ai .)(cid:12)(cid:12)(cid:12)(cid:12) ij = Pα∈Is,n{j}(cid:12)(cid:12)(cid:12)(cid:12)(cid:16)A2(cid:0)A2(cid:1)∗(cid:17)j. Pα∈Is,n(cid:12)(cid:12)A2 (A2)∗(cid:12)(cid:12)α ij = Pβ∈Js,n{i}(cid:12)(cid:12)(cid:12)(cid:16)(cid:0)A2(cid:1)∗ (a.j )(cid:12)(cid:12)(cid:12) A2(cid:17).i Pβ∈Js,n(cid:12)(cid:12)(A2)∗ A2(cid:12)(cid:12)β a #(cid:13),r a #(cid:13),l α , β β , α β α where ai . is the ith row of A = Ak(Ak+1)∗ and a.j is the jth column of A = (Ak+1)∗Ak. If IndA = 1, then A #(cid:13) = (cid:0)a #(cid:13),r determinantal representations, respectively, where ai . is the ith row of A = A(A2)∗ and a.j is the jth column of A = (A2)∗A. 3.3 Determinantal representations of the core DMP and MPD inverses The concept of the DMP inverse in complex matrices was introduced in [30] by S. Malik and N. Thome that can be expended to quaternion matrices as follows. Definition 3.11. Suppose A ∈ Hn×n and Ind A = k. A matrix X ∈ Hn×n is called the DMP inverse of A if it satisfies the conditions XAX = X, XA = AdA, and AkX = AkA†. (33) It is denoted Ad,†. 14 It is proven [30] that the matrix satisfying system of equations (33) is unique and it has the following representation Ad,† = AdAA†. (34) In accordance with the order of use the Drazin inverse (D) and the Moore- Penrose (MP) inverse, its name is the DMP inverse. Theorem 3.12. Let A ∈ Hn×n inverse Ad,† =(cid:16)ad,† ij (cid:17) has the following determinantal representations. s , Ind A = k, and rk(Ak) = s1. Then its DMP (i) If A is an arbitrary matrix, then ad,† ij = rdetj((AA∗)j.(ui.))α α Pα∈Is,n{j} Pα∈Is1 ,n(cid:12)(cid:12)A2k+1 (A2k+1)∗(cid:12)(cid:12)α α Pα∈Is,n , AA∗α α (35) that where eui. is the ith row of eU := UA2A∗, and U = (uif ) ∈ Hn×n such uif = Xα∈Is1 ,n{f } rdetf(cid:18)(cid:16)A2k+1(cid:0)A2k+1(cid:1)∗(cid:17).f (ai .)(cid:19)α α , where ai . is the ith row of A = A(A2k+1)∗. (ii) If A is Hermitian, then ad,† ij = Pα∈Is,n{j} Pβ∈Js1 ,n = Pβ∈Js1,n{i} Pβ∈Js1,n (36) (37) Ak+1β α = rdetj(cid:0)(A2)j.(vi.)(cid:1)α cdeti(cid:0)(cid:0)Ak+1(cid:1). i (u.j)(cid:1)β A2α α β Ak+1β , A2β β β Pα∈Is,n β Pβ∈Js,n β ∈ H1×n, f = 1, . . . , n (cid:17)(cid:17)β α ∈ Hn×1, )(cid:17)α l = 1, . . . , n, (k+2) .f where vi. = Xβ∈Js1 ,n{i} u.j = Xα∈Is,n{j} cdeti(cid:16)(cid:0)Ak+1(cid:1).i(cid:16)a rdetj(cid:16)(A2)j.(a(k+2) l. are the row-vector and the column-vector, respectively. Proof. By (34), ad,† ij = ad ilplj. nXl=1 15 (i) Let A ∈ Hn×n s be an arbitrary matrix. Then, using the determinantal representations (13) and (11) for respectively Ad and PA = AA†, we obtain ad,† ij = nXl=1 nPf =1 Pα∈Is1 ,n{f } af l × α α (ai .)(cid:19)α rdetj((AA∗)j.(al.))α rdetf(cid:18)(cid:16)A2k+1(cid:0)A2k+1(cid:1)∗(cid:17).f nPf =1 Pα∈Is1 ,n{f } Pα∈Is1 ,n(cid:12)(cid:12)A2k+1 (A2k+1)∗(cid:12)(cid:12)α Pα∈Is,n{j} Pα∈Is,n (ai .)(cid:19)α rdetf(cid:18)(cid:16)A2k+1(cid:0)A2k+1(cid:1)∗(cid:17).f α Pα∈Is,n{j} Pα∈Is1 ,n(cid:12)(cid:12)A2k+1 (A2k+1)∗(cid:12)(cid:12)α α Pα∈Is,n AA∗α α AA∗α α = α rdetj((AA∗)j.(af.))α α , where ai . is the ith row of A = A(A2k+1)∗ andeaf. is the f th row of eA := A2A∗. Denote by uif := Xα∈Is1 ,n{f } rdetf(cid:18)(cid:16)A2k+1(cid:0)A2k+1(cid:1)∗(cid:17).f (ai .)(cid:19)α α for all i, f = 1, . . . , n. Now, we construct the matrix U = (uif ) ∈ Hn×n, and denote U := U A = UA2A∗. It follows that uif Xα∈Is,n{j} rdetj(cid:16)(AA∗)j. (eaf.)(cid:17)α = Xα∈Is,n{j} Xf where eui. is the ith row of U. Thus we have (35). (ii) Let, now, A ∈ Hn×n α be Hermitian. Then using (14 ) for the determi- nantal representation of Ad and (9) for the determinantal representation of A†, we obtain s rdetj(cid:16)(AA∗)j. (eui.)(cid:17)α α , ad,† ij = nXl=1 nXl=1 nXf =1 nXf =1 Pβ∈Js1 ,n{i} Pβ∈Js1 ,n{i} (k) .l (cid:17)(cid:17)β β Ak+1β β cdeti(cid:16)(cid:0)Ak+1(cid:1). i(cid:16)a Pβ∈Js1,n cdeti(cid:0)(cid:0)Ak+1(cid:1). i (e.l)(cid:1)β Pβ∈Js1,n Ak+1β β β alf Pα∈Is,n{j} Pα∈Is,n{j} a(k+2) lf α A2α α rdetj(cid:0)(A2)j.(af.)(cid:1)α Pα∈Is,n rdetj(cid:0)(A2)j.(ef.)(cid:1)α Pα∈Is,n A2α α α = , where e.l and el. are the unit column-vector and the unit row-vector, respec- tively, such that all their components are 0, except the lth components which are 1; a(k+2) is the (lf )th element of the matrix Ak+2. lf If we denote by 16 vif := nXl=1 Xβ∈Js1,n{i} Xβ∈Js1,n{i} cdeti(cid:0)(cid:0)Ak+1(cid:1).i (e.l)(cid:1)β (cid:17)(cid:17)β cdeti(cid:16)(cid:0)Ak+1(cid:1).i(cid:16)a(k+2) .f β β a(k+2) lf = the f th component of a row-vector vi. = [vi1, . . . , vin], then nXf =1 vif Xα∈Is,n{j} rdetj(cid:0)(A2)j.(ef.)(cid:1)α α = Xα∈Is,n{j} rdetj(cid:0)(A2)j.(vi.)(cid:1)α α. So, we have (36). If we denote by ulj := rdetj(cid:0)(A2)j.(ef.)(cid:1)α a(k+2) nXf =1 Xα∈Is,n{j} lf Xα∈Is,n{j} rdetj(cid:16)(A2)j.(a(k+2) )(cid:17)α α l. α = the lth component of a column-vector u.j = [u1j, . . . , unj], then nXl=1 Xβ∈Js1,n{i} cdeti(cid:0)(cid:0)Ak+1(cid:1). i (e.l)(cid:1)β β ulj = Xβ∈Js1 ,n{i} cdeti(cid:0)(cid:0)Ak+1(cid:1). i (u.j)(cid:1)β β . So, we get (37). In that connection, it would be logical to consider the following definition. Definition 3.13. Suppose A ∈ Hn×n and Ind A = k. A matrix X ∈ Hn×n is called the MPD inverse of A if it satisfies the conditions XAX = X, AX = AAd, and XAk = A†Ak. It is denoted A†,d. The matrix A†,d is unique, and it can be represented as A†,d = A†AAd. (38) Theorem 3.14. Let A ∈ Hn×n inverse A†,d =(cid:16)a†,d ij (cid:17) have the following determinantal representations. (i) If A is an arbitrary matrix, then s , IndA = k and rk Ak = s1. Then its MPD a†,d ij = Pβ∈Js,n Pβ∈Js,n{i} A∗Aβ β cdeti((A∗A).i(ev.j))β β Pβ∈Js1 ,n(cid:12)(cid:12)(A2k+1)∗ A2k+1(cid:12)(cid:12)β β , 17 where ev.j is the jth column of eV := A∗A2V, and V = (vf j ) ∈ Hn×n such that , β β a†,d (a.j)(cid:19)β (ii) If A is Hermitian, then A2k+1(cid:17).f vf j = Xβ∈Js1,n{f } cdetf(cid:18)(cid:16)(cid:0)A2k+1(cid:1)∗ where ba.j is the jth column of bA = (A2k+1)∗A. = Pα∈Is1 ,n{j} Pα∈Is, n α Pα∈Is1 ,n α ∈ Hn×1, (cid:17)(cid:17)α β ∈ H1×n, )(cid:17)β cdeti(cid:0)(A2).i(v.j )(cid:1)β β Pβ∈Is1 ,n rdetj(cid:16)(cid:0)Ak+1(cid:1)j.(cid:16)a cdeti(cid:16)(A2).i(a(k+2) ij = Pβ∈Js,n{i} Pβ∈Js,n v.j = Xα∈Is1 ,n{j} ui. = Xβ∈Js,n{i} Ak+1β β Ak+1α (k+2) l. A2β where .f Proof. The proof is similar to the proof of Theorem 3.12. By (38), l = 1, . . . , n. rdetj(cid:16)(cid:0)Ak+1(cid:1)j. (ui.)(cid:17)α α A2α α l = 1, . . . , n , (39) a†,d ij = qilad lj, nXl=1 (i) If A ∈ Hn×n is an arbitrary matrix, we use (10) for the determinantal rep- resentation of QA = A†A = (qij ) and (12) for the determinantal representation of Ad in (39). s (ii) If A ∈ Hn×n is Hermitian, then we substitute in the equation (39) the de- terminantal representation (8) for A† and for the determinantal representation (15) for Ad . s The next corollary gives determinantal representations of the DMP and MPD inverses for complex matrices. Corollary 3.4. Let A ∈ Cn×n DMP inverse has determinantal representations s with Ind A = k and rk(cid:0)Ak(cid:1) = s1. Then its ad,† ij = where , AA∗α α = Ak+1β AA∗α α (AA∗)j.(vi.)α α β Pα∈Is,n Pα∈Is,n{j} Pβ∈Js1,n vi. = Xβ∈Js1,n{i}(cid:12)(cid:12)(cid:12)(cid:0)Ak+1(cid:1).i(cid:16)a(k+2) u.j = Xα∈Is,n{j}(cid:12)(cid:12)(cid:12)(AA∗)j.(a(k+2) )(cid:12)(cid:12)(cid:12) .f l. β Ak+1β Pβ∈Js1,n{i}(cid:12)(cid:12)(cid:0)Ak+1(cid:1). i (u.j)(cid:12)(cid:12)β β Pβ∈Is,n Pβ∈Js1,n β ∈ C1×n, f = 1, . . . , n (cid:17)(cid:12)(cid:12)(cid:12) α ∈ Cn×1, l = 1, . . . , n. α β 18 And, its MPD inverse has determinantal representations a†,d ij = where , Ak+1α α α α = A∗Aβ Ak+1α α (A∗A).i(v.j )β β β Pα∈Is1 ,n Pβ∈Js,n{i} Pβ∈Js,n v.j = Xα∈Is1 ,n{j}(cid:12)(cid:12)(cid:12)(cid:0)Ak+1(cid:1)j.(cid:16)a(k+2) ui. = Xβ∈Js,n{i}(cid:12)(cid:12)(cid:12)(A∗A).i(a(k+2) )(cid:12)(cid:12)(cid:12) .f l. A∗Aβ β Pα∈Is1 ,n Pα∈Is1 ,n{j}(cid:12)(cid:12)(cid:12)(cid:0)Ak+1(cid:1)j. (ui.)(cid:12)(cid:12)(cid:12) Pβ∈Js,n α ∈ Cn×1, (cid:17)(cid:12)(cid:12)(cid:12) β ∈ C1×n, f = 1, . . . , n, . l = 1, . . . , n α β 3.4 Determinantal representations of the CMP inverse Recently, a new generalized inverse was investigated in [31] by M. Mehdipour and A. Salemi that can be extended to quaternion matrices as follows. Definition 3.15. Suppose A ∈ Hn×n the core-nilpotent decomposition A = A1 + A2, where Ind A1 = Ind A, A2 is nilpotent and A1A2 = A2A1 = 0. The CMP inverse of A is called the matrix Ac,† := A†A1A†. Similarly to complex matrices can be proved the next lemma. Lemma 3.4. Let A ∈ Hn×n. The matrix X = Ac,† is the unique matrix that satisfies the following system of equations: XAX = X, AXA = A1, AX = A1A†, and XA = A†A1. Moreover, Ac,† = QAAdPA. (40) Taking into account (40), it follows the next theorem about determinantal representations of the quaternion CMP inverse. , Ind A = m, and rk (Am) = s1. Then the ij (cid:17) can be expressed s as Theorem 3.16. Let A ∈ Hn×n ac,† ij = (i) when A is an arbitrary matrix determinantal representations of its CMP inverse Ac,† =(cid:16)ac,† .j )(cid:17)β cdeti(cid:16)(A∗A).i(v,(l) Pβ∈Js,n{i} β!2 Pβ∈Js1,n(cid:12)(cid:12)(A2m+1)∗ A2m+1(cid:12)(cid:12)β i. )(cid:17)α rdetj(cid:16)(AA∗)j. (w (l) Pα∈Is,n{j} α!2 Pβ∈Js1,n(cid:12)(cid:12)(A2m+1)∗ A2m+1(cid:12)(cid:12)β Pβ∈Js,n Pα∈Is,n AA∗α A∗Aβ = α β β β = (41) (42) 19 for all l = 1, 2, where v(1) w (1) .j = Xα∈Is,n{j} i. = Xβ∈Js,n{i} .j = Xα∈Is,n{j} i. = Xβ∈Js,n{i} v(2) w(2) rdetj(cid:16)(AA∗)j. (ut.)(cid:17)α cdeti ((A∗A).i (u.k))β rdetj(cid:16)(AA∗)j. (gt.)(cid:17)α cdeti ((A∗A).i (g.k))β α ∈ Hn×1, t = 1, . . . , n, β ∈ H1×n, k = 1, . . . , n, α ∈ Hn×1, t = 1, . . . , n, β ∈ H1×n, k = 1, . . . , n. (43) (44) (45) (46) Here ut. is the tth row and u.k is the kth column of U := UAm+1A∗, gt. is the tth row and g.k is the kth column of G := A∗A2G, and the matrices U = (uij) ∈ Hn×n and G = (gij) ∈ Hn×n are such that uij = Xα∈Is1 ,n{j} gij = Xβ∈Js1,n{i} rdetj(cid:18)(cid:16)A2m+1(cid:0)A2m+1(cid:1)∗(cid:17)j. A2m+1(cid:17).i(cid:16)a cdeti(cid:16)(cid:16)(cid:0)A2m+1(cid:1)∗ α i. )(cid:19)α .j (cid:17)(cid:17)β (2) β (a(1) , , where a(1) of A2 := (A2m+1)∗Am+1A∗. i. is the ith row of A1 := A∗Am+1(A2m+1)∗ and a(2) .j is the jth column (ii) If A is Hermitian, then ac,† ij = ac,† ij = for all l = 1, 2, where A2β Pβ∈Js,n{i} Pβ∈Js,n Pα∈Is,n{j} Pα∈Is,n β .j )(cid:17)β cdeti(cid:16)(A2).i(v(l) β!2 Pβ∈Js1,n(cid:12)(cid:12)(cid:12)(Am+1)β β(cid:12)(cid:12)(cid:12) i. )(cid:17)α rdetj(cid:16)(cid:0)A2(cid:1)j. (w(l) α!2 Pβ∈Js1,n(cid:12)(cid:12)(cid:12)(Am+1)β β(cid:12)(cid:12)(cid:12) α A2α = (47) (48) 20 v(1) w(1) .j = Xα∈Is,n{j} i. = Xβ∈Js,n{i} .j = Xα∈Is,n{j} i. = Xβ∈Js,n{i} rdetj(cid:16)(cid:0)A2(cid:1)j. (ut.)(cid:17)α cdeti(cid:0)(cid:0)A2(cid:1).i (u.k)(cid:1)β rdetj(cid:16)(cid:0)A2(cid:1)j. (gt.)(cid:17)α cdeti(cid:0)(cid:0)A2(cid:1).i (g.k)(cid:1)β α ∈ Hn×1, t = 1, . . . , n, β ∈ H1×n, k = 1, . . . , n, α ∈ Hn×1, t = 1, . . . , n, β ∈ H1×n, k = 1, . . . , n. v(2) w(2) (49) (50) (51) (52) Here but. is the tth row and bu.k is the kth column of widehatU := UA2, gt. is the tth row and bg.k is the kth column of bG := A2G, and the matrices U = (uij) ∈ Hn×n and G = (gij ) ∈ Hn×n are such that uij = Xα∈Is1 ,n{j} gij = Xβ∈Js1 ,n{i} , i. α rdetj(cid:16)(cid:0)Am+1(cid:1)j. (a(m+2) )(cid:17)α (cid:17)(cid:17)β cdeti(cid:16)(cid:0)Am+1(cid:1).i(cid:16)a(m+2) nXl=1 nXk=1 lkpkj , qilad .j β ac,† ij = . (53) Proof. By (40), where QA = (qil), Ad = (ad il), and PA = (pil). be an arbitrary matrix. (i) Let A ∈ Hn×n a) Using (13), (10), and (11) for the determinantal representations of Ad, s QA = A†A, and PA = AA†, respectively, we obtain ac,† ij = Pβ∈Js,n{i} nXl=1 nXk=1 nPt=1 Pα∈Is1 ,n{t} Pα∈Is,n{j} Pα∈Is,n cdeti ((A∗A).i ( a.l))β β × A∗Aβ β Pβ∈Jr,n rdett(cid:16)(cid:16)A2m+1(cid:0)A2m+1(cid:1)∗(cid:17).t Pα∈Is1 ,n(cid:12)(cid:12)A2m+1 (A2m+1)∗(cid:12)(cid:12)α α rdetj((AA∗)j.(ak.))α α AA∗α α = 21 α! a(m) tk (al.)(cid:17)α × =Xl Xt Pα∈Is1 ,n{t} Pα∈Is,n{j} β A∗Aβ β Pβ∈Js,n cdeti ((A∗A).i ( a.l))β Pβ∈Js,n{i} rdett(cid:16)(cid:16)A2m+1(cid:0)A2m+1(cid:1)∗(cid:17).t Pα∈Is1 ,n(cid:12)(cid:12)A2m+1 (A2m+1)∗(cid:12)(cid:12)α Pα∈Is,n rdetj((AA∗)j.(at.))α α AA∗α α α , × (al.)(cid:17)α α × where al. is the lth row of A := Am(A2m+1)∗ and at. is the tth row of A := Am+1A∗. Denote A∗Am+1(A2m+1)∗ =: A1 = (a(1) ij ). So, it is clear that cdeti ((A∗A).i (e.t))β β ac,† ij =Xl Xt Xk Pα∈Is1 ,n{l} Pα∈Is,n{j} A∗Aβ β Pβ∈Js,n Pβ∈Js,n{i} rdetl(cid:16)(cid:16)A2m+1(cid:0)A2m+1(cid:1)∗(cid:17)l. Pα∈Is,n(cid:12)(cid:12)A2m+1 (A2m+1)∗(cid:12)(cid:12)α Pα∈Is,n rdetj((AA∗)j.(al.))α AA∗α α α α , a(1) tk × (ek.)(cid:17)α α × where e.t and ek. are the unit column and row vectors. the tth component of a column-vector u. l = [u1l, . . . , unl], then cdet ((A∗A).i (e.t))β cdeti ((A∗A).i (u.l))β β. If we denote by utl :=Xk a(1) Xα∈Is1 ,n{j} Xt Xβ∈Js,n{i} It follows that ac,† ij =Xl Pβ∈Js,n{i} Pβ∈Js,n α (1) (a rdetl(cid:16)(cid:16)A2m+1(cid:0)A2m+1(cid:1)∗(cid:17)l. t. )(cid:17)α tk Xα∈Is1 ,n{j} rdetl(cid:16)(cid:16)A2m+1(cid:0)A2m+1(cid:1)∗(cid:17)l. β utl = Xβ∈Js,n{i} β Pα∈Is,n{j} β Pα∈Is1 ,n(cid:12)(cid:12)A2m+1 (A2m+1)∗(cid:12)(cid:12)α cdeti ((A∗A).i (u.l))β A∗Aβ Construct the matrix U = (utl) ∈ Hn×n, where utl is given by (54). Denote U A = UAm+1A∗ =: U = (utk). Then, taking into account that A∗Aβ β = AA∗α α, we have 22 (ek.)(cid:17)α α = (54) rdetj((AA∗)j.(al.))α α . AA∗α α α Pα∈Is,n ij =PtPk Pβ∈Js,n{i} ac,† If we denote by v(1) tj :=Xk utk Xα∈Is,n{j} cdeti ((A∗A).i (e.t))β rdetj((AA∗)j.(ek.))α α β utk Pα∈Is,n{j} β(cid:17)2 Pα∈Is1 ,n(cid:12)(cid:12)A2m+1 (A2m+1)∗(cid:12)(cid:12)α α Pβ∈Js,n(cid:16)A∗Aβ rdetj(cid:16)(AA∗)j. (ek.)(cid:17)α α . = Xα∈Is,n{j} 1j , . . . , v(1) rdetj(cid:16)(AA∗)j. (ut.)(cid:17)α nji, then α .j (cid:17)(cid:17)β cdeti(cid:16)(A∗A).i(cid:16)v(1) β . (1) .j =hv(1) tj = Xβ∈Js,n{i} the tth component of a column-vector v Xt Xβ∈Js,n{i} cdet ((A∗A).i (e.t))β β v(1) Thus we have (41) with v(1) .j from (43). If we denote by w(1) ik :=Xt Xβ∈Js,n{i} cdeti ((A∗A).i (e.t))β cdeti ((A∗A).i (u.k))β β β utk = Xβ∈Js,n{i} i. =hw(1) = Xα∈Is,n{j} i1 , . . . , w(1) ini, then i. )(cid:17)α rdetj(cid:16)(AA∗)j. (w(1) α . the kth component of a row-vector w(1) Xk w(1) ik Xα∈Is,n{j} rdetj(cid:16)(AA∗)j. (ek.)(cid:17)α α Thus we have (42) with w(1) from (44). i. b) By using the determinantal representation (12) for Ad in (53), we get ac,† ij = Pβ∈Js,n{i} Pβ∈Js1 ,n{t} cdeti ((A∗A).i ( a.l))β β × A∗Aβ β Pβ∈Js,n cdett(cid:16)(cid:16)(cid:0)A2m+1(cid:1)∗ Pβ∈Js1,n(cid:12)(cid:12)(A2m+1)∗ A2m+1(cid:12)(cid:12)β β rdetj((AA∗)j.(ak.))α α lt a(m) nXk=1 nXl=1 nPt=1 Pα∈Is,n{j} A2m+1(cid:17).t (a.k)(cid:17)β β × Pα∈Is,n AA∗α α = 23 =Xk Xt Pβ∈Js1,n{t} Pα∈Is,n{j} β A∗Aβ β Pβ∈Js,n cdeti ((A∗A).i (a.t))β Pβ∈Js,n{i} A2m+1(cid:17).t cdett(cid:16)(cid:16)(cid:0)A2m+1(cid:1)∗ Pβ∈Js1 ,n(cid:12)(cid:12)(A2m+1)∗ A2m+1(cid:12)(cid:12)β Pα∈Is,n rdetj((AA∗)j.(ak.))α AA∗α α α β , × (a.k)(cid:17)β β × where a.t is the tth column of A := A∗A2 and a.k is the kth column of A = (A2m+1)∗A. Denote A2 = (a(2) ij ) = (A2m+1)∗A2A∗. So, it is clear that cdeti ((A∗A).i (a.t))β β ac,† ij =Xl Xk Xt Pβ∈Js1 ,n{t} Pα∈Is,n{j} A∗Aβ β Pβ∈Js,n{i} Pβ∈Js,n A2m+1(cid:17).t cdett(cid:16)(cid:16)(cid:0)A2m+1(cid:1)∗ β(cid:12)(cid:12)(cid:12) Pβ∈Js1,n(cid:12)(cid:12)(cid:12)(cid:0)(A2m+1)∗ A2m+1(cid:1)β Pα∈Is,n rdetj((AA∗)j.(el.))α α AA∗α α , × (e.k)(cid:17)β β a(2) kl × where e.k and el. are the unit column and row vectors, respectively. If we denote by gtl :=Xl Xβ∈Js1,n{t} = Xβ∈Js1 ,n{t} cdett(cid:16)(cid:16)(cid:0)A2m+1(cid:1)∗ cdett(cid:16)(cid:16)(cid:0)A2m+1(cid:1)∗ A2m+1(cid:17).t A2m+1(cid:17).t(cid:16)a (e.k)(cid:17)β .l (cid:17)(cid:17)β (2) β β a(2) kl = (55) the lth component of a row-vector gt. = [gt1, . . . , gtn], then rdetj((AA∗)j.(el.))α rdetj((AA∗)j.(gt.))α α. It follows that Xl gtl Xα∈Is,n{j} Pβ∈Js,n{i} Pβ∈Jr,n ac,† ij =Xt A∗Aβ α = Xα∈Is,n{j} β Pα∈Is,n{j} β Pα∈Is1 ,n(cid:12)(cid:12)A2m+1 (A2m+1)∗(cid:12)(cid:12)α 24 cdeti ((A∗A).i (a.t))β rdetj((AA∗)j.(gt.))α α . AA∗α α α Pα∈Is,n Construct the matrix G = (gtl) ∈ Hn×n, where gtf is given by (55). Denote AG = A∗A2G =: G = (gtl). Then ij =PtPk Pβ∈Js,n{i} ac,† If we denote by v(2) tj :=Xk atk Xα∈Is,n{j} . cdeti ((A∗A).i (e.t))β rdetj((AA∗)j.(ek.))α α β gtk Pα∈Is,n{j} β(cid:17)2 Pα∈Is1 ,n(cid:12)(cid:12)A2m+1 (A2m+1)∗(cid:12)(cid:12)α α Pβ∈Jr,n(cid:16)A∗Aβ rdetj(cid:16)(AA∗)j. (ek.)(cid:17)α α = Xα∈Is,n{j} rdetj(cid:16)(AA∗)j. (gt.)(cid:17)α nji, then α .j (cid:17)(cid:17)β cdeti(cid:16)(A∗A).i(cid:16)v(2) β . .j =hv(2) tj = Xβ∈Js,n{i} the tth component of a column-vector v(2) 1j , . . . , v(2) Xt Xβ∈Js,n{i} cdet ((A∗A).i (e.t))β β v2 Thus we have (41) with v(2) .j from (45). If we denote by w(2) ik :=Xt Xβ∈Js,n{i} cdeti ((A∗A).i (e.t))β cdeti ((A∗A).i (g.k))β β (2) β gtk = Xβ∈Js,n{i} i. =hw(2) = Xα∈Is,n{j} i1 , . . . , w(2) ini, then i. )(cid:17)α rdetj(cid:16)(AA∗)j. (w(2) α . the kth component of a row-vector w Xk w(2) ik Xα∈Is,n{j} rdetj(cid:16)(AA∗)j. (ek.)(cid:17)α α Thus we have (42) with w from (46). (2) i. s be Hermitian. (ii) Let A ∈ Hn×n a) By using the determinantal representations (15) for Ad, (10) for QA = A†A, and (11) for PA = AA†, and taking into account Hermicity of A, we have ac,† ij =Xl Xt Pα∈Is1 ,n{l} β A2β β cdeti(cid:16)(cid:0)A2(cid:1).i(cid:16)a(2) .t (cid:17)(cid:17)β Pβ∈Js,n{i} Pβ∈Js,n rdetl(cid:16)(Am+1)l.(cid:16)a(m) t. (cid:17)(cid:17)α Pα∈Is1 ,n (Am+1) α α α × Pα∈Is,n{j} rdetj(cid:16)(A2)j.(a(2) l. )(cid:17)α Pα∈Is,n A2α α α , where a(2) tth row of Am. So, it is clear that .t and a(2) l. are the tth column and the lth row of A2, and a(m) t. is the 25 ac,† ij =Xl Xt Xk Pα∈Is1 ,n{l} β A2β β cdeti(cid:0)(cid:0)A2(cid:1).i (e.t)(cid:1)β Pβ∈Js,n Pα∈Is,n{j} Pβ∈Js,n{i} rdetl(cid:0)(cid:0)Am+1(cid:1)l. (ek.)(cid:1)α Pα∈Is1 ,n Am+1α α α a(m+2) tk × rdetj(cid:16)(A2)j.(a(2) l. )(cid:17)α Pα∈Is,n A2α α α where e.t and ek. are the unit column and row vectors, respectively. If we denote by the tth component of a column-vector u. l = [u1l, . . . , unl], then rdetl(cid:0)(cid:0)Am+1(cid:1)l. (ek.)(cid:1)α α = )(cid:17)α α a(m+2) tk utl :=Xk = Xα∈Is1 ,n{j} t. Xα∈Is1 ,n{j} rdetl(cid:16)(cid:0)Am+1(cid:1)l. (a(m+2) utl = Xβ∈Js,n{i} β Pα∈Is,n{j} Am+1α β cdet(cid:0)(cid:0)A2(cid:1).i (e.t)(cid:1)β Pβ∈Js,n{i} cdeti(cid:0)(cid:0)A2(cid:1).i (u.l)(cid:1)β β Pα∈Is1 ,n Pβ∈Jr,n A2β Xt Xβ∈Js,n{i} It follows that ac,† ij =Xl (56) , . . β α cdeti(cid:0)(cid:0)A2(cid:1).i (u.l)(cid:1)β rdetj(cid:16)(A2)j.(a(2) l. )(cid:17)α α Pα∈Is,n β =(cid:12)(cid:12)A2(cid:12)(cid:12)α utk Pα∈Is,n{j} rdetj(cid:0)(A2)j.(ek.)(cid:1)α α, we have A2α α α . Construct the matrix U = (utl) ∈ Hn×n, where utl is given by (56). Denote β ac,† U := UA2. Then, taking into account that(cid:12)(cid:12)A2(cid:12)(cid:12)β cdeti(cid:0)(cid:0)A2(cid:1).i (e.t)(cid:1)β ij =PtPk Pβ∈Js,n{i} Pβ∈Js,n(cid:16)A2β β(cid:17)2 Pα∈Is1 ,n rdetj(cid:16)(cid:0)A2(cid:1)j. (ek.)(cid:17)α :=Xk utk Xα∈Is,n{j} Xα∈Is,n{j} If we denote by Am+1α α v(1) tj α = the tth component of a column-vector v(1) Xt Xβ∈Js,n{i} cdet(cid:0)(cid:0)A2(cid:1).i (e.t)(cid:1)β β α rdetj(cid:16)(cid:0)A2(cid:1)j. (ut.)(cid:17)α .j =hv(1) tj = Xβ∈Js,n{i} v(1) 26 1j , . . . , v(1) nji, then .j (cid:17)(cid:17)β cdeti(cid:16)(cid:0)A2(cid:1).i(cid:16)v(1) β . Thus we have (47) with v1 .j from (49). If we denote by w(1) ik :=Xt Xβ∈Js,n{i} Xβ∈Js,n{i} cdeti(cid:0)(cid:0)A2(cid:1).i (e.t)(cid:1)β cdeti(cid:0)(cid:0)A2(cid:1).i (u.k)(cid:1)β β β utk = the kth component of a row-vector w(1) Xk w(1) ik Xα∈Is,n{j} rdetj(cid:16)(cid:0)A2(cid:1)j. (ek.)(cid:17)α α i1 , . . . , w(1) i. =hw(1) = Xα∈Is,n{j} ini, then rdetj(cid:16)(cid:0)A2(cid:1)j. (w(1) i. )(cid:17)α α . Thus we have (48) with w(1) from (50). i. b) By using the determinantal representation (14) for Ad in (53), we get ac,† ij =Xk Xt Pβ∈Js1,n{t} It is clear that ac,† ij =Xl Xk Xt Pβ∈Js1,n{t} If we denote by × β β (m) A2β β .t (cid:17)(cid:17)β cdeti(cid:16)(cid:0)A2(cid:1).i(cid:16)a(2) Pβ∈Js,n{i} Pβ∈Js,n cdett(cid:16)(cid:0)Am+1(cid:1).t(cid:16)a .k (cid:17)(cid:17)β Pβ∈Js,n(cid:12)(cid:12)(cid:12)(Am+1)β β(cid:12)(cid:12)(cid:12) cdeti(cid:16)(cid:0)A2(cid:1).i(cid:16)a(2) .t (cid:17)(cid:17)β Pβ∈Js,n{i} Pβ∈Js,n cdett(cid:0)(cid:0)Am+1(cid:1).t (e.k)(cid:1)β Pβ∈Js1,n(cid:12)(cid:12)(cid:12)(Am+1)β β(cid:12)(cid:12)(cid:12) gtl :=Xl Xβ∈Js1,n{t} Xβ∈Js1 ,n{t} a(m+2) A2β β kl β β × β cdett(cid:0)(cid:0)Am+1(cid:1).t (e.k)(cid:1)β (cid:17)(cid:17)β cdett(cid:16)(cid:0)Am+1(cid:1).t(cid:16)a(m+2) α = Xα∈Is,n{j} .l β 27 the lth component of a row-vector gt. = [gt1, . . . , gtn], then Xl gtl Xα∈Is,n{j} rdetj(cid:0)(A2)j.(el.)(cid:1)α rdetj(cid:0)(A2)j.(gt.)(cid:1)α α. Pα∈Is,n{j} rdetj(cid:16)(A2)j.(a(2) k. )(cid:17)α Pα∈Is,n A2α α α . Pα∈Is,n{j} rdetj(cid:0)(A2)j.(el.)(cid:1)α Pα∈Is,n A2α α α . a(m+2) kl = (57) It follows that Pβ∈Js,n{i} ac,† ij =Xt cdeti(cid:16)(cid:0)A2(cid:1).i(cid:16)a(2) .t (cid:17)(cid:17)β β Pα∈Is1 ,n Pβ∈Jr,n A2β rdetj(cid:0)(A2)j.(gt.)(cid:1)α β Pα∈Is,n{j} α Pα∈Is,n A2α α Am+1α α . Construct the matrix G = (gtl) ∈ Hn×n, where gtl is given by (57). Denote G := AG = A2G. Then If we denote by ij =PtPk Pβ∈Js,n{i} ac,† β cdeti(cid:0)(cid:0)A2(cid:1).i (e.t)(cid:1)β β(cid:17)2 Pα∈Is1 ,n Pβ∈Js,n(cid:16)A2β :=Xk atk Xα∈Is,n{j} rdetj(cid:16)(cid:0)A2(cid:1)j. (gt.)(cid:17)α v(2) tj α Xα∈Is,n{j} gtk Pα∈Is,n{j} rdetj(cid:0)(A2)j.(ek.)(cid:1)α α . Am+1α α rdetj(cid:16)(cid:0)A2(cid:1)j. (ek.)(cid:17)α α = the tth component of a column-vector v(2) Xt Xβ∈Js,n{i} cdet(cid:0)(cid:0)A2(cid:1).i (e.t)(cid:1)β β Thus we have (47) with v2 .j from (51). If we denote by 1j , . . . , v(2) .j =hv(2) tj = Xβ∈Js,n{i} v(2) nji, then .j (cid:17)(cid:17)β cdeti(cid:16)(cid:0)A2(cid:1).i(cid:16)v(2) β . w(2) ik :=Xt Xβ∈Js,n{i} Xβ∈Js,n{i} cdeti(cid:0)(cid:0)A2(cid:1).i (e.t)(cid:1)β cdeti(cid:0)(cid:0)A2(cid:1).i (g.k)(cid:1)β β β gtk = the kth component of a row-vector w(2) Xk w(2) ik Xα∈Is,n{j} rdetj(cid:16)(cid:0)A2(cid:1)j. (ek.)(cid:17)α i1 , . . . , w(2) i. =hw(2) = Xα∈Is,n{j} α ini, then rdetj(cid:16)(cid:0)A2(cid:1)j. (w(2) i. )(cid:17)α α . Thus we have (48) with w(2) i. from (52). The following corollary gives determinantal representations of the CMP in- verse for complex matrices. Corollary 3.5. Let A ∈ Cn×n inverse Ac,† =(cid:16)ac,† ij (cid:17) has the following determinantal representations s , Ind A = m and rk Am = s1. Then its CMP 28 for all l = 1, 2, where = = β β α α ac,† ij = A∗Aβ AA∗α β!2 .j )(cid:12)(cid:12)(cid:12) Pβ∈Js,n{i}(cid:12)(cid:12)(cid:12)(A∗A).i(v(l) Pβ∈Js,n Pβ∈Js1 ,n(cid:12)(cid:12)(cid:12)(Am+1)β β(cid:12)(cid:12)(cid:12) i. )(cid:12)(cid:12)(cid:12) Pα∈Is,n{j}(cid:12)(cid:12)(cid:12)(AA∗)j. (w(l) α!2 Pα∈Is,n Pβ∈Js1,n(cid:12)(cid:12)(cid:12)(Am+1)β β(cid:12)(cid:12)(cid:12) α ∈ Cn×1, t = 1, . . . , n, .j = Xα∈Is,n{j}(cid:12)(cid:12)(cid:12)(AA∗)j. (ut.)(cid:12)(cid:12)(cid:12) β ∈ C1×n, k = 1, . . . , n, i. = Xβ∈Js,n{i} α ∈ Cn×1, t = 1, . . . , n, .j = Xα∈Is,n{j}(cid:12)(cid:12)(cid:12)(AA∗)j. (gt.)(cid:12)(cid:12)(cid:12) β ∈ C1×n, k = 1, . . . , n, i. = Xβ∈Js,n{i} (A∗A).i (u.k)β (A∗A).i (g.k)β α α v(1) w(1) v(2) w(2) Here ut. is the tth row and u.k is the kth column of U = UA2, gt. is the tth row and g.k is the kth column of G = A2G, and the matrices U = (uij) ∈ Cn×n and G = (gij) ∈ Cn×n are such that i. )(cid:12)(cid:12)(cid:12) uij = Xα∈Is1 ,n{j}(cid:12)(cid:12)(cid:12)(cid:0)Am+1(cid:1)j. (a(m+2) (cid:17)(cid:12)(cid:12)(cid:12) gij = Xβ∈Js1,n{i}(cid:12)(cid:12)(cid:12)(cid:0)Am+1(cid:1).i(cid:16)a(m+2) .j α α , β β . 4 An example Given the matrix Since A = A∗A = 0 k i 0 0 0 −j j 0 . 1 , 0 k 1 0 2 0 k 0 29 then (determinantal) rk(A∗A) = rk(A) = 2. Similarly, it can be find A2 = j 0 −k 0 i 1 − j 0 0 1 , A2(cid:0)A2(cid:1)∗ = 2 0 0 2 −2i 0 2i 0 2 , and rk(A2) = 2. So, Ind A = 1 and we shall find A #(cid:13) and A #(cid:13) by (31) and (32), respectively. Since then by (31) Similarly, a #(cid:13),r 21 = = a #(cid:13),r 31 = = a #(cid:13),r 12 = = a #(cid:13),r 22 = = a #(cid:13),r 32 = = a #(cid:13),r 13 = = a #(cid:13),r 23 = = a #(cid:13),r 33 = = So, By analogy, due to (32), we have A = A(A2)∗ = 11 = Pα∈I2,3{1} a #(cid:13),r α 0 0 0 0 −2j −2k −1 + j −i + k 0  , rdet1(cid:16)(cid:16)A2(cid:0)A2(cid:1)∗(cid:17)1. (a1.)(cid:17)α Pα∈I2,3(cid:12)(cid:12)A2 (A2)∗(cid:12)(cid:12)α (cid:21) + rdet1(cid:20) 0 −2i 2(cid:21)(cid:19) = 0. 2(cid:21)(cid:19) = −4k, −2i 2(cid:21)(cid:19) = 0, 2(cid:21) + rdet1(cid:20) 2 (cid:21) + rdet1(cid:20) 0 −2i 2i = 0 0 2 0 2 α 0 0 0 −i + k(cid:21) + rdet2(cid:20)−i + k 0 −2k 0(cid:21) + rdet2(cid:20)0 −2j 2 (cid:21)(cid:19) = 0, 0 −1 + j(cid:21) + rdet2(cid:20)−1 + j 0 0(cid:21) + rdet2(cid:20) −i + k 2(cid:21)(cid:19) = 0, −2k −2j(cid:21) + rdet2(cid:20)2 0(cid:21) + rdet2(cid:20) 2i 2i 0 2 0 0 0 −2j(cid:21)(cid:19) = −4j, 0 0 0 0 1 1 1 1 1 1 0 8(cid:18)rdet1(cid:20)0 −i + k 8(cid:18)rdet1(cid:20)−2k 0 8(cid:18)rdet1(cid:20)0 −1 + j 8(cid:18)rdet1(cid:20)2 8(cid:18)rdet1(cid:20) 2 8(cid:18)rdet1(cid:20)2 8(cid:18)rdet1(cid:20)2 8(cid:18)rdet1(cid:20) 2 8(cid:18)rdet1(cid:20)2 A #(cid:13) = A #(cid:13) = 0.5i − 0.5k −0.5k 2i 1 0 0 1 0 0 0 0 1 −0.5i + 0.5k −0.5 + 0.5j 0 0 2 0 −0.5j −1 + j 0(cid:21)(cid:19) = 0. 0  .  . 0.5 − 0.5j 0 0 0 −0.5i 0.5j 30 2(cid:21)(cid:19) = −4i + 4k, 2(cid:21)(cid:19) = −4 + 4j, We can verify the results, for example, by the representation (ii) from Lemma 3.2 because by Theorem 2.3 A† = −0.5i 0 0.5j −0.5j 0 0  , (A #(cid:13))† = 0 −0.5i + 0.5k i 0 0.5 + 0.5j 0 0 −j 0 , −0.5k 0 0 and QAA = (A #(cid:13))†. 5 Conclusion Notions of the core inverse, the core EP inverse, the DMP and MPD inverses, and the CMP inverse have been extended to quaternion matrices in this paper. Due to noncommutativity of quaternions, these generalized inverses in quater- nion matrices have some features in comparison to complex matrices. We have obtained their determinantal representations within the framework of the the- ory of column-row determinants previously introduced by the author. As the special case, their determinantal representations in complex matrices have been obtained as well. A numerical example to illustrate the main result has given. References [1] H. Aslaksen, Quaternionic determinants, Math. Intellig. 18(3) (1996) 57-65. [2] O.M. Baksalary, G. Trenkler, Core inverse of matrices, Linear Multilinear Algebra 58 (2010) 681-697. [3] O.M. Baksalary, G. Trenkler, On a generalized core inverse, Appl. Math. Comput. 236 (2014) 450-457. [4] R.B. Bapat, K.P.S. Bhaskara Rao, K. Manjunatha Prasad, Generalized inverses over integral domains, Linear Algebra Appl. 140 (1990) 181-196. [5] K.P.S. Bhaskara Rao, Generalized inverses of matrices over integral do- mains, Linear Algebra Appl. 49 (1983) 179-189. [6] 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. [7] N. Cohen, S. De Leo. The quaternionic determinant, Electron. J. Linear Algebra 7 (2000) 100-111. [8] Y.F. Gao, J.L. Chen, Pseudo core inverses in rings with involution, Comm. Algebra 46 (2018) 38-50. [9] A. Guterman, A. Herrero, N. Thome, New matrix partial order based on spectrally orthogonal matrix decomposition, Linear Multilinear Algebra 64(3) (2016) 362-374. [10] D.E. Ferreyra, F.E. Levis, N. Thome, Maximal classes of matrices deter- mining generalized inverses, Appl. Math. Comput. 333 (2018) 42-52. 31 [11] D.E. Ferreyra, F.E. Levis, N. Thome, Revisiting the core EP inverse and its extension to rectangular matrices, Quaest. Math. 41 (2) (2018) 265-281. [12] I. Kyrchei, Analogs of the adjoint matrix for generalized inverses and cor- responding Cramer rules, Linear Multilinear Algebra 56(4) (2008) 453-469. [13] I. Kyrchei, Explicit formulas for determinantal representations of the Drazin inverse solutions of some matrix and differential matrix equations, Appl. Math. Comput. 219 (2013) 7632-7644. [14] I. Kyrchei, Cramer's rule for generalized inverse solutions. In: I. Kyrchei (Ed.), Advances in Linear Algebra Research, pp. 79 -- 132, Nova Sci. Publ., New York, 2015. [15] I. Kyrchei, Cramer's rule for quaternionic systems of linear equations, J. Math. Sci. 155(6) (2008) 839 -- 858. [16] I. Kyrchei, The theory of the column and row determinants in a quater- nion linear algebra. In: Albert R. Baswell (Ed.), Advances in Mathematics Research 15, pp. 301-359, Nova Sci. Publ., New York, 2012. [17] I. Kyrchei, Determinantal representations of the Moore-Penrose inverse over the quaternion skew field and corresponding Cramer's rules, Linear Multilinear Algebra 59(4) (2011) 413-431. [18] I. Kyrchei, Determinantal representations of the Drazin inverse over the quaternion skew field with applications to some matrix equations, Appl. Math. Comput. 238 (2014) 193-207. [19] I. Kyrchei, Determinantal representations of the W-weighted Drazin inverse over the quaternion skew field, Appl. Math. Comput. 264 (2015) 453-465. [20] I. Kyrchei, Explicit determinantal representation formulas of W-weighted Drazin inverse solutions of some matrix equations over the quaternion skew field, Math. Probl. Eng. ID 8673809 (2016) 13 p. [21] I. Kyrchei, Explicit determinantal representation formulas for the solution of the two-sided restricted quaternionic matrix equation, J. Appl. Math. Comput. 58(1-2) (2018) 335 -- 365. [22] I. Kyrchei, Determinantal representations of the Drazin and W-weighted Drazin inverses over the quaternion skew field with applications. In: Sandra Griffin (Ed.), Quaternions: Theory and Applications, pp.201-275, Nova Sci. Publ., New York, 2017. [23] I. Kyrchei, Weighted singular value decomposition and determinantal repre- sentations of the quaternion weighted Moore-Penrose inverse, Appl. Math. Comput. 309 (2017) 1-16. [24] I.I. Kyrchei, Determinantal representations of the quaternion weighted Moore-Penrose inverse and its applications, In: Albert R.B. (Ed.), Ad- vances in Mathematics Research 23, pp.35 -- 96, Nova Sci. Publ., New York, 2017. 32 [25] I. Kyrchei, Determinantal representations of solutions to systems of quater- nion matrix equations, Adv. Appl. Clifford Algebras 28(1) (2018) 23. [26] I.I. Kyrchei, Cramer's rules for Sylvester quaternion matrix equation and its special cases, Adv. Appl. Clifford Algebras 28(4) (2018) 90. [27] I. Kyrchei, Determinantal representations of solutions and Hermitian solu- tions to some system of two-sided quaternion matrix equations, Journal of Mathematics ID 6294672 (2018) 12 p. [28] I. Kyrchei, Determinantal representations of solutions and Hermitian solu- tions to some system of two-sided quaternion matrix equations, Abstract and Applied Analysis ID 5926832 (2019) 14 p. [29] X. Liu, N. Cai, High-order iterative methods for the DMP inverse, Journal of Mathematics 8175935 (2018) 6 p. [30] S. Malik, N. Thome, On a new generalized inverse for matrices of an arbi- trary index, Appl. Math. Comput. 226 (2014) 575-580. [31] M. Mehdipour, A. Salemi, On a new generalized inverse of matrices, Linear Multilinear Algebra 66 (5) (2018) 1046-1053. [32] J. Mielniczuk, Note on the core matrix partial ordering, Discuss. Math. Probab. Stat. 31 (2011) 71-75. [33] D. Mosi´c, C. Deng, H. Ma, On a weighted core inverse in a ring with involution, Comm. Algebra 46(6) (2018) 2332-2345. [34] K.M. Prasad, K.S. Mohana, Core EP inverse, Linear Multilinear Algebra 62(3) (2014) 792-802. [35] K.M. Prasad, M.D. Raj, Bordering method to compute Core-EP inverse, Spec. Matrices 6 (2018) 193-200. [36] M. Zhou, J. Chen, T. Li, D. Wang, Three limit representations of the core- EP inverse, http://arxiv.org/abs/1804.05206v1. [37] 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. [38] P.S. Stanimirovi´c, General determinantal representation of pseudoinverses of matrices. Mat. Vesnik. 48 (1996) 1-9. [39] P.S. Stanimirovi´c, D.S. Djordjevic, Full-rank and determinantal represen- tation of the Drazin inverse, Linear Algebra Appl. 311 (2000) 131-151. [40] G.J. Song, Determinantal representations of the generalized inverses A(2) T,S over the quaternion skew field with applications, J. Appl. Math. Comput. 39 (2012) 201-220. [41] G.J. Song, Bott-Duffin inverse over the quaternion skew field with applica- tions, J. Appl. Math. Comput. 41 (2013) 377-392. 33 [42] G.J. Song, Q.W. Wang, Condensed Cramer rule for some restricted quater- nion linear equations, Appl. Math. Comp. 218 (2011) 3110-3121. [43] G.J. Song, Characterization of the W-weighted Drazin inverse over the quaternion skew field with applications, Electron. J. Linear Algebra 26 (2013) 1-14. [44] G.J. Song, Q.W. Wang, S.W. Yu, Cramer's rule for a system of quaternion matrix equations with applications, Appl. Math. Comp. 336 (2018) 490-499. [45] 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. [46] H.X. Wang, Core-EP decomposition and its applications, Linear Algebra Appl. 508 (2016) 289-300. 34
1503.08705	3	1503	2016-03-11T12:58:55	Leavitt $R$-algebras over countable graphs embed into $L_{2,R}$	[ "math.RA", "math.OA" ]	For a commutative ring $R$ with unit we show that the Leavitt path algebra $L_R(E)$ of a graph $E$ embeds into $L_{2,R}$ precisely when $E$ is countable. Before proving this result we prove a generalised Cuntz-Krieger Uniqueness Theorem for Leavitt path algebras over $R$.	math.RA	math	LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R NATHAN BROWNLOWE AND ADAM P W SØRENSEN Abstract. For a commutative ring R with unit we show that the Leavitt path algebra LR(E) of a graph E embeds into L2,R precisely when E is countable. Before proving this result we prove a generalised Cuntz-Krieger Uniqueness Theorem for Leavitt path algebras over R. 1. Introduction In [22] Leavitt introduced a class of rings R, which fail to possess the Invariant Basis Number Property in a strong sense. A ring R does not have invariant basis number if there exist m, n ∈ N, m (cid:54)= n, with Rm ∼= Rn as modules; such a ring is said to be of module type (m, n) if m is the smallest natural number such that Rm ∼= Rk for some k > m and if n > m is the minimal natural number with Rm ∼= Rn. For K a field and m, n any pair of natural numbers with 1 ≤ m < n, Leavitt introduced what is now called the Leavitt algebra LK(m, n), which is an algebra over K of module type (m, n). Furthermore the Leavitt algebras are universal for their module type. In [14] Cuntz introduced a class of C∗-algebras now referred to as Cuntz al- gebras. While it was not realised at the time the Cuntz algebras On are natural C∗-algebraic analogues of the Leavitt algebras LK(1, n) - they are universal for the same defining relations (in each their category) and LC(1, n) is dense in On. Partly for this reason we denote LK(1, n) by Ln,K from here on out. Cuntz algebras have by now been generalised in many different ways, including by the Cuntz-Krieger algebras of [16] and the more general C∗-algebras associated to directed graphs [18, 21]. Graph C∗-algebras have become their own industry in C∗-algebra theory, and have recently been used in the ongoing classification of C∗-algebras program [17, 27]. Inspired by the success of graph C∗-algebras, Abrams and Aranda Pino [3] and, independently, Ara, Moreno and Pardo [10] generalised Leavitt algebras by introducing Leavitt path algebras associated to directed graphs. Since the appear- ance of [3, 10] Leavitt path algebras have enjoyed a similar amount of attention as their C∗-algebraic cousins, including Tomforde's work on Leavitt path algebras with coefficients in arbitrary (unital) commutative rings [29]. For an account of the history of Leavitt path algebras see the excellent survey paper [1]. One of the main trends in Leavitt path algebras has been a desire to get a purely algebraic version of the celebrated classification theorem of Kirchberg and Phillips. The Kirchberg-Phillips Theorem (see [19, 24] and [20]) shows that a class of purely infinite C∗-algebras are completely classified by their K-theory. In a precursor to his proof of this classification theorem, Kirchberg proved three seminal theorems, which have been labelled "Kirchberg's Geneva Theorems" (see MR1780426): (1) A separable C∗-algebra is exact if and only if it is a sub-C∗-algebra of the Cuntz algebra O2; 2010 Mathematics Subject Classification. 16B99, 46L05, 46L55. Key words and phrases. Leavitt path algebra, graph C∗-algebra. 1 2 NATHAN BROWNLOWE AND ADAM P W SØRENSEN (2) A ⊗ O2 is isomorphic to O2 if and only if A is a unital, simple, separable (3) if A is a separable, simple, nuclear C∗-algebra, then A is isomorphic to and nuclear C∗-algebra; and A ⊗ O∞ if and only if A is purely infinite. While there are many similarities between the theories of graph C∗-algebras and Leavitt path algebras, the algebraic analogues of the Geneva Theorems have pro- vided some of the most striking differences between them so far. Ara and Cortinas studied the Hochschild homology of tensor products of Leavitt path algebras in [8], and concluded that the natural algebraic version of Geneva Theorems (2) and (3) do not hold. In particular, Ara and Cortinas show that for K any field L2,K ⊗ L2,K is not1 Morita equivalent to L2,K, thus providing a counterexample to (2). In fact they show that L2,K ⊗ L2,K cannot be isomorphic to any Leavitt path algebra. They also show that L∞,K ⊗ LK(E) is not Morita equivalent to LK(E) for a class of graphs E, thus providing a counterexample to (3). The result L2,K⊗L2,K (cid:54)∼= L2,K was also proved independently by Dicks and by Bell and Bergman. Both proofs are unpublished but are described by Abrams in [1, Section 3.5]. In this paper we commence an investigation into a possible algebraic analogue of Geneva Theorem (1). Rather than attempt a complete generalization, we take the more modest approach of investigate whether all Leavitt path algebras over R embed into L2,R. Our main result is that the Leavitt path algebra LR(E) of any countable graph E does indeed embed into L2,R. It is not hard to see that countability is a necessary assumption: a result of [5] implies that L2,R has at most countable dimension and Leavitt path algebras of uncountable graphs contain uncountable linearly independent subsets, which means they cannot possibly embed into L2,R. To establish the injectivity of the embeddings LR(E) (cid:44)→ L2,R we have proved a generalised Cuntz-Krieger Uniqueness Theorem for arbitrary graphs, in the spirit of Szymanski's Unqueness Theorem for graph C∗-algebras [28, Theorem 1.2]. Our Uniqueness Theorem generalises the Cuntz-Krieger Uniqueness Theorem proved by Tomforde [29, Theorem 6.5] for graphs in which every cycle has an exit (Condition (L)). 2. Background A directed graph E = (E0, E1, r, s) consists of sets E0 and E1, and maps r, s : E1 → E0. We call a graph E countable if E0 and E1 are countable, and un- countable otherwise. We call elements of E0 vertices, and elements of E1 edges. The maps r and s are called the range and source maps, respectively. A vertex v ∈ E0 is called a sink if s−1(v) = ∅, and is called an infinite emitter if s−1(v) = ∞. A path in a graph E is a sequence of edges α = e1e2 . . . en with r(ei) = s(ei+1) for all 1 ≤ i ≤ n − 1. A path of n edges is said to have length n. The set of paths of length n is denoted En, and the set of all finite paths is denoted E∗ = ∪∞ n=0En. The range and source maps extend to finite paths in the obvious way. A cycle α = e1 . . . en is a path of length at least one which satisfies s(α) = r(α). Cycles are also called closed paths in the literature. A cycle is called vertex simple if no vertices are repeated. An exit for a path α = e1 . . . en is an edge e such that s(e) = s(ei) and e (cid:54)= ei for some 1 ≤ i ≤ n. A graph E satisfies Condition (L) if every cycle has an exit. r(ei) = s(ei+1) for all i ≥ 1. We denote the set of infinite paths by E∞. An infinite path in a graph E is an infinite sequence of edges e1e2 . . . with 1This is in contrast to O2 ⊗ O2 ∼= O2, which was used in the proof of Theorems (1) and (2), but was published by Rørdam in [26] (and proved earlier by Elliott). LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R 3 Definition 2.1. Given a graph E we define (E1)∗ to be the set of formal symbols {e∗ : e ∈ E1}. In the literature, elements e∗ are called ghost edges, and elements of E1 are called real edges. We also require the set of formal symbols {v∗ : v ∈ E0}, but we insist that v∗ = v. For α = e1 . . . en we define α∗ = e∗ n . . . e∗ 1. The following definition first appeared in [4] (and appeared earlier in [3] in a less general setting). Definition 2.2. Let E be a directed graph and K a field. The Leavitt path algebra of E with coefficients in K, denoted LK(E), is the universal K-algebra generated by a family {v, e, e∗ : v ∈ E0, e ∈ E1} satisfying (1) {v : v ∈ E0} consists of pairwise orthogonal idempotents; (2) s(e)e = er(e) = e for all e ∈ E1; (3) r(e)e∗ = e∗s(e) = e∗ for all e ∈ E1; (4) e∗f = δe,f r(e) for all e, f ∈ E1; and s(f )=v f f∗ for all v ∈ E0 with 0 < s−1(v) < ∞. (5) v =(cid:80) If S is a ring and {v, e, e∗ : v ∈ E0, e ∈ E1} ⊆ S is a collection satisfying (1)–(5) above, we call this collection a Leavitt E-family in S. In [29, Definition 3.1] the definition of a Leavitt path algebra was generalised to include more general coefficients. For E a directed graph and R a commutative ring with unit, the Leavitt path algebra with coefficients in R, denoted LR(E), is the universal R-algebra generated by a Leavitt E-family. Recall from [29, Propo- sition 3.4] that the elements {v, e, e∗ : v ∈ E0, e ∈ E1} of the Leavitt E-family generating LR(E) are all nonzero. Moreover, that LR(E) = spanR{αβ∗ : α, β ∈ E∗ with r(α) = r(β)}, and rv (cid:54)= 0 for all v ∈ E0 and 0 (cid:54)= r ∈ R. Also recall that by [29, Proposition 4.9] the set E∗ and the set of all ghost paths are both linearly independent sets. The map αβ∗ (cid:55)→ βα∗ extends to an R-linear involution of LR(E). Standing Assumption. Throughout this paper R will always be a commutative ring with unit. We will be focussed on the graph (2.1) a b and its Leavitt path algebra L2,R, which is the universal R-algebra generated by elements a, a∗, b, b∗ satisfying a∗a = b∗b = 1 = aa∗ + bb∗. (2.2) From these relations we see that aa∗a = a and b∗bb∗ = b∗, and hence b∗a = b∗(aa∗ + bb∗)a = b∗a + b∗a. So b∗a = 0, and a similar argument gives a∗b = 0. Hence the universal R-algebra generated by a and b satisfying (2.2) is indeed the Leavitt path algebra for the above graph. Note that when R is a field K, L2,K is the Leavitt algebra of module type (1, 2). Definition 2.3. Let A be an involutive R-algebra. We call an element u ∈ A a unitary if u∗u = uu∗ = 1. We call an element p ∈ A a projection if p = p2 = p∗. We say that two projections are Murray-von Neumann equivalent, written p1 ∼ p2, if there exist x such that p1 = xx∗ and p2 = x∗x. If two projections p1, p2 are 4 NATHAN BROWNLOWE AND ADAM P W SØRENSEN Murray-von Neumann equivalent, we can choose a witness x such that xx∗x = x (see [12, Proposition 4.2.2]). Remark 2.4. The equivalence relation introduced above is often referred to as ∗- equivalence to distinguish it from the equivalence relation on idempotents given by letting e, f be equivalent if there exist x, y such that e = xy and f = yx. In general these notions of equivalence do not coincide, for instance they do not coincide on M2(Q), see [6, Theorem 1.12] for details and a much more general statement. The authors thanks Pere Ara for pointing out the difference to them. Definition 2.5. Let R be a commutative ring with unit and let A be an involutive R-algebra. A partial unitary is an element u ∈ A with u∗u = uu∗ = p for some projection p ∈ A. The subalgebra of A generated by the partial unitary u and p is a ∗-homomorphic image of LR[z, z−1], the Laurent polynomials over R. This allows one to apply any Laurent polynomial to u. More elaborately: Remark 2.6. LR[z, z−1] is the universal R-algebra generated by the partial uni- if u ∈ A is a partial unitary there exists a ∗-homomorphism tary z; that is, φ : LR[z, z−1] → A such that φ(z) = u, φ(z−1) = u∗, Let q be a polynomial in LR[z, z−1] and write and φ(1) = p. m(cid:88) n=−m q(z) = knzn where each kn ∈ R. Then φ(q) = −1(cid:88) n=−m kn(u∗)−n + p + m(cid:88) n=1 knun. In our notation we will omit the map φ and simply write q(u). Definition 2.7. Let R be a commutative ring with unit and let A be an involutive R-algebra. We say a partial unitary u ∈ A has full spectrum2 if q(u) (cid:54)= 0 for all nonzero q ∈ R[x]. Note that if u has full spectrum, then q(u) (cid:54)= 0 for all nonzero Laurent polyno- mials q. Example 2.8. The most basic example of a full spectrum partial unitary in a Leavitt path algebra is a cycle α without an exit. This follows because the set of paths of real edges is linearly independent [29, Proposition 4.9], and hence q(α) (cid:54)= 0 for all non-zero q ∈ R[x]. 3. A generalised Cuntz-Krieger Uniqueness Theorem We now prove a uniqueness theorem for Leavitt path algebras with coefficients in a unital, commutative ring R. This result generalises the uniqueness theorem of Tomforde [29, Theorem 6.5], in which graphs are assumed to satisfy Condition (L), and is the algebraic analogue of Szymanski's uniqueness theorem for graph C∗-algebras [28, Theorem 1.2]. (See also Nagy and Reznikoff's uniquess theorem in [23].) 2This term comes from the notion of full spectrum in C∗-algebra theory. LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R 5 Theorem 3.1 (Generalised Cuntz-Krieger Uniqueness Theorem). Let E be a graph, R a commutative ring with unit, and φ a ring homomorphism of LR(E) into a ring S. Then φ is injective if and only if the following conditions are satisfied: (G1) φ(rv) (cid:54)= 0 for all v ∈ E0 and all r ∈ R \ {0}; and (G2) φ(q(α)) (cid:54)= 0 for all cycles α without an exit, and all nonzero polynomials q ∈ R[x]. A version of this Theorem for fields can be derived either from [25, Proposition 2] or from [11, Proposition 3.1]. We work over rings rather than fields, so must do some extra work. To prove Theorem 3.1 we study the ideal structure of Leavitt path algebras over a commutative ring. In Section 7 of [29] Tomforde carries out an investigation into so called basic ideals, which are ideals I with the property that if rv ∈ I for some nonzero r then v ∈ I. Here we will be interested in a different class of ideals, namely those that contain no elements of the form rv. Definition 3.2. An ideal I ⊆ LR(E) is called vertex free if for r ∈ R and v ∈ E0 we have rv ∈ I =⇒ r = 0. We now prove some structure results for vertex free ideals. We follow the proof of the Cuntz-Krieger Uniqueness Theorem [29, Theorem 6.5], but we keep track of what happens at cycles without exits. Our goal is to show that, in a certain sense, vertex free ideals come from cycles without exits, see Proposition 3.5 for a precise statement. In the following two lemmas E is a graph, R is a commutative ring with unit and we work in LR(E). Lemma 3.3. For every nonzero vertex free ideal I there exists a vertex u ∈ E0, cycles β1, β2, . . . , βn based at u, and elements r0, r1, . . . , rn ∈ R \ {0} such that 0 (cid:54)= r0u + riβi ∈ I. n(cid:88) i=1 m(cid:88) j=1 k(cid:88) l=1 Proof. Since I is nonzero, there is a nonzero x ∈ I. By [29, Lemma 6.4] we can assume that x is a polynomial in real edges. As E0 form a set of local units for LR(E), we can find some u ∈ E0 with y = ux (cid:54)= 0. Since I is vertex free, y has the form y = s0u + sjγj, for sj ∈ R, nonzero for j (cid:54)= 0, and γj distinct paths of positive length with source u. We now claim that there exists a nonzero z ∈ I of the form (3.1) z = t0u + tlβl, where t0, t1, . . . , tk are all nonzero, and βl distinct paths of positive length with source u. If s0 (cid:54)= 0, then z = y does the job. If s0 = 0, we relabel the terms of y so that γ1 ≤ γ2 ≤ ··· ≤ γm, and consider γ∗ m(cid:88) 1 y. Then sjγ∗ 1 γj. γ∗ 1 y = s1r(γ1) + Since I is vertex free, at least one of the γ∗ length among the γj, whenever γ∗ 1 γj is nonzero, and since γ1 has minimal 1 γj is nonzero it is a path of positive length with j=2 6 NATHAN BROWNLOWE AND ADAM P W SØRENSEN source r(γ1). Thus n(cid:88) rlβl γ∗ 1 y = s1r(γ1) + for some non-zero rl ∈ R and distinct paths βl with source r(γ1). By [29, Proposi- tion 4.9] E∗ is linearly independent in LR(E), so since s1 (cid:54)= 0 and since all the βl have positive length we have γ∗ 1 y (cid:54)= 0. Hence z = γ∗ 1 y has the desired form. Now, fix z of the form (3.1) and consider w = zu ∈ I. We have l=2 (cid:88) w = t0u + tlβl. {l:s(βl)=u=r(βl)} Since t0 (cid:54)= 0 and I is vertex free, the set {l : s(βl) = u = r(βl)} is nonempty and so w has the desired form. As we noted in our construction of z, since t0 (cid:54)= 0 and all the βl have positive length, we have w (cid:54)= 0. (cid:3) Lemma 3.4. Let β1, β2, . . . , βn be distinct cycles all based at some vertex u ∈ E0. If cycles based at u have exits, then there exists γ ∈ E∗ with s(γ) = u and such that γ∗βiγ = 0 for all i = 1, 2, . . . , n. Note that either all cycles based at a vertex u have exits, or no cycle based at u has exits, which explains the wording of the above lemma. In the latter case there is a cycle α of minimal length based at u and all other cycles are of the form αn. Proof of Lemma 3.4. Pick a cycle τ based at u of minimal length. Since τ has an exit we can write τ = µν for µ ∈ E∗, ν ∈ E∗ \ E0 with s−1(r(µ)) ≥ 2. Let f ∈ s−1(r(µ)) be distinct from the first edge in ν. Fix an m ∈ N such that τ m is longer than all the βi, and put γ = τ mµf. To compute γ∗βiγ we look at two cases. First suppose that βi is an initial segment of τ m. Since τ is a cycle of minimal length and βi is a cycle we have βi = τ k for some 1 ≤ k < m. Hence γ∗βiγ = f∗µ∗(τ∗)mτ kτ mµf = f∗µ∗τ kµf = f∗µ∗µντ k−1µf = f∗ντ k−1µf = 0, since f is distinct from the first edge in ν. Now suppose βi is not an initial segment of τ m. Then (τ m)∗βi = 0, and hence (cid:3) γ∗βiγ = 0. Proposition 3.5. Let E be a graph and R a commutative ring with unit. If I ⊆ LR(E) is a non-zero vertex free ideal, then I contains a nonzero element of the form s0u + s1αi, for some vertex u ∈ E0, a cycle with no exits α based at u, and s0, s1, . . . , sn ∈ R not all zero. Proof. Since I is vertex free we can apply Lemma 3.3 to find a nonzero w ∈ I of the form w = r0u + riβi, for some vertex u ∈ E0, cycles β1, β2, . . . , βn based at u, and r0, r1, . . . , rn ∈ R\{0}. If cycles based at u had exits, we could apply Lemma 3.4 to find a path γ ∈ E∗ with s(γ) = u and γ∗βiγ = 0 for all i = 1, 2, . . . , n. Hence γ∗wγ = r0γ∗uγ + riγ∗βiγ = r0r(γ), n(cid:88) i=1 n(cid:88) i=1 n(cid:88) i=1 LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R 7 which contradicts that I is vertex free. We conclude that cycles based at u do not have exits. Since this is the case, there is a cycle α based at u of minimal length. Furthermore all the βi must have the form βi = αki for some ki ∈ N. So w has the (cid:3) desired form (by padding the sum with zero elements if necessary). Proof of Theorem 3.1. The "only if" direction is immediate. For the "if" direction suppose that φ : LR(E) → S satisfies (G1) and (G2). Suppose for contradiction that I = ker φ is nonzero. We know from (G1) that I is a vertex free ideal. Then by Proposition 3.5 there is a nonzero element x ∈ I of the form n(cid:88) x = s0u + siαi, in R not all zero. This means the nonzero polynomial q(x) = (cid:80)n for some vertex u ∈ E0, a cycle with no exits α based at u, and elements s0, s1, . . . , sn i=0 sixi in R[x] i=1 satisfies φ(q(α)) = 0, which contradicts (G2). Hence we must have I = {0}, and φ is injective. (cid:3) We can rephrase the Generalised Cuntz-Krieger Uniqueness Theorem for ∗- homomorphisms into involutive R-algebras. It is this result we apply in the next section. Corollary 3.6. Let E be a graph, R a commutative ring with unit, and φ a ∗- homomorphism of LR(E) into an involutive R-algebra A. Then φ is injective if and only if the following conditions are satisfied: (1) φ(rv) (cid:54)= 0 for all v ∈ E0 and all r ∈ R \ {0}; and (2) φ(α) has full spectrum in A for every cycle α without an exit. Proof. Since for any cycle α without an exit, and for any q(x) ∈ R[x], we have φ(q(α)) = q(φ(α)), it follows that (2) is equivalent to (G2). The result now follows (cid:3) from Theorem 3.1. 4. Embedding LR(E) into L2,R The main result of this paper is the following. Theorem 4.1. Let R be a commutative ring with unit and let E be a directed graph. If E is countable, then there is a ∗-algebraic embedding of the Leavitt path algebra LR(E) into L2,R, and if E0 is finite, then this embedding can be chosen to be unital. If E is uncountable then there is no embedding of LR(E) into L2,R as R-modules. As discussed in the introduction, a dimension argument shows that LR(E) cannot embed into L2,R for uncountable E. The first step in proving that the countability of E is sufficient for such an embedding is to show that the unit in L2,R may be broken into countably many pairwise orthogonal projections that are all Murray-von Neumann equivalent to the unit. This will be done in Proposition 4.3, but first we prove a lemma. Lemma 4.2. Let R be a commutative ring with unit and let p ∈ L2,R be a projec- tion. If p ∼ 1, in the sense of Definition 2.3, then (i) rp (cid:54)= 0 for all r ∈ R \ {0} and (ii) pL2,Rp ∼= L2,R as ∗-algebras. 8 NATHAN BROWNLOWE AND ADAM P W SØRENSEN Proof. Since p ∼ 1 we can find a t ∈ L2,R such that t∗t = 1 and tt∗ = p. For (i) suppose that rp = 0 for some r ∈ R. Then 0 = t∗0t = t∗rpt = rt∗tt∗t = r1. By [29, Proposition 4.9] this implies that r = 0. The map φ : L2,R → L2,R given by φ(x) = txt∗ is a ∗-homomorphism. Since φ(1) = p it follows from (i) and the Cuntz-Krieger Uniqueness Theorem ([29, The- orem 6.5]) that φ is injective. As pt = t and t∗p = t∗ we have that the image of φ is contained in pL2,Rp, and as φ(t∗xt) = pxp for all x ∈ L2,R, the image of φ is all (cid:3) of pL2,Rp. Hence φ shows that L2,R ∼= pL2,Rp as ∗-algebras, i.e. (ii) holds. Proposition 4.3. Let R be a commutative ring with unit. For every nonempty subset I ⊆ N there exist pairwise orthogonal nonzero projections {pi : i ∈ I} ⊂ L2,R such that (i) pi ∼ 1, in the sense of Definition 2.3, for all i; (ii) rpi (cid:54)= 0 for all i and for all r ∈ R \ {0}; (iii) piL2,Rpi (iv) if I is finite, then {pi : i ∈ I} can be chosen so that(cid:80) ∼= L2,R for all i; and i∈I pi = 1 also holds. Proof. For each i ∈ I we define ti = aib and pi = tit∗ i . Since t∗ the pi satisfy (ii) and (iii). For i > j we have t∗ we have t∗ i ti = b∗(ai)∗aib = 1 we have that pi ∼ 1. It follows from Lemma 4.2 that i tj = b∗(ai−j)∗b = 0, and for i < j For (iv) we may assume that I = {1, . . . , n}. We will prove by induction on n i tj = b∗aj−ib = 0. Hence the pi are pairwise orthogonal. that there exist orthogonal projections pi satisfying (i)–(iv). If n = 1 then we simply put p1 = 1. Suppose now that our inductive hypotheses holds for all k ≤ n − 1 and pick projections q1, q2, . . . , qn−1 satisfying (i)–(iv). Let t be such that t∗t = 1 and tt∗ = qn−1. Define pi = qi if i < n − 1 and pn−1 = ta(ta)∗, pn = tb(tb)∗. Since (ta)∗(ta) = a∗t∗ta = a∗a = 1 it follows from Lemma 4.2 that pn−1 satisfy (i)–(iii). Similarly pn satisfies (i)–(iii). We see that pn−1pn = taa∗t∗tbb∗t∗ = taa∗bb∗t∗ = 0. For i < n − 1 we have pipn−1 = qitaa∗t∗ = qitt∗taa∗t∗ = qiqn−1taa∗t∗ = 0, and similarly pipn = 0. Thus p1, p2, . . . , pn are orthogonal projections, and since we have pn−1 + pn = taa∗t∗ + tbb∗t∗ = t(aa∗ + bb∗)t∗ = tt∗ = qn−1, n(cid:88) pi = 1. Which completes the induction and therefore shows that (iv) holds. i=1 (cid:3) To deal with graphs that have cycles without exits, we also need to find a full spectrum unitary in L2,R. Proposition 4.4. There exists a full spectrum unitary in L2,R. LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R 9 Proof. Define u = aaa∗ + aba∗b∗ + bb∗b∗. A routine calculation shows that u∗u = uu∗ = 1. To see that u has full spectrum, we note that ua = a2, so for any n ≥ 1 we have una = an+1. (4.1) Let q ∈ R[x] satisfy q(u) = 0. We aim to show that q must be the zero polynomial. We write k(cid:88) (cid:32) k(cid:88) n=0 n=0 q(x) = where each rn ∈ R. Using (4.1) we get rnxn, 0 = 0 · a = q(u) · a = rnun (cid:33) k(cid:88) a = rnan+1. n=0 Since we know from [29, Proposition 4.9] that {an : n ∈ N} is a linearly independent set, we have rn = 0 for 0 ≤ n ≤ k. So q(x) is the zero polynomial. Since no nonzero (cid:3) polynomial in u is zero we conclude that u has full spectrum. Remark 4.5. How does one arrive at the particular u presented in Proposition 4.4? The crucial point in proving that u has full spectrum was that u2a = a2. To get a unitary with this property, we first looked for a unital endomorphism φ of L2,R such that φ(a) = a2. It is well known, see e.g. [15], that endomorphisms of the Cuntz algebra O2 are in one-to-one correspondence with unitaries. Applying these ideas to Leavitt algebras we define u as u = φ(a)a∗ + φ(b)b∗. We arrived at our particular u by putting φ(b) = aba∗ + bb∗. Choosing endomor- phisms ψn of L2,R with ψn(a) = an one can construct many different full spectrum unitaries. Proposition 4.4 lets us embed LR[z, z−1] into L2,R. So if we let Cn denote the graph consisting of single vertex simple cycle of length n, then we have that LR(C1) embeds into L2,R. The next lemma gives a concrete embedding of LR(Cn) into L2,R for all n. Lemma 4.6. Given pairwise orthogonal projections p1, p2, . . . , pn in L2,R with pi ∼ 1 for 1 ≤ i ≤ n there exists t1, t2, . . . , tn in L2,R such that i ti = pi+1, for i < n, and t∗ i = pi, for all i, i ti = ti, for all i, and, (1) t∗ (2) tit∗ (3) tit∗ (4) v = t1t2 ··· tn is a full spectrum partial unitary with vv∗ = p1 = v∗v. ntn = p1, Proof. Because the projections are Murray-von Neumann equivalent, we can find t1, t2, . . . , tn−1 ∈ L2,R that satisfy (1)–(3) and an s ∈ L2,R such that ss∗ = pn and s∗s = 1. By Proposition 4.4 there is a full spectrum unitary(cid:101)u ∈ L2,R. Let u = s(cid:101)us∗ and put ∗ tn = u (t1t2 ··· tn−1) . Observe that u∗u = pn = uu∗. We check that tn satisfies (1)–(3). Since i = tipi+1t∗ titi+1t∗ i+1t∗ i = tit∗ i tit∗ i = tit∗ i , 10 NATHAN BROWNLOWE AND ADAM P W SØRENSEN for i < n − 1, we have that ntn = (t1t2 ··· tn−1) u∗u (t1t2 ··· tn−1) t∗ ··· t∗ 2t∗ = t1t2 ··· tn−2tn−1t∗ = t1t∗ n−1t∗ 1 = p1. n2 1 ∗ So tn satisfies (1). Similarly, we see that i titi+1 = t∗ i+1pi+1ti+1 = t∗ i+1ti+1t∗ i+1ti+1 = t∗ i+1ti+1, t∗ i+1t∗ for i < n − 1. Hence tnt∗ n2 ∗ (t1t2 ··· tn−1) u∗ n = u (t1t2 ··· tn−1) ··· t∗ 1t1t2 ··· tn−1u∗ = ut∗ 2t∗ = ut∗ = uu∗uu∗ = pn. n−1t∗ n−1tn−1u∗ = upnu∗ pnu = ss∗s(cid:101)us∗ = s(cid:101)us∗ = u, I.e. tn satisfies (2). Combining the above with the fact that we get that tnt∗ ntn = pntn = pnu (t1t2 ··· tn−1) ∗ = u (t1t2 ··· tn−1) ∗ So tn satisfies (3). Let v = t1t2 ··· tn and let t = t1t2 ··· tn−1. Then v = (ts)(cid:101)u(ts)∗. Computations = tn. like above show that ts(ts)∗ = tss∗t∗ = tpnt∗ = tt∗ = p1, and that Hence and (ts)∗ts = s∗t∗ts = s∗pns = s∗ss∗s = 1. vv∗ = (ts)(cid:101)u(ts)∗(ts)(cid:101)u∗(ts)∗ = (ts)(cid:101)u(cid:101)u∗(ts)∗ = ts(ts)∗ = p1, v∗v = (ts)(cid:101)u∗(ts)∗ts(cid:101)u(ts)∗ = (ts)(cid:101)u∗(cid:101)u(ts)∗ = ts(ts)∗ = p1. So v is a partial unitary. To see that v has full spectrum let q ∈ R[z] be such that q(v) = 0. Since vm = ((ts)(cid:101)u(ts)∗)m = (ts)(cid:101)um(ts)∗, and similarly we see that (v∗)m = (ts)((cid:101)u∗)m(ts)∗ 0 = q(v) = (ts)q((cid:101)u)(ts)∗. An argument similar to the proof of (ii) in Lemma 4.2 shows that conjugation by ts is an isomorphism of L2,R onto p1L2,Rp1. So we must have q((cid:101)u) = 0. Since (cid:101)u has full spectrum we conclude that q is the zero polynomial, and therefore v has (cid:3) full spectrum. Proof of Theorem 4.1. Arguing as in [5, Proposition 5 and Corollary 6], which is for fields but works for commutative rings, one sees that L2,R has countable dimension as an R-module. Suppose that E is uncountable, then E0 is uncountable or E1 is uncountable. If E0 is uncountable then it forms an uncountable set of linearly independent elements of LR(E). If E0 is at most countable and E1 is uncountable, then there is some vertex u ∈ E0 that emits uncountably many edges. In this case {ee∗ : s(e) = u} is an uncountable set of orthogonal projections, so it forms an uncountable set of linearly independent elements of LR(E). In both cases we have LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R 11 an uncountable collection of linearly independent elements in LR(E), hence there is no module embedding of LR(E) into L2,R. Suppose instead that E is countable. Since E0 is countable we can use Proposi- tion 4.3 to get pairwise orthogonal nonzero projections {pv : v ∈ E0} ⊂ L2,R such ∼= L2,R for all v ∈ E0. If E0 is finite, we know from (iv) that pv ∼ 1 and pvL2,Rpv of Proposition 4.3 that these projections can be chosen to sum to the identity. For ∼= L2,R, use Proposition 4.3 each v ∈ E0 with s−1(v) (cid:54)= ∅ we can, since pvL2,Rpv to choose pairwise orthogonal nonzero projections {qv e : e ∈ s−1(v)} ⊆ pvL2,Rpv such that qv e ∼ pv, and if s−1(v) < ∞, such that qv e . pv = (cid:88) e∈s−1(v) Then for each e ∈ E1 we have e ∼ ps(e) ∼ 1 ∼ pr(e). qs(e) e , t∗ ete = pr(e), and tet∗ If e is not part of a cycle without an exit we use the definition of ∼ to choose te such that tet∗ ete = te. For each vertex simple cycle α = e1e2 ··· en without an exit, we use Lemma 4.6 to pick elements tei such that tei = tei and te1 te2 ··· ten is a partial unitary teit∗ with full spectrum. e : e ∈ E1} form a Leavitt E-family. By We claim that {pv : v ∈ E0} and {te, t∗ e = qs(e) , t∗ tei = pr(ei), teit∗ = qs(ei) ei ei ei ei construction the pv are orthogonal projections. Furthermore we have ps(e)te = ps(e)tet∗ ete = ps(e)qs(e) e te = qs(e) e te = tet∗ ete = te and Applying the involution we get tepr(e) = tet∗ ete = te. For e (cid:54)= f we have and we have pr(e)t∗ e = t∗ eps(e) = t∗ e. t∗ etf = t∗ eqs(e) e qs(f ) f tf = 0, Finally, if 0 < s−1(v) < ∞ then(cid:88) t∗ ete = pr(e). (cid:88) tet∗ e = qs(e) e = ps(e). e∈s−1(v) e∈s−1(v) e : e ∈ E1} do indeed form a Leavitt E-family. So {pv : v ∈ E0} and {te, t∗ The universal property of LR(E) gives a ∗-homomorphism φ : LR(E) → L2,R which takes the vertex projections in LR(E) to the projections {pv : v ∈ E0}, and we know from part (iii) of Proposition 4.3 that rpv (cid:54)= 0 for each v and for all r ∈ R\{0}. Since for all cycles α without an exit the image φ(α) is a full spectrum partial unitary by construction, it now follows from Corollary 3.6 that φ is injective. Finally, in the case that E0 is finite, we chose the projections pv to sum to the v∈E0 pv = 1. So φ is unital, as claimed. (cid:3) identity. Then φ(1) = φ((cid:80) v∈E0 v) =(cid:80) Theorem 4.1 raises two natural questions. Firstly, keeping in mind that we are trying to get an algebraic version of the first of Kirchberg's Geneva Theorems, we wonder if there are any R-algebras with a countable basis that do not embed into L2,R. In [13] we show that L2,Z ⊗ L2,Z does not embed into L2,Z. However, we have no counterexamples outside of this specific case, so we ask the following question. 12 NATHAN BROWNLOWE AND ADAM P W SØRENSEN Question 4.7. Let K be a field. Does there exist a K-algebra A with a countable basis such that A does not embed into L2,K? The authors have been unable to come up with any such A, but we would find an affirmative answer to the above question quite surprising. The second natural question to ask was originally asked of us by the anonymous referee: is L2,R the only Leavitt path algebra that admits embeddings of all other Leavitt path algebras of countable graphs? We thank the referee for this question, as well as suggesting Corollaries 4.9 and 4.10 below. The answer to this question is no. As an easy example note that L2,R embeds unitally into L2,R ⊕ L2,R as a ∗-algebra by the map x (cid:55)→ (x, x). Hence by The- orem 4.1 all Leavitt path algebras of countable graphs embed (unitally, when it makes sense) into L2,R ⊕ L2,R as ∗-algebras. We end this section with a few re- sults about other Leavitt path algebras that admit embeddings of all Leavitt path algebras of countable graphs. Lemma 4.8. Let K be a field and let A be a unital K-algebra. Then L2,K embeds unitally into A (as a K-algebra) if and only if A has module type (1, 2). Proof. Recall (for instance from [22]) that A has module type (1, 2) if and only if there exists elements x1, x2, y1, y2 ∈ A such that y1x1 = 1 = y2x2 and x1y1 + x2y2 = 1. If such elements exist, then by the universal property of L2,K we can define a unital K-algebra homomorphism φ : L2,K → A. Since L2,K is simple it is necessarily an embedding. On the other hand, if L2,K embeds unitally into A, then the images of (cid:3) a, a∗, b, and b∗ will show that A has module type (1, 2). Corollary 4.9. Let K be a field, and let F be a graph with finitely many vertices. All Leavitt path algebras over countable graphs will embed (as K-algebras) into LK(F ) if and only if LK(F ) has module type (1, 2). Furthermore, if LK(F ) has module type (1, 2) then any unital Leavitt path algebra over a countable graph will embed unitally into LK(F ). Corollary 4.10. Let K be a field. Let F be a graph with finitely many vertices such that LK(E) is purely infinite simple and the class of the unit in K0(LK(F )) is 0. For any countable graph E the Leavitt path algebra LK(E) embeds into LK(F ) (as K-algebras), and if E has finitely many vertices, then the embedding can be chosen unital. Proof. We say that two idempotents e, f are equivalent if there exists elements x, y such that e = xy and f = yx, we denote this by e ≈ f . (For the relation between ∼ and ≈ see Remark 2.4.) LK(F ) is purely infinite, so we can find orthogonal idempotents e, f ∈ LK(f ) such that e ≈ 1, f (cid:54)= 0 and 1 = e + f . Since the class of 1 in K0(LK(F )) is 0, the class of f is also 0, so by [9, Proposition 2.2] f ≈ 1. Therefore we can find elements x1, x2, y1, y2 ∈ LK(F ) such that x1y1 = e, x2y2 = f, and y1x1 = 1 = y2x2. That is LK(F ) has module type (1, 2). The conclusion now follows from Corollary (cid:3) 4.9. There are many algebras that satisfy the conditions in Corollary 4.10, for ex- amples see for instance [2, Section 4]. We can even get algebras with non-trivial K-theory. We note that Corollary 4.10 only lets us conclude the existence of a K-algebra embedding, not, as in Theorem 4.1 a ∗-algebra embedding. The issue is that in LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R 13 definition of module type and K-theory for rings, we deal with equivalence of idem- potents rather than projections. However, in certain specific cases, we can directly prove that a ∗-algebra embedding exists. Example 4.11. Let R be a unital commutative ring and let F be the graph (4.2) u e f g h v We claim that for any countable graph E the Leavitt path algebra LR(E) embeds into LR(F ) (as ∗-algebras), and if E has finitely many vertices, then the embedding can be chosen unital. To see this, it suffices to find a unital ∗-algebra embedding of L2,R into LR(F ). Let s = e + g, t = f + v. Then we have that s∗s = (e∗ + g∗)(e + g) = e∗e + g∗g = u + v = 1, ss∗ = (e + g)(e∗ + g∗) = ee∗ + gg∗, t∗t = (f∗ + v)(f + v) = f∗f + v = u + v = 1, t∗t = (f + v)(f∗ + v) = f f∗ + v. By the universal property of L2,R we get a unital ∗-homomorphism φ : L2,R → LR(F ) such that φ(a) = s, φ(b) = t. By the Cuntz-Krieger Uniqueness Theorem ([29, Theorem 6.5]) φ is an embedding. Note that when R is a principal ideal domain it follows from [7, Corollary 7.7] that K0(LR(F )) = Z/2Z, and in particular it is non-zero. 5. Concrete embeddings The proof of Theorem 4.1 is constructive, in that it gives a recipe for how to construct embeddings of LR(E) into L2,R. In the case where E satisfies Condition (L), the recipe is as follows: 4.3). such that qv,e ∼ pv and such that pv =(cid:80) (1) Pick orthogonal projections {pv v ∈ E0} such that pv ∼ 1 (Proposition (2) For each v, that isn't a sink, pick orthogonal projections {qv,e e ∈ s−1(v)} (3) Pick partial isometries te such that t∗ e = qs(e),e (such (4) Then {pv, te} is a Cuntz-Krieger E-family in L2,R and the ∗-homomorphism partial isometries exists since qs(e),e ∼ pr(e)). s(e)=v qv,e (Proposition 4.3). ete = pr(e) and tet∗ they define from LR(E) to L2,R is injective. In the proof we use Proposition 4.3 to get the desired projections in L2,R, but we can often make easier choices in concrete cases. To help us do that we introduce the notion of a cylinder set. We denote the set of finite paths in the graph underlying L2,R by {a, b}∗, and the set of infinite paths by {a, b}N . For α ∈ {a, b}∗ and ξ ∈ denotes the obvious concatenation. For each path α ∈ {a, b}∗ {a, b}N we define the cylinder set of α, denoted Z(α), as , αξ ∈ {a, b}N Z(α) = {αµ µ ∈ {a, b}N} ⊆ {a, b}N , 14 NATHAN BROWNLOWE AND ADAM P W SØRENSEN It is a well known consequence of relation (5) in Definition 2.2 that if α1, α2, . . . , αn is a collection of paths with (cid:116)iZ(αi) = {a, b}N then And similarly if (cid:116)iZ(αi) = Z(β) for some path β then n(cid:88) n(cid:88) i=1 αiα∗ i = 1. αiα∗ i = ββ∗. If two paths α, β have disjoint cylinder sets then α∗β = 0. i=1 We can now describe a concrete embedding of finite graphs that satisfy Condition (L). Proposition 5.1. Let E be a finite graph that satisfies Condition (L). Suppose we are given paths • {αv ∈ {a, b}∗ v ∈ E0}, and • {βe ∈ {a, b}∗ e ∈ E1}, v∈E0 Z(αv) = {a, b}N e∈s−1(v) Z(βe) = Z(αv), for each v that is not a sink. , and such that • (cid:70) • (cid:70) Then pv = αvα∗ v, and te = βeα∗ r(e) form a Cuntz-Krieger E family in L2,R. Furthermore the ∗-homomorphism φ : LR(E) → L2,R given by is unital and injective. φ(v) = pv, and φ(e) = te, Proof. Since the cylinder sets Z(αv) are disjoint the pv are pairwise orthogonal projections, and since α∗ vαv = 1 we have that pv ∼ 1. For each e ∈ E1, we let Then βe witness the equivalence qs(e),e ∼ 1, so qs(e),e ∼ ps(e). Since the cylinder sets Z(βe) are disjoint the qv,e are orthogonal projections and we have qs(e),e = βeβ∗ e . (cid:88) e∈s−1(v) (cid:71) e∈s−1(v) qv,e = pv Z(βe) = Z(αv). as Then and This shows that pv, qv,e satisfies the first two points in the recipe. We now define te = βeα∗ r(e). t∗ ete = αr(e)β∗ e βeα∗ r(e) = αr(e)α∗ r(e) = pr(e), tet∗ e = βeα∗ r(e)αr(e)β∗ e = βeβ∗ e = qs(e),e. Hence the te satisfies the third point in the recipe. So by the fourth point {pv, te} is a Cuntz-Krieger E family, and the ∗-homomorphism they define is injective. (cid:3) We use this to give concrete embeddings of some known Leavitt path algebras. LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R 15 Example 5.2 (Laurent polynomials). We know from Proposition 4.4 that u = aaa∗ + aba∗b∗ + bb∗b∗ is a full spectrum unitary in L2,R. As noted earlier, this means we have an embedding LR[z, z−1] (cid:44)→ L2,R mapping the polynomial z to u. Example 5.3 (Ln,R). Recall that Ln,R is the Leavitt path algebra of the graph with one vertex and n loops. We call the vertex u and the loops e1, e2, . . . , en. We wish to use Proposition 5.1 to define an embedding. Since there is only one vertex we let α = be the empty path. We now need to choose n paths β1, β2, . . . , βn such that (cid:116)iZ(βi) = {a, b}N . One way to do this is to put βi = ai−1b, for i = 1, 2, . . . , n − 1, and βn = an−1. It now follows from Proposition 5.1 that the map φ : Ln,R → L2,R given on generators by φ(u) = αα∗ = 1, (cid:40) φ(ei) = βi = ai−1b, an−1, i = 1, 2, . . . , n − 1 i = n . is a unital ∗-homomorphic embedding. Example 5.4 (The line graphs). Let An be the "line graph" with n vertices. (5.1) u1 u2 u3 un−1 un e1 e2 en−1 Label the vertices and edges as above. We again wish to apply Proposition 5.1. This time we first need to find n paths α1, α2, . . . , αn such that (cid:116)iZ(αi) = {a, b}N . Similar to the above example we define (cid:40) αi = ai−1b, an−1, i = 1, 2, . . . , n − 1 i = n. . For each j = 1, 2, . . . , n − 1 we let βj = αi. Then we are in a position to apply Proposition 5.1, which tells us that the map φ : LR(An) → L2,R given on generators by φ(ui) = αiα∗ i , φ(ej) = αjα∗ j+1, is an injective, unital ∗-homomorphism. Example 5.5 (The Toeplitz algebra). Let T be the graph pictured below (5.2) e u v f We call the Leavitt path algebra LR(T ) a Toeplitz algebra. To embed LR(T ) into L2,R we define α1 = a, α2 = b, β1 = aa, and, β2 = ab. and Z(β1) (cid:116) Z(β2) = Z(α1), so by Proposition 5.1 Then Z(α1) (cid:116) Z(α2) = {a, b}N the map φ : LR(T ) → L2,R given on generators by φ(u) = aa∗, φ(v) = bb∗, φ(e) = aaa∗ is a unital ∗-homomorphic embedding. and φ(f ) = abb∗, 16 NATHAN BROWNLOWE AND ADAM P W SØRENSEN Acknowledgements The authors thank Efren Ruiz for pointing out the reference [5] to them. The second-named author is grateful to Wojciech Szyma´nski for enlightening conversa- tions about endomorphisms of O2. The authors thank Gene Abrams, Pere Ara, and Enrique Pardo for helpful com- ments on an earlier version of this paper that helped improve it. During work on this project the first named author visited the University of Oslo. The first named author would like to thank the Institute for Mathematics & Its Applications at the University of Wollongong for providing financial support, and the University of Oslo and Nadia Larsen for also providing financial support, and for their hospitality. Part of this work was done while the second named author was supported by an individual post doctoral grant from the Danish Council for Independent Research Natural Sciences. References [1] G. Abrams. Leavitt path algebras: the first decade. Bull. Math. Sci., 5(1):59–120, 2015. [2] G. Abrams, P. N. ´Anh, A. Louly and E. Pardo. The classification question for Leavitt path algebras J. Algebra, 320(5):1983–2026, 2008. [3] G. Abrams and G. Aranda Pino. The Leavitt path algebra of a graph. J. Algebra, 293(2):319– 334, 2005. [4] G. Abrams and G. Aranda Pino. The Leavitt path algebras of arbitrary graphs. Houston J. Math., 34(2):423–442, 2008. [5] G. Abrams and K. M. Rangaswamy. Row-infinite equivalents exist only for row-countable graphs. In New trends in noncommutative algebra, volume 562 of Contemp. Math., pages 1–10. Amer. Math. Soc., Providence, RI, 2012. [6] P. Ara. Matrix rings over ∗-regular rings and pseudorank functions. Pacific J. Math., 129(2):209–241, 1987. [7] P. Ara, M. Brustenga and G. Cortinas. K-theory of Leavitt path algebras. Munster J. Math., 2:5–33, 2009. [8] P. Ara and G. Cortinas. Tensor products of Leavitt path algebras. Proc. Amer. Math. Soc., 141(8):2629–2639, 2013. [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. Repre- sent. Theory, 10(2):157–178, 2007. [11] G. Aranda Pino, D. Mart´ın Barquero, C. Mart´ın Gonz´alez, and M. Siles Molina. The socle of a Leavitt path algebra. J. Pure Appl. Algebra, 212(3):500–509, 2008. 1977. Institute Publications. Springer-Verlag, New York, 1986. [12] B. Blackadar. K-theory for operator algebras, volume 5 of Mathematical Sciences Research [13] N. Brownlowe and A. P. W. Sørensen. L2,Z ⊗ L2,Z does not embed in L2,Z. [14] J. Cuntz. Simple C∗-algebras generated by isometries. Comm. Math. Phys., 57(2):173–185, [15] J. Cuntz. Automorphisms of certain simple C∗-algebras. In Quantum fields-algebras, pro- cesses (Proc. Sympos., Univ. Bielefeld, Bielefeld, 1978), pages 187–196. Springer, Vienna, 1980. [16] J. Cuntz and W. Krieger. A class of C∗-algebras and topological Markov chains. Invent. [17] S. Eilers, G. Restorff, and E. Ruiz. Classifying C∗-algebras with both finite and infinite [18] M. Enomoto and Y. Watatani. A graph theory for C∗-algebras. Math. Japon., 25(4):435–442, [19] E. Kirchberg. The classification of purely infinite simple C∗-algebras using Kasparov's theory. [20] E. Kirchberg and N. C. Phillips. Embedding of exact C∗-algebras in the Cuntz algebra O2. subquotients. J. Funct. Anal., 265(3):449–468, 2013. Math., 56(3):251–268, 1980. 1980. 3rd draft, 1994. J. Reine Angew. Math., 525:17–53, 2000. [21] A. Kumjian, D. Pask, I. Raeburn, and J. Renault. Graphs, groupoids, and Cuntz-Krieger algebras. J. Funct. Anal., 144(2):505–541, 1997. [22] W. G. Leavitt. The module type of a ring. Trans. Amer. Math. Soc., 103:113–130, 1962. LEAVITT R-ALGEBRAS OVER COUNTABLE GRAPHS EMBED INTO L2,R 17 [23] G. Nagy and S. Reznikoff, Abelian core of graph algebras. J. Lond. Math. Soc., 85:889–908, [24] N. C. Phillips. A classification theorem for nuclear purely infinite simple C∗-algebras. Doc. 2012. Math., 5:49–114, 2000. [25] K. M. Rangaswamy. On generators of two-sided ideals of Leavitt path algebras over arbitrary graphs. Comm. Algebra, 42(7):2859–2868, 2014. [26] M. Rørdam. A short proof of Elliott's theorem: O2 ⊗ O2 ∼= O2. C. R. Math. Rep. Acad. Sci. Canada, 16(1):31–36, 1994. [27] E. Ruiz, A. Sims, and A. P. W. Sørensen. UCT-Kirchberg algebras have nuclear dimension one. arXiv:1406.2045, 2014. [28] W. Szyma´nski. General Cuntz-Krieger uniqueness theorem. Internat. J. Math., 13(5):549– 555, 2002. [29] M. Tomforde. Leavitt path algebras with coefficients in a commutative ring. J. Pure Appl. Algebra, 215(4):471–484, 2011. (Nathan Brownlowe) School of Mathematics and Applied Statistics, University of Wollongong, Australia E-mail address: [email protected] (Adam P W Sørensen) Department of Mathematics, University of Oslo, Norway, and School of Mathematics and Applied Statistics, University of Wollongong, Australia E-mail address: [email protected]
1502.06097	1	1502	2015-02-21T11:57:49	A note on bilateral semidirect product decompositions of some monoids of order-preserving partial permutations	[ "math.RA" ]	In this note we consider the monoid $\mathcal{PODI}_n$ of all monotone partial permutations on $\{1,\ldots,n\}$ and its submonoids $\mathcal{DP}_n$, $\mathcal{POI}_n$ and $\mathcal{ODP}_n$ of all partial isometries, of all order-preserving partial permutations and of all order-preserving partial isometries, respectively. We prove that both the monoids $\mathcal{POI}_n$ and $\mathcal{ODP}_n$ are quotients of bilateral semidirect products of two of their remarkable submonoids, namely of extensive and of co-extensive transformations. Moreover, we show that $\mathcal{PODI}_n$ is a quotient of a semidirect product of $\mathcal{POI}_n$ and the group $\mathcal{C}_2$ of order two and, analogously, $\mathcal{DP}_n$ is a quotient of a semidirect product of $\mathcal{ODP}_n$ and $\mathcal{C}_2$.	math.RA	math	A note on bilateral semidirect product decompositions of some monoids of In this note we consider the monoid PODI n of all monotone partial permutations on {1, . . . , n} and its submonoids DP n, POI n and ODP n of all partial isometries, of all order-preserving partial permutations and of all order-preserving partial isometries, respectively. We prove that both the monoids POI n and ODP n are quotients of bilateral semidirect products of two of their remarkable submonoids, namely of extensive and of co-extensive transformations. Moreover, we show that PODI n is a quotient of a semidirect product of POI n and the group C2 of order two and, analogously, DP n is a quotient of a semidirect product of ODP n and C2. 2010 Mathematics subject classification: 20M20, 20M07, 20M10, 20M35 Keywords: transformations, partial isometries, order-preserving, semidirect products, pseudovarieties. ∗This work was developed within the FCT Project PEst-OE/MAT/UI0143/2014 of CAUL, FCUL, and of Departamento de Matem´atica da Faculdade de Ciencias e Tecnologia da Universidade Nova de Lisboa. †This work was developed within the FCT Project PEst-OE/MAT/UI0143/2014 of CAUL, FCUL, and of Instituto Superior de Engenharia de Lisboa. Abstract July 1, 2018 V´ıtor H. Fernandes∗ and Teresa M. Quinteiro† order-preserving partial permutations K R A M R E T A W S −→ S s 7−→ u.s T −→ T u 7−→ us ϕ : S −→ T (T ) δ : T −→ T (S) s 7−→ ϕs : u 7−→ δu : 1 Introduction and preliminaries Strongly motivated by automata theoretic ideas, in [25] Kunze studied the notion of bilateral semidirect product of two semigroups (see [26, 27] for applications in Automata Theory) and proved in [28] that the full transfor- mation semigroup on a finite set X is a quotient of a bilateral semidirect product of the symmetric group on X and the semigroup of all order-preserving full transformations on X, for some linear order on X. Also in [28], Kunze showed that the semigroup of all order-preserving full transformations on a finite chain is a quotient of a bilateral semidirect product of two of its subsemigroups. These results as well as applications to Formal Languages were also discussed by Kunze in [29]. Bilateral semidirect products were also considered by Lavers [32] who gave conditions under which a bilateral semidirect product of two finitely presented monoids is itself finitely presented, by exhibiting explicit presentations, under some conditions. In this note we construct bilateral semidirect decompositions, i.e. a representation of monoid S as a quotient of a bilateral semidirect product of two proper submonoids of S, of certain monoids of partial permutations. Denote by T (X) the semigroup (under composition) of all full transformations of a set X. Let S and T be two semigroups. Let be an anti-homomorphism of semigroups (i.e. (uv).s = u.(v.s), for s ∈ S and u, v ∈ T ) and let 5 1 0 2 b e F 1 2 ] . A R h t a m [ 1 v 7 9 0 6 0 . 2 0 5 1 : v i X r a ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a be a homomorphism of semigroups (i.e. usr = (us)r, for s, r ∈ S and u ∈ T ) such that: (SPR) (uv)s = uv.svs, for s ∈ S and u, v ∈ T (Sequential Processing Rule); and (SCR) u.(sr) = (u.s)(us.r), for s, r ∈ S and u ∈ T (Serial Composition Rule). Within these conditions, we say that δ is a left action of T on S and that ϕ is a right action of S on T . In [25], Kunze proved that the set S × T is a semigroup with respect to the following multiplication: for s, r ∈ S and u, v ∈ T . We denote this semigroup by Sδ ⋊⋉ ϕT (or, if it is not ambiguous, simply by S ⋊⋉ T ) and call it the bilateral semidirect product of S and T associated with δ and ϕ. If S and T are monoids and the actions δ and ϕ preserve the identity (i.e. 1.s = s, for s ∈ S, and u1 = u, for u ∈ T ) and are monoidal (i.e. u.1 = 1, for u ∈ T , and 1s = 1, for s ∈ S), then S ⋊⋉ T is a monoid with identity (1, 1). Here, we will just consider bilateral semidirect products of monoids associated to monoidal actions. Notice that, if the right action ϕ is a trivial action (i.e. (S)ϕ = {idT }) then S ⋊⋉ T = S ⋊ T is an usual semidirect product, if the left action δ is a trivial action (i.e. (T )δ = {idS}) then S ⋊⋉ T coincides with a reverse semidirect product S ⋉T and if both actions are trivial then S ⋊⋉ T is the usual direct product S × T . Observe also that the bilateral semidirect product is quite different from the double semidirect product by Rhodes and Tilson [35], wherein the second components multiply always as in the direct product. A partial transformation s on the chain Xn = {1 < 2 < · · · < n}, n ∈ N, is said to be order-preserving (respectively, order-reversing) if i ≤ j implies is ≤ js (respectively, is ≥ js) for all i, j ∈ Dom(s). Order- preserving and order-reversing partial transformations are also called monotone. Semigroups of order-preserving transformations have been considered in the literature since the 1960s. In 1962, Aızenstat [1] and Popova [34] exhibited presentations for On, the monoid of all order-preserving full transformations on Xn, and for POn, the monoid of all order-preserving partial transformations on Xn, In 1971, Howie [23] studied some combinatorial and algebraic properties of On and, in 1992, respectively. together with Gomes [20] revisited the monoids On and POn. More combinatorial properties of these two monoids were presented by Laradji and Umar in [30, 31]. Certain classes of divisors of the monoid On were determined in 1995 by Higgins [21] and by Vernitskiı and Volkov [36], in 1997 by Fernandes [9] and in 2010 by Fernandes and Volkov [18]. In [28] Kunze proved that the monoid On is a quotient of a bilateral semidirect product of its subsemigroups O− n = {s ∈ On i ≤ is, for i ∈ Xn}. See also [29, 15, 16]. (s, u)(r, v) = (s(u.r), urv), K R A M R E T A W n = {s ∈ On is ≤ i, for i ∈ Xn} and O+ The study of semigroups of finite partial isometries was initiated by Al-Kharousi et al. in [2, 3]. The first of these two papers was dedicated to investigate some combinatorial properties of the monoid DP n of all partial isometries on Xn and of its submonoid ODP n of all order-preserving partial isometries, in particular, their cardinalities. The second one presented the study of some of their algebraic properties, namely Green's structure and ranks. On the other hand, in [17] the authors exhibited presentations for both monoids DP n and ODP n. Observe that ODP n, POI n, DP n and PODI n are all inverse submonoids of the symmetric inverse monoid (i.e. the monoid of all partial permutations) In on Xn (see [3, 14]). Obviously, POI n ⊆ PODI n and The injective counterpart of On, i.e. the monoid POI n of all injective members of POn, has been object of study by the first author in several papers [9, 10, 11, 12, 13], by Derech in [8], by Cowan and Reilly in [5], by Ganyushkin and Mazorchuk in [19], among other authors. Presentations for the monoid POI n and for its extension PODI n, the monoid of all monotone partial permutations on Xn, were given by Fernandes [11] in 2001 and by Fernandes et al. [14] in 2004, respectively. The first author together with Delgado [6, 7] have also computed the abelian kernels of the monoids POI n and PODI n. Next, let s be a partial permutation on Xn. We say that s is an isometry if is − js = i − j, for all i, j ∈ Dom(s). ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a 2 ODP n = DP n ∩ POI n and, as observed by Al-Kharousi et al. [3], we also have DP n ⊆ PODI n. Moreover, it is easy to check that ODP n = {s ∈ In is − js = i − j, for i, j ∈ Dom(s)}. n = {s ∈ POI n i ≤ is, for i ∈ Dom(s)} and POI − In this paper, in Section 1, we obtain a bilateral semidirect decomposition of POI n in terms of its submonoids POI + n = {s ∈ POI n is ≤ i, for i ∈ Dom(s)} of extensive and of co-extensive transformations, respectively. A similar decomposition is constructed for the monoid ODP n by considering its submonoids ODP + n . On the other hand, in Section 2, we prove that PODI n and DP n are quotients of semidirect products of the form POI n ⋊C2 and ODP n⋊C2, respectively, where C2 denotes the group of order two. In both sections we extract consequences for pseudovarieties generated by some of these families of partial permutations monoids. n = ODP n ∩ POI − n and ODP − Recall that a pseudovariety of monoids is a class of finite monoids closed under formation of finite direct products, submonoids and homomorphic images. The semidirect product V⋊W of the pseudovarieties of monoids V and W is the pseudovariety generated by all monoidal semidirect products M ⋊N , where M ∈ V and N ∈ W. Similarly, we define the reverse semidirect product V ⋉ W and the bilateral semidirect product V ⋊⋉ W of the pseudovarieties of monoids V and W. n and O− n are isomorphic monoids, by {O− Let O and J be the pseudovarieties of monoids generated by {On n ∈ N} and by {O+ n n ∈ N} (or, since O+ It is well-known that J is the pseudovariety of J-trivial monoids and that it also is generated by the syntactic monoids of piecewise testable languages (see e.g. [33]). Let A be the pseudovariety of all aperiodic (i.e. H-trivial) monoids. It is easy to show that J ⋊⋉ J ⊆ A and, as an immediate consequence of Kunze's result [28] above mentioned, we have O ⊆ J ⋊⋉ J (see [15]). On the other hand, let Ecom be the pseudovariety of all idempotent commuting monoids (recall that a celebrated Theorem of Ash [4] states that Ecom is generated by all finite inverse monoids) and let POI and PODI be the pseudovarieties generated by {POI n n ∈ N} and by {PODI n n ∈ N}, respectively. Notice that POI ⊂ O ⊂ A [9] and that J ∩ Ecom is the pseudovariety generated by {POI + n and POI − n n ∈ N}) [22]. Finally, consider the pseudovariety of monoids Ab2 generated by C2 (a pseudovariety of Abelian groups). n are isomorphic monoids, by {POI − n n ∈ N} (or, since POI + n = ODP n ∩ POI + n n ∈ N}), respectively. K R A M R E T A W Dom(us) = {1, . . . , Im(us)} Im(u.s) = {1, . . . , Dom(us)}, and us ∈ POI + n . Im(u.s) = Dom(us) Dom(u.s) = Dom(us) Im(us) = Im(us). u.s ∈ POI − n and 11 = 1. n ⋊⋉ POI + n . 1s = 1, and 1.s = s, u1 = u, u.1 = 1, 1.1 = 1 and For for basic notions on Semigroup Theory, we refer the reader to Howie's book [24]. For simplicity, from now on we consider n ≥ 3. Observe that any element of POI n is well defined by its domain and image. Notice that 1 On the monoids POI n and ODP n In this section we show that POI n and ODP n are homomorphic images of certain bilateral semidirect products of the form POI − n ⋊⋉ POI + n and ODP − n , respectively. n ⋊⋉ ODP + We begin by constructing a bilateral semidirect product POI − Let s, u ∈ POI n \ {1}. Define the elements u.s, us ∈ POI n by ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a and, clearly, Define also 3 Consider the following two (well defined) functions: and We have: Lemma 1.1. The above defined functions δ and ϕ are an anti-homomorphism of monoids and a homomorphism of monoids, respectively. Proof. First, notice that δ1 and ϕ1 are, clearly, the identity maps of POI − n and of POI + n , respectively. Now, let u, v ∈ POI + n and s, r ∈ POI − It is immediate that, if any of the elements u, v or s is the identity, then (uv).s = u.(v.s), and if any of the elements u, s or r is the identity, then usr = (us)r. So, let us suppose that none of the elements u, v, s and r is the identity. In order to prove that (uv).s = u.(v.s), it suffices to show that Dom((uv).s) = Dom(u.(v.s)). In fact, we have On the other hand, in order to prove that usr = (us)r, it suffices to show that Im(usr) = Im((us)r): Im(usr) = Im(u(sr)) = Im((us)r) = (Im(us) ∩ Dom r)r = (Im(us) ∩ Dom r)r = Im(usr) = Im((us)r), Before proving that δ and ϕ also verify sequential processing and serial composition rules, we observe that (1) δ : POI + n −→ T (POI − n ) 7−→ δu : POI − n −→ POI − n u s s u 7−→ u.s 7−→ us. ϕ : POI − (uv).s = u.(v.s) and usr = (us)r. n . Then, we must prove that n −→ T (POI + n ) 7−→ ϕs : POI + n −→ POI + n Dom((uv).s) = Dom((uv)s) = Dom(u(vs)) = (Im(u) ∩ Dom(vs))u−1 = (Im(u) ∩ Dom(v.s))u−1 = Dom(u(v.s)) = Dom(u.(v.s)). K R A M R E T A W n and u, v ∈ POI + n . Then: (u.s)us = us, 4 as required. it is easy to check the equality for all u ∈ POI + n and s ∈ POI − n . Lemma 1.2. Let s, r ∈ POI − (SPR) (uv)s = uv.svs; (SCR) u.(sr) = (u.s)(us.r). ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a Proof. (SPR) We begin by noticing that if any of the elements s, u or v is the identity then the equality (uv)s = uv.svs is obvious. Thus, admit that none of the elements s, u or v is the identity. Since Im(u(v.s)) ⊆ Im(v.s) and taking in account Lemma 1.1, we obtain Then, in particular, Im((uv)s) = Im(uv.svs). Hence, in order to prove that (uv)s = uv.svs, it suffices to show, for instance, the inclusion Im(uv.svs) ⊆ Im((uv)s). Let y ∈ Im(uv.svs). Then there exists x ∈ Dom(uv.svs) such that y = x(uv.svs). It follows that xuv.s ∈ Im(uv.s) = Im(u(v.s)) and so xuv.s = a(u(v.s)), for some a ∈ Dom(u(v.s)). Thus, by using (1), we have y = (xuv.s)vs = (a(u(v.s)))vs = a(u((v.s)vs)) = a(u(vs)) = a((uv)s) ∈ Im((uv)s) = Im((uv)s), which proves the required inclusion. (SCR) As for (SPR), if any of the elements s, r or u is the identity then the equality u.(sr) = (u.s)(us.r) In view of the inclusion is trivial. Therefore, let us assume that none of these elements is the identity. Dom(usr) ⊆ Dom(us) and of Lemma 1.1, we have Dom(uv.svs) = (Im(uv.s) ∩ Dom(vs))(uv.s)−1 = (Im(u(v.s)) ∩ Im(v.s))(uv.s)−1 = (Im(u(v.s)))(uv.s)−1 = (Im(uv.s))(uv.s)−1 = Dom(uv.s) = Im(u.(v.s)) = Im((uv).s) = Dom((uv)s). Im((u.s)(us.r)) = (Im(u.s) ∩ Dom(us.r))(us.r) = (Dom(us) ∩ Dom(usr))(us.r) = (Dom(usr))(us.r) = (Dom(us.r))(us.r) = Im(us.r) = Dom((us)r) = Dom(usr) = Im(u.(sr)). K R A M R E T A W n is an aperiodic monoid. On the other hand, POI − 1(cid:1) and f = (cid:0)12 n . As (u.r)ur = ur, we have n −→ POI n 7−→ su . n ⋊⋉ POI + (s, u) µ : POI − ((s, u)(r, v))µ = (s(u.r), urv)µ = s(u.r)urv = surv = (s, u)µ(r, v)µ, 5 It follows, in particular, that Dom((u.s)(us.r)) = Dom(u.(sr)) and so it remains to show, for instance, that Dom((u.s)(us.r)) ⊆ Dom(u.(sr)). Let x ∈ Dom((u.s)(us.r)). Then x(u.s) ∈ Dom(us.r) = Dom(usr), whence x(u.s)us ∈ Dom(r) and so, by (1), x(us) = x(u.s)us ∈ Dom(r). Thus x ∈ Dom((us)r) = Dom(u(sr)) = Dom(u.(sr)), as required. Now, by Lemma 1.1 and Lemma 1.2, we can consider the bilateral semidirect product POI − n ⋊⋉ POI + n associated with δ and ϕ. Since POI − n are monoids and the actions δ and ϕ preserve the identity and are monoidal, then POI − n ⋊⋉ POI + n and POI + n are (isomorphic) J-trivial monoids and any bilateral semidirect product of J-trivial monoids is an aperiodic semigroup, whence POI − n is not regular n and POI + n is also a monoid. Moreover, as we already observed, POI − n ⋊⋉ POI + and is not an idempotent commuting semigroup. For instance, if e = (cid:0)1 show that (e, ∅) is not regular, (1, e) and (f, f ) are idempotents and (1, e)(f, f ) = (e, e) 6= (f, e) = (f, f )(1, e). n ⋊⋉ POI + 12(cid:1), it is routine matter to Next, consider the following function ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a Let (s, u), (r, v) ∈ POI − n ⋊⋉ POI + and so µ is a homomorphism. In addition, given t ∈ POI n, we may define elements s ∈ POI − by n and u ∈ POI + n Dom(s) = Dom(t), Im(s) = {1, . . . , Dom(t)} = Dom(u) and Im(u) = Im(t), and we obtain t = su = (s, u)µ. Hence µ is onto homomorphism and we have: Theorem 1.3. The monoid POI n is a homomorphic image of POI − As an immediate consequence of this result and the above observed fact that POI − n ⋊⋉ POI + n 6∈ Ecom, we have the following property: Corollary 1.4. POI ( (J ∩ Ecom) ⋊⋉ (J ∩ Ecom) 6⊆ Ecom. Next, we construct a bilateral semidirect product ODP − n , just by slightly modifying the definition of the previous actions. Although with different meanings, we will use the same notations in this new context. Let s, u ∈ ODP n \ {1} and suppose that Dom(us) = {i1, . . . , ik}, for some 1 ≤ i1 < · · · < ik ≤ n and n ⋊⋉ ODP + 0 ≤ k < n. Define the elements u.s, us ∈ POI n by Dom(u.s) = Dom(us) and Im(u.s) = {1, 1 + i2 − i1, . . . , 1 + ik − i1}, Dom(us) = {1, 1 + i2us − i1us, . . . , 1 + ikus − i1us} and Im(us) = Im(us) 1.s = s, u1 = u, u.1 = 1, n ⋊⋉ POI + n . u.s ∈ ODP − n Im(u.s) = Dom(us). and us ∈ ODP + n . K R A M R E T A W n −→ T (ODP − n ) 7−→ δu : ODP − n −→ ODP − n n −→ T (ODP + n ) 7−→ ϕs : ODP + n −→ ODP + n n , and we may consider the following two functions: n −→ ODP n 7−→ su , n ⋊⋉ ODP + (s, u) n on ODP − (u.s)us = us, ϕ : ODP − 1s = 1, 1.1 = 1 and 11 = 1. µ : ODP − δ : ODP + 7−→ us. 7−→ u.s u s s u 6 and ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a By exact replication of the proofs of Lemma 1.1 and Lemma 1.2, we may prove the following lemma: Lemma 1.5. The functions δ and ϕ are a monoidal left action of ODP + action of ODP − n , respectively. n on ODP + n and a monoidal right This lemma allows us to consider the bilateral semidirect product ODP − n associated with δ and ϕ, n , a non regular and non idempotent commuting aperiodic monoid. We may n ⋊⋉ ODP + n ⋊⋉ POI + which is, likewise POI − also consider the function (considering u.s = us = ∅ if us = ∅). Notice that, clearly, Moreover Define also for all u ∈ ODP + n and s ∈ ODP − As for the first studied case, it is easy to check the equality (2) which is, by (2), clearly a homomorphism. Moreover, let t ∈ ODP n be such that Dom(t) = {i1, . . . , ik}, for some 1 ≤ i1 < · · · < ik ≤ n and 0 ≤ k ≤ n, and define elements s ∈ ODP − n and u ∈ ODP + n by Im(s) = {1, 1 + i2 − i1, . . . , 1 + ik − i1} = {1, 1 + i2t − i1t, . . . , 1 + ikt − i1t} = Dom(u) Then t = su = (s, u)µ and so µ is onto homomorphism. Hence, we have the following result, with which we finish this section. Theorem 1.6. The monoid ODP n is a homomorphic image of ODP − n ⋊⋉ ODP + n . 2 On the monoids PODI n and DP n and Let ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a Then h ∈ DP n (and so h ∈ PODI n). Moreover, the identity (on Xn) and h are the only permutations of PODI n (and so of DP n). On the other hand, given α ∈ PODI n, it is clear that α is an order-reversing transformation if and only if hα (and αh) is an order-preserving transformation (see [14]). Hence, as α = h2α = h(hα), it follows that the monoids PODI n and DP n are generated by POI n ∪ {h} and ODP n ∪ {h}, respectively. Furthermore, we may see the cyclic group of order two C2 = {1, h} as a submonoid of both the monoids PODI n and DP n. Notice that, given x, y ∈ C2, we have xy = yx and x2 = y2 = 1. First, we turn our attention to the monoid PODI n. We obtain a semidirect decomposition of it in terms of its submonoids POI n and C2. For each x ∈ C2 and s ∈ POI n, define the element x.s = xsx ∈ POI n. Then, consider the function Dom(s) = Dom(t), 1 (cid:19) . Im(u) = Im(t). 2 · · · n − 1 n n n − 1 · · · 2 h = (cid:18) 1 K R A M R E T A W x 7−→ δx : POI n −→ POI n 7−→ x.s . µ : POI n ⋊C2 −→ PODI n δ : C2 −→ T (POI n) (s, x) 7−→ sx . s 7 Since (xy).s = xysxy = xysyx = x.(ysy) = x.(y.s) and 1.s = s, for x, y ∈ C2 and s ∈ POI n, then δ is an anti-homomorphism of monoids. On the other hand, for x ∈ C2 and s, r ∈ POI n, we have x.(sr) = xsrx = xs1rx = xsxxrx = (x.s)(x.r) and x.1 = x1x = x2 = 1. Thus δ induces a semidirect product POI n ⋊C2. It is easy to prove that POI n ⋊C2 is an inverse monoid. In fact, it is a routine matter to check that the idempotents of POI n⋊C2 commute (the idempotents of POI n⋊C2 are of the form (e, 1), with e an idempotent of POI n) and, given (s, x) ∈ POI n⋊C2, the element (xs−1x, x) of POI n⋊C2 is an (and so the) inverse of (s, x). Moreover, we have: Theorem 2.1. The monoid PODI n is a homomorphic image of POI n ⋊C2. Proof. Consider the function Then, for s, r ∈ POI n and x, y ∈ C2, we have ((s, x)(r, y))µ = (s(x.r), xy)µ = (sxrx, xy)µ = sxrx2y = sxry = (s, x)µ(r, y)µ. Thus µ is a homomorphism. On the other hand, let t ∈ PODI n. If t ∈ POI n then t = t1 = (t, 1)µ, otherwise th ∈ POI n and t = (th)h = (th, h)µ. Hence µ is surjective. Observe that, clearly, µ also separates idempotents, i.e. the restriction of µ to the set of the idempotents of POI n ⋊C2 is an injective function. The next result follows immediately from Theorem 2.1. Corollary 2.2. PODI ⊆ POI⋊Ab2. On the other hand, we also have: Lemma 2.3. POI n ⋊C2 ∈ PODI. Proof. It is easy to show that the function is an injective homomorphism. Supported by this result, we formulate the following conjecture: Conjecture 2.4. PODI = POI⋊Ab2. Notice that, since C2 is a commutative monoid, the left action of C2 on POI n may also be considered as a right action. Furthermore, similar results to Theorem 2.1 and Corollary 2.2 (and Lemma 2.3) also hold for reverse semidirect products. We finish this section by establishing the analogous result to Theorem 2.1 for the monoid DP n. This aim will be accomplish by noticing that DP n is a submonoid of PODI n that fits in the general framework described below. Let S be a monoid and let S1 and S2 be two submonoids of S. Let δ be a left action of S2 on S1 such that the function is a homomorphism. Let T be a submonoid of S, T1 a submonoid of S1 and T2 a submonoid of S2. It is a routine matter to check that, if (s)(u)δ ∈ T1, for all s ∈ T1 and u ∈ T2, then δ induces a (restriction) left action of T2 on T1 and the corresponding semidirect product T1⋊T2 is a submonoid of S1⋊S2. If, in addition, T = T1T2 then (s, x) 7−→ (sx, x) POI n ⋊C2 −→ PODI n × C2 K R A M R E T A W µT1⋊T2 : T1 ⋊T2 −→ T 7→ su µ : S1 ⋊S2 −→ S 7→ su (s, u) (s, u) 8 is a surjective homomorphism. For s ∈ ODP n and x ∈ C2, it is clear that x.s = xsx ∈ ODP n. Thus, we may consider the semidirect product ODP n ⋊ C2 induced by the left action δ of C2 on POI n. Moreover, since DP n = ODP nC2, then µODP n⋊C2 : ODP n ⋊ C2 −→ DP n is a surjective homomorphism and so we have: Theorem 2.5. The monoid DP n is a homomorphic image of ODP n ⋊C2. [1] A.Ya. Aızenstat, The defining relations of the endomorphism semigroup of a finite linearly ordered set, Sibirsk. Mat. 3 (1962), 161 -- 169 (Russian). [2] F. Al-Kharousi, R. Kehinde and A. Umar, Combinatorial results for certain semigroups of partial isometries of a finite chain, Australas. J. Combin. 58(3) (2014), 365 -- 375. References ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a [3] F. Al-Kharousi, R. Kehinde and A. Umar, On the semigroup of partial isometries of a finite chain, Com- munications in Algebra. To appear. [4] C.J. Ash, Finite semigroups with commuting idempotents, J. Austral. Math. Soc. Ser. A 43 (1987) 81 -- 90. [5] D.F. Cowan and N.R. Reilly, Partial cross-sections of symmetric inverse semigroups, Int. J. Algebra Com- put. 5 (1995) 259 -- 287. [6] M. Delgado and V.H. Fernandes, Abelian kernels of some monoids of injective partial transformations and an application, Semigroup Forum 61 (2000) 435 -- 452. [7] M. Delgado and V.H. Fernandes, Abelian kernels of monoids of order-preserving maps and of some of its extensions, Semigroup Forum 68 (2004) 335 -- 356. [8] V.D. Derech, On quasi-orders over certain inverse semigroups, Sov. Math. 35 (1991) No.3 74 -- 76; translation from Izv. Vyssh. Uchebn. Zaved., Mat. 1991 (1991) No.3 (346) 76 -- 78. [9] V.H. Fernandes, Semigroups of order-preserving mappings on a finite chain: a new class of divisors, Semi- group Forum 54 (1997) 230 -- 236. [10] V.H. Fernandes, Normally ordered inverse semigoups, Semigroup Forum 58 (1998) 418 -- 433. [11] V.H. Fernandes, The monoid of all injective order preserving partial transformations on a finite chain, Semigroup Forum 62 (2001) 178-204. [12] V.H. Fernandes, Semigroups of order-preserving mappings on a finite chain: another class of divisors, Izvestiya VUZ. Matematika 3 (478) (2002) 51 -- 59 (Russian). English translation in: Russ. Math. Izv. VUZ 46 No.3 47 -- 55 (2002) [13] V.H. Fernandes, Normally ordered semigroups, Glasg. Math. J. 50 (2008) 325 -- 333. K R A M R E T A W 9 [14] V.H. Fernandes, G.M.S. Gomes and M.M. Jesus, Presentations for some monoids of injective partial trans- formations on a finite chain, Southeast Asian Bull. Math. 28 (2004) no. 5 903 -- 918. [15] V.H. Fernandes and T.M. Quinteiro, Bilateral semidirect product decompositions of transformation monoids, Semigroup Forum 82 (2) (2011) 171 -- 187. [16] V.H. Fernandes and T.M. Quinteiro, On the monoids of transformations that preserve the order and a uniform partition, Comm. Algebra 39 (8) (2011) 2798 -- 2815. [17] V.H. Fernandes and T.M. Quinteiro, Presentations for monoids of finite partial isometries, submitted. [18] V.H. Fernandes and M. V. Volkov, On divisors of semigroups of order-preserving mappings of a finite chain, Semigroup Forum 81 (2010) 551 -- 554. [19] O. Ganyushkin and V. Mazorchuk, On the structure of IOn, Semigroup Forum 66 (3) (2003) 455 -- 483. [20] G.M.S. Gomes and J.M. Howie, On the ranks of certain semigroups of order-preserving transformations, Semigroup Forum 45 (1992) 272 -- 282. [21] P.M. Higgins, Divisors of semigroups of order-preserving mappings on a finite chain, Int. J. Algebra Comput. 5 (1995) 725 -- 742. [22] P.M. Higgins, Pseudovarieties generated by classes of transformation semigroups, Proc. St. Petersburgh Semigroup Conference Russian State Hydrometeorological Inst. (1999) 85 -- 94. ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a [27] M. Kunze, Lineare Parallelrechner II [Linear parallel processing machines II], Elektron. Informationsverarb. [28] M. Kunze, Bilateral semidirect products of transformation semigroups, Semigroup Forum 45 (1992) 166 -- [29] M. Kunze, Standard automata and semidirect products of transformation semigroups, Theoret. Comput. Kybernet. 20 (1984) 9 -- 39 (German). Kybernet. 20 (1984) 111 -- 147 (German). 182. Sci. 108 (1993) 151 -- 171. [23] J.M. Howie, Product of idempotents in certain semigroups of transformations, Proc. Edinburgh Math. Soc. 17 (1971) 223 -- 236. [24] J.M. Howie, Fundamentals of Semigroup Theory, Oxford, Oxford University Press, 1995. [25] M. Kunze, Zappa products, Acta Math. Hungar. 41 (1983) 225 -- 239. [26] M. Kunze, Lineare Parallelrechner I [Linear parallel processing machines I], Elektron. Informationsverarb. [30] A. Laradji and A. Umar, Combinatorial results for semigroups of order-preserving partial transformations, J. Algebra 278 (2004) No. 1 342 -- 359. [31] A. Laradji and A. Umar, Combinatorial results for semigroups of order-preserving full transformations, Semigroup Forum 72 (2006) No. 1 51 -- 62. [32] T.G. Lavers, Presentations of general products of monoids, J. Algebra 204 (1998) 733 -- 741. [33] J.-E. Pin, Varieties of Formal Languages, Plenum, London, 1986. [34] L.M. Popova, The defining relations of the semigroup of partial endomorphisms of a finite linearly ordered set, Leningradskij gosudarstvennyj pedagogicheskij institut imeni A. I. Gerzena, Uchenye Zapiski 238 (1962) 78 -- 88 (Russian). [35] J. Rhodes and B. Tilson, The kernel of monoid morphisms, J. Pure Appl. Algebra 62 (1989) 227 -- 268. [36] A.S. Vernitskiı and M.V. Volkov, A proof and generalisation of Higgins' division theorem for semigroups of order-preserving mappings, Izvestiya VUZ. Matematika No.1 (1995) 38 -- 44. (Russian). English translation in: Russ. Math. Izv. VUZ 39 No.1 34 -- 39 (1995) V´ıtor H. Fernandes, Departamento de Matem´atica, Faculdade de Ciencias e Tecnologia, Universidade Nova de Lisboa, Monte da Caparica, 2829-516 Caparica, Portugal; also: Centro de ´Algebra da Universidade de Lisboa, Av. Prof. Gama Pinto 2, 1649-003 Lisboa, Portugal; e-mail: [email protected] Teresa M. Quinteiro, Instituto Superior de Engenharia de Lisboa, Rua Conselheiro Em´ıdio Navarro 1, 1950-062 Lisboa, Portugal; also: Centro de ´Algebra da Universidade de Lisboa, Av. Prof. Gama Pinto 2, 1649-003 Lisboa, Portugal; e-mail: [email protected] K R A M R E T A W 10 ) 5 1 0 2 2 0 9 1 . d e t t i m b u s . 1 v ( v i X r a
1705.04157	1	1705	2017-05-11T13:17:09	Algebraic computation of genetic patterns related to three-dimensional evolution algebras	[ "math.RA" ]	The mitosis process of an eukaryotic cell can be represented by the structure constants of an evolution algebra. Any isotopism of the latter corresponds to a mutation of genotypes of the former. This paper uses Computational Algebraic Geometry to determine the distribution of three-dimensional evolution algebras over any field into isotopism classes and hence, to describe the spectrum of genetic patterns of three distinct genotypes during a mitosis process. Their distribution into isomorphism classes is also determined in case of dealing with algebras having a one-dimensional annihilator.	math.RA	math	Algebraic computation of genetic patterns related to three-dimensional evolution algebras1 O. J. Falc´on1, R. M. Falc´on2 J. N´unez3 1 , 3 Faculty of Mathematics, University of Seville, Spain. 2School of Building Engineering, University of Seville, Spain. E-mail: [email protected], [email protected], [email protected] Abstract. The mitosis process of an eukaryotic cell can be represented by the structure constants of an evolution algebra. Any isotopism of the latter corresponds to a mutation of genotypes of the former. This paper uses Computational Algebraic Geometry to determine the distribution of three-dimensional evolution algebras over any field into isotopism classes and hence, to describe the spectrum of genetic patterns of three distinct genotypes during a mitosis process. Their distribution into isomorphism classes is also determined in case of dealing with algebras having a one- dimensional annihilator. Keywords: Computational Algebraic Geometry, evolution algebra, clas- sification, 2000 MSC: 17D92, 68W30. isomorphism. isotopism, 1 Introduction Evolution algebras were introduced by Tian and Vojtechovsky [22, 23] to simulate algebraically the self-reproduction of alleles in non-Mendelian Ge- netics. In the last years, these algebras have been widely studied without probabilistic restrictions on their structure constants [3, 6, 10, 8, 13, 17, 19, 18, 21]. A main problem to be solved in this regard is the distribution of evolution algebras into isomorphism classes. Besides, Holgate and Campos [7, 16] exposed the importance in Genetics of considering also the distri- bution of such algebras under isotopism classes, because they constitute a way to formulate algebraically the mutation of genotypes in the inheritance process. The mentioned distribution into isomorphism classes has already been dealt with for two-dimensional evolution algebras over the complex field [5, 9] 1Preliminary version. Accepted manuscript for publication in Applied Mathematics and Computation. DOI: 10.1016/j.amc.2017.05.045. 1 and for nilpotent evolution algebras of dimension up to four over arbitrary fields [15]. More recently, the authors [14] have characterized the isomor- phism classes of two-dimensional evolution algebras over arbitrary fields and have established their distribution into isotopism classes. The latter gives rise to the spectrum of genetic patterns of two distinct genotypes during mitosis of eukaryotic cells. More specifically, if u and v are two distinct non-zero elements of the algebra under consideration, then the mentioned spectrum is formed by the next four genetic patterns: (0, 0), for which no offspring exists; (u, 0), for which exactly one of the two genotypes does not produce offspring; (u, u), for which the offspring has always the same geno- type, whatever the initial one is; and (u, v), for which the genotype of the offspring depends directly on that of the cell parent. This spectrum was obtained in [14] by means of distinct aspects on Computational Algebraic Geometry, all of them based on the fact that the algebraic law defined by the structure constants of any evolution algebra, together with the relations among basis vectors described by any isotopism of algebras, constitutes the set of generators of an ideal of polynomials whose reduced Grobner basis es- tablishes the algebraic relations that must hold, up to mutation, the genetic patterns of the mitosis process. This paper also focuses on the computation of such relations, particularly in case of dealing with three-dimensional evo- lution algebras. This enables us to distribute these algebras into isotopism classes, whatever the base field is, and describe mathematically the spectrum of genetic patterns of three distinct genotypes during a mitosis process. The distribution of such algebras into isomorphism classes is also determined in case of dealing with algebras having a one-dimensional annihilator. 2 Preliminaries In this section we expose some basic concepts and results on evolution al- gebras, isotopisms and Computational Algebraic Geometry that are used throughout the paper. For more details about these topics we refer the reader to the respective manuscripts of Tian [22], Albert [1] and Cox et al. [11]. 2.1 Evolution algebras A gene is the molecular unit of hereditary information. This consists of deoxyribonucleic acid (DNA), which contains the code to determine the at- tributes or phenotypes that characterize each organism. Genes are disposed in sequential order giving rise to long strands of DNA called chromosomes. 2 Genes related to a given phenotype can have distinct forms, which are called alleles. They always appear at the same position in chromosomes and con- stitute the genotype of the organism with respect to such a phenotype. In eukaryotic cells, DNA is primordially contained in the nucleus, al- though it also appears in organelles as mitochondria and chloroplasts, which are located in the cytoplasm. Mitosis is an asexual form of reproduction that consists of the division of an eukaryotic cell into two daughter cells in such a way that the nuclear genetic material of the former duplicates giving rise, up to rare mutation, to homologous nuclear chromosomes. In the final stage, the two daughter cells split having each resulting cell its corresponding copy of nuclear genetic material, whereas the extra-nuclear genetic material in the cytoplasm of the parent cell is randomly distributed between both of them by vegetative division. There exist distinct probabilistic laws that reg- ulate the theoretical influence of all of this genetic material in the genotype of the offspring. Tian and Vojtechovsky [23] introduced evolution algebras to represent mathematically these laws. Specifically, an evolution algebra defined on a set β = {e1, . . . , en} of distinct genotypes with respect to a given phenotype of an asexual organism is an n-dimensional algebra of basis j=1 cijej, for some ci1, . . . , cin ∈ K. The elements cij are called the structure constants of the algebra. Here, the product eiej = 0, for i 6= j, is due to the uniqueness of the genotype of the parent cell; the product eiei represents self-replication in the mitosis process; and each structure constant cij constitutes the probability, due to vegetative division, that the genotype ei becomes the genotype ej in the next generation. β over a field K such that eiej = 0, if i 6= j, and eiei =Pn 2.2 Isotopisms of evolution algebras Two n-dimensional algebras A and A′ are isotopic [1] if there exist three non- singular linear transformations f , g and h from A to A′ such that f (u)g(v) = h(uv), for all u, v ∈ A. The triple (f, g, h) is called an isotopism between A and A′. If f = g, then the triple (f, f, h) is called a strong isotopism and the algebras are said to be strongly isotopic. If f = g = h, then the isotopism constitutes an isomorphism, which is denoted by f instead of (f, f, f ). To be isotopic, strongly isotopic or isomorphic are equivalence relations among algebras. Throughout the paper, we refer the former and the latter as ∼ and ∼=, respectively. Isotopisms of evolution algebras can be interpreted as mutations of the genetic material of parent and daughter cells in the mitosis process with 3 respect to a given phenotype. Specifically, if (f, g, h) is an isotopism between two n-dimensional evolution algebras A and A′, then f and g represent the respective possible mutation of each one of the two homologous chromosomes in which the nuclear genetic material of the parent cell duplicates during the first part of the mitosis process, whereas h represents a possible mutation of the genotype of the offspring in the final step of the process. For each j=1 aijej involves the genotype ei to mutate to ej with probability aij. Since A′ is also an evolution algebra, the mitosis process only finishes if the genotypes of both homologous chromosomes that have been created after the mutations f and g coincide. If these genotypes do not coincide, then there is no offspring. α ∈ {f, g, h}, the corresponding expression α(ei) = Pn Hereafter, En(K) denotes the set of n-dimensional evolution algebras over the base field K with basis {e1, . . . , en}, whereas Tn(K) denotes the direct i=1h e1, . . . , en i. Every evolution algebra in En(K) is uniquely determined by an structure tuple T = (t1, . . . , tn) ∈ Tn(K), where ti = eiei, for all i ≤ n. The structure tuple T also determines the genetic pattern of the corresponding mitosis process. product Qn Proposition 2.1 ([14]) Let K be a field. The next results hold. a) Any two structure tuples in Tn(K) that are equal up to permutation of their components and basis vectors give rise to a pair of strongly isotopic evolution algebras. b) Let T be a structure tuple in Tn(K). There always exists a structure tuple T ′ = (Pn j=1 c1jej, . . . ,Pn j=1 cnjej) ∈ Tn(K) such that a) If cii = 0, for some i ≥ 1, then cjk = 0, for all j, k ≥ i. b) If cii 6= 0, for some i ≥ 1, then cij = 0, for all j 6= i. c) The evolution algebra in En(K) of structure tuple T ′ is strongly isotopic to that one of structure tuple T . Let A be an evolution algebra in En(K). Isotopisms preserve the di- mension of the derived algebra A2 = {uv u, v ∈ A} ⊆ A and that of the annihilator Ann(A) = {u ∈ A uv = 0, for all v ∈ A}. Hereafter, En;m(K) denotes the subset of evolution algebras in En(K) with an m-codimensional annihilator. Theorem 2.2 ([14]) Let K be a field. The next results hold. a) Let m < m′. None evolution algebra in En;m(K) is isotopic to an evolu- tion algebra in En;m′(K). 4 b) The set En;0(K) is only formed by the n-dimensional trivial algebra, which have all its structure constants equal to zero. c) Any algebra in En;1(K) is isotopic to the algebra described as e1e1 = e1. If the former is not isomorphic to the latter, then it is isomorphic to the evolution algebra described as e1e1 = e2. d) Any algebra in En;2(K) is isotopic to the algebra described as e1e1 = e2e2 = e1 or to that described as e1e1 = e1 and e2e2 = e2, and is isomor- phic to an evolution algebra in En;2(K) such that e1e1 ∈ {e1, e2, e1 + e2}. 2.3 Computational Algebraic Geometry Let I be an ideal of a multivariate polynomial ring K[X]. The algebraic set defined by I is the set of common zeros of all its polynomials. The largest monomial of a polynomial in I with respect to a given monomial term ordering is its leading monomial. The reduced Grobner basis of I is the only subset G of monic polynomials in I whose leading monomials generate the ideal also generated by all the leading monomials of I and such that no monomial of a polynomial in G is generated by the leading monomials of the rest of polynomials in the basis. This can always be computed from the sequential multivariate division of polynomials described by Buchberger's algorithm [2], which is extremely sensitive to the number of variables [20]. Computational Algebraic Geometry can be used to determine the set of isotopisms and isomorphisms between two evolution algebras A and A′ in En(K), with respective structure constants cij and c′ ij. To this end, we define the sets of variables Fn = {fij i, j ≤ n}, Gn = {gij i, j ≤ n} and Hn = {hij i, j ≤ n}, which play the role of the entries in the regular matrices related to a possible isotopism (f, g, h) between A and A′. The coefficients of each basis vector el in the expression f (ei)g(ej ) = h(eiej) enable us to ensure k=1 fikgjkc′ that Pn kl is equal to 0, if i 6= j, or to Pn k=1 hklcik, otherwise. Theorem 2.3 ([14]) The set of isotopisms and isomorphisms between A and A′ are respectively identified with a) The subset of zeros (f11, . . . , fnn, g11, . . . , gnn, h11, . . . , hnn) ∈ K3n2 in the algebraic set defined by the ideal IA,A′ of K[Fn ∪ Gn ∪ Hn] described as h n Xk=1 fikgjkc′ kl i, j, l ≤ n; i 6= j i + h fikgikc′ kl − n Xk=1 n Xk=1 hklcik i, l ≤ n i, giving rise to non-singular matrices F = (fij), G = (gij ) and H = (hij). 5 b) The subset of zeros (f11, . . . , fnn) ∈ Kn2 in the algebraic set defined by the ideal JA,A′ of K[Fn] described as h n Xk=1 fikfjkc′ kl i, j, l ≤ n; i 6= j i + h f2 ikc′ kl − n Xk=1 n Xk=1 fklcik i, l ≤ n i, giving rise to a non-singular matrix F = (fij). The computation of the reduced Grobner bases of both ideals in Theo- rem 2.3 has recently been implemented [14] in the open computer algebra system for polynomial computations Singular [12]. Particularly, in case of dealing with the finite field K = Fq, with q a power prime, the respec- tive complexity times that Buchberger's algorithm requires to compute the bases under consideration are qO(3n2) + O(n8) and qO(n2) + O(n8). We have made use of the mentioned procedure to determine all the isotopisms and isomorphisms that appear throughout this paper. The run time and memory usage that are required to compute each one of the reduced Grobner bases indicated in the paper are respectively 0 seconds and 0 MB in a computer system with an Intel Core i7-2600, with a 3.4 GHz processor and 16 GB of RAM, except for the computation related to Proposition 3.1.d, for which these two measures of computational efficiency have been 3 seconds and 1 MB, respectively. 3 Distribution of En;3(K) into isotopism classes Let n ≥ 3 and let K be a field. In order to determine the distribution of the set En;3(K) into isotopism classes, Proposition 2.1.b enables us to focus on those n-dimensional evolution algebras described as Aa,b :=  for some α, β ∈ K \ {0}, or e1e1 = e1, e2e2 = αe1, e3e3 = βe1, e1e1 = e1, e2e2 = e2, e3e3 = αe1 + βe2 + γe3, Bα,β,γ :=  for some (α, β, γ) ∈ K3 \ {(0, 0, 0)}. 6 Proposition 3.1 The next results hold. a) Aα,β ∼ A1,1, for all α, β ∈ K \ {0}. b) Let α, β, γ ∈ K be such that γ 6= 0. Then, Bα,β,γ ∼ B0,0,1. c) Let (α, β) ∈ K2 \ {(0, 0)}. Then, Bα,β,0 ∼ B1,β ′,0, for some β ′ ∈ K. d) B1,β,0 ∼ B1,1,0, for all β ∈ K \ {0}. e) The evolution algebras B1,1,0 and B1,0,0 are not isotopic. Proof. Let us prove each assertion separately. a) The triple (f, Id, Id) such that f (e1) = e1, f (e2) = αe2 and f (e3) = βe3 is an isotopism between the evolution algebras Aα,β and A1,1. b) Since isotopisms preserve the dimension of derived algebras, Proposi- tion 2.1.b establishes that any evolution algebra Bα,β,γ with a three- dimensional derived algebra (that is, such that γ 6= 0) is isotopic to the evolution algebra B0,0,1. c) From Proposition 2.1.a, we can suppose α 6= 0. Otherwise, it is enough to switch the vectors e1 and e2. The triple (f, Id, Id) such that f (ei) = ei, for i ∈ {1, 2}, and f (e3) = αe3, is then an isotopism between the evolution algebras Bα,β,0 and B1,β/α,0. d) The triple (f, g, h) related to the non-singular matrices 0 β 0 1 0 0 0 0 β F =    0 1 0 1 0 0 0 0 1 G =    H =  0 β 0 1 0 0 0 0 1   is an isotopism between the evolution algebras B1,β,0 and B1,1,0. e) Let (f, g, h) be an isotopism between the evolution algebras B1,1,0 and B1,0,0 and let H = (hij) be the non-singular matrix that is related to the linear transformation h. The computation of the reduced Grobner basis related to the ideal IB1 0 in Theorem 2.3.a enables us to ensure in particular that h12 = h13 = h22 = h23 = 0. But then, the matrix H is singular, which contradicts the fact of being (f, g, h) an isotopism. Hence, the algebras B1,1,0 and B1,0,0 are not isotopic. (cid:3) 0,B1 1 , , 0 , , Theorem 3.2 There exists four isotopism classes in En;3(K), whatever the base field is. They correspond to the evolution algebras A1,1, B0,0,1, B1,1,0 and B1,0,0. 7 Proof. The result follow straightforwardly from Proposition 3.1. (cid:3) Corollary 3.3 There exist eight isotopism classes of three-dimensional evo- lution algebras over any field. They correspond to the evolution algebras of structure tuples (0, 0, 0), (e1, 0, 0), (e1, e1, 0), (e1, e2, 0), (e1, e1, e1), (e1, e2, e3), (e1, e2, e1 + e2) and (e1, e2, e1). Proof. The result is an immediate consequence of Theorems 2.2 and (cid:3) 3.2. Let us describe the spectrum of genetic patterns of three distinct geno- types during a mitosis process that are related to the classification exposed in Corollary 3.3. To this end, let u, v and w be three distinct elements of the evolution algebra under consideration. The spectrum is formed by the next eight genetic patterns: (0, 0, 0), for which no offspring exists; (u, 0, 0), for which only one of the genotypes gives rise to offspring; (u, u, 0), for which exactly one of the genotypes does not produce offspring, whereas the other two give rise to offspring with the same genotype; (u, v, 0), for which ex- actly one of the genotypes does not produce offspring, whereas the other two give rise to offspring with distinct genotypes; (u, u, u), for which the off- spring has always the same genotype, whatever the initial one is; (u, v, w), for which the genotype of the offspring depends directly on that of the cell parent; (u, v, 1 2 v), for which the third genotype gives rise to each one of the genotypes produced by the other two with the same probability; and (u, v, u), for which two of the genotypes produce offspring with the same genotype. 2 u + 1 4 Distribution of three-dimensional evolution al- gebras with a non-trivial annihilator into iso- morphism classes Similarly to the reasoning exposed in the previous section, the distribution of the set E3(K) into isomorphism classes can be partitioned into that of the subsets E3;m(K), for m ≤ 3. In particular, Theorem 2.2 describes these partitions in case of dealing with m ∈ {0, 1}. Specifically, the only evolution algebra in E3;0(K) is the trivial three-dimensional algebra, whereas there exist exactly two isomorphism classes in E3;1(K). They correspond to the three-dimensional algebra described by e1e1 = e1 and that one described by e1e1 = e2. This section deals, therefore, with the computation of isomor- phism classes in E3;2(K). To this end, Theorem 2.2.d enables us to focus on 8 those three-dimensional evolution algebras described as or Cα,β,γ,δ,ǫ :=(e1e1 = αe1 + βe2, e2e2 = γe1 + δe2 + ǫe3, Dα,β,γ,δ,ǫ :=(e1e1 = αe1 + βe2, e3e3 = γe1 + δe2 + ǫe3, for some (α, β) ∈ {0, 1}2 \ {(0, 0)} and (γ, δ, ǫ) ∈ K3 \ {(0, 0, 0)}. The next result establishes when an evolution algebra of type Dα,β,γ,δ,ǫ is isomorphic to one of type Cα,β,γ,δ,ǫ. Throughout the distinct proofs of this and the subsequent results, we denote, respectively, as F = (fij) and G, the non- singular matrix related to an isomorphism f between the algebras under consideration and the corresponding reduced Grobner basis described in Theorem 2.3.b. Proposition 4.1 Let (α, β) ∈ {0, 1}2\{(0, 0)} and (γ, δ, ǫ) ∈ K3\{(0, 0, 0)}. Then, a) D1,0,γ,δ,ǫ b) D1,1,γ,δ,ǫ ∼= C1,0,γ,ǫ,δ ∼= C1,0,γ,ǫ,δ−γ. c) If ǫ 6= 0, then D0,1,γ,δ,ǫ ∼= Cǫ,γ,0,0,1. d) If γ 6= 0, then D0,1,γ,δ,0 ∼= C0,γ,0,0,1. e) Let δ′ ∈ K \ {0}. Then, i. The evolution algebras D0,1,0,δ′,0 and Cα,β,γ,δ,ǫ are not isomorphic. ii. If δ 6= 0, then the evolution algebras D0,1,0,δ,0 and D0,1,0,δ′,0 are iso- morphic if and only if there exists m ∈ K \ {0} such that δ = m2δ′. Proof. Let us prove each assertion separately. a) It is enough to switch the basis vectors e2 and e3. b) The result follows from (a) and the fact of being D1,1,γ,δ,ǫ ∼= D1,0,γ,δ−γ,ǫ by means of the isomorphism f such that f (e1) = e1 − e2 and f (ei) = ei, for all i ∈ {2, 3}. c) It is enough to consider the isomorphism f such that f (e1) = e2, f (e2) = e3 and f (e3) = e1 − δe3/ǫ. 9 d) Both algebras are isomorphic by means of the isomorphism f such that f (e1) = e2 − δe3/γ, f (e2) = e3 and f (e3) = e1. e) i. The computation of the related reduced Grobner basis G implies that δ′f 2 6= 0, we have that f21 = f22 = f23 = 0, which contradicts the fact of being F a non-singular matrix. 23 = 0. Since δ′ 21 = δ′f 2 22 = δ′f 2 ii. Here, the corresponding reduced Grobner basis G enables us to en- sure that f12 = f21 = f23 = f31 = f32 = 0, f22 = f 2 11 and δ = f22δ′. Hence, the isomorphism f exists if and only if there exists m ∈ K\{0} such that δ = m2δ′. In such a case, it is enough to consider the lin- ear transformation f such that f (e1) = me1, f (e2) = m2e2 and f (e3) = e3. (cid:3) Let us focus now on those evolution algebras of type Cα,β,γ,δ,ǫ. Since isomorphisms preserve the dimension of derived algebras, it is remarkable the possibility of distinguishing whether the parameter ǫ is equal to 0 or not. In the first case, the distribution of the set E3;2(K) into isomorphism classes is equivalent to that of the set E2;2(K), which has recently been discussed by the authors [14]. Theorem 4.2 Let (α, β) ∈ {0, 1}2 \ {(0, 0)} and (γ, δ) ∈ K2 \ {(0, 0)}. The evolution algebra Cα,β,γ,δ,0 ∈ E3;2(K) is isomorphic to exactly one of the next algebras • C1,0,γ,0,0, with γ ∈ K \ {0}. Here, C1,0,γ,0,0 ∼= C1,0,γ ′,0,0 if and only if γ ′ = γm2 for some m ∈ K \ {0}. • C1,1,−1,−1,0. • C1,0,γ,δ,0, with δ 6= 0. Here, C1,0,γ,δ,0 δ2γ. ∼= C1,0,γ ′,δ′,0 if and only if γδ′2 = • C0,1,γ,δ,0, with γ 6= 0. Here, C0,1,γ,δ,0 ∼= C0,1,γ ′,δ′,0 if and only if there exists an element m ∈ K \ {0} such that γ = γ ′m3 and δ = δ′m2, or, γ = γ ′2m3 and δ = δ′ = 0. • C1,1,γ,δ,0, with γ 6= d. Here, if γ 6= 0 6= δ, then C1,1,γ,δ,0 and only if γ ′ = γ2/δ3 and δ′ = γ/δ2. ∼= C1,1,γ ′,δ′,0 if We focus here, therefore, on the second case. 10 Lemma 4.3 Let (α, β) ∈ {0, 1}2 \ {(0, 0)} and let γ, δ, ǫ ∈ K be such that ǫ 6= 0. Then, Cα,β,γ,δ,ǫ ∼= Cα,β,γ,δ,1. Proof. both algebras that is described as f (ei) = ei, if i ∈ {1, 2}, and f (e3) = 1 (cid:3) It is enough to consider the linear transformation f between ǫ e3. Lemma 4.3 enables us to focus on the study of evolution algebras of the form Cα,β,γ,δ,1. The next result establishes which ones of these algebras are isomorphic to the evolution algebras described in Theorem 4.2. Proposition 4.4 Let (α, β) ∈ {0, 1}2 \ {(0, 0)} and let γ, δ ∈ K. Then, a) The evolution algebra Cα,β,γ,δ,1 is not isomorphic to any algebra C1,0,γ ′,0,0, with γ ′ ∈ K \ {0}, or to C1,1,−1,−1,0. b) Let γ ′, δ′ ∈ K be such that δ′ 6= 0. The evolution algebra Cα,β,γ,δ,1 is isomorphic to the algebra C1,0,γ ′,δ′,0 only if β = 0. Specifically, ∼= C1,0,γ,δ,0, whenever δ 6= 0. i. C1,0,γ,δ,1 ii. The evolution algebra C1,0,γ,0,1 and C1,0,γ ′,δ′,0 are not isomorphic. c) Let γ ′, δ′ ∈ K be such that γ ′ 6= 0. Then, ∼= C0,1,γ,δ,0, whenever γ 6= 0. i. The evolution algebras C1,0,γ,0,1 and C0,1,γ ′,δ′,0 are not isomorphic. ii. C0,1,γ,δ,1 iii. The evolution algebra C0,1,0,δ,1 and C0,1,γ ′,δ′,0 are not isomorphic. iv. The evolution algebra C1,1,γ,δ,1 is isomorphic to the algebra C0,1,γ ′,δ′,0 In particular, C1,1,γ,0,1 ∼= if and only if δ = 0 and γ ′2 = γδ′3. C0,1,γ 2,γ,0, for all γ ∈ K \ {0}. d) Let γ ′, δ′ ∈ K be such that γ ′ 6= δ′. Then, i. None of the evolution algebras C1,0,γ,0,1, C0,1,0,δ,1 and C1,1,0,0,1 is isomorphic to the algebra C1,1,γ ′,δ′,0. ii. Let δ 6= 0. Then, • C1,1,γ,δ,1 • C1,1,0,δ,1 ∼= C1,1,γ 2/δ3,γ/δ2,0 if and only if δ 6= γ 6= 0. ∼= C1,1,0,δ,0. Proof. Let us prove each assertion separately. 11 a) The result follows straightforwardly from the fact that isomorphisms pre- serve the dimension of derived algebras. In the case under considera- tion, the derived algebra of Cα,β,γ,δ,1 is two-dimensional, whereas those of C1,0,γ ′,0,0 and C1,1,−1,−1,0 are one-dimensional. b) The computation of the corresponding reduced Grobner basis G enables us to ensure that f31δ′ = f32δ′ = 0 and hence, f31 = f32 = 0, because δ′ 6= 0. After imposing this condition to the polynomials in G, we obtain that f11f22 = f12f21 = f21βγ = f11βγ = 0. From the non-singularity of the matrix F , we have, therefore, that βγ = 0. Now, if β = 1 and γ = 0, then the basis G involves that f11 = f22 = 0 and f21f33 = 0, which is a contradiction with being F a non-singular matrix. Hence, we can focus on the case β = 0. i. If δ 6= 0, then the evolution algebras C1,0,γ,δ,1 and C1,0,γ,δ,0 are iso- morphic by means of the isomorphism related to the non-singular matrix ii. The computation of the basis G involves that f22δ′ = 0 and hence, f22 = 0, because δ′ 6= 0. After imposing this condition to the polyno- mials in G, we also obtain that f11 = f21 = 0, which is a contradiction with being F a non-singular matrix. c) i. The computation of the basis G involves in this case that f (e3) = 0, which is a contradiction with being F a non-singular matrix. ii. If γ 6= 0, then the evolution algebras C0,1,γ,δ,1 and C0,1,γ,δ,0 are iso- morphic by means of the isomorphism related to the non-singular matrix 0 1 0 0 1 1 0 0 −δ     . 1 1 0 0 1 0 0 0 −γ     . iii. The computation of the basis G enables us to ensure that f11 = f12 = f31 = f32 = 0, which contradicts the fact of being F a non-singular matrix. iv. The computation of the basis G involves now that f11 = f22 = f31 = f32 = f12δ = 0. From the non-singularity of the matrix F , we have that δ = 0. After imposing these conditions to the polynomials in G, we obtain that the isotopism f exists if and only γ ′2 = γδ′3. In 12 such a case, γ 6= 0 6= δ′ and it is enough to define f as the linear transformation related to the non-singular matrix 0 γ ′/δ′2 0 1/δ′ 1 0 −1 0 −γ     . d) i. In the three cases, the computation of the corresponding reduced Grobner basis B involves that f (e3) = 0, which contradicts the fact of being F a non-singular matrix. ii. The computation of the basis G enables us to ensure in this case that f31(γ ′ − δ′) = 0 and hence, f31 = 0, because γ ′ 6= δ′. After imposing this condition to the polynomials in G, we obtain that f32δ′ = 0 and hence, f32 = 0, because δ′ 6= 0. The non-singularity of the matrix F involves then that f33 = f13(δ − γ) 6= 0, with δ 6= γ. Then, from the generators in G, exactly one of the next two cases holds. • f11 = f22 = 0, γ = f21 6= 0 and f12 = 1/δ′. According to the basis G, the isomorphism f exists in this case if and only if γ ′ = γ2/δ3 and δ′ = γ/δ2. In particular, it must be γ 6= 0. In such a case, it is enough to define f as the linear transformation related to the non-singular matrix 0 δ2/γ δ 0 0 0 1 −1 δ − γ   • f12 = f21 = 0 and f11 = f22 = 1.   . In this case, the linear transformation f related to the non-singular matrix 1 0 1 0 1 −1 0 0 δ     is an isomorphism between C1,1,0,δ,1 and C1,1,0,δ,0. (cid:3) Let us focus now on those evolution algebras of the form Cα,β,γ,δ,1 that are not isomorphic to any of the algebras described in Theorem 4.2, that is, on the evolution algebras of the form C1,0,γ,0,1, C0,1,0,δ,1 and C1,1,γ,γ,1. Lemma 4.5 Let (α, β) and (α′, β ′) be two pairs in {0, 1}2 \ {(0, 0)} and let γ, δ, γ ′, δ′ ∈ K be such that the evolution algebras Cα,β,γ,δ,1 and Cα′,β ′,γ ′,δ′,1 are isomorphic. Then, f (e3) = me3, for some m ∈ K \ {0}. 13 Proof. The coefficients of the basis vector e3 in the expression f (e3)f (e3) = f (e3e3) = 0 involves that f32 = 0. Due to it, the coefficients of the ba- sis vectors e1 and e2 in the previous expression enable us to ensure that f31α′ = f31β ′ = 0. Since (α′, β ′) 6= (0, 0), it is f31 = 0. Hence, f (e3) = f33e3, where f33 6= 0, because the matrix F is non-singular. (cid:3) Proposition 4.6 Let γ, δ, γ ′ ∈ K. Then, a) The evolution algebras C1,0,γ,0,1 and C1,0,γ ′,0,1 are isomorphic if and only if there exists m ∈ K \ {0} such that γ = m2γ ′. b) C0,1,0,δ,1 ∼= C1,0,δ,0,1, whenever δ 6= 0. c) The evolution algebras C0,1,0,0,1 and C1,1,γ,γ,1 are not isomorphic. Be- sides, none of them is isomorphic to the algebra C1,0,γ ′,0,1. d) If γ 6= γ ′, then the evolution algebras C1,1,γ,γ,1 and C1,1,γ ′,γ ′,1 are isomor- phic if and only if γγ ′ = 1. Proof. Let us prove each assertion separately. a) The computation of the corresponding reduced Grobner basis G enables us to ensure that f11 = 1, f12 = f13 = f21 = f31 = f32 = 0, f33 = f 2 22 and γ = f33γ ′. From Lemma 4.5, the isomorphism f exists if and only if there exists m ∈ K \ {0} such that γ = m2γ ′. In such a case, it is enough to consider the linear transformation f such that f (e1) = e1, f (e2) = me2 and f (e3) = m2e3. b) The linear transformation related to the non-singular matrix 0 0 1 1 δ 0 0 0 −δ     is an isomorphism between these two algebras. c) In any of the possible cases under consideration, the computation of the corresponding reduced Grobner basis G involves that f33 = 0, which contradicts Lemma 4.5. d) The computation of the corresponding reduced Grobner basis G involves that f11 = f22, f12 = f21 = γ(1 − f22) and f31 = f32 = 0 = (γ ′ − γ)f22. Hence, since γ 6= γ ′, we have that f11 = f22 = 0 and f12 = f21 = γ. 14 After imposing these conditions to the polynomials in G, we obtain that γ ′f33 = −γ2 and γ2γ ′ = γ. The first condition, together with Lemma 4.5, enables us to ensure that γ 6= 0 6= γ ′. Then, the second condition involves that γγ ′ = 1 and hence, f33 = −γ3. We also obtain from the reduced Grobner basis G that f13 + f23 = γ2. In particular, the linear transformation related to the non-singular matrix γ2 0   0 γ γ 0 0 0 −γ3   is an isomorphism between the evolution algebras C1,1,γ,γ,1 and C1,1,γ ′,γ ′,1, whenever γγ ′ = 1. (cid:3) The next theorem gathers together all the results that we have exposed throughout this section. Theorem 4.7 Any evolution algebra in the set E3;2(K) is isomorphic to exactly one of the algebras of Theorem 4.2 or to one of the next algebras • C1,0,γ,0,1, for some γ ∈ K. Here, C1,0,γ,0,1 ∼= C1,0,γ ′,0,1 if and only if there exists m ∈ K \ {0} such that γ = m2γ ′. • C1,1,γ,γ,1, for some γ ∈ K. Here, C1,1,γ,γ,1 ∼= C1,1,γ ′,γ ′,1 if and only if γγ ′ = 1. • D0,1,0,δ,0, for some δ ∈ K \ {0}. Here, D0,1,0,δ,0 if there exists m ∈ K \ {0} such that δ = m2δ′. ∼= D0,1,0,δ′,0 if and only 5 Conclusion and further studies This paper has dealt with the use of Computational Algebraic Geometry to determine the distribution of three-dimensional evolution algebras over any field into isotopism and isomorphism classes. We have proved in particular that the set E3(K) is distributed into eight isotopism classes, whatever the base field is, and we have characterized their isomorphism classes in case of dealing with algebras with a one-dimensional annihilator. We have also described the spectrum of genetic patterns of three distinct genotypes dur- ing mitosis of eukaryotic cells. The distribution into isotopism classes of the set E4(K) and the characterization of their isomorphism classes in case of dealing with algebras with one-dimensional annihilator is established as further work. 15 Acknowledgements The authors are very grateful to the Editors and to the referees for their valuable comments and suggestions that have improved the original version of the paper. Particularly, the authors gratefully thank to one of the referees for the suggestion of considering for our study the recent work developed by Cabrera et al. [4], which has enabled us to detect a pair of omitted cases of study in the preliminary version of the paper. References [1] A. A. Albert, Non-Associative Algebras: I. Fundamental Concepts and Isotopy, Ann. of Math., Second Series 43 (1942) 685 -- 707. [2] B. Buchberger, An algorithm for finding the basis elements of the residue class ring of a zero dimensional polynomial ideal, J. Symbolic Comput. 41 (2006) 475 -- 511. [3] Y. Cabrera Casado, M. Siles Molina, M. V. Velasco, Evolution algebras of arbitrary dimension and their decompositions, Linear Algebra App. 495 (2016) 122 -- 162. [4] Y. Cabrera Casado, M. Siles Molina, M. V. Velasco, Description of three-dimensional evolution algebras. arXiv:1701.07219 (2017). [5] L. M. Camacho, J. R. G´omez, B. A. Omirov, R. M. Turdibaev, Some properties of evolution algebras, Bull. Korean Math. Soc. 50 (2013) 1481 -- 1494. [6] L. M. Camacho, J. R. G´omez, B. A. Omirov, R. M. Turdibaev, The derivations of some evolution algebras, Linear Multilinear A. 61 (2013) 309 -- 322. [7] T. M. M. Campos, P. Holgate, Algebraic Isotopy in Genetics, IMA J. Math. Appl. Med. Biol. 4 (1987) 215 -- 222. [8] J. M. Casas, M. Ladra, B. A. Omirov, U. A. Rozikov, On nilpotent index and dibaricity of evolution algebras, Linear Algebra Appl. 439 (2013) 90 -- 05. [9] J. M. Casas, M. Ladra, B. A. Omirov, U. A. Rozikov, On Evolution Algebras, Algebra Colloq. 21 (2014) 331 -- 342. [10] J. M. Casas, M. Ladra, U. A. Rozikov, A chain of evolution algebras, Linear Algebra Appl. 435 (2011) 852 -- 870. 16 [11] D. A. Cox, J. B. Little, D. O'Shea, Using Algebraic Geometry, Springer- Verlag, New York, 1998. [12] W. Decker, G. M. Greuel, G. Pfister, H. Schonemann, Singular 4-1-0 -- A computer algebra system for polynomial computations, 2017. [13] A. Dzhumadildaev, B. A. Omirov, U. A. Rozikov, Constrained evolu- tion algebras and dynamical systems of a bisexual population, Linear Algebra Appl. 496 (2016) 351 -- 380. [14] O. J. Falc´on, R. M. Falc´on, J. N´unez, Classification of asexual diploid organisms by means of strongly isotopic evolution algebras defined over any field, J. Algebra 472 (2017) 573 -- 593. [15] A. S. Hegazi, H. Abdelwahab, Nilpotent evolution algebras over arbi- trary fields, Linear Algebra App. 486 (2015) 345 -- 360. [16] P. Holgate, Genetic algebras associated with polyploidy, Proc. Edin- burgh Math. Soc. 15 (1966) 1 -- 9. [17] A. Labra, M. Ladra, U. A. Rozikov, An evolution algebra in population genetics, Linear Algebra Appl. 457 (2014) 348 -- 362. [18] M. Ladra, B. A. Omirov, U. A. Rozikov, Dibaric and evolution algebras in biology, Lobachevskii J. Math. 35 (2014) 198 -- 210. [19] M. Ladra, U. A. Rozikov, Evolution algebra of a bisexual population, J. Algebra 378 (2013) 153 -- 172. [20] Y. N. Lakshman, On the complexity of computing a Grobner basis for the radical of a zero dimensional ideal. In: H. Ortiz (ed.) Proceedings of the twenty-second annual ACM Symposium on Theory of computing, STOC'90, New York, 1990; 555 -- 563. [21] B. A. Omirov, U. A. Rozikov, K. M. Tulenbayev, On real chains of evolution algebras, Linear Multilinear Algebra 63 (2015) 586 -- 600. [22] J. P. Tian, Evolution Algebras and their Applications, Lect. Notes Math. 1921, Springer-Verlag, Berlin, 2008. [23] J. P. Tian, P. Vojtechovsky, Mathematical concepts of evolution alge- bras in non-mendelian genetics, Quasigroups Related Systems 14 (2006) 111 -- 122. 17
1106.2203	3	1106	2012-02-28T18:32:05	Lattices of quasi-equational theories as congruence lattices of semilattices with operators, Part I	[ "math.RA", "math.LO" ]	We show that for every quasivariety K of structures (where both functions and relations are allowed) there is a semilattice S with operators such that the lattice of quasi-equational theories of K (the dual of the lattice of sub-quasivarieties of K) is isomorphic to Con(S,+,0,F). As a consequence, new restrictions on the natural quasi-interior operator on lattices of quasi-equational theories are found.	math.RA	math	LATTICES OF QUASI-EQUATIONAL THEORIES AS CONGRUENCE LATTICES OF SEMILATTICES WITH OPERATORS, PART I KIRA ADARICHEVA AND J. B. NATION Abstract. We show that for every quasivariety K of structures (where both functions and relations are allowed) there is a semilattice S with operators such that the lattice of quasi-equational theories of K (the dual of the lattice of sub-quasivarieties of K) is isomorphic to Con(S, +, 0, F). As a consequence, new restrictions on the natural quasi-interior operator on lattices of quasi-equational theories are found. 2 1 0 2 b e F 8 2 ] . A R h t a m [ 3 v 3 0 2 2 . 6 0 1 1 : v i X r a 1. Motivation and terminology Our objective is to provide, for the lattice of quasivarieties contained in a given quasivariety (Q-lattices in short), a description similar to the one that characterizes the lattice of subvarieties of a given variety as the dual of the lattice of fully invariant congruences on a countably generated free algebra. Just as the result for varieties is more naturally expressed in terms of the lattice of equational theories, rather than the dual lattice of varieties, so it will be more natural to consider lattices of quasi-equational theories rather than lattices of quasivarieties. The basic result is that the lattice of quasi-equational theories extend- ing a given quasi-equational theory is isomorphic to the congruence lattice of a semilattice with operators preserving join and 0. These lattices sup- port a natural quasi-interior operator, the properties of which lead to new restrictions on lattices of quasi-equational theories. This is the first paper in a series of four. Part II shows that if S is a semilattice with both 0 and 1, and G is a group of operators on S such that each operator in G fixes both 0 and 1, then there is a quasi-equational theory T such that Con(S, +, 0, G) is isomorphic to the lattice of quasi-equational theories extending T. The third part [30] shows that if S is any semilattice with operators, then Con S is isomorphic to the lattice of implicational the- ories extending some given implicational theory, but in a language that may Date: July 1, 2019. 2010 Mathematics Subject Classification. 08C15, 08A30, 06A12. Key words and phrases. quasivariety, quasi-equational theory, congruence lattice, semilattice. The authors were supported in part by a grant from the U.S. Civilian Research & Development Foundation. The first author was also supported in part by INTAS Grant N03-51-4110. 1 2 KIRA ADARICHEVA AND J. B. NATION not include equality. The fourth paper [23], with T. Holmes, D. Kitsuwa and S. Tamagawa, concerns the structure of lattices of atomic theories in a language without equality. The setting for varieties is traditionally algebras, i.e., sets with opera- tions, whereas work on quasivarieties normally allows structures, i.e., sets with operations and relations. Some adjustments are required for the more general setting. Let us review the universal algebra of structures, following Section 1.4 of Gorbunov [17]; see also Gorbunov and Tumanov [19, 20] and Gorbunov [15]. The type of a structure is determined by its signature σ = (cid:104)F, R, ρ(cid:105) where F is a set of function symbols, R is a set of relation symbols, and ρ : F∪R → ω assigns arity. A structure is then A = (cid:104)A, FA, RA(cid:105) where A is the carrier set, FA is the set of operations on A, and RA is the set of relations on A. For structures A and B of the same type, a map h : A → B is a ho- momorphism if it preserves operations and h(RA) ⊆ RB for each relation symbol R. An endomorphism of A is a homomorphism ε : A → A. The kernel ker h of a homomorphism h is a pair κ = (cid:104)κ0, κ1(cid:105) where • κ0 is the equivalence relation on A induced by h, i.e., (x, y) ∈ κ0 iff • κ1 =(cid:83) h(x) = h(y), R∈R κR 1 where κR 1 = h−1(RB) = {s ∈ Aρ(R) : h(s) ∈ RB}. • θ1 =(cid:83) Equality is treated differently because, in standard logic, equality is assumed to be a congruence relation. Indeed, the statements that ≈ is reflexive, sym- metric, transitive, and compatible with the functions of F and the relations of R, are universal Horn sentences. Thus in normal quasi-equational logic we are working in the quasivariety given by these laws. This is not necessary: see Parts III and IV [30, 23]. A congruence on a structure A = (cid:104)A, FA, RA(cid:105) is a pair θ = (cid:104)θ0, θ1(cid:105) where • θ0 is an equivalence relation on A that is compatible with the oper- 1 where each θR 1 ⊆ Aρ(R) and RA ⊆ θR R∈R θR 1 and b ∈ Aρ(R) and a θ0 b componentwise, then b ∈ θR 1 . ations of FA, and 1 , i.e., the original relations of A are contained in those of θ1, and for each R ∈ R, if a ∈ θR Note that if h : A → B is a homomorphism, then ker h is a congruence on A. The collection of all congruences on A forms an algebraic lattice Con A under set containment. A subset S ⊆ A is a subuniverse if it is closed under the operations of A. A substructure of A is S = (cid:104)S, FS, RS(cid:105) where S is a subuniverse of A, for each operation symbol f ∈ F the operation f S is the restriction of f A to Sρ(f ), and for each relation symbol R ∈ F the relation RS is RA ∩ Sρ(R). Given a congruence θ on a structure A, we can form a quotient structure A/θ by defining operations and relations on the θ0-classes of A in the natural way. The isomorphism theorems carry over to this more general setting. In particular, if h : A → B is a homomorphism, then h(A) is a substructure of B, and h(B) is isomorphic to A/ ker h. Q-LATTICES AS CONGRUENCE LATTICES 3 A congruence is fully invariant if, for every endomorphism ε of A, • a θ0 b implies ε(a) θ0 ε(b), and • for each R ∈ R, a ∈ θR 1 implies ε(a) ∈ θR 1 . The lattice of fully invariant congruences is denoted Ficon A. The congruence generation theorems are straightforward to generalize. Let C ⊆ A2 and let D be a set of formulae of the form R(a) with R ∈ R and a ∈ Aρ(R). The congruence generated by C ∪ D, denoted con(C ∪ D), is the least congruence θ = (cid:104)θ0, θ1(cid:105) such that C ⊆ θ0 and a ∈ θR 1 for all R(a) ∈ D. The equivalence relation θ0 is given by the usual Mal'cev construction applied to C, and θ1 is the closure of D ∪ RA with respect to θ0, i.e., if R(a) ∈ D and a θ0 b componentwise, then b ∈ θR 1 . A variety is a class closed under homomorphic images, substructures and direct products. Varieties are determined by laws of the form s ≈ t and R(s) where s, t and the components of s are terms. That is, a variety is the class of all similar structures satisfying a collection of atomic formulae. If V is a variety of structures and F is the countably generated free structure for V, then the lattice Lv(V) of subvarieties of V is dually isomorphic to the lattice of fully invariant congruences of F, i.e., Lv(V) ∼=d Ficon F. In the case of varieties of algebras (with no relational symbols in the language), this is equivalent to adding the endomorphisms of F to its operations and taking the usual congruence lattice, so that Lv(V) ∼=d Con (F, F ∪ End F). For structures in general, this simplification does not work. (These standard results are based on Birkhoff [8].) A quasivariety is a class of structures closed under substructures, direct products and ultraproducts (equivalently, substructures and reduced prod- ucts). Quasivarieties are determined by laws that are quasi-identities, i.e., Horn sentences &1≤i≤nαi =⇒ β where the αi and β are atomic formulae of the form s ≈ t and/or R(s). If K is a quasivariety and A a structure, then a congruence θ on A is said to be a K-congruence if A/θ ∈ K. Since the largest congruence is a K-congruence, and K-congruences are closed under intersection, the set of K-congruences on A forms a complete meet subsemilattice of Con A, denoted ConK A. Moreover, ConK A is itself an algebraic lattice. Let us adopt some notation to reflect the standard duality between the- ories and models. For a variety V, let ATh(V) denote the lattice of "equa- tional" (really, atomic) theories extending the theory of V, so that ATh(V) ∼=d Lv(V). Likewise, for a quasivariety K, let QTh(K) denote the lattice of quasi- equational theories containing the theory of K, so that QTh(K) ∼=d Lq(K). Gorbunov and Tumanov described the lattice Lq(K) of quasivarieties con- tained in a given quasivariety K in terms of algebraic subsets. This descrip- tion requires some definitions. • Given K, let F = FK(ω) be the countably generated K-free struc- ture. Then ConK F denotes the lattice of all K-congruences of F. 4 KIRA ADARICHEVA AND J. B. NATION • Define the isomorphism relation I and embedding relation E on ConK F by ϕ I ψ if F/ψ ∼= F/ϕ ϕ E ψ if F/ψ ≤ F/ϕ. • For a binary relation R on a complete lattice L, let Sp(L, R) denote the lattice of all R-closed algebraic subsets of L. (Recall that S ⊆ L is algebraic if it is closed under arbitrary meets and nonempty directed joins. The set S is R-closed if s ∈ S and s R t implies t ∈ S.) The characterization theorem of Gorbunov and Tumanov [20] then says that Lq(K) ∼= Sp(ConK F, I) ∼= Sp(ConK F, E). See Section 5.2 of Gorbunov [17]; also cf. Hoehnke [21]. By way of comparison, we might say that the description of the lattice of subvarieties by Lv(V) ∼=d Ficon F reflects equational logic, whereas the representation Lq(K) ∼= Sp(ConK F, E) say reflects structural properties the former for quasivarieties, ideally something of the form Lq(K) ∼=d Con S (closure under S, P and direct limits). We would like to find an analogue of for some semilattice S with operators, reflecting quasi-equational logic. This is done below. Indeed, while our emphasis is on the structure of Q-lattices, Bob Quackenbush has used the same general ideas to provide a nice algebraic proof of the completeness theorem for quasi-equational logic [33]. The lattice QTh(K) of theories of a quasivariety is algebraic and (com- pletely) meet semidistributive. Most of the other known properties of these lattices can be described in terms of the natural equa-interior operator, which is the dual of an equational closure operator on QTh(K). See Appen- dix II or Section 5.3 of Gorbunov [17]. A.M. Nurakunov [31], building on earlier work of R. McKenzie [28] and R. Newrly [29], has recently provided a nice algebraic description of the lattices ATh(V), where V is a variety of algebras, as congruence lattices of monoids with two additional unary operations satisfying certain properties. See Appendix III. Finally, let us note two (related) major differences between quasivarieties of structures versus algebras. Firstly, the greatest quasi-equational theory in QTh(K) need not be compact if the language of K has infinitely many relations. Secondly, many nice representation theorems for quasivarieties use one-element structures, whereas one-element algebras are trivial. Indeed, in light of Theorem 2 below, Theorem 5.2.8 of Gorbunov [17] (from Gorbunov and Tumanov [18]) can be stated as follows. Theorem 1. The following are equivalent for an algebraic lattice L. (1) L ∼= Con(S, +, 0) for some semilattice S. (2) L ∼= QTh(K) for some quasivariety K of one-element structures. Q-LATTICES AS CONGRUENCE LATTICES 5 Congruence lattices of semilattices are coatomistic, i.e., every element is a meet of coatoms. Thus the Q-lattices for the special quasivarieties in the preceding theorem are correspondingly atomistic. 2. Congruence lattices of semilattices Let Sp(L) denote the lattice of algebraic subsets of a complete lattice L. If L is an algebraic lattice, let Lc denote its semilattice of compact elements. This is a join semilattice with zero. The following result of Fajtlowicz and Schmidt [11] directly generalizes the Freese-Nation theorem [13]. See also [12], [22], [34]. Theorem 2. If L is an algebraic lattice, then Sp(L) ∼=d Con Lc. Proof. For an arbitrary join 0-semilattice S = (cid:104)S, +, 0(cid:105) we set up a Galois correspondence between congruences of S and algebraic subsets of the ideal lattice I(S) as follows. For θ ∈ Con S, let h(θ) be the set of all θ-closed ideals of S. For H ∈ Sp(I(S)), let x ρ(H) y if {I ∈ H : x ∈ I} = {J ∈ H : y ∈ J}. It is straightforward to check that h and ρ are order-reversing, that h(θ) ∈ Sp(I(S)) and ρ(H) ∈ Con S. To show that θ = ρh(θ), we note that if x < y (w.l.o.g.) and (x, y) /∈ θ, then {z ∈ S : x + z θ x} is a θ-closed ideal containing x and not y. Hence (x, y) /∈ ρh(θ). To show that H = hρ(H), consider an ideal J /∈ H. For any x ∈ S, let x =(cid:84){I ∈ H : x ∈ I}, noting that x ∈ H. Then {x : x ∈ J} is up-directed, whence(cid:83){x : x ∈ J} ∈ H. Therefore the union properly contains J, so that there exist x < y with x ∈ J and y ∈ x− J, and J is not ρ(H)-closed. Thus J /∈ H implies J /∈ hρ(H), as desired. (cid:3) Compare this with the following result of Adaricheva, Gorbunov and Tu- manov ([5] Theorem 2.4, also [17] Theorem 4.4.12). Theorem 3. Let L be a join semidistributive lattice that is finitely presented within the class SD∨. Then L ≤ Sp(A) for some algebraic and dually algebraic lattice A. On the other hand, Example 4.4.15 of Gorbunov [17] gives a 4-generated join semidistributive lattice that is not embeddable into any lower continuous lattice satisfying SD∨. Keith Kearnes points out that the class ES of lattices that are embeddable into congruence lattices of semilattices is not first order. Indeed, every finite meet semidistributive lattice is in ES, and ES is closed under S and P. Now the quasivariety SD∧ is generated by its finite members (Tumanov [35], Theorem 4.1.7 in [17]), while ES is properly contained in SD∧. Hence ES is not a quasivariety, which means it must not be closed under ultraproducts. This result has been generalized in Kearnes and Nation [25]. 6 KIRA ADARICHEVA AND J. B. NATION 3. Connection with Quasivarieties In this section, we will show that for each quasivariety K of structures, the lattice of quasi-equational theories Qth(K) is isomorphic to the congruence lattice of a semilattice with operators. Given a quasivariety K, let F = FK(ω) be the K-free algebra on ω gener- ators, and let ConK F be the lattice of K-congruences of F. For a set S of atomic formulae, recall that the K-congruence generated by S is (cid:92){ψ ∈ Con F : F/ψ ∈ K and S ⊆ ψ}. conK S = (cid:87) Then let T = TK denote the join semilattice of compact K-congruences in ConK F. Thus T = (ConK FK(ω))c consists of finite joins of the form j ϕj, with each ϕj either conK (s, t) or conK R(s) for terms s, t, si ∈ F and a relation R. Let X be a free generating set for FK(ω). Any map σ0 : X → F can be extended to an endomorphism σ : F → F in the usual way. Since the image σ(F) is a substructure of F, the kernel of an endomorphism σ is a K-congruence. The endomorphisms of F form a monoid End F. The endomorphisms of F act naturally on T. For ε ∈ End F, define (cid:98)ε(conK (s, t)) = conK (εs, εt) (cid:98)ε(conK R(s)) = conK R(εs) (cid:95) j (cid:98)εϕj. (cid:95) (cid:98)ε( j ϕj) = implies The next lemma is used to check the crucial technical details that(cid:98)ε is well- (cid:98)ε(conK α) ≤(cid:95)(cid:98)ε(conK βj). defined, and hence join-preserving. Lemma 4. Let K be a quasivariety, F a K-free algebra, and ε ∈ End F. Let α, β1, . . . , βm be atomic formulae. In ConK F, conK α ≤(cid:95) conK βj Proof. For an atomic formula α and a congruence θ, let us write α ∈ θ to mean either (1) α is s ≈ t and (s, t) ∈ θ0, or (2) α is R(s) and s ∈ θR 1 . So for the lemma, we are given that if F/ψ ∈ K and β1, . . . , βm ∈ ψ, then α ∈ ψ. We want to show that if F/θ ∈ K and εβ1, . . . , εβm ∈ θ, then εα ∈ θ. Let θ ∈ Con F be a congruence such that F/θ ∈ K, and let h : F → F/θ be the natural map. Then hε : F → F/θ, and since hε(F) is a substructure of h(F), the image is in K. Now β1, . . . , βm ∈ ker hε, and so α ∈ ker hε. Thus εα ∈ ker h = θ, as desired. (cid:3) Now let ξ be a compact K-congruence. Suppose that ξ = (cid:87) ξ =(cid:87) for each i we have ϕi ≤ (cid:87) j(cid:98)εψj by Lemma 4. Thus i(cid:98)εϕi, and so (cid:98)εξ = (cid:87) i(cid:98)εϕi ≤ (cid:87) (cid:87) j(cid:98)εψj = (cid:87) i(cid:98)εϕi is well-defined. i ϕi and j ψj in T, with each ϕi and ψj being a principal K-congruence. Then j ψj, whence (cid:98)εϕi ≤ (cid:87) j(cid:98)εψj ≤ (cid:87) j(cid:98)εψj. Symmetrically (cid:87) in T, then Q-LATTICES AS CONGRUENCE LATTICES It then follows from the definition of (cid:98)ε that if ϕ =(cid:87) ϕi ∨ (cid:95) (cid:95) (cid:98)ε(ϕ ∨ ψ) =(cid:98)ε( (cid:98)εϕi ∨ (cid:95) (cid:95) (cid:98)εψj) =(cid:98)εϕ ∨(cid:98)εψ. ψj) = i j j i Thus (cid:98)ε preserves joins. Also note that for the zero congruence we have (cid:98)ε(0) = 0. Let(cid:98)E = {(cid:98)ε : ε ∈ End F}, and consider the algebra S = SK = (cid:104)T,∨, 0,(cid:98)E(cid:105). By the preceding remarks, the operations of(cid:98)E are operators on S, i.e., (∨, 0)- homomorphisms, so S is a join semilattice with operators. With this setup, we can now state our main result. Theorem 5. For a quasivariety K, where S = (cid:104)T,∨, 0,(cid:98)E(cid:105) with T the semilattice of compact congruences of Lq(K) ∼=d Con S ConK F, E = End F, and F = FK(ω). In Part II, we will use this technical variation. Theorem 6. Let K be a quasivariety and let n ≥ 1 be an integer. Then the lattice of all quasi-equational theories that (1) contain the theory of K, and (2) are determined relative to K by quasi-identities in at most n vari- is isomorphic to Con Sn, where Sn = (cid:104)Tn,∨, 0,(cid:98)E(cid:105) with Tn the semilattice ables, of compact congruences of ConK F, E = End F, and F = FK(n). We shall prove Theorem 5, and afterwards discuss the modification re- quired for Theorem 6, which is essentially just replacing FK(ω) by FK(n). For the proof of Theorem 5, and for its application, it is natural to use two structures closely related to the congruence lattice instead. For an algebra A with a join semilattice reduct, let Don A be the lattice of all reflexive, transitive, compatible relations R such that ≥⊆ R, i.e., x ≥ y implies x R y. Let Eon A be the lattice of all reflexive, transitive, compatible relations R such that (1) R ⊆≤, i.e., x R y implies x ≤ y, and (2) if x ≤ y ≤ z and x R z, then x R y. i ϕi and ψ =(cid:87) 7 j ψj Lemma 7. If A = (cid:104)A,∨, 0, F(cid:105) is a semilattice with operators, then Con A ∼= Don A ∼= Eon A. Proof. Let δ : Con A → Don A via δ(θ) = θ◦ ≥, so that x δ(θ) y iff x θ x ∨ y 8 KIRA ADARICHEVA AND J. B. NATION and let γ : Don A → Con A via γ(R) = (R ∩ ≤) ◦ (R ∩ ≤) (cid:96) , so that x γ(R) y iff x R x ∨ y & y R x ∨ y. Now we check that, for θ ∈ Con A and R ∈ Don A, (1) δ(θ) ∈ Don A, (2) γ(R) ∈ Con A, (3) δ and γ are order-preserving, (4) γδ(θ) = θ, (5) δγ(R) = R. . (cid:96) This is straightforward and only slightly tedious. Similarly, let ε : Don A → Eon A via ε(R) = R ∩ ≤, and δ(cid:48) : Eon A → Don A via δ(cid:48)(S) = S ◦ ≥, and check the analogous statements for this pair, which is again routine. Note that for a congruence relation θ the corresponding eon-relation is εδ(θ) = θ ∩ ≤, while for S ∈ Eon A we have the congruence γδ(cid:48)(S) = S ◦ S (cid:3) Now we define a Galois connection between T 2 and structures A ∈ K. (The collection of structures A ∈ K forms a proper class. However, every quasivariety is determined by its finitely generated members. So we could avoid any potential logical difficulties by restricting our attention to struc- tures A defined on some fixed infinite set large enough to contain an isomor- phic copy of each finitely generated member of K.) For a pair (β, γ) ∈ T 2 and A ∈ K, let (β, γ) Ξ A if, whenever h : F → A is a homomorphism, β ≤ ker h implies γ ≤ ker h. Then, following the usual rubric for a Galois connection, for X ⊆ T 2 let κ(X) = {A ∈ K : (β, γ) Ξ A for all (β, γ) ∈ X}. Likewise, for Y ⊆ K, let ∆(Y ) = {(β, γ) ∈ T 2 : (β, γ) Ξ A for all A ∈ Y }. We must check that the following hold for X ⊆ T 2 and Y ⊆ K. (1) κ(X) ∈ Lq(K), (2) ∆(Y ) ∈ Don S, (3) ∆κ(X) = X if X ∈ Don S, (4) κ∆(Y ) = Y if Y ∈ Lq(K). To prove (1), we show that κ(X) is closed under substructures, direct products and ultraproducts. Closure under substructures is immediate, and closure under direct products follows from the observation that if h : F → (cid:81) i Ai then ker h = (cid:84) ker πih. So let Ai ∈ κ(X) for i ∈ I, let U be an ultrafilter on I, and let h : F →(cid:81) Ai/U be a homomorphism. Since F is free, we can find f : F →(cid:81) Ai such that h = gf where g :(cid:81) Ai →(cid:81) Ai/U is the standard map. Let (β, γ) ∈ X with β =(cid:87) ϕj and γ =(cid:87) ψk, where these are finite joins and each ϕ and ψ is of the form conK α for an atomic formula α. Each α in turn is of the form either s ≈ t or R(s). Q-LATTICES AS CONGRUENCE LATTICES 9 Assume β ≤ ker h. Then h(αj) holds for each j, so that for each j we have {i ∈ I : πif (αj)} ∈ U . Taking the intersection, {i ∈ I : ∀j πif (αj)} ∈ U . In other words, {i ∈ I : β ≤ ker πif} ∈ U , and so the same thing holds for γ. Now we reverse the steps to obtain γ ≤ ker h, as desired. Thus κ(X) is also closed under ultraproducts, and it is a quasivariety. To prove (2), let Y ⊆ K. It is straightforward that ∆(Y ) ⊆ T 2 is a relation that is reflexive, transitive, and contains ≥. Moreover, if (β, γ) ∈ ∆(Y ) and β ∨ τ ≤ ker h for an appropriate h, then γ ∨ τ ≤ ker h, so ∆(Y ) respects joins. Again let (β, γ) ∈ ∆(Y ) and h : F → A with A ∈ Y . Let (cid:98)ε ∈ (cid:98)E and assume that(cid:98)εβ ≤ ker h. This is equivalent to β ≤ ker hε, as both mean that hε(αj) holds for all j, where β = (cid:87) conK αj. Hence γ ≤ ker hε, yielding (cid:98)εγ ≤ ker h. Thus ∆(Y ) is compatible with the operations of(cid:98)E. We conclude holding in Y . Set β =(cid:87) conK αj and γ = conK ζ, and let h : F → A be that ∆(Y ) ∈ Don S. Next consider (4). Given that Y is a quasivariety, we want to show that κ∆(Y ) ⊆ Y . Let A ∈ κ∆(Y ), and let &j αj =⇒ ζ be any quasi-identity a homomorphism. Then (β, γ) ∈ ∆(Y ), whence as A ∈ κ∆(Y ) we have β ≤ ker h implies γ ≤ ker h. Thus A satisfies the quasi-identity in question, which shows that κ∆(Y ) ⊆ Y , as desired. Part (3) requires the most care (we must show that relations in Don S correspond to theories of quasivarieties). Given X ∈ Don S, we want to prove that ∆κ(X) ⊆ X. Let (µ, ν) ∈ T 2 − X. Define a congruence θ on F as follows. θ0 = µ θk+1 = θk ∨(cid:95){γ(β, γ) ∈ X and β ≤ θk} (cid:95) θ = θk. k of (βi, γi) ∈ X with each βi ≤ θk. Let ξ =(cid:87) βi, so that ξ is compact and (ξ,(cid:87) γi) ∈ X. Now inductively (µ, ξ) ∈ X, and so (µ,(cid:87) γi) ∈ X. Let C = F/θ. We want to show that C ∈ κ(X) and that ν (cid:2) θ. Claim a. If ψ is compact and ψ ≤ θ, then (µ, ψ) ∈ X. We prove by induction that if compact ψ ≤ θk, then (µ, ψ) ∈ X. For k = 0 this is trivial. Assume the statement holds for k. Suppose we have a finite collection βi ≤ ξ ≤ θk. Then (ξ, βi) ∈ X, so by transitivity (ξ, γi) ∈ X for all i. Hence Claim b. If (β, γ) ∈ X and β ≤ θ, then γ ≤ θ. This holds by construction and compactness. Claim c. F/θ ∈ κ(X). Suppose h : F → F/θ, (β, γ) ∈ X and β ≤ ker h. Let f : F → F/θ be the standard map with ker f = θ. There exists an endmorphism ε of F such that h = f ε. Then, using Claim b and an argument 10 above, KIRA ADARICHEVA AND J. B. NATION β ≤ ker h = ker f ε =⇒ (cid:98)εβ ≤ ker f = θ =⇒ (cid:98)εγ ≤ θ = ker f =⇒ γ ≤ ker f ε = ker h. Claim d. (µ, ν) /∈ ∆κ(X). This is because C ∈ κ(X) by Claim c and µ ≤ θ = ker f , while ν (cid:2) θ by Claim a. This completes the proof of (3), and hence Theorem 5. Only a slight modification is required for Theorem 6. Consider the col- lection of quasivarieties C satisfying the conditions of the theorem: (1) C ⊆ K, and (2) C is determined relative to K by quasi-identities in at most n vari- ables. These properties mean that a structure C is in C if and only if and (1)(cid:48) Every map f0 : ω → C extends to a homomorphism f : FK(ω) → C, (2)(cid:48) Every map g0 : n → C extends to a homomorphism g : FC(ω) → C. Quasivarieties satisfying conditions (1) and (2) are closed under arbitrary joins, and thus under containment they form a lattice which we will denote by Ln q (K). This is a complete join subsemilattice of Lq(K); the correspond- ing dual lattice of theories is a complete meet subsemilattice QThn(K) of QTh(K). The proof of Theorem 5 gives us QTh(K) as the congruence lattice of a semilattice with operators obtained from FK(ω). In view of condition (2)(cid:48), the same construction with FK(ω) replaced throughout by FK(n) yields QThn(K). 4. Interpretation of principal congruences, say β = (cid:87) conK αj and γ = (cid:87) conK ζk, where The foregoing analysis is rather structural and omits the motivation, which we supply here. Let β and γ be elements of T, i.e., compact K- congruences on the free structure F. Then these are finite joins in ConK F each α and ζ is an atomic formula of the form s ≈ t or R(s). The basic idea is that the congruence con(β, β ∨ γ), on the semilattice S of compact K- congruences of F with the endomorphisms as operators, should correspond to the conjunction over the indices k of the quasi-identities &j αj =⇒ ζk, and that furthermore the quasi-equational consequences of combining impli- cations (modulo the theory of K) behaves like the join operation in Con S. But β ≥ γ should mean that β =⇒ γ, so it is really Don S that we want. On the other hand, all the nontrivial information is contained already in Eon S, and these three lattices are isomorphic. Let H(β, γ) denote the set of all quasi-identities &j αj =⇒ ζk where the atomic formulae αj and ζk come from join representations β =(cid:87) conK αj and γ = (cid:87) conK ζk. Let ∆ and κ be the mappings from the Galois con- Q-LATTICES AS CONGRUENCE LATTICES 11 nection in the proof of Theorem 5. The semantic versions of the structural results of the preceding section then take the following form. Lemma 8. Let Q be a quasivariety contained in K. The set of all pairs (β, γ) such that Q satisfies each of the sentences in H(β, γ) is in Don S, where S = (cid:104)T,∨, 0,(cid:98)E(cid:105) with T the semilattice of compact congruences of ConK F, E = End F, and F = FK(ω). Lemma 9. Let Y be a collection of structures contained in K. The following are equivalent. (1) (β, γ) ∈ ∆(Y ). (2) Every A ∈ Y satisfies all the implications in H(β, γ). (3) The quasivariety SPU(Y ) satisfies all the implications in H(β, γ). Lemma 10. Let X ⊆ T 2, where T is as in Lemma 8. The following are equivalent for a structure A. (1) A ∈ κ(X). (2) For every pair (β, γ) ∈ X, A satisfies all the quasi-identities of H(β, γ). As always, it is good to understand both the semantic and logical view- point. 5. Congruence lattices of semilattices with operators Let us examine more closely lattices of the form Con(S, +, 0, F). The following theorem summarizes some fundamental facts about their structure. Theorem 11. Let (S, +, 0, F) be a semilattice with operators. (1) An ideal I of S is the 0-class of some congruence relation if and only if f (I) ⊆ I for every f ∈ F. (2) If the ideal I is F-closed, then the least congruence with 0-class I is η(I), the semilattice congruence generated by I. It is characterized by x η(I) y iff x + i = y + i for some i ∈ I. (3) There is also a greatest congruence with 0-class I, which we will denote by τ (I). To describe this, let F† denote the monoid generated by F, including the identity function. Then x τ (I) y iff (∀h ∈ F† ) h(x) ∈ I ⇐⇒ h(y) ∈ I. The proof of each part of the theorem is straightforward. As a sample application, it follows that if S is a simple semigroup with one operator, then S = 2. The maps η and τ from Theorem 11 induce operations on the entire congruence lattice Con(S, +, 0, F). If θ is a congruence with 0-class I, de- fine η(θ) = η(I) and τ (θ) = τ (I). The map η is known as the natural 12 KIRA ADARICHEVA AND J. B. NATION equa-interior operator on Con(S, +, 0, F). This terminology will be justified below. The natural equa-interior operator induces a partition of Con(S, +, 0, F). Theorem 12. Let S = (cid:104)S, +, 0, F(cid:105) be a semilattice with operators. The natural equa-interior operator partitions Con(S) into intervals [η(θ), τ (θ)] consisting of all the congruences with the same 0-class (which is an F-closed ideal). The natural equa-interior operator on the congruence lattice of a semilat- tice with operators plays a role dual to that of the equaclosure operator for lattices of quasivarieties. Adaricheva and Gorbunov [4], building on Dziobiak [9], described the natural equational closure operator on Q-lattices. In the dual language of theories, the restriction of quasi-equational theories to atomic formulae gives rise to an equa-interior operator (defined below) on QTh(K). Finitely based subvarieties of a quasi-variety K are given by quasi-identities that can be written as x ≈ x =⇒ &k βk for some atomic formulae βk. By Lemma 9, the corresponding congruences are of the form con(0, θ) where θ is a compact K-congruence on the free algebra FK(ω). More generally, subvarieties of K θ∈I con(0, θ) for some ideal I of the semilattice of compact K-congruences. Thus we should expect the map η to be the analogous interior operator on congruence lattices of semilattices with operators. correspond to joins of these, i.e., to congruences of the form(cid:87) We now define an equa-interior operator abstractly to have those prop- erties that we know to hold for the natural equa-interior operator on the lattice of theories of a quasivariety. One of our main goals, in this section and the next two, is to extend this list of known properties using the repre- sentation of the lattice of theories as the congruence lattice of a semilattice with operators. An equa-interior operator on an algebraic lattice L is a map η : L → L satisfying the following properties. (I1) η(x) ≤ x (I2) x ≥ y implies η(x) ≥ η(y) (I3) η2(x) = η(x) (I4) η(1) = 1 (I6) η(x) ∨ (y ∧ z) = (η(x) ∨ y) ∧ (η(x) ∨ z) (I7) The image η(L) is the complete join subsemilattice of L generated (I8) There is a compact element w ∈ L such that η(w) = w and the interval [w, 1] is isomorphic to the congruence lattice of a semilattice. (Thus the interval [w, 1] is coatomistic.) (I5) η(x) = u for all x ∈ X implies η((cid:87) X) = u by η(L) ∩ Lc. Property (I5) means that the operation τ is implicitly defined by η, via (cid:95){z ∈ L : η(z) = η(x)}. τ (x) = Q-LATTICES AS CONGRUENCE LATTICES 13 Thus τ (x) is the largest element z such that η(z) = η(x). Likewise, proper- ties (I1) and (I3) insure that η(x) is the least element z(cid:48) such that η(z(cid:48)) = η(x). By (I2), if η(x) ≤ y ≤ τ (x), then η(y) = η(x). Thus the kernel of η, defined by x ≈ y iff η(x) = η(y), is an equivalence relation that partitions L into disjoint intervals of the form [η(x), τ (x)]. We will refer to this as the equa-partition of L. Now τ is not order-preserving in general. However, it does satisfy a weak order property that can be useful. Lemma 13. Let L be an algebraic lattice, and assume that η satisfies con- ditions (I1) -- (I5). Define τ as above. Then for any subset {xj : j ∈ J} ⊆ L, τ ( (cid:94) τ xj) ≤(cid:94) xj) ≥ (cid:94) ητ xj ≤(cid:94) j∈J j∈J τ (xj). (cid:94) Proof. We have xj ≤(cid:94) and that's all in one block of the equa-partition, while(cid:86) xj ≤ τ ((cid:86) xj), which is the top of the same block. Thus(cid:86) τ xj ≤ τ ((cid:86) xj). τ xj (cid:3) η( Property (I7) has some nice consequences. Lemma 14. Let η be an equa-interior operator on an algebraic lattice L. (1) The image η(L) is an algebraic lattice, and x is compact in η(L) iff (2) If X is up-directed, then η((cid:87) X) =(cid:87) η(X). x ∈ η(L) and x is compact in L. For any quasivariety K, the natural equa-interior operator on the lattice of theories of K satisfies the eight listed basic properties. Congruence lattices of semilattices with operators come close. For an ideal I in a semilattice with operators, let conSL(I) denote the semilattice congruence generated by collapsing all the elements of I to 0. Theorem 15. If S = (cid:104)S, +, 0, F(cid:105) is a semilattice with operators, then the map η on Con S given by η(θ) = conSL(0/θ) satisfies properties (I1) -- (I7). Proof. Property (I6) is the hard one to verify. Let α, β, γ ∈ Con S and let ξ = η(α). Then x ξ y if and only if there exists z ∈ S such that z α 0 and x + z = y + z. (This is the semilattice congruence but it's compatible with F.) We want to show that (ξ ∨ β) ∧ (ξ ∨ γ) ≤ ξ ∨ (β ∧ γ). Let a, b ∈ LHS. Then there exist elements such that a β c1 ξ c2 β c3 . . . b a γ d1 ξ d2 γ d3 . . . b. Let z be the join of the elements witnessing the above ξ-relations. Then a ξ a + z β c1 + z = c2 + z β c3 + z = . . . b + z ξ b 14 KIRA ADARICHEVA AND J. B. NATION so that a ξ a+z β b+z ξ b, and similarly a ξ a+z γ b+z ξ b. Thus a, b ∈ RHS, (cid:3) as desired. Property (I8), on the other hand, need not hold in the congruence lattice of a semilattice with operators. The element w of (I8), called the pseudo-one, in lattices of quasi-equational theories corresponds to the identity x ≈ y. For an equa-interior operator on a lattice L with 1 compact, we can take w = 1; in particular, this applies when the semilattice has a top element, in which case we can take w = con(0, 1). But in general, there may be no candidate for the pseudo-one. Note that property (I8) implies that a lattice is dually atomic (or coatomic). Let x < 1 in L. If x ∨ w < 1 then it is below a coatom, while if x ∨ w = 1 then by the compactness of w there is a coatom above x that is not above w. In particular, the lattice of theories of a quasivariety is coatomic (Corol- lary 5.1.2 of Gorbunov [17]). Consider the semilattice Ω = (ω,∨, 0, p) with p(0) = 0 and p(x) = x − 1 for x > 0. Then Con Ω ∼= ω + 1, which has no pseudo-one (regardless of how η is defined). Thus Con Ω is not the dual of a Q-lattice. Likewise, Con Ω fails to be dually atomic. In each of the next two sections we will discuss an additional property of the natural equa-interior operator on semilattices with operators. The point of this is that an algebraic lattice cannot be the dual of a Q-lattice unless it admits an equa-interior operator satisfying all these conditions. Indeed, we should really consider the representation problem in the context of pairs (L, η), rather than just the representation of a lattice with an unspecified equa-interior operator. For the sake of clarity, let us agree that the term equa-interior operator refers to conditions (I1) -- (I8) for the remainder of the paper, even though we are proposing that henceforth a ninth condition should be included in the definition. 6. A new property of natural equa-interior operators The next theorem gives a property of the natural equa-partition on con- gruence lattices of semilattices with operators that need not hold in all lattices with an equa-interior operator. Theorem 16. Let S = (cid:104)S, +, 0, F(cid:105) be a semilattice with operators, and let η, τ denote the bounds of the natural equa-partition on Con S. If the congruences ζ, γ, χ satisfy η(ζ) ≤ η(γ) and τ (χ) ≤ τ (γ), then η(η(ζ) ∨ τ (ζ ∧ χ)) ≤ η(γ). Proof. Assume that ζ, γ, χ satisfy the hypotheses, and let 0/ζ = Z, 0/γ = C and 0/χ = X be the corresponding ideals. So Z ⊆ C and τ (X) ⊆ τ (C). For notation, let α = τ (Z ∩ X). Q-LATTICES AS CONGRUENCE LATTICES 15 We want to show that 0/(η(Z) ∨ α) ⊆ C, so let w ∈ LHS. For any z ∈ Z we have (z, w) ∈ η(Z)∨ α. Fix an element z0 ∈ Z. We claim that there exist elements z∗ ∈ Z and w∗ ∈ S such that z0 ≤ z∗ ≤ w∗, w ≤ w∗ and z∗ α w∗. There is a sequence z0 = s0 η(Z) s1 α s2 η(Z) s3 . . . sk = w. Let tj = s0 + ··· + sj for 0 ≤ j ≤ k. Thus we obtain z0 = t0 η(Z) t1 α t2 η(Z) t3 . . . tk with t0 ≤ t1 ≤ t2 ≤ t3 ≤ ··· ≤ tk. Put z(cid:48) = t1 and w(cid:48) = tk, so that with z0 ≤ z(cid:48) ∈ Z and w ≤ w(cid:48). Moreover, we may assume that k is minimal for such a sequence. If k > 2, then z(cid:48) = t1 α t2 η(Z) t3 α t4. By the definition of η(Z), there exists u ∈ Z such that t2 + u = t3 + u. Joining with u yields the shorter sequence (cid:48)(cid:48) z = t1 + u α t2 + u = t3 + u α t4 + u . . . contradicting the minimality of k. Thus k ≤ 2, which yields the conclusion of the claim with z∗ = t1 and w∗ = t2. Next, we claim that (z∗, w∗) ∈ τ (X). This follows from the sequence of implications: ∗ f (z ) ∈ X ∩ Z ) ∈ X =⇒ f (z ∗ =⇒ f (w ) ∈ X ∩ Z ∗ =⇒ f (w ) ∈ X ∗ =⇒ f (z ) ∈ X ∗ which hold for any f ∈ F, using the F-closure of Z, (z∗, w∗) ∈ τ (X ∩ Z) and z∗ ≤ w∗. Thus (z∗, w∗) ∈ τ (X) ⊆ τ (C). But z∗ ∈ Z ⊆ C = 0/τ (C), whence w∗ ∈ C and w ∈ C, as desired. (cid:3) For an application of this condition, consider the lattice K in Figure 1. It is straightforward to show that K has a unique equa-interior operator, with h(t) = 0 if t ≤ a and h(t) = t otherwise. Indeed, any equa-interior operator on K must have h(a)∨ (x∧ z) = (h(a)∨ x)∧ (h(a)∨ z), from which it follows easily that h(a) = 0. But then we cannot have h(x) = 0, else h(1) = h(a ∨ x) = 0, a contradiction. Thus h(x) = x and symmetrically h(z) = z. This in turn yields that h(c) = c. But K is not the congruence lattice of a semilattice with operators. The only candidate for the equa-interior operator fails the condition of Theo- rem 16 with the substitution ζ (cid:55)→ z, γ (cid:55)→ c, χ (cid:55)→ x. Therefore K is not the lattice of theories of a quasivariety. We could have also derived this latter fact by noting that K is not dually biatomic: in K we have a ≥ x ∧ z which is not refinable to a meet of coatoms. 16 KIRA ADARICHEVA AND J. B. NATION Figure 1. K On the other hand, K can be represented as a filterable sublattice of Con(B3, +, 0), where B3 is the Boolean lattice on three atoms. (See Appen- dix II for this terminology.) Indeed, if the atoms of B3 are p, q, r then we can take a (cid:55)→ [0] [p, q, r, p ∨ q, p ∨ r, q ∨ r, 1] c (cid:55)→ con(0, p ∨ q) x (cid:55)→ con(0, p) z (cid:55)→ con(0, q). We will pursue the comparison of congruence lattices and lattices of algebraic sets in the appendices. Taking a cue from this example, we continue investigating the conse- quences of the condition of Theorem 16. Recall that, whenever η satisfies (I1) -- (I5), we have η(y) = η(x) iff η(x) ≤ y ≤ τ (x). The condition can be written as follows, where we use the fact that η(u) ≤ c iff η(u) ≤ η(c). (†) τ (x) ≤ τ (c) & η(z) ≤ c =⇒ η(η(z) ∨ τ (x ∧ z)) ≤ c This holds for the natural equa-interior operator on congruence lattices of semilattices with operators, and we want to see how it applies to pairs (L, h) where h is an arbitrary equa-interior operator on L. putting c = η(z) ∨ τ (x). There is a two-variable version of the condition, which is obtained by (‡) η(η(z) ∨ τ (x ∧ z)) ≤ η(z) ∨ τ (x) This appears to be slightly weaker than (†). Consider the Boolean lattice B3 with atoms x, y, z and the equa-interior operator with h(y) = 0 and h(t) = t otherwise. Then (B3, h) fails the condi- tion (‡), though B3 is a dual Q-lattice with another equa-interior operator by Theorem 1. acxzFigure1.K1Q-LATTICES AS CONGRUENCE LATTICES 17 There are two additional conditions on equa-interior operators that are known to hold in the duals of Q-lattices: bicoatomicity and the four-coatom condition. (See Section 5.3 of Gorbunov [17].) Unfortunately, congruence lattices of semilattices with operators need not be coatomic (there is an example in the discussion of property (I8) in Section 5), but duals of Q- lattices are, so we will impose this as an extra condition. In that case, we will see that (†) implies both of these properties. A lattice L is bicoatomic (or dually biatomic) if whenever p is a coatom of L and p ≥ u ∧ v properly, then there exist coatoms c ≥ u and d ≥ v such that p ≥ c ∧ d. Theorem 17. Let L be a coatomic lattice and let h be an equa-interior operator on L. If (L, h) satisfies property (†), then L is bicoatomic. Proof. Assume 1 (cid:31) p ≥ u ∧ v properly in L. We want to find elements c, z with 1 (cid:31) c ≥ u, z ≥ v, and c ∧ z ≤ p. (Then apply the argument a second time.) Note that p ≥ η(p) ∨ (u ∧ v) = (η(p) ∨ u) ∧ (η(p) ∨ v). Put x = η(p) ∨ u and z = η(p) ∨ v. Let 1 (cid:31) c ≥ τ (x) and note τ (x) ≥ x ≥ u. Suppose c∧ z (cid:2) p. Put z(cid:48) = c∧ z. Then η(z(cid:48)) (cid:2) p, for else since η(p) ≤ z(cid:48) we would have η(z(cid:48)) = η(p) = η(z(cid:48) ∨ p) = η(1) = 1, a contradiction. Now we apply (†). Surely τ (x) ≤ c and η(z(cid:48)) ≤ z(cid:48) ≤ c. Moreover η(p) ≤ z(cid:48) ∧ x ≤ z ∧ x ≤ p whence η(z(cid:48) ∧ x) = η(p), and thus τ (z(cid:48) ∧ x) = p. But then η(η(z(cid:48)) ∨ τ (x ∧ z(cid:48))) = η(η(z(cid:48)) ∨ p) = η(1) = 1, again a contradiction. Therefore c ∧ z ≤ p, as desired. (cid:3) The dual of the four-coatom condition played a significant role in the characterization of the atomistic, algebraic Q-lattices. This too is a conse- quence of property (†). For coatoms a, d we write a ∼ d to indicate that ↑ (a ∧ d) = 4, in which case the filter ↑ (a ∧ d) is exactly {1, a, d, a ∧ d}. A lattice L with an equa-interior operator η satisfies the four-coatom condition if, whenever a, b, c, d are coatoms of L such that a ∼ d, η(a) (cid:2) d, η(c) ≤ d and η(c) = η(a ∧ b), then η(c) = η(b ∧ d). Theorem 18. The four-coatom condition holds in a lattice with an equa- interior operator η satisfying (†). Proof. As η(c) ≤ b, d is given, we need that η(b ∧ d) ≤ c. Supposing not, substitute x = a ∧ d, z = η(b ∧ d), and the element d into (†). Note that τ (a ∧ d) (cid:54)= a has η(a) (cid:2) d. Thus τ (a ∧ d) ≤ d, and of course η(b ∧ d) ≤ d. But we also have η(c) ≤ a ∧ b ∧ d ≤ a ∧ b and η(a ∧ b) = η(c), so η(η(b ∧ d) ∨ τ (a ∧ b ∧ d)) = η(η(b ∧ d) ∨ c) = η(1) = 1, a contradiction. Thus η(b ∧ d) ≤ c, as desired. (cid:3) 7. Coatomistic congruence lattices and a stronger property One of the most intriguing hypotheses about lattices of quasivarieties is formulated for atomistic lattices. Dually, it can be expressed as follows: 18 KIRA ADARICHEVA AND J. B. NATION Can every coatomistic lattice of quasi-equational theories be represented as Con(S, +, 0), i.e., without operators? This hypothesis is shown to be valid in the case when the lattice of quasi- equational theories is dually algebraic [3]. The problem provides a mo- tivation for investigating which coatomistic lattices can be represented as lattices of equational theories, or congruence lattices of semilattices, with or without operators. Consider the class M of lattices dual to Subf M, where M is an infinite semilattice with 0, and Subf M is the lattice of finite subsemilattices of M, topped by the semilattice M itself. Evidently, lattices in M are coatomistic, and they are algebraic but not dually algebraic. Besides, it is straightforward to show that they cannot be presented as Con(S, +, 0). Thus, it would be natural to ask whether such lattices can be presented as Con(S, +, 0, F), for a non-empty set of operators on S. In many cases the answer is "no" simply because there might be no equa-interior operator. For example, let M be a meet semilattice such that the dual of Subf M admits an equa-interior operator. If a is an element of M that can be expressed as a meet in infinitely many ways, then η(a) = 0 by Lemma 22 below. Hence M can contain at most one such element. It turns out to be feasible to show that certain lattices from M, that do ad- mit an equa-interior operator, still cannot be represented as Con(S, +, 0, F). The crucial factor here is to understand the behavior of infinite meets of coatoms, or more generally infinite meets of elements τ (x), in the congruence lattice of a semilattice with operators. The restriction given by Theorem 19 can be expressed as a ninth basic property of the natural equa-interior op- erator (as it implies (†)). Aside: Coatoms arise naturally in another context, that does not make the lattice coatomistic. Suppose S = (cid:104)S, +, 0, F(cid:105) has the property that for each F-closed ideal I, every f ∈ F, and every x ∈ S, f (x) ∈ I =⇒ x ∈ I. Then the congruence τ (I) partitions S into I and S − I, and hence is a coatom. In particular, this property holds whenever • F is empty, or • F is a group, or • every f ∈ F is increasing, i.e., x ≤ f (x) for all x ∈ S. In all these cases, τ (θ) is a coatom for every θ ∈ Con S. We will be partic- ularly concerned with the case when F is a group in Part II [7]. Theorem 19. Let S = (cid:104)S, +, 0, F(cid:105) be a semilattice with operators, I an arbitrary index set, and χ, γ, and ζi for i ∈ I congruences on S. The natural (cid:86) equa-interior operator on Con S has the following property: if η(χ) ≤ γ and i∈I τ (ζi) ≤ τ (γ), then η(η(χ) ∨(cid:94) i∈I τ (χ ∧ ζi)) ≤ γ. Q-LATTICES AS CONGRUENCE LATTICES 19 For the proof, it is useful to write down abstractly the two parts of the argument of the proof of Theorem 16. Lemma 20. Let α, χ, ζ ∈ Con(S, +, 0, F) and let X be the 0-class of χ. u ≤ u∗ ∈ X, v ≤ v∗, u∗ ≤ v∗, and (u∗, v∗) ∈ α. (1) If u ∈ X and (u, v) ∈ χ ∨ α, then there exist elements u∗, v∗ with (2) If u ∈ X, u ≤ v and (u, v) ∈ τ (χ ∧ ζ), then (u, v) ∈ τ (ζ). Now, under the assumptions of the theorem, let u ∈ X and (u, v) ∈ η(χ)∨ (cid:86) τ (χ∧ζi), so that v is in the 0-class of the LHS. Then by Lemma 20(1), there exist u∗, v∗ with u ≤ u∗ ∈ X, v ≤ v∗, u∗ ≤ v∗ and (u∗, v∗) ∈(cid:86) τ (χ ∧ ζi). for every i, so that (u∗, v∗) ∈(cid:86) τ (ζi). we have u∗ ∈ X ⊆ C, and (u∗, v∗) ∈(cid:86) τ (ζi) ≤ τ (γ), so v∗ ∈ C as well. A Then (u∗, v∗) ∈ τ (χ∧ζi) for every i, whence by Lemma 20(2) (u∗, v∗) ∈ τ (ζi) Let X and C denote the 0-classes of χ and γ, respectively. By assumption, fortiori, v ∈ C, as desired. (I9) For any index set I, if η(x) ≤ c and (cid:86) τ (zi) ≤ τ (c), then η(η(x) ∨ This proves Theorem 19. Thus we obtain the ninth fundamental prop- erty of the natural equa-interior operator on the congruence lattice of a semilattice with operators. i∈I τ (x ∧ zi)) ≤ c. (cid:86) As before, there is also a slightly simpler (and weaker) variation: η(η(x) ∨(cid:94) τ (x ∧ zi)) ≤ η(x) ∨(cid:94) τ (zi). (cid:48) (I9 ) i∈I Clearly, if I = 1 then property (I9) reduces to property (†). In fact, for I finite, (†) implies (I9). But for I infinite, property (I9) seems to carry a rather different sort of information, as we shall see below. Consider the case when I = 2; the argument for the general finite case is similar. Assume that η(x) ≤ c and τ (y) ∧ τ (z) ≤ τ (c). Using (I6), (†), and the fact that η(u ∧ v) = η(η(u) ∧ η(v)), we calculate η(η(x) ∨ (τ (x ∧ y) ∧ τ (x ∧ z))) = η((η(x) ∨ (τ (x ∧ y)) ∧ (η(x) ∨ τ (x ∧ z)))) ≤ η((η(x) ∨ (τ (y)) ∧ (η(x) ∨ τ (z)))) = η(η(x) ∨ (τ (y) ∧ τ (z))) ≤ c as desired. With property (I9) as a tool-in-hand, we turn to a thorough investiga- tion of the (dual) dependence relation for coatoms of Con(S, +, 0, F); see Theorems 23 and 24 below. Throughout the remainder of this section, χ, ζ and α will denote distinct coatoms of the congruence lattice. Repeatedly, we use the basic property of equa-interior operators that ηx ∨ (y ∧ z) = (ηx ∨ y) ∧ (ηx ∨ z). Our goal is to generalize (to whatever extent possible) the following property of finite sets of coatoms. 20 KIRA ADARICHEVA AND J. B. NATION Theorem 21. Let L be a lattice with an equa-interior operator. If for coatoms x, z1, . . . , zk, a1, . . . , ak of L we have x ∧ zi ≤ ai properly, then ηx ∨(cid:86)k i=1 zi = 1. The proof uses the next lemma. Lemma 22. Suppose x ∧ z ≤ a properly for coatoms in a lattice with an equa-interior operator. Then ηa ≤ x ∧ z, and thus (1) τ (x ∧ z) = a, (2) ηx (cid:2) a, (3) ηx (cid:2) z. Proof. If say ηa (cid:2) x, then ηa ∨ x = 1, and using (I6) we would have ηa ∨ z = (ηa ∨ x) ∧ (ηa ∨ z) = ηa ∨ (x ∧ z) ≤ a whence z ≤ a, a contradiction. So ηa ≤ x, and symmetrically ηa ≤ z. Since ηa ≤ x ∧ z ≤ a = τ a, we have τ (x ∧ z) = a. It follows that we cannot have ηx ≤ a, else ηa = η(x ∧ z) ≤ ηx ≤ a, implying that ηx = ηa, and thus ηa = η(x ∨ a) = η1 = 1 by (I5) and (I4), a contradiction. Therefore also ηx (cid:2) z, else ηx ≤ x ∧ z ≤ a. (cid:3) The theorem now follows immediately, because ηx ∨ k(cid:94) k(cid:94) zi = (ηx ∨ zi) = 1. i=1 i=1 The property of Theorem 21 can fail when there are infinitely many zi's, even in the congruence lattice of a semilattice. Let Q be the join semilattice in Figure 2. Consider the ideals X = {0, u1, u2, u3, . . .} Zi =↓ vi Ai =↓ ui for i ∈ ω, and let χ = τ (X), ζi = τ (Zi) and αi = τ (Ai). Then an easy calculation shows that(cid:86) ζi = 0, and the infinite version of the property of the theorem fails. Nonetheless, we shall show that a couple of infinite versions do hold. Theorem 23. Let L be a lattice with an equa-interior operator satisfying property (I9). If for coatoms a, x and zi (i ∈ I) of L we have x ∧ zi ≤ a properly, then ηx ∨(cid:86) ηx ∨(cid:86) τ (x ∧ zi) = 1. Then property (I9(cid:48)) gives the conclusion immediately.(cid:3) Proof. By Lemma 22, we have τ (x ∧ zi) = a for every i, and ηx (cid:2) a. Hence i∈I zi = 1. Q-LATTICES AS CONGRUENCE LATTICES 21 Figure 2. Con(S, +, 0) does not satisfy the infinite analogue of Theorem 21. Theorem 24. Let L be a lattice with an equa-interior operator satisfying property (I9). Let x, ai and zi be coatoms of L with x∧ zi ≤ ai properly for all i ∈ I. If (cid:86) i∈I ai (cid:2) x, then(cid:86) i∈I zi (cid:2) x. Proof. Again, by Lemma 22, we have τ (x ∧ zi) = ai for every i. Now apply (cid:3) (I9) directly with c = x. Let us now use these results to show that certain coatomistic lattices are not lattices of quasi-equational theories. Call an infinite (∧)-semilattice M cute if it has an element a and different elements m, mj ∈ M\{a}, j ∈ ω, with m ∧ mj = a. Examples of cute semilattices are M∞: countably many mi covering the least element a, or M2: a chain {mj, j ∈ ω} in addition to elements m, a, satisfying m∧mj = a for all j. It was asked in [2] (p. 175), in connection with the hypothesis about the atomistic Q-lattices mentioned above in the dual form, whether Subf M∞ is a Q-lattice. The following result, an immediate application of Theorem 23, answers this question in the negative. Theorem 25. If M is a cute semilattice, then the dual of Subf M is not representable as Con(S, +, 0, F). Hence Subf M is not a Q-lattice. It would be desirable to extend Theorem 25 to all lattices from M. In particular, we may ask about possibility to represent L = (Subf P1)d, where the semilattice P1 consists of two descending chains {bi, i ∈ ω}, {ai, i ∈ ω} with defining relations ai+1 = ai ∧ bi+1, b0 > a0. Every equa-interior operator η on L would satisfy: η({ai}) = [ai, b0], η({bi}) ≥ [bi, b0]. In particular, η(c) = 0, c ∈ L, implies c = 0 (equivalently, 01v3v2v1u1u2u3122 KIRA ADARICHEVA AND J. B. NATION τ (0) = 0). This makes P1 drastically different from cute semilattices. Is the dual of Subf P1 representable as Con(S, +, 0, F)? Another interesting case to consider would be Subf C where C is an infinite chain, so that every finite subset of C is a subsemilattice. 8. Appendix I: Complete sublattices of subalgebras In the first two appendices, we analyze conditions that were used in older descriptions of lattices of quasivarieties; see Gorbunov [17]. Note that Con(S, +, 0, F) is a complete sublattice of Con(S, +, 0), which is dually isomorphic to Sp(I(S)), which is the lattice of subalgebras of an infinitary algebra. (Joins of non-directed sets can be set to 1.) In this context we are considering complete sublattices of Sub A where A is a semilattice, or a complete semilattice, or a complete algebra of algebraic subsets. Let ε be a binary relation on a set S. A subset X ⊆ S is said to be ε-closed if c ∈ X and c ε d implies d ∈ X. Recall that a quasi-order ε on a semilattice S = (cid:104)S,∧, 1(cid:105) is distributive if (1) If c1 ∧ c2 ε d then there exist elements d1, d2 such that ci ε di and it satisfies the following conditions. d = d1 ∧ d2. (2) If 1 ε d then d = 1. The effect of the next result is that for a semilattice S, any complete sublattice of Sub S can be represented as the lattice of all ρ-closed subsemi- lattices, for some distributive quasi-order ρ. Theorem 26. Let S = (cid:104)S,∧, 1(cid:105) be a semilattice with 1, and let ε be a distributive quasi-order on S. Then Sub (S, ε), the lattice of all ε-closed subsemilattices (with 1), is a complete sublattice of Sub S. Conversely, let T be a complete sublattice of Sub S. Define a relation ρ on S by c ρ d if for all X ∈ T we have c ∈ X =⇒ d ∈ X. Then ρ is a distributive quasi-order, and T consists precisely of the ρ-closed subsemilattices of S. Furthermore, ρ satisfies the following conditions. (3) If c ρ d1, d2 then c ρ d1 ∧ d2. (4) For all c ∈ S, c ρ 1. The correspondence between complete sublattices of Sub S and distribu- tive quasi-orders satisfying (3) and (4) is a dual isomorphism. The proof is relatively straightforward. The description of all complete sublattices of Sub S, the lattice of all complete subsemilattices of a complete semilattice S, is almost identical, except that complete meets appear in the conditions. (1)(cid:48) If(cid:86) ci ε d then there exist elements di such that ci ε di and d =(cid:86) di. (3)(cid:48) If c ε ci for all i, then c ε(cid:86) ci. Complete semilattices satisfying (1)(cid:48) are called Brouwerian by Gorbunov [17]. The results can be summarized thusly. Q-LATTICES AS CONGRUENCE LATTICES 23 Theorem 27. Let S = (cid:104)S,∧, 1(cid:105) be a complete semilattice. Then there is a dual isomorphism between complete sublattices of Sub S and quasi-orders satisfying conditions (1)(cid:48), (2), (3)(cid:48) and (4). For complete sublattices of Sp(A), the lattice of algebraic subsets of an algebraic lattice A, we must also deal with joins of nonempty up-directed subsets, and once A fails the ACC matters get more complicated. A quasi- order ε on A is said to be continuous if it has the following property. (5) If C is a directed set and(cid:87) C ε d, then there exists a directed set D such that d =(cid:87) D and for each d ∈ D there exists c ∈ C with c ε d. This is a very slight weakening of Gorbunov's definition [17]. As above, we have this result of Gorbunov. Theorem 28. Let ε be a continuous Brouwerian quasi-order on a complete lattice A. Then Sp(A), the lattice of ε-closed algebraic subsets, is a complete sublattice of Sp(A). Now for any algebra B we can define the embedding relation E on Con B by θ E ψ if B/ψ ≤ B/θ. A fundamental result of Gorbunov characterizes Q-lattices in terms of the embedding relations (Corollaries 5.2.2 and 5.6.8 of [17]). Theorem 29. Let K be a quasivariety and let F = FK(ω). The embedding relation is a continuous Brouwerian quasi-order on ConK F, and Lq(K) ∼= Sp(ConK(F, E)). For comparison, we note that the isomorphism relation need not be con- tinuous; see Gorbunov [17], Example 5.6.6. We do not know (and doubt) that the relation ρ corresponding to a com- plete sublattice of Sp(A) need always be continuous. However, our repre- sentation of Con(S, +, 0, F) as dually isomorphic to a complete sublattice of Sp(I(S)) could be unraveled to give the ρ relation explicitly in that case. Are these particular relations always continuous? 9. Appendix II: Filterability and equaclosure operators The natural equational closure operator on Lq(K) is given by the map h(Q) = H(Q) ∩ K for quasivarieties Q ⊆ K. That is, h(Q) consists of all members of K that are in the variety generated by Q, or equivalently, that are homomorphic images of FQ(X) for some set X. For the corresponding map on Sp(Con FK(ω)), let X be the algebraic subset of all Q-congruences and the filter ↑ ϕ is the algebraic subset associated with h(Q), that is, all h(Q)-congruences of Con FK(ω). of Con FK(ω). Then ϕ =(cid:86) X is the natural congruence with F/ϕ ∼= FQ(ω), Then it is not hard to see that the map h(X) =↑(cid:86) X on Sp(A, ε) will satisfy the duals of conditions (I1) -- (I7) so long as ↑(cid:86) X is ε-closed for every Abstractly, let ε be a distributive quasi-order on an algebraic lattice A. X ∈ Sp(A, ε). A quasi-order that satisfies this crucial condition, c ≥(cid:94) X & c ε d =⇒ d ≥(cid:94) 24 KIRA ADARICHEVA AND J. B. NATION erator h(X) =↑(cid:86) X on Sp(A, ε) is again called the natural closure operator is said to be filterable. If the quasi-order ε is filterable, then the closure op- X determined by ε. We can also speak of a complete sublattice of Sp(A) as being filterable if the quasi-order it induces via Theorem 26 is so. Dually, a sublattice T ≤ Con(S, +, 0) is filterable if, for each θ ∈ T, the semilattice congruence generated by the 0-class of θ is in T. As we have observed, this is the case when T = Con(S, +, 0, F) for some set of operators F. Thus we obtain a slightly different perspective on Theorem 15. Theorem 30. For a semilattice S with operators, T = Con(S, +, 0, F) is a filterable complete sublattice of Con(S, +, 0). Thus T supports the natural interior operator h(θ) = con(0/θ), which satisfies conditions (I1) -- (I7). In fact, the natural interior operator on Con(S, +, 0, F) also satisfies con- dition (I9). However, as we saw in Section 6, a filterable sublattice of Con(S, +, 0) may fail condition (†), which is the finite index case of (I9), even with S finite. Thus being a congruence lattice of a semilattice with operators is a stronger property than just being a filterable sublattice of Con(S, +, 0). 10. Appendix III: Lattices of equational theories In this appendix, we summarize what is known about lattices of equational theories. Throughout the section, V will denote a variety of algebras, with no relation symbols in the signature. For this situation, atomic theories really are equational theories. The lattice of equational theories is, of course, dual to the lattice of subvarieties of V. From the basic representation ATh(V) ∼= Ficon FV(ω), we see that the lattice is algebraic. Its top element 1 has the basis x ≈ y, and thus 1 is compact. On the other hand, J. Jezek proved that any algebraic lattice with countably many compact elements is isomorphic to an interval in some lattice of equational theories [24]. R. McKenzie showed that every lattice of equational theories is isomorphic to the congruence lattice of a groupoid with left unit and right zero [28]. N. Newrly refined these ideas, showing that a lattice of equational theories is isomorphic to the congruence lattice of a monoid with a right zero and one additional unary operation [29]. A. Nurakunov added a second unary operation and proved a converse: a lattice is a lattice of equational theories if and only if it is the congruence lattice of a monoid with a right zero and two unary operations satisfying certain properties [31]. Nurakunov's conditions are rather technical, but they just codify the prop- erties of the natural operations on the free algebra FV(X) that they model. If X = {x0, x1, x2, . . .} and s, t are terms, then s · t = t(s, x1, x2, . . . ). Q-LATTICES AS CONGRUENCE LATTICES 25 The two unary operations are the endomorphism ϕ+ and ϕ−, where ϕ+(xi) = xi+1 for all i, while ϕ−(x0) = x0 and ϕ−(xi) = xi−1 for i > 0. W.A. Lampe used McKenzie's representation to prove that lattices of equational theories satisfy a form of meet semidistributivity at 1, the so- called Zipper Condition [26]: (cid:95) i∈I If ai ∧ c = z for all i ∈ I and ai = 1, then c = z. A similar but stronger condition was found by M. Ern´e [10] and G. Tar- dos (independently), which was refined yet further by Lampe [27]. These results show that the structure of lattices of equational theories is quite con- strained at the top, whereas Jezek's theorem shows that this is not the case globally. Confirming this heuristic, D. Pigozzi and G. Tardos proved that every algebraic lattice with a completely join irreducible greatest element 1 is isomorphic to a lattice of equational theories [32]. Again, we propose that one should investigate ATh(V) for varieties of structures. The authors would like to thank the referee for many helpful comments. References [1] M. Adams, K. Adaricheva, W. Dziobiak and A. Kravchenko, Open questions related to the problem of Birkhoff and Maltsev, Studia Logica 78 (2004), 357 -- 378. [2] K.V. Adaricheva, Lattices of algebraic subsets, Alg. Univ. 52 (2004), 167 -- 183. [3] K.V. Adaricheva, W. Dziobiak, and V.A. Gorbunov, Algebraic atomistic lattices of quasivarieties, Algebra and Logic 36 (1997), 213 -- 225. [4] K.V. Adaricheva and V.A. Gorbunov, Equational closure operator and forbidden semidistributive lattice, Sib. Math. J. 30 (1989), 831 -- 849. [5] K.V. Adaricheva, V.A. Gorbunov and V.I. Tumanov, Join-semidistributive lattices [6] K.V. Adaricheva and J.B. Nation, Equaclosure operators on join semidistributive and convex geometries, Advances in Math. 173 (2003), 1 -- 49. lattices, manuscript available at www.math.hawaii.edu/∼jb. [7] K.V. Adaricheva and J.B. Nation, Lattices of quasi-equational theories as con- gruence lattices of semilattices with operators, Part II, preprint available at www.math.hawaii.edu/∼jb. [8] G. Birkhoff, On the structure of abstract algebras, Proc. Camb. Phil. Soc. 31 (1935), 433 -- 354. [9] W. Dziobiak, On atoms in the lattice of quasivarieties, Algebra Universalis 24 (1987), 32 -- 35. [10] M Ern´e, Weak distributive laws and their role in lattices of congruences and equational theories, Alg. Univ. 25 (1988), 290 -- 321. [11] S. Fajtlowicz and J. Schmidt, B´ezout families, join-congruences and meet-irreducible ideals, Lattice Theory (Proc. Colloq., Szeged, 1974), pp. 51 -- 76, Colloq. Math. Soc. Janos Bolyai, Vol. 14, North Holland, Amsterdam, 1976. [12] R. Freese, K. Kearnes and J. Nation, Congruence lattices of congruence semidistribu- tive algebras, Lattice Theory and its Applications, Darmstadt 1991), pp. 63 -- 78, Res. Exp. Math. 23, Heldermann, Lemgo, 1995. [13] R. Freese and J. Nation, Congruence lattices of semilattices, Pacific J. Math 49 (1973), 51 -- 58. 26 KIRA ADARICHEVA AND J. B. NATION [14] V. Gorbunov, Covers in lattices of quasivarieties and independent axiomatizability, Algebra Logika, 14 (1975), 123 -- 142. [15] V. Gorbunov, The cardinality of subdirectly irreducible systems in quasivarieties, Al- gebra and Logic, 25 (1986), 1 -- 34. [16] V. Gorbunov, The structure of lattices of quasivarieties, Algebra Universalis, 32 (1994), 493 -- 530. [17] V. Gorbunov, Algebraic Theory of Quasivarieties, Siberian School of Algebra and Logic, Plenum, New York, 1998. [18] V. Gorbunov and V. Tumanov, A class of lattices of quasivarieties, Algebra and Logic, 19 (1980), 38 -- 52. [19] V. Gorbunov and V. Tumanov, On the structure of lattices of quasivarieties, Sov. Math. Dokl., 22 (1980), 333 -- 336. [20] V. Gorbunov and V. Tumanov, Construction of lattices of quasivarieties, Math. Logic and Theory of Algorithms, 12 -- 44, Trudy Inst. Math. Sibirsk. Otdel. Adad. Nauk SSSR, 2 (1982), Nauka, Novosibirsk. [21] H.-J. Hoehnke, Fully invariant algebraic closure systems of congruences and quasi- varieties of algebras, Lectures in Universal Algebra (Szeged, 1983), 189 -- 207, Colloq. Math. Soc. J´anos Bolyai, 43, North-Holland, Amsterdam, 1986. [23] T. Holmes, D. Kitsuwa, J. Nation and S. Tamagawa, Lattices of atomic theories in [22] R. Hofmann, M. Mislove and A. Stralka, The Pontryagin Duality of Compact 0- dimensional Semilattices and its Applications, Lecture Notes in Math., Vol. 396, Springer, Berlin, 1974. languages without equality, preprint available at www.math.hawaii.edu/∼jb. [24] J. Jezek, Intervals in the lattice of varieties, Alg. Univ. 6 (1976), 147 -- 158. [25] K. Kearnes and J.B. Nation, Axiomatizable and nonaxiomatizable congruence preva- rieties, Alg. Univ. 59 (2008), 323 -- 335. [26] W.A. Lampe, A property of the lattice of equational theories, Alg. Univ. 23 (1986), 61 -- 69. [27] W.A. Lampe, Further properties of lattices of equational theories, Alg. Univ. 28 (1991), 459 -- 486. [28] R. McKenzie, Finite forbidden lattices, Universal Algebra and Lattice Theory, Lecture Notes in Mathematics 1004(1983), Springer-Verlag, Berlin, 176 -- 205. [29] N. Newrly, Lattices of equational theories are congruence lattices of monoids with one additional unary operation, Alg. Univ. 30 (1993), 217 -- 220. [30] J.B. Nation, Lattices of theories in languages without equality, to appear in Notre Dame Journal of Formal Logic. [31] A.M. Nurakunov, Equational theories as congruences of enriched monoids, Alg. Univ. 58 (2008), 357 -- 372. [32] D. Pigozzi and G. Tardos, The representation of certain abstract lattices as lattices of subvarieties, manuscript, 1999. [33] R. Quackenbush, Completeness theorems for universal and implicational logics of algebras via congruences, Proc. Amer. Math. Soc. 103 (1988), 1015 -- 1021. [34] E.T. Schmidt, Kongruenzrelationen algebraische Strukturen, Deutsch. Verlag Wis- sensch., Berlin, 1969. [35] V. Tumanov, On quasivarieties of lattices, XVI All-Union Algebraic Conf., Part 2, 135, Leningrad, 1981. Q-LATTICES AS CONGRUENCE LATTICES 27 Department of Mathematical Sciences, Yeshiva University, New York, NY 10016, USA E-mail address: [email protected] Department of Mathematics, University of Hawaii, Honolulu, HI 96822, USA E-mail address: [email protected]
1203.1405	1	1203	2012-03-07T09:00:49	A Combinatorial Discussion on Finite Dimensional Leavitt Path Algebras	[ "math.RA" ]	Any finite dimensional semisimple algebra A over a field K is isomorphic to a direct sum of finite dimensional full matrix rings over suitable division rings. In this paper we will consider the special case where all division rings are exactly the field K. All such finite dimensional semisimple algebras arise as a finite dimensional Leavitt path algebra. For this specific finite dimensional semisimple algebra A over a field K, we define a uniquely detemined specific graph - which we name as a truncated tree associated with A - whose Leavitt path algebra is isomorphic to A. We define an algebraic invariant {\kappa}(A) for A and count the number of isomorphism classes of Leavitt path algebras with {\kappa}(A)=n. Moreover, we find the maximum and the minimum K-dimensions of the Leavitt path algebras of possible trees with a given number of vertices and determine the number of distinct Leavitt path algebras of a line graph with a given number of vertices.	math.RA	math	A Combinatorial Discussion on Finite Dimensional Leavitt Path Algebras A. Ko¸c(1), S. Esin(2), I. Guloglu(2), M. Kanuni(3) (1) Istanbul Kultur University Department of Mathematics and Computer Sciences (2) Dogu¸s University Department of Mathematics (3) Bogazi¸ci University Department of Mathematics November 7, 2018 Abstract Any finite dimensional semisimple algebra A over a field K is isomor- phic to a direct sum of finite dimensional full matrix rings over suitable division rings. In this paper we will consider the special case where all division rings are exactly the field K. All such finite dimensional semisim- ple algebras arise as a finite dimensional Leavitt path algebra. For this specific finite dimensional semisimple algebra A over a field K, we de- fine a uniquely detemined specific graph - which we name as a truncated tree associated with A - whose Leavitt path algebra is isomorphic to A. We define an algebraic invariant κ(A) for A and count the number of isomorphism classes of Leavitt path algebras with κ(A) = n. Moreover, we find the maximum and the minimum K-dimensions of the Leavitt path algebras of possible trees with a given number of vertices and determine the number of distinct Leavitt path algebras of a line graph with a given number of vertices. Keywords: Finite dimensional semisimple algebra, Leavitt path algebra, Trun- cated trees, Line graphs. 1 Introduction By the well-known Wedderburn-Artin Theorem [2], any finite dimensional semisimple algebra A over a field K is isomorphic to a direct sum of finite dimensional full matrix rings over suitable division rings. In this paper we will consider the special case where all division rings are exactly the field K. All such finite dimensional semisimple algebras arise as a finite dimensional Leavitt 1 path algebra as studied in [1]. The Leavitt path algebras are introduced by Abrams and Aranda Pino in 2005, [3]. Many papers on Leavitt path algebras appeared in literature since then. In the following discussion, we are particularly interested in answering some combinatorial questions on the finite dimensional Leavitt path algebras. We start by recalling the definitions of a path algebra and a Leavitt path algebra, see [1]. A directed graph E = (E0, E1, r, s) consists of two countable sets E0, E1 and functions r, s : E1 → E0. The elements E0 and E1 are called vertices and edges, respectively. For each e ∈ E0, s(e) is the source of e and r(e) is the range of e. If s(e) = v and r(e) = w, then we say that v emits e and that w receives e. A vertex which does not receive any edges is called a source, and a vertex which emits no edges is called a sink. A graph is called row- finite if s−1(v) is a finite set for each vertex v. For a row-finite graph the edge set E1 of E is finite if its set of vertices E0 is finite. Thus, a row-finite graph is finite if E0 is a finite set. A path in a graph E is a sequence of edges µ = e1 . . . en such that r(ei) = s(ei+1) for i = 1, . . . , n − 1. In such a case, s(µ) := s(e1) is the source of µ and r(µ) := r(en) is the range of µ, and n is the length of µ, i.e., l(µ) = n. If s(µ) = r(µ) and s(ei) 6= s(ej) for every i 6= j, then µ is called a cycle. If E does not contain any cycles, E is called acyclic. For n ≥ 2, define En to be the set of paths of length n, and E∗ = Sn≥0 the set of all paths. En The path K-algebra over E is defined as the free K-algebra K[E0 ∪ E1] with the relations: (1) vivj = δij vi for every vi, vj ∈ E0. (2) ei = eir(ei) = s(ei)ei for every ei ∈ E1. This algebra is denoted by KE. Given a graph E, define the extended graph i ei ∈ E1} of E as the new graph bE = (E0, E1 ∪ (E1)∗, r′, s′) where (E1)∗ = {e∗ and the functions r′ and s′ are defined as r′E 1 = r, s′E 1 = s, r′(e∗ i ) = s(ei) and s′(e∗ i ) = r(ei). The Leavitt path algebra of E with coefficients in K is defined as the path (CK1) e∗ algebra over the extended graph bE, with relations: (CK2) vi =P{ej ∈E 1 s(ej )=vi} eje∗ i ej = δij r(ej ) for every ej ∈ E1 and e∗ j i ∈ (E1)∗. for every vi ∈ E0 which is not a sink. This algebra is denoted by LK(E). The conditions (CK1) and (CK2) are called the Cuntz-Krieger relations. In particular condition (CK2) is the Cuntz- Krieger relation at vi. If vi is a sink, we do not have a (CK2) relation at vi. Note that the condition of row-finiteness is needed in order to define the equation (CK2). The main structure theorem in [1] can be summarized as follows: For any v ∈ E0, we define n(v) = {α ∈ E∗ r(α) = v} . 2 Proposition 1 : 1. The Leavitt path algebra LK(E) is a finite-dimensional K-algebra if and only if E is a finite and acyclic graph. sLi=1 2. If A = Mni(K) , then A ∼= LK(E) for a graph E having s con- nected components each of which is an oriented line graph with ni vertices, i = 1, 2, · · · , s. 3. A finite dimensional K-algebra A arises as a LK(E) for a graph E if and only if A = Mni(K). sLi=1 sLi=1 4. If A = Mni(K) and A ∼= LK(E) for a finite, acyclic graph E, then the number of sinks of E is equal to s, and each sink vi (i = 1, 2, · · · , s) has n(vi) = ni with a suitable indexing of the sinks. 2 Truncated Trees For a finite dimensional Leavitt path algebra LK(E) of a graph E, we would like to construct a distinguished graph F having the Leavitt path algebra isomorphic to LK(E) as follows: Theorem 2 Let E be a finite, acyclic graph with no isolated points. Let s = S(E) where S(E) is the set of sinks of E and N = max{n(v) v ∈ S(E)}. Then there exists a unique (up to isomorphism) tree F with exactly one source and s + N − 1 vertices such that LK(E) ∼= LK(F ). Proof. Let the sinks v1, v2, . . . , vs of E be indexed such that 2 ≤ n(v1) ≤ n(v2) ≤ . . . ≤ n(vs) = N. Define a graph F = (F 0, F 1, r, s) as follows: F 0 = {u1, u2, . . . , uN , w1, w2, . . . ws−1} F 1 = {e1, e2, . . . , eN −1, f1, f2, . . . , fs−1} s(ei) = ui s(fi) = un(vi)−1 and r(ei) = ui+1 and r(fi) = wi i = 1, . . . , N − 1 i = 1, . . . , s − 1. 3 Clearly, F is a directed tree with unique source u1 and s + N − 1 vertices. F has exactly s sinks, namely uN , w1, w2, . . . ws−1 with n(uN ) = N, n(wi) = n(vi), i = 1, . . . , s − 1. Therefore, LK(E) ∼= LK(F ). For the uniqueness part, take a tree T with exactly one source and s + N − 1 vertices such that LK(E) ∼= LK(T ). Since N = max{n(v) v ∈ S(E)} which is equal to the square root of the maximum of the K-dimensions of the minimal ideals of LK(E) and hence LK(T ), there exists a sink v in T with {µi ∈ T ∗ r(µi) = v} = N. On the other hand, since T is a tree with a unique source and hence any vertex is connected to the unique source by a uniquely determined path, we see that the unique path joining v to the source must contain exactly N vertices, say a1, ..., aN −1, v where a1 is the unique source and the length of the path joining ak to a1 being equal to k − 1 for any k = 1, 2, ..., N − 1. As LK(E) = Mni(K) with s summands, the remaining s − 1 vertices must then all be sinks by Proposition 1 (4), say b1, ..., bs−1. Since for any vertex a different from the unique source we have n(a) > 1 we see that for each i = 1, . . . , s − 1 there exists an edge gi with r(gi) = bi. Since s(gi) is not a sink we see that s(gi) ∈ {a1, a2, ..., aN −1}, more precisely s(gi) = an(bi)−1, i = 1, 2, ..., s − 1. Thus T is isomorphic to F. Observe that the F constructed in Theorem 2 is the tree with one source and smallest possible number of vertices (s + N − 1) having LK(F ) isomorphic to LK(E). We call F constructed in Theorem 2 as the truncated tree associated with E. Proposition 3 With the above definition of F , T 0 < F 0 such that LK(T ) ∼= LK(F ). there is no tree T with sLi=1 sLi=1 Proof. Notice that since T is a tree, any vertex contributing to a sink represents a unique path ending at that sink. Assume on the contrary there exists a tree T with n vertices and LK(T ) ∼= A = Mni(K) such that n < s + N − 1. Since N is the maximum of ni's there exists a sink with N vertices contributing. But in T the number n − s of vertices which are not sinks is less than N − 1. Hence the maximum contribution to any sink can be at most n − s + 1 which is strictly less than N. This is the desired contradiction. However if we omit the tree assumption then it is possible to find a graph G with smaller number of vertices having LK(G) isomorphic to LK(E) as the next example illustrates. Example 4 Both LK(G) ∼= M3(K) ∼= LK(F ) and G0 = 2 where as F 0 = 3. 4 Given F1, F2 truncated trees associated with graphs G1 and G2 respectively, then F1 ∼= F2 iff LK(F1) ∼= LK(F2) so there is a one-to-one correspondence between the Leavitt path algebra and truncated trees. For a given finite dimensional Leavitt path algebra A = Mni(K) with sLi=1 2 ≤ n1 ≤ n2 ≤ . . . ≤ ns = N, the number s is the number of minimal ideals of A and N 2 is the maximum of the dimensions of these ideals. Therefore κ(A) = s + N − 1 is a uniquely determined algebraic invariant of A. Given m ≥ 2, the number of isomorphism classes of finite dimensional Leavitt path algebras A which do not have any ideals isomorphic to K and κ(A) = m is equal to the number of distinct truncated trees with m vertices by the previous paragraph. The next proposition computes this number. Definition 5 Define a function d : E0 → N such that for any u ∈ E0, d(u) = {v n(v) ≤ n(u)} . Observe that in a truncated tree, the restriction of the function d on the set of vertices which are not sinks is one to one. Proposition 6 The number of distinct truncated trees with n vertices is 2n−2. Proof. For every truncated tree E with n vertices we assign an n-vector α(E) = (α1, α2, · · · , αn) where αi ∈ {0, 1} as follows: • α(E) contains exactly N − 1 many 1's where N − 1 is the number of non-sinks of E. • To define that vector it is sufficient to know which component is 1. • To each vertex v which is not a sink, we assign a 1 appearing in the d(v)-th component. • Remaining components are all zero. Hence α(E) starts with 1 and ends with 0. Given any {0, 1} sequence β of length n starting with 1 and ending with 0, there exists clearly a unique truncated tree E with n vertices such that α(E) = β. Hence the number of distinct truncated trees with n vertices is equal to the number of all {0, 1}-sequences of length n in which the first and last components are constant which is equal to 2n−2. For a tree F with n vertices the K-dimension of LK(F ) is not uniquely determined by the number of vertices only. However, we can compute the maximum and the minimum K-dimensions of LK(F ) where F ranges over all possible trees with n vertices. Lemma 7 The maximum K-dimension of LK(E) where E ranges over all pos- sible trees with n vertices and s sinks is equal to s(n − s + 1)2. 5 Proof. Assume E is a tree with n vertices. Then LK(E) ∼= Mni(K), by Proposition 1 (3) where s is the number of sinks in E and ni ≤ n − s + 1 for all i = 1, . . . s. Hence sLi=1 dim LK(E) = sXi=1 n2 i ≤ s(n − s + 1)2. Notice that there exists a tree E as sketched below with n vertices and s sinks such that dim LK(E) = s(n − s + 1)2. Theorem 8 The maximum K-dimension of LK(E) where E ranges over all possible trees with n vertices is given by f (n) where f (n) =   n(2n + 3)2 27 if n ≡ 0 (mod 3) 1 27 (n + 2) (2n + 1)2 if n ≡ 1 (mod 3) 4 27 (n + 1)3 if n ≡ 2 (mod 3) Proof. Assume E is a tree with n vertices. Then LK(E) ∼= Mni where s is the number of sinks in E. Now, to find max dim LK(E) we need only to deter- mine maximum value of for s = 1, 2, . . . , n − 1. Extending the domain of f (s) to real numbers 1 ≤ s ≤ n − 1 we get a continuous function, hence we can find its maximum value. function f (s) = s(n − s + 1)2 the sLi=1 f (s) = s(n − s + 1)2 ⇒ d ds(cid:0)s(n − s + 1)2(cid:1) = (n − 3s + 1) (n − s + 1) is the only critical point in the interval [1, n − 1] and since ) < 0, it is a local maximum. In particular f is increasing on n + 1 3 n + 1 3 n + 1 Then s = d2f ds2 ( (cid:20)1, 3 (cid:21) and decreasing on (cid:20) n + 1 3 (cid:19) = K-dimension of LK(E) is f(cid:18) n + 1 3 for each i = 1, 2, . . . , s. 6 Case 1: n ≡ 2 (mod 3). In this case s = is an integer and maximum , n − 1(cid:21) . We have three cases: n + 1 3 4 27 (n + 1)3 and we have ni = 2(n + 1) 3 , Case 2: n ≡ 0 (mod 3). Then we have: n 3 = t < t + 1 3 = s < t + 1 and 3(cid:17) = f(cid:16) n (2n + 3)2n 27 = α1 and f(cid:16) n 3 + 1(cid:17) = 4n2(n + 3) 27 = α2. Note that, α1 > α2. So α1 is maximum K-dimension of LK(E) and we have ni = n + 1, for each i = 1, 2, . . . , s. 2 3 n − 1 3 = t < t + 2 3 = s < t + 1 and Case 3: n ≡ 1 (mod 3). Then and f(cid:18) n − 1 f(cid:18) n + 2 3 (cid:19) = 3 (cid:19) = 4 27 1 27 (n + 2)2 (n − 1) = β1 (2n + 1)2 (n + 2) = β2. In this case β2 > β1 and so β2 gives the maximum K-dimension of LK(E) and we have ni = , for each i = 1, 2, . . . , s. 2n + 1 3 Theorem 9 The minimum K-dimension of LK(E) where E ranges over all possible trees with n vertices and s sinks is equal to r(q + 2)2 + (s − r)(q + 1)2, where n − 1 = qs + r, 0 ≤ r < s. Proof. We call a graph a bunch tree if it is obtained by identifiying the unique sources of the finitely many oriented finite line graphs. Let E(n, s) be the set of all bunch trees with n vertices and s sinks. Every element of E(n, s) can be uniquely represented by an s-tuple (t1, t2, ..., ts) where each ti is the number of vertices contributing only to the i-th sink with 1 ≤ t1 ≤ t2 ≤ ... ≤ ts and t1 + t2 + ... + ts = n − 1. Let E ∈ E(n, s) with ts − t1 ≤ 1. This E is represented by the s-tuple (q, . . . , q, q + 1, . . . , q + 1) where n − 1 = sq + r, 0 ≤ r < s. Now we claim that the dimension of E is the minimum of the set {dim LK(F ) : F tree with s sinks and n vertices} . 7 If we represent U ∈ E(n, s) by the s-tuple (u1, u2, ..., us) then E 6= U implies that us − u1 ≥ 2. Consider the s-tuple (t1, t2, ..., ts) where (t1, t2, ..., ts) is obtained from (u1 + 1, u2, ..., us−1, us − 1) by reordering the components in increasing order. In this case the dimension dU of U is dU = (u1 + 1)2 + . . . + (us + 1)2. Similarly, the dimension dT of the bunch graph T represented by the s-tuple (t1, t2, ..., ts), is dT = (t1 + 1)2 + . . . + (ts + 1)2 = (u1 + 2)2 + . . . + u2 s−1 + us 2. Hence dU − dT = 2(us − u1) − 2 > 0. Repeating this process sufficiently many times we see that the process has to end at the exceptional bunch tree E showing that its dimension is the smallest among the dimensions of all elements of E(n, s). Now let F be an arbitrary tree with n vertices and s sinks. As above we assign to F the s-tuple (n1, n2, ..., ns) with ni = n(vi) − 1 where the sinks vi, i = 1, 2, . . . , s are indexed in such a way that ni ≤ ni+1, i = 1, . . . , s − 1. Observe that n1 + n2 + · · ·+ ns ≥ n− 1. Let β = ni − (n− 1). Since s ≤ n− 1, sXi=1 (ni − 1). Either n1 − 1 ≥ β or there exists a unique k ∈ {2, . . . , s} such β ≤ that sXi=1 k−1Xi=1 (ni − 1) < β ≤ (ni − 1). If n1 − 1 ≥ β, then let kXi=1 mi =(cid:26) n1 − β , ni , i = 1 i > 1 . Otherwise, let mi =   nk −(cid:18)β − 1 k−1Pi=1 ni , (ni − 1)(cid:19) , i ≤ k − 1 i = k . , i ≥ k + 1 In both cases, the s-tuple (m1, m2, . . . , ms) that satisfies 1 ≤ mi ≤ ni, m1 ≤ m2 ≤ · · · ≤ ms and m1 + m2 + · · · + ms = n − 1 is obtained. So, there ex- ists a bunch tree M namely the one corresponding uniquely to (m1, m2, . . . , ms) which has dimension dM ≤ dF . This implies that dF ≥ dE. Hence the result follows. Lemma 10 The minimum K-dimension of LK(E) where E ranges over all possible trees with n vertices occurs when the number of sinks is n − 1 and is equal to 4(n − 1). 8 Proof. By the previous theorem we see that dim LK(E) ≥ r(q + 2)2 + (s − r)(q + 1)2 where n − 1 = qs + r, 0 ≤ r < s. We have r(q + 2)2 + (s − r)(q + 1)2 = (n − 1)(q + 2) + qr + r + s. Thus (n − 1)(q + 2) + qr + r + s − 4(n − 1) = (n − 1)(q − 2) + qr + r + s ≥ 0 if q ≥ 2. If q = 1,then −(n − 1) + 2r + s = −(n − 1) + r + (n − 1) = r ≥ 0. Hence dim LK(E) ≥ 4(n − 1). Notice that there exists a truncated tree E with n vertices and dim LK(E) = 4(n − 1) as sketched below : 3 Line Graphs The total-degree of the vertex v is the number of edges that either have v as its source or as its range, that is, tot deg(v) =(cid:12)(cid:12)s−1(v) ∪ r−1(v)(cid:12)(cid:12) . A finite graph E is a line graph if it is connected, acyclic and totdeg(v) ≤ 2 for every v ∈ E0. Remark 11 In [1], the proposition 5.7 shows that a semisimple finite dimen- sional algebra A = Mni(K) over the field K can be described as a Leavitt path algebra L(E) defined by a line graph E, if and only if A has no ideals of K−dimension 1 and the number of minimal ideals of A of K dimension 22 is at most 2. On the other hand, if A ∼= L(E) for some n line graph E then n − 1 = (ni − 1), that is, n is an algebraic invariant of A. sLi=1 sPi=1 Therefore the following proposition answers a reasonable question. Proposition 12 The number An of isomorphism classes of Leavitt path alge- bras defined by line graphs having exactly n vertices is An = P (n − 1) − P (n − 4) where P (m) is the number of partitions of the natural number m. 9 Proof. Any n-line graph has n − 1 edges. In a line graph, for any edge e there exists a unique sink v so that there exists a path from s(e) to v. In this case we say that e is directed towards v. The number of edges directed towards v is clearly equal to n(v) − 1. Let E and F be two n-line graphs. LK(E) ∼= LK(F ) if and only if there exists a bijection φ : S(E) → S(F ) such that for each v in S(E), we have n(v) = n(φ(v)). Therefore the number of isomorphism classes of Leavitt path algebras determined by n-line graphs is the number of partitions of n − 1 edges in which the number of parts having exactly one edge is at most two. Since the number of partitions of k objects having at least three parts each of which containing exactly one element is P (k − 3), we get the result An = P (n − 1) − P (n − 4). References [1] G. Abrams, G. Aranda Pino, M. Siles Molina, Finite-dimensional Leavitt path algebras, J. Pure Appl. Algebra 209 (2007) 753 - 762. [2] T.Y. Lam, A First Course In Noncommutative Rings, Springer-Verlag 2001. [3] G. Abrams, G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2) (2005), 319 -- 334. 10
1504.00194	2	1504	2015-12-15T10:50:57	Tensor products of nonassociative cyclic algebras	[ "math.RA" ]	We study the tensor product of an associative and a nonassociative cyclic algebra. The condition for the tensor product to be a division algebra equals the classical one for the tensor product of two associative cyclic algebras by Albert or Jacobson, if the base field contains a suitable root of unity. Stronger conditions are obtained in special cases. Applications to space-time block coding are discussed.	math.RA	math	TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS S. PUMPL UN Abstract. We study the tensor product of an associative and a nonassociative cyclic algebra. The condition for the tensor product to be a division algebra equals the classical one for the tensor product of two associative cyclic algebras by Albert or Jacobson, if the base field contains a suitable root of unity. Stronger conditions are obtained in special cases. Applications to space-time block coding are discussed. Introduction Nonassociative cyclic algebras of degree n are canonical generalizations of associative cyclic algebras of degree n and were first introduced over finite fields by Sandler [17]. Nonassociative quaternion algebras (the case n = 2) constituted the first known exam- ples of a nonassociative division algebra (Dickson [5]). Nonassociative cyclic algebras were investigated over arbitrary fields by Steele [19], [20], see also [14]. In the following we study the tensor product A = D0 ⊗F0 D1 of an associative and a nonassociative cyclic algebra over a field F0 and give conditions for A to be a division algebra. We discover that these tensor product algebras are used in space-time block coding [11], [12], [13], and employed to construct some of the iterated codes by Markin and Oggier [9]. After recalling results needed in the paper in Section 1, results by Petit [10] are used to show that the iterated algebras Itm R (D, τ, d) introduced in [11], [12], [13] can be defined using polynomials in skew-polynomial rings over D when D is a cyclic division algebra (Theorem 7) in Section 2. In Section 3 we show that for an associative division algebra D = (L/F0, σ, c) ⊗F0 F , L and F linearly disjoint over F0, (L/F0, σ, c) ⊗F0 (F/F0, τ, d) ∼= Sf ∼= Itm R (D, τ, d), where the twisted polynomial f (t) = tm−d ∈ D[t;eτ−1], eτ an automorphism of D canonically extending τ , is used to construct the algebra Sf (Theorem 11). Section 3 contains the main results: if D0 is an associative cyclic algebra over F0 such that D = D0 ⊗F0 F is a division algebra, and D1 = (F/F0, τ, d) a nonassociative cyclic algebra of degree m, L and F linearly disjoint over F0, then D0⊗F0 D1 is a division algebra if and only if f (t) = tm−d is irreducible in D[t;eτ−1] (Theorem 16). Date: 14.12.2015. 2010 Mathematics Subject Classification. Primary: 17A35; Secondary: 16S36, 94B05. Key words and phrases. cyclic algebra, nonassociative cyclic algebra, nonassociative quaternion algebra, tensor product, division algebra. 1 2 S. PUMPL UN Assume further that m is prime and if m 6= 2, 3, that F0 contains a primitive mth root of unity. Then (L/F0, σ, c) ⊗F0 (F/F0, τ, d) is a division algebra if and only if d 6= zeτ (z)···eτ m−1(z) for all z ∈ D. This generalizes the classical condition for the tensor product of two associative cyclic algebras [6, Theorem 1.9.8], (see also Theorem 13), to the nonassociative setting. Some more detailed conditions are obtained for special cases. For instance, if char F0 6= 2, D0 is an associative quaternion algebra over F0 which remains a division algebra over F , and D1 a nonassociative quaternion algebra, such that D0 and D1 do not share a common subfield, then D0 ⊗F0 D1 is always a division algebra over F0 (Theorem 17). In Section 4, we discuss how our results can be applied to systematically construct fully diverse fast-decodable space- time block codes for nm transmit and m receive antennas. We thus generalize the set-up discussed in [9, Section V.A] which is limited to tensor products of a cyclic algebra of degree three and a nonassociative quaternion algebra, i.e. to n = 3, m = 2, and see that Theorems 16, 17 and 18 provide conditions for an iterated code consisting of larger matrices (where m > 2) to be fully diverse. We then design a new family of fast-decodable fully diverse 4× 2- codes with non-vanishing determinant starting with the Silver code, which have decoding complexity O(M 4.5), i.e. the same decoding complexity as all state-of the art fast-decodable fully diverse 4 × 2 codes. Thus our construction allows us to systematically design codes whose decoding complexity is competitive with the ones designed 'ad hoc' like the SR-code [16, IV.B] or the code in [8] and which also have non-vanishing determinant. 1. Preliminaries 1.1. Nonassociative algebras. Let F be a field and let A be an F -vector space. We call A 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. 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 is finite-dimensional, A is a division algebra if and only if A has no zero divisors [18, pp. 15, 16]. For an F -algebra A, 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 mid- dle nucleus is Nucm(A) = {x ∈ A [A, x, A] = 0}, the right nucleus is Nucr(A) = {x ∈ A [A, A, x] = 0} and the nucleus is Nuc(A) = {x ∈ A [x, A, A] = [A, x, A] = [A, A, x] = 0}. It 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 commuter of A is defined as Comm(A) = {x ∈ A xy = yx for all y ∈ A} and the center of A is C(A) = {x ∈ A x ∈ Nuc(A) and xy = yx for all y ∈ A}. For two nonassociative algebras C and D over F , Nuc(C) ⊗F Nuc(D) ⊂ Nuc(C ⊗F D). TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS 3 Thus we can consider the tensor product A = C ⊗F D as a right R-module over any ring R ⊂ Nuc(C) ⊗F Nuc(D). 1.2. Associative and nonassociative cyclic algebras. Let K/F be a cyclic Galois ex- tension of degree n with Galois group Gal(K/F ) = hσi. An associative cyclic algebra (K/F, σ, c) of degree n over F , c ∈ F ×, is an n-dimensional K-vector space with multiplication given by the relations (K/F, σ, c) = K ⊕ eK ⊕ e2K ⊕ ··· ⊕ en−1K, en = c, le = eσ(l), for all l ∈ K. (K/F, σ, c) is division for all c ∈ F ×, such that cs 6∈ NK/F (K×) for all s which are prime divisors of n, 1 ≤ s ≤ n − 1. For c ∈ K\F , we define a unital nonassociative algebra (K/F, σ, c) (Sandler [17]) as the n-dimensional K-vector space (K/F, σ, c) = K ⊕ eK ⊕ e2K ⊕ ··· ⊕ en−1K, where multiplication is given by the following rules for all a, b ∈ K, 0 ≤ i, j, < n, which then are extended linearly to all elements of A: (eia)(ejb) =   ei+jσj(a)b e(i+j)−ndσj(a)b if i + j < n, if i + j ≥ n. We call D = (K/F, σ, c) with c ∈ K \ F a nonassociative cyclic algebra of degree n. D has nucleus K and center F . D is not (n + 1)th power associative since (en−1e)e = eσ(a) and e(en−1e) = ea. If [K : F ] is prime, D always is a division algebra. If [K : F ] is not prime, D is a division algebra for any choice of c such that 1, c, . . . , cn−1 are linearly independent over F [20]. For n = 2, (K/F, σ, c) = Cay(K, c) is an associative (if c ∈ F ) or nonassociative (if c ∈ K \ F ) quaternion algebra over F , cf. [2], [15] or [21]. From now on, when we say D = (K/F, σ, c) is a cyclic algebra, we mean an associative or nonassociative cyclic algebra over F without always explicitly stating that we also allow c ∈ K×. We call {1, e, e2, . . . , en−1} the standard basis of (K/F, σ, c). D = (K/F, σ, c) is a K-vector space of dimension n (since K = Nuc(D) if the algebra is nonassociative) and, after a choice of a K-basis, we can embed the K-vector space EndK(D) into Matn(K). The left multiplication of elements of D with y = y0+ey1+···+en−1yn−1 ∈ D (yi ∈ K) induces the K-linear embedding λ : D → Matn(K). 1.3. Iterated algebras. From now on, we will use the following notation: Let F and L be fields and let K be a cyclic field extension of both F and L such that (1) Gal(K/F ) = hσi and [K : F ] = n, (2) Gal(K/L) = hτi and [K : L] = m, (3) σ and τ commute: στ = τ σ. 4 S. PUMPL UN Define F0 = F ∩ L. Let D = (K/F, σ, c) be a cyclic algebra of degree n over F with reduced norm ND/F . For x = x0 + ex1 + e2x2 + ··· + en−1xn−1 ∈ D (xi ∈ K, 1 ≤ i ≤ n), define the L-linear map eτ : D → D via eτ (x) = τ (x0) + eτ (x1) + e2τ (x2) + ··· + en−1τ (xn−1). If c ∈ F0 then for all x, y ∈ D, where for any matrix X = λ(x) representing left multiplication with x, τ (X) means applying τ to each entry of the matrix. eτ (xy) = eτ (x)eτ (y) and λ(eτ (x)) = τ (λ(x)) Definition 1. (cf. D-module via the rules [12], [13]) Pick d ∈ F ×, c ∈ F0. Define a multiplication on the right D ⊕ f D ⊕ f 2D ⊕ ··· ⊕ f m−1D, (f ix)(f j y) =   if i + j < m f i+jeτ j(x)y f (i+j)−meτ j (x)yd if i + j ≥ m R (D, τ, d). for all x, y ∈ D, and call the resulting iterated algebra Itm Itm R (D, τ, d) is a nonassociative algebra over F0 of dimension m2n2 with unit element (1D, 0, . . . , 0) and contains D as a subalgebra. We call {1, e, e2, . . . , en−1, f, f e, f e2, . . . , f m−1en−1} the standard basis of the K-vector space Itm R (D, τ, d). Example 1. Let A = It3 (d, 0, 0) = f f 2 and the multiplication in A is given by R(D, τ, d) and put f = (0, 1D, 0). Then f 2 = (0, 0, 1D), f 2f = (u, v, w)(u′, v′, w′) = (   v u deτ (w) eτ (u) w eτ (v) deτ 2(v) deτ 2(w) eτ 2(u)     u′ v′ w′  )T = (uu′ + deτ (w)v′ + deτ 2(v)w′, vu′ +eτ (u)v′ + deτ 2(w)w′, wu′ +eτ (v)v′ + eτ 2(u)w′) for u, v, w, u′, v′, w′ ∈ D. From now on, let A = Itm R (D, τ, d). Lemma 2. (i) The cyclic algebra (K/L, τ, d) over L, viewed as an algebra over F0, is a subalgebra of A, and is nonassociative if d ∈ F \ F0. (ii) Let m be even. Then It2 R(D, τ, d) is isomorphic to a subalgebra of A. In particular, the quaternion algebra (K/L, τ, d) = Cay(K, d) over L, viewed as algebra R(D, τ, d), which is nonassociative and division if d ∈ F \ F0. over F0, is a subalgebra of It2 Proof. (i) This can be seen by restricting the multiplication of A to K ⊕ ··· ⊕ K. (ii) Suppose that m = 2s for some integer s. Then It2 which is a subalgebra of A under the multiplication inherited from A. R(D, τ, d) is isomorphic to D ⊕ f sD, (cid:3) TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS 5 We can embed EndK(A) into the module Matnm(K). Left multiplication Lx with x ∈ A is a right K-endomorphism, so that we obtain a well-defined additive map L : A → EndK(A) ֒→ Matnm(K), x 7→ Lx 7→ L(x) = X which is injective if A is division. Take the standard basis {1, e, . . . , en−1, f, f e, . . . , f m−1en−1} of the K-vector space A. Then xy = (λ(M (x))yT )T , where (1) λ(M (x)) =   λ(x0) λ(x1) ... dτ (λ(xm−1)) τ (λ(x0)) ... λ(xm−1) τ (λ(xm−2)) ··· ··· . . . ··· dτ m−1(λ(x1)) dτ m−1(λ(x2)) ... τ m−1(λ(x0))   is obtained by taking the matrix λ(xi), xi ∈ D, representing left multiplication in D of each entry in the matrix M (x). λ(M (x)) represents the left multiplication by the element x in A. Define MA : A → K, MA(x) = det(λ(M (x))). If x ∈ A is nonzero and not a left zero divisor in A, then MA(x) 6= 0 by [13, Theorem 9]. Remark 3. It is clear that A is a division algebra if and only if MA(x) 6= 0 for all x 6= 0: If A is a division algebra then Lx is bijective for all x 6= 0 and thus λ(M (x)) invertible, i.e. MA(x) 6= 0. Conversely, if MA(x) 6= 0 for all x 6= 0 then for all x, y ∈ A, x 6= 0, y 6= 0, also xy = (λ(M (x))yT )T 6= 0. 2. Division algebras obtained from skew-polynomial rings In the following, we recall results from [6] and [10]. Let D be a unital division ring and σ a ring isomorphism of D. The twisted polynomial ring D[t; σ] is the set of polynomials a0 + a1t + ··· + antn with ai ∈ D, where addition is defined term-wise and multiplication by That means, ta = σ(a)t (a ∈ D). atnbtm = aσn(b)tn+m and tna = σn(a)tn for all a, b ∈ D [6, p. 2]. R = D[t; σ] is a left principal ideal domain and there is a right division algorithm in R [6, p. 3], i.e. for all g, f ∈ R, g 6= 0, there exist unique r, q ∈ R such that deg(r) < deg(f ) and g = qf + r. R = D[t; σ] is also a right principal ideal domain [6, p. 6] with a left division algorithm in R [6, p. 3 and Prop. 1.1.14]. (We point out that our terminology is the one used by Petit [10] and Lavrauw and Sheekey [7]; it is different from Jacobson's [6], who calls what we call right a left division algorithm and vice versa.) 6 S. PUMPL UN Thus R = D[t; σ] is a (left and right) principal ideal domain (PID). An element f ∈ R is irreducible in R if it is no unit and it has no proper factors, i.e there do not exist g, h ∈ R with deg(g), deg(h) < deg(f ) such that f = gh [6, p. 11]. [10, (7)]) Let f ∈ D[t; σ] be of degree m and let modrf denote the Definition 2. (cf. remainder of right division by f . Then the vector space Rm = {g ∈ D[t; σ] deg(g) < m} together with the multiplication g ◦ h = gh modrf becomes a unital nonassociative algebra Sf = (Rm,◦) over F0 = {z ∈ D zh = hz for all h ∈ Sf}. The multiplication is well-defined because of the right division algorithm and F0 is a subfield of D. Since σ is a ring isomorphism, we can use the left division algorithm to define an algebra as well: Let f ∈ D[t; σ] be of degree m and let modlf denote the remainder of left division by f . Then Rm together with the multiplication g ◦ h = gh modlf becomes a nonassociative algebra f S = (Rm,◦), which, however, turns out to be anti- isomorphic to a suitable algebra Sg for some g ∈ R′ in some twisted polynomial ring R′. Remark 4. (i) When deg(g)deg(h) < m, the multiplication of g and h in Sf is the same as the multiplication of g and h in R [10, (10)]. (ii) Given a cyclic Galois field extension K/F of degree m with Gal(K/F ) = hσi, the cyclic algebra (K/F, σ, d) is the algebra Sf with f (t) = tm − d ∈ R = K[t; σ−1] [10, p. 13-13]. (iii) Let D be a finite-dimensional central division algebra over F and σ an automorphism of D of order m. In [6], the associative algebras E(f ) = {g ∈ D[t; σ] deg(g) < m, f right divides f g} for f = tm − d ∈ D[t; σ], were investigated. E(f ) is division iff f is irreducible. Theorem 5. (cf. [10, (2), p. 13-03, (7) (9), (15), (17), (18), (19)]) Let f = tm−Pm−1 i=0 diti ∈ R = D[t; σ]. (i) If Sf is not associative then and Nucl(Sf ) = Nucm(Sf ) = D Nucr(Sf ) = {g ∈ Sf f g ∈ Rf} = E(f ). (ii) If f is irreducible then Nucr(Sf ) is an associative division algebra. (iii) Let f ∈ R be irreducible and Sf a finite-dimensional F0-vector space or a finite- dimensional right Nucr(Sf )-module. Then Sf is a division algebra. (iv) f (t) = t2 − d1t − d0 is irreducible in D[t; σ] if and only if σ(z)z − d1z − d0 6= 0 for all TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS 7 z ∈ D. (v) f (t) = t3 − d2t2 − d1t − d0 is irreducible in D[t; σ] if and only if σ2(z)σ(z)z − σ2(z)σ(z)d2 − σ2(z)σ(d1) − σ2(d0) 6= 0 and σ2(z)σ(z)z − d2σ(z)z − d1z − d0 6= 0 for all z ∈ D. (vi) Suppose m is prime and C(D) ∩ Fix(σ) contains a primitive mth root of unity. Then f (t) = tm − d is irreducible in D[t; σ] if and only if d 6= σm−1(z)··· σ(z)z and σm−1(d) 6= σm−1(z)··· σ(z)z for all z ∈ D. Theorem 5, parts (v) and (vi), can be improved as follows [4]: Theorem 6. (i) f (t) = t3 − d is irreducible in D[t; σ] if and only if d 6= σ2(z)σ(z)z for all z ∈ D. (ii) f (t) = t4 − d is irreducible in D[t; σ] if and only if d 6= σ3(z)σ2(z)σ(z)z and σ2(z1)σ(z1)z1 + σ2(z0)z1 + σ2(z1)σ(z0) 6= 0 or σ2(z0)z0 + σ2(z1)σ(z0)z0 6= d for all z0, z1 ∈ D. (iii) Suppose m is prime and C(D) ∩ Fix(σ) contains a primitive mth root of unity. Then f (t) = tm − d ∈ D[t; σ] is irreducible if and only if d 6= σm−1(z)··· σ(z)z for all z ∈ D. Proof. (i) If f (t) = t3 − d is reducible then either f is divisible by a linear factor from the left or from the right. A straightforward calculation shows that f is divisible on the left by t − z, z ∈ D, iff σ2(d) = σ2(z)σ(z)z iff zσ−1(z)σ−2(z) = d. By [6, 1.3.11], f is divisible on the right by t − z, z ∈ D, iff 0 = σ2(z)σ(z)z − d, which is the remainder of right division of f by t − z. If f (t) = t3 − d is irreducible then t − z does not divide f (t) from the right for any z ∈ D and thus d 6= σ2(z)σ(z)z for all z ∈ D. If f (t) = g(t)(t − b) then d = σ2(b)σ(b)b. Conversely, assuming f (t) is reducible then there is a linear factor dividing f from the left or from the right. If f (t) = (t − c)g(t) this is equivalent to d = cσ−1(c)σ−2(c). This implies that (t2 + ct + cσ−1)(t − σ−2(c)) = t3 − cσ−1(c)σ−2(c) = d and that t − σ−2(c) divides f (t) from the right. Put b = σ−2(c) to see there is b ∈ D such that σ2(b)σ(b)b = d. (ii) If f (t) = t4 − d is reducible then either f is divisible by a linear factor from the left, from the right, or f = g1(t)g2(t) for two irreducible polynomials g1, g2 ∈ R of degree 2. By [6, 1.3.11], f is divisible on the right by a factor t − z, z ∈ D, iff d = σ3(z)σ2(z)σ(z)z. A 8 S. PUMPL UN straightforward calculation shows that f is divisible on the left by a factor t − z, z ∈ D, iff 0 = σ3(z)σ2(z)σ(z)z − σ3(d), which is the remainder of left division of f by t− z. Moreover, f is divisible on the right by g(t) = t2 − z1t − z0 ∈ R iff [σ2(z1)σ(z1)z1 + σ2(z0)z1 + σ2(z1)σ(z0)]t + [σ2(z0)z0 + σ2(z1)σ(z0)z0 − d] = 0 which is the remainder of right division of f (t) by g(t). This is equivalent to σ2(z1)σ(z1)z1 + σ2(z0)z1 + σ2(z1)σ(z0) = 0 and σ2(z0)z0 + σ2(z1)σ(z0)z0 = d. Thus f is irreducible if and only if d 6= zσ(z)σ2(z)σ(z) and σ3(d) 6= zσ(z)σ2(z)σ3(z) and σ2(z1)σ(z1)z1 + σ2(z0)z1 + σ2(z1)σ(z0) 6= 0 or σ2(z0)z0 + σ2(z1)σ(z0)z0 6= d for all z0, z1 ∈ D. Now observe that the case that f (t) is divisible on the left by some t − z is equivalent to d = zσ−1(z)σ−2(z)σ−3(z) and putting w = σ−3(z), we obtain d = zσ−1(z)σ−2(z)σ−3(z) = σ3(w)σ2(w)σ(w)w, which is equivalent to f being divisible on the right by t − σ−3(z), and we have proved the assertion. (iii) By [3, Ex no 4, p. 344], f is either irreducible or a product of m linear factors. Thus f is irreducible if and only if for all z ∈ D, t − z does not divide f on the right, which is equivalent to the assertion. (cid:3) The iterated algebras Itm R (D, τ, d) were originally introduced for space-time coding ([9], [16], [13]), and can be obtained from skew-polynomial rings: Theorem 7. Let F and L be fields, F0 = F ∩ L, and let K be a cyclic field extension of both F and L such that (1) Gal(K/F ) = hσi and [K : F ] = n, (2) Gal(K/L) = hτi and [K : L] = m, (3) σ and τ commute: στ = τ σ. Let D = (K/F, σ, c) be an associative cyclic division algebra over F of degree n, c ∈ F0 and d ∈ D×. Then Itm R (D, τ, d) = Sf where R = D[t;eτ−1] and f (t) = tm − d. Proof. Let f = (0, 1D, 0, . . . , 0) ∈ A = Itm f 2D ⊕ ··· ⊕ f m−1D is given by R (D, τ, d). The multiplication on A = D ⊕ f D ⊕ (f ix)(f j y) =   if i + j < m f i+jeτ j(x)y f (i+j)−meτ j (x)yd if i + j ≥ m for all x, y ∈ D which corresponds to the multiplication of the algebra Sf . (cid:3) Theorems 5, 6 and 7 imply: TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS 9 Corollary 8. Assume the setup of Theorem 7. (i) If d 6∈ F0 then Nucl(Itm R (D, τ, d)) = Nucm(Itm R (D, τ, d)) = D and Nucr(Itm R (D, τ, d)) = {g ∈ Sf f g ∈ Rf}. (ii) Itm (iii) It4 R (D, τ, d) is a division algebra if and only if f (t) is irreducible in D[t;eτ−1]. R(D, τ, d) is a division algebra if and only if and d 6= zeτ (z)eτ 2(z)eτ 3(z) eτ 2(z1)eτ 3(z1)z1 +eτ 2(z0)z1 +eτ 2(z1)eτ 3(z0) 6= 0 or eτ 2(z0)z0 +eτ 2(z1)eτ 3(z0)z0 6= d for all z, z0, z1 ∈ D. (iv) Suppose that m is prime and in case m 6= 2, 3, additionally that F0 contains a primitive mth root of unity. Then Itm R (D, τ, d) is a division algebra if and only if d 6= zeτ (z)···eτ m−1(z) for all z ∈ D. Lemma 9. Assume the setup of Theorem 7 and d ∈ F ×. zeτ (z)···eτ m−1(z) for all z ∈ D. The proof generalizes the idea of the proof of [9, Proposition 13]: If τ (dn) 6= dn, then d 6= Proof. If d = zeτ (z)···eτ m−1(z) for some z ∈ D, then for Z = λ(z) this means Zτ (Z)··· τ m−1(Z) = dIn and therefore det(Z) det(τ (Z))··· det(τ m−1(Z)) = dn. Since the left-hand-side is fixed by τ i, this implies that τ i(dn) = dn for 1 ≤ i < m, in particular, τ (dn) = dn. Corollary 10. Assume the setup of Theorem 7 and d ∈ F ×. Suppose that m is prime and in case m 6= 2, 3, additionally that F0 contains a primitive mth root of unity. (i) If τ (dn) 6= dn then Itm (ii) If d ∈ F such that dn 6∈ ND/F0(D×), then Itm for all d ∈ F \ F0 with dn 6∈ F0, Itm Proof. (i) is clear. R (D, τ, d) is a division algebra. In particular, R (D, τ, d) is a division algebra. R (D, τ, d) is a division algebra. (cid:3) (ii) Since c ∈ F0 ⊂ Fix(τ ) = L we have ND/F (eτ (x)) = τ (ND/F (x)) for all x ∈ D by [11, Proposition 4]. Assume that d = zeτ (z)···eτ m−1(z), then ND/F (d) = ND/F (z)ND/F (eτ (z))··· ND/F (eτ m−1(z)) = ND/F (z)τ (ND/F (z))··· τ m−1(ND/F (z)). Put a = ND/F (z), note that aτ (a)··· τ m−1(a) = NF/F0 (a) = NF/F0 (ND/F (z)) = ND/F0(z) ∈ F0, and use that ND/F (d) = dn for d ∈ F . (cid:3) 10 S. PUMPL UN 3. The tensor product of an associative and a nonassociative cyclic algebra 3.1. Let L/F0 be a cyclic Galois field extension of degree n with Gal(L/F0) = hσi, and F/F0 be a cyclic Galois field extension of degree m with Gal(F/F0) = hτi. Let L and F be linearly disjoint over F0 and let K = L ⊗F0 F = L · F be the composite of L and F over F0, with Galois group Gal(K/F0) = hσi × hτi, where σ and τ are canonically extended to K. In the following, let D0 = (L/F0, σ, c) and D1 = (F/F0, τ, d) be two cyclic algebras over F0 such that D0 is associative and D1 is nonassociative, i.e. c ∈ F ×0 and d ∈ F ×. Let A = (L/F0, σ, c) ⊗F0 (F/F0, τ, d). Then K is a subfield of A of degree mn over F0 and K = L ⊗F0 F ⊂ Nuc(A). Let {1, e, e2, . . . , en−1} be the standard basis of the L-vector space D0 and {1, f, f 2, . . . , f m−1} be the standard basis of the F -vector space D1. A is a K-vector space with basis {1 ⊗ 1, e ⊗ 1, . . . , en−1 ⊗ 1, 1 ⊗ f, e ⊗ f, . . . , en−1 ⊗ f m−1}. Identify A = K ⊕ eK ⊕ ··· ⊕ en−1K ⊕ f K ⊕ ef K ⊕ ··· ⊕ en−1f m−1K. Note that D0 ⊗F0 F = (K/F, σ, c). An element in λ(A) has the form dτ m−1(Y1) dτ m−1(Y2) dτ (Yn−1) dτ 2(Yn−2) dτ 2(Yn−1) τ (Y0) . . . . . . (2)   Y0 Y1 ... Yn−2 Yn−1 ... ... τ (Yn−3) τ (Yn−2) τ 2(Yn−4) τ 2(Yn−3) . . . . . . dτ m−1(Yn−1) τ m−1(Y0)   with Yi ∈ λ(D0⊗F0 F ). That means, Yi ∈ Matn(K), and when the entries in Yi are restricted to elements in L, Yi ∈ λ(D0) (multiplication with d in the upper right triangle of the matrix means simply scalar multiplication with d). Theorem 11. (i) (L/F0, σ, c) ⊗F0 (F/F0, τ, d) ∼= Itm R (D0 ⊗F0 F, τ, d). (ii) Suppose that D = (L/F0, σ, c) ⊗F0 F is a division algebra. Then Sf ∼= (L/F0, σ, c) ⊗F0 (F/F0, τ, d) where R = D[t;eτ−1] and f (t) = tm − d. Proof. (i) The matrices in (2) also represent left multiplication with an element in the algebra Itm R ((K/F, σ, c), τ, d), see (1). Thus the multiplications of both algebras are the same. (ii) If D0 ⊗F0 F is a division algebra then Sf ∼= Itm F )[t;eτ−1] and f (t) = tm − d by Theorem 7. R ((K/F, σ, c), τ, d) with R = (D0 ⊗F0 (cid:3) Corollary 12. (i) The cyclic algebras viewed as algebras over F0, are subalgebras of (K/L, τ, d) and (K/F, σ, c) (L/F0, σ, c) ⊗F0 (F/F0, τ, d) TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS 11 of dimension m2n, resp. n2m. (ii) The subalgebra (K/L, τ, d) is nonassociative and thus division if m is prime or, if m is not prime, if 1, d, . . . , dm−1 are linearly independent over L. (iii) If m = st and Fs = Fix(τ s) then is isomorphic to a subalgebra of (L/F0, σ, c) ⊗F0 (F/Fs, τ s, d) (L/F0, σ, c) ⊗F0 (F/F0, τ, d) = Itm R (D0 ⊗F0 F, τ, d). Proof. (i) This is Lemma 2 and [13], Lemma 5 (which also holds if D0⊗F0 F is not division). (ii) This follows from (i), since (F/F0, τ, d) is nonassociative if and only if d ∈ F \ F0. This means d ∈ K \ L. The same argument holds for nonassociative (L/F0, σ, c). (iii) This follows from [20], Theorem 3.3.2, see also [19]. (cid:3) 3.2. Conditions on the tensor product to be a division algebra. To see when the tensor product of two associative algebras is a division algebra we have the classical result by Jacobson [6, Theorem 1.9.8], see also Albert [1, Theorem 12, Ch. XI]: Theorem 13. Let (F/F0, τ, d) be a cyclic associative division algebra of prime degree m. Suppose that D0 is a central associative algebra over F0 such that D = D0⊗F0 F is a division algebra. Then D0 ⊗F0 (F/F0, τ, d) is a division algebra if and only if d 6= eτ m−1(z)···eτ (z)z for all z ∈ D. Note that here d 6= eτ m−1(z)···eτ (z)z is equivalent to d 6= zeτ (z)···eτ m−1(z) since d ∈ F0. This classical result has the following immediate generalization to the nonas- sociative setting: Theorem 14. Let (F/F0, τ, d) be a nonassociative cyclic algebra of degree m. Let D0 = (L/F0, σ, c) be an associative cyclic algebra over F0 of degree n, such that D = D0 ⊗F0 F = (K/F, σ, c) is a division algebra. Let L and F be linearly disjoint over F0. Assume m is prime and in case m 6= 2, 3, additionally that F0 contains a primitive mth root of unity. Then is a division algebra if and only if (L/F0, σ, c) ⊗F0 (F/F0, τ, d) d 6= zeτ (z)···eτ m−1(z) for all z ∈ D. Proof. This is Theorem 11 together with Theorem 6 (resp., [11], Theorem 3.2 for n = 2). (cid:3) More generally, we conclude: 12 S. PUMPL UN Theorem 15. Let (F/F0, τ, d) be an associative or nonassociative cyclic algebra of degree m. Let D0 = (L/F0, σ, c) be an associative cyclic algebra over F0 of degree n, such that D = D0 ⊗F0 F = (K/F, σ, c) is a division algebra. Let L and F be linearly disjoint over F0. Then is a division algebra if and only if (L/F0, σ, c) ⊗F0 (F/F0, τ, d) f (t) = tm − d ∈ D[t;eτ−1] is irreducible. This and Corollary 10 yields: Theorem 16. Let (F/F0, τ, d) be a nonassociative cyclic algebra of degree m. Let D0 = (L/F0, σ, c) be an associative cyclic algebra over F0 of degree n, such that D = D0 ⊗F0 F = (K/F, σ, c) is a division algebra. Let L and F be linearly disjoint over F0. (a) Let m = 4. Then (L/F0, σ, c) ⊗F0 (F/F0, τ, d) is a division algebra if and only if and d 6= zeτ (z)eτ 2(z)eτ 3(z) eτ 2(z1)eτ 3(z1)z1 +eτ 2(z0)z1 +eτ 2(z1)eτ 3(z0) 6= 0 or eτ 2(z0)z0 +eτ 2(z1)eτ 3(z0)z0 6= d for all z0, z1 ∈ D. (b) Suppose that m is prime and in case m 6= 2, 3, additionally that F0 contains a primitive mth root of unity. (i) If τ (dn) 6= dn then (L/F0, σ, c) ⊗F0 (F/F0, τ, d) is a division algebra. (ii) If dn 6∈ ND/F0(D×), then is a division algebra. (iii) For all dn 6∈ F0, is a division algebra. (L/F0, σ, c) ⊗F0 (F/F0, τ, d) (L/F0, σ, c) ⊗F0 (F/F0, τ, d) In special cases, Theorem 14 yields straightforward conditions for the tensor product to be a division algebra: Theorem 17. Let F0 be of characteristic not 2. Let (a, c)F0 be a quaternion algebra over F0 which is a division algebra over F = F0(√b), and (F0(√b)/F0, τ, d) a nonassociative quaternion algebra. Then the tensor product is a division algebra over F0. (a, c)F0 ⊗F0 (F0(√b)/F0, τ, d) σ(√a) = −√a, τ (√a) = √a, σ(√b) = σ(√b), τ (√b) = − √b, L = F0(√a) and D = (a, c)F0 ⊗ F . For z = z0 + iz1 + jz2 + kz3 ∈ D, zi ∈ F0(√b), i2 = a, j2 = c, we get zeτ (z) = (z0τ (z0) + az1τ (z1) + cz2τ (z2) − acz3τ (z3)) +i(z0τ (z1) + z1τ (z0) − cz2τ (z3) + cz3τ (z2)) +j(z0τ (z2) + z2τ (z3) + az1τ (z3) − az3τ (z1)) +k(z0τ (z3) + z3τ (z2) + z1τ (z2) − z2τ (z1)). Since (F0(√b)/F0, τ, d) is nonassociative, d ∈ F0(√b)\F0. Hence if we assume that d = zeτ(z) for some z ∈ D then d = z0τ (z0) + aσ(z1)τ (z1) + cσ(z2)τ (z2) − acσ(z3)τ (z3) = NF/F0 (z0) + aNF/F0 (z1) + cNF/F0 (z2) − acNF/F0 (z3) ∈ F0, a contradiction. Thus, by Theorem 14, the tensor product (a, c)F0 ⊗F0 (F0(√b)/F0, τ, d) is a division algebra. Theorem 18. Let F0 be of characteristic not 2, F = F0(√b). Let D0 = (L/F0, σ, c) be a cyclic algebra over F0 of degree 3 and (F0(√b)/F0, τ, d) a nonassociative quaternion algebra. Let d = d0 + √bd1 ∈ F \ F0 with d0, d1 ∈ F0. (cid:3) (i) If 3d2 0 + bd2 1 6= 0, then TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS 13 Proof. Here, K = F0(√a,√b) with Galois group G = Gal(K/F0) = {id, σ, τ, στ}, where (L/F0, σ, c) ⊗F0 (F0(√b)/F0, τ, d) is a division algebra over F0. (ii) Let F0 = Q. If b > 0, or if b < 0 and − b 3 6∈ Q×2 then (L/F0, σ, c) ⊗F0 (F0(√b)/F0, τ, d) is a division algebra over F0. Proof. Here, F = F0(√b) and K = F0(√b). 1 + √bd1(3d2 (i) We know d3 = d3 equivalent to 3d2 0 + bd2 (ii) is a direct consequence from (i): for F0 = Q, 3d2 1 = 0 if and only if ( d0 assertion is true since 3d2 d1 0 + bd2 0 + bd2 0 + 3bd0d2 1 6= 0. The assertion follows from Theorem 16 (b). 1), so if we want that d3 6= eτ (d3), this is 1 > 0 for all b > 0. For b < 0, the 0 + bd2 )2 = − b 3 . (cid:3) 14 S. PUMPL UN Remark 19. For A = (L/F0, σ, c)⊗F0 (F/F0, τ, d), the map MA(x) = det(Lx) = det(λ(M (x))) can be seen as a generalization of the norm of an associative central simple algebra, since MA = NA/F if both cyclic algebras in the tensor product A are associative. For all X = λ(M (x)) = λ(x) ∈ λ(A) ⊂ Matnm(K), D = D0 ⊗F0 F , we have det X ∈ F (cf. [12] or [11, Corollary 2] for m = 2). Thus MA : A → F . We also have MA(x) = ND/F (x)τ (ND/F (x))··· τ (ND/F (x)) = NF/F0(ND/F (x)) for all x ∈ (K/F, σ, c) (which is easy to see from applying the determinant to the matrix of Lx in Equation (4) for some x ∈ D). We conclude with the observation that the generalization of Albert and Jacobson's con- dition is a necessary condition for d in the general case: Proposition 20. Let D0 = (L/F0, σ, c) be a an associative cyclic algebra of degree n over F0, such that D = D0⊗F0 F is a division algebra. If D0⊗F0 (F/F0, τ, d) is a division algebra then for all z ∈ D. d 6= zeτ (z)···eτ m−1(z) 4. Applications to space-time block coding Space-time block codes are used for reliable high rate transmission over wireless digital channels with multiple antennas transmitting and receiving the data. For instance, four transmit and two receive antennas can be used in digital video broadcasting for portable TV devices, or for transmitting data to mobile phones. A space-time block code (STBC) is a set C of complex n× m matrices (the codebook ), that satisfies a number of properties which determine how well the code performs. Each column is transmitted from n transmit antennas simultaneously, after n columns are transmitted, the receive antennas process the data and try to recover them. We consider linear codes: if X, X′ ∈ C then X ± X′ ∈ C. There are some basic design criteria which are desirable for good performance of a linear code: The code should be fully diverse, i.e. detX 6= 0 for all 0 6= X ∈ C; and for all X ∈ C, detX2 should be as large as possible to minimize the pairwise error probability. The linear code has non-vanishing determinant (NVD) if min{X∈ C}detX2 is bounded away from 0. Central simple (associative) division algebras, in particular cyclic division algebras A = (K/F, σ, c) of degree m with K an imaginary number field, have been highly successfully used to systematically build STBCs, using the fact that their left regular representation λ : A ֒→ EndK(A) ⊂ Matm(K), x 7→ Lx 7→ X over a maximal subfield K yields a fully diverse linear STBC C = λ(A). The idea to take the matrix representing left multiplication (over a suitable subfield of the algebra) to obtain a fully diverse STBC can be extended to nonassociative division algebras, provided they have a large enough nucleus, as first observed in [15]. TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS 15 The previous results establish a general framework and some new simplified conditions for constructing fully diverse STBCs using the left multiplication in tensor products of any associative cyclic algebra over F0 of degree n and a nonassociative cyclic algebra of degree m. We thus have the potential to systematically construct STBCs for given numbers of transmit/receive antennas which are not just fully diverse but fast-decodable, provided we start with a suitable cyclic algebra (whose left regular representation yields a fast-decodable code), as discussed in [13, Section 6.1]. We use the nm2 degrees of freedom of the channel to transmit nm2 complex informa- tion symbols per codeword X ∈ C. If mn channels are used, our space-time block code C consisting of mn × mn matrices X of the form (2) with entries in K therefore has a rate of m complex symbols per channel use, which is maximal for m receive antennas. If the associated tensor product algebra is division (which can be checked using Theorem 11 and the resulting easier conditions for special cases), the code is fully diverse. This generalizes the set-up discussed in [9, Section V.A] which only covers n = 3, m = 2. Theorem 11 implies also that the fast-decodable iterated codes for 6 transmit and 3 receive antennas constructed seemingly ad hoc in [9, Section V.A] consist of the 6×6-matrices representing left multiplication in the tensor product of a degree three cyclic division algebra D and a nonassociative quaternion algebra. The division algebra A = (−1,−1)Q ⊗Q (Q(θ)/Q, τ, θ) 2 is a primitive third root of unity, ζ7 is a primitive 7th root of unity, θ = ζ7 + ζ−1 is behind the fast-decodable fully diverse code designed in [13, Section 6.2], where ω = −1+√3i 7 = 2 cos( 2π 7 ), and Q(ω, θ)/Q(ω) is a cubic cyclic field extension whose Galois group is generated by the automorphism τ : ζ7 + ζ−1 7 . For m = 3 and n = 2 we can now systematically build other fast-decodable fully diverse iterated codes of maximal rate for 6 transmit and 3 receive antennas out of tensor products if desired. 7 + ζ−2 7 7→ ζ2 Using Theorem 17, we can construct a new family of fast-decodable 4 × 2-codes starting with the Silver code. These codes have the same decoding complexity as the SR-code [16, IV.B] and as the code presented in [8], which is O(M 4.5). Three other state-of the-art 'ad hoc' constructions of 4×2-codes of the same decoding complexity are given in [8, Table I]. To our knowledge there are no fast-decodable fully diverse 4 × 2 codes known for 4x2 multiple- input multiple-output transmission with a better decoding complexity than O(M 4.5): Example 21. Let F = Q(√−7), K = Q(i,√−7) and L = Q(i). D = (−1,−1)F is the quaternion division algebra used in the Silver code, and for a, b ∈ Q(√−7), σ(a+ib) = a−ib. The entries of the matrices of the Silver code are elements of the order OK ⊕ jOK in D = (−1,−1)Q(√−7), where OK = Z[i] ⊕ Z[i]( 1+√−7 ) is the ring of integers of K. 2 16 S. PUMPL UN By Theorem 17, A = (−1,−1)Q ⊗Q (Q(√−7)/Q, τ, d) is a division algebra for all choices of d ∈ Q(√−7) \ Q. Therefore the code C given by the matrix x0 −σ(x1) dτ (y0) −dτ (σ(y1)) dτ (y1) dτ (σ(y0)) x1 σ(x0) τ (x0) −τ (σ(x1)) y0 −σ(y1) τ (σ(x0)) τ (x1) σ(y0) y1     , representing left multiplication in A, xi, yi ∈ Q(√−7), is fully diverse and has NVD since det X ∈ Q(√−7) [12], for all choices of d ∈ Q(√−7)\Q. Therefore its minimum determinant is at least 1. This rough estimate puts our code family already among the top three places in [8, Table I], along with the code proposed in [8] and the SR-code, which compares the fast-decodable codes with best minimum determinant and decoding complexity. This is because C has decoding complexity at most O(M 2m2−3m/2) = O(M 5) no matter the choice of d ∈ Q(√−7) \ Q, if the matrix entries take values from M -QAM ⊂ Z[i], thus is fast-decodable of maximal rate 2 by [13, Section 6.1]. In particular, it has optimal diversity-multiplexing gain trade-off (DMT). The usual hard-limiting as done for the SR-code then lowers its decoding complexity to O(M 4.5). As a further comparison, the iterated Silver code in [9, Section IV.A], which does not arise from a tensor product (since there σ = τ ), initially has decoding complexity at most O(M 13) which is then further reduced to O(M 10) by scaling. References [1] A. A. Albert, "Structure of algebras", in American Mathematical Society Colloquium Publications, Provdence, RI, USA AMS 24, 1961. [2] V. Astier, S. Pumplun, Nonassociative quaternion algebras over rings, Israel J. Math. 155 (2006), 125- 147. [3] N. Bourbaki, Alg`ebre, Chapter 8: Modules et anneaux semi-simples, Second revised edition of the 1958 edition, Springer Verlag, Berlin, 2012. [4] C. Brown, PhD thesis, University of Nottingham, in preparation. [5] L.E. Dickson, Linear algebras in which division is always uniquely possible, Trans. AMS 7 (3) (1906), 370-390. [6] N. Jacobson, "Finite-dimensional division algebras over fields," Springer Verlag, Berlin-Heidelberg-New York, 1996. [7] M. Lavrauw, J. Sheekey, Semifields from skew-polynomial rings, Adv. Geom. 13 (4) (2013), 583-604. [8] M. Liu, M. Hlard, J.-F. Hlard, M. Crussire, A fast decodable full-rate STBC with high coding gain for 4x2 MIMO systems. 2013 IEEE 24th International Symposium on Personal Indoor and Mobile Radio Communications (PIMRC), London, United Kingdom (2013). [9] N. Markin, F. Oggier, Iterated space-time code constructions from cyclic algebras, IEEE Trans. Inform. Theory (9) 59, September 2013, 5966-5979. [10] 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. [11] S. Pumplun, How to obtain algebras used for fast-decodable space-time block codes, Adv. Math. Comm. 8 (3) (2014), 323-342. [12] S. Pumplun, A. Steele, Fast-decodable MIDO codes from nonassociative algebras. Int. J. of Information and Coding Theory (IJICOT) 3 (1) 2015, 15-38. TENSOR PRODUCTS OF NONASSOCIATIVE CYCLIC ALGEBRAS 17 [13] S. Pumplun, A. Steele, The nonassociative algebras used to build fast-decodable space-time block codes, Adv. Math. Comm. 9 (4) 2015, 449-469. [14] S. Pumplun, A. Steele, Algebras with semi-multiplicative maps, online at http://molle.fernuni-hagen.de/~loos/jordan/index.html [15] S. Pumplun, T. Unger, Space-time block codes from nonassociative division algebras, Adv. Math. Comm. 5 (3) (2011), 609-629. [16] K. P. Srinath, B. S. Rajan, Fast-decodable MIDO codes with large coding gain, IEEE Transactions on Information Theory (2) 60 2014, 992-1007. [17] R. Sandler, Autotopism groups of some finite non-associative algebras, Amer. J. Math. 84 (1962), 239-264. [18] R.D. Schafer, "An Introduction to Nonassociative Algebras," Dover Publ., Inc., New York, 1995. [19] A. Steele, Nonassociative cyclic algebras, Israel J. Math. 200 (1) (2014), 361-387. [20] A. Steele, Some new classes of division algebras and potential applications to space-time block coding, PhD Thesis, Nottingham, 2013. [21] W.C. Waterhouse, Nonassociative quaternion algebras, Algebras Groups Geom. 4 (3) (1987), 365-378. E-mail address: [email protected] School of Mathematical Sciences, University of Nottingham, University Park, Nottingham NG7 2RD, United Kingdom
1702.03922	3	1702	2018-11-22T23:56:29	On free Gelfand--Dorfman--Novikov superalgebras and a PBW type theorem	[ "math.RA" ]	We construct a linear basis of a free GDN superalgebra over a field of characteristic $\neq 2$. As applications, we prove a PBW theorem, that is, any GDN superalgebra can be embedded into its universal enveloping commutative associative differential superalgebra. An Engel theorem under some assumptions is given.	math.RA	math	ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† Abstract. We construct linear bases of free GDN superalgebras. As applications, we prove a Poincar´e -- Birkhoff -- Witt type theorem, that is, every GDN superalgebra can be embedded into its universal enveloping associative differential supercommuative algebra. An Engel theorem is given. 1. Introduction We recall that a superalgebra over a field k is a vector space A with a direct sum decomposition A = A0 ⊕ A1 together with a bilinear multiplication ◦: A × A 7→ A such that Ai ◦ Aj ⊆ Ai+j, where the subscripts are elements of Z2. The parity x of every element x in A0 is 0, and the parity x of every nonzero element x in A1 is 1. If a superalgebra A satisfies the following two identities (x ◦ (y ◦ z)) − ((x ◦ y) ◦ z) = (−1)xy((y ◦ (x ◦ z)) − ((y ◦ x) ◦ z)) (left supersymmetry), and ((x ◦ y) ◦ z) = (−1)yz((x ◦ z) ◦ y) (right supercommutativity) for all elements x, y, z in A0 ∪A1, then A is called a (left) Novikov superalgebra [17]. (There is a "right" version of using right supersymmetry and left supercommutativity.) Moreover, if a (left) Novikov superalgebra A equals to its even part, i.e., A = A0, then A is just an ordinary (left) Novikov algebra [3, 11, 13]. Since Novikov algebras were invented by Balinskii and Novikov [3], and independently by Gelfand and Dorfman [11], we also call Novikov algebra as Gelfand -- Dorfman -- Novikov algebra (GDN algebra) and call Novikov superalgebra as Gelfand -- Dorfman -- Novikov superalgebra (GDN superalgebra). A rich structure and combinatorial theory of GDN algebras have been done up to now. Zelmanov solved Novikov's problem on classification of simple GDN algebras over an algebraically closed field: There are no such algebras besides trivial [18]. Osborn and Zelmanov classified simple GDN algebras A over an algebraically closed field of 1991 Mathematics Subject Classification. 16S15, 17A30, 17A70, 17B30. Key words and phrases. GDN superalgebra; Poincar´e-Birkhoff-Witt theorem; nilpotency. ‡Supported by the NNSF of China (11571121), the NSF of Guangdong Province (2017A030313002) and the Science and Technology Program of Guangzhou (201707010137). ∗ Supported by the Innovation Project of Graduate School of South China Normal University. ♯ Corresponding author. † Supported by Russian Science Foundation (project 14-21-00065). 1 2 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† characteristic 0 with a maximal subalgebra H such that A/H has a finite dimensional irreducible H-submodule [14]. Xu gave a complete classification of finite dimensional simple GDN algebras and their irreducible modules over an algebraically closed field with prime characteristic [15], and he introduced some quadratic GDN superalgebras connecting with Gelfand -- Dorfman (Ω-bi) algebras [16] (Gelfand -- Dorfman (Ω-bi) algebras were invented in [11]). See also, for example, Bai and Meng [1, 2], Burde and Dekimpe [6], Chen, Niu and Meng [7], Kang and Chen [12], Zhu and Chen [19], Bokut, Chen and Zhang [4, 5]. Dzhumadildaev and Lofwall proved that the set of all the Novikov tableaux (we call them GDN tableaux because of the above reason) over a well-ordered set X forms a linear basis of a free GDN algebra generated by X by using trees and by appealing to the connection with free commutative associative differential algebra [8] (the idea of this connection was given by S.I. Gelfand, see [11]). And we wonder what would a basis of a free GDN superalgebra be like. The method of using trees developed in [8] can not be directly applied for GDN superalgebras, but the idea of tracing a root of a tree can be modified to define the root number of a term. Moreover, the definition of GDN tableau can be easily extended to a definition of GDN supertableau, see Definition 2.6. One of the results we prove below is as follow: Theorem A. The set of all the GDN supertableaux over a well-ordered set X = X0 ∪ X1 forms a linear basis of the free GDN superalgebra GDNs(X) generated by X, where every element of the set X0 is of parity 0 and every element of the set X1 is of parity 1. We also prove a Poincar´e -- Birkhoff -- Witt (PBW) type theorem for GDN superalgebras: Theorem B. Every GDN superalgebra can be embedded into its universal enveloping associative differential supercommutative algebra. As a corollary, we show that every GDN superalgebra generated by a finite set of elements of parity 1 is nilpotent. Several results concerning the nilpotency of certain GDN algebras have been found up to now. Zelmanov proved that, if A is a left-nilpotent finite dimensional (right) GDN algebra over a field of characteristic zero, then A2 is nilpotent [18]. Filippov proved that a right-nil algebra of bounded index over a field of characteristic zero is right nilpotent provided that it is right symmetric and is nilpotent provided that it is a right GDN algebra [10]. Dzhumadildaev and Tulenbaev proved that if a (right) GDN algebra A over a field of characteristic p is left-nil of bounded index n and p = 0 or p > n, then A2 is nilpotent [9]. Again, to some extent, this kind of result can be extended to the case of GDN superalgebras, and we prove the following Engel theorem: Theorem C. Let A = A0 ⊕ A1 be a (left) GDN superalgebra over a field of characteristic 0 generated by X = X0 ∪ X1, where every element of the set X0 is of parity 0 and every element of the set X1 is of parity 1. If for some integer n > 0, the even part A0 is right-nil of bounded index n and X1 is a finite set, then A2 is nilpotent. The paper is organized as follows. In section 2, we construct a linear generating set Tabs(X) for a free GDN superalgebra generated by a well-ordered set X over a field of characteristic 6= 2. (For the case of characteristic 2, a GDN superalgebra is the same as ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 3 a GDN algebra, so a linear basis is already known [8]). In section 3, we show the linear independence of Tabs(X), and we also prove a PBW type theorem for GDN superalgebras. In section 4, we prove Theorem C, an Engel type theorem for GDN superalgebras. 2. A linear generating set of GDNs(X) Our aim in this section is to construct a specific linear generating set of the free GDN superalgebra GDNs(X) generated by a well-ordered set X. (We shall show that the set we constructed is indeed linearly independent in the next section.) The idea of our construction is reminiscent of what was done for GDN algebras in [8]. However, the original method of using trees is not extended directly. So we develop a new notion of root number of a term. In the whole paper, we assume that X = X0 ∪ X1 is a fixed well-ordered set, where every element of the set X0 is of parity 0 and every element of the set X1 is of parity 1. We also assume that the characteristic char(k) of the field k is not 2. 2.1. The root number of a term. In this subsection, we first recall the definition of terms. Then we define the root map from the set of all terms over a set X to the set of nonnegative integers. Finally, we develop several handy properties of the root map. They will be useful in the sequel when we try to develop a method of writing an arbitrary term into a linear combination of some specified terms (hereafter called GDN supertableaux). We recall that terms over X are defined by the following induction: (i) Every element a of X is a term over X; (ii) If µ and ν are terms over X, then (µ ◦ ν) is a term over X. Denote by X (∗) the set of all terms over X. For every term µ in X (∗), the length ℓ(µ) of µ is defined to be 1 if µ lies in X, and ℓ(µ) is defined to be ℓ(µ1) + ℓ(µ2) if µ = (µ1 ◦ µ2) for some terms µ1 and µ2 in X (∗). Similar to the definition of length, for every term µ in X (∗), the parity µ of µ satisfies the following claims: (i) µ = 0 if µ lies in X0, and µ = 1 if µ lies in X1. (ii) µ = µ1 + µ2 modulo 2 if µ = (µ1 ◦ µ2). Definition 2.1. We define a root map r from the set X (∗) to the set Z>0 of nonnegative integers defined inductively as follows: (i) r(a) = 0 for every element a in X; (ii) r((µ ◦ ν)) = r(µ) + 1 if ν lies in X, and r((µ ◦ ν)) = r(µ) + r(ν) if ν does not lie in X. For every term µ in X (∗), we call r(µ) the root number of µ to indicate that our idea is based on [8], in which the authors appealed to the tool of trees (and roots of trees). For all terms µ1, ..., µn in X (∗), to make the notations shorter, define [µ1, ..., µn]L = ((...((µ1 ◦ µ2) ◦ µ3)◦ ···) ◦ µn) (left-normed bracketing), [µ1, ..., µn]R = (µ1 ◦ (··· ◦(µn−2 ◦ (µn−1 ◦ µn))...)) (right-normed bracketing). Moreover, if µ1, ..., µn are elements of X, then we call [µ1, ..., µn]R a simple term over X of length n. Below we offer an instance of counting the root number of a term in X (∗). 4 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† Example 2.2. For every positive integer n, for all elements ai (1 6 i 6 n) in X, we have (i) r([a1, ..., an]R) = 1; (ii) r([a1, ..., an]L) = n − 1. In general, the root number of a term µ is not uniquely decided by the length ℓ(µ). The following lemma shows that the root number r(µ) is bounded above by ℓ(µ) − 1. Lemma 2.3. For every term µ in X (∗), we have r(µ) 6 ℓ(µ) − 1, with equality only if µ = [a1, ..., aℓ(µ)]L for some elements a1, ..., aℓ(µ) in X. Proof. Use induction on ℓ(µ). For ℓ(µ) = 1, we have r(µ) = 0 = ℓ(µ) − 1. For ℓ(µ) > 1, we have µ = (µ1 ◦ µ2) for some terms µ1 and µ2 in X (∗). If ℓ(µ2) > 1, then by induction hypothesis, we have r(µ) = r(µ1) + r(µ2) 6 ℓ(µ1) − 1 + ℓ(µ2) − 1 < ℓ(µ) − 1. On the other hand, if ℓ(µ2) = 1, then by induction hypothesis, we have r(µ) = r(µ1) + 1 6 ℓ(µ1) − 1 + 1 = ℓ(µ1) = ℓ(µ) − 1, with the equality only if r(µ1) = ℓ(µ1) − 1. The claim follows by induction hypothesis. (cid:3) The following lemma offers a formula for counting the root number of a term. Lemma 2.4. For every term µ = (µ1, µ2) in X (∗), we have r(µ) = r(µ1) + max(1, r(µ2)) > 1, with r(µ) = 1 only if µ = [a1, ..., aℓ(µ)]R for some elements a1, ..., aℓ(µ) in X. Proof. Use induction on ℓ(µ). For ℓ(µ) = 2, the terms µ1 and µ2 lie in X, so the claim follows. For ℓ(µ) > 2, if µ2 lies in X, then max(1, r(µ2)) = 1; if ℓ(µ2) > 2, then by induction hypothesis, we have r(µ2) > 1 and so max(1, r(µ2)) = r(µ2). Therefore, we obtain r(µ) = r(µ1) + max(1, r(µ2)) > 1. If the equality r(µ) = 1 holds, then the induction hypothesis forces ℓ(µ1) = 1. The claim (cid:3) follows. The following lemma shows that the root map is compatible with the right supercom- mutativity, and to some extent, the root map is also compatible with the product ◦. Lemma 2.5. For all terms µ1, µ2 and µ3 in X (∗), we have (i) r([µ1, µ2, µ3]L ) = r([µ1, µ3, µ2]L); (ii) If r(µ1) > r(µ2), then r((µ1 ◦ µ3)) > r((µ2 ◦ µ3)); (iii) If r(µ1) > r(µ2) and r(µ1) > 1, then r((µ3 ◦ µ1)) > r((µ3 ◦ µ2)); (iv) If r(µ1) > r(µ2) and r(µ1) = 1, then r((µ3 ◦ µ1)) = r((µ3 ◦ µ2)). Proof. The lemma follows immediately from Lemma 2.4. (cid:3) ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 5 2.2. GDN supertableaux. Now we are ready to define the notion of a GDN supertableau, which is directly reminiscent of the notation of a GDN tableau. Our aim in this subsection is to show that, if char(k) 6= 2, then the set of all GDN supertableaux over X forms a linear generating set of the free GDN superalgebra GDNs(X). Definition 2.6. We call a term µ a Gelfand -- Dorfman -- Novikov supertableau (GDN su- pertableau) over a well-ordered set X = X0 ∪ X1 if, for some letter a in X, for some nonnegative integer n, and for some simple terms µi = [ai,ri, ..., ai,1]R (1 6 i 6 n) over X of length ri > 1, we have (2.1) µ = [a, µ1, ..., µn]L such that the following conditions hold: (i) The integers r1, ..., rn satisfy that r1 > · · · > rn; (ii) If ri = ri+1, then ai,1 > ai+1,1 holds; (iii) The inequality a > a1,r1 > · · · > a1,2 > a2,r2 > · · · > a2,2 > · · · > an,rn > · · · > an,2 holds; (iv) If for some integers i, t, j and l satisfying i 6 n, t 6 n, 2 6 j 6 rj and 2 6 l 6 rt, the letters ai,j and at,l lie in X1, then the inequality ai,j 6= at,l holds; (v) If for some integer i 6 n − 1, the elements ai,1 and ai+1,1 lie in X1, and ri = ri+1, then the inequality ai,1 6= ai+1,1 holds. Every term of the form (2.1) satisfying Points (i)-(iii) is called a Novikov tableau [8] over X, and we call it a GDN tableau because of the reason explained in the introduction. Denote by Tabs(X) the set of all the GDN supertableaux over X. It is quite easy to show that every term in X (∗) ⊆ GDNs(X) of the form (2.1) can be written as a linear combination of terms satisfying Points (i) and (ii) by the right supercommutativity, but what remains becomes complicated and we will need the notion of root number of a term. The strategy for rewriting is to apply the right supercommutativity and the left super- symmetry. Unfortunately, whenever we apply the left supersymmetry to a term, we shall get three other terms in return, and thus this process becomes complicated. So a simplified notation is needed. Because of this reason, we introduce the following notation. Definition 2.7. For all terms µ and ν in X (∗) such that r(µ) = r(ν) and ℓ(µ) = ℓ(ν), for all nonzero elements α and β in the field k, the polynomials αµ and βν in GDNs(X) are said to be equivalent, denoted by if αµ − βν = Pi αiµi in GDNs(X) for some elements αi in k and terms µi in X (∗) such that r(µ) < r(µi) and ℓ(µ) = ℓ(µi) for every i. αµ ∼ βν, It is clear that if αµ ∼ βν and βν ∼ α′µ′, then we get αµ ∼ α′µ′. Recall that for every element a of X, the parity a of a is i if a lies in Xi with i = 0, 1. Moreover, for all elements a1, ..., an (n > 1) of X, we define the parity a1...an of the string a1...an to be a1+ ··· +an modulo 2, extended with ε = 0 for the empty string ε. The following lemma shows that, for every simple term [ar, ..., a1]R with r > 3, we can rearrange ar, ..., a2 at the expense of adding a linear combination of terms of length r 6 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† and with root numbers > 1. We shall see in Lemma 2.10 that the added terms do not increase the difficulty of rewriting an arbitrary term into a linear combination of GDN supertableaux. Lemma 2.8. For all elements a1, ..., ar in X, for every simple term µ = [ar, ..., a1]R, the following claims hold: (i) For every integer j such that 2 6 j < r, we have µ ∼ (−1)aj aj+1...ar[aj, ar, ..., aj+1, aj−1, ..., a1]R; (ii) For all integers i and j such that 2 6 j < i 6 r, we have µ ∼ (−1)aiaj ...ai−1+ajaj+1...ai−1[ar, ..., ai+1, aj, ai−1, ..., aj+1, ai, aj−1, ..., a1]R. Proof. We shall just prove Point (i), because Point (ii) can be proved in a similar way. Assume ν = [aj−1, ..., a1]R. Then by the left supersymmetry, we have µ = (−1)aj aj+1[ar, ..., aj+2, aj, aj+1, ν]R + [ar, ..., aj+2, (aj+1 ◦ aj), ν]R − (−1)aj aj+1[ar, ..., aj+2, (aj ◦ aj+1), ν]R . Since r([ar, ..., aj+2, (aj+1 ◦ aj), ν]R) = r([ar, ..., aj+2, (aj ◦ aj+1), ν]R) = 2 > 1 = r(µ) = r([ar, ..., aj+2, aj, aj+1, ν]R), we have By induction on r − j, we obtain µ ∼ (−1)aj aj+1[ar, ..., aj+2, aj, aj+1, ν]R. µ ∼ (−1)aj aj+1[ar, ..., aj+2, aj, aj+1, ν]R ∼ (−1)aj aj+1...ar[aj, ar, ..., aj+1, aj−1, ..., a1]R. The proof is completed. (cid:3) For every simple term µ = [ar, ..., a1]R in X (∗), for every integer i such that 2 6 i 6 r, we define and µ ai = [ar, ..., ai+1, ai−1, ..., a1]R µai7→bj = [ar, ..., ai+1, bj, ai−1, ..., a1]R. The following lemma is crucial to the construction of a linear basis of the free GDN superalgebra GDNs(X). It shows that, for the product of two simple terms, we can "interchange" certain letters of the two simple terms in the sense of adding a linear combination of some nonessential terms. We shall see that, as a result of the following lemma, the set of all the GDN tableaux over X is not linearly independent in GDNs(X) provided that X1 is nonempty and the characteristic of the field is not 2. ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 7 Lemma 2.9. For all elements a1, ..., ar+1, b1, ..., bm (r > 2, m > 2) in X, for all integers i, j such that 2 6 i 6 r + 1 and 2 6 j 6 m, for all simple terms µ = [ar+1, ..., a1]R and ν = [bm, ..., b1]R, we can interchange ai and bj in (µ ◦ ν) in the sense that (2.2) (µ ◦ ν) ∼ (−1)aiai−1...a1bm...bj+bjai−1...a1bm...bj+1(µai7→bj ◦ νbj 7→ai). In particular, if r = m, a1 = b1 and a1 = 1, then we get (µ ◦ ν) ∼ −(µ ◦ ν). Since char(k) 6= 2, the term (µ ◦ ν) can be written as a linear combination of terms that are of root numbers > r(µ) + r(ν) and with lengths ℓ(µ) + ℓ(ν). Proof. By Lemmas 2.5 and 2.8, we get (µ ◦ ν) ∼ (−1)aiai+1...ar+1((ai ◦ µ ai) ◦ ν) ∼ (−1)aiai+1...ar+1+µ ai ν((ai ◦ ν) ◦ µai) ∼ (−1)aiai+1...ar+1+µ ai ν+aibj ...bm+bjbj+1...bm((bj ◦ νbj7→ai) ◦ µ ai) ∼ (−1)aiai+1...ar+1+µν+aib1...bj−1+bjbj+1...bm+µai νbj 7→ai ((bj ◦ µ ai) ◦ νbj7→ai) ∼ (−1)aiai+1...ar+1b1...bj−1+µν+bjai+1...ar+1bj+1...bm+µai νbj 7→ai (µai7→bj ◦ νbj 7→ai) ∼ (−1)aiai−1...a1bm...bj+1+bjai...a1bj+1...bm(µai7→bj ◦ νbj 7→ai). In particular, if r = m, a1 = b1 and a1 = 1, then we obtain (µ ◦ ν) ∼ (−1)ar ar−1...a1+brar ...a1(µar 7→br ◦ νbr7→ar ) ∼ (−1)ar ar−1...a1+brbr−1ar...a1+ar−1ar ar−1...a1((µar 7→br )ar−17→br−1 ◦ (νbr7→ar )br−17→ar−1) ∼ · · · ∼ (−1)ar ar−1...a2+ar ar−1...a2ar...a1+brbr−1...b2ar...a1((ar+1 ◦ νb17→a1) ◦ (µar+1)a17→b1) ∼ (−1)ar ...a2a1+br...b2ar ...a1+ar ar−1...a2b1brbr−1...b2a1((ar+1 ◦ (µar+1)a17→b1) ◦ νb17→a1). Since a1 = b1, we obtain (ar+1◦(µar+1)a17→b1) = µ and νb17→a1 = ν. Moreover, a1 = b1 = 1 implies that (−1)ar ...a2a1+br...b2ar...a1+arar−1...a2b1brbr−1...b2a1 = (−1). Therefore, we (cid:3) obtain (µ ◦ ν) ∼ −(µ, ν). Since char(k) 6= 2, the claim follows. Now we are in a position to show that the set of all the GDN supertableaux over a well-ordered set X = X0 ∪ X1 forms a linear generating set of the free GDN superalge- bra GDNs(X) generated by X. Lemma 2.10. For every term λ in X (∗), we have λ = Pi αiλi for some elements αi in the field k and for some GDN supertableaux λi such that ℓ(λi) = ℓ(λ) and r(λi) > r(λ). Proof. We use induction on ℓ(λ). For ℓ(λ) 6 2, it is clear. For ℓ(λ) > 2, we use a second (downward) induction on r(λ). For r(λ) = ℓ(λ) − 1, by Lemma 2.3 and by the right supercommuativity, we may assume λ = [a1, ..., aℓ(λ)]L and a2 >···> aℓ(λ). If Condition Definition 2.6(v) is not satisfied, then λ = 0, otherwise, [a1, ..., aℓ(λ)]L is already a GDN supertableau. For r(λ) < ℓ(λ) − 1, we may assume that λ = (µ ◦ ν) for some terms µ, ν with ℓ(µ) < ℓ(λ) and ℓ(ν) < ℓ(λ). By induction hypothesis, both µ and ν can be written as j. Then linear combinations of GDN supertableaux, say µ = Pi αiµ′ i and ν = Pj βjν′ 8 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† for all i, j, we have r((µ′ and ℓ((µ′ j)) = ℓ((µ ◦ ν)). i ◦ ν′ j )) > r(µ) + max(1, r(ν)) = r((µ ◦ ν)) i ◦ ν′ j)) = r(µ′ i) + max(1, r(ν′ Now we can assume that µ and ν are GDN supertableaux. Suppose that µ = [a, µ1, ..., µp]L and ν = [b, ν1, ..., νq]L, where all the µi and νj are simple terms, and a, b are elements in X. For ℓ(µ) > 1, we have p > 0 and λ = (µ ◦ ν) = (−1)νµp([a, µ1, ..., µp−1, ν]L ◦ µp). By induction hypothesis again, we can write the term [a, µ1, ..., µp−1, ν]L as a linear combination of GDN supertableaux. Therefore, we may assume that ℓ(µ) > 1 and ν is a simple term. In other words, we can assume that λ = [a, λ1, ..., λn]L (n > 1), where a is an element in X and each λi = [ai,ri, ..., ai,1]R is a simple term. We first show that in this case λ can be written as a linear combination of GDN supertableaux with the claimed conditions. This is the main case with which we should deal. Applying the right supercommuativity whenever necessary, we may assume that the conditions of Definition 2.6(i)-(ii) are satisfied. By Lemmas 2.5, 2.8 and 2.9, and by induction hypothesis, the conditions of Definition 2.6(iii) can also be obtained. For instance, say a2,2 < a4,3. Then we need to interchange a2,2 and a4,3 in the sense of Equation (2.2). Since [a, λ1, ..., λn]L = (−1)λ2λ1+λ4λ3+λ4λ1[a, λ2, λ4, λ1, λ3, λ5, ..., λn]L, we can apply Lemma 2.9 to the term ((a ◦ λ2) ◦ λ4). Suppose that the conditions of Definition 2.6(iv) are destroyed. If ai,j = ai,j+1 ∈ X1 for some integers i and j such that 1 6 i 6 n and 2 6 j < ri, then by left supersymmetry, we obtain λi − [ai,ri, ..., ai,j+2, (ai,j+1 ◦ ai,j), ai,j−1, ..., ai,1]R = −1(λi − [ai,ri, ..., ai,j+2, (ai,j ◦ ai,j+1), ai,j−1, ..., ai,1]R). Since char(k) 6= 2, we have λi = [ai,ri, ..., ai,j+2, (ai,j+1 ◦ ai,j), ai,j−1, ..., ai,1]R in GDNs(X) and r([ai,ri, ..., ai,j+2, (ai,j+1 ◦ ai,j), ai,j−1, ..., ai,1]R) = 2 > 1 = r(λi). By the second induction hypothesis on root numbers, we are done. Therefore, we may assume that, for every integer j > 2, we have ai,j 6= ai,j+1 if ai,j lies in X1. Similarly, we may assume that a 6= a1,r1 if a lies in X1. On the other hand, if ai,2 = ai+1,ri+1, then we have λ = α1[a, λi, λi+1, λ1, , ..., , λi−1, λi+2, ..., λn]L ∼ α2[ai,2, (λi)ai,27→a, λi+1, λ1, , ..., , λi−1, λi+2, ..., λn]L ∼ α3[ai,2, λi+1, (λi)ai,27→a, λ1, , ..., , λi−1, λi+2, ..., λn]L for some elements α1, α2 and α3 in k. The claim follows by the above reasoning. Finally, by Lemma 2.9, right supercommutativity and by induction hypothesis, the conditions of Definition 2.6(v) can also be satisfied. For ℓ(µ) = 1, we have ℓ(ν) > 2 and thus q > 1. For q = 1, the term λ = (µ ◦ ν) is a simple term. By Lemma 2.8 and induction hypothesis on root number, we are done. For q > 1, we shall resort to the case of ℓ(µ) > 1. By left supersymmetry, we have λ = (µ ◦ [b, ν1, ..., νq]L) = ((µ ◦ [b, ν1, ..., νq−1]L ) ◦ νq) ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 9 + (−1)µ[b,ν1,...,νq−1]L (([b, ν1, ..., νq−1]L ◦ (µ ◦ νq)) − (([b, ν1, ..., νq−1]L ◦ µ) ◦ νq)). By induction hypothesis and the above reasoning for the case of ℓ(µ) > 1, the result (cid:3) follows. 3. A linear basis of GDNs(X) and a Poincar´e-Birkhoff-Witt type Theorem Our aim in this section is to show that the set Tabs(X) of all the GDN supertableaux over a well-ordered set X = X0∪X1 forms a linear basis of the free GDN superalgebra GDNs(X). We already know that it is a linear generating set of GDNs(X), so what remains is to prove the linear independence. We shall also prove a PBW type theorem for GDN superalgebra, that is, every GDN superalgebra can be embedded into its universal enveloping associative differential supercommutative algebra. 3.1. Associative differential supercommutative algebra. In this subsection, we shall first construct the free associative differential supercommuative algebra generated by X. It will be instrumental in proving the linear independence of the set Tabs(X) of all the GDN supertableaux over X. Recall that a supercommutative algebra is a superalgebra A satisfying the following identity: x · y = (−1)xyy · x for all elements x, y in A0 ∪ A1, and an associative differential supercommutative algebra is an associative supercommutative algebra (A, ·, D) with a linear derivation D of parity 0 satisfying that D(Ai) ⊆ Ai (i = 0, 1) and the identity: D(x · y) = D(x) · y + x · D(y), for all elements x, y in A0 ∪ A1. S.I. Gelfand [11] pointed out that, every associative differential commutative alge- bra (A, ·, D) becomes a GDN algebra under the new operation ◦ defined by x◦y := x·D(y), and with the help of this discovery, Dzhumadildaev and Lofwall proved that the set of all the GDN tableaux over a well-ordered set X forms a linear basis of the free GDN algebra generated by X. This idea motivates us to establish the connection of GDN superalgebra and associative differential supercommutative algebra. The proof for the following observation is straightforward and thus omitted. Lemma 3.1. For every associative differential supercommutative algebra (A, ·, D), if we define a new bilinear operation on A by the rule: x ◦ y = x · D(y) for all elements x and y in A, then (A, ◦) becomes a GDN superalgebra. Let A be an associative differential supercommutative algebra over k generated by a set X = X0 ∪ X1. We say A is free on X if, for every map ψ of X into an associative differential supercommuative algebra B = B0 ⊕ B1 such that ψ(Xi) ⊆ Bi (i = 0, 1), there exists a unique homomorphism ϕ : A → B extending ψ. We shall construct the free associative differential supercommutative algebra ks{X} generated by a set X directly. 10 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† Define D0(a) = a for every a in X. Define Y = {Dn(a) a ∈ X, n > 0, n ∈ N} and let Y + be the free semigroup (without unit) generated by Y . For every u = Di1(a1)...Din(an) in Y +, define the parity u of u to be a1+ ··· +an modulo by 2, and define Di(a) < Dj(b) if (i, a) < (j, b) lexicographically. Finally, define Ds[X] := {Di1(a1)...Din (an) ∈ Y + Di1(a1), ..., Din (an) ∈ Y , Di1(a1) 6···6 Din(an), if ap = aq ∈ X1 for some integers p 6= q 6 n, then ip 6= iq}. Let kDs[X] be the k linear space with a k-basis Ds[X]. Define a bilinear operation · on the space kDs[X] as follows: For all u = Di1(a1)...Din (an), v = Dj1(b1)...Djm(bm) and Dj(b) in Ds[X], (3.1) if b lies in X1 and Dj(b) = Dit(at) for some integer t 6 n, then u · Dj(b) is defined to be 0. Otherwise, assume that Di1(a1)...Dit−1 (at−1)Dj(b)Dit(at)...Din (an) lies in Ds[X] for some integer t satisfying 1 6 t 6 n + 1, where t = 1 (or t = n + 1, resp.) means Di1(a1)...Dit−1(at−1) (or Dit(at)...Din (an), resp.) is an empty sequence. Then define u · Dj(b) to be (−1)Pp(apb)Di1(a1)...Dit−1 (at−1)Dj(b)Dit(at)...Din (an), where the sum is over all the integer p such that Dip(ap) > Dj(b). Next, the product u · v is defined inductively as follows: u · v := (u · Dj1(b1)) · Dj2(b2)...Djm (bm). Finally, define a unary linear operation D on kDs[X] as follows: D(u) = X (Di1(a1)...Dit−1(at−1) · Dit+1(at)) · Dit+1(at+1)...Din (an). 16t6n The following lemma offers an explicit formula for calculating the product of arbitrary two elements in Ds[X]. Lemma 3.2. Let u and v be as in Equation (3.1). If u · v 6= 0, then we have u · v = (−1)P(p,q)(apbq)Dl1(d1)...Dln+m (dn+m), where Dl1(d1), ..., Dln+m (dn+m) is a reordering of Di1(a1), ..., Din (an), Dj1(b1), ..., Djm (bm) such that Dl1(d1)...Dln+m (dn+m) lies in Ds[X], and the sum is over all the pairs (p, q) such that Dip(ap) > Djq (bq). Moreover, the equality u · v = 0 holds if and only if, for some integers t 6 n and l 6 m, we have it = jl and at = bl ∈ X1. Proof. The second claim is clear, so we just prove the first one. Use induction on m. For m = 1, the claim follows by the definition of the operation ·. For m > 1, since the inequality Dj1(b1) 6 Dj2(b2) 6···6 Djm(bm) holds, we obtain u · v = (u · Dj1(b1)) · Dj2(b2)...Djm (bm) = (−1)Pp(apb1)Di1(a1)...Dit−1 (at−1)Dj1(b1)Dit(at)...Din (an) · Dj2(b2)...Djm(bm) = (−1)P(p,q)(apbq)Dl1(d1)...Dln+m (dn+m) ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 11 with the desired properties. (cid:3) Now we are in a position to show that, endowed with the defined operations · and D, the vector space kDs[X] becomes a free associative differential supercommutative algebra. Lemma 3.3. The algebra (kDs[X], ·, D) is isomorphic to the free associative differential supercommutative algebra ks{X} generated by X. In particular, if we define a linear operation ◦ on (kDs[X], ·, D) by the rule: u ◦ v = u · D(v) for all u and v in Ds[X], then (kDs[X], ◦) becomes a GDN superalgebra. Proof. We first show that (kDs[X], ·, D) is an associative differential supercommutative al- gebra. By Lemma 3.2, the associativity is straightforward. As for the supercommutativity, let u and v be as in Equation (3.1). For u · v = 0, it is clear that v · u = 0 = (−1)uvu · v. For u · v 6= 0, with the same notation of Lemma 3.2, we have u · v = (−1)P(p,q)(apbq)Dl1(d1)...Dln+m (dn+m), where the sum is over all the pairs (p, q) such that Dip(ap) > Dj(b). Similarly, we get v · u = (−1)P(p′,q)(ap′ bq)Dl1(d1)...Dln+m (dn+m), where the sum is over all the pairs (p′, q) such that Dip′ (ap′) < Dj(b). Combining the above two formulas, we get u · v = (−1)P(p,q)(apbq)v · u, where the sum is over all the pairs (p, q) such that Dip(ap) 6= Dj(b). Moreover, if ip = jq and ap = bq for some integers p, q such that 1 6 p 6 n and 1 6 q 6 m, then ap lies in X0 and thus (−1)apbq = 1. So we obtain v · u = (−1)uvu · v. To show that D(u · v) = D(u) · v + u · D(v), we use induction on m. For m = 1, we have D(u · Dj1(b1)) = (−1)Pp(apb1)D(Di1(a1)...Dit−1(at−1)Dj(b)Dit(at)...Din (an)) = (−1)Pp(apb1)((Di1(a1)...Dit−1(at−1) · Dj1+1(b1)) · Dit(at)...Din(an) (Di1(a1)...Diq−1 (aq−1) · Diq+1(aq)) · Diq+1(aq+1)...Dj1(b1)...Din (an) + X + X 16q6t−1 (Di1(a1)...Dj1(b1)...Dip−1(ap−1) · Diq+1(aq)) · Diq+1(aq+1)...Din (an)) t6q6n = D(u) · Dj1(b1) + u · Dj1+1(b1) (by applying associativity and supercommutativity). For m > 1, we obtain D(u · v) = D((u · Dj1(b1)) · Dj2(b2)...Djm(bm)) = D(u · Dj1(b1)) · Dj2(b2)...Djm(bm) + (u · Dj1(b1)) · D(Dj2(b2)...Djm(bm)) = (D(u) · Dj1(b1)) · Dj2(b2)...Djm(bm) + (u · Dj1+1(b1)) · Dj2(b2)...Djm(bm) + (u · Dj1(b1)) · D(Dj2(b2)...Djm (bm)) = D(u) · v + u · D(v). 12 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† Therefore, (kDs[X], ·, D) is an associative differential supercommutative algebra. It remains to show that (kDs[X], ·, D) is free on X. By applying associativity and supercommutativity in ks{X}, it is easy to see that the set of all the monomials of the form: (((...(Di1 (a1) · Di2(a2))· ···) · Din−1(an−1)) · Din(an)) (left-normed bracketting) such that Di1(a1)...Din(an) lies in Ds[X] forms a linear generating set of ks{X}. Define a map ψ: X → kDs[X] by ψ(a) = a for every a in X, and extend ψ to a superalgebra homomorphism eψ: ks{X} → kDs[X]. Then eψ((((...(Di1 (a1) · Di2(a2))· ···) · Din−1(an−1)) · Din(an))) = Di1(a1)...Din (an). Since the set Ds[X] is linearly independent in kDs[X], the homomorphism eψ is an isomor- phism. (cid:3) Thanks to Lemma 3.3, we can identify ks{X} with kDs[X]. 3.2. The linear independence of the set Tabs(X). Our aim in this subsection is to show that the set of all the GDN supertableaux Tabs(X) over X is linearly independent. Our strategy is to construct an GDN superalgebra homomorphism from (GDNs(X), ◦) to (kDs[X], ◦), and show that the image of Tabs(X) is linearly independent in kDs[X], where the operation ◦ is defined in Lemma 3.1. We define an ordering < on Ds[X] as follows: For all u and v be as in Equation (3.1), we define (∗) u < v ⇔ (n, in, an, ..., i1, a1) < (m, jm, bm, , ..., j1, b1) lexicographically, and define the length ℓ(u) of u to be n. For every element f = P16i6n αiui with each αi 6= 0 in k and u1 > u2 > ... > un in Ds[X], we call f := u1 the leading monomial of f , and call lc(f ) := α1 the leading coefficient of f . Now we are ready to show that the set Tabs(X) is linearly independent in GDNs(X). Recall that by Lemma 3.1, (kDs[X], ◦) is a GDN superalgebra. Theorem 3.4. Let ϕ: (GDNs(X), ◦) → (kDs[X], ◦) be a GDN superalgebra homomorphism induced by ϕ(a) = a for every element a in X. Then ϕ is injective. Moreover, the set Tabs(X) of all the GDN supertableaux over X forms a linear basis of the free GDN superalgebra GDNs(X). Proof. We first show that ϕ is injective. Let µ be a GDN supertableau as in Equation (2.1). Then it is easy to see that Therefore, it is straightforward to show that ϕ(µ) = ϕ(a) · D(ϕ(µ1))· ··· ·D(ϕ(µn)). ϕ(µ) = an,2...an,rnan−1,2...an−1,rn−1 ...a1,2...a1,r1aDrn(an,1)...Dr1(a1,1), and lc(ϕ(µ)) = 1 or lc(ϕ(µ)) = −1. ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 13 Therefore, for all GDN supertableaux µ and ν, if µ 6= ν, then we obtain ϕ(µ) 6= ϕ(ν). Suppose that for some pairwise different GDN supertableaux µ1, ..., µn in Tabs(X), for some nonzero elements α1, ..., αn in k, we have P αiµi = 0. Then the equality P ϕ(αiµi) = 0 contradicts to the fact that ϕ(µi) are pairwise different. Therefore, the set Tabs(X) is linearly independent and the homorphism ϕ is injective. In particular, by Lemma 2.10, the (cid:3) set Tabs(X) is a linear basis of GDNs(X). 3.3. A Poincar´e-Birkhoff-Witt type Theorem. We call an associative differential supercommutative algebra B = B0 ⊕ B1 a universal enveloping algebra of a GDN su- peralgebra A = A0 ⊕ A1 if, there is a linear map ψ: A → B satisfying ϕ(Ai) ⊆ Bi (i = 0, 1) and (3.2) ψ(x ◦ y) = ψ(x) · D(ψ(y)) for all x and y in A, and the following holds: for an arbitrary associative differential supercommutative algebra C = C0 ⊕ C1, for every linear map ψ′: A → C satisfying the equation ψ′(x ◦ y) = ψ′(x) · D(ψ′(y)) for all x and y in A, and ψ′(Ai) ⊆ Ci (i = 1, 2), there exists a unique homomorphism of associative differential supercommutative alge- bras ϕ : B → C such that ϕ ◦ ψ = ψ′. It is easy to see that whenever such an universal enveloping algebra B exists, then it is unique up to isomorphism. Let A = A0 ⊕ A1 be a superalgebra and let S be a subset of A. We call S a homogeneous set if S is a subset of A0 ∪ A1. For every homogenous subset S of GDNs(X), the notation GDNs(XS) means the quotient superalgebra GDNs(X)/Id(S), where Id(S) means the ideal of GDNs(X) generated by S. Let ϕ be as that in Theorem 3.4, and denote by IdD[ϕ(S)] the associative differential supercommutative algebra ideal of (kDs[X], ·, D) generated by ϕ(S). By convention, the notation kDs[Xϕ(S)] means the associative differential supercommutative algebra generated by X with the set ϕ(S) of defining re- lations, that is, the quotient superalgebra kDs[X]/IdD[ϕ(S)]. Then it is easy to see that, for every GDN superalgebra GDNs(XS), the associative differential supercommutative algebra kDs[Xϕ(S)] is the universal enveloping algebra of GDNs(XS). Our aim in this subsection is to show that every GDN superalgebra can be embedded into its universal enveloping associative differential supercommutative algebra. We first consider the subalgebra of (kDs[X], ◦) (as GDN superalgebra) generated by X. For every monomial u = Di1(a1)...Din(an) in Ds[X], define the weight wt(u) of u to be (P16j6n ij) − n + 1. Then it is easy to see that wt(u · v) = wt(u) + wt(v) − 1 for all u and v in Ds[X] such that u · v 6= 0. The following lemma offers another linear basis of the free GDN superalgebra generated s [X] of all the monomials of weight 0 in Ds[X]. s [X] be the subspace of kDs[X] spanned by all the monomials of s [X], ◦) is the subalgebra of (kDs[X], ◦) generated by X. s [X] be the GDN superalgebra homomorphism induced by X, that is, the set D0 Lemma 3.5. Let kD0 weight 0 in Ds[X]. Then (kD0 Moreover, let ϕ : GDNs(X) −→ kD0 by ϕ(a) = a for every a in X. Then ϕ is an isomorphism. Proof. We first show that (kD0 for all u and v in D0 s [X], the product u · D(v) lies in kD0 s [X], ◦) is a GDN superalgebra. It is enough to show that, s [X]. Assume that u and v are as 14 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† in Equation 3.1 such that u · Dv 6= 0 and wt(u) = wt(v) = 0. Then by Lemma 3.2 and by the definition of the operation D, we obtain that each monomial in u · D(v) is of weight X it + X jl + 1 − n − m + 1 = ( X it − n + 1) + ( X jl − m + 1) = 0. 16t6n 16l6m 16t6n 16l6m To show that every monomial u of weight 0 lies in the subalgebra of (kDs[X], ◦) generated by X, we use induction on u with respect to the order < defined by (∗). For u = a ∈ X, it is obvious. For u = Dj1(b1)...Djm(bm) in Ds[X] such that (P16l6m jl) − m + 1 = 0, we have j1 = 0, because (j1, b1) 6···6 (jm, bm) lexicographically forces j1 6···6 jm. Therefore, we may assume that u = an,2...an,rnan−1,2...an−1,rn−1...a1,2...a1,r1aDrn(an,1)...Dr1(a1,1) ∈ Ds[X], where 1 6 rn 6···6 r1 and n > 1. Let µ be a GDN supertableau as in Equation (2.1). Then µ lies in the subalgebra of (kDs[X], ◦) generated by X and it is straightforward to show that µ = u. By induction hypothesis, the element u − lc(µ)−1µ lies in the subalgebra of (kDs[X], ◦) generated by X. The first claim of the lemma follows. As for the second claim, notice that by the proof of Theorem 3.4, the homomorphism ϕ (cid:3) is an injection. By the first claim, the homomorphism ϕ is an epimorphism. By Lemma 3.5, we can identify (kD0 indicates some properties inherited from kD0 offers a sufficient condition under which GDNs(X) is nilpotent. s [X], ◦) with (GDNs(X), ◦). This identification s [X]. For instance, the following corollary Corollary 3.6. If X = X1 is a finite set, where each element of X1 is of parity 1, then GDNs(X) is nilpotent. In particular, for every GDN superalgebra A = A0 ⊕ A1, if A is generated by finite elements of A1, then A is nilpotent. Proof. It is enough to show that, there is some positive integer n such that for every GDN supertableau µ, the inequality ℓ(µ) > n implies that µ = 0. Let ϕ be as in Lemma 3.5. For every GDN supertableau µ, we have ϕ(µ) = P16p6q αpup for some nonzero elements αi in k, and for some monomials up in Ds[X] such that wt(up) = 0 and ℓ(up) = ℓ(µ). Say u = up for some integer p 6 q. Then we may assume that u = c1...cmD(b1)...D(bt)Dr1(a1)...Drn(an) for some elements ai, bj, cl in X1 such that 2 6 r1 6···6 rn. Then we have n 6 (r1 − 1) + (r2 − 1)+ ··· +(rn − 1) = m − 1. Therefore, we obtain ℓ(µ) = n + m + t 6 2m + t − 1. So if ℓ(µ) > 3(♯X1), where ♯X1 is the cardinality of X1, then t > (♯X1) or m > (♯X1), both of which imply that ϕ(µ) = 0. (cid:3) Since ϕ is an isomorphism, we get µ = 0. by Id0 by kD0 Let S be a homogeneous subset of GDNs(X) and let ϕ be as that in Lemma 3.5. Denote s [X], ◦) generated by ϕ(S) and denote D[ϕ(S)]. D[ϕ(S)] the GDN superalgebra ideal of (kD0 s [X], ◦) and Id0 s [Xϕ(S)] the quotient of (kD0 s [Xϕ(S)] is isomorphic to GDNs(XS). Therefore, s [Xϕ(S)], ◦) can always be embedded for the embedding, it is enough to prove that (kD0 By Lemma 3.5, it is clear that kD0 ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 15 into (kDs[Xϕ(S)], ◦). We shall first investigate the elements of IdD[ϕ(S)] and investigate those of Id0 D[ϕ(S)]. Since S is homogeneous, it is easy to see that ϕ(S) is also homogeneous, in particular, for every element s in S, the parity ϕ(s) of ϕ(s) is well-defined. Therefore, for every monomial u in Ds[X], we have u · ϕ(s) = (−1)ϕ(s)uϕ(s) · u. By applying right supercom- mutativity, we obtain IdD[ϕ(S)] = spank{u · Dt(ϕ(s)) u ∈ Ds[X], t ∈ Z>0, s ∈ S}, where D0(ϕ(s)) is defined to be ϕ(s). We are now ready to describe the ideal of (kD0 generated by the set ϕ(S). s [X], ◦) Lemma 3.7. Let S be a homogeneous subset of GDNs(X) and let ϕ be as in Lemma 3.5. Suppose that Id0 s [X], ◦) generated by ϕ(S). Then we have (3.3) Id0 D[ϕ(S)] = spank{u · Dt(ϕ(s)) u ∈ Ds[X], t ∈ Z>0, s ∈ S, wt(u · Dt(ϕ(s))) = 0}. D[ϕ(S)] is the ideal of the GDN superalgebra (kD0 Proof. Since S is homogeneous, it is clear that the right part of Equation 3.3 is an ideal including ϕ(S). So to prove the lemma, it is enough to show that u · Dt(ϕ(s)) lies in Id0 D[ϕ(S)] whenever wt(u · Dt(ϕ(s))) = 0. Since every monomial in the expansion of Dt(ϕ(s)) has weight t, we may suppose that u = a1...amb1...btDr1(c1)...Drn(cn) lies in Ds[X] such that m = r1 + · · · + rn − n and rn > rn−1 > . . . > r1 > 1. Then in kD0 s [X], we have u · Dt(ϕ(s)) = αDt(s) · b1...bt · a1...amDr1(c1)...Drn(cn) for some integer α. So the lemma will be clear if we show that the following two claims hold: (i) The polynomial Dt(ϕ(s)) · b1...bt lies in Id0 (ii) The polynomial f ·a1...ar−1 ·Dr(c) lies in Id0 D[ϕ(S)] if s lies in S; D[ϕ(S)] if f is a homogeneous polynomial To prove (i), we use induction on t. For t = 0, we get Dt(ϕ(s))·b1...bt = ϕ(s) ∈ Id0 D[ϕ(S)]. in Id0 D[ϕ(S)]. For t > 0, the polynomial Dt(s) · b1...bt = (−1)b1sb1 · Dt(s) · b2 · · · bt = (−1)b1s(b1 ◦ (Dt−1(s) · b2...bt)) − (−1)b1sb1 · X = (−1)b1s(b1 ◦ (Dt−1(s) · b2...bt)) − X 26i6t (−1)bibi+1···bt((Dt−1(s) · b1...bi−1bi+1...bt) ◦ bi) 26i6t lies in Id0 D[ϕ(S)] by induction hypothesis. To prove (ii), we use induction on r. For r = 1, we obtain f · D(c) = f ◦ c ∈ Id0 D[ϕ(S)]. For r > 1, the polynomial (Dt−1(s) · b2...bi−1 · (Dbi) · bi+1...bt) 16 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† f ·a1...ar−1Dr(c) = (f ◦(a1...ar−1Dr−1(c)))− X = (f ◦(a1...ar−1Dr−1(c)))− X 16i6r−1 (−1)aiai+1...ar−1c((f ·a1...ai−1ai+1...ar−1Dr−1(c))◦ai) (f ·a1...ai−1·Dai·ai+1...ar−1Dr−1(c)) lies in Id0 D[ϕ(S)] by induction hypothesis. 16i6r−1 (cid:3) We then have the following Poincar´e-Birkhoff-Witt type theorem. Theorem 3.8. Every GDN superalgebra GDNs(XS) can be embedded into its universal enveloping associative differential supercommutative algebra kDs[Xϕ(S)], where ϕ : GDNs(X) −→ kDs[X] is the GDN superalgebra homomorphism induced by ϕ(a) = a for every a in X. Proof. By Lemmas 3.5 and 3.7, we obtain GDNs(XS) ∼= kD0 s [Xϕ(S)] = The lemma follows. ∼= (kD0 Id0 s [X], ◦) D[ϕ(S)] kD0 = (kD0 s [X], ◦) IdD[ϕ(S)] ∩ kD0 s [X] s [X] + IdD[ϕ(S)] IdD[ϕ(S)] 6 (kDs[X], ◦) IdD[ϕ(S)] = kDs[Xϕ(S)]. (cid:3) 4. Engel Theorem Our aim in this section is to prove an Engel theorem for GDN superalgebras (Theo- rem 4.4), which is based on what was done for GDN algebras [9]. In this section, we assume that the characteristic char(k) of the field k is 0, and assume that A = A0 ⊕ A1 is a GDN superalgebra. For every x in (A, ◦), let ρx be the right multiplication operator ρx : A −→ A, ρx(y) = (y ◦ x) for every y in A. Then A is called right-nil of bound index if, for some positive integer n, for every x ∈ A, we have ρn−1 (x) = 0. We use the notation xn (x). For all x1, ..., xn in A, define L for ρn−1 x [x1, ..., xn]L = ((...((x1 ◦ x2) ◦ x3)◦ ··· ◦) ◦ xn) (left-normed bracketing). x For all subspace V1, ..., Vn of A, define [V1, ..., Vn]L = spank{[x1, ..., xn]L xi ∈ Vi, 1 6 i 6 n}. In particular, we obtain V n L = [V, ..., V {z } n times ]L and [V1, V2]L = (V1 ◦ V2) = spank{(x1 ◦ x2) x1 ∈ V1, x2 ∈ V2}. We call an algebra A right-nilpotent if An L = 0 for some positive integer n. Finally, for every subspace V of A, for every integer n > 1, define the subspace V n of A inductively as follows: ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 17 We call an algebra A nilpotent if An = 0 for some positive integer n. (i) V 1 = V and V 2 = (V ◦ V); (ii) V n = P16i6n−1(V i ◦ V n−i). Since A0 is an ordinal GDN algebra, by Lemmas 6 and 7 in [9], we get the following lemma, which shows that every right-nil GDN algebra of bound index is right nilpotent. For the convenience of the readers, we quickly repeat the argument. Lemma 4.1. [9] Let A = A0 ⊕ A1 be a GDN superalgebra over a field of characteristic 0. If for some positive integer n, for every x ∈ A0, we have xn L = 0, then (A0)n+1 L = 0. Proof. For all x1, ..., xt in A0, define S(x1, x2, . . . , xt) = X [xσ(1), xσ(2), ..., xσ(t)]L , σ∈St where St is the symmetric group of order t. Then for every term µ occurred in the polynomial (x1 + x2 + · · · + xt)t L − S(x1, x2, . . . , xt), there is some integer i 6 t such that the letter xi does not occur in µ. By the inclusion-exclusion properties, we get (x1 + x2 + · · · + xt)t (−1)t−r+1(xi1 + · · · + xir )t L. L − S(x1, x2, . . . , xt) = X ∅6={i1,i2,···ir}${1,...,t} Therefore, for t > n, we get S(x1, x2, . . . , xt) = 0. Moreover, using right (super)commutativity, it is straightforward to show that S(x1, . . . , xt+1) = t(S(x2, . . . , xt+1) ◦ x1) + t![x1, ..., xt+1]L. Since char(k) = 0, we have [x1, ..., xt+1]L = 0 for every t > n + 1. (cid:3) The following lemma shows that, if a GDN superalgebra A is right nilpotent, then A2 is nilpotent. This result is directly reminiscent of that for GDN algebras [9]. Lemma 4.2. Let A be a GDN superalgebra. Then for every positive integer n, the space An is an ideal of A, and we have (A2)n ⊆ An+1 Proof. We first use induction on n to show that An clear. For n > 2, we have L forms an ideal of A. For n = 1, it is . L L (An L ◦ A) = An+1 L ⊆ An L, and by induction hypothesis, we also have (A ◦ An L) ⊆ ((A ◦ An−1 L ) ◦ A) + (An−1 L ◦ (A ◦ A)) + ((An−1 L ◦ A) ◦ A) ⊆ An L. Now we use induction on n to show (A2)n ⊆ An+1 L . For n = 1, it is clear. For n > 2, we obtain (A2)n = X 16i6n−1 16i6n−1 The proof is completed. ((A2)i ◦ (A2)n−i) ⊆ X ⊆ X [A, An−i+1 L 16i6n−1 ) ◦ An−i+1 L L (Ai+1 ]L ⊆ X 16i6n−1 , A, ..., A {z } i times [An−i+1 L ]L = An+1 L , A, ..., A {z } i times . (cid:3) 18 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† We want to show that, under certain conditions, every right-nil GDN superalgebra of bounded index is right nilpotent. And the main difficulty lies in how to deal with the space [A0, A1, A1, ..., A1]L. The idea is to "split" A1 in the following sense. Lemma 4.3. Let A = A0 ⊕ A1 be a GDN superalgebra generated by X = X0 ∪ X1. Then for every integer q > 1, we have [A0, A1, ..., A1 } q times {z ]L ⊆ , A1, ..., A1 , kX1, ..., kX1 [A0, A0, ..., A0 } } } m times p times t times {z {z {z ]L, X 2t+m+p=q,t,m>0,p∈{1,2} where kX1 is the subspace of A1 spanned by X1. Proof. We use induction on q. For q 6 2, it is clear. For q = 3, we shall show that [A0, A1, A1, A1]L ⊆ [A0, A1, A0]L + [A0, A1, A1, kX1]L. It is enough to show that, for every term µ over X of parity 0, for all terms µ1, µ2 and µ3 over X of parity 1, we have [µ, µ1, µ2, µ3]L ∈ [A0, A1, A0]L + [A0, A1, A1, kX1]L. We use induction on ℓ(µ1). For ℓ(µ1) = 1, the claim follows by right supercommutativity. For ℓ(µ1) > 1, suppose that µ1 = (µ11 ◦ µ12). If µ12 lies in A1, then µ11 lies in A0, and by induction hypothesis, we have [µ, µ12, (µ11 ◦ µ2), µ3]L ∈ [A0, A1, A0]L + [A0, A1, A1, kX1]L. Therefore, we obtain [µ, µ1, µ2, µ3]L = [µ, (µ11 ◦ µ12), µ2, µ3]L = [µ, µ11, µ12, µ2, µ3]L − [µ11, µ, µ12, µ2, µ3]L + [µ11, (µ ◦ µ12), µ2, µ3]L = [µ, µ12, µ2, µ3, µ11]L − [µ11, µ12, µ2, µ3, µ]L + [µ11, (µ ◦ µ12), µ2, µ3]L = [µ, µ12, µ2, µ3, µ11]L − [µ11, µ12, µ2, µ3, µ]L + [µ11, ((µ ◦ µ12) ◦ µ2), µ3]L + [(µ ◦ µ12), µ11, µ2, µ3]L − [(µ ◦ µ12), (µ11 ◦ µ2), µ3]L = [((µ ◦ µ12) ◦ µ2), µ3, µ11]L − [((µ11 ◦ µ12) ◦ µ2), µ3, µ]L + [µ11, µ3, ((µ ◦ µ12) ◦ µ2)]L + [((µ ◦ µ12) ◦ µ2), µ3, µ11]L − [µ, µ12, (µ11 ◦ µ2), µ3]L ∈ [A0, A1, A0]L + [A0, A1, A1, kX1]L. If µ12 lies in A0, then µ11 lies in A1, and we obtain [µ, µ1, µ2, µ3]L = [µ, (µ11 ◦ µ12), µ2, µ3]L = [µ, µ11, µ12, µ2, µ3]L − [µ11, µ, µ12, µ2, µ3]L + [µ11, (µ ◦ µ12), µ2, µ3]L = [((µ ◦ µ11) ◦ µ2), µ3, µ12]L + [(µ11 ◦ µ2), µ3, (µ ◦ µ12)]L − [((µ11 ◦ µ) ◦ µ2), µ3, µ12]L. So [µ, µ1, µ2, µ3]L lies in [A0, A1, A0]L. For q > 4, by right supercommutativity and the case q = 3, we have ]L ⊆ [A0, A1, ..., A1, [A0, A1, ..., A1 } } q − 2 times q times {z {z A0]L + [A0, A1, ..., A1, } q − 1 times {z kX1]L ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 19 ⊆ + X X 2t+m+p=q−2,t,m>0,p∈{1,2} 2t+m+p=q−1,t,m>0,p∈{1,2} ⊆ The claim follows. {z {z t times p times m times , A1, A1, kX1, ..., kX1 [A0, A0, ..., A0, {z } } } , A1, A1 kX1, ..., kX1 [A0, A0, ..., A0, {z } } } [A0, A0, ..., A0, kX1, ..., kX1, } } m times m times p times X t times t times {z {z {z {z A0]L , kX1]L 2t+m+p=q,t,m>0,p∈{1,2} ]L. A1, A1 {z } p times (cid:3) We are now in a position to prove the following Engel theorem. Theorem 4.4. Let A = A0 ⊕ A1 be a GDN superalgebra generated by X = X0 ∪ X1 over a field of characteristic 0, where every element of the set X0 is of parity 0 and every element of the set X1 is of parity 1. If X1 is a finite set and the even part A0 is right-nil of bounded index n, then A is right nilpotent, in particular, the ideal A2 of A is nilpotent. Proof. By Lemma 4.1, we have (A0)n+1 We shall show that Aq L = 0. Let n0 = max(♯(X1), n+1) and let q = 3n0 + 1. L = 0. It is enough to show the following two claims: [A0, A1, ..., A1 } j times {z and [A1, A1, ..., A1 } j times {z , A0, ..., A0 } q − 2 − j times {z , A0, ..., A0 } q − 1 − j times {z ]L = 0 for every integer j such that 0 6 j 6 q − 2, ]L = 0 for every integer j such that 0 6 j 6 q − 1. For the first claim, if j = 0, then by Lemma 4.1, we get (A0)q−1 L = 0. For every integer j such that 1 6 j 6 q − 2, by Lemma 4.3, we have [A0, A1, ..., A1 , A0, ..., A0 } } q − 2 − j times j times ]L {z X {z [A0, ⊆ 2t+m+p=j,t,m>0,p∈{1,2} If m > n0, then we obtain A0, ..., A0, } {z t + q − 2 − j times kX1, ..., kX1, } m times {z ]L. A1, A1 {z } p times [A0, {z If m 6 n0, then we obtain t = 1 2 (j − m − p) and t + q − 2 − j times A0, ..., A0, } kX1, ..., kX1, } m times {z ]L = 0. A1, A1 {z } p times t + q − 2 − j = 1 2 (j − m − p) + q − j − 2 = 1 2 (−m − p + q) + 1 2 (q − j) − 2 > 1 2 (−n0 − 2 + 3n0 + 1) + 1 2 × 2 − 2 = n0 − 3 2 . 20 ZERUI ZHANG∗, YUQUN CHEN♯, AND LEONID A. BOKUT† Since t + q − 2 − j is an integer, we have t + q − 2 − j > n0 − 1. So we get [A0, A0, ..., A0, } {z t + q − 2 − j times kX1, ..., kX1, } m times {z ]L = 0. A1, A1 {z } p times The first claim follows. For the second claim, if j > 1, then by the first claim, we get [A1, A1, ..., A1 } j times {z , A0, ..., A0 } q − 1 − j times {z ]L j [A0, A1, ..., A1, } j − 1 times {z A0, ..., A0 } {z q − 2 − (j − 1) times ]L = 0. If j = 0, then we first use induction on q to show that, for every q > 3, we have [A1, A0, · · · , A0 } q − 1 times {z ]L ⊆ (A1 ◦ (A0)q−1 L ) + ((A0)q−1 L ◦ A1) + ((A0)q−2 L ◦ (A1 ◦ A0)). For q = 3, by left supersymmetry and right supercommutativity, we get [A1, A0, A0]L ⊆ (A1 ◦ (A0 ◦ A0)) + ((A0 ◦ A1) ◦ A0) + (A0 ◦ (A1 ◦ A0)) ⊆ (A1 ◦ (A0 ◦ A0)) + ((A0 ◦ A0) ◦ A1) + (A0 ◦ (A1 ◦ A0)). For q > 3, by induction hypothesis, we get [A1, A0, · · · , A0 } q − 1 times ⊆ ((A1 ◦ (A0)q−2 ]L = ([A1, A0, · · · , A0 } ) ◦ A0) + (((A0)q−2 q − 2 times {z {z L L ]L ◦ A0) ◦ A1) ◦ A0) + (((A0)q−3 L ◦ (A1 ◦ A0)) ◦ A0) ⊆ (A1 ◦ ((A0)q−2 L ◦ A0)) + (((A0)q−2 ⊆ (A1 ◦ (A0)q−1 ◦ A1) ◦ A0) + ((A0)q−2 ) + ((A0)q−1 ◦ A1) + ((A0)q−2 L L L L L ◦ (A1 ◦ A0)) ◦ (A1 ◦ A0)). (cid:3) Finally, since (A0)q−2 L = 0, the second claim follows. References [1] C. Bai and D. Meng, The classification of Novikov algebras in low dimensions, J. Phys. A: Math. Gen. 34(8) (2001) 1581 -- 1594. [2] C. Bai and D. Meng, Transitive Novikov algebras on four-dimensional nilpotent Lie algebras, International Journal of Theoretical Physics 40(10) (2001) 1761 -- 1768. [3] A.A. Balinskii and S.P. Novikov, Poisson brackets of hydrodynamics type, Frobenius algebras and Lie algebras (Russian), Dokl. Akad. Nauk SSSR 283(5) (1985) 1036 -- 1039. [4] L.A. Bokut, Y.Q. Chen and Z. Zhang, Grobner -- Shirshov bases method for Gelfand -- Dorfman -- Novikov algebras, Journal of Algebra and Its Applications, 16(1) (2017) 1750001 (22 pages), DOI: 10.1142/S0219498817500013 [5] L.A. Bokut, Y.Q. Chen and Z. Zhang, On free Gelfand -- Dorfman -- Novikov -- Poisson algebras and a PBW theorem, Journal of Algebra 500 (2018) 153 -- 170. [6] D. Burde and K. Dekimpe, Novikov structures on solvable Lie algebras, Journal of Geometry and Physics 56(9) (2006) 1837-1855. ON FREE GELFAND -- DORFMAN -- NOVIKOV SUPERALGEBRAS AND A PBW TYPE THEOREM‡ 21 [7] L. Chen, Y. Niu and D. Meng, Two kinds of Novikov algebras and their realizations, Journal of Pure and Applied Algebra 212(4) (2008) 902 -- 909. [8] A.S. Dzhumadildaev and C. Lofwall, Trees, free right-symmetric algebras, free Novikov algebras and identities, Homology, Homotopy and Applications 4(2) (2002) 165 -- 190. [9] A.S. Dzhumadildaev and K.M. Tulenbaev, Engel theorem for Novikov algebras, Communications in Algebra 34(3) (2006) 883 -- 888. [10] V.T. Filippov, On right-symmetric and Novikov nil algebras of bounded index, (Russian. Russian summary) Mat. Zametki 70(2) (2001) 289 -- 295; translation in Math. Notes 70(2) (2001) 258-263. [11] I.M. Gelfand and I.Ya. Dorfman, Hamiltonian operators and algebraic structures related to them, Functional Analysis and Its Applications 13(4) (1979) 248 -- 262. [12] Y. Kang and Z. Chen, Novikov superalgebras in low dimensions, Journal of Nonlinear Mathematical Physics 16(3) (2009) 251 -- 257. [13] S.P. Novikov, The geometry of conservative systems of hydrodynamic type. The method of averaging for field-theoretical systems , Uspekhi Mat. Nauk 40(4)(244) (1985) 79 -- 89; Russian Math. Surveys, 40(4) (1985) 85 -- 98. [14] J.M. Osborn and E.I. Zelmanov, Nonassociative algebras related to Hamiltonian operators in the formal calculus of variations, Journal of Pure and Applied Algebra 101(3) (1995) 335 -- 352. [15] X. Xu, On simple Novikov algebras and their irreducible modules, Journal of Algebra 185(3) (1996) 905 -- 934. [16] X. Xu, Quadratic Conformal Superalgebras, Journal of Algebra 231(1) (2000) 1 -- 38. [17] X. Xu, Variational calculus of supervariables and related algebraic structures, Journal of Algebra 223(2) (2000) 396 -- 437. [18] E.I. Zelmanov, On a class of local translation invariant Lie algebras, Soviet Math. Dokl. 35(1) (1987) 216 -- 218. [19] F. Zhu and Z. Chen, Novikov superalgebras with A0 = A1A1, Czechoslovak Mathe- matical Journal 60(4) (2010) 903 -- 907. Z.Z., School of Mathematical Sciences, South China Normal University, Guangzhou 510631, P. R. China E-mail address: [email protected] Y.C., School of Mathematical Sciences, South China Normal University, Guangzhou 510631, P. R. China E-mail address: [email protected] L.A.B., School of Mathematical Sciences, South China Normal University Guangzhou 510631, P. R. China; Sobolev Institute of mathematics, Novosibirsk, 630090, Russia; Novosi- birsk State University, Novosibirsk 630090, Russia E-mail address: [email protected]
1907.00803	1	1907	2019-06-28T15:03:16	Classification of 3-Dimensional BiHom-Associative and BiHom-Bialgebras	[ "math.RA", "math.AC" ]	The purpose of this paper is to study the structure and the algebraic varieties of BiHom-associative algebras. We provide a classication of n-dimensional BiHom-associative and BiHom-bialgebras and BiHom Hopf algebras for n $\le$ 3.	math.RA	math	CLASSIFICATION OF 3-DIMENSIONAL BIHOM-ASSOCIATIVE AND BIHOM-BIALGEBRAS. AHMED ZAHARI Abstract. The purpose of this paper is to study the structure and the algebraic varieties of BiHom-associative algebras. We provide a classification of n-dimensional BiHom-associative and BiHom-bialgebras and BiHom Hopf algebras for n ≤ 3. 9 1 0 2 n u J 8 2 ] . A R h t a m [ 1 v 3 0 8 0 0 . 7 0 9 1 : v i X r a Introduction The first motivation to study nonassociative BiHom-algebras comes from quasi-deformations of Lie algebras of vector fields, in particular q-deformations of Witt and Virasoro algebras. It was observed in the pioneering works, mainly by physicists, that in these examples a twisted Jacobi identity holds. Motivated by these examples and their generalization on the one hand, and the desire to be able to treat within the same framework such well-known generalizations of Lie algebras as the color and Lie superalgebras on the other hand, quasi-Lie algebras and subclasses of quasi-hom- Lie algebras and hom-Lie algebras were introduced by Hartwig, Larsson and Silvestrov in [5, 6]. The BiHom-associative algebras play the role of associative algebras in the BiHom-Lie setting. Usual functors between the categories of Lie algebras and associative algebras were extended to Hom-setting, see [10] for the construction of the enveloping algebra of a BiHom-Lie algebra. A Hom-associative algebra (A, µ, α, β) consists of a vector space, a multiplication and two linear endomorphism. It may be viewed as a deformation of an associative algebras, in which the associa- tivity condition is twisted by the linear maps α and β in a certain way such that when α = id and β = id, the BiHom-associative algebras degenerate to exactly associative algebras. In this paper we aim to study the structure of BiHom-associative algebras. Let A be an n-dimensional K-linear space and {e1, e2, · · · , en} be a basis of A. A BiHom-algebra structure on A with product µ is ij, were µ(ei, ej) = Pn determined by n3 structure constants C k ijek and by α and β which is given by 2n2 structure constants aij and bij, where α(ei) = Pn k=1 bkjek. Requiring the algebra structure to be BiHom-associative and unital gives rise to sub-variety Hn (resp. Hun) of kn3+2n2 k=1 C k j=1 ajiej and β(ej ) = Pn . Base changes in A result in the natural transport of the structure action of GLn(k) on Hn. Thus the isomorphism classes of n-dimensional BiHom-algebras are in one-to-one correspondence with the orbits of the action of GLn(k) on Hn (rep. Hun). Key words and phrases. BiHom-associative algebra, BiHom-Bialgebra, BiHom-Hopf algebra, Classification. 1 2 AHMED ZAHARI Furthermore, we shall consider the class of BiHom-bialgebras which are Hom-associative algebras equipped with a compatible BiHomCoalgebra structure, in particular BiHomHopf algebras where the additional structures like the comultiplication ∆ and the counit ǫ can be expressed in a base in a similar way as the multiplication above. We shall also give a classification of these algebras up to isomorphism in low dimension n ≤ 3. The paper is organized as follows. In the first section we give the basics about BiHom-associative algebras and provide some new properties. Moreover, we discuss unital BiHom-associative algebras. In Section 2 is dedicated to describe algebraic varieties of BiHom-associative algebras and provide classifications, up to isomorphism, of 2-dimensional (resp. 3-dimensional) BiHom-associative alge- bras. Moreover, in Section 3 we shall recall the definitions of BiHom-bialgebras and BiHomHopf algebras and present the classification up to dimension 3. Acknowledgements. I'd like to thank my thesis supervisor Abdenacer Makhlouf and Martin Bordomann for giving me the problem, for many fruitful discussions and for their constant support. 1. Structure of BiHom-associative algebras Let K be an algebraically closed field of characteristic 0, A be a linear space over K. We refer to a Hom-algebra by a 4-tuple (A, µ, α, β), where µ : A × A → A is a bilinear map (multiplication) and α and β be two homomorphisms of A (twist map). 1.1. Definitions. Definition 1.1. [9]. A BiHom-associative algebra is a 4-tuple (A, µ, α, β) consisting of a linear space A, a bilinear map µ : A × A → A and two linear space homomorphism α : A → A and β : A → A satisfying (1.1) (1.2) (1.3) µ(α(x), µ(y, z)) = µ(µ(x, y), β(z)). α(µ(x, y)) = µ(α(x), α(y)) and β(µ(x, y)) = µ(β(x), β(y)). α ◦ β = β ◦ α. Usually such BiHom-associative algebras are called multiplicative. Since we are dealing only with multiplicative BiHom-associative algebras, we shall call them BiHom-associative algebras for sim- plicity. We denote the set of all BiHom-associative algebras by H. In the language of Hopf algebras, the multiplication of a BiHom-associative algebra over A consists of a linear map µ : A ⊗ A → A, and Condition (1.1) can be written as (1.4) µ(α(x) ⊗ µ(y ⊗ z)) = µ(µ(x ⊗ y) ⊗ β(z)). 3 Definition 1.2. Let (A, µA, αA, βA) and (B, µB , αB, βB) be two BiHom-associative algebras. A linear map ϕ : A → B is called a BiHom-associative algebras morphism if (1.5) ϕ ◦ µA = µB(ϕ ⊗ ϕ), αB ◦ ϕ = ϕ ◦ αA and βB ◦ ϕ = ϕ ◦ βA. In particular, BiHom-associative algebras (A, µA, αA, βA) and (B, µB , αB, βB) are isomorphic if ϕ is also bijective. Definition 1.3. A unital BiHom-associative algebra (A, µ, α, β, u) is called unital if there exists an element uA ∈ A (called a unit) such that α(uA) = uA, β(uA) = uA and auA = α(a) and uAa = β(a), ∀a ∈ A. A morphism of unital BiHom-associative algebras φ : A −→ B is called unital if φ(uA) = uB. 1.2. Structure of BiHom-associative algebras. We state in this section some properties on the structure of BiHom-associative algebras which are not necessarily multiplicative. Proposition 1.4 ([11]). Let (A, µ, α, β) be a BiHom-associative algebra and γ : A → A be a BiHom-associative algebra morphism. Then (A, γµ, γα, γβ) is a BiHom-associative algebra. Definition 1.5. Let (A, µ, α, β) be a BiHom-associative algebra. If there is an associative algebra (A, µ′) such that µ(x, y) = (α ⊗ β)µ′(x, y), ∀x, y ∈ A, we say that (A, µ, α, β) is of associative type and (A, µ′) is its compatible associative algebra or the untwist of (A, µ, α, β). Proposition 1.6. Let (A, µ, α, β) be an n-dimensional BiHom-associative algebra and φ : A → A be an invertible linear map. Then there is an isomorphism with an n-dimensional BiHom-associative algebra (A, µ′, φαφ−1, φβφ−1) where µ′ = φ◦µ◦(φ−1⊗φ−1). Furthermore, if (cid:8)C k ij(cid:9) are the structure constants of µ with respect to the basis {e1, . . . , en}, then µ′ has the same structure constants with respect to the basis {φ(e1), . . . , φ(en)}. Proof. We prove for any invertible linear map φ : A → A, (A, µ′, φαφ−1, φβφ−1) is a BiHom- associative algebra. µ′(µ′(x, y), φβφ−1(z)) = φµ(φ−1 ⊗ φ−1)(φµ(φ−1 ⊗ φ−1)(x, y), φβφ−1(z)) = φµ(µ(φ−1(x), φ−1(y)), βφ−1(z)) = φµ(αφ−1(x), µ(φ−1(y), φ−1(z))) = φµ(φ−1 ⊗ φ−1)(φ ⊗ φ)(αφ−1(x), µ(φ−1 ⊗ φ−1)(y, z))) = φµ(φ−1 ⊗ φ−1)(φαφ1(x), φµ(φ−1 ⊗ φ−1)(y, z))) = µ′(φαφ−1(x), µ′(y, z)). So (A, µ′, φαφ−1, φβφ−1) is a BiHom-associative algebra. It is also multiplicative. Indeed, for α φαφ−1µ′(x, y) = φαφ−1φµ(φ−1 ⊗ φ−1)(x, y) = φαµ(φ−1 ⊗ φ−1)(x, y) = φµ(αφ−1(x), αφ−1(y)) = φµ(φ−1 ⊗ φ−1)(φ ⊗ φ)(αφ−1(x), αφ−1(y)) = µ′(φαφ−1(x), φαφ−1(y)). 4 AHMED ZAHARI We have also for β φβφ−1µ′(x, y) = φβφ−1φµ(φ−1 ⊗ φ−1)(x, y) = φβµ(φ−1 ⊗ φ−1)(x, y) = φµ(βφ−1(x), βφ−1(y)) = φµ(φ−1 ⊗ φ−1)(φ ⊗ φ)(βφ−1(x), βφ−1(y)) = µ′(φβφ−1(x), φβφ−1(y)). Therefore φ : (A, µ, α, β) → (A, µ′, φαφ−1, φβφ−1) is a BiHom-associative algebras morphism, since φ ◦ µ = φ ◦ µ ◦ (φ−1 ⊗ φ−1) ◦ (φ ⊗ φ) = µ′ ◦ (φ ⊗ φ) and (φαφ−1) ◦ φ = φ ◦ α and (φβφ−1) ◦ φ = φ ◦ β. It is easy to see that {φ(ei), · · · , φ(en)} is a basis of A. For i, j = 1, · · · , n, we have µ2(φ(ei), φ(ej)) = φµ1(φ−1(ei), φ−1(ej)) = φµ(ei, ej) = Pn k=1 C k ijφ(ek). (cid:3) Remark 1.7. A BiHom-associative algebra (A, µ, α, β) is isomorphic to an associative algebra if and only if α = β = id. Indeed, φ ◦ α ◦ φ−1 = φ ◦ β ◦ φ−1 = id is equivalent to α = β = id. Remark 1.8. Proposition 1.6 is useful to make a classification of BiHom-associative algebras. Indeed, we have to consider the class of morphisms which are conjugate. Representations of these classes are given by Jordan forms of the matrix. Any n × n matrix over K is equivalent up to basis change to Jordan's canonical form, then we choose φ such that the matrix of φαφ−1 = γ and λ = φβφ−1, where γ and λ are Jordan canonical forms. Hence, to obtain the classification, we consider only Jordan forms for the structure map of Hom- associative algebras. Proposition 1.9. Let (A, µ, α, β) be a Hom-associative algebra over K. Let (A, µ′, φαφ−1, φβφ−1) be its isomorphic Hom-associative algebra described in Proposition 1.6. If ψ is an automorphism of (A, µ, α, β), then φψφ−1 is an automorphism of (A, µ, φαφ−1, φβφ−1). Proof. Note that γ = φαφ−1. We have φψφ−1γ = φψφ−1φαφ−1 = φψαφ−1 = φαψφ−1 = φαφ−1φψφ−1 = γφψφ−1 . For β, we pose λ = φβφ−1, we have φψφ−1λ = φψφ−1φβφ−1 = φψβφ−1 = φβψφ−1 = φβφ−1φψφ−1 = γφψφ−1. For any x, y ∈ A, φψφ−1µ′(φ(x), φ(y)) = φψφ−1φµ(x, y) = φψµ(x, y) = φµ(ψ(x), ψ(y)) = µ′(φψ(x), φψ(y)) = µ′(φψφ−1(φ(x)), φψφ−1(φ(y))). By Definition, φψφ−1 is an automorphism of (A, µ′, φαφ−1, φβφ−1). (cid:3) The following characterization was given for Hom-Lie algebras in [12]. Proposition 1.10. Given two BiHom-associative algebras (A, µA, αA, βA) and (B, µB , αB, βB) over field K, there is a BiHom-associative algebra (A ⊕ B, µA⊕B, αA + αB, βA + βB), where µA⊕B 5 the usual multiplication (a + b)(a′ + b′) = (a, a′) + (b, b′). µA⊕B(., .) : (A ⊕ B) × (A ⊕ B) → (A ⊕ B) is given by µA⊕B(a + b, a′ + b′) = (µA(a, a′), µB(b, b′)), ∀ a, a′ ∈ A, ∀ b, b′ ∈ B, and the linear map (αA + αB, βA + βB) : A ⊕ B → A ⊕ B is given by (αA + αB, βA + βB)(a, b) = ((αA + βA)a, (αB + βB)b) ∀a ∈ A, b ∈ B. Proof. For any a, a′, a′′ ∈ A, b, b′, b′′ ∈ B, by direct computation, we get µA⊕B((αA + βA, αB + βB, )(a, b), µA⊕B(a′ + b′, a′′ + b′′)) = = µA⊕B(((αA + βA)a, (αB + βB)b), (µA(a′, a′′), µB(b′, b′′))) = (µA(((αA + βA)a), µA(a′, a′′)), µB((αB + βB)b, µB(b′, b′′))) = (µA(µA(a, a′), (αA + βA)a′′), µB(µB(b, b′), (α + β)b)) = (µA⊕B(µA⊕B(a + b, a′ + b′), (αA + βA, αB + βB)(a′′, b′′))). This ends the proof. (cid:3) A morphism of BiHom-associative algebras φ : (A, µA, αA, βA) → (B, µB, αB, βB) is a linear map φ : A → B such that φ◦ µA(a, b) = µB ◦ (φ(a), φ(b)), ∀a, b ∈ A, αB ◦ φ = φ◦ αA and βB ◦ φ = φ◦ βB Denote by ξφ ⊂ A ⊕ B, the graph of the linear map φ : A → B. Proposition 1.11. A linear map φ : (A, µA, αA, βA) → (B, µB, αB, βB) is a BiHom-associative algebras morphism if and only if the graph ξφ ⊂ A ⊕ B is a BiHom-associative sub-algebras of (A ⊕ B, µA⊕B, αA + αB, βA + βB). Proof. Let φ : (A, µA, αA, βA) → (B, µB , αB, βB) be a BiHom-associative algebra morphism. Then for any a, b ∈ A, we have µA⊕B((a, φ(a)), (b, φ(b)) = (µA(a, b), µB(φ(a), φ(b))) = (µA(a, b), φµA(a, b)). Thus the graph ξφ is closed under the product µA⊕B. Furthermore, since αB ◦ φ = φ ◦ αA and βB ◦ φ = φ ◦ βA we have (αA + αB, βA + βB)(a, φ(a)) = ((αA + βA)(a), (αB + βB) ◦ φ(a)) = ((αA + βA)(a), φ ◦ (αA + βA(a))), which implies that (αA + αB, βA + βB) ⊂ ξφ. Thus ξφ is a BiHom-associative sub-algebra of (A ⊕ B, µA⊕B, αA + αB, βA + βB). 6 AHMED ZAHARI Conversely, if the graph ξφ ⊂ A ⊕ B is a Hom-associative sub-algebra of (A ⊕ B, µA⊕B, αA + αB, βA + βB), then we have µA⊕B((a, φ(a)), (b, φ(b)) = (µA(a, b), µB(φ(a), φ(b)) ∈ ξφ, which implies that µB(φ(a), φ(b)) = φ ◦ µA(a, b). Furthermore, (αA + βA, αB + βB)(ξφ) ⊂ ξφ yields that (αA + αB, βA + βB)(a, φ(a)) = ((αA + βA)a, φ ◦ (αA + βA)a) ∈ ξφ, which is equivalent to the condition αB ◦ φ(a) = φ ◦ αA(a) and βB ◦ φ(a) = φ ◦ βA(a). Therefore, φ is a BiHom-associative algebra morphism. (cid:3) 1.3. Unital BiHom-associative algebras. In this section we discuss unital BiHom-associative algebras. We denote by Hun the set of n- dimensional unital BiHom-associative algebras. Definition 1.12. A BiHom-associative algebra (A, µ, α, β) is called unital if there exists an element u ∈ A such that µ(x, u) = α(x) and µ(u, x) = β(x) for all x ∈ A. Proposition 1.13. Let (A, µ, α, β) be a BiHom-associative algebra. We set A = span(A, u) the vector space generated by elements of A and u. Assume µ(x, u) = α(x), and µ(u, x) = β(x) ∀ x ∈ A, α(u) = u and β(u) = u. Then ( A, µ, α, β, u) is a unital Hom-associative algebra. Proof. It is straightforward to check the Hom-associativity. For example µ(µ(x, y), β(u)) = µ(µ(x, y), u) = α(µ(x, y)) = µ(α(x), α(y)) = µ(α(x), µ(y, u)). (cid:3) Remark 1.14. Some unital BiHom-associative cannot be obtained as an extension of a non unital BiHom-associative algebra. Remark 1.15. Let (A, µ, α, β, u) be an n-dimensional unital BiHom-associative algebra and φ : A → A be an invertible linear map such that φ(u) = u. Then it is isomorphic to a n-dimensional BiHom- associative algebra (A, µ′, φαφ−1, φβφ−1, u) where µ′ = φ ◦ µ ◦ (φ−1 ⊗ φ−1). Moreover, if (cid:8)C k ij(cid:9) are the structure constants of µ with respect to the basis {e1, . . . , en} with e1 = u being the unit, then µ′ has the same structure constants with respect to the basis {φ(e1), . . . , φ(en)} with u the unit element. Indeed, we use Proposition 1.6 and Definition 1.3. The unit is conserved since µ′(x, e1) = φ◦µ(φ−1(x), φ−1(e1)) = φ◦α◦φ−1(x) and µ′(e1, x) = φ◦µ(φ−1(e1), φ−1(x)) = φ◦β◦φ−1(x) Proposition 1.16. Let (A, µA, αA, βA, uA) and (B, µB, αB , βB, uB) be two unital BiHom-associative algebras with φ(uA) = uB. Suppose there exists a BiHom-associative algebra morphism φ : A → B. If (A, µ′ A) is an untwist of (A, µA, αA, βA, uA) then there exists an untwist of (B, µB , αB, βB, uB) A, u′ such that φ : (A, µ′ A, u′ A) → (B, µ′ B, u′ B) is an algebra morphism. Proof. Because φ is a homomorphism from (A, µA, αA, βA, uA) to (B, µB , αB, βB, uB). Then αBφ = φαA, and βB ◦ φ = φ ◦ βA for all x ∈ A, we have µB(φ(x), φ(uA)) = µB(φ(x), uB) = αB ◦ φ(x) and µB(φ(uA), φ(x)) = µB(uB, φ(x)) = βB ◦ φ(x). We have also φ ◦ µA(x, uA) = φ ◦ αA(x) and φ ◦ µA(uA, x) = φ ◦ βA(x). By Proposition 1.13, we can see that (AB , µB, uB) is also an associative algebra. Furthermore 7 µ′ B(φ(x), φ(uA)) = µ′ B(φ(x), uB) = φ ◦ α′ A ◦ φ(x) = φ ◦ αA ◦ µA(x, uA) = αB ◦ φ ◦ µA(x, uA) = αB ◦ µB(φ(x), uB) and B(φ(uA), φ(x)) = µ′ µ′ B(uB, φ(x)) = φ ◦ β′ A ◦ φ(x) = φ ◦ αA ◦ µA(uA, x) = βB ◦ φ ◦ µA(uA, x) = βB ◦ µB(uB, φ(x)). (cid:3) 2. Algebraic varieties of BiHom-associative algebras and Classifications In this section, we deal with Algebraic varieties of BiHom-associative algebras with a fixed dimension. A BiHom-associative algebra is identified with its structures constants with respect to a fixed basis. Their set corresponds to an algebraic variety where the ideal is generated by polynomials corresponding to the BiHom-associativity condition. 2.1. Algebraic varieties Hn and their action of linear group. . Let A be a n-dimensional K-linear space and {e1, · · · , en} be a basis of A. A BiHom-algebra struc- k=1 C k ture on A with product µ and a structure map α and β is determined by n3 structure constants Ck ij where µ(ei, ej) = Pn and β(ej ) = Pn limits the set of structure constants (C k Kn3+2n2 j=1 ajiej k=1 bkj ek. If we require this algebra structure to be BiHom-associative, then this ij , aij, bjk) to a cubic sub-variety of the affine algebraic variety ij ek and by 2n2 structure constants aji and bkj, where α(ei) = Pn defined by the following polynomial equations system : (2.1)   ijCs lm = 0, Pn l=1 Pn Pn p=1 aspC p Pn p=1 brpC p Pn lm − bmkC l jkC s m=1 ailCm ij − Pn p=1 Pn jk − Pn p=1 Pn pq = 0, j=1 bjiakj − Pn j=1 ajibkjbqk == 0, q=1 apiaqjC s q=1 bpjbqkC s pq = 0, ∀ i, j, k, s ∈ {1, . . . , n} , ∀ i, j, s ∈ {1, . . . , n} , ∀ j, k, r ∈ {1, . . . , n} , ∀ j, k ∈ {1, . . . , n} . Moreover if µ is commutative, we have C k ij = C k ji i, j, k = 1, · · · , n. The first set of equation cor- respond to the BiHom-associative condition µ(α(ei), µ(ej, ek)) = µ(µ(ei, ej), β(ek)) and the second set to multiplicativity condition α ◦ µ(ei, ej) = µ(α(ei), α(ej )) and β ◦ µ(ej, ek) = µ(β(ej ), β(ek)). We denote by Hn the set of all n-dimensional multiplicative Hom-associative algebras. The group GLn(K) acts on the algebraic varieties of BiHom-structures by the so-called transport of structure action defined as follows. Let A = (A, µ, α, β) be a n-dimensional BiHom-associative algebra defined by multiplication µ and a linear map α and β. Given f ∈ GLn(K), the action f · A 8 AHMED ZAHARI transports the structure, Θ : GLn(K) × Hn −→ Hn (f, (A, µ, α, β)) 7−→ (A, f −1 ◦ µ ◦ (f ⊗ f ), f ◦ α ◦ f −1, f ◦ β ◦ f −1) defined for x, y ∈ A by (2.2) f · µ(x, y) = f −1µ(f (x), f (y)), f · α(x) = f −1α(f (x)), f · β(x) = f −1β(f (x)). The conjugate class is given by Θ(f, (A, µ, α, β)) = (A, f −1 ◦ µ ◦ (f ⊗ f ), f ◦ α ◦ f −1, f ◦ β ◦ f −1)) for f ∈ GLn(K). The orbit of a Hom-associative algebra A of Hasn is given by ϑ(A) = {A′ = f · A, f ∈ GLn(K)} . The orbits are in 1-1 correspondence with the isomorphism classes of n-dimensional BiHom- associative algebras. The stabilizer is Stab((A, µ, α, β)) = (cid:8)f ∈ GLn(K)(f −1 ◦ µ ◦ (f ⊗ f ) = µ and f ◦ α = α ◦ f and f ◦ β = β ◦ f(cid:9) . We characterize -- in terms of structure constants -- the fact that two BiHom-associative algebras are in the same orbit (or isomorphic). Let (A, µA, αA, βA) and (B, µB, αB, βB) be two n-dimensional BiHom-associative algebras. They are isomorphic if there exists ϕ ∈ GLn(K) such that (2.3) ϕ ◦ µA = µB(ϕ ⊗ ϕ), αB ◦ ϕ = ϕ ◦ αA and βB ◦ ϕ = ϕ ◦ βA. Remark 2.1. Conditions (2.3) are equivalent to µA = ϕ−1 ◦ µB ◦ ϕ ⊗ ϕ, αA = ϕ−1 ◦ αB ◦ ϕ and βA = ϕ−1 ◦ βB ◦ ϕ. We set with respect to a basis {ei}i=1,··· ,n: ϕ(ei) = Pn µA(ei, ej) = Pn p=1 dpiep, α(ei) = Pn j=1 ajiej, ijek, µB(ei, ej) = Pn k=1 C k β(ei) = Pn C k ijek, k=1 late to the following system k=1 bkiek, i ∈ {1, . . . , n} i, j ∈ {1, . . . , n} . Conditions (2.3) trans- (2.4)   Pn k=1 Pn ijdpk − Pn k=1 C k Pn p=1(dpiaqp − apidpq), Pn p=1(dpibqp − bpidqp), q=1 dkidqj C s kq, ∀ i, q ∈ {1, . . . , n} , ∀ i, j, p ∈ {1, . . . , n} , ∀ i, q ∈ {1, . . . , n} . We shall check whether the previous are isomorphic. In particular, we shall provide all 2- dimensional BiHom-associative algebras, corresponding to solutions of the system (3.6). To this end, we use a computer algebra system. 2.2. Classification in low dimensions. Theorem 2.2. Every 2-dimensional multiplicative BiHom-associative algebra is isomorphic to one of the following pairwise non-isomorphic BiHom-associative algebras (A, ∗, α, β) appearing in Table 1, where ∗ is the multiplication and α and β the structure maps. We set {e1, e2} to be a basis of 9 K2. Table 1 H2 1 : e1 ∗ e1 = e2, e1 ∗ e2 = e2, e2 ∗ e1 = −e1, e2 ∗ e2 = e2, α(e1) = e1, α(e2) = e2, β(e1) = −e1, β(e2) = e2; H2 2 : e1 ∗ e1 = e2, e1 ∗ e2 = e1, e2 ∗ e1 = −e1, e2 ∗ e2 = −e2, α(e1) = −e1, α(e2) = e2, β(e1) = e1, β(e2) = e2; H2 3 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e2 ∗ e1 = e2, e2 ∗ e2 = e2, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e2; H2 4 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e2 ∗ e1 = e2, e2 ∗ e2 = e1, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e2; H2 5 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e2 ∗ e1 = e1, e2 ∗ e2 = e2, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e2; H2 6 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e1 = e1, e2 ∗ e2 = e2, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e2; H2 7 : e1 ∗ e2 = e1, e2 ∗ e1 = −e1, e2 ∗ e2 = −e2, α(e1) = −e1, α(e2) = e2, β(e1) = e1, β(e2) = e2; H2 8 : e1 ∗ e1 = −e1, e1 ∗ e2 = −e2, e2 ∗ e2 = e2, α(e1) = e1, α(e2) = −e2, β(e1) = e1, β(e2) = e2; H2 9 : e1 ∗ e1 = e1, e1 ∗ e2 = −e2, e2 ∗ e1 = e2, α(e1) = e1, α(e2) = e2, β(e1) = e2, β(e2) = −e2; H2 10 : e1 ∗ e2 = e1, e2 ∗ e1 = e1, e2 ∗ e2 = e1 + e2, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e2; H2 11 : e1 ∗ e1 = e2, e1 ∗ e2 = e1, e2 ∗ e1 = e1, e2 ∗ e2 = e2, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e2; H2 12 : e1 ∗ e2 = e1, e2 ∗ e1 = e1, e2 ∗ e2 = e1, β(e2) = e1; H2 13 : e2 ∗ e2 = e1, α(e1) = e1, α(e2) = e1 + e2, β(e1) = e1, β(e2) = e2. Theorem 2.3. Every 2-dimensional unital multiplicative BiHom-associative algebra is isomorphic to one of the following pairwise non-isomorphic BiHom-associative algebras (A, ∗, α, β) appearing in Table 2, where ∗ is the multiplication and α and β the structure maps. We set {e1, e2} to be a basis of K2 where e1 is the unit : Table 2 Hu2 1 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e2 ∗ e1 = e2, e2 ∗ e2 = e1 + e2, α(e1) = β(e1) = e1, α(e2) = β(e2) = e2; 10 AHMED ZAHARI Hu2 2 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e2 ∗ e1 = e2, α(e1) = β(e1) = e1, α(e2) = β(e2) = e2; Hu2 3 : e1 ∗ e1 = e1, e1 ∗ e2 = −e2, e2 ∗ e1 = −e2, e2 ∗ e2 = e1, α(e1) = β(e1) = e1, α(e2) = β(e2) = −e2; Hu2 4 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e2 ∗ e1 = e2, e2 ∗ e2 = e2, α(e1) = β(e1) = e1, α(e2) = β(e2) = e2. Theorem 2.4. Every 3-dimensional BiHom-associative algebra is isomorphic to one of the follow- ing pairwise non-isomorphic BiHom-associative algebras (A, ∗, α, β) where ∗ is the multiplication and α and β the structure maps. We set {e1, e2, e3} to be a basis of K3. H3 1 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e3 ∗ e2 = e3, e3 ∗ e3 = e3, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e1 − e2; H3 2 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e1 ∗ e3 = e3, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e2 ∗ e3 = e3, e3 ∗ e1 = e3, e3 ∗ e2 = e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1; H3 3 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e2 = e2, e1 ∗ e3 = −e3, e3 ∗ e2 = e1 − e2, α(e1) = e1, α(e2) = e2, α(e3) = −e3, β(e1) = e1, β(e2) = e1; H3 4 : e1 ∗ e1 = e1, e1 ∗ e2 = e1 − e2, e2 ∗ e1 = e2, e2 ∗ e2 = −e1, α(e1) = e1, α(e2) = e2, β(e1) = e1; H3 5 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e3 ∗ e1 = e3, e3 ∗ e2 = e3, e3 ∗ e3 = e1 − e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1; H3 6 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e1 ∗ e3 = e3, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e2 ∗ e3 = e3, e3 ∗ e1 = e3, e3 ∗ e2 = e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1; H3 7 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e3 ∗ e2 = e3, e3 ∗ e3 = e3, α(e1) = e1, α(e2) = e2, β(e1) = e1; H3 8 : e1 ∗ e3 = e1 − e2, e2 ∗ e3 = e1 − e2, e3 ∗ e3 = e1 − e2, α(e1) = e1, α(e2) = e1 − e2, α(e3) = e2 − e3, β(e1) = e1, β(e2) = e1 − e2, β(e3) = e2 − e3; H3 9 : e2∗e3 = −e1, e3∗e2 = e1, e3∗e3 = e1, α(e1) = e1, α(e2) = e1+e2, α(e3) = e2+e3, β(e1) = e1, β(e2) = e1 + e2, β(e3) = e2 + e3; H3 10 : e1∗e3 = e1+e2, e2∗e3 = e1+e2, e3 ∗e1 = e1 −e2, e3∗e2 = e1−e2, e3∗e3 = e1−e2, α(e1) = e1, α(e2) = e1 − e2, α(e3) = e2 − e3, β(e1) = e1, β(e2) = e1 − e2, β(e3) = e2 − e3; H3 11 : e2 ∗ e3 = −e1, e3 ∗ e2 = e1, e3 ∗ e3 = e1, α(e1) = e1, α(e2) = e1 + e2, α(e3) = e1 + e2 + e3, β(e1) = e1, β(e2) = e1 + e2, β(e3) = e1 + e2 + e3; H3 12 : e1 ∗ e3 = e1 − e2, e2 ∗ e3 = e1 − e2, e3 ∗ e3 = e1 − e2, α(e1) = e1, α(e2) = e1, α(e3) = 11 e2, β(e1) = e1, β(e2) = e1, β(e3) = e2; H3 13 : e1∗e3 = e1−e2, e2∗e3 = e1−e2, e3 ∗e1 = e1 −e2, e3∗e2 = e1−e2, e3∗e3 = e1−e2, α(e1) = e1, α(e2) = e1, α(e3) = e2, β(e1) = e1, β(e2) = e1, β(e3) = e2. Theorem 2.5. Every 3-dimensional unital BiHom-associative algebra is isomorphic to one of the following pairwise non-isomorphic BiHom-associative algebras (A, ∗, α, β) where ∗ is the multipli- cation and α and β the structure maps. We set {e1, e2, e3} to be a basis of K3. Hu3 1 : e1 ∗ e1 = e1, e1 ∗ e2 = e1 − e2, e2 ∗ e1 = e2, e2 ∗ e2 = −e1, e3 ∗ e3 = −e3 α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e1 − e2; Hu3 2 : e1∗e1 = e1, e1∗e2 = e1, e2∗e1 = e2, e2∗e2 = e2, e2∗e3 = e1−e2, e3∗e1 = e3, e3∗e2 = e3, e3 ∗ e3 = e1 − e2, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1; Hu3 3 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e3 ∗ e1 = e3, e3 ∗ e2 = e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1; Hu3 4 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e3 ∗ e3 = e3, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e1; Hu3 5 : e1 ∗ e1 = e1, e1 ∗ e2 = e1 − e2, e2 ∗ e1 = e2, e3 ∗ e1 = e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1 − e2; Hu3 6 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e2 ∗ e3 = e1 − e2, e3 ∗ e1 = e3, e3 ∗ e2 = e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1; Hu3 7 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e1 = e2, e2 ∗ e2 = e2 + e3, e3 ∗ e1 = e3, e3 ∗ e2 = e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1; Hu3 8 : e1 ∗ e1 = e1, e1 ∗ e2 = e1 − e2, e2 ∗ e1 = e2, e3 ∗ e1 = e3, e3 ∗ e2 = e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1 − e2; 12 AHMED ZAHARI H3 9 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e3 ∗ e1 = e3, e3 ∗ e2 = e3, α(e1) = e1, α(e2) = e2, α(e3) = e3, β(e1) = e1, β(e2) = e1; Hu3 10 : e1 ∗ e1 = e1, e1 ∗ e2 = e1, e1 ∗ e3 = e3, e2 ∗ e1 = e2, e2 ∗ e2 = e2, e2 ∗ e3 = e3, α(e1) = e1, α(e2) = e2, β(e1) = e1, β(e2) = e1, β(e3) = e3; Hu3 11 : e1 ∗ e1 = e1, e1 ∗ e3 = e3, e2 ∗ e2 = e2, e3 ∗ e2 = e2, α(e1) = e1, β(e1) = e1, β(e3) = e3; Hu3 12 : e1 ∗ e1 = e1, e1 ∗ e2 = e2, e2 ∗ e3 = e3, e3 ∗ e3 = e3, α(e1) = e1, β(e1) = e1, β(e2) = e2; Hu3 13 : e1 ∗ e1 = e1, e2 ∗ e2 = e2, e2 ∗ e3 = e2, e3 ∗ e1 = e3, α(e1) = e1, α(e3) = e3, β(e1) = e1. 3. BiHom-coassociative coalgebras and BiHom-bialgebras In this Section, we show that for a fixed dimension n, the set of BiHom-bialgebras is endowed with a structure of an algebraic variety and a natural structure transport action which describes the set of isomorphic BiHom-algebras. Solving such systems of polynomial equations leads to the classification of such structures. We shall now introduce the dual concept to BiHom-associative algebras: Definition 3.1. A BiHom-coassociative coalgebra is 4-tuple (A, ∆, ψ, ω) in which A is a linear space, ψ, ω : A −→ A and ∆ : A −→ A ⊗ A are linear maps, such that (3.1) ψ ◦ ω = ω ◦ ψ, (ψ ⊗ ψ) ◦ ∆ = ∆ ◦ ψ, (ω ⊗ ω) ◦ ∆ = ∆ ◦ ω, (∆ ⊗ ψ) ◦ ∆ = (ω ⊗ ∆) ◦ ∆. We call ψ and ω (in this order) the structure maps of A. Let V be an n-dimensional vector space over K. Fixing a basis {ei}i={1,...,n} of V , a multiplication µ (resp. linear maps α, β, ψ, ω and a comultiplication ∆) is identified with its n3, 2n2 structure con- ij ∈ K (resp. aji, bji, ξji, Djk stants C k β(ei) = Pn j,k=1 Djk counit ε is identified with its n structure constants ζi. We assume that e1 is the unit. i and γji) where µ(ei ⊗ ej) = Pn j=1 γjiej and ∆(ei) = Pn j=1 bjiej, ψ(ei) = Pn j=1 ξjiej, ω(ei) = Pn k=1 C k ijek, α(ei) = Pn j=1 ajiej, i ej ⊗ ek. The A family n(C k i ), . . . , i, j, k ∈ {1, . . . , n}o represents a BiHom-coassociative coalgebra if the underlying family satisfies the appropriate conditions which translate to the follow- ij , aji, bji, ξji, γji, Djk ing polynomial equations: 13 ∀ i, p ∈ {1, . . . , n} , (3.2)   k=1 γkiξpk − Pn Pn Pn j=1 Pn k=1 Djk j=1 Pn Pn k=1 Djk Pn j=1 Pn k=1(Djk i ξpjξqk − Pn i γpjγqk − Pn i ξrk − Djk j=1 ξjiγpj = 0, j=1 ξjiDqp k=1 γkiDpq j = 0, k = 0, i γrjDpq k ) = 0, ∀ i, p, q ∈ {1, . . . , n} , ∀ i, p, q ∈ {1, . . . , n} , ∀ i, p, q, r ∈ {1, . . . , n} . A morphism f : (A, ∆A, ψA, ωA) −→ (B, ∆B , ψB, ωB) of BiHom-coassociative coalgebras is a linear map f : A −→ B such that ψB ◦ f = f ◦ ψA, ωB ◦ f = f ◦ ωA and (f ⊗ f ) ◦ ∆A = ∆B ◦ f. (3.3)   Pn j=1(djiξkj − ξjidkj ) = 0, Pn j=1(djiγkj − γjidkj ) = 0, k=1 Djk i dpjdqk − Pn j=1 djiDqp Pn j=1 Pn ∀ i, k ∈ {1, . . . , n} , ∀ i, k ∈ {1, . . . , n} , j = 0, ∀ i, p, q ∈ {1, . . . , n} . A BiHom-coassociative coalgebra (A, ∆, ψ, ω) is called counital if there exists a linear map ε : A → K (called a counit) such that (3.4) ε ◦ ψ = ε, ε ◦ ω = ε, (idA ⊗ ε) ◦ ∆ = ω and (ε ⊗ idA) ◦ ∆ = ψ. We have Pn j=1 ξjiζj = ζi, Pn j=1 γjiζj = ζi, Pn p=1 Pn q=1 Dpq i εq = aij , Pn p=1 Pn q=1 Dqr i εr = bij. A morphism of counital BiHom-coassociative coalgebras f : A −→ B is called counital if f = εA, where εA and εB are the counits of A and B, respectively. Definition 3.2. A BiHom-bialgebra is a 7-tuple (H, µ, ∆, α, β, ψ, ω), with the property that (H, µ, α, β) is a BiHom-associative algebra, (H, ∆, ψ, ω) is a BiHom-coassociative coalgebra and moreover the following relations are satisfied, for h, h′ ∈ H : ∆(hh′) = h1h′ 1 ⊗ h2h′ 2, α ◦ ψ = ψ ◦ α, α ◦ ω = ω ◦ α, (3.5) β ◦ ψ = ψ ◦ β, β ◦ ω = ω ◦ β, (α ⊗ α) ◦ ∆ = ∆ ◦ α, (β ⊗ β) ◦ ∆ = ∆ ◦ β, ψ(hh′) = ψ(h)ψ(h′), ω(hh′) = ω(h)ω(h′). 14 AHMED ZAHARI   (3.6) (3.7) Pn k=1 C k i,jDpq k − Pn i Duv ruCq sv = 0, j C p ∀ i, k ∈ {1, . . . , n} , ∀ i, j, p, q ∈ {1, . . . , n} , s=1 Pn u=1 Pn r=1 Pn v=1 Drs Pn j=1(ξjiakj − ajiξkj ) = 0, Pn j=1(γjiakj − ajiγkj ) = 0, Pn j=1(ξjibkj − bjiξkj ) = 0, Pn j=1(γjibkj − bjiγkj ) = 0, q=1 Dpq q=1 Dpq ij ξpk − Pn ij γpk − Pn i arpasq − Pn i brpbsq − Pn q=1 ξkiξqjC p q=1 γkiγqjC p k=1 Pn k=1 Pn j=1 ajiDrs j=1 bjiDrs Pn p=1 Pn Pn p=1 Pn Pn k=1 C k Pn k=1 C k ∀ i, k ∈ {1, . . . , n} , ∀ i, k ∈ {1, . . . , n} , ∀ i, k ∈ {1, . . . , n} , j = 0, j = 0, kq = 0, kq = 0, ∀ i, r, s ∈ {1, . . . , n} , ∀ i, r, s ∈ {1, . . . , n} , ∀ i, j, p ∈ {1, . . . , n} , ∀ i, j, p ∈ {1, . . . , n} . We say that H is a unital and counital BiHom-bialgebra if, in addition, it admits a unit uH and a counit εH such that ∆(uH ) = uH ⊗ uH , εH(uH ) = 1, ψ(uH ) = uH , ω(uH) = uH , εH ◦ α = εH , εH ◦ β = εH, εH (hh′) = εH (h)εH (h′), ε(e1) = 1, ψ(e1) = e1, ω(e1) = e1, Pn ∀h, h′ ∈ H. j=1 ajiζj = ζi, and Pn j=1 bjiηj = We have ∆(e1) = e1⊗e1, ηi. Let us record the formula expressing the BiHom-coassociativity of ∆ : (3.8) ∆(h1) ⊗ ψ(h2) = ω(h1) ⊗ ∆(h2), h ∈ H. 3.1. Classification in Dimension 2 in H2 i . Thanks to computer Hom-algebra, we obtain the following Hom-coalgebras associated to the pre- vions Hom algebras in order to obtain a Hom-bialgebra structures. We denote the comultiplications by ∆2 i,j , where i indicates the item of the multiplication and j the item of comultiplication. Theorem 3.3. The set of 2-dimensional BiHom-Bialgebras algebras yields two non-isomorphic algebras. Let {e1, e2} be a basis of K2, then the BiHom-Bialgebras are given by the following non- trivial comultiplications. 1. ∆(e1) = e1 ⊗ e2 + e2 ⊗ e2, ∆(e2) = e1 ⊗ e1 − e2 ⊗ e2, ψ(e1) = −e1, ψ(e2) = e2, ω(e1) = −e1, ω(e2) = e2; 2. ∆(e1) = e1 ⊗e1, ∆(e2) = e1 ⊗e1 +e1 ⊗e2 +e2 ⊗e1 +e2 ⊗e2, ψ(e1) = e1, ω(e1) = e1, ω(e2) = e2; 3. ∆(e1) = e1⊗e1−e1⊗e2−e2⊗e1+e2⊗e2, ∆(e2) = e1⊗e1−e1⊗e2−e2⊗e1+e2⊗e2, ψ(e1) = e1, ψ(e2) = e2, ω(e1) = e1 − e2, ω(e2) = e1 − e2; 4. ∆(e1) = e1 ⊗e1−e1⊗e2−e2⊗e1 +e2⊗e2, ∆(e2) = e1 ⊗e1−e1⊗e2−e2⊗e1 +e2⊗e2, ψ(e1) = e1 − e2, ψ(e2) = e1 − e2, ω(e1) = e1, ω(e2) = e2; 15 5. ∆(e1) = e1 ⊗e1−e1⊗e2−e2⊗e1 +e2⊗e2, ∆(e2) = e1 ⊗e1−e1⊗e2−e2⊗e1 +e2⊗e2, ψ(e1) = e1 − e2, ψ(e2) = e1 − e2, ω(e1) = e1 − e2, ω(e2) = e1 − e2; 6. ∆(e1) = e1 ⊗e1−e1⊗e2−e2⊗e1 +e2⊗e2, ∆(e2) = e1 ⊗e1−e1⊗e2−e2⊗e1 +e2⊗e2, ψ(e1) = −e1 + e2, ψ(e2) = −e1 + e2, ω(e1) = e1, ω(e2) = e2; 7. ∆(e1) = e1 ⊗e1−e1⊗e2−e2⊗e1 +e2⊗e2, ∆(e2) = e1 ⊗e1−e1⊗e2−e2⊗e1 +e2⊗e2, ψ(e1) = e1, ψ(e2) = e2; 8. ∆(e1) = e1 ⊗e1+e1⊗e2+e2⊗e1 +e2⊗e2, ∆(e2) = e1 ⊗e1+e1⊗e2+e2⊗e1 +e2⊗e2, ψ(e1) = e1 − e2, ψ(e2) = −e1 + e2; 9. ∆(e1) = e1 ⊗e1+e1⊗e2+e2⊗e1 +e2⊗e2, ∆(e2) = e1 ⊗e1+e1⊗e2+e2⊗e1 +e2⊗e2, ψ(e1) = e1 − e2, ψ(e2) = −e1 + e2, ω(e1) = e1 − e2, ω(e2) = −e1 + e2; 10. ∆(e1) = e1 ⊗e1+e1 ⊗e2+e2⊗e1 +e2⊗e2, ∆(e2) = e1 ⊗e1+e1 ⊗e2+e2⊗e1 +e2⊗e2, ω(e1) = e1 + e2, ω(e2) = e1 + e2. Remark 3.4. There is no BiHom-Bialgabra whose underlying BiHom-associative algebra is given by H2 1. Theorem 3.5. The set of 2-dimensional unital BiHom-Bialgebras yields two non-isomorphic al- gebras. Let {e1, e2} be a basis of K2, then the unital BiHom-Bialgebras are given by the following non-trivial comultiplications. 1. ∆2 1,1(e1) = e1 ⊗ e1, ∆2 1,1(e2) = −e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1 + e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e2, ω(e1) = e1, ω(e2) = e2, ε(e1) = 1, ε(e2) = 2; 2. ∆2 1,2(e1) = e1 ⊗ e1, ∆2 1,2(e2) = −e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1 + e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e1, ω(e1) = e1, ω(e2) = e1, ε(e1) = 1, ε(e2) = 2; 3. ∆2 1,3(e1) = e1 ⊗ e1, ∆2 1,3(e2) = e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1 − e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e1, ω(e1) = e1, ω(e2) = e1, ε(e1) = 1, ε(e2) = −1; 4. ∆2 1,4(e1) = e1 ⊗ e1, ∆2 1,4(e2) = e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1 − e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = −e1, ω(e1) = e1, ω(e2) = −e1, ε(e1) = 1, ε(e2) = −1; 5. ∆2 2,1(e1) = e1 ⊗ e1, ∆2 2,2(e2) = −e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1 + e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e1, ω(e1) = e1, ω(e2) = e1, ε(e1) = 1, ε(e2) = 1; 16 AHMED ZAHARI 6. ∆2 2,2(e1) = e1 ⊗ e1, ∆2 2,3(e2) = e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1 − e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e2, ω(e1) = e1, ω(e2) = e2, ε(e1) = 1, ε(e2) = 1; 7. ∆2 4,1(e1) = e1 ⊗ e1, ∆2 e2, ε(e1) = 1, ε(e2) = 1; 3,1(e2) = e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e2, ω(e1) = e1, ω(e2) = 8. ∆2 4,2(e1) = e1 ⊗ e1, ∆2 e1, ε(e1) = 1, ε(e2) = 1; 4,2(e2) = e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e1, ω(e1) = e1, ω(e2) = 9. ∆2 4,3(e1) = e1 ⊗ e1, ∆2 4,3(e2) = e1 ⊗ e2 + e2 ⊗ e1 − 2e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e2, ω(e1) = e1, ω(e2) = e2, ε(e1) = 1, ε(e2) = 1; 10. ∆2 4,4(e1) = e1 ⊗ e1, ∆2 4,4(e2) = e1 ⊗ e2 + e2 ⊗ e1 − e2 ⊗ e2, ψ(e1) = e1, ψ(e2) = e1, ω(e1) = e1, ω(e2) = e1, ε(e1) = 1; Remark 3.6. There is no BiHom-Bialgabra whose underlying BiHom-associative algebra is given by Hu2 3. 3.2. Classification in Dimension 3 in H3 i . Thanks to computer Hom-algebra, we obtain the following Hom-coalgebras associated to the pre- vions Hom algebras in order to obtain a Hom-bialgebra structures. We denote the comultiplications by ∆3 i,j , where i indicates the item of the multiplication and j the item of comultiplication. Theorem 3.7. The set of 3-dimensional BiHom-Bialgebras yields two non-isomorphic algebras. Let {e1, e2, e3} be a basis of K3, then the BiHom-Bialgebras are given by the following non-trivial comultiplications. 1. ∆(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e1 + e1 ⊗ e2 − e1 ⊗ e3 + e2 ⊗ e1 + e2 ⊗ e3 − e3 ⊗ e1 − e3 ⊗ e3, ∆(e3) = e3 ⊗ e3, ψ(e1) = e1, ψ(e2) = e1 + e2, ψ(e3) = e3, ω(e1) = e1, ω(e2) = e1 + e2, ω(e3) = e3; 2. ∆(e1) = e1 ⊗e1, ∆(e2) = e1 ⊗e1 +e1 ⊗e2 +e1 ⊗e3 +e2 ⊗e1 −e2 ⊗e2 +e2 ⊗e3 −e3 ⊗e1 +e3 ⊗e2 ∆(e3) = e3 ⊗ e3, ψ(e1) = e1, ψ(e2) = e1, ψ(e3) = e3, ω(e1) = e1, ω(e2) = e1, ω(e3) = e3; 3. ∆(e1) = e1⊗e1+e3⊗e3, ∆(e2) = e1⊗e2+e2⊗e1, ∆(e3) = e1⊗e3+e3⊗e1, ψ(e1) = e1, ψ(e3) = e3, ω(e1) = e1, ω(e3) = e3; 4. ∆(e1) = e1 ⊗e1 +e3 ⊗e3, ∆(e2) = e1 ⊗e2 +e2 ⊗e1, ∆(e3) = −e1 ⊗e3 −e3 ⊗e1, ψ(e1) = e1, ψ(e3) = e3, ω(e1) = e1, ω(e3) = e3; 17 5. ∆(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e2 + e2 ⊗ e1 + e2 ⊗ e2, ∆(e3) = e1 ⊗ e3 + e2 ⊗ e3 + e3 ⊗ e1 + e3 ⊗ e2 − e3 ⊗ e3, ψ(e1) = e1, ω(e1) = e1; 6. ∆(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e2 + e2 ⊗ e1 + e2 ⊗ e2, ∆(e3) = e1 ⊗ e3 + e2 ⊗ e3 + e3 ⊗ e1 + e3 ⊗ e2 − e3 ⊗ e3, ψ(e1) = e1, ψ(e2) = e2, ω(e1) = e1, ω(e2) = e2; 7. ∆(e1) = 0, ∆(e2) = −e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1, ∆(e3) = e1 ⊗ e3 + e3 ⊗ e1, ψ(e1) = e1, ψ(e2) = e1 − e2, ψ(e3) = −e3, ω(e1) = e1, ω(e2) = e1 − e2, ω(e3) = −e3; 8. ∆(e1) = 0, ∆(e2) = −e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1, ∆(e3) = e1 ⊗ e3 + e3 ⊗ e1, ψ(e1) = e1, ψ(e2) = e2, ψ(e3) = e3, ω(e1) = e1, ω(e2) = e1 − e2, ω(e3) = −e3; 9. ∆(e1) = 0, ∆(e2) = −e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1, ∆(e3) = e1 ⊗ e3 + e3 ⊗ e1, ψ(e1) = e1, ψ(e2) = e1 − e2, ψ(e3) = −e3, ω(e1) = e1, ω(e2) = e2, ω(e3) = e3. Remark 3.8. There is no BiHom-Bialgabra whose underlying BiHom-associative algebra is given by H3 3, H3 5. Theorem 3.9. The set of 3-dimensional unital BiHom-Bialgebras yields two non-isomorphic alge- bras. Let {e1, e2, e3} be a basis of K3, then the unital BiHom-Bialgebras are given by the following non-trivial comultiplications. 1. ∆3 2,1(e1) = e1 ⊗ e1, ∆(e2) = −e1 ⊗ e1 + e1 ⊗ e2 + e2 ⊗ e1, ∆(e3) = e1 ⊗ e1 − e1 ⊗ e2 + 2e1 ⊗ e3 − e2 ⊗ e1 + e2 ⊗ e2 − e2 ⊗ e2 + 2e3 ⊗ e1 − e3 ⊗ e2 + e3 ⊗ e3, ψ(e1) = e1, ψ(e2) = e1, ω(e1) = e1, ω(e2) = e1, ε(e1) = ε(e2) = 1; 2. ∆3 3,1(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e1 + e1 ⊗ e2 + e1 ⊗ e3 + e2 ⊗ e1 − e2 ⊗ e2 − e2 ⊗ e3 − e3 ⊗ e1 + e3 ⊗ e2, ∆(e3) = −ae3 ⊗ e3, ψ(e1) = e1, ψ(e2) = e2, ψ(e3) = e3, ω(e1) = e1, ω(e2) = e2, ω(e3) = e3, ε(e1) = ε(e2) = 1; 3. ∆3 4,1(e1) = e1 ⊗ e1, ∆(e2) = e2 ⊗ e2, ∆(e3) = e1 ⊗ e3 + e3 ⊗ e1 + e3 ⊗ e3, ψ(e1) = e1, ψ(e2) = e2, ψ(e3) = e3, ω(e1) = e1, ω(e2) = e2, ω(e3) = e3, ε(e1) = ε(e2) = 1; 4. ∆3 12,1(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e2 + e2 ⊗ e1, ∆(e3) = e1 ⊗ e3 + e3 ⊗ e1, ψ(e1) = e1, ω(e1) = e1, ε(e1) = 1; 5. ∆3 15,1(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e1 + e1 ⊗ e2 − e1 ⊗ e3 + e2 ⊗ e1 − e2 ⊗ e3 − e3 ⊗ e1 − e3 ⊗ e2 + 2e3 ⊗ e3, ∆(e3) = −e2 ⊗ e2 + e2 ⊗ e3 + e3 ⊗ e2, ψ(e1) = e1, ψ(e3) = e1, ω(e1) = e1, ω(e3) = e1, ε(e1) = ε(e3) = 1; 6. ∆3 15,2(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e2 + e2 ⊗ e1 − ae2 ⊗ e2 − e2 ⊗ e3 − e3 ⊗ e2, ∆(e3) = e1 ⊗ e3 + be2 ⊗ e2 − e3 ⊗ e1 − e3 ⊗ e3, ψ(e1) = e1, ψ(e2) = ce2, ψ(e3) = de2 + e3, ω(e1) = e1, ω(e2) = ce2, ω(e3) = de2 + e3, ε(e1) = 1; 18 AHMED ZAHARI 7. ∆3 20,1(e1) = e1 ⊗e1, ∆(e2) = e2⊗e2, ∆(e3) = e1 ⊗e3+e3⊗e1+e3⊗e3, ψ(e1) = e1, , ψ(e2) = e2, ψ(e3) = e3, ω(e1) = e1, ω(e2) = e2, ω(e3) = e3, ε(e1) = ε(e2) = 1; 8. ∆3 21,1(e1) = e1 ⊗e1, ∆(e2) = e1 ⊗e2 +e2 ⊗e1 +e2 ⊗e2, ∆(e3) = e3 ⊗e3, ψ(e1) = e1, ψ(e2) = e2, ω(e1) = e1, ω(e2) = e2, ε(e1) = 1, ε(e2) = 1; 9. ∆3 21,2(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e2 + e2 ⊗ e1, ∆(e3) = e1 + ⊗e3 + e3 ⊗ e1, ψ(e1) = e1, ω(e1) = e1, ε(e1) = 1; 10. ∆3 22,1(e1) = e1 ⊗ e1, ∆(e2) = e1 ⊗ e2 + e2 ⊗ e1 + e2 ⊗ e2 + e2 ⊗ e3 + e3 ⊗ e2, ∆(e3) = e1 + ⊗e3 + e3 ⊗ e1 + e3 ⊗ e3, ψ(e1) = e1, ψ(e3) = e3, ω(e1) = e1, ω(e3) = e3, ε(e1) = 1; 11. ∆3 22,2(e1) = e1 ⊗ e1, ∆(e2) = e2 ⊗ e2, ∆(e3) = e1 + ⊗e3 + e3 ⊗ e1 + e3 ⊗ e3, ψ(e1) = e1, ω(e1) = e1, ε(e1) = 1; 12. ∆3 22,3(e1) = e1 ⊗ e1, ∆(e2) = e2 ⊗ e2, ∆(e3) = e1 + ⊗e3 + e3 ⊗ e1 + e3 ⊗ e3, ψ(e1) = e1, ψ(e3) = e3, ω(e1) = e1, ω(e3) = e3, ε(e1) = 1; 13. ∆3 22,4(e1) = e1 ⊗ e1, ∆(e2) = e2 ⊗ e2, ∆(e3) = e1 + ⊗e3 + e3 ⊗ e1 + e3 ⊗ e3, ψ(e1) = e1, ψ(e2) = e2, ψ(e3) = e3, ω(e1) = e1, ω(e2) = e2, ω(e3) = e3, ε(e1) = ε(e2) = 1. Remark 3.10. There is no unital BiHom-Bialgabra whose underlying unital BiHom-associative al- gebra is given by Hu3 1, Hu3 5, Hu3 7, Hu3 8, Hu3 10. We introduce the concept of BiHom-Hopf algebras. Definition 3.11. Let (H, µ, ∆, α, β) be a unital and counital BiHom-bialgebra with a unit 1H and a co-unit εH . A linear map S : H → H is called an antipode if it commutes with all the maps α, β, ψ, ω and it satisfies the relation : ψω(S(h1))αβ(h2) = εH (h)1H = βψ(h1)αω(S(h2)), ∀h ∈ H. A BiHom-Hopf algebra is a unital and counital BiHom-bialgebra with an antipode. Pn Pn s=1 Pn s=1 Pn q,r=1 Pn q,r=1 Pn l,p=1 Pn l,p=1 Pn j,k=1 Djk j,k=1 Djk i SljgpkfqlarpbqrC t i fljSljbqlgrpasrC t sr − ξi = 0, qr − ξi = 0, ∀ i, t ∈ {1, n} ; ∀ i, t ∈ {1, n} . Proposition 3.12. The BiHom-bialgebra structures on K2 which are BiHom-Hopf algebras are given by the following pairs of multiplication and comultiplication with the appropriate unit and conits : (Hu2 1, ∆2 1,1), (Hu2 1, ∆2 1,2), (Hu2 1, ∆2 1,4), (Hu2 2, ∆2 2,1), (Hu2 2, ∆2 2,2), (Hu2 4, ∆2 4,1), (Hu2 4, ∆2 4,2), (Hu2 4, ∆2 4,3), (Hu2 4, ∆2 4,4). Proposition 3.13. The BiHom-bialgebra structures on K3 which are BiHom-Hopf algebras are given by the following pairs of multiplication and comultiplication with the appropriate unit and conits : (Hu3 (Hu3 3, ∆3 22, ∆3 3,1), (Hu3 22,2), (Hu3 4,1), (Hu3 12, ∆3 22,3), (Hu3 12,1), (Hu3 22, ∆3 22,4). 22, ∆3 4, ∆3 15, ∆3 15,1), (Hu3 15, ∆3 15,2), (Hu3 21, ∆3 21,1), (Hu3 21, ∆3 21,2), (Hu3 22, ∆3 22,1), References 19 [1] G. Graziani, A. Makhlouf, C. Menini, F. Panaite BiHom-asosociative algebras, BiHom-Lie algebras and BiHom- bialgebras, Symetry, Integrability and gepmerty, SIGMA 11(2015), 086 34 pages [2] A. Armour, H. Chen and Y. Zhang, Classification of 4-dimensional superalgebras, Comm. in Algebra 37(2009), 3697-3728. [3] M. Goze and A. Makhlouf, Classification and rigid associative algebras in low dimensions, Lois d'Algèbres et variétés algébriques (Colmar, 1991), 5-22, Travaux en cours, 50 Hermann, Paris, (1996). [4] M. Goze, and A. Makhlouf, On the rigid complex associative algebra, Comm. Algebra 18(12)(1990)4031-4046. [5] J.T. Hartwig, D. Larsson and S. Silvestrov, Deformations of Lie algebras using σ- derivations, J. Algebra 295 (2006), 314Ð361. [6] D. Larsson and S. Silvestrov, Quasi-hom-Lie algebras, central extensions and 2-cocycle- like identities, J. Algebra 288 (2005), 321 -- 344. [7] X.X. Li, Structures of multiplicative Hom-Lie algebras, Advances in Mathematics (China), 43(6)(2014)817-823. [8] A. Makhlouf and S. Silvestrov, Notes on formal deformations of Hom-associative and Hom-Lie algebras, Forum Mathematicum, vol. 22 (4) (2010) 715 -- 759. [9] A. Makhlouf and S. Silvestrov, Hom-algebra structures, J. Gen. Lie Theory Appl. Vol.2 (2008), No.2, 51-64. [10] D. Yau, Enveloping algebra of Hom-Lie algebras, J. Gen. Lie Theory Appl. 2 (2008), 95 -- 108. [11] D. Yau, Hom-algebras and homology, J. Lie Theory, 19 (2009), 409 -- 421. [12] Y. Sheng, Representations of hom-Lie algebras, Algebras and Representation Theory 15 (2012), 1081 -- 1098. Université de Haute Alsace, Laboratoire de Mathématiques, Informatique et Applications, 4, rue des Frères Lumière F-68093 Mulhouse, France E-mail address: [email protected]
1803.01140	1	1803	2018-03-03T10:31:43	A note on homotopy categories of FP-Injectives	[ "math.RA", "math.CT" ]	For a locally finitely presented Grothendieck category $\mathcal{A}$, we consider a certain subcategory of the homotopy category of FP-injective objects in $\mathcal{A}$ which we show is compactly generated. In the case where $\mathcal{A}$ is locally coherent, we identify this subcategory with the derived category of FP-injective objects in $\mathcal{A}$. Our results are, in a sense, dual to the ones obtained by Neeman on the homotopy category of flat modules. Our proof is based on extending a characterization of the pure acyclic complexes which is due to Emmanouil.	math.RA	math	A NOTE ON HOMOTOPY CATEGORIES OF FP-INJECTIVES GEORGIOS DALEZIOS Abstract. For a locally finitely presented Grothendieck category A, we con- sider a certain subcategory of the homotopy category of FP-injectives in A which we show is compactly generated. In the case where A is locally coher- ent, we identify this subcategory with the derived category of FP-injectives in A. Our results are, in a sense, dual to the ones obtained by Neeman [21] on the homotopy category of flat modules. Our proof is based on extending a characterization of the pure acyclic complexes which is due to Emmanouil [7]. 1. Introduction The work of Neeman [21] on the homotopy category of flat modules has led to interesting advances in ring theory and homological algebra. Neeman was inspired by work of Iyengar and Krause [16], who proved that over a Noetherian ring R with D⊗R− −−−−→ K(Inj R) is a dualizing complex D, the composite K(Proj R) ֌ K(Flat R) an equivalence, which restricts to Grothendieck duality. Neeman in [21] focuses on the embedding K(Proj R) ֌ K(Flat R) for a general ring R and shows that the category K(Proj R) is ℵ1–compactly generated and that the composite of canonical maps K(Proj R) ֌ K(Flat R) −→ D(Flat R) is an equivalence. Related work on homotopy categories and the existence of adjoints between them is done by Krause [17], Murfet and Salarian [18], Saor´ın and Sťov´ıcek [23], and others. Closely related to the notion of flatness, is the notion of purity [4]. A submodule A ≤ B is called pure if any finite system of linear equations with constants from A and a solution in B, has a solution in A. This condition can be expressed diagrammatically, and is equivalent to asking for the sequence A ֌ B ։ B/A to remain exact after applying, for any finitely presented module F , the functor HomR(F, −), or equivalently the functor F ⊗R −. Such sequences are called pure exact and they are of interest since they form the smallest class of short exact sequences which is closed under filtered colimits. It follows from this discussion that a module M is flat if and only if any epimorphism with target M is pure. Thus flatness can be defined in any additive category which has an appropriate notion of finitely presented objects, namely locally finitely presented additive categories [2, 5]. If A is such a category, it is well known that the relation between purity and flatness can be given formally via the equivalence A ∼= Flat(fp(A)op, Ab); A 7→ 1, see [5, 1.4]. Thus, roughly speaking, the study of purity can HomA(−, A)fp(A) be reduced to the study of flat (left exact) functors, and Neeman's results have analogues in the context of purity, see Emmanouil [7], Krause [17], Simson [25] and Sťov´ıcek [28]. 2010 Mathematics Subject Classification. 18E30 (Primary) 16E35, 18G25 (Secondary). The author is supported by the Fundaci´on S´eneca of Murcia 19880/GERM/15. 1 fp(A) denotes a set of isomorphism classes of finitely presented objects in A, see 2.1. 1 2 GEORGIOS DALEZIOS The dual notion of flatness, in a locally finitely presented Grothendieck category A, is that of FP-injectivity. Namely, an object A in A is called FP-injective if any monomorphism with source A is pure. We denote the class of FP-injective objects by FPI(A). FP-injective modules were studied first by Stenstrom in [26]. One reason why they are of importance is because over (non-Noetherian) rings where injectives fail to be closed under coproducts, one can work with FP-injectives which are always closed under coproducts. Moreover, a ring is coherent if and only if the class of FP-injective modules is closed under filtered colimits, in strong analogy with the dual situation, where coherent rings are characterized by the closure of flat modules under products. In this note our goal is to provide, in a sense, duals of the above mentioned results of Neeman, that is, to obtain analogous results for the homotopy category of FP-injectives. For this we look at the tensor embedding functor of a mod- ule category to FP-injective (right exact) functors, that is, the functor Mod-R → A := (R-mod, Ab); M 7→ (M ⊗R −)R-mod, which identifies pure exact sequences in Mod-R with short exact sequences of FP-injective (right exact) functors, and induces an equivalence Mod-R ∼= FPI(A) [13, §1]. It is easy to observe that under this equivalence, the pure projective modules (the projectives with respect to the pure exact sequences) correspond to functors in the class FPI(A) ∩ ⊥ FPI(A)2. We point out that by work of Eklof and Trlifaj [6], we know that this class consists of those FP-injectives which are (summands of) transfinite extensions of finitely presented objects (see 2.5). We can now state our main result (3.5). Theorem. Let A be a locally finitely presented Grothendieck category and denote by FPI(A) the class of FP-injective objects in A. Then the homotopy category K(FPI(A) ∩ ⊥ FPI(A)) is compactly generated. Moreover, if A is locally coherent, the composite functor K(FPI(A) ∩ ⊥ FPI(A)) ֌ K(FPI(A)) can−−→ D(FPI(A)) is an equivalence of triangulated categories. Our proof is based on Neeman's strategy. Let C := FPI(A) ∩ ⊥ FPI(A). Since K(C) is compactly generated and admits coproducts, we obtain a right adjoint of the inclusion K(C) ֌ K(FPI(A)), and in the case where A is locally coherent, we identify its kernel with the pure acyclic complexes of FP-injective objects in A. From this it follows that any chain map from a complex in K(C) to a pure acyclic complex of FP-injectives is null homotopic. In fact, in 3.2 we prove something more general, namely that for any locally finitely presented Grothendieck category A, any chain map from a complex in K(⊥ FPI(A)) to a pure acyclic complex is null homotopic. This extends a result of Emmanouil [7]. Finally, we point out that Sťov´ıcek [28] has also studied the category D(FPI(A)) and in the locally coherent case has proved the existence of a model category struc- ture with homotopy category D(FPI(A)). Using our main result we can identify the cofibrant objects in this model structure with the category K(C) (see 3.7, 3.8). 2 ⊥ FPI(A) denotes the left orthogonal class of FP-injectives with respect to the Ext1 A(−, −) functor, see 2.3. A NOTE ON HOMOTOPY CATEGORIES OF FP-INJECTIVES 3 2. Preliminaries [2, 5] In an additive 2.1. Locally finitely presented additive categories. category A, an object X is called finitely presented if the functor HomA(X, −) : A → Ab preserves filtered colimits. A is called locally finitely presented if it is cocomplete, the isomorphism classes of finitely presented objects in A form a set fp(A), and every object in A is a filtered colimit of objects in fp(A). An abelian category A is locally finitely presented if and only if it is a Grothendieck category with a generating set of finitely presented objects [2, Satz 1.5]. A locally finitely presented Grothendieck category A is called locally coherent if the subcategory fp(A) is abelian. If A is a locally finitely presented additive category, a sequence 2.2. Purity. 0 → X → Y → Z → 0 in A is called pure exact if it is HomA(fp(A), −)–exact, that is, if for any A ∈ fp(A), the sequence 0 → HomA(A, X) → HomA(A, Y ) → HomA(A, Z) → 0 is an exact sequence of abelian groups. An object X ∈ A is called pure projective if any pure exact sequence of the form 0 → Z → Y → X → 0 splits, and dually X is called pure injective if any pure exact sequence of the form 0 → X → Y → Z → 0 splits. We will denote the class of pure projective objects in A by PProj(A). Pure exact sequences induce on the category A the structure of a (Quillen) exact category, i.e. we equip A with an exact structure [3, Dfn. 2.1] where the conflations are the pure exact sequences; see Crawley-Boevey [5, 3.1]. We will refer to this exact structure as the pure exact structure on A. 2.3. Cotorsion Pairs. ([22], see also [12]) Let X be a class of objects in an exact category A. Put X ⊥ := {A ∈ A ∀X ∈ X , Ext1 A(X, A) = 0} and define ⊥X analogously. A pair (X , Y) of classes in A is called a cotorsion pair if X ⊥ = Y and ⊥Y = X . A cotorsion pair is said to be generated by a set 3 if it is of the form (⊥(S ⊥), S ⊥) where S is a set of objects in A. A cotorsion pair (X , Y) is called complete if for every object A in A there exists a short exact sequence 0 → Y → X → A → 0 with X ∈ X and Y ∈ Y, and also a short exact sequence 0 → A → Y → X → 0 with X ∈ X and Y ∈ Y. It is called hereditary if X is closed under kernels of epimorphisms and Y is closed under cokernels of monomorphisms. We recall a fundamental result on cotorsion pairs generated by a set from the work of Eklof and Trlifaj [6]. First a definition. Definition 2.4. Let A be an abelian category and S a class of objects in A. An object A in A is called S-filtered if there exists a chain of subobjects 0 = A0 ⊆ A1 ⊆ ... ⊆ [α<σ Aα = A where σ is an ordinal, Aλ = ∪β<λAβ for all limit ordinals λ, and Aα+1/Aα ∈ S for all α < σ. The class of S-filtered objects will be denoted by Filt(S). Fact 2.5. ([6], see also [12, 3.2]) Let S be a (small) set of objects in a Grothendieck category and assume that S contains a generator. Then the following hold: 3This terminology is in accordance with Gobel and Trlifaj [12, Dfn. 2.2.1]. 4 GEORGIOS DALEZIOS (i) The cotorsion pair (⊥(S ⊥), S ⊥) is complete. (ii) The class ⊥(S ⊥) consists of direct summands of S-filtered objects, that is, for all X ∈ ⊥(S ⊥) we have P ∼= X ⊕ K where P ∈ Filt(S). Moreover, in this decomposition K can be chosen in ⊥(S ⊥) ∩ S ⊥. 2.6. FP-Injectives. [26] Let A be a locally finitely presented Grothendieck cat- egory. The objects in the class fp(A)⊥ = {A ∈ A ∀F ∈ fp(A), Ext1 A(F, A) = 0} are called FP-injective objects. We will denote this class by FPI(A). Note that fact 2.5, applied on the cotorsion pair (⊥ FPI(A), FPI(A)), tells us that the class ⊥ FPI(A) consists of direct summands of fp(A)-filtered modules. We recall the following well known facts for the class of FP-injectives. Fact 2.7. ([26], see also [28, App.B]) Let A be a locally finitely presented Grothendieck category. Then the following hold: (i) The class FPI(A) is closed under extensions, direct unions, products, co- products, and pure subobjects. (ii) An object A ∈ A belongs to FPI(A) if and only if any monomorphism with source A is pure. Moreover, the category A is locally coherent if and only if the class FPI(A) is closed under filtered colimits if and only if the class FPI(A) is closed under cokernels of monomorphisms. [19] Let E be a (Quillen) 2.8. The derived category of an exact category. exact category and denote by C(E) the corresponding category of chain complexes. C(E) has a canonical exact structure in which a diagram X ֌ Y ։ Z in C(E) is a conflation if and only if Xn ֌ Yn ։ Zn is a conflation in E for every n ∈ Z; see Buhler [3, Lem. 9.1]. We will refer to this exact structure as the induced exact structure on C(E). dn+1−−−→ Xn A complex X = · · · → Xn+1 dn−→ Xn−1 → · · · in C(E) is called acyclic (with respect to the exact structure of E) if each map dn decomposes in E as a deflation Xn ։ Zn−1(X) followed by an inflation Zn−1(X) ֌ Xn−1 and such that the induced sequence Zn(X) ֌ Xn ։ Zn−1(X) is a conflation in E. Denote by Kac(E) the homotopy category of acyclic complexes. If the exact category E has split idempotents, then Kac(E) is an ´epaisse (thick) subcategory of K(E) [19, 1.2] and then by definition [19, 1.5] the derived category of E is the Verdier quotient D(E) := K(E)/Kac(E). 3. On the homotopy category of fp-injectives Let A is a locally finitely presented Grothendieck category, viewed as an exact category with the pure exact structure, as in 2.2. Then the acyclic complexes in A (with respect to this exact structure) are called pure acyclic complexes and are de- noted by Cpac(A). Moreover, the subcategory of FP-injective objects in A is closed under extensions, therefore it is an exact category. It is easy to see that the acyclic complexes in C(FPI(A)) (with respect to the exact category FPI(A)) are the acyclic complexes (in the usual sense) with cycles in FPI(A). Equivalently, since the class FPI(A) is closed under pure subobjects (2.7), they are the pure acyclic complexes with components FP-injectives. Thus we will denote them by Cpac(FPI(A)). Then by definition (2.8) we have D(FPI(A)) := K(FPI(A))/Kpac(FPI(A)). A NOTE ON HOMOTOPY CATEGORIES OF FP-INJECTIVES 5 In Theorem 3.5, we identify D(FPI(A)) with a certain homotopy category. The key ingredient is to extend a characterization of the pure acyclic complexes which is due to Emmanouil. In [7] Emmanouil proves that a complex X is pure acyclic if and only if any chain map from a complex of pure projectives to X is null homotopic. Emmanouil's proof is self-contained, while Simson [25] and also Sťov´ıcek [28] give a functorial proof of this result by reducing it to Neeman's [21, Thm. 8.6]. We first recall a useful and well known lemma we will need. Lemma 3.1. Let A be an exact category and consider C(A) with the induced exact structure (as in 2.8). If X, Y ∈ C(A), denote by Ext1 C(A)(X, Y ) the abelian group of (Yoneda) extensions with respect to the induced exact structure, and by Ext1 dw(A)(X, Y ) the subgroup consisting of extensions Y ֌ T ։ X which are degreewise split. Then we have natural isomorphisms Ext1 dw(A)(X, Σ−(n+1)Y ) ∼= Hn HomA(X, Y ) ∼= HomK(A)(X, Σ−nY ). Proposition 3.2. (Compare with [7]) Let A be a locally finitely presented Grothendieck category and let X be a chain complex in A. Then the following are equivalent: (i) X is a pure acyclic complex. (ii) Any chain map from a complex in C(PProj(A)) to X is null-homotopic. (iii) Any chain map from a complex in C(⊥ FPI(A)) to X is null-homotopic. In particular, any pure acyclic complex with components in ⊥ FPI(A) is contractible. Proof. As we discussed above the assertions (i) ⇔ (ii) have been proved in [7]. Moreover, (iii) ⇒ (ii) is trivial, thus we are left with (ii) ⇒ (iii). First consider the case where we are given a chain map Y → X, with Y having components in Filt(fp(A)). From the fact that each component of Y is fp(A)-filtered, a result of Sťov´ıcek [27, Prop. 4.3] implies that Y itself is C−(fp(A))-filtered, that is, Y is given as a continuous chain of subcomplexes 0 = Y0 ⊆ Y1 ⊆ ... ⊆ [α<σ Yα = Y where σ is an ordinal, Yλ = ∪β<λYβ for all limit ordinals λ, and for all α < σ the quotient Yα+1/Yα is a bounded below complex with components finitely presented objects. Now, denote by C(A)pure the exact category of chain complexes with the induced pure exact structure (as in 2.8). For all ordinals α < σ we have Ext1 C(A)pure (Yα+1/Yα, X) = Ext1 dw(A)(Yα+1/Yα, X) ∼= HomK(A)(Yα+1/Yα, Σ1X) = 0 where the first equality holds because each degreewise pure extension of a complex with pure projective components is degreewise split exact, the isomorphism is ob- tained by Lemma 3.1 and the last equality follows by assumption. Hence Eklof's lemma [6, Lemma 1], in its version for exact categories [23, Prop. 2.12], gives the result. Now consider the case where Y has components in ⊥ FPI(A). Then from 2.5 we know that for all n ∈ Z there exists Jn such that Yn ⊕ Jn ∼= Fn, where Fn is fp(A)-filtered. Consider for each n the disc complex Dn(Jn) = 0 → Jn = Jn → 0, 6 GEORGIOS DALEZIOS which is concentrated in homological degrees n and n − 1. Then the complex Y ′ := Y ⊕ Mn∈Z Dn(Jn)! ⊕ Σ−1Y has components of the form Fn ⊕ Fn+1, and these are fp(A)-filtered. Then from the previous treated case we have that HomK(A)(Y ′, X) = 0, thus HomK(A)(Y, X) = 0 too. Finally, by what we have proved, if X is a pure acyclic complex with components in the class ⊥ FPI(A), then the identity map on X is null homotopic, in other words X is contractible. (cid:3) As a corollary we obtain a result on pure periodicity which extends the following fact: if M is a module fitting into a pure exact sequence 0 → M → P → M → 0 with P pure projective, then M is pure projective as well. In other words, every PProj(A)–pure periodic module is pure projective. This result was first proved by Simson [24, Thm. 1.3] and recently by Emmanouil [7, Cor. 3.6]. We point out that in [1] the authors provide a proof of this result and also a proof of the dual statement. Our version below extends the case of PProj(A)–pure periodicity to the case of ⊥ FPI(A)–pure periodicity. Corollary 3.3. (Compare with [7, 24]) Let A be a locally finitely presented Grothendieck category and let M be an object in A admitting a pure short exact sequence of the form 0 → M → F → M → 0 where F ∈ ⊥ FPI(A). Then M ∈ ⊥ FPI(A). In other words, any ⊥ FPI(A)–pure periodic object belongs to the class ⊥ FPI(A). Proof. The argument is identical as in [7, Cor. 3.8], but invoking 3.2. Namely, we may splice copies of the given short exact sequence to obtain a pure acyclic complex with components in ⊥ FPI(A), thus a contractible complex. Hence M is a summand of F ∈ ⊥ FPI(A) and the assertion follows since the class ⊥ FPI(A) is closed under summands. (cid:3) We now relate Proposition 3.2 with the theory of cotorsion pairs. Lemma 3.4. Let A be a locally finitely presented Grothendieck category and let C := C(FPI(A) ∩ ⊥ FPI(A)) and W := Cpac(FPI(A)). Then the following hold. (i) (C(⊥ FPI(A)), W) is a cotorsion pair in C(A). (ii) If A is locally coherent, then the cotorsion pair of (i) is complete. Moreover, in this case the pair (C, W) is a complete and hereditary cotorsion pair in C(FPI(A)). Proof. (i) Recall that by definition (⊥ FPI(A), FPI(A)) is a cotorsion pair which is generated by a set, therefore by 2.5 it is complete. Thus, from work of Gillespie [9, Prop. 3.6], there exists an induced cotorsion pair (⊥W, W) in the abelian category C(A) where the class ⊥W can be identified with ⊥W = {X ∈ C(⊥ FPI(A)) ∀W ∈ W, HomK(A)(X, W ) = 0}. Since every complex in W is pure acyclic, 3.2 implies that ⊥W = C(⊥ FPI(A)), which proves the claim. A NOTE ON HOMOTOPY CATEGORIES OF FP-INJECTIVES 7 (ii) Assume that A is locally coherent. In this case, from 2.7, we obtain that the complete cotorsion pair (⊥ FPI(A), FPI(A)) is also hereditary. Thus, [10, Cor. 3.7] implies that the cotorsion pair of (i) is complete. Now, we prove that (C, W) is a cotorsion pair in C(FPI(A)). Let C ∈ C and W ∈ W. Invoking Lemma 3.1 we have Ext1 C(FPI(A))(C, W ) = Ext1 dw(A)(C, W ) ∼= HomK(A)(C, Σ1W ). Since W is pure acyclic, from part (i) we obtain ⊥W = C and W ⊆ C ⊥. To prove the inclusion C ⊥ ⊆ W, let X ∈ C(FPI(A)) be such that, for all C ∈ C, Ext1 C(FPI(A))(C, X) = 0. We need to show that X ∈ W. Since the cotorsion pair (C(⊥ FPI(A)), W) is complete, there exists a short exact sequence X ֌ W ։ C with W ∈ W and C ∈ C(⊥ FPI(A)). Since A is locally coherent, from 2.7 we obtain that the complex C has components in FPI(A). By the assumption on X this short exact sequence splits, therefore the fact that W is closed under direct summands implies that X ∈ W. Completeness of the cotorsion pair (C, W) in C(FPI(A)) follows easily from the completeness of the cotorsion pair in (i). We show that (C, W) is hereditary. The category C, as a subcategory of C(FPI(A)), is easily seen to be closed under kernels of epimorphisms. To see that the class W is closed under cokernels of monomorphisms let 0 → A → B → C → 0 be a short exact sequence in C(FPI(A)) with A, B ∈ W. Then C is an exact complex and using the fact that in the coherent case FPI(A) is closed under cokernels of monomorphisms (2.7), we obtain that C has cycles in FPI(A), thus C ∈ W. (cid:3) Recall that if T is a triangulated category with set-indexed coproducts, an object S ∈ T is called compact if for any family {Xi}i∈I of objects in T , the natural map `i∈I HomT (Xi, S) → HomT (`i∈I Xi, S) is an isomorphism. T is called compactly generated if there exists a set S of compact objects in T , such that for any non-zero T ∈ T there exists a non-zero morphism S → T for some S ∈ S. Theorem 3.5. Let A be a locally finitely presented Grothendieck category. Then the homotopy category K(FPI(A) ∩ ⊥ FPI(A)) is compactly generated. Moreover, if A is locally coherent, the composite functor K(FPI(A) ∩ ⊥ FPI(A)) ֌ K(FPI(A)) can−−→ D(FPI(A)) is an equivalence of triangulated categories. Proof. Put C := FPI(A) ∩ ⊥ FPI(A). We will make use of [14, Thm. 3.1], which asserts that for any class of objects C which is closed under (set indexed) coprod- ucts and direct summands, the homotopy category K(C) is compactly generated, provided the following hold: (i) Every finitely presented object A has a right C-resolution [8, Dfn. 8.1.2], which by definition means that there exists a sequence 0 → A → C0 → C1 → · · · with Ci ∈ C which is exact after applying functors of the form HomA(−, C). (ii) Every pure exact sequence consisting of objects in C is split exact. Their result holds for modules over associative rings, but it is clear that it generalizes to our setup. To check condition (i), recall that the cotorsion pair 8 GEORGIOS DALEZIOS (⊥ FPI(A), FPI(A)) is complete, therefore for any A ∈ fp(A), we can construct an exact sequence C(A) := 0 → A ∂−1 −−→ C0 ∂0 −→ C1 ∂1 −→ C2 → · · · where ∂−1 is a (special) FP-injective preenvelope of A with cokernel ǫ0 : C0 ։ Z0 ∈ ⊥ FPI(A), ∂0 = d0 ◦ ǫ0 where d0 is an FP-injective envelope of Z0 with cokernel C1 ։ Z1 ∈ ⊥ FPI(A) etc. Since fp(A) is contained in ⊥ FPI(A) and the latter class is closed under extensions, we deduce that for all i = 0, 1, ...; Ci ∈ ⊥ FPI(A). The sequence constructed has all the Ci's in C and clearly is HomA(−, C)–exact, thus it is a right C–resolution of A. To check condition (ii), let C := · · · → Cn+1 → Cn → Cn−1 → · · · be a pure exact sequence consisting of objects in C. In particular, C is a pure acyclic complex with components in ⊥ FPI(A), hence by 3.2 it is contractible. Thus employing [14, Thm. 3.1] we obtain that K(C) is compactly generated by the set {ΣiC(A) A ∈ fp(A), i ∈ Z}. We now assume that A is locally coherent. Since K(C) is compactly generated and the inclusion j! : K(C) → K(FPI(A)) preserves coproducts (which exist because FPI(A) is closed under coproducts), by Neeman's Brown representability theorem [20, Thm. 4.1], the functor j! admits a right adjoint j∗ : K(FPI(A)) → K(C). The kernel of this right adjoint is ker(j∗) = {Y ∀X ∈ K(C), HomK(FPI(A))(X, Y ) = 0}, which by 3.4 (ii) is precisely the category Kpac(FPI(A)). Therefore, well known arguments (see for instance [21, Remark 2.12]) imply that j!−→ K(FPI(A)) can−−→ D(FPI(A)) is an equivalence of triangu- the composite K(C) lated categories and that the canonical map K(FPI(A)) → D(FPI(A)) is equivalent (up to natural isomorphism) with j∗. (cid:3) Remark 3.6. For any locally finitely presented Grothendieck category A, Krause in [17, Example 7] shows the existence of a left adjoint of the canonical map K(FPI(A)) → D(FPI(A)). In the proof of 3.5 above, we obtain such a left ad- joint after restricting ourselves to the case where A is locally coherent, and we identify its essential image with K(FPI(A) ∩ ⊥ FPI(A)). Before closing this note, we mention that our theorem 3.5 has an interpreta- tion in the language of (Quillen) model categories. By the work of Hovey [15] (resp. Gillespie [11]) we know that certain cotorsion pairs on an abelian (resp. ex- act) category A correspond bijectively to the so-called abelian (resp. exact) model structures on the category A. If A is a locally coherent Grothendieck category, it is not hard to see that the cotorsion pair on the category C(FPI(A)) we obtained in 3.4, corresponds (via the aforementioned Hovey–Gillespie theory) to an exact model structure on the category C(FPI(A)) with Quillen homotopy category D(FPI(A)). The precise statement is as follows. Theorem 3.7. Let A be a locally coherent Grothendieck category and let C(FPI(A)) denote the category of chain complexes with components FP-injective objects. Then there exists an (exact) model structure on C(FPI(A)), where - the cofibrant objects are the chain complexes in C(FPI(A) ∩ ⊥ FPI(A)). - every chain complex in C(FPI(A)) is fibrant. - the trivial objects are the pure acyclic complexes with FP-injective compo- nents. A NOTE ON HOMOTOPY CATEGORIES OF FP-INJECTIVES 9 The homotopy category of this model structure is equivalent to D(FPI(A)). Remark 3.8. Let A be a locally coherent Grothendieck category. Sťov´ıcek in [28, Thm. 6.12] shows the existence of a model structure with Quillen homotopy category D(FPI(A)) and also proves an equivalence D(FPI(A)) ∼= K(Inj(A)). In 3.7 we identify the cofibrant objects of this model structure with the category C(FPI(A) ∩ ⊥ FPI(A)). Thus, combining 3.7 with [28, Thm. 6.12] we obtain equiv- alences K(FPI(A) ∩ ⊥ FPI(A)) ∼= D(FPI(A)) ∼= K(Inj(A)). Acknowledgement The author would like to thank his PhD supervisors, Sergio Estrada from the University of Murcia and Henrik Holm from the University of Copenhagen. References 1. Silvana Bazzoni, Manuel Cort´es Izurdiaga, and Sergio Estrada, Periodic modules and acyclic complexes, preprint, 2017, https://arxiv.org/abs/1704.06672. 2. Siegfried Breitsprecher, Lokal endlich prasentierbare Grothendieck-Kategorien, Mitt. Math. Sem. Giessen Heft 85 (1970), 1–25. MR0262330 3. Theo Buhler, Exact categories, Expo. Math. 28 (2010), no. 1, 1–69. MR2606234 4. P.M. Cohn, On the free product of associative rings, Math. Z. 71 (1959), 380–398. MR0106918 5. William Crawley-Boevey, Locally finitely presented additive categories, Comm. Algebra 22 (1994), no. 5, 1641–1674. MR1264733 6. Paul C. Eklof and Jan Trlifaj, How to make Ext vanish, Bull. London Math. Soc. 33 (2001), no. 1, 41–51. MR1798574 7. Ioannis Emmanouil, On pure acyclic complexes, J. Algebra 465 (2016), 190–213. MR3537821 8. Edgar E. Enochs and Overtoun M. G. Jenda, Relative homological algebra, de Gruyter Expo- sitions in Mathematics, vol. 30, Walter de Gruyter & Co., Berlin, 2000. MR1753146 9. James Gillespie, The flat model structure on Ch(R), Trans. Amer. Math. Soc. 356 (2004), no. 8, 3369–3390. MR2052954 10. 11. , The flat model structure on complexes of sheaves, Trans. Amer. Math. Soc. 358 (2006), no. 7, 2855–2874. MR2216249 , Model structures on exact categories, J. Pure Appl. Algebra 215 (2011), no. 12, 2892–2902. MR2811572 12. Rudiger Gobel and Jan Trlifaj, Approximations and endomorphism algebras of modules, de Gruyter Expositions in Mathematics, vol. 41, Walter de Gruyter GmbH & Co. KG, Berlin, 2006. MR2251271 13. 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 14. Henrik Holm and Peter Jørgensen, Compactly generated homotopy categories, Homology, Ho- motopy Appl. 9 (2007), no. 1, 257–274. MR2280295 15. Mark Hovey, Cotorsion pairs, model category structures, and representation theory, Math. Z. 241 (2002), no. 3, 553–592. MR1938704 16. Srikanth Iyengar and Henning Krause, Acyclicity versus total acyclicity for complexes over Noetherian rings, Doc. Math. 11 (2006), 207–240. MR2262932 17. Henning Krause, Approximations and adjoints in homotopy categories, Math. Ann. 353 (2012), no. 3, 765–781. MR2923949 18. Daniel Murfet and Shokrollah Salarian, Totally acyclic complexes over Noetherian schemes, Adv. Math. 226 (2011), no. 2, 1096–1133. MR2737778 19. Amnon Neeman, The derived category of an exact category, J. Algebra 135 (1990), no. 2, 388–394. MR1080854 20. , The Grothendieck duality theorem via Bousfield's techniques and Brown repre- sentability, J. Amer. Math. Soc. 9 (1996), no. 1, 205–236. MR1308405 10 21. GEORGIOS DALEZIOS , The homotopy category of flat modules, and Grothendieck duality, Invent. Math. 174 (2008), no. 2, 255–308. MR2439608 22. Luigi Salce, Cotorsion theories for abelian groups, (1979), 11–32. MR565595 23. Manuel Saor´ın and Jan Sťov´ıcek, On exact categories and applications to triangulated adjoints and model structures, Adv. Math. 228 (2011), no. 2, 968–1007. MR2822215 24. Daniel Simson, Pure-periodic modules and a structure of pure-projective resolutions, Pacific J. Math. 207 (2002), no. 1, 235–256. MR1974474 25. , Flat complexes, pure periodicity and pure acyclic complexes, J. Algebra 480 (2017), 298–308. MR3633309 26. Bo Stenstrom, Coherent rings and F P -injective modules, J. London Math. Soc. (2) 2 (1970), 323–329. MR0258888 27. Jan Sťov´ıcek, Deconstructibility and the Hill lemma in Grothendieck categories, Forum Math. 25 (2013), no. 1, 193–219. MR3010854 28. , On purity and applications to coderived and singularity categories, preprint, 2014, https://arxiv.org/abs/1412.1615. Departamento de Matem´aticas, Universidad de Murcia, 30100 Murcia, Spain E-mail address: [email protected] Department of Mathematical Sciences, University of Copenhagen, Universitets- parken 5, 2100 Copenhagen Ø, Denmark
1704.02437	1	1704	2017-04-08T04:14:26	Classification of certain types of maximal matrix subalgebras	[ "math.RA" ]	Let $M_n(K)$ denote the algebra of $n \times n$ matrices over a field $K$ of characteristic zero. A nonunital subalgebra $N \subset M_n(K)$ will be called a nonunital intersection if $N$ is the intersection of two unital subalgebras of $M_n(K)$. Appealing to recent work of Agore, we show that for $n \ge 3$, the dimension (over $K$) of a nonunital intersection is at most $(n-1)(n-2)$, and we completely classify the nonunital intersections of maximum dimension $(n-1)(n-2)$. We also classify the unital subalgebras of maximum dimension properly contained in a parabolic subalgebra of maximum dimension in $M_n(K)$.	math.RA	math	Classification of certain types of maximal matrix subalgebras John Eggers Department of Mathematics University of California at San Diego La Jolla, CA 92093-0112 [email protected] Ron Evans Department of Mathematics University of California at San Diego La Jolla, CA 92093-0112 [email protected] Mark Van Veen Varasco LLC 2138 Edinburg Avenue Cardiff by the Sea, CA 92007 [email protected] 2010 Mathematics Subject Classification. 15B33, 16S50 Key words and phrases. matrix ring over a field, intersection of matrix subalgebras, nonunital intersections, subalgebras of maximum dimension, parabolic subalgebra, semi-simple Lie algebra, radical 1 Abstract Let Mn(K) denote the algebra of n × n matrices over a field K of characteristic zero. A nonunital subalgebra N ⊂ Mn(K) will be called a nonunital intersection if N is the intersection of two unital subalgebras of Mn(K). Appealing to recent work of Agore, we show that for n ≥ 3, the dimension (over K) of a nonunital intersection is at most (n − 1)(n − 2), and we completely classify the nonunital intersections of maximum dimension (n − 1)(n − 2). We also classify the unital subalgebras of maximum dimension properly contained in a parabolic subalgebra of maximum dimension in Mn(K). 1 Introduction Let Mn(F ) denote the algebra of n × n matrices over a field F . For some interesting sets Λ of subspaces S ⊂ Mn(F ), those S ∈ Λ of maximum dimension over F have been completely classified. For example, a theorem of Gerstenhaber and Serezhkin [6, Theorem 1] states that when Λ is the set of subspaces S ⊂ Mn(F ) for which every matrix in S is nilpotent, then each S ∈ Λ of maximum dimension is conjugate to the algebra of all strictly upper triangular matrices in Mn(F ). For another example, it is shown in [1, Prop. 2.5] that when Λ is the set of proper unital subalgebras S ⊂ Mn(F ) and F is an algebraically closed field of characteristic zero, then each S ∈ Λ of maximum dimension is a parabolic subalgebra of maximum dimension in Mn(F ). The goal of this paper is to classify the elements in Λ of maximum di- mension in the cases Λ = Γ and Λ = Ω, where the sets Γ and Ω are defined below. In Isaac's text [3, p. 161], every ring is required to have a unity, but the unity in a subring need not be the same as the unity in its parent ring. Under this definition, a ring may have subrings whose intersection is not a subring. This motivated us to study examples of pairs of unital subrings in Mn(K) whose intersection N is nonunital, where K is a field of characteristic zero. We call such N a nonunital intersection and we let Γ denote the set of all nonunital intersections N ⊂ Mn(K). Note that Γ is closed under transposition and conjugation, i.e., if N ∈ Γ, then N T ∈ Γ and S −1N S ∈ Γ for any invertible S ∈ Mn(K). In order to define Ω, we need to establish some notation. For brevity, write M = Mn = Mn(K). In the spirit of [2, p. viii], we define a subalgebra of 2 M to be a vector subspace of M over K closed under the multiplication of M (cf. [2, p. 2]); thus a subalgebra need not have a unity, and the unity of a unital subalgebra need not be a unity of the parent algebra. Subalgebras A, B ⊂ M are said to be similar if A = {S −1BS : B ∈ B} for some invertible S ∈ M. The notation M[Rn] will be used for the subalgebra of M consisting of those matrices whose n-th row is zero. Similarly, M[Rn, Cn] indicates that the n-th row and n-th column are zero, etc. For 1 ≤ i, j ≤ n, let Ei,j denote the elementary matrix in M with a single entry 1 in row i, column j, and 0 in each of the other n2 − 1 positions. The identity matrix in M will be denoted by I. For the maximal parabolic subalgebra P := M[Rn] + KEn,n in M, define Ω to be the set of proper subalgebras B of P with B 6= M[Rn]. We now describe Theorems 3.1 -- 3.3, our main results. Theorem 3.1 shows that dim N ≤ (n − 1)(n − 2) for each N ∈ Γ. Theorem 3.2 shows that up to similarity, W := M[Rn, Rn−1, Cn] and W T := M[Rn, Cn−1, Cn] are the only subalgebras in Γ having maximum dimension (n − 1)(n − 2). In Theorem 3.3, we show that dim B ≤ n2 − 2n + 3 for each B ∈ Ω, and we classify all B ∈ Ω of maximum dimension n2 − 2n + 3. The proofs of our theorems depend on four lemmas which are proved in Section 2. Lemma 2.1 shows that W (and hence also W T) is a nonunital intersection of dimension (n − 1)(n − 2) when n ≥ 3. Lemmas 2.2 and 2.3 show that dim L ≤ n(n − 1) for any nonunital subalgebra L ⊂ M, and when equality holds, L must be similar to M[Rn] or M[Cn]. (Thus if Λ denotes the set of nonunital subalgebras L ⊂ M, Lemmas 2.2 and 2.3 classify those L ∈ Λ of maximum dimension.) Lemma 2.4 shows that if U ⊂ M is a subalgebra with unity different from I, then some conjugate of U is contained in M[Rn, Cn]. 2 Lemmas Recall the definition W := M[Rn, Rn−1, Cn]. Lemma 2.1. For n ≥ 3, W ∈ Γ and dim W = (n − 1)(n − 2). Proof. For n > 1, define A ∈ M by A = I + En,n−1 . Note that A−1 = I − En,n−1 . A straightforward computation shows that for M ∈ M[Rn, Cn], the conjugate AM A−1 is obtained from M by replacing the (zero) bottom row of M by the (n − 1)-th row of M. Since the bottom two rows of AM A−1 3 are identical, it follows that AM A−1 ∈ M[Rn, Cn] ∩ AM[Rn, Cn]A−1 if and only if AM A−1 ∈ W. Since W = A−1WA, this shows that W is the intersection of the unital subalgebras A−1M[Rn, Cn]A and M[Rn, Cn]. To see that W is nonunital, note that E1,n−1 is a nonzero matrix in W for which E1,n−1W is the zero matrix for each W ∈ W; thus W cannot have a right identity, so W ∈ Γ. Finally, it follows from the definition of W that dim W = (n − 1)(n − 2). Remark: The same proof shows that W ∈ Γ holds when the field K is replaced by an arbitrary ring R with 1 6= 0. If moreover R happens to be commutative, then the dimension of the algebra W over R is well-defined [7, p. 483] and it equals (n − 1)(n − 2). Lemma 2.2. For any nonunital subalgebra L ⊂ M, dim L ≤ n(n − 1). Proof. It cannot happen that L+KI = M, otherwise L would be a two-sided proper ideal of M, contradicting the fact that M is a simple ring [7, p. 280]. Since L + KI is a proper subalgebra of M containing the unity I, it follows from Agore [1, Cor. 2.6] that dim L = −1 + dim (L + KI) ≤ n(n − 1). Lemma 2.3. Any nonunital subalgebra L ⊂ M with dim L = n(n − 1) must be similar to either M[Rn] or M[Cn] = M[Rn]T. Proof. Consider the two parabolic subalgebras P, P ′ ⊂ M defined by P = PK = M[Rn] + KEn,n , P ′ = P ′ K = M[C1] + KE1,1 . Note that P ′ is similar to the transpose P T. Since L + KI is a proper subalgebra of M of dimension n(n−1)+1, it follows from Agore [1, Prop. 2.5] that L + KI is similar to P or P ′, under the condition that K is algebraically closed. However, Nolan Wallach [8] has proved that this condition can be dropped; see the Appendix. Thus, replacing L by a conjugate if necessary, we may assume that L + KI = P or L + KI = P T. We will assume that L + KI = P, since the proof for P T is essentially the same. It suffices to show that L is similar to M[Rn] or M[C1], since M[C1] is similar to M[Cn]. Assume temporarily that each L ∈ L has all entries 0 in its upper left (n−1)×(n−1) corner. Then n = 2, because if n ≥ 3, then every matrix in P would have a zero entry in row 1, column 2, contradicting the definition of P. 4 Since L ⊂ M2[C1] and both sides have dimension 2, we have L = M2[C1], which proves the theorem under our temporary assumption. When the temporary assumption is false, there exists L ∈ L with the entry 1 in row i, column j for some fixed pair i, j with 1 ≤ i, j ≤ n − 1. Since Ei,i and Ej,j are in P = L + KI and L is a two-sided ideal of P, we have Ei,j = Ei,iLEj,j ∈ L. Consequently, Ea,b = Ea,iEi,jEj,b ∈ L for all pairs a, b with 1 ≤ a ≤ n − 1 and 1 ≤ b ≤ n. Therefore M[Rn] = n−1 n X X a=1 b=1 KEa,b ⊂ L, and since both M[Rn] and L have the same dimension n(n − 1), we conclude that L = M[Rn]. Remark: Any subalgebra B ⊂ M properly containing M[Rn] must also contain I. To see this, note that B contains a nonzero matrix of the form B := n X i=1 ciEn,i , ci ∈ K. If cj = 0 for all j < n, then En,n ∈ B, so I ∈ B. On the other hand, if cj 6= 0 for some j < n, then En,n = c−1 j BEj,n ∈ B, so again I ∈ B. Lemma 2.4. Suppose that a subalgebra U ⊂ M has a unity e 6= I. Then S −1U S ⊂ M[Rn, Cn] for some invertible S ∈ M. Proof. Let r be the rank of the matrix e. Note that e is idempotent, so by [5, p. 27], there exists an invertible S ∈ M for which S −1eS = Dr, where Dr is a diagonal matrix with entries 1 in rows 1 through r, and entries 0 elsewhere. Replacing U by S −1U S if necessary, we may assume that e = Dr. Since r ≤ n − 1, we have U = e U e ⊂ eMe = DrMDr ⊂ Dn−1MDn−1 = M[Rn, Cn]. 3 Theorems Recall that Γ is the set of all nonunital intersections in M. 5 Theorem 3.1. If N ∈ Γ, then dim N ≤ (n − 1)(n − 2). Proof. Let N ∈ Γ, so that N = U ∩ V for some pair of unital subalgebras U , V ⊂ M. Since N is nonunital, one of U , V, say U, does not contain I. Thus U contains a unity e 6= I. Define S as in Lemma 2.4. Replacing U, V, N by S −1U S, S −1VS, S −1N S, if necessary, we deduce from Lemma 2.4 that U is contained in M[Rn, Cn]. Since N is a nonunital subalgebra of U ⊂ M[Rn, Cn], it follows from Lemma 2.2 with (n − 1) in place of n that dim N ≤ (n − 1)(n − 2). Theorem 3.2. Let n ≥ 3. Then up to similarity, W and W T are the only subalgebras of M in Γ having dimension (n − 1)(n − 2). Proof. By Lemma 2.1, every subalgebra of M similar to W or W T lies in Γ and has dimension (n − 1)(n − 2). Conversely, let N ∈ Γ with dim N = (n − 1)(n − 2). We must show that N is similar to W or W T. We may assume, as in the proof of Theorem 3.1, that N is a nonunital subalgebra of M[Rn, Cn]. Let L be the subalgebra of Mn−1 consisting of those matrices in the upper left (n − 1) × (n − 1) corners of the matrices in N . Since dim L = dim N = (n − 1)(n − 2), it follows from Lemma 2.3 that L is similar to Mn−1[Rn−1] or Mn−1[Cn−1]. Thus N is similar to W = M[Rn, Rn−1, Cn] or W T = M[Rn, Cn−1, Cn]. Recall that Ω denotes the set of proper subalgebras B 6= M[Rn] in P. Theorem 3.3. Let B ∈ Ω. Then dim B ≤ n2 − 2n + 3. If B has maximum dimension n2 − 2n + 3, then B is similar to one of MEn,n + M[Rn, C1] + KE1,1, MEn,n + M[Rn, Rn−1] + KEn−1,n−1. Proof. Let e ∈ M denote the diagonal matrix of rank n − 1 with entry 0 in row n and entries 1 in the remaining rows. Because e is a left identity in M[Rn] and Be ⊂ M[Rn, Cn], it follows that Be is an algebra. First suppose that Be = M[Rn, Cn]. Then P = C + D, where C = B + KEn,n, D = M[Rn]En,n. We proceed to show that C ∩ D is zero. Assume for the purpose of contra- diction that there exists a nonzero matrix B ∈ C ∩ D. Then B ∈ B. We have BB = D, since the matrices in B have all possible submatrices in their 6 upper left (n − 1) by (n − 1) corners. Thus D ⊂ B ⊂ C, which implies that M[Rn] ⊂ B and P = C = B + KEn,n. If KEn,n ⊂ B, then B = P, and if KEn,n is not contained in B, then B = M[Rn]; either case contradicts the fact that B ∈ Ω. Since C ∩ D is zero, dim B ≤ dim C = dim P − dim D = (n2 − n + 1) − (n − 1) = n2 − 2n + 2. Thus dim B < n2 − 2n + 3, so the desired upper bound for dim B holds when Be = M[Rn, Cn]. Next suppose that Be is a proper subalgebra of M[Rn, Cn]. We proceed to show that by showing that d := dim Be ≤ (n − 1)(n − 2) + 1, dim L ≤ (n − 1)(n − 2) + 1 for every proper subalgebra L of Mn−1. If L is nonunital, then dim L ≤ (n − 1)(n − 2) < (n − 1)(n − 2) + 1 by Lemma 2.2 (with n − 1 in place of n). If L contains a unit different from the identity of Mn−1, then by Lemma 2.4 (with L in place of U), dim L ≤ dim M[Rn−1, Cn−1] = (n − 2)2 < (n − 1)(n − 2) + 1. If L contains the identity of Mn−1, then by Agore [1, Cor. 2.6], dim L ≤ (n − 1)(n − 2) + 1. This completes the demonstration that d ≤ (n − 1)(n − 2) + 1. Let B1e, B2e, . . . , Bde be a basis for Be, with Bi ∈ B. Since B is a subspace of the vector space spanned by the d + n matrices B1, . . . , Bd, E1,n, . . . , En,n, it follows that dim B ≤ d + n ≤ (n − 1)(n − 2) + 1 + n = n2 − 2n + 3. 7 Thus the desired upper bound for dim B holds in all cases. The argument above shows that when we have the equality dim B = d + n = (n − 1)(n − 2) + 1 + n = n2 − 2n + 3, then B = Be + MEn,n. Moreover, from the equality d = dim Be = (n − 1)(n − 2) + 1, it follows from Agore [1, Prop. 2.5] (again appealing to the Appendix to dispense with the condition of algebraic closure) that there is an invertible matrix S in the set M[Rn, Cn] + En,n such that S −1BeS is equal to one of M[Rn, Cn, C1] + KE1,1, M[Rn, Cn, Rn−1] + KEn−1,n−1. Since S −1MEn,nS = MEn,n, we achieve the desired classification of Ω. 4 Appendix Let F be a field of characteristic 0 with algebraic closure F . Given a proper subalgebra C ⊂ Mn(F ) of maximum dimension, Agore [1, Prop. 2.5, Cor. 2.6] proved that the ¯F -span of C is similar over ¯F to the ¯F -span of D for some parabolic subalgebra D of maximum dimension in Mn(F ). The purpose of this Appendix is to deduce that C is similar over F to D. Lemma 4.1. (Wallach) Let A be a subspace of Mn(F ) of dimension n − 1 such that A ⊗F F has basis of one of the following two forms: a) x1 ⊗ λ1, x2 ⊗ λ1, ..., xn−1 ⊗ λ1, with λ1 ∈ ( ¯F n)∗, xj ∈ ¯F n and λ1(xj) = 0, b) x1 ⊗ λ1, x1 ⊗ λ2, ..., x1 ⊗ λn−1, with λj ∈ ( ¯F n)∗, x1 ∈ ¯F n and λj(x1) = 0. Then in case a) A is F −conjugate (i.e. under GL(n, F )) to the span of the matrices Ei,n with i = 1, ..., n − 1, and in case b) A is F −conjugate to the span of the matrices En,i with i = 1, ..., n − 1. Proof. In either case, if X, Y ∈ A then XY = 0 and X has rank 1. For X of rank 1, we have XF n = F y for some y 6= 0. Thus there exists µ ∈ (F n)∗ with Xz = µ(z)y = (y ⊗ µ) (z) for all z. We conclude that A has a basis over F of the form Xi = yi ⊗ µi for i = 1, ..., n − 1. We now assume that case a) is true (the argument for the other case is essentially the same). In case a), there exists z ∈ ¯F n such that {X1(z), ..., Xn−1(z)} 8 is linearly independent over ¯F . This implies that µ1(z) · · · µn−1(z)y1 ∧ · · · ∧ yn−1 6= 0. Thus y1, ..., yn−1 are linearly independent. But 0 = XiXj = µi(yj)yi ⊗ µj. Thus µi(yj) = 0 for all j = 1, ..., n − 1. Let ν be a non-zero element of (F n)∗ such that ν(yi) = 0 for all i = 1, ..., n − 1. Then ν is unique up to non-zero scalar multiple. Thus yi ⊗ ν, i = 1, ..., n − 1 is an F -- basis of A. Clearly there exists g ∈ GL(n, F ) such that if e1, ..., en is the standard basis and ξ1, ..., ξn is the dual basis then gyi = ei and ν ◦ g = ξn. This completes the proof in case a). Proposition 4.2. (Wallach) Suppose that L ⊂ Mn(F ) is a subalgebra such that L ⊗F F is either: a) conjugate to the parabolic subalgebra P ¯F , b) conjugate to the parabolic subalgebra (P ¯F )T . In case a) L is F -- conjugate to PF . In case b) L is F -- conjugate to P T F . Proof. We just do case a) as case b) is proved in the same way. We look upon L as a Lie algebra over F . Then Levi's theorem [4, p. 91] implies that L = S ⊕ R with S a semi-simple Lie algebra and R the radical (the maximal solvable ideal). Thus L ⊗F F = S ⊗F F ⊕ R ⊗F F . Therefore R ⊗F F is the radical of L ⊗F F . If we conjugate L ⊗F F to P ¯F via h ∈ GL(n, ¯F ), then we see that h[R ⊗F F , R ⊗F F ]h−1 has basis Ei,n , i = 1, ..., n − 1. Thus hypothesis a) of Lemma 4.1 is satisfied for A = [R, R]. There exists therefore g ∈ GL(n, F ) such that gAg−1 has basis Ei,n , i = 1, ..., n − 1. Assume that we have replaced L with gLg−1. Then A has basis Ei,n , i = 1, ..., n − 1. Since [L, A] ⊂ A and PF is exactly the set of elements X of Mn(F ) such that [X, A] ⊂ A, we have L ⊂ PF . Thus L = PF , as both sides have the same dimension. Acknowledgment We are very grateful to Nolan Wallach for providing the Appendix and for many helpful suggestions. 9 References [1] [2] [3] [4] [5] [6] [7] [8] A. L. Agore, The maximal dimension of unital subalgebras of the matrix algebra, Forum Math. 29 (2017), 1 -- 5. B. J. Gardner and R. Wiegandt, Radical Theory of Rings, M. Dekker, NY, 2004. I. M. Isaacs, Algebra: A Graduate Course, Brooks/Cole, Pacific Grove, CA, 1994. N. Jacobson, Lie Algebras, Courier Dover Publications, NY, 1979. K. O'Meara, J. Clark, and C. Vinsonhaler, Advanced Topics in Linear Algebra, Oxford University Press, 2011. C. S. Pazzis, On Gerstenhaber's theorem for spaces of nilpotent matrices over a skew field, Linear Algebra Appl. 438 (2013), 4426 -- 4438. J. Rotman, Advanced Modern Algebra: Third Edition, Part I, AMS, Providence, RI, 2015. N. Wallach, email dated 2/20/2017. 10
1601.07388	1	1601	2016-01-26T04:00:51	A Lie conformal algebra of Block type	[ "math.RA", "math-ph", "math-ph", "math.QA", "math.RT" ]	The aim of this paper is to study a Lie conformal algebra of Block type. In this paper, conformal derivation, conformal module of rank 1 and low-dimensional comohology of the Lie conformal algebra of Block type are studied. Also, the vertex Poisson algebra structure associated with the Lie conformal algebra of Block type is constructed.	math.RA	math	1 A Lie conformal algebra of Block type Lamei Yuan Academy of Fundamental and Interdisciplinary Science, Harbin Institute of Technology, Harbin 150080, China [email protected] Abstract: The aim of this paper is to study a Lie conformal algebra of Block type. In this paper, conformal derivation, conformal module of rank 1 and low-dimensional comohology of the Lie conformal algebra of Block type are studied. Also, the vertex Poisson algebra structure associated with the Lie conformal algebra of Block type is constructed. Keywords: Lie conformal algebra, vertex Lie algebra, cohomology, vertex Poisson algebra MR(2000) Subject Classification: 17B65, 17B69 1 Introduction The notion of Lie conformal algebra, introduced by Kac [9], encode an axiomatic description of the operator product expansions of chiral fields in conformal field theory. It is a pow- erful tool for the study of infinite-dimensional Lie (super)algebras, associative algebras and their representations. Lie conformal algebras have been extensively studied, including the classification problem [5, 6], cohomology theory [2, 12] and representation theory [3]. The Lie conformal algebras are closely related to vertex algebras. Primc [11] introduced and studied a notion of vertex Lie algebra, which is a special case of a more general notion of local vertex Lie algebra [4]. As it was explained in [10], the notion of Lie conformal algebra and the notion of vertex Lie algebra are equivalent. In this paper, we shall use Lie conformal algebra and vertex Lie algebra synonymously. With the notion of vertex Lie algebra, one arrives at the notion of vertex Poisson algebra, which is a combination of a differential algebra structure and a vertex Lie algebra structure, satisfying a natural compatibility condition. The symmetric algebra of a vertex Lie algebra is naturally a vertex Poisson algebra [7]. A general construction theorem of vertex Poisson algebras was given in [10]. Applications of vertex Poisson algebras to the theory of integrable systems were studied in [1]. In the present paper, we study a nonsimple Lie conformal algebra of infinite rank, which is endowed with a C[∂]-basis {Jii ∈ Z+}, such that [Ji λJj] = ((i + 1)∂ + (i + j + 2)λ)Ji+j, for i, j ∈ Z+. (1.1) The corresponding formal distribution Lie algebra is a Block type Lie algebra, which is the associated graded Lie algebra of the filtered Lie algebra W1+∞ [13, 14, 15, 16, 18]. Thus we call this Lie conformal algebra a Lie conformal algebra of Block type and denote it by B in this paper. It is a conformal subalgebra of gr gc1 studied in [17]. In addition, it contains the Virasoro conformal algebra Vir = C[∂]J0 with [J0 λJ0] = (∂ + 2λ)J0 as a subalgebra. 2 The paper is organized as follows. In Section 2, we recall the notions of Lie conformal algebra and vertex Lie algebra. In Section 3, we study conformal derivations of the Lie conformal algebra of Block type B. In Section 4, we recall the notions of conformal module and comohology of Lie conformal algebras. Then we study conformal module of rank 1 and low-dimensional comohology of B with coefficients in B-modules. In Section 5, we equip a vertex Lie algebra structure (Y−, ∂) with B and establish an association of a vertex Poisson algebra structure to the vertex Lie algebra (B, Y−, ∂). 2 Preliminaries Throughout this paper, all vector spaces and tensor products are over the complex field C. We use notations Z for the set of integers and Z+ for the set of nonnegative integers. Definition 2.1. A Lie conformal algebra R is a C[∂]-module with a C-bilinear map, called the λ-bracket, and satisfying the following axioms (a, b, c ∈ R), R ⊗ R → C[λ] ⊗ R, a ⊗ b 7→ [aλb], (conformal sesquilinearity) [∂aλb] = −λ[aλb], [aλ∂b] = (∂ + λ)[aλb], (skew-symmetry) [aλb] = −[b−λ−∂a], (Jacobi identity) [aλ[bµc]] = [[aλb]λ+µc] + [bµ[aλc]]. If we consider the expansion [aλb] = Pj∈Z+ λj j! (a(j)b), (2.1) (2.2) (2.3) (2.4) the coefficients of λj and the axioms (2.1) -- (2.3) can be written in terms of them as follows: j! are called the j-product satisfying a(n)b = 0 for n sufficiently large, ∂a(n)b = −na(n−1)b, a(n)∂b = ∂(a(n)b) + na(n−1)b, a(n)b = − Pi∈Z+ (−1)n+i 1 i! ∂ib(n+i)a, a(m)b(n)c = b(n)a(m)c + m Pi=0(cid:0)m i(cid:1)(a(i)b)(m+n−i)c. (2.5) (2.6) (2.7) In terms of generating functions, Mirko Primc in [11] presented an equivalent definition of a Lie conformal algebra under the name of vertex Lie algebra (see also [10]). Let V be any vector space. Following [11], for a formal series u(m1, · · · , mn)x−m1−1 · · · x−mn−1 ∈ V [[x±1 1 , · · · , x±1 n ]], 1 n f (x1, · · · , xn) = Pm1,··· ,mn∈Z we set Singf (x1, · · · , xn) = Pm1,··· ,mn∈Z+ Clearly, for 1 ≤ i ≤ n, u(m1, · · · , mn)x−m1−1 1 · · · x−mn−1 n . (2.8) ∂ ∂xi Singf (x1, · · · , xn) = Sing ∂ ∂xi f (x1, · · · , xn). (2.9) Definition 2.2. A vertex Lie algebra is a vector space A equipped with a linear operator ∂ called the derivation and a linear map 3 Y−(·, z) : A → z−1(End A)[[z−1]], a 7→ Y−(a, z) = Pn≥0a(n)z−n−1, satisfying the following conditions for a, b ∈ A, n ∈ Z+: a(n)b = 0 for n sufficiently large, [∂, Y−(a, z)] = Y−(∂a, z) = d dz Y−(a, z), and the half Jacobi identity holds: Y−(a, z)b = Sing(cid:0)ez∂Y−(b, −z)a(cid:1), Sing(cid:16)z−1 0 δ( z1 − z2 z0 )Y−(a, z1)Y−(b, z2) − z−1 = Sing(cid:16)z−1 z1 − z0 2 δ( )Y−(Y−(a, z0)b, z2)(cid:17). z2 0 δ( z2 − z1 −z0 )Y−(b, z1)Y−(a, z2)(cid:17) (2.10) (2.11) (2.12) (2.13) Relation (2.12) is called the half skew-symmetry. It was shown in [11] that the half Jacobi identity (2.13) amounts to the following half commutator formula: Y−(a, z1)Y−(b, z2) − Y−(b, z1)Y−(a, z2) = Sing(cid:16)Pi∈Z+(z1 − z2)−i−1Y−(a(i)b, z2)(cid:17). (2.14) As it was explained in [10, Remark 2.6], the notion of Lie conformal algebra is equivalent to the notion of vertex Lie algebra. We often denote a vertex Lie algebra by (A, Y−, ∂) and refer to (Y−, ∂) as the vertex Lie algebra structure. A vertex Lie algebra (A, Y−, ∂) is said to be free, if A is a free C[∂]-module over a vector space V , namely, A = C[∂]V ∼= C[∂] ⊗C V. 3 Conformal derivation Let C denote the ring C[∂] of polynomials in the indeterminate ∂. Definition 3.1. Let V and W be two C -modules. A linear map φ : V → C [λ] ⊗C W , denoted by φλ : V → W , is called a conformal linear map, if φλ(∂v) = (∂ + λ)(φλv), for v ∈ V. (3.1) The space of conformal linear maps between C -modules V and W is denoted by Chom(V, W ) and it can be made into an C -module via (∂φ)λv = −λφλv, for v ∈ V. Definition 3.2. Let R be a Lie conformal algebra. A conformal linear map dλ : R → R is called a conformal derivation of R if dλ[aµb] = [(dλa)λ+µb] + [aµ(dλb)], for a, b ∈ R. (3.2) 4 The space of all conformal derivations of R is denoted by CDer(R). For any a ∈ R, one can define a conformal derivation (ad a)λ : R → R by (ad a)λb = [aλb] for b ∈ R. Such conformal derivation is called inner. Denote by CInn(R) the space of all conformal inner derivations of R. Proposition 3.3. Every conformal derivation of the Lie conformal algebra B is inner. Proof. Let d be any conformal derivation of B. Denote L = J0. Assume that there exists j=1fij (∂, λ)Jij , where fij (∂, λ) ∈ C[∂, λ]. Condition (3.2) requires dλ([LµL]) = [Lµ(dλL)] + [(dλL)λ+µL]. This is equivalent to a finite subset I = {i1, · · · , in} ⊆ Z+ such that dλL = Pn n n (∂ + λ + 2µ) n fij (∂, λ)Jij − Pj=1 (∂ + (ij + 2)µ)fij (∂ + µ, λ)Jij Pj=1 = ((ij + 1)∂ + (ij + 2)(λ + µ))fij (−λ − µ, λ)Jij . Pj=1 (3.3) For each j, (∂ + λ + 2µ)fij (∂, λ) − (∂ + (i + 2)µ)fij (∂ + µ, λ) = ((ij + 1)∂ + (ij + 2)(λ + µ))fij (−λ − µ, λ). (3.4) Write fij (λ, ∂) = Pm k=0 aij ,k(λ)∂k with aij ,m(λ) 6= 0. Then, assuming m > 1, if we equate terms of degree m in ∂, we have (λ − i − mµ)aij ,m(λ) = 0 and thus aij ,m(λ) = 0. This contra- dicts aij ,m(λ) 6= 0. Thus deg∂fij (λ, ∂) ≤ 1, and fij (λ, ∂) = aij ,0(λ) + aij ,1(λ)∂. Substituting it into (3.4) gives aij ,0(λ) = ij+2 ij+1λaij ,1(λ). Therefore, dλL = aij ,1(λ) ij + 1 n Pj=1 ((ij + 1)∂ + (ij + 2)λ)Jij . Replacing dλ by dλ − (ad h)λ with h = Pn j=1 For k > 0, assume that dλJk = Pl and using dλ(L) = 0, we obtain aij ,1(−∂) ij +1 Jij , we get dλ(L) = 0. i=1 fki(∂, λ)Jki. Applying dλ to [LµJk] = (∂ +(k+2)µ)Jk l l (∂ + λ + (k + 2)µ) fki(∂, λ)Jki = Pi=1 (∂ + (ki + 2)µ)fki(∂ + µ, λ)Jki, Pi=1 (3.5) and thus (∂ + λ + (k + 2)µ)fki(∂, λ) = (∂ + (ki + 2)µ)fki(∂ + µ, λ), for 1 ≤ i ≤ l. (3.6) Comparing the highest degree of λ gives fki(∂, λ) = 0 for 1 ≤ i ≤ l. Hence dλ(Jk) = 0 for k > 0. This concludes the proof. (cid:3) 4 Cohomology Definition 4.1. A module M over a Lie conformal algebra R is a C[∂]-module endowed with a bilinear map R ⊗ M → M[[λ]], a ⊗ v 7→ aλv such that (a, b ∈ R, v ∈ M) aλ(bµv) − bµ(aλv) = [aλb]λ+µv, (∂a)λv = −λaλv, aλ(∂v) = (∂ + λ)aλv, 5 (4.1) (4.2) If aλv ∈ M[λ] for all a ∈ R, v ∈ M, then the R-module M is said to be conformal. If M is finitely generated as C[∂]-module, then M is simply called finite. Since we only consider conformal modules, we will simply shorten the term "conformal module" to "module". The one-dimensional vector space C can be viewed as a module (called the trivial module) over any conformal algebra R with both the action of ∂ and the action of R being zero. In addition, for a fixed nonzero complex constant a, there is a natural C[∂]-module Ca, which is the one-dimensional vector space C such that ∂v = av for v ∈ Ca. Then Ca becomes an R-module on which all elements of R act by zero. For the Virasoro conformal algebra Vir, it was proved in [3] that all free nontrivial Vir- modules of rank 1 are the following ones (∆, α ∈ C): M∆,α = C[∂]v, Lλv = (∂ + α + ∆λ)v. (4.3) The module M∆,α is irreducible if and only if ∆ 6= 0, the module M0,α contains a unique nontrivial submodule (∂ + α)M0,α isomorphic to M1,α, and the modules M∆,α with ∆ 6= 0 exhaust all finite irreducible nontrivial conformal Vir-modules. Proposition 4.2. All free nontrivial B-modules of rank 1 are as follows (∆, α ∈ C): M∆,α = C[∂]v, J0λv = (∂ + α + ∆λ)v, Ji λv = 0, for i > 0. Proof. By (4.3), J0 λv = (∂ + α + ∆λ)v for some ∆, α ∈ C. By [17, Lemma 5.1], we can suppose that k is the smallest nonnegative integer such that Jk λv 6= 0, Jk+1 λv = 0. Assume k > 0 and write Jk λv = g(λ, ∂)v, for some g(λ, ∂) ∈ C[λ, ∂]. Since [Jk λJk]λ+µv = 0, g(λ, ∂)g(µ, λ + ∂) = g(µ, ∂)g(λ, µ + ∂). (4.4) This implies degλg(λ, ∂) + deg∂g(λ, ∂) = degλg(λ, ∂). Thus deg∂g(λ, ∂) = 0. Then we have g(λ, ∂) = g(λ) for some g(λ) ∈ C[λ]. The fact that [J0 λJk]λ+µv = ((k + 1)λ − µ)Jkλ+µv yields ((k + 1)λ − µ)g(λ + µ)v = (∂ + α + ∆λ)g(µ)v − (∂ + µ + α + ∆λ)g(µ)v = −µg(µ)v, which gives g(µ) = 0. Hence, Jk λv = 0, a contradiction. Thus k = 0 and J1 λv = 0. It follows immediately that Ji λv = 0 for all i ≥ 1. (cid:3) In the following we study cohomology of the Lie conformal algebra B with coefficients in B-modules C, Ca and M∆,α, respectively. For completeness, we shall present the definition of cohomology of Lie conformal algebras given in [2]. Definition 4.3. An n-cochain (n ∈ Z+) of a Lie conformal algebra R with coefficients in an R-module M is a C-linear map γ : R ⊗n → M[λ1, · · · , λn], a1 ⊗ · · · ⊗ an 7→ γλ1,··· ,λn(a1, · · · , an) satisfying 6 (1) γλ1,··· ,λn(a1, · · · , ∂ai, · · · , an) = −λiγλ1,··· ,λn(a1, · · · , an) (conformal antilinearity), (2) γ is skew-symmetric with respect to simultaneous permutations of ai's and λi's, namely, γλ1,··· ,λi−1,λi+1,λi,λi+2,··· ,λn(a1, · · · , ai−1, ai+1, ai, αi+2, · · · , an) = −γλ1,··· ,λi−1,λi,λi+1,λi+2,··· ,λn(a1, · · · , ai−1, ai, ai+1, ai+2, · · · , an). (4.5) As usual, let R ⊗0 = C, so that a 0-cochain is an element of M. Denote by C n(R, M) the set of all n-cochains. The differential d of an n-cochain γ is defined by (dγ)λ1,··· ,λn+1(a1, · · · , an+1) (−1)i+1ai λiγλ1,··· , λi,··· ,λn+1(a1, · · · , ai, · · · , an+1) n+1 = + n+1 Pi=1 Pi,j=1;i<j (−1)i+jγλi+λj ,λ1,··· , λi,··· , λj ,··· ,λn+1(cid:0)[ai λiaj], a1, · · · , ai, · · · , aj, · · · , an+1(cid:1), (4.6) where γ is extended linearly over the polynomials in λi. In particular, if γ ∈ M is a 0-cochain, then (dγ)λ(a) = aλγ. It is proved in [2] that the operator d preserves the space of cochains and d2 = 0. Thus the cochains of a Lie conformal algebra R with coefficients in R-module M form a complex, which is called the basic complex and will be denoted by Moreover, define a C[∂]-module structure on C •(R, M) by C •(R, M) = Ln∈Z+ C n(R, M). (∂γ)λ1,··· ,λn(a1, · · · , an) = (∂M + n λi)γλ1,··· ,λn(a1, · · · , an), Pi=1 (4.7) where ∂M denotes the action of ∂ on M. Then d∂ = ∂d and thus ∂ C •(R, M) ⊂ C •(R, M) forms a subcomplex. The quotient complex C •(R, M) = C •(R, M)/∂ C •(R, M) = Ln∈Z+ C n(R, M) is called the reduced complex. Definition 4.4. The basis cohomology H •(R, M) of a Lie conformal algebra R with coeffi- cients in R-module M is the cohomology of the basis complex C •(R, M) and the (reduced) cohomology H •(R, M) is the cohomology of the reduced complex C •(R, M). Remark 4.5. The basic cohomology H •(R, M) is naturally a C[∂]-module, whereas the reduced cohomology H •(R, M) is a complex vector space. For a q-cochain γ ∈ C q(R, M), we call γ a q-cocycle if d(γ) = 0; a q-coboundary or a trivial q-cocycle if there is a (q − 1)-cochain φ ∈ C q−1(R, M) such that γ = d(φ). Two cochains γ1 and γ2 are called equivalent if γ1 − γ2 is a coboundary. Denote by Dq(R, M) and Bq(R, M) the spaces of q-cocycles and q-boundaries, respectively. By Definition 4.4, Hq(R, M) = Dq(R, M)/ Bq(R, M) = {equivalent classes of q-cocycles}. The main results of this section are the following theorem. 7 Theorem 4.6. For the Lie conformal algebra B, the following statements hold. (1) For the trivial module C, we have and q dim H (B, C) = (cid:26) 1 if q = 0, 0 if q = 1, or 2, dim Hq(B, C) = (cid:26) 1 if q = 0, or 2, 0 if q = 1. (2) If a 6= 0, then dim H•(B, Ca) = 0. (3) If α 6= 0, then dim H•(B, M∆,α) = 0. Proof. (1) Since a 0-cochain γ is an element of C and (dγ)λ(a) = aλγ = 0 for a ∈ B, we have D0(B, C) = C 0(B, C) = C and B0(B, C) = 0. Thus H0(B, C) = D0(B, C)/ B0(B, C) = C, and H0(B, C) = C because ∂ C 0(B, C) = ∂C = 0. Let γ ∈ C 1(B, C) and dγ ∈ ∂ C 2(B, C), namely, there is φ ∈ C 2(B, C) such that d(γ) = ∂φ. By (4.6), (4.7) and ∂C = 0, γλ1+λ2([aλ1b]) = −(dγ)λ1,λ2(a, b) = −(∂φ)λ1,λ2(a, b) = −(λ1 + λ2)φλ1,λ2(a, b), a, b ∈ B,(4.8) By (1.1), (4.8) and Definition 4.3 (1), ((i + 1)λ1 − λ2)γλ1+λ2(Ji) = −(λ1 + λ2)φλ1,λ2(J0, Ji), i ≥ 0. (4.9) Setting λ = λ1 + λ2 in (4.9) gives ((i + 1)λ − (i + 2)λ2)γλ(Ji) = −λφλ1,λ2(L, Ji), i ≥ 0, which implies that γλ(Ji) is divisible by λ. We can define a 1-cochain γ′ ∈ C 1(B, C) by λ(Ji) = λ−1γλ(Ji), for i ≥ 0. γ′ (4.10) Since ∂C = 0, γ = ∂γ′ ∈ ∂ C 1(B, C). Hence H1(B, C) = 0. If γ is a 1-cocycle, namely, φ = 0 in (4.8), then (4.9) gives γ = 0. Thus H1(B, C) = 0. Let ψ ∈ D2(B, C) be a 2-cocycle. We have 0 = (dψ)λ1,λ2,λ3(Ji, J0, J0)λ3=0 = −(λ1 − (i + 1)λ2)ψλ1+λ2,λ3(Ji, J0)λ3=0 + (λ1 − (i + 1)λ3)ψλ1+λ3,λ2(Ji, J0)λ3=0 −(λ2 − λ3)ψλ2+λ3,λ1(J0, Ji)λ3=0 = −((λ1 − (i + 1)λ2)ψλ1+λ2,0(Ji, J0) + (λ1 + λ2)ψλ1,λ2(Ji, J0). (4.11) Setting λ = λ1 + λ2 in (4.11) gives ((λ − (i + 2)λ2)ψλ,0(Ji, J0) = λψλ1,λ2(Ji, J0). Thus ψλ,0(Ji, J0) is divisible by λ. Define a 1-cochain f by fλ1(Ji) = λ−1 1 ψλ1,λ(Ji, J0)λ=0, for i ≥ 0. Set γ = ψ + df , which is also a 2-cocycle. For all i ≥ 0, γλ1,λ(Ji, J0)λ=0 = ψλ1,λ(Ji, J0)λ=0 − λ1fλ1(Ji) = 0. (4.12) i=1 λi, namely, it is divided by Pq γ′ λ1,··· ,λq(a1, · · · , aq) = (Pq i=1λi. Thus we get a q-cochain i=1λi)−1γλ1,··· ,λq (a1, · · · , aq), (4.15) 8 By (4.12) and (4.11) with γ in place of ψ, we have (λ1 + λ2)γλ1,λ2(Ji, J0) = 0. Therefore γλ1,λ2(Ji, J0) = 0 = γλ1,λ2(Ji, J0). With this, 0 = (dγ)λ1,λ2,λ(J0, Ji, Jk)λ=0 = −γλ1+λ2,λ([J0 λ1Ji], Jk)λ=0 + γλ1+λ,λ2([J0 λ1Jk], Ji)λ=0 − γλ2+λ,λ1([Ji λ2Jk], J0)λ=0 = −((i + 1)λ1 − λ2)γλ1+λ2,0(Ji, Jk) − (k + 1)λ1γλ2,λ1(Ji, Jk). (4.13) Setting λ1 = 0 in (4.13) gives γλ2,0(Ji, Jk) = 0 and thus γλ2,λ1(Ji, Jk) = 0. This proves γ = 0. Hence H2(B, C) = 0. It remains to compute H2(B, C). Following [12], we define a linear map σγ : B⊗q → C[λ1, · · · , λq−1] for q ≥ 2 by (σγ)( a1 ⊗ · · · ⊗ an) = γλ1,··· ,λq(a1, · · · , aq)λq=−λ1−···−λq−1, a1, · · · , aq ∈ B. (4.14) We define σγ = γ if q = 0 and σγ(a1) = γλ1(a1)λ1=0 if q = 1. Set C ′q(B, C) = {σγγ ∈ C q(B, C)}. Obviously, σ : C q(B, C) → C ′q(B, C) is a surjective map. If γ ∈ ∂ C q(B, C) = i=1λi) C q(B, C), then σγ = 0. That is, σ factors to a map σ : C q(B, C) → C ′q(B, C). We claim that σ : C q(B, C) → C ′q(B, C) is an isomorphism as vector spaces. In fact, if σγ = 0 for a q-cochain γ, then, by (4.14), γλ1,··· ,λq(a1, · · · , aq) as a polynomial in λq has a (Pq root λq = −Pq−1 and γ = (Pq i=1λi)γ′ ∈ ∂ C q(B, C), which proves that σ is injective. Hence the claim is true. In the following we can identify C q(B, C) with C ′q(B, C). We still call an element in C ′q(B, C) a reduced q-cochain. By defining the operator d′ : C ′q(B, C) → C ′q+1(B, C) by d′(σγ) = σdγ, we have similar notions of reduced q-cocycle and q-coboundary. For convenience, we will abbreviate γλ1,··· ,λq (a1, · · · , aq)λq=−λ1−···−λq−1 to γλ1,··· ,λq−1(a1, · · · , aq). Let ψ′ = σψ ∈ C ′2(B, C) be a reduced 2-cochain. By (4.6) and (4.14), (dψ′)λ1,λ2(a1, a2, a3) = −ψ′ λ1+λ2([a1 λ1a2], a3) + ψ′ −λ2([a1 λ1a3], a2) − ψ′ −λ1([a2 λ2a3], a1), for a1, a2, a3 ∈ B. Define a reduce 1-cochain f ′ = σf ∈ C ′1(B, C) by f ′(Ji) = (i + 2)−1 d dλ ψ′ λ(Ji, J0)λ=0, for i ≥ 0. (4.16) (4.17) Note that f ′(a) = fλ(a)λ=0. Thus f ′ is simply a linear function from B to C, satisfying f ′(∂a) = fλ(∂a)λ=0 = −λfλ(a)λ=0 = 0, and (df ′)λ(a1, a2) = −f ′([a1 λa2]), for a1, a2 ∈ B. (4.18) If ψ′ is a reduced 2-cocycle, then γ′ = ψ′ + df ′ is a reduced 2-cocycle, equivalent to ψ′. By (4.17) and (4.18), d dλ γ′ λ(Ji, J0)λ=0 = 0, for i ≥ 0. (4.19) This, along with (4.16) and (4.5), gives 9 ∂ ∂λ ∂ 0 = = = (dγ′)λ1,λ(Ji, Jk, J0)λ=−λ1 ∂ λ1+λ([Ji λ1Jk], J0) + γ′ ∂λ(cid:0) − γ′ ∂λ(cid:0)((i + 1)(λ + λ1) + λ1)γ′ = (i + k + 3)γ′ λ1(Ji, Jk) − λ1 −λ([Ji λ1J0], Jk) − γ′ −λ1([Jk λJ0], Ji)(cid:1)λ=−λ1 −λ(Ji, Jk) − ((k + 1)(λ + λ1) + λ)γ′ ∂ ∂λ1 γ′ λ1(Ji, Jk). −λ1(Jk, Ji)(cid:1)λ=−λ1. (4.20) (4.21) (4.22) Thus, λ(Ji, Jk) = ci,kλi+k+3, for some ci,k ∈ C. γ′ By (4.16) with γ′ in place of ψ′ and (4.21), 0 = −γ′ λ1+λ2([Ji λ1Jj], Jk) + γ′ −λ2([Ji λ1Jk], Jj) − γ′ = −((j + 1)λ1 − (i + 1)λ2)ci+j,k(λ1 + λ2)i+j+k+3 +((i + 1)λ2 + (i + k + 2)λ1)ci+k,j(−λ2)i+j+k+3 −((j + 1)λ1 + (j + k + 2)λ2)cj+k,i(−λ1)i+j+k+3. −λ1([Jj λ2Jk], Ji) Taking i = j = 0 in (4.22), we get c0,k(λ1 − λ2)(λ1 + λ2)k+3 = ck,0(cid:0)(λ2 + (k + 2)λ1)(−λ2)k+3 − (λ1 + (k + 2)λ2)(−λ1)k+3(cid:1).(4.23) Setting λ2 = 0 in (4.23) gives c0,k = (−1)kck,0. Comparing coefficients of λ2 in (4.23) gives c0,k = ck,0 = 0 for k ≥ 1. Setting i = 0 in (4.22) and comparing coefficients of λj+k+4 , we obtain cj,k = 0 for j, k ≥ 1. Thus, by (4.21), there exists a nonzero complex number c, such that 1λk+2 2 1 λ(J0, J0) = cλ3, γ′ γ′ λ(Ji, Jj) = 0, for i, j ≥ 1. (4.24) Therefore, dim H2(B, C) = 1, which proves (1). (2) Define an operator τ : C q(B, Ca) → C q−1(B, Ca) by (τ γ)λ1,··· ,λq−1(a1, · · · , aq−1) = (−1)q−1γλ1,··· ,λq−1,λ(a1, · · · , aq−1, J0)λ=0, (4.25) for a1, · · · , aq−1 ∈ B. By the fact that ∂ C q(B, Ca) = (a +Pq i=1 λi) C q(B, Ca) and (4.25), ((dτ + τ d)γ)λ1,··· ,λq(Jn1, · · · , Jnq) = (cid:0)Pq i=1λi(cid:1)γλ1,··· ,λq (Jn1, · · · , Jnq) ≡ −aγλ1,··· ,λq(Jn1, · · · , Jnq) (mod ∂ C q(B, Ca).(4.26) Suppose that γ is a q-cochain such that dγ ∈ ∂ C q+1(B, Ca), namely, there is a (q + 1)- i=1 λi)τ φ ∈ ∂ C q(B, Ca). By(4.26), γ ≡ −d(a−1τ γ) is a reduced coboundary because a 6= 0. Thus Hq(B, Ca) = 0 for q ≥ 0. This proves (2). cochain φ such that dγ = (a +Pq+1 i=1 λi)φ. By (4.25), τ dγ = (a +Pq 10 (3) Note that ∂ C q(B, M∆,α) = (∂ +Pq we define an operator κ : C q(B, M∆,α) → C q−1(B, M∆,α) by i=1 λi) C q(B, M∆,α). Similarly to the proof of (2), (κγ)λ1,··· ,λq−1(a1, · · · , aq−1) = (−1)q−1γλ1,··· ,λq−1,λ(a1, · · · , aq−1, J0)λ=0, for a1, · · · , aq−1 ∈ B. By Theorem 4.2, ((dκ + κd)γ)λ1,··· ,λq(Jn1, · · · , Jnq) q = J0 λγλ1,··· ,λq (Jn1, · · · , Jnq)λ=0 +(cid:0) λi(cid:1)γλ1,··· ,λq(Jn1, · · · , Jnq) Pi=1 λi(cid:1)γλ1,··· ,λq(Jn1, · · · , Jnq) = (cid:0)∂ + α + Pi=1 ≡ αγλ1,··· ,λq (Jn1, · · · , Jnq) (mod ∂ C q(B, M∆,α)). q (4.27) Pq+1 i=1 λi)ϕ. In this case, κdγ = (∂ +Pq If γ is a reduced q-cocycle, then there is a (q + 1)-cochain ϕ such that dγ = ∂ϕ = (∂ + i=1 λi)κφ ∈ ∂ C q(B, M∆,α). It follows from (4.27) that γ ≡ d(α−1κγ) is a reduced q-coboundary, since we assume α 6= 0. Hence Hq(B, M∆,α) = 0 for q ≥ 0. This completes the proof of Theorem 4.6. (cid:3) Corollary 4.7. There is a unique nontrivial universal central extension B = B ⊕ Cc of the Lie conformal algebra B, satisfying [J0 λJ0] = (∂ + 2λ)J0 + λ3 [Ji λJj] = ((i + 1)∂ + (i + j + 2)λ)Ji+j, for i, j > 0. c, Remark 4.8. The formal distribution Lie algebra corresponding to B is a well-known Lie algebra of Block type studied in [16, 18]. 5 Vertex Poisson algebra structure associated to B Denote V = Li∈Z+ CJi. Thus the Lie conformal algebra of Block type B is a free C[∂]- module over V . By (2.4), the λ-bracket (1.1) is equivalent to the following j-products Ji (0)Jk = (i + 1)∂Ji+k, Ji (1)Jk = (i + k + 2)Ji+k, Ji (n)Jk = 0, (5.1) for i, k ∈ Z+, n ≥ 2. Define a linear map Y−(·, z) from V to z−1(End V )[[z−1]] by Y−(Ji, z) = Pn∈Z+ Ji (n)z−n−1, for i ≥ 0. (5.2) From (5.1) and (5.2), we have Y−(Ji, z)Jk = (i + 1)∂Ji+kz−1 + (i + k + 2)Ji+kz−2, for i, k ∈ Z+. (5.3) Extending the map Y−(·, z) to the whole B = C[∂] ⊗C V by Y−(f (∂)Ji, z)(cid:0)∂mJk(cid:1) = f (d/dz) (−1)l∂m−l(d/dz)lY−(Ji, z)Jk, m Pl=0 (5.4) then (B, Y−, ∂) forms a free vertex Lie algebra. Indeed, (5.3) and (5.4) guarantee axioms (2.10) and (2.11). It suffices to check (2.12) and (2.14) on the generators. For i, j ∈ Z+, Sing(ez∂Y−(Jj, −z)Ji) = (j + 1)∂Ji+j(−z)−1 + (i + j + 2)Ji+jz−2 + (i + j + 2)∂Ji+jz−1 11 = (i + 1)∂Ji+jz−1 + (i + j + 2)Ji+jz−2 = Y−(Ji, z)Jj, which proves (2.12). Furthermore, for any i, j, k ∈ Z+, Y−(Ji, z1)Y−(Jj, z2)Jk = (j + 1)(i + 1)∂2Ji+j+kz−1 1 z−1 +2(j + 1)(i + j + k + 2)Ji+j+kz−3 1 z−1 +(j + k + 2)(i + j + k + 2)Ji+j+kz−2 2 + (j + 1)(2i + j + k + 3)∂Ji+j+kz−2 1 z−1 2 2 + (i + 1)(j + k + 2)∂Ji+j+kz−1 1 z−2 2 1 z−2 2 , and (5.5) (5.6) Sing(cid:0)Pn≥0(z1 − z2)−n−1Y−(Ji (n)Jj, z2)(cid:1)Jk = Sing(cid:0)(z1 − z2)−1Y−(Ji (0)Jj, z2) + (z1 − z2)−2Y−(J i(1)Jj, z2)(cid:1)Jk = Sing(cid:0)(i + 1)(z1 − z2)−1Y−(∂Ji+j, z2) + (i + j + 2)(z1 − z2)−2Y−(Ji+j, z2)(cid:1)Jk 2 + z−2 1 z−2 = −(i + 1)(i + j + 1)∂Ji+j+k(z−1 −2(i + 1)(i + j + k + 2)Ji+j+k(z−1 1 z−3 +(i + j + 2)(i + j + k + 2)Ji+j+k(z−2 2 ) + (i + j + 2)(i + j + 1)∂Ji+j+kz−1 1 z−2 2 + 2z−3 = Y−(Ji, z1)Y−(Jj, z2)Jk − Y−(Jj, z1)Y−(Ji, z2)Jk, 1 z−1 2 + z−2 1 z−2 2 + z−3 1 z−1 2 ) 1 z−1 2 ) 1 z−2 2 (5.7) where the last equality follows from (5.6). This proves (2.14) and thus (B, Y−, ∂) is a vertex Lie algebra, called the vertex Lie algebra of Block type. Remark 5.1. By [8], the vertex Lie algebra (B, Y−, ∂) is equivalent to a Novikov algebra (V, ◦) with the operation ◦ defined by Ji ◦ Jj = (j + 1)Ji+j, for i, j ≥ 0. Moreover, the commutator on (V, ◦) of the form [Ji, Jj] = (j − i)Ji+j, for i, j ≥ 0, (5.8) (5.9) make V into a Lie algebra. The Lie algebra (V, [·, ·]) is a subalgebra of the one-sided Witt algebra W + 1 with a basis {Jii ≥ −1}, which occurs in the study of conformal field theory. By a differential algebra, we mean a commutative associative algebra A (with 1) equipped with a derivation ∂, denoted by (A, ∂). A subset U of A is said to generate A as a differential algebra if ∂nU for n ≥ 0 generate A as an algebra. The following notion of vertex Poisson algebra is due to [7]. Definition 5.2. A vertex Poisson algebra is a differential algebra (A, ∂) equipped with a vertex Lie algebra structure (Y−, ∂) such that for a, b, c ∈ A, Y−(a, z)(bc) = (Y−(a, z)b)c + b(Y−(a, z)c), (5.10) where Y−(a, z) = Pn≥0anz−n−1. In terms of components, relation (5.10) is equivalent to an(bc) = (anb)c + b(anc) for a, b, c ∈ A, n ≥ 0, namely, these an are derivations of A. Therefore, Y−(a, z) ∈ z−1( DerA)[[z−1]], for a ∈ A, 12 (5.11) (5.12) which implies Y−(a, z)1 = 0. Then Y−(1, z)a = 0 by (2.12), namely, Y−(1, z) = 0. Let U be a vector space. Denote by A = S(C[∂] ⊗ U) the symmetric algebra over the free C[∂]-module C[∂] ⊗ U. The operator ∂ can be uniquely extended to a derivation of A. Then (A, ∂) forms a differential algebra, which is referred to as the free differential algebra over U. Following [10], a week pre-vertex Poisson structure on A is a linear map Y 0 − from U × U to z−1A[z−1] such that Y 0 −(a, z)b = Sing(ez∂Y 0 −(b, −z)a), for a, b ∈ A. (5.13) By [10, Proposition 3.10], the operator Y 0 − can be uniquely extended to a linear map Y− : A → Hom(A, z−1A[z−1]), a 7→ Y−(a, z) = Pn∈Z+ anz−n−1 such that for a, b ∈ A, Y−(a, z) = Pn≥0anz−n−1 ∈ z−1(DerA)[[z−1]], [∂, Y−(a, z)] = Y−(∂a, z) = Y−(a, z), d dz Y−(a, z)b = Sing(cid:0)ez∂Y−(b, −z)a(cid:1). (5.14) (5.15) (5.16) The following result, due to [10, Theorem 3.11], gives a general construction of vertex Poisson algebras from free differential algebras. Theorem 5.3. Let A be the free differential algebra over U and let Y 0 − be a week pre-vertex Poisson structure on A such that for u, v, w ∈ U from an ordered basis of U with u ≤ v ≤ w, Y 0 −(u, z1)Y 0 −(v, z2)w − Y 0 −(v, z2)Y 0(u, z1)w = Sing(cid:0)ez2∂ Y 0 −(w, −z2)Y 0 −(u, z1 − z2)v(cid:1), (5.17) where Y 0 −(u, z) ∈ z−1(Der)[[z−1]] is uniquely determined by Y 0 −(u, z)ez1∂v = ez1∂e−z1 d dz Y 0 −(u, z)v, for v ∈ U . (5.18) Then Y 0 − uniquely extends to a vertex Poisson structure Y− on A. The following result is an application of the above theorem. Proposition 5.4. Let (B, Y−, ∂) be the vertex Lie algebra of Block type and A = S(B) the free differential algebra over V = Li∈Z+ CJi. Define −(Ji, z)Jj = (i + 1)∂Ji+jz−1 + (i + j + 2)Ji+jz−2, for i, j ≥ 0. Y 0 (5.19) Then Y 0 − uniquely extends to a vertex Poisson algebra structure on A. Proof. By (5.3) and (5.19), Y 0 − satisfies (5.13) and so it is a week pre-vertex Poisson structure on A. By (5.18) and (5.19), for i, j, k ∈ Z+, −(J i, z)J j = Y−(J i, z)J j for i, j ∈ Z+. Thus Y 0 Y 0 −(Ji, z1)Y 0 −(Jj, z2)Jk 13 = (j + 1)(i + 1)∂2Ji+j+kz−1 1 z−1 1 z−1 +2(j + 1)(i + j + k + 2)Ji+j+kz−3 +(j + k + 2)(i + j + k + 2)Ji+j+kz−2 2 + (j + 1)(2i + j + k + 3)∂Ji+j+kz−2 1 z−1 2 1 z−2 2 + (i + 1)(j + k + 2)∂Ji+j+kz−1 2 1 z−2 2 . (5.20) By (5.18) -- (5.20), −(Ji, z1 − z2)Jj(cid:1) Sing(cid:0)ez2∂ Y 0 −(Jk, −z2)Y 0 = Sing(cid:0)ez2∂Y 0 −(Jk, −z2)(cid:0)(i + 1)∂Ji+j(z1 − z2)−1 + (i + j + 2)Ji+j(z1 − z2)−2(cid:1)(cid:1) = Sing(cid:0)(i + 1)Y−(∂Ji+j, z2)Jk(z1 − z2)−1 + (i + j + 2)Y−(Ji+j, z2)Jk(z1 − z2)−2(cid:1) = Sing(cid:0) − (i + 1)(cid:0)(i + j + 1)∂Ji+j+kz−2 2 Ji+j+k(cid:1)(z1 − z2)−1 2 + (i + j + k + 2)z−2 +(i + j + 2)(cid:0)(i + j + 1)∂Ji+j+kz−1 2 Ji+j+k(cid:1)(z1 − z2)−2(cid:1) 2 ) + (i + j + 2)(i + j + 1)∂Ji+j+kz−1 1 + z−2 2 z−1 = −(i + 1)(i + j + 1)∂Ji+j+k(z−2 2 z−3 2 z−1 −2(i + 1)(i + j + k + 2)Ji+j+k(z−3 2 z−2 1 ) 1 + 2z−3 +(i + j + 2)(i + j + k + 2)Ji+j+k(z−2 −(Li, z1)Y 0 −(Lj, z2)Y 0 −(Li, z1)Lk. 1 z−1 1 + z−2 2 z−2 2 + 2(i + j + k + 2)z−3 1 + z−1 1 z−1 2 ) 2 z−2 1 = Y 0 −(Lj, z2)Lk − Y 0 This proves the result by Theorem 5.3. (cid:3) Acknowledgements This work was supported by National Natural Science Foundation grants of China (11301109) and the Research Fund for the Doctoral Program of Higher Education (20132302120042). References [1] Barakat A., De Sole A., Kac V., Poisson vertex algebras in the theory of Hamiltonian equations, Japan. J. Math., 4 (2009) 141 -- 252. [2] Bakalov B., Kac V., Voronov A., Cohomology of conformal algebras, Comm. Math. Phys., 200 (1999) 561 -- 598. [3] Cheng S.-J., Kac V.G., Conformal modules, Asian J. Math., 1997, 1(1):181 -- 193. [4] C. Dong, H.-S. Li, G. Mason, Vertex Lie algebra, vertex Poisson algebra and vertex operator algebras, in Recent Development in Infinite-Dimensional Lie Algebras and Conformal Field Theory, Proc. of an International Conference, May 23 -- 27, 2000, University of Virginia, eds. by S. Berman, P. Fendley, Y.-Z. Huang, K. Misra and B. Parshall, Contemporary Math. 297, Amer. Math. Soc., 2002, 69 -- 96. [5] D'Andrea A., Kac V.G., Structure theory of finite conformal algebras, Sel. Math., New Ser., 1998, 4:377 -- 418. [6] Fattori D., Kac V., Classification of finite simple Lie conformal superalgebras, J. Algebra, 258(1) (2002) 23 -- 59. [7] E. Frenkel, D. Ben-Zvi, Vertex algebras and algebraic curves, Mathematical Surveys and Mono- graphs, Vol. 88, Amer. Math. Soc., 2001. [8] I. M. Gel'fand , I. Ya. Dorfman, Hamiltonian operators and algebraic structures related to them, Functs. Anal. i Prilozhen., 13(4)(1979), 13 -- 30. 14 [9] V.G. Kac, Vertex Algebras for Beginners, 2nd ed. University Lecture Series Vol. 10 (American Mathematical Society, 1998). [10] Haisheng Li, Vertex algebras and vertex Piosson algebras, Commun. Contemp. Math., 6(1) (2004), 61 -- 110. [11] M. Primc, Vertex algebras generated by Lie algebras, J. Pure Appl. Algebra 135 (1999), 253 -- 293. [12] Su Y., Low dimensional cohomology of general conformal algebras gcN , J. Math. Phys., 45 (2004) 509 -- 524. [13] Y. Su, Quasifinite representations of a Lie algebra of Block type, J. Algebra 276 (2004) 117 -- 128. [14] Y. Su, Quasifinite representations of a family of Lie algebra of Block type, J. Pure Appl. Algebra 192 (2004) 293 -- 305. [15] Y. Su, C. Xia, Y. Xu, Quasifinite representations of a class of Block type Lie algebra B(q), J. Pure Appl. Algebra 216 (2012) 923 -- 934. [16] Y. Su, C. Xia, Y. Xu, Classification of quasifinite representations of a Lie algebra related to Block type, J. Algebra, 393 (2013) 71 -- 78. [17] Su Yucai, Yue Xiaoqing, Filtered Lie conformal algebras whose associated graded algebras are isomorphic to that of general conformal algebra gc1, J. Algebra, 2011, 340:182 -- 198. [18] Wang Q., Tan S., Quasifinite modules of a Lie algebra related to Block type, J. Pure Appl. Algebra, 2007, 211:596 -- 608.
1512.03570	1	1512	2015-12-11T10:00:49	Nontriviality results for the characteristic algebra of a DGA	[ "math.RA", "math.SG" ]	Assume that we are given a semifree noncommutative differential graded algebra (DGA for short) whose differential respects an action filtration. We show that the canonical unital algebra map from the homology of the DGA to its characteristic algebra, i.e. the quotient of the underlying algebra by the two-sided ideal generated by the boundaries, is a monomorphism. The main tool that we use is the weak division algorithm in free noncommutative algebras due to P. Cohn.	math.RA	math	NONTRIVIALITY RESULTS FOR THE CHARACTERISTIC ALGEBRA OF A DGA GEORGIOS DIMITROGLOU RIZELL c e D 1 1 ] . A R h t a m [ 1 v 0 7 5 3 0 . 2 1 5 1 : v i X r a Abstract. Assume that we are given a semifree noncommutative differential graded algebra (DGA for short) whose differential respects an action filtration. We show that the canonical unital algebra map from the homology of the DGA to its characteristic algebra, i.e. the quotient of the underlying algebra by the two-sided ideal generated by the boundaries, is a monomorphism. The main tool that we use is the weak division algorithm in free noncommutative algebras due to P. Cohn. 1. Introduction Differential graded algebras (DGAs for short) appear in algebraic topology. For example, they appear in the formulation of rational homotopy theory in [Sul] due to D. Sullivan. They also appear in modern symplectic and contact topology, such as in Legendrian contact homology by Y. Chekanov [Che] as well as in the more general theory of symplectic field theory by Y. Eliashberg, H. Hofer, and A. Givental [EGH]. We will be interested in DGAs that naturally appear in the latter setting, of which we give a very rough outline below in Section 1.3. This is also the setting in which all applications known to the author can be found. However, we emphasise that the results in this paper are purely algebraic. We will be mainly interested in DGAs that are finitely generated and semifree, and thus in particular fully noncommutative, as they appear in the geometric con- text of Legendrian contact homology. These DGAs will also be equipped with an action filtration which is respected by the differential, which naturally appears in the latter context. We refer to Section 3 for the precise algebraic definitions. Although easily described, it is in general not an easy problem to distinguish two DGAs. In the geometric context of Legendrian contact homology considered here, one is in particular interested in the stable-tame isomorphism class of the DGA (see Section 3.2). In [Ng] L. Ng introduced the characteristic algebra as a tool to study DGAs under this relation. The question that we will given an answer to here is: to what extent does the characteristic algebra remember the homology algebra of the DGA? In addition, we briefly discuss acyclic DGAs in Section 2. 1.1. Main results. In the following we consider a finitely generated semifree, thus fully noncommutative, DGA (A, ∂) over a field k which is of the form considered in Section 3, together with its associated characteristic algebra C := A/A∂(A)A. In The author is supported by the grant KAW 2013.0321 from the Knut and Alice Wallenberg Foundation. Part of this work was done during a visit of the author to the Institut Mittag-Leffler (Djursholm, Sweden). 1 2 GEORGIOS DIMITROGLOU RIZELL particular we assume that the differential ∂ respects an action filtration as postu- lated by condition (F) in the same section. For short, we say that (A, ∂) is a DGA with action filtration. Our main result in this setting is as follows. Theorem 1.1. Consider a DGA (A, ∂) with action filtration as above, whose cor- responding characteristic algebra is denoted by C. The natural unital algebra mor- phism induced by the inclusion of the cycles is a monomorphism. H(A, ∂) → C The proof of the main theorem relies heavily on the fact that the DGA has an action filtration respected by the differential. It is not clear to the author if this condition can be omitted. As an immediate consequence, we obtain the following useful result. Corollary 1.2. A DGA with action filtration is acyclic if and only if its charac- teristic algebra is trivial. We note that, in its full generality, [DR, Theorem 1.6] by the author depends on this result, which was stated as [DR, Lemma 3.1] in the same article, but appeared there without a proof. 1.2. Results in the super-commutative case. In the following we will investi- gate what can be said in the case of a super-commutative DGA. In other words, a DGA as defined in Section 3, but where we have imposed the commutativity relation a · b = (−1)abb · a. Clearly, any element b of odd degree satisfies b2 = 0 unless the ground field k is of characteristic two. In general, the statement analogous to Theorem 1.1 is not satisfied for a super- commutative DGA; see Example 1.5 below. However, it is not difficult to establish the following result which is analogous to Corollary 1.2. Proposition 1.3. Let (A, ∂) be a super-commutative DGA over the ground field k for which either: (1) The field satisfies char k 6= 2, and the grading is taken in the group Z/2Z; or (2) The field satisfies char k = 2, and the grading is arbitrary. It follows that the characteristic algebra A/A∂(A)A is trivial if and only if the DGA is acyclic. In the first case, the commutative unital algebra defined as the quotient C/Codd, where Codd ⊂ C denotes the two-sided ideal generated by the elements in odd degree is, moreover, trivial if and only if the DGA is acyclic. Proof. Suppose that the characteristic algebra is trivial, which implies that 1 = x1∂(y1) + . . . + xn∂(yn) for elements xi, yi ∈ A, i = 1, . . . , n, which all may be assumed to be of homoge- neous degrees. (1): Recall that the differential ∂ is of degree −1, that 1 = 0, and that (−1)xixi∂(yi) + (−1)yi(xi−1)yi∂(xi) = ∂(xiyi). NONTRIVIALITY OF THE CHARACTERISTIC ALGEBRA 3 Using these relations, may write 1 = u1∂(v1) + . . . + un∂(vn) + ∂(w), for elements ui, wi ∈ A satisfying ui = ∂(wi) = 1 ∈ Z/2Z, i = 1, . . . , n. 1 − ∂(w) = u1∂(v1) + . . . + un∂(vn) The equation implies that (1 − ∂(w))2 = (u1∂(v1) + . . . + un∂(vn))2 = 0 is satisfied by degree reasons, from which it follows that 1 = ∂(2w − w∂(w)). Hence, (A, ∂) is acyclic as sought. In the same manner, we also see that 1 ∈ A is contained in the two-sided ideal generated by the odd elements if and only if (A, ∂) is acyclic. (2): In the case when char k = 2 we can square both sides of the first equation above, giving rise to the relation 1 = (x1∂(y1) + . . . + xn∂(yn))2 = x2 1∂(y1)2 + . . . + x2 n∂(yn)2. Since every square is a cycle in this characteristic, i.e. ∂(x2) = x∂(x) + ∂(x)x = 0 by commutativity, the Leibniz rule again implies that 1 = ∂(x2 1y1∂(y1)) + . . . + ∂(x2 nyn∂(yn)) is a boundary. (cid:3) Corollary 1.4. Given that (A, ∂) is of one of the forms as prescribed by Proposition 1.3, it follows that there exists an augmentation into a field F ⊃ k, i.e. a unital DGA morphism ε : (A, ∂) → (F, 0) considered as a unital DGA with an empty generating set, if and only if (A, ∂) is not acyclic. Proof. The existence of an augmentation is equivalent to the existence of a unital algebra map C → F from the characteristic algebra to a field. Such a map exists since C admits a unital algebra map to a commutative unital algebra in both of the cases covered by Proposition 1.3. (cid:3) Example 1.5. Consider the super-commutative DGA (A = hb, b1, b2, ci, ∂) over Q generated by one generator c of degree c = −1, and three generators b1, b2, b of degree b1 = b2 = b = 1. We prescribe the relations ∂(b1) = bc, ∂(b2) = bc, while the other generators are cycles. It immediately follows that ∂2 = 0. Moreover, ∂(b1)b2 = bcb2 is a cycle which is not a boundary, but whose image inside the characteristic algebra C = A/A∂(A)A clearly vanishes. 4 GEORGIOS DIMITROGLOU RIZELL In order to see that bcb2 is not a boundary, we argue as follows. First, observe that any word of length 5 or more automatically vanishes in this DGA. Second, we consider the computations ∂(b1b2) = bcb2 − b1bc, ∂(bibi) = 0, ∂(bib) = 0, ∂(bic) = 0, ∂(bc) = 0, ∂(bibjb) = 0, ∂(bibjc) = 0, ∂(bibjbc) = 0, for any i, j ∈ {1, 2}. 1.3. A brief introduction to the geometric setup in which our DGAs arise. The Chekanov-Eliashberg algebra associated to a Legendrian knot of a con- tact manifold, as introduced independently in [Che] by Chekanov and [EGH] by Eliashberg-Givental-Hofer, is in the basic geometric setup a natural example of a finitely generated semifree DGA. Given a Legendrian submanifold, the theory as- sociates to it the so-called Chekanov-Eliashberg algebra (A, ∂) over a field k. The differential ∂ is defined by a count of associated rigid pseudoholomorphic polygons. The homotopy type of this DGA has been shown to be a powerful Legendrian iso- topy invariant. An important algebraic feature of this DGA, due to this geometric setup, is that the differential respects an action filtration. This will later turn out to be important. The basic case where the technical details of Legendrian contact homology has been carried out is that for a Legendrian submanifold Λ ⊂ (P × R, dz + θ) of a contactisation of a 2n-dimensional Liouville manifold (P, dθ); see [Che] and [EES3]. A Liouville manifold is a particular exact symplectic manifold (P, dθ), where the latter is a pair consisting of a non-compact smooth 2n-dimensional manifold P together with an exact two-form dθ satisfying the property that dθ∧n is a volume form on P . Recall that a submanifold of a (2n + 1)-dimensional contactisation as above is Legendrian given that it is of dimension n, and that the contact one-form dz + θ vanishes along it. Example 1.6. The standard contact (2n + 1)-space (R2n+1, dz − (y1dx1 + . . . + yndxn)) is the archetypal example, as well as a local model, of a contact manifold. The study of Legendrian submanifolds up to Legendrian isotopy, i.e. smooth iso- topy through Legendrian submanifolds, has been shown to be a both subtle and rich field; see e.g. [Che], [Ng], [EES2], and [Siv], among others. The Chekanov- Eliashberg algebra is the main invariant used, which is considerably more powerful than classical topological invariants such as the rotation number and Thurston- Bennequin invariant. More precisely, for a Legendrian submanifold of a contact manifold as above, the DGA-homotopy type, and even the so-called stable-tame isomorphism type, of the Chekanov-Eliashberg algebra is invariant under Legen- drian isotopy [EES3]. Since a semifree DGA typically is an infinite-dimensional noncommutative alge- bra, it is in general not easily studied. For that reason one usually tries to derive NONTRIVIALITY OF THE CHARACTERISTIC ALGEBRA 5 finite-dimensional invariants from it. For instance, given an augmentation of the DGA, i.e. a unital DGA morphism (A, ∂) → (k, 0), one can use Chekanov's lin- earisation procedure in [Che] to produce a finite-dimensional complex. The set of isomorphism classes of the homologies of all linearisations is an invariant of the DGA up to DGA homotopy. The above linearisation procedure produces computable Legendrian isotopy in- variants. However, far from all interesting Legendrian submanifolds have Chekanov- Eliashberg algebras admitting augmentations. On one hand, as shown by Proposi- tion 2.1, there is a unique acyclic Chekanov-Eliashberg algebra up to stable tame isomorphism. On the other hand, being non-acyclic is a necessary but clearly not a sufficient condition for admitting an augmentation. In [Ng] NG introduced the so-called characteristic algebra associated to a DGA (A, ∂), namely the quotient C := A/A∂(A)A under the two-sided ideal gen- erated by the boundaries. In the same article, he also successfully used this algebra in order to obtain invariants from the DGA in cases when there are no augmenta- tions. Augmentations clearly factorise through the characteristic algebra, which can be considered as the "universal" (not necessarily commutative) augmentation (A, ∂) → (C, 0). The first important question that now arises is under what conditions the characteristic algebra is nontrivial, which is what we study here. For more details and applications concerning augmentations in the characteristic algebra, we refer to [DR] and [DRG] due to the author and the author together with R. Golovko, respectively. We also refer to [Siv] for computations of Chekanov- Eliashberg algebras of interesting Legendrian knots due to S. Sivek. For instance, examples of Legendrian knots are produced for which the characteristic algebra admits a two-dimensional but no one-dimensional representation, as well as knots for which the characteristic algebra admits no finite-dimensional representations. 2. Results concerning acyclic DGAs First we show the basic result that the stable-tame isomorphism class of an acyclic DGA does not contain any interesting information. Proposition 2.1. Consider two DGAs, where each DGA has a generator whose boundary is equal to the unit. Two such acyclic DGAs are tame isomorphic if and only if there is a degree-preserving bijection between their generators. Proof. Let b be a generator for which ∂(b) = 1, and let a be any other generator. The elementary automorphism Φ defined by a 7→ a + b∂(a) satisfies the property that Φ ◦ ∂ ◦ Φ−1(a) = 0. After applying a suitable tame isomorphism, we may thus assume that all generators except b are cycles in both DGAs considered. (cid:3) Corollary 2.2. Two acyclic DGAs (Ai = ha1, . . . , aki , b1, . . . , blii, ∂i), i = 0, 1, with generators aj and bj in even and odd degrees, respectively, are stable-tame isomorphic if and only if l1 − k1 = l0 − k0. Proof. Consider an acyclic DGA (A, ∂) and choose an element x ∈ A of degree 1 satisfying ∂(x) = 1. After taking the free product with a stabilisation (S1, ∂S1 ) in degree 1 (see Section 3.2), we find generators a and b of degrees 1 and 2, respectively, for which ∂(a) = 0. Observe that such a free product does not affect the difference 6 GEORGIOS DIMITROGLOU RIZELL between the number of odd and even generators. After the elementary automor- phism Φ determined by a 7→ a − x, the differential satisfies 1 = Φ ◦ ∂ ◦ Φ−1(a) for the new generator a. Applying this argument to the DGAs (Ai, ∂i), i = 0, 1, we may assume that they both have a generator whose boundary is equal to 1 ∈ Ai. The above proposition combined with Lemma 2.3 finally implies the existence of the sought stable-tame isomorphism. (cid:3) The following lemma is standard. Lemma 2.3. Two DGAs (Ai = ha1, . . . , aki , b1, . . . , blii, ∂i), i = 0, 1, with genera- tors aj and bj in even and odd degrees, respectively, are stable-tame isomorphic as graded algebras (i.e. forgetting the differential) if and only if l1 − k1 = l0 − k0. On the other hand, there are examples of acyclic DGAs that are not isomorphic, but whose generators can be identified by a bijection which preserves the grading. We produce such an example just out of curiosity, and note that it is irrelevant for the application of invariants of Legendrian submanifolds. Namely, in this set- ting it is the stable-tame isomorphism class of the DGA that contains invariant information. Example 2.4. Consider the following two DGAs whose underlying graded algebra is given by A = ha1, a2, b1, b2i, where the elements a1, a2 are of degree a1 = a2 = 0, while the elements b1, b2 are of degree b1 = b2 = 1. (1) The first DGA has differential ∂0 defined by prescribing ∂0(a1) = 0, ∂0(a2) = 0, ∂0(b1) = 1, ∂0(b2) = 0, which clearly yields an acyclic DGA. (2) The second DGA has differential ∂ defined by prescribing ∂(a1) = 0, ∂(a2) = 0, ∂(b1) = 1 + a1a2, ∂(b2) = a2 1. Since ∂(b1 − a1b1a2 + b2a2 2) = 1, it follows that this DGA is acyclic as well. Proposition 2.5. The two semifree DGAs described in Example 2.4 are not iso- morphic. Proof. The first DGA has the property that the elements in ∂−1 geneous degree 1 all can be written as 0 (1) ⊂ A of homo- x = b1 + uib2vi + mXi=1 xib1yi, nXi=1 NONTRIVIALITY OF THE CHARACTERISTIC ALGEBRA 7 for arbitrary elements ui, vi ∈ ha1, a2i ⊂ A, i = 1, . . . , m, together with elements i=1 xiyi = 0. Consider the quotient xi, yi ∈ ha1, a2i, i = 1, . . . , n, satisfying Pm projection φ(r1,r2) : A → A/AR(r1,r2)A, R(r1,r2) := {a1 − r1, a2 − r2} ⊂ A, (r1, r2) ∈ k2, of algebras together with the canonical identification ψ(r1,r2) : A/AR(r1,r2)A ≃−→ hb1, b2i. It follows that for any pair (r1, r2) ∈ k2 and element x ∈ ∂−1 0 (1), the subset {ψ(r1,r2) ◦ φ(r1,r2)(x), b2} ⊂ hb1, b2i is a free generating set of the algebra hb1, b2i. Namely, ψ(r1,r2) ◦ φ(r1,r2)(x) = ψ(r1,r2) ◦ φ(r1,r2) b1 + uivib2! mXi=1 by the above. On the other hand, for any fixed element s1b1 + s2b2 ∈ hb1, b2i, si ∈ k, of degree 1, we can readily find a pair (r1, r2) ∈ k2 for which {ψ(r1,r2) ◦ φ(r1,r2)(b1 − a1b1a2 + b2a2 2), s1b1 + s2b2} ⊂ hb1, b2i is linearly dependent over k (and hence, in particular does not generate hb1, b2i). 2 ∈ ∂−1(1), it follows that the two DGAs cannot be isomor- Since b1 − a1b1a2 + b2a2 phic. (cid:3) 3. The algebraic setup In this section we give a precise definition of the algebraic setup used. In the fol- lowing k will denote an arbitrary field. By ha1, . . . , aki we will denote the free, fully noncommutative, associative, and unital k-algebra generated by elements a1, . . . , ak. For a subset B ⊂ A, we use AB, BA, and ABA to denote the corre- sponding left, right, and two-sided ideals generated by B, respectively. The free product of two such algebras ha1, . . . , aki and hak+1, . . . , ak+li is the free algebra ha1, . . . , ak, ak+1, . . . , ak+li. 3.1. Semifree DGAs. The main object of interest will be a so-called semifree differential graded algebra (DGA for short) (A, ∂) over a field k. The under- lying algebra will be the free unital algebra A = ha1, . . . , ami, freely generated by elements which each are equipped with a grading ai ∈ Z/Zµ. In addition, we will also require each generator to have an associated action ℓ(ai) ∈ R>0. The grading and the action are both extended to all monomials in the above generators of A using the formulae together with r = 0, r ∈ k; ai1 · . . . · aik = ai1 + . . . + aik ∈ Z/Zµ, ℓ(0) = −∞; ℓ(r) = 0, r ∈ k \ {0}; ℓ(ai1 · . . . · aik ) = ℓ(ai1 ) + . . . + ℓ(aik ) ∈ R>0. 8 GEORGIOS DIMITROGLOU RIZELL We will be writing Ai ⊂ A for the vector space over k spanned by the monomials of degree i ∈ Z/Zµ. The differential ∂ : A• → A•−1 is k-linear of degree −1, satisfies ∂2, ∂k = 0, together with the graded Leibniz rule ∂(ab) = ∂(a)b + (−1)aa∂(b), for any monomials a, b ∈ A. Finally, we will also assume that: (F): For any generator ai, i = 1, . . . , m, the action of a monomial appearing in ∂(ai) with a non-zero coefficient is strictly less than ℓ(ai). Remark 3.1. For the Chekanov-Eliashberg algebra of a Legendrian submanifold Λ ⊂ (P × R, dz + θ), the generators a1, . . . , am are given by the so-called Reeb chords on Λ (which here are assumed to be generic). Recall that a Reeb chord is an integral curve of ∂z having both endpoints on Λ. The grading is induced by the Conley-Zehnder index associated to a Reeb chord, which is well-defined modulo the Maslov number µ ∈ Z of Λ, while the action simply is given by the length ℓ(ai) :=Zai (dz + θ) ∈ R>0 of the Reeb chord. We refer to [EES2] and [EES1] for more details. The fact that the differential is of degree −1 and satisfies (F) (i.e. that it decreases the action) follows from the fact that it is defined by a count of rigid pseudoholomorphic polygons in P having boundary on the canonical projection ΠLag(Λ) ⊂ P of P and corners at the double-points. Recall that the latter double-points are in a natural bijective correspondence with the Reeb chords. More precisely, the degree follows from the dimension formula for the moduli space of pseudo-holomorphic polygons expressed in terms of the Conley-Zehnder indices, while property (F) follows from the formula for their symplectic areas, which necessarily is positive. 3.2. Stable-tame isomorphism. For the applications that we have in mind, the strongest known invariant that can be extracted from the DGA is its stable-tame isomorphism class. We here proceed to give a definition. An elementary auto- morphism of a semifree algebra A = ha1, . . . , ami with a choice of generators is a unital algebra isomorphism of the form ai 7→(ai, rai0 + A, i 6= i0, i = i0, for some 1 ≤ i0 ≤ m, and elements r ∈ k \ {0}, A ∈ ha1, . . . , ai0−1, ai0+1, . . . , ami. A tame isomorphism of two DGAs is one which can be decomposed as a sequence of elementary automorphisms, after first identifying their sets of generators. As a side note, it seems to be unknown whether there are DGA isomorphisms which are not tame. Consider the stabilisation in degree i ∈ Z/Zµ, which is the DGA (Si, ∂Si) with underlying algebra Si := ha, bi, grading given by a = i and b = i + 1, and differential defined by ∂Si (b) = a. Two DGAs are stable-tame isomorphic if they become tame isomorphic after taking free products with a number of stabilisations (Si, ∂Si) for different i ∈ Z/Zµ. NONTRIVIALITY OF THE CHARACTERISTIC ALGEBRA 9 By [Ng, Section 3.1] it follows that, up to algebra isomorphism and taking the free product with a free algebra, the characteristic algebra of a DGA is invariant under stable-tame isomorphisms of the DGA. (In fact, tameness is irrelevant here.) 3.3. Degree functions and the weak algorithm. Here we recall some tech- niques used in the study of free algebras. The weak algorithm is a generalisation of Euclid's division algorithm to the noncommutative setting due to P. Cohn. We refer to [Coh, Section 2] for an introduction. Below follows an overview of the results needed. Let A = ha1, . . . , ami be the free algebra with a choice of m generators over k. Recall the definition of a degree function ν : A → N ∪ {−∞} given in [Coh, Section 2.4], which is a function satisfying the following properties: (1) ν(x) ≥ 0 whenever x 6= 0, ν(0) = −∞; (2) ν(x − y) ≤ max{ν(x), ν(y)}; (3) ν(xy) = ν(x) + ν(y); and (4) ν(x) = 0 if and only if x ∈ k \ {0}. As an example, the usual degree ν0 of an element of A, considered as a noncom- mutative polynomial in the indeterminates a1, . . . , am, clearly satisfies the above properties. Let {ai,1 · . . . · ai,ji }i∈N, be the infinite k-basis of monomials in the generators a1, . . . , am ordered by, say, the lexicographical order. The formal degree of an element x ∈ A induced by ν and the above set of generators is given by x =Xi∈N riai,1 · . . . · ai,ji ∈ A, νfor(x) := max ν(riai,1 · . . . · ai,ji ), i where the ri ∈ k above are uniquely determined and vanishing for all but finitely many values of N. Observe that ν = νfor holds by definition in the case ν = ν0. A family {x1, . . . , xn} ⊂ A of elements is called (left) ν-dependent given that there exist elements {y1, . . . , yn} ⊂ A for which is said to be (left) ν-dependent on the family {x1, . . . , xn}. Definition 3.2. The degree-function ν on A is said to satisfy the (left) weak algorithm if any left ν-dependent family {x1, . . . , xn} ⊂ A of elements satisfying ν(x1) ≤ . . . ≤ ν(xn) has the property that xi is left ν-dependent on {x1, . . . , xi−1} for some 1 ≤ i ≤ n. The main result needed from the theory of noncommutative rings is the following theorem. Moreover, we call such a family (left) ν-independent if it is not (left) ν-dependent. An element x ∈ A satisfying the relation i=1,...,n ν(yixi). yixi! < max ν nXi=1 yixi! < ν(x), ν(yixi) ≤ ν(x), i = 1, . . . , n, ν x − nXi=1 10 GEORGIOS DIMITROGLOU RIZELL Theorem 3.3 (Theorems 2.5.1 (§2.5) and 2.4.6 (§2.4) in [Coh]). Assume that we are given a degree function as above which coincides with the induced formal degree, i.e. that ν = νfor holds. Then: (1) The degree-function ν satisfies the weak algorithm. (2) Any finitely generated left ideal a = Ax1 + . . .+ Axm can be freely generated by a ν-independent family {b1, . . . , bn} ⊂ A of generators. Strictly speaking, we only rely on part (1) of the above theorem, while part (2) is stated for completeness. On the other hand, note that the crucial result Proposition 4.1 proven below is a refinement of the latter statement to a special setting needed here. 3.4. A degree-function induced by the action filtration. Assume that we are given a DGA (A = ha1, . . . , ami, ∂) with an action filtration ℓ taking values in R≥0 ∪{−∞} defined on the monomials. Again, consider the basis {ai,1 ·. . .·ai,ji }i∈N consisting of the monomials. We can extend the action to arbitrary elements by ℓ Xi riai,1 · . . . · ai,ji! := max i ℓ(riai,1 · . . . · ai,ji ), where ri ∈ k \ {0} are non-zero for finitely many values of i ∈ N. For a DGA with action filtration, we approximate the value of ℓ(ai) for each generator, i = 1, . . . , m, by a rational number in Q>0. In this way we can always Finally, we set assume that the action is Q-valued, i.e. that we have a map eℓ : A → Q≥0 ∪ {−∞}. to be the action defined as above, but where we have assigned the value Neℓ(ai) ∈ the actionseℓ(a1), . . . ,eℓ(am) ∈ Q≥0 of the generators. Lemma 3.4. The map ν satisfies the assumptions of a degree-function, as defined in Section 3.3 above, for which ν = νfor moreover is satisfied. N>0 to the generator ai, where N ∈ N>0 is the greatest common denominator of ν : A → N ∪ {−∞} This degree-function is moreover compatible with the differential in the following sense. It then follows that, for any non-zero x ∈ A, the small for each i = 1, . . . , m. inequality Lemma 3.5. Suppose thateℓ was constructed so that ℓ(ai)−eℓ(ai) > 0 is sufficiently holds for the degree-function ν : A → N ∪ {−∞} obtained from eℓ as above. Proof. For a generator ak, take a monomial ai,1 · . . . · ai,ji in the generators arising in the expression ∂(ak) with a non-zero coefficient. Since ν(∂(x)) < ν(x) holds by assumption (F), we conclude that ℓ(ak) − (ℓ(ai,1) + . . . + ℓ(ai,ji )) > 0, eℓ(ak) − (eℓ(ai,1) + . . . +eℓ(ai,ji )) > 0 holds for the action defined above as well, given that ℓ(ai)−eℓ(ai) > 0 is sufficiently small for each i = 1, . . . , m. Since ν(ak) − (ν(ai,1) + . . . + ν(ai,ji )) = Neℓ(ai) − (Neℓ(b1) + . . . + Neℓ(bn)) > 0 NONTRIVIALITY OF THE CHARACTERISTIC ALGEBRA 11 also is satisfied, the statement now follows. To that end, using the Leibniz rule, it suffices to obtain inequalities of the above form for the finite monomials arising with non-zero coefficients in the expressions ∂(ai), i = 1, . . . , m. (cid:3) 4. The proof of Theorem 1.1 Consider a cycle x ∈ (A, ∂) which vanishes inside the characteristic algebra, i.e. [x] = 0 ∈ C. The goal is showing that x = ∂(y) for some element y ∈ A. By definition, there exist elements u1, v1, wn, . . . , un, vn, wn ∈ A for which x = u1∂(v1)w1 + . . . + un∂(vn)wn. Using the Leibniz rule, we may now write x = u1∂(v1w1) + . . . + un∂(vnwn) − ((−1)v1u1v1∂(w1) + . . . + (−1)vnunvn∂(wn)). In particular, x is contained inside the left-ideal a := A∂(v1w1) + . . . + A∂(vnwn) + A∂(w1) + . . . + A∂(wn) generated by boundaries. By Proposition 4.1 below, we conclude that this left-ideal is freely generated by a set of boundaries ∂(y1), . . . , ∂(ym) ∈ A. Writing x = x1∂(y1) + . . . + xm∂(ym), the fact that x is closed together with the Leibniz rule implies that ∂(x) = ∂(x1)∂(y1) + . . . + ∂(xm)∂(ym) = 0. Since the above set of generators is free, we must have ∂(xi) = 0 for each i = 1, . . . , m. We have thus shown that x = ∂((−1)x1x1y1 + . . . + (−1)xmxmym), which finishes the claim. What is left is showing the following proposition, which is the core of the argu- ment. Proposition 4.1. Any left ideal a ⊂ A of the form a = A∂(x1) + . . . + A∂(xn) in a DGA (A, ∂) with action filtration can be freely generated by a family ∂(y1), . . . , ∂(ym) of boundaries. 4.1. The proof of Proposition 4.1. In the following, we let ν be the degree- function induced by the action ℓ, as constructed in Section 3.4 above. The fact that the differential ∂ strictly decreases the degree ν will turn out to be crucial; see Lemma 3.5. First observe that, for each C ≥ 0, the k-subspace AC := ν−1([−∞, C]) ⊂ A is finite-dimensional. We write and also M := max i=1,...,n ν(∂(xi)) bC := b ∩ ν−1([−∞, C]), C ≥ 0, for any subset b ⊂ A. Using this notation, it follows that ∂(A) ∩ aM generates a as a left-ideal. 12 GEORGIOS DIMITROGLOU RIZELL Given an ideal an0 = A∂(y1) + . . . + A∂(yn0) ⊂ a contained inside a which is generated by the boundaries ∂(yi), i = 1, . . . , n0, we define the following two properties: (A): The left ideal an0 is freely generated by a ν-independent family b1, . . . , bn0 satisfying where there moreover are elements ui j ∈ A for which ν(b1) ≤ . . . ≤ ν(bn0) ≤ M, bi = ∂(yi) − (ui 1b1 + . . . + ui i−1bi−1), ν(bi) ≤ ν(∂(yi)), ν(ui jbj) ≤ ν(∂(yi)), is satisfied for each i = 1, . . . , n0. (B): For any boundary ∂(y) ∈ a \ an0 we have ν(∂(y) + x) ≥ ν(bn0) for each element x ∈ an0. From (A) we easily deduce: (A'): The boundaries ∂(y1), . . . , ∂(yn0) form a free generating set for the left-ideal an0. In particular, ∂(an0) ⊂ an0. To see (A') one can argue by induction. Namely, given that v1∂(y1) + . . . + vi−1∂(yi−1) + vi∂(yi) = 0 is satisfied, Property (A) implies the existence of elements v′ which 1, . . . , v′ i−1 ∈ A for v′ 1b1 + . . . + v′ i−1bi−1 + vibi = 0, and hence vi = 0 follows. The proposition will be proven by the following induction argument. We start by choosing a non-zero boundary ∂(y1) ∈ a satisfying the property that, for any non-zero boundary ∂(x) ∈ a, we have (1) Such an element exists by the finite-dimensionality of aM , together with the fact that ∂(A)∩aM generates a. We are now ready to establish the base of the induction. 0 ≤ ν(∂(y1)) ≤ ν(∂(x)). Lemma 4.2 (The base case). The left-ideal a1 := A∂(y1) ⊂ a, satisfies (A) and (B). Proof. Property (A) is immediate with b1 := ∂(y1), while (B) is shown as follows. Assume by contradiction that there exists a boundary ∂(y) ∈ a \ a1 which satisfies (2) ν(∂(y) − v1∂(y1)) < ν(b1) = ν(∂(y1)), for some v1 ∈ A, and write b := ∂(y) − v1∂(y1). The identity ∂(b) = −∂(v1)∂(y1), NONTRIVIALITY OF THE CHARACTERISTIC ALGEBRA 13 combined with Lemma 3.5 and Formula (2) now implies that ∂(v1) = 0. Hence, we conclude that b = ∂(y − v1y1) ∈ a \ a1 + a1 = a \ a1 which, since ν(b) < ν(∂(y1), leads to a contradiction with Formula (1). (cid:3) The following lemma provides us with the induction step. Lemma 4.3 (The induction step). Whenever there is a left-ideal an0 ( a satis- fying (A) and (B), there exists a boundary ∂(yn0+1) ∈ a \ an0 for which again satisfies properties (A), (B). an0+1 := an0 + A∂(yn0+1) We claim that the proposition follows from this induction step. Indeed, recall that aM ⊃ (aN )M is a finite-dimensional vector space. Property (A) states that aN is freely generated by ν(b1) ≤ ν(b2) ≤ . . . ≤ (bN ) ≤ M, which implies that aN = a for some N ≥ 0. The proposition now follows from (A'). What remains is proving the induction step. Proof of Lemma 4.3. Since a is generated by ∂(A) ∩ aM , the subset (a \ an0)M ⊂ aM must contain a boundary. Here we have used the assumption that an0 ( a. The finite-dimensionality of (a \ an0)M implies that we may choose an element ∂(yn0+1) ∈ (a \ an0)M satisfying the following properties: (∗) Writing m := min vi∈A ν(∂(yn0+1) − (v1b1 + . . . + vn0 bn0)) ∈ N, the number m ≥ 0 is minimal in the sense that (3) min vi∈A ν(∂(y) − (v1b1 + . . . + vn0 bn0)) ≥ m, is satisfied for any ∂(y) ∈ a \ an0. By property (B) of an0 we conclude that the inequality (4) m ≥ ν(bn0 ) is satisfied. Furthermore, from the ν-independence of the family b1, . . . , bn0 we establish the following property. Given any ∂(y) ∈ a \ an0 and vi ∈ A, i = 1, . . . , n0, satisfying ν(∂(y) − (v1b1 + . . . + vn0 bn0)) = m ≤ ν(∂(yn0+1)), the inequalities ν(vibi) ≤ ν(∂(y)) must be satisfied for each i = 1, . . . , n0. Property (A) for an0+1: We fix vi ∈ A, i = 1, . . . , n0, as above for which ν(∂(yn0+1) − (v1b1 + . . . + vn0 bn0)) = m ≥ ν(bn0 ) and write bn0+1 := ∂(yn0+1) − (v1b1 + . . . + vn0 bn0). 14 GEORGIOS DIMITROGLOU RIZELL It suffices to show that b1, . . . , bn0, bn0+1 is ν-independent. By the weak algorithm, which holds by part (1) of Theorem 3.3, it thus suffices to show that bn0+1 is not ν-dependent on the family b1, . . . , bn0. To that end, recall that the latter family is ν-independent by assumption. Indeed, given elements ui ∈ A, i = 1, . . . , n0, the inequality ν(bn0+1 − (u1b1 + . . . + un0bn0)) ≥ m = ν(bn0+1) must hold since, otherwise, we would obtain ν(∂(yn0+1) − ((v1 + u1)b1 + . . . + (vn0 + un0)bn0 )) < m, which clearly is in contradiction the choice of ∂(yn0+1) in (∗). Property (B) for an0+1: This property is shown using the previously estab- lished Property (A). By contradiction, take ∂(y) ∈ a \ an0+1 and v1, . . . , vn0+1 ∈ A satisfying ν(∂(y) − (v1b1 + . . . + vn0 bn0 + vn0+1bn0+1)) < m = ν(bn0+1). It is immediate from the construction of m ≥ 0 in (∗) that vn0+1 6= 0. Property (A) for an0+1 shows that we can write 1b1 + . . . + v′ n0 bn0 + vn0+1∂(yn0+1), v1b1 + . . . + vn0 bn0 + vn0+1bn0+1 = v′ i ∈ A, i = 1, . . . , n0. After using Property (A') for an0 (5) for suitable elements v′ together with the Leibniz rule, we compute x := ∂(v′ n0bn0 + ∂(vn0+1)bn0+1, for suitable elements v′′ i ∈ A, i = 1, . . . , n0. Lemma 3.5 implies that ν(x) < m = ν(bn0+1). By the ν-independence of b1, . . . , bn0+1 together with ν(bn0+1) = m it follows that ∂(vn0+1) = 0. 1b1 + . . . + v′ n0bn0 + vn0+1∂(yn0+1)) = v′′ 1 b1 + . . . + v′′ In conclusion, we can construct a boundary ∂(y − vn0+1yn0+1) ∈ a \ an0+1 + an0+1 = a \ an0+1 which by Formula (5) satisfies ν(∂(y − vn0+1yn0+1) − (v′ 1b1 + . . . + v′ n0 bn0 )) < m. This is clearly in contradiction with the construction of m ≥ 0 in (∗). (cid:3) References [Che] Y. Chekanov. Differential algebra of Legendrian links. Invent. Math., 150(3):441 -- 483, 2002. [Coh] P. M. Cohn. Free ideal rings and localization in general rings, volume 3 of New Mathe- matical Monographs. Cambridge University Press, Cambridge, 2006. [DR] G. Dimitroglou Rizell. Exact Lagrangian caps and non-uniruled Lagrangian submanifolds. Ark. Mat., 53(1):37 -- 64, 2015. [DRG] G. Dimitroglou Rizell and R. Golovko. Estimating the number of Reeb chords using a linear representation of the characteristic algebra. Algebr. Geom. Topol., 15(5):2887 -- 2920, 2015. [EES1] T. Ekholm, J. Etnyre, and M. Sullivan. The contact homology of Legendrian submanifolds in R2n+1. J. Differential Geom., 71(2):177 -- 305, 2005. [EES2] T. Ekholm, J. Etnyre, and M. Sullivan. Non-isotopic Legendrian submanifolds in R2n+1. J. Differential Geom., 71(1):85 -- 128, 2005. [EES3] T. Ekholm, J. Etnyre, and M. Sullivan. Legendrian contact homology in P × R. Trans. Amer. Math. Soc., 359(7):3301 -- 3335 (electronic), 2007. [EGH] Y. Eliashberg, A. Givental, and H. Hofer. Introduction to symplectic field theory. Geom. Funct. Anal., (Special Volume, Part II):560 -- 673, 2000. GAFA 2000 (Tel Aviv, 1999). NONTRIVIALITY OF THE CHARACTERISTIC ALGEBRA 15 [Ng] [Siv] L. Ng. Computable Legendrian invariants. Topology, 42(1):55 -- 82, 2003. S. Sivek. The contact homology of Legendrian knots with maximal Thurston-Bennequin invariant. J. Symplectic Geom., 11(2):167 -- 178, 2013. [Sul] D. Sullivan. Infinitesimal computations in topology. Inst. Hautes ´Etudes Sci. Publ. Math., (47):269 -- 331 (1978), 1977. Georgios Dimitroglou Rizell Centre for Mathematical Sciences University of Cambridge Wilberforce Road Cambridge, CB3 0WB United Kingdom [email protected]
1708.06648	1	1708	2017-08-21T15:49:16	A note on matrices mapping a positive vector onto its element-wise inverse	[ "math.RA" ]	For any primitive matrix $M\in\mathbb{R}^{n\times n}$ with positive diagonal entries, we prove the existence and uniqueness of a positive vector $\mathbf{x}=(x_1,\dots,x_n)^t$ such that $M\mathbf{x}=(\frac{1}{x_1},\dots,\frac{1}{x_n})^t$. The contribution of this note is to provide an alternative proof of a result of Brualdi et al. (1966) on the diagonal equivalence of a nonnegative matrix to a stochastic matrix.	math.RA	math	A NOTE ON MATRICES MAPPING A POSITIVE VECTOR ONTO ITS ELEMENT-WISE INVERSE SÉBASTIEN LABBÉ Abstract. For any primitive matrix M ∈ Rn×n with positive diagonal entries, we prove the existence and uniqueness of a positive vector x = (x1, . . . , xn)t such that Mx = ( 1 )t. The contribution of this note is to provide an alternative proof of a result of Brualdi et al. (1966) on the diagonal equivalence of a nonnegative matrix to a stochastic matrix. x1 , . . . , 1 xn 1. Introduction 7 1 0 2 g u A 1 2 ] . A R h t a m [ 1 v 8 4 6 6 0 . 8 0 7 1 : v i X r a xn )t. In this note, we consider matrices mapping a vector with positive entries onto its element-wise inverse. We prove unicity and existence of such a vector for primitive matrices, that is nonnegative matrices some power of which is positive, with positive diagonal entries. The main result is: Theorem 1. Let M ∈ Rn×n≥0 be a primitive matrix with positive diagonal entries. Then there exists a unique vector x = (x1, . . . , xn)t with positive entries such that Mx = ( 1 x1 , . . . , 1 It turns out that this question was already answered in 1966 under an equivalent form. In [BPS66], it was proved that if A is a nonnegative square matrix with positive diagonal entries, then there exists a unique diagonal matrix D with positive diagonal entries such that DAD is row stochastic (see also [Sin66] who proved it for positive matrices A). The equivalence is explained below in Lemma 2. As a consequence, the contribution of this note is to provide an alternative proof of the above result. Unicity for primitive matrices is obtained as a consequence of Perron theorem whereas existence for nonnegative matrices with positive diagonal entries is deduced from the Brouwer fixed-point theorem. This note is structured as follows. In Section 2, we present an equivalent system of quadratic equations to be solved. In Section 3, we deduce unicity for primitive matrices from Perron theorem. In Section 4, we reduce the question to finding fixed-points of a function and we recall Brouwer and Banach fixed-point theorems in Section 5. In Section 6, we use Brouwer fixed-point theorem to prove existence for nonnegative matrices with positive diagonal entries. In Section 7, we use Banach fixed-point theorem to prove existence and unicity for nonnegative matrices with relatively large enough diagonal entries including matrices which are not primitive (Proposition 12). Date: October 24, 2018. 2010 Mathematics Subject Classification. Primary 15B48; Secondary 47H10 and 15B51. Key words and phrases. Primitive matrices and Stochastic matrices and Fixed-Point theorems and Perron theorem. 1 2 S. LABBÉ Figure 1. Left: the two quadratic curves x2 2 = 1, intersect in a unique point in the box [0, 1]2. Right: the three quadratic surfaces x2 1 + 2x1x2 + 2x1x3 = 1, x1x2 + x2 3 = 1 intersect in a unique point in the box [0, 1]3. 2 + x2x3 = 1, x1x3 + 3x2x3 + x2 1 + 3x1x2 = 1, 5x1x2 + 2x2 2. A system of quadratic equations We say that a vector or a matrix is nonnegative (resp. positive) if all of its entries are nonnegative (resp. positive). Lemma 2. Let M = (mij) ∈ Rn×n be a nonnegative matrix and x = (x1, . . . , xn)t be a positive vector. The following conditions are equivalent. x1 , . . . , 1 (i) Mx = ( 1 (ii) diag(x)Mdiag(x) is a stochastic matrix, (iii) for every i ∈ {1, . . . , n}, )t, xn nX j=1 (1) xi mijxj = 1. Proof. (i) ⇐⇒ (ii). The matrix diag(x)Mdiag(x) is stochastic if and only if (1, . . . , 1)t is a right eigenvector with eigenvalue 1, that is, (2) which is equivalent to Mx = diag(x)−1(1, . . . , 1)t = ( 1 diag(x)Mdiag(x)(1, . . . , 1)t = (1, . . . , 1)t )t x1 , . . . , 1 (ii) ⇐⇒ (iii). Let ri be the i-th row of the matrix M. We develop (2) and we get diag(x)Mx = diag(x)(r1 · x, . . . , rn · x)t = (x1r1 · x, . . . , xnrn · x)t = (1, . . . , 1)t. This equation is verified if and only if, for each i ∈ {1, . . . , n}, the quadratic Equation (1) (cid:3) in x1, . . . , xn holds. The system of equations (1) for i ∈ {1, . . . , n} is illustrated in Figure 1 for n = 2 and xn n = 3. 00.20.40.60.8100.20.40.60.81A NOTE ON MATRICES MAPPING A POSITIVE VECTOR ONTO ITS ELEMENT-WISE INVERSE3 3. Uniqueness for primitive matrices A primitive matrix is a nonnegative matrix some power of which is positive. >0. If M 2 v = v, then M 2(M v) = M(M 2v) = M v. be a primitive matrix and v ∈ Rn Lemma 3. Let M ∈ Rn×n≥0 M v = v. Proof. We already have that v is a positive vector fixed by M 2 which is primitive. But so is M v: By Perron's theorem, v and M v must be colinear, that is, there exists λ ∈ R such that v = λM v. Then, v = λ2M 2v = λ2v and thus λ2 = 1. Since v and M v are positive, we (cid:3) deduce λ = 1. Proposition 4. Let M ∈ Rn×n be a primitive matrix. If there exists a positive vector x = (x1, . . . , xn)t ∈ Rn )t, then it is unique. x1 , . . . , 1 Proof. Let x = (x1, . . . , xn)t, y = (y1, . . . , yn)t ∈ Rn >0. Suppose that X = diag(x) and Y = diag(y) are such that XM X and Y M Y are both stochastic. The product of diagonal matrices commutes, so we have >0 such that Mx = ( 1 xn (XM Y )2 = XM(Y X)M Y = XM(XY )M Y = (XM X)(Y M Y ). We conclude that (XM Y )2 is stochastic. From Lemma 3, we conclude that XM Y is stochastic. Thus we have XM Y (1, . . . , 1)t = (1, . . . , 1)t and Y M Y (1, . . . , 1)t = (1, . . . , 1)t and (x−1 Therefore x = y. The conclusion follows from Lemma 2. n ) = X−1(1, . . . , 1)t = M Y (1, . . . , 1)t = Y −1(1, . . . , 1)t = (y−1 1 , . . . , x−1 1 , . . . , y−1 n ). (cid:3) 4. Solutions are fixed points It can be seen in Figure 1 that the surfaces of each equation in the positive octant Let M = (mij) ∈ Rn×n be a nonnegative matrix. For each i ∈ {1, . . . , n} and are functions of the form y = f(x) or z = f(x, y). We now formalize and prove this. bi =X (x1, . . . , xi−1, xi+1, . . . , xn) ∈ Rn−1≥0 , we denote (3)  For each i ∈ {1, . . . , n}, we define a function f (M) i b−1 −bi + : Rn−1≥0 → R>0: q (x1, . . . , xi−1, xi+1, . . . , xn) = if mii = 0, if mii 6= 0. mijxj. (M) i (4) f Lemma 5. Let M = (mij) ∈ Rn×n be a nonnegative real matrix, i ∈ {1, . . . , n} and assume x = (x1, . . . , xn)t ∈ Rn >0. The vector x satisfies Equation (1) if and only if xi = f (x1, . . . , xi−1, xi+1, . . . , xn). (M) i j6=i i i + 4mii b2 2mii 4 Proof. Equation (1) can be seen as a quadratic equation of the variable xi: (5) S. LABBÉ where bi =P then bixi = 1 and there is only one solution xi = b−1 If mii 6= 0, then there are exactly two real solutions i + bixi − 1 = 0 miix2 j6=i mijxj is the coefficient of xi in this quadratic polynomial. If mii = 0, to Equation (5). Moreover xi > 0. i −bi −q i + 4mii b2 2mii < 0 and −bi + q i + 4mii b2 2mii (cid:3) > 0  to Equation (5), the second one being positive. For every matrix M ∈ Rn×n, let F (M) : Rn≥0 → Rn >0 : x 7→  (M) f 1 (M) f 2 . . . f (M) n (x2, x3, . . . , xn) (x1, x3, . . . , xn) (x1, x2, . . . , xn−1) We now have a new equivalent statement. Lemma 6. Let M ∈ Rn×n be a nonnegative matrix and x = (x1, . . . , xn)t be a positive vector. Then Mx = ( 1 )t if and only if x is a fixed point of F (M). Proof. Let x = (x1, . . . , xn)t > 0. We have that F (M)(x) = x if and only if x1 , . . . , 1 xn xi = f (M) i (x1, . . . , xi−1, xi+1, . . . , xn) for every i ∈ {1, . . . , n} if and only if x satisfies Equation (1) for every i ∈ {1, . . . , n} (cid:3) from Lemma 5 if and only if Mx = ( 1 )t from Lemma 2. x1 , . . . , 1 xn From [Zei86], we recall some classical fixed-point theorems. 5. Fixed-Point theorems Theorem 7 (Brouwer Fixed Point Theorem). Every continuous function from a closed ball of a Euclidean space into itself has a fixed point. We consider closed balls for the ∞-norm. For every a = (a1, . . . , an), b = (b1, . . . , bn) ∈ Rn, the closed ball with center (a + b)/2 and radius max{0, (ai − bi)/2} for every i-th coordinate, 1 ≤ i ≤ n, is denoted by Box(a, b) = {(x1, . . . , xn) ∈ Rn ai ≤ xi ≤ bi, 1 ≤ i ≤ n}. A function F : Rn≥0 → Rn≥0 is said decreasing if F(x + t) ≤ F(x) for every x, t ∈ Rn≥0. Corollary 8. Let F : Rn≥0 → Rn≥0 be a continuous and decreasing function. Then there exists a vector x ∈ Rn≥0 such that x = F(x). Proof. Since F is continuous and decreasing, we have that, for every a, b ∈ Rn≥0, F(Box(a, b)) ⊆ Box(F(b), F(a)). A NOTE ON MATRICES MAPPING A POSITIVE VECTOR ONTO ITS ELEMENT-WISE INVERSE5 Moreover, F reaches its maximal value at a = 0 so that F(Rn≥0) ⊆ Box(0, F(0)). Then F(Box(0, F(0))) ⊆ Box(F 2(0), F(0)) ⊆ Box(0, F(0)) (cid:3) and Brouwer fixed point theorem applies since Box(0, F(0)) is a closed ball. 5.1. Banach Fixed-Point Theorem. Let (X, d) be a metric space. Then a map T : X → X is called a contraction mapping on X if there exists q ∈ [0, 1) such that d(T(x), T(y)) ≤ qd(x, y) for all x, y ∈ X. Theorem 9 (Banach Fixed Point Theorem). Let (X, d) be a non-empty complete metric space with a contraction mapping T : X → X. Then T admits a unique fixed-point x in X. 6. Existence for nonnegative matrices with positive diagonal entries Now we compute the gradient of f : (M) i −b−2 i (mi1, . . . , mi,i−1, mi,i+1, . . . , min)t 1 2mii  (mi1, . . . , mi,i−1, mi,i+1, . . . , min)t  − 1 if mii = 0, if mii 6= 0,  (6) ~∇f (M) i (x) = biq i + 4mii b2 i ≤ 0. and we conclude that ~∇f (M) xn )t. x1 , . . . , 1 We now prove the existence of a fixed point of F (M) using Brouwer fixed-point theo- rem. Proposition 10. Let M = (mij) ∈ Rn×n be a nonnegative real matrix such that mii > 0 for every i with 1 ≤ i ≤ n. Then there is a positive vector x = (x1, . . . , xn)t > 0 such that Mx = ( 1 Proof. The function F (M) : Rn≥0 → Rn >0 is continuous since mii 6= 0 for all i such that 1 ≤ i ≤ n. It is decreasing since the entries of its gradient are zero or negative. Thus, Corollary 8 applies and there exists a vector x ∈ Rn≥0 such that x = F (M)(x). From the definition of F (M), we conclude that the entries of x are positive, i.e., x ∈ Rn >0. From Lemma 6, we conclude the existence of a positive vector x = (x1, . . . , xn)t such that (cid:3) Mx = ( 1 )t. x1 , . . . , 1 xn Example 11. Let  M = 1 1 0 1 1 1  1 0 0 1, 5 + 1 2 , We verify that √ Mx = and x = √ q 5 + 2 √ 5 + 22 + 1 4 1, t = q √ 5 − 1 2 , √ 5 + 22 − √ 2 5 − 1 4 t . 1, 2√ 5 − 1, q 2 √ 4 5 + 22 − √ 5 − 1 t . S. LABBÉ 6 Proof of Theorem 1. Unicity follows from Proposition 4 since M is primitive. Existence (cid:3) follows from Proposition 10 since diagonal entries of M are positive. Proposition 10 does not include primitive matrices with zero entries on the diagonal since we can't apply Brouwer fixed-point theorem when mii = 0 for some i: f is not continuous at 0 in this case. But the result still holds (see [BPS66, Theorem 8.2]). For example, let (M) i M = We verify that Mx =(cid:16) 1√  0 0 1 1 0 0 0 1 1 2, √ 2,  2(cid:17)t. √ √ !t . 2, 1√ 2, 1√ 2 and x = 7. Uniqueness when diagonal entries are relatively large To prove uniqueness in some cases including matrix M from Example 11 which is not primitive, we can use Banach fixed-point theorem. Note that it is not possible to prove that the map F (M) is a contraction for every nonnegative matrix M. For example, consider  1 2m 2m 2m 1 2m 2m 2m 1  M = for some m > 0. We get that the gradient of F (M) at x = 0 (in which case bi = 0 in Equation (6)) is ∂F (M) ∂xi ! (0) i=1,2,3  =  0 −m −m 0 −m 0 −m −m −m which can get as large as m is. For some matrices M, the map F (M) is a contraction as we show now. Proposition 12. Let M = (mij) ∈ Rn×n be a nonnegative real matrix such that 2mii > mij for every i, j with 1 ≤ i, j ≤ n. Then there exists a unique positive vector x = (x1, . . . , xn)t > 0 such that Mx = ( 1 )t. x1 , . . . , 1 xn Proof. Existence follows from Proposition 10. To prove uniqueness we use the Banach Fixed Point Theorem and we show that F (M) is a contraction. From the hypothesis, there exists a constant C > 0 such that for every i, j with 1 ≤ i, j ≤ n. Thus from Equation (6) and since 0 ≤ mij 2mii ≤ C < 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) ≤ 1 − 1 biq i + 4mii b2 (cid:13)(cid:13)(cid:13)(mi1, . . . , mi,i−1, mi,i+1, . . . , min)t(cid:13)(cid:13)(cid:13)∞ ≤ C we get that (7) (cid:13)(cid:13)(cid:13)~∇f (M) i (x)(cid:13)(cid:13)(cid:13)∞ ≤ 1 2mii A NOTE ON MATRICES MAPPING A POSITIVE VECTOR ONTO ITS ELEMENT-WISE INVERSE7 for every x ∈ Rn−1≥0 and 1 ≤ i ≤ n. The function f is differentiable on Rn−1≥0 . Using the Mean value theorem in several variables, for every a, b ∈ Rn−1≥0 there exists c ∈ [0, 1] such that Therefore, by the Cauchy-Schwarz inequality (x · y ≤ kxkkyk), ((1 − c)a + cb) · (b − a). (a) = ∇f (b) − f (M) i (M) i (M) i f (M) i (cid:12)(cid:12)(cid:12)f (M) i (b) − f (M) i (a)(cid:12)(cid:12)(cid:12) ≤ C kb − ak∞ . (M) i Thus f conclusion is deduced from Lemma 6. is a contraction for every i with 1 ≤ i ≤ n. Then F (M) is a contraction. The (cid:3) Proposition 12 seems to hold when 2mii ≤ mij. A possible option in this case is to show that some power of F (M) is a contraction and use a stronger version of Banach theorem: if some iterate T n of T is a contraction, then T has a unique fixed point. More work has to be done. Acknowledgements I am thankful to Daniel S. Maynard for fruitful exchanges. The question came up in his research on interference competition in fungi [MBL+17] exploring how spatial structure and species diversity interactively structure the community. References [BPS66] Richard A. Brualdi, Seymour V. Parter, and Hans Schneider. The diagonal equivalence of a nonnegative matrix to a stochastic matrix. Journal of Mathematical Analysis and Applications, 16(1):31 -- 50, 1966. [MBL+17] Daniel S. Maynard, Mark A. Bradford, Daniel L. Lindner, Linda T. A. van Diepen, Serita D. Frey, Jessie A. Glaeser, and Thomas W. Crowther. Diversity begets diversity in compe- tition for space. Nature Ecology & Evolution, May 2017. http://dx.doi.org/10.1038/ s41559-017-0156. Richard Sinkhorn. A relationship between arbitrary positive matrices and stochastic ma- trices. Canadian Journal of Mathematics, 18(0):303 -- 306, January 1966. Eberhard Zeidler. Nonlinear functional analysis and its applications. I. Springer-Verlag, New York, 1986. Fixed-point theorems, Translated from the German by Peter R. Wadsack. [Sin66] [Zei86] (S. Labbé) CNRS, LaBRI, UMR 5800, F-33400 Talence, France E-mail address: [email protected] URL: http://www.slabbe.org/
1707.07765	4	1707	2018-04-06T23:02:51	Elementary matrix-computational proof of Quillen-Suslin theorem for Ore extensions	[ "math.RA" ]	In this short note we present an elementary matrix-constructive proof of Quillen-Suslin theorem for Ore extensions: If $K$ is a division ring and $A:=K[x;\sigma,\delta]$ is an Ore extension, with $\sigma$ bijective, then every finitely generated projective $A$-module is free. We will show an algorithm that computes the basis of a given finitely generated projective module. The algorithm has been implemented in a computational package, and some illustrative examples are included.	math.RA	math	Elementary matrix-computational proof of Quillen-Suslin theorem for Ore extensions William Fajardo∗ Oswaldo Lezama† Seminario de ´Algebra Constructiva - SAC2 Departamento de Matem´aticas Universidad Nacional de Colombia, Bogot´a, COLOMBIA [email protected] 8 1 0 2 r p A 6 ] . A R h t a m [ 4 v 5 6 7 7 0 . 7 0 7 1 : v i X r a Abstract In this short note we present an elementary matrix-constructive proof of Quillen-Suslin theorem for Ore extensions: If K is a division ring and A := K[x; σ, δ] is an Ore extension, with σ bijective, then every finitely generated projective A-module is free. We will show an algorithm that computes the basis of a given finitely generated projective module. The algorithm has been implemented in a computational package, and some illustrative examples are included. Key words and phrases. Projective modules, Ore extensions, non-commutative computational algebra. 2010 Mathematics Subject Classification. Primary: 16Z05. Secondary: 16D40, 15A21. 1 Introduction When a new type of ring is defined, it is an interesting problem to investigate if the finitely generated projective modules over it are free. This problem becomes classical after the formulation in 1955 of the famous Serre's problem about the freeness of finitely generated projective modules over the polynomial ring K[x1, . . . , xn], K a field (see [1], [2], [3], [6]). The Serre's problem was solved positively, and independently, by Quillen in USA, and by Suslin in Leningrad, USSR (St. Petersburg, Russia) in 1976 ([7], [8]). Definition 1.1. Let S be a ring. S is a PF ring if every finitely generated (f.g.) projective S-module is free. Theorem 1.2 (Quillen-Suslin; [7], [8]). K[x1, . . . , xn] is PF . The goal of this short paper is to present an elementary matrix-constructive proof of Quillen-Suslin theorem for single Ore extensions over division rings, i.e, if K is a division ring and A := K[x; σ, δ] is an Ore extension, with σ a bijective endomorphism of K and δ a σ-derivation, then A is PF . Our proof is supported in a matrix characterization of PF rings given in [5]. ∗The first author was supported by Instituci´on Universitaria Polit´ecnico Grancolombiano †The second author was supported by the project New trends of non-commutative algebra and skew PBW extensions, HERMES CODE 26872, Universidad Nacional de Colombia. 1 Proposition 1.3 ([5]). Let S be a ring. S is P F if and only if for every s ≥ 1, given an idempotent matrix F ∈ Ms(S), there exists a matrix U ∈ GLs(S) such that where r = dim(hF i), 0 ≤ r ≤ s, and hF i represents the left S-module generated by the rows of F . Moreover, a basis of M is given by the last r rows of U . U F U −1 =(cid:20)0 0 0 Ir(cid:21) , (1.1) 2 Quillen-Suslin theorem: Elementary matrix proof In this section we will prove that the Ore extension K[x; σ, δ] is PF. Despite of this fact is well-known (see [4]), our proof is elementary and matrix-constructive, and allow to exhibit an algorithm that computes the basis of a given finitely generated projective modules. Theorem 2.1 (Quillen-Suslin). Let K be a division ring and A := K[x; σ, δ], with σ bijective. Then A is PF. Proof. Let s ≥ 1 and let F = [fij] ∈ Ms(A) be an idempotent matrix, the proof is by induction on s and we will follow a procedure as in Proposition 64 of [5]. We will use the relations that satisfy the entries of F , in particular, the following two relations: f 2 11 + f12f21 + f13f31 + · · · + f1sfs1 = f11, f11f12 + f12f22 + f13f32 + · · · + f1sfs2 = f12. s=1: In this case F = [f ]; since A is a domain, its idempotents are trivial, then f = 1 or f = 0 and hence U = [1]. s ≥ 2: Now suppose that the result holds for s − 1 and let F = [fij] ∈ Ms(A) be an idempotent matrix. We have two possibilities. (A) All elements in the first row and in the first column of F are zero. Then we apply induction. (B) Suppose that there exists at least one non zero element in the first row (the reasoning for the first column is similar); we can assume that this element is f11 (if f11 = 0 and f1j 6= 0 then we can change F by T F T −1 with T := Is − Ej1). Then arise two possibilities. (B1) deg(f11) = 0, so f11 ∈ K − 0, i.e., f11 is invertible. Then taking f −1 11 f12 1 0 f −1 11 f13 0 1 −f21f −1 11 −f31f −1 11 U := 1 ...   −fs1f −1 11 0 0 we have that U ∈ GLs(A) and its inverse is U −1 = −f12 11 f12 + 1 −f31f −1 11 f12 f11 f21 −f21f −1 f31 ... fs1 −fs1f −1 11 f12 −f13 −f21f −1 11 f13 −f31f −1 11 f13 + 1 −fs1f −1 11 f13   · · · · · · · · · · · · · · · f −1 11 f1s 0 0 1   · · · · · · · · · −f1s −f21f −1 −f31f −1 11 f1s 11 f1s · · · · · · −fs1f −1 11 f1s + 1 .   Moreover, U F U −1 = (cid:20) 1 apply induction. 0s−1,1 01,s−1 F1 (cid:21), where F1 ∈ Ms−1(A) is an idempotent matrix, therefore we can 2 T F T −1 = F ′ =  f ′ 11 f21 ... fs1 f ′ 12 f ′ 22 ... f ′ s2 f ′ 13 f23 ... fs3 . . . . . . . . . . . . f ′ 1s f2s ... fss LF L−1 = F ′′ =  f ′′ 11 f ′′ 21 ... f ′′ s1 f12 f ′′ 22 ... fs2 f13 f ′′ 23 ... fs3 . . . . . . . . . . . . f1s f ′′ 2s ... fss ,   ,   (B2) deg(f11) := n ≥ 1; since A is a domain at least one non diagonal entry in the first row and in the first column of F are non zero: In fact, if f12 = · · · = f1s = 0, then f11 = 1 or f11 = 0, false; similarly if f21 = · · · = fs1 = 0. Using elementary and permutation matrices, no affecting the entry f11, we can reduce the degrees of f12, . . . , f1s until the situation in which f12 6= 0 and f13 = · · · = f1s = 0 (a similar reasoning apply for the first column); then we have f 2 11 + f12f21 = f11 and f21 6= 0; note that deg(f 2 11) = 2n, so deg(f21) := p ≤ n or deg(f12) := q ≤ n; let an := lc(f11), cp := lc(f21) and bq := lc(f12). If p ≤ n then with T := Is − anσn−p(c−1 then arise two options: deg(f ′ or deg(f ′ 12) ≤ deg f ′ 11. If p > n but q ≤ n then p )xn−pE12; note that F ′ is idempotent; moreover f ′ 11) = 0, i.e., f ′ 11 ∈ K −0 or 1 ≤ deg(f ′ 11 6= 0 11) ≤ n−1 and again deg(f21) ≤ deg f ′ 11 11 = 0 or f ′ 11 6= 0; if f ′ with L := Is + σ−q(b−1 then arise two options: deg(f ′′ or deg(f ′′ 21) ≤ deg f ′′ 11. q an)xn−qE21; note that F ′′ is idempotent; moreover f ′′ 11) = 0, i.e., f ′′ 11 ∈ K −0 or 1 ≤ deg(f ′′ 11 6= 0 11) ≤ n−1 and again deg(f12) ≤ deg f ′′ 11 11 = 0 or f ′′ 11 6= 0; if f ′′ We can repeat this reasoning for F ′ and F ′′ and we obtain an idempotent matrix G = [gij] similar to F with g11 = 0 or g11 ∈ K − 0; if g11 ∈ K − 0 we conclude using the case (B1). Then assume that g11 = 0; if all elements in the first row and in the first column of G are zero, then we can apply induction and we finish. If not, then in a similar way as was remarked above, using elementary and permutation matrices, no affecting the first column, in particular the entry g11, we can reduce the degrees of g12, . . . , g1s until the situation in which g12 6= 0 and g13 = · · · = g1s = 0 (a similar reasoning apply for the first column); thus, from g12g22 = g12 we obtain that g22 = 1 and hence by the permutation matrix P12 we finish using the case (B1). 3 The algorithm In this section we present the algorithm for computing the matrix U in the proof of Quillen-Suslin theorem for Ore extensions (Theorem 2.1); the algorithm also calculates the basis of a given finitely generated projective module (Proposition 1.3). We present two versions of the algorithm, a constructive simplified version, and a more complete computational version over fields. The computational version was implemented using Mapler 2016 (see Remark 4.2 below). 3 Algorithm for the Quillen-Suslin theorem: Constructive version INPUT: An Ore extension A := K[x, σ, δ] (K a division ring, σ bijective); F ∈ Ms(A) an idempotent matrix. OUTPUT: Matrices U , U −1 and a basis X of hF i, where U F U −1 =(cid:20)0 0 Ir(cid:21) and r = dim(hF i). 0 (3.1) INITIALIZATION: F1 := F . FOR k from 1 to n − 1 DO 1. Follow the reduction procedures (B1) and (B2) in the proof of The- and Fk+1 such that orem 2.1 in order to compute matrices U ′ k, U ′−1 k 0 Fk+1(cid:21) , where αk ∈ {0, 1}. U ′ kFkU ′−1 0 k =(cid:20)αk k(cid:21) Uk−1; compute U −1 k . 0 U ′ 2. Uk :=(cid:20)Ik−1 0 3. By permutation matrices modify Un−1. RETURN U := Un−1, U −1 satisfying (3.1), and a basis X of hF i. Example 3.1. For A := K[x, σ, δ], with K := C, σ(z) := z and δ := 0, we consider in M4(A) the idempotent matrix F =   1 − ix − x2 + (1 + i)x3 −1 + (2 − i)x2 + (−1 − i)x3 −i − x + (1 + i)x2 1 + ix + (−1 + i)x2 −ix + (1 + i)x3 ix + (1 − i)x2 + (−1 − i)x3 −i + (1 + i)x2 1 + (−1 + i)x2 ix2 − x2 x3 −x − ix2 −ix + (1 − i)x2 − x3 1 + ix x2 − x x 1 + ix + ix2   . We apply the constructive version of the Quillen-Suslin algorithm, i.e., following the reductions (B1) and (B2), we compute the matrices Uk and Fk, for 1 ≤ k ≤ 3: U1 =   1 − ix − x2 + (1 + i)x3 −1 + (2 − i)x2 + (−1 − i)x3 −i − x + (1 + i)x2 1 + ix + (−1 + i)x2 x 0 0 −i − x 1 0 1 0 0 i 0 1   , i + x + (−1 − i)x2 1 0 −x 1 + ix − x2 + (1 − i)x3 0 0 0 U −1 1 =  0 1 0 0 i + x −i 1 0   , U1F U −1 0 0 1 0 0 −i + (1 + i)x2 0 x2 − x 1 =  0 0 0 0 1 0 0 1   1 − ix − x2 + (1 + i)x3 ix + (−1 − i)x3 −1 + (2 − i)x2 + (−1 − i)x3 −ix + (−1 + i)x2 + (1 + i)x3 0 −i + (1 + i)x2 x2 − x , F2 =  0 1 0 0 0 1  ; −i − x + (1 + i)x2 i + (−1 − i)x2 1 + ix + (−1 + i)x2 −1 + (1 − i)x2 1 0 i 1   , U2 =   x 0 −i − x 0 4 U −1 2 =  2 =  U2F U −1 0 −1 1 0 −x −i − x 0 0 i + x + (−1 − i)x2 i + (−1 − i)x2 −ix2 0 1 0 0 0 1 0 0 0 0 0 0 x2 − x 1 0 0 0 , F3 =(cid:20)   , 0 0 −i 1   x2 − x 1(cid:21) ; 0 0 U3 =   1 − ix − x2 + (1 + i)x3 ix + (−1 − i)x3 −x3 + x2 x −1 + (2 − i)x2 + (−1 − i)x3 −ix + (−1 + i)x2 + (1 + i)x3 ix + (−1 + i)x2 + x3 −i − x −i − x + (1 + i)x2 1 + ix + (−1 + i)x2 i + (−1 − i)x2 −x2 + x 1 −1 + (1 − i)x2 −1 − ix − ix2 i   , U −1 3 =  0 −1 1 0 −x −i − x 0 0 U3F U −1 1 0 0 0 3 =  Finally, using permutation matrices, we get 0 0 i −1 0 0 1 0 0 1 0 0 ,   i + x + (−1 − i)x2 i + (−1 − i)x2 −ix −x2 + x 0 0 0 0   , F4 =(cid:2)0(cid:3) . U =   x −i − x 1 1 − ix − x2 + (1 + i)x3 ix + (−1 − i)x3 −1 + (2 − i)x2 + (−1 − i)x3 −ix + (−1 + i)x2 + (1 + i)x3 −i − x + (1 + i)x2 i + (−1 − i)x2 −x3 + x2 ix + (−1 + i)x2 + x3 −x2 + x i 1 + ix + (−1 + i)x2 −1 + (1 − i)x2 −1 − ix − ix2   , i + x + (−1 − i)x2 i + (−1 − i)x2 −ix −x2 + x 0 −1 1 0 −x −i − x 0 0 U −1 =  , 0 0 i −1   U F U −1 =  0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 1 .   So, r = 3 and the last three rows of U conform a basis X = {x1, x 2, x 3} of hF i, x 1 = (1 − ix − x2 + (1 + i)x3, −1 + (2 − i)x2 + (−1 − i)x3, −i − x + (1 + i)x2, 1 + ix + (−1 + i)x2), x 2 = (ix + (−1 − i)x3, −ix + (−1 + i)x2 + (1 + i)x3, i + (−1 − i)x2, −1 + (1 − i)x2), x 3 = (−x3 + x2, ix + (−1 + i)x2 + x3, −x2 + x, −1 − ix − ix2). Next we present a second illustration of the constructive algorithm. Example 3.2. Let M4(A), where A := K[x, σ, δ], K := Q(t), σ := idQ(t) and δ := d idempotent matrix F := [F (1) F (2) F (3) F (4)], F (i) the ith column of F , where dt ; we consider the F (1) =:  2 + 2t + (13t2 − 5t)x + (8t3 − 6t2)x2 + t3(t − 1)x3 2t2 + t + (13t3 − 8t2)x + (8t4 − 7t3)x2 + t4(t − 1)x3 3t + 2 + (14t2 − 8t)x + (8t3 − 7t2)x2 + t3(t − 1)x3 t2 + t + (t3 + 6t2)x + 6t3x2 + t4x3 5 ,   −t3x3 − 5t2x2 − 3tx + 1 t + (−3t2 + 2t)x + (−5t3 + t2)x2 − t4x3 −t3x3 − 5t2x2 − 3tx + 1 −t3x3 − 5t2x2 − 3tx + 1 t3x3 + 5t2x2 + 3tx − 1 t4x3 + 5t3x2 + 2t2x − 2t −t − 1 + (−t2 + 5t)x + 6t2x2 + t3x3 −t2 + t + (−t3 + 6t2)x + 2t3x2 ,   ,   F (2) =:  F (3) =:  F (4) =:  0 tx tx 1 + (t2 − 2t)x − t2x2 .   Applying the algorithm we obtain U (1) =:  2t + 1 + (10t2 − 5t)x + (7t3 − 6t2)x2 + (t4 − t3)x3 −3t − 2 + (−14t2 + 8t)x + (−8t3 + 7t2)x2 + (−t4 + t3)x3 −2t + 2 − t(t − 1)x −2t2 + 7t − 2 − t(4t2 − 21t + 10)x − t2(t2 − 10t + 7)x2 + t3(t − 1)x3 ,   U (2) =:  −t3x3 − 4t2x2 − tx t3x3 + 5t2x2 + 3tx − 1 tx + 1 2t(t − 3)x + t2(t − 6)x2 − t3x3 ,   −t − 1 + (−t2 + 3t)x + 5t2x2 + t3x3 t + 2 + (t2 − 5t)x − 6t2x2 − t3x3 −tx − 1 −t + 1 − t(2t − 7)x − t2(t − 6)x2 + t3x3 U (3) =:  ,   U (4) =:  tx −tx 0 1 ;   tx + 1 (U −1)(1) =:  (U −1)(2) =:  t − 2 + t(t − 1)x 0 −t + 2 − t(t − 4)x + t2x2 tx + 1 t − 1 + t(t − 1)x 1 1 + (−t2 + 3t)x + t2x2 −t2x2 − 2tx + 1 ,     , (U −1)(4) =:  0 tx tx 1 + (t2 − 2t)x − t2x2 6 .   (U −1)(3) =:   t + (−4t2 + 4t)x + (−2t3 + 5t2)x2 + t3x3 1 + (−2t2 + t)x + (−t3 + 4t2)x2 + t3x3 − 18t2)x2 + (2t4 − 5t)x + (−t4 + 11t3 − 9t3)x3 1 + (−2t3 + 8t2   , − t4x4 With these computations we have U F U −1 =  0 0 0 0 thus, r = 2 and a base of hF i is X = {x1, x 2}, with 0 0 0 0 0 0 0 0 1 0 0 1 ,   x 2 = (−2t2 + 7t − 2 − t(4t2 − 21t + 10)x − t2 (t2 − 10t + 7)x2 + t3(t − 1)x3 , 2t(t − 3)x + t2 (t − 6)x2 − t3 x3 , −t + 1 − t(2t − 7)x − t2 (t − 6)x2 + t3 x3 , 1). x1 = (−2t + 2 − t(t − 1)x, tx + 1, −tx − 1, 0), Algorithm for the Quillen-Suslin theorem: Computational version REQUIRE: A := K[x; σ, δ] and an idempotent matrix F ∈ Ms(A). 1: k := 0, F ′ := F ; 2: WHILE k < s − 1 DO 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: 14: 15: 16: 17: k := k + 1 IF max{deg(f ′ ij ) i = 1 or j = 1} = −∞ THEN F ′ := SubM atrix(F ′, 2..s, 2..s); ELSE (B): IF f ′ 11 = 0 THEN if (f ′ if (f ′ 1k 6= 0) F ′ := Tk1(−1)F ′Tk1(−1)−1; k1 6= 0) F ′ := T1k(−1)F ′T1k(−1)−1; END IF (B1): IF f ′ 11 ∈ K − {0} THEN Apply: OrderReduction1; ELSE Apply: (B2) OrderReduction2; END IF END IF 18: END WHILE 19: RETURN Matrices U, U −1, U F U −1; a basis X of hF i; process step by step. Example 3.3. In this example we will illustrate the computational version of the Quillen-Suslin algo- rithm; let M3(A), where A := K[x, σ, δ], K := Q(t), σ( p(t) q(t−1) and δ := 0; we have the idempotent matrix q(t) ) := p(t−1) 1+t x 2t − 2t(3+2t) 1+t x 3+2t 1+t x 1 − 2t 1 1+t x t 1+t x F =  2t (1+t)2 x (1+t)2 x −1 .   −t + t(3+2t) 1+t x 1 − t (1+t)2 x Let F ′ := F , along the example, we will replace the matrices F ′, U and U −1 for the new versions given by the procedures of the algorithm. Step 1. Since f ′ Step1.1 : The idea is to convert f ′ 11 = 1 − 2t 1,i = 0 for i > 2 and f ′ 1,2 6= 0. 1+t x, we will apply the reduction procedure of (B2), i.e, OrderReduction2: 7 Applying first T2,3( −1 t(1+2t) ), then T3,2(t(1 + 2t) − t(3+2t)(1+2t) 1+t x), and finally permuting the rows and columns 2 and 3, we get where Step 1.2. Since the new F ′ is U F U −1 =  1 − 2t 1+t x t(1+2t) 1+t x − 2t(1+2t) t+2 x2 2t (1+2t)(1+t) x 2 1+2t 0 2t(1+2t) (3+2t)(1+t) x 0 1 −2 (1+2t)2 ,   1 0 0 1 0 0 U =  U −1 =  F ′ =  0 t(1 + 2t) − t(3+2t)(1+2t) 1+t 1 0 x t(1+2t) −1 (1+t)2 x t(1+2t) 0 1 t(1+2t) 1 0 3+2t 1+t x −t(1 + 2t) + t(3+2t)(1+2t) 1+t ,     x . 1 − 2t 1+t x t(1+2t) 1+t x − 2t(1+2t) t+2 x2 2t (1+2t)(1+t) x 2 1+2t 2t(1+2t) (3+2t)(1+t) −2 (1+2t)2 , 0 0 1   we want to reduce the degree of f ′ 1,1; for this we apply T2,1( −t(1+2t) (1+t) x) and we obtain U F U −1 =  1 0 0 2 1+2t 0 −2 (1+2t)2 0 0 1  , where the new U and U −1 are U =  U −1 =  −t(1+2t) 1 1+t x t(1 + 2t) − t(3+2t)(1+2t) 0 1+t 0 1 0 x t(1+2t) −1 (1+t)2 x t(1+2t) 1 x 0 1 t(1+2t) 1+t 1+t x t(1+2t) 1 0 3+2t 1+t x −t(1 + 2t) + t(3+2t)(1+2t) x 1+t ,     . since f ′ 1,1 = 1 we apply (B1), i.e., OrderReduction1, for this we consider the matrices Step 2. The new F ′ is 1 0 0 S =  and then 2 0 0 1+2t 1 0 0 0 −2 (1+2t)2 F ′ = 1  ;    , and S−1 =  1 0 0 0 0 1 1+2t 2 1 0 S F ′S−1 =  1 0 0 0 0 −2 (1+2t)2 . 0 0 1  8 −2 1+2t 1 0 0 0 1   , Therefore, the new F ′ is F ′ =(cid:20) 0 −2 (1+2t)2 where the new U and U −1 are 0 1(cid:21) , and U F U −1 =  1 0 0 0 0 −2 (1+2t)2 , 0 0 1  U =  1 x t(1+2t) 1+t 1+t x U −1 =  1 − 2t 1+t x −t(1+2t) 2t − 2t(3+2t) 1+t x 1+t x t(1 + 2t) − t(3+2t)(1+2t) 0 1+t 1 2t (1+t)2 x x t(1+2t) (1+t)2 x −1 t(1+2t) ,   0 3+2t 1+t x −2 1+2t (1+t)(3+2t) x 2 1 t(1+2t) − 1 − 2t(1+2t) (1+t)(3+2t) x −t(1 + 2t) + t(3+2t)(1+2t) 1+t .   x Since f ′ 1,1 = 0, we apply T1,2(−1), we get U F U −1 =  1 0 0 where the new U and U −1 are 0 2 (1+2t)2 −2 (1+2t)2 0 −4t2−4t+1 (1+2t)2 4t2+4t−1 (1+2t)2   and F ′ =" 2 (1+2t)2 −2 (1+2t)2 −4t2−4t+1 (1+2t)2 4t2+4t−1 (1+2t)2 # , 2t − 2t(3+2t) 1+t x 1+t x 2t2 + t − 1 − t(3+2t)(1+2t) 0 1+t 1 x 1 − 2t 1+t x −t(1+2t) U =  U −1 =  2 1 1 1+t x t(1+2t) 1+t x −2 1+2t (1+t)(3+2t) x 2 1 t(1+2t) − 1 − 2t(1+2t) (1+t)(3+2t) x 1 t(1+2t) + t(1+2t) (1+t)2 x 2t (1+t)2 x −1 t(1+2t) ,   −2 1+2t (1+t)(3+2t) x t(1+2t) + 4t2+12t+7 1 −2t2 − t + 1 + t(1+2t)(4t2+12t+7) (1+t)(3+2t) .   x (1+2t)2 is invertible, we apply OrderReduction1 with matrices Since f ′ 1,1 = so T =(cid:20)1 −2t2 − 2t + 1 1 1 2 (1+2t)2 −2 (1+2t)2 4t2+4t−1 (1+2t)2 (1+2t)2 # , 2 2 (cid:21) and T −1 =" 0  1 0 0 1 0 0 T F ′T −1 =  0 0 . Thus, the new F ′ is where the new U and U −1 are F ′ =(cid:2)0(cid:3) and U F U −1 =  1 0 0 , 0 0 1 0 0 0  1 − 2t 1+t x −t(1+2t) 2t − 2t(3+2t) 2 − t(3+2t)(1+2t) 1+t x −t − 1 1+t x 2t2 + t − t(3+2t)(1+2t) 1+t x −t(1+2t) 1+t 1+t U =  9 2t (1+t)2 x x 1+2t 2t + t(1+2t) (1+t)2 x , x t(1+2t) (1+t)2 x   U −1 =  1 x t(1+2t) 1+t 1+t x Permuting, we have finally 0 −2 (1+t)(3+2t) x 1+2t − 2t(1+2t) (1+t)(3+2t) x 2t −2 1+2t 1 t(1+2t) 1 1+2t .   where the new U and U −1 are U F U −1 =  0 0 0 1 0 0 0 0 1  , −t(1+2t) 1 − 2t 1+t x −t(1+2t) U =  U −1 =  1+t x 2t2 + t − t(3+2t)(1+2t) x 1+t 2t − 2t(3+2t) 1+t x 2 − t(3+2t)(1+2t) 1+t x 1+2t 1+t x −t − 1 −2 1+2t 1 t(1+2t) 1 1+2t 1 1 1+t x t(1+2t) 1+t x 0 −2 (1+t)(3+2t) x 1+2t − 2t(1+2t) (1+t)(3+2t) x 2t ,   t(1+2t) (1+t)2 x (1+t)2 x 2t 2t + t(1+2t) (1+t)2 x   . Therefore, r = 2 and the last two rows of U conform a basis X = {x1, x 2}, of hF i, x1 = (1 − 2t 1+t x, 2t − 2t(3+2t) 1+t x, 2t (1+t)2 x), x2 = ( −t(1+2t) 1+t x, −t − 1 2 − t(3+2t)(1+2t) 1+t x, 1+2t 2t + t(1+2t) (1+t)2 x). Example 3.4. Let M4(A), where A := K[x, σ, δ], K := Q(t), σ(f (t)) := f (qt) and δ(f (t)) := f (qt)−f (t) , where q ∈ K − {0, 1}; we consider the idempotent matrix F := [F (1) F (2) F (3) F (4)], F (i) the ith column of F and a ∈ Q, where t(q−1) (−ta + 2 t) x − 2 a + 2 F (1) =  F (2) =  −t2qx2 tx + 2 −1 −2 tx + 2 −tx − 2 tx + 2 ,   ,   −t2qx2 + (ta − 4 t) x + 2 a − 1 Applying the algorithm we obtain −tx − 2 t2qx2 + (−ta + 4 t) x − 2 a + 2 F (3) = (cid:0)−2 t2qa + 3 t2q(cid:1) x2 +(cid:0)a2t − 8 ta + 8 t(cid:1) x + 2 a2 − 3 a + 1   F (4) = −t3q3x3 +(cid:0)−q2t2 − 3 t2q(cid:1) x2 + (−ta + t) x − 2 a + 2   −t3q3x3 +(cid:0)−q2t2 − 5 t2q(cid:1) x2 − 5 tx + 2 (ta − 2 t) x + 2 a − 2 t2qx2 + 2 tx − 1 tx + 2 . ,   tx + 1 0 t2qx2 + 2 tx − 1 t2qx2 + 3 tx −tx − 2 (−ta + 2 t) x − 2 a + 2 −t2qx2 − 2 tx + 2 t2qx2 + 2 tx − 1 t2qx2 + a − 1 U =  1 tx − 1 1 1 0 tx 10 tx + 1 ,   U −1 =   tx −1 −tx − 2 a − 1 −tx + a − 1 −t2 qx2 + (ta − 4 t) x + 2 a − 1 t3 q3 x3  − (−q + a − 4) t2 qx2 + (−3 ta + 3 t) x + 1 0 −1 0 −1 1 −tx − 2 tx + 2 t2 qx2 + 3 tx −t2 qx2 − 2 tx + 1 ,  U F U −1 =  0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 1 ,   Therefore, r = 2 and the last two rows of U conform a basis X = {x1, x 2}, of hF i, x1 = (tx − 1, 1, t2qx2 + a − 1, t2qx2 + 2 tx − 1), x2 = (1, 0, tx, tx + 1). 4 Some remarks about the implementation In this final section we present some comments about the implementation of the computational version of the Quillen-Suslin algorithm. Remark 4.1. The OrderReduction1 is based in the implementation of the procedure (B1) in the proof of Theorem 2.1; for the OrderReduction2, the following algorithm describes its functionality: Algorithm OrderReduction2 REQUIRE: A := K[x; σ, δ] and an idempotent matrix F ∈ Ms(A) with deg(f11) ≥ 1. 1: Make f1,j = 0 for j > 2 and f1,2 6= 0; 2: 3: 4: 5: 6: 7: 8: 9: 10: 11: 12: 13: Reduce degree of f1,1; IF f1,1 = 0 IF max{deg(fi,j ) > 0 i = 1 or j = 1} > 0 Make f1,j = 0 for j > 2 and f1,2 6= 0; F := P12 F P12; Apply: OrderReduction1; ELSE F ′ := SubM atrix(F, 2..s, 2..s); ENDIF ELSE Apply: OrderReduction1; ENDIF 14: RETURN Matrices U, U −1, F ′ and U F U −1 = (cid:20)α 0 F ′(cid:21), with α ∈ {0, 1}. 0 Remark 4.2. For the implementation of the Quillen-Suslin algorithm we used Mapler 2016, and we create a library called OrePolyToolKit.lib consisting in two packages: • OrePolyUtility: This is a new useful collection of functions for operating matrices, vectors and lists over an UnivariateOreRing K[x; σ, δ]; the UnivariateOreRing structure was taken from the library OreTools within the standard Maple libraries. • OrePolyQS: This is the most important new collection of functions related to the Quillen-Suslin algorithm over K[x; σ, δ]; the main routine of the algorithm was implemented here, the following functions of this package are fundamentals: 11 -- GenerateIdemp: This function generates idempotent matrices over K[x; σ, δ], the arguments are the matrix order and the UnivariateOreRing, and return an idempotent matrix of the given dimension over the respective UnivariateOreRing. -- QSAlgKsd: This is the main function of the algorithm, it shows the sequence of all steps of the Quillen-Suslin algorithm presented in this paper; the arguments are the idempotent matrix and the UnivariateOreRing, and return the matrix U F U −1 in the form of Theorem 2.1, the matrices U and U −1, the basis of hF i and the complete process step by step. References [1] Artamonov, V., Serre's quantum problem, Russian Math. Surveys, 53(4), 1998, 657-730. [2] Artamonov, V., On projective modules over quantum polynomials, Journal of Mathematical Sci- ences, 93(2), 1999, 135-148. [3] Bass, H., Proyective modules over algebras, Annals of Math. 73, 532-542, 1962. [4] Cohn, P., Free Ideal Rings and Localizations in General Rings, Cambridge University Press, 2006. [5] Gallego, C. and Lezama, O., Projective modules and Grobner bases for skew P BW extensions, Dissertationes Mathematicae, 521, 2017, 1-50. [6] Lam, T.Y., Serre's Problem on Projective Modules , Springer Monographs in Mathematics, Springer, 2006. [7] Quillen, D., Proyective modules over polynomial rings, Invent. Math., 36, 1976, 167-171. [8] Suslin, A.A., Proyective modules over polynomial rings are free, Soviet Math. Dokl., 17, 1976, 1160-1164. 12
1906.07134	1	1906	2019-06-17T17:15:45	Pre-Calabi-Yau algebras and double Poisson brackets	[ "math.RA" ]	We give an explicit formula showing how the double Poisson algebra introduced in \cite{VdB} appears as a particular part of a pre-Calabi-Yau structure, i.e. cyclically invariant, with respect to the natural inner form, solution of the Maurer-Cartan equation on $A\oplus A^*$. Specific part of this solution is described, which is in one-to-one correspondence with the double Poisson algebra structures. The result holds for any associative algebra $A$ and emphasizes the special role of the fourth component of a pre-Calabi-Yau structure in this respect. As a consequence we have that appropriate pre-Calabi-Yau structures induce a Poisson brackets on representation spaces $({\rm Rep}_n A)^{Gl_n}$ for any associative algebra $A$.	math.RA	math	Pre-Calabi-Yau algebras and double Poisson brackets Natalia Iyudu, Maxim Kontsevich, Yannis Vlassopoulos Abstract We give an explicit formula showing how the double Poisson algebra introduced in [16] appears as a particular part of a pre-Calabi-Yau structure, i.e. cyclically invariant, with respect to the natural inner form, solution of the Maurer-Cartan equation on A ⊕ A∗. Specific part of this solution is described, which is in one-to-one correspon- dence with the double Poisson algebra structures. The result holds for any associative algebra A and emphasizes the special role of the fourth component of a pre-Calabi-Yau structure in this respect. As a consequence we have that appropriate pre-Calabi-Yau structures induce a Poisson brackets on representation spaces (RepnA)Gln for any as- sociative algebra A. 9 1 0 2 n u J 7 1 ] . A R h t a m [ 1 v 4 3 1 7 0 . 6 0 9 1 : v i X r a 16A22, 16S37, 16Y99, 16G99, 16W10, 17B63 MSC: Keywords: A-infinity structure, pre-Calabi-Yau algebra, inner product, cyclic invariance, graded pre-Lie algebras, necklace bracket, Maurer-Cartan equation, double Poisson brackets. 1 Introduction We consider the structures introduced in [9], [10], [14], which are cyclically invariant with respect to the natural inner form solutions of the Maurer-Cartan equation on the algebra A ⊕ A∗, for any graded associative algebra A. This structure is called pre-Calabi-Yau algebra. We show, how the double Poisson bracket [16] appear as a particular part of a pre-Calabi-Yau structure. It was suggested in [8] that cyclic structure on A∞-algebra with respect to non-degenerate inner form should be considered as a symplectic form on the formal noncommutative manifold. Here we demonstrate that a natural inner form on A ⊕ A∗, namely pre-Calabi-Yau structure gives rise to a noncommutative version of a Poisson bracket. m(1) Indeed, we check that pre-Calabi-Yau structure of appropriate kind on A induce a Gln invariant Poisson bracket on the representation spaces (RepnA) of A. More precisely, we found the way to associate to a pre-Calabi-Yau structure, defined by an A∞-structure on A ⊕ A∗ = (A ⊕ A∗, m = ∞P ) a bracket, which satisfies all axioms of the double Poisson algebra [16]. This allows to i=2,i6=4 consider pre-Calabi-Yau structures as a noncommutative version of Poisson structures according to the ideology introduced and developed in [8, 11], saying that noncommutative structure should manifest as a corresponding commutative structure on representation spaces. We show how dou- m(1) ble Poisson bracket can be constructed from a pre-CY algebra A = (A ⊕ A∗, m = ) i ∞P i=2,i6=4 i with m4 = 0. This indicates the case when Jacobi identity can be obtained form Maurer-Cartan equation exactly, without any additional correcting terms. Thus to clarify the picture and to do 1 it for arbitrary algebras we consider this case separately here, in spite the noncmmutative Poisson structure can be obtained from an arbitrary pre-Calabi-Yau structure, without this restriction on m4 (as, for example, we show in []). The way we establish the correspondence between the two structures is the following. First, we associate to a solution of the Maurer-Cartan equation of type B the Poisson algebra structure. Most subtle point here is the choice of the definition of the bracket via the pre-Calabi-Yau structure. Theorem 1.1. Let we have A∞-structure on (A ⊕ A∗, m = ∞P i=2,i6=4 m(1) i ). Define the bracket by the formula hg ⊗ f, {{b, a}}i := hm3(a, f, b), gi, where a, b ∈ A, f, g ∈ A∗ and m3(a, f, b) = c ∈ A corresponds to the component of type B of the solution to the Maurer-Cartan, i.e. m3: A × A∗ × A → A. Then this bracket does satisfy all axioms of the double Poisson algebra. Moreover, pre-Calabi-Yau structures of type B (corresponding to the tensor A ⊗ A∗ ⊗ A ⊗ A∗ or A∗ ⊗ A ⊗ A∗ ⊗ A) with mi = 0, i > 4 are in one-to-one correspondence with the double Poisson brackets {{·, ·}} : A ⊗ A → A ⊗ A for an arbitrary associative algebra A. Here we concentrate on the non-graded version of the double Poisson structure. We show how it could be obtained as a part of the solution of the Maurer-Cartan equation on the algebra A ⊕ A∗, which is already a graded object otherwise the Maurer-Cartan equation would be trivial). Namely, we consider the grading on R = A ⊕ A∗, where R0 = A, R1 = A∗, and find it quite amazing how graded continuation of an arbitrary non-graded associative algebra can induce an interesting non- graded structure on A itself. In order the non-graded version of the double Poisson algebra to be induced on associative algebra A (sitting in degree zero) it have to be 2-pre-Calabi-Yau structure in case of this grading. Analogous results for an arbitrary graded associative algebra will be treated elsewhere. The contents of this paper is somewhat extended version of the preprint [4]. The structure of the paper is the following. We explain the notions of strong homotopy asso- ciative algebra (A∞-algebra) and of pre-Calabi-Yau structure on A∞ or graded associative algebra in Section 2. In Section 4 we suggest a way to define a double Poisson bracket out of the part of the solutions to the Maurer-Cartan equation corresponding to exclusively operations of the type A⊗A∗ ⊗A⊗A∗, that is of type B. The axioms (double Jacobi identity) of double Poisson bracket obtained in such a way from pre-CY structures with m4 = 0 are proved. This shows that pre-CY structures with m4 = 0 induce the double Poisson bracket. Moreover, for the structures of type B with mi = 0, i > 4 there is a one-to-one correspondence between those structures and Poisson brackets, defined by our formula from Theorem 4.2. In Section 5 we discuss how pre-Calabi-Yau structures via double Poisson bracket induce a Poisson structures on representation spaces of an arbitrary associative algebra. 2 Finite and infinite dimensional pre-Calabi-Yau alge- bras We deal here with the definition of a d-pre-Calabi-Yau structure on A∞-algebra. Further in the text we consider mainly pre-Calabi-Yau structures on an associative algebra A. Since in the definition of 2 pre-Calabi-Yau structure the main ingredient is A∞-structure on A⊕A∗ we start with the definition of A∞-algebra, or strong homotopy associative algebra introduced by Stasheff [15]. In fact, there are two accepted conventions of grading of an A∞-algebra. They differ by a shift in numeration of graded components. In one convention, we call it Conv.1, each operation has degree 1, while the other is determined by making the binary operation to be of degree 0, and the degree of operation of arity n, mn to be 2 − n. This second convention will be called Conv.0. If the degree of element x in Conv.0 is degx = x, then shifted degree in Ash = A[1], which fall into Conv.1, will be degshx = x′, where x′ = x − 1, since x ∈ Ai = A[1]i+1. The formulae for the graded Lie bracket, Maurer-Cartan equations and cyclic invariance of the inner form are different in different conventions. Since we mainly will use the Conv.1, but in a way need Conv.0 as well we sometimes present both of them. Let A be a Z graded vector space A = ⊕ n∈Z Hom(A[1]⊗l, A[1]), for l > 0, C •(A, A) = Q An. Let C l(A, A) be Hochschild cochains C l(A, A) = C l(A, A). k>1 On C •(A, A)[1] there is a natural structure of graded pre-Lie algebra, defined via composition: ◦ : C l1(A, A) ⊗ C l2(A, A) → C l1+l2−1(A, A) : f ◦ g(a1 ⊗ ... ⊗ al1+l2−1) = X(−1) g i−1 P j=1 aj f (a1 ⊗ ... ⊗ ai−1 ⊗ g(a1 ⊗ ... ⊗ ai+l2+1) ⊗ ... ⊗ al1+l2−1) The operation ◦ defined in this way does satisfy the graded right-symmetric identity: where (f, g, h) = (−1)gh(f, h, g) (f, g, h) = (f ◦ g) ◦ h − f ◦ (g ◦ h). As it was shown in [?] the graded commutator on a graded pre-Lie algebra defines a graded Lie algebra structure. Thus the Gerstenhaber bracket [−, −]G: [f, g]G = f ◦ g − (−1)f gg ◦ f makes C •(A) into a graded Lie algebra. Equipped with the derivation d = ad m2, (C •(A), m2) becomes a DGLA, which is a Hochschild cohomological complex. With respect to the Gerstenhaber bracket [−, −]G we have the Maurer-Cartan equation [m(1), m(1)]G = X p−1X (−1)εmp(x1, . . . , xi−1, mq(xj , . . . , xi+q−1), . . . , xk) = 0, (2.1) p+q=k+1 i=1 where ε = x1′ + . . . + xi−1′, xi′ = xi − 1 = degxi − 1 The Maurer-Cartan in Conv.0 is: 3 [m(1), m(1)] = X p−1X p+q=k+1 i=1 where (−1)εmp(x1, . . . , xi−1, mq(xj , . . . , xi+q−1), . . . , xk) = 0, (2.2) ε = i(q + 1) + q(x1 + . . . + xi−1, Definition 2.1. An element m(1) ∈ C •(A, A)[1] which satisfies the Maurer-Cartan equation [m(1), m(1)]G with respect to the Gerstenhaber bracket [−, −]G is called an A∞-structure on A. Equivalently, it can be formulated in a more compact way as a coderivation on the coalgebra of the bar complex of A. In particular, associative algebra with zero derivation A = (A, m = m(1) 2 ) is an example of A∞- algebra. The component of the Maurer-Cartan equation of arity 3, M C3 will say that the binary operation of this structure, the multiplication m2 is associative: (ab)c − a(bc) = dm3(a, b, c) + (−1)σm3(da, b, c) + (−1)σm3(a, db, c) + (−1)σm3(a, b, dc) We can give now definition of pre-Calabi-Yau structure (in Conv.1). Definition 2.2. A d-pre-Calabi-Yau structure on a finite dimensional A∞-algebra A is (I). an A∞-structure on A ⊕ A∗[1 − d], (II). cyclic invariant with respect to natural non-degenerate pairing on A ⊕ A∗[1 − d], meaning: hmn(α1, ..., αn), αn+1i = (−1)α1′(α2′+...+αn+1′)hmn(α2, ...αn+1), α1)i where the inner form h, i on A + A∗ is defined naturally as h(a, f ), (b, g)i = f (b) + (−1)g′a′ for a, b ∈ A, f, g ∈ A∗ g(a) (III) and such that A is A∞-subalgebra in A ⊕ A∗[1 − d]. The signs in this definition written in Conv.1, are assigned according to the Koszul rule. It is not quite the case for Conv.0, where the cyclic invariance with respect to the natural non-degenerate pairing on A ⊕ A∗[1 − d], from (II) sounds: hmn(α1, ..., αn), αn+1i = (−1)n+α1′(α2′+...+αn+1′)hmn(α2, ...αn+1), α1i The appearance of the arity n, which influence the sign in this formula, does not really fit with the Koszul rule, this is the feature of the Conv.0, and this is why it is more convenient to work with the Conv.1. As we will need to refer to these later, let us define separately the cyclic invariance condition and inner form symmetricity in Conv.1: hmn(α1, ..., αn), αn+1i = (−1)α1′(α2′+...+αn+1′)hmn(α2, ...αn+1), α1)i hx, yi = −(−1)x′ y′ hy, xi (2.3) (2.4) The notion of pre-Calabi-Yau algebra introduced in [10], [14] use the fact that A is finite dimensional, since there is no natural grading on the dual algebra A∗ = Hom(A, K), induced 4 form the grading on A in infinite dimensional case. The general definition suitable for infinite dimensional algebra was given in [10], [9], and it is equivalent to the definition, where the Hom(A, K) is substituted with the graded version: A∗ = ⊕(An)∗ = Hom(A, K), if graded components are finite dimensional. We will give this general definition and show the equivalence further in this section. Example. The most simple example of pre-Calabi-Yau structure demonstrates that this struc- ture does exist on any associative algebra. Namely, the structure of associative algebra on A can be extended to the associative structure on A ⊕ A∗[1 − d] in such a way, that the natural inner form is (graded)cyclic with respect to this multiplication. This amounts to the following fact: for any A-bimodule M the associative multiplication on A ⊕ M is given by (a + f )(b + g) = ab + af + gb. In this simplest situation both structures on A and on A + A∗ are in fact associative algebras. More examples one can find in [3], [13], [1]. One can reformulate the above definition without A∗, using the inner product, to change inputs and outputs of operations, and by this to substitute A∗ with A, as it was done in [10]. The A∞-structure on A ⊕ A∗ means first of all the bunch of linear maps Such a map splits as a collection of linear maps of the type mN : (A ⊕ A∗)N → A ⊕ A∗. ξ = mq1,...,ql p1,...,pl : A⊗p1 ⊗ A∗⊗q1 ⊗ ... ⊗ A⊗pl ⊗ A∗ ⊗ ql → A (or A∗) where P pi + qi = N , 0 6 pi 6 N . These could be interpreted, using the inner product, as tensors of the type, A⊗p1 ⊗ A∗⊗q1 ⊗ ... ⊗ A⊗pl ⊗ A∗ ⊗ ql, P pi + qi = N + 1, and graphically depicted as operations where incoming edges correspond to elements of A, outgoing edges to elements of A∗ and the marked point correspond to the output of operation mq1,...,ql p1,...,pl. (Of course, due to cyclic invariance operations with different marked points are equal up to a sign, but to keep track of the signs, marked point is needed). This gives rise to the definition below. First, we shell define higher Hochschild cochains and generalised necklace bracket. Definition 2.3. For k > 1 the space of k-higher Hochschild cochains is defined as C (k)(A) := Y Hom(A[1]⊗r, A[1]⊗k) = Y r1,...,rk>0 r>0 Hom( k ⊗ i=1 A[1]⊗ri , A[1]⊗k) Denote by C (•)(A) = Qk>1 C (k)(A) the space of all higher Hochschild cochains. Note, that C (1)(A) = C •(A, A) is the space of usual Hochschild cochains. We can see that element of the higher Hochschild cochain can be interpreted as operation with r incoming and k outgoing edges. There is a marked point in the picture as well, and due to the cyclic invariance condition one can move this marked point with the change of a sign. Indeed, suppose that the last position is marked, then it can be moved to the one but last using the formula: (−1)xn(x1+...+xn−1) < x1, m(x2, ..., xn) >=< xn, m(x1, ..., xn−1) > where xi ∈ A or A∗. Thus, just higher Hochschild cochains, without a specified point will appear in the definition of pre-Calabi-Yau structure. The composition of two operations of this kind translates according to definition 2.2 to the explained above picture via the notion of generalised necklace bracket: 5 Definition 2.4. The generalised necklace bracket between two elements f, g ∈ C (k)(A) is given as [f, g]gen.neckl = f ◦ g − (−1)σg ◦ f, where composition f ◦ g consists of inserting all outputs of g to all inputs from f with signs assigned according to the Koszul rule. Again, since the defined above composition f ◦ g makes C (•) into a graded pre-Lie algebra, the generalised necklace bracket obtained from it as a graded commutator, makes C (•) into a graded Lie algebra. Definition 2.5. Let A be a Z-graded space A = ⊕An. The pre-Calabi-Yau structure on A is an element from the space of higher Hochschild cochains C (•), m = Pk>0 m(k), m(k) ∈ C (k)(A), which is a solution to the Maurer-Cartan equation [m, m]gen.neckl = 0 with respect to generalised necklace bracket. Any such solution makes C (•)(A) into a DGLA with the differential ad m. Definition 2.6. The pre-Calabi-Yau structure on a Z-graded space A = ⊕An is a cyclically invariant A∞-structure on A ⊕ A∗[1 − d], where A∗ is understood as A∗ = ⊕(An)∗ = Hom(A, K). Proposition 2.7. The definitions of pre-Calabi-Yau structures 2.5 and 2.6 are equivalent, when dimAn < ∞ ∀n. Proof. To demonstrate this we will start with an element m = Pk>0 m(k), m(k) ∈ C (k)(A), m(k) = A⊗r → A⊗k depicted as an operation with r incoming arrows, k outgoing arrows, and one marked point. From this data we construct a collection of operations mn : (A ⊕ A∗)⊗n → A ⊕ A∗, to form an A∞-structure on A ⊕ A∗. So let us have an element ξ in the tensor product (in some order) of r copies of A and k copies of A∗, where the last position is specified. Thus we have an operation E : A⊗r → A⊗k with one fixed entry. This defines an element bξ ∈ (A∗)⊗r ⊗ A⊗k (by means of the natural pairing) such that hE(a1 ⊗ ... ⊗ ar), f1 ⊗ ... ⊗ fki = hbξ, a1 ⊗ ... ⊗ ar ⊗ f1 ⊗ ... ⊗ fki Note that here we use the equality A∗∗ = A, which is true only for finite dimensional spaces. We should make sure that we use duals satisfying A∗∗ = A, as it is done in definition 2.6, when graded components are finite dimensional. Now we can define an operation from the A∞-structure on A ⊕ A∗ corresponding to the above operation E, if the marked point have an outgoing edge and mn−1(a1, f1, . . . , bfk) if the marked point has an incoming edge. Here n = k + r and the order of entries of elements from A and from A∗ is dictated by the order in ξ. In these two cases we define mn−1 as follows: mn−1(a1, f1, . . . , bar) hfk, mn−1(a1, f1, . . . , bfk)i = hbξ, a1 ⊗ · · · ⊗ ar ⊗ f1 ⊗ · · · ⊗ fki; har, mn−1(a1, f1, . . . , bar)i = hbξ, a1 ⊗ · · · ⊗ ar ⊗ f1 ⊗ · · · ⊗ fki. In spite definition 2.5 looks more beautiful and reveals nice graphically presented connection with A-infinity structure, we will use definition 2.6, since we find it easier to work with and make sure all details are correct. 6 3 Structure of the Maurer-Cartan equations The general Maurer-Cartan equations on C = A ⊕ A∗ for the operations mn : C[1]n → C[1] have the shape X p−1X (−1)εmp(x1, . . . , xi−1, mq(xj , . . . , xi+q−1), . . . , xk), p+q=k+1 i=1 where ε = x1′ + . . . + xi−1′, xs′ = degxs − 1 The equations we get from the Maurer-Cartan in arities four and five, which are relevant for comparing with the Leibniz and Jacibi identities for the double bracket, will look as follows. In arity 4, Maurer-Cartan equation, M C4 reads: m3(x1x2, x3, x4) + (−1)x1′ m3(x1, x2x3, x4) + (−1)x1′+x2′ m3(x1, x2, x3x4)+ (−1)x1′ m2(x1, m3(x2, x3, x4)) + m2(m3(x1, x2, x3), x4) = 0 In arity 5, Maurer-Cartan equation, M C4 reads: m3(m3(x1, x2, x3), x4, x5) + (−1)x1′ m3(x1, m3(x2, x3, x4), x5)+ (−1)x1′+x2′ m3(x1, x2, m3(x3, x4, x5)) = 0 Operations of arity 4 are absent due to our condition that m4 = 0 in A∞-structure on A ⊕ A∗. Since we have Maurer-Cartan equations on A ⊕ A∗, it essentially means that any equation splits into the set of equations with various distributions of inputs/outputs from A and A∗. Note that solutions of the Maurer-Cartan which are interesting for us correspond to operations A⊗A → A⊗A (which can serve as a double bracket). These are operations from tensors with exactly two Ath and two A∗th. Remind that an operation, say, A∗ × A∗ × A → A∗ can be naturally interpreted as an element of the space A ⊗ A ⊗ A∗ ⊗ A∗ and this tensor due to cyclic invariance of the structure equals to its cyclic permutations up to sign, in this case A∗ ⊗ A ⊗ A ⊗ A∗, A∗ ⊗ A∗ ⊗ A ⊗ A and A ⊗ A∗ ⊗ A∗ ⊗ A. There is another type of tensor from A ⊗ A∗ ⊗ A ⊗ A∗ for which there is only one cyclic permutation A∗ ⊗ A ⊗ A∗ ⊗ A. Due to cyclic invariance hm3(f, a, g), bi = ±hm3(b, f, a), gi operation A∗ × A × A∗ → A∗ corresponding to tensor A ⊗ A∗ ⊗ A ⊗ A∗ is the same as operation A × A∗ × A → A corresponding to tensor A∗ ⊗ A ⊗ A∗ ⊗ A. These tensors encode the second type of operations. Two types of operations mentioned above which are different up to cyclic permutation on tensors will serve as variables in the equations we obtain from the Maurer-Cartan. Let us list 6 tensors corresponding to 2 types of operations, of which we will think as of two types of main variables in MC equations. Type A A∗ ⊗ A ⊗ A ⊗ A∗, A × A∗ × A∗ → A∗, A∗ ⊗ A∗ ⊗ A ⊗ A, A × A × A∗ → A, A ⊗ A∗ ⊗ A∗ ⊗ A, A∗ × A × A → A, A ⊗ A ⊗ A∗ ⊗ A∗, A∗ × A∗ × A → A∗. Type B A ⊗ A∗ ⊗ A ⊗ A∗, A∗ × A × A∗ → A∗, A∗ ⊗ A ⊗ A∗ ⊗ A, A × A∗ × A → A, 7 Definition 3.1. We say that operations corresponding to the tensor A⊗A⊗A∗ ⊗A∗ (and its cyclic permutations) are operations of type A, and operations corresponding to the tensor A⊗ A∗ ⊗ A⊗ A∗ (and its cyclic permutations) are operations of type B. These two variables, being cyclicly invariant tensors can be depicted as follows: Graphically these cyclically invariant operations could be depicted as follows for type A and B respectively. A A∗ A A∗ A∗ A A A∗ But we will mainly use for calculations the above row notations, since they are more suitable for following the signs, which are crucially important in some of our calculations, for example, for the result on one-to-one correspondence between part of pre-CY structure and a double Poisson structure. The other operations which are also variables in the Maurer-Cartan equation, correspond to the tensors of length four, containing not exactly two A and two A∗. We call them secondary type variables, as opposed to the main type, consisting of variables of type A and B. So, secondary type variables correspond to the cyclic invariant tensors with one A∗: A ⊗ A ⊗ A ⊗ A∗, we call this C1 or with one A: A∗ ⊗ A∗ ⊗ A∗ ⊗ A, this we call C2 (there are eight corresponding operations), as well as two operations corresponding to each of tensors A ⊗ A ⊗ A ⊗ A and A∗ ⊗ A∗ ⊗ A∗ ⊗ A∗, which are variables C3 and C4 respectively. Graphically these cyclically invariant operations could be depicted as follows. C1, C2 : C3, C4 : A A A A∗ A A∗ A∗ A∗ 8 A∗ A A A A A∗ A∗ A∗ Again, for our calculations it will be more convenient to present them as operations, written in a row (e.i. with the fixed starting point), so in these denotations we have the following operations corresponding to variables C1, C2, C3, C4: variable C4 Secondary type variable C3 A ⊗ A ⊗ A ⊗ A∗, A∗ × A∗ × A∗ → A∗, A∗ ⊗ A∗ ⊗ A∗ ⊗ A, A × A × A → A, A∗ ⊗ A ⊗ A ⊗ A, A × A∗ × A∗ → A, A ⊗ A∗ ⊗ A ⊗ A, A∗ × A × A∗ → A, A ⊗ A ⊗ A∗ ⊗ A, A∗ × A∗ × A → A, variable C1 A ⊗ A ⊗ A ⊗ A, A∗ × A∗ × A∗ → A, A∗ ⊗ A∗ ⊗ A∗ ⊗ A∗, A × A × A → A∗. A ⊗ A∗ ⊗ A∗ ⊗ A∗, A∗ × A × A → A∗, A∗ ⊗ A ⊗ A∗ ⊗ A∗, A × A∗ × A∗ → A∗, A∗ ⊗ A∗ ⊗ A ⊗ A∗, A × A × A∗ → A∗, variable C2 Let us look at what we can get from the Maurer-Cartan in arity 5. First consider the input row containing 4 or more entries from A (or A∗). It is easy to check that in this case all terms of equations we get contain secondary type variables. For example, consider the input A, A, A∗, A, A. The term m3(m3(a, f, b), c, d) is zero if m3(a, b, f ) ∈ A, since A is associative algebra and m3(a1, a2, a3) = 0 for all a1, a2, a3 ∈ A. If m3(a, b, f ) ∈ A∗ then the operation is of secondary type, from tensor A∗ ⊗ A∗ ⊗ A ⊗ A∗. Another group of equations correspond to input containing three A (or A∗). These are divided according to what is the output of the corresponding operation of arity 5. In case of operations with 3 inputs from A, two inputs from A∗ and output from A∗ as well as 3 inputs from A∗, two inputs from A and output from A, all terms of the equations still contain at least one variable of secondary type. This property of equations will allow us to restrict any solution of the Maurer-Cartan to the ones containing only main variables (take the projection of solution to the space of main variables, and ensure that we have a solution again). In the cases of operations with 3 inputs from A, two inputs from A∗ and output from A as well as 3 inputs from A∗, two inputs from A and output from A∗, all terms of the equations contain only variables of the main type. Each of these cases corresponds to 10 (5 choose 2) equations on main variables. We consider their structure in more detail. These equations on main variables contain both variables of types A and of B. Call variables of type B by X's and of type A by Y 's. Then the system of equations again splits into those, each term of which contains a Y variable and those which are equations only on X's. Lemma 3.2. Any equation on main variables coming from M C5 either containing only terms XX, i.e. only variables of type B or each term contains at least one variable Y - variable of type A. 9 Proof. Let us see from which inputs terms of type XX can appear. There are two operations of type B: I. A × A∗ × A → A and II. A∗ × A × A∗ → A∗. Consider the case of composition of the type m3(x1, m3(x2, x3, x4), x5)). In case I. to have a composition of two operations of type B we forced to start with input row A∗(AA∗A)A∗. In case II. to have a composition of two operations of type B we forced to start with input row A(A∗AA∗)A. The remaining types of compositions: m3(m3(x1, x2, x3), x4, x5)) and m3(x1, x2, m3(x3, x4, x5)) analogously give the same result. Thus the only rows of inputs from which XX term can appear are those two rows. We see moreover that no compositions containing variable Y (operation of type A) appear from this row of input. Thus variables X and Y are separated in the above sense in this system of equations. This structure of the system of equations on operations which constitute an unwrapped Maurer- Cartan equation will be a key to relate any pre-Calabi-Yau structure concentrated in appropriate arities to the double Poisson bracket. Each equation that we get from M C5 consists of 'quadratic' terms, meaning terms involving two operations. This system of equations has the feature that in no equation both terms containing two X variables and XY or Y Y terms appear. These terms are separated. The above arguments allow us to see that Proposition 3.3. Projection of any M C5 solution to the B-type component is also a solution of M C5. 4 Solutions to the Maurer-Cartan equations in arity four and five and double Poisson bracket In this section we show that the pre-Calabi-Yau structures of type B, namely the ones which are solutions of type B (corresponding to the tensor A ⊗ A∗ ⊗ A ⊗ A∗ or A∗ ⊗ A ⊗ A∗ ⊗ A) of the Maurer-Cartan equation on A ⊕ A∗, are in one-to-one correspondence with the non-graded double Poisson brackets. We choose the main example of grading on A + A∗ in order to get correspondence with the non-graded double Poisson bracket. Namely, in order to have multiplication on A to be of degree 0 (as it should be in Conv.0 ), we have to have A0 = A. Then, in order for the type B operations (the most interesting part of the solution of the Maurer-Cartan, which is a ternary operation) to make sense, i.e. according to the Conv.0, to be of degree −1, we need A∗ to be in the component of degree 1. That is, R = A ⊕ A∗ is graded by R0 = A and R1 = A∗. Now we shift this grading by one, to use more convenient formulae of Conv.1. Thus we get in 0 = A∗, that is A will have Ash = A[1] Ash degree −1, and A∗, degree 0, when we are in shifted situation, and in Conv.1. −1 = A, and Rsh = Ash + A∗sh is graded by Rsh −1 = A, Rsh Let A be an arbitrary associative algebra A = (A, m = m(1) 2 ) with a pre-Calabi-Yau struc- ture given as a cyclicly symmetric A∞-structure on A ⊕ A∗: (A ⊕ A∗, m = m(1) 3 ). We define the double Poisson bracket via the pre-Calabi-Yau structure, more precisely its component corresponding to the tensor A ⊗ A∗ ⊗ A ⊗ A∗, as follows. 2 + m(1) Definition 4.1. The double bracket is defined as: hg ⊗ f, {{b, a}}i := hm3(a, f, b), gi, where a, b ∈ A, f, g ∈ A∗ and m3(a, f, b) = c ∈ A corresponds to the component of m3: A × A∗ × A → A. 10 By choosing this definition we set up a one-to-one correspondence between pre-Calabi-Yau structures of type B and double Poisson brackets from [16]. This choice have been done in such a way that it would be possible, having the Maurer-Cartan equation to show, that the double bracket defined above indeed satisfies all axioms of double Poisson bracket. Moreover, no other identities follows in case mi = 0, i > 4. Note, that it is most subtle point, since there are many possibilities for this choice. We will check that double bracket defined in this way satisfies all axioms of the double Poisson bracket. Anti-symmetry: {{a, b}} = −{{b, a}}op (4.1) bi ⊗ ci, then {{b, a}}op = i (4.2) (4.3) Here {{b, a}}op means the twist in the tensor product, i.e. if {{b, a}} = P P ci ⊗ bi. i Double Leibniz: and double Jacobi identity: {{a, bc}} = b{{a, c}} + {{a, b}}c {{a, {{b, c}}}}L + τ(123){{b{{c, a}}}}L + τ(132){{c{{a, b}}}}L Here for a ∈ A ⊗ A ⊗ A, and σ ∈ S3 The {{ }}L defined as τσ(a) = aσ−1 (1) ⊗ aσ−1(2) ⊗ aσ−1(3). {{b, a1 ⊗ an}}L = {{b, a1}}L ⊗ a1 ⊗ ... ⊗ an Theorem 4.2. Let we have A∞-structure on (A ⊕ A∗, m = ∞P i=2,i6=4 m(1) i ). Define the bracket by the formula hg ⊗ f, {{b, a}}i := hm3(a, f, b), gi, where a, b ∈ A, f, g ∈ A∗ and m3(a, f, b) = c ∈ A corresponds to the component of type B of the solution to the Maurer-Cartan, i.e. m3: A × A∗ × A → A. Then this bracket does satisfy all axioms of the double Poisson algebra. Moreover, pre-Calabi-Yau structures of type B (corresponding to the tensor A ⊗ A∗ ⊗ A ⊗ A∗ or A∗ ⊗ A ⊗ A∗ ⊗ A) with mi = 0, i > 4 are in one-to-one correspondence with the double Poisson brackets {{·, ·}} : A ⊗ A → A ⊗ A for an arbitrary associative algebra A. Proof. Anti-symmetry of the double Poisson bracket reads in these notations: So we need to check that hf ⊗ g, {{b, a}}i = −hg ⊗ f, {{a, b}}i, hm3(b, g, a), f i = −hm3(a, f, b), gi. 11 Indeed, using cyclic invariance, we have hf ⊗ g, {{b, a}}i = hm3(a, f, b), gi = (−1)b′(g′+a′+f ′)hm3(g, a, f ), bi = = (−1)b′(g′+a′+f ′)(−1)g′(a′+f ′+b′)hm3(a, f, b), gi = −hm3(a, f, b), gi = −hg ⊗ f, {{a, b}}i We used the fact that in our grading ∀f ∈ A∗, f ′ = 0 and ∀a ∈ A, a′ = −1. By this the anti-symmetry of obtained in this way from pre-Calabi-Yau structure (non-graded) bracket is proven. Now we deduce the Leibnitz identity from the part of the arity 4 of the Maurer -- Cartan equations with inputs from A,A, A∗ and A. General Maurer -- Cartan in arity 4 reads: m3(x1x2, x3, x4) + (−1)x1′ m3(x1, x2x3, x4) + (−1)x1′+x2′ m3(x1, x2, x3x4)+ (−1)x1′ x1m3(x2, x3, x4) + m3(x1, x2, x3)x4 = 0 Applying this to the input a, b, f, c from A,A, A∗, A we have m3(ab, f, c) + (−1)a′ m3(a, bf, c) + (−1)a′+b′ m3(a, b, f c) + (−1)a′ am3(b, f, c) + m3(a, b, f )c = 0. Since we consider solutions containing only B-type components, two terms in the equation (m3(a, b, f )c and m3(a, b, f c)) vanish, leaving us with m3(ab, f, c) − m3(a, bf, c) − am3(b, f, c) = 0. (4.4) after we applied our grading, where a′ = −1 for all a ∈ A. Now we pair the above equality obtained from M C4 with g (the equality holds if and only if it holds for any pairing with an arbitrary g ∈ A∗): hm3(ab, f, c), gi − hm3(a, bf, c), gi − ham3(b, f, c), gi = 0. and express the three terms appearing there via the double bracket. For doing this we need the following lemma. Lemma 4.3. The following equalities hold: (R) hg ⊗ af, {{b, c}}i = hg ⊗ f, {{b, c}}ai (L) hga ⊗ f, {{b, c}}i = −hg ⊗ f, a{{b, c}}i Proof. We use here Sweedler notations: {{b, c}} = P bi ⊗ ci = P b′ ⊗ c′′. (R) hg ⊗ af, {{b, c}}i = Xhg ⊗ af, b′ ⊗ c′′i = Xhg, b′ihaf, c′′i = Xhg, b′ihf, c′′ai = hg ⊗ f,X b′ ⊗ c′′ai = hg ⊗ f, {{b, c}}ai (4.5) 12 We use here: (2.3) = (−1)a′(f ′+c′′′)hf c′′, ai haf, c′′i = (−1)a′(f ′+c′′′)(−1)f ′(c′′′+a′)hc′′a, f i (2.3) (2.4) = (−1)a′(f ′+c′′′)(−1)f ′(c′′′+a′) · −(−1)c′′a′f ′ hf, c′′ai and in our grading, where for all a ∈ A, f ∈ A∗, a′ = −1, f ′ = 0, we get haf, c′′i = hf, c′′ai (4.5) for all a, c′′ ∈ A, f ∈ A∗ (L) hga ⊗ f, {{b, c}}i = Xhga ⊗ f ihb′ ⊗ c′′i = = Xhab′, gihf, c′′i Xhga ⊗ b′ihf ⊗ c′′i (4.6) (4.7) = −Xhg, ab′ihf, c′′i = −Xhg ⊗ f, ab′ ⊗ c′′i = −hg ⊗ f, a{{b, c}}i We use here: hga, b′i (2.3) = (−1)g′(a′+b′′)hab′, gi ha, b′gi (2.4) = (−1)ab′′g′ hg, ab′i and in our grading, where for all a ∈ A, f ∈ A∗, a′ = −1, f ′ = 0, we get and respectively, for all a, b′ ∈ A, g ∈ A∗ hga, b′i = hab′, gi hab′, gi = −hg, ab′i (4.6) (4.7) Now we are ready to express three terms of the Maurer-Cartan equation 4.4 via the bracket. hm3(ab, f, c), gi hm3(a, bf, c), gi ham3(b, f, c), gi def = hg ⊗ f, {{c, ab}}i; def = hg ⊗ bf, {{c, a}}i R=hg ⊗ f, {{c, a}}bi; cycl.m2= −hm3(b, f, c), gai def = −hga ⊗ f, {{c, b}}i L =hg ⊗ f, a{{c, b}}i According to these the Maurer-Cartan can be rewritten as hg ⊗ f, {{c, ab}}i − hg ⊗ f, {{c, a}}bi − hg ⊗ f, a{{c, b}}i which is exactly the Leibniz identity: 13 Now it remains to prove that the double bracket defined via the solution of the Maurer-Cartan (of type B) as {{c, ab}} = {{c, a}}b + a{{c, b}} hg ⊗ f, {{b, a}}i = hm(a, f, b), gi for all a, b ∈ A, f, g ∈ A∗, does satisfy the Jacobi identity. The appropriate part of the Maurer-Cartan equation to consider is the part of arity 5, with inputs from A,A∗, A, A∗ and A. General Maurer -- Cartan in arity 5 reads: (−1)0m3(m3(x1, x2, x3), x4, x5) + (−1)x1′ m3(x1, m3(x2, x3, x4), x5)+ m3(x1, x2, m3(x3, x4, x5)) = 0 Applying this to the input a, f, b, g, c from A,A∗, A, A∗, A we get (−1)x1′+x2′ m3(m3(a, f, b), g, c) + (−1)a′ m3(a, m3(f, b, g), c) + (−1)a′ (−1)a′+f ′ m3(a, f, m3(b, g, c)) = 0. Thus from the Maurer-Cartan we have. m3(m3(a, f, b), g, c) − m3(a, m3(f, b, g), c) − m3(a, f, m3(b, g, c)) = 0. (4.8) Since we are going to prove the double Jacobi identity: {{a, {{b, c}}}}L + τ123{{b, {{c, a}}}}L + τ132{{c, {{a, b}}}}L = 0. we need to express double commutators via the operations - solutions of the Maurer-Cartan equation. Lemma 4.4. For any a, b, c ∈ A and α, β, γ ∈ A∗ the from the definition 4.1 it follows: hα ⊗ β ⊗ γ, {{a, {{b, c}}}}Li = hm3(m3(c, γ, b), β, a), αi Proof. hα ⊗ β ⊗ γ, {{a, {{b, c}}}}Li = hα ⊗ β, {{a, hid ⊗ γ, {{b, c}}i}}i = hm3(hid ⊗ γ, {{b, c}}i, β, a), αi = hm3(m3(c, γ, b), β, a), αi Clearly (4.8) is equivalent to hm3(m3(a, f, b), g, c), hi − hm3(a, m3(f, b, g), c), hi − hm3(a, f, m3(b, g, c)), hi = 0. (4.9) for any a, b, c ∈ A and f, g, h ∈ A∗. By Lemma (4.4), the first summand in (4.9) is given by hm3(m3(a, f, b), g, c), hi = hh ⊗ g ⊗ f, {{c, {{b, a}}}}Li. (4.10) 14 We show now that the second term in (4.9) is expressed via double commutator as: hm3(a, m3(f, b, g), c), hi = −hg ⊗ f ⊗ h, {{b, {{a, c}}}}Li. (4.11) Indeed, using cyclic invariance and graded symmetry of the inner product, we see hm3(a, m3(f, b, g), c), hi = (−1)a′(m(f,b,g)′+c′+h′)hm3(m3(f, b, g), c), h), ai = (−1)a′(m(f,b,g)′+c′+h′)(−1)m(f,b,g)′(c′+h′+a′)hm3(c, h, a), m3(f, b, g))i = (−1)a′(m(f,b,g)′+c′+h′)(−1)m(f,b,g)′(c′+h′+a′)·−(−1)m(c,h,a)′m(f,b,g)′ hm3(f, b, g), m3(c, h, a)i = (−1)a′(m(f,b,g)′+c′+h′)(−1)m(f,b,g)′(c′+h′+a′) · −(−1)m(c,h,a)′m(f,b,g)′ (−1)f ′(b′+g′+m(c,h,a)′)hm3(b, g, m3(c, h, a)), f )i = (−1)a′(m(f,b,g)′+c′+h′)(−1)m(f,b,g)′(c′+h′+a′) · −(−1)m(c,h,a)′m(f,b,g)′ (−1)f ′(b′+g′+m(c,h,a)′)(−1)b′(g′+m(c,h,a)′+f ′)hm3(g, m3(c, h, a), f ), b)i = (−1)a′(m(f,b,g)′+c′+h′)(−1)m(f,b,g)′(c′+h′+a′) · −(−1)m(c,h,a)′m(f,b,g)′ (−1)f ′(b′+g′+m(c,h,a)′)(−1)b′(g′+m(c,h,a)′+f ′)(−1)g′(m(c,h,a)′+f ′+b′)hm3(m3(c, h, a)), f, b), gi. Taking into account that in our grading m(f, b, g)′ = f ′ + b′ + g′ + 1 = 0 and m(a, f, b)′ = a′ + f ′ + b′ + 1 = −1 for all a, b ∈ A, f, g ∈ A∗ we see that the latter sign is '-', hence we get the required: hm3(a, m3(f, b, g), c), hi = −hg ⊗ f ⊗ h, {{b, {{a, c}}}}Li, since due to Lemma 4.4 hm3(m3(c, h, a)), f, b), gi = hg ⊗ f ⊗ h, {{b, {{a, c}}}}L i. Now we consider the third term in (4.9) and show that it is expressed via double commutator as: hm3(a, f, m3(b, g, c))), hi = −hf ⊗ h ⊗ g, {{a, {{c, b}}}}Li. (4.12) Indeed, using cyclic invariance, we see hm3(a, f, m3(b, g, c)), hi = (−1)a′(f ′+m(b,g,c)′+h′)hm3(f, m3(b, g, c), h), ai = (−1)a′(f ′+m(b,g,c)′+h′)(−1)f ′(m(b,g,c)′+h′+a′)hm3(m3(b, g, c), h, a), f i Taking into account signs in our grading, we see that the letter sign is "-", thus by 4.4 (−1)a′(f ′+m(b,g,c)′+h′)(−1)f ′(m(b,g,c)′+h′+a′)hm3(m3(b, g, c), h, a), f i −hm3(m3(b, g, c), h, a), f i = −hf ⊗ h ⊗ g, {{a, {{c, b}}}}Li. 15 Thus 4.9 can be rewritten as: hh ⊗ g ⊗ f, {{c, {{b, a}}}}Li + hg ⊗ f ⊗ h, {{b, {{a, c}}}}Li + hf ⊗ h ⊗ g, {{a, {{c, b}}}}L i = 0 We see that permutations of functionals f, g, h ∈ A∗ in our formulas match with the permutation on the images of the bracket in the double Jacobi identity: τ(123)(h ⊗ g ⊗ f ) = g ⊗ f ⊗ h, τ(132)(h ⊗ g ⊗ f ) = f ⊗ h ⊗ g. Thus we get the required identity 4.3: {{c{{b, a}}}}L + τ(123){{b{{a, c}}}}L + τ(132){{a{{c, b}}}}L = 0. By this the proof of the first part of the theorem is completed. Now we prove the second part of the theorem, i.e. that the double Poisson algebras are in bijection with the pre-Calabi-Yau algebras of type B, given by A∞-structures on A ⊕ A∗ = (A ⊕ A∗, m = m(1) 3 ). We need to check that all remaining components of the Maurer-Cartan equations does not give any other identities, but the double bracket axioms. 2 + m(1) Thus we consider identities which appears from M C4 and M C5 for all possible types of input. We will show that from M C4 on type B operations we get only Leibniz identity, written in one of two forms: {{a, bc}} = b · {{a, c}} + {{a, b}} · c {{bc, a}} = b ⋆ {{c, a}} + {{b, a}} ⋆ c where · denotes the outer multiplication on A−A bimodule A⊗A: c·(a⊗b) = ca⊗b, (a⊗b)·c = a⊗bc, and ⋆ denotes the inner multiplication on A−A bimodule A⊗A: c⋆(a⊗b) = a⊗cb, (a⊗b)⋆c = ac⊗b. Whenever we have anti-symmetry identity: {{a, b}} = −{{b, a}}op, these two forms of Leibniz identity are equivalent. Then we show that from all M C5 equations on type B variables, the only nontrivial identities we get are two copies of the double Jacobi identity. Note, that if m4 = 0, but all higher operations mk, k > 5 are arbitrary, we get Leibnitz from M C4 and Jacobi identity from M C5, however we can get extra identities from M C6, those which connect m2 and m5, from M C7, those which connect m2 and m6, m3 and m5, etc. Thus we emphasize that the second part of the theorem holds only for the pre-Calabi-Yau-structures of type B, given by the A∞-structures (A ⊕ A∗, m = m(1) 2 + m(1) 3 . Let us start with M C4. We need to consider all possible inputs for the arity 4 Maurer-Cartan, each will give a separate equation. A priori there are 24 of such inputs, however, we will see, that when we consider ternary operations m3 on A ⊗ A∗ with only nonzero components corresponding to tensors of type B: A∗ ⊗ A ⊗ A∗ ⊗ A or A ⊗ A∗ ⊗ A ⊗ A∗, some terms of these equations vanish. We also take in account grading, chosen on R = A ⊕ A∗. For example, having this grading means that for any element f, g ∈ A∗, f g = 0, since f g′ = f ′ + g′ + 1 because deg m2 = 1 in our conv.1, but f ′ + g′ + 1 = 1, hence f g′ = 1, which for element f g ∈ A∗ is possible only if f g = 0. For inputs from A × A × A × A M C4 is trivial. Let a, b, c, d ∈ A, then from M C4 we have an equation: 16 m3(ab, c, d) ± m3(a, bc, d) ± m3(a, b, cd) ± am3(b, c, d) + m3(a, b, c)d = 0. All terms in this equation vanish since neither of them correspond to the B-component of the structure, that is one of the operations A∗ × A × A∗ → A∗ or A × A∗ × A → A, so they are equal to zero for all the type B variables. The vanishing of the equation will happen for the inputs: A∗ ×A×A×A, A×A×A×A∗; A∗×A∗×A∗×A, A×A∗×A×A∗, A∗ ×A×A∗×A∗, A×A∗×A∗×A∗; A × A∗ × A∗ × A, A × A × A∗ × A∗, A∗ × A∗ × A × A; A∗ × A∗ × A∗ × A∗, A × A × A × A. Let us give an argument just for one of those as an example, less trivial than the one above, where all inputs were from A. Let us consider the input f, a, g, h from A∗ × A × A∗ × A∗. The M C4 will look like m3(f a, g, h) ± m3(f, ag, h) ± m3(f, a, gh) ± f m3(a, g, h) ± m3(f, a, g)h = 0 The terms m3(f a, g, h), m3(f, ag, h) and f m3(a, g, h) are vanishing, since they correspond to oper- ations of secondary type, which are absent from the B-type solution we are looking for. The term m3(f, a, gh) contains among the arguments element gh ∈ A∗ · A∗, which is according to our grading and conv.1 should have degree 1, and since the only nonzero graded components are of degree −1 and 0, elements from A∗ · A∗ are equal to zero. The last term m3(f, a, g)h itself is an element from A∗ · A∗, because whenever m3(f, a, g) is an operation of B-type, it takes value in A∗. Hence this last term is vanishing as well. Vanishing of the equations for the other inputs from the above list can be shown analogously. The following inputs will give nontrivial equations: A × A × A∗ × A, A × A∗ × A × A; A∗ × A × A × A∗, A × A∗ × A × A∗, A∗ × A × A∗ × A. To write down the corresponding equations in terms of the double bracket defined as above, we need in addition to the lemma 4.3 on the outer multiplication of the A − A bimodule A ⊗ A (and respectively the inner multiplication on the dual bimodule (A ⊗ A)∗)), the following lemma on the inner multiplication ⋆ on the A − A bimodule A ⊗ A (and respectively the inner multiplication on the dual bimodule (A ⊗ A)∗)). Lemma 4.5. The following equalities hold: (R∗) (L∗) hg ⊗ f a, {{b, c}}i = −hg ⊗ f, a ⋆ {{b, c}}i hag ⊗ f, {{b, c}}i = hg ⊗ f, {{b, c}} ⋆ ai Proof. We use here Sweedler notations: {{b, c}} = P bi ⊗ ci = P b′ ⊗ c′′. (R) We used here: hg ⊗ f a, {{b, c}}i = Xhg ⊗ f a, b′ ⊗ c′′i = Xhg, b′ihf a, c′′i = −Xhg, b′ihf, ac′′i = −hg ⊗ f,X b′ ⊗ ac′′i = −hg ⊗ f, a ⋆ {{b, c}}i hf a, c′′i (4.6) = hac′′, f i (4.7) = −hf, ac′′i 17 (L) hag ⊗ f, {{b, c}}i = Xhag ⊗ f ihb′ ⊗ c′′i = Xhag ⊗ b′ihf ⊗ c′′i = Xhg, b′aihf, c′′i = Xhg ⊗ f, b′a ⊗ c′′i = hg ⊗ f, {{b, c}} ⋆ ai (4.5) Now we need to check that indeed from the remaining four inputs we get exactly one of the two mentioned above forms of the Leibniz identity. We start with the input a, f, b, c from A × A∗ × A × A. m3(af, b, c) + (−1)a′ m3(a, f b, c) + (−1)a′+f ′ m3(a, f, bc) + am3(f, b, c) + m3(a, f, b)c = 0 Taking into account vanishing terms, and our grading, it reads, after pairing with arbitrary element g ∈ A∗. −hm3(a, f b, c), gi − hm3(a, f, bc), gi + hm3(a, f, b)c, gi = 0 Using the definition of the bracket, this means hg ⊗ f b, {{c, a}}i − hg ⊗ f, {{bc, a}}i + hcg ⊗ f, {{b, a}}i = 0 (4.13) For the third term we also used cyclic invariance with respect to m2 and anti-symmetry: hm3(a, f, b)c, gi = (−1)m3(a,f,b)(c′+g′)hcg, m3(a, f, b)i = (−1)m3(a,f,b)′(c′+g′) · −(−1)m3(a,f,b)′cg′ hm3(a, f, b), cgi = hm3(a, f, b), cgi = hcg ⊗ f, {{b, a}}i. Now each of the first and the last terms can be rewritten using lemma 4.5 as follows, to match the Leibniz rule (in the second form, written via inner multiplication). hg ⊗ f b, {{c, a}}i R∗ = −hg ⊗ f, b ⋆ {{c, a}}i hcg ⊗ f, {{b, a}}i L∗ = −hg ⊗ f, {{b, a}} ⋆ ci. and Thus 4.13 reads hg ⊗ f, b ⋆ {{c, a}}i − hg ⊗ f, {{bc, a}}i + hg ⊗ f, {{b, a}} ⋆ ci = 0, which means Thus we got Leibniz identity, written via the inner product. {{bc, a}} = b ⋆ {{c, a}} + {{b, a}} ⋆ c. Consider now input f, a, b, g from A∗ × A × A × A∗. m3(f a, b, g) + (−1)f ′ m3(f, ab, g) + (−1)f ′+a′ m3(f, a, bg) + (−1)f ′ f m3(a, b, g) + m3(f, a, b)g = 0 18 Which in type B and for our grading becomes: m3(f a, b, g) + m3(f, ab, g) − m3(f, a, bg) = 0 These three terms are elements from A∗, so after pairing with an arbitrary element c ∈ A we get an equivalent equality: hm3(f a, b, g), ci + hm3(f, ab, g), ci − hm3(f, a, bg), ci = 0 The first term can be rewritten as hm3(f a, b, g), ci = (−1)f a′(b′+g′+c′)hm3(b, g, c), f ai (4.1) = hf a ⊗ g, {{c, b}}i. The second: The third: Thus we got hm3(f, ab, g), ci = (−1)f ′(ab′+g′+c′)hm3(b, g, c), f ai (4.1) = hf ⊗ g, {{c, ab}}i. −hm3(f, a, bg), ci = −(−1)f ′ hm3(a, bg, c), f i (4.1) = −hf ⊗ bg, {{c, a}}i. hf a ⊗ g, {{c, b}}i + hf ⊗ g, {{c, ab}}i − hf ⊗ bg, {{c, a}}i = 0. Applying Lemma 4.3 to the first and last term we get −hf ⊗ g, a{{c, b}}i + hf ⊗ g, {{c, ab}}i − hf ⊗ g, {{c, a}}bi = 0. This ensures the Leibniz identity (written via outer product). Consider now input a, f, b, g from A × A∗ × A × A∗. m3(af, b, g) + (−1)a′ m3(a, f b, g) + (−1)a′+f ′ m3(a, f, bg) + (−1)a′ am3(f, b, g) + m3(a, f, b)g = 0 Two terms m3(a, f b, g) and m3(a, f, bg) which are not of type B vanishes and after pairing with c ∈ A we have: hm3(af, b, g), ci − ham3(f, b, g), ci + hm3(a, f, b)g, ci = 0 Using cyclic invariance: hm3(af, b, g), ci = (−1)af ′(b′+g′+c′)hm3(b, g, c), af i = (−1)af ′(b′+g′+c′)(−1)b′(g′+c′+af ′)hm3(g, c, af ), bi 19 (−1)af ′(b′+g′+c′)(−1)b′(g′+c′+af ′)(−1)g′(c′+af ′+b′)hm3(c, af, b), gi = −hm3(c, af, b), gi = −hg ⊗ af, {{b, c}}i R=hg ⊗ f, {{b, c}}ai The second term: −ham3(f, b, g), ci cycl.m2= −(−1)a′ cycl.m2= −(−1)a′hm3(f, b, g)c, ai (−1)m3(f,b,g)′ hca, m3(f, b, g)i = −(−1)a′ (−1)m3(f,b,g)′ · −(−1)ca′m3(f,b,g)′ hm3(f, b, g), cai = −hm3(f, b, g), cai = −(−1)f ′(b′+g′+ca′)hm3(b, g, ca), f i = −(−1)f ′(b′+g′+ca′)(−1)b′(g′+ca′+f ′)hm3(g, ca, f ), bi = −(−1)f ′(b′+g′+ca′)(−1)b′(g′+ca′+f ′)(−1)g′(ca′+f ′+b′)hm3(ca, f, b), gi = The last term hm3(ca, f, b), gi = hg ⊗ f, {{b, ca}}i. hm3(a, f, b)g, ci cycl.m2= (−1)m3(a,f,b)′(g′+c′)hgc, m3(a, f, b)i = −(−1)m3(a,f,b)′gc′ hm3(a, f, b), gci = hm3(a, f, b), gci, which by definition of the bracket is: hm3(a, f, b), gci = hgc ⊗ f, {{b, a}}i (L) = −hg ⊗ f, c{{b, a}}i Combining the three terms we get the Leibniz identity (for outer product). And finally, to ensure that from M C4 we got nothing but Leibnitz identity, we have to consider the last non-trivial input: f, a, g, b from A∗ × A × A∗ × A. m3(f a, g, b) + (−1)f ′ m3(f, ag, b) + (−1)f ′+a′ m3(f, a, gb) + m3(f, a, g)b + (−1)f f m3(a, g, b) = 0 Which in type B and for our grading becomes: m3(f, a, g)b + f m3(a, g, b) − m3(f, a, gb) = 0 These three terms are elements from A∗, so after pairing with an arbitrary element c ∈ A we get an equivalent equality: hm3(f, a, g)b, ci + hf m3(a, g, b), ci − hm3(f, a, gb), ci = 0 The first term can be rewritten as 20 hm3(f, a, g)b, ci (cycl.m2) = (−1)m3(f,a,g)′(b′+c′)hbc, m3(f, a, g)i = hbc, m3(f, a, g)i = −(−1)m3(f,a,g)′bc′ hm3(f, a, g), bci = −(−1)m3(f,a,g)′bc′ (−1)f ′(a′+g′+bc′ hm3(a, g, bc), f i = −(−1)m3(f,a,g)bc′ (−1)f ′(a′+g′+bc′ (−1)a′(g′+bc′+f ′ hm3(g, bc, f ), ai = −(−1)m3(f,a,g)′bc′ (−1)f ′(a′+g′+bc′ (−1)a′(g′+bc′+f ′ (−1)g′(bc′+f ′+a′ hm3(bc, f, a), gi The second: = hm3(bc, f, a), gi = hg ⊗ f, {{a, bc}}i hf m3(a, g, b), ci cycl.m2= (−1)f ′(m(a,g,b)′+c′)hm3(a, g, b)c, f i = (−1)f ′(m3(a,g,b)′+c′)(−1)m3(a,g,b)′(c′+f ′)hcf, m3(a, g, b)i = (−1)f ′(m3(a,g,b)′+c′)(−1)m3(a,g,b)′(c′+f ′) · −(−1)cf ′m3(a,g,b)′ hm3(a, g, b), cf i = hm3(a, g, b), cf i = (−1)a′(g′+b′+cf ′ hm3(g, b, cf ), ai = (−1)a′(g′+b′+cf ′ (−1)g′(b′+cf ′+a′ hm3(b, cf, a), gi = −hm3(b, cf, a), gi (4.1) = −hg ⊗ cf, {{a, b}}i R= −hg ⊗ f, {{a, b}}ci The third: −hm3(f, a, gb), ci = −(−1)f ′(a′+gb′+c′)hm3(a, gb, c), f i = −(−1)f ′(a′+gb′+c′)(−1)a′(gb′+c′+f ′)hm3(gb, c, f ), ai = −(−1)f ′(a′+gb′+c′)(−1)a′(gb′+c′+f ′)(−1)gb′(c′+f ′+a′)hm3(c, f, a), gbi = hm3(c, f, a), gbi (4.1) = hgb ⊗ f, {{a, c}}i L= −hg ⊗ f, b{{a, c}}i Thus we got hg ⊗ f, {{a, bc}}i = hg ⊗ f, {{a, b}}ci + hg ⊗ f, b{{a, c}}i = 0. This ensures the Leibniz identity (written via outer product). By this we complete the proof of the second part of the Theorem 4.2 21 5 Polyderivations from pre-Calabi-Yau structures in- duce a Poisson bracket on representation spaces Consider polyderivations, that is maps A1 ⊗ · · · ⊗ Ar → A1 ⊗ · · · ⊗ Ak, which satisfy kind of Leibniz identities. Definition 5.1. Let P olyDer(A⊗n, M ), for any A⊗n-bimodule M , be the space of polyderivations, that is linear maps δ : A ⊗ ... ⊗ A → M satisfying the Leibniz identity: δ(a1⊗...⊗a′ ia′′ i ⊗...⊗an = (1⊗...⊗a′ i⊗...⊗1)δ(a1⊗...⊗a′′ i ⊗...⊗an)+δ(a1⊗...⊗a′ i⊗...⊗an)(1⊗...⊗a′′ i ⊗...⊗1) Definition 5.2. We call a solution of the Maurer-Cartan equation on A ⊕ A∗ (and a corresponding linear map δ : A⊗A → A⊗A) a restricted polyderivation, δ ∈ RP olyDer(A⊗2, A⊗2), if its projection to B-component is a polyderivation. Now we can see that pre-Calabi-Yau structure, that is a cyclicly invariant A∞-structure on i ), which is a restricted polyderivation gives rise to the Poisson bracket on (A ⊕ A∗, m = m(1) ∞P i=2,i6=4 the space Rep(A, m). Moreover the bracket is GLn invariant. This follows from Theorem 5.3. Pre-Calabi-Yau structure on A = (A ⊕ A∗, m = ∞P i=2,i6=4 m(1) i ), which is additionally a restricted polyderivation δ : A ⊗ A → A ⊗ A, δ ∈ RP olyDer(A⊗2, A⊗2) gives rise to the double Poisson bracket. Proof. As we have shown in Theorem 4.2, one can construct a double Poisson brackets from a pre- Calabi-Yau structures of type B We now want to show that for any solution of M C5, its projection to the space of solutions of type B is also a solution. Thus this projection will satisfy double Jacobi identity and thus create a double Poisson bracket from any pre-CY structure, which is a restricted polyderivation. This comes from the consideration of Section 2 on the structure of equations arising from M C5. We showed that all equations but those which are entirely on operations of type B contain in each term at least one operation of type A. Hence if we replace in a given solution of the Maurer-Cartan equation all Y s (corresponding to operations of type A) by zero, we get all equations with Y in them automatically satisfied. System of equations arising from M C5 turns into its restriction to those equations which are on operations of type B only. The latter, as we know (Theorem 4.2) under the assignment hg ⊗ f, {{a, b}}i := hg, m3(a, f, b)i (which automatically has antisymmetry) coincides with the double Jacobi identity. Unfortunately, the analogous procedure of projection of any solution onto the components of type B does not work in the same way for M C4. This is why we additionally asked for our arbitrary pre-Calabi-Yau structure to be a restricted polyderivation. Now the type B component of any Maurer-Cartan solution gives us a double Poisson bracket. Corollary 5.4. Any pre-Calabi-Yau structure of an arbitrary associative algebra A, given by an A∞-structure on (A ⊕ A∗, m = i ), which is a restricted polyderivation ρ : A ⊗ A → A ⊗ A m(1) ∞P i=2,i6=4 on ternary level, induces a Poisson structure on representation spaces Rep(A, n), which is Gln- invariant. 22 Proof. It comes as a direct consequence of Van den Bergh's construction for double Poisson bracket [16], after the application of Theorem 5.3. Remark. In spite for general pre-CY structure (m4 non necessary zero), we have from M C5 not precisely the Jacobi identity, but only up to certain terms involving cyclic tensor A ⊗ A∗ ⊗ A ⊗ A∗ ⊗A, the noncommutative Poisson structure still can be constructed and the Poisson structure on representation spaces induced (however it requiters some restrictions on A), as it was shown in [5]. In case m4 = 0 the situation is much more transparent, and it shows clearly how the correcting term for the Jacobi identity appear, so we consider it separately in this paper. Moreover, if m = m2 + m3 we have even one-to-one correspondence between pre-CY-structures and double Poisson brackets. Remark. Note that there was a considerable freedom in the choice of definition for the double bracket via the pre-Calabi-Yau structure (in spite it is in many ways defined by various features of the double bracket), but only the one presented in definition 4.1, together with appropriate choice of the grading on A ⊕ A∗ allowed to deduce the axioms of double bracket. By this choice we thus found an embedding of a double Poisson structures into pre-Calabi-Yau structures. 6 Acknowledgements We are grateful to IHES and MPIM where this work partially was done for hospitality and support. This work is funded by the ERC grant 320974, and partially supported by the project PUT9038 and EPSRC grant EP/M008460/1/. References [1] C. Brav, T. Dyckerhoff, Relative Calabi-Yau structures, arXiv:1606.00619 [2] M.Gerstenhaber, On the deformation of rings and algebras, Ann. of Math. 78 (1964), 59-103. [3] N.Iyudu, Examples of Pre-Calabi-Yau algebras, associated operads, and homology, preprint IHES M/17/02 (2017). [4] N.Iyudu, M.Kontsevich, Pre-Calabi-Yau algebras as noncommutative Poisson structures, preprint IHES M/18/04 (2018). [5] N.Iyudu, M.Kontsevich, Y.Vlassopoulos, Higher Hochshild cohomology, preprint. [6] M.Kontsevich, Y.Soibelman, Notes on A∞-algebras, A∞-categories and non-commutative ge- ometry, ArXiv MathRA/0606241 (2006) [7] M.Kontsevich, Deformation quantization of Poisson manifolds, I, Letters Math.Physics 66(2003), no 3, 157-216. [8] M.Kontsevich, Formal non-commutative symplectic geometry, in 'The Gelfand mathematical Seminars 1990 -- 1992', Birkhauser, (1993), 173-187. [9] M.Kontsevich, Weak Calabi-Yau algebras, Conference on Homological Mirror Symmetry, Mi- ami, 2013 [10] M.Kontsevich, Y.Vlassopoulos, Pre-Calabi-Yau algebras and topological quantum field theories, preprint. [11] M.Kontsevich, A.Rosenberg, Noncommutative smooth spaces, in 'The Gelfand mathematical Seminars 1996 -- 1999', Birkhauser, (1993), 85-108. [12] J-L.Loday, B.Valette, Algebraic operads, Springer 2013 23 [13] A.Odesskii, V.Rubtsov, V.Sokolov, Double Poisson brackets on free associative algebras, arXiv:1208.2935 [14] P.Seidel, Fukaya A∞ structures associated to Lefschetz fibrations I, ArXiv MathSG/0912.3932v2 [15] J.Stasheff, J. Homotopy associativity of H-spaces I, II, Trans. Amer. Math. Soc. 108(1963), 275-312. [16] M.Van den Bergh, Double Poisson algebras, Trans. Amer. Math. Soc. 360 (2008), 5711-5769 Natalia Iyudu School of Mathematics The University of Edinburgh James Clerk Maxwell Building The King's Buildings Peter Guthrie Tait Road Edinburgh Scotland EH9 3FD E-mail address: [email protected] Maxim Kontsevich Institut des Hautes ´Etudes Scientifiques 35 route de Chartres F - 91440 Bures-sur-Yvette E-mail address: [email protected] Yannis Vlassopoulos Institut des Hautes ´Etudes Scientifiques 35 route de Chartres F - 91440 Bures-sur-Yvette E-mail address: [email protected] 24
1112.2065	4	1112	2013-07-23T14:59:38	Groebner bases and gradings for partial difference ideals	[ "math.RA", "math.AP" ]	In this paper we introduce a working generalization of the theory of Gr\"obner bases for algebras of partial difference polynomials with constant coefficients. One obtains symbolic (formal) computation for systems of linear or non-linear partial difference equations arising, for instance, as discrete models or by the discretization of systems of differential equations. From an algebraic viewpoint, the algebras of partial difference polynomials are free objects in the category of commutative algebras endowed with the action by endomorphisms of a monoid isomorphic to $\N^r$. Then, the investigation of Gr\"obner bases in this context contributes also to the current research trend consisting in studying polynomial rings under the action of suitable symmetries that are compatible with effective methods. Since the algebras of difference polynomials are not Noetherian ones, we propose in this paper a theory for grading them that provides a Noetherian subalgebras filtration. This implies that the variants of the Buchberger's algorithm we developed for difference ideals terminate in the finitely generated graded case when truncated up to some degree. Moreover, even in the non-graded case, we provide criterions for certifying completeness of eventually finite Gr\"obner bases when they are computed within sufficiently large bounded degrees. We generalize also the concepts of homogenization and saturation, and related algorithms, to the context of difference ideals. The feasibily of the proposed methods is shown by an implementation in Maple that is the first to provide computations for systems of non-linear partial difference equations. We make use of a test set based on the discretization of concrete systems of non-linear partial differential equations.	math.RA	math	GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS ROBERTO LA SCALA∗ Abstract. In this paper we introduce a working generalization of the theory of Grobner bases for algebras of partial difference polynomials with constant coefficients. One obtains symbolic (formal) computation for systems of lin- ear or non-linear partial difference equations arising, for instance, as discrete models or by the discretization of systems of differential equations. From an algebraic viewpoint, the algebras of partial difference polynomials are free ob- jects in the category of commutative algebras endowed with the action by endomorphisms of a monoid isomorphic to Nr. Then, the investigation of Grobner bases in this context contributes also to the current research trend consisting in studying polynomial rings under the action of suitable symmetries that are compatible with effective methods. Since the algebras of difference polynomials are not Noetherian ones, we propose in this paper a theory for grading them that provides a Noetherian subalgebras filtration. This implies that the variants of the Buchberger's algorithm we developed for difference ideals terminate in the finitely generated graded case when truncated up to some degree. Moreover, even in the non-graded case, we provide criterions for certifying completeness of eventually finite Grobner bases when they are computed within sufficiently large bounded degrees. We generalize also the concepts of homogenization and saturation, and related algorithms, to the context of difference ideals. The feasibily of the proposed methods is shown by an implementation in Maple that is the first to provide computations for systems of non-linear partial difference equations. We make use of a test set based on the discretization of concrete systems of non-linear partial differential equations. 1. Introduction An important idea at the intersection of many algebraic theories consists in studying algebraic structures under the action of operators of different nature, typ- ically automorphisms and derivations. Classical roots of this idea can be found clearly in invariant and representation theory, as well as in the study of polynomial identities satisfied by associative algebras. Recently, topics like algebraic statistic [4, 16] or entanglement theory [24] have given new impulse and applications to the research on such themes. Another fundamental source of inspiration is the theory of differential and difference algebras introduced in the pioneeristic work of Ritt [25, 26] and afterwards developed by Kolchin [17], Cohn [6], Levin [23] and many others. From the point of view of computational methods, starting from the algo- rithms proposed by Ritt himself, a considerable advancement can be recorded in 2000 Mathematics Subject Classification. Primary 12H10. Secondary 13P10, 16W22, 16W50. Key words and phrases. Partial difference equations; Grobner bases; Actions on algebras; Gradings on algebras. Partially supported by Universit`a di Bari. 1 2 R. LA SCALA the differential case (see for instance [27]). Much less has been achieved for the algebras of difference polynomials where working algorithms can be found mainly in the linear case [12]. Nevertheless, the interest for such computations is relevant because of applications in the discretization of systems of differential equations like the automatic generation of finite difference schemes or the consistency analysis of finite difference approximations [9, 11, 21]. The present paper contributes to this research trend by concerning the development of effective methods for systems of linear or non-linear partial difference equations with constant coefficients. We provide also an implementation of such methods which is the first to allow com- putations in the non-linear case. Specifically, we generalize the theory of Grobner bases and related algorithms for ideals of the algebra of partial difference polyno- mials. We are able to do this in a general and systematic way, by defining classes of suitable monomial orderings, by extending the Buchberger's algorithm and the concept of grading to difference ideals, by defining truncated homogeneous compu- tations and even by introducing a suitable notion of homogenization for such ideals. First contributions to such theory can be found in [9, 19, 20]. In particular, owing to the notion of "letterplace correspondence/embedding" introduced in [18, 19, 20], note that Grobner bases computations for ideals of the free associative algebra are a subclass of the same computations for ideals of the algebra of ordinary difference polynomials. The algebras of partial difference polynomials are free algebras in the class of commutative algebras that are invariant under the action by endomorphisms of a monoid isomorphic to Nr. Then, the study of Grobner bases for such algebras be- longs to the general investigation of computational methods for commutative rings or modules that have suitable symmetries. Moreover, from the viewpoint of appli- cations, the algebras of partial difference polynomials are fundamental structures in the formal theory of partial difference equations where a set of unknown mul- tivariate functions is assumed algebraically independent together with all partial shifts of them. To provide symbolic computation for systems of such equations is hence essential to introduce Grobner bases methods. Based on a suitable definition of monomial orderings that are compatible with shifts action and the description of large classes of them, the present paper introduces variants of the Buchberger's algo- rithm for partial difference ideals. These procedures take advantage of the monoid symmetry essentially by killing all S-polynomials in a orbit except for a minimal one. Note that the algebras of difference polynomials are not Noetherian since they are polynomial rings in an infinite number of variables and hence termination is not generally guaranteed for the proposed algorithms. With the aim of improving this situation, we define suitable gradings that are compatible with the monoid action and provide filtrations of the algebra of partial difference polynomials with finitely generated subalgebras. We obtain therefore the termination for finitely generated graded difference ideals when computations are performed within some bounded degree. For non-graded ideals but for monomial orderings compatible with such gradings, we prove also criterions able to certify that a Grobner basis computation performed over a suitable finite set of variables that is within a sufficiently large de- gree, is a complete one. Finally, the paper generalizes the notion of homogenization and saturation to difference ideals with respect to the given gradings and provides the algorithms to perfom this ideal operations. As a byproduct, one obtains an alternative algorithm to compute Grobner bases of non-graded difference ideals via GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 3 homogeneous computations. By means of an implementation in Maple, all these methods are finally experimented on difference ideals obtained by the discretization of systems of non-linear differential equations. 2. Algebras of partial difference polynomials Fix K any field and let Σ be a monoid (semigroup with identity) that we denote multiplicatively. Let A be a commutative K-algebra and denote EndK(A) the monoid of K-algebra endomorphisms of A. We call A a Σ-invariant algebra or briefly a Σ-algebra if there is a monoid homomorphism ρ : Σ → EndK(A). In this case, we denote σ · x = ρ(σ)(x), for all σ ∈ Σ and x ∈ A. Let A, B be Σ-algebras and ϕ : A → B be a K-algebra homomorphism. We say that ϕ is a Σ-algebra homomorphism if ϕ(σ · x) = σ · ϕ(x), for all σ ∈ Σ and x ∈ A. Let A be a Σ-algebra and let I ⊂ A be an ideal. We call I a Σ-invariant ideal or simply a Σ-ideal if Σ · I ⊂ I. Clearly, all kernels of Σ-algebra homomorphisms are Σ-ideals. Definition 2.1. Let A be a Σ-algebra and let X ⊂ A be a subset. We say that A is Σ-generated by X if A is generated by Σ · X as K-algebra. In other words, A coincides with the smallest Σ-subalgebra of A containing X. In the same way, one defines Σ-generation for the Σ-ideals. In the category of Σ-invariant algebras one can define free objects. In fact, let X be a set and denote x(σ) each element (x, σ) of the product set X(Σ) = X × Σ. Define P = K[X(Σ)] the polynomial algebra in the commuting variables x(σ). For any element σ ∈ Σ consider the K-algebra endomorphism ¯σ : P → P such that x(τ ) 7→ x(στ ), for all x(τ ) ∈ X(Σ). Then, one has a faithful monoid representation ρ : Σ → EndK(P ) such that ρ(σ) = ¯σ ad hence P is a Σ-algebra. Note that if Σ is a left-cancellative monoid then all maps ρ(σ) are injective. Proposition 2.2. Let A be a Σ-algebra and let f : X → A be any map. Then, there is a unique Σ-algebra homomorphism ϕ : P → A such that ϕ(x(1)) = f (x), for all x ∈ X. Proof. It is sufficient to define ϕ(x(σ)) = σ · f (x), for all x ∈ X and σ ∈ Σ. In fact, one has ϕ(τ · x(σ)) = ϕ(x(τ σ)) = τ σ · f (x) = τ · (σ · f (x)) = τ · ϕ(x(σ)), for any τ ∈ Σ. (cid:3) Definition 2.3. We call P = K[X(Σ)] the free Σ-algebra generated by X. In fact, P is Σ-generated by the subset X(1) = {xi(1) xi ∈ X}. In other words, the algebra P is an essential tool in the theory of Σ-algebras because any such algebra A that is Σ-generated by a set X can be obtained as a quotient Σ-algebra P/I, where I is a Σ-ideal of P . For instance, from the viewpoint of computational methods, if one develops them for P then such methods can be extended to any quotient P/I as it is done in the classical theory of Grobner bases for affine algebras. Note also that if Σ is defined as the monoid Inc(N) = {f : N → N f strictly increasing}, or some power of this, one obtains an environment for computations in algebraic statistic [4, 16]. We want now to go in the direction of developing fundamental structures for sym- bolic (formal) computation on systems of partial difference equations with constant coefficients. From now on, we assume that X = {x0, x1, . . .} is a finite or countable set and Σ is a free commutative monoid generated by a finite set, say {σ1, . . . , σr}. 4 R. LA SCALA Then, we consider the free Σ-algebra P = K[X(Σ)]. Note that (Σ, ·) is a cancella- tive monoid isomorphic to (Nr, +) and the monomorphisms ρ(σ) : P → P have infinite order for all σ 6= 1. For any xi(σ) ∈ X(Σ), we call i and σ respectively the index and the weight of the variable xi(σ). If we put X(σ) = {xi(σ) xi ∈ X} and xi(Σ) = {xi(σ) σ ∈ Σ} one has clearly P = Nσ∈Σ K[X(σ)] = Nxi∈X K[xi(Σ)], where all subalgebras K[X(σ)] are isomorphic to K[X] and all subalgebras K[xi(Σ)] to K[Σ]. Definition 2.4. The free Σ-algebra P = K[X(Σ)] (Σ = hσ1, . . . , σri) is called the algebra of partial difference polynomials with constant coefficients. The motivation for such name is in the formal theory of partial difference equa- tions [6, 23]. In this theory, in fact, the indeterminates xi(1) are by definition algebraically independent unknown functions ui(t1, . . . , tr) in the variables tj and the maps ρ(σk) are the shift operators ui(t1, . . . , tr) 7→ ui(t1, . . . , tk + h, . . . , tr) where h is a parameter (mesh step). then the indeterminates xi(σ) = σ · xi(1) correspond to the (algebraically independent) shifted functions ui(t1 + α1h, . . . , tr + αrh) = σ · ui(t1, . . . , tr). Then, a Σ-ideal I ⊂ P is also called a partial difference ideal and a Σ-basis of I corresponds to a system of partial differ- ence equations in the unknown functions ui(t1, . . . , tr). One uses the term ordinary difference when r = 1. Note that the algebras of difference polynomials are not Noetherian rings since they are polynomial rings in an infinite number of variables. One has therefore that difference ideals have bases or Σ-bases which are generally infinite. If σ = Qi σαi i In the next sections we generalize the Grobner basis theory to the free Σ-algebra P = K[X(Σ)] of partial difference polynomials. Clearly, one reobtains the classical theory when Σ = {1} (r = 0) that is P = K[X]. The starting point is to define monomial orderings of P which are compatible with the action of the monoid Σ. 3. Monomial Σ-orderings Denote by M = Mon(P ) the set of all monomials of P . Note that even if the set X(Σ) is infinite (in fact countable), one can endow P by monomial orderings. This is an important consequence of the Higman's Lemma [15] which can be stated in the following way (see for instance [1], Corollary 2.3 and remarks at beginning of page 5175). Proposition 3.1. Let ≺ be a total ordering on M such that (i) 1 (cid:22) m for all m ∈ M ; (ii) ≺ is compatible with multiplication on M , that is if m ≺ n then tm ≺ tn, for any m, n, t ∈ M . Then ≺ is also a well-ordering of M that is a monomial ordering of P if and only if the restriction of ≺ to the variables set X(Σ) is a well-ordering. Clearly, it is easy to construct well-orderings for the set X(Σ) which is in bijective correspondence to Nr+1. Note that the monoid Σ stabilizes the monomials set M since it stabilizes X(Σ). We introduce then the following notion. Definition 3.2. Let ≺ be a monomial ordering of P . We call ≺ a (monomial) Σ-ordering of P if ≺ is compatible with the Σ-action on M , that is m ≺ n implies that σ · m ≺ σ · n for all m, n ∈ M and σ ∈ Σ. GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 5 A straightforward consequence of this definition is the following result. Proposition 3.3. Let ≺ be a monomial Σ-ordering of P . Then m (cid:22) σ · m for all m ∈ M and σ ∈ Σ. Proof. By contradiction, assume that there are m, σ such that m ≻ σ · m. Then, σ · m ≻ σ2 · m and by induction one obtains the infinite descending chain m ≻ σ · m ≻ σ2 · m ≻ . . . which contradicts that ≺ is a well-ordering. (cid:3) The orderings on the variables set X(Σ) that can be extended to monomial Σ-orderings are as follows. Definition 3.4. Let ≺ be a well-ordering of X(Σ). We call ≺ a (variable) Σ- ranking of P if ≺ is compatible with the Σ-action on X(Σ), that is u ≺ v implies that σ · u ≺ σ · v for all u, v ∈ X(Σ) and σ ∈ Σ. As for Proposition 3.3, we have that if ≺ is a Σ-ranking then u (cid:22) σ · u for all u ∈ X(Σ) and σ ∈ Σ. Moreover, if X is a finite set then condition u (cid:22) σ · u for all u, σ together with Σ-compatibility implies that ≺ is a well-ordering by applying Dickson's Lemma (or Higman's Lemma) to Σ which is isomorphic to Nr. However, note that in this paper the set X may be also countable. Owing to the decompositions X(Σ) = Sσ∈Σ X(σ) = Sxi∈X xi(Σ) of the vari- ables set of the ring P , we can define Σ-rankings of P in a natural way. Denote by Q the monoid K-algebra defined by the free commutative monoid Σ = hσ1, . . . , σri. In other words, Q = K[σ1, . . . , σr] is the polynomial algebra in the commutative variables σi. From now on, we assume that Σ is endowed with a monomial ordering < of Q. By abuse, we call < a monomial ordering of Σ. Definition 3.5. Fix < a monomial ordering of Σ. For all xi(σ), xj (τ ) ∈ X(Σ), we define: (i) xi(σ) ≺ xj (τ ) if and only if σ < τ or σ = τ and i < j. In other words, X(σ) ≺ X(τ ) when σ < τ . (ii) xi(σ) ≺′ xj (τ ) if and only if i < j or i = j and σ < τ . In other words, xi(Σ) ≺′ xj(Σ) when i < j. Clearly ≺ and ≺′ are both Σ-rankings of P that we call respectively weight and index Σ-ranking defined by a monomial ordering of Σ. For all xi ∈ X and σ ∈ Σ denote P (σ) = K[X(σ)], M (σ) = Mon(P (σ)) and P (xi) = K[xi(Σ)], M (xi) = Mon(P (xi)). Owing to the tensor decomposi- tions P = Nσ∈Σ P (σ) = Nxi∈X P (xi), one has that a monomial m ∈ M can be uniquely written as m = m(δ1) · · · m(δk) = m(xi1 ) · · · m(xil ), where m(δp) ∈ M (δp), m(xip ) ∈ M (xip ) and δ1 > . . . > δk, i1 > . . . > il. By means of such presentations we can define block monomial orderings of P extending weight and index ranking. Recall that ρ : Σ → EndK(P ) is the faithful monoid representation defined by the action of Σ over P . For any σ ∈ Σ one has that the map ρ(σ) defines an isomorphism between the monoids M (1), M (σ) and hence between the algebras P (1), P (σ). In other words, we have M (σ) = σ · M (1), P (σ) = σ · P (1). Definition 3.6. Fix ≺ a monomial ordering of the subalgebra P (1) ⊂ P and extend it to all subalgebras P (σ) (σ ∈ Σ) by the isomorphisms ρ(σ). In other words, we put σ · m ≺ σ · n if and only if m ≺ n, for any m, n ∈ M (1). Then, for all m, n ∈ M, m = m(δ1) · · · m(δk), n = n(δ1) · · · n(δk) with δ1 > . . . > δk we define 6 R. LA SCALA m ≺w n if and only if m(δj) = n(δj) if j < i and m(δi) ≺ n(δi) for some 1 ≤ i ≤ k. Clearly, the restriction of ≺w to the variables of P is just the weight Σ-ranking. Proposition 3.7. The ordering ≺w is a Σ-ordering of P . Proof. Note that if m = m(δ1) · · · m(δk) ∈ M with m(σi) ∈ M (σi) and δ1 > . . . > δk then σ · m = m(σδ1) · · · m(σδk), where m(σδi) = σ · m(δi) ∈ M (σδi) and σδ1 > . . . > σδk since < is a monomial ordering of Σ. Assume m ≺w n that is m(δj) = n(δj) for j < i and m(δi) ≺ n(δi). Clearly m(σδj ) = n(σδj) for j < i and one has m(δi) ≺ n(δi) if and only if m(1) ≺ n(1) if and only if m(σδi) ≺ n(σδi). Then, we conclude that σ · m ≺w σ · n. (cid:3) Note that we have also a monoid faithful representation φ : N → EndK(P ) such that the endomorphism φ(i) is defined as xj(σ) 7→ xi+j (σ) for any i, j ≥ 0 and σ ∈ Σ. Clearly φ(i) induces isomorphism between the monoids M (x0), M (xi) and the algebras P (x0), P (xi). The algebra P (x0) can be easily endowed with a Σ- ordering. For instance, since P (x0) = Nσ∈Σ K[x0(σ)] one can define a lexicographic ordering as in Definition 3.6. Definition 3.8. Fix ≺ a monomial Σ-ordering of the subalgebra P (x0) ⊂ P and extend it to all subalgebras P (xi) (xi ∈ X) by the isomorphisms φ(i). For any m, n ∈ M, m = m(xi1 ) · · · m(xik ), n = n(xi1 ) · · · n(xik ) with i1 > . . . > ik we put m ≺i n if and only if m(xiq ) = n(xiq ) if q < p and m(xip ) ≺ n(xip ) for some 1 ≤ p ≤ k. Note that the restriction of ≺i to the variables of P is the index Σ-ranking. Proposition 3.9. The ordering ≺i is a Σ-ordering of P . Proof. Note that if m = m(xi1 ) · · · m(xik ) ∈ M with m(xip ) ∈ M (xip ) and i1 > . . . > ik then σ · m = m′(xi1 ) · · · m′(xik ) where m′(xip ) = σ · m(xip ) ∈ M (xip ). Suppose m ≺i n that is m(xiq ) = n(xiq ) if q < p and m(xip ) ≺ n(xip ). We have clearly that m′(xiq ) = n′(xiq ). Moreover, since ≺ is a Σ-ordering of P (x0) and therefore of P (xip ), one has also m′(xip ) ≺ n′(xip ) that is σ · m ≺i σ · n. (cid:3) We call the above monomial Σ-orderings ≺w, ≺i of P respectively weight Σ- ordering defined by a monomial ordering of P (1) and index Σ-ordering of P defined by a monomial Σ-ordering of P (x0). Clearly, both these orderings depend also on a monomial ordering of Σ. Note that index Σ-orderings are suitable for generation of finite difference schemes for partial differential equations [10, 11]. The weight Σ- orderings are instead compatible with the gradings of the Σ-algebra P we introduce in Section 6. For this reason they are suitable for obtaining complete Grobner bases from truncated computations (see Proposition 6.14). To make things more explicit, we give now an example of a weight and an index Σ-ordering. Fix X = {x, y, z} and Σ = hσ1, σ2i. To simplify the notation, we 1σj identify the monoid (Σ, ·) with (N2, +) by means of the isomorphism σi 2 7→ (i, j). Then, we fix the degrevlex monomial ordering on Σ with σ1 > σ2 that is . . . > (2, 0) > (1, 1) > (0, 2) > (1, 0) > (0, 1) > (0, 0) and assume P (x) = K[x(i, j) i, j ≥ 0] be endowed with the lex monomial ordering such that . . . ≻ x(2, 0) ≻ x(1, 1) ≻ x(0, 2) ≻ x(1, 0) ≻ x(0, 1) ≻ x(0, 0). GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 7 Finally, we fix also the lex ordering on P (0, 0) = K[x(0, 0), y(0, 0), z(0, 0)] with x(0, 0) ≻ y(0, 0) ≻ z(0, 0). By isomorphisms, one has clearly the same ordering on P (y), P (z) and P (i, j) = K[x(i, j), y(i, j), z(i, j)], for all i, j ≥ 0, (i, j) 6= (0, 0). Then, a weight Σ-ordering is defined on P = K[x(i, j), y(i, j), z(i, j) i, j ≥ 0] as the block monomial ordering corresponding to the tensor decomposition P = . . . ⊗ P (2, 0) ⊗ P (1, 1) ⊗ P (0, 2) ⊗ P (1, 0) ⊗ P (0, 1) ⊗ P (0, 0). In a similar way, one defines an index Σ-ordering on P owing to the decomposition P = P (x) ⊗ P (y) ⊗ P (z). Similar Σ-orderings have been used for the examples contained in Section 5 and 8 and for the computational experiments presented in Section 9 (see also the Appen- dix). 4. Grobner Σ-bases From now on, we consider P endowed with a monomial Σ-ordering ≺. Let f = Pi cimi ∈ P with mi ∈ M, ci ∈ K, ci 6= 0. We denote as usual lm(f ) = mk = max≺{mi}, lc(f ) = ck and lt(f ) = lc(f )lm(f ). If G ⊂ P we put lm(G) = {lm(f ) f ∈ G, f 6= 0} and we denote as LM(G) the ideal of P generated by lm(G). Proposition 4.1. Let G ⊂ P . Then lm(Σ · G) = Σ · lm(G). In particular, if I is a Σ-ideal of P then LM(I) is also a Σ-ideal. Proof. Since P is endowed with a Σ-ordering, one has that lm(σ · f ) = σ · lm(f ) for any f ∈ P, f 6= 0 and σ ∈ Σ. Then, Σ · lm(I) = lm(Σ · I) ⊂ lm(I) and therefore LM(I) = hlm(I)i is a Σ-ideal. (cid:3) Definition 4.2. Let I ⊂ P be a Σ-ideal and G ⊂ I. We call G a Grobner Σ-basis of I if lm(G) is a Σ-basis of LM(I). In other words, Σ · G is a Grobner basis of I as P -ideal. Since the monoid Σ is assumed isomorphic to Nr that is Σ-ideals are partial difference ideals, we may say that Grobner Σ-bases are partial difference Grobner bases [9]. Another possible name is Σ-equivariant Grobner bases [4]. Simplicity and generality lead us to the previous definition that already appeared in [20]. Let f, g ∈ P, f, g 6= 0 and put lt(f ) = cm, lt(g) = dn with m, n ∈ M and c, d ∈ K. If l = lcm(m, n) we define as usual the S-polynomial spoly(f, g) = (l/cm)f −(l/dn)g. Clearly spoly(f, g) = −spoly(g, f ) and spoly(f, f ) = 0. Proposition 4.3. For all f, g ∈ P, f, g 6= 0 and for any σ ∈ Σ one has σ · spoly(f, g) = spoly(σ · f, σ · g). Proof. Since Σ acts on the variables set X(Σ) by injective maps, it is sufficient to note that σ · lcm(m, n) = lcm(σ · m, σ · n) for all m, n ∈ M and σ ∈ Σ. (cid:3) The following definition is a standard tool in Grobner bases theory. Definition 4.4. Let f ∈ P, f 6= 0 and G ⊂ P . If f = Pi figi with fi ∈ P, gi ∈ G and lm(f ) (cid:23) lm(fi)lm(gi) for all i, we say that f has a Grobner representation respect to G. 8 R. LA SCALA Note that if f = Pi figi is a Grobner representation then σ ·f = Pi(σ ·fi)(σ ·gi) is also a Grobner representation, for any σ ∈ Σ. In fact, since ≺ is a Σ-ordering of P one has that lm(f ) (cid:23) lm(fi)lm(gi) implies that lm(σ · f ) = σ · lm(f ) (cid:23) (σ · lm(fi))(σ · lm(gi)) = lm(σ · fi)lm(σ · gi) for all i. A celebrated result from Bruno Buchberger [5] is the following. Proposition 4.5 (Buchberger's criterion). Let G be a basis of the ideal I ⊂ P . Then, G is a Grobner basis of I if and only if for all f, g ∈ G, f, g 6= 0 the S- polynomial spoly(f, g) has a Grobner representation with respect to G. Usually the above result, see for instance [8], is stated when P is a polynomial algebra with a finite number of variables and G is a finite set. In fact, such as- sumptions are not needed since Noetherianity is not used in the proof, but only the existence of a monomial ordering for P . See also the comprehensive Bergman's paper [2] where the "Diamond Lemma" is proved without any restriction on the finiteness of the variables set. We want now to prove a generalization of the Buch- berger's criterion for Grobner Σ-bases of P . For this purpose it is useful to introduce the following notations. Definition 4.6. Let σ = Qi σαi where γi = min(αi, βi), for any i. i ∈ Σ. We denote gcd(σ, τ ) = Qi σγi i , τ = Qi σβi i Proposition 4.7 (Σ-criterion). Let G be a Σ-basis of a Σ-ideal I ⊂ P . Then, G is a Grobner Σ-basis of I if and only if for all f, g ∈ G, f, g 6= 0 and for any σ, τ ∈ Σ such that gcd(σ, τ ) = 1, the S-polynomial spoly(σ · f, τ · g) has a Grobner representation with respect to Σ · G. Proof. We prove that Σ·G is a Grobner basis of I and we make use of the Proposition 4.5. Then, consider any pair of elements σ·f, τ ·g ∈ Σ·G where f, g ∈ G, f, g 6= 0 and σ, τ ∈ Σ. Put δ = gcd(σ, τ ) and hence σ = δσ′, τ = δτ ′ with σ′, τ ′ ∈ Σ, gcd(σ′, τ ′) = 1. By Proposition 4.3 we have spoly(σ·f, τ ·g) = δ·spoly(σ′ ·f, τ ′ ·g). By hypothesis, assume that spoly(σ′ · f, τ ′ · g) = h = Pν fν(ν · gν), with ν ∈ Σ, fν ∈ P, gν ∈ G, is a Grobner representation with respect to Σ · G. Since ≺ is a Σ-ordering of P , we conclude that we have also the Grobner representation spoly(σ · f, τ · g) = δ · h = Pν (δ · fν)(δν · gν). (cid:3) For the purpose of obtaining an effective Buchberger's algorithm from the above criterion, note that all usual criteria (product criterion, chain criterion, etc) can be used also in such procedure. In particular, the arguments contained in the proof of Proposition 6.14 (see comments after this proof) imply that for any pair of elements f, g ∈ G and for all σ, τ ∈ Σ there are only a finite number of S-polynomials spoly(σ· f, τ · g) satisfying both the criteria gcd(σ, τ ) = 1 and gcd(σ · lm(f ), τ · lm(g)) 6= 1. A standard subroutine in the Buchberger's algorithm is the following. GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 9 Algorithm 4.1 Reduce Input: G ⊂ P and f ∈ P . Output: h ∈ P such that f − h ∈ hGi and h = 0 or lm(h) /∈ LM(G). h := f ; while h 6= 0 and lm(h) ∈ LM(G) do choose g ∈ G, g 6= 0 such that lm(g) divides lm(h); h := h − (lt(h)/lt(g))g; end while; return h. Note that the termination of Reduce is provided since ≺ is a monomial ordering of P . In particular, even if G is an infinite set, there are only a finite number of elements g ∈ G, g 6= 0 such that lm(g) divides lm(h) and hence lm(g) (cid:22) lm(h). It is well-known that if Reduce(f, G) = 0 then f has a Grobner representation with respect to G. Moreover, if Reduce(f, G) = h 6= 0 then clearly one has Reduce(f, G ∪ {h}) = 0. Therefore, from Proposition 4.7 and product criterion it follows immediately the correctness of the following algorithm. Algorithm 4.2 SigmaGBasis Input: H, a Σ-basis of a Σ-ideal I ⊂ P . Output: G, a Grobner Σ-basis of I. G := H; B := {(f, g) f, g ∈ G}; while B 6= ∅ do choose (f, g) ∈ B; B := B \ {(f, g)}; for all σ, τ ∈ Σ such that gcd(σ, τ ) = 1, gcd(σ · lm(f ), τ · lm(g)) 6= 1 do h := Reduce(spoly(σ · f, τ · g), Σ · G); if h 6= 0 then B := B ∪ {(g, h), (h, h) g ∈ G}; G := G ∪ {h}; end if ; end for; end while; return G. 10 R. LA SCALA Note that the above algorithm can be viewed as a variant of the usual Buch- berger's procedure applied for the basis Σ · H, where an additional criterion to avoid "useless pairs" is given by Proposition 4.7. Unfortunately, owing to Non- Noetherianity of the ring P , the termination of SigmaGBasis is not provided in general and this is, in fact, one of the main problems in differential/difference al- gebra. Precisely, even if a Σ-ideal I ⊂ P has a finite Σ-basis this may be not the case for the initial Σ-ideal LM(I) that is all Grobner Σ-bases of I are infinite sets. Despite this bad general case, in Section 6 we introduce suitable gradings for the algebra P which provides that truncated versions of the algorithm SigmaGBasis with homogeneous input stops in a finite number of steps. Note finally that some variant of SigmaGBasis appeared in [9] and before in [19, 20] for the ordinary difference case. In fact, the notion of "letterplace correspondence/embedding" in- troduced in these latter papers strictly relates non-commutative Grobner bases to Grobner Σ-bases of ordinary difference ideals (see also [18]). 5. An illustrative example In this section we apply the algorithm SigmaGBasis to a simple example in order to provide a concrete computation with it. Let X = {x, y}, Σ = hσ1, σ2i and consider the algebra of partial difference polynomials P = K[X(Σ)]. To simplify the notation, we identify the monoid (Σ, ·) with (N2, +) by means of the isomorphism 1σj σi 2 7→ (i, j). Then, we denote the variables of P as x(i, j), y(i, j), for all i, j ≥ 0. We consider now the Σ-ideal (difference ideal) I ⊂ P that is Σ-generated by the difference polynomials g1 = y(1, 1)y(1, 0) − 2x(0, 1)2, g2 = y(2, 0) + x(0, 0)x(1, 0). In other words, this Σ-basis (difference basis) encodes a system of non-linear differ- ence equations with constant coefficients in two unknown bivariate functions. By symbolic (formal) computations, we want to substitute this system with a comple- tion of it, namely a Grobner Σ-basis. We may want to do this for the purposes of checking membership of other equations to the Σ-ideal, elimination of unknowns, etc. The main problem is that such basis may be infinite, but it is not the case for this example. We fix then the degrevlex ordering on Σ with σ1 > σ2 that is on N2 where (1, 0) > (0, 1). Moreover, we consider the lex monomial ordering on K[x(0, 0), y(0, 0)] with x(0, 0) ≻ y(0, 0). A weight Σ-ordering (Definition 3.6) is hence defined on P = N(i,j)∈N2 K[x(i, j), y(i, j)] as a block monomial ordering. In practice, it is the lexicographic monomial ordering based on the following weight Σ-ranking . . . ≻ x(2, 0) ≻ y(2, 0) ≻ x(1, 1) ≻ y(1, 1) ≻ x(0, 2) ≻ y(0, 2) ≻ x(1, 0) ≻ y(1, 0) ≻ x(0, 1) ≻ y(0, 1) ≻ x(0, 0) ≻ y(0, 0). We use this monomial Σ-ordering of P for computing a Grobner Σ-basis of I. Such basis consists of the elements g1, g2 together with the difference polynomials g3 = y(1, 2)x(0, 1)2 − y(1, 0)x(0, 2)2, g4 = 2x(1, 1)2 − x(0, 0)x(1, 0)x(0, 1)x(1, 1). Note that in all these elements the first monomial is the leading one with respect to the given ordering of P . Let us see how the algorithm SigmaGBasis is able to GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 11 obtain such Grobner Σ-basis. Since the Σ-ideal I is Σ-generated by G = {g1, g2} then I is generated as an ideal of P by Σ · G that are the polynomials (i, j) · g1 = y(i + 1, j + 1)y(i + 1, j) − 2x(i, j + 1)2, (i, j) · g2 = y(i + 2, j) + x(i, j)x(i + 1, j), for all i, j ≥ 0. By applying the product criterion, we have to consider three kinds of S-polynomials spoly((i, j + 1) · g1, (i, j) · g1) = (i, j) · spoly((0, 1) · g1, g1), spoly((i + 1, j) · g1, (i, j) · g2) = (i, j) · spoly((1, 0) · g1, g2), spoly((i + 1, j + 1) · g1, (i, j + 2) · g2) = (i, j + 1) · spoly((1, 0) · g1, (0, 1) · g2). The Σ-criterion implies therefore that one has to reduce with respect to the basis Σ · G just the S-polynomials s1 = spoly((0, 1) · g1, g1), s2 = spoly((1, 0) · g1, g2), s3 = spoly((1, 0) · g1, (0, 1) · g2). The reduction of the S-polynomial s1 leads to the new element g3 and the current Σ-basis of I is now G = {g1, g2, g3}. The additional S-polynomials that survive to product and Σ-criterion are s4 = spoly((0, 1) · g1, g3), s5 = spoly((0, 2) · g1, g3), s6 = spoly((0, 2) · g2, (1, 0) · g3). We have that s4 → 0 and s2 → g4 with respect to Σ · G. The Σ-basis is then G = {g1, g2, g3, g4} and one has a new S-polynomial s7 = spoly((1, 0) · g3, g4). Finally, we have that all S-polynomials s3, s5, s6, s7 reduce to zero with respect to Σ · G and hence G is a Grobner Σ-basis of I. Note that G is in fact a minimal such basis and also that in this simple example there is no use of the chain criterion that can be always applied together with the other criteria. 6. Gradings of P compatible with Σ-action We want now to introduce some gradings of the algebra P = K[X(Σ)] which are compatible with Σ-action and formation of least common multiples in M = Mon(P ). As before, we fix a monomial order < of Σ. We start extending the structure (Σ, max, ·) in the following way. Definition 6.1. Let 0 be an element disjoint with Σ and put Σ = Σ ∪ {0}. Then, we define a commutative idempotent monoid ( Σ, +) with identity 0 that extends the monoid (Σ, max) (with identity 1) by imposing that 0+σ = σ, for any σ ∈ Σ. More- over, we define a commutative monoid ( Σ, ·) with identity 1 extending the monoid (Σ, ·) by putting 0 · σ = 0, for all σ ∈ Σ. Since multiplication clearly distributes over addition, one has that ( Σ, +, ·) is a commutative idempotent semiring, also known as commutative dioid [13]. Note that the faithful monoid representation ρ : Σ → EndK(P ) can be extended to Σ where ρ(0) : P → P is the algebra endomorphism such that xi(σ) 7→ 0, for all xi(σ) ∈ X(Σ). 12 R. LA SCALA Definition 6.2. Let w : M → Σ be the unique mapping such that (i) w(1) = 0; (ii) w(mn) = w(m) + w(n), for any m, n ∈ M ; (iii) w(xi(σ)) = σ, for all i ≥ 0 and σ ∈ Σ. Note that (i),(ii) state that w is a monoid homomorphism from the free commutative monoid (M, ·) to ( Σ, +). We call w the weight function of P . More explicitely, if m = xi1 (δ1)α1 · · · xik (δk)αk is any monomial of P different from 1 then w(m) = δ1 + · · · + δk = max<(δ1, . . . , δk). We denote Mσ = {m ∈ M w(m) = σ} and define Pσ ⊂ P the subspace spanned by Mσ, for any σ ∈ Σ. Because w : (M, ·) → ( Σ, +) is a monoid homomorphism one has that P = Lσ∈ Σ Pσ is a grading of the algebra P over the commutative monoid ( Σ, +). If f ∈ Pσ we say that f is a w-homogeneous element and we put w(f ) = σ. Recall that for any σ ∈ Σ we denoted P (σ) = K[X(σ)] which is a subalgebra of P = K[X(Σ)] isomorphic to K[X]. If we put P (0) = P0 = K then one has that P (σ) = Lτ ≤σ Pσ = Nτ ≤σ P (τ ) is a subalgebra of P . In particular, we have that P (1) = P0 ⊕ P1 = P (0) ⊗ P (1) = P (1) is isomorphic to the polynomial algebra K[X]. Definition 6.3. A monomial order < of Σ is said to be sequential if {τ ∈ Σ τ ≤ σ} is a finite set, for all σ ∈ Σ. It is important to note that if X is a finite set and < is a sequential ordering of Σ then the sequence {P (σ) σ ∈ Σ} is a filtration of P consisting of Noether- ian subalgebras. For such reason, from now on we assume Σ be endowed with a sequential monomial ordering. Proposition 6.4. The weight function satisfies the following properties: (i) w(σ · m) = σw(m), for any σ ∈ Σ and m ∈ M ; (ii) w(lcm(m, n)) = w(mn) = w(m) + w(n), for all m, n ∈ M . Then, m n implies that w(m) ≤ w(n). Proof. If m = 1 then w(σ · m) = w(m) = 0 = σw(m). If otherwise m = xi1 (δ1)α1 · · · xik (δk)αk with δ1 > . . . > δk then σ · m = xi1 (σδ1)α1 · · · xik (σδk)αk where σδ1 > . . . > σδk since < is a monomial ordering of Σ. We conclude that w(σ · m) = σδ1 = σw(m). To prove (ii) it is sufficient to note that the weight of a monomial does not depend on the exponents of the variables occuring in it. (cid:3) Note that the property (i) implies that the map w is a homomorphism with respect to the action of Σ on M and Σ. In other words, one has that σPτ ⊂ Pστ for any σ ∈ Σ, τ ∈ Σ. Moreover, the property (ii) means that w is also a monoid homomorphism from (M, lcm) to ( Σ, +). Definition 6.5. Let I be an ideal of P . We call I a w-graded ideal if I = Pσ Iσ with Iσ = I ∩ Pσ. In this case, if I is also a Σ-ideal then σ · Iτ ⊂ Iστ for all σ ∈ Σ, τ ∈ Σ. Owing to the w-grading of P , one can show that a truncated version of the algo- rithm SigmaGBasis admits termination. If f, g ∈ P, f 6= g are w-homogeneous el- ements then the S-polynomial h = spoly(f, g) is clearly w-homogeneous too. More- over, by property (ii) of Proposition 6.4, we have that w(h) = w(f ) + w(g) and hence if w(f ), w(g) ≤ δ then also w(h) ≤ δ, for some δ ∈ Σ. By means of this remark, one obtains immediately the following result. GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 13 Proposition 6.6 (Truncated termination over the weight). Let I ⊂ P be a w- graded Σ-ideal and fix δ ∈ Σ. Assume I has a w-homogeneous basis H such that Hδ = {f ∈ H w(f ) ≤ δ} is a finite set. Then, there is a w-homogeneous Grobner Σ-basis G of I such that Gδ is also a finite set. In other words, if we consider for the algorithm SigmaGBasis a selection strategy of the S-polynomials based on their weights ordered by <, we obtain that the δ-truncated version of SigmaGBasis stops in a finite number of steps. Proof. First of all, note that the algorithm SigmaGBasis computes essentially a subset G of a Grobner basis Σ · G obtained by applying the Buchberger's algorithm to the basis Σ·H of I. Moreover, by Proposition 6.4 the elements of Σ·H and hence of Σ · G are all w-homogeneous. Denote H ′ δ = {σ · f σ ∈ Σ, f ∈ H, σw(f ) ≤ δ}. Since < is a sequential monomial order of Σ and Hδ is a finite set one has that H ′ δ is also a finite set. We consider therefore Xδ the finite set of variables of P occurring in the elements of H ′ In fact, the δ- truncated algorithm SigmaGBasis computes a subset of a Grobner basis of the ideal I(δ) ⊂ P(δ) generated by H ′ δ. By Noetherianity of the finitely generated polynomial ring P(δ) we clearly obtain termination. (cid:3) δ and define P(δ) = K[Xδ] ⊂ P . Clearly the above result provides algorithmic solution to the ideal membership problem for finitely generated w-graded Σ-ideals. Note that if r = 0 that is Σ = {1} then the algorithm SigmaGBasis coincides with classical Buchberger's algorithm and Proposition 6.6 states that if I is a finitely generated ideal of P = P0 ⊕ P1 = K[x0, x1, . . .] then I has also a finite Grobner basis. According with the above proof, this is a consequence of the fact that the Buchberger's algorithm runs over the finite number of variables occuring in the generators of I. Another useful grading of P can be introduced in the following way. Consider the set N = N ∪ {−∞} endowed with the binary operations max and +. Then ( N, max, +) is clearly a commutative idempotent semiring (or commutative dioid or max-plus algebra). Define deg : Σ → N the mapping such that deg(0) = −∞ and deg(σ) = Pi αi, for any σ = Qi σαi . Clearly deg is a monoid homomorphism from ( Σ, ·) to ( N, +). i Definition 6.7. Let ord : M → N be the unique mapping such that (i) ord(1) = −∞; (ii) ord(mn) = max(ord(m), ord(n)), for any m, n ∈ M ; (iii) ord(xi(σ)) = deg(σ), for all i ≥ 0 and σ ∈ Σ. Clearly (i),(ii) state that ord is a monoid homomorphism from (M, ·) to ( N, max). We call ord the order function of P . For any monomial m = xi1 (δ1)α1 · · · xik (δk)αk different from 1 we have that ord(m) = max(deg(δ1), . . . , deg(δk)). Clearly, the order function defines a grading P = Ld∈N Pd of the algebra P over the commutative monoid ( N, max). Define P (d) = Li≤d Pi = Ndeg(σ)≤d P (σ) which is a subalgebra of P . Then, if X is a finite set we have that the sequence {P (d) d ∈ N} is a filtration of P with Noetherian subalgebras where P (0) = P−∞ ⊕ P0 is isomorphic to K[X]. Definition 6.8. A monomial order < of Σ is said to be compatible with deg when deg(σ) < deg(τ ) implies that σ < τ , for any σ, τ ∈ Σ. 14 R. LA SCALA If < is compatible with deg, note that < is a sequential ordering of Σ and ord(m) = deg(w(m)), for all m ∈ M . Finally, one has that the weight and order functions clearly coincide when r = 1. Proposition 6.9. The order function satisfies the following: (i) ord(σ · m) = deg(σ) + ord(m), for any σ ∈ Σ and m ∈ M ; (ii) ord(lcm(m, n)) = ord(mn) = max(ord(m), ord(n)), for all m, n ∈ M . There- fore, if m n then ord(m) ≤ ord(n). Proof. For m = 1 one has ord(σ · m) = ord(m) = −∞ = deg(σ) + ord(m). If oth- erwise m = xi1 (δ1)α1 · · · xik (δk)αk then σ · m = xi1 (σδ1)α1 · · · xik (σδk)αk and hence ord(σ · m) = max(deg(σδ1), . . . , deg(σδk)) = deg(σ) + max(deg(δ1), . . . , deg(δk)) = deg(σ) + ord(m). Property (ii) follows immediately as in Proposition 6.4. (cid:3) Definition 6.10. Let I be an ideal of P . We call I a ord-graded ideal if I = Pi Ii with Ii = I ∩ Pi. If I is also a Σ-ideal then σ · Ii ⊂ Ideg(σ)+i for any σ ∈ Σ and i ∈ N. Consider now f, g ∈ P, f 6= g two ord-homogeneous elements. The S-polynomial h = spoly(f, g) is clearly ord-homogeneous and ord(h) = max(ord(f ), ord(g)). Then ord(f ), ord(g) ≤ d implies that ord(h) ≤ d, for some d ∈ N and one proves the following result as for Proposition 6.6. Proposition 6.11 (Truncated termination over the order). Let I ⊂ P be a ord- graded Σ-ideal and fix d ∈ N. Assume I has a ord-homogeneous basis of H such that Hd = {f ∈ H ord(f ) ≤ d} is a finite set. Then, there is a ord-homogeneous Grobner Σ-basis G ⊂ I such that Gd is also a finite set. In other words, if we consider for SigmaGBasis a selection strategy of the S-polynomials based on their orders, we have that the d-truncated version of SigmaGBasis terminates in a finite number of steps. By means of weight and order functions one has criterions, also in the non-graded case, that provide that a Grobner Σ-basis is the eventually finite complete one even if it has been computed within some bounded weight or order for the algebra P that is over a finite number of variables. This is of course important because actual computations can be only performed in such a way. As before, we fix a sequential monomial ordering < on Σ. Definition 6.12. Let ≺ be a monomial Σ-ordering of P . We call ≺ compatible with the weight function if w(m) < w(n) implies that m ≺ n, for all m, n ∈ M . In a similar way, one defines when ≺ is compatible with the order function. Proposition 6.13. Let ≺w be a weight Σ-ordering as in Definition 3.6. Then ≺w is compatible with the weight function. In particular, if the monomial order < of Σ is compatible with deg then ≺w is also compatible with the order function. Proof. Let m = m(δ1) · · · m(δk), n = n(δ1) · · · n(δk) two monomials of P with m(δi), n(δi) ∈ M (δi) and δ1 > . . . > δk. Assume m ≺w n that is m(δj) = n(δj) if j < i and m(δi) ≺ n(δi) for some 1 ≤ i ≤ k. If i > 1 or m(δi) 6= 1 then clearly w(m) = w(n) = δ1. Otherwise, we conclude w(m) < δ1 = w(n). Moreover, if < is compatible with deg then ord(m) = deg(w(m)) < deg(w(n)) = ord(n) implies that w(m) < w(n) and hence m ≺w n. (cid:3) GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 15 For any δ ∈ Σ, d ∈ N define now Σδ = {σ ∈ Σ σ ≤ δ} and Σd = {σ ∈ Σ deg(σ) ≤ d}. Proposition 6.14 (Finite Σ-criterion). Assume the Σ-ordering of P is compatible with the weight function. Let G ⊂ P be a finite set and denote I the Σ-ideal generated by G. Moreover, define δ = max<{w(lm(g)) g ∈ G}. Then, G is a Grobner Σ-basis of I if and only if for all f, g ∈ G and for any σ, τ ∈ Σ such that gcd(σ, τ ) = 1 and gcd(σ·lm(f ), τ ·lm(g)) 6= 1, the S-polynomial spoly(σ·f, τ ·g) has a Grobner representation with respect to the finite set Σδ2 · G. In the same way, if the Σ-ordering of P is compatible with the order function and d = max{ord(lm(g)) g ∈ G}, then G is a Grobner Σ-basis of I when the above S-polynomials have a Grobner representation with respect to Σ2d · G. Proof. Let spoly(σ · f, τ · g) = h = Pν fν(ν · gν) be a Grobner representation with respect to Σ · G that is lm(h) (cid:23) lm(fν )(ν · lm(gν)) for all ν. We want to bound the elements ν ∈ Σ with respect to the ordering <. Put m = lm(f ), n = lm(g) and hence lm(σ · f ) = σ · m, lm(σ · g) = σ · n. By product criterion, we can assume that u = gcd(σ · m, τ · n) 6= 1. Then, there is a variable xi(σα) = xi(τ β) that divides u where xi(α) divides m and hence α ≤ w(m) ≤ δ and xi(β) divides n and therefore β ≤ w(n) ≤ δ. Then σα = τ β and one has that σ β, τ α because gcd(σ, τ ) = 1. We conclude that σ, τ ≤ δ and if v = lcm(σ · m, τ · m) then w(v) = max(σw(m), τ w(n)) ≤ δ2. Clearly v ≻ lm(h) (cid:23) ν · lm(gν) and hence δ2 ≥ w(v) ≥ νw(lm(gν )) ≥ ν. In a similar way, one argues for the order function. (cid:3) The above criterion implies that with respect to Σ-orderings compatible with weight or order functions one has an algorithm able to compute a finite Grobner Σ- basis, whenever this exists, in a finite number of steps. To fix ideas, let us consider If G is a finite Σ-basis of I and δ = max<{w(lm(g)) g ∈ G}, only weights. we may start considering all S-polynomials spoly(σ · f, τ · g) with f, g ∈ G and σ, τ ∈ Σ such that gcd(σ, τ ) = 1 and gcd(σ · lm(f ), τ · lm(g)) 6= 1. Note that such S-polynomials are in a finite number since in the above proof we observed that σ, τ ≤ δ and the monomial ordering of Σ is sequential. Moreover, one has also that these S-polynomials can be reduced only by elements of the finite set Σδ2 · G. If, as a result of some reduction, a new element f 6= 0 has to be added to the Σ-basis G and w(lm(f )) = δ′ > δ then it is sufficient to update the weight bound δ to δ′. 7. Homogenizing with respect to order function The purpose of this section is to analyze (de)homogenization processes in the context of Σ-ideals. Such methods are generally developed to have structures and computations that are homogeneous with respect to some grading, even if the input data are not such. Besides to the theoretical advantages as the concept of projective closure, these techniques usually imply computational benefits (see for instance [3]). Note that for Σ-ideals it is completely useless to consider classical gradings (total degree, multidegree, etc) since they not provide compatibility conditions with the Σ-action like the ones contained in Proposition 6.4 and Proposition 6.9. We decided then to present (de)homogenization methods only for the grading defined by the order function because univariate homogenizations are usually more efficient than multivariate ones since leading monomials are preserved by the homogenization process. Note finally that a major difference of the theory we present here with 16 R. LA SCALA the classical one is that the kernel of the dehomogenizing homomorphism contains a non-trivial graded ideal which implies that the homogenization process has to be considered modulo such ideal. Let t be a new variable disjoint with X. Define ¯X = X ∪ {t}, ¯X(Σ) = ¯X × Σ, ¯P = K[ ¯X(Σ)] and finally ¯M = Mon( ¯P ). Consider the algebra endomorphism ϕ : ¯P → ¯P such that xi(σ) 7→ xi(σ) and t(σ) 7→ 1, for all i, σ. Clearly ϕ2 = ϕ and P = ϕ( ¯P ). Moreover, one has that ϕ is a Σ-algebra endomorphism. Then ϕ defines a bijective correspondence between all Σ-ideals of P and Σ-ideals of ¯P containing ker ϕ = ht(1) − 1iΣ. Definition 7.1. Denote by N = Nord the largest ord-graded Σ-ideal contained in ker ϕ that is the ideal generated by all ord-homogeneous elements f ∈ ¯P such that ϕ(f ) = 0. Proposition 7.2. The ideal N ⊂ ¯P is generated by the elements (i) t(σ) − t(τ ) for all σ, τ ∈ Σ, σ 6= τ, deg(σ) = deg(τ ); (ii) t(σ)t(τ ) − t(σ), x(σ)t(τ ) − x(σ) for any σ, τ ∈ Σ, deg(σ) ≥ deg(τ ). Proof. Let f ∈ ¯P be a ord-homogeneous element such that ϕ(f ) = 0. Since the polynomials of type (i),(ii) clearly belongs to N , we have to prove that f is congruent to 0 modulo (i),(ii). Assume first that all variables of f belong to t(Σ) = {t(σ) σ ∈ Σ}. Recall that if m = t(δ1)α1 · · · t(δk)αk is any monomial of f then d = ord(f ) = max(deg(δ1), . . . , deg(δk)). Therefore, one has that f is congruent modulo (ii) to f ′ = Pi cit(τi) where τi ∈ Σ, deg(τi) = d and ci ∈ K, Pi ci = 0. By applying identity (i) it follows that f is congruent to (Pi ci)t(σ) = 0 for some fixed σ such that deg(σ) = d. Consider now the general case when the variables of f belong to ¯X(Σ). Fix σ ∈ Σ such that deg(σ) = d. Modulo the identities (i),(ii), one has that f is congruent to a polynomial f ′ whose monomials are either of type m ∈ M such that ord(m) = d or of type t(σ)n where n ∈ M, ord(n) < d. We show that in fact f ′ = 0. Denote f ′ = t(σ)g − h where g, h are polynomials in P , h is ord-homogeneous and ord(h) = d. Since 0 = ϕ(f ′) = g − h one has that f ′ = (t(σ) − 1)g. If we assume g 6= 0 then the monomials n of g are such that ord(n) = d which is a contradiction. (cid:3) We want now to define a bijective correspondence between all Σ-ideals of P and some class of ord-graded Σ-ideals of ¯P containing N . Definition 7.3. Let I be any Σ-ideal of P . We define I ∗ ⊂ ¯P the largest ord- graded Σ-ideal contained in the preimage ϕ−1(I) that is I ∗ is the ideal generated by all ord-homogeneous elements in ϕ−1(I). Clearly N = 0∗ ⊂ I ∗. We call I ∗ the ord-homogenization of the Σ-ideal I. Definition 7.4. Let f ∈ P, f 6= 0 and denote f = Pd fd the decomposition of f in its ord-homogeneous components. We define topord(f ) = d′ = max{d}. If f ∈ K that is d′ = −∞ we put f ∗ = f . Otherwise, we denote f ∗ = t(σ)f where σ ∈ Σ such that deg(σ) = d′. We call topord(f ) the top order of f and f ∗ its ord-homogenization. Note that ϕ(f ∗) = f and hence the element f ∗ is essentially defined modulo the ideal N (see also the next result). Owing to generators (i) of N in Proposition 7.2, all variables t(σ) such that deg(σ) = d′ are congruent modulo N . We don't need then to specify which of these variables we use for defining f ∗ = t(σ)f . GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 17 Proposition 7.5. Let I be a Σ-ideal of P . Then I ∗ = hf ∗ f ∈ I, f 6= 0i + N . Proof. Denote J = hf ∗ f ∈ Ii + N . Clearly J ⊂ I ∗. Let g ∈ I ∗ be a ord- homogeneous element and define f = ϕ(g) ∈ I. If f = 0 then g ∈ N ⊂ J. Otherwise, denote d = topord(f ) and d′ = ord(g). Since clearly d′ ≥ d one has that g is congruent modulo N to h = t(σ)f , where σ ∈ Σ such that deg(σ) = d′. Hence, if d′ = d then h is congruent exactly to f ∗. Otherwise, the polynomial h is congruent to t(σ)f ∗. In both cases, we conclude that g is congruent modulo N ⊂ J to an element of J and therefore g ∈ J. (cid:3) If I ⊂ P is a Σ-ideal one has clearly that ϕ(I ∗) = I. Moreover, if J ⊂ ¯P is a ord-graded Σ-ideal containing N then in general J ⊂ ϕ(J)∗. Definition 7.6. Let N ⊂ J ⊂ ¯P be a ord-graded Σ-ideal. Define J ′ = ϕ(J)∗ = hϕ(f )∗ f ∈ J, f /∈ N, f ord-homogeneousi + N . Then J ⊂ J ′ ⊂ ¯P is a ord-graded Σ-ideal that we call the saturation of J. Definition 7.7. Let J ⊂ ¯P be a ord-graded Σ-ideal containing N . We say that J is saturated if J coincides with its saturation ϕ(J)∗ that is for any ord-homogeneous element f ∈ J, f /∈ N one has that ϕ(f )∗ ∈ J. If I is a Σ-ideal of P then its ord-homogenization I ∗ is clearly a saturated ideal. Therefore, a bijective correspondence is given between all Σ-ideals of P and the saturated ord-graded Σ-ideals of ¯P containing N . We want now to analyze the behaviour of Grobner Σ-bases under homogenization and dehomogenization. Note that the arguments of Proposition 7.2 implies clearly that the polynomials (i),(ii) are in fact a Grobner basis of the ideal N with respect to any monomial ordering of ¯P . For this reason we introduce the following notion. Definition 7.8. A monomial m ∈ ¯M is said to be normal modulo N if m ∈ M or m = t(σ)n with n ∈ M, σ ∈ Σ such that d = deg(σ) > ord(n). Moreover, we require that t(σ) = min≺{t(τ ) deg(τ ) = d}. A polynomial f ∈ ¯P is in normal form modulo N if all its monomials are normal modulo N . Note that owing to generators (i) of the ideal N , we choose t(σ) = min≺{t(τ ) deg(τ ) = d} since in this case lm(t(σ) − t(τ )) = t(τ ). Definition 7.9. Let ≺ be a Σ-ordering of ¯P compatible with the order function. We call ≺ a ord-homogenization Σ-ordering if t(σ)m ≺ n for all m, n ∈ M, σ ∈ Σ such that deg(σ) = ord(n) > ord(m). It is easy to define one of the above orderings. Fix for instance the lex or degrevlex monomial order on the polynomial ring ¯P (1) = K[x0(1), x1(1), . . . , t(1)] where x0(1) ≻ x1(1) ≻ . . . ≻ t(1). Moreover, fix a monomial ordering on Σ which is compatible with deg and define the weight Σ-ordering ≺w of ¯P as in Definition 3.6. Clearly ≺w is a ord-homogenization Σ-ordering. From now on, we assume ¯P be endowed with a ord-homogenization Σ-ordering. Proposition 7.10. Let p, q ∈ ¯M be two normal monomials modulo N such that ord(p) = ord(q). Then p ≺ q implies that ϕ(p) ≺ ϕ(q). Proof. By definition, the monomials p, q are of type m ∈ M or t(σ)m with deg(σ) > ord(m). Since ≺ is a ord-homogenization order, when comparing two of such mono- mials of the same order one has only the following cases: m ≺ n, t(σ)m ≺ t(σ)n 18 R. LA SCALA or t(σ)m ≺ n. Then, we have to prove ϕ(p) = m ≺ n = ϕ(q) only when t(σ)m ≺ n. This follows immediately from ≺ is compatible with the order function and ord(m) < ord(n) = deg(σ). (cid:3) From now on, for any f ∈ P, f 6= 0 we denote by f ∗ the normal form of t(σ)f modulo N where σ ∈ Σ, deg(σ) = topord(f ). Proposition 7.11. Let f ∈ ¯P , f 6= 0 be a ord-homogeneous polynomial in normal form modulo N . Then lm(ϕ(f )) = ϕ(lm(f )). Moreover, we have that lm(f ∗) = lm(f ) for all f ∈ P, f 6= 0. Proof. The first part of the statement follows immediately from Proposition 7.10. Moreover, if σ ∈ Σ, deg(σ) = topord(f ) then t(σ) cannot appear in the leading monomial of f ∗ and hence lm(f ∗) = lm(f ). (cid:3) Definition 7.12. Let N ⊂ J ⊂ ¯P be a Σ-ideal. Moreover, let G ⊂ J be a subset of polynomials in normal form modulo N . We say that G is a Grobner Σ-basis of J modulo N if G ∪ N is a Grobner Σ-basis of J. Proposition 7.13. Let N ⊂ J ⊂ ¯P be a ord-graded Σ-ideal. If G is a ord- homogeneous Grobner Σ-basis of J modulo N then ϕ(G) is a Grobner Σ-basis of ϕ(J). Proof. Since G is a Grobner Σ-basis of J modulo N we have that for any ord- homogeneous polynomial f ∈ J, f 6= 0 in normal form modulo N there is an element g ∈ G and σ ∈ Σ such that σ · lm(g) lm(f ). Then, by applying the Σ-algebra endomorphism ϕ one obtains that σ · lm(ϕ(g)) lm(ϕ(f )) that is ϕ(G) is a Grobner Σ-basis of ϕ(J). (cid:3) Proposition 7.14. Let I ⊂ P be a Σ-ideal and let G be a Grobner Σ-basis of I. Then G∗ = {g∗ g ∈ G} is a ord-homogeneous Grobner basis of I ∗ modulo N . Moreover, one has that lm(G∗) = lm(G). Proof. Let f ′ ∈ I ∗ be a ord-homogeneous element in normal form modulo N and put f = ϕ(f ′) ∈ I. Then, either f ′ = f ∗ or f ′ = t(σ)f ∗ with ord(f ′) = deg(σ) > ord(f ∗) = topord(f ). Since G is a Grobner Σ-basis of I there is g ∈ G, τ ∈ Σ such that τ · lm(g) lm(f ). By Proposition 7.11 one has that lm(f ) = lm(f ∗) and lm(g) = lm(g∗). Therefore, τ · lm(g∗) divides lm(f ∗) and this monomial clearly divides lm(f ′). (cid:3) By the above propositions we obtain immediately what follows. Corollary 7.15. Let N ⊂ J ⊂ ¯P be a ord-graded Σ-ideal and denote J ′ = ϕ(J)∗ its saturation. Moreover, let G be a ord-homogeneous Grobner Σ-basis of J modulo N . Then G′ = ϕ(G)∗ = {ϕ(g)∗ g ∈ G} is a ord-homogeneous Grobner Σ-basis of J ′ modulo N . Moreover, we have lm(G′) = lm(ϕ(G)). Let I ⊂ P be any Σ-ideal. The previous results suggest an alternative method to calculate a Grobner Σ-basis of I which is based only on ord-homogeneous com- putations. Assume H is any Σ-basis of I and denote as before H ∗ = {f ∗ f ∈ H}. Clearly J = hH ∗iΣ + N is a ord-homogeneous Σ-ideal of ¯P containing N such that ϕ(J) = I. Assume now we compute G a ord-homogeneous Grobner Σ-basis of J modulo N . Then, ϕ(G) is a Grobner Σ-basis of I. Note that by using a ord-based selection strategy for the S-polynomials, the Grobner Σ-basis G can be obtained GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 19 order by order automatically minimal that is σ · lm(f ) not divides lm(g) for all f, g ∈ G, f 6= g and σ ∈ Σ. This is clearly a computational advantage, but since generally lm(G) 6= lm(ϕ(G)) one has that ϕ(G) may be not minimal. In the worst case, the ideal J may have an infinite and hence uncomputable minimal Grobner Σ-bases but I has a finite one. This is clearly not the case when one considers a saturated ideal J ′ = I ∗ since we have lm(G′) = lm(ϕ(G′)) when G′ is a minimal Grobner Σ-basis of J ′. Note that this nice property depends on the fact that we deal with a univariate homogenization. A drawback is that if one computes the saturation J ′ by means of the ideal J according to Corollary 7.15, one has again to compute a Grobner Σ-basis of J. Then, a better approach consists in computing "on the fly" the Grobner Σ-basis of J ′ starting from the generating set {f ∗ f ∈ H}. In other words, any time that a new generator g of the ord-homogeneous Grobner Σ-basis arises from the reduction of an S-polynomial, we saturate g that is we sub- stitute this polynomial with ϕ(g)∗. In formal terms, the algorithm one obtains is the following one. Algorithm 7.1 SigmaGBasis2 Input: H, a Σ-basis of a Σ-ideal I ⊂ P . Output: ϕ(G), a Grobner Σ-basis of I such that lm(G) = lm(ϕ(G)). G := H ∗; B := {(f, g) f, g ∈ G}; while B 6= ∅ do choose (f, g) ∈ B; B := B \ {(f, g)}; for all σ, τ ∈ Σ such that gcd(σ, τ ) = 1 do h := Reduce(spoly(σ · f, τ · g), Σ · G ∪ N ); if h 6= 0 then h = ϕ(h)∗ B := B ∪ {(g, h), (h, h) g ∈ G}; G := G ∪ {h}; end if ; end for; end while; return ϕ(G). 20 R. LA SCALA Proposition 7.16. The algorithm SigmaGBasis2 is correct. Proof. Note that at each step we are inside an ideal J such that ϕ(J) = I that is whose saturation is J ′ = I ∗. Moreover, for any ord-homogeneous element h ∈ ¯P which is in normal form modulo N one has that h′ = ϕ(h)∗ divides h. This implies that if an S-polynomial is reduced to zero by adding h to the basis G, the same holds if we substitute h with h′. In case of termination, owing to the set G is a ord- homogeneous Grobner Σ-basis of J modulo N whose elements are all saturated, by Corollary 7.15 we may conclude that J = J ′ and hence ϕ(G) is a Grobner Σ-basis of I such that lm(G) = lm(ϕ(G)). (cid:3) About termination or just termination up to some order d, this is not provided in general for the above algorithm. The reason is that even if all computations are ord-homogeneous, because of the saturation h = ϕ(h)∗ that may decrease the order we can't be sure at some suitable step that we will not get additional elements of order ≤ d in the steps that follow. 8. An illustrative example (continued) We apply now the algorithm SigmaGBasis2 to the same Σ-ideal that has been considered in Section 5 for illustrating SigmaGBasis. Recall that such ideal is I = hg1, g2i ⊂ P = K[X(Σ)] where X = {x, y}, Σ = N2 and g1 = y(1, 1)y(1, 0) − 2x(0, 1)2 g2 = y(2, 0) + x(0, 0)x(1, 0). Let now ¯X = {x, y, t} and define the polynomial algebra ¯P = K[ ¯X(Σ)] with vari- ables x(i, j), y(i, j), t(i, j), for all i, j ≥ 0. We consider the Σ-algebra endomorphism ϕ : ¯P → ¯P such that x(i, j) 7→ x(i, j), y(i, j) 7→ y(i, j) and t(i, j) 7→ 1. In Propo- sition 7.2 we proved that the largest ord-graded Σ-ideal contained in ker ϕ is the ideal N ⊂ ¯P generated by the polynomials t(i, j) − t(k, l) where (i, j) 6= (k, l), i + j = k + l; t(i, j)t(k, l) − t(i, j), x(i, j)t(k, l) − x(i, j) where i + j ≥ k + l. Moreover, we define a ord-homogenization Σ-ordering of ¯P (Definition 7.9) as the weight Σ-ordering given by the degrevlex ordering of Σ (σ1 > σ2) and the lex monomial ordering of K[x(0, 0), y(0, 0), t(0, 0)] (x(0, 0) ≻ y(0, 0) ≻ t(0, 0)). Note that such Σ-ordering extends the one defined for P ⊂ ¯P in Section 5. In practice, it is the lexicographic monomial ordering of ¯P such that . . . ≻ x(2, 0) ≻ y(2, 0) ≻ t(2, 0) ≻ x(1, 1) ≻ y(1, 1) ≻ t(1, 1) ≻ x(0, 2) ≻ y(0, 2) ≻ t(0, 2) ≻ x(1, 0) ≻ y(1, 0) ≻ t(1, 0) ≻ x(0, 1) ≻ y(0, 1) ≻ t(0, 1) ≻ x(0, 0) ≻ y(0, 0) ≻ t(0, 0). Recall that a Grobner Σ-basis of I with respect to such ordering is given by the elements g1, g2 together with g3 = y(1, 2)x(0, 1)2 − y(1, 0)x(0, 2)2, g4 = 2x(1, 1)2 − x(0, 0)x(1, 0)x(0, 1)x(1, 1). We introduce then ord-homogenizations of the polynomials g1, g2 that are 1 = y(1, 1)y(1, 0) − 2t(0, 2)x(0, 1)2, g∗ g∗ 2 = y(2, 0) + t(0, 2)x(0, 0)x(1, 0). GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 21 We start applying SigmaGBasis2 to G = {g∗ s1 = spoly((0, 1) · g∗ s2 = spoly((1, 0) · g∗ s3 = spoly((1, 0) · g∗ 2} by considering the S-polynomials 1), 2), 1, g∗ 1, g∗ 1, g∗ 1, (0, 1) · g∗ 2). By reducing s1 with respect to Σ · G ∪ N one obtains exactly the polynomial 3 = y(1, 2)x(0, 1)2 − t(0, 3)y(1, 0)x(0, 2)2. g∗ Clearly g∗ new S-polynomials 3 is already a saturated element. Then G = {g∗ 1, g∗ 2, g∗ 3} and we form the s4 = spoly((0, 1) · g∗ s5 = spoly((0, 2) · g∗ s6 = spoly((0, 2) · g∗ 3), 3), 1, g∗ 1, g∗ 2, (1, 0) · g∗ 3). With respect to Σ · G ∪ N , one has the reductions s4 → 0 and s2 → h where h = 2t(0, 3)x(1, 1)2 − t(0, 3)x(0, 0)x(1, 0)x(0, 1)x(1, 1) = t(0, 3)g∗ 4. 3, g∗ 2, g∗ 4, we update G = {g∗ 1, g∗ By saturating this element as ϕ(h)∗ = g∗ another S-polynomial is defined as 4} and s7 = spoly((1, 0) · g∗ 3, g∗ 4). All remaining S-polynomials s3, s5, s6, s7 reduce to zero with respect to Σ · G ∪ N and we conclude that ϕ(G) = {g1, g2, g3, g4} is a Grobner Σ-basis of I ⊂ P . 9. Testing and timings In this section we present a set of tests for the algorithms SigmaGBasis and SigmaGBasis2 which is based on an experimental implementation of them in the language of Maple. This is actually the first implementation of algorithms for com- puting Grobner bases of systems of linear or non-linear partial difference equations. Note that for the linear case one has the packages LDA (Linear Difference Algebra) [12] and Ore algebra[shift algebra] in the Maple distribution. The main idea that lead us when coding the proposed algorithms is that they can be considered variants of the classical Buchberger's algorithm where some amount of computations can be avoided by means of the symmetry defined by the monoid Σ. In fact, as explained in the previous sections, a "basic" approach to calculate a Grobner Σ-basis of a Σ-ideal I generated by a Σ-basis H consists in applying the Buchberger's algorithm to the basis Σ · H. One obtains therefore a Grobner basis G′ of I from which a Grobner Σ-basis G ⊂ G′ can be extracted such that Σ · lm(G) = lm(G′). Clearly, chain and coprime criterions can be used in the usual way in the procedure. Then, the algorithm SigmaGBasis can be understood as the variant that prescribes the appli- cation also of the Σ-criterion (Proposition 4.7) to the S-polynomials spoly(σ ·f, τ ·g) and to add the set of all shifts Σ · h to the current basis when a new element h arises from the reduction of an S-polynomial. Then, the Grobner Σ-basis of I is simply the union of the initial basis H with the new elements h. Recall that the procedure is correct only if one uses a monomial Σ-ordering. Clearly, from the set Σ is infinite it follows that actual computations can be only performed with a finite subset of Σ that is over a finite set of variables of P = K[X(Σ)]. Typically, one fixes a bound d for the degree of the elements of Σ that is for the order of the variables xi(σ). Owing to the finite Σ-criterion (Proposition 6.14), a basis obtained with a monomial ordering compatible with the order function is certified to be a 22 R. LA SCALA complete Grobner Σ-basis if the order bound is at least the double of the maximum top order of its elements. In addition to the basic procedure for the computation of Grobner Σ-bases and the algorithm SigmaGBasis, for the experiments we consider also a variant of the latter method where the Σ-criterion is suppressed but one continues to shift the reduced form of the S-polynomials. This procedure is tested to the aim of under- standing the contribution of any of the implemented strategies. Finally, we propose an implementation of the algorithm SigmaGBasis2 based on the saturation of a Σ-ideal with respect to the grading defined by the order function. In practice, once one has homogenized the initial generators, the saturation ϕ(h)∗ is performed be- fore the application of shifting, for each new element h obtained by the reduction of an S-polynomial. In output one returns the dehomogenization of the computed basis. Note that this procedure is correct only if one uses a Σ-ordering which is compatible with the order function and if the polynomials are kept in normal form modulo N during the computations. The monomial Σ-orderings of P that we consider for the tests are defined in the following way. One has initially to fix a monomial ordering for Σ and we choose degrevlex in order to provide compatibility with the degree. Then, one fixes a monomial ordering, for us lex, over the subring P (1) = K[X(1)] or P (x0) = K[x0(Σ)] that is extended as a block ordering to the polynomial ring P = K[X(Σ)] according to the choice of a variables ranking based on weight or index respectively. For a detailed description of these orderings in the considered examples see the Appendix. In the table of tests, we distinguish weight or index ranking by the letters "w" and "i". The integer that comes before these letters refers to the fixed order bound. Note that the algorithm SigmaGBasis2 is compatible only with rankings of type weight. For the basic variant of the Grobner Σ-bases algorithm, one can clearly use any implementation of the Buchberger's algorithm as, for instance, the one contained in the package Groebner of Maple. We have preferred instead to develop ourselves all different variants in order to have the same implementation and hence the same efficiency, for the fundamental subroutines of the algorithms. In this way, for the basic version we have been also able to access to important parameters of the com- putation as the total number of S-polynomial reductions. This number is for us the sum of the actual S-polynomials with the initial generators that are interreduced. Note that our implementation of the Buchberger's algorithm is in fact generally comparable with the built-in ones of Maple. For instance, the test falkow-6w-basic takes 9 hours, but using Groebner[Basis] it takes 11 hours with method=buchberger and 8.5 hours with method=maplef4. Other parameters that are considered for the experiments are the number of input and output generators. Note that for the basic algorithm we count generators and not Σ-generators. Finally, the parameter "mi- nout" refers to the number of elements of a minimal Grobner Σ-basis. All examples have been computed with Maple 12 running on a server with a four core Intel Xeon at 3.16GHz and 64 GB RAM. The timings are given in hour-minute-second format. GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 23 in 4 4 4 out minout pairs 5 5 5 Example falkow-6w-sigma falkow-6w-nocrit falkow-6w-sigma2 falkow-6w-basic falkow-6i-sigma falkow-6i-nocrit falkow-6i-basic navier-8w-sigma navier-8w-nocrit navier-8w-sigma2 navier-8w-basic navier-8i-sigma navier-8i-nocrit navier-8i-basic heat-12w-sigma heat-12w-nocrit heat-12w-sigma2 heat-12w-basic eq26-12w-sigma eq26-12w-nocrit eq26-12w-sigma2 eq26-12w-basic eq27-12w-sigma eq27-12w-nocrit eq27-12w-sigma2 eq27-12w-basic 157 157 10 10 157 163 4 4 4 4 4 86 4 4 86 5 5 5 6 6 6 - 9 9 86 5 5 5 378 246 43 43 43 208 28 28 28 121 1 1 1 10 1 1 1 9 time 18s 57m1s 61s 9h8m11s 1m45s 1m53s 1m44s 26s 3h46m29s 6m4s > 3 days 12s 16s 10s 1m10s 1m46s 2m15s 1m33s 2m4s 1m50s 24m33s 6m40s 14s 22s 25s 11s 5 8 5 157 25 34 172 9 22 9 - 15 37 86 7 137 7 378 557 790 557 1673 609 923 609 726 5 5 5 5 9 9 9 5 5 5 - 4 4 4 5 5 5 5 28 28 28 28 18 18 18 18 We give now some informations about the examples we have used for the ex- periments. See the Appendix of the paper for an explicit description of the test set. All considered examples are non-linear systems of ordinary or partial difference equations with constant coefficients which are of interest in literature. For instance, the tests falkow are obtained by the discretization of the Falkowich-Karman equa- tion which is a non-linear two-dimensional differential equation describing transonic flow in gas dynamics. The discretization we used are equations (41) in [11]. Then, the navier examples are based on equations e1, e2, e3, e4 of the system (13) in the paper [10] that are a finite difference scheme corresponding to the discretization (9) of the Navier-Stokes equations for two-dimensional viscous incompressible fluid flows. The tests heat are the discretization of the one-dimensional heat equation as described in the equations (10) and (11) of [22]. Finally, eq26 and eq27 are the equations (2.6) and (2.7) at page 24 of [14] which are examples of ordinary difference equations that have periodic solutions. By analyzing the experiments, it is sufficiently clear that the strategy imple- mented in SigmaGBasis is the safest one and hence on the average, the most efficient one. In fact, by decreasing the number of S-polynomials this strategy avoids the dramatical effects of involved reductions as for the tests falkow-6w and navier-8w. For simpler examples the four strategies appear essentially equivalent. The algorithms SigmaGBasis and SigmaGBasis2 lead to practically identical 24 R. LA SCALA computations but the latter method suffers of some overhead which is probably due to our still experimental implementation. For instance, even if the normal form modulo the ideal N is described in the Definition 7.8, in our implementation we obtain it by computations that is by adding a Grobner basis of N to the input basis for SigmaGBasis2. The proposed algorithms usually provide only partial informations about the structure of Grobner Σ-bases since they are in general infinite. Nevertheless, it is interesting to note that by means of the finite Σ-criterion we have been able to certify that the examples falkow, navier and heat have finite bases with respect to the weight ranking. In particular, the elements of the Grobner basis of falkow have maximum top order equal to 4 and hence they are certified in order 8 in about 4 minutes. The example navier has max top order equal to 6 and its certification is obtained in order 12 in less than one hour. Finally, the example heat has max top order 2 and it gets certification in order 4 in 0 seconds. 10. Conclusions and future directions This paper shows that one can not only generalize in a systematic way the Grobner bases theory and related algorithms to the algebras of partial difference polynomials but also make these methods really work by introducing suitable grad- ings for such algebras. In fact, weight and order functions provide a Noetherian subalgebras filtration that implies termination and completeness certification for actual computations that are performed within some bounded degree that is over a finite number of variables. We have then developed the first implementation of a variant of the Buchberger's algorithm for systems of linear or non-linear partial dif- ference equations. Even if such implementation is just experimental, the approach corresponding to the algorithm SigmaGBasis is strong enough to let it able to work with discretizations of real world systems of non-linear differential equations. In this paper we consider difference equations with constant coefficients and hence a next step along this line of research is to extend the proposed methods to systems of difference equations with non-constant coefficients that is to assume that Σ acts on the base field K in a non-trivial way. Moreover, since the algebras of partial difference polynomials are free objects in the category of commutative algebras endowed with the action of a monoid Σ isomorphic to Nr, a natural future direction consists in extending the ideas introduced here to other types of monoid symmetry over commutative algebras as the ones used, for instance, in algebraic statistic [4, 16]. Starting from Grobner bases, classical directions are the compu- tation of the Hilbert series and free resolutions that one may generalize to partial difference ideals or other types of invariant ideals. Finally, we aim to have the pro- posed algorithms implemented in the kernel of computer algebra systems in order to tackle involved problems related with the discretization of systems of partial differential equations [9, 11, 12]. Acknowledgments The author would like to thank Vladimir Gerdt for introducing him to the theory of difference algebras and supporting the preparation of testing examples. He is also grateful to the research group of Singular [7] for the courtesy of allowing access to their servers for performing computational experiments. Thanks also to the reviewer for all valuable remarks that have helped to make the paper more readable. GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 25 Falkovich-Karman (falkow) Appendix: the test set The base field is F = Q(h, τ, K, γ) (field of rational functions with rational co- efficients) where h is the space mesh step, τ is the time mesh step and K, γ are parameters of the corresponding differential equation. The algebra of partial dif- ference polynomials is P = F [ϕx(i, j, k), ϕy(i, j, k), ϕt(i, j, k), ϕ(i, j, k) i, j, k ≥ 0]. The monomial ordering of P is the lexicographic one based on the following weight ranking . . . ≻ ϕx(2, 0, 0) ≻ ϕy(2, 0, 0) ≻ ϕt(2, 0, 0) ≻ ϕ(2, 0, 0) ≻ ϕx(1, 1, 0) ≻ ϕy(1, 1, 0) ≻ ϕt(1, 1, 0) ≻ ϕ(1, 1, 0) ≻ ϕx(0, 2, 0) ≻ ϕy(0, 2, 0) ≻ ϕt(0, 2, 0) ≻ ϕ(0, 2, 0) ≻ ϕx(1, 0, 1) ≻ ϕy(1, 0, 1) ≻ ϕt(1, 0, 1) ≻ ϕ(1, 0, 1) ≻ ϕx(0, 1, 1) ≻ ϕy(0, 1, 1) ≻ ϕt(0, 1, 1) ≻ ϕ(0, 1, 1) ≻ ϕx(0, 0, 2) ≻ ϕy(0, 0, 2) ≻ ϕt(0, 0, 2) ≻ ϕ(0, 0, 2) ≻ ϕx(1, 0, 0) ≻ ϕy(1, 0, 0) ≻ ϕt(1, 0, 0) ≻ ϕ(1, 0, 0) ≻ ϕx(0, 1, 0) ≻ ϕy(0, 1, 0) ≻ ϕt(0, 1, 0) ≻ ϕ(0, 1, 0) ≻ ϕx(0, 0, 1) ≻ ϕy(0, 0, 1) ≻ ϕt(0, 0, 1) ≻ ϕ(0, 0, 1) ≻ ϕx(0, 0, 0) ≻ ϕy(0, 0, 0) ≻ ϕt(0, 0, 0) ≻ ϕ(0, 0, 0), or on the following index ranking . . . ≻ ϕx(2, 0, 0) ≻ ϕx(1, 1, 0) ≻ ϕx(0, 2, 0) ≻ ϕx(1, 0, 1) ≻ ϕx(0, 1, 1) ≻ ϕx(0, 0, 2) ≻ ϕx(1, 0, 0) ≻ ϕx(0, 1, 0) ≻ ϕx(0, 0, 1) ≻ ϕx(0, 0, 0) ≻ . . . ≻ ϕy(2, 0, 0) ≻ ϕy(1, 1, 0) ≻ ϕy(0, 2, 0) ≻ ϕy(1, 0, 1) ≻ ϕy(0, 1, 1) ≻ ϕy(0, 0, 2) ≻ ϕy(1, 0, 0) ≻ ϕy(0, 1, 0) ≻ ϕy(0, 0, 1) ≻ ϕy(0, 0, 0) ≻ . . . ≻ ϕt(2, 0, 0) ≻ ϕt(1, 1, 0) ≻ ϕt(0, 2, 0) ≻ ϕt(1, 0, 1) ≻ ϕt(0, 1, 1) ≻ ϕt(0, 0, 2) ≻ ϕt(1, 0, 0) ≻ ϕt(0, 1, 0) ≻ ϕt(0, 0, 1) ≻ ϕt(0, 0, 0) ≻ . . . ≻ ϕ(2, 0, 0) ≻ ϕ(1, 1, 0) ≻ ϕ(0, 2, 0) ≻ ϕ(1, 0, 1) ≻ ϕ(0, 1, 1) ≻ ϕ(0, 0, 2) ≻ ϕ(1, 0, 0) ≻ ϕ(0, 1, 0) ≻ ϕ(0, 0, 1) ≻ ϕ(0, 0, 0). The input partial difference polynomials for the computational experiments are E1 = −4hp(2, 1, 1) − hτ (γ + 1)px(2, 1, 0)2 + 2hτ Kpx(2, 1, 0) + 4hp(2, 1, 0)+ 2hτ py(1, 2, 0) − 4h2pt(1, 1, 0) + 4hp(0, 1, 1) − 2hτ py(1, 0, 0)+ hτ (γ + 1)px(0, 1, 0)2 − 2hτ Kpx(0, 1, 0) − 4hp(0, 1, 0), E2 = h 2 (px(1, 0, 0) + px(0, 0, 0)) − p(1, 0, 0) + p(0, 0, 0), E3 = h 2 (py(0, 1, 0) + py(0, 0, 0)) − p(0, 1, 0) + p(0, 0, 0), E4 = 2τ pt(0, 0, 1) − p(0, 0, 2) + p(0, 0, 0). Navier-Stokes (navier) The base field is K = Q(h, τ, Re) where Re is the Reynolds number. The algebra of partial difference polynomials is P = K[p(i, j, k), u(i, j, k), v(i, j, k) i, j, k ≥ 0]. The lexicographic monomial ordering of P is defined by following weight ranking . . . ≻ p(2, 0, 0) ≻ u(2, 0, 0) ≻ v(2, 0, 0) ≻ p(1, 1, 0) ≻ u(1, 1, 0) ≻ v(1, 1, 0) ≻ p(0, 2, 0) ≻ u(0, 2, 0) ≻ v(0, 2, 0) ≻ p(1, 0, 1) ≻ u(1, 0, 1) ≻ v(1, 0, 1) ≻ p(0, 1, 1) ≻ u(0, 1, 1) ≻ v(0, 1, 1) ≻ p(0, 0, 2) ≻ u(0, 0, 2) ≻ v(0, 0, 2) ≻ p(1, 0, 0) ≻ u(1, 0, 0) ≻ v(1, 0, 0) ≻ p(0, 1, 0) ≻ u(0, 1, 0) ≻ v(0, 1, 0) ≻ p(0, 0, 1) ≻ u(0, 0, 1) ≻ v(0, 0, 1) ≻ p(0, 0, 0) ≻ u(0, 0, 0) ≻ v(0, 0, 0), 26 R. LA SCALA or by the following index ranking . . . ≻ p(2, 0, 0) ≻ p(1, 1, 0) ≻ p(0, 2, 0) ≻ p(1, 0, 1) ≻ p(0, 1, 1) ≻ p(0, 0, 2) ≻ p(1, 0, 0) ≻ p(0, 1, 0) ≻ p(0, 0, 1) ≻ p(0, 0, 0) ≻ . . . ≻ u(2, 0, 0) ≻ u(1, 1, 0) ≻ u(0, 2, 0) ≻ u(1, 0, 1) ≻ u(0, 1, 1) ≻ u(0, 0, 2) ≻ u(1, 0, 0) ≻ u(0, 1, 0) ≻ u(0, 0, 1) ≻ u(0, 0, 0) ≻ . . . ≻ v(2, 0, 0) ≻ v(1, 1, 0) ≻ v(0, 2, 0) ≻ v(1, 0, 1) ≻ v(0, 1, 1) ≻ v(0, 0, 2) ≻ v(1, 0, 0) ≻ v(0, 1, 0) ≻ v(0, 0, 1) ≻ v(0, 0, 0). The input partial difference polynomials are Re )u(0, 2, 2) − 8τ h3(u(0, 1, 2)2+ Re h2(u(0, 0, 2) + u(0, 2, 0)), Re h2(v(0, 4, 2) + v(0, 2, 4)) + 16h4v(1, 2, 2) + 8τ h3(v(0, 2, 3)2+ Re h2(u(0, 4, 2) + u(0, 2, 4)) + 16h4u(1, 2, 2) + 8τ h3(u(0, 3, 2)2+ E1 = 2h(u(0, 2, 1) − u(0, 0, 1) + v(0, 1, 2) − v(0, 1, 0)), E2 = −4 τ u(0, 2, 3)v(0, 2, 3) + p(0, 3, 2)) + 16h2(−h2 + τ u(0, 2, 1)v(0, 2, 1) + p(0, 1, 2)) − 4 τ E3 = −4 τ u(0, 3, 2)v(0, 3, 2) + p(0, 2, 3)) + 16h2(−h2 + τ u(0, 1, 2)v(0, 1, 2) + p(0, 2, 1)) − 4 τ E4 = 4h2(v(0, 2, 4)2 + u(0, 4, 2)2 + v(0, 2, 0)2 + u(0, 0, 2)2 + p(0, 4, 2)+ p(0, 2, 4) + p(0, 2, 0) + p(0, 0, 2)) + 8h2(u(0, 3, 3)v(0, 3, 3) − v(0, 2, 2)2− u(0, 2, 2)2 − u(0, 3, 1)v(0, 3, 1) − u(0, 1, 3)v(0, 1, 3) + u(0, 1, 1)v(0, 1, 1))− 16h2p(0, 2, 2). Re )v(0, 2, 2) − 8τ h3(v(0, 2, 1)2+ Re h2(v(0, 2, 0) + v(0, 0, 2)), Heat (heat) The base field is by definition K = Q(h, τ ) and the partial difference polynomial algebra is P = K[x(i, j), t(i, j), u(i, j) i, j ≥ 0]. The lexicographic monomial order of P is obtained by the following weight ranking . . . ≻ x(2, 0) ≻ t(2, 0) ≻ u(2, 0) ≻ x(1, 1) ≻ t(1, 1) ≻ u(1, 1) ≻ x(0, 2) ≻ t(0, 2) ≻ u(0, 2) ≻ x(1, 0) ≻ t(1, 0) ≻ u(1, 0) ≻ x(0, 1) ≻ t(0, 1) ≻ u(0, 1) ≻ x(0, 0) ≻ t(0, 0) ≻ u(0, 0), The input is given by E1 = (u(1, 0) − u(0, 0))(x(0, 1) − x(0, 0))2 − (u(0, 2) − 2u(0, 1)+ u(0, 0))(t(1, 0) − t(0, 0)), E2 = x(1, 0) − x(0, 0), E3 = x(0, 1) − x(0, 0) − h, E4 = t(1, 0) − t(0, 0) − τ, E5 = t(0, 1) − t(0, 0). Equation 2.6 (eq26) One considers the algebra of ordinary difference polynomials P = Q[x(i) i ≥ 0] endowed with the lexicographic monomial ordering such that . . . ≻ x(4) ≻ x(3) ≻ x(2) ≻ x(1) ≻ x(0). The input ordinary difference polynomial is E = x(3)x(0) − x(2) − x(1) − 1. Equation 2.7 (eq27) We consider the same algebra P = Q[x(i) i ≥ 0] endowed with the same monomial ordering as in the previous example. The input polynomial is here E = x(4)x(2)x(0) − x(3)x(1). GR OBNER BASES AND GRADINGS FOR PARTIAL DIFFERENCE IDEALS 27 References [1] Aschenbrenner, M.; Hillar, C.J., Finite generation of symmetric ideals. Trans. Amer. Math. Soc., 359 (2007), no. 11, 5171 -- 5192. [2] Bergman, G. M., The diamond lemma for ring theory. Adv. in Math., 29 (1978), no. 2, 178 -- 218. [3] Bigatti, A.M.; Caboara, M.; Robbiano, L., Computing inhomogeneous Grobner bases. J. Symbolic Comput., 46 (2011), no. 5, 498 -- 510. [4] Brouwer, A.E.; Draisma, J., Equivariant Grobner bases and the Gaussian two-factor model, Math. Comp., 80, (2011), no. 274, 1123 -- 1133. [5] Buchberger, B., Ein algorithmisches Kriterium fur die Losbarkeit eines algebraischen Gle- ichungssystems.(German), Aequationes Math., 4 (1970), 374 -- 383. [6] Cohn, R.M., Difference algebra. Interscience Publishers John Wiley & Sons, New York- London-Sydney, 1965. [7] Decker, W.; Greuel, G.-M.; Pfister, G.; Schonemann, H.: Singular 3-1-5 -- A computer algebra system for polynomial computations (2012). http://www.singular.uni-kl.de [8] Eisenbud, D., Commutative algebra with a view toward algebraic geometry. Graduate Texts in Mathematics, 150. Springer-Verlag, New York, 1995. [9] Gerdt, V.P., Consistency Analysis of Finite Difference Approximations to PDE Systems. In: Adam G. et al. (Eds.), Proc. of Mathematical Modeling and Computational Physics. MMCP 2011, 28 -- 42, Lecture Notes in Comput. Sci., 7175, Springer, Heidelberg, 2012. [10] Gerdt, V.P.; Blinkov, Y.A., Involution and Difference Schemes for the Navier-Stokes Equa- tions. In: Gerdt V.P. et al. (Eds.), Proc. of Computer Algebra in Scientific Computing. CASC 2009, 94 -- 105, Lecture Notes in Comput. Sci., 5743, Springer, Berlin, 2009. [11] Gerdt, V.P.; Blinkov, Y.A.; Mozzhilkin, V.V., Grobner bases and generation of difference schemes for partial differential equations. SIGMA Symmetry Integrability Geom. Methods Appl., 2, (2006), Paper 051, 26 pp. [12] Gerdt, V.P.; Robertz, D., A Maple Package for Computing Grobner Bases for Linear Re- currence Relations. Nucl. Instrum. Methods, 599 (2006), 215 -- 219. http://wwwb.math.rwth- aachen.de/Janet [13] Gondran, M.; Minoux, M., Graphs, dioids and semirings. New models and algorithms. Oper- ations Research/Computer Science Interfaces Series, 41. Springer, New York, 2008. [14] Grove, E.A.; Ladas, G., Periodicities in Nonlinear Difference Equations. Advances in Discrete Mathematics and Applications, 4. Chapman & Hall/CRC, Boca Raton, FL, 2005. [15] Higman, G., Ordering by divisibility in abstract algebras. Proc. London Math. Soc. (3), 2 (1952), 326 -- 336. [16] Hillar, C.J.; Martin del Campo, A., Finiteness theorems and algorithms for permutation invariant chains of Laurent lattice ideals. J. Symbolic Comput., 50 (2013), 314 -- 334. [17] Kolchin, E.R., Differential algebra and algebraic groups. Pure and Applied Mathematics, Vol. 54. Academic Press, New York-London, 1973. [18] La Scala, R., Extended letterplace correspondence for nongraded noncommutative ideals and related algorithms, preprint (2012), 1 -- 20. arXiv:1206.6027 [19] La Scala, R.; Levandovskyy, V., Letterplace ideals and non-commutative Grobner bases. J. Symbolic Comput., 44 (2009), 1374 -- 1393. [20] La Scala, R.; Levandovskyy, V., Skew polynomial rings, Grobner bases and the letterplace embedding of the free associative algebra, J. Symbolic Comput., 48 (2013), 110 -- 131. [21] Levandovskyy V.; Martin B., A Symbolic Approach to Generation and Analysis of Finite Difference Schemes of Partial Differential Equations. In: Langer U. et al. (Eds.), Numerical and Symbolic Scientific Computing: Progress and Prospects. Springer (2012), 123 -- 156. [22] Levi, D., Lie symmetries for lattice equations. Note Mat., 23 (2004/05), no. 2, 139 -- 156. [23] Levin, A., Difference algebra. Algebra and Applications, 8. Springer, New York, 2008. [24] Meyer D.A; Wallach N., Invariants for multiple qubits: the case of 3 qubits, Mathematics of quantum computation, Comput. Math. Ser., Chapman & Hall/CRC, Boca Raton, FL, 2002, 77 -- 97. [25] Ritt, J.F., Differential Equations From the Algebraic Standpoint. Amer. Math. Soc. Collo- quium Publ., Vol. 14, AMS, New York, 1932. [26] Ritt, J.F., Differential Algebra. Amer. Math. Soc. Colloquium Publ., Vol. 33, AMS, New York, 1950. 28 R. LA SCALA [27] Seiler, W.M. Involution. The formal theory of differential equations and its applications in computer algebra. Algorithms and Computation in Mathematics, Vol. 24. Springer-Verlag, Berlin, 2010. ∗ Dipartimento di Matematica, via Orabona 4, 70125 Bari, Italia E-mail address: [email protected]
1602.00296	1	1602	2016-01-31T18:33:33	A Factorization Algorithm for G-Algebras and Applications	[ "math.RA", "cs.SC", "math.OA" ]	It has been recently discovered by Bell, Heinle and Levandovskyy that a large class of algebras, including the ubiquitous $G$-algebras, are finite factorization domains (FFD for short). Utilizing this result, we contribute an algorithm to find all distinct factorizations of a given element $f \in \mathcal{G}$, where $\mathcal{G}$ is any $G$-algebra, with minor assumptions on the underlying field. Moreover, the property of being an FFD, in combination with the factorization algorithm, enables us to propose an analogous description of the factorized Gr\"obner basis algorithm for $G$-algebras. This algorithm is useful for various applications, e.g. in analysis of solution spaces of systems of linear partial functional equations with polynomial coefficients, coming from $\mathcal{G}$. Additionally, it is possible to include inequality constraints for ideals in the input.	math.RA	math	A Factorization Algorithm for G-Algebras and Applications David R. Cheriton School of Computer Science Albert Heinle University of Waterloo 200 University Avenue West Waterloo, ON N2L 3G1, Canada [email protected] Viktor Levandovskyy Lehrstuhl D für Mathematik RWTH Aachen University Pontdriesch 16 52062 Aachen, Germany [email protected] ABSTRACT It has been recently discovered by Bell, Heinle and Levan- dovskyy that a large class of algebras, including the ubiqui- tous G-algebras, are finite factorization domains (FFD for short). Utilizing this result, we contribute an algorithm to find all distinct factorizations of a given element f ∈ G, where G is any G-algebra, with minor assumptions on the underlying field. Moreover, the property of being an FFD, in combination with the factorization algorithm, enables us to propose an analogous description of the factorized Grobner basis algo- rithm for G-algebras. This algorithm is useful for various applications, e.g. in analysis of solution spaces of systems of linear partial functional equations with polynomial coeffi- cients, coming from G. Additionally, it is possible to include inequality constraints for ideals in the input. Keywords G-Algebras, Grobner Bases, Factorization 1. INTRODUCTION Notations: Throughout the paper we denote by K a field. In the algorithmic part we will assume K to be a computable field. N0 = N ∪ {0} is the set of natural numbers including zero. For a K-algebra R we denote by U (R) the group of invertible (unit) elements of R, which is nonabelian in gen- eral. For f ∈ R we denote by Rf the left ideal, generated by f . The main focus in this paper lies in so called G-algebras, which are defined as follows. Definition 1. For n ∈ N and 1 ≤ i < j ≤ n consider the units cij ∈ K∗ and polynomials dij ∈ K[x1, . . . , xn]. Sup- pose, that there exists a monomial total well-ordering ≺ on K[x1, . . . , xn], such that for any 1 ≤ i < j ≤ n either dij = 0 or the leading monomial of dij is smaller than xixj with re- spect to ≺. The K-algebra A := Khx1, . . . , xn {xj xi = Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. ISSAC '16 Waterloo, Ontario, Canada Copyright 20XX ACM X-XXXXX-XX-X/XX/XX ...$15.00. cij xixj + dij : 1 ≤ i < j ≤ n}i is called a G-algebra, if {xα1 1 n : αi ∈ N0} is a K-basis of A. · . . . · xαn G-algebras [1, 19] are also known as algebras of solvable type [18, 20] and as PBW algebras [4, 5]. G-algebras are Noethe- rian domains of finite global, Krull and Gel'fand-Kirillov dimensions. We assume that the reader is familiar with the basic termi- nology in the area of Grobner bases, both in the commuta- tive as well as in the non-commutative case. We recommend [3, 5, 19] as literature on this topic. Recall, that r ∈ R \ {0} is called irreducible, if in any factorization r = ab either a ∈ U (R) or b ∈ U (R) holds. Otherwise, we call r reducible. Definition 2 (cf. [2]). Let A be a (not necessarily com- mutative) domain. We say that A is a finite factorization domain (FFD, for short), if every nonzero, non-unit ele- ment of A has at least one factorization into irreducible el- ements and there are at most finitely many distinct factor- izations into irreducible elements up to multiplication of the irreducible factors by central units in A. 2. MOTIVATION AND APPLICATIONS Problem 1. Let A be a finite factorization domain and a K-algebra. Given f ∈ A \ (U (A) ∪ {0}), compute all its factorizations of the form f = c · f1 · · · fn, where c ∈ U (A) and fi ∈ A \ U (A) are irreducible. This paper is devoted in part to the algorithmic solution of Problem 1 for a broad class of G-algebras. With this al- gorithm one can approach a number of important problems, which we discuss in details. Let A be a K-algebra and 0 6= L ⊂ A a finitely generated left ideal. Problem 2. Compute a proper left ideal N ) L. Unfortunately, it is not known in general, whether the prob- lem of left maximality of a given ideal with respect to in- clusion is decidable. Therefore we are interested in the local negative form of it. Namely, if Problem 2 can be solved, then L is not left maximal. Moreover, for any N as above, we have a surjection from A/L to its proper factor-module A/N , in other words the exact sequence of left A-modules 0 → N/L → A/L → A/N → 0 which contributes to the knowledge of the structure of A/L. Suppose that f ∈ A \ {0}, f /∈ L has finitely many factor- izations f = gihi up to multiplication by central units, where i ∈ I, I < ∞ and gi, hi /∈ U (A). We do not require that gi or hi are irreducible. Suppose, that L ( L + Af ( A. Then L + Af ⊆Ti∈I (L + Ahi), hence there is a natural surjective homomorphism of left A-modules A L + Af → A Ti∈I (L + Ahi) → 0 Relations with solution spaces: Let F be an arbi- trary (in particular, not necessarily finitely generated) left A-module, which one can think about as of the space of solu- tions for A-modules. Then, for fixed K and an A-module M one denotes Sol(M, F) := HomA(M, F), which is a K-vector space and an EndA(M )-module [22]. By invoking the Noether-Malgrange isomorphism [22], we obtain for an the natural injective map of K-vector spaces 0 → Sol(cid:18) A Ti∈I (L + Ahi) , F(cid:19) → Sol(cid:18) A L + Af , F(cid:19) . The latter sheds light on the structure of the space of solu- tions of A/(L + Af ). Note, that the following version of the left Chinese re- mainder theorem for modules holds: For a general S, it is unknown, whether LS is computable. If A is the nth Weyl algebra, S = K[x1, . . . , xn] \ {0} and L ⊂ A has finite holonomic rank, then there is an algorithm [24, 23] to compute LS (known as the Weyl closure of L). The factorization can be used in the process of computing LS as follows. Let A be an FFD. Given ℓ ∈ L, one computes finitely many factorizations ℓ = aibi, i ∈ I for some finite indexing set I. Then let J := {j ∈ I (aj, bj ) ∈ S × L}. If J 6= ∅, one has L + {Abj : j ∈ J} ⊆ LS. In such a way one obtains an approximation to LS. Note, that LS = A if and only if L ∩ S 6= ∅. 3. HOW TO FACTOR IN G-ALGEBRAS 3.1 General Algorithm In a recent publication [2, Theorem 1.3], it was proven that each G-algebra G is a finite factorization domain. In the same paper, an outline was given how one could find all possible factorizations of an element in G. In this section, we will provide a thorough description of an algorithm to find all possible factorizations of an element in a G-algebra G, up to multiplication by central units. For this, we need to make a further assumption on our field K, which holds for most practical choices of K. Theorem 1. Let A be a K-algebra, I a finite set of in- dices and {Li : i ∈ I} are left ideals in A. Consider the homomorphism Assumption: There exists an algorithm to determine if a polynomial p in K[x] has roots in K. If p has roots in K, then this algorithm can produce all K-roots of p. Li A/\i∈I φ −→Mi∈I A/Li, Then the following holds a+\i∈I 1) φ is injective Li 7→ (a+L1, . . . , a+LI). 2) if ∀i, j ∈ I, i 6= j holds Li + Lj = A, then φ is surjec- tive. Of course, one can assume that Li are proper nonzero ideals. In the second item of the Theorem 1 one says that the collection {Li : i ∈ I} is left comaximal. Then φ is an isomorphism and one has a finite direct sum decomposition decomposition of the solution spaces of the module A/Ti∈I Li. Hence, there is a direct sum Sol A/\i∈I Li, F! = Sol Mi∈I A/Li, F! =Mi∈I Sol (A/Li, F) . Note, that the right hand side can be a direct sum even if the condition (2) is not satisfied, see Example 3. Another application: Let A be a domain and S ⊂ A be multiplicatively closed Ore set. By Ore's Theorem the local- ization S−1A exists and there is an injective homomorphism A → S−1A [5]. A left ideal L ⊂ A is called left S-closed if LS = L, where LS := {a ∈ A ∃s ∈ S sa ∈ L} ⊇ L is the S-closure of L. There is another characterization of S-closedness: LS = ker(A → S−1(A/L)), where the latter homomorphism of A- modules is a 7→ 1−1a + S−1L. Then LS/L is the S-torsion submodule of A/L and A/LS has no S-torsion. Problem 3. Given S ⊂ A an Ore set and L ⊂ A, give an algorithm to compute LS. n : α ∈ Nn 1 · . . . · xαn Recall, that {xα = xα1 0 } is a K-basis of G. With respect to an admissible monomial ordering ≺ on G we can uniquely write every g ∈ G \ {0} as g = cαxα + tg with cα ∈ K \ {0}. Moreover, either tg = 0 or xβ ≺ xα for any summand cβxβ, cβ 6= 0 of tg. Then lm(g) = xα is the leading monomial of g and lc(g) = cα is the leading coefficient of g. A polynomial g 6= 0 with lc(g) = 1 is called monic. It is important to recall [19], that lm(xα · xβ) = xα+β holds ∀α, β ∈ Nn 0 in a G-algebra. Proof of Algorithm 1. Let us begin with discussing the termination. The set M in line 2 is finite, as it is a permutation of a finite product of the variables in G. Since G is a G-algebra, the set of total well-orderings on it, satisfying the Definition 1, is nonempty. By [4], in this set there is a weighted degree total ordering, say ≺w with strictly positive weights. Without loss of generality let us assume this is the ordering we are working with. Thus for any monomial there are only finitely many monomials which are smaller with respect to ≺w. In particular, this applies to lt(a) and lt(b) in line 5. The variety V will be a finite set due to the fact that G is an FFD. Thus, the set R in line 9 will also be finite. The recursive call will also terminate, since in each step we either discover that we cannot refine our factorization any more, or we split a given factor into two factors of strictly smaller degrees. For the correctness discussion of our algorithm, we need to show that we can calculate the variety V in line 8. We know, since G is an FFD, that the ideal generated by F is either zero-dimensional over K[x] or it is an intersection of such with a higher-dimensional ideal H, whereas the variety of H does not contain points from an affine space over K. Hence we proceed with the zero-dimensional component F0 of F . Algorithm 1 Factoring an element g in a G-algebra G Input: g ∈ G \ K. Output: {(g1, . . . , gm) m ∈ N, gi ∈ G \ K for i ∈ {1, . . . , m}, g1 · · · gm = g} (up to multiplication of each factor by a central unit). Assumption: An admissible monomial ordering ≺ on G is fixed and g is monic with respect to it. 1: R := {} 2: M := {(p1, . . . , pν) ν ∈ N, pi ∈ {x1, . . . , xn}, lm(p1 · . . . · pν) = lm(g)} 3: for (p1, . . . , pν) ∈ M do for i := 1 to ν − 1 do 4: 5: 6: 7: 8: 9: Set up an ansatz for the K-coefficients of a · b = g with lt(a) = p1 · . . . · pi and lt(b) = pi+1 · . . . · pν. F := the reduced Grobner basis w.r.t. an elimina- tion ordering of the ideal generated by the coeffi- cients of a · b − g. if F 6= {1} then V := Variety of hF i in an affine space over K. R := R ∪ {(a, b) a, b ∈ G, a · b = g, where the coefficients of a, b are given by v ∈ V } end if end for 10: 11: 12: end for 13: if R = {} then return {(g)} 14: 15: else 16: 17: end if 18: return R Recursively factor a and b for each (a, b) ∈ R. Our assumption above states that we can find all K-roots of a univariate polynomial. Since F0 is zero-dimensional, for any variable xi there is the corresponding univariate polyno- mial, generating the principal ideal F0∩K[xi]. By backwards substitution, we obtain the entire K-variety of the ideal gen- erated by F0. Example 1. Let us consider the universal enveloping al- gebra U (sl2) of sl2 [11], represented by Khe, f, h f e = ef − h, he = eh + 2e, hf = f h − 2f i. In U (sl2), we want to factorize the element p :=e3f + e2f 2 − e3 + e2f + 2ef 2 − 3e2h − 2ef h − 8e2 + ef + f 2 − 4eh − 2f h − 7e + f − h. We fix the lexicographic ordering on U (sl2), i.e. the leading term of p is e3f . Therefore the set M in line 2 is given as M := {(e, e, e, f ), (e, e, f, e), (e, f, e, e), (f, e, e, e)}. When choosing (e, e, e, f ), for i = 1 one obtains the fac- torization p = (e + 1) · (e2f + ef 2 − 3eh − 2f h − e2 + f 2 − 7e + f − h). By picking (e, e, f, e), for i = 3 one obtains two more fac- torizations, namely p = (e2f + 2ef − 2eh − e2 − 4e + f − 2h − 3) · (e + f ) and p = (e2f + ef 2 − 2eh − e2 + f 2 − 3e − f − 2h) · (e + 1). All the other combinations either produce the same factor- izations or none. When recursively calling the algorithm for each factor in the found factorizations, we discover that the first two fac- torizations have a reducible factor. In the end, one obtains the following two distinct factorizations of p into irreducible factors: p =(e2f + ef 2 − 2eh − e2 + f 2 − 3e − f − 2h) · (e + 1) =(e + 1) · (ef − e + f − 2h − 3) · (e + f ). 3.2 Implementation We have developed an experimental implementation of Al- gorithm 1 in the computer algebra system Singular [10]. We will make it available as part of ncfactor.lib. Our newly implemented procedures factorize elements in any G- algebra, whose ground field is F(q1, . . . , qn), where F is either Q or a finite prime field and qi are transcendental over F. We designed the software in a modular way, so that during runtime our function checks if a more efficient factorization algorithm is available for the specific given G-algebra and/or input polynomial. If this is the case, the input is re-directed to this function. In this way, the user can call the general function to factor elements in any one of the supported G- algebras, and runs the available optimized algorithms, where available, without calling them individually. 3.3 Possible Improvements Algorithm 1 solves the problem of finding all possible fac- torizations of an element in a G-algebra, but it will not be very efficient in general. This is not only due to the complex- ity of the necessary calculation of a Grobner basis [21], but also the size of the set M is a bottleneck. In [12, 13], an algo- rithm for factoring elements in the nth Weyl algebra is pre- sented, which is similar to Algorithm 1. The main difference is that the Zn-graded structure is utilized. There, the homo- geneous polynomials of degree zero form a K-algebra A(0) n , which is isomorphic to a commutative multivariate polyno- mial ring. The set of homogeneous polynomials of degree z ∈ Zn \ {0n} has the structure of a cyclic A(0) n -bimodule. Hence, factorization of homogeneous polynomials with re- spect to the Zn-grading reduces to factoring commutative polynomials with minor additional combinatorial steps. An inhomogeneous polynomial f has now the highest graded part α(f ) and the lowest graded part ω(f ), both of them rather polynomials than monomials. Hence α(f ), ω(f ) have potentially smaller numbers of different factorizations than the permutations of the leading term collected in M in Algo- rithm 1. Indeed, it suffices to consider firstly factorizations into two polynomials and for each candidate pair an ansatz is made for the graded terms between the highest and the lowest graded parts. This means, that the set M has smaller size in general when using this technique. Additionally, this approach takes the lowest graded part into account, which allows to eliminate certain invalid cases beforehand. The performance increase is reflected by the benchmarks pre- sented in [12, 17]. Hence, for practical implementations of Algorithm 1, one should examine each possible G-algebra separately and take advantage of potential extra structure, like the presence of nontrivial Zn-grading or an isomorphism to an algebra with this structure. We will conclude this section by summarizing the condi- tions that can lead to an improved version of Algorithm 1. Let A be a K-algebra, which possesses a nontrivial (i.e. not all weight vectors are zero) Zn-(multi)grading. Then one can infer the following additional information: 1. For z ∈ Zn, Az := {a ∈ A : deg(a) = z} ∪ {0} is a K- vector space. Moreover, ⊕zAz = A and AiAj ⊆ Ai+j for all i, j ∈ Zn. 2. A0n , the graded part of degree zero, is a K-algebra itself (since A0A0 ⊆ A0). 3. For z ∈ Zn \ {0n}, the z-th graded part Az is an A0n - bimodule (since A0Az, AzA0 ⊆ Az). In order to be useful for factorizing purpose, this grading should have the following properties: 4. The graded part of degree zero, A0n , which is a K- algebra, is additionally an FFD with "easy" factor- ization, preferably the commutative polynomial ring. Furthermore, for keeping the set M in Algorithm 1 small, it would be desirable if in A0n a randomly cho- sen polynomial is irreducible with high probability. 5. The irreducible elements in A0n, that are reducible in A, can be identified and factorized in an efficient manner. Preferably, one has a finite number of monic elements of such type. 6. For z ∈ Zn \ {0n}, the z-th graded part Az is a finitely generated A0n -bimodule, preferably a cyclic bimodule. Then the Algorithm 1 can be modified along the lines of algorithms from [12, 13], which we have also sketched above. Let us illustrate this approach by a concrete example. Example 2. As in Example 1, let A = U (sl2), that is any of ϕ or ψ would have a non-trivial A0 factor, we would obtain again that p is reducible in A0. This leaves as only options p = ef or p = f e = ef − h, as claimed. Thus, we have shown that irreducible elements in A0, which are reducible in A, can be easily identified and factored. Now consider the same polynomial p as in Example 1. With respect to the (1, −1, 0)-grading it decomposes into the following graded parts: α(p) = −e3, ω(p) = f 2 (as we see, in this case we have monomials in graded parts, while in general rather polynomials appear) and the intermediate parts are e3f − 3e2h − 8e2 + e2f − 4eh − 7e + + e2f 2 − 2ef h + ef − h + 2ef 2 − 2f h + f . deg:2 {z {z deg:0 } } deg:1 {z } deg:−1 {z } Among the factorizations of α(p) = −e3 and ω(p) = f 2 into two factors, consider the case (−e2) · e and f · f . Thus, we're looking for a, b ∈ A with α(a) = e2, ω(a) = f and α(b) = e, ω(b) = f and p = ab holds. In b we have only one possible intermediate graded part b0(ef, h), namely of degree 0 since deg α(b) = 1 and deg ω(b) = −1. In a we have to specify the parts of degrees 1 resp. 0, that is a1(ef, h)·e resp. a0(ef, h). After the multiplication, we obtain the following graded decomposition of intermediate graded terms of ab: −e2b0 + a1e2 + a1eb0 + a0e − e2f + + a1ef + a0b0 + ef − h + f b0 + a0f . deg:2 {z } {z deg:0 deg:1 {z } } deg:−1 {z } By fixing the maximal possible degree of a0, a1, b0 ∈ K[ef, h], we can create and solve a system of equations which the co- efficients of a0, a1, b0 have to satisfy. In this example an ansatz in terms of 1, h, ef , i.e. 9 unknown coefficients, leads to the system of 18 at most quadratic equations, which leads to the unique solution: b0(ef, h) = 0, a0(ef, h) = 2ef −2h−3 and a1(ef, h) = ef − h − 2. Substituting the polynomials, we arrive at the following factorization with polynomials sorted according to the grading: A = Khe, f, h f e = ef − h, he = eh + 2e, hf = f h − 2f i. p = (−e2 + e2f − 2eh − 4e + 2ef − 2h − 3 + f ) · (e + f ) Af first, let us determine which gradings are possible. Let we, wf and wh be the weights of the variables, not all zero. The two last relations of A imply that wh = 0, and the first one implies we + wf = wh = 0, that is wf = −we. Hence a Z-grading (we, wf , wh) = (1, −1, 0) is enough for our purposes, since A0 = K[ef, h] is commutative and the z-th graded part is a cyclic A0-bimodule, generated by ez if z > 0 and by f z otherwise. This property guarantees, that ∀r ∈ K[ef, h] and ∀z ∈ N there exists q1, q2 ∈ K[ef, h], such that rez = ezq1 and ezr = q2ez and the same holds for the multiplication by f z. Note, that deg(qi) = deg(r). We claim that the only monic irreducible elements in A0, which are reducible in A, are given by ef and ef − h. The proof to this claim is similar to the one for [13, Lemma 2.4], which we outline here: Let p be an irreducible element in A0, which reduces into p = ϕ · ψ in A, where ϕ, ψ ∈ A \ K are monic. Since A is a domain, the factors ϕ, ψ are homogeneous with deg(ϕ) = k and deg(ψ) = −k for some k ∈ Z. If k > 1 or k = 0, p would be reducible in A0, which violates our assumption. Hence only k = 1 is possible. If This is already known to us from the Example 1. In an anal- ogous way one can address other factorizations. Note, that in the ansatz we made, significantly less variables for un- known coefficients and a system of less equations of smaller total degree were used, compared to the general Algorithm. 4. THE FACTORIZED GRÖBNER BASIS AL- GORITHM FOR G-ALGEBRAS In what follows, by the term ideal we always mean a left ideal (unless otherwise specified). The factorized Grobner approach has been studied ex- tensively for the commutative case [8, 7, 9, 14, 15], and implementations are e.g. provided in the computer algebra systems Singular [10] and Reduce [16]. The leading motivation is to split a Grobner basis com- putation into smaller pieces with respect to the degrees of their generators. The union of the varieties of the ideals generated by these smaller pieces equals the variety of the initial system. In the commutative case, there is also a way to constrain the solution space. One can provide an extra set of elements, that should not be reducible by the computed Grobner ba- sises. In this way, one excludes certain unwanted solutions, which is useful in practice. The search for varieties in the commutative case translates to the search for solutions in the non-commutative case: All G-algebras are finite factorization domains and a general factorization algorithm via Algorithm 1 is given. Many of them are abstractions of algebras of operators, and one is interested to find common solutions of certain sets of opera- tors, written as polynomials. Right hand factors of elements correspond to partial solutions, and hence a split similar to the commutative case is helpful to recover partial solutions. Motivated by this observation, we attempt to generalize the factorized Grobner basis algorithm to the G-algebra case in this section. Our algorithm includes the possibility to intro- duce constraints, similar to the methods in the commutative case. Unfortunately, not all nice properties transfer into the non-commutative case, as the following example depicts. Example 3. In the commutative case, one has the prop- erty that the radical of the input ideal will be equal to the intersection of the radicals of all ideals computed by the fac- torized Grobner basis algorithm. We will show via a counter-example that we do not have the same property for G-algebras. Consider p =(x6 + 2x4 − 3x2)∂2 − (4x5 − 4x4 − 12x2 − 12x)∂ + (6x4 − 12x3 − 6x2 − 24x − 12) ∈ A1. This polynomial appears in [24] and has two different fac- torizations, namely p =(x4∂ − x3∂ − 3x3 + 3x2∂ + 6x2 − 3x∂ − 3x + 12)· (x2∂ + x∂ − 3x − 1) =(x4∂ + x3∂ − 4x3 + 3x2∂ − 3x2 + 3x∂ − 6x − 3)· (x2∂ − x∂ − 2x + 4) A reduced Grobner basis of hx2∂ + x∂ − 3x − 1i ∩ hx2∂ − x∂ − 2x + 4i, computed in Singular [10], is given by {3x5∂2 + 2x4∂3 − x4∂2 − 12x4∂ + x3∂2 − 2x2∂3 + 16x3∂ + 9x2∂2 + 18x3 + 4x2∂ + 4x∂2 − 42x2 − 4x∂ − 12x − 12, 2x4∂4 − 2x4∂3 + 11x4∂2 + 12x3∂3 − 2x2∂4 − 2x3∂2 + 10x2∂3 − 44x3∂ − 17x2∂2 + 64x2∂ + 12x∂2 + 66x2 + 52x∂ + 4∂2 − 168x − 16∂ − 60}. Hence, one main difference will be that we do not claim that the union of all solutions of our smaller pieces in the factorizing Grobner basis algorithm will always be equal to all common solutions of the initial set of polynomials. In general, we will only find a subset of all solutions using our method. However, in this example, the space of holomorphic solutions of the differential equation associated to p in fact coincides with the union of the solution spaces of the two generators of the intersection stated above. Definition 3. Let B, C be finite subsets in G. We call the tuple (B, C) a constrained Grobner tuple, if B is a Algorithm 2 Factorized Grobner bases Algorithm for G- Algebras (FGBG) Input: B := {f1, . . . , fk} ⊂ G, C := {g1, . . . , gl} ⊂ G. Output: R := {( B, C) ( B, C) is factorized constrained Grobner tuple} with Assumption: All elements in B and C are monic. hBi ⊆T( B, C)∈Rh Bi 1: for i = 1 to k do 2: 3: if fi is reducible then i i 4: 5: i , f (2) M := {(f (1) lc(f (2) if there exists (a, b), (a, b) ∈ M with a 6= a then ∈ G \ K, lc(f (1) is irreducible} f (1) i = fi, f (1) ) = 1, f (1) · f (2) , f (2) i i i i ) = i return [(a,b)∈M FGBG (B \ {fi}) ∪ {b}, C ∪ [(a,b)∈M  b6=b {b}  end if 6: 7: end if 8: end for 9: P := {(fi, fj ) i, j ∈ {1, . . . , k}, i < j} 10: while P 6= ∅ do Pick (f, g) ∈ P 11: P := P \ {(f, g)} 12: s := S-polynomial of f and g 13: h := NF(s, B) 14: if h 6= 0 then 15: 16: 17: 18: 19: 20: 21: 22: 23: 24: 25: end while 26: return {(B, C)} end if P := P ∪ {(h, f ) f ∈ B} B := B ∪ {h} return FGBG(B ∪ {h}, C) if h is reducible then return ∅ end if end if if there exists i ∈ {1, . . . , l} with NF(gi, B) = 0 then Grobner basis of hBi, and NF(g, B) 6= 0 for every g ∈ C. We call a constrained Grobner tuple factorized, if every f ∈ B is either irreducible or has a unique irreducible left divisor. It is possible to strengthen the assumptions on a factorized constrained Grobner tuples by only allowing completely irre- ducible elements in B, which might be preferable depending on the concrete problem. However, in our application, we allow elements with only one factorization. In this way, we increase the number of solutions we can find for a certain system B ⊂ G by using our generalized factorized Grob- ner basis algorithm. This methodology also appears in the context of semifirs, where the concept of so called block fac- torizations or cleavages has been introduced to study the reducibility of a principal ideal [6, Chapter 3.5]. Proof of Algorithm 2. We will first discuss the termi- nation aspect of Algorithm 2. Since M as calculated in line 3 is of finite cardinality, the existence check in line 4 can be done in a finite number of steps. Line 5 consists of a finite number of recursive calls to FGBG. The algorithm reaches this line if there is an element f in B, which is reducible and has a non-unique irreducible left divisor. In each recursive call, the algorithm is called with an altered version of the set B, where f is being replaced in B by b ∈ G, where b is chosen such that there exists an irreducible a in G with f = ab. Therefore, after a finite depth of recursion, FGBG will be called with a set B containing elements that are ei- ther irreducible or have an unique irreducible left divisor. We can make this assumption on B when FGBG reaches line 9. Lines 10 -- 25 describe the Buchberger algorithm to compute a Grobner basis, with two differences: 1. If the normal form h of an S-polynomial with respect to B is not 0, we check h for reducibility. If h is reducible, we call FGBG recursively, adding h to B. 2. We check the system for consistency, i.e. if there is an element in C that reduces with respect to B, we return the empty set. Each recursive call will terminate, since we add an element to B that will reduce an S-polynomial to zero, which could not be reduced to zero before. For the correctness discussion, one observes that lines 1 -- 8 serve the purpose to split the computation based on the re- ducibility of the elements in the initial set B. If an element f ∈ B factorizes in more than one way, we recursively call FGBG with (B \ {f }) ∪ {b} as the generator set for each maximal right hand factor b of f . Hence, the left ideal gen- erated by (B \ {f }) ∪ {b} will contain hBi, and thus hBi will be contained in the intersection of all of them, as required. As already mentioned in the termination discussion, lines 10 -- 25 describe the Buchberger algorithm. After computing an S-polynomial h, we check for its reducibility. If there is more than one maximal right factor r of h, we call FGBG recursively and add h to our set B. Here, we have again a guarantee that the left ideal generated by B is a subset of the left ideal generated by B ∪ {h}. The additional constraints that we impose on each recur- sive call enable us to minimize our computations, but do not violate the subset property. In the end, it is ensured that in all computed constrained Grobner tuples ( B, C), no element in C lies in the left ideal generated by B. Example 4. Let us execute FGBG on an example. Let B := {∂4 + x∂2 − 2∂3 − 2x∂ + ∂2 + x + 2∂ − 2, x∂3 + x2∂ − x∂2 + ∂3 − x2 + x∂ − 2∂2 − x + 1} be a subset of the first Weyl algebra A1. We assume that C := {∂ − 1}, and that our ordering is the degree reverse lexicographic one with ∂ > x. This example is taken from the Singular manual [10] (and it is a Grobner basis for the left ideal h∂2 + xi ∩ h∂ − 1i; hence we would expect the output with our chosen C to be h∂2 + xi). Each element factors separately as f1 :=∂4 + x∂2 − 2∂3 − 2x∂ + ∂2 + x + 2∂ − 2 =(∂3 + x∂ − ∂2 − x + 2) · (∂ − 1) =(∂ − 1) · (∂3 + x∂ − ∂2 − x + 1) respectively f2 :=x∂3 + x2∂ − x∂2 + ∂3 − x2 + x∂ − 2∂2 − x + 1 =(x∂2 + x2 + ∂2 + x − ∂ − 1) · (∂ − 1) =(x∂ − x + ∂ − 2) · (∂2 + x). Hence, in line 5, FGBG will return two recursive calls of itself, namely • FGBG({∂ − 1, f2}, {∂ − 1, ∂3 + x∂ − ∂2 − x + 1}) • FGBG({∂3 + x∂ − ∂2 − x + 1, f2}, C) For simplicity, we will ignore the first call, as C contains ∂ − 1, which also appears in the generator list. Furthermore, the new element b1 := ∂3 + x∂ − ∂2 − x + 1 only has one possible factorization. Therefore, we consider now the factorizations of f2. This leads again in line 5 to two recursive calls: • FGBG({b1, ∂ − 1}, {∂ − 1, ∂2 + x}) • FGBG({b1, ∂2 + x}, C) As before, we can ignore the first recursive call. Thus, we are left with ({b1, ∂2 + x}, C) to proceed on line 9. The normal form of the S-polynomial of b1 and ∂2 + x is equal to zero. Further, the normal form of b1 with respect to h∂2 + xi, is equal to zero, i.e. ∂2 + x is a right divisor of b1. Hence, we can omit b1 and our complete Grobner basis is given by {∂2 + x}. Since NF(∂ − 1, h∂2 + xi) 6= 0, our algorithm returns {({∂2 + x}, C)} as final output. Note, that if we would have chosen C = ∅ in the beginning, the output of our algorithm would have been {({∂ − 1}, {b1}), ({∂2 + x}, {∂ − 1})}, i.e. we recover hBi = h∂2 + xi ∩ h∂ − 1i in this case. Remark 1. One can also insert an early termination cri- terion inside Algorithm 2, namely after at least one factor- ized constrained Grobner tuple has been found. This is in the commutative case motivated by the fact that in practice users are often not interested in all the elements in a vari- ety but would be content with at least one. For example, the computer algebra system Reduce can be instructed to stop after finding one factorized Grobner basis (see [16]). In the non-commutative case, we can only hope for partial solutions in general, but a mechanism to stop a computation once at least one is found is also desirable. 5. CONCLUSIONS An algorithm for factoring elements in G-algebras, where the underlying field K has the property that we are able to extract all possible K-roots of any polynomial in K[x], has been shown (Algorithm 1). This algorithm and the FFD-property of G-algebras en- able us to propose a generalization of the factorized Grobner basis algorithm for G-algebras (Algorithm 2). A future work would be to identify improvements to Al- gorithm 1 for practically interesting G-algebras. This has been studied e.g. for partial q-differential, differential and difference operators in [12, 13], where the Zn graded struc- ture resp. a certain embedding has been utilized. In the meantime, we have implemented the unimproved version in the Singular library ncfactor.lib, which will be made Symbolic and Algebraic Computation, pages 194 -- 201. ACM, 2014. [13] M. Giesbrecht, A. Heinle, and V. Levandovskyy. Factoring linear partial differential operators in n variables. Journal of Symbolic Computation, pages -- , 2015. [14] H.-G. Grabe. On factorized grobner bases. In Computer algebra in science and engineering, pages 77 -- 89. World Scientific. Citeseer, 1995. [15] H.-G. Grabe. Triangular systems and factorized Grobner bases. Springer, 1995. [16] A. Hearn. Reduce user's manual, version 3.8 (2004). [17] A. Heinle and V. Levandovskyy. Factorization of Z-homogeneous polynomials in the first (q)-weyl algebra. arXiv preprint arXiv:1302.5674, 2013. [18] A. Kandri-Rody and V. Weispfenning. Non-commutative grobner bases in algebras of solvable type. 9(1):1 -- 26, 1990. [19] V. Levandovskyy. Non-commutative computer algebra for polynomial algebras: Grobner bases, applications and implementation. Doctoral Thesis, Universitat Kaiserslautern, 2005. [20] H. Li. Noncommutative Grobner bases and filtered-graded transfer. Springer, 2002. [21] E. W. Mayr and A. R. Meyer. The complexity of the word problems for commutative semigroups and polynomial ideals. Advances in mathematics, 46(3):305 -- 329, 1982. [22] W. M. Seiler. Involution. The formal theory of differential equations and its applications in computer algebra. Algorithms and Computation in Mathematics 24. Berlin:Springer, 2010. [23] H. Tsai. Algorithms for algebraic analysis. PhD thesis, University of California at Berkeley, 2000. [24] H. Tsai. Weyl closure of a linear differential operator. Journal of Symbolic Computation, 29:747 -- 775, 2000. available shortly. Our implementation identifies beforehand if an improved method is already included in ncfactor.lib for a specific algebra and, if this is the case, re-directs the input there. This modular design allows us to update the function once an improved algorithm is available for a cer- tain G-algebra. The use of the function stays the same after such an update. Another interesting future direction would be to charac- terize further the connection between the solution space of a polynomial system B ⊂ G and the union of the solution spaces of the output of Algorithm 2 when called with B. Especially, it would be interesting to identify properties of G and B, under which both spaces coincide. An implementation of Algorithm 2 would also be of prac- tical interest, which the authors intend to provide in the near future. 6. ACKNOWLEDGMENTS We thank Eugene Zima, Jason P. Bell and Mark Gies- brecht for insightful discussions we had with them. We are grateful to Wolfram Koepf for the information on internals of Reduce. 7. REFERENCES [1] J. Apel. Grobnerbasen in nichtkommutativen Algebren und ihre Anwendung. Dissertation, Universitat Leipzig, 1988. [2] J. P. Bell, A. Heinle, and V. Levandovskyy. On noncommutative finite factorization domains. To Appear in the Transactions of the American Mathematical Society; arXiv preprint arXiv:1410.6178, 2014. [3] B. Buchberger. Introduction to Grobner Bases. In Grobner Bases and Applications, pages 3 -- 31. Berlin: Springer, 1997. [4] J. Bueso, J. G´omez-Torrecillas, and F. Lobillo. Re-filtering and exactness of the Gelfand-Kirillov dimension. Bull. Sci. Math., 125(8):689 -- 715, 2001. [5] J. Bueso, J. G´omez-Torrecillas, and A. Verschoren. Algorithmic methods in non-commutative algebra. Applications to quantum groups. Dordrecht: Kluwer Academic Publishers, 2003. [6] P. M. Cohn. Free ideal rings and localization in general rings, volume 3. Cambridge University Press, 2006. [7] S. R. Czapor. Solving algebraic equations: combining buchberger's algorithm with multivariate factorization. Journal of Symbolic Computation, 7(1):49 -- 53, 1989. [8] S. R. Czapor. Solving algebraic equations via buchberger's algorithm. In Eurocal'87, pages 260 -- 269. Springer, 1989. [9] J. H. Davenport. Looking at a set of equations. Thechnical report, pages 87 -- 06, 1987. [10] W. Decker, G.-M. Greuel, G. Pfister, and H. Schonemann. Singular 4-0-2 -- A computer algebra system for polynomial computations. http://www.singular.uni-kl.de, 2015. [11] J. Dixmier. Enveloping algebras, volume 14. Newnes, 1977. [12] M. Giesbrecht, A. Heinle, and V. Levandovskyy. Factoring linear differential operators in n variables. In Proceedings of the 39th International Symposium on
1502.06902	2	1502	2015-03-14T08:10:32	A Determinantal Inequality for the Geometric Mean with an Application in Diffusion Tensor Imaging	[ "math.RA", "math.ST", "math.ST" ]	We prove that for positive semidefinite matrices $A$ and $B$ the following determinantal inequality holds: \[ \det(I+A\#B)\le \det(I+A^{1/2}B^{1/2}), \] where $A\#B$ is the geometric mean of $A$ and $B$. We apply this inequality to the study of interpolation methods in diffusion tensor imaging.	math.RA	math	A Determinantal Inequality for the Geometric Mean with an Application in Diffusion Tensor Imaging Koenraad M.R. Audenaert Department of Mathematics, Royal Holloway University of London, Egham TW20 0EX, United Kingdom Department of Physics and Astronomy, Ghent University, S9, Krijgslaan 281, B-9000 Ghent, Belgium Abstract We prove that for positive semidefinite matrices A and B the following determinan- tal inequality holds: det(I + A#B) ≤ det(I + A1/2B1/2), where A#B is the geometric mean of A and B. We apply this inequality to the study of interpolation methods in diffusion tensor imaging. 1 Introduction The topic of this paper has arisen in the study of interpolation methods for image processing in diffusion tensor imaging (DTI). DTI is an imaging method used in medical magnetic resonance imaging (MRI) whereby the data to be imaged usually consists of a field of 3×3 statistical covariance matrices D(r) ∈ P, where the points r lie on a 3-dimensional grid. In order to improve the visual quality of the image it is necessary to interpolate and/or extrapolate between neighbouring D-values. As there is no unique way to do so, it is important to have a mathematical framework within which to describe the various methods as well as their measures of goodness. Such a framework has recently been discussed in [3,9], and the present work grew out of this. Email address: [email protected] (Koenraad M.R. Audenaert). Preprint submitted to Elsevier 15 September 2018, 2:17 The interpolation/extrapolation problem is easily formulated as follows. In this context, all covariance matrices are real, symmetric, positive semidefinite 3 × 3 matrices. Let D1 and D2 be two covariance matrices. Construct a path p 7→ D(p), where p ∈ [0, 1] for interpolation or p < 0 or p > 1 for extrapolation, such that D(0) = D1, D(1) = D2 and D(p) ≥ 0, for all p in the interval of interest, and whereby certain quality criteria have to be satisfied. Without going in too much detail, the quality criterion used in [9] is based on the cube root of the determinant of D(p) as this is one of the quantities that provides structural details of the sample being imaged. It is argued by some that interpolated/extrapolated values of this determinant should not be "too large", as this might lead to so-called "swelling" of certain features in the reconstructed image. One further requirement on the path D(p) is that it be the shortest path between D1 and D2, in a sense to be specified. This is easiest to satisfy by defining a suitable metric d(·,·) on P and let the path be a geodesic path. The requirement that all D(p) on the path be positive semidefinite can then be enforced by choosing a metric specific to the curved space P; this excludes the Euclidean metric d(A, B) = A− B2 for example as its geodesics do not stay within P. Still, many choices remain. Probably the most studied P-metric in matrix analysis is the Riemannian metric dR = log(D−1/2 )2 [2, Chapter 6]. Accordingly, this metric has been given due consideration in DTI as well [4]. However, one of its drawbacks is that it is inordinately sensitive to very small eigenvalues of the covariance matrices. In particular, all rank- deficient matrices are infinitely far apart in the Riemannian sense, no matter how close they are in the Euclidean sense. For this and other reasons, other metrics besides the Riemannian one are being studied for DTI. 1 D2D−1/2 1 The metrics studied in the present work are the so-called "Euclidean root met- ric", dH, and the "Procrustes size-and-shape metric" dS; we will not explain this nomenclature here. Both metrics are based on a reparameterisation of the covariance matrices to enforce positive semidefiniteness, in that they give the Euclidean distance between certain square roots of the covariance matrices, S1 − S22. Here, a square root of a positive semidefinite matrix D is any matrix S for which D = S ∗S. Taking for example the Cholesky decomposi- tion for S (S being upper triangular with positive diagonal entries) yields the Cholesky metric dC. One gets the Euclidean root metric dH by taking positive square roots; i.e. Si = Qi := D1/2 . i The Procrustes metric is the minimal one in the sense that it minimises S1− S22 over all possible choices of square roots. That is, dS(D1, D2) = min R {Q1 − RQ22 : R unitary}. In other words, we look for that square root of D2 that is closest to the 2 1 + Q2 2 = Tr(Q2 positive square root of D1. This minimisation can easily be done analytically, since Q1 − RQ22 2) − 2 Re Tr RQ2Q1. Hence, the optimal R in the above minimisation is the unitary matrix for which RQ2Q1 = Q2Q1, where X = √X ∗X denotes the modulus of the matrix X. Thus the optimal R is given by R = U ∗ where U is the unitary factor in the polar decomposition of Q2Q1 = UQ2Q1. The geodesics induced by these metrics are obtained by considering linear paths in the square root space, and taking the square to map them back to P. That is, for the Euclidean root metric, DH(p) = pQ1 + (1 − p)Q22 and for the Procrustes metric DS(p) = pQ1 + (1 − p)U ∗Q22. The question that we wanted to answer about these metrics is how the de- terminants of the interpolated values of the various D(p) behave. Our main result shows that for 0 ≤ p ≤ 1 the Procrustes path always produces deter- minants that are smaller or equal to those of the Euclidean root path, i.e. the Procrustes interpolation is less prone to swelling. For the extrapolation (p < 0 or p > 1) numerical calculations confirmed our intuition that there can be no guaranteed ordering between the determinants of the two paths. 2 Main Result The following theorem answers the question posed in the introduction; in fact it is more general than required as it holds for all complex positive semidefinite matrices, of any dimension. Theorem 1 Let Q1 and Q2 be n× n positive semidefinite matrices, Let Q2Q1 have polar decomposition Q2Q1 = UQ2Q1. Then det(Q1 + U ∗Q2) is real and non-negative and det(Q1 + U ∗Q2) ≤ det(Q1 + Q2). Here we have absorbed the interpolation parameter p in Q1 and 1 − p in Q2. For 0 ≤ p ≤ 1 this does not change the signature of Q1 or Q2, nor does it change the unitary factor U. 3 That det(Q1 +U ∗Q2) is real and non-negative is easy to show. Indeed, we have U ∗Q2Q1 = Q2Q1, so that det(Q1 + U ∗Q2) det(Q1) = det(Q2 1 +Q2Q1) >= 0. Dividing by the positive number det(Q1) shows that det(Q1 + U ∗Q2) ≥ 0. The proof of the inequality of this theorem relies on a number of concepts from matrix analysis, which we introduce first in some detail (for the benefit of those readers working in diffusion tensor imaging). In Section 3 we consider the related concepts of majorisation and log-majorisation, presenting their definitions and their most important properties. In Section 4 we consider the matrix generalisation of the geometric mean. Finally, in Section 5 we present the proof of the theorem. 3 Majorisation In this section we consider vectors x = (x1, . . . , xn) and y = (y1, . . . , yn) in Rn. We denote by x↓ the vector consisting of the elements of x, sorted in non-ascending order. Thus, x↓ k is the k-th largest element of x. We will now introduce several relations between x and y that come under the heading of majorisation. The standard work about the theory and applications of majorisation is undoubtedly [7], to which we refer for more details. A more concise treatment can be found in [1, Chapter II]. We say that x weakly majorises y, denoted y ≺w x, if and only if, for all k, the sum of the k largest elements of x dominates the sum of the k largest elements of y: y ≺w x ⇐⇒ k Xi=1 y↓ i ≤ x↓ i , k Xi=1 k = 1, 2, . . . , n. If in addition the sum of all elements of x equals the sum of all elements of y, then we say that x majorises y: y ≺ x ⇐⇒   i=1 y↓ i=1 y↓ Pk Pn i=1 x↓ i , i=1 x↓ i . i ≤ Pk i = Pn k = 1, 2, . . . , n − 1; A closely related concept is log-majorisation, which concerns vectors in Rn +. We say that x weakly log-majorises y, denoted y ≺w,log x if and only if log y ≺w log x. Thus, y ≺w,log x ⇐⇒ k Yi=1 y↓ i ≤ x↓ i , k Yi=1 k = 1, 2, . . . , n. 4 Likewise, x log-majorises y, denoted y ≺log x if and only if log y ≺ log x. Next we discuss which functions preserve the (weak) majorisation ordering. A function Φ : Rn → Rm is called strongly isotone if and only if it preserves weak majorisation: y ≺w x ⇒ Φ(y) ≺w Φ(x). A function is called isotone if and only if y ≺ x ⇒ Φ(y) ≺w Φ(x). We will need the following characterisation of isotony in the case that m = 1 [1, Theorem II.3.14]. Lemma 1 (Schur) A differentiable function Φ : Rn → R is isotone if and only if it satisfies (1) Φ is permutation invariant, (2) for all x ∈ Rn and for all i, j: ∂Φ ∂xj (xi − xj) ∂Φ (x)! ≥ 0. (x) − ∂xi 4 Geometric Mean Recall that the geometric mean of two positive real numbers x and y is given by √xy. As is the case with all means, when one wishes to generalise the geometric mean to two positive semidefinite matrices A and B, one is faced with the usual problem of non-commutativity: there exists an infinite number of expressions involving A and B that reduce to √AB when A and B commute. For example, √AB and √A√B are in general different matrices, but both are equal in the commutative case. To resolve this problem, one has to impose a number of conditions in order to obtain a unique generalisation. Kubo and Ando [6] have developed a very nice theory of matrix means that does exactly that. It is now standard to define the matrix geometric mean as follows: Definition 1 Let A and B be n × n positive definite matrices. Then their geometric mean, denoted A#B, is defined as A#B := A1/2(A−1/2BA−1/2)1/2A1/2. 5 When A and/or B are rank-deficient, the geometric mean is defined via a limiting procedure. While this is not obvious from the definition, the geometric mean is symmetric in its arguments; that is, A#B = B#A. Clearly, for a, b > 0 we have (aA)#(bB) = √ab(A#B). One can show that A#B emerges as the solution of the following optimisation problem: the set of positive semidefinite matrices X for which the block matrix A X X B     is itself positive semidefinite, has a maximum in the positive semidefinite or- dering, and this maximum is exactly A#B. In the present work, we will need two more properties of the geometric mean: Lemma 2 (Monotonicity) Let 0 ≤ A and 0 ≤ B1 ≤ B2. Then A#B1 ≤ A#B2. Lemma 3 Let A, B ≥ 0 and r ≥ 1. If A#B ≤ 1, then Ar#Br ≤ 1. The proof of this last statement can be found in [5], as part of the proof of its Theorem 4.6.9. 5 Proof of the theorem. We start with a lemma relating the eigenvalues of the matrix geometric mean to the eigenvalues of what could be designated as a "naıve" matrix geometric mean. This is actually a special case of a more general inequality due to Matharu and Aujla [8, Theorem 2.10], but we provide a stand-alone proof for the benefit of the reader. Henceforth we will denote the vector of eigenvalues of a matrix X by λ(X). Lemma 4 Let A, B ≥ 0. Then λ(A#B) ≺log λ(√A √B). 6 Proof. Throughout the proof we assume that A is invertible. The general case follows from continuity considerations. We first show that the inequality holds for the largest eigenvalue λ1; that is: λ1(A#B) ≤ λ1(√A √B). (1) Let a = λ1(√A √B). Thus B1/2 ≤ aA−1/2. By monotonicity of the geometric mean, we then have A1/2#B1/2 ≤ √a (A1/2#A−1/2) = √a. Using Lemma 3 with r = 2, we obtain A#B ≤ a. This says that λ1(A#B) ≤ a, as required. To prove the statement of the lemma, we use the so-called "Weyl-trick". Let A∧k be the k-th antisymmetric tensor power of A; this is the restriction of the k-th tensor power of A, A⊗k, to the totally antisymmetric subspace of (Cn)⊗k. The Weyl-trick exploits two facts about these powers. Firstly, for A ≥ 0 the largest eigenvalue of A∧k is given by λ1(A∧k) = λ1(A)λ2(A) · · · λk(A). Secondly, any expression involving products and/or fractional matrix powers "commutes" with taking the k-th antisymmetric tensor power. In particular, (A#B)∧k = A∧k#B∧k. Thus, exploiting (1), we get k Yi=1 λi(A#B) = λ1((A#B)∧k) = λ1(A∧k#B∧k) ≤ λ1(√A∧k√B∧k) = λ1((√A √B)∧k) = λi(√A √B). k Yi=1 This already shows that we have a weak log-majorisation: λ(A#B) ≺w,log λ(√A √B). Because of the Cauchy-Binet theorem, we also have det(A#B) = qdet(A) det(B) = det(√A√B), majorisation relation can be strengthened to a log-majorisation. ✷ i=1 λi(√A √B). Thus, the weak log- which says that Qn Theorem 2 Let A, B ≥ 0. Then i=1 λi(A#B) = Qn det(I + A#B) ≤ det(I + A1/2B1/2). (2) 7 Proof. Note that, for any X ≥ 0, we have log det X = Tr log X = Pn So we equivalently need to show i=1 log λi(X). Tr log(I + A#B) ≤ Tr log(I + A1/2B1/2). (3) By Lemma 4, log λ(A#B) ≺ log λ(A1/2B1/2). This implies (3) if we can show that the function Φ(x) := Pn is isotone. i=1 log(1+exp(xi)) Clearly, this function is permutation symmetric. Furthermore, ∂Φ/∂xi = exp(xi)/(1 + exp(xi)) = 1/(1 + exp(−xi)), which is monotonously increasing in xi. Hence, the second condition for isotony is also satisfied. ✷ Proof of Theorem 1. Let us consider the matrices Q1, Q2 and U given in the statement of the theorem. Thus U ∗Q2Q1 = Q2Q1. We will assume that Q1 is invertible, so det(Q1) 6= 0. The general case follows from continuity of the determinant. Because of the Cauchy-Binet theorem, the statement of the theorem is equiv- alent with det((Q1 + U ∗Q2)Q1) ≤ det(Q1(Q1 + Q2)), which becomes det(Q2 1 + Q2Q1) ≤ det(Q2 1 + Q1Q2). Applying the Cauchy-Binet theorem a second time we can divide out Q2 each side in a well-chosen way, and get the equivalent statement 1 from det(I + Q−1 1 (Q1Q2 2Q1)1/2Q−1 1 ) ≤ det(I + Q−1/2 1 Q2Q−1/2 1 ). Using the substitutions A = Q−2 1 and B = Q2 2, this can be rewritten as det(I + A1/2(A−1/2BA−1/2)1/2A1/2) ≤ det(I + A1/4B1/2A1/4), 8 or, in terms of the geometric mean, det(I + A#B) ≤ det(I + A1/2B1/2). The validity of this inequality is just Theorem 2. ✷ Acknowledgments We acknowledge support by an Odysseus grant from the Flemish FWO. Many thanks to A. Koloydenko for suggesting this problem. References [1] R. Bhatia, Matrix Analysis, Springer, Berlin (1997). [2] R. Bhatia, "Positive semidefinite matrices", Princeton (2007). [3] I.L. Dryden, A.A. Koloydenko and D. Zhou, "Non-Euclidean statistics for covariance matrices, with applications to Diffusion Tensor Imaging", The Annals of Applied Statistics 3(3), 1102 -- 1123 (2009). [4] P.T. Fletcher and S. Joshi, "Riemannian geometry for the statistical analysis of diffusion tensor data", Signal Processing 87, 250 -- 262 (2007). [5] F. Hiai, "Matrix Analysis: Matrix Monotone Functions, Matrix Means, and Majorization", Interdisciplinary Information Sciences 16(2), 139 -- 248, Graduate School of Information Sciences, Tohoku University (2010). [6] F. Kubo and T. Ando, "Means of positive linear operators", Math. Ann. 246, 205 -- 224 (1980). [7] A.W. Marshall, I. Olkin and B.C. Arnold, Inequalities: theory of majorization and its applications, Second Edition, Springer (2011). [8] J.S. Matharu and J.S. Aujla, "Some inequalities for unitarily invariant norm", Linear Algebra Appl. 436(6), 1623 -- 1631 (2012). [9] D. Zhou, I.L. Dryden, A.A. Koloydenko, K.M.R. Audenaert and L. Bai, "Regularisation, interpolation and visualisation of diffusion tensor images using non-Euclidean statistics", submitted (2015). 9
1510.08207	1	1510	2015-10-28T06:55:31	Characterizing some rings of finite order	[ "math.RA" ]	In this paper, we compute the number of distinct centralizers of some classes of finite rings. We then characterize all finite rings with $n$ distinct centralizers for any positive integer $n \leq 5$. Further we give some connections between the number of distinct centralizers of a finite ring and its commutativity degree.	math.RA	math	Characterizing some rings of finite order t c O 8 2 ] . A R h t a m [ 1 v 7 0 2 8 0 . 0 1 5 1 : v i X r a Jutirekha Dutta, Dhiren K. Basnet and Rajat K. Nath∗ Department of Mathematical Sciences, Tezpur University, Napaam-784028, Sonitpur, Assam, India. Emails: [email protected], [email protected] and [email protected]* Abstract In this paper, we compute the number of distinct centralizers of some classes of finite rings. We then characterize all finite rings with n distinct centralizers for any positive integer n ≤ 5. Further we give some connections between the number of distinct centralizers of a finite ring and its commutativity degree. Key words: finite ring, centralizer, commutativity degree. 2010 Mathematics Subject Classification: 16U70. 1 Introduction Finite abelian groups have been completely characterized up to isomorphism for a long time but finite rings have yet to be characterized. The problem of characterizing finite rings up to isomorphism has received considerable attention in recent years (see [2, 8, 9, 11, 12]) starting from the works of Eldridge [10] and Raghavendran [15]. In this paper we characterize finite rings in terms of their number of distinct centralizers. Given a ring R and ∗Corresponding author 1 an element r ∈ R, the subrings C(r) = {s ∈ R : rs = sr} and Z(R) = {s ∈ R : rs = sr for all r ∈ R} are known as centralizer of r in R and center of R respectively. We write Cent(R) to denote the set of all centralizers in R. Firstly we compute the order of Cent(R) for some classes of finite rings R. Motivated by the works of Belcastro and Sherman [3] and Ashrafi [1], we define n-centralizer ring for any positive integer n. A ring R is said to be n-centralizer ring if Cent(R) = n, for any positive integer n. We then characterize n-centralizer finite rings for all n ≤ 5, adapting similar techniques that are used by Belcastro and Sherman [3] in order to characterize n-centralizer finite groups for n ≤ 5. Further, we conclude the paper by noting some interesting connections between d(R) and Cent(R). Note that for any finite ring R, the ratio d(R) = C(r) is the probability that a randomly chosen pair of elements of R 1 R2 Pr∈R commute. This ratio is known as commutativity degree of finite ring R and it was introduced by MacHale [13] in the year 1975. Some characterizations of finite rings in terms of commutativity degree can be found in [13, 5, 6]. Throughout the paper R denotes a finite ring. For any subring S of R, R/S denotes the additive quotient group and R : S denotes the index of the additive subgroup S in the additive group R. Note that the isomorphisms considered are the additive group isomorphisms. Also for any two non-empty subsets A and B of a ring R, we write A + B = {a + b : a ∈ A, b ∈ B}. We shall use the fact that for any non-commutative ring R, the additive group Z(R) is not a cyclic group (see [13, Lemma 1]). R 2 Some computations of Cent(R) In this section, we compute Cent(R) for some classes of finite rings. How- ever, first we prove some results which are useful for subsequent results as well as for the next sections. Proposition 2.1. R is a commutative ring if and only if R is a 1-centralizer ring. Proof. The proposition follows from the fact that a ring R is commutative if and only if C(r) = R for each r ∈ R. Proposition 2.2. Let R, S be two rings, then Cent(R × S) = Cent(R) × Cent(S). 2 Proof. It can be easily seen that C((r, s)) = C(r) × C(s) for any r ∈ R and s ∈ S. This proves the proposition. The following lemmas play an important role in finding lower bound of Cent(R) for any non-commutative ring R. Lemma 2.3. Let R be a ring. Then Z(R) is the intersection of all central- izers in R. Proof. It is clear that Z(R) ⊆ ∩ r∈R C(r). Now, for any s ∈ ∩ r∈R C(r) we have rs = sr for all r ∈ R. Therefore s ∈ Z(R). Hence the lemma follows. Lemma 2.4. If R is a ring, then R is the union of centralizers of all non- central elements of R. Proof. It is clear that ∪ r∈R−Z(R) C(r) ⊆ R. Again, for any s ∈ Z(R), we have by Lemma 2.3, s ∈ C(r) for all r ∈ R. So s ∈ ∪ C(r). Also for any s ∈ R − Z(R), we have s ∈ C(s) and so s ∈ follows. r∈R−Z(R) ∪ r∈R−Z(R) C(r). Hence the lemma Lemma 2.5. A ring R cannot be written as a union of two of its proper subrings. Proof. The lemma follows from the well-known fact that a group can not be written as a union of two of its proper subgroups. Theorem 2.6. For any non-commutative ring R, Cent(R) ≥ 4. Proof. Since R is non-commutative, so Cent(R) ≥ 2. If Cent(R) = 2, then, by Lemma 2.4, R is equal to a proper subset of itself, which is not possible. Also by Lemma 2.5, Cent(R) 6= 3. Hence the theorem follows. Note that the ring R = (cid:26)(cid:20)0 0 0 0(cid:21) ,(cid:20)1 0 1 0(cid:21) ,(cid:20)0 1 0 1(cid:21) ,(cid:20)1 1 1 1(cid:21)(cid:27), where 0, 1 ∈ Z2, has 4 distinct centralizers. So the above result is the best one possible. At this point, the following question, similar to the question posed by Belcastro and Sherman [3, p. 371], arises naturally. Question 2.7. Does there exist an n-centralizer ring for any positive integer n 6= 2, 3? Can we characterize an n-centralizer ring? 3 The following results show the existence of n-centralizer rings for some values of n. Proposition 2.8. There exists a (p + 2)-centralizer ring for any prime p. 0 0(cid:21) : a, b ∈ Zp(cid:27). For any element Proof. We consider the ring R = (cid:26)(cid:20)a b 0 0(cid:21)(cid:19) we have xb − ay = 0. 0 0(cid:21) of C(cid:18)(cid:20)a b (cid:20)x y 0 0(cid:21)(cid:19) = R. Using simple calculations, we have for any Clearly, C(cid:18)(cid:20)0 0 C(cid:18)(cid:20)a 0 C(cid:18)(cid:20)la a 0(cid:21) : x ∈ Zp(cid:27). Hence Cent(R) = p + 2. 0 0(cid:21) : x ∈ Zp(cid:27) and a 6= 0 and l ∈ Zp, 0 0(cid:21)(cid:19) = (cid:26)(cid:20)x 0 0(cid:21)(cid:19) = (cid:26)(cid:20)lx x 0 0 The above proposition is a particular case of the following theorem. Theorem 2.9. Let R be a non-commutative ring of order p2, where p is a prime. Then Cent(R) = p + 2. Proof. For any x ∈ R − Z(R), we consider C(x). As C(x) is an additive subgroup of R we have C(x) = 1, p or p2. Clearly, C(x) 6= 1, p2, as x, 0R ∈ C(x) and R is non-commutative, where 0R is the additive identity in R. Hence C(x) is additive cyclic group of order p and so Z(R) = {0R}. Let x, y ∈ R − Z(R). If there exists an element t(6= 0R) ∈ C(x) ∩ C(y) then C(x) = C(y), as C(x), C(y) are additive cyclic groups of order p. Thus for any x, y ∈ R − Z(R) we have either C(x) ∩ C(y) = {0R} or C(x) = C(y). Therefore the number of centralizers of non-central elements is R − Z(R) p − 1 = p2 − 1 p − 1 = p + 1. Hence Cent(R) = p + 2. Theorem 2.10. Let p be a prime number and R be a non-commutative ring of order p3 with unity. Then Cent(R) = p + 2. Proof. Let x be an arbitrary element of R − Z(R). Then C(x) is an additive subgroup of R and so C(x) = 1, p, p2 or p3. Here C(x) 6= 1, p3 as x, 0R ∈ C(x), where 0R is the additive identity in R and R is non-commutative. If C(x) = p then Z(R) = 1, which is not possible as 0R, 1R ∈ Z(R). So C(x) = p2 and this gives Z(R) = p. 4 Now, we suppose that y ∈ R − Z(R) and y ∈ C(x). Let z ∈ C(x) be an arbitrary element. We know that Z(R) ⊂ Z(C(x)) and so Z(C(x)) > 1, therefore by Lemma 3 of [14], C(x) is commutative. Therefore z ∈ C(y), as y ∈ C(x). So C(x) ⊆ C(y). Also C(x) = C(y). Hence, C(x) = C(y); and if y /∈ C(x) then C(x) ∩ C(y) = Z(R). Therefore the number of centralizers of non-central elements of R is Cent(R) = p + 2. R − Z(R) C(x) − Z(R) = p3 − p p2 − p = p + 1. Thus As an application of the above theorem, it follows that the ring R = 0 c(cid:21) a, b, c ∈ Zp(cid:27) having order p3 is a (p + 2)-centralizer ring. The (cid:26)(cid:20)a b following theorem, which is generalization of Theorem 2.9, gives another class of (p + 2)-centralizer rings . Theorem 2.11. Let R be a ring and R Z(R) Then Cent(R) = p + 2. ∼= Zp × Zp, where p is a prime. Proof. We write Z := Z(R). Since R/Z ∼= Zp × Zp we have R Z = hZ + a, Z + b : p(Z + a) = p(Z + b) = Z; a, b ∈ Ri. If S/Z is additive non-trivial subgroup of R/Z then S/Z = p. Therefore any additive proper subgroup of R properly containing Z has p disjoint right cosets. Hence the proper additive subgroups of R properly containing Z are Sm = Z ∪ (Z + (a + mb)) ∪ (Z + 2(a + mb)) ∪ · · · ∪ (Z + (p − 1)(a + mb)), where 1 ≤ m ≤ (p − 1), Sp = Z ∪ (Z + a) ∪ (Z + 2a) ∪ · · · ∪ (Z + (p − 1)a) and Sp+1 = Z ∪ (Z + b) ∪ (Z + 2b) ∪ · · · ∪ (Z + (p − 1)b). Now for any x ∈ R − Z, we have Z + x is equal to Z + k for some k ∈ {ma, mb, a+ mb, 2(a+ mb), . . . , (p −1)(a+ mb) : 1 ≤ m ≤ (p −1)}. Therefore C(x) = C(k). Again, let y ∈ Sj − Z for some j ∈ {1, 2, . . . , (p + 1)}, then C(y) 6= Sq, where 1 ≤ q(6= j) ≤ (p + 1). Thus C(y) = Sj. Hence Cent(R) = p + 2. Further, we have the following theorem analogous to Lemma 2.7 of [1]. 5 Theorem 2.12. Let R be a non-commutative ring whose order is a power of a prime p. Then Cent(R) ≥ p + 2, and equality holds if and only if R Z(R) ∼= Zp × Zp. Proof. Let R be a non-commutative ring whose order is a power of a prime p. Suppose k = Cent(R). Let A1, . . . , Ak be the distinct centralizers of R such that A1 ≥ · · · ≥ Ak and A1 = R. So R = k ∪ i=2 Ai and by Cohn’s theorem in [7], we have R ≤ where i 6= 1. Hence k Pi=3 Ai (as Ai’s are additive groups). Also Ai ≤ R p , R ≤ + · · · + R p (k−2)−times {z R p } For the equality, which implies R ≤ (k − 2) R p and so k ≥ p + 2. That is Cent(R) ≥ p + 2. ∼= Zp × Zp then by Theorem 2.11, we have Cent(R) = p + 2. Conversely, we assume that l = Cent(R) = p + 2. Suppose A1, A2, . . . , Al are distinct centralizers of R such that A1 ≥ · · · ≥ Z(R) if R Al and A1 = R. So R = l ∪ i=2 Ai and by Cohn’s theorem in [7], we have R ≤ Ai. Also Ai ≤ R p , where i 6= 1. Suppose, there exists an Ai such l Pi=3 that Ai < R p for 3 ≤ i ≤ l then R < + · · · + = (l − 2) R p R p = R, R p (l−2)−times } {z p , . . . , Al = R a contradiction. Hence A3 = R Ai = R p , where 2 ≤ i ≤ l. Hence p . Also A2 ≥ · · · ≥ Al, so Ai = (l − 2) R p = R. Therefore l Pi=3 Ai = R if and only if A2 + Am = R, for all m 6= 2 and Ak ∩ Al ⊆ A2 for all k 6= l (By Cohn’s Theorem in [7]). Interchanging Ai’s we have A2 ∩ A3 = Z(R). Thus l Pi=3 R = A2 + A3 = A2A3 A2 ∩ A3 = R2 p2Z(R) 6 which gives R : Z(R) = p2. Hence commutative. This completes the proof. R Z(R) ∼= Zp × Zp, since R is non- We conclude this section by the following result. Proposition 2.13. There exists an 8-centralizer ring. Proof. We consider the ring R = {a + bi + cj + dk : a, b, c, d ∈ Zp, i2 = j2 = k2 = −1, ij = k, jk = i, ki = j, ji = −k, kj = −i, ik = −j}. If b = c = d = 0, then clearly C(a) = R. If c = d = 0 and b 6= 0, then C(a + bi) = {x + yi : x, y ∈ Zp}. If b = d = 0 and c 6= 0, then C(a + cj) = {x + zj : x, z ∈ Zp}. If b = c = 0 and d 6= 0, then C(a + dk) = {x + wk : x, w ∈ Zp}. If d = 0 and b, c 6= 0, then C(a+bi+cj) = {x+yi+zj : bz = cy, x, y, z ∈ Zp}. If c = 0 and b, d 6= 0, then C(a + bi + dk) = {x + yi + wk : bw = dy, x, y, w ∈ Zp}. If b = 0 and c, d 6= 0, then C(a + cj + dk) = {x + zj + wk : cw = dz, x, z, w ∈ Zp}. If b, c, d 6= 0, then C(a + bi + cj + dk) = {x + yi + zj + wk : bz = cy, dy = bw, dz = cw, x, y, z, w ∈ Zp}. Hence Cent(R) = 8. 3 4-centralizer rings In this section, we give a characterization of finite 4-centralizer rings analo- gous to Theorem 2 of [3]. The following lemma which is useful in character- ization of 4-centralizer rings. Lemma 3.1. Let R be a 4-centralizer finite ring. Then at least one of the centralizers of non-central elements has index 2 in R. Proof. Let A, B, C be the three proper centralizers of R. Suppose none of A, B, C has index 2, that is R : A ≥ 3, R : B ≥ 3, R : C ≥ 3. Then as R = A ∪ B ∪ C, we have R ≤ A + B + C − 2Z(R) ≤ R 3 + R 3 + R 3 − 2Z(R) < R, which is a contradiction. Hence the lemma follows. We have the following characterization of finite 4-centralizer rings. Theorem 3.2. Let R be a non-commutative finite ring. Then Cent(R) = 4 if and only if R Z(R) ∼= Z2 × Z2. 7 Proof. If R Z(R) ∼= Z2 × Z2 then by Theorem 2.11, we have Cent(R) = 4. Conversely, let Cent(R) = 4 then R has exactly four distinct centraliz- ers, say R, A, B, C where A, B, C are centralizers of three distinct non-central elements of R. By Lemma 2.5, R cannot be written as the union of two of its proper subrings of R. Therefore we may choose a ∈ A−(B ∪C), b ∈ B −(C ∪A), c ∈ C − (A ∪ B) respectively. It can be easily seen that C(a) = A, C(b) = B, C(c) = C. By Lemma 3.1, at least one of the centralizers A, B, C, say A has index 2 in R, that is R : A = 2. Now, let x ∈ (A∩B)−Z(R) then C(x) 6= R. If C(x) = A then a, b ∈ C(x). So, C(x) 6= A. Similarly it can be seen that C(x) 6= B. If C(x) = C then x ∈ A∩B∩C = Z(R) (using Lemma 2.3), which is a contradiction. Therefore Cent(R) must be at least 5, which is again a contradiction. So A ∩ B = A∩B∩C = Z(R). Similarly it can be seen that B∩C = Z(R), A∩C = Z(R). Again A, B, C are additive subgroups of R, therefore R ≥ A + B = AB A ∩ B = AB Z(R) which gives B ≤ 2Z(R). Since Z(R) ⊂ B, so B B = 2Z(R). Similarly C = 2Z(R). Therefore 2 ≤ Z(R) < B. Hence R = A + B + C − 2Z(R) = R 2 + 2Z(R) which gives R : Z(R) = 4 and hence R Z(R) ∼= Z2 × Z2. 4 5-centralizer rings In this section, we give a characterization of finite 5-centralizer rings analo- gous to Theorem 4 of [3]. The following lemmas are useful in this regard. Lemma 4.1. Let R be a ring and R = A∪B∪C, where A, B, C are the proper distinct subrings. We put K = A ∩ B ∩ C, L = A ∩ B − K, M = A ∩ C − K, N = B ∩ C − K and A′ = A − (B ∪ C), B′ = B − (A ∪ C), C ′ = C − (A ∪ B). Then (a) L = M = N = φ, (b) A′ + B′ ⊆ C ′, B′ + C ′ ⊆ A′ and C ′ + A′ ⊆ B′, 8 (c) A′ + A′ ⊆ K, B′ + B′ ⊆ K and C ′ + C ′ ⊆ K, (d) R : K = 4. Proof. (a) We consider l ∈ L and c′ ∈ C ′. Then c′ + l ∈ A or B or C. If c′ + l ∈ A then c′ + l + (−l) = c′ ∈ A, a contradiction. If c′ + l ∈ B then c′ +l +(−l) = c′ ∈ B, a contradiction. If c′ +l ∈ C then (−c′)+c′ +l = l ∈ C, a contradiction. Since C ′ 6= φ, we must have L = φ. Similarly M = N = φ. (b) Let a′ ∈ A′, then a′ ∈ A ⇒ −a′ ∈ A ⇒ −a′ ∈ K or A′. If −a′ ∈ K then a′ ∈ K, a contradiction. Hence −a′ ∈ A′. Similarly if b′ ∈ B′ then −b′ ∈ B′ and if c′ ∈ C ′ then −c′ ∈ C ′. Suppose a′ ∈ A′, b′ ∈ B′ then a′ + b′ ∈ K or A′ or B′ or C ′. If a′ + b′ ∈ A′ ⊆ A then b′ = −a′ + a′ + b′ ∈ A, If a′ + b′ ∈ B′ ⊆ B, then a′ = a′ + b′ + (−b′) ∈ B, a a contradiction. If a′ + b′ ∈ K, then a′ + b′ ∈ A, a contradiction. Hence contradiction. a′ + b′ ∈ C ′. Thus A′ + B′ ⊆ C ′. Similarly it can be seen that B′ + C ′ ⊆ A′ and C ′ + A′ ⊆ B′. (c) Let a′, a1 ′ ∈ A′ ⊆ A. So a′ + a1 ′ ∈ A′. We consider b′ + a′ + a1 a′ + a1 we have b′ + (a′ + a1 a contradiction. Similarly we can show the other two. ′) ∈ C ′ and (b′ + a′) + a1 ′ ∈ A ⇒ a′ + a1 ′ ∈ A′ or K. Let ′, for some b′ ∈ B′. Then by second part ′ ∈ B′ ∩ C ′, ′ ∈ B′. So b′ + a′ + a1 (d) From part (a), we have R = K ∪ A′ ∪ B′ ∪ C ′. Let k + a′ ∈ K + a′ where k ∈ K, a′ ∈ A′ then k + a′ ∈ A = K ∪ A′. If k + a′ ∈ K then a′ ∈ K, a contradiction. So K + a′ ⊆ A′. Again x′ ∈ A′ gives x′ + (−a′) ∈ K (by part (c)). So, x′ ∈ K + a′. Hence K + a′ = A′. Similarly it can be seen that K + b′ = B′, K + c′ = C ′, where b′ ∈ B′, c′ ∈ C ′. Therefore R : K = 4. Lemma 4.2. Let R be a 5-centralizer finite ring and A, B, C, D be the four proper centralizers of R. Then (a) R = A + B + C + D − 3Z(R). (b) If S and T are distinct proper centralizers of R, then ST R ≤ Z(R) ≤ R 6 . Proof. Let a ∈ A − (B ∪ C), b ∈ B − (A ∪ C) and c ∈ C − (A ∪ B). Suppose there does not exist any a ∈ A − (B ∪ C) such that C(a) = A. Then C(a) = D for all a ∈ A − (B ∪ C). Therefore A − (B ∪ C) ⊆ D − (B ∪ C). Interchanging the roles of A and D we get A − (B ∪ C) = D − (B ∪ C), 9 which gives A ∪ B ∪ C = D ∪ B ∪ C = R. Again, by Lemma 4.1 (a), we have B ∩ C = C ∩ D and so Z(R) = A ∩ B ∩ C. Therefore, by Lemma 4.1 (d), we have R/Z(R) ∼= Z2 × Z2. This gives Cent(R) = 4, contradiction. Hence C(a) = A. Similarly C(b) = B and C(c) = C. (a) Let us assume without loss of generality that D is a subset of A∪B∪C. Then R = A∪B∪C∪D = A∪B∪C. Now, by Lemma 4.1, we have R : K = 4 where K = A ∩ B ∩ C = Z(R). Thus by Theorem 3.2, Cent(R) = 4, which is a contradiction. Therefore no one of A, B, C or D is contained in the union of the other three. Let r ∈ (A ∩ B) − (C ∪ D) then r ∈ C(a) ∩ C(b) which gives a, b ∈ C(r). But a /∈ C(b), so C(r) 6= A, B. Again r /∈ C, D; so C(r) 6= C, D. Also C(r) 6= R, since r ∈ R − Z(R). Therefore Cent(R) must be at least 6, a contradiction. Hence (A ∩ B) − (C ∪ D) = φ. This shows that no element of R is in exactly two proper centralizers. Let r ∈ (A ∩ B ∩ C) − D then r ∈ C(a) ∩ C(b) ∩ C(c). Therefore a, b, c ∈ C(r). But b /∈ C(a), c /∈ C(b). So C(r) 6= A, B, C. Also C(r) 6= D, R; as r /∈ D and r /∈ Z(R). Therefore Cent(R) must be at least 6, a contradiction. Hence A ∩ B ∩ C − D = φ. Thus no element of R is in exactly three proper centralizers. From above, it can be seen clearly that R = A ∪ B ∪ C ∪ D = A + B + C + D − 3Z(R). (b) Note that for any two proper centralizers S and T of R we have S ∩ T = Z(R), since no element of R is in exactly two as well as three proper centralizers. Also any proper centralizers of R are additive subgroups of R, so ST S+T = S ∩ T = Z(R). Since S + T ⊆ R we have Z(R) ≥ ST R . Again by part (a), R = A + B + C + D − 3Z(R) ≥ 2Z(R) + 2Z(R) + 2Z(R) + 2Z(R) − 3Z(R). Thus R : Z(R) ≥ 5. If R : Z(R) = 5 then R Z(R) Therefore Z(R) ≤ R R ≤ Z(R) ≤ R 6 . 6 . So, ST ∼= Z5, a contradiction. We would like to mention here that the group theoretic analogues of Lemma 4.1 and Lemma 4.2 can be found in [4] and [3] respectively. Now we prove the main theorem of this section which characterizes finite 5-centralizer rings. 10 Theorem 4.3. Let R be a finite ring. Then Cent(R) = 5 if and only if R Z(R) ∼= Z3 × Z3. Proof. Let R Z(R) ∼= Z3 × Z3, then by Theorem 2.11, we get Cent(R) = 5. R ≤ Z(R) ≤ R Conversely, let Cent(R) = 5. Let A, B, C, D be the four proper cen- tralizers of R. Then by Lemma 4.2 (b), AB 6 . Our aim is to get more near lower bound for Z(R). We may assume without loss of 3 , as 1 < A ≤ R generality that A ≥ B ≥ C ≥ D. Suppose A < R 2 . That is A ≤ R 4 . Now by Lemma 4.2 (a), R ≤ R − 3Z(R) < R, a contradiction. Hence A = R 2 , then R = A + B + C + D − 3Z(R) gives R 6 < B. Also, applying Lemma 4.2 (b) on A and B we have R 3 . So B is one of R 2 < B + C + D and so R 6 < B ≤ R 5 . Reapplying Lemma 4.2 (b) on A and B we have, 2 or A = R 3 . If A = R 4 or R 3 , R AB R ≤ Z(R) ≤ R 6 10 ≤ Z(R) ≤ R 9 7 , R which gives R R 10 . Let Z(R) = R Z(R) = R then R Z(R) 6 divides B. If B = R B = R So C+D ≤ R 2 < 5R divides B. If B = R B = R therefore C + D ≤ 2R 7 , R 3 , R 8 , R 6 , R 6 . Thus Z(R) is one of R ∼= Z6, a contradiction. Let Z(R) = R 9 or then 2 divides 7 and 9, which is not possible. If then R 8 then 3, 5 divides 8, a contradiction. Therefore 8 = C+D. Also B ≥ C ≥ D. 10 , then R 10 4 then 3, 4 divides 10, a contradiction. Therefore 10 . Also B ≥ C ≥ D, 8 = C+D, a contradiction. If Z(R) = R 3 , R 5 8 5 . Now Lemma 4.2(a) gives, C + D = 6R 4 . By Lemma 4.2 (a), we have 5R 5 < 6R 10 = C + D, a contradiction. If A = R 3 then Lemma 4.2 (a) gives, 2R 2R 3 < 3B and so B ≥ R applying Lemma 4.2 (b) on A and B we get, 4 . Also A ≥ B, so B = R 3 < B + C + D which gives 4 . Again, 3 or R AB R ≤ Z(R) ≤ R 6 which gives R R 11 or R 12 ≤ Z(R) ≤ R 6 . Therefore Z(R) is one of R 9 , R 10 , 11 then 3 divides 7, 8, 10, 11, a 6 , R 7 , R 8 , R 12 . Now if Z(R) = R 7 , R 8 , R 10 , R 11 then R Z(R) contradiction. Let Z(R) = R 6 Z(R) = R 9 applying Lemma 4.2 (b) on A and B we have, R B = R 4 so C + D ≤ 3R Z3 × Z3. ∼= Z3 × Z3. Let Z(R) = R 9 ≤ R then Lemma 4.2 (a) gives, C + D = 4R then as above we get a contradiction. Let 12 and B = R then 12 , a contradiction. If 6 . Also C, D ≤ R 4 , ∼= 3 6 < 4R 6 = C + D, which is not possible. Hence R Z(R) 5 Relation between Cent(R) and d(R) Note that d(R) = 1 if and only if R is commutative. Therefore, by Propo- sition 2.1, we have Cent(R) = 1 if and only if d(R) = 1. By Theorem 3.2 and Theorem 1 of [13], we have the following result. Proposition 5.1. Let R be a non-commutative finite ring. Then Cent(R) = 4 if and only if d(R) = 5 8 . In [13], MacHale also proved the following theorem: Theorem 5.2. Let R be a non-commutative finite ring and p the smallest prime dividing the order of R. Then d(R) ≤ 1 p3 (p2 + p − 1), with equality if and only if R : Z(R) = p2. Now by Theorem 2.11 and Theorem 5.2, we have the following interesting connection between d(R) and Cent(R). Proposition 5.3. Let R be a non-commutative finite ring and p the smallest prime dividing the order of R. If d(R) = 1 p3 (p2+p−1) then Cent(R) = p+2. We conclude the paper by noting that the converse of Proposition 5.3 holds for some finite non-commutative rings. In particular, by Theorem 2.12 and Theorem 5.2, we have the following result. Proposition 5.4. Let R be a non-commutative ring whose order is a power of a prime p. If Cent(R) = p + 2 then d(R) = 1 p3 (p2 + p − 1). References [1] A. R. Ashrafi, On finite groups with a given number of centralizers, Al- gebra Colloq., 7 (2) (2000), 139–146. 12 [2] M. Behboodi, R. Beyranvand, A. Hashemi and H. Khabazian, Classifica- tion of finite rings: theory and algorithm, Czechoslovak Math. J., 64(3) (2014), 641–658. [3] S. M. Belcastro and G. J. Sherman, Counting centralizers in finite groups, Math. Magazine, 67 (5) (1994), 366–374. [4] M. Bruckheimer, A. C. Bryan and A. Muir, Groups which are the union of three subgroups, Amer. Math. Monthly, 77 (1970), 52-57. [5] S. M. Buckley, D. MacHale and A. N´i Sh´e, Finite rings with many com- muting pairs of elements, Preprint. [6] S. M. Buckley and D. MacHale, Contrasting the commuting probabilities of groups and rings, Preprint. [7] J. H. E. Cohn, On n-sum groups, Math. Scand., 75 (1994), 44–58. [8] C. J. Chikunji, A Classification of a certain class of completely pri- mary finite rings, Ring and Module Theory, Trends in Mathematics 2010, Springer Basel, pp 83–90. [9] J. B. Derr, G. F. Orr and P. S. Peck, Noncommutative rings of order p4, J. Pure Appl. Algebra, 97(2) (1994), 109–116. [10] K. E. Eldridge, Orders for finite noncommutative rings with unity, Amer. Math. Monthly, 75 (1968), 512–514. [11] B. Fine, Classification of finite rings of order p2, Math. Magazine, 66(4) (1993), 248–252. [12] R.W. Goldbach and H.L. Claasen, Classification of not commutative rings with identity of order dividing p4, Indag. Math., 6 (1995), 167–187. [13] D. MacHale, Commutativity in finite rings, Amer. Math. Monthly, 83(1976), 30–32. [14] G. R. Omidi and E. Vatandoost, On the commuting Graph of rings, J. Algebra and Appl., 10 3(2011), 521–527. [15] R. Raghavendran, A class of finite rings, Compositio Math., 22 (1970), 49–57. 13
1107.4819	2	1107	2016-02-02T20:51:03	Bisimple monogenic orthodox semigroups	[ "math.RA" ]	We give a complete description of the structure of all bisimple orthodox semigroups generated by two mutually inverse elements.	math.RA	math	Semigroup Forum (2009) 78, 310 -- 325 DOI 10.1007/s00233-008-9101-5 arXiv version: fonts, pagination and layout differ from the SF published version; moreover, the published version contains two typos which have been corrected here -- in the paragraph following Remark 3, it is now correctly written that (ab)an = an and bn(ab) = bn, and in the paragraph following Lemma 2.6 the word 'bisimple' has been inserted on line 4 between the words 'any' and 'orthodox'. 6 1 0 2 b e F 2 ] . A R h t a m [ BISIMPLE MONOGENIC ORTHODOX SEMIGROUPS SIMON M. GOBERSTEIN* Abstract. We give a complete description of the structure of all bisimple orthodox semi- groups generated by two mutually inverse elements. 2000 Mathematics Subject Classification: 20M19, 20M10. 2 v 9 1 8 4 . 7 0 1 1 : v i X r a Introduction It is well known that cyclic groups and the bicyclic semigroup are the only bisimple monogenic inverse semigroups. On the other hand, the class of bisimple orthodox semigroups generated by two mutually inverse elements is substantially more diverse. The purpose of this paper is to determine the structure of all semigroups of that class using only some standard facts about orthodox semigroups. Partial results in this direction, based on the study of the lattice of congruences on the free orthodox semigroup generated by a pair of mutually inverse elements, can be found in [5] -- see Section 1 below for more details. Let S be an orthodox semigroup generated by two mutually inverse elements a and b. In Section 2, we determine the structure of all such bisimple semigroups S when a (and therefore b) does not belong to any subgroup of S (see Theorem 2.9, which is the main result of the paper). In Section 3, we describe the structure of S when a and b are group elements of S (in which case, S is necessarily bisimple). In the sequel to this article, we will apply its results to the study of lattice isomorphisms of bisimple orthodox semigroups generated by a pair of mutually inverse elements. *Partially supported by a CSU research grant. Department of Mathematics and Statistics, California State University, Chico, CA 95929-0525, U.S.A. e-mail: [email protected]. 1 2 SIMON M. GOBERSTEIN 1. Preliminaries Let S be a semigroup. We say that x ∈ S is a group element of S if it belongs to some subgroup of S; otherwise x is a nongroup element of S. The set of nongroup elements of S will be denoted by NS, and the set of idempotents of S by ES. If x = yz for some x, y, z ∈ S, according to standard terminology, y is a left and z a right divisor of x. As usual, hXi stands for the subsemigroup of S generated by its subset X, and hxi for the cyclic subsemigroup of S generated by x ∈ S. The order of an element x of S will be denoted by o(x); if x has infinite order, we will write o(x) = ∞. If w = w(x1, . . . , xn) is a word in the alphabet {x1, . . . , xn} ⊆ S, we will say shortly that w is a word in x1, . . . , xn, and if no confusion is likely, w will be identified with its value in S. For any x ∈ S, we define x0 to be the identity of the semigroup S1, so x0y = yx0 = y for all y ∈ S (exceptions of this agreement will occur when S contains a subsemigroup U with an identity e and it will be convenient to put x0 = e for x ∈ U; all such situations will be explicitly identified). Recall that R is a left and L a right congruence on S. These and other standard results concerning Green's relations on S will be used without reference. As shown by Hall [8, Result 9], if U is a regular subsemigroup of S, then using superscripts to distinguish Green's relations on U from those on S, we have KU = KS ∩ (U × U) for K ∈ {L, R, H}. This result will also be applied without mention. If H is the identity relation on S, it is common to say that S is combinatorial, so a regular semigroup is combinatorial if and only if it has no nontrivial subgroups. The bicyclic semigroup B(a, b) is usually defined as a semigroup with identity 1 generated by the two-element set {a, b} and given by one defining relation ab = 1. Disposing of the identity in the definition of B(a, b), one can define it as a semigroup given by the following presentation: B(a, b) = ha, b aba = a, bab = b, a2b = a, ab2 = bi. The structure of B(a, b) is well known -- it is a combinatorial bisimple inverse semigroup, each of its elements has a unique representation in the form bman where m and n are nonnegative integers (and a0 = b0 = ab), the semilattice of idempotents of B(a, b) is a chain: ab > ba > b2a2 > · · · , and for each bman ∈ B(a, b), we have Rbman = {bmal : l ≥ 0} and Lbman = {bkan : k ≥ 0} (see [2, Lemma 1.31 and Theorem 2.53]). These facts will be used below without reference. Let S be a semigroup. Following Clifford [1], we write x ⊥ y to indicate that xyx = x and yxy = y for x, y ∈ S, and the phrase "x ⊥ y in S" will be used to express briefly the fact that x, y ∈ S and x ⊥ y. As usual, the set of all inverses of x ∈ S will be denoted by VS(x), sometimes without the subscript S if no confusion is likely. Thus y ∈ VS(x) precisely when x ⊥ y in S. An orthodox semigroup is a regular semigroup in which the idempotents form a subsemigroup. By [10, Theorem VI.1.1], the following conditions are equivalent for a regular semigroup S: (a) S is orthodox; (b) VS(e) ⊆ ES for all e ∈ ES; (c) VS(b)VS(a) ⊆ VS(ab) for all a, b ∈ S. Thus if S is orthodox and x ⊥ y in S, then x ∈ NS if and only if y ∈ NS, and xn ⊥ yn for all n ∈ N (we denote by N the set of all positive integers), so that o(x) = o(y). Moreover, if S is any semigroup and xy = x2y2 for some x, y ∈ S, it is clear that xy = xnyn for all n ∈ N. These simple facts will be used below without comment. Let S be an orthodox semigroup, and let Y = {(x, y) ∈ S × S : V (x) = V (y)}. Then Y is the smallest inverse semigroup congruence on S [10, Theorem VI.1.12], so that S/Y is the BISIMPLE MONOGENIC ORTHODOX SEMIGROUPS 3 maximum inverse semigroup homomorphic image of S. As noted, for instance, in [7, p. 72], it is easily seen (and well known) that S is combinatorial if and only if S/Y is such. In [4, 5], a semigroup S was termed elementary if S = hA ∪ Bi for two nonempty subsets A and B of S such that a ⊥ b for all a ∈ A and b ∈ B, and an orthodox semigroup S = ha, bi with a ⊥ b was referred to as "an elementary orthodox semigroup on two mutually inverse generators". This notion was introduced as a generalization of (and by analogy with) "ele- mentary generalized groups" studied by Gluskin in [6]. In modern terminology, "elementary generalized groups" are known as "monogenic inverse semigroups" (by definition, an inverse semigroup is monogenic if it is generated by a pair of mutually inverse elements). Since the phrase "an elementary orthodox semigroup on two mutually inverse generators" is too cumbersome, we introduce a different term for such semigroups. Namely, we will say that S is a monogenic orthodox semigroup (by analogy with the inverse semigroup case) if S is an orthodox semigroup such that S = ha, bi for some a, b ∈ S with a ⊥ b. Of course, it is important to require in this definition that S be orthodox in addition to being generated by a pair of mutually inverse elements. By [2, Exercise 2.3.5], a semigroup given by the presentation hp, q pqp = p, qpq = qi is not regular; in fact, {p, q, pq, qp} is its only regular D-class. Moreover, as shown by Clifford (see [4, Example 1.6]), even if a ⊥ b in a regular semigroup S, then ha, bi need not be a regular subsemigroup of S. On the other hand, an elementary subsemigroup of an orthodox semigroup is itself orthodox [4, Proposition 1.1], so if S is orthodox and a ⊥ b in S, then ha, bi is a monogenic orthodox semigroup. In what follows, the phrase "let S = ha, bi be a monogenic orthodox semigroup" will always mean that S is an orthodox semigroup generated by a pair of mutually inverse elements a, b ∈ S. In view of the above, it is natural to ask: If S = hx, yi is a semigroup such that x ⊥ y, under what conditions is S orthodox? The answer follows easily from [5, Theorem 1.1], stating that the free monogenic orthodox semigroup FO has the following presentation: FO = hp, q pn ⊥ qn for all n ∈ Ni; we will also denote this semigroup by FO(p, q) if we wish to name the generators explicitly. Lemma 1.1. Let S = hx, yi be a semigroup such that x ⊥ y. Then (i) S is orthodox if and only if xn ⊥ yn for all n ∈ N; (ii) S is orthodox if xy = x2y2. Proof. (i) If xn ⊥ yn for all n ∈ N, then S is a homomorphic image of FO(x, y), and hence, by [10, Lemma II.4.7], S is orthodox. The converse is obvious. (ii) Suppose that xy = x2y2. Then for any n ≥ 1, xnynxn = (xnyn)xn = (xy)xn = xn and (cid:3) ynxnyn = yn(xnyn) = yn(xy) = yn. Therefore, by (i), S is orthodox. In [5, Section II], using the description of FO from [5, Section I] and a number of re- sults from [3] and a few other papers, Eberhart and Williams analyzed the lattice of con- gruences on FO and constructed several interesting monogenic orthodox semigroups. In particular, in [5, Results 2.3, 2.4] they exhibited the following two infinitely presented bisim- ple monogenic orthodox semigroups: FO/α′ = hp, q pn ⊥ qn (∀ n ∈ N) and pq = p2q2i and FO/σ′ = hp, q pn ⊥ qn (∀ n ∈ N) and pq = p2q2, qp = q2p2i, where α′ and σ′ are certain congru- ences on FO defined in [5] in several steps using the description of congruences on the free 4 SIMON M. GOBERSTEIN monogenic inverse semigroup and the results about the lattice of congruences on FO (the interested reader is referred to [5] for definitions of α′ and σ′; they will not be needed in this paper). It is immediate from Lemma 1.1 that FO/α′ and FO/σ′ can be finitely presented: Corollary 1.2. The semigroups FO/α′ and FO/σ′ given in [5, Results 2.3 and 2.4], re- spectively, have the following finite presentations: FO(p, q)/α′ = h p, q p ⊥ q, pq = p2q2 i and FO(p, q)/σ′ = h p, q p ⊥ q, pq = p2q2, qp = q2p2 i. It is stated without proof in [5] that each element of FO/α′ has a unique representation in the form piqmpnqj, where m, n ≥ 0, i, j ∈ {0, 1}, m ≥ i, and n ≥ j (the uniqueness assertion here is false since, for instance, pq = p1q1p0q0 = p0q0p1q1). Note also that the eggbox picture of FO/α′ given in [5, Figure 2] contains a number of misprints (among other things, Rq is typed there twice). After correcting these inaccuracies, one could, in principle, obtain a description of all bisimple monogenic orthodox semigroups by using [5, Results 2.3, 2.4] together with several other results about the lattice of congruences on FO stated in [5, Section II] (mostly without proofs). Nevertheless [5] contains no such description, and at any rate it is certainly of interest to get it independently of all the theorems on the structure of the lattice of congruences on FO used in [5]. The main goal of this paper is to obtain such a description directly, using only some basic results about orthodox semigroups. The following simple fact is probably well-known. We record it for convenience of reference and include its proof for completeness. Lemma 1.3. Let S be an orthodox semigroup, a an arbitrary element of S, and b ∈ V (a). (i) The following conditions are equivalent: (a) a2 is a left divisor of a; (b) a = a2(b2a); (c) ab = a2b2; (d) b = (ba2)b2; (e) b2 is a right divisor of b. (ii) If ab = a2b2 and ba = b2a2, then Ha and Hb are groups with identity elements ab2a and ba2b, respectively. Conversely, if Ha (and thus Hb) is a group, then ab = a2b2 and ba = b2a2. Proof. (i) Suppose a2 is a left divisor of a. Then a = a2x for some x ∈ S, and hence a2(b2a) = (a2b2a2)x = a2x = a. Thus (a) implies (b). Since the converse is obvious, (a)⇔(b). By symmetry, (d)⇔(e), and the equivalences (b)⇔(c) and (c)⇔(d) hold trivially. (ii) Suppose ab = a2b2 and ba = b2a2. Then (ab2a)a = a = a(ab2a) whence (a, ab2a) ∈ R and (a, ab2a) ∈ L, that is, (a, ab2a) ∈ H. Since (ab2a)2 = a(b2a2b2)a = ab2a, by Green's Theorem [2, Theorem 2.16], Ha is a group and ab2a is its identity element. By symmetry, Hb is a group with identity ba2b. Conversely, assume that Ha is a group. Since b ⊥ a and S is orthodox, Hb is also a group (cid:3) and b2 ⊥ a2. Thus, by [2, Lemmas 2.12 and 2.15], a2b2 = ab and b2a2 = ba. Lemma 1.4. Let S be an orthodox semigroup and a ⊥ b in S. The following conditions are equivalent: (a) ab = a2b2 and ba 6= b2a2; (b) ab = a2b2 and a ∈ NS; (c) a2 is a left divisor of a and a ∈ NS; (d) a2 is a left divisor of a and ba 6= b2a2. If one (and hence each) of these conditions holds, then o(a) = o(b) = ∞, hai ∪ haib ⊆ Ra, and hbi ∪ ahbi ⊆ Lb. Proof. It is immediate from Lemma 1.3 that conditions (a), (b), (c), and (d) are equiva- lent. Suppose (c) holds. Then a = a2x for some x ∈ S. If o(a) < ∞, there is m ∈ N such that BISIMPLE MONOGENIC ORTHODOX SEMIGROUPS 5 am ∈ ES and, by putting e = am, we get a = a2x = a3x2 = · · · = amxm−1 = exm−1 whence a = ea = am+1, which shows that hai is a group; a contradiction. Therefore o(a) = ∞ = o(b). Note that a = a2x implies aRa2, and so aRan for any n ∈ N. Moreover, anbRan for all n ∈ N since (anb)a = an. Hence hai ∪ haib ⊆ Ra and, by symmetry, hbi ∪ ahbi ⊆ Lb. (cid:3) The equivalence of conditions (a) -- (d) of Lemma 1.4 will be used below without mention. Lemma 1.5. Let S be an orthodox semigroup, a ∈ NS, b ∈ V (a), and ab = a2b2. Then (i) the subsemigroup hba2, b2ai of S is bicyclic and ba is its identity element, that is, hba2, b2ai = B(ba2, b2a); (ii) (ba2)k = bak+1 and (b2a)k = bk+1a for all k ≥ 0, so each element of hba2, b2ai has a unique representation in the form bm+1an+1 for some nonnegative integers m and n; (iii) if k, l, m, and n are nonnegative integers such that k + l ≥ 1 and m + n ≥ 1, then bkalRbman if and only if k = m, and bkalLbman if and only if l = n. Proof. (i) This is immediate since (ba2)(b2a) = ba, (b2a)(ba2) = b2a2 6= ba, and ba is a two-sided identity for ba2 and b2a. (ii) Take an arbitrary nonnegative integer k. If k ≥ 1, then (ba2)k = ba(aba)k−1a = bak+1 and (b2a)k = b(bab)k−1ba = bk+1a. Since ba is the identity element of B(ba2, b2a), we define (ba2)0 = (b2a)0 = ba, so the assertion holds for k = 0 as well. Therefore each element of hba2, b2ai can be uniquely written as (b2a)m(ba2)n, and thus as (bm+1a)(ban+1) = bm+1an+1 for some nonnegative integers m and n. (iii) Let k, l, m, and n be nonnegative integers such that k + l ≥ 1 and m + n ≥ 1. If k, l, m, n ≥ 1, then bkal, bman ∈ B(ba2, b2a) and therefore bkalRbman if and only if k = m, and bkalLbman if and only if l = n. Suppose l = 0 (and hence k ≥ 1). Let bkRbman. If m, n ≥ 1, using the fact that bkRbka, we get bkaRbman, so that k = m. If n = 0 (and thus m ≥ 1), then bkRbm and so, assuming without loss of generality that k ≤ m, we obtain b = b(akbk)Rb(akbk)bm−k = bm−k+1 whence m = k in view of Green's Theorem [2, Theorem 2.16], Lemma 1.4, and the fact that b ∈ NS. Note that we cannot have m = 0 and n ≥ 1, for otherwise bkRan which implies bk+1Rban and thus, as shown above, k = 0; a contradiction. We have proved that bkRbman implies k = m, and if m ≥ 1, it is easily seen that bmRbman for all n ≥ 0, so the converse also holds. Now let bkLbman. If m ≥ 1, then bkaLbman+1 and since bka, bman+1 ∈ B(ba2, b2a), we have n = 0. Here again we cannot have m = 0 and n ≥ 1, for otherwise bkLan and, using the fact that anLban, we obtain bkLban so that, as has just been shown, n = 0; a contradiction. Thus bkLbman implies n = 0, and the converse follows from Lemma 1.4. By symmetry, the result also holds in the case when l ≥ 1 and k = 0. The proof is complete. (cid:3) 2. Bisimple monogenic orthodox semigroups with nongroup generators Let S be an orthodox semigroup, a ∈ NS, b ∈ V (a), and ab = a2b2. Clearly, every x ∈ S can be written in the form x = (ak1bl1) · · · (akr blr ) for some r ∈ N and nonnegative integers ki, li (i = 1, . . . , r) such that ki + li ≥ 1 for all 1 ≤ i ≤ r. We call ak1bl1, . . . , akrblr the syllables 6 SIMON M. GOBERSTEIN of (ak1bl1) · · · (akr blr ) and refer to (ak1bl1) · · · (akr blr ) as an r-syllable word in a, b. If m, n ∈ Z and i, j ∈ {0, 1} are such that either (I) m > i and n = j = 0, or (II) m = i = 0 and n ≥ 1, or (III) m > i and n > j, we say that aibmanbj is an abridged word in a, b (in this order!) of type I, II, or III, respectively (or simply an abridged word in a, b when there is no need to indicate its type), and if x = aibmanbj, we will also call aibmanbj an abridged form of x. Remark 1. Let S = ha, bi be a monogenic orthodox semigroup with ab = a2b2 and ba 6= b2a2. As we will see shortly, each element of S can be written in an abridged form although, in general, not uniquely. Later, by imposing certain additional conditions on S, we will associate with each x ∈ S a unique abridged word for which we reserve the term "reduced form" of x. Observe also that definitions of type I and type II abridged words in a, b are not entirely symmetric -- in a type I word aibm we must have m > i but in a type II word anbj the equality n = j = 1 is allowed. As will be seen later, this ensures that ab ∈ S is always represented by a unique abridged word in a, b (namely, by a type II word a0b0a1b1). Thus the dual of an abridged word b1a1 of type II in b, a is an abridged word a1b1 of type II in a, b; this is the only exception of the following rule -- the dual of an abridged word in b, a of type I [type II, type III] is an abridged word in a, b of type II [type I, type III]. Lemma 2.1. Let S be an orthodox semigroup, a ∈ NS, b ∈ V (a), and ab = a2b2. Then (i) each 1-syllable word in a, b can be written as an abridged word in a, b of type I or II; (ii) the product of two abridged words in a, b of type I [type II] can be written as an abridged word in a, b of type I [type II]; (iii) if x = aibm and y = anbj are abridged words in a, b of types I and II, respectively, then xy equals an abridged word in a, b of type I or type III, while yx can be written as a 1-syllable word in a, b and thus as an abridged word in a, b of type I or type II. Proof. (i) Let x = akbl for some k, l ≥ 0 with k + l ≥ 1. If k = l, then x equals ab, an abridged word in a, b of type II. If k = 0 or l = 0, then x equals bl or ak, an abridged word in a, b of type I or II, respectively. If l > k ≥ 1, then x = akbkbl−k = abl−k+1, an abridged word in a, b of type I, and if k > l ≥ 1, then x = ak−lalbl = ak−l+1b, an abridged word in a, b of type II. (ii) If aibm and ai′bm′ are abridged words in a, b of type I, then aibmai′bm′ = aibm+m′−i′, which is an abridged word in a, b of type I. Similarly, if anbj and an′bj ′ are abridged words in a, b of type II, then anbjan′bj ′ = an+n′−jbj ′, which is an abridged word in a, b of type II. (iii) Let x = aibm and y = anbj be abridged words in a, b of types I and II, respectively. If n = j = 1, then xy = aibmab = aibm, and if n > j, then xy = aibmanbj, which, by definition, is an abridged word in a, b of type III. Finally, yx = anbm if i = j, and yx = an+1−jbm+1−i if i 6= j, so that, by (i), yx can be written as an abridged word in a, b of type I or type II. (cid:3) Lemma 2.2. Let S = ha, bi be a monogenic orthodox semigroup such that a ∈ NS and ab = a2b2. Then for each x ∈ S there is an abridged word aibmanbj in a, b such that x = aibmanbj. Proof. Let x ∈ S, and let r be the smallest positive integer such that x = (ak1bl1) · · · (akrblr) for some r-syllable word (ak1bl1) · · · (akrblr ) in a, b. In view of Lemma 2.1(i), we can assume that akibli is an abridged word in a, b of type I or II for each i = 1, . . . , r. Since x cannot be represented by a q-syllable word for q < r, by Lemma 2.1(ii) adjacent syllables akj blj and BISIMPLE MONOGENIC ORTHODOX SEMIGROUPS 7 akj+1blj+1 (j = 1, . . . , r − 1) are of different types, and by Lemma 2.1(iii) a syllable written as an abridged word in a, b of type II cannot precede a syllable which is an abridged word in a, b of type I. Therefore either r = 1 and ak1bl1 is an abridged word in a, b of type I or II, or r = 2 and ak1bl1ak2bl2 is an abridged word in a, b of type III. (cid:3) 1 · · · xin Remark 2. By [5, Lemma 1.3], each element of FO(a, b) can be written uniquely as a word xi1 n in a, b, where ik > min{ik−1, ik+1} for k = 2, . . . , n−1 (and, of course, xj 6= xj+1 for j = 1, . . . , n − 1). Lemma 2.2 can be proved using that result but when this alternative proof is rigorously written, it is neither shorter nor better than the proof given above. Lemma 2.3. Let S = ha, bi be a monogenic orthodox semigroup such that ab = a2b2 and ba 6= b2a2. Then (i) if aibmanbj is an abridged word in a, b of type III (that is, i, j ∈ {0, 1}, m > i, n > j), then aibmRaibmanbj and aibmanbjLanbj; (ii) S is bisimple and combinatorial. Proof. (i) Suppose aibmanbj is an abridged word in a, b of type III. It is easily seen that aibm = aibmanbjajbn, which implies aibmRaibmanbj. Similarly, since anbj = ambiaibmanbj, we have aibmanbjLanbj. (ii) Let s ∈ S. By Lemma 2.2, s = aibmanbj where aibmanbj is an abridged word in a, b. If aibmanbj is of type I, n = j = 0 and m > i ∈ {0, 1} whence s = aibmLbDa by Lemma 1.4. If aibmanbj is of type II, m = i = 0 and n ≥ 1, so that s = anbjRa again by Lemma 1.4. Finally, if aibmanbj is of type III, by part (i) and Lemma 1.4, aibmanbjLanbjRa. Thus S = Da. Let x = aY and y = bY. Then S/Y = hx, yi, x ⊥ y in S/Y, and xy = x2y2, which implies (xy)y = (xy)(yx)y = (yx)(xy)y = y(x2y2) = y, and, by symmetry, x(xy) = x. Therefore xy is the identity of S/Y. If xy = yx, then (ab)Y = (ba)Y, so that ab ⊥ ba in S whence ba = b(ab)(ba)(ab)a = b2a2; a contradiction. Thus yx 6= xy and, by [2, Lemma 1.31], S/Y = B(x, y). It follows that S is combinatorial. (cid:3) Remark 3. In the proof of Lemma 2.3(ii), it was shown that S/Y = B(x, y), and we could have deduced from this that S is bisimple by applying, for instance, [9, Theorem 10]. However, a simple direct proof of that fact given above seems to be preferable. Let S = ha, bi be a monogenic orthodox semigroup such that ab = a2b2 and ba 6= b2a2. If n ∈ N, it is clear that (ab)an = an and bn(ab) = bn but for the products an(ab) and (ab)bn in S there are the following possibilities: 1) an(ab) 6= an and (ab)bn 6= bn for all n ∈ N, in which case we denote S by O(∞,∞)(a, b); 2) (ab)bm 6= bm for all m ∈ N but ak(ab) = ak for some k ∈ N, and letting n be the smallest of such integers k, we denote S by O(n,∞)(a, b); 3) an(ab) 6= an for all n ∈ N but (ab)bl = bl for some l ∈ N, and with m standing for the smallest of such integers l, we denote S by O(∞,m)(a, b); 4) an(ab) = an and (ab)bm = bm for some n, m ∈ N, and letting n and m be the smallest integers with these properties, we denote S by O(n,m)(a, b) (of course, O(1,1)(a, b) is just another notation for the bicyclic semigroup B(a, b)). 8 SIMON M. GOBERSTEIN Remark 4. In view of Remark 1, the dual of O(∞,∞)(a, b) is O(∞,∞)(b, a), the dual of O(n,∞)(a, b) is O(∞,n)(b, a), and the dual of O(n,m)(a, b) is O(m,n)(b, a). Thus all results about O(∞,m)(a, b) are obtained automatically from the corresponding results about O(m,∞)(b, a). Proposition 2.4. Let S = ha, bi be a monogenic orthodox semigroup such that a ∈ NS and ab = a2b2. Then every nontrivial relation that holds in S (other than ab = a2b2) is equivalent either to an+1b = an, or to abm+1 = bm, or to an+1b = an and abm+1 = bm for some m, n ∈ N. Proof. By Lemma 1.4, o(a) = o(b) = ∞. Suppose that S satisfies a nontrivial relation In view of Lemma 2.2, we may w(a, b) = w′(a, b), which is not equivalent to ab = a2b2. assume that w(a, b) and w′(a, b) are abridged words in a, b. Case 1: w(a, b) and w′(a, b) are both of type I. In this case, w(a, b) = aibm and w′(a, b) = ai′bm′ with i, i′ ∈ {0, 1}, m > i, and m′ > i′. If i = i′, then bm = bm′ whence m = m′, so that w(a, b) = w′(a, b) is a trivial relation; a contradiction. Therefore i 6= i′. Assume without loss of generality that i = 0 and i′ = 1. Then bm = abm′ whence bm+1 = bm′ and so m′ = m + 1. Thus the given relation is abm+1 = bm. Case 2: w(a, b) is of type I and w′(a, b) of type II. Here we have aibm = an′bj ′ with i, j ′ ∈ {0, 1}, m > i, and n′ ≥ 1. Thus bm−i+1 = ban′bj ′, which implies b(m−i)+(n′ −j ′)+1 = bm−i+1bn′−j ′ = ban′bn′ = b. Hence (m − i) + (n′ − j ′) = 0, contrary to the conditions m − i > 0 and n′ − j ′ ≥ 0. Therefore this case cannot happen. Case 3: w(a, b) is of type I and w′(a, b) of type III. In this case, aibm = ai′bm′an′bj ′ where i, i′, j ′ ∈ {0, 1}, m > i, m′ > i′, and n′ > j ′. Then b(aibm)a = b(ai′bm′an′bj ′)a, that is, bm−i+1a = bm′−i′+1an′−j ′+1. By Lemma 1.5, m − i = m′ − i′ and n′ − j ′ = 0, contradicting the fact that n′ > j ′. Hence this case is also impossible. Case 4: w(a, b) and w′(a, b) are both of type II. Here anbj = an′bj ′ with j, j ′ ∈ {0, 1}, n ≥ 1, and n′ ≥ 1, and hence anbja = an′bj ′a, that is, an−j+1 = an′−j ′+1, which implies n − j = n′ − j ′. Since the relation anbj = an′bj ′ is nontrivial, we have j 6= j ′. Without loss of generality assume that j = 0 and j ′ = 1. Then an = an′b whence an+1 = an′ and therefore n′ = n + 1. Thus the given relation is an+1b = an. Case 5: w(a, b) is of type II and w′(a, b) of type III. In this case, anbj = ai′bm′an′bj ′ with j, i′, j ′ ∈ {0, 1}, n ≥ 1, m′ > i′, and n′ > j ′. Then b(anbj)a = b(ai′bm′an′bj ′)a, that is, ban−j+1 = bm′−i′+1an′−j ′+1. By Lemma 1.5, n − j = n′ − j ′ and m′ − i′ = 0, but the latter contradicts the fact that m′ > i′. Thus this case is impossible. Case 6: w(a, b) and w′(a, b) are both of type III. The given relation is aibmanbj = ai′bm′an′bj ′ with i, j, i′, j ′ ∈ {0, 1}, m > i, n > j, m′ > i′, and n′ > j ′. Then b(aibmanbj)a = b(ai′bm′an′bj ′)a, that is, bm−i+1an−j+1 = bm′−i′+1an′−j ′+1 and so, by Lemma 1.5, m − i = m′ − i′ and n − j = n′ − j ′. Since the given relation is nontrivial, we cannot have both i = i′ and j = j ′. 6a) Suppose that i = i′ and j 6= j ′. Assuming without loss of generality that j = 0 and j ′ = 1, we have the relation aibman = aibm′an′b. Hence b(aibman)a = b(aibm′an′b)a, that is, bm−i+1an+1 = bm′−i+1an′. By Lemma 1.5, m′ = m and n′ = n + 1, so aibman = aibm′an′b is, in fact, aibman = aibman+1b. Then an = ambman = am−iaibman = am−iaibman+1b = an+1b. BISIMPLE MONOGENIC ORTHODOX SEMIGROUPS 9 Since an = an+1b implies aibman = aibman+1b, the given relation is equivalent to an = an+1b. 6b) By symmetry with 6a), if i 6= i′ and j = j ′, the relation is equivalent to bm = abm+1. 6c) Suppose that i 6= i′ and j 6= j ′. Assume without loss of generality that i = 0 and i′ = 1. Suppose, first, that j = 0 and j ′ = 1. Since bman = abm′an′b implies bm+1an+1 = bm′an′ and so, by Lemma 1.5, m′ = m + 1 and n′ = n + 1, the given relation is bman = abm+1an+1b. Hence bm = bmanbn = abm+1an+1bn+1 = abm+1 and an = ambman = am+1bm+1an+1b = an+1b. On the other hand, bm = abm+1 and an = an+1b together clearly imply bman = abm+1an+1b. Therefore the given relation is equivalent to two relations, bm = abm+1 and an = an+1b. Assume now that j = 1 and j ′ = 0. Since bmanb = abm′an′ implies bm+1an = bm′an′+1, by Lemma 1.5, m′ = m + 1 and n = n′ + 1, so the given relation is bmanb = abm+1an−1. Then bm = bmanbn = abm+1an−1bn−1 = abm+1 and an−1 = am+1bm+1an−1 = ambmanb = anb. Since bm = abm+1 and an−1 = anb together clearly imply bmanb = abm+1an−1, the given relation is equivalent to two relations, bm = abm+1 and an−1 = anb. This completes the proof. (cid:3) It is asserted (without proof) in [5] that [5, Figure 2] gives an eggbox picture of FO(a, b)/α′ but, as pointed out in the paragraph following Corollary 1.2, [5, Figure 2] contains quite a few misprints. Here is its corrected version (with p and q replaced by a and b, respectively): ... ... ... ... ... ... ... · · · ab4a4b ab4a3b ab4a2b ab4 ab4a ab4a2 ab4a3 · · · ab3a4b ab3a3b ab3a2b ab3 ab3a ab3a2 ab3a3 · · · ab2a4b ab2a3b ab2a2b ab2 ab2a ab2a2 ab2a3 · · · · · · · · · · · · ... ... ... a2 ba2 b2a2 b3a2 ... a3 ba3 b2a3 b3a3 ... a3b ba3b b2a3b b3a3b a2b ba2b b2a2b b3a2b a4b ba4b b2a4b b3a4b a ba b2a b3a ... ab b b2 b3 ... ... ... ... ... ... ... ... ... ... · · · · · · · · · · · · · · · · · · · · · ... ... ... Figure 1 It follows from Proposition 2.4 that O(∞,∞)(a, b) = h a, b a ⊥ b, ab = a2b2 i. Therefore, by Corollary 1.2, O(∞,∞)(a, b) = FO(a, b)/α′ which means (in view of the above mentioned assertion in [5] about the eggbox picture of FO(a, b)/α′) that Figure 1 exhibits the eggbox picture of O(∞,∞)(a, b). We will include a proof of this fact both for the sake of completeness and because it is independent of the results about the lattice of congruences on FO contained in [5, Section II]. Lemma 2.5. The semigroup O(∞,∞)(a, b) is a combinatorial bisimple monogenic orthodox semigroup whose eggbox picture is given in Figure 1. 10 SIMON M. GOBERSTEIN Proof. Let S = O(∞,∞)(a, b). According to Lemma 2.3, S is bisimple and combinatorial. The latter implies that Ra ∩Lb = {ab}. By Lemma 1.4, hai∪haib ⊆ Ra and hbi∪ahbi ⊆ Lb. It is clear that hai ∪ haib coincides with the set of all abridged words in a, b of type II, and from the definition of O(∞,∞)(a, b) and the proof of Proposition 2.4 (see Case 4), it is immediate that anbj = an′bj ′ if and only if n = n′ and j = j ′ for all anbj, an′bj ′ ∈ hai ∪ haib. It follows also from the definition of O(∞,∞)(a, b) and the proof of Proposition 2.4 that if i, i′ ∈ {0, 1} and k, k′ ∈ N, then aibk = ai′bk′ if and only if i = i′ and k = k′, and it is obvious that the set of all abridged words in a, b of type I coincides with (hbi ∪ ahbi) \ {ab}. Suppose that aibmanbj is an abridged word of type I or III such that aibmanbj ∈ Ra. By Lemma 2.3(i), aibmRaibmanbj and hence aibm ∈ Ra ∩ Lb = {ab}; a contradiction. Therefore Ra = hai ∪ haib and, similarly, Lb = hbi ∪ ahbi, which also shows that S \ (Ra ∪ Lb) coincides with the set of all abridged words in a, b of type III. Finally, if aibmanbj is any abridged word of type III, Lemma 2.3(i) and the above remarks imply that {aibmanbj} = Raibm ∩ Lanbj . It follows that Figure 1 gives the eggbox picture of O(∞,∞)(a, b). The proof is complete. (cid:3) Remark 5. It is immediate from Lemma 2.5 that for each x ∈ O(∞,∞)(a, b) there is a unique abridged word aibmanbj in a, b such that x = aibmanbj, and we will say that aibmanbj is a reduced word representing x (or the reduced form of x). It is also easily seen that O(∞,∞)(a, b) is the disjoint union of four bicyclic semigroups and two infinite cyclic semi- groups: O(∞,∞)(a, b) = B(a2b, ab2) ∪ B(ab2a2, ab3a) ∪ B(ba3b, b2a2b) ∪ B(ba2, b2a) ∪ hai ∪ hbi. Let n ∈ N. By definition of O(n,∞)(a, b), we have (ab)bm 6= bm for all m ∈ N, an(ab) = an, and al(ab) 6= al for all 1 ≤ l < n. Thus O(n,∞)(a, b) = ha, b a ⊥ b, ab = a2b2, an+1b = an i, which shows that O(n,∞)(a, b) is a homomorphic image of O(∞,∞)(a, b), and the eggbox pic- ture of O(n,∞)(a, b) is easily obtained from that of O(∞,∞)(a, b); it is given in Figure 2. ... ... ab4anb ab4an−1b ab3anb ab3an−1b ab2anb ab2an−1b anb banb b2anb b3anb ... an−1b ban−1b b2an−1b b3an−1b ... ... ... ... ... ... ... ... · · · ab4a2b ab4 ab4a ab4a2 ab4a3 · · · ab3a2b ab3 ab3a ab3a2 ab3a3 · · · ab2a2b ab2 ab2a ab2a2 ab2a3 · · · · · · · · · · · · ... ... ... a3 ba3 b2a3 b3a3 ... a2 ba2 b2a2 b3a2 ... a2b ba2b b2a2b b3a2b a ba b2a b3a ... ab b b2 b3 ... ... ... ... ... ... · · · · · · · · · · · · · · · · · · · · · ... ... ... Note that for each x ∈ O(n,∞)(a, b) there is a unique abridged word aibmalbj with l ≤ n such that x = aibmalbj; we will call it the reduced form of x. By duality, if m ∈ N, we obtain the eggbox picture of O(∞,m)(a, b) and observe that each x ∈ O(∞,m)(a, b) is represented by Figure 2 BISIMPLE MONOGENIC ORTHODOX SEMIGROUPS 11 a unique abridged word aibkanbj with k ≤ m, called the reduced form of x. Now take any m, n ∈ N. By definition of O(n,m)(a, b), we have an(ab) = an and (ab)bm = bm but al(ab) 6= al for all 1 ≤ l < n and (ab)bk 6= bk for all 1 ≤ k < m. Therefore O(n,m)(a, b) is given by the presentation: O(n,m)(a, b) = ha, b a ⊥ b, ab = a2b2, an+1b = an, abm+1 = bm i, and so, in particular, O(n,m)(a, b) is a homomorphic image of O(n,∞)(a, b). This enables us to obtain the eggbox picture of O(n,m)(a, b) from that of O(n,∞)(a, b); it is shown in Figure 3. It is clear that for every x ∈ O(n,m)(a, b) there is a unique abridged word aibkalbj such that k ≤ m, l ≤ n, and x = aibkalbj; we will say that aibkalbj is the reduced form of x. abman−1b abmanb abm−1anb abm−1an−1b ... ab3anb ab2anb anb banb b2anb ... ... ab3an−1b ab2an−1b an−1b ban−1b b2an−1b ... ab3a2b ab2a2b ... ... abma2b abm abma abma2 · · · · · · abm−1a2b abm−1 abm−1a abm−1a2 ... ... ... · · · · · · · · · · · · · · · ... ... ... ... ab3 ab2 ab b b2 ... a2 ba2 b2a2 ... a2b ba2b b2a2b ... ab3a ab2a ... ab3a2 ab2a2 a ba b2a ... · · · · · · ... ... ... · · · · · · · · · · · · · · · ... ... ... Figure 3 For convenience of reference, we summarize the above observations in the following lemma. Lemma 2.6. Let m, n ∈ N. The semigroups O(n,∞)(a, b) and O(n,m)(a, b) are combinatorial bisimple monogenic orthodox semigroups whose eggbox pictures are given in Figures 2 and 3, respectively, and the semigroup O(∞,m)(a, b) is the dual of O(m,∞)(b, a). Recall that a band E is said to be uniform if eEe ∼= f Ef for all e, f ∈ E. As shown by Hall [8, Main Theorem], a band E is the band of a bisimple orthodox semigroup if and only if E is uniform. In particular, the easy part of the cited theorem guarantees that the band of idempotents of any bisimple orthodox semigroup is uniform (see [8, Result 7] or [10, Proposition VI.3.1]). Therefore, for any m, n ∈ N, the bands of idempotents of semigroups O(∞,∞)(a, b), O(n,∞)(a, b), O(∞,m)(a, b), and O(n,m)(a, b) are uniform, and using the diagrams of these bands we can explicitly illustrate this fact. In view of duality, it is sufficient to consider the bands of idempotents of the semigroups O(∞,∞)(a, b), O(n,∞)(a, b), and O(n,m)(a, b) with m ≥ n; they are shown in parts (a), (b), and (c), respectively, of Figure 4 with the bold line segments representing the covering relation of the natural partial order and the thin line segments indicating the R- and L-relations on each of these bands. (The diagrams of O(n,∞)(a, b) and O(n,m)(a, b) are drawn under the assumptions that n > 1 and m > n > 1, respectively; modifications for n = 1 and for 12 SIMON M. GOBERSTEIN m = n > 1 or m > n = 1 are obvious.) It is easily seen that if E is any of the bands shown in Figure 4 and e ∈ E, then eEe is isomorphic to the chain of idempotents of the bicyclic semigroup (that is, to the chain e0 > e1 > e2 > · · · ), and hence eEe ∼= f Ef for all e, f ∈ E. ab t ab t ab t ab2a2b ★★ ★ L ★ ★ t ba2b ab3a3b ★★ ★ L ★ ★ t b2a3b t t R R ... ... ... R ab2a t ★★ ★ L t ★ L ab2a2b ★★ ... ... ★ t ★ ba2b R R ★ L ★ ★ ba t ab2a t ★★ ... ... ★ ★ ba t R ★ ★ b2a2 t ab3a2 t ★★ ★ L ... ... ... ★ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ t bn−1anb abnanb ★★ t ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ★ L ★ R ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ R ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ... ... ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ★ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ★ t t t R R R ★ ★ ★ L ★ L ★ ba t t ★ ba2b abnanb ★★ ab2a ab2a2b t ★★ ★★ ... ... ... ... ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ★ abn+1an ★★ ... L ★ abmam−1 ★★ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ... ... ... ... ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ❛❛❛❛❛❛❛❛ ★ ... ★ abnan−1 ★★ bn−1an−1 ★ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ★ L ★ L bnan ★ t ★ ★ R t t t t bm−1am−1 t L ★ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ bmam t bm+1am+1 t ... (c) abnan−1 t ★ t bn−1anb abn+1an t abn+2an+1 t ★★ ★ L ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ★★ L ★ ★★ ... L ★ bn−1an−1 ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ ❜ bnan ★ t ★ ★ bn+1an+1 t ★ ... (a) (b) Figure 4 We have shown that if a monogenic orthodox semigroup S = ha, bi such that a, b ∈ NS satisfies ab = a2b2, then S coincides with one of the bisimple combinatorial semigroups O(∞,∞)(a, b), O(n,∞)(a, b), O(∞,m)(a, b), or O(n,m)(a, b) for some m, n ∈ N, described by Lem- mas 2.5 and 2.6. We will prove now that there are no other bisimple orthodox semigroups generated by two mutually inverse nongroup elements and thus obtain a complete description of all monogenic orthodox semigroups with nongroup generators. By [7, Lemma 2.7] (deduced in [7] as a corollary to [5, Section II]), if S is an orthodox semigroup, a ∈ NS, and b ∈ V (a), then either {a, b, ab, ba} is the top D-class of ha, bi (which means that ha, bi \ {a, b, ab, ba} is an ideal of ha, bi), or o(a) = ∞ = o(b). As mentioned prior to the statement of [7, Lemma 2.7], it could have been proved independently of [5]. Now we will give such a direct proof establishing, in fact, a sharper result. Lemma 2.7. (An extension of [7, Lemma 2.7]) Let S be an orthodox semigroup, a an arbitrary nongroup element of S, and b ∈ V (a). Then either {a, b, ab, ba} is the top D-class of ha, bi, or (perhaps after interchanging a and b) ab = a2b2 and ba 6= b2a2, and hence o(a) = o(b) = ∞, hai ∪ haib ⊆ Ra, and hbi ∪ ahbi ⊆ Lb. BISIMPLE MONOGENIC ORTHODOX SEMIGROUPS 13 Proof. Denote A = ha, bi, and put, as always, I A(a) = J A(a) \ J A a (the superscript A is used to distinguish subsets of A from the corresponding subsets of S). Suppose a2 ∈ I A(a). Since I A(a) is an ideal of A, we conclude that J A a ⊆ J A a . Thus J A a = {a, b, ab, ba}. Moreover, A \ {a, b, ab, ba} coincides with I A(a) and hence is an ideal of A. Therefore {a, b, ab, ba} is the top D-class of ha, bi. For the rest of the proof, we will assume that a2 6∈ I A(a), that is, (a, a2) ∈ J A. a = DA a = J A(a)\I A(a) = {a, b, ab, ba} ⊆ DA Suppose a2 is neither a left nor a right divisor of a. By assumption, a ∈ Aa2A. Therefore we can choose shortest possible words u and v in a, b such that a = ua2v. Since b = b(ua2v)b and since, by Lemma 1.3, b2 is neither a left nor a right divisor of b, the first letter of u as well as the last letter of v is a. In fact, in view of our assumption about a2, we must have u = abu′ and v = v′ba where u′ and v′ are nonempty words in a, b. The first letter of u′ cannot be a, for otherwise u′ = au′′ for some nonempty word u′′ in a, b, and hence a = (au′′)a2v with au′′ being shorter than u, a contradiction. Similarly, the last letter of v′ cannot be a. Thus b is the first letter of u′ and the last letter of v′. However, b = bu′a2v′b and so b2 is both a left and a right divisor of b, again a contradiction. Therefore a2 is either a left or a right divisor of a and, by Lemma 1.3, b2 is, respectively, a right or a left divisor of b. Thus, interchanging a and b if necessary, we may assume that a2 is a left divisor of a. Then, according to Lemma 1.3, ab = a2b2 and ba 6= b2a2 whence, by Lemma 1.4, o(a) = o(b) = ∞, hai ∪ haib ⊆ Ra, and hbi ∪ ahbi ⊆ Lb. (cid:3) The author gratefully acknowledges that he learned the argument used in the second paragraph of the above proof from Mark Sapir (in a private conversation many years ago). Corollary 2.8. Let S = ha, bi be a bisimple monogenic orthodox semigroup such that a (and therefore b) is a nongroup element of S. Then (perhaps after interchanging a and b) we have ab = a2b2 and ba 6= b2a2. The principal results of this section are combined in the following theorem. Theorem 2.9. Let S = ha, bi be a monogenic orthodox semigroup such that a (and hence b) is a nongroup element of S. Then S is bisimple if and only if S (or the dual of S) coincides with one of the semigroups O(∞,∞)(a, b), O(n,∞)(a, b), O(∞,m)(a, b), or O(n,m)(a, b) for some m, n ∈ N. 3. Monogenic orthodox semigroups with group generators Let S = ha, bi be a monogenic orthodox semigroup such that a (and hence b) is a group element of S. To avoid repetition, this assumption will be retained through the rest of this section. By Lemma 1.3(ii), ab = a2b2 and ba = b2a2, and the identity elements of the groups Ha and Hb are ab2a and ba2b, respectively. Since ab = a2b2 and ba = b2a2, it is easily seen that each element of S can be written (in general, not uniquely) in the form bianbj or aibnaj for some i, j ∈ {0, 1} and n ∈ N. Denote ea = ab2a and eb = ba2b. Then eaeb = ab ∈ Ra ∩ Lb and ebea = ba ∈ La ∩ Rb, and hence ES is a rectangular band with at most four elements: ea, eb, ab, and ba. It follows that S is isomorphic to the direct product of Ha and ES (see, for 14 SIMON M. GOBERSTEIN example, [10, Exercise III.12]) so, in particular, S is bisimple. There are four possibilities: 1) a(ab) 6= a and (ab)b 6= b, which happens precisely when the idempotents ea, eb, ab, and ba are pairwise distinct, and thus ES = {ea, ab, ba, eb} is a 2 × 2 rectangular band; 2) a(ab) 6= a and (ab)b = b, which is equivalent to the condition ea = ba 6= ab = eb, and hence ES = {ea, eb} is a 2-element right zero semigroup; 3) a(ab) = a and (ab)b 6= b, which is equivalent to ea = ab 6= ba = eb, so that ES = {ea, eb} is a 2-element left zero semigroup; 4) a(ab) = a and (ab)b = b, that is, ea = ab = ba = eb, and thus ES is a singleton. The eggbox pictures of S in cases 1 -- 4 are shown, respectively, in parts (a) -- (d) of Figure 5. Ha Hab Hba Hb Ha Hb (a) (b) Ha Hb (c) Ha (d) Figure 5 Case 1: a(ab) 6= a and (ab)b 6= b. Suppose that o(a) = o(b) = ∞. Since ab3a is the inverse of a in the group Ha and (ab3a)n = abn+2a for all n ∈ N, we have Ha ⊇ ha, ab3ai = {. . . , ab4a, ab3a, ab2a, a, a2, . . .} ∼= Z. Dually, Hb ⊇ hb, ba3bi = {. . . , ba4b, ba3b, ba2b, b, b2, . . .} ∼= Z. Note that a2b is the inverse of ab2 in the group Hab, and (a2b)n = an+1b and (ab2)n = abn+1 for all n ∈ N. It follows that Hab ⊇ {. . . , ab3, ab2, ab, a2b, a3b, . . .} = ha2b, ab2i ∼= Z. By symmetry, we also have Hba ⊇ {. . . , ba3, ba2, ba, b2a, b3a, . . .} = hb2a, ba2i ∼= Z. Since each element of S has the form bianbj or aibnaj for some i, j ∈ {0, 1} and n ∈ N, we conclude that Ha = ha, ab3ai, Hab = ha2b, ab2i, Hba = hb2a, ba2i, and Hb = hb, ba3bi. Note that in this case S coincides with the semigroup FO(a, b)/σ′ of [5, Result 2.4]. Now let o(a) = o(b) = m ∈ N. Then it is clear that Ha, Hb, Hab, and Hba are cyclic groups of order m. More specifically, Ha = {ab2a, a, . . . , am−1} = hai, Hb = {ba2b, b, . . . , bm−1} = hbi, Hab = {ab, a2b, . . . , am−1b} = ha2bi, and Hba = {ba, b2a, . . . , bm−1a} = hb2ai. Case 2: a(ab) 6= a and (ab)b = b. Here, as observed above, ab2a = ba 6= ab = ba2b, so ES = {ba, ab} is a 2-element right zero semigroup, and S ∼= Ha×ES, that is, S is a right group. Note that bna = abn+1a and an = ban+1 for all n ∈ N. If o(a) = o(b) = ∞, then Ha = Hba = ha, b2ai = {. . . , b3a, b2a, ba, a, a2, . . .} ∼= Z and Hb = Hab = hb, a2bi = {. . . , a3b, a2b, ab, b, b2, . . .} ∼= Z. Now assume that o(a) = o(b) = m ∈ N. ∼= Hab = Hb or, in more detail, Ha = Hba = {ba, a, . . . , am−1} and Then Ha = Hba Hab = Hb = {ab, b, . . . , bm−1}. ∼= Zm Case 3: a(ab) = a and (ab)b 6= b. Here the description of the structure of S is obtained from that of Case 2 by duality. Case 4: a(ab) = a and (ab)b = b. In this case, ES is a singleton and S = Ha is a cyclic group. BISIMPLE MONOGENIC ORTHODOX SEMIGROUPS 15 The results of this section are summarized in the following theorem. Theorem 3.1. Let S = ha, bi be a monogenic orthodox semigroup such that a (and hence b) is a group element of S. (1) Let a(ab) 6= a and (ab)b 6= b. Then ES = {ab2a, ab, ba, ba2b} is a 2 × 2 rectangular band and S ∼= Ha × ES is a bisimple semigroup whose eggbox picture is shown in Figure 5(a). If o(a) = ∞, then Ha = ha, ab3ai, Hab = ha2b, ab2i, Hba = hb2a, ba2i, and Hb = hb, ba3bi are infinite cyclic groups, and if o(a) = m for some m ∈ N, then the H-classes of S are finite cyclic groups of order m: Ha = {ab2a, a, . . . , am−1} = hai, Hb = {ba2b, b, . . . , bm−1} = hbi, Hab = {ab, a2b, . . . , am−1b} = ha2bi, and Hba = {ba, b2a, . . . , bm−1a} = hb2ai. (2) Let a(ab) 6= a and (ab)b = b. Then ES = {ba, ab} is a 2-element right zero semigroup, and S ∼= Ha ×ES is a right group whose eggbox picture is shown in Figure 5(b). If o(a) = ∞, then Ha = ha, b2ai and Hb = hb, a2bi are infinite cyclic groups, and if o(a) = m ∈ N, then Ha = {ba, a, . . . , am−1} = hai ∼= Zm and Hb = {ab, b, . . . , bm−1} = hbi ∼= Zm. (3) If a(ab) = a and (ab)b 6= b, then the structure of the dual of S is described in (2). (4) If a(ab) = a and (ab)b = b, then S is a cyclic group. Remark 6. The author is grateful to a reader of the preprint for proposing to use in the proof of Lemma 2.3(ii) the fact that if S = ha, bi is a monogenic orthodox semigroup with ab = a2b2 and ba 6= b2a2, then S/Y is bicyclic, and for suggesting that combinatorial bisimple monogenic orthodox semigroups can be described by investigating the idempotent- pure congruences on the semigroup ha, b ai ⊥ bi for each i ≥ 1, ab = a2b2i. Following the first suggestion, the author has shortened the proof that S is combinatorial in Lemma 2.3(ii), originally proved by showing directly that Hba = 1. However, after examining the second suggestion, the author has concluded that it would not lead to a shorter way of describing combinatorial bisimple monogenic orthodox semigroups than that given in Section 2. References [1] A. H. Clifford, The fundamental representation of a regular semigroup, Semigroup Forum, 10 (1975), 84 -- 92. [2] A. H. Clifford and G. B. Preston, The Algebraic Theory of Semigroups, Math. Surveys No. 7, Amer. Math. Soc., Providence, R.I.; Vol. I, 1961; Vol. II, 1967. [3] C. Eberhart and W. Williams, Congruences on an orthodox semigroup via the minimum inverse semi- group congruence, Glasgow Math. J., 18 (1977), 181 -- 192. [4] C. Eberhart and W. Williams, Elementary semigroups, Semigroup Forum, 15 (1978), 295 -- 309. [5] C. Eberhart and W. Williams, Elementary orthodox semigroups, Semigroup Forum, 29 (1984), 351 -- 364. [6] L. M. Gluskin, Elementary generalized groups, Matem. Sbornik, 41 (1957), 23 -- 36. [7] S. M. Goberstein, On orthodox semigroups determined by their bundles of correspondences, Pacific J. Math., 153 (1992), 71 -- 84. [8] T. E. Hall, On orthodox semigroups and uniform and antiuniform bands, J. Algebra, 16 (1970), 204 -- 217. [9] T. E. Hall, Congruences and Green's relations on regular semigroups, Glasgow Math. J., 13 (1972), 167 -- 175. [10] J. M. Howie, An Introduction to Semigroup Theory, Academic Press, London, 1976.
1610.00583	5	1610	2018-07-25T13:38:22	Resolutions for Twisted Tensor Products	[ "math.RA", "math.RT" ]	We build resolutions for general twisted tensor products of algebras. These bimodule and module resolutions unify many constructions in the literature and are suitable for computing Hochschild (co)homology and more generally Ext and Tor for (bi)modules. We analyze in detail the case of Ore extensions, consequently obtaining Chevalley-Eilenberg resolutions for universal enveloping algebras of Lie algebras (defining the cohomology of Lie groups and Lie algebras). Other examples include semidirect products, crossed products, Weyl algebras, Sridharan enveloping algebras, and Koszul pairs.	math.RA	math	RESOLUTIONS FOR TWISTED TENSOR PRODUCTS A.V. SHEPLER AND S. WITHERSPOON Abstract. We build resolutions for general twisted tensor products of algebras. These bi- module and module resolutions unify many constructions in the literature and are suitable for computing Hochschild (co)homology and more generally Ext and Tor for (bi)modules. We analyze in detail the case of Ore extensions, consequently obtaining Chevalley-Eilenberg resolutions for universal enveloping algebras of Lie algebras (defining the cohomology of Lie groups and Lie algebras). Other examples include semidirect products, crossed products, Weyl algebras, Sridharan enveloping algebras, and Koszul pairs. 1. Introduction Motivated by questions in noncommutative geometry, Cap, Schichl, and Vanzura [5] introduced a very general twisted tensor product of algebras to replace the (commutative) tensor product. Their examples included noncommutative 2-tori and crossed products of C∗-algebras with groups. Many other algebras of interest arise as twisted tensor product algebras: crossed products with Hopf algebras, algebras with triangular decomposition (e.g., universal enveloping algebras of Lie algebras and quantum groups), braided tensor products defined by R-matrices, and other biproduct constructions. In fact, twisted tensor product algebras are rather copious: If an algebra is isomorphic to A ⊗ B as a vector space for two of its subalgebras A and B under the canonical inclusion maps, then it must be isomorphic to a twisted tensor product A ⊗τ B for some twisting map τ : B ⊗ A → A ⊗ B (see [5]). Modules over a twisted tensor product algebra arise from tensoring together modules for the individual algebras: If M and N are modules over algebras A and B, respectively, compatible with a twisting map τ , then M ⊗N adopts the structure of a module over A⊗τB. We describe in this note a general method to twist together resolutions of A-modules and B-modules in order to construct resolutions for the corresponding modules over the twisted tensor product A ⊗τ B. A similar method works for bimodules. In particular, we twist together resolutions of algebras over a field to obtain a resolution for a twisted tensor product algebra as a bimodule over itself. We are motivated by a desire to understand deformations of twisted tensor products and to describe the homology theory in terms of the homology of the original factor algebras. For example, under some finiteness assumptions, the Hochschild cohomology of a tensor product of algebras is the tensor product of their Hochschild cohomology rings. A similar statement is true of the cohomology of augmented algebras. Both results hold because the Date: July 24, 2018. 2010 Mathematics Subject Classification. 16E40. Key words: twisted tensor product, projective resolutions, Hochschild (co)homology, augmented algebras. The first author was partially supported by Simons grant 429539. The second author was partially supported by NSF grant DMS-1401016. 1 2 A.V. SHEPLER AND S. WITHERSPOON tensor product of projective resolutions for the factor algebras is a projective resolution for the tensor product of the algebras. In some particular settings, similar homological constructions have appeared for modified versions of the tensor product of algebras. We mention just a few examples. Gopalakrish- nan and Sridharan [7] constructed resolutions for modules of Ore extensions. Bergh and Oppermann [1] twisted resolutions when the twisting arises from a bicharacter on grading groups. Jara Martinez, L´opez Pena, and S¸tefan [12] worked with Koszul pairs. Guccione and Guccione [8, 9] built resolutions for twisted tensor products, in particular crossed products with Hopf algebras, out of bar and Koszul resolutions of the factor algebras. We adapted this last construction in [16] to handle more general resolutions for the case of skew group algebras in order to understand deformations. Walton and the second author generalized these resolutions to smash products with Hopf algebras in [18]. In this paper, we unify many of these previous constructions and provide methods use- ful in new settings for finding resolutions of modules over twisted tensor product algebras: We show very generally that projective resolutions for bimodules of two factor algebras can be twisted together to construct a projective resolution for the resulting bimodule for the twisted tensor product given some compatibility conditions. This twisting of resolu- tions provides an efficient means for computing and handling Hochschild (co)homology in particular. A similar construction applies to projective (left) module resolutions used, for example, to compute (co)homology of augmented algebras. We verify that many known resolutions may be viewed as twisted resolutions in this way, including some of those mentioned above. We give details in the case of Ore extensions. In particular, the bimodule Koszul resolution of a universal enveloping algebra U(g) is a twisted resolution when g is a finite dimensional supersolvable Lie algebra. More general Lie algebras can be handled via triangular decomposition. Our method also leads to standard resolutions for Weyl algebras and some Sridharan enveloping algebras. For an Ore extension, we adapt results of Gopalakrishnan and Sridharan [7] to construct twisted product resolutions of modules. We thus regard the Chevalley-Eilenberg complex of U(g) as a twisted product resolution. This defines Lie algebra and Lie group cohomology in terms of an iterative twisting of resolutions. In Section 2, we give definitions and some preliminary results. Bimodule twisted tensor product complexes are constructed in Section 3 and we show they give projective resolutions in Theorem 3.10. Section 4 gives applications to some types of Ore extensions. We construct twisted tensor product complexes for resolving modules in Section 5, and we show these complexes are projective resolutions in Theorem 5.12. Applications to Ore extensions appear in Section 6. We fix a field k of arbitrary characteristic throughout. All tensor products are over k unless otherwise indicated, i.e., ⊗ = ⊗k, and all algebras are k-algebras. Modules are left modules unless otherwise described. 2. Twisted tensor product algebras and compatible resolutions In this section, we recall twisted tensor product algebras from [5] and define a compat- ibility condition necessary for twisting resolutions together. Examples include skew group algebras and crossed products with Hopf algebras [13], twisted tensor products given by TWISTED TENSOR PRODUCTS 3 bicharacters of grading groups [1], braided products arising from R-matrices [11], two- cocycle twists of Hopf algebras [15], and more. Let A and B be associative algebras over k with multiplication maps mA : A ⊗ A → A and mB : B ⊗ B → B and multiplicative identities 1A and 1B, respectively. We write 1 for the identity map on any set. Twisted tensor products. A twisting map τ : B ⊗ A → A ⊗ B is a bijective k-linear map for which τ (1B ⊗ a) = a ⊗ 1B and τ (b ⊗ 1A) = 1A ⊗ b for all a ∈ A and b ∈ B, and (2.1) τ ◦ (mB ⊗ mA) = (mA ⊗ mB) ◦ (1 ⊗ τ ⊗ 1) ◦ (τ ⊗ τ ) ◦ (1 ⊗ τ ⊗ 1) as maps B ⊗ B ⊗ A ⊗ A → A ⊗ B. The twisted tensor product algebra A ⊗τ B is the vector space A ⊗ B together with multiplication mτ given by such a twisting map τ . By [5, Proposition/Definition 2.3], the algebra A ⊗τ B is associative. Note that the left-right distinction in a twisted tensor product algebra is artificial since A ⊗τ B ∼= B ⊗τ −1 A. Indeed, one might identify A ⊗τ B with the algebra generated by A and B (so that A and B are subalgebras) with relations given by Equation (2.1). If A and B are N-graded algebras, we take the standard N-grading on A ⊗ B and B ⊗ A and say a twisting map τ is strongly graded if it takes Bj ⊗ Ai to Ai ⊗ Bj for all i, j following Conner and Goetz [4]. (Note that [12] leave off the adjective strongly.) In this case, the twisted tensor product algebra A ⊗τ B is N-graded. Example 2.2. The Weyl algebra W = khx, yi/(xy − yx − 1) is isomorphic to the twisted tensor product A ⊗τ B of A = k[x] and B = k[y] with twisting map τ : B ⊗ A → A ⊗ B defined by τ (y ⊗ x) = x ⊗ y − 1 ⊗ 1. Likewise, the Weyl algebra Wn on 2n indeterminates, Wn = khx1, . . . , xn, y1, . . . , yni/(xixj − xjxi, yiyj − yjyi, xiyj − yjxi − δi,j : 1 ≤ i, j ≤ n) , is isomorphic to a twisted tensor product. These are examples of (iterated) Ore extensions, which we consider in detail in Section 4. Example 2.3. A skew group algebra S ⋊ G for a finite group G acting on an algebra S by automorphisms is isomorphic to the twisted tensor product kG ⊗τ S of the group algebra kG and of S. The twisting map τ is defined by τ (s ⊗ g) = g ⊗ g−1(s) for s ∈ S and g ∈ G. We consider the special case that S is a Koszul algebra at the end of Section 3. Bimodules over twisted tensor products. We fix a twisting map τ : B ⊗ A → A ⊗ B for k-algebras A and B. Definition 2.4. An A-bimodule M is compatible with τ if there is a bijective k-linear map τB,M : B ⊗ M → M ⊗ B commuting with the bimodule structure of M and multiplication in B, i.e., as maps on B ⊗ B ⊗ M and on B ⊗ A ⊗ M ⊗ A, respectively, (2.5) (2.6) τB,M (mB ⊗ 1) = (1 ⊗ mB)(τB,M ⊗ 1)(1 ⊗ τB,M ) and τB,M (1 ⊗ ρA,M ) = (ρA,M ⊗ 1)(1 ⊗ 1 ⊗ τ )(1 ⊗ τB,M ⊗ 1)(τ ⊗ 1 ⊗ 1) , 4 A.V. SHEPLER AND S. WITHERSPOON where ρA,M : A ⊗ M ⊗ A → M is the bimodule structure map. If A is graded and M is a graded A-bimodule, we say that M is compatible with a strongly graded twisting map τ if there is a map τB,M as above that takes Bi ⊗ Mj to Mj ⊗ Bi for all i, j. Remark 2.7. Note that the above definition applies to B-bimodules as well as A-bimodules by reversing the role of A and B. Indeed, we apply the definition to the algebra B, the twisted tensor product B ⊗τ −1 A, and the twisting map τ −1 to obtain conditions for a B-bimodule N to be compatible with τ −1. We may rewrite these conditions using the convenient notation τN,A = (τ −1 A,N )−1. We obtain an equivalent right version of the above definition: A given B-bimodule N is compatible with τ −1 when there is some bijective k-linear map τN,A : N ⊗ A → A ⊗ N satisfying (2.8) τN,A(1 ⊗ mA) = (mA ⊗ 1)(1 ⊗ τN,A)(τN,A ⊗ 1) and τN,A(ρB,N ⊗ 1) = (1 ⊗ ρB,N )(τ ⊗ 1 ⊗ 1)(1 ⊗ τN,A ⊗ 1)(1 ⊗ 1 ⊗ τ ) , (2.9) as maps on N ⊗ A ⊗ A and on B ⊗ N ⊗ B ⊗ A, respectively, where ρB,N : B ⊗ N ⊗ B → N is the bimodule structure map. In light of the last remark, we will say a bimodule is compatible with τ when it is either an A-bimodule compatible with τ or a B-bimodule compatible with τ −1, since one often identifies A ⊗τ B and the isomorphic algebra B ⊗τ −1 A in practice. Remark 2.10. An A-bimodule M is compatible with the twisting map τ exactly when there is a bijective k-linear map τB,M : B ⊗ M → M ⊗ B making the following diagram commute: (2.11) B ⊗ M ⊗ B B ⊗ B ⊗ M M ⊗ B ⊗ B mB ⊗1 1⊗τB,M 4❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ &◆◆◆◆◆◆◆◆◆◆◆◆◆ 8♣♣♣♣♣♣♣♣♣♣♣♣♣ &◆◆◆◆◆◆◆◆◆◆◆◆◆ B ⊗ M τ ⊗1⊗1 A ⊗ B ⊗ M ⊗ A 1⊗ρA,M τB,M ⊗1 ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ x♣♣♣♣♣♣♣♣♣♣♣♣♣ f◆◆◆◆◆◆◆◆◆◆◆◆◆ 8♣♣♣♣♣♣♣♣♣♣♣♣♣ / M ⊗ B 1⊗mB 1⊗1⊗τ ρA,M ⊗1 / A ⊗ M ⊗ B ⊗ A . τB,M 1⊗ τB,M ⊗1 B ⊗ A ⊗ M ⊗ A A ⊗ M ⊗ A ⊗ B A similar diagram expresses compatibility of a B-bimodule N with τ . Example 2.12. Let M = A, an A-bimodule via multiplication. Then A is compatible with τ via τB,A = τ . Similarly N = B is compatible with τ . 4 & x / & 8 f / 8 TWISTED TENSOR PRODUCTS 5 Bimodule structure. When M and N are compatible with τ , the tensor product M ⊗ N is naturally an A ⊗τ B-bimodule via the following composition of maps: (2.13) A⊗τ B ⊗M ⊗N ⊗A⊗τ B 1⊗ τB,M ⊗ τN,A⊗1 / A⊗M ⊗B ⊗A⊗N ⊗B 1⊗1⊗ τ ⊗1⊗1 / A⊗M ⊗A⊗B ⊗N ⊗B ρA,M ⊗ ρB,N / M ⊗ N . Bimodule resolutions. For any k-algebra A, let Ae = A ⊗ Aop be its enveloping algebra, with Aop the opposite algebra to A. We view an A-bimodule M as a left Ae-module. In Lemma 3.1 below, we will construct a projective resolution of an (A ⊗τ B)e-module M ⊗ N from individual resolutions of M and N using some compatibility conditions. Let P q(M ) be an Ae-projective resolution of M and let P q(N ) be a Be-projective resolution of N : (2.14) (2.15) · · · → P2(M ) → P1(M ) → P0(M ) → M → 0, · · · → P2(N ) → P1(N ) → P0(N ) → N → 0. Bar and reduced bar resolutions. For example, M could be A itself and P q(A) could be the bar resolution Bar q(A) given by Barn(A) = A⊗(n+2) with differential a0 ⊗ a1 ⊗ · · · ⊗ an+1 7→ n X i=0 (−1)ia0 ⊗ · · · ⊗ aiai+1 ⊗ · · · ⊗ an+1 for all n ≥ 0 and a0, a1, . . . , an+1 ∈ A. We also use the reduced bar resolution Bar q(A) with Barn(A) = A ⊗ ¯A⊗n ⊗ A for ¯A = A/k1A and differential given by the same formula. Compatibility conditions. We now define what it means for resolutions to be compatible with the twisting map τ . We tensor arbitrary resolutions (2.15) and (2.14) with A and B on the right and left to obtain complexes P q(N ) ⊗ A, A ⊗ P q(N ), P q(M ) ⊗ B, and B ⊗ P q(M ) . Viewing these simply as exact sequences of vector spaces, we note that any k-linear maps τN,A : N ⊗ A → A ⊗ N and τB,M : B ⊗ M → M ⊗ B can be lifted to k-linear chain maps (2.16) τP q(N ),A : P q(N ) ⊗ A → A ⊗ P q(N ) and τB,P q(M ) : B ⊗ P q(M ) → P q(M ) ⊗ B . For simplicity in the sequel, we will write τi,A = τPi(N ),A and τB,i = τB,Pi(M ), for each i, when no confusion will arise. We will use such maps to glue the two resolutions together provided they satisfy the following compatibility conditions. These conditions just state that the chain maps commute with multiplication and with bimodule structure maps. There are many settings in which compatible chain maps do exist, as we will see. / / / 6 A.V. SHEPLER AND S. WITHERSPOON Definition 2.17. Let M be an A-bimodule that is compatible with τ . A projective A- bimodule resolution P q(M ) is compatible with the twisting map τ if each Pi(M ) is compatible with τ via a map τB,i : B ⊗ Pi(M ) −→ Pi(M ) ⊗ B with τB, q a chain map lifting τB,M . If A is graded, M is a graded A-bimodule, and P q(M ) is a graded projective A-bimodule resolution, we say that P q(M ) is compatible with a strongly graded twisting map τ if there are maps τB,i as above taking Bj ⊗ (Pi(M ))l to (Pi(M ))l ⊗ Bj for all j, l. Remark 2.18. The above definition applies to B-bimodule resolutions as well as A-bimodule resolutions by reversing the role of A and B in the definition, again as A ⊗τ B = B ⊗τ −1 A. For a B-bimodule N that is compatible with τ , the definition implies that a projective B-bimodule resolution P q(N ) of N is compatible with the twisting map τ when each Pi(N ) is compatible with τ via a map τi,A : Pi(N ) ⊗ A → A ⊗ Pi(N ), with τ q,A a chain map lifting τN,A. Thus we say a resolution is compatible with τ if it is either an A-bimodule resolution or a B-bimodule resolution compatible with τ . We give some small examples later: Example 2.21 (Weyl algebra) and Example 3.13 (skew group algebra). First a remark on embedding resolutions and some general results. Remark 2.19. Note that compatibility is preserved under embedding of resolutions so long as the extensions of the twisting map τ preserve the embedding. Specifically, assume φ q : Q q(A) ֒→ P q(A) is an embedding of resolutions of the algebra A, and P q(A) is compatible with a twisting map τ : B ⊗ A → A ⊗ B via chain maps τB,i : B ⊗ Pi(A) → Pi(A) ⊗ B. If the maps τB,i preserve the embedding in the obvious sense that each τB,i restricts to a surjective map B ⊗ Im(φ) ։ Im(φ) ⊗ B, then Q q(A) is compatible with τ via these restrictions. Compatibility of bar and Koszul resolutions. If A and B are both Koszul algebras and τ is a strongly graded twisting map, then the algebra A ⊗τ B is known to be Koszul (see [14, Example 4.7.3], [12, Corollary 4.1.9], or [19, Proposition 1.8]). Conner and Goetz [4] examine the situation when τ is not strongly graded. We show next that both bar and Koszul resolutions are compatible with twisting maps. We always assume our Koszul algebras are connected graded algebras, so that they are quotients of tensor algebras on generating vector spaces in degree 1. Note that the roles of A and B may be exchanged in the next proposition. Proposition 2.20. Let τ be a twisting map for some k-algebras A and B. (i) The bar resolution Bar q(A) is compatible with τ . (ii) The reduced bar resolution Bar q(A) is compatible with τ . (iii) If A is a Koszul algebra, B is a graded algebra, and τ is strongly graded, then the Koszul resolution Kos q(A) is compatible with τ . TWISTED TENSOR PRODUCTS 7 Proof. (i) The bar resolution of A may be twisted by repeated application of the map τ , e.g., define τB,i : B ⊗ A⊗(i+2) → A⊗(i+2) ⊗ B by applying τ to the first two tensor factors on the left, then applying τ to next two tensor factors, and so on: τB,i = (1 ⊗ · · · ⊗ 1 ⊗ τ )(1 ⊗ · · · ⊗ 1 ⊗ τ ⊗ 1) · · · (1 ⊗ τ ⊗ 1 ⊗ · · · ⊗ 1)(τ ⊗ 1 ⊗ · · · ⊗ 1). Then Bar q(A) is compatible with τ via τB,i, as may be verified directly by repeated use of equation (2.1). (ii) Write the terms in the bar complex Bar q(A) as Pi = A⊗(i+2) for each i, and define the terms in the reduced bar complex Bar q(A) by ¯Pi = A ⊗ ¯A⊗i ⊗ A. For each i, let Ti be the kernel of the quotient map Bari(A) → Bari(A). Then T q is a subcomplex of Bar q(A) and Bar q(A) ∼= Bar q(A)/T q. By definition of twisting map τ , the multiplicative identity 1A commutes with elements of B under τ , implying that τB,i of part (i) takes B ⊗Ti onto Ti ⊗B for each i. Let ¯τB,i : B ⊗ Bari(A) → Bari(A) ⊗ B be the corresponding map on quotients. Then Bar q(A) is compatible with τ via the maps ¯τB,i. (iii) The proof of [19, Proposition 1.8] shows that the embedding Kos q(A) ֒→ Bar q(A) of bimodule resolutions is preserved by the iterated twisting in part (i) above (see Re- mark 2.19). Thus Kos q(A) satisfies compatibility. (cid:3) We give an example next showing how Proposition 2.20 can be used for Koszul resolutions even when the twisting map τ is not strongly graded. Example 2.21. As in Example 2.2, let W be the Weyl algebra on x, y with A = k[x] and B = k[y]. Let Kos q(A) be the Koszul resolution of A as an A-bimodule, 0 → A ⊗ V ⊗ A d1−→ A ⊗ A m −→ A → 0 , where V = Spank{x} ⊂ A, d1(1 ⊗ x ⊗ 1) = x ⊗ 1 − 1 ⊗ x, and m is multiplication. Let ¯τ : B ⊗ V → V ⊗ B be the swap map b ⊗ v 7→ v ⊗ b for all b in B and v in V , and define by iterations of τ and ¯τ : ¯τB, q : B ⊗ Kos q(A) → Kos q(A) ⊗ B ¯τB,0 : B ⊗ A ⊗ A τ ⊗1 −−−−−−→ A ⊗ B ⊗ A 1⊗τ −−−−−−→ A ⊗ A ⊗ B , and ¯τB,1 : B ⊗ A ⊗ V ⊗ A τ ⊗1⊗1 −−−−−−→ A ⊗ B ⊗ V ⊗ A 1⊗¯τ ⊗1 −−−−−−→ A ⊗ V ⊗ B ⊗ A 1⊗1⊗τ −−−−−−→ A ⊗ V ⊗ A ⊗ B . Define ¯τ q,A : Kos q(B) ⊗ A → A ⊗ Kos q(B) similarly for the Koszul resolution Kos q(B) of B. Note that τ is not strongly graded, so part (iii) of Proposition 2.20 does not apply even though both A and B are Koszul algebras. Instead, we appeal to part (ii) and Remark 2.19 after taking canonical embeddings Kos q(A) ֒→ Bar q(A) and Kos q(B) ֒→ Bar q(B). (For example, view A ⊗ V ⊗ A as a subspace of A ⊗ ¯A ⊗ A; the terms in other degrees are either 0 or the same as in the bar resolution.) The maps ¯τB, q and ¯τ q,A above are the restrictions to B ⊗ Kos q(A) and Kos q(B) ⊗ A of the maps of the same name in the proof of Proposition 2.20(ii) (after identifying V with its image under the quotient map A → ¯A). In this way, we see that the Koszul resolutions Kos q(A) and Kos q(B) are compatible with the twisting map τ via ¯τB, q and ¯τ q,A. We extend these ideas in Theorem 4.2. 8 A.V. SHEPLER AND S. WITHERSPOON 3. Twisted product resolutions for Bimodules Again, we fix k-algebras A and B with a twisting map τ : B ⊗ A → A⊗ B and consider an A-bimodule M and B-bimodule N . We build a projective resolution of M ⊗N as a bimodule over A ⊗τ B from resolutions P q(M ) and P q(N ) under our compatibility assumptions. We give the construction in Lemma 3.1, prove exactness in Lemma 3.5, and show in Lemma 3.9 that the modules in the construction are indeed projective under an additional assumption. Lemma 3.1. Let M be an A-bimodule and let N be a B-bimodule, both compatible with a twisting map τ . Let P q(M ) and P q(N ) be projective A- and B-bimodule resolutions of M and N , respectively, that are compatible with τ . For each i, j ≥ 0, let (3.2) Xi,j = Pi(M ) ⊗ Pj(N ) , an A ⊗τ B-bimodule via diagram (2.13). Then X q, q is a bicomplex of A ⊗τ B-bimodules with horizontal and vertical differentials given by dh i,j = (−1)i ⊗ dj, where di and dj denote the differentials of the appropriate resolutions: i,j = di ⊗ 1 and dv ... X0,2 dv 0,2 X0,1 dv 0,1 X0,0 dh 1,2 dh 1,1 dh 1,0 ... X1,2 dv 1,2 X1,1 dv 1,1 X1,0 dh 2,2 dh 2,1 dh 2,0 ... X2,2 · · · dv 2,2 X2,1 · · · dv 2,1 X2,0 · · · Proof. The k-vector spaces Xi,j form a tensor product bicomplex with differentials as stated. The bimodule action of A⊗τB on Xi,j commutes with the horizontal and vertical differentials since τ q,B and τA, q are chain maps. Therefore this is an A ⊗τ B-bimodule bicomplex. (cid:3) Definition 3.3. The twisted product complex X q is the total complex of X q, q, i.e., when augmented by M ⊗ N , it is the complex (3.4) with Xn = ⊕i+j=nXi,j, and nth differential Pi+j=n di,j where di,j = di ⊗ 1 + (−1)i ⊗ dj. Lemma 3.5. The twisted product complex (3.4) is exact. · · · → X2 → X1 → X0 → M ⊗ N → 0 Proof. By the Kunneth Theorem [20, Theorem 3.6.3], for each n there is an exact sequence 0 −→ M Hi (cid:0)P q(M )(cid:1) ⊗ Hj (cid:0)P q(N )(cid:1) −→ Hn (cid:0)P q(M ) ⊗ P q(N )(cid:1) i+j=n −→ M i+j=n−1 Tork 1 (cid:16) Hi (cid:0)P q(M )(cid:1), Hj (cid:0)P q(N )(cid:1)(cid:17) −→ 0 . o o o o o o o o o o o o o o o o o o TWISTED TENSOR PRODUCTS 9 Now P q(M ) and P q(N ) are exact other than in degree 0, where they have homology M and N , respectively. Therefore Hi (cid:0)P q(M )(cid:1) = 0 for all i > 0 and Hj (cid:0)P q(N )) = 0 for all j > 0 . The Tor term is 0 since k is a field. Thus for all n > 0, Hn (cid:0)P q(M ) ⊗ P q(N )(cid:1) = 0, and H0 (cid:0)P q(M ) ⊗ P q(N )(cid:1) ∼= H0 (cid:0)P q(M )(cid:1) ⊗ H0 (cid:0)P q(N )(cid:1) ∼= M ⊗ N as vector spaces. Thus the complex (3.4) is exact. (cid:3) In practice, one often can show directly that each Xi,j is projective as an A⊗τB-bimodule, for example, when working with bar resolutions and/or Koszul resolutions. For the general case, we need an extra compatibility assumption, which we explain next. As each Pi(N ) is a projective B-bimodule, it embeds into a free Be-module (Be)⊕J for some indexing set J. In the following definition, we use the map (τ ⊗ 1)(1 ⊗ τ ) : Be ⊗ A → A ⊗ Be. Definition 3.6. A chain map τi,A : Pi(N ) ⊗ A → A ⊗ Pi(N ) is compatible with a chosen embedding Pi(N ) ֒→ (Be)⊕J (for some indexing set J) if the corresponding diagram is commutative: Pi(N ) ⊗ A ֒ (Be)⊕J ⊗ A τi,A ((τ ⊗1)(1⊗τ ))⊕J A ⊗ Pi(N ) ֒ / A ⊗ (Be)⊕J . Similarly, the map τB,i of (2.16) is compatible with a chosen embedding of Pi(M ) into a free Ae-module (Ae)⊕I (for some indexing set I) if the corresponding diagram is commutative, i.e., if τB,i is the restriction of the map ((1 ⊗ τ )(τ ⊗ 1))⊕I to B ⊗ Pi(M ). Remark 3.7. In many settings, one sees directly that each Xi,j is projective, in which case one need not consider this extra compatibility condition, as the next lemma is not needed. This is the case, for example, when twisting by a bicharacter on grading groups (see [1, Lemma 3.3]). In other settings, τi,A and τB,i are automatically compatible with chosen embeddings into free modules, for example if A and B are Koszul algebras and the embeddings are standard embeddings into bar resolutions (see [19, Proposition 1.8]). Example 3.8. As in Examples 2.2 and 2.21, let W ∼= A ⊗τ B be the Weyl algebra on x, y, A = k[x], and B = k[y]. By construction, each map ¯τi,A is compatible with the canonical embedding Kosi(A) ֒→ Bari(A) and likewise ¯τB,i is compatible with Kosi(B) ֒→ Bari(B). Lemma 3.9. If τB,i and τj,A are compatible with chosen embeddings of Pi(M ) and Pj(N ) into free modules, then Xi,j = Pi(M ) ⊗ Pj(N ) is a projective A ⊗τ B-bimodule. Proof. First we verify the lemma in case Pi(M ) = Ae, Pj(N ) = Be, and the chosen embed- dings are the identity maps. In this case, Xi,j = Ae ⊗ Be = A ⊗ Aop ⊗ B ⊗ Bop. One checks that the map 1 ⊗ τ ⊗ 1 : A ⊗ B ⊗ (A ⊗ B)op −→ A ⊗ Aop ⊗ B ⊗ Bop / / / 10 A.V. SHEPLER AND S. WITHERSPOON is an isomorphism of (A ⊗τ B)e-modules by equation (2.1) and the definition of the action given in the proof of Lemma 3.1. If Pi(M ) and Pj(N ) are arbitrary free modules, and the chosen embeddings are identity maps, we apply the above map to each summand Ae ⊗ Be of Pi(M ) ⊗ Pj(N ) to see that Xi,j is a free (A ⊗τ B)e-module. Now we consider the general case, including the possibility that at least one of Pi(M ), Pj(N ) is free but the corresponding chosen embedding into a (possibly different) free module is not the identity map. The first part of the proof together with the compatibility hypoth- esis implies that the embedding of k-vector spaces Pi(M )⊗ Pj (N ) ֒→ (Ae)⊕I ⊗ (Be)⊕J given by the tensor product of the two embedding maps is a map of (A ⊗τ B)e-modules. (cid:3) We combine the lemmas to obtain the following theorem. Theorem 3.10. Let A and B be k-algebras, and let τ : B ⊗ A → A ⊗ B be a twisting map. Let M be an A-bimodule and N a B-bimodule with projective A- and B-bimodule resolutions P q(M ) and P q(N ), respectively. Assume that M , N , P q(M ), and P q(N ) are compatible with τ and the corresponding maps τB,i and τj,A are compatible with chosen embeddings of Pi(M ) and Pi(N ) into free modules. Then the twisted product complex with Xn = ⊕i+j=n Xi,j for Xi,j = Pi(M ) ⊗ Pj(N ) gives a projective resolution of M ⊗ N as A ⊗τ B-bimodule: · · · → X2 → X1 → X0 → M ⊗ N → 0 . Proof. The result follows from Lemmas 3.1, 3.5, and 3.9. (cid:3) Remark 3.11. The theorem generally unifies known constructions of resolutions in several different contexts, for example, twisted tensor products given by bicharacters of grading groups [1], crossed products [9], skew group algebras (semidirect products) of Koszul alge- bras and finite groups [16], and smash products of Koszul algebras with Hopf algebras [18]. Theorem 3.10 combined with Proposition 2.20 and Remark 3.7 implies that a twisted product resolution of A ⊗τ B as a bimodule always exists, since bar resolutions may always be twisted (and likewise Koszul resolutions, when one or both of the algebras is Koszul, see also [12, 14, 19]): Corollary 3.12. Let A and B be k-algebras with twisting map τ : B ⊗ A → B ⊗ A. The following are projective resolutions of A ⊗τ B as a bimodule over itself. • The twisted product complex of two bar resolutions. • The twisted product complex of two Koszul resolutions when A and B are Koszul algebras and τ is strongly graded. • The twisted product complex of one bar resolution and one Koszul resolution in case one of A or B is Koszul and the other is graded, for τ strongly graded. Moreover, bar resolutions may be replaced by reduced bar resolutions in the above statements. TWISTED TENSOR PRODUCTS 11 Examples: Skew group algebras. We give some details for a class of examples intro- duced in Example 2.3. The resolutions in [16] for S ⋊ G, where G is a finite group acting by graded automorphisms on a Koszul algebra S, appear different from but are equivalent to (3.4) when M = kG (the group algebra) and N = S. Note that kG ⊗ S is isomorphic to S ⋊ G as an (S ⋊ G)-bimodule via the twisting map τ . In [16], the modules Xi,j are given as j ⊗ (S ⋊ G) i ⊗ D′ (S ⋊ G) ⊗ C ′ j⊗S are free (kG)e- and Se-modules determined where Pi(kG) = kG⊗C ′ i⊗kG, Pj(S) = S⊗D′ by vector spaces C ′ i, D′ j, respectively. We assume Pi(kG) is G-graded and the grading is compatible with the kG-bimodule action. We assume Pj(S) is a kG-module in such a way that the differentials are kG-module homomorphisms, and this action is compatible with that of S, so that Pj(S) becomes an S ⋊ G-module. Compatibility with τ follows from these assumptions. There is an isomorphism of S ⋊ G-bimodules, (kG ⊗ C ′ i ⊗ kG) ⊗ (S ⊗ D′ j ⊗ S) ∼ −→ (S ⋊ G) ⊗ C ′ i ⊗ D′ j ⊗ (S ⋊ G) , similar to that used in the proof of [16, Theorem 4.3], given by g ⊗ x ⊗ g′ ⊗ s ⊗ y ⊗ s′ 7→ g((hg′)s) ⊗ x ⊗ (g′y) ⊗ g′s′ for all g, g′ ∈ G, s, s′ ∈ S, x in the h-component of C ′ i, and y ∈ D′ j. Example 3.13. In particular, [16, Example 4.6] involves a resolution that is neither a Koszul resolution nor a bar resolution and yet satisfies compatibility. In that example, k is a field of positive characteristic p, S = k[x, y], and G = hgi is a group of order p acting on S by g · x = x, g · y = x + y. The resolution P q(S) is the Koszul resolution Kos q(S) of S, 0 → S ⊗ V2 V ⊗ S → S ⊗ V1 V ⊗ S → S ⊗ S → S → 0 , where V = Spank{x, y}. The resolution P q(kG) is the bimodule resolution of kG, (3.14) · · · η· −→ kG ⊗ kG γ· −→ kG ⊗ kG η· −→ kG ⊗ kG γ· −→ kG ⊗ kG m −→ kG −→ 0 , where γ = g ⊗ 1 − 1 ⊗ g, η = gp−1 ⊗ 1 + gp−2 ⊗ g + · · · + 1 ⊗ gp−1, and m is multiplication. Compatibility follows from Proposition 2.20(i) using Remark 2.19 after taking the standard embedding Kos q(S) ֒→ Bar q(S) and embedding (3.14) into Bar q(kG) (see, e.g., [3]). 4. Bimodule resolutions of Ore extensions Many algebras of interest are Ore extensions of other algebras. We show how to twist bimodule resolutions for such extensions in this section. Ore extensions as twisted tensor products. Let R be a k-algebra and fix a k-algebra automorphism σ of R. Let δ : R → R be a left σ-derivation of R, that is, (4.1) δ(rs) = δ(r)s + σ(r)δ(s) for all r, s ∈ R . The Ore extension R[x; σ, δ] is the algebra with underlying vector space R[x] and multipli- cation determined by that of R and of k[x] and the additional Ore relation xr = σ(r)x + δ(r) for all r ∈ R . 12 A.V. SHEPLER AND S. WITHERSPOON An Ore extension R[x; σ, δ] is thus isomorphic to a twisted tensor product A ⊗τ B where A = R, B = k[x], and the twisting map τ : B ⊗ A → A ⊗ B satisfies τ (x ⊗ r) = σ(r) ⊗ x + δ(r) ⊗ 1 for all r ∈ R . Free resolutions for iterated Ore extensions. We will work with general Ore extensions in Section 6. Here for simplicity we restrict to the case that the automorphism on R is the identity, σ = 1R, so the Ore relation sets commutators xr−rx equal to elements in R. In this case, the Ore extension is also known as a ring of formal differential operators. We consider an iterated Ore extension S = (· · · (k[x1][x2; 1, δ2]) · · · )[xt; 1, δt], which we abbreviate as S = k[x1, . . . , xt; δ2, . . . , δt] = khx1, . . . , xti/(xjxi − xixj − δj(xi) : 1 ≤ i < j ≤ t) with S ∼= k[x1, . . . , xt] as a k-vector space. We assume that S is a filtered algebra with deg(xi) = 1 for all i. Then each δj is a filtered map, i.e., δj(xi) ∈ k⊕k-span{x1, . . . , xj−1} for i < j. This setting includes Weyl algebras and universal enveloping algebras of supersolvable Lie algebras. Theorem 4.2. Consider an iterated Ore extension S = k[x1, . . . , xt; δ2, . . . , δt] with identity automorphisms σi = 1 and filtered derivations δi. There is an iterated twisted product resolution K q that is a free resolution of S as a bimodule over itself: Kn = S ⊗ Vn V ⊗ S for V = k-span{x1, . . . , xt} with differentials given by (for 1 ≤ l1 < · · · < ln ≤ t) dn(1 ⊗ xl1 ∧ · · · ∧ xln ⊗ 1) = X 1≤i≤n (−1)i+1(cid:0)xli ⊗ xl1 ∧ · · · ∧ xli ∧ · · · ∧ xln ⊗ 1 − 1 ⊗ xl1 ∧ · · · ∧ xli ∧ · · · ∧ xln ⊗ xli(cid:1) + X 1≤i<j≤n (−1)j ⊗ xl1 ∧ · · · ∧ xli−1 ∧ ¯δlj (xli) ∧ xli+1 ∧ · · · ∧ xlj ∧ · · · ∧ xln ⊗ 1 , where ¯δlj (xli) is the image of δlj (xli) under the projection k ⊕ V ։ V . Proof. We induct on t. For each i, the Koszul resolution of k[xi] is embedded in the (reduced) bar resolution of k[xi] as (4.3) 0 → k[xi] ⊗ Spank{xi} ⊗ k[xi] d1−→ k[xi] ⊗ k[xi] m −→ k[xi] → 0 , where d1(1⊗xi ⊗1) = xi ⊗1−1⊗xi and m is multiplication. For t = i = 1, the complex (4.3) is a resolution of S satisfying the statement of the theorem. Now assume t ≥ 2 and that the iterated Ore extension A = k[x1, . . . , xt−1; δ2, . . . , δt−1] has a free bimodule resolution P q(A) as in the theorem. Let B = k[xt] and let P q(B) be the Koszul resolution (4.3) for i = t. Then S = A ⊗τ B where τ (xt ⊗ a) = a ⊗ xt + δt(a) ⊗ 1 for all a ∈ A . TWISTED TENSOR PRODUCTS 13 Embedding into the reduced bar resolution. We embed P q(A) into the reduced bar resolution Bar q(A) and then define twisting maps for P q(A) via this embedding: Let φn : Pn(A) → A⊗(n+2) be the standard symmetrization map defined by φn(1 ⊗ xl1 ∧ · · · ∧ xln ⊗ 1) = X σ∈Symn sgn σ ⊗ xlσ(1) ⊗ · · · ⊗ xlσ(n) ⊗ 1 for all 1 ≤ l1 < · · · < ln ≤ t − 1. This is a chain map from P q(A) to Bar q(A). Compose with the quotient map Bar q(A) → Bar q(A) to obtain a chain map ¯φ q : P q(A) −→ Bar q(A) . Note that the image of P q(A) in the bar resolution Bar q(A), under φ q, intersects the kernel of this quotient map trivially. Thus the induced map ¯φ q is injective. Iterated twisting. The reduced bar resolution is compatibile with τ via the map from the proof of Proposition 2.20(ii). We argue that ¯τB, q restricts to a surjective map ¯τB, q : B ⊗ Bar q(A) −→ Bar q(A) ⊗ B τB, q : B ⊗ P q(A) −→ P q(A) ⊗ B by verifying that it preserves the image of ¯φ q, i.e., ¯τB,n takes B ⊗ Im( ¯φn) onto Im( ¯φn) ⊗ B for all n. We apply ¯τB,n to xt ⊗ ¯φn(a0 ⊗ y1 ∧ · · · ∧ yn ⊗ an+1) = X sgn π (xt ⊗ a0 ⊗ yπ(1) ⊗ · · · ⊗ yπ(n) ⊗ an+1), π∈Symn for some a0, an+1 in A, in order to move xt to the far right, obtaining (cid:16) X π∈Symn (sgn π) a0 ⊗ yπ(1) ⊗ · · · ⊗ yπ(n) ⊗ an+1(cid:17) ⊗ xt ∈ Im( ¯φn) ⊗ B plus additional terms that arise from the relation τ (xt ⊗yπ(i)) = yπ(i) ⊗xt +δt(yπ(i))⊗1. (We use the same notation for elements of A and their images under the quotient map A → ¯A in cases where no confusion can arise.) Since τ (1 ⊗ yj) = yj ⊗ 1 for all j, these additional terms sum to X π∈Symn (sgn π) δt(a0) ⊗ yπ(1) ⊗ · · · ⊗ yπ(n) ⊗ an+1 ⊗ 1 + X π∈Symn X 1≤i≤n (sgn π) a0 ⊗ yπ(1) ⊗ · · · ⊗ ¯δt(yπ(i)) ⊗ yπ(i+1) ⊗ · · · ⊗ yπ(n) ⊗ an+1 ⊗ 1 + X π∈Symn (sgn π) a0 ⊗ yπ(1) ⊗ · · · ⊗ yπ(n) ⊗ δt(an+1) ⊗ 1 = ¯φn(cid:0)δt(a0) ⊗ y1 ∧ · · · ∧ yn ⊗ an+1(cid:1) ⊗ 1 + ¯φn(cid:0)a0 ⊗ y1 ∧ · · · ∧ yn ⊗ δt(an+1)(cid:1) ⊗ 1 ¯φn(cid:0)a0 ⊗ y1 ∧ · · · ∧ ¯δt(yi) ∧ yi+1 ∧ yn ⊗ an+1(cid:1) ⊗ 1 ∈ Im( ¯φn) ⊗ B . + X 1≤i≤n We may replace xt by xm t τ (xm t ⊗ xi) = (1 ⊗ mB)τ (xt ⊗ (cid:0)τ (xm−1 in the above computation using induction after noting that ⊗ xi)(cid:1) for i < t. The above arguments can be t 14 A.V. SHEPLER AND S. WITHERSPOON modified to apply to ¯τ −1 restricts to a surjective chain map τB, q : B ⊗ P q(A) −→ P q(A) ⊗ B as claimed. B,i as well. Thus the chain map ¯τB, q preserves the image of ¯φ q and Compatibility on one side. The complex P q(A) inherits compatibility with τ from the compatibility of the reduced bar complex Bar q(A) with τ . Indeed, since Bar q(A) is com- patible with τ via a map ¯τB, q which preserves the embedding ¯φ q : P q(A) ֒→ Bar q(A), the complex P q(A) is compatible with τ via the restriction τB, q of ¯τB, q to B ⊗ P q(A). (See Proposition 2.20(ii) and its proof and Remark 2.19.) Compatibility on the other side. Define a chain map τ q,A : P q(B) ⊗ A → A ⊗ P q(B) by setting τ0,A = (τ ⊗ 1)(1 ⊗ τ ) and τ1,A((1 ⊗ xt ⊗ 1) ⊗ xi) = xi ⊗ (1 ⊗ xt ⊗ 1) and then extending (uniquely) to P1(B)⊗ A by requiring that compatibility conditions (2.8) and (2.9) hold. A calculation shows that τ q,A is a chain map and that P q(B) is compatible with τ . By their definitions, τ0,A and τ1,A are compatible with the embeddings of P0(B) and P1(B) into corresponding terms of the (reduced) bar resolution. Twisted product resolution. By construction, the twisted product resolution K q arising from P q(A) and P q(B) in degree n is isomorphic to S ⊗ Vn V ⊗ S as an S-bimodule via the isomorphisms A ⊗ Vi Spank{x1, . . . , xt−1} ⊗ A ⊗ B ⊗ Vj Spank{xt} ⊗ B ∼ −→ A ⊗ B ⊗ Vi Spank{x1, . . . , xt−1} ⊗ Vj Spank{xt} ⊗ A ⊗ B , for j = 0, 1, given by applying τ −1 (properly interpreted for each factor) to the innermost tensor factors A and B. We check the differentials: On Xn,0, the differential is just that arising from the factor Pn(A). Now consider on Xn−1,1, again writing xli = yi for some indices 1 ≤ l1 < · · · < ln ≤ t − 1: dn(1 ⊗ y1 ∧ · · · ∧ yn−1 ⊗ 1 ⊗ 1 ⊗ xt ⊗ 1) = (cid:16) X 1≤i≤n−1 (−1)i+1(cid:0)yi ⊗ y1 ∧ · · · yi ∧ · · · ∧ yn−1 ⊗ 1 − 1 ⊗ y1 ∧ · · · ∧ yi ∧ · · · ∧ yn−1 ⊗ yi(cid:1) + X 1≤i<j≤n−1 (−1)j ⊗ y1 ∧ · · · ∧ ¯δj(yi) ∧ · · · ∧ yj ∧ · · · ∧ yn−1 ⊗ 1(cid:17) ⊗ (1 ⊗ xt ⊗ 1) + (−1)n−1(1 ⊗ y1 ∧ · · · ∧ yn−1 ⊗ 1) ⊗ (xt ⊗ 1 − 1 ⊗ xt) , which may be rewritten, under the above isomorphism, as TWISTED TENSOR PRODUCTS 15 X 1≤i≤n−1 (−1)i+1yi ⊗ y1 ∧ · · · ∧ yi ∧ · · · ∧ yn−1 ⊗ xt ⊗ 1 − X 1≤i≤n−1 (−1)i+1 ⊗ y1 ∧ · · · ∧ yi ∧ · · · ∧ yn−1 ⊗ xt ⊗ yi + X 1≤i<j≤n−1 (−1)j ⊗ y1 ∧ · · · ∧ ¯δj(yi) ∧ · · · ∧ yj ∧ · · · ∧ yn−1 ⊗ xt ⊗ 1 + (−1)n−1xt ⊗ y1 ∧ · · · ∧ yn−1 ⊗ 1 + (−1)n ⊗ y1 ∧ · · · ∧ yn−1 ⊗ xt + (−1)n X 1≤i≤n−1 1 ⊗ y1 ∧ · · · ∧ ¯δt(yi) ∧ · · · ∧ yn−1 ⊗ 1 . Once one sets yn = xt, identifies y1 ∧ · · · ∧ yn−1 ⊗ xt with y1 ∧ · · · ∧ yn−1 ∧ xt, and makes other similar identifications, this agrees with the differential in the statement. (cid:3) Examples. The theorem applies in particular to the universal enveloping algebra U(g) of a finite dimensional solvable Lie algebra g. Here, we assume the underlying field k is algebraically closed, else g should be supersolvable; see [6, 1.3.14] and [2, Section 3]. The theorem gives a bimodule Koszul resolution of U(g). Semisimple Lie algebras can then be handled via triangular decomposition. Other examples include Weyl algebras and Sridharan enveloping algebras [17]. 5. Twisted product resolutions for (left) modules We now consider a twisted product resolution of left modules instead of bimodules. We give the one-sided version of bimodule constructions in Sections 2 and 3. Again, we fix k-algebras A and B with a twisting map τ : B ⊗ A → A ⊗ B. In the constructions below, we consider compatible A-modules, but note that we as easily could have started with compatible B-modules instead of A-modules using the inverse twisting map τ −1 instead of τ in order to lift (left) modules of A and B to (left) modules of A ⊗ B = B ⊗τ −1 A. Let M be an A-module with module structure map ρA,M : A ⊗ M → M and recall the multiplication map mB : B ⊗ B → B. Definition 5.1. The A-module M is compatible with the twisting map τ if there is a bijective k-linear map τB,M : B ⊗ M → M ⊗ B such that (5.2) (5.3) τB,M (mB ⊗ 1) = (1 ⊗ mB)(τB,M ⊗ 1)(1 ⊗ τB,M ) and τB,M (1 ⊗ ρA,M ) = (ρA,M ⊗ 1)(1 ⊗ τB,M )(τ ⊗ 1) as maps on B ⊗ B ⊗ M and on B ⊗ A ⊗ M , respectively. Note that this definition is equivalent to the commutativity of a diagram similar to (2.11), where ρA,M is replaced by a one-sided module structure map. 16 A.V. SHEPLER AND S. WITHERSPOON Let N be a B-module with module structure map ρB,N : B ⊗ N → N . In case M is compatible with τ , the tensor product M ⊗ N may be given the structure of an A ⊗τ B- module via the following composition of maps: (5.4) A⊗τ B ⊗M ⊗N 1⊗ τB,M ⊗1 / A⊗M ⊗B ⊗N ρA,M ⊗ρB,N / M ⊗ N . Let P q(M ) be an A-projective resolution of M and P q(N ) a B-projective resolution of N : · · · → P2(M ) → P1(M ) → P0(M ) → k → 0 , · · · → P2(N ) → P1(N ) → P0(N ) → k → 0 . Definition 5.5. Let M be an A-module that is compatible with τ . The projective module resolution P q(M ) of the A-module M is compatible with the twisting map τ if each Pi(M ) is compatible with τ via maps τB,i for which τB, q : B ⊗ P q(M ) → P q(M ) ⊗ B is a k-linear chain map lifting τB,M : B ⊗ M → M ⊗ B. Under the assumption of compatibility, we make the following definition. Definition 5.6. Let M be an A-module compatible with τ and P q(M ) a projective resolu- tion of M that is compatible with τ . Let N be a B-module. The twisted product complex Y q is the total complex of the bicomplex Y q, q defined by (5.7) Yi,j = Pi(M ) ⊗ Pj(N ) , with A ⊗τ B-module structure given by the maps τB, q as in equation (5.4) and with vertical and horizontal differentials given by dh i,j = (−1)i ⊗ dj. In other words, Yn = ⊕i+j=nYi,j with dn = Pi+j=n di,j where di,j = dh i,j = di ⊗ 1 and dv i,j + dv i,j. Lemma 5.8. Assume M and P q(M ) are compatible with τ . Then the twisted product complex Y q is a complex of A ⊗τ B-modules. Proof. Each space Yi,j is given the structure of an A ⊗τ B-module via diagram (5.4). The differentials are module homomorphisms since τB, q is a chain map. (cid:3) Lemma 5.9. The twisted product complex · · · → Y2 → Y1 → Y0 → M ⊗ N → 0 is exact. Proof. As in the proof of Lemma 3.5, apply the Kunneth Theorem to obtain Hn(Y q) = 0 for all n > 0 and H0(Y q) ∼= M ⊗ N . (cid:3) We wish to prove in general that the modules Yi,j are projective, so we make an additional assumption in the next lemma. Since P q(M ) is a projective resolution of M as an A-module, each Pi(M ) embeds in a free A-module A⊕I . Definition 5.10. For each i ≥ 0, the map τB,i is compatible with a chosen embedding Pi(M ) ֒→ A⊕I (for some indexing set I) if the corresponding diagram is commutative: B ⊗ Pi(M ) ֒ τB,i B ⊗ A⊕I τ ⊕I Pi(M ) ⊗ B ֒ / A⊕I ⊗ B . / / / / / TWISTED TENSOR PRODUCTS 17 In many settings, one proves directly that the modules Yi,j are projective -- e.g. the Ore extensions in the next section -- and so one does not need this additional compatibility assumption, nor the next lemma. Lemma 5.11. For i ≥ 0, if τB,i is compatible with a chosen embedding of Pi(M ) into a free A-module, then Yi,j = Pi(M ) ⊗ Pj(N ) is a projective A ⊗τ B-module. Proof. By the hypothesis, it suffices to prove the lemma in case Pi(A) = A and Pj(B) = B. In that case, A ⊗ B is the right regular module A ⊗τ B by definition, and so is free. (cid:3) Combining Lemmas 5.8, 5.9, and 5.11, we obtain the following theorem. Theorem 5.12. Let A and B be k-algebras with twisting map τ : B⊗A → A⊗B. Let P q(M ) and P q(N ) be projective A- and B-module resolutions of M and N , respectively. Assume M and P q(M ) are compatible with τ and that the corresponding maps τB,i are compatible with chosen embeddings of Pi(M ) into free A-modules. Then the twisted product complex with Yn = ⊕i+j=nYi,j for Yi,j = Pi(M ) ⊗ Pj(N ) gives a projective resolution of M ⊗ N as a module over the twisted tensor product A ⊗τ B: · · · → Y2 → Y1 → Y0 → M ⊗ N → 0 . Examples. Resolutions that may be constructed in this way include the Koszul resolution of k for a twisted tensor product of two Koszul algebras (see the proof of [19, Proposi- tion 1.8]) and a resolution for a twisted tensor product of algebras whose twisting map is given by a bicharacter on grading groups (see [1]). We give another class of examples in the next section. 6. Resolutions for Ore extensions In Section 4, we considered resolutions of an Ore extension algebra as a bimodule over itself. Here, we consider (left) modules over an Ore extension and show how to construct projective resolutions of these modules by regarding the Ore extension as a twisted tensor product. Gopalakrishnan and Sridharan [7] studied Ore extensions R[x; σ, δ] in case σ is the identity automorphism. They showed that if M is a (left) module over R[x; 1, δ], then an R-projective resolution of M lifts to an R[x; 1, δ]-projective resolution. Here we allow arbitrary automorphisms σ of R and give conditions under which an R-projective resolution of an R[x; σ, δ]-module M lifts to an R[x; σ, δ]-projective resolution. Again, let R be a k-algebra and σ a k-algebra automorphism of R. Let δ be a left σ- derivation of R (see (4.1)) and consider the Ore extension R[x; σ, δ]. Let A = R, B = k[x], and τ : B ⊗ A → A ⊗ B be the twisting map determined by τ (x ⊗ r) = σ(r) ⊗ x + δ(r) ⊗ 1 for all r ∈ R, as in Section 4, so that R[x; σ, δ] is the twisted tensor product A ⊗τ B. 18 A.V. SHEPLER AND S. WITHERSPOON Modules over Ore extensions. Consider an R[x; σ, δ]-module M . Assume that on re- striction to R, there is an isomorphism of R-modules, φ : M ∼−→ M σ, where M σ is the vector space M with R-module action given by r ·σ m = σ(r) · m for all r ∈ R and m ∈ M . Then M is compatible with τ : We define τB,M := B ⊗ M → M ⊗ B by setting τB,M (1 ⊗ m) = m ⊗ 1 , τB,M (x ⊗ m) = φ(m) ⊗ x + xm ⊗ 1 for all m ∈ M and extending by applying compatibility condition (5.2). That is, since the algebra B = k[x] is free on the generator x, for each element m of M , we may define τB,M (xn ⊗m) by applying (5.2) to x ⊗ xn−1 ⊗ m. We check that (5.3) holds for elements of the form x ⊗ r ⊗ m, where r ∈ R and m ∈ M . Then a careful induction on the power of x shows that (5.3) holds for all elements of the form xn ⊗ r ⊗ m. For example, if R[x; σ, δ] is an augmented algebra with augmentation ε : R[x; σ, δ] → k for which εσ = ε, then εδ = 0 and the field k as a module over R[x; σ, δ] via ε has the property that k ∼= kσ, and so k is compatible with τ . Projective resolutions. Let P q(M ) be a projective resolution of M as an R-module: For each i, set P σ · · · d2−→ P1(M ) i (M ) = (Pi(M ))σ. Then d1−→ P0(M ) µ −→ M → 0. · · · d2−→ P σ 1 (M ) d2−→ P σ 0 (M ) φ−1µ −−−→ M → 0 is also a projective resolution of M as an R-module. By the Comparison Theorem, there is an R-module chain map from P q(M ) to P σ q (M ) lifting the identity map M → M , which we view as a k-linear chain map (6.1) σ q : P q(M ) → P q(M ) with σi(rz) = σ(r)σi(z) for all i ≥ 0, r ∈ R, and z ∈ Pi(M ). We will assume for Theorem 6.6 below that each σi is bijective. Let P q(B) be the Koszul resolution of k for B = k[x], (6.2) 0 → k[x] x· −→ k[x] ǫ −→ k → 0 , where ǫ(x) = 0. The following two lemmas are proven as in [7] (where Gopalakrishnan and Sridharan proved the special case σ = 1). We include details for completeness. Lemma 6.3. Let P be a projective R-module. There is an R[x; σ, δ]-module structure on P that extends the action of R. Proof. First consider the case that P = R, the left regular module. Let x act on R by x · r = δ(r) for all r ∈ R. One checks that the action of xr in R[x; σ, δ] agrees with that of σ(r)x + δ(r) on P , for all r ∈ R. Next, if P is a free module, it is a direct sum of copies of R, and x acts on each copy in this way. Finally, in general, P is a direct summand of a free R-module F . Let ι : P → F and π : F → P be R-module homomorphisms for which πι is the identity map. Define x · p = π(x · ι(p)) for all p ∈ P , where the action of x on ι(p) is as given previously for a free module. Again one checks that the actions of xr and of σ(r)x + δ(r) agree, and so P is an R[x; σ, δ]-module as claimed. (cid:3) TWISTED TENSOR PRODUCTS 19 Compatibility requirements. We will use the next lemma to show that the resolution P q(M ) of M as an R-module is compatible with the twisting map τ (see Lemma 6.5). Let f : M → M be the function given by the action of x on the R[x; σ, δ]-module M . Lemma 6.4. There is a k-linear chain map δ q : P q(M ) → P q(M ) lifting f : M → M such that for each i ≥ 0, δi(rz) = σ(r)δi(z) + δ(r)z for all r ∈ R and z ∈ Pi(M ). Proof. If i = 0, let δ′ 0 be the action of x on P0(M ) given by Lemma 6.3. Then δ′ 0(rz) − σ(r)δ′ 0(z) = δ(r)z 0 − f µ : P0(M ) → M σ is an R-module ho- for r ∈ R, z ∈ P0(M ). One checks that µδ′ momorphism. As P0(M ) is a projective R-module, there is an R-module homomorphism 0 : P0(M ) → P σ δ′′ 0 (M ) such that µδ′ 0 . One may check this satisfies the equation in the lemma. 0 . Let δ0 = δ′ 0 − f µ = µδ′′ 0 − δ′′ Now fix i > 0 and assume there are k-linear maps δj : Pj(M ) → Pj(M ) such that δj(rz) = σ(r)δj(z) + δ(r)z and djδj = δj−1dj for all j, 0 ≤ j < i, and r ∈ R, z ∈ Pj(M ). Let δ′ i(rz) = σ(r)δ′ i : Pi(M ) → Pi(M ) be the action of x on Pi(M ) given in Lemma 6.3, so that δ′ i(z) + δ(r)z for all r ∈ R, z ∈ Pi(M ). Consider the map diδ′ i − δi−1di : Pi(M ) −→ P σ i−1(M ) . A calculation shows that it is an R-module homomorphism. Since δi−1 is a chain map, di−1(diδ′ i − δi−1di) = 0 , and so the image of diδ′ R-module, there is an R-homomorphism δ′′ Let δi = δ′ z ∈ Pi(M ), i − δi−1di lies in Ker (di−1) = Im(di). Since Pi(M ) is projective as an i−δi−1di = diδ′′ i . i , so that diδi = δi−1di by construction. One checks that for all r ∈ R and i (M ) such that diδ′ i : Pi(M ) → P σ i − δ′′ δi(rz) = δ′ i (z) = σ(r)δi(z) + δ(r)z . i(rz) − δ′′ i (rz) = σ(r)δ′ i(z) + δ(r)z − σ(r)δ′′ (cid:3) Lemma 6.5. The resolution P q(M ) is compatible with the twisting map τ . Proof. Define τB,i : B ⊗ Pi(M ) → Pi(M ) ⊗ B by τB,i(1 ⊗ z) = z ⊗ 1 , τB,i(x ⊗ z) = σi(z) ⊗ x + δi(z) ⊗ 1 for all z ∈ Pi(M ) , where σ q is the chain map of (6.1), δ q is the chain map of Lemma 6.4, and we extend τB,i to B ⊗ Pi(M ) as before by requiring that compatibility conditions (5.2) and (5.3) hold. We check condition (5.3) in one case as an example: τB,i(x ⊗ rz) = σi(rz) ⊗ x + δi(rz) ⊗ 1 = σ(r)σi(z) ⊗ x + σ(r)δi(z) ⊗ 1 + δ(r)z ⊗ 1 , for all r ∈ R, and z ∈ Pi(M ), while on the other hand, (ρA,i ⊗ 1)(1 ⊗ τB,i)(τ ⊗ 1)(x ⊗ r ⊗ z) = (ρA,i ⊗ 1)(1 ⊗ τB,i)(σ(r) ⊗ x ⊗ z + δ(r) ⊗ 1 ⊗ z) = (ρA,i ⊗ 1)(σ(r) ⊗ σi(z) ⊗ x + σ(r) ⊗ δi(z) ⊗ 1 + δ(r) ⊗ z ⊗ 1) = σ(r)σi(z) ⊗ x + σ(r)δi(z) ⊗ 1 + δ(r)z ⊗ 1 ; 20 A.V. SHEPLER AND S. WITHERSPOON Condition (5.3) holds for all xn ⊗ rz by induction on n. (cid:3) Twisting resolutions for an Ore extension. We now construct a projective resolution of M as an R[x; σ, δ]-module from a projective resolution of M as an R-module. We take the twisted product of two resolutions: the R-projective resolution of M and the Koszul resolution (6.2) of k as a module over B = k[x]. Theorem 6.6. Let R[x; σ, δ] be an Ore extension. Let M be an R[x; σ, δ]-module for which M σ ∼= M as R-modules. Consider a projective resolution P q(M ) of M as an R-module and suppose that each map σi : Pi(M ) → Pi(M ) of (6.1) is bijective. For each i ≥ 0, set Yi,0 = Yi,1 = Pi(M ) ⊗ k[x] and Yi,j = 0 for all j > 1 as in Lemma 5.8. Then Y q is a projective resolution of M as an R[x; σ, δ]-module. Proof. By Lemma 6.5, P q(M ) is compatible with τ , and so by Lemmas 5.8 and 5.9, the complex · · · → Y1 → Y0 → M → 0 is an exact complex of R[x; σ, δ]-modules. We verify directly that each Yi,j is a projective module: For each i ≥ 0 and j = 0, 1, (6.7) Yi,j ∼= R[x; σ, δ] ⊗R Pi(M ) via the R[x; σ, δ]-homomorphism given by R[x; σ, δ] ⊗R Pi(M ) −→ Yi,j , x ⊗ z 7→ σi(z) ⊗ x + δi(z) ⊗ 1 , with inverse map given by z ⊗ x 7→ x ⊗ σ−1 i (z) − 1 ⊗ δi(σ−1 i (z)) . Then R[x; σ, δ] ⊗R Pi(M ) is projective since it is a tensor-induced module and R[x; σ, δ] is flat over R. (cid:3) Remark 6.8. When σ is the identity, the complex Y q is precisely that of Gopalakrishnan and Sridharan [7, Theorem 1], under the isomorphism (6.7) above. As a specific class of examples, we obtain in this way, via iterated Ore extension, the Chevalley-Eilenberg resolution of the U(g)-module k for a finite dimensional supersolvable Lie algebra g. Acknowledgments. The authors thank Andrew Conner and Peter Goetz for helpful com- ments on an earlier version of this paper, and a referee for suggesting improvements. References [1] P. A. Bergh and S. Oppermann, "Cohomology of twisted tensor products," J. Algebra 320 (2008), 3327 -- 3338. [2] K.A. Brown, S. O'Hagan, J.J. Zhang, and G. Zhuang, "Connected Hopf algebras and iterated Ore extensions", J. Pure Appl. Algebra 219 (2015), no. 6, 2405 -- 2433. [3] The Buenos Aires Cyclic Homology Group, "Cyclic homology of algebras with one generator," K- Theory 5 (1991), 51 -- 69. [4] A. Conner and P. Goetz, "The Koszul Property for graded twisted tensor products," submitted. See https://arxiv.org/pdf/1708.02514.pdf. [5] A. Cap, H. Schichl, and J. Vanzura, "On twisted tensor products of algebras," Comm. Algebra 23 (1995), no. 12, 4701 -- 4735. TWISTED TENSOR PRODUCTS 21 [6] J. Dixmier, Enveloping Algebras. (English Summary) Revised reprint of the 1977 translation. Graduate Studies in Mathematics, 11. American Mathematical Society, Providence, RI, 1996. xx+379 pp. ISBN: 0-8218-0560-6. [7] N. S. Gopalakrishnan and R. Sridharan, "Homological dimension of Ore-extensions," Pacific J. Math. 19 (1966), no. 1, 67 -- 75. [8] J. A. Guccione and J. J. Guccione, "Hochschild homology of twisted tensor products," K-Theory 18 (1999), 363 -- 400. [9] J. A. Guccione and J. J. Guccione, "Hochschild (co)homology of Hopf crossed products," K-Theory 25 (2002), no. 2, 139 -- 169. [10] P. J. Hilton and U. Stammbach, A Course in Homological Algebra, Springer-Verlag, New York, 1971. [11] Yu. I. Manin, Quantum Groups and Noncommutative Geometry, Universit´e de Montr´eal, Centre de Recherches Math´ematiques, Montr´eal, 1988. [12] P. Jara Martinez, J. L´opez Pina, and D. S¸tefan, "Koszul pairs and applications," arXiv:1011.4243. [13] S. Montgomery, Hopf Algebras and Their Actions on Rings, 82, Regional Conference Series in Mathe- matics, AMS, Providence, RI, 1993. [14] A. Polishchuk and L. Positselski, Quadratic Algebras, University Lecture Series 37, Amer. Math. Soc., 2005. [15] D. E. Radford and H. J. Schneider, "Biproducts and two-cocycle twists of Hopf algebras," Modules and Comodules, 331 -- 355, Trends Math., Birkhauser Verlag, Basel, 2008. [16] A.V. Shepler and S. Witherspoon, "A Poincar´e-Birkhoff-Witt Theorem for quadratic algebras with group actions," Trans. Amer. Math. Soc. 366 (2014), no. 12, 6483 -- 6506. [17] R. Sridharan, "Filtered algebras and representations of Lie algebras," Trans. Amer. Math. Soc. 100 (1961), 530 -- 550. [18] C. Walton and S. Witherspoon, "Poincar´e-Birkhoff-Witt deformations of smash product algebras from Hopf actions on Koszul algebras," Algebra and Number Theory 8 (2014), no. 7, 1701 -- 1731. [19] C. Walton and S. Witherspoon, "PBW deformations of braided products," J. Algebra 504 (2018), 536 -- 567. [20] C. A. Weibel, An Introduction to Homological Algebra, Cambridge Studies in Advanced Mathematics, volume 38, Cambridge University Press, Cambridge, 1994. Department of Mathematics, University of North Texas, Denton, Texas 76203, USA E-mail address: [email protected] Department of Mathematics, Texas A&M University, College Station, Texas 77843, USA E-mail address: [email protected]
1801.05774	1	1801	2018-01-16T16:49:45	Product of three octonions	[ "math.RA" ]	This paper is devoted to octonions that are the eight-dimensional hypercomplex numbers characterized by multiplicative non-associativity. The decomposition of the product of three octonions with the conjugated central factor into the sum of mutually orthogonal anticommutator, commutator and associator, is introduced in an obvious way by commuting of factors and alternating the multiplication order. The commutator is regarded as a generalization of the cross product to the case of three arguments both for quaternions and for octonions. It is verified that the resulting additive decomposition is equivalent to the known solution derived and presented by S. Okubo in a cumbersome form.	math.RA	math	UDC 511.2 M.V. KHARINOV PRODUCT OF THREE OCTONIONS Khariniv M.V. Product of three octonions. Abstract. This paper is devoted to octonions that are the eight-dimensional hyper- complex numbers characterized by multiplicative non-associativity. The decomposition of the product of three octonions with the conjugated central factor into the sum of mutually orthogo- nal anticommutator, commutator and associator, is introduced in an obvious way by commut- ing of factors and alternating the multiplication order. The commutator is regarded as a gener- alization of the cross product to the case of three arguments both for quaternions and for oc- tonions. It is verified that the resulting additive decomposition is equivalent to the known solu- tion derived and presented by S. Okubo in a cumbersome form. Keywords: quaternions, octonions, triple cross product, additive decomposition, mutually orthogonal terms. 1. Introduction. The famous generalization of real numbers by W.R. Hamilton surpassed by time the introduction of conventional cross product in mathematics and the discovery of four-dimensional space-time in physics [1]. Nowadays many applications of hypercomplex numbers are spawned not only in the field of classical physics and mathematics, but also in variety branches of modern science and technology, where the utilization of quater- nions is constantly expanding. Nevertheless, according to opinions of many experts, the potential of four-dimensional quaternions and especially capa- bilities of their eight-dimensional generalization (octonions) are far from been exhausted. Unlike the quaternions, the treatment of non-associative octonions, although occurs [2], but comparatively much less frequently, since it is nec- essary to deal with non-associative vector multiplication. A possible way to solve the problem is to generalize the basic method of additive decomposi- tion of the product of hypercomplex numbers into mutually orthogonal symmetric and antisymmetric components, which is the topic of the paper. The paper is intended for a reader familiar with the hypercomplex numbers (normalized algebras with multiplicative unit). Otherwise, the gen- eral information on hypercomplex numbers can be obtained from Wikipe- dia. All the necessary information in an accessible form is set forth in the popular book [3]. Many useful rules for working with hypercomplex num- bers are given in [4]. In this paper we list only the most relevant formulas. Unnumbered formulas are given mostly to explain the meaning of notations. 2. Elementary information. By 0i we denote the multiplicative unit that commutes with any hypercomplex number and, when multiplied, leaves it unchanged: Let u denotes the hypercomplex number which is conjugate to the = ui = 0 ui 0 u . number u , and is related to u by the formula [4]: u = ( ) i iu 00,2 − u , . wherein ( )0,iu is the inner product of vector u and vector 0i )uu, In four algebras of hypercomplex numbers (real numbers, complex numbers, quaternions and octonions), the square of the length ( of the vector u is introduced as the product of the vector u and the conjugate 2u coincides with 1u and vector u , and the inner product of the vectors 12uu : the half-sum of the vector and the conjugate vector ( ) 02 iuu , 1 uu ≡ From the last relations it is easy to establish the useful equality: 21uu uu 21 ( ) uuu uuu 2 = 121 1 1 121 uuu is written without brackets, since it does not wherein the product depend on the order of multiplication of hypercomplex factors: + 2 ( ) 2 uuu 1 ( ) iuu , 0 uu 12 uu 12 uu 21 (1) + 2 uu ⇔ ≡ = = − . , , , 1 2 In order to reliably relate to multiplication of quaternions and oc- uuu 121 = = . )12 uuu 1 ( ( ) uuu 1 21 tonions, it is useful to keep in mind the following elementary formulas. So, the next formula expresses that conjugation changes the order of factors: uu 21 And the following formula for the inner product ( vector and vector 0i expresses the transfer rule of the factor with simultane- 21uu ous conjugation, from one to another part of the inner product: ,iuu 21 uu 12 )0 = . ≡ ( iuu , 12 0 ) = ( uu , 1 )2 . ( uuu 32 1 ( ) , i ) 0 ≡ 2u . So, the prod- written without brackets implies choosing either of the two al- states that the order of multiplying of three hypercomplex factors and uct ternative ways of their arrangement, just as in left part of (1). 3u does not affect the inner product of ( 321 uuu ) 3 uuu 21 and 0i 1u , 0 ( iuuu , 321 )0 ( iuu , 21 ) 0 Finally, the formula: ) = ( ( iuuu , 21 ) 3 3. Additive decomposition of the product of two hypercomplex numbers. 21uu The product of two hypercomplex numbers 1u and is decomposed into the sum of two mutually orthogonal terms { [ 1,uu ]2 : 2u usually 1,uu and }2 uu 21 = uu 21 + 2 uu 12 uu 21 + uu 12 − 2 = { uu , 1 2 } [ + uu , 1 ]2 , (2) 2 2 }2 } [ , ] ) 0 = uu , 1 uu , 1 , the anticommutator ( wherein { { 1,uu , is a linear combination of the arguments tive unit 0i { uu , 1 ( ) uiu , 1 02 : } uu 12 uu 21 + = 2 2 + 2 uu 21 = ( ) uiu , 01 uu 12 and the commutator product of a pair of hypercomplex numbers 1u and , which is denoted as [ 2u . − 2 uu 21 1u , uu 12 + , denoted as 2 2u , and multiplica- − ( ) 02 iuu 1 , , (3) 1,uu ]2 , is a cross Taking into account (3) the product of two hypercomplex numbers 21uu , is expressed in expanded form by the formula [5]: [ ]2 ( ) ) ( uu uiu uiu , , , + 01 1 1 02 2 Cross product [ ]2 1,uu of hypercomplex numbers ( ) iuu 02 1 uu 21 − + = , . 1u and 2u is or- and vanishes to zero when the multiplica- (4) thogonal to multiplicative unit 0i tive unit 0i is used as one or another argument: 0 , . ] 0 = In other respects, the cross product [ 1,uu [ uu , ( 2 1 [ ] ui , = 0 ] =i ) , 0 [ iu , 01 2 ]2 preserves the properties of the conventional cross product, which is introduced in three-dimensional space using intuitively perceived "right hand rule". It means that, if we agree to denote by the prime an annihilating of the real component of hy- percomplex number u : u u −=′ ]2 then the cross product [ of the vectors 1u and 1,uu conventional cross product [ 1,uu ′ dimensional subspace that is orthogonal to multiplicative unit 0i 2u coincides with the 2u′ of the three- . In particu- of the vectors 1u′ and ]2 ′ ( iu 0, ) 0 i , lar, the square of the length of the cross product is expressed by the sym- metric determinant from the inner products of said three-dimensional vec- tors: [ ( uu , 1 ] [ uu , , ′ 1 [ ( uu , ′ 1 ] ) =′ 2 det . ] ) = 2 2 ( uu , ′ 1 ( uu , ′ 1 ′ 1 ′ 2 ) ) ⎡ ⎢ ⎣ ( uu , ′ ′ 1 2 ) ( uu , ′ ′ 2 ) 2 ⎤ ⎥ ⎦ 2 ] [ uu , , 1 ( uu , ′ ′ ⎡ 1 1 ⎢ ( uu , ′ ′ ⎣ 1 2 ) ) ′ 2 ) ⎤ ( uuuu , ′=⎥ ) 1 ⎦ )( ′ 1 , ( uu , ′ ′ 1 2 ( uu , ′ ′ 2 2 wherein det At the same time, square of the length of the anticommutator { . ′ 2 ) ( uu , ′−′ ′ 1 2 1 )2 1,uu 2 = } ) uu , 1 uuuu ( , 1 ( uu , ′ ′ 2 1 ( uu , ′ ′ 2 2 So the axiomatic property of normalized algebras: uu ( , , 1 ( uu , ′ ′ 1 1 ( uu , ′ ′ 1 2 ] [ uu , , 1 [ ( uu , 1 { ( uu , 1 uu , 1 } { , det ) ) } ) ⎡ ⎢ ⎣ ] ) )( ≡ = + − ) ) , 2 2 2 2 2 2 1 . ) ⎤ )⎥ ⎦ uu )( , 2 2 1 ) is expressed as: } { , { ( uu , 1 2 ( uuuu 21 21 is fulfilled. }2 + 2 4. Additive decomposition of the product of three hypercomplex numbers. The product of three octonions obviously can be written in the form: ( ) uuu 21 3 = ( ) uuu 21 3 ( ) uuu 1 23 ( ) uuu 21 3 ( uuu 12 3 ) ( uuu 12 3 ) ( ) uuu 1 23 + + . (5) NOTE: the conjugation of central factor provides mutual orthogonality of summands and is necessary to obtain laconic relations. As any number is conjugate to conjugated one this doesn't reduce the generality of reasoning. is decomposed into the sum In this case the triple product ( ) 3 uuu 21 − 2 − 2 of mutually orthogonal terms: , ( ) uuu 21 3 { uuu , 1 3 = 2 } [ + uuu , 3 1 , 2 ] + uuu , 1 3 , 2 , (6) 3 3 3 2 3 2 2 , , , , , 3 2 , = } , ] ) } [ , uuu , 1 uuu , 1 ( { uuu , 1 ) 0 . uuu , = 1 3 ]3 uuu and , 1 are determined by the relations (7)-(9) with an alternative order wherein ( ) { { ( uuu uuu , , = 1 2 1 2 The mutually orthogonal summands { }3 uuu , , 1 uuu , 1 of multiplication of the factors: ( ) uuu + 1 23 2 − 2 , , (7) (8) { uuu , 1 [ uuu , 1 ( ) uuu 1 23 ( ) uuu 21 ( ) uuu 21 } , 3 , [ uuu 3 12 uuu 32 1 uuu 12 3 uuu 32 1 + 2 − 2 , , } ≡ = = ≡ ] ( ) ( ) ( ) ( ) , , 2 2 3 3 2 3 3 2 2 uuu , 1 , 2 ( ) uuu 21 3 = 3 ( uuu 32 1 ) ≡ ( uuu 12 3 ) ( ) uuu 1 23 . (9) − 2 − 2 2 2 , 1 1 ] 3 , 2 , 2 , 2 , 2 2 3 . )[ ]2 ]3 wherein the terms on the right-hand side of (9), besides the alternative order of arranging the parentheses, are also replaced with each other to fit expres- sion (5). The anticommutator { }3 is expressed by a linear combination uuu is expressed by a linear com- , , 1 and conventional cross products of arguments. So, uuu , 1 of arguments, and the commutator [ bination of the unit 0i the definitions (7)-(8) presented as follows: ) 1 ) } ( { ) ( ( uuu uuu uuu uuu , , , , , , (10) − + = 1 2 2 2 3 3 1 1 3 [ )[ ] )[ [ ] ] ( ) ( ( ( uuu uuiu uuiu iuuu uuiu , , , , , , , , (11) = − − + 1 3 2 1 03 3 02 3 03 01 2 Formula (10) for { }3 uuu is derived from equality (1) by substitu- , , 1 1u . Expression (11) for [ 1 u u + tion in place of is derived di- rectly from definition (9) and expression (4) for the product of two hyper- complex numbers. uuu , 1 uuu , 1 5. Interpretation. Interpretation of the summands { [ ]3 uuu and , 1 ) 3 product ( uuu 21 uuu , 1 of three octonions is described by the following. }3 }3 , in the additive decomposition (5)-(6) of the According to definition (7) and formula (10), { is an anti- commutator, which is expressed by a linear combination of the arguments 3u and does not change when the first and last arguments, namely 1u , 2u , 1u , and 3u , are interchanged. When the multiplicative unit 0i substitutes the central argument, the anticommutator { converts into the anti- commutator { for the product of two hypercomplex numbers, de- scribed by (3). uuu , 1 uuu , 1 1,uu }3 }3 ]3 2 , ]3 The commutator [ uuu , 1 , defined in (8) and represented in (11) as a linear combination of multiplicative unit 0i and conventional cross prod- ucts of pairs of arguments, is a generalization of the conventional cross product to the case of three arguments both for quaternions and octonions. When one of the arguments is replaced by multiplicative unit 0i , the triple cross product [ up to a sign coincides with the conventional cross product of two other arguments. When the multiplicative unit 0i substitutes uuu , 1 ]3 , 2 2 3 , 2 , 2 , 2 , the central argument commutator [ ]3 1,uu 2u , the commutator [ uuu , 1 , 2 ]3 is turned into the for the product of two hypercomplex numbers. 2 3 , uuu , 1 The associator is well known from the literature [4]. Here it is introduced according to formulas (9), where, unlike conventional definitions, the conjugate central argument is used, and the associator 1 . The mentioned differences uuu , 1 2 do not affect the basic properties of the associator. is computed with a coefficient , 2 3 6. Key properties. As in the case of the cross product of two vec- have and the associator , uuu , 1 , 2 ]3 3 tors, the triple cross product [ 2 the following common properties: uuu , 1 2 3 3 2 3 , , The triple cross product and the associator are antisymmetric, i.e. ) ) = = = = [ ] ( uuuu , , , 2 1 1 ( u uuu , , , 1 1 when two arguments are the same, cross product vanishes to zero: [ ] ( ) 0 uuuu , , , = 3 1 3 ( ) 0 uuu u . , , = 3 3 1 [ ( uuu Then scalar quadruple mixed product of vectors , , 1 uuu , 1 )4 are antisymmetric and change sign when any two of [ ] ( uuuu , , 2 1 ( uuu u , , 1 and ( the four arguments exchange by places, for example: )1 u , )1 u . , )4 [ ( uuu , , 1 2 ( uuu , , 1 [ ( uuu , −= 4 3 ( uuu , −= 4 ] , 3 ] , u u ] , , ) ) ) ) , , , , u u , , 3 2 2 2 4 2 3 2 2 2 3 2 4 3 2 3 2 2 , , ]3 uuu , 1 Here the latter properties are a consequence of the previous two. In comparison with the triple cross product [ , the associator uuu satisfies more orthogonality and zeroing conditions. At that, in , 1 3 addition to orthogonality to each of its arguments, the associator is orthogonal to the multiplicative unit 0i and to the conventional cross products of its arguments. In total, the orthogonality of the associator to seven vectors is guaranteed: ) u , = 1 [ uu , 1 uuu , , , 3 1 ( uuu , , = 1 2 In general case, the associator , uuu u , , 2 1 3 ) ( . ] uuu , 0 = 1 3 vanishes to zero if its ar- ) ( u = 2 [ uu , , 1 2 uuu , 1 uuu , 1 3 uuu , 1 ) 0 , = [ uu , , 2 uuu , 1 uuu 1 ( ) ] ( ( , , 2 , ) ] , , i 0 ) ( = = , 2 3 3 2 2 3 , , , 3 3 2 3 2 3 3 2 guments belong to the same quaternion subalgebra, for example: uui , 0 , 2 = 3 [ uu , 2 3 ] , uu , 2 3 = 0 . The squares of the lengths of the anticommutator { 2 , ]3 uuu , 1 commutator [ uuu , and the associator 1 2 2u and terms of the inner products of the vectors 1u , ( uu , ⎡ 1 1 ⎢ ( uu , ⎢ 1 ( uu ⎢ , ⎣ 1 { uuu , 1 uu , )( 2 uu , )( 3 uu , ( 1 1 ) ) ) det } = − ) , , 3 2 3 2 2 3 2 3 3u as follows: ( ( uu uu , , 1 2 3 1 ( ( uu uu , , 2 2 ( ( uu uu , , 3 2 ) ) ) 3 2 3 3 ) ⎤ ⎥ ) ⎥ )⎥ ⎦ }3 , uuu , 1 , the are expressed in 2 = , [ ] uuu , 2 3 1 ( uu , ⎡ 1 1 ⎢ ( uu , ⎢ 1 ( uu ⎢ , ⎣ 1 det 2 ) ) ) 3 2 + uuu , 1 , 2 3 det 2 2 2 u [ ] uu ( , ) , + 1 3 ( ( ) uu uu , , 2 1 3 1 ( ( ) uu uu , , 2 2 ) ( ( uu uu , , 3 3 2 ( ) uu , ′ ′ ⎡ 1 1 ⎢ ( ) uu , ′ ′ ⎢ 1 2 ( ) uu ⎢ , ′ ′ ⎣ 1 3 = 2 3 − ) ⎤ ⎥ ) ⎥ ) ⎥ ⎦ 3 ( uu , ′ ′ 2 1 ( uu , ′ ′ 2 2 ( uu , ′ ′ 2 3 ) ) ) ⎡ ⎢ ⎢ ⎢ ⎣ ( uu , ′ ′ 1 1 ( uu , ′ ′ 2 1 ( uu , ′ ′ 3 1 ( uu , ′ ′ 1 3 ( uu , ′ ′ 2 3 ( uu , ′ ′ 3 3 ) ) ) ) ⎤ ⎥ ) ⎥ ) ⎥ ⎦ ( uu , ′ ′ 2 1 ( uu , ′ ′ 2 2 ( uu , ′ ′ 3 2 ) ) ) , ( uu , ′ ′ 3 1 ( uu , ′ ′ 3 2 ( uu , ′ ′ 3 3 ) ⎤ ⎥ ) ⎥ )⎥ ⎦ − [ uu , ( 1 2 ] , u 3 2 ) . 2u for any hypercomplex vector u is just a re-designation of (12) , (13) (14) wherein ( )uu, . So that in (12)-(14): { } 2 uuu , , 3 1 [ uuu , 1 2 uuu , , 1 2 , ] 3 3 2 2 ≡ 2 2 { ( uuu , , 1 [ ( uuu , , ≡ 1 3 ( uuu , 1 , 2 2 ≡ 3 , } { } )3 uuu , , 1 3 2 ] [ ] )3 uuu , , , , 1 )3 . uuu , , 1 , 2 2 , Using (3), it is easy to verify that in the particular case of the multi- 2 } plicative unit 0i = { uu , 1 substituted in (12) as the central factor: )( { ) uiu uu , , + 1 2 1 )1 ) ( 2 uu , 1 )( ( ) iu iu , , 2 − 2 0 0 1 ) ( ( ) 2 iu uu , , + 2 2 0i in (13) as the central factor (3), ( uu , 1 2 ( iu , + 1 0 Substituting multiplicative unit } = , 2 2 0 2 2 0 2 we obtain: [ uu , 1 2 ( iu , − 1 2 2 , [ ] ] uiu , = 1 2 ) ( ( 2 iu uu , , 2 3 1 0 + ) 3 0 ( )( uu uu , , = 3 1 1 )( )( uu iu , , 1 3 0 0 ) 3 ) − ( uu , − 1 ( iu , 3 0 3 2 ) 3 ) ( 2 − uu , 1 )1 REMARK: pay attention to the variability of the presentation of the squared lengths , which should be systematized in future works. Summing (12)-(14) and taking (6) into account, we obtain: ( ) uuu 21 3 2 ≡ ( ) ( uuu 21 3 , ( ) uuu 21 3 ) = ( uuuu , 1 )( , 1 2 2 )( )3 uu , 3 , matching the axiomatic property of normalized algebras for any hypercom- plex vectors : 1,uu 2 ( uuuu 21 21 , ) = ( uuuu , 1 )( , 1 2 )2 . It is characteristic that the triple cross product [ changes sign, when conjugated with simultaneous conjugation of arguments, while the anticommutator { remain un- changed: and the associator uuu , 1 3 uuu , 1 uuu , 1 }3 ]3 , , , 2 2 2 3 2 { uuu , , 1 uuu , , 1 2 [ uuu , , 1 2 2 }3 } { uuu , = 1 uuu , = 1 3 [ ]3 uuu , −= 1 , , , 2 2 , , . 3 ] 3 2 3 2 3 2 3 2 , , , , ] , + = )12 ( uuu 3 ( uuu 3 12 Applying the operation of general conjugation with simultaneous conjugation of the arguments to (6), we obtain the decomposition of the product of three hypercomplex numbers in the form: ) { uuu , 1 uuu , 1 uuu , 1 } [ − uuu , 1 (15) that was taken into account in the development of the definition (8) for the triple cross product [ ]3 and in the study of the permutation invari- ance of mirror symmetry [6,7]. 7. Okubo solution. In [4] Susumu Okubo considered the product ( ) 3 3u and solved the problem of selection uuu 21 of antisymmetric additive term [ . The solution is obtained 2 through cumbersome calculations and is given in [4] on the page 22 by the following expression: [ ] ( ) uuu uuu , 3 1 21 3 Ocubo wherein [ ]Okubo , . ( ) uuu , 1 2 presented as follows: of three octonions 1u , 2u , uuu , , 1 , = 2 uuu , 1 ( uuiu 2 31 2 ( ) 2 uuu 1 ( ) uuu 1 ]Ocubo , − + − + ) 3 2 3 , , , 3 3 2 3 0 ] [ uuu , , = 3 2 1 Okubo ( uuu uuiu , , , , −= + 1 2 3 01 3 )[ 2 ] + ( uuiu , , 1 02 )[ 3 ] + ( uuiu , , 2 03 )[ 1 ] − ] ( ) 0 iuuu 3 [ , , 2 1 . three octonions ( tonions ( ) 3 uuu 21 the Table. To simplify the comparison of the decomposition of the product of ) 3 uuu in this paper with the decomposition of three oc- 21 according to [4], the notation matching is established in Table. Notation matching Notation in this paper 0i [ ]2 1,uu uuu , 1 , 2 3 Notion Multiplicative unit Inner product Conventional cross product of two vectors Associator − in [4] e 2 uu 1 1 2 1 ( uuu , 1 2 , 2 )3 , Taking into account the expressions (10) for { [ ]3 uuu , 2 1 ( ) uuu 21 uuu , , 1 , the solution [4] is expressed by the formula: uuu , , 1 3 ) 3 which is equivalent to formula (6) for the decomposition of ( uuu 21 ]3 , commutator [ the sum of the anticommutator { uuu , , 1 associator ( uuiu 2 , 31 02 { uuu , 1 uuu , 1 uuu , 1 } [ − }3 . − = − ] ) , , , , 2 2 2 2 3 2 3 2 3 uuu , 1 , 2 3 }3 and (11) for into and 3 2 2 2 , , . ]3 }3 , [ uuu , 1 ) 3 uuu 21 uuu and , , 1 Thus, formulas (6)-(11) allow us in an obvious way to formulate the idea of decomposing ( into the sum of mutually orthogonal { uuu by commutation and changing the , 1 order of multiplication of arguments, and formulas (10)-(11) ensures by minimum effort to enhance the results, originally formulated in [4 ] for ) 3 ( uuu 21 8. Conclusion. Unique multiplicative properties of advanced hyper- complex numbers, namely quaternions and octonions, ensure their wide application not only in mathematics and physics, but also in modern infor- mation technologies for solving problems of navigation, robotics, computer graphics, image processing, etc. A well-known method of algebraic computation in terms of hyper- complex numbers is the additive decomposition of the product of two num- bers into mutually orthogonal symmetric and antisymmetric components i.e. into anticommutator and a commutator (cross product). To develop the tools of hypercomplex numbers and their applica- tions, it is interesting to generalize the additive expansion for the product of three arguments. Susumu Okubo proposed an expansion of the product of three octonions [4], wherein the associator is isolated, but the remaining components are not expressed through mutually orthogonal anticommutator and commutator (cross product) generalized to the case of three arguments. ) 3 uuu 21 Perhaps, it is just not enough simple form of representation of Okubo's re- markable result [4] for triple product ( that prevents its widest ap- plication in the tools of hypercomplex numbers. It seems possible that the transparent idea of the decomposition of the product ( of three hy- percomplex numbers into the sum of mutually orthogonal anticommutator, commutator and associator will help to remove this obstacle and to activate the utilization of octonions. ) 3 uuu 21 Kharinov Mikhail Vyacheslavovich - PhD., associate professor; senior researcher, Labora- tory of Applied Informatics, SPIIRAS. Research interests: digital information analysis, infor- mation quantity estimation, numerical representation system, hierarchical data structures, im- age segmentation and invariant representation for recognition purposes, color transformations of images. The number of publications - 140. [email protected]; SPIIRAS, 39, 14-th Line V.O., St. Petersburg, 199178, Russia; office phone +7(812)328-1919, fax +7(812)328-4450. References Hamilton W.R. Lectures on quaternions. – Hodges and Smith, 1853. – 736 pp. Flaut C. Some remarks regarding Quaternions and Octonions//arXiv preprint arXiv:1711.10434. – 2017. – 19 pp. – URL:https://arxiv.org/abs/1711.10434 Kantor I. L., Solodovnikov A. S. Hypercomplex numbers: an elementary introduction to algebras. – Springer, 1989. – 169 pp. Okubo S. Triple products and Yang–Baxter equation. I. Octonionic and quaternionic triple systems // Journal of mathematical physics. 1993. – Vol. 34. – №. 7. – pp. 3273- 3291. – URL: https://arxiv.org/pdf/hep-th/9212051.pdf Die Mathematischen Hilfsmittel des Physikers. By Dr. Erwin Madelung. Berlin, Julius Springer, – 1922. – 247 pp. Kharinov M.V. Color logical puzzle game "DatoRainbow" USSR patent No. 1799274 from 31.10.90 – 38 pp. (in Russian) Kharinov M.V. Permutational and hidden symmetry on the example of isomorphic Hadamard matrices. Applications in the field of artificial intelligence // Means of mathematical modeling / Proceedings of Second Int. Conf. - S.-P .: publishing house SPbSTU, 1999. -Vol. 5 - pp. 247-254. (in Russian) 1. 2. 3. 4. 5. 6. 7.
1012.1887	2	1012	2010-12-15T23:01:35	Some remarks on structural matrix rings and matrices with ideal entries	[ "math.RA", "math.CO" ]	Associating to each pre-order on the indices 1,...,n the corresponding structural matrix ring, or incidence algebra, embeds the lattice of n-element pre-orders into the lattice of n x n matrix rings. Rings within the order-convex hull of the embedding, i.e. matrix rings that contain the ring of diagonal matrices, can be viewed as incidence algebras of ideal-valued, generalized pre-order relations. Certain conjugates of the upper or lower triangular matrix rings correspond to the various linear orderings of the indices, and the incidence algebras of partial orderings arise as intersections of such conjugate matrix rings.	math.RA	math	Some remarks on structural matrix rings and matrices with ideal entries Tampere University of Technology, PL 553, 33101 Tampere, Finland Stephan Foldes [email protected] Gerasimos Meletiou TEI of Epirus, PO Box 110, 47100 Arta, Greece [email protected] December 2010 Abstract Associating to each pre-order on the indices 1, ..., n the corresponding structural matrix ring, or incidence algebra, embeds the lattice of n-element pre-orders into the lattice of n × n matrix rings. Rings within the order- convex hull of the embedding, i.e. matrix rings that contain the ring of diag- onal matrices, can be viewed as incidence algebras of ideal-valued, generalized pre-order relations. Certain conjugates of the upper or lower triangular ma- trix rings correspond to the various linear orderings of the indices, and the incidence algebras of partial orderings arise as intersections of such conjugate matrix rings. Keywords: structural matrix ring, incidence algebra, pre-order, quasi- order, triangular matrix, conjugation, semiring, ideal lattice, subring lattice 1 Conjugate subrings Rings are understood to be possibly non-commutative, and to have a unit (multiplicatively neutral) element, which is assumed to be distinct from the zero (null) element and is denoted by 1 or I or a similar symbol. Subrings are understood to contain the unit element. An n × n matrix is viewed as a map defined on the set n2 = {1, ..., n}2 . (Here n ≥ 1 is assumed.) The ring of n × n matrices over a ring R is denoted by Mn(R). 1 A pre-order, also called quasi-order, is a reflexive and transitive binary relation . on a set S, an order (or partial order ) is an anti-symmetric pre- order, and a linear ( or total) order is an order in which any two elements are comparable. Instead of the generic notation ., specific pre-orders may be denoted by other symbols such as θ. Given a ring R and a pre-order . on n, the structural matrix ring Mn(., R) over R is defined by Mn(., R) = { A ∈ Mn(R) : ∀i, j A(i, j) = 0 unless i . j } The full matrix ring Mn(R), the subrings of all upper triangular (respectively lower triangular), and of all diagonal matrices, are examples of structural matrix rings. Structural matrix rings are essentially the same as incidence algebras of finite pre-ordered sets, although the latter term is sometimes used under the assumption that the base ring R is a field [F] or that the pre-order in question is an order, possibly on an infinite set, as in [R]. Ring-theoretical properties of incidence algebras most relevant to the present context were studied in [DW, F, MSW, W1, W2]. The set of all pre-orders defined on any given set constitutes a lattice whose minimum is the equality relation. Proposition 1 For any ring R, the map associating to each pre-order . on n the corresponding structural matrix ring Mn(., R) provides an embed- ding of the lattice of pre-orders on n into the lattice of subrings of the matrix ring Mn(R). The embedding preserves also infinite greatest lower and least upper bounds. Proof Preservation of greatest lower bounds is obvious. The preservation of upper bounds is a consequence of the description of the least upper bound of a family of pre-orders as the transitive closure of the least binary relation that is implied by the pre-orders in the family. (cid:3) The lattice embedding described in the proposition above is generally not surjective, and for n ≥ 2 its range is order-convex if and only if R is a division ring. Section 2 will provide a description of the order-convex hull of the embedding's range. Subrings S and T of Mn(R) are said to be permutation conjugates if there is a permutation matrix P such that S = (cid:8)P AP −1 : A ∈ T (cid:9) 2 Such subrings are obviously isomorphic under the automorphism A 7→ P AP −1 of Mn(R). All the n! linear orders on the finite set n are isomorphic, and the iso- morphisms among them are precisely the self-bijections of the underlying set n. Consequently we have: Proposition 2 For any pre-order . on n and any ring R, the following conditions are equivalent: (i) . is a linear order, (ii) Mn(., R) is a permutation conjugate of the ring of upper triangular matrices, (iii) Mn(., R) is a permutation conjugate of the ring of lower triangular matrices. (cid:3) The well-known fact that every partial order is the intersection of its linear extensions yields: Proposition 3 For any pre-order . on n and any ring R, the following conditions are equivalent: (i) . is an order, (ii) Mn(., R) is the intersection of some permutation conjugates of the ring of upper triangular matrices, (iii) Mn(. R) is the intersection of some permutation conjugates of the ring of lower triangular matrices. (cid:3) These considerations were first developed, in [FM1]. They were related to a ring property concerning one-sided and two- sided inverses in [FSW], addressing questions originating in [C] and [SW]. in the context of fields, 2 Matrices with ideal entries The order-convex hull of the embedding provided by Proposition 1 con- sists obviously of those matrix rings that contain the ring of diagonal matri- ces. We now show that these rings can be viewed as generalized incidence algebras, corresponding to generalized relations that are the analogues of pre-order relations in a reticulated semiring-valued framework. 3 By a semiring we understand a set endowed with a binary operation called sum (denoted additively) and a binary operation called product (denoted multiplicatively) such that: (i) the sum operation defines a commutative semigroup, (ii) the product operation defines a semigroup, (iii) both distributivity laws a(b + c) = ab + ac and (b + c)a = ba + ca hold for all members a, b, c of the underlying set. The set I(R) of (bilateral) ideals of any ring R is a semiring under ideal sum and product of ideals. This semiring is lattice-ordered by inclusion. The set Mn(I(R)) of n × n matrices with entries in I(R) is again a semiring under the obvious sum and product operations, also lattice-ordered by entry- wise inclusion: lattice join coincides with semiring sum, denoted +, while the lattice meet operation ∧ is entry-wise intersection. The lattice Mn(I(R)) is complete, it is isomorphic to the n2-th Cartesian power of the complete lattice I(R). The matrix I, with the improper ideal (1) = R in diagonal positions and the trivial ideal (0) in off-diagonal positions, is multiplicatively neutral in the semiring Mn(I(R)). For every matrix U with ideal entries, U ∈ Mn(I(R)), consider the set of matrices G = {A ∈ Mn(R) : ∀i, j A(i, j) ∈ U(i, j)} This set is always an additive subgroup of Mn(R), and it is a subring if and only if U2 + I ≤ U in the ordered semiring of n × n matrices with ideal entries. In that case, we say that the subring G is defined by U. There is a natural lattice embedding from the lattice of all pre-orders on n into the lattice Mn(I(R)), namely the map θ 7→ U where U(i, j) is the improper or trivial ideal of R according to whether the relation iθj holds or not. The matrix U so obtained always satisfies U2 + I ≤ U, and the subring of Mn(R) defined by it coincides with the structural matrix ring Mn(θ, R). For this reason we denote by Mn(R, U) the matrix ring defined by any U ∈ Mn(I(R)) satisfying U2 + I ≤ U. Any U ∈ Mn(I(R)) is viewed as an I(R)-valued relation on the n−element set n, the inequality U2 + I ≤ U generalizes transitivity and reflexivity of relations, and Mn(R, U) may then be thought of as the incidence algebra of the generalized pre-order U. Matrices with ideal entries U ∈ Mn(I(R)) satisfying the generalized tran- sitivity condition U2 ≤ U only were used by van Wyk [W1], and recently 4 again by Meyer, Szigeti and van Wyk [MSW], in the description of ideals of stuctural matrix rings . Any matrix with ideal entries U ∈ Mn(I(R)) satisfying U2 + I ≤ U is said to be reflexive-transitive. If R is a division ring, then the range of the map θ 7→ U described above, embedding the pre-order lattice into Mn(I(R)), is exactly the set of reflexive-transitive matrices. For any ring R, the set of reflexive-transitive matrices with ideal entries constitutes a complete lattice under the ordering of Mn(I(R)) (but not a sublattice of Mn(I(R)) in general). Proposition 4 For any ring R, the map U 7→Mn(R, U) establishes a lattice isomorphism between: (i) the lattice {U ∈ Mn(I(R)) : U2 + I ≤ U} of n × n reflexive-transitive matrices with ideal entries, (ii) the lattice of subrings of Mn(R) containing all diagonal matrices. Proof Obviously if U2 + I ≤ U then Mn(R, U) contains all diagonal matrices, and if U ⊆ V then Mn(R, U) ⊆ Mn(R, V). Conversely, let N ⊆ Mn(R) be any matrix ring including all diagonal matrices. For every 1 ≤ i, j ≤ n the set Uij = {A(i, j) : A ∈ N} is an ideal of R, the matrix U = (Uij) with ideal entries can be seen to satisfy U2 + I ≤ U, and Mn(R, U) = N. (cid:3) The proof above is presented in [FM2] in somewhat greater detail, to- gether with some consequences. For division rings, where there is only the trivial and the improper ideal, the structure of the lattice of subrings of Mn(R) containing all diagonal matrices is independent of the choice of the particular division ring R. In the general case, it is the ideal lattice structure of R that determines the structure of this upper section of the subring lattice of Mn(R). References [C] P.M. Cohn, Reversible rings, Bull. London Math. Soc. 31 (1999) 641 -- 648 5 [DW] S. Dascalescu, L. van Wyk, Do isomorphic structural matrix rings have isomorphic graphs? Proc. Amer. Math. Soc. 124 (1996) 1385 -- 1391. [F] R.B. Feinberg, Polynomial identities of incidence algebras, Proc. Amer. Math. Soc. 55 (1976) 25 -- 28. [FM1] S. Foldes, G. Meletiou, On incidence algebras and triangular ma- trices, Rutcor Res.Report 35-2002, Rutgers University, 2002. Available at rutcor.rutgers.edu/rrr [FM2] S. Foldes, G. Meletiou, On matrix rings containing all diagonal ma- trices, Tampere University of Technology, August 2007 http://math.tut.fi/algebra/ [FSW] S. Foldes, J. Szigeti, L. van Wyk, Invertibility and Dedekind finite- ness in structural matrix rings. Forthcoming in Linear and Multilin. Algebra 2011. Manuscript iFirst 2010, 1 -- 7, http://dx.doi.org/10.1080/03081080903357653 [MSW] J. Meyer, J. Szigeti, L. van Wyk: On ideals of triangular matrix rings, Periodica Math. Hung. Vol. 59 (1) (2009), 109-115 [R] G.-C. Rota, On the foundations of combinatorial theory I. Theory of Mobius functions, Zeitschrift Wahrscheinlichkeitstheorie 2 (1964) 340 -- 368 [SW] J. Szigeti, L. van Wyk, Subrings which are closed with respect to taking the inverse, J. Algebra 318 (2007) 1068 -- 1076 [W1] L. van Wyk, Special radicals in structural matrix rings, Communi- cations Alg. 16 (1988) 421-435 [W2] L. van Wyk, Matrix rings satisfying column sum conditions versus structural matrix rings, Linear Algebra Appl. 249 (1996) 15 -- 28 6
1306.1620	1	1306	2013-06-07T05:40:08	Clifford (Geometric) Algebra Wavelet Transform	[ "math.RA", "math.RT" ]	While the Clifford (geometric) algebra Fourier Transform (CFT) is global, we introduce here the local Clifford (geometric) algebra (GA) wavelet concept. We show how for $n=2,3 (\mod 4)$ continuous $Cl_n$-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 $Cl_{n}$ CFT 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 explicit example, we introduce Clifford Gabor wavelets. We further invent a generalized Clifford wavelet uncertainty principle. Extensions of CFTs and Clifford wavelets to $Cl_{0,n'}, n' = 1,2 (\mod 4)$ appear straight forward. Keywords: Clifford geometric algebra, Clifford wavelet transform, multidimensional wavelets, continuous wavelets, similitude group.	math.RA	math	Clifford (Geometric) Algebra Wavelet Transform Eckhard Hitzer, Department of Applied Physics, University of Fukui, 910-8507 Japan September 21, 2018 Abstract While the Clifford (geometric) algebra Fourier Transform (CFT) is global, we introduce here the local Clifford (geometric) algebra (GA) wavelet concept. We show how for n = 2, 3(mod 4) continuous 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 CFT 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 explicit example, we introduce Clifford Gabor wavelets. We further invent a generalized Clifford wavelet uncertainty principle. Extensions of CFTs and Clifford wavelets to Cl0,n′ , n′ = 1, 2(mod 4) appear straight forward. Keywords: Clifford geometric algebra, Clifford wavelet transform, multidimensional wavelets, con- tinuous wavelets, similitude group. AMS Subj. Class.: 15A66, 42C40, 94A12. 1 Introduction The meaning and importance of wavelets is clearly seen in a biographical note on J. P. Morlet: Fol- lowing in the footsteps of Denis Gabor (father of holography), Morlet was disconcerted by the poor results he [Gabor] obtained; but, being inquisitive and persistent, he asked himself, "Why?" and immediately provided the answer. Gabor paved the time-frequency plane in uniform cells and associated each cell with a wave shape of invariant envelope with a carrier of variable frequency. Morlet kept the constraint resulting from the uncertainty principle applied to time and frequency, but he perceived that it was the wave shape that must be invariant to give uniform resolution in the entire plane. For this he adapted the sampling rate to the Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial ad- vantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. frequency, thereby creating, in effect, a changing time scale producing a stretching of the wave shape. Today the wavelet transform is also called the "time-scale analysis" approach, which is comparable to the conventional time-frequency analysis. . . . It has been rediscovered as a very useful tool, partic- ularly in data compression where it can produce significant savings in storage and transmission costs but also in mathematics, data processing, communications, image analysis, and many other engineering problems.[1] In order to favorably combine wavelet techniques with Clifford (geometric) algebra, which provides a complete algebra of a vector space and all its sub- spaces, several efforts have been undertaken. They include Clifford multi resolution analysis (MRA) [2], quaternion MRA [4], Clifford wavelet net- works, quaternion wavelet transforms (QWT) ap- plied to image analysis (using the QWT phase concept), image processing and motion estimation [5], quaternion-valued admissible wavelets, Clifford algebra-valued admissible (continuous) wavelets us- ing complex Fourier transforms for the spectral representation [6], monogenic wavelets over the unit ball [7], Clifford continuous wavelet transforms (ContWT) in L0,2, L0,3, wavelets on the 3D sphere with Cauchy kernel in Clifford analysis (2009), dif- fusion wavelets [8], ContWT in Clifford analysis, wavelet frames on the sphere, benchmarking of 3D Clifford wavelet functions, metric dependent Clif- ford analysis, new multivariable polynomials and as- sociated ContWT: Clifford versions of Hermite, Her- mitean Clifford-Hermite, bi-axial Clifford-Hermite, Jacobi, Gegenbauer, Laguerre, and Bessel polyno- mials [3]. Fourier transformations have been successfully developed in the framework of real Clifford (geo- metric) algebra (GA), replacing the imaginary unit i ∈ C by a geometric (GA) square root of −1 [9]. These Clifford Fourier transformations (CFT) [10 -- 12] have already found interesting applications in vector field analysis and pattern matching [17]. A special case are the socalled quaternion Fourier transforms (QFT) [13 -- 15]. We now use the spectral CFT representation in order to develop real Clifford GA wavelets in di- mensions n = 2, 3(mod 4). We dimensionally ex- tend [16] and elaborate the short summary given in [18] by adding proofs and generalizations. In Section 2 Clifford (geometric) algebra is in- troduced including multivector signal functions, the Clifford Fourier transform, and the similitude group of dilations, rotations and translations. Section 3 defines Clifford mother and daughter wavelets, spec- tral representation, discusses admissibility, the Clif- ford wavelet transformation and its spectral CFT representation. This is followed by a detailed dis- cussion of Clifford wavelet properties, i.e. linearity, covariance w.r.t. dilation, rotation and translation, inner product and norm relations, the inverse Clif- ford wavelet transform, a reproducing kernel and a Clifford wavelet uncertainty principle. Finally the example of Clifford Gabor wavelets is given. 2 Clifford (geometric) algebra and multivector signals 2.1 Clifford (geometric) algebra Clifford (geometric) algebra is based on the geomet- ric product of vectors a , b ∈ Rp,q, p + q = n a b = a · b + a ∧ b, (1) and the associative algebra Clp,q thus generated with R and Rp,q as subspaces of Clp,q. a · b is the symmetric inner product of vectors and a ∧ b is Grassmann's outer product of vectors representing the oriented parallelogram area spanned by a , b. As an example we take the Clifford geometric al- gebra Cl3 = Cl3,0 of three-dimensional (3D) Eu- clidean space R3 = R3,0. R3 has an orthonormal basis {e1, e2, e3}. Cl3 then has an eight-dimensional basis of {1, e1, e2, e3 , e2e3, e3e1, e1e2 , i = e1e2e3 }. (2) vectors area bivectors volume trivector {z } {z } {z } Here i denotes the unit trivector, i.e. the oriented volume of a unit cube, with i2 = −1. The even grade subalgebra Cl+ 3 is isomorphic to Hamilton's quaternions H. Therefore elements of Cl+ 3 are also called rotors (rotation operators), rotating vectors and multivectors of Cl3. In general Clp,q, p+q = n is composed of so-called r-vector subspaces spanned by the induced bases {e k1 e k2 . . . e kr 1 ≤ k1 < k2 < . . . < kr ≤ n}, (3) each with dimension (cid:0)r the Clp,q therefore becomesPn n(cid:1). The total dimension of General elements called multivectors M ∈ Clp,q, p + q = n, have k-vector parts (0 ≤ k ≤ n): scalar part Sc(M ) = hM i = hM i0 = M0 ∈ R, vec- tor part hM i1 ∈ Rp,q, bi-vector part hM i2, . . . , and r=0(cid:0)r n(cid:1) = 2n. pseudoscalar part hM in ∈Vn Rp,q M = MAe A = hM i+hM i1 +hM i2+. . .+hM in . 2nXA=1 (4) (5) The reverse of M ∈ Clp,q defined as (−1) nXk=0 k(k−1) 2 hM ik, fM = often replaces complex conjugation and quaternion conjugation. Taking the reverse is equivalent to re- versing the order of products ob basis vectors in the basis blades of (3). The scalar product of two mul- tivectors M, eN ∈ Clp,q is defined as M ∗ eN = hMeN i = hM eN i0. For M, eN ∈ Cln = Cln,0 we get M ∗ eN = PA MANA. The modulus M of a multivector M ∈ Cln is defined as (6) M 2 = M ∗fM =XA M 2 A. (7) For n = 2(mod 4) and n = 3(mod 4) the pseu- doscalar is in = e 1e 2 . . . e n with (also valid for1 Cl0,n′, n′ = 1, 2(mod 4)) i2 n = −1. (8) A blade Bk = b 1 ∧ b 2 ∧ . . . ∧ b k, b l ∈ Rp,q, 1 ≤ l ≤ k ≤ n = p + q describes a k-dimensional vector subspace VB = {x ∈ Rp,qx ∧ B = 0}. Its dual blade B∗ = Bi−1 n (9) (10) describes the complimentary (n − k)-dimensional vector subspace V ⊥ B . The pseudoscalar in ∈ Cln is central for n = 3(mod 4) in M = M in, ∀M ∈ Cln. (11) For the Clifford geometric algebra Fourier trans- formation (CFT) [11] 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 transform2 F {f }: Rn → Cln, n = 2, 3(mod 4) is given by F {f }(ω) = bf (ω) =ZRn for multivector functions f : Rn → Cln. f (x) e−inω·x dnx, (19) The CFT (19) is inverted by f (x ) = F −1[F {f }(ω)] 1 (2π)nZRn 1For an extension of the current real Clifford (geometric) algebra wavelet approach to Clifford algebras Cl0,n′ the defi- nition of reversion has to be modified to include sign changes of negative definite basis vectors e k → ee k = −e k, 1 ≤ k ≤ n′. L2(G; Cln) = {f : G → Cln kf k < ∞}. (27) 2 The CFT can be defined analogously for Cl0,n′ , n′ = 1, 2(mod 4). But for even n we get due to non-commutativity [11] of the pseudoscalar in ∈ Cln for all M ∈ Cln, λ ∈ R = F {f }(ω) einω·x dnω. (20) inM = Meven in − Modd in , einλM = Meven einλ + Modd e−inλ. (12) (13) The similitude group G = SIM (n) of dilations, rotations and translations is a subgroup of the affine group of Rn 2.2 Multivector signal functions A multivector valued function f : Rp,q → Clp,q, p + q = n, has 2n blade components (fA : Rp,q → R) G = R+ × SO(n) ⋊ Rn = {(a, rθ, b)a ∈ R+, rθ ∈ SO(n), b ∈ Rn}. (21) The left Haar measure on G is given by fA(x )e A. (14) dλ = dλ(a, θ, b) = dµ(a, θ)dnb, f (x ) =XA We define the inner product of Rn → Cln functions f, g by fA(x )gB(x ) dnx , (f, g) =ZRn f (x )gg(x ) dnx e Afe BZRn =XA,B kf k2 = h(f, f )i =ZRn and the L2(Rn; Cln)-norm f (x )2dnx =XA ZRn A(x ) dnx , f 2 L2(Rn; Cln) = {f : Rn → Cln kf k < ∞}. (15) (16) (17) (18) dµ = dµ(a, θ) = dadθ an+1 , (22) (23) where dθ is the Haar measure on SO(n). For exam- ple dθ =( dθ n = 2 2π , 1 8π2 sin θ1dθ1dθ2dθ3, n = 3 . (24) We define the inner product of f, g : G → Cln by (f, g) =ZG f (a, θ, b) ^ g(a, θ, b) dλ(a, θ, b), (25) and the L2(G; Cln)-norm kf k2 = h(f, f )i =ZG f (a, θ, b)2dλ, (26) is an invertible multivector constant and finite at a.e. ω ∈ Rn. We must therefore have bψ(ω = 0) = 0 ψ(0) =ZRn =XA ZRn ψ(x)ein0·x dnx =ZRn ψA(x) dnx eA = 0, ψ(x) dnx (34) and therefore for all 2n Clifford mother wavelet com- ponents ψA(x ) dnx = 0. (35) ZRn By construction Cψ = fCψ. 2, 3(mod 4) Hence for n = Cψ = hCψi0 + hCψi1 + hCψi4 + hCψi5 + . . . (hCψi4k + hCψi4k+1), = and [n/4]Xk=0 hCψi0 =ZRn =ZRn h{bψ(ξ)}∼bψ(ξ)i0 bψ(ξ)2 dξn > 0. ξn (36) (37) 1 ξn dξn signal f ∈ The variations of a multivector L2(Rn; Cln) in position x ∈ Rn and frequency ω ∈ Rn are related by the CFT uncertainty prin- ciple [11, 19 -- 21] kx f k2 L2(Rn;Cln) kω f k2 L2(Rn;Cln) ≥ n (2π)n 4 kf k4 L2(Rn;Cln). (28) 3 Clifford GA wavelets 3.1 Real admissible continuous Clif- ford GA wavelets We represent the transformation group G = SIM (n) by applying translations, scaling and ro- tations to a so-called Clifford mother wavelet ψ : Rn → Cln ψ(x ) 7−→ ψa,θ,b (x ) = 1 an/2 ψ(r−1 θ ( x − b a )). (29) The family of wavelets ψa,θ,b are so-called Clifford daughter wavelets. Lemma 1 (Norm identity). The factor a−n/2 in ψa,θ,b ensures (independent of a, θ, b) that kψa,θ,bkL2(Rn;Cln) = kψkL2(Rn;Cln). (30) Proof. kψa,θ,b k2 L2(Rn;Cln) 1 1 an =ZRnXA anZRnXA =ZRnXA = ψ2 A(rθ −1( x − b ) dnx ) a = z {z } ψ2 A(z)an det(rθ) dnz ψ2 A(z) dnz = kψkL2(Rn;Cln,0). The invertibility of Cψ depends on its grade content, e.g. for n = 2, 3, Cψ is invertible, if and only if 1 6= hCψi2 hCψi2 0 : C −1 ψ = hCψi0 − hCψi1 hCψi2 0 − hCψi2 1 . (38) (31) Definition 2 (Clifford GA wavelet transformation (CWT) ). For an admissible GA mother wavelet ψ ∈ L2(Rn; Cln) and a multivector signal function f ∈ L2(Rn; Cln) The spectral CFT representation of Clifford daughter wavelets is F {ψa,θ,b }(ω) = a θ (ω))e−inb ·ω . (32) n 2 bψ(ar−1 In the proof of (32) the CFT properties of scaling, x -shift and rotation are applied. A Clifford mother wavelet ψ ∈ L2(Rn; Cln) is admissible if Cψ =ZR+ZS0(n) =ZRn ebψ(ω)bψ(ω) ωn an{bψ(ar−1 θ (ω))}∼bψ(ar−1 θ (ω)) dµ dnω, Tψ : L2(Rn; Cln) → L2(G; Cln), f 7→ Tψf (a, θ, b) =ZRn f (x) ^ ψa,θ,b(x) dnx. (39) (40) • Because of (12) we need to restrict the mother wavelet ψ for n = 2(mod 4) to even or odd grades: Either we have a spinor wavelet ψ ∈ L2(Rn; Cl+ n ) with ε = 1, or we have an odd parity wavelet ψ ∈ L2(Rn; Cl− n ) with ε = −1. (33) • For n = 3(mod 4), no grade restrictions exist. We then always have ε = 1. NB: The admissibility constant Cψ is always scalar for n = 2, for the spinor wavelet as well as for the odd parity vector wavelet. The spectral (CFT) representation of the Clifford wavelet transform is Dilation covariance: If 0 < c ∈ R then [Tψf (c ·)](a, θ, b) = 1 c n 2 Tψf (ca, θ, cb) . (45) Proof. By definition (41) [Tψf (c ·)](a, θ, b) θ (ω))}∼eεinb ·ω dnω. Tψf (a, θ, b) = 1 (2π)nZRnbf (ω) a n 2 {bψ(ar−1 Proof. Tψf (a, θ, b) IP= (f, ψa,θ,b )L2(Rn;Cln,0) P T= 1 1 1 (2π)n (bf , \ψa,θ,b )L2(Rn;Cln,0) (2π)nZRn f (ω)h\ψa,θ,b (ω)i∼ (2π)nZRn (2π)nZRn 2 hbψ(ar−1 f (ω) einb ·ωa f (ω) a 1 n n 2 hbψ(ar−1 θ (ω))i∼ = F T= = (42) θ (ω))i∼ dnω eεinb ·ωdnω, with IP = inner product (15), P T = CFT Plancherel theorem [11], and F T = F {ψa,θ,b } of (32). NB: The CFT for n = 2(mod 4) preserves even and odd grades. 3.2 Properties of real Clifford GA wavelets We immediately see from Definition 2 that the Clif- ford GA wavelet transform is left linear with respect to multivector constants λ1, λ2 ∈ Cln. We further have the following set of properties. Translation covariance: If the argument of Tψf (x ) is translated by a constant x 0 ∈ Rn then [Tψf (· − x 0)](a, θ, b) = Tψf (a, θ, b − x 0) . (43) Proof. By definition dnω = Tψf (ac, θ, bc). (46) 1 a 2 (cid:20)ψ(r−1 n dnx y c − b a a 1 θ ( x − b ))(cid:21)∼ cn "ψ r−1 2 (cid:20)ψ(cid:18)r−1 θ ( θ ( ac y − bc )!#∼ )(cid:19)(cid:21)∼ dny dny f (y) 1 a n 2 1 f (y) (ac) n f (cx) y = cx =ZRn = ZRn 2 ZRn = n 1 c 1 c n 2 Rotation covariance: If r = rθ, r0 = rθ0 and r′ = rθ ′ = r0r = rθ0rθ are rotations, then [Tψf (rθ0 ·)](a, θ, b) = Tψf (a, θ′, rθ0 b) . (47) Proof. By definition and with substitution y = r0x [Tψf (r0·)](a, θ, b) =ZRn =ZRn =ZRn =ZRn =ZRn f (r0x ) ^ ψa,θ,b (x ) dnx x − b a r−1 0 y − b f (r0x )(cid:20)ψ(r−1( f (y )(cid:20)ψ(cid:18)r−1( f (y )(cid:20)ψ(cid:18)r−1r−1 f (y )(cid:20)ψ(cid:18)(r0r)−1( 0 ( dnx ))(cid:21)∼ )(cid:19)(cid:21)∼ )(cid:19)(cid:21)∼ )(cid:19)(cid:21)∼ a a a y − r0b y − r0b = Tψf (a, θ′, r0b). det−1(r) dny dny dny (48) [Tψf (· − x0)](a, θ, b) =ZRn =ZRn =ZRn f (x − x0) ^ ψa,θ,b (x) dnx f (x − x0 θ ( n ) 1 a 2 (cid:20)ψ(r−1 {z }=y 2 (cid:20)ψ(cid:18)r−1 1 a θ ( n f (y) a x − b a y − (b − x0) dnx )(cid:21)∼ )(cid:19)(cid:21)∼ = Tψf (a, θ, b − x0). Now we will see some differences from the classical wavelet transforms. The next property is an inner product relation: Let f, g ∈ L2(Rn; Cln) arbitrary. Then we have (Tψf, Tψg)L2(G;Cln) = (f Cψ, g)L2(Rn;Cln)(49) In the following proof of (49) we will use the ab- dny breviations (44) Fa,θ(ω) = a Ga,θ(ω′) = a n 2 f (ω){ ψ(ar−1 2 g(ω′){ ψ(ar−1 θ (ω))}∼, θ (ω′))}∼ , n (50) (51) and the spectral representations (41) Tψf (a, θ, b) = Tψg(a, θ, b) = F {Fa,θ}(−εb) (2π)n , (52) F {Ga,θ}(−εb) (2π)n . (53) Proof. Using the abbreviations (50), (51) and spec- tral representations (52), (53) we get (Tψf, Tψg)L2(G;Cl3,0) = 1 (2π)2nZR+ZS0(n)(cid:18)ZRn F {Fa,θ}(−εb) P T 1 = ZR+ZS0(n) (2π)nZRn(cid:18)ZR+ 1 = ψ(ar−1 θ (ξ))}∼ f (ξ){ ψ(ar−1 Fa,θ(ξ) ^ (2π)n(cid:18)ZRn anZS0(n) {F {Ga,θ}(−εb)}∼ dnb(cid:19)dµ Ga,θ(ξ) dnξ(cid:19) dµ θ (ξ))gg(ξ)dnξ(cid:19) dµ θ (ξ))dµ(cid:19)= Cψgg(ξ) dnξ f (x)Cψgg(x) dnx = (f Cψ, g)L2(Rn;Cln,0), f (ξ)(cid:18)ZR+ZS0(n) f (ξ)Cψgg(ξ) dnξ an{ ψ(ar−1 θ (ξ))}∼ ψ(ar−1 (54) = 1 (2π)nZRn 1 = (2π)nZRn P T= ZRn where P T denotes the CFT Plancherel theorem. For the second equality we have also used the fact, that a substitution b ′ = −εb, ε = ±1, as in RRn h(−εb) dnb = RRn h(b ′)dnb ′, does not change the overall sign. As a corollary we get the following norm relation: kTψf k2 L2(G;Cln) = Sc(f Cψ, f )L2(Rn;Cln) = Cψ ∗ (f, f )L2(Rn;Cln) . (55) (56) We can further derive the Theorem 1 (Inverse Clifford Cln wavelet trans- form). Any f ∈ L2(Rn; Cln) can be decomposed with respect to an admissible Clifford GA wavelet as f (x) =ZG =ZG Tψf (a, θ, b) ψa,θ,b C −1 ψ dµdnb (57) (f, ψa,θ,b)L2(Rn;Cln)ψa,θ,bC −1 ψ dµdnb, the integral converging in the weak sense. Proof. For any g ∈ L2(Rn; Cln,0) (Tψf, Tψg)L2(G;Cln,0) Tψf (a, θ, b){Tψg(a, θ, b)}∼ dµdnb =ZG =ZGZRn =ZRnZG =(cid:18)ZG Tψf (a, θ, b)ψa,θ,b (x)gg(x) dnxdµdnb Tψf (a, θ, b)ψa,θ,b (x) dµdnb gg(x) dnx Tψf (a, θ, b)ψa,θ,b dµdnb , g(cid:19)L2(Rn;Cln,0) IP R= (f Cψ , g)L2(Rn;Cln,0), (58) where IP R denotes the inner product relation (49). Because (58) holds for any g ∈ L2(Rn; Cln,0) we get Tψf (a, b, θ)ψa,θ,b (x) dµdnb, (59) or equivalently the inverse CWT Tψf (a, b, θ)ψa,θ,b (x) C −1 ψ dµdnb. (60) f (x)Cψ =ZG f (x) =ZG Next is the reproducing kernel : We define for an admissible Clifford mother wavelet ψ ∈ L2(Rn; Cln) Kψ(a, θ, b; a′, θ′, b ′) =(cid:16)ψa,θ,b C −1 ψ , ψa′,θ′,b ′(cid:17)L2(Rn;Cln) . (61) Then Kψ(a, θ, b; a′, θ′, b ′) is a reproducing kernel in L2(G, dλ), i.e, Tψf (a′, θ′, b ′) =ZG Tψf (a, θ, b)Kψ(a, θ, b; a′, θ′, b ′)dλ . (62) Proof. By inserting for f (x ) the inverse CWT (57) into the definition of the CWT we obtain Tψf (a′, θ′, b′) Tψf (a, θ, b) ψa,θ,b (x) dλ C −1 ψ (cid:19) {ψa′,θ′,b ′(x)}∼ dnx Tψf (a, θ, b) ψa,θ,b (x)C −1 ψ {ψa′,θ′,b ′ (x)}∼ dnx(cid:19)=Kψ Tψf (a, b, θ)Kψ(a, θ, b; a′, θ′, b′) dλ. dλ (63) =ZRn(cid:18)ZG =ZG (cid:18)ZRn =ZG Theorem 2 (Generalized Clifford GA wavelet un- certainty principle). Let ψ be an admissible Clif- ford algebra mother wavelet. Then for every f ∈ L2(Rn; Cln), the following inequality holds kbTψf (a, θ, b)k2 ≥ n(2π)n 4 L2(G;Cln)Cψ ∗ (fω f ,fω f )L2(Rn;Cln) (cid:2)Cψ ∗ (f, f )L2(Rn;Cln)(cid:3)2 . (64) NB: The integrated variance It follows with (33) that the Clifford Gabor wavelet admissibility constant Cψc =ZRn = eAAZRn {cψc(ξ)}∼cψc(ξ) ξn φ(ξ)2 ξn dnξ . dnξ (69) If e.g. A is a vector or a product of vectors (versor), then Cψc will be scalar. ZR+ZSO(n) kωF {Tψf (a, θ, . )}k2 L2(Rn;Cln)dµ (65) 4 Conclusion is independent of the wavelet parity ε. Otherwise the proof is similar to the one for n = 3 in [16]. For scalar admissibility constant this reduces to Corollary 1 (Uncertainty principle for Clifford GA wavelet). Let ψ be a Clifford algebra wavelet with scalar admissibility constant Cψ ∈ Rn. Then for ev- ery f ∈ L2(Rn; Cln), the following inequality holds We have introduced real Clifford (geometric) algebra wavelets for multivector signals taking values in Cln. Real means that we completely avoid to use the field of complex numbers C. This also applies to the use of a real Clifford (geometric) algebra Fourier trans- form for the spectral representation. An extension to Cl0,n′ , n′ = 1, 2(mod 4) appears straight forward. kb Tψf (a, θ, b)k2 L2(G;Cln) kω f k2 L2(Rn;Cln) ≥ nCψ (2π)n 4 kf k4 L2(Rn;Cln). (66) • This shows indeed, that Theorem 2 represents a multivector generalization of the uncertainty principle for Clifford wavelets with scalar ad- missibility constant. • Compare with the (direction independent) un- certainty principle (28) for the CFT. 3.3 Example of Clifford GA Gabor wavelets Finally Clifford (geometric) algebra Gabor Wavelets are defined as (variances σk, 1 ≤ k ≤ n, for n = 2(mod 4) : A ∈ Cl+ n or A ∈ Cl− n ) ψc(x ) = (einω0 ·x − e− 1 2 Pn k=1 σ2 kω2 0,k ), x2 k σ2 k k=1 σk A e (2π) n − 1 2 Pk 2 Qn x , ω0 ∈ Rn, constant A ∈ Cln . constant {z } (67) The spectral (CFT) representation of the Clifford Gabor wavelets (67) is F {ψc}(ω) =cψc(ω) = A(e− 1 2 Pk σ2 =φ(ω)∈R {z k(ωk−ω0,k)2 − e− 1 2 Pk σ2 k(ω2 k+ω2 0,k) (68) } Acknowledgments Soli deo gloria. I do thank my dear family, B. Mawardi, G. Scheuermann, D. Hildenbrand and V. Skala. References [1] P. Goupillaud, Biography of Jean P. Morlet, http://www.mssu.edu/seg-vm/bio_jean_p__morlet.html [2] M. Mitrea, Clifford Wavelets, Singular Integrals and Hardy Spaces. Lect. Notes in Math. 1575, Springer, New York, 1994. [3] Brackx et al., Ghent Clifford Analy- (2001-2007), sis Wavelet http://cage.ugent.be/crg/cliffordpublicaties.PDF Publications [4] L. Traversoni, Quaternion Wavelet Problems. Proc. of 8th Int. Symp. on Approx. The- ory, Texas A&M University, Jan. 1995. Im- age Analysis Using Quaternion Wavelets. in E. Bayro-Corrochano, G. Sobczyk (eds.), Proc. of AGACSE 1999, Birkhauser, Basel, 2001. ) . [5] E. Bayro-Corrochano, List of Publications, http://www.gdl.cinvestav.mx/edb/publications.html [6] J. Zhao, and L. Peng. Quaternion-valued Ad- missible Wavelets Associated with the 2D Eu- clidean Group with Dilations. J. of Nat. Geom., 2001; J. Zhao. Clifford algebra-valued Ad- missible Wavelets Associated with Admissi- ble Group. Acta Sci. Nat. Univ. Pek., 41(5), (2005). [7] Kahler et al. Monogenic Wavelets over Unit Ball. ZAA, 24, 813 -- 824 (2005) . [8] S. Bernstein, T.E. Simos et al. (eds.). Clifford Continuous Wavelet Transforms in L0,2 and L0,3. AIP Proc. of ICNAAM 2008, 1048, 634 -- 637 (2008); S. Bernstein and S. Ebert, Spherical Wavelets, Kernels and Symmetries. AIP Proc. of ICNAAM 2009, 1168, 761 -- 764 (2009); S. Bernstein, S. Ebert and R.S. Krausshar, Dif- fusion Wavelets on Conformally Flat Cylin- ders and Tori. AIP Proc. of ICNAAM 2009, 1168, 773 -- 776 (2009); S. Bernstein, Spheri- cal Singular Integrals, Monogenic Kernels and Wavelets on the 3D Sphere. AACA, 19(2), 173 -- 189 (2009). [9] E. Hitzer and R. Ab lamowicz, Geometric Roots of −1 in Clifford Algebras Cl(p, q) with p + q ≤ 4. Submitted to Adv. App. Cliff. Alg., May 2009. Preprint version: Technical Report 2009-3, Department of Mathematics, Tennessee Technological University, Tennessee, USA, May 2009. Also available as: arXiv:0905.3019v1 [math.RA] [10] B. Mawardi and E. Hitzer, Clifford Fourier Transformation and Uncertainty Principle for the Clifford Geometric Algebra Cl(3, 0). Adv. App. Cliff. Alg., 16(1), 41 -- 61 (2006). [11] E. Hitzer and B. Mawardi, Clifford Fourier Transform on Multivector Fields and Uncer- tainty Principles for Dimensions n = 2 (mod 4) and n = 3 (mod 4). Adv. App. Cliff. Alg. Vol. 18(S3,4), 715 -- 736 (2008). [12] B. Mawardi, E. Hitzer and S. Adji, Two- Dimensional Clifford Windowed Fourier Trans- form. acc. for G. Scheuermann, E. Bayro- Corrochano (eds.), Appl. Geom. Algs. in Comp. Sc. and Engineering, Springer, New York, 2010. [13] T. A. Ell, Quaterion-Fourier Transform for Analysis of Two-dimensional Linear Time- Invariant Partial Differential Systems. Proc. 32nd IEEE Conf. on Decision and Control, Dec. 1993, 1830 -- 1841. [14] E. Hitzer, Quaternion Fourier Transform on Quaternion Fields and Generalizations. Adv. App. Cliff. Alg., 17(3), 497 -- 517 (2007); E. Hitzer, Directional Uncertainty Principle for Quaternion Fourier Transforms. Adv. App. Cliff. Alg., Online First, 14 pp. (2009). [15] B. Mawardi, E. Hitzer, R. Ashino and R. Vail- lancourt, Windowed Fourier transform of two- dimensional quaternionic signals. Submitted to Appl. Math. and Comp., March 2009. [16] B. Mawardi and E. Hitzer, Clifford Algebra Cl3,0-valued Wavelet Transformation, Clifford Wavelet Uncertainty Inequality and Clifford Gabor Wavelets. Int. J. of Wavelets, Multires. and Inf. Proces., 5(6), 997 -- 1019 (2007). [17] J. Ebling and G. Scheuermann, Clifford Fourier Transform on Vector Fields. IEEE Trans. on Vis. and Comp. Graph., 11(4), (July/Aug. 2005); Clifford Convolution And Pattern Matching On Vector Fields. Proc. 14th IEEE Vis. 2003 (VIS'03), p. 26, (Oct. 22-24, 2003). [18] E. Hitzer, T.E. Simos et al. (eds.). Real Clifford Algebra Cln,0, n = 2, 3(mod 4) Wavelet Trans- form. AIP Proc. of ICNAAM 2009, 1168, 781 -- 784 (2009). [19] E. Hitzer and B. Mawardi, Uncertainty Princi- ple for the Clifford Geometric Algebra Cl(3, 0) based on Clifford Fourier Transform. in TE. Simos, G. Sihoyios, C. Tsitouras (eds.), Proc. of ICNAAM 2005, Wiley-VCH, Weinheim, 922 -- 925 (2005). [20] E. Hitzer and B. Mawardi, Uncertainty Princi- ple for Clifford Geometric Algebras Cln,0, n = 3(mod 4) based on Clifford Fourier Transform. in T. Qian, M.I. Vai, X. Yusheng (eds.), Wavelet Analysis and Applications, Springer (SCI) Book Series Applied and Numerical Har- monic Analysis, Springer, 45 -- 54 (2006). [21] B. Mawardi, E. Hitzer, A. Hayashi and R. Ashino, An Uncertainty Principle for Quater- nion Fourier Transform, Comp. & Math. with Appl., 56, 2398 -- 2410 (2008).
1601.00553	2	1601	2016-04-12T01:12:20	Averaging algebras, rewriting systems and Gr$\"o$bner-Shirshov bases	[ "math.RA" ]	In this paper, we study the averaging operator by assigning a rewriting system to it. We obtain some basic results on the kind of rewriting system we used. In particular, we obtain a sufficient and necessary condition for the confluence. We supply the relationship between rewriting systems and Grobner-Shirshov bases based on bracketed polynomials. As an application, we give a basis of the free unitary averaging algebra on a non-empty set.	math.RA	math	AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES XING GAO AND TIANJIE ZHANG Abstract. In this paper, we study the averaging operator by assigning a rewriting system to it. We obtain some basic results on the kind of rewriting system we used. In particular, we obtain a sufficient and necessary condition for the confluence. We supply the relationship between rewriting systems and Grobner-Shirshov bases based on bracketed polynomials. As an application, we give a basis of the free unitary averaging algebra on a non-empsty set. Contents Introduction 1. 2. Grobner-Shirshov bases and rewriting systems 2.1. Free averaging algebras 2.2. Grobner Shirshov bases 2.3. Term-rewriting systems 2.4. Term-rewriting systems and Grobner-Shirshov bases 3. A basis of the free averaging algebra References 1 2 3 5 7 11 12 23 1. Introduction The averaging operators are generalizations of conditional expectation in probability theory [31], and are closely related to Reynolds operators, symmetric operators and Rota-Baxter operators [16, 39, 10]. The study of averaging operators originated from a famous paper on turbulence theory by Reynolds in 1895 [34]. The explicit definition of averaging operators was given in 1930s [26]. Since then there is an extensive literature on averaging operators under various contexts, which can be grouped into two classes. The first one is mainly analytic and for different varieties of averaging algebras; see the references [6, 15, 25, 30, 31, 36]. The other class is from an algebraic point of view. Cao [12] constructed the free commutative averaging algebras and characterized the naturally induced Lie algebra structures from averaging operators. Aguiar proved that the diassociative algebra -- the enveloping algebra of the Leibniz algebra [29] -- can be obtained from the averaging associative algebra [1]. Recently, Guo et al. acquired a relationship between averaging operators and Rota-Baxter operators: the algebraic structures resulted from the actions of the two operators are Koszul dual to each other [22]. It is worth mentioning that the Rota-Baxter operator has broad connections with many areas in mathematics and mathematical physics [3, 4, 21]. In [22], Guo et al. also constructed the free nonunitary (noncommutative) averaging algebra on a non-empty set in terms of a class of bracketed words, by checking the Date: January 12, 2021. 1 2 XING GAO AND TIANJIE ZHANG universal property. It is natural to construct further the free unitary (noncommutative) averaging algebra on a non-empty set -- our main object of study in the present paper. Grobner and Grobner-Shirshov bases theory was initiated independently by Shirshov [38], Hironaka [24] and Buchberger [11]. It has been proved to be very useful in different branches of mathematics, including commutative algebras and combinatorial algebras, see [7, 8, 9]. Ab- stract rewriting system is a branch of theoretical computer science, combining elements of logic, universal algebra, automated theorem proving and functional programming [2, 32]. The theories of Grobner-Shirshov bases and rewriting systems are successfully applied to study operators and operator polynomial identities [18, 23]. In the present paper along this line, using the theories of Grobner-Shirshov bases and rewriting systems, we construct a basis of the free unitary (noncommutative) averaging algebra on a non- empty set. Terminating and confluence are essential and desirable properties of a rewriting system. To use the tools of Grobner-Shirshov bases and rewriting systems, we obtain a sufficient and necessary condition for the confluence of the kind of rewriting system we used. We supply the relationship between Grobner-Shirshov bases and rewriting systems based on bracketed poly- nomials. Applying the method we obtained for checking confluence, we successfully prove that the rewriting system associated to the averaging operator is confluent and then convergent with a suitable order. Let us emphasize that there are a lot of forks in the process of checking confluence. We handle technically most of them in a unified way. These techniques can also be used to study other operators. It is well known that in the category of any given algebraic structure, the free objects play a central role in study other objects. Thus as an application, we give a basis of the free unitary (noncommutative) averaging algebra on a non-empty set. Our characterization of averaging operators in terms of Grobner-Shirshov bases and rewriting systems reveals the power of this approach. It would be interesting to further study operators and operator polynomial identities by making use of the two related theories: Grobner-Shirshov bases and rewriting systems. The layout of the paper is as follows. In Section 2, we first recall the concepts of averaging algebras and free operated algebras. We next recall some necessary backgrounds of Grobner- Shirshov bases and rewriting systems. We obtain some basic results on the kind of rewriting system we used. In particular, we obtain a sufficient and necessary condition to characterize the confluence (Theorem 2.36). We end this section by supplying the relationship between the two powerful tools -- Grobner-Shirshov bases and rewriting systems (Theorem 2.41). Section 3 is devoted to a basis of the free unitary averaging algebra on a non-empty set. In order to achieve this purpose, we assign a rewriting system to the averaging operator (Eq. (19)). We show this rewriting system is convergent (Theorem 3.10). We end this section by giving a basis of the free unitary (noncommutative) averaging algebra on a non-empty set (Theorem 3.11). Some remark on notation. We fix a domain k and a non-empty set X. Denote by k× := k \ {0} the subset of nonzero elements. We denote the k-span of a set Y by kY. For an algebra, we mean a unitary associative noncommutative k-algebra, unless specified otherwise. For any set Y, let M(Y) be the free monoid on Y with identity 1. We use ⊔ for disjoint union. 2. Grobner-Shirshov bases and rewriting systems In this section, we first recall the definition of averaging algebras and characterize free av- eraging algebras as quotients of free operated algebras. We then recall some backgrounds on Grobner-Shirshov bases and rewriting systems. AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 3 2.1. Free averaging algebras. An averaging algebra in the noncommutative context is given as follows. Definition 2.1. A linear operator A on a k-algebra R is called an averaging operator if A(u1)A(u2) = A(A(u1)u2) = A(u1A(u2)) for all u1, u2 ∈ R. A k-algebra R together with an averaging operator A on R is called an averaging algebra. To characterize the free averaging algebra, let us recall the free operated algebra [9, 20, 28]. Definition 2.2. An operated monoid (resp. operated k-algebra, resp. operated k-module) is a monoid (resp. k-algebra, resp. k-module) U together with a map (resp. k-linear map, resp. k-linear map) PU : U → U. A morphism from an operated monoid (resp. k-algebra, resp. k-module) (U, PU) to an operated monoid (resp. k-algebra, resp. k-module) (V, PV) is a monoid (resp. k-algebra, resp. k-module) homomorphism f : U → V such that f ◦ PU = PV ◦ f . For any set Y, define ⌊Y⌋ := {⌊y⌋ y ∈ Y}, which is a disjoint copy of Y. The following is the construction of the free operated monoid on the set X, proceeding via the finite stage Mn(X) recursively defined as follows. Define M0(X) := M(X) and M1(X) := M(X ⊔ ⌊M0(X)⌋). Then the inclusion X ֒→ X ⊔ ⌊M0⌋ induces a monomorphism i0 : M0(X) = M(X) ֒→ M1(X) = M(X ⊔ ⌊M0⌋) of monoids through which we identify M0(X) with its image in M1(X). Suppose that Mn−1(X) has been defined and the embedding in−2,n−1 : Mn−2(X) ֒→ Mn−1(X) has been obtained for n > 2 and consider the case of n. Define Since Mn−1(X) = M(cid:0)X ⊔ ⌊Mn−2(X)⌋(cid:1) is the free monoid on the set X ⊔ ⌊Mn−2(X)⌋, the injection Mn(X) := M(cid:0)X ⊔ ⌊Mn−1(X)⌋(cid:1). X ⊔ ⌊Mn−2(X)⌋ ֒→ X ⊔ ⌊Mn−1(X)⌋ induces a monoid embedding Mn−1(X) = M(cid:0)X ⊔ ⌊Mn−2(X)⌋(cid:1) ֒→ Mn(X) = M(cid:0)X ⊔ ⌊Mn−1(X)⌋(cid:1). Finally we define the monoid M(X) := lim −→ Mn(X) = [ n>0 Mn(X). The elements in M(X) are called bracketed words or bracketed monomials on X. When X is finite, we may also just list its elements, as in M(x1, x2) if X = {x1, x2}. For any u ∈ M(X) \ {1}, u can be written uniquely as a product: (1) u = u1 · · · un, for some n > 1, ui ∈ X ⊔ ⌊M(X)⌋, 1 6 i 6 n. The breadth of u, denoted by u, is defined to be n. If u = 1, define u = 0. 4 XING GAO AND TIANJIE ZHANG Let kM(X) be the free module with the basis M(X). Using k-linearity, the concatenation product on M(X) can be extended to a multiplication on kM(X), turning kM(X) into a k-algebra. Define an operator ⌊ ⌋ : M(X) → M(X) by assigning u 7→ ⌊u⌋, u ∈ M(X). By k-linearly, the operator ⌊ ⌋ : M(X) → M(X) can be extended to a linear operator ⌊ ⌋ : kM(X) → kM(X), turning (kM(X), ⌊ ⌋) into an operated k-algebra. The elements in kM(X) are called bracketed polynomials or operated polynomials on X. Lemma 2.3. [20, Coro. 3.6, 3.7] With structures as above, (a) the (M(X), ⌊ ⌋) together with the natural embedding i : X → M(X) is the free operated monoid on X; and (b) the (kM(X), ⌊ ⌋) together with the natural embedding i : X → kM(X) is the free operated k-algebra on X. Definition 2.4. Let (R, P) be an operated k-algebra. (a) An element φ(x1, . . . , xk) ∈ kM(X) (or φ(x1, . . . , xk) = 0) is called an operated polynomial identity (OPI), where k > 1 and x1, . . . , xk ∈ X. (b) Let φ = φ(x1, . . . , xk) ∈ kM(X) be an OPI. Given any u1, . . . , uk ∈ R, there is a set map f : xi 7→ ui, 1 6 i 6 k and we define φ(u1, . . . , uk) := ef (φ(x1, . . . , xk)), where ef : kM(x1, . . . , xk) → R is the unique morphism of operated algebras that extends the set map f , using the universal property of kM(x1, . . . , xk) as the free operated k-algebra on {x1, . . . , xk}. Informally, φ(u1, . . . , uk) is the element of R obtained from φ(x1, . . . , xk) by replacing each xi by ui, 1 6 i 6 k. (c) Let Φ ⊆ kM(X) be a set of OPIs. We call Φ is satisfied by R if φ(u1, . . . , uk) = 0, ∀φ(x1, . . . , xk) ∈ Φ, ∀u1, . . . , uk ∈ R. In this case, we speak that R is a Φ-algebra and P is a Φ-operator. (d) Let S ⊆ kM(X) be a set. The operated ideal Id(S ) of kM(X) generated by S is the smallest operated ideal containing S . Let us proceed some examples. Example 2.5. The differential operator as an algebraic abstraction of derivation in analysis leads to the differential algebra, which is an algebraic study of differential equations and has been largely successful in many important areas [27, 33, 35]. The differential operator d = ⌊ ⌋ fulfils the following OPI φ(x1, x2) = ⌊x1x2⌋ − ⌊x1⌋x2 − x1⌊x2⌋. Example 2.6. The Rota-Baxter operator P = ⌊ ⌋ of weight λ has played important role in mathe- matics and physics[4, 21, 37], satisfying the OPI φ(x1, x2) = ⌊x1⌋⌊x2⌋ − ⌊x1⌊x2⌋⌋ − ⌊⌊x1⌋x2⌋ − λ⌊x1x2⌋, where λ ∈ k is a fixed constant. AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 5 Example 2.7. From Definition 2.1, the averaging operator A = ⌊ ⌋ (noncommutative) is defined by the OPIs (2) φ(x1, x2) = ⌊x1⌋⌊x2⌋ − ⌊⌊x1⌋x2⌋, ψ(x1, x2) = ⌊x1⌊x2⌋⌋ − ⌊⌊x1⌋x2⌋. Example 2.8. O. Reynolds [34] introduced the concept of Reynolds operators into fluid dynam- ics, and Kamp´e de F´eriet [14] named it in his study on the various spaces of functions. The Reynolds operator is defined by the OPI Definition 2.9. (3) φ(x1, x2) = ⌊⌊x1⌋⌊x2⌋⌋ + ⌊x1⌋⌊x2⌋ − ⌊x1⌊x2⌋⌋ − ⌊⌊x1⌋x2⌋. (a) Let φ = φ(x1, . . . , xk) ∈ kM(X) be an OPI with k > 1. Define S φ(X) := { φ(u1, . . . , uk) u1, . . . , uk ∈ M(X) }. (b) Let Φ be a set of OPIs. Define (4) It is well-known that S Φ(X) := [ φ∈Φ S φ(X). Proposition 2.10. [13, Prop. 1.3.6] Let Φ ⊆ kM(X) a set of OPIs. Then the quotient operated algebra kM(X)/Id(S Φ(X)) is the free Φ-algebra on X. In particular, we have Proposition 2.11. Let φ(x1, x2), ψ(x1, x2) defined in Eq. (2). Then the quotient operated algebra kM(X)/Id(S φ(X) ∪ S ψ(X)) is the free averaging algebra on X. 2.2. Grobner Shirshov bases. In this subsection, we provide some backgrounds on Grobner- Shirshov bases [9, 19, 23]. Definition 2.12. Let ⋆ be a symbol not in X and X⋆ = X ⊔ {⋆}. (a) By a ⋆-bracketed word on X, we mean any bracketed word in M(X⋆) with exactly one occurrence of ⋆, counting multiplicities. The set of all ⋆-bracketed words on X is denoted by M⋆(X). obtained by replacing the symbol ⋆ in q by u. (b) For q ∈ M⋆(X) and u ∈ M(X), we define qu := q⋆7→u to be the bracketed word on X (c) For q ∈ M⋆(X) and s = Pi ciui ∈ kM(X), where ci ∈ k and ui ∈ M(X), we define qs := X i ciqui . (d) A bracketed word u ∈ M(X) is a subword of another bracketed word w ∈ M(X) if w = qu for some q ∈ M⋆(X). Generally, with ⋆1, ⋆2 distinct symbols not in X, set X⋆2 := X ⊔ {⋆1, ⋆2}. (e) We define an (⋆1, ⋆2)-bracketed word on X to be a bracketed word in M(X⋆2) with exactly one occurrence of each of ⋆i, i = 1, 2. The set of all (⋆1, ⋆2)-bracketed words on X is denoted by M⋆1,⋆2(X). (f) For q ∈ M⋆1,⋆2(X) and u1, u2 ∈ kM⋆1,⋆2(X), we define qu1,u2 := q⋆17→u1,⋆27→u2 to be obtained by replacing the letters ⋆i in q by ui for i = 1, 2. 6 XING GAO AND TIANJIE ZHANG Remark 2.13. Recall [23] that qu1,u2 = (q⋆1u1)u2 = (q⋆2u2)u1, where q⋆1 is viewed as a ⋆1- bracketed word on X ⊔ {⋆2} and q⋆2 as a ⋆2-bracketed word on X ⊔ {⋆1}. We record the following obvious properties of subwords, which will be used later. Lemma 2.14. Let u, v, w ∈ M(X). (a) If u is a subword of ⌊v⌋, then either u = ⌊v⌋ or u is a subword of v. (b) If ⌊u⌋ is a subword of vw, then either ⌊u⌋ is a subword of v or ⌊u⌋ is a subword of w. Proof. (a) Suppose u , ⌊v⌋. Since u is a subword of ⌊v⌋, then ⌊v⌋ = qu for some q ∈ M⋆(X) by Definition 2.12 (d). Since u , ⌊v⌋, it follows that q , ⋆. Thus q = ⌊p⌋ for some p ∈ M⋆(X) by ⌊v⌋ = qu. Therefore ⌊v⌋ = qu = ⌊pu⌋ and so v = pu, as required. (b). This is followed by the breadth of ⌊u⌋ is 1. The operated ideals in kM(X) can be characterized by ⋆-bracketed words [9, 23]. (cid:3) . Lemma 2.15. ([23, Lem. 3.2]) Let S ⊆ kM(X). Then (5) Id(S ) =  nX i=1 ciqisi (cid:12)(cid:12)(cid:12)(cid:12) n > 1 and ci ∈ k×, qi ∈ M⋆(X), si ∈ S for 1 6 i 6 n Definition 2.16. A monomial order on M(X) is a well-order 6 on M(X) such that u < v =⇒ qu < qv, ∀u, v ∈ M(X), ∀q ∈ M⋆(X). Definition 2.17. Let s ∈ kM(X) and 6 a linear order on M(X). (a) Let s < k. The leading monomial of s, denoted by s, is the largest monomial appearing in s. The leading coefficient of s, denoted by cs, is the coefficient of s in s. (b) If s ∈ k, we define the leading monomial of s to be 1 and the leading coefficient of s to be cs = s. (c) s is called monic with respect to 6 if s < k and cs = 1. A subset S ⊆ kM(X) is called monic with respect to 6 if every s ∈ S is monic with respect to 6. (d) Define R(s) := css − s. So s = css − R(s). We will not need the precise definition of Grobner-Shirshov bases for our construction. So we will not recall it for now and the authors are refereed to [7] and references therein. Suffices it to say that we need the Composition-Diamond Lemma, the corner stone of Grobner-Shirshov basis theories. Lemma 2.18. (Composition-Diamond Lemma [9, 23]) Let 6 a monomial order on M(X) and S ⊆ kM(X) monic with respect to 6. Then the following conditions are equivalent. (a) S is a Grobner-Shirshov basis in kM(X). (b) η(Irr(S )) is a k-basis of kM(X)/Id(S ), where η : kM(X) → kM(X)/Id(S ) is the canonical homomorphism of k-modules and (6) Irr(S ) := M(X) \ {qs s ∈ S }. More precisely as k-modules, kM(X) = kIrr(S ) ⊕ Id(S ). AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 7 2.3. Term-rewriting systems. In this subsection, we give a method for checking confluence of term-rewriting systems. Let us recall some basic notations and results [18]. Definition 2.19. Let V be a free k-module with a given k-basis W and f , g ∈ V. (a) The support Supp( f ) of f is the set of monomials (with non-zero coefficients) of f . Here we use the convention that Supp(0) = ∅. (b) We write f ∔ g to indicate that Supp( f ) ∩ Supp(g) = ∅ and say f + g is a direct sum of f and g. If this is the case, we also use f ∔ g for the sum f + g. (c) For w ∈ Supp( f ) with the coefficient cw, we define Rw( f ) := cww − f ∈ V and so f = cww ∔ (−Rw( f )). Lemma 2.20. [18, Lem. 2.12] Let V be a free k-module with a k-basis W and f , g ∈ V. If f ∔ g, then c f ∔ dg for any c, d ∈ k. Remark 2.21. Using the notation ∔, the equation s = css − R(s) in Definition 2.17 (d) can be written in more detail as s = css ∔ (−R(s)). The following is the concept of term-rewriting systems. Definition 2.22. Let V be a free k-module with a k-basis W. A term-rewriting system Π on V with respect to W is a binary relation Π ⊆ W × V. An element (t, v) ∈ Π is called a (term) rewriting rule of Π, denoted by t → v. The term-rewriting system Π is called simple if t ∔ v for all t → v ∈ Π. Remark 2.23. Now we explain the requirement that the term-rewriting system Π is simple. Suppose Π is not simple. Then by Definition 2.22, there is a rewriting rule t → v such that t ∈ Supp(v). Assume v = ct ∔ (−Rt(v)) for some c ∈ k×. Then t →Π v = ct ∔ (−Rt( f )) →Π cv − Rt(v) = c2t ∔ (−c − 1)Rt(v) →Π · · · . So as long as c is not a nilpotent element, Π is not terminating. In the remainder of this paper, we always assume that the term-rewriting system is simple, unless specified otherwise. Definition 2.24. Let V be a free k-module with a k-basis W, Π a simple term-rewriting system on V with respect to W and f , g ∈ V. (a) We speak that f rewrites to g in one-step, denoted by f →Π g or f (t,v) −→Π g, if f = ctt ∔ (−Rt( f )) and g = ctv − Rt( f ) for some ct ∈ k× and t → v ∈ Π. (b) The reflexive-transitive closure of the binary relation →Π on V is denoted by ∗ If f rewrites (resp. doesn't rewrite ) to g with respect →Π. ∗ →Π g (resp. f 6 ∗ →Π g), we speak that f to Π. (c) We call f and g are joinable, denoted by f ↓Π g, if there exists h ∈ V such that f and g ∗ →Π h. (d) We say f a normal form if no more rewriting rules can apply. Remark 2.25. Let f , g ∈ V. (a) By Definition 2.24 (b), f ∗ →Π f and f ∗ → g ⇐⇒ f =: f0 →Π f1 →Π · · · →Π fn := g for some n > 0, fi ∈ V, 0 6 i 6 n. ∗ →Π h 8 (b) If f XING GAO AND TIANJIE ZHANG ∗ →Π g, then f ↓Π g by g ∗ →Π g. In particular, f ↓Π f by f ∗ →Π f . The following definitions are adapted from abstract rewriting systems [2, 5]. Definition 2.26. Let V be a free k-module with a k-basis W, Π a simple term-rewriting system on V with respect to W. (a) Π is terminating if there is no infinite chain of one-step rewriting f0 →Π f1 →Π f2 · · · . (b) f ∈ V is locally confluent if for every local fork (h Π← f →Π g), we have g ↓Π h. (c) f ∈ V is confluent if for every fork (h Π (d) Π is locally confluent (resp. confluent) if every f ∈ V is locally confluent (resp. confluent). (e) Π is convergent if it is both terminating and confluent. ∗ →Π g), we have g ↓Π h. ∗ ← f A well-known result on rewriting systems is Newman's Lemma. Lemma 2.27. ([2, Lem. 2.7.2]) A terminating rewriting system is confluent if and only if it is locally confluent. The following result will be used later. Lemma 2.28. ([18, Thm. 2.20]) Let V be a free k-module with a k-basis W and Π a simple If Π is confluent, then, for all m > 1 and term-rewriting system on V with respect to W. f1, . . . , fm, g1, . . . , gm ∈ V, fi ↓Π gi (1 6 i 6 m), and mX i=1 gi = 0 =⇒  mX i=1 fi ∗ →Π 0. Remark 2.29. If Π is confluent and f ↓Π g, then f − g ∗ →Π 0 by −g ↓Π −g and Lemma 2.28. The following is a concept strong than locally confluence and similar to Buchberger's S - polynomials. Definition 2.30. Let V be a free k-module with a k-basis W, Π a simple term-rewriting system on V with respect to W. (a) A local base-fork is a fork (cv1 Π← ct →Π cv2), where c ∈ k× and t → v1, t → v2 ∈ Π are rewriting rules. (b) The term-rewriting system Π is called locally base-confluent if for every local base-fork (cv1 Π← ct →Π cv2), we have c(v1 − v2) ∗ →Π 0. (c) Π is compatible with a linear order 6 on W if v < t for each t → v ∈ Π. Lemma 2.31. ([18, Lem. 2.22]) Let V be a free k-module with a k-basis W and let Π be a simple term-rewriting system on V which is compatible with a well order 6 on W. If Π is locally base-confluent, then it is locally confluent. The following concept is followed from general abstract rewriting systems [5, Def. 1.1.6]. Definition 2.32. Let V be a free k-modules with a k-basis W and let Π be a simple term-rewriting system on V with respect to W. Let Y ⊆ W and ΠkY := Π ∩ (Y × kY). We call ΠkY a sub-term- rewriting system of Π on kY with respect to Y, denoted by ΠkY 6 Π, if kY is closed under Π, i.e., for any f ∈ kY and any g ∈ V, f →Π g implies g ∈ kY. AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 9 Remark 2.33. Since Π is simple, ΠkY is also simple. Indeed, let t → v ∈ ΠkY be a rewriting rule with t ∈ Y and v ∈ kY. Then t → v ∈ Π by ΠkY ⊆ Π. Since Π is simple, we have t < Supp(v) by Definition 2.22 and so ΠkY is simple. We record the following properties. Lemma 2.34. Let V be a free k-module with a k-basis W, and let Π be a simple term-rewriting system on V with respect to W. (a) If t ∈ Supp(c f ) with t ∈ W, c ∈ k× and f ∈ V, then t ∈ Supp( f ). (b) If c f →Π g with c ∈ k× and f , g ∈ V, then g = cg′ for some g′ ∈ V. (c) If c f = 0 with c ∈ k× and f ∈ V, then f = 0. (d) For c ∈ k× and f , g ∈ V with f , g, f →Π g if and only if c f →Π cg. Proof. (a) Suppose for a contrary that t < Supp( f ). Since W is a k-basis of V, by Defini- tion 2.19 (a), we may write f = Pi ciwi, where each ci ∈ k× and wi ∈ W \ {t}. Then c f = Pi cciwi. Since wi , t for each i, we have t < Supp(c f ), a contradiction. (t,v) −→Π g for some t → v ∈ Π. Then t ∈ Supp(c f ) and so t ∈ Supp( f ) by Item (a). (b) Suppose c f Write f = ctt ∔ (−Rt( f )) with ct ∈ k×. Then by Lemma 2.20, c f = cctt ∔ (−cRt( f )) (t,v) −→Π cctv − cRt( f ) = c(ctv − Rt( f )) = g, as required. (c) Since W is a k-basis of V, we may write f = Pi ciwi with ci ∈ k and wi ∈ W for each i. Then c f = Pi cciwi = 0 and so cci = 0 for each i. Since k is a domain by our hypothesis and c , 0, we have ci = 0 for each i, that is, f = 0. (d) Suppose f (t,v) −→Π g for some t → v ∈ Π. By Definition 2.24 (a), we may write f = dt ∔ (−Rt( f )) and g = dv − Rt( f ) for some d ∈ k×. Then by Lemma 2.20, c f = cdt ∔ (−cRt( f )) and cg = cdv − cRt( f ) and so c f so t ∈ Supp( f ) by Item (a). Write f = ctt ∔ (−Rt( f )) with ct ∈ k×. Then from Lemma 2.20, (t,v) −→Π cg for some t → v ∈ Π. Then t ∈ Supp(c f ) and (t,v) −→Π cg. Conversely, suppose c f c f = cctt ∔ (−cRt( f )) (t,v) −→Π cctv − cRt( f ) = cg. Since c ∈ k×, we get ctv − Rt( f ) = g by Item (c) and so f →Π g. Lemma 2.35. Let V be a free k-module with a k-basis W, and let Π be a simple term-rewriting system on V with respect to W. Let f , g ∈ V and c ∈ k×. Then f ∗ →Π g if and only if c f ∗ →Π cg. (cid:3) Proof. (⇒) If f = g, then c f = cg and c f be the least number such that f rewrites to g in n steps. Then ∗ →Π cg by Remark 2.25 (a). Suppose f , g. Let n > 1 (7) for some distinct fi ∈ V, 0 6 i 6 n and so by Lemma 2.34 (d), (8) f = f0 →Π f1 →Π · · · →Π fn = g c f = c f0 →Π c f1 →Π · · · →Π c fn = cg. Hence c f ∗ →Π cg. 10 XING GAO AND TIANJIE ZHANG (⇐) If c f = cg, then f = g by Lemma 2.34 (c) and so f ∗ →Π g by Remark 2.25 (a). Suppose c f , cg. Let n > 1 be the least number such that c f rewrites to cg in n steps. Then by Lemma 2.34 (b), Eq (8) holds for some distinct c fi ∈ V, 0 6 i 6 n. Using Lemma 2.34 (c), fi ∈ V are distinct for 0 6 i 6 n. From Lemma 2.34 (d), Eq. (7) is valid and so f (cid:3) ∗ →Π g. Theorem 2.36. Let V be a free k-module with a k-basis W and let Π be a simple terminating term-rewriting system on V with respect to W. Suppose 6 is a well-order on W compatible with Π. Then Π is confluent if and only if w is locally confluent for any w ∈ W. Proof. (⇒) Since Π is confluent, Π is locally confluent by Definition 2.26, that is, every element in V is locally confluent. From W ⊆ V, w is locally confluent for any w ∈ W. (⇐) To show Π is confluent, it is enough to show Π is locally confluent by Lemma 2.27. In view of Lemma 2.31, we are left to prove that Π is locally base-confluent, that is, for any local ∗ →Π 0. Suppose for a contrary that Π is not base-fork (cv1 Π← cw →Π cv2), we have cv1 − cv2 locally base-confluent. Then the set C = {w ∈ W (cid:12)(cid:12)(cid:12)(cid:12) there is a local fork base-fork (cv1 Π← cw →Π cv2) for some c ∈ k×, v1, v2 ∈ V such that cv1 − cv2 6 ∗ →Π 0} is non-empty. Since 6 is a well-order, C has the least element w with respect to 6. Thus there is a local base-fork (9) such that (10) Let (11) (cv1 Π← cw →Π cv2) with w → v1, w → v2 ∈ Π cv1 − cv2 6 ∗ →Π 0 for some c ∈ k×, v1, v2 ∈ V. Y := {y ∈ W y < w} and ΠkY = Π ∩ (Y × kY). Since 6 is compatible with Π, we have Supp(v1), Supp(v2) ⊆ Y and so Y , ∅. Furthermore, ∗ →Π g with f ∈ kY, since 6 is ΠkY 6 Π is a sub-term-rewriting system of Π. Indeed, let f compatible with Π, we get g 6 f < w and so g ∈ kY. Thus ΠkY is closed under Π and so ΠkY 6 Π by Definition 2.32. For any local base-fork (du1 ΠkY ← dy →ΠkY du2) of ΠkY with d ∈ k×, y ∈ Y and u1, u2 ∈ kY, it induces a local base-fork (du1 Π← dy →Π du2) by ΠkY ⊆ Π. Since y ∈ Y, we have y < w and y < C by the minimality of w. So du1 − du2 ∗ →Π 0 by the definition of C. Claim (12) f ∗ →Π g =⇒ f ∗ →ΠkY g for f , g ∈ kY. ∗ Since du1 − du2 ∈ kY by u1, u2 ∈ kY, we have du1 − du2 →ΠkY 0 by the Claim. Thus ΠkY is locally base-confluent and so is locally confluent by Lemma 2.31. Since Π is terminating and ΠkY 6 Π, ΠkY is terminating. Therefore ΠkY is confluent by Lemma 2.27. For the local fork in Eq. (9), it induces a local fork (v1 Π← w →Π v2) by Lemma 2.34 (d). Since ∗ →Π u w ∈ W is confluent by our hypothesis, it follows that v1 ↓Π v2. So there is u ∈ V such that v1 and v2 ∗ →Π u by Definition 2.24 (c). From Lemma 2.35, cv1 ∗ →Π cu and cv2 ∗ →Π cu. AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 11 From cv1 ∈ kY and ΠkY 6 Π is closed under Π, we have cu ∈ kY. So by the Claim of Eq. (12), ∗ →ΠkY cu. This means that cv1 ↓ΠkY cv2. Since ΠkY is confluent, cv1 − cv2 cv1 − cv2 ∗ →ΠkY cu and cv2 ∗ →Π 0 by ΠkY ⊆ Π, contradicting Eq. (10). We are left to prove the Claim. ∗ →ΠkY 0 by Remark 2.29. Hence (cid:3) cv1 proof of Claim. We want to show Eq. (12). Suppose f f ∗ →ΠkY g by Remark 2.25 (a). Assume f , g and let n > 1 be least number such that ∗ →Π g with f , g ∈ kY. If f = g, then f =: f0 →Π f1 →Π · · · →Π fn := g with fi ∈ V are distinct, 0 6 i 6 n. Since f0 = f ∈ kY and Π is compatible with 6, we have fi ∈ kY for 0 6 i 6 n. We prove the (t,v) → Π f1 = g for some Claim by induction on n > 1. For the initial step of n = 1, suppose f = f0 t → v ∈ Π. Then t ∈ Supp( f ) ⊆ Y. This follows that t < w by Eq. (11). Since Π is compatible with 6, we have v < t < w and so v ∈ kY. Thus t → v ∈ Y ×kY and so t → v ∈ Π∩(Y ×kY) = ΠkY. (t,v) −→Π f1. For the induction step, This implies that f = f0 ∗ we have f = f0 →ΠkY g, as required. (cid:3) f1 = g by f0, f1 ∈ kY and f0 ∗ →ΠkY fn = g by induction hypothesis and so f (t,v) −→ΠkY f1 and f1 ∗ →ΠkY 2.4. Term-rewriting systems and Grobner-Shirshov bases. In this subsection, we supply the relationship between Grobner-Shirshov bases and term-rewriting systems based on bracketed polynomials. A term-rewriting system can be assigned to a given set S of OPIs [18]. Definition 2.37. Let 6 be a linear order on M(X) and S ⊆ kM(X) monic with respect to 6. Define a term-rewriting system associated to S as (13) ΠS := { qs → qR(s) s = s ∔ (−R(s)) ∈ S , q ∈ M⋆(X) } ⊆ M(X) × kM(X). For notation clarify, we denote →ΠS (resp. ∗ →S , resp. ↓S ). In more detail when a specific s ∈ S is used in one step rewriting, we replace →S by →s. If 6 is a monomial order on M(X), we have qR(s) = qR(s) < qs by R(s) < s. So ΠS is compatible with 6 in the sense in Definition 2.30 (c). Remark 2.38. Let f , g ∈ kM(X). ∗ →ΠS , resp. ↓ΠS ) by →S (resp. (a) If f →S g, then we can write f = cqs ∔ f ′ and g = cqR(s) + f ′ for some c ∈ k×, q ∈ M⋆(X), s ∈ S and f ′ ∈ kM(X) by Definition 2.24 (a). So f − g = cqs−R(s) = cqs ∈ Id(S ) by Lemma 2.15. ∗ →S g, then f =: f0 →S f1 →S · · · →S fn := g for some n > 0, fi ∈ kM(X), 0 6 i 6 n. (b) If f If n = 0, then f = g and f − g ∈ Id(S ). If n > 1, then by Item (a), f − g = ( f0 − f1) + ( f1 − f2) + · · · + ( fn−1 − fn) ∈ Id(S ). Lemma 2.39. If u ∔ v, then qu ∔ qv for any q ∈ M⋆(X) and u, v ∈ kM(X). Proof. Write u = Pi ciui and v = P j d jv j, where each ci, d j ∈ k× and ui, v j ∈ M(X). Then d jqv j . j qu = X ciqui and qv = X i Suppose for a contrary that qu ∔ qv fails. Then qui = qv j by Definition 2.19 for some i, j. This implies that ui = v j ∈ Supp(u) ∩ Supp(v), contradicting that u ∔ v. (cid:3) 12 XING GAO AND TIANJIE ZHANG The following results are characterized in [17]. For completeness, we record the proof here. Lemma 2.40. Let 6 be a linear order on M(X) and S ⊆ kM(X) monic with respect to 6. (a) If ΠS is confluent, then, u ∈ Id(S ) if and only if u ∗ (b) If ΠS is confluent, then Id(S ) ∩ kIrr(S ) = 0. (c) If ΠS is terminating and Id(S ) ∩ kIrr(S ) = 0, then ΠS is confluent. (d) If ΠS is terminating, then kM(X) = Id(S ) + kIrr(S ), →ΠS 0. where Irr(S ) = M(X) \ {qs s ∈ S }. Proof. Note that kIrr(S ) is precisely the set of normal forms of ΠS . (a) If u ∗ have →ΠS 0, then u ∈ Id(S ) by Remark 2.38 (b). Conversely, let u ∈ Id(S ). By Eq. (5), we u = nX i=1 ciqisi, where ci ∈ k×, si ∈ S , qi ∈ M⋆(X), 1 6 i 6 n. For each si = si ∔ (−R(si)) with 1 6 i 6 n, it follows from Lemmas 2.20 and 2.39 that ∔ (−ciqiR(si)) →ΠS ciqiR(si) − ciqiR(si) = 0 and so ciqisi ↓ΠS 0 ciqisi = ciqisi by Remark 2.25 (b). Since ΠS is confluent, u = Pn (b) If Id(S ) ∩ kIrr(S ) , 0, let 0 , w ∈ Id(S ) ∩ kIrr(S ). Since w ∈ kIrr(S ), w is of normal form. On the other hand, from w ∈ Id(S ) and Item (a), we have w ∗ →ΠS 0. So w has two normal forms w and 0, contradicting that ΠS is confluent. ∗ →ΠS 0 by Lemma 2.28. i=1 ciqisi (c) Suppose for a contrary that ΠS is not confluent. Since ΠS is terminating, there is w ∈ kM(X) such that w has two distinct normal forms, say u and v. Thus u, v ∈ kIrr(S ) and so u − v ∈ kIrr(S ). Since w ∗ → vΠ, we have w − u, w − v ∈ Id(S ) by Remark 2.38 (b). Hence 0 , u − v ∈ Id(S ) ∩ kIrr(S ), a contradiction. →Π u and w ∗ (d) Let w ∈ kM(X), since ΠS is terminating, w has a normal form u ∈ kIrr(S ) and w ∗ →Π u. (cid:3) From Remark 2.38 (b), we have w − u ∈ Id(S ) and so w ∈ Id(S ) + kIrr(S ). Theorem 2.41. Let 6 be a monomial order on M(X) and S ⊆ kM(X) monic with respect to 6. Then the followings are equivalent. (a) ΠS is convergent. (b) ΠS is confluent. (c) Id(S ) ∩ kIrr(S ) = 0. (d) Id(S ) ⊕ kIrr(S ) = kM(X). (e) S is a Gr obner-Shirshov basis in kM(X), where Irr(S ) = M(X) \ {qs s ∈ S }. Proof. Since 6 is a monomial order on M(X), ΠS is terminating [18]. So Item (a) and Item (b) are equivalent. The equivalence of Item (b) and Item (c) is followed from Items (b) and (c) in Lemma 2.40. Clearly, Item (d) implies Item (c). The converse is employed Item (d) in Lemma 2.40. At last, (cid:3) the equivalence of Item (d) and Item (e) is obtained from Lemma 2.18. 3. A basis of the free averaging algebra In this section, we give a basis of the free averaging algebra. We begin with a lemma. Lemma 3.1. Let S ⊆ kM(X), q ∈ M⋆(X) and 6 a linear order on M(X). Then AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 13 (a) If u ∗ (b) If u ↓S v, then qu ↓S qv. →S v with u, v ∈ kM(X), then qu ∗ →S qv. ∗ Proof. (a) If u = v, then qu = qv and qu →S qv by Remark 2.25 (a). Suppose u , v. Let m > 1 be the least number such that u rewrites to v in m steps. We prove the result by induction on m. For the initial step m = 1, since u →S v, we may write u = cps ∔ u′ and v = cpR(s) + u′ for some c ∈ k×, s ∈ S , p ∈ M⋆(X), u′ ∈ kM(X). Then from Lemma 2.39, qu = c(qp)s ∔ qu′ →S c(qp)R(s) + qu′ = qcpR(s) +u′ = qv. Assume the result is true for m 6 n and consider the case of m = n + 1 > 2. Then we can write u →S w ∗ →S v for some u , w ∈ kM(X). By the minimality of m, we have w , v. Using induction hypothesis, we get qu ∗ →S qv. This implies that qu ∗ →S qw and qw (b) Since u ↓S v, we may suppose by Definition 2.24 (c) that u ∗ w ∈ kM(X). Then by Item (a), we have qu the proof. ∗ →S qw and qv ∗ →S qv, as required. →S w and v ∗ →S w for some ∗ →S qw. So qu ↓S qv. This completes (cid:3) The following is a concept finer than subwords, including the information of placements [40]. Definition 3.2. Let w ∈ M(X) such that (14) The two placements (u1, q1) and (u2, q2) are called q1u1 = w = q2u2 for some u1, u2 ∈ M(X), q1, q2 ∈ M⋆(X). (a) separated if there exist p ∈ M⋆1,⋆2(X) and a, b ∈ M(X) such that q1⋆1 = p⋆1, b, q2⋆2 = pa, ⋆2, and w = pa, b; (b) nested if there exists q ∈ M⋆(X) such that either q2 = q1q or q1 = q2q; (c) intersecting if there exist q ∈ M⋆(X) and a, b, c ∈ M(X)\{1} such that w = qabc and either (i) q1 = q⋆c and q2 = qa⋆; or (ii) q1 = qa⋆ and q2 = q⋆c. Lemma 3.3. exactly one of the following is true : [40, Thm. 4.11] Let w ∈ M(X). For any two placements (u1, q1) and (u2, q2) in w, (a) (u1, q1) and (u2, q2) are separated ; (b) (u1, q1) and (u2, q2) are nested ; (c) (u1, q1) and (u2, q2) are intersecting. Now we fix some notations which will be used through out the remainder of the paper. For any u ∈ M(X), define recursively ⌊u⌋(1) := ⌊u⌋ and ⌊u⌋(k+1) := ⌊⌊u⌋(k)⌋ for k > 1. Recall from Example 2.7 that φ(x1, x2) := ⌊x1⌋⌊x2⌋ − ⌊⌊x1⌋x2⌋ and ψ(x1, x2) := ⌊x1⌊x2⌋⌋ − ⌊⌊x1⌋x2⌋ are the OPIs defining the averaging operator. Let 6 be a well-order on X such that x1 < x2. Then 6 can be extended to the monomial order 6db on M(X) [18], which will be used through out in the remainder of the paper. With respect to 6db, we have (15) φ(x1, x2) =⌊x1⌋⌊x2⌋, R(φ(x1, x2)) = ⌊⌊x1⌋x2⌋, ψ(x1, x2) =⌊x1⌊x2⌋⌋, R(ψ(x1, x2)) = ⌊⌊x1⌋x2⌋. 14 XING GAO AND TIANJIE ZHANG The term-rewriting system associated to φ(x1, x2), ψ(x1, x2) is not confluent. For example, for the element ⌊⌊x1⌋⌊x2⌋⌋ ∈ M(X), on the one hand, which is in normal form. On the other hand, ⌊⌊x1⌋⌊x2⌋⌋ →φ(x1,x2) ⌊⌊⌊x1⌋x2⌋⌋ = ⌊⌊x1⌋x2⌋(2), ⌊⌊x1⌋⌊x2⌋⌋ →ψ(x1,x2) ⌊⌊⌊x1⌋⌋x2⌋ = ⌊⌊x1⌋(2)x2⌋, ϕ(x1, x2) := ⌊⌊x1⌋x2⌋(2) − ⌊⌊x1⌋(2)x2⌋ and Φ := {φ(x1, x2), ψ(x1, x2), ϕ(x1, x2)}. which is in normal form. So the element ⌊⌊x1⌋⌊x2⌋⌋ is not confluent. For the desired confluence, we need more rewriting rules. Let (16) With respect to 6db, we have (17) Let u1, u2 ∈ M(X). Then by Eq. (3), ϕ(x1, x2) = ⌊⌊x1⌋x2⌋(2) and R(ϕ(x1, x2)) = ⌊⌊x1⌋(2)x2⌋. φ(u1, u2) = ⌊u1⌋⌊u2⌋ − ⌊⌊u1⌋u2⌋ ∈ S φ(X), and by Lemma 2.15, ⌊⌊u1⌋⌊u2⌋⌋ − ⌊⌊u1⌋u2⌋(2) = ⌊⋆⌋ φ(u1,u2) ∈ Id(S φ(X)) ⊆ Id(S φ(X) ∪ S ψ(X)). With the same argument, ⌊⌊u1⌋⌊u2⌋⌋ − ⌊⌊u1⌋(2)u2⌋ = ψ(⌊u1⌋, u2) ∈ S ψ(X) ⊆ Id(S ψ(X)) ⊆ Id(S φ(X) ∪ S ψ(X)). This implies that ϕ(u1, u2) = ⌊⌊u1⌋u2⌋(2) − ⌊⌊u1⌋(2)u2⌋ =⌊⌊u1⌋⌊u2⌋⌋ − ⌊⌊u1⌋(2)u2⌋ − (⌊⌊u1⌋⌊u2⌋⌋ − ⌊⌊u1⌋u2⌋(2)) ∈ Id(S φ(X) ∪ S ψ(X)) and so Id(S ϕ(X)) ⊆ Id(S φ(X) ∪ S ψ(X)). Hence by Eqs. (4) and (16), (18) Id(S Φ(X)) = Id(S φ(X) ∪ S ψ(X)). Remark 3.4. If u2 = 1, then ϕ(u1, u2) degenerates to ϕ(u1, u2) = ⌊⌊u1⌋u2⌋(2) − ⌊⌊u1⌋(2)u2⌋ = ⌊u1⌋(3) − ⌊u1⌋(3) = 0. So we always assume u2 , 1 in ϕ(u1, u2). This is our running hypothesis in the remainder of the paper. Remark 3.5. From Eqs. (15) and (17), we have (a) for any α(x1, x2) ∈ Φ and u1, u2 ∈ M(X), R(α(u1, u2)) ∈ M(X) is a monomial. (b) for any u1, u2 ∈ M(X), the breadth φ(u1, u2) = 2 and ψ(u1, u2) = ϕ(u1, u2) = 1. Recall Φ is fixed in Eq. (16). In Eq. (13), taking S = S Φ(X) defined in Eq. (4), we get a ΠΦ := ΠS Φ(X) = { qα(u1,u2) → qR(α(u1,u2)) α(x1, x2) ∈ Φ, q ∈ M⋆(X), u1, u2 ∈ M(X)}. term-rewriting system associated to Φ (with respect to 6db) (19) For notation clarity, we abbreviate →α(u1,u2) as →α. Now we are in the position to consider the confluence of the term-rewriting system ΠΦ. By Theorem 2.36, we only need to consider the confluence of basis elements. Take a local fork of a basis element w ∈ M(X): (q1R(α(u1,u2)) α ←q1α(u1,u2) = w = q2β(v1,v2) →β q2R(β(v1,v2))), AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 15 where α(x1, x2), β(x1, x2) ∈ Φ, ui, vi ∈ M(X), i = 1, 2. According to Lemma 3.3, the two placements (α(u1, u2), q1) and (β(v1, v2), q2) are separated, or intersecting, or nested. We consider firstly the former two cases. Lemma 3.6. Let α(x1, x2), β(x1, x2) ∈ Φ and q1α(u1,u2) = q2β(v1,v2) for some q1, q2 ∈ M⋆(X) and ui, vi ∈ M(X), i = 1, 2. If the placements (α(u1, u2), q1) and (β(v1, v2), q2) are separated, then q1R(α(u1,u2)) ↓Φ q2R(β(v1,v2)). Proof. In view of Definition 3.2 (a), there exists p ∈ M⋆1,⋆2(X) such that q1⋆1 = p⋆1, β(v1,v2) and q2⋆2 = pα(u1,u2), ⋆2 . q1R(α(u1,u2)) = pR(α(u1,u2)), β(v1,v2) →β pR(α(u1,u2)), R(β(v1,v2)), On the one hand, (20) where the last step employs the facts that R(α(u1, u2)) is a monomial by Remark 3.5 (a) and so is pR(α(u1,u2)), β(v1,v2). On the other hand, q2R(β(v1,v2)) = pα(u1,u2), R(β(v1,v2)) →α pR(α(u1,u2)), R(β(v1,v2)). (21) Comparing Eqs (20) and (21), we conclude that q1R(α(u1,u2)) ↓Φ q2R(β(v1,v2)). Lemma 3.7. Let α(x1, x2), β(x1, x2) ∈ Φ and q1α(u1,u2) = q2β(v1,v2) for some q1, q2 ∈ M⋆(X) and ui, vi ∈ M(X), i = 1, 2. If the placements (α(u1, u2), q1) and (β(v1, v2), q2) are intersecting, then q1R(α(u1,u2)) ↓Φ q2R(β(v1,v2)). (cid:3) Proof. If the two placements (α(u1, u2), q1) and (β(v1, v2), q2) are intersecting, by symmetry, we may assume that Item (c) (i) in Definition 3.2 holds. Then q1 , q2, because if q1 = q2, then ⋆c = a⋆, a contradiction. So (22) and qα(u1,u2) c = q1α(u1,u2) = q2β(v1,v2) = qa β(v1,v2) = qabc α(u1, u2) c = a β(v1, v2) = abc. This implies that α(u1, u2) = ab and β(v1, v2) = bc. (23) If the breadth α(u1, u2) = 1, then a = 1 or b = 1, both contradicting that a, b , 1 in Defini- tion 3.2 (c). Similarly, if the breadth β(v1, v2) = 1, then b = 1 or c = 1, again a contradiction. So α(u1, u2) , 1 and β(v1, v2) , 1. Hence by Remark 3.5 (b), α(x1, x2) = β(x1, x2) = φ(x1, x2) = ⌊x1⌋⌊x2⌋ − ⌊⌊x1⌋x2⌋. From Eq. (23), we have α(u1, u2) = ⌊u1⌋⌊u2⌋ = ab and β(v1, v2) = ⌊v1⌋⌊v2⌋ = bc and so ⌊u1⌋ = a, ⌊u2⌋ = b = ⌊v1⌋, u2 = v1 and ⌊v2⌋ = c. From Eqs. (15) and (17), R(α(u1, u2))c = R(φ(u1, u2))c = ⌊⌊u1⌋u2⌋⌊v2⌋ →φ ⌊⌊⌊u1⌋u2⌋v2⌋ and aR(β(v1, v2)) = aR(φ(v1, v2)) = aR(φ(u2, v2)) = ⌊u1⌋⌊⌊u2⌋v2⌋ →φ ⌊⌊u1⌋⌊u2⌋v2⌋ →φ ⌊⌊⌊u1⌋u2⌋v2⌋. 16 XING GAO AND TIANJIE ZHANG So R(α(u1, u2))c ↓Φ aR(β(v1, v2)). This follows from Eq. (22) and Lemma 3.1 (b) that q1R(α(u1,u2)) = qR(α(u1,u2)) c ↓Φ qa R(β(v1,v2)) = q2R(β(v1,v2)), as required. (cid:3) Next, let us turn to consider the nested case. We need the following lemmas. The first is on the leading monomials of OPIs in Φ. Lemma 3.8. Let α(x1, x2), β(x1, x2) ∈ Φ and α(u1, u2) = β(v1, v2) for some ui, vi ∈ M(X), i = 1, 2. Then exactly one of the following is true: (a) α(x1, x2) = β(x1, x2), u1 = v1, u2 = v2; (b) α(x1, x2) = ψ(x1, x2), β(x1, x2) = ϕ(x1, x2), u1 = 1, u2 = ⌊v1⌋v2; (c) α(x1, x2) = ϕ(x1, x2), β(x1, x2) = ψ(x1, x2), v1 = 1, v2 = ⌊u1⌋u2. Proof. According to whether α and β are equal, we have the following cases to consider. Case 1. α(x1, x2) = β(x1, x2). Then Items (b) and (c) fail. We show Item (a) is valid. Consider firstly that α(x1, x2) = φ(x1, x2). Then ⌊u1⌋⌊u2⌋ = α(u1, u2) = β(v1, v2) = ⌊v1⌋⌊v2⌋. By the unique decomposition of bracketed words in Eq. (1), we have ⌊u1⌋ = ⌊v1⌋ and ⌊u2⌋ = ⌊v2⌋. This implies u1 = v1 and u2 = v2. Consider secondly that α(x1, x2) = ψ(x1, x2). Then ⌊u1⌊u2⌋⌋ = α(u1, u2) = β(v1, v2) = ⌊v1⌊v2⌋⌋ and so u1⌊u2⌋ = v1⌊v2⌋. This also implies u1 = v1, ⌊u2⌋ = ⌊v2⌋ and u2 = v2. At last, consider α(x1, x2) = ϕ(x1, x2). Then ⌊⌊u1⌋u2⌋(2) = α(u1, u2) = β(v1, v2) = ⌊⌊v1⌋v2⌋(2) and so ⌊⌊u1⌋u2⌋ = ⌊⌊v1⌋v2⌋. Thus ⌊u1⌋u2 = ⌊v1⌋v2 and so u1 = v1 and u2 = v2. Case 2. α(x1, x2) , β(x1, x2). Then Item (a) fails. Suppose firstly that one of α(x1, x2) and β(x1, x2) is φ(x1, x2). By symmetry, we may let α(x1, x2) = φ(x1, x2). Then β(x1, x2) , φ(x1, x2). From Remark 3.5 (b), α(u1, u2) = φ(u1, u2) = 2 and β(v1, v2) = 1. This implies that α(u1, u2) , β(v1, v2), contradicting our hypothesis. Suppose α(x1, x2), β(x1, x2) , φ(x1, x2). Then we have the following two subcases. Case 2.1. α(x1, x2) = ψ(x1, x2) and β(x1, x2) = ϕ(x1, x2). Then Item (c) fails and ⌊u1⌊u2⌋⌋ = ψ(u1, u2) = α(u1, u2) = β(v1, v2) = ϕ(v1, v2) = ⌊⌊v1⌋v2⌋(2). So u1⌊u2⌋ = ⌊⌊v1⌋v2⌋. This implies that u1 = 1, ⌊u2⌋ = ⌊⌊v1⌋v2⌋ and u2 = ⌊v1⌋v2 and so Item (b) is valid. Case 2.2. α(x1, x2) = ϕ(x1, x2) and β(x1, x2) = ψ(x1, x2). Then Item (b) fails and ⌊⌊u1⌋u2⌋(2) = ϕ(u1, u2) = α(u1, u2) = β(v1, v2) = ψ(v1, v2) = ⌊v1⌊v2⌋⌋. This follows that ⌊⌊u1⌋u2⌋ = v1⌊v2⌋. So v1 = 1, v2 = ⌊u1⌋u2 and Item (c) is valid. Lemma 3.9. Let α(x1, x2), β(x1, x2) ∈ Φ and q1α(u1,u2) = q2β(v1,v2) for some q1, q2 ∈ M⋆(X) and ui, vi ∈ M(X), i = 1, 2. If q2 = q1q for some q ∈ M⋆(X) and β(v1, v2) is a subword of u1 or u2, then q1R(α(u1,u2)) ↓Φ q2R(β(v1,v2)). (cid:3) Proof. For clarity, write α := α(u1, u2) and β := β(v1, v2). AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 17 By symmetry we may assume that β is a subword of u1 and so u1 = q′β for some q′ ∈ M⋆(X). As α(x1, x2) is linear on each variable and R(α(u1, u2)) is a monomial by Remark 3.5 (a), we may write (24) Since q2 = q1q by our hypothesis, we have α = α(u1, u2) = pu1,u2 and R(α) = R(α(u1, u2)) = p′u1,u2 for some p, p′ ∈ M⋆(X). q1α = q2β = q1q β , and so qβ = α = pu1,u2 = pq′ β,u2 = (pq′,u2)β. Hence (25) where the second equation employs Eq. (24). So on the one hand, we have q = pq′,u2 = α(q′, u2), q1R(α) = q1p′u1,u2 = q1p′q′ (26) where the first step is followed from Eq. (24). On the other hand, we have (27) where the first step is followed from the hypothesis q2 = q1q, the second from Eq. (25) and the last from Eq. (24). Comparing Eqs (26) and (27), we obtain q1R(α) ↓Φ q2R(β). This completes the proof. q2R(β) = q1qR(β) = q1α(q′R(β), u2) →α q1R(α(q′R(β), u2)) = q1p′q′R(β) , u2 , , u2 β →β q1p′q′ R(β) , u2 , As an application of Theorem 2.36, we have Theorem 3.10. The term-rewriting system ΠΦ defined in Eq. (19) is convergent. Proof. Since 6db we used is a monomial order on M(X), ΠΦ is terminating [18]. By Defini- tion 2.26 (e), we are left to show that ΠΦ is confluent. From Theorem 2.36, it is sufficient to prove that ΠΦ is locally confluent for any basis element. Let (cid:3) (q1R(α(u1,u2)) Φ ←q1α(u1,u2) = w = q2β(v1,v2) →Φ q2R(β(v1,v2))) be an arbitrary local fork of a basis element w, where α(x1, x2), β(x1, x2) ∈ Φ, ui, vi ∈ M(X), i = 1, 2. We only need to show that (28) q1R(α(u1,u2)) ↓Φ q2R(β(v1,v2)). According to Lemma 3.3, the two placements (α(u1, u2), q1) and (β(v1, v2), q2) are separated, or nested, or intersecting. If they are separated or intersecting, then by Lemmas 3.6 and 3.7, Eq. (28) holds. If the two placements (α(u1, u2), q1) and (β(v1, v2), q2) are nested, by symmetry in Definition 3.2 (b), we may assume that q2 = q1q. If β(v1, v2) is a subword of u1 or u2, then by Lemma 3.9, Eq. (28) holds. Suppose β(v1, v2) is not a subword of u1 and u2. Note (29) q1α(u1,u2) = q2β(v1,v2) = q1q β(v1,v2) and so α(u1, u2) = qβ(v1,v2). Since q2 = q1q, Eq. (28) is equivalent to q1R(α(u1,u2)) ↓Φ q1qR(β(v1,v2)). 18 XING GAO AND TIANJIE ZHANG So to prove Eq. (28), by Lemma 3.1 (b), it is enough to show that (30) R(α(u1, u2)) ↓Φ qR(β(v1,v2)). If q = ⋆, then α(u1, u2) = β(v1, v2). By Lemma 3.8, exactly one of the three items there holds. If Item (a) holds, then R(α(u1, u2)) = R(β(v1, v2)) and Eq. (30) is valid by q = ⋆. Since Item (b) and Item (c) are symmetric, we consider that Item (b) holds. Then α(x1, x2) = ψ(x1, x2), β(x1, x2) = ϕ(x1, x2), u1 = 1, u2 = ⌊v1⌋v2. This follows from Eqs. (15) and (17) that R(α(u1, u2)) = ⌊⌊u1⌋u2⌋ = ⌊⌊1⌋⌊v1⌋v2⌋ →φ ⌊⌊⌊1⌋v1⌋v2⌋ and qR(β(v1,v2)) = ⋆ ⌊⌊v1⌋(2)v2⌋ = ⌊⌊v1⌋(2)v2⌋ = ⌊⌊1⌊v1⌋⌋v2⌋ →ψ ⌊⌊⌊1⌋v1⌋v2⌋. Hence Eq. (30) is valid. Summing up, we are left to consider the case of that q2 = q1q, α(u1, u2) = qβ(v1,v2), q , ⋆ and β(v1, v2) is not a subword of u1 and u2. (31) Then (32) We have the following cases to consider. Case 1. α(x1, x2) = φ(x1, x2). Then α(u1, u2) = ⌊u1⌋⌊u2⌋ by Eq. (15). q1 , q2 and α(u1, u2) , β(v1, v2). If β(x1, x2) = φ(x1, x2), then ⌊u1⌋⌊u2⌋ = α(u1, u2) = qβ(v1,v2) = q⌊v1⌋⌊v2⌋, ⌊u1⌋⌊u2⌋ = α(u1, u2) = qβ(v1,v2) = q⌊v1⌊v2⌋⌋, that is, ⌊v1⌋⌊v2⌋ is a subword of ⌊u1⌋⌊u2⌋. By Eq. (32), ⌊v1⌋⌊v2⌋ , ⌊u1⌋⌊u2⌋. So ⌊v1⌋⌊v2⌋ is a subword of ⌊u1⌋ or ⌊u2⌋. Since ⌊v1⌋⌊v2⌋ , ⌊u1⌋, ⌊u2⌋ by comparing the breadth, ⌊v1⌋⌊v2⌋ is a subword of u1 or u2 by Lemma 2.14 (a), contradicting Eq. (31). So β(x1, x2) , φ(x1, x2). Subcase 1.1. β(x1, x2) = ψ(x1, x2). In this subcase, we have (33) that is, ⌊v1⌊v2⌋⌋ is a subword of ⌊u1⌋⌊u2⌋. By Lemma 2.14 (b), either ⌊v1⌊v2⌋⌋ is a subword of ⌊u1⌋ or ⌊v1⌊v2⌋⌋ is a subword of ⌊u2⌋. Note that β(v1, v2) = ⌊v1⌊v2⌋⌋ is not a subword of u1 and u2 by Eq. (31). From Lemma 2.14 (a) and Eq. (33), either (34) or (35) For the former case of Eq. (34), we have ⌊v1⌊v2⌋⌋ = ⌊u2⌋ and q = ⌊u1⌋ ⋆ . ⌊v1⌊v2⌋⌋ = ⌊u1⌋ and q = ⋆⌊u2⌋, R(φ(u1, u2)) = ⌊⌊u1⌋u2⌋ = ⌊⌊v1⌊v2⌋⌋u2⌋ →ψ ⌊⌊⌊v1⌋v2⌋u2⌋ and qR(ψ(v1,v2)) = (⋆⌊u2⌋) ⌊⌊v1⌋v2⌋ = ⌊⌊v1⌋v2⌋⌊u2⌋ →φ ⌊⌊⌊v1⌋v2⌋u2⌋. Hence R(φ(u1, u2)) ↓Φ qR(ψ(v1,v2)) and Eq. (30) holds, as needed. For the later case of Eq. (35), we have u2 = v1⌊v2⌋. So R(φ(u1, u2)) = ⌊⌊u1⌋u2⌋ = ⌊⌊u1⌋v1⌊v2⌋⌋ →ψ ⌊⌊⌊u1⌋v1⌋v2⌋ AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 19 and qR(ψ(v1,v2)) = (⌊u1⌋⋆) ⌊⌊v1⌋v2⌋ = ⌊u1⌋⌊⌊v1⌋v2⌋ →φ ⌊⌊u1⌋⌊v1⌋v2⌋ →φ ⌊⌊⌊u1⌋v1⌋v2⌋. ⌊u1⌋⌊u2⌋ = α(u1, u2) = qβ(v1,v2) = q⌊⌊v1⌋v2⌋(2), Hence R(φ(u1, u2)) ↓Φ qR(ψ(v1,v2)) and Eq. (30) holds, as needed. Subcase 1.2. β(x1, x2) = ϕ(x1, x2). In this subcase, we have (36) that is, ⌊⌊v1⌋v2⌋(2) is a subword of ⌊u1⌋⌊u2⌋. By Lemma 2.14 (b), either ⌊⌊v1⌋v2⌋(2) is a subword of ⌊u1⌋ or ⌊⌊v1⌋v2⌋(2) is a subword of or ⌊u2⌋. Since ⌊⌊v1⌋v2⌋(2) is not a subword or u1 and u2 by Eq. (31), from Lemma 2.14 (a) and Eq (36), either (37) or (38) Consider firstly the former case of Eq. (37). We have ⌊⌊v1⌋v2⌋(2) = ⌊u1⌋ and q = ⋆⌊u2⌋ ⌊⌊v1⌋v2⌋(2) = ⌊u2⌋ and q = ⌊u1⌋ ⋆ . R(φ(u1, u2)) = ⌊⌊u1⌋u2⌋ = ⌊⌊⌊v1⌋v2⌋(2)u2⌋ →ϕ ⌊⌊⌊v1⌋(2)v2⌋u2⌋ and qR(ϕ(v1,v2)) =(⋆⌊u2⌋) ⌊⌊v1⌋(2)v2⌋ = ⌊⌊v1⌋(2)v2⌋⌊u2⌋ →φ ⌊⌊⌊v1⌋(2)v2⌋u2⌋. Hence R(φ(u1, u2)) ↓Φ qR(ϕ(v1,v2)) and Eq. (30) holds. For the later case of Eq. (38), we have u2 = ⌊⌊v1⌋v2⌋. Then R(φ(u1, u2)) = ⌊⌊u1⌋u2⌋ = ⌊⌊u1⌋⌊⌊v1⌋v2⌋⌋ →φ ⌊⌊⌊u1⌋⌊v1⌋v2⌋⌋ = ⌊⌊u1⌋⌊v1⌋v2⌋(2) →φ ⌊⌊⌊u1⌋v1⌋v2⌋(2) →ϕ ⌊⌊⌊u1⌋v1⌋(2)v2⌋ →ϕ ⌊⌊⌊u1⌋(2)v1⌋v2⌋ and qR(ϕ(v1,v2)) = (⌊u1⌋⋆) ⌊⌊v1⌋(2)v2⌋ = ⌊u1⌋⌊⌊v1⌋(2)v2⌋ →φ ⌊⌊u1⌋⌊v1⌋(2)v2⌋ →φ ⌊⌊⌊u1⌋⌊v1⌋⌋v2⌋ →φ ⌊⌊⌊⌊u1⌋v1⌋⌋v2⌋ = ⌊⌊⌊u1⌋v1⌋(2)v2⌋ →ϕ ⌊⌊⌊u1⌋(2)v1⌋v2⌋. Hence R(φ(u1, u2)) ↓Φ qR(ϕ(v1,v2)) and Eq. (30) holds, as needed. Case 2. α(x1, x2) = ψ(x1, x2). Then α(u1, u2) = ⌊u1⌊u2⌋⌋ by Eq. (15). Case 2.1. β(x1, x2) = φ(x1, x2). In this subcase, we have (39) that is, ⌊v1⌋⌊v2⌋ is a subword of ⌊u1⌊u2⌋⌋. Since ⌊v1⌋⌊v2⌋ , ⌊u1⌊u2⌋⌋ by Eq. (32), it follows from Lemma 2.14 (a) that ⌊v1⌋⌊v2⌋ is a subword of u1⌊u2⌋. Note ⌊v1⌋⌊v2⌋ is not a subword of u1 or u2 by Eq. (31). So a⌊v1⌋⌊v2⌋ = u1⌊u2⌋ for some a ∈ M(X) and q = ⌊a⋆⌋ by Eq. (39). Then a⌊v1⌋ = u1, ⌊v2⌋ = ⌊u2⌋, v2 = u2. This follows that ⌊u1⌊u2⌋⌋ = α(u1, u2) = qβ(v1,v2) = q⌊v1⌋⌊v2⌋, R(ψ(u1, u2)) = ⌊⌊u1⌋u2⌋ = ⌊⌊a⌊v1⌋⌋u2⌋ →ψ ⌊⌊⌊a⌋v1⌋u2⌋ and qR(φ(v1,v2)) = (⌊a⋆⌋) ⌊⌊v1⌋v2⌋ = ⌊a⌊⌊v1⌋v2⌋⌋ = ⌊a⌊⌊v1⌋u2⌋⌋ →ψ ⌊⌊a⌋⌊v1⌋u2⌋ →φ ⌊⌊⌊a⌋v1⌋u2⌋. Hence R(ψ(u1, u2)) ↓Φ qR(φ(v1,v2)) and Eq. (30) holds, as needed. Case 2.2 β(x1, x2) = ψ(x1, x2). In this subcase, we have (40) ⌊u1⌊u2⌋⌋ = α(u1, u2) = qβ(v1,v2) = q⌊v1⌊v2⌋⌋, 20 XING GAO AND TIANJIE ZHANG that is, ⌊v1⌊v2⌋⌋ is a subword of ⌊u1⌊u2⌋⌋. By Lemma 2.14 (a) and ⌊v1⌊v2⌋⌋ , ⌊u1⌊u2⌋⌋ from Eq. (32), ⌊v1⌊v2⌋⌋ is a subword of u1⌊u2⌋. Note ⌊v1⌊v2⌋⌋ is not a subword of u1 and u2 by Eq. (31). So by Lemma 2.14 (b), ⌊v1⌊v2⌋⌋ is a subword of ⌊u2⌋. From Lemma 2.14 (a), we have ⌊v1⌊v2⌋⌋ = ⌊u2⌋, v1⌊v2⌋ = u2 and q = ⌊u1⋆⌋ by Eq. (40). Thus R(ψ(u1, u2)) = ⌊⌊u1⌋u2⌋ = ⌊⌊u1⌋v1⌊v2⌋⌋ →ψ ⌊⌊⌊u1⌋v1⌋v2⌋ and qR(ψ(v1,v2)) = (⌊u1⋆⌋) ⌊⌊v1⌋v2⌋ = ⌊u1⌊⌊v1⌋v2⌋⌋ →ψ ⌊⌊u1⌋⌊v1⌋v2⌋ →φ ⌊⌊⌊u1⌋v1⌋v2⌋. Hence R(ψ(u1, u2)) ↓Φ qR(ψ(v1,v2)) and Eq. (30) holds, as needed. Case 2.3. β(x1, x2) = ϕ(x1, x2). In this subcase, we have (41) that is, ⌊⌊v1⌋v2⌋(2) is a subword of ⌊u1⌊u2⌋⌋. By Lemma 2.14 (a) and ⌊⌊v1⌋v2⌋(2) , ⌊u1⌊u2⌋⌋ from Eq. (32), ⌊⌊v1⌋v2⌋(2) is a subword of u1⌊u2⌋. Note from Eq. (31), ⌊⌊v1⌋v2⌋(2) is not a subword of u1 and u2. So by Lemma 2.14 (b), ⌊⌊v1⌋v2⌋(2) is a subword of ⌊u2⌋. By Lemma 2.14 (a), ⌊⌊v1⌋v2⌋(2) = ⌊u2⌋ and then q = ⌊u1⋆⌋ by Eq. (41). This implies ⌊⌊v1⌋v2⌋ = u2. Thus ⌊u1⌊u2⌋⌋ = α(u1, u2) = qβ(v1,v2) = q⌊⌊v1⌋v2⌋(2), R(ψ(u1, u2)) = ⌊⌊u1⌋u2⌋ = ⌊⌊u1⌋⌊⌊v1⌋v2⌋⌋ →φ ⌊⌊⌊u1⌋⌊v1⌋v2⌋⌋ = ⌊⌊u1⌋⌊v1⌋v2⌋(2) →φ ⌊⌊⌊u1⌋v1⌋v2⌋(2) →ϕ ⌊⌊⌊u1⌋v1⌋(2)v2⌋ →ϕ ⌊⌊⌊u1⌋(2)v1⌋v2⌋ and qR(ϕ(v1,v2)) = (⌊u1⋆⌋) ⌊⌊v1⌋(2)v2⌋ = ⌊u1⌊⌊v1⌋(2)v2⌋⌋ →ψ ⌊⌊u1⌋⌊v1⌋(2)v2⌋ →φ ⌊⌊⌊u1⌋⌊v1⌋⌋v2⌋ →φ ⌊⌊⌊⌊u1⌋v1⌋⌋v2⌋ = ⌊⌊⌊u1⌋v1⌋(2)v2⌋ →ϕ ⌊⌊⌊u1⌋(2)v1⌋v2⌋. Hence R(ψ(u1, u2)) ↓Φ qR(ϕ(v1,v2)) and Eq. (30) holds, as needed. Case 3. α(x1, x2) = ϕ(x1, x2). Then α(u1, u2) = ⌊⌊u1⌋u2⌋(2) by Eq. (17). Subcase 3.1. β(x1, x2) = φ(x1, x2). In this subcase, (42) that is, ⌊v1⌋⌊v2⌋ is a subword of ⌊⌊u1⌋u2⌋(2). As ⌊v1⌋⌊v2⌋ , ⌊⌊u1⌋u2⌋(2) by Eq. (31), ⌊v1⌋⌊v2⌋ is a subword of ⌊⌊u1⌋u2⌋ by Lemma 2.14 (a). Again using Lemma 2.14 (a), ⌊v1⌋⌊v2⌋ is a subword of ⌊u1⌋u2 by ⌊v1⌋⌊v2⌋ , ⌊⌊u1⌋u2⌋. From Eq. (31), β(v1, v2) = ⌊v1⌋⌊v2⌋ is not a subword of u1 and u2. Hence ⌊v1⌋⌊v2⌋a = ⌊u1⌋u2 for some a ∈ M(X) and so q = ⌊⋆a⌋(2) by Eq. (42). This implies that ⌊v1⌋ = ⌊u1⌋, v1 = u1 and ⌊v2⌋a = u2. Hence ⌊⌊u1⌋u2⌋(2) = α(u1, u2) = qβ(v1,v2) = q⌊v1⌋⌊v2⌋, R(ϕ(u1, u2)) = ⌊⌊u1⌋(2)u2⌋ = ⌊⌊u1⌋(2)⌊v2⌋a⌋ →φ ⌊⌊⌊u1⌋(2)v2⌋a⌋ and qR(φ(v1,v2)) = (⌊⋆a⌋(2)) ⌊⌊v1⌋v2⌋ = ⌊⌊⌊v1⌋v2⌋a⌋(2) = ⌊⌊⌊u1⌋v2⌋a⌋(2) →ϕ ⌊⌊⌊u1⌋v2⌋(2)a⌋ →ϕ ⌊⌊⌊u1⌋(2)v2⌋a⌋. Hence R(ϕ(u1, u2)) ↓Φ qR(φ(v1,v2)) and Eq. (30) holds, as needed. Subcase 3.2. β(x1, x2) = ψ(x1, x2). In this subcase, (43) that is, ⌊v1⌊v2⌋⌋ is a subword of ⌊⌊u1⌋u2⌋(2). Since ⌊v1⌊v2⌋⌋ , ⌊⌊u1⌋u2⌋(2) by Eq. (32), ⌊v1⌊v2⌋⌋ is a subword of ⌊⌊u1⌋u2⌋ by Lemma 2.14 (a). Again using Lemma 2.14 (a), either ⌊v1⌊v2⌋⌋ = ⌊⌊u1⌋u2⌋ or ⌊v1⌊v2⌋⌋ is a subword of ⌊u1⌋u2. ⌊⌊u1⌋u2⌋(2) = α(u1, u2) = qβ(v1,v2) = q⌊v1⌊v2⌋⌋, AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 21 For the former case of ⌊v1⌊v2⌋⌋ = ⌊⌊u1⌋u2⌋, we have q = ⌊⋆⌋ by Eq. (43) and v1⌊v2⌋ = ⌊u1⌋u2. This implies that v1 = ⌊u1⌋v′ 1 and u2 = u′ 2⌊v2⌋ for some v′ ⌊u1⌋v′ 1⌊v2⌋ = v1⌊v2⌋ = ⌊u1⌋u2 = ⌊u1⌋u′ 1, u′ 2 ∈ M(X). Then 1 = u′ 2⌊v2⌋ and so v′ 2 =: a. Then v1 = ⌊u1⌋a and u2 = a⌊v2⌋. This follows that R(ϕ(u1, u2)) = ⌊⌊u1⌋(2)u2⌋ = ⌊⌊u1⌋(2)a⌊v2⌋⌋ →ψ ⌊⌊⌊u1⌋(2)a⌋v2⌋ and qR(ψ(v1,v2)) = ⌊⋆⌋ ⌊⌊v1⌋v2⌋ = ⌊⌊⌊v1⌋v2⌋⌋ = ⌊⌊v1⌋v2⌋(2) = ⌊⌊⌊u1⌋a⌋v2⌋(2) →ϕ ⌊⌊⌊u1⌋a⌋(2)v2⌋ →ϕ ⌊⌊⌊u1⌋(2)a⌋v2⌋. Hence R(ϕ(u1, u2)) ↓Φ qR(ψ(v1,v2)) and Eq. (30) holds, as needed. Consider the latter case that ⌊v1⌊v2⌋⌋ is a subword of ⌊u1⌋u2. By Eq. (31), β(v1, v2) = ⌊⌊v1⌋v2⌋ is not a subword of u1 and u2. So from Lemma 2.14 (b), ⌊v1⌊v2⌋⌋ is a subword of ⌊u1⌋. Using Lemma 2.14 (a), we have ⌊v1⌊v2⌋⌋ = ⌊u1⌋ and so q = ⌊⋆u2⌋(2) by Eq. (43). Then v1⌊v2⌋ = u1. So we have R(ϕ(u1, u2)) = ⌊⌊u1⌋(2)u2⌋ = ⌊⌊v1⌊v2⌋⌋(2)u2⌋ →ψ ⌊⌊⌊v1⌋v2⌋(2)u2⌋ →ϕ ⌊⌊⌊v1⌋(2)v2⌋u2⌋ and qR(ψ(v1,v2)) = (⌊⋆u2⌋(2)) ⌊⌊v1⌋v2⌋ = ⌊⌊⌊v1⌋v2⌋u2⌋(2) →ϕ ⌊⌊⌊v1⌋v2⌋(2)u2⌋ →ϕ ⌊⌊⌊v1⌋(2)v2⌋u2⌋. Hence R(ϕ(u1, u2)) ↓Φ qR(ψ(v1,v2)) and Eq. (30) holds, as needed. Subcase 3.3. β(x1, x2) = ϕ(x1, x2). In this subsection, we have (44) ⌊⌊u1⌋u2⌋(2) = α(u1, u2) = qβ(v1,v2) = q⌊⌊v1⌋v2⌋(2), that is, ⌊⌊v1⌋v2⌋(2) is a subword of ⌊⌊u1⌋u2⌋(2). By Eq. (32), ⌊⌊v1⌋v2⌋(2) , ⌊⌊u1⌋u2⌋(2). So from Lemma 2.14 (a), ⌊⌊v1⌋v2⌋(2) is a subword of ⌊⌊u1⌋u2⌋. Again using Lemma 2.14 (a), either ⌊⌊v1⌋v2⌋(2) = ⌊⌊u1⌋u2⌋ or ⌊⌊v1⌋v2⌋(2) is a subword of ⌊u1⌋u2. For the former case, we have q = ⌊⋆⌋ by Eq. (44) and ⌊⌊v1⌋v2⌋ = ⌊u1⌋u2. This implies that u2 = 1, ⌊⌊v1⌋v2⌋ = ⌊u1⌋ and ⌊v1⌋v2 = u1. Then R(ϕ(u1, u2)) = ⌊⌊u1⌋(2)u2⌋ = ⌊u1⌋(3) = ⌊⌊v1⌋v2⌋(3) →ϕ ⌊⌊v1⌋(2)v2⌋(2) →ϕ ⌊⌊v1⌋(3)v2⌋ and qR(ϕ(v1,v2)) = ⌊⋆⌋ ⌊⌊v1⌋(2)v2⌋ = ⌊⌊⌊v1⌋(2)v2⌋⌋ = ⌊⌊v1⌋(2)v2⌋(2) →ϕ ⌊⌊v1⌋(3)v2⌋. Hence R(ϕ(u1, u2)) ↓Φ qR(ϕ(v1,v2)) and Eq. (30) holds, as needed. Consider the later case of that ⌊⌊v1⌋v2⌋(2) is a subword of ⌊u1⌋u2. By Eq. (31), β(v1, v2) = ⌊⌊v1⌋v2⌋(2) is not a subword of u1 and u2. So from Lemma 2.14 (b), ⌊⌊v1⌋v2⌋(2) is a subword of ⌊u1⌋. Using Lemma 2.14 (a), we have ⌊⌊v1⌋v2⌋(2) = ⌊u1⌋ and so q = ⌊⋆u2⌋(2) by Eq. (44). Thus we have R(ϕ(u1, u2)) = ⌊⌊u1⌋(2)u2⌋ = ⌊⌊⌊v1⌋v2⌋(3)u2⌋ →ϕ ⌊⌊⌊v1⌋(2)v2⌋(2)u2⌋ →ϕ ⌊⌊⌊v1⌋(3)v2⌋u2⌋ and qR(ϕ(v1,v2)) = (⌊⋆u2⌋(2)) ⌊⌊v1⌋(2)v2⌋ = ⌊⌊⌊v1⌋(2)v2⌋u2⌋(2) →ϕ ⌊⌊⌊v1⌋(2)v2⌋(2)u2⌋ →ϕ ⌊⌊⌊v1⌋(3)v2⌋u2⌋ Hence R(ϕ(u1, u2)) ↓Φ qR(ϕ(v1,v2)) and Eq. (30) holds, as needed. This completes the proof. (cid:3) 22 XING GAO AND TIANJIE ZHANG Recall from Remark 3.4 that u2 , 1 in ϕ(u1, u2). So we define (45) M :={qφ(u1,u2), qψ(u1,u2) q ∈ M⋆(X), u1, u2 ∈ M(X)}, N :={qϕ(u1,u2) q ∈ M⋆(X), u1 ∈ M(X), u2 ∈ M(X) \ {1}}, N1 :={qϕ(u1,u2) q ∈ M⋆(X), u1, u2 ∈ M(X)}, N2 :={qϕ(u1,1) q ∈ M⋆(X), u1 ∈ M(X)}. Then N = N1 \ N2. From Eqs. (15) and (17), and so N2 ⊆ M. Thus qϕ(u1,1) = q⌊u1⌋(3) = q⌊1⌊u1⌋(2)⌋ = qψ(1, ⌊u1⌋) ∈ M M ∪ N = M ∪ (N1 \ N2) = M ∪ N1. Hence (46) {qs q ∈ M⋆(X), s ∈ S Φ(X)} ={qs q ∈ M⋆(X), s ∈ S φ(X) ∪ S ψ(X)} ∪ {qs q ∈ M⋆(X), s ∈ S ϕ(X)} =M ∪ N = M ∪ N1 ={q⌊u1⌋⌊u2⌋, q⌊u1⌊u2⌋⌋, q⌊⌊u1⌋u2⌋(2) q ∈ M⋆(X), u1, u2 ∈ M(X)}, where the second step employs Remark 3.4. Now we are ready to give our main result. From Proposition 2.11 and Eq. (18), kM(X)/Id(S Φ(X)) is the free averaging algebra on X. Theorem 3.11. The Irr(S Φ(X)) is a k-basis of the free (unitary) averaging algebra kM(X)/Id(S Φ(X)) on X. More precisely, kM(X) = Id(S Φ(X)) ⊕ kIrr(S Φ(X)), where Irr(S Φ(X)) = M(X) \ {q⌊u1⌋⌊u2⌋, q⌊u1⌊u2⌋⌋, q⌊⌊u1⌋u2⌋(2) q ∈ M⋆(X), u1, u2 ∈ M(X)}. Proof. By Theorem 3.10, ΠΦ = ΠS Φ(X) is convergent. Using Theorems 2.41 to S = S Φ(X), we have kM(X) = Id(S Φ(X)) ⊕ kIrr(S Φ(X)), where Irr(S Φ(X)) =M(X) \ {qs q ∈ M⋆(X), s ∈ S Φ(X)} =M(X) \ {q⌊u1⌋⌊u2⌋, q⌊u1⌊u2⌋⌋, q⌊⌊u1⌋u2⌋(2) q ∈ M⋆(X), u1, u2 ∈ M(X)} by Eq. (46). (cid:3) Acknowledgements: This work was supported by the National Natural Science Foundation of China (Grant No. 11201201, 11371177 and 11371178) and the Natural Science Foundation of Gansu Province (Grant No. 1308RJZA112). AVERAGING ALGEBRAS, REWRITING SYSTEMS AND GR OBNER-SHIRSHOV BASES 23 References [1] M. Aguiar, Pre-Poisson algebras, Lett. Math. Phys. 54 (2000) 263-277. 1 [2] F. Baader and T. Nipkow, Term Rewriting and All That (Cambridge University Press, Cambridge, 1998). 2, 8 [3] C. Bai, A unified algebraic approach to the classical Yang-Baxter equations, J. Phys. A: Math. Theor. 40 (2007) 11073 -- 11082. 1 [4] G. Baxter, An analytic problem whose solution follows from a simple algebraic identity, Pacific J. Math. 10 (1960) 731 -- 742. 1, 4 [5] M. Bezem, J. W. Klop, R. de Vrijer and Terese, Term rewriting systems (Cambridge University Press, 2003). 8 [6] G. Birkhoff, Moyennes de fonctions born´ees, Coil. Internat. Centre Nat. Recherthe Sci. (Paris), Alg´ebre Th´eforie Nombres 24 (1949) 149-153. 1 [7] L. A. Bokut and Y. Chen, Grobner-Shirshov bases and their calculations, Bulletin of Mathematical Sciences 4 (2014) 325-395. 2, 6 [8] L. A. Bokut, Y. Chen and Y. Chen, Composition-Diamond lemma for tensor product of free algebras, J. Algebra 323 (2010) 2520-2537. 2 [9] 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. 2, 3, 5, 6 [10] N. H. Bong, Some Apparent Connection Between Baxter and Averaging Operators, J. Math. Anal. Appl. 56 (1976) 330-345. 1 [11] B. Buchberger, An algorithm for finding a basis for the residue class ring of a zero-dimensional polynomial ideal, PhD thesis, University of Innsbruck, 1965 (in German). 2 [12] W. Cao, An algebraic study of averaging operators, PhD thesis, Rutgers University at Newark, 2000. 1 [13] P.M. Cohn, Further Algebra and Applications (Springer, second edition 2003). 5 [14] J. Kamp´e de F´eriet, Introduction to the statistical theory of turbulence, correlation and spectrum, Lecture Series No.8, prepared by S. I. Pai, The Institute of Fluid Dynamics and Applied Mathematics, University of Maryland (1950-51). 5 [15] W. Fechner, Inequalities conected with averaging operators, Indag. Math. 24 (2013) 305-312. 1 [16] J. L. B. Gamlen and J. B. Miller, Averaging and Reynolds Operators on Banach Algebras II. Spectral Properties of Averaging Operators, J. Math. Anal. Appl. 23 (1968) 183-197. 1 [17] X. Gao and L. Guo, Operators, rewriting systems and Grobner-Shirshov bases, Preprint. 12 [18] X. Gao, L. Guo, W. Sit and S. Zheng, Rota-Baxter type operators, rewriting systems and Grobner- Shirshov bases, http://arxiv.org/pdf/1412.8055v1.pdf, received by J. Symb. Comput. 2, 7, 8, 11, 12, 13, 17 [19] X. Gao, L. Guo and S. Zheng, Construction of free commutative integro-differential algebras by the method of Grobner-Shirshov bases, J. Algebra and Its Applications 13 (2014) 1350160. 5 [20] L. Guo, Operated semigroups, Motzkin paths and rooted trees, J Algebra Comb. 29 (2009) 35-62. 3, 4 [21] L. Guo, An Introduction to Rota-Baxter Algebra (International Press (US) and Higher Education Press (China), 2012). 1, 4 [22] L. Guo, J. Pei, Averaging algebras, Schroder numbers, rooted trees and operads, J Algebra Comb. 42 (2015) 73-109. 1 [23] L. Guo, W. Sit and R. Zhang, Differemtail Type Operators and Grobner-Shirshov Bases, J. Symb. Comput. 52 (2013) 97-123. 2, 5, 6 [24] H. Hironaka, Resolution of singulatities of an algebraic variety over a field if characteristic zero, I, II, Ann. Math. 79 (1964) 109-203, 205-326. 2 [25] J. L. Kelley, Averging operators on C∞(X), Illinois J. Math. 2 (1958) 214-223. 1 [26] J. Kamp´e de F´eriet, L´etat actuel du probl´eme de la turbulaence (I and II), La Sci. A´erienne 3 (1934) 9-34, 4 (1935) 12-52. 1 [27] E. Kolchin, Differential algebraic groups (Academic Press, Inc., Orlando, FL, 1985). 4 [28] A. G. Kurosh, Free sums of multiple operator algebras, Siberian. Math. J. 1 (1960) 62-70 (in Russian). 3 [29] J. L. Loday, Dialgebras, in Dialgebras and related operads, Lecture Notes in Math. 1763 (2002) 7-66. 1 24 XING GAO AND TIANJIE ZHANG [30] J. B. Miller, Averaging and Reynolds operators on Banach algebra I, Representation by derivation and antiderivations, J. Math. Anal. Appl. 14 (1966) 527-548. 1 [31] S. T. C. Moy, Characterizations of conditional expectation as a transformation on function spaces, Pacific J. Math. 4 (1954) 47-63. 1 [32] E. Ohlebusch, Advanced topics in term rewriting (Springer, New York, 2002). 2 [33] M. van der Put and M. Singer, Galois Theory of Linear Differential Equations, ( Springer, 2003). 4 [34] O. Reynolds, On the dynamic theory of incompressible viscous fluids and the determination of the criterion, Phil. Trans. Roy. Soc. A 136 (1895) 123-164. 1, 5 [35] J. F. Ritt, Differential Algebra, Colloquium publications, Vol. 33 (Amer. Math. Soc., New York, 1950). 4 [36] G. C. Rota, Reynolds operators, Proceedings of Symposia in Applied Mathematics, Vol. XVI (1964), Amer. Math. Soc., Providence, R.I., 70-83. 1 [37] G. C. Rota, Baxter algebras and combinatorial identities I, II, Bull. Amer.Math. Soc. 75 (1969) 325-329, 330-334. 4 [38] A. I. Shirshov, Some algorithmic problem for ǫ-algebras, Sibirsk. Mat. Z. 3 (1962) 132-137. 2 [39] A. Triki, Extensions of Positive Projections and Averaging Operators, J. Math. Anal. 153 (1990) 486-496. 1 [40] S. Zheng and L. Guo, Relative locations of subwords in free operated semigroups and Motzkin words, Frontiers of Mathematics in China 10 (2015) 1243-1261. 13 School of Mathematics and Statistics, Key Laboratory of Applied Mathematics and Complex Systems, Lanzhou University, Lanzhou, 730000, P.R. China E-mail address: [email protected] School of Mathematics and Statistics, Lanzhou University, Lanzhou, 730000, P.R. China E-mail address: [email protected]
1608.08282	2	1608	2017-06-07T22:00:45	Infinite-Dimensional Triangularization	[ "math.RA" ]	The goal of this paper is to generalize the theory of triangularizing matrices to linear transformations of an arbitrary vector space, without placing any restrictions on the dimension of the space or on the base field. We define a transformation T of a vector space V to be "triangularizable" if V has a well-ordered basis such that T sends each vector in that basis to the subspace spanned by basis vectors no greater than it. We then show that the following conditions (among others) are equivalent: (1) T is triangularizable, (2) every finite-dimensional subspace of V is annihilated by f(T) for some polynomial f that factors into linear terms, (3) there is a maximal well-ordered set of subspaces of V that are invariant under T, (4) T can be put into a crude version of the Jordan canonical form. We also show that any finite collection of commuting triangularizable transformations is simultaneously triangularizable, we describe the closure of the set of triangularizable transformations in the standard topology on the algebra of all transformations of V , and we extend to transformations that satisfy a polynomial the classical fact that the double-centralizer of a matrix is the algebra generated by that matrix.	math.RA	math	Infinite-Dimensional Triangularization Zachary Mesyan March 13, 2018 Abstract The goal of this paper is to generalize the theory of triangularizing matrices to linear transformations of an arbitrary vector space, without placing any restrictions on the dimension of the space or on the base field. We define a transformation T of a vector space V to be triangularizable if V has a well-ordered basis such that T sends each vector in that basis to the subspace spanned by basis vectors no greater than it. We then show that the following conditions (among others) are equivalent: (1) T is triangularizable, (2) every finite-dimensional subspace of V is annihilated by f (T ) for some polynomial f that factors into linear terms, (3) there is a maximal well-ordered set of subspaces of V that are invariant under T , (4) T can be put into a crude version of the Jordan canonical form. We also show that any finite collection of commuting triangularizable transformations is simultaneously triangularizable, we describe the closure of the set of triangularizable transformations in the standard topology on the algebra of all transformations of V , and we extend to transformations that satisfy a polynomial the classical fact that the double-centralizer of a matrix is the algebra generated by that matrix. Keywords: triangular matrix, linear transformation, simultaneous triangularization, canonical form, function topology, endomorphism ring, locally artinian module, double- centralizer 2010 MSC numbers: 15A04, 15A21 (primary), 16S50, 16W80 (secondary) 1 Introduction The following summarizes much of the existing wisdom on triangularizing a linear trans- formation of a finite-dimensional vector space. Our main goal is to generalize this to trans- formations of vector spaces of arbitrary dimension over an arbitrary field. Theorem 1 (Classical Triangularization Theorem). Let k be a field, V a finite-dimensional k-vector space, and T a linear transformation of V . Then the following are equivalent. (1) T has an upper-triangular representation as a matrix with respect to some basis for V . (1′) T has a lower-triangular representation as a matrix with respect to some basis for V . (2) There is a polynomial p(x) ∈ k[x] \ k that factors into linear terms in k[x], such that p(T ) = 0. 1 (3) There exists a well-ordered set of T -invariant subspaces of V , which is maximal as a well-ordered set of subspaces of V . (3′) There exists a totally ordered set of T -invariant subspaces of V , which is maximal as a totally ordered set of subspaces of V . (4) T has a representation as a matrix in Jordan canonical form with respect to some basis for V . Proof. (1) ⇔ (1′) The transformation T , viewed as a matrix, is upper-triangular with respect to a basis v1, . . . , vn for V if and only if it is lower-triangular with respect to vn, . . . , v1. (1) ⇒ (2) Suppose that T can be represented as an n × n upper-triangular matrix with diagonal entries a11, . . . , ann ∈ k, and let p(x) = (x − a11) · · · (x − ann). Then p(T ) = 0, by the Cayley-Hamilton theorem. (2) ⇒ (4) See, e.g., [2, Section 12.3, Theorem 22]. (4) ⇒ (1) A matrix in Jordan canonical form is necessarily upper-triangular. (1) ⇔ (3′) See, e.g., [7, page 1]. (3) ⇔ (3′) A finite set is totally ordered if and only if it is well-ordered. Given a field k and a k-vector space V , we denote by Endk(V ) the k-algebra of all linear transformations of V . We define a transformation T ∈ Endk(V ) to be triangularizable if V has a well-ordered basis (B, ≤) such that T sends each vector v ∈ B to the subspace spanned by {u ∈ B u ≤ v}. This definition clearly generalizes condition (1) in the above theorem. We then show, in Theorem 8, that for T ∈ Endk(V ) (with k and V arbitrary) being triangularizable is equivalent to satisfying condition (3) above, as well as to satisfying each of the following (along with another condition). (2′) For every finite-dimensional subspace W of V there is a polynomial p(x) ∈ k[x] \ k that factors into linear terms in k[x], such that p(T ) annihilates W . (4′) V = La∈kS∞ i=1 ker((T − aI)i), where I ∈ Endk(V ) is the identity transformation. If k is algebraically closed, then T being triangularizable is also equivalent to the following. (5) Every finite-dimensional subspace of V is contained in a finite-dimensional T -invariant subspace of V . (6) V is locally artinian, when viewed as a k[x]-module, where x acts on V as T . Condition (2′), of course, is a direct generalization of (2), while (4′) is a crude version of (4). There is extensive literature on triangularization of bounded linear operators on Banach spaces, where condition (3′) is taken to be the definition of "triangularizable" (see [7]). This leads to many beautiful results about when collections of operators can be simultaneously triangularized. However, (a statement equivalent to) the stronger condition (3) was cho- sen as the definition of "triangularizable" here, since much more of the intuition regarding triangularization, as summarized in Theorem 1, can be preserved this way. As we show in Example 9, for a transformation T of a vector space, satisfying (3′) is generally not equivalent to satisfying (2′), (3), and (4′). 2 With the basics of infinite-dimensional triangularization established, we then show that any finite collection of commuting triangularizable elements of Endk(V ) is simultaneously triangularizable (Theorem 15), generalizing a well-known fact from finite-dimensional linear algebra. Next we show that the inverse, if it exists, of a triangularizable transformation is also triangularizable with respect to the same well-ordered basis, as one would hope (Proposition 16). Then, after reviewing the standard topology on Endk(V ) in Section 6, we characterize, in Theorem 19, the closure of the set of triangularizable transformations in Endk(V ). In particular, if the field k is algebraically closed, then the closure of this set is Endk(V ), which generalizes the fact, known as Shur's theorem, that over an algebraically closed field every matrix is triangularizable. Then, in Proposition 20, we give a number of equivalent characterizations of topologically nilpotent transformations in Endk(V ), i.e., transformations T such that the sequence (T i)∞ i=1 converges to 0 in the topology on Endk(V ). These generalize the familiar fact that a matrix is nilpotent if and only if it is similar to a strictly upper-triangular matrix if and only if 0 is its only eigenvalue (over the algebraic closure of the base field). In Section 7 we discuss the transformations in Endk(V ) which satisfy a single polynomial on the entire space V . Finally, in Theorem 30 we generalize to transformations that satisfy a polynomial on the entire space the classical result that the double-centralizer of a matrix is the algebra generated by that matrix. 2 Preliminaries We begin with the following standard fact from finite-dimensional linear algebra, which will be useful for our purposes. The usual proof also works for a vector space of arbitrary dimension, but we provide it here for completeness. Lemma 2. Let k be a field, V a k-vector space, and T ∈ Endk(V ). Also suppose that f1(x), . . . , fn(x) ∈ k[x] are pairwise relatively prime polynomials, and set S = f1(T ) · · · fn(T ). Then ker(S) = Ln i=1 ker(fi(T )). Proof. For each i ∈ {1, . . . , n} let gi(x) = f1(x) · · · fi−1(x)fi+1(x) · · · fn(x). Since the fi(x) are pairwise relatively prime, {g1(x), . . . , gn(x)} is relatively prime in k[x]. Thus there exist h1(x), . . . , hn(x) ∈ k[x] such that 1 = Pn Let us now show that the sum Pn for some vi ∈ ker(fi(T )). Then for each j ∈ {1, . . . , n} we have i=1 ker(fi(T )) is direct. Thus suppose that 0 = Pn i=1 vi i=1 hi(x)gi(x). 0 = gj(T )(cid:18) n Xi=1 vi(cid:19) = gj(T )(vj), since f1(T ), . . . , fn(T ) commute with each other. Thus gj(T )(vi) = 0 for all i, j ∈ {1, . . . , n}, and therefore n vj = 1 · vj = Xi=1 hi(T )gi(T )(vj) = 0 3 for all j ∈ {1, . . . , n}. It follows that n Xi=1 ker(fi(T )) = n Mi=1 ker(fi(T )). Since the fi(T ) commute with each other, clearly ker(S) ⊇ Ln it suffices to show that the reverse inclusion holds. Let v ∈ ker(S). Then i=1 ker(fi(T )), and hence 0 = hi(T )S(v) = hi(T )gi(T )fi(T )(v) = fi(T )hi(T )gi(T )(v), and hence hi(T )gi(T )(v) ∈ ker(fi(T )) for all i ∈ {1, . . . , n}. Thus v = n Xi=1 hi(T )gi(T )(v) ∈ n Mi=1 ker(fi(T )), giving the desired conclusion. Recall that a binary relation ≤ on a set X is a partial order if it is reflexive, antisymmetric, and transitive. If, in addition, x ≤ y or y ≤ x for all x, y ∈ X, then ≤ is a total order. If, moreover, every non-empty subset of X has a least element with respect to ≤, then ≤ is a well order. We shall require the following standard set-theoretic fact. Lemma 3. Let (Λ, ≤Λ) be a well-ordered set, and for each λ ∈ Λ let (Ωλ, ≤λ) be a well- as follows: ordered set, such that the Ωλ are pairwise disjoint. Define a binary relation ≤ on Sλ∈Λ Ωλ for all α1, α2 ∈ Sλ∈Λ Ωλ, with α1 ∈ Ωλ1 and α2 ∈ Ωλ2, let α1 ≤ α1 if either λ1 = λ2 and α1 ≤λ1 α2, or λ1 <Λ λ2. Then (Sλ∈Λ Ωλ, ≤) is a well-ordered set. Proof. It is routine to check that ≤ is reflexive, antisymmetric, and transitive. That ≤ is a total order then follows from its definition and the fact that ≤Λ and each ≤λ is a total order. Now, let Γ be a nonempty subset of Sλ∈Λ Ωλ. Since Λ is well-ordered, there is a least λ ∈ Λ (with respect to ≤Λ) such that Γ ∩ Ωλ 6= ∅. Since Ωλ is well-ordered, there is a least α ∈ Ωλ (with respect to ≤λ) such that α ∈ Γ ∩ Ωλ. Then α must be the least element of Γ with respect to ≤, which shows that ≤ is a well order. Throughout the paper Z will denote the set of the integers, Z+ the set of the positive integers, and N the set of the natural numbers (including 0). We shall implicitly rely, whenever appropriate, on the fact that Z is totally ordered by its usual ordering, while Z+ and N are well-ordered. 3 Triangularization We now extend the notion of "upper-triangular" to transformations of an arbitrary vector space. Given a subset X of a vector space, we denote by hXi the subspace generated by X. 4 Definition 4. Let k be a field, V a k-vector space, T ∈ Endk(V ), B a basis for V , and ≤ a partial ordering on B. We say that T is triangular with respect to (B, ≤) if T (v) ∈ h{u ∈ B u ≤ v}i for all v ∈ B, and that T is strictly triangular with respect to (B, ≤) if T (v) ∈ h{u ∈ B u < v}i for all v ∈ B. If T is triangular, respectively strictly triangular, with respect to some well-ordered basis for V , then we say that T is triangularizable, respectively strictly triangularizable. The condition that T (v) ∈ h{u ∈ B u ≤ v}i for all v ∈ B, in the above definition, is based on the defining property of upper-triangular matrices. We could have used the lower-triangular analog instead: T (v) ∈ h{u ∈ B v ≤ u}i for all v ∈ B. This would have resulted in an equivalent notion of "triangular", for given a partially ordered basis (B, ≤), one has T (v) ∈ h{u ∈ B u ≤ v}i for all v ∈ B if and only if T (v) ∈ h{u ∈ B v ≤′ u}i for all v ∈ B, where ≤′ is the opposite partial ordering of ≤. (I.e., v ≤′ u if and only if u ≤ v, for all u, v ∈ B.) Even though we allowed ≤ to be an arbitrary partial order in the above definition, for occasional convenience, our primary interest will be in transformations that are triangular with respect to a well-ordered basis, which is why we reserve "triangularizable" for that case alone. We focus on this case since, as we shall see, such transformations behave very much like triangular matrices, and to a significantly greater extent than transformations that are triangular with respect to a merely totally ordered basis. Still, it may be of interest to investigate other sorts of partially ordered bases in this context. For example, one could describe the transformations that are triangular with respect to orderings that are opposite to well orderings, which would produce a theory substantially different (and more messy) than the one presented here. (An instance of such a transformation can be found in Example 18.) In order to avoid a lengthy digression, however, we shall not discuss such possibilities in detail. Sometimes, we shall find it more convenient to index bases with ordered sets rather than ordering the bases themselves, when dealing with triangularization. Also, given k- vector spaces W ⊆ V and a transformation T ∈ Endk(V ) we say that W is T -invariant if T (W ) ⊆ W . In Theorem 8 we shall give a number of equivalent characterizations of triangularizable transformations, but we require a few preliminary results. Lemma 5. Let k be a field, V a k-vector space, and T ∈ Endk(V ). Then T is triangularizable if and only if there exists a well-ordered (by inclusion) set of T -invariant subspaces of V , which is maximal as a well-ordered set of subspaces of V . Proof. Suppose that T is triangularizable. Then there is a well-ordered set (Ω, ≤) and a basis B = {vα α ∈ Ω} for V such that T (vα) ∈ h{vβ β ≤ α}i for all α ∈ Ω. Since every well-ordered set is order-isomorphic to an ordinal, we may assume that Ω is an ordinal. For each α ∈ Ω set Vα = h{vβ β < α}i, where V0 is understood to be the zero space (0 being the least element of Ω). Then for all α1, α2 ∈ Ω we have Vα1 ⊆ Vα2 if and only if α1 ≤ α2. Since (Ω, ≤) is well-ordered, it follows that X = {Vα α ∈ Ω+} is well-ordered by set inclusion, where Ω+ = Ω∪{Ω} is the successor of Ω and V = VΩ. Moreover, T (vβ) ∈ Vα for all α, β ∈ Ω satisfying β < α, from which it follows that each element of X is T -invariant. It remains 5 to show that X is maximal. First, note that for each α ∈ Ω+ we have Sβ<α Vβ ⊆ Vα, with equality if α is a limit ordinal, and Vα/(Sβ<α Vβ) one-dimensional otherwise. Now, let W ⊆ V be a subspace that is comparable under set inclusion to Vα for each α ∈ Ω+. Since Ω+ is well-ordered, there is a least α ∈ Ω+ such that W ⊆ Vα. Since W is comparable to each element of X, from the choice of Vα it follows that Vβ ⊂ W for all β < α. If α is a limit ordinal, then Vα = Sβ<α Vβ ⊆ W , and hence W = Vα ∈ X. Otherwise, there is a β ∈ Ω+ such that α is the successor of β, and Vβ ⊂ W ⊆ Vα. But in this case Vα/Vβ is one-dimensional, and therefore W = Vα ∈ X once again. Thus X is a maximal well-ordered set of subspaces of V . Conversely, suppose that there exists a well-ordered set (Ω, ≤), which we may assume to be an ordinal, and a set X = {Vα α ∈ Ω} of T -invariant subspaces of V , such that Vα1 ⊆ Vα2 if and only if α1 ≤ α2 (for all α1, α2 ∈ Ω), and X is maximal as a well-ordered set of subspaces of V . Let α ∈ Ω be any element. If α is a successor ordinal, with predecessor β, then Vα/Vβ must be one-dimensional, by the maximality of X. If α is a limit ordinal, then for all β < α we have Vβ ⊂ Sγ<α Vγ ⊆ Vα. Again, by the maximality of X, this implies that Sγ<α Vγ ∈ X, and hence Sγ<α Vγ = Vα. Now for each β ∈ Ω with successor α ∈ Ω let vα ∈ V be such that vα + Vβ spans Vα/Vβ. Also, let Γ = {α ∈ Ω α is a successor ordinal}. As a subset of a well-ordered set, Γ is itself well-ordered by (the restriction of) ≤. We claim that {vα α ∈ Γ} is a basis for V with respect to which T is triangular. Since for each α ∈ Ω we have Vα = h{vβ β ≤ α, β ∈ Γ}i, and since X must contain V , by virtue of being maximal, if follows that {vα α ∈ Γ} spans V . From the fact that the spaces Vα are distinct it also follows that {vα α ∈ Γ} is linearly independent, and hence is a basis for V . Now let α ∈ Γ be any element. Then T (vα) ∈ Vα, since Vα is T -invariant, and hence T (vα) ∈ h{vβ β ≤ α, β ∈ Γ}i. That is, T is triangular with respect to {vα α ∈ Γ}. Proposition 6. Let k be a field, V a k-vector space, (B, ≤) a well-ordered basis for V , and T ∈ Endk(V ) a transformation triangular with respect to B. Then the following hold. (1) If W ⊆ V is a finite-dimensional subspace, then W is contained in a finite-dimensional T -invariant subspace of V . (2) There is a partial ordering (cid:22) on B such that T is triangular with respect to (B, (cid:22)) and {u ∈ B u (cid:22) v} is finite for all v ∈ B. Proof. (1) Let U1 ⊆ B be a finite subset such that W ⊆ hU1i. Now for each i > 1 (i ∈ Z+) define recursively Ui = {v ∈ B πvT (Ui−1) 6= 0} ∪ Ui−1, where πv ∈ Endk(V ) is the projection onto hvi with kernel hB \ {v}i. Since U1 is finite, it follows by induction that every Ui is finite. Also, we have T (Ui) ⊆ hUi+1i for all i ∈ Z+. We claim that the chain U1 ⊆ U2 ⊆ U3 ⊆ · · · If not, then for each i ∈ Z+ let vi ∈ B be the must stabilize after finitely many steps. maximal element, with respect to ≤, such that vi ∈ Ui \ Ui−1 (where U0 is understood to be 6 the empty set). This is well-defined since each Ui \ Ui−1 is finite but nonempty. Then for each i > 1, there exists u ∈ Ui−1 \ Ui−2 such that πviT (u) 6= 0, by the definition of Ui. Since T is triangular with respect to B, we have T (u) ∈ h{w ∈ B w ≤ u}i, from which it follows that vi ≤ u ≤ vi−1. Moreover, since vi ∈ Ui \ Ui−1 and vi−1 ∈ Ui−1, we have vi < vi−1. Thus v1 > v2 > v3 > · · · is an infinite strictly descending chain of elements of B. This contradicts B being well- ordered, since {v1, v2, v3, . . . } has no least element. Hence there exists n ∈ Z+ such that Un = Un+1, and therefore T (Un) ⊆ hUn+1i = hUni. It follows that T (hUni) ⊆ hUni, where W ⊆ hUni and hUni is finite-dimensional, as desired. (2) Given u, v ∈ B, we write u (cid:22) v if either u = v or there exist w1, . . . , wn ∈ B, where w1 = v and wn = u, such that πwnT πwn−1T πwn−2 · · · πw2T πw1 6= 0. (Note that this product being nonzero is equivalent to each of πwnT πwn−1, . . . , πw2T πw1 being nonzero, since the image of each of the projections involved is 1-dimensional.) Let v ∈ B, and let U1 = {v}. Then defining Ui for all i > 1 as in the proof of (1), for any u ∈ B we have u (cid:22) v if and only if u ∈ Ui for some i ∈ Z+. Since the chain U1 ⊆ U2 ⊆ U3 ⊆ · · · must stabilize after finitely many steps, and each Ui is finite, it follows that {u ∈ B u (cid:22) v} is finite for all v ∈ B. It remains to show that (cid:22) is a partial order. The binary relation (cid:22) is reflexive, by definition. To show that (cid:22) is antisymmetric, first we note that given u, v ∈ B, if πuT πv 6= 0, then u ≤ v, since T is triangular with respect to (B, ≤), and hence u (cid:22) v implies that u ≤ v. (I.e., ≤ extends (cid:22).) Thus, if u (cid:22) v and v (cid:22) u for some u, v ∈ B, then u ≤ v and v ≤ u, from which it follows that u = v. Finally, to show that (cid:22) is transitive, suppose that u (cid:22) v and v (cid:22) w for some u, v, w ∈ B. Then there exist x1, . . . , xn, y1, . . . , ym ∈ B, where x1 = v = ym, xn = u, y1 = w, such that πxnT πxn−1 · · · πx2T πx1 6= 0 and πymT πym−1 · · · πy2T πy1 6= 0. It follows that πuT πxn−1 · · · πx2T (v) = au and πvT πym−1 · · · πy2T (w) = bv for some a, b ∈ k \ {0}. Therefore, πuT πxn−1 · · · πx2T πvT πym−1 · · · πy2T (w) = abu 6= 0, and hence u (cid:22) w, as required. Lemma 7. Let k be a field, V a k-vector space, and T ∈ Endk(V ). If V = S∞ then T is strictly triangularizable. i=1 ker(T i), 7 Proof. For each i ≥ 1 let Ui ⊆ ker(T i) \ ker(T i−1) be a linearly independent set such that i=1 Ui is Ui + ker(T i−1) is a basis for ker(T i)/ ker(T i−1), where Ui is possibly empty. Then Sn a basis for ker(T n), for each n ∈ Z+, and hence B = S∞ i=1 Ui is a basis for V . Now for each i ∈ Z+, let ≤i be a well-ordering on Ui (chosen arbitrarily), and define a binary relation ≤B on B as follows. Given u1, u2 ∈ B, where u1 ∈ Ui1 and u2 ∈ Ui2 (i1, i2 ∈ Z+), let u1 ≤B u2 if either i1 = i2 and u1 ≤i1 u2, or i1 < i2. Then, by Lemma 3, (B, ≤B) is well-ordered. Finally, let u ∈ B, and let n ∈ Z+ be such that u ∈ Un (⊆ ker(T n)). Then T (u) ∈ ker(T n−1) = (cid:28) n−1 [i=1 Ui(cid:29) ⊆ h{v ∈ B v <B u}i, by the definition of ≤B, which shows that T is strictly triangular with respect to B. We are now ready for our main result, which characterizes the triangularizable transfor- mations of an arbitrary vector space. The conditions (1)–(3) in the statement generalize the corresponding ones in Theorem 1, while condition (4) is a crude version of Theorem 1(4). We recall that given a commutative ring R, an R-module M is called locally artinian if every finitely-generated R-submodule of M is artinian. Theorem 8. Let k be a field, V a k-vector space, and T ∈ Endk(V ). Then the following are equivalent. (1) T is triangularizable. (2) For every finite-dimensional subspace W of V there is a polynomial p(x) ∈ k[x] \ k that factors into linear terms in k[x], such that p(T ) annihilates W . (3) There exists a well-ordered set of T -invariant subspaces of V , which is maximal as a well-ordered set of subspaces of V . (4) V = La∈kS∞ i=1 ker((T − aI)i), where I ∈ Endk(V ) is the identity transformation. (5) There is a partially ordered basis (B, (cid:22)) for V such that T is triangular with respect to (B, (cid:22)) and {u ∈ B u (cid:22) v} is finite for all v ∈ B. Moreover, if k is algebraically closed, then these are also equivalent to the following. (6) Every finite-dimensional subspace of V is contained in a finite-dimensional T -invariant subspace of V . (7) V is locally artinian, when viewed as a k[x]-module, where x acts on V as T . Proof. By Proposition 6, (1) implies (5) and (6). Also, (1) and (3) are equivalent, by Lemma 5. We shall prove that (5) ⇒ (2) ⇒ (4) ⇒ (1), and then treat (6) and (7) at the end. (5) ⇒ (2) Let (B, (cid:22)) be as in (5), and let W be a finite-dimensional subspace of V . We can find a finite subset X of B such that W ⊆ hXi. Then, by hypothesis, the set Y = {u ∈ B ∃v ∈ X (u (cid:22) v)} 8 is finite. Since (cid:22) is transitive, for all u ∈ Y and v ∈ B such that v (cid:22) u, we have v ∈ Y . Hence, the assumption that T is triangular with respect to (B, (cid:22)) implies that hY i is T - invariant. By the order-extension principle, the restriction of (cid:22) to Y ⊆ B can be extended to a total order ≤ on Y . Then T (v) ∈ h{u ∈ Y u (cid:22) v}i ⊆ h{u ∈ Y u ≤ v}i for all v ∈ Y . Hence the restriction of T to hY i is triangular with respect to (Y, ≤), and so can be represented as a (finite) upper-triangular matrix. Therefore, by Theorem 1, there is a polynomial p(x) ∈ k[x] \ k that factors into linear terms in k[x], such that p(T ) annihilates hY i and hence also W ⊆ hXi ⊆ hY i, proving (2). (2) ⇒ (4) Suppose that (2) holds, and let P ⊆ k[x] \ k be the subset consisting of all the polynomials that factor into linear terms. Then V = [p∈P ker(p(T )) = Ma∈k [i=1 ker((T − aI)i), ∞ by Lemma 2, and hence (4) holds. enough to show that for each a ∈ k there is a well-ordered basis for S∞ respect to which T is triangular (when restricted to the T -invariant subspace S∞ aI)i) of V ). Thus, let us assume that V = S∞ (4) ⇒ (1) Suppose that (4) holds. Upon well-ordering k, by Lemma 3, to prove (1), it is i=1 ker((T − aI)i) with i=1 ker((T − i=1 ker((T − aI)i) for some a ∈ k, and let S = T − aI. Then, by Lemma 7, there is a well-ordered basis (B, ≤) for V , with respect to which S is triangular. Thus, for any v ∈ B we have T (v) = S(v) + av ∈ h{u ∈ B u ≤ v}i, showing that T is triangular with respect to (B, ≤). We have shown that (1)–(5) are equivalent. Next, let us suppose that k is algebraically closed and that (6) holds, and show that (2) also holds. Let W be a finite-dimensional subspace of V . Then, by (6), there is a finite-dimensional T -invariant subspace W ′ of V containing W . Viewing the restriction of T to W ′ as a (finite) matrix, there is a polynomial p(x) ∈ k[x] \ k such that p(T ) annihilates W ′ (e.g., by the Cayley-Hamilton theorem). In particular, p(T ) annihilates W . Since k is algebraically closed, p(x) factors into linear terms in k[x], showing that (2) holds. Thus, when k is algebraically closed, (1)–(6) are equivalent. To conclude the proof, we shall show that (6) and (7) are equivalent. Thus suppose that (6) holds, and let M be a finitely-generated k[x]-submodule of V , where x acts as T . Let W ⊆ M be a finite set such that M = k[x]W . Then W is contained in a finite-dimensional T -invariant subspace M ′ of V , by (6). But T -invariant subspaces of V are precisely the k[x]- submodules of V , which shows that M is contained in the finite-dimensional k[x]-submodule M ′ of V . Thus M is finite-dimensional as a k-vector space, and hence artinian, proving (7). Conversely, suppose that (7) holds, and let W be a finite-dimensional subspace of V . Then, by (7), the k[x]-submodule M = k[x]W of V is artinian. Since M is a T -invariant subspace of V , to conclude that (6) holds it suffices to show that M is finite-dimensional. But since k[x] is a principal ideal domain and M is a finitely-generated k[x]-module, M ∼= k[x]r ⊕ k[x]/hf1(x)i ⊕ · · · ⊕ k[x]/hfn(x)i, 9 where r ∈ N, f1(x), . . . , fn(x) ∈ k[x] \ {0}, and hfi(x)i denotes the ideal of k[x] generated by fi(x). (See, e.g., [2, Section 12.1, Theorem 5].) Since M is artinian, we must have r = 0, and hence M is finite-dimensional as a k-vector space, giving the desired conclusion. As mentioned in the Introduction, in the literature on bounded linear operators on Banach spaces, a transformation T is said to be "triangularizable" if there is a chain (i.e., totally ordered set) of T -invariant subspaces of the Banach space which is maximal as a chain of subspaces (see [7, Definition 7.1.1]). That is, for such operators, condition (3′) from Theorem 1 is used to generalize the notion of "triangular" from finite-dimensional spaces to infinite-dimensional ones. By using the stronger condition (3) instead (which, by the previous theorem, is equivalent to T being triangularizable, as we have defined the term) in our generalization of "triangular" we acquire much greater control over the behavior of transformations, as the next example demonstrates. To facilitate the discussion, we say that a transformation T of a vector space V is chain- triangularizable if there is a chain of T -invariant subspaces of V , which is maximal as a chain of subspaces of V . Example 9. Let k be a field and V a k-vector space with basis {vi i ∈ Z}. Define T ∈ Endk(V ) by T (vi) = vi−1 for each i ∈ Z, and extend linearly to all of V . Also for each i ∈ Z let Vi = h{vj j ≤ i}i. Then · · · ⊆ V−1 ⊆ V0 ⊆ V1 ⊆ · · · is a maximal chain of subspaces of V (since every Vi/Vi−1 is 1-dimensional, V = Si∈Z Vi, and 0 = Ti∈Z Vi), each T -invariant. Thus T is chain-triangularizable. However, T satisfies none of the seven conditions in Theorem 8. To see this, let W = hv0i. Then any T -invariant subspace of V that contains W must contain v−1, v−2, . . . , and hence also V0. Therefore T does not satisfy condition (6) in Theorem 8, and is hence not triangularizable, by Proposition 6. It follows that T does not satisfy any of the conditions (1)–(7) in Theorem 8. (cid:3) Let us next derive a useful consequence of Theorem 8. Given k-vector spaces W ⊆ V and a transformation T ∈ Endk(V ), we denote by T W the restriction of T to W . Corollary 10. Let k be a field, V a k-vector space, T ∈ Endk(V ) triangular with respect to some well-ordered basis for V , and W ⊆ V a T -invariant subspace. (1) T W is triangular with respect to some well-ordered basis for W . (2) Let T ∈ Endk(V /W ) be the transformation defined by T (v + W ) = T (v) + W . Then T is triangular with respect to some well-ordered basis for V /W . Proof. (1) Let U ⊆ W be a finite-dimensional subspace. Since U ⊆ V and T is triangular- izable, by Theorem 8, there is a polynomial p(x) ∈ k[x] \ k that factors into linear terms in k[x], such that p(T ) annihilates U. Since T (W ) ⊆ W , we have p(T )W = p(T W ), and hence p(T W ) annihilates U. Therefore, by Theorem 8, T W is triangular with respect to some well-ordered basis for W . 10 (2) First, note that T is well-defined. For if v1 + W = v2 + W for some v1, v2 ∈ V , then v1 − v2 = w for some w ∈ W . Hence T (v1 + W ) = T (v1) + W = T (v2) + T (w) + W = T (v2) + W = T (v2 + W ). It is routine to verify that T is also linear. Now, let W ⊆ U ⊆ V be a subspace such that U/W is finite-dimensional. Then there exist u1, . . . , un ∈ U such that U/W = hu1 + W, . . . , un + W i. Since T is triangularizable, by Theorem 8, there is a polynomial p(x) ∈ k[x] \ k that factors into linear terms in k[x], such that p(T ) annihilates hu1, . . . , uni ⊆ V . Then for any i ∈ {1, . . . , n} we have p(T )(ui + W ) = p(T )(ui) + W = W, showing that p(T ) annihilates U/W . Thus, T is triangular with respect to some well-ordered basis for V /W , by Theorem 8. To complement our description of triangularizable transformations we also give a more specialized description of diagonalizable ones. This is part of [3, Proposition 4.13], but we present a more direct proof here. Proposition 11. Let k be a field, V a k-vector space, and T ∈ Endk(V ). Then the following are equivalent. (1) T is diagonalizable. (I.e., there is a basis for V consisting of eigenvectors of T .) (2) For every finite-dimensional subspace W of V there is a polynomial p(x) ∈ k[x] \ k that factors into distinct linear terms in k[x], such that p(T ) annihilates W . Proof. Suppose that T is diagonalizable, and let B be a basis for V consisting of eigenvectors of T . To prove (2), let W ⊆ V be a finite-dimensional subspace. Then we can find v1, . . . , vn ∈ B such that W ⊆ hv1, . . . , vni. By hypothesis, for each i ∈ {1, . . . , n} there exists ai ∈ k such that T (vi) = aivi. Let a1, . . . , al be the distinct elements of {a1, . . . , an} (upon reindexing, if necessary), and set S = (T −a1I) · · · (T −alI), where I ∈ Endk(V ) the identity transformation. Then, since the factors T −aiI commute with each other, S(hv1, . . . , vni) = 0. Thus letting p(x) = (x − a1) · · · (x − al), we have p(T )(W ) = 0. Conversely, suppose that (2) holds, for each a ∈ k let Ba be a basis for ker(T − aI), and let B = Sa∈k Ba. Then, by Lemma 2, B is a basis for V , and clearly B consists of eigenvectors of T . 4 Simultaneous Triangularization Our next goal is to show that any finite commuting collection of triangularizable trans- formations is triangular with respect to a common well-ordered basis. This generalizes the classical fact that any commuting collection of triangularizable transformations of a finite- dimensional vector space is upper-triangular with respect to some basis for that vector space. The following notation and observations will be useful. 11 Definition 12. Given a ring R and a subset X ⊆ R we denote by CR(X) (or C(X), if there is no danger of ambiguity) the centralizer (or commutant) {r ∈ R rx = xr for all x ∈ X} of X in R. Given r ∈ R we shall also write CR(r) to mean CR({r}). Lemma 13. Let k be a field, V a nonzero k-vector space, and T ∈ Endk(V ) triangularizable. Then there exists a ∈ k such that W = ker(T − aI) is nonzero, where I ∈ Endk(V ) is the identity transformation. Moreover any such W satisfies C(T )(W ) ⊆ W . Proof. Let (B, ≤) be a well-ordered basis with respect to which T is triangular. Since V 6= 0, we have B 6= ∅. Let v ∈ B be the least element with respect to ≤. Then T (v) ∈ hvi, and hence T (v) = av for some a ∈ k. Letting W = ker(T − aI), we see that W 6= 0, since v ∈ W . Now let S ∈ C(T ) be any element. Then (T − aI)S(W ) = S(T − aI)(W ) = 0, and hence S(W ) ⊆ W . It follows that C(T )(W ) ⊆ W . Lemma 14. Let k be a field, V a nonzero k-vector space, and X ⊆ Endk(V ) a finite commutative collection of transformations. If each element of X is triangularizable, then there exists a 1-dimensional subspace W ⊆ V such that X(W ) ⊆ W . Proof. Write X = {T1, . . . , Tn}. It suffices to construct a nonzero subspace of V on which each Ti acts as a scalar multiple of the identity, since any subspace U of such a space would satisfy X(U) ⊆ U, and in particular, any 1-dimensional subspace. By Lemma 13, there exists a1 ∈ k such that W1 = ker(T1 − a1I) satisfies 0 6= W1 and X(W1) ⊆ W1. In particular, T1 acts as a scalar multiple of the identity on W1. By Corollary 10, the restriction XW1 of X to W1 is a commutative collection of transformations in Endk(W1), each triangularizable. Applying Lemma 13 again, we find a2 ∈ k such that W2 = ker(T2W1 − a2I) ⊆ W1 satisfies 0 6= W2 and XW1(W2) ⊆ W2, and hence also X(W2) ⊆ W2. Now both T1 and T2 act as scalar multiples of the identity on W2. Continuing in this fashion, the construction will yield a nonzero subspace Wm of V (m ≤ n) on which every Ti acts this way. Theorem 15. Let k be a field, V a k-vector space, and X ⊆ Endk(V ) a finite commutative collection of transformations. If each element of X is triangularizable, then there exists a well-ordered basis for V with respect to which every element of X is triangular. Proof. We begin by constructing recursively for each ordinal α a subspace Vα ⊆ V that is invariant under X, and for each successor ordinal α a vector vα ∈ V . Set V0 = 0. Now let α be an ordinal and assume that Vγ has been defined for every γ < α. If α is a limit ordinal, then let Vα = Sγ<α Vγ. Since each Vγ is assumed to be invariant under X, their union Vα will also be invariant under X. Next, if α is a successor ordinal, then let β be its predecessor. By Corollary 10, the transformation on V /Vβ induced by each element of X is triangular with respect to some basis for V /Vβ. Thus, by Lemma 14, there is a 1-dimensional subspace W/Vβ of V /Vβ invariant under each transformation on V /Vβ induced by an element of X 12 (assuming that V 6= Vβ). Let vα ∈ V be such that {vα + Vβ} is a basis for W/Vβ, and define Vα = hVβ ∪ {vα}i. Then Vα must be invariant under X, because of the invariance of Vβ and W . We proceed in this fashion until V = Sα∈Λ Vα for some ordinal Λ. Now let Γ = {α ∈ Λ α is a successor ordinal}, and let B = {vα α ∈ Γ}. Since we introduced new vectors only at successor steps in our construction, V = [α∈Γ Vα = [α∈Γ h{vγ γ ≤ α, γ ∈ Γ}i, and hence V = hBi. Since Vα/Vβ = hvα + Vβi is 1-dimensional for all α ∈ Γ with predecessor β, we conclude that B is a basis for V . Also, since Vα = h{vγ γ ≤ α, γ ∈ Γ}i is invariant under X for all α ∈ Γ, it follows that X(vα) ∈ h{vγ γ ≤ α, γ ∈ Γ}i for all α ∈ Γ. Thus, every element of X is triangular with respect to B, a basis for V indexed by the well-ordered set Γ. In [3, Example 4.17] there is a construction of a countably infinite commutative set E of transformations of a countably infinite-dimensional vector space V , over an arbitrary field, such that each transformation in E is diagonalizable (an idempotent, actually), but such that no 1-dimensional subspace of V is invariant under E. Thus, there is no well-ordered basis for V with respect to which every element of E is triangular, since the least element of such a basis would be an eigenvector of every element of E. Hence, Theorem 15 cannot be extended to arbitrary infinite commutative collections of triangularizable transformations. 5 Inverses In the following proposition we generalize the facts that an upper-triangular matrix is invertible if and only if it has only nonzero diagonal entries, and that the inverse of an upper-triangular matrix is also upper-triangular. These are simple observations, but they further reinforce the idea that our notion of "triangularizable" preserves intuition from finite- dimensional linear algebra. Proposition 16. Let k be a field, V a k-vector space, and T ∈ Endk(V ) triangular with respect to some well-ordered basis (B, ≤) for V . Also for each v ∈ B let πv ∈ Endk(V ) be the projection onto hvi with kernel hB \ {v}i. Then the following are equivalent. (1) T is invertible. (2) The restriction of T to any finite-dimensional T -invariant subspace of V is invertible. (3) T is injective. (4) T (hu ∈ B u ≤ vi) = hu ∈ B u ≤ vi for all v ∈ B. (5) πvT πv 6= 0 for all v ∈ B. Moreover, if T is invertible, then its inverse is triangular with respect to (B, ≤). 13 Proof. We shall show that (1) ⇒ (2) ⇒ (3) ⇒ (4) ⇒ (5) ⇒ (3) ⇒ (1). For the rest of the proof let Uv = hu ∈ B u ≤ vi for each v ∈ B. (1) ⇒ (2) Let W be a finite-dimensional T -invariant subspace of V . If T is invertible, then ker(T ) = 0, and hence also ker(T W ) = 0. Standard finite-dimensional linear algebra then gives that T W is invertible. (2) ⇒ (3) Suppose that T (v) = 0 for some v ∈ V . Then by Proposition 6, v is an element of some finite-dimensional T -invariant subspace W of V . Now, by (2) T W is invertible, and therefore T (v) = 0 = T W (v) implies that v = 0. Thus T is injective. (3) ⇒ (4) Suppose that T is injective. Since T is triangular with respect to B, we have T (Uv) ⊆ Su≤v Uu = Uv for all v ∈ B. Now suppose that Uw 6⊆ T (Uw) for some w ∈ B. Since B is well-ordered, we may assume that w is the least element of B with this property. Thus for all v ∈ B such that v < w, we have v ∈ Uv = T (Uv) ⊆ T (Uw), and therefore w 6∈ T (Uw). Since T is triangular with respect to B this implies that T (w) ∈ hu ∈ B u < wi. Thus either T (w) = 0 or T (w) ∈ Uv for some v ∈ B such that v < w. But, by hypothesis, Uv = T (Uv) for any v < w, and hence either T (w) = 0 or T (u) = T (w) for some u ∈ Uv, both of which would contradict T being injective. Therefore T (Uv) = Uv for all v ∈ B. (4) ⇒ (5) Suppose that T (Uv) = Uv for all v ∈ B. Then, given any v ∈ B, there i=1 aiui for some ai ∈ k and ui ∈ B, such that u1 < u2 < · · · < un = v. Since T is triangular with respect to B, we have exists w ∈ Uv such that T (w) = v. Write w = Pn T (Pn−1 i=1 aiui) ∈ Uun−1 (if n > 1). Hence v = πvT (w) = πvT(cid:18) n−1 Xi=1 aiui(cid:19) + anπvT (un) = anπvT (v), and therefore πvT πv 6= 0. i=1 aivi for some ai ∈ k \ {0} and vi ∈ B, such that v1 < v2 < · · · < vn. Since T is triangular with respect to (5) ⇒ (3) Let w ∈ V \ {0}, and suppose that T (w) = 0. Write w = Pn B, we have T (Pn−1 i=1 aivi) ∈ Uvn−1 (if n > 1). Hence 0 = πvnT (w) = πvnT(cid:18) n−1 Xi=1 aivi(cid:19) + anπvnT (vn) = anπvnT (vn), which implies that πvT πv = 0, since an 6= 0. Thus if πvT πv 6= 0 for all v ∈ B, then T must be injective. (3) ⇒ (1) Supposing that T is injective, we also have T (Uv) = Uv for all v ∈ B, by (3) ⇒ (4). Thus, v ∈ T (Uv) for all v ∈ B, and therefore B ⊆ T (V ), which implies that T is surjective. Therefore T is a bijection. The desired conclusion now follows from the easy fact that the inverse of any k-linear bijection from V to V is necessarily k-linear. For the final claim, suppose that T is invertible, with inverse T −1 ∈ Endk(V ). Then we have T (Uv) = Uv for all v ∈ B, by the equivalence of (1) and (3). Therefore T −1(Uv) = Uv for all v ∈ B, and in particular T −1(v) ∈ Uv. Hence T −1 is also triangular with respect to B. We note that a triangularizable transformation can be surjective without being invertible, in contrast to the situation with injectivity discussed above. For example, let k be a field, 14 V a k-vector space with basis B = {vi i ∈ N}, and T ∈ Endk(V ) such that T (v0) = 0 and T (vi) = vi−1 for all i ≥ 1. Then clearly T is (strictly) triangular with respect to B and surjective, but it is not injective. The next two examples show that chain-triangularizable transformations are not nearly as well-behaved with respect to inversion as triangularizable ones. Example 17. Let k be a field and V a k-vector space with basis {vi i ∈ Z}. Define T ∈ Endk(V ) by T (vi) = vi−1 for each i ∈ Z, and extend linearly to all of V . As seen in Example 9, T is chain-triangularizable but not triangularizable. Clearly T is invertible, with inverse T −1 defined by T −1(vi) = vi+1 for all i ∈ Z. Letting Vi = h{vj j ≤ i}i for each i ∈ Z, as we showed in Example 9, · · · ⊆ V−1 ⊆ V0 ⊆ V1 ⊆ · · · is a maximal chain of T -invariant subspaces of V , each T -invariant. However none of the Vi is T −1-invariant, since T −1(vi) = vi+1 /∈ Vi for each i ∈ Z. Thus, a "triangularizing chain" for T need not be one for T −1. Or, to put it another way, T is triangular with respect to the totally ordered basis {vi i ∈ Z}, but T −1 is not. (cid:3) Example 18. Let k be a field and V a k-vector space with basis {vi i ∈ N}. Define T ∈ Endk(V ) by T (vi) = vi+1 for each i ∈ N, and extend linearly to all of V . Also for each i ∈ N let Vi = h{vj j ≥ i}i. Then V0 ⊇ V1 ⊇ V2 ⊇ · · · is a maximal chain of subspaces of V , each T -invariant (by the same argument as in Exam- ple 9). Thus T is chain-triangularizable. On the other hand, T is not triangularizable, by Proposition 6, since the only finite-dimensional T -invariant subspace of V is the zero space. Now, given the previous observation, it is vacuously true that the restriction of T to any finite-dimensional T -invariant subspace of V is invertible. But unlike triangularizable transformations with this property, T itself is certainly not invertible, since v0 /∈ T (V ), and hence T is not surjective. (cid:3) 6 Topology We begin this section by recalling the standard topology on the ring Endk(V ), which will help us with subsequent results. Let X and Y be sets, and let Y X denote the set of all functions X → Y . The function (or finite) topology on Y X has a base of open sets of the following form: {f ∈ Y X f (x1) = y1, . . . , f (xn) = yn} (x1, . . . , xn ∈ X, y1, . . . , yn ∈ Y ). It is straightforward to see that this coincides with the product topology on Y X = QX Y , where each component set Y is given the discrete topology. As a product of discrete spaces, this space is Hausdorff. Now let V be a vector space over a field k. Then Endk(V ) ⊆ V V inherits a topology from the function topology on V V , which we shall also call the function topology. Under this 15 topology Endk(V ) is a topological ring (see, e.g., [9, Theorem 29.1]), i.e., a ring R equipped with a topology that makes + : R × R → R, − : R → R, and · : R × R → R continuous. Alternatively, we may describe the function topology on Endk(V ) as the topology having a base of open sets of the following form: {S ∈ Endk(V ) SW = T W } (T ∈ Endk(V ), W ⊆ V a finite-dimensional subspace). Observe that when V is finite-dimensional, Endk(V ) is discrete in this topology. Next, we describe the closure of the set of triangularizable transformations in Endk(V ) with respect to the above topology. This result generalizes (as did Theorem 8) Shur's theo- rem, which says that every (finite) matrix over an algebraically closed field is triangularizable. Theorem 19. Let k be a field and V a k-vector space. Define T ⊆ Endk(V ) to be the subset of all triangularizable transformations, and let T ⊆ Endk(V ) be the closure of T in the function topology. Then for all T ∈ Endk(V ), we have T ∈ T if and only if the restriction of T to any finite- dimensional T -invariant subspace of V is triangularizable. In particular, if k is algebraically closed, then T = Endk(V ). Proof. Suppose that T ∈ T , and let W ⊆ V be a finite-dimensional T -invariant subspace. Since T ∈ T , there exists S ∈ T that agrees with T on W . Since S is triangularizable, by Theorem 8, there is a polynomial p(x) ∈ k[x] \ k that factors into linear terms, such that p(S) annihilates W . It follows that p(T W ) annihilates W as well, and hence, by Theorem 1 (or Theorem 8), T W is triangularizable. Conversely, suppose that the restriction of T to any finite-dimensional T -invariant sub- space of V is triangularizable, and let U be an open neighborhood of T . Passing to a subset, if necessary, we may assume that U = {H ∈ Endk(V ) HW = T W } for some finite-dimensional subspace W of V . We shall show that U contains a triangular- izable transformation, from which the desired conclusion follows. We may view V as a k[x]-module, where x acts on V as T . Then M = k[x]W is a finitely- generated k[x]-submodule of V . Since k[x] is a principal ideal domain, M ∼= k[x]r ⊕N, where r ∈ N and N is a torsion k[x]-module. (See, e.g., [2, Section 12.1, Theorem 5].) Hence there exist subspaces M1, M2 ⊆ V such that M = M1 ⊕ M2, M1 has a basis of the form {T i(vj) 1 ≤ j ≤ r, i ∈ N} (where T i1(vj1) 6= T i2(vj2) whenever (i1, j1) 6= (i2, j2)), and for every w ∈ M2 there is In particular, every w ∈ M2 is contained some p(x) ∈ k[x] \ k such that p(T )(w) = 0. in a finite-dimensional T -invariant subspace of V . (Specifically, the space spanned by w, T (w), T 2(w), . . . , T n−1(w), where n is the degree of a polynomial p(x) ∈ k[x] \ k such that p(T )(w) = 0, is invariant under T .) Since W is finite-dimensional, we can find finite-dimensional subspaces W1 ⊆ M1 and W2 ⊆ M2 such that W ⊆ W1 ⊕ W2. Since, by the above, W2 is contained in a finite- dimensional T -invariant subspace of V , upon enlarging W2, if necessary, we may assume that it is T -invariant. Hence, by hypothesis, T W2 is triangularizable. 16 Upon enlarging W1, if necessary, we may assume that W1 has a basis of the form {T i(vj) 1 ≤ j ≤ r, 0 ≤ i ≤ nj}, for some n1, . . . , nr ∈ N. Let W + 1 = h{T i(vj) 1 ≤ j ≤ r, 0 ≤ i ≤ nj + 1}i. Define S ∈ Endk(V ) on {T i(vj) 1 ≤ j ≤ r, 0 ≤ i ≤ nj + 1} by S(T i(vj)) = (cid:26) T i+1(vj) 0 if i ≤ nj if i = nj + 1 , and extend S to a transformation on V by letting it act as T on W2 and as the zero transformation on a complement of W + 1 ⊕W2. Then S agrees with T on W , and hence S ∈ U. Moreover, S is triangularizable, since T W2 is triangularizable, while SW + is nilpotent, and hence triangularizable, by Theorem 1 (or Lemma 7 or Theorem 8). 1 For the final claim, suppose that k is algebraically closed, and let T ∈ Endk(V ). Suppose also that W ⊆ V is a finite-dimensional T -invariant subspace. Then, by the Cayley-Hamilton theorem, T satisfies a polynomial on W . Since k is algebraically closed, this polynomial can be factored into linear terms in k[x]. Hence T W is triangularizable, by Theorem 1. It follows that T ∈ T , and hence T = Endk(V ). Using the function topology and Theorem 8 we can generalize the standard fact that a matrix is nilpotent if and only if it is similar to a strictly upper-triangular matrix if and only if 0 is its only eigenvalue (over the algebraic closure of the base field). Proposition 20. Let k be a field and V a nonzero k-vector space. The following are equiv- alent for any T ∈ Endk(V ). (1) T is topologically nilpotent with respect to the function topology on Endk(V ). That is, the sequence (T i)∞ i=1 converges to 0. (2) V = S∞ i=1 ker(T i). (3) T is strictly triangularizable. (4) T is triangularizable, and if (B, ≤) is a well-ordered basis for V with respect to which T is triangular, then T is strictly triangular with respect to (B, ≤). (5) T is triangularizable, and ker(T − aI) 6= 0 if and only if a = 0, for all a ∈ k. Proof. We shall show that (1) ⇔ (2) ⇒ (4) ⇒ (3) ⇒ (5) ⇒ (2). (1) ⇔ (2) T is topologically nilpotent if and only if for every open neighborhood U of 0 there exists n ∈ Z+ such that T n ∈ U. By our description of the function topology, this is equivalent to: for every finite-dimensional subspace W of V there exists n ∈ Z+ such that T n(W ) = 0. That statement is clearly equivalent to V = S∞ (2) ⇒ (4) If T satisfies (2), then it is triangularizable, by Lemma 7. Now let (B, ≤) be a well-ordered basis for V with respect to which T is triangular, and let v ∈ B. Write i=1 ker(T i). 17 T (v) = av +Pu<v auu for some u ∈ B and a, au ∈ k, and suppose that a 6= 0. Then for all n ∈ Z+ we have T n(v) = anv + w for some w ∈ h{u ∈ B u < v}i, and hence T n(v) 6= 0, producing a contradiction. Therefore a = 0, and hence T (v) ∈ h{u ∈ B u < v}i for all v ∈ B. That is, T is strictly triangular with respect to (B, ≤). (4) ⇒ (3) This is a tautology. (3) ⇒ (5) Suppose that T is strictly triangular with respect to a well-ordered basis (B, ≤) for V . Then T is triangularizable, by definition. Now, let v ∈ V , and write v = auu + Pw<u aww for some u, w ∈ B and au, aw ∈ k. Then T being strictly triangular with respect to (B, ≤) implies that the coefficient of u in T (v), when expressed as a linear combination of elements of B, is zero. Therefore, given a ∈ k, we can have T (v) = av only if a = 0. That is, a = 0 whenever ker(T − aI) 6= 0. On the other hand, since T is triangularizable and V 6= 0, we have ker(T − aI) 6= 0 for some a ∈ k, by Lemma 13, from which (5) follows. (5) ⇒ (2) Suppose that T satisfies (5), and let v ∈ V . By Theorem 8, there is a polynomial p(x) ∈ k[x] \ k that factors into linear terms in k[x], such that v ∈ ker(p(T )). By Lemma 2 and (5), this means that p(x) can be taken to be xn for some n ∈ Z+, and therefore T n(v) = 0. It follows that V = S∞ i=1 ker(T i). 7 Transformations Satisfying a Polynomial As we saw in Theorem 8, every triangularizable transformation T satisfies a polynomial on each finite-dimensional T -invariant subspace. It is therefore natural to ask whether more can be said about transformations that satisfy a single polynomial on the entire space. That indeed can be quickly accomplished with the help of the following classical result from [4]. (See also [1, Corollary 3.3] for a noncommutative generalization.) Theorem 21 (Kothe). Let R be a commutative artinian ring. Then every R-module is a direct sum of cyclic R-modules if and only if R is a principal ideal ring. Applying Kothe's theorem to the linear algebra setting yields the following extension of the rational canonical form to transformations of an arbitrary vector space that satisfy a polynomial. This was also observed by Radjabalipour in [6, Theorem 1.5], using a more elementary approach. Corollary 22. Let k be a field, V a k-vector space, T ∈ Endk(V ), and p(x) ∈ k[x] \ k such that p(T ) = 0. Then V = Mλ∈Λ h{vλ, T (vλ), . . . , T n−1(vλ)}i for some vλ ∈ V , where n is the degree of p(x). Proof. Since k[x] is a principal ideal domain, R = k[x]/(p(x)) is a (commutative) principal ideal ring (as its ideals correspond to the ideals of k[x] containing (p(x))). Moreover, since R is finite-dimensional as a k-vector space (being spanned by {1, x, x2, . . . , xn−1}), it is also artinian. Hence, by Theorem 21, every R-module is a direct sum of cyclic R-modules. Now, viewing V as an R-module, by letting x act as T , we see that V = Lλ∈Λ Rvλ for some vλ ∈ V , from which the desired conclusion follows. 18 The next definition will help us apply Corollary 22 to triangularizable transformations, and thereby extend the Jordan canonical form to transformations of an arbitrary vector space that satisfy a polynomial. Definition 23. Let k be a field, V a finite-dimensional k-vector space, and T ∈ Endk(V ). If there is a basis {v0, v1, . . . , vn} for V such that T (vi) = vi−1 for all 1 ≤ i ≤ n and T (v0) = 0, then we say that T acts as a left shift transformation on V . Corollary 24. Let k be a field, V a k-vector space, p(x) ∈ k[x] \ k a polynomial that factors into linear terms in k[x], and T ∈ Endk(V ) such that p(T ) = 0. Then there are finite- dimensional subspaces Vλ ⊆ V and aλ ∈ k (λ ∈ Λ), such that V = Lλ∈Λ Vλ and T − aλI acts as a left shift transformation on Vλ, for each λ ∈ Λ. Proof. By Corollary 22, V can be written as a direct sum of finite-dimensional T -invariant subspaces. The desired conclusion now follows from applying Theorem 1 to each of these subspaces. 8 Double-Centralizer We conclude the paper by generalizing the following result for (finite) matrices to trans- formations of vector spaces of arbitrary dimension. See, e.g., [8, Chapter 1, Theorem 7] or [5, Theorem 1] for proofs of this result. Theorem 25 (Classical Double-Centralizer Theorem). Let k be a field, n ∈ Z+, Mn(k) the k-algebra of all n × n matrices over k, and T ∈ Mn(k). Then C(C(T )) = k[T ]. We require a couple of standard lemmas. Lemma 26. If R is a Hausdorff topological ring, then the centralizer of any subset of R is closed in R. Proof. Let X be a subset of R, and let C(X) denote the closure of C(X) in R. Suppose that C(X) 6= C(X). Then there must be some r ∈ C(X) \ C(X), and hence rx − xr 6= 0 for some x ∈ X. Since the topology is Hausdorff, there must be an open neighborhood U of rx − xr such that 0 /∈ U. By the continuity of the operations, we can find an open neighborhood V of r such that Vx − xV ⊆ U. Since C(X) is dense in C(X), there is some r′ ∈ C(X) such that r′ ∈ V. But then 0 = r′x − xr′ ∈ Vx − xV ⊆ U, contradicting 0 /∈ U. Thus C(X) = C(X), i.e., C(X) is closed. Lemma 27. Let k be a field, let V = W ⊕ U be k-vector spaces, and let T ∈ Endk(V ). If W and U are T -invariant, then W and U are also invariant under every element of C(C(T )). Proof. Let π ∈ Endk(V ) be the projection of V onto W with kernel U, and let S ∈ C(C(T )). Since W and U are T -invariant, we have π ∈ C(T ), and hence Sπ = πS. Thus S(W ) = Sπ(W ) = πS(W ) ⊆ π(V ) = W, and similarly S(U) ⊆ U. 19 Proposition 28. Let k be a field, V a k-vector space, and T ∈ Endk(V ). Suppose that there are finite-dimensional T -invariant subspaces Vλ ⊆ V (λ ∈ Λ) such that V = Lλ∈Λ Vλ. Then C(C(T )) = k[T ]. Proof. Clearly k[T ] ⊆ C(C(T )). Since, by Lemma 26, C(C(T )) is closed, it follows that k[T ] ⊆ C(C(T )). For the opposite inclusion, let S ∈ C(C(T )), and let U be an open neighborhood of S. Passing to a subset, if necessary, we may assume that U = {F ∈ Endk(V ) F U = SU } for some finite-dimensional subspace U of V . We can find some λ1, . . . , λm ∈ Λ such that U ⊆ Vλ1 ⊕ · · · ⊕ Vλm. Letting W = Vλ1 ⊕ · · · ⊕ Vλm, we have S(W ) ⊆ W , by Lemma 27. Let H ′ ∈ Endk(W ) be such that H ′T W = T W H ′. Extending H ′ to a map H ∈ Endk(V ) by letting H(Lλ∈Λ\{λ1,...,λm} Vλ) = 0, we see that HT = T H, and hence also SH = HS. Since S(W ) ⊆ W , we have SW H ′ = H ′SW . Since H ′ ∈ C(T W ) was arbitrary, this shows that SW ∈ C(C(T W )). Thus, by Theorem 25, we have SW ∈ k[T W ]. Since W is T - invariant, this implies that there is some polynomial p(x) ∈ k[x] such that SW = p(T )W . Therefore p(T ) ∈ U, and hence S is a limit point of k[T ]. It follows that S ∈ k[T ], and thus C(C(T )) ⊆ k[T ]. With the help of the next lemma, we can give another generalization of Theorem 25 to infinite-dimensional vector spaces. Lemma 29. Let k be a field, V a k-vector space, p(x) ∈ k[x] \ k, and T ∈ Endk(V ) such that p(T ) = 0. Then k[T ] = k[T ]. Proof. Clearly k[T ] ⊆ k[T ]. To show the opposite inclusion, let us take S ∈ k[T ], and prove that S ∈ k[T ]. By the properties of the function topology, for each finite-dimensional subspace W ⊆ V there exists q(x) ∈ k[x] such that SW = q(T )W . For each W let qW (x) ∈ k[x] be such a polynomial of least degree. Since p(T ) = 0, we have deg(qW ) < deg(p) for each W , by the division algorithm. Thus, we can find a finite-dimensional subspace U ⊆ V such that deg(qU ) ≥ deg(qW ) for all W . Now let W ⊆ V be any finite-dimensional subspace. Then qU +W (T )U = SU = qU (T )U , and hence deg(qU +W ) = deg(qU ), by our definition of the qW and choice of U. Thus, q(x) = qU (x) − qU +W (x) is a polynomial of degree at most deg(qU ) such that q(T )U = 0. If q(x) were nonzero, then this would imply, upon applying the division algorithm to qU (x) and q(x), that there is a polynomial q′(x) ∈ k[x] such that deg(q′) < deg(q) ≤ deg(qU ) and SU = q′(T )U , contradicting the minimality of the degree of qU . Therefore q(x) = 0, and hence qU +W (x) = qU (x), which implies that SW = qU +W (T )W = qU (T )W . Since W was arbitrary, this means that S = qU (T ), and hence S ∈ k[T ]. Theorem 30. Let k be a field, V a k-vector space, and T ∈ Endk(V ). p(x) ∈ k[x] \ k such that p(T ) = 0, then C(C(T )) = k[T ] = k[T ]. If there exists Proof. By Corollary 22, if T satisfies the above condition, then it also satisfies the hypotheses of Proposition 28, and hence C(C(T )) = k[T ]. The desired conclusion now follows from Lemma 29. 20 The next example shows that the conclusion of Theorem 30 does not hold for chain- triangularizable transformations. Example 31. Let k be a field and V a k-vector space with basis {vi i ∈ Z}. Define T ∈ Endk(V ) by T (vi) = vi−1 for each i ∈ Z, and extend linearly to all of V . As seen in Example 9, T is chain-triangularizable but not triangularizable. We shall show that C(C(T )) = C(T ) = k[T, T −1] = k[T, T −1] 6= k[T ]. Since T is clearly invertible, k[T, T −1] ⊆ C(T ), and since, by Lemma 26, C(T ) is closed, i=−n aivi we have k[T, T −1] ⊆ C(T ). Now let S ∈ C(T ) be any element, and write S(v0) = Pm for some n, m ∈ N and ai ∈ k. Then for every l ∈ Z, we have S(vl) = ST −l(v0) = T −lS(v0) = m Xi=−n aiT −l(vi) = m Xi=−n aivi+l = m Xi=−n aiT −i(vl), and therefore S = Pm k[T, T −1] ⊆ C(T ), we conclude that i=−n aiT −i ∈ k[T, T −1]. Thus C(T ) ⊆ k[T, T −1]. Combining this with C(T ) = k[T, T −1] = k[T, T −1]. Since C(C(T )) is the center of the ring C(T ), and C(T ) = k[T, T −1] is commutative, we also have C(T ) = C(C(T )). It remains to show that T −1 /∈ k[T ], from which we can conclude that k[T, T −1] 6= k[T ]. Suppose, on the contrary, that T −1 ∈ k[T ]. Then there must be some S ∈ k[T ] such that i=0 aiT i for some m ∈ N and ai ∈ k, and hence S(v0) = v1 = T −1(v0). But S = Pm Xi=0 S(v0) = m aiT i(v0) = m Xi=0 aiv−i 6= v1, producing a contradiction. (cid:3) The previous theorem and example leave us with the following question. Question 32. Let k be a field, V a k-vector space, and T ∈ Endk(V ) a triangularizable trans- formation (or, more generally, a transformation such that every finite-dimensional subspace of V is annihilated by p(T ) for some p(x) ∈ k[x]). Is it the case that C(C(T )) = k[T ]? Acknowledgements I am grateful to George Bergman for his numerous comments on an earlier version of this paper, which have led to significant improvements, including the addition of condition (5) to Theorem 8. I also would like to thank the referee for a very thoughtful review, and particularly for the suggestion to add condition (7) to Theorem 8. Finally, I would like to thank Greg Oman for helpful conversations about this material. 21 References [1] M. Behboodi, A. Ghorbani, A. Moradzadeh-Dehkordi, and S. H. Shojaee, On Left Kothe Rings and a Generalization of the Kothe-Cohen-Kaplansky Theorem, Proc. Amer. Math. Soc. 142 (2014) 2625–2631. [2] D. S. Dummit and R. M. Foote, Abstract Algebra, 3rd Edition, John Wiley and Sons, Inc., Hoboken, New Jersey, 2004. [3] M. C. Iovanov, Z. Mesyan, and M. L. Reyes, Infinite-Dimensional Diagonalization and Semisimplicity, Israel J. Math. 215 (2016) 801–855. [4] G. Kothe, Verallgemeinerte Abelsche Gruppen mit hyperkomplexem Operatorenring, Math. Z. 39 (1935) 31–44. [5] P. Lagerstrom, A Proof of a Theorem on Commutative Matrices, Bull. Amer. Math. Soc. 51 (1945) 535–536. [6] M. Radjabalipour, Infinite-Dimensional Versions of the Primary, Cyclic and Jordan Decompositions, Bull. Iranian Math. Soc. 41 (2015) 175–183. [7] H. Radjavi and P. Rosenthal, Simultaneous Diagonalization, Springer-Verlag, New York, 2000. [8] D. A. Suprunenko and R. I. Tyshkevich, Commutative Matrices, Academic Press, New York, 1968. [9] S. Warner, Topological Rings, North-Holland Mathematical Studies, 178, North- Holland, Amsterdam, 1993. Department of Mathematics, University of Colorado, Colorado Springs, CO, 80918, USA [email protected] 22
1002.3280	1	1002	2010-02-17T14:39:03	Wild Pfister forms over Henselian fields, K-theory, and conic division algebras	[ "math.RA" ]	The epicenter of this paper concerns Pfister quadratic forms over a field $F$ with a Henselian discrete valuation. All characteristics are considered but we focus on the most complicated case where the residue field has characteristic 2 but $F$ does not. We also prove results about round quadratic forms, composition algebras, generalizations of composition algebras we call conic algebras, and central simple associative symbol algebras. Finally we give relationships between these objects and Kato's filtration on the Milnor $K$-groups of $F$.	math.RA	math	WILD PFISTER FORMS OVER HENSELIAN FIELDS, K-THEORY, AND CONIC DIVISION ALGEBRAS SKIP GARIBALDI AND HOLGER P. PETERSSON Abstract. The epicenter of this paper concerns Pfister quadratic forms over a field F with a Henselian discrete valuation. All characteristics are consid- ered but we focus on the most complicated case where the residue field has characteristic 2 but F does not. We also prove results about round quadratic forms, composition algebras, generalizations of composition algebras we call conic algebras, and central simple associative symbol algebras. Finally we give relationships between these objects and Kato's filtration on the Milnor K-groups of F . Introduction The theory of quadratic forms over a field F with a Henselian discrete valuation is well understood in case the residue field F has characteristic different from 2 thanks to Springer [46]. But when F is imperfect of characteristic 2, the theorems are much more complicated -- see, e.g., [50] and [22] -- reflecting perhaps the well- known fact that quadratic forms are not determined by valuation-theoretic data, as illustrated below in Example 11.14. However, more can be said when one focuses on Pfister quadratic forms over F and more generally round forms, see Part II below. In Part III we change our focus to Kato's filtration on the mod-p Milnor K- theory of a Henselian discretely valued field of characteristic zero where the residue field has characteristic p. (Again, if F has characteristic different from p, the mod-p Galois cohomology and Milnor K-theory of F are easily described in terms of F , see [18, pp. 17 -- 19] and [19, 7.1.10].) We give translations between valuation-theoretic properties of Pfister forms, octonion algebras, central simple associative algebras of prime degree (really, symbol algebras), and cyclic field extensions of prime degree over F on the one hand and properties of the corresponding symbols in Milnor K-theory on the other. Along the way, we prove some results that are of independent interest, which we now highlight. Part I treats quadratic forms and composition algebras over arbitrary fields. It includes a Skolem-Noether Theorem for purely inseparable sub- fields of composition algebras (Th. 5.7) and a result on factoring quadratic forms (Prop. 3.12). We also give a new family of examples of what we call conic division algebras, which are roughly speaking division algebras where every element satisfies a polynomial of degree 2, see Example 6.6. More precisely, we show that -- contrary to what is known, e.g., over the reals -- a Pfister quadratic form of characteristic 2 is anisotropic if and only if it is the norm of such a division algebra (Cor. 6.5). 2000 Mathematics Subject Classification. Primary 17A75; secondary 11E04, 16W60, 17A45, 19D45, 19F15. 1 2 SKIP GARIBALDI AND HOLGER P. PETERSSON Part II focuses on round quadratic forms and composition algebras over a field F with a 2-Henselian discrete valuation with residue field of characteristic 2. From this part, our Local Norm Theorem 8.10 has already been applied in [56]. We also relate Tignol's height ω from [51] with Saltman's level hgtcom from [45], with a nonassociative version of Saltman's level that we denote by hgtass, and with valuation-theoretic properties of composition algebras, see Th. 12.11 and Cor. 19.3. We show that composition division algebras over F having pre-assigned valuation data, subject to a few obvious constraints, always exist (Cor. 11.13), though they are far from unique up to isomorphism (Example 11.14). Part III gives a K-theoretic proof of the Local Norm Theorem (Th. 15.2); it has an easy proof but stronger hypotheses than the version in Part II. Finally, our Gathering Lemma 16.1 is independent of the rest of the paper and says that one may rewrite symbols in a convenient form. Let us now discuss what happens "under the hood". The basic technical result in Part I is a non-orthogonal analogue of the classical Cayley-Dickson construction for algebras of degree 2. It is used to prove the Skolem-Noether Theorem mentioned above as well as to construct the examples of conic division algebras. In Part II, we proceed to a more arithmetic set-up by considering a base field F , discretely valued by a normalized discrete valuation λ : F → Z ∪ {∞}, which is 2-Henselian in the sense that it satisfies Hensel's lemma for quadratic polynomials. Our main goal is to understand composition algebras and Pfister quadratic forms over F . For this purpose, the results of [41] -- where the base field was assumed to be complete rather than Henselian -- carry over to this more general (and also more natural) setting virtually unchanged; we use them here here without further ado. Moreover, we will be mostly concerned with "wild" composition algebras over F (see 7.15 below for the precise definition of this term in a more general context) since a complete description of the "tame" ones in terms of data living over the residue field of F has been given in [41]. (Alternatively -- and from a different perspective -- one has a good description of the tame part of the Witt group of F from [16].) The approach adopted here owes much to the work of Kato [23, § 1], Saltman [45] and particularly Tignol [51] on wild associative division algebras of degree the residual characteristic p > 0 of their (possibly non-discrete) Henselian base field. Moreover, our approach is not confined to composition algebras but, at least to a certain extent, works more generally for (non-singular) pointed quadratic spaces that are round (e.g., Pfister) and anisotropic. We attach valuation data to these spaces, among which not so much the usual ones (ramification index (7.9 (b)) and pointed quadratic residue space (7.9 (c))), but wildness-detecting invariants like the trace exponent (8.1) play a significant role. After imitating the quadratic defect [35, 63A] for round and anisotropic pointed quadratic spaces (8.8) and extending the local square theorem [35, 63:1] to this more general setting (Thm. 8.10), we proceed to investigate the behavior of our valuation data when passing from a wild, round and anisotropic pointed quadratic space P having ramification index 1 as input to the output Q := hhµii ⊗ P , for any non-zero scalar µ ∈ F (Section 9). In all cases except one, the output, assuming it is anisotropic, will again be a wild pointed quadratic space. Remarkably, the description of the exceptional case (Thm. 9.9), where the input is assumed to be Pfister and the output turns out to be tame, when specialized to composition algebras, relies critically on the non- orthogonal Cayley-Dickson construction encountered in the first part of the paper WILD PFISTER FORMS 3 (Cor. 10.18). This connection is due to the fact that our approach also lends itself to the study of what we call λ-normed and λ-valued conic algebras (Section 10), the latter forming a class of conic division algebras over F that generalize ordinary composition algebras and turn out to exist in all dimensions 2n, n = 0, 1, 2, . . . , once F has been chosen appropriately (Examples 10.7,10.15). There is yet another unusual feature of the exceptional case: though exclusively belonging to the theory of quadratic forms (albeit in an arithmetic setting), it can be resolved here only by appealing to elementary properties of flexible conic algebras (Thm. 10.17). The second part of the paper concludes with extending Tignol's notion of height [51], which agrees with Saltman's notion of level [45], to composition division algebras over F and relating them to the valuation data introduced before (Thm. 12.11). In the third part of the paper, we consider the case where F has characteristic zero and a primitive p-th root of unity and has a Henselian discrete valuation with residue field of characteristic p. In that setting, Kato, Bloch, and Gabber gave a description of the mod-p Milnor K-groups kq(F ) in [23, 24, 25, 4]; we use [10] as a convenient reference. We relate properties of a symbol in kq(F ) with valuation- theoretic properties of the corresponding algebra, see Prop. 19.1 and Th. 19.2. Introduction Contents Part I. Base fields of characteristic 2 1. Standard properties of conic algebras 2. Flexible and alternative conic algebras. 3. 4. A non-orthogonal Cayley-Dickson construction 5. The Skolem-Noether theorem for inseparable subfields 6. Conic division algebras in characteristic 2. Interlude: Pfister bilinear and quadratic forms in characteristic 2 Part II. 2-Henselian base fields 7. Pointed quadratic spaces over 2-Henselian fields. 8. Trace and norm exponent. 9. Valuation data under enlargements. 10. λ-normed and λ-valued conic algebras. 11. Applications to composition algebras. 12. Types of composition algebras and heights. Part III. Connections with K-theory Introduction to Part III 13. 14. The filtration on K-theory 15. The Local Norm Theorem 8.10 revisited 16. Gathering the depth 17. Ramification index for symbols 18. Theorem 9.9 revisited 19. Dictionary between K-theory and algebras and quadratic forms 20. Proof of Proposition 19.1: case q = 1 21. Proof of Proposition 19.1: cases q = 2 22. Proofs of Theorem 19.2 and Corollary 19.3 1 4 4 8 12 16 20 25 27 27 33 38 45 52 59 66 66 67 68 70 72 73 74 75 77 78 4 SKIP GARIBALDI AND HOLGER P. PETERSSON References 79 Part I. Base fields of characteristic 2 1. Standard properties of conic algebras Although composition algebras are our main concern in this paper, quite a few of our results remain valid under far less restrictive conditions. The appropriate framework for some of these conditions is provided by the category of conic algebras. They are the subject of the present section. We begin by fixing some terminological and notational conventions about non- associative algebras in general and about quadratic forms. For the time being, we let k be a field of arbitrary characteristic. Only later on (Sections 3−6) will we confine ourselves to base fields of characteristic 2. 1.1. Algebras. Non-associative (= not necessarily associative) algebras play a dominant role in the present investigation. For brevity, they will often be referred to simply as algebras (over k) or as k-algebras. A good reference for the standard vocabulary is [57]. Left and right multiplication of a k-algebra A will be denoted by x 7→ Lx and x 7→ Rx, respectively. A is called unital if it has an identity (or unit) element, denoted by 1A. A subalgebra of A is called unital if it contains the identity element of A. Algebra homomorphisms are called unital if they preserve identity elements. Commutator and associator of A will be denoted by [x, y] = xy − yx and [x, y, z] = (xy)z − x(yz), respectively. If A is unital, then Nuc(A) :=(cid:8)x ∈ A [A, A, x] = [A, x, A] = [x, A, A] = {0}(cid:9) is a unital associative subalgebra of A, called its nucleus, and Cent(A) :=(cid:8)x ∈ Nuc(A) [A, x] = {0}(cid:9) is a unital commutative associative subalgebra of A, called its centre. We say A is central (resp. has trivial nucleus) if Cent(A) = k1A (resp. Nuc(A) = k1A). A k-algebra A is called flexible if it satisfies the flexible law (1) xyx := (xy)x = x(yx). A is said to be alternative if the associator is an alternating (trilinear) function of its arguments. This means that A is flexible and satisfies the left and right alternative laws (2) x(xy) = x2y, (yx)x = yx2. Furthermore, the left, middle and right Moufang identities (3) hold. x(cid:0)y(xz)(cid:1) = (xyx)z, x(yz)x = (xy)(zx), (cid:0)(zx)y(cid:1)x = z(xyx) WILD PFISTER FORMS 5 1.2. Quadratic forms. Our main reference in this paper for the algebraic theory of quadratic forms is [15], although our notation will occasionally be different and we sometimes work in infinite dimensions. Let V be vector space over k, possibly infinite-dimensional. Deviating from the notation used in [15], we write the polar form of a quadratic form q : V → k, also called the bilinear form associated with q or its bilinearization, as ∂q, so ∂q : V × V → k is the symmetric bilinear form given by ∂q(x, y) := q(x + y) − q(x) − q(y) (x, y ∈ V ). Most of the time we simplify notation and write q(x, y) := ∂q(x, y) if there is no danger of confusion. The quadratic form q is said to be non-singular if it has finite dimension and its polar form is non-degenerate in the usual sense, i.e., for any x ∈ V , the relations ∂q(x, y) = 0 for all y ∈ V imply x = 0. Recall that non- singular quadratic forms have even dimension if the characteristic is 2 [15, Chap. 2, Remark 7.22]. 1.3. Conic algebras. We consider a class of non-associative algebras that most authors refer to as quadratic [57] or algebras of degree 2 [33]. In order to avoid confusion with Bourbaki's notion of a quadratic algebra [5], we adopt a different terminology. A k-algebra C is said to be conic if it has an identity element 1C 6= 0 and there exists a quadratic form n : C → k with x2 − t(x)x + n(x)1C = 0 for all x ∈ C, where t is defined by t := ∂n(1C,−) : C → k and hence is a linear form. The quadratic form n is uniquely determined by these conditions and is called the norm of C, written as nC . We call tC := t = ∂nC(1C ,−) the trace of C and have (1) x2 − tC(x)x + nC (x)1C = 0, nC (1C ) = 1, tC (1C ) = 2 (x ∈ C). Finally, the linear map (2) ιC : C −→ C, x 7−→ ιC (x) := x∗ := tC(x)1C − x, called the conjugation of C, has period 2 and is characterized by the condition (3) 1∗ C = 1C , xx∗ = nC (x)1C (x ∈ C). The property of an algebra to be conic is inherited by unital subalgebras. Injective unital homomorphisms of conic algebras are automatically norm preserving. A conic algebra C over k is said to be non-degenerate if the polar form ∂nC has this property. Thus finite-dimensional conic algebras are non-degenerate iff their norms are non-singular as quadratic forms. Orthogonal complementation in C always refers to ∂nC. We say C is simple as an algebra with conjugation if only the trivial (two-sided) ideals I ⊆ C satisfy I ∗ = I. 1.4. Invertibility in conic algebras. Let C be a conic algebra over k. By (1.3.1), the unital subalgebra of C generated by an element a ∈ C, written as k[a], is commutative associative and spanned by 1C , a as a vector space over k; in particular it has dimension at most 2. We say a is invertible in C if this is so in k[a], i.e., if there exists an element a−1 ∈ k[a] (necessarily unique and called the inverse of a in C) such that aa−1 = 1C . For a to be invertible in C it is necessary and sufficient that nC(a) 6= 0, in which case a−1 = nC(a)−1a∗. The set of invertible elements in C will always be denoted by C×. As usual, a non-associative k-algebra A is called a division algebra if for all a, b ∈ A, a 6= 0, the equations ax = b, ya = b can be solved uniquely in A. The 6 SKIP GARIBALDI AND HOLGER P. PETERSSON quest for conic division algebras is an important topic in the present investigation. The following necessary criterion, though trivial, turns out to be useful. 1.5. Proposition. The norm of a conic division algebra is anisotropic. (cid:3) The converse of this proposition does not hold (cf. 6.1). For char(k) 6= 2, conditions that are necessary and sufficient for a conic algebra to be division have been given by Osborn [36, Thm. 3]. 1.6. Inseparable field extensions. Exotic examples of conic algebras arise in connection with inseparability. Suppose k has characteristic 2 and let K/k be a purely inseparable field extension of exponent at most 1, so K 2 ⊆ k. Then K is a conic k-algebra with nK(u) = u2 for all u ∈ K, ∂nK = 0, tK = 0 and ιK = 1K. In particular, inseparable field extensions of exponent at most 1 over k are degenerate, hence singular, conic division algebras. 1.7. Composition algebras. Composition algebras form the most important class of conic algebras. Convenient references, including base fields of characteristic 2, are [29, 47], although [29] introduces a slightly more general notion. An algebra C over k is said to be a composition algebra if it is non-zero, contains a unit element and carries a non-singular quadratic form n : C → k that permits composition: n(xy) = n(x)n(y) for all x, y ∈ C. Composition algebras are automatically conic. In fact, the only quadratic form on C permitting composition is the norm of C in its capacity as a conic algebra. 1.8. Basic properties of composition algebras. Composition algebras exist only in dimensions 1, 2, 4, 8 and are alternative. They are associative iff their di- mension is at most 4, and commutative iff their dimension is at most 2. The base field is a composition algebra if and only if it has characteristic different from 2. Composition algebras of dimension 2 are the same as quadratic ´etale algebras. Composition algebras of dimension 4 (resp. 8) are called quaternion (resp. octonion or Cayley) algebras. Two composition algebras are isomorphic if and only if their norms are isometric (as quadratic forms). The conjugation of a composition algebra C over k is an algebra involution. What is denied to arbitrary conic algebras holds true for composition algebras: 1.9. Norm criterion for division algebras. A composition algebra C over k is a division algebra if and only if its norm is anisotropic. Otherwise its norm is hyperbolic, in which case we say C is split. Up to isomorphism, split composi- tion algebras are uniquely determined by their dimension, and their structure is explicitly known. 1.10. The Cayley-Dickson construction. The main tool for dealing with conic algebras in general and composition algebras in particular is the Cayley-Dickson construction. Its inputs are a conic algebra B and a non-zero scalar µ ∈ k. Its output is a conic algebra C := Cay(B, µ) that is given on the vector space direct sum C = B ⊕ Bj of two copies of B by the multiplication (1) (u1 + v1j)(u2 + v2j) := (u1u2 + µv∗ 2v1) + (v2u1 + v1u∗ 2)j (ui, vi ∈ B, i = 1, 2). WILD PFISTER FORMS 7 Norm, polarized norm, trace and conjugation of C are related to the corresponding data of B by the formulas (2) (3) (4) nC (u + vj) = nB(u) − µnB(v), nC (u1 + v1j, u2 + v2j) = nB(u1, u2) − µnB(v1, v2), tC (u + vj) = tB(u), (u + vj)∗ = u∗ − vj (5) for all u, v, ui, vi ∈ B, i = 1, 2. Note that B embeds into C as a unital conic subalgebra through the first summand; we always identify B ⊆ C accordingly. The Cayley-Dickson construction Cay(B, µ) is clearly functorial in B, under injective unital homomorphisms. It is a basic fact that C is a composition algebra iff B is an associative compo- sition algebra. Conversely, we have the following embedding property, which fails for arbitrary conic algebras, cf. Example 10.8 below. 1.11. Embedding property. Any proper composition subalgebra B of a compo- sition algebra C over k is associative and admits a scalar µ ∈ k× such that the inclusion B ֒→ C extends to an embedding Cay(B, µ) → C of conic algebras. More precisely, µ ∈ k× satisfies this condition iff µ = −nC (y) for some y ∈ B⊥ ∩ C×. 1.12. The Cayley-Dickson process. Let B be a conic k-algebra. Using non-zero scalars µ1, . . . , µn ∈ k× (n ≥ 1), we write inductively C := Cay(B; µ1, . . . , µn) := Cay(cid:0)Cay(B; µ1, . . . , µn−1), µn(cid:1) for the corresponding iterated Cayley-Dickson construction starting from B. It is a conic k-algebra of dimension 2ndimk(B). We say C arises from B and the µ1, . . . , µn by means of the Cayley-Dickson process. The norm of C is given by (1) nC = hhµ1, . . . , µnii ⊗ nB. Here are the most important special cases of the Cayley-Dickson process. Case 1. B = k, char(k) 6= 2. Then nC = hhµ1, . . . , µnii is an n-Pfister quadratic form. C is a composition algebra iff n ≤ 3. Case 2. B = k, char(k) = 2. Then nC = hhµ1, . . . , µniiq is a quasi-Pfister (quadratic) form [15, § 10, p. 56]. Moreover, nC is anisotropic iff C = k(√µ1, . . . ,√µn) is an extension field of k, necessarily purely inseparable of exponent 1, hence never a composition algebra. Case 3. B is a quadratic ´etale k-algebra. Then nB = hhµK for some µ ∈ k [15, Example 9.4] and (1) shows that nC = hhµ1, . . . , µn, µK is an (n + 1)-Pfister quadratic form over k. Moreover, C is a composition algebra iff n ≤ 2. Composition algebras other than the base field itself always contain quadratic ´etale subalgebras. Hence, by Cases 1, 3 above and by the embedding property 1.11, they may all be obtained from each one of these, even from the base field itself if the characteristic is not 2, by the Cayley-Dickson process. The preceding discussion 8 SKIP GARIBALDI AND HOLGER P. PETERSSON also shows that all Pfister and all quasi-Pfister quadratic forms are the norms of appropriate conic algebras. 1.13. Inseparable subfields. Let C be a composition division algebra over k. A unital subalgebra of C is either a composition (division) algebra itself or an inseparable extension field of k; in the latter case, k has characteristic 2 and the extension is purely inseparable of exponent at most 1 [53]. The extent to which this case actually occurs may be described somewhat more generally as follows. Suppose k has characteristic 2, B is a conic k-algebra and µ1, . . . , µn ∈ k× are such that the norm of (1) C := Cay(B; µ1, . . . , µn) is anisotropic, so all non-zero elements of C are invertible (1.4). Then, by Case 2 of 1.12, K := Cay(k; µ1, . . . , µn) ⊆ C is a purely inseparable subfield of degree 2n and exponent 1. Specializing this observation to n = 2 and B quadratic ´etale over k, we conclude that every octonion division algebra over a field of characteristic 2 contains an inseparable subfield of degree 4. 2. Flexible and alternative conic algebras. This section is devoted to some elementary properties of flexible and alterna- tive conic algebras. In particular, we derive expansion formulas for the norm of commutators and associators that turn out to be especially useful in subsequent applications. Phrased with appropriate care, most of the results obtained here remain valid over any commutative associative ring of scalars. For simplicity, however, we con- tinue to work over a field k of arbitrary characteristic. We fix a conic algebra C over k and occasionally adopt the abbreviations 1 = 1C , n = nC , t = tC . 2.1. Identities in arbitrary conic algebras. The following identities, some of which have been recorded before, are assembled here for the convenience of the reader and either hold by definition or are straightforward to check. (1) (2) (3) (4) (5) (6) (7) (8) nC (1C) = 1C, tC (1C) = 2, tC(x) = nC (1C, x), x2 = tC (x)x − nC(x)1C , x ◦ y := xy + yx = tC (x)y + tC (y)x − nC(x, y)1C , x∗ = tC (x)1C − x, xx∗ = nC (x)1C , nC(x∗) = nC (x). WILD PFISTER FORMS 9 2.2. Identities in flexible conic algebras. We now assume that C is flexible. By McCrimmon [33, 3.4,Thm. 3.5], this implies the following relations: nC(xy, x) = nC (x)tC (y) = nC (yx, x), nC (x, zy∗) = nC (xy, z) = nC (y, x∗z), tC (xy) = tC(yx), nC (x, y) = tC (xy∗) = tC (x)tC (y) − tC(xy), tC (xyz) := tC(cid:0)(xy)z(cid:1) = tC(cid:0)x(yz)(cid:1). (1) (2) (3) (4) Moreover, the conjugation is an algebra involution of C, so we have (xy)∗ = y∗x∗ for all x, y ∈ C. Dealing with flexible conic algebras has the additional advantage that this property is preserved under the Cayley-Dickson construction [33, Thm. 6.8]. 2.3. Remark. By [33, 3.4], each one of the four(!) identities in (2.2.1),(2.2.2) is actually equivalent to C being flexible. The norm of a flexible conic algebra will in general not permit composition. But we have at least the following result. 2.4. Proposition. Let C be a flexible conic algebra over k. Then (1) nC (xy) = nC(yx), nC ([x, y]) = 4nC(xy) − tC (x)2nC (y) − tC (y)2nC(x) + tC (xy)tC (xy∗) (2) for all x, y ∈ C. Proof. Expanding the expression n(x ◦ y) by means of (2.1.5),(2.1.3) yields n(x ◦ y) = t(x)2n(y) + t(y)2n(x) + n(x, y)2 − t(x)t(y)n(x, y), where flexibility allows us to invoke (2.2.3); we obtain n(x ◦ y) = t(x)2n(y) + t(y)2n(x) − t(xy)t(xy∗). (3) Now let ε = ±1. Then n(xy + εyx) = n(xy) + εn(xy, yx) + n(yx) = n(xy) + (1 − 2ε)n(yx) + εn(x ◦ y, yx), and combining (2.1.5) with (2.2.1)(2.2.3),(2.2.4), we conclude (4) n(xy + εyx) = n(xy) + (1 − 2ε)n(yx)+ ε(cid:0)t(x)2n(y) + t(y)2n(x) − t(xy)t(xy∗)(cid:1). Comparing (3) and (4) for ε = 1 yields (1), while (1) and (4) for ε = −1 yield (2). (cid:3) 2.5. Proposition. Let C be a non-degenerate conic algebra over k. (a) C is simple as an algebra with conjugation (cf. 1.3). (b) If ιC is an algebra involution of C, in particular, if C is flexible, then C is either simple or split quadratic ´etale. Proof. (a) Let I ⊆ C be an ideal with I ∗ = I. For x ∈ I, y ∈ C we linearize (2.1.7) and obtain nC(x, y)1C = xy∗ + yx∗ ∈ I. Then either I = C or nC(x, y) = 0 for all x ∈ I, y ∈ C, forcing I = {0} by non-degeneracy. (b) Assuming C is not simple, we must show it has dimension 2. Since (C, ιC ) is simple as an algebra with involution by (a), there exists a k-algebra A such that 10 SKIP GARIBALDI AND HOLGER P. PETERSSON (C, ιC ) ∼= (Aop ⊕ A, ε) as algebras with involution, ε being the exchange involution of Aop ⊕ A. Then A, embedded diagonally into Aop ⊕ A, and H := H(C, ιC ) := {x ∈ C x = x∗} ⊆ C identify canonically as vector spaces over k. In particular, not only the dimension of H but also its codimension in C agree with the dimension of A. Applying (2.1.6), we conclude that H has dimension 1 for char(k) 6= 2 and codimension 1 for char(k) = 2. In both cases, C must be 2-dimensional. (cid:3) 2.6. Proposition. Let C be a flexible conic algebra over k whose norm is anisotropic. Then either C is non-degenerate or ∂nC = 0. Proof. Since, by 1.4, all non-zero elements of C are invertible, C is a simple algebra. On the other hand, (2.2.2) shows that I := C⊥ ⊆ C is an ideal. If I = {0}, then C is non-degenerate. If I = C, then ∂nC = 0. (cid:3) 2.7. Identities in conic alternative algebras. If C is a conic alternative algebra, then by [33, p. 97] its norm permits composition: (1) nC(xy) = nC(x)C n(y). Linearizing (1), we obtain (2) (3) (4) nC (x1y, x2y) = nC(x1, x2)nC (y), nC (xy1, xy2) = nC(x)nC (y1, y2), nC (x1y1, x2y2) + nC(x1y2, x2y1) = nC(x1, x2)nC (y1, y2). Moreover, by [33, Prop. 3.9] and (2.2.3), (5) xyx = nC (x, y∗)x − nC(x)y∗ = tC (xy)x − nC (x)y∗. 2.8. Theorem. Let C be a conic alternative algebra over k. Then (1) nC([x1, x2, x3]) = 4nC(x1)nC (x2)nC (x3) −X tC(xi)2nC (xj )nC (xl)+ j )nC (xl) − tC (x1x2)tC (x2x3)tC (x3x1)+ X tC (xixj)tC (xix∗ tC (x1x2x3)tC (x2x1x3) for all x1, x2, x3 ∈ C, where both summations on the right of (1) are taken over the cyclic permutations (ijl) of (123). Proof. Expanding n([x1, x2, x3]) = n((x1x2)x3) − n((x1x2)x3, x1(x2x3)) + n(x1(x2x3)) and applying (2.7.1), we conclude (2) n([x1, x2, x3]) = 2n(x1)n(x2)n(x3) − n(cid:0)(x1x2)x3, x1(x2x3)(cid:1). Turning to the second summand on the right of (2), we obtain, by (2.7.4), where applying (2.2.1) to the first summand on the right yields n(cid:0)(x1x2)x3, x1(x2x3)(cid:1) = n(x1x2, x1)n(x3, x2x3) − n(cid:0)(x1x2)(x2x3), x1x3(cid:1), n(cid:0)(x1x2)x3, x1(x2x3)(cid:1) = t(x2)2n(x3)n(x1) − n(cid:0)(x1x2)(x2x3), x1x3(cid:1). (3) WILD PFISTER FORMS 11 Manipulating the expression (x1x2)(x2x3) by means of (2.1.5) and the Moufang identities (1.1.3), we obtain (x1x2)(x2x3) = (x1x2) ◦ (x2x3) − (x2x3)(x1x2) = t(x1x2)x2x3 + t(x2x3)x1x2 − n(x1x2, x2x3)1 − x2(x3x1)x2, where (2.7.5),(2.2.4) yield (4) (x1x2)(x2x3) = t(x1x2)x2x3 + t(x2x3)x1x2 − n(x1x2, x2x3)1− t(x1x2x3)x2 + n(x2)(x3x1)∗. Here we use (2.2.2),(2.1.6) to compute n(x1x2, x2x3) = n(x1, x2x3x∗ 2) = t(x2)n(x1, x2x3) − n(x1, x2x3x2) and (2.2.3),(2.7.5) give n(x1x2, x2x3) = t(x2)n(x1, x2x3) − t(x2x3)n(x1, x2) + t(x3x1)n(x2) = t(x1)t(x2)t(x2x3) − t(x2)t(x1x2x3) − t(x1)t(x2)t(x2x3)+ t(x1x2)t(x2x3) + t(x3x1)n(x2), hence n(x1x2, x2x3) = t(x1x2)t(x2x3) − t(x2)t(x1x2x3) + t(x3x1)n(x2). Inserting this into (4), and (4) into the second term on the right of (3), we conclude n(cid:0)(x1x2)(x2x3), x1x3(cid:1) = t(x1x2)n(x2x3, x1x3) + t(x2x3)n(x1x2, x1x3)− t(x1x2)t(x2x3)t(x3x1) + t(x2)t(x3x1)t(x1x2x3)− t(x3x1)2n(x2) − t(x1x2x3)n(x2, x1x3)+ n(cid:0)(x3x1)∗, x1x3(cid:1)n(x2) t(x1x2)t(x2x3)t(x3x1) + t(x2)t(x3x1)t(x1x2x3)− t(x3x1)2n(x2) − t(x2)t(x3x1)t(x1x2x3)+ t(x1x2x3)t(x2x1x3) + t(x3x2 2)n(x3) + t(x2x3)t(x2x∗ = t(x1x2)t(x1x∗ 3)n(x1)− 1x3)n(x2), where we may use (2.1.4),(2.2.3),(2.2.2) to expand t(x3x2 1x3)n(x2) − t(x3x1)2n(x2) = t(x2 3x2 1)n(x2) − t(x3x1)2n(x2) t(x3x1)2n(x2) = t(cid:0)[t(x3)x3 − n(x3)1][t(x1)x1 − n(x1)1](cid:1)n(x2)− = t(x3)t(x1)t(x3x1)n(x2) − t(x1)2n(x2)n(x3)− t(x3)2n(x1)n(x2) + 2n(x1)n(x2)n(x3)− t(x3x1)2n(x2) = t(x3x1)t(x3x∗ 1)n(x2) − t(x1)2n(x2)n(x3)− t(x3)2n(x1)n(x2) + 2n(x1)n(x2)n(x3). 12 SKIP GARIBALDI AND HOLGER P. PETERSSON Inserting the resulting expression n(cid:0)(x1x2)(x2x3), x1x3(cid:1) = X t(xixj)t(xix∗ j )n(xl) − t(x1x2)t(x2x3)t(x3x1)+ t(x1x2x3)t(x2x1x3) − t(x1)2n(x2)n(x3)− t(x3)2n(x1)n(x2) + 2n(x1)n(x2)n(x3) into (3) and (3) into (2), the theorem follows. (cid:3) Remark. The associator of a conic alternative algebra being alternating, its norm must be totally symmetric in all three variables. Since the expression tC (xy∗) is symmetric in x, y ∈ C by (2.2.3), this fact is in agreement with the right-hand side of (2.8.1). 3. Interlude: Pfister bilinear and quadratic forms in characteristic 2 Working over an arbitrary field k of characteristic 2, the main purpose of this section is to collect a few results on Pfister bilinear and Pfister quadratic forms that are hardly new but, at least in their present form, apparently not in the literature. The final result, Prop. 3.12, concerns quadratic forms in arbitrary characteristic and may be amusing even for experts. Recall the following from, for example, [21, 8.5(iv)]. If one is given an anisotropic n-Pfister bilinear form b ∼= hhα1, . . . , αnii, then K := k(√α1, . . . ,√αn) is a field of dimension 2n over k such that K 2 ⊆ k. Conversely, given such an extension K/k, we can view K as a k-vector space endowed with a k-valued quadratic form q : x 7→ x2, and one checks that q is isomorphic to the quadratic form v 7→ b(v, v). This defines a bijection between isomorphism classes of extensions K/k of dimension 2n with K 2 ⊆ k and anisotropic n-quasi-Pfister quadratic forms [15, 10.4]. 3.1. Unital linear forms. We now refine this bijection. Fix an extension K/k as in the previous paragraph, and a k-linear form s : K → k such that s(1K) = 1, i.e., s is a retraction of the inclusion k ֒→ K. (We say that s is unital.) Define a symmetric bilinear bK,s : K × K −→ k via bK,s(u, v) := s(uv) for u, v ∈ K; it is the transfer s∗h1i in the notation of [15, §20.A]. Moreover, it is anisotropic (hence non-degenerate) since bK,s(u, u) = u2 for u ∈ K. We remark that if both s and t are unital linear forms on K, then (by the nondegeneracy of bK,s) there is some u ∈ K × such that bK,s(u, -- ) = t, i.e., s(uv) = t(v) for all v ∈ K. 3.2. Pfister bilinear forms. Fix a finite extension K/k such that K 2 ⊆ k as assumed above. We compute bK,s for a particular s. Fix a 2-basis √α1, . . . ,√αn of K. The monomials √α1 in with ij ∈ {0, 1} are a k-basis for K and we define s to be 1 on 1k and 0 on the other monomials. One sees immediately that bK,s is isomorphic to the Pfister bilinear form hhα1, . . . , αnii. In fact, the s constructed above is the general case. Suppose we are given a unital linear form s : K → k; we will construct a 2-basis so that s is as in the previous paragraph. Start with A = ∅ and repeat the following loop: If k(A) = K, then we are done. Otherwise, [K : k(A)] is at least 2, hence there is some a 6= 0 in the intersection of the k-subspaces Ker(s) and k(A)⊥, orthogonal complementation i1 ···√αn WILD PFISTER FORMS 13 relative to bK,s. As a is not in k(A), A ∪ {a} is p-free by [6, §V.13.1, Prop. 3]. We replace A by A ∪ {a} and repeat. In this way, we have proved: If [K : k] = 2n, then bK,s is an anisotropic n- Pfister bilinear form. Conversely, if we are given an anisotropic symmetric bilinear form hhα1, . . . , αnii over k, then √α1, . . . ,√αn is a 2-basis for a purely inseparable extension K/k as in the coarser correspondence recalled at the beginning of the section. The preceding considerations can be made more precise by looking at the cat- egory of pairs (K, s), where K/k is a finite purely inseparable field extension of exponent at most 1, s : K → k is a unital linear form, and morphisms are (auto- matically injective) k-homomorphisms of field extensions preserving unital linear forms. The map (K, s) 7→ bK,s defines a functor from this category into the cate- gory of anisotropic Pfister bilinear forms over k where morphisms are (automatically injective) isometries of bilinear forms. 3.3. Proposition. The functor just defined is an equivalence of categories. Proof. Given two pairs (K, s), (K ′, s′) of finite purely inseparable field extensions of exponent at most 1 over k with unital linear forms, we need only show that any isometry ϕ : bK,s → bK ′,s′ is, in fact, a field homomorphism preserving unital linear forms. We have s′(ϕ(u)ϕ(v)) = s(uv) for all u, v ∈ K, hence ϕ(u)2 = s′(ϕ(u)2) = s(u2) = u2 by unitality of s, s′, which implies ϕ(uv)2 = (uv)2 = u2v2 = (ϕ(u)ϕ(v))2 and therefore ϕ(uv) = ϕ(u)ϕ(v) since we are in characteristic 2. Thus ϕ is a k-homomorphism of fields preserving unital linear forms in view of s′(ϕ(u)) = s′(ϕ(1K)ϕ(u)) = s(1Ku) = s(u). (cid:3) 3.4. The passage to Pfister quadratic forms. Let α ∈ k be a scalar and s : K → k a unital linear form. We define the quadratic form qK;α,s to be the transfer s∗(hhαK ⊗ K), where, as usual, hhαK stands for the binary quadratic form given on k ⊕ kj by the matrix ( 1 1 0 α ), so hhαK(β + γj) = β2 + βγ + αγ2 for β, γ ∈ k. By the projection formula [15, p. 84], qK;α,s is isomorphic to bK,s⊗hhαK. More concretely, on the vector space direct sum K ⊕ Kj of two copies of K over k, we can define qK;α,s by the formula (u, v ∈ K). qK;α,s(u + vj) := u2 + s(uv) + αv2 (1) 3.5. Example. For K and s as in the first paragraph of 3.2 and αn+1 ∈ k, the form qK;αn+1,s is isomorphic to hhα1, . . . , αn, αn+1K. 3.6. Theorem. For α ∈ k and s : K → k a unital linear form, qK;α,s is an (n + 1)-Pfister quadratic form over k. Conversely, every anisotropic (n + 1)-Pfister quadratic form over k is isomorphic to qK;α,s for some purely inseparable field extension K/k of exponent at most 1 and degree 2n, some α ∈ k and some unital linear form s : K → k. Proof. Combine 3.2 and Example 3.5. (cid:3) The Pfister quadratic forms qK;α,s of 3.4 need not be anisotropic, nor can isometries between two of them be described as easily as in the case of Pfister bilinear forms (cf. Prop. 3.3). 14 SKIP GARIBALDI AND HOLGER P. PETERSSON 3.7. Proposition. Given scalars α, β ∈ k and unital linear forms s, t : K → k, the following conditions are equivalent. (i) The Pfister quadratic forms qK;α,s and qK;β,t are isometric. (ii) There exist elements u0, v0 ∈ K such that t(u) = s(uv0) 0 + s(u0v0) + αv2 0, β = u2 (u ∈ K). (1) Proof. We identify K ⊆ K ⊕ Kj canonically through the first summand. If q := qK;α,s and q′ := qK;β,t are isometric, Witt's theorem [15, (i) =⇒ (ii). Thm. 8.3] yields a bijective k-linear isometry ϕ : K ⊕ Kj ∼→ K ⊕ Kj from q′ to q (so q ◦ ϕ = q′) that is the identity on K. Then there are elements u0, v0 ∈ K such that (2) and there are k-linear maps f, g : K → K such that (3) ϕ(j) = u0 + v0j, ϕ(u + vj) =(cid:0)u + f (v)(cid:1) + g(v)j Combining (2), (3), we conclude (u, v ∈ K). (4) while evaluating q ◦ ϕ = q′ at u + vj ∈ K ⊕ Kj with the aid of (3) and (1) yields f (1K) = u0, g(1K) = v0, t(uv) + βv2 = f (v)2 + s(cid:0)ug(v)(cid:1) + s(cid:0)f (v)g(v)(cid:1) + αg(v)2 for all u, v ∈ K. Setting u = 0, v = 1K and observing (4), we obtain the first equation of (1). The second one now follows by setting v = 1K again but keeping u ∈ K arbitrary. tion using (1) shows that the map (ii) =⇒ (i). s(v0) = t(1K) = 1 implies v0 ∈ K ×, and a straightforward verifica- K ⊕ Kj ∼−→ K ⊕ Kj, is a bijective isometry from q′ to q. u + vj 7−→ (u + u0v) + (v0v)j, (cid:3) 3.8. Corollary. Let s : K → k be a unital linear form. Given β ∈ k and a unital linear form t : K → k, there exists an element α ∈ k such that qK;β,t ∼= qK;α,s. Proof. By 3.1, some v0 ∈ K satisfies t(u) = s(uv0) for all u ∈ K. Now Prop. 3.7 applies. (cid:3) 3.9. The Artin-Schreier map. Let s : K → k be a unital linear form. Then ℘K,s : K −→ k, u 7−→ ℘K,s(u) := u2 + s(u) is called the Artin-Schreier map of K/k relative to s. It is obviously additive and becomes the usual Artin-Schreier map (in characteristic p = 2), simply written as ℘, for K = k. Given another unital linear form t : K → k, we conclude from 3.1 that there is a unique element v0 ∈ K × satisfying s(v0) = 1 and t(u) = s(uv0) for all u ∈ K. Hence ℘K,t(u) = v2 0℘K,s(uv−1 0 ) since the left-hand side is equal to u2 + s(uv0) = v2 (u ∈ K) 0 )2 + s(uv−1 0 )]. 0[(uv−1 WILD PFISTER FORMS 15 With K ′ := Ker(s), we obtain the orthogonal splitting K = k1K ⊥ K ′ relative to K,s := bK,sK ′×K ′ up to isometry is uniquely bK,s, so the symmetric bilinear form b′ determined by bK,s ∼= h1i ⊥ b′ K,s. For α ∈ k, u′ ∈ K ′ we obtain (1) ℘K,s(α1K + u′) = ℘(α) + b′ K,s(u′, u′) = ℘(α) + u′2. 3.10. Corollary. Let s : K → k be a unital linear form and α, β ∈ k. (a) qK;α,s is isotropic if and only if α ∈ Im(℘K,s). (b) qK;α,s and qK;β,s are isometric if and only if α ≡ β mod Im(℘K,s). Proof. (a) If q := qK;α,s is isotropic, (3.4.1) shows u2 + s(uv) + αv2 = 0 for some u, v ∈ K not both zero, which implies v 6= 0 and then α = (uv−1)2 + s(uv−1) ∈ Im(℘K,s). Conversely, α ∈ Im(℘K,s) implies α = u2 + s(u) for some u ∈ K, forcing q(u + j) = 0 by (3.4.1). (b) By Prop. 3.7, the two quadratic forms are isometric iff (3.7.1) holds with t = s. But this forces v0 = 1K, and (3.7.1) becomes equivalent to α ≡ β mod Im(℘K,s). (cid:3) 3.11. Remark. (a) For a symmetric bilinear form b on a vector space V over k, we adopt the usual notation from the theory of quadratic forms by writing D(b) := {b(v, v)v ∈ V } [15, Def. 1.12] as an additive subgroup of k (recall char(k) = 2). With this notation, and observing (3.9.1), Cor. 3.10 (a) amounts to saying that the quadratic form qK;α,s is isotropic if and only if α belongs to Im(℘) + D(b′ K,s); written in this way, Cor. 3.10 (a) agrees with [15, Lemma 9.11]. (b) The definition of the quadratic forms qK;α,s in 3.4 as well as of the Artin-Schreier map in 3.9 make sense also if K has infinite degree over k, and the isomorphism qK;α,s ∼= bK,s ⊗ hhαK as well as Cor. 3.10 (a) continue to hold under these more general circumstances. We close this section with an observation that will play a useful role in the discussion of types for composition algebras over 2-Henselian fields, cf. in particular 12.7 below. For the remainder of this section, we dispense ourselves from the overall restriction that k have characteristic 2. 3.12. Proposition. Let q be a Pfister quadratic form over a field k of arbitrary characteristic and suppose q1, q2 are Pfister quadratic subforms of q with dim(q1) ≤ dim(q2). Then there are Pfister bilinear forms b1, b2 over k such that b2 ⊗ b1 ⊗ q1 ∼= q ∼= b2 ⊗ q2. In particular, if dim(q1) = dim(q2), then b ⊗ q1 ∼= q ∼= b ⊗ q2 for some Pfister bilinear form b over k. Proof. We may assume that q is anisotropic. Writing q : V → k, qi = qVi : Vi → k for i = 1, 2, with a vector space V of dimension n over k and subspaces Vi ⊆ V of dimension ni (0 ≤ n1 ≤ n2 ≤ n), we then argue by induction on r := n − n1. If r = 0, we put b1 := b2 := h1i. Now suppose r > 0. Since the qi represent 1, there are ei ∈ Vi with q(ei) = qi(ei) = 1. Now Witt's theorem [15, Thm. 8.3] yields an orthogonal transformation f ∈ O(V, q) with f (e1) = e2, and replacing q1 by 16 SKIP GARIBALDI AND HOLGER P. PETERSSON q1◦ f −1f (V1) : f (V1) → k if necessary, we may assume e1 = e2 ∈ V1 ∩ V2. Assuming n2 < n, and taking orthogonal complements relative to ∂q, we obtain dimk(V ⊥ 1 ∩ V ⊥ 2 ) = dimk(cid:0)(V1 + V2)⊥(cid:1) = 2n − dimk(V1 + V2) = 2n − dimk(V1) − dimk(V2) + dimk(V1 ∩ V2) > 2n − 2n1 − 2n2 ≥ 2n − 2n2+1 ≥ 0. Hence we can find a non-zero vector j ∈ V ⊥ Setting µ := −q(j) ∈ k×, we conclude from [15, Lemma 23.1] that q′ a subform of q, and the same holds true for q′ case we put q′ bilinear forms b′ 1 ∼= q ∼= b′ b′ 1 ⊗ q′ if n2 < n. 1 := hhµii⊗ q1 is 2 := hhµii ⊗ q2 unless n2 = n, in which In any event, the induction hypothesis yields Pfister 2 = h1i for n2 = n and 2 := q2 = q. 1, b′ 2 ⊗ q′ 2 over k such that b′ 2. Hence b1 := b′ that also belongs to V ⊥ 2 2 is a subform of b′ 1, b′ 1 ⊗ hhµii, 1 b2 :=(h1i b′ 2 ⊗ hhµii for n2 = n, for n2 < n are Pfister bilinear forms that satisfy b1⊗ q1 ∼= b′ 1⊗ q′ completing the induction step since b2 obviously is a subform of b1. 1⊗hhµii⊗ q1 ∼= b′ 1 ∼= q ∼= b2⊗ q2, (cid:3) 4. A non-orthogonal Cayley-Dickson construction By the embedding property 1.11, the Cayley-Dickson construction may be re- garded as a tool to recover the structure of a composition algebra C having di- mension 2n, n = 1, 2, 3, from a composition subalgebra B ⊆ C of dimension 2n−1: there exists a scalar µ ∈ k× with C ∼= Cay(B, µ). In the present section, a similar construction will be developed achieving the same objective when B is replaced by an inseparable subfield of degree 2n−1; concerning the existence of such subfields, see 1.13. The construction we are going to present is less general but more intrinsic than the one recently investigated by Pumplun [43]. Throughout this section, we work over a field k of characteristic 2. 4.1. Unital linear forms and conic algebras. Let K/k be a purely inseparable field extension of exponent at most 1, C a conic alternative k-algebra containing K as a unital subalgebra and let l ∈ C. Since K has trivial conjugation by 1.6 and C satisfies (2.2.3), the relation (1) sl(u) := nC (u, l) = tC (ul) holds for all u ∈ K and defines a linear form sl : K → k, which is unital in the sense of 3.1 if tC(l) = 1. In this case, sl is called the unital linear form on K associated with l. The following proposition paves the way for the non-orthogonal Cayley-Dickson construction we have in mind. 4.2. Proposition. Let C be a conic alternative algebra over k and K ⊆ C a purely inseparable subfield of exponent at most 1. Suppose l ∈ C satisfies tC (l) = 1, put µ := nC (l) ∈ k and write s := sl for the unital linear form on K associated with WILD PFISTER FORMS 17 l. Then B := K + Kl ⊆ C is the subalgebra of C generated by K and l. More precisely, the relations (1) (2) (3) (vl)u = s(u)v + uv + (uv)l, u(vl) = s(uv)1C + s(u)v + s(v)u + uv + (uv)l, (v1l)(v2l) = s(v1v2)1C + s(v1)v2 + s(v2)v1 + (1 + µ)v1v2+ (cid:0)s(v1v2)1C + s(v1)v2 + v1v2(cid:1)l hold for all u, v, v1, v2 ∈ K. Proof. The assertion about B will follow once we have established the relations (1)−(3). To do so, we first prove (4) u(vl) + (vl)u = s(uv)1C + s(v)u. Since K has trace zero, we combine the definition of s with (2.1.5),(2.2.2) and obtain u(vl) + (vl)u = tC (u)vl + tC (vl)u − nC (u, vl)1C = s(v)u − nC (v∗u, l)1C = s(v)u + nC (uv, l)1C = s(uv)1C + s(v)u, giving (4). In order to establish (1), it suffices to show (5) (vx)u = tC(ux)v + tC (x)uv + (uv)x (u, v ∈ K, x ∈ C) since this implies (1) for x = l. The assertion being obvious for u = v = 1, we may assume K 6= k, forcing k to be infinite. By Zariski density, we may therefore assume that x is invertible. Then the Moufang identities (1.1.3) combine with (2.7.5), (2.1.4), the right alternative law (1.1.2) and (4.1.1) to imply (cid:0)(vx)u(cid:1)x = v(xux) = v(cid:0)tC (xu)x − nC (x)u∗(cid:1) = tC (ux)vx + nC(x)uv = tC(ux)vx + (uv)(cid:0)tC (x)x + x2(cid:1) =(cid:0)tC(ux)v + tC (x)uv + (uv)x(cid:1)x, hence (5). Combining (4) and (1), we now obtain (2). Finally, making use of the Moufang identities again and of (1),(4),(4.1.1), we conclude (v1l)(v2l) = (v1l + lv1)(v2l) + (lv1)(v2l) =(cid:0)s(v1)1C + s(1C )v1(cid:1)(v2l) + l(v1v2)l = s(v1)v2l + v1(v2l) + tC(cid:0)(v1v2)l(cid:1)l − nC(l)(v1v2)∗ = s(v1)v2l + v1(v2l) + s(v1v2)l + µv1v2. Applying (2) to the second summand on the right gives (3). (cid:3) 4.3. Embedding inseparable field extensions into conic algebras. We now look at the converse of the situation described in Prop. 4.2. Let K/k be a purely inseparable field extension of exponent at most 1. Suppose we are given a unital linear form s : K → k and a scalar µ ∈ k. Inspired by the relations (4.2.1)−(4.2.3), we now observe the obvious fact that the vector space direct sum K ⊕ Kj of two copies of K carries a unique unital non-associative k-algebra structure C := Cay(K; µ, s) 18 SKIP GARIBALDI AND HOLGER P. PETERSSON into which K embeds as a unital subalgebra through the first summand such that the relations (vj)u = (cid:0)s(u)v + uv(cid:1) + (uv)j, u(vj) = (cid:0)s(uv)1K + s(u)v + s(v)u + uv(cid:1) + (uv)j, (v1j)(v2j) = (cid:0)s(v1v2)1K + s(v1)v2 + s(v2)v1 + (1 + µ)v1v2(cid:1)+ (cid:0)s(v1v2)1K + s(v1)v2 + v1v2(cid:1)j hold for all u, v, v1, v2 ∈ K. Adding (1) to (2), we obtain (4) u ◦ (vj) = u(vj) + (vj)u = s(uv)1K + s(v)u. 4.4. Proposition. With the notations and assumptions of 4.3, C = Cay(K; µ, s) is a non-degenerate flexible conic k-algebra with norm, polarized norm, trace, con- jugation respectively given by the formulas nC (u + vj) = u2 + s(uv) + µv2, nC (u1 + v1j, u2 + v2j) = s(u1v2) + s(u2v1), (1) (2) (3) (1) (2) (3) tC (u + vj) = s(v), (u + vj)∗ = s(v)1K + u + vj (4) for all u, u1, u2, v, v1, v2 ∈ K. Moreover, nC = qK;µ,s (cf. 3.4), and this is an (n + 1)-Pfister quadratic form if K has finite degree 2n over k. Proof. The right-hand side of (1) defines a quadratic form n : C → k whose polarization is given by the right-hand side of (2). In particular, setting t := ∂n(1C,- −), we obtain t(u + vj) = s(v) and then, using (4.3.3),(4.3.4), (u + vj)2 = u2 + u ◦ (vj) + (vj)2 = u2 + s(uv)1C + s(v)u + v2 + (1 + µ)v2 + s(v)vj = s(v)(u + vj) +(cid:0)u2 + s(uv) + µv2(cid:1)1C = t(u + vj)(u + vj) − n(u + vj)1C . Thus C is indeed a conic k-algebra, and (1)−(4) hold. Since K is a field and s is unital, forcing bK,s to be non-degenerate by 3.1, we conclude from (2) that ∂nC is non-degenerate as well. The final statement of the proposition follows from comparing (1) with (3.4.1) and applying Thm. 3.6. It therefore remains to show that C is flexible. We do so by invoking Remark 2.3 and verifying the first relation of (2.2.1). Letting u, v, w ∈ K be arbitrary and setting x = u + vj, we may assume y = w or y = wj. Leaving the former case to the reader, we apply (1),(3),(4.3.1−3) and compute nC(xy, x) − nC(x)tC (y) = nC(cid:0)u(wj) + (vj)(wj), u + vj(cid:1) − nC(u + vj)tC (wj) = nC(cid:16)s(uw)1K + s(u)w + s(w)u + uw + s(vw)1K + s(v)w + s(w)v + (1 + µ)vw +(cid:0)uw + s(vw)1K + s(v)w + vw(cid:1)j, u + vj(cid:17) +(cid:0)u2 + s(uv) + µv2(cid:1)s(w) = u2s(w) + s(u)s(vw) + s(v)s(uw) + s(uvw) + s(v)s(uw) + s(u)s(vw) + s(w)s(uv) + s(uvw) + s(v)s(vw) + s(v)s(vw) + v2s(w) + v2s(w) + µv2s(w) + u2s(w) + s(w)s(uv) + µv2s(w) = 0, WILD PFISTER FORMS which completes the proof. 19 (cid:3) Prop. 4.2 not only serves to illuminate the intuitive background of the non- orthogonal Cayley-Dickson construction presented in 4.3 and Prop. 4.4. It also allows for the following application. 4.5. Proposition. Let C be a conic alternative algebra over k and K ⊆ C a purely inseparable subfield of exponent at most 1. Suppose l ∈ C satisfies tC (l) = 1, put µ := nC (l) ∈ k and write s := sl for the unital linear form on K associated with l. Then there is a unique homomorphism ϕ : Cay(K; µ, s) −→ C extending the identity of K and satisfying ϕ(j) = l. Moreover, ϕ is injective, and its image is the subalgebra of C generated by K and l. Proof. The uniqueness assertion being obvious, define ϕ : Cay(K; µ, δ) → C by ϕ(u + vj) := u + vl for u, v ∈ K. Then ϕ is a k-linear map whose image by Prop. 4.2 agrees with the subalgebra of C generated by K and l. But ϕ is also a homomorphism of algebras, which follows by comparing (4.3.1)−(4.3.3) with the corresponding relations (4.2.1)−(4.2.3). It remains to show that ϕ is injective, i.e., that u + vl = 0 for u, v ∈ K implies u = v = 0. Otherwise, we would have v 6= 0, which implies 0 = v−1(u + vl) = v−1u + l and applying tC yields a contradiction. (cid:3) It is a natural question to ask for conditions that are necessary and sufficient for a non-orthogonal Cayley-Dickson construction as in 4.3 to be a composition algebra. The answer is given by the following result. 4.6. Theorem. Let K/k be a purely inseparable field extension of exponent at most 1, s : K → k a unital linear form and µ ∈ k a scalar. We put C := Cay(K; µ, s). (a) The map fC : K 4 → k defined by fC(u1, u2, u3, u4) := s(u1u2u3u4) +X s(ui)s(ujulum) +X s(uiuj)s(ulu4) for u1, u2, u3, u4 ∈ K, where the first (resp. second) sum on the right is taken over all cyclic permutations ijlm (resp. ijl) of 1234 (resp. 123), is an alternating quadri-linear map. Setting K := K/k1K, fC induces canonically an alternating quadri-linear map fC : K 4 → k. Moreover, the relation nC (xy) = nC (x)nC (y) + fC (u1, u2, v1, v2) (1) holds for all x = u1 + v1j, y = u2 + v2j ∈ C with ui, vi ∈ K, i = 1, 2. (b) The following conditions are equivalent. (i) C is a composition algebra. (ii) fC = 0. (iii) [K : k] ≤ 4. Proof. (a) Setting f := fC, which is obviously quadri-linear, it is straightforward to check that it is alternating as well (hence symmetric since we are in characteristic 2) and satisfies f (1K, u2, u3, u4) = 0 for all ui ∈ K, i = 1, 2, 3. This proves existence f with the desired properties. It remains to establish (1). Subtracting the first of 20 SKIP GARIBALDI AND HOLGER P. PETERSSON summand on the right from the left, and expanding the resulting expression in the obvious way, we conclude that it decomposes into the sum of terms nC(cid:0)(vj)u(cid:1)− nC (vj)nC (u), nC(cid:0)u(vj)(cid:1)− nC (u)nC (vj), nC(cid:0)(v1j)(v2j)(cid:1)− nC (v1j)nC (v2j), nC(cid:0)u1u2, u1(v2j)(cid:1)− nC (u1)nC (u2, v2j), nC (u1u2, (v1j)u2)(cid:1)− nC (u1, v1j)nC (u2), nC(cid:0)u1(v2j), (v1j)(v2j)(cid:1)− nC (u1, v1j)nC (v2j), nC(cid:0)(v1j)u2, (v1j)(v2j)(cid:1)− nC (v1j)nC (u2, v2j), nC(cid:0)u1u2, (v1j)(v2j)(cid:1) + nC(cid:0)u1(v2j), (v1j)u2(cid:1)− nC (u1, v1j)nC (u2, v2j). A tedious but routine computation, involving (4.2.1)−(4.2.3) and (4.4.1),(4.4.2) shows that all but the very last one of these expressions are equal to zero. Hence (1) follows from nC(cid:0)u1u2, (v1j)(v2j)(cid:1) + nC(cid:0)u1(v2j), (v1j)u2(cid:1) = nC(cid:16)u1u2,(cid:0)s(v1v2)1K + s(v1)v2 + s(v2)v1 + (1 + µ)v1v2(cid:1)+ (cid:0)s(v1v2)1K + s(v1)v2 + v1v2(cid:1)j(cid:17)+ nC(cid:16)(cid:0)s(u1v2)1K + s(u1)v2 + s(v2)u1 + u1v2(cid:1) + (u1v2)j,(cid:0)s(u2)v1 + u2v1(cid:1) + (u2v1)j(cid:17) = s(u1u2)s(v1v2) + s(v1)s(u1u2v2) + s(u1u2v1v2) + s(u1v2)s(u2v1)+ s(u1)s(u2v1v2) + s(v2)s(u1u2v1) + s(u1u2v1v2) + s(u2)s(u1v1v2) + s(u1u2v1v2) = s(u1u2v1v2) + s(u1)s(u2v1v2) + s(u2)s(v1v2u1) + s(v1)s(v2u1u2) + s(v2)s(u1u2v1)+ s(u1u2)s(v1v2) + s(u2v1)s(u1v2) + s(v1u1)s(u2v2) + s(u1v1)s(u2v2) = f (u1, u2, v1, v2) + nC (u1, v1j)nC (u2, v2j). (b) (i) =⇒ (iii) follows from the fact that composition algebras have dimension at most 8 and that the dimension of C is twice the degree of K/k. (iii) =⇒ (ii) follows from the fact that K has dimension at most 3, so any alternating quadri-linear linear map on K must be zero. (ii) =⇒ (i) follows immediately from (1) since ∂nC by Prop. 4.4 is non-degenerate. (cid:3) Remark. Thm. 4.6 (b) can be proved without recourse to any a priori knowledge of composition algebras. To do so, it suffices to show directly that [K : k] > 4 implies fC 6= 0, which can be done quite easily. We omit the details. 5. The Skolem-Noether theorem for inseparable subfields The non-orthogonal Cayley-Dickson construction introduced in the preceding section will be applied in two ways. Recalling from [47] that every isomorphism between composition subalgebras of a composition algebra C can be extended to an automorphism of C, the aim of our first application will be to derive an analogous result where the composition subalgebras are replaced by inseparable subfields. We WILD PFISTER FORMS 21 begin by exploiting more fully our description of Pfister quadratic forms presented in Section 3 within the framework of composition algebras. Throughout we continue to work over an arbitrary field k of characteristic 2. 5.1. Comparing non-orthogonal Cayley-Dickson constructions. Let K/k be a purely inseparable field extension of exponent at most 1 and degree 2n−1, 1 ≤ n ≤ 3. Suppose we are given scalars µ, µ′ ∈ k and unital linear forms s, s′ : K → k. By Thm. 4.6, C := Cay(K; µ, s) and C′ := Cay(K; µ′, s′) are composition algebras over k; the notational conventions of 4.3 will remain in force for C and will be extended to C′ = K ⊕ Kj′ in the obvious manner. By a K-isomorphism from C to C′ we mean an isomorphism that induces the identity on K; we say C and C′ are K-isomorphic if a K-isomorphism from C to C′ exists. 5.2. Proposition. With the notations and assumptions of 5.1, we have: (a) C is split if and only if µ ∈ Im(℘K,s). (b) C and C′ are K-isomorphic if and only if there exist u0, v0 ∈ K such that (1) s′(u) = s(uv0) 0 + s(u0v0) + µv2 0, µ′ = u2 (u ∈ K). Proof. Prop. 4.4. (a) C is split iff nC is isotropic iff µ ∈ Im(℘K,s) by Cor. 3.10 (a) and (b) If C and C′ are K-isomorphic, their norms are isometric, so by Prop. 3.7, some u0, v0 ∈ K satisfy (1). Conversely, let this be so. Setting l := u0 + v0j ∈ C and combining (4.4.1)−(4.4.3)with (1), we conclude tC (l) = s(v0) = s′(1K) = 1, nC(l) = u2 0 = µ′ and nC(u, l) = s(uv0) = s′(u) for all u ∈ K, so Prop. 4.5 yields a unique K-isomorphism C′ ∼→ C sending j′ to l. 0 + s(u0v0) + µv2 (cid:3) 5.3. Remark. While the set-up described in 5.1 above extends to the case of allow- ing purely inseparable extensions of exponent 1 and arbitrary degree 2n−1, n ≥ 1, in the obvious manner (replacing composition algebras by flexible conic ones in the process), our methods of proof become unsustainable in this generality. A typical example is provided by the proof of Prop. 5.2 (b), where a K-isomorphism from C′ to C sending j′ to l in general does not exist unless n ≤ 3. Indeed, assuming that every element l ∈ C of trace 1 allows a K-isomorphism ϕ : Cay(cid:0)K; nC(l), ∂nC(−, l)(cid:1) ∼−→ C with ϕ(j′) = l, one computes the expression ϕ((vj′)u)− ϕ(vj′)ϕ(u) = 0 and arrives at the conclusion that the trilinear map defined by (1) H(u1, u2, u3) := s(u1u2u3)1K +X(cid:16)s(uiuj)ul + s(ui)ujul(cid:17) + u1u2u3 for u1, u2, u3 ∈ K is zero, the sum on the right being extended over all cyclic permutations ijl of 123. Since H is obviously alternating and satisfies the relations H(1K, u2, u3) = H(u1, u2, u1u2) = 0 for all u1, u2, u3 ∈ K, it is easy to check (see the corresponding argument in the proof of Thm. 4.6) that H vanishes for n ≤ 3 (as it should). But for n > 3, we may choose u1, u2, u3 ∈ K to be 2-independent over k [6, V §13.2, Thm. 2], forcing the right-hand side of (1) to be a k-linear combination of linear independent vectors over k [6, V §13.2, Prop. 1], whence H(u1, u2, u3) 6= 0. 22 SKIP GARIBALDI AND HOLGER P. PETERSSON 5.4. Proposition. Let K/k be a purely inseparable field extension of exponent at most 1 and degree 2n−1, 1 ≤ n ≤ 3. Furthermore, let s : K → k be a unital linear form. (a) If C is a composition algebra of dimension 2n over k containing K as a unital subalgebra, then there exists a scalar µ ∈ k such that C and Cay(K; µ, s) are K- isomorphic. (b) For µ, µ′ ∈ k, the following conditions are equivalent. (i) Cay(K; µ, s) and Cay(K; µ′, s) are K-isomorphic. (ii) Cay(K; µ, s) ∼= Cay(K; µ′, s). (iii) µ ≡ µ′ mod Im(℘K,s). Proof. (a) Pick any l ∈ C of trace 1. Then C = K ⊕ Kl, and Prop. 4.5 yields a K-isomorphism C ∼→ Cay(K; µ′, s′), for some µ′ ∈ k and some unital linear form s′ : K → k. Hence there is a unique v0 ∈ K × such that s′(u) = s(uv0) for all u ∈ K. Setting µ := v−2 0 ) and consulting Prop. 5.2 (b), we obtain a K-isomorphism Cay(K; µ′, s′) ∼→ Cay(K; µ, s). (b) While (i) ⇒ (ii) is obvious, (ii) ⇒ (iii) follows from Cor. 3.10 (b). It remains to show (iii) ⇒ (i). But (iii) implies µ′ = µ + ℘K,s(u0) for some u0 ∈ K, so (5.2.1) holds with v0 = 1K. This gives (i). 0 µ′ + ℘K,s(u0v−1 (cid:3) Every element of an octonion algebra over a field is contained in a suitable quater- nion subalgebra [47, Prop. 1.6.4]. However, it doesn't seem entirely obvious that, if the octonion algebra is split, the quaternion subalgebra can be chosen to be split as well. But, in fact, it can: 5.5. Proposition. Let C be a split octonion algebra over an arbitrary field F . Then every element of C is contained in a split quaternion subalgebra of C. Proof. Let x ∈ C. We may assume that R := k[x] has dimension 2 over k. There are three cases [5, III §2 Prop. 3]. Case 1. R is the algebra of dual numbers. In particular, it contains zero divisors. Hence so does every quaternion subalgebra of C containing x, which therefore must be split. Case 2. R is (quadratic) ´etale. Since C up to isomorphism is uniquely determined by splitness, it may be obtained from R by the Cayley-Dickson process as C ∼= Cay(R; 1, 1), which implies that x is contained in the split quaternion subalgebra Cay(R, 1) of C. Case 3. We are left with the most delicate possibility that F = k has characteristic 2 and K := R is a purely inseparable field extension of k having exponent at most 1. Let c ∈ C be an idempotent different from 0, 1C, which exists since C is split, and write B for the subalgebra of C generated by K and c. Since c has trace 1, Prop. 4.5 yields a scalar µ ∈ k, a unital linear form s : K → k and an isomorphism from B′ := Cay(K; µ, s) onto B. But B′ is a quaternion algebra by Thm. 4.6 whereas B, containing c, has zero divisors. Hence B is a split quaternion subalgebra of C containing x. (cid:3) And finally, we need a purely technical result. WILD PFISTER FORMS 23 ∼→ L2. ∼→ 5.6. Lemma. Let C be an octonion division algebra over k and suppose ϕ : K1 K2 is an isomorphism of inseparable quadratic subfields K1, K2 ⊆ C. Then there exist inseparable subfields L1, L2 ⊆ C of degree 4 over k such Ki ⊆ Li for i = 1, 2 and ϕ extends to an isomorphism ψ : L1 Proof. Given any elements y ∈ C, xi ∈ Ki \ k1C for i = 1, 2, denote by Li the subalgebra of C generated by Ki and y. Since C has no zero divisors, Li is either a composition algebra or an inseparable field extension, the latter possibility being equivalent to the trace of C vanishing identically on Li. In any event, the dimension of Li is either 2 or 4. Moreover, Li is spanned by 1C, xi, y, xiy as a vector space over k. Summing up, Li/k is therefore an inseparable field extension of degree 4 if and only if y /∈ Ki satisfies the condition tC (y) = tC(xiy) = 0. To choose y appropriately, we now write C0 for the space of trace zero elements in C and consider the hyperplane intersection V := C0 ∩ x1C0 ∩ x2C0 ⊆ C, which is a subspace of dimension at least 5. Since a group cannot be the union of two proper subgroups, we conclude V * K1 ∪ K2 and, accordingly, pick a y ∈ V that neither belongs to K1 nor to K2. Then y has trace zero and y = xizi for some zi ∈ C0, so xiy = x2 i zi = nC (xi)zi ∈ kzi has trace zero as well, and by the above, Li/k is an inseparable field extension of degree 4 containing Ki. Moreover, K1 ∼= K2 implies 1 = K 2 K 2 2 , hence L2 1 = K 2 1 + K 2 so there is an isomorphism ψ : L1 k-embedding K1 → L2 is the one induced by ϕ. 2 + K 2 1 y2 = K 2 ∼→ L2, which necessarily extends ϕ since the only 2 y2 = L2 2, (cid:3) After these preparations, we can now establish the Skolem-Noether theorem for inseparable subfields of composition algebras. 5.7. Theorem. (Skolem-Noether) Let C be a composition algebra over k. Then every isomorphism between inseparable subfields of C can be extended to an auto- morphism of C. ∼→ K2 be an isomorphism Proof. Write dimk(C) = 2n, 0 ≤ n ≤ 3 and let ϕ : K1 of inseparable subfields K1, K2 ⊆ C having degree 2n′ , 0 ≤ n′ < n, over k. We may assume n′ > 0 and first reduce to the case n′ = n − 1. To do so, suppose the theorem holds for n′ = n − 1 and let n′ < n − 1. Then n′ = 1, n = 3, so C is an octonion algebra containing K1, K2 as inseparable quadratic subfields. If C is split, there are split quaternion subalgebras Bi ⊆ C containing Ki for i = 1, 2 (Prop. 5.5). But B1, B2 are isomorphic, hence conjugate under the automorphism group of C, by the classical Skolem-Noether theorem for composition algebras. Hence up to conjugation by automorphisms of C, we may assume B1 = B2 =: B. But then ϕ extends to an automorphism of B, which in turn extends to an automorphism of C. We are left with the case that C is a division algebra. By Lemma 5.6, there are inseparable subfields Ki ⊆ Li ⊆ C of degree 4 (i = 1, 2) such that ϕ extends ∼→ L2. But ψ in turn extends to an automorphism of to an isomorphism ψ : L1 C, completing the reduction to the case n′ = n − 1. From now on we assume n′ = n − 1 and fix a unital linear form s2 : K2 → k. Then s1 := s2 ◦ ϕ : K1 → k is a unital linear form as well. For i = 1, 2, Prop. 5.4 (a) yields scalars µi ∈ k and Ki-isomorphisms ψi : Cay(Ki; µi, si) ∼→ C. Now observe that the non-orthogonal Cayley-Dickson construction of 4.3 is functorial in the parameters involved. Hence 24 SKIP GARIBALDI AND HOLGER P. PETERSSON ϕ determines canonically an isomorphism ψ := Cay(ϕ) : Cay(K1; µ1, s1) ∼−→ Cay(K2; µ1, s1 ◦ ϕ−1) = Cay(K2; µ1, s2). Putting things together, we thus obtain an isomorphism 2 ◦ ψ1 ◦ ψ−1 : Cay(K2; µ1, s2) ∼−→ Cay(K2; µ2, s2). ψ−1 Applying Prop. 5.4 (b), we see that there exists a K2-isomorphism χ : Cay(K2; µ2, s2) ∼→ Cay(K2; µ1, s2), giving rise to the automorphism φ := ψ2◦ χ−1◦ ψ◦ ψ−1 ψ2 = χ−1 = 1 on K2 and ψ = ϕ, ψ−1 1 = 1 on K1. 1 of C, which extends ϕ since (cid:3) Remark. Let C be an octonion algebra over k and K ⊆ C an inseparable subfield of degree 4. Changing scalars to the algebraic closure, ¯k, of k, K ⊗k ¯k becomes a unital ¯k-subalgebra of ¯C := C ⊗k ¯k containing a 3-dimensional subalgebra N that consists entirely of nilpotent elements. Hence N ⊆ ¯C is a Borel subalgebra in the sense of [38]. The fact that all Borel subalgebras of ¯C are conjugate under its automorphism group [38, § 2, 1.] corresponds nicely with the Skolem-Noether theorem. It is a natural question to ask how our non-orthogonal Cayley-Dickson construc- tion can be converted into the classical orthogonal one. When dealing within the framework of composition (and not of arbitrary conic) algebras, here is a simple answer. 5.8. Orthogonalizing the non-orthogonal Cayley-Dickson construction. Let K/k be a purely inseparable field extension of exponent at most 1 and degree 2n, n = 1, 2. Suppose we are given an intermediate subfield k ⊆ K ′ ⊆ K of degree 2n−1, a scalar µ ∈ k and a unital linear form s : K → k. Then s′ := sK ′ : K ′ → k is a unital linear form on K ′ and B := Cay(K ′; µ, s′) is a composition subalgebra of C := Cay(K; µ, s). Moreover, (4.4.2) implies (1) B⊥ = K ′⊥ ⊕ K ′⊥j, orthogonal complementation in K (resp. C) being taken relative to bK,s (resp. ∂nC). From (1) we conclude (2) C = Cay(B,−nC (u)) = Cay(B; u2) for any non-zero element u ∈ K ′⊥. To be more specific, let C be an octonion algebra over k and K ⊆ C an insep- arable subfield of degree 4. Pick a 2-basis a = (a1, a2) of K/k. By Prop. 5.4 (a), there exists a scalar µ ∈ k with C ∼= Cay(K; µ, sa). On the other hand, L = Cay(k; µ, 1k) = k[t]/(t2 + t + µ) is a quadratic ´etale k-algebra, and (2) yields C = Cay(L; a2 1, a2 2) as an ordinary Cayley-Dickson process starting from L. WILD PFISTER FORMS 25 6. Conic division algebras in characteristic 2. We now turn to a second application of the non-orthogonal Cayley-Dickson con- struction which consists in finding new examples of conic division algebras. A few comments on the historical context seem to be in order. 6.1. Conic division algebras over arbitrary fields. Among all conic division algebras, it is the composition division algebras that are particularly well under- stood and particularly easy to construct: by 1.9, it suffices to ensure that their norms be anisotropic. Of course, composition division algebras exist only in di- mensions 1, 2, 4, 8, as do all non-associative division algebras over the reals, by the Bott-Kervaire-Milnor theorem [13, Kap. 10,§ 2]; in particular, the Cayley-Dickson process 1.12 leads to conic algebras Cay(R; µ1, . . . , µn), µ1 = ··· = µn = −1 (n ∈ Z, n ≥ 1) over the reals whose norms are positive definite (hence anisotropic) but which fail to be division algebras unless n ≤ 3. Hence it is natural to ask for examples of conic division algebras in dimensions other than 1, 2, 4, 8, over fields other than the reals. From the point of view of non-associative algebras, conic division algebras that are not central, like purely inseparable field extensions of characteristic 2 and ex- ponent 1 as in 1.6, are not particularly interesting. Over appropriate fields of characteristic not 2, the first examples of central conic division algebras in all di- mensions 2n, n = 0, 1, 2 . . . are apparently due to Brown [7, pp. 421-422]. They all arise from the base field, an iterated Laurent series field in finitely many variables, by the Cayley-Dickson process; generalizations of these examples will be discussed in Example 10.7 below. Other examples of central conic division algebras in dimen- sion 16, using a refinement of the Cayley-Dickson construction, have been exhibited by Becker [3, Satz 16]. Examples of central commutative conic division algebras in characteristic 2 and all dimensions 2n, n ≥ 0, have been constructed by Albert [2, Thm. 2]. These algebras are closely related to purely inseparable field exten- sions of exponent 1 since their norms bilinearize to zero and hence degenerate when extending scalars to the algebraic closure. In view of the preceding results one may ask whether the dimension of a finite- dimensional conic division algebra is always a power of 2. Though a feeble result along these lines has been obtained by Petersson [39], the answer to this question doesn't seem to be known. In this paper, two classes of conic division algebras in all dimensions 2n, n = 0, 1, 2 . . . , will be constructed. The examples of the first class, to be discussed in the present section, depend strongly on the non-orthogonal Cayley-Dickson con- struction, hence exist only in characteristic 2 but differ from Albert's by being central and highly non-commutative and by allowing arbitrary anisotropic Pfister quadratic forms as their norms, which in particular remain non-singular under all scalar extensions. The second class of examples will be discussed in 10.7 and 10.15 below. 6.2. Notations and conventions. For the remainder of this section, we fix a base field k of characteristic 2, a purely inseparable field extension K/k of exponent at most 1, a scalar µ ∈ k and a unital linear form s : K → k to consider the non-orthogonal Cayley-Dickson construction C := Cay(K; µ, s) as in 4.3. We put 26 SKIP GARIBALDI AND HOLGER P. PETERSSON [K : k] = 2n, n = 0, 1, 2, . . . and explicitly allow the possibility n = ∞, i.e., that K has infinite degree over k. The following proposition paves the way for the application we have in mind. 6.3. Proposition. With the notations and conventions of 6.2, the following asser- tions hold. (a) C is locally finite-dimensional. (b) For n ≥ 2, every element of C belongs to an octonion subalgebra of C. (c) C is central simple for n ≥ 1 and has trivial nucleus for n ≥ 2. (a) It suffices to note that finitely many elements xi = ui + vij ∈ Proof. C with ui, vi ∈ K (1 ≤ i ≤ m) are contained in Cay(K ′; µ, sK ′) (K ′ := k(u1, v1, . . . , um, vm)), which is a subalgebra of C having dimension at most 22m+1. (b) Let x = u + vj ∈ C, u, v ∈ K. Then there is a subfield K ′ ⊆ K containing u, v and having degree 4 over k, so C′ := Cay(K ′; µ, sK ′) ⊆ C by Thm. 4.6 is an octonion subalgebra containing x. (c) Standard properties of composition algebras allow us to assume n > 2. Com- bining Props. 2.5, 4.4 we see that C is simple. Let x ∈ Nuc(C) and apply (b) to pick an octonion subalgebra C′ ⊆ C containing x. Since C′ has trivial nucleus [47, Prop. 1.9.2], we conclude x ∈ k1C , so C has trivial nucleus as well; in particular, C is central. (cid:3) Referring the reader to our version of the Artin-Schreier map (3.9, Remark 3.11), we can now state the main result of this section. 6.4. Theorem. With the notations and conventions of 6.2, the following conditions are equivalent. (i) C is a division algebra. (ii) nC is anisotropic. (iii) µ /∈ Im(℘K,s). Proof. The implication (i) ⇒ (ii) follows from 1.5, while (ii) and (iii) are equivalent by Cor. 3.10 (a) and Remark 3.11 (b). It remains to prove (ii) =⇒ (i). Since C is locally finite-dimensional by Prop. 6.3 (a), it suffices to show that there are no zero divisors, so suppose x1, x2 ∈ C satisfy x1x2 = 0. By (ii) and (2.4.1), this implies x2x1 = 0, and from (2.1.5) we conclude 0 = x1x2 + x2x1 = tC(x1)x2 + tC (x2)x1 − nC (x1, x2)1C . If tC (x1) 6= 0, this yields x2 ∈ k[x1], hence x2 = 0 since nC being anisotropic implies that k[x1] is a field. By symmetry, we may therefore assume tC(x1) = tC (x2) = nC (x1, x2) = 0. Write xi = ui + vij with ui, vi ∈ K for i = 1, 2. Then s(vi) = 0 by (4.4.3), s(u1v2) = s(u2v1) by (4.4.2), and if v1 = 0 or v2 = 0, Thm. 4.6 (a) yields nC (x1)nC (x2) = nC (x1x2) = 0, hence x1 = 0 or x2 = 0. We are thus reduced to the case v1 6= 0 6= v2, s(v1) = s(v2) = 0, (1) Next we use (4.3.1−3) and (1) to expand (u1 + v1j)(u2 + v2j) = 0. A short computation gives s(u1v2) = s(u2v1). u1u2 + s(u1v2)1K + s(u1)v2 + u1v2 + s(u2)v1 + u2v1 + s(v1v2)1K + (1 + µ)v1v2 = 0, (2) u1v2 + u2v1 + s(v1v2)1K + v1v2 = 0. WILD PFISTER FORMS 27 Adding these two relations, we obtain (3) u1u2 + s(u1v2)1K + s(u1)v2 + s(u2)v1 + µv1v2 = 0 and note that (2),(3) are symmetric in the indices 1, 2 by (1). We now claim: (∗) The subfield K ′ of K/k generated by u1, u2, v1, v2 is spanned by (4) 1K, u1, u2, v1, v2, u1v2, u2v1 as a vector space over k Suppose for the time being that this claim has been proved. Then the field extension K ′/k has degree at most 7. Being purely inseparable at the same time, it has, in fact, degree at most 4. By Thm. 4.6 (b), C′ := Cay(K ′; µ, sK ′) ⊆ C is therefore a composition subalgebra containing x1, x2 and inheriting its anisotropic norm from C. Thus C′ is a division algebra, and x1x2 = 0 implies x1 = 0 or x2 = 0, as desired. We are thus reduced to showing (∗). Writing V for the linear span of the vectors in (4), it suffices to show that V ⊆ K is a k-subalgebra, i.e., that the product of any two distinct elements in (4) belongs to V . Since v1v2 ∈ V by (2), hence u1u2 ∈ V by (3), this will follow once we have shown that v2 puqvq, upuqvp, upvpvq ({p, q} = {1, 2}) all belong to V . By symmetry, we may assume p = 1, q = 2. Multiplying (2) by u2v1, we obtain u1u2v1v2 + u2 2v2 1 + s(v1v2)u2v1 + v2 1u2v2 = 0, forcing v2 V , so we have v2 v1, yields 1u2v2 ≡ u1u2v1v2 mod V . But multiplying (3) by v1v2 implies u1u2v1v2 ∈ 1u2v2 ∈ V as well. Moreover, multiplying (2) first by u1, then by u1u2v1 = u2 1v2 + s(v1v2)u1 + u1v1v2 ≡ v2 1u2 + s(v1v2)v1 + v2 1v2 ≡ 0 mod V. Hence also u1v1v2 = u2 1v2 + u1u2v1 + s(v1v2)u1 ∈ V , which completes the proof. (cid:3) 6.5. Corollary. For a Pfister quadratic form q over a field of characteristic 2 to be the norm of a conic division algebra it is necessary and sufficient that q be anisotropic. (cid:3) 6.6. Examples. Letting k be the field of rational functions in countably many variables t1, t2, t3, . . . over any field of characteristic 2, e.g., over F2, the (n + 1)- Pfister quadratic forms hht1, . . . , tn, tn+1K, n ≥ 0, by standard arguments are easily seen to be anisotropic, and we obtain central flexible conic division algebras over k in all dimensions 2n, n = 0, 1, 2, . . . . Part II. 2-Henselian base fields 7. Pointed quadratic spaces over 2-Henselian fields. In this section, we recast the conceptual foundations for the study of quadratic forms over Henselian fields in the setting of pointed quadratic spaces. Our subse- quent considerations also fit naturally into the valuation theory of Jordan division rings [40] when specialized to the Jordan algebras of pointed quadratic spaces over Henselian fields. 28 SKIP GARIBALDI AND HOLGER P. PETERSSON 7.1. Round quadratic forms. For the time being, we work over a field k that is completely arbitrary. We recall from [15, § 9A, p. 52] that a finite-dimensional quadratic form q over k is said to be round if all its non-zero values are precisely its similarity factors; in particular, they form a subgroup of k×. The most important examples of round quadratic forms are Pfister forms and quasi-Pfister forms [15, Cor. 9.9, Cor. 10.13]. 7.2. Pointed quadratic spaces over arbitrary fields. Adopting the terminol- ogy of Weiss [55, Def. 1.1], by a pointed quadratic space over k we mean a triple Q = (V, q, e) consisting of a finite-dimensional vector space V over k, a quadratic form q : V → k and a vector e ∈ V which is a base point for q in the sense that q(e) = 1. Morphisms of pointed quadratic spaces are isometries preserving base points. Q is said to be non-singular (resp. anisotropic, round, Pfister, . . . ) if q is. Given a conic algebra C over k, we obtain in QC := (C, nC , 1C) a pointed quadratic space and every pointed quadratic space arises in this manner (Loos [31], see also Rosemeier [44]). Notationally, we do not always distinguish carefully between C and QC. If Q = (V, q, e) is any pointed quadratic space over k, we therefore find it conve- nient to put VQ = V as a vector space over k and to call nQ := q the norm, 1Q := e the unit element , tQ := ∂nQ(1Q,−) the trace, ιQ : Q → Q, x 7→ x∗ := tQ(x)1Q − x the conjugation of Q. We also put V × When dealing with quadratic forms representing 1, insisting on pointedness is not so much a matter of necessity but one of convenience, making the language of non- associative algebras the natural mode of communication. Just as for composition algebras, Witt's theorem [15, Thm. 8.3] implies that pointed quadratic spaces are classified by their norms: Q := {x ∈ Q nQ(x) 6= 0}. 7.3. Proposition. For non-singular pointed quadratic spaces over k to be isomor- phic it is necessary and sufficient that their underlying quadratic forms be isometric. (cid:3) 7.4. Enlargements. Let P be a pointed quadratic space over k and µ ∈ k×. Writing VP ⊕ VP j for the direct sum of two copies of the vector space VP over k as in 1.10, and identifying VP ⊆ VP ⊕ VP j as a subspace through the first summand, Q := hhµii ⊗ P := (VQ, nQ, 1Q), 1Q := 1P with VQ := VP ⊕ VP j, nQ := hhµii ⊗ nP , (1) (cid:0)hhµii ⊗ nP(cid:1)(u + vj) = nP (u) − µnP (v) (u, v ∈ VP ) is again a pointed quadratic space over k whose trace and conjugation are given by the formulas (2) (3) tQ(u + vj) = tP (u), (u + vj)∗ = u∗ − vj for all u, v ∈ VP . Moreover, if P is round (resp. Pfister), so is Q. Finally, a comparison with the Cayley-Dickson construction 1.10 shows QCay(B,µ) = hhµii⊗QB for any conic k-algebra B and any µ ∈ k×. The following two statements are standard facts about round quadratic forms, translated into the setting of pointed quadratic spaces; we refer to [15, Prop. 9.8, Lemma 23.1] for details. WILD PFISTER FORMS 29 7.5. Proposition. Let P be a non-singular round pointed quadratic space over k. (a) For µ ∈ k×, the following conditions are equivalent. (i) hhµii ⊗ P is isotropic. (ii) µ ∈ nP (V × P ). (iii) hhµii ⊗ P is hyperbolic. (b) Let µ1, µ2 ∈ k×. Then hhµ1ii ⊗ P ∼= hhµ2ii ⊗ P if and only if µ1 = µ2nP (u) for some u ∈ P ×. 7.6. Proposition. (Embedding property) Let Q be a pointed Pfister quadratic space over k and P ⊂ Q a proper pointed Pfister quadratic subspace. Then the inclusion P ֒→ Q extends to an embedding from hhµii ⊗ P to Q, for some µ ∈ k×. (cid:3) (cid:3) 7.7. 2-Henselian fields. Let F be a field of arbitrary characteristic that is en- dowed with a normalized discrete valuation λ, so λ : F → Z∞ := Z ∪ {∞} is a surjective map satisfying the following conditions, for all α, β ∈ F . λ is definite: λ(α) = ∞ ⇐⇒ α = 0. λ is sub-additive: λ(α + β) ≥ min{λ(α), λ(β)}. λ is multiplicative: λ(αβ) = λ(α) + λ(β). As convenient references for the theory of valuations we mention [14, 17], and particularly [32] for the discrete case. We write o ⊆ F for the valuation ring of F relative to λ, p ⊆ o for its valuation ideal and ¯F := o/p for the residue field of F . The natural map from o to ¯F will always be indicated by α 7→ ¯α. Throughout the remainder of this paper, we fix a prime element π ∈ o. The quantity (1) which is either a non-negative integer or ∞, will play an important role in the sequel. If ¯F has characteristic 2, then eF > 0 agrees with what is usually called the absolute ramification index of F . eF := λ(2 · 1F ), Due to the quadratic character of the gadgets we are interested in (composition and conic algebras, pointed quadratic spaces), requiring F to be Henselian (with respect to λ) is too strong a condition. It actually suffices to assume that F be 2-Henselian in the sense of Dress [12] or [17, §4.2], i.e., that F satisfies the following two equivalent conditions [12, Satz 1]. (i) For any quadratic field extension K/F , there is a unique extension of λ to a discrete valuation λK of K taking values in Q∞ = Q ∪ {∞}. (ii) For all α0, α1, α2 ∈ o with α0 ∈ p, α1 /∈ p, the polynomial α0t2 + α1t + α2 ∈ F [t] is reducible. In this case, the extension λK of λ in (i) is given by (2) λK(u) = 1 2 λ(cid:0)NK/F (u)(cid:1) (u ∈ K). From now on, F is assumed to be a fixed 2-Henselian field with respect to a normalized discrete valuation λ. For simplicity, all algebras, quadratic forms etc. over F are assumed to be finite-dimensional. The characteristic of F is arbitrary. 30 SKIP GARIBALDI AND HOLGER P. PETERSSON 7.8. Quadratic forms over 2-Henselian fields. Let q : V → F be a quadratic form over F and suppose q is anisotropic. Following Springer [46] (in the case of a complete rather than Henselian valuation), λ(cid:0)q(x + y)(cid:1) ≥ min{λ(cid:0)q(x)(cid:1), λ(cid:0)q(y)(cid:1)} For convenience, we give the easy proof of this inequality. It evidently suffices to show (x, y ∈ V ). λ(cid:0)q(x, y)(cid:1) ≥ min{λ(cid:0)q(x)(cid:1), λ(cid:0)q(y)(cid:1)} for all x, y ∈ V . Suppose there are x, y ∈ V such that (2) does not hold. Then q(x, y) 6= 0, and the polynomial 1 q(x, y) q(tx + y) = q(x) q(x, y) t2 + t + q(y) q(x, y) ∈ F [t] has no zero in F since q is anisotropic, but is reducible by 7.7 (ii), a contradiction. (cid:3) Relation (2) may be strengthened to 2λ(cid:0)q(x, y)(cid:1) ≥ λ(cid:0)q(x)(cid:1) + λ(cid:0)q(y)(cid:1) (x, y ∈ V ), either by appealing to a general result of Bruhat-Tits [8, Thm. 10.1.15], or by using an ad-hoc argument from the valuation theory of Jordan rings [40] in disguise: for x, y ∈ V , x 6= 0, we obtain Qxy := q(x, y)x − q(x)y = −q(x)τx(y), where τx : V → V is the reflection in the hyperplane perpendicular to x. In par- ticular, τx leaves q invariant, which implies λ(q(Qxy)) = 2λ(q(x)) + λ(q(y)), and since q(x, y)x = Qxy + q(x)y, we obtain (3). 7.9. Valuation data for pointed quadratic spaces. For the rest of this section, we fix a pointed quadratic space Q over F which is round and anisotropic. (a) The map (1) λQ : VQ −→ Q∞, x 7−→ λQ(x) := 1 2 λ(cid:0)nQ(x)(cid:1), is a norm of VQ as an F -vector space in the sense of Bruhat-Tits [9, 1.1], that is, the following relations hold for all α ∈ F , x, y ∈ VQ. (2) λQ is definite: λQ(x) = ∞ ⇐⇒ x = 0, λQ is sub-additive: λQ(x + y) ≥ min{λQ(x), λQ(y)}, λQ is scalar-compatible: λQ(αx) = λ(α) + λQ(x), (1) (2) (3) (3) (4) (5) (6) (7) where (3) is a consequence of (7.8.1). Moreover, λQ(x∗) = λQ(x), λ(cid:0)nQ(x, y)(cid:1) ≥ λQ(x) + λQ(y), λ(cid:0)tQ(x)(cid:1) ≥ λQ(x), for all x, y ∈ VQ, where (5) follows from conjugation invariance of nQ, (6) from (7.8.3), and (7) from (6) for y = 1Q. WILD PFISTER FORMS 31 (b) nQ being round, ΓQ := λQ(V × implies Q ) is an additive subgroup of Q for which (1) (8) Z ⊆ ΓQ ⊆ 1 2 Z, eQ/F := [ΓQ : Z] ∈ {1, 2}, ΓQ = 1 eQ/F Z. We call eQ/F the ramification index of Q. (c) We put (9) (10) pQ := {x ∈ VQ λQ(x) > 0} = {x ∈ VQ λQ(x) ≥ oQ := {x ∈ VQ λQ(x) ≥ 0}, 1 eQ/F } ⊆ oQ, which are both full o-lattices in VQ, and (11) o× Q := oQ \ pQ = {x ∈ VQ λQ(x) = 0}, which is just a subset of oQ containing 1Q. By abuse of language, the elements of Q are called units of oQ. An element Π ∈ V × o× Q such that λQ(Π) > 0 generates the infinite cyclic group ΓQ belongs to pQ and is called a prime element of oQ. Writing x 7→ ¯x for the natural map from oQ to oQ/pQ and setting ¯Q := (V ¯Q, n ¯Q, 1 ¯Q), V ¯Q := oQ/pQ, 1 ¯Q := 1Q, where n ¯Q : ¯Q −→ ¯F , ¯x 7−→ n ¯Q(¯x) := nQ(x) is the first residue form of nQ, we obtain a pointed quadratic space over ¯F , called the pointed quadratic residue space of Q, which is round and anisotropic. Here only roundness of n ¯Q demands a proof, so let u ∈ o× Q. Since nQ is round, there exists a linear bijection f : VQ → VQ with nQ(f (x)) = nQ(u)nQ(x) for all x ∈ VQ. Hence f stabilizes oQ as well as pQ and thus canonically induces a similarity transformation V ¯Q → V ¯Q relative to n ¯Q with multiplier n ¯Q(¯u). We call (12) fQ/F := dim ¯F (V ¯Q) the residue degree of Q. For convenience, we collect some of the above properties of ¯Q in the following proposition, which is really stating the obvious. 7.10. Proposition. Norm, trace and conjugation of the pointed quadratic residue space ¯Q of Q are given by the formulas (1) (2) n ¯Q(¯x) = nQ(x), t ¯Q(¯x) = tQ(x), (3) for all x ∈ oQ. Moreover, n ¯Q is round and anisotropic. Here is the easiest example that is not totally trivial. ¯x∗ = x∗ (cid:3) 7.11. Example. Let Q = (V, q, e) be such that V has basis e, j and q(ue + vj) = u2 − µv2 for some µ ∈ o× and all u, v ∈ F . (This is the case P = h1i of 7.4; q ∼= hhµii.) If ¯µ is not a square in F , then oQ and pQ are the o- and p-spans of e, j respectively and the pointed quadratic residue space of Q is naturally identified with the quadratic extension F (√¯µ) with quadratic form the squaring map z 7→ z2 and base point 1 ∈ F . 32 SKIP GARIBALDI AND HOLGER P. PETERSSON 7.12. Lemma. The map 0 ≤ λQ(y) ≤ 1 − 1 eQ/F (cid:27) −→ V × Q , (m, y) 7−→ πmy, Q (cid:12)(cid:12)(cid:12)(cid:12) Z ×(cid:26)y ∈ V × . is surjective. Proof. For x ∈ V × meQ/F + r, m, r ∈ Z, 0 ≤ r ≤ eQ/F − 1, the element y := π−mx ∈ V × 0 ≤ λQ(y) ≤ 1 − 1 Q , some n ∈ Z has λQ(x) = n by (7.9.8), and writing n = Q satisfies eQ/F eQ/F (cid:3) 7.13. Proposition. eQ/F fQ/F = dimF (VQ). If eQ/F = 1, then pQ = πoQ, and the dimension of V ¯Q = oQ ⊗o ¯F over ¯F Proof. agrees with the dimension of VQ over F . We may therefore assume eQ/F = 2. Let Π be a prime element of oQ and f : VQ → VQ a norm similarity with multiplier nQ(Π). Then λQ(f (x)) = 1 Q , which implies f (oQ) = pQ, f (pQ) = πoQ, and f induces canonically an ¯F -linear bijection from V ¯Q onto pQ/πoQ. Combined with the filtration oQ ⊗o ¯F = oQ/πoQ ⊃ pQ/πoQ ⊃ {0} and the isomorphism (oQ/πoQ)/(pQ/πoQ) ∼= V ¯Q as vector spaces over ¯F , this implies dimF (VQ) = 2dim ¯F (V ¯Q), as desired. Remark. Prop. 7.13 becomes false if Q is not assumed to be round, for example if eQ/F = 2 and dimF (VQ) is odd, see [40, Satz 6.3] for generalization. 2 + λQ(x) for all x ∈ V × (cid:3) 7.14. Connecting with conic algebras. Let C be a finite-dimensional conic al- gebra over F and suppose its norm is round and anisotropic; for example, C could itself be a composition algebra, or it could arise from a composition algebra by means of the Cayley-Dickson process. Applying as we may the preceding consider- ations to QC, the pointed quadratic space associated with C via 7.2, we systemat- ically adhere to the following convention: all notation and terminology developed up to now and later on for pointed quadratic spaces will be applied without further comment to C in place of QC , modifying subscripts accordingly whenever possible. For example, (1) λC : C −→ Q∞, x 7−→ λC (x) := 1 2 λ(cid:0)nC (x)(cid:1), is a norm of C as an F -vector space, eC/F := eQC /F is the ramification index , fC/F := fQC /F is the residue degree and ¯C := QC is the pointed quadratic residue space of C. But note that ¯C in general is not a conic algebra over ¯F in a natural way unless C is a composition algebra; see 7.15 and Section 10 below for further discussion. 7.15. Tame and wild pointed quadratic spaces. If t ¯Q is non-zero, then Q is said to be tame. Otherwise, i.e., if t ¯Q = 0, then Q is said to be wild. For Q to be wild it is clearly necessary that ¯F have characteristic 2. Applying [41, Prop. 1], and bearing in mind the conventions of 7.14, a composition division algebra C over F is tame (resp. wild) iff ¯C is a composition algebra (resp. a purely inseparable field extension of exponent at most 1) over ¯F . Extending the terminology of [41] to the present more general set-up, we call Q unramified (resp. ramified ) if Q is tame with eQ/F = 1 (resp. eQ/F = 2). For Q to be unramified it is necessary and WILD PFISTER FORMS 33 sufficient that the quadratic form nQ have good reduction with respect to λ in the sense of Knebusch [27, 28]. This will follow from Prop. 8.2 (c) below. The preceding definitions seem to assign a distinguished role to the base point of a pointed quadratic space. But this is not so as will be seen in Prop. 8.2 below. 8. Trace and norm exponent. This section serves a double purpose. Working with a fixed non-singular, round and anisotropic pointed quadratic space Q over a 2-Henselian field F , we attach wildness-detecting invariants to Q. Moreover, we present a device measuring how far a scalar in o× is removed from being the norm of an appropriate element in VQ. 8.1. Trace ideal and trace exponent. Since oQ ⊆ VQ is a full o-lattice, its image under the trace of Q by non-singularity is a non-zero ideal in o, called the trace ideal of Q. But o is a discrete valuation ring, so there is a unique integer texp(Q) ≥ 0 such that (1) tQ(oQ) = ptexp(Q). We call texp(Q) the trace exponent of Q and have (2) texp(Q) = min{λ(cid:0)tQ(x)(cid:1) x ∈ oQ}. The image of oQ ⊗o oQ under the linear map x ⊗ y 7→ nQ(x, y) is an ideal in o denoted by ∂nQ(oQ ⊗o oQ). The following result relates the ideals just defined to one another but also to wild and tame pointed quadratic spaces. 8.2. Proposition. (a) Q is wild if and only if texp(Q) > 0. (b) If P ⊆ Q is pointed quadratic subspace that is round and non-singular, then texp(P ) ≥ texp(Q). (c) tQ(oQ) = ∂nQ(oQ ⊗o oQ). (d) Q is tame if and only if ¯Q is non-singular. Proof. (a) and (b) are obvious. Before proving (c),(d), we let x ∈ Q× and, by using a Jordan isotopy argument in disguise, pass to Qx := (VQ, nQ(x)−1nQ, x), which is a non-singular quadratic space over F . Since nQ is round, the norms of Q and Qx are isometric, forcing Q and Qx to be isomorphic as pointed quadratic spaces by Prop. 7.3. This implies (1) tQ(oQ) = tQx (oQx) = (cid:8)nQ(cid:0)x, nQ(x)−1y(cid:1) y ∈ VQ, λQ(y) ≥ λQ(x)(cid:9) = {nQ(x, y) y ∈ VQ, λQ(y) ≥ −λQ(x)}. (c) The left-hand side is clearly contained in the right, so it suffices to show nQ(x, y) ∈ tQ(oQ) for all x, y ∈ oQ, x 6= 0. But this follows from (1) and λQ(x) ≥ 0. (d) Non-singularity of ¯Q is clearly sufficient for Q to be tame. Conversely, sup- pose Q is tame and let x ∈ o× Q. Then (1) produces an element y ∈ oQ with nQ(x, y) = 1 ∈ tQ(oQ), hence n ¯Q(¯x, ¯y) 6= 0. (cid:3) 34 SKIP GARIBALDI AND HOLGER P. PETERSSON 8.3. Trace generators. An element w0 ∈ oQ where the minimum in (8.1.2) is attained, i.e., with λ(tQ(w0)) = texp(Q) is called a trace generator of Q. If even tQ(w0) = πtexp(Q), we speak of a normalized trace generator, dependence on π being understood. Trace generators always exist (as do normalized ones) and their traces generate the trace ideal of C. Moreover, they satisfy the inequalities (1) 0 ≤ λQ(w0) ≤ 1 − 1 eQ/F . Indeed, assuming λQ(w0) > 1 − 1 eQ/F implies λQ(π−1w0) = λQ(w0) − 1 > − 1 eQ/F , 1 eQ/F Z, we conclude π−1w0 ∈ oQ. But now and since λQ(π−1w0) belongs to tQ(π−1w0) = πtexp(Q)−1 /∈ ptexp(Q) leads to a contradiction. 8.4. Tignol's invariant ω(C). Another invariant that fits into the present set-up is due to Tignol [51, pp. 9,17] for separable field extensions and central associative division algebras of degree p over Henselian fields (not necessarily discrete) having residual characteristic p > 0. As a straightforward adaptation of Tignol's definition to the situation we are interested in, we define (1) ω(Q) := min{λ(cid:0)tQ(x)(cid:1) − λQ(x) x ∈ V × Q }; thanks to (7.9.7), it is a non-negative rational number. Moreover, it is closely related to the trace exponent, as the following proposition shows. 8.5. Proposition. (a) ω(Q) = texp(Q) or ω(Q) = texp(Q) − 1 2 . (b) There exists a trace generator w0 of Q with (1) ω(Q) = texp(Q) − λQ(w0). (c) If eQ/F = 1, then ω(Q) = texp(Q). Proof. Since the map ϕ : V × Q −→ ΓQ = 1 eQ/F Z, x 7−→ ϕ(x) := λ(cid:0)tQ(x)(cid:1) − λQ(x), is homogeneous of degree zero, so ϕ(αx) = ϕ(x) for all α ∈ F ×, x ∈ V × Q , Lemma 7.12 implies ω(Q) = min{ϕ(x) x ∈ S}, where Q 0 ≤ λQ(x) ≤ 1 − S := {x ∈ V × 1 eQ/F }. Accordingly, let w0 ∈ S satisfy (2) ϕ(w0) = ω(Q). Given any trace generator w′ 0 of Q, the chain of inequalities (3) 0 ≤ λQ(w′ 0) = λ(cid:0)tQ(w′ = λQ(w0) ≤ 1 − 0)(cid:1) − ϕ(w′ 1 eQ/F 0) ≤ texp(Q) − ω(Q) ≤ λ(cid:0)tQ(w0)(cid:1) − ϕ(w0) WILD PFISTER FORMS 35 implies (a) and (c), while in (b) we may assume eQ/F = 2 since otherwise (1) holds for w′ 2 Z, we either have λQ(w′ 0, i.e., for any trace generator of Q. Since all quantities in (3) belong to 1 0) = texp(Q) − ω(Q) or texp(Q) − ω(Q) = λ(cid:0)tQ(w0)(cid:1) − ϕ(w0) = λQ(w0) = 1 2 . In the latter case, (2) shows that w0 is a trace generator of Q, and the proof of (b) is complete. (cid:3) 8.6. Regular trace generators. Trace generators of Q satisfying (8.5.1) are called regular . If Q has ramification index 1, then every trace generator by (8.3.1) and Prop. 8.5 (c) is regular but, in general, this need not be so, cf. Cor. 11.7 and Thm. 12.3 below. 8.7. Example. If F has characteristic not 2, it is a composition division algebra over itself, with norm and trace given by nF (α) = α2, tF (α) = 2α for all α ∈ F . Hence (7.7.1), (8.1.1) and (8.4.1) imply (1) texp(F ) = ω(F ) = eF , and w0 = πeF regular. From (1) and Props. 8.2 (c), 8.5 we now conclude 2 ∈ o× is the unique normalized trace generator of F ; it is obviously (2) 0 ≤ ω(Q) ≤ texp(Q) ≤ eF , which for trivial reasons also holds in characteristic 2. 8.8. The norm exponent. We wish to measure how far a given unit in the valua- tion ring of F is removed from being the norm of an element in VQ or, equivalently, in o× Q. To this end, we observe that, given α ∈ o× and an integer d ≥ 0, the following conditions are equivalent. (i) α is a norm of o× Q mod pd, i.e., there exists an element v ∈ o× Q with (ii) There exist elements β ∈ o, v ∈ o× α − nQ(v) ∈ pd. Q with α = (1 − πdβ)nQ(v). Thus the set NQ(α) of non-negative integers d satisfying (i)/(ii) above contains 0, and d ∈ NQ(α) implies d′ ∈ NQ(α) for all integers d′, 0 ≤ d′ ≤ d. We therefore put (1) nexpQ(α) := sup NQ(α) and call this the norm exponent of α relative to Q; it is either a non-negative integer or ∞. Roughly speaking, the bigger the norm exponent becomes, the closer α gets to being a norm of Q. More precisely, 2− nexpQ(α) is the (minimum) distance of α from the subset nQ(o× Q) ⊆ F × relative to the metric induced by the absolute value ξ 7→ ξ := 2−λ(ξ). A number of useful elementary properties of the norm exponent are collected in the following proposition, whose straightforward proof is left to the reader. 36 SKIP GARIBALDI AND HOLGER P. PETERSSON 8.9. Proposition. With the notations and assumptions of 8.8, let α, α′ ∈ o×. (a) If α ∈ nQ(o× (b) If d = nexpQ(α) is finite, then α can be written in the form Q), then nexpQ(α) = ∞. v ∈ o× Q, α = (1 − πdβ)nQ(v), β ∈ o, and every such representation of α satisfies β ∈ o×. (c) nexpQ(α) = 0 if and only if ¯α /∈ n ¯Q(V × ¯Q ). (d) α ≡ α′ mod nQ(V × (e) nexpP (α) ≤ nexpQ(α) for any round non-singular pointed quadratic subspace P ⊆ Q. Q ) implies nexpQ(α) = nexpQ(α′). (cid:3) By Prop. 8.9 (a), the elements of nQ(o× Q) have infinite norm exponent. While the converse is also true, we can do better than that by showing that the norm exponents of elements in o× \ nQ(o× Q) are uniformly bounded from above. Indeed, we have the following result. 8.10. Local Norm Theorem. Let α ∈ o× \ nQ(o× nexpQ(α) ≤ 2 ω(Q). Q) = o× \ nQ(V × Q ). Then More precisely, letting w0 ∈ oQ be a normalized regular trace generator of Q, every β ∈ o admits a γ ∈ o with (1) 1 − π2 ω(Q)+1β = nQ(cid:0)1Q + πtexp(Q)γw0(cid:1). Proof. Arguing indirectly, and using roundness of nQ, the first part of the theorem follows from the second. To establish the second part, we apply Prop. 8.5 (a),(b) and obtain Since F is 2-Henselian, the polynomial 0 ≤ d := λ(cid:0)nQ(w0)(cid:1) = 2λQ(w0) = 2(cid:0) texp(Q) − ω(Q)(cid:1) ≤ 1. g := π1−dnQ(w0)t2 + t + β ∈ o[t] ⊆ F [t] by 7.7 (ii) is reducible, and the two roots δ, δ′ ∈ F of πd−1nQ(w0)−1g satisfy the relation δδ′ = πd−1nQ(w0)−1β ∈ p−1. Thus we may assume δ ∈ o and have π1−dnQ(w0)δ2 + δ + β = 0. Setting γ := π1−dδ ∈ o, we therefore obtain nQ(cid:0)1 + πtexp(Q)γw0(cid:1) = 1 + πtexp(Q)+1−dδtQ(w0) + π2 texp(Q)π2(1−d)δ2nQ(w0) = 1 + π2 texp(Q)−d+1(cid:0)δ + π1−dnQ(w0)δ2(cid:1) = 1 − π2 ω(Q)+1β. (cid:3) For a version of this result addressed to central associative division algebras of degree p = char( ¯F ) > 0, see Kato [23, Prop. 2 (iii)]. 8.11. Corollary. Let µ, µ′ ∈ o×. Then µ ≡ µ′ mod nQ(Q×) ⇐⇒ ∃ x ∈ o× Q : µ ≡ µ′nQ(x) mod p2 ω(Q)+1. (cid:3) WILD PFISTER FORMS 37 8.12. The connection with the quadratic defect. We assume char(F ) 6= 2 and return to the composition division algebra C = F of 8.7. Comparing 8.8 with [35, §63A], we see that pnexpF (α) is the quadratic defect of α ∈ o×. Moreover, texp(F ) = ω(F ) = eF , and w0 := πeF 2 ∈ o× is the unique normalized regular trace π2eF β′, generator of F . In particular, γ = 4 π2eF ∈ o×, and the change of variables β = − 4 π2eF γ′ converts (8.10.1) into the relation 4 1 + 4πβ′ = (1 + 2πγ′)2. Hence the local norm theorem becomes the local square theorem of [35, 63:1] or [30, VI.2.19] in the special case Q = QF . See also [10, Prop. 4.1.2] for an extension of this result to residual characteristics other than 2. The local norm theorem has numerous applications. One of these can already be given in this section; a useful technical lemma prepares the way. 8.13. Lemma. Suppose Q is wild and has ramification index eQ/F = 1. For d ∈ Z, 0 ≤ d ≤ 2 texp(Q) = 2 ω(Q) and β ∈ o×, the following conditions are equivalent. (i) 1 − πdβ ∈ nQ(o× Q). (ii) 1 − πdβ ∈ o× and there are elements m ∈ Z, w ∈ o× β = −nQ(w) + π−mtQ(w). d = 2m, Q with (1) Proof. (i) =⇒ (ii). There exists an element v ∈ o× Q with 1− πdβ = nQ(v) ∈ o×. For d = 0 we put w = 1Q− v ∈ oQ and obtain 1− β = nQ(1Q− w) = 1− tQ(w)+ nQ(w), hence β = −nQ(w) + tQ(w). But tQ(w) ∈ ptexp(Q) ⊆ p since Q is wild, forcing w ∈ o× Q. Thus (ii) holds with m = 0. We may therefore assume d > 0. Then 1 = nQ(v) = n ¯Q(¯v), forcing ¯v = 1 since n ¯Q is anisotropic and n ¯Q(1 ¯Q − ¯v) = 0 by wildness of Q. Combining with eQ/F = 1, we find an integer m > 0 and a unit w ∈ o× Q with v = 1Q−πmw. Expanding the right-hand side of 1−πdβ = nQ(1Q−πmw), we conclude (2) πdβ = −π2mnQ(w) + πmtQ(w), which in turn yields the estimate 2 texp(Q) ≥ d = λ(πdβ) = λ(cid:0) − π2mnQ(w) + πmtQ(w)(cid:1) ≥ min{2m, m + texp(Q)}. This implies (3) and (2) attains the form m ≤ texp(Q), d ≥ 2m, (4) πdβ = −π2mnQ(w)(1 − γ), γ := π−mnQ(w)−1tQ(w) ∈ ptexp(Q)−m. If m < texp(Q), then (1) follows from (2),(4). On the other hand, if m = texp(Q), then (3) implies d = 2 texp(Q) = 2m, and (1) follows from (2). (ii) =⇒ (i). Setting v := 1Q − πmw ∈ oQ and applying (1), we conclude nQ(v) = 1 − πdβ ∈ o×, hence v ∈ o× Remark. For d > 0, the condition 1 − πdβ ∈ o× in (ii) is of course automatic. Q, and (i) holds. (cid:3) 38 SKIP GARIBALDI AND HOLGER P. PETERSSON 8.14. Proposition. Let P be a pointed quadratic space over F that is non-singular, round and anisotropic with eP/F = 1, and let d be an odd integer with 0 ≤ d < 2 texp(P ). (a) If β ∈ o, then µ := 1 − πdβ ∈ o× and d = nexpP (µ) ⇐⇒ β ∈ o×. (b) If βi ∈ o× and µi := 1 − πdβi for i = 1, 2, then hhµ1ii ⊗ P ∼= hhµ2ii ⊗ P =⇒ β1 = β2. Proof. β ∈ o×. Then µ /∈ nP (V × implies d ≤ d′ representation µ = µ′nP (v′), µ′ = 1 − πd′ d < d′ would imply that µµ′ = 1 − πdγ, γ := β + πd′−dβ′ − πd′ belong to nP (V × (a) d = nexpP (µ) implies β ∈ o× by Prop. 8.9 (b). Conversely, suppose P ) by Lemma 8.13. Thus the local norm theorem 8.10 := nexpP (µ) ≤ 2 texp(P ), and from Prop. 8.9 (b) we obtain a P . Assuming ββ′ ∈ o×, does not (b) Arguing indirectly, let us assume β1 6= β2. Then µ1µ2 = 1 − πdβ, where β = β1 + β2− πdβ1β2 ∈ o satisfies ¯β = β1 − β2 6= 0, hence β ∈ o×. By (a), the norm exponent of µ1µ2 relative to P is d, forcing µ1µ2 /∈ nP (V × P ) (this also follows from Lemma 8.13). Hence hhµ1ii ⊗ P and hhµ2ii ⊗ P are not isomorphic by Prop. 7.5 (b). P ), again by Lemma 8.13, in contradiction to µµ′ = nP (µ′v′). β′ for some β′ ∈ o×, v′ ∈ o× (cid:3) We will see in Example 9.11 (b) below that the converse of Prop. 8.14 (b) does not hold. 8.15. Lemma. Let β ∈ o, w ∈ oQ and suppose d, m are non-negative integers. Determine α ∈ o by tQ(w) = πtexp(Q)α. Then (1 − πdβ)nQ(1Q − πmw) = 1 − πdβ + π2mnQ(w) − πtexp(Q)+mα + πtexp(Q)+d+mαβ − πd+2mβnQ(w). Proof. Expand the left-hand side in the obvious way. (cid:3) 9. Valuation data under enlargements. In this section we will be concerned with the question of what happens to the valuation data ramification index (7.9 (b)), pointed quadratic residue space (7.9 (c)) and trace exponent (8.1) when passing from a pointed quadratic space P to hhµii⊗P , µ ∈ F ×. We will answer this question not in full generality but only under the additional hypothesis that P have ramification index 1. This hypothesis derives its justification from the fact that, if F has characteristic zero, every anisotropic pointed (n+1)-Pfister quadratic space over F contains a pointed n-Pfister quadratic subspace of ramification index 1. We will prove this in Prop. 17.2 and Thm. 19.2(i) below by using methods from algebraic K-theory. It would be interesting to know whether the result in question also holds for F having characteristic 2. 9.1. The general set-up. (a) We fix a 2-Henselian field F and a pointed quadratic space P over F which is non-singular, round and anisotropic. We also assume throughout that P has ramification index eP/F = 1, which implies (1) ω(P ) = texp(P ) WILD PFISTER FORMS 39 by Prop. 8.5 (c) and (2) ΓP = λP (P ×) = Z, λ(cid:0)nP (P ×)(cid:1) = 2Z by 7.9 (b) and (7.9.1). (b) We are interested in pointed quadratic spaces Q = hhµii ⊗ P , µ ∈ F ×, as in 7.4; in particular, we recall VQ = VP +VP j as vector spaces over F . By Prop. 7.5 (b), we may and always will assume λ(µ) ∈ {0, 1}, so µ is either a unit or a prime element in o. There are two harmless cases which we treat first. One of them arises when µ is a prime, the other when µ is a unit and P is tame. 9.2. Proposition. If µ is a prime element in o, then Q := hhµii ⊗ P is a non- singular, round and anisotropic pointed quadratic space over F with (1) (2) (3) λQ(u + vj) = min{λP (u), λP (v) + oQ = oP ⊕ oP j, pQ = pP ⊕ oP j, ¯Q = ¯P , texp(Q) = texp(P ). eQ/F = 2, 1 2} (u, v ∈ VP ), Proof. From (9.1.2) we conclude µ /∈ nP (P ×). Thus Q is not only round and non-singular but also anisotropic, so λQ satisfying (7.9.1)−(7.9.4) exists. Since λQ(vj) = λP (v) + 1 P , by (7.4.1),(7.9.1) and again by (9.1.2), we obtain (1), hence (2) and the first two relations of (3), while the last one is immediately implied by (7.4.2). (cid:3) 2 6= λP (u) for all u, v ∈ V × 9.3. Proposition. If P as in 9.1 (a) is tame and µ ∈ o×, then Q = hhµii ⊗ P is a non-singular and round pointed quadratic space over F . Moreover the following conditions are equivalent. (i) Q is anisotropic. (ii) µ /∈ nP (V × P ). (iii) ¯µ /∈ n ¯P (V × ¯P ). In this case, (1) (u, v ∈ P ), λQ(u + vj) = min{λP (u), λP (v)} oQ = oP ⊕ oP j, pQ = pP ⊕ pP j, (2) and Q is tame of ramification index eQ/F = 1 with ¯Q ∼= hh¯µii ⊗ ¯P . Proof. While the first statement is obvious, the equivalence of (i),(ii),(iii) fol- lows from Prop. 7.5 (a) and Cor. 8.11 since ω(P ) = texp(P ) = 0 by (9.1.1) and tameness of P . If (i),(ii),(iii) hold, then λQ exists and it suffices to show that (1) holds. Since λQ(vj) = λP (v) by (7.4.1),(7.9.1), we certainly have λQ(u + vj) ≥ min{λP (u), λP (v)}. To prove equality, Lemma 7.12 allows us to assume λP (u) = λP (v) = 0. Here λQ(u + vj) > 0 would imply n ¯P (¯u) = ¯µn ¯P (¯v), hence ¯µ ∈ n ¯P ( ¯P ×) since n ¯P is round, and we obtain a contradiction to (iii). (cid:3) The remaining cases where P is wild and µ ∈ o is a unit are much more troublesome. 40 SKIP GARIBALDI AND HOLGER P. PETERSSON 9.4. Some easy reductions. For the rest of this section, we assume that P as given in 9.1 (a) is wild, so ω(P ) = texp(P ) > 0 by Prop. 8.2 (a). We are interested in the pointed quadratic spaces hhµii ⊗ P , µ ∈ o×, only when they are anisotropic. By Prop. 7.5 (a) and the local norm theorem 8.10, this is equivalent to µ having norm exponent ≤ 2 texp(P ), so by Prop. 8.9 (b) it will be enough to consider units in o having the form µ = (1 − πdβ)nP (v), where d ∈ Z satisfies 0 ≤ d ≤ 2 texp(P ), β ∈ o× and v ∈ o× P . Here Prop. 7.5 (b) allows us to assume v = 1P . We are thus reduced to working with scalars µ that may be written as (1) Setting (2) µ = 1 − πdβ, d ∈ Z, 0 ≤ d ≤ 2 texp(P ), β ∈ o×. 0 ≤ m :=(cid:22) d 2(cid:23) ≤ texp(P ), Θd := π−m(1P + j) ∈ VQ = VP + VP j, we define, (1.10.1)), inspired by the Cayley-Dickson construction of conic algebras (cf. (3) vΘd := π−m(v + vj) (v ∈ VP ), VP Θd := {vΘd v ∈ VP} and obtain after a straightforward computation, involving (7.4.1),(7.4.2), (4) (5) (6) VQ = VP ⊕ VP Θd, nQ(u + vΘd) = nP (u) + π−mnP (u, v) + πd−2mβnP (v), tQ(u + vΘd) = tP (u) + π−mtP (v) for all v ∈ VP . 9.5. Pointed quadratic residue spaces and inseparable extensions. (cf. [15, Remark 10.4]) For P as in 9.1,9.4, we claim that V ¯P carries a unique structure of a purely inseparable extension field over ¯F having exponent at most 1 such that n ¯P (u′) = u′2 for all u′ ∈ V ¯P . To see this, it suffices to note that n ¯P is round and anisotropic with ∂n ¯P = 0, making n ¯P (V ¯P ) a subfield of ¯F , so an ¯F -bilinear multiplication V ¯P × V ¯P → V ¯P , (u′, v′) 7→ u′v′, gives a purely inseparable extension field structure as indicated iff n ¯P (u′v′) = n ¯P (u′)n ¯P (v′) for all u′, v′ ∈ V ¯P . The purely inseparable extension field thus constructed will again be denoted by V ¯P if there is no danger of confusion. 9.6. Proposition. Let d be an odd integer with 0 ≤ d ≤ 2 texp(P ) and β ∈ o×. Then µ := 1 − πdβ ∈ o× and Q := hhµii⊗ P is a non-singular, round and anisotropic pointed quadratic space over F . Moreover, Q is wild and (1) Π := Θd = π− d−1 2 (1B + j) ∈ VQ WILD PFISTER FORMS 41 is a prime element of oQ with (2) (3) (4) (5) nQ(Π) = πβ, tQ(Π) = 2π− d−1 2 , λQ(u + vΠ) = min{λP (u), λP (v) + VQ = VP ⊕ VP Π, 1 2} (u, v ∈ VP ), oQ = oP ⊕ oP Π, pQ = pP ⊕ oP Π, texp(Q) = texp(P ) − d − 1 2 . eQ/F = 2, ¯Q = ¯P , Proof. By Prop. 8.14 (a), nexpP (µ) = d is finite, forcing Q to be anisotropic. Applying (9.4.4)−(9.4.6), we obtain (2); in particular, λQ(Π) = 1 2 , so Π is a prime element of oQ and the first formula of (5) holds. We proceed to establish (3). If u, v ∈ V × P , then λQ(u) = λP (u) is an integer by (9.1.2), while λQ(vΠ) = λP (v) + 1 2 is not. This not only proves (3) but also (4) and the second formula of (5). The last one follows from the fact that (9.4.6) establishes ptexp(P ) + ptexp(P )− d−1 2 = ptexp(P )− d−1 (cid:3) 2 as the trace ideal of Q. 9.7. Proposition. Let d be an even integer with 0 ≤ d < 2 texp(P ) and suppose β ∈ o satisfies the condition ¯β /∈ V 2 (1) ¯P (cf. 9.5). Then µ = 1 − πdβ ∈ o× and Q := hhµii⊗ P is a non-singular, round and anisotropic pointed quadratic space over F . Moreover, Q is wild and (2) is a unit of oQ with Ξ := Θd = π− d 2 (1P + j) ∈ VQ (3) (4) (5) (6) nQ(Ξ) = β, tQ(Ξ) = 2π− d 2 , VQ = VP ⊕ VP Ξ, λQ(u + vΞ) = min {λP (u), λP (v)} oQ = oP ⊕ oP Ξ, pQ = pP ⊕ pP Ξ, ¯Q = hh ¯βii ⊗ ¯P , texp(Q) = texp(P ) − d 2 . eQ/F = 1, (u, v ∈ VP ), ¯P implies ¯µ = 1 ¯F − ¯β /∈ V 2 Proof. The assertion µ ∈ o× is trivial for d > 0 but holds also for d = 0 since in this case ¯β /∈ V 2 ¯P by wildness of P . Next we show that Q is anisotropic. Otherwise, µ ∈ nP (VP ) by Prop. 7.5 (a), and Lemma 8.13 yields an element w ∈ o× 2 , where the second summand belongs to ptexp(P )−m ⊆ p by the hypothesis on d. Thus ¯β = n ¯P ( ¯w) = ¯w2, a contradiction, and we have proved that Q is indeed anisotropic. Consulting (9.4.4- −6) for u = 0, v = 1P , we end up with (3). Turning to (4), it suffices to show, by Lemma 7.12, that u, v ∈ o× λQ(u + vΞ) = P with β = −nP (w) + π−mtP (w), m = d P implies u + vΞ ∈ o× Q. Otherwise, observing (9.4.5), 1 2 λ(cid:0)nP (u) + π−mnP (u, v) + βnP (v)(cid:1) were strictly positive, and since π−mnP (u, v) ∈ p by Prop. 8.2 (c), we would again arrive at the contradiction ¯β ∈ V 2 ¯P . Thus (4) holds, which directly implies (5), while (6) follows from (5) and (9.4.5,6). (cid:3) 42 SKIP GARIBALDI AND HOLGER P. PETERSSON In 9.4, particularly (9.4.1), we are left with the case d = 2 texp(P ), which turns out to be the most delicate. In order to get started, we require the following elementary but crucial observations. 9.8. Setting the stage for the case d = 2 texp(P ). (a) Let K/k be a purely inseparable field extension of characteristic 2, exponent at most 1 and finite degree. Consider a scalar α ∈ k and a unital linear form s : K → k. We denote by by QK;α,s the pointed quadratic space over k with norm the Pfister quadratic form qK;α,s of 3.4 and with base point 1K ∈ K ⊆ K ⊕ Kj. Recall from Prop. 4.4 that QK;α,s = QCay(K;α,s) is the pointed quadratic space corresponding to the flexible conic algebra Cay(K; α, s) that arises form K, α, s by means of the non-orthogonal Cayley-Dickson construction. (b) Put m := texp(P ) and let w0 be a normalized trace generator of P . Then w0 ∈ o× P by (8.3.1) and the map sw0 : VP → F , u 7→ π−mnP (u, w0) is a unital linear form with sw0(oP ) ⊆ o, sw0 (pP ) ⊆ p by Prop. 8.2 (c) and since pP = poP . Thus we obtain an induced unital linear form ¯sw0 : V ¯P −→ ¯F , (1) and given any α′ ∈ ¯F , the notational conventions of (a) apply when V ¯P / ¯F is viewed via 9.5 as a purely inseparable field extension of exponent at most 1. ¯u 7−→ ¯sw0(¯u) = π−mnP (u, w0), 9.9. Theorem. With the notations of 9.8, let (1) β ∈ o, µ := 1 − π2 texp(P )β ∈ o×, β0 := nP (w0)β ∈ o, Q := hhµii ⊗ P. Then the following conditions are equivalent. (i) Q is anisotropic and unramified. (ii) Q is anisotropic and tame. (iii) Q is anisotropic. (iv) nexpP (µ) = 2 texp(P ). Moreover, if P is a pointed Pfister quadratic space, then these conditions are also equivalent to (v) β0 /∈ Im(℘V ¯P ,¯sw0 ). Proof. In this section, we will not be able to give the proof in full but must restrict ourselves to showing the equivalence of (i)−(iv), relegating the rest to the next section. Since the implications (i) ⇒ (ii) ⇒ (iii) are obvious, it suffices to show (iv) ⇔ (iii) ⇒ (i). (iv) ⇐⇒ (iii). From 8.8 we deduce nexpP (µ) ≥ 2 texp(P ). Hence (iv) holds iff nexpP (µ) ≤ 2 texp(P ) iff Q is anisotropic by the local norm theorem 8.10 and Prop. 8.9 (a). (iii) =⇒ (i). Setting d := 2 texp(P ), we apply (9.4.5,6) for u = 0, v = w0 and obtain Ξ0 := w0Θd ∈ o× Q and tQ(Ξ0) = 1. Thus Q is tame, and since P is wild, we conclude that ¯P is a proper pointed quadratic subspace of ¯Q. Hence eQ/F = 1, forcing Q to be unramified. (cid:3) Given β ∈ o, d ∈ Z with 0 ≤ d ≤ 2 texp(P ), it follows from 8.8 that µ := 1−πdβ has norm exponent at least d, and if d is odd, Prop. 8.14 (a) yields a characterization in terms of β when equality holds. While a similar characterization for d = 2 texp(P ) is presented in Thm. 9.9 (v) (though as yet unproved), we are now able to provide one for d even, d < 2 texp(P ). WILD PFISTER FORMS 43 9.10. Corollary. Let d be an even integer such that 0 ≤ d < 2 texp(P ). (a) With β ∈ o and µ := 1 − πdβ, the following conditions are equivalent. (i) µ ∈ o× and nexpP (µ) = d. (ii) ¯β /∈ V 2 ¯P . (b) If d > 0, βi ∈ o× and µi := 1 − πdβi ∈ o× for i = 1, 2, then hhµ1ii ⊗ P ∼= hhµ2ii ⊗ P =⇒ β1 ≡ β2 mod V 2 ¯P . Proof. (a) (i) =⇒ (ii). We have β ∈ o× by Prop. 8.9 (b). For d = 0 the assertion follows from Prop. 8.9 (c). Now suppose d 6= 0. Arguing indirectly, we assume that P with ¯β = ¯w2 = nP (w). Then nP (w) = β + πβ′ for there exists an element w ∈ o× some β′ ∈ o, and Lemma 8.15 with m := d 2 yields µ = 1 − πdβ = (1 − πd+1β′′)nP (v′′)−1 P where, setting r := texp(P ), v′′ = 1P − πmw ∈ o× β′′ = −β′ + πr−m−1α − πr+m−1αβ + π2m−1βnP (w) ∈ o, since 0 < m < r. Now roundness of P and the definition of the norm exponent imply nexpP (µ) ≥ d + 1, a contradiction. (ii) =⇒ (i). By (9.7.1) we have µ ∈ o× and Q is anisotropic by Prop. 9.7, forcing µ /∈ nP (V × P ) (Prop. 7.5 (a)) and d ≤ d′ := nexpP (µ) ≤ 2 texp(P ) (Thm. 8.10). Furthermore, by Prop. 8.9 (b), µ = µ′nP (v′), µ′ = 1 − πd′ β′ for some β′ ∈ o×, P . In particular, µ′ ∈ o× and Q ∼= Q′ := hhµ′ii ⊗ P . This implies eQ′/F = v′ ∈ o× eQ/F = 1 by (9.7.6), so d′ < 2 texp(P ) (Thm. 9.9) is even (Prop. 9.6), and we are allowed to apply Prop. 9.7 to Q′ by the implication (i) ⇒ (ii) already established. Thus (9.7.6) yields d = d′. ¯P contains β1 but not β2, then µ1, µ2 have different norm exponents by (a), so Q1, Q2 cannot be isomor- phic (Props. 8.9 (d), 7.5 (b)). We may therefore assume β1, β2 /∈ V 2 ¯P . As in the proof of Cor. 8.14 we have µ := µ1µ2 = 1 − πdβ, β := β1 + β2 − πdβ1β2 ∈ o. Assuming ¯β = β1 − β2 /∈ V 2 ¯P would force hhµii ⊗ P to be anisotropic by Prop. 9.7, hence µ1, µ2 to fall into distinct norm classes relative to P . But then Q1, Q2 would not be isomorphic, a contradiction. (cid:3) := hhµiii ⊗ P for i = 1, 2. (b) We put Qi If V 2 9.11. Examples. (a) Let m be an integer with 0 ≤ m < texp(P ) and suppose we are given an element γ ∈ o× such that ¯γ ∈ V 2 ¯P . If µ := 1 − π2mγ is a unit in o (automatic unless m = 0), Cor. 9.10 (a) implies nexpP (µ) > 2m, so it is a natural question to ask whether a more precise estimate for the norm exponent of µ can be given. Unless specific properties of γ are taken into account, the answer is no. To see this, let d ∈ Z with d > 2m, β ∈ o, and w ∈ o× P with ¯w 6= 1 ¯P . Then Lemma 8.15 implies for 1 − π2mγ ≡ 1 − πdβ mod nP (P ×), ¯γ = ¯w2 γ = −nP (w) + πd−2mβ + πtexp(P )−mα − πtexp(P )+d−mαβ + πdβnP (w) ∈ o×. Hence nexpP (1 − π2mγ) = ∞ for d > 2 texp(P ) by the local norm theorem 8.10, and, by Cors. 8.14,9.10, β may be so chosen that nexpP (1−π2mγ) attains any finite pre-assigned value d with 2m < d < 2 texp(P ) provided V 2 ¯P 6= ¯F . 44 SKIP GARIBALDI AND HOLGER P. PETERSSON (b) Let d ∈ Z, 0 < d ≤ 2 texp(P ), β, β′ ∈ o× and put µ := 1− πdβ, µ′ = 1− πdβ′ ∈ o×. We wish to refute the converse of Cors. 8.14 (b) and 9.10 (b) by showing that ¯β = ¯β′ does not imply µ ≡ µ′ mod nP (V × P ). Indeed, if β 6= β′, then ¯β = ¯β′ amounts to the same as µ′ = 1− πdβ− πqγ for some integer q > d and some γ ∈ o×, allowing us to conclude µµ′ = µ2(1 − πqµ−1γ) ≡ 1 − πqµ−1γ mod nP (V × P ). Therefore, • µ ≡ µ′ mod nP (V × P ) for q > 2 texp(P ) (Thm. 8.10), but • µ 6≡ µ′ mod nP (V × q < 2 texp(P ) is even (Cors. 8.14,9.10), and ¯w2 d = 2 texp(P ) (Thm. 9.9). P ) for d < q ≤ 2 texp(P ), provided ¯γ /∈ V 2 0 ¯γ /∈ Im(℘V ¯P ;¯sw0 ¯P if ) if (cid:3) The preceding results on the behavior of the ramification index, the pointed qua- dratic residue space and the trace exponent under the passage from P to hhµii ⊗ P , µ ∈ F ×, can be stated in a particularly concise way when addressed to pointed Pfis- ter quadratic spaces by combining them with the embedding property of Prop. 7.6. In order to do so, we introduce the following terminology. 9.12. Scalars of standard type. We say a scalar µ ∈ F has standard type rel- ative (or with respect) to P if it satisfies one of the following mutually exclusive conditions. (a) µ is a prime element of o (possibly distinct from π). (b) µ = 1 − πdβ for some odd integer d with 0 ≤ d < 2 texp(B) and some (c) µ = 1 − πdβ for some even integer d with 0 ≤ d < 2 texp(B) and some β ∈ o×. β ∈ o with ¯β /∈ V 2 ¯P . 9.13. Theorem. Let P be a pointed n-Pfister quadratic space over F that is anisotropic and wild of ramification index eP/F = 1. For Q to be a wild anisotropic pointed (n + 1)-Pfister quadratic space over F into which P embeds as a pointed quadratic subspace it is necessary and sufficient that Q be a pointed quadratic space isomorphic to hhµii ⊗ P , for some scalar µ ∈ F of standard type relative to P . In this case, precisely one of the following implications holds. (a) If µ is a prime element in o, then eQ/F = 2, ¯Q ∼= ¯P , texp(Q) = texp(P ). (b) If µ = 1 − πdβ for some odd integer d with 0 ≤ d < 2 texp(P ) and some β ∈ o×, then eQ/F = 2, ¯Q ∼= ¯P , texp(Q) = texp(P ) − d − 1 2 . (c) If µ = 1− πdβ for some even integer d, 0 ≤ d < 2 texp(P ) and some β ∈ o with ¯β /∈ V 2 ¯P , then eQ/F = 1, ¯Q ∼= hh ¯βii ⊗ ¯P , texp(Q) = texp(P ) − d 2 . WILD PFISTER FORMS 45 Proof. By Props. 9.2,9.6,9.7, the condition is sufficient, and (a)−(c) hold. Con- versely, suppose Q is a pointed anisotropic wild (n + 1)-Pfister quadratic space over F containing P as a pointed quadratic subspace. Up to isomorphism, Q = hhµii⊗ P for some µ ∈ F × by the embedding property (Prop. 7.6), where the reduction of 9.1 (b) allows us to assume that µ ∈ o× is a unit in o. Q being anisotropic implies 0 ≤ d := nexpP (µ) ≤ 2 texp(P ) by the local norm theorem 8.10 and without loss µ = 1 − πdβ for some β ∈ o. Since Q is wild, we conclude d < 2 texp(P ) from the part of Thm. 9.9 already established, and if d is odd, then β ∈ o× (Prop. 8.14 (a)), while if d is even, then ¯β /∈ V 2 In any event, µ has standard type relative to P . (cid:3) ¯P (Cor. 9.10). 9.14. Corollary. With P as in Thm. 9.13, let Q be a pointed quadratic space over F . (a) The following conditions are equivalent. (i) Q is an anisotropic pointed (n + 1)-Pfister quadratic space into which P embeds as a pointed quadratic subspace such that eQ/F = 2 and texp(Q) = texp(P ). (ii) Q ∼= hhπβii ⊗ P or Q ∼= hh1 − πβii ⊗ P for some β ∈ o×. In this case ¯Q ∼= ¯P . (b) The following conditions are equivalent. (i) Q is an anisotropic pointed (n + 1)-Pfister quadratic space into which P embeds as a pointed quadratic subspace such such that eQ/F = 1 and texp(Q) = texp(P ). (ii) Q ∼= hhµii ⊗ P for some µ ∈ o with ¯µ /∈ V 2 ¯P . In this case, ¯Q ∼= hh¯µii ⊗ P . Proof. the assertions follow immediately from Thm. 9.13. In (a) and (b), condition (i) implies that Q is wild (Prop. 8.2 (a)). Hence (cid:3) There is an analogue of Thm. 9.13 dealing with tame rather than wild enlargements of pointed n-Pfister quadratic spaces. We omit the proof since it proceeds along the same lines as the one of Thm. 9.13, applying Thm. 9.9 in full rather than Props. 9.2,9.6,9.7. 9.15. Theorem. Keeping the notations of 9.8 (b), let P be a pointed n-Pfister quadratic space over F that is anisotropic and wild of ramification index eP/F = 1. For Q to be a tame and anisotropic (n + 1)-Pfister pointed quadratic space over F into which P embeds as a pointed quadratic subspace it is necessary and sufficient that Q be a pointed quadratic F -space and there exist an element β ∈ o with Q ∼= hhµii ⊗ P, µ := 1 − π2 texp(B)β, β0 /∈ Im(℘V ¯P ,¯sw0 ), β0 := nP (w0)β. 10. λ-normed and λ-valued conic algebras. In order to illuminate the intuitive background of the present section, we recall the notion of an absolute-valued algebra. Following Albert [1] (see also Palacios [37]), an absolute-valued algebra is a non-associative real algebra A equipped with a norm x 7→ kxk that permits composition: kxyk = kxk kyk for all x, y ∈ A. Since absolute valued algebras obviously have no zero divisors, the finite-dimensional ones (cid:3) 46 SKIP GARIBALDI AND HOLGER P. PETERSSON are division algebras, hence exist only in dimensions 1, 2, 4, 8 (Albert [1] gave an ad- hoc proof of this result, the Bott-Kervaire-Milnor thoerem not having been known at the time). By contrast, natural analogues of absolute-valued algebras over 2- Henselian fields will be discussed in the present section that exist in all dimensions 2n, n = 0, 1, 2 . . . . Throughout we continue to work over a fixed 2-Henselian field F as in 7.7 and alert the reader to the terminological conventions of 7.14. All vector spaces, alge- bras, etc. over F are tacitly assumed to be finite-dimensional. 10.1. The basic concepts. A conic algebra C over F is said to be λ-normed if the following conditions hold. (i) C is non-singular, round and anisotropic. (ii) λC is sub-multiplicative: λC (xy) ≥ λC (x) + λC (y) for all x, y ∈ C. We speak of a λ-valued conic algebra if C satisfies (i), and if instead of (ii) the following stronger condition holds: (iii) λC is multiplicative: λC (xy) = λC (x) + λC (y) for all x, y ∈ C. The norm of a λ-valued conic algebra C over F will typically not permit composition (for example, if the dimension of C differs from 1, 2, 4, 8) but, remarkably, its failure to do so is not detected by λ since λC being multiplicative by (7.14.1) amounts to λ(nC (xy)) = λ(nC (x)nC (y)) for all x, y ∈ C. This looks like a pretty far-fetched phenomenon but, in fact, turns out to be quite common. We start with a trivial but useful observation. 10.2. Proposition. (a) λ-valued conic algebras over F are division algebras. (b) A composition algebra over F is a λ-valued conic algebra if and only if it is a division algebra. (cid:3) 10.3. Proposition. Let C be a non-singular, round and anisotropic conic algebra over F . Then C is λ-normed if and only if oC ⊆ C is an o-subalgebra and pC ⊆ oC C ⊆ poC. In this case, ¯C := oC/pC is a conic algebra over ¯F is an ideal with p2 whose norm, trace and conjugation are given by the formulas (1) (2) n ¯C (¯x) = nC (x), t ¯C (¯x) = tC (x), (¯x)∗ = (x∗) (3) for all x ∈ oC. Moreover, the norm of ¯C is round and anisotropic. Proof. By Lemma 7.12, sub-multiplicativity of λC amounts to λC (xy) ≥ λC (x) + λC (y) (x, y ∈ C, 0 ≤ λC (x), λC (y) ≤ 1 2 ). The first part of the proposition follows from this at once. The second part is a restatement of Prop. 7.10. (cid:3) Remark. The conic algebra ¯C = oC/pC is called the residue algebra of C. If C is wild, we do not know whether this residue algebra always agrees with the purely inseparable extension field of ¯F attached to C via 9.5, though it does if C is a composition algebra [41, Prop. 1]. WILD PFISTER FORMS 47 10.4. Corollary. With the notations of Prop. 10.3 suppose in addition that C has ramification index eC/F = 1. Then: (a) C is λ-normed if and only if oC ⊆ C is an o-subalgebra. (b) C is λ-valued if and only if C is λ-normed and ¯C is a division algebra. Proof. (a) pC = poC. C iff C is λ-normed and ¯C is a division algebra. (b) Consulting Lemma 7.12 again, λC is multiplicative iff λC (xy) = 0 for all (cid:3) We now proceed to re-examine the main results of the preceding section within x, y ∈ o× the framework of λ-normed and λ-valued conic algebras. 10.5. Convention. For the remainder of this section, we fix a λ-normed conic algebra B over F having ramification index eB/F = 1. 10.6. Proposition. If µ is a prime element in o, then C := Cay(B, µ) is a λ- normed conic algebra over F with ¯C = ¯B as conic ¯F -algebras. Moreover, C is λ-valued if and only if B is λ-valued. Proof. By Prop. 9.2, C is non-singular, round and anisotropic. Combining Prop. 10.3 with (1.10.1), we conclude that C is λ-normed. It remains to show that if B is λ-valued, so is C. By Lemma 7.12, we must show λC (x1x2) = λC (x1)+ λC (x2) for all xi = ui + vij ∈ C, ui, vi ∈ B, 0 ≤ λC (xi) ≤ 1 2 , i = 1, 2. There are four cases: (i) λC (x1) = λC (x2) = 0, (ii) λC (x1) = 0, λC (x2) = 1 2 , (iii) λC (x1) = 1 2 , λC (x2) = 0, (iv) λC (x1) = λC (x2) = 1 2 . We only treat (iv) and leave the other three cases to the reader. From (9.2.1) we deduce ui ∈ pB, vi ∈ o× B for i = 1, 2 and (1.10.1) yields v = v2u1 + v1u∗ 2, where λB(u) = 1, λB(v) ≥ 1, hence λC (x1x2) = 1 = λC (x1) + λC (x2) 10.7. Example. Specializing Prop. 10.6 to (iterated) Laurent series fields of char- acteristic not 2, we recover examples of λ-valued conic algebras that originally go back to to Brown [7, pp. 421-422]. In a slightly more general vein, let k be any field, L/k a separable quadratic field extension and write u = u1u2 + µv∗ x1x2 = u + vj, 2v1, (cid:3) A = Cay(L; µ1, . . . , µn−1) (n ∈ Z, n ≥ 1) for the k-algebra arising from L and scalars µ1, . . . , µn−1 ∈ k× by means of the Cayley-Dickson process as in 1.12. Then A is a flexible conic algebra with norm an n-Pfister quadratic form. We now assume that A is a division algebra, forcing A to be non-singular, round and anisotropic. Consider the field F = k((t)) of formal Laurent series in a variable t with coefficients in k, which is complete and therefore Henselian under the standard discrete valuation λ : F → Z∞. Setting B := A ⊗k F = A((t)), we obtain a flexible conic division F -algebra whose norm is an anisotropic n-Pfister quadratic form over F . Using (7.9.1), a straightforward verification shows ∞ λB(cid:16) Xr≫−∞ artr(cid:17) = min{r ∈ Z ar 6= 0} (ar ∈ A, r ∈ Z), which immediately implies that B is an unramified λ-valued conic algebra over F . By Prop. 10.6 we thus find in C := Cay(B, t) a λ-valued conic algebra over F 48 SKIP GARIBALDI AND HOLGER P. PETERSSON having dimension 2n+1. Starting from A = L (i.e., from n = 1) and continuing in this way, we obtain λ-valued conic algebras over appropriate iterated Laurent series fields in all dimensions 2n, n = 0, 1, 2, . . . . 10.8. Example. We specify Example 10.7 a bit further by setting k = R, L = C, n = 3, µ1 = µ2 = −1. Then A = O, the real algebra of Graves-Cayley octonions, and B = O((t)) is the unique unramified octonion division algebra over F = R((t)). Moreover, the 16-dimensional conic division algebra C = Cay(B, t) = Cay(F ;−1,−1,−1, t) over F contains B′ := Cay(F ;−1,−1, t) as a ramified octonion subalgebra. In particular, B and B′ are not isomorphic, allowing us to conclude from [7, Thm. 2] that the subalgebra B′ ⊆ C does not satisfy the embedding property 1.11. 10.9. Proposition. If B as in 10.5 is tame and µ ∈ o× \ nB(B×), then C := Cay(B, µ) is a λ-normed conic algebra over F with ¯C = Cay( ¯B, ¯µ) as conic ¯F - algebras. Moreover, C is λ-valued if and only if ¯C is a division algebra. Proof. By Prop. 9.3, C is a tame non-singular, round and anisotropic conic algebra over F with eC/F = 1. Moreover, oB being a o-subalgebra of B by Prop. 10.3, we conclude from (9.3.2) that oC is an o-subalgebra of C. Now everything follows from Cor. 10.4. (cid:3) Dealing with the case that B as in 10.5 is wild turns out to be more troublesome. We not only need a few preparations but also have to add an extra hypothesis by requiring that the conic algebras involved be flexible. 10.10. Proposition. Let C be a flexible λ-normed conic algebra over F . Then (1) (2) (3) λ(cid:0)tC (x)(cid:1) ≥ ω(C) + λC (x), λC ([x1, x2]) ≥ ω(C) + λC (x1) + λC (x2), λC (x − x∗) ≥ ω(C) + λC (x) for all x, x1, x2 ∈ C×. Proof. (1) follows immediately from (8.4.1). To establish (2), we combine (1) with (2.4.2),(7.9.5),(8.7.2) and use the fact that λC is sub-multiplicative. Finally, applying (1) and (2.1.6), we obtain λC (x − x∗) = λC(cid:0)2x − tC (x)1C(cid:1) ≥ min{eF + λC (x), ω(C) + λC (x)}, and (3) follows from (8.7.2). (cid:3) 10.11. Lemma. Suppose B as in 10.5 is wild and µ ∈ o× has the form (1) β ∈ o×. µ = 1 − πdβ, Setting C = Cay(B, µ) and d ∈ Z, 0 ≤ d ≤ 2 texp(B), (2) 0 ≤ m :=(cid:22) d 2(cid:23) < texp(B), Θd := π−m(1B + j) ∈ C WILD PFISTER FORMS 49 as in (9.4.1), the relations u(vΘd) = π−m[u, v] + (vu)Θd, (vΘd)u = π−mv(u − u∗) + (vu∗)Θd, (3) (4) (5) 2 v1(cid:17) + π−m(cid:16)tB(v2)v1 − [v1, v2](cid:17)Θd (v1Θd)(v2Θd) = (cid:16)π−2m[v1, v2 − v∗ hold for all u, v, v1, v2 ∈ B. Moreover, if B is flexible, then O := oB ⊕ oBΘd is an o-subalgebra of C. 2 ] − πd−2mβv∗ Proof. A slightly involved but straightforward computation using the transition formulas u + vΘd = (u + π−mv) + π−mvj, u + vj = (u − v) + πmvΘd implies (3)−(5). Combining these with Prop. 10.10 leads to the final assertion of the lemma. (cid:3) 10.12. Theorem. Suppose B as in 10.5 is flexible and wild, µ = 1 − πdβ, d ∈ Z, 0 ≤ d < 2 texp(B), β ∈ o×, and d is odd. Then C := Cay(B, µ) is a flexible λ-normed conic algebra over F with ¯C = ¯B as conic ¯F -algebras. Moreover, C is λ-valued if and only if B is λ-valued. Proof. The proof is similar to, but a bit more complicated than, the one of Prop. 10.6. By Prop. 9.6, C is non-singular, round and anisotropic. Combining the final statement of Lemma 10.11 with (9.6.4), we conclude that oC ⊆ C is an o-subalgebra which, thanks to (10.11.3−5) and to d being odd contains pC as an ideal with p2 C ⊆ poC. Thus C is λ-normed (Prop. 10.3), and it remains to show that if B is λ-valued, so is C. Let xi = ui + viΠ ∈ C×, ui, vi ∈ B, i = 1, 2 and x1x2 = u + vΠ, where (10.11.3−5) imply u = u1u2 + π−m(cid:0)[u1, v2] + v1(u2 − u∗ (1) 2)(cid:1) + π−2m[v1, v2 − v∗ v = v2u1 + v1u∗ 2 ] − πβv∗ 2 v1, (2) We must show λC (x1x2) = λC (x1)+ λC (x2). To this end, invoking Lemma 7.12, we may assume 0 ≤ λC (xi) ≤ 1 2 , i = 1, 2. Since conjugation is an algebra involution of C leaving λC invariant, there are three cases: (i) λC (x1) = λC (x2) = 0, (ii) λC (x1) = 0, λC (x2) = 1 2 . Among these cases, we treat only (iii) since the other ones can be treated analogously. In (iii) we have u1, u2 ∈ pB, v1, v2 ∈ o× B by (1), hence λB(u) = 1, while (2) yields λB(v) ≥ 1. Therefore λC (x1x2) = 1 = λC (x1)+λC (x2). B, observe Prop. 10.10 and obtain u ≡ −πβv∗ 2 , (iii) λC (x1) = λC (x2) = 1 2 v1 mod p2 (cid:3) 2 + π−m(cid:0)tB(v2)v1 − [v1, v2](cid:1). 10.13. Theorem. Suppose B as in 10.5 is flexible and wild, β ∈ o, 0 ≤ d < 2 texp(B), µ = 1 − πdβ, d ∈ Z, (1) ¯β /∈ V 2 ¯B and d is even. Then C := Cay(B, µ) is a wild λ-normed conic algebra over F with ¯C ∼= Cay( ¯B, ¯β) as conic ¯F -algebras. Moreover, C is λ-valued if and only if ¯C is a division algebra. Proof. By Prop. 9.7, C is a non-singular, round and anisotropic conic F -algebra. Moreover, C is wild of ramification index 1. The final statement of Lemma 10.11 50 SKIP GARIBALDI AND HOLGER P. PETERSSON combined with (9.7.5) shows that oC ⊆ C is an o-subalgebra, forcing C to be λ- normed (Cor. 10.4 (a)). Moreover, writing Cay( ¯B, ¯β) = ¯B ⊕ ¯Bj′, j′2 = ¯β1 ¯B as in 1.10, consulting (10.11.3−5) and observing d < texp(B) = ω(B), Prop. 10.10 shows that (in the notations of Prop. 9.7) the assignment u + vΞ 7→ ¯u + ¯vj′ determines an isomorphism ¯C ∼→ Cay( ¯B, ¯β) of ¯F -algebras. The final statement of the theorem follows immediately from Cor. 10.4 (b). (cid:3) 10.14. Corollary. With the notations and assumptions of Thm. 10.13, suppose in addition that ¯B/ ¯F is a purely inseparable field extension of exponent at most 1. Proof. Then C is a λ-valued conic algebra over F with ¯C ∼= ¯B(p ¯β). ¯B(p ¯β) (cf. 1.12, Case 2) a division algebra. It suffices to note that the last condition of (10.13.1) makes Cay( ¯B, ¯β) = (cid:3) Remark. The additional hypothesis in Cor. 10.14 is fulfilled if, e.g., B is a compo- sition algebra (Remark to 10.3). 10.15. Examples. In Brown's examples of conic division algebras (cf. 10.7), one basically keeps building up ramified λ-valued conic algebras over iterated Laurent series fields of characteristic not 2. By contrast, we will now be able to construct wild λ-valued conic algebras of ramification index 1 over appropriate Henselian fields of characteristic zero. Let k be any field of characteristic 2 and write K for the field of rational functions in an infinite number of variables over k. Then [K : K 2] = ∞. Pick an infinite chain K = K0 ⊂ K1 ⊂ ··· ⊂ Kn−1 ⊂ Kn ⊂ ··· of purely inseparable field extensions of K having exponent at most 1 and [Kn : K] = 2n for all integers n ≥ 0. Following Teichmuller [49], there is an essentially unique complete field F under a discrete valuation λ : F → Z∞ such that F has characteristic zero, residue field ¯F = K and absolute ramification index eF = 1. For n ≥ 1 choose βn ∈ o× such that Kn = Kn−1(p ¯βn), put µn = 1 − βn, observe (8.7.1) and apply Cor. 10.14 successively for d = 0 and n = 1, 2, 3, . . . to conclude that the Cayley-Dickson process leads to a wild λ-valued conic algebra Cn := Cay(F ; µ1, . . . , µn) over F having dimension 2n and ramification index 1 such that ¯Cn ∼= Kn. 10.16. Corollary. Let Q be a pointed Pfister quadratic space over F that is anisotropic and wild of ramification index eQ/F = 1. Then there exists a flexible λ-valued conic algebra C over F such that QC ∼= Q and ¯C/ ¯F is a purely inseparable field extension of exponent at most 1. Proof. Arguing by induction, we let Q be a pointed (n + 1)-Pfister quadratic space and pick a pointed n-Pfister quadratic subspace P ⊆ Q. Clearly, P is wild with eP/F = 1. By Theorem 9.13, some scalar µ of standard type relative to P satisfies Q = hhµii⊗ P up to isomorphism, and since the implications (a),(b) of that theorem do not hold for µ, implication (c) does. On the other hand, the induction hypothesis leads to a flexible λ-valued conic algebra B over F with QB ∼= P such that ¯B/ ¯F is a purely inseparable field extension of exponent at most 1. By Cor. 10.14, C := Cay(B, µ) is a flexible λ-valued conic algebra over F with QC = hhµii ⊗ P = Q and ¯C ∼= ¯B(p ¯β). (cid:3) WILD PFISTER FORMS 51 We still haven't closed the gap in our proof of Thm. 9.9 but will now be able to do so by appealing to the connection with conic algebras. In view of Cors. 10.14,10.16, the missing equivalence of (v) and (i)−(iv) in Thm. 9.9 will be a consequence of the following result. 10.17. Theorem. Suppose B as in 10.5 is flexible, wild and λ-normed having ¯B/ ¯F as a purely inseparable field extension of exponent at most 1 and, with the notations of 9.8 (b), let (1) β ∈ o, µ := 1 − π2 texp(B)β ∈ o×, β0 := nB(w0)β. Then C := Cay(B, µ) is anisotropic if and only if β0 /∈ Im(℘ ¯B,¯sw0 setting ). In this case, (2) we obtain the relations Ξ0 := π− texp(B)(w0 + w0j) ∈ C, (3) (4) tC (Ξ0) = 1, nC (Ξ0) = β0, C = B ⊕ BΞ0, λC (u + vΞ0) = min {λB(u), λB(v)} pC = pB ⊕ pBΞ0 oC = oB ⊕ oBΞ0, (u, v ∈ B), . (5) and C is λ-normed with Q ¯C ∼= Q ¯B; ¯β0,¯sw0 Proof. As in the proof of the implication (iii) ⇒ (i) in Thm. 9.9, we put d := 2 texp(B), Ξ := Θd and have Ξ0 = w0Ξ. Thus the first two relations of (3) follow from (9.4.5,6), while the last one is a straightforward consequence of the fact that B is λ-valued by Cor. 10.4 (b), hence a division algebra (Prop. 10.2 (a)). Setting m := texp(B), we let v ∈ B and compute, using (1.10.1,2),(2.4.1) and (1), nC(vΞ0) = π−2mnC(cid:0)v(w0 + w0j)(cid:1) = π−2mnC(cid:0)vw0 + (w0v)j(cid:1) = π−2m(cid:0)nB(vw0) − µnB(w0v)(cid:1) = π−2mnB(vw0)(1 − µ) = nB(vw0)β. But since ¯B/ ¯F is a purely inseparable field extension of exponent at most 1, we have nB(vw0) ≡ nB(v)nB(w0) mod p and conclude (6) Suppose first that C is anisotropic. Then ¯C is an anisotropic conic algebra over ¯F containing ¯B as a subalgebra and l := Ξ0 ∈ ¯C as a distinguished element with t ¯C(l) = 1, n ¯C (l) = β0, n ¯C(¯u, l) = ¯sw0 (¯u) for all u ∈ oB. Moreover, (6) implies n ¯C(¯vl) = β0¯v2 for all v ∈ oB. Writing ¯B ⊕ ¯Bj′ for the vector space underlying the , the assignment ¯u + ¯vj′ 7→ ¯u + ¯vl therefore and pointed quadratic space Q ¯B; ¯β0,¯sw0 to Q ¯C of pointed quadratic spaces. by (2.2.2) gives an embedding ϕ from Q ¯B; ¯β0,¯sw0 ∼→ Q ¯C Comparing dimensions (observe eC/F = 1 by (i) of Thm. 9.9), ϕ : Q ¯B; ¯β0,¯sw0 . Now is, in fact, an isomorphism, and since Q ¯C is anisotropic, so is Q ¯B; ¯β0,¯sw0 β0 /∈ Im(℘ ¯B,¯sw0 ) follows from Cor. 3.10 (a). Moreover, we claim that (4) holds (which immediately implies (5)). As usual, we may assume u, v ∈ o× B, which yields λC (u+vΞ0) ≥ 0, and if this were strictly positive, we would end up with ϕ(¯u+¯vj′) = ¯u + ¯vl = 0, forcing ¯u = ¯v = 0, a contradiction. Suppose next β0 /∈ Im(℘ ¯B,¯sw0 ) and consider the full o-lattice nC (vΞ0) ≡ nB(v)β0 mod p. (7) O := oB ⊕ oBΞ0 ⊆ C, 52 SKIP GARIBALDI AND HOLGER P. PETERSSON on which nC takes integral values. More precisely, (6) and (2.2.2) imply nC(u + vΞ0) ≡ nB(u) + π−mnB(v∗u, Ξ0) + nB(v)β0 mod p (8) for all u, v ∈ oB. Reducing mod p, we obtain a pointed quadratic space ¯O := O ⊗o ¯F = ¯B ⊕ ¯Bl′, l′ := Ξ0 ⊗o 1 ¯F , over ¯F , and the assignment ¯u ⊕ ¯vj′ 7→ ¯u + ¯vl′ by (7),(8) gives an isomorphism from onto ¯O. The former being anisotropic by Cor. 3.10 (a), so is the latter. Q ¯B; ¯β0,¯sw0 But then C must be anisotropic as well since every non-zero element x ∈ C satisfies πmx ∈ O \ pO for some integer m. It remains to show that C is λ-normed provided it is anisotropic. By Cor. 10.4 (a), it suffices to show that oC ⊆ C is an o-subalgebra. In order to do so, we use Lemma 10.11 to derive the following formulas by a straightforward computation, for all u, v, v1, v2 ∈ B. (9) vΞ0 = π−m[v, w0] + (w0v)Ξ, w0 v] + (L−1 vΞ = π−m[w0, L−1 w0 v)Ξ0, (10) (11) (12) (13) u(vΞ0) = π−m(cid:0)u[v, w0] + [u, w0v](cid:1) +(cid:0)(w0v)u(cid:1)Ξ, (vΞ0)u = π−m(cid:0)[v, w0]u + (w0v)(u − u∗)(cid:1) +(cid:0)(w0v)u∗(cid:1)Ξ (v1Ξ0)(v2Ξ0) = π−2m(cid:16)[v1, w0][v2, w0] + [[v1, w0], w0v2]+ β(w0v∗ [w0v1, w0v2 − (w0v2)∗] + (w0v1)(cid:0)[v2, w0] − [v2, w0]∗(cid:1)(cid:17)− 2)(w0v1) + π−m(cid:16)(w0v2)[v1, w0] + (w0v1)[v2, w0]∗+ tB(w0v2)w0v1 − [w0v1, w0v2](cid:17)Ξ. Since B is λ-valued, (10) implies oBΞ ⊆ oC, and then (11)−(13) combine with Prop. 10.10 to establish oC as an o-subalgebra of C. (cid:3) 10.18. Corollary. Suppose in Thm. 10.17 that B is an associative composition division algebra and β0 /∈ Im(℘ ¯B; ¯β0,¯sw0 ). Then C is an unramified composition division algebra over F with ¯C ∼= Cay( ¯B; β0, ¯sw0 ) as a non-orthogonal Cayley- Dickson construction. Proof. Composition algebras are classified by their norms. (cid:3) Remark. If C as in Thm. 10.17 is anisotropic, its pointed quadratic residue space is described explicitly by the theorem. But C is also a λ-normed conic algebra, making ¯C canonically a conic algebra (over ¯F ) in its own right. It would be interesting to obtain an equally explicit description of that algebra. Cor. 10.18 provides one if C is a composition algebra but it is not at all clear whether this description prevails in the general case, nor whether C is always a λ-valued conic algebra. 11. Applications to composition algebras. There are obvious and less obvious applications of the preceding results to compo- sition algebras. Working over a fixed 2-Henselian field F of arbitrary characteristic as before (cf. 7.7), the obvious ones may be described as follows. WILD PFISTER FORMS 53 11.1. Translations. Since composition division algebras over F are classified by their norms and are λ-valued conic F -algebras by Prop. 10.2 (b), the results of Sections 8, 9 translate immediately into this more special setting, where the ones in Section 9 in particular yield explicit descriptions of how the valuation data ramifi- cation index, residue algebra and trace exponent behave under the Cayley-Dickson construction. Rather than carrying out these translations in full detail, suffice it to point out that all one has to do is replace division algebra B over F having ramification index eB/F = 1, • the pointed quadratic space P of 8.14,9.1,9.4 by an associative composition • the pointed quadratic space Q by a composition algebra C, and the con- • the passage from P to hhµii⊗ P , µ ∈ F ×, by the Cayley-Dickson construc- dition of Q being anisotropic by the one of C being a division algebra, tion Cay(B, µ). The less obvious applications of our results to composition algebras are all related, in one way or another, to the following innocuous observation, which we have not been able to extend to pointed Pfister quadratic spaces. 11.2. Proposition. Let C be a composition division algebra over F and B′ ⊆ ¯C a unital subalgebra. Assume char(F ) 6= 2 or dim ¯F (B′) > 1. Then there exists a composition subalgebra B ⊆ C having ramification index eB/F = 1 and satisfying ¯B = B′. Proof. One adapts the proof of [41, Lemma 3] to the present more general set- up; for completeness, we include the details. Since B′ is either a composition division algebra over ¯F or a purely inseparable field extension of characteristic 2 and exponent at most 1, it has dimension 2m, m ∈ Z, 0 ≤ m ≤ 3. Moreover B′ is generated by m elements x1, . . . , xm, for some x1, . . . , xm ∈ oC, where we may assume m ≥ 1 since m = 0 implies char(F ) 6= 2 by hypothesis and B := F does the job. The elements of non-zero trace in C form a Zariski open and dense subset, which therefore is open and dense in the valuation topology as well, so we may assume tC (x1) 6= 0. Then B, the unital subalgebra of C generated by x1, . . . , xm, is a composition division algebra with dimF (B) ≤ 2m, B′ ⊆ ¯B. Now Prop. 7.13 implies ¯B = B′ and eB/F = 1. (cid:3) The preceding result can be refined in various ways. For example, given a com- position division algebra C over F , we will exhibit (chains of) proper composition subalgebras of C having ramification index 1 and the same trace exponent as C. From this we derive normal forms for octonion and quaternion algebras over F and show that quantities subject to a few obvious constraints are the valuation data of an appropriate composition division algebra. We begin by listing a few properties of wild separable quadratic field extensions which should be well known but seem to lack a convenient reference. We therefore include the details. 11.3. Proposition. Let L be an F -algebra and suppose ¯F has characteristic 2. Then the following conditions are equivalent. (i) L/F is a wild separable quadratic field extension and eL/F = 1. (ii) There are a positive integer r and elements α ∈ o×, β ∈ o such that ¯β /∈ ¯F 2 and L ∼= F [t]/(t2 − πrαt + β). 54 SKIP GARIBALDI AND HOLGER P. PETERSSON If these conditions hold, ¯L = ¯F (p ¯β). Moreover, setting ϑ := t mod (t2 − πrαt + β) ∈ L (1) in (ii), the following relations hold. (2) (3) (4) λL(γ + δϑ) = min {λ(γ), λ(δ)} oL = o1L ⊕ oϑ, pL = p1L ⊕ pϑ, texp(L) = min {eF , r}. (γ, δ ∈ F ), Proof. (i) =⇒ (ii). By (i) there exists an element β ∈ o such that ¯β /∈ ¯F 2 and ¯L = ¯F (p ¯β). Pick an element ϑ ∈ oL satisfying ¯ϑ = p ¯β. Then tL(ϑ) ∈ p, and replacing β by nL(ϑ) if necessary, we may assume nL(ϑ) = β. Here ¯β /∈ ¯F 2 forces L = F (ϑ) = F [ϑ]. We claim there is no harm in assuming tL(ϑ) 6= 0. Indeed, for char(F ) = 2, this is automatic while, if char(F ) = 0, we may replace ϑ by ϑ′ := 1L + ϑ. Thus, without loss, tL(ϑ) 6= 0. But this yields a unit α ∈ o× such that tL(ϑ) = πrα, r := λ(tL(ϑ)) ∈ Z, r > 0, and L has the form described in (ii). (ii) =⇒ (i). By the hypotheses on β, the monic polynomial f := t2 − πrαt + β ∈ o[t] ⊆ F [t] is irreducible over F , and L/F is a separable quadratic field extension. L , and ¯ϑ2 = ¯β, which Define ϑ as in (1). We have nL(ϑ) = β ∈ o×, hence ϑ ∈ o× implies ¯L = ¯F (p ¯β), so L is wild and has ramification index eL/F = 1. Hence (i) holds. A standard argument now yields (2), which immediately implies the remaining (cid:3) assertions of the proposition. 11.4. Corollary. For a separable quadratic field extension L/F to be wild and to have ramification index 1 it is necessary and sufficient that texp(L) > 0 and there exist a trace generator u of L with nL(u) /∈ ¯F 2. In this case, L = k[u] and u may be so chosen as to satisfy the additional relation (1) λL(u − u∗) = texp(L). Proof. Necessity and the final statement. If L is wild and eL/F = 1, we obtain texp(L) > 0 and first deal with the case eF > texp(L). Then we find a positive integer r and elements α ∈ o×, β ∈ o as in Prop. 11.3 (ii) and conclude texp(L) = r from (11.3.4). Moreover, u := ϑ as defined in (11.3.1) is a trace generator of L satisfying nL(u) /∈ ¯F 2 and L = F [u]. Finally, λL(u − u∗) = λL(2u − πrα1L), where λL(2u) = eF > r = λL(πrα1L), which implies (1) as well. By (8.7.2), we are left with the case eF = texp(L) < ∞. Then F has characteristic 0, allowing us to apply Cor. 9.14 (b) with B = P = F : there exists a scalar µ ∈ o such that ¯µ /∈ ¯F 2 and L = F (√µ), so some y ∈ o× L has L = F [y], tL(y) = 0, nL(y) /∈ ¯F 2. (2) Hence u := 1L + y ∈ oL is a trace generator of L satisfying nL(u) /∈ ¯F 2. Moreover, since y∗ = −y by (2), λL(u − u∗) = λL(2y) = eF = texp(L), and the proof is complete. Sufficiency. We have L = k[u], tL(u) = πrα, r := texp(L), α ∈ o×, and condition (ii) of Prop. 11.3 holds with β = nL(u). (cid:3) WILD PFISTER FORMS 55 11.5. Proposition. Let L be an F -algebra and suppose ¯F has characteristic 2. Then the following conditions are equivalent. (i) L/F is a separable quadratic field extension of ramification index eL/F = 2. (ii) There are a positive integer r and elements α, β ∈ o× such that L ∼= F [t]/(t2 − πrαt + πβ). In this case, for any prime element Π ∈ oL (e.g., for (1) Π := t mod (t2 − πrαt + πβ)) the following relations hold. (2) (3) (4) λL(γ + δΠ) = min{λ(γ), λ(δ) + pL = p ⊕ oΠ, oL = o ⊕ oΠ, texp(L) = min{eF , r}. 1 2} (γ, δ ∈ F ), Proof. Everything is standard once it has been shown in (i) ⇒ (ii) that oL contains prime elements Π with tL(Π) 6= 0. But this follows from the fact that the set of elements in L with non-zero trace, by separability being open and dense in the Zariski topology, is open and dense in the valuation topology as well. (cid:3) 11.6. Theorem. Let C be a composition division algebra of dimension 2n, n = 2, 3, over F . Then there exists a separable quadratic subfield L ⊆ C having ramification index 1 and the same trace exponent as C: eL/F = 1, texp(L) = texp(C). Proof. Setting r := texp(C), we proceed in four steps. 10. Let us first consider the case r = 0. Then ¯C is a composition division algebra of dimension at least 2 over ¯F and hence contains a separable quadratic subfield L′ ⊆ ¯C, which by Prop. 11.2 or [41, Lemma 3], lifts to a separable quadratic subfield L ⊆ C with eL/F = 1, texp(L) = 0 = r. We may therefore assume from now on that r > 0, so C is wild. 20. Next we deal with the case eC/F = 1. Pick a trace generator w0 of C, which belongs to o× C by (8.3.1). If w0 /∈ F 1C, then L := F [w0] is a separable quadratic subfield of C satisfying 1 ≤ eL/F ≤ eC/F = 1, hence eL/F = 1. From pr = πro = tL(ow0) ⊆ tL(oL) = ptexp(L) we conclude texp(L) ≤ r, which implies texp(L) = r by Prop. 8.2 (b). On the other hand, if w0 ∈ F 1C, then r = eF and any separable quadratic subfield L ⊆ C satisfies eL/F = 1 as well as r ≤ texp(L) ≤ eF = r by Prop. 8.2 (b) and (8.7.2). 30. We are left with the case eC/F = 2. Then Prop. 11.2 yields a compositions subalgebra B ⊆ C with eB/F = 1 and dimF (B) = 2n−1. If texp(B) = r, we are done for n = 2 and my apply 20 to B for n = 3 to arrive at the desired conclusion. By Prop. 8.2 (b), we may therefore assume texp(B) > r. But then Thm. 9.13 justifies the assumption C = Cay(B, µ), µ = 1 − πdβ, for some odd integer d, 0 ≤ d < 2 texp(B), and some unit β ∈ o×. In other words, we are in the situation of Prop. 9.6. 40. Applying 20 to B if B is a quaternion algebra, we find a separable quadratic subfield L ⊆ B satisfying eL/F = 1, texp(L) = texp(B). From Cor. 11.4 we therefore conclude that there is an element u ∈ o× L such that tL(u) = πtexp(B), 56 SKIP GARIBALDI AND HOLGER P. PETERSSON nL(u) /∈ ¯F 2. Now consider the element w := u + uΠ, which is a unit of oC by (9.6.4). Moreover, setting m := d−1 2 and observing Prop. 9.6, tC(w) = πrε, ε := 1 + πtexp(B)−r ∈ o×, nC(w) = nL(u) + 2π−mnL(u) + πβnL(u), hence nC (w) = nL(u) /∈ ¯F 2, so w by Prop. 11.3 generates a separable quadratic subfield L′ ⊆ C with eL′/F = 1, texp(L′) = r. 11.7. Corollary. Every composition division algebra C of dimension 2n, n = 2, 3, over F contains a trace generator which is a unit in oC. Proof. Picking L ⊆ C as in Thm. 11.6, every trace generator of L is one of C and by (8.3.1) belongs to o× (cid:3) (cid:3) L ⊆ o× C. 11.8. Corollary. Let C be a composition division algebra over F with fC/F > 1. Then texp(C) = min {λC(u − u∗) u ∈ oC} = min{λC (u − u∗) u ∈ o× C}. Proof. For u ∈ oC we apply (2.1.6),(7.7.1),(8.7.2) and obtain λC (u − u∗) = λC(cid:0)2u − tC (u)1C(cid:1) ≥ min{eF + λC (u), λ(cid:0)tC (u)(cid:1)} ≥ texp(C). Hence it suffices to show that there is an element u ∈ o× C with λC (u−u∗) = texp(C). If C is tame, the conjugation of ¯C cannot be the identity, so λC (u − u∗) = 0 = texp(C) for some u ∈ o× C . We may therefore assume that C is wild. Combining the hypothesis fC/F > 1 with Thm. 11.6, we find a separable quadratic subfield L ⊆ C with eL/F = 1 and texp(L) = texp(C). Now Cor. 11.4 yields and element u ∈ o× We will see in Example 12.6 below that the hypothesis fC/F > 1 in the preceding corollary cannot be avoided. C such that λC (u − u∗) = λL(u − u∗) = texp(L) = texp(C). L ⊆ o× (cid:3) 11.9. Corollary. Let C be a composition division algebra of dimension 2n, n = 2, 3, over F . Then there exists a composition subalgebra B ⊆ C with dimF (B) = 2n−1, eB/F = 1, texp(B) = texp(C). Proof. For n = 2, this is just Thm. 11.6, so we may assume n = 3, i.e., that C is an octonion algebra. By [41, Lemma 3], we may also assume texp(C) > 0 and, applying Thm. 11.6 again, we find a separable quadratic subfield L ⊆ C with eL/F = 1, texp(L) = texp(C). Hence it suffices to prove the following lemma. 11.10. Lemma. Let C be a wild octonion division algebra over F and L ⊆ C a separable quadratic subfield such that eL/F = 1, texp(L) = texp(C). Then there exists a quaternion subalgebra L ⊆ B ⊆ C with eB/F = 1, texp(B) = texp(C). Proof. Since C is wild, its residue algebra ¯C is a purely inseparable field extension of exponent 1 and degree at least 4 over ¯F containing ¯L as a quadratic subfield. C satisfying ¯y ∈ ¯C \ ¯L. Then L and y generate a composition Pick an element y ∈ o× subalgebra B ⊆ C of dimension at most 4 whose residue algebra contains ¯L(¯y). Hence B ⊆ C is a quaternion subalgebra containing L and having eB/F = 1, texp(C) = texp(L) ≥ texp(B) ≥ texp(C) by Prop. 8.2 (b). (cid:3) WILD PFISTER FORMS 57 11.11. Normal Form Theorem: octonion algebras. For 0 < r ≤ eF and an F -algebra C, the following conditions are equivalent. (i) C is an octonion division algebra over F having trace exponent texp(C) = r. (ii) There exist (I) a separable quadratic field extension L/F such that eL/F = 1 and texp(L) = r (in particular, L is wild ), (II) a scalar α ∈ o such that ¯α /∈ ¯L2, (III) either a scalar β ∈ o such that ¯β /∈ ¯L(√¯α)2 and C ∼= Cay(L; α, β) or a scalar β ∈ o× such that C ∼= Cay(L; α, 1 − πβ) or C ∼= Cay(L; α, πβ). If this is so, then eC/F = 1, ¯C ∼= ¯L(√ ¯α,p ¯β) in the first alternative of (III), while eC/F = 2, ¯C ∼= ¯L(√ ¯α) otherwise. Proof. (i) =⇒ (ii). Thm. 11.6 yields a separable quadratic subfield L ⊆ C satisfying (I), and applying Lemma 11.10, we find a quaternion subalgebra L ⊆ B ⊆ C with eB/F = 1, texp(B) = texp(C) = r = texp(L). Now Cor. 9.14 (b) yields a quantity α satisfying (II) and B ∼= Cay(L, α), ¯B ∼= ¯L(√ ¯α), while (a) and (b) of the same corollary yield a quantity β satisfying (III) as well as the remaining assertions of the theorem. (ii) =⇒ (i). This follows immediately from Prop. 9.2 and Cor. 9.14. (cid:3) Basically the same arguments, inserting the obvious simplifications at the appro- priate places, leads to the following quaternionic version of the preceding result. 11.12. Normal Form Theorem: quaternion algebras. For 0 < r ≤ eF and an F -algebra B, the following conditions are equivalent. (i) B is a quaternion division algebra over F having trace exponent texp(B) = r. (ii) There exist (I) a separable quadratic field extension L/F such that eL/F = 1 and texp(L) = r (in particular, L is wild ), (II) either a scalar α ∈ o such that ¯α /∈ ¯L2 and B ∼= Cay(L, α), or a scalar α ∈ o× such that B ∼= Cay(L; 1 − πα) or B ∼= Cay(L, πα). If this is so, then eB/F = 1, ¯B ∼= ¯L(√¯α) in the first alternative of (II), while eB/F = 2, ¯B ∼= ¯L otherwise. On the basis of the preceding results, it can now be shown that the three valuation data we are interested in can be pre-assigned in advance pretty much arbitrarily once the obvious constraints are taken into account: (cid:3) 58 SKIP GARIBALDI AND HOLGER P. PETERSSON 11.13. Corollary. Let e, n, r be integers with n ≥ 0 and A an ¯F -algebra of dimen- sion 2n. There exists a composition division algebra C over F satisfying ¯C ∼= A, eC/F = e, texp(C) = r if and only if the following conditions are fulfilled. (a) e ∈ {1, 2}, 0 ≤ n ≤ 4 − e, 0 ≤ r ≤ eF , r = eF (for n = 0, e = 1). (b) A is a composition division algebra for r = 0, and A/ ¯F is a purely in- separable field extension of characteristic 2 and exponent at most 1 for r > 0. Proof. If C is a composition division algebra of dimension 2m, 0 ≤ m ≤ 3, over F having the prescribed valuation data, then e = eC/F ∈ {1, 2} by (7.9.8), 0 ≤ r = texp(C) ≤ eF by (8.7.2) and n = m or n = m − 1 according as eC/F is 1 or 2 by Prop. 7.13; moreover, n = 0 and e = 1 imply C ∼= F . Summing up and observing (8.7.1), we obtain (a), while (b) follows from Prop. 8.2 (d) combined with the remark to Prop. 10.3. Conversely, suppose (a) and (b) are fulfilled. If r = 0, the existence of a composition division algebra over F with the desired properties follows from from [41, Thms. 1,2]. We may therefore assume r > 0, which by (b) implies that A/ ¯F is a purely inseparable field extension of characteristic 2, exponent at most 1 and degree 2n. We first consider the case n = 0. If e = 1, then C = F has the prescribed valuation data since r = eF by (a). If e = 2, Prop. 11.5 yields a separable quadratic field extension L/k having eL/F = 2, texp(L) = r. We may therefore assume n > 0. By Prop. 11.3, there exists a separable quadratic field extension L/F such that eL/F = 1, texp(L) = r and ¯L ⊆ A. Hence A = ¯L, or A = ¯L(√¯α) for some α ∈ o, ¯α /∈ ¯L2, or A = ¯L(√ ¯α,p ¯β ) for some α, β ∈ o, ¯α /∈ ¯L2, ¯β /∈ ¯L(√¯α)2 according as n = 1, 2, 3. In each case, Prop. 11.3 and Thms. 11.11, 11.12 yield a composition division algebra C over F having the prescribed valuation data. (cid:3) Not surprisingly, however, the above valuation data are far away from classifying composition division algebras over 2-Henselian fields. This fact is underscored by the following example. 11.14. Example. Suppose ¯F has characteristic 2 and let ¯F ⊆ K ′ ⊆ L′ be a chain of purely inseparable field extensions having exponent at most 1 over ¯F such that [L′ : ¯F ] = 2n, [K ′ : ¯F ] = 2n−1, n = 2, 3. Furthermore, suppose r, s ∈ Z satisfy the relations 0 < s < r ≤ eF . Then Cor. 11.13 yields a composition division algebra B over F with (1) eB/F = 1, ¯B ∼= K ′, texp(B) = r. We claim there are an infinite number of mutually non-isomorphic composition divi- sion algebras C over F containing B as a subalgebra and having eC/F = 1, ¯C ∼= L′, texp(C) = s. This is in stark contrast to the fact that unramified composition divi- sion algebras up to isomorphism are uniquely determined by their residue algebras [41, Thm. 1]. To prove our claim, we fix an element β ∈ o with L′ = K ′(p ¯β) and have ¯β /∈ ¯B2 by (1), which implies αβ /∈ ¯B2 for any α ∈ o× with ¯α ∈ ¯B2. Setting d := 2(r − s) = 2 (texp(B) − s) > 0, µα := 1 − πdαβ ∈ o, Cα := Cay(B, µα), we conclude from Prop. 9.7 that µα ∈ o× and Cα is a composition division algebra over F satisfying eCα/F = 1, Cα ∼= L′, texp(Cα) = s. WILD PFISTER FORMS 59 It therefore suffices to show, for any additional element α′ ∈ o× with α′ ∈ ¯B2, that Cα ∼= Cα′ implies ¯α = α′. To see this, we write ¯α = δ2, α′ = δ′2 with δ, δ′ ∈ ¯B and invoke Cor. 9.10 (b) to derive the following chain of implications. Cα ∼= Cα′ =⇒ αβ ≡ α′β mod ¯B2 =⇒ ∃ γ ∈ ¯B : δ2 ¯β = δ′2 ¯β + γ2 =⇒ ∃ γ ∈ ¯B : (δ − δ′)q ¯β + γ = 0 =⇒ δ = δ′ =⇒ ¯α = α′. (cid:3) 12. Types of composition algebras and heights. The trace exponent has been our principal tool so far to detect wildness in pointed quadratic spaces and related objects. Other tools of this kind may be obtained by consulting the literature. It is the purpose of the present section to recast these tools in the setting of composition algebras and to compare them with the trace exponent. We begin by describing what will turn out later (see Cor. 12.4 below) to be a dichotomy of composition division algebras over a 2-Henselian field F . 12.1. Types of composition algebras. A composition division algebra C over F is said to be of unitary (resp. of primary) type if there exist an associative composition division algebra B over F with eB/F = 1 and a scalar µ which is a unit (resp. a prime element) in o such that C ∼= Cay(B, µ). We record a few easy but useful observations. (a) By Prop. 11.2, C is of unitary or of primary type, provided fC/F > 1 or char(F ) 6= 2. (b) For C to be of primary type it is necessary by Prop. 9.2 that C have ramification index eC/F = 2. (c) If C is tame, then C is of unitary (resp. of primary) type if and only if C is unramified (resp. ramified) [41]. (d) Suppose F has characteristic 0 and ¯F has characteristic 2. A quadratic field extension of F may or may not be wild. But if it is, it is of primary (resp. of unitary) type if and only if it is tamely ramified (resp. wildly ramified or wildly unramified) in the sense of [22, p. 60]. 12.2. Remark. At this stage we cannot rule out the possibility that a composition division algebra C over F is both of unitary and of primary type: conceivably, there could exist composition division algebras B, B′ over F of ramification index 1, and a unit µ as well as a prime element µ′ in o such that Cay(B, µ) ∼= C ∼= Cay(B′, µ′). For showing that this scenario is actually impossible, the following improvement of Prop. 8.5 will play a crucial role. 12.3. Theorem. Let C be a composition division algebra over F . (a) If C is of primary type, then ω(C) = texp(C). (b) Consider the following conditions on C. (i) ω(C) = texp(C) − 1 2 . 60 SKIP GARIBALDI AND HOLGER P. PETERSSON (ii) There are trace generators of C belonging to pC. (iii) C is wild of unitary type and ramification index 2. Then the implications (i) ⇐⇒ (ii) ⇐= (iii) hold. Moreover, if fC/F > 1 or char(F ) 6= 2, then all three conditions are equivalent. Proof. We begin with the first part of (b). λC (w0) > 0, hence w0 ∈ pC. λC (w0) < texp(C), and Prop. 8.5 (a) gives (ii). (i) =⇒ (ii). Let w0 be a regular trace generator of C. Then (i) and (8.5.1) show (ii) =⇒ (i). Let w0 ∈ pC be a trace generator of C. Then ω(C) ≤ λ(tC (w0)) − (iii) =⇒ (ii). If C is wild of unitary type and ramification index 2, we have C = Cay(B, µ), B an associative composition division algebra over F with eB/F = 1, µ ∈ o×, and Thm. 9.13 shows that we are in the situation of Prop. 9.6 with Q = C, P = B (cf. the translation formalism 11.1). Picking a trace generator u of B, we apply (9.6.1),(9.6.5) to obtain λ(cid:0)tC(uΠ)(cid:1) = λ(cid:16)tC(cid:0)π− d−1 = texp(B) − 2 2 (u + uj)(cid:1)(cid:17) = λ(cid:0)tB(u)(cid:1) − d − 1 = texp(C), d − 1 2 so uΠ ∈ pC is a trace generator of C, showing (ii). Before completing the proof of (b), we turn to (a) By hypothesis, we are in the situation of Prop. 9.2 with Q = C, P = B as before, and by (9.2.2), an element of pC has the form y = u + vj, u ∈ pB, v ∈ oB. Hence π−1u ∈ oB since eB/F = 1, and from (9.2.3) we conclude λ(cid:0)tC (y)(cid:1) = λ(cid:0)tB(u)(cid:1) = λ(cid:0)tB(π−1u)(cid:1) + 1 ≥ texp(B) + 1 = texp(C) + 1 texp(C). It remains to show (i) ⇒ (iii) in (b) under the assumption fC/F > 1 or char(F ) 6= 2. Then eC/F = 2 by Prop. 8.5 (c), and texp(C) = ω(C) + 1 2 > 0 by (i) forces C to be wild. Moreover, by 12.1 (a), it is of unitary or of primary type, the latter case being excluded by (i) and (a). (cid:3) Thus pC does not contain trace generators of C, violating (ii), hence (i), in (b). Now Prop. 8.5 (a) implies (a). 12.4. Corollary. A composition division algebra C over F cannot be both of uni- tary and of primary type. Proof. If C is tame, the assertion follows immediately from 12.1 (c). If C is wild of ramification index 1, it cannot be of primary type by 12.1 (b). Hence we are left with the case that C is wild of ramification index 2. Then, by Thm. 12.3, ω(C) = texp(C) is an integer if C is of primary type, while ω(C) = texp(C) − 1 2 is not if C is of unitary type. (cid:3) 12.5. Corollary. Let C be a composition division algebra of primary type over F and suppose B ⊆ C is a composition subalgebra with eB/F = 1, dimk(B) = Then texp(B) = texp(C). 1 2 dimk(C). WILD PFISTER FORMS 61 Proof. Since C is not of unitary type by Cor. 12.4, we obtain C ∼= Cay(B, µ) for some prime element µ ∈ o. Hence texp(B) = texp(C) by (9.2.3). Both Cor. 11.8 and Thm. 12.3 (b) fail in the exceptional cases stated therein. This is the upshot of the following example. (cid:3) 12.6. Example. Let L/F be a separable quadratic field extension of ramification index 2 and suppose char( ¯F ) = 2. Then we are in the situation of of Prop. 11.5, and if r ≤ eF , then texp(L) = r by (11.5.4). Moreover, (11.5.1) gives λ(tL(Π)) = λ(πrα) = r = texp(L), so Π ∈ pL is a trace generator of L; in particular, for char(F ) = 2, parts (i),(ii) of Thm. 12.3 (b) hold but (iii) doesn't. On the other hand, if r > eF , then F has characteristic zero and L = F(cid:0)√πγ(cid:1) = Cay(F, πγ), γ := −β + π2r−1 4 α2 ∈ o×, is of primary type, forcing ω(L) = texp(L) by Thm. 12.3 (a). Let u = γ + δΠ ∈ oL, γ, δ ∈ o (cf. (11.5.3)). Applying (11.5.2), λL(u − u∗) = λL(−πrαδ + 2δΠ) = min {r + λ(δ), eF + λ(δ) + 1 2} ≥ min{r, eF + 1 2}, and this minimum is attained for, e.g., u = Π. Thus, by(11.5.4), min{λL(u − u∗) u ∈ oL} =(texp(L) texp(L) + 1 2 for r ≤ eF , for r > eF , so in the latter case, the conclusion of Cor. 11.8 does not hold. We remark in closing that Cor. 11.8 also fails if L/F is tame of ramification index 2 since this implies texp(L) = 0 while ∗ induces the identity on ¯L, so λL(u− u∗) > 0 for all u ∈ oL. 12.7. Remark. There is an alternate way of proving Cor. 12.4, by working with quadratic forms. Let C be a composition division algebra over F and suppose C is of unitary and of primary type. Then there are composition division algebras B, B′ over F of ramification index 1 and a unit µ as well as a prime element µ′ in o such that Cay(B, µ) ∼= C ∼= Cay(B′, µ′). Then Prop. 3.12 yields a scalar γ ∈ F × satisfying (1) Cay(B, µ) ∼= Cay(B, γ), Cay(B′, µ′) ∼= Cay(B′, γ) Since eB/F = eB ′/F = 1, we conclude from (9.1.2) that λ(nB(B×)) = λ(nB ′ (B′×)) = 2Z. Hence Prop. 7.5 (b) and the first relation of (1) show that λ(γ) is even, while Prop. 7.5 (b) and the second relation of (1) show that λ(γ) is odd, a contradiction. With the aim of generalizing Thm. 11.6, Cor. 11.9 and Lemma 11.10, we next turn to the problem of finding (chains of) subalgebras having ramification index 1 and pre-assigned trace exponents. Once the obvious constraints are taken into account (provided, e.g., by Prop. 8.2 (b) and Cor. 12.5), we will show that chains of such subalgebras always exist. 62 SKIP GARIBALDI AND HOLGER P. PETERSSON 12.8. Theorem. Suppose ¯F has characteristic 2 and let C be a composition divi- sion algebra of dimension 2n, n = 2, 3, over F that is not a quaternion division algebra of primary type. Given r ∈ Z, texp(C) ≤ r ≤ eF , there exists a separable quadratic subfield L ⊆ C with eL/F = 1, texp(L) = r. Proof. The case r = texp(C) having been settled by Thm. 11.6, we may assume r > texp(C). Applying Cor. 11.9 and then Thm. 11.6, we find a composition subalgebra B ⊆ C and a separable quadratic subfield K ⊆ B with dimF (B) = 2n−1, eK/F = eB/F = 1, texp(K) = texp(B) = texp(C). Suppose for the time being that the case n = 2 has been solved and let n = 3. The quaternion algebra B, having ramification index 1, cannot be of primary type, allowing us to apply the case n = 2 to B in place of C and leading us to the desired conclusion. We are thus reduced to the case n = 2, which we will assume for the rest of the proof. Then K = B and C has dimension 4, hence is not of primary type by hypothesis. We are therefore lead to a unit µ ∈ o× with C = Cay(K, µ) = K ⊕ Kj. (1) Let us first assume texp(K) = texp(C) = 0, so K is tame. Since C is a division algebra, we conclude ¯µ /∈ n ¯K( ¯K ×) from Prop. 9.3. In particular, we have ¯µ /∈ ¯F 2. The hypothesis texp(K) = 0 yields an element v ∈ oK having tK(v) = 1. Observing (1), we put w := πrv + j ∈ C and obtain tC(w) = πr, nC(w) = π2rnK(v) − µ, nC (w) = ¯µ /∈ ¯F 2. In particular, w ∈ o× C \ F 1C and we conclude from Prop. 11.3 that L = F [w] ⊆ C is a separable quadratic subfield of the desired kind. We are left with the case texp(K) = texp(C) > 0. Then K is wild, and Cor. 11.4 yields an element u ∈ o× (2) K with tK(u) = πtexp(K), nK(u) /∈ ¯F 2. By (1) and Cor. 9.14, we may assume (3) ¯µ /∈ ¯F 2 or µ = 1 − πβ for some β ∈ o× according as C has ramification index 1 or 2. We now put w :=(πr−texp(K)u + j ∈ C if eC/F = 1, πr−texp(K)u + uj ∈ C if eC/F = 2. Then tC (w) = πr by (2), and for eC/F = 1 we obtain nC(w) = ¯µ /∈ ¯F 2 by (3). Similarly, for eC/F = 2, we obtain nC(w) = nK(u) /∈ ¯F 2 by (2),(3). In either case, w ∈ o× C \ k1C generates a separable quadratic subfield L := F [w] ⊆ C of ramification index eL/F = 1 and trace exponent texp(L) = r (Prop. 11.3). (cid:3) 12.9. Corollary. Suppose ¯F has characteristic 2 and let C be an octonion division algebra over F that is not of primary type. Given r, s ∈ Z, texp(C) ≤ r ≤ s ≤ eF , there exists a filtration L ⊆ B ⊆ C consisting of a quaternion subalgebra B ⊆ C and a separable quadratic subfield L ⊆ B such that eL/F = eB/F = 1 and texp(B) = r, texp(L) = s. WILD PFISTER FORMS 63 It suffices to construct a quaternion subalgebra B ⊆ C with eB/F = 1, Proof. texp(B) = r because Thm. 12.8 applies to such a B and also yields an L with the desired properties. To construct B, we first invoke Thm. 11.6 and Cor. 11.9 to find a quaternion eL1/F = eB1/F = 1, subalgebra B1 ⊆ C and a separable quadratic subfield L1 ⊆ B1 satisfying (1) By hypothesis, (1) and Cor. 9.14 yield units µ1, µ2 ∈ o× with B1 = Cay(L1, µ1) = L1 ⊕ L1j1, µ1 /∈ ¯F 2, (2) C = Cay(B1, µ2) = B1 ⊕ B1j2, texp(L1) = texp(B1) = texp(C). where µ2 /∈ ¯F 2 for eC/F = 1 and µ2 = 1 − πβ2, β2 ∈ o× for eC/F = 2. By the same token, B2 := Cay(L1, µ2) = L1 ⊕ L1j2 ⊆ C is a quaternion subalgebra not of primary type by Cor. 12.4 with texp(B2) = texp(L1) = texp(C) ≤ r ≤ eF , eB2/F = eC/F . Hence Thm. 12.8 leads us to a separable quadratic subfield L ⊆ B2 with eL/F = 1, texp(L) = r and (2) combines with Cor. 9.14 (b) to show that the quaternion subalgebra B = Cay(L, µ1) = L ⊕ Lj1 ⊆ C satisfies eB/F = 1, texp(B) = texp(L) = r. (cid:3) 12.10. Heights. Over a Henselian field having residual characteristic p > 0, the height as an important invariant of a central associative division algebra of degree p over F has been considered by Saltman [45, pp. 1757, 1765-6] (who uses the term "level"), Kato [23, § 1] (who calls it the "ramification number"), and Tignol [51, 3.2]. It is, in particular, Tignol's approach that suggests two immediate translations to the setting of composition algebras. Let C be a composition division algebra over our 2-Henselian field F . We use the maps hcom : C× × C× → Q∞, hass : C× × C× × C× → Q∞ given by (1) hcom(x, y) := λC ([x, y]) − λC (x) − λC (y) ≥ 0, hass(x, y, z) := λC ([x, y, z]) − λC (x) − λC (y) − λC (z) ≥ 0 (2) for all x, y, z ∈ C× to define (3) (4) hgtcom(C) := inf {hcom(x, y) x, y ∈ C×}, hgtass(C) := inf {hass(x, y, z) x, y, z ∈ C×} and to call these numbers the commutative height and the associative height of C, respectively. If C has dimension at most 2, then hgtcom(C) = hgtass(C) = ∞. On the other hand, if C is a quaternion algebra, then hgtcom(C) < ∞ = hgtass(C) and hgtcom(C) agrees with what Tignol calls its height, while if C is an octonion algebra, its commutative and its associative height are both finite. A general theorem of Tignol [51, 3.12] implies hgtcom(C) = ω(C) for any quater- nion division algebra C over F . This special observation is part of a much more general picture that will be summarized in the following theorem, whose proof in the quaternionic case is independent of [51] and, in fact, works uniformly in the octonionic case as well. 64 SKIP GARIBALDI AND HOLGER P. PETERSSON 12.11. Theorem. If C is a quaternion division algebra over F , then If C is an octonion division algebra over F , then hgtcom(C) = ω(C). hgtcom(C) = hgtass(C) = ω(C). Proof. Let C be a composition division algebra of dimension 2n, n = 2, 3, over F . We must show (1) (2) hgtcom(C) = ω(C), hgtcom(C) = hgtass(C) = ω(C) (for n = 3). To do so, we combine Thm. 2.8 with (10.10.1) to obtain λC ([x1, x2, x3]) ≥ ω(C) + λC (x1) + λC (x2) + λC (x3) for all x1, x2, x3 ∈ C. Combining this and (10.10.2) with (12.10.1−4), we conclude (3) hgtcom(C) ≥ ω(C), hgtass(C) ≥ ω(C). To complete the proof of (1),(2), it therefore suffices to show that (4) (5) hcom(x1, x2) = ω(C) hass(x1, x2, x3) = ω(C) (for some x1, x2 ∈ C), (for n = 3 and some x1, x2, x3 ∈ C). For this purpose, we require two additional formulas: suppose C = Cay(B, µ) = B⊕Bj is a Cayley-Dickson construction as in 1.10, for some associative composition algebra B over F and some scalar µ ∈ F ×. Then a straightforward application of (1.10.1) yields (6) (7) [u, j] = (u − u∗)j [u1, u2, j] = [u1, u2]j (u ∈ B), (u1, u2 ∈ B). In order to establish (4),(5), we distinguish the following cases. Case 1. eC/F = 1. Then ω(C) = texp(C) by Prop. 8.5 (c). If C is tame, the ¯C is a composition division algebra of dimension 2n over ¯F , so there are x1, x2, x3 ∈ o× C , and even [x1, x2, x3] ∈ o× C for n = 3. By (12.10.1),(12.10.2), this implies hcom(x1, x2) = 0 = texp(C), and even hass(x1, x2, x3) = 0 = texp(C) for n = 3, proving (4),(5) in the tame case. C with [x1, x2] ∈ o× If C is wild, it must be of unitary type since eC/F = 1, and Thm. 9.13 combined with Prop. 11.2 implies C = Cay(B, µ), B an associative composition division algebra over F with eB/F = 1, µ = 1 − πdβ, d ∈ Z even, 0 ≤ d < 2 texp(B), β ∈ o, ¯β /∈ ¯B2. In particular, taking into account 11.1, we are in the situation of Prop. 9.7. Applying Cor. 11.8, we find an element u ∈ o× B such that λB(u − u∗) = texp(B). Hence (9.7.2),(9.7.3),(9.7.6) and (6) yield hcom(u, Ξ) = λC ([u, Ξ]) − λC (u) − λC (Ξ) = λC (π− d 2 [u, j]) d 2 = texp(C). = λB(u − u∗) − = texp(B) − d 2 Thus (4) holds for x1 = u, x2 = Ξ, and we have established (1) in Case 1. WILD PFISTER FORMS 65 If n = 3 (C still assumed to be wild), Case 1 applies to B, so (4) yields elements u1, u2 ∈ B such that hcom(u1, u2) = texp(B). Hence (7) and (9.7.6) imply hass([u1, u2, Ξ]) = λC (π− d 2 [u1, u2, j]) − λB(u1) − λB(u2) = hcom(u1, u2) − = texp(B) − = texp(C), d 2 d 2 giving (5) for x1 = u1, x2 = u2, x3 = Ξ, and settling Case 1 completely. Case 2. eC/F = 2. If C is of primary type, then ω(C) = texp(C) by Thm. 12.3 (a), and C = Cay(B, µ) with B as before and µ a prime element in o. Applying Thm. 9.13 (a) and picking u ∈ o× B with texp(C) = texp(B) = λB(u − u∗), we obtain hcom(u, j) = λC ([u, j]) − λB(u) − λC (j) = λC(cid:0)(u − u∗)j(cid:1) − λC (j) = λB(u − u∗) = texp(C), and (4) holds for x1 = u, x2 = j. Moreover, for n = 3, Case 1 applies to B and yields u1, u2 ∈ B× having hcom(u1, u2) = texp(B) = texp(C), allowing us to compute hass(u1, u2, j) = λC ([u1, u2]j) − λB(u1) − λB(u2) − λC (j) = hcom(u1, u2) = texp(C) and completing the proof of (5) for x1 = u1, x2 = u2, x3 = j. Now suppose C is of unitary type. Then C is wild since eC/F = 2, and Thm. 12.3 (b) shows ω(C) = texp(C) − 1 2 . This time, Thm. 9.13 implies C = Cay(B, µ) with B as before, µ = 1 − πdβ, d ∈ Z odd, 0 ≤ d < 2 texp(B), β ∈ o×, so in view of 11.1 we are in the situation of Prop. 9.6. Picking again an element u ∈ o× B with texp(B) = λB(u − u∗), we obtain hcom(u, Π) = λC ([u, Π]) − λB(u) − λC (Π) = λC (π− d−1 = λC(cid:0)(u − u∗)j(cid:1) − d − 1 2 − = texp(B) − d − 1 2 − 1 2 1 2 = λB(u − u∗) − = ω(C), 1 2 = texp(C) − 2 [u, j]) − d − 1 1 2 2 − 1 2 and (4) holds for x1 = u, x2 = Π. If, in addition, n = 3, then Case 1 applies to B, yielding u1, u2 ∈ B× with hcom(u1, u2) = texp(B). Hence = λC (π− d−1 hass(u1, u2, Π) = λC ([u1, u2, Π]) − λB(u1) − λB(u2) − λC (Π) 1 2 [u1, u2, j]) − λB(u1) − λB(u2) − 2 1 2 − λB(u1) − λB(u2) − 2 d − 1 2 − 2 − = ω(C). = λC ([u1, u2]j) − = hcom(u1, u2) − 1 = texp(C) − 2 d − 1 d − 1 = texp(B) − 1 2 1 2 (cid:3) 66 SKIP GARIBALDI AND HOLGER P. PETERSSON Part III. Connections with K-theory 13. Introduction to Part III 13.1. The goal of this part of the paper is to translate the results of Part II into the language of Kato's filtration on Milnor K-theory mod 2. Because it costs little extra, we will also give a dictionary relating traditional valuation-theoretic terms on associative division algebras of prime degree p with Milnor K-theory mod p. For the remainder of the paper, we fix a field F of characteristic zero that has a Henselian discrete valuation λ and residue field F of prime characteristic p. We assume that F contains a primitive p-th root of unity ζ and set m := p · λ(p) p − 1 . Recall that the Milnor K-ring of F , denoted K M One writes {a1, . . . , aq} for the image of a1 ⊗ ··· ⊗ aq in K M This is an integer divisible by p because λ(p)/(p − 1) = λ(ζ − 1), see, e.g., [10, 4.1.2(i)]. ∗ (F ), is the tensor algebra (over Z) of the abelian group F × modulo the "Steinberg relation" a ⊗ (1 − a) = 0 for a ∈ F , a 6= 0, 1. q (F ). We put kq(F ) for K M q (F )/p, and we abuse notation by writing {a1, . . . , aq} also for the image of that element in kq(F ); such a class in kq(F ) is called a symbol. See, e.g., [19] for basic properties. Kato, Bloch, and Gabber proved that kq(F ) is isomorphic to H q(F, µ⊗q p ) via the "Galois symbol", which sends {a1, . . . , aq} 7→ (a1)· (a2)··· (aq); this identifies nonzero symbols in kq(F ) with nonzero symbols in H q(F, µ⊗q p ). We are mainly interested in the following cases: (2) q = 2: H 2(F, µ⊗2 (1) q = 1: k1(F ) and H 1(F, µp) are naturally identified with F ×/F ×p. A nonzero element xF ×p defines a degree p extension F (χ) such that χp = x. p ) is identified (via ζ) with H 2(F, µp), i.e., the p-torsion in the Brauer group of F . We fix the identification with the Brauer group so that the nonzero symbol {x, y} in k2(F ) is sent to the associative central division F -algebra of dimension p2 generated by elements χ, ψ satisfying χp = x, ψp = y, and χψ = ζψχ. (3) p = 2: In this case, there is a bijection between symbols in kq(F ) and q-Pfister (quadratic) forms given by sending {a1, . . . , aq} to hha1, . . . , aqii. For q ≤ 3, we can of course further identify anisotropic q-Pfister forms with composition algebras of dimension 2q. For a nonzero symbol γ ∈ kq(F ) from cases (1) or (2), write D for the corre- sponding division F -algebra. As the valuation λ is Henselian, it extends to a dis- crete valuation λD on D via the usual formula λD(x) := λ(NF (x)/F (x))/[F (x) : F ], cf. (7.14.1). The definition of residue division algebra D, ramification index eD/F , etc., is the same as for quaternion algebras, and the fundamental relation of Prop. 7.13 holds, see [54, p. 393] for references. 13.2. Below, we will recall the filtration on kq(F ) and define invariants eγ (= 1 or p) and depth(γ) for a symbol γ ∈ kq(F ) in terms of K-theory. We will prove K-theoretic analogues of the Local Norm Theorem 8.10 (§15), the Normal Form Theorems 11.11 and 11.12 (§16), and Theorem 9.9 (§18). These proofs use the WILD PFISTER FORMS 67 background results on quadratic forms over Henselian fields from §7 but not the deeper results from the rest of Part II. In the final sections of the paper (§§19 -- 22) we give a dictionary between prop- erties of symbols in kq(F ) in the cases q = 1, q = 2, or p = 2. The proofs in the case p = 2 rely heavily on the full strength of the results in Part II. 14. The filtration on K-theory We now recall the Kato filtration on Milnor K-theory over F , together with the isomorphisms of the graded components with various modules of differential forms, etc., over F . 14.1. Filtration and depth. Write o for the valuation ring on F and p for its maximal ideal. One can filter o as o \ {0} = U0 ⊇ U1 ⊇ U2 ⊇ ··· where Ui := 1 + pi for i ≥ 1. for u ∈ Ui defines a filtration For q ≥ 1, setting U ikq(F ) to be additively generated by elements {u} · kq−1(F ) kq(F ) = U 0kq(F ) ⊇ U 1kq(F ) ⊇ ··· For i > m, Ui consists of p-th powers [10, 4.1.2(ii)], so U ikq(F ) is zero. The depth of γ ∈ kq(F ) is the supremum of {i γ ∈ U ikq(F )}. The only element of depth > m is zero, which has depth ∞. The filtration is compatible with the product in the sense that U ikr(F ) · U jks(F ) ⊆ U i+jkr+s(F ) by [10, 4.1.1b]. Said differently, for elements α ∈ kr(F ) and β ∈ ks(F ), we have depth(α) + depth(β) ≤ depth(α · β); (1) this inequality can be strict, see Example 20.4. 14.2. Kato isomorphisms. For nonzero γ ∈ kq(F ) with q ≥ 1 of depth d, we consider the (nonzero) image of γ in grd kq(F ) := U dkq(F )/U d+1kq(F ); this is the initial form of γ. The results of Kato, et al, include specific isomorphisms: grd kq(F ) ∼=  kq(F ) ⊕ kq−1(F ) Ωq−1 Zq−1 ⊕ Ωq−2 Ωq−1 H 1(F , ν(q − 1)) ⊕ H 1(F , ν(q − 2)) Zq−2 if d = 0; if 0 < d < m and p does not divide d; if 0 < d < m and p divides d; if d = m. Here Ω1 denotes the F -vector space of derivations F → F , Ωq := ∧qΩ1 for q ≥ 1, Ω0 = F and Ω−1 = {0}. The subspace Z q is the kernel of the differential Ωq → Ωq+1, i.e., Z q is the subspace of exact forms. The groups ν(q) are defined in terms of the Cartier operator [10, pp. 4,5]; they are chosen so that H 1(F , ν(q − 1)) in characteristic p plays the role of the Galois cohomology group H q(K, µ⊗(q−1) ) for K of characteristic 6= p. p 68 SKIP GARIBALDI AND HOLGER P. PETERSSON We refer to these isomorphisms as the "Kato isomorphisms". Fix a uniformizer π for λ and write ai and b for elements of o×. The isomorphisms are: d d = 0 p does not divide d p divides d and d 6= 0, m d = m map {a1, a2, . . . , aq}7→({¯a1, . . . , ¯aq}, 0) {π, a1, . . . , aq−1}7→(0,{¯a1, . . . , ¯aq−1}) ¯a1 ∧ ··· ∧ d¯aq−1 ¯a1 ∧ ··· ∧ d¯aq−1 , 0) ¯a1 ∧ ··· ∧ d¯aq−2 ¯a1 ∧ ··· ∧ d¯aq−1 , 0) ¯a1 ∧ ··· ∧ d¯aq−2 {1 + bπd, a1, . . . , aq−1}7→¯b d¯a1 {1 + bπd, a1, . . . , aq−1}7→(¯b d¯a1 {π, 1 + bπd, a1, . . . , aq−2}7→(0, ¯b d¯a1 {1 + b(ζ − 1)p, a1, . . . , aq−1}7→(¯b d¯a1 {π, 1 + b(ζ − 1)p, a1, . . . , aq−2}7→(0, ¯b d¯a1 ¯aq−2 ¯aq−1 ¯aq−1 ¯aq−2 ¯aq−1 ) ) The description of gr0 kq(F ) is a result of Bass-Tate that holds without restriction on the characteristic of F . For depth m, of course (ζ − 1)p has value m, and in case p = 2 the expression (ζ − 1)p is 4. 14.3. We mention for later reference a useful fact about kq(F ) for q ≥ 1. As with any field, there is a group homomorphism K M q (F ) → Ωq defined by {x1, . . . , xq} 7→ x1 ∧ ··· ∧ dxq dx1 (one checks the Steinberg relation). But F has characteristic p, so this homomorphism induces a homomorphism ψ : kq(F ) → Ωq. Moreover, ψ is injective by [4, 2.1] or [19, 9.7.1]. In summary, we have: for x1, . . . , xq ∈ F , the following are equivalent: xq × (i) The symbol {x1, . . . , xq} is zero in kq(F ). (ii) dx1 ∧ ··· ∧ dxq is zero in Ωq. (iii) The elements p√x1, . . . , p√xq are not p-free over F . The equivalence of (ii) and (iii) is [6, §V.13.2, Th. 1]. For further statements along these lines, see e.g. [20, 8.1]. × with dx1∧···∧dxq 6= 0 in Ωq, the preceding equivalence implies that y dx1 ∧ ··· ∧ dxq = 0 in Ωq/Z q if and only if y is a p-th power in F ( p√x1, . . . , p√xq). Given y, x1, . . . , xq ∈ F 15. The Local Norm Theorem 8.10 revisited We now prove an analogue of the Local Norm Theorem 8.10. We continue the notation of 13.1, and focus on a nonzero symbol γ ∈ kq(F ) where q = 1, q = 2, or p = 2. The symbol γ corresponds to a Galois field extension of F of degree p, a (central) associative division algebra of dimension p2 over F , or a q-Pfister quadratic form over F . Write V for the underlying vector space, which has dimension pq. In each case, there is a canonical choice of homogeneous polynomial f : V → F of degree p and representing 1: the norm, the reduced norm, or the quadratic form itself. Further, the valuation λ extends to a valuation λV on V via the formula λV (v) := λ(f (v))/p, and we write o× V for the set of v ∈ V with value zero. For a given a ∈ o×, we ask: How close is f to representing a? Imitating the definition in 8.8, we put It is obviously equivalent to define nexpγ(a) to be the supremum of all d ≥ 0 such that there exist β ∈ o and v ∈ o× nexpγ(a) := sup(cid:8)λ(a − f (v)) v ∈ o× V(cid:9) . V such that a = (1 − πdβ) f (v). WILD PFISTER FORMS 69 15.1. Example. As f represents 1, depth{a} ≤ nexpγ(a) (a ∈ o×). Moreover, we have equality in case p = 2 and q = 0 (so necessarily f is the quadratic form h1i and γ = 1 ∈ Z/2Z = k0(F )). 15.2. Local Norm Theorem. Suppose p = 2 or 1 ≤ q ≤ 2. For a ∈ o× and a symbol γ ∈ kq(F ), we have: (1) nexpγ(a) + depth γ ≤ depth({a} · γ). Furthermore, the following are equivalent: (i) {a} · γ = 0. (ii) a ∈ f (V ). (iii) nexpγ(a) > m − depth γ. (iv) nexpγ(a) = ∞. The number m was defined in 13.1 to be p λ(p)/(p− 1). Note that the inequality (15.2.1) apparently strengthens (14.1.1) by Example 15.1. Proof. (i) and (ii) are known to be equivalent: for q = 1, it is [19, 4.7.5] and for p = 2 it is Prop. 7.5. For q = 2, the implication (ii) ⇒ (i) is elementary and the converse is due to Merkurjev-Suslin [34, 12.2]. Assume (ii), i.e., f (v) = a for some v ∈ V . Then λV (v) = λ(a)/p = 0, so v ∈ o× We now prove equation (1). Suppose a = (1 − πdβ) f (v) for some d ≥ 0, β ∈ o and (iv) is obvious, hence also (iii). V and v ∈ o× V . Then {a} · γ = {f (v)} · γ + {1 − πdβ} · γ. But the first term on the right side is zero by the equivalence of (i),(ii) already established. Hence depth({a} · γ) = depth({1 − πdβ} · γ) ≥ d + depth γ, and we have proved (1). Finally, suppose (iii). By (1), the symbol {a} · γ has depth > m, hence the (cid:3) symbol is zero, proving (i). In order to compare this result with the Local Norm Theorem 8.10, we must relate ω with depth; for the purposes of this discussion, let us focus on the case p = 2 and put ω(γ) := m − (depth γ)/2. (This agrees with the definition of ω for composition algebras in case additionally q = 2 or 3 by Cor. 19.3(ii) below.) Translating Equation 15.2.1 into this notation gives: That is, Theorem 15.2 sharpens Theorem 8.10. nexpγ(a) ≤ 2 ω(γ) − 2 ω({a} · γ). 15.3. Remark. If every finite extension of F has dimension a power of p ("F is p-special") for some prime p, then Theorem 15.2 holds for that prime p and all q ≥ 1 if one adjusts slightly the statement of (ii). The adjusted statement should be in terms of a norm variety for γ as is obvious from [48, Prop. 2.4]; we leave the details to the reader. The paper [48] also provides the proof of the equivalence of (i) and the adjusted form of (ii). 70 SKIP GARIBALDI AND HOLGER P. PETERSSON 16. Gathering the depth We maintain the notation of 13.1. Recall that U dkq(F ) is generated as an abelian group by Ud · kq−1(F ). In this section, we prove that a symbol in U dkq(F ) can be written as {u} · α where u ∈ Ud and α is a symbol in kq−1(F ) (and not just as a sum of such things). More precisely, we have: 16.1. Gathering Lemma. For every nonzero symbol γ ∈ kq(F ) with q ≥ 2, there is a u ∈ o× and a symbol α ∈ kq−1(F ) such that γ = {u}· α, depth{u} = depth γ, and depth α = 0. The symbol α may be chosen to be {a2, . . . , aq} with 0 ≤ λ(a2) < p and λ(ai) = 0 for 3 ≤ i ≤ q. If depth γ is not divisible by p, we may further arrange that λ(a2) has any pre-assigned value j = 0, . . . , p − 1 as desired. One should compare this lemma in the case p = 2 and q = 2, 3 with the Normal Form Theorems 11.11 and 11.12. Heuristically speaking, here we "gather the depth in the first slot" u. The Normal Form Theorems (in view of Th. 19.2 below and with u replaced by L) do the same, except when C is of unitary type, where they take depth L = depth C − 1. We first amass some preliminary results; the proof of the Gathering Lemma will come at the end of the section. We use only background material on K-theory including the material summarized in §14; we don't use anything else from this paper. 16.2. We write O(πi) for an unspecified element (possibly zero) of o divisible by πi. We have the trivial but useful observation: u + O(πj ) = u(1 + O(πj )) for u ∈ o× and j ≥ 0. Indeed, for b ∈ o, we have: u + bπj = u(cid:0)1 + (u−1b)πj(cid:1). 16.3. Example (cf. [4, p. 122]). Suppose that y ∈ o has ¯y = ¯cp for some c ∈ o and fix 0 ≤ s < λ(p)/(p − 1) such that 1 + yπps ∈ o×. Then 1 + yπps = 1 + cpπps + O(πps+1) = (1 + cπs)p + O(πps+1) where the second equality is because λ(cid:0)πis(cid:0)p i(cid:1)(cid:1) ≥ is + λ(p) > is + (p − 1)s ≥ ps for 1 ≤ i < p. (1) As 1 + yπps is a unit, so is (1 + cπs)p, and 16.2 gives that 1 + yπps = (1 + cπs)p (1 + O(πps+1)), hence {1+yπps} = {1+O(πps+1)} in k1(F ). Noting that the hypotheses on y obviously depend only on ¯y (or applying 16.2 once more), we find: {1 + yπps + O(πps+1)} = {1 + O(πps+1)} in k1(F ). 16.4. Lemma. Let a, b ∈ o, i ≥ 0, j ≥ 1 with 1 + aπi ∈ o×. Then in kq(F ) we have: {1 + aπi, 1 + bπj} =n1 + cπi+j, dπi(p−1)o for some nonzero c, d ∈ o. Further, if a, b ∈ o×, then also c, d ∈ o×. Proof. The computations in the proof of [10, 4.1.1b] or [4, p. 122] yield: {1 + aπi, 1 + bπj} = −(cid:26)1 + ab 1 + aπi πi+j ,−aπi(1 + bπj)(cid:27) ∈ k2(F ). WILD PFISTER FORMS 71 (cid:3) πi(p−1)(cid:27) . As −1 = p − 1 in k0(F ), we have: {1 + aπi, 1 + bπj} =(cid:26)1 + ab 1 + aπi πi+j ,(cid:0)−a(1 + bπj)(cid:1)p−1 16.5. Lemma. Suppose that {1+bπps, a2, . . . , aq} satisfies d¯b∧d¯a2∧d¯a3∧···∧d¯aq = 0 in Ωq, with b, 1 + bπps, a2, . . . , aq ∈ o× and 0 ≤ s < λ(p)/(p − 1). Then {1 + i ∈ o× where depth{u′} > bπps, a2, . . . , aq} is equal to {u′, a′ ps. Proof. We may assume that d¯b is not zero -- hence that ¯b is not a p-th power -- by Example 16.3. This settles the q = 1 case. q} for some u′, a′ 2, . . . , a′ Suppose q ≥ 2 and {¯a2, . . . , ¯aq} is zero in kq−1(F ), so d¯a2∧···∧ d¯aq = 0 by 14.3. We apply the q − 1 case of the lemma with u = aq and s = 0 (so u = aq = 1 + c with c = aq − 1, hence d¯c = d¯aq) to see that {a2, . . . , aq} = {a′ q} where depth{a′ So we may assume that {¯a2, . . . , ¯aq} is not zero. Write [i] for a (q − 1)-tuple (i2, . . . , iq) with 0 ≤ ij < p and put ¯a[i] for ¯ai2 q ∈ F . By 14.3 there are c[i] ∈ o such that pp¯a[i](cid:19)p q} is positive. Then Lemma 16.4 gives the claim. 3 ··· ¯aiq If it happens that ¯c[i] = 0 for all nonzero [i], then ¯b is a p-th power in F and we are done. We now show, roughly speaking, that we can make ¯c[i] zero for all nonzero [i]. ¯b =(cid:18)X[i] =X[i] 2, . . . , a′ ¯c[i] ¯cp [i]¯a[i]. 2 ¯ai3 More precisely, fix a nonzero [i] with ¯c[i] nonzero; choose a specific j0 such that ij0 6= 0. Take E/F to be the extension obtained by adjoining a p-th root α of ij0 (−aj)ij ; obviously α is integral. Take v := c[i]α. As ¯c[i] is not zero, j0 Qj6=j0 a the residue of v does not belong to F , hence vπs is not in F and so has minimal polynomial xp − (c[i]απs)p in F [x]. By degree count, this is also the characteristic polynomial chpolyvπs (x) of vπs as an element of the F -algebra E. It follows that (1) NE/F (1 − vπs) = chpolyvπs(1) = 1 + (−1)pNE/F (v)πps. Next observe that E kills γ := {a2, . . . , aq}: we renumber the aj's so that j0 = 2. In kq−1(E), we have: i2γ = {ai2 2 , a3, . . . , aq} = q Xj=3 (p − ij){−aj, a3, . . . , aq} = 0, where the middle equality is because α is in E. As i2 is not divisible by p, we deduce that γ is zero in kq−1(E), as required. Now the projection formula [19, 7.2.7] gives: 0 = NE/F ({1 − vπs} · γ) = {NE/F (1 − vπs)} · γ in kq(F ). Combining this with (1), we find: {1 + bπps} · γ = {1 + (b + (−1)pNE/F (v))πps + O(πps+1)} · γ. But the residue of NE/F (v) is ¯vp = ¯cp [i]¯a[i]. In this way, we have replaced b with a new one that has fewer nonzero coefficients ¯c[i]. Repeating this process completes the proof. (cid:3) 72 SKIP GARIBALDI AND HOLGER P. PETERSSON Proof of the Gathering Lemma 16.1. First, consider a symbol {x, y} ∈ k2(F ). Sup- pose that neither λ(x) nor λ(y) are divisible by p. Then there is some s such that λ(x) ≡ sλ(y) (mod p) and {x, y} = {x(−y)−s, y} because {−y, y} is zero in k2(F ). As λ(x(−y)−s) = λ(x) − sλ(y), we may assume that x has value 0. Second, we may shuffle the entries in a symbol {x1, . . . , xq} by a permutation σ. We have {x1, . . . , xq−1, xq} = {xσ(1), . . . , xσ(q−1), xsgn σ σ(q) }. Combining the two preceding paragraphs shows that we may write γ = {u} · α for α = {a2, . . . , aq} with u, ai ∈ o× for 3 ≤ i ≤ q. Amongst all such ways of writing γ, fix one with depth{u} maximal. For sake of contradiction, suppose that d := depth{u} < depth γ ≤ m. Put r = 2 if λ(a2) = 0 and r = 3 if 0 < λ(a2) < p. We now inspect the Kato isomorphism at depth d. By hypothesis, γ is zero in grd kq(F ), so has zero image. If d = 0, then {¯u, ¯ar, . . . , ¯aq} is zero in k∗(F ), hence d(¯u − 1F ) ∧ d¯ar ∧ ··· ∧ d¯aq is zero by 14.3. Lemma 16.5 gives a contradiction. The case where d = ps for some 0 < s = d/p < (depth γ)/p < λ(p)/(p − 1) is similar. Finally we suppose d is not divisible by p. We claim that λ(a2) may be freely Indeed, for j ∈ Z, λ(a2) − j ≡ sd (mod p) for some s ≥ 0. Writing chosen. u = 1 + bπd for b ∈ o×, we have: {1 + bπd, a2} = {1 + bπd, a2} − s{1 + bπd,−bπd} = {1 + bπd, a2(−b)−sπ−sd}, where λ(a2(−b)−sπ−sd) ≡ j (mod p), proving the claim. The hypothesis that {u}·α has depth greater than d implies that d¯a2 ∧ ··· ∧ d¯aq is zero, and applying Lemma 16.5 to α with s = 0 shows that we may assume that one of the ai has residue 1; Lemma 16.4 gives a contradiction. (cid:3) 16.6. Remark. We can quickly deduce a useful restatement of the Gathering Lemma. Fix some r ≥ 1 and a permutation σ of {2, . . . , q} and put β := {u, aσ(2), . . . , aσ(r)} and δ := {aσ(r+1), aσ(r+2), . . . , aσ(q)}. Because of the identity {x, y} = −{y, x} in K M depth β = depth γ, γ = ±β · δ, 2 (F ), we find: and depth δ = 0. Indeed, for the second and third equalities, ≥ is obvious and ≤ follows from the first equality and equation (14.1.1). We will apply this in the case p = 2, so the sign in the first equation will be irrelevant. 17. Ramification index for symbols 17.1. Definition. For a class γ ∈ kq(F ), we put eγ = p if • depth(γ) is not divisible by p, or • depth(γ) is divisible by p and its initial form has nonzero projection in the second summand of grd kq(F ). (Note that this condition does not depend on the choice of uniformizer π.) Otherwise -- or if γ = 0 -- we put eγ = 1. The "ramification index" eγ is more subtle than in the case of good residue characteristic, see Example 20.4 below. But we do have the following positive results: 17.2. Proposition. Let γ ∈ kq(F ), q ≥ 2, be a non-zero symbol. Then there exist a symbol β ∈ kq−1(F ) and an element a ∈ F × such that γ = β · {a}, eβ = 1 and one of the following holds. WILD PFISTER FORMS 73 (i) depth β = 0, depth{a} = depth γ, a ∈ o× and eγ = 1 ⇐⇒ depth γ ≡ 0 mod p. (ii) depth β = depth γ, 0 < λ(a) < p and eγ = p. Proof. Write γ as in the Gathering Lemma 16.1, where we may assume a2, . . . , aq−1 ∈ o×, 0 ≤ λ(aq) < p. If λ(aq) = 0, then the Kato isomorphism at depth zero shows eα = 1, so with β := α and a = u±1 we are in Case (i) since the Kato isomorphism at depth d := depth γ gives eγ = 1 iff d is divisible by p. Now suppose λ(aq) > 0; by the final statement of the Gathering Lemma, we may also assume d ≡ 0 mod p. Arguing as before, in particular consulting the Kato isomorphisms again for the determination of eγ, β := {u, a2, . . . , aq−1} has depth d and eβ = 1, so we are in Case (ii). 17.3. Proposition. For a symbol γ ∈ kq(F ) for q ≥ 2, the following are equivalent: (i) γ = β · {a} for a nonzero symbol β ∈ kq−1(F ) with eβ = 1 and a ∈ F × of (ii) eγ = p and depth γ is divisible by p. value not divisible by p. (cid:3) If these conditions hold, then additionally depth γ = depth β. Proof. The proof amounts to looking at the explicit formulas for the Kato isomor- phisms. (i) ⇒ (ii): The Kato isomorphisms send β to a nonzero symbol ¯β (in some coho- mology group over F ), because eβ = 1. We have depth(γ) ≥ depth(β) by (14.1.1), and an examination of the isomorphisms show that the isomorphism at the depth of β (divisible by p) sends β · {a} to λ(a) ¯β in the second component of the image, which is not zero because all of the targets of the Kato isomorphisms are abelian groups killed by p; that is, eγ = p and depth γ = depth β. (ii) ⇒ (i): By (ii), alternative (i) of Prop. 17.2 does not hold. Hence alternative (ii) does. (cid:3) 18. Theorem 9.9 revisited We will now prove a version of Theorem 9.9 for K-theory mod-2 when F has characteristic 2; in the notation of §14 we restrict to the case p = 2. We replace the anisotropic round quadratic form P with eP/F = 1 from Th. 9.9 with a nonzero symbol γ ∈ kq(F ) with q ≥ 1 and eγ = 1 (in particular, depth γ is even). We replace the hypothesis on texp with the hypothesis depth{µ} + depth γ = m (recall that m = 2λ(2)), the largest possible depth for a nonzero element of kq(F ). The extreme cases where depth γ is 0 or m are comparatively easy, so we focus on the middle case. We will use the following: ¯a2 ∧ ··· ∧ d¯aq 18.1. Technique. If one has a nonzero class ¯b d¯a2 ¯aq ∈ Ωq−1/Z q−1, we may apply d and obtain the nonzero symbol x0 := d¯b ∧ d¯a2 in Ωq. The F -span of this symbol contains x := ¯b−1x0, which lies in ν(q). Indeed, it is the unique nonzero element of F x0 with this property, because ν(q) is defined to be ker(γ − 1) for γ the inverse Cartier operator [10, pp. 123, 124] and for c ∈ F we have (γ − 1)(cx) = (c2 − c)x ∈ Ωq. A canonical isomorphism identifies x with the class of the anisotropic symmetric bilinear form B = hh¯b, ¯a2, . . . , ¯aqii in the graded ¯a2 ∧ ··· ∧ d¯aq ¯aq 74 SKIP GARIBALDI AND HOLGER P. PETERSSON Witt ring [26], hence with B itself [15, 6.20]. The equivalence from Prop. 3.3 takes B and gives an extension K/F of degree 2q with a unital linear form s : K → F . Let γ ∈ kq(F ) be a nonzero symbol of even depth d 6= 0, m and suppose that eγ = 1. Then the initial form of γ is a nonzero symbol in Ωq−1/Z q−1 and the technique in the preceding paragraph gives a (K, s) derived from γ. 18.2. Proposition. Let γ ∈ kq(F ) be a nonzero symbol of even depth d 6= 0, m with eγ = 1. Write (K, s) as in 18.1, and write γ = {1 + xπd} · α with x ∈ o× as in the Gathering Lemma 16.1. For µ = 1 − bπm−d with b ∈ o×, we have: {µ} · γ = 0 in kq+1(F ) if and only if the residue of xbπm/4 is in the image of ℘K,s. Proof. Write α = {a2, . . . , aq} for some a2, . . . , aq ∈ o×. Using Lemma 16.4, we calculate: 1 + xπd πm,−x(1 − bπm−d)(cid:27) . {µ} · {1 + xπd} = {1 + xπd, 1 − bπm−d} =(cid:26)1 − We see from this that the initial form of {µ} · γ is ¯x¯b¯ε d¯x where we have set ε := πm/4 to simplify the notation. This determines the class of the quadratic Pfister form hh¯x, ¯a2, . . . , ¯aq, ¯x¯b¯εK in the graded Witt group of quadratic forms [26]. The Arason-Pfister Hauptsatz [15, 23.7(1)] implies that this class is zero (equivalently, {µ} · γ is zero) if and only if the Pfister form is isotropic, if and only if ¯x¯b¯ε is in the image of ℘K,s by Cor. 3.10(a), proving the claim. ¯a2 ∧ ··· ∧ d¯aq ¯x ∧ d¯a2 bx ¯aq (cid:3) 19. Dictionary between K-theory and algebras and quadratic forms In the cases q = 1, q = 2, or p = 2, we have a close relationship between properties of symbols in kq(F ) relative to Kato's filtration and valuation-theoretic properties on the corresponding algebras. Specifically: 19.1. Proposition. In cases q = 1 and q = 2 we have: (i) eγ = eD/F . (ii) D is a separable division algebra over F and distinct from F if and only if depth(γ) = m. That is, you can determine eD/F and whether or not D is tame by examining the corresponding symbol in kq(F ). We prove the proposition in §20 and 21.2 below. The calculations appearing in the proof of this result are similar to some in sections 1.1 and 2.1 of [52]. Roughly speaking, our statements here differ from those in [52] by using the language of Kato's filtration. In contrast with Proposition 19.1, the following theorem relies heavily on the results of Part II. It is proved in §22. 19.2. Theorem. Let p = 2. Fix a nonzero symbol γ ∈ kq(F ) for some q ≥ 1 and let Q be the corresponding anisotropic q-Pfister form. Then: (i) eγ = eQ/F . (ii) Q is nonsingular (i.e., Q is tame) if and only if depth(γ) = 2λ(2). (iii) texp(Q) = λ(2) −j depth(γ) 2 k. In the statement of the theorem, Q is a quadratic form, whereas the results in §§7 -- 9 (including the definition of tame and texp) are in terms of pointed quadratic spaces. This is harmless in view of Prop. 7.3. The theorem leads to: WILD PFISTER FORMS 75 19.3. Corollary. Fix a division quaternion or octonion algebra C over F and write γ for the corresponding symbol in k∗(F ). We have: (i) C is of primary type if and only if eγ = 2 and depth(γ) is even. (Otherwise 2 C is of unitary type.) (ii) ω(C) = λ(2) − depth(γ) That is, one can read off properties of the composition algebra C -- including the invariant ω(C) studied by Saltman and Tignol -- from the properties of the corresponding symbol γ in Milnor K-theory. We prove Theorem 19.2 and Cor. 19.3 in §22. 19.4. Remark. It is natural to wonder if on can extend Prop. 19.1 to include the case p = q = 3, where the corresponding algebraic objects are Albert (Jordan) algebras obtained from the first Tits construction. The answer is no, because we do not know if two such Albert algebras corresponding to the same symbol in k3(F ) are necessarily isomorphic. (This is a special case of open problem #4 from [42].) If this is indeed the case, then one can easily extend Prop. 19.1 to include the case p = q = 3 by taking advantage of the valuation theoretic results in [40] and by imitating the proof in the cases given below. 20. Proof of Proposition 19.1: case q = 1 20.1. Example (q = 1 and nonzero depth divisible by p). Fix {x} ∈ k1(F ) of depth d divisible by p, so x = 1 + uπd for some u of value 0. As the "second summand" Ω−1/Z −1 or H 1(F , ν(−1)) in the Kato isomorphism is zero, e{x} = 1. The algebra D corresponding to {x} is F (χ) where χp = x. For α := π−d/p(χ − 1) ∈ D (α + π−d/p)p = π−dx = u + π−d, we have so i(cid:19)αiπ−(p−i)d/p − u = 0. p (1) Xi=1(cid:18)p For 1 ≤ i < p, the prime p divides (cid:0)p i(cid:1)π−(p−i)d/p(cid:1) ≥ λ(p) − λ(cid:0)(cid:0)p i(cid:1), so (p − i) = d p (m − d)(p − 1) + d(i − 1) p . As i ≥ 1 and d ≤ m, this is at least 0, hence α is integral. Further, the coefficient of αi in (1) has residue zero for 2 ≤ i < p in all cases and also for i = 1 if d < m. (Clearly, ¯α is not zero, so λ(α) = 0.) Therefore, if d < m, D contains ¯α satisfying ¯αp − ¯u = 0. As x has depth d, the Kato isomorphism shows that ¯u is not a p-th power in F , and we conclude that D is the proper extension F ( p√¯u) and eD/F = 1. In case d = m, we set η := ζ − 1. As λ(η) = m/p = λ(p)/(p − 1), we have x = 1 + bηp for b = uπm ηp . 76 SKIP GARIBALDI AND HOLGER P. PETERSSON We put β := η−1(χ − 1) and apply the same reasoning as in the case d < m with π−d/p replaced with η−1. The element β satisfies (2) p Xi=1(cid:18)p i(cid:19)βiη−(p−i) − b = 0. Again, the coefficients of β are integral because (3) λ(cid:18)(cid:18)p i(cid:19)η−(p−i)(cid:19) ≥ λ(p) − (p − i) λ(p) p − 1 = λ(p) Taking residues of (2) kills the terms with 1 < i < p. For the i = 1 term, we note i=2 ηi and that expanding the equation (1 + η)p = 1 gives Pp Xi=2(cid:18)p pη−(p−1) = pη1−p = − p i − 1 p − 1 ≥ 0 for 1 ≤ i < p. i(cid:1)ηi = 0, so pη = −Pp i=1(cid:0)p i(cid:19)η−(p−i). Taking residues and applying (3), only the i = p term is nonzero on the right side, so pη−(p−1) has residue −1. Taking residues of (2), we find the equation ¯βp − ¯β − ¯b = 0 in D. The fact that x has depth d asserts that the element ¯b is nonzero in H 1(F , ν(0)) ∼= F /℘(F ), i.e., ¯β generates a proper extension of F and we conclude that D/F is unramified. We have verified Proposition 19.1 for γ ∈ k1(F ) of nonzero depth divisible by p. 20.2. Remark. It might be illuminating to compare Example 20.1 in the case p = 2 with the material in Part II. In case 0 < d < m = 2λ(2) = 2 texp(F ), we compare the example to Prop. 9.7 with P = F . The Kato isomorphism at depth d sends {x} to the image of ¯u in ¯F / ¯F 2, and we conclude ¯u /∈ ¯F 2, in agreement with Prop. 9.7. Moreover, identifying F (√x) = Cay(F, x) = F ⊕ F j and j = −χ, we obtain that −α = Ξ in the sense of (9.7.2); it is a normalized trace generator. The situation with d = 2λ(2) is slightly more delicate; we compare the example to Thm. 9.9 (resp. Cor. 10.18) for P (resp. C) = F . First of all, we have ζ = −1, η = −2, β = χ−1 2 ∈ o× is the unique normalized trace generator = ℘ to be the usual Artin-Schreier map of F . Moreover, ¯sw0 = 1 ¯F , forcing ℘ ¯F ,¯sw0 on ¯F . We have u = −β, u0 = nF (w0)u = −β0 in the sense of (9.9.1) and b = u0. The Kato isomorphism at depth m sends {x} to ¯u0 = ¯β0 ∈ H 1( ¯F , ν(0)) = ¯F /℘( ¯F ), forcing ¯β0 /∈ Im(℘) in agreement with the equivalence (iii) ⇔ (v) in Thm. 9.9. Furthermore, by Cor. 10.18, 2 , and w0 = πλ(2) ¯C = ¯F [t]/(t2 − t − ¯u0) = ¯F [t]/(t2 − t − ¯b) = F (√x), in agreement with the second part of Example 20.1. 20.3. Proof of Proposition 19.1 for q = 1: Fix a nonzero {x} ∈ k1(F ) and write D = F (χ) as in Example 20.1. Suppose first that x has depth 0; we may assume that 0 ≤ λ(x) < p. The Kato isomorphism gr0 k1(F ) ∼−→ k1(F ) ⊕ k0(F ) ∼= F ×p × /F ⊕ Z/pZ is the direct sum of the specialization map sπ and the tame symbol ∂, described concretely in [19], e.g. As ∂(x) ≡ λ(x) (mod p), we have e{x} = p if and only if λ(x) is not zero. As to the extension D/F , the element χ satisfies χp − x = 0, hence is integral. If λ(x) is not zero, then ¯χp is zero, hence ¯χ is zero, D = F , and WILD PFISTER FORMS 77 eD/F = p. If λ(x) is zero, then since x has depth 0, ¯x is not a p-th power in F and F ( ¯χ) is a proper extension of F ; in this case D = F ( p√¯x) and eD/F = 1. Suppose that x has depth d not divisible by p; we may take x of the form 1 + uπd where u has value 0. Then ND/F (χ − 1) = (−1)p−1(x − 1) = (−1)p−1uπd, so the value of χ − 1 is not an integer and eD/F = p = e{x}. Note that in both of these cases, D is not a proper separable extension of F . The remaining case where d has nonzero depth divisible by p was treated in Example 20.1, which concludes the proof. (cid:3) 20.4. Example. Let F := F2((x)), Laurent series over the field with 2 elements. Construct a field F of characteristic zero by taking the absolutely unramified Te- ichmuller extension of F as in Example 10.15 and adjoining √2. Then F has residue field F and √2 is a uniformizer. Consider the elements a = 1 +√2 and b = 1 + 2u in F , where u ∈ F is a unit with residue x. The symbols {a} and {b} have depth 3 and 2. The valuation λ ramifies on F (√a) and does not ramify on F (√b). A reader familiar with the situation of good residue characteristic might assume that the quaternion algebra (a, b) is division over F , but in fact it is split. This is easily seen because the symbol {a, b} has depth at least 5 by (14.1.1), which is greater than 2λ(2). 3 21. Proof of Proposition 19.1: cases q = 2 We now prove Proposition 19.1 and Theorem 19.2 in the case q = 2. We consider a nonzero symbol {x, y} in k2(F ) and write D for the corresponding symbol algebra. 21.1. Lemma. eD/F = p if and only if some subfield L of D has eL/F = p. Proof. Easy. If such an L exists, then there is some ℓ ∈ L× such that λ(NL/F (ℓ)) is not divisible by p. But NL/F (ℓ) = NrdD/F (ℓ) by [11, p. 150]. This proves "if". For "only if", note that a nonzero element of D generates a subfield L/F . (cid:3) 21.2. Proof of Proposition 19.1 for q = 2: Suppose first that the depth d of {x, y} is not divisible by p (so e{x,y} = p). By the Gathering Lemma 16.1 we can arrange that x is in Ud and y ∈ o× has residue that is not a p-th power. As in 20.3, eF (χ)/F = p, hence eD/F = p and the dimension of D/F is p. The residue field of F (ψ) is the proper extension F ( p√¯y) of F , and by dimension count it is all of D and D is wild. So we may assume that the depth of {x, y} is divisible by p. Suppose now that e{x,y} = p. By Prop. 17.3, we may assume that e{x} = 1 and y = uπn for some u ∈ o and n not divisible by p, and depth{x, y} = depth{x}. On the one hand, D contains F (ψ) on which the valuation ramifies, so eD/F = p as claimed. On the other hand, D contains F (χ) whose residue algebra is a proper extension of F that is inseparable (if depth{x, y} < m) or separable (if depth{x, y} = m) by Example 20.1; this proves the claim. We are left with the case where the depth is divisible by p and e{x,y} = 1. If the depth of {x, y} is zero, then by the Kato isomorphisms {¯x, ¯y} is not . Following 20.3, ¯χ and ¯ψ generate zero in k2(F ), hence ¯x, ¯y are p-free over F purely inseparable extensions of F in D. The value of χψ − ψχ = (ζ − 1)ψχ is p 78 SKIP GARIBALDI AND HOLGER P. PETERSSON eD/F = 1, and D is wild. λ(ζ − 1) = m/p > 0, so ¯χ and ¯ψ commute in D. We deduce that D is F ( p√¯x, p√¯y), If 0 < depth{x, y} < m, then we can choose x = 1 + aπd where d = depth{x, y} so that the initial form of {x, y} in Ω1/Z 1 is ¯a d¯y ¯y is not in Z 1, ¯y is not 0. It follows that dimF F ( p√¯a, p√¯y) = p2 by 14.3. We claim that i.e., d¯a ∧ d¯y this field is D. By dimension count, it suffices to note that for α := π−d/p(χ − 1), we have ¯y . As the depth is d, ¯a d¯y ¯αp = ¯a, ¯ψp = ¯y, and ¯α ¯ψ = ¯ψ ¯α. The first equation is as in Example 20.1 and the second is obvious. The third follows because αψ − ψα = π−d/p(χψ − ψχ) = π−d/p(ζ − 1)ψχ. But ζ − 1 has value m/p > d/p, hence this commutator has positive value and ¯α, ¯ψ commute. This shows that eD/F = 1 and D is wild. Finally suppose that the depth of {x, y} is m. Then we write x = 1 + bηp and β = η−1(χ − 1) as in Example 20.1; again ¯βp − ¯β − ¯b = 0. Further, βψ − ψβ = η−1(χψ − ψχ) = ψχ, so ¯β ¯ψ = ¯ψ( ¯β + 1). The elements ¯β and ¯ψ generate a division algebra of dimension p2 over its center F [19, p. 36], so it must be D. Therefore, eD/F = 1 and D is tame. (cid:3) 22. Proofs of Theorem 19.2 and Corollary 19.3 The following proofs amount to translating results of §§8, 9 into K-theory. Proof of Theorem 19.2. It suffices to establish (i) and (iii) since by Prop. 8.2(a) (iii) implies (ii). Case q = 1: Suppose first that q = 1, i.e., γ = {µ} for some µ ∈ F × and Q is the 1-Pfister hhµii. Put P := h1i; it is wild because char F = 2 and texp P = λ(2) by Example 8.7. We may assume that µ has value 0 or 1. If µ has value 1, then depth γ = 0, eγ = 2, and the theorem holds by Prop. 9.2.3. If µ has value 0 and ¯µ is a nonsquare in F , then depth γ = 0, eγ = 1, and the theorem holds by Prop. 9.7.6. Otherwise µ is a square in F , so multiplying µ by a square in F we may assume that µ = 1. If depth γ < 2λ(2), then (i) and (iii) hold by Prop. 9.6.5 or Prop. 9.7.6. Finally suppose that depth γ = 2λ(2). The symbol γ corresponds to the qua- dratic extension F (√µ) and the residue algebra of this extension was computed in Example 20.1; this verifies (i) and that Q is tame (hence (ii)). Then texp Q = 0 by Prop. 8.2(d), proving (iii). Case q ≥ 2: We argue by induction on q and decompose γ as in Prop. 17.2 (with p = 2). Writing P (resp. Q) for the Pfister quadratic form corresponding to β (resp. γ), we conclude Q ∼= hhaii ⊗ P and put d := depth γ. Then eP/F = 1 and texp(P ) = λ(2) − (depth β)/2 by the induction hypothesis. If alternative (ii) of Prop. 17.2 holds, then (9.2.3) shows eγ = 2 = eQ/F and texp(Q) = texp(P ) = k since depth γ = depth β is even. Now suppose alternative (i) of λ(2) −j depth γ Prop. 17.2 holds. Since β has depth zero and ramification index 1, we can write β = {a1, . . . , aq−1}, ai ∈ o×, 1 ≤ i < q. If d = 2λ(2), then eγ = 1 and hhaii is a tame 2 WILD PFISTER FORMS 79 1-Pfister quadratic subspace of Q ∼= hha, a1, . . . , aq−1ii with ehhaii/F = 1. Applying Prop. 9.3 q − 1 times yields eQ/F = 1 = eγ and that Q is tame as well, forcing texp(Q) = 0 = λ(2) − ⌊d/2⌋. We are left with the case 0 ≤ d < 2λ(2). Assertions (i) and (iii) follow by combining the effect of the Kato isomorphism at depth d on γ with 14.3 and (9.6.5) (for d odd) or with (9.7.6) and Cor. 9.10(a) (for d even). (cid:3) Proof of Corollary 19.3. We now prove Corollary 19.3. Claim (i) amounts to refor- mulating Proposition 17.3. For (ii), one combines Theorem 19.2(iii) relating depth γ with texp C with the relation between texp C and ω(C) demonstrated in the proof of Theorem 12.11. (cid:3) Acknowledgements. The first author was partially supported by National Science Foun- dation grant no. DMS-0653502. Both authors thank Eric Brussel, Detlev Hoffmann, Daniel Krashen, Ottmar Loos, Pat Morandi, David Saltman, and Jean-Pierre Tignol for useful comments and suggestions. Special thanks are due to Richard Weiss, whose stimulat- ing questions (related to his recent work [56] on affine buildings) inspired us to look for successively improved versions of the Local Norm Theorems 8.10 and 15.2. References [1] A. A. Albert, Absolute valued real algebras, Ann. of Math. (2) 48 (1947), 495 -- 501. MR 0020550 (8,561d) [2] , On nonassociative division algebras, Trans. Amer. Math. Soc. 72 (1952), 296 -- 309. MR 0047027 (13,816d) [3] E. Becker, Uber eine Klasse flexibler quadratischer Divisionsalgebren, J. Reine Angew. Math. 256 (1972), 25 -- 57. MR 0320099 (47 #8640) [4] S. Bloch and K. Kato, p-adic ´etale cohomology, Publ. Math. IHES 63 (1986), 107 -- 152. [5] N. Bourbaki, Elements of mathematics. Algebra, Part I: Chapters 1-3, Hermann, Paris, 1974, Translated from the French. [6] [7] R. B. Brown, On generalized Cayley-Dickson algebras, Pacific J. Math. 20 (1967), 415 -- 422. , Algebra II, Springer-Verlag, 1988. MR 0215891 (35 #6726) [8] F. Bruhat and J. Tits, Groupes r´eductifs sur un corps local, Inst. Hautes ´Etudes Sci. Publ. Math. (1972), no. 41, 5 -- 251. MR 0327923 (48 #6265) [9] , Sch´emas en groupes et immeubles des groupes classiques sur un corps local, Bull. Soc. Math. France 112 (1984), no. 2, 259 -- 301. MR 788969 (86i:20064) [10] J.-L. Colliot-Th´el`ene, Cohomologie galoisienne des corps valu´es discrets henseliens, d'apr`es K. Kato et S. Bloch, Algebraic K-theory and its applications (Trieste, 1997) (H. Bass, A.O. Kuku, and C. Pedrini, eds.), 1999, pp. 120 -- 163. [11] P. K. Draxl, Skew fields, London Math. Soc. Lecture Note Series, vol. 81, Cambridge Univer- sity Press, Cambridge-New York, 1983. [12] A. Dress, Metrische Ebenen uber quadratisch perfekten Korpern, Math. Z. 92 (1966), 19 -- 29. MR 0206804 (34 #6620) [13] H.-D. Ebbinghaus, H. Hermes, F. Hirzebruch, M. Koecher, K. Mainzer, J. Neukirch, A. Pres- tel, and R. Remmert, Numbers, Graduate Texts in Mathematics, vol. 123, Springer, 1991. [14] I. Efrat, Valuations, orderings, and Milnor K-theory, Mathematical Surveys and Mono- graphs, vol. 124, American Mathematical Society, Providence, RI, 2006. MR 2215492 (2007g:12006) [15] R. Elman, N. Karpenko, and A. Merkurjev, The algebraic and geometric theory of quadratic forms, American Mathematical Society Colloquium Publications, vol. 56, American Mathe- matical Society, Providence, RI, 2008. MR 2427530 [16] M.A. Elomary and J.-P. Tignol, Springer's theorem for tame quadratic forms over Henselian fields, preprint, 2010. [17] A. J. Engler and A. Prestel, Valued fields, Springer Monographs in Mathematics, Springer- Verlag, Berlin, 2005. MR 2183496 (2007a:12005) 80 SKIP GARIBALDI AND HOLGER P. PETERSSON [18] S. Garibaldi, A. Merkurjev, and J-P. Serre, Cohomological invariants in Galois cohomology, University Lecture Series, vol. 28, American Mathematical Society, Providence, RI, 2003. MR 1999383 (2004f:11034) [19] P. Gille and T. Szamuely, Central simple algebras and Galois cohomology, Cambridge studies in Advanced Math., vol. 101, Cambridge, 2006. [20] D.W. Hoffmann, Diagonal forms of degree p in characteristic p, Contemporary Math. 344 (2004), 135 -- 183. [21] D.W. Hoffmann and A. Laghribi, Quadratic forms and Pfister neighbors in characteristic 2, Trans. Amer. Math. Soc. 356 (2004), 4019 -- 4053. [22] B. Jacob, Quadratic forms over dyadic valued fields. I. The graded Witt ring, Pacific J. Math. 126 (1987), no. 1, 21 -- 79. MR 868606 (87m:11039) [23] K. Kato, A generalization of local class field theory by using K-groups. I, J. Fac. Sci. Univ. Tokyo IA 26 (1979), 303 -- 376. [24] [25] [26] , A generalization of local class field theory by using K-groups. II, J. Fac. Sci. Univ. Tokyo IA 27 (1980), 603 -- 683. , Galois cohomology of complete discrete valuation fields, Lecture Notes in Math., vol. 967, pp. 215 -- 238, Springer, 1982. , Symmetric bilinear forms, quadratic forms and Milnor K-theory in characteristic two, Invent. Math. 66 (1982), no. 3, 493 -- 510. [27] M. Knebusch, Generic splitting of quadratic forms. I, Proc. London Math. Soc. (3) 33 (1976), no. 1, 65 -- 93. MR 0412101 (54 #230) [28] , Specialization forms, http://www-nw.uni-regensburg.de/~.knm22087.mathematik.uni-regensburg.de/book.pdf, 2007. symmetric bilinear of quadratic and [29] M.-A. Knus, A. Merkurjev, M. Rost, and J.-P. Tignol, The book of involutions, American Mathematical Society Colloquium Publications, vol. 44, American Mathematical Society, Providence, RI, 1998, With a preface in French by J. Tits. MR 2000a:16031 [30] T.Y. Lam, Introduction to quadratic forms over fields, Graduate Studies in Mathematics, vol. 67, Amer. Math. Soc., Providence, RI, 2005. [31] O. Loos, The scheme of quaternionic algebras, To appear. [32] P. J. McCarthy, Algebraic extensions of fields, second ed., Dover Publications Inc., New York, 1991. MR MR1105534 (92a:11138) [33] K. McCrimmon, Nonassociative algebras with scalar involution, Pacific J. Math. 116 (1985), no. 1, 85 -- 109. [34] A.S. Merkurjev and A.A. Suslin, K-cohomology of Severi-Brauer varieties and the norm residue homomorphism, Math. USSR Izv. 21 (1983), no. 2, 307 -- 340. [35] O. T. O'Meara, Introduction to quadratic forms, Grundlehren der mathematischen Wis- senschaften, Vol. 117, Springer-Verlag, New York, 1963. [36] J. M. Osborn, Quadratic division algebras, Trans. Amer. Math. Soc. 105 (1962), 202 -- 221. MR 0140550 (25 #3968) [37] ´A. Rodr´ıguez Palacios, Absolute-valued algebras, and absolute-valuable Banach spaces, Ad- vanced courses of mathematical analysis I, World Sci. Publ., Hackensack, NJ, 2004, pp. 99 -- 155. MR 2162414 (2006j:46013) [38] H. P. Petersson, Borel subalgebras of alternative and Jordan algebras, J. Algebra 16 (1970), 541 -- 560. MR 52 #8209 [39] [40] [41] , Eine Bemerkung zu quadratischen Divisionsalgebren, Arch. Math. (Basel) 22 (1971), 59 -- 61. MR 46 #9132 , Jordan-Divisionsalgebren und Bewertungen, Math. Ann. 202 (1973MR50:422), 215 -- 243. MR 50 #422 , Composition algebras over a field with a discrete valuation, J. Algebra 29 (1974), 414 -- 426. [42] H.P. Petersson and M.L. Racine, Albert algebras, Jordan algebras (Berlin) (W. Kaup, K. Mc- Crimmon, and H.P. Petersson, eds.), de Gruyter, 1994, (Proceedings of a conference at Ober- wolfach, 1992), pp. 197 -- 207. [43] S. Pumplun, A non-orthogonal Cayley-Dickson doubling, J. Algebra Appl. 5 (2006), no. 2, 193 -- 199. MR 2223466 (2007a:17009) [44] F. Rosemeier, Semiquadratische algebren, Ph.D. thesis, FernUniversitat in Hagen, 2002. WILD PFISTER FORMS 81 [45] D. J. Saltman, Division algebras over discrete valued fields, Comm. Algebra 8 (1980), no. 18, 1749 -- 1774. MR MR589083 (83b:16015) [46] T. A. Springer, Quadratic forms over fields with a discrete valuation. I. Equivalence classes of definite forms, Nederl. Akad. Wetensch. Proc. Ser. A. 58 = Indag. Math. 17 (1955), 352 -- 362. MR 0070664 (17,17a) [47] T. A. Springer and F. D. Veldkamp, Octonions, Jordan algebras and exceptional groups, Springer Monographs in Mathematics, Springer-Verlag, Berlin, 2000. [48] A. Suslin and S. Joukhovitski, Norm varieties, J. Pure Appl. Algebra 206 (2006), no. 1-2, 245 -- 276. [49] O. Teichmuller, Diskret bewertete perfekte Korper mit unvollkommenem Restklassenkorper, J. Reine Angew. Math. 176 (1937), 141 -- 152. [50] U.P. Tietze, Zur Theorie quadratischer Formen uber Hensel-Korpern, Arch. Math. 17 (1955), 352 -- 362. [51] J.-P. Tignol, Alg`ebres `a division et extensions de corps sauvagement ramifi´ees de degr´e pre- mier, J. Reine Angew. Math. 404 (1990), 1 -- 38. [52] , Classification of wild cyclic field extensions and division algebras of prime de- gree over a Henselian field, Proceedings of the International Conference on Algebra, Part 2 (Novosibirsk, 1989) (Providence, RI), Contemp. Math., vol. 131, Amer. Math. Soc., 1992, pp. 491 -- 508. [53] F. van der Blij and T. A. Springer, The arithmetics of octaves and of the group G2, Nederl. Akad. Wetensch. Proc. Ser. A 62 = Indag. Math. 21 (1959), 406 -- 418. MR 0152555 (27 #2533) [54] A.R. Wadsworth, Valuation theory on finite dimensional division algebras, Valuation theory and its applications, Vol. I (Saskatoon, SK, 1999), Fields Inst. Commun., vol. 32, Amer. Math. Soc., Providence, RI, 2002, pp. 385 -- 449. [55] R. M. Weiss, Quadrangular algebras, Mathematical Notes, vol. 46, Princeton University Press, Princeton, NJ, 2006. MR 2177056 (2007a:17010) [56] , On the existence of certain affine buildings of type E6 and E7, J. Reine Angew. Math. (To appear). [57] K. A. Zhevlakov, A. M. Slin′ko, I. P. Shestakov, and A. I. Shirshov, Rings that are nearly associative, Pure and Applied Mathematics, Vol. 104, Academic Press, New York, 1982. (Garibaldi) Department of Mathematics and Computer Science, Emory University, Atlanta, GA 30322, USA E-mail address: [email protected] (Petersson) Fakultat fur Mathematik und Informatik, FernUniversitat in Hagen, D-58084 Hagen, Germany E-mail address: [email protected]
1302.3345	2	1302	2013-03-01T11:51:05	On some basic properties of Leibniz algebras	[ "math.RA", "math.DG" ]	This paper gives an overview of some basic properties of Leibniz algebras. Some of the results were known earlier, but in the article they are accompanied by new simple proofs. Some of the results are new. The article can be viewed as a digest or a mini-manual for the basic theory of Leibniz algebras	math.RA	math	ON SOME BASIC PROPERTIES OF LEIBNIZ ALGEBRAS V.Gorbatsevich Abstract. This paper gives an overview of some basic properties of Leibniz alge- bras. Some of the results were known earlier, but in the article they are accompanied by new simple proofs. Some of the results are new. The article can be viewed as a digest or a mini-manual for the basic theory of Leibniz algebras 3 1 0 2 r a M 1 ] . A R h t a m [ 2 v 5 4 3 3 . 2 0 3 1 : v i X r a The main topic of this article is a liezation of Leibniz algebras (the procedure of transition to Lie algebras), which allows to prove many of the properties of Leibniz algebras with aid of known results about Lie algebras. We prove some results recently obtained by various authors. Most of the results below are direct analogs of known results for Lie algebras and almost all of them are proven in many recent articles on the theory of Leibniz algebras (we do not give detailed references here, as the results are very scattered and not always published in the print media, some of them - only in the electronic form). New in this paper is a unified approach and the allocation of some fundamental statements about Leibniz algebras. The article can be considered as a sort of digest or a mini-tutorial on the basics of the theory of Leibniz algebras. We will consider finite dimensional algebras L over a field k of characteristic 0 (in fact it is only important that this characteristic is not equal to 2). Linear mapping D : L → L is called a derivation of L, if D(x · y) = Dx · y + x · D(y) for all x, y ∈ L (where "·" denotes the multiplication in L). The set of all derivations of L is denoted by Der(L). Due to the operation of commutation of linear operators Der(L) is a Lie algebra. Naturally can be defined the automorphism group Aut(L) of the algebra L. If the field k is R or C, then the automorphism group is a Lie group and the Lie algebra Der(L) be its Lie algebra. One can consider Aut(L) as an algebraic group (or as a group of k-points of an algebraic group defined over a field k). An algebra L over k is called a left Leibniz algebra if for every x ∈ L the cor- responding operator lx of left multiplication is a derivation of L, i.e. Lx ∈ Der(L) (Leibniz axiom). Similarly we define the right Leibniz algebra - here operators rx of right multiplication have to be derivations of L. Leibniz algebra is a gen- eralization of the notion of Lie algebras, which are both left and right Leibniz algebra and their multiplications are skew. So we can say that the Leibniz algebra is a "nonanticommutative" analog of Lie algebra. The term "nonanticommutative" looks very cumbersome (and contains a double negative), sometimes it is used the term "non-commutative" analogue of Lie algebras, although the term is inaccurate - it implicitly assumes that the main operation in the Lie algebras is commutative. Leibniz algebra at first introduced and investigated in the papers of A.M.Bloh [1] [2] (he called this algebras as D− algebras, directly pointing to their connection 1 Typeset by AMS-TEX with derivations). They have not received a worthy continuation immediately after writing. Then Leibniz algebras were independently rediscovered by Jean-L.Loday (12.01.1946 - 06.06.2012, drowned in the sea riding on a yacht) [3], who called them the Leibniz algebras, since it was Leibniz who proved "Leibniz rule" for differentia- tion of functions (although there is no information about any "Leibniz algebra" in the works of Leibniz). In the works of Loday (partially - with co-authors), the the- ory of Leibniz algebras got developed. Some authors call these algebra as Loday's algebras, although the Loday (who sometimes wrote under a pseudonym "Guil- laume William Zinbiel", here Zinbiel - is a surname Leibniz, read in reverse order) in [4] pointed that it is a wrong name. An important role in drawing attention to the Leibniz algebra was played by paper [5] (which was followed by many other works). A large number of properties (including the fundamental one's) of Leibniz alge- bras scattered in articles and notes of various authors. While some authors consider the left Leibniz algebra, others - the right one's, making it difficult to read one of these articles. In this paper a listing of some basic definitions and statements of Leibniz algebras is given. However, most of them - with proofs, built on the notion of liezation - a transition to Lie algebras. It turns out that the use of the known properties of Lie algebras dramatically simplify many published proofs in the theory of Leibniz algebras. The concepts of left and right Leibniz algebras are parallel. For example, we may pass from the right to the left Leibniz algebra by considering a new multiplication x ◦ y = y · x. It will be convenient for me to consider left Leibniz algebras, since they have more visible relationship with differentiation of products (in which the differential operator is written to the right of a differentiable object). However, in many studies on Leibniz algebra the authors prefer to consider right Leibniz algebra. In this case, the main identity is usually written in the form x · (y · z) = (x · y) · z − (x · z) · y. This identity is of course equivalent to the identity, which expresses the property of differentiation for right multiplication by z, but to me it seems less natural than based on the concept of differential algebra. In addition, in the study of Leibniz algebras the multiplication is often written (by analogy with the Lie algebras) in bracket form [x, y]. This article will be used in the future such a symbol of multiplication in Leibniz algebras. For left Leibniz algebras the set of all left multiplication lx form a Lie algebra. We denote this algebra as adl(L) (similarly for right Leibniz algebras). Let us consider some simple examples of Leibniz algebras - Leibniz algebras of dimension 1 and 2. Let dimk(L) = 1 and a - a non-zero element in L. If [a, a] = 0, then L is an abelian Lie algebra. If [a, a] 6= 0, then [a, a] = αa for some nonzero α ∈ k. But then the Leibniz identities (both right and left) gives us a contradiction. Therefore, there is only one-dimensional Leibniz algebra - and it is a Lie algebra with the trivial (zero) multiplication. Now let dimk(L) = 2. Then it is easy to verify that there are only four (up to isomorphism, of course) Leibniz algebra: 1. Two Lie algebras: a2 (abelian) and solvable r2 (given in a suitable basis a, b by the relations [a, b] = −[b, a] = b). 2. Two (left) Leibniz algebra, which are not Lie algebras. In a suitable basis a, b they are given by such relations (we specify only the cases of non-zero products): (i) [b, b] = a 2 (ii) [b, a] = a, [b, b] = a Note that the algebra (i) is the left and right Leibniz algebra (obviously), and (ii) only the left one - you can check it by direct calculation of identities. But is easier to use a consequence of Leibniz algebras given below. In fact, for our algebra we have [[b, b], b] = [a, b] = 0 (as it should be for the left Leibniz algebra), but [b, [b, b]] = [b, a] = a, but for the right Leibniz algebra it should be 0. So this two-dimensional left Leibniz algebra is not a right Leibniz one. The identity of left Leibniz algebra follow some useful consequences of this iden- tity. Here are some of them: r[a,b] = rb ◦ ra + la ◦ rb (for the identity [c, [a, b]], taking c as an argument) r[ab] = la ◦ rb − rb ◦ la (from the identity for [a, [c, b]], taking c as an argument) These identities imply that rb ◦ ra = −lb ◦ ra, i.e. for any a, b, c ∈ L, we have [a, b], c]] = −[[b, a], c]. In particular, for any a, b, we have [[a, a], b] = 0. The right Leibniz algebras have the identity of the same type - just rearrange the relevant factors. In particular, for the right Leibniz algebra we have [a, [b, b]] = 0. Here is a direct proof of the same identities. Take the main identity and the second one obtained from the first by permutation of the arguments: [a, [b, c]] = [[a, b], c] + [b, [a, c]] [b, [a, c]] = [[b, a], c] + [a, [b, c]] or [a, [b, c]] = [b, [a, c]] − [[b, a], c] A comparison of these identities gives us [[a, b], c] = −[[b, a], c]. In the other words, although the operation [−, −] in the left Leibniz algebra is not skew one, but becomes so after right multiplication by any element. A similar can be said about the right Leibniz algebra - there will need to use multiplication by an element on the left. Let L be any left Leibniz algebra. Consider a subset consisting of elements of the form [a, a] for all possible a ∈ L. Let us consider the linear subspace generated by this subset (we denote it by Ker(L); reason for such notation and other possible its name age written below) - it is an ideal in L, left and right one's, abelian and lefty-center (the multiplication [Ker(L), L] is 0.) The fact that it is the right ideal follows from such identity: [a, [b, b]] = [a + [b, b], a + [b, b]]] − [a, a]. But the ideal Ker(L) is not always right-central. This ideal (using other notation) was firstly introduced in [2]. Here is an another proof of the fact that Ker(L) is a right ideal: at first we show that Ker(L) is spanned by elements of the form [a, b] + [b, a] (this follows from the identity [a, b] + [b, a] = [a + b, a + b] − [a, a] − [b, b]), and then we use the identity [a, [b, b]] = [c, b] + [b, c], where c = [a, b]. The ideal Ker(L) always is different from the L, as it abelian and abelian Leibniz algebra is a Lie algebra and therefore for it Ker(L) = {0}. Factor algebra L/Ker(L) is a non-zero algebra, where Ker(L) is the smallest ideal in L, the quotient algebra by which is a Lie algebra (both for left and for right Leibniz algebras). Therefore Ker(L) can be viewed as a "non-Lie core" and perhaps could be called the "liezator" for L. The ideal Ker(L) can be described in another way - as the linear span of all elements of the form [x, y] + [y, x]. Factor algebra L/Ker(L) is called as the liezation of L and denotes by L⋆. There is a natural action of the Lie algebra L⋆ on a vector space Ker(L) (multiplication in which is trivial). If the Leibniz algebra L is commutative (i.e. [x, y] = [y, x]), then the subset [x, y] + [y, x] coincides with the commutant [L, L] = {[x, y]x, y ∈ L}. But then liezation of the algebra L is just an abelian Lie algebra L/[L, L]. Commutative 3 algebras are nilpotent Leibniz algebras (about which see the next section), their class of nilpotency equals to 2. This kind of Leibniz algebra can be described in some detail. Now let us consider the left center Z l(L) = {x ∈ L[x, L] = 0}. Similarly we introduce the concept of the right center Z k(L) (both this centers can be considered for the left and right Leibniz algebras). For the left Leibniz algebra Z l(L) is an ideal, moreover - two-sided ideal (since [[x, y], L] = −[[y, x], L]). The right center Z k(L) is an subalgebra (because [x, [u, v]] = [[x, u], v] + [u, [x, v]] = 0 for all x ∈ L, u, v ∈ Z r(L), and in general, the left and right centers are different; they even may have different dimensions. By the above, we have Ker(L) ⊂ Z l(L). Therefore L/Z l(L) is the Lie algebra, which is isomorphic to the Lie algebra adl(L) mentioned above. For Leibniz algebras may be naturally introduced many of the concepts studied in the theory of Lie algebras. For example, consider series (derived series) of com- mutants Dn(L): D1(L) = [L, L], Dk+1(L) = [Dk(L), Dk(L)]. It is important to note that the members of this series are two-sided ideals (both for the right and for the left Leibniz algebras) - it is an easy consequence of the Leibniz identity. Leibniz algebra is called solvable if the derived series comes to {0} at some finite step. It is easy to check that the sum of solvable ideals in Leibniz algebra is also a solvable ideal. Therefore, in L there exists a largest solvable ideal R (containing all other soluble ideals). Naturally, it is called the radical of the Leibniz algebra L. Since Ker(L) is an ideal with trivial multiplication, then it is contained in the radical of R of Leibniz algebra L. Next, we consider the decreasing central series for Leibniz algebra L: C n(L): C1(L) = [L, L], C k+1(L) = [L, C k(L)]. Despite a certain lack of symmetry of the definition (multiplication by L only from the left) members of this series are two- sided ideals, since by the Leibniz identity (any - right or left), we have [C k(L), L] = [L, C k(L)]. Therefore, C k(L) consists of linear combinations of the products of k elements with arbitrarily spaced brackets. In particular, it implies such inclusion [C p(L), C q(L)] ⊂ C p+q(L). Leibniz algebra is called nilpotent if its central series reaches zero in a finite number of steps. As it is follows from definition the centers (left and right one's) are nonzero. Proposition 1. Any Leibniz algebra L has a maximal nilpotent ideal (containing all the nilpotent ideals of L) Proof. Let I be some nilpotent ideal in L. Then the sum I + Ker(L) is a nilpotent Ideal too - this follows immediately from the left centrality of Ker(L). Therefore, in discussing the question of the existence of maximal nilpotent ideal is sufficient to consider only nilpotent ideals containing Ker(L). Consider the epimorphism of liezation L : L → L/Ker(L) = L⋆. For the Lie algebra L/Ker(L) the existence of a maximal nilpotent ideal is known. Its inverse image under the mapping L → L also is nilpotent (by the centrality of Ker(L)) and it is maximal because of what is said above about the maximum nilpotent ideals in L. By nilradical in a Leibniz algebra L is called a maximal nilpotent ideal in L (which exists by Proposition 1). By its definition, the nilradical is a characteristic ideal, i.e. It remains invariant under all automorphisms of the Leibniz algebra L. 4 It is obviously contained in the radical of the Leibniz algebra and it equals to the nilradical of the solvable radical of L. The nilradical contains right center, as well as the ideal Ker(L). We also consider the notion of normalizer. The left normalizer N l L(U ) of subset U ⊂ L in Leibniz algebra L is the set of such x ∈ L, that [a, U ] ⊂ U . For the right normalizer N r L(U ) it requires that [U, a] ⊂ U . For the left Leibniz algebra a left normalizer of some subalgebra M is a subalgebra, but the right normalizer of M is not necessarily a subalgebra (but see Lemma below). Proposition 2. (Engel's theorem for Leibniz algebras) If all operators lx of left multiplication for the left Leibniz algebra L are nilpotent, then the algebra L is nilpotent In particular, for left multiplications there is a common eigenvector with zero eigenvalue In some basis matrices of all lx are upper-triangular nilpotent matrices. Proof. For the Lie algebra L/Ker(L) the Engel's theorem is well known. Therefore, the assertion of Proposition 2 holds for it. But Ker(L) is the central ideal - it action by left multiplication on L is trivial. Therefore Leibniz algebra L can be considered as a central extension of a nilpotent Lie algebra - so it is also nilpotent. Reducing of matrix of left multiplications to upper-triangular nilpotent form is obvious now (using induction). There is a stronger version [6] of this proposition: if for left Leibniz algebra all left multiplication are nilpotent, then all right multiplications also are nilpotent and in a suitable bases linear operators of all left and all right multiplications have upper-triangular nilpotent matrices. Moreover, both left and right multiplications have a common eigenvector with eigenvalue equals to 0. The proof of this version is based on the identity (rx)n = (−1)nrx(lx)n−1, which is proved by induction. For n = 2, it follows from the identity mentioned above: [[x, a], a] = −[[a, x], a]. Then take (rx)3(a) = [[[a, x], x], x] and transform it similarly and then we may use an induction. Corollary 1. The normalizers (both - left and right) of some subalgebra M in a nilpotent Leibniz algebra L are not equal to the subalgebra M (they both strictly contain M ) Proof. At first we consider the left normalizer. The reasoning here is exactly the same as for Lie algebras - we consider the action induced by left multiplication on the factor space L/M . The set of linear operators of left multiplications forms a Lie algebra adl(L), which is nilpotent in our case. So by Engel's theorem for Lie algebras this action has a common eigenvector with 0 as an eigenvalue. The vector in L which corresponding to this eigenvector is contained in NL(M ) and is out of M . For the right normalizer we use the stronger version of Engel theorem mentioned above. Corollary 2. A subspace V ⊂ L generates a Leibniz algebra if and only if V + [L, L] = L Proof. Let V + [L, L] = L. As M we denote the subalgebra of L, generated by the subspace V . We want to prove that M = L. The proof uses inductions on codim(M ) and dim(L). For codim(M ) = 0 the statement is obvious, because in this case M = L. For dim(L) = 0 this statement is evident too. 5 Let codim(M ) > 0. Consider the normalizer M ′ = NL(M ) - it is strictly greater than M . Therefore codim(M ′) < codim(M ) and so by the induction we have M ′ = L. This means that M is the left ideal in L, and it is supplemental to the [L, L]. Now we are going to prove that M may be supposed to be a right ideal too. We use one easy lemma. L(I) is a subalgebra in L. Lemma. Let I be a left ideal in a left Leibniz algebra L. Then the right normalizer N r Proof of Lemma. Let us consider two elements n, n′ ∈ N r L(M ). We have [M, [n, n′]] = [[M, n], n′] + [n, [M, n′]]. But [M, n] and [M, n′] both contain in M (due to the definition of the normalizer) and [n, M ] ⊂ M because M is the left ideal in L. So [M, [n, n′]] ⊂ M , [n, n′] ∈ N r L(M ), i.e. M is the subalgebra in L. In our case M is the left ideal in L. Due to Lemma N r L(M ) is a subalgebra in L (and it strictly contains M - see Corollary 1). We continue to use an induction on codimL(M ). We have codimL(N r L(M )) < codimL(M ). Therefore by induction we have N r L(M ) = L, i.e. M is the right ideal in L. Therefore we may suppose now that M is a two-sided ideal in L. We use now a induction on dim(L). Let L⋆ = L/Ker(L) and consider the natural epimorphism L : L → L⋆. It is clear that L([L, L]) = [L⋆, L⋆]. Let V ⋆ = L(V ). It is clear that V ⋆ + [L⋆, L⋆] = L⋆. Since the dimension of the Lie algebra L⋆ less than the dimension of the Lie algebra L, then we can hold the induction step. Therefore, we conclude that the subspace V ⋆ generates a Lie algebra L⋆. But then we have M + Ker(L) = L. For the commutant of L we have [L, L] = [M + Ker(L), M + Ker(L)] = [M, M ] + [M, Ker(L)] + [Ker(L), M ] + [Ker(L), Ker(L)]. However, the subspaces [Ker(L), Ker(L)] and [Ker(L), M ] are zero by left centrality of Z. Also we have [M, Ker(L)] ⊂ M , since M also is the right ideal in L. So [L, L] = [M, M ]+M ⊂ M , and therefore M contains [L, L]. Since M + [L, L] = L, we see that M = L. Conversely, if the subspace of V generates a Leibniz algebra L, then its image under the natural epimorphism L → L/[L, L] generates abelian Leibniz algebra L/[L, L]. Since the multiplication in L/[L, L] is trivial, then the subspace of this algebra can generate it only if this subspace coincides with the whole algebra. But this is equivalent to V + [L, L] = L It is interesting to note that not all properties of nilpotent Lie algebras, even a simple and well-known one's, are hold for the case of Leibniz algebras. For example, there is a simple statement for nilpotent Lie algebras of dimension 2 or more: "codimension of commutant is ≥ 2". For Leibniz algebras it is not true (though not only for nilpotent, but for all solvable Leibniz algebras we have codimL[L, L] > 0). For example, two-dimensional Leibniz algebra < a, b >, for which [a, a] = b (see above) is nilpotent, but its commutant has codimension 1. This is due to the fact that its liezation is one dimensional. But for one-dimensional Lie algebras above mentioned statement is incorrect. We obtain the useful corollary. Corollary 3. If Leibniz algebra L is nilpotent and codimL([L, L]) = 1, the algebra L generated by one element. So for codimL([L, L]) = 1 a nilpotent algebra is a kind of "cyclic". The study of such nilpotent algebras is the specifics of the theory of Leibniz algebras; Lie algebras has no analogue. Such cyclic L may be explicitly described. 6 Corollary 4. The minimal number of generators of Leibniz algebra L equals to dimL/[L, L] We recall now that a complete flag of subspaces of a finite dimensional vector space V is a family of nested subspaces V0 = {0} ⊂ V1 ⊂ · · · ⊂ Vn = V such that dimVk = k. Proposition 3. (weak version of Lie Theorem for solvable Leibniz algebras) A left Leibniz algebra L over C has a complete flag of subspaces which is invariant under left multiplication. In other words, all linear operators lx of left multiplications can be simultaneously reduced to triangular form. Proof. For Lie algebras this theorem in well known (in one of the variants of its formulation). Now let L be an arbitrary solvable Leibniz algebra. Since liezator Ker(L) is central, then the solvability of the Leibniz algebra L is equivalent to the solvability of the Lie algebra L/Ker(L). By the classical Lie theorem L/Ker(L) has a complete flag, which is invariant under the multiplication (both left and right). Let x ∈ L be some element. Its left action on Ker(L) is trivial and therefore in Ker(L) there is a complete flag, which is invariant under left multiplication (for the right multiplication on the left Leibniz algebra such reasoning does not pass). Adding to it the full invariant flag in the factor space until the full flag in L, we obtain a complete flag in L, which is invariant under left multiplication. Proposition 4. Let R - radical in the Leibniz algebra L, and N - nilradical of L. Then [L, R] ⊂ N Proof. For Lie algebras L that statement is true. Now let L be an arbitrary Leibniz algebra. Then Ker(L) ⊂ N ⊂ R. Let L⋆ = L/Ker(L) (the liezation of algebra L), R⋆ = R/Ker(L), N ⋆ = N/Ker(L). From the definitions of the radical and nilradical it follows that R⋆ andN ⋆ are radical and nilradical of Lie algebra L⋆ respectively (we remember that Ker(L) ⊂ Z l(L)). Since L⋆ is the Lie algebra, then [L⋆, R⋆] ⊂ N ⋆. But then, by the fact that Ker(L) ⊂ N ⊂ R, we get [L, R] ⊂ N I note that the statements contained in Propositions 3 and 4 are not met me in the papers on the theory of Leibniz algebras (although they are well known for Lie algebras, especially Lie Theorem). Corollary 5. [R, R] ⊂ N . In particular, [R, R] nilpotent. Corollary 6. Leibniz algebra L is solvable if and only if [L, L] is nilpotent. Proof. In one direction the assertion of Corollary 6 is proved in Corollary 5. Let us consider the converse. Let [L, L] be nilpotent. Consider the radical R of algebra L. The Lie algebra L/R is the semi-simple Lie algebra (as it contains no solvable ideals). But semisimple Lie algebra coincides with its commutator. Therefore, in the case where [L, L] is nilpotent, the this semisimple algebra should be trivial. But this means that L is solvable. Proposition 5. Multiplications (right and left one's) in the Leibniz algebra are degenerate linear operators (their determinants equal to 0). Proof. For Lie algebras this degeneration is obvious, since [x, x] = 0 for any x ∈ L, i.e. the vector x is the root vector for the zero eigenvalue. 7 In any Leibniz algebra Ker(L) is the two-sided ideal different from L (see above) and L/Ker(L) is a Lie algebra of positive dimension. Let x be some non-zero element in L. For lx and rx the subspace Ker(L) invariant. On the factor space which is a Lie algebra, the linear operators induced by these multiplications, are degenerated (their determinants equal to 0). But then these multiplications are degenerate in the whole space L. There is an interesting question about the existence of faithful representations of Leibniz algebras. Linear representation (sometimes referred as module) of Leibniz algebra is a vector space V , for which we have two actions (left and right) of the Leibniz algebra L [−, −] : L × M → V and [−, −] : M × L → V , such that the identity [x, [y, z]] = [[x, y], z] + [y, [x, z]] is true if one (any) of the variables is in M , and two others - in L. In some other formulation (and without using bracketed notation for the action of the algebra L on V ) we have such conditions conditions: a(bm) = [ab]m + b(am) a(mb) = (am)b + m[ab] m[ab] = (ma)b + a(mb) Note that the concept of representations of Lie algebras and Leibniz algebras are different. Therefore, such an important theorem in the theory of Lie algebras, as the Ado theorem on the existence of faithful representation in the case of Leibniz algebras was proved much easier and gives a stronger result. It is because the ker- nel of the Leibniz algebra representation is the intersection of kernels (in general, different one's) of right and left actions, in contrast to representations of Lie alge- bras, where these kernels are the same. Therefore, an faithful representation of the Leibniz can be obtained easier than faithful representation of the Lie algebra. Proposition 6 (see [7]). Any Leibniz algebra has a faithful representation of dimension ≤ dim(L) + 1. The structure of Leibniz algebras may be described by such result. Proposition 7 (Levi theorem for Leibniz algebras, see [8]). For a Leibniz algebra L there exists a subalgebra S (which is a semisimple Lie algebra), which gives the decomposition L = S + R, where R - radical. Proof. Consider the Lie algebra L⋆ = L/Ker(L). By the classical Levi theorem there is a semisimple algebra S⋆ ⊂ L⋆, which gives the semidirect sum decomposi- tion L⋆ = S⋆ + R⋆. Here R⋆ is the radical of L⋆. Let R be the radical of Leibniz algebra L and L : L → L⋆ be the natural epimorphism. Then it is easy to un- derstand, that (R) = R⋆. Let F = L−1(S⋆). Then F is the Leibniz subalgebra in L which contains Ker(L), and the quotient algebra F/Ker(L) is isomorphic to a semisimple Lie algebra S⋆. For the Lie algebra S⋆ the abelian Leibniz algebra Ker(L) is S⋆− module. For semisimple Lie algebra S⋆, it follows that the subalgebra F , considered as an ex- tension of S⋆ using Ker(L), is split (by Whitehead's Lemma for semisimple Lie algebras). But this means that in F there exists a subalgebra (semisimple Lie alge- bra) which is complementary to the abelian ideal Ker(L). Clearly, the S + R = L and we get the required decomposition. 8 By examples (see [8]) the non-uniqueness of the subalgebra S it can be shown In the case of Lie (minimum dimension of Leibniz algebra there equals to 6). algebras the semi-simple Levy factor is unique up to conjugation. Corollary 7. Let L be a Leibniz algebra such that its left center lies in the right center (i.e., that [L, Z l(L)] = {0}). Then if the Lie algebra L/Z l(L) is a semisimple one, that L is the reductive Lie algebra (so its radical is central) Proof. Let us consider the Levi decomposition L = S +R. Since, by hypothesis, the algebra L/Z r(L) is semisimple, then R = Z l(L). The left action of the subalgebra S on Z l(L) is always trivial. But since [L, Z l(L)] = {0}, the right action of S on Z l(L) is also trivial. Therefore, S commutes with the radical and therefore Levi decomposition of L is a direct one. The radical of R = Z l(L) is abelian, so L is the reductive Lie algebra. Statement of Corollary 7 is proved in [5] for the special case when S is a classical simple Lie algebra. The author is grateful to B.Omirov for helpful discussions about the theory of Leibniz algebras. Bibliography 1. Bloh A.M., On a generalization of the concept of Lie algebra, Dokl. Akad. Nauk SSSR 165 (1965), 471 -- 473.. 2. Bloh A.M., Cartan-Eilenberg homology theory for a generalized class of Lie algebras, Dokl. Akad. Nauk SSSR 175 (1967)), 824 -- 826. 3. Loday J.-L., Une version non commutative des algebres de Lie: les algebres de Leibniz, Enseign. Math. vol 39 (1993), 269 -- 293. 4. Zinbiel, Guillaume W., Encyclopedia of types of algebras 2010, Bai, Chengming; Guo, Li; Lo- day, Jean-Louis, Operads and universal algebra, Nankai Series in Pure, Applied Mathematics and Theoretical Physics, 9, 2012, pp. 217 -- 298,. 5. Ayupov Sh., Omirov B., On Leibniz algebras, Algebra and operator theory. Proc. of the colloquium in Tashkent, 1997, Dordrecht; Boston; London: Kluwer Acad. Publ., 1998, pp. 1 -- 13. 6. Patsourakos A., On nilpotent properties of Leibniz algebras, Commun. Algebra 35 (2007), 3828 -- 3834. 7. Barnes D., Faithful representations of Leibniz algebras, arXiv:1111.2627 (2011), 1 -- 2. 8. Barnes D., On Levi's Theorem for Leibniz algebras, Bulletin of the Australian Mathematical Society 86 (2012), 184 -- 185. 9
1607.03178	2	1607	2017-01-01T17:06:34	Quasi-duo differential polynomial rings	[ "math.RA" ]	In this article we give a characterization of left (right) quasi-duo differential polynomial rings. In particular, we show that a differential polynomial ring is left quasi-duo if and only if it is right quasi-duo. This yields a partial answer to a question posed by Lam and Dugas in 2005. We provide non-trivial examples of such rings and give a complete description of the maximal ideals of an arbitrary quasi-duo differential polynomial ring. Moreover, we show that there is no left (right) quasi-duo differential polynomial ring in several indeterminates.	math.RA	math	QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS MAI HOANG BIEN1,2 AND JOHAN OINERT2 1Department of Basic Sciences, University of Architecture, 196 Pasteur Str., Dist. 1, Ho Chi Minh City, Vietnam 2Blekinge Institute of Technology, Department of Mathematics and Natural Sciences, SE-37179 Karlskrona, Sweden Abstract. In this article we give a characterization of left (right) quasi-duo differential polynomial rings. In particular, we show that a differential polynomial ring is left quasi- duo if and only if it is right quasi-duo. This yields a partial answer to a question posed by Lam and Dugas in 2005. We provide non-trivial examples of such rings and give a complete description of the maximal ideals of an arbitrary quasi-duo differential polynomial ring. Moreover, we show that there is no left (right) quasi-duo differential polynomial ring in several indeterminates. 1. Introduction Throughout this article, all rings are assumed to be unital and associative. Following [3], a ring S is said to be left (right) duo if every left (right) ideal of S is a two-sided ideal. More generally, S is said to be left (right) quasi-duo if every maximal left (right) ideal of S is a two-sided ideal (see e.g. [13]) or, equivalently, if every left (right) primitive homomorphic image of S is a division ring (see e.g. [11, Proposition 4]). A ring which is both left and right quasi-duo is called quasi-duo. Quasi-duo rings appear in various places in ring theory, e.g. in the investigation of the Kothe conjecture (see e.g. [10, Proposition 2.5]). There are many open problems concerning left (right) quasi-duo rings, one of which is due to Lam and Dugas who ask whether there exists a right quasi-duo ring which is not left quasi-duo (see [6, Question 7.7]). Our main result (Theorem 1.1) shows that such an example can not be found in the class of differential polynomial rings. E-mail address: [email protected], [email protected]. Date: 2017-01-01. 2010 Mathematics Subject Classification. 16S32, 16W70, 16D25. Key words and phrases. Ore extension, differential polynomial ring, quasi-duo ring, maximal ideal. This research was supported by the Crafoord Foundation, research grant no. 20150871. 1 2 QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS Recall that an Ore extension R[x; σ, δ] is constructed from a ring R, a ring endomorphism σ : R → R (respecting 1R) and a σ-derivation δ : R → R, i.e. an additive map satisfying δ(rs) = σ(r)δ(s) + δ(r)s, ∀r, s ∈ R. As a left R-module R[x; σ, δ] is equal to the usual polynomial ring R[x]. The multiplication on R[x; σ, δ] is defined by the rule xr = σ(r)x + δ(r) for r ∈ R. This turns the Ore extension R[x; σ, δ] into a unital and associative ring (see e.g. [9]). If δ = 0, then R[x; σ, 0] is said to be a skew polynomial ring. If, on the other hand, σ = idR, then δ is called a derivation and R[x; idR, δ] is said to be a differential polynomial ring and will simply be denoted by R[x; δ]. In [7], Leroy, Matczuk and Puczylowski obtained a complete characterization of left (right) quasi-duo skew polynomial rings. In [8], the same authors continued their investi- gation and gave a complete characterization of left (right) quasi-duo Z-graded rings. In this article we direct our attention to another type of Ore extensions, namely the differential polynomial rings. Our main result is the following. Theorem 1.1. Let S = R[x; δ] be a differential polynomial ring, and put J0 = J(S) ∩ R. The following five assertions are equivalent: (i) S is left quasi-duo; (ii) S is right quasi-duo; (iii) Every left ideal of S containing the Jacobson radical J(S) is two-sided, i.e. S/J(S) is left duo; (iv) Every right ideal of S containing the Jacobson radical J(S) is two-sided, i.e. S/J(S) is right duo; (v) The quotient ring R/J0 is commutative and δ(R) ⊆ J0. This result provides a complete characterization of left (right) quasi-duo differential polynomial rings. In particular, it shows that a differential polynomial ring is left quasi- duo if and only if it is right quasi-duo, thereby yielding a partial answer to [6, Question 7.7]. Notice that for skew polynomial rings, which were studied in [7], the same question is still wide open. This article is organized as follows. In Section 2 we prove Theorem 1.1. In Section 3 we show that if R[x; δ] is quasi-duo, then if R belongs to certain classes of rings, we can conclude that R[x; δ] is commutative (see Proposition 3.2). This means, in particular, that δ = 0 and hence R[x; δ] is a polynomial ring. We also provide examples of quasi-duo differential polynomial rings which are not polynomial rings (see Example 3.3). In Section 4 we give a complete description of the maximal ideals of quasi-duo differential polynomial rings (see Theorem 4.3). In Section 5 we consider differential polynomial rings in several indeterminates, R[X; D], defined by a (countable) set of variables X and a family D of derivations on R. We show that R[X; D] can never be quasi-duo if X consist of more than one variable (see Theorem 5.3). QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS 3 2. Proof of the main result In this section we give a proof of our main result, Theorem 1.1. We begin by showing that a left (right) quasi-duo differential polynomial ring over a simple ring is necessarily a polynomial ring. Lemma 2.1. Let S = R[x; δ] be a left (right) quasi-duo differential polynomial ring. If R is a simple ring, then R is a field and δ = 0. Proof. Let L be a maximal left (right) ideal of S containing x. By assumption, L is a two-sided ideal of S and hence, for any r ∈ R, we get δ(r) = xr − rx ∈ L ∩ R. Notice that L ∩ R is a two-sided ideal of R. Using that R is a simple ring and that L 6= S, we conclude that L ∩ R = {0} and hence δ(r) = 0, for all r ∈ R. Take a ∈ R and let M be a maximal left (right) ideal of S containing x−a. By assumption M is a two-sided ideal of S and hence, for any b ∈ R, we get ab−ba = b(x−a)−(x−a)b ∈ M. Using the same argument as before we get M ∩ R = {0} and hence ab − ba = 0. This shows that R is a commutative and simple ring, i.e. a field. (cid:3) The following proposition gives us a necessary condition on the ring R and the derivation δ in order for the differential polynomial ring R[x; δ] to be left (right) quasi-duo. Proposition 2.2. Let S = R[x; δ] be a differential polynomial ring, and put J0 = J(S)∩R. If S is left (right) quasi-duo, then R/J0 is commutative and δ(R) ⊆ J0. Proof. Suppose that S is left quasi-duo. (The right quasi-duo case is treated analogously.) There are now two cases: (1) there exists a maximal left ideal M of S such that M ∩ R = {0}; or (2) M ∩ R 6= {0} for any non-zero maximal left ideal M of S. Case 1. Suppose that there exists a maximal left ideal M of S such that M ∩ R = {0}. In this case, J0 = J(S) ∩ R = {0}. We claim that R is a division ring. Indeed, since S is left quasi-duo, M is two-sided. Hence, the factor ring K = S/M is a division ring. Let a ∈ R\{0}. Since a /∈ M, a is invertible in K. Let a0, a1, · · · , an ∈ R such that a(a0 + a1x + · · · + anxn) = 1 in K. Then aa0 = 1 in K. Hence, aa0 − 1 ∈ M. By the assumption R ∩ M = {0}, we get that aa0 = 1. This means that every non-zero element of R has a right inverse in R, which implies that R is a division ring. By Lemma 2.1 we now conclude that R is a field and that δ = 0. Case 2. Suppose that for any non-zero maximal left ideal M of S, M ∩ R 6= {0} holds. Let M be a non-zero maximal left ideal of S. Put M0 = M ∩ R. For any a ∈ M0, one has δ(a) = xa − ax ∈ M. Hence, δ(a) ∈ M0 for any a ∈ M0. This means that M0 is δ-invariant. Therefore, δ induces a derivation δ on R/M0, namely, δ(a) = δ(a), for a ∈ R. Consider the map ϕ : R[x; δ] → (R/M0)[x; δ], a0 + a1x + . . . + anxn 7→ a0 + a1x + . . . + anxn. It is easy to check that ϕ is a surjective ring morphism. Thus, in view of [6, Page 245], (R/M0)[x; δ] is left quasi-duo. Moreover, the ideal M = ϕ(M) is a maximal left ideal of (R/M0)[x; δ] and (R/M0) ∩ M = {0}. Now, by applying Case 1, we get that R/M0 is commutative and that δ = 0. 4 QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS It remains to show that R/J0 is commutative and that δ(R) ⊆ J0 holds. To see this, notice that we have already proved that for any maximal left ideal M, δ(a) = 0 and ab = ba for any a, b ∈ R/M0 where M0 = M ∩ R. As a corollary, δ(a) ∈ M and ab − ba ∈ M, for all a, b ∈ R and every maximal left ideal M of S. Therefore, δ(a), ab − ba ∈ J(S) ∩ R = J0 for any a, b ∈ R. Thus, δ(R) ⊆ J0 and R/J0 is commutative. This concludes the proof. (cid:3) We shall now prove Theorem 1.1 and thereby get a complete characterization of quasi- duoness of R[x; δ]. Proof of Theorem 1.1 We will only prove the left sided case, (ii)⇔(iv)⇔(v), is treated analogously. We will now show that (i)⇒(v)⇒(iii)⇒(i). (i)⇒(v): This implication follows from Proposition 2.2. (v)⇒(iii): Consider the morphism ϕ as defined in Case 2 of Proposition 2.2: (i)⇔(iii)⇔(v). The right sided case, i.e. i.e. ϕ : R[x; δ] → (R/J0)[x; δ], a0 + a1x + . . . + anxn 7→ a0 + a1x + . . . + anxn. In our case, R/J0 is commutative and δ = 0. Hence, (R/J0)[x, δ] is commutative. Notice that ϕ is surjective and that ker(ϕ) = J0[x, δ] ⊆ J(S). Hence, S/(J0[x, δ]) ∼= (R/J0)[x, δ] which is commutative. Therefore, every left ideal of S containing J(S) is two-sided. (iii)⇒(i): This is trivial. (cid:3) Remark 2.3. Theorem 1.1 shows that a differential polynomial ring is left quasi-duo if and only if it is right quasi-duo. Henceforth, we need not make a distinction between the left and the right properties and shall simply use the notion quasi-duo. 3. Trivial and non-trivial quasi-duo differential polynomial rings By Lemma 2.1 we have observered that if the differential polynomial ring R[x; δ] is quasi-duo and R is simple, then R[x; δ] is necessarily a commutative polynomial ring. In this section we show that the same conclusion holds for large classes of rings R which are not necessarily simple (see Proposition 3.2). We will also show that there exist quasi-duo differential polynomial rings which are not classical polynomial rings (see Example 3.3). Lemma 3.1. Let S = R[x; δ] be a differential polynomial ring, and denote the nilradical of R by Nil(R). Suppose that R satisfies at least one of the following conditions: (i) Nil(R) = {0} and R is a PI-ring, i.e. R satisfies a polynomial identity; (ii) Nil(R) = {0} and R satisfies the ascending chain condition on right annihilators. Then, S is semiprimitive, i.e. J(S) = {0}. Proof. This is just a corollary of [12]. (cid:3) By the preceding lemma, the class of rings R over which differential polynomial rings R[x; δ] are semiprimitive includes e.g. semiprime commutative rings, domains, and noe- therian rings with Nil(R) = {0}. QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS 5 Proposition 3.2. Let R be a ring satisfying Nil(R) = {0}. If R is a PI-ring or satisfies the ascending chain condition on right annihilators, then the following two assertions are equivalent: (i) R[x; δ] is quasi-duo; (ii) R[x; δ] is commutative. Proof. The desired conclusion follows from Lemma 3.1 and Theorem 1.1. (cid:3) The following example demonstrates a quasi-duo differential polynomial ring which is non-trivial, i.e. not a polynomial ring. 0 c(cid:21)(cid:12)(cid:12)(cid:12)(cid:12) Example 3.3. Let R = (cid:26)(cid:20)a b a, b, c ∈ R(cid:27). It is not difficult to see that the Jacob- 0 0(cid:21)(cid:12)(cid:12)(cid:12)(cid:12) son radical of R is J(R) = (cid:26)(cid:20)0 b a, c ∈ R(cid:27) is 0 c(cid:21)(cid:12)(cid:12)(cid:12)(cid:12) b ∈ R(cid:27) . Hence, R/J(R) = (cid:26)(cid:20)a 0 commutative. Take A ∈ J(R)\{0} and let δ be the inner derivation on R defined by δ(B) = AB − BA, for B ∈ R. Clearly, δ(R) ⊆ J(R). Now consider the corresponding dif- ferential polynomial ring R[x; δ]. Using that R is a PI-ring over R, a field of characteristic zero, [1, Theorem 1.2] yields that J(R[x; δ]) = Nil(R)[x; δ]. Since every element of R is a root of a non-zero polynomial over R, by [5, Corollary 4.19], J(R) = Nil(R). This means that J0 = J(R[x; δ]) ∩ R = J(R). Therefore, R[x; δ] satisfies Theorem 1.1(v) and hence R[x; δ] is left and right quasi-duo. Remark 3.4. One may replace R by an arbitrary ring of upper triangular n by n matrices and mimick the above construction. This gives us a whole family of non-trivial quasi-duo differential polynomial ring. 4. Maximal ideals and the Jacobson radical In this section, we shall describe all maximal ideals of S = R[x; δ] in the case when S is quasi-duo. Notice that in [7], Leroy et al. gave a complete characterization of left (right) quasi-duo skew polynomial rings R[x; σ, 0] where σ is an automorphism of R. In order to do so, they first described all maximal ideals of R[x; σ, 0], and then used their description to characterize quasi-duoness. In this article we work in the opposite direction. In fact, we use our main result (Theorem 1.1) to find all maximal ideals of S = R[x; δ] as well as the Jacobson radical of R[x; δ]. Remark 4.1. Suppose that R[x; δ] is quasi-duo. Using the same argument as in the proof of Theorem 1.1, for any maximal ideal M of R[x; δ], if M0 = M ∩ R, then R/M0 is a field and δ(R) ⊆ M0. Let I be a maximal ideal of R such that R/I is a field and δ(R) ⊆ I. Then the map ΦI : R[x; δ] → (R/I)[x], a0 + a1x + · · · + anxn 7→ a0 + a1x + · · · + anxn is a (well-defined) surjective ring morphism. Moreover, ker ΦI = I[x; δ]. Hence, R[x; δ]/I[x; δ] ∼= (R/I)[x]. 6 QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS In particular, R[x; δ]/I[x; δ] is semiprimitive by Lemma 3.1. These facts will be used several times in this section. Proposition 4.2. Let R[x; δ] be a quasi-duo differential polynomial ring, and let I and ΦI be as in Remark 4.1. Then for any maximal ideal N of S containing I, we have Φ−1 I (ΦI(N)) = N. Proof. Clearly, N ⊆ Φ−1 Φ−1 I (ΦI(N)) or Φ−1 anxn ∈ N such that I (ΦI (N)). Because of the maximality of N, we have either N = I (ΦI(N)) = S, then there exists a0 + a1x + . . . + I (ΦI (N)) = S. If Φ−1 1 = a0 + a1x + . . . + anxn. Hence, 1 − a0, a1, . . . , an ∈ I. For any i ≥ 1 we have ai ∈ I and hence aixi ∈ N. This implies that a0 ∈ N. From the fact that 1 − a0 ∈ N, we conclude that 1 ∈ N. This is a contradiction. Hence, N = Φ−1 (cid:3) I (ΦI(N)). Given a differential polynomial ring R[x; δ], denote by M(R) the set of maximal ideals I of R such that R/I is a field and δ(R) ⊆ I. Recall that a monic polynomial in R[x; δ] is an element whose highest degree coefficient is equal to 1. For any I ∈ M(R), the set of all irreducible monic polynomials in (R/I)[x] is denoted by P(R/I). Theorem 4.3. Let S = R[x; δ] be a quasi-duo differential polynomial ring. Consider the following two maps: (i) Associate with any pair A = (I, xn + an−1xn−1 + · · · + a0) ∈ (M(R), P(R/I)) the maximal ideal M(A) = I[x, δ] + hxn + an−1xn−1 + · · · + a0iS of S; (ii) Associate with any maximal ideal M of S the pair A(M) = (M0, p(x)) ∈ (M(R), P(R/M0)) where M0 = M ∩ R and p(x) ∈ (R/M0)[x] is such that hp(x)i(R/M0)[x] = ΦM0(M). Then, these maps yield two mutually inverse bijections between the set of all maximal ideals of S and the set {(I, p(x)) I ∈ M(R) and p(x) ∈ P(R/I)}. Proof. We must first show that the maps from (i) and (ii) are well-defined, that is M(A) is a maximal ideal of S for any pair A = (I, p(x)), and A(M) is an element of the set {(I, p(x)) I ∈ M(R) and p(x) ∈ P(R/I)} for any maximal ideal M of S. Let A be a pair (I, p(x)) where K = R/I is a field, p(x) = xn + an−1xn−1 + · · · + a0 is an irreducible polynomial in K[x]. We have L = hp(x)iK[x], i.e. the ideal of K[x] generated by p(x). Since p(x) is irreducible, L is maximal in K[x]. Notice that ΦI (xn + an−1xn−1 + · · · + a0) = p(x). Hence, if ΦI(f ) ∈ L, then f ∈ hxn + an−1xn−1 + · · · + a0iS + I[x; δ] (using that R[x; δ]/I[x; δ] ∼= K[x] via ΦI in Remark 4.1). This implies that Φ−1 I (L) = M(A). Again, since R[x; δ]/I[x; δ] ∼= K[x] via ΦI, the ideal M(A) is a maximal ideal of R[x; δ]/I[x; δ]. Notice that I[x; δ] ⊆ M(A), and hence M(A) is a maximal ideal of S. Thus, the map from (i) is well-defined. Now assume that M is a maximal ideal of S. Put M0 = M ∩ R. By Remark 4.1, K = R/M0 is a field and δ(R) ⊆ M0. Hence, M0 ∈ M(R). Since M is maximal in S, the QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS 7 ideal ΦM0(M) is also maximal in K[x], by Remark 4.1. Therefore, there exists a unique irreducible monic polynomial p(x) ∈ K[x] such that ΦM0(M) = hp(x)i(R/M0)[x]. Hence, the map from (ii) is well-defined. Now we will show that M(A(M)) = M holds for any maximal ideal M of S, and that A(M(A)) = A holds for any pair A ∈ {(I, p(x)) I ∈ M(R) and p(x) ∈ P(R/I)}. Let M be a maximal ideal of S. Assume that A(M) = (M0, p(x)) where M0 = M ∩ R and p(x) = xn + an−1xn−1 + . . . + a0 ∈ P(R/M0). Notice that M0 ⊆ M, so that M0[x; δ] ⊆ M. Therefore, xn + an−1xn−1 + . . . + a0 ∈ M, which yields that M0[x; δ] + hxn + an−1xn−1 + . . . + a0iS ⊆ M. Equivalently, M(A(M)) ⊆ M. By the maximality of M(A(M)) in S we get M(A(M)) = M. Let A = (I, p(x)) where I is an ideal in M(R) and p(x) = xn + an−1xn−1 + . . . + a0 ∈ P(R/I). Then, M(A) = I[x; δ] + hxn + an−1xn−1 + . . . + a0iS. It is clear that I = M(A) ∩ R and that ΦI(M(A)) = hp(x)iK[x]. Thus, A = A(M(A)). (cid:3) We will now use the preceding theorem to describe the Jacobson radical of R[x; δ] and obtain a result which resembles [1, Theorem 1.2]. Corollary 4.4. Let S = R[x; δ] be a quasi-duo differential polynomial ring. Put K = TI ∈M(R) Then the Jacobson radical of S is J(S) = K[x; δ] = (J(S) ∩ R)[x; δ]. I. Proof. By Theorem 4.3, J(S) is the intersection of all ideals of the form I[x; δ] + hxn + an−1xn−1 + . . . + a0iS where I ranges over M(R) and the polynomial xn +an−1xn−1 +. . .+a0 ranges over P(R/I). That is, J(S) = \I ∈M(R) \ xn+an−1xn−1+...+a0∈P(R/I) (I[x, δ] + hxn + . . . + a0iS). For any I ∈ M(R), we define LI = \ xn+an−1xn−1+...+a0∈P(R/I) (I[x; δ] + hxn + an−1xn−1 + . . . + a0iS). According to the maps in Theorem 4.3, when xn + an−1xn−1 + . . . + a0 ranges over P(R/I), then I[x; δ] + hxn + an−1xn−1 + . . . + a0iS ranges over the set of all maximal ideals of R[x; δ]/I[x; δ]. Hence, LI is the Jacobson radical of R[x; δ]/I[x; δ]. Thus, LI = I[x; δ] using that R[x; δ]/I[x; δ] is semiprimitive. Therefore, J(S) = K[x; δ] = (J(S) ∩ R)[x; δ]. (cid:3) 8 QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS 5. Differential polynomial rings in several indeterminates In this section we shall show that differential polynomial rings in several indeterminates can never be quasi-duo (see Theorem 5.3). Let us begin by recalling the definition of a differential polynomial ring in a set of indeterminates. Let I be a non-empty (possibly infinite) countable set, let D = {δi i ∈ I} be a family of derivations on R (by "a family" we mean that all δi's need not be distinct), and let X = {xi i ∈ I} be a set of distinct non-commuting indeterminates. Given R, D and X, we can define the ring R[X; D] which is the set of all polynomials in the indeterminates xi ∈ X with coefficients from R. The addition in R[X; D] is the natural one and the multiplication is generated by the commutation rule xia = axi + δi(a), for i ∈ I. The ring R[X, D] is called a differential polynomial ring in several indeterminates. Readers are referred to [2, 12] for more details on this class of rings. In particular, every element f ∈ R[X; D] can be written in the form f = a1t1 + a2t2 + · · · + antn, where a1, a2, · · · , an ∈ R\{0} and t1, t2, . . . , tn are distinct monomials in X, i.e. finite words in the alphabet X. In this case, the support of f is defined as supp(f ) = {t1, t2, . . . , tn}. If δi = 0, for all i ∈ I, then R[X] is called a free polynomial ring. A subring B of a ring A is called a corner subring of A if B is unital, possibly with 1A 6= 1B, and if there exists an additive subgroup C of A such that A = B ⊕ C and BC, CB ⊆ C. The subgroup C is called a complement of B. We say that B is a left corner of A if the complement C satisfies BC ⊆ C only. The notion of a right corner is defined analogously. A classical example of a corner subring is given by B = eAe, where e ∈ A is an idempotent. The following lemma shows that quasi-duoness of differential polynomial rings in several indeterminates can be inherited by certain subrings. Lemma 5.1. Let R be a ring, let I be a non-empty countable set, let D = {δi i ∈ I} be a family of derivations on R, and let X = {xi i ∈ I} be a set of non-commuting indeterminates. For any subset J ⊆ I, put XJ = {xi i ∈ J} and DJ = {δi i ∈ J}. The ring SJ = R[XJ ; DJ] is a right (left) corner of S = R[X; D]. In particular, if S is left (right) quasi-duo, then SJ is left (right) quasi-duo for any subset J ⊆ I. Proof. Take J ⊆ I. Denote by X + the set of all "words" xm1 J the set of all nontrivial monomials in XJ , that is X + jt where xji ranges over Xj, and t, mi > 0. Now put j1 xm2 J is j2 · · · xmt C = { f ∈ S supp(f ) ∩ X + J = ∅ }. To demonstrate that SJ is a right corner of S, we will show C is an additive group, SJ ⊕ C = S and CSJ ⊆ C. It is easy to show the first statement since C is an additive subgroup of S generated by all elements of the following form axt1 im ∈ S with xt1 i1 xt2 J . Again, by the definitions of SJ and C, one has S = SJ + C and SJ ∩ C = {0}. Therefore, S = SJ ⊕ C. Now we must show that CSJ ⊆ C. If f = xt1 i1 xt2 J , then gf ∈ C. Notice that SJ is the set of J and g = xq1 im /∈ X + im ∈ X + i2 · · · xtm i2 · · · xtm j1 xq2 j2 · · · xql jl /∈ X + i2 · · · xtm i1 xt2 QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS 9 all finite sums αf where α ∈ R and f ∈ X + where β ∈ R and g /∈ X + [7, Theorem 1.2]. J , and that C is the set of all finite sums βg J . Thus, CSJ ⊆ C. The last conclusion now follows directly from Analogously, one can show that SJ is a left corner of S and that right quasi-duoness of (cid:3) S implies right quasi-duoness of SJ . Lemma 5.2. Let I be a non-empty countable set and let S = R[X; D] be a left (right) quasi-duo differential polynomial ring in several indeterminates (as above). If R is a simple ring, then R is a field, δ = 0 and I = 1. Proof. Suppose that S is left quasi-duo. (The right quasi-duo case can be treated anal- ogously and is therefore omitted.) Take i ∈ I. By Lemma 5.1, Si = K[xi; δi] is left quasi-duo. Using Lemma 2.1 we conclude that R is a field and that δi = 0. It remains to show that I = 1. If I > 1, then it is well-known that the free polynomial algebra K[X] is left primitive (see e.g. [4, Page 36]). In view of [6, Proposition 4.1], K[X] is a division ring. This is a contradiction. Therefore, I = 1. (cid:3) Theorem 5.3. Let I be a non-empty countable set and let S = R[X; D] be a differential polynomial ring in several indeterminates (as above). If S is left (right) quasi-duo, then I = 1. Proof. The proof is essentially the same as the proof of Proposition 2.2, and we will there- fore omit some details. As in the proof of Proposition 2.2 we need to consider two cases: Case 1. This will lead to that R is a division ring. Thus, by Lemma 5.2, we get that I = 1. Case 2. Analogously to the proof of Proposition 2.2 we will define a surjective ring morphism ϕ : R[X; D] → (R/M0)[X; D], a0 + a1t1 + · · · + antn 7→ a0 + a1t1 + · · · + antn and use it to conclude that R/M0[X; D] is left (right) quasi-duo. By invoking case 1, we conclude that I = 1. (cid:3) References [1] J. P. Bell, B. W. Madill and F. Shinko, Differential polynomial rings over rings satisfying a polynomial identity, J. Algebra 423 (2015), 28 -- 36. [2] V. D. Burkov, Differentially prime rings (Russian), Uspekhi Mat. Nauk 35 (1980), 219 -- 220. [3] E. H. Feller, Properties of primary noncommutative rings, Trans. Amer. Math. Soc. 89 (1958), 79 -- 91. [4] N. Jacobson, Structure of Rings, Revised edition, AMS Colloquium Publications, Vol. 37, Providence (1964). [5] T. Y. Lam, A First Course in Noncommutative Rings, Graduate Texts in Mathematics, Vol. 131, Springer-Verlag, Berlin (1991). [6] T. Y. Lam and A. S. Dugas, Quasi-duo rings and stable range descent, J. Pure Appl. Algebra 195(3) (2005), 243 -- 259. [7] A. Leroy, J. Matczuk and E. R. Puczy lowski, Quasi-duo skew polynomial rings, J. Pure Appl. Algebra 212(8) (2008), 1951 -- 1959. [8] A. Leroy, J. Matczuk and. E. R. Puczy lowski, A description of quasi-duo Z-graded rings, Comm. Algebra 38(4) (2010), 1319 -- 1324. 10 QUASI-DUO DIFFERENTIAL POLYNOMIAL RINGS [9] P. Nystedt, A combinatorial proof of associativity of Ore extensions, Discrete Math. 313(23) (2013), 2748 -- 2750. [10] E. R. Puczy lowski, Questions related to Koethe's nil ideal problem, Algebra and its applications, 269 -- 283, Contemp. Math., 419, Amer. Math. Soc., Providence, RI, (2006). [11] E. R. Puczy lowski, On quasi-duo rings, Noncommutative rings and their applications, 243 -- 252, Con- temp. Math., 634, Amer. Math. Soc., Providence, RI, (2015). [12] Y.-T. Tsai, T.-Y. Wu and C.-L. Chuang, Jacobson radicals of Ore extensions of derivation type, Comm. Algebra 35(3) (2007), 975 -- 982. [13] H.-P. Yu, On quasi-duo rings, Glasgow Math. J. 37(1) (1995), 21 -- 31.
1808.06523	1	1808	2018-08-17T05:12:16	Gr\"obner-Shirshov bases for Temperley-Lieb algebras of the complex reflection group of type $G(d,1,n)$	[ "math.RA" ]	We construct a Gr\"obner-Shirshov basis of the Temperley-Lieb algebra $\mathfrak{T}(d,n)$ of the complex reflection group $G(d,1,n)$, inducing the standard monomials expressed by the generators $\{E_i\}$ of $\mathfrak{T}(d,n)$. This result generalizes the one for the Coxeter group of type $B_n$ in \cite{KimSSLeeDI}. We also give a combinatorial interpretation of the standard monomials of $\mathfrak{T}(d,n)$, relating to the fully commutative elements of the complex reflection group $G(d,1,n)$. In this way, we obtain the dimension formula of $\mathfrak{T}(d,n)$.	math.RA	math	GR OBNER-SHIRSHOV BASES FOR TEMPERLEY-LIEB ALGEBRAS OF THE COMPLEX REFLECTION GROUP OF TYPE G(d, 1, n) JEONG-YUP LEE1, DONG-IL LEE2,∗ AND SUNGSOON KIM3 Abstract. We construct a Grobner-Shirshov basis of the Temperley-Lieb algebra T (d, n) of the complex reflection group G(d, 1, n), inducing the standard monomials expressed by the generators {Ei} of T (d, n). This result generalizes the one for the Coxeter group of type Bn in [13]. We also give a combinatorial interpretation of the standard monomials of T (d, n), relating to the fully commutative elements of the complex reflection group G(d, 1, n). In this way, we obtain the dimension formula of T (d, n). 1. Introduction The Temperley-Lieb algebra appears originally in the context of statistical mechanics [26], and later its structure has been studied in connection with knot theory, where it is known to be a quotient of the Hecke algebra of type A in [10]. Our approach to understanding the structure of Temperley-Lieb algebras is from the noncommutative Grobner basis theory, or the Grobner-Shirshov basis theory more pre- cisely, which provides a powerful tool for understanding the structure of (non-)associative algebras and their representations, especially in computational aspects. With the ever- growing power of computers, it is now viewed as a universal engine behind algebraic or symbolic computation. The main interest of the notion of Grobner-Shirshov bases stems from Shirshov's Com- position Lemma and his algorithm [22] for Lie algebras and independently from Buch- berger's algorithm [4] of computing Grobner bases for commutative algebras. In [2], Bokut applied Shirshov's method to associative algebras, and Bergman mentioned the diamond lemma for ring theory [1]. The main idea of the Composition-Diamond lemma is to establish an algorithm for constructing standard monomials of a quotient algebra by a two-sided ideal generated by a set of relations called Grobner-Shirshov basis. Our set of standard monomials in this algorithm is a minimal set of monomials which are indivisible 2010 Mathematics Subject Classification. Primary 20F55, Secondary 05E15, 16Z05. Key words and phrases. Temperley-Lieb algebra, Grobner-Shirshov basis, Catalan number, fully com- mutative element. 1 This research was supported by NRF Grant # 2017078374. 2 This research was supported by NRF Grant # 2018R1D1A1B07044111. * Corresponding author. 3 This author is grateful to KIAS for its hospitality during this work. 1 2 J.-Y. LEE, D.-I. LEE AND S. KIM by any leading monomial of the polynomials in the Grobner-Shirshov basis. The details on the Grobner-Shirshov basis theory are given in Section 2. The Grobner-Shirshov bases for Coxeter groups of classical and exceptional types were completely determined by Bokut, Lee et al. in [3, 16, 18, 19, 25]. The cases for Hecke algebras and Temperley-Lieb algebras of type A as well as for Ariki-Koike algebras were calculated by Lee et al. in [11, 12, 17]. This paper consists of two principal parts as follows : 1) In the first part of this paper, extending the result for type Bn in[13], we construct a Grobner-Shirshov basis for the Temperley-Lieb algebra T (d, n) of the complex reflec- tion group of type G(d, 1, n) and compute the dimension of T (d, n), by enumerating the standard monomails which are in bijection with the fully commutative elements. The main Theorem goes as follows : Theorem 1.1 (Main theorem 4.3). The algebra T (d, n) has a Grobner-Shirshov basis bRT (d,n) with respect to our monomial order < (i.e. degree-lexicographic order with E0 < E1 < · · · < En−1) : Ed 0 − (d − 1)δE0, EiE0 − E0Ei 0 E1 − (k + 1)E0E1 0 E1E0 − (k + 1)E1E0 0 E1,jEi − Ei−2,1Ek E0E1Ek E1Ek 0 E1,jEi for 1 < i ≤ n − 1, for 1 ≤ k < d, for 1 ≤ k < d, for i > j + 1 ≥ 1, for 1 ≤ i ≤ n − 1, for i > j + 1 > 1, for i > j > 0, for i > j > 0. bRT (d,n) : Ei,1Ek E2 i − δEi EiEj − EjEi Ei,jEi − Ei−2,jEi EjEi,j − EjEi,j+2 The cardinality of the set of bRT (d,n)-standard monomials is where Fn,k(x) =Pk in [20, §2.3]. dim T (d, n) = (d − 1)(Fn,n−1(d) − 1) + dCn s=0 C(n, s)xk−s is the (n, k)th Catalan triangle polynomial, introduced We remark here that by specializing d = 2, we recover the formula for the Temperley- Lieb algebra T (Bn) of the Bn. 2) In the second part of this paper, we try to understand some combinatorial aspects on the dimension of the Temperley-Lieb algebra T (d, n). In [14, 15], Kleshchev and Ram constructed a class of representations called homoge- neous representations of the KLR algebra (or a quiver Hecke algebra) and showed that the homogeneous representations can be parametrized by the set of fully commutative TEMPERLEY-LIEB ALGEBRAS 3 elements of the corresponding Coxeter group. These elements were studied by Fan [5] and Stembridge [23] for Coxeter groups and parametrize the bases of the corresponding Temperley-Lieb algebras. Motivated by the bijective correspondance of the homogeneous representations of KLR algebras (or a quiver Hecke algebras) with fully commutative elements, Feinberg and Lee studied the fully commutative elements of the Coxeter groups of types A and D in their papers [7] and [8] and obtained a dimension formula of the homogeneous representations by using Dyck paths. {Standard Monomials of T (d, n)} bijective { generalized Dyck paths } {fully commutative elements of G(d, 1, n)} Figure 1. These three sets are bijective for G(d, 1, n) Their combinatorial strategy is as follows : For type A, we decompose the fully commutative elements into natural subsets accord- ing to the lengths of fully commutative elements and show that the fully commutative elements of a given length k can be parametrized by the Dyck paths of semi-length n with the property that (the sum of peak heights)−(the number of peaks)= k using the canonical form of reduced words for fully commutative elements. For type D, from the set of fully commutative element written in canonical form, we decompose the fully com- mutative elements according to the same type of prefixes and call the set with exactly the same prefix a collection. Then they prove that some collections have the same number of elements by showing that those collections contain the same set of prefixes. We then group those collections together and call the group a Packet. This decomposition process is called the packet decomposition. In the article [6], we generalize the above strategy to the complex reflection group of type G(d, 1, n) for the enumeration of the fully commutative elements of G(d, 1, n). The main result on the combinatorial aspects in Section 5 is that there is a bijection between the standard monomials of T (d, n) in the first part of this paper and the fully commutative elements of G(d, 1, n) in the second part of this paper. In this way, we realize an explicit computation of the dimension of T (d, n). Our canonical forms for the reduced elements will be slightly in different form from the ones in [6] and so is the packet decomposition process. Though, the fully commutative 4 J.-Y. LEE, D.-I. LEE AND S. KIM elements in this paper are better adapted for the standard monomials induced from a Grobner-Shirshov basis. 2. Preliminaries 2.1. Grobner-Shirshov basis. We recall a basic theory of Grobner-Shirshov bases for associative algebras so as to make the paper self-contained. Some properties listed below without proofs are well-known and the readers are invited to see the references cited next to each claim for further detailed explanations. Let X be a set and let hXi be the free monoid of associative words on X. We denote the empty word by 1 and the length (or degree) of a word u by l(u). We define a total-order < on hXi, called a monomial order as follows ; if x < y implies axb < ayb for all a, b ∈ hXi. Fix a monomial order < on hXi and let FhXi be the free associative algebra generated by X over a field F. Given a nonzero element p ∈ FhXi, we denote by p the monomial (called the leading monomial) appearing in p, which is maximal under the ordering <. Thus p = αp +P βiwi with α, βi ∈ F, wi ∈ hXi, α 6= 0 and wi < p for all i. If α = 1, p is Let S be a subset of monic elements in FhXi, and let I be the two-sided ideal of FhXi said to be monic. generated by S. Then we say that the algebra A = FhXi/I is defined by S. Definition 2.1. Given a subset S of monic elements in FhXi, a monomial u ∈ hXi is said to be S-standard (or S-reduced) if u cannot be expressed as asb, that is u 6= asb, for any s ∈ S and a, b ∈ hXi. Otherwise, the monomial u is said to be S-reducible. Lemma 2.2 ([1, 2]). Every p ∈ FhXi can be expressed as (2.1) p =X αiaisibi +X βjuj, where αi, βj ∈ F, ai, bi, uj ∈ hXi, si ∈ S, aisibi ≤ p, uj ≤ p and uj are S-standard. Remark. The termP βjuj in the expression (2.1) is called a normal form (or a remainder) of p with respect to the subset S (and with respect to the monomial order <). In general, a normal form is not unique. As an immediate corollary of Lemma 2.2, we obtain: Proposition 2.3. The set of S-standard monomials spans the algebra A = FhXi/I defined by the subset S, as a vector space over F. Let p and q be monic elements in FhXi with leading monomials p and q. We define the composition of p and q as follows. Definition 2.4. (a) If there exist a and b in hXi such that pa = bq = w with l(p) > l(b), then we define (p, q)w := pa − bq, called the composition of intersection. TEMPERLEY-LIEB ALGEBRAS 5 (b) If there exist a and b in hXi such that a 6= 1, apb = q = w, then we define (p, q)a,b := apb − q, called the composition of inclusion. Let p, q ∈ FhXi and w ∈ hXi. We define the congruence relation on FhXi as follows: p ≡ q mod (S; w) if and only if p − q =P αiaisibi, where αi ∈ F, ai, bi ∈ hXi, si ∈ S, and aisibi < w. Definition 2.5. A subset S of monic elements in FhXi is said to be closed under com- position if (p, q)w ≡ 0 mod (S; w) and (p, q)a,b ≡ 0 mod (S; w) for all p, q ∈ S, a, b ∈ hXi whenever the compositions (p, q)w and (p, q)a,b are defined. The following theorem is a main tool for our results in the subsequent sections. Theorem 2.6 (Composition Lemma [1, 2]). Let S be a subset of monic elements in FhXi. Then the following conditions are equivalent : (a) S is closed under composition. (b) For each p ∈ FhXi, a normal form of p with respect to S is unique. (c) The set of S-standard monomials forms a linear basis of the algebra A = FhXi/I defined by S. Definition 2.7. A subset S of monic elements in FhXi satisfying one of the equivalent conditions in Theorem 2.6 is called a a Grobner-Shirshov basis for the algebra A defined by S. 3. Temperley-Lieb algebras of types An−1 and Bn 3.1. Temperley-Lieb algebra of type An−1. First, we review the results on Temperley- Lieb algebras T (An−1) (n ≥ 2). Define T (An−1) to be the associative algebra over the complex field C, generated by X = {E1, E2, . . . , En−1} with defining relations: E2 i = δEi RT (An−1) : EiEj = EjEi EiEjEi = Ei for 1 ≤ i ≤ n − 1, (idempotent relations) for i > j + 1, (commutative relations) (untwisting relations) for j = i ± 1, where δ ∈ C is a parameter. We call the first and second relations to be the quadratic and commutative relations, respectively. Our monomial order < is taken to be the degree- lexicographic order with E1 < E2 < · · · < En−1. We write Ei,j = EiEi−1 · · · Ej for i ≥ j (hence Ei,i = Ei). By convention Ei,i+1 = 1 for i ≥ 1. 6 J.-Y. LEE, D.-I. LEE AND S. KIM Proposition 3.1. ([11, Proposition 6.2]) The Temperley-Lieb algebra T (An−1) has a Grobner-Shirshov basis bRT (An−1)as follows: i − δEi E2 (3.1) bRT (An−1) : EiEj − EjEi Ei,jEi − Ei−2,jEi EjEi,j − EjEi,j+2 for 1 ≤ i ≤ n − 1, for i > j + 1, for i > j, for i > j. The corresponding bRT (An−1)-standard monomials are of the form Ei1,j1Ei2,j2 · · · Eip,jp (0 ≤ p ≤ n − 1) (3.2) where 1 ≤ i1 < i2 < · · · < ip ≤ n − 1, i1 ≥ j1, i2 ≥ j2, . . . , ip ≥ jp 1 ≤ j1 < j2 < · · · < jp ≤ n − 1, (the case of p = 0 is the monomial 1). We denote the set of bRT (An−1)-standard monomials by MT (An−1). Note that the number of bRT (An−1)-standard monomials equals the nth Catalan number, i.e. MT (An−1) = 1 n + 1(cid:18)2n n(cid:19) = Cn. Remark. There are many combinatorial ways to realize the Catalan number Cn, but among those, It is well-known that Cn represents the number of Dyck paths of length 2n starting from the point (0, 0) ending at (n, n) not passing over the diagonal of the n × n-lattice plane. Example 3.2. (1) Note that MT (A2) = C3 = 5. Explicitly, the bRT (A2)-standard mono- 1, E1, E2,1, E2, E1E2. mials are as follows: We will give combinatorial interpretations of this set via Dyck paths and fully commutative elements in the last section of this article. (2) Another example with n = 3 : we have MT (A3) = C4 = 14. Explicitly, the bRT (A3)-standard monomials are as follows: 1, E1, E2,1, E2, E1E2, E3,1, E3,2, E3, E1E3,2, E1E3, E2,1E3,2, E2,1E3, E2E3, E1E2E3. 3.2. Temperley-Lieb algebra of type Bn. Now we consider the Coxeter diagram for type Bn : (3.3) • 0 • 1 • • n−2 • . n−1 Let W (Bn) be the Weyl group with generators {si}0≤i<n and the following defining relations : TEMPERLEY-LIEB ALGEBRAS 7 - quadratic relations : s2 i = 1 for 0 ≤ i ≤ n − 1 sisj = sjsi sisi+1si = si+1sisi+1 s1s0s1s0 = s0s1s0s1. for i − j > 1 (i, j = 1, . . . , n), for 1 ≤ i < n − 1, (3.4) - braid relations : Then W (Bn) is the Weyl group of type Bn, which is isomorphic to (Z/2Z)n ⋊ Sn. Let H := H(Bn) be the Iwahori-Hecke algebra of type Bn, which is the associative algebra over A := Z[q, q−1], generated by {Ti}0≤i<n with defining relations : - quadratic relations : (Ti − q)(Ti + q−1) = 0 for 0 ≤ i ≤ n − 1, TiTj = TjTi TiTi+1Ti = Ti+1TiTi+1 T1T0T1T0 = T0T1T0T1 for i − j > 1 (i, j = 0, . . . , n − 1), for 1 ≤ i < n − 1, - braid relations : where q is a parameter and Ti := Tsi, the generator corresponding to the reflection si. Let T (Bn) (n ≥ 2) be the Temperley-Lieb algebra of type Bn, which is a quotient of the Hecke algebra H(Bn). T (Bn) is the associative algebra over the complex field C, generated by X = {E0, E1, . . . , En−1} with defining relations : (3.5) RT (Bn) : E2 i = δEi EiEj = EjEi EiEjEi = Ei EiEjEiEj = 2EiEj for 0 ≤ i ≤ n − 1, for i > j + 1, for j = i ± 1, i, j > 0, for {i, j} = {0, 1}, where δ ∈ C is a parameter. Fix our monomial order < to be the degree-lexicographic order with E0 < E1 < · · · < En−1. We write Ei,j = EiEi−1 · · · Ej for i ≥ j ≥ 0, and Ei,j = EiEi+1 · · · Ej for i ≤ j. By convention, Ei,i+1 = 1 and Ei+1,i = 1 for i ≥ 0. We can easily prove the following lemma. Lemma 3.3. ([13, §4]) The following relations hold in T (Bn): Ei,0E1,jEi = Ei−2,0E1,jEi for i > j + 1 ≥ 1. monomials. Among the monomials in MT (Bn), we consider the monomials which are not Let bRT (Bn) be the collection of defining relations (3.5) combined with (3.1) and the relations in Lemma 3.3. From this, we denote by MT (Bn) the set of bRT (Bn)-standard bRT (An−1)-standard. That is, we take only bRT (Bn)-standard monomials which are not of the 8 J.-Y. LEE, D.-I. LEE AND S. KIM form (3.2). This set is denoted by M 0 E0. We decompose the set M 0 T (Bn) into two parts as follows : T (Bn). Note that each monomial in M 0 T (Bn) contains M 0 T (Bn) = M 0+ T (Bn) ∐ M 0− T (Bn) where the monomials in M 0+ T (Bn) are of the form E0Ei1,j1Ei2,j2 · · · Eip,jp (0 ≤ p ≤ n − 1) with 1 ≤ i1 < i2 < · · · < ip ≤ n − 1, i1 ≥ j1, i2 ≥ j2, . . . , ip ≥ jp, and jk > 0 (1 ≤ k < p) implies jk < jk+1 0 ≤ j1 ≤ j2 ≤ · · · ≤ jp ≤ n − 1, (the case of p = 0 is the monomial E0), and the monomials in M 0− T (Bn) are of the form with E′ i1,j1Ei2,j2 · · · Eip,jp (1 ≤ p ≤ n − 1) E′ i,j = Ei,0E1,j and the same restriction on i's and j's as above. It can be easily checked that M 0 T (Bn) is Counting the number of elements in M 0 the set of bRT (Bn)-standard monomials which are not bRT (An−1)-standard. Theorem 3.4. ([13, §4]) The algebra T (Bn) has a Grobner-Shirshov basis bRT (Bn) with T (Bn), we obtain the following theorem. respect to our monomial order <: E2 i − δEi bRT (Bn) : EiEj − EjEi Ei,jEi − Ei−2,jEi EjEi,j − EjEi,j+2 EiEjEiEj − 2EiEj Ei,0E1,jEi − Ei−2,0E1,jEi for 0 ≤ i ≤ n − 1, for i > j + 1, for i > j > 0, for i > j > 0. for {i, j} = {0, 1}, for i > j + 1 ≥ 1. The cardinality of the set MT (Bn), i.e. the set of bRT (Bn)-standard monomials, is dim T (Bn) = (n + 2)Cn − 1. Example 3.5. For n = 3, we have MT (B3) = (3 + 2)C3 − 1 = 24. We enumerate the standard monomials in M 0 T (B3) = MT (B3) \ MT (A3), that is, the standard monomials containing E0, of cardinality 24 − 5 = 19 as follows (using the notation E′ i,j := Ei,0E1,j) : TEMPERLEY-LIEB ALGEBRAS 9 E0, E0E1,0, E1,0, E0E1, E′ 2,1 = E2E1E0E1, E0E2, E′ E′ E0E1,0E2,1, E1,0E2,1, E0E1,0E2, E1,0E2, E0E1E2, E′ 2 = E2E1E0E1E2, E0E1,0E2,0, E1,0E2,0, 1E2 = E1E0E1E2. 1 = E1E0E1, E0E2,0, E2,0, E0E2,1, 4. Grobner-Shirshov bases for Temperley-Lieb algebras of the complex reflection group of type G(d, 1, n) In this section, we try to generalize the result of the previous section to the complex reflection group of type G(d, 1, n), where d ≥ 2 and n ≥ 2. Let Wn := G(d, 1, n) be the wreath product (Z/dZ)n ⋊ Sn of the cyclic group Z/dZ and the symmetric group Sn, with the following diagram : d (cid:13) (cid:13) s1 s0 · · · · · · (cid:13) sn−2 (cid:13) sn−1 In other words, Wn is the group with generators {si}0≤i<n and the following defining relations : s2 i = 1 for 1 ≤ i ≤ n − 1 and sd 0 = 1, braid relations : sisj = sjsi sisi+1si = si+1sisi+1 s1s0s1s0 = s0s1s0s1 if i − j > 1 (i, j = 0, . . . , n − 1), if 1 ≤ i < n − 1, Let Hn := H(Wn) be the Ariki-Koike algebra of type G(d, 1, n), cyclotomic Hecke algebra which is the associative algebra over A := Z[ζ, q, q−1], generated by {Ti}0≤i<n with defining relations : (Ti − q)(Ti + q−1) = 0 for 0 ≤ i ≤ n − 1, (T0 − qm0)(T0 − qm1ζ)(T0 − qm2ζ 2) · · · (T0 − qmd−1ζ d−1) = 0, TiTj = TjTi TiTi+1Ti = Ti+1TiTi+1 T1T0T1T0 = T0T1T0T1 for i − j > 1 (i, j = 0, . . . , n − 1), for 1 ≤ i < n − 1, braid relations : where ζ is a primitive dth root of unity and Ti := Tsi, the generator corresponding to the reflection si, and mi ∈ N. Similarly, we define the Temperley-Lieb algebra of G(d, 1, n) (d, n ≥ 2), denoted by T (d, n), as a quotient of the Ariki-Koike algebra of G(d, 1, n) (d, n ≥ 2). 10 J.-Y. LEE, D.-I. LEE AND S. KIM Definition 4.1. The Temperley-Lieb algebra T (d, n) is the associative algebra over the complex field C, generated by X = {E0, E1, . . . , En−1} with defining relations: E2 i = δEi for 1 ≤ i ≤ n − 1, (4.1) RT (d,n) : Ed 0 = (d − 1)δE0, EiEj = EjEi EiEjEi = Ei 0 E1 = (k + 1)E0E1 0 E1E0 = (k + 1)E1E0 E0E1Ek E1Ek for i > j + 1, for j = i ± 1, i, j > 0, for 1 ≤ k < d, for 1 ≤ k < d, where δ ∈ C is a parameter. Fix the same monomial order < as in the previous section, i.e. the degree-lexicographic order with E0 < E1 < · · · < En−1. Lemma 4.2. The following relations hold in T (d, n); 0 E1,jEi = Ei−2,1Ek Ei,1Ek 0 E1,jEi for i > j + 1 ≥ 1 and 1 ≤ k < d. Proof. Since 2 ≤ i ≤ n − 1 and 0 ≤ j ≤ i − 2, we calculate that Ei,1Ek 0 E1,jEi = (EiEi−1Ei)Ei−2,1Ek 0 E1,j = EiEi−2,1Ek 0 E1,j = Ei−2,1Ek 0 E1,jEi by the commutative relations and EiEi−1Ei = Ei. (cid:3) The set of defining relations (4.1) combined with (3.1) and the relations in Lemma 4.2 is denoted by bRT (d,n). We enumerate bRT (d,n)-standard monomials containing E0 considering the following two types of standard monomials. The first type of standard monomials is : (Ei1,1Ek1 0 )(Ei2,1Ek2 0 ) · · · (Eiq,1Ekq 0 )Eiq+1,jq+1 · · · Eip,jp (1 ≤ p ≤ n − 1, 1 ≤ q ≤ p) (4.2) with 1 ≤ k1, k2, . . . , kq < d, 1 ≤ jq+1 < jq+2 < · · · < jp ≤ n − 1, 0 ≤ i1 < i2 < · · · < iq < iq+1 < · · · < ip ≤ n − 1, iq+1 ≥ jq+1, iq+2 ≥ jq+2, . . . , ip ≥ jp. In a similar way as we did in the proof of [13, Theorem 4.2], to count the number of standard monomials of the above form (4.2), we associate those standard monomials bijectively to the paths defined below. To each monomial 0 )Eiq+1,jq+1 · · · Eip,jp, we associate a unique path, which we call a G(d, 1, n)-Dyck path, 0 ) · · · (Eiq ,1Ekq (Ei1,1Ek1 (i1, 0)k1 → · · · → (iq, 0)kq → (iq+1, jq+1) → · · · → (ip, jp) → (n, n). TEMPERLEY-LIEB ALGEBRAS 11 Here, a path consists of moves to the east or to the north, not above the diagonal in the lattice plane. The move from (i, j) to (i′, j′) (i < i′ and j < j′) is a concatenation of eastern moves followed by northern moves. As an example, the monomial E2 0 E1,0E2,1 in T (3, 3) corresponds to the figure below A(0, 0)2 → B(1, 0)→C(2, 1) → (3, 3). (3,3) C (2,1) C A B (0, 0)2 (1,0) Table 1. The path corresponding to E2 0E1,0E2,1 in T (3, 3) We note that in the move from B(1, 0) to C(2, 1), there is a concatenation of an eastern move followed by a northern move as well as in the move from C(2, 1) to the plot (3, 3) as shown in Table 1. The second type of standard monomials is: (Ei1,1Ek 0 E1,j1)Ei2,j2Ei3,j3 · · · Eip,jp (1 ≤ p ≤ n − 1) (4.3) with 1 ≤ k < d, 1 ≤ j1 < j2 < · · · < jp ≤ n − 1, 1 ≤ i1 < i2 < · · · < ip ≤ n − 1, i1 ≥ j1, i2 ≥ j2, . . . , ip ≥ jp. Now we let MT (d,n) be the set of standard monomials of the forms (4.2) and (4.3) combined with (3.2). The following is our main theorem. 12 J.-Y. LEE, D.-I. LEE AND S. KIM Theorem 4.3. The algebra T (d, n) has a Grobner-Shirshov basis bRT (d,n) with respect to our monomial order < (i.e. degree-lexicographic order with E0 < E1 < · · · < En−1) : Ed 0 − (d − 1)δE0, EiE0 − E0Ei 0 E1 − (k + 1)E0E1 0 E1E0 − (k + 1)E1E0 0 E1,jEi − Ei−2,1Ek E0E1Ek E1Ek 0 E1,jEi E2 i − δEi EiEj − EjEi Ei,jEi − Ei−2,jEi EjEi,j − EjEi,j+2 for 1 < i ≤ n − 1, for 1 ≤ k < d, for 1 ≤ k < d, for i > j + 1 ≥ 1, for 1 ≤ i ≤ n − 1, for i > j + 1 > 1, for i > j > 0, for i > j > 0. bRT (d,n) : Ei,1Ek in [20, §2.3]. dim T (d, n) = (d − 1)(Fn,n−1(d) − 1) + dCn The cardinality of the set MT (d,n), i.e. the set of bRT (d,n)-standard monomials, is where Fn,k(x) =Pk composition, we have only to check for the relations containing E0, to show that bRT (d,n) Proof. Since the set of relations (3.1) which do not contain E0 is already closed under s=0 C(n, s)xk−s is the (n, k)th Catalan triangle polynomial, introduced is closed under composition. (I) First, consider the compositions between Ed 0 − (d − 1)δE0 and another relation. = −E0EiEd−1 1) (EiE0 − E0Ei)Ed−1 0 − Ei(cid:0)Ed 0 − (d − 1)δE0(cid:1) 0 + (d − 1)δEiE0 = −Ei(cid:0)Ed 0 − (d − 1)δE0(cid:1) = 0. (cid:0)E0E1Ek 0 E1 − (k + 1)E0E1(cid:1) 0 E1(cid:0)Ed 0 − (d − 1)δE0(cid:1) 0 − (d − 1)δE0(cid:1) E1Ek 0 E1E0 − (k + 1)E1E0(cid:1) Ed−1 = −(d − 1)δE0E1Ek = −(d − 1)δ(k + 1)E0E1 + (k + 1)(d − 1)δE0E1 = 0. = −(k + 1)E1Ed = −(k + 1)(d − 1)δE1E0 + (d − 1)δ(k + 1)E1E0 = 0. 0 E1 − Ed−1 0 E1 + (k + 1)Ed 2) (cid:0)Ed 3) (cid:0)E1Ek 0 + (d − 1)δE1Ek 0 − E1Ek 0 E1E0 0 E1 0 (II) Next, we take the relation EiE0 − E0Ei and another relation. 1) (E2 2) (EiE0 − E0Ei)E1Ek = −E0EiE1Ek i − δEi)E0 − Ei(EiE0 − E0Ei) = −δEiE0 + EiE0Ei = E0(−δE0 + E2 i ) = 0. 0 E1 − Ei(cid:0)E0E1Ek 0 E1 + (k + 1)EiE0E1 = −Ei(cid:0)E0E1Ek 0 E1,jEi)E0 − Ei,1Ek 0 E1 − (k + 1)E0E1(cid:1) 0 E1 − (k + 1)E0E1(cid:1) = 0. 0 E1,j(EiE0 − E0Ei) 3) (Ei,1Ek 0 E1,jEi − Ei−2,1Ek = −Ei−2,1Ek = (−Ei−2,1Ek 0 E1,jEiE0 + Ei,1Ek 0 E1,jEi + Ei,1Ek 0 E1,jE0Ei 0 E1,jEi)E0 = 0. 4) (EiEj − EjEi)E0 − Ei(EjE0 − E0Ej) = −EjEiE0 + EiE0Ej = 0. TEMPERLEY-LIEB ALGEBRAS 13 5) (Ei,jEi − Ei−2,jEi)E0 − Ei,j(EiE0 − E0Ei) = −Ei−2EiE0 + Ei,jE0Ei = (−Ei−2Ei + Ei,jEi)E0 = 0. 6) (EjEi,j − EjEi,j+2)E0 − EjEi,j+1(EjE0 − E0Ej) = −EjEi,j+2E0 + EjEi,j+1E0Ej = (−EjEi,j+2 + EjEi,j)E0 = 0. (III) Calculate for the relation E0E1Ek 0 E1 − (k + 1)E0E1 and another relation. 0 E1 − E0E1Ek−1 0 E1 = 0. 0 (cid:0)E0E1Ek 0 E1E0 − (k + 1)E1E0(cid:1) 0(cid:0)E1Ek (cid:0)E0E1Ek 0 E1(cid:0)E0E1Ek 0 E1 − (k + 1)E0E1(cid:1) 0 E1E0 − (k + 1)E1E0(cid:1) 0 E1 − (k + 1)E0E1(cid:1) 0 E1 − (k + 1)E0E1(cid:1) 0 E1E0 = 0. 0 0 E1E0 − E0E1Ek = −(k + 1)E0E1Ek = −(k + 1)E0E1Ek 0 E1 + (k + 1)E0E1Ek 0 E1E0 + (k + 1)E0E1Ek = −(k + 1)E0E1E0 + (k + 1)E0E1E0 = 0. 1) (cid:0)E0E1Ek 2) (cid:0)E0E1Ek 3) (cid:0)E0E1Ek 4) (cid:0)E1Ek 5) (cid:0)E1Ek 6) (cid:0)E0E1Ek 7) (cid:0)E0E1Ek 0 E1 − (k + 1)E0E1(cid:1) Ek 0 E1 − (k + 1)E0E1(cid:1) E0 − E0(cid:0)E1Ek 0 E1 − (k + 1)E0E1(cid:1) Ek 0 E1E0 − (k + 1)E1E0(cid:1) Ek−1 0 E1E0 − (k + 1)E1E0(cid:1) E1Ek 0 E1 − (k + 1)E0E1(cid:1) E1 − E0E1Ek 0 E1 − (k + 1)E0E1(cid:1) Ei,1 − E0E1Ek 0 E1 + (k + 1)E1Ek = −(k + 1)E0E2 = −(k + 1)E1Ek 1 + δE0E1Ek 0 E1 = 0. 0 E1 − E1Ek = −(k + 1)E1E0E1Ek = −(k + 1)E1(k + 1)E0E1 + (k + 1)2E1E0E1 = 0. 0 E1 + (k + 1)E1Ek 0 E1E0E1 0 E1 − E1Ek−1 0 (E2 = −(k + 1)E0E1Ei,1 + E0E1Ek = −(k + 1)E0E1Ei,1 + (k + 1)E0E1Ei,3 = −(k + 1)E0(E1Ei,1 − E1Ei,3) = 0. 0 E1Ei,3 1 − δE1) 0 E1 = −(k + 1)δE0E1 + δ(k + 1)E0E1 = 0. 0 (E1Ei,1 − E1Ei,3) (IV) Check for the relation E1Ek 0 E1E0 − (k + 1)E1E0 and another relation. 0 E1E0 − (k + 1)E1E0(cid:1) 3) (EiE1 − E1Ei)Ek 1) (cid:0)E1Ek 2) (E2 = −(k + 1)E1Ek 1 − δE1)Ek = −δE1Ek 0 E1E0 + (k + 1)E2 0 E1E0 + (k + 1)E1Ek 0 E1E0 − E1Ek 0 E1E0 = 0. 0 E1E0 − (k + 1)E1E0(cid:1) Ek−1 0(cid:0)E1Ek 0 E1E0 − E1(cid:0)E1Ek 0 E1E0 − (k + 1)E1E0(cid:1) 0 E1E0 − Ei(cid:0)E1Ek 0 E1E0 − (k + 1)E1E0(cid:1) 0 E1E0 + (k + 1)EiE1E0 = −Ei(cid:0)E1Ek 0 E1E0 − E1Ei,2(cid:0)E1Ek 0 E1E0 + (k + 1)E1Ei,1E0 = −Ei,3(cid:0)E1Ek = −E1Ei,3Ek = −E1EiEk 4) (E1Ei,1 − E1Ei,3)Ek 0 E1E0 − (k + 1)E1E0(cid:1) 0 E1E0 − (k + 1)E1E0(cid:1) = 0. 0 E1E0 − (k + 1)E1E0(cid:1) = 0. 1E0 = −δ(k + 1)E1E0 + (k + 1)δE1E0 = 0. (V) Finally, consider Ei,1Ek 0 E1,jEi − Ei−2,1Ek 0 E1,jEi and another relation. 1) (Ei,1Ek 0 E1,jEi − Ei−2,1Ek = −Ei−2,1Ek 0 E1,jE2 i + δEi,1Ek 0 E1,jEi)Ei − Ei,1Ek 0 E1,j(E2 0 E1,jEi = δ(−Ei−2,1Ek i − δEi) 0 E1,jEi + Ei,1Ek 2) (Ei,1Ek 0 E1,jEi − Ei−2,1Ek 0 E1,jEi)Ej − Ei,1Ek 0 E1,j(EiEj − EjEi) 0 E1,jEi) = 0. 3) (Ei,1Ek 0 E1,jEi − Ei−2,1Ek 0 E1,jEi)Ei−1,jEi − Ei,1Ek 0 E1,j(Ei,jEi − Ei−2,jEi) = −Ei−2,1Ek = (−Ei−2,1Ek 0 E1,jEiEj + Ei,1Ek 0 E1,jEi + Ei,1Ek 0 E1,jEjEi 0 E1,jEi)Ej = 0. = −Ei−2,1Ek = (−Ei−2,1Ek 0 E1,jEi,jEi + Ei,1Ek 0 E1,jEi + Ei,1Ek 0 E1,jEi−2,jEi 0 E1,jEi)Ei−2,j = 0. 14 J.-Y. LEE, D.-I. LEE AND S. KIM 4) (Ei,1Ek 0 E1,jEi − Ei−2,1Ek = −Ei−2,1Ek = (−Ei−2,1Ek 0 E1,jEiEℓ,i + Ei,1Ek 0 E1,jEi + Ei,1Ek 5) (EjEi,j − EjEi,j+2)Ej+1,1Ek 0 E1,jEi)Eℓ,i − Ei,1Ek 0 E1,jEiEℓ,i+2 0 E1,jEi)Eℓ,i+2 = 0. 0 E1,jEi − Ej(Ei,1Ek 0 E1,jEi 0 E1,jEi + EjEi−2,1Ek = −EjEi,j+2Ej+1,1Ek = −Ej(Ei,1Ek 0 E1,jEi − Ei−2,1Ek 0 E1,jEi) = 0. 0 E1,j(EiEℓ,i − EiEℓ,i+2) 0 E1,jEi − Ei−2,1Ek 0 E1,jEi) Hence Theorem 2.6 yields that bRT (d,n) is a Grobner-Shirshov basis for T (d, n). Remark. We can easily check that all bRT (d,n)-standard monomials are the ones in MT (d,n). Following the same procedure as in the proof of [13, Theorem 4.2], the number of mono- mials of the form (4.2) is (cid:3) (d − 1) n−1Xs=0 C(n, s)dn−1−s = (d − 1)Fn,n−1(d) by counting the monomials in (4.2) according to iq(= n − 1 − s) = 0, 1, 2, . . . , n − 1 via G(d, 1, n)-Dyck paths. More precisely, if iq = 0 then the monomials are of the form Ek 0 Ei2,j2 · · · Eip,jp, so the number is (d − 1)Cn = (d − 1)C(n, n − 1). If iq = 1 then the number of the monomials is d(d − 1)C(n, n − 2). If iq = 2 then the number is d2(d − 1)C(n, n − 3). In general, if iq = ℓ ≤ n − 1 then the number is dℓ(d − 1)C(n, n − 1 − ℓ). On the other hand, the number of monomials of the form (4.3) is Thus the cardinality of the set MT (d,n) is (d − 1)(Cn − 1). dim T (d, n) = MT (d,n) = Cn + (d − 1)Fn,n−1(d) + (d − 1)(Cn − 1) = (d − 1)(Fn,n−1(d) − 1) + dCn. In particular, with specialisation d = 2, we recover the result for type Bn: dim T (Bn) = (n + 2)Cn − 1. Example 4.4. Consider the cases of n = 3. The the dimension of T (d, 3) is (d − 1)(F3,2(d) − 1) + dCn = (d − 1)(d2 + 3d + 4) + 5d = d3 + 2d2 + 6d − 4. Notice that, for n = 3 and d = 3, MT (3,3) \ MT (B3) = 59 − 24 = 35. The explicit and complete list of the standard monomials in MT (3,3) \ MT (B3), that is, the standard TEMPERLEY-LIEB ALGEBRAS 15 monomials containing E2 0 , is as follows: 0 E1, E2 0, E2 E2 E2 0E1,0, E2 0), E0(E1E2 0 E2,0, E2 E2 0, E0(E2,1E2 0E1,0E2,1, E2 0 )E2,1, E0(E1E2 0E1,0E2,0, (E1E2 0 ), E2 0(E2,1E2 E0(E1E2 E2,1E2 0 E1E2, E2 0E1,0E2, E1E2 0E2,1, E2 0E2, 0, (E1E2 0 )E2,1, (E1E2 0), E2 0(E1E2 0)E2,0, E2 0)E2, 0 )E2,1, E2 0(E1E2 0 )E2,0, 0)E2, E2 0(E1E2 0)E2,0, E0(E1E2 0(E1E2 0)E2, 0 ), E1,0(E2,1E2 0 ), (E1E2 0 )(E2,1E2 0), E0E1,0(E2,1E2 0), E2 0E1,0(E2,1E2 0 ), E0(E1E2 E1E2 0E1, (E1E2 0 )(E2,1E2 0E1)E2, E2,1E2 0 ), E2 0(E1E2 0E1, E2,1E2 0)(E2,1E2 0 E1,2. 0 ), 5. Combinatorial aspects - connections to Fully commutative elements and Dyck paths In this section, we introduce some combinatorial tools which are useful for the enumera- tioin of standard monomials induced from a Grobner-Shirshov basis for a Temperley-Lieb algebra of Coxeter groups. We later try to set up a bijective correspondence between the standard monomials and the fully commutative elements. Let W be a Coxeter group. An element w ∈ W is said to be fully commutative if any reduced word for w can be obtained from any other by interchanges of adjacent commut- ing generators. The fully commutative elements play an important role particularly for computing the dimension of the Temperley -- Lieb algebra of the Coxeter groups. Stembridge [23] classified all of the Coxeter groups which have finitely many fully com- mutative elements. His results completed the work of Fan [5], which was done only for the simply-laced types. In the same paper [5], Fan showed that the fully commutative elements parameterize natural bases corresponding to certain quotients of Hecke algebras. In type An, these give rise to the Temperley -- Lieb algebras (see [10]). Fan and Stembridge also enumerated the set of fully commutative elements. In particular, they showed the following property. Proposition 5.1 ([5, 24]). Let Cn be the nth Catalan number, i.e. Cn = 1 the numbers of fully commutative elements in the Coxeter group of types An, Dn and Bn respectively are given as follows: n+1(cid:0)2n n(cid:1). Then Cn+1 n+3 2 × Cn − 1 (n + 2) × Cn − 1 if the type is An, if the type is Dn, if the type is Bn.  Remark. There are several combinatorial ways to realize the Catalan number Cn. Among those, we consider the Dyck paths in n×n-lattice plane starting from the point at (0, 0) and ending at the point at (n, n) using only with northern and eastern directions at each step 16 J.-Y. LEE, D.-I. LEE AND S. KIM and never passing above the diagonal line. Seeing that the dimension of the Temperley- Lieb algebra T (An−1) of the Coxeter group of type An−1 is the Catalan number Cn which is the number of fully commutative elements, we can realize a bijective correspondence between the Dyck paths and the fully commutative elements for T (An−1) as we can see in the article [8]. The number of fully commutative elements in Coxeter groups of type B and D re- spectively in Proposition 5.1 above can also be understood in pure combinatorial way by using the notions of Catalan's triangle and Packets as mentioned in the articles [7] and [6]. In this section, we recall the definitions and some combinatorial properties of Catalan numbers and Catalan triangles as well as the main results in [7] and [6]. The Catalan numbers can be defined in a recursive way as follows : we set the first entry C(0, 0) = 1. For n ≥ 0 and 0 ≤ k ≤ n, we denote C(n, k) (0 ≤ k ≤ n) the entry in the nth row and kth column of Table 2 below, called the Catalan's Triangle, verifying the following recursive formula. C(n, k) = 1 C(n, k − 1) + C(n − 1, k) C(n − 1, 0) C(n, n − 1) = C(n, n) = Cn 0 if n = 0; if 0 < k < n; if k = 0; if k = n. if k > n.  14 28 ... 1 1 1(= C1) 1 1 1 1 ... 2 3 4 5 ... 2(= C2) 5 9 14 ... 5(= C3) 14(= C4) C(n, k) = (n + k)!(n − k + 1) k!(n + 1)! 42(= C5) 42 ... ... Table 2. Catalan's Triangle ... ... . . . Remark. We consider the Dyck paths in n × n-lattice plane starting from the point at (0, 0) and ending at the point at (n, n) using only with northern and eastern directions at each step and never passing above the diagonal line. Then the coefficient C(n, k) in the Catalan's triangle can be understood as the number of these Dyck paths passing through exactly the point (n, k) (k ≤ n). In [7], Feinberg and Lee showed that the set of fully commutative elements can be decomposed into subsets called collections depending on the shape of the suffix of each reduced element written in canonical form. Then they proved that some collections have TEMPERLEY-LIEB ALGEBRAS 17 the same set of prefixes, i.e. together and call the group a packet and we call this process the packet decomposition. the same cardinality. Now, we group those collections Following the notations in (3.3) and (3.4), for 0 ≤ i < n, we define the words sij and [i, i − 1, . . . , j] := si · · · sj [i] := si [ ] := 1 if i > j, if i = j, if i < j. sjsj+1 · · · si [i] := si [ ] := 1 if i > j, if i = j, if i < j. Lemma 5.2. ([12, Proposition 2.3], cf. [3, Lemma 4.2]) Any element of the Coxeter group of type Bn can be uniquely written in the reduced form 2,a2 · · · s(kn−1) s(k0) 0,a0s(k1) 1,a1s(k2) sji = sj,i by: sij = where 1 ≤ ai ≤ i + 1, ki = 0 or 1, and sk n−1,an−1 ij =(sij si,0s1,j−1 if k = 0, if k = 1. Remark. Notice that there are i + 1 choices for each ai since 1 ≤ ai ≤ i + 1 while i = 0, 1, . . . , n − 1, thus n! choices for the values of ai's altogether. There are 2 ways for each exponent ki of ai, and thus there are 2n choices for the exponents. Thus we have n! · 2n elements in the canonical reduced form. We recall that there are the same number of elements in the type Bn Coxeter group. Among the above canonical elements, since, for 1 ≤ i1 < i2 and j1 > 1, any element s(1) i1,j1s(1) i2,j2 = si1,0s1,j1−1si2,0s1,j2−1 = si1,0s1,j1−2si2,j1+1(sj1−1sj1sj1−1)sj1−2,0s1,j2−1 contains a braid relation sj−1sjsj−1, thus any element containing this form is not fully commutative. Also, for j > 1, any element containing the form s0s(1) i,j = s0si,0s1,j−1 = si,2(s0s1s0s1)s2,j−1 is non-fully-commutative element. Now we consider only the elements of the form (5.1) si1,0si2,0 · · · siℓ,0siℓ+1,aiℓ+1 · · · sn−1,an−1 where 0 ≤ i1 < i2 < · · · < iℓ < n for ℓ ≥ 0, or (5.2) with i ≥ 1. (si,0s1,ai)si+1,ai+1 · · · sn−1,an−1 The left factors si1,0si2,0 · · · siℓ,0 in (5.1), and si,0 in (5.2) will be called the prefix. Similarly the right factors siℓ+1,aiℓ +1 · · · sn−1,an−1 in (5.1), and s1,aisi+1,ai+1 · · · sn−1,an−1 in (5.2) will be called the suffix of the reduced word. Given a reduced word w in the extracted canonical form, we will denote by wp the prefix and by ws the suffix of w and we can write, in a unique way, w = wpws. 18 J.-Y. LEE, D.-I. LEE AND S. KIM We notice that every prefix is a fully commutative element. Some of the collections in the set of fully commutative elements of W (Bn) have the same number of elements and thus we group them together into a set called a (n, k)-packet depending on the form of the prefixes as follows : Definition 5.3. For 0 ≤ k ≤ n, we define the (n, k)-packet of collections: • The (n, 0)-packet is the set of collections labeled by prefixes of the form si1,0si2,0 · · · siℓ,0sn−1,0 (ℓ ≥ 1). • The (n, k)-packet, 1 ≤ k ≤ n − 2, is the set of collections labeled by sn−k,0 or prefixes of the form si1,0si2,0 · · · siℓ,0sn−k−1,0 (ℓ ≥ 1). • The (n, n−1)-packet contains only the collection labeled by s0 = [0] or s1,0 = [1, 0]. • The (n, n)-packet contains only the collection labeled by the empty prefix [ ]. We will denote the (n, k)-packet by PB(n, k). As an example, Table 3 shows all of collections of the packets for type B3. Proposition 5.4 ([6]). The cardinality of the packet PB(n, k) for type Bn is PB(n, k) = 2n−1 − 1 2n−k−1 2 1 if k = 0, if 1 ≤ k ≤ n − 2, if k = n − 1, if k = n. Therefore Pn k=0 PB(n, k) = 2n. We remark that for a fixed k with 0 ≤ k ≤ n (n ≥ 3), each collection with same prefix in the packet PB(n, k) has the same cardinality which is C(n, k). Corollary 5.5. For n ≥ 3, we obtain the identity: (5.3) nXk=0 C(n, k) PB(n, k) = (n + 2) × Cn − 1. We generalize the above method for the complex reflection group of type G(d, 1, n). Lemma 5.6. (Canonical Form for G(d, 1, n) [12, Proposition 2.3]) Any element of the Coxeter group of type G(d, 1, n) can be uniquely written in the reduced form s(k0) 0,a0s(k1) 1,a1s(k2) 2,a2 · · · s(kn−1) n−1,an−1 where 1 ≤ ai ≤ i + 1, 0 ≤ ki ≤ d − 1, and sk ij =(sij si,1sk 0s1,j−1 if k = 0, if k = 1, 2, . . . , d − 1. TEMPERLEY-LIEB ALGEBRAS 19 Remark. Notice that there are n! choices of ai's (i = 0, 1, . . . , n − 1), and there are dn ways to choose ki's. Therefore we have n! · dn elements in the canonical reduced form. We recall that there are the same number of elements in the complex reflection group of type G(d, 1, n). The fully commutative elements of the complex reflection group of type G(d, 1, n) are of the form si1,1sk1 0 si2,1sk2 0 · · · siq ,1skq 0 siq+1,jq+1 · · · sip,jp 0 ≤ q ≤ p, 0 ≤ p ≤ n, 0 ≤ i1 < i2 < · · · < ip ≤ n − 1, iq+1 ≥ jq+1, . . . , ip ≥ jp, 1 ≤ k1, k2, . . . , kq < d, 0 < jq+1 < · · · ≤ jp ≤ n − 1, and (si1,1sk 0s1,j1)si2,j2si3,j3 · · · sip,jp (5.4) with or (5.5) with 1 ≤ k < d, 1 ≤ p ≤ n − 1, 1 ≤ i1 < i2 < i3 < · · · < ip ≤ n − 1, i1 ≥ j1, i2 ≥ j2, . . . , ip ≥ jp. 1 ≤ j1 < j2 < j3 < · · · < jp ≤ n − 1, and The left factor si1,1sk1 0 si2,1sk2 0 · · · siq ,1skq 0 with q ≥ 0 in (5.4) and (5.5) will be called the prefix of the fully commutative element of type G(d, 1, n). Definition 5.7. For 0 ≤ s ≤ n, we define the (n, s)-packet of collections: • The (n, 0)-packet is the set of collections labeled by prefixes of the form si1,1sk1 0 si2,1sk2 0 · · · siq−1,1skq−1 0 sn−1,1skq 0 (q ≥ 2). • The (n, s)-packet, 1 ≤ s ≤ n − 2, is the set of collections labeled by sn−s,1sk 0 or prefixes of the form si1,1sk1 0 si2,1sk2 0 · · · skq−1 iq−1,1sn−s−1,1skq 0 (q ≥ 2). • The (n, n − 1)-packet contains only the collections labeled by sk 0 = [0k] or s1sk 0 = [1, 0k]. • The (n, n)-packet contains only the collection labeled by the empty prefix [ ]. These elements cover the elements of Sn ⊂ G(d, 1, n). We will denote the (n, s)-packet by P(n, s). For clarifying the packet decomposition, we denote (n, s)-packet of type G(d, 1, n) which is not contained in the (n, s)-packet of type Bn by P(n, s)′ = P(n, s) \ PB(n, s). Proposition 5.8. (cf. [6, §4]) (a) Every collection in the packet P(n, s) has C(n, s) elements. 20 J.-Y. LEE, D.-I. LEE AND S. KIM (b) The size of the packet P(n, s) is .P(n, s) = (d − 1)(dn−1 − 1) (d − 1)dn−1−s 2(d − 1) 1 if s = 0, if 1 ≤ s ≤ n − 2, if s = n − 1, if s = n. Proof. (a) The collections in P(n, 0) have no other element than the prefix. It's also clear that the collections in P(n, n − 1) have Cn = C(n, n − 1) elements and the collection in P(n, n) has Cn = C(n, n) elements. For the collections in P(n, s) with 1 ≤ s ≤ n − 2, the number of suffixes attached to the given prefix is exactly C(n, s) by counting the Dyck paths from (n − s, 0) to (n, n) via the Dyck paths, reflected relative to y = n − x in the xy-plane, from (0, 0) to (n, s). (cid:3) (b) Counting the number of prefixes in each packet, we get its packet size. In the same way as the formula (5.3), we obtain the identity for the complex reflection group of type G(d, 1, n): nXs=0 C(n, s) P(n, s) = (d − 1)(dn−1 − 1) + C(n, s)(d − 1)dn−1−s + 2(d − 1)Cn + Cn n−2Xs=1 = (d − 1) (Fn,n−1(d) − 1) + dCn, which is equal to dim T (d, n). As an example, Table 3 shows all of collections of the packets for type G(3, 1, 3). 6. Temperley-Lieb algebras of types G(d, d, n) and G(d, r, n) Since we obtain a C-basis of T (d, n) and its multiplication structure, we define the subalgebras T (d, d, n) and T (d, r, n) in general as follows. Definition 6.1. (a) The subalgebra T (d, d, n) is a subalgebra of T (d, n), whose C-basis consists of  We remark that the elements in (3.2), the elements in (4.2) satisfying k1 + · · · + kq ≡ 0 the elements in (4.3) with k = 1. (mod d), dim T (d, d, n) = Cn + (d − 1)Fn,n−2(d) + Cn − 1 = (d − 1)Fn,n−2(d) + 2Cn − 1. In particular, for d = 2, we recover the same formula as we had in [13] : dim T (2, 2, n) = dim T (Dn) = n + 3 2 Cn − 1. TEMPERLEY-LIEB ALGEBRAS 21 Standard Monomials Fully Commutative Elements MT (A2) 1, E1, E2, E2,1 = E2E1, E1E2 P(3, 3) = C[ ] = {id, s1, s2, s2s1, s1s2} M 0 T (B3) (E0, E0E2, E0E1E2, E0E1, E0E2E1, E1,0, E1,0E2, E1,0E1E2, E1,0E1, E1,0E2,1 PB(3, 2) =(C[0] = {s0, s0s2, s0s1s2, s0s1, s0s2s1} C[1,0] = {s1s0, s1s0s2, s1s0s1s2, s1s0s1, s1s0s2s1} (E2,0, E2,0E1, E2,0E1,2, E0E1,0, E0E1,0E2, E0E1,0E2,1 PB(3, 1) =(C[2,1,0] = {s2s1s0, s2s1s0s1, s2s1s0s1s2} C[0,1,0] = {s0s1s0, s0s1s0s2, s0s1s0s2s1} E0E2,0, E1,0E2,0, E0E1,0E2,0 Monomials containing E0 C[0,2,1,0] = {s0s2s1s0} C[1,0,2,1,0] = {s1s0s2s1s0} C[0,1,0,2,1,0] = {s0s1s0s2s1s0} 0 , E2 E1E2 0 E2, E2 0 , E1E2 0 E1,2, E2 0 E2, E1E2 0 E1, E2 0 E2,1, 0 E1,2, E1E2 0 E1, E1E2 0 E2,1 0s2, s2 0, s1s2 0s1s2, s2 0s2, s1s2 0s1, s2 0s2s1} 0s1s2, s1s2 0s1, s1s2 0s2s1} 0 , E2,1E2 0 E1, E2,1E2 0 E1,0, E2 0 E1,0E2, E2 0 E1,2, 0 E1,0E2,1, 0 E2, E0E1E2 0 E2, E2 0 E1E2 0 E2,1, 0 E2,1 0 , E0E1E2 0 , E2 0 E1E2 0 ), 0 ) Monomials containing E2 0 Table 3. Bijection between Standard Monomials and Fully Commutative Elements of G(3, 1, 3)   T (3,3) (E2 M 02 E2,1E2 E2 E0E1E2 E2 0 E1E2   0 E2,0, 0 E2,0, 0 E1,0E2,0, 0 )E2,0, 0 )E2,0, E2 E1E2 E2 E0(E1E2 E2 0 (E1E2 E0(E2,1E2 0 ), E2 0 (E2,1E2 0 ), E1,0(E2,1E2 (E1E2 0 )(E2,1E2 0 ), E0E1,0(E2,1E2 0 ), E2 0 E1,0(E2,1E2 0 ), E0(E1E2 0 )(E2,1E2 0 )(E2,1E2 E2 0 (E1E2 0 ),   PB(3, 0) =  P(3, 2)′ =(C[02] = {s2 0, s2 C[1,02] = {s1s2 P(3, 1)′ = 0s1s2} 0s1s0s2s1} 0s2s1} 0s2s1}   0s2s1s0} 0s1s0, s2 0, s2s1s2 0s1, s2s1s2 0s1s0s2, s2 0, s0s1s2 0, s2 0s1s2 0s2, s0s1s2 0s2, s2 0s1s2 0s2s1s0} 0s1s0s2s1s0} 0s2s1s0} 0s2s1s0} C[2,1,02] = {s2s1s2 C[02,1,0] = {s2 C[0,1,02] = {s0s1s2 C[02,1,02] = {s2 0s1s2 C[02,2,1,0] = {s2 C[1,02,2,1,0] = {s1s2 C[02,1,0,2,1,0] = {s2 C[0,1,02,2,1,0] = {s0s1s2 C[02,1,02 ,2,1,0] = {s2 0s1s2 C[0,2,1,02] = {s0s2s1s2 0} C[02,2,1,02] = {s2 0s2s1s2 0} C[1,0,2,1,02] = {s1s0s2s1s2 0} C[1,02,2,1,02 ] = {s1s2 0s2s1s2 0} C[0,1,0,2,1,02] = {s0s1s0s2s1s2 0} C[02,1,0,2,1,02 ] = {s2 0s1s0s2s1s2 0} C[0,1,02,2,1,02 ] = {s0s1s2 0s2s1s2 0} C[02,1,02 ,2,1,02] = {s2 0s1s2 0s2s1s2 0}   P(3, 0)′ = 22 J.-Y. LEE, D.-I. LEE AND S. KIM (b) More generally, the subalgebra T (d, r, n) (rd and r ≥ 2) is a subalgebra of T (d, n), whose C-basis consists of the elements in (3.2), the elements in (4.2) satisfying k1 + · · · + kq ≡ 0 the elements in (4.3) with k ∈ {1, r, 2r, · · · , d − 2r, d − r}. (mod r),  We remark that dim T (d, r, n) = Cn + (d − 1)Fn,n−2(d) + (Cn − 1) d r = d r d r d (d − 1)Fn,n−2(d) +(cid:18)1 + r(cid:19) Cn − d r . References [1] G. M. Bergman, The diamond lemma for ring theory, Adv. Math. 29 (1978), 178 -- 218. [2] L. A. Bokut, Imbedding into simple associative algebras, Algebra and Logic 15 (1976), 117 -- 142. [3] L. A. Bokut, L.-S. Shiao, Grobner-Shirshov bases for Coxeter groups, Comm. Algebra 29 (2001), 4305 -- 4319. [4] B. Buchberger, An algorithm for finding the basis elements of the residue class ring of a zero di- mensional polynomial ideal, Ph.D. thesis, Univ. of Innsbruck, 1965 (in German). Translated in: J. Symbolic Comput. 41 (2006), 475 -- 511. [5] C. K. Fan, A Hecke Algebra quotient and some combinatorial applications, J. Algebraic Combin. 5 (1996), 175 -- 189. [6] G. Feinberg, S. Kim, K.-H. Lee, S.-j. Oh, Fully commutative elements of the complex reflection groups, arXiv:1808.04269 [math.GR]. [7] G. Feinberg, K.-H. Lee, Fully commutative elements of type D and homogeneous representations of KLR-algebras, J. Comb. 6 (2015), 535 -- 557. [8] , Homogeneous representations of type A KLR-algebras and Dyck paths, S´em. Lothar. Combin. 75 (2016), Art. B75b. [9] D. Hill, G. Melvin, D. Mondragon, Representations of quiver Hecke algebras via Lyndon bases, J. Pure Appl. Algebra 216 (2012), 1052 -- 1079. [10] V. F. R. Jones, Hecke algebra representations of braid groups and link polynomials, Ann. Math. 126 (1987), 335 -- 388. [11] S.-J. Kang, I.-S. Lee, K.-H. Lee, H. Oh, Hecke algebras, Specht modules and Grobner-Shirshov bases, J. Algebra 252 (2002), 258 -- 292. [12] , Representations of Ariki-Koike algebras and Grobner-Shirshov bases, Proc. London Math. Soc. 89 (2004), 54 -- 70. [13] S. Kim, D.-I. Lee, Grobner-Shirshov bases for Temperley-Lieb algebras of types B and D, submitted (2017), arXiv:1808.05026 [14] A. Kleshchev, A. Ram, Homogeneous representations of Khovanov-Lauda algebras, J. Eur. Math. Soc. 12 (2010), 1293 -- 1306. [15] , Representations of Khovanov-Lauda-Rouquier algebras and combinatorics of Lyndon words, Math. Ann. 349 (2011), 943 -- 975. TEMPERLEY-LIEB ALGEBRAS 23 [16] D. Lee, Grobner-Shirshov bases and normal forms for the Coxeter groups E6 and E7, In: Advances in Algebra and Combinatorics (Guangzhou, 2007), World Sci. Publ., 2008, pp. 243 -- 255. [17] D.-I. Lee, Cyclotomic Hecke algebras of G(r, p, n), Algebr. Represent. Theory 13 (2010), 705 -- 718. [18] [19] D.-I. Lee, J.-Y. Lee Grobner-Shirshov bases for non-crystallographic Coxeter groups, submitted , Standard monomials for the Weyl group F4, J. Algebra Appl. 15 (2016), 1650146:1 -- 8. (2018). [20] K.-H. Lee, S.-j. Oh, Catalan triangle numbers and binomial coefficients, Contemp. Math., to appear. arXiv:1601.06685v2 [21] Sequence A009766, The On-Line Encyclopedia of Integer Sequences, published electronically at http://oeis.org. [22] A. I. Shirshov, Some algorithmic problems for Lie algebras, Sibirk. Math. Z. 3 (1962), 292 -- 296 (in Russian). Translated in: ACM SIGSAM Bull. Commun. Comput. Algebra 33 (1999) no. 2, 3 -- 6. [23] J. R. Stembridge, On the fully commutative elements of Coxeter groups, J. Algebraic Combin. 5 (1996), 353 -- 385. [24] , Some combinatorial aspects of reduced words in finite Coxeter groups, Trans. Amer. Math. Soc. 349 (1997), 1285 -- 1332. [25] O. Svechkarenko, Grobner-Shirshov bases for the Coxeter group E8, Master thesis, Novosibirsk State Univ., 2007. [26] H. N. V. Temperley, E. H. Lieb, Relations between percolation and colouring problems and other graph theoretical problems associated with regular planar lattices: some exact results for the percolation problem, Proc. Roy. Soc. London Ser. A 322 (1971), 251 -- 280. Department of Mathematics Education, Catholic Kwandong University, Gangwondo 25601, Korea E-mail address: [email protected] Department of Mathematics, Seoul Women's University, Seoul 01797, Korea E-mail address: [email protected] LAMFA-CNRS UMR 7352, UPJV, 33 rue St. Leu, 80039 Amiens, France, (membre assoc. ´equipe des groupes, IMJ-PRG Univ. Paris 7) E-mail address: [email protected]
1501.03675	1	1501	2015-01-15T13:39:36	One-parameter formal deformations of Hom-Lie-Yamaguti algebras	[ "math.RA" ]	This paper studies one-parameter formal deformations of Hom-Lie-Yamaguti algebras. The first, second and third cohomology groups on Hom-Lie-Yamaguti algebras extending ones on Lie-Yamaguti algebras are provided. It is proved that first and second cohomology groups are suitable to the deformation theory involving infinitesimals, equivalent deformations and rigidity. However, the third cohomology group is not suitable for the obstructions.	math.RA	math	One-parameter formal deformations of Hom-Lie-Yamaguti algebras Yao Ma1, Liangyun Chen1, Jie Lin2 1 School of Mathematics and Statistics, Northeast Normal University, Changchun, 130024, CHINA 2 Sino-European Institute of Aviation Engineering,Civil Aviation University of China, Tianjin, 300300, CHINA Abstract This paper studies one-parameter formal deformations of Hom-Lie-Yamaguti algebras. The first, second and third cohomology groups on Hom-Lie-Yamaguti algebras extending ones on Lie-Yamaguti algebras are provided. It is proved that first and second cohomology groups are suitable to the deformation theory involving infinitesimals, equivalent deformations and rigidity. However, the third cohomology group is not suitable for the obstructions. Key words: Hom-Lie-Yamaguti algebra, cohomology, deformation MSC(2010): 17A99, 17B56, 16S80 1 Introduction Lie-Yamaguti algebras were introduced by Yamaguti in [26] to give an algebraic in- terpretation of the characteristic properties of the torsion and curvature of homogeneous spaces with canonical connection in [22]. He called them "generalized Lie triple systems" at first, which were later called "Lie triple algebras". Recently, they were renamed as "Lie-Yamaguti algebras". A Hom-type algebra is a kind of algebras whose identities defining the structures are twisted by a linear homomorphism (the twisting map). When the twisting map is the Corresponding author(L. Chen): [email protected]. Supported by NNSF of China (No.11171055 and No.11471090), NSF of Jilin province (No.201115006), Scientific Research Foundation for Returned Scholars Ministry of Education of China and the Fundamen- tal Research Funds for the Central Universities (No.12SSXT139). 1 identity map, one recovers the original algebra. The notion of Hom-type algebras was initially introduced in [12] as "Hom-Lie algebras" to describe the q-deformation of the Witt and the Virasoro algebras. For more information on various Hom-type algebras one may refer to [2, 3, 13, 16, 18, 23, 24, 27, 28]. In particular, the notion of Hom-Lie Yamaguti algebras was introduced by Gaparayi and Nourou in [6]. A Hom-Lie-Yamaguti algebra (HLYA for short) is a quadruple (L, [·, ·], {·, ·, ·}, α) in which L is a vector space over a field K, "[·, ·]" a binary operation and "{·, ·, ·}" a ternary operation on L, and α : L → L a linear map such that for all x, y, z, u, v ∈ L, α([xy]) = [α(x)α(y)], α({xyz}) = {α(x)α(y)α(z)}, (1.1) (1.2) (1.3) (1.4) (1.5) (1.6) (1.7) {α2(u)α2(v){xyz}} = {{uvx}α2(y)α2(z)}+{α2(x){uvy}α2(z)}+{α2(x)α2(y){uvz}}, (1.8) {α(x)α(y)[uv]} = [{xyu}α2(v)] + [α2(u){xyv}], [xx] = 0, {xxy} = 0, (cid:9)x,y,z ([[xy]α(z)] + {xyz}) = 0, (cid:9)x,y,z ({[xy]α(z)α(u)}) = 0, where (cid:9)x,y,z denotes the sum over cyclic permutations of x, y, z. HLYAs generalize Hom-Lie triple systems and Hom-Lie algebras in the same way as Lie-Yamaguti algebras generalize Lie triple systems and Lie algebras, i.e., if {xyz} = 0 for all x, y, z ∈ L, then (L, [·, ·], α) becomes a Hom-Lie algebra; if [xy] = 0 for all x, y ∈ L, then (L, {·, ·, ·}, α2) becomes a Hom-Lie triple system. When α = idL, (L, [·, ·], {·, ·, ·}) becomes a Lie-Yamaguti algebra. There are more examples and properties about HLYA in [6, 7]. A deformation is a tool to study a mathematical object by deforming it into a fam- ily of the same kind of objects depending on a certain parameter. Deformation problems appear in various areas of mathematics, especially in algebra, algebraic and analytic geom- etry, and mathematical physics. The deformation theory was introduced by Kodaira and Spencer to study complex structures of higher dimensional manifolds (see [14]), which was extended to rings and algebras by Gerstenhaber in [8 -- 11] and to Lie algebras by Nijenhuis and Richardson in [21]. They connected deformation theory for associative al- gebras and Lie algebras with Hochschild cohomology and Chevally-Eilenberg cohomology, respectively. See also [1, 4, 5, 15, 17, 19, 20] for more deformation theory. The aim of this paper is to consider the cohomology theory and the one-parameter formal deformation theory of HLYAs based on some work in [15, 17, 23, 25]. The rest of this paper is organized as follows. In section 2, we define the first, second and third cohomology groups on HLYAs and show that the first cohomology group corresponds to the derivations space of a HLYA. Section 3 concerns the one-parameter formal deformation theory of HLYAs. We show that the first and second cohomology groups defined in Section 2 fits this one-parameter formal deformation theory but the third one does not. Throughout this paper K denotes an arbitrary field. 2 2 First, second and third cohomology groups of a Hom-Lie-Yamaguti algebra Inspired by the cohomology theory of Lie-Yamaguti algebras in [25], we introduce the first, second and third cohomology groups of HLYAs. Definition 2.1. Let (L, [·, ·], {·, ·, ·}, α) be a HLYA. An n-linear map f : L × · · · × L n times {z } → L (2.1) (2.2) is called an n-Hom-cochain, if f satisfies f (x1, · · · , x2i−1, x2i, · · · , xn) = 0, for x2i−1 = x2i, f (α(x1), · · · , α(xn)) = α ◦ f (x1, · · · , xn). The set of n-Hom-cochains is denoted by HomC n(L, L), for n ≥ 1. Definition 2.2. Suppose that (L, [·, ·], {·, ·, ·}, α) is a HLYA. (i) A 1-coboundary operator of (L, [·, ·], {·, ·, ·}, α) is a pair of maps (δ1 I , δ1 II) : HomC 1(L, L) × HomC 1(L, L) −→ HomC 2(L, L) × HomC 3(L, L) for f ∈ HomC 1(L, L) and (f, f ) 7−→ (δ1 I f, δ1 IIf ) δ1 I f (x, y) = [xf (y)] + [f (x)y] − f ([xy]), δ1 II f (x, y, z) = {f (x)yz} + {xf (y)z} + {xyf (z)} − f ({xyz}). (ii) A 2-coboundary operator of (L, [·, ·], {·, ·, ·}, α) is a pair of maps (δ2 I , δ2 II) : HomC 2(L, L) × HomC 3(L, L) −→ HomC 4(L, L) × HomC 5(L, L) (f, g) 7−→ (δ2 I f, δ2 IIg) for f ∈ HomC 2(L, L), g ∈ HomC 3(L, L) and δ2 I f (x, y, z, u) ={α(x)α(y)f (z, u)} − f ({xyz}, α2(u)) − f (α2(z), {xyu}) + g(α(x), α(y), [zu]) − [α2(z)g(x, y, u)] − [g(x, y, z)α2(u)], δ2 II g(x, y, u, v, w) ={α2(x)α2(y)g(u, v, w)} − {g(x, y, u)α2(v)α2(w)} − {α2(u)g(x, y, v)α2(w)} − {α2(u)α2(v)g(x, y, w)} + g(α2(x), α2(y), {uvw}) − g({xyu}, α2(v), α2(w)) − g(α2(u), {xyv}, α2(w)) − g(α2(u), α2(v), {xyw}). (iii) A 3-coboundary operator of (L, [·, ·], {·, ·, ·}, α) is a pair of maps (δ3 I , δ3 II) : HomC 4(L, L) × HomC 5(L, L) −→ HomC 6(L, L) × HomC 7(L, L) 3 (f, g) 7−→ (δ3 I f, δ3 IIg) for f ∈ HomC 4(L, L), g ∈ HomC 5(L, L) and δ3 I f (x1, · · · , x6) ={α3(x1)α3(x2)f (x3, · · · , x6)} − {α3(x3)α3(x4)f (x1, x2, x5, x6)} (−1)kf (α2(x1), · · · , \α2(x2k−1), \α2(x2k), · · · , {x2k−1x2kxi}, · · · , α2(x6)) − g(α(x1), · · · , α(x4), [x5x6]) + [α4(x5)g(x1, · · · , x4, x6)] + [g(x1, · · · , x5)α4(x6)], + 2Xk=1 6Xi=2k+1 δ3 II g(x1, · · · , x7) = 3Xk=1 (−1)k+1{α4(x2k−1)α4(x2k)g(x1, · · · , [x2k−1,cx2k, · · · , x7) 3Xk=1 7Xi=2k+1 + + {g(x1, · · · , x5)α4(x6)α4(x7)} − {g(x1, · · · , x4, x6)α4(x5)α4(x7)}, (−1)kg(α2(x1), · · · , \α2(x2k−1), \α2(x2k), · · · , {x2k−1x2kxi}, · · · , α2(x7)) where the sign indicates that the element below must be omitted. Theorem 2.3. The coboundary operators (δi I , δi II) are well defined, for i = 1, 2, 3. Proof. Take (f, f ) ∈ HomC 1(L, L) × HomC 1(L, L). It is clear that δ1 I f and δ1 II f satisfy (2.1). Notice that δ1 I f (α(x), α(y)) =[α(x)f (α(y))] + [f (α(x))α(y)] − f ([α(x)α(y)]) =α([xf (y)]) + α([f (x)y]) − α(f ([xy])) = α ◦ δ1 I f (x, y) and δ1 II f (α(x), α(y), α(z)) ={f (α(x))α(y)α(z)} + {α(x)f (α(y))α(z)} + {α(x)α(y)f (α(z))} − f ({α(x)α(y)α(z)}) =α({f (x)yz} + {xf (y)z} + {xyf (z)} − f ({xyz})) =α ◦ δ1 II f (x, y, z). Then (δ1 I , δ1 II) is well defined. Now let (f, g) ∈ HomC 2(L, L) × HomC 3(L, L). Then δ2 I f and δ2 II g satisfy (2.1) and δ2 I f (α(x), α(y), α(z), α(u)) ={α2(x)α2(y)f (α(z), α(u))} − f (α({xyz}), α3(u)) − f (α3(z), α({xyu})) + g(α2(x), α2(y), α([zu])) − [α3(z)g(α(x), α(y), α(u))] − [g(α(x), α(y), α(z))α3(u)] 4 =α({α(x)α(y)f (z, u)}) − α ◦ f ({xyz}, α2(u)) − α ◦ f (α2(z), {xyu}) + α ◦ g(α(x), α(y), [zu]) − α([α2(z)g(x, y, u)]) − α([g(x, y, z)α2(u)]) =α ◦ δ2 I f (x, y, z, u). Similarly, δ2 II g(α(x), α(y), α(u), α(v), α(w)) = α ◦ δ2 II g(x, y, u, v, w). One proves (δ3 I f, δ3 II g) ∈ HomC 6(L, L) × HomC 7(L, L) if (f, g) ∈ HomC 4(L, L) × HomC 5(L, L) in the same way. Hence the theorem follows. Moreover, for (f, g) ∈ HomC 2(L, L)×HomC 3(L, L), we define another 2-coboundary operator of (L, [·, ·], {·, ·, ·}, α) as (d2 I , d2 II) : HomC 2(L, L) × HomC 3(L, L) −→ HomC 3(L, L) × HomC 4(L, L) where (f, g) 7−→ (d2 If, d2 II g) d2 I f (x, y, z) = (cid:9)x,y,z ([f (x, y)α(z)] + f ([xy], α(z)) + g(x, y, z)), d2 IIg(x, y, z, u) = (cid:9)x,y,z ({f (x, y)α(z)α(u)} + g([xy], α(z), α(u))). It is easy to prove that (d2 I, d2 II) is well defined. Theorem 2.4. With notations as above, we have (δ2 I , δ2 II)(δ1 I , δ1 II) = (0, 0), (d2 I , d2 II)(δ1 I , δ1 II) = (0, 0), (δ3 I , δ3 II)(δ2 I , δ2 II) = 0. Proof. Suppose (f, f ) ∈ HomC 1(L, L) × HomC 1(L, L). Then (δ2 I , δ2 II)(δ1 I , δ1 II)(f, f ) = (δ2 I , δ2 II)(δ1 I f, δ1 II f ) = (δ2 I δ1 I f, δ2 IIδ1 II f ). Using (1.7), (1.8) and (2.2), we have δ2 I δ1 I f (x, y, z, u) ={α(x)α(y)δ1 I f (z, u)} − δ1 I f ({xyz}, α2(u)) − δ1 I f (α2(z), {xyu}) + δ1 II f (α(x), α(y), [zu]) − [α2(z)δ1 II f (x, y, u)] − [δ1 II f (x, y, z)α2(u)] ={α(x)α(y)[zf (u)]} + {α(x)α(y)[f (z)u]} − {α(x)α(y)f ([zu])} − [{xyz}f α2(u)] − [f ({xyz})α2(u)] + f ([{xyz}α2(u)]) − [α2(z)f ({xyu})] − [f α2(z){xyu}] + f ([α2(z){xyu}]) + {f α(x)α(y)[zu]} + {α(x)f α(y)[zu]} + {α(x)α(y)f ([zu])} − f ({α(x)α(y)[zu]}) − [α2(z){f (x)yu}] − [α2(z){xf (y)u}] − [α2(z){xyf (u)}] + [α2(z)f ({xyu})] − [{f (x)yz}α2(u)] − [{xf (y)z}α2(u)] − [{xyf (z)}α2(u)] + [f ({xyz})α2(u)] =0 5 and II δ1 δ2 II f (x, y, u, v, w) ={α2(x)α2(y)δ1 IIf (u, v, w)} − {δ1 IIf (x, y, u)α2(v)α2(w)} − {α2(u)δ1 II f (x, y, v)α2(w)} − {α2(u)α2(v)δ1 − δ1 II f (x, y, w)} + δ1 II f (α2(u), {xyv}, α2(w)) − δ1 II f (α2(x), α2(y), {uvw}) − δ1 IIf (α2(u), α2(v), {xyw}) II f ({xyu}, α2(v), α2(w)) ={α2(x)α2(y){f (u)vw}} + {α2(x)α2(y){uf (v)w}} + {α2(x)α2(y){uvf (w)}} − {α2(x)α2(y)f ({uvw})} − {{f (x)yu}α2(v)α2(w)} − {{xf (y)u}α2(v)α2(w)} − {{xyf (u)}α2(v)α2(w)} + {f ({xyu})α2(v)α2(w)} − {α2(u){f (x)yv}α2(w)} − {α2(u){xf (y)v}α2(w)} − {α2(u){xyf (v)}α2(w)} + {α2(u)f ({xyv})α2(w)} − {α2(u)α2(v){f (x)yw}} − {α2(u)α2(v){xf (y)w}} − {α2(u)α2(v){xyf (w)}} + {α2(u)α2(v)f ({xyw})} + {f α2(x)α2(y){uvw}} + {α2(x)f α2(y){uvw}} + {α2(x)α2(y)f ({uvw})} − f ({α2(x)α2(y){uvw}}) − {f ({xyu})α2(v)α2(w)} − {{xyu}f α2(v)α2(w)} − {{xyu}α2(v)f α2(w)} + f ({{xyu}α2(v)α2(w)}) − {f α2(u){xyv}α2(w)} − {α2(u)f ({xyv})α2(w)} − {α2(u){xyv}f α2(w)} + f ({α2(u){xyv}α2(w)}) − {f α2(u)α2(v){xyw}} − {α2(u)f α2(v){xyw}} − {α2(u)α2(v)f ({xyw})} + f ({α2(u)α2(v){xyw}}) =0. Moreover, by (1.5) and (1.6), one obtains d2 I δ1 I f (x, y, z) = (cid:9)x,y,z ([δ1 I f (x, y)α(z)] + δ1 I f ([xy], α(z)) + δ1 II f (x, y, z)) = (cid:9)x,y,z ([[xf (y)]α(z)] + [[f (x)y]α(z)] − [f ([xy])α(z)]) + (cid:9)x,y,z ([[xy]f α(z)] + [f ([xy])α(z)] − f ([[xy]α(z)])) + (cid:9)x,y,z ({f (x)yz} + {xf (y)z} + {xyf (z)} − f ({xyz})) =0 and d2 IIδ1 II f (x, y, z, u) = (cid:9)x,y,z ({δ1 = (cid:9)x,y,z ({[xf (y)]α(z)α(u)} + {[f (x)y]α(z)α(u)} − {f ([xy])α(z)α(u)}) I f (x, y)α(z)α(u)} + δ1 II f ([xy], α(z), α(u))) + (cid:9)x,y,z ({f ([xy])α(z)α(u)} + {[xy]f α(z)α(u)} + {[xy]α(z)f α(u)}−f ({[xy]α(z)α(u)})) ={[xf (y)]α(z)α(u)} + {[yf (z)]α(x)α(u)} + {[zf (x)]α(y)α(u)} + {[f (x)y]α(z)α(u)} + {[f (y)z]α(x)α(u)} + {[f (z)x]α(y)α(u)} + {[xy]f α(z)α(u)} + {[yz]f α(x)α(u)} + {[zx]f α(y)α(u)} =0. For (f, g) ∈ HomC 2(L, L) × HomC 3(L, L), δ3 I δ2 I f (x1, · · · , x6) 6 ={α3(x1)α3(x2)δ2 I f (x3, · · · , x6)} − {α3(x3)α3(x4)δ2 I f (x1, x2, x5, x6)} (−1)kδ2 I f (α2(x1), · · · , \ α2(x2k−1), \ α2(x2k), · · · , {x2k−1x2kxi}, · · · , α2(x6)) + 2Xk=1 6Xi=2k+1 I g(x1, · · · , x5)α4(x6)] I g(x1, · · · , x4, x6)] + [δ2 I g(α(x1), · · · , α(x4), [x5x6]) + [α4(x5)δ2 −δ2 ={α3(x1)α3(x2){α(x3)α(x4)f (x5, x6)}} − {α3(x3)α3(x4){α(x1)α(x2)f (x5, x6)}} −{{α(x1)α(x2)α(x3)}α3(x4)α2f (x5, x6)} − {α3(x3){α(x1)α(x2)α(x4)}α2f (x5, x6)} +f ({{x1x2x3}α2(x4)α2(x5)}, α4(x6)) + f ({α2(x3){x1x2x4}α2(x5)}, α4(x6)) +f ({α2(x3)α2(x4){x1x2x5}}, α4(x6)) − f ({α2(x1)α2(x2){x3x4x5}}, α4(x6)) +f (α4(x5), {{x1x2x3}α2(x4)α2(x6)}) + f (α4(x5), {α2(x3){x1x2x4}α2(x6)}) +f (α4(x5), {α2(x3)α2(x4){x1x2x6}}) − f (α4(x5), {α2(x1)α2(x2){x3x4x6}}) −{α3(x1)α3(x2)[α2(x5)g(x3, x4, x6)]} + [{α2(x1)α2(x2)α2(x5)}α2g(x3, x4, x6)] +[α4(x5){α2(x1)α2(x2)g(x3, x4, x6)}] − {α3(x1)α3(x2)[g(x3, x4, x5)α2(x6)]} +[α2g(x3, x4, x5){α2(x1)α2(x2)α2(x6)}] + [{α2(x1)α2(x2)g(x3, x4, x5)}α4(x6)] +{α3(x3)α3(x4)[α2(x5)g(x1, x2, x6)]} − [{α2(x3)α2(x4)α2(x5)}α2g(x1, x2, x6)] −[α4(x5){α2(x3)α2(x4)g(x1, x2, x6)}] + {α3(x3)α3(x4)[g(x1, x2, x5)α2(x6)]} −[α2g(x1, x2, x5){α2(x3)α2(x4)α2(x6)}] − [{α2(x3)α2(x4)g(x1, x2, x5)}α4(x6)] −g(α3(x3), α3(x4), [{x1x2x5}α2(x6)]) − g(α3(x3), α3(x4), [α2(x5){x1x2x6}]) +g(α3(x3), α3(x4), {α(x1)α(x2)[x5x6]}) + g(α3(x1), α3(x2), [{x3x4x5}α2(x6)]) +g(α3(x1), α3(x2), [α2(x5){x3x4x6}]) − g(α3(x1), α3(x2), {α(x3)α(x4)[x5x6]}) +{αg(x1, x2, x3)α3(x4)[α2(x5)α2(x6)]} − [α4(x5){g(x1, x2, x3)α2(x4)α2(x6)}] −[{g(x1, x2, x3)α2(x4)α2(x5)}α4(x6)] + {α3(x3)αg(x1, x2, x4)[α2(x5)α2(x6)]} −[α4(x5){α2(x3)g(x1, x2, x4)α2(x6)}] − [{α2(x3)g(x1, x2, x4)α2(x5)}α4(x6)] =0 and δ3 II δ2 II g(x1, · · · , x7) = + 3Xk=1 3Xk=1 7Xi=2k+1 (−1)k+1{α4(x2k−1)α4(x2k)δ2 II g(x1, · · · , [x2k−1,cx2k, · · · , x7) (−1)kδ2 IIg(α2(x1), · · · , \α2(x2k−1), \α2(x2k), · · · , {x2k−1x2kxi}, · · · , α2(x7)) IIg(x1, · · · , x4, x6)α4(x5)α4(x7)} IIg(x1, · · · , x5)α4(x6)α4(x7)} − {δ2 +{δ2 ={α4(x1)α4(x2){α2(x3)α2(x4)g(x5,x6,x7)}}−{α4(x3)α4(x4){α2(x1)α2(x2)g(x5,x6,x7)}} −{α2({x1x2x3})α4(x4)α2g(x5, x6, x7)} − {α4(x3)α2({x1x2x4})α2g(x5, x6, x7)} −{α4(x1)α4(x2){g(x3, x4, x5)α2(x6)α2(x7)}} + {α2g(x3, x4, x5)α2({x1x2x6})α4(x7)} +{α2g(x3, x4, x5)α4(x6)α2({x1x2x7})} + {{α2(x1)α2(x2)g(x3, x4, x5)}α4(x6)α4(x7)} −{α4(x1)α4(x2){α2(x5)g(x3, x4, x6)α2(x7)}} + {α2({x1x2x5})α2g(x3, x4, x6)α4(x7)} 7 +{α4(x5)α2g(x3, x4, x6)α2({x1x2x7})} + {α4(x5){α2(x1)α2(x2)g(x3, x4, x6)}α4(x7)} −{α4(x1)α4(x2){α2(x5)α2(x6)g(x3,x4,x7)}}+{α4(x5)α4(x6){α2(x1)α2(x2)g(x3,x4,x7)}} +{α2({x1x2x5})α4(x6)α2g(x3, x4, x7)} + {α4(x5)α2({x1x2x6})α2g(x3, x4, x7)} +{α4(x3)α4(x4){g(x1, x2, x5)α2(x6)α2(x7)}} − {α2g(x1, x2, x5)α2({x3x4x6})α4(x7)} −{α2g(x1, x2, x5)α4(x6)α2({x3x4x7})} − {{α2(x3)α2(x4)g(x1, x2, x5)}α4(x6)α4(x7)} +{α4(x3)α4(x4){α2(x5)g(x1, x2, x6)α2(x7)}} − {α2({x3x4x5})g(x1, x2, x6)α4(x7)} −{α4(x5)α2g(x1, x2, x6)α2({x3x4x7})} − {α4(x5){α2(x3)α2(x4)g(x1, x2, x6)}α4(x7)} +{α4(x3)α4(x4){α2(x5)α2(x6)g(x1,x2,x7)}}−{α4(x5)α4(x6){α2(x3)α2(x4)g(x1,x2,x7)}} −{α2({x3x4x5})α4(x6)α2g(x1, x2, x7)} − {α4(x5)α2({x3x4x6})α2g(x1, x2, x7)} −{α4(x5)α4(x6){g(x1, x2, x3)α2(x4)α2(x7)}} + {α2g(x1, x2, x3)α4(x4)α2({x5x6x7})} −{{g(x1,x2,x3)α2(x4)α2(x5)}α4(x6)α4(x7)}−{{α4(x5){g(x1,x2,x3)α2(x4)α2(x6)}α4(x7)} −{α4(x5)α4(x6){α2(x3)g(x1, x2, x4)α2(x7)}} + {α4(x3)α2g(x1, x2, x4)α2({x5x6x7})} −{{α2(x3)g(x1,x2,x4)α2(x5)}α4(x6)α4(x7)}−{α4(x5){α2(x3)g(x1,x2,x4)α2(x6)}α4(x7)} +g({{x1x2x3}α2(x4)α2(x5)},α4(x6),α4(x7))+g({α2(x3){x1x2x4}α2(x5)},α4(x6),α4(x7)) +g({α2(x3)α2(x4){x1x2x5}},α4(x6),α4(x7))−g({α2(x1)α2(x2){x3x4x5}},α4(x6),α4(x7)) +g({α4(x5),{{x1x2x3}α2(x4)α2(x6)},α4(x7))+g({α4(x5),{α2(x3){x1x2x4}α2(x6)},α4(x7)) +g({α4(x5),{α2(x3)α2(x4){x1x2x6}},α4(x7))−g({α4(x5),{α2(x1)α2(x2){x3x4x6}},α4(x7)) +g(α4(x5), α4(x6), {{x1x2x3}α2(x4)α2(x7)}) + g(α4(x5), α4(x6), {α2(x3){x1x2x4}α2(x7)}) +g(α4(x5), α4(x6), {α2(x3)α2(x4){x1x2x7}}) − g(α4(x5), α4(x6), {α2(x1)α2(x2){x3x4x7}}) −g(α4(x3), α4(x4), {{x1x2x5}α2(x6)α2(x7)}) − g(α4(x3), α4(x4), {α2(x5){x1x2x6}α2(x7)}) −g(α4(x3), α4(x4), {α2(x5)α2(x6){x1x2x7}}) + g(α4(x3), α4(x4), {α2(x1)α2(x2){x5x6x7}}) +g(α4(x1), α4(x2), {{x3x4x5}α2(x6)α2(x7)}) + g(α4(x1), α4(x2), {α2(x5){x3x4x6}α2(x7)}) +g(α4(x1), α4(x2), {α2(x5)α2(x6){x3x4x7}}) − g(α4(x1), α4(x2), {α2(x3)α2(x4){x5x6x7}}) =0, where the items that could be canceled in pairs are omitted. The proof is completed. Define HomZ1(L,L)×HomZ1(L,L) = {(f,f ) ∈ HomC1(L,L)×HomC1(L,L) (δ1 I ,δ1 II)(f,f ) = (0,0)}, HomZ 2(L, L) × HomZ 3(L, L) ={(f, g) ∈ HomC 2(L, L) × HomC 3(L, L) (δ2 I , δ2 II)(f, g) = (d2 I, d2 II)(f, g) = (0, 0)}, HomZ 4(L, L) × HomZ 5(L, L) ={(f, g) ∈ HomC 4(L, L) × HomC 5(L, L) (δ3 I , δ3 II)(f, g) = (0, 0)}, HomB2(L, L) × HomB3(L, L) = {(δ1 I , δ1 II)(f, f ) f ∈ HomC 1(L, L)}, 8 HomB4(L, L)×HomB5(L, L) = {(δ2 I , δ2 II)(f, g) (f, g) ∈ HomC 2(L, L)×HomC 3(L, L)}. Then by Theorem 2.4, HomB2(L, L) × HomB3(L, L) ⊆ HomZ 2(L, L) × HomZ 3(L, L), HomB4(L, L) × HomB5(L, L) ⊆ HomZ 4(L, L) × HomZ 5(L, L). So one could define HomH 1(L, L) × HomH 1(L, L) = HomZ 1(L, L) × HomZ 1(L, L), HomH 2(L, L) × HomH 3(L, L) = HomZ 2(L, L) × HomZ 3(L, L) HomB2(L, L) × HomB3(L, L) and HomH 4(L, L) × HomH 5(L, L) = HomZ 4(L, L) × HomZ 5(L, L) HomB4(L, L) × HomB5(L, L) as the first, second and third cohomology groups of (L, [·, ·], {·, ·, ·}, α), respectively. Definition 2.5. A linear map D : L → L is called the αk-derivation of (L, [·, ·], {·, ·, ·}, α), if D satisfies D ◦ α = α ◦ D and D([xy]) = [αk(x)D(y)] + [D(x)αk(y)], D({xyz}) = {D(x)αk(y)αk(z)} + {αk(x)D(y)αk(z)} + {αk(x)αk(y)D(z)}, k {z } where αk = α ◦ · · · ◦ α and α0 = idL. It is straightforward to show that D is an α0- derivation of (L, [·, ·], {·, ·, ·}, α) if and only if (D, D) ∈ HomH 1(L, L) × HomH 1(L, L). Denote by Derαk(L) the set of all αk-derivations of (L, [·, ·], {·, ·, ·}, α). Theorem 2.6. Set Der(L) =Lk≥0 Derαk (L). Then Der(L) is a Lie algebra. Proof. It is sufficient to prove [Derαk (L), Derαs(L)] ⊆ Derαk+s(L). Now suppose that D ∈ Derαk (L) and D′ ∈ Derαs(L). Then [D, D′] ◦ α = D ◦ D′ ◦ α − D′ ◦ D ◦ α = α ◦ (D ◦ D′ − D′ ◦ D) = α ◦ [D, D′]. Note that [D, D′]([xy]) =D([αs(x)D′(y)] + [D′(x)αs(y)]) − D′([αk(x)D(y)] + [D(x)αk(y)]) =[Dαs(x)αkD′(y)] + [αk+s(x)DD′(y)] + [DD′(x)αk+s(y)] + [αkD′(x)Dαs(y)] −[D′αk(x)αsD(y)]−[αk+s(x)D′D(y)]−[D′D(x)αk+s(y)]−[αsD(x)D′αk(y)] =[[D, D′](x)αk+s(y)] + [αk+s(x)[D, D′](y)], 9 and [D, D′]({xyz}) =D({D′(x)αs(y)αs(z)} + {αs(x)D′(y)αs(z)} + {αs(x)αs(y)D′(z)}) − D′({D(x)αk(y)αk(z)} + {αk(x)D(y)αk(z)} + {αk(x)αk(y)D(z)}) ={DD′(x)αk+s(y)αk+s(z)} + {αkD′(x)Dαs(y)αk+s(z)} + {αkD′(x)αk+s(y)Dαs(z)} + {Dαs(x)αkD′(y)αk+s(z)} + {αk+s(x)DD′(y)αk+s(z)} + {αk+s(x)αkD′(y)Dαs(z)} + {Dαs(x)αk+s(y)αkD′(z)} + {αk+s(x)Dαs(y)αkD′(z)} + {αk+s(x)αk+s(y)DD′(z)} − {D′D(x)αk+s(y)αk+s(z)} − {αsD(x)D′αk(y)αk+s(z)} − {αsD(x)αk+s(y)D′αk(z)} − {D′αk(x)αsD(y)αk+s(z)} − {αk+s(x)D′D(y)αk+s(z)} − {αk+s(x)αsD(y)D′αk(z)} − {D′αk(x)αk+s(y)αsD(z)} − {αk+s(x)D′αk(y)αsD(z)} − {αk+s(x)αk+s(y)D′D(z)} ={[D, D′](x)αk+s(y)αk+s(z)} +{αk+s(x)[D, D′](y)αk+s(z)} +{αk+s(x)αk+s(y)[D, D′](z)}. It follows that [D, D′] ∈ Derαk+s(L). 3 Deformations of a Hom-Lie-Yamaguti algebra Suppose that (L, [·, ·], {·, ·, ·}, α) is a HLYA over K. Let K[[t]] be the ring of formal power series over K and L[[t]] be the set of formal power series over L. Then for a K- bilinear map f : L × L → L and a K-trilinear map g : L × L × L → L, it is natural to extend them to be a K[[t]]-bilinear map f : L[[t]] × L[[t]] → L[[t]] and a K[[t]]-trilinear map g : L[[t]] × L[[t]] × L[[t]] → L[[t]] by f Xi≥0 xiti,Xj≥0 xiti,Xj≥0 yjtj,Xk≥0 yjtj! = Xi,j≥0 zktk! = Xi,j,k≥0 g Xi≥0 f (xi, yj)ti+j, g(xi, yj, zk)ti+j+k. Definition 3.1. Suppose that (L, [·, ·], {·, ·, ·}, α) is a HLYA over K. A one-parameter formal deformation of (L, [·, ·], {·, ·, ·}, α) is a pair of formal power series (ft, gt) of the form ft = [·, ·] +Xi≥1 fiti, gt = {·, ·, ·} +Xi≥1 giti, where each fi : L × L → L is a K-bilinear map (extended to be K[[t]]-bilinear) and each gi : L × L × L → L is a K-trilinear map (extended to be K[[t]]-trilinear) such that (L[[t]], ft, gt, α) is a HLYA over K[[t]]. Set f0 = [·, ·] and g0 = {·, ·, ·}, then ft and gt can be written as ft =Pi≥0 fiti and gt =Pi≥0 giti, respectively. 10 Note that (L[[t]], ft, gt, α) is required to be a HLYA. Then the following equations must be satisfied: α ◦ ft(x, y) = ft(α(x), α(y)), α ◦ gt(x, y, z) = gt(α(x), α(y), α(z)), ft(x, x) = 0, gt(x, x, y) = 0, (cid:9)x,y,z (ft(ft(x, y), α(z)) + gt(x, y, z)) = 0, (cid:9)x,y,z gt(ft(x, y), α(z), α(u)) = 0, gt(α(x), α(y), ft(z, u)) = ft(gt(x, y, z), α2(u)) + ft(α2(z), gt(x, y, u)), (3.1) (3.2) (3.3) (3.4) (3.5) (3.6) (3.7) gt(α2(u), α2(v), gt(x, y, z)) =gt(gt(u, v, x), α2(y), α2(z)) + gt(α2(x), gt(u, v, y), α2(z)) + gt(α2(x), α2(y), gt(u, v, z)). Equations (3.1)-(3.8) are equivalent to α ◦ fn(x, y) = fn(α(x), α(y)), α ◦ gn(x, y, z) = gn(α(x), α(y), α(z)), (3.8) (3.1′) (3.2′) (3.3′) (3.4′) (3.5′) (3.6′) fn(x, x) = 0, gn(x, x, y) = 0, (cid:9)x,y,z Xi+j=n (cid:9)x,y,z Xi+j=n fi(fj(x, y), α(z)) + gn(x, y, z)! = 0, gi(fj(x, y), α(z), α(u)) = 0, Xi+j=n Xi+j=n gi(α(x), α(y), fj(z, u)) = Xi+j=n(cid:16)fi(gj(x, y, z), α2(u)) + fi(α2(z), gj(x, y, u))(cid:17), (3.7′) gi(α2(u), α2(v), gj(x, y, z)) = Xi+j=n(cid:16)gi(gj(u, v, x), α2(y), α2(z)) +gi(α2(x), gj(u, v, y), α2(z))+gi(α2(x), α2(y), gj(u, v, z))(cid:17), (3.8′) respectively. These equations are called the deformation equations of a HLYA. Equa- tions (3.1′)-(3.4′) imply that (fi, gi) ∈ HomC 2(L, L) × HomC 3(L, L). Let n = 1 in (3.5′)-(3.8′). Then 0 =(cid:9)x,y,z ([f1(x, y)α(z)] + f1([xy], α(z)) + g1(x, y, z)) , 0 =(cid:9)x,y,z ({f1(x, y)α(z)α(u)} + g1([xy], α(z), α(u))) , 0 ={α(x)α(y)f1(z, u)} + g1(α(x), α(y), [zu]) − [g1(x, y, z)α2(u)] − f1({xyz}, α2(u)) − [α2(z)g1(x, y, u)] − f1(α2(z), {xyu}), 11 0 ={α2(u)α2(v)g1(x, y, z)} − {g1(u, v, x)α2(y)α2(z)} − {α2(x)g1(u, v, y)α2(z)} − {α2(x)α2(y)g1(u, v, z)} + g1(α2(u), α2(v), {xyz}) − g1({uvx}, α2(y), α2(z)) − g1(α2(x), {uvy}, α2(z)) − g1(α2(x), α2(y), {uvz}), which imply (δ2 I , δ2 II)(f1, g1) = (d2 I , d2 II)(f1, g1) = (0, 0), i.e., (f1, g1) ∈ HomZ 2(L, L) × HomZ 3(L, L). The first order term (f1, g1) is called the infinitesimal of (ft, gt). Definition 3.2. Suppose that (ft, gt) and (f ′ mations of (L, [·, ·], {·, ·, ·}, α). They are called equivalent, denoted by (ft, gt) ∼ (f ′ there is a linear isomorphism of HLYA Φt =Pi≥0 φiti : (L[[t]], ft, gt, α) → (L[[t]], f ′ such that t) are two one-parameter formal defor- t), if t, α) t, g′ t, g′ t, g′ φ0 = idL, Φt ◦ α = α ◦ Φt, Φt ◦ ft(x, y) = f ′ t(Φt(x), Φt(y)), Φt ◦ gt(x, y, z) = g′ t(Φt(x), Φt(y), Φt(z)). In the case (f1, g1) = (f2, g2) = · · · = (0, 0), (ft, gt) = (f0, g0) is called the null deforma- tion. A one-parameter formal deformation (ft, gt) is said to be trivial if (ft, gt) ∼ (f0, g0). A HLYA (L, [·, ·], {·, ·, ·}, α) is called analytically rigid if every one-parameter formal de- formation (ft, gt) is trivial. Theorem 3.3. Let (ft, gt) and (f ′ of (L, [·, ·], {·, ·, ·}, α). Then (f1, g1) and (f ′ HomH 2(L, L) × HomH 3(L, L). t, g′ t) be equivalent one-parameter formal deformations 1) belong to the same cohomology class in 1, g′ Proof. It is sufficient to prove (f1 − f ′ Suppose that Φt =Pi≥0 φiti : (L[[t]], ft, gt, α) → (L[[t]], f ′ 1, g1 − g′ that 1) ∈ HomB2(L, L) × HomB3(L, L). t, α) is an isomorphism such t , g′ φ0 = idL, Φt ◦ α = α ◦ Φt, Φt ◦ ft(x, y) = f ′ t(Φt(x), Φt(y)), Φt ◦ gt(x, y, z) = g′ t(Φt(x), Φt(y), Φt(z)). Then φ1 ∈ HomC 1(L, L) and Xi≥0 φiti Xj≥0 φiti Xj≥0 gj(x, y, z)tj! =Xi≥0 fj(x, y)tj! =Xi≥0 i Xk≥0 i Xk≥0 φk(x)tk,Xl≥0 f ′ g′ φk(x)tk,Xl≥0 φl(y)tl! , φl(y)tl,Xm≥0 φm(z)tm! . Xi≥0 Hence f1(x, y) + φ1([xy]) = [xφ1(y)] + [φ1(x)y] + f ′ 1(x, y), g1(x, y, z) + φ1({xyz}) = {φ1(x)yz} + {xφ1(y)z} + {xyφ1(z)} + g′ 1(x, y, z). Therefore, (f1 − f ′ 1, g1 − g′ 1) = (δ1 I , δ1 II)(φ1, φ1) ∈ HomB2(L, L) × HomB3(L, L). 12 Theorem 3.4. Suppose that (L, [·, ·], {·, ·, ·}, α) is a HLYA. Then (L, [·, ·], {·, ·, ·}, α) is analytically rigid, if HomH 2(L, L) × HomH 3(L, L) = 0. Proof. Let (ft, gt) be a one-parameter formal deformation of (L, [·, ·], {·, ·, ·}, α). Sup- pose ft = f0 +Pi≥r fiti and gt = g0 +Pi≥r giti. Set n = r in (3.5′)-(3.8′). It follows that (fr, gr) ∈ HomZ 2(L, L) × HomZ 3(L, L) = HomB2(L, L) × HomB3(L, L). Then there exists hr ∈ HomC 1(L, L) such that (fr, gr) = (δ1 I hr, δ1 II hr). Consider Φt = idL−hrtr. Then Φt : L → L is a linear isomorphism and Φt◦α = α◦Φt. Let t(x, y) = Φ−1 f ′ t ft(Φt(x), Φt(y)), t(x, y, z) = Φ−1 g′ t gt(Φt(x), Φt(y), Φt(z)). Assume that f ′ i ti and use the fact Φtf ′ t(x, y) = ft(Φt(x), Φt(y)). Then f ′ i (x, y)ti = f0 +Xi≥r fiti! (x − hr(x)tr, y − hr(y)tr), t =Pi≥0 f ′ (idL − hrtr)Xi≥0 i (x, y)ti −Xi≥0 fi(x, y)ti −Xi≥r that is, f ′ Xi≥0 +Xi≥r hr ◦ f ′ i (x, y)ti+r =f0(x, y) − f0(hr(x), y)tr − f0(x, hr(y))tr + f0(hr(x), hr(y))t2r fi(hr(x), y)ti+r −Xi≥r fi(x, hr(y))ti+r +Xi≥r fi(hr(x), hr(y))ti+2r. So f ′ 0(x, y) = f0(x, y) = [xy], f ′ 1(x, y) = · · · = f ′ r−1(x, y) = 0 and f ′ r(x, y) − hr([xy]) = −[hr(x)y] − [xhr(y)] + fr(x, y). I hr(x, y) + fr(x, y) = 0, which implies f ′ r(x, y) = −δ1 Hence f ′ same way, we have g′ formal deformation of (L, [·, ·], {·, ·, ·}, α) and (ft, gt) ∼ (f ′ (ft, gt) ∼ (f0, g0). Therefore, (L, [·, ·], {·, ·, ·}, α) is analytically rigid. t = {·, ·, ·} +Pi≥r+1 g′ iti. It is clear that (f ′ t, g′ t = [·, ·] +Pi≥r+1 f ′ i ti. In the t) is a one-parameter t). By induction, one gets t, g′ In the deformation theory of other algebraic structures, the obstructions are in the same cohomology theory as the infinitesimal deformations but one dimension higher. But the usual approach doesn't work for HLYAs: Let (f0, g0) = ([·, ·], {·, ·, ·}) and (f1, g1) ∈ HomZ 2(L, L) × HomZ 3(L, L). Then (f0, g0), (f1, g1) satisfy the deformation equations (3.1′)-(3.8′) for n = 1. Set F (x, y, z, u) = f1(g1(x, y, z), α2(u)) + f1(α2(z), g1(x, y, u)) − g1(α(x), α(y), f1(z, u)), G(u, v, x, y, z) =g1(g1(u, v, x), α2(y), α2(z)) + g1(α2(x), g1(u, v, y), α2(z)) + g1(α2(x), α2(y), g1(u, v, z)) − g1(α2(u), α2(v), g1(x, y, z)). 13 Then (F, G) ∈ HomZ 4(L, L)×HomZ 5(L, L). If HomH 4(L, L)×HomH 5(L, L) = 0, then there exists a pair (f2, g2) ∈ HomC 2(L, L) × HomC 3(L, L) such that (δ2 II)(f2, g2) = (−F, −G). Based on one's experience in other algebras, (f0, g0), (f1, g1) and (f2, g2) would satisfy the deformation equations (3.1′)-(3.8′) for n = 2. Note that (3.1′)-(3.4′) clearly hold since (fi, gi) ∈ HomC 2(L, L) × HomC 3(L, L) and it is straightforward to verify (f0, g0), (f1, g1) and (f2, g2) satisfying (3.7′) and (3.8′) for n = 2. However, one could not prove that (f0, g0), (f1, g1) and (f2, g2) satisfy (3.5′) or (3.6′) for n = 2. Therefore, the obstructions of a HYLA involve other cohomology theory instead of the one carried over from Lie-Yamaguti algebras in [25], directly. I , δ2 ACKNOWLEDGEMENTS The authors would like to thank the referee for valu- able comments and suggestions on this article. References [1] F. Ammar, Z. Ejbehi and A. Makhlouf, Cohomology and deformations of Hom- algebras. J. Lie Theory 21 (2011), no. 4, 813 -- 836. [2] F. Ammar, S. Mabrouk and A. Makhlouf, Representations and cohomology of n-ary multiplicative Hom-Nambu-Lie algebras. J. Geom. Phys. 61 (2011), no. 10, 1898 -- 1913. [3] S. Benayadi and A. Makhlouf, Hom-Lie algebras with invariant nondegenerate bilin- ear forms. J. Geom. Phys., 76 (2014), no. 2, 38 -- 60. [4] M. Elhamdadi and A. Makhlouf, Deformations of Hom-alternative and Hom-Malcev algebras. Algebras Groups Geom. 28 (2011), no. 2, 117 -- 145. [5] M. Flato, M. Gerstenhaber and A. Voronov, Cohomology and deformation of Leibniz pairs. Lett. Math. Phys. 34 (1995), no. 1, 77 -- 90. [6] D. Gaparayi and A. Nourou Issa, A twisted generalization of Lie-Yamaguti algebras. Int. J. Algebra 6 (2012), no. 5 -- 8, 339 -- 352. [7] D. Gaparayi and A. Nourou Issa, Hom-Lie-Yamaguti structures on Hom-Leibniz al- gebras. ArXiv: 1208.6038 (2012). [8] M. Gerstenhaber, On the deformation of rings and algebras. Ann. of Math. (2) 79 (1964), 59 -- 103. [9] M. Gerstenhaber, On the deformation of rings and algebras.II. Ann. of Math. 84 (1966), 1 -- 19. [10] M. Gerstenhaber, On the deformation of rings and algebras.III. Ann. of Math. (2) 88 (1968), 1 -- 34. 14 [11] M. Gerstenhaber, On the deformation of rings and algebras.IV. Ann. of Math. (2) 99 (1974), 257 -- 276. [12] J. Hartwig, D. Larsson and S. Silvestrov, Deformations of Lie algebras using σ- derivations. J. Algebra 295 (2006), no. 2, 314 -- 361. [13] N. Hu, Q-Witt algebras, q-Lie algebras, q-holomorph structure and representations. Algebra Colloq. 6 (1999), no. 1, 51 -- 70. [14] K. Kodaira and D. Spencer, On deformations of complex analytic structures. I, II. Ann. of Math. (2) 67 (1958), 328 -- 466. [15] F. Kubo and Y. Taniguchi, A controlling cohomology of the deformation theory of Lie triple systems. J. Algebra 278 (2004), no. 1, 242 -- 250. [16] D. Larsson and S. Silvestrov, Quasi-hom-Lie algebras, central extensions and 2- cocycle-like identities. J. Algebra 288 (2005), no. 2, 321 -- 344. [17] J. Lin, L. Chen and Y. Ma, On the Deformation of Lie-Yamaguti algebras. Acta Math. Sin. (Engl. Ser.) (Accepted). [18] Y. Liu, L. Chen and Y. Ma, Hom-Nijienhuis operators and T*-extensions of hom-Lie superalgebras. Linear Algebra Appl. 439 (2013), no. 7, 2131 -- 2144. [19] A. Makhlouf and S. Silvestrov, Notes on 1-parameter formal deformations of Hom- associative and Hom-Lie algebras. Forum Math. 22 (2010), no. 4, 715 -- 739. [20] Y. Ma, L. Chen and J. Lin, Cohomology and 1-parameter formal deformations of Hom-Lie triple systems. arXiv:1309.3347 (2013). [21] A. Nijenhuis and R. Richardson, Cohomology and deformations in graded Lie alge- bras. Bull. Amer. Math. Soc. 72 (1966), 1 -- 29. [22] K. Nomizu, Invariant affine connections on homogeneous spaces. Amer. J. Math. 76 (1954), 33 -- 65. [23] Y. Sheng, Representations of hom-Lie algebras. Algebr. Represent. Theory 15 (2012), no. 6, 1081 -- 1098. [24] Y. Sheng and D. Chen, Hom-Lie 2-algebras. J. Algebra 376 (2013), 174 -- 195. [25] K. Yamaguti, On cohomology groups of general Lie triple systems. Kumamoto J. Sci. Ser. A 8 (1967/1969), 135 -- 146. [26] K. Yamaguti, On the Lie triple system and its generalization. J. Sci. Hiroshima Univ. Ser. A 21 (1957/1958), 155 -- 160. [27] D. Yau, Hom-algebras and homology. J. Lie Theory 19 (2009), no. 2, 409 -- 421. [28] D. Yau, On n-ary Hom-Nambu and Hom-Nambu-Lie algebras. J. Geom. Phys. 62 (2012), no. 2, 506 -- 522. 15
1404.2855	4	1404	2015-04-23T13:58:42	On certain modules of covariants in exterior algebras	[ "math.RA", "math.RT" ]	We study the structure of the space of covariants $B:=\left(\bigwedge (\mathfrak g/\mathfrak k)^\otimes \mathfrak g\right)^{\mathfrak k},$ for a certain class of infinitesimal symmetric spaces $(\mathfrak g,\mathfrak k)$ such that the space of invariants $A:=\left(\bigwedge (\mathfrak g/\mathfrak k)^\right)^{\mathfrak k}$ is an exterior algebra $\wedge (x_1,...,x_r),$ with $r=rk(\mathfrak g)-rk(\mathfrak k)$. We prove that they are free modules over the subalgebra $A_{r-1}=\wedge (x_1,...,x_{r-1})$ of rank $4r$. In addition we will give an explicit basis of $B$. As particular cases we will recover same classical results. In fact we will describe the structure of $\left(\bigwedge (M_n^{\pm})^*\otimes M_n\right)^G$, the space of the $G-$equivariant matrix valued alternating multilinear maps on the space of (skew-symmetric or symmetric with respect to a specific involution) matrices, where $G$ is the symplectic group or the odd orthogonal group. Furthermore we prove new polynomial trace identities.	math.RA	math	On certain modules of covariants in exterior algebras Salvatore Dolce∗ September 9, 2018 Abstract We study the structure of the space of covariants B := (V(g/k)∗ ⊗ g)k , for a certain class of infinitesimal symmetric spaces (g, k) such that the space of invariants A := (V(g/k)∗)k is an exterior algebra ∧(x1, ..., xr), with r = rk(g) − rk(k). We prove that they are free modules over the subalgebra Ar−1 = ∧(x1, ..., xr−1) of rank 4r. In addition we will give an explicit basis of B. As particular cases we will recover same classical results. In fact we will describe the structure of (cid:0)V(M ± , the space of the G−equivariant matrix valued alternating multilinear maps on the space of (skew-symmetric or symmetric with respect to a spe- cific involution) matrices, where G is the symplectic group or the odd orthogonal group. Furthermore we prove new polynomial trace identities. ⊗ Mn(cid:1)G n )∗ 4 v 5 5 8 2 . 4 0 4 1 : v i X r a Introduction In this paper we study the ring of invariant skew symmetric multilinear functions on a linear representation of an algebraic group G, that is from a geometric viewpoint constant coefficient invariant differential forms. It is a classical fact that these forms give the cohomology of compact Lie groups and more generally compact symmetric spaces. (See for example [3], [7], [8]) In the first part of the paper we are mostly concerned with classical groups. Let G = GL(n) be the group of invertible n × n complex matrices. G acts on the space Mn of complex n × n matrices by conjugation. We first study, essentially following Procesi [22], the algebra of invariant skew symmetric multilinear functions on Mn. This turns out to be closely related to the theory of rings with Polynomial Identities. A fundamental role for our purposes will be played by the the space of G-equivariant multi- linear alternating matrix valued maps on the m−tuples of matrices. We will endow this space with a natural structure of algebra, by defining a suitable skew- symmetric product. It will be worthwhile to consider this algebra as a module on the algebra of invariants. Bresar, Procesi and Spenko [6] have shown, in the case of the linear group GL(n), that this algebra is a free module on a certain subalgebra of invariants, with a natural explicit basis. Section 1 is devoted to recollect these facts. ∗Dipartimento di Matematica, Universit`a La Sapienza di Roma, P.le Aldo Moro 5, 00185 Rome, Italy; E-mail address: [email protected]; 2010 Mathematics Subject Classification 17B20; Key words and phrases. Invariant theory, symmetric spaces, exterior algebras, polynomial trace identities. 1 Most of our paper is devoted to extend this and related results to the case of certain symmetric spaces. Starting with GL(n) (or better SL(n)) consider the orthogonal (respectively, when n is even, symplectic) involution σ and denote by SO(n) (respectively Sp(n)) the special orthogonal (respectively symplectic) groups of elements in SL(n) fixed by σ. σ induces a linear involution on the space Mn which decomposes as the direct sum M + n of the +1 and −1 eigenspaces. In section 2 we give a precise description of the rings of invariant skew symmetric multilinear n and of the ring of Mn valued invariant skew symmetric multilinear maps on M ± n maps on M ± at least in the case of the symplectic involution and of the orthogonal involution for n odd. n ⊕ M − We will prove that, similarly to the linear group, the space of the G-equivariant matrix valued alternating multilinear maps of the space of m−tuples of matrices (symmetric or skew- symmetric) is a free module on a certain subalgebra of invariants. Also in these cases, we exhibit an explicit basis of this module. Furthermore we recover some classical results analog to the Amitsur-Levitzki theorem. More precisely, for the symplectic group we recover a result of Rowen (see [29]) which states that the skew-symmetric standard polynomial of degree 4n − 2 vanishes on 2n × 2n matrices, sym- metric with respect to the symplectic involution. For the odd orthogonal group we recover the corresponding result of Hutchinson (see [17]) which states that the skew-symmetric standard polynomial of degree 4n is zero when restricted to skew-symmetric matrices 2n + 1 × 2n + 1. The case of the even orthogonal group deserves a study on its own. We will deal with it in a subsequent paper ([13]). Inspired by the strategy of [10], in which the result of Bresar et al. has been extended to a general result for a simple Lie algebra g concerning the space of invariant skew symmetric g valued multilinear maps on g, we extend our results to a certain class of symmetric pairs (see section 3 for details). Indeed this will not be too hard since, apart from the symplectic and orthogonal (for odd n) involutions, the only other cases we need to consider are the pairs (so(2n), so(2n−1)) and (e6, f4). The first case is easy to deal with since the corresponding symmetric space is an odd dimensional sphere. The second is treated combining ad hoc reasoning and computer aided computation (we use the software LiE). Acknowledgements I want to thank professors C.De Concini and C.Procesi for precious suggestions and advices. 1 General setting, ideas and goals 1.1 Antisymmetry In this work we will study multilinear antisymmetric identities. Let us first introduce the appro- priate setting for this. By the antisymmetrizer we mean the operator that sends a multilinear An important example for us is obtained applying the antisymmetrizer to the noncommuta- ǫσf (xσ(1), . . . , xσ(h)). application f (x1, . . . , xh) into the antisymmetric application Pσ∈Sh tive monomial x1 · · · xh. We get the standard polynomial of degree h, Sth(x1, . . . , xh) = Xσ∈Sh ǫσxσ(1) . . . xσ(h). 2 Up to a scalar multiple, this is the only multilinear antisymmetric noncommutative polynomial of degree h. Let R be any algebra (not necessarily associative) over a field F, and let V be a finite dimensional vector space over F. The set of multilinear antisymmetric functions from V k to R can be identified in a natural way with Vk V ∗ ⊗ R. Using the algebra structure of R we have a wedge product of these functions; for G ∈Vh V ∗ ⊗ R, H ∈Vk V ∗ ⊗ R we define 1 (G ∧ H)(v1, . . . , vh+k) := ǫσG(vσ(1), . . . , vσ(h))H(vσ(h+1), . . . , vσ(h+k)) h!k! Xσ∈Sh+k = Xσ∈Sh+k/Sh×Sk ǫσG(vσ(1), . . . , vσ(h))H(vσ(h+1), . . . , vσ(h+k)). It is easy to show (see [22]) that Sta ∧ Stb = Sta+b. With this multiplication the algebra of multilinear antisymmetric functions from V to R is algebra. Assume now that R is an associative algebra and V ⊂ R. The inclusion map X : V → R is of isomorphic to the tensor product algebra V V ∗ ⊗ R. We shall denote by ∧ the product in this course antisymmetric, since the symmetric group on one variable is trivial, hence X ∈V V ∗ ⊗ R. By iterating the definition of wedge product we have the important fact: Proposition 1.1. As a multilinear function, each power X a := X ∧a equals the standard poly- nomial Sta computed in V . We consider now an example which will be useful in the following. We apply the previous considerations to V = R = Mn(C) := Mn; the group G = P GL(n, C) acts on this space, and hence on functions, by conjugation and it is interesting to study the algebra of G -- equivariant maps B := (^ M ∗ n ⊗ Mn)G. (1) This among other topics is discussed in [6]. Here we provide a slightly different approach to the study of B and we generalize it to the other classical groups. n)G. We see that B is naturally a A-module. It follows from results on Chevalley trasgression [8] and Dynkyn [14] that A is the exterior algebra in the elements: Let us first consider the space of invariants A := (V M ∗ Th := T r(St2h+1(x1, ..., x2n+1)), i = 0, ..., n − 1. We remark that we use only traces of the standard polynomials of odd degree since, as it is well-known, T r(St2h(x1, ..., x2h)) = 0 for every h, see [28]. Consider now the graded super algebra A[t], with deg t = 1. Notice that for each i, Tit = −tTi. Define the graded algebra homomorphism π : A[t] → B by Xj ajtj 7→Xj aj ∧ X j, aj ∈ A. By classical results (see for example [23]) we know that π is surjective. Furthermore by [6],[22] we have: 3 Theorem 1.2. 1. The algebra B is a free module on the subalgebra An−1 ⊂ A generated by the elements Ti, i = 0, ..., n − 2, with basis 1, ..., X 2n−1. 2. The kernel of the canonical homomorphim π : A[t] → B is the principal ideal generated by t2iTn−i−1 − nt2n−1. n−1 Xi=0 The proof of this theorem uses two main results. The first is a result of Kostant (see [19]) on the dimension of B and the second is the Amitsur-Levitzki's Theorem (see [1]). An interesting remark is that, assuming theorem 1.2, one could deduce the Amitsur-Levitzki's identity. In fact multiplying on the left by t the identity above we obtain nt2n −P t2i+1Tn−i−1, while multiplying on the right by t we obtain nt2n −P t2iTn−i−1t. So t2n ∈ Ker(π) (by the fact that in A[t] we have Tit = −tTi). It is easy to see that, in our setting, X 2n = 0 is equivalent to the Amitsur-Levitzki's identity. We will use this trick to deduce other interesting identities in the case of symmetric or skew-symmetric matrices. 1.2 Covariants in exterior algebras of Lie algebras with involution In the following we will use previous ideas to show similar results for a certain class of Lie algebras with involution. So let g be a complex finite dimensional semisimple Lie algebras and let σ : g → g be an indecomposable involution. We denote by g = k ⊕ p its Cartan decomposition, where k is the fixed point set of σ. By classical results (see for example [15]) we know that g is either simple or a sum of two simple ideals switched by the flip involution. algebra One of the main result of [10] is the following theorem: For the case g non simple, it is a celebrated theorem of Hopf, Samelson and Koszul that the where the integers mi, with m1 ≤ ... ≤ mr are the exponents of ∆ and r is the rank of k. The main object of our interest will be the space (V p∗ ⊗ g)k for pairs (g, σ) such that the algebra of invariants (V p∗)k is an exterior algebra. (V p∗)k = (V k∗)k is always an exterior algebra over primitive generators Pi of degree 2mi + 1, Theorem 1.3. The algebra B := (V k∗ ⊗ k)k is a free module of rank 2r on the subalgebra Ar−1 ⊂ A := (V k∗)k generated by P1, ...Pr−1. pairs (g, σ) such that (V p∗)k is an exterior algebra are: For the case g simple we have that from classical results (see [15], [3], [30]) the only four 1. (sl(2n), −s), where s is the symplectic transposition and k = sp(2n), 2. (sl(2n + 1), −t), where t is the usual transposition and k = so(2n + 1), 3. (so(2n), σ1), where σ1 is an involution such that k = so(2n − 1), 4. (e6, σ2), where σ2 is an involution such that k = f4. 4 In this paper we study the covariants in these cases so that our results plus Theorem 1.3 will give the following: Theorem 1.4. Let g be a complex finite dimensional semisimple Lie algebra. Let σ : g → g be the xi's are ordered by their degree and r is such that r := rk(g) − rk(k). We have that the an indecomposable involution such that (V p∗)k is an exterior algebra of type ∧(x1, ..., xr), where algebra B := (V p∗ ⊗ g)k is a free module of rank 4r on the subalgebra of (V p∗)k generated by the elements x1, ..., xr−1. 2 Invariant theory in exterior algebras for the symplectic and orthogonal groups 2.1 Symplectic case We denote, for all k ∈ N, by 1k the identity matrix of order k and by At the usual transposition of a matrix A. We consider the skew-symmetric matrix J :=(cid:18) 0 −1n 1n 0(cid:19). On the space V := C2n we have the skew-symmetric form (u, v) := utJv, where u and v are column vectors in C2n. The symplectic transposition, A 7→ As, is defined by As := −JAtJ. Explicitly we have that, if M =(cid:18)A B C D(cid:19) with A, B, C, D n × n, its symplectic transpose is M s =(cid:18) Dt −Bt −Ct At (cid:19) . (2) Let M + transposition. Notice that the map A 7→ AJ gives a linear isomorphism of the space M + 2n = {A As = A} be the space of symmetric matrices with respect to the symplectic 2n onto the spaceV2 V of skew-symmetric matrices (with respect to the usual transposition). The group GL(2n, C) = GL(V ) acts on V2 V by X ◦ A := XAX t, for X ∈ GL(V ) and A ∈V2 V . Let G ⊂ GL(V ) be the symplectic group. By definition G is the group of transformations preserving the symplectic form, that is G := {X ∈ GL(V ) J = X tJX}. This is equivalent to require that X −1 = X s. So we can see the symplectic group as the fixed point set of the involution of GL(V ) given by X 7→ (X s)−1. The Lie algebra M − 2n of G is the space of the skew-symmetric matrices, with respect to the symplectic form, M − 2n := {A As = −A}. The map A 7→ AJ gives a linear isomorphism of M − 2n onto the space S2(V ) of symmetric matrices (with respect to the usual transposition). Also in this case GL(V ) acts on S2(V ) by X ◦ A := XAX t, for X ∈ GL(V ) and A ∈ S2V . The conjugation action of G commutes with the map A 7→ AJ. In fact if X ∈ G we have XJX −1 = J, so we can state that 5 Proposition 2.1. The action of the symplectic group G on M + 2n (resp. on M − 2n) can be identified with the restriction to G of the usual action of the linear group GL(V ) onV2 V (resp. on S2(V )). 2n ≃ S2(V ) is irreducible, M + 2n is not irreducible. Indeed it decomposes as the direct sum of the one dimensional space of scalar matrices and of the space P0 of traceless matrices. Under the isomorphism with Remark that with respect to the action of G, while the Lie algebra L := M − V2 V the space of scalar matrices maps to the space spanned by J. 2.2 Invariants of the representation V(M ± 2.2.1 Dimension 2n)∗ ⊗ M2n Here and below we index the irreducible representations of GL(V ) by Young diagrams with at Let us start from the case M + ing to the diagram λ will be denoted by Sλ(V ). most n columns (the row of length k corresponds to Vk V ). The irreducible module correspond- 2n ≃ V2 V . By the Plethysm formulas (see [21]) we know how to decompose V[V2 V ] with respect to the action of the linear group GL(V ). We have that V[V2 V ] is the direct sum of the irreducible representations H − a1,a2,...,ak (V ) = H − a (V ) := Sλ(a)(V ), with 2n > a1 > a2 . . . > ak > 0. Notice that H − ∅ is the trivial one dimensional representation. The Young diagram λ(a) is built by nesting the hook diagrams hai whose column is of length ai and whose row is of the length ai + 1,. As an example, the diagram λ(4, 3, 1) is V M + Using this, we can compute the dimension of the invariants for the symplectic group G in 2n. We know that, for each diagram λ, dim(Sλ(V ))G) ≤ 1 and dim(Sλ(V )G) = 1 if and only if every row of λ is even, (see [23]). The rows of a representation H − a (V ) are even if and only if a = b1, b1 −1, b2, b2 −1, . . . , bs, bs − 1, . . ., or a = b1, b1−1, b2, b2−1, . . . , bs, 1 with the bj's odd and bs > 1. So we have to compute the number of decreasing sequences of odd numbers smaller than 2n. This is the same to compute the number of decreasing sequences of numbers taken from 1, ..., n, that is the number of subsets of {1, . . . , n}. Thus Proposition 2.2. The dimension of the space of invariants (V M + 2n is quite similar. We set L := M − The case M − 2n ≃ S2V . 2n)G is 2n. Remark 2.3. Remark that L is the Lie algebra of type Cn and in this part of the work we will recover general results for covariants in the exterior algebra of a Lie algebra g as in [10]. By the Plethism formulas (see [21]) we know how to decompose V[S2V ] with respect to the action of the linear group GL(V ). 6 We have that V[S2V ] is the direct sum of the irreducible representations H + a1,a2,...,ak (V ) = H + a (V ) = Sλ(a)(V ), (3) with 2n > a1 > a2 . . . > ak ≥ 0. This time however the Young diagram λ(a) is built by nesting the hook diagrams hai whose column is of length ai + 2 and whose row is of the length ai + 1. Now (analogous to the previous case) we want to compute the dimension of the invariants for the symplectic group. We have that the rows of a representation H + a (V ) are even if and only if the sequence is of the type a, a − 1, b, b − 1, c, c − 1, . . ., with a, b, c, ... odd. So as before we can state: Proposition 2.4. The dimension of the space of invariants (V M − We now pass to determine the dimension of the space B := (V M ± 2n ⊗ M2n)G. Before we proceed, let us make a few remarks. We know that to study an isotypic component relative to an irreducible representation N in a representation M of a reductive group G, we need to study the space homG(N, M ) that we can identify to (N ∗ ⊗ M )G. The dimension of this space will be the number of copies of N in M . 2n)G is 2n. We decompose the space of matrices M2n = M + 2n ⊕ M − 2n = C ⊕ P0 ⊕ M − 2n. Thus we have 2n ⊗ M2n(cid:17)G =(cid:16)^ M ± 2n ⊗ C(cid:17)G ⊕(cid:16)^ M ± 2n ⊗ P0(cid:17)G 2n ⊗ M − ⊕(cid:16)^ M ± 2n(cid:17)G Since P0 and M − on the isotypic components relative to these two representations. 2n ⊗ C)G is the space of invariants whose dimension we have already computed. 2n are both irreducible and self dual, our computation will give us information Now identify, as a representation of G, M2n ≃ V ⊗V . We thus have to study the G-invariants (cid:16)^ M ± The space (V M ± in 2n) ⊗ V ⊗ V = ⊕aH ∓ a (V ) ⊗ V ⊗ V. ^(M ± By Pieri's formulas we know how to decompose Sλ(V )⊗V ⊗V . First we decompose Sλ(V )⊗V . We have that Sλ(V ) ⊗ V = ⊕Sλi (V ) where λi runs along the diagrams whose first row is of length at most 2n and which are obtained from λ by adding one box. So, iterating the same process, to compute the dimension of the invariants we have to study when, adding two squares, starting from one of the diagrams H − we obtain a diagram with even rows. Theorem 2.5. The dimension of B is (2n − 1)2n (resp. (2n)2n) in the case M + 2n (resp. M − 2n). Proof. We prove only the case M + 2n. The proof of the case M − 2n is similar. We proceed by induction on n. The case n = 1 is trivial so we assume n > 1. We have to compute the sequences of type: 2n > a1 > a2 > ... > ak > 0, such that when we add two squares to the corresponding diagram we obtain a diagram with even rows. We can have four different situations. Let us start by considering the case a1 < 2(n−1) = 2n−2. By induction the number of diagrams 7 with even rows whose first row has at most 2(n − 1) boxes is (2(n − 1) − 1)2n−1. However if a1 = 2n − 3 we can add two squares to the first row so that its length is (a1 + 3 = 2n). Thus we have a contribution from each sequence a = 2n − 3 > 2n − 4 > b2 > b2 − 1 > . . . > bs − 1 or a = 2n − 3 > 2n − 4 > b2 > b2 − 1 > . . . > bs − 1 > 1 with the bj's odd and bs > 1. Reasoning exactly as in the proof of Proposition 2.2 we see that there are exactly 2n−2 such diagrams. The second case is a1 = 2n − 1 and a2 = 2n − 2, In this case we have two possibilities. We can add two squares at the bottom X X and we get a contribution from each sequence a = 2n − 3 > 2n − 4 > b2 > b2 − 1 > . . . > bs − 1 or a = 2n − 1 > 2n − 2 > b2 > b2 − 1 > . . . > bs − 1 > 1 with the bj's odd and bs > 1 and there are 2n−1 such sequences. Otherwise we add the two squares to the diagram associated to the sequence 2(n − 1) > a3 > a4 > ... > ak > 0, so that, by induction on n, we get (2(n − 1) − 1)2n−1 contributions. The third case is: a1 = 2n − 1 and a2 < 2n − 2, in this case the only way to add two squares is to add one on the second row and one to the second column X X so we would get 2n−2 contributions, but we can exchange the order in which we add the squares, so we get 2 · 2n−2 contributions. In the case we have a1 = 2n − 2, so that in order to get a diagram with even rows, we need to add a box to the first row. If a2 < 2n − 4, there are at least two rows of length 1 so that we cannot obtain a diagram with even rows by adding a single box. If a2 = 2n − 3 we need to add the second square to the second row X X and one easily sees that we get 2n−2 contributions. 8 If a2 = 2n − 4 we need to add a box to the second column: X X and in this case we get 2 · 2n−2 contributions, since we can exchange the order in which the boxes are inserted. Finally adding all the contributions we obtain (2(n − 1) − 1)2n−1 + 2n−2 + 2n−1 + (2(n − 1) − 1)2n−1 + 2n−1 + 2n−2 + 2n−1 = (2n − 1) · 2n. This is our claim. 2.2.2 Structure of algebras of invariants of the representation V(M + both V∗(M + Let M2n be, as in the previous section, the space of complex matrices 2n × 2n and let M + 2n := {x ∈ M2nx = xs} be the symmetric part with respect to the symplectic involution. Since 2n)∗ ⊗ M2n have a structure of graded associative algebras, also the 2n)∗ ⊗ M2n algebras 2n)∗!G ^(M + and B := ∗ ^(M + 2n)∗ ⊗ M2n!G 2n)∗ and V∗(M + A := ∗ have a natural structure of graded associative algebras. Furthermore B is clearly an A-module. These are the structures we want to investigate. As an algebra A is generated by the element T r(St2h+1(x1, ..., x2h+1)). Indeed by classi- cal invariant theory, we know that the polynomial invariant functions on the space M2n are generated by traces of the monomials in the variables xi, xs i (FFT for matrices, see [23]); fur- thermore since M + 2n is a G−stable subspace of M2n, we have that every G−invariant polynomial function on M + 2n is the restriction of a G−invariant polynomial on M2n (since G is a linear reductive algebraic group). So by multilinearizing and alternating we have our claim (recall that T r(St2h(x1, ..., x2n)) = 0). 2n⊕M + Since M2n = M − 2n)∗ ⊗ M + 2n)∗⊗M2n =(cid:0)V(M + and, passing to the invariants, B = B+ ⊕ B−, with B∓ = (V(M + 2n we have thatV(M + Further, we can define the element 2n(cid:1)⊕(cid:0)V(M + 2n)G. 2n)∗ ⊗ M − 2n)∗ ⊗ M ± 2n(cid:1) (4) X ∈ B1 by X(x) = x for any x ∈ M + some crucial properties of A and B: 2n (recall that M + 2n ⊂ M2n). The following simple Proposition gives Proposition 2.6. We have: 1. The element T r(St4k+3(x1, ..., x4k+3)) is zero for each k ≥ 0. 2. X k ∈ B− if and only if k ≡ 0, 1 modulo 4 9 3. X k ∈ B+if and only if k ≡ 2, 3 modulo 4. Proof. We prove part 1, the others being similar. By the explicit form of the symplectic transposition (1) we have T r(x) = T r(xs), so but T r(St2h+1(x1, ..., x2h+1)) = T r(St2h+1(x1, ..., x2h+1)s), St2h+1(x1, ..., x2h+1)s = Xσ∈S2h+1 = Xσ∈S2h+1 ǫσ(xσ(1) · · · xσ(2h+1))s ǫσxσ(2h+1) · · · xσ(1), and if we consider the permutation η : (1, ..., 2h + 1) → (2h + 1, ..., 1) we see that it has the same sign of the parity of h. So everything follows. We deduce the following well know result (see for example [2]): Theorem 2.7. The algebra A is the exterior algebra, of dimension 2n, in the elements T0, T1, ..., Tn−1, where Th := T r(S4h+1(x1, ..., x4h+1)) ∈ A4h+1. Proof. By the previous Proposition, A is generated by the elements T r(St4h+1(x1, ..., x4h+1)). By the Amitzur-Levitzki theorem, Str(x1, ..., xr) = 0 for r ≥ 4n. It follows that A is generated by T0, T1, ..., Tn−1, so it is the quotient of an exterior algebra on n generators. Since dim A = 2n, our claim follows. Notice that we can define the trace function T r : V(M + 2n by extending the invariant function on matrices. By equivariance, on B the trace function takes values in A. In particular notice that T r(X h)(x1, . . . xk) = T r(Sth(x1, . . . xk)). 2n)∗ ⊗ M2n → V M + Theorem 2.8. We have: 1. As a algebra, B is generated by A and the element X. 2. B is a free module on the the exterior algebra An−1 ⊂ A generated by T0, ..., Tn−2, with basis 1, X, ..., X 4n−3. Proof. 1) follows from classical invariant theory (see [24],[23],[6]). 2) By Proposition 2.5, it suffices to see that the elements 1, X, ..., X 4n−3 are linearly inde- pendent over An−1. So, let 4n−3 Pj ∧ X j = 0, Xh=0 Ph ∈ An−1 for each h = 0, . . . , 4n − 3. Assume by contradiction that not all Ph's are 0. Let j be the minimum such that Pj 6= 0. Multiply by X 4n−3−j and take traces. We get Pj ∧ Tn−1 = 0, since for h ≥ 4n, X h = 0, T r(X 4n−2) = T r(X 4n−1) = 0 by Proposition 2.6 and the fact that 4n − 2 is even. Pj ∈ An−1, so Pj ∧ Tn−1 = 0 if and only if Pj = 0, a contradiction. 10 Consider now the graded superalgebra A[t], with deg t = 1. Notice that for each i, Tit = −tTi. Define the graded algebra homomorphism π : A[t] → B by aj ∈ A. By the previous theorem we see that π is surjective. Let us describe its kernel. Xj ajtj 7→Xj ajX j, Theorem 2.9. The Kernel I of the homomorphism π is the principal ideal generated by the element n−1 nt4n−3 − Xi=0 1 2 Tn−i−1t4i. Proof. Let us first start by showing that the element nt4n−3 −Pn−1 the second part of Theorem 2.7, we necessarily have a homogeneous relation i=0 1 2 Tn−i−1t4i lies in I. By Tn−1 = 4n−3 Xh=0 Ph ∧ X h with Ph ∈ An−1 of degree 4n − 3 − h. Let us compute P4n−3. Notice that T r(P4n−4 h=0 Ph ∧ X h) ∈ An−1, while T r(Tn−1) = 2nTn−1 It follows that (2n−P4n−3)∧Tn−1 ∈ An−1 namely P4n−3 = 2n. Assume now 0 ≤ j ≤ 4n − 4 and multiply by X 4n−3−j. Taking the trace and reasoning as in the proof of Theorem 2.8, we get that Tn−1 ∧ (T r(X 4n−3−j) + (−1)jPj) ∈ An−1 If j = 0 we get P0 = 0. Assume j > 0. This implies T r(X 4n−3−j) + (−1)jPj = 0 that is Pj =(0 if j = 2h + 1, 4h + 2 if i = 4h + 1 Th Thus dividing by 2 we obtain 1 2 Tn−1 = nX 4n−3 − 1 2 n−1 Xj=1 Tn−j−1 ∧ X 4j (5) which is our relation. Denote by J the ideal generated by nt4n−3 − 1 2 n−1 Xi=0 Tn−i−1t4i. We have seen that J ⊂ I. Now we show that t2n−2 ∈ J. To see this let us write the relation (5) as Tn−h−1t4h. nt2n−3 − n−1 Xh=0 11 Remark that multiplying on the right by t we get nt2n−2 − n−1 Xh=0 Tn−h−1t4h+1 ∈ J. If we multiply by t on the left we get nt2n−2 − n−1 Xh=0 tTn−h−1t4h = nt2n−2 + Tn−h−1t4h+1. n−1 Xh=0 Adding the two relation we thus obtain Now consider A/J. Using (6), it is clear that the image of 1, t, . . . t4n−3 span A/J as a A module. However by (5) we deduce that 2nt2n−2 ∈ J. (6) Tn−1 = 2nt4n−3 − Tn−i−1t4i, n−1 Xi=1 so that, substituting, the same elements span A/J as a An−1 module. It follows that dim A/J ≤ (4n − 2)2n−1 = (2n − 1)2n. On the other hand we know by Proposition 2.5 that dim A/I= dim B = (2n − 1)2n. We deduce that I = J. Remark 2.10. The fact that X 2n−2 = 0 means that the standard polynomial St4n−2(x1, ..., x4n−2) is identically 0 for x1, . . . , x4n−2 ∈ M + 2n. This is a result of Rowen [29]. Finally using Proposition 2.6 we deduce, Corollary 2.11. 1) The elements 1, X, X 4, X 5, ..., X 4n−3 are a basis of B− as a free module over the exterior algebra An−1. 2) The elements X 2, X 3, X 6, X 7, ..., X 4n−5 are a basis B+ as a free module over the exterior algebra An−1. By the classical invariant theory we have that polynomial functions G-equivariant from m- copies of L to M2n, are an associative algebra generated by coordinates Yi and by the traces of 2.2.3 Structure of invariants of the representation V(M − monomials in Yi. So we have that [V L∗ ⊗ M2n]G is generated by the element Y, Y s = −Y and T r(Y i). Further we can decompose in a natural way B = B+ ⊕ B− as 2n)∗ ⊗ M2n We have an analog of the Proposition 2.6: B := [^ L∗ ⊗ M2n]G = [^ L∗ ⊗ L]G ⊕ [^ L∗ ⊗ M + 2n]G. Proposition 2.12. 1. The element T r(St4k+1(y1, ..., y4k+1)) is zero for each k ≥ 0. 2. Y k ∈ B− if and only if k ≡ 0, 3 modulo 4 12 3. Y k ∈ B+ if and only if k ≡ 1, 2 modulo 4. Proof. As in the Proposition 2.6 we just prove the first fact. We recall that but T r(St2h+1(y1, ..., y2h+1)) = T r(St2h+1(y1, ..., y2h+1)s), St2h+1(y1, ..., y2h+1)s = Xσ∈S2h+1 = Xσ∈S2h+1 ǫσ(yσ(1) · · · yσ(2h+1))s (−1)2h+1ǫσyσ(2h+1) · · · yσ(1), and if we consider the permutation η : (1, ..., 2h + 1) → (2h + 1, ..., 1) we see that it has the same sign of the parity of h. So everything follows. We set we have, by the computation of the dimension and by the Amitsur-Levitzki's Theorem, that Th := T r(St4h+3(Y1, ..., Y4h+3)) (7) Lemma 2.13. The algebra A := (V M − So we can describe the structure of covariants: 2n)G is the exterior algebra in the elements T0, ..., Tn−1. Theorem 2.14. 1. The algebra B is a free module on the subalgebra An−1 ⊂ A generated by the elements Ti, i = 0, ..., n − 2, with basis 1, ..., Y 4n−1. 2. With analogous notations to the previous section we have that the kernel of the canonical homomorphim π : A[t] → B is the principal ideal generated by t4i ∧ Tn−i−1 − 2nt4n−1. n−1 Xi=0 Proof. By computing dimensions we only need a formula to compute the multiplication by the element Tn−1 = T r(Y 4(n−1)+3) = T r(Y 4n−1), but in this case we can use the general formulas for the matrices (see [2]). We start from the universal formula for M2n: 2n−1 1 ∧ T (4n − 1) = T (4n − 1) = − Z 2i ∧ T (2(2n − i) − 1) + 2nZ 4n−1, (8) where Z : M2n → M2n is the identity map and T (h) := T r(Sth(x1, ..., xh)). In our case Z ❀ Y and we have (by 2.12) that T (2(2n − i) − 1) = 0 unless that −2i − 1 = 3 mod(4). So i is even. We deduce: Xi=1 Tn−1 = − n−1 Xi=1 Y 4i ∧ Tn−i−1 + 2nY 4n−1, (9) 13 so Y j ∧ Tn−1 = − n−[ j 4 ] Xi=1 Y 4i+j ∧ Tn−i−1. (10) So by this last result and by Proposition 2.12 we have (consistently to the general theory for Lie algebras developed in [10]): Corollary 2.15. 1. The 2n elements Y, Y 2, Y 5, Y 6, . . . , Y 4n−3, Y 4n−2 are a basis of B+ on the exterior algebra generated by the n − 1 elements Ti, i = 0, . . . , n − 2. 2. The 2n elements 1, Y 3, Y 4, Y 7, . . . , Y 4n−4, Y 4n−1 are a basis of B− on the exterior algebra generated by the n − 1 elements Ti, i = 0, . . . , n − 2 2.3 The odd orthogonal case In the last part of this section we want to generalize our considerations to the case in which the group G considered is not Sp(2n), but O(2n + 1). Results are very similar and, almost always, the proofs are the same as in the symplectic case, so we leave details to the reader. 2n+1)∗ ⊗ M2n+1)G, where M ± 2n+1 indicates the space of the symmetric or skew-symmetric matrices with respect to the usual transposition. In this case we start to investigate the space (V(M ± 2.3.1 Dimension We start analyzing the skew-symmetric case. We recall the isomorphism M − V = C2n+1. We have the usual decomposition: 2n+1 ≃V2 V , where ^ 2 ^ V! =M H − a1,a2,...,ak (V ), with 2n + 1 > a1 > · · · ak > 0. The only difference with the symplectic case regards the invariants. This time, there is the invariant only if the diagram has even columns (see [23]). We have: Proposition 2.16. The dimension of the space of invariants V(cid:0)M − Proof. We want that the diagram has even columns. So the sequence a1 > a2 > ... > ak must be of type 2n + 1 > b1 > b1 − 1 > b2 > b2 − 1 > ..., with bi even (not zero). It follows that the number of these sequences is the number of different sequences composed by even numbers minor than 2n + 1, therefore likewise the symplectic case we have our claim. is 2n. 2n+1(cid:1)G The symmetric orthogonal case is very simple by previous considerations. We have M + 2n+1 ≃ S2(V ). So by the decomposition (3), to compute the dimension of invariants we have to consider sequences of type 2n + 1 > b1 > b1 − 1 . . . > bk > bk − 1 > 0, with each bi even or of type 2n + 1 > b1 > b1 − 1 . . . bk−1 > bk−1 − 1 > bk = 0. So we have: Proposition 2.17. The dimension of the space of invariants (cid:0)V M + 2n+1(cid:1)G is 2n+1. 14 Furthermore we can compute the dimension of covariants. Proposition 2.18. The dimension of the space (V(M − M2n+1)G) is n2n+1 (resp. (2n + 1)2n+1). 2n+1)∗ ⊗ M2n+1)G (resp. (V(M + 2n+1)∗ ⊗ Proof. We will prove only the case M − from Theorem 2.5. 2n+1, the other case is similar. The proof follows easily In fact we can reinterpret the orthogonal case from the point of view of the symplectic case. By transposing diagrams we are in the situation of the skewsymmetric symplectic case for 2n up to: 1. subtract the contribution given by the case in which the first two columns have maximal length and we add two boxes to these. 2. add the contribution given by the case in which the first row has maximal length and we add two boxes to this. So we have that the dimension is n2n+1 − 2n−1 + 2n−1 = n2n+1. 2.4 Structure 2n+1)∗ ⊗ M2n+1)G. We can see such 2n+1)∗)G which is, as an algebra, generated by the elements T r(St2k+1) and similarly to the symplectic case can be described as the exterior algebra. We proceed to describe explicitly of the space B := (V(M − algebra as a left module on the algebra A := (V(M − Proposition 2.19. The algebra A = (V(M − 2n+1)∗)G is the exterior algebra in the elements Th = T r(St4h+3), with h = 0, 1, ..., n − 1. Proof. By the proof of Proposition 2.12 and the Amitsur-Levitzki's Theorem we have that the elements Th, h = 0, ..., n − 1 generate the invariants. Since the dimension is exactly 2n, we have our claim. To study covariants let us recall the by classical invariant theory we have that the space 2n+1)∗)G by the elements Y i. So reasoning as in 2.8 and 2.9 we can state the following. 2n+1)∗ ⊗ M2n+1)G is generated as a module on A = (V(M − B = (V(M − Theorem 2.20. 1. As a algebra, B is generated by A and the element Y . 2. B is a free module on the the exterior algebra An−1 ⊂ A generated by T0, ..., Tn−2, with basis 1, Y, ..., Y 4n−1. With notations of Section 2 we can consider the canonical surjective homomorphism π : A[t] → B. We have: 3. The Kernel of the homomorphism π is the principal ideal generated by the element Tn−i−1 ∧ t4i, (11) (2n + 1)t4n−1 − n−1 Xi=0 15 Remark that multiplying by t the generator in (11), we obtain an identity between an element of even degree t4n and elements of odd degree t4i+1. By the relation Tn−i−1 ∧ t4i+1 = −t4i+1 ∧ Tn−i−1, we can deduce t4n = 0. We have recovered a well known result of Hutchinson (see [17]): Proposition 2.21. The skew-symmetric standard polynomial St4n(x1, ..., x4n) is zero on the space M − 2n+1. From the Proposition 2.12 we also get: Corollary 2.22. algebra A− 1. The algebra (V(M − n−1 with basis Y, Y 2, Y 5, Y 6, ..., Y 4n−3, Y 4n−2. 2n+1)∗ ⊗ M − 2n+1)G is a free module on the exterior 2n+1)∗ ⊗ M + 2n+1)G is a free module on the exterior algebra A− n−1 with 2. The algebra (V(M − basis 1, Y 3, Y 4, Y 7, ..., Y 4n−4, Y 4n−1. Let us pass to the symmetric case. We can describe easily the structure of invariants. By classical invariant theory we have that this algebra is generated by the elements T r(St2k+1), but by Propositions 2.6, 2.17 and Amitsur-Levitzki we have: elements T0, ..., Tn (Ti := T r(St4i+1)). Lemma 2.23. The algebra of invariants A := (cid:0)V(M + So we can describe the covariants B := (V(M + Theorem 2.24. 2n+1)∗(cid:1)G 2n+1)∗ ⊗ M2n+1)G. 1. As a algebra B is generated by A and the element X. is the exterior algebra in the 2. B is a free module on the the exterior algebra An−1 ⊂ A generated by T0, ..., Tn−1, with basis 1, X, ..., X 4n+1. With notations of Section 2 we can consider the canonical surjective homomorphism π : A[t] → B. We have: 3. The Kernel of the homomorphism π is the principal ideal generated by the element Proof. We need to prove only the third part. By the analog of the general formula (8) we have (2n + 2)t4n+1 − n Xi=0 Tn−i ∧ t4i+2, (12) 1 ∧ T (4n + 1) = T (4n + 1) = − so we deduce i is odd and we have: n 2n+1 Xi=1 Z 2i ∧ T (2(2n + 2 − i) − 1) + (2n + 2)Z 4n+1, Tn = − Xi=0 X 4i+2 ∧ Tn−i + (2n + 2)X 4n+1, which is our claim. Corollary 2.25. 2n+1)∗ ⊗ M + with basis 1, X, ..., X 4n, X 4n+1. 2. The algebra (V(M + algebra An−1, with basis X 2, X 3, ..., X 4n−2, X 4n−1. 1. The algebra (V(M + 2n+1)∗ ⊗ M − 2n+1)G is a free left module on the exterior 2n+1)G is a free left module on the exterior algebra An−1, 16 3 Nonclassical cases 3.1 The case so(2n)/so(2n − 1) In this part of the work we want to investigate the remaining cases which we have discussed in the section 1.2.3. In this case g = k ⊕ p, where g = so(2n), k = so(2n − 1) and p = C2n−1. Let us start considering the invariants of the the space so(2n − 1)(cid:19)∗ ^(cid:18) so(2n) , under di action of so(2n − 1) (acting by derivation), we will denote this space by A. If we denote V := C2n−1 we have that the action of so(2n − 1) is the natural action on ∧V . From the point of view of the classical invariant theory, we have to look at multilinear skew- symmetric invariants under simultaneous conjugation of the group K = SO(2n−1). By the FFT we deduce they are the exterior algebra in the element p = [v1, ..., v2n−1], i.e., the determinant. Let us investigate the space of covariants. We have: (cid:16)^ p∗ ⊗ k(cid:17)K ≃ ^(V )∗ ⊗ 2 ^ V!K , we can deduce easily that this space has dimension 2. Then we can consider in(cid:16)V2n−3(V ) ⊗V2 V(cid:17)∗ the unique (up to scalar) element p = [v1, ..., v2n−1] and we take the corresponding covariant element: . 2 ^ (V )∗ ⊗ ^ V!K ω1 =X[v1, ...v2n−3, ei, ej]ei ∧ ej ∈ 2n−3 On the other hand if we consider in (cid:16)V2(V ) ⊗V2 V(cid:17)∗ xi ∈ so(2n) we have the corresponding element ω2 ∈(cid:16)V2(V )∗ ⊗V2 V(cid:17)K Theorem 3.1. The space B+ := (V p ⊗ k)K is the vector space of dimension 2 with basis (over We can do similar considerations for the case B− := (V p ⊗ p)K. From the point of view of the classical invariant theory we have to look at the space the element ([x1, x2], [x3, x4]), where So we have: C) ω1, ω2. . (cid:16)^(V )∗ ⊗ V(cid:17)K . As before we have that this space has dimension 2 and a basis is given respectively by the and the covariant θ2 covariant element θ1 corresponding to [v1, ..., v2n−1] ∈ (cid:16)V2n−2(V ) ⊗ V(cid:17)∗ corresponding to (v1, v2) ∈(cid:16)V1(V ) ⊗ V(cid:17)∗ Theorem 3.2. The space B− := (V p ⊗ p)K is the vector space of dimension 2 with basis (over . So we have: C) θ1, θ2. 17 3.2 Exceptional case The last case we consider is the symmetric space E6/F4. Let us start recalling some general results for Lie algebras (see for example [16], [20]). Let g be a finite dimensional Lie algebra over C and let u ⊂ g a subalgebra reductive in g, i.e. the adjoint representation u × g → g is completely reducible. We denote by p a stable complement under the adjoint action, so g = u ⊕ p and as a vector space p ≃ g/u. In this setting we can state the following (see [16] Theorem 12): Theorem 3.3. Let g, u as before, and assume furthermore that the restriction homomorphism maps (V g)g onto (V u)u. Then we have (cid:16)^ u(cid:17)u ⊗(cid:16)^ p(cid:17)u ≃(cid:16)^ g(cid:17)g . In our case we have that g = Lie(E6) and u = Lie(F4). If we denote by N the algebra of invariants (V g)g = (V g)G, we know by classical results that N = ∧(p3, p9, p11, p15, p17, p23), where pi is a primitive invariant of degree i. While (V u)U = ∧(q3, q11, q15, q23), with qi = π+(pi), where π+ : (V g)G → (V u)U is the natural map induced by p+ : g → u. So by the Theorem 3.3 we have that A := (V p)U is the exterior algebra ∧(r9, r17), where ri = π−(pi) and π− : N → A Now we can consider M = hom(V g, g)G and B+ = hom(V p, u)U . Further we can decompose in a natural way M = M − ⊕ M + as is induced by p− : g → p. M =(cid:16)^ g ⊗ p(cid:17)G ⊕(cid:16)^ g ⊗ u(cid:17)G , so we can define an homorphism Γ : M → B+ such that φ 7→ π−(φ+). We recall (See [10]) that M is a free module on the subalgebra ∧(p3, p9, p11, p15, p17) ⊂ N with basis certain elements ui−2, fi−1, where i runs over the degrees of the generators of invariants and uj, fj has degree j. Using the software "Lie" we have the following table that describes the graduated module B+: degree dimension 0 0 1 0 2 1 3 0 4 0 5 0 6 0 7 1 8 0 9 0 10 1 11 1 12 0 13 0 (13) Remark that the dimension of p is 26, so by the Poincar´e duality the previous table it is sufficient to describe the all space. We deduce dim(B+) = 8 and each covariant has different degree. Let us consider gi := Γ(fi), with i = 2, 10 and vi := Γ(ui), with i = 7, 15. We want to prove that B+ is a free module on the one-dimensional algebra generated by r9 with basis g2, g10, v7, v15. So we only need to prove that all gi, vi and their products by r9 is nonzero. We denote by (·, ·)g the Killing form of g, we know that it induces, by restriction, a non degenerate invariant form on u, which will be denoted by (·, ·)u. Our goal is to prove that, up to a non zero scalar, (g2, v15)u = (v7, g10)u = r17, (14) so our claim will follow. Let us start considering the first scalar product (v7, g10)u. We know (see [10]) that, up to a non zero scalar, we have: (u7, f10)g = p17. 18 Let us write u7 = u+ 7 + u− 7 . Using the software "Lie" we have the following table describing B− := (V p ⊗ p)H: degree dimension 0 0 1 1 2 0 3 0 4 0 5 0 6 0 7 0 8 1 9 1 10 1 11 0 12 0 13 0 (15) We deduce π−(u− 7 ) = 0. On the other hand, we have p17 = (u7, f10)g 7 + u− 7 , f + = (u+ = (u+ 7 , f + 10 + f − 10)g 7 , f − 10)g, 10)g + (u− so applying π− we have r17 = π−(p17) = π−((u7, f10)g) = (π−(u+ = (v7, g10)h. 7 ), π−(f + 10))h We can do the same considerations for (g2, v15)u. Furthermore, by linearity we have: (r9g2, v15)u = (r9v7, g10)u = (g2, r9v15)u = −(v7, r9g10)u = r9r17 6= 0. We can do similar considerations for the space B−. Let us consider vi := π−(u− i ), i = 1, 9 and gi := π−(f − i ) = 0, i = 1, 9. As before, the Killing form (·, ·)g induces a non degenerate invariant form on p, we denote it by (·, ·)p and we have, up to a nonzero scalar, that i ), i = 8, 16. By the table (13) we deduce π−(u+ (v1, g16)p = (g8, u9)p = r17, and that (r9g8, v9)p = (r9v1, g16)p = (g8, r9v9)p = −(v1, r9g16)p = r9r17 6= 0. We can summarize our considerations in the following: Theorem 3.4. 1. The space B+, of dimension 2 · 22, is a free module on the subalgebra ∧(r9) ⊂ A, with basis g2, g10, v7, v15. 2. The space B−, of dimension 2 · 22, is a free module on the subalgebra ∧(r9) ⊂ A, with basis g8, g16, v1, v9. 4 Conclusions We now summarize the previous results and deduce Theorem 1.4. Let us start analyzing our results about spaces of type B+. Let g be a simple complex Lie algebra, σ : g → g an indecomposable involution and g ≃ p ⊕ k the Cartan decoposition with k the fixed point set of σ. Let us assume that (V p∗)k is an exterior algebra of type ∧(x1, ..., xr), where the xi's are ordered by their degree and r is such that r := rk(g) − rk(k). Then we have: 19 Theorem 4.1. The algebra B+ := (V p∗ ⊗ k)k is a free module of rank 2r on the subalgebra of (V p∗)k generated by the elements x1, ..., xr−1. Proof. The only thing we have to prove is that our results about the orthogonal and symplectic groups can be seen from this point of view in a way that implies our theorem. The case B+ is particularly simple. In fact by little changes in the proof of the calculus of the dimensions we see that dim(B+) = (n − 1)2n−1 for the symplectic case and n2n for the odd orthogonal case. Then our claim follows choosing as a basis the elements Y i := X i − 1 m T r(X i), with m = 2n, 2n + 1 and from the fact that for traceless matrices T0 = 0. Further we have the same result for B−. We just have to remark that 1 /∈ B− and in both symplectic and orthogonal cases we can obtain the generator of higher degree, respectively X 4n−3 − 1 2n+1 T r(X 4n+1) by the characterizing identities we have proved. We deduce: 2n T r(X 4n−3) and X 4n+1 − 1 Theorem 4.2. The algebra B− := (V p∗ ⊗ p)k is a free module of rank 2r on the subalgebra of (V p∗)k generated by the elements x1, ..., xr−1. References [1] S.Amitsur, J.Levitzki, Minimal identities for algebras, Proc. Amer. Math. Soc. 1, (1950),449- 463. [2] A.Borel, (1967) [1954], Halpern, Edward, ed., Topics in the homology theory of fibre bundles, Lecture notes in mathematics 36, Berlin, New York: Springer-Verlag [3] A.Borel, Sur la cohomologie des especes fibres principaux et des espaces homogenes de groupes de Lie compacts, Ann. of Math. 57 (1953), 115-207 [4] A.Borel, F.Hirzebruch, Characteristic classes and homogeneous spaces, Amer. J. Math. 80 (1958), 459-538. [5] N.Bourbaki, Groupes et algebres de Lie, Hermann, Paris, 1968. [6] M. Bresar, C.Procesi, S.Spenko, Quasi-identities on matrices and the Cayley-Hamilton poly- nomial (2014) arXiv:1212.4597 [7] H.Cartan, La transgression dans un groupe de Lie et dans un espace fibre principal. Colloque de topologie (espaces fibres), Bruxelles, 1950, pp. 5771. Georges Thone, Liege; Masson et Cie., Paris, 1951 [8] C. Chevalley, The Betti numbers of the exceptional Lie groups, in Proc. International Congress of Mathematicians 1950, Vol. II, 21-24. [9] C. De Concini, P. Moseneder Frajria, P. Papi, C. Procesi, On special covariants in the exterior algebra of a simple Lie algebra, Rend. Lincei Mat. Appl. 25 (2014), 331-334 [10] C.De Concini, P.Papi, C.Procesi, The adjoint representation inside the exterior algebra of a simple Lie algebra (2014), arXiv:1311.4338 20 [11] C.De Concini, P.Papi, C.Procesi, Invariants of E8, available at: http://www1.mat.uniroma1.it/people/papi/E8.zip. [12] C.De Concini, C.Procesi, A characteristic free approach to invariant theory. Advances in Math. 21 (1976), no. 3, 330-354. [13] S.Dolce, On covariants in exterior algebras for the even special orthogonal group, in prepa- ration. [14] E.B.Dynkin, Homologies of compact Lie groups, Amer. Math. Soc. Transl. 12 (1959), 251- 300. [15] S.Helgason, Differential Geometry, Lie Groups, and Symmetric Spaces, Graduate Studies in Mathematics, vol. 34, 2001 (AMS edition). [16] G.Hochschild, J-P.Serre, Cohomology of Lie Algebras, Annals of Mathematics, Second Se- ries, Vol. 57, No. 3 (May, 1953), pp. 591-603. [17] J.P.Hutchinson, Eulerian Graphs and Polynomial Identities for Sets of Matrices, Proc. Nat. Acad. Sci. USA Vol. 71, No. 4, pp. 1314-1316, April 1974. [18] B.Kostant, A theorem of Froboenius, a theorem of Amitsur-Levitzki and cohomology theory, Indiana J. Math. (and Mech.) 7(1958), 237-264. [19] B.Kostant, Clifford algebra analogue of the Hopf-Koszul-Samelson theorem, the ρ- decomposition C(g) = EndVρ ⊗ C(P ), and the g-module structure of V g , Adv. Math. 125 (1997), 275-350 [20] J.L.Koszul, Homologie et cohomologie des algebres de Lie, Bull. Soc. Math. France 78 (1950), 66-127. [21] I.G.MacDonald, Symmetric Functions and Hall Polynomials. Oxford Mathematical Mono- graphs. [22] C.Procesi, On the theorem of Amitsur-Levitzki (2013), arXiv:1308.2421 [23] C.Procesi, Lie Groups, an approach through Invariants and Representations, Universitex, Springer. [24] C.Procesi, The invariant theory of n × n matrices, Advances in Math. 19 (1976), 306-381. [25] C.Procesi, A formal inverse to the Cayley-Hamilton theorem, J. Algebra 107 (1987), 63-74. [26] Yu.P.Razmyslov,Trace identities of full matrix algebras over a field of characteristic zero, Math. USSR-Izv. 8(1974), 727-760.. [27] M.Reeder, Exterior powers of the adjoint representation, Canad. J. Math. 49 (1997), 133- 159. [28] S.Rosset, A new proof of the Amitsur-Levitski identity, Israel J. Math. 23 (1976), 187-188. 21 [29] L.H.Rowen, A simple proof of Kostant's Theorem and an analogue for the symplectic invo- lution, Contemp. Math. 13 (1982), 207-215. [30] M.Takeuchi, On Pontrjagin classes of compact symmetric spaces, J.Fac.Sci.Univ.Tokyo Sect. I 9 1962 313-328 (1962). 22
1209.5660	1	1209	2012-09-25T16:10:50	A Poincare-Birkhoff-Witt theorem for quadratic algebras with group actions	[ "math.RA", "math.RT" ]	Braverman and Gaitsgory gave necessary and sufficient conditions for a nonhomogeneous quadratic algebra to satisfy the Poincare-Birkhoff-Witt property when its homogeneous version is Koszul. We widen their viewpoint and consider a quotient of an algebra that is free over some (not necessarily semisimple) subalgebra. We show that their theorem holds under a weaker hypothesis: We require the homogeneous version of the nonhomogeneous quadratic algebra to be the skew group algebra (semidirect product algebra) of a finite group acting on a Koszul algebra, obtaining conditions for the Poincare-Birkhoff-Witt property over (nonsemisimple) group algebras. We prove our main results by exploiting a double complex adapted from Guccione, Guccione, and Valqui (formed from a Koszul complex and a resolution of the group), giving a practical way to analyze Hochschild cohomology and deformations of skew group algebras in positive characteristic. We apply these conditions to graded Hecke algebras and Drinfeld orbifold algebras (including rational Cherednik algebras and symplectic reflection algebras) in arbitrary characteristic, with special interest in the case when the characteristic of the underlying field divides the order of the acting group.	math.RA	math	A POINCARE-BIRKHOFF-WITT THEOREM FOR QUADRATIC ALGEBRAS WITH GROUP ACTIONS A. V. SHEPLER AND S. WITHERSPOON Abstract. Braverman and Gaitsgory gave necessary and sufficient conditions for a nonhomogeneous quadratic algebra to satisfy the Poincar´e-Birkhoff-Witt property when its homogeneous version is Koszul. We widen their viewpoint and consider a quotient of an algebra that is free over some (not necessarily semisimple) subalgebra. We show that their theorem holds under a weaker hypothesis: We require the homogeneous version of the nonhomogeneous qua- dratic algebra to be the skew group algebra (semidirect product algebra) of a finite group acting on a Koszul algebra, obtaining conditions for the Poincar´e- Birkhoff-Witt property over (nonsemisimple) group algebras. We prove our main results by exploiting a double complex adapted from Guccione, Guccione, and Valqui (formed from a Koszul complex and a resolution of the group), giv- ing a practical way to analyze Hochschild cohomology and deformations of skew group algebras in positive characteristic. We apply these conditions to graded Hecke algebras and Drinfeld orbifold algebras (including rational Cherednik al- gebras and symplectic reflection algebras) in arbitrary characteristic, with spe- cial interest in the case when the characteristic of the underlying field divides the order of the acting group. 1. Introduction Poincar´e-Birkhoff-Witt properties are used to isolate convenient canonical bases of algebras, identify algebras with formal deformations, and depict associated graded structures of algebras explicitly. They reveal how a set of generators for an ideal of relations defining an algebra may capture the graded structure im- plied by the entire ideal. In 1996, Braverman and Gaitsgory [5] gave necessary and sufficient conditions for a nonhomogeneous quadratic algebra to satisfy the Poincar´e-Birkhoff-Witt property when a homogeneous version is Koszul. Poincar´e- Birkhoff-Witt theorems are often established by inspection of monomial orderings, noncommutative Grobner bases, or other methods from noncommutative compu- tational algebra. In contrast, Braverman and Gaitsgory used Hochschild coho- mology to streamline arguments. They identified conditions that often provide an Date: September 25, 2012. Key Words: Koszul algebras, skew group algebras, Hochschild cohomology, Drinfeld orbifold algebras; AMS Subject Classification Codes: 16S37, 16E40, 16S80, 16S35. The first author was partially supported by NSF grants #DMS-0800951 and #DMS-1101177. The second author was partially supported by NSF grants #DMS-0800832 and #DMS-1101399. 1 2 A. V. SHEPLER AND S. WITHERSPOON elegant alternative to direct application of Bergman's Diamond Lemma [3] (which can be tedious) for determining when a Poincar´e-Birkhoff-Witt property holds. Thus their theory has been used widely to investigate various algebras in many settings. (See [22] and references therein.) Etingof and Ginzburg [8] noted that the conditions for a Poincar´e-Birkhoff- Witt property developed by Braverman and Gaitsgory [5] may be generalized by replacing the ground field k by any semisimple ring (for example, a group ring kG) using the theory of Beilinson, Ginzburg, and Soergel [4] of Koszul rings. They exploited this generalization in investigations of symplectic reflection algebras. Later others applied it even more generally, for example, Halbout, Oudom, and Tang [16]. Yet the theorem of Braverman and Gaitsgory and the generalization observed by Etingof and Ginzburg [8] do not apply in some interesting settings, in particular, when replacing the ground field by a ring that is not semisimple. For example, a theory of symplectic reflection algebras and graded Hecke algebras over fields of arbitrary characteristic would include exploration of quotient algebras over group rings kG when the characteristic of the field k divides the order of the finite group G. We show in Theorem 5.4 that the conditions of Braverman and Gaitsgory for a Poincar´e-Birkhoff-Witt property hold over any finite group algebra (including nonsemisimple group algebras). We establish necessary and sufficient conditions for a filtered quadratic algebra over kG to satisfy the PBW property when its homogeneous version is a skew group algebra (semidirect product algebra) formed from a finite group G acting on a Koszul algebra S. The proof uses deformation theory and an explicit bimodule resolution X r. This resolution, defined in Section 4 and adapted from Guccione, Guccione, and Valqui [15], is comprised of both the Koszul resolution for S and the bar resolution for kG. Semisimplicity does not play a crucial role. However, there are potentially many graded deformations of the skew group algebra S#G in positive characteristic that this theory does not identify, corresponding to components of the resolution X r that are not fully explored here. Other methods have been employed in the modular setting in some special cases; for example, see [14] for representation-theoretic techniques and [26] for application of the Diamond Lemma. We compare our techniques with those in the nonmodular setting: In charac- teristic zero, the group algebra kG has trivial cohomology and thus the Koszul resolution of S is sufficient for analyzing the Hochschild cohomology of S#G (by a spectral sequence argument, see [9]). But in positive characteristic, the group algebra may exhibit nontrivial cohomology influencing the Hochschild cohomology of S#G, and one seeks a more sophisticated replacement for the Koszul resolution of S which nevertheless remains practical for determining concrete results (such as establishing a Poincar´e-Birkhoff-Witt property). In this article, we highlight the complex X r as a first tool in understanding the Hochschild cohomology of S#G in arbitrary characteristic and its deformation theory. Our general construction of POINCAR ´E-BIRKHOFF-WITT THEOREM 3 X r in Section 4 takes as input arbitrary resolutions of S and of kG satisfying some natural conditions; the Koszul resolution of S (in case S is Koszul) and the bar resolution of kG are but two of the potentially many useful choices. Our approach may lead to better understanding of many algebras currently of interest, such as rational Cherednik algebras in positive characteristic. In Section 6 we illustrate these applications by focusing on a collection of such algebras, showing how known results in the modular setting may now be obtained directly from the methods of Braverman and Gaitsgory. There are several papers containing generalizations of Koszul algebras for var- ious purposes, those most relevant to our setting being [13, 19, 29]. We will not need these here, however, as the known properties of Koszul algebras over fields are sufficient to obtain our results. It would be interesting to determine whether it is possible to generalize the theory of Braverman and Gaitsgory more directly by using some equivalent definition of Koszul rings over (nonsemisimple) rings involving a bimodule complex (cf. [5, Proposition A.2(b)]). Throughout this article, k denotes a field (of arbitrary characteristic) and ten- sor symbols without subscript denote tensor product over k: ⊗ = ⊗k. (Tensor products over other rings will always be indicated). We assume all k-algebras have unity and all modules are left modules, unless otherwise specified. We use the notation N = Z≥0, the nonnegative integers. 2. Filtered and homogeneous quadratic algebras Quadratic algebras and their variations traditionally arise from taking a free al- gebra modulo a set of (nonhomogeneous) relations of degree two. Examples include symmetric algebras, commutative polynomial rings, skew/quantum polynomial rings, Weyl algebras, Clifford and exterior algebras, and enveloping algebras of fi- nite dimensional Lie algebras. But quadratic shape arises from any N-graded alge- bra modulo an ideal generated in filtered degree two. Specifically, let T =Lm T m be an N-graded k-algebra. (We have in mind the tensor algebra (i.e., free algebra) Tk(V ) of a k-vector space V , or the tensor algebra TB(U) of a B-bimodule U over a k-algebra B, for example, a group algebra.) Let F m(T ) = T 0 ⊕ T 1 ⊕ . . . ⊕ T m be the m-th filtered component of T and fix a two-sided ideal I of T . A quotient T /I is called a filtered quadratic algebra (or a nonhomogeneous quadratic algebra) if I can be generated in filtered degree 2, i.e., if I = hP i for some P ⊂ F 2(T ). In this case, P is called a set of filtered quadratic relations. Quadratic algebras are filtered algebras, with m-th filtered component F m(T /I) = (F m(T ) + I)/I induced from that on T . When a set of filtered quadratic relations R resides exactly in the second graded component (homogeneous of degree 2), i.e., R ⊂ T 2, we call the quotient T /hRi a homogeneous quadratic algebra. Homogeneous quadratic algebras are not only filtered, but also graded: T /I = ⊕m(T m + I)/I for I = hRi. 4 A. V. SHEPLER AND S. WITHERSPOON One may easily associate to every filtered quadratic algebra T /I two different graded versions. We might cross out lower order terms in a generating set of relations for the algebra, or, instead, cross out lower order terms in each element of the entire ideal of relations. The Poincar´e-Birkhoff-Witt conditions are precisely those under which these two graded versions of the original algebra coincide, as we now recall. On one hand, we may simply ignore those parts of each defining relation in P of degree less than two to obtain a simplified version of the original filtered algebra which is homogeneous quadratic. Formally, we let π denote projection onto the second graded component of the tensor algebra: π : T → T 2. If P is a set of filtered quadratic relations defining the algebra T /I, then π(P ) defines a graded quadratic algebra T /hπ(P )i , called the homogeneous quadratic algebra determined by P (or the induced homogeneous quadratic algebra). Although this construction depends on the choice of generators P for I, often a choice can be made to capture the degree two aspect of the entire ideal of relations. On the other hand, we may consider the traditional associated graded alge- bra gr(cid:0)T /I(cid:1) = Mm Fm/Fm−1 , where Fm = F m(T /I). This graded version of the original algebra does not de- pend on the choice of generators P of the ideal I of relations. We realize the associated graded algebra concretely also as a homogeneous version of the original filtered quadratic algebra by projecting each element in the ideal I onto its leading homogeneous part and taking the quotient by the resulting ideal: The associated homogeneous quadratic algebra (also called the leading homogeneous algebra) is defined as T /hLH(I)i , where LH(I) = {LH(f ) : f ∈ I} and LH(f ) picks off the highest homogeneous i=1 fi with each fi in T i and fd nonzero, LH(f ) = fd, and LH(0) = 0.) This associated homogeneous quadratic algebra is isomorphic as a graded k-algebra to the associated graded algebra (see Li [18, Theorem 3.2]): part of f in T . (For f =Pd gr(cid:0)T /I(cid:1) ∼= T /hLH(I)i . We say the original filtered quadratic algebra has Poincar´e-Birkhoff-Witt type when the associated homogeneous quadratic algebra and the homogeneous qua- dratic algebra determined by P coincide, and thus both give the associated graded algebra. (This terminology arises in analogy with the original PBW Theorem for universal enveloping algebras of Lie algebras.) More precisely, let φ : T → gr(T /I) POINCAR ´E-BIRKHOFF-WITT THEOREM 5 be the natural k-algebra epimorphism. Then (again, see Li [18, Theorem 3.2]) Ker φ = hLH(I)i , which contains hπ(P )i when I = hP i. Thus, a natural surjection p always arises from the homogeneous quadratic algebra determined by P to the associated graded algebra of the filtered quadratic algebra: p : T /hπ(P )i → gr(cid:0)T /hP i(cid:1) . A filtered quadratic algebra exhibits PBW type (with respect to P ) exactly when it can be written as T /hP i for some set of filtered quadratic relations P with p an isomorphism of graded algebras, i.e., when hπ(P )i = Ker φ = hLH(I)i and thus the homogeneous versions of the filtered quadratic algebra all coincide: In this case, we say that P is a set of PBW generating relations. T /hπ(P )i ∼= gr(cid:0)T /hP i(cid:1) ∼= T /hLHhP ii . The definition of PBW type depends on P , as we see in Example 2.3 below. But if T 0 = k, we may require (without loss of generality) that P be a k-subspace, in which case a set of PBW generating relations P is unique. In fact, a set of PBW generating relations is always unique up to additive closure over the degree zero component of T , as we explain in the next proposition: Proposition 2.1. PBW filtered quadratic algebras have unique PBW filtered qua- dratic relations up to addition and multiplication by degree zero elements: If P, P ′ are each PBW filtered quadratic relations defining the same filtered quadratic alge- bra T /I (that is, hP i = I = hP ′i), then P and P ′ generate the same T 0-bimodule. Thus, if both P and P ′ are closed under addition and under multiplication by degree zero elements in T , then P = P ′. Proof. We check that for any set P of PBW relations, (2.2) hP i ∩ F 1(T ) = {0} and hP i ∩ F 2(T ) = T 0P T 0, where T 0P T 0 is the T 0-bimodule generated by P . The first claim is immediate: Since π(P ) ⊂ T 2, the algebra T /hπ(P )i contains F 1(T ) as a subspace; if hP i ∩ F 1(T ) 6= {0}, then gr(T /hP i) does not contain F 1(T ) as a subspace, and so p cannot be an isomorphism. Now let x be any element of hP i ∩ F 2(T ) with x 6∈ T 0P T 0. Note that for all y ∈ T 0P T 0, π(x) 6= π(y). Otherwise, some nonzero x − y would lie in F 1(T ) (as π(x − y) = 0), implying that hP i ∩ F 1(T ) is nonzero. Hence π(x) 6∈ π(T 0P T 0) = T 0π(P )T 0. But this contradicts the PBW property, which implies that π(x) ∈ hLHhP ii = hπ(P )i (a homogeneous ideal) and hence π(x) ∈ T 0π(P )T 0. 6 A. V. SHEPLER AND S. WITHERSPOON Lastly, if P and P ′ are both PBW filtered quadratic relations generating the ideal I, then T 0P T 0 = I ∩ F 2(T ) = T 0P ′T 0, and P and P ′ generate the same T 0-bimodule. (cid:3) Example 2.3. We give an example of a filtered quadratic algebra that exhibits PBW type with respect to one generating set of relations but not another. Let V = Spank{x, y} and let I be the two-sided ideal in the tensor algebra T = T (V ) (over k) generated by P = Spank{xy − x, yx − y} (we suppress tensor signs in T ). The filtered quadratic algebra T (V )/I = T (V )/hP i is not of PBW type with respect to P . Indeed, the graded summand of the associated graded algebra of degree 2 has dimension 0 over k since a quick calculation verifies that x2 − x and y2 − y both lie in the ideal hP i, implying that all of T 2(V ) = V ⊗ V projects to zero in T (V )/hP i after passing to the associated graded algebra. But π(P ) = Spank{xy, yx} ⊂ V ⊗ V defines a homogeneous quadratic algebra T (V )/hxy, yxi whose graded summand of degree 2 has dimension 2 over k (with basis {x2, y2}). Note that the filtered quadratic algebra T (V )/I exhibits PBW type with respect to a different set of generators for the ideal I. If we extend P to P ′ = Spank{xy − x, yx − y, x2 − x, y2 − y} , then T (V )/I is of PBW type with respect to P ′, as its associated graded algebra is isomorphic to T (V )/hxy, yx, x2, y2i ∼= T (V )/hπ(P ′)i . In Section 5 we will analyze filtered and homogeneous quadratic algebras when T is a free algebra. Traditional quadratic algebras arise as quotients of a tensor al- gebra T of a k-vector space V . We expand this view and consider (more generally) bimodules V over an arbitrary k-algebra B in order to include constructions of algebras of quadratic shape naturally appearing in other settings. First we recall Hochschild cohomology and deformations in the next section, and construct the needed resolution X r in Section 4. 3. Deformations and Koszul algebras Let A be a k-algebra. Let M be an A-bimodule, equivalently, an Ae-module, where Ae = A ⊗ Aop, the enveloping algebra of A. As k is a field, the Hochschild cohomology of M is HHn(A, M) = Extn Ae(A, M). This cohomology can be examined explicitly using the bar resolution, a free reso- lution of the Ae-module A: (3.1) · · · δ3−→ A ⊗ A ⊗ A ⊗ A δ2−→ A ⊗ A ⊗ A δ1−→ A ⊗ A δ0−→ A → 0 POINCAR ´E-BIRKHOFF-WITT THEOREM 7 where (3.2) δn(a0 ⊗ · · · ⊗ an+1) = nXi=0 (−1)ia0 ⊗ · · · ⊗ aiai+1 ⊗ · · · ⊗ an+1 for all n ≥ 0 and a0, . . . , an+1 in A. HHn(A, A). If M = A, we abbreviate, HHn(A) := A deformation of A over k[t] is an associative k[t]-algebra with underlying vector space A[t] and multiplication determined by (3.3) a1 ∗ a2 = a1a2 + µ1(a1 ⊗ a2)t + µ2(a1 ⊗ a2)t2 + · · · where a1a2 is the product of a1 and a2 in A and each µk : A ⊗ A → A is a k-linear map. (Only finitely many terms are nonzero for each pair a1, a2 in the above expansion.) We record some needed properties of µ1 and µ2. Note that Hom k(A ⊗ A, A) ∼= Hom Ae(A⊗A⊗A⊗A, A) since the Ae-module A⊗A⊗A⊗A is (tensor) induced from the k-module A⊗A, and we identify µ1 with a 2-cochain on the bar resolution (3.1). Associativity of ∗ implies that µ1 is a Hochschild 2-cocycle, i.e., that (3.4) a1µ1(a2 ⊗ a3) + µ1(a1 ⊗ a2a3) = µ1(a1a2 ⊗ a3) + µ1(a1 ⊗ a2)a3 for all a1, a2, a3 ∈ A, or, equivalently, that δ∗ 3(µ1) = 0: One need only expand each side of the equation a1 ∗ (a2 ∗ a3) = (a1 ∗ a2) ∗ a3 and compare coefficients of t. Comparing coefficients of t2 instead yields δ∗ 3(µ2)(a1 ⊗ a2 ⊗ a3) = µ1(a1 ⊗ µ1(a2 ⊗ a3)) − µ1(µ1(a1 ⊗ a2) ⊗ a3) for all a1, a2, a3 ∈ A. Thus we consider µ2 to be a cochain on the bar resolution whose coboundary is given as above. Generally, for all i ≥ 1, i−1Xj=1 (3.5) δ∗ 3(µi)(a1 ⊗ a2 ⊗ a3) = µj(µi−j(a1 ⊗ a2) ⊗ a3) − µj(a1 ⊗ µi−j(a2 ⊗ a3)). We call the right side of the last equation the (i − 1)-th obstruction. Now we recall graded deformations. Assume that the algebra A is N-graded. Let t be an indeterminate and extend the grading to A[t] by assigning deg t = 1. A graded deformation of A over k[t] is a deformation At of A which is also graded; each map µj : A ⊗ A → A is necessarily homogeneous of degree −j in this case. An i-th level graded deformation of A is a graded associative k[t]/(ti+1)- algebra Ai whose underlying vector space is A[t]/(ti+1) and whose multiplication is determined by a1 ∗ a2 = a1a2 + µ1(a1 ⊗ a2)t + µ2(a1 ⊗ a2)t2 + · · · + µi(a1 ⊗ a2)ti for some maps µj : A ⊗ A → A extended to be linear over k[t]/(ti+1). We call µj the j-th multiplication map of the deformation Ai; note that it is homogeneous of degree −j as Ai is graded. We say that an (i + 1)-st level graded deformation 8 A. V. SHEPLER AND S. WITHERSPOON Ai+1 of A extends (or lifts) an i-th level graded deformation Ai of A if the j-th multiplication maps agree for all j ≤ i. For any N-graded A-bimodule M, the bar resolution (3.1) induces a grading on Hochschild cohomology HHn(A, M) in the following way: Let Hom i k(A⊗n, M) be the space of all homogeneous k-linear maps from A⊗n to M of degree i, i.e., mapping the degree j component of A⊗n (for any j) to the degree j + i component of M, where deg(a1 ⊗ · · · ⊗ an) in A⊗n is deg a1 + . . . + deg an for a1, . . . , an homogeneous in A. The differential on the bar complex (3.1) has degree zero, so the spaces of coboundaries and cocycles inherit this grading. We denote the resulting i-th graded component of HHn(A, M) by HHn,i(A, M), so that HHn(A, M) =Mi HHn,i(A, M) . The following proposition is a consequence of [5, Proposition 1.5]. Proposition 3.6. First level graded deformations of the algebra A define ele- ments of the Hochschild cohomology space HH2,−1(A). Two such deformations are (graded) isomorphic if, and only if, the corresponding cocycles are cohomologous. All obstructions to lifting an (i − 1)-th level graded deformation to the next level lie in HH3,−i(A); an (i − 1)-th level deformation lifts to the i-th level if and only if its (i − 1)-th obstruction cocycle is zero in cohomology. We are most interested in deformations of skew group algebras formed from Koszul algebras with group actions, and we recall Koszul algebras (over the field k) next. Let V be a finite dimensional vector space. Let Tk(V ) = k ⊕ V ⊕ (V ⊗ V ) ⊕ (V ⊗ V ⊗ V ) ⊕ . . . , Let R be a subspace of T 2 the tensor algebra of V over k with i-th graded piece T i k (V ) = k. k (V ), and let S = Tk(V )/hRi be the corresponding homogeneous quadratic algebra. Let K 0(S) = k, K 1(S) = V , and for each n ≥ 2, define k(V ) := V ⊗i and T 0 K n(S) = Set n−2\j=0 (V ⊗j ⊗ R ⊗ V ⊗(n−2−j)). eK n(S) = S ⊗ K n(S) ⊗ S . resolution (3.1). We apply the differential δn, defined in (3.2), to an element in Then eK 0(S) ∼= S ⊗ S, eK 1(S) = S ⊗ V ⊗ S, eK 2(S) = S ⊗ R ⊗ S, and, in general, eK n(S) embeds as an Se-submodule into S ⊗(n+2), the n-th component of the bar eK n(S) and note that the terms indexed by i = 1, 2, . . . , n − 1 vanish (due to the factors in the space of relations R). The remaining terms are clearly in eK n−1(S). POINCAR ´E-BIRKHOFF-WITT THEOREM 9 We may thus consider eK r(S) to be a complex with differential d (restricted from the bar complex, dn := δn eK n(S)), called the Koszul complex: (3.7) By definition, S is a Koszul algebra if the related complex ¯K r(S), defined by ¯K n(S) := S ⊗ K n(S), is a resolution of the S-module k on which each element . . . d3−→ eK 2(S) d2−→ eK 1(S) d1−→ eK 0(S) d0−→ S → 0 . of V acts as 0. (Note that ¯K n(S) ∼= eK n(S) ⊗S k; differentials are dn ⊗ id .) It is well-known that S is a Koszul algebra if, and only if, eK r(S) is a resolution of the Se-module S. (See e.g. [5, Proposition A.2] or [17].) 4. Resolutions for skew group algebras In this section, we consider any finite group G and any k-algebra S upon which G acts by automorphisms. We work in arbitrary characteristic to develop techniques helpful in the modular setting. We explain how to construct a resolution of the skew group algebra A = S#G from resolutions of S and of kG, generalizing a construction of Guccione, Guccione, and Valqui [15, §4.1]. It should be compared to the double complexes in [24, §2.2] and more generally in [27, Corollary 3.4] that are used to build spectral sequences. We apply the resolution in the next section to generalize the result of Braverman and Gaitsgory [5, Theorem 4.1] on the Poincar´e-Birkhoff-Witt property. Some of our assumptions in this section will seem restrictive, however the large class of examples to which we generalize their result in Theorem 5.4 all satisfy the assumptions, as do the modular versions of Drinfeld orbifold algebras, graded Hecke algebras, and symplectic reflection algebras in Section 6. In characteristic zero, HH r(S#G) ∼= HH r(S, S#G)G as a consequence of a spec- tral sequence argument, and the latter may be obtained from a resolution of S. (In the special case that S is Koszul, one thus has a convenient resolution at hand for analyzing the cohomology of S#G and its deformation theory.) But in the modular setting, when the characteristic of k divides the order of G, the spectral sequence no longer merely produces G-invariants (as the group may have non- trivial cohomology) and thus this technique for simplifying the cohomology of the skew group algebra fails. The resolution of S#G constructed in this section re- tains some of the flavor of the Koszul resolution of S, so as to allow similar Koszul techniques to be applied in the modular setting. We first recall the definition of a skew group algebra. The skew group algebra S#G is a semidirect product algebra: It is the k-vector space S ⊗ kG together with multiplication given by (r ⊗ g)(s ⊗ h) = r( gs) ⊗ gh for all r, s in S and g, h in G, where gs is the image of s under the automorphism g. (We are most interested in skew group algebras arising from the linear action of G on a finite dimensional vector space V and its induced actions on tensor algebras, symmetric algebras, Koszul algebras, and homogeneous quadratic algebras all generated by V .) 10 A. V. SHEPLER AND S. WITHERSPOON We need the notion of Yetter-Drinfeld modules: A vector space V is a Yetter- Drinfeld module over G if V is G-graded, that is, V = ⊕g∈GVg, and V is a kG-module with h(Vg) = Vhgh−1 for all g, h ∈ G. Any kG-module V is trivially a Yetter-Drinfeld module by letting V1 = V and Vg = 0 for all nonidentity group elements g. Similarly, any algebra S on which G acts by automorphisms is a Yetter-Drinfeld module in this way. Alternatively, the group algebra kG itself may be considered to be a Yetter-Drinfeld module by letting the g-component be all scalar multiples of g, for each g ∈ G, and by letting G act on kG by conjugation. The skew group algebra A = S#G is a Yetter-Drinfeld module by combining these two structures: A =Mg∈G Ag where Ag = Sg and G acts on A by conjugation (an inner action). A morphism of Yetter-Drinfeld modules is a kG-module homomorphism that preserves the G- grading. We use a braiding on the category of Yetter-Drinfeld modules: Given any two Yetter-Drinfeld modules V, W over G, their tensor product V ⊗ W is again a Yetter-Drinfeld module with (V ⊗ W )g =Mxy=g (Vx ⊗ Wy) and the usual G-action on a tensor product: g(v ⊗ w) = gv ⊗ gw for all g ∈ G, v ∈ V , w ∈ W . There is an isomorphism of Yetter-Drinfeld modules cV,W : V ⊗ W → W ⊗ V given by cV,W (v ⊗ w) = gw ⊗ v for all v ∈ Vg, w ∈ W , g ∈ G, called the braiding. We will now construct an Ae-free resolution of A = S#G from a (kG)e-free resolution of kG and an Se-free resolution of S. First let · · · → C1 → C0 → kG → 0 be a (kG)e-free resolution of kG. We assume that each Ci is G-graded with grading preserved by the bimodule structure, that is g((Ci)h)l = (Ci)ghl for all g, h, l in G. Note this implies that Ci is a Yetter-Drinfeld module where gc = gcg−1 for all g in G and c in Ci. We also assume that the differentials preserve the G-grading. Assume that as a free (kG)e-module, Ci = kG ⊗ C ′ i ⊗ kG for a G-graded vector space C ′ i whose G-grading induces that on Ci under the usual tensor product of G-graded vector spaces. For example, the bar resolution of kG satisfies all these properties. Another instance can be found in Example 4.6 below. Next let · · · → D1 → D0 → S → 0 be an Se-free resolution of S consisting of left kG-modules for which the differ- entials are kG-module homomorphisms, and the left actions of S and of kG are compatible in the sense that they induce a left action of A = S#G. We consider POINCAR ´E-BIRKHOFF-WITT THEOREM 11 each Di to be a Yetter-Drinfeld module by setting it all in the component of the identity. Write the free Se-module Di as S ⊗ D′ i. For example, the bar resolution of S satisfies these properties under the usual action of G on tensor products. If S is a Koszul algebra on which G acts by graded automorphisms, the Koszul resolution of S also satisfies these properties. i ⊗ S, for a vector space D′ We now induce both C r and D r to A by tensoring with A in each degree. We induce C r from kG to A on the left: Since A is free as a right kG-module under multiplication, and A ⊗kG kG ∼= A, we obtain an exact sequence of A ⊗ (kG)op- modules, · · · → A ⊗kG C1 → A ⊗kG C0 → A → 0. Similarly, we induce D r from S to A on the right: Since A is free as a left S-module, and S ⊗S A ∼= A, · · · → D1 ⊗S A → D0 ⊗S A → A → 0 is an exact sequence of S ⊗ Aop-modules. We extend the actions on the modules in each of these two sequences so that they become sequences of Ae-modules. This will allow us to take their tensor product over A. We extend the right kG-module structure on A ⊗kG C r to a right A-module structure by using the braiding to define a right action of S: For all a ∈ A, g ∈ G, x ∈ (Ci)g, s ∈ S, we set (a ⊗ x) · s := a(gs) ⊗ x. We combine this right action of S with the right action of kG; under our assump- tions, this results in a right action of A on A ⊗kG Ci. To see that it is well-defined, note that if h ∈ G, then (ah ⊗ x) · s = ah(gs) ⊗ x = a(hgs) ⊗ hx = (a ⊗ hx) · s. Thus A ⊗kG Ci is an A-bimodule and the action commutes with the differentials by the assumption that the differentials on C r preserve the G-grading. We extend the left S-module structure on Di ⊗S A to a left A-module structure by defining a left action of kG: g · (y ⊗ a) := gy ⊗ ga for all g ∈ G, y ∈ Di, a ∈ A. It is well-defined since gs = ( gs)g for all s ∈ S, and indeed gives a left action of A on Di ⊗S A. Again this action commutes with the differentials, by our assumption that the differentials on D r are kG-module homomorphisms. We use the A-bimodule structures on A⊗kG C r and on D r⊗S A defined above and consider each as a complex of Ae-modules. (Note that we have not assumed they consist of projective Ae-modules.) We take their tensor product over A, setting X r, r := (A ⊗kG C r) ⊗A (D r ⊗S A), that is, for all i, j ≥ 0, (4.1) Xi,j := (A ⊗kG Ci) ⊗A (Dj ⊗S A), with horizontal and vertical differentials dh i,j : Xi,j → Xi−1,j and dv i,j : Xi,j → Xi,j−1 12 A. V. SHEPLER AND S. WITHERSPOON given by dh X r, r, i.e., the complex i,j := di ⊗ id and dv i,j := (−1)i id ⊗ dj. Let X r be the total complex of (4.2) · · · → X2 → X1 → X0 → A → 0 with Xn = ⊕i+j=nXi,j. Theorem 4.3. Let S be a k-algebra with action of a finite group G by automor- phisms and set A = S#G. Let X r be the complex defined in (4.2) from factors Ci = kG ⊗ C ′ i ⊗ S as above. Then X r is a free resolu- tion of the Ae-module A, and for each i, j, the Ae-module Xi,j is isomorphic to A ⊗ C ′ i ⊗ kG and Di = S ⊗ D′ i ⊗ D′ j ⊗ A. In the case that C r is the (normalized) bar resolution of kG and D r is the Koszul resolution of S(V ), our resolution X r is precisely that in [15, §4]. Proof. We first check that for each i, j, the Ae-module Xi,j is free. By construction, Xi,j = (A ⊗kG kG ⊗ C ′ i ⊗ kG) ⊗A (S ⊗ D′ j ⊗ S ⊗S A) ∼= (A ⊗ C ′ i ⊗ kG) ⊗A (S ⊗ D′ i ⊗ D′ j ⊗ A). j ⊗ A as an Ae-module. To see We claim that this is isomorphic to A ⊗ C ′ this, first define a map as follows: (4.4) (A ⊗ C ′ i ⊗ kG) × (S ⊗ D′ j ⊗ A) → A ⊗ C ′ i ⊗ D′ j ⊗ A (a ⊗ x ⊗ g, s ⊗ y ⊗ b) 7→ a(hgs) ⊗ x ⊗ gy ⊗ g where x ∈ (C ′ A-balanced: If r ∈ S, ℓ ∈ G, then on the one hand, i)h. This map is bilinear by its definition, and we check that it is ((a ⊗ x ⊗ g) · (rℓ), s ⊗ y ⊗ b) = (a(hgr) ⊗ x ⊗ gℓ, s ⊗ y ⊗ b) 7→ a(hgr)(hgℓs) ⊗ x ⊗ gℓy ⊗ gℓb, while on the other hand, ((a ⊗ x ⊗ g, (rℓ) · (s ⊗ y ⊗ b)) = (a ⊗ x ⊗ g, r(ℓs) ⊗ ℓy ⊗ ℓb) 7→ a(hgr)(hgℓs) ⊗ x ⊗ gℓy ⊗ gℓb). Therefore there is an induced map i ⊗ D′ It is straightforward to verify that an inverse map is given by i ⊗ kG) ⊗A (S ⊗ D′ j ⊗ A) → A ⊗ C ′ (A ⊗ C ′ j ⊗ A. (4.5) a ⊗ x ⊗ y ⊗ b 7→ (a ⊗ x ⊗ 1) ⊗ (1 ⊗ y ⊗ b). Therefore the two spaces are isomorphic as claimed. We wish to apply the Kunneth Theorem, and to that end we check that each term in the complex A ⊗kG C r consists of free right A-modules, and that the image POINCAR ´E-BIRKHOFF-WITT THEOREM 13 of each differential in the complex is projective as a right A-module. This may be proved inductively, starting on one end of the complex · · · f2−→ A ⊗kG C1 f1−→ A ⊗kG C0 f0−→ A → 0. To see directly that each A ⊗kG Ci is free as a right A-module, write A ⊗kG Ci = A ⊗kG (kG ⊗ C ′ i ⊗ kG. i ⊗ kG) ∼= A ⊗ C ′ Choose a k-linear finite basis {xm 1 ≤ m ≤ ri} of C ′ homogeneous with respect to the G-grading, and ri = dimk C ′ works if the C ′ to assume that the C ′ A ⊗kG Ci as a right A-module is i for which each xm is i. (A similar idea i are infinite dimensional, however since G is finite, it is reasonable i are finite dimensional.) Then a set of free generators of {g ⊗ xm ⊗ 1 g ∈ G, 1 ≤ m ≤ ri}. Indeed, if we fix g in G and xm as above, with xm in the ℓ-component (ℓ ∈ G), then for each s in S and h in G, (g ⊗ xm ⊗ 1) · (ℓ−1g−1 s)h = sg ⊗ xm ⊗ h, and consequently the full subspace Sg ⊗ xm ⊗ kG is generated from this single element. It also follows that they are independent. Since A is right A-projective, and f0 is surjective, the map f0 splits so that Ker f0 is a direct summand of A ⊗kG C0 as a right A-module. Therefore Ker f0 = Im f1 is right A-projective. Repeat the argument with A ⊗kG C0 replaced by A ⊗kG C1 and A replaced by Im f1, and so on, to complete the check. The Kunneth Theorem [28, Theorem 3.6.3] then gives for each n an exact se- quence 0 −→ Mi+j=n Hi(A ⊗kG C r) ⊗A Hj(D r ⊗S A) −→ Hn((A ⊗kG C r) ⊗A (D r ⊗S A)) −→ Mi+j=n−1 TorA 1 (Hi(A ⊗kG C r), Hj(D r ⊗S A)) → 0. Now A⊗kG C r and D r⊗S A are exact other than in degree 0, where their homologies are each A. Thus Hj(A ⊗kG C r) = Hi(D r ⊗S A) = 0 unless i = j = 0. The Tor term for i = j = 0 is also zero as TorA 1 (A, A) = 0 (since A is flat over A). Thus Hn((A ⊗kG C r) ⊗A (D r ⊗S A)) = 0 for all n > 0 and H0((A ⊗kG C r) ⊗A (D r ⊗S A)) ∼= H0(A ⊗kG C r) ⊗A H0(D r ⊗S A) ∼= A ⊗A A ∼= A. Thus X r is an Ae-free resolution of A. (cid:3) 14 A. V. SHEPLER AND S. WITHERSPOON The resolution X r, being more general than the one given in [15, §4], has an advantage: One may use any convenient resolution of the group algebra kG in the construction. The resolution in [15] uses the (normalized) bar resolution of kG, resulting in a potentially larger complex X r. In the example below, we show that X r may be quite tractable when a smaller resolution of kG is chosen. Example 4.6. Let G be a cyclic group of prime order p, generated by g. Let k be a field of characteristic p. Let V = k2 with basis v1, v2. Let g act as the matrix (cid:18) 1 1 0 1 (cid:19) on the ordered basis v1, v2. Let S = Sk(V ), the symmetric algebra, and let D r : 0 → S ⊗V2 V ⊗ S → S ⊗V1 V ⊗ S → S ⊗ S → S → 0 be the Koszul resolution of S, where we identify V1 V with V and V2 V with R = {v ⊗ w − w ⊗ v v, w ∈ V }. The differentials commute with the G-action. Let · · · C r : v·−→ kG ⊗ kG u·−→ kG ⊗ kG v·−→ kG ⊗ kG u·−→ kG ⊗ kG m−→ kG → 0 where u = g⊗1−1⊗g, v = gp−1⊗1+gp−2⊗g+· · ·+1⊗gp−1, and m is multiplication. Then C r is a (kG)e-free resolution of kG; exactness may be verified by constructing an explicit contracting homotopy. We consider kG ⊗ kG in even degrees to be a Yetter-Drinfeld module in the usual way: (kG ⊗ kG)gi = Spank{x ⊗ y xy = gi}. But in odd degrees let (kG ⊗ kG)gi = Spank{x ⊗ y xy = gi−1}. This will ensure that the differentials preserve the G-grading. By Theorem 4.3, X r, r = (A ⊗kG C r) ⊗A (D r ⊗S A) yields a free Ae-resolution X r (the total complex) of A. By our earlier analysis, for all i ≥ 0 and 0 ≤ j ≤ 2, and the differentials are d = dh i,j = di ⊗ id + (−1)iid ⊗ dj. Xi,j i,j + dv ∼= A ⊗Vj V ⊗ A For our applications to Koszul algebras, we will need chain maps between the resolution X r and the bar resolution of A with the properties stated in the next lemma. Lemma 4.7. Let S be a finitely generated graded Koszul algebra over k on which a finite group G acts by graded automorphisms. Let C r be the bar resolution of kG, let D r be the Koszul resolution of S, let A = S#G, and let X r be as in (4.2). Then there exist chain maps φ r : X r → A⊗( r+2) and ψ r : A⊗( r+2) → X r of degree 0 for which ψnφn is the identity map on the subspace X0,n of Xn for each n ≥ 0. A chain map was given explicitly in case S = Sk(V ) by Guccione, Guccione, and Valqui [15, §4.2]. POINCAR ´E-BIRKHOFF-WITT THEOREM 15 Proof. Both X r and A⊗( r+2) are free resolutions of A as an Ae-module whose differ- entials are maps of degree 0. We first argue inductively that there exists a chain map φn : Xn → A⊗(n+2) of degree 0 for which φnX0,n is induced by the standard embedding of the Koszul complex into the bar complex. Suppose S is generated by a finite dimensional k-vector space V with quadratic relations R: S = Tk(V )/hRi (see Section 3). Define φ0 = id ⊗id = ψ0, the identity map from A ⊗ A to itself. Consider X0, r as a subcomplex (not necessarily acyclic) of X r. An inductive argument shows that we may define φ r so that when restricted to X0, r it corresponds to the standard embedding of the Koszul complex into the bar complex: For n = 1, this is the embedding of A ⊗ V ⊗ A into A ⊗ A ⊗ A, and ∼= (A ⊗ V ⊗ A) ⊕ (A ⊗ kG ⊗ A) may be one checks that φ1 on X1 = X0,1 ⊕ X1,0 defined by φ1(1 ⊗ v ⊗ 1) = 1 ⊗ v ⊗ 1 and φ1(1 ⊗ g ⊗ 1) = 1 ⊗ g ⊗ 1 for all v ∈ V , g ∈ G. For n ≥ 2, (4.8) X0,n ∼= A ⊗ n−2\i=0 V ⊗i ⊗ R ⊗ V ⊗(n−i−2)! ⊗ A , i ⊗ D′ a free Ae-submodule of A⊗(n+2) by its definition. For each i, j with i + j = n, choose a basis of the vector space C ′ j, whose elements are necessarily of degree j. By construction, after applying φn−1dn to these basis elements, we obtain elements of degree j in the kernel of δn−1, which is the image of δn. Choose corresponding elements in the inverse image of Im(δn) to define φn. If we start with an element in X0,n, we may choose its canonical image in A⊗(n+2) (see (4.8)). Elements of Xi,j (i > 0) have different degree, so their images under φn may be chosen independently of those of X0,n. ∼= A ⊗ V ⊗ A and X1,0 Now we show inductively that each ψn may be chosen to be a degree 0 map for which ψnφn is the identity map on X0,n. In degree 0, this is true as φ0, ψ0 are ∼= A ⊗ kG ⊗ A. Note identity maps. In degree 1, X0,1 that V ⊕ kG is a direct summand of A as a vector space. We may therefore define ψ1(1 ⊗ v ⊗ 1) = 1 ⊗ v ⊗ 1 in X0,1 for all v ∈ V and ψ1(1 ⊗ g ⊗ 1) = 1 ⊗ g ⊗ 1 in X1,0 for all g ∈ G. We define ψ1 on elements of the form 1 ⊗ z ⊗ 1, for z ranging over a basis of a chosen complement of V ⊕ kG as a vector subspace of A, arbitrarily subject to the condition that d1ψ1(1 ⊗ z ⊗ 1) = ψ0δ1(1 ⊗ z ⊗ 1). Since ψ0, d1, δ1 all have degree 0 as maps, one may also choose ψ1 to have degree 0. Note that ψ1φ1 is the identity map on X0,1. (In fact, it is the identity map on all of X1.) Now let n ≥ 2 and assume that ψn−2 and ψn−1 have been defined to be degree 0 maps for which dn−1ψn−1 = ψn−2δn−1 and ψjφj is the identity map on X0,j for j = n − 2, n − 1. To define ψn, first note that A⊗(n+2) contains as an Ae- submodule the space X0,n (see (4.8)) and the image of each Xi,j under φn (n = i+j, i ≥ 1). By construction, their images intersect in 0 (being generated by elements of different degrees), the image of X0,n under φn is free, and moreover φn is injective on restriction to X0,n. Choose a set of free generators of φn(X0,n), and choose a 16 A. V. SHEPLER AND S. WITHERSPOON set of free generators of its complement in A⊗(n+2). For each chosen generator x of X0,n, we define ψn(φn(x)) to be x. Since dn(x) is in X0,n−1 by definition, we have by induction ψn−1φn−1dn(x) = dn(x). As δnφn(x) = φn−1dn(x), we now have dnψnφn(x) = ψn−1δnφn(x). That is, on these elements, ψn extends the chain map from degree n − 1 to degree n. On the remaining free generators of A⊗(n+2), define ψn arbitrarily subject to the requirement that it be a chain map of degree 0. (cid:3) 5. Deformations of quadratic algebras Let B be an arbitrary k-algebra. Let U be a B-bimodule that is free as a left B-module and as a right B-module, and set T := TB(U) = B ⊕ U ⊕ (U ⊗B U) ⊕ (U ⊗B U ⊗B U) ⊕ · · · , the tensor algebra of U over B with i-th graded component T i := U ⊗B i and T 0 = B. Let F i(T ) be the i-th filtered component: F i(T ) = T 0 ⊕ T 1 ⊕ · · · ⊕ T i. We call a B-subbimodule P of F 2(T ) a set of filtered quadratic relations over B and we call the quotient TB(U)/hP i a filtered quadratic algebra over B generated by U. By Proposition 2.1, if the relations are of PBW type, then they are unique. Set R = π(P ) where (recall) π is the projection F 2(T ) → U ⊗B U, so that T /hRi is the homogeneous quadratic algebra determined by P . Note that R is a B-subbimodule of U ⊗B U. We give below some conditions sufficient to guarantee that P and the quadratic algebra TB(U)/hP i it defines are of PBW type. First we present two lemmas. It is not difficult to see that any quadratic algebra over B of PBW type must be defined by a B-subbimodule P ⊂ T devoid of elements of filtered degree one. We record this observation and more in the next lemma. We choose labels consistent with those in [5] for ease of comparison. The proof (see (2.2)) of Proposition 2.1 implies: Lemma 5.1. Suppose P ⊂ T is a set of filtered quadratic relations over B defining a filtered quadratic algebra T /hP i of PBW type (with respect to P ). Then (I) P ∩ F 1(T ) = {0}, and (J) (F 1(T )P F 1(T )) ∩ F 2(T ) = P . If Condition (I) of Lemma 5.1 holds, then each (nonhomogeneous) generating relation defining the quadratic algebra T /hP i may be expressed as a unique ele- ment of homogeneous degree 2 plus linear and constant terms. We record these terms with functions α and β: Condition (I) implies existence of k-linear maps α : R → U and β : R → B for which P = {x − α(x) − β(x) x ∈ R}. Since P is a B-subbimodule of T , so is R, and it is not hard to see that the maps α and β are B-bimodule homomorphisms. We may now use the maps α and β to explore the PBW property using cohomology instead of examining overlap POINCAR ´E-BIRKHOFF-WITT THEOREM 17 polynomials and ambiguities (see [6], for example) explicitly. Note that since U is free (and thus flat) as a left B-module and as a right B-module, the spaces R⊗B U and U ⊗B R may be identified with subspaces of U ⊗B 3. Lemma 5.2. Suppose P ⊂ T is a set of filtered quadratic relations over B defining a filtered quadratic algebra T /hP i of PBW type (with respect to P ). Then Im(α ⊗B id − id ⊗B α) ⊂ R, (i) (ii) α ◦ (α ⊗B id − id ⊗B α) = −(β ⊗B id − id ⊗B β), (iii) β ◦ (α ⊗B id − id ⊗B α) = 0, where the maps α ⊗B id − id ⊗B α and β ⊗B id − id ⊗B β are defined on the subspace (R ⊗B U) ∩ (U ⊗B R) of T . Proof. By Lemma 5.1, Conditions (I) and (J) hold. We show that (I) and (J) imply (i), (ii), and (iii). Let x ∈ (R ⊗B U) ∩ (U ⊗B R). By definition of α and β, x − (α ⊗B id + β ⊗B id )(x) ∈ P T 1 ⊂ F 1(T )P F 1(T ), x − (id ⊗B α + id ⊗B β)(x) ∈ T 1P ⊂ F 1(T )P F 1(T ). We subtract these two expressions and check degrees to see that Condition (J) implies (α ⊗B id − id ⊗B α + β ⊗B id − id ⊗B β)(x) ∈ (F 1(T )P F 1(T )) ∩ F 2(T ) = P. Again considering the degrees of the above elements, we must have (α ⊗B id − id ⊗B α)(x) ∈ R, α((α ⊗B id − id ⊗B α)(x)) = −(β ⊗B id − id ⊗B β)(x), β((α ⊗B id − id ⊗B α)(x)) = 0. (cid:3) Example 5.3. We return to Example 2.3 in which B = k and the B-bimodule U is the vector space V = Spank{x, y} with P ′ = Spank{xy −x, yx−y, x2 −x, y2−y}. As β is identically zero, an easy check of Conditions (i), (ii), and (iii) amounts to checking overlap relations in P ′ and verifying that P ′ is a noncommutative Grobner basis for the ideal it generates in Tk(V ). In the remainder of this section, we turn to the case B = kG, a finite group algebra. We show in the next theorem that the above conditions, adapted from Braverman and Gaitsgory [5], are both necessary and sufficient in the case that the homogeneous quadratic algebra determined by P is isomorphic to S#G for some Koszul algebra S. (Precisely, we set U = V ⊗ kG for a finite dimensional k-vector space V and view U as a bimodule over the group algebra B = kG.) Theorem 5.4. Let S be a finitely generated graded Koszul algebra over k on which a finite group G acts by graded automorphisms. Suppose a filtered quadratic algebra A′ over kG is defined by a set of filtered quadratic relations that determine 18 A. V. SHEPLER AND S. WITHERSPOON a homogeneous quadratic algebra isomorphic to S#G. Then A′ is of PBW type if and only if Conditions (I), (i), (ii), and (iii) hold. Proof. Suppose the Koszul algebra S is generated by the k-vector space V with some k-vector space of quadratic relations R′ ⊂ V ⊗ V , i.e., S = Tk(V )/hR′i. Let U = V ⊗ kG, a kG-bimodule with right action given by multiplication on the rightmost factor kG only and left action given by g(v ⊗ h) = gv ⊗ gh for all v ∈ V , g, h ∈ G. Set R = R′ ⊗ kG, similarly a kG-bimodule (as R′ is a kG-module). Then TkG(U)/hRi ∼= (Tk(V )/hR′i)#G = S#G ∼= (Tk(V )#G)/hR′i, as a consequence of the canonical isomorphism between TkG(U) and Tk(V )#G given by identifying elements of V and of G and moving all group elements far right; here hR′i denotes the ideal of Tk(V ) or of Tk(V )#G generated by R′, re- spectively. We may assume that A′ is also generated by U over kG, i.e., A′ = TkG(U)/hP i for a set of filtered quadratic relations P over kG. Note that π(P ) = R since both π(P ) and R are kG-bimodules generating the same ideal in TkG(U) (use Proposition 2.1, for example). The conditions are then necessary by Lemmas 5.1 and 5.2. It remains to prove that they are sufficient, so assume Conditions (I), (i), (ii), and (iii) hold for P = {x − α(x) − β(x) x ∈ R}. We adapt the proof of Braverman and Gaitsgory [5, §4] to our setting using the resolution X r given by (4.2) after choosing C r to be the bar resolution (3.1) of kG and D r to be the Koszul resolution (3.7) of S. By Theorem 4.3, X r calculates the Hochschild cohomology HH r(A) cataloging deformations of A. We first extend the maps α and β to cochains on X r. We then use chain maps between X r and the bar resolution for A to convert α and β to Hochschild 2-cochains which can define multiplication maps for a potential deformation. We modify the cochains as necessary to preserve the conditions of the theorem. Using these conditions, we build a second level graded deformation of A. We then extend to a graded de- formation of A and conclude the PBW property using properties of the resolution X r. We note that Conditions (i), (ii), and (iii) may be interpreted as conditions on a tensor product over k for the extensions of the maps α, β to X r. We first extend α and β to X r. In degree 2, X2 contains the direct summand ∼= A⊗R′⊗A (apply Theorem 4.3 with i = 0, j = 2). Note that R = R′⊗kG ⊆ X0,2 V ⊗ V ⊗ kG ∼= U ⊗kG U, and we thus view the kG-bilinear maps α, β : R → A as maps on R′ ⊗ kG. Extend them to Ae-module maps from A ⊗ R′ ⊗ A ∼= A ⊗ R ⊗ S to A by composing with the multiplication map, and, by abuse of notation, denote these extended maps by α, β as well. Set α and β equal to 0 on the summands X2,0 and X1,1 of X2 so that they further extend to maps α, β : X2 → A. Condition (i) implies that α is 0 on the image of the differential on X0,3. Since α is a kG-bimodule homomorphism by its definition, α is G-invariant. We claim that this implies it is also 0 on the image of the differential on X1,2: Let a, b ∈ A, POINCAR ´E-BIRKHOFF-WITT THEOREM 19 g ∈ G, and r ∈ R′, and consider a⊗g⊗r⊗b as an element of X1,2 (apply Theorem 4.3 with i = 1, j = 2). We apply (4.5): ∼= A⊗kG⊗R′ ⊗A d(a ⊗ g ⊗ r ⊗ b) = d((a ⊗ g ⊗ 1) ⊗ (1 ⊗ r ⊗ b)) = d(a ⊗ g ⊗ 1) ⊗ (1 ⊗ r ⊗ b) − (a ⊗ g ⊗ 1) ⊗ d(1 ⊗ r ⊗ b). The second term lies in X1,1, but α is 0 on X1,1 by definition. Therefore α(d(a ⊗ g ⊗ r ⊗ b)) = α((ag ⊗ 1 − a ⊗ g) ⊗ (1 ⊗ r ⊗ b)) = α(ag ⊗ r ⊗ b − a ⊗ gr ⊗ gb) = agα(r)b − aα(gr)gb, where we use (4.4). But gα(r) = α(gr)g, since α is G-invariant, and thus α is zero on the image of d on X1,2. It follows that α is a 2-cocycle on X r and defines a Hochschild cohomology class of HH2(A). Thus α yields a first level deformation A1 of A (i.e., an infinitesimal deformation of A) with some first multiplication map µ1 : A ⊗ A → A. In fact, we may apply the chain map ψ r of Lemma 4.7 and choose µ1 = ψ∗ 2(α). Note that α is homogeneous of degree −1 by its definition, and therefore so is µ1. We claim that φ∗ 2(µ1) = α as cochains. To verify this, first let x ∈ X0,2. By Lemma 4.7, ψ2φ2(x) = x, and hence µ1φ2(x) = αψ2φ2(x) = α(x). Now let x be a free generator of X1,1 or of X2,0, so that it has degree 1 or 0. Then ψ2φ2(x) has degree 1 or 0, implying that its component in X0,2 is 0, from which it follows that µ1φ2(x) = αψ2φ2(x) = 0 = α(x). Therefore φ∗ 2(µ1) = α. Condition (ii) implies that −d∗ 3(β) = α ◦ (α ⊗ id − id ⊗ α) as cochains on X r. (Again, since β is G-invariant, it will be 0 on the image of the differential on X1,2.) Let µ2 = ψ∗ 2(µ2) = β. However, we want µ1, µ2 to satisfy the differential condition that α, β satisfy, i.e., 2(β). By a similar argument as that above for α, we have φ∗ (5.5) − δ∗(µ2) = µ1 ◦ (µ1 ⊗ id − id ⊗ µ1) as cochains on the bar resolution. We modify µ2 as necessary to satisfy this condition. Let γ = δ∗(µ2) + µ1 ◦ (µ1 ⊗ id − id ⊗ µ1). The cochain φ∗(γ) is zero on X0,3 by Condition (ii), since the image of φ on X0,3 is contained in A ⊗ ((R′ ⊗ V ) ∩ (V ⊗ R′)) ⊗ A, and φ∗(µ1) = α. Additionally, φ∗(γ) is 0 on X2,1 and on X3,0 since it is a map of degree −2. To see that it is also 0 on X1,2, note that as an Ae-module, the image of X1,2 under φ is generated by elements of degree 2. Since α contains kG in its kernel and µ1 = ψ∗ 2(α), the map µ1 ◦ (µ1 ⊗ id − id ⊗ µ1) must be 0 on the image of X1,2 under φ. Since β is G-invariant and φ∗(µ2) = β, we have that φ∗δ∗(µ2) = d∗φ∗(µ2) = d∗(β) is 0 on X1,2. Therefore φ∗(γ) is 0 on X1,2. We have shown that φ∗(γ) is 0 on all of X3, and so γ defines the zero cohomology class on the bar complex as well, i.e., 20 A. V. SHEPLER AND S. WITHERSPOON it is a coboundary. Thus there is a 2-cochain µ of degree −2 on the bar complex with δ∗(µ) = γ. If we were to replace µ2 with µ2 − µ, it would satisfy the desired differential condition (5.5). However, φ∗(µ2 − µ) may not agree with β on X2. We subtract off another term: Since d∗φ∗(µ) = φ∗δ∗(µ) = φ∗(γ) = 0, the 2-cochain φ∗(µ) is a cocycle on the complex X r and thus lifts to a cocycle µ′ of degree −2 on the bar complex, i.e., µ′ satisfies φ∗(µ′) = φ∗(µ). Then φ∗(µ2 − µ + µ′) = β and δ∗(µ2 − µ + µ′) + µ1 ◦ (µ1 ⊗ id − id ⊗ µ1) is zero on the bar resolution as δ∗µ′ = 0 and −δ∗(µ2 − µ) = µ1 ◦ (µ1 ⊗ id − id ⊗ µ1). We hence replace µ2 by µ2 − µ + µ′. We have now constructed maps µ1, µ2 satisfying the differential condition (5.5) required to obtain a second level graded deformation: There exists a second level graded deformation A2 of A (with multiplication defined by µ2) extending A1. By Condition (iii) and degree considerations, the obstruction µ2 ◦ (µ1 ⊗ id − id ⊗ µ1) + µ1 ◦ (µ2 ⊗ id − id ⊗ µ2) to lifting A2 to a third level deformation A3 is 0 as a cochain on X r under the cochain map φ∗. Therefore, as a cochain on the bar resolution, this obstruction is a coboundary, and so represents the zero cohomology class in HH3(A). Thus there exists a 2-cochain µ3 of degree −3 satisfying the obstruction equation (3.5) for i = 3, and the deformation lifts to the third level by Proposition 3.6. The obstruction for a third level graded deformation A3 of A to lift to the fourth level lies in HH3,−4(A) (again by Proposition 3.6). We apply φ∗ to this obstruction (the right side of equation (3.5) with i = 4) to obtain a cochain on X3. But there are no cochains of degree −4 on X3 by definition (as it is generated by elements of degree 3 or less), hence the obstruction is automatically zero. Therefore the deformation may be continued to the fourth level. Similar arguments show that it can be continued to the fifth level, and so on. Let At be the (graded) deformation of A that we obtain in this manner. Then At is the k-vector space A[t] with multiplication a1 ∗ a2 = a1a2 + µ1(a1 ⊗ a2)t + µ2(a1 ⊗ a2)t2 + µ3(a1 ⊗ a2)t3 + . . . , where a1a2 is the product in the homogeneous quadratic algebra A = S#G ∼= TkG(U)/hRi ∼= (Tk(V )#G)/hR′i and each µi : A ⊗ A → A is a k-linear map of homogeneous degree −i. (The sum terminates for each pair a1, a2 by degree con- siderations.) Then for any r in R, µ1(r) = α(r) and µ2(r) = β(r) by construction, and µi(r) = 0 for i ≥ 3 by considering degrees. We now argue that A′ is isomorphic as a filtered algebra to the fiber of the deformation At at t = 1. Let A′′ = (At)t=1. First note that A′′ is generated by V and G, since it is a filtered algebra (as At is graded) with associated graded algebra POINCAR ´E-BIRKHOFF-WITT THEOREM 21 A = S#G. Next note that products of pairs of elements in G, or of an element of V paired with an element of G, are the same in A′′ as they are in Tk(V )#G: g ∗ v = gv, v ∗ g = vg, and g ∗ h = gh for all g, h in G. To verify this observation, one need only check that µ1 vanishes on such pairs, as At is a graded deformation with group elements in degree 0 and vectors in degree 1. But µ1 must vanish on low degree pairs by our construction of chain maps: µ1 = ψ∗ 2(α), the chain map ψ2 preserves degree, and X0,2 has free basis as an Ae-module consisting of elements of degree 2. Thus the canonical surjective algebra homomorphism TkG(U) ∼= Tk(V )#G → A′′ arises, mapping each v in V and g in G to their copies in A′′. The elements of P lie in the kernel (by definition of A′′), and thus the map induces a surjective algebra homomorphism: A′ = TkG(U)/hP i → A′′. We claim this map is an isomorphism of filtered algebras. First compare the dimensions in each filtered component of these two algebras. Recall that algebra A′ has dimension at most that of S#G in each filtered component (as its associated graded algebra is a quotient of S#G). Then since there is a surjective algebra homomorphism from A′ to A′′ and gr(A′′) = S#G, dimk(F m(S#G)) ≥ dimk(F m(A′)) ≥ dimk(F m(A′′)) = dimk(F m(S#G)) for each degree m, where F m denotes the m-th filtered component. Thus the inequalities are forced to be equalities, and A′ ∼= A′′, a specialization of a defor- mation of S#G. Consequently A′ is of PBW type. (cid:3) The next result points out a correspondence between PBW filtered quadratic algebras and fibers of graded deformations of a particular type: Corollary 5.6. Let A = S#G for a finitely generated, graded Koszul algebra S over k carrying the action of a finite group G by graded automorphisms. Every graded deformation At of A for which the kernel of µ1 contains kG⊗V and V ⊗kG has fiber at t = 1 isomorphic (as a filtered algebra) to a filtered quadratic algebra over kG of PBW type with induced quadratic algebra isomorphic to A. Conversely, every such filtered quadratic algebra is isomorphic to the fiber at t = 1 of a graded deformation of A for which the kernel of µ1 contains kG ⊗ V and V ⊗ kG. Proof. Write the algebra A = S#G as TkG(U)/hRi where U = V ⊗ kG for some finite dimensional k-vector space V generating S and set of filtered quadratic relations R ⊂ U ⊗kG U. Suppose At is a graded deformation of A with mul- tiplication map µ1 vanishing on all v ⊗ g and g ⊗ v for g in G and v in V . (Higher degree maps automatically vanish on such input as At is graded.) We may reverse engineer the filtered quadratic algebra A′ = TkG(U)/hP i by setting 22 A. V. SHEPLER AND S. WITHERSPOON P = {r − µ1(r) − µ2(r) : r ∈ R}. The argument at the end of the proof of Theo- rem 5.4 implies that A′ is of PBW type and isomorphic to the fiber of At at t = 1 as a filtered algebra, as this fiber is a quotient of Tk(V )#G. Conversely, any PBW filtered quadratic algebra satisfies the conditions of Theorem 5.4; its proof con- structs a graded deformation At, for which the kernel of µ1 contains kG ⊗ V and V ⊗ kG, and whose fiber A′′ at t = 1 is isomorphic to A′ as a filtered algebra. (cid:3) 6. Applications: Drinfeld orbifold algebras, graded Hecke algebras, and symplectic reflection algebras We now apply our results from previous sections to Drinfeld orbifold algebras, which include graded Hecke algebras, rational Cherednik algebras, symplectic re- flection algebras, and Lie orbifold algebras as special cases. These algebras present as a certain kind of quotient of a skew group algebra. Let G be a finite group acting linearly on a finite dimensional k-vector space V , and consider the induced action on Sk(V ) and on Tk(V ). Drinfeld orbifold algebras are deformations of the skew group algebra Sk(V )#G (see [26]). Many articles investigate their properties and representation theory, in particular, when k is the field of real or complex numbers, when G acts faithfully, and when G acts symplectically. We assume that the characteristic of k is not 2 throughout this section. We are especially interested in the case when the characteristic of k divides G, where the theory is much less developed. Drinfeld orbifold algebras arise as quotients of the form Hκ = Tk(V )#G/hv1 ⊗ v2 − v2 ⊗ v1 − κ(v1, v2) : v1, v2 ∈ V i where κ is a parameter function on V ⊗ V taking values in the group algebra kG, or possibly in V , or some combination of kG and V . We abbreviate v ⊗ 1 by v and 1 ⊗g by g in the skew group algebra Tk(V )#G (which is isomorphic to Tk(V ) ⊗kG as a k-vector space). We take κ to be any alternating map from V ⊗ V to the first filtered component of Tk(V )#G: κ : V ⊗ V → kG ⊕ (V ⊗ kG) , and write κ(v1, v2) for κ(v1⊗v2) for ease of notation. The associated graded algebra of Hκ is a quotient of Sk(V )#G. We say that Hκ is a Drinfeld orbifold algebra if it satisfies the Poincar´e-Birkhoff-Witt property: gr(Hκ) ∼= Sk(V )#G as graded algebras. We explained in [26] how every Drinfeld orbifold algebra defines a formal deformation of Sk(V )#G and we also explained which deformations arise this way explicitly. Various authors explore conditions on κ guaranteeing that the quotient Hκ satisfies the PBW property. Such Drinfeld orbifold algebras are often called: • Rational Cherednik algebras when G is a real or complex reflection group acting diagonally on V = X ⊕ X ∗, for X the natural reflection representation, and κ has image in kG and a particular geometric form, POINCAR ´E-BIRKHOFF-WITT THEOREM 23 • Symplectic reflection algebras when G acts on any symplectic vector space V and κ has image in kG, • Graded affine Hecke algebras when G is a Weyl group (or Coxeter group) and κ has image in kG, • Drinfeld Hecke algebras when G is arbitrary and κ has image in kG, • Quantum Drinfeld Hecke algebras when G is arbitrary and the non- homogeneous relation v1 ⊗ v2 = v2 ⊗ v1 + κ(v1, v2) is replaced by a quantum version vi ⊗ vj = qijvj ⊗ vi + κ(vi, vj) (for a system of quantum parameters {qij}) and κ has image in kG, • Lie orbifold algebras when κ has image in V ⊕kG (defining a deformation of the universal enveloping algebra of a Lie algebra with group action). Terminology arises from various settings. Drinfeld [7] originally defined these algebras for arbitrary groups and for κ with image in kG. Around the same time, Lusztig (see [20, 21], for example) explored a graded version of the affine Hecke algebra for Coxeter groups. Ram and Shepler [23] showed that Lusztig's graded affine Hecke algebras are a special case of Drinfeld's construction. Etingof and Ginzburg [8] rediscovered Drinfeld's algebras for symplectic groups in the context of orbifold theory. The general case (when κ maps to the filtered degree 1 piece of Tk(V )#G and G acts with an arbitrary representation) is explored in [26]; see [16] in case the field is the real numbers. The original conditions of Braverman and Gaitsgory were adapted and used for determining which κ define Drinfeld orbifold algebras, but arguments relied on the fact that the skew group algebra Sk(V )#G is Koszul as an algebra over the semisimple ring kG. (This was the approach taken by Etingof and Ginzburg [8].) Indeed, we used this theory in [26] to establish PBW conditions in the nonmodular setting. However the technique fails in modular characteristic (i.e., when the characteristic of k divides the order of G). The results in previous sections allow us to overlook the nonsemisimplicity of the group algebra in determining which quotients Hκ satisfy the PBW property. We consider each Drinfeld orbifold algebra as a nonhomogeneous quadratic alge- bra whose homogeneous version is the skew group algebra formed from a Koszul algebra and a finite group. We are now able to give a new, shorter proof of The- orem 3.1 in [26] using the methods of Braverman and Gaitsgory but in arbitrary characteristic. Thus we bypass tedious application of the Diamond Lemma [3]. (Details of a long computation using the Diamond Lemma over arbitrary fields were largely suppressed in [26].) Decompose the alternating map κ : V ⊗ V → (k ⊕ V ) ⊗ kG into its linear and constant parts by writing κ(v1, v2) =Xg∈G(cid:16)κC g (v1, v2) + κL g (v1, v2)(cid:17) ⊗ g 24 A. V. SHEPLER AND S. WITHERSPOON for maps κC the alternating group on three symbols. g : V ⊗ V → k and κL g : V ⊗ V → V , for all g in G. Let Alt3 denote Proposition 6.1. [26, Theorem 3.1] Let G be a finite group acting linearly on a finite dimensional vector space V over a field k of arbitrary characteristic. The quotient algebra Hκ = Tk(V )#G/hv1 ⊗ v2 − v2 ⊗ v1 − κ(v1, v2) : v1, v2 ∈ V i satisfies the Poincar´e-Birkhoff-Witt property if and only if • κ is G-invariant, κL g (vσ(2), vσ(3))(vσ(1) − gvσ(1)) = 0 in Sk(V ), • Xσ∈Alt3 • Xσ∈Alt3,h∈G • Xσ∈Alt3,h∈G κL κC gh−1(cid:16)vσ(1) + hvσ(1), κL = 2 Xσ∈Alt3 gh−1(cid:16)vσ(1) + hvσ(1), κL h (vσ(2), vσ(3))(cid:17) h (vσ(2), vσ(3))(cid:17) = 0, κC g (vσ(2), vσ(3))(vσ(1) − gvσ(1)), for all v1, v2, v3 in V and g in G. Proof. Let U = V ⊗ kG. Consider Tk(V )#G ∼= TkG(U) to be a kG-bimodule under the action g1(v ⊗ g2)g3 = g1v ⊗ g1g2g3 for gi in G and v in Tk(V ). As before, we filter T := TkG(U) by setting F i(T ) = T 0 + T 1 + . . . + T i where T i = U ⊗kG i for i > 0 and T 0 = kG. Extend κ to a map κ : T 2 kG(U) → kG ⊕ U defined by κ((v1 ⊗ g1) ⊗kG (v2 ⊗ g2)) = κ(v1, g1v2) ⊗ g1g2 for vi in V and g in G. (Note that κ extends to a unique kG-bimodule map κ : T 2 kG(U) → kG ⊕ U if, and only if, κ is G-invariant: g(κ(g−1 v)) = κ(u, v) for all u, v in V and g in G. ) u, g−1 We set P to be the generating nonhomogeneous relations parametrized by κ: Let P be the kG-subbimodule of F 2(T ) generated by all v1 ⊗kG v2 − v2 ⊗kG v1 − κ(v1, v2) ∼= TkG(U)/hP i. for v1, v2 in V . We then have an isomorphism of filtered algebras, Hκ Thus, Hκ satisfies the PBW property if and only if TkG(U)/hP i exhibits PBW type with respect to P as a filtered quadratic algebra over kG. We now consider the homogeneous version A determined by P . Set R = π(P ), the kG-subbimodule of T 2 kG(U) generated by all v1 ⊗kG v2 − v2 ⊗kG v1 for v1, v2 in V . Set A := TkG(U)/hRi and note that A ∼= Sk(V )#G ∼= TkG(U)/hπ(P )i. Then as Sk(V ) is Koszul, the conditions of Theorem 5.4 apply to give explicit conditions on κ under which Hκ is of PBW type. We apply the conditions as in the first proof of Theorem 3.1 of [26] without needing the extra assumption there that kG is semisimple. (cid:3) POINCAR ´E-BIRKHOFF-WITT THEOREM 25 In the next corollary, we set κL g ≡ 0 for all g in G to recover a modular version of a result that is well-known in characteristic zero (stated in [7], then confirmed in [8] and [23]). The positive characteristic result was first shown by Griffeth [14] by construction of an explicit Hκ-module, as in the classical proof of the PBW theorem for universal enveloping algebras of Lie algebras. (See also Bazlov and Berenstein [2] for a generalization.) Our approach yields a different proof. Note that several authors (for example, Griffeth [14] and Balagovic and Chen [1]) study the representations of rational Cherednik algebras in the modular setting. Corollary 6.2. Let G be a finite group acting linearly on a finite dimensional vector space V over a field k of arbitrary characteristic. Let κ : V ⊗ V → kG be an alternating map. The quotient algebra Hκ = Tk(V )#G/hv1 ⊗ v2 − v2 ⊗ v1 − κ(v1, v2) : v1, v2 ∈ V i satisfies the Poincar´e-Birkhoff-Witt property if and only if κ is G-invariant and κg(vσ(2) ⊗ vσ(3)) ( gvσ(1) − vσ(1)) 0 = Xσ∈Alt3 for all v1, v2 in V and g in G. References [1] M. Balagovic and H. Chen, "Representations of Rational Cherednik Algebras in Positive Characteristic," arXiv:1107.0504v1. [2] Y. Bazlov, A. Berenstein, "Braided doubles and rational Cherednik algebras," Adv. Math. 220 (2009), no. 5, 1466 -- 1530. [3] G. Bergman, "The diamond lemma for ring theory," Adv. Math. 29 (1978) no. 2, 178 -- 218. [4] A. Beilinson, V. Ginzburg, and W. Soergel, "Koszul duality patterns in representation theory," J. Amer. Math. Soc. 9 (1996), no. 2, 473 -- 527. [5] A. Braverman and D. Gaitsgory, "Poincar´e-Birkhoff-Witt Theorem for quadratic algebras of Koszul type," J. Algebra 181 (1996), 315 -- 328. [6] J. L. Bueso and J. G´omez-Torrecillas and A. Verschoren, Algorithmic methods in non- commutative algebra. Applications to quantum groups. Mathematical Modelling: Theory and Applications, 17, Kluwer Academic Publishers, Dordrecht, 2003. [7] V. G. Drinfeld, "Degenerate affine Hecke algebras and Yangians," Funct. Anal. Appl. 20 (1986), 58 -- 60. [8] P. Etingof and V. Ginzburg, "Symplectic reflection algebras, Calogero-Moser space, and deformed Harish-Chandra homomorphism," Invent. Math. 147 (2002), no. 2, 243 -- 348. [9] M. Farinati, "Hochschild duality, localization, and smash products," J. Algebra 284 (2005), no. 1, 415 -- 434. [10] M. Gerstenhaber, "On the deformation of rings and algebras," Ann. of Math. 79 (1964), 59 -- 103. [11] M. Gerstenhaber and S. Schack, "On the deformation of algebra morphisms and diagrams," Trans. Amer. Math. Soc. 279 (1983), 1 -- 50. [12] M. Gerstenhaber and S. Schack, "Algebraic cohomology and deformation theory," Defor- mation theory of algebras and structures and applications (Il Ciocco, 1986), NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 247, Kluwer Acad. Publ., Dordrecht (1988), 11 -- 264. 26 A. V. SHEPLER AND S. WITHERSPOON [13] E. Green, I. Reiten, and Ø. Solberg, "Dualities on generalized Koszul algebras," Mem. Amer. Math. Soc. 159 (2002). [14] S. Griffeth, "Towards a combinatorial representation theory for the rational Cherednik algebra of type G(r, p, n)," Proceedings of the Edinburgh Mathematical Society (Series 2), 53 (2010), 419 -- 445, see arxiv:0612733. [15] J. A. Guccione, J. J. Guccione, and C. Valqui, "Universal deformation formulas and braided module algebras," J. Algebra 330 (2011), 263 -- 297. [16] G. Halbout, J.-M. Oudom, and X. Tang, "Deformations of orbifolds with noncommutative linear Poisson structures," Int. Math. Res. Not. (2011), no. 1, 1 -- 39. [17] U. Krahmer, "Notes on Koszul algebras," available at http://www.maths.gla.ac.uk/∼ukraehmer/#research. [18] H. Li, Grobner Bases in Ring Theory, World Scientific Publishing Company, 2012. [19] L. Li, "A generalized Koszul theory and its application," to appear in Trans. Amer. Math. Soc. [20] G. Lusztig, "Cuspidal local systems and graded Hecke algebras. I," Inst. Haute Etudes Sci. Publ. Math. 67 (1988), 145 -- 202. [21] G. Lusztig, "Affine Hecke algebras and their graded version," J. Amer. Math. Soc. 2 (1989), no. 3, 599 -- 635. [22] A. Polishchuk and L. Positselski, Quadratic Algebras, University Lecture Series 37, Amer. Math. Soc., 2005. [23] A. Ram and A.V. Shepler, "Classification of graded Hecke algebras for complex reflection groups," Comment. Math. Helv. 78 (2003), 308 -- 334. [24] K. Sanada, "On the Hochschild cohomology of crossed products," Comm. Algebra 21 (8) (1993), 2717 -- 2748. [25] A.V. Shepler and S. Witherspoon, "Quantum differentiation and chain maps of bimodule complexes," Algebra and Number Theory, 5-3 (2011), 339 -- 360. [26] A.V. Shepler and S. Witherspoon, "Drinfeld orbifold algebras," Pacific J. Math. 259-1 (2012), 161 -- 193. [27] D. S¸tefan, "Hochschild cohomology on Hopf Galois extensions," J. Pure Appl. Algebra 103 (1995), 221 -- 233. [28] C. Weibel, An Introduction to Homological Algebra, Cambridge Univ. Press, 1994. [29] D. Woodcock, "Cohen-Macalay complexes and Koszul rings," J. London Math. Soc. (2) 57 (1998), no. 2, 398 -- 410. Department of Mathematics, University of North Texas, Denton, Texas 76203, USA E-mail address: [email protected] Department of Mathematics, Texas A&M University, College Station, Texas 77843, USA E-mail address: [email protected]
1706.04879	1	1706	2017-06-13T18:08:21	On the idempotent semirings such that $\mathcal{D}^\bullet$ is the least distributive lattice congruence	[ "math.RA" ]	Here we describe the least distributive lattice congruence $\eta$ on an idempotent semiring in general and characterize the varieties $D^\bullet, L^\bullet$ and $R^\bullet$ of all idempotent semirings such that $\eta=\mathcal{D}^\bullet, \mathcal{L}^\bullet$ and $\mathcal{R}^\bullet$, respectively. If $S \in D^\bullet [L^\bullet, R^\bullet]$, then the multiplicative reduct $(S, \cdot)$ is a [left, right] normal band. Every semiring $S \in D^\bullet$ is a spined product of a semiring in $L^\bullet$ and a semiring in $R^\bullet$ with respect to a distributive lattice.	math.RA	math	On the idempotent semirings such that D• is the least distributive lattice congruence M. K. Sena, A. K. Bhuniyab∗ and R. Debnathc a) Department of Pure Mathematics, University of Calcutta, Kolkata, India b) Department of Mathematics, Visva-Bharati, Santiniketan-731235, India. c) Department of Mathematics, Kurseong college, Kurseong-734203, India. E-mail addresses: [email protected], [email protected] and [email protected]. Abstract Here we describe the least distributive lattice congruence η on an idempotent semiring in general and characterize the varieties D•, L• and R• of all idempotent semirings such that η = D•, L• and R•, respectively. If S ∈ D•[L•, R•], then the multiplicative reduct (S, ·) is a [left, right] normal band. Every semiring S ∈ D• is a spined product of a semiring in L• and a semiring in R• with respect to a distributive lattice. Keywords : idempotent semiring; least distributive lattice congruence; normal band, spined prod- uct, Malcev's product. AMS Mathematics Subject Classification: 16Y60. 1 Introduction A semiring (S, +, ·) is an algebra with two binary operations + and · such that both the addi- tive reduct (S, +) and the multiplicative reduct (S, ·) are semigroups and such that the following distributive laws hold: ∗Corresponding author x(y + z) = xy + xz and (x + y)z = xz + yz. 1 If moreover, both the reducts (S, +) and (S, ·) are bands, then S is called an idempotent semiring. Thus the class of all idempotent semirings is an equational class satisfying two additional identities: The variety of all idempotent semirings will be denoted by I. x + x ≈ x and x · x ≈ x. We now fix the notation for some varieties of idempotent semirings and give their determining identity within I: Notation Defining identity within I Notation Defining identity within I R+ LZ+ LZ• x + y + x ≈ x, x + y ≈ x, xy ≈ x, R• RZ+ RZ• xyx ≈ x, x + y ≈ y, xy ≈ y, LNB• xyz ≈ xzy, RNB• xyz ≈ yxz, LQBI x + xy + x ≈ x, RQBI x + yx + x ≈ x, LN N x + xyx ≈ x, x + xyx + x ≈ x, RN Sl+ xyx + x ≈ x, x + y ≈ y + x. For an idempotent semiring (S, +, ·) the Green's relations L, R and D on the additive [mul- tiplicative] reduct (S, +)[(S, ·)] will be denoted by L+, R+ and D+[L•, R•, D•]. Since the multi- plicative reduct (S, ·) of an idempotent semiring S is a band, D•, L• and R• are given by: for a, b ∈ S, aD•b ⇔ aba = a, bab = b, aL•b ⇔ ba = b, ab = a and aR•b ⇔ ab = b, ba = a. The relations D+, L+ and R+ are given dually. The variety I of all idempotent semirings contains the variety D of all distributive lattices; and hence there is the least distributive lattice congruence on each idempotent semiring S. Throughout this article, we denote the least distributive lattice congruence on S by ηS or precisely by η. Then (S/η, +) is a semilattice. Since (S, +) is a band, D+ is the least congruence on the idempotent semiring S such that (S/D+, +) is a semilattice. Hence D+ ⊆ η. In [13], Sen, Guo and Shum characterized in depth the idempotent semirings S such that η = D+. 2 Lemma 1.1. [13] Let S be an idempotent semiring. Then the following conditions are equivalent. 1. D+ is the least distributive lattice congruence on S; 2. S satisfies the identities: x + xy + x ≈ x and x + yx + x ≈ x; (1.1) (1.2) 3. S ∈ R+ ◦ D. An idempotent semiring S is said to be a band semiring if D+ is the least distributive lattice congruence on S, equivalently if it satisfies the identities (1.1) and (1.2). We will denote the variety of all band semirings by BI. We define an idempotent semiring to be a left [right] quasi-band semiring if it satisfies the identity (1.1) [(1.2)]. Thus BI = LQBI ∩ RQBI. Additive reduct (S, +) of a band semiring S is a regular band [14]. The idempotent semirings for which η = D• have been characterized by Pastijn and Zhao [9]. There are several articles characterizing the variety I of idempotent semirings by Green's relations [7], [9], [13] - [16]. We extend the problem to the idempotent semirings S such that the least distributive lattice congruence η on S is either of the Green's relations L+, R+, L• and R•. For this, we first describe explicitly the least distributive lattice congruence η on an idempotent semiring in general, which makes it possible to characterize the idempotent semirings for which η = D+ [D•, L+, L•, R+, R•] in an unified and simple approach. We present our results in two parts. The idempotent semirings S such that η = D+ [L+, R+] have been characterized in [1]. The following result has some use in the last section of this articel. Lemma 1.2. [1] Let S be an idempotent semiring. Then the following conditions are equivalent. 1. L+ is the least distributive lattice congruence on S; 2. S ∈ LN and D• ⊆ L+; 3. S satisfies the identities: 4. S ∈ LZ+ ◦ D. x + yxy ≈ x; (1.3) 3 In Section 2 of this article, we characterize the least distributive lattice congruence on an idempotent semiring. The idempotent semirings S such that η = D• [L•, R•] have been studied in Section 3. In Section 4, we show that the multiplicative reduct (S, ·) of an idempotent semiring S such that η = D• is a normal band, and such a semiring S is a spined product of an idempotent semiring S1 such that η = L• and an idempotent semiring S2 such that η = R• with respect to a distributive lattice. We refer to [4] for the information concerning semigroup theory, [3] for background on semirings and [5] for notions concerning universal algebra and lattice theory. 2 The least distributive lattice congruence on an idempo- tent semiring Let S be an idempotent semiring. Define a binary relation σ on S by: for a, b ∈ S, aσb if and only if aba = aba + a + aba and bab = bab + b + bab. Lemma 2.1. Let S be an idempotent semiring. Then the congruence relation generated by σ is the least distributive lattice congruence on S. Proof. Let a, b ∈ S. Then we have (a + b)(b + a)(a + b) + a + b + (a + b)(b + a)(a + b) = (a + b)(b + a + a + b + b + a)(a + b) = (a + b)(b + a)(a + b), and similarly (b + a)(a + b)(b + a) = (b + a)(a + b)(b + a) + a + b + (b + a)(a + b)(b + a), which implies that (a + b)σ(b + a). Also (ab)σ(ba) for all a, b ∈ S. Now, a(a + ab)a + a + a(a + ab)a = (a + aba) + a + (a + aba) = a + aba = a(a + ab)a and (a + ab)a(a + ab) + a + ab + (a + ab)a(a + ab) = (a + ab)a(a + ab) implies that (a + ab)σa for all a, b ∈ S. Thus any congruence on S which contains σ is a distributive lattice congruence on S. Consider a distributive lattice congruence ρ on S and a, b ∈ S such that aσb. Then aba = aba + a + aba implies that bρ(b + aba) = (b + aba + a + aba)ρ(b + a)ρ(a + b). Similarly from bab = bab + b = bab we have aρ(a + b) and it follows that aρb. Thus σ is contained in every distributive lattice congruence on S. Hence the result follows. Now we show that σ is not transitive on a semiring in general. For this consider the following example [2] of an idempotent semiring, 4 + a b c a a b c b c b c b b b c · a b c a a a a b b b c a b b c Then aσb and bσc but aσc shows that σ is not transitive. The transitive closure σ∗ of σ is given by: aσ∗b if and only if there exists x ∈ S such that axbxa = axbxa + a + axbxa and bxaxb = bxaxb + b + bxaxb. A semiring (S, +, ·) is called almost idempotent if (S, +) is a semilattice and a + a2 = a2 for every a ∈ S. In [12], we proved that σ∗ is the least distributive lattice congruence on an almost idempotent semiring. Since every idempotent semiring S with commutative addition is an almost idempotent semiring, σ∗ is the least distributive lattice congruence on S. The proof that σ∗ remains the least distributive lattice congruence on an idempotent semiring S even if the addition in S is not commutative, is almost similar and so we omit the proof here. Theorem 2.2. Let S be an idempotent semiring. Then σ∗ is the least distributive distributive congruence on S. Almost all varieties of idempotent semirings considered here are subvarieties of N. For this, the following result will have an important roll throughout the rest of this article. Lemma 2.3. An idempotent semiring S ∈ N if and only if it satisfies the identity: xz + xyz + xz ≈ xz. (2.1) Proof. First assume that S ∈ N and a, b, c ∈ S. Then ac = (a + abca + a)c = ac + abcac + ac = ac + abca(c + cabc + c) + ac = ac + abcac + abc + abcac + ac implies that ac + abc + abcac + ac = ac [add abc + abcac + ac to both sides from right]. Similarly this implies that ac = ac + abc + ac. Converse follows directly. Theorem 2.4. If an idempotent semiring S ∈ N, then σ is the least distributive lattice congruence on S. 5 Proof. Let a, b ∈ S be such that aηb. Then there exists x ∈ S such that axbxa = axbxa + a + axbxa and bxaxb = bxaxb + b + bxaxb. Since S ∈ N, ab + axb + ab = ab and ba + bxa + ba = ba, by Lemma 2.3. Then we have aba = a(ba + bxa + ba) = aba + bxa + aba = aba + (ab + axb + ab)xa + aba = aba + abxa + axbxa + abxa + aba = aba + abxa + axbxa + a + axbxa + abxa + aba = aba + abxa + axbxa + a + (aba + abxa+ axbxa + a + axbxa + abxa + aba) since (S, +) is a band = aba + abxa + axbxa + a + aba = aba + a + aba since (S, +) is a band. Similarly, bab = bab + b + bab. Thus aσb and so η ⊆ σ. Also σ is contained in every distributive lattice congruence. Hence σ = η is the least distributive lattice congruence on S. 3 The idempotent semirings such that η = D•, L•, R• In this section we show that the class of all idempotent semirings for which η = D•[L•, R•] is an equational class and so a variety. Here we find several systems of identities defining these varieties. On an idempotent semiring (S, +, ·) one may introduce the relations ≤l +, ≤l ·, ≤r +, ≤r · and ≤+, ≤· by the following: for a, b ∈ S, a ≤l + b ⇔ b = a + b a ≤l · b ⇔ a = ba a ≤r + b ⇔ b = b + a a ≤r · b ⇔ a = ab ≤+=≤l + ∩ ≤r + and ≤·=≤l · ∩ ≤r · The relations ≤l +, ≤l ·, ≤r +, ≤r · are quasi-orders and the relations ≤+ and ≤· are partial orders [6]. In [9], Pastijn and Zhao characterized the idempotent semirings S such that D• is the least distributive lattice congruence on S. Following result is already proved in [9]. Use of Theorem 2.2 shorten the proof which we would like to include here. 6 Theorem 3.1. Let S be an idempotent semiring. Then the following conditions are equivalent. 1. D• is the least distributive lattice congruence on S; 2. S ∈ N and D+ ⊆ D•; 3. S satisfies the identity Proof. (1) ⇒ (2) : Let a, b ∈ S. Since D• is a distributive lattice congruence, (a + ab)D•a. Then x ≈ xyx + x + xyx. (3.1) we have and so S ∈ N. a = a(a + ab)a ⇒ a = a + aba ⇒ a = a + aba + a Now let a, b ∈ S such that aD+b. Then a = a + b + a and b = b + a + b implies that bab = bab + b + bab and aba = aba + a + aba, and so aσb. Thus D+ ⊆ σ ⊆ η = D•. (2) ⇒ (3) : Similar to the proof of the Theorem 2.17 [9]. (3) ⇒ (1) : Let a, b ∈ S such that aηb. Since S satisfied the identity (3.1, it follows that S ∈ N and so η = σ, by Theorem 2.4. Hence we have aba = aba + a + aba and bab = bab + b + bab. Since S satisfies the identity (3.1), it follows that a = aba and b = bab. Thus aD•b and so η ⊆ D•. Also D• ⊆ σ = η. Therefore D• = η and so D• is the least distributive lattice congruence on S. Now wecharacterize the idempotent semirings for which L• is the least distributive lattice congruence. First we prove the following lemma. Lemma 3.2. Let S be an idempotent semiring. Then the following conditions are equivalent. 1. S satisfies the identity (1.1); 2. S ∈ N and R• ⊆ D+. Proof. (1) ⇒ (2) : Let a, b ∈ S such that aR•b. Then a = ba and b = ab. Now a = a + ab + a = a + b + a and b = b + ba + b = b + a + b implies that aD+b. Thus R• ⊆ D+. Also it follows trivially 7 that S ∈ N. (2) ⇒ (1) : Let a, b ∈ S. Then abR•aba implies that abD+aba. This implies that aba + ab + aba = aba ⇒ a + aba + ab + aba + a = a + aba + a ⇒ a + aba + ab + aba + a = a, (since S ∈ N) ⇒ a + aba + ab + aba + a = a + aba + ab + a ⇒ a = a + aba + ab + a ⇒ a + ab + a = a + aba + ab + a ⇒ a + ab + a = a Thus S satisfies the identity (1.1). Theorem 3.3. Let S be an idempotent semiring. Then the following conditions are equivalent. 1. L• is the least distributive lattice congruence on S; 2. D+ ⊆ L• and S satisfies the identity (1.1); 3. S ∈ N and R• ⊆ D+ ⊆ L•; 4. ≤l ·⊆≤+ for S; 5. S satisfies the identity 6. S satisfies the identity x ≈ xy + x + xy; x ≈ x(y + x + y). (3.2) (3.3) Proof. Equivalence of (2) and (3) follows from the Lemma 3.2. Equivalence of (5) and (6) follows trivially. (1) ⇒ (2) : Let a, b ∈ S. Then (a+ab)L•a implies that a = a(a+ab) = a+ab and so a = a+ab+a for all a, b ∈ S. Thus S satisfies the identity (1.1). Since D+ is the least semilattice congruence on the additive reduct (S, +), D+ ⊆ L•. (2) ⇒ (5) : Let a, b ∈ S. Then (a + b)D+(b + a) and so (a + b)L•(b + a). Therefore a + b = (a + b)(b + a) = ab + a + b + ba and b + a = (b + a)(a + b) = ba + b + a + ab 8 Also S satisfies the identity (1.1). Thus we have a = a(a + b) + a(b + a) ⇒ a = a(ab + a + b + ba) + a(ba + b + a + ab) ⇒ a = ab + a + ab + aba + ab + a + ab ⇒ ab + a + ab = ab + a + ab + aba + ab + a + ab ⇒ a = ab + a + ab for all a, b ∈ S. (5) ⇒ (1) : For every a, b ∈ S, we have a = ab + a + ab. This implies that a = aba + a + aba. Therefore D• is the least distributive lattice congruence on S, by Theorem 3.1. Also for every a, b ∈ S, a = ab + a + ab ⇒ a = a + ab + a + ab + a ⇒ a = a + ab + a Now let a, b ∈ S such that aD•b. Then a = aba and b = bab. Now a = aba ⇒ a + ab + a = aba + ab + aba ⇒ a = ab Similarly b = ba. Hence aL•b and so D• = L•. Thus L• is the least distributive lattice congruence on S. (4) ⇒ (5) : Let a, b ∈ S. Then ab ≤l · a. This implies that ab ≤+ a. Then ab + a = a = a + ab and so a = ab + a + ab. Thus S satisfies the identity (3.2. (5) ⇒ (4) : Let a, b ∈ S such that a ≤l · b. Then a = ba. Also a, b ∈ S implies that b = ba + b + ba. This implies that and so a ≤+ b. Therefore ≤l ·⊆≤+. b = a + b + a ⇒ b + a = b = a + b 9 The left-right dual of this result is stated as follows. Since the proof is similar to the above theorem, we omit. Theorem 3.4. Let S be an idempotent semiring. Then the following conditions are equivalent. 1. R• is the least distributive lattice congruence on S; 2. D+ ⊆ R• and S satisfies the identity (1.2); 3. S ∈ N and L• ⊆ D+ ⊆ R•; 4. ≤r · ⊆≤+ for S; 5. S satisfies the identity 6. S satisfies the identity x ≈ yx + x + yx; x ≈ (y + x + y)x. (3.4) (3.5) 4 Joins and Malcev's Products: The variety of all idempotent semirings S such that D•[L•, R•] is the least distributive lattice congruence on S will be denoted by D•[L•, R•]. Thus it follows, by Theorem 3.1, Theorem 3.3 and Theorem 3.4, that the varieties D•, L• and R• are determined by the additional identities (3.1), (3.2) and (3.4), respectively. For subvarieties V and W of I, the Mal'cev product V ◦ W of V and W (within I) is the class of all idempotent semirings S on which there exists a congruence ρ such that S/ρ ∈ W and such that the ρ-classes belong to V. In this section, different characterizations of the varieties D•, L• and R• by Malcev's product are given. Theorem 4.1. (i) L• = LZ• ◦ D. (ii) R• = RZ• ◦ D. Proof. (i) First we assume that S ∈ LZ• ◦ D. Then there exists a congruence relation δ on S such that S/δ ∈ D and such that each δ-class belongs to LZ•. Let a, b ∈ S. Since S/δ ∈ D, we have 10 (a + ab)δa and (ab + a)δa. Since the δ-class of a is a left zero band for the multiplication, we have a = a(ab + a) = ab + a and a = a(a + ab) = a + ab, and so a = (ab + a) + (a + ab) = ab + a + ab. Thus S satisfies the identity 3.2. Hence S ∈ L•, by Theorem 3.3. Thus LZ• ◦ D ⊆ L•. The reverse inclusion follows trivially. Hence L• = LZ• ◦ D. (ii) The proof is left-right dual to (i). Thus for S ∈ L• the multiplicative reduct (S, ·) is a left regular band and for S ∈ R• the multiplicative reduct (S, ·) is a right regular band. Lemma 4.2. For an idempotent semiring S the following conditions are equivalent: 1. S ∈ LN. 2. D• is a congruence on S and S/D• ∈ LZ+ ◦ D. Proof. (1) ⇒ (2) : Let S ∈ LN. Then S ∈ N and so D• is a congruence on S, by Theorem 2.11 [9]. Hence S/D• ∈ LN and D• S/D• ⊆ L+ S/D•. Thus S/D• ∈ LZ+ ◦ D. (2) ⇒ (1) : Assume that S/D• ∈ LZ+ ◦ D. Then L+ S/D• is the least distributive lattice congruence on S/D•, by Lemma 1.2. Let a be the D•-class of a in S. Then for all a, b ∈ S, we have (a + ab)L+a ⇒ a = a + ab ⇒ a = a + aba and so S ∈ LN. Theorem 4.3. (i) LN = R• ◦ (LZ+ ◦ D). (ii) RN = R• ◦ (RZ+ ◦ D). Proof. (Revise) Let S ∈ LN. Then by the above lemma, S/D• ∈ LZ+ ◦ D. Hence LN ⊆ R•◦(LZ+ ◦D). If S ∈ R• ◦(LZ+ ◦D) then there exists a congruence ρ on S such that S/ρ ∈ LZ+ ◦D and each ρ-class is in R•. Hence S/ρ ∈ LN, by Lemma 1.2, and so (a + aba)ρa for all a, b ∈ S. Since each ρ-class is in R•, it follows that a = a + aba. Thus LN = R• ◦ (LZ+ ◦ D). Let S1 and S2 be two semirings and D a distributive lattice. If there are two homomorphisms φ1 : S1 −→ D and φ2 : S2 −→ D onto D then the semiring S = {(s1, s2) ∈ S1 ×S2 φ1(s1) = φ2(s2)} is called a spined product of the two semirings S1 and S2 with respect to the spine D. 11 We show that every semiring S ∈ D• is a spined product of a semiring S1 ∈ L• and a semiring S2 ∈ R•. For this first we prove that the multiplicative reduct of each idempotent semiring S ∈ D• is a normal band. We do the groundwork by providing a sequence of useful lemmas. Lemma 4.4. Every idempotent semiring satisfies the following two identities: xyzx ≈ xyzx + xyxzx + xyzx and xyxzx ≈ xyxzx + xyzx + xyxzx. (4.1) (4.2) Proof. Let S be an idempotent semiring. Then D+ is a congruence on S and S/D+ ∈ Sl+. Hence the multiplicative reduct (S/D+, ·) is a regular band [10]. Then for all a, b, c ∈ S, abcaD+abaca and so abca = abca + abaca + abca and abaca = abaca + abca + abaca Lemma 4.5. For an idempotent semiring S the following conditions are equivalent: 1. S ∈ D•; 2. S satisfies both the identities: xz ≈ xz + xyz and xz ≈ xyz + xz; 3. S satisfies the identity: xz ≈ xyz + xz + xyz. (4.3) Proof. It is clear that (2) ⇒ (3) and (3) ⇒ (1). Assume that (1) holds. Then xz ≈ (xyzx + x + xyzx)(zxyz + z + zxyz) ≈ xyzxzxyz + xyzxz + xyzxzxyz + xzxyz + xz + xzxyz + xyzxzxyz + xyzxz + xyzxzxyz ≈ xyz + xyzxz + xyzxzxyz + xzxyz + xz + xzxyz + xyzxzxyz + xyzxz + xyz. This implies that xz ≈ xz + xyz and xz ≈ xyz + xz. Lemma 4.6. If S ∈ D• then S satisfies the identity: xyzx ≈ xzyx + xyzx + xzyx. (4.4) 12 Proof. Let S ∈ D•. Then for x, y, z ∈ S, xyzx ≈ (xzy + xy + xzy)zx ≈ xzyzx + xyzx + xzyzx ≈ xzy(zyx + zx + zyx) + xyzx + xzy(zyx + zx + zyx) ≈ xzyx + xzyzx + xzyx + xyzx + xzyx + xzyzx + xzyx, which implies that xyzx + xzyx ≈ xyzx and hence xzyx + xyzx + xzyx ≈ xyzx. Theorem 4.7. For S ∈ D• the multiplicative reduct (S, ·) is a normal band. Proof. Let S ∈ D• and x, y, z, ∈ S. Then xyzx ≈ xzyx + xyzx + xzyx. Also we have xzyx ≈ xzyx + xzyx + xzyx ⇒ xzyx ≈ xzyx + xyzx + xzyx + xyzx + xzyx ⇒ xzyx + xyzx + xzyx ≈ xzyx ⇒ xyzx ≈ xzyx. Theorem 4.8. For every idempotent S the following conditions are equivalent: 1. S ∈ LNB• ∩ D•; 2. S satisfies the identity: 3. S ∈ L•. xz ≈ xzy + xz + xzy; Proof. Equivalence of (2) and (3) is trivial. (1) ⇒ (2) : Assume that S ∈ LNB• ∩ D•. Then we have Since S ∈ LNB•, xz ≈ xyz + xz + xyz. xz ≈ xzy + xz + xzy. 13 (2) ⇒ (1) : Let x, y, z ∈ S. Then xz ≈ xzy + xz + xzy and xz ≈ xyz + xz + xyz. Also we have xyz ≈ xyzy + xyz + xyzy ⇒ xyz ≈ (xzy + xy + xzy)zy + xyz + (xzy + xy + xzy)zy ⇒ xyz ≈ xzy + xyzy + xzy + xyz + xzy + xyzy + xzy ⇒ xzy + xyz + xzy ≈ xyz. Now xzy ≈ xzy + xzy + xzy implies that xzy ≈ xzy + xyzy + xzy + xyzy + xzy ⇒ xzy ≈ xzy + xyzy + xzy ⇒ xzy ≈ xzy + xyzyz + xyzy + xyzyz + xzy ⇒ xzy ≈ xzy + xyz + xzy. Thus we get xyz ≈ xzy and so S ∈ LNB•. Also it follows that S ∈ D•. Thus the multiplicative reduct (S, ·) of every idempotent semiring S ∈ L• is a left normal band. Similarly, the multiplicative reduct (S, ·) of every idempotent semiring S ∈ R• is a right normal band. In D• we have the following derivation: x ≈ xyx + x + xyx ⇒ x ≈ x + xyx + x + xyx + x ⇒ x ≈ x + xyx + x Theorem 4.9. An idempotent semiring S ∈ D• if and only if S is a spined product of an idempotent semiring S1 ∈ L• and an idempotent semiring S2 ∈ R• with respect to a distributive lattice D. Proof. Let S ∈ D•. Then the multiplicative reduct (S, ·) is a regular band. So both L• and R• are congruences on the multiplicative reduct (S, ·). Let a, b, c ∈ S and aL•b. Then a = ab, b = ba and 14 caL•cb, acL•bc. Hence we have, (a + c)(b + c) = ab + ac + cb + c = a + ac + cbca + c (since caL•cb) = a + ac + (c + cac + c)bca + c = a + ac + cbca + cacbca + cbca + c = a + ac + cbca + ca + cbca + c (since caL•cb and L• ⊆ D•) = a + ac + (cbc + c + cbc)a + c = a + ac + ca + c = (a + c)2 = a + c and similarly (b + c)(a + c) = b + c. Thus (a + c)L•(b + c). Similarly (c + a)L•(c + b). Hence L• is a congruence on S. Now for any a, b ∈ S, a = aba + a + aba ⇒ a = a(ba + a + ba) implies that aL•(ba+ a+ ba). Hence S/L• ∈ R•. Similarly R• is a congruence on S and S/R• ∈ L•. Denote S1 = S/L• and S2 = S/R•. Also D = S/D• is a distributive lattice. Since L•, R• ⊆ D•, it follows that φ1 : S1 −→ D and φ2 : S2 −→ D defined by φ1(La) = Da and φ2(Ra) = Da are well defined surjective homomorphisms, where La [Ra, Da] is the L• [ R•, D•] -class containing a. Thus R = {(La, Ra) ∈ S1 × S2 a D• b} is a spined product of S1 and S2 with respect to D. Then the mapping θ : S −→ R defined by θ(a) = (La, Ra) is a monomorphism. Again if (La, Ra) ∈ R, then we have a D• b. Also since D• = L•oR•, there exists c ∈ S such that a L• c and c R• b. This implies that θ(c) = (La, Ra). Thus θ is onto. Thus D• ⊆ L• ∨ R•. Also L• ⊆ D• and R• ⊆ D• implies that L• ∨ R• ⊆ D•. Then we have the following corollary. Corollary 4.10. D• = L• ∨ R•. PROBLEMS Characterize the class of all idempotent semirings S such that σ is the least distributive lattice congruence on S. 15 References [1] Bhuniya, A. K. and Debnath, R., On band orthorings, communicated [arxiv: ] [2] Ghosh, S., Pastijn, F. and Zhao, X. Z., Varieties generated by ordered bands I, Or- der 22(2)(2005), 109-128. [3] J. S. Golan, The Theory of Semirings with Applications in Mathematics and Theoretical Com- puter Science, Pitman Monographs and Surveys in Pure and Applied Mathematics 54, Longman Scientific (1992). [4] Howie, J. M.: Fundamentals of Semigroup Theory, Clarendon Press, Oxford, 1995. [5] McKenzie, R. N., McNulty, G. F. and Taylor, W. F.: Algebras, Lattices, Varieties, Vol. 1, Wadsworth and Brooks/Cole, Monterey, 1987. [6] Nambooripad, K. S. S.: Structure of Regular Semigroups-I, Memoirs of the American Mathe- matical Society, No-224. [7] Pastijn, F. and Guo, Y. Q.: The lattice of idempotent distributive semiring varieties, Science in China(Series A), 29(1999), 391-407. [8] Pastijn, F. and Romanowska, A.: Idempotent distributive semirings I, Acta Sci. Math.(Szeged), 44(1982), 239-253. [9] Pastijn, F. and Zhao, Z. X.: Green's D-relation for the multiplicative reduct of an idempotent semiring, Archivum Mathematicum(Brno), 36(2000), 77-93. [10] Pastijn, F. and Zhao, Z. X.: Varieties of idempotent semirings with commutative addition, Algebra Universalis, 54(2005), 301-321. [11] Petrich, M.: Lectures in Semigroups, John Wiley & Sons, 1977. [12] Sen, M. K. and A. K. Bhuniya, The structure of almost idempotent semirings Algebra Collo- quium, 17 (Spec 1) (2010), 851 - 864. [13] Sen, M. K., Guo, Y. Q. and Shum, K. P.: A class of idempotent semirings, Semigroup Forum, 60(2000), 351-367. [14] Wang, Z. P., Zhou, Y. L. and Guo, Y. Q.: A note on band semirings, Semigroup Forum, 71(2005), 439-442. [15] Zhao, X. Z., Guo, Y. Q. and Shum, K. P.: D-subvarieties of the variety of idempotent semirings, Algebra Colloquium, 9(2002), 15-28. [16] Zhao, X. Z., K. P. Shum and Y. Q. Guo: L-subvarieties of the variety of idempotent semirings, Algebra Universalis, 46(2001), 75-96. 16
1909.08946	2	1909	2019-09-25T12:31:11	Free (tri)dendriform family algebras	[ "math.RA", "math.CO" ]	In this paper, we first prove that a Rota-Baxter family algebra indexed by a semigroup induces an ordinary Rota-Baxter algebra structure on the tensor product with the semigroup algebra. We show that the same phenomenon arises for dendriform and tridendriform family algebras. Then we construct free dendriform family algebras in terms of typed decorated planar binary trees. Finally, we generalize typed decorated rooted trees to typed valently decorated Schr\"oder trees and use them to construct free tridendriform family algebras.	math.RA	math	FREE (TRI)DENDRIFORM FAMILY ALGEBRAS YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ Abstract. In this paper, we first prove that a Rota-Baxter family algebra indexed by a semigroup induces an ordinary Rota-Baxter algebra structure on the tensor product with the semigroup alge- bra. We show that the same phenomenon arises for dendriform and tridendriform family algebras. Then we construct free dendriform family algebras in terms of typed decorated planar binary trees. Finally, we generalize typed decorated rooted trees to typed valently decorated Schroder trees and use them to construct free tridendriform family algebras. Contents Introduction Free dendriform family algebras 1. 2. Rota-Baxter family algebras and (tri)dendriform family algebras 2.1. Rota-Baxter family algebras 2.2. Dendriform family algebras and tridendriform family algebras 3. 3.1. Free dendriform algebras 3.2. Free dendriform family algebras 4. Free tridendriform family algebras 4.1. Free tridendriform algebras 4.2. Free tridendriform family algebras References 5. Appendix 1 3 3 4 7 7 8 16 16 17 25 26 1. Introduction The concept of Rota-Baxter family algebra is a generalization of Rota-Baxter algebras [18], which was proposed by Guo. It arises naturally in renormalization of quantum field theory ([7, Proposition 9.1] and [19, Theorem 3.7.2] ). The free objects were constructed in [28], in which the authors described free commutative Rota-Baxter family algebras, and also described free non- commutative Rota-Baxter family algebras via the method of Grobner-Shirshov bases. Dendriform algebras were introduced by Loday [20] in 1995 with motivation from algebraic K- theory. They have been studied quite extensively with connections to several areas in mathematics and physics, including operads [24], homology [9], arithmetics [23], Hopf algebras [21, 25, 26] 9 1 0 2 p e S 5 2 ] . A R h t a m [ 2 v 6 4 9 8 0 . 9 0 9 1 : v i X r a 0* Corresponding author. Date: September 26, 2019. 2010 Mathematics Subject Classification. 16W99, 16S10, 13P10, 08B20, Key words and phrases. Rota-Baxter algebra, Rota-Baxter family algebra, (tri)dendriform algebra, (tri)dendriform family algebra, typed decorated planar binary trees, typed valently decorated rooted trees. 1 2 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ and quantum field theory [8, 10], in connection with the theory of renormalization of Connes and Kreimer [4, 5, 6]. Some years later after [20], Loday and Ronco introduced the concept of tridendriform alge- bra in the study of polytopes and Koszul duality [25]. The construction of free (tri)dendriform algebras can be referred to [20, 21, 25]. The free objects play a crucial role in the study of any algebraic structures, such as the construction of free differential algebras in terms of differential monomials, the construction of free Rota-Baxter algebras [15] which is more involved, the free differential Rota-Baxter algebras [16] composing the construction of free differential algebras fol- lowed by that of the free Rota-Baxter algebras, and the free integro-differential algebras [12, 14] for analyzing the underlying algebraic structures of boundary problems for linear ordinary differ- ential equations. In [28], the authors proposed the concepts of (tri)dendriform family algebras and they obtained that Rota-Baxter family algebras induce (tri)dendriform family algebras. Besides, they construct free commutative (tri)dendriform family algebras. It is natural to ask how to use reduced planar rooted trees (also known as Schroder trees) to con- struct free (tri)dendriform family algebras. Here we exhibit a way to construct free (tri)dendriform family algebras via typed decorated Schroder trees, whose vertices are decorated by elements of a set X and edges are decorated by elements of a semigroup Ω. Typed decorated trees are used by Bruned, Hairer and Zambotti in [2] to give a systematic description of a canonical renormalisation procedure of stochastic PDEs. They also appear [27] in a context of low dimension topology and also appear in [3] for the description of combinatorial species. The concept of typed decorated planar binary trees enables us to construct free dendriform family algebras, and the concept of typed valently decorated Schroder trees will be used to construct free tridendriform family alge- bras. The layout of the paper is as follows. In Section 1, we prove that any Rota-Baxter family algebra R on a base ring k indexed by a semigroup Ω induces an ordinary Rota-Baxter algebra structure on R ⊗ kΩ (Theorem 2.4). The semigroup Ω at hand is not necessarily commuta- tive. Then we prove that Rota-Baxter family algebras can induce (tri)dendriform family algebras (Proposition 2.10). Finally, we give the relationship between (tri)dendriform family algebras and (tri)dendriform algebras (Theorem 2.11). Section 3 is devoted to typed decorated planar binary trees. We derive a useful recursive expression for typed decorated planar binary trees endowed with the operations {≺ω,(cid:31)ω ω ∈ Ω} introduced in the third section (Definition 3.7). We first con- struct dendriform family algebras (Proposition 3.10) in terms of typed decorated planar binary trees, then we prove the freeness of this dendriform family algebra (Theorem 3.12). In Section 4, closely related to Section 3, we first introduce the concept of typed valently decorated Schroder trees. Secondly, we construct tridendriform family algebras (Proposition 4.8) in terms of typed valently decorated Schroder trees, and finally prove the freeness of this tridendriform family al- gebra (Theorem 4.10). The related notions of matching Rota-Baxter algebra, matching dendriform algebra and match- ing pre-Lie algebra are addressed in the recent paper [13]. The main difference is that the param- eter at hand runs into a set Ω without any semigroup structure: see Remark 2. 2 (b) therein. FREE (TRI)DENDRIFORM FAMILY ALGEBRAS 3 Notation: In this paper, we fix a ring k and assume that an algebra is a k-algebra. Denote by Ω a semigroup, unless otherwise specified. 2. Rota-Baxter family algebras and (tri)dendriform family algebras In this section, we first recall the concepts of Rota-Baxter family algebras and (tri)dendriform family algebras. We then give a method to induce Rota-Baxter (resp. (tri)dendriform) algebras from Rota-Baxter (resp. (tri)dendriform) family algebras. 2.1. Rota-Baxter family algebras. Rota-Baxter algebras (first called Baxter algebras) are orig- inated in the work of the American mathematician Glen E. Baxter [1] in the realm of probability theory. Definition 2.1. Let λ be a given element in k. A Rota-Baxter algebra of weight λ is a pair (R, P) consisting of an algebra R with a linear operator P : R → R that satisfies the Rota-Baxter equation P(a)P(b) = P(cid:0)P(a)b + aP(b) + λab(cid:1), for a, b ∈ R. Then P is called a Rota-Baxter operator of weight λ. If further R is commutative, then (R, P) is called a commutative Rota-Baxter algebra of weight λ. The following is the concept of Rota-Baxter family algebra proposed by Guo, which arises ([7, 17]) Let Ω be a semigroup and λ ∈ k be given. A Rota-Baxter family of naturally in renormalization of quantum field theory [7, Proposition 9.1], see also [19]. Definition 2.2. weight λ on an algebra R is a collection of linear operators {Pω ω ∈ Ω} on R such that (cid:0)Pα(a)b + aPβ(b) + λab(cid:1), for a, b ∈ R and α, β ∈ Ω. Pα(a)Pβ(b) = Pαβ Then the pair (R, {Pω ω ∈ Ω}) is called a Rota-Baxter family algebra of weight λ. If further R is commutative, then (R, {Pω ω ∈ Ω}) is called a commutative Rota-Baxter family algebra of weight λ. Definition 2.3. Let (R, {Pω ω ∈ Ω}) and (R(cid:48), {P(cid:48) ω ω ∈ Ω}) be two Rota-Baxter family algebras of weight λ. A map f : R → R(cid:48) is called a Rota-Baxter family algebra morphism if f is an algebra homomorphism and f ◦ Pω = P(cid:48) The following theorem precises the link between Rota-Baxter family algebras and ordinary Rota- Baxter algebras. Theorem 2.4. Let (R,{Pω ω ∈ Ω}) be a Rota-Baxter family algebra of weight λ. Then (R⊗kΩ, P) is a Rota-Baxter algebra of weight λ, where P : R ⊗ kΩ → R ⊗ kΩ, x ⊗ ω (cid:55)→ Pω(x) ⊗ ω. Proof. For x, y ∈ R and α, β ∈ Ω, we have ω ◦ f for each ω ∈ Ω. P(x ⊗ α)P(y ⊗ β) =(cid:0)Pα(x) ⊗ α(cid:1)(Pβ(y) ⊗ β) =(cid:0)Pα(x)Pβ(y)(cid:1) ⊗ αβ (cid:0)Pα(x)y + xPβ(y) + λxy(cid:1) ⊗ αβ (cid:16)(cid:0)Pα(x)y + xPβ(y) + λxy(cid:1) ⊗ αβ (cid:17) = P(cid:0)Pα(x)y ⊗ αβ + xPβ(y) ⊗ αβ + λxy ⊗ αβ(cid:1) (cid:16)(cid:0)Pα(x) ⊗ α(cid:1)(y ⊗ β) + (x ⊗ α)(cid:0)Pβ(y) ⊗ β(cid:1) + λ(x ⊗ α)(y ⊗ β) (cid:17) = P(cid:0)P(x ⊗ α)(y ⊗ β) + (x ⊗ α)P(y ⊗ β) + λ(x ⊗ α)(y ⊗ β)(cid:1), = Pαβ = P = P as required. (cid:3) 4 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ 2.2. Dendriform family algebras and tridendriform family algebras. The concept of dendri- form family algebras was proposed in [28], as a generalization of dendriform algebras invented by Loday [20] in the study of algebraic K-theory. Definition 2.5. [20] A dendriform algebra is a k-module D with two binary operations ≺,(cid:31) such that for x, y, z ∈ D, (x ≺ y) ≺ z = x ≺ (y ≺ z + y (cid:31) z), (x (cid:31) y) ≺ z = x (cid:31) (y ≺ z), (x ≺ y + x (cid:31) y) (cid:31) z = x (cid:31) (y (cid:31) z). Definition 2.6. [28] Let Ω be a semigroup. A dendriform family algebra is a k-module D with a family of binary operations {≺ω,(cid:31)ω ω ∈ Ω} such that for x, y, z ∈ D and α, β ∈ Ω, (1) (2) (3) (x ≺α y) ≺β z = x ≺αβ (y ≺β z + y (cid:31)α z), (x (cid:31)α y) ≺β z = x (cid:31)α (y ≺β z), (x ≺β y + x (cid:31)α y) (cid:31)αβ z = x (cid:31)α (y (cid:31)β z). The concept of tridendriform family algebras was also introduced in [28], which is a general- ization of tridendriform algebras invented by Loday and Ronco [25] in the study of polytopes and Koszul duality. Definition 2.7. [25] A tridendriform algebra is a k-module T equipped with three binary oper- ations ≺,(cid:31) and · such that for x, y, z ∈ T, (x ≺ y) ≺ z = x ≺ (y ≺ z + y (cid:31) z + y · z), (x (cid:31) y) ≺ z = x (cid:31) (y ≺ z), (x ≺ y + x (cid:31) y + x · y) (cid:31) z = x (cid:31) (y (cid:31) z), (x (cid:31) y) · z = x (cid:31) (y · z), (x ≺ y) · z = x · (y (cid:31) z), (x · y) ≺ z = x · (y ≺ z), (x · y) · z = x · (y · z). Definition 2.8. [28] Let Ω be a semigroup. A tridendriform family algebra is a k-module T equipped with a family of binary operations {≺ω,(cid:31)ω ω ∈ Ω} and a binary operation · such that for x, y, z ∈ T and α, β ∈ Ω, (4) (5) (6) (7) (8) (9) (10) (x ≺α y) ≺β z = x ≺αβ (y ≺β z + y (cid:31)α z + y · z), (x (cid:31)α y) ≺β z = x (cid:31)α (y ≺β z), (x ≺β y + x (cid:31)α y + x · y) (cid:31)αβ z = x (cid:31)α (y (cid:31)β z), (x (cid:31)α y) · z = x (cid:31)α (y · z), (x ≺α y) · z = x · (y (cid:31)α z), (x · y) ≺α z = x · (y ≺α z), (x · y) · z = x · (y · z). FREE (TRI)DENDRIFORM FAMILY ALGEBRAS 5 Remark 2.9. When the semigroup Ω is taken to be the trivial monoid with one single element, a dendriform (resp. tridendriform) family algebra is precisely a dendriform (resp. tridendriform) algebra. Let RBFλ be the category of Rota-Baxter family algebras of weight λ, and let DDF (resp. DTF) be the category of dendriform (resp. tridendriform) family algebras. A functor ε : RBFλ → DTF has been introduced in [28, Theorem 4.4]. Further we have Proposition 2.10. Let Ω be a semigroup. (a) A Rota-Baxter family algebra (R, {Pω ω ∈ Ω}) of weight λ induces a dendriform family algebra (R, {≺ω,(cid:31)ω ω ∈ Ω}), where x ≺ω y := xPω(y) + λxy and x (cid:31)ω y := Pω(x)y, for x, y ∈ R. (b) A tridendriform family algebra (T,{≺ω,(cid:31)ω ω ∈ Ω},·) induces a dendriform family algebra (T,{≺(cid:48) ω,(cid:31)(cid:48) ω ω ∈ Ω}), where x ≺(cid:48) ω y := x ≺ω y + x · y and x (cid:31)(cid:48) ω y := x (cid:31)ω y, for x, y ∈ T. Proof. (a) For any x, y, z ∈ R and α, β ∈ Ω, (x ≺α y) ≺β z = (xPα(y) + λxy) ≺β z = (xPα(y) + λxy)Pβ(z) + λ(xPα(y) + λxy)z = xPα(y)Pβ(z) + λxyPβ(z) + λxPα(y)z + λ2xyz = xPαβ(yPβ(z) + Pα(y)z + λyz) + λx(yPβ(z) + Pα(y)z + λyz) = x ≺αβ (yPβ(z) + Pα(y)z + λyz) = x ≺αβ (y ≺β z + y (cid:31)α z), (x (cid:31)α y) ≺β z = Pα(x)y ≺β z = Pα(x)yPβ(z) + λPα(x)yz = Pα(x)(yPβ(z) + λyz) = Pα(x)(y ≺β z) = x (cid:31)α (y ≺β z), x (cid:31)α (y (cid:31)β z) = x (cid:31)α (Pβ(y)z) = Pα(x)(Pβ(y)z) = (Pα(x)Pβ(y))z = Pαβ(Pα(x)y + xPβ(y) + λxy)z = (Pα(x)y + xPβ(y) + λxy) (cid:31)αβ z = (x (cid:31)α y + x ≺β y) (cid:31)αβ z. (cid:3) (b) The proof is straightforward. Let η : RBFλ → DDF and γ : DTF → DDF be the functors obtained in Proposition 2.10. Then we have the commutative diagram: RBFλ ε / DTF η γ DDF The following result precises the link between (tri)dendriform family algebras and ordinary (tri)dendriform algebras. $ $ / 6 Theorem 2.11. dendriform algebra, where YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ (a) Let (D,{≺ω,(cid:31)ω ω ∈ Ω}) be a dendriform family. Then (D⊗ kΩ,≺,(cid:31)) is a (x ⊗ α) ≺ (y ⊗ β) :=(x ≺β y) ⊗ αβ (x ⊗ α) (cid:31) (y ⊗ β) :=(x (cid:31)α y) ⊗ αβ, for x, y ∈ D and α, β ∈ Ω. (b) Let (T,{≺ω,(cid:31)ω ω ∈ Ω},·) be a tridendriform family. Then (T ⊗ kΩ,≺,(cid:31),•) is a tridendri- form algebra , where (x ⊗ α) ≺ (y ⊗ β) :=(x ≺β y) ⊗ αβ, (x ⊗ α) (cid:31) (y ⊗ β) :=(x (cid:31)α y) ⊗ αβ, (x ⊗ α) • (y ⊗ β) :=(x · y) ⊗ αβ, for x, y ∈ T and α, β ∈ Ω. Proof. We only prove Item (b), as the proof of Item (a) is similar and easier. For x⊗α, y⊗β, z⊗γ ∈ (cid:16) T ⊗ kΩ, we get (x ⊗ α) ≺ (y ⊗ β) (cid:17) ≺ (z ⊗ γ) = (y ≺γ z + y (cid:31)β z + y · z) ⊗ βγ (cid:17) (y ≺γ z) ⊗ βγ + (y (cid:31)β z) ⊗ βγ + (y · z) ⊗ βγ (y ⊗ β) ≺ (z ⊗ γ) + (y ⊗ β) (cid:31) (z ⊗ γ) + (y ⊗ β) • (z ⊗ γ) , (cid:17) (cid:17) = = (cid:17) ⊗ αβγ (cid:16) (x (cid:31)α y) ≺γ z (cid:17) ≺ (z ⊗ γ) (cid:17) ⊗ αβγ (cid:17) ⊗ αβγ (by Eq. (4)) (cid:17) ≺ (z ⊗ γ) = (cid:17) ⊗ αβγ (by Eq. (5)) (cid:17) (cid:17) (y ≺γ z) ⊗ βγ (cid:17) (y ⊗ β) ≺ (z ⊗ γ) (y (cid:31)β z) ⊗ βγ (cid:16) (cid:16) (x ≺β y) ⊗ αβ (cid:16) (x ≺β y) ≺γ z = (x ⊗ α) ≺(cid:16) x ≺βγ (y ≺γ z + y (cid:31)β z + y · z) = (x ⊗ α) ≺(cid:16) = (x ⊗ α) ≺(cid:16) (cid:16) (cid:16) (x (cid:31)α y) ⊗ αβ = (x ⊗ α) (cid:31)(cid:16) x (cid:31)α (y ≺γ z) = (x ⊗ α) (cid:31)(cid:16) = (x ⊗ α) (cid:31)(cid:16) (cid:16) (cid:16) x (cid:31)α (y (cid:31)β z) (cid:16) (x ≺β y + x (cid:31)α y + x · y) (cid:31)αβ z (cid:16) (x ≺β y + x (cid:31)α y + x · y) ⊗ αβ (cid:16) (x ≺β y) ⊗ αβ + (x (cid:31)α y) ⊗ αβ + (x · y) ⊗ αβ (cid:16) (x ⊗ α) ≺ (y ⊗ β) + (x ⊗ α) (cid:31) (y ⊗ β) + (x ⊗ α) • (y ⊗ β) (cid:16) (x (cid:31)α y) ⊗ αβ = (x ⊗ α) (cid:31)(cid:16) (x (cid:31)α y) · z (cid:17) ⊗ αβγ (by Eq. (6)) (cid:17) (cid:31) (z ⊗ γ) (cid:17) (cid:31) (z ⊗ γ) (cid:17) ⊗ αβγ (by Eq. (7)) (cid:17) • (z ⊗ γ) (cid:17) ⊗ αβγ = (cid:16) (cid:17) (y · z) ⊗ βγ (cid:17) ⊗ αβγ x (cid:31)α (y · z) = = = = , = = (cid:17) (cid:31) (z ⊗ γ), (cid:16) (x ⊗ α) (cid:31) (y ⊗ β) (cid:17) ≺ (z ⊗ γ) = (x ⊗ α) (cid:31)(cid:16) (cid:17) (y ⊗ β) (cid:31) (z ⊗ γ) (cid:16) (x ⊗ α) (cid:31) (y ⊗ β) (cid:17) • (z ⊗ γ) = = (cid:16) (x ⊗ α) ≺ (y ⊗ β) (cid:17) • (z ⊗ γ) = (cid:16) (x ⊗ α) • (y ⊗ β) (cid:17) ≺ (z ⊗ γ) = (cid:16) (x ⊗ α) • (y ⊗ β) (cid:17) • (z ⊗ γ) = (cid:17) ⊗ αβγ (cid:17) ⊗ αβγ = FREE (TRI)DENDRIFORM FAMILY ALGEBRAS = (x ⊗ α) (cid:31)(cid:16) (cid:16) (cid:16) (x ≺β y) ⊗ αβ = (x ⊗ α) •(cid:16) x · (y (cid:31)β z) = (x ⊗ α) •(cid:16) (cid:16) (cid:16) (x · y) ⊗ αβ = (x ⊗ α) •(cid:16) x · (y ≺γ z) = (x ⊗ α) •(cid:16) (cid:16) (cid:16) (x · y) ⊗ αβ = (x ⊗ α) •(cid:16) x · (y · z) = (x ⊗ α) •(cid:16) (cid:17) (cid:17) • (z ⊗ γ) = (cid:16) (y ⊗ β) • (z ⊗ γ) , (cid:17) ⊗ αβγ (by Eq. (8)) (x ≺β y) · z (cid:17) (cid:17) (y (cid:31)β z) ⊗ βγ (cid:17) ≺ (z ⊗ γ) = (cid:16) (y ⊗ β) (cid:31) (z ⊗ γ) , (cid:17) ⊗ αβγ (by Eq. (9)) (x · y) ≺γ z (cid:17) (cid:17) (y ≺γ z) ⊗ βγ (cid:17) • (z ⊗ γ) = (cid:16) (y ⊗ β) ≺ (z ⊗ γ) , (cid:17) ⊗ αβγ (by Eq. (10)) (x · y) · z (cid:17) (cid:17) (y · z) ⊗ βγ (y ⊗ β) • (z ⊗ γ) = = . (cid:17) ⊗ αβγ 7 (cid:3) This completes the proof. 3. Free dendriform family algebras In this section, we construct free dendriform family algebras. For this, let us first briefly recall the construction of free dendriform algebras. For details, see [20, 21, 22, 26]. 3.1. Free dendriform algebras. Let X be a set. For n ≥ 0, let Yn, X be the set of planar binary trees with n + 1 leaves and with internal vertices decorated by elements of X. The unique tree (cid:41) (cid:12)(cid:12)(cid:12)(cid:12) x, y ∈ X with one leaf is denoted by . So we have Y0, X = {}. Here are the first few of them. (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) x, y, z ∈ X Y0, X = {}, Y1, X = (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X , Y2, X = Y3, X = (cid:40) (cid:26) (cid:27) x y x z y z y z y y x y x x z y x x , x , , , , ,  z  . For T ∈ Ym, X, U ∈ Yn, X and x ∈ X, the grafting ∨x of T and U over the vertex x is defined to be the planar binary tree T ∨x U ∈ Ym+n+1, X obtained by adding a new vertex decorated by x and joining the roots of T and U to the new vertex. Remark 3.1. In the graphical representation above, the edge pointing downwards is the root, the upper edges are the leaves. The other edges, joining two internal vertices, are called internal edges. 8 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ Example 3.2. Let T = and U = y with y ∈ X. Then T ∨x U = ∨x y = y x . Given a planar binary tree T ∈ Yn, X not equal to , there is a unique decomposition T = T l∨x T r for some x ∈ X. For example, Let DD(X) :=(cid:76) x = ∨x , y x y = ∨x , y x = ∨x y . kYn, X. Define binary operations ≺ and (cid:31) on DD(X) recursively by n≥1 (a) (cid:31) T := T ≺ := T and ≺ T := T (cid:31) := 0 for T ∈ Yn, X with n ≥ 1. (b) For T = T l ∨x T r and U = Ul ∨y Ur, define T ≺ U := T l ∨x (T r ≺ U + T r (cid:31) U), T (cid:31) U := (T ≺ Ul + T (cid:31) Ul) ∨y Ur. Let j be the unique linear map from X into DD(X) defined by j(x) = x for x ∈ X. The following result is well-known. Theorem 3.3. [21] Let X be a set. Then (DD(X),≺,(cid:31)), together with the map j, is the free dendriform algebra on X. 3.2. Free dendriform family algebras. In this subsection, we apply typed decorated planar bi- nary trees to construct free dendriform family algebras. For this, let us first recall typed decorated rooted trees studied in [2, 11]. For a rooted tree T, denote by V(T) (resp. E(T)) the set of its vertices (resp. edges). Definition 3.4. [2] Let X and Ω be two sets. An X-decorated Ω-typed (abbreviated typed decorated) rooted tree is a triple T = (T, dec, type), where (a) T is a rooted tree. (b) dec : V(T) → X is a map. (c) type : E(T) → Ω is a map. In other words, vertices of T are decorated by elements of X and edges of T are decorated by elements of Ω. As we can see from the examples drawn above, the graphical representation of planar binary trees is slightly different from the graphical representation of rooted trees used in [2, 11]: the root and the leaves are now edges rather than vertices. Here the set V(T) must be replaced by the set IV(T) of internal vertices of T, and the set E(T) must be replaced by the set IE(T) of internal edges, i.e. edges which are neither a leaf nor the root. Hence we propose the following definition: Definition 3.5. Let X and Ω be two sets. An X-decorated Ω-typed planar binary tree is a triple T = (T, dec, type), where (a) T is a planar binary tree. (b) dec : IV(T) → X is a map. (c) type : IE(T) → Ω is a map. FREE (TRI)DENDRIFORM FAMILY ALGEBRAS 9 Note that in the definition of dendriform family algebras, Ω is a semigroup. However in the following construction of free dendriform family algebras, if the semigroup Ω has no identity element, it will be convenient to consider the monoid Ω1 := Ω(cid:116){1} obtained from Ω by adjoining an identity: 1ω := ω1 := ω, for ω ∈ Ω and 11 := 1. Let X be a set and let Ω be a semigroup. For n ≥ 0, let Yn := Yn, X, Ω be the set of X-decorated Ω1-typed planar binary trees with n + 1 leaves, such that leaves are decorated by the identity 1 in Ω1 and internal edges are decorated by elements of Ω. Let us expose some examples for better understanding. For convenience, we omit the decoration 1 in the sequel. (cid:26) x , Y0 = {}, Y1 = z β y α x Y3 =  (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X , Y2 = (cid:40) z β y α x , z β y αx , y α x , y z βx α y α x , (cid:12)(cid:12)(cid:12)(cid:12) x, y ∈ X, α ∈ Ω z βx y α , . . . , (cid:41)  . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) x, y, z ∈ X, α, β ∈ Ω For T ∈ Ym, U ∈ Yn, x ∈ X and α, β ∈ Ω1, the grafting ∨x, (α, β) of T and U over x and (α, β) is defined to be the tree T ∨x, (α, β) U ∈ Ym+n+1, obtained by adding a new vertex decorated by x and joining the roots of T and U with the new vertex via two new edges decorated by α and β respectively. The depth dep(T) of a rooted tree T is the maximal length of linear chains of vertices from the root to the leaves of the tree. For example, (cid:18) dep x (cid:19) (cid:18) y x (cid:19) α = 1 and dep y with x, y ∈ X. For z ∈ X and α, β ∈ Ω, = 2. Example 3.6. Let T = x and U = T ∨z, (α, β) U = x ∨z, (α, β) y x y = α β z . Given a typed decorated planar binary tree T ∈ Yn not equal to , there is a unique decomposi- tion T = T l ∨x, (α, β) T r for some x ∈ X and α, β ∈ Ω1. ∨x, (α, 1) , y α = ∨x, (1, α) x y , For example, x = ∨x, (1, 1) , y α x y z α β x y y = = Denote by z . ∨x, (α, β) (cid:77) n≥1 DD(X, Ω) := kYn. Definition 3.7. Let X be a set and let Ω be a semigroup. We define binary operations {≺ω,(cid:31)ω ω ∈ Ω} on DD(X, Ω) recursively on dep(T) + dep(U) by (a) (cid:31)ω T := T ≺ω := T and ≺ω T := T (cid:31)ω := 0 for ω ∈ Ω and T ∈ Yn with n ≥ 1. 10 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ (b) For T = T l ∨x, (α1, α2) T r and U = Ul ∨y, (β1, β2) Ur, define T ≺ω U := T l ∨x, (α1, α2ω) (T r ≺ω U + T r (cid:31)α2 U), T (cid:31)ω U := (T ≺β1 Ul + T (cid:31)ω Ul) ∨y, (ωβ1, β2) Ur, where ω ∈ Ω. In the following, we employ the convention that (cid:31)1 T := T ≺1 := T and ≺1 T := T (cid:31)1 := 0. Example 3.8. Let T = x and U = y with x, y ∈ X. For ω ∈ Ω, we have (11) (12) (13) x x ≺ω (cid:31)ω y y (cid:18) ≺ω = ( ∨x, (1, 1) ) ≺ω = ∨x, (1, ω) = ∨x, (1, ω) y y y + (cid:31)1 y (cid:19) (by Eq. (11)) (by Item (a) of Definition 3.7 and Eq. (13)) , y ω x x x x (cid:18) = = = = yx = ω . (cid:31)ω ( ∨y, (1, 1) ) ≺1 + ∨y, (ω, 1) x (cid:31)ω (cid:19) ∨y, (ω, 1) (by Eq. (12)) (by Item (a) of Definition 3.7 and Eq. (13)) u Example 3.9. Let T = α β x z y with x, y, z, u ∈ X and α, β ∈ Ω. For ω ∈ Ω, (cid:18) T ≺ω U = = = z z z and U = ∨x, (α, β) ∨x, (α, βω) ∨x, (α, βω) (cid:18) (cid:18) y z (cid:19) ≺ω ≺ω y + , (cid:18) u u u ω yu βx βω (cid:19) y y (cid:19) yu β + u (cid:31)β y (by Example 3.8) z u α β x z ≺1 + α β x u (cid:31)ω (cid:19) ∨y, (ω, 1) (by Item (a) of Definition 3.7 and Eq. (13)) z = α u ωx βω + α (cid:31)ω ( ∨y, (1, 1) ) = ∨y, (ω, 1) T (cid:31)ω U = α β x u z z x u = α β z u βα ω = x y . FREE (TRI)DENDRIFORM FAMILY ALGEBRAS 11 Proposition 3.10. Let X be a set and let Ω be a semigroup. Then (DD(X, Ω),{≺ω,(cid:31)ω ω ∈ Ω}) is a dendriform family algebra, generated by x . (cid:26) (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X Proof. We proceed in two steps to prove the result. Step 1: We first prove (DD(X, Ω),{≺ω,(cid:31)ω ω ∈ Ω}) is a dendriform family algebra. Let T = T l ∨x, (α1, α2) T r, U = Ul ∨y, (β1, β2) Ur, W = Wl ∨z, (γ1, γ2) Wr ∈ DD(X, Ω). Then we apply induction on dep(T) +dep(U) +dep(W) ≥ 3. For the initial step dep(T) +dep(U) + dep(W) = 3, we have T = x , U = y and W = z and so = = = y x (cid:18) (T ≺β U) ≺γ W (cid:32) (cid:18) ∨x, (1, β) ≺β ∨x, (1, β) (cid:19) ≺γ (cid:19) ≺γ y (cid:18) ( ∨x, (1, 1) ) ≺β (cid:19)(cid:33) (cid:19) ≺γ y z ≺γ z (by Eq. (11)) z + (cid:31)1 = y (by Item (a) in Definition 3.7 and Eq. (13)) y + (cid:31)β z (by Eq. (11)) (cid:19) y (cid:18) ≺β (cid:18) (cid:18) (cid:32) y y ≺βγ z γ (cid:18) (cid:18) z z (cid:19) ≺γ y + β z = ∨x, (1, βγ) = ∨x, (1, βγ) = ∨x, (1, βγ) (by Example 3.8) (cid:18) y z γ y + β z + (cid:31)1 y z γ y + β z (cid:19)(cid:33) (cid:19) (cid:19) (by Item (a) in Definition 3.7 and Eq. (13)) (cid:18) = ( ∨x, (1, 1) ) ≺βγ (cid:18) = x ≺βγ ≺βγ = x z γ y y z (cid:19) y z γ y + β y + β ≺γ z z verifying Eq. (1). Next, x y = T ≺βγ (U ≺γ W + U (cid:31)β W), (cid:18) (cid:33) ≺1 (cid:19) ∨y, (α, 1) = (cid:18) (T (cid:31)α U) ≺γ W (cid:32)(cid:18) (cid:18) (cid:31)α (cid:31)α + (cid:19) ≺γ ∨y, (α, 1) (cid:19) ≺γ (cid:18) ≺γ ∨y, (α, γ) + (cid:31)1 z z z x x x x x z = = = = ≺γ (cid:19) (by Eq. (11)) (cid:19) y + (cid:31)β z (cid:19) ≺γ (cid:31)α ( ∨y, (1, 1) ) z z (by Eq. (12)) (by Item (a) in Definition 3.7 and Eq. (13)) (by Eq. (11)) YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ (by Item (a) in Definition 3.7 and Eq. (13)) z (by Item (a) in Definition 3.7 and Eq. (13)) (by Item (a) in Definition 3.7 and Eq. (13)) (cid:19) y ≺γ z (cid:18) 12 = = = = = x x x x x ≺1 (cid:19) ∨y, (α, γ) (cid:19) (cid:18) ≺γ z z z x ∨y, (α, γ) (cid:18) ∨y, (1, γ) (cid:31)α + (cid:32) (cid:31)α (cid:18) (cid:31)α (cid:31)α ∨y, (1, γ) ( ∨y, (1, 1) ) ≺γ z (cid:19) + (cid:31)1 = (by Eq. (12)) (cid:19)(cid:33) (cid:18) (cid:31)α = T (cid:31)α (U ≺γ W), verifying Eq. (2). Finally, z x (cid:19) = y (by Item (a) in Definition 3.7 and Eq. (13)) ≺β (by Eq. (12)) y z y y x x x x x y y = = = = + (cid:31)β (cid:31)β + (cid:18) T (cid:31)α (U (cid:31)β W) (cid:32)(cid:18) (cid:31)α (cid:18) (cid:31)α (cid:31)α (cid:31)α ∨z, (β, 1) (cid:19) y (cid:19) ∨z, (αβ, 1) y (cid:19) (cid:31)αβ + (cid:18) y (cid:19) (cid:31)αβ ( ∨z, (1, 1) ) (cid:18) (cid:18) (cid:32)(cid:18) (cid:18) (cid:18) = (T (cid:31)α U + T ≺β U) (cid:31)αβ W. (cid:31)α ≺β yx α yx α yx α + + = = = + = x β x β y x x y x (cid:31)α (cid:18) (cid:33) ≺1 (cid:19) ∨z, (β, 1) (cid:19) ∨z, (αβ, 1) (cid:33) y (cid:19) ≺1 (cid:19) (cid:31)αβ (by Eq. (12)) + y x β z (by Item (a) in Definition 3.7 and Eq. (13)) yx α + x β (by Example 3.8) (cid:19) (cid:31)β ( ∨, (z, 1,1) ) (by Eq. (12)) ∨z, (αβ, 1) (by Example 3.8) This completes the proof of the initial step. Assume that the conclusion holds for dep(T) + dep(U) + dep(W) = k and consider the induction step of dep(T) + dep(U) + dep(W) = k + 1 ≥ 4. We have = (cid:17) ≺γ W (by Eq. (11)) (cid:16) (T ≺β U) ≺γ W (cid:16) T l ∨x, (α1, α2β) (T r ≺β U + T r (cid:31)α2 U) (T r ≺β U + T r (cid:31)α2 U) ≺γ W + (T r ≺β U + T r (cid:31)α2 U) (cid:31)α2β W = T l ∨x, (α1, α2βγ) (cid:16) (cid:17) (T r ≺β U) ≺γ W + (T r (cid:31)α2 U) ≺γ W + T r (cid:31)α2 (U (cid:31)β W) (by the induction hypothesis and Eq. (3)) = T l ∨x, (α1, α2βγ) (by Eq. (11)) (cid:17) = T l ∨x, (α1, α2βγ) FREE (TRI)DENDRIFORM FAMILY ALGEBRAS (cid:16) (cid:17) T r ≺βγ (U ≺γ W + U (cid:31)β W) + T r (cid:31)α2 (U ≺γ W) + T r (cid:31)α2 (U (cid:31)β W) (cid:16) (cid:17) (by the induction hypothesis and Eq. (1)) T r ≺βγ (U ≺γ W + U (cid:31)β W) + T r (cid:31)α2 (U ≺γ W + U (cid:31)β W) 13 = T l ∨x, (α1, α2βγ) = (T l ∨x, (α1, α2) T r) ≺βγ (U ≺γ W + U (cid:31)β W) = T ≺βγ (U ≺γ W + U (cid:31)β W). (by Eq. (11)) This proves Eq. (1) in Definition 2.6. Next, we prove Eq. (2): = (cid:16) (T (cid:31)α Ul + T ≺β1 Ul) ∨y, (αβ1, β2) Ur(cid:17) ≺γ W (by Eq. (12)) (T (cid:31)α U) ≺γ W (cid:16) (cid:17) = (T (cid:31)α Ul + T ≺β1 Ul) ∨y, (αβ1, β2γ) (Ur ≺γ W + Ur (cid:31)β2 W) (cid:16) = T (cid:31)α Ul ∨y, (β1, β2γ) (Ur ≺γ W + Ur (cid:31)β2 W) = T (cid:31)α (Ul ∨y, (β1, β2) Ur) ≺γ W = T (cid:31)α (U ≺γ W). (by Eq. (11)) (by Eq. (12)) (cid:17) (by Eq. (11)) = = = Finally, we prove Eq. (3): = (by Eq. (12)) (by Eq. (12)) (cid:16) (cid:17) T (cid:31)α (U (cid:31)β W) (cid:16) (U (cid:31)β Wl + U ≺γ1 Wl) ∨z, (βγ1, γ2) Wr(cid:17) = T (cid:31)α U (cid:31)β (Wl ∨z, (γ1, γ2) Wr) (cid:16) (cid:17) ∨z, (αβγ1, γ2) Wr = T (cid:31)α (cid:16) (cid:17) ∨z, (αβγ1, γ2) Wr T (cid:31)α (U (cid:31)β Wl + U ≺γ1 Wl) + T ≺βγ1 (U (cid:31)β Wl + U ≺γ1 Wl) (cid:16) (T (cid:31)α U + T ≺β U) (cid:31)αβ Wl + (T (cid:31)α U) ≺γ1 Wl + (T ≺β U) ≺γ1 Wl(cid:17) ∨z, (αβγ1, γ2) Wr T (cid:31)α (U (cid:31)β Wl) + T (cid:31)α (U ≺γ1 Wl) + T ≺βγ1 (U (cid:31)β Wl + U ≺γ1 Wl) (cid:16) (T (cid:31)α U + T ≺β U) (cid:31)αβ Wl + (T (cid:31)α U + T ≺β U) ≺γ1 Wl(cid:17) ∨z, (αβγ1, γ2) Wr = (T (cid:31)α U + T ≺β U) (cid:31)αβ (Wl ∨z, (γ1, γ2) Wr) = (T (cid:31)α U + T ≺β U) (cid:31)αβ W. (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X This completes the proof of step 1. . Let T = T l ∨x, (α1, α2) T r with Step 2: We show that DD(X, Ω) is generated by (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X x ∈ X and α1, α2 ∈ Ω1. We employ induction on dep(T) ≥ 1. For the initial step of dep(T) = 1, we have (by the induction hypothesis and Eqs. (1)-(3)) (cid:26) ∈(cid:26) T = ∨x, (1, 1) = (by Eq. (12)) x x x . Assume that the conclusion holds for T with dep(T) = k and consider the induction step of dep(T) = k + 1 ≥ 2. Since T l and T r can not be simultaneously, there exist three cases to consider. Case 1: T l = and T r (cid:44) . Then α1 = 1 and T = T l ∨x, (α1, α2) T r = ∨x, (1, α2) T r 14 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ (by Item (a) in Definition 3.7 and Eq. (13)) (cid:26) (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X and so T is generated by = ∨x, (1, α2) ( ≺α2 T r + (cid:31)1 T r) = ( ∨x, (1, 1) ) ≺α2 T r (by Eq. (11)) = x ≺α2 T r. (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X By the induction hypothesis, T r is generated by (cid:26) Case 2: T l (cid:44) and T r = . This case is similar to Case 1. Case 3: T l (cid:44) and T r (cid:44) . Then x x . T = T l ∨x, (α1, α2) T r = T l ∨x, (α1, α2) ( ≺α2 T r + (cid:31)1 T r) (cid:16) (T l (cid:31)α1 + T l ≺1 ) ∨x, (α1, 1) (cid:17) ≺α2 T r = (T l ∨x, (α1, 1) ) ≺α2 T r (cid:16) (cid:18) T l (cid:31)α1 ( ∨x, (1, 1) ) T l (cid:31)α1 (cid:17) ≺α2 T r (cid:19) ≺α2 T r. (by Eq. (11)) = = = x (by Eq. (12)) (cid:26) by x (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X By the induction hypothesis, T l and T r are generated by . This completes the proof. (by Item (a) in Definition 3.7 and Eq. (13)) (by Item (a) in Definition 3.7 and Eq. (13)) (cid:26) x (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X and so T is generated (cid:3) Now we arrive at our main result in this section. Definition 3.11. Let X be a set and let Ω be a semigroup. A free dendriform family algebra on X is a dendriform family algebra (D,{≺ω,(cid:31)ω ω ∈ Ω}) together with a map j : X → D that satisfies the following universal property: for any dendriform family algebra (D(cid:48),{≺(cid:48) ω,(cid:31)(cid:48) ω ω ∈ Ω}) and map f : X → D(cid:48), there is a unique dendriform family algebra morphism ¯f : D → D(cid:48) such that f = ¯f ◦ j. The free dendriform family algebra on X is unique up to isomorphism. Let j : X → DD(X, Ω) be the map defined by j(x) = Theorem 3.12. Let X be a set and let Ω be a semigroup. Then (DD(X, Ω),{≺ω,(cid:31)ω ω ∈ Ω}), together with the map j, is the free dendriform family algebra on X. Proof. By Proposition 3.10, we are left to show that (DD(X, Ω),{≺ω,(cid:31)ω ω ∈ Ω}) satisfies the ω ω ∈ Ω}) be a dendriform family algebra. First of all, universal property. For this, let (D,{≺(cid:48) if there exists a dendriform family algebra morphism ¯f : DD(X, Ω) → D extending f : X → D (cid:19) ) = f (x) for any x ∈ X, then such an ¯f is unique, due to the fact that the in the sense that ¯f ( collection of trees generate the dendriform family algebra DD(X, Ω). for x ∈ X. ω,(cid:31)(cid:48) (cid:18) x x x x∈X Now let us define a linear map ¯f : DD(X, Ω) → D, T (cid:55)→ ¯f (T) FREE (TRI)DENDRIFORM FAMILY ALGEBRAS 15 by induction on dep(T) ≥ 1. Let us write T = T l ∨x, (α1, α2) T r with x ∈ X and α1, α2 ∈ Ω1. For the initial step dep(T) = 1, we have T = x (14) Assume that ¯f (T) has been defined for T with dep(T) = k and consider the induction step of dep(T) = k + 1 ≥ 2. Note that T l and T r can not be simultaneously and define for some x ∈ X and define ¯f (T) := f (x). ¯f (T) := ¯f (T l ∨x, (α1, α2) T r) ¯f (T r), α1 f (x),  f (x) ≺(cid:48) (cid:0) ¯f (T l) (cid:31)(cid:48) α1 f (x)(cid:1) ≺(cid:48) ¯f (T l) (cid:31)(cid:48) := α2 α2 if T l = (cid:44) T r; if T l (cid:44) = T r; if T l (cid:44) (cid:44) T r. ¯f (T r), (15) We are left to prove that ¯f is a morphism of dendriform family algebras: ¯f (U) and ¯f (T (cid:31)ω U) = ¯f (T) (cid:31)(cid:48) ¯f (T ≺ω U) = ¯f (T) ≺(cid:48) ω ¯f (U), ω in which we only prove the first equation by induction on dep(T) + dep(U) ≥ 2, as the proof of the second one is similar. Write T = T l ∨x, (α1, α2) T r and U = Ul ∨y, (β1, β2) Ur. and U = x For the initial step dep(T) + dep(U) = 2, we have T = So y for some x, y ∈ X. = ¯f (cid:18) ≺ω (cid:18) (cid:32) ( ∨x, (1, 1) ) ≺ω ¯f (T ≺ω U) = ¯f (cid:18) ∨x, (1, ω) ∨x, (1, ω) (cid:18) = ¯f (cid:19) ≺(cid:48) (cid:18) = f (x) ≺(cid:48) = ¯f = ¯f (T) ≺(cid:48) ¯f ¯f (U). (cid:18) ¯f ω ω y y x ω (cid:19) (cid:19) y y y (cid:19) + (cid:31)1 (cid:19)(cid:33) y (by Eq. (11)) (by Item (a) in Definition 3.7 and Eq. (13)) (by Eq. (15)) (cid:19) (by Eq. (14)) Assume that the conclusion holds for T and U in DD(X, Ω) with dep(T) + dep(U) = k and consider the case when dep(T) + dep(U) = k + 1 ≥ 3, there are four cases to consider. Case 1: T l = and T r (cid:44) . Then α1 = 1 and ¯f (T ≺ω U) = ¯f(cid:0)( ∨x, (1, α2) T r) ≺ω U(cid:1) (cid:0) ¯f (T r) ≺(cid:48) ¯f (T r)(cid:1) ≺(cid:48) = ¯f(cid:0) ∨x, (1, α2ω) (T r ≺ω U + T r (cid:31)α2 U)(cid:1) =(cid:0) f (x) ≺(cid:48) ¯f (T r ≺ω U + T r (cid:31)α2 U) = f (x) ≺(cid:48) = f (x) ≺(cid:48) (by the induction hypothesis on dep(T) + dep(U)) ¯f (U) + ¯f (T r) (cid:31)(cid:48) ¯f (U)(cid:1) (by Eq. (15)) α2ω α2ω α2 ω (by Eq. (11)) ω ¯f (U) ¯f (U) (by Eq. (1)) (by Eq. (15)) α2 = ¯f ( ∨x, (1, α2) T r) ≺(cid:48) = ¯f (T) ≺(cid:48) ¯f (U). ω ω 16 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ Case 2: T l (cid:44) and T r = . This case is similar to Case 1. Case 3: T l (cid:44) and T r (cid:44) . Then ¯f (T ≺ω U) = ¯f(cid:0)(T l ∨x, (α1, α2) T r) ≺ω U(cid:1) = ¯f(cid:0)T l ∨x, (α1, α2ω) (T r (cid:31)α2 U + T r ≺ω U)(cid:1) =(cid:0) ¯f (T l) (cid:31)(cid:48) =(cid:0) ¯f (T l) (cid:31)(cid:48) (cid:16)(cid:0) ¯f (T l) (cid:31)(cid:48) α1 f (x)(cid:1) ≺(cid:48) α1 f (x)(cid:1) ≺(cid:48) α1 f (x)(cid:1) ≺(cid:48) (cid:0) ¯f (T r) (cid:31)(cid:48) (cid:17) ≺(cid:48) α2ω α2ω = α2 α2 (by the induction hypothesis on dep(T) + dep(U)) (by Eq. (1)) ¯f (U) ¯f (T r (cid:31)α2 U + T r ≺ω U) ¯f (U) + ¯f (T r) ≺(cid:48) ω (by Eq. (11)) ¯f (U)(cid:1) (by Eq. (15)) ¯f (T r) ¯f (U) ω (by Eq. (15)) = ¯f (T l ∨x, (α1, α2) T r) ≺(cid:48) = ¯f (T) ≺(cid:48) ¯f (U). ω ω x for some x ∈ X and we have ¯f (T ≺ω U) = ¯f(cid:0) ∨x,(1,ω) U(cid:1) = f (x) ≺(cid:48) = ¯f (T) ≺(cid:48) ω ω ¯f (U) ¯f (U) Case 4: T l = and T r = . Then T = according to (15). This completes the proof. (cid:3) 4. Free tridendriform family algebras x is decorated by x ∈ X while the vertex of In this section, we construct free tridendriform family algebras in terms of typed valently dec- orated Schroder trees. For this, let us first recall the construction of free tridendriform algebras. See [21, 25] for more details. 4.1. Free tridendriform algebras. Let X be a set. For n ≥ 0, let Tn, X be the set of planar rooted trees with n + 1 leaves and with vertices valently decorated by elements of X, in the sense that if a vertex has valence k, then the vertex is decorated by an element in Xk−1. For example, the vertex is decorated by (x, y) ∈ X2. Here are of the first few of them. T0, X = {}, (cid:41) (cid:12)(cid:12)(cid:12)(cid:12) x, y ∈ X  (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) x, y, z ∈ X For T (i) ∈ Tni, X, 0 ≤ i ≤ k, and x1, . . . , xk ∈ X, the grafting(cid:87) of T (i) over (x1, . . . , xk) is (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X  . T2, X = T3, X = T1, X = (cid:40) zyx (cid:27) (cid:26) , . . . y z x y x y x y x x x y z y x y y x x y , x z z , , z , , , , , , Any planar rooted tree T can be uniquely expressed as such a grafting of lower depth trees. For example, (cid:95)k+1 (T (0), . . . , T (k)). x1,...,xk (cid:95) x y = (, , ). x,y (a) (cid:31) T := T ≺ := T, ≺ T := T (cid:31) := 0 and · T := T · := 0 for T ∈ Tn, X with n ≥ 1. FREE (TRI)DENDRIFORM FAMILY ALGEBRAS 17 kTn, X. Define binary operations ≺,(cid:31) and · on DT(X) recursively on dep(T) + n≥1 dep(U) by Let DT(X) :=(cid:76) (b) For(cid:87)m+1 x1,...,xm(T (0), . . . , T (m)) and(cid:87)n+1 (cid:95)m+1 (cid:95)n+1 (cid:95)m+n+1 T ≺ U := T (cid:31) U := T · U := Let j : X → DT(X) be the map defined by j(x) = x1,...,xm−1,xm, y1,...,yn x1,...,xm−1,xm y1,y2,...,yn y1,...,yn(U(0), . . . , U(n)), define (T (0), . . . , T (m−1), T (m) (cid:31) U + T (m) ≺ U + T (m) · U), (T (cid:31) U(0) + T ≺ U(0) + T · U(0), U(1), . . . , U(n)), (T (0), . . . , T (m−1), T (m) (cid:31) U(0) + T (m) ≺ U(0) + T (m) · U(0), U(1), . . . , U(n)). for any x ∈ X. The following result is x well-known. Theorem 4.1. [21, 25] Let X be a set. Then (DT(X),≺,(cid:31),·), together with j, is the free tridendri- form algebra on X. 4.2. Free tridendriform family algebras. In this subsection, we introduce typed valently dec- orated Schroder trees to construct free tridendriform family algebras. Let us first propose Definition 4.2. Let X and Ω be two sets. A X-valently decorated Ω-typed (abbreviated typed valently decorated) Schroder tree is a triple T = (T, dec, type), where (a) T is a Schroder tree. (b) dec : IV(T) → (cid:116)n≥1Xn is a map sending any vertex of arity n + 1 to Xn. (c) type : IE(T) → Ω is a map. Remark 4.3. The trees in DT(X) employed to construct free tridendriform algebras are indeed X-valently decorated {(cid:63)}-typed Schroder trees, where {(cid:63)} is any set with one single element. Let X be a set and let Ω be a semigroup. For n ≥ 0, let Tn := Tn, X, Ω be the set of X-valently decorated Ω1-typed Schroder trees with n + 1 leaves, such that the leaves are decorated by the identity 1 in Ω1 and internal edges are decorated by elements of Ω. Let us expose some examples. Note that the edge decoration 1 is omitted for convenience. x y (cid:41) (cid:12)(cid:12)(cid:12)(cid:12) x, y ∈ X, α ∈ Ω  . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) x, y, z ∈ X, α, β ∈ Ω , (cid:26) (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X , (cid:40) y α x , z yα x y x , x x , z β y z y α y α T2 = T1 = T3 = z βx α T0 = {},  Ω1, the grafting(cid:87) of T (i) over (x1, . . . , xk) and (α0, . . . , αk) is (cid:77) DT(X, Ω) := Denote by kTn. zyx n≥1 x α , , , , . . . (16) T = (T (0), . . . , T (k)) (cid:95)k+1; α0,...,αk x1,...,xk For typed valently decorated Schroder trees T (i) ∈ Tni, 0 ≤ i ≤ k, and x1, . . . , xk ∈ X, α0, . . . , αk ∈ 18 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ obtained by joining the k + 1 roots of T (i) to a new vertex valently decorated by (x1, . . . , xk) and decorating the new edges from left to right by α0, . . . , αk respectively. Example 4.4. Let T (0) = , T (1) = y , T (2) = z (cid:19) = , u, v ∈ X and α, β, γ ∈ Ω. Then y β u v x α z γ . x , y , z Any typed valently decorated Schroder tree T ∈ DT(X, Ω) can be uniquely expressed as such a grafting of lower depth typed valently decorated Schroder trees in Eq. (16), and we define the breadth of T as the arity of the internal vertex adjacent to the root, i.e. bre(T) := k + 1. For example, x y = T = x,y y zα x bre (,,); , (cid:95)3; 1,α,β u β = x,z (cid:18) (cid:18), (cid:19) x y = 3, and (cid:19) , y u , bre(T) = 3. x (cid:95)3; α,β,γ (cid:18) u,v (cid:95)3; 1,1,1 Definition 4.5. Let X be a set and let Ω be a semigroup. Define binary operations {≺ω,(cid:31)ω ω ∈ Ω} and · on DT(X, Ω) recursively on dep(T) + dep(U) by (a) (cid:31)ω T := T ≺ω := T, ≺ω T := T (cid:31)ω := 0 and · T := T · := 0 for ω ∈ Ω and T ∈ Tn with n ≥ 1. (b) Let (cid:95)m+1; α0,...,αm (cid:95)n+1; β0,...,βn x1,...,xm y1,...,yn T = U = (T (0), . . . , T (m)) ∈ Tm, (U(0), . . . , U(n)) ∈ Tn, define (17) T ≺ω U := (18) T (cid:31)ω U := (19) T · U := x1,...,xm−1,xm (cid:95)m+1; α0,...,αm−1,αmω (cid:95)n+1; ωβ0,β1,...,βn (cid:95)m+n+1; α0,...,αm−1,αmβ0,β1,...,βn y1,y2,...,yn x1,...,xm−1,xm,y1,...,yn U(1), . . . , U(n)). (T (0), . . . , T (m−1), T (m) (cid:31)αm U + T (m) ≺ω U + T (m) · U), (T (cid:31)ω U(0) + T ≺β0 U(0) + T · U(0), U(1), . . . , U(n)), (T (0), . . . , T (m−1), T (m) (cid:31)αm U(0) + T (m) ≺β0 U(0) + T (m) · U(0), Here in Eq. (19) when T (m) = = U(0), we employ the convention that (20) and (21) Let us expose some examples for better understanding. (cid:31)1 T := T ≺1 := T and ≺1 T := T (cid:31)1 := 0. ≺1 + (cid:31)1 + · := , FREE (TRI)DENDRIFORM FAMILY ALGEBRAS 19 and U = (,) ≺ω y with x, y ∈ X. For ω ∈ Ω, y + (cid:31)1 y + · y (cid:19) (by Eq. (17)) (by Item (a) in Definition 4.5 and Eq. (21)) · ,(cid:19) (by Eq. (18)) x (cid:31)ω + x (by Item (a) in Definition 4.5 and Eq. (21)) Example 4.6. Let T = x x x ≺ω (cid:31)ω y y = = = = = = = (cid:95)2; 1,1 (cid:95)2; 1,ω (cid:95)2; 1,ω x x x y ω x , x (cid:18) (cid:95)2; ω,1 (cid:31)ω (cid:18) (cid:95)2; ω,1 y y y y (cid:18), ≺ω (cid:18), (cid:19) (cid:95)2; 1,1 (,) ,(cid:19) ≺1 + (,) ·(cid:95)2; 1,1 x x y y (,) yx = ω , (cid:95)2; 1,1 (cid:95)3; 1,1,1 (cid:95)3; 1,1,1 x,y x x,y x y . x · y = = = = (, ≺1 + (cid:31)1 + · ,) (,,) (by Eq. (20)) (by Eq. (19)) z z x x x y y α = = y (cid:95)2; 1,1 Example 4.7. Let T = α T (cid:31)β U = α (cid:31)β ,(cid:19) (cid:18) (cid:95)2; β,1 ,(cid:19) ≺β (cid:18) (cid:95)2; α,1 (cid:18) (cid:95)2; α,β (cid:18) (,) ·(cid:95)2; α,1 (cid:95)2; 1,1 ,(cid:19) (cid:18), (cid:95)3; 1,α,1 T ≺β U = U · T = = = y y y z , x x x z z,x , U = (,) = z with x, y, z ∈ X and α ∈ Ω. For β ∈ Ω, (cid:18) (cid:95)2; β,1 y y y (cid:31)β + α x ≺1 + α x · ,(cid:19) x z α y α x β z z = = α y ,(cid:19) y xα z (cid:19) y = y , ≺β z + (cid:31)1 z + · z (cid:18), (cid:31)1 y + ≺α y + · y (cid:19) ,(cid:19) . (cid:95)2; α,β (cid:18) x z βx . (cid:95)3; 1,α,1 = z,x . 20 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ Proposition 4.8. Let X be a set and let Ω be a semigroup. Then (DT(X, Ω), {≺ω,(cid:31)ω ω ∈ Ω},·) is a tridendriform family algebra, generated by x . (cid:26) (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X Proof. We divide into two steps to prove the result. Step 1: we first prove (DT(X, Ω),{≺ω,(cid:31)ω ω ∈ Ω},·) is a tridendriform family algebra. Let (cid:95)m+1; α0,...,αm (cid:95)n+1; β0,...,βn (cid:95)(cid:96)+1; γ0,...,γ(cid:96) x1,...,xm y1,...,yn z1,...,z(cid:96) T = U = W = (T (0), . . . , T (m)), (U(0), . . . , U(n)) and (W(0), . . . , W((cid:96))) = = = = z1 . . . z(cid:96) (cid:33) z1 . . . z(cid:96) (cid:33) (cid:33) . . . yn z1 . . . z(cid:96) z1 . . . z(cid:96) . . . xm ≺γ y1 . . . yn y1 . . . yn x1,...,xm y1 . . . yn y1 . . . yn y1 . . . yn x1 . . . xm + · (cid:33)(cid:33) , U = ≺γ y1 x1,...,xm−1,xm and W = be in DT(X, Ω). Then we apply induction on dep(T) + dep(U) + dep(W) ≥ 3. For the initial step of dep(T) + dep(U) + dep(W) = 3, we have T = for some x1, . . . xm, y1, . . . , yn, z1, . . . , z(cid:96) ∈ X. Then (cid:32) x1 (T ≺β U) ≺γ W ≺β (cid:32)(cid:95)m+1; 1,...,1 (, . . . ,) ≺β (cid:32)(cid:95)m+1; 1,...,1,β (cid:32) , . . . ,, ≺β (cid:32) (cid:95)m+1; 1,...,1,β , . . . ,, (cid:32) (cid:95)m+1; 1,...,1,βγ , . . . ,, (cid:32) (cid:95)m+1; 1,...,1,βγ , . . . ,, ≺βγ (cid:32) y1 yn + (cid:31)1 (cid:32) y1 (cid:95)m+1; 1,...,1 (by Item (a) in Definition 4.5 and Eq. (21)) z(cid:96) + (cid:31)1 (by Eq. (17)) z(cid:96) ≺γ (by Item (a) in Definition 4.5 and Eq. (21)) (cid:31)β (, . . . ,) ≺βγ (by Eq. (17)) (cid:32) y1 (cid:32) y1 x1,...,xm−1,xm x1,...,xm−1,xm x1,...,xm−1,xm ≺γ yn (cid:31)β yn (cid:33)(cid:33) + · y1 . . . yn y1 . . . yn y1 . . . yn y1 . . . yn y1 . . . yn y1 . . . yn y1 . . . yn y1 . . . yn ≺γ y1 . . . yn ≺γ ≺γ y1 . . . yn ≺γ ≺γ (cid:31)β (cid:31)β (cid:31)β z1 . . . z(cid:96) . . . yn . . . yn z1 . . . z(cid:96) z1 . . . z(cid:96) . . . yn z1 . . . z(cid:96) y1 . . . z1 . . . z(cid:96) y1 . . . z1 . . . z(cid:96) (cid:33) z1 . . . z(cid:96) z1 . . . z(cid:96) z1 . . . z(cid:96) (cid:33) z1 . . . z(cid:96) z1 . . . (cid:33) z1 . . . z(cid:96) (cid:33) z1 . . . z(cid:96) . . . z1 . . . z(cid:96) z1 . . . . . . = z1 z(cid:96) = = · · + + + + + + · + · + · + + x1,...,xm (by Eq. (17)) (cid:32) y1 x1 . . . xm = ≺βγ FREE (TRI)DENDRIFORM FAMILY ALGEBRAS . . . yn ≺γ z1 . . . z(cid:96) y1 . . . yn + (cid:31)β z1 . . . z(cid:96) y1 . . . yn + z1 . . . z(cid:96) · (cid:33) 21 = = = = x1,...,xm−1,xm x1,...,xm−1,xm x1,...,xm−1,xm (by Eq. (17)) = T ≺βγ (U ≺γ W + U (cid:31)β W + U · W). The proofs of Eqs. (5)-(10) are similar. For the induction step of dep(T) + dep(U) + dep(W) ≥ 4, we have (cid:95)m+1; α0,...,αm−1,αmβ (T ≺β U) ≺γ W (cid:95)m+1; α0,...,αm−1,αmβγ (cid:16) (T (0), . . . , T (m−1), T (m) (cid:31)αm U + T (m) ≺β U + T (m) · U) ≺γ W (by Eq. (17)) (cid:17) T (0), . . . , T (m−1), (T (m) (cid:31)αm U + T (m) ≺β U + T (m) · U) ≺γ W + (T (m) (cid:31)αm U + T (m) ≺β U + T (m) · U) (cid:31)αmβ W + (T (m) (cid:31)αm U + T (m) ≺β U + T (m) · U) · W (cid:95)m+1; α0,...,αm−1,αmβγ (cid:16) T (0), . . . , T (m−1), (T (m) (cid:31)αm U) ≺γ W + (T (m) ≺β U) ≺γ W + (T (m) · U) ≺γ W (cid:95)m+1; α0,...,αm−1,αmβγ (cid:16) + T (m) (cid:31)αm (U (cid:31)β W) + (T (m) (cid:31)αm U + T (m) ≺β U + T (m) · U) · W (cid:17) T (0), . . . , T (m−1), T (m) (cid:31)αm (U ≺γ W) + T (m) ≺βγ (U ≺γ W + U (cid:31)β W + U · W) + (T (m) · U) ≺γ W + T (m) (cid:31)αm (U (cid:31)β W) + (T (m) (cid:31)αm U) · W + (T (m) ≺β U) · W + (T (m) · U) · W (cid:95)m+1; α0,...,αm−1,αmβγ (cid:16) (cid:17) T (0), . . . , T (m−1), T (m) ≺βγ (U ≺γ W + U (cid:31)β W + U · W) + T (m) (cid:31)αm (U ≺γ W+ (cid:95)m+1; α0,...,αm U (cid:31)β W + U · W) + T (m) · (U ≺γ W + U (cid:31)β W + U · W) = T ≺βγ (U ≺γ W + U (cid:31)β W + U · W). The proofs of Eqs. (5)-(10) are similar and are moved to Appendix. Step 2: We show that DT(X, Ω) is generated by dep(T) ≥ 1. For the initial step dep(T) = 1, we have T = (T (0), . . . , T (m)) ≺βγ (U ≺γ W + U (cid:31)β W + U · W) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X (by the induction hypothesis) . We employ induction on (by Eq. (17)) · . . . · x1,...,xm−1,xm x1,...,xm−1,xm x1,...,xm x1 . . . xm = = x1 = · x2 (cid:27) (cid:17) (cid:26) x xm x xi ∈(cid:26) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X (cid:27) where secondary induction on bre(T) ≥ 2. (cid:95)2; 1,α1 are three cases to consider. Case 1: T (0) = and T (1) (cid:44) . Then α0 = 1 and (, T (1)) T = x (T (0), T (1)) = (, ≺α1 T (1) + (cid:31)1 T (1) + · T (1)) (,) ≺α1 T (1) (by Eq. (17)) (cid:95)2; α0,α1 (cid:95)2; 1,α1 (cid:95)2; 1,1 x x x = = with 1 ≤ i ≤ m. For any induction step dep(T) ≥ 2, we apply a For the initial step bre(T) = 2, since dep(T) ≥ 2, we have T (cid:44) x for any x ∈ X and there (by Item (a) in Definition 4.5 and Eq. (21)) 22 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ ≺α1 T (1). (cid:27) By the induction hypothesis, T (1) is generated by (cid:26) Case 2: T (0) (cid:44) and T (1) = . This case is similar to Case 1. Case 3: T (0) (cid:44) and T (1) (cid:44) . Then x x . (cid:26) (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X and so T is generated by (T (0), T (1)) (T (0), ≺α1 T (1) + (cid:31)1 T (1) + · T (1)) (T (0),) ≺α1 T (1) (by Eq. (17)) (cid:95)2; 1,1 (T (0) (cid:31)α0 + T (0) ≺1 + T (0) · ,) (by Eq. (18)) (cid:19) ≺α1 T (1) (cid:19) ≺α1 T (1). (,) x x (by Item (a) in Definition 4.5 and Eq. (21)) (cid:19) ≺α1 T (1) (by Item (a) in Definition 4.5 and Eq. (21)) = x x (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X (cid:95)2; α0,α1 (cid:95)2; α0,α1 (cid:95)2; α0,1 (cid:18)(cid:95)2; α0,1 (cid:18) (cid:18) T (0) (cid:31)α0 T (0) (cid:31)α0 x x x T = = = = = = (cid:26) x (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X . x1,...,xm by x T = x1,x2,...,xm and so T is generated (T (0), . . . , T (m)), there exist four cases to consider. By the induction hypothesis, T (0) and T (1) are generated by Case 4: T (0) = and T (1) (cid:44) . Then α0 = 1 and (, T (1), . . . , T (m)) (cid:27) (cid:26) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X write T =(cid:87)m+1; α0,...,αm Now let us consider the case when dep(T) = k + 1 ≥ 2 and bre(T) = m + 1 ≥ 3. Then we can (cid:95)m+1; 1,α1,...,αm (cid:18)(cid:95)2; 1,α1 (, T (1)) (cid:18) (cid:26) (cid:19) ·(cid:18)(cid:95)2; 1,α2 (cid:19) ·(cid:18) ≺α1 T (1) (cid:27) (cid:19) · . . . ·(cid:18)(cid:95)2; 1,αm (cid:19) · . . . ·(cid:18) (by Item (a) in Definition 4.5 and Eq. (19)) x1 By the induction hypothesis, T (i) with 1 ≤ i ≤ m are generated by generated by Case 5: T (0) (cid:44) and T (1) = . This case is similar to Case 4. Case 6: T (0) (cid:44) and T (1) (cid:44) . Then xm ≺αm T (m) (cid:26) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X ≺α2 T (2) (by Case 1) (, T (m)) and so T is (, T (2)) (cid:19) . x2 x2 = = (cid:19) x (cid:27) x1 x xm, . (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X (cid:19) ·(cid:18)(cid:95)2; 1,α2 x2 (T (0), T (1), . . . , T (m)) (cid:95)m+1; α0,α1,...,αm (cid:18)(cid:95)2; α0,α1 x1,x2,...,xm (T (0), T (1)) x1 T = = (cid:19) · . . . ·(cid:18)(cid:95)2; 1,αm (cid:19) (, T (m)) xm, (, T (2)) (by Item (a) in Definition 4.5 and Eq. (19)) (cid:19) . xm ≺αm T (m) (cid:26) x (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X 23 (by Case 1 and Case 3) and so T is x2 x1 = (cid:33) FREE (TRI)DENDRIFORM FAMILY ALGEBRAS ·(cid:18) ≺α2 T (2) (cid:19) · . . . ·(cid:18) (cid:32)(cid:18) T (0) (cid:31)α0 (cid:26) (cid:19) ≺α1 T (1) (cid:27) (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X By the induction hypothesis, T (i) with 1 ≤ i ≤ m are generated by generated by (cid:95)m+1; 1,1,α2,...,αm Case 7: T (0) = and T (1) = . Then (cid:18)(cid:95)2; 1,1 (,) ·(cid:18) (cid:19) · . . . ·(cid:18)(cid:95)2; 1,αm (cid:19) · . . . ·(cid:18) (cid:19) ·(cid:18)(cid:95)2; 1,α2 (,, T (2), . . . , T (m)) (, T (2)) x1,x2,x3,...,xm T = = xm x2 x1 x . (cid:19) (, T (m)) (cid:19) (cid:26) . x2 = x1 (by Case 1) ≺α2 T (2) (by Item (a) in Definition 4.5 and Eq. (19)) xm ≺αm T (m) By the induction hypothesis, T (i) with 1 ≤ i ≤ m are generated by This completes the proof. The concept of free tridendriform family algebra is given as usual. Definition 4.9. Let X be a set and let Ω be a semigroup. A free tridendriform family algebra on X is a tridendriform family algebra (T,{≺ω,(cid:31)ω ω ∈ Ω},· ) together with the map j : X → T that satisfies the following universal property: for any tridendriform family algebra (T(cid:48),{≺(cid:48) ω ω ∈ Ω},·(cid:48) ) and map f : X → T(cid:48), there is a unique tridendriform family algebra morphism ¯f : T → T(cid:48) such that f = ¯f ◦ j. The free tridendriform family algebra is unique up to isomorphism. (cid:12)(cid:12)(cid:12)(cid:12) x ∈ X and so is T. (cid:3) ω,(cid:31)(cid:48) (cid:27) x Let j : X → DT(X, Ω) be the map defined by j(x) = x for x ∈ X. Now we arrive at our main result in this section. Theorem 4.10. Let X be a set and let Ω be a semigroup. The tridendriform family algebra (DT(X, Ω),{≺ω,(cid:31)ω ω ∈ Ω},·), together with the map j, is the free tridendriform family algebra on X. Proof. By Proposition 4.8, we are left to prove (DT(X, Ω),{≺ω,(cid:31)ω ω ∈ Ω},·) satisfies the univer- sal property. For this, let (T(cid:48),{≺(cid:48) ω ω ∈ Ω},·(cid:48) ) be a tridendriform family algebra. First of all, if there exists a tridendriform family algebra morphism ¯f : DT(X, Ω) → T(cid:48) extending f : X → T(cid:48) (cid:19) ) = f (x) for any x ∈ X, then such an ¯f is unique, due to the fact that in the sense that ¯f ( generate the tridendriform family algebra DT(X, Ω). Now we the collection of trees x∈X define a linear map ¯f : DT(X, Ω) → T(cid:48) by induction on dep(T) ≥ 1 and prove that it is a tridendriform family algebra morphism. We use the unique decomposition T (cid:55)→ ¯f (T) ω,(cid:31)(cid:48) (cid:18) x x , (cid:95)m+1; α0,...,αm (cid:95)2; α0,α1 x1,...,xm x1 (T (0), . . . , T (m)) (T (0), T (1)) ·(cid:95)2; 1,α2 (, T (2)) · . . . ·(cid:95)2; 1,αm x2 (, T (m)), xm T = = (22) 24 together with: (cid:95)2; α0,α1 x YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ (V, W) = x if V = W = , = V (cid:31)α0 = x = V (cid:31)α0 x if V (cid:44) and W = , ≺α1 W if V = and W (cid:44) , x ≺α1 W if V (cid:44) and W (cid:44) , which uniquely defines ¯f (T) in terms of ¯f (T (0)), . . . , ¯f (T (m)) and the operations ({≺(cid:48) Ω},·(cid:48)). We are going to prove that ¯f is a morphism of tridendriform family algebras: ω ω ∈ in which we only prove the first equation, as the proofs of the two other are similar. Starting from ¯f (T ≺ω U) = ¯f (T) ≺(cid:48) (cid:95)m+1; α0,...,αm ω T = x1,...,xm ¯f (U), ¯f (T (cid:31)ω U) = ¯f (T) (cid:31)(cid:48) ω (T (0), . . . , T (m)), U = ω,(cid:31)(cid:48) ¯f (U) and ¯f (T · U) = ¯f (T) ·(cid:48) ¯f (U), (cid:95)n+1; β0,...,βn (U(0), . . . , U(n)) y1,...,yn x1 (T (0), T (1)) ·(cid:95)2; 1,α2 (cid:19) (T (0), . . . , T (m−1), T (m) (cid:31)αm U + T (m) ≺ω U + T (m) · U) x1,...,xm = ¯f we compute: (cid:18)(cid:95)m+1; α0,...,αm−1,αmω ¯f (T ≺ω U) (cid:18)(cid:95)2; α0,α1 = ¯f (cid:18)(cid:95)2; α0,α1 (cid:18)(cid:95)2; α0,α1 = ¯f x1 x1 (cid:19) (, T (m) (cid:31)αm U + T (m) ≺ω U + T (m) · U) xm (, T (2)) · . . . ·(cid:95)2; 1,αmω (cid:18) (cid:18)(cid:16) x2 (from (22)) . . . ·(cid:48) ¯f (from the definition of ¯f ) . . . ·(cid:48) ¯f (T (0), T (1)) (T (0), T (1)) (23) = ¯f (from Axiom (4)). (cid:19) xm ≺αmω (T (m) (cid:31)αm U + T (m) ≺ω U + T (m) · U) (cid:19) ·(cid:48) (cid:19) ·(cid:48) Now we proceed by induction on dep(T)+dep(U), together with a secondary induction on bre(T). If dep(T) + dep(U) = 2 (main initial step), then T (0) = ··· = T (m) = , and Equation (23) boils (cid:16) (cid:1) ·(cid:48) down to (cid:17) ·(cid:48) (cid:16) ¯f (T ≺ω U) = ¯f (cid:18) (cid:17) ·(cid:48) (cid:16) = ¯f x1 = ¯f (T) ≺(cid:48) (cid:18)(cid:95)2; α0,α1ω (cid:18) T (0) (cid:31)α0 If dep(T) + dep(U) ≥ 3 and bre(T) = 2 (secondary initial step), then we have ¯f (T ≺ω U) = ¯f = ¯f xm ≺αm T (m)(cid:17) ≺ω U (cid:19) (cid:18) (cid:16) xm−1 (cid:17) ·(cid:48) ¯f (cid:16) xm−1 (cid:17) ·(cid:48)(cid:18) (cid:16) (cid:16) xm−1 (cid:17) ·(cid:48) ¯f (cid:16) . . . ·(cid:48) ¯f . . . ·(cid:48) ¯f . . . ·(cid:48) ¯f (cid:0)T 0), T (1) (cid:31)α1 U + T (1) ≺ω U + T (1) · U(cid:1)(cid:19) (cid:0)T (1) (cid:31)α1 U + T (1) ≺ω U + T (1) · U(cid:1)(cid:19) ≺α1ω ¯f (U) (from an iteration of Axiom (9)) (from the definition of ¯f ) (cid:17) ≺(cid:48) (cid:17)(cid:19) ≺(cid:48) xm ≺ω U ¯f xm xm ¯f (U). ω ¯f (U) ω = ¯f x1 x1 x1 x1 (cid:19) (cid:19) ω FREE (TRI)DENDRIFORM FAMILY ALGEBRAS ¯f (cid:16) (cid:16) (cid:18)(cid:18) (cid:18) ¯f (cid:16) (cid:16) ¯f ¯f x1 x1 x1 x1 α1ω (cid:17) ≺(cid:48) (cid:17) ≺(cid:48) (cid:17) ≺(cid:48) (cid:17) ≺(cid:48) α1ω α1 25 ¯f(cid:0)T (1) (cid:31)α1 U + T (1) ≺ω U + T (1) · U(cid:1) (by definition of ¯f ) (cid:16) ¯f (T (1)) (cid:31)(cid:48) (cid:17) ¯f (U) + ¯f (T (1)) ·(cid:48) ¯f (U) (cid:19) ≺(cid:48) (cid:19)(cid:19) ≺(cid:48) (by the main induction hypothesis) ¯f (T (1)) ¯f (U) + ¯f (T (1)) ≺(cid:48) ¯f (U) ¯f (T (1)) ¯f (U) (cid:19) α1 ω ω α0 ¯f (U) (again by definition of ¯f ). α1 ω α0 α0 = ¯f (T (0)) (cid:31)(cid:48) = ¯f (T (0)) (cid:31)(cid:48) (cid:18) = ¯f (T (0)) (cid:31)(cid:48) ¯f (T (0)) (cid:31)(cid:48) = ¯f (T) ≺(cid:48) = α0 ω Finally, if dep(T) + dep(U) ≥ 3 and bre(T) ≥ 3, we can push further Compilation (23) by using the secondary initial step for (cid:18) ¯f (T ≺ω U) = ¯f ¯f (cid:16)(cid:95)2; α0,α1 xm ≺αm T (m), which is of breadth two: (cid:16)(cid:95)2; α0,α1 (cid:18)(cid:16)(cid:95)2; α0,α1 (cid:16) (cid:16) ¯f . . . ·(cid:48) ¯f (by iterating Axiom (9)) xm ≺αm T (m)(cid:17) ≺(cid:48) xm ≺αm T (m)(cid:17)(cid:19) ≺(cid:48) xm ≺αm T (m)(cid:17)(cid:19) ≺(cid:48) . . . ·(cid:48)(cid:18) (cid:17) ·(cid:48) (cid:17) ·(cid:48) (cid:17) · . . . ·(cid:16) (T (0), T (1)) (T (0), T (1)) (T (0), T (1)) ¯f (U) ¯f (U) ¯f (U) = ¯f (cid:19) = x1 x1 ω ω ω x1 = ¯f (T) ≺(cid:48) ω ¯f (U). (by definition of ¯f ) The verification of the two identities ¯f (T (cid:31)ω U) = ¯f (T) (cid:31)(cid:48) similar and left to the reader. ω ¯f (U) and ¯f (T · U) = ¯f (T) ·(cid:48) ¯f (U) is (cid:3) Acknowledgments: This work was supported by the National Natural Science Foundation of China (Grant No. 11771191 and 11861051). References [1] G. Baxter, An analytic problem whose solution follows from a simple algebraic identity, Pacific J. Math. 10 (1960), 731-742. 3 [2] Y. Bruned, M. Hairer and L. Zambotti, Algebraic renormalisation of regularity structures, Invent. math. https://doi.org/10.1007/s0022-018-0841-x. 2, 8 [3] F. Bergeron, G. Labelle and P. Leroux, Combinatorial species and tree-like structures, Encyclopedia of Math- ematics and its Application, vol. 67, Cambridge University Press, Cambridge, 1998, Translated from the 1994 French original by Margaret Readdy, With a foreword by Gian-Carlo Rota. 2 [4] A. Connes and D. Kreimer, Hopf algebras, renormalization and non-commutative geometry, Commun. Math. Phys. 199 (1998), 203-242. 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, Commun. Math. Phys. 210 (2000), 249-273. 2 [6] A. Connes and D. Kreimer, Renormalization in quantumn field theory and the Riemann-Hilbert problem. II. The β-function, diffemorphisms and the renormalization group, Comm. Math. Phys. 216 (2001), 215-241. 2 [7] K. Ebrahimi-Fard, J. Gracia-Bondia and F. Patras, A Lie theoretic approach to renormalization, Comm. Math. Phys. 276 (2007), 519-549. 1, 3 708-727. 2 [8] K. Ebrahimi-Fard, D. Manchon and F. Patras, New identies in dendriform algebras, J. Algebra 320 (2008), [9] A. Frabetti, Leibniz homology of dialgebras of matrices, J. Pure. Appl. Algebra 129 (1998), 123-141. 1 26 YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ [10] L. Foissy, Les alg`ebres de Hopf des arbres enracin´es d´ecor´es II, Bull. Sci. Math. 126 (2002), 249-288. 2 [11] L. Foissy, Algebraic structures on typed decorated rooted trees, arXiv: 1811.07572vl [math.RA] 19 Nov 2018. [12] X. Gao, L. Guo and M. Rosenkranz, Free intergro-differential algebras and Grobner-Shirshov basis, J. Algebra [13] X. Gao, L. Guo and Y. Zhang, Matching Rota-Baxter algebras, matching dendriform algebras and matching 442 (2015), 354-396. 2 pre-Lie algebras, arxiv:1909.10577 (2019). 2 8 2 [14] X. Gao, L. Guo and S. H. Zheng, Construction of free commutative integro-differential algebras by the method of Grobner-Shirshov basis, J. Algebra Appl. 5 (2014), 1350160. 2 [15] L. Guo and W. Keigher, On free Baxter algebras: completions and the internal construction, Adv. Math. 151 (2000), 101-127. 2 [16] L. Guo and W. Keigher, On differential Rota-Baxter algebras, J. Pure Appl. Algebra 212 (2008), 522-540. 2 [17] L. Guo, Operated monoids, Motzkin paths and rooted trees, J. Algebraic Combin. 29 (2009), 35-62. 3 [18] L. Guo, An introduction to Rota-Baxter algebra, International Press, 2012. 1 [19] D. Kreimer and Erik Panzer, Hopf-algebraic renormalization of Kreimer's toy model, Master thesis, Handbook, https://arxiv.org/abs/1202.3552. 1, 3 [20] J. -L. Loday, Une version non commutative des alg`ebre de Lie: les alg`ebres de Leibniz, Ens. Math. 39 (1993), [21] J. -L. Loday and M. R. Ronco, Hopf algebra of the planar binary trees, Adv. Math. 39 (1998), 293-309. 1, 2, 7, 269-293. 1, 2, 4, 7 8, 16, 17 [22] J. -L. Loday, Dialgebra, in Dialgebras and related operad, Lecture Notes in Math. 1763 (2002), 7-66. 7 [23] J. -L. Loday, Arithmetree, J. Algebra 258 (2002), 275-309. 1 [24] J. -L. Loday, Scindement d'associativit´e et alg`ebres de Hopf, the Proceedings of the Conference in hounor of Jean Leray, Nantes, (2002), S´eminaire et Congr`es(SMF) 9 (2004), 155-172. 1 [25] J. -L. Loday and M. O. Ronco, Trialgebras and families of polytopes, in "Homotopy theoty: relations with algebraic geometry, group cohomology, and algebraic K-theory", Contemp. Math. 346 (2004), 369-398. 1, 2, 4, 16, 17 [26] M. R. Ronco, Eulerian idempotents and Milnor-Moore theorem for certain non-commutative Hopf algebras, J. Algebra 254 (2002), 152-172. 1, 7 [27] Richard J. Mathar, Topologically distinct sets of non-intersecting circles in the plane, arXiv:1603.00077,2016. [28] Y. Y. Zhang and X. Gao, Free Rota-Baxter family algebras and (tri)dendriform family algebras, accepted by Pacific J. Math.. 1, 2, 4, 5 5. Appendix = = y1,y2,...,yn The remaining proofs of Step 1 in Proposition 4.8. (cid:95)n+1; αβ0,β1,...,βn T (cid:31)α U(0) + T ≺β0 U(0) + T · U(0)), U(1), . . . , U(n)(cid:17) ≺γ W (cid:16) (T (cid:31)α U) ≺γ W (cid:95)n+1; αβ0,β1,...,βn−1,βnγ (cid:16) T (cid:31)α U(0) + T ≺β0 U(0) + T · U(0), U(1), . . . , U(n−1), (cid:16)(cid:95)n+1; β0,...,βn−1,βnγ U(n) (cid:31)βn W + U(n) ≺γ W + U(n) · W = T (cid:31)α = T (cid:31)α (U ≺γ W). (U(0), . . . , U(n−1), U(n) (cid:31)βn W + U(n) ≺γ W + U(n) · W y1,y2,...,yn−1,yn y1,...,yn−1,yn (cid:17) (cid:17) (cid:16)(cid:95)(cid:96)+1; βγ0,γ1,...,γ(cid:96) T (cid:31)α (U (cid:31)β W) = T (cid:31)α z1,z2,...,z(cid:96) (cid:17) (U (cid:31)β W(0) + U ≺γ0 W(0) + U · W(0), W(1), . . . , W(l)) T (cid:31)α (U (cid:31)β W(0) + U ≺γ0 W(0) + U · W(0)) + T ≺βγ0 (U (cid:31)β W(0) + U ≺γ0 W(0) 27 z1,z2,...,z(cid:96) (cid:16) (cid:16) (cid:16) FREE (TRI)DENDRIFORM FAMILY ALGEBRAS (cid:95)(cid:96)+1; αβγ0,γ1,...,γ(cid:96) + U · W(0)) + T · (U (cid:31)β W(0) + U ≺γ0 W(0) + U · W(0)), W(1), . . . , W((cid:96))(cid:17) (cid:95)(cid:96)+1; αβγ0,γ1,...,γ(cid:96) + T · (U (cid:31)β W(0)) + T · (U ≺γ0 W(0)) + T · (U · W(0)), W(1), . . . , W((cid:96))(cid:17) (cid:95)(cid:96)+1; αβγ0,γ1,...,γ(cid:96) + (T (cid:31)α U + T ≺β U + T · U) · W(0), W(1), . . . , W((cid:96))(cid:17) z1,z2,...,z(cid:96) z1,z2,...,z(cid:96) (cid:95)(cid:96)+1; γ0,...,γ(cid:96) z1,...,z(cid:96) T (cid:31)α (U (cid:31)β W(0)) + T (cid:31)α (U ≺γ0 W(0)) + T (cid:31)α (U · W(0)) + (T ≺β U) ≺γ0 W(0) (T (cid:31)α U + T ≺β U + T · U) (cid:31)αβ W(0) + (T (cid:31)α U + T ≺β U + T · U) ≺γ0 W(0) z1,...,z(cid:96) (cid:17) (cid:16) (cid:16) y1,y2,...,yn y1,y2,...,yn,z1,z2,...z(cid:96) = T (cid:31)α y1,...,yn−1,yn,z1,...,z(cid:96) (W(0), . . . , Wl) (W(0), . . . , W(l)) U(0), . . . , U(n−1), U(n) (cid:31)βn W(0) + U(n) ≺γ0 W(0) + U(n) · W(0), T (cid:31)α U(0) + T ≺β0 U(0) + T · U(0), U(1), . . . , U(n−1), (cid:16) T (cid:31)α U(0) + T ≺β0 U(0) + T · U(0), U(1), . . . , U(n)(cid:17) ·(cid:95)(cid:96)+1; γ0,...,γ(cid:96) = (T (cid:31)α U + T ≺β U + T · U) (cid:31)αβ ( = (T (cid:31)α U + T ≺β U + T · U) (cid:31)αβ W. (cid:95)n+1; αβ0,β1,...,βn (T (cid:31)α U) · W (cid:95)n+1+(cid:96); αβ0,β1,...,βn−1,βnγ0,γ1,...,γ(cid:96) (cid:16) U(n) (cid:31)βn W(0) + U(n) ≺γ0 W(0) + U(n) · W(0), W(1), . . . , W(l)(cid:17) (cid:95)n+1+(cid:96); β0,...,βn−1,βnγ0,z1,...,z(cid:96) W(1), . . . , W(l)(cid:17) = T (cid:31)α (U · W). (cid:95)m+1; α0,...,αm−1,αmβ (T ≺β U) · W ·(cid:95)(cid:96)+1; γ0,...,γ(cid:96) (cid:95)m+1+(cid:96); α0,...,αm−1,αmβγ0,γ1,...,γ(cid:96) (cid:16) T (0), . . . , T (m−1), (T (m) (cid:31)αm U + T (m) ≺β U + T (m) · U) (cid:31)αmβ W(0) + T (m) · U) · W(0), W(1), . . . , W((cid:96))(cid:17) + (T (m) (cid:31)αm U + T (m) ≺β U + T (m) · U) ≺γ0 W(0) + (T (m) (cid:31)αm U + T (m) ≺β U (cid:95)m+1+(cid:96); α0,...,αm−1,αmβγ0,γ1,...,γ(cid:96) (cid:16) T (0), . . . , T (m−1), T (m) (cid:31)αm (U (cid:31)β W(0) + U ≺γ0 W(0) + U · W(0)) W(1), . . . , W((cid:96))(cid:17) + T (m) ≺βγ0 (U (cid:31)β W(0) + U ≺γ0 W(0) + U · W(0)) + T (m) · (U (cid:31)β W(0) + U ≺γ0 W(0) + U · W(0)), (cid:95)m+1; α0,...,αm = T ·(cid:95)(cid:96)+1; βγ0,γ1,...,γ(cid:96) (T (0), . . . , T (m)) ·(cid:95)(cid:96)+1; βγ0,γ1,...,γ(cid:96) (cid:16) U (cid:31)β W(0) + U ≺γ0 W(0) + U · W(0), W(1), . . . , W(l)(cid:17) U (cid:31)β W(0) + U ≺γ0 W(0) + U · W(0), W(1), . . . , W(l)(cid:17) T (0), . . . , T (m−1), T (m) (cid:31)αm U + T (m) ≺β U + T (m) · U (W(0), . . . , W(l)) x1,...,xm−1,xm,z1,...,z(cid:96) x1,...,xm−1,xm,z1,...,z(cid:96) = = = = x1,...,xm−1,xm z1,...,z(cid:96) x1,...,xm z1,z2,...,z(cid:96) (cid:16) = = = = = z1,z2,...,z(cid:96) = T · (U (cid:31)β W). (T · U) ≺γ W x1,...,xm y1,...,yn−1,yn YUANYUAN ZHANG, XING GAO, AND DOMINIQUE MANCHON∗ T (0), . . . , T (m−1), T (m) (cid:31)αm U(0) + T (m) ≺β0 U(0) + T (m) · U(0), U(1), . . . , U(n)(cid:17) (cid:16) (cid:17) (cid:17) U(0), . . . , U(n−1), U(n) (cid:31)βn W + U(n) ≺γ W + U(n) · W T (0), . . . , T (m−1), T (m) (cid:31)αm U(0) + T (m) ≺β0 U(0) + T (m) · U(0), 28 = = = x1,...,xm−1,xm,y1,...,yn−1,yn (T (0), . . . , T (m)) ·(cid:95)n+1; β0,...,βn−1,βnγ (cid:95)m+1+n; α0,...,αm−1,αmβ0,β1,...,βn (cid:16) (cid:95)m+1+n; α0,...,αm−1,αmβ0,β1,...,βn−1,βnγ (cid:95)m+1; α0,...,αm (cid:16) U(1), . . . , U(n−1), U(n) (cid:31)βn W + U(n) ≺γ W + U(n) · W = T · (U ≺γ W). (cid:95)m+1; α0,...,αm−1,αmβ0,β1,...,βn ·(cid:95)(cid:96)+1; γ0,...,γ(cid:96) (cid:95)m+1+n+(cid:96); α0,...,αm−1,αmβ0,β1,...,βn−1,βnγ0,γ1,...,γ(cid:96) U(1), . . . , U(n−1), U(n) (cid:31)βn W(0) + U(n) ≺γ0 W(0) + U(n) · W(0), W(1), . . . , W(l)(cid:17) (cid:95)m+1; α0,...,αm + U(n) ≺γ0 W(0) + U(n) · W(0), W(1), . . . , W((cid:96))(cid:17) (T (0), . . . , T (m)) ·(cid:95)n+1; β0,...,βn−1,βnγ0,γ1,...,γ(cid:96) x1,...,xm−1,xm,y1,...,yn−1,yn,z1,...,z(cid:96) (W(0), . . . , W(l)) (T · U) · W x1,...,xm−1,xm,y1,y2,...,yn y1,...,yn−1,yn,z1,...,z(cid:96) = = = (cid:16) T (0), . . . , T (m−1), T (m) (cid:31)αm U(0) + T (m) ≺β0 U(0) + T (m) · U(0), U(0), . . . , U(n)(cid:17) T (0), . . . , T (m−1), T (m) (cid:31)αm U(0) + T (m) ≺β0 U(0) + T (m) · U(0), (cid:16) U(0), . . . , U(n−1), U(n) (cid:31)βn W(0) (cid:16) x1,...,xm,y1,y2,...,yn z1,...,z(cid:96) x1,...,xm = T · (U · W). School of Mathematics and Statistics, Lanzhou University, Lanzhou, 730000, P. R. China E-mail address: [email protected] School of Mathematics and Statistics, Key Laboratory of Applied Mathematics and Complex Systems, Lanzhou University, Lanzhou, 730000, P.R. China E-mail address: [email protected] Laboratoire de Math´ematiques Blaise Pascal, CNRS -- Unibersit´e Clermont-Auvergne, 3 place Vasar´ely, CS 60026, F63178 Aubi`ere, France E-mail address: [email protected]
1301.6819	2	1301	2013-04-16T03:29:40	Generalized Yetter-Drinfel'd module categories for regular multiplier Hopf algebras	[ "math.RA" ]	For a regular multiplier Hopf algebra $A$, the Yetter-Drinfel'd module category ${}_{A}\mathcal{YD}^{A}$ is equivalent to the centre $Z({}_{A}\mathcal{M})$ of the unital left $A$-module category ${}_{A}\mathcal{M}$. Then we introduce the generalized $(\alpha, \beta)$-Yetter-Drinfel'd module categories ${}_{A}\mathcal{GYD}^{A}(\alpha, \beta)$, which are treated as components of a braided $T$-category. Especially when $A$ is a coFrobenius Hopf algebra, ${}_{A}\mathcal{YD}^{A}(\alpha, \beta)$ is isomorphic to the unital $\hat{A} \bowtie A(\alpha, \beta)$-module category ${}_{\hat{A} \bowtie A(\alpha, \beta)}\mathcal{M}$. Finally for a Yetter-Drinfel'd $A$-module algebra $H$, we introduce Yetter-Drinfel'd $(H, A)$-module category, which is a monoidal.	math.RA	math	Generalized Yetter-Drinfel'd module categories for regular multiplier Hopf algebras Tao Yang ∗, Xuan Zhou † Abstract For a commutative regular multiplier Hopf algebra A, the Yetter-Drinfel'd module category AYDA is equivalent to the centre Z(AM) of the unital left A-module category AM. Then we introduce the generalized (α, β)-Yetter-Drinfel'd module categories AGYDA(α, β), which are treated as components of a braided T -category. Especially when A is a coFrobenius Hopf algebra, AYDA(α, β) is isomorphic to the unital bA ⊲⊳ A(α, β)-module category bA⊲⊳A(α,β) M. Finally for a Yetter-Drinfel'd A-module algebra H, we introduce Yetter-Drinfel'd (H, A)-module category, which is a monoidal. Key words Multiplier Hopf Algebra, Yetter-Drinfel'd Module. Mathematics Subject Classification: 16W30 · 17B37 1 Introduction A Yetter-Drinfel'd module (or crossed bimodule, Yang-Baxter module) over Hopf al- gebra is a module and a comodule satisfying a certain compatibility condition. The main feature of this definition is that Yetter-Drinfel'd modules form a pre-braided monoidal category. Under favourable conditions (e.g. the antipode of Hopf algebra is bijective), the category is even braided (or quasisymmetric). Via (pre-)braiding structure, the notion of Yetter-Drinfel'd module plays a part in the relations between quantum group and knot theory. As Hopf algebra is finite dimensional, the Yetter-Drinfel'd module category has some special properties, e.g., it is isomorphic to the category of left modules over the Drinfel'd double of this Hopf algebra. This property no longer holds in the infinite dimensional case. However by the multiplier Hopf algebra theory, the isomorphism exists when the ∗Corresponding author. College of Science, Nanjing Agricultural University, Nanjing 210095, Jiangsu, CHINA. E-mail: [email protected] †Department of Mathematics, Jiangsu Institute of Education, Nanjing 210013, Jiangsu, CHINA. E-mail: [email protected] 1 Hopf algebra is coFrobenius (see [12]). Now, multiplier Hopf algebra becomes a valuable tool to deal with infinite dimensional Hopf algebra cases. In [10], new kinds of Yetter-Drinfeld modules over a regular multiplier Hopf algebra were introduced via extended module structure. We also use them to construct new braided crossed categories, which are related to homotopy invariants. In this paper, we continue to consider the (left-right) Yetter-Drinfel'd module category AYDA defined in [10], and show that it is isomorphic to the center of unital left A- module category. Then we consider two kinds of generalizations of Yetter-Drinfel'd module categories and get some properties. The paper is organized in the following way. In section 2, we recall some notions which we will use in the following, such as multiplier Hopf algebras, extended modules, comodules and centre of a strict tensor category. In section 3, we mianly show the relationship between the Yetter-drinfel'd module category AYDA and the centre Z(AM) of the unital A-module category AM. In section 4, we introduced the generalized (α, β)-Yetter-Drinfel'd module categories AGYDA(α, β), which are treated as components of a braided T -category. Especially when A is a coFrobenius Hopf algebra, AYDA(α, β) is isomorphic to the unital bA ⊲⊳ A(α, β)- module category bA⊲⊳A(α,β) M. In section 5, let H be a Yetter-Drinfel'd A-module algebra, we introduce Yetter- Drinfel'd (H, A)-module category, which is a monoidal. 2 Preliminaries We begin this section with a short introduction to multiplier Hopf algebras. Throughout this paper, all spaces we considered are over a fixed field K (such as field C of complex numbers). Algebras may or may not have units, but always should be non- degenerate, i.e., the multiplication maps (viewed as bilinear forms) are non-degenerate. For an algebra A, the multiplier algebra M (A) of A is defined as the largest algebra with unit in which A is a dense ideal (see the appendix in [5]). Now, we recall the definitions of a multiplier Hopf algebra (see [5] for details). A comultiplication on an algebra A is a homomorphism ∆ : A −→ M (A ⊗ A) such that ∆(a)(1 ⊗ b) and (a ⊗ 1)∆(b) belong to A ⊗ A for all a, b ∈ A. We require ∆ to be coassociative in the sense that (a ⊗ 1 ⊗ 1)(∆ ⊗ ι)(∆(b)(1 ⊗ c)) = (ι ⊗ ∆)((a ⊗ 1)∆(b))(1 ⊗ 1 ⊗ c) for all a, b, c ∈ A (where ι denotes the identity map). A pair (A, ∆) of an algebra A with non-degenerate product and a comultiplication ∆ on A is called a multiplier Hopf algebra, if the linear maps T1, T2 : A ⊗ A −→ A ⊗ A 2 defined by are bijective. T1(a ⊗ b) = ∆(a)(1 ⊗ b), T2(a ⊗ b) = (a ⊗ 1)∆(b) (2.1) A multiplier Hopf algebra (A, ∆) is called regular if (A, ∆cop) is also a multiplier Hopf algebra, where ∆cop denotes the co-opposite comultiplication defined as ∆cop = τ ◦ ∆ with τ the usual flip map from A ⊗ A to itself (and extended to M (A ⊗ A)). In this case, ∆(a)(b ⊗ 1) and (1 ⊗ a)∆(b) ∈ A ⊗ A for all a, b ∈ A. Multiplier Hopf algebra (A, ∆) is regular if and only if the antipode S is bijective from A to A (see [6], Proposition 2.9). In this situation, the comultiplication is also determined by the bijective maps T3, T4 : A ⊗ A −→ A ⊗ A defined as follows T3(a ⊗ b) = ∆(a)(b ⊗ 1), T4(a ⊗ b) = (1 ⊗ a)∆(b) (2.2) for all a, b ∈ A. We will use the adapted Sweedler notation for regular multiplier Hopf algebras (see [7]). We will e.g., write P a(1) ⊗ a(2)b for ∆(a)(1 ⊗ b) and P ab(1) ⊗ b(2) for (a ⊗ 1)∆(b), sometimes we omit the P. Define two linear operators T and T ′ acting on A ⊗ A introduced by the formulae T (a ⊗ b) = b(2) ⊗ aS(b(1))b(3), T ′(a ⊗ b) = b(1) ⊗ S(b(2))ab(3) for any a, b ∈ A. This two operators above are well-defined, since T (a ⊗ b) = T4(S ⊗ ι)T3(ι ⊗ S−1)τ (a ⊗ b), T ′(a ⊗ b) = (ι ⊗ S)T4τ (ι ⊗ S−1)T4(a ⊗ b). They are obviously invertible, and the inverses can be written as follows T −1(a ⊗ b) = bS−1(a(3))a(1) ⊗ a(2), T ′−1(a ⊗ b) = a(3)bS−1(a(2)) ⊗ a(1). For any a, b ∈ A, T ◦ T2 = T4. If A is commutative, then T ′ = τ , and if A is cocommutative, then T = τ . Proposition 2.1 Operators T and T ′ satisfy the braided equation (T ⊗ ι)(ι ⊗ T )(T ⊗ ι) = (ι ⊗ T )(T ⊗ ι)(ι ⊗ T ), (T ′ ⊗ ι)(ι ⊗ T ′)(T ′ ⊗ ι) = (ι ⊗ T ′)(T ′ ⊗ ι)(ι ⊗ T ′). 3 2.1 Extended modules Let A be an (associative) algebra. Suppose X is a left A-module with the module structure map · : A ⊗ X −→ X. We will always assume that the module is non-degenerate, this means that x = 0 if x ∈ X and a · x = 0 for all a ∈ A. If the module is unital (i.e., A · X = X), then we can get an extension of the module structure to M (A), this means that we can define f · x, where f ∈ M (A) and x ∈ X. In fact, since x ∈ X = A · X, then x = Pi ai · xi and f · x = Pi(f ai) · xi. In this setting, we can easily get 1 · x = x, where 1 is the unit of multiplier algebra M (A). Denote by Y the space of linear maps ρ : A → X satisfying ρ(aa′) = a · ρ(a′) for all a, a′ ∈ A. Then Y becomes a left A-module if we define a · ρ for a ∈ A and ρ ∈ Y by (a · ρ)(a′) = ρ(a′a). For any x ∈ X, we have an element ρx ∈ Y defined by ρx(a) = a · x when a ∈ A. The injective map x 7→ ρx is a map of left modules, from this X becomes a submodule of Y . We call Y the extended module of X. From the definition, we have A · Y ⊆ X, and A · Y = X if A · X = X. Since A2 = A, Y is still non-degenerate. If A has a unit, then Y = X, in the other case, mostly Y is strictly bigger than X. Because the module Y is the completion of the original left A-module X for the strict topology (see [7]), sometimes it is also called the complete module (or completion) of X, and denoted Y as Ml(X). We can also do the same for a right A-module X and denote the extended module of the right A-module X as Mr(X). Let X be a non-degenerate A-bimodule. Denote by Z the space of pair (λ, ρ) of linear maps from A to X satisfying a · λ(a′) = ρ(a) · a′ for all a, a′ ∈ A. From the non-degeneracy, it follows that ρ(aa′) = a · ρ(a′) and λ(aa′) = λ(a) · a′ for all a, a′ ∈ A. Also ρ is completely determined by λ and vice versa. We can consider Z as the intersection of two extensions of X (as a left and a right modules). Z becomes an A-bimodule, if we define a · z and z · a for a ∈ A and z = (λ, ρ) ∈ Z by a · z = (a · λ(·), ρ(·a)) and z · a = (λ(a·), ρ(·) · a). If we define (λx, ρx) for x ∈ X by λx(a) = x · a and ρx(a) = a · x, we get X as a submodule of Z. We call Z the extended module of A-bimodule X, and denote it as M0(X). Let V be a vector space and X = V ⊗ A. We consider the left and right module actions of A on X respectively given by a · (v ⊗ a′) = v ⊗ aa′ and (v ⊗ a′) · a = v ⊗ a′a for a, a′ ∈ A and v ∈ V . We denote the extended module of left A-module X as Ml(V ⊗A), and extended module of the right one as Mr(V ⊗ A). This two actions are compatible, and make X an A-bimodule, we denote its extend module as M0(V ⊗ A). 4 As vector spaces, Ml(V ⊗ A) and M0(V ⊗ A) are bigger than V ⊗ A, since V ⊗ R(A) ⊆ Ml(V ⊗ A) and V ⊗ M (A) ⊆ M0(V ⊗ A), where R(A) is the right multipliers of A and M (A) is the multipliers of A (see Appendix in [5]). If V is finite-dimensional, then Ml(V ⊗ A) = V ⊗ R(A). 2.2 Comodules of a regular multiplier Hopf algebra Let A be a regular multiplier Hopf algebra and V a vector space. A right coaction of A on V is an injective linear map Γ : V → Mr(V ⊗ A), where Mr(V ⊗ A) is the extended module of the right A-module V ⊗ A, such that (Γ ⊗ ι)Γ = (ι ⊗ ∆)Γ. We call V a right A-comodule. Of course, we need to give a precise meaning to the last equation. Indeed, similar to Proposition 2.10 in [7], the map ι ⊗ ∆ and Γ ⊗ ι, defined on V ⊗ A, have natural extensions to maps from Mr(V ⊗ A) to Mr(V ⊗ A ⊗ A). For any v ∈ V , Γ(v) ∈ Mr(V ⊗A) and Γ(v)(1⊗a) ∈ V ⊗A. We will also use the adapted version of the Sweedler notation for this coaction, and write Γ(v)(1 ⊗ a) = v(0) ⊗ v(1)a for all a ∈ A and v ∈ V . Similarly, we can also define comodules by the other two kind of extended modules. These definitions are essentially the same, except for the ranges of the coactions. 2.3 Centre of a strict tensor category Let (C, ⊗, I) be a strict tensor category. We first recall the definition of the centre Z(C) and some of its properties (see [3]). An object of Z(C) is a pair (V, C−,V ), where V is an object of C (i.e., V ∈ C) and C−,V is a family of natural isomorphisms C−,V : X ⊗ V → V ⊗ X defined for all objects X in C such that for all objects X, Y in C we have CX⊗Y,V = (CX,V ⊗ ι)(ι ⊗ CY,V ). The naturality means that for any morphism f : X → Y in C, (ι ⊗ f )CX,V = CY,V (f ⊗ ι). A morphism from (V, C−,V ) to (W, C−,W ) is a morphism f : V → W such that for each object X of C we have (f ⊗ ι)CX,V = CX,W (ι ⊗ f ). Let (C, ⊗, I) be a strict tensor category. then Z(C) is a strict tensor category, where (1) the unit is (I, ι); (2) the tensor product of (V, C−,V ) and (W, C−,W ) is given by (V, C−,V ) ⊗ (W, C−,W ) = (V ⊗ W, C−,V ⊗W ), where C−,V ⊗W : X ⊗ V ⊗ W → V ⊗ W ⊗ X is the morphism of C defined for all objects X in C by CX,V ⊗W = (ι ⊗ CX,W )(CX,V ⊗ ι); (3) and the braiding is given by CV,W : (V, C−,V ) ⊗ (W, C−,W ) → (W, C−,W ) ⊗ (V, C−,V ). 5 3 A categorical interpretation of Yetter-Drinfel'd modules Let (A, ∆, ε, S) be a regular multiplier Hopf algebra. We first recall from [10] that a (left-right) Yetter-Drinfel'd module over A is a vector space V satisfying the following conditions • (V, ·) is an unital left A-module, i.e., A · V = V . • (V, Γ) is a (right) A-comodule, where Γ : V → M0(V ⊗ A) denotes the right coaction of A on V , M0(V ⊗ A) denote the extended module shown in section 2. • Γ and · satisfy the following compatible conditions (a · v)(0) ⊗ (a · v)(1)a′ = a(2) · v(0) ⊗ a(3)v(1)S−1(a(1))a′. (3.1) The Yetter-Drinfel'd module category AYDA is defined as follows. The objects in AYDA are left-right Yetter-Drinfel'd modules and the morphisms are linear maps that intertwine with the left action and the right coaction of A on M , i.e., the morphisms between two objects are left A-linear and right A-colinear maps. More precisely, let V, W ∈ AYDA and f : V → W be a morphism, then f (a · v) = a · f (v), ΓW ◦ f (v)(1 ⊗ a′) = (f ⊗ ι)ΓV (v)(1 ⊗ a′), (3.2) for all a, a′ ∈ A and v ∈ V , where ΓW (ΓV ) is the right coaction on W (V ). AYDA is a braided monoidal category in the following way. For V, W ∈ AYDA, the action and coaction of A on V ⊗ W is given by a · (v ⊗ w) = a(1) · v ⊗ a(2) · w, Γ(v ⊗ w)(1 ⊗ a′) = v(0) ⊗ w(0) ⊗ w(1)v(1)a′. (3.3) (3.4) for any v ∈ V, w ∈ W and a, a′ ∈ A. Let A be a regular multiplier Hopf algebra and denote by AM the unital left A-module category. The module structure is same as what in Yetter-Drinfel'd category. We will try to show the equivalence between Z(AM) and AYDA. Proposition 3.1 Let V be an object in AYDA. For any left A-module X, define CX,V (x ⊗ v) = v(0) ⊗ v(1) · x where x ∈ X and v ∈ V , then (V, C−,V ) ∈ Z(AM). Proof. Indeed, CX,V is well-defined, and clearly it is invertible C −1 X,V (v ⊗ x) = S(v(1)) · x ⊗ v(0). For any unital left A-module Y , x ∈ X, y ∈ Y , and v ∈ V , CX⊗Y,V (x ⊗ y ⊗ v) = v(0) ⊗ v(1) · x ⊗ v(2) · y 6 = CX,V (x ⊗ v(0)) ⊗ v(1) · y = (CX,V ⊗ ι)(x ⊗ v(0) ⊗ v(1) · y) = (CX,V ⊗ ι)(ι ⊗ CY,V )(x ⊗ y ⊗ v), we get that CX⊗Y,V = (CX,V ⊗ ι)(ι ⊗ CY,V ). Let f : X → Y be a left A-module map, CY,V (f ⊗ ι)(x ⊗ v) = CY,V (f (x) ⊗ v) = v(0) ⊗ v(1) · f (x) = v(0) ⊗ f (v(1) · x) = (ι ⊗ f )CX,V (x ⊗ v), thus C−,V is natural. CX,V is a left A-module map, since CX,V (a · (x ⊗ v)) = CX,V (a(1) · x ⊗ a(2) · v) = (a(2) · v)(0) ⊗ (a(2) · v)(1)a(1)x = a(1) · v(0) ⊗ a(2)v(1)x = a · CX,V (x ⊗ v), where the third equation holds because of the Yetter-Drinfel'd compatible condition. (cid:4) Let f : V → W be a morphism in AYDA, i.e., f is a module and comodule map. Then (f ⊗ ι)CX,V (x ⊗ v) = f (v(0)) ⊗ v(1) · x = f (v)(0) ⊗ f (v)(1) · x = CX,W (x ⊗ f (v)) = CX,W (ι ⊗ f )(x ⊗ v), so (f ⊗ ι)CX,V = CX,W (ι ⊗ f ). We can define a functor G : AYDA → Z(AM) as follows, G(V ) = (V, C−,V ), G(f ) = f, where CX,V (x ⊗ v) = v(0) ⊗ v(1) · x, f : V → W is a morphism in AYDA. Also we can get that G preserves tensor product, since for V, W ∈ AYDA, X ∈ AM , x ∈ X, v ∈ V, w ∈ W , CX,V ⊗W (x ⊗ v ⊗ w) = v(0) ⊗ w(0) ⊗ w(1)v(1) · x = v(0) ⊗ CX,W (v(1) · x ⊗ w) = (ι ⊗ CX,W )(v(0) ⊗ v(1) · x ⊗ w) = (ι ⊗ CX,W )(CX,V ⊗ ι)(x ⊗ v ⊗ w). Proposition 3.2 Let A be a regular multiplier Hopf algebra. For any object (V, C−,V ) in Z(AM), if Mr(V ⊗A) = M0(V ⊗A) as vector spaces, then V is a Yetter-Drinfel'd module over A. Proof. For any object (V, C−,V ) in Z(AM), we can endow V with the structure of a Yetter-Drinfel'd module. For any unital left A-module X and any element x ∈ X, there exist a local unit e ∈ A such that e · x = x, define the left A-module map ¯x : A → X by ¯x(a) = a · x. For each v ∈ V and a ∈ A, by the natruality of CX,V , we have CX,V (x ⊗ v) = CX,V (e · x ⊗ v) 7 = CX,V (¯x ⊗ ι)(e ⊗ v) = (ι ⊗ ¯x)CA,V (e ⊗ v) ∈ V ⊗ A · x. Set X = A with the multiplication as module action, then CA,V (ba⊗v) = CA,V (b⊗v)(1⊗a) for all a, b ∈ A and v ∈ V . If we treat A ⊗V and V ⊗A as right A-module with structures acting by multiplication on the A-component, then obviously the right module action is compatible with the left one, and CA,V (b ⊗ v) · a = CA,V ((b ⊗ v) · a). There is a natural extension Mr(A ⊗ V ) −→ Mr(V ⊗ A), denoted by C ′ A,V , C ′ A,V (y) · a = CA,V (y · a), y ∈ Mr(A ⊗ V ) for all a ∈ A. So by the assumption we can get C ′ A,V (1 ⊗ v) ∈ M0(V ⊗ A), where 1 is the unit of M (A). If we define ρ : V → M0(V ⊗ A) by ρ(v) = C ′ A,V (1 ⊗ v), and write CA,V (a ⊗ v) = A,V (1 ⊗ v) · a = ρ(v)(1 ⊗ a) = v(0) ⊗ v(1)a, then V is a right A-comodule, since by the C ′ definition of Z(C), CA⊗A,V (x ⊗ y ⊗ v) = v(0) ⊗ v(1) · (x ⊗ y) = v(0) ⊗ v(1)(1) · x ⊗ v(1)(2) · y = (CA,V ⊗ ι)(ι ⊗ CA,V )(x ⊗ y ⊗ v) = v(0)(0) ⊗ v(0)(1) · x ⊗ v(1) · y, then v(0) ⊗ v(1)(1)x ⊗ v(1)(2)y = v(0)(0) ⊗ v(0)(1)x ⊗ v(1)y for all x, y ∈ A, which proves that V is a right A-comodule. V is also a Yetter-Drinfel'd module. Since CX,V is a natural isomorphism in AM, CX,V is left A-linear, i.e., for all a ∈ A, x ∈ X and v ∈ V , a · CX,V (x ⊗ v) = CX,V (a · (x ⊗ v)). a · CX,V (x ⊗ v) = a · (v(0) ⊗ v(1) · x) = a(1) · v(0) ⊗ a(2)v(1) · x, CX,V (a · (x ⊗ v)) = CX,V (a(1) · x ⊗ a(2) · v) = (a(2) · v)(0) ⊗ (a(2) · v)(1)a(1) · x. From above, we can get the compatible condition of the Yetter-Drinfel'd module, i.e., for any a′ ∈ A, (a(2) · v)(0) ⊗ (a(2) · v)(1)a(1)a′ = a(1) · v(0) ⊗ a(2)v(1)a′. (cid:4) Assume that f : (V, C−,V ) → (W, C−,W ) is a morphism in Z(AM). Of course f is a module map. Since (f ⊗ ι)CA,V = CA,W (ι ⊗ f ), f is also a comodule map. Hence, we can define a functor F from Z(AM) to AYDA as follows: F ((V, C−,V )) = V, F (f ) = f. We can easily check that F preserve the tensor product. For any v ∈ V and w ∈ W , ρ(v ⊗ w)(1 ⊗ a) = CA,V ⊗W (a ⊗ v ⊗ w) = (ι ⊗ CA,W )(CA,V ⊗ ι)(a ⊗ v ⊗ w) = (ι ⊗ CA,W )(v(0) ⊗ v(1)a ⊗ w) = v(0) ⊗ w(0) ⊗ w(1)v(1)a, so F ((V ⊗ W, C−,V ⊗W )) = V ⊗ W . 8 Theorem 3.3 Let A be a regular multiplier Hopf algebra with Mr(V ⊗A) = M0(V ⊗A) for any vector space V . Then the Yetter-drinfel'd module category AYDA is equivalent to the centre Z(AM) of left unital A-module category AM. Proof. From above, F and G preserve the braiding. Also F G = 1 and GF = 1. So we have established the equivalence between Z(AM) and AYDA. (cid:4) The assumption Mr(V ⊗ A) = M0(V ⊗ A) as vector spaces in Proposition 3.2 makes A,V (1 ⊗ v) belong to M0(V ⊗ A), which is exactly what we need. This assumption C ′ sometimes naturally holds. If A has a unit, Mr(V ⊗A) = V ⊗A = M0(V ⊗A), the assumption holds, then Theorem 3.3 actually shows that for a Hopf algebra A the Yetter-drinfel'd module category AYDA is equivalent to the centre Z(AM) of A-module category AM. If A is commutative, the result Mr(V ⊗ A) = M0(V ⊗ A) as vector spaces also holds, and we get the following corollary Corollary 3.4 Let A be a commutative regular multiplier Hopf algebra. Then the Yetter-drinfel'd module category AYDA is equivalent to the centre Z(AM) of AM. From the proof in Proposition 3.2, CX,V is left and right A-linear, and these two structure are compatible. We can define a · C ′ A,V (1 ⊗ v) by (a · C ′ A,V (1 ⊗ v)) · b = a · (C ′ A,V (1 ⊗ v) · b) = a · CA,V (b ⊗ v) = CA,V (a(1)b ⊗ a(2) · v) = CA,V (a(1) ⊗ a(2) · v) · b. A,V (1⊗v) = CA,V (a(1) ⊗a(2) ·v) ∈ V ⊗A. If V is finite-dimensional, C ′ A,V (1⊗v) ∈ Then a·C ′ Mr(V ⊗ A) = V ⊗ L(A), then C ′ A,V (1 ⊗ v) ∈ V ⊗ M (A) = M0(V ⊗ A). Denote by AYDA f.d (AMf.d) the subcategory of AYDA (AM), in which the objects are all finite-dimensional. Then we have Theorem 3.5 Let A be a regular multiplier Hopf algebra. Then the category AYDA f.d is equivalent to the centre Z(AMf.d) of the category AMf.d. 4 Generalized Yetter-Drinfel'd modules over a regular mul- tiplier Hopf algebra Using the extended modules (of the right modules) shown in section 2, we first give the generalized definition of (left-right) Yetter-Drinfel'd module over a regular multiplier Hopf algebra, which generalizes the notion before (e.g. in [10]). We call it generalized Yetter-Drinfel'd module, and also we can define it by the extended modules of the right 9 ones similarly. Definition 4.1 Let (A, ∆, ε, S) be a regular multiplier Hopf algebra and V a vector space. Then V is called a generalized (left-right) Yetter-Drinfel'd module over A, if the following conditions hold: (1) (V, ·) is a unital left A-module, i.e., A · V = V ; (2) (V, Γ) is a right A-comodule, where Γ : V → Mr(V ⊗ A) denotes the right coaction of A on V and Mr(V ⊗ A) is the extended module of right A-module V ⊗ A; (3) Γ and · satisfy the following compatible condition: (a · v)(0) ⊗ (a · v)(1)a′ = a(2) · v(0) ⊗ a(3)v(1)S−1(a(1))a′. (4.1) for all a, a′ ∈ A and v ∈ V . Remark 4.2 (1) Although the compatible conditions are same, the Yertter-Drinfel'd module in [10] is different from this one, and can be considered as a special case of the above definition. Indeed, M0(V ⊗ A), the range of coaction, is the intersection of extended modules of right and left A-module V ⊗ A (see section 2.3 in [7] for more details). (2) The compatible condition make sense, since (a · v)(0) ⊗ (a · v)(1)a′ = Γ(a · v)(1 ⊗ a′) ∈ Indeed, S−1(a(1))a′ ⊗ a(2) ⊗ V ⊗ A, and the right-hand side also belongs to V ⊗ A. a(3) ∈ A ⊗ M (A ⊗ A), and v(0) ⊗ v(1)S−1(a(1))a′ = Γ(v)(1 ⊗ S−1(a(1))a′) ∈ V ⊗ A, then a(2) · v(0) ⊗ a(3)v(1)S−1(a(1))a′ ∈ V ⊗ A. (3) The compatible condition (4.1) is equivalent to the following condition (a(2) · v)(0) ⊗ (a(2) · v)(1)a(1)a′ = a(1) · v(0) ⊗ a(2)v(1)a′. (4) Let A be a Hopf algebra with unit 1, then Mr(V ⊗ A) = V ⊗ A. In this situation, the Yetter-Drinfel'd module for a Hopf algebra is a special case of Yetter-Drinfel'd module for a multiplier Hopf algebra. We will consider this at the end of this section. By the definition of generalized Yetter-Drinfel'd modules, we can define generalized Yetter-Drinfel'd module categories AGYDA. The objects in AGYDA are generalized left- right Yetter-Drinfel'd modules, and the morphisms are linear maps which interwine with the left action and the right coaction of A on V , i.e., the morphisms between two objects are left A-linear and right A-colinear maps. More precisely, let V, W ∈ AGYDA and f : V → W be a morphism, then f (a · v) = a · f (v), (ΓW ◦ f )(v)(1 ⊗ a′) = (f ⊗ ι)ΓV (v)(1 ⊗ a′), 10 for all a, a′ ∈ A and v ∈ V , where ΓW (ΓV ) is the right coaction on W (V ). The other three kind of generalized Yetter-Drinfel'd module categories can be also de- fined as in [10]. The results in [10] can be naturally generalized to this case, such as the following. Theorem 4.3 The category AGYDA is a braided monoidal category. Denote by Aut(A) the set of all isomorphisms α from A to A that are algebra maps satisfying (∆ ◦ α)(a) = (α ⊗ α) ◦ ∆(a) for all a ∈ A. Then for α, β ∈ Aut(A), we also can define generalized (α, β)-Yetter-Drinfel'd modules by the extended modules, and construct a new class of braided T -categories. The procedure is totally same as in [7]. Let (A, ∆, ε, S) be a regular multiplier Hopf algebra and V a vector space. Then V is called a (left-right) generalized (α, β)-Yetter-Drinfel'd module over A, if (1) (V, ·) is an unital left A-module, i.e., A · V = V . (2) (V, Γ) is a (right) A-comodule, where Γ : M → Mr(V ⊗ A) denotes the right coaction of A on V , and Mr(V ⊗ A) denote the extended module of right A-module V ⊗ A. (3) Γ and · satisfy the following compatible conditions (a · v)(0) ⊗ (a · v)(1)a′ = a(2) · v(0) ⊗ β(a(3))v(1)α(S−1(a(1)))a′. (4.2) Similarly, we can define the generalized (α, β)-Yetter-Drinfel'd module category, de- note it as AGYDA(α, β), by which we can also construct a new braided T -category. Define GYD(A) as the disjoint union of all AGYDA(α, β) with (α, β) ∈ G, where G = Aut(A) × Aut(A) is the twisted semi-direct square of group Aut(A) with product (α, β)#(γ, δ) = (αγ, δγ−1βγ). Then we have Theorem 4.4 GYD(A) is a braided T -category with the structures as follows. • Tensor product: if V ∈ AGYDA(α, β) and W ∈ AGYDA(γ, δ) with α, β, γ, δ ∈ Aut(A), then V ⊗ W ∈ AGYDA(αγ, δγ−1βγ), with the structures as follows: a · (v ⊗ w) = γ(a(1)) · v ⊗ γ−1βγ(a(2)) · w, Γ(v ⊗ w)(1 ⊗ a′) = (v ⊗ w)(0) ⊗ (v ⊗ w)(1)a′ = (v(0) ⊗ w(0)) ⊗ w(1)v(1)a′ for all v ∈ V, w ∈ W and a, a′ ∈ A. • Crossed functor: Let W ∈ AGYDA(γ, δ), define ϕ(α,β)(W ) = (α,β)W = W as vector space, with structures: for all a, a′ ∈ A and w ∈ W a ⇀ w = γ−1βγα−1(a) · w, 11 Γ(w)(1 ⊗ a′) =: w<0> ⊗ w<1>a′ = w(0) ⊗ αβ−1(w(1))a′. Then (α,β)W ∈ AGYDA((α, β)#(γ, δ)#(α, β)−1 ) = AGYDA(αγα−1, αβ−1δγ−1βγα−1). The functor ϕ(α,β) acts as identity on morphisms. • Braiding: If V ∈ AGYDA(α, β), and W ∈ AGYDA(γ, δ). Take V W = (α,β)W , define a map CV,W : V ⊗ W −→ V W ⊗ V by CV,W (v ⊗ w) = w(0) ⊗ β−1(w(1)) · v for all v ∈ V and w ∈ W . When A has a unit, i.e., A is a Hopf algebra, Mr(V ⊗A) = V ⊗A, and the AGYDA(α, β) is exactly AYDA(α, β) in the Hopf algebra case. Especially when A is a coFrobenius Hopf algebra with cointegral t in A and integral ϕ on A satisfying ϕ(t) = 1. Then we have AYDA(α, β) ∼= bA⊲⊳A(α,β) M, (4.3) where bA⊲⊳A(α,β) diagonal crossed product with the multiplication (see Definition 2.4 in [9]) M is the unital bA ⊲⊳ A(α, β)-module category, and bA ⊲⊳ A(α, β) is the (p ⊲⊳ a)(p′ ⊲⊳ a′) = p(α(a(1)) ◮ p′ ◭ S−1β(a(3))) ⊗ a(2)a′ for a, a′ ∈ A and p, p′ ∈ bA. Indeed, the correspondence is give as follows. If M ∈ AYDA(α, β), then M ∈ bA⊲⊳A(α,β) M with the structure (p ⊲⊳ a) · m = p((a · m)(1))(a · m)(0). Conversely, if M ∈ bA⊲⊳A(α,β) M, then M ∈ AYDA(α, β) with structures a · m = (ε ⊲⊳ a) · m, m 7→ m(0) ⊗ m(1) = (ϕ(·t(2)) ⊲⊳ 1) · m ⊗ S−1(t(1)). If α = β = ι, the result (5.4) is just Theorem 12 shown in [12]. 5 Yetter-Drinfel'd (H, A)-modules over a regular multiplier Hopf algebra Let A be a regular multiplier Hopf algebra, and R be an algebra with or without identity, but with non-degenerate product. Assume that R is an unital left A-module, we recall from [2, 8] that R is called a (left) A-module algebra if a · (xx′) = (a(1) · x)(a(2) · x′) (5.1) 12 for all a ∈ A and x, x′ ∈ R. We can extend the action of A on R to M (R) by the following formulas (a · x)x′ = a(1)(cid:0)x(S(a(2)) · x′)(cid:1), x(a · x′) = a(2)(cid:0)(S−1(a(1)) · x)x′(cid:1). (5.2) (5.3) It follows that M (R) is a left A-module, the action of A on M (R) is still non-degenerate but no longer unital, and a · 1 = ε(a)1. By a right coaction of A on R, we mean an injective map Γ : R −→ M (R⊗A) satisfying (i) Γ(R)(1 ⊗ A) ⊆ R ⊗ A and (1 ⊗ A)Γ(R) ⊆ R ⊗ A, (ii) (Γ ⊗ ι)Γ = (ι ⊗ ∆)Γ. If furthermore Γ is a homomorphism, then R is called a (right) A-comodule algebra. The injectivity of Γ is equivalent to the counitary property (ι ⊗ ε)Γ = ι. A (left-right) Yetter-Drinfel'd A-module algebra R is an algebra R (with or without unit) over the field K, which is an unital left A-module algebra and a right A-comodule algebra satisfying the compatibility condition (4.1). In the another word, R is an algebra (with or without the unitary property) in Yetter-Drinfel'd module category QA. A Yetter-Drinfel'd A-module algebra R is called A-commutative, if for all x, y ∈ R, xy = y(0)(y(1) · x). Let A be a quasitriangular multiplier Hopf algebra with quasitriangular structure R and H be a left A-module algebra. Then there is a natural right A-comodule structure on H, namely ρ(h) = τ (R)(h ⊗ 1). (5.4) Remark here that (5.4) makes sense. Since the flip map τ : A ⊗ A −→ A ⊗ A is a homo- morphism, it can be extended to M (A ⊗ A). For h ∈ H, there exists finite ai and hi such that h = Pi ai · hi. So τ (R)(h ⊗ 1) = Pi τ (R)(ai ⊗ 1)(hi ⊗ 1) ∈ M (R ⊗ A). Proposition 5.1 Let A be a quasitriangular multiplier Hopf algebra and H be a left A-module algebra with a A-comodule structure as (5.4). Then H is a right A-comodule algebra, and a Yetter-Drinfel'd A-module algebra. As shown in [10], the (left-right) Yetter-Drinfel'd module category QA is a braided monoidal category. From the above proposition, if A is quasitriangular, then the unital A-module category AM is a braided monoidal subcategory of AYDA, and the braiding on 13 AM is giveb by CM,N : M ⊗ N −→ N ⊗ M , CM,N (m ⊗ n) = τ (R)(n ⊗ m). Let A be a regular multiplier Hopf algebra and H be a Yetter-Drinfel'd module algebra. In the following, we introduced a new kind of modules: Yetter-Drinfel'd (H, A)-modules. Definition 5.2 Let H be a Yetter-Drinfel'd A-module algebra. A Yetter-Drinfel'd (H, A)-module M is a K-module, which has the structures of Yetter-Drinfel'd (A-)module and of left H-module with left H-action (denoted by ⇀) such that the following compat- ibility conditions hold: for all h ∈ H, m ∈ M and a, a′ ∈ A, a · (h ⇀ m) = (a(1) · h) ⇀ (a(2) · m), ρ(h ⇀ m)(1 ⊗ a′) = h(0) ⇀ m(0) ⊗ m(1)h(1)a′. (5.5) (5.6) Denote by H QA the category of Yetter-Drinfel'd (H, A)-modules and Yetter-Drinfel'd (H, A)-module maps. Note that if H is the trivial Yetter-Drinfel'd module algebra K, then the Yetter- Drinfel'd (H, A)-module is nothing but Yetter-Drinfel'd A-module. If A is a coFrobenius Hopf algebra, by [12] QA ∼= D(A)M. If H ia a D(A)-module algebra, then H QA is equiva- lent to the unital module category H#D(A)M, where D(A) = bA ⊲⊳ A the Drinfel'd double of A, and H#D(A) the usual smash product. Now we consider the case in which H ia an unital A-commutative Yetter-Drinfel'd module algebra. Take M in H QA, we may define a right action of H on M as follows: for any m ∈ M and h ∈ H, m ↼ h = h(0) ⇀ (h(1) · m) (5.7) It is no hard to check that this action is well-defined and along with the left H-module structure of M makes M into an H-bimodule. For M, N ∈ H QA, we form a tensor product M ⊗H N , and endow it with the following H-action and A-structures: for any m ∈ M , n ∈ N , a, a′ ∈ A and h ∈ H, a · (m ⊗ n) = a(1) · m ⊗ a(2) · n, χ(m ⊗ n)(1 ⊗ a′) = m(0) ⊗ n(0) ⊗ n(1)m(1)a′, h ⇀ (m ⊗ n) = h ⇀ m ⊗ n. Then M ⊗H N with the above structures is a Yetter-Drinfel'd (H, A)-module, denoted by M e⊗H N . Note that the right H-module structure of M e⊗H N , defined as in (5.7), is the same as the one coming from the right H-module N , i.e. (m ⊗ n) ↼ h = m ⊗ n ↼ h for m ∈ M , 14 n ∈ N and h ∈ H. Therefore, the standard map for Yetter-Drinfel'd (H, A)-modules X, Y and Z, Φ : (X e⊗H Y )e⊗H Z −→ Xe⊗H(Y e⊗H Z) is H-linear, and hence an isomorphism in H QA. By the construction above, we have Theorem 5.3 With notation above, (H QA, e⊗H , H) forms a monoidal category. Acknowledgements The work was partially supported by the NNSF of China (No. 11226070), the NJAUF (No. LXY201201019, LXYQ201201103). The first author Tao Yang is very grateful to Professor A. Van Daele for his lectures and discussions on the multiplier Hopf algebra theory. References [1] Caenepeel, S., Van Oystaeyen, F., and Zhang, Y. H. (1994). Quantum Yang-Baxter module algebras. K-Theory 8: 231-255. [2] Drabant, B., Van Daele, A. and Zhang, Y. H. (1999). Actions of multiplier hopf algebras. Communications in Algebra 27(9): 4117-4172. [3] Kassel, C. (1995). Quantum groups. Graduate Texts in Mathematics 155. [4] Panaite F. and Staic Mihai D. (2007). Generalized (anti) Yetter-Drinfel'd modules as components of a braided T-category. Israel Journal of Mathematics 158: 349-365. [5] Van Daele, A. (1994). Multiplier Hopf algebras. Transaction of the American Math- ematical Society 342(2): 917-932. [6] Van Daele, A. (1998). An algebraic framework for group duality. Advances in Math- ematics 140(2): 323-366. [7] Van Daele, A. (2008). Tools for working with multiplier Hopf algebras. The Arabian Journal for Science and Engineering 33(2C): 505-527. [8] Van Daele, A. and Zhang Y. H. (1999), Galois theory for multiplier Hopf algebras with integrals. Algebras and Representation Theory 2: 83-106. [9] Yang, T. and Wang, S. H. (2011). A lot of quasitriangular group-cograded multiplier Hopf algebras. Algebras and Representation Theory 14(5): 959-976. [10] Yang, T. and Wang, S. H. (2011). Constructing new braided T -categories over regular multiplier Hopf algebras. Communications in Algebra, 39(9): 3073-3089. 15 [11] Wang, S. H. (2002). Braided monoidal categories associative to Yetter-Drinfeld cate- gories. Communications in Algebra, 30(11): 5111-5124. [12] Zhang, Y. H. (1999). The quantum double of a coFrobenius Hopf algebra. Commu- nications in Algebra, 27(3): 1413-1427. 16
1107.3511	3	1107	2012-03-15T23:51:31	Category equivalences involving graded modules over path algebras of quivers	[ "math.RA" ]	Let kQ be the path algebra of a quiver Q with its standard grading. We show that the category of graded kQ-modules modulo those that are the sum of their finite dimensional submodules, QGr(kQ), is equivalent to several other categories: the graded modules over a suitable Leavitt path algebra, the modules over a certain direct limit of finite dimensional multi-matrix algebras, QGr(kQ') where Q' is the quiver whose incidence matrix is the n^{th} power of that for Q, and others. A relation with a suitable Cuntz-Krieger algebra is established. All short exact sequences in the full subcategory of finitely presented objects in QGr(kQ), split so that subcategory can be given the structure of a triangulated category with suspension functor the Serre degree twist (-1); it is shown that this triangulated category is equivalent to the "singularity category" for the radical square zero algebra kQ/kQ_{\ge 2}.	math.RA	math	CATEGORY EQUIVALENCES INVOLVING GRADED MODULES OVER PATH ALGEBRAS OF QUIVERS S. PAUL SMITH Abstract. Let Q be a finite quiver with vertex set I and arrow set Q1, k a field, and kQ its path algebra with its standard grading. This paper proves some category equivalences involving the quotient category QGr(kQ) := Gr(kQ)/Fdim(kQ) of graded kQ-modules modulo those that are the sum of their finite dimensional submodules, namely QGr(kQ) ≡ ModS(Q) ≡ GrL(Q◦) ≡ ModL(Q◦)0 ≡ QGr(kQ(n)). EndkI (kQ⊗n 1 ) is a direct limit of finite dimensional Here S(Q) = lim −→ semisimple algebras; Q◦ is the quiver without sources or sinks that is obtained by repeatedly removing all sinks and sources from Q; L(Q◦) is the Leavitt path algebra of Q◦; L(Q◦)0 is its degree zero component; and Q(n) is the quiver whose incidence matrix is the nth power of that for Q. It is also shown that all short exact sequences in qgr(kQ), the full sub- category of finitely presented objects in QGr(kQ), split. Consequently qgr(kQ) can be given the structure of a triangulated category with sus- pension functor the Serre degree twist (−1); this triangulated category is equivalent to the "singularity category" Db(Λ)/Dperf (Λ) where Λ is the radical square zero algebra kQ/kQ≥2, and Db(Λ) is the bounded derived category of finite dimensional left Λ-modules. 1. Introduction 1.1. Throughout k is a field and Q a finite quiver (directed graph) with vertex set I. Loops and multiple arrows between vertices are allowed. We write kQ for the path algebra of Q. We make kQ an N-graded algebra by declaring that a path is homo- geneous of degree equal to its length. The category of Z-graded left kQ- modules with degree-preserving homomorphisms is denoted by Gr(kQ) and we write Fdim(kQ) for its full subcategory of modules that are the sum of their finite-dimensional submodules. Since Fdim(kQ) is a localizing subcat- egory of Gr(kQ) we may form the quotient category QGr(kQ) := Gr(kQ) Fdim(kQ) . 1991 Mathematics Subject Classification. 16W50, 16E50, 16G20, 16D90 . Key words and phrases. path algebra; quiver; directed graph; graded module; quotient category; equivalence of categories; Leavitt path algebra. This work was partially supported by the U.S. National Science Foundation grant 0602347. 1 2 S. PAUL SMITH By [14, Prop. 4, p. 372], the quotient functor π∗ : Gr(kQ) → QGr(kQ) has a right adjoint that we will denote by π∗. We define O := π∗(kQ). The main result in this paper is the following theorem combined with an explicit description of the algebra S(Q) that appears in its statement. Theorem 1.1. The endomorphism ring of O in QGr(kQ) is an ultramatri- cial k-algebra, S(Q), and HomQGr(kQ)(O, −) is an equivalence with the category of right S(Q)-modules. QGr(kQ) ≡ ModS(Q) 1.2. Definition and description of S(Q). We write Qn for the set of paths of length n and kQn for the linear span of Qn. With this notation kQ =kI ⊕ kQ1 ⊕ kQ2 ⊕ · · · =TkI (kQ1) where TkI (kQ1) is the tensor algebra of the kI-bimodule kQ1. The ring of left kI-module endomorphisms of kQn is denoted Sn := EndkI (kQn). Since kQn+1 = kQ1 ⊗kI kQn the functor kQ1 ⊗kI − gives k-algebra homo- morphisms θn : Sn → Sn+1. Explicitly, if x1, . . . , xn+1 ∈ kQ1, f ∈ Sn, and ⊗ = ⊗kI, then The θns give rise to a directed system kI = S0 → S1 → · · · , and we define θn(f )(cid:0)x1 ⊗ · · · ⊗ xn+1(cid:1) := x1 ⊗ f (x2 ⊗ · · · ⊗ xn+1). S(Q) := lim −→ Sn. As a k-algebra, kI is isomorphic to a product of I copies of k. Every left kI-module is therefore a direct sum of 1-dimensional kI-submodules and the endomorphism ring of a finite dimensional kI-module is therefore a direct sum of ≤ I matrix algebras Mr(k) where the rs that appear are determined by the multiplicities of the simple kI-modules. Hence S(Q) is a direct limit of products of matrix algebras. Such algebras are called ultramatricial. Theorem 1.1 will follow from the following result. Theorem 1.2. The object O is a finitely generated projective generator in QGr(kQ) and EndQGr(kQ) O ∼= S(Q). The functor implementing the equivalence in Theorem 1.1 is HomQGr(kQ)(O, −). Ultramatricial algebras are described by Bratteli diagrams [4] (see also [18]). The Bratteli diagram for S(Q), and hence S(Q), is described explicitly in Proposition 5.1 in terms of the incidence matrix for Q. CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 3 1.3. Relation to Leavitt path algebras and Cuntz-Krieger algebras. Apart from the path algebra kQ two other algebras are commonly associated to Q, the Leavitt path algebra L(Q) and the Cuntz-Krieger algebra OQ. The algebra L(Q), which can be defined over any commutative ring, is an algebraic analogue of the C∗-algebra OQ because (when the base field is C) OQ contains L(Q) as a dense subalgebra. Theorem 1.3. Let Q◦ be the quiver without sources or sinks that is obtained by repeatedly removing all sinks and sources from Q. Then (1) QGr(kQ) ≡ QGr(kQ◦); (2) S(Q◦) ∼= L(Q◦)0; (3) L(Q◦) is a strongly graded ring; (4) QGr(kQ) ≡ ModS(Q) ≡ GrL(Q◦) ≡ ModL(Q◦)0. After proving this theorem the author learned that Roozbeh Hazrat had previously given necessary and sufficient conditions for L(Q◦) to be a strongly graded ring [19, Thm. 3.15]. The idea in our proof of (3) differs from that in Hazrat's paper. 1.4. Coherence. A ring R is left coherent if the kernel of every homomor- phism f : Rm → Rn between finitely generated free left R-modules is finitely generated. If R is left coherent we write modR for the full subcategory of ModR consisting of finitely presented modules; modR is then an abelian category. To prove R is left coherent it suffices to show that every finitely generated left ideal is finitely presented. A ring in which every left ideal is projective is left coherent so kQ is left coherent. A direct limit of left coherent rings is left coherent so S(Q) is left coherent. Because kQ is left coherent the full subcategory consisting of finitely presented graded left kQ-modules is abelian. The cat- gr(kQ) ⊂ Gr(kQ) egory fdim(cid:0)kQ(cid:1) :=(cid:0)gr(kQ)(cid:1) ∩(cid:0)Fdim(kQ)(cid:1) is the full subcategory of gr(kQ) consisting of finite dimensional modules. We now define define qgr(kQ) := ⊂ QGr(kQ). gr(kQ) fdim(kQ) Proposition 1.4. The equivalence in Theorem 1.1 restricts to an equiva- lence qgr(kQ) ≡ modS(Q). By [22, Prop. A.5, p. 113], qgr R consists of the finitely presented objects in QGr R and every object in QGr R is a direct limit of objects in qgr R.1 1An object M in an additive category A is finitely presented if HomA(M, −) commutes with direct limits; is finitely generated if whenever M = P Mi for some directed family of subobjects Mi there is an index j such that M = Mj ; is coherent if it is finitely presented and all its finitely generated subobjects are finitely presented. 4 S. PAUL SMITH 1.5. qgr(kQ) as a triangulated category. One of the main steps in prov- ing Theorems 1.1 and 1.2 is to prove the following. Proposition 1.5. Every short exact sequence in qgr(kQ) splits. If Σ is an auto-equivalence of an abelian category A in which every short exact sequence splits, then A can be given the structure of a triangulated category with Σ being the translation: one declares that the distinguished triangles are all direct sums of the following triangles: M → 0 → ΣM id −→ ΣM, M id −→ M → 0 → ΣM, 0 → M id−→ M → 0, as M ranges over the objects of A. Hence qgr(kQ) endowed with the Serre twist (−1) is a triangulated category. In section 7 we use a result of Xiao-Wu Chen to prove the following. Theorem 1.6. Let Q be a quiver and Λ the finite dimensional algebra kQ/kQ≥2. There is an equivalence of triangulated categories (cid:0) qgr(kQ), (−1)(cid:1) ≡ Db(modΛ)/Db perf (modΛ). 1.6. Equivalences of categories. It can happen that QGr(kQ) is equiva- lent to QGr(kQ′) with Q and Q′ being non-isomorphic quivers. Theorem 1.7. (See section 4.) If Q and Q′ become the same after repeatedly removing vertices that are sources or sinks, then QGr(kQ) ≡ QGr(kQ′). Let A be a Z-graded algebra. If m is a positive integer the algebra A(m) = ⊕i∈ZAim is called the mth Veronese subalgebra of A. When A is a commutative N-graded algebra the schemes Proj A and Proj A(m) are isomorphic. Verevkin proved a non-commutative version of that result: QGr A ≡ QGr A(m) if A is an N-graded ring generated by A0 and A1 [31, Thm. 4.4]. Theorem 1.8. Let Q be a quiver with incidence matrix C. Let Q(m) be the quiver with incidence matrix C m, m ≥ 1; i.e., Q(m) has the same vertices as Q but the arrows in Q(m) are the paths of length m in Q. Then QGr(kQ) ≡ QGr(kQ(m)). Proof. This follows from Verevkin's result because kQ(m) = (kQ)(m). (cid:3) The referee pointed out the following alternative proof of Theorem 1.8. First, Sn(Q(m)) = Snm(Q) so the directed system used to define S(Q(m)) is equal to the directed system obtained by taking every mth term of the directed system for S(Q). Hence S(Q) = S(Q(m)). Therefore QGr(kQ) ≡ ModS(Q) = ModS(Q(m)) ≡ QGr(kQ(m)). CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 5 We call Q(n) the nth Veronese of Q. In symbolic dynamics Q(n) is called the nth higher power graph of Q [23, Defn. 2.3.10]. Other equivalences involving strong shift equivalence of incidence matri- ces, a notion from symbolic dynamics, appear in [28]. Acknowledgements. Part of this paper was written during a visit to Bielefeld University in April 2011. I thank Henning Krause for the invitation and the university for providing excellent working conditions. Conversations with Gene Abrams, Pere Ara, Ken Goodearl, Henning Krause, Mark Tomforde, and Michel Van den Bergh, shed light on the mate- rial in this paper and I thank them all for sharing their ideas and knowledge. I am especially grateful to Xiao-Wu Chen, Roozbeh Hazrat, and the ref- eree, for reading earlier versions of this paper and pointing out typos and obscurities. Their comments and suggestions have improved the final ver- sion of this paper. I thank Chen for sending me an early version of his paper [5] and Hazrat for bringing his paper [19] to my attention (see the remark after Theorem 1.3). 2. The endomorphism ring of O Recall that O denotes π∗(kQ), the image of the graded left module kQ in the quotient category QGr(kQ). Notation. In addition to the notation set out at the beginning of section 1.2 we write Q≥n for the set of paths of length ≥ n and kQ≥n for its linear span. We note that kQ≥n is a graded two-sided ideal in kQ. We write ei for the trivial path at vertex i, Ei for the simple module at vertex i, and Pi = (kQ)ei. If p is a path in Q we write s(p) for its starting point and t(p) for the vertex at which it terminates. We write pq to denote the path q followed by the path p. Lemma 2.1. Let I 0 = {i ∈ I the number of paths starting at i is finite}. Let I ∞ := I − I 0 and let Q∞ be the subquiver of Q consisting of the vertices in I ∞ and all arrows that begin and end at points in I ∞. Let T be the sum of all finite-dimensional left ideals in kQ. Then (1) T is a two sided ideal; (2) T = (ei i ∈ I 0); (3) kQ/T ∼= kQ∞; (4) the only finite-dimensional left ideal in kQ∞ is {0}; (5) if f : kQ≥n → T is a homomorphism of graded left kQ-modules, then f (kQ≥n+r) = 0 for r ≫ 0. Proof. The result is obviously true if dimk kQ < ∞ so we assume this is not the case, i.e., Q has arbitrarily long paths; equivalently, Q∞ 6= ∅. (1) If L is a finite-dimensional left ideal in kQ so is Lx for all x ∈ kQ, whence T is a two-sided ideal. 6 S. PAUL SMITH (2), (3), (4). Since the paths beginning at a vertex i are a basis for (kQ)ei, dimk(kQ)ei < ∞ if and only if i ∈ I 0. Hence T contains {ei i ∈ I 0}. It is clear that ∼= kQ∞. kQ (ei i ∈ I 0) Let p be a path in Q∞. Then there is an arrow a ∈ Q∞ such that ap 6= 0. It follows that dimk(kQ∞)p = ∞. It follows that the only finite-dimensional left ideal in kQ∞ is {0}. Therefore T /(ei i ∈ I 0) = 0. (5) Let f : kQ≥n → T be a homomorphism of graded left kQ-modules. Every finitely generated left ideal contained in T has finite dimension so f (kQ≥n) has finite dimension. Hence kQ≥n/ ker f is annihilated by kQ≥r for r ≫ 0. In other words, ker f ⊃ (kQ≥r).(kQ≥n) = kQ≥n+r. (cid:3) The ideal T in Lemma 2.1 need not have finite dimension; for example, if Q is the quiver in Proposition 6.3, dimk T = ∞. Lemma 2.2. Let I be a graded left ideal of kQ. If kQ/I is the sum of its finite dimensional submodules, then I ⊃ kQ≥n for n ≫ 0. Proof. The image of 1 in kQ/I belongs to a finite sum of finite dimensional submodules of kQ/I so the submodule it generates is finite dimensional. Hence dimk(kQ/I) < ∞. Therefore kQ/I is non-zero in only finitely many degrees; thus I contains kQ≥n for n ≫ 0. (cid:3) By definition, the objects in QGr(kQ) are the same as those in the Gr(kQ) and the morphisms are HomQGr(kQ)(π∗M, π∗N ) = lim −→ HomGr(kQ)(M ′, N/N ′) where the direct limit is taken as M ′ and N ′ range over all graded submod- ules of M and N such that M/M ′ and N ′ belong to Fdim(kQ). Proposition 2.3. If N ∈ Gr(kQ), then HomQGr(kQ)(O, π∗N ) = lim −→ HomGr(kQ)(kQ≥n, N/N ′) where the direct limit is taken over all integers n ≥ 0 and all submodules N ′ of N such that N ′ is the sum of its finite dimensional submodules. Proof. This follows from Lemma 2.2. (cid:3) Lemma 2.4. Consider kQn as a left kI-module. The restriction map Φ : EndGr(kQ)(kQ≥n) −→ EndkI (kQn), Φ(f ) = f kQn, is a k-algebra isomorphism with inverse given by applying the functor kQ⊗kI − to each kI-module endomorphism of kQn. Proof. Each f ∈ EndGr(kQ)(kQ≥n) sends kQn to itself. Since f is a left kQ-module homomorphism it is a left kI-module homomorphism. Hence Φ is a well-defined algebra homomorphism. Since kQ≥n is generated by kQn as a left kQ-module, Φ is injective. Since kQ≥n ∼= kQ ⊗kI kQn every left CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 7 kI-module homomorphism kQn → kQn extends in a unique way to a kQ- module homomorphism kQ≥n → kQ≥n (by applying the functor kQ ⊗kI −). Hence Φ is surjective. (cid:3) Theorem 2.5. There is a k-algebra isomorphism EndQGr(kQ) O ∼= lim −→ HomGr(kQ)(kQ≥n, kQ≥n) = S(Q). Proof. By the definition of morphisms in a quotient category, (2-1) EndQGr(kQ) O = lim −→ HomGr(kQ)(I, kQ/T ′) where I runs over all graded left ideals such that dimk(kQ/I) < ∞ and T ′ runs over all graded left ideals such that dimk T ′ < ∞. If T ′ is a graded left ideal of finite dimension it is contained in the ideal T that appears in Lemma 2.1. The system of graded left ideals of finite codimension in kQ is cofinal with the system of left ideals kQ≥n. These two facts imply that EndQGr(kQ) O = lim −→ n HomGr(kQ)(kQ≥n, kQ/T ). Since kQ≥n is a projective left kQ-module the map HomGr(kQ)(kQ≥n, kQ) → HomGr(kQ)(kQ≥n, kQ/T ) is surjective. This leads to a surjective map (2-2) lim −→ n HomGr(kQ)(kQ≥n, kQ) → lim −→ n HomGr(kQ)(kQ≥n, kQ/T ). Suppose the image of a map f ∈ HomGr(kQ)(kQ≥n, kQ) is contained in T . By Lemma 2.1(5), the restriction of f to kQ≥n+r is zero for r ≫ 0. The map in (2-2) is therefore injective and hence an isomorphism. Since morphisms in HomGr(kQ)(kQ≥n, kQ) preserve degree the natural map HomGr(kQ)(kQ≥n, kQ≥n) → HomGr(kQ)(kQ≥n, kQ) is an isomorphism. It follows that EndQGr(kQ) O = lim −→ n HomGr(kQ)(kQ≥n, kQ≥n). However, HomGr(kQ)(kQ≥n, kQ≥n) ∼= EndkI (kQn) by Lemma 2.4 so the re- sult follows from the definition of S(Q). (cid:3) 3. Proof that O is a progenerator in QGr(kQ) Each M ∈ Gr(kQ) has a largest submodule belonging to Fdim(kQ), namely τ M := the sum of all finite-dimensional graded submodules of M . 8 S. PAUL SMITH 3.1. Up to isomorphism and degree shift, the indecomposable projective graded left kQ-modules are Pi = (kQ)ei ∼= kQ ⊗kI kei, i ∈ I, It follows that every where kei is the simple left kI-module at vertex i. projective module in Gr(kQ) is isomorphic to kQ ⊗kI V for a suitable graded kI-module V . Lemma 3.1. Let P, P ′ ∈ Gr(kQ) be graded projective modules generated by their degree n components. Every injective degree-preserving homomorphism f : P → P ′ splits. Proof. Without loss of generality we can assume n = 0, P = kQ ⊗kI U , and P ′ = kQ ⊗kI V . The natural map kQ ⊗kI − : HomkI(U, V ) → HomGr(kQ)(kQ ⊗kI U, kQ ⊗kI V ) is an isomorphism with inverse given by restricting a kQ-module homo- morphism to the degree-zero components. An injective homomorphism f : kQ ⊗kI U → kQ ⊗kI V in Gr(kQ) restricts to an injective kI-module homomorphism U → V which splits because kI is a semisimple ring. (cid:3) Part (1) of the next result is implied by [3, Thm. 3.14] but because we only prove it for graded modules a simpler proof is possible. Proposition 3.2. (1) Let M be a finitely generated graded left kQ-module. Then M is finitely presented if and only if for all n ≫ 0 M≥n ∼=Mi∈I Pi(−n)⊕mi for some integers mi depending on M and n. (2) If 0 → L → M → N → 0 is an exact sequence in gr(kQ), then 0 → L≥n → M≥n → N≥n → 0 splits for n ≫ 0. (3) If M ∈ qgr(kQ), there is a projective M ∈ gr(kQ) such that M ∼= π∗M . (4) Every short exact sequence in qgr(kQ) splits. (5) All objects in qgr(kQ) are injective and projective. Proof. (1) (⇐) Each Pi is finitely presented because it is a finitely gener- ated left ideal of the coherent ring kQ. Hence every Pi(−n) is in gr(kQ). Therefore, if there is an integer n such that M≥n is a finite direct sum of various Pi(−n)s, then M≥n is in gr(kQ) too. The hypothesis that M is finitely generated implies that M/M≥n is finite dimensional. But every fi- nite dimensional graded kQ-module is a quotient of a direct sum of twists of the finitely presented finite dimensional module kQ/kQ≥1 and is therefore in gr(kQ). In particular, M/M≥n ∈ gr(kQ). Since gr(kQ) is closed under extensions, M is in gr(kQ) too. CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 9 (⇒) Let M ∈ gr(kQ). Then there is an exact sequence 0 → F ′ −→ F → M → 0 in gr(kQ) with F and F ′ finitely generated graded projective kQ- modules. Since F ′, F , and M , are finitely generated, for all sufficiently large ≥n, F≥n, and M≥n, are generated as kQ-modules by F ′ n the modules F ′ n, Fn, and Mn, respectively. But kQ is hereditary so F≥n and F ′ ≥n are graded projective. Now Lemma 3.1 implies that the restriction f : F ′ ≥n → F≥n splits. Hence M≥n is a direct summand of F≥n. The result follows. f (2) By (1), N≥n is projective for n ≫ 0, hence the splitting. (3) There is some M in gr(kQ) such that M ∼= π∗M . But π∗M ∼= π∗(M≥n) for all n so (3) follows from (1). (4) By [14, Cor. 1, p. 368], every short exact sequence in qgr(kQ) is of the form (3-1) 0 −→ π∗L π∗f −→ π∗M π∗g −→ π∗N −→ 0 for some exact sequence 0 −→ L −→ N −→ 0 in gr(kQ). But (3-1) is also obtained by applying π∗ to the restriction 0 → L≥n → M≥n → N≥n → 0 which splits for n ≫ 0. Hence (3-1) splits. (cid:3) −→ M f g 3.2. By Proposition 3.2(5), O is a projective object in qgr(kQ). Lemma 3.3. O is a projective object in QGr(kQ). Proof. As noted in the proof of Proposition 3.2(4), an epimorphism in QGr(kQ) is necessarily of the form π∗g : π∗M → π∗N for some surjective homomorphism g : M → N in Gr(kQ). Let η : O → π∗N be a morphism in QGr(kQ). Then η ∈ lim −→ HomGr(kQ)(kQ≥n, N/N ′) where the direct limit is taken over all n ∈ N and all N ′ ⊂ N such that N/N ′ is the sum of its finite dimensional submodules, so η = π∗h for some n, some N ′, and some h : kQ≥n → N/N ′. Since kQ≥n is a projective object in Gr(kQ), h factors through N and for the same reason h factors through g. Hence there is a morphism γ : O → π∗M such that η = (π∗g) ◦ γ. (cid:3) Lemma 3.4. HomQGr(kQ)(O, −) commutes with all direct sums in QGr(kQ). Proof. Let Mλ M =Mλ∈Λ be a direct sum in QGr(kQ). Let Mλ, λ ∈ Λ, be graded kQ-modules such that π∗Mλ = Mλ. Because π∗ has a right adjoint it commutes with direct sums. Hence M = π∗M where M = ⊕λ∈ΛMλ. Because kQ≥n is a finitely 10 S. PAUL SMITH generated module we obtain the second equality in the computation HomQGr(kQ)(O, M) = lim −→ n HomGr(kQ)(kQ≥n, ⊕λ∈ΛMλ) = lim −→ n Mλ∈Λ HomGr(kQ)(kQ≥n, Mλ) =Mλ∈Λ =Mλ∈Λ HomGr(kQ)(kQ≥n, Mλ) lim −→ n HomQGr(kQ)(O, Mλ). (cid:3) This proves the lemma. 3.2.1. Notation. We write Pi = π∗Pi for the images of the indecomposable projectives in QGr(kQ). 3.2.2. If S is a set of objects in an additive category A we write add(S) for the smallest full subcategory of A that contains S and is closed under direct summands and finite direct sums. Lemma 3.5. For every positive integer n, O(−n) ∈ add(O). Proof. Let Im :={v ∈ I there is a path of length m that ends at v}, m ≥ 1, I0 :=I − I1, ∞ I∞ := Im, \m=1 Tm :=add{Pi i ∈ Im}, 0 ≤ m ≤ ∞. The vertices in I0 are the sources. A vertex is in I∞ if and only if for every m ≥ 1 there is a path of length m ending at it. If m ≫ 0, then I1 ⊃ I2 ⊃ · · · ⊃ Im = Im+1 = · · · = I∞ and, consequently, T1 ⊃ T2 ⊃ · · · ⊃ Tm = Tm+1 = · · · = T∞. To prove the lemma it suffices to show that O(−1) ∈ add(O) because, if it is, an induction argument would complete the proof: O(−1) ∈ add(O) implies that O(−2) ∈ add(O(−1)) ⊂ add(O), and so on. But O(−1) = ⊕i∈I Pi(−1) is a direct sum of an object in T0(−1) and an object in T1(−1) so it suffices to show that add(O) contains T0(−1) and T1(−1). If j is a sink, then Pj = kej so Pj = 0. Suppose j ∈ Im and j is not a sink. There is an exact sequence Pt(a)(−1) (·a) −→ Pj → kej → 0 0 → Ma∈s−1(j) CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 11 where kej denotes the simple module concentrated at vertex j. Therefore Pj ∼= Ma∈s−1(j) P t(a)(−1) If a ∈ s−1(j), then t(a) ∈ Im+1. Therefore Tm ⊂ for every vertex j. Tm+1(−1). On the other hand, if m ≥ 1 and i ∈ Im+1, there is an arrow a such that t(a) = i and s(a) ∈ Im so Pi(−1) is a direct summand of P s(a). Hence Tm+1(−1) ⊂ Tm. The previous paragraph shows that T0 ⊂ T1(−1) and Tm = Tm+1(−1) for all m ≥ 1. Thus (3-2) T1 = T2(−1) = · · · = Tm(−m + 1) for all m ≥ 1. For m ≫ 0, Tm = Tm+1 so Tm(−1) = Tm+1(−1) = Tm. Hence, for m ≫ 0, Tm = Tm(n) for all n ∈ Z. Therefore (3-2) implies T1 = T1(n) for all n ∈ Z. Since O = ⊕i∈I Pi, T1 ⊂ add(O). Therefore T1(−1) = T1 ⊂ add(O) and (cid:3) T0(−1) ⊂ T1(−2) = T1(−1) ⊂ add(O). Proposition 3.6. qgr(kQ) = add(O). Proof. Let M ∈ qgr(kQ). There is some M in gr(kQ) such that M ∼= π∗M . But π∗M ∼= π∗(M≥n) for all n so, by Proposition 3.2(1), if n ≫ 0 there are integers mi such that Pi(−n)⊕mi. M ∼=Mi∈I Each Pi(−n) belongs to add(O) by Lemma 3.5 so M ∈ add(O). (cid:3) Theorem 3.7. Let ModS(Q) be the category of right S(Q)-modules. The functor HomQGr(kQ)(O, −) provides an equivalence of categories QGr(kQ) ≡ ModS(Q) that sends O to S(Q). This equivalence restricts to an equivalence between qgr(kQ) and modS(Q), the category of finitely presented S(Q)-modules. Proof. By Proposition 3.6, O is a generator in qgr(kQ). Every object in QGr(kQ) is a direct limit of objects in qgr(kQ) so O is also a generator in QGr(kQ). Since O is a finitely generated, projective generator in the Grothendieck category QGr(kQ), HomQGr(kQ)(O, −) : QGr(kQ) → Mod(cid:0) EndQGr(kQ)(O)(cid:1). is an equivalence of categories [29, Example X.4.2]. The result now follows from the isomorphism EndQGr(kQ)(O) ∼= S(Q) in Theorem 2.5. (cid:3) 12 S. PAUL SMITH 4. Sinks and sources can be deleted 4.1. A vertex is a sink if no arrows begin at it and a source if no arrows end at it. Theorem 4.1. Suppose quivers Q and Q′ become the same after repeatedly removing sources and sinks and attached arrows. Let (Q◦, I ◦) be the quiver without sources or sinks that is obtained from Q by this process. Then there is an equivalence of categories QGr(kQ) ≡ QGr(kQ◦) ≡ QGr(kQ′). The equivalence of categories is induced by sending a representation (Mi, Ma; i ∈ I, a ∈ Q1) of Q to the representation (Mi, Ma; i ∈ I ◦, a ∈ Q◦ 1) of Q◦. A quasi-inverse to this is induced by the functor that sends a repre- 1) of Qe to the representation (Mi, Ma; i ∈ sentation (Mi, Ma; i ∈ I ◦, a ∈ Q◦ I, a ∈ Q1) of Q where Mi = 0 if i /∈ I ◦ and Ma = 0 if a /∈ Q◦ 1. 4.2. The fact that sinks and sources can be deleted is reminiscent of three other results in the literature. The category QGr(kQ) is related to the dynamical system with topological space the bi-infinite paths in Q viewed as a subspace of QZ 1 and automor- phism the edge shift σ defined by σ(f )(n) = f (n + 1). No arrow that begins at a source and no arrow that ends at a sink appears in any bi-infinite path so, as remarked after Example 2.2.8 in [23], since Q◦ "contains the only part of Q used for symbolic dynamics, we will usually confine our attention to [quivers such that Q = Q◦]". See also [23, Prop. 2.2.10]. Second, as remarked on page 18 of [24], "Cuntz-Krieger algebras are the C∗-algebras of finite graphs with no sinks or sources". Third, in section 4 of [5] it is shown that the singularity category of an artin algebra is not changed by deleting or adding sources or sinks. 4.3. Theorem 4.1 follows from the next two results. Proposition 4.2. Let t be a sink in (Q, I) and Q′ the quiver with vertex set I ′ := I − {t} and arrows Q′ 1 := {arrows in Q that do not end at t}. Then the functor i∗ : Gr(kQ′) → Gr(kQ) that sends a representation of Q′ to the "same" representation of kQ obtained by putting 0 at vertex t induces an equivalence of categories that sends O′ to O. QGr(kQ′) ≡ QGr(kQ) Proof. Since t is a sink, (kQ)et = ket. Hence et(kQ) is a two-sided ideal of kQ and, if M is a left kQ module, then etM is a submodule of M . It is clear that i∗ is the forgetful functor induced by the homomorphism kQ → kQ/(et) = kQ′. The functor i∗ has a left adjoint i∗ and a right adjoint CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 13 i!. The functor i∗ sends a kQ-module M to M/etM . It is easy to see that the counit i∗i∗ → idGr(kQ′) is an isomorphism. Both i∗ and i∗ are exact. Since i∗ and i∗ send direct limits of finite dimensional modules to direct limits of finite dimensional modules they induce functors ι∗ : QGr(kQ′) → QGr(kQ) and ι∗ : QGr(kQ) → QGr(kQ′). Because i∗i∗ ∼= idGr(kQ′) we have ι∗ι∗ ∼= idQGr(kQ′). If M ∈ Gr(kQ) there is an exact sequence 0 → etM → M → i∗i∗M → If a is any arrow, then aet = 0. Therefore etM is a direct sum of 1- 0. dimensional left kQ-modules, hence in Fdim(kQ). It follows that the unit idGr(kQ) → i∗i∗ induces an isomorphism idQGr(kQ) ∼= ι∗ι∗. Hence ι∗ and ι∗ are mutually quasi-inverse equivalences. Since kQ′ = kQ/etkQ, i∗ sends kQ′ to kQ/etkQ. The natural homo- morphism kQ → kQ/etkQ = i∗(kQ′) becomes an isomorphism in QGr(kQ) because etkQ ∈ Fdim(kQ). Hence ι∗O′ = O. (cid:3) Proposition 4.3. Let s be a source in (Q, I) and Q′ be the quiver with vertex set I ′ := I − {s} and arrows Q′ 1 := {arrows in Q that do not begin at s}. The functor i∗ : Gr(kQ′) → Gr(kQ) that sends a representation of Q′ to the "same" representation of kQ obtained by putting 0 at vertex s induces an equivalence, ι∗, of categories QGr(kQ′) ≡ QGr(kQ). Furthermore, O ∼= ι∗O′ ⊕ Ps where Ps = π∗(kQes) and π∗ is the quotient functor Gr(kQ) → QGr(kQ). Proof. Every kQ′-module becomes a kQ-module through the homomor- phism ϕ : kQ → kQ/(es) = kQ′; this is the exact fully faithful embedding i∗. A right adjoint to i∗ is given by the functor i!, i!M := HomkQ(kQ′, M ) = {m ∈ M esm = 0} = (1 − es)M. It is clear that the unit idGr(kQ′) → i!i∗ is an isomorphism of functors. Both i∗ and i! are exact. Since i∗ and i! send direct limits of finite dimensional modules to di- rect limits of finite dimensional modules there are unique functors ι∗ : QGr(kQ′) → QGr(kQ) and ι! : QGr(kQ) → QGr(kQ′) such that the dia- grams Gr(kQ′) QGr(kQ′) i∗ ι∗ Gr(kQ) and Gr(kQ) / QGr(kQ) QGr(kQ) i! ι! Gr(kQ′) / QGr(kQ′) / / / / / / 14 S. PAUL SMITH III.1]. (The vertical arrows in the diagrams are the commute [14, Sect. quotient functors.) Because idGr(kQ′) If M ∈ Gr(kQ), there is an exact sequence 0 → i∗i!M → M → ¯M → 0 in which ¯M is supported only at the vertex s; a module supported only at s is a sum of 1-dimensional kQ-modules so belongs to Fdim(kQ). It follows that the unit i∗i! → idGr(kQ) induces an isomorphism idQGr(kQ) ∼= i!i∗, ι!ι∗ ∼= idQGr(kQ′). ∼= ι∗ι!. Hence ι∗ and ι! are mutually quasi-inverse equivalences. The isomorphism O ∼= ι∗O′ ⊕ Ps is proved in section 4.5 below. (cid:3) It need not be the case that the equivalence ι∗ in Proposition 4.3 sends 4.4. O′ to O. We will show that ι∗O′ 6∼= O for the quivers Q′ = Q = and / v b a s v b Since kQ′ is a polynomial ring in one variable, QGr(kQ′) ≡ Modk and this equivalence sends O′, the image of kQ′ in QGr(kQ′), to k. Since ι∗ is an equivalence it follows that ι∗O′ is indecomposable. Now kQ is isomorphic as a graded left kQ-module to the direct sum of the projectives Ps and Pv. Right multiplication by the arrow a gives an isomorphism Pv → (Ps)≥1(1) in Gr(kQ). Let Ps = π∗Ps and Pv = π∗Pv. Then O = Pv ⊕ Ps ∼= Pv ⊕ Pv(−1). Hence O 6∼= ι∗O′. Right multiplication by b induces an isomorphism Pv ∼ −→ Pv(−1). 4.5. We now prove the last sentence of Proposition 4.3. Because s is a source, the two-sided ideal (es) is equal to kQes. Hence as a graded left kQ-module i∗(kQ′) is isomorphic to kQ/kQes which is iso- morphic to kQ(1 − es). The claim that O ∼= ι∗O′ ⊕ Ps now follows from the decomposition kQ = kQ(1 − es) ⊕ kQes. 4.5.1. Because O ∼= ι∗O′ ⊕ Ps, ι!O ∼= O′ ⊕ ι!Ps. Moreover, ι!Ps is isomorphic to ⊕s(a)=sP′ t(a) is the image in QGr(kQ′) of (kQ′)et(a). t(a)(−1) where P′ 5. Description of S(Q) We will give two different descriptions of S(Q). In section 5.1, we describe S(Q) in terms of its Bratteli diagram. See [4] and [12] for information about Bratteli diagrams. In section 5.3, we show that when Q has no sinks or sources S(Q) is isomorphic to the degree zero component of the Leavitt path algebra, L(Q), associated to Q and, because L(Q) is strongly graded, QGr(kQ) ≡ GrℓL(Q) ≡ ModℓL(Q)0 where the subscript ℓ means left modules. It is well-known that L(Q) is a dense subalgebra of the Cuntz-Krieger algebra OQ, associated to Q. The philosophy of non-commutative geometry / e e e e CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 15 suggests that kQ is a homogeneous coordinate ring for a non-commutative scheme whose underlying non-commutative topological space has OQ as its ring of "continuous C-valued functions". 5.1. The Bratteli diagram for S = S(Q). Because kI has I isoclasses of simple modules and Sn is the endomorphism ring of a left kI-module, Sn is a product of at most I matrix algebras of various sizes; although fewer than I matrix algebras may occur in the product it is better to think there are I of them with the proviso that some (those corresponding to sources) might be 0 × 0 matrices. The nth level of the Bratteli diagram for S(Q) therefore consists of I vertices, each denoted by •, that we label (n, i), i ∈ I. The vertex labelled (n, i) represents the summand EndkI(ei(kQ)) of Sn; this endomorphism ring is isomorphic to a matrix algebra Mr(k) for some integer r; it is common practice to replace the symbol • at (n, i) by the integer r; we then say that r is the number at vertex (n, i) . We do this for some of the examples in section 6. We will see that the number of edges from (n, i) to (n + 1, j) is the same as the number of edges from i to j in Q. The Bratteli diagram is there- fore stationary in the terminology of [13], and the following example, which appears in [21], illustrates how to pass from the quiver to the associated Bratteli diagram. Q = Sn Sn+1 • • • • • ❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ qqqqqqqqqqqqqq qqqqqqqqqqqqqq ✂✂✂✂✂✂✂✂ ✂✂✂✂✂✂✂✂ ✂✂✂✂✂✂✂✂ • • • • • • • 5.2. Let C := (cij)i,j∈I be the incidence matrix for Q with the convention that cij = the number of arrows from j to i. The ij-entry in C n, which we denote by c[n] n from j to i. The number of paths of length n ending at vertex i is ij , is the number of paths of length c[n] ij . pn,i :=Xj∈I Proposition 5.1. The sum of the left kI-submodules of kQn isomorphic to Ei is equal to ei(kQn). Its dimension is equal to pn,i and (5-1) Sn ∼=Mi∈I EndkI(cid:0)ei(kQn)(cid:1) ∼=Mi∈I Mpn,i(k). o o o o f f o o f f b b 16 S. PAUL SMITH Referring to the Bratteli diagram for lim −→ (n, i) is pn,i, and the number of edges from (n, i) to (n + 1, j) is cji. Sn, the number at the vertex labelled Composing with the inclusions and projections in (5-1), the components of the map θn : Sn → Sn+1 are the maps EndkI(cid:0)ei(kQn)(cid:1) → Sn θn−→ Sn+1 → EndkI(cid:0)ej(kQn+1)(cid:1) that send a matrix in EndkI(cid:0)ei(kQn)(cid:1) to cji "block-diagonal" copies of itself in EndkI(cid:0)ej(kQn+1)(cid:1). Proof. The irreducible representation of Q at vertex i is Ei = kei. Since a path p ends at vertex i if and only if p = eip, the multiplicity of Ei in a composition series for kQn as a left kI-module is [kQn : Ei] = dimk ei(kQn) = the number of paths of length n ending at i = pn,i. The existence of the left-most isomorphism in (5-1) follows at once from the fact that kI is a semisimple ring; the second isomorphism follows from the analysis in the first part of this paragraph. Up to isomorphism, {E∗ i = Homk(Ei, k) i ∈ I} is a complete set of simple right kQ modules. It follows that the kI-bimodules Eij := Ei ⊗ E∗ j , i, j ∈ I, form a complete set of isoclasses of simple kI-bimodules. If a is an arrow from j to i there is a kI-bimodule isomorphism ka ∼= Eij so the multiplicity of Eij in kQ1 is the number of arrows from j to i, i.e., [kQ1 : Eij] = cij. More explicitly, E⊕cij ∼= ei(kQ1)ej. ij The image in Sn+1 of a map f ∈ Sn is the map kQ1 ⊗ f . Hence if f is belongs to the component EndkI(cid:0)ei(kQn)(cid:1) of Sn, the component of kQ1 ⊗ f in EndkI(cid:0)ej(kQn+1)(cid:1) is ejkQ1ei ⊗ f . But the dimension of ejkQ1ei is [kQ1 : Eji] = cji. (cid:3) 5.3. Leavitt path algebras. Goodearl's survey [17] is an excellent intro- duction to Leavitt path algebras. Since QGr(kQ) is unchanged when Q is replaced by the quiver obtained by repeatedly deleting sources and sinks, the essential case is when Q has no sinks or sources. For the remainder of section 5 we assume Q has no sinks or sources. This is equivalent to the hypothesis that Q = Q◦. 5.3.1. Under the hypothesis that Q = Q◦, (1) the Leavitt path algebra of Q, L(Q), is a universal localization of kQ in the sense of [7, Sect. 7.2] or [25, Ch. 4]; CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 17 (2) L(Q) is a strongly Z-graded ring and L(Q)0 ∼= S(Q)op so GrℓL(Q) ≡ ModℓL(Q)0 ≡ ModrS(Q) ≡ QGr(kQ) where the subscripts ℓ and r denote left and right modules. (3) L(Q) is a dense subalgebra of the Cuntz-Krieger algebra OQ. 5.3.2. Statement (1) in 5.3.1 holds for every finite Q (i.e., the hypothesis Q = Q◦ is not needed) [3]. For 5.3.1(3), see [1] and [24]. The fact that L(Q) is strongly graded when Q = Q◦ is proved in [19, Thm. 3.11]. We give an alternative proof of this in Proposition 5.5(6). A Z-graded ring R is strongly graded if RnR−n = R0 for all n. When R is strongly graded GrR is equivalent to ModR0 via the functor M M0 [9, Thm. 2.8]. That explains the left-most equivalence in (2). The second equivalence in 5.3.1(2) follows from the fact that S(Q)op ∼= L(Q)0 which we will prove in Theorem 5.4 below. 5.4. Let Ei be the 1-dimensional left kQ-module supported at vertex i and concentrated in degree zero. Because Q has no sinks Ei is not projective and its minimal projective resolution is (5-2) 0 −→ Ma∈s−1(i) Pt(a) fi−→ Pi −→ Ei −→ 0 where Pj = (kQ)ej and the direct sum is over all arrows starting at i. Elements in the direct sum will be written as row vectors (xa, xb, . . .) with xa ∈ Pt(a), xb ∈ Pt(b), and so on. The map fi is right multiplication by the column vector (a, b, . . .)T where a, b, . . . are the arrows starting at i, i.e., (5-3) fi(xa, xb, . . .) = (xa, xb, . . .)  a b ...   = xaa + xbb + · · · ∈ (kQ)ei. 5.4.1. Definition of L(Q) as a universal localization. We refer the reader to [7, Sect. 7.2] and [25, Ch. 4] for details about universal localization. Let Σ = {fi i ∈ I} and let L(Q) := Σ−1(kQ) be the universal localization of kQ at Σ. Since Q will not change in this section we will often write L for L(Q). If x ∈ kQ we continue to write x for its image in L under the universal Σ-inverting map kQ → L. Let P ′ := Ma∈s−1(i) Let(a). 18 S. PAUL SMITH The defining property of L is that the map kQ → L is universal subject to the condition that applying L ⊗kQ − to (5-2) produces an isomorphism q   a b ...   / Lei idL ⊗fi : P ′ for all i. Thus L ⊗kQ Ei = 0 for all i ∈ I. Every L-module homomorphism Lei → Lej is right multiplication by an element of L so [25, Thm. 4.1] tells us L is generated by kQ and elements a∗, b∗, . . . such that the inverse of idL ⊗fi is right multiplication by the row vector (a∗, b∗, . . .) where a∗ = eia∗et(a), etc. In particular, the defining relations for L are given by = idLei and (a∗, b∗, . . .) = idP ′ . Since idLei is right multiplication by ei and idP ′ is right multiplication by (a∗, b∗, . . .)  a b ...   0 0 · · · et(b) 0 · · · et(a) 0 ...   L = kQ ha∗ a ∈ Q1i modulo the relations a b ...       0 0 ... , es(a)a∗et(a) = a∗ aa∗ = et(a) ab∗ = 0 for all arrows a ∈ Q1, for all arrows a ∈ Q1, if a and b are different arrows, ei = Xa∈s−1(i) a∗a for all i ∈ I. 5.4.2. Our L(Q) is not defined in the same way as the algebra Lk(Q) defined in [17, Sect. 1]. Because our notational convention for composition of paths is the reverse of that in [17, Sect. 1.1] the relations for L(Q) just above are the opposite of those for Lk(Q) in [17, Sect. 1.4]. (If we had defined L(Q) by inverting homomorphisms between right instead of left modules we would have obtained the relations in [17, Sect. 1.4] but then our convention for composition of paths would have created the problems discussed at the end of [17, Sect. 1.8].) As a consequence, our L(Q) is anti-isomorphic to Lk(Q). However, as explained at the end of [17, Sect. 1.7], Lk(Q) is anti-isomorphic to itself via a map that sends each arrow a to a∗ and fixes ei for each vertex i. Thus, our L(Q) is isomorphic to the algebra Lk(Q) defined in [17]. Proposition 5.2. The algebra L(Q) = Σ−1(kQ) is isomorphic and anti- isomorphic to the Leavitt path algebra Lk(Q) defined in [17, Sect. 1.4]. / CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 19 5.4.3. Our convention for composition of paths is that used by (most of) the finite dimensional algebra community and by Raeburn [24, Rmk. 1.1.3]. However, it is not the convention adopted in [1] (see [1, Defn. 2.9]). If a, b, . . . , c, d ∈ Q1 and p = dc . . . ba we define p∗ := a∗b∗ . . . c∗d∗. If 5.5. p and q are paths of the same length, then (5-4) pq∗ = δp,qet(q) = δp,qet(p). If p is a path in Q of length n we write p = n. We give L(Q) a Z-grading by declaring that deg a∗ = −1 for all a ∈ Q1. For completeness we include the following well-known fact. deg a = 1 and Lemma 5.3. The degree-n component of L(Q) is Ln = span{p∗q p and q are paths such that q − p = n}. Proof. Certainly L(Q) is spanned by words in the letters a and a∗, a ∈ Q1. Let w be a non-zero word and ab∗ a subword of w with a, b ∈ Q1. Since w 6= 0, ab∗ = aa∗ = et(a); but et(a) can be absorbed into the letters on either side of aa∗ so, repeating this if necessary, w = p∗q for some paths p and q. The degree of p∗q is q − p so the result follows. (cid:3) Theorem 5.4. The algebras L(Q)0 and S(Q) are anti-isomorphic, L(Q)0 ∼= S(Q)op. Proof. By definition, S(Q) is the ascending union of its subalgebras Sn = EndkI (kQn). We will sometimes write L for L(Q). It is clear that L0 is the ascending union of its subspaces L0,n := span{p∗q p, q ∈ Qn} and each L0,n is a subalgebra of L because (p∗q)(x∗y) = δxqp∗y. It is also clear that L0,n ⊂ L0,n+1 because p∗q = Xa∈s−1(t(q)) p∗a∗aq (this uses the fact that Q has no sinks). By [6, Prop. 4.1], the linear map kQ → L(Q), p 7→ p, is injective for any quiver Q. As a consequence, there is a well-defined linear map Φn : L0,n → EndkI(kQn), Φn(p∗q)(r) := rp∗q (= δrpet(p)q), for r ∈ Qn. Since Φn(p∗q)Φn(x∗y)(r) = rx∗yp∗q = Φn(x∗yp∗q)(r) Φn is an algebra anti-homorphism. Since Φn(p∗q)(p) = q and Φn(p∗q)(r) = 0 if r 6= p, Φn is injective. 20 S. PAUL SMITH We will now show that L0,n and EndkI(kQn) have the same dimension. This will complete the proof that Φn is an algebra isomorphism. Since (5-5) it follows that On the other hand, {p∗q p, q ∈ eiQn}. {non-zero p∗q p, q ∈ Qn} = Gi∈I dimk L0,n =Xi∈I eiQn2. kQn =Mi∈I ei(kQn) =Mi∈I keiQn and keiQn is isomorphic as a left kI-module to a direct sum of eiQn copies of the simple kI-module kei. Hence EndkI(kQn) =Mi∈I Endk(keiQn) ∼=Mi∈I MeiQn(k) is the direct sum of I matrix algebras of sizes eiQn, i ∈ I. This completes the proof that L0,n and EndkI(kQn) have the same dimension. Hence Φn is an isomorphism. Rather than counting dimensions one can give a more honest proof by observing that the elements in {p∗q p, q ∈ eiQn} are a set of matrix units for Endk(keiQn) with respect to the basis eiQn. To complete the proof of the theorem we will show the Φns induce an isomorphism between the direct limits by showing that the diagram L0,n Φn Sn / L0,n+1 Φn+1 f 7→id ⊗f / Sn+1 commutes. To this end, let p∗q ∈ L0,n where p, q ∈ Qn, and let r ∈ Qn and a ∈ Q1 be such that ar 6= 0. Thus ar ∈ Qn+1 and a ⊗ r ∈ kQ1 ⊗kI kQn. Going clockwise around the diagram, Φn+1(p∗q)(ar) = arp∗q. Going anti- clockwise around the diagram, (idkQ1 ⊗Φn)(p∗q)(a ⊗ r) = a ⊗ Φn(p∗q)(r) = a ⊗ rp∗q = arp∗q. The diagram commutes. (cid:3) 5.6. Most of the next result, but not part (6), is covered by [2, Sect. 2.3]. It makes use of a construction in [2] that we now recall. Let R be a ring and φ : (Z≥1, +) → EndRing(R) a monoid homomor- phism to the monoid of ring endomorphisms of R. Let Z[t+] and Z[t−] be the polynomial rings on the indeterminates t+ and t−. We define the ring R[t−, t+; φ] to be the quotient of the free coproduct Z[t−] ∗Z R ∗Z Z[t+] / / CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 21 modulo the ideal generated by the relations r ∈ R, r ∈ R, t−r =φ(r)t−, rt+ =t+φ(r), t−t+ =φ(1) t+t− =1. The image of the map φ in the next result is the subalgebra t−Lt+ = eLe where e is the idempotent t−t+. Proposition 5.5. We continue to assume Q has neither sinks nor sources and L denotes L(Q). For each i ∈ I pick an arrow ai ending at i and define ai t+ :=Xi∈I and t− := t∗ +. Define a non-unital ring homomorphism φ : L → L by φ(x) := t−xt+. Then (1) t+t− = 1; (2) If n > 0, then Ln = tn (3) L(Q) is generated by L0 and t+ and t−; (4) in the notation of [2], L = L0[t−, t+; φ]; (5) if n is positive, Ln = (L1)n and L−n = (L−1)n; (6) L(Q) is strongly graded. +L0 and L−n = L0tn −; Proof. (1) This follows from the fact that aia∗ (2) Suppose n > 0 and let b ∈ Ln and c ∈ L−n. Then b = tn j = 0 if i 6= j. −b and + ∈ L0. This proves (2) and (3) is an immediate i = ei and aia∗ +tn −, and tn −b, ctn +tn c = ctn consequence. (4) See [2, Sect. 2 and Lem. 2.4]. (5) This is proved by induction. For example, assuming n > 0 and starting with (2) and the induction hypothesis, Ln = tn +L0, we have (L1)n+1 = (L1)nL1 = tn +L0L1 = tn +L1 = tn+1 + L0 = Ln+1. The proof for L−n is similar. − ∈ LnL−n so LnL−n = L0. (6) Suppose n > 0. Then 1 = tn Now +tn 1 =Xi∈I ei =Xi∈I Xa∈s−1(i) a∗a = Xa∈Q1 a∗a so 1 ∈ L−1L1. It now follows from (5) that L−nLn = L0. (cid:3) Dade's Theorem [9, Thm. 2.8] on strongly graded rings gives the next result. Corollary 5.6. If Q has neither sinks nor sources, there is an equivalence of categories GrL(Q) ≡ ModL0. 22 S. PAUL SMITH Because L(Q) is strongly graded, [9, (2.12a)] tells us that each Ln is an invertible L0-bimodule and the multiplication in L gives L0-bimodule isomorphisms Lm ⊗L0 Ln ∼ −→ Lm+n for all m and n. In other words, if we use the multiplication in L to identify 1 with L−1 and define L⊗(−r) L−1 := (L−1)⊗r for all r > 0, then 1 ∞ where the tensor product is taken over L0. L = L⊗n 1 Mn=−∞ 5.7. We have now completed the proof that QGr(kQ) ≡ ModrS(Q) ≡ ModℓL(Q)0 ≡ GrℓL(Q) when Q has no sinks or sources. It is possible to prove the equivalence QGr(kQ) ≡ GrℓL(Q) directly by modifying the arguments in section 4 of [27] for the free algebra khx0, . . . , xni so they apply to kQ. The required changes are minimal and straightforward so we leave the details of the next three results to the reader. The next result is proved in [3, Thm. 4.1] but the following proof is more direct. Proposition 5.7. The ring L is flat as a right kQ-module. Proof. Since L is the ascending union of the finitely generated free right kQ-modules (5-6) Fn = Xp∈Qn p∗(kQ) = Mp∈Qn p∗(kQ) (cid:3) it is a flat right kQ-module. A version of the following result for finitely presented not-necessarily- graded modules is given in [3, Sect. 6]. Proposition 5.8. If M ∈ Gr(kQ), then L ⊗kQ M = 0 if and only if M ∈ Fdim(kQ). Proof. The argument in [27, Prop. 4.3] works provided one replaces "free module" by "projective module". (cid:3) A version of the next result for finitely presented not-necessarily-graded modules is given in [3, Sect. 6]. Theorem 5.9. Let π∗ : Gr(kQ) → QGr(kQ) be the quotient functor and i∗ = L ⊗kQ − : Gr(kQ) → GrL. Then via a functor α∗ : QGr(kQ) → GrL such that α∗π∗ = i∗. QGr(kQ) ≡ GrL Proof. The argument in [27, Thm. 4.4] works here. (cid:3) CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 23 5.8. The referee pointed out that the equivalence between QGr(kQ) and GrL(Q) can be understood using ideas about perpendicular subcategories of Gr(kQ), as in [16], and ideas about universal localization that are implicit in [15]. We defined L(Q) as the universal localization Σ−1(kQ). Let ϕ : kQ → L(Q) be the universal Σ-inverting map. As Schofield remarks, [25, p.56], ϕ is an epimorphism in the category of rings. It is also an epimorphism in the category of graded rings. The restriction functor ϕ∗ : Gr(L(Q)) → Gr(kQ) therefore embeds Gr(L(Q)) as a fully exact subcategory (see [15, p. 280] for the definition) of Gr(kQ). Because L(Q) is flat as a right kQ-module (Prop. 5.7) ϕ : kQ → L(Q) satisfies the slightly stronger property of being a homological epimorphism in the category of graded rings (i.e., the equivalent properties of [16, Thm. 4.4] are satisfied). 6.1. If dimk kQ < ∞, then S(Q) = 0. 6. Examples 6.2. If Q is the cyclic quiver 1 / 2 / · · · / n then S(Q) ∼= kn. 6.3. By [27], if Q has one vertex and r arrows, then kQ is the free algebra khx1, . . . , xri, the Bratteli diagram for S(Q) is 1 · · / r · · / r2 · · / r3 · · / r4 · · · where there are r arrows between adjacent vertices, and S(Q) ∼= lim −→ n Mr(k)⊗n. 6.4. Different quivers can have a common Veronese quiver. For example, •9 is the 2-Veronese quiver of both •9 F Q = • • and Q′ = • ⊔ • It now follows from Theorem 1.8 that QGr(kQ) ≡ QGr(kQ′) and, by the comments after Theorem 1.8, S(Q) = S(Q(2)) = S(Q′) ∼=(cid:18) lim −→ n M2(k)⊗n(cid:19) ⊕(cid:18) lim −→ n M2(k)⊗n(cid:19). / / / h h / / / / / / / / / / / / 9 Y Y y y 9 Y Y y y / / o o ` ` Y Y Y Y 24 S. PAUL SMITH It is not obvious that QGr(khx, yi/(y2)) is equivalent to QGr(kQ) for 6.5. some quiver Q. Proposition 6.1. Let Q = 1 2 The Bratteli diagram for S(Q) is 1 1 ❃❃❃❃❃❃❃❃ @ 2 1 ❃❃❃❃❃❃❃❃ @ 3 2 ❃❃❃❃❃❃❃❃ @ 5 3 ❃❃❃❃❃❃❃❃ @ 8 5 ❆❆❆❆❆❆❆❆ >⑥⑥⑥⑥⑥⑥⑥⑥ · · · · · · and QGr(kQ) ≡ QGr khx, yi (y2) . Proof. We will use the notation Ei and Eij that appears in the proof of Proposition 5.1. The powers of the incidence matrix for Q are C n =(cid:18) fn fn−1 fn−2(cid:19) fn−1 where f−1 = 0, f0 = f1 = 1, and fn+1 = fn + fn−1 for n ≥ 1. As a kI-bimodule kQn ∼= Efn 11 ⊕ Efn−1 12 ⊕ Efn−1 21 ⊕ Efn−2 22 and as a left kI-module kQn ∼= Efn+1 1 ⊕ Efn 2 . It follows that Sn ∼= Mfn(k) ⊕ Mfn−1(k) and the Bratteli diagram is as claimed. But this Bratteli diagram also arises in [26] where it is shown that ModS(Q) ≡ QGr khx, yi (y2) . It also follows from the main result in [21], which was written after this paper, that QGr(khx, yi/(y2)) is equivalent to QGr(kQ). (cid:3) As explained in [26], we can interpret khx, yi/(y2), and therefore kQ, as a non-commutative homogeneous coordinate ring of the space of Penrose tilings of the plane. This is consistent with Connes' view [8, Sect. II.3] of the norm closure of S(Q), over C, as a C∗-algebra coordinate ring for the space of Penrose tilings. For each integer r ≥ 1 there is a quiver Q such that QGr(kQ) is equivalent QGr khx, yi/(yr+1); see section 7.2 and [21]. * * 9 9 j j / / / / / / / / / / @ @ @ @ > CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 25 6.6. The previous example can be generalized as follows. Proposition 6.2. Let Q = m q q where there are m arrows from the left-hand vertex to itself. The Bratteli diagram for S(Q) is 1 1 m &▲▲▲▲▲▲▲▲▲▲▲▲▲ 8rrrrrrrrrrrrr q1 q0 m &▼▼▼▼▼▼▼▼▼▼▼▼▼ 8qqqqqqqqqqqqq q2 q1 m &▼▼▼▼▼▼▼▼▼▼▼▼▼ 8qqqqqqqqqqqqq q3 q2 m &▼▼▼▼▼▼▼▼▼▼▼▼▼ 8qqqqqqqqqqqqq · · · · · · where there are m arrows from qn to qn+1 in the top row, and the numbers qn are given by q0 = q−1 = 1 and qn+1 = mqn + qn−1 for n ≥ 0. Furthermore, the Hilbert series of kQ, viewed as an element of K0(kI)[[t]] = (Z × Z)[[t]], is HkQ(t) = 1 − mt + t2 (1 + t, 1 + (1 − m)t) 1 with the first component of HkQ(t) giving the multiplicity in kQn of the simple kI-module that is supported at the left-most vertex. If Q and Q′ become the same after repeatedly deleting sources and 6.7. sinks, then QGr(kQ) ≡ QGr(kQ′) by Theorem 4.1. Therefore S(Q) and S(Q′) are Morita equivalent, but they need not be isomorphic as the next example shows. The quivers Q and Q′ are formed by adjoining a sink, respectively, a source, to (6-1) By Theorem 4.1 Q◦ = • x QGr(kQ) ≡ QGr(kQ′) ≡ QGr(kQ◦) ≡ Qcoh(Proj k[x]) ≡ Modk. In this example, Q′ = Qop. Proposition 6.3. Let Q = 1 2 and Q′ = 19 2. The Bratteli diagram for S(Q) is 1 1 @ 1 1 @ 1 1 @ 1 1 and that for S(Q′) is 1 · · · >⑥⑥⑥⑥⑥⑥⑥⑥ · · · / 2 Furthermore, S(Q) ∼= k and S(Q′) ∼= M2(k). / 2 / 2 2 / · · · * * : : j j / / & / / & / / & / / & 8 8 8 8 e e / / 9 9 9 o o / / @ / / @ / / @ / / > / / / / 26 S. PAUL SMITH Proof. The incidence matrices for Q and Q′ are so C =(cid:18)1 0 1 0(cid:19) = C n and C ′ =(cid:18)1 1 0 0(cid:19) = (C ′)n kQn ∼= E11 ⊕ E21 and kQ′ n ∼= E11 ⊕ E12. The result follows easily from this. (cid:3) Here is another way to show QGr(kQ) ≡ Modk for the Q in Proposition 6.3. Write x for the loop at vertex 1 and w for the arrow from 1 to 2. The path algebra is The two-sided ideal generated by e2 is 0 kQ =(cid:18) k[x] T =(cid:18) 0 wk[x] k(cid:19) . wk[x] k(cid:19) , 0 which is the ideal T in Lemma 2.1, T is annihilated on the left by so, as a left kQ-module, T is a sum of finite dimensional modules. Therefore 0 (cid:18) k[x] wk[x] 0(cid:19) QGr(kQ) ≡ QGr(cid:16) kQ T (cid:17) ≡ QGr(k[x]) ≡ Qcoh(Proj k[x]) ≡ Modk. 6.8. Let K(H) be the C∗-algebra of compact operators on an infinite di- mensional separable Hilbert space H. The direct limit in the category of C∗-algebras of the directed system with Bratteli diagram (6-2) is isomor- phic to K(H) ⊕ C idH. The algebra S(Q) in Proposition 6.4 is the algebraic analogue of K(H) ⊕ C idH. Proposition 6.4. Let Q = x w 1 2 y The Bratteli diagram for S(Q) is (6-2) and 1 1 @ / 2 1 @ / 3 1 @ / 4 1 @ / 5 1 >⑥⑥⑥⑥⑥⑥⑥⑥ / · · · · · · S(Q) ∼= M∞(k) ⊕ k.I where M∞(k) is the algebra without unit consisting of N × N matrices with only finitely many non-zero entries and M∞(k) ⊕ k.I is the algebra of N × N matrices that differ from a scalar multiple of the N × N identity matrix in only finitely many places. # # o o { { / / / / / / / @ / / @ / / @ / / @ / / > CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 27 Proof. One can see directly that kQ ∼=(cid:18)k[x] k[x] ⊗ w ⊗ k[y] k[y] 0 (cid:19) . As a kI-bimodule, kQ1 = kx ⊕ kw ⊕ ky ∼= E11 ⊕ E12 ⊕ E22. The nth power of the incidence matrix for Q is so, as a kI-bimodule, kQn ∼= E11 ⊕ E⊕n 12 ⊕ E22 and as a left kI-module (cid:18)1 n 0 1(cid:19) kQn ∼= E⊕(n+1) 1 ⊕ E2. Therefore EndkI(kQn) ∼= Mn+1(k) ⊕ k. In order to give an explicit description of the homomorphisms θn : EndkI(kQn) → EndkI (kQn+1) we take the ordered basis for kQn consisting of the n + 2 elements xn, xn−1w, xn−2wy, . . . , xwyn−2, wyn−1, yn. The linear span of xn is a kI-bimodule isomorphic to E11. The linear span of the next n elements, those with a w in them, is a kI-bimodule isomorphic to E⊕n 12 . The linear span of yn is a kI-bimodule isomorphic to E22. We will write an element of f ∈ Sn = EndkI(kQn) as f = (A, λ) ∈ Mn+1(k) ⊕ k where A represents the restriction of f to E⊕(n+1) basis, and f (yn) = λyn. 1 with respect to the ordered The homomorphism θn : Sn → Sn+1 is defined in section 1.2. In this example, kQn+1 = kQ1 ⊗kI kQn =(cid:0)E11 ⊗k En+1 1 = x ⊗ span{xn, xn−1w, xn−2wy, . . . , xwyn−2, wyn−1} (cid:1) ⊕(cid:0)E12 ⊗k E2(cid:1) ⊕(cid:0)E22 ⊗k E2(cid:1) ⊕ (kw ⊗ kyn) ⊕ (ky ⊗ kyn). Therefore θn(A, λ) = (A + λen+2,n+2, λ) = (cid:18)A 0 0 λ(cid:19) , λ!. Define φn : Sn → M∞(k) ⊕ kI by φn(A, λ) := A + λIn+1 where In = I − (e11 + · · · + enn) ∈ M∞(k) ⊕ kI. Since AIn+1 = In+1A = 0 and I 2 n+1 = 0, φn is a homomorphism of k-algebras sending the identity to the identity. It is straightforward to check that φn+1θn = φn. It follows that 28 S. PAUL SMITH all the φns factor through a single homomorphism φ : S(Q) = lim −→ M∞(k) ⊕ kI. We leave the reader to check that φ is an isomorphism. Sn → (cid:3) 7. Relation to finite dimensional algebras with radical square zero 7.1. The work of Xiao-Wu Chen [5]. The singularity category of a left coherent ring R, denoted Dsg(R), is the quotient of the derived category Db(modR) of bounded complexes of finitely presented left R-modules by its full subcategory of perfect complexes. The following is a simplified version of the main result in [5]. Theorem 7.1 (X.-W. Chen). [5] Let Λ be a finite dimensional k-algebra and J its Jacobson radical. Suppose J 2 = 0. Viewing J as a left Λ-module, define and the S(Λ)-bimodule S(Λ) := lim −→ EndΛ(J ⊗n) where the maps in the directed systems are f 7→ idJ ⊗f . Then B := lim −→ HomΛ(J ⊗n, J ⊗n−1) • B is an invertible S(Λ)-bimodule with inverse lim −→ • S(Λ) is a von Neumann regular ring; • J is a progenerator in Dsg(Λ) with endomorphism ring S(Λ); • HomDsg(Λ)(J, −) is an equivalence of triangulated categories HomΛ(J ⊗n, J ⊗n+1); (cid:0)Dsg(Λ), [1](cid:1) ≡(cid:0) proj S(Λ), − ⊗S(Λ) B(cid:1) where proj S(Λ) is the category of finitely generated projective right S(Λ)-modules, and −⊗S(Λ) B is the translation functor on proj S(Λ). If the field k in Chen's theorem is algebraically closed then Λ is Morita equivalent to kQ/kQ≥2 for a suitable quiver Q. Theorem 7.2. Let k be a field, Q a quiver, and Λ = kQ/kQ≥2. The rings S(Λ) and S(Q) are isomorphic and there is an equivalence of triangulated categories Proof. The Jacobson radical of Λ is J = kQ≥1/kQ≥2. We identify J with kQ1 . This identification is compatible with the kI-bimodule structures. (cid:0)Dsg(Λ), [1](cid:1) ≡(cid:0) qgr(kQ), (−1)(cid:1). Since J 2 = 0, J ⊗Λ · · · ⊗Λ J =J ⊗Λ/J · · · ⊗Λ/J J =(kQ1) ⊗kI · · · ⊗kI (kQ1) so EndΛ(J ⊗n) = EndkI(kQn); i.e., the individual terms in the the directed systems defining S(Λ) and S(Q) are the same. But the maps in the directed systems are of the form f 7→ id ⊗f in both cases so the direct limits S(Λ) and S(Q) are isomorphic. CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 29 Since S(Λ) is von Neumann regular proj S(Λ) is equal to modrS(Λ), the category of finitely presented right S(Λ)-modules. The translation functor on proj S(Λ) is M 7→ M ⊗S(Λ) B. Hence Chen's Theorem says that (cid:0)Dsg(Λ), [1](cid:1) ≡(cid:0)modrS(Q), − ⊗S(Q) B(cid:1). We write Σ for the translation functor − ⊗S(Q) B. An auto-equivalence of an abelian category having a generator is deter- mined its effect on the generator. The generator O for qgr(kQ) corresponds to the generator S(Q) under the equivalence HomQGr(O, −) : qgr(kQ) ≡ modS(Q); since Σ(S(Q)) = B, the auto-equivalence of qgr(kQ) that corre- sponds to Σ is the unique auto-equivalence σ such that HomQGr(O, σ(O)) = B. The calculation in the next paragraph shows that σ(O) = O(−1), so the auto-equivalence of qgr(kQ) that that corresponds to Σ is F 7→ F(−1). The equivalence qgr(kQ) → modrS(Q) sends O(−1) to HomQGr(O, O(−1)) = lim −→ = lim −→ = lim −→ = lim −→ = lim −→ = B. HomGr(kQ)(kQ≥n, kQ(−1)) HomGr(kQ)(kQ≥n, kQ(−1)≥n) HomkI(kQn, kQ(−1)n) HomkI(kQn, kQn−1) HomkI(J ⊗n, J ⊗(n−1)) This completes the proof that (modS(Λ), − ⊗S(Λ) B) ≡ (qgr(kQ), (−1)). (cid:3) 7.2. An example. Fix an integer r ≥ 1, let (7-1) Q = 09 / 1 / 2 / · · · · · · / r and define Λr := kQ/kQ≥2. The algebra khx, yi/(yr+1) in the next result is studied in [26]. When r = 1 it behaves as a non-commutative homogeneous coordinate ring for the space of Penrose tilings of the plane. Thus the equivalence of categories in the next result says that the path algebra of the quiver in (7-1) for r = 1 is also a homogeneous coordinate ring for the space of Penrose tilings. Proposition 7.3. The following categories are equivalent: (yr+1)!. Dsg(Λr) ≡ qgr(kQ) ≡ qgr khx, yi 9 / i i / f f / / d d 30 S. PAUL SMITH Proof. The incidence matrix for Q is the (r + 1) × (r + 1) matrix (7-2) C =  .   1 0 0 ... 1 0 1 1 1 · · · 1 0 0 · · · 0 1 0 · · · ... 0 0 0 · · ·  kQ1 ∼= E10 ⊕ E21 ⊕ · · · ⊕ Er r−1 ⊕(cid:18) r Mi=0 E0i(cid:19). It follows that kQ1 ∼= Er+1 kI-bimodule, 0 ⊕ E1 ⊕ · · · ⊕ Er as a left kI-module and, as a Thus, the dimension vector for kQ1 as a left kI-module is (r + 1, 1, . . . , 1)T and the dimension vector for kQn as a left kI-module is C n−1(r+1, 1, . . . , 1)T. Define d0 = r + 1, d1 = d2 = · · · = dr = 1, and dn+1 = dn + · · · + dn−r for n ≥ r. The dimension vector for kQn+1 as a left kI-module is therefore (dn+1, dn, . . . , dn−r+1) and the Bratteli diagram for S(Q), written from top to bottom, is repeated copies of Sn−1 θn Sn dn dn−1 dn−2 ❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞❞ ❋❋❋❋❋❋❋❋❋ ❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦❦ ❑❑❑❑❑❑❑❑❑❑ ❋❋❋❋❋❋❋❋ ①①①①①①①① ❉❉❉❉❉❉❉❉❉ dn−r+1 · · · · · · dn+1 dn dn−1 dn−r dn−r+1. In [26] it was show that (7-3) QGr khx, yi (yr+1) ≡ ModRr where Rr is the ultramatricial algebra associated to a Bratteli diagram that has the same underlying (unlabelled) graph as that for S(Q). By Proposition 7.4 below, Rr and S(Q) are Morita equivalent. (cid:3) We illustrate the remark in the last paragraph of the previous proof. The Bratteli diagram for S(Λ2), written from left to right, is 1 2 2 ✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿ rrrrrrrrrrrr ▲▲▲▲▲▲▲▲▲▲▲▲ rrrrrrrrrrrr 2 2 5 ✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿ rrrrrrrrrrrr ▲▲▲▲▲▲▲▲▲▲▲▲ rrrrrrrrrrrr 2 5 9 ✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿ rrrrrrrrrrrr ▲▲▲▲▲▲▲▲▲▲▲▲ rrrrrrrrrrrr 5 9 16 ✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿✿ rrrrrrrrrrrr ▲▲▲▲▲▲▲▲▲▲▲▲ rrrrrrrrrrr 9 16 30 · · · · · · / · · · S1(Λ2) / S2(Λ2) / S3(Λ2) / S4(Λ2) / S5(Λ2) / / / / / / / CATEGORY EQUIVALENCES INVOLVING PATH ALGEBRAS OF QUIVERS 31 The Bratteli diagram for the ring R2 in (7-3), written from left to right, is 0 0 1 ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ❃❃❃❃❃❃❃❃ 0 1 1 ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ❃❃❃❃❃❃❃❃ 1 1 2 ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ❃❃❃❃❃❃❃❃ 1 2 4 ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ✳ ❃❃❃❃❃❃❃❃ 2 4 7 4 7 ⑦⑦⑦⑦⑦⑦⑦⑦ ❅❅❅❅❅❅❅ ⑦⑦⑦⑦⑦⑦⑦⑦ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ ✵ 13 · · · · · · · · · 7.3. The proof of the next result was shown to me by Ken Goodearl and I thank him for allowing me to include it. Although the result is implicit in [1], [11], and [30], Goodearl's proof is simple and direct. Proposition 7.4. Suppose A and B are ultramatricial k-algebras formed from Bratteli diagrams on the same underlying (i.e., unlabelled) directed graph. Then A is Morita equivalent to B. (Goodearl) The directed graph determines a directed system of Proof. free abelian groups whose direct limit as an ordered group is isomorphic to K0(A) and to K0(B). Since K0(A) and K0(B) are isomorphic as ordered groups Elliott's results show that A and B are Morita equivalent. To see this directly, choose an ordered group isomorphism f : K0(A) → K0(B), and let P be a finitely generated projective right A-module such that [P ] = f −1([B]). Since [B] is an order-unit in K0(B), [P ] is an order- unit in K0(A). This implies that P is a generator in ModA. The category equivalence given by − ⊗C P , where C = EndA P , takes the category of finitely generated projective right C-modules to the the category of finitely generated projective right A-modules with C mapping to P . The composi- tion K0(C) −⊗C P / K0(A) f / K0(B) is an isomorphism of ordered abelian groups that sends [C] to [B], i.e., it is an isomorphism of ordered abelian groups with order unit. Elliott's classification theorem therefore implies C ∼= B. But C and A are Morita equivalent via P . (cid:3) References [1] G. Abrams and M. Tomforde, Isomorphism and Morita equivalence of graph algebras, Trans. Amer. Math. Soc., 363 (2011) 3733-3767. [2] P. Ara, M.A. Gonz´alez-Barroso, K.R. Goodearl, and E. Pardo, Fractional skew monoid rings, J. Algebra, 278 (2004), no. 1, 104-126. [3] P. Ara and M. Brustenga, Module theory over Leavitt path algebras and K-theory, J. Pure and App. Alg., 214 (2010) 1131-1151. [4] O. Bratteli, Inductive limits of finite dimensional C∗-algebras, Trans. Amer. Math. Soc., 171 (1979) 195-234. [5] X.-W. Chen, The singularity category of an algebra with radical square zero, Docu- menta Mathematica, 16 (2011) 921-936. [6] X.-W. Chen, Irreducible representations of Leavitt path algebras, arXiv: 1108.3726v1. / / 32 S. PAUL SMITH [7] P. M. Cohn, Free rings and their relations, 2nd ed., Academic Press, London, 1985. [8] A. Connes, Noncommutative Geometry, Academic Press, San Diego, 1994. [9] E. C. Dade, Group Graded Rings and Modules, Math. Zeit., 174 (1980) 241-262. [10] K.R. Davidson, C∗-algebras by Example, Fields Institute Monographs, Amer. Math. Soc., Rhode Island, 1996. [11] D. Drinen, Viewing AF-algebras as graph algebras, Proc. Amer. Math. Soc., 128 (2000) 1991-2000. [12] E.G. Effros, Dimensions and C∗-algebras, CBMS Regional Conf. Ser. in Math., No. 46, Amer. Math. Soc., Providence, R. I., 1981. [13] G.A. Elliott, On the classification of inductive limits of sequences of semisimple finite- dimensional algebras, J. Algebra, 38 (1976) 29-44. [14] P. Gabriel, Des Cat´egories Ab´eliennes, Bull. Soc. Math. Fr., 90 (1962) 323-448. [15] P. Gabriel and J.-A. de la Pena, Quotients of representation-finite algebras, Comm. in Alg., 15 (1987) 279-307. [16] W. Geigle and H. Lenzing, Perpendicular Categories with Applications to Represen- tations and Sheaves, J. Alg., 144 (1991) 273-343. [17] K.R. Goodearl, Leavitt path algebras and direct limits, in Rings, modules and rep- resentations, pp. 165-187, Contemp. Math., 480, Amer. Math. Soc., Providence, RI, 2009. [18] K.R. Goodearl, Von Neumann Regular Rings, Monographs and Studies in Mathe- matics, No. 4, Pitman, London-San Francisco-Melbourne, 1979. [19] R. Hazrat, The graded structure of Leavitt Path algebras, arXiv: 1005.1900. [20] C. Holdaway and S. P. Smith, An equivalence of categories for graded modules over monomial algebras and path algebras of quivers, Journal of Algebra 353 (2011) 249- 260. [21] J.H. Hong and W. Szyma´nski, Quantum spheres and projective spaces as graph al- gebras, Comm. Math. Phys., 232 (2002) 157-188. [22] H. Krause, The spectrum of a module category, Memoirs of the Amer. Math. Soc., No. 707, Vol. 149, 2001. [23] D. Lind and B. Marcus, An Introduction to Symbolic Dynamics and Coding, Camb. Univ. Press, Cambridge, 1995. [24] I. Raeburn, Graph algebras, CBMS Regional Conference Series in Mathematics, 103, Published for the Conference Board of the Mathematical Sciences, Washington, DC, by the Amer. Math. Soc., Providence, RI, 2005. [25] A.H. Schofield, Representations of Rings over Skew Fields, Lond. Math. Soc., Lecture Notes Series, Vol. 92, Cambridge University Press, Cambridge, UK, 1985. [26] S.P. Smith, The space of Penrose tilings and the non-commutative curve with homoge- neous coordinate ring khx, yi/(y2), J. Noncomm. Geom., to appear. arXiv:1104.3811 [27] S.P. Smith, The non-commutative scheme having a free algebra as a homogeneous coordinate ring, arXiv:1104.3822. [28] S.P. Smith, Shift equivalence and a category equivalence involving graded modules over path algebras of quivers, arXiv:1108.4994. [29] B. Stenstrom, Rings of Quotients, Die Grundlehren der mathematischen Wis- senschaften in Einzeldarstellungen, vol. 217, Springer Verlag, Berlin, 1975. [30] J. Tyler, Every AF-algebra is Morita equivalent to a graph algebra, Bull. Austral. Math. Soc., 69 (2004) 237-240. [31] A.B. Verevkin, On a non-commutative analogue of the category of coherent sheaves on a projective scheme, Amer. Math. Soc. Transl., (2) 151 (1992). Department of Mathematics, Box 354350, Univ. Washington, Seattle, WA 98195 E-mail address: [email protected]
1102.2493	5	1102	2011-09-14T20:15:34	Large affine spaces of non-singular matrices	[ "math.RA", "math.RT" ]	Let K be an arbitrary (commutative) field with at least three elements. It was recently proven that an affine subspace of M_n(K) consisting only of non-singular matrices must have a dimension lesser than or equal to n(n-1)/2. Here, we classify, up to equivalence, the subspaces whose dimension equals n(n-1)/2. This is done by classifying, up to similarity, all the n(n-1)/2-dimensional linear subspaces of M_n(K) consisting of matrices with no non-zero invariant vector, reinforcing a classical theorem of Gerstenhaber. Both classifications only involve the quadratic structure of the field K.	math.RA	math	Large affine spaces of non-singular matrices Cl´ement de Seguins Pazzis∗† July 31, 2018 Abstract Let K be an arbitrary (commutative) field with at least three elements. It was recently proven that an affine subspace of Mn(K) consisting only of Here, we classify, up to equivalence, the subspaces whose dimension equals non-singular matrices must have a dimension lesser than or equal to (cid:0)n 2(cid:1). (cid:0)n 2(cid:1). This is done by classifying, up to similarity, all the (cid:0)n 2(cid:1)-dimensional linear subspaces of Mn(K) consisting of matrices with no non-zero invariant vector, reinforcing a classical theorem of Gerstenhaber. Both classifications only involve the quadratic structure of the field K. AMS Classification : 15A03, 15A30 Keywords : affine subspaces, non-zero eigenvalues, alternate matrices, simulta- neous triangularization, non-isotropic quadratic forms, Gerstenhaber theorem 1 Introduction 1.1 Introduction and basic definitions In this article, we let K be an arbitrary (commutative) field. We denote by Mn(K) the algebra of square matrices with n rows and entries in K, and by GLn(K) its group of invertible elements. We also denote by Mn,p(K) the vector space of matrices with n rows, p columns and entries in K. The transpose of a matrix M is denoted by M T . ∗Lyc´ee Priv´e Sainte-Genevi`eve, 2, rue de l'´Ecole des Postes, 78029 Versailles Cedex, FRANCE. †e-mail address: [email protected] 1 1 1 0 2 p e S 4 1 ] . A R h t a m [ 5 v 3 9 4 2 . 2 0 1 1 : v i X r a An affine subspace V of Mn(K) is the translate of a linear subspace V of Mn(K): then V is uniquely determined by V (it is the set of all matrices M such that M + V = V) and is called the translation vector space of V. Given two linear (or affine) subspaces V and W of Mn(K), we say that V and W are equivalent, and we write V ∼ W , if W = P V Q for some (P, Q) ∈ GLn(K)2; we say that V and W are similar, and we write V ≃ W , if W = P V P −1 for some P ∈ GLn(K). Two matrices A and B of Mn(K) are called congruent, and we write A ≈ B, if A = P BP T for some P ∈ GLn(K). Finally, two quadratic forms q and q′ on vector spaces over K are called similar if q′ is equivalent to λ q for some λ ∈ K r {0}. Here, we are concerned with the geometry of GLn(K) ∪ {0} as a cone in the vector space Mn(K). From the linear algebraist's viewpoint, the natural questions that one may ask are the following ones: • What is the minimal linear (resp. affine) subspace of Mn(K) containing GLn(K)? • What is the minimal linear subspace of Mn(K) containing Mn(K)rGLn(K)? • What are the maximal linear (resp. affine) subspaces included in GLn(K)∪ {0}? • What are the maximal linear (resp. affine) subspaces included in Mn(K) r GLn(K)? The first two problems have easy answers: GLn(K) always spans Mn(K), the affine subspace it generates is Mn(K) unless n = 1 and # K = 2, and Mn(K) r GLn(K) spans Mn(K) unless n = 1. The last two questions have no clear answer however and depend widely on the field K. For example, GLn(C) ∪ {0} contains no 2-dimensional linear subspace, whilst GL2n(R) ∪ {0} always does. As for singular linear subspaces (i.e. linear subspaces included in Mn(K) r GLn(K)), a classification of them is generally considered to be out of reach, even for an algebraically closed field, although a lot of progress has been made in understanding their structure in the 1980's (see the works of Atkinson, Lloyd and Stephens [1, 2, 3, 4] and our recent [12]). Rather than try to classify all the linear (resp. affine) subspaces contained in GLn(K) or Mn(K) r GLn(K), a more modest approach is to find the maximal dimension for such a subspace and to classify the linear (resp. affine) subspaces 2 with a maximal dimension. To this day, this problem has been almost entirely solved: • A linear subspace included in GLn(K)∪{0} has dimension at most n; linear subspaces in GLn(K) ∪ {0} with dimension n correspond to the structures of (possibly non-associative and non-unital) division algebras on Kn that are compatible with its vector space structure (see e.g. the last section of [13]). Note that no such subspace exists when n ≥ 2 and K is algebraically closed. • An affine subspace included in Mn(K) r GLn(K) has dimension at most n(n − 1). If its dimension is n(n − 1), then it is equivalent to the space of matrices with zero as last column or to its transpose (unless n = 2 and # K = 2 in which case there is an additional equivalence class). This is a classical result of Dieudonn´e [5] (see also [11] for a simplified proof) which may be used to classify the endomorphisms of the vector space Mn(K) that stabilize GLn(K) (see [13]). Here, we will focus on the affine subspaces of Mn(K) that are included in GLn(K). Let V be such a subspace, and choose P ∈ V. Then P −1V is also included in GLn(K), contains the identity matrix In and has the same dimension as V. Denoting by H its translation vector space, we see that In −λ M ∈ GLn(K) for every λ ∈ K and M ∈ H, hence the linear subspace H has the two following equivalent properties: (i) For every M ∈ H, one has Sp(M ) ⊂ {0}, where Sp(M ) denotes the set of eigenvalues of M in the field K. (ii) No matrix of H possesses a non-zero invariant vector in Kn. Definition 1. A linear subspace H of Mn(K) is said to have a trivial spectrum if no matrix of H possesses a non-zero invariant vector in Kn. Note that for such a linear subspace H with a trivial spectrum, the affine subspace In + H is included in GLn(K), and so is any subspace equivalent to it. For example, if we denote by NTn(K) the space of strictly upper triangular matrices of Mn(K), then In + NTn(K) is an affine subspace of non-singular 2(cid:1). matrices with dimension(cid:0)n It follows that classifying up to equivalence the affine subspaces of non- singular matrices essentially amounts to classifying up to similarity the linear 3 subspaces of Mn(K) with a trivial spectrum. In the case K is algebraically closed, the linear subspaces with a trivial spectrum are the linear subspaces of nilpotent matrices: a famous theorem of Gerstenhaber [6] states that the dimension of such a subspace is bounded above by(cid:0)n spaces similar to NTn(K). It is only very recently that the upper bound(cid:0)n 2(cid:1) and that equality occurs only for sub- 2(cid:1) has been shown to apply to linear subspaces with a trivial spectrum for an arbitrary field (see the works of Quinlan [8] and our own [10]): Theorem 1. Let V be a linear subspace of Mn(K) with a trivial spectrum. Definition 2. A linear subspace of Mn(K) with a trivial spectrum is called Then dim V ≤(cid:0)n 2(cid:1). maximal1 if its dimension is(cid:0)n 2(cid:1). Our aim is to classify the maximal linear subspaces of Mn(K) with a trivial spectrum. Unlike the case of nilpotent linear subspaces, the structure of the ground field K plays a large part in this classification. For example, if there exists a polynomial t2 − at − b ∈ K[t] with degree two and no root in K, then the line spanned by the companion matrix(cid:20)0 b which has a trivial spectrum and dimension (cid:0)n 1 a(cid:21) is obviously a maximal linear 2(cid:1), although it is not similar to subspace of M2(K) with a trivial spectrum and it is not similar to NT2(K). Another example is given by the space An(R) of skew-symmetric real matrices, NTn(R) if n ≥ 2. 1.2 Reducibility Notation 3. Let V and W be respective subsets of Mn(K) and Mp(K). Set V ∨ W :=(cid:26)(cid:20)A B 0 C(cid:21) (A, B, C) ∈ V × Mn,p(K) × W(cid:27) ⊂ Mn+p(K). then V ∨ W is a linear subspace with a trivial spectrum and dimension (cid:0)n 2(cid:1) + 2 (cid:1), hence it is maximal. Notice also that the composition law ∨ (cid:0)p 2(cid:1) + pn =(cid:0)n+p Note that if V and W are maximal linear subspaces with a trivial spectrum, is associative. 1This should not be confused with the concept of maximality in the set of linear subspaces with a trivial spectrum ordered by the inclusion of subsets. 4 Definition 4. A maximal linear subspace of Mn(K) with a trivial spectrum is called irreducible if the only linear subspaces of Kn it stabilizes are {0} and Kn (and we call it reducible otherwise). Conversely, let H be a maximal linear subspace of Mn(K) with a trivial spectrum. Assume that there is a p ∈ [[1, n − 1]] such that F := Kp × {0} is stabilized by every matrix of H. Then we may write every matrix of H as M =(cid:20)f (M ) g(M ) h(M )(cid:21) 0 for some (f (M ), g(M ), h(M )) ∈ Mp(K) × Mp,n−p(K) × Mn−p(K). Therefore V := f (H) and W := h(H) are linear subspaces respectively of Mp(K) and Mn−p(K), each with a trivial spectrum, and since (cid:18)n 2(cid:19) = dim H ≤ dim V +dim W +dim g(H) ≤(cid:18)p 2(cid:19)+(cid:18)n − p 2 (cid:19)+p(n−p) =(cid:18)n 2(cid:19), we find that both V and W are maximal. Hence H ⊂ V ∨ W , and since the dimensions are equal, we deduce that H = V ∨ W . Conjugating H with an appropriate invertible matrix, this generalizes as if H is not irreducible, then H ≃ V ∨ W for some maximal linear follows: subspaces V and W with trivial spectra. This yields: Proposition 2. Let H be a maximal linear subspace of Mn(K) with a trivial spectrum. Then there are irreducible maximal linear subspaces V1, . . . , Vp with trivial spectra such that H ≃ V1 ∨ V2 ∨ · · · ∨ Vp. This suggests that we focus our attention on the irreducible maximal subspaces. 1.3 Main theorems Denote by An(K) the set of alternate matrices of Mn(K), i.e. the skew-symmetric ones with a zero diagonal, i.e. the ones for which ∀X ∈ Kn, X T AX = 0. Definition 5. A matrix P ∈ Mn(K) is called non-isotropic if the quadratic form X 7→ X T P X is non-isotropic, i.e. ∀X ∈ Kn r {0}, X T P X 6= 0. Notice, in that case, that P is non-singular and that P −1 is non-isotropic. The subspace P An(K) then has dimension(cid:0)n deed, given A ∈ An(K) and X ∈ Kn, 2(cid:1) and has a trivial spectrum: in- P AX = X ⇒ P −1X = AX ⇒ X T P −1X = 0 ⇒ X = 0. We may now state our main results. 5 Theorem 3. Assume that #K ≥ 3. Let n be a positive integer. Then the irreducible maximal linear subspaces of Mn(K) with a trivial spectrum are the subspaces of the form P An(K) for a non-isotropic matrix P ∈ GLn(K). Theorem 4 (Classification theorem for maximal linear subspaces with a trivial spectrum). Assume that #K ≥ 3. Let V be a maximal linear subspace of Mn(K) with a trivial spectrum. Then there is a list (P1, . . . , Pp) ∈ GLn1(K) × · · · × GLnp(K) of non-isotropic matrices such that V ≃ P1 An1(K) ∨ · · · ∨ Pp Anp(K). The integer p is uniquely determined by V and, for every k ∈ [[1, p]], the matrix Pk is uniquely determined by V up to congruence and multiplication by a non-zero scalar. Moreover, given another list (Q1, . . . , Qp) ∈ GLn1(K) × · · · × GLnp(K), if Qk is congruent to a scalar multiple of Pk for each k ∈ [[1, p]], then V ≃ Q1 An1(K) ∨ · · · ∨ Qp Anp(K). If K is quadratically closed, it follows that there is no irreducible maximal linear subspace of Mn(K) with a trivial spectrum for n ≥ 2. If K is finite (with at least three elements), then every 3-dimensional quadratic form over K is isotropic, hence Mn(K) contains no irreducible maximal linear subspace with a trivial spectrum for n ≥ 3. We deduce the following corollaries: Corollary 5. Let K be a quadratically closed field. Then NTn(K) is, up to similarity, the sole maximal linear subspace of Mn(K) with a trivial spectrum. Corollary 6. Let K be a finite field with at least three elements. Let V be a maximal linear subspace of Mn(K) with a trivial spectrum. Then there are matrices M1, . . . , Mp, either equal to 0 ∈ M1(K) or belonging to M2(K) with no eigenvalue in K, such that V ≃ K M1 ∨ · · · ∨ K Mp. Each Mk is then uniquely determined by V up to similarity and multiplication by a non-zero scalar. We may finally state the structure theorem for affine subspaces of non- singular matrices. 6 Theorem 7 (Classification theorem for large affine subspaces of non-singular matrices). Assume that #K ≥ 3. Let V be a(cid:0)n Mn(K) included in GLn(K). Then there is a list (P1, . . . , Pp) ∈ GLn1(K) × · · · × GLnp(K) of non-isotropic matrices such that n = n1 + · · · + np and 2(cid:1)-dimensional affine subspace of V ∼ In +(cid:0)P1 An1(K) ∨ · · · ∨ Pp Anp(K)(cid:1). The integer p is uniquely determined by V and, for 1 ≤ k ≤ p, the similarity class of the non-isotropic quadratic form X 7→ X T PkX is uniquely determined by V. Moreover, given another list (Q1, . . . , Qp) ∈ GLn1(K) × · · · × GLnp(K), if X 7→ X T QkX is similar to X 7→ X T PkX for each k ∈ [[1, p]], then V ∼ In +(cid:0)Q1 An1(K) ∨ · · · ∨ Qp Anp(K)(cid:1). Note that the existence of (P1, . . . , Pp) is a trivial consequence of Theorem 4 using the considerations of Paragraph 1.1. As a consequence, (cid:0)n GLn(K) are classified, up to equivalence, by the lists of the form ([ϕ1], . . . , [ϕp]) where the ϕk's are finite-dimensional non-isotropic quadratic forms over K, the 2(cid:1)-dimensional affine subspaces of Mn(K) included in [ϕk]'s are their similarity classes, and dim ϕk = n. For the field of real numbers, this has the following striking corollary: pPk=1 Corollary 8. Let V be an affine subspace of Mn(R) included in GLn(R) with 2(cid:1). Then there is a unique list (n1, . . . , np) of positive integers such dimension(cid:0)n that n = n1 + · · · + np and V ∼ In +(cid:0)An1(R) ∨ · · · ∨ Anp(R)(cid:1). 1.4 Totally intransitive action of a space of matrices Proving the previous theorems will require an extensive use of the following concept and of the subsequent remark: Definition 6. Let V be a linear subspace of Mn(K). For X ∈ Kn, set V X :=(cid:8)M X X ∈ V(cid:9). Note that V X is always a linear subspace of Kn. We say that V acts totally intransitively on Kn if V X 6= Kn for every X ∈ Kn, which is equivalent to having dim(V X) < n for every X ∈ Kn. 7 Remark 1. If V has a trivial spectrum, then X 6∈ V X for every X ∈ Kn r {0}, hence V acts totally intransitively on Kn. Moreover V T :=(cid:8)M T M ∈ V(cid:9) also has a trivial spectrum, hence ∀X ∈ Kn, dim(V X) < n and dim(V T X) < n. 1.5 Structure of the paper We will start (Section 2) with general considerations on the spaces of the type P An(K) with P ∈ GLn(K). Using some of the obtained results, we will then prove the uniqueness statements in Theorems 4 and 7 (Section 3). The proof of Theorem 3 will be carried out in Section 4 by induction on n, starting from n = 2 and using a recent lemma that was proved in [10]: this is, by far, the most technical part of the paper. In Section 5, we will easily derive Gerstenhaber's theorem from Theorem 4 in the case #K ≥ 3. In Section 6, we will show that Theorem 3 fails for n = 3 and K ≃ F2. The case #K = 2 remains a very exciting challenge that we will not undertake here. 2 Basic properties of the spaces P An(K) We consider first P An(K) for an arbitrary P ∈ GLn(K). To start with, note that, for every Q ∈ GLn(K), one has P An(K) Q = P (QT )−1QT An(K) Q =(cid:0)P (QT )−1(cid:1) An(K) which immediately shows that {P An(K) P ∈ GLn(K)} is an equivalence class (for the equivalence of spaces of matrices). In order to move forward, we need some basic properties of An(K): for this, we equip Kn with the non-degenerate symmetric bilinear form (X, Y ) 7→ X T Y . Lemma 9. For any X ∈ Kn r {0}, one has An(K)X = {X}⊥ and in particular dim(An(K)X) = n − 1. Proof. This is obvious if X is the first vector e1 of the canonical basis of Kn. In the general case, we may find some P ∈ GLn(K) such that P e1 = X, and note 8 that An(K)X = (P T )−1P T An(K)P e1 = (P T )−1 An(K)e1 = (P T )−1{e1}⊥ = {P e1}⊥ = {X}⊥. We may now determine, amongst the spaces of the above form, those with a trivial spectrum: Lemma 10. Let P ∈ GLn(K). Then P An(K) has a trivial spectrum if and only if P is non-isotropic. Proof. The "if" part has already been dealt with in the beginning of Section 1.3. Assume that P is isotropic. Then obviously (P T )−1 is also isotropic, hence we find a non-zero vector X ∈ Kn such that X T (P T )−1X = 0, i.e. P −1X ∈ {X}⊥. Then Lemma 9 shows that P −1X = AX for some A ∈ An(K) hence (P A)X = X, which shows that P An(K) does not have a trivial spectrum. Proposition 11. Let P ∈ GLn(K) be a non-isotropic matrix. Then P An(K) is an irreducible maximal subspace with a trivial spectrum. Proof. It only remains to show that P An(K) is irreducible. We use a reductio ad absurdum by assuming that it has a non-trivial stable subspace F ⊂ Kn with dimension p ∈ [[1, n − 1]]. Then F ⊥ is stabilized by (P An(K))T = An(K)P T . Choosing an arbitrary non-zero vector X ∈ F , we have dim(P An(K)X) = dim{X}⊥ = n − 1 hence p = n − 1. However, choosing a non-zero vector Y ∈ F ⊥ yields dim(An(K)P T Y ) = n − 1 hence n − p = n − 1. This yields n = 2 and p = 1, in which case every matrix of P An(K) must be nilpotent (since it has an eigenvector and 0 is the sole possible eigenvalue in K), contradicting the fact that every non-zero matrix of P A2(K) is non-singular. We now investigate when two spaces of the form P An(K) are similar. Here is our basic result: Lemma 12. Let P ∈ GLn(K). Then P An(K) = An(K) if and only if P is a scalar multiple of the identity. 9 Proof. The "if" part is trivial. Assume conversely that P An(K) = An(K). Let X ∈ Kn r {0}. Then P An(K)X = An(K)X yields that P stabilizes the hyperplane {X}⊥, hence P T stabilizes span(X). Since this holds for every non- zero X ∈ Kn, this shows that P T is a scalar multiple of the identity, hence P also is. The following corollary will be our starting point for the uniqueness state- ment in Theorem 4: Proposition 13. Let (P, Q) ∈ GLn(K)2. Then P An(K) ≃ Q An(K) if and only if P ≈ λQ for some λ ∈ K r {0}. Proof. If P = λ RQRT for some R ∈ GLn(K) and some λ ∈ K r {0}, then P An(K) = RQRT An(K) = R(QRT An(K)R)R−1 = R(Q An(K))R−1. Conversely, assume that P An(K) = R(Q An(K))R−1 for some R ∈ GLn(K). Then the above computation yields (RQRT )−1P An(K) = An(K) hence Lemma 12 yields a non-zero scalar λ such that (RQRT )−1P = λIn. Therefore P = R(λ Q)RT . Remark 2 (A crucial remark). Let E be a finite dimensional vector space and b a (possibly non-symmetric) bilinear form on E such that ∀x ∈ Er{0}, b(x, x) 6= 0. Given a non-zero vector x ∈ E, the hyperplane H := {y ∈ E : b(x, y) = 0} is then a complementary subspace of span(x) in E. By induction on the dimension of spaces, it follows that there exists a basis (f1, . . . , fn) of E which is right- orthogonal for b, i.e. b(fi, fj) = 0 for every (i, j) ∈ [[1, n]]2 satisfying i < j. For a non-isotropic matrix P ∈ GLn(K), this may be interpreted as follows: P is congruent to a lower-triangular matrix T , and hence P An(K) is similar to T An(K). This remark will play a major part in our proof of Theorem 3. Now, given non-isotropic matrices P and Q of GLn(K), we may examine when the two affine subspaces In + P An(K) and In + Q An(K) are equivalent. Proposition 14. Let P and Q be non-isotropic matrices of GLn(K). Then In + P An(K) ∼ In + Q An(K) if and only if the quadratic forms X 7→ X T P X and X 7→ X T QX are similar. Proof. • Assume first that In + P An(K) ∼ In + Q An(K), and choose a pair (R, S) ∈ GLn(K)2 such that R(In + P An(K)) = (In + Q An(K))S. Obvi- ously S belongs to (In + Q An(K))S, hence S = R(In + P A) for some A ∈ 10 An(K). By comparing the translation vector spaces of R(In+P An(K)) and (In+Q An(K))S, we also find that RP An(K) = Q An(K)S = Q(ST )−1 An(K). Therefore Proposition 13 yields a non-zero scalar λ such that RP = λ Q(ST )−1. It follows that ST = (In − AP T )RT and λ Q = RP ST = RP (In − AP T )RT = RP RT − (RP )A(RP )T . Since A is alternate, we find that λ X T QX = X T (RP RT )X = (RT X)T P (RT X) for every X ∈ Kn, and the quadratic forms X 7→ X T QX and X 7→ X T P X are similar because RT is non-singular. • Conversely, assume that X 7→ X T QX and X 7→ X T P X are similar. Then there is a non-singular matrix R ∈ GLn(K), a non-zero scalar λ and an alternate matrix A′ such that λQ = RP RT + A′. The matrix A := −(RP )−1A′((RP )T )−1 is congruent to −A′ and is therefore alternate. We set S := R(In + P A). Note that S = RP (P −1 + A) is non-singular: indeed, ∀X ∈ Kn r {0}, X T (P −1 + A)X = X T P −1X 6= 0 since P −1 is non-isotropic, hence P −1 + A is non-singular. However ST = (In − AP T )RT , therefore RP ST = RP RT − (RP )A(RP )T = RP RT + A′ = λ Q. We deduce that R(cid:0)P An(K)(cid:1) = λ Q(ST )−1 An(K) =(cid:0)Q An(K)(cid:1)S. We have just proven that the affine subspaces R(In + P An(K)) and (In + Q An(K))S have S as common point and have the same translation vector space, hence they are equal. This yields In + P An(K) ∼ In + Q An(K). Finally, the following lemma will be a major key to unlock our proof of Theorem 3: Lemma 15. Let n ≥ 3. Assume #K ≥ 3. Let V be a (cid:0)n subspace of Mn(K) which acts totally intransitively on Kn. Assume that there exists a linear hyperplane H of V such that H ⊂ An(K). Then V = An(K). 2(cid:1)-dimensional linear 11 Proof. Let A ∈ V . We prove that A is alternate, i.e. that the quadratic form q : X 7→ X T AX is zero. We denote by (e1, . . . , en) the canonical basis of Kn. Let X ∈ Kn r {0}. If dim(HX) = n − 1 then AX ∈ HX since HX ⊂ V X ( Kn, and hence q(X) = 0. If dim(HX) = n − 1 for every X ∈ Kn r {0}, then we readily have q = 0. Assume now that dim(HX1) < n − 1 for some X1 ∈ Kn r {0}. This shows that there exists X2 ∈ Kn r span(X1) such that X T 2 M X1 = 0 for every M ∈ H. Let X3 ∈ Kn r span(X1, X2). We may choose a non-singular matrix P ∈ GLn(K) such that P ei = Xi for every i ∈ [[1, 3]]. Then V ′ := P T V P acts totally intransitively on Kn and contains the hyperplane H ′ := P T HP ⊂ An(K). We now have eT 2 M e1 = 0 for every M ∈ H ′, hence H ′ is included in the space V1 of all alternate matrices A = (ai,j) of Mn(K) such that 2(cid:1)−1, and therefore H ′ = V1. a2,1 = 0. The dimension of this space is obviously(cid:0)n Then it is obvious that dim(H ′e3) = n − 1 and hence dim(HX3) = n − 1. We have therefore proven that ∀X ∈ Kn r span(X1, X2), q(X) = 0. It now suffices to show that q vanishes everywhere on span(X1, X2). Let X ∈ span(X1, X2) r {0}. We choose an arbitrary vector X3 ∈ Kn r span(X1, X2). The plane span(X, X3) satisfies span(X, X3) ∩ span(X1, X2) = span(X). Since # K > 2, this plane has at least four distinct 1-dimensional sub- spaces, three of which are different from span(X). We deduce that the quadratic form q vanishes on at least three 1-dimensional subspaces of span(X, X3). Clas- sically, this shows that q vanishes everywhere on span(X, X3) (indeed, a non-zero homogeneous polynomial of degree 2 on K2 has at most 2 zeroes in the projective line P(K2)). In particular q(X) = 0. We deduce that q = 0, which completes our proof. 3 The uniqueness statement in the two classification theorems The uniqueness statement in Theorem 4 is equivalent to the following result, which we prove right away: Proposition 16. Let (P1, . . . , Pp) and (Q1, . . . , Qq) be two families of non- isotropic matrices, respectively of GLn1(K) × · · · × GLnp(K) and GLm1 (K) × 12 · · · × GLmq (K). In order that P1 An1(K) ∨ · · · ∨ Pp Anp(K) ≃ Q1 Am1 (K) ∨ · · · ∨ Qq Amq (K), it is necessary and sufficient that q = p and Pk be congruent to a scalar multiple of Qk for every k ∈ [[1, p]]. Proof. The "sufficient condition" statement follows immediately from Proposi- tion 13. For the converse statement, set V := P1 An1(K) ∨ · · · ∨ Pp Anp(K) and W := Q1 Am1 (K) ∨ · · · ∨ Qq Amq (K). For k ∈ [[1, p]], set Fk := Kn1+···+nk × {0} ⊂ Kn , where n := n1 + · · · + np. Set also F0 = {0} and denote by (e1, . . . , en) the canonical basis of Kn. Set k ∈ [[1, p]]. Our key statement is the set of equalities: ∀X ∈ Fk r Fk−1, dim(V X) = n1 + · · · + nk − 1. Note first that the case X = en1+···+nk−1+1 follows trivially from Lemma 9. Consider now an arbitrary vector X ∈ FkrFk−1. Then e1, . . . , en1+···+nk−1 , X are linearly independent, and may therefore be completed as a basis (e1, . . . , en1+···+nk−1, X, f2, . . . , fnk ) of Fk. Therefore B := (e1, . . . , en1+···+nk−1, X, f2, . . . , fnk , en1+···+nk+1, . . . , en) is a basis of Kn and the matrix of coordinates R of B in the canonical basis of Kn belongs to GLn1(K) ∨ · · · ∨ GLnp(K) and satisfies Ren1+···+nk−1+1 = X. Propo- sition 13 thus yields a list of non-isotropic matrices (P ′ p) ∈ GLn1(K) × · · · × GLnp(K) for which 1, . . . , P ′ RV R−1 ⊂ P ′ 1 An1(K) ∨ · · · ∨ P ′ p Anp(K) and therefore RV R−1 = P ′ 1 An1(K) ∨ · · · ∨ P ′ p Anp(K) as the dimensions equal It follows that yields dim(V X) = dim(RV X) = dim(RV R−1)(RX) = n1 + · · · + nk − 1. 2(cid:1) on both sides. Applying the special case of en1+···+nk−1+1 to RV R−1 then (cid:0)n (cid:8)dim(V X) X ∈ Kn(cid:9) = {0, n1 − 1, n1 + n2 − 1, . . . , n1 + · · · + np − 1} has cardinality p+1. The same holds for W instead of V with the mj's in place of the nk's. Since V is similar to W , one has(cid:8)dim(V X) X ∈ Kn(cid:9) =(cid:8)dim(W X) 13 Now, set P ∈ GLn(K) such that W = P −1V P . For every k ∈ [[1, p]], remark that X ∈ Kn(cid:9) and we deduce successively that q = p and (n1, . . . , np) = (m1, . . . , mq). (cid:8)X ∈ Kn : dim V X ≤ n1+· · ·+nk−1(cid:9) = Fk =(cid:8)X ∈ Kn : dim W X ≤ n1+· · ·+nk−1(cid:9), hence P stabilizes Fk. This shows that P ∈ GLn1(K) ∨ · · · ∨ GLnp(K), which in turn proves that Pk Ank (K) is similar to Qk Ank (K) for every k ∈ [[1, p]]. Proposition 13 finally yields that Pk is congruent to a scalar multiple of Qk, for every k ∈ [[1, p]]. Proposition 17. Let (P1, . . . , Pp) and (Q1, . . . , Qq) be two families of non- isotropic matrices, respectively in GLn1(K) × · · · × GLnp(K) and GLm1(K) × · · · × GLmq (K). In order that (cid:0)In1+P1 An1(K)(cid:1)∨· · ·∨(cid:0)Inp+Pp Anp(K)(cid:1) ∼(cid:0)Im1+Q1 Am1(K)(cid:1)∨· · ·∨(cid:0)Imq +Qq Amq (K)(cid:1), it is necessary and sufficient that q = p and that the (non-isotropic) quadratic form X 7→ X T PkX be similar to X 7→ X T QkX for every k ∈ [[1, p]]. Proof. The "sufficient condition" statement follows trivially from Proposition 14. For the converse statement, let us set V := (In1 + P1 An1(K)) ∨ · · · ∨ (Inp + Pp Anp(K)) and W := (Im1 + Q1 Am1(K)) ∨ · · · ∨ (Imq + Qq Amq (K)), and assume that V ∼ W. Choose two non-singular matrices R and S such that W = RVS. Denote by V (resp. by W ) the translation vector space of V (resp. of W), and set n := nk. Then pPk=1 W = (RS)S−1(In + V )S = (RS)(In + S−1V S). In particular RS ∈ W and the comparison of translation vector spaces yields S−1V S = (RS)−1W . The first result yields that RS is upper block-triangular with diagonal blocks R1, . . . , Rq where Rk ∈ GLmk (K) for every k ∈ [[1, q]]. Thus S−1V S = (RS)−1W = (R−1 1 Q1) Am1(K) ∨ · · · ∨ (R−1 q Qq) Amq (K) and the R−1 k Qk's are necessarily non-isotropic since S−1V S has a trivial spec- trum. We deduce from Proposition 16 that (n1, . . . , np) = (m1, . . . , mq). With the line of reasoning from the proof of Proposition 16, we also find that S ∈ 14 GLn1(K)∨· · ·∨GLnp(K). However we already know that RS belongs to GLn1(K)∨ · · · ∨ GLnp(K) and hence R = (RS)S−1 ∈ GLn1(K) ∨ · · · ∨ GLnp(K). Returning to RVS = W finally entails that Ink + Qk Ank (K) is equivalent to Ink + Pk Ank (K) for each k ∈ [[1, p]], and Proposition 14 then yields that X 7→ X T PkX is similar to X 7→ X T QkX for each k ∈ [[1, p]]. 4 Structure of the irreducible maximal spaces with a trivial spectrum In the whole section, we assume #K ≥ 3. We will prove Theorem 3 by induction. The case n = 1 needs no explanation. 4.1 The case n = 2 Let V be an irreducible maximal linear subspace of M2(K) with a trivial spec- trum. Then V = span(M ) for some M ∈ M2(K) r {0} with no non-zero eigen- value. If 0 is an eigenvalue of M , then M is triangularizable and V is not irreducible. 1 Hence M is non-singular. Setting K :=(cid:20) 0 −1 0(cid:21) and P := M K −1, we readily have P A2(K) = span(M ) = V and Lemma 10 shows that P is non-isotropic. 4.2 Setting things up Let n ≥ 2 and assume that the result of Theorem 3 holds for any positive integer k ≤ n. Let V ⊂ Mn+1(K) be a maximal subspace with a trivial spectrum. Denote by (e1, . . . , en+1) the canonical basis of Kn+1. We wish to show that V is reducible or similar to P An+1(K) for some P ∈ GLn+1(K), in which case Lemma 10 guarantees that P must be non-isotropic. Of course, this amounts to finding a basis of Kn+1 in which all the endo- morphisms X 7→ M X of Kn+1, for M ∈ V , have a "reduced" shape that is essentially the one described in Theorem 4. The first problem is how to select the last vector fn+1 of such a basis. Since the rank of an alternate matrix is even, an obvious necessary condition is that V should not contain any matrix with span(fn+1) as column space. Our starting point is that such a vector exists (and may even be chosen amongst the canonical basis of Kn+1). This has already been proven in [10, Proposition 10]: we reproduce a proof since it is short and the result is crucial to our study. 15 Lemma 18. Let W be a linear subspace of Mp(K) with a trivial spectrum. Then there exists a non-zero vector X ∈ Kp such that W contains no matrix M with span(X) as column space. Proof. Denote by (e1, . . . , ep) the canonical basis of Kp. For X ∈ Kp r {0}, set WX :=(cid:8)M ∈ W : Im(M ) ⊂ span(X)(cid:9). For (i, j) ∈ [[1, p]]2, denote by Ei,j the matrix of Mp(K) with zero entries everywhere except at the (i, j)-spot where the entry is 1. We prove, by induction on p, that there exists an index i ∈ [[1, p]] such that Wei = {0}. The case p = 1 is trivial. Assume that Wei 6= {0} for every i ∈ [[1, p]], denote by W ′ the linear subspace of W consisting of its matrices with zero as last row, and write every M ∈ W ′ as M =(cid:20) J(M ) [0]1×(p−1) [?](p−1)×1 0 (cid:21) with J(M ) ∈ Mp−1(K). Then J(W ′) is a linear subspace of Mp−1(K) with a trivial spectrum. The induction hypothesis yields an index i ∈ [[1, p − 1]] such that J(W ′)ei = {0}. Since Wei 6= {0}, we find a matrix M ∈ W such that Im(M ) = span(ei). Then M ∈ W ′ and it follows from J(W ′)ei = {0} that M is a non-zero scalar multiple of Ei,p. Therefore Ei,p ∈ W . Now, taking an arbitrary permutation matrix P ∈ GLn(K) and applying the previous step to P W P −1 yields the following generalization: for every j ∈ [[1, p]], there exists an integer f (j) ∈ [[1, p]] r {j} such that Ef (j),j ∈ W . We choose a cycle for the map f : [[1, p]] → [[1, p]], i.e. a list (j1, . . . , jr) of distinct elements of [[1, p]] such that f (j1) = j2, . . . , f (jr−1) = jr and f (jr) = j1. The matrix A := Ef (jk),jk then belongs to W although 1 is an eigenvalue of it (a rPk=1 corresponding eigenvector being ejk ). This is a contradiction, which shows that Wei = {0} for some i ∈ [[1, p]]. rPk=1 By conjugating V with an appropriate invertible matrix, we then lose no generality assuming that no matrix of V has span(en+1) as column space and that V en+1 ⊂ span(e1, . . . , en) (since en+1 6∈ V en+1). This means that every matrix of V has a 0 entry at the (n + 1, n + 1)-spot. In order to complete the choice of a "good" basis for V , we now turn to the first n vectors f1, . . . , fn. The basic idea is to find the projections of f1, . . . , fn 16 onto span(e1, . . . , en) and alongside span(en+1) by applying the induction hy- pothesis to a subspace of Mn(K) that is deduced from V (the space Vul defined below), and then apply the induction hypothesis once more to find the projec- tions of f1, . . . , fn onto span(en+1) alongside span(e1, . . . , en). Consider the subspace W of V consisting of its matrices with zero as last column. For M ∈ W , write M =(cid:20)K(M ) L(M ) [0]n×1 0 (cid:21) with K(M ) ∈ Mn(K) and L(M ) ∈ M1,n(K), and set Vul := K(W ) (the subscript "ul" stands for "upper left"). The rank theorem shows that dim V = dim W + dim(V en+1) and dim W = dim Ker K + dim Vul. However, our assumptions mean that Ker K = {0}, hence dim V = dim Vul + dim(V en+1). Kn+1. We deduce that Obviously, Vul is a linear subspace of Mn(K) with a trivial spectrum hence dim Vul ≤(cid:0)n 2(cid:1). Moreover dim(V en+1) ≤ n since V acts totally intransitively on (cid:18)n + 1 2 (cid:19) = dim V = dim Vul + dim(V en+1) ≤(cid:18)n 2(cid:19) + n =(cid:18)n + 1 2 (cid:19), hence dim Vul =(cid:18)n 2(cid:19) and dim(V en+1) = n. In this reduced situation, we conclude that: 1. Vul is a maximal linear subspace of Mn(K) with a trivial spectrum. 2. V en+1 = span(e1, . . . , en). Applying the induction hypothesis to Vul together with Remark 2 shows that we may find non-isotropic lower-triangular matrices P1, . . . , Pr such that Vul ≃ P1 An1(K) ∨ · · · ∨ Pr Anr (K). 17 for some R ∈ GLn(K), we lose no generality assuming that This shows that, by conjugating V with a well-chosen matrix of the form(cid:20) R (cid:21) Vul = P1 An1(K) ∨ · · · ∨ Pr Anr (K) and P1 =(cid:20) 1 [0]1×(n1−1) C ′ 1 P ′ 1 [0]1×n [0]n×1 1 (cid:21) for some lower-triangular matrix P ′ column matrix C ′ Remark 3 (An important remark on block-diagrams). From now on, and unless specified otherwise, every matrix M of V will be systematically seen with the following 3 × 3 block decomposition: 1 ∈ Mn1−1(K) (possibly of size 0) and some 1 ∈ Mn1−1,1(K). M = ? [?]1×(n−1) ? [?](n−1)×1 [?]n−1 [?](n−1)×1 ? [?]1×(n−1) ?   ;  .   ? L 0 0 ? ? 0 ? ? 0 0 0 0 U 0 ? 0 ? 18 i.e. the four question marks represent single entries, whilst the others represent submatrices with sizes as indicated by the subscript (where the central subscript n − 1 denotes a (n − 1) × (n − 1) block). 1, P2, . . . , Pr). If n1 > 1, we set s := r, (i1, . . . , is) := (n1 − 1, n2, . . . , nr) and (R1, . . . , Rs) := (P ′ If n1 = 1, we set s := r − 1, (i1, . . . , is) := (n2, . . . , nr) and (R1, . . . , Rs) := (P2, . . . , Pr). In any case, we set Vm := R1 Ai1(K) ∨ · · · ∨ Rs Ais(K) (the subscript "m" stands for "middle"). Here are two consequences of the above reductions (with the block decompositions laid out in Remark 3): (i) For every L ∈ M1,n−1(K), the subspace V contains a matrix of the form (ii) For every U ∈ Vm, the subspace V contains a matrix of the form Proof of statement (i). Let L1 ∈ M1,n1−1(K). Then P1 ×(cid:20) 0 −LT 1 L1 [0]n1−1(cid:21) =(cid:20) ? [?](n1−1)×1 L1 [?]n1−1(cid:21) deduce that, for every L ∈ M1,n−1(K), the subspace Vul contains a matrix of L1 [0]n1−1(cid:21) is alternate. Since Vul = P1 An1(K) ∨ · · · ∨ Pr Anr (K), we [?]n−1(cid:21), and the conclusion follows from the definition of L ? [?](n−1)×1 −LT 1 and(cid:20) 0 the form(cid:20) Vul. Proof of statement (ii). We will only tackle the case n1 > 1, the case n1 = 1 being essentially similar (and even simpler). For every M ∈ P2 An2(K) ∨ · · · ∨ Pr Anr (K) and every N ∈ Mn1−1,n−n1(K), we know that Vul contains the matrix 0 [0](n1−1)×1 [0](n−n1)×1 [0]1×(n1−1) [0]n1−1 [0](n−n1)×(n1−1)  [0]1×(n−n1) N M and it follows that Vul contains a matrix of the form 0 [0](n1−1)×1 [0]1×(n1−1) . Let A ∈ An1−1(K). Then 1A (cid:21) (cid:21) =(cid:20)  . [0](n1−1)×(n−n1) [0](n−n1)×(n−n1) [0]1×(n−n1) P ′ P1 ×(cid:20)  0 [0]1×(n1−1) [0](n1−1)×1 A 0 [0](n1−1)×1 [0](n−n1)×1 [0]1×(n1−1) P ′ 1A [0](n−n1)×(n1−1) With the respective definitions of Vm and Vul, point (ii) follows easily. Let now C ∈ Mn−1,1(K). Since V en+1 = span(e1, . . . , en), we know that V contains a matrix of the form ? ? 0 ? ? C 0 ? ?   . Adding an appropriate matrix given by statement (i), and remembering that 0 is the only possible eigenvalue for a matrix in V , we deduce: 19 (iii) V contains a matrix of the form 0 0 0 ? ? C 0 ? ?   . Denote now by V ′ the subspace of V consisting of its matrices with zero as first row. For M ∈ V ′, write M =(cid:20) 0 [?]n×1 J (M )(cid:21) with J (M ) ∈ Mn(K), [0]1×n and set Vlr := J (V ′) C (the subscript "lr" stands for "lower right"). Note that the subspace Vlr of Mn(K) has a trivial spectrum and that it contains: ment (ii)); [?]1×(n−1) 0 U [0](n−1)×1 [?]1×(n−1) ment (iii)). (cid:21) for every U ∈ Vm (by state- (a) A matrix of the form(cid:20) (b) A matrix of the form(cid:20) [?]n−1 0(cid:21) for every C ∈ Mn−1,1(K) (by state- 2(cid:1). However Since dim Vm =(cid:0)n−1 2(cid:1) since Vlr has a trivial spectrum. It thus follows from statements dim Vlr ≤(cid:0)n (cid:21); (c) Vlr contains, for every U ∈ Vm, a unique matrix of the form(cid:20)U [0](n−1)×1 (cid:21) (d) Every matrix of Vlr with zero as last column has the form(cid:20)U [0](n−1)×1 2 (cid:1), we deduce that dim Vlr ≥(cid:0)n−1 2 (cid:1) + (n − 1) =(cid:0)n (a) and (b) that: ? ? 0 0 for some U ∈ Vm. A key point now is that Vlr is a maximal linear subspace of Mn(K) with a trivial spectrum. One may thus be tempted to apply the induction hypothesis to Vlr. However, the problem is that using a new change of basis blindingly risks destroying the previous reduced form of Vul! As we shall now see, the fact that Vm is already reduced forces Vlr to be already in the reduced form of Theorem 4 (i.e. no further change of basis is necessary at this point). 20 Claim 1. The subspace Vlr has a "roughly reduced" shape i.e. there exists an integer q ≥ 1, a non-isotropic matrix Q ∈ GLq(K) and a maximal subspace W of Mn−q(K) with a trivial spectrum such that Vlr = W ∨ Q Aq(K). Proof. Applying the induction hypothesis to Vlr, we recover a matrix P ∈ GLn(K), a non-isotropic matrix Q′ ∈ GLq(K) (possibly with q = n) and a maximal subspace W ′ of Mn−q(K) with a trivial spectrum such that P VlrP −1 = W ′ ∨ Q′ Aq(K). Note, using statement (b), that dim(Vlren) = n − 1 whereas dim(P VlrP −1x) < n − 1 for every x ∈ span(e1, . . . , en−q) (since W ′ acts totally intransitively on Kn−q). Hence P en 6∈ span(e1, . . . , en−q). Multiplying P with a well-chosen matrix of GLn−q(K) ∨ GLq(K), we lose no generality assuming that P en = en. Assume first that q = 1. Then Vlren = span(e1, . . . , en−1) = (P VlrP −1)en whilst P VlrP −1en = P (Vlren), which shows that P stabilizes span(e1, . . . , en−1). Therefore P ∈ GLn−1(K) ∨ {1} and Vlr = W ∨ A1(K) for some maximal linear subspace W of Mn−1(K) with a trivial spectrum. Assume, for the rest of the proof, that q > 1. Our aim is to prove that P ∈ GLn−q(K) ∨ GLq(K), and it will follow that Vlr = W ∨ Q Aq(K) for some maximal linear subspace W of Mn−q(K) with a trivial spectrum and some non- isotropic matrix Q ∈ GLq(K). Set H :=(cid:8)M ∈ Vlr : M en = 0(cid:9) i.e. H is the set of all matrices of Vlr with 0 as last column. Notice that P HP −1 = (cid:8)M ∈ P VlrP −1 : M en = 0(cid:9) since P en = en. Notice also that span(e1, . . . , en−1−is) ⊂ span(e1, . . . , en−1) = Vlren (this uses statement (b) and the fact that Vlren 6= Kn) and that span(e1, . . . , en−q) ⊂ (P VlrP −1)en . • Case 1: is > 1. 21 -- We first claim that ∀x ∈ Vlren, dim Hx < n − 2 ⇔ x ∈ span(e1, . . . , en−1−is). (1) Indeed, let x ∈ span(e1, . . . , en−1) seen as a vector of Kn−1 with the canonical identification Kn−1 ≃ Kn−1 × {0} ⊂ Kn. By statements (c) and (d), one has dim Vmx ≤ dim Hx ≤ 1 + dim Vmx. If x ∈ span(e1, . . . , en−1−is), then the line of reasoning from the proof of Proposition 16 yields dim Vmx ≤ n − is − 2 and hence dim Hx ≤ n − is − 1 < n − 2; otherwise dim Vmx = n − 2 and hence dim Hx ≥ n − 2. -- Moreover, we claim that ∀x ∈ (P VlrP −1)en, dim(P HP −1x) < n−2 ⇔ x ∈ span(e1, . . . , en−q). (2) The implication ⇐ follows from P VlrP −1 = W ′ ∨ Q′ Aq(K) since W ′ acts totally intransitively on Kn−q and q > 1. For the converse implication, notice first that the equality P VlrP −1 = W ′ ∨ Q′ Aq(K) yields (P VlrP −1)en = span(e1, . . . , en−q) ⊕ G for some (q−1)-dimensional subspace G of span(en−q+1, . . . , en) which does not contain en (note that (P VlrP −1)en cannot contain en since P VlrP −1 has a trivial spectrum). Consider a vector x ∈ G r {0}. The subspace P HP −1 contains, for every A ∈ Aq−1(K), and every B ∈ Mn−q,q(K) with zero as last column, the matrix (cid:20) [0]n−q [0]q×(n−q) C(cid:21) where C = Q′ ×(cid:20) B A [0](q−1)×1 [0]1×(q−1) 0 (cid:21). Since x belongs to span(en−q+1, . . . , en) and is linearly independent from en, it easily follows that dim(P HP −1)x ≥ n − 2. Let now x ∈ (P VlrP −1)en r span(e1, . . . , en−q). Then we have a decomposition x = z + y with z ∈ span(e1, . . . , en−q) and y ∈ G r {0}. Obviously, there exists a non-singular matrix R ∈ {In−q} ∨ {Iq} such that Rx = y. Replacing P with RP , we thus reduce the situation to the one where x ∈ Gr{0}, which we have treated before. Implication ⇒ in statement (2) follows. 22 Since X 7→ P X is linear, P en = en, span(e1, . . . , en−q) ⊂ (P VlrP −1)en, and span(e1, . . . , en−1−is) ⊂ Vlren, we deduce from statements (1) and (2) that X 7→ P X induces an isomorphism from span(e1, . . . , en−1−is) to span(e1, . . . , en−q), hence is = q − 1 and P ∈ GLn−q(K) ∨ GLq(K). • Case 2: is = 1. -- Notice first that span(e1, . . . , en−1) = Vlren and ∀x ∈ Vlren, dim(Hx ∩ Vlren) < n − 2 if x ∈ span(e1, . . . , en−2). (3) Indeed, for every x ∈ span(e1, . . . , en−2), statements (c) and (d) show that dim(Hx ∩ Vlren) ≤ dim(Vmx) (where x is naturally seen as a vector of Kn−1), and the definition of Vm shows, since is = 1, that dim(Vmx) < n − 2. -- On the other hand, we claim that Indeed, for any x ∈ span(e1, . . . , en−q), one has ∀x ∈ (P VlrP −1)en, dim(cid:0)(P HP −1)x∩(P VlrP −1)en(cid:1) < n−2 ⇔ x ∈ span(e1, . . . , en−q). dim(cid:0)(P HP −1)x∩(P VlrP −1)en(cid:1) ≤ dim(P VlrP −1)x ≤ n−q−1 < n−2. Conversely, let x ∈ (P VlrP −1)en r span(e1, . . . , en−q). Note first that (P HP −1)x ⊂ (P VlrP −1)en. In order to see this, we naturally iden- tify Kn with Kn−q ⊕ Kq: the identity P VlrP −1 = W ′ ∨ Q′ Aq(K) (4) [0](q−1)×1 0 [0]1×(q−1) M ∈ P HP −1, the column space of M is included in Kn−q ×Q′ Im(N ), yields (P VlrP −1)en = Kn−q ×(cid:2)Q′(Kq−1 × {0})(cid:3) whilst, for every (cid:21) for some AM ∈ Aq−1(K); this where N = (cid:20) AM prove that dim(P HP −1)x = n − 2, and hence dim(cid:0)(P HP −1)x ∩ (P VlrP −1)en(cid:1) = dim(P HP −1)x = n − 2. Therefore statement (4) is shows that Im M ⊂ (P VlrP −1)en for every M ∈ P HP −1. With the same arguments as in the proof of statement (2), one may established. From statements (3) and (4), we deduce that the linear injection X 7→ P X maps span(e1, . . . , en−2) into span(e1, . . . , en−q), which shows that q = 2, is = 1 = q − 1 and P ∈ GLn−q(K) ∨ GLq(K). This finishes our proof. 23 Now that we know that Vlr is "roughly reduced", we may use the shape of Vm to better grasp the one of Vlr. Take W , q and Q as in Claim 1. If q = 1, then obviously W = Vm. Assume now that q > 1 and split Q = (cid:20) Q1 [?]1×(q−1) [?](q−1)×1 ? (cid:21) with Q1 ∈ Mq−1(K). Then Q1 is still non-isotropic and statement (d) shows that Vm con- tains W ∨ Q1 Aq−1(K), and hence Vm = W ∨ Q1 Aq−1(K) since the dimensions are equal on both sides. By applying the induction hypothesis to W and by using the same arguments as in the proof of Proposition 16, we deduce that W = R1 Ai1(K) ∨ · · · ∨ Rs−1 Ais−1(K) and Q1 Aq−1(K) = Rs Ais(K). Therefore Vlr =(R1 Ai1(K) ∨ · · · ∨ Rs Ais(K) ∨ A1(K) R1 Ai1(K) ∨ · · · ∨ Rs−1 Ais−1(K) ∨ Q Aq(K) if q = 1 if q > 1. Assume again that q > 1. Then Q need not be lower-triangular, so we have to reduce the situation a little further. Since V en+1 = span(e1, . . . , en), we find that Q Aq(K)eq = span(e1, . . . , eq−1) [?](q−1)×1 which shows that Q stabilizes span(e1, . . . , eq−1), i.e. Q =(cid:20) for some T0 ∈ GLq−1(K) and some α ∈ K r {0}. Note, since Q is non-singular, that a matrix of the form M = QA, with A ∈ Aq(K), has zero as last column if and only if A has zero as last column. It then follows from the shape of Vm that T0 Aq−1(K) = Rs Ais(K). Therefore (R−1 s T0) Aq−1(K) = Aq−1(K), and we deduce from Lemma 12 that T0 is a scalar multiple of Rs. Since we may replace Q with a scalar multiple of itself, we lose no generality assuming that T0 = Rs. [0]1×(q−1) α (cid:21) T0 Finally we define T1 := (cid:20) Q′ := (T1)T QT1 is lower-triangular; we replace V with RV R−1 for R :=(cid:20) 1(cid:21) ∈ GLq(K), where C ′ := [0]1×(q−1) Iq−1 C α C ′ , so that In+1−q [0]q×(n+1−q) [0](n+1−q)×q T T 1 (cid:21). For the sake of convenience (and symmetry), we now set P := P1 and p := n1. Let us see how the situation looks like after all those reductions: 24 (i) We still have and Vul = P Ap(K) ∨ P2 An2(K) ∨ · · · ∨ Pr Anr (K) Vm = R1 Ai1(K) ∨ · · · ∨ Rs Ais(K) with the above notations (nothing has changed there). Recall that (i1, . . . , is) = (n2, . . . , nr) if p = 1, otherwise (i1, . . . , is) = (n1 − 1, n2, . . . , nr). (ii) Either q = 1 and then Vlr = R1 Ai1(K) ∨ · · · ∨ Rs Ais(K) ∨ Q A1(K) with Q = 1, or q > 1 and then Vlr = R1 Ai1(K) ∨ · · · ∨ Rs−1 Ais−1(K) ∨ Q Aq(K) and Q =(cid:20)Rs L1 [0]is×1 α (cid:21) with α ∈ K r {0} and L1 ∈ M1,q−1(K). We set α := 1 if q = 1. (iii) Recall finally that if p > 1, then P = (cid:20) 1 C1 Mp−1,1(K). [0]1×(p−1) R1 (cid:21) for some C1 ∈ (iv) No matrix of V has span(en+1) as column space (no change there). However, one important thing has changed: if q > 1, we no longer have V en+1 = span(e1, . . . , en), rather V en+1 = span(e1, . . . , en+1−q) ⊕ H for some linear hy- perplane H of span(en+2−q, . . . , en+1) which does not contain en+1. We still have e1 ∈ V en+1, nevertheless. Set finally Z := R1 (0) . . . (0) Rs From there, V will remain essentially fixed. We will prove separately:  ∈ GLn−1(K). 25 • That the case p = n = q (i.e. Vul and Vlr are glued) leads to the equivalence of V with An+1(K); • That the case p 6= n or q 6= n (i.e. Vul and Vlr are unglued) leads to the reducibility of V . Prior to studying the two cases separately, we continue with general considera- tions that apply to both of them. 4.3 Special types of matrices in V With the matrices L1 and C1 from the previous paragraph2, set fL1 :=(cid:2)[0]1×(n−q) L1(cid:3) ∈ M1,n−1(K) and fC1 :=(cid:20) C1 [0](n−p)×1(cid:21) ∈ Mn−1,1(K). Notation 7. For an arbitrary L ∈ M1,n−1(K), we define L as the matrix of M1,n−1(K) with the same first p − 1 entries as L and all the other ones equal to zero. For an arbitrary C ∈ Mn−1,1(K), we define C as the matrix of Mn−1,1(K) with the same last q − 1 entries as C and all the other ones equal to zero. Using the respective shapes of Vul, Vm and Vlr, we now find important classes of matrices in V , together with an isolated matrix. First of all, taking arbitrary row matrices L0 ∈ M1,p−1(K) and L′ 0 ∈ M1,n−p(K), we know that Vul contains N [0]p−1(cid:21) and N = a matrix of the form (cid:20) P A [0](p−1)×(n−p)(cid:21). Therefore, using the block decomposition of matrices of V (cid:20) [0]n−p(cid:21) with A = (cid:20) 0 [0](n−p)×p −LT 0 L′ 0 L0 explained in Remark 3, we find that: • For every L ∈ M1,n−1(K), there is a unique3 AL ∈ V of the form AL = 0 0 L  , T fC1L 0 −ZL f (L) ϕ(L) 0 and f : M1,n−1(K) → K and ϕ : M1,n−1(K) → M1,n−1(K) are linear maps. 2Setting L1 := 0 if q = 1, and C1 := 0 if p = 1. 3As the map K from the beginning of Section 4.2 is one-to-one. 26 [?]n−1−is,is RsA (cid:21) [0]is,n−1−is with A ∈ As(K). With the respective structures of Vul and Vm and the fact that V contains no matrix with column space span(en+1), we know that V contains Let U ∈ Vm, which we write as a block-triangular matrix U =(cid:20) [?]n−1−is a unique matrix of the form (cid:20) V hash? fL1Z −1U 0i as last row. Therefore: (cid:21) = (cid:21), the structure of Vlr yields that the above matrix of . Since Q ×(cid:20) • For every U ∈ Vm, there is a unique EU ∈ V of the form 0 0 0 U 0 ? 0 RsA s (RsA) [0](is−1)×1 [0]1×(is−1) L1R−1 A [0](is−1)×1 0 ? 0 0 EU = 0 0 0 U 0 0  . as last column. Summing • The subspace V contains a matrix it with a well-chosen matrix of type AL, we deduce: h(U ) fL1Z −1U 0 We know that some matrix of V has(cid:2)1 0 · · · 0(cid:3)T  with (a, b) ∈ K2 and (L′ J = 2 (cid:1) and dim Vlr = (cid:0)n knew that dim V = (cid:0)n+1 0 1 a C ′ ? 0 1 b L′ 1 0 1, C ′ With the above matrices AL and J, we find that dim(eT that the map J from Section 4.2 yields an isomorphism from the subspace of all matrices of V with zero as first row to Vlr. Using the structure of Vlr with the same method as in the definition of the AL matrices, we thus find one last important class of matrices in V : 1 V ) ≥ n. We already 2(cid:1), hence the rank theorem shows 1) ∈ M1,n−1(K) × Mn−1,1(K). • For every C ∈ Mn−1,1(K), there is a unique BC ∈ V of the form BC = 0 ψ(C) g(C) −α C 0 0 T 0 C  (Z −1)T fL1Z −1C and g : Mn−1,1(K) → K and ψ : Mn−1,1(K) → Mn−1,1(K) are linear maps. 27 Remark 4. The above matrices span V : a straightforward computation shows 2 (cid:1) + 1 =(cid:0)n+1 indeed that the linear subspaces(cid:8)AL L ∈ M1,n−1(K)(cid:9),(cid:8)BC C ∈ Mn−1,1(K)(cid:9), (cid:8)EU U ∈ Vm(cid:9) and span(J) are independent, and the sum of their dimensions is (n − 1) + (n − 1) +(cid:0)n−1 From now on, our main task is to refine our understanding of the matrices of the types AL, BC, EU and J: the basic strategy is to form well-chosen linear combinations of those special matrices and use the fact that none of them may have a non-zero eigenvalue. Most of the time, we will simply apply the fact that both V and V T act totally intransitively on Kn+1. Let us start by considering the maps ϕ and ψ in the AL and BC matrices. 2 (cid:1) = dim V . Claim 2. The maps ϕ and ψ are scalar multiples of the identity. Proof. Let C ∈ Mn−1,1(K) and L ∈ M1,n−1(K). Denote by x (resp. y) the vector of span(e2, . . . , en) with coordinate matrix C (resp. LT ) in the basis (e2, . . . , en). We prove that LC = 0 ⇒(cid:0)ϕ(L) C = 0 and L ψ(C) = 0(cid:1). Assume that LC = 0. Notice then that both AL and BC stabilize the plane span(x, en+1) and that the respective matrices of their induced endomorphisms in the basis (x, en+1) are(cid:20) V has a trivial spectrum, we deduce that 1 t2(cid:21) for some (t1, t2) ∈ K2. Since (5) 0 0 t1 ϕ(L)C 0(cid:21) and(cid:20) 0 (cid:12)(cid:12)(cid:12)(cid:12) t1 + λ ϕ(L)C 1 + t2(cid:12)(cid:12)(cid:12)(cid:12) 6= 0, 1 1 ∀λ ∈ K, hence ϕ(L)C = 0. Similarly, notice that AT 0 0 L and BT C both stabilize span(e1, y) and the respective matrices of their induced endomorphisms in the basis (e1, y) are (cid:20)0 s1 (cid:20)0 Lψ(C) 1 s2(cid:21) and (cid:21) for some (s1, s2) ∈ K2. With the above line of reasoning, we deduce that Lψ(C) = 0. We may now conclude. For the non-degenerate bilinear mapping (L, C) 7→ LC on M1,n−1(K) × Mn−1,1(K), we deduce from (5) that ϕ stabilizes the orthogonal subspace of every linear hyperplane of M1,n−1(K), hence ϕ stabilizes every 1- dimensional linear subspace of M1,n−1(K), which shows that ϕ is a scalar multiple of the identity. With the same line of reasoning, we see that ψ is also a scalar multiple of the identity. 28 We now have two scalars λ and µ such that: ∀L ∈ M1,n−1(K), AL = 0 L 0 −ZL f (L) T fC1L 0 λ L 0 and ∀C ∈ Mn−1,1(K), BC = 0 µ C g(C) −α C 0 0 T Claim 3. The map h vanishes everywhere on Vm.  0 C  . (Z −1)T fL1Z −1C Proof. Choose t ∈ K such that µ + t 6= 0 and a + t 6= 0 (this is feasible since #K ≥ 3). Remark then that ∀U ∈ Vm, ∀C ∈ Mn−1,1(K),  0 [0](n−1)×1 h(U ) 0 (µ + t) C   EU (e1 + t en+1) = BC(e1 + t en+1) = J(e1 + t en+1) = a + t ? ? ?  . However V (e1 + ten+1) is a strict linear subspace of Kn+1. Judging from the vectors BC(e1+t en+1) and the vector J(e1+t en+1), we deduce that V (e1+ten+1) cannot contain en+1. This shows that h(U ) = 0 for every U ∈ Vm. It follows that ∀U ∈ Vm, EU = 0 0 0 0 0 U  . 0 fL1Z −1U 0 From there, we need to study the glued and unglued cases separately. 29 4.4 The case Vul and Vlr are glued In this section, we assume p = q = n. In this case, we simply havefL1 = L1,fC1 = C1, Z = R1 = Rs, Vm = Z An−1(K) and ∀(L, C) ∈ M1,n−1(K) × Mn−1,1(K), L = L and C = C. Our aim is to prove that V is equivalent to An+1(K). Claim 4. One has ∀(L, C) ∈ M1,n−1(K) × Mn−1,1(K), f (L) = −L1LT and g(C) = µL1Z −1C. Proof. Let t ∈ Kr{−a}. Note that J(e1 +ten+1) has a+t as first entry, whereas  ∀L ∈ M1,n−1(K), AL(e1 + ten+1) = ∀C ∈ Mn−1,1(K), BC(e1 + ten+1) = 0 −ZLT f (L)  0 (µ + t)C g(C) + tL1Z −1C  . Judging from J(e1 + ten+1), the vector space V (e1 + ten+1) cannot contain span(e2, . . . , en+1). Thus V (e1 + ten+1) ∩ span(e2, . . . , en+1) =(cid:8)AL(e1 + ten+1) L ∈ M1,n−1(K)(cid:9) (since the first space has a dimension lesser than n and obviously contains the second one). Using the BC matrices, it follows that ∀C ∈ Mn−1,1(K), g(C) + tL1Z −1C = (µ + t) f(cid:0)−C T (Z −1)T(cid:1). Since this holds for several values of t, we deduce that ∀C ∈ Mn−1,1(K), g(C) = −µf (C T (Z −1)T ) which obviously yields the claimed results. and L1Z −1C = −f(cid:0)C T (Z −1)T(cid:1),  0 C 0 0 0 µ C µ L1Z −1C −α C T (Z −1)T L1Z −1C Therefore, for any (L, C, U ) ∈ M1,n−1(K) × Mn−1,1(K) × Z An−1(K), we have AL = and 0 L 0 −ZLT C1L 0 −L1LT λ L 0  ; BC = EU = 30 0 U 0 0 0 0 0 L1Z −1U 0  . Set now 0 1 0 C1 Z 0 λ L1 α T := T −1AL T ′ = −LT 0 0 and  ∈ GLn+1(K) and T ′ :=  ; T −1BC T ′ = T −1EU T ′ = 0 0 0 Z −1U 0 0 0 L 0 0 0 0 0 0 0  . 1 0 −µ 0 0 In−1 0 1 0 0 0 0 0 0 −(Z −1C)T  ∈ GLn+1(K).  Z −1C 0 0 A straightforward computation shows that, for every (L, C, U ) ∈ M1,n−1(K) × Mn−1,1(K) × (Z An−1(K)): Therefore T −1V T ′ contains a linear hyperplane of An+1(K). Since V acts totally intransitively on Kn+1, this is also the case of T −1V T ′, hence Lemma 15 shows that T −1V T ′ = An+1(K). We deduce that V is equivalent to An+1(K) and may thus be written as Y An+1(K) for some Y ∈ GLn+1(K), and Lemma 10 yields that Y is non-isotropic. This completes the case where Vul and Vlr are glued. 4.5 The case Vul and Vlr are unglued Here, we assume that p < n or q < n. Note that this means that p = 1 or q = 1 or there are several diagonal blocks R1 Ai1(K), . . . , Rs Ais(K) in the block decomposition of Vm discussed earlier. Note in particular that p + q ≤ n + 1. Our aim is to prove that V is reducible. Since the matrices AL, BC, EU and J span V , it suffices to find a non-trivial linear subspace of Kn+1 which is stabilized by all of them. In that prospect, we start by analyzing f and g. Claim 5. One has f = 0, and g(C) = 0 for every C ∈ Mn−1,1(K) such that C = 0. Proof. We start by proving that ∀L ∈ M1,n−1(K), L = 0 ⇒ f (L) = 0 and ∀C ∈ Mn−1,1(K), C = 0 ⇒ g(C) = 0. 31 We choose t ∈ K such that µ+t 6= 0 and a+t 6= 0. Then, for every L ∈ M1,n−1(K) such that L = 0, one has hence f (L) = 0 with the same argument as in the proof of Claim 3. Choose now x ∈ K such that λ + x 6= 0 and x 6= 0. Then 0 f (L) [0](n−1)×1 AL(e1 + ten+1) =  , (xe1 + en+1)T AL =hf (L) (λ + x)L 0i (xe1 + en+1)T J =h? [?]1×(n−1) xi . Since V T (xe1 + en+1) is a strict linear subspace of Kn+1, those matrices show that e1 cannot belong to V T (xe1 + en+1). However ∀L ∈ M1,n−1(K),  ∀C ∈ Mn−1,1(K), (xe1 + en+1)T BC =hg(C) −α C Therefore, if C = 0, then fL1Z −1C = 0 and hence g(C) = 0. T T Let L ∈ M1,n−1(K). The column matrix C := ZL the p-th, and since p + q ≤ n + 1, this yields C = 0. Therefore g(C) = 0 and has null entries starting from (Z −1)T fL1Z −1Ci . (cid:0)(µ + t)AL + BC(cid:1)(e1 + ten+1) = The above argument then shows that f (L) = 0. In particular, we have 0 [0](n−1)×1 (µ + t)f (L)  . ∀L ∈ M1,n−1(K), AL =  . 0 L 0 −ZL 0 T fC1L 0 λ L 0 We now distinguish between two cases, whether p < n or p = n. Claim 6. If p < n, then V e1 ⊂ span(e1, . . . , ep). 32 Proof. Assume that p < n. Write C ′ T T 1 = c′ 1 ... c′ n−1 . Let i ∈ [[p, n − 1]] (note that such an integer exists). have: ∀L ∈ M1,n−1(K), Let (x, y, z) ∈ K3 such that x + λz 6= 0. Denote by C ′′ i ∈ Mn−1,1(K) the column matrix with all entries 0 except the i-th which equals 1. Note that, for every have zero entries starting , this is obvious if p = 1 because then L = 0, otherwise this comes from the fact that Z stabilizes Kp−1 × {0} ⊂ Kn−1 (as i1 = p − 1) and that L ∈ Kp−1 × {0}. It follows that the i , we therefore L ∈ M1,n−1(K), both column matrices fC1 and ZL from the p-th: forfC1, this comes from its very definition; for ZL (i + 1)-th row of every AL matrix is zero. Setting γ :=fL1Z −1C ′′ (xe1 + yei+1 + zen+1)T AL =h0 (x + λz) L 0i (xe1 + yei+1 + zen+1)T J =hax + c′ y + γz(cid:12)(cid:12)(cid:12)(cid:12) = 0. (cid:12)(cid:12)(cid:12)(cid:12) Notice that, with an arbitrary (y, z) ∈ K2 being fixed, the above equation is linear in x and has several solutions, hence Since V T (xe1 + yei+1 + zen+1) 6= Kn+1, we deduce that =hµy + g(C ′′ (xe1 + yei+1 + zen+1)T BC ′′ ax + c′ µy + g(C ′′ iy + bz i )z  iy + bz i )z i x (c′ iy + bz)(y + γz) = 0 and a(y + γz) − (µy + g(C ′′ i )z) = 0. Both equations have a degree lesser than or equal to 2 in both variables. Since # K > 2, we deduce that c′ i = 0 ; c′ iγ + b = 0 ; a = µ and aγ = g(C ′′ i ). Therefore a = µ, b = 0 and c′ stabilizes Kp−1 × {0}, we may thus find L ∈ M1,n−1(K) such that C ′ n−1 = 0. Since Z is non-singular and T . 1 = ZL p = · · · = c′ [?]1×(n−1) xi [?]1×(n−1) y + γzi . The first column of AL + J is a 0 ... 0  33 , therefore a = 0 (because AL + J has no non- zero eigenvalue). It follows that µ = 0 and g(C ′′ i ) = 0 for every i ∈ [[p, n − 1]]. Since g is linear, g(C) = 0 whenever C = 0 (by Claim 5), and p − 1 < n − q + 1, we deduce that g = 0. For any matrix of type AL, BC, EU or J, we have therefore found that its first column has null entries starting from the (p + 1)-th. This yields our claim since these matrices span V . Claim 7. Assume that p = n (and therefore q = 1). Then λ = b = 0 and L′ 1 = 0. This shows that all the matrices AL, BC, EU and J have zero as last row in the case p = n. 0 0 0 µ C 0 C 0 0 0  . 0 0 · · · l′ 1 i ∈ M1,n−1(K) the leads to f = 0 and g = 0 by Claim 5. Therefore Proof. Since q = 1, one hasfL1 = 0, whilst C = 0 for every C ∈ M1,n−1(K). This ∀(L, C) ∈ M1,n−1(K)×Mn−1,1(K), AL =  and BC = L 0 ? 0 λ L 0 −ZLT Write L′ row matrix with all entries zero except the i-th which equals one. Let (x, z) ∈ K2 such that µx + z 6= 0. Then n−1(cid:3). Let i ∈ [[1, n − 1]]. Denote by L′′ 1 =(cid:2)l′ ∀C ∈ Mn−1,1(K), BC(xe1 + ei+1 + zen+1) = (xe1 + ei+1 + zen+1) = J(xe1 + ei+1 + zen+1) = i(cid:12)(cid:12)(cid:12)(cid:12) = 0. Since, for a given x ∈ K, this holds for several    . We deduce that(cid:12)(cid:12)(cid:12)(cid:12) values of z, we successively deduce that λ = 0 and ∀x ∈ K, b x + l′ yields λ = b = l′ i = 0. Therefore L′  1 ax + z λ bx + l′ i = 0, which [?](n−1)×1 [?](n−1)×1 (µx + z)C bx + l′ i 0 0 1 λ ax + z AL′′ i 1 = 0. 34 In two special cases, we may now conclude that V is reducible: if p = 1 then Claim 6 shows that span(e1) is stabilized by V ; if p = n, then Claims 5 and 7 show that span(e1, . . . , en) is stabilized by V (indeed, in that case q = 1 and Assume finally that 1 < p < n. Then V e1 ⊂ span(e1, . . . , ep) by Claim 6. hence fL1 = 0 and C = 0 for every C ∈ Mn−1,1(K)). Note that the change of basis matrix R =(cid:20)In+1−q 0 0 T T 1(cid:21) from Section 4.2 leaves span(e1, . . . , ep) invariant as p ≤ n + 1 − q. Therefore we also have (R−1V R)e1 ⊂ span(e1, . . . , ep), and some of our recent findings may be summed up as follows: Proposition 19. Let V be a maximal subspace of Mn+1(K) with a trivial spec- trum such that: (i) V en+1 = span(e1, . . . , en); (ii) There are lower-triangular non-isotropic matrices P ∈ GLp(K), P2 ∈ GLn2(K), . . . , Pr ∈ GLnr (K), with 1 < p < n, such that Vul = P Ap(K) ∨ P2 An2(K) ∨ · · · ∨ Pr Anr (K). Then V e1 ⊂ span(e1, . . . , ep). Note that the fact that V contains no matrix with column space span(en+1), our starting point in Section 4.2, is a consequence of assumptions (i) and (ii) of Proposition 19 (using the rank theorem to compute the dimension of V from that of Vul, as in the beginning of Section 4.2). Now, all we need to complete the unglued case is to show that any V satis- fying the assumptions of Proposition 19 is reducible. Let V be such a subspace, with the above notations. Let x ∈ span(e1, . . . , ep)r{0}. Recall that the bilinear form b : (X, Y ) ∈ (Kp)2 7→ X T P Y is non-isotropic, and hence non-degenerate. Denote by X0 the matrix of coordinates of x in (e1, . . . , ep). In the hyperplane H := {Y ∈ Kp : X T 0 Y = 0}, we may therefore find a "right-sided orthogonal basis" (f2, . . . , fp), i.e. b(fi, fj) = 0 for every (i, j) ∈ [[2, p]]2 with i < j. We then choose a non-zero vector f1 such that b(f1, fj) = 0 for every j ∈ [[2, p]]. It follows that (f1, . . . , fp) is a basis of Kp. Denoting by S the matrix of coordinates of (f1, f2, . . . , fp) in (e1, . . . , ep), the matrix P ′ := ST P S is lower-triangular and ST(cid:0)P Ap(K)(cid:1)(ST )−1 = P ′ Ap(K). 35 Set then T2 :=(cid:20)ST 0 0 In+1−p(cid:21) ∈ GLn+1(K) and V ′ := T2 V T −1 2 . Notice finally that T2 stabilizes span(e1, . . . , en), fixes en+1, and obviously V ′ ul = P ′ Ap(K) ∨ P2 An2(K) ∨ · · · ∨ Pr Anr (K). Thus Proposition 19 applied to V ′ shows that V ′e1 ⊂ span(e1, . . . , ep). However S maps span(e2, . . . , ep) to span(f2, . . . , fp), hence ST X0 ∈ span(e1) r {0}. This yields V x ⊂ span(e1, . . . , ep). We conclude that span(e1, . . . , ep) is a non-trivial invariant subspace for V , hence V is reducible. This completes our proof of Theorem 3. 5 On large spaces of nilpotent matrices In this short section, we show that the following famous theorem of Gerstenhaber on linear subspaces of nilpotent matrices is an easy consequence of Theorem 4: Theorem 20 (Gerstenhaber's theorem). Let K be a field with at least three elements, and V be a linear subspace of Mn(K) such that dim V =(cid:0)n matrix of V is nilpotent. Then V is similar to NTn(K). 2(cid:1) and every See [6] for the original proof under the more restrictive assumption #K ≥ n, [7] for a very elegant proof using trace maps and a theorem of Jacobson, and [14] for a proof with no restriction on the cardinality of K. Proof. The assumptions show that V is a maximal linear subspace of Mn(K) with a trivial spectrum. Then V ≃ P1 An1(K) ∨ · · · ∨ Pp Anp(K) for non-isotropic ma- trices P1, . . . , Pp. Since every matrix of V is nilpotent, every matrix of Pk Ank (K) is nilpotent for every k ∈ [[1, p]]. Let q ≥ 2 be a positive integer and P ∈ GLq(K), and assume that P is non- isotropic and every element of P Aq(K) is nilpotent. Note that q is odd since Aq(K) contains non-singular matrices when q is even. Then tr(P A) = 0 for every A ∈ Aq(K), which shows that P is symmetric. Since q is odd and P is non-singular, P is not alternate hence it is congruent to a non-singular diagonal matrix D (even if K has characteristic 2, see [9, Chapter 35]). Thus D Aq(K) 36 is similar to P Aq(K) and must therefore have a trivial spectrum. Finally, set K :=(cid:20) 0 −1 0(cid:21) and A :=(cid:20) K 1 [0](q−2)×2 [0]2×(q−2) [0]q−2 (cid:21) ∈ Aq(K), and note that DA is obviously non-nilpotent, a contradiction. Returning to V , we deduce that n1 = · · · = np = 1 hence V ≃ NTn(K). 6 On the exceptional case of F2 In the proof of Theorem 4, we have repeatedly used the assumption that the field K had at least 3 elements. The reader will therefore not be surprised by the following counterexample which shows that Theorem 4 fails for the field F2. Remark first that there is no non-isotropic matrix in GL3(F2) (since every 3- dimensional quadratic form over a finite field is isotropic), hence no maximal linear subspace of M3(F2) with a trivial spectrum has the form P A3(F2). Consider the following matrices of M3(F2):  and C := 0 0 0 0 1 1 1 1 0  . A := 0 1 0 0 0 0 0 1 0  ; B := 1 0 1 1 0 0 1 0 0 Using the identities ∀x ∈ F2, x + x = 0 and x2 = x, a straightforward computa- tion yields ∀(x, y, z) ∈ F3 2, det(I3 + x A + y B + z C) = 1. Therefore the 3-dimensional subspace V := span(A, B, C) has a trivial spectrum. The fact that A + B is non-singular shows however that V is irreducible. If V were reducible indeed, then there would exist a 1-dimensional subspace W of M2(F2) such that V ≃ {0} ∨ W or V ≃ W ∨ {0}, and in both cases every matrix of V would be singular. The classification of the irreducible maximal subspaces of Mn(F2) with a trivial spectrum thus remains an unresolved issue. References [1] M. D. Atkinson, S. Lloyd, Large spaces of matrices of bounded rank, Quart. J. Math. Oxford (2) 31 (1980) 253-262. 37 [2] M. D. Atkinson, S. Lloyd, Primitive spaces of matrices of bounded rank, J. Austral. Math. Soc. 30 (1981) 473-482. [3] M. D. Atkinson, S. Lloyd, Primitive spaces of matrices of bounded rank II, J. Austral. Math. Soc. 30 (1983) 306-315. [4] M. D. Atkinson, N. M. Stephens, Spaces of matrices of bounded rank, Quart. J. Math. Oxford (2) 29 (1978) 221-223. [5] J. Dieudonn´e, Sur une g´en´eralisation du groupe orthogonal `a quatre vari- ables, Arch. Math. 1 (1949) 282-287. [6] M. Gerstenhaber, On Nilalgebras and Linear Varieties of Nilpotent Matrices (I), Amer. J. Math. 80 (1958) 614-622. [7] B. Mathes, M. Omladic, H. Radjavi, Linear spaces of nilpotent matrices, Linear Algebra Appl. 149 (1991) 215-225. [8] R. Quinlan, Spaces of matrices without non-zero eigenvalues in their field of definition, and a question of Szechtman, Linear Algebra Appl. 434 (2011) 1580-1587. [9] C. de Seguins Pazzis, Invitation aux formes quadratiques, Calvage & Mounet, Paris (2011). [10] C. de Seguins Pazzis, On the matrices of given rank in a large subspace, Linear Algebra Appl. 435-1 (2011) 147-151. [11] C. de Seguins Pazzis, The affine preservers of non-singular matrices, Arch. Math. 95 (2010) 333-342. [12] C. de Seguins Pazzis, The classification of large spaces of matrices with bounded rank, arXiv preprint http://arxiv.org/abs/1004.0298 [13] C. de Seguins Pazzis, The singular linear preservers of non-singular matri- ces, Linear Algebra Appl. 433 (2010) 483-490. [14] V. N. Serezhkin, On linear transformations preserving nilpotency, Izv. Akad. Nauk BSSR, Ser. Fiz.-Mat. Nauk 6 (1985) 46-50. 38
1211.6310	2	1211	2013-02-25T14:40:23	A model for the relatively free graded algebra of block triangular matrices with entries from a graded algebra	[ "math.RA" ]	Let G be a group and A be a G-graded algebra satisfying a polynomial identity. We buid up a model for the relative free G-graded algebra and we obtain, as an application, the "factoring" property for the T_G-ideals of block triangular matrices with entries from the finite dimensional Grassmann algebra E for some particular Z_2-grading.	math.RA	math	A MODEL FOR THE RELATIVELY FREE GRADED ALGEBRA OF BLOCK-TRIANGULAR MATRICES WITH ENTRIES FROM A GRADED PI-ALGEBRA LUCIO CENTRONE AND THIAGO CASTILHO DE MELLO Abstract. Let G be a group and A be a G-graded algebra satisfying a poly- nomial identity. We construct a model for the relatively free G-graded algebra of the G-graded algebra of block-triangular matrices with entries from A. We obtain, as an application, the "factoring" property for the TZ2 -ideals of block- triangular matrices with entries from the infinite dimensional Grassmann al- gebra E for some particular Z2-grading. 1. introduction Let F be a field and we denote by F hXi the free associative algebra freely generated by the set X over F . We refer to elements of F hXi as polynomi- als in the non-commutative variables of X. If A is an F -algebra we say that A satisfies a polynomial identity (or A is a PI-algebra) if there exists an element f = f (x1, . . . , xn) ∈ F hXi such that f (a1, . . . , an) = 0, for any a1, . . . , an ∈ A. Such f is called a polynomial identity of A. The set of all polynomial identities of A, denoted by T (A), is called the T-ideal of A. It is an ideal of F hXi which is invari- ant under endomorphisms of the algebra F hXi. The algebra F hXi/(F hXi ∩ T (A)) is called the relatively free algebra of A (see the paper of Belov [2] for a detailed account on relatively free algebras and their representability in any characteristic). We recall that if R is a block-triangular matrix algebra such that R =(cid:18) A M 0 B (cid:19) , where A and B are PI algebras and M is an A, B-bimodule, then R is a PI-algebra too. We say that R (or also T (R)) satisfies the factoring property if T (R) = T (A)T (B), where the product of T-ideals is the usual product of ideals. We refer to the papers [10] and [9] by Drensky for a conjecture regarding the factoring property of T-ideals of block-triangular matrices with entries from minimal algebras already solved positively by Giambruno and Zaicev in [12]. In this paper we focus on the factoring property of block-triangular matrices in the graded case. Let G be a group and let us consider a block triangular matrix algebra R =(cid:18) A M 0 B (cid:19) , where A and B are PI G-graded algebras and M is an A, B-bimodule. In [7] Di Vincenzo and La Scala gave a proof for the Theorem of Lewin in the graded case. T. C. de Mello was partially supported by grants from Capes (AUX-PE-PRODOC-2548/2010) and from FAPESP (No. 2012/16838-0). 1 2 LUCIO CENTRONE AND THIAGO CASTILHO DE MELLO Moreover, if A is a G-graded subalgebra of Mn(F ) they introduce the notion of G-regularity for A. We recall that a map · : {1, . . . , n} → G induces a G-grading over Mn(F ) such that for each i, j the homogeneous degree of the matrix unit eij is ji−1. If A is a graded subalgebra of Mn(F ), graded by a finite abelian group G, we have that A is G-regular if · is surjective and its fibers are equipotent (see Theorem 5.4 of [7]). They also proved that if A ⊆ Mn(F ) and B ⊆ Mm(F ) are matrix algebras graded by a finite abelian group G, then R has the factoring property for its ideal of graded polynomial identities if and only if one between A and B is G-regular. In the proof of the Theorem of Lewin, both in the ordinary and in the graded case, one prominent role is played by the relatively free (graded) algebra. In [19] Procesi showed the k-generated relatively free algebra of Mn(F ) is isomorphic to the k-generated algebra of generic matrices over the polynomial ring, and such construction can be generalized for relatively free algebras of finite dimensional PI- algebras (see [18]). Moreover in [3] Berele constructed the k-generated relatively free algebras of the minimal algebras Mn(E) and Ma,b(E). It turned out that they are isomorphic to some k-generated subalgebra of generic matrices over the supercommutative polynomial algebra. Let A be an algebra over a field F . If d1, d2, . . . , dm are positive integers, we denote by U T (d1, . . . , dm; A) the subalgebra of the matrix algebra Md1+···+dm(A) consisting of matrices of the type . . . A1m . . . A2m ... . . . . . . Amm ,   . . . A11 A12 A22   0 ... 0 where Aij ∈ Mdi×dj (A) for each i, j. One such algebra is called the algebra of block- triangular matrices of size d1, . . . , dm over A. We may observe that if in addition A is a PI-algebra, then U T (d1, . . . , dm; A) is a PI-algebra, too. In this paper we consider a PI G-graded algebra A and we construct a model for the relatively free G- graded algebra of U T (d1, . . . , dm; A). It turns out that the relatively free G-graded algebra of U T (d1, . . . , dm; A) is isomorphic to the algebra of generic matrices with entries from the graded relatively free algebra of A. We use the model to prove that U T (d1, . . . , dm; A) has the factoring property when the relatively free graded algebra of A has a partially multiplicative basis. We also observe that the factoring property fails if we consider U T (d1, . . . , dm; E) with the grading induced by the map · k∗ of the work of Di Vincenzo and Da Silva (see [8]). The paper is organized as follows. Sections 2 is dedicated to the main definitions concerning the graded polynomial identities. In Sections 3 we present the main tool of the paper: the graded version of the Theorem of Lewin obtained by Di Vincenzo and La Scala in [7]. In Section 4 we present the results regarding the Z2-graded identities of the infinite dimensional Grassmann by Di Vincenzo and Da Silva (see [8]). The results of Sections 4 will be used for applications of the main Theorem. Section 5 is devoted to the construction of the model for the relatively free graded algebra of the upper triangular block matrices with entries from a G-graded PI algebra. We obtain the factoring property for U T (d1, . . . , dm; A), where A is a G-regular algebra having the same TG ideal of Mn(F ) when G is a finite abelian group. In Section 6 we present the main theorem and, as a consequence, we obtain the factoring property RELATIVELY FREE GRADED ALGEBRAS 3 for U T (d1, . . . , dm; E) when Z2-graded by the Z2-grading induced by the natural grading over E. We also mention that the factoring property for U T (d1, . . . , dm; E) has been proved by Berele and Regev in the ungraded case (see [4]). 2. Graded structures All fields we refer to are assumed to be of characteristic zero and all algebras we consider are associative and unitary. Let (G, ·) = {g1, . . . , gr} be any finite group, and let F be a field. If A is an associative F -algebra, we say that A is a G-graded algebra (or G-algebra) if there are subspaces Ag for each g ∈ G such that Ag and AgAh ⊆ Agh. A = Mg∈G If 0 6= a ∈ Ag we say that a is homogeneous of G-degree g or simply that a has degree g, and we write deg(a) = g. We consider the following subset of G Supp(A) = {g ∈ GAg 6= 0}. The latter is called the support of the G-graded algebra A. We define a free object; let {X g g ∈ G} be a family of disjoint countable sets. Put X =Sg∈G X g and denote by F hXGi the free associative algebra freely generated by the set X. An indeterminate x ∈ X is said to be of homogeneous G-degree g, written deg(x) = g, if x ∈ X g. We always write xg if x ∈ X g. The homogeneous G-degree of a monomial m = xi1 xi2 · · · xik is defined to be deg(m) = deg(xi1 ) · deg(xi2 ) · · · · · deg(xik ). For every g ∈ G, we denote by F hXGig the subspace of F hXGi spanned by all the monomials having homogeneous G-degree g. Notice that F hXGigF hXGig′ for all g, g′ ∈ G. Thus ⊆ F hXGigg′ F hXGi = Mg∈G F hXGig proves F hXGi to be a G-graded algebra. The elements of the G-graded alge- bra F hXGi are referred to as G-graded polynomials or, simply, graded polyno- mials. An ideal I of F hXGi is said to be a TG-ideal if it is invariant under all F -endomorphisms ϕ : F hXGi → F hXGi such that ϕ (F hXGig) ⊆ F hXGig for all g ∈ G. If A is a G-graded algebra, a G-graded polynomial f (x1, . . . , xn) is said to be a graded polynomial identity of A if f (a1, a2, · · · , at) = 0 for all a1, a2, · · · , at ∈Sg∈G Ag such that ak ∈ Adeg(xk), k = 1, · · · , t. If A has a non-zero graded polynomial identity, we say that A is a G-graded polynomial identity algebra (GPI-algebra). We denote by TG(A) the ideal of all graded polynomial identities of A. It is a TG-ideal of F hXGi. If A is ungraded, i.e., graded by the trivial group, we talk about polynomial identities and T-ideal of A. We recall that if the group G is finite and A is a G-graded PI-algebra, then it satisfies a polynomial identity (see [1], [5]). Moreover, we recall that if two GP I-algebras A and B satisfy the same graded identities, i.e., TG(A) = TG(B), then they satisfy the same identities, i.e., T (A) = T (B). When one deals with TG-ideals, given a subset Y ⊆ X one can talk about the least TG-ideal of F hXGi which contains the set Y . Such TG-ideal will be denoted by hY iTG and will be called the TG-ideal generated by Y . We say that elements of hY iTG are consequences of elements of Y , or simply that they follow from Y . Given a G-algebra A one of the main problems in PI-theory is to find a finite set Y 4 LUCIO CENTRONE AND THIAGO CASTILHO DE MELLO such that Tg(A) = hY iTG . Such a Y is called a basis for the G-graded polynomial identities of A. F hXGi F hXGi ∩ TG(A) We denote by UG(A) the factor algebra and we shall call it the relatively free G-algebra of A. We observe that if G = {1G}, we obtain the definition of the relatively free algebra. 3. Z2-graded identities for the Grassmann algebra In this section we recall the main tools and definitions that are necessary for the study of graded polynomial identities of the Grassmann algebra. We shall indicate the infinite dimensional Grassmann algebra by E. The algebra E can be constructed as follows. Let F hXi be the free algebra of countable rank on X = {x1, x2, . . .}. If I is the two-sided ideal of F hXi generated by the set of polynomials {xixj + xjxii, j ≥ 1}, then E = F hXi/I. If we write ei = xi + I for i = 1, 2, . . . , then E has the following presentation: E = h1, e1, e2, . . . eiej = −ejei, for all i, j ≥ 1i. We say that the vector space V generated by X over F is the generating vector space of E. Moreover, the set B = {1, ei1 · · · eik 1 ≤ i1 < · · · < ik} is a basis of E over F . Sometimes it is convenient to write E in the form E = E0 ⊕ E1, where E0 := span{1, ei1 · · · ei2k 1 ≤ i1 < · · · < i2k, k ≥ 0}, E1 := span{ei1 · · · ei2k+11 ≤ i1 < · · · < i2k+1, k ≥ 0}. It is easily verified that the decomposition E = E0 ⊕ E1 is a Z2-grading of E called the natural grading. Notice that E0 coincides with the center of E. We give a look at the whole class of homogeneous Z2-grading of E. For more details we refer to the work of Di Vincenzo and Da Silva ([8]). For a homogeneous Z2-grading of E we mean any Z2-grading such that the generating vector space V is a homogeneous subspace. This is equivalent to consider a map deg : V → Z2. If w = ei1ei2 · · · ein ∈ E, then the set Supp(w) := {ei1, ei2, . . . , ein } is the support of w and we define the Z2-grading of w by deg(ei1 ei2 · · · ein ) = deg(ei1 ) + · · · + deg(ein ). If, for all ei ∈ B, one has deg(ei) = 1 ∈ Z2, then we obtain the natural Z2-grading on E. In this case, let E0 be the homogeneous component of Z2-degree 0 and let E1 be the component of degree 1. As we said above, E0 = Z(E) is the center of E and ab + ba = 0 for all a, b ∈ E1. This means that E satisfies the following graded polynomial identities: [y1, y2], [y1, z1], z1z2 + z2z1. Now, let us consider the Z2-gradings on E induced by the maps deg(·)k∗, deg(·)∞, and deg(·)k, defined respectively by: deg(ei)k∗ =(cid:26) 1 for i = 1, . . . , k 0 otherwise, RELATIVELY FREE GRADED ALGEBRAS 5 0 otherwise, deg(ei)∞ =(cid:26) 1 for i odd deg(ei)k =(cid:26) 0 for i = 1, . . . , k 1 otherwise. By [8], we have the following result. Theorem 1. Let Y be a countable set of indeterminates of degree 0 and Z be a countable set of indeterminates of degree 1 and put X = Y ∪ Z. Then: (1) The TZ2 -ideal of E graded by deg(·)∞ is generated by the polynomial [x1, x2, x3]. (2) The TZ2 -ideal of E graded by deg(·)k∗ is generated by the polynomials (3) The TZ2 -ideal of E graded by deg(·)k is generated by the polynomials [x1, x2, x3], z1z2 · · · zk+1. [x1, x2, x3], [y1, y2] · · · [yk−1, yk][yk+1, x] (if k is even), [y1, y2] · · · [yk, yk+1] (if k is odd), gk−l+2(z1, . . . , zk−l+2)[y1, y2] · · · [yl−1, yl] (if l ≤ k), [gk−l+2(z1, . . . , zk−l+2), y1][y2, y3] · · · [yl−1, yl] (if l ≤ k, l is odd), gk−l+2(z1, . . . , zk−l+2)[z, y1][y2, y3] · · · [yl−1, yl] (if l ≤ k, l is odd), where the gm = gm(z1, . . . , zm) are some polynomials in the odd variables z1, . . . , zm. Remark 2. We note that among the three Z2-gradings defined above, the Grass- mann algebra satisfies a graded monomial identity only if its Z2-grading is induced by deg(·)k∗ . Remark 3. It is easily verified that a basis for the relatively free Z2-graded algebra of E with the grading induced by deg(·)∞ is the following: yi1 yi2 · · · yin zj1 · · · zjm [xl1 , xl2 ] · · · [xl2s−1 , xl2s ], where n ≥ 0, i1 ≤ · · · ≤ in, m ≥ 0, j1 ≤ · · · ≤ jm, s ≥ 0 and l1 < · · · < l2s. On the other hand, a basis for the relatively free Z2-graded algebra of E with the grading induced by deg(·)k∗ is the following: yi1 yi2 · · · yin zj1 · · · zjm [xl1 , xl2 ] · · · [xl2s−1 , xl2s ], where n ≥ 0, i1 ≤ · · · ≤ in, 0 ≤ m ≤ k, j1 ≤ · · · ≤ jm, s ≥ 0 and l1 < · · · < l2s. 4. The graded Thorem of Lewin We resume the work of Di Vincenzo and La Scala (see [7]) for the generalization of the Theorem of Lewin (see [15]) in the graded case. In what follows the grading group is supposed to be finite. Let A, B be G-graded algebras and M be an A,B-bimodule, then it is possible to consider the G-algebra R =(cid:18) A M 0 B (cid:19) . Let us call {xi}'s the generators of the relatively free graded algebra of A and B. Then the following result holds. 6 LUCIO CENTRONE AND THIAGO CASTILHO DE MELLO Proposition 4. Let A and B be GPI-algebras. If M contains a countable free set {ui} of homogeneous elements such that deg(xi) = deg(ui) for any i ≥ 1, then TG(R) = TG(A)TG(B). We fix a map · : {1, 2, · · · , n} → G. Then · induces a grading on Mn(F ) by setting eij = ji−1, for all matrix units eij. Indeed this is an elementary grading defined by (1, . . . , n). If we assume A to be a matrix algebra, i.e., A ⊆ Mn(F ), with an elementary G-grading, the authors introduce the notion of G-regularity. In the case G is abelian, we have the following equivalence result. Theorem 5. The G-algebra A is G-regular if and only if the map · is surjective and all its fibers are equipotent. Note that the G-regularity of A is verified when the order of G is exactly n and the map · is bijective. This is the case, for instance, when we consider the Di Vincenzo and Vasilovsky Zn-grading of Mn(F ) (see [6] and [20]). Moreover, for the ordinary case, that is for G = {1G}, the algebra A is regular. We close the section with the main result of [7]. Theorem 6. Let R be the G-graded block-triangular matrix algebra defined as above, where A ⊆ Mn(F ), B ⊆ Mm(F ) are G-algebras and M is an A,B-bimodule. If one between A and B is G-regular, then the TG-ideal TG(R) factorizes as: TG(R) = TG(A)TG(B). Corollary 7. Let A11 A12 A22 ... 0 0 ... 0 . . . A1m . . . A2m ... . . . . . . Amm R =    be a G-graded subalgebra of some matrix algebra. Suppose that each Aii is a G- regular G-graded subalgebra of some G-graded matrix algebra Mdi(F ) and that Aij = Mdi×dj (F ), for each i and j. Then TG(R) = TG(A11)TG(A22) · · · TG(Amm). Proof. We prove the statement by induction on n. If n = 2, we are in the hypothesis of Theorem 6 and we are done. Suppose the assertion true for m − 1, where m ≥ 3. We consider A =  then 0 ... 0 A11 A12 A22 . . . A1m−1 A2m−1 . . . . . . . . . . . . Am−1m−1 ... ,   R =(cid:18) A M 0 Amm (cid:19) . Due to the fact that R is a G-algebra, we have A is a G-algebra too. Now we are again in the hypothesis of Theorem 6 and the proof follows. (cid:3) RELATIVELY FREE GRADED ALGEBRAS 7 5. A model for the relatively free graded algebra of block-triangular matrices In this section we construct a model for the relatively free graded algebra of U T (d1, . . . , dm; A), where A is a GPI-algebra. We recall that in the finite dimen- sional case there is a standard way to construct such a model. For more details we refer to the book of Rowen [18]. Let A be a PI G-algebra over a field F . If d1, d2, . . . , dm are positive integers, we denote by U T (d1, . . . , dm; A) the subalgebra of the matrix algebra Md1+···+dm(A) consisting of matrices of the type A11 A12 A22 ... 0 0 ... 0 . . . A1m . . . A2m ... . . . . . . Amm   ,   where Aij ∈ Mdi×dj (A) for each i, j. One such algebra is called the algebra of block-triangular matrices of size d1, . . . , dm over A. In what follows we are going to construct a model for the relatively free graded algebra of U T (d1, . . . , dm; A), where A is any GPI-algebra. We shall use the following notation: if f (x1, . . . , xn) is a graded polynomial of F hXGi, we shall indicate by x the string of the homogeneous indeterminates ap- pearing in f , i.e., x = (x1, . . . , xn), and we shall write f (x) instead of f (x1, . . . , xn). Moreover, if we are dealing with any graded substitution of the type xdeg(x1) 7→ adeg (x1) 1 , we shall indicate the valuation of f by f (a). 1 The model is based on the following construction. For each k ∈ N and g ∈ G, we define the matrix ξ(g) k ∈ U T (d1, . . . , dm; UG(A)) by ξ(g) k =   d1×d2,k d2×d2,k d1×d1,k B(g) B(g) B(g) 0 ... ... 0 0 d1×dm,k d2×dm,k · · · B(g) · · · B(g) . . . · · · B(g) ... dm×dm,k   Here, each B(g) dr×ds,k is a dr×ds matrix whose entry (i, j) is x(g) ij,k +TG(A) ∈ UG(A), with d1 + · · · + dr−1 + 1 ≤ i ≤ d1 + · · ·+ dr and d1 + · · ·+ ds−1 + 1 ≤ j ≤ d1 + · · ·+ ds. We denote by U(d1, . . . , dm; A) the subalgebra of U T (d1, . . . , dm; UG(A)), gen- k , k ∈ N, g ∈ G, defined above (we omit an index G in erated by the matrices ξ(g) order to simplify the notation). Lemma 8. The algebra U(d1, . . . , dm; A) is a generic model for the relatively free graded algebra of U T (d1, . . . , dm; A), i.e., U(d1, . . . , dm; A) ∼= F hXGi TG(U T (d1, . . . , dm; A)) . Proof. Let X = Sg∈G X g, where the union is disjoint and each X g is a countable set of homogeneous indeterminates. Define the homomorphism ϕ : F hXGi −→ U(d1, . . . , dm; A) x(g) k 7→ ξ(g) k . 8 LUCIO CENTRONE AND THIAGO CASTILHO DE MELLO Of course ϕ is a graded homomorphism onto U(d1, . . . , dm; A). We shall show ker ϕ = TG(U T (d1, . . . , dm; A)). First, observe that ker ϕ ⊆ TG(U T (d1, . . . , dm; A)). Indeed, if f = f (x) ∈ ker ϕ, then ϕ(f ) = f (ξ) = 0. Since UG(A) is the relatively free algebra of A, we have that for any {a(g) ij,k} ⊆ A, there exists a graded homomorphism UG(A) −→ A, such that x(g) ij,k. Since f (ξ) = 0 in U(d1, . . . , dm; A), the image of any entry of the matrix f (ξ) under any homomorphisms UG(A) −→ A is zero. Then f ∈ TG(U T (d1, . . . , dm; A)). ij + TG(A) 7→ a(g) In order to show the reverse inclusion, we take f (x) ∈ F hXGi a graded poly- nomial identity for U T (d1, . . . , dm; A) and we consider the matrix M = f (ξ). Each entry (r, s) of M has the form mrs(x(g) ij,k) + TG(A), for some polynomials mrs ∈ F hXGi, with 1 ≤ r, s ≤ d1 + · · · + dm. We claim that M = 0. Indeed, since f (u) = 0, for any u ∈ U T (d1, . . . , dm; A), we have that mrs(a(g) ij,k) = 0, for any a(g) ij,k ∈ A and this shows that mrs ∈ TG(A), for any r and s, and hence M = 0, which concludes the lemma. (cid:3) Remark 9. Notice that the above construction is still valid if we consider algebras over arbitrary fields. Now we present some applications of the above construction. In particular, we are interested in studying the "factoring" property of T-ideals. Lemma 10. Let A and B be GPI-algebras such that TG(A) = TG(B). Then TG(U T (d1, . . . , dm; A)) = TG(U T (d1, . . . , dm; B)). As a consequence, for n ∈ N, we have TG(Mn(A)) = TG(Mn(B)). Proof. Let ξ(g) and ηk be the generators of U(d1, . . . , dm; B). Let k be the generators of U(d1, . . . , dm; A) as in the previous theorem, f (x) ∈ TG(U T (d1, . . . , dm; A)), then we claim that f ∈ TG(U T (d1, . . . , dm; B)). Let mrs(x(g) (r, s) of the matrix f (ξ), where mrs are polynomials in the variables x(g) f ∈ TG(U T (d1, . . . , dn)), by Lemma 8 we have f (ξ) = 0, then mrs(x(g) 0, i.e., mrs ∈ TG(A) = TG(B). Hence, mrs(x(g) Then we have f (η) = (mrs(x(g) The other inclusion is analogous. ij,k +TG(A)) be the entry ij,k. Since ij,k + TG(A)) = ij,k + TG(B)) = 0, for every (r, s). ij,k + TG(B))) = 0, and f ∈ TG(U T (d1, . . . , dm; B)). (cid:3) Theorem 11. Let G be a finite abelian group and A be a GPI-algebra such that TG(A) = TG(Mn(F )), where Mn(F ) is G-regular. Then for any set of positive integers {d1, . . . , dm}, we have TG(U T (d1, . . . , dm; A)) = TG(Md1(A))TG(Md2(A)) · · · TG(Mdm(A)). Proof. If TG(A) = TG(Mn(F )), by Lemma 8 we have TG(U T (d1, . . . , dm; A)) = TG(U T (d1, . . . , dm; Mn(F ))). In light of the fact that Mn(F ) is G-regular, the grading is induced by the n-tuple g = (g1, . . . , gn). The n-tuple g gives rise to the map · : {1, . . . , n} → G such that RELATIVELY FREE GRADED ALGEBRAS 9 its fibers are equipotent. It is easy to see that Mk(Mn(F )) is G-graded isomorphic to Mnk(F ), for any k, where the G-grading is induced by the kn-tuple g′ = (g1, . . . , gn, g1, . . . , gn, . . . , g1, . . . , gn). Hence TG(U T (d1, . . . , dm; Mn(F ))) = TG(U T (nd1, . . . , ndm; F ))). Since each of the Mndi(F )'s is G-regular, it follows by Corollary 7 that the right- hand side of the above equation is equal to TG(Mnd1(F )) · · · T (Mndm(F )). By Lemma 10, for each i, we have TG(Mndi(F )) = TG(Mdi(Mn(F ))) = TG(Mdi(A)), and the result follows. (cid:3) 6. The factoring property In this section we shall use the model we constructed in the previous section in order to obtain a "factoring" theorem for block-triangular matrices with entries from a G-graded algebra A, whose relatively free graded algebra has a partially multiplicative basis. As a consequence, we show that the Grassmann algebra E, when Z2-graded by the grading inherited by the natural Z2-grading of E, has the factoring property. Let A be a GPI-algebra and consider the automorphism ϕ of UG(A) defined by ϕ(x(g) ij,k) = x(g) i+1j+1,k, where the sum on the indexes is taken modulo n. Roughly speaking, we consider the automorphism of UG(A), which sends each entry (i, j) of a generic matrix to the next variable in its diagonal. The next lemma shows how the entries of the elements of U(n; A) behave with respect to this automorphism. Lemma 12. Let A be a GPI-algebra and M = (mij ) ∈ U(n; A). Then, for each i and j, we have ϕ(mij ) = mi+1j+1. Proof. We use the same notations of Lemma 8. It is enough to prove the lemma for monomials and we do it by induction on the degree of such monomials. We consider the monomial f = f (B1, . . . , Br), where the Bt's are the generic matrices in U(n; A). If f has degree one, then f is a scalar multiple of one of such generic matrices and the result holds. Suppose the result is true for monomials of degree strictly less than m. If f has degree m, we write f = gBk, where g has degree m − 1 and g = g(B1, . . . , Br) = (pij). If Bk = (x(g) As a consequence, the next result shows that in order to verify whether the elements of UG(n; A) are linearly independent, it is enough to verify the linear independence of their k-th columns. t+1j+1,k that is the (i + 1, j + 1) entry of f , and we are done. ij,k), we have f = (cid:16)Pn tj,k) = Pn t=1 pitx(g) t=1 pitx(g) t=1 ϕ(pit)ϕ(x(k) tj,k(cid:17). The tj ) = (cid:3) (i, j) entry of f is Pn Pn t=1 pi+1t+1x(g) t=1 pitx(g) tj,k. Hence ϕ(Pn 10 LUCIO CENTRONE AND THIAGO CASTILHO DE MELLO Corollary 13. Let A be a GPI-algebra, UG(n; A) be the graded generic algebra of Mn(A) and {f1, . . . fn} ⊆ UG(n; A). If f k is the k-th column of fi then the set of i column vectors with entries from UG(A), {f k 1 , . . . , f k n} is linearly independent over F if and only if {f1, . . . fn} is linearly independent over F . Remark 14. With the above notations, one can see that for each k, the map πk : n Xi,j=1 aijeij 7→ aikeik n Xi=1 is an injective map from UG(n; A) into Mn(UG(A)). Before stating our main theorem, we introduce the concept of partially multi- plicative basis. Definition 15. If A is a GPI-algebra, we say that the basis B of UG(A) is partially multiplicative, if for any S1, S2 ⊆ B such that the elements of S1 and the elements of S2 are polynomials in disjoint sets of graded variables, the set S1S2 = {s1s2 s1 ∈ S1, s2 ∈ S2} is linearly independent over F . Theorem 16. Let A be a GPI-algebra, graded by a group G such that Supp(A) = G. Suppose that its relatively free graded algebra, UG(A), has a partially multiplicative basis. If d1, . . . , dm are positive integers, then TG(U T (d1, . . . , dm; A)) = TG(Md1(A)) · · · TG(Mdm(A)). Proof. We prove the theorem by induction on m. If m = 1 the result is obvious. Suppose the assertion true for m − 1, where m > 1. We use the same notation of the proof of Lemma 8. By Lemma 8 we have TG(U T (d1, . . . , dm; A)) = TG(U(d1, . . . , dm; A)). The algebra U(d1, . . . , dm; A) is generated by the matrices d1×d2,k d2×d2,k d1×d1,k B(g) B(g) B(g) 0 ... ... 0 0 d1×dm,k d2×dm,k · · · B(g) · · · B(g) . . . · · · B(g) ... dm×dm,k   , k ≥ 1, where g ranges over G. We consider now the algebra C generated by the matrices ξ(g) k =  k =  ω(g) d1×d2,k d2×d2,k B(g) d1×d1,k B(g) B(g) 0 ... ... 0 0 · · · · · · . . . · · · B(g) B(g) B(g) d1×dm−1,k d2×dm−1,k ... dm−1×dm−1,k , k ≥ 1,   and the algebra B generated by the matrices η(g) dm×dm,k, for k ≥ 1 and g ∈ G. We recall that by Lemma 8, the algebra C is a relatively free G-graded algebra and B is a relatively of U T (d1, . . . , dm−1; A) freely generated by the matrices ω(g) k free algebra of Mdm(A), freely generated by the matrices η(g) k . k = B(g) RELATIVELY FREE GRADED ALGEBRAS 11 Let d = d1 + · · · + dm−1, and let M be the (C, B)-bimodule generated by the d × dm matrices µ(g) k =  d1×dm,k d2×dm,k B(g) B(g) ... B(g) dm−1×dm,k   , k ≥ 1, with action given by the usual product of matrices. By Proposition 4 (the graded theorem of Lewin), if M is freely generated by the µ(g) k , we have TG(U(d1, . . . , dm; A)) = TG(U(d1, . . . , dm−1; A))TG(U(dm; A)) and the proof follows by induction. Due to the fact that Supp(A) = G, we have that the µ(g) to show that the bimodule M is freely generated by the µ(g) finite subset of {µ(g) k 's are non-zero. In order k , we take {µ1, . . . , µm} a k }, and we suppose {f1, . . . , fn} ⊆ B to be a linearly independent j=1 hijµjfi = 0, then i=1 hijµjfi) = 0, and each µj depends on i=1 hij µjfi depends on j=1 hijµjfi) = 0, we also have that i=1 hijµjfi = 0, for each j, then it is enough to prove the assertion for only one of the µj, say µ, i.e., we need to prove that if {f1, . . . , fn} ⊆ B is linearly independent set over F and {hij} ⊆ C. We need to show that if Pn i=1Pm j=1 (Pn each hij is zero in C. We have Pm disjoint sets of variables. Then for each j ∈ {1, . . . , m}, Pn i=1(Pm a different set of variables. Hence, if Pn Pn over F and {h1, . . . , hn} ⊆ C is such that Pn We put d = d1 + · · · + dm−1 and rename the variables x(g) i=1 hiµfi = 0, then each hi is zero. j ≤ d + dm from µ(g) k by y(g) d + 1 ≤ i, j ≤ d + dm from η(g) For each t, ht = ht(ω(g1) 1 ij,k, 1 ≤ i ≤ d, d + 1 ≤ ij,k, 1 ≤ i ≤ d, 1 ≤ j ≤ dm and the variables x(g) ij,k, k by z(g) , . . . , ω(gt) ) is a matrix of the form (hrs,t), where each ij,k, for 1 ≤ i, j ≤ d and k ≥ 1. ij,k, 1 ≤ i, j ≤ dm. kt hrs,t = hrs,t(x(g) The matrix µ has the form (ypq), 1 ≤ p ≤ d and 1 ≤ q ≤ dm. ij,k) is a polynomial in the variables x(g) With this notation, multiplying matrices we have htµ = d Xl=1 hrl,tyls!1≤r≤d, 1≤s≤dm . If ft = ft(η(g1) , . . . , η(gt) 1 fpq,t is a polynomial in the variables z(g) t, we have nt ), we can write it as a matrix of the form (fpq,t), where ij,k, 1 ≤ i, j ≤ dm and k ≥ 1. Then for each htµft = dm Xk=1 d Xl=1 hrl,tylkfks,t!1≤r≤d, 1≤s≤dm . t=1 htµft = 0, we obtain If Pn 0 = n Xt=1 htµft = n Xt=1 dm Xk=1 d Xl=1 hrl,tylkfks,t!1≤r≤d, 1≤s≤dm . 12 LUCIO CENTRONE AND THIAGO CASTILHO DE MELLO Hence each entry of the above matrix is zero, i.e., for 1 ≤ r ≤ d and 1 ≤ s ≤ dm, n dm d Xt=1 Xk=1 Xl=1 hrl,tylkfks,t = 0. Since the polynomials hlr,t depend only on the variables x's, and the polynomials fks,t depend only on the variables z's, comparing the degree of the variables ykl in the above sum, we have that for each k and l hrl,tylkfks,t = 0. n Xt=1 For r = s = l = 1 we have the system of equations h11,ty1sfs1,t = 0 1 ≤ s ≤ d. Since {f1, . . . , fn} is linearly independent over F , Corollary 13 implies that the n Xt=1 f11,t ... fd1,t set       , 1 ≤ t ≤ n  is also linearly independent over F . For each t ∈ {1, . . . , n}, and s ∈ {1, . . . , d} write ⊆ Md×1(UG(A)) fs1,t = α(i) s1,tbi and h11,t = β(j) t cj, m Xi=1 Xj=1 r where bi and cj are elements of a partially multiplicative basis of UG(A). Since the f ′s and the h′s depend on disjoint sets of variables, the set {cjy1sbi, i, j} is linearly independent in UG(A). This means that in the above system of equations, we have n r m that is equivalent to Xi=1 Xj=1 Xt=1 Xi=1 n Xt=1 m r Xj=1 α(i) s1,tβ(j) t cjy1sbi = 0, 1 ≤ s ≤ d, α(i) s1,tβ(j) t ! cjy1sbi = 0, 1 ≤ s ≤ d. Since the set {cjy1sbi} is linearly independent over F , the above system is equiv- alent to (1) n Xt=1 α(i) s1,tβ(j) t = 0 for each 1 ≤ s ≤ d, 1 ≤ i ≤ m, 1 ≤ j ≤ r. It is enough to prove that the only solution of the above system is β(j) t = 0, for 1 ≤ t ≤ n and 1 ≤ j ≤ r. RELATIVELY FREE GRADED ALGEBRAS 13 Fix 1 ≤ j ≤ m; since the set f11,t ... fd1,t       , 1 ≤ t ≤ n  ⊆ Md×1(UG(A)) is linearly independent over F , n Xt=1 β(j) t   f11,t ... fd1,t   = 0 if and only if β(j) to the system of equations (1). t = 0 for every t. We show now that the above equation is equivalent Substituting the expressions for the fs1,t, the above equation is equivalent to β(j) 1 (α(1) β(j) 1 (α(1) 11,1b1 + · · · + α(m) 21,1b1 + · · · + α(m) 11,1bm) + · · · + β(j) 21,1bm) + · · · + β(j) n (α(1) n (α(1) 11,nb1 + · · · + α(m) 21,1b1 + · · · + α(m) 11,nbm) 21,1bm) ... β(j) 1 (α(1) d1,1b1 + · · · + α(m) d1,1bm) + · · · + β(j) n (α(1) d1,1b1 + · · · + α(m) d1,1bm) = 0.         Factoring the bi, we obtain 11,1 + · · · + β(j) 21,1 + · · · + β(j) 1 α(1) 1 α(1) (β(j) (β(j) n α(1) n α(1) 11,n)b1 + · · · + (β(j) 21,n)b1 + · · · + (β(j) 1 α(m) 1 α(m) 11,1 + · · · + β(j) 21,1 + · · · + β(j) n α(m) n α(m) 11,n)bm 21,n)bm = 0. ... (β(j) 1 α(1) d1,1 + · · · + β(j) n α(1) d1,n)b1 + · · · + (β(j) 1 α(m) d1,1 + · · · + β(j) n α(m) d1,n)bm Due to the fact that the bi are linearly independent over F , we conclude that t α(i) β(j) s1,t = 0, for each 1 ≤ s ≤ d, 1 ≤ i ≤ m, 1 ≤ j ≤ r n Xt=1 that is exactly the system of equations (1), i.e., we have shown that the only solution to the system (1) is the trivial one. As a consequence, we have that h11,t = 0, for every 1 ≤ t ≤ n. Similar calculations show that hpq,t = 0, for every 1 ≤ p, q ≤ d and 1 ≤ t ≤ n, (cid:3) which concludes the proof of the theorem. 7. Conclusions We observe that the model we presented in this paper transfers the informa- tion regarding the graded identities of a graded algebra A into the ideal of graded polynomial identities of a block-triangular matrix U T (d1, . . . , dm; A). We may ob- serve that the map deg(·)∞ induces a grading over E such that its relatively free Z2-graded algebra has a partially multiplicative basis. We can state the following. Theorem 17. Let E be the Grassmann algebra endowed with the Z2-grading in- duced by the map deg(·)∞. If d1, . . . , dm are positive integers, then TZ2(U T (d1, . . . , dm; E)) = TZ2(Md1(E)) · · · TZ2(Mdm(E)). 14 LUCIO CENTRONE AND THIAGO CASTILHO DE MELLO As we already noted in Remark 2, the only homogeneous Z2-grading over E that gives rise to a monomial identity is the one induced by the map deg(·)k∗ . Indeed, the relatively free Z2-graded algebra does not have a partially multiplicative basis. In fact, the Theorems 16 and 17 cannot be generalized. We have the following. Proposition 18. Let R be the Z2-graded algebra where the Z2-grading is induced by that of E. If E is graded by deg(·)k∗ , then R :=(cid:18) E E 0 E (cid:19) , TZ2(E)TZ2 (E) TZ2(R). Proof. It is easy to see that z1 · · · zk+1 is a graded identity of R but it is clearly not a consequence of the product TZ2(E)TZ2 (E) because TZ2(E) contains the identity z1 · · · zk+1 (see Theorem 1). (cid:3) It is well known that the (ungraded) Grassmann algebra, E, has a partially mul- tiplicative basis for its relatively free-graded algebra. As a consequence, Theorem 16 gives that U T (d1, . . . , dm; E) has the factoring property for its T-ideal. A natural question arises: "what are the PI-algebras such that their relatively free algebras do not have a partially multiplicative basis?" References [1] Yu. Bahturin, A. Giambruno, D. Riley, Group-graded algebra with polynomial identity, Isr. J. Math. 104 (1998), 145-155. [2] A. Belov, Local finite basis property and local representability of varieties of associative rings, Izvestia of Russian Academia of science 1 (2010), 3-134. [3] A. Berele, Generic verbally prime PI-algebras and their GK-dimensions, Commun. Algebra 21(5) (1993), 1487-1504. [4] A. Berele, A. Regev, Exponential growth for codimensions of some p.i. algebras, J. Algebra 241(1) (2001), 118-145. [5] J. Bergen, M. Cohen, Action of commutative Hopf algebras, Bull. London Math. Soc. 18 (1986), 159-164. [6] O. M. Di Vincenzo, On the graded identities of M1,1(E), Isr. J. Math. 80 (1992), 323-335. [7] O. M. Di Vincenzo, R. La Scala, Block-triangular matrix algebras and factorable ideals of graded polynomial identities, J. Algebra 279 (2004), 260-279. [8] O. M. Di Vincenzo, V. R. T. Da Silva, On Z2-graded polynomial identities of the Grassmann algebra, Linear Algebra Appl. 431(1-2) (2009), 56-72. [9] V. Drensky, Extremal varieties algebras II, Serdica 14 (1988), 20-27 (Russian). [10] V. Drensky, Extremal varieties algebras I, Serdica 13 (1987), 320-332 (Russian). [11] A. Giambruno, M. V. Zaicev, Polynomial Identities and Asymptotic Methods, Math. Surveys and Monog. 122 (2005), AMS. [12] A. Giambruno, M. V. Zaicev, Codimension growth and minimal superalgebras, Trans. Amer. Math. Soc. 355 (2003), 5091-5117. [13] I. N., Herstein, Noncommutative rings. The Carus Mathematical Monographs, 15. Published by The Mathematical Association of America; distributed by John Wiley Sons, Inc., New York 1968. [14] A. R. Kemer, Varieties and Z2-graded algebras (Russian), Izv. Akad. Nauk SSSR, Ser. Mat. 48 (1984), 1042-1059. Translation: Math. USSR, Izv. 25 (1985), 359-374. [15] J. Lewin, A matrix representation for associative algebras I, Trans. Amer. Math. Soc. 188 (1974), 293-308. [16] C. Nastasescu, F. van Oystaeyen, Graded ring theory, Amsterdam (1982). [17] A. Regev, Existence of identities in A ⊗ B, Isr. J. Math. 11 (1972), 131-152. [18] L. H. Rowen, Polynomial identities in ring theory, Pure Applied Math., 1980, Academic Press, New York. RELATIVELY FREE GRADED ALGEBRAS 15 [19] C. Procesi, Non-commutative affine rings. Atti Accad. Naz. Lincei Mem. Cl. Sci. Fis. Mat. Natur. Sez. I (8) 8 (1967), 237-255. [20] S. Y. Vasilovsky, Zn-graded polynomial identities of the full matrix algebra of order n, Proc. Amer. Math. Soc. 127(12) (1999), 3517-3524. IMECC, Universidade Estadual de Campinas, Rua S´ergio Buarque de Holanda, 651, Cidade Universit´aria "Zeferino Vaz", Distr. Barao Geraldo, Campinas, Sao Paulo, Brazil, CEP 13083-859 E-mail address: [email protected] Instituto de Matematica e Estat´ıstica, Universidade de Sao Paulo Caixa Postal 66281, Sao Paulo, SP, 05315-970 Brasil E-mail address: [email protected]
1711.05163	1	1711	2017-11-14T16:12:13	On congruence-semisimple semirings and the $K_0$-group characterization of ultramatricial algebras over semifields	[ "math.RA", "math.KT" ]	In this paper, we provide a complete description of congruence-semisimple semirings and introduce the pre-ordered abelian Grothendieck groups $K_0(S)$ and $SK_0(S)$ of the isomorphism classes of the finitely generated projective and strongly projective S-semimodules, respectively, over an arbitrary semiring S. We prove that the $SK_0$-groups and $K_0$-groups are complete invariants of, i.e., completely classify, ultramatricial algebras over a semifield F. Consequently, we show that the $SK_0$-groups completely characterize zerosumfree congruence-semisimple semirings.	math.RA	math	ON CONGRUENCE-SEMISIMPLE SEMIRINGS AND THE K0-GROUP CHARACTERIZATION OF ULTRAMATRICIAL ALGEBRAS OVER SEMIFIELDS YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Abstract. In this paper, we provide a complete description of congruence- semisimple semirings and introduce the pre-ordered abelian Grothendieck groups K0(S) and SK0(S) of the isomorphism classes of the finitely gener- ated projective and strongly projective S-semimodules, respectively, over an arbitrary semiring S. We prove that the SK0-groups and K0-groups are com- plete invariants of, i.e., completely classify, ultramatricial algebras over a semi- field F . Consequently, we show that the SK0-groups completely characterize zerosumfree congruence-semisimple semirings. 1. Introduction As is well-known (see, for example, [5]), projective modules play a fundamental role in developing of algebraic K-theory which, in turn, has crucial outcomes in many areas of modern mathematics such as topology, geometery, number theory, functional analysis, etc. In short, algebraic K-theory is a study of groups of the isomorphism classes of algebraic objects, the first of which is K0(R), Grothendieck's group of the isomorphism classes of finitely generated projective R-modules, and that is used to create a sort of dimension for R-modules that lack a basis. There- fore, the structure theory of projective modules is certainly of a great interest and importance. Semirings, semimodules, and their applications, arise in various branches of mathematics, computer science, quantum physics, and many other areas of science (see, for example, [12] and [11]). As algebraic structures, semirings certainly are the most natural generalization of such (at first glance different) algebraic concepts as rings and bounded distributive lattices, and therefore, they form a very natural and exciting ground for furthering the structure theory of projective (semi)modules in a "non-additive" categorical setting. And, in fact, the structure theory of pro- jective semimodules has been recently considered by several authors (see, e.g., [25], [34], [27], [19], [23], [33], [22] and [20]). Also, in the last one or two decades, there can be observed an intensively growing substantial interest in additively idempo- tent semirings, which particularly include the Boolean and tropical semifields and have a fundamental meaning in such relatively new, "non-traditional", and fasci- nating areas of modern mathematics such as tropical geometry [35] and [10], tropi- cal/supertropical algebra [21], F1-geometry [6], and the geometry of blueprints [31]. 2010 Mathematics Subject Classification. Primary 16Y60, 16E20, 18G05. The second author is supported by the Vietnam National Foundation for Science and Technol- ogy Development (NAFOSTED) under Grant 101.04-2017.19. 1 2 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Although, in general, describing the structure of (finitely generated) projective semimodules seems to be a quite difficult task, recently there have been obtained a number of interesting results regarding structures of projective semimodules over special classes of semirings among which we mention, for example, the following ones. Il'in et al. [20] initiated a homological structure theory of semirings and in- vestigated semirings all of whose cyclic semimodules are projective; Izhakian et al. [23] characterized finitely generated projective semimodules over a tropical semi- field in terms of rank functions of semimodules; and Macpherson [33] classified projective semimodules over additively idempotent semirings that are free on a monoid. Further, motivated by direct sum decompositions of subsemimodules of free semimodules over a tropical semifield and related structures, Izhakian et al. [22] developed a theory of the decomposition socle, dsoc(M ), for zerosumfree semimod- ules M . In particular, they provided a criterion for zerosumfree semirings S when dsoc(S) = S ([22, Thm. 3.3]) and established the uniqueness of direct sum decom- positions for some special finitely generated projective semimodules ([22, Cor. 3.4]), called 'strongly projective' in the present paper, over such semirings. Moreover, Elliott [8] classified/characterized ulramatricial algebras over an arbi- trary field by means of their pointed ordered Grothendieck groups K0. This fun- damental result implies a C∗-algebra technique and initiated very fruitful research lines in algebra and operator algebra, not to mention that the Elliott program of classifying simple nuclear separable C∗-algebras by K-theoretic invariants became a profoundly active area of research (see, e.g., the survey paper by Elliott and Toms [9]). In light of the two previous paragraphs and motivated by the Elliott program of classifying C∗-algebras in terms of K-theory, our paper has a twofold goal: to characterize the decomposition socles and structure of (finitely generated) projec- tive semimodules over a semiring S in terms of the Grothendieck group K0(S) of a semiring S; and to extend Elliott's classification of ultramatricial algebras to a "non-additive" semiring setting. Let us a briefly clarify the latter: If F is a (semi)field and C is a class of unital F -algebras, then one says that the K0- group is a complete invariant for algebras in C, or that K0 completely classifies F -algebras in C, if any F -algebras R and S from C are isomorphic as F -algebras iff there is a group isomorphism K0(R) ∼= K0(S) which respects the natural pre-order structure of the K0-groups and their order-units. It should be mentioned that the "blueprints" of Lorscheid [31] contain commutative semirings as a full subcategory, which eventually leads to a K-theory of blueprints, including a K-theory of com- mutative semirings as a special case, and the group K0(S) of a semiring S has been introduced by Di Nola and Russo [7]. However, the considerations of K0(S) in our paper are distinguished from those in [31] and [7] -- we consider two quite different types of K0-groups and, to the extend of our knowledge, at the first time use them as complete invariants for classifying algebras of a non-additive category in the spirit of the Elliot program. The article is organized as follows. In Section 2, for the reader's convenience, we briefly collect the necessary notions and facts on semirings and semimodules. Subsequently, we provide in Section 3 a full description of congruence-semisimple semirings (Theorem 3.4) and show that zerosumfree congruence-semisimple semir- ings are precisely matricial algebras over the Boolean semifield B (Corollary 3.5). In Section 4, beyond of some basic considerations of strongly projective semimodules CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 3 under change of semirings (Propositions 4.8 and 4.10), we give a complete descrip- tion of the strongly projective semimodules over an arbitrary semisimple semiring (Theorems 4.5 and 4.9). Based on the results of Section 4 and [13, Ch. 15], in Section 5 we introduce and establish fundamental properties of the monoids V(S) (SV(S)) of the isomorphism classes of finitely generated (strongly) projective semimodules over a semiring S and show that those monoids completely characterize the class of ultramatricial algebras over a semifield F (Theorems 5.10 and 5.11). In Section 6, using the results of Sections 4 and 5, we consider the pre-ordered abelian groups K0(S) and SK0(S) -- which are the Grothendieck groups on the monoids V(S) and SV(S), respectively -- for an arbitrary semiring S, and, using the concept of 'weak dimension' of semimodules, describe division semirings D having the groups K0(D) and SK0(D) to be isomorphic (Theorem 6.10). Also, it is shown (Proposition 6.7) that, for any additively idempotent commutative semiring S, the group K0(S) always contains a free abelian group with countably infinite basis; and it is given (Theorem 6.14) for semirings S having dsoc(S) = S, i.e., congruence- semisimple semirings here, a K-theory version of [22, Cor. 3.4]. Finally, we extend Elliott's classification theorem for ultramatricial algebras over fields [8] and show that the SK0-groups and K0-groups are complete invariants of ultramatricial al- gebras over semifields (Theorems 6.21 and 6.23), as well as that SK0 completely classifies zerosumfree congruence-semisimple semrings (Theorem 6.22). All notions and facts of categorical algebra, used here without any comments, can be found in [32]; for notions and facts from semiring theory we refer to [12]. 2. Preliminaries Recall [12] that a semiring is an algebra (S, +, ·, 0, 1) such that the following conditions are satisfied: (1) (S, +, 0) is a commutative monoid with identity element 0; (2) (S, ·, 1) is a monoid with identity element 1; (3) multiplication distributes over addition from either side; (4) 0s = 0 = s0 for all s ∈ S. Given two semirings S and S′, a map ϕ : S → S′ is a homomorphism if it satisfies ϕ(0) = 0, ϕ(1) = 1, ϕ(x + y) = ϕ(x) + ϕ(y) and ϕ(xy) = ϕ(x)ϕ(y) for all x, y ∈ S. A semiring S is commutative if (S, ·, 0) is a commutative monoid; and S is entire if ab = 0 implies that a = 0 or b = 0 for all a, b ∈ S. The semiring S is a division semiring if (S \ {0}, ·, 1) is a group; and S is a semifield if it is a commutative division semiring. An element e in a given semiring S is idempotent if e2 = e; and an idempotent e ∈ S is strong if there exists an idempotent f ∈ S such that e+f = 1 and ef = 0 = f e. Two idempotents e, f ∈ S are orthogonal if ef = 0 = f e. An idempotent is primitive if it cannot be written as the sum of two nonzero orthogonal idempotents. As usual, a right S-semimodule over a given semiring S is a commutative monoid (M, +, 0M ) together with a scalar multiplication (m, s) 7→ ms from M × S to M which satisfies the identities m(ss′) = (ms)s′, (m + m′)s = ms + m′s, m(s + s′) = ms+ms′, m1 = m, 0M s = 0M = m0 for all s, s′ ∈ S and m, m′ ∈ M . Left semimod- ules over S and homomorphisms between semimodules are defined in the standard manner. An S-semimodule M is called a module if its additive reduct (M, +, 0M ) is an abelian group. Let henceforth M be the variety of commutative monoids, and 4 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL let MS and SM denote the categories of right and left S-semimodules, respectively, over a semiring S. Recall [24, Def. 3.1] the tensor product bifunctor − ⊗ − : MS ×S M → M, which for a right semimodule A ∈ MS and a left semimodule B ∈ SM can be described as the factor monoid F/σ of the free monoid F ∈ M, generated by the Cartesian product A × B, factorized with respect to the congruence σ on F generated by the ordered pairs having the form h(a1 + a2, b), (a1, b) + (a2, b)i, h(a, b1 + b2), (a, b1) + (a, b2)i, h(as, b), (a, sb)i, with a1, a2 ∈ A, b1, b2 ∈ B and s ∈ S. An S-semimodule M is called (additively) idempotent (resp., zerosumfree) if m + m = m for all m ∈ M (resp., if m + m′ = 0 implies m = m′ = 0 for all m, m′ ∈ M ); clearly, every idempotent semimodule is zerosumfree. In particular, a semiring S is additively idempotent (resp., zerosumfree, a ring) if SS ∈ MS as a semimodule is idempotent (resp., zerosumfree, a module). Two well-known impor- tant examples of additively idempotent semirings are the Boolean semifield B := ({0, 1}, max, min, 0, 1) and the tropical semifield T := (R ∪ {−∞}, max, +, −∞, 0). By an S-algebra A over a given commutative semiring S we mean the data of an S-semimodule A and of an associate multiplication on A that is bilinear with respect to the operations of the S-semimodule A. For example, every semiring may be considered as a Z+-algebra and any additively idempotent semiring as a B-algebra. An S-algebra A is called unital if the multiplication on A has a neutral element 1A, i.e., a1A = a = 1Aa for all a ∈ A. As usual (see, for example, [12, Ch. 17]), if S is a semiring, then in the cate- gory MS, a free right semimodule F with basis set I is a direct sum (a coproduct) of I copies of SS, i.e., F = Li∈I Si where Si ∼= SS for i ∈ I. Accordingly, a projective semimodule in MS is defined to be a retract of a free semimodule, i.e., a right semimodule P is called projective if there is a free right semimodule F with homomorphisms f : F → P and g : P → F such that f ◦ g = idP . And a semi- module MS is finitely generated if it is a homomorphic image of a free semimodule with finite basis set. Moreover, a semiring S is said to have the IBN ("invariant basis number") property (cf. [26, Def. 2.8]) if, for any natural numbers n, m, the free semimodules Sm and Sn are isomorphic in MS if and only if m = n. Note that the "left" version of the IBN property is equivalent to this right version, see [26, Prop. 3.1]. Congruences on a right S-semimodule M are defined in the standard manner. Any subsemimodule L of a right S-semimodule M induces a congruence ≡L on M , known as the Bourne congruence, by setting m ≡L m′ iff m + l = m′ + l′ for some l, l′ ∈ L. The subsemimodule L is subtractive if a ∈ L and x + a ∈ L (where x ∈ M ) implies x ∈ L. In this case, the zero class [0] with respect to ≡L coincides with L. A nonzero right S-semimodule M is called congruence-simple if its only congruences are the identity congruence △M := {(m, m) m ∈ M } and the universal congruence M×M -- in this case, its only subtractive subsemimodules are {0} and M . Accordingly, a right ideal I of a semiring S is called congruence-simple if the right S-semimodule I is congruence-simple. And a semiring S is called right (left) congruence-semisimple if S is a direct sum of congruence-simple right (left) ideals. Finally, a nonzero right S-semimodule M is called minimal if it has no proper nonzero subsemimodules, and the S-semimodule M is said to be semisimple if it CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 5 is a direct sum of minimal subsemimodules; in particular, a semiring S is said to be right (left) semisimple if the right (left) regular semimodule is semisimple. As is well-known (see, for example, [16, Thm. 7.8] or [28, Thm. 4.5]), the celebrated Artin-Wedderburn theorem generalized to semirings states that a semiring S is (right, left) semisimple if and only if S ∼= Mn1(D1) × . . . × Mnr (Dr) , where Mni(Di) is the semiring of ni × ni-matrices over a division semiring Di for each i = 1, . . . , r. In the sequel, we refer to such an isomorphism as a direct product representation of a semisimple semiring S. 3. Congruence-semisimple semirings Providing a full description of the class of all congruence-semisimple semirings constitutes a main goal of this section; and to accomplish it, we need the following useful facts. Lemma 3.1 ([18, Prop. 1.2]). If M ∈ MS is a congruence-simple right S- semimodule, then M is either idempotent or a module. Lemma 3.2 ([18, Lem. 1.1]). Let M ∈ MS be an idempotent right S-semimodule. Then the relation x ∼ y ⇐⇒ Ann(x) = Ann(y), where Ann(m) := {s ∈ S ms = 0M } is the annihilator of an element m ∈ M , is a congruence on M . Proposition 3.3. Let S be a semiring. (1) The endomorphism semiring EndS(M ) of any cyclic congruence-simple S- semimodule M ∈ MS is either a division ring or the Boolean semifield B. (2) Let I be a congruence-simple right ideal of a semiring S such that I = eS for some idempotent e ∈ S, and let M ∈ MS be a cyclic congruence-simple S-semimodule. Then, M ∼= I or HomS(M, I) = 0. Proof. (1) By Lemma 3.1, M is either a module or idempotent. If M is a module, then using similar arguments as in the classical Schur lemma [29, Lem. 1.3.6], one readily sees that EndS(M ) is a division ring. Now suppose that M is an idempotent semimodule such that M = mS for some 0 6= m ∈ M , and let f ∈ EndS(M ) be a nonzero endomorphism. It is clear (see also [1, Prop. 2.1 (1)]) that f is injective. If m′ := f (m), then Ann(m) = Ann(m′): Indeed, since f is injective, for all s ∈ S, ms = 0 iff m′s = f (ms) = 0. And, by Lemma 3.2, m = m′ = f (m), thus f = idM , whence EndS(M ) ∼= B. (2) Assume there exists a nonzero homomorphism f : M → I. Again, it is clear (see also [1, Prop. 2.1 (1)]) that f is injective. We claim that f is also surjective, whence M ∼= I. If M is a module, then 0 6= f (M ) is a subtractive submodule of I, hence f (M ) = I as desired. Suppose then that M is not a module, hence M is idempotent by Lemma 3.1. Then 0 6= f (M ) is also idempotent, thus I is not a module either, so that I is idempotent by Lemma 3.1 as well. Now, let M = mS for some 0 6= m ∈ M , and a := f (m) ∈ I. Since f (M ) is a nonzero subsemimodule of I, the Bourne congruence ≡f (M) on I is the universal one. So, e ≡f (M) 0, and hence, e+as = as′ and e+ase = as′e for some s, s′ ∈ S. From the latter and since I is idempotent, one immediately gets that Ann(e) = Ann(as′e); and by Lemma 3.2, e = as′e = f (ms′e), and hence, f is a again surjective. (cid:3) 6 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Recall that a semiring S is right (left) congruence-semisimple if S is a direct sum of its congruence-simple right (left) ideals. Notice that a ring is a right (left) congruence-semisimple ring iff it is a classical semisimple ring, i.e., it is a direct sum of its minimal one-sided ideals. However, in a semiring setting, this fact is not true in general. The following theorem, constituting the main result of this section, gives a full description of all congruence-semisimple semirings and also demonstrates that the class of congruence-semisimple semirings is a proper subclass of the class of semisimple semirings. Theorem 3.4. For any semiring S, the following statements are equivalent: (1) S is a right congruence-semisimple semiring; (2) S ∼= Mn1(B) × · · · × Mnk (B) × Mm1(D1) × · · · × Mmr (Dr), where B is the Boolean semifield, D1, . . . , Dr are division rings, k ≥ 0, r ≥ 0, and ni (i = 1, . . . , k) and mj (j = 1, . . . , r) are positive integers; (3) S is a left congruence-semisimple semiring. Proof. (1) =⇒ (2). Let S be a right congruence-semisimple semiring, thus S is a finite direct sum of its congruence-simple right ideals. By applying Lemma 3.1 and grouping those summands according to their isomorphism types as right S- semimodules, we obtain SS ∼= I n1 1 ⊕ · · · ⊕ I nk k ⊕ I m1 k+1 ⊕ · · · ⊕ I mr k+r , with mutually nonisomorphic congruence-simple right ideals I1, . . . , Ik+r of S, where the Ii for i = 1, . . . , k are additively idempotent S-semimodules and the additive reducts of Ii for i = k + 1, . . . , k + r are abelian groups. Notice that each Ii for i = 1, . . . , k + r is a direct summand of SS, so Ii = eiS for some idempotent ei ∈ S. By Proposition 3.3, EndS(Ii) ∼= B for i = 1, . . . , k, Dj := EndS(Ik+j ) for j = 1, . . . , r are division rings, and HomS(Ii, Ij ) = 0 for all distinct i, j = 1, . . . , k + r. Since elements of EndS(SS) are presented by multiplications on the left by ele- ments of S, and as HomS(Ii, Ij) = 0 for i 6= j, we infer S ∼= EndS(SS) ∼= EndS(I n1 ∼= EndS(I n1 1 ) × · · · × EndS(I nk 1 ⊕ · · · ⊕ I nk k ⊕ I m1 k ) × EndS(I m1 k+1 ⊕ · · · ⊕ I mr k+1) × · · · × EndS(I mr k+r) m+r) . Finally, by noting that EndS(M m) ∼= Mm(EndS(M )) for any M ∈ MS and postive integer m, we conclude that S ∼= Mn1(B) × · · · × Mnk (B) × Mm1(D1) × · · · × Mmr (Dr) . (2) =⇒ (1). It suffices to show the congruence-semisimpleness of a matrix semiring S := Mn(K) with K to be either a division ring or the Boolean semi- field. To this end, let eii for i = 1, . . . , n be the matrix units in Mn(K), so that S = e11S ⊕ · · · ⊕ ennS with eiiS ∼= K n as right S-semimodules for each i. As was shown in [25, Thm. 5.14], the functors F : MS ⇆ MK : G given by F (A) = Ae11 and G(B) = Bn establish an equivalence of the semimodule cate- gories MS and MK. Therefore, taking into consideration that K is a congruence- simple right K-semimodule and [1, Lem. 3.8], we have that each eiiS ∼= K n = G(K) for i = 1, . . . , n is a congruence-simple right S-semimodule as well, whence S is a right congruence-semisimple semiring. The equivalence (1) ⇐⇒ (3) follows by symmetry. (cid:3) CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 7 Recently in [22], introducing and studying the decomposition socle for semimod- ules over zerosumfree semirings, the authors characterize zerosumfree semirings R such that the regular semimodule RR is a finite direct sum of indecomposable pro- jective R-subsemimodules (see [22, Thm. 3.3]). In light of this and as a corollary of Theorem 3.4, we note a description of such semirings in the class of congruence- semisimple semirings as matricial algebras -- a central subject of our considerations in the following sections -- over the Boolean semifield. Corollary 3.5. For any zerosumfree semiring S, the following are equivalent: (1) S is a right congruence-semisimple semiring; (2) S ∼= Mn1(B) × · · · × Mnk(B), with B the Boolean semifield and k, n1, . . . , nk some positive integers; (3) S is a left congruence-semisimple semiring. 4. Strongly projective semimodules In [22], a very natural variation of the concept of a projective semimodule has been introduced in a semiring setting, which we here call "strongly projective semi- module". In the present section, we thoroughly investigate such kind of semimod- ules over general semirings, semifields and semisimple semirings. Definition 4.1 (cf. [22, Def. 3.1]). A semimodule P ∈ MS is (finitely generated) strongly projective if it is isomorphic to a direct summand of a (finitely generated) free right S-semimodule. Remark 4.2. We note a few easy facts on strongly projective semimodules. (1) Any (finitely generated) strongly projective semimodule is a (finitely gen- erated) projective semimodule as well, and the concepts of "projectivity" and "strong projectivity" for modules over rings coincide. (2) A strongly projective semimodule P ∈ MS over a zerosumfree semiring S (3) Let (Pi)i∈I be a family of right S-semimodules. Then, the right S-semimodule is zerosumfree as well. Li∈I Pi is strongly projective iff Pi is strongly projective for all i ∈ I. The next observation provides a simple criterion for (strong) projectivity. Lemma 4.3. A finitely generated right S-semimodule P is (strongly) projective iff there exist a positive integer n and a (strongly) idempotent matrix A ∈ Mn(S) such that A(Sn) ∼= P , where A(Sn) is the subsemimodule of the right S-semimodule Sn generated by all column vectors of A. Proof. If P is a finitely generated projective right S-semimodule, then there is some positive integer n and a homomorphism f : Sn → P with a right inverse ho- momorphism g : P → Sn, i.e., f ◦ g = idP . Then α := g ◦ f : Sn → Sn is an idempotent endomorphism with α(Sn) ∼= P . If the semimodule P is in addition strongly projective, there exists, for some n and some right S-semimodule P ′, an isomorphism Sn → P ⊕ P ′. Hence, the corresponding projections are homomor- phisms f : Sn → P , f ′ : Sn → P ′ with right inverses g : P → Sn, g′ : P → Sn, such that α := g ◦ f and α′ := g′ ◦ f ′ are idempotent endomorphisms of Sn satisfying α + α′ = idSn and α ◦ α′ = 0 = α′ ◦ α. Applying now the standard interpretation of endomorphisms of the free right S-semimodule Sn as n × n matrices over S yields a (strongly) idempotent matrix A ∈ Mn(S) such that A(Sn) ∼= P . The converse direction is obvious. (cid:3) 8 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL A description of strongly projective semimodules over semifields is our next goal, for which we need the following useful fact. Recall first (see, e.g., [12, p. 154]) that a subsemimodule K of a semimodule M ∈ MS is said to be strong if m + m′ ∈ K implies m, m′ ∈ K for all m, m′ ∈ M . Proposition 4.4. Let M = Li∈I Ti ∈ MS be a direct sum of minimal S- subsemimodules Ti ∈ MS. Then, for every strong subsemimodule K ⊆ M , there exists a subset IK ⊆ I such that M = K ⊕ (cid:0) L Ti(cid:1) i∈IK and K ∼= L Ti. i∈I\IK Ti) = 0; let N := K + (Li∈IK Proof. By Zorn's lemma, there exists a maximal subset IK ⊆ I satisfying the property K ∩ (Li∈IK Ti). We claim that M = N . Indeed, for any i ∈ I, as Ti is minimal, we have either N ∩ Ti = 0 or N ∩ Ti = Ti. Suppose that N ∩ Ti = 0 for some i ∈ I. Then let J := IK ∪ {i}, and for any x ∈ K ∩ (Li∈J Ti) we can write x = x1 + x2 with x1 ∈ Li∈IK Ti and x2 ∈ Ti. Since K is strong, we get that x1, x2 ∈ K, and thus x1 ∈ K ∩ (Li∈IK Ti) and x2 ∈ K ∩ Ti. From the latter one has x1 = 0 = x2 and thus x = 0. Thus, K ∩ (Li∈J Ti) = 0, contradicting the maximality of the subset IK ⊆ I. Therefore, N ∩ Ti = Ti for all i ∈ I, whence M = N . 2, where x1, x′ Now let x1 + x2 = x′ 1 ∈ K and x2, x′ 2 ∈ Li∈IK Ti. Using the 1 + x′ direct sum (∗) M = (cid:0) L i∈I\IK Ti(cid:1) ⊕ (cid:0) L i∈IK Ti(cid:1), 1 ∈ Li∈I\IK 1 = y′ we can write x1 = y1 + y2 and x′ Li∈IK and x′ x2 = x′ corresponding Bourne congruence, that K ∼= M/Li∈IK Ti. As K is strong, we infer that y2, y′ 1 = y′ 2, whence M = K ⊕ (Li∈IK 2 ∈ Ti) = 0, thus x1 = y1 2. By (∗) this implies x1 = y1 = y′ 1 and Ti). The direct sum (∗) also shows, using the Ti ∼= Li∈I\IK (cid:3) 2 with y1, y′ 2 ∈ K ∩ (Li∈IK 1, so that y1 + x2 = y′ Ti and y2, y′ 1 = x′ 1 + y′ 1 + x′ Ti. Theorem 4.5. Every strongly projective right D-semimodule over a division semir- ing D is free. In particular, a finitely generated semimodule P ∈ MD is strongly projective if and only if there exists a unique nonnegative integer n such that P ∼= Dn. Proof. It is clear that D is either a division ring or a zerosumfree division semiring. Also, the statement is the well-known "classical" result when D is a division ring. So let D be a zerosumfree division semiring and let P ∈ MD be a strongly projective semimodule. There is a free semimodule F ∈ MD, which obviously is zerosumfree, such that F = P ⊕ Q for some semimodule Q ∈ MD, and we claim that P is a strong subsemimodule of F . Indeed, let x, y ∈ F such that x + y ∈ P . Then, x = p + q and y = p′ + q′ for some p, p′ ∈ P and q, q′ ∈ Q, hence x + y = (p + p′) + (q + q′) ∈ P ⊕ Q . Therefore, we have that q + q′ = 0, so that q = 0 = q′, since F is zerosumfree. This implies that x = p ∈ P and y = p′ ∈ P , and thus P is strong. Now noticing that F = Li∈I Di, where Di ∼= D for all i ∈ I, is a direct sum of minimal right D-subsemimodules, by applying Proposition 4.4 we get that P ∼= Li∈J Di for some subset J ⊆ I, whence P is a free semimodule. CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 9 If P ∈ MD is a finitely generated strongly projective semimodule, then there exists a positive integer m such that Dm = P ⊕ Q for some Q ∈ MD. By the observation above, there exists a nonnegative integer n ≤ m such that P ∼= Dn. The uniqueness of such a number n follows from the IBN property of division semirings (see [15, Thm. 5.3] or Corollary 5.2 below). (cid:3) Next we illustrate that the concepts of "projectivity" and "strong projectivity" for semimodules, in general, are quite different. It is clear that any B-semimodule M ∈ MB is an idempotent semimodule and an upper semilattice under the partial ordering ≤ on M defined for any two elements x, y ∈ M by x ≤ y iff x + y = y. Let us recall the following projectivity criterion for B-semimodules. Fact 4.6 ([17, Thm. 5.3]). A B-semimodule M is projective if and only if M is a distributive lattice and {m ∈ M m ≤ x} is finite for all x ∈ M . By applying Theorem 4.5 and Fact 4.6 it is fairly easy to provide counterex- amples demonstrating the difference of the concepts of "projectivity" and "strong projectivity" for general semimodules. Example 4.7. Consider the subsemimodule PB := {(0, 0), (0, 1), (1, 1)} of the free semimodule B2 ∈ MB. By Fact 4.6, the semimodule P is finitely generated projective. However, it is obvious that there is no positive integer n with P ∼= Bn, and therefore, by Theorem 4.5, P is not a strongly projective B-semimodule. Now let us consider strongly projective semimodules under a change of semirings. We need the following two functors introduced in [25]. Given any semirings R, S and a homomorphism π : R → S, every right S-semimodule BS may be considered as a right R-semimodule by pullback along π, i.e., by defining b · r := b · π(r) for any b ∈ B, r ∈ R. The resulting R-semimodule is written π#B, and it is easy to see that the assignment B 7→ π#B naturally constitutes a restriction functor π# : MS → MR. The restriction functor π# for left semimodules is similarly defined. In particular, the restriction functor π# : SM → RM, applied to the left S-semimodule SS, gives the R-S-bisemimodule RSS = π#S. Then, tensoring by π#S we have the extension functor π# := −⊗R π#S = −⊗R S : MR → MS, which is a left adjoint to the restriction functor π# : MS → MR, by [25, Prop. 4.1]. Using [24, Prop. 3.8], we obtain the following observation, which will prove to be useful. Proposition 4.8. Let π : R → S be a semiring homomorphism. (1) The extension functor π# : MR → MS preserves the subcategory of (finitely generated) strongly projective semimodules. (2) The restriction functor π# : MS → MR preserves the subcategory of (finitely generated) strongly projective semimodules if and only if π#(S) is a (finitely generated) strongly projective right R-semimodule. Proof. (1) Let P be a strongly projective R-semimodule. There is then a right R-semimodule Q such that R(I) ∼= P ⊕ Q for some basis set I. Now according to [24, Prop. 3.8], we obtain that S(I) ∼= R(I) ⊗R S ∼= (P ⊕ Q) ⊗R S ∼= (P ⊗R S) ⊕ (Q ⊗R S), whence π#(P ) = P ⊗R S is a strongly projective right S-semimodule. (2) (=⇒). It is obvious. 10 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL (⇐=). Assume that π#(S) is a strongly projective right R-semimodule and let P be a strongly projective right S-semimodule. Then, R(I) ∼= S ⊕ A and S(J) ∼= P ⊕ B for some right R-semimodule A and some right S-semimodule B. This implies R(I×J) ∼= S(J) ⊕ A(J) ∼= P ⊕ B ⊕ A(J) as right R-semimodules, thus π#(P ) is strongly projective. (cid:3) Applying Propositions 4.4 and 4.8, the next result gives a full description of the (finitely generated) strongly projective semimodules over semisimple semirings. Theorem 4.9. Let S be a semisimple semiring with direct product representation S ∼= Mn1(D1) × · · · × Mnr (Dr) , where D1, . . . , Dr are division semirings. For each 1 ≤ j ≤ r and 1 ≤ i ≤ nj, let e(j) ii denote the ni × ni matrix units in Mnj (Dj). Then, the following holds: (1) A right S-semimodule is strongly projective if and only if it is isomorphic to (e(1) 11 S)(I1) ⊕ · · · ⊕ (e(r) 11 S)(Ir ) for some sets I1, . . . , Ir. (2) A finitely generated right S-semimodule is strongly projective if and only if it can be uniquely written in the form 11 S)k1 ⊕ · · · ⊕ (e(r) where k1, . . . , kr are nonnegative integers. (e(1) 11 S)kr , Proof. For each 1 ≤ j ≤ r, let πj : S → Mnj (Dj) be the canonical projection. Then π# j (Mnj (Dj)) is obviously a (finitely generated) strongly projective right S- semimodule and e(j) ii Mnj (Dj) for 1 ≤ i ≤ nj is a (finitely generated) strongly projective right Mnj (Dj)-semimodule. Therefore, by Proposition 4.8 (2), e(j) ii S = j (e(j) π# ii Mnj (Dj)) is a (finitely generated) strongly projective right S-semimodule. From this and Remark 4.2 (3), we immediately see that the sufficient conditions of statements (1) and (2) are true. Assuming that P ∈ MS is a (finitely generated) strongly projective semimod- ule, we may write it in the form P = (π1)#(P ) ⊕ · · · ⊕ (πr)#(P ) . By Proposition 4.8 (1), (πi)#(P ) is a (finitely generated) strongly right Mni(Di)- semimodule for each 1 ≤ i ≤ r. Now we consider the structure of a (finitely generated) strongly projective right semimodule over a matrix semiring Mm(D) for some positive integer m and division semiring D. For 1 ≤ i, j ≤ m, let eij be the m × m matrix units in Mm(D). Then each eiiMm(D), for i = 1, . . . , m, is a minimal right Mm(D)-semimodule, and Mm(D) = Lm i=1 eiiMm(D). As it is clear that, for all i, j = 1, . . . , m, the Mm(D)- semimodules eiiMm(D) and ejjMm(D) are isomorphic, the semimodules Mm(D) and (e11Mm(D))m are isomorphic as right Mm(D)-semimodules, too. Again, it is easy to see that D is either a division ring or a zerosumfree divi- sion semiring. In the first scenario, it is a well-known "classical" result (e.g., [29, CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 11 Thm. 1.3.3]) that every (finitely generated) projective right Mm(D)-module can be uniquely written in the form (e11Mm(D))(I) for some (finite) set I. Thus from now on, let D be a zerosumfree division semiring, and hence, Mm(D) is a zerosumfree semisimple semiring. Let P be a (finitely generated) strongly projective right Mm(D)-semimodule, i.e., P is a direct summand of a free right Mm(D)-semimodule Mm(D)(J) =: F for some (finite) set J. As in the proof of Theorem 4.5, it is easy to show that P is a strong subsemimodule of F . Also, since Mm(D) ∼= (e11Mm(D))m, we have that F is a semisimple semimodule and F ∼= (e11Mm(D))(K) for some set K (which can be taken to be finite in case P is finitely generated). From the latter, by Proposition 4.4, there exists a set I ⊆ K such that P ∼= (e11Mm(D))(I) (and I is finite when P is finitely generated). If (e11Mm(D))n ∼= (e11Mm(D))k as right Mm(D)-semimodules, then it easy to see that (e11Mm(D))n ∼= (e11Mm(D))k as right D-semimodules, which means that Dmn ∼= Dmk as right D-semimodules. From the latter, by the IBN property of division semirings (see [15, Thm. 5.3] or Corollary 5.2 below), one gets mn = mk and m = k. Therefore, every finitely generated strongly projective right Mm(D)- semimodule can be uniquely written in the form (e11Mm(D))k for some nonnegative integer k. Finally, we notice that e(j) 11 Mnj (Dj)) as right S-semimodules for all j = 1, . . . , r, and conclude that the necessary conditions of statements (1) and (2) are true as well. (cid:3) 11 S ∼= π# j (e(j) At the end of this section we establish some preparatory results regarding strongly projective semimodules over semirings that are direct limits of directed families of semirings. Let us recall a few general notions from universal algebra (see, e.g., [14, Ch. 3]) in a semiring context. A partially ordered set I is directed if any two ele- ments of I have an upper bound in I. Denoting by SR the category of semirings, a direct system {Si ϕij} of semirings over a directed set I consists of a family {Si}i∈I of semirings Si ∈ SR, together with semiring homomorphisms ϕij : Si → Sj for i ≤ j, such that, for all i, j, k ∈ I, if i ≤ j ≤ k, then ϕjkϕij = ϕik and ϕii = idSi. If one defines a binary relation "≡" on the disjoint union Si∈I Si of the sets Si by x ≡ y iff x ∈ Si, y ∈ Sj for some i, j ∈ I, and there exists z ∈ Sk such that i, j ≤ k and ϕik(x) = z = ϕjk(y), then this is easily seen to be an equivalence relation. Considering its set S := {[x] x ∈ Si∈I Si} of equivalence classes, it is not hard to verify that by defining [x] + [y] := [ϕik(x) + ϕjk(y)] and [x] · [y] := [ϕik(x) · ϕjk(y)] , where x ∈ Si, y ∈ Sj and i, j ≤ k, one obtains a semiring lim := S = −→I (S, +, ·, [0], [1]) called the direct limit of the direct system {Si ϕij } of semirings. It is also easy that there is a family {ϕi}i∈I of canonical homomorphisms ϕi : Si → S defined by ϕi(x) := [x] for any x ∈ Si, so that ϕi = ϕjϕij for all i ≤ j; and if all ϕij for i ≤ j are embeddings, then all ϕi, i ∈ I, are embeddings, too. Si Our next result, needed in a sequel, is of a "technical" nature and can be justified by using Lemma 4.3 and repeating verbatim the proof of [13, Lemma 15.10] in the ring setting. However, for the reader's convenience, we briefly sketch here an 12 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL alternative, more homological, proof based on the tensor product construction and related results considered in [24] and [25]. Proposition 4.10 (cf. [13, Lem. 15.10]). Let S be a direct limit of a direct system {Si ϕij } of semirings. Then the following statements are true: (1) If P is a finitely generated (strongly) projective right S-semimodule, then, for some m, there exists a finitely generated (strongly) projective right Sm- semimodule Q such that Q ⊗Sm S ∼= P . (2) If P ⊗Si S ∼= Q ⊗Si S for some i and finitely generated (strongly) projective right Si-semimodules P, Q, then P ⊗Si Sk ∼= Q ⊗Si Sk for some k ≥ i. Proof. (1) Since the semimodule P ∈ MS is a finitely generated summand of a free S-semimodule, we can consider all components of a finite generator P0 to be elements of some semiring Sm, and let Q := P0Sm ∈ MSm. It is easy to SSi = SS in the see that Q is a (strongly) projective Sm-semimodule and lim −→I category SM. Then, P = P0S = P0S ⊗S S = P0S ⊗S lim (P0S ⊗S −→I SSi = SSi) = lim −→I Q ⊗Sm lim −→I (P0Sm ⊗Sm SS ⊗S SSi) = lim −→I SSi = Q ⊗Sm S. (P0Sm ⊗Sm SSi) = P0Sm ⊗Sm lim −→I SSi = lim −→I (2) Since semimodules P, Q ∈ MSi are finitely generated summands of free Si-semimodules, the semimodules P ⊗Si S, Q ⊗Si S ∈ MS are finetely gener- ated summands of free S-semimodules as well. Since any isomorphism between S-semimodules P ⊗Si S and Q ⊗Si S is defined by the finite number of their gen- erators and a finite number of elements of S, and taking into consideration the nature of the congruence relation in the construction of the tensor product, we can consider that all elements of S involved into the isomorphism P ⊗Si S ∼= Q⊗Si S are elements of some semiring Sk with k ≥ i. Therefore, P ⊗Si S ∼= P ⊗Si (Sk ⊗Sk S) ∼= (P ⊗Si Sk) ⊗Sk S and Q ⊗Si S ∼= Q ⊗Si (Sk ⊗Sk S) ∼= (P ⊗Si Sk) ⊗Sk S, and it is clear that P ⊗Si Sk ∼= Q ⊗Si Sk. (cid:3) 5. Characterizing ultramatricial algebras by monoids of isomorphism classes of projective semimodules In this section, we introduce the monoids V(S) and SV(S) of isomorphism classes of finitely generated projective and strongly projective, respectively, semimodules over a semiring S and demonstrate their roles in the characterization of the class of ultramatricial algebras over a semifield. The proof of the main result is essentially based on the presentation in [13, Ch. 15]. From now on, let V(S) be the set of isomorphism classes of finitely generated pro- jective right S-semimodules and, for a finitely generated projective S-semimodule P ∈ MS let P ∈ V(S) denote the class of finitely generated projective right S- semimodules isomorphic to P . Furthermore, defining for any isomorphism classes P and Q an addition "+" by P + Q := P ⊕ Q, it is easy to see that the set V(S) becomes a commutative monoid (V(S), +, 0) with the zero element 0. The monoid V(S) is always a zerosumfree monoid (or a strict cone in the terminology of [13, p. 202]), i.e., x + y = 0 implies x = y = 0. Let SV(S) ⊆ V(S) denote the submonoid of the monoid V(S) consisting of all classes P with a finitely generated strongly projective S-semimodule P ∈ MS, i.e., SV(S) := {P ∈ V(S) P is a strongly projective S-semimodule}. CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 13 Before presenting "computational" examples of the monoids V(S) and SV(S), we start with some useful observations. Recall that a semiring S has the IBN property if Sm ∼= Sn (in MS) implies m = n, for any natural numbers m, n. Lemma 5.1. If ϕ : S → T is a semiring homomorphism and T has the IBN prop- erty, then S has it as well. Proof. Indeed, if Sm ∼= Sn in MS, then by [24, Prop. 3.8] and [25, Thm. 3.3], one readily has T m ∼= Sm ⊗S T ∼= Sn ⊗S T ∼= T n in MT , whence m = n. (cid:3) Corollary 5.2. Division semirings and commutative semirings satisfy IBN. Proof. The case when S is a division semiring was justified in [15, Thm. 5.3]; alternatively, one can use Lemma 5.1 and the fact that for any zerosumfree division semiring S there is a semiring homomorphism ϕ : S → B into the finite IBN semiring B, given by ϕ(0) = 0 and ϕ(s) = 1 for 0 6= s ∈ S. If S is a commutative semiring, by Zorn's lemma there exists a maximal congru- ence ρ on S, so that T := S/ρ is a congruence-simple commutative semiring (i.e., T has only the trivial congruences). By [4, Thm. 10.1], T is either a field or the Boolean semifield B, and hence T has the IBN property. From Lemma 5.1 it follows that S has the IBN property, too. (cid:3) Examples 5.3. Cases of semirings S, for which the monoids V(S) and SV(S) are essentially known, include the following. (1) By Theorem 4.5, we have SV(D) ∼= Z+ for any division semiring D. In par- ticular, SV(B) ∼= Z+. However, Example 4.7 shows that SV(B) is a proper submonoid of V(B). In fact, the monoid V(B) contains a free commutative monoid with countable basis. (2) For any semiring S, obviously V(S) = SV(S) iff all finitely generated pro- jective right S-semimodules are strongly projective. In particular, for any division semiring D, we have V(D) = SV(D) iff D is weakly cancellative, i.e., a + a = a + b implies a = b, for all a, b ∈ S ([19, p. 4026]). Indeed, by Theorem 4.5, one has V(D) = SV(D) iff every finitely generated projective right D-semimodule is free, that is, by [19, Prop. 3.1, Thm. 3.2], iff D is a weakly cancellative division semiring. (3) Let S be a semiring for which every finitely generated projective right S- semimodule is free. (For example, in [34] and [19], polynomial semirings are considered having this property.) Then the monoids V(S) and SV(S) are cyclic monoids generated by the element S. If in addition S has the IBN property (e.g., if S is a commutative semiring), then the monoids V(S) and SV(S) are exactly Z+. Notice that if ϕ : R → S is a semiring homomorphism, then, taking into ac- count Proposition 4.8, we see that ϕ induces a well-defined monoid homomorphism V(ϕ) : V(R) → V(S) such that V(ϕ)(P ) = ϕ#(P ) = P ⊗R S; furthermore, it holds that V(ϕ)(SV (R)) ⊆ SV(S). From these observations, it is routine to check that V and SV give covariant functors from the category of semirings to the category of commutative monoids. Recall that a commutative monoid M is conical if x + y = 0 implies x = 0 = y, for any x, y ∈ M . An order-unit in the monoid M is an element u in M such that for every x ∈ M there exist y ∈ M and a positive integer n such that x + y = nu. 14 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Throughout this section, we denote by C the category consisting of all pairs (M, u), where M is a conical monoid and u is an order-unit in M , with morphisms from an object (M, u) to an object (M ′, u′) to be the monoid homomorphisms f : M → M ′ satisfying f (u) = u′. For any semiring S, observe that S is an order-unit in the conical monoid SV(S). (Note that S is no order-unit in the monoid V(S), unless V(S) = SV(S).) There- fore, we have an object (SV(S), S) in the category C defined above. Given any semiring homomorphism ϕ : R → S, note that SV(ϕ) maps R to S, so that SV(ϕ) is a morphism in C from (SV(R), R) to (SV(S), S). Thus, (SV(−), −) defines a covariant functor from the category of semirings to the category C. In order to apply SV to direct limits and finite products of semirings, we consider direct limits and finite products in the category C. Given a direct system of objects (Mi, ui) and morphisms fij in C, we first form the direct limit M of the commutative monoids Mi and let fi : Mi → M denote the canonical homomorphisms. One can easily check that M is a conical monoid. Since fij(ui) = uj whenever i ≤ j, there is a unique element u ∈ M such that fi(ui) = u for all i, and we observe that u is an order-unit in M . Thus, (M, u) is an object in C, and each fi is a morphism from (Mi, ui) to (M, u). It is easy to see that (M, u) is the direct limit of the (Mi, ui). It is standard how to form finite products in the category C. Namely, given objects (M1, u1), . . . , (Mn, un) in C, we set M = M1 × · · · × Mn, which is a conical monoid, together with u = (u1, . . . , un), which is an order-unit in M . It is easy to check that (M, u) is the product of the (Mi, ui) in C. Proposition 5.4. The functor (SV(−), −) : SR → C preserves direct limits and finite products. Proof. We adapt the proof of [13, Prop. 15.11] to our situation. Let S be the direct limit of a direct system {Si ϕij } of semirings, and for each i let ϕi : Si → S be the canonical homomorphism. Let (M, u) be the direct limit of the monoids (SV(Si), Si) in C, and for each i let fi : (SV(Si), Si) → (M, u) be the canonical homomorphism. We have morphisms SV(ϕi) : (SV(Si), Si) → (SV(S), S) such that SV(ϕj)SV(ϕij ) = SV(ϕi) whenever i ≤ j; hence, there exists a unique monoid homomorphism g : (M, u) → (SV(S), S) such that gfi = SV(ϕi) for all i. We are going to prove that g is an isomorphism. Given P ∈ SV(S), we see from Proposition 4.10 (1) that there is a finitely gener- ated strongly projective right Si-semimodule Q for some i with Q ⊗Si S ∼= P . Then Q ∈ SV(Si), and so fi(Q) ∈ M , and also gfi(Q) = SV(ϕi)(Q) = Q ⊗Si S = P . This implies that g is surjective. Now, let x, y ∈ M be such that g(x) = g(y). Then there exist i and j such that fi(P ) = x and fj(Q) = y for some P ∈ SV(Si) and Q ∈ SV(Sj). Choosing k such that i, j ≤ k, we have that SV(ϕk)SV(ϕik)(P ) = gfi(P ) = g(x) = g(y) = gfj(Q) = SV(ϕk)SV(ϕjk)(Q) , which means SV(ϕik)(P ) ⊗Sk S = SV(ϕjk)(Q) ⊗Sk S. By Proposition 4.10 (2), SV(ϕik)(P ) ⊗Sk St = SV(ϕjk)(Q) ⊗Sk St CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 15 for some t ≥ k. Then, SV(ϕit)(P ) = SV(ϕkt)SV(ϕik)(P ) = SV(ϕik)(P ) ⊗Sk St = SV(ϕjk)(Q) ⊗Sk St = SV(ϕkt)SV(ϕjk)(Q) = SV(ϕjt)(Q), and hence, x = fi(P ) = ftSV(ϕit)(P ) = ftSV(ϕjt)(Q) = fj(Q) = y . Thus g is injective. Using the same argument above and Proposition 4.8, we get that the functor (cid:3) (SV(−), −) preserves finite products. The following fact shows that every free commutative monoid of finite rank occurs as a monoid of isomorphism classes of strongly projective semimodules of a zerosumfree semisimple semiring. Proposition 5.5. Let S be a semisimple semiring with direct product representa- tion S ∼= Mn1(D1) × · · · × Mnr (Dr) , where D1, . . . , Dr are division semirings. For each 1 ≤ j ≤ r and 1 ≤ i ≤ nj, let e(j) ii be the ni × ni matrix units in Mnj (Dj). Then, SV(S) is a free commutative (r)S}, and (SV(S), S) ∼= (cid:0)(Z+)r, (n1, . . . , nr)(cid:1). monoid with basis {e11 (1)S, . . . , e11 Proof. According to Proposition 5.4, we have SV(S) ∼= SV(Mn1(D1)) ⊕ · · · ⊕ SV(Mnr (Dr)) . Moreover, for each 1 ≤ j ≤ r, the monoid SV(Mnj (Dj)) is a free commutative (j)S}, by Theorem 4.9 (2). Using those observations, we monoid with basis {e11 immediately get the statement. (cid:3) From Corollary 3.5 and Proposition 5.5 we readily see that every free com- mutative monoid of finite rank appears as a monoid SV(S) for some additively idempotent congruence-semisimple semiring S. Motivated by this remark and the Realization Problem, which constitutes a very active area in non-stable K-theory (we refer the reader to [3] and the references given there for a recent progress on the Realization Problem), it is natural to pose the following problem. Problem 1. Describe commutative monoids which can be realized as either a monoid SV(S) or V(S) for an additively idempotent semiring S. Now we define a central notion for the present article, which has been investigated in Section 3 for a special case. Definition 5.6. Let F be a semifield. (1) A matricial F -algebra is an F -algebra isomorphic to Mn1(F )×· · ·×Mnr (F ), for some positive integers n1, . . . , nr. (2) An F -algebra is said to be ultramatricial if it is isomorphic to the direct limit (in the category of unital F -algebras) of a sequence S1 → S2 → · · · of matricial F -algebras. We note the following simple observation, pointing out that its justification dif- fers significantly from the arguments in the "classical" ring case. Proposition 5.7. An algebra S over a semifield F is ultramatricial if and only if S is the union of an ascending sequence S1 ⊆ S2 ⊆ · · · of matricial subalgebras. 16 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Proof. Suppose that S is the direct limit of a sequence S1 → S2 → . . . of matricial F -algebras with canonical homomorphisms ϕi : Si → S, then S is the union of the ascending sequence ϕ1(S1) ⊆ ϕ2(S2) ⊆ · · · . Therefore, it is left to show that each ϕi(Si) is a matricial F -algebra. This follows, as is easy to verify, from the following claim. If R and S are matricial F -algebras, where R = R1 × · · · × Rr and Ri = Mni(F ) for some positive integers ni, and if ϕ : R → S is any algebra homomorphism, then ϕ(R) ∼= Qj∈J Rj for some subset J ⊆ {1, . . . , r}; thus ϕ(R) is also a matricial F -algebra. To prove this claim, note that ϕ : R → S as above induces an isomorphism R/ ker(ϕ) → ϕ(R), r 7→ ϕ(r), where ker(ϕ) = {(x, y) ∈ R × R ϕ(x) = ϕ(y)} is its kernel congruence. Using the congruences ρi := ker(ϕ) ∩ Ri ×Ri, we have ϕ(R) ∼= R/ ker(ϕ) ∼= R1/ρ1 × · · · × Rr/ρr . We argue that each ρi is a trivial congruence on Ri, which proves the claim. Now the semifield F is either a field or a zerosumfree semifield. If F is a field, then ρi is obviously a trivial congruence, since Ri is a simple ring. Assume then that F is a zerosumfree semifield, and that ρi is not the identity one. There are distinct elements A = (ajk) and B = (bjk) in Ri such that A ρi B, thus a := ajk 6= bjk =: b for some j, k ∈ {1, . . . , ni}. Denoting by Ejk the matrix units in Ri and using EjkEkt = Ejt, we readily infer that aEjk ρi bEjk for all j, k, whence aIni ρi bIni . This implies that aϕ(Ini ) = ϕ(aIni ) = ϕ(bIni ) = bϕ(Ini ) . If ϕ(Ini ) 6= 0, then since S is a matricial F -algebra we must have that a = b, giving a contradiction. Therefore, ϕ(Ini ) = 0, i.e., Ini ρi 0, which implies C = CIni ρi 0Ini = 0 for all C ∈ Ri, so that ρi = Ri × Ri is the universal one. (cid:3) The subsequent fact, which immediately follows from Propositions 5.4 and 5.5, provides some information on the monoid SV(S) for an ultramatricial F -algebra over a semifield F . Remark 5.8. Let S be a direct limit of a sequence S1 → S2 → . . . of matri- cial F -algebras with canonical homomorphisms ϕi : Si → S, and suppose that Si = Mni r(i), for each i, (F ). Then SV(S) is a cancellative denoting by e11 monoid generated by the set {ϕi(e11 (j,i) ∈ Si the matrix units in Mni (F ) for some positive integers ni (j,i))S 1 ≤ j ≤ r(i), i = 1, 2, . . . }. (F ) × · · · × Mni 1 r(i) 1, . . . , ni j In order to establish the main results of this section, we state the following useful lemma, which proof is essentially based on the one of [13, Lem. 15.23]. Lemma 5.9. Let F be a semifield, let S be a matricial F -algebra, and let T be any unital F -algebra. (1) For any morphism f : (SV(S), S) → (SV(T ), T ) in the category C, there exists an F -algebra homomorphism ϕ : S → T such that SV(ϕ) = f . (2) Let ϕ, ψ : S → T be F -algebra homomorphisms. If SV(ϕ) = SV(ψ), then there exists an inner automorphism θ of S such that ϕ = θψ. Moreover, if in addition T is an ultramatricial F -algebra, then SV(ϕ) = SV(ψ) if and only if there exists an inner automorphism θ of S such that ϕ = θψ. Proof. There are orthogonal central idempotents e1, . . . , er ∈ S with e1+. . .+er = 1 and each eiS ∼= Mni(F ) for some positive integer ni. For each i, denoting by CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 17 e(i) jk ∈ eiS the matrix units, we have that e(i) sition 5.5, SV(S) is a free commutative monoid with basis {e11 11 +. . .+e(i) nini = ei. According to Propo- (1)S, . . . , e11 (r)S}. (1) For each i, we have eiS ∈ SV(S) and so f (eiS) ∈ SV(T ), i.e., f (eiS) = Pi for some finitely generated strongly projective right T -semimodule Pi. Since P1 ⊕ . . . ⊕ Pr = P1 + . . . + Pr = f ( rP i=1 eiS) = f (S) = T , we have P1 ⊕ · · · ⊕ Pr ∼= T as right T -semimodules. Consequently, there exist orthogonal idempotents g1, . . . , gn ∈ T such that g1+· · ·+gn = 1 and each giT ∼= Pi. Note that each giT = Pi = f (eiS). Furthermore, for each i, we have f (e11 (i)S) = Qi for some finitely generated strongly projective right T -semimodule Qi. Because (i)S) = f (e11 i = f (ni e11 (i)S + . . . + enini ∼= giT . As a result, there exist ni × ni matrix units g(i) Qni we have Qni jk ∈ giT gi such i that g(i) (i)S). For every i, there is a unique F -algebra homomorphism from eiS into giT gi sending e(i) jk to g(i) jk , for all j, k = 1, . . . , ni. Consequently, there is a unique F -algebra homomorphism ϕ : S → T such that ϕ(e(i) 11 T ∼= Qi and g(i) nini = gi. Note that g11 (i)S) = f (eiS) = giT , (i)T = Qi = f (e11 11 +. . .+g(i) jk for all i, j, k. Then SV(ϕ)(e11 (i)S) = ϕ(e11 (i)T = f (e11 (i)S) jk ) = g(i) (i))T = g11 for all i = 1, . . . , r. {e11 Since SV(S) is a free commutative monoid with basis (r)S}, we conclude that SV(ϕ) = f . (1)S, . . . , e11 (2) Assume that SV(ϕ) = SV(ψ). Set g(i) i=1 Pni all i, j, k. Note that the g(i) Pr jj = 1, and similarly for the h(i) j=1 g(i) jk ) for jj are pairwise orthogonal idempotents in T such that jk = ϕ(e(i) jk ) and h(i) jk = ψ(e(i) jj . For i = 1, . . . , r, we have g(i) 11 T = SV(ϕ)(e(i) 11 T . Consequently, there are elements xi ∈ g(i) 11 S) = SV(ψ)(e(i) 11 S) = h(i) 11 T ; 11 T h(i) 11 and yi ∈ 11 such that xiyi = g(i) 11 T ∼= h(i) hence, g(i) h(i) 11 T g(i) Set x = Pr i=1 Pni j=1 g(i) rP niP xy = nkP j1 xih(i) j1 xih(i) g(i) i,k=1 j=1 m=1 11 and yixi = h(i) 11 . 1j and y = Pr i=1 Pni m1ykg(k) 1j h(k) rP j1 yig(i) 1j = niP 1m i=1 j=1 = rP i=1 niP j=1 j1 xih(i) g(i) 1j h(i) j1 g(i) g(i) 11 g(i) 1j = rP i=1 niP j=1 g(i) jj = 1 , j=1 h(i) j1 yig(i) 1j . We then have and, similarly, yx = 1. As a result, there exists an inner automorphism θ of T given by the rule θ(a) = xay for all a ∈ T . For all i, j, k, we compute that xh(i) jk = rP nsP t=1 s=1 = g(i) 1t h(i) jk = g(i) t1 xsh(s) g(s) nP nsP 1k = s=1 t=1 jk g(i) k1 xih(i) j1 xih(i) 1j h(i) jk g(i) jk g(s) t1 xsh(s) 1t = g(i) jk x , whence θψ(e(i) we get that θψ = ϕ. jk ) = xh(i) jk y = g(i) jk = ϕ(e(i) jk ). Since the e(i) jk form a basis for S over F , 18 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Finally, assume that T is the limit of a sequence of matricial F -algebras, and that there is a unit x ∈ T such that θ(a) = xax−1 for all a ∈ T . Given any strongly idempotent element e ∈ T , we have that xe = xex−1xe ∈ θ(e)T e and ex−1 = ex−1xex−1 ∈ eT θ(e), where (xe)(ex−1) = θ(e) and (ex−1)(xe) = e, so that θ(e)T ∼= eT , and therefore SV(θ)(eT ) = θ(e)T = eT . Using this and Remark 5.8, we obtain that SV(θ) is the identity map on SV(S). Thus, SV(ϕ) = SV(θ)SV(ψ) = SV(ψ), and we have finished the proof. (cid:3) Now we are ready to state the main results of this section. The following the- orem shows a class of semirings in which the monoid SV(S) determines S up to isomorphism, namely the class of ultramatricial algebras over a semifield. Note that its proof is essentially based on the one of [13, Thm. 15.26]. Theorem 5.10. Let S and T be ultramatricial algebras over a semifield F . Then (SV(S), S) ∼= (SV(T ), T ) if and only if S ∼= T as F -algebras. Proof. The direction (⇐=) is obvious, so we show the direction (=⇒). Assume that f : (SV(S), S) → (SV(T ), T ) is an isomorphism in C. Using Propo- sition 5.7, we may assume that S and T are the union of an ascending sequence S1 ⊆ S2 ⊆ · · · and T1 ⊆ T2 ⊆ · · · , respectively, of matricial subalgebras. For each n = 1, 2, . . . , let ϕn : Sn → S and ψn : Tn → T denote the corresponding inclusion maps. We prove first the following useful claims. Claim 1 : If α : Tk → Sn is an F -algebra homomorphism such that SV(ϕnα) = f −1SV(ψk), then there exist an integer j > k and an F -algebra homomorphism β : Sn → Tj such that ψj βα = ψk and SV(ψj β) = f SV(ϕn). Proof of the claim. By Lemma 5.9 (1), there exists an F -algebra homomorphism β′ : Sn → T such that SV(β′) = f SV(ϕn). Since Sn is a free F -semimodule of finite rank, we must have β′(Sn) ⊆ Ti for some i. Then β′ defines an F -algebra homomorphism β′′ : Sn → Ti such that ψiβ′′ = β′, and we have SV(ψiβ′′) = f SV(ϕn). This implies that SV(ψiβ′′α) = f SV(ϕn)SV(α) = f SV(ϕnα) = SV(ψk) . Applying Lemma 5.9 (2), there exists an inner automorphism θ of T such that ψk = θψiβ′′α. Since Ti is also a free F -semimodule of finite rank, there exists an integer j > k such that θ(Ti) ⊆ Tj. Then θ defines an F -algebra homomorphism θ′ : Ti → Tj such that ψjθ′ = θψi. Set β = θ′β′′, so that β is an F -algebra homomorphism from Sn into Sj and ψjβα = ψjθ′β′′α = θψiβ′′α = ψk. Using Lemma 5.9 (2), we see that SV(ψjβ) = SV(ψθ′β′′) = SV(θψiβ′′) = SV(ψiβ′′) = f SV(ϕn) . Thus, the claim is proved. Similarly, we get the following claim. Claim 2 : If α : Sn → Tk is an F -algebra homomorphism such that SV(ψkα) = f SV(ϕn), then there exist an integer m > n and an F -algebra homomorphism β : Tk → Sm such that ϕmβα = ϕn and SV(ϕmβ) = f −1SV(ψk). We next construct positive integers n(1) < n(2) < · · · and F -algebra homomor- phisms βk : Sn(k) → T such that: (a) For all k = 1, 2, . . . , we have Tk ⊆ βk(Sn(k)) and SV(βk) = f SV(ϕn(k)). (b) For all k = 1, 2, . . . , it is βk injective and βk+1 an extension of βk. By Lemma 5.9 (1), there exists an F -algebra homomorphism α′ : T1 → S such that SV(α′) = f −1SV(ψ1). Because T1 is free F -semimodule of finite rank, we have CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 19 α′(T1) ⊆ Sn(1) for some positive integer n(1). Then α′ defines an F -algebra homo- morphism α : T1 → Sn(1) such that ϕn(1)α = α′, and SV(ϕn(1)α) = f −1SV(ψ1). By Claim 1, there exists an integer j > 1 and an F -algebra homomorphism β : Sn(1) → Sj such that ψjβα = ψ1 and SV(ψjβ) = f SV(ϕn(1)). Then β1 := ψjβ is an F -algebra homomorphism from Sn(1) to T such that β1α = ψ1 and SV(β1) = f SV(ϕn(1)). Moreover, T1 = ψ1(T1) = β1α(T1) ⊆ β1(Sn(1)), and we see that (a) is satisfied for k = 1. Assume that we have n(1), . . . , n(k) and β1, . . . , βk for some positive integer k such that (a) is satisfied up to k and (b) up to k − 1. Since Sn(k) is a free F - semimodule of finite rank, there is an integer i > k such that βk(Sn(k)) ⊆ Ti. Then βk defines an F -algebra homomorphism β′ : Sn(k) → Ti such that ψiβ′ = βk, and we note that SV(ψiβ′) = f SV(ϕn(k)). By Claim 2, there exist a positive integer n(k+1) > n(k) and an F -algebra homomorphism δ : Ti → Sn(k+1) such that ϕn(k+1)δβ′ = ϕn(k) and SV(ϕn(k+1)δ) = f −1SV(ψi). Since ϕn(k+1)δβ′ = ϕn(k), we get that β′ is injective; hence, βk = ψiβ′ is also injective. Applying Claim 1, there exists an integer j > i and an F -algebra homomorphism γ : Sn(k+1) → Sj such that ψjγδ = ψi and SV(ψj γ) = f SV(ϕn(k+1)). Then βk+1 := ψjγ is an F -algebra homomorphism from Sn(k+1) into T such that SV(βk+1) = f SV(ϕn(k+1)). From βk+1δ = ψjγδ = ψi and i ≥ k + 1, we get Tk+1 = ψi(Tk+1) = βk+1δ(Tk+1) ⊆ βk+1δ(Ti) ⊆ βk+1(Sn(k+1)) . Finally, since ϕn(k+1)δβ′ = ϕn(k) and βk+1δβ′ = ψiβ′ = βk, we obtain that βk+1 is an extension of βk. Therefore, (a) holds for k + 1 and (b) for k, so that the induction works. Since n(1) < n(2) < · · · and n(k) ≥ k, we immediately get S Sn(k) = S. As a result, the βk induce an injective F -algebra homomorphism β : S → T such that βϕn(k) = βk. Then Tk ⊆ βk(Sn(k)) = β(Sn(k)) ⊆ β(S) for all k, hence, T = β(S). Thus β is surjective and, therefore, establishes an isomorphism S ∼= T . (cid:3) Finally, we deduce that ultramatricial algebras over a semifield are characterized also by their monoid of finitely generated projective semimodules. Theorem 5.11. Let S and T be ultramatricial algebras over a semifield F . Then, there exists a monoid isomorphism f : V(S) → V(T ) such that f (S) = T if and only if S ∼= T as F -algebras. Proof. Again, the direction (⇐=) is obvious, and we show the direction (=⇒). Let P be a finitely generated strongly projective right S-semimodule. Then there is a positive integer n such that Sn ∼= P ⊕ Q for some finitely generated projective right S-semimodule, i.e., we have nS = P + Q in V(S), whence nT = f (nS) = f (P ) + f (Q) in V(T ). This implies that f (P ) ∈ VS(T ), so f induces an isomorphism from (SV(S), S) onto (SV(T ), T ) in the category C. Now, applying Theorem 5.10, we immediately get that S ∼= T as F -algebras, as desired. (cid:3) 6. The K0-group characterization of ultramatricial algebras over semifields The main goal of this section is to investigate Grothendieck's K0-groups on finitely generated projective semimodules, whose study was initiated by Di Nola and 20 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Russo [7], and to introduce and examine K0-theory on finitely generated strongly projective semimodules, as well as to establish "semiring" analogs of Elliott's cele- brated classification theorem for ultramatricial algebras over an arbitrary field [8]. Consequently, we classify zerosumfree congruence-semisimple semirings in terms of K0-theory. We begin this section by recalling the K0-group of a semiring which was men- tioned by Di Nola and Russo in [7, Sec. 4]. Definition 6.1 (cf. [7, Sec. 4]). Let S be a semiring. The Grothendieck group K0(S) is the additive abelian group presented by the set of generators V(S) and the following set of relations: P ⊕ Q = P + Q for all P , Q ∈ V(S). Remark 6.2. The Grothendieck group K0(S) can be described as follows. (1) Let G be the free abelian group generated by P ∈ V(S), and H the subgroup of G generated by P ⊕ Q − P − Q , where P , Q ∈ V(S). Then K0(S) = G/H, and we denote by [P ] the image of P in K0(S). (2) A general element of K0(S) has the form x = [P1] + . . . + [Pm] − [Q1] − . . . − [Qn] = [P ] − [Q] , with P := P1 ⊕ · · · ⊕ Pm, Q := Q1 ⊕ · · · ⊕ Qn and P , Q ∈ V(S). (3) Defining on X := V(S)×V(S) an equivalence relation by (P , Q) ∼ (P ′, Q′) ⇐⇒ ∃T ∈ V(S) : P ⊕ Q′ ⊕ T ∼= P ′ ⊕ Q ⊕ T , it is a routine matter to check that X/∼ := {[P , Q] P , Q ∈ V(S)} becomes an abelian group by defining [P , Q]+[T , U ] := [P ⊕T , Q⊕U ] for P , Q, T , U ∈ V(S), and that there is a group isomorphism given by K0(S) → X/∼ , [P ] − [Q] 7→ [P , Q] . (4) For any finitely generated projective right S-semimodules P and Q we have [P ] = [Q] ∈ K0(S) if and only if P ⊕ T ∼= Q ⊕ T for some finitely generated projective right S-semimodule T . This follows immediately from (3); see also [30, Prop. I.6.1] for a direct proof given in the case of rings, which serves in our semiring setting as well. Notice that if ϕ : R → S is a semiring homomorphism, then ϕ induces a well- defined group homomorphism K0(ϕ) : K0(R) → K0(S) such that K0(ϕ)([P ]) = [ϕ#(P )] = [P ⊗R S]. From this observation, we easily check that K0 gives a covariant functor from the category of semirings to the category of abelian groups. This fact was also mentioned by Di Nola and Russo [7]. Furthermore, we note the following useful fact. The proof is quite similar as it was done in the one of Proposition 5.4 (also, we can refer to [13, Prop. 15.11, Prop. 15.13]); hence, we will not reproduce it here. Proposition 6.3. The functor K0(−) : SR → A preserves direct limits and finite products, where A is the category of abelian groups. We next consider the K0-group of finitely generated strongly projective semimod- ules over a semiring. Similarly to the group K0(S) of a semiring S, we introduce the following notion. CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 21 Definition 6.4. Let S be a semiring. The Grothendieck group SK0(S) is the additive abelian group presented by the set of generators SV(S) and the following set of relations: P ⊕ Q = P + Q for all P , Q ∈ SV(S). The Grothendieck group SK0(S) can also be described as follows. Let G′ be the free abelian group generated by P ∈ SV(S), and H ′ the subgroup of G′ generated by P ⊕ Q − P − Q , where P , Q ∈ SV(S). Then SK0(S) = G′/H ′, and we denote by bP the image of P in SK0(S). Similarly to the case of the group K0(S), a general element of SK0(S) may be written in the form x = bP − bQ , where P and Q are finitely generated strongly projective right S-semimodules. We may also choose a finitely generated strongly projective right S-semimodule Q′ such that Q ⊕ Q′ ∼= Sn for some n, and rewrite x = \P ⊕ Q′ − \Q ⊕ Q′ = cP ′ − cSn , where P ′ = P ⊕ Q′. Again, it is not hard to see that SK0(−) defines a covariant functor from the category of semirings to the category of abelian groups. Furthermore, we have the following lemma, whose proof is done similarly to the ones of Remark 6.2, and hence, we will not reproduce it here. Lemma 6.5 (cf. [30, Prop. I.6.1]). Let P and Q be finitely generated strongly projective right S-semimodules. Then, the following are equivalent: (1) bP = bQ ∈ SK0(S); (2) P ⊕ T ∼= Q ⊕ T for some finitely generated strongly projective right S- semimodule T ; (3) there exists a positive integer n such that P ⊕ Sn ∼= Q ⊕ Sn. Remark 6.6. Let us note the following simple facts. (1) For any semiring S, there is always the canonical group homomorphism  : SK0(S) → K0(S), defined by ( bP ) = [P ]. (2) Let S be a semiring all of whose finitely generated projective right modules are free. Then the groups SK0(S) and K0(S) are cyclic groups generated by bS and [S], respectively. And, if in addition S has the IBN property, then those groups are exactly Z. (3) The group SK0(B) is isomorphic to the free abelian group Z, but K0(B) contains as a subgroup a free abelian group with countably infinite basis. Indeed, the first fact follows from Example 5.3 (1), and the latter follows from the following proposition. Proposition 6.7. For an additively idempotent commutative semiring S, the group K0(S) contains a free abelian group with countable basis as a subgroup. Proof. We first prove the statement for the case when S = B. For any prime number p, consider the subset Qp := {0, 1, . . . , p − 1} of Z+. From Fact 4.6 we see that the monoid (Qp, max) is a projective B-semimodule. We denote by G the 22 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Indeed, assume that Pr subgroup of K0(S) generated by all elements [Qp] and show that the countably infinite set {[Qp] p is prime} is a basis of G. i=1 ni[Qpi ] = 0 in K0(B), where the ni are integers and the pi are pairwise distinct prime numbers. We may assume that n1, . . . , nk ≥ 0 and nk+1, . . . , nr ≤ 0, so that, writing mj = −nj for k+1 ≤ j ≤ r, we have kP pi ⊕ T = Lr i=1 i=1 Qni ni[Qpi] = rP j=k+1 mj[Qpj ] , j=k+1 Qmj pj ⊕ T for some finitely generated projective and hence, Lk B-semimodule T . Since every finitely generated projective B-semimodule is finite, we get that Lk j=k+1 Qmj i=1 Qni pj . This implies that pn1 , and hence, we must have ni = 0 for all i = 1, . . . , r. Therefore, K0(B) contains the subgroup G which is isomorphic to the free abelian group with countable basis {p ∈ Z+ p is prime}. pj ⊕ T , and hence, Lk k+1 . . . pmr pi ⊕ T = Lr 1 . . . pnk k = pmk+1 i=1 Qni pi = Lr j=k+1 Qmj r Consider now the case when S is an arbitrary additively idempotent commu- tative semiring. One may readily find a maximal congruence ρ on S by using Zorn's lemma. We then have that the additively idempotent commutative semiring T := S/ρ has only the trivial congruences. By [4, Thm. 10.1], T is the Boolean semifield B. Let ı : B → S and π : S → B be the canonical injection and surjection, respectively. Since π ◦ ı = idB, we must have that K0(π)K0(ı) = idK0(B), by the functorial property of K0. This implies that K0(ı) : K0(B) → K0(S) is an injective group homomorphism, and hence, we may consider K0(B) as a subgroup of K0(S) and finish the proof. (cid:3) The homomorphism  of Remark 6.6 (1) is, in general, not an isomorphism in semiring setting. Lemma 6.8. The canonical homomorphism  : SK0(B) → K0(B) is injective but not surjective. Proof. Assume that x = bP − bQ ∈ SK0(B) such that (x) = 0. We then have that [P ] = [Q] ∈ K0(B), so P ⊕ T ∼= Q ⊕ T for some finitely generated projective B-semimodule T , by Remark 6.2 (4). For P and Q are finitely generated strongly projective B-semimodule and Theorem 4.5, there exist nonnegative integers m and n such that P ∼= Bm and Q ∼= Bn. Furthermore, since T is a finitely generated B- semimodule, we immediately get that T is a finite set. Then, from the equality P ⊕ T ∼= Q ⊕ T , we must have that Bm ⊕ T = Bn ⊕ T ; hence, m = n. This implies that P ∼= Q, that means, x = 0 ∈ SK0(B). Therefore,  is injective. From Proposition 6.7 and since SK0(B) ∼= Z by Remark 6.6 (2), we see that  is not surjective. For the reader's convenience, we also give a direct argument. Assume that  is surjective, and consider the projective B-semimodule P = {(0, 0), (0, 1), (1, 1)} as in Example 4.7. We then have that there exists an element x = bA − bB ∈ SK0(B) such that (x) = [P ], that means, A ⊕ C ∼= B ⊕ P ⊕ C for some finitely generated projective B-semimodule C. Since A, B and C are finite, we get that A = B ⊕ P . By Theorem 4.5, we write A and B of the form A ∼= Bm and B ∼= Bn for some nonnegative integers m and n. From these observations, CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 23 we must have that P = 2m−n, a contradiction. Thus  is not surjective, as claimed. (cid:3) In the next theorem we provide a criterion for checking the isomorphism prop- erty of the homomorphism  in a division semiring setting. Before doing this, we need some useful notions and facts. Following [2], a family X of elements in a semi- module M over a semiring S is weakly linearly independent if there is no element in X that can be expressed as a linear combination of other elements of X. We define the weak dimension of M , denoted by dimw(M ), as the minimum cardinality of a weakly linearly independent generating family of M . It is not hard to see that the weak dimension of a semimodule M is equal to the minimum cardinality of a minimal generating family, or the minimum cardinality of any generating family of M . Recall that a semiring S is entire if ab = 0 implies that a = 0 or b = 0 for any a, b ∈ S. Lemma 6.9. Let S be a zerosumfree entire semiring, and P, Q finitely generated projective right S-semimodules. Then dimw(P ⊕ Q) = dimw(P ) + dimw(Q). Proof. Notice that dimw(A ⊕ B) ≤ dimw(A) + dimw(B) holds for any finitely gen- erated right S-semimodules A and B, so it suffices to prove the converse inequality for finitely generated projective right S-semimodules, which we may assume to be nonzero. It is easy to see that every projective right S-semimodule T is zerosumfree, and satisfies that xs = 0 implies x = 0 or s = 0, for x ∈ T , s ∈ S. Now let d := dimw(P ⊕ Q) and let X = {(x1, y1), . . . , (xd, yd)} be a minimal generating family of P ⊕ Q for some xi ∈ P , yi ∈ Q. We may assume that X = (cid:8)(x1, 0), . . . , (xk, 0), (0, yk+1), . . . , (0, yn), (xn+1, yn+1), . . . , (xd, yd)(cid:9) with all xi, yj 6= 0, for some 0 ≤ k ≤ n ≤ d. (xj, 0) = Pd i=1(xi, yi)si for some si ∈ S, so that For each n + 1 ≤ j ≤ d, we have (xj , 0) ∈ P ⊕ Q and hence we can write xj = dP i=1 xisi ∈ P and 0 = dP i=1 yisi ∈ Q . As Q is zerosumfree we have yisi = 0 for all i, and since yk+1, . . . , yd 6= 0 we get that sk+1 = . . . = sd = 0; thus we obtain (xj , 0) = Pk i=1(xi, 0)si. Similarly, we have that (0, yj) = Pn j=k+1(0, yj)rj for some rj ∈ S, whence (xj, yj) = kP i=1 (xi, 0)si + nP j=k+1 (0, yj)rj . From this observation and the minimality of X, we infer that n = d. It is then easy to see that {x1, . . . , xk} and {yk+1, . . . , yd} are generating families of P and Q, respectively. This implies that k ≥ dimw(P ) and d − k ≥ dimw(Q), whence dimw(P ⊕ Q) = d ≥ dimw(P ) + dimw(Q), as desired. (cid:3) Theorem 6.10. For an arbitrary division semiring D the following is true: (1) The canonical homomorphism  : SK0(D) → K0(D) is injective. (2) The homomorphism  : SK0(D) → K0(D) is an isomorphism if and only if D is a weakly cancellative division semiring. 24 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Proof. Note first that D is either a division ring or a zerosumfree division semiring. Also, if D is a division ring, then the group homomorphism  : SK0(D) → K0(D) is always an isomorphism, since every right D-semimodule is free; that means, the statements are obvious. Consider now the case when D is a zerosumfree division semiring. We then have a semiring homomorphism π : D → B, defined by π(0) = 0 and π(x) = 1 for all 0 6= x ∈ D, and the following diagram SK0(D) SK0(π) SK0(B)   K0(D) K0(π) / K0(B) is commutative. By Proposition 5.5, we obtain that SK0(D) is the free abelian group with basis bD. (1) Assume that (n bD) = [0] ∈ K0(D) for some nonnegative integer n. Since the diagram above is commutative and  : SK0(B) → K0(B) is injective by Lemma 6.8, we get that SK0(π)(n bD) = b0 ∈ SK0(B), that means, cBn = \π#(Dn) = b0 ∈ SK0(B) , so that Bn ⊕ Bm ∼= Bm for some nonnegative integer m, by Lemma 6.5. This implies that n = 0, whence  : SK0(D) → K0(D) is injective. (2) The sufficient condition follows immediately from [19, Thm. 3.2] which shows that every projective right semimodule over a weakly cancellative division semiring is always free. In order to prove the necessary condition, we assume that the group homomorphism  : SK0(D) → K0(D) is an isomorphism, and D is not a weakly cancellative division semiring. Then there exists an element s ∈ D such that s 6= 1 and 1 + 1 = 1 + s. Suppose that s = 0. Then D contains B as a subsemiring. Letting ı : B → D be the canonical injection, we have that π ◦ ı = idB, whence K0(π)K0(ı) = idK0(B). This implies that K0(π) is surjective, so that  : SK0(B) → K0(B) is also surjective, contradicting Lemma 6.8. Therefore, we must have that s 6= 0. Consider the subsemimodule P = {(d, d)a + (ds, d)b a, b ∈ D} of the free right D-semimodule D2, where d = (1 + 1)−1. We then have that P ∈ MD is finitely d d (cid:1) is an generated projective, since d + d = 1 = d + ds and thus the matrix (cid:0) d ds idempotent one. Since  : SK0(D) → K0(D) is an isomorphism, there exists a nonnegative integer n such that (n bD) = [P ] ∈ K0(D), and hence, Dn ⊕ Q ∼= P ⊕ Q for some finitely generated projective right D-semimodule Q. This implies that K0(π)([Dn ⊕ Q]) = K0(π)([P ⊕ Q]), that is, π#(Dn) ⊕ π#(Q) ⊕ T ∼= π#(P ) ⊕ π#(Q) ⊕ T for some finitely generated projective right B-semimodule T . Furthermore, we have that π#(Dn) ∼= Bn and π#(P ) ∼= B, so that Bn ⊕ π#(Q) ⊕ T ∼= B ⊕ π#(Q) ⊕ T . Consequently, we must have that n = 1; hence, D ⊕ Q ∼= P ⊕ Q. Now, applying Lemma 6.9, we obtain that dimw(P ) = 1, so that P is free. On the other hand, by [19, Prop. 3.1], P is not a free semimodule, and with this (cid:3) contradiction we end the proof. / / / CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 25 Theorem 6.10 provokes a quite natural and, in our view, interesting question. Problem 2. Describe the class of all semirings S having the homomorphism  : SK0(S) → K0(S) to be an isomorphism. Is this class axiomatizable in the first order semiring language? In order to discuss the combined structure on SK0(S) and K0(S), we require the following definitions. Recall [13, p. 203] that a cone in an abelian group G is an additively closed subset C such that 0 ∈ C. Any cone C in G determines a pre- order ≤ (i.e., a reflexive, transitive relation) on G, which is translation-invariant (i.e., x ≤ y implies x + z ≤ y + z), by letting x ≤ y if and only if y − x ∈ C. Conversely, any translation-invariant pre-order ≤ on G arises in this fashion, from the cone {x ∈ G x ≥ 0}. A pre-ordered abelian group is a pair (G, ≤), where G is an abelian group and ≤ is a translation-invariant pre-order on G. When there is no danger of confusion as to the pre-order being used, we refer to G itself as a pre-ordered abelian group, and we write G+ for the cone {x ∈ G x ≥ 0}. An order-unit in a pre-ordered abelian group G is an element u ∈ G+ such that for any x ∈ G, there exists a positive integer n with x ≤ nu. We note that if G has an order-unit, then any element x ∈ G can be written as the difference of two elements of G+, whence G is generated as a group by G+. Throughout this section, we use P to denote the following category. The objects of P are pairs (G, u) such that G is a pre-ordered abelian group and u is an order- unit in G. The morphisms in P from an object (G, u) to an object (H, v) are the monotone (i.e., order-preserving) group homomorphisms f : G → H such that f (u) = v. Unspecificed categorical terms applied to pre-ordered abelian groups are to be interpreted in P. For example, (G, u) ∼= (H, v) means that there is a group isomorphism f : G → H with f (u) = v and f, f −1 monotone (equivalently, there exists a group isomorphism f : G → H such that f (u) = v and f (G+) = H +). Definition 6.11 (cf. [13, Def., p. 203]). Given a semiring S, there is a natural way to make SK0(S) into a pre-ordered abelian group with order-unit, as follows. First, let X denote the class of all finitely generated strongly projective right S- semimodules, and define SK0(S)+ = { bP P ∈ X}. It is clear that SK0(S)+ is a cone in SK0(S), and we refer to the pre-order on SK0(S) determined by this cone as the natural pre-order on SK0(S). Explicitly, we have bA − bB ≤ bC − bD in SK0(S) if and only if A ⊕ D ⊕ E ⊕ Sn ∼= C ⊕ B ⊕ Sn (as right S-semimodules) for some E ∈ X and some positive integer n. For any semiring S, observe that bS is an order-unit in the pre-ordered abelian group SK0(S). Therefore, we have an object (SK0(S), bS) in the category P defined above. Given any semiring homomorphism ϕ : R → S, note that SK0(ϕ) maps SK0(R)+ into SK0(S)+, and that SK0(bR) = bS, so that SK0(ϕ) is a morphism in P from (SK0(R), bR) to (SK0(S), bS). Thus, (SK0(−), b−) defines a covariant functor from the category of semirings to the category P. We present the following examples in order to illustrate these notions. Examples 6.12. Let F be any semifield. Then we have: (1) (SK0(F ), bF ) ∼= (Z, 1); (2) (SK0(Mn(F )), \Mn(F )) ∼= (Z, n) ∼= ( 1 n Z, 1), for any positive integer n. 26 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Proof. (1) By Theorem 4.5, each element of SK0(F ) is uniquely written in the form n bF with n ∈ Z, giving an isomorphism (SK0(F ), bF ) ∼= (Z, 1), n bF 7→ n. (2) Set S := Mn(F ). As was shown in [25, Theorem 5.14], the functors G : MS ⇆ MF : H given by G(A) = Ae11 and H(B) = Bn establish an equivalence of the semimodule categories MS and MF , where e11 is the matrix unit in Mn(F ). Then the restriction αSK0(S) : SK0(S) → SK0(F ) is a group isomorphism with α(bS) = cF n = n bF in SK0(F ). This shows that (SK0(S), bS) ∼= (SK0(F ), n bF ), and hence, we readily get that (SK0(S), bS) ∼= (Z, n), using (1). The group isomorphim ϕ : Z → 1 n n , induces an isomorphism (Z, n) ∼= ( 1 Z, defined by ϕ(1) = 1 Z, 1). n (cid:3) For any a semiring S, similarly to the case of SK0(S), we may make K0(S) into a pre-ordered abelian group by defining the cone K0(S)+ to be the set of all [P ], where P is a finitely generated projective right S-semimodule. Explicitly, we have [A] − [B] ≤ [C] − [D] in K0(S) if and only if A ⊕ D ⊕ E ⊕ F ∼= C ⊕ B ⊕ F (as right S-semimodules) for some finitely generated projective right S-semimodules E and F . However, K0(S) has no order-units in general. Proposition 6.13. For any additively idempotent commutative semiring S, the pre-ordered group K0(S) has no order-units. Proof. We first prove the statement for the case when S = B. Assume that [A] is an order-unit in K0(B). For any prime number p consider the subset Qp := {0, 1, . . . , p−1} of Z+ and the monoid (Qp, max), which is a projective B-semimodule by Fact 4.6. Since [A] is an order-unit, we have that [Qp] ≤ m[A] in K0(B) for some positive integer m, which means that Qp ⊕ B ⊕ C ∼= Am ⊕ C for some finitely generated, thus finite, projective B-semimodules B and C. We deduce that Qp ⊕ B ⊕ C = Am ⊕ C, so that Am = Qp ⊕ B = QpB. This implies that p = Qp divides A for all primes p, which is a contradiction. Now if S is an arbitrary additively idempotent commutative semiring, by the same reasoning as in the proof of Proposition 6.7, we have homomorphisms ı : B → S and π : S → B such that K0(π)K0(ı) = idK0(B). Suppose U to be an order-unit in K0(S). Then, for each X ∈ K0(B), there exists a positive integer n such that K0(ı)(X) ≤ nU in K0(S), whence X = K0(π)K0(ı)(X) ≤ nK0(π)(U ) in K0(B), which means that K0(π)(U ) is an order-unit in K0(B), contradicting the fact above, finishing our proof. (cid:3) Izhakian, Knebusch and Rowen [22] develop a theory of the decomposition socle for zerosumfree semimodules; and consequently, the authors prove that a zerosum- free semiring S is a finite direct sum of indecomposable projective semimodules if and only if S has a finite set of orthogonal primitive idempotents whose sum is 1S ([22, Thm. 3.3]). In this case, every finitely generated strongly projective S-semimodule is uniquely decomposed by these orthogonal primitive idempotents ([22, Cor. 3.4]). The following theorem allows us to express these results from the point of view of K-theory. Theorem 6.14. For any zerosumfree semiring S, the following are equivalent: (1) S has a finite set of orthogonal primitive idempotents whose sum is 1S; (2) the monoid SV(S) is cancellative and there exists a positive integer n such that (SK0(S), bS) ∼= (Zn, (1, . . . , 1)). CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 27 Proof. (1) =⇒ (2). Let {e1, . . . , en} be a set of orthogonal primitive idempotents of S whose sum is 1S. By [22, Cor. 3.4], every finitely generated strongly projective right S-semimodule P is uniquely written in the form P ∼= Ln i=1(eiS)ni for some non-negative integers ni. This implies that the monoid SV(S) is cancellative, and the map f : SK0(S) → Zn, defined by f (deiS) = ǫi for all i = 1, . . . , n, is a group isomorphism, where {ǫ1, . . . , ǫn} is the canonical basis of the free abelian group Zn. Obviuosly, S = Ln i=1 ǫi = (1, . . . , 1) ∈ Zn. From these observations, we get that (SK0(S), bS) ∼= (Zn, (1, . . . , 1)). (2) =⇒ (1). Let f : (SK0(S), bS) → (Zn, (1, . . . , 1)) be an isomorphism in P, and let {ǫ1, . . . , ǫn} be the canonical basis of the free abelian group Zn. Then, for each 1 ≤ i ≤ n, there is a unique element bPi ∈ SK0(S)+ such that f ( bPi) = ǫi. This implies that i=1 eiS, hence, f (bS) = Pn \nL Pi) = f ( i=1 nP i=1 f ( bPi) = nP i=1 ǫi = (1, . . . , 1) = f (bS) , so that \Ln i=1 Pi = bS, since f is injective. By Lemma 6.5, there exists a positive integer m such that (Ln i=1 Pi ∼= S, since the monoid SV(S) is cancellative. Consequently, there exists a set of orthogonal idempotents {e1, . . . , en} of S such that Pn i=1 Pi) ⊕ Sm ∼= Sm+1, and hence, Ln i=1 ei = 1 and Pi ∼= eiS for all i. It is left to prove that each ei is primitive. To this end, assume that there exist idempotent elements e, e′ ∈ S such that ee′ = 0 = e′e and ei = e + e′. We then have that eiS = eS ⊕ e′S, and both x := f (ceS) and y := f ( ce′S) are elements in (Z+)n, due to our hypothesis that f (SK0(S)+) = (Z+)n. From the equality x + y = f (ceS) + f ( ce′S) = f (deiS) = ǫi , we immediately see that either x = ǫi, y = 0 or x = 0, y = ǫi. Suppose that y = 0, so that f ( ce′S) = 0, which implies ce′S = 0 in SK0(S). By Lemma 6.5, there exists a positive integer k such that e′S ⊕ Sk ∼= Sk, and therefore, e′S = 0 (since the monoid SV(S) is cancellative), that is, e′ = 0. A similar argument applies in the case x = 0. Thus ei is a primitive idempotent, as desired. (cid:3) In order to apply SK0 to direct limits of semirings, we consider direct limits in the category P. Given a direct system of objects (Gi, ui) and morphisms fij in P, we first form the direct limit G of the abelian groups Gi. For each i, let gi : Gi → G denote the canonical homomorphism. Setting G+ = S gi(G+ i ), we obtain a cone G+ in G, using which G becomes a pre-ordered abelian group. Notice that the homomorphisms gi are all monotone. Inasmuch as fij(ui) = uj whenever i ≤ j, there is a unique element u ∈ G such that gi(ui) = u for all i, and we observe that u is an order-unit in G. Thus, (G, u) is an object in P, and each gi is a morphism from (Gi, ui) to (G, u) in P. It is easy to see that (G, u) is the direct limit of the (Gi, ui). 1 × . . . × G+ Finite products in the category P can be formed in a standard manner. Namely, given objects (G1, u1), . . . , (Gn, un) in P, we set G = G1 × . . . × Gn and G+ = G+ n . Then G+ is a cone in G, using which G becomes a pre-ordered abelian group. Also, u = (u1, . . . , un) is an order-unit in G, whence (G, u) is an object in P. It is easy to check that (G, u) is the product of the (Gi, ui) in P (see, also, [13, p. 210] for details). 28 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL Proposition 6.15 (cf. [13, Prop. 15.11, Prop. 15.13]). The functor (SK0(−), b−) : SR → P preserves direct limits and direct products. Proof. Similarly as it was done in the proof of Proposition 5.4, we may show that the functor (SK0(−), b−) preserves direct limits; hence, we will not reproduce it here. (Also, using Proposition 4.10 and repeating verbatim the proof of [13, Prop. 15.11], we immediately get the statement.) Using Proposition 4.8 and repeating the proof of [13, Prop. 15.13], we obtain that the functor (SK0(−), b−) preserves direct products, and which, for the reader's convenience, we provide here. Namely, let S be the direct product of semirings S1, . . . , Sn, and for each i let ϕi : S → Si be the canonical projection. Let (G, u) be the product of the (SK0(Si), bSi) in P, and for each i let pi denote the canonical projection (G, u) → (SK0(Si), bSi). We have morphisms SK0(ϕi) : (SK0(S), bS) → (SK0(Si), bSi) for each i; hence, there is a unique mor- phism f : (G, u) → (SK0(S), bS) such that pif = SK0(ϕi) for all i. We are going to prove that f is an isomorphism in P. Given x ∈ G+, there exists a finitely generated strongly projective right Si- semimodule Pi for each i such that pi(x) = bPi. Then, by Proposition 4.8 and Remark 4.2 (3), P = P1 × . . . × Pn is a finitely generated strongly projective right S-semimodule such that P ⊗S Si ∼= Pi for all i. As a result, we get bP ∈ SK0(S)+ such that pif ( bP ) = SK0(ϕi)( bP ) = \P ⊗S Si = bPi = pi(x) for all i, whence f ( bP ) = x. Thus, f (SK0(S)+) = G+. Specially, it follows that f is surjective. Now consider any bP − bQ ∈ ker(f ). For all i we have \P ⊗S Si − \Q ⊗S Si = SK0(ϕi)( bP − bQ) = pif (bP − bQ) = 0; whence (P ⊗S Si) ⊕ Ski i Lemma 6.5. Then we obtain a positive integer k = max{k1, . . . , kn} such that for some a positive integer ki, by ∼= (Q ⊗S Si) ⊕ Ski i (P ⊕ Sk) ⊗S Si ∼= (P ⊗S Si) ⊕ Sk i ∼= (Q ⊗S Si) ⊕ Sk i ∼= (Q ⊕ Sk) ⊗S Si for all i. Consequently, we get that P ⊕ Sk ∼= Q ⊕ Sk, whence bP − bQ = 0, by Lemma 6.5. Thus f is injective, and the proof is finished. (cid:3) We are going to present semiring analogs of Elliott's theorem [8] for ultramatricial algebras over an abitrary semifield. In order to establish the main theorems of this section, we need some preparatory facts and notions. Proposition 6.16. Let S be a semisimple semiring with direct product represen- tation S ∼= Mn1(D1) × · · · × Mnr (Dr) , where D1, . . . , Dr are division semirings. For each 1 ≤ j ≤ r and 1 ≤ i ≤ nj, let e(j) ii denote the ni × ni matrix unit in Mnj (Dj). Then SK0(S) is a free abelian group with basis { \e11 Proof. It follows from Propositions 5.5 and 6.15, and Examples 6.12 (2). (r)S}, and (SK0(S), bS) ∼= (Zr, (n1, . . . , nr)). (1)S, . . . , \e11 (cid:3) Notice that by Corollary 3.5 and Proposition 6.16, we immediately get that every free abelian group of finite rank may appear as a SK0(S) for some additively idempotent congruence-semisimple semiring S. There are also pre-ordered abelian CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 29 groups not being free, which appear as a SK0(S) for an additively idempotent semiring S. We illustrate this by presenting the following example. Example 6.17. Fix an additively idempotent semifield F (e.g., the Boolean or the tropical semifield) and consider the semiring Sn = M2n (F ) (for n ≥ 0) of square matrices of order 2n over F . We may consider Sn as a subsemiring of Sn+1 by identifying a 2n × 2n-matrix M with the 2n+1 × 2n+1-matrix (cid:0) M 0 0 M (cid:1). In this way, we have a chain of additively idempotent semirings S0 ⊆ S1 ⊆ S2 ⊆ · · · . Denoting by S the direct limit of the directed system {Sn n ∈ Z+}, then S is an additively idempotent semiring. Moreover, for all n we have (SK0(Sn), cSn) ∼= ( 1 by Examples 6.12, so that using Proposition 6.15 we see that (SK0(S), bS) ∼= (lim −→ 2n m ∈ Z and n ∈ Z+}. Example 6.17 and the facts above motivate the following natural question. Z, 1); it is routine to check that lim −→ Z = { m Z, 1) 1 2n 2n 1 2n Problem 3. Describe all pre-ordered abelian groups which can be either SK0(S) or K0(S) for some additively idempotent semiring S. Proposition 6.18. Let F be a semifield and S a matricial F -algebra. Then the canonical homomorphism  : SK0(S) → K0(S) is injective. Proof. It is easy to see that F is either a field or a zerosumfree semifield. If F is a field, then  is obviously an isomorphism, since the functors SK0 and K0 are the same. Assuming that F is a zerosumfree semifield, write S in the form S = Mn1(F ) × . . . × Mnr (F ) , (1)S, . . . , \e11 11 S] for all j = 1, . . . , r. and for each 1 ≤ j ≤ r and 1 ≤ i ≤ nj let e(j) ii be the nj × nj matrix unit in Mnj (F ). Then SK0(S) is a free abelian group with basis { \e11 (r)S}, by Proposition 6.16. Therefore, the canonical homomorphism  : SK0(S) → K0(S) is exactly given by ( \e11 (j)S) = [e(j) Next we show that  is injective for the case when F is the Boolean semifield B. Indeed, assume that 0 = (Pr j=1 kj[e(j) 11 S] ∈ K0(S) for some nonnegative integers k1, . . . , kn. We then have that (Lr j=1(e(j) 11 S)kj ) ⊕ C ∼= C for some finitely generated projective right S-semimodule C. Since S is finite and C is a finitely generated right S-semimodule, C is also finite. We deduce that Lr j=1(e(j) We now consider the case when F is an arbitrary zerosumfree semifield. There exists a surjective semiring homomorphism π : F → B, defined by π(0) = 0 and π(x) = 1 for all 0 6= x ∈ F , which induces a surjective semiring homomorphism θ : S → Mn1(B) × . . . × Mnr (B) =: T . We then have a commutative diagram: 11 S)kj = 0, whence kj = 0 for all j. Therefore,  is injective. (j)S) = Pr \e11 j=1 kj SK0(S) SK0(θ) SK0(T )   K0(S) K0(θ) / K0(T ) Assume that (Pr (j)S) = 0 ∈ K0(S) for some integers k1, . . . , kn ≥ 0. Since the above diagram is commutative and  : SK0(T ) → K0(T ) is injective, we \e11 j=1 kj / / / 30 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL have that SK0(θ)(Pr 0 ∈ SK0(T ); hence Lr This implies that kj = 0 for all j, whence  : SK0(S) → K0(S) is injective. (j)S) = 0 ∈ SK0(T ); that means, Pr \e11 (j)T = 11 T )kj ) ⊕ T m ∼= T m for some nonnegative integer m. j=1 kj j=1(e(j) j=1 kj \e11 (cid:3) Before stating the main results of this section, we name another useful notion. Definition 6.19. A semiring S is said be have cancellation of projectives if the monoid SV(S) is cancellative. Remark 6.20. Let us note the following simple facts. (1) Every semisimple semiring has cancellation of projectives. Indeed, this follows immediately from Proposition 6.15. (2) It is not hard to see that a semiring S has cancellation of projectives if and only if the natural map SV(S) → SK0(S), P 7−→ bP , is an injective monoid homomorphism (by using Lemma 6.5). This shows that we may consider SV(S) to be the cone SK0(S)+ in SK0(S), for any semiring S having cancellation of projectives. (3) Let F be a semifield and S a ultramatricial F -algebra. Then S has cancel- lation of projectives. Indeed, this follows immediately from Remark 5.8. Now we are ready to present the main theorems of this section. Theorem 6.21. Let S and T be ultramatricial algebras over a semifield F . Then (SK0(S), bS) ∼= (SK0(T ), bT ) if and only if S ∼= T as F -algebras. Proof. (⇐=). It is clear. (=⇒). By Remark 6.20 (2) and (3), we obtain that SV(S) = SK0(S)+ and SV(T ) = SK0(T )+. Let f : (SK0(S), bS) → (SK0(T ), bT ) be an isomorphism in the category P. We then have that f (SK0(S)+) = SK0(T )+, and hence, f (SV(S)) = SV(T ). This implies that f induces an isomorphism from (SV(S), S) onto (SV(T ), T ) in category C. Now, applying Theorem 5.10, we immediately get that S ∼= T as F -algebras, as desired. (cid:3) Theorem 6.21 is, in general, not valid for semisimple semirings (even, additively idempotent ones). For example, (SK0(B), bB) ∼= (Z, 1) ∼= (SK0(T), bT), but the Boolean semifield B and the tropical semifield T are not isomorphic. However, as a corollary of Theorem 6.21, the following result permits us to classify zerosumfree congruence-semisimple semirings in terms of their SK0-groups. Theorem 6.22. Let S and T be zerosumfree congruence-semisimple semirings. Then (SK0(S), bS) ∼= (SK0(T ), bT ) if and only if S ∼= T as B-algebras. Proof. By Corollary 3.5, every zerosumfree semiring is a matricial B-algebra. Also, every matrical B-algebra is an ultramatricial B-algebra. From these observations and Theorem 6.21, we immediately get the statement. (cid:3) Finally, according to Propositions 6.3, 6.15 and 6.18, as well as Theorem 6.21, we obtain the following result. Theorem 6.23 (cf. [13, Thm. 15.26]). Let S and T be ultramatricial algebras over a semifield F . Then, there exists a group isomorphism f : K0(S) → K0(T ) such that f ([S]) = [T ] and f (K0(S)+) = K0(T )+ if and only if S ∼= T as F -algebras. CONGRUENCE-SEMISIMPLE SEMIRINGS AND K0-GROUP CHARACTERIZATION 31 Proof. (⇐=). It is obvious. (=⇒). Assume that f : K0(S) → K0(T ) is a group homomorphism such that f ([S]) = [T ] and f (K0(S)+) = K0(T )+. Let S be the union of an ascending sequence S1 ⊆ S2 ⊆ · · · of matricial subalgebras, and T the union of an ascending sequence T1 ⊆ T2 ⊆ · · · of matricial subalgebras. We then have a commutative diagram in the category of abelian groups as follows: SK0(S1) SK0(S2) SK0(S3) · · ·    K0(S1) / K0(S2) / K0(S3) / · · · From Proposition 6.18 we have that the homomorphisms  : SK0(Sn) → K0(Sn) are injective for all n. Then, taking into account Propositions 6.3 and 6.15, we may consider SK0(S) to be a subgroup of K0(S); and similarly, SK0(T ) may be consid- ered as a subgroup of K0(T ). Then, by our hypothesis, we immediately get that f induces an isomorphism from (SK0(S), bS) onto (SK0(T ), bT ) in the category P, whence we obtain the statement by Theorem 6.21. (cid:3) Motivated by Elliott's program and Theorems 6.22 and 6.23, we finish this article by posting the following problem. Problem 4. Describe all additively idempotent semirings for which the SK0- groups and K0-groups are complete invariants. References [1] J. Y. Abuhlail, S. N. Il'in, Y. Katsov and T. G. Nam, On V-semirings and semirings all of whose cyclic semimodules are injective, Comm. Algebra 43 (2015), 4632 -- 4654. [2] M. Akian, S. Gaubert and A. Guterman, Linear independence over tropical semirings and beyond, in: Tropical and idempotent mathematics, Contemp. Math., vol. 495, Amer. Math. Soc., Providence, RI, 2009, pp. 1 -- 38. [3] P. Ara and K. R. Goodearl, The realization problem for some wild monoids and the Atiyah problem, Trans. Amer. Math. Soc. 369 (2017), 5665 -- 5710. [4] R. El Bashir, J. Hurt, A. Janca r´ık and T. Kepka, Simple commutative semirings, J. Algebra 236 (2001), 277 -- 306. [5] H. Bass, Algebraic K-theory, W. A. Benjamin, Inc., New York-Amsterdam, 1968. [6] A. Connes and C. Consani, Schemes over F1 and zeta functions, Compos. Math. 146 (2010), 1383 -- 1415. [7] A. Di Nola and C. Russo, The semiring-theoretic approach to MV-algebras: a survey, Fuzzy Sets and Systems 281 (2015), 134 -- 154. [8] G. A. Elliott, On the classification of inductive limits of sequences of semisimple finite- dimensional algebras, J. Algebra 38 (1976), 29 -- 44. [9] G. A. Elliott and A. S. Toms, Regularity properties in the classification program for separable amenable C ∗-algebras, Bull. Amer. Math. Soc. (N.S.) 45:2 (2008), 229 -- 245. tropical varieties, Duke [10] J. Giansiracusa and N. Giansiracusa, Equations of Math. J. 165:18 (2016), 3379 -- 3433. [11] K. G lazek, A guide to the literature on semirings and their applications in mathemat- ics and information sciences, Dordrecht-Boston-London: Kluwer Academic Publish- ers, 2002. [12] J. S. Golan, Semirings and their Applications, Kluwer Academic Publishers, Dordrecht-Boston-London, 1999. [13] K. R. Goodearl, von Neumann regular rings, Pitman, London, 1979. / / / / / / / / / 32 YEFIM KATSOV, TRAN GIANG NAM, AND JENS ZUMBR AGEL [14] G. Gratzer, Universal Algebra, 2nd ed., Springer-Verlag, New York-Berlin, 1979. [15] U. Hebisch and H. J. Weinert, On the rank of semimodules over semirings, Collectanea Mathematica 46 (1998), 83 -- 95. [16] U. Hebisch and H. J. Weinert, Semirings and Semifields, in: Handbook of Algebra, Vol. 1, Elsevier, Amsterdam, 1996, pp. 425 -- 462. [17] A. Horn and N. Kimura, The category of semilattices, Algebra Universalis 1 (1971), 26 -- 38. [18] S. N. Il'in, V-semirings, Sib. Math. J. 53:2 (2012), 222 -- 231. [19] S. N. Il'in and Y. Katsov, On Serre's problem on projective semimodules over poly- nomial semirings, Comm. Algebra 42 (2014), 4021 -- 4032. [20] S. N. Il'in, Y. Katsov and T. G. Nam, Toward homological structure theory of semi- modules: On semirings all of whose cyclic semimodules are projective, J. Algebra 476 (2017), 238 -- 266. [21] Z. Izhakian and L. Rowen, Supertropical algebra, Adv. Math. 225:4 (2010), 2222 -- 2286. [22] Z. Izhakian, M. Knebusch and L. Rowen, Decompositions of modules lacking zero sums, Israel J. Math., to appear; see also arXiv:1511.04041. [23] Z. Izhakian, M. Johnson and M. Kambites, Pure dimension and projectivity of tropical polytopes, Adv. Math. 303:5 (2016), 1236 -- 1263. [24] Y. Katsov, Tensor products and injective enveloples of semimodules over additively regular semirings, Algebra Colloquium 4 (1997), 121 -- 131. [25] Y. Katsov, Toward homological characterization of semirings: Serre's conjecture and Bass's perfectness in a semiring context, Algebra Universalis 52 (2004), 197 -- 214. [26] Y. Katsov, R. Lipyanski and B. Plotkin, Automorphisms of categories of free modules, free semimodules, and free Lie modules, Comm. Algebra 35 (2007), 931 -- 952. [27] Y. Katsov and T. G. Nam, Morita Equivalence and Homological Characterization of Semirings, J. Algebra Appl. 10 (2011), 445 -- 473. [28] Y. Katsov, T. G. Nam and N. X. Tuyen, On subtractive semisimple semirings, Algebra Colloquium 16 (2009), 415-426. [29] T. Y. Lam, A first course in noncommutative rings, 2nd ed., Springer-Verlag, New York-Berlin, 2001. [30] T. Y. Lam, Serre's problem on projective modules, Springer Monographs in Mathe- matics, Springer-Verlag, Berlin, 2006. [31] O. Lorscheid, The geometry of blueprints: Part I: Algebraic background and scheme theory, Adv. Math. 229:3 (2012), 1804 -- 1846. [32] S. Mac Lane, Categories for the Working Mathematician, Springer-Verlag, New York- Berlin, 1971. [33] A. W. Macpherson, Projective modules over polyhedral semirings, arXiv:1507.07213v1. [34] A. Patchkoria, Projective semimodules over semirings with valuations in nonnegative integers, Semigroup Forum 79 (2009), 451 -- 460. [35] J. Richter-Gebert, B. Sturmfels and T. Theobald, First steps in tropical geometry, in: Idempotent Mathematics and Mathematical Physics, Contemp. Math., vol. 377, Amer. Math. Soc., Providence, RI, 2005, pp. 289 -- 317. Department of Mathematics, Hanover College, Hanover, IN 47243 -- 0890, USA E-mail address: [email protected] Institute of Mathematics, VAST, 18 Hoang Quoc Viet, Cau Giay, Hanoi, Vietnam E-mail address: [email protected] Faculty of Computer Science and Mathematics, University of Passau, Germany E-mail address: [email protected]
1102.1776	1	1102	2011-02-09T04:01:27	Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra	[ "math.RA" ]	In this paper, we considered the theory of quasideterminants and row and column determinants. We considered the application of this theory to the solving of a system of linear equations in quaternion algebra. We established correspondence between row and column determinants and quasideterminants of matrix over quaternion algebra.	math.RA	math	Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei Abstract. In this paper, we considered the theory of quasideterminants and row and column determinants. We considered the application of this theory to the solving of a system of linear equations in quaternion algebra. We es- tablished correspondence between row and column determinants and quaside- terminants of matrix over quaternion algebra. Contents Introduction to the Theory of the Row and Column Determinants 1. Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Convention about Notations . . . . . . . . . . . . . . . . . . . . . . . . 3. Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. Quaternion Algebra 5. . . 6. Quasideterminant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. Correspondence between Row-Column Determinants and Quasidetermi- nants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8. References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10. Special Symbols and Notations . . . . . . . . . . . . . . . . . . . . . . 1 2 3 5 6 15 17 19 21 22 1. Preface Linear algebra is a powerful tool that we use in different areas of mathematics, in- cluding calculus, analytic and differential geometry, theory of differential equations, optimal control theory. Linear algebra has accumulated a rich set of different meth- ods. Since some methods have a common final result, this gives us the opportunity to choose the most effective method, depending on the nature of calculations. At transition from linear algebra over a field to linear algebra over division ring, we want to save as much as possible tools that we regularly use. Already in the 2010 Mathematics Subject Classification. Primary 15A33, 15A15, 15A24. Key words and phrases. quaternion algebra, column determinant, quasideterminant, system of linear equations, Cramer's rule. Aleks [email protected]. [email protected]. Pidstrygach Institute for Applied Problems of Mechanics and Mathematics, str.Naukova 3b, Lviv, Ukraine, 79005. 1 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei early XX century, shortly after Hamilton created quaternion algebra, mathemati- cians began to search the answer how looks like the algebra with noncommutative multiplication. In particular, there is a problem how to determine a determinant of a matrix with elements belonging to noncommutative ring. Such determinant is also called noncommutative determinant. There were a lot of approaches to the definition of the noncommutative determi- nant. However none of the introduced noncommutative determinants maintained all those properties that determinant possessed for matrices over a field. Moreover, in paper [11], J. Fan proved that there is no unique definition of determinant which would expands the definition of determinant of real matrices for matrices over di- vision ring of quaternions. Therefore, search for a solution of the problem to define a noncommutative determinant is still going on. In this paper, we consider two approaches to define noncommutative determi- nant. Namely, we explore row-column determinants and quasideterminant. Row-column determinants are an extension of the classical definition of the deter- minant, however we assume predetermined order of elements in each of the terms of the determinant. Using row-column determinants, we obtain the solution of system of linear equations over quaternion algebra according to Cramer's rule. Quasideterminant appeared from the analysis of the procedure of matrix in- version. Using quasideterminant, the procedure of solving of a system of linear equations over quaternion algebra is similar to the method of Gauss. There is common in definition of row and column determinants and quasideter- In both cases, we have not one determinant in correspondence to qua- minant. dratic matrix of order n with noncommutative entries, but certain set (there are n2 quasideterminant, n row determinants, and n column determinants). Today there is wide application of quasideterminants in linear algebra ([18, 21]), and in physics ([12, 13, 20]). Row and column determinants ([14]) introduced rel- atively recently are less well known. Purpose of this paper is establishment of the correspondence between row and column determinants and quasideterminants of matrix over quaternion algebra. The authors are hopeful that the establishment of this correspondence can provide mutual development of both theory quasidetermi- nants and theory of row and column determinants. 2. Convention about Notations There are different forms to write elements of matrix. In this paper, we denote aij an element of the matrix A. Index i labels rows, and index j labels columns. We use the following notation for different minors of the matrix A ai . AS . : the row with index i : the minor obtained from A by selecting rows with index from the set S Ai . AS . S a. j A. T : the minor obtained from A by deleting row ai . : the minor obtained from A by deleting rows with index from the set : column with index j : the minor obtained from A by selecting columns with index from the set T A. j : the minor obtained from A by deleting column a. j 2 Aleks Kleyn and Ivan Kyrchei Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra A. T : the minor obtained from A by deleting columns with index from the set T A.j (b) : the matrix obtained from matrix A by replacing its column with number j by column b Ai. (b) : the matrix obtained from matrix A by replacing its row with number i by row b Considered notations can be combined. For instance, the record Aii k.(b) means replacing of the row with number k by matrix b followed by removal of the row with number i and by removal of the column with number i. As was noted in section [17]-2.2, we can define two types of matrix products: either product of rows of first matrix over columns of second one, or product of columns of first matrix over rows of second one. However according to the theorem [17]-2.2.5 this product is symmetric relative operation of transposition. Hence in the paper, we will restrict ourselves by traditional product of rows of first matrix over columns of second one; and we do not indicate clearly the operation like it was done in [17]. 3. Preliminaries Theory of determinants of matrices with noncommutative elements can be di- vided into three groups regarding their methods of definition. Denote M(n, K) the ring of matrices with elements from the ring K. One of the ways to determine determinant of a matrix of M (n, K) is following ([1, 4, 5]). Definition 3.1. Let the functional d : M (n, K) → K satisfy the following axioms. Axiom 1. d (A) = 0 iff A is singular (irreversible). Axiom 2. ∀A, B ∈ M (n, K), d (A · B) = d (A) · d (B). Axiom 3. If we obtain a matrix A′ from matrix A either by adding of an arbitrary row multiplied on the left with its another row or by adding of an arbitrary column multiplied on the right with its another column, then d (A′) = d (A) Then the value of the functional d is called determinant of matrix A ∈ M (n, K). (cid:3) Known determinants of Dieudonn´e and Study are examples of such functional. In [1], Aslaksen proved that determinants that satisfy axioms 1, 2, 3, take their value in some commutative subset of the ring. Also it makes no sense for them such property of conventional determinants as the expansion relative to minors of a row or column. Therefore, the determinant representation of the inverse matrix using only these determinants is impossible. This is the reason that causes to introduce determinant functionals that do not satisfy all the above axioms. However in [5], Dyson considers the axiom 1 as necessary to determine the determinant. In another approach, the determinant of a square matrix over noncommutative ring is considered as a rational function of the entries of matrix. The greatest success is achieved by Gelfand and Retakh in the theory of quasideterminants ([9, 10]). We present introduction into the theory of quasideterminants in section 6. 3 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei In third approach, the determinant of a square matrix over noncommutative ring is considered as alternating sum of n! products of the entries of matrix; however, it assumed certain fixed order of factors in each term. E. H. Moore was first who achieved implementation of the key axiom 1 using such definition of noncommu- tative determinant. Moore had done this not for all square matrices, but only for Hermitian. He defined the determinant of Hermitian matrix1 A = (aij )n×n over division ring with involution by induction over n following way (see [5]) (3.1) Here εkj = ( 1, −1, MdetA =  i = j i 6= j a11, n n = 1 εijaij Mdet (A(i → j)) , n > 1 Pj=1 , and using A(i → j) we denoted the matrix obtained from the matrix A by successive replacement of column with number j by col- umn with number i and removing from matrix of row with number i and column with number i. Another definition of this determinant is presented in [1] using permutations: Mdet A = Xσ∈Sn σan11n12 · . . . · an1l1 n11 ·an21n22 · . . . · anrl1 nr1 . Here Sn is symmetric group of n elements. A cycle decomposition of a permutation σ has form: σ = (n11 . . . n1l1 ) (n21 . . . n2l2) . . . (nr1 . . . nrlr ) . However, there was no any generalization of the definition of determinant by Moore for arbitrary square matrix. Freeman J. Dyson pointed out the importance of this problem in the paper [5]. In papers [2, 3], L. Chen offered the following definition of determinant of square matrix over division ring of quaternions A = (aij ) ∈ M (n, H): det A = Pσ∈Sn ε (σ) an1i2 · ai2i3 . . . · aisn1 · . . . · anr k2 · . . . · aklnr , σ = (n1i2 . . . is) . . . (nrk2 . . . kr) , n1 > i2, i3, . . . , is; . . . , nr > k2, k3, . . . , kl, n = n1 > n2 > . . . > nr ≥ 1. Despite the fact that this determinant does not satisfy the axiom 1, L. Chen re- ceived the determinant representation of the inverse matrix. However we cannot also expand his determinant relative to minors of rows or column (except for n- th row). Therefore, L. Chen also did not receive a classical adjoint matrix. For A = (α1, . . . , αm) over division ring of quaternions H, if kAk := det(A∗A) 6= 0, then ∃A−1 = (bjk), where bjk = 1 kAk ωkj , (j, k = 1, 2, . . . , n) , ωkj = det (α1 . . . αj−1αnαj+1 . . . αn−1δk)∗ (α1 . . . αj−1αnαj+1 . . . αn−1αj ) . 1Hermitian matrix is such matrix A = (aij ) that aij = aji. 4 Aleks Kleyn and Ivan Kyrchei Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Here αi is i-th column of A, δk is n-dimensional column with 1 in k-th row and 0 in other ones. L. Chen defined kAk := det(A∗A), as double determinant. The solution of the right system of linear equations αjxj = β Xn j=1 over H in case kAk 6= 0 represented by the following formula, which the author calls Cramer's rule xj = kAk−1Dj , ∀j = 1, n where Dj = det α∗ 1 ... α∗ α∗ n α∗ ... α∗ β ∗   j−1 j+1 n−1   (cid:16) α1 . . . αj−1 αn αj+1 . . . αn−1 αj (cid:17) . Here αi is i-th column of matrix A, α∗ vector-row conjugated with β. i is i-th row of matrix A∗, β ∗ is n-dimensional In this paper, we explore the theory of row and column determinants which develops the classical approach to the definition of the determinant of the matrix, but with a predetermined order of factors in each of the terms of the determinant. 4. Quaternion Algebra A quaternion algebra H(a, b) (we also use notation (cid:18) a, b F (cid:19) ) is four- dimensional vector space over the field F with basis {1, i, j, k} and the following multiplication rules: i2 = a, j2 = b, ij = k, ji = −k. The field F is the center of the quaternion algebra H(a, b). In the algebra H(a, b) there are following mappings. • A quadratic form such that n : x ∈ H → n(x) ∈ F n(x · y) = n(x)n(y) x, y ∈ H is called the norm on quaternion algebra H. • The linear mapping t : x = x0 + x1i + x2j + x3k ∈ H → t(x) = 2x0 ∈ F 5 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei is called trace of quaternion. Trace satisfies permutability property of the trace: From the theorem [17]-10.3.3, it follows t (q · p) = t (p · q) (4.1) (4.2) • A linear mapping t(x) = 1 2 (x − ixi − jxj − kxk) x → x = t(x) − x is an involution. Involution has following properties x = x x + y = x + y x · y = y · x A quaternion x is called the conjugate of quaternion x ∈ H. Norm and involution satisfy the following condition: n (q) = n(q) Trace and involution satisfy the following condition: From equations (4.1), (4.2), it follows that t (x) = t(x) x = − 1 2 (x + ixi + jxj + kxk) Depending on the choice of the field F, a and b, on the set of quaternion algebra there are only two possibilities [19]: 1. (cid:18) a, b 2. (cid:18) a, b F (cid:19) is a division algebra. F (cid:19) is isomorphic to the algebra of all 2 × 2 matrices with entries from the field F. In this case, quaternion algebra is splittable. Consider some non-isomorphic quaternion algebra with division. R (cid:19), where R is real field, is isomorphic to the Hamilton quaternion skew 1. (cid:18) a, b field H whenever a < 0 and b < 0. Otherwise (cid:18) a, b division quaternion algebras (cid:18) a, b R (cid:19) is splittable. 2. If F is the rational field Q, then there exist infinitely many nonisomorphic Q (cid:19) depending on choice of a < 0 and b < 0. 3. Let Qp be the p-adic field where p is a prime number. For each prime number p there is a unique division quaternion algebra. 5. Introduction to the Theory of the Row and Column Determinants The row and column determinants are introduced for matrices with entries in quaternion algebra H. To introduce the row and column determinants, we need the following definitions in the theory of permutations. 6 Aleks Kleyn and Ivan Kyrchei Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Definition 5.1. Let Sn be symmetric group on the set In = {1, 2, ..., n} ([6], p. 30). If two-line notation of a permutation σ ∈ Sn corresponds to its some cycle notation, then we say that the permutation σ ∈ Sn forms the direct product of disjoint cycles (5.1) σ = n11 n12 n12 n13 . . . n1l1 . . . n11 . . . nr1 nr2 . . . nr2 nr3 . . . nrlr . . . nr1! (cid:3) Definition 5.2. We say that cycle notation (5.1) of the permutation σ is left- ordered, if elements, which closes each of its independent cycles, are recorded first on the left in each cycle σ = (n11n12 . . . n1l1 ) (n21n22 . . . n2l2 ) . . . (nr1nr2 . . . nrlr ) . (cid:3) Definition 5.3. We say that cycle notation (5.1) of the permutation σ is right- ordered, if elements, which closes each of its independent cycles, are recorded first on the right in each cycle σ = (n12 . . . n1l1 n11) (n22 . . . n2l2 n21) . . . (nr2 . . . nrlr nr1) . (cid:3) Definition 5.4. Let Sn be symmetric group on the set In = {1, 2, ..., n}. Let i ∈ 1, n. The ith row determinant of matrix A = (aij) ∈ M (n, H) is defined as expression (−1)n−rai ik1 aik1 ik1 +1 . . .aik1 +l1 i . . . aikr ikr +1 . . . aikr +lr ikr rdetiA = Xσ∈Sn This expression is defined as the alternative sum of n! monomials compounded from entries of matrix A such that the index permutation σ ∈ Sn forms the direct product of disjoint independent cycles σ = (i ik1 ik1+1 . . . ik1+l1 ) (ik2 ik2+1 . . . ik2+l2) . . . (ikr ikr +1 . . . ikr +lr ) In left-ordered cycle notation of the permutation σ, the index i starts the first cycle from the left and other cycles satisfy the following conditions ik2 < ik3 < . . . < ikr , ikt < ikt+s, (cid:0)∀t = 2, r(cid:1) , (cid:0)∀s = 1, lt(cid:1) . (cid:3) Definition 5.5. Let Sn be symmetric group on the set Jn = {1, 2, ..., n}. Let j ∈ 1, n. The jth column determinant of matrix A = (aij ) ∈ M (n, H) is defined as expression (−1)n−r ajkr jkr +lr . . . ajkr +1ikr . . .aj jk1 +l1 . . . ajk1 +1jk1 ajk1 j. cdetj A = Xτ ∈Sn This expression is defined as the alternative sum of n! monomials compounded from entries of matrix A such that the index permutation τ ∈ Sn forms the direct product of disjoint independent cycles τ = (jkr +lr . . . jkr +1jkr ) . . . (jk2+l2 . . . jk2+1jk2 ) (jk1+l1 . . . jk1+1jk1 j) 7 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei In right-ordered cycle notation of the permutation τ , the index j starts the first cycle from the right and other cycles satisfy the following conditions jk2 < jk3 < . . . < jkr , jkt < jkt+s, (cid:0)∀t = 2, r(cid:1) , (cid:0)∀s = 1, lt(cid:1) . (cid:3) Remark 5.6. The peculiarity of the column determinant is that, at the direct cal- culation, factors of each of the monomials are written from right to left. (cid:3) In lemmas 5.7 and 5.8, we consider recursive definition of column and row de- terminants. This definition is an analogue of the expansion of determinant by rows and columns in commutative case. Lemma 5.7. Let Ri j be the right ijth cofactor of entry ai j of matrix A = (aij) ∈ M (n, H), namely Then rdeti A = n Pj=1 ai j · Ri j i = 1, n .j(a. i)) rdetk Aii Ri j =( −rdetj (Aii k =(2 i = 1 i > 1 1 i 6= j i = j where we obtain the matrix (Aii .j(a. i)) from A by replacing its jth column with the ith column b and after that by deleting both the ith row and the ith column. (cid:3) Lemma 5.8. Let Li j be the left ijth cofactor of entry ai j of matrix A = (aij ) ∈ M (n, H), namely Then cdetj A = n Pi=1 Li j · ai j j = 1, n i. (aj .)) cdetk Ajj Li j =( −cdeti (Ajj k =(2 j = 1 j > 1 1 i 6= j i = j where we obtain the matrix (Ajj jth and after that by deleting both the jth row and the jth column. i. (aj .)) from A by replacing its ith row with the (cid:3) Remark 5.9. Clearly, any monomial of each row or column determinant of a square matrix corresponds to a certain monomial of another row or column determinant such that both of them have the same sign, consists of the same factors and differ only in their ordering. If the entries of an arbitrary matrix A are commutative, then rdet1 A = . . . = rdetnA = cdet1 A = . . . = cdetnA. (cid:3) Consider the basic properties of the column and row determinants of a square matrix over H. Their proofs immediately follow from the definitions. 8 Aleks Kleyn and Ivan Kyrchei Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Theorem 5.10. If one of the rows (columns) of the matrix A ∈ M (n, H) consists of zeros only, then for all i = 1, n, rdeti A = 0, cdeti A = 0. Theorem 5.11. If the ith row of the matrix A ∈ M (n, H) is left-multiplied by b ∈ H, then for all i = 1, n rdeti Ai .(b · a. i) = b · rdeti A. Theorem 5.12. If the jth column of the matrix A ∈ M (n, H) is right-multiplied by b ∈ H, then for all j = 1, n cdetj A. j (aj . · b) = cdetj A · b. Theorem 5.13. If for A ∈ M (n, H) there exists index k ∈ In such that akj = bj +cj for all j = 1, n, then for any i = 1, n rdeti A = rdeti Ak . (b) + rdeti Ak . (c) , cdeti A = cdeti Ak . (b) + cdeti Ak . (c) . where b = (b1, . . . , bn), c = (c1, . . . , cn). Theorem 5.14. If for A ∈ M (n, H) there exists index k ∈ In such that ai k = bi+ci for all i = 1, n, then for any j = 1, n rdetj A = rdetj A. k (b) + rdetj A. k (c) , cdetj A = cdetj A. k (b) + cdetj A. k (c) . where b = (b1, . . . , bn)T , c = (c1, . . . , cn)T . Theorem 5.15. Let A∗ be the Hermitian adjoint matrix of A ∈ M (n, H), then rdeti A∗ = cdeti A for any i = 1, n. (cid:3) The following theorem is crucial in the theory of the row and column determi- nants. Theorem 5.16. If A = (aij) ∈ M (n, H) is a Hermitian matrix, then rdet1A = . . . = rdetnA = cdet1A = . . . = cdetnA ∈ F. (cid:3) Remark 5.17. Since all column and row determinants of a Hermitian matrix over division ring H are equal, we can define the determinant of a Hermitian matrix A ∈ M (n, H) det A := rdeti A = cdeti A, (cid:0)∀i = 1, n(cid:1) . (cid:3) Remark 5.18. Since we represent determinant of the Hermitian matrix as i-th row determinant (i is arbitrary), then according to lemma 5.7, we have (5.2) k =(2 1 ai j · rdetj (Aii .j(a. i)) + ai i · rdetk Aii det A = − Pσ∈Sn i = 1 i > 1 By comparing expressions (3.1) and (5.2) for Hermitian matrix A ∈ M (n, H), we conclude that the row determinant of a Hermitian matrix coincides with the 9 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei Moore determinant. Hence the row and column determinants extend the Moore determinant to an arbitrary square matrix. (cid:3) Properties of the determinant of a Hermitian matrix is completely explored in [14] by its row and column determinants. Among all, consider the following. Theorem 5.19. If the ith row of the Hermitian matrix A ∈ M (n, H) is replaced with a left linear combination of its other rows ai . = c1ai1 . + . . . + ckaik . where cl ∈ H for all l = 1, k and {i, il} ⊂ In, then for all i = 1, n cdetiAi . (c1 · ai1 . + . . . + ck · aik . ) = rdetiAi . (c1 · ai1 . + . . . + ck · aik . ) = 0. (cid:3) Theorem 5.20. If the jth column of a Hermitian matrix A ∈ M (n, H) is replaced with a right linear combination of its other columns a. j = a. j1 c1 + . . . + a. jk ck where cl ∈ H for all l = 1, k and {j, jl} ⊂ Jn , then for all j = 1, n cdetj A. j (a. j1 · c1 + . . . + a. jk · ck) = rdetj A. j (a. j1 · c1 + . . . + a. jk · ck) = 0. (cid:3) The following theorem on the determinantal representation of the matrix inverse of the Hermitian one follows directly from these properties. Theorem 5.21. There exist a unique right inverse matrix (RA)−1 and a unique left inverse matrix (LA)−1 of a nonsingular Hermitian matrix A ∈ M (n, H), (det A 6= 0), where (RA)−1 = (LA)−1 =: A−1. Right inverse and left inverse matrices has following determinantal representation R11 R21 R12 R22 · · · · · · R1n R2n L11 L21 L12 L22 · · · · · · L1n L2n     · · · Rn1 · · · Rn2 · · · · · · · · · Rnn · · · Ln1 · · · Ln2 · · · · · · · · · Lnn     . (RA)−1 = 1 det A (LA)−1 = 1 det A where Rij , Lij are right and left ij-th cofactors of A respectively for all i, j = 1, n. (cid:3) To obtain the determinantal representation for an arbitrary inverse matrix over division ring H, we consider the right AA∗ and left A∗A corresponding Hermitian matrix. Theorem 5.22 ([14]). If an arbitrary column of matrix A ∈ Hm×n is a right linear combination of its other columns, or an arbitrary row of matrix A∗ is a left linear combination of its other rows, then det A∗A = 0. (cid:3) 10 Aleks Kleyn and Ivan Kyrchei Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Since the principal submatrices of a Hermitian matrix are also Hermitian, then the basis principal minor may be defined in this noncommutative case as a principal nonzero minor of maximal order. We also introduce a notion the rank of a Her- mitian matrix by principal minors, as maximal order of a principal nonzero minor. The following theorem establishes the correspondence between the rank by principal minors of a Hermitian matrix and the rank of matrix defined as the max- imum number of right-linearly independent columns or left-linearly independent rows, which form a basis. Theorem 5.23 ([14]). A rank by principal minors of a Hermitian matrix A∗A is equal to its rank and a rank of A ∈ Hm×n. (cid:3) Theorem 5.24 ([14]). If A ∈ Hm×n, then arbitrary column of matrix A is a right linear combination of its basic columns or arbitrary row of matrix A is a left linear combination of its basic rows. (cid:3) This implies a criterion for the singularity of a corresponding Hermitian matrix. Theorem 5.25 ([14]). The right linearly independence of columns of the matrix A ∈ Hm×n or the left linearly independence of rows of A∗ is the necessary and sufficient condition for det A∗A 6= 0 Theorem 5.26 ([14]). If A ∈ M (n, H), then det AA∗ = det A∗A. (cid:3) Example 5.27. Consider the matrix a21 a22! A = a11 a12 a12 a22! A∗ = a11 a21 Then Respectively, we have AA∗ = a11a11 + a12a12 a11a21 + a12a22 a21a11 + a22a12 a21a21 + a22a22! a12a11 + a22a21 a12a12 + a22a22! A∗A = a11a11 + a21a21 a11a12 + a21a22 To assess the determinants of obtained Hermitian matrices, we consider for example the first row determinant of each matrix. According to the theorem 5.16 and the remark 5.17 we have det AA∗ = rdet1AA∗ det A∗A = rdet1A∗A 11 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra According to the lemma 5.7 Aleks Kleyn and Ivan Kyrchei (5.3) (5.4) det AA∗ = (AA∗)11(AA∗)22 − (AA∗)12(AA∗)21 = (a11a11 + a12a12)(a21a21 + a22a22) −(a11a21 + a12a22)(a21a11 + a22a12) = a11a11a21a21 + a12a12a21a21 +a11a11a22a22 + a12a12a22a22 −a11a21a21a11 − a12a22a21a11 −a11a21a22a12 − a12a22a22a12 = a12a12a21a21 + a11a11a22a22 −a12a22a21a11 − a11a21a22a12 det A∗A = (A∗A)11(A∗A)22 − (A∗A)12(A∗A)21 = (a11a11 + a21a21)(a12a12 + a22a22) −(a11a12 + a21a22)(a12a11 + a22a21) = a11a11a12a12 + a21a21a12a12 +a11a11a22a22 + a21a21a22a22 −a11a12a12a11 − a21a22a12a11 −a11a12a22a21 − a21a22a22a21 = a21a21a12a12 + a11a11a22a22 −a21a22a12a11 − a11a12a22a21 Positive terms in the equations (5.3), (5.4) are real numbers and they obviously coincide. To prove equation (5.5) a12a22a21a11 + a11a21a22a12 = a21a22a12a11 + a11a12a22a21 we use the rearrangement property of the trace of elements of the quaternion alge- bra, t(pq) = t(qp). Indeed, a12a22a21a11 + a11a21a22a12 = a12a22a21a11 + a12a22a21a11 = t(a12a22a21a11), a21a22a12a11 + a11a12a22a21 = a11a12a22a21 + a11a12a22a21 = t(a11a12a22a21) Then by the rearrangement property of the trace, we obtain (5.5). (cid:3) According to the theorem 5.26 we introduce the concept of double determinant. For the first time this concept was introduced by L. Chen ([2]). Definition 5.28. Determinant of the Hermitian matrix AA∗ is called double determinant of the matrix A ∈ M (n, H) ddetA := det (A∗A) = det (AA∗) (cid:3) If H is the classical quaternion skew field H over the real field, then the following theorem establishes the validity of Axiom 1 for the double determinant. Theorem 5.29. If {A, B} ⊂ M (n, H), then ddet (A · B) = ddetA · ddetB. (cid:3) Unfortunately, if non-Hermitian matrix is not full rank, then nothing can be said about singularity of its row and column determinant. We show it in the following example. 12 Aleks Kleyn and Ivan Kyrchei Example 5.30. Consider the matrix Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra A = i j −i! j Its second row is obtained from the first row by left-multiplying by k. Then, by Theorem 5.25, ddetA = 0. Indeed, A∗A = −i −j i ! · i 2 ! . j −i! = 2 −2k −j 2k j Then ddetA = 4 + 4k2 = 0; however cdet1A = cdet2A = rdet1A = rdet2A = −i2 − j2 = 2 At the same time rankA = 1, that corresponds to the theorem 5.23. (cid:3) The correspondence between the double-determinant and the non-commutative determinants of Moore, Stady and Dieudonn´e are obtained ddetA = Mdet (A∗A) = SdetA = Ddet2A Definition 5.31. Let ddetA = cdetj (A∗A) =Pi Lij · aij ∀j = 1, n then Lij A ∈ M (n, H). is called the left double ijth cofactor of entry aij of the matrix (cid:3) Definition 5.32. Let ddetA = rdeti (AA∗) =Pj aij · Rij ∀i = 1, n then Rij A ∈ M (n, H). is called the right double ijth cofactor of entry aij of the matrix (cid:3) Theorem 5.33. The necessary and sufficient condition of invertibility of matrix A = (aij ) ∈ M(n, H) is ddetA 6= 0. Then ∃A−1 = (LA)−1 = (RA)−1, where (5.6) (LA)−1 = (A∗A)−1 A∗ = 1 ddetA (5.7) (RA)−1 = A∗ (AA∗)−1 = 1 ddetA∗ and Lij = cdetj(A∗A). j (A∗ Corollary 5.34. Let ddetA 6= 0. Then     L11 L21 L12 L22 . . . . . . L1n L2n R 11 R 21 R 12 R 22 . . . . . . R 1n R 2n . . . Lnn . . . Ln1 . . . Ln2 . . . . . .     j .(cid:1) ,(cid:0)∀i, j = 1, n(cid:1). . . . R n1 . . . R n2 . . . . . . . . . R nn . i), R ij = rdeti(AA∗)i.(cid:0)A∗ Lij = Rij 13 (cid:3) (cid:3) Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei Remark 5.35. In the theorem 5.33, the inverse matrix A−1 of an arbitrary matrix A ∈ M(n, H) under the assumption of ddetA 6= 0 is represented by the analog of the classical adjoint matrix. If we denote this analog of the adjoint matrix over H by Adj[[A]], then the next formula is valid over H: A−1 = Adj[[A]] ddetA . (cid:3) An obvious consequence of a determinantal representation of the inverse matrix by the classical adjoint matrix is the Cramer's rule. Theorem 5.36. Let (5.8) A · x = y be a right system of linear equations with a matrix of coefficients A ∈ M(n, H), a column of constants y = (y1, . . . , yn)T ∈ Hn×1, and a column of unknowns x = (x1, . . . , xn)T . If ddetA 6= 0, then the solution to the system of linear equations (5.8) has a unique solution that is represented by equation: (5.9) where f = A∗y. Theorem 5.37. Let (5.10) xj = cdetj(A∗A).j (f ) ddetA ∀j = 1, n (cid:3) x · A = y be a left system of linear equations with a matrix of coefficients A ∈ M(n, H), a column of constants y = (y1, . . . , yn) ∈ H1×n and a column of unknowns x = (x1, . . . , xn). If ddetA 6= 0, then the system of linear equations (5.10) has a unique solution that is represented by equation: (5.11) where z = yA∗. xi = rdeti (AA∗)i. (z) ddetA ∀i = 1, n (cid:3) Equations (5.9) and (5.11) are the obvious and natural generalizations of Cramer's rule for systems of linear equations over quaternion algebra. As follows from the theorem 5.21, the closer analog to Cramer's rule can be obtained in the following specific cases. Theorem 5.38. Let the matrix of coefficients A ∈ M(n, H) in the right system of linear equations (5.8) over division ring H be Hermitian. Then the system of linear equations (5.8) has the unique solution represented by the equation: xj = cdetj A.j (y) det A , (cid:0)∀j = 1, n(cid:1) . (cid:3) Theorem 5.39. Let the matrix of coefficients A ∈ M(n, H) in the left system of linear equations (5.10) over division ring H be Hermitian. Then the system of linear equations (5.10) has the unique solution represented by the equation: xi = rdetiAi. (y) det A , (cid:0)∀i = 1, n(cid:1) . 14 (cid:3) Aleks Kleyn and Ivan Kyrchei Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra In the framework of the row-column determinants also obtained Cramer's rule for the right AX = B, left XA = B and two-sided AXB = C matrix equations are obtained ([15]). Determinantal representations of Moore-Penrose inverse matrix and Cramer's rule for the normal solution of the left and right system of linear equations are given as well ([16]). Theorem 6.1. Suppose matrix 6. Quasideterminant a11 ... an1 ... a1n ... ... ... ann A =    has inverse matrix A−1.2 Then minor of inverse matrix satisfies the following equation, provided that the inverse matrices exist ((A−1)IJ )−1 = AJI − A.I J.(AJI )−1AJ. (6.1) .I Proof. Definition of inverse matrix leads to the system of linear equations (6.2) (6.3) AJI (A−1)I. J.(A−1)I. A.I .J + AJ. .I (A−1)IJ = 0 .J + AJI (A−1)IJ = E We multiply (6.2) by (cid:0)AJI(cid:1)−1 (A−1)I. (6.4) .J + (AJI )−1AJ. .I (A−1)IJ = 0 Now we can substitute (6.4) into (6.3) AJI (A−1)IJ − A.I (6.5) J.(AJI )−1AJ. .I (A−1)IJ = E (6.1) follows from (6.5). (cid:3) Corollary 6.2. Suppose matrix A has inverse matrix. Then elements of inverse matrix satisfy to the equation (6.6) ((A−1)ij )−1 = aji − A.i j.(Aji)−1Aj. .i Example 6.3. Consider matrix a21 a22! A = a11 a12 According to (6.6) (6.7) (6.8) (6.9) (6.10) (A−1)11 = (a11 − a12(a22)−1 a21)−1 (A−1)21 = (a21 − a22(a12)−1 a11)−1 (A−1)12 = (a12 − a11(a21)−1 a22)−1 (A−1)22 = (a22 − a21(a11)−1 a12)−1 (cid:3) (cid:3) 2This statement and its proof are based on statement 1.2.1 from [7] (page 8) for matrix over free division ring. 15 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei We call matrix (6.11) HA = ((HA)ij) = ((aji)−1) Hadamard inverse of matrix 3 a11 ... an1 ... a1n ... ... ... ann A =    Definition 6.4. (ji)-quasideterminant of the matrix A of n × n matrix A is formal expression (6.12) Aji = (HA−1)ji = ((A−1)ij)−1 We consider (ji)-quasideterminant as an element of the matrix A , which is called quasideterminant. (cid:3) Theorem 6.5. Expression for (ji)-quasideterminant has form (6.13) (6.14) Aji = aji − A.i Aji = aji − A.i j.(Aji)−1Aj. j. HAji Aj. .i .i Proof. The statement follows from (6.6) and (6.12). (cid:3) Theorem 6.6. Let (6.15) Then (6.16) A = 1 0 0 1! A−1 = 1 0 0 1! (A−1)11 = 1 and (A−1)22 = 1. Proof. It is clear from (6.7) and (6.10) that However expression for (A−1)21 and (A−1)12 cannot be defined from (6.8) and (6.9) since (a21 − a22(a12)−1 a11)−1 = (a12 − a11(a21)−1 a22)−1 = 0. We can transform these expressions. For instance (A−1)21 = (a21 − a22(a12)−1 a11)−1 = (a11((a11)−1 a12 − (a21)−1 a22))−1 = ((a21)−1 a11(a21(a11)−1 a12 − a22))−1 = (a11(a21(a11)−1 a12 − a22))−1 a21 It follows immediately that (A−1)21 = 0. (A−1)12 = 0. This completes the proof of (6.16). In the same manner we can find that (cid:3) From the proof of the theorem 6.6 we see that we cannot always use the equation (6.6) to find elements of inverse matrix and we need more transformations to solve this problem. From the theorem [17]-4.6.3, it follows that if a11 ... an1 rank  ≤ n − 2 ... a1n ... ... ... ann   16 3[8]-page 4 Aleks Kleyn and Ivan Kyrchei Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra then Aij , i = 1, ..., n, j = 1, ..., n, is not defined. From this, it follows that although quasideterminant is a powerful tool, use of a determinant is a major advantage. Theorem 6.7. Let matrix A have inverse matrix. Then for any matrices B and C equation (6.17) follows from the equation (6.18) B = C BA = CA Proof. Equation (6.17) follows from the equation (6.18) if we multiply both parts of the equation (6.18) over A−1. (cid:3) Theorem 6.8. Solution of nonsingular system of linear equations (6.19) is determined uniquely and can be presented in either form4 Ax = b (6.20) (6.21) x = A−1b x = HA b Proof. Multiplying both sides of equation (6.19) from left by A−1 we get (6.20). Using definition 6.4, we get (6.21). Since theorem 6.7 the solution is unique. (cid:3) 7. Correspondence between Row-Column Determinants and Quasideterminants Theorem 7.1. If A ∈ M(n, H) is an invertible matrix, then, for arbitrary p, q = 1, ..., n, we have the following representation of quasideterminant (7.1) (7.2) (7.3) (7.4) Apq = ddetA (Lpq)−1 = ddetA (cdetq(A∗A). q(A∗ . p))−1 = ddetA n(cdetq(A∗A). q(A∗ . p)) cdetq(A∗A). q(A∗ . p) Apq = ddetA (Rpq)−1 = ddetA (rdetp(AA∗)p .(A∗ q .))−1 = ddetA n(rdetp(AA∗)p .(A∗ q .)) rdetp(AA∗)p .(A∗ q .) Proof. Let A−1 = (bij ) be matrix inversed to the matrix A. The equation (6.12) reveals the relationship between quasideterminant A p,q of matrix A ∈ M(n, H) and elements of the inverse matrix A−1 = (bij ), namely A pq= b−1 qp for every p, q = 1, ..., n. At the same time, the theory of row and column determi- nants (the theorem 5.33) gives us representation of inverse matrix through its left (5.6) and right (5.7) double cofactors. Thus, accordingly, we obtain (7.5) A pq= b−1 qp =(cid:18) Lpq ddetA(cid:19)−1 = cdetq(A∗A). q(cid:0)A∗ . p(cid:1) ddetA !−1 , 4See similar statement in the theorem [7]-1.6.1. 17 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra (7.6) A pq= b−1 qp =(cid:18) Rpq ddetA(cid:19)−1 Aleks Kleyn and Ivan Kyrchei = rdetp(AA∗)p .(cid:0)A∗ q .(cid:1) ddetA !−1 . Since ddetA 6= 0 ∈ F, then ∃(ddetA)−1 ∈ F. It follows that cdetq(A∗A). q(cid:0)A∗ rdetp(AA∗)p .(cid:0)A∗ . p(cid:1)−1 q .(cid:1)−1 = = . p(cid:1) cdetq(A∗A). q(cid:0)A∗ . p(cid:1)) n(cdetq(A∗A). q(cid:0)A∗ q .(cid:1) rdetp(AA∗)p .(cid:0)A∗ n(rdetp(AA∗)p .(cid:0)A∗ q .(cid:1)) , . (7.7) (7.8) Substituting (7.7) into (7.5), and (7.8) into (7.6), we accordingly obtain (7.2) and (7.4). We proved the theorem. (cid:3) Equation (7.2) gives an explicit representation of quasideterminant A p,q of the matrix A ∈ M(n, H) for all p, q = 1, ..., n through the column determinant of its corresponding left Hermitian matrix A∗A, and (7.4) does through the row determinant of its corresponding right Hermitian matrix AA∗. Example 7.2. Consider a matrix A = a11 a12 a21 a22! According to (6.13) (7.9) a21 − a22(a12)−1 a11 a22 − a21(a11)−1 a12! A = a11 − a12(a22)−1 a21 a12 − a11(a21)−1 a22 Our goal is to find this quasideterminant, using the theorem 7.1. It is evident that A∗ = a11 a21 a12a11 + a22a21 n(a12) + n(a22)! . a12 a22! A∗A = n(a11) + n(a21) a11a12 + a21a22 Calculate the necessary determinants ddetA = rdet1(A∗A) = (n(a11) + n(a21)) · (n(a12) + n(a22)) −(a11a12 + a21a22) · (a12a11 + a22a21) = n(a11)n(a12) + n(a11)n(a22) + n(a21)n(a12) + n(a21)n(a22) −a11a12a12a11 − a11a12a22a21 − a21a22a12a11 − a21a22a22a21 = n(a11)n(a22) + n(a21)n(a12) − (a11a12a22a21 + a11a12a22a21) = n(a11)n(a22) + n(a21)n(a12) − t(a11a12a22a21) cdet1(A∗A).1(a∗ .2) = cdet1 a21 a11a12 + a21a22 a22 n(a12) + n(a22)! = n(a12)a21 + n(a22)a21 − a11a12a22 − a21a22a22 = n(a12)a21 − a11a12a22. Then cdet1(A∗A).1(a∗ .2) = n(a12)a21 − a22a12a11, 18 Aleks Kleyn and Ivan Kyrchei Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra n(cdet1(A∗A).1(a∗ .2)) = cdet1(A∗A).1(a∗ .2) · cdet1(A∗A).1(a∗ .2) = (n(a12)a21 − a22a12a11) · (n(a12)a21 − a11a12a22) = n2(a12)n(a21) − n(a12)a21a11a12a22 −n(a12)a22a12a11a21 + a22a12a11a11a12a22 = n(a12)(n(a12)n(a21) − t(a11a12a22a21) + n(a21)n(a12)) = n(a12)ddetA. Following the formula (7.2), we obtain ddetA A21 = cdet1(A∗A).1(a∗ .2) n(cdet1(A∗A).1(a∗ .2)) ddetA n(a12)ddetA cdet1(A∗A).1(a∗ .2) (7.10) = = = 1 n(a12) 1 n(a12) · cdet1(A∗A).1(a∗ .2) · (n(a12)a21 − a22a12a11) = a21 − a22(a12)−1a11. The last expression in (7.10) coincides with the expression A21 in (7.9). (cid:3) 8. References [1] H. Aslaksen. Quaternionic determinants Math. Intelligencer 18(3), pp.57- 65, (1996). [2] L. Chen, Definition of determinant and Cramer solutions over quaternion field, Acta Math. Sinica (N.S.) 7, pp.171-180, (1991). [3] L. Chen, Inverse matrix and properties of double determinant over quater- nion field, Sci. China, Ser. A 34, pp.528-540, (1991). [4] N. Cohen, S. De Leo, The quaternionic determinant, The Electronic Journal Linear Algebra 7, pp.100-111, (2000). [5] F. J. Dyson, Quaternion determinants, Helvetica Phys. Acta 45, pp. 289- 302, (1972). [6] Serge Lang, Algebra, Springer, 2002 [7] I. Gelfand, S. Gelfand, V. Retakh, R. Wilson, Quasideterminants, eprint arXiv:math.QA/0208146 (2002) [8] I.Gelfand, V.Retakh, Quasideterminants, I, eprint arXiv:q-alg/9705026 (1997) [9] I. Gelfand and V. Retakh, Determinants of Matrices over Noncommutative Rings, Funct. Anal. Appl. 25 (1991), no. 2, 91-102 [10] I. Gelfand and V. Retakh, A Theory of Noncommutative Determinants and Characteristic Functions of Graphs, Funct. Anal. Appl. 26 (1992), no. 4, 1-20 [11] J. Fan, Determinants and multiplicative functionals on quaternion matrices, Linear Algebra and Its Applications 369, pp. 193-201, (2003). [12] C.R.Gilson, J.J.C.Nimmo, Y.Ohta, Quasideterminant solutions of a non- Abelian Hirota-Miwa equation, Journal of Physics A: Mathematical and Theoretical 40(42), pp. 12607-12617,(2007). 19 Correspondence between Row-Column Determinants and Quasideterminants of Matrices over Quaternion Algebra Aleks Kleyn and Ivan Kyrchei [13] B. Haider, M. Hassan, Quasideterminant solutions of an integrable chiral model in two dimensions, Journal of Physics A: Mathematical and Theo- retical 42 (35), art. no. 355211, (2009). [14] I.I. Kyrchei, Cramer's rule for quaternion systems of linear equations, Jour- nal of Mathematical Sciences 155(6), 839-858, (2008). Translated from Fundamental and Appl. Math. 13(4), pp.67-94, (2007). (in Russian) eprint arXiv:math.RA/0702447 (2007) [15] I.I. Kyrchei, Cramer's rule for some quaternion matrix equations, Applied Mathematics and Computation 217(5), pp.2024-2030, (2010). eprint arXiv:math.RA/arXiv:1004.4380 (2010) [16] I.I. Kyrchei,Determinantal representations of the Moore-Penrose inverse over the quaternion skew field and corresponding Cramer's rules, eprint arXiv:math.RA/1005.0736 (2010) [17] Aleks Kleyn, Lectures on Linear Algebra over Division Ring, eprint arXiv:math.GM/0701238 (2010) [18] A. Lauve, Quantum- and quasi-Plucker coordinates, Journal of Algebra 296(2), pp.440-461, (2006). [19] Lewis D. W. Quaternion algebras and the algebraic legacy of Hamilton's quaternions, Irish Math. Soc. Bulletin 57, pp. 41-64, (2006). [20] C.X.Li, J.J.C. Nimmo, Darboux transformations for a twisted derivation and quasideterminant solutions to the super KdV equation, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences 466 (2120), pp. 2471-2493, (2010) [21] T. Suzuki, Noncommutative spectral decomposition with quasideterminant, Advances in Mathematics 217(5), pp. 2141-2158, (2008) 20 9. Index column determinant 7 conjugate of quaternion x 6 double determinant 12 Hadamard inverse of matrix 16 hermitian matrix 4 involution in quaternion algebra 6 (ji)-quasideterminant 16 left cofactor of entry of matrix 8 left double cofactor of entry of matrix 13 left-ordered cycle notation of permutation 7 norm in quaternion algebra 5 permutability property of trace 6 quasideterminant 16 quaternion algebra 5 rank of Hermitian matrix by principal minors 11 right cofactor of entry of matrix 8 right double cofactor of entry of matrix 13 right-ordered cycle notation of permutation 7 row determinant 7 trace of quaternion 6 21 10. Special Symbols and Notations A. T minor 2 AS . minor 2 A. T minor 3 AS . minor 2 Aji (ji)-quasideterminant of matrix A 16 A.j (b) transformation of matrix 3 Ai. (b) transformation of matrix 3 A quasideterminant of matrix A 16 cdetj A jth column determinant of matrix A 7 ddetA double determinant of matrix A 12 (cid:18) a, b F (cid:19) quaternion algebra over field F 5 HA Hadamard inverse of matrix 16 H(a, b) quaternion algebra 5 Li j Lij left ijth cofactor of entry of matrix 8 left double ijth cofactor of entry of matrix 13 n(x) norm of quaternion x 5 Ri j right ijth cofactor of entry of matrix 8 Rij right double ijth cofactor of entry of matrix 13 rdetiA ith row determinant of matrix A 7 Sn symmetric group 7 t(x) trace of quaternion x 5 22 1 1 0 2 b e F 9 ] . A R h t a m [ 1 v 6 7 7 1 . 2 0 1 1 : v i X r a Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Аннотация. В работе рассматриваются элементы теорий квазидетерминан- тов и строчно-столбцовых определителей и их применение к решению си- стем линейных уравнений в кватернионной алгебре. Установлено соответ- ствие между строчно-столбцовыми определителями и квазидетерминанта- ми матриц над кватернионной алгеброй. Содержание 1. Предисловие . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Соглашение об обозначениях . . . . . . . . . . . . . . . . . . . . . . . 3. Предварительные замечания . . . . . . . . . . . . . . . . . . . . . . . 4. Кватернионная алгебра . . . . . . . . . . . . . . . . . . . . . . . . . . 5. Введение в теорию строчных и столбцовых определителей . . . . . 6. Квазидетерминант . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. Соответствие между строчно-столбцовыми определителями и квази- детерминантами . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8. Список литературы . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. Предметный указатель . . . . . . . . . . . . . . . . . . . . . . . . . . 10. Специальные символы и обозначения . . . . . . . . . . . . . . . . . 1 2 3 6 7 16 19 21 23 24 1. Предисловие Линейная алгебра является мощным инструментом, которым мы пользуемся в различных областях математики, включая математический анализ, аналити- ческую и дифференциальную геометрии, теорию дифференциальных уравне- ний, теорию оптимального управления. Линейная алгебра накопила богатый набор различных методов. Так как некоторые методы имеют общий конечный результат, это даёт нам возможность выбрать наиболее эффективный метод в зависимости от характера расчётов или выполняемых построений. 2010 Mathematics Subject Classification. Primary 15A33, 15A15, 15A24; УДК 512.643.2. Key words and phrases. кватернионная алгебра, некоммутативный определитель, квази- детерминант, система линейных уравнений, правило Крамера. [email protected]. [email protected]. Институт Прикладных Проблем Механики и Математики им. Я.С.Пидстрыгача НАН Украины, ул. Наукова 3а, Львов, 79053, Украина. 1 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей При переходе от линейной алгебры над полем к линейной алгебре над телом мы хотим сохранить как можно больше инструментария, которым мы регу- лярно пользуемся. Уже в начале XX века, вскоре после после создания Га- мильтоном алгебры кватернионов, математики стали искать ответ, как выгля- дит алгебра с некоммутативным умножением. В частности возникла проблема как определить детерминант матрицы с элементами, принадлежащими неком- мутативному кольцу. Такой детерминант также называют некоммутативным детерминантом. Возникло немало подходов к определению некоммутативного детерминанта. Но ни один из введённых некоммутативных детерминантов в полной мере не сохранял те свойства, которыми он обладал для матриц над полем. Более того, в работе [11] доказано, что не существует единого детерминантного функцио- нала, который бы расширял определение детерминанта вещественных матриц для матриц над телом кватернионов. Поэтому поиски решения проблемы опре- деления некоммутативного детерминанта продолжаются до сих пор. В данной работе мы рассматриваем два подхода к определению некоммута- тивного детерминанта: строчно-столбцовые определители и квазидетерминан- ты. Строчно-столбцовые определители являются расширением классического опре- деления детерминанта, но с заранее определенным порядком элементов в каж- дом из слагаемых детерминанта. При использовании строчно-столбцовых опре- делителей, мы получаем решение системы линейных уравнений над кватерни- онной алгеброй согласно правилу Крамера. Квазидетерминант появился в результате анализа процесса обращения мат- рицы. При использовании квазидетерминанта, решение системы линейных урав- нений над кватернионной алгеброй аналогично методу Гаусса. Общим в определении строчно-столбцовых определителей и квазидетерми- нантов является, что в обоих случаях квадратной матрице n-го порядка с некоммутативными элементами ставится в соответствие не один определитель, а некоторое их множество (n2 квазидетерминантов и соответственно n строч- ных и n столбцовых определителей). Квазидетерминанты нашли в последнее время широкое применение, как в области линейной алгебры ([18, 21]), так и физики ([12, 13, 20]). Сравнитель- но недавно введенные строчно-столбцовые определители ([14]) менее извест- ны. Целью данной статьи является установление соответствия между строчно- столбцовыми определителями и квазидетерминантами матриц над кватерни- онной алгеброй. Авторы полны надежды что установленное соответствие мо- жет дать обоюдное развитие как теории квазидетерминантов, так и строчно- столбцовых определителей. 2. Соглашение об обозначениях Существуют различные формы записи элементов матрицы. В данной статье элемент матрицы A мы обозначать aij . При этом индекс i нумерует строки, а индекс j нумерует столбцы. Для обозначения различных миноров матрицы A мы будем пользоваться следующей записью ai . : строка с индексом i 2 Александр Клейн and Иван Кирчей Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. AS . S Ai . AS . : минор, полученный из A выбором строк с индексом из множества : минор, полученный из A удалением строки ai . : минор, полученный из A удалением строк с индексом из множе- ства S a. j A. T : столбец с индексом j : минор, полученный из A выбором столбцов с индексом из мно- жества T A. j A. T : минор, полученный из A удалением столбца a. j : минор, полученный из A удалением столбцов с индексом из мно- жества T A.j (b) : матрица, которую получим из матрицы A заменой ее j-го столбца столбцом b Ai. (b) : матрица, которую получим из матрицы A заменой ее i-й стро- ки строкой b Рассмотренные обозначения могут быть скомбинированы. Например, запись Aii k.(b) означает замену строки с номером k матрицей b с последующим удалением строки с номером i и столбца с номером i. Как отмечалось в разделе [17]-2.2 мы можем определить два вида произве- дения матриц: либо произведение строк первой матрицы на столбцы второй, либо произведение столбцов первой матрицы на строки второй. Однако со- гласно теореме [17]-2.2.5 это произведение симметрично относительно операции транспонирования. Поэтому в статье мы ограничимся традиционным произве- дением строк первой матрицы на столбцы второй, и мы не будем указывать явно операцию как это было сделано в [17]. 3. Предварительные замечания Теорию определителей матриц с некоммутативными элементами, можно услов- но разделить на три группы относительно методов их определения. Обозначим через M(n, K) кольцо матриц с элементами из кольца K. Один из способов определения детерминанта матрицы из M (n, K) (cid:22) следующий ([1, 4, 5]). Определение 3.1. Пусть функционал d : M (n, K) → K удовлетворяет следующим аксиомам. Аксиома 1. d (A) = 0 тогда и только тогда, когда A -- вырожденная (необратимая). Аксиома 2. ∀A, B ∈ M (n, K), d (A · B) = d (A) · d (B). Аксиома 3. Если матрица A′ получается из матрицы A прибавлением ее произвольной строки умноженной слева с ее другой строкой или ее произвольного столбца умноженного справа с ее другим столбцом, тогда d (A′) = d (A) Тогда значение функционала d называется определителем матрицы A ∈ M (n, K). (cid:3) 3 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Примерами таких функционалов являются известные определители Дье- донне и Стади. В [1] доказано, что определители, которые удовлетворяют ак- сиомы 1, 2, 3, принимают свое значение в некотором коммутативном подмно- жестве кольца. Также для них не имеет смысла такое свойство обычных опре- делителей как разложение по минорам строки или столбца. Поэтому детер- минантное представление обратной матрицы с помощью только таких опреде- лителей невозможно. Всё это является причиной, которая заставляет вводить детерминантные функционалы, не удовлетворяющие всем вышеприведенным аксиомам. При этом аксиома 1 рассматривается в [5] как необходимая для определения детерминанта. При другом подходе определитель квадратной матрицы над некоммутатив- ным кольцом строится как рациональная функция от элементов матрицы. Наи- большего успеха здесь достигли И. М. Гельфанд и В. С. Ретах своей теорией квазидетерминантов ([9, 10]). Введение в теорию квазидетерминантов будет изложено в разделе 6. При третьем подходе детерминант матрицы с некоммутативными элемен- тами определяется, как альтернированная сумма n! произведений элементов матрицы, но с определенным фиксированным порядком множителей в каж- дом из них. Е. Г. Мур был первый, кто достиг выполнения ключевой аксиомы 1 при таком определении некоммутативного детерминанта. Мур сделал это не для всех квадратных матриц, а только для эрмитовых. Он определил детер- минант эрмитовой матрицы1 A = (aij)n×n над телом с инволюцией индукцией по n следующим образом (см. [5]) (3.1) Здесь εkj =( 1, −1, MdetA =  i = j i 6= j a11, n n = 1 εijaij Mdet (A(i → j)) , n > 1 Pj=1 , а через A(i → j) обозначена матрица, которая по- лучается из матрицы A последовательным применением замены j-го столбца i-м и вычеркивания i-ых строки и столбца. Другое определение этого детер- минанта представлено в [1] в терминах подстановок: Mdet A = Xσ∈Sn σan11n12 · . . . · an1l1 n11 ·an21n22 · . . . · anrl1 nr1 . Здесь Sn -- симметрическая группа n элементов. Разложение подстановки σ в циклы имеет вид: σ = (n11 . . . n1l1 ) (n21 . . . n2l2) . . . (nr1 . . . nrlr ) . Однако, не существовало никакого обобщения определения детерминанта Мура для произвольных квадратных матриц. Ф. Д. Дайсон отмечал важность этой задачи в работе [5]. 1Эрмитова матрица - это такая матрица A = (aij ), что aij = aji. 4 Александр Клейн and Иван Кирчей Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. В статьях [2, 3] Чен Лонгхуан предложил следующее определение детерми- нанта квадратной матрицы над телом кватернионов A = (aij ) ∈ M (n, H): det A = Pσ∈Sn ε (σ) an1i2 · ai2i3 . . . · aisn1 · . . . · anr k2 · . . . · aklnr , σ = (n1i2 . . . is) . . . (nrk2 . . . kr) , n1 > i2, i3, . . . , is; . . . , nr > k2, k3, . . . , kl, n = n1 > n2 > . . . > nr ≥ 1. Несмотря на то, что такой определитель не удовлетворяет аксиоме 1, Л. Чен получил детерминантное представление обратной матрицы. Но его определи- тель также нельзя разложить по минорам строки или столбца (за исключе- нием n-й строки). Поэтому Л. Чен также не получил классическую присо- единенную матрицу. Для A = (α1, . . . , αm) над телом кватернионов H, если kAk := det(A∗A) 6= 0, тогда ∃A−1 = (bjk), где bjk = 1 kAk ωkj , (j, k = 1, 2, . . . , n) , ωkj = det (α1 . . . αj−1αnαj+1 . . . αn−1δk)∗ (α1 . . . αj−1αnαj+1 . . . αn−1αj ) . Здесь αi -- i-й столбец A, δk -- n-мерный столбец с 1 в k-й строке и 0 в других. Л. Чен определил kAk := det(A∗A), как двойной детерминант. Решение правой системы линейных уравнений над H, если kAk 6= 0, представлено следующей формулой, которую автор на- зывает крамеровской αjxj = β Xn j=1 где xj = kAk−1Dj , ∀j = 1, n Dj = det (cid:16) α1 . . . αj−1 αn αj+1 . . . αn−1 αj (cid:17) . α∗ 1 ... α∗ α∗ n α∗ ... α∗ β ∗   j−1 j+1 n−1   Здесь αi -- i-й столбец матрицы A, α∗ вектор-строка сопряженная с β. i (cid:22) i-я строка матрицы A∗, β ∗ (cid:22) n-мерная В данной работе мы рассматриваем теорию строчных и столбцовых опреде- лителей, которая развивает классический подход к определению детерминанта матрицы, но с заранее определенным порядком множителей в каждом из сла- гаемых детерминанта. 5 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей 4. Кватернионная алгебра Кватернионная алгебра H(a, b) (она также обозначается как (cid:18) a, b - это четырехмерное векторное пространство над полем F с базисом {1, i, j, k} и следующими правилами умножения: F (cid:19) ) i2 = a, j2 = b, ij = k, ji = −k. Поле F является центром кватернионной алгебры H(a, b). В алгебре H(a, b) определены следующие отображения. • Квадратичная форма n : x ∈ H → n(x) ∈ F такая что n(x · y) = n(x)n(y) x, y ∈ H называется нормой в кватернионной алгебре H. • Линейное отображение t : x = x0 + x1i + x2j + x3k ∈ H → t(x) = 2x0 ∈ F называется следом кватерниона. След удовлетворяет условию пе- рестановочности следа Из теоремы [17]-10.3.3 следует t (q · p) = t (p · q) (4.1) (4.2) t(x) = 1 2 (x − ixi − jxj − kxk) • Линейное отображение x → x = t(x) − x является инволюцией. Инволюция имеет следующие свойства x = x x + y = x + y x · y = y · x При этом кватернион x будем называть сопряженным к кватерни- ону x ∈ H. Норма и инволюция удовлетворяют следующему условию След и инволюция удовлетворяют следующему условию n (q) = n(q) Из равенств (4.1), (4.2) следует t (x) = t(x) x = − 1 2 (x + ixi + jxj + kxk) 6 Александр Клейн and Иван Кирчей Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. В зависимости от выбора поля F и элементов a, b на множестве всех кватер- нионных алгебр возможны два случая [19]: 1.(cid:18) a, b 2. (cid:18) a, b F (cid:19) является алгеброй с делением. F (cid:19) изоморфна алгебре всех 2 × 2 матриц над полем F. В этом случае кватернионная алгебра - расщепляемая. Рассмотрим некоторые не изоморфные кватернионные алгебры с делением. 1. (cid:18) a, b телу кватернионов H, когда a < 0 и b < 0. В противном случае, (cid:18) a, b морфных кватернионных алгебр с делением (cid:18) a, b R (cid:19), где R - поле действительных чисел, изоморфна классическому R (cid:19) - Q (cid:19) в зависимости от выбора 2. Над полем рациональных чисел Q существует бесконечно много не изо- расщепляемая. a < 0 и b < 0. 3. Пусть Qp - поле p-адичних чисел, где p - простое число. Для каждого простого числа p существует единственная кватернионная алгебра с делением. 5. Введение в теорию строчных и столбцовых определителей Строчные и столбцовые определители вводятся для матриц над кватерни- онной алгеброй H. Следующие определения в теории подстановок необходимы нам для введения строчно-столбцовых определителей. Определение 5.1. Пусть Sn -- симметрическая группа на множестве In = {1, 2, ..., n} ([6], с. 70, упражнения 9, 10). Будем говорить, что подстановка σ ∈ Sn образует прямое произведение независимых циклов, если ее запись в обычной двухрядной форме соответствует ее разложению в независимые цик- лы (5.1) σ = n11 n12 n12 n13 . . . n1l1 . . . n11 . . . nr1 nr2 . . . nr2 nr3 . . . nrlr . . . nr1! (cid:3) Определение 5.2. Будем говорить, что представление (5.1) подстановки σ произведением независимых циклов является упорядоченным слева, если элементы, которые замыкают каждый из ее независимых циклов, записывают- ся первыми слева в каждом из циклов σ = (n11n12 . . . n1l1 ) (n21n22 . . . n2l2 ) . . . (nr1nr2 . . . nrlr ) . (cid:3) Определение 5.3. Будем говорить, что представление (5.1) подстановки σ произведением независимых циклов является упорядоченным справа, если элементы, которые замыкают каждый из ее независимых циклов, записывают- ся первыми справа в каждом из циклов σ = (n12 . . . n1l1 n11) (n22 . . . n2l2 n21) . . . (nr2 . . . nrlr nr1) . (cid:3) 7 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Определение 5.4. Пусть Sn -- симметрическая группа на множестве In = {1, 2, ..., n}. Пусть i ∈ 1, n. Строчным определителем по i-й строке матрицы A = (aij ) ∈ M (n, H) будем называть выражение rdetiA = Xσ∈Sn (−1)n−rai ik1 aik1 ik1 +1 . . .aik1 +l1 i . . . aikr ikr +1 . . . aikr +lr ikr Это выражение является альтернированной суммой n! мономов, составленных из элементов матрицы A, в каждом из которых подстановка σ ∈ Sn индексов образует прямое произведение независимых циклов σ = (i ik1 ik1+1 . . . ik1+l1 ) (ik2 ik2+1 . . . ik2+l2) . . . (ikr ikr +1 . . . ikr +lr ) В упорядоченном слева разложении подстановки σ первый слева цикл начина- ется слева индексом i, а все следующие циклы удовлетворяют условиям: ik2 < ik3 < . . . < ikr , ikt < ikt+s, (cid:0)∀t = 2, r(cid:1) , (cid:0)∀s = 1, lt(cid:1) . (cid:3) Определение 5.5. Пусть Sn -- симметрическая группа на множестве Jn = {1, 2, ..., n}. Пусть j ∈ 1, n. Столбцовым определителем по j-му столбцу матрицы A = (aij) ∈ M (n, H) будем называть выражение (−1)n−r ajkr jkr +lr . . . ajkr +1ikr . . .aj jk1 +l1 . . . ajk1 +1jk1 ajk1 j. cdetj A = Xτ ∈Sn Это выражение является альтернированной суммой n! мономов, составленных из элементов матрицы A, в каждом из которых подстановка τ ∈ Sn индексов образует прямое произведение независимых циклов τ = (jkr +lr . . . jkr +1jkr ) . . . (jk2+l2 . . . jk2+1jk2 ) (jk1+l1 . . . jk1+1jk1 j) В упорядоченном справа разложении подстановки τ первый справа цикл начи- нается справа индексом j, а все следующие циклы удовлетворяют условиям: jk2 < jk3 < . . . < jkr , jkt < jkt+s, (cid:0)∀t = 2, r(cid:1) , (cid:0)∀s = 1, lt(cid:1) . (cid:3) Замечание 5.6. Особенностью столбцовых определителей является то, что при непосредственном их вычислении множители каждого из мономов записыва- ются справа налево. (cid:3) В леммах 5.7 и 5.8 рассмотрено рекуррентное определение столбцовых и строчных определителей. Это определение является аналогом разложения опре- делителя по строкам и столбцам в коммутативном случае. Лемма 5.7. Пусть Ri j -- правое алгебраическое дополнение элемента ai j матрицы A = (aij) ∈ M (n, H), а именно rdeti A = n Pj=1 ai j · Ri j i = 1, n 8 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Тогда .j(a. i)) rdetk Aii Ri j =( −rdetj (Aii k =(2 i = 1 i > 1 1 i 6= j i = j где матрица (Aii ны j-го столбца i-м и вычеркивания i-х строки и столбца. .j(a. i)) получается из A последовательным применением заме- (cid:3) Лемма 5.8. Пусть Li j -- левое алгебраическое дополнение элемента ai j матрицы A = (aij) ∈ M (n, H), а именно Тогда cdetj A = n Pi=1 Li j · ai j j = 1, n i. (aj .)) cdetk Ajj Li j =( −cdeti (Ajj k =(2 j = 1 j > 1 1 i 6= j i = j где матрица (Ajj ны i-й строки j-й и вычеркивания j-х строки и столбца. i. (aj .)) получается из A последовательным применением заме- (cid:3) Замечание 5.9. Очевидно, что каждому моному любого определенного выше детерминанта квадратной матрицы отвечает моном любого другого детерми- нанта, строчного или столбцового, такой, что оба они имеют одинаковый знак, состоят из одних и тех же множителей, - элементов матрицы, и отличаются только порядком их размещения. Кроме того, если элементы матрицы комму- тируют, то rdet1 A = . . . = rdetnA = cdet1 A = . . . = cdetnA. (cid:3) Рассмотрим основные свойства строчного и столбцового детерминанта про- извольной квадратной матрицы над телом H, доказательства которых непо- средственно следуют из определений. Теорема 5.10. Если одна из строк (столбцов) матрицы A ∈ M (n, H) состо- ит из нолей, тогда для всех i = 1, n rdeti A = 0, cdeti A = 0. Теорема 5.11. Если i-я строка матрицы A = (aij ) ∈ M (n, H) умножается слева на любое b ∈ H, тогда для всех i = 1, n rdeti Ai .(b · a. i) = b · rdeti A. Теорема 5.12. Если j-й столбец матрицы A = (aij) ∈ M (n, H) умножается справа на любое b ∈ H, тогда для всех j = 1, n cdetj A. j (aj . · b) = cdetj A · b. 9 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Теорема 5.13. Если для A = (aij ) ∈ M (n, H) найдется такое k ∈ In, что akj = bj + cj для всех j = 1, n, тогда для любого i = 1, n rdeti A = rdeti Ak . (b) + rdeti Ak . (c) , cdeti A = cdeti Ak . (b) + cdeti Ak . (c) . где b = (b1, . . . , bn), c = (c1, . . . , cn). Теорема 5.14. Если для A = (aij ) ∈ M (n, H) найдется такое k ∈ In, что ai k = bi + ci для всех i = 1, n, тогда для любого j = 1, n rdetj A = rdetj A. k (b) + rdetj A. k (c) , cdetj A = cdetj A. k (b) + cdetj A. k (c) . где b = (b1, . . . , bn)T , c = (c1, . . . , cn)T . Теорема 5.15. Пусть A∗ -- матрица, эрмитово сопряженная к A ∈ M (n, H), тогда rdeti A∗ = cdeti A для любого i = 1, n. (cid:3) Следующая теорема имеет ключевое значение в теории строчных и столб- цовых определителей. Теорема 5.16. Если A = (aij) ∈ M (n, H) -- эрмитовая матрица, тогда rdet1A = . . . = rdetnA = cdet1A = . . . = cdetnA ∈ F. (cid:3) Замечание 5.17. Поскольку все столбцовые и строчные определители эрмито- вой матрицы над телом H равны между собой, то мы можем однозначно ввести понятие определителя эрмитовой матрицы A ∈ M (n, H): det A := rdeti A = cdeti A, (cid:0)∀i = 1, n(cid:1) . (cid:3) Замечание 5.18. Представляя детерминант эрмитовой матрицы как строчный определитель по произвольной i-й строке по лемме 5.7 имеем ai j · rdetj (Aii .j(a. i)) + ai i · rdetk Aii (5.2) k =(2 1 det A = − Pσ∈Sn i = 1 i > 1 Сравнивая выражения (3.1) и (5.2) для эрмитовой матрицы A ∈ M (n, H), с очевидностью получаем, что введенный строчный определитель для эрмито- вой матрицы совпадает с детерминантом Мура, а строчные и столбцовые опре- делители для произвольных матриц являются его обобщением на множестве произвольных квадратных матриц. (cid:3) Свойства определителя эрмитовой матрицы полностью рассматриваются в [14] посредством ее строчных и столбцовых определителей. Среди всех рас- смотрим следующие. Теорема 5.19. Если i-ю строку эрмитовой матрицы A ∈ M (n, H) заменить левой линейной комбинацией других ее строк ai . = c1ai1 . + . . . + ckaik . 10 Александр Клейн and Иван Кирчей Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. где cl ∈ H для всех l = 1, k и {i, il} ⊂ In, тогда для всех i = 1, n cdetiAi . (c1 · ai1 . + . . . + ck · aik . ) = rdetiAi . (c1 · ai1 . + . . . + ck · aik . ) = 0. (cid:3) Теорема 5.20. Если j-й столбец эрмитовой матрицы A ∈ M (n, H) заменить правой линейной комбинацией других ее столбцов a. j = a. j1 c1 + . . . + a. jk ck где cl ∈ H для всех l = 1, k и {j, jl} ⊂ Jn , тогда для всех j = 1, n cdetj A. j (a. j1 · c1 + . . . + a. jk · ck) = rdetj A. j (a. j1 · c1 + . . . + a. jk · ck) = 0. (cid:3) Следующая теорема о детерминантном представлении матрицы обратной к эрмитовой непосредственно следует из этих свойств. Теорема 5.21. Для эрмитовой невырожденной матрицы A ∈ M (n, H), ( det A 6= 0), существует единственная правая (RA)−1 и единственная ле- вая обратная матрица (LA)−1, при этом (RA)−1 = (LA)−1 =: A−1, которые обладают следующими детерминантными представлениями: R11 R21 R12 R22 · · · · · · R1n R2n L11 L21 L12 L22 · · · · · · L1n L2n     · · · Rn1 · · · Rn2 · · · · · · · · · Rnn · · · Ln1 · · · Ln2 · · · · · · · · · Lnn     . (RA)−1 = 1 det A (LA)−1 = 1 det A Здесь Rij, Lij являются соответственно левым и правым алгебраическими дополнениями для всех i, j = 1, n. (cid:3) Для того, чтобы получить детерминантное представление для произвольной обратной матрицы над телом H рассматриваются ее правая AA∗ и левая A∗A соответственные эрмитовые матрицы. Теорема 5.22 ([14]). Если произвольный столбец матрицы A ∈ Hm×n яв- ляется правой линейной комбинацией ее других столбцов, или произвольная строка матрицы A∗ является левой линейной комбинацией ее других строк, тогда det A∗A = 0. (cid:3) Поскольку главная подматрица эрмитовой матрицы эрмитова, то главный базисный минор может быть определен по аналогии с коммутативным случаем, как ненулевой определитель ее главной подматрицы максимального порядка. Также вводим понятие ранга эрмитовой матрицы по главным минорам, как максимальный порядок отличного от нуля главного минора. Следующая теорема устанавливает соответствие между ним и рангом матрицы определен- ным как максимальное количество линейно независимых столбцов справа или линейно независимых строк слева, которые и образуют базис. 11 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Теорема 5.23 ([14]). Ранг по главным минорам эрмитовой матрицы A∗A равняется ее рангу, а также рангу матрицы A ∈ Hm×n. (cid:3) Теорема 5.24 ([14]). Если A ∈ Hm×n, тогда произвольный столбец мат- рицы A является правой линейной комбинацией ее базисных столбцов или произвольная строка матрицы A является левой линейной комбинацией ее базисных строк. (cid:3) Отсюда следует критерий вырожденности соответственной эрмитовой мат- рицы. Теорема 5.25 ([14]). Правая линейная независимость столбцов матрицы A ∈ Hm×n или левая линейная независимость строк матрицы A∗ является необходимым и достаточным условием того, что det A∗A 6= 0 Теорема 5.26 ([14]). Если A ∈ M (n, H), тогда det AA∗ = det A∗A. (cid:3) Пример 5.27. Рассмотрим матрицу a21 a22! A = a11 a12 a12 a22! A∗ = a11 a21 Тогда Соответственно a21a11 + a22a12 a21a21 + a22a22! AA∗ = a11a11 + a12a12 a11a21 + a12a22 A∗A = a11a11 + a21a21 a11a12 + a21a22 a12a11 + a22a21 a12a12 + a22a22! Чтобы оценить детерминанты полученных эрмитовых матриц, рассмотрим на- пример первый строчный определитель каждой матрицы. Согласно теореме 5.16 и замечанию 5.17 det AA∗ = rdet1AA∗ det A∗A = rdet1A∗A 12 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Согласно лемме 5.7 (5.3) (5.4) det AA∗ = (AA∗)11(AA∗)22 − (AA∗)12(AA∗)21 = (a11a11 + a12a12)(a21a21 + a22a22) −(a11a21 + a12a22)(a21a11 + a22a12) = a11a11a21a21 + a12a12a21a21 +a11a11a22a22 + a12a12a22a22 −a11a21a21a11 − a12a22a21a11 −a11a21a22a12 − a12a22a22a12 = a12a12a21a21 + a11a11a22a22 −a12a22a21a11 − a11a21a22a12 det A∗A = (A∗A)11(A∗A)22 − (A∗A)12(A∗A)21 = (a11a11 + a21a21)(a12a12 + a22a22) −(a11a12 + a21a22)(a12a11 + a22a21) = a11a11a12a12 + a21a21a12a12 +a11a11a22a22 + a21a21a22a22 −a11a12a12a11 − a21a22a12a11 −a11a12a22a21 − a21a22a22a21 = a21a21a12a12 + a11a11a22a22 −a21a22a12a11 − a11a12a22a21 Положительные слагаемые в равенствах (5.3), (5.4) являются действительными числами и, очевидно, совпадают. Чтобы доказать равенство (5.5) a12a22a21a11 + a11a21a22a12 = a21a22a12a11 + a11a12a22a21 используем свойство перестановочности следа элементов кватернионной алгеб- ры, t(pq) = t(qp). Действительно, a12a22a21a11 + a11a21a22a12 = a12a22a21a11 + a12a22a21a11 = t(a12a22a21a11), a21a22a12a11 + a11a12a22a21 = a11a12a22a21 + a11a12a22a21 = t(a11a12a22a21) Тогда из свойства перестановочности следа, следует (5.5). (cid:3) На основании теоремы 5.26 вводится понятие двойного определителя. Впер- вые это понятие было введено Л. Ченом ([2]). Определение 5.28. Определитель эрмитовой матрицы AA∗ называется двой- ным определителем матрицы A ∈ M (n, H) ddetA := det (A∗A) = det (AA∗) (cid:3) В случае если H - классическое тело кватернионов H над полем действитель- ных чисел, тогда имеет место теорема, которая устанавливает справедливость Аксиомы 1 для двойного определителя. Теорема 5.29. Если {A, B} ⊂ M (n, H), тогда ddet (A · B) = ddetA · ddetB. (cid:3) 13 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей В общем случае не существует взаимосвязей между вырожденностью двой- ного определителя матрицы и её строчными и столбцовыми определителями. Продемонстрируем это в следующем примере. Пример 5.30. Рассмотрим матрицу A = i j −i! j Ее вторая строка получена из первой умножением слева на k. Тогда согласно теореме 5.25 ddetA = 0. Действительно, A∗A = −i −j i ! · i j −i! = 2 −2k 2 ! . −j 2k j Тогда ddetA = 4 + 4k2 = 0, но cdet1A = cdet2A = rdet1A = rdet2A = −i2 − j2 = 2 B то же время rankA = 1, что соответствует теореме 5.23. (cid:3) Установлено соответствие между двойным определителем и некоммутатив- ными определителями Мура, Стади и Дьедонне, ddetA = Mdet (A∗A) = SdetA = Ddet2A Определение 5.31. Пусть ddetA = cdetj (A∗A) =Pi Lij · aij ∀j = 1, n тогда будем называть Lij левым двойным алгебраическим дополнением элемента aij матрицы A ∈ M (n, H). (cid:3) Определение 5.32. Пусть ddetA = rdeti (AA∗) =Pj aij · Rij ∀i = 1, n тогда будем называть Rij правым двойным алгебраическим дополне- нием элемента aij матрицы A ∈ M (n, H). (cid:3) Теорема 5.33. Необходимым и достаточным условием обратимости матри- цы A = (aij ) ∈ M(n, H) является ddetA 6= 0. Тогда ∃A−1 = (LA)−1 = (RA)−1, где (5.6) (LA)−1 = (A∗A)−1 A∗ = 1 ddetA (5.7) (RA)−1 = A∗ (AA∗)−1 = 1 ddetA∗     L11 L21 L12 L22 . . . . . . L1n L2n . . . Ln1 . . . Ln2 . . . . . . . . . Lnn R 11 R 21 R 12 R 22 . . . . . . . . . R n1 . . . R n2 . . . . . . R 1n R 2n . . . R nn     и Lij = cdetj(A∗A). j (A∗ . i), R ij = rdeti(AA∗)i.(cid:0)A∗ j .(cid:1) ,(cid:0)∀i, j = 1, n(cid:1). (cid:3) 14 Александр Клейн and Иван Кирчей Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Следствие 5.34. Пусть ddetA 6= 0. Тогда Lij = Rij Замечание 5.35. Теорема 5.33 представляет обратную матрицу A−1 произволь- ной квадратной матрицы A ∈ M(n, H), если ddetA 6= 0, через аналог классиче- ской присоединенной. Если мы обозначим ее Adj[[A]], тогда над телом H имеет место следующая формула: (cid:3) A−1 = Adj[[A]] ddetA . (cid:3) Очевидным следствием детерминантного представления обратной матрицы через аналог классической присоединенной матрицы является правило Краме- ра. Теорема 5.36. Пусть (5.8) A · x = y правая система линейных уравнений с матрицей коэффициентов A ∈ M(n, H), столбцом свободных элементов y = (y1, . . . , yn)T ∈ Hn×1 и столбцом неиз- вестных x = (x1, . . . , xn)T . Если ddetA 6= 0, тогда система линейных уравне- ний (5.8) имеет единственное решение, которое представляется формулой: (5.9) где f = A∗y. xj = cdetj(A∗A).j (f ) ddetA ∀j = 1, n (cid:3) Теорема 5.37. Пусть (5.10) x · A = y -- левая система линейных уравнений с матрицей коэффициентов A ∈ M(n, H), строкой свободных элементов y = (y1, . . . , yn) ∈ H1×n и строкой неизвестных x = (x1, . . . , xn). Если ddetA 6= 0, тогда система линейных уравнений (5.10) имеет единственное решение, которое представляется формулой: (5.11) где z = yA∗. xi = rdeti (AA∗)i. (z) ddetA ∀i = 1, n (cid:3) Формулы (5.9) и (5.11) являются естественным и очевидным обобщением правила Крамера для квадратных систем линейных уравнений над кватерни- онной алгеброй с делением. Еще более близкую аналогию можно получить в следующих частных случаях, которые следуют из теоремы 5.21. Теорема 5.38. Пусть в правой системе линейных уравнений (5.8) матрица коэффициентов A ∈ M(n, H) -- эрмитова. Тогда система имеет единственное решение, которое представляется формулой: xj = cdetj A.j (y) det A , (cid:0)∀j = 1, n(cid:1) . 15 (cid:3) Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Теорема 5.39. Пусть в левой системе линейных уравнений (5.10) матрица коэффициентов A ∈ M(n, H) -- эрмитова. Тогда система имеет единственное решение, которое представляется формулой: xi = rdetiAi. (y) det A , (cid:0)∀i = 1, n(cid:1) . (cid:3) В рамках теории строчных-столбцовых определителей получено также пра- вило Крамера для правого AX = B, левого XA = B и двустороннего AXB = C матричных уравнений ([15]). Также получено детерминантное представление Мура-Пенроуза обратной матрицы над телом кватернионов и правило Крамера для нормального решения правой и левой систем линейных уравнений ([16]). Теорема 6.1. Предположим, что матрица2 6. Квазидетерминант a11 ... an1 ... a1n ... ... ... ann A =    имеет обратную матрицу A−1. Тогда минор обратной матрицы удовлетво- ряет следующему равенству, при условии, что рассматриваемые обратные матрицы существуют, (6.1) ((A−1)IJ )−1 = AJI − A.I J.(AJI )−1AJ. .I Доказательство. Определение обратной матрицы приводит к системе линей- ных уравнений (6.2) (6.3) AJI (A−1)I. J.(A−1)I. A.I .J + AJ. .I (A−1)IJ = 0 .J + AJI (A−1)IJ = E Мы умножим (6.2) на (cid:0)AJI(cid:1)−1 (A−1)I. (6.4) .J + (AJI )−1AJ. .I (A−1)IJ = 0 Теперь мы можем подставить (6.4) в (6.3) (6.5) AJI (A−1)IJ − A.I J.(AJI )−1AJ. .I (A−1)IJ = E (6.1) следует из (6.5). (cid:3) Следствие 6.2. Предположим, что матрица A имеет обратную матрицу. Тогда элементы обратной матрицы удовлетворяют равенству (6.6) ((A−1)ij )−1 = aji − A.i j.(Aji)−1Aj. .i 2Это утверждение и его доказательство основаны на утверждении 1.2.1 из [7] для матриц над свободным кольцом с делением. (cid:3) 16 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей Пример 6.3. Рассмотрим матрицу a21 a22! A = a11 a12 Согласно (6.6) (6.7) (6.8) (6.9) (6.10) (A−1)11 = (a11 − a12(a22)−1 a21)−1 (A−1)21 = (a21 − a22(a12)−1 a11)−1 (A−1)12 = (a12 − a11(a21)−1 a22)−1 (A−1)22 = (a22 − a21(a11)−1 a12)−1 Мы называем матрицу (6.11) HA = ((HA)ij) = ((aji)−1) обращением Адамара матрицы 3 (cid:3) a11 ... an1 ... a1n ... ... ... ann A =    Определение 6.4. (ji)-квазидетерминант матрицы A - это формальное выражение (6.12) Aji = (HA−1)ji = ((A−1)ij)−1 Мы можем рассматривать (ji)-квазидетерминант как элемент матрицы A , которую мы будем называть квазидетерминантом. (cid:3) Теорема 6.5. Выражение для (ji)-квазидетерминанта имеет форму (6.13) (6.14) Aji = aji − A.i Aji = aji − A.i j.(Aji)−1Aj. j. HAji Aj. .i .i Доказательство. Утверждение следует из (6.6) и (6.12). (cid:3) Теорема 6.6. Пусть (6.15) Тогда (6.16) A = 1 0 0 1! A−1 = 1 0 0 1! Доказательство. Из (6.7) и (6.10) очевидно, что (A−1)11 = 1 и (A−1)22 = 1. Тем не менее выражение для (A−1)21 и (A−1)12 не может быть определено 3[8]-стр. 4 17 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей из (6.8) и (6.9) так как (a21 − a22(a12)−1 a11)−1 = (a12 − a11(a21)−1 a22)−1 = 0. Мы можем преобразовать эти выражения. Например (A−1)21 = (a21 − a22(a12)−1 a11)−1 = (a11((a11)−1 a12 − (a21)−1 a22))−1 = ((a21)−1 a11(a21(a11)−1 a12 − a22))−1 = (a11(a21(a11)−1 a12 − a22))−1 a21 Мы непосредственно видим, что (A−1)21 = 0. Таким же образом мы можем найти, что (A−1)12 = 0. Это завершает доказательство (6.16). (cid:3) Из доказательства теоремы 6.6 видно, что мы не всегда можем пользоваться равенством (6.6) для определения элементов обратной матрицы и необходимы дополнительные преобразования для решения этой задачи. Из теоремы [17]- 4.6.3 следует, что если a11 ... an1 rank  ... a1n ... ... ... ann   ≤ n − 2 то Aij , i = 1, ..., n, j = 1, ..., n, не определён. Из этого следует, что хотя квазидетерминант является мощным инструментом, возможность пользовать- ся определителем является серьёзным преимуществом. Теорема 6.7. Если матрица A имеет обратную матрицу, то для любых матриц B и C из равенства (6.17) следует равенство (6.18) BA = CA B = C Доказательство. Равенство (6.18) следует из (6.17), если обе части равенства (6.17) умножить на A−1. (cid:3) Теорема 6.8. Решение невырожденной системы линейных уравнений (6.19) определено однозначно и может быть записано в любой из следующих форм4 Ax = b (6.20) (6.21) x = A−1b x = HA b Доказательство. Умножая обе части равенства (6.19) слева на A−1, мы полу- чим (6.20). Пользуясь определением 6.4, мы получим (6.21). Решение системы единственно в силу теоремы 6.7. (cid:3) 4Смотри аналогичное утверждение в теореме [7]-1.6.1. 18 Александр Клейн and Иван Кирчей Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. 7. Соответствие между строчно-столбцовыми определителями и квазидетерминантами Теорема 7.1. Если A ∈ M(n, H) - обратимая матрица, то для произвольных p, q = 1, ..., n мы имеем следующие представления квазидетерминанта (7.1) (7.2) (7.3) (7.4) Apq = ddetA (Lpq)−1 = ddetA (cdetq(A∗A). q(A∗ . p))−1 = ddetA n(cdetq(A∗A). q(A∗ . p)) cdetq(A∗A). q(A∗ . p) Apq = ddetA (Rpq)−1 = ddetA (rdetp(AA∗)p .(A∗ q .))−1 = ddetA n(rdetp(AA∗)p .(A∗ q .)) rdetp(AA∗)p .(A∗ q .) Доказательство. Пусть A−1 = (bij ) - матрица обратная матрице A. Равенство (6.12) раскрывает связь между квазидетерминантом A p,q матрицы A ∈ M(n, H) и элементами обратной матрицы A−1 = (bij), а именно A pq= b−1 qp для всех p, q = 1, ..., n. В тоже время теория строчных-столбцовых определите- лей (теорема 5.33) дает нам представления обратной матрицы через ее двойные левые (5.6) и правые (5.7) алгебраические дополнения. Таким образом, соот- ветственно, получим !−1 !−1 , . , . (7.5) (7.6) (7.7) (7.8) A pq= b−1 A pq= b−1 qp =(cid:18) Lpq qp =(cid:18) Rpq ddetA(cid:19)−1 ddetA(cid:19)−1 ddetA = cdetq(A∗A). q(cid:0)A∗ . p(cid:1) = rdetp(AA∗)p .(cid:0)A∗ q .(cid:1) ddetA Так как ddetA 6= 0 ∈ F, то ∃(ddetA)−1 ∈ F. В свою очередь cdetq(A∗A). q(cid:0)A∗ rdetp(AA∗)p .(cid:0)A∗ . p(cid:1)−1 q .(cid:1)−1 = = cdetq(A∗A). q(cid:0)A∗ . p(cid:1) . p(cid:1)) n(cdetq(A∗A). q(cid:0)A∗ q .(cid:1) rdetp(AA∗)p .(cid:0)A∗ q .(cid:1)) n(rdetp(AA∗)p .(cid:0)A∗ Подставив (7.7) в (7.5), а (7.8) в (7.6), соответственно получим (7.2) и (7.4). Теорема доказана. (cid:3) Формула (7.2) дает явное представление квазидетерминанта A p,q мат- рицы A ∈ M(n, H) для всех p, q = 1, ..., n через столбцовый определитель ее соответственной левой эрмитовой матрицы A∗A, а (7.4) - через строчный опре- делитель ее соответственной правой эрмитовой AA∗. Пример 7.2. Рассмотрим матрицу a21 a22! A = a11 a12 19 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Согласно (6.13) Александр Клейн and Иван Кирчей (7.9) A = a11 − a12(a22)−1 a21 a12 − a11(a21)−1 a22 a21 − a22(a12)−1 a11 a22 − a21(a11)−1 a12! Наша задача найти этот квазидетерминант, пользуясь теоремой 7.1. Очевидно, что A∗ = a11 a21 a12a11 + a22a21 n(a12) + n(a22)! . a12 a22! A∗A = n(a11) + n(a21) a11a12 + a21a22 Вычисляем необходимые определители ddetA = rdet1(A∗A) = (n(a11) + n(a21)) · (n(a12) + n(a22)) −(a11a12 + a21a22) · (a12a11 + a22a21) = n(a11)n(a12) + n(a11)n(a22) + n(a21)n(a12) + n(a21)n(a22) −a11a12a12a11 − a11a12a22a21 − a21a22a12a11 − a21a22a22a21 = n(a11)n(a22) + n(a21)n(a12) − (a11a12a22a21 + a11a12a22a21) = n(a11)n(a22) + n(a21)n(a12) − t(a11a12a22a21) cdet1(A∗A).1(a∗ .2) = cdet1 a21 a11a12 + a21a22 a22 n(a12) + n(a22)! = n(a12)a21 + n(a22)a21 − a11a12a22 − a21a22a22 = n(a12)a21 − a11a12a22. Тогда, n(cdet1(A∗A).1(a∗ cdet1(A∗A).1(a∗ .2)) = cdet1(A∗A).1(a∗ .2) = n(a12)a21 − a22a12a11, .2) · cdet1(A∗A).1(a∗ .2) = (n(a12)a21 − a22a12a11) · (n(a12)a21 − a11a12a22) = n2(a12)n(a21) − n(a12)a21a11a12a22 −n(a12)a22a12a11a21 + a22a12a11a11a12a22 = n(a12)(n(a12)n(a21) − t(a11a12a22a21) + n(a21)n(a12)) = n(a12)ddetA. Следуя формуле (7.2), получим ddetA n(cdet1(A∗A).1(a∗ .2)) ddetA n(a12)ddetA cdet1(A∗A).1(a∗ .2) cdet1(A∗A).1(a∗ .2) (7.10) A21 = = = = 1 n(a12) 1 n(a12) · cdet1(A∗A).1(a∗ .2) · (n(a12)a21 − a22a12a11) = a21 − a22(a12)−1a11. Последнее выражение в (7.10) совпадает с выражением A21 в (7.9). (cid:3) 20 Александр Клейн and Иван Кирчей Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. 8. Список литературы [1] H. Aslaksen. Quaternionic determinants Math. Intelligencer 18(3), pp.57- 65, (1996). [2] L. Chen, Definition of determinant and Cramer solutions over quaternion field, Acta Math. Sinica (N.S.) 7, pp.171-180, (1991). [3] L. Chen, Inverse matrix and properties of double determinant over quaternion field, Sci. China, Ser. A 34, pp.528-540, (1991). [4] N. Cohen, S. De Leo, The quaternionic determinant, The Electronic Journal Linear Algebra 7, pp.100-111, (2000). [5] F. J. Dyson, Quaternion determinants, Helvetica Phys. Acta 45, pp. 289- 302, (1972). [6] Серж Ленг, Алгебра, М. Мир, 1968 [7] I. Gelfand, S. Gelfand, V. Retakh, R. Wilson, Quasideterminants, eprint arXiv:math.QA/0208146 (2002) [8] I.Gelfand, V.Retakh, Quasideterminants, I, eprint arXiv:q-alg/9705026 (1997) [9] И. М. Гельфанд, В. С. Ретах, Детерминанты матриц над некоммута- тивными кольцами, Функциональный анализ и его приложения 25(2), c. 13-35, (1991). [10] И. М. Гельфанд, В. С. Ретах, Теория некоммутативных детерминантов и характеристические функции графов, Функциональный анализ и его приложения 26 (4), c. 33-45, (1992). [11] J. Fan, Determinants and multiplicative functionals on quaternion matrices, Linear Algebra and Its Applications 369, pp. 193-201, (2003). [12] C.R.Gilson, J.J.C.Nimmo, Y.Ohta, Quasideterminant solutions of a non- Abelian Hirota-Miwa equation, Journal of Physics A: Mathematical and Theoretical 40(42), pp. 12607-12617,(2007). [13] B. Haider, M. Hassan, Quasideterminant solutions of an integrable chiral model in two dimensions, Journal of Physics A: Mathematical and Theoretical 42 (35), art. no. 355211, (2009). [14] I.I. Kyrchei, Cramer's rule for quaternion systems of linear equations, Journal of Mathematical Sciences 155(6), 839-858, (2008). Translated from Fundamental and Appl. Math. 13(4), pp.67-94, (2007). (in Russian) eprint arXiv:math.RA/0702447 (2007) [15] I.I. Kyrchei, Cramer's rule for some quaternion matrix equations, Applied Mathematics and Computation 217(5), pp.2024-2030, (2010). eprint arXiv:math.RA/arXiv:1004.4380 (2010) [16] I.I. Kyrchei,Determinantal representations of the Moore-Penrose inverse over the quaternion skew field and corresponding Cramer's rules, eprint arXiv:math.RA/1005.0736 (2010) [17] Александр Клейн, Лекции по линейной алгебре над телом, eprint arXiv:math.GM/0701238 (2010) [18] A. Lauve, Quantum- and quasi-Plucker coordinates, Journal of Algebra 296(2), pp.440-461, (2006). [19] Lewis D. W. Quaternion algebras and the algebraic legacy of Hamilton's quaternions, Irish Math. Soc. Bulletin 57, pp. 41-64, (2006). [20] C.X.Li, J.J.C. Nimmo, Darboux transformations for a twisted derivation and quasideterminant solutions to the super KdV equation, Proceedings of 21 Соответствие между строчно-столбцовыми определителями и квазидетерминантами матриц над кватернионной алгеброй. Александр Клейн and Иван Кирчей the Royal Society A: Mathematical, Physical and Engineering Sciences 466 (2120), pp. 2471-2493, (2010) [21] T. Suzuki, Noncommutative spectral decomposition with quasideterminant, Advances in Mathematics 217(5), pp. 2141-2158, (2008) 22 9. Предметный указатель (ji)-квазидетерминант 17 двойной определитель 13 инволюция в кватернионной алгебре 6 квазидетерминант 17 кватернион, сопряженный к кватерниону x 6 кватернионная алгебра 6 левое алгебраическое дополнение элемента матрицы 9 левое двойное алгебраическое дополнение элемента матрицы 14 норма в кватернионной алгебре 6 обращение Адамара матрицы 17 правое алгебраическое дополнение элемента матрицы 8 правое двойное алгебраическое дополнение элемента матрицы 14 ранг эрмитовой матрицы по главным минорам 11 след кватерниона 6 столбцовый определитель 8 строчный определитель 8 упорядоченное слева представление подстановки произведением циклов 7 упорядоченное справа представление подстановки произведением циклов 7 условие перестановочности следа 6 эрмитова матрица 4 23 10. Специальные символы и обозначения A. T минор 3 AS . минор 3 A. T минор 3 AS . минор 3 Aji (ji)-квазидетерминант матрицы A 17 A.j (b) преобразование матрицы 3 Ai. (b) преобразование матрицы 3 A квазидетерминант матрицы A 17 cdetj A столбцовый определитель по j-му столбцу матрицы A 8 ddetA двойной определитель матрицы A 13 (cid:18) a, b F (cid:19) кватернионная алгебра над полем F 6 HA обращение Адамара матрицы 17 H(a, b) кватернионная алгебра 6 Li j левое алгебраическое дополнение элемента ai j матрицы 9 Lij левое двойное алгебраическое дополнение элемента aij матрицы 14 n(x) норма кватерниона x 6 Ri j правое алгебраическое дополнение элемента ai j матрицы 8 Rij правое двойное алгебраическое дополнение элемента aij матрицы 14 rdetiA строчный определитель по i-й строке матрицы A 8 Sn симметрическая группа 7 t(x) след кватерниона x 6 24
1311.2094	2	1311	2013-12-14T19:51:26	Invariant bilinear forms of algebras given by faithfully flat descent	[ "math.RA", "math.AG", "math.RT" ]	The existence of nondegenerate invariant bilinear forms is one of the most important tools in the study of Kac-Moody Lie algebras and extended affine Lie algebras. In practice, these forms are created, or shown to exist, either by assumption or in an ad hoc basis. The purpose of this work is to describe the nature of the space of invariant bilinear forms of certain algebras given by faithfully flat descent (which includes the affine Kac-Moody Lie algebras, as well as Azumaya algebras and multiloop algebras) within a functorial framework. This will allow us to conclude the existence, uniqueness and nature of invariant bilinear forms for many important classes of algebras.	math.RA	math	INVARIANT BILINEAR FORMS OF ALGEBRAS GIVEN BY FAITHFULLY FLAT DESCENT E. NEHER, A. PIANZOLA, D. PRELAT, AND C. SEPP Abstract. The existence of nondegenerate invariant bilinear forms is one of the most important tools in the study of Kac-Moody Lie algebras and extended affine Lie algebras. In practice, these forms are created, or shown to exist, either by assumption or in an ad hoc basis. The purpose of this work is to describe the nature of the space of invariant bilinear forms of certain algebras given by faithfully flat descent (which includes the affine Kac-Moody Lie algebras, as well as Azumaya algebras and multiloop algebras) within a functorial framework. This will allow us to conclude the existence, uniqueness and nature of invariant bilinear forms for many important classes of algebras. 1. Introduction One of the key ingredients for the study of Kac-Moody Lie algebras is the (generalized) Casimir operator. The existence of this operator is in turn based upon the existence of a symmetric invariant nondegenerate bilinear form on the Lie algebra. Kac-Moody Lie algebras admitting such bilinear forms are called symmetrizable, and the most important example of these is given by the affine Kac-Moody Lie algebras. It is also known [Pi2] that the affine algebras are precisely the twisted forms (given by Galois, hence also faithfully flat descent) of the "split" affine algebras, namely of algebras of the form g ⊗k k[t±1] where g is a finite dimensional simple Lie algebra over an algebraically closed field k of characteristic 0. In this paper we shall establish the representability (in a functorial sense) and explicit description of the space of invariant k-bilinear forms for a large class of (in general infinite dimensional) algebras given by faithfully flat descent. The techniques developed in this paper not only apply to Lie algebras, but to other classes of algebras as well, such as Azumaya algebras, octonion, alternative or Jordan algebras; see §6 for an (incomplete) list of examples. Just as in the theory of Kac-Moody Lie algebras, the existence of a nondegenerate or even nonsingular invariant bilinear form on these algebras has important structural consequences. We therefore follow the approach of [NP] and study invariant bilinear forms of arbitrary (nonassociative) algebras. Among other things we will recover [MSZ, Lemma 2.3], which considers this question for Lie algebras in the untwisted case. The need to consider graded invariant forms in the study of extended affine Lie algebras will require a "graded version" of our main results. This will be given towards the end of the paper in §7, where we will provide, among other things, a classification-free proof of Yoshii's Theorem [Yo2] for multiloop Lie tori stating that graded invariant bilinear forms are unique up to scalars. We will also obtain new (as well as shed new light on known) results on Azumaya algebras (Theorem 6.10). The technique that we use to describe the nature of invariant bilinear forms of algebras is based on two crucial ingredients: Date: September 20, 2018. 2010 Mathematics Subject Classification. Primary 17B67; Secondary 17B01, 12G05, 20G10. Key words and phrases. Invariant bilinear form, Twisted form, Centroid, Faithfully flat descent. E. Neher wishes to thank NSERC for partial support through a Discovery grant. A. Pianzola wishes to thank NSERC and CONICET for their continuous support. D. Prelat is supported by a Research Grant from Universidad CAECE. C. Sepp is supported by CONICET and Universidad CAECE. 1 2 Version September 20, 2018 • That invariant bilinear forms are functorial in nature and that this functor is representable. • Descent theory. This leads us to outline a general theory of descent within a functorial setting which we find to be It is developed in §2 and later applied to the functor IBF representing of independent interest. invariant bilinear forms. Acknowledgement. We thank Ottmar Loos for his many useful suggestions and comments. The first two authors gratefully acknowledge the hospitality of BIRS (Banff) and the Fields Institute (Toronto) where part of the research for this paper was carried out. 2. Descent for functors stable under base change The goal of this section is to show that functors from the category of algebras to the category of modules that are stable under base change preserve descended forms. Relevant to our paper is the special case of the "functor of invariant bilinear forms" IBF as stated in Corollary 4.4. To write the proof in this particular case, however, obscures the more general nature of the construction that is taking place. We have thus chosen a framework that allows the essence of the argument to come across. Our setting, while somewhat abstract, is sufficiently ample for our purposes and of independent interest. By appealing to fibered categories, an even more general set up is possible in which the concept of functors stable under base change can be defined. It is nevertheless a delicate question to identify, within this more general setting, which arrows play the (crucial) role of faithfully flat base change and their accompanying descent theory. We leave it to the interested reader to explore such more general scenarios. Definition 2.1 (Categories k-alg, k-ALG, k-MOD). We fix once and for all a commutative asso- ciative unital ring k. We denote by k-alg the category of commutative associative unital k-algebras with unital k-linear algebra homomorphisms as morphisms. The symbol R ∈ k-alg means that R is an object in k-alg. We also use the notation R/k to describe this situation. By definition, R comes accompanied with a "structure" ring homomorphism σR,k : k → R under which R is viewed as a k-module. Of course since R is a commutative ring, we also have the category R-alg. An object S ∈ R-alg will be viewed as an object of k-alg via σS,k = σS,R ◦ σR,k. The arrows of R-alg are then arrows of k-alg and we have a natural forgetful functor R-alg → k-alg. Let α : R → S be a morphism in k-alg, M be an R-module and N an S-module. We say that a map f : M → N is an α-semilinear module homomorphism if it is additive and satisfies f (rm) = α(r)f (m) for all r ∈ R and m ∈ M . Such a map f is necessarily k-linear, namely if M and N are viewed as k-modules by means of the structure maps σR,k and σS,k then f (cm) = cf (m) for all c ∈ k and m ∈ M . If M and N have algebra structures and f preserves multiplication, then we say that f is an α-semilinear algebra homomorphism. Given an α-semilinear module homomorphism f : M → N as above and a morphism β : S → T in k-alg, there exists a unique β-semilinear module homomorphism f ⊗ β : M ⊗R S → N ⊗S T satisfying (f ⊗ β)(m ⊗ s) = f (m) ⊗ β(s). It is clear that if f is an algebra homomorphism, then so is f ⊗ β. The objects of the category k-ALG are pairs (R, A) consisting of an R ∈ k-alg and an R-algebra A. By this we mean an R-module A together with an R-bilinear map A×A → A, (a1, a2) 7→ a1a2. In particular, we do not require any further identities (even though Lie algebras are our main interest, all algebras are considered). A morphism in k-ALG, written as (α, f ) : (R, A) → (S, B), is a pair consisting of a morphism α : R → S in k-alg together with an α-semilinear algebra homomorphisms f : A → B.1 The category k-MOD is defined in complete analogy to k-ALG, with algebras replaced by modules and algebra homomorphisms replaced by module homomorphisms. Thus, an object in k-MOD is a pair (R, M ) consisting of some R ∈ k-alg and an R-module M and a morphism in 1 The category k-ALG appears in similar form in [LPR, 3.2]. Version September 20, 2018 3 k-MOD is a pair (α, f ) : (R, M ) → (S, N ) consisting of a morphism α : R → S in k-alg and an α-semilinear module homomorphism f : M → N . Definition 2.2 (Base change). Both categories k-ALG and k-MOD admit base change by objects of k-alg. We explain this for k-ALG and at the same time also set up our notation. Let α : R → S be a morphism in k-alg and let A be an R-algebra. Viewing S as an R-algebra via α, the tensor product A ⊗R S is an S-algebra whose product is given by (a1 ⊗ s1)(a2 ⊗ s2) = (a1a2) ⊗ (s1s2) for ai ∈ A and si ∈ S. Since we will in this section repeatedly use different algebra homomorphisms between R and S, it will be useful to temporarily employ the more precise notation A ⊗α S instead of the traditional A ⊗R S. We will also appeal to this notation elsewhere in the paper whenever it is necessary to emphasize the structure map α. We define αA : A → A ⊗α S, a 7→ a ⊗ 1S, and note that (α, αA) : (R, A) → (S, A ⊗α S) is a morphism in k-ALG. Moreover, for any mor- phism (α, f ) : (R, A) → (S, B) in k-ALG there exists a unique S-linear algebra homomorphism f α : A ⊗α S → B satifying f α(a ⊗ s) = sf (a). It is clear from the definitions that (2.1) (R, A) (α,f ) (S, B) &▲▲▲▲▲▲▲▲▲▲ (α,αA) 8rrrrrrrrrr (IdS ,f α) (S, A ⊗α S) is a commutative diagram in k-ALG. Assume that β : S → T is another morphism in k-alg. We can then add to the diagram (2.1) the morphism (β, βB) : (S, B) → (T, B ⊗β T ) and obtain the diagram (R, A) (α,f ) (S, B) (2.2) (α,αA) (β,βB) (S, A ⊗α S) / (T, B ⊗β T ) (β,f ⊗β) which commutes since (β, f ⊗ β) = (β, βB) ◦ (IdS, f α). One should view this diagram as the base change of (α, f ) by β. Definition 2.3 (Functors over k-alg). The projection onto the first component defines functors ΠA : k-ALG → k-alg and ΠM : k-MOD → k-alg. We say that a functor F : k-ALG → k-MOD is a functor over k-alg if ΠM ◦ F = ΠA, i.e., k-ALG F %❏❏❏❏❏❏❏❏❏❏ ΠA k-alg k-MOD yssssssssss ΠM is a commutative diagram. Such a functor F maps an object (R, A) ∈ k-ALG to F (R, A) = (R, FR(A)) for some R-module FR(A), and it sends a morphism (α, f ) : (R, A) → (S, B) to F (α, f ) = (α, Fα(f )) for some α-semilinear map Fα(f ) : FR(A) → FS(B). Convention. In what follows and without further explication, if R = S and α = IdS then we will simply write F (f ) instead of Fα(f ). Given a morphism α : R → S in k-alg and (R, A) ∈ k-ALG, we can apply F to the morphism (α, αA) : (R, A) → (S, A ⊗α S) in k-ALG of 2.2 and get a morphism in k-MOD F (α, αA) = (α, Fα(αA)) : (R, FR(A)) → (S, FS(A ⊗α S)). Since this map is α-semilinear, it induces an S-linear map (2.3) A,α : FR(A) ⊗α S → FS(A ⊗α S), m ⊗ s 7→ sFα(αA)(m) νF / / & 8 / / / / / % y 4 Version September 20, 2018 of S-modules. These maps will play an essential role in our work for they constitute the essential ingredient in the definition of functors stable under base change. For convenience in what follows, if F is fixed in the discussion, we will denote νF A,α simply by νA,α. Lemma 2.4. Suppose that F : k-ALG → k-MOD is a functor over k-alg. (a) (F and ν commute) Let (α, f ) : (R, A) → (S, B) be a morphism in k-ALG and let β : S → T be a morphism in k-alg. Then the diagram FR(A) ⊗α S Fα(f )⊗β FS(B) ⊗β T νA,α νB,β FS(A ⊗α S) Fβ (f ⊗β) / FT (B ⊗β T ) commutes. (b) (Transitivity) Let R α−→ S diagram β −→ T be morphisms in k-alg and let (R, A) ∈ k-ALG. Then the FR(A) ⊗α S IdFR (A) ⊗β FR(A) ⊗β◦α T νA,α νA,β◦α FS(A ⊗α S) Fβ (IdA ⊗β) / FT (A ⊗β◦α T ) commutes. Proof. (a) The maps f ⊗ β and Fα(f ) ⊗ β are described in Definitions 2.1 and 2.2. The bottom horizontal arrow is obtained by applying F to (β, f ⊗ β). For m ∈ FR(A) and s ∈ S we have (cid:0)νB,β ◦ (Fα(f ) ⊗ β)(cid:1)(m ⊗ s) = β(s)Fβ (βB)(cid:0)Fα(f )(m)(cid:1), while (cid:0)Fβ(f ⊗ β) ◦ νA,α(cid:1)(m ⊗ s) = β(s)Fβ (f ⊗ β)(cid:0)Fα(αA)(m)(cid:1). It therefore suffices to show that Fβ(βB) ◦ Fα(f ) = Fβ(f ⊗ β) ◦ Fα(αA). But this follows by applying the functor F to the commutative diagram (2.2). (b) With the notation of (a) we have (cid:0)νA,β◦α ◦ (Id ⊗β)(cid:1)(m ⊗ s) = β(s) Fβ◦α(cid:0)(β ◦ α)A(cid:1)(m), while (Fβ (IdA ⊗β) ◦ νA,α) (m ⊗ s) = β(s)Fβ (IdA ⊗β) Fα(αA) (m). It is therefore sufficient to show that F(cid:0)(β ◦ α)A(cid:1) = Fβ(IdA ⊗β) ◦ Fα(αA). By functoriality, this is a consequence of (cid:0)(β ◦ α), (β ◦ α)A(cid:1) = (β, IdA ⊗β) ◦ (α, αA). Definition 2.5 (Functors stable under base change). Let F : k-ALG → k-MOD be a functor over k-alg. We will say F is stable under base change if for all morphisms α ∈ k-alg and all (R, A) ∈ k-ALG the S-module homomorphism νF A,α : FR(A) ⊗α S → FS(A ⊗α S) defined in (2.3) is an isomorphism. (cid:3) Example 2.6. An example of a functor stable under base change is the "invariant bilinear form functor" IBF of 3.5, see Proposition 4.3(b), which is most relevant to our work. Of course the quintessential example of a functor F : k-ALG → k-MOD over k-alg stable under base change is the "tensor product" functor that attaches to (R, A) the R-module A ⊗R A with the natural definition of F at the level of arrows. The stability of base change is given by the canonical S- module isomorphism (A ⊗R A) ⊗R S ≃ (A ⊗R S) ⊗S (A ⊗R S). To justify the generality of this section we give another example. Let A be an R-algebra. Recall that its derived algebra is defined by D(A) = SpanR{a1a2 : ai ∈ A} = SpanZ{a1a2 : ai ∈ A}. If (α, f ) : (R, A) → (S, B) is a morphism in k-ALG, then f(cid:0)D(A)(cid:1) ⊂ D(B). Hence by restriction If we view D(A) as a submodule of the R-module A we obtain a map Dα(f ) : D(A) → D(B). / / / / / / Version September 20, 2018 5 it is immediate that (cid:0)α, Dα(f )(cid:1) : (cid:0)R, D(A)(cid:1) → (cid:0)S, D(B)(cid:1) is a morphism in k-MOD and that the assignment defines a functor D : k-ALG → k-MOD over k-alg. (R, A) 7→(cid:0)R, D(A)(cid:1) and (α, f ) 7→(cid:0)α, Dα(f )(cid:1) The explicit nature of the map ν D A,α : D(A)⊗αS → D(A⊗αS) is clear: a1a2⊗s 7→ (a1⊗s)(a2⊗1) = a1a2 ⊗ s = (a1 ⊗ 1)(a2 ⊗ s). Thus νD A,α is always surjective. This map, however, need to be injective since there is no reason for the natural map D(A) ⊗α S → A ⊗α S to be so (it is, for example, if α : R → S is flat). Consider a new functor F : k-ALG → k-MOD over k-alg which assigns to (R, A) the pair (cid:0)R, A/D(A)(cid:1) and is defined at the level of arrows in the natural way. We leave it to the reader to A,α is an isomorphism of S-modules. Thus F A,α easily implies that νF check that the surjectivity of νD is stable under base change. Remark 2.7. The results of this section can be generalized by replacing k-alg, k-ALG and k-MOD by subcategories stable under base change and by modifying the Definition 2.3 correspondingly. As we shall see, the most relevant case for us is that of functors which are stable under faithfully flat base change. We will leave it to the interested reader to work out the necessary axioms. Descent Theory 2.8 (Faithfully flat descent of modules and algebras). We give a short review of the descent theory of modules and algebras. Our ultimate objective is to outline a descent theory in the setting of functors stable under faithfully flat base change. We also use the opportunity to introduce notation and a presentation of descent theory that is implicitly, but not explicitly used in the standard references ([KO, SGA1, Wa]), cf. [Pi1]. Without these the formalism for descent in the functorial setting is impossible to redact. Assume that S/R is faithfully flat. We let S′′ = S ⊗R S and denote by αi : S → S′′, i = 1, 2, the "projections" defined by α1(s) = s ⊗ 1 and α2(s) = 1 ⊗ s, which allow us to view S′′ as an S-algebra in two different ways. Note that since α : R → S is faithfully flat, α1 ◦ α = α2 ◦ α. Suppose M and N are R-modules and that N is an S/R-form of M . Thus there exists an S-module isomorphism θ : (M ⊗α S) → (N ⊗α S). To θ and i = 1, 2 we associate the S′′-module isomorphisms θi defined by the following commutative diagram. (2.4) (M ⊗α S) ⊗αi S′′ τ M i ≃ M ⊗αi◦α S′′ θ⊗αi IdS′′ ≃ θi ≃ (N ⊗α S) ⊗αi S′′ τ N i≃ / N ⊗αi◦α S′′ Here τi : S⊗αi S′′ → S′′ is defined by τi(s1⊗s2⊗s3) = αi(s1)(s2⊗s3), e.g. τ2(s1⊗s2⊗s3) = (s2⊗s1s3), while τ M i are defined similarly. In what follows, τ M i (cid:0)(m ⊗ s1) ⊗ s2 ⊗ s3(cid:1) = m ⊗ αi(s1)(s2 ⊗ s3). The maps τ N i are viewed as S′′-linear maps. and τ N i The situation can be summarized by the following commutative diagram (2.5) 0 0 / M ≃ M ⊗IdR R IdM ⊗α / N ≃ N ⊗IdR R IdN ⊗α / M ⊗α S ≃ θ / N ⊗α S IdM ⊗α1 IdM ⊗α2 IdN ⊗α1 IdN ⊗α2 M ⊗αi◦α S′′ θ2 θ1 / N ⊗αi◦α S′′ The rows are exact since α : R → S is faithfully flat, e.g., IdM ⊗α is injective and its image M ⊗ 1S is the R-submodule M ⊗α S where IdM ⊗α1 and IdM ⊗α2 agree. The S/R-cocycle u defining the S/R-form N is u = θ−1 2 ◦ θ1 ∈ AutS′′ (M ⊗α1◦α S′′) = AutS′′(M ⊗R S′′) / / / / / / / / / / / / / / 6 Version September 20, 2018 as we now explain.2 Let (2.6) L = {x ∈ M ⊗α S : u(cid:0)(IdM ⊗α1)(x)(cid:1) = (IdM ⊗α2)(x)}. It is clear that L is an R-submodule of M ⊗α S = M ⊗R S, and a simple diagram chase in (2.5) above shows that the restriction of θ to L induces an isomorphism with N ⊗ 1S ⊂ N ⊗α S = N ⊗R S. In other words, up to R-module isomorphism, our module N corresponds to the cocycle u. It is well-known (and easy to check in any case) that u is a cocycle, i.e. that u ⊗ α1,3 = (u ⊗ α2,3) ◦ (u ⊗ α1,2) where the αi,j : S′′ → S′′′ = S ⊗α S ⊗α S are the natural S′′-algebra morphisms defined by putting 1S in the position l 6= i, j. The cocycle condition can be rewritten in the form θ1,3 = θ2,3 ◦ θ1,2 where the θi,j are automorphisms of the S′′′-module M ⊗αi,j ◦αi◦α S′′′ = M ⊗R S′′′ defined using a diagram similar to (2.4). For R-algebras the situation is identical. Say that both A and B are R-algebras and that our isomorphism θ above is now an S-algebra isomorphism. Then u is an S′′-algebra automorphism of A ⊗R S′′, the descended R-module L is an R-subalgebra of A ⊗α S = A ⊗R S and the restriction of θ induces an R-algebra isomorphism between L and B ≃ B ⊗ 1S. Theorem 2.9. Let A be an R-algebra, and let B be an S/R form of A determined by the cocycle u as described in 2.8 above. If F : k-ALG → k-MOD is a functor over k-alg stable under base change, then FR(B) is an S/R-form of the R-module FR(A) which is isomorphic as an R-module to the one given by the cocycle ν−1 A,α2◦α ◦ F (u) ◦ νA,α1◦α ∈ AutS′′(cid:0)FR(A) ⊗R S′′(cid:1). Proof. We fix an S-algebra isomorphism θ : A ⊗α S → B ⊗α S. The cocycle determining B (up to R-algebra isomorphism) is then u = θ−1 2 ◦ θ1 ∈ AutS′′ (A ⊗R S′′). The result to establish can thus be rephrased by saying that (a) z :=(cid:0)ν−1 a cocycle. A,α2◦α ◦ F (θ2)−1 ◦ νB,α2◦α(cid:1) ◦ (cid:0)ν−1 B,α1◦α ◦ F (θ1) ◦ νA,α1◦α(cid:1) ∈ AutS′′(cid:0)FR(A) ⊗R S′′(cid:1) is (b) The R-module determined by the cocycle z is isomorphic to FR(B). Let us define F ν(θ) = ν−1 B,α ◦ F (θ) ◦ νA,α by means of the diagram FR(A) ⊗α S νA,α F ν (θ) FR(B) ⊗α S (νB,α)−1 FS(A ⊗α S) F (θ) / FS(B ⊗α S) In view of the descent of modules construction explained in 2.8, to prove (a) and (b) it will suffice to show that (2.7) ν−1 B,αi◦α ◦ F (θi) ◦ νA,αi◦α =(cid:0)F ν(θ)(cid:1)i for i = 1, 2. 2Note that u can indeed be viewed as an S ′′-module automorphism M ⊗α1 ◦α S ′′ because α1 ◦ α = α2 ◦ α. In what follows we will view S ′′ as an R-algebra via either one of these two (equal) maps. The notation M ⊗R S ′′ responds to this convention. / / / O O Version September 20, 2018 7 Both cases i = 1, 2 are similar and we check in detail the case i = 1 only. We will use the following commutative diagram of S′′-module isomorphisms. FR(A) ⊗α1◦α S′′ (cid:0)F ν (θ)(cid:1)1 FR(B) ⊗α1◦α S′′ (IdFR (A) ⊗τ1)−1 (IdFR (B) ⊗τ1)−1 FR(A) ⊗α S ⊗α1 S′′ F ν (θ)⊗IdS′′ FR(B) ⊗α S ⊗α1 S′′ νA,α⊗IdS′′ νB,α⊗IdS′′ νA,α1 ◦α FS(A ⊗α S) ⊗α1 S′′ F (θ)⊗IdS′′ FS(B ⊗α S) ⊗α1 S′′ νB,α1 ◦α νA⊗α S,α1 νB⊗α S,α1 FS′′(cid:0)(A ⊗α S) ⊗α1 S′′(cid:1) F (θ⊗α1 IdS′′ ) FS′′(cid:0)(B ⊗α S) ⊗α1 S′′(cid:1) F (τ A 1 ) F (τ B 1 ) FS′′ (A ⊗α1◦α S) F (θ1) / FS′′ (B ⊗α1◦α S′′) The top rectangle commutes by definition of F ν (θ1); the second rectangle commutes by applying the base change α1 : S → S′′ to the diagram defining F ν (θ); the third rectangle commutes by Lemma 2.4(a) for (R, A), (S, B), (α, f ) and β replaced by (S, A ⊗α S), (S, B ⊗α S), (IdS, θ) and IdS′′ respectively; the bottom rectangle commutes by applying F to the diagram (2.4) defining θ1. For the proof of (2.7) it is therefore sufficient to show that the dotted maps equal νA,α1◦α and νB,α1◦α respectively. We check the case of νA,α1◦α by following explicitly the arrows on the left of the diagram. The case of νB,α1◦α, which is similar, is left to the reader. Let m ∈ FR(A) and s1, s2 ∈ S. Then m ⊗ s1 ⊗ s2 (IdFR (A) ⊗τ1)−1 −−−−−−−−−−→ m ⊗ s1 ⊗ 1S ⊗ s2 νA⊗α S,α1 −−−−−−→ (1S ⊗ s2)(cid:16)Fα1 (αA⊗αS 1 νA,α⊗IdS′′ −−−−−−−→(cid:0)s1Fα(αA)(m)(cid:1) ⊗ (1S ⊗ s2) )(cid:0)s1Fα(αA)(m)(cid:1)(cid:17) = (1S ⊗ s2)(cid:16)α1(s1)Fα1 (αA⊗αS = (1S ⊗ s2)(s1 ⊗ 1S)(cid:16)Fα1 (αA⊗αS = (s1 ⊗ s2)(cid:16)Fα1 (αA⊗αS 1 1 1 )(cid:0)Fα(αA)(m)(cid:1)(cid:17) )(cid:0)Fα(αA)(m)(cid:1)(cid:17) )(cid:0)Fα(αA)(m)(cid:1)(cid:17) F (τ A 1 ) −−−−→ (s1 ⊗ s2)F (τ A 1 )(cid:16)Fα1 (αA⊗αS 1 )(cid:0)Fα(αA)(m)(cid:1)(cid:17) = (s1 ⊗ s2) Fα1◦α(cid:0)(α1 ◦ α)A(m)(cid:1). This completes the proof since by definition νA,α1◦α(m⊗s1⊗s2) = (s1 ⊗s2) Fα1◦α(cid:0)(α1 ◦α)A(m)(cid:1). (cid:3) Remark 2.10. In the case that we are most interested in, the R-algebra A is of the form A = a⊗k R for some k-algebra a. The isomorphism θ can now be thought as an S-algebra isomorphism (also denoted by θ) θ : a ⊗k S ≃ (a ⊗k R) ⊗R S = A ⊗R S → B ⊗R S where we have denoted ⊗α by ⊗R. Under the canonical S′′-isomorphism a ⊗k S′′ ≃ (a ⊗k R) ⊗R S′′ (where we recall that S′′ is viewed as an R-algebra via α1 ◦α = α2 ◦α) we can view the corresponding cocycle u as an S′′-algebra automorphism of a ⊗k S′′. The descended module L is then given by (2.8) L = {x ∈ a ⊗k S : u(cid:0)(Ida ⊗α1)(x)(cid:1) = (Ida ⊗α2)(x)}. / / $ $ z z / / / / / / / (3.1) ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ≃ y 5❦❦❦❦❦❦❦❦❦❦❦❦❦❦ ≃ z HomR(cid:0)M, Homk(M, V )(cid:1) 8 Version September 20, 2018 For future reference we record the following "descent setting" that covers the case that we are most interested in: (2.9)   a is an algebra over k; R ∈ k-alg is such that R/k is flat; S ∈ R-alg is such that S/R is faithfully flat; B is an R-algebra which is an S/R-form of A = a ⊗k R. 3. Invariant functions Unless stated otherwise, k is a commutative associative unital ring, R is a commutative associative unital k-algebra, namely R ∈ k-alg in the notation of §2, M is an R-module, V is a k-module and B is an arbitrary (not necessarily associative, unital...) R-algebra, i.e., (R, B) ∈ k-ALG. Our goal is to study invariant bilinear functions B × B → V . We begin with pertinent definitions. Definition 3.1 ((R, k)-bilinear functions). A k-bilinear function β : M × M → V is called (R, k)- bilinear if β(rm1, m2) = β(m1, rm2) holds for all mi ∈ M and r ∈ R. We denote by L 2 (R,k)(M ; V ) the R-module of (R, k)-bilinear maps M × M → V . Its R-module structure is given by (rβ)(m1, m2) = β(rm1, m2) for r ∈ R and β ∈ L 2 R(M ) = L 2 It is immediate from [B:A, II, §4.1] that one has a commutative triangle (R,k)(M ; V ). We abbreviate L 2 (R,R)(M ; R). Homk(M ⊗R M, V ) x ≃ L 2 (R,k)(M ; V ) This is an R-linear map with respect to the R-action on Homk(M ⊗RM, V ) given by (rϕ)(m1 ⊗m2) = of R-linear isomorphisms: For ϕ ∈ Homk(M ⊗R M, V ) one defines (cid:0)x(ϕ)(cid:1)(m1, m2) = ϕ(m1 ⊗ m2). ϕ(rm1 ⊗ m2) for r ∈ R. Also, (cid:0)y(ϕ)(cid:1)(m1) maps m2 onto ϕ(m1 ⊗ m2). The map y is R-linear HomR(cid:0)M, Homk(M, V )(cid:1) one associates the (R, k)-bilinear function z(β) defined by(cid:0)z(β)(cid:1)(m1, m2) = (cid:0)β(m1)(cid:1)(m2). if we view Homk(M, V ) as an R-module via (rh)(m) = h(rm) for h ∈ Homk(M, V ). To β ∈ (R,k)(M ; k) nondegenerate (respectively nonsingular ) if z−1(β) ∈ HomR(M, M ∗), M ∗ = Homk(M, k), is injective (respectively bijective).3 In more familiar terms β ∈ L 2 (R,k)(M ; k) is nondegenerate if and only if β(b, M ) = 0 implies b = 0. If k is a field, the existence of a nonsingular bilinear form on a k-algebra B forces B to be of finite dimension. Moreover, for a finite-dimensional B a form is nondegenerate if and only if it is nonsingular. See Lemma 6.16 for a generalization. One calls β ∈ L 2 Definition 3.2 (Invariant functions). We call β ∈ L 2 (R,k)(B; V ) invariant if (3.2) β(ab, c) = β(a, bc) = β(b, ca) holds for all a, b, c ∈ B. Clearly, the set IBF(R,k)(B; V ) of all invariant (R, k)-bilinear functions B × B → V is a submod- (R,k)(B; V ). The following special cases of IBF(R,k)(B; V ) are of particular ule of the R-module L 2 interest: IBFk(B; V ) : = IBF(k,k)(B; V ), IBF(R,k)(B) : = IBF(R,k)(B; k), IBFk(B) : = IBFk(B; k) = IBF(k,k)(B). 3The asymmetry in these definitions (one should, strictly speaking, speak of left and right nondegeneracy and nonsingularity) will not play a major role in this paper since our main interest later will be in invariant bilinear forms of perfect algebras which, by Remark 3.4, are symmetric. / / 5 Version September 20, 2018 9 The elements of IBFk(B) are called invariant k-bilinear forms. Of particular importance is the case k = R; these are the invariant R-bilinear forms on B. Remark 3.3. The above definition works equally well for invariant bilinear functions on dimodules of algebras (for the definition of a dimodule see [NP] as well as Lemma 3.7 supra). This is not without interest as these types of bilinear forms are an important tool for the study of the representation theory of the algebras in question, for example for the existence of the Jantzen filtration of Verma modules. We will not pursue this more general set up in this work. Remark 3.4. If B is perfect, namely if B = BB, where BB = SpanR{ab : a, b, ∈ B}, every invariant bilinear function is symmetric: β(ab, c) = β(a, bc) = β(b, ca) = β(bc, a) = β(c, ab). Moreover, any k-bilinear function is already (R, k)-bilinear: IBFk(B; V ) = IBF(R,k)(B; V ) (B perfect). Indeed, β(cid:0)r(ab), c(cid:1) = β(cid:0)a(rb), c(cid:1) = β(cid:0)a, (rb)c(cid:1) = β(cid:0)a, b(rc)(cid:1) = β(cid:0)ab, rc(cid:1). Definition 3.5 (Universal invariant function). Let IBFR(B) be the quotient of the R-module B ⊗R B by the submodule ibf R(B) = SpanR{ab ⊗ c − a ⊗ bc, ab ⊗ c − b ⊗ ca : a, b, c ∈ B} = SpanZ{ab ⊗ c − a ⊗ bc, ab ⊗ c − b ⊗ ca : a, b, c ∈ B} (3.3) = SpanZ{ab ⊗ c − a ⊗ bc, a ⊗ bc − bc ⊗ a : a, b, c ∈ B}, where the last equality follows from (bc⊗a−a⊗bc)+(ab⊗c−b⊗ca) = (ab⊗c−a⊗bc)+(bc⊗a−b⊗ca). Denoting by iB : ibfR(B) → B ⊗R B the inclusion and by qB the canonical quotient map qB : B ⊗R B → IBFR(B), a ⊗ b 7→ a ⊗ b, we have an exact sequence of R-modules (3.4) 0 → ibf R(B) iB−→ B ⊗R B qB−−→ IBFR(B) → 0. We define an invariant R-bilinear function (3.5) called the universal invariant R-bilinear function.4 This terminology is justified because of the following natural R-module isomorphism βuni : B × B → IBFR(B), βuni(a, b) = a ⊗ b, (3.6) Its inverse (3.7) Homk(IBFR(B), V ) ≃−→ IBF(R,k)(B; V ), f 7→ f = f ◦ βuni. IBF(R,k)(B; V ) ≃−→ Homk(IBFR(B), V ), β 7→ ¯β assigns to β ∈ IBF(R,k)(B, V ) the unique k-linear map (3.8) ¯β : IBFR(B) → V, ¯β(a ⊗ b) = β(a, b). In other words, IBFR(B) represents the obvious functor IBF(B; −) : k-mod → R-mod. The isomorphism (3.6) determines IBFR(B) up to a unique k-linear isomorphism. We will describe IBFR(B) for several cases of interest in this paper. The most important situation is captured by the following result that we state in the form of a principle. IBF-Principle 3.6. Assume B is an R-algebra for some R ∈ k-alg. Let β ∈ IBFR(B) be such that the induced map ¯β : IBFR(B) → R, b1 ⊗ b2 7→ β(b1, b2) is an R-module isomorphism. Then for any k-module V the map (3.9) Homk(R, V ) → IBF(R,k)(B; V ), ϕ 7→ ϕ ◦ β is an isomorphism of R-modules. In particular, (a) the map R∗ = Homk(R, k) → IBF(R,k)(B), ϕ 7→ ϕ ◦ β is an isomorphism, and (b) every γ ∈ IBFR(B) is of the form γ = rβ for a unique r ∈ R. 4We want to thank K.-H. Neeb for bringing this concept to our attention. 10 Version September 20, 2018 Proof. This follows from the isomorphism (3.6) and the equality β = ¯β ◦ βuni. (cid:3) We will say that (B, β) satisfies the IBF-principle if the assumptions and hence also the conclu- sions of 3.6 hold. Note that in this case we have a precise and explicit description of all invariant (R, k)-bilinear functions on B. *** While the connection between invariant bilinear forms and centroids does not feature prominently in this paper, it has nevertheless been an important guiding principle for our work. Besides the conceptual importance of this connection, another reason for elaborating on it is Corollary 3.7 relating invariant forms and the centroid of an algebra. Not only will this provide the reader with a means to determine the module IBF(R,k)(B), but it will also be useful in §6 when we will be looking at algebras with a "1-dimensional" IBF(R,k)(B). In preparation for these results, we first present the necessary background. Using the terminology of [NP], we recall that a (B, R)-dimodule is an R-module M together with R-bilinear maps B × M → M and M × B → M . For example, B itself is a (B, R)-dimodule with respect to the left and right multiplications of the algebra B, called the regular dimodule and denoted Breg. Also the R-module Homk(B, V ) is a (B, R)-dimodule with respect to the B-actions(cid:0)b1 · ϕ(cid:1)(b2) = ϕ(b2b1) =(cid:0)ϕ · b2(cid:1)(b1). For any (B, R)-dimodule M the centroid of B with values in M is CtdR(B, M ) = {χ ∈ HomR(B, M ) : χ(b1b2) = b1 · χ(b2) = χ(b1) · b2 for all b1, b2 ∈ B}. Taking as M the regular dimodule Breg, we recover the usual notion of the centroid of B: CtdR(B) = CtdR(B, Breg). We note that CtdR(B) is isomorphic to the usual centre if B is a unital algebra, where, we recall, the centre of an arbitrary algebra B consists of those c ∈ B which commute with all b ∈ B, i.e. cb = bc, and which associate with all b1, b2 ∈ B, i.e., (c, b1, b2) = (b1, c, b2) = (b1, b2, c) = 0 where (x, y, z) = (xy)z − x(yz). There exists a natural map R → CtdR(B) given by r 7→ χr where χr(b) = rb for all b ∈ B. We call B a central R-algebra if this map is an isomorphism. In this situation we will often identify R with CtdR(B) without any further explicit reference. We can now describe the connection between invariant functions and centroids. Lemma 3.7. The restriction of the R-isomorphism z of (3.1) to CtdR(cid:0)B, Homk(B, V )(cid:1) induces an R-linear isomorphism CtdR(B, Homk(B, V )) ≃ IBF(R,k)(B; V ). In particular, CtdR(B, B∗) ≃ IBF(R,k)(B). Proof. The result is a straightforward consequence of the various definitions. (cid:3) Corollary 3.8. Assume β ∈ IBF(R,k)(B) is nonsingular. Then CtdR(B) ≃ IBF(R,k)(B) (isomorphism of R-modules). If furthermore B is a central R-algebra, then IBF(R,k)(B) is a free R-module of rank 1 admitting β as a basis: (3.10) IBF(R,k)(B) = Rβ ≃ R. Proof. By assumption χβ = z−1(β) ∈ CtdR(B, B∗) is an isomorphism of R-modules. The fact that χβ is centroidal amounts to saying that Breg ≃ B∗ as dimodules, whence IBF(R,k)(B) ≃ CtdR(B, B∗) ≃ CtdR(B) ≃ R. Via these isomorphism we have β 7→ χβ 7→ χ−1 β ◦ χβ = IdB 7→ 1R. Since 1R is a basis of R, β is a basis of IBF(R,k)(B). (cid:3) As we will see in §6, there are many natural types of algebras satisfying the assumptions of Corollary 3.8. The following result will allow us to transition from the general setting to the specific examples. Proposition 3.9. Assume that B is a central R-algebra, that β ∈ IBFR(B) is nonsingular with β(B, B) = R, and that IBFR(B) is projective. Then the IBF-principle 3.6 holds for (B, β). Version September 20, 2018 11 Proof. By assumption ¯β is surjective. Denoting by K the kernel of ¯β, we obtain a split exact sequence 0 → K → IBFR(B) of R-modules and consequently a split exact sequence ¯β −→ R → 0 0 → HomR(R, R) β −→ HomR(IBFR(B), R) → HomR(K, R) → 0 with β(r IdR) = r IdR ◦ ¯β = r ¯β for r ∈ R. The R-module isomorphism (3.6) sends ¯β to β. We also know that IBFR(B) ≃ R by Corollary 3.8 (applied to the case k = R). Hence the diagram HomR(R, R) β / HomR(IBFR(B), R) r IdR ✤ ≃ R ≃ ≃ / IBFR(B) r❴ / r ¯β ❴ / rβ commutes, proving that β is an isomorphism. This in turn forces K (∗) := HomR(K, R) = {0}.5 Since IBFR(B) ≃ K ⊕ R is projective, so is K. But then the canonical map K → (K (∗))(∗) = 0 is injective. Thus K = 0. (cid:3) 4. Functorial nature of IBF As in the previous section k will denote a commutative associative unital ring and R ∈ k-alg. The main purpose of this section is to describe the functorial nature of IBF and study its behaviour under base change. Let (α, f ) : (R, M ) → (S, N ) be a morphism in k-MOD (see 2.1), i.e., α : R → S is a morphism in (S,k)(N ; V ) k-alg and f : M → N is α-semilinear, and suppose V is a k-module. Then for any κ ∈ L 2 the map (4.1) f ∗(κ) : M × M → V, is (R, k)-bilinear. We obtain in this way a k-linear map (m1, m2) 7→ κ(cid:0)f (m1), f (m2)(cid:1) f ∗ : L 2 (S,k)(N ; V ) → L 2 (R,k)(M ; V ), κ 7→ f ∗(κ). If (β, g) : (S, N ) → (T, P ) is another morphism in k-MOD, it is immediate that (g ◦ f )∗ = f ∗ ◦ g∗. Observe that this in particular defines a right action of the group GLR(M ) on L 2 (R,k)(M ; V ) (see 5.2 for a functorial version of this observation). Base change 4.1. Let κ : M × M → R be an R-bilinear form and let α : R → S be a morphism in k-alg. There exists a unique S-bilinear form κα : M ⊗α S × M ⊗α S → S satisfying κα(m1 ⊗ s1, m2 ⊗ s2) = α(cid:0)κ(m1, m2)(cid:1)s1s2. In case α is clear from the context, we will denote κα = κS and call κS the base change of κ by S. We then have the equation κS(m1 ⊗ s1, m2 ⊗ s2) = κ(m1, m2)s1s2. Base change can also be understood in terms of the isomorphism x : HomR(M ⊗R M, R) ≃−→ R(M ) of (3.1). Let x−1(κ) = κ : M ⊗R M → R be the R-linear map associated to κ. Then x−1(κS) = fκS is obtained from κ ⊗ IdS with the aid of two canonical S-module isomorphisms, namely L 2 eκ⊗IdS−−−−→ R ⊗R S ≃ S. The following lemma collects some results using base change. fκS : (M ⊗R S) ⊗S (M ⊗R S) ≃ (M ⊗R M ) ⊗R S Lemma 4.2. (a) (Transitivity of base change) Let κ ∈ L 2 k-alg, and let ζ : (M ⊗α S)⊗β T → M ⊗β◦α T , m⊗ s⊗ t 7→ m⊗(cid:0)β(s)(cid:1)t be the canonical isomorphism of T -modules. Then R(M ), let R α−→ S β −→ T be morphisms in (4.2) ζ ∗κβ◦α = (κα)β. 5To avoid any possible confusion we use (∗) as opposed to ∗ to denote the R-dual given that, by convention, ∗ always refers to the k-dual. / O O / / O O ✤ / 12 Version September 20, 2018 (b) Let M and N be R-modules, λ ∈ L 2 R(N ), f : M → N an R-linear map and α : R → S a morphism in k-alg. Then (4.3) (cid:0)f ∗(λ)(cid:1)α = (f ⊗ IdS)∗(λα). (c) Let κ, κ′ ∈ L 2 R(M ). Then κ = κ′ if and only if κS = κ′ S for some faithfully flat extension S ∈ R-alg. (d) Assume that M is finitely presented. Let κ ∈ L 2 R(M ) and let S ∈ R-alg be such that S is a flat R-module. If κ is nondegenerate (resp. nonsingular), then κS is nondegenerate (resp. nonsingular). The converse holds in both cases if S/R is faithfully flat. Proof. The proofs of (a) and (b) are immediate from the definitions. In (c) suppose κS = κ′ some faithfully flat S ∈ R-alg. Then (cid:0)x−1(κ)(cid:1)S = x−1(κS) = x−1(κ′ (M ⊗R S) ⊗S (M ⊗R S) → S by 4.1. But then x−1(κ) = x−1(κ′) by faithfully flat descent, whence κ = κ′. S) = (cid:0)x−1(κ′)(cid:1)S as maps S for (d) Since M is finitely presented and S/R is flat, the canonical map ω : HomR(M, R) ⊗R S → HomS(M ⊗RS, S) is an isomorphism. Recall from (3.1) the R-linear map z−1(κ) : M → HomR(M, R). It is immediate from the definitions that M ⊗R S z−1(κ)⊗IdS HomR(M, R) ⊗R S (◗◗◗◗◗◗◗◗◗◗◗◗◗ z−1(κS) u❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ω HomS(M ⊗R S, S) is a commutative diagram. Hence z−1(κS) is injective (resp. bijective) if and only if z−1(κ) is so. The claim then follows from standard properties of flat (resp. faithfully flat) extensions. (cid:3) Proposition 4.3 (IBF and ibf as functors). (a) Let (α, f ) : (R, B) → (S, C) be a morphism in k-ALG. The map f ⊗ f : B ⊗R B → C ⊗S C, b1 ⊗ b2 7→ f (b1) ⊗ f (b2) is α-semilinear and maps ibf R(B) to ibf S(C). We denote by ibf α(f ) : ibf R(B) → ibf S(C) the re- stricted map and by IBFα(f ) : IBFR(B) → IBFS(C) the induced quotient map. (b The assignments (R, B) →(cid:0)R, IBFR(B)(cid:1) and (α, f ) 7→(cid:0)α, IBFα(f )(cid:1) define a functor IBF : k-ALG → k-MOD over k-alg which is stable under base change in the sense of 2.5. (c) The assignments define a functor ibf : k-ALG → k-MOD over k-alg which is stable under flat base change. (R, B) →(cid:0)R, ibf R(B)(cid:1) and (α, f ) 7→(cid:0)α, ibf α(f )(cid:1) Proof. (a) is straightforward. We will prove (b) and (c) at the same time. It is easy to verify that IBF and ibf are functors over k-alg. We have already noted in Example 2.6 that B 7→ B ⊗R B and (α, f ) 7→ (α, f ⊗ f ) defines a functor over k-alg which is stable under base change. Indeed, for any morphism α : R → S the map νB,α of (2.3) is the well-known isomorphism νB,α : (B ⊗R B) ⊗R S → (B ⊗R S) ⊗S (B ⊗R S), b1 ⊗ b2 ⊗ s 7→ b1 ⊗ 1R ⊗ b2 ⊗ s. In the following we will abbreviate ν = νB,α and ¯ν = ¯νB,α : IBFR(B) ⊗α S → IBFS(B ⊗α S). We have the following diagram with exact rows (c) /❴❴❴ 0 ibf R(B) ⊗R S iB ⊗IdS / / (B ⊗R B) ⊗R S qB ⊗IdS / IBFR(B) ⊗R S (4.4) ν≃ ν 0 / ibf S(B ⊗R S) iB⊗R S / (B ⊗R S) ⊗S (B ⊗R S) qB⊗R S / IBFS(B ⊗R S) / 0 / 0 / / ( u / ✤ ✤ ✤ / / / / / / Version September 20, 2018 13 where the top row is obtained by tensoring (3.4) with S and the bottom row is (3.4) for B ⊗R S (under assumption (c), namely when S/R is flat, then iB ⊗ IdS is injective. This is reflected by the dashed line at the top left of the diagram). One easily verifies that ν sends the S-submodule (iB ⊗ IdS)(ibf R(B) ⊗R S) of (B ⊗R B) ⊗ S onto ibfS(B ⊗R S) ⊂ (B ⊗R S) ⊗S (B ⊗R S). It follows that by restriction ν induces an S-module isomorphism between (iB ⊗ IdS)(ibf R(B) ⊗R S) and ibf S(B ⊗R S). Since from the definitions of ν and ν the right hand side square of diagram (4.4) commutes, a simple diagram chase shows that ν is an isomorphism. In case S is a flat extension, the dashed vertical arrow is injective, hence bijective. (cid:3) Corollary 4.4. Let B be an S/R-form of A = a ⊗k R as in the descent setting 2.9. Let u ∈ Aut(a)(S′′) be a cocycle determining B up to R-algebra isomorphism (see Remark 2.10). We denote by ν : IBFk(a) ⊗k S′′ → IBFS′′ (a ⊗k S′′) the isomorphism (2.3) for F = IBF. Then IBFR(B) is an S/R-form of IBF(a ⊗k R) which is isomorphic as an R-module to the one given by the cocycle ν−1 ◦ IBF(u) ◦ ν. Proof. This is a special case of Theorem 2.9. (cid:3) Remark 4.5. In the above corollary we can take k = R. The result then applies to an arbitrary R-algebra A. Lemma 4.6. Let B be an R-algebra, β ∈ IBFR(B), and S ∈ R-alg. (a) We denote by ν : IBFR(B) ⊗R S → IBFS(B ⊗R S) the isomorphism (2.3) for F = IBF, by ( ¯β)S the base change of ¯β : IBFR(B) → R and by (βS) the map (3.8) associated to the bilinear form βS. Then, after identifying R ⊗R S = S, we have ( ¯β)S = (βS) ◦ ν : IBFR(B) ⊗R S ( ¯β)S / R ⊗R S ν ≃ IBFS(B ⊗R S) (βS) / S (b) Assume that B is finitely presented as a R-module and that S/R is flat. If (B, β) satisfies the IBF-principle 3.6, then so does (BS, βS). The converse is true whenever S/R is faithfully flat. Proof. Part (a) is immediate from the definitions. In (b) we know from (a) that βS is an isomorphism. The assertion about faithful flatness is standard. (cid:3) 5. Descent of bilinear forms In this section we will study the descent of bilinear forms in the setting of (2.9). We are interested in having a full understanding of all k-bilinear forms on B. The guiding principle is that this ought to be possible if one knows the nature of the k-bilinear forms of a, for example if a is a finite-dimensional central-simple Lie algebra over a field of characteristic 0 (Theorem 6.5). That this may be possible at all, is somehow surprising. The twisted nature and descent theory related to B views B as an object over R (note that it is not the case that B is in any meaningful way a twisted form of a ⊗k R or a as k-algebras). Yet the information that we will get is about k-bilinear forms B × B → k. As mentioned before, it is the k- (and not R-) bilinear forms on B which are often of interest (such as in the case of infinite dimensional Lie theory as exemplified, for example, by the affine Kac-Moody Lie algebras. See also §6 and §7 below). Assume that κ is a k-bilinear form on a. We want to know when, in a natural fashion, we can attach to κ an R-bilinear form κB on B. It will be κB that will lead us to fully understand all k-bilinear functions on B. The key assumption that makes this construction possible and natural is that κ be invariant under algebra automorphisms. We define this concept before proceeding with the main results. Automorphism invariance 5.1. For an R-algebra B we denote by Aut(B) the automorphism group functor of B. We remind the reader that Aut(B) is the functor from the category R-alg to the category of groups that attaches to S ∈ R-alg the group AutS(B ⊗R S) of automorphisms of the / / 14 Version September 20, 2018 S-algebra B ⊗R S, and to an arrow S → T in R-alg and f ∈ AutS(B ⊗R S) the automorphism f ⊗IdT of B ⊗R T ≃ (B ⊗R S) ⊗S T . We say that β ∈ L 2 R(B) is Aut(B)-invariant if f ∗(βS) = βS holds for all S ∈ R-alg and all f ∈ Aut(B)(S), where we remind the reader that f ∗(βS)(b1 ⊗ s1, b2 ⊗ s2) = βS(cid:0)f (b1 ⊗ s1), f (b2 ⊗ s2)(cid:1). In other words, β is Aut(B)-invariant if and only if βS is AutS(B ⊗R S)- invariant in the obvious sense for all S ∈ R-alg. Remark 5.2. Automorphism invariance has a functorial interpretation. Namely, we have a functor L 2(B) : R-alg → R-MOD which assigns to the extension S/R the S-module L 2 S (B ⊗R S), and is given at the level of arrows by base change 4.1. The automorphism group functor Aut(B) acts on the functor L 2(B) from the right. A bilinear form β ∈ L 2 R(B) is Aut(B)-invariant if and only if it is invariant under this action. In particular, the above considerations apply to modules, viewed as trivial algebras. If M is an R-module, the R-group functor GL(M ) acts naturally on L 2 R(M ). Automorphism invariance behaves nicely with respect to base change and faithfully flat descent. Lemma 5.3. Let B be an R-algebra, β ∈ L 2 βS is Aut(B ⊗R S)-invariant. The converse holds if S/R is faithfully flat. R(B) and S ∈ R-alg. If β is Aut(B)-invariant, then Proof. (I) We begin with a general observation. Let T be an extension of S. The canonical T -linear algebra isomorphism ζ : (B ⊗R S) ⊗S T → B ⊗R T of Lemma 4.2(a) induces a group isomorphism AutT (cid:0)(B ⊗R S) ⊗S T(cid:1) ≃−→ AutT (B ⊗R T ), f 7→ ζ ◦ f ◦ ζ −1 (we view T as an object in R-alg in the obvious way). Since (ζ −1)∗(cid:0)(βS)T(cid:1) = βT by (4.2), it follows that (βS)T is AutT (cid:0)(B ⊗R S) ⊗S T(cid:1)-invariant if and only if βT is AutT (B ⊗R T )-invariant. (II) It is immediate from (I) and the definitions that if β is Aut(B)-invariant then βS is Aut(B ⊗ S)-invariant. (III) Assume now that S/R is faithfully flat and that βS is Aut(B ⊗R S)-invariant. To prove that β is Aut(B)-invariant, let S′ ∈ R-alg and let f ∈ Aut(B)(S′). The extension T = S ⊗R S′ of S′ is faithfully flat. Hence, by Lemma 4.2(c), we have f ∗(βS′) = βS′ as soon as (cid:0)f ∗(βS′ )(cid:1)T = T(cid:0)(B ⊗R S′) ⊗S′ T(cid:1). Note that(cid:0)f ∗(βS′ )(cid:1)T = (f ⊗ IdT )∗(cid:0)(βS′ )T(cid:1) by (4.3). Applying the (βS′ )T ∈ L 2 considerations of (I) to the isomorphism ζ ′ : (B ⊗R S′) ⊗S′ T → B ⊗R T shows that we need to prove that βT is AutT (B ⊗R T )-invariant. But by (I) again this is indeed the case. (cid:3) Theorem 5.4 (Descent of Aut-invariant forms). Assume that we are under the descent setting of (2.9): a is a k-algebra, R ∈ k-alg is flat, and we are given an R-algebra B which is a twisted form of A = a ⊗k R, hence split by some faithfully flat extension of R. Assume that κ ∈ L 2 k (a) is an Aut(a)-invariant bilinear form. (a) There exists a unique R-bilinear form κB ∈ L 2 R(B) such that (κB)S = θ∗(κS) whenever S/R is faithfully flat and θ : B ⊗R S → a ⊗k S is an isomorphism of S-algebras. Moreover, κB is Aut(B)-invariant. (b) If κ is invariant, then so is κB. (c) If a is finitely presented and κ is nondegenerate (resp. nonsingular), then so is κB. Proof. (a) Throughout the proof we fix S and θ as in (a). We let α : R → S be the structure map and κS the base change of κ to S. It will be convenient to first point out a general observation, which is immediate from the definitions. (5.1) If β : S → T is a homomorphism in k-alg, then (Ida ⊗β)∗(κT ) = β ◦ κS. We first show the existence of an R-bilinear form κθ B)S = θ∗(κS). The B indicates that, a priori, the form depends on θ (which in turn involves an S). According and we notation κθ to 2.8 and 2.10 the cocycle corresponding to the S-algebra isomorphism θ−1 is u = θ2θ−1 have 1 B ∈ L 2 R(B) satisfying (κθ (5.2) θ(B ⊗ 1) = {x ∈ a ⊗k S : u(cid:0)(Ida ⊗α1)(x)(cid:1) = (Ida ⊗α2)(x)}. Version September 20, 2018 15 We identify B ⊂ B ⊗R S via b 7→ b ⊗ 1 and claim that the restriction of θ∗(κS) to B × B, which a priori takes values in S, actually takes values in R. In other words, for b, b′ ∈ B and x = θ(b), x′ = θ(b′) ∈ a ⊗k S we claim θ∗(κS)(b, b′) = κS(x, x′) ∈ R. With αi : S → S′′ as before we have, using (5.1), (5.2) and the automorphism-invariance of κS′′, α1(cid:0)θ∗(κS)(b, b′)(cid:1) = α1(cid:0)κS(x, x′)(cid:1) = κS′′(cid:0)(Ida ⊗α1)(x), (Ida ⊗α1)(x′)(cid:1) (5.3) = κS′′(cid:16)u(cid:0)(Ida ⊗α1)(x)(cid:1), u(cid:0)(Ida ⊗α1)(x′)(cid:1)(cid:17) = κS′′(cid:0)(Ida ⊗α2)(x), (Ida ⊗α2)(x′)(cid:1) = α2(cid:0)κS(x, x′)(cid:1) = α2(cid:0)θ∗(κS)(b, b′)(cid:1). This shows that θ∗(κS)(b, b′) belongs to the equalizer of α1 and α2 in S, but these are precisely the B of θ∗(κS) to B is an R-bilinear elements of R (viewed as elements of S). Hence the restriction κθ form on B. Clearly, by its very definition, (κθ B)S = θ∗(κS). We next aim to show that κθ B is independent of the trivialization θ. Thus, let S′/R be another faithfully flat extension, say with structure map α′ : R → S′, and let θ′ : B ⊗R S′ → a ⊗k S′ be an S′-algebra isomorphism. By what we just have shown, we have an R-bilinear form κθ′ B satisfying (κθ′ B . The vehicle to show this will be the algebra T = S ⊗R S′ which we view in an obvious way as an S- and S′-algebra, say with structure maps β and β′ respectively. For simplicity we denote by ρ : k → R the structure map σR,k of R. We have the following commutative diagram: B )S = (θ′)∗(κS), and we claim κθ B = κθ′ k ρ / R α >⑦⑦⑦⑦⑦⑦⑦⑦ ❅❅❅❅❅❅❅❅ α′ S S′ β ❅❅❅❅❅❅❅❅ ?⑦⑦⑦⑦⑦⑦⑦ β′ T We remind the reader that T is faithful flat over S, S′ and R. Let ξ : (a ⊗α◦ρ S)⊗β T → a ⊗β◦α◦ρ T = a ⊗k T be the canonical T -linear algebra isomorphism, and define θ : a ⊗β◦α◦ρ T → B ⊗β◦α T by the composition of maps θ : a ⊗β◦α◦ρ T ξ−1 / (a ⊗α◦ρ S) ⊗β T (θ⊗Id)−1 / (B ⊗α S) ⊗β T ζ / B ⊗β◦α T . Then θ∗ maps (κθ B)β◦α ∈ L 2 T (B ⊗β◦α T ) onto a bilinear form in L 2 θ∗(cid:0)(κθ (4.2) B)β◦α(cid:1) = (ζ ◦ (θ−1 ⊗ Id) ◦ ξ−1)∗ ((κθ = (ξ−1 ∗ ◦ (θ−1 ⊗ Id)∗)(cid:0)((κθ = (cid:0)ξ−1 ∗ ◦ (θ−1 ⊗ Id)∗ ◦ (θ ⊗ Id)∗(cid:1)(cid:0)(κα◦ρ)β(cid:1) = ξ−1 ∗(cid:0)(κα◦ρ)β) B)α)β(cid:1) = (ξ−1 ∗ ◦ (θ−1 ⊗ Id)∗)(cid:0)(θ∗κα◦ρ)β(cid:1) B)β◦α) = (ξ−1 ∗ ◦ (θ−1 ⊗ Id)∗ ◦ ζ ∗) (κθ (4.3) T (a ⊗k T ). In fact, B)β◦α) (4.2) = κβ◦α◦ρ = κT . We define θ′ : a ⊗k T → Bβ′◦α′ T by replacing θ ⊗ Id by θ′ ⊗ Id in the definition of θ above. By symmetry (cid:0)θ′ ∗(cid:1)(κθ′ B )β′◦α′ = κT . Since θ−1 ◦ θ′ ∈ Aut(a)(T ), we get B )β′◦α′ = (θ′)−1 ∗(κT ) = θ−1 ∗(κT ) = (κθ (κθ′ B)β◦α so that Lemma 4.2(c) implies κθ B = κθ′ B . We are now justified to define κB = κθ B. Finally, we use Lemma 5.3 to establish that κB is Aut(B)-invariant: κ is Aut(a)-invariant =⇒ κS is Aut(aS)-invariant =⇒ θ∗(κS) = (κB)S is Aut(B ⊗α S)-invariant =⇒ κB is Aut(B)- invariant. (b) κB is invariant since invariance of bilinear forms is stable under base ring extensions and, by Lemma 4.2(c), also under faithfully flat descent. (c) The argument is analogous to that of (b), using Lemma 4.2(d) and the fact that being finitely presented is a property which is invariant under arbitrary base change and faithfully flat descent. (cid:3) / > ? / / / 16 Version September 20, 2018 Corollary 5.5. Let B be an S/R-form of A = a ⊗k R as in the descent setting 2.9. We further assume that a is finitely presented as a k-module and that κ ∈ IBFk(a) is Aut(a)-invariant. Let κB be the R-bilinear form on B associated to κ as described in Theorem 5.4. If the IBF-principle holds for (a, κ), then the IBF-principle holds for (B, κB) a well. Proof. By faithfully flat descent, the associated map κB : IBFR(B) → R is an isomorphism as soon as the extended map (κB)S has this property. By Lemma 4.6(a) applied to β = κB we have (κB)S = (κB)S ◦ ν. Since ν is an isomorphism, we are reduced to showing that (κB)S is an isomorphism. By the theorem, (κB)S = θ∗(κS). From the definitions, the reader easily checks that θ∗(κS) = κS ◦ IBF(θ). By functoriality IBF(θ) is an isomorphism while κS is an isomorphism by Lemma 4.6(b). Hence (κB)S = κS ◦ IBF(θ) is an isomorphism. This completes the proof. (cid:3) 6. Applications to Lie and other classes of algebras We now look in detail at our general results in some important special cases. Unless stated otherwise, we use our basic setting: k is a commutative associative unital ring and R ∈ k-alg. If f is an endomorphism of a finitely generated and projective R-module, we denote by tr(f ) its trace. For details on this notion, see for example [B:A, II, §4.3] or [KO]. 6.1. Lie Algebras. We start by discussing the Killing form of a Lie algebra L, defined by κ(l1, l2) = tr(cid:0)(ad l1) ◦ (ad l2)(cid:1) for li ∈ L. Proposition 6.1. Let L be a Lie algebra over R whose underlying R-module is finitely generated and projective. (a) The Killing form κ of L is an invariant and Aut(L)-invariant R-bilinear form. For any S ∈ R-alg the base change κS is the Killing form of the Lie algebra L ⊗R S. Proof. (a) The invariance of κ follows from tr(f g) = tr(gf ) for endomorphisms f, g of L. This (b) Suppose f is an α-semilinear automorphism of L for some α ∈ Autk(R). Then κ(cid:0)f (l1), f (l2)(cid:1) = α(cid:0)κ(l1, l2)(cid:1) holds for li ∈ L. identity also implies that κ is AutR(L)-invariant: For f ∈ AutR(L) we have ad(cid:0)f (l)(cid:1) = f (ad l)f −1 and hence κ(cid:0)f (l1), f (l2)(cid:1) = tr(cid:0)f (ad l1)(ad l2)f −1(cid:1) = κ(l1, l2). The adjoint maps of the Lie S-algebra L ⊗R S are obtained by base change from the adjoint maps of L. Since the trace is invariant under base change, the Killing form of the Lie algebra L ⊗R S is the base change of κ by S. It then follows that κ is Aut(L)-invariant. (b) We first prove an auxiliary formula. Namely, assume that g is an α−1-semilinear endomor- phism of L. Then (6.1) tr(f g) = α(cid:0) tr(gf )(cid:1) (observe that both f g and gf are R-linear). By descent it is sufficient to show (6.1) in case L is free of finite rank. Let F and G be matrices representing f and g in some R-basis of L. We denote by α·G the matrix obtained from G by applying α to all its entries. Then f g and gf are represented by F (α · G) and G(α−1 · F ) respectively, whence tr(f g) = tr(cid:0)F (α · G)(cid:1) = tr(cid:0)(α · G)F(cid:1) = tr(cid:0)α · (G(α−1 · F ))(cid:1) = α(cid:0) tr(gf )(cid:1). We can now establish (b): κ(cid:0)f (l1), f (l2)(cid:1) = tr(cid:0)f (ad l1 ad l2 f −1)(cid:1). Applying (6.1) with g = ad l1 ad l2 f −1 shows that tr(cid:0)f (ad l1 ad l2 f −1)(cid:1) = α(cid:0) tr(ad l1 ad l2 f −1f )(cid:1) = α(cid:0)κ(l1, l2)(cid:1). Corollary 6.2. In the descent setting (2.9) suppose that a is a Lie algebra whose underlying k- module is finitely generated and projective. Let κ be the Killing form of a. Then (cid:3) (a) B is a finitely generated projective R-module and the unique R-bilinear form κB on B associ- ated to κ in Theorem 5.4 is the Killing form of the Lie algebra B. If B is realized as an R-subalgebra of a ⊗k S as explained in (2.8),6 the form κB is the restriction of the Killing form κS of a ⊗k S to B. (b) If κ is nonsingular, then the Killing form of B is non-singular. 6Which is always possible up to R-isomorphism -- this is the content of (2.8) Version September 20, 2018 17 (c) If κ is nonsingular and a is a central k-algebra, then B is a central R-algebra, and IBFR(B) is free R-module of rank 1 admitting κB as a basis. In particular IBFR(B) = RκB. Proof. (a) By Proposition 6.1, κS is the Killing form of the S-Lie algebra a ⊗k S. The property of being finitely generated and projective is stable under arbitrary base change and faithfully flat descent. Hence the R-module B is finitely generated and projective. By Proposition 6.1, βS is the Killing form of BS. Let θ : B ⊗R S → a ⊗k S be a trivialization. Since the isomorphism θ preserves Killing forms, we get βS = θ∗(κS). Now (a) follows from the uniqueness assertion in Theorem 5.4. (b) This follows from (a) and Lemma 4.2(d) (we remind the reader that every finitely generated projective module is finitely presented). (c) By [Pi1, Lemma 3.1] the S-algebra a ⊗k S is central. The faithfully flat descent reasoning of [GP, Lemma 4.6(3)] then shows that the R-algebra B is a central. The last claim now follows from Corollary 3.8. (cid:3) At this point a very natural question arises: What are interesting examples of Lie algebras for which the IBF-principle 3.6 holds with respect to the Killing form? To convince the reader that (over rings) one cannot expect easy answers, we will look at one of the simplest and innocent looking Lie algebras. Example 6.3 (sl2(k)). Let sl2(k) be the Lie algebra of all traceless 2 × 2-matrices with entries in our ring k. Its underlying module is free of rank 3, with the following matrices forming the standard basis: f =(cid:18)0 0 1 0(cid:19) . A straightforward calculation shows that ibfk(cid:0)sl2(k)(cid:1) is spanned by 0 −1(cid:19) , h =(cid:18)1 0(cid:19) , e =(cid:18)0 0 1 0 h ⊗ e, e ⊗ h, f ⊗ h, h ⊗ f, 2e ⊗ e, 2f ⊗ f, h ⊗ h − 2f ⊗ e, h ⊗ h − 2e ⊗ f. Hence Consequently: IBFk(cid:0)sl2(k)(cid:1) =(cid:0)(k/2k)e ⊗ e(cid:1) ⊕ (cid:0)(k/2k)f ⊗ f(cid:1) ⊕ Spank{e ⊗ f , f ⊗ e}. • If 2k = 0, then IBFk(cid:0)sl2(k)(cid:1) is free of rank 4, with basis {e ⊗ e, f ⊗ f , e ⊗ f , f ⊗ e}. • If 2 ∈ k×, i.e., 2 ∈ k is invertible, then IBFk(cid:0)sl2(k)(cid:1) is free of rank 1, e.g. with basis {h ⊗ h}. Using the isomorphism (3.6) and the description of IBF(cid:0)sl2(k)(cid:1) above, we can define an invariant bilinear form γ ∈ IBFk(cid:0)sl2(k)(cid:1), sometimes called the normalized Killing form or the normalized invariant form, by γ(e, e) = 0 = γ(f, f ), γ(e, f ) = 1 = γ(f, e). Note that γ(h, h) = 2 and that all other values of γ on the standard basis of sl2(k) are zero, in particular γ is symmetric. The description of IBFk(cid:0)sl2(k)(cid:1) above implies that: If 2 is not a zero divisor in k, then IBFk(cid:0)sl2(k)(cid:1) = kγ is free of rank 1. Moreover, by calculating the discriminant of γ one obtains: γ is nonsingular ⇐⇒ 2 ∈ k× ⇐⇒ the IBF-principle holds for (sl2(k), γ). In this case, γ is Aut(cid:0)sl2(k)(cid:1)-invariant, which can be seen by noting that (ad x)3 − 2γ(x, x) ad x = 0 is the generic minimal polynomial of sl2(k). It is straightforward that 12γ is the Killing form of sl2(k). In particular, the Killing form vanishes if 2k = 0 (not surprising since then sl2(k) is a 2-step nilpotent Lie algebra) or if 3k = 0 (somewhat surprising since sl2(k) is a simple Lie algebra when k is a field of characteristic 3). The conclusion is that for the setting of this paper the normalized Killing form γ is better behaved than the Killing form itself. A case in point is a revised version of Corollary 6.2 for a = sl2(k) and 2 ∈ k×. Since sl2(k) is then central,7 the proof of loc. cit. shows that: • If 2 ∈ k×, any S/R-form B of sl2(R) is central and has a nonsingular invariant bilinear form β (not necessarily the Killing form), for which IBFR(B) = Rβ is free of rank 1. 7Centrality only requires that 2 not be a zero divisor. 18 Version September 20, 2018 It is instructive to summarize what we have shown above for the special case k = Z. • The k-module IBFZ(cid:0)sl2(Z)(cid:1) is neither projective nor cyclic, in particular, the IBF-principle does not hold for sl2(Z). Yet IBFZ(cid:0)sl2(Z)(cid:1) = Zγ is free of rank 1. • All invariant bilinear forms of sl2(Z) are symmetric (even though sl2(Z) is not perfect, cf. Remark 3.4). • All non-zero invariant bilinear forms of sl2(Z) are nondegenerate, but none of them is non- singular. Remark 6.4 (Generalizations of Example 6.3). We note that Example 6.3 can be generalized by replacing sl2(k) = sl2(Z) ⊗Z k by G ⊗Z k where G is the Lie algebra of a split simple simply- connected Chevalley-Demazure group scheme. In terms of Lie algebras, G is a Chevalley order of a split simple Lie algebra (g, h) over Q, say with root system ∆, which is compatible with the root space decomposition of (g, h) and satisfies G ∩ h = SpanZ{hα : α ∈ ∆} (with the standard notation). In this setting the existence of an invariant bilinear form γ as above follows from [SS, GN]. It is uniquely determined by the condition γ(hα, hα) = 2 for any long root α. Details will be left to the reader. In what follows we restrict our presentation to base fields of characteristic 0. To abide by standard notation we denote our algebra a, which is now a finite-dimensional semisimple Lie algebra defined over a field k of characteristic 0, by g. We are interested in twisted forms of g ⊗k R for some R ∈ k-alg. In the case when k is algebraically closed, g is simple and R is the Laurent polynomial ring k[t±1 n ], the twisted forms in question are related to the affine Kac-Moody Lie algebras (the case n = 1) and more generally to multiloop algebras (see [ABFP, GP, Ne4, Pi1] and §7 for further details and references). 1 , . . . , t±1 Theorem 6.5. Let g be a finite-dimensional semisimple Lie algebra over a field k of characteristic 0. Let B be a twisted form of g ⊗k R, split by a faithfully flat extension S/R. (a) Then B is a finitely generated projective R-module and perfect as a Lie algebra. The Killing form of the R-algebra B coincides with the bilinear form κB associated to the Killing form κ of g in Theorem 5.4. In particular the Killing form of B is nonsingular and Aut(B)-invariant. If B is realized as an R-subalgebra of g ⊗k S, the Killing form of B is the restriction of the Killing form κg⊗S to B. (b) Assume henceforth that g is central, hence central-simple. Then B is a central R-algebra, IBFR(B) is a free R-module of rank 1 admitting κB as a basis, and (B, κB) satisfies the IBF- principle 3.6. Proof. With the exception of the perfectness statement, (a) and the first part of (b) is a re-statement of Corollary 6.2. But being perfect is a property (of arbitrary algebras) which is stable under arbitrary base change and faithfully flat descent. Since B ⊗R S ≃ g ⊗k S and the latter is perfect, B is perfect. That (B, κB) satisfies the IBF-principle is a special case of Corollary 5.5 keeping in mind that (g, κ) satisfies the IBF-principle in view of Proposition 3.9. (cid:3) Remark 6.6. In the special case when R = S and B = g ⊗k R, our result says that (6.2) IBF(g ⊗k R) ≃ R ≃ IBFk(g) ⊗k R, This formula is also a special case of [Zu, Th. 4.1] which, using methods from Lie algebra homology, determines the "predual" of the space of symmetric bilinear forms of a Lie algebra of type L ⊗k R for k a field of characteristic 6= 2 and L any Lie algebra over k. Assuming for comparison reasons that L is perfect, we know (Remark 3.4) that all invariant bilinear maps are symmetric and thus [Zu, Th. 4.1] becomes (6.2) for Lie algebras of type L ⊗k R. We emphasize that the approach of [Zu] cannot be applied to the case of twisted forms of g ⊗k R. Developing methods that would apply to these algebras was the original motivation for our work. As already observed, such twisted algebras already arise in the affine Kac-Moody setting and are crucial for EALA theory. An immediate consequence of (6.2) is that every invariant bilinear form β ∈ IBF(R,k)(g ⊗k R) = IBFk(g ⊗k R) has the form ϕ ◦ κR for a unique ϕ ∈ R∗ = Homk(R, k), i.e., β(x1 ⊗ r1, x2 ⊗ r2) = Version September 20, 2018 19 κ(x1, x2) ϕ(r1r2) for xi ∈ g and ri ∈ R. This latter fact has recently been re-proven in [MSZ, Lemma 2.3] in case k is an algebraically closed field of characteristic 0, using the structure theory of g. The untwisted case was, out of necessity, the first objective of our work. The methods to be developed, however, had to be compatible with descent theory so that results about twisted algebras could be obtained. In retrospect, we "knew" that the functor on k-spaces IBFk(g ⊗k S, −) is repre- sented by S (one can reinterpret [Zu] or [MSZ] this way). But how does one recover S from g ⊗k S? The answer is as its centroid. By descent, the centroid of B is in this case naturally isomorphic to R. Our result shows that the representability of IBFk(B, −) in terms of the centroid is indeed the correct point of view. Remark 6.7. For crucial use in [PPS], we note the following. Since the IBF-principle holds by Theorem 6.5(b), composing the isomorphism (3.9) with the inverse of the isomorphism given in Lemma 3.7, we have an isomorphism Homk(R, V ) → CtdR(cid:0)B, Homk(B, V )(cid:1), ϕ 7→ eϕ such that eϕ(b)(b′) = ϕ(cid:0)κB(b, b′)(cid:1). 6.2. Unital algebras. In this subsection we will discuss invariant bilinear forms of unital algebras B defined over some R ∈ k-alg. To do so, we will use the associator module and commutator module defined for an arbitrary algebra B by (B, B, B) = SpanZ{(a, b, c, ) : a, b, c ∈ B} and [B, B] = SpanZ{[a, b] : a, b ∈ B} respectively, where (a, b, c) = (ab)c − a(bc) is the associator and [a, b] = ab − ba is the commutator in B.8 It is immediate that (B, B, B) and [B, B] are R-submodules of B. We define ac(B) = (B, B, B) + [B, B] and AC(B) = B/ac(B). Let 1B ∈ B be the identity element of B. Thus b 1B = b = 1B b for all b ∈ B. A unital algebra is perfect, whence IBF(R,k)(B; V ) = IBFk(B; V ) for any k-module V by Remark 3.4. For convenience for the remainder of this section we will denote the identity element of B by 1. Lemma 6.8. Let B be a unital R-algebra. Then the multiplication map µ : B ⊗R B → B, µ(a ⊗ b) = ab, induces an isomorphism with inverse given by ¯a 7→ 1 ⊗ a = a ⊗ 1. Hence, for any k-module V the natural map ¯µ : IBFR(B) → AC(B), ¯µ(a ⊗ b) = ab Homk(AC(B), V ) → IBFk(B; V ), which assigns to ϕ ∈ Homk(AC(B), V ) the bilinear function (a, b) 7→ ϕ(ab), is an isomorphism of R-modules. Its inverse is given by assigning to β the linear function ¯b 7→ β(b, 1), where b ∈ B. Proof. By (3.3), ibf R(B) is spanned by elements of the form ab ⊗ c − a ⊗ bc and a ⊗ b − b ⊗ a. It is clear that µ(cid:0)ibf R(B)(cid:1) = ac(B). Hence ¯µ is well-defined and surjective. Let ν : B → B ⊗k B be defined by ν(a) = 1 ⊗ a. Then we have ν(cid:0)(a, b, c)(cid:1) = 1 ⊗ (ab)c − 1 ⊗ a(bc) ≡ ab ⊗ c − a ⊗ bc ≡ 0 mod ibf R(B), ν(cid:0)[a, b](cid:1) = 1 ⊗ ab − 1 ⊗ ba ≡ a ⊗ b − b ⊗ a ≡ 0 mod ibf R(B). and We thus get a well-defined k-linear map ¯ν : AC(B) → IBFR(B) satisfying ¯ν(¯b) = b ⊗ 1 = 1 ⊗ b. Because of (ν ◦ µ)(a ⊗ b) = 1 ⊗ ab = (1 ⊗ ab − 1a ⊗ b) + a ⊗ b ≡ a ⊗ b mod ibf R(B), we have ¯ν◦ ¯µ = IdIBF(B), proving injectivity and thus bijectivity of ¯µ. Under the isomorphism ¯µ, the universal bilinear map βuni : B × B → IBFR(B) becomes βuni,u : B × B → AC(B) with βuni,u(a, b) = ab. In view of (3.7) this implies the last claim. (cid:3) 8If B happens to be a Lie algebra, the commutator as defined here is twice the Lie algebra product. This notational conflict should not cause any problems since in the following we will employ the notation [a, b] for non-Lie algebras only. 20 Version September 20, 2018 Corollary 6.9. Let B be a unital R-algebra and assume that B = Rb0 ⊕ ac(B) for some b0 ∈ B where Rb0 is free with basis {b0}. Let π : B → R be defined by b = π(b)b0 ⊕ bac where bac ∈ ac(B), and define Then β0 ∈ IBFR(B), and (B, β0) satisfies the IBF-principle. Furthermore β0 : B × B → R, β0(a, b) = π(ab). β0 : AC(B) → R, ¯b 7→ β0(b, 1) is a well-defined R-module isomorphism. Proof. The proof is straightforward. (cid:3) 6.3. Azumaya algebras. This subsection fits within the descent setting (2.9): k is base ring, R ∈ k-alg is flat as a k-module (for example k = R), and S ∈ R-alg is a faithfully flat R-module. We let a = Mn(k). Then our S/R-form B of Mn(k) ⊗k R = Mn(R) is an Azumaya algebra over R of constant rank n2. We start by recording some facts about a. As a unital algebra, a is perfect. It is also well-known that a is central. Moreover, a has a natural invariant bilinear form, the trace form κ defined by κ(x, y) = tr(xy) where this last is the usual trace of the matrix xy. It is easy to see (using the standard dual basis of the elementary matrices Eij ) that κ is nonsingular. Moreover, κ is Aut(a)-invariant. Indeed, since a ⊗k K = Mn(K) for any K ∈ k-alg, it suffices to verify that κ is automorphism-invariant. Thus let σ ∈ Autk(a) and x, y ∈ a. To show that xy and σ(x)σ(y) have the same trace, it is of Mn(kp) have = σp(xp)σp(yp) where σp = σ ⊗ Idkp. We may therefore assume that k is a local ring. But then, by the Skolem-Noether Theorem for local rings ([KO, IV, Cor. 1.3]), σ is given by conjugation by an invertible matrix M ∈ GLn(k), whence σ(x)σ(y) = M xyM −1, and so clearly xy and σ(x)σ(y) have the same trace. We also have enough to prove that for all p ∈ Spec(k) the two elements (xy)p and (cid:0)σ(x)σ(y)(cid:1)p the same trace. Clearly (xy)p = xpyp and (cid:0)σ(x)σ(y)(cid:1)p a = kE11 ⊕ [a, a], [a, a] = {x ∈ a : tr(x) = 0} = ac(a) since any x =Pi,j xij Eij can be uniquely written as (6.3) x =(cid:0)x11 +P1<i xii(cid:1)E11 +P1<i xii(Eii − E11) +Pi6=j xij Eij and [a, a] is spanned by matrices of type [Eii, Eij ] = Eij and [Eij , Eji] = Eii − Ejj = (Eii − E11) − (Ejj −E11) for i 6= j. Formula (6.3) implies that the trace form κ is the bilinear form of Corollary 6.9. Hence (a, κ) satisfies the IBF-principle. Theorem 6.10. Let B be an S/R-form of Mn(R) and let κB be the bilinear form associated to the trace form κ of Mn(k) in Theorem 5.4. (a) Then κB is a nonsingular, invariant and Aut(B)-invariant bilinear form and a basis of IBFR(B). (b) IBFR(B) ≃ AC(B) ≃ R, and the map IBFR(B) 7→ R, b ⊗ b′ 7→ κB(b, b′) is an isomorphism of R-modules. Hence (B, κB) satisfies the IBF-principle 3.6. (c) If B is realized as an R-subalgebra of Mn(S), see Remark 2.10, κB coincides with the restriction of the trace form of Mn(S) to B. Proof. (a) and (b) follow from Theorem 5.4, Corollary 5.5 and Corollary 6.9. For the proof of (c) one uses the automorphism invariance of the trace of Mn(S′′) and the reasoning in (5.3) to conclude that the restriction λ of the trace form of Mn(S) to B has values in R. Since λS is the trace form of Mn(S) and thus coincides with (κB)S, we get λ = κB from uniqueness in Theorem 5.4 (or from Lemma 4.2(c)). (cid:3) Remark 6.11. The form κB is, by definition, nothing but the reduced trace form of the Azumaya algebra B as defined in [KO]. This proves (without the construction of the characteristic polynomial as done in [KO]) that the reduced trace form, which a priori takes values in S, does take values in R. Version September 20, 2018 21 Corollary 6.12. Every Azumaya algebra B over R has a nonsingular, invariant and Aut(B)- invariant bilinear form κB such that (B, κB) satisfies the IBF-principle 3.6. In particular, IBFR(B) is a free R-module with basis {κB}. Proof. If B has constant rank, then B is an S/R-form of some Mn(R) and the results follows from Theorem 6.10. In general, we can decompose the identity element 1R of R into a sum 1R = e1+· · ·+es of orthogonal idempotents ei ∈ R such that B = B1 ⊞ · · · ⊞ Bs is a direct product of ideals Bi = eiB, each Bi is an Azumaya algebra of constant rank ρi over Ri = eiR, and ρi 6= ρj for i 6= j (if ρi = ρj then we replace ei, ej by ei + ej). We then define κB as the orthogonal sum of the forms κBi i=1 HomRi(Bi, Ri) and the nonsingularity of the κBi. Finally, Aut(B)-invariance holds since the decomposition B = B1 ⊞ · · · ⊞ Bs is preserved under base ring extensions and automorphisms. (cid:3) constructed previously. Nonsingularity follows from HomR(B1 ⊞ · · · ⊞ Bs, R) ≃Ls 6.4. Octonion algebras. As in the previous subsection, k here is an arbitrary base ring and R ∈ k-alg. Following [Bi2, LPR, Pe] we call an algebra B over R an octonion algebra if its underlying R-module is projective of constant rank 8, contains an identity element 1B, and admits a quadratic form nB : B → R, the norm of B, satisfying the following two conditions. (i) The associated bilinear form nB : B × B → R, nB(a, b) = nB(a + b) − nB(a) − nB(b), is nonsingular, and (ii) nB(ab) = nB(a) nB(b) holds for all a, b ∈ B. For an octonion algebra B the linear form tB = nB(1B, −) is called the trace of B. An example of an octonion algebra is the algebra Zor(R) of Zorn vector matrices, defined on the R-module with product x α2(cid:21) (cid:20)β1 (cid:20)α1 u y Z = Zor(R) =(cid:20) R R3 R3 R(cid:21) β2(cid:21) =(cid:20) β1x + α2y + u × v α1β1 − tuy v α1v + β2u + x × y −txv + α2β2 (cid:21) for αi, βi ∈ R and u, v, x, y ∈ R3. Here tuy and x × y are the usual scalar and vector product of vectors in R3. For this octonion algebra and a = [ α1 u x α2 ] ∈ Z one has 1Z =(cid:20)1 0 0 1(cid:21) , nZ(a) = α1α2 + tux, trZ(a) = α1 + α2. For our approach to octonion algebras it is important that an R-algebra B is an octonion algebra if and only if there exists a faithfully flat (even faithfully flat and ´etale) S ∈ R-alg such that B ⊗R S ≃ Zor(S) ([LPR, Cor. 4.11]). The algebra Zor(R) is referred to as split octonions. Theorem 6.13. Let B be an octonion algebra over R. (a) B is a central R-algebra satisfying ac(B) = (B, B, B) = [B, B]. (b) The bilinear form τ : B × B → R, defined by τ (x, y) = tr(xy), is an invariant, nonsingular and Aut(B)-invariant R-bilinear form. It coincides with the form τB associated in Theorem 5.4 to the bilinear form τ of Zor(R) with respect to any splitting B ⊗R S ≃ Zor(S) of B. (c) (B, τ ) satisfies the IBF-principle. Proof. The algebra B fits into our descent setting with k = R and a = Zor(R). We first prove all assertions for a. Straightforward calculations (admittedly tedious in the case of the associator module) show that (i) [a, a] = (a, a, a) = ac(a), a = Re ⊕ ac(a) for e = [ 1 0 (ii) For a, b as in the product formula above we have τ (a, b) = α1β1 + α2β2 − tuy − txv. Hence 0 0 ], a is central. τ is nonsingular and symmetric. (iii) τ(cid:0)ac(a)(cid:1) = 0, τ (e) = 1, whence τ is the bilinear form associated to the decomposition a = Re ⊕ ac(a) in Corollary 6.9. 22 Version September 20, 2018 This corollary now implies that (a, τ ) satisfies the IBF-principle. To establish that τ is Aut(a)- invariant, we recall that since a is a quadratic algebra the trace linear form tr of a is uniquely determined by the unital algebra a ([Pe, Lem. 1.2]). For any extension K ∈ k-alg, the base change τK is therefore the bilinear form τa⊗K of the K-algebra a ⊗R K = Zor(K). Uniqueness of the trace then implies that trK is AutK(aK)-invariant. We now consider an arbitrary octonion algebra B over R and choose a faithfully flat extension S ∈ R-alg such that B⊗RS ≃ Zor(S) as S-algebras. In the first part of the proof we have established all claims for a = Zor(R), whence also for Zor(S). The assertions in (a) now hold for B since they are all preserved by faithfully flat descent. In (b) it suffices to establish the second part, but this follows from the fact that the base change of the trace form τ of B to S is the trace form of Zor(S). (c) is a special case of Corollary 5.5. (cid:3) Remark 6.14 (Quadratic algebras). The experts will undoubtedly have noticed that the automor- phism-invariance of the bilinear form κB comes from the fact that octonions are quadratic algebras, see e.g. [Pe, 1.1]. Hence our techniques can also be applied to certain quadratic algebras whose trace forms are invariant. 6.5. Alternative algebras. We consider alternative algebras, always assumed to be unital, over some base ring R. Recall ([Bi2]) that an alternative algebra B is called separable if for every algebraically closed field K in R-alg the K-algebra B ⊗R K is finite-dimensional and a direct sum of simple ideals. Equivalently, the unital universal multiplication envelope of B is a separable associative algebra. By [Bi2, Prop. 2.11], an R-algebra B is central separable and alternative if and only if R = R1 ⊞ R2 is a direct sum of two ideals such that B1 = R1B is an Azumaya algebra over R1 and B2 = R2B is an octonion algebra over R2. It is now straightforward to extend the results of §6.3 and §6.4 to central separable alternative algebras. We leave the details to the reader and only mention the following. Corollary 6.15. A central separable alternative B over R has a nonsingular invariant bilinear form κB such that IBFR(B) is a free R-module with basis {κB}. 6.6. Jordan algebras. Central separable Jordan algebras over rings R containing 1 are another class of algebras to which our results apply. 2 ([Bi1, Lo2]) By definition, a unital Jordan algebra J over R is separable if and only if J ⊗R K is finite- dimensional semisimple for all fields K ∈ R-alg. A central separable Jordan algebra J is generically algebraic ([Lo3, Ex. 2.4(d)]). Let tr ∈ HomR(J, R) be its generic trace. By Prop. 2.7 of loc. cit. the associated bilinear form τ , defined by τ (a, b) = tr(ab), is invariant, Aut(J)-invariant and commutes with extensions and faithfully flat descent. Since separability and being generically algebraic is invariant under base ring extensions, τ is in fact Aut(J)-invariant. By [Lo1, Cor. 16.16], τ is nondegenerate for separable Jordan algebras over fields. From Lemma 6.16 we then get that τ is nonsingular. Lemma 6.16. Let M be a finitely generated projective R-module and let β ∈ L 2 nonsingular if and only if βK is nondegenerate for all K ∈ k-alg which are fields. R(M ). Then β is Proof. This is an application of [B:AC, II, §3.3 Th. 1, §3.2 Cor. de la Prop. 6 and §5.3 Th. 2]. (cid:3) Thus, in view of Corollary 3.8 we have the following. Theorem 6.17. The generic trace form τ of a central separable Jordan algebra J over a ring R containing 1 2 is an invariant nonsingular and Aut(J)-invariant bilinear form, and IBFR(J) is a free R-module admitting {τ } as a basis. We leave it to the interested reader to look into the following possible improvement of this last result. Question 6.18. For J as in Proposition 6.17, is the R-module IBFR(J) projective? Since J is central and therefore a faithful R-module, we have κJ (J, J) = R by [DI, I, Cor. 1.10]. Hence, if the question has a positive answer, Proposition 3.9 applies and yields that (J, τ ) satisfies the IBF-principle. In particular, Theorem 6.17 then becomes a corollary. Version September 20, 2018 23 7. Graded invariant bilinear forms In this section we classify graded invariant bilinear forms, which are particularly important for infinite-dimensional Lie theory . We will therefore concentrate on these (except for preliminaries considerations), and leave the extension to other classes of algebras to the reader. We begin with some generalities about gradings and graded forms. Unless specified otherwise, we continue with our standard setting: k is a base ring and B is an arbitrary R-algebra for some R ∈ k-alg. Throughout, Λ is an abelian group. Definition 7.1 (Graded algebras and graded invariant bilinear forms). A Λ-graded algebra is a pair (C, C ) consisting of a k-algebra C and a family C = (C λ)λ∈Λ of k-submodules C λ of C satisfying C = Lλ∈Λ C λ and C λC µ ⊂ C λ+µ for all λ, µ ∈ Λ. We will say that a k-algebra C is Λ-graded if (C, C ) is a Λ-graded algebra for some family C . We point out that it is allowed that some of the homogeneous submodules C λ vanish. If C is a unital algebra, then necessarily 1C ∈ C0. Assume C is Λ-graded. We call κ ∈ L 2 k (C) a graded bilinear form if κ(C λ, C µ) = 0 whenever λ + µ 6= 0. Definition 7.2 (Graded S/R-forms of algebras). In the following we assume that R ∈ k-alg is Λ-graded, say R =Lλ∈Λ Rλ. An R-algebra B is then called a Λ-graded R-algebra if B =Lλ∈Λ Bλ is Λ-graded as a k-algebra and the Λ-gradings of R and B are compatible in the sense that RλBµ ⊂ Bλ+µ for all λ, µ ∈ Λ. For example, for any k-algebra a the R-algebra a ⊗k R is canonically a Λ-graded R-algebra by defining the λ-homogeneous submodule (a ⊗k R)λ = a ⊗k Rλ. In the descent setting (2.9) we suppose that R ∈ k-alg is Λ-graded and that S ∈ R-alg is a Λ- graded R-algebra. We view a ⊗k S with its canonical Λ-grading. An S/R-form B of a ⊗k R is called graded if B is a Λ-graded R-algebra and there exists an S-algebra isomorphism θ : B ⊗R S → a ⊗k S which respects the gradings: θ(bλ ⊗ sµ) ∈ a ⊗k Sλ+µ for bλ ∈ Bλ and sµ ∈ Sµ. We now specialize to Lie algebras and derive a graded version of Theorem 6.5. Following the notation used in loc. cit. we change a to g. Proposition 7.3. Let g be a finite-dimensional semisimple Lie algebra over a field k of characteristic 0 with Killing form κ. Assume that R ∈ k-alg and S ∈ R-alg are Λ-graded, S/R is faithfully flat and B is a graded S/R-form of g ⊗k R. (a) If κB is the form attached to κ in Theorem 5.4, then κB is the Killing form of the R-algebra B and satisfies κB(Bλ, Bµ) ⊂ Rλ+µ for all λ, µ ∈ Λ. (b) If g is central (and hence simple), every graded invariant bilinear form β ∈ IBFk(B) can be written in the form ϕ ◦ κB for a unique ϕ ∈ {ϕ ∈ R∗ : ϕ(Rλ) = 0 for λ 6= 0} ≃ (R0)∗. Proof. (a) That κB is the Killing form of B, was established in Corollary 6.2. Since there exists an S-algebra isomorphism θ : B ⊗R S → g ⊗k S respecting the gradings, there is no harm to assume that B ⊂ g ⊗k S with Bλ = Bλ ⊗ 1S ⊂ g ⊗ Sλ. Recall from Corollary 6.2 that κB = κg⊗kS B × B where κg⊗kS is the Killing form of the S-algebra g ⊗k S, and that κg⊗kS coincides with the base change κS of κ by S. Because κS(cid:0)g ⊗k Sλ, g ⊗k Sµ) ⊂ Sλ+µ, it suffices to show that Rλ = Sλ ∩ R for all λ ∈ Λ. To see this last statement, we use that 1S ∈ S0, so that Rλ = Rλ1S ∈ RλS0 ⊂ Sλ. Then Rλ ⊂ Sλ ∩ R follows. The other inclusion is immediate. (b) We have seen in Theorem 6.5(b) that (B, κB) satisfies the IBF-principle. It follows that R = SpanZ{κB(b1, b2) : bi ∈ B}. Let now β ∈ IBFk(B) be a graded invariant bilinear form. Again by Theorem 6.5 there exists a unique ϕ ∈ R∗ such that β = ϕ ◦ κB. We claim ϕ(r) = 0 for any r ∈ Rλ, λ 6= 0. Indeed, there exist finitely many bi ∈ Bµi and b′ i). i) = 0. That, conversely, every ϕ ∈ (R0)∗ gives rise to a graded invariant (cid:3) i ∈ Bλ−µi such that r =Pi κB(bi, b′ Hence ϕ(r) = Pi β(bi, b′ bilinear form, is of course obvious. Proposition 7.3 can be applied to multiloop algebras based on simple finite-dimensional Lie al- gebras. We recall their definition: g is a finite-dimensional simple Lie algebra over an algebraically closed field k of characteristic 0, and σ = (σ1, . . . , σN ) is a family of commuting automorphisms of 24 Version September 20, 2018 g, which have finite orders m1, . . . , mN respectively. We fix a set of primitive m-th roots of unity ζm ∈ k which are compatible in the sense that ζ l ln = ζn, and put R = k[t±1 1 , . . . , t±1 N ] ⊂ S = k[t ± 1 m1 1 , . . . , t ± 1 mN N ]. The multiloop algebra L = L(g, σ) associated to these data is the Lie algebra L =Li1,...,iN ∈ZN gi1,...,iN ⊗k t i1 m1 1 . . . t iN mN N where gi1,...,iN = {x ∈ g : σj(x) = ζ ij mj x for all j}. Since gi1,...,iN = gi1+k1,...,iN +kN for (k1, . . . , kN ) ∈ m1Z ⊕ · · · ⊕ mN Z, it is clear that L is an R-Lie algebra. It is in fact an S/R-form of g ⊗k R ([ABP, Th. 3.6]). To enter the grading into the picture, we let Λ = 1 m1 Z. Then Λ ≃ ZN and we have a natural Λ-grading on S and R (the reader will note that the homogeneous elements of R have degrees in Z × · · · × Z ⊂ Λ). The Lie algebra g ⊗k S is naturally Λ-graded and this makes L also naturally into a Λ-graded Lie algebra. It is immediate from the definitions that L is R-graded and a graded S/R-form of g ⊗k S. Since in our situation R0 = k, Proposition 7.3 yields the first part of the following. Z × · · · × 1 mN Corollary 7.4. Let L be a multiloop Lie algebra based on a simple finite-dimensional Lie algebra over an algebraically closed field k if characteristic 0. Then, up to scalars in k, the ZN -graded Lie algebra L has a unique graded invariant k-bilinear form β. It is given by β(x ⊗ t j1 m1 1 . . . t jN mN N , y ⊗ t l1 m1 1 . . . t lN mN N ) = κ(x, y)δj1+l1,0 . . . δjN +lN ,0 where κ is the Killing form of g. The form β is nondegenerate. Every k-linear automorphism of L is orthogonal with respect to β. Proof. The nondegeneracy of κ implies that β Lλ × L−λ is a nondegenerate pairing for all λ, which in turn forces β to be nondegenerate. Let now f ∈ Autk(L). The map χ 7→ f ◦ χ ◦ f −1 is an automorphism of the centroid of L. Since L is central, f is α-semilinear for some k-linear automorphism α of R. By Proposition 6.1(b), we then know κL ◦ (f × f ) = α ◦ κL for κL the Killing form of the R-algebra L. The form β is obtained by composing κL with the canonical projection ǫ : R → R0 = k. It is therefore enough to show ǫ = ǫ ◦ α. But this is indeed the case: Every k-linear automorphism of R fixes R0 pointwise and permutes the Rλ, λ 6= 0. To see this, we realize Autk(R) as GLN (Z) ⋉ (k×)N in the natural fashion. (cid:3) This corollary is of interest for the construction of extended affine Lie algebras based on centreless Lie tori, which heavily depends on the existence of a graded invariant nondegenerate bilinear form on a Lie torus. The reader is referred to [AABGP, Ne2, Ne3, Ne4] for background material on extended affine Lie algebras and Lie tori. More precisely, we will deal here with Lie-ZN -tori. Corollary 7.5. Up to scalars, a centreless Lie-ZN -torus has a unique graded invariant nondegen- erate bilinear form. Proof. By [Ne1] a centreless Lie-ZN -torus is either finitely generated over its centroid or is a Lie torus with a root system of type A. Concerning the latter type, one knows from [BGK, BGKN, Yo1] that they are graded isomorphic to sln(kq) for a quantum torus kq, and for these types of Lie algebras the claim follows from [Ne4, 7.10]. If L is a centreless Lie-ZN -torus which is finitely generated over its centroid, then [ABFP, Th. 3.3.1] says that L is graded-isomorphic to a multiloop algebra so that we can apply the previous Corollary 7.4. (cid:3) Corollary 7.5 has been proven in [Yo2, Th. 7.1] for Lie tori graded by a torsion-free group Λ, using the structure theory of these types of Lie tori. Version September 20, 2018 25 References [AABGP] B. Allison, S. Azam, S. Berman, Y. Gao and A. Pianzola, Extended affine Lie algebras and their root systems, Mem. Amer. Math. Soc. 126 #603 (1997). [ABFP] B. Allison, S. Berman, J. Faulkner, A. Pianzola, Multiloop realization of extended affine Lie algebras and [ABP] [BGK] [BGKN] [Bi1] [Bi2] [B:A] [B:AC] [DI] [GP] [GN] [KO] [Lo1] [Lo2] [Lo3] [LPR] [MSZ] [Ne1] [Ne2] [Ne3] [Ne4] [NP] [Pe] [Pi1] [Pi2] [PPS] [SGA1] [SS] [Wa] [Yo1] [Yo2] [Zu] Lie tori, Trans. Amer. Math. Soc. 361 (2009), 4807 -- 4842. B. Allison, S. Berman, and A. Pianzola, Covering algebras. II. Isomorphism of loop algebras, J. Reine Angew. Math. 571 (2004), 39 -- 71. 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. R. Bix, Separable Jordan algebras over commutative rings I, J. Algebra 57 (1979), 111 -- 143. Separable alternative algebras over commutative rings, J. Algebra, 92 (1985), 81 -- 103. N. Bourbaki, Alg`ebre, Chapitres 1 -- 3, Diffusion C.C.L.S., Paris 1970. , Alg`ebre commutative. Chapitres 1 `a 4. Masson Paris, 1985. F. DeMeyer and E. Ingraham, Separable algebras over commutative rings, Lecture Notes in Mathematics, Vol. 181, Springer-Verlag Berlin 1971. P. Gille and A. Pianzola, Galois cohomology and forms of algebras over Laurent polynomial rings, Math. Ann. 338 (2) (2007) 497 -- 543. B. Gross and G. Nebe, Globally maximal arithmetic groups, J. Algebra 272 (2004), 625 -- 642. M.-A. Knus, and M. Ojanguren, Th´eorie de la descente et alg`ebres d'Azumaya, Lecture Notes in Mathe- matics 389, Springer-Verlag Berlin, 1974, iv+163 pp. O. Loos, Jordan pairs, Lecture Notes in Mathematics 460, Springer-Verlag Berlin, 1975, xvi+218 pp. , Separable Jordan pairs over commutative rings, Math. Ann. 233 (1978), 137 -- 144. , Generically algebraic Jordan algebras over commutative rings, J. Algebra 297 (2006), 474 -- 529. O. Loos, H. Petersson, and M. Racine, Inner derivations of alternative algebras over commutative rings, Algebra Number Theory 2 (2008), 927 -- 968. F. Montaner, A. Stolin, and E. Zelmanov, Classification of Lie bialgebras over current algebras, Selecta Math. (N.S.) 16 (2010), 935 -- 962. E. Neher, Lie tori, C. R. Math. Acad. Sci. Soc. R. Can. 26 (2004), 84 -- 89. , Extended affine Lie algebras, C. R. Math. Acad. Sci. Soc. R. Can. 26 (2004), 90 -- 96. , Extended affine Lie Algebras -- An Introductrion to Their Structure Theory, in: Geometric repre- sentation theory and extended affine Lie algebras, Fields Inst. Commun. 59 (2011), 107 -- 167, Amer. Math. Soc., Providence, RI. , Extended affine Lie algebras and other generalizations of affine Lie algebras -- a survey, in: Devel- opments and trends in infinite-dimensional Lie theory, Progr. Math. 288, 53 -- 126, Birkhauser Boston Inc., Boston, MA (2011). E. Neher and A. Pianzola, ´Etale Descent of Derivations, Transformations Groups, 18 (2013), pages 1189 -- 1205. H. Petersson, Composition algebras over algebraic curves of genus zero, Trans. Amer. Math. Soc. 337 (1993), 473 -- 493. A. Pianzola, Derivations of certain algebras defined by ´etale descent. Math. Z. 264 (2010), 485 -- 495. , Vanishing of H 1 for Dedekind rings and applications to loop algebras, C. R. Acad. Sci. Paris 340 (2005) 633 -- 638. A. Pianzola, D. Prelat, and C. Sepp, Variations on themes of C. Kassel and R. Wilson, in preparation. S´eminaire de G´eom´etrie alg´ebrique de l'I.H.E.S., Revetements ´etales et groupe fondamental, dirig´e par A. Grothendieck, Lecture Notes in Math. 224. Springer (1971). T. A. Springer and R. Steinberg, Conjugacy classes, in: Seminar on Algebraic Groups and Related Finite Groups, The Institute for Advanced Study, Princeton, NJ, 1968/69, Lecture Notes in Math. 131, Springer- Verlag, 1970, pp. 167266. W.C. Waterhouse, Introduction to Affine Group Schemes, Graduate Texts in Mathematics, vol. 66.Springer, New York (1979) Y. Yoshii, Coordinate Algebras of Extended Affine Lie Algebras of Type A1, J. Algebra 234 (2000), 128 -- 168. , Lie tori -- a simple characterization of extended affine Lie algebras, Publ. Res. Inst. Math. Sci. 42 (2006), 739 -- 762. P. Zusmanovich, The second homology group of current Lie algebras, K-theory (Strasbourg, 1992), Ast´erisque 226 (1994), 435 -- 452. 26 Version September 20, 2018 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, and Centro de Altos Estudios en Ciencias Exactas, Avenida de Mayo 866, (1084) Buenos Aires, Argentina. E-mail address: [email protected] Centro de Altos Estudios en Ciencias Exactas, Avenida de Mayo 866, (1084) Buenos Aires, Argentina. E-mail address: [email protected] Centro de Altos Estudios en Ciencias Exactas, Avenida de Mayo 866, (1084) Buenos Aires, Argentina. E-mail address: [email protected]
0910.5409	3	0910	2010-09-26T19:05:43	Galois connection for sets of operations closed under permutation, cylindrification and composition	[ "math.RA" ]	We consider sets of operations on a set A that are closed under permutation of variables, addition of dummy variables and composition. We describe these closed sets in terms of a Galois connection between operations and systems of pointed multisets, and we also describe the closed sets of the dual objects by means of necessary and sufficient closure conditions. Moreover, we show that the corresponding closure systems are uncountable for every A with at least two elements.	math.RA	math	GALOIS CONNECTION FOR SETS OF OPERATIONS CLOSED UNDER PERMUTATION, CYLINDRIFICATION AND COMPOSITION MIGUEL COUCEIRO AND ERKKO LEHTONEN Abstract. We consider sets of operations on a set A that are closed under permutation of variables, addition of dummy variables and composition. We describe these closed sets in terms of a Galois connection between operations and systems of pointed multisets, and we also describe the closed sets of the dual objects by means of necessary and sufficient closure conditions. Moreover, we show that the corresponding closure systems are uncountable for every A with at least two elements. 1. Preliminaries Throughout this paper, let A be an arbitrary nonempty set. An operation on A is a map f : An → A for some integer n ≥ 1, called the arity of f . For n ≥ 1, the set of all n-ary operations on A is denoted by O(n) A , and the set of all operations on A is A . For a subset F ⊆ OA and an integer n ≥ 1, the n-ary A . For each n ≥ 1 and 1 ≤ i ≤ n, the n-ary on A defined by (a1, . . . , an) 7→ ai is called the i-th n-ary projection part of F is defined as F (n) := F ∩ O(n) operation en,A on A. We denote the set of all projections on A by EA := {en,A denoted by OA :=Sn≥1 O(n) 1 ≤ i ≤ n}. i i We denote the set of natural numbers by ω := {0, 1, 2, . . . }, and we regard its elements as ordinals, i.e., n ∈ ω is the set of lesser ordinals {0, 1, . . . , n−1}. Thus, an n-tuple a ∈ An is formally a map a : {0, 1, . . . , n−1} → A. The notation (ai i ∈ n) means the n-tuple mapping i to ai for each i ∈ n. The notation (a1, . . . , an) means the n-tuple mapping i to ai+1 for each i ∈ n. We view an m × n matrix M ∈ Am×n with entries in A as an n-tuple of m- tuples M := (a1, . . . , an). The m-tuples a1, . . . , an are called the columns of M. 1, . . . , ai Mi := (ai i=1 ni matrix (a1 1, . . . , a1 n1 , a2 has no columns and is denoted by (). ni ) is an m × ni matrix, then we denote by [M1M2 · · · Mp] the np ). An empty matrix For i ∈ m, the n-tuple (cid:0)a1(i), . . . , an(i)(cid:1) is called row i of M. If for 1 ≤ i ≤ p, m × Pp by f M the m-tuple (cid:0)f (a1(i), . . . , an(i))(cid:12)(cid:12) i ∈ m(cid:1) in Bm, in other words, f M is the For a function f : An → B and a matrix M := (a1, . . . , an) ∈ Am×n, we denote Mal'cev [12] introduced the operations ζ, τ , ∆, ∇, ∗ on the set OA of all opera- m-tuple obtained by applying f to the rows of M. n2 , . . . , ap 1, . . . , a2 1, . . . , ap tions on A, defined as follows for arbitrary f ∈ O(n) A , g ∈ O(m) A : (ζf )(x1, x2, . . . , xn) := f (x2, x3, . . . , xn, x1), (τ f )(x1, x2, . . . , xn) := f (x2, x1, x3, . . . , xn), (∆f )(x1, x2, . . . , xn−1) := f (x1, x1, x2, . . . , xn−1) 1 2 MIGUEL COUCEIRO AND ERKKO LEHTONEN for n > 1, ζf = τ f = ∆f := f for n = 1, and (∇f )(x1, x2, . . . , xn+1) := f (x2, . . . , xn+1), (f ∗ g)(x1, x2, . . . , xm+n−1) := f(cid:0)g(x1, x2, . . . , xm), xm+1, . . . , xm+n−1(cid:1). The operations ζ and τ are collectively referred to as permutation of variables, ∆ is called identification of variables (also known as diagonalization), ∇ is called addition of a dummy variable (or cylindrification), and ∗ is called composition. The algebra (OA; ζ, τ, ∆, ∇, ∗) of type (1, 1, 1, 1, 2) is called the full iterative algebra on A, and its subalgebras are called iterative algebras on A. A subset F ⊆ OA is called a clone on A, if it is the universe of an iterative algebra on A that contains all projections en,A , 1 ≤ i ≤ n. i A Galois connection between sets A and B is a pair (σ, π) of mappings σ : P(A) → P(B) and π : P(B) → P(A) between the power sets P(A) and P(B) such that for all X, X ′ ⊆ A and all Y, Y ′ ⊆ B the following conditions are satisfied: and or, equivalently, X ⊆ X ′ =⇒ σ(X) ⊇ σ(X ′), Y ⊆ Y ′ =⇒ π(Y ) ⊇ π(Y ′), X ⊆ π(σ(X)), Y ⊆ σ(π(Y )), X ⊆ π(Y ) ⇐⇒ σ(X) ⊇ Y. Galois connections can be equivalently described as certain mappings induced by polarities, i.e., relations R ⊆ A × B, as the following well-known theorem shows (for early references, see [6, 13]; see also [4, 10]): Theorem 1.1. Let A and B be nonempty sets and let R ⊆ A × B. Define the mappings σ : P(A) → P(B), π : P(B) → P(A) by σ(X) := {y ∈ B ∀x ∈ X : (x, y) ∈ R}, π(Y ) := {x ∈ A ∀y ∈ Y : (x, y) ∈ R}. Then the pair (σ, π) is a Galois connection between A and B. A prototypical example of a Galois connection is given by the Pol -- Inv theory of functions and relations. For m ≥ 1, we denote R(m) A := {R R ⊆ Am} = P(Am) and RA := [m≥1 R(m) A . Let R ∈ R(m) A . For a matrix M ∈ Am×n, we write M ≺ R to mean that the columns of M are m-tuples from the relation R. An operation f : An → A is said to preserve R (or f is a polymorphism of R, or R is an invariant of f ), denoted f ⊲ R, if for all m × n matrices M ∈ Am×n M ≺ R implies f M ∈ R. For a relation R ∈ RA, we denote by Pol R the set of all operations f ∈ OA that preserve the relation R. For a set Q ⊆ RA of relations, we let Pol Q :=TR∈Q Pol R. GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 3 The sets Pol R and Pol Q are called the sets of all polymorphisms of R and Q, respectively. Similarly, for an operation f ∈ OA, we denote by Inv f the set of all relations R ∈ RA that are preserved by f . For a set F ⊆ OA of functions, we let of f and F , respectively. Inv F :=Tf ∈F Inv f . The sets Inv f and Inv F are called the sets of all invariants By Theorem 1.1, (Inv, Pol) is the Galois connection induced by the relation ⊲ between the set OA of all operations on A and the set RA of all relations on A. It was shown by Geiger [7] and independently by Bodnarcuk, Kaluznin, Kotov and Romov [1] that for finite sets A, the closed subsets of OA under this Galois connec- tion are exactly the clones on A. These authors also described the Galois closed subsets of RA by defining an algebra on RA and showing that the closed sets of re- lations are exactly the relational clones, i.e., the subuniverses of the aforementioned algebra on RA. Theorem 1.2 (Geiger [7]; Bodnarcuk, Kaluznin, Kotov and Romov [1]). Let A be a finite nonempty set. (i) A set F ⊆ OA of operations is the set of polymorphisms of some set Q ⊆ RA of relations if and only if F is a clone on A. (ii) A set Q ⊆ RA of relations is the set of invariants of some set F ⊆ OA of operations if and only if Q is a relational clone on A. On arbitrary, possibly infinite sets A, the Galois closed sets of operations are the locally closed clones, as shown by Szab´o [18] and independently by Poschel [15]. A set F ⊆ OA of operations is said to be locally closed, if it holds that for all f ∈ OA, say of arity n, f ∈ F whenever for all finite subsets F ⊆ An, there exists a function g ∈ F (n) such that f F = gF . These results were generalized to iterative algebras (with or without projections) by Harnau [8] who defined a polarity between operations and relation pairs. An m-ary relation pair on A is a pair (R, R′) where R, R′ ∈ R(m) A for some m ≥ 1 and R′ ⊆ R. For m ≥ 1, denote and H(m) A := {(R, R′) R′ ⊆ R ⊆ Am} H(m) A . HA = [m≥1 An operation f ∈ OA is said to preserve a relation pair (R, R′) ∈ H(m) A , denoted f ⊲ (R, R′), if for all matrices M ∈ Am×n, M ≺ R implies f M ∈ R′. In light of Theorem 1.1, the preservation relation ⊲ induces a Galois connection between the sets OA and HA. Harnau showed that the closed sets of operations are exactly the universes of iterative algebras. He defined certain operations on the set HA of relation pairs and showed that the Galois closed subsets of relation pairs are precisely the subsets that are closed under these operations. In [11], the subalgebras of the reduct (OA; ζ, τ, ∇, ∗) of the full iterative algebra containing all projections were completely characterized in terms of a preservation relation between operations and so-called clusters. In analogy with Harnau's ap- proach to iterative algebras (i.e., considering all subalgebras of (OA; ζ, τ, ∆, ∇, ∗) with or without projections and thus extending the Pol -- Inv theory of clones and re- lations), in this paper we relax the closure system to all subalgebras of (OA; ζ, τ, ∇, ∗), 4 MIGUEL COUCEIRO AND ERKKO LEHTONEN not necessarily containing all projections. This is achieved within a Galois frame- work where the dual objects are systems of pointed multisets. We will also describe the Galois closed sets of these dual objects in terms of explicit closure conditions. Furthermore, we will show that the respective closure systems are uncountable for all A ≥ 2. Such a relaxation is both natural and noteworthy. Motivating examples are those sets of operations obtained from clones by removing all those operations that have arity at most m for some fixed m ≥ 1. Clearly, these sets are closed under permutation of variables, addition of dummy variables, and composition, and hence they constitute subalgebras of the reduct (OA; ζ, τ, ∇, ∗). This line of research has been carried out by several authors (see [1, 2, 7, 8, 9, 11, 14, 15, 18]; a brief survey of the topic is provided in [11]; see also chapter "Galois connections for operations and relations" by R. Poschel in [4], pp. 231 -- 258). Sections 1 and 2 of the current paper are based on [3]. 2. Sets of operations closed under permutation of variables, addition of dummy variables, and composition We now consider the problem of characterizing the sets of operations on an arbitrary nonempty set A that are closed under permutation of variables, addition of dummy variables, and composition (but not necessarily under identification of variables). A finite multiset S on a set A is a map νS : A → ω, called a multiplicity function, such that the set {x ∈ A νS(x) 6= 0} is finite. Then the sum Px∈A νS(x) is a well-defined natural number, and it is called the cardinality of S and denoted by S. The number νS(x) is called the multiplicity of x in S. We may represent a finite multiset S by giving a list enclosed in set brackets, i.e., {a1, . . . , an}, where each element x ∈ A occurs νS(x) times. If S′ is another multiset on A corresponding to νS ′ : A → ω, then we say that S′ is a submultiset of S, denoted S′ ⊆ S, if νS ′(x) ≤ νS(x) for all x ∈ A. We denote the set of all finite multisets on A by M(A). We also denote, for each p ≥ 0, by M(p)(A) the set of all finite multisets on A of cardinality at most p, i.e., M(p)(A) := {S ∈ M(A) S ≤ p}. The set M(A) is partially ordered by the multiset inclusion relation "⊆". The join S ⊎ S′ and the difference S \ S′ of multisets S and S′ are determined by the multiplicity functions νS⊎S ′ (x) := νS(x) + νS ′ (x), νS\S ′ (x) := max{νS(x) − νS ′ (x), 0}, respectively. The empty multiset on A is the zero function, and it is denoted by ε. A partition of a finite multiset S on A is a multiset {S1, . . . , Sn} (on the set of all finite multisets on A) of nonempty finite multisets on A such that S = S1 ⊎ · · · ⊎ Sn. A pointed multiset on a set A is a pair (x, S) ∈ A × M(A). The multiset {x} ⊎ S is called the underlying multiset of (x, S). We define the cardinality of a pointed multiset (x, S) to be equal to the cardinality of its underlying multiset, i.e., (x, S) := {x} ⊎ S = S + 1. If (x, S) and (x′, S) are pointed multisets on A, then we say that (x, S) is a pointed submultiset of (x′, S′), denoted (x, S) ⊆ (x′, S′), if x = x′ and S ⊆ S′. For an m × n matrix M ∈ Am×n, the multiset of columns of M is the multiset M∗ on Am defined by the function χM which maps each m-tuple a ∈ Am to the GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 5 number of times a occurs as a column of M. A matrix N ∈ Am×n′ of M ∈ Am×n if N∗ ⊆ M∗, i.e., χN(a) ≤ χM(a) for all a ∈ Am. is a submatrix For an integer m ≥ 1, an m-ary system of pointed multisets on A (a system for short) is a pair (Φ, Φ′) ∈ P(M(Am)) × P(Am × M(Am)), where the antecedent Φ ⊆ M(Am) is a set of finite multisets on Am and the consequent Φ′ ⊆ Am×M(Am) is a set of pointed multisets on Am, satisfying the following two conditions: (1) (x, S) ∈ Φ′ and S′ ⊆ S imply (x, S′) ∈ Φ′; (2) (x, S) ∈ Φ′ implies {x} ⊎ S ∈ Φ. For m ≥ 1, we denote the set of all m-ary systems of pointed multisets on A by W (m) A , and we denote the set of all systems of pointed multisets on A by WA := [m≥1 W (m) A . For (Φ, Φ′), (Ψ, Ψ′) ∈ P(M(Am)) × P(Am ×M(Am)), we write (Φ, Φ′) ⊆ (Ψ, Ψ′) to mean componentwise inclusion, i.e., Φ ⊆ Ψ and Φ′ ⊆ Ψ′. If M ∈ Am×n and Φ is a set of multisets over Am, we write M ≺ Φ to mean that the multiset M∗ of columns of M is an element of Φ. If M = (m1, . . . , mn) ∈ Am×n, n ≥ 1, and Φ′ is a set of pointed multisets (x, S) ∈ Am × M(Am), then we write M ≺ Φ′ to mean that (m1, {m2, . . . , mn}) ∈ Φ′. If f ∈ On A , we say that f preserves (Φ, Φ′), denoted f ⊲ (Φ, Φ′), if for every matrix M ∈ Am×p for some p ≥ 0, it holds that whenever M ≺ Φ and M = [M1M2] where M1 has n columns and M2 may be empty, we have that [f M1M2] ≺ Φ′. A and (Φ, Φ′) ∈ W (m) In light of Theorem 1.1, the relation ⊲ establishes a Galois connection between the sets OA and WA. We say that a set F ⊆ OA of operations on A is characterized by a set W ⊆ WA of systems of pointed multisets, if F = {f ∈ OA ∀(Φ, Φ′) ∈ W : f ⊲ (Φ, Φ′)}, i.e., F is precisely the set of operations on A that preserve every system of pointed multisets in W. Similarly, we say that W is characterized by F , if W = {(Φ, Φ′) ∈ WA ∀f ∈ F : f ⊲ (Φ, Φ′)}, i.e., W is precisely the set of systems of pointed multisets that are preserved by every operation in F . Thus, the Galois closed sets of operations (systems of pointed multisets) are exactly those that are characterized by systems of pointed multisets (operations, respectively). Example 2.1. Let p ≥ 1 be an integer and consider the set O(≥p) of operations on A of arity at least p. It is clearly a subalgebra of (OA; ζ, τ, ∇, ∗) which does not contain all projections. Let S be any multiset on A of cardinality A characterizes O(≥p) p − 1. A . For, every operation on A of arity at least p vacuously preserves ({S}, ∅), but no operation of arity less than p can preserve ({S}, ∅). It is easy to verify that the system ({S}, ∅) ∈ W (1) A := Sn≥p O(n) A Example 2.2. Let R ⊆ Am be an m-ary relation R on A. Let ΦR be the set of all finite multisets S on Am such that νS(a) 6= 0 only if a ∈ R, and let Φ′ R := R × ΦR. It is easy to verify that f ⊲ R if and only if f ⊲ (ΦR, Φ′ R). Hence, if C is a clone on A such that C = Pol Q for some set Q of relations on A, then C is characterized by the set {(ΦR, Φ′ R) R ∈ Q} of systems of pointed multisets. 6 MIGUEL COUCEIRO AND ERKKO LEHTONEN As a corollary, for a clone C = Pol Q and an integer p ≥ 1, the set C(≥p) := C ∩ O(≥p) R) R ∈ Q} ∪ {({S}, ∅)} of systems of pointed multisets, where S is any multiset on A with S = p − 1. of at least p-ary members of C is characterized by the set {(ΦR, Φ′ A Lemma 2.3. Let F ⊆ OA be a locally closed set of operations that is closed under permutation of variables, addition of dummy variables, and composition. Then for every g ∈ OA \ F , there exists a system (Φ, Φ′) ∈ WA that is preserved by every operation in F but not by g. Proof. The statement is vacuously true for F = OA, and it is easily seen to hold for F = ∅. Thus, we can assume that ∅ ( F ( OA. Suppose that g ∈ OA \ F is n-ary. Since F is locally closed, there is a finite subset F ⊆ An such that gF 6= f F for every f ∈ F (n). Clearly F is nonempty. Let M be a F × n matrix whose rows are the elements of F in some fixed order. Let µ be the smallest integer such that F (µ) 6= ∅. Note that since F is closed under addition of dummy variables, F (m) 6= ∅ for all m ≥ µ. Let X be any submultiset of M∗. (Recall that M∗ denotes the multiset of columns of M.) Let Π := (M1, . . . , Mq) be a sequence of submatrices of M such that {M∗ 1, . . . , M∗ q} is a partition of M∗ \ X where each block M∗ i has cardinality at least µ. For 1 ≤ i ≤ q, let di ∈ FMi, and let D := (d1, . . . , dq). (Note that each FMi is nonempty, because F contains functions of arity M∗ i ≥ µ. Observe also that if Mi is a submatrix of Mj, then FMi ⊆ FMj, because F is closed under permutation of variables and addition of dummy variables; in particular, each FMi is a subset of FM.) Denote ⌈X, Π, D⌋ := D∗ ⊎ X, and for X 6= M∗, denote hX, Π, Di := (d1, {d2, . . . , dq} ⊎ X) ∈ Am × M(Am). Note that if hX, Π, Di = (x, S), then ⌈X, Π, D⌋ = {x} ⊎ S. We define Φ′ to be the set of all hX, Π, Di for all possible choices of X, Π, and D such that X 6= M∗, and we define Φ to be the set of all ⌈X, Π, D⌋ for all possible choices of X, Π and D, i.e., Φ := {x ⊎ S (x, S) ∈ Φ′} ∪ {M∗}. We first verify that (Φ, Φ′) is indeed a system of pointed multisets. The first condition in the definition of a system of pointed multisets is clearly satisfied, by the definition of (Φ, Φ′). For the second condition, let (x, S) = hX, Π, Di ∈ Φ′, where Π := (M1, . . . , Mq) and D := (d1, . . . , dq). A simple inductive argument proves that (x, S′) ∈ Φ′ for all S′ ⊆ S, and it suffices to show that (x, S′) ∈ Φ′ whenever S = S′ ⊎ {y} for some y ∈ Am. If y ∈ X, then we let X ′ := X \ {y}, M′ 1 := [M1y], Π′ := (M′ 1, M2, . . . , Mq). Since F is closed under addition of dummy variables, we have that d1 ∈ FM′ 1, and hence we have (x, S′) = hX ′, Π′, Di ∈ Φ′. If y ∈ {d2, . . . , dq}, say, y = dj, then we let M′ 1 := [M1Mj], Π′ := (M1, . . . , Mj−1, Mj+1, . . . , Mq), GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 7 and again, since F is closed under addition of dummy variables, we have that d1 ∈ FM′ 1 and we can let D′ := (d1, . . . , dj−1, dj+1, . . . , dq), and we have (x, S′) = hX, Π′, D′i ∈ Φ′. Observe first that g 6⊲ (Φ, Φ′). For, we have that M ≺ Φ by the definition of Φ. On the other hand, since gM /∈ FM, we have that gM is not the first component of hX, Π, Di for all X, Π, D, and hence gM ⊀ Φ′. It remains to show that f ⊲ (Φ, Φ′) for all f ∈ F . Assume that f is n-ary. If N := [N1N2] ≺ Φ, where N1 has n columns, then N∗ = ⌈X, Π, D⌋ for some X, Π := (M1, . . . , Mq), D := (d1, . . . , dq), where {M∗ q} is a partition of M∗ \ X and for 1 ≤ i ≤ q, di ∈ FMi. We will show by induction on q that for every f ∈ F , there exist X ′, Π′, D′ such that (f N1, N∗ 2) = hX ′, Π′, D′i and hence [f N1N2] ≺ Φ′. 1, . . . , M∗ If q = 0, then we have X = M∗, Π = (), D = (), and ⌈X, Π, D⌋ = M∗, and the condition N∗ = M∗ implies that N1 is a submatrix of M. Then f N1 ∈ F N1 and (f N1, N∗ 2) = hM∗ \ N∗ 1, (N1), (f N1)i. Assume that the claim holds for q = k ≥ 0, and consider the case that q = k + 1. 1 ⊆ X, then f N1 ∈ F N1 We assume that N = [N1N2] and N∗ = ⌈X, Π, D⌋. If N∗ and (f N1, N∗ 2) = hX \ N∗ 1, (M1, . . . , Mk+1, N1), (d1, . . . , dk+1, f N1)i. Otherwise, for some i ∈ {1, . . . , k + 1}, di is a column of N1. Denote by N′ 1 the matrix obtained from N1 by deleting the column di. Since F is closed under per- mutation of variables, there is an operation f ′ ∈ F (n) such that f N1 = f ′[diN′ 1]. By the definition of di, there is an operation h ∈ F such that hMi = di, and we have that (2.1) 1]. Since F is closed under composition, f ′ ∗ h ∈ F . Furthermore, 1] = (f ′ ∗ h)[MiN′ 1] = f ′[hMiN′ f ′[diN′ [MiN′ 1N2]∗ = ⌈X ⊎ M∗ i , (M1, . . . , Mi−1, Mi+1, . . . , Mk+1), (d1, . . . , di−1, di+1, . . . , dk+1)⌋. By the induction hypothesis, there exist X ′, Π′, D′ such that ((f ′ ∗ h)[MiN′ 1], N∗ 2) = hX ′, Π′, D′i. By (2.1), we have that (f ′ ∗h)[MiN′ 1] = f N1, and hence (f N1, N∗ 2) = hX ′, Π′, D′i. (cid:3) Theorem 2.4. Let A be an arbitrary, possibly infinite nonempty set. For any set F ⊆ OA of operations, the following two conditions are equivalent: (i) F is locally closed and closed under permutation of variables, addition of dummy variables, and composition. (ii) F is characterized by a set W ⊆ WA of systems of pointed multisets. Proof. (ii) ⇒ (i): It is straightforward to verify that the set of operations preserving a set of systems of pointed multisets is closed under permutation of variables and addition of dummy variables. To see that it is closed under composition, let f ∈ F (n) and g ∈ F (p), and consider f ∗ g : An+p−1 → A. Let (Φ, Φ′) ∈ W, and let 8 MIGUEL COUCEIRO AND ERKKO LEHTONEN M := [M1M2M3] ≺ Φ, where M1 has p columns and M2 has n − 1 columns. Since g ⊲ (Φ, Φ′), we have that [gM1M2M3] ≺ Φ′ and hence, by property (2) of the definition of a system of pointed multisets, we have [gM1M2M3] ≺ Φ. Then Φ′. But f [gM1M2] = (f ∗ g)[M1M2], [gM1M2] has n columns, and since f ⊲ (Φ, Φ′), we have that (cid:2)f [gM1M2](cid:12)(cid:12)M3(cid:3) ≺ so (cid:2)(f ∗ g)[M1M2](cid:12)(cid:12)M3(cid:3) ≺ Φ′, and thus f ∗ g ⊲ (Φ, Φ′). It remains to show that F is locally closed. It is clear that OA is locally closed, so we may assume that F 6= OA. Suppose on the contrary that there is a g ∈ OA \ F , say of arity n, such that for every finite subset F ⊆ An, there is an f ∈ F (n) such that gF = f F . Since F is characterized by W and g /∈ F , there is a system (Φ, Φ′) ∈ W such that g 6⊲ (Φ, Φ′), and hence for some matrix M := [M1M2] ≺ Φ where M1 has n columns, we have that [gM1M2] ⊀ Φ′. Let F be the finite set of rows of M1. By our assumption, there is an f ∈ F (n) such that gF = f F , and hence f M1 = f F M1 = gF M1 = gM1, and so [f M1M2] ⊀ Φ′, which contradicts the assumption that f ⊲ (Φ, Φ′). (i) ⇒ (ii): It follows from Lemma 2.3 that for every operation g ∈ OA \ F , there exists a system (Φ, Φ′) ∈ WA that is preserved by every operation in F but not by g. The set of all such "separating" systems of pointed multisets, for each g ∈ OA \ F , characterizes F . (cid:3) 3. Closure conditions for systems of pointed multisets In this section we will describe the sets of systems of pointed multisets that are characterized by sets of operations in terms of explicit closure conditions. We will follow Couceiro and Foldes's [2] proof techniques and adapt their notion of con- junctive minor to systems of pointed multisets. We first introduce several technical notions and definitions that will be needed in the statement of Theorem 3.9 and in its proof. For maps f : A → B and g : C → D, the composition g ◦ f is defined only if B = C. Removing this restriction, the concatenation of f and g is defined to be the map gf : f −1[B ∩ C] → D given by the rule (gf )(a) = g(cid:0)f (a)(cid:1) for all a ∈ f −1[B ∩ C]. Clearly, if B = C, then gf = g ◦ f ; thus functional composition is subsumed and extended by concatenation. Concatenation is associative, i.e., for any maps f , g, h, we have h(gf ) = (hg)f . For a family (gi)i∈I of maps gi : Ai → Bi such that Ai∩Aj = ∅ whenever i 6= j, we define the (piecewise) sum of the family (gi)i∈I to be the map Pi∈I gi : Si∈I Ai → Si∈I Bi whose restriction to each Ai coincides with gi. If I is a two-element set, say I = {1, 2}, then we write g1 + g2. Clearly, this operation is associative and commutative. Concatenation is distributive over summation, i.e., for any family (gi)i∈I of maps on disjoint domains and any map f , (cid:0)Xi∈I gi(cid:1)f =Xi∈I (gif ) and f(cid:0)Xi∈I gi(cid:1) =Xi∈I (f gi). In particular, if g1 and g2 are maps with disjoint domains, then (g1 + g2)f = (g1f ) + (g2f ) and f (g1 + g2) = (f g1) + (f g2). GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 9 Let g1, . . . , gn be maps from A to B. The n-tuple (g1, . . . , gn) determines a vector-valued map g : A → Bn, given by g(a) := (cid:0)g1(a), . . . , gn(a)(cid:1) for every a ∈ A. For f : Bn → C, the composition f ◦ g is a map from A to C, denoted by f (g1, . . . , gn), and called the composition of f with g1, . . . , gn. Suppose that A∩A′ = n are maps from A′ to B. Let g and g′ be the vector-valued maps ∅ and g′ determined by (g1, . . . , gn) and (g′ n), respectively. We have that f (g + g′) = (f g) + (f g′), i.e., 1, . . . , g′ 1, . . . , g′ f(cid:0)(g1 + g′ 1), . . . , (gn + g′ n)(cid:1) = f (g1, . . . , gn) + f (g′ 1, . . . , g′ n). For B ⊆ A, ιAB denotes the canonical injection (inclusion map) from B to A. Thus the restriction f B of any map f : A → C to the subset B is given by f B = f ιAB. Remark 3.1. Observe that the notation f M introduced in Section 1 is in accordance with the notation for concatenation of mappings. Since a matrix M := (a1, . . . , an) is an n-tuple of m-tuples ai : m → A, 1 ≤ i ≤ n, the composition of the vector- valued map (a1, . . . , an) : m → An with f : An → B gives rise to the m-tuple f (a1, . . . , an) : m → B. Let m and n be positive integers (viewed as ordinals, i.e., m = {0, . . . , m − 1}). Let h : n → m ∪ V where V is an arbitrary set of symbols disjoint from the ordinals, called existentially quantified indeterminate indices, or simply indeterminates, and let σ : V → A be any map, called a Skolem map. Then each m-tuple a ∈ Am, being a map a : m → A, gives rise to an n-tuple (a + σ)h =: (b0, . . . , bn−1) ∈ An, where bi :=(ah(i), σ(h(i)), if h(i) ∈ {0, 1, . . . , m − 1}, if h(i) ∈ V . Let H := (hj)j∈J be a nonempty family of maps hj : nj → m ∪ V , where each nj is a positive integer. Then H is called a minor formation scheme with target m, indeterminate set V , and source family (nj)j∈J . Let (Φj)j∈J be a family of sets of multisets (or pointed multisets), each Φj on the set Anj , and let Φ be a set of multisets (or pointed multisets, respectively), on Am. We say that Φ is a restrictive conjunctive minor of the family (Φj )j∈J via H, if, for every m × n matrix M := (a1, . . . , an) ∈ Am×n, On the other hand, if, for every m × n matrix M := (a1, . . . , an) ∈ Am×n, M ≺ Φ =⇒ (cid:2)∃σ1, . . . , σn ∈ AV ∀j ∈ J : (cid:0)(a1 + σ1)hj, . . . , (an + σn)hj(cid:1) ≺ Φj(cid:3). (cid:2)∃σ1, . . . , σn ∈ AV ∀j ∈ J : (cid:0)(a1 + σ1)hj, . . . , (an + σn)hj(cid:1) ≺ Φj(cid:3) =⇒ M ≺ Φ, then we say that Φ is an extensive conjunctive minor of the family (Φj)j∈J via H. If Φ is both a restrictive conjunctive minor and an extensive conjunctive minor of the family (Φj)j∈J via H, i.e., for every m × n matrix M := (a1, . . . , an) ∈ Am×n, M ≺ Φ ⇐⇒ (cid:2)∃σ1, . . . , σn ∈ AV ∀j ∈ J : (cid:0)(a1 + σ1)hj, . . . , (an + σn)hj(cid:1) ≺ Φj(cid:3), then Φ is said to be a tight conjunctive minor of the family (Φj)j∈J via H. If (Φ, Φ′) ∈ WA is a system of pointed multisets on A and (Φj, Φ′ j)j∈J is a family of systems of pointed multisets on A (of various arities) such that Φ is a restrictive conjunctive minor of the family (Φj)j∈J of multisets via a scheme H and Φ′ is an extensive conjunctive minor of the family (Φ′ j)j∈J of pointed multisets via the same scheme H, then (Φ, Φ′) is said to be a conjunctive minor of the family (Φj, Φ′ j)j∈J 10 MIGUEL COUCEIRO AND ERKKO LEHTONEN via H. If both Φ and Φ′ are tight conjunctive minors of the respective families via H, then (Φ, Φ′) is said to be a tight conjunctive minor of the family (Φj, Φ′ j)j∈J via H. If the minor formation scheme H := (hj)j∈J and the family (Φj, Φ′ j)j∈J are indexed by a singleton J := {0}, then a tight conjunctive minor (Φ, Φ′) of a family consisting of a single system of pointed multisets (Φ0, Φ′ 0) is called a simple minor of (Φ0, Φ′ 0). j)j∈J of members of WA, and let f ∈ OA. If f ⊲ (Φj , Φ′ (Φ, Φ′) be a conjunctive minor of a nonempty family j) for all j ∈ J, Lemma 3.2. Let (Φj, Φ′ then f ⊲ (Φ, Φ′). Proof. Let (Φ, Φ′) be an m-ary conjunctive minor of the family (Φj, Φ′ j)j∈J via the scheme H := (hj)j∈J , hj : nj → m∪V . Let M := (a1, . . . , an) be an arbitrary m×n matrix such that M ≺ Φ. We want to prove that f M ≺ Φ′. Since Φ is a restrictive conjunctive minor of (Φj )j∈J via H = (hj )j∈J , there exist Skolem maps σi : V → A, Since Φ′ is an extensive conjunctive minor of (Φ′ 1 ≤ i ≤ n, such that for every j ∈ J, Mj :=(cid:0)(a1 + σ1)hj, . . . , (an + σn)hj(cid:1) ≺ Φj. j)j∈J via the same scheme H = (hj)j∈J , to prove that f M ≺ Φ′, it suffices to give a Skolem map σ : V → A such that, for all j ∈ J, (f M + σ)hj ≺ Φ′ j. Let σ := f (σ1, . . . , σn). We have that, for each j ∈ J, (f M + σ)hj =(cid:0)f (a1, . . . , an) + f (σ1, . . . , σn)(cid:1)hj =(cid:0)f (a1 + σ1, . . . , an + σn)(cid:1)hj = f(cid:0)(a1 + σ1)hj, . . . , (an + σn)hj(cid:1) = f Mj. j), so we have f Mj ≺ Φ′ j. By our assumption f ⊲ (Φj, Φ′ (cid:3) We say that a set W ⊆ WA of systems of pointed multisets is closed under j)j∈J is a nonempty family of formation of conjunctive minors if whenever (Φj, Φ′ members of W, all conjunctive minors of the family (Φj, Φ′ j)j∈J are also in W. Let (Φ, Φ′), (Ψ, Ψ′) ∈ W (m) A . If Φ ⊆ Ψ and Φ′ = Ψ′, then we say that (Φ, Φ′) is obtained from (Ψ, Ψ′) by restricting the antecedent. If Φ = Ψ and Φ′ ⊇ Ψ′, then we say that (Φ, Φ′) is obtained from (Ψ, Ψ′) by extending the consequent. The formation of conjunctive minors subsumes the formation of simple minors as well as the operations of restricting the antecedent and extending the consequent. Simple minors in turn subsume permutation of arguments, projection, identification of arguments, and addition of a dummy argument, operations which can be defined for systems of pointed multisets in an analogous way as for Pippenger's [14] constraints or Hellerstein's [9] generalized constraints. The m-ary trivial system of pointed multisets on A is Ωm := (M(Am), Am × M(Am)). For p ≥ 0, the m-ary trivial system of pointed multisets on A of breadth p is Ω(p) m := (M(p)(Am), Am × M(p−1)(Am)). The m-ary empty system on A is the pair (∅, ∅). Note that Ω(0) m . The m-ary equality system on A, denoted Em, is the system Em := (Em, E′ m 6= ∅, because ({ε}, ∅) is the unique member of Ω(0) m), where Em := {S ∈ M(Am) νS(a1, . . . , am) 6= 0 =⇒ a1 = · · · = am}, E′ m := {(a, . . . , a) ∈ Am a ∈ A} × Em. Lemma 3.3. Let W ⊆ WA be a set of systems of pointed multisets that contains the binary equality system and the unary empty system. If W is closed under formation GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 11 of conjunctive minors, then it contains all trivial systems, all equality systems, and all empty systems. Proof. The unary trivial system is a simple minor of the binary equality system via the scheme H := {h}, where h : 2 → 1 is given by h(0) = h(1) = 0 (by identification of arguments). The m-ary trivial system is a simple minor of the unary trivial system via the scheme H := {h}, where h : 1 → m is given by h(0) = 0 (by addition of m − 1 dummy arguments). For m ≥ 2, the m-ary equality system is a conjunctive minor of the binary equality system via the scheme H := (hi)i∈m−1, where hi : 2 → m is given by hi(0) = i, hi(1) = i + 1 (by addition of n − 2 dummy arguments, restricting the antecedents and intersecting the consequents). The m-ary empty system is a simple minor of the unary empty system via the scheme H := {h}, where h : 1 → m is given by h(0) = 0 (by addition of m − 1 dummy arguments). (cid:3) We define the union of systems of pointed multisets componentwise, if j)j∈J is a family of pointed multisets on A of a common arity m, then the i.e., (Φj, Φ′ union of (Φj, Φ′ j)j∈J is [j∈J (Φj, Φ′ j) :=(cid:0)[j∈J Φj, [j∈J Φ′ j(cid:1). j). Lemma 3.4. Let (Φj, Φ′ multisets on A. If f : An → A preserves (Φj, Φ′ j)j∈J be a nonempty family of m-ary systems of pointed j) for every j ∈ J, then f preserves Sj∈J (Φj, Φ′ Proof. Let M := [M1M2] ≺ Sj∈J Φj. Then M ≺ Φj for some j ∈ J. By the Sj∈J Φ′ Let Φ ⊆ M(Am), Φ′ ⊆ Am × M(Am), S ∈ M(Am). The quotient of the set Φ j, and hence [f M1M2] ≺ (cid:3) j), we have [f M1M2] ≺ Φ′ assumption that f ⊲ (Φj, Φ′ j. of multisets by the multiset S is defined as Φ/S := {S′ ∈ M(Am) S ⊎ S′ ∈ Φ}. The quotient of the set Φ′ of pointed multisets by the multiset S is defined as Φ′/S := {(x, S′) ∈ Am × M(Am) (x, S ⊎ S′) ∈ Φ′}. The quotient of the pair (Φ, Φ′) by S is defined componentwise, i.e., (Φ, Φ′)/S := (Φ/S, Φ′/S). It is easy to verify that if (Φ, Φ′) is a system of pointed multisets on A, then so it (Φ, Φ′)/S for every S ∈ M(Am). Lemma 3.5. Let Φ, Ψ ⊆ M(Am) and S ∈ M(Am). Then (i) X ∈ Φ/S if and only if X ⊎ S ∈ Φ; (ii) (Φ ∪ Ψ)/S = (Φ/S) ∪ (Ψ/S). Proof. (i) Immediate from the definition. (ii) By part (i) and the definition of union, we have X ∈ (Φ ∪ Ψ)/S ⇐⇒ X ⊎ S ∈ Φ ∪ Ψ ⇐⇒ X ⊎ S ∈ Φ ∨ X ⊎ S ∈ Ψ ⇐⇒ X ∈ Φ/S ∨ X ∈ Ψ/S ⇐⇒ X ∈ (Φ/S) ∪ (Ψ/S). The claimed equality thus follows. (cid:3) 12 MIGUEL COUCEIRO AND ERKKO LEHTONEN Lemma 3.6. Let (Φ, Φ′) be an m-ary system of pointed multisets on A. If f : An → A preserves (Φ, Φ′), then f preserves (Φ, Φ′)/S for every multiset S ∈ M(Am). Proof. Let [M1M2] ≺ Φ/S. Let N be a matrix such that N∗ = S. Then [M1M2N] ≺ Φ. By our assumption that f ⊲ (Φ, Φ′), we have [f M1M2N] ≺ Φ′. Thus, [f M1M2] ≺ Φ′/S, and we conclude that f ⊲ (Φ, Φ′)/S. (cid:3) Lemma 3.7. Assume that (Φ, Φ′) is an m-ary system on A such that Ω(p) (Φ, Φ′). preserves (Φ, Φ′). Proof. Let [M1M2] ≺ Φ, where M1 has n columns and M2 has n′ columns. If n′ < p, then the number of columns of [f M1M2] is n′ + 1 ≤ p, and hence [f M1M2] ≺ Φ′. Otherwise n′ ≥ p and, by our assumption, f ⊲ (Φ, Φ′)/M∗ 2. Thus, since M1 ≺ Φ/M∗ 2. Therefore [f M1M2] ≺ Φ′, and we conclude that f ⊲ (Φ, Φ′). (cid:3) m ⊆ If f : An → A preserves all quotients (Φ, Φ′)/S where S ≥ p, then f 2, we have that f M1 ≺ Φ′/M∗ For an m-ary system (Φ, Φ′) ∈ W (m) A and p ≥ 0, set (Φ, Φ′)(p) := (Φ(p), Φ′(p)), where Φ(p) := Φ ∩ M(p)(Am), Φ′(p) := Φ′ ∩ (Am × M(p−1)(Am)), m , is obtained from (Φ, Φ′) by restricting the breadth that is, (Φ, Φ′)(p) := (Φ, Φ′)∩Ω(p) to p. Lemma 3.8. Let (Φ, Φ′) be an m-ary system of pointed multisets on A. Then f : An → A preserves (Φ, Φ′) if and only if f preserves (Φ, Φ′)(p) for all p ≥ 0. Proof. Assume first that f ⊲ (Φ, Φ′). Let [M1M2] ≺ Φ(p). Since Φ(p) ⊆ Φ, we have that [M1M2] ≺ Φ, and hence [f M1M2] ≺ Φ′ by our assumption. The number of columns of [f M1M2] is at most p, so we have that [f M1M2] ≺ Φ′(p). Thus, f ⊲ (Φ, Φ′)(p). Assume then that f ⊲ (Φ, Φ′)(p) for all p ≥ 0. Let M := [M1M2] ≺ Φ, and let q be the number of columns in M. Then [M1M2] ≺ Φ(q), and hence [f M1M2] ≺ Φ′(q) by our assumption. Since Φ′(q) ⊆ Φ′, we have that [f M1M2] ≺ Φ′, and we conclude that f ⊲ (Φ, Φ′). (cid:3) We say that a set W ⊆ WA of systems of pointed multisets is • closed under quotients, if for any (Φ, Φ′) ∈ W, every quotient (Φ, Φ′)/S is also in W; • closed under dividends, if for every system (Φ, Φ′) ∈ WA, say of arity m, m ⊆ (Φ, Φ′) and (Φ, Φ′)/S ∈ W for it holds that (Φ, Φ′) ∈ W whenever Ω(p) every multiset S on Am of cardinality at least p; • locally closed, if (Φ, Φ′) ∈ W whenever (Φ, Φ′)(p) ∈ W for all p ≥ 0; • closed under unions, if Sj∈J (Φj , Φ′ nonempty family of m-ary systems in W; j) ∈ W whenever (Φj, Φ′j)j∈J is a • closed under formation of conjunctive minors, if all conjunctive minors of nonempty families of members of W are members of W. Theorem 3.9. Let A be an arbitrary, possibly infinite nonempty set. For any set W ⊆ WA of systems of pointed multisets on A, the following two conditions are equivalent: GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 13 (i) W is locally closed and contains the binary equality system, the unary empty system, and all unary trivial systems of breadth p ≥ 0, and it is closed under formation of conjunctive minors, unions, quotients, and dividends. (ii) W is characterized by some set F ⊆ OA of operations. In order to prove Theorem 3.9, we need to extend the notions of tuple and system of pointed multisets and allow them to have infinite arities, as will be explained below. Functions remain finitary. These extended definitions have no bearing on Theorem 3.9 itself; they are only needed as a tool in its proof. For any nonzero, possibly infinite ordinal m (an ordinal m is the set of lesser ordinals), an m-tuple a ∈ Am is formally a map a : m → A. The arities of tuples and systems of pointed multisets are thus allowed to be arbitrary nonzero, possibly infinite ordinals. In minor formation schemes, the target m and the members nj of the source family are also allowed to be arbitrary nonzero, possibly infinite ordinals. For systems of pointed multisets, we shall use the terms restrictive conjunctive ∞-minor, extensive conjunctive ∞-minor, conjunctive ∞-minor and simple ∞- minor to indicate a restrictive conjunctive minor, an extensive conjunctive minor, a conjunctive minor, or a simple minor via a scheme whose target and source ordinals may be infinite or finite. Thus in the sequel the use of the term "minor" without the prefix "∞" continues to mean the respective minor via a scheme whose target and source ordinals are all finite. Matrices can also have infinitely many rows but only a finite number of columns; an m × n matrix M ∈ Am×n, where n is finite but m may be finite or infinite, is an n-tuple of m-tuples M := (a1, . . . , an) where ai : m → A for 1 ≤ i ≤ n. Let H := (hj )j∈J be a minor formation scheme with target m, indeterminate set V and source family (nj )j∈J , and, for each j ∈ J, let Hj := (hi j)i∈Ij be a scheme with target nj, indeterminate set Vj and source family (ni j)i∈Ij . Assume that V is disjoint from the Vj's, and for distinct j's the Vj's are also pairwise disjoint. Then the composite scheme H(Hj j ∈ J) is the scheme K := (ki j )j∈J, i∈Ij defined as follows: (i) the target of K is the target m of H, (ii) the source family of K is (ni j)j∈J, i∈Ij , (iii) the indeterminate set of K is U := V ∪ (Sj∈J Vj), j : ni injection (inclusion map) from Vj to U . (iv) ki j → m ∪ U is defined by ki j := (hj + ιU Vj )hi j, where ιU Vj is the canonical For a set W of systems of pointed multisets on A of arbitrary, possibly infinite arities, we denote by W ∞ the set of those systems which are conjunctive ∞-minors of families of members of W. This set W ∞ is the smallest set of systems of pointed multisets containing W which is closed under formation of conjunctive ∞-minors, and it is called the conjunctive ∞-minor closure of W. Considering the formation of repeated conjunctive ∞-minors, we can show that the following lemma and corollary hold; these are analogues of Couceiro and Foldes's Claim 1 and Fact 1 in the proof of Theorem 3.2 in [2]. Lemma 3.10. If (Φ, Φ′) is a conjunctive ∞-minor of a nonempty family (Φj, Φ′ j)j∈J of systems of pointed multisets on A via the scheme H, and, for each j ∈ J, (Φj, Φ′ j) is a conjunctive ∞-minor of a nonempty family (Φji, Φ′ ji)i∈Ij via the scheme Hj, then (Φ, Φ′) is a conjunctive ∞-minor of the nonempty family (Φji, Φ′ ji)j∈J, i∈Ij via the composite scheme K := H(Hj j ∈ J). 14 MIGUEL COUCEIRO AND ERKKO LEHTONEN Corollary 3.11. Let W ⊆ WA be a set of finitary systems of pointed multisets, and let W ∞ be its conjunctive ∞-minor closure. If W is closed under formation of conjunctive minors, then W is the set of all finitary systems belonging to W ∞. We are now ready to prove the key result needed in the proof of Theorem 3.9. Lemma 3.12. Let A be an arbitrary, possibly infinite nonempty set. Let W ⊆ WA be a locally closed set of finitary systems of pointed multisets that contains the binary equality system, the unary empty system, and all unary trivial systems of breadth p ≥ 0, and is closed under formation of conjunctive minors, unions, quotients, and dividends. Let W ∞ be the conjunctive ∞-minor closure of W. Let (Φ, Φ′) ∈ WA\W be finitary. Then there exists a function in OA which preserves every system in W ∞ but does not preserve (Φ, Φ′). Proof. We shall construct a function g that preserves all systems in W ∞ but does not preserve (Φ, Φ′). Note that, by Corollary 3.11, (Φ, Φ′) cannot be in W ∞. Let m be the arity of (Φ, Φ′). Since W is locally closed and (Φ, Φ′) /∈ W, there is an integer p such that (Φ, Φ′)(p) := (Φ, Φ′) ∩ Ω(p) m /∈ W; let n be the smallest such integer. By Lemma 3.8, each function not preserving (Φ, Φ′)(n) does not preserve (Φ, Φ′) either, so we can consider (Φ, Φ′)(n) instead of (Φ, Φ′). Due to the minimality of n, the breadth of (Φ, Φ′)(n) is n. Observe that (Φ, Φ′) is not the trivial system of breadth n nor the empty system, because these are members of W. Thus, n ≥ 1. We can assume that (Φ, Φ′) is a minimal nonmember of W with respect to identification of rows, i.e., every simple minor of (Φ, Φ′) obtained by identifying some rows of (Φ, Φ′) is a member of W. If this is not the case, then we can identify some rows of (Φ, Φ′) to obtain a minimal nonmember ( Φ, Φ′) of W and consider the cluster ( Φ, Φ′) instead of (Φ, Φ′). Note that by Lemma 3.2, each function not preserving ( Φ, Φ′) does not preserve (Φ, Φ′) either. We can also assume that (Φ, Φ′) is a minimal nonmember of W with respect to taking quotients, i.e., whenever S 6= ε, we have that (Φ, Φ′)/S ∈ K. If this is not the case, then consider a minimal nonmember (Φ, Φ′)/S of W instead of (Φ, Φ′). By Lemma 3.6, each function not preserving (Φ, Φ′)/S does not preserve (Φ, Φ′) either. The fact that (Φ, Φ′) is a minimal nonmember of W with respect to taking quotients implies that Ω(1) m ⊆ (Φ, Φ′). Since all quotients (Φ, Φ′)/S where S ≥ 1 are in W and W is closed under dividends, we have that (Φ, Φ′) ∈ W, a contradiction. m 6⊆ (Φ, Φ′). For, suppose, on the contrary, that Ω(1) Let (Ψ, Ψ′) := S{(P, P ′) ∈ W (P, P ′) ⊆ (Φ, Φ′)}, i.e., (Ψ, Ψ′) is the largest system in W such that (Ψ, Ψ′) ⊆ (Φ, Φ′). Note that this is not the empty union, because the empty system is a member of W. It is clear that (Ψ, Ψ′) 6= (Φ, Φ′). Furthermore, Ψ ( Φ, for if it were the case that Ψ = Φ, then (Φ, Φ′) would be a conjunctive minor of (Ψ, Ψ′) by extending the consequent and hence (Φ, Φ′) would be a member of W, a contradiction. Since n was chosen to be the smallest integer satisfying (Φ, Φ′)(n) /∈ W, we have that (Φ, Φ′)(n−1) ∈ W and since (Φ, Φ′)(n−1) ⊆ (Φ, Φ′)(n), it holds that (Φ, Φ′)(n−1) ⊆ (Ψ, Ψ′). Thus there is a multiset Q ∈ Φ \ Ψ with Q = n. Let D := (d1, . . . , dn) be an m × n matrix whose multiset of columns equals Q. The rows of D are pairwise distinct. Suppose, for the sake of contradiction, that rows i and j of D coincide. Since (Φ, Φ′) is a minimal nonmember of W with GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 15 respect to identification of rows, by identifying rows i and j of (Φ, Φ′), we obtain a system ( Φ, Φ′) that is in W. By adding a dummy row in the place of the row that got deleted when we identified rows i and j, and finally by intersecting with the conjunctive minor of the binary equality cluster whose rows i and j are equal (the overall effect of the operations performed above is the selection of exactly those multisets in Φ and pointed multisets in Φ′ whose rows i and j coincide), we obtain a system ( ¯Φ, ¯Φ′) ∈ W such that Q ∈ ¯Φ and ( ¯Φ, ¯Φ′) ⊆ (Φ, Φ′). But this is impossible by the choice of Q. Let ( Φ, Φ′) := (Φ, Φ′) ∪ Ω(1) m . We claim that for S 6= ε, ( Φ, Φ′)/S = (Φ, Φ′)/S or ( Φ, Φ′)/S = (Φ, Φ′)/S ∪ Ω(0) m . For, by Lemma 3.5 we have ( Φ, Φ′)/S = ((Φ, Φ′) ∪ Ω(1) m )/S = (Φ, Φ′)/S ∪ Ω(1) m /S. If S > 1, then Ω(1) ({ε}, ∅) = Ω(0) m ; hence ( Φ, Φ′)/S = (Φ, Φ′)/S ∪ Ω(0) m . m /S = (∅, ∅); hence ( Φ, Φ′) = (Φ, Φ′). If S = 1, then Ω(1) m /S = Since (Φ, Φ′) is a minimal nonmember of W with respect to quotients, Ω(0) m ∈ W and W is closed under unions, by the above claim we have that ( Φ, Φ′)/S ∈ W whenever S ≥ 1. Since W is closed under dividends, we have that ( Φ, Φ′) ∈ W. Let (Υ, Υ′) := T{(P, P ′) ∈ W Q ∈ P }, i.e., (Υ, Υ′) is the smallest system in W such that Q ∈ Υ. Note that this is not the empty intersection, because ( Φ, Φ′) is a member of W, as shown above, and Q ∈ Φ; thus (Υ, Υ′) ⊆ ( Φ, Φ′). We claim that Υ′ 6⊆ Φ′. Suppose, on the contrary, that Υ′ ⊆ Φ′. Then we must have that Φ 6⊆ Υ. For, if it were the case that Φ ⊆ Υ, then (Φ, Φ′) would be a conjuctive minor of (Υ, Υ′) (by restricting the antecedent and extending the consequent) and hence (Φ, Φ′) would be a member of W, a contradiction. Consider (Λ, Λ′) := (Ψ, Ψ′) ∪ (Υ ∩ Φ, Υ′). Let us first verify that the pair (Υ ∩ Φ, Υ′) is actually a system of pointed multisets. Since (Υ, Υ′) ∈ WA, it holds that Υ′ is downward closed and {x} ⊎ S ∈ Υ for every (x, S) ∈ Υ′. By the assumption that Υ′ ⊆ Φ′ and by the fact that (Φ, Φ′) ∈ WA, it also holds that {x} ⊎ S ∈ Φ for every (x, S) ∈ Υ′. Indeed, (Υ ∩ Φ, Υ′) ∈ WA as claimed. The system (Υ ∩ Φ, Υ′) is a conjunctive minor of (Υ, Υ′) ∈ W (by restricting the antecedent) and hence it is a member of W. Since (Ψ, Ψ′) is also a member of W and W is closed under unions, we have that (Λ, Λ′) ∈ W. We have that (Ψ, Ψ′) ( (Λ, Λ′) ( (Φ, Φ′), where the first inclusion clearly holds by the definition of (Λ, Λ′), and the inclusion is strict, because Q ∈ Υ ∩ Φ but Q /∈ Ψ. The second inclusion holds, because (Ψ, Ψ′) ⊆ (Φ, Φ′) by the definition of (Ψ, Ψ′), Υ ∩ Φ ⊆ Φ by the definition of intersection and Υ′ ⊆ Φ′ by our assumption. The second inclusion is strict, because (Λ, Λ′) ∈ W but (Φ, Φ′) /∈ W. We have reached a contradiction, because (Ψ, Ψ′) is by definition the largest member of W that is componentwise included in (Φ, Φ′). This completes the proof of the claim that Υ′ 6⊆ Φ′. We have shown above that Υ′ ⊆ Φ′ but Υ′ 6⊆ Φ′. We conclude that there exists an m-tuple s ∈ Am such that (s, ε) ∈ Υ′ \ Φ′. Let M := (m1, . . . , mn) be a µ × n matrix whose first m rows are the rows of D (i.e., (cid:0)m1(i), . . . , mn(i)(cid:1) = (cid:0)d1(i), . . . , dn(i)(cid:1) for every i ∈ m) and whose other rows are the remaining distinct n-tuples in An; every n-tuple in An is a row of M 16 MIGUEL COUCEIRO AND ERKKO LEHTONEN and there is no repetition of rows in M. Note that m ≤ µ and µ is infinite if and only if A is infinite. Let (Θ, Θ′) := T{(P, P ′) ∈ W ∞ M ≺ P }. There must exist a µ-tuple u := (ut t ∈ µ) in Aµ such that u(i) = s(i) for all i ∈ m and (u, ε) ∈ Θ′. For, suppose that this is not the case. Let ( Θ, Θ′) be the projection of (Θ, Θ′) to its first m coordinates. Then ( Θ, Θ′) ∈ W and Q ∈ Θ but (s, ε) /∈ Θ′. This contradicts the choice of s. We can now define a function g : An → A by the rule gM = u. The definition is valid, because every n-tuple in An occurs exactly once as a row of M. It is clear that g 6⊲ (Φ, Φ′), because D ≺ Φ but gD = s ⊀ Φ′. We need to show that every system in W ∞ is preserved by g. Suppose, on the 0) ∈ W ∞, possibly infinitary, which ) ≺ Φ0, 0. Let 1) ∈ W. We 1) either. Define contrary, that there is a ρ-ary system (Φ0, Φ′ is not preserved by g. Thus, for some ρ × n′ matrix N := (c1, . . . , cn′ with N0 := (c1, . . . , cn), N1 := (cn+1, . . . , cn′ (Φ1, Φ′ have that N0 ≺ Φ1 but gN0 ⊀ Φ′ h : ρ → µ to be any map such that 1. Since W is closed under quotients, (Φ1, Φ′ 1, so g does not preserve (Φ1, Φ′ ), we have [gN0N1] ⊀ Φ′ 1) := (Φ0, Φ′ 0)/N∗ (cid:0)c1(i), . . . , cn(i)(cid:1) =(cid:0)(m1h)(i), . . . , (mnh)(i)(cid:1) for every i ∈ ρ, i.e., row i of N0 is the same as row h(i) of M, for each i ∈ ρ. Let (Φh, Φ′ 1) via H := {h}. Note that (Φh, Φ′ h) be the µ-ary simple ∞-minor of (Φ1, Φ′ h) ∈ W ∞. We claim that M ≺ Φh. To prove this, by the definition of simple ∞-minor, it is enough to show that (m1h, . . . , mnh) ≺ Φ1. In fact, we have for 1 ≤ j ≤ n, mjh = (mjh(i) i ∈ ρ) = (cj(i) i ∈ ρ) = cj, and (c1, . . . , cn) = N0 ≺ Φ1. Next we claim that (u, ε) /∈ Φ′ it is enough to show that uh ⊀ Φ′ h. For this, by the definition of simple ∞-minor, 1. For every i ∈ ρ, we have (uh)(i) =(cid:0)g(m1, . . . , mn)h(cid:1)(i) Thus uh = gN0. Since gN0 ⊀ Φ′ = g(cid:0)(m1h)(i), . . . , (mnh)(i)(cid:1) = g(cid:0)c1(i), . . . , cn(i)(cid:1). 1, we conclude that (u, ε) /∈ Φ′ h. Thus, (Φh, Φ′ h) ∈ W ∞, M ≺ Φh but (u, ε) /∈ Φ′ impossible, and we have reached a contradiction. h. By the choice of u, this is (cid:3) Proof of Theorem 3.9. (ii) =⇒ (i): It is clear that every function preserves the equality, empty, and trivial systems. By Lemmas 3.2, 3.4, 3.6, and 3.7, W is closed under formation of conjunctive minors, unions, quotients, and dividends. It remains to show that W is locally closed. Suppose on the contrary that there is a system (Φ, Φ′) ∈ WA \W, say of arity m, such that (Φ, Φ′)(p) = (Φ, Φ′)∩Ω(p) m ∈ W for all p ≥ 0. By (ii), there is an operation f : An → A that preserves every system in W but does not preserve (Φ, Φ′). Thus, there is a p ≥ 0 and an m × p matrix M := [M1M2] ≺ Φ such that [f M1M2] ⊀ Φ′. By our assumption, (Φ, Φ′)(p) ∈ W, but we have that [M1M2] ≺ Φ(p) and [f M1M2] ⊀ Φ′(p), which is a contradiction to the fact that f ⊲ (Φ, Φ′)(p). (i) =⇒ (ii): By Lemma 3.12, for every system (Φ, Φ′) ∈ WA \ W, there is a function in OA which preserves every system in W but does not preserve (Φ, Φ′). GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 17 The set of these "separating" functions, for each (Φ, Φ′) ∈ WA \ W, characterizes W. (cid:3) 4. The number of closed sets In this section, we will show that the closure system of the subalgebras of (OA; ζ, τ, ∇, ∗) is uncountable whenever A ≥ 2. We will first recall basic no- tions related to terms and term operations, following the notation and terminology presented in [5]. For a natural number n ≥ 1, let Xn := {x1, . . . , xn} be a set of variables. Let {fi i ∈ I} be a set of operation symbols, disjoint from the variables, and assign to each operation symbol fi a natural number ni, called the arity of fi. The sequence τ := (ni)i∈I is called a type. The n-ary terms of type τ are defined in the following inductive way: (i) Every variable xi ∈ Xn is an n-ary term. (ii) If fi is an ni-ary operation symbol and t1, . . . , tni are n-ary terms, then fi(t1, . . . , tni) is an n-ary term. (iii) The set Wτ (Xn) of all n-ary terms is the smallest set which contains the variables x1, . . . , xn and which is closed under the finite application of (ii). Every n-ary term is also an m-ary term for every m ≥ n. Let X :=Sn≥1 Xn = {x1, x2, . . . }. We denote by Wτ (X) the set of all terms of type τ over the countably infinite alphabet X: Wτ (X) := [n≥1 Wτ (Xn). A term is linear, if it contains no multiple occurrences of the same variable. We τ (Xn) the set of all n-ary linear terms of type τ over the alphabet denote by W lin Xn, and we denote by W lin τ (X) the set of all linear terms of type τ over X. The number of occurrences of operation symbols in a term is called the complexity of the term. A term s is a subterm of a term t if t = usv for some words u and v. The subterms of a linear term are linear. Let A = (A; (fi)i∈I ) be an algebra of type τ , i.e., each fundamental operation fi has arity ni, and let t be an n-ary term of type τ over X. The term t induces an n-ary operation tA on A (see Definition 5.2.1 in [5]). We call an operation induced by a linear term a linear term operation. The set of all n-ary linear term operations of the algebra A is denoted by W lin τ (Xn)A, and the set of all finitary linear term operations of the algebra A is denoted by W lin τ (X)A. For a set F ⊆ OA, the universe of the subalgebra of (OA; ζ, τ, ∇, ∗) generated by F is denoted by hF i. The following theorem shows that linear terms are related to subuniverses of (OA; ζ, τ, ∇, ∗) much in the same way as terms are related to clones. Theorem 4.1. Let A = (A; (f A the set of all linear terms of type τ over X. Then W lin (OA; ζ, τ, ∇, ∗) that contains all projections on A. Moreover, W lin i ∈ I} ∪ EAi. τ (X) be τ (X)A is a subuniverse of τ (X)A = h{fi i )i∈I ) be an algebra of type τ , and let W lin τ (X)A contains all projections on A. Let f, g ∈ W lin Proof. Since for all 1 ≤ i ≤ n, xi ∈ Xn, we have xA W lin is m-ary. Then there exist linear terms t ∈ W lin f A = f , sA = g. Then i = en,A i ∈ W lin τ (X)A. Thus τ (X)A, say f is n-ary, g τ (Xm) such that τ (Xn), s ∈ W lin 18 MIGUEL COUCEIRO AND ERKKO LEHTONEN • t(x2, x3, . . . , xn, x1) ∈ W lin • t(x2, x1, x3, . . . , xn) ∈ W lin • t(x2, . . . , xn+1) ∈ W lin • t(s, xm+1, xm+2, . . . , xm+n−1) ∈ W lin t(s, xm+1, xm+2, . . . , xm+n−1)A = f ∗ g. τ (Xn) and t(x2, x3, . . . , xn, x1)A = ζf , τ (Xn) and t(x2, x1, x3, . . . , xn)A = τ f , τ (Xn+1) and t(x2, . . . , xn+1)A = ∇f , τ (Xm+n−1) and Thus, ζf, τ f, ∇f, f ∗ g ∈ W lin (OA; ζ, τ, ∇, ∗). τ (X)A. Therefore, W lin τ (X)A is a subuniverse of i ∈ I} ∪ EA ⊆ W lin It is clear that {f A i ∈ I} ∪ EAi ⊆ i τ (X)A. We will show the converse inclusion by induction on the complexity ∈ h{f A i ∈ I} ∪ EAi. Otherwise i τ (X)A. Then there exist numbers W lin of a term t. If t = xi ∈ Xn, then tA = en,A t = fi(t1, . . . , tni ) is a linear term and tA ∈ W lin m1, . . . , mni ≥ 1 and an injective map τ (X)A, and so h{f A i i σ : {(j, k) ∈ ω × ω 1 ≤ j ≤ ni, 1 ≤ k ≤ mj} → {1, . . . , n} such that the variables occurring in the linear term tj (1 ≤ j ≤ ni) are precisely xσ(j,1), . . . , xσ(j,mj ). For 1 ≤ j ≤ ni, let uj be the mj-ary term that is obtained by replacing the occurrence of xσ(j,ℓ) by xℓ for each 1 ≤ ℓ ≤ mj. Note that then we clearly have that tA j = uA j (en,A σ(j,1), . . . , en,A σ(j,mj )) = uj(xσ(j,1), . . . , xσ(j,mj ))A. It is clear that f A holds that uA ∗ show that the functions 1 , . . . , uA i ∈ h{f A i ni ∈ h{f A i i ∈ I} ∪ EAi, and by our induction hypothesis it also i ∈ I} ∪ EAi. Then repeated applications of ζ and ζf A i , ζf A i ∗ uA ni, i ∗ uA ζ(ζf A ni), ni) ∗ uA i ∗ uA ζ(ζf A ni) ∗ uA i ∗ uA ζ(ζ(ζf A ni) ∗ uA i ∗ uA ζ(ζ(ζf A ni−1, ni−1), ni−1) ∗ uA ni−2, ... ζ(ζ(· · · (ζ(ζf A i ∗ uA ni) ∗ uA ni−1) ∗ · · · ) ∗ uA 2 ) ∗ uA 1 GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 19 are all in h{f A i i ∈ I} ∪ EAi. Note that ζf A ζf A i (a1, . . . , ani ) = f A i ∗ uA ni(a1, . . . , amni +ni−1) = i (a2, . . . , ani, a1), ζ(ζf A ζ(ζf A )), ni(a1, . . . , amni f A i (amni +1, . . . , amni +ni−1, uA i ∗ uA ni)(a1, . . . , amni +ni−1) = f A i (amni +2, . . . , amni +ni−1, a1, uA i ∗ uA f A i (amni−1+mni +1, . . . , amni−1+mni +ni−2, ni−1(a1, . . . , amni−1+mni +ni−2) = ni) ∗ uA ni(a2, . . . , amni +1)), ni−1(a1, . . . , amni−1), uA uA ni(amni−1+1, . . . , amni−1+mni )), ζ(ζ(· · · (ζ(ζf A i ∗ uA ni−1) ∗ · · · ) ∗ uA 2 ) ∗ uA 1 (a1, . . . , am1+m2+···+mni ) = ... ni) ∗ uA 1 (a1, . . . , am1 ), uA f A i (uA 2 (am1+1, . . . , am1+m2 ), . . . , uA ni(am1+m2+···+mni−1+1, . . . , am1+m2+···+mni )). Furthermore, repeated applications of ζ, τ and ∇ yield that the n-ary function g given by g(a1, . . . , an) = f A i (uA 1 (aσ(1,1), . . . , aσ(1,m1)), . . . , uA ni(aσ(ni,1), . . . , aσ(ni,mni ))) is in h{f A i I} ∪ EA ⊇ W lin i ∈ I} ∪ EAi. We clearly have that tA = g. Therefore {f A i τ (X)A, and the claimed equality holds. i ∈ (cid:3) Assume that 0 and 1 are distinct elements of A. For each integer n ≥ 3, define the function µn : An → A by µn(a1, . . . , an) =  1 0 if (a1, . . . , an) ∈ {0, 1}n and {i ∈ {1, . . . , n} : ai = 1} ∈ {1, n − 1}, otherwise. In the particular case that A = {0, 1}, the µn are the Boolean functions defined by Pippenger [14, Proposition 3.4]. Observe that µn(0, . . . , 0) = 0 for every n ≥ 3. Lemma 4.2. Let I ⊆ ω\{0, 1, 2} and k ∈ ω\{0, 1, 2}. Then µk ∈ h{µi i ∈ I}∪EAi if and only if k ∈ I. Proof. If k ∈ I, then obviously µk ∈ h{µi i ∈ I} ∪E{0,1}i. Assume then that k /∈ I. Let A = (A; (µi)i∈I ). By Theorem 4.1, it is enough to show that there is no linear term t of type τ := (i)i∈I such that tA = µk. Thus, let t be a k-ary linear term of type τ . It is clear that a k-ary linear term does not contain any operation symbols of arity greater than k, and since k /∈ I, the operation symbols occurring in t have arity less than k. If t = xj, then tA = ek,A , but clearly µk is not a projection. Otherwise, t has a subterm p of the form fℓ(xi1 , . . . , xiℓ ), where ℓ < k and i1, . . . , iℓ are pairwise distinct. Let x ∈ {0, 1}k be a k-tuple with 1's at exactly ℓ − 1 positions among i1, . . . , iℓ and 0's at all remaining positions. j Claim. For every subterm s of t that contains p as a subterm, sA(x) = 1. 20 MIGUEL COUCEIRO AND ERKKO LEHTONEN Proof of Claim. We first define the depth d(s, t) of a subterm s in the linear term t recursively as follows: d(t, t) = 0; and if s = fi(t1, . . . , tni) is a subterm of t with d(s, t) = d, then d(tj, t) = d + 1 for 1 ≤ j ≤ ni. We proceed by induction on the depth d of the subterm p in s. If d = 0 then s = p, and so sA(x) = pA(x) = µℓ(x) = 1. Assume that the claim holds for d = q for some q ≥ 0. Let then d = q + 1. Then s = fm(t1, . . . , tm) for some m < k, and there is an n ∈ {1, . . . , m} such that p is a subterm of ti and the depth of p in ti is d. By the induction hypothesis, tA n (x) = 1. Furthermore, we have that for all p 6= n, tA p (x) = tA p (0, . . . , 0) = 0, where the first equality holds because the variables xi1 , . . . , xiℓ do not occur in tp since t is a linear term; and the second equality holds because the fundamental operations of A preserve 0, and hence so do all term operations on A. Thus, sA(x) = f A m(x)) = µm(0, . . . , 0, 1, 0, . . . , 0) = 1, m(tA 1 (x), . . . , tA as claimed. ⋄ By the Claim, we have in particular that tA(x) = 1. However, since 3 ≤ ℓ < k, (cid:3) we have that 1 < ℓ − 1 < k − 1; hence µk(x) = 0. Theorem 4.3. Let A ≥ 2. (i) The set of subalgebras of (OA; ζ, τ, ∇, ∗) containing all projections is uncount- able. (ii) The set of subalgebras of (OA; ζ, τ, ∇, ∗) is uncountable. Proof. (i) By Lemma 4.2, if I, J ⊆ ω \ {0, 1, 2} and I 6= J, then h{µi i ∈ I} ∪EAi 6= h{µi i ∈ J}∪EAi. Thus, there are uncountably many subalgebras of (OA; ζ, τ, ∇, ∗) containing all projections. (ii) An immediate consequence of (i). (cid:3) Table 1 summarizes the Galois connections that describe closure systems of sub- algebras of various reducts of (OA; ζ, τ, ∆, ∇, ∗) considered in the literature, up to our knowledge. It is well-known that the closure system of line 2 of Table 1 (and, in particular, that of line 1) is countably infinite in the case A = 2 (see, e.g., [10]). Pippenger [14] showed that the closure system of line 3 is uncountable whenever A ≥ 2. By Theorem 4.3, this is also the case for the closure system of line 5. From this it follows that the closure systems of lines 4 and 6 are uncountable as well whenever A ≥ 2. Looking at possible directions for future work, we are inevitably drawn to con- sider the remaining reducts of (OA; ζ, τ, ∆, ∇, ∗). This asks for analogous descrip- tions of the subalgebras of these reducts in terms of Galois connections and the sizes of the respective closure systems. Acknowledgements. The authors wish to express their gratitude to Ivo Rosen- berg for suggesting the problem of finding a characterization of the closure system of subalgebras of (OA; ζ, τ, ∇, ∗). References [1] V. G. Bodnarcuk, L. A. Kaluznin, V. N. Kotov, B. A. Romov, Galois theory for Post algebras. I, II (in Russian), Kibernetika 3 (1969) 1 -- 10, 5 (1969) 1 -- 9. English translation: Cybernetics 5 (1969) 243 -- 252, 531 -- 539. GALOIS CONNECTION FOR CLOSED SETS OF OPERATIONS 21 algebra dual objects reference (OA; ζ, τ, ∆, ∇, ∗) -- subalgebras with projections (clones) relations R -- all subalgebras relation pairs (R, R′) with R′ ⊆ R (OA; ζ, τ, ∆, ∇) constraints (R, S) 1 2 3 4 (OA; ζ, τ, ∇) generalized constraints (φ, S) Geiger [7]; Bodnarcuk, Kaluznin, Kotov, Romov [1] (finite domains), Szab´o [18]; Poschel [15] (general) Harnau [8] (finite domains) Pippenger [14] (finite domains), Couceiro, Foldes [2] (general) Hellerstein [9] (finite domains), Lehtonen [11] (general) (OA; ζ, τ, ∇, ∗) -- subalgebras with projections -- all subalgebras 5 6 clusters Φ Lehtonen [11] (general) systems of pointed multisets (Φ, Φ′) Theorems 2.4, 3.9 Table 1. Galois theories for function algebras. [2] M. Couceiro, S. Foldes, On closed sets of relational constraints and classes of functions closed under variable substitution, Algebra Universalis 54 (2005) 149 -- 165. [3] M. Couceiro, E. Lehtonen, Classes of operations closed under permutation, cylindrification and composition, 40th IEEE International Symposium on Multiple-Valued Logic (ISMVL 2010), IEEE Computer Society, Los Alamitos, 2010, pp. 117 -- 121. [4] K. Denecke, M. Ern´e, S. L. Wismath (eds.), Galois Connections and Applications, Kluwer Academic Publishers, Dordrecht, 2004. [5] K. Denecke, S. L. Wismath, Universal Algebra and Applications in Theoretical Computer Science, Chapman & Hall/CRC, Boca Raton, 2002. [6] C. J. Everett, Closure operations and Galois theory in lattices, Trans. Amer. Math. Soc. 55 (1944) 514 -- 525. [7] D. Geiger, Closed systems of functions and predicates, Pacific J. Math. 27 (1968) 95 -- 100. [8] W. Harnau, Ein verallgemeinerter Relationenbegriff fur die Algebra der mehrwertigen Logik, Teil I (Grundlagen), Teil II (Relationenpaare), Teil III (Beweis), Rostock. Math. Kolloq. 28 (1985) 5 -- 17, 31 (1987) 11 -- 20, 32 (1987) 15 -- 24. [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, Closed classes of functions, generalized constraints and clusters, Algebra Uni- versalis (2010), doi:10.1007/s00012-010-0071-6. [12] A. I. Mal'cev, Iterative Post Algebras (in Russian), Nauka, Novosibirsk, 1976. [13] Ø. Ore, Galois connexions, Trans. Amer. Math. Soc. 55 (1944) 493 -- 513. [14] N. Pippenger, Galois theory for minors of finite functions, Discrete Math. 254 (2002) 405 -- 419. [15] R. Poschel, Concrete representation of algebraic structures and a general Galois theory. In: H. Kautschitsch, W. B. Muller, W. Nobauer (eds.), Contributions to General Algebra (Proc. Klagenfurt Conf., 1978), Verlag Johannes Heyn, Klagenfurt, 1979, pp. 249 -- 272. 22 MIGUEL COUCEIRO AND ERKKO LEHTONEN [16] R. Poschel, A general Galois theory for operations and relations and concrete characteri- zation of related algebraic structures, Report R-01/80, Zentralinstitut fur Mathematik und Mechanik, Akademie der Wissenschaften der DDR, Berlin, 1980. [17] R. Poschel, L. A. Kaluznin, Funktionen- und Relationenalgebren: Ein Kapitel der diskreten Mathematik, Birkhauser, Basel, Stuttgart, 1979. [18] L. Szab´o, Concrete representation of related structures of universal algebras. I, Acta Sci. Math. (Szeged) 40 (1978) 175 -- 184. [19] ´A. Szendrei, 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. (M. Couceiro) Mathematics Research Unit, University of Luxembourg, 6, rue Richard Coudenhove-Kalergi, L-1359 Luxembourg, Luxembourg E-mail address: [email protected] (E. Lehtonen) Computer Science and Communications Research Unit, University of Luxembourg, 6, rue Richard Coudenhove-Kalergi, L-1359 Luxembourg, Luxembourg E-mail address: [email protected]
1003.2352	1	1003	2010-03-11T15:57:49	Lectures on extended affine Lie algebras	[ "math.RA", "math.QA", "math.RT" ]	We give an introduction to the structure theory of extended affine Lie algebras, which provide a common framework for finite-dimensional semisimple, affine and toroidal Lie algebras. The notes are based on a lecture series given during the Fields Institute summer school at the University of Ottawa in June 2009.	math.RA	math	LECTURES ON EXTENDED AFFINE LIE ALGEBRAS ERHARD NEHER Abstract. We give an introduction to the structure theory of extended affine Lie algebras, which provide a common framework for finite-dimensional semisim- ple, affine and toroidal Lie algebras. The notes are based on a lecture series given during the Fields Institute summer school at the University of Ottawa in June 2009. 0 1 0 2 r a M 1 1 ] . A R h t a m [ 1 v 2 5 3 2 . 3 0 0 1 : v i X r a Contents Introduction 1. Affine Lie algebras and some generalizations 1.1. Realization (construction of affine Kac-Moody Lie algebras) 1.2. Multiloop and toroidal Lie algebras 1.3. Appendix on central extensions of Lie algebras 2. Extended affine Lie algebras: Definition and first examples 2.1. Definition of an extended affine Lie algebra 2.2. Some elementary properties of extended affine Lie algebras 2.3. Extended affine Lie algebras of nullity 0 2.4. Affine Kac-Moody Lie algebras again 2.5. Higher nullity examples 3. The structure of the roots of an EALA 3.1. Affine reflection systems: Definition 3.2. Examples of affine reflection systems 3.3. The Structure Theorem of affine reflection systems 3.4. Extended affine root systems 4. The core and centreless core of an EALA 4.1. Lie tori: Definition 4.2. Some basic properties of Lie tori 4.3. The core of an EALA 4.4. Lie tori of type Al, l ≥ 3 4.5. Some more easy examples of Lie tori 5. The construction of all EALAs 5.1. Degree maps 5.2. The centroid of Lie algebras, in particular of Lie tori 5.3. Centroidal derivations of Lie tori 5.4. The general construction References 2 4 4 6 10 14 15 17 19 19 22 23 23 25 31 36 38 38 40 43 45 48 50 51 53 56 58 60 Date: May 28, 2018. 1991 Mathematics Subject Classification. Primary 17B. This work is partially supported by the Natural Sciences and Engineering Research Council (NSERC) of Canada through the author's Discovery Grant. 1 2 ERHARD NEHER Introduction Extended affine Lie algebras form a category of Lie algebras containing finite- dimensional semisimple, affine, toroidal and some other interesting classes of Lie algebras. Like finite-dimensional simple Lie algebras, extended affine Lie algebras are de- fined by a set of axioms prescribing their internal structure, rather than a potentially elusive presentation. The structure of extended affine Lie algebras is now well un- derstood, and is quite similar to the construction of affine Lie algebras: They are obtained from a generalized loop algebra, a so-called invariant Lie torus, by taking a central extension and adding some derivations: central extension of L (another Lie torus) add derivations extended affine Lie algebra invariant Lie torus L Invariant Lie tori have been classified. Although there are some rather sophisticated examples, many of them have a concrete matrix realization or can be described in terms of familiar objects like finite-dimensional simple Lie algebras and Laurent polynomial rings. This makes extended affine Lie algebras easily accessible. Since they are an emerging new area, there are many open questions, opportunities for research and applications, for example in physics. A short history of extended affine Lie algebras is given in section 2.1, in particular it describes the role physicists have played. The goal of these notes is to provide a survey of the structure theory of extended affine Lie algebras, accessible to graduate students. The emphasis is on examples, and not on an exposition containing all proofs. Such an exposition will appear elsewhere. Thus, while we have endeavored to present a complete picture of the theory by giving precise definitions and theorems, most of the proofs have been left out. But references to proofs are provided, as far as possible. Outline. Section 1 reviews the construction of affine Kac-Moody algebras and discusses some natural generalizations, like toroidal algebras. It also contains an exposition of central extensions of Lie algebras, which are crucial for the theory. The following section 2 starts with the definition of an extended affine Lie algebra and then presents some easily proven properties. We also give examples of extended affine Lie algebras: finite-dimensional split simple, affine Kac-Moody and untwisted multi-loop algebras. Part of the axioms for an extended affine Lie algebra is the existence of a root space decomposition. Section 3 describes the structure of the roots occurring in an extended affine Lie algebra, naturally called extended affine root systems. They turn out to be special types of so-called affine reflection systems. In section 4 we reverse the picture above: We start with an extended affine Lie algebra and, using the structure of affine reflection systems, we associate to it a graded ideal, the so-called core, and its central quotient, the centreless core. Both are Lie tori. This section also presents properties of Lie tori and examples. Finally, in section 5 we survey the general construction of extended affine Lie algebras, as summarized in the picture above. / / / o / o / o / o / o / o / o / o / o / o / o O O O O O LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 3 Prerequisites. We assume that the reader is familiar with the basic structure theory of complex finite-dimensional semisimple Lie algebras, as for example de- veloped in [Hu]. Some familiarity with affine Kac-Moody algebras, e.g. chapters 7 and 8 of [Kac], is helpful but not essential, since section 1.1 will give a short review of the necessary background. Similarly, knowing split simple Lie algebras will facilitate reading the notes, but is not required. A short summary of the facts used here is presented in section 2.3. Notation and setting. With some rare exceptions (in 1.1, 5.1 and 5.2), all vector spaces and algebras are defined over a field F of characteristic 0. We will not assume that F is algebraically closed , since this is not needed and would not do proper justice to the theory to be explained here. Thus, F could be, but need not be the field C of complex numbers or the field R of real numbers or the field of rational numbers Q or ... Unless specified otherwise, linear maps will always be F -linear. All unadorned tensor products will be over F . The symbol g will always denote a split simple finite-dimensional Lie algebra. We let Z(L) = {z ∈ L : [z, L] = 0} denote the centre of a Lie algebra L. We will say that L is centreless if Z(L) = 0. If K is a subspace of a Lie algebra E, the centralizer of K in E is CE(K) = {c ∈ E : [c, K] = 0}. With the exception of some remarks, all algebras will be associative or Lie al- gebras. For an F -algebra A we denote by DerF (L) the Lie algebra of all deriva- tions of L (recall that an F -linear map d : L → L is a derivation if d([l1, l2]) = [d(l1), l2] + [l1, d(l2)] holds for all l1, l2 ∈ L). The algebras considered here will often be graded by some abelian group, usually denoted Λ and always written additively. A Λ-grading of a vector space V by the is such a Λ-graded vector space. Then the Λ-support of V is defined as suppΛ V = {λ ∈ Λ : V λ 6= 0}. A graded subspace of V is a subspace U of V satisfying abelian group Λ is a decomposition V =Lλ∈Λ V λ into subspaces V λ. Suppose V U =Lλ(U ∩ V λ). We will say that V has finite bounded dimension if there exists a constant M such that for all λ ∈ Λ we have dim V λ ≤ M . Note that this is a stronger condition than requiring that V has finite homogeneous dimension, which by definition just means that every V λ, λ ∈ Λ, is finite-dimensional. F -linear map f : V → W has degree λ if f (V µ) ⊂ W λ+µ holds for all µ ∈ Λ. We denote by HomF (V, W )λ the linear maps of degree λ and put Given two Λ-graded vector spaces V = Lλ V λ and W = Lλ W λ, we say an grHomF (V, W ) =Lλ∈Λ HomF (V, W )λ We note that grEndF (V ) is a Λ-graded associative algebra with respect to compo- sition of maps. We give F the trivial grading F = F 0 and define the graded dual space of V as and grEndF (V ) = grHomF (V, V ). V gr∗ = grHomF (V, F ) =Lλ∈Λ(V gr∗)λ. Observe that (V gr∗)λ consists of those linear forms ϕ : V → F which satisfy ϕ(V µ) = 0 whenever λ + µ 6= 0 and can therefore be identified with the usual dual space (V −λ)∗. Given a symmetric bilinear form on a vector space V , an endomorphism d of V is called skew-symmetric if (d(v) v) = 0 for all v ∈ V . Since we assume that our base field has characteristic 0, this is equivalent to the condition (d(v1) v2) + (v1 d(v2)) = 0 for all v1, v2 ∈ V . A bilinear form is nondegenerate if (v u) = 0 for all u ∈ V implies v = 0. 4 ERHARD NEHER If A is an algebra, a Λ-grading of the algebra A is a Λ-grading of the underlying vector space A, say A =Lλ∈Λ Aλ, for which in addition AλAµ ⊂ Aλ+µ holds for all λ, µ ∈ Λ. Since we will often deal with algebras with two gradings, it is convenient to use superscripts and subscripts to distinguish them. These notes grew out of my notes for a lecture series during the Fields Insti- tute summer school on Geometric Representation Theory and Extended Affine Lie Algebras, held at the University of Ottawa in June 2009. I would like to thank all the participants of the summer school for their interest and questions. I also thank Bruce Allison and Juana S´anchez Ortega for their careful reading of an earlier version of these notes. 1. Affine Lie algebras and some generalizations We will always assume that F is a field of characteristic 0. Occasionally we will need some roots of unity in F , so certainly an algebraically closed field like C will do. We denote by g a split simple finite-dimensional Lie algebra over F . For example, if F is algebraically closed then this just means that g is a simple and finite- dimensional. Their structure theory is explained in most standard textbooks, for example in [Hu]. For more general fields, an example of a split simple g is the Lie algebra sln(F ) of n × n-matrices over F which have trace 0. These types of Lie algebra are investigated in [Bou3, Ch. VII], [D, Ch. 1] or [J, Ch. IV]. 1.1. Realization (construction of affine Kac-Moody Lie algebras). Let ζ ∈ F be a primitive mth root of 1. In other words, the multiplicative subgroup of F generated by ζ is isomorphic to Z/mZ. For example, in F = C we can take ζ = exp(2πi/m). Let σ be an automorphism of g of finite order m ∈ N. Thus, the subgroup hσi of the automorphism group of g is isomorphic to Z/mZ. For example, if g = sln(F ) an example of such an automorphism is σ(x) = axa−1, where a is an n × n-matrix of order m, and an example of such a matrix is a = ζEn where En is the n × n identity matrix. Observe that σ is diagonalizable. Indeed, its minimal polynomial divides the polynomial tm = 1 and therefore has no multiple roots in F . For a general field F this would of course only say that σ is a semisimple endomorphism. But since as we assumed that F contains all roots of unity which we need, σ is diagonalizable over F . To describe its eigenspaces we need some notation. In anticipation of the later developments we put and denote the canonical map Λ → ¯Λ by λ 7→ ¯λ. That σ is diagonalizable, means Λ = Z and ¯Λ = Z/mZ, (1.1) for g¯λ = {x ∈ g : σ(x) = ζ λx} g =L¯λ∈ ¯Λ g¯λ Of course, some of the g¯λ could be zero. The eigenspaces of σ are precisely the non-zero among the subspaces g¯λ. It is also appropriate to note that g¯λ is well- defined: if ¯λ = ¯µ then ζ λ = ζ µ. Finally we point out that the decomposition (1.1) is a ¯Λ-grading, which means that it satisfies (1.2) [g¯λ, g¯µ] ⊂ g¯λ+¯µ for all ¯λ, ¯µ ∈ ¯Λ. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 5 Let F [t±1] be the ring of Laurent polynomials. This is a unital associative commutative F -algebra with F -basis {tλ : λ ∈ Z} and multiplication rule tλtµ = tλ+µ. The loop algebra associated to the data (g, σ) is the Lie algebra (1.3) with product (1.4) L = L(g, σ) =Lλ∈Λ g¯λ ⊗ F tλ [u¯λ ⊗ tλ, v¯µ ⊗ tµ] = [u¯λ, v¯µ] ⊗ tλ+µ. We will sometimes use more precise terminology: If σ = Id, i.e., m = 1, we will call L(g, Id) = g ⊗ F [t±1] the untwisted loop algebra, and we will call L(g, σ) a twisted loop algebra if it is clear that σ 6= Id and we want to emphasize this. We point out that we consider L(g, σ) as a Lie algebra over F . It is therefore It is also important to note that L is a Λ-graded algebra, infinite-dimensional. whose homogenous spaces are Lλ = g¯λ ⊗ F tλ for λ ∈ Λ. For the reader with some background in algebraic geometry, a more geometric definition of L(g, σ) is the following: It is (isomorphic to) the Lie algebra of equivariant maps F × → g, where σ acts on F × by σ(x) = ζx.) Let κ be the Killing form of g, i.e., κ(u, v) = tr(ad u ◦ ad v), and define (1.5) ψ : L × L → F, ψ(u ⊗ tλ, v ⊗ tµ) = λ δλ,−µ κ(u, v) where δλ,−µ is the Kronecker delta: otherwise. It has the value 1 if λ = −µ and is zero Exercise 1.1. Check that the map ψ of (1.5) is a 2-cocycle of L, i.e. an F -bilinear map satisfying (1.6) ψ(l, l) = 0 = ψ([l1, l2], l3) + ψ([l2, l3], l1) + ψ([l3, l1], l2) for l, li ∈ L. A consequence of Exercise 1.1 is that we can enlarge our Lie algebra L by ad- joining a 1-dimensional space, denoted F c here: (1.7) L = L(g, σ) = L(g, σ) ⊕ F c is a Lie algebra over F with respect to the product [l1 ⊕ s1c, l2 ⊕ s2c] L = [l1, l2]L ⊕ ψ(l1, l2)c for li ∈ L and si ∈ F . We have added subscripts on the products to emphasize where the product is calculated, in L or in L. It is obvious from the product formula, that it is important to know in which Lie algebra the product is being calculated. But in the future we will leave out the subscripts, if it is clear in which algebra the product is calculated. The equations (1.6) are exactly what is needed to make L a Lie algebra. The map u : L → L, u(l ⊕ sc) = l is a surjective Lie algebra homomorphism with kernel Ker(u) = F c = Z( L), the centre of L. In other words, u is a central extension (see 1.3 for a short review of central extensions). In fact, u is the "biggest" central extension, the so-called universal central extension, see [G] and [Wi] for a proof. 6 ERHARD NEHER The Lie algebra L has a canonical derivation d, the so-called degree derivation (1.8) Hence we can form the semidirect product L = L(g, σ) = L ⋊ F d with product d(cid:0)(u ⊗ tλ) ⊕ sc(cid:1) = λu ⊗ tλ, (λ ∈ Z, u ∈ g¯λ, s ∈ F ). [l1 ⊕ s1d, l2 ⊕ s2d] L = [l1, l2] L + s1d(l2) − s2d(l1) for li ∈ L and si ∈ F . In untangled form, is the Lie algebra with product L =(cid:0)Lλ∈Z(g¯λ ⊗ F tλ)(cid:1) ⊕ F c ⊕ F d (1.9) (1.10) [u¯λ ⊗ tλ ⊕ s1c ⊕ s′ 1d, v¯µ ⊗ tµ ⊕ s2c ⊕ s′ 2d] 1v¯µ ⊗ tµ − λs′ =(cid:0)[u¯λ, v¯µ] ⊗ tλ+µ + µs′ 2u¯λ ⊗ tλ(cid:1) ⊕ λ δλ,−µ κ(u¯λ, v¯µ) c. Exercise 1.2. Show [ L, L] = L and Z( L) = F c = Z( L). The importance of the Lie algebras L(g, σ) stems from the following. Theorem 1.3. (Realization Theorem [Kac, Th. 7.4, Th. 8.3, Th. 8.5]) Suppose F is algebraically closed. (a) The Lie algebra L(g, σ) is an affine Kac-Moody Lie algebra, and every affine Kac-Moody Lie algebra is isomorphic (as F -algebra) to some L(g, σ). (b) L(g, σ) ∼= L(g, σ′) where σ′ is a diagram automorphism with respect to some Cartan subalgebra of g. We note that diagram automorphisms have order 1, 2 or 3, with the latter case only occurring for g of type D4. 1.2. Multiloop and toroidal Lie algebras. We will discuss some (straightfor- ward) generalizations of L = L(g, σ), the central extension L and the big Lie algebra L. The first idea is to replace the Laurent polynomial ring F [t±1] by a ring with similar properties. Instead of one variable we will use the Laurent polynomial ring F [t±1 n ] in n variables. This ring has indeed very similar properties to the ring F [t±1]. We put Λ = Zn and define 1 , . . . , t±1 tλ = tλ1 1 · · · tλn n for λ = (λ1, . . . , λn) ∈ Λ 1 , . . . , t±1 1 , . . . , t±1 Then {tλ : λ ∈ Λ} is an F -basis of F [t±1 n ] and the multiplication rule in F [t±1 n ] is tλtµ = tλ+µ, which is the "same" as in the 1-variable case. Also, F [t±1 n ] is still a unital commutative associative F -algebra. We can therefore define the untwisted multiloop algebra, the "several variable" generalization of the untwisted loop algebra of 1.1 as 1 , . . . , t±1 (1.11) L(g) = g ⊗ F [t±1 1 , . . . , t±1 n ], which becomes a Lie algebra with respect to the product [u ⊗ tλ, v ⊗ tµ] = [u, v] ⊗ tλ+µ for u, v ∈ g and λ, µ ∈ Zn. We will meet this Lie algebra again in Example 4.31. To continue the analogy we let σ = (σ1, . . . , σn) be a family of n commuting finite order automorphisms of g, say σi has order mi ∈ N+. Let ζi ∈ F be a LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 7 primitive mi-th root of 1 (recall that we assumed that F has an ample supply of them). We put ¯Λ = (Z/m1Z) ⊕ · · · ⊕ (Z/mnZ) and let λ 7→ ¯λ be the obvious map. The automorphisms σi are simultaneously diagonalizable: (1.12) As in the one-variable case, the decomposition (1.12) is a ¯Λ-grading: [g¯λ, g¯µ] ⊂ g¯λ+¯µ for ¯λ, ¯µ ∈ ¯Λ. It follows from this that g =L¯λ∈ ¯Λ g¯λ, g¯λ = {u ∈ g : σi(u) = ζ λi i u, 1 ≤ i ≤ n}. (1.13) L(g, σ) =Lλ∈Λ g¯λ ⊗ F tλ is a subalgebra of g⊗F [t±1 n ], called the multiloop algebra associated to g and σ. If all σi = Idg we will (of course) call it an untwisted multiloop algebra. Multiloop algebras are investigated in the papers [ABFP1], [ABFP2], [ABP1], [ABP2] and [ABP3]. 1 , . . . , t±1 Following our procedure in section 1.1 we should now make a central extension to get a bigger Lie algebra L and then add some derivations: (1.14) L = L ⊕ C add derivations L = L ⋊ D central extension L = L(g, σ) To define the Lie algebra product on L we would use a 2-cocycle ψ : L × L → C where C is some vector space and then put (1.15) for li ∈ L and ci ∈ C. The Lie algebra L should be a semidirect product with D acting on L by derivations. [l1 ⊗ c1, l1 ⊗ c2] L = [l1, l2]L ⊕ ψ(l1, l2) But here is where the problems start, or things become interesting depending on one's taste. In the one-variable case the 2-cocycle ψ of (1.5) was the only possible choice up to scalars, i.e., the universal central extension L of L had a 1-dimensional centre C = F c. This is no longer true in the case of several variables. It is not so surprising that there exists a 2-cocycle with values in F n: We can simply use the same formula as in (1.5). Exercise 1.4. Let L = L(g, σ) be a multiloop algebra and embed Λ ⊂ F n canon- ically. Then ψ : L × L → F n, given by (1.16) ψ(u ⊗ tλ, v ⊗ tµ) = δλ+µ,0 κ(u, v) λ , is a 2-cocycle of L. However, this is still not the "biggest" possible. Rather, the centre of the univer- sal central extension is infinite-dimensional and the so-called universal 2-cocycle, i.e., the 2-cocycle used in (1.15) to describe the universal central extension L of L, is described in the following result. / / 6 6 6 v 6 v 6 v 6 v 6 v 6 v 6 v 6 v 6 v 6 v 6 v 6 v 6 v 6 v 8 ERHARD NEHER Theorem 1.5 ([Ne6]). Let L = L(g, σ) be a multiloop algebra. We embed Λ ⊂ F n canonically, put Γ = m1Z ⊕ · · · ⊕ mnZ and let C =Lγ∈Γ Cγ where Cγ = F n/F γ. Then the universal 2-cocycle is ψu : L × L → C for which the γ-component of ψu is (1.17) ψu(u ⊗ tλ, y ⊗ tµ)γ = κ(u, v)δλ+µ,−γ ¯λ ∈ Cγ. Observe that (1.16) is just the 0-component of (1.17). The theorem is well-known in the untwisted case (all σi = Idg, so Γ = Λ), in which it can be deduced from the description of the universal central extension of the Lie algebra g ⊗ A where A is any unital commutative associative F -algebra, see [Kas] and [MRK]. (In these references the centre C of the universal central extension is described as ΩA/dA where ΩA is the module of Kahler differentials, which is also the same as the first cyclic homology group HC1(A).) In the untwisted case, the universal central extension L was termed the n-toroidal Lie algebra based on g. The reader should however be warned that this terminology is not standard. It is sometimes used for the Lie algebra L with the 2-cocycle of exercise 1.4, and sometimes also for the Lie algebras of the form L = L ⊕ C ⊕ D for an appropriate subalgebra D of derivations, e.g. in [DFP]. Thus, there are many possibilities for C in the diagram (1.14), and it is not clear which one is the best possible choice. (In fact, we will later allow any central extension). Assuming that we have settled for some C, which D should we take? For sim- plicity we will discuss this only in the untwisted case. If n = 1 we added the degree derivation d described in (1.8). This is far from being an arbitrary derivation. The full derivation algebra of the Lie algebra g ⊗ A for A is described in [BM, Th. 1]: (1.18) DerF (g ⊗ A) =(cid:0) DerF (g) ⊗ A(cid:1) ⊕(cid:0)F Id ⊗ DerF (A)(cid:1) = IDer(g ⊗ A) ⊕ F Id ⊗ DerF (A). where DerF (g) ⊗ A and F ⊗ DerF (A) = F Id ⊗ DerF (A) act on g ⊗ A in the obvious way. Since g ⊗ A is perfect, up to a canonical isomorphism, this is then also the derivation algebra of the universal central extension of g ⊗ A (see for example [BM, Th. 2.2]). From Der F [t±1] = F [t±1]d we see that we added a rather special derivation, one which can be used to define the Λ-grading of L (see also Ex. 1.7). We can do something similar in multi-variable case. Define the i-th degree deriva- tion ∂i of L(g) ⊕ C by ∂i(u ⊗ tλ ⊕ c) = λi u ⊗ tλ for λ = (λ1, . . . , λn) ∈ Λ = Zn (1.19) and put the space of degree derivations. Possible (interesting) choices for D are: D = spanF {∂i : 1 ≤ i ≤ n}, (1) D = D, (2) F [t±1 1 , . . . , t±1 n ]D (in physics parlance: "all vector fields"), and (3) Lλ∈Λ F tλ{Pn i=1 si∂i :Pi si = 0} (the "divergence 0 vector fields"). LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 9 It will turn out that for the Lie algebras which we are going to study in the next chapters, the choices (1) and (3) are the correct ones. In addition, there will be a surprise: semidirect products in (1.14) will not be enough! Exercise 1.6. Recall that a bilinear form (··) on a Lie algebra L is called invariant if ([l1, l2] l3) = (l1 [l2, l3]) holds for all li ∈ L. Show: (a) The set IF(L) of invariant bilinear forms on L is a vector space with respect to the obvious scalar multiplication and addition defined by (β1 + β2)(l1, l2) = β1(l1, l2) + β2(l1, l2) for βi ∈ IF(L). (b) If L is perfect, any invariant bilinear form is symmetric. (c) Let S be a unital associative F -algebra. A bilinear form b on S is called invariant if b(s1s2, s3) = b(s1, s2s3) = b(s2, s3s1) for si ∈ S. (i) The set IF(S) of invariant bilinear forms on S is a vector space with respect to the obvious operations. (ii) Any linear form λ ∈ S∗ with λ([S, S]) = 0 gives rise to an invariant bilinear form bλ on S, defined by bλ(s1, s2) = λ(s1s2). (iii) The map (S/[S, S])∗ → IF(S), given by λ 7→ bλ, is a vector space isomor- phism. (d) Let L be a perfect Lie algebra with a 1-dimensional space IF(L), say IF(L) = F κ. Also, let S be a unital associative commutative F -algebra. We consider L ⊗ S as Lie algebra with respect to the product [l1 ⊗ s1, l2 ⊗ s2] = [l1, l2] ⊗ s2s2, cf. (1.5). For λ ∈ IF(S) define a bilinear form κ ⊗ λ on L ⊗ S by (κ ⊗ λ) (l1 ⊗ s2, l2 ⊗ s2) = κ(l1, l2) λ(s1, s2). Then κ ⊗ λ ∈ IF(L ⊗ S) and the map IF(S) → IF(L ⊗ S), given by λ 7→ κ ⊗ λ, is an isomorphism of vector spaces. Exercise 1.7. Define the ith degree derivation ∂i of the Laurent polynomial ring S = F [t±1 n ] by ∂i(tλ) = λitλ, so that the ∂i of (1.19) becomes ∂i(u ⊗ tλ) = u ⊗ ∂i(tλ) = (Id ⊗∂i)(u ⊗ tλ) (this double meaning of ∂i should not create any confusion). Show: 1 , . . . , t±1 (a) The derivation algebra DerF (S) of S is given by where, as above, D = spanF {∂i : 1 ≤ i ≤ n}. The derivation algebra is a Zn-graded Lie algebra with Lie algebra product determined by DerF (S) = SD =Lλ∈Zn F tλD [tλ∂i, tµ∂j] = tλ+µ(µi∂j − λj∂i). Thus, for n = 1 we obtain the usual Witt algebra, see for example [MP, 1.4]. (b) (tλ tµ) = δλ+µ,0 defines a nondegenerate symmetric bilinear form (··) on S which is invariant in the sense that (abc) = (a bc) for all a, b, c ∈ S. (c) Let SDerF (S) be the subalgebra of derivations of S, which are skew-sym- metric with respect to the form (··) of (b). Then SDerF (S) =Lλ∈Zn F tλ(cid:8)Pn i=1 si∂i :Pi siλi = 0(cid:9). In particular, for n = 1 we get SDerF (F [t±1]) = F d for d = ∂1. 10 ERHARD NEHER 1.3. Appendix on central extensions of Lie algebras. Central extensions will turn out to be an important tool in the construction of extended affine Lie algebras. Although this provides one with a bigger and hence potentially more complicated Lie algebra, central extensions turn up naturally in the general theory and the biggest of them (the universal central extension) is in fact quite "nice". For example, universal central extensions often have a simpler presentation and a much richer representation theory than the original Lie algebra. In this appendix we review the necessary background. Definition 1.8 (Extensions). An extension of a Lie algebra L is a surjective ho- momorphism f : K → L of Lie algebras. A homomorphism from an extension f : K → L to another extension f ′ : K ′ → L is a Lie algebra homomorphism g : K → K ′ satisfying f = f ′ ◦ g. In other words, the diagram below is commuta- tive. (1.20) K @@@@@@@ f g L K ′ ~}}}}}}}} f ′ We will use abelian extensions, i.e., extensions f : K → L with Ker f an abelian ideal in the construction of an extended affine Lie algebra in section 5.4. Definition 1.9 (central extensions). A central extension of L is an extension f : K → L whose kernel Ker f is contained in the centre Z(K) of K. A central extension f : K → L is called a covering if K is perfect, i.e., K = [K, K]. It is traditional (but not always advisable) to not specify the morphism f and simply say that K is a central extension of L or a covering. A central extension u : L → L is called a universal central extension if there exists a unique homomorphism from u : L → L to any other central extension f : K → L of L. It is obvious from the universal property that two universal central extensions of L are isomorphic as central extensions and hence in particular their underlying Lie algebras are isomorphic. We denote the universal central extension of L by u : uce(L) → L or simply uce(L). Theorem 1.10 ([vdK, Prop. 1.3], [G, §1]). A Lie algebra L has a universal central extension if and only if L is perfect. In this case, the universal central extension u : uce(L) → L is perfect too, i.e., u is a covering. The process of taking universal central extensions stops at uce(L), due to the following equivalent conditions for a Lie algebra L: (i) Id : L → L is a universal central extension, i.e., uce(L) = L, (ii) every central extension f : K → L is direct product K = L × Ker f such that f L is an isomorphism between L and L. If (i) and (ii) hold, one calls L centrally closed. Examples 1.11. (a) It is an immediate corollary of the Levi-Malcev Theorem that every finite-dimensional semisimple Lie algebra is centrally closed ([Bou3, VII, §6.8, Cor. 3] or [We, Cor. 7.9.5]). (b) An example of a universal central extension is the Virasoro algebra: It is the universal central extension of the Witt algebra DerF (F [t±1]), see for example [MP, / / ~ LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 11 I.9, Prop. 4]. Hence the Virasoro algebra is centrally closed, while DerF (F [t±1]) is not. On the other hand, the higher rank Witt algebra DerF (F [t±1 n ]), n > 1, is centrally closed ([RSS, V, Th. 5.1]). 1 , . . . , t±1 Definition 1.12 (Central extensions via 2-cocycles.). We have already seen in §1.1 that one can construct central extensions of a Lie algebra L by using 2-cocycles, which, we recall, are bilinear maps ψ : L × L → C into a vector space C satisfying for all l, l1, l2, l3 ∈ L (1.21) ψ(l, l) = 0 and ψ([l1, l2], l3) + ψ([l2, l3], l1) + ψ([l3, l1], l2) = 0. The first equation is of course equivalent to ψ(l1, l2) = −ψ(l2, l1). Given a 2-cocycle ψ : L × L → C, the algebra (1.22) K = L ⊕ C by [l1 ⊕ c1, l2 ⊕ c2]K = [l1, l2]L ⊕ ψ(l1, l2) (li ∈ L, ci ∈ C) is a Lie algebra and prL : K → L, prL(l ⊕ c) = l, is a central extension of L, which we will denote by E(L, C, ψ) or E(L, ψ) for short. Conversely, given a central extension f : K → L, let s : L → K be a section of f in the category of vector spaces, i.e. a linear map s : L → K such that f ◦ s = IdL. Such a section always exists: We can choose a subspace L′ of K, which is complementary to C = Ker f , and take s = (f L′)−1 which makes sense since (f L′) : L′ → L is an invertible linear map (but in general not a Lie algebra homomorphism since L′ need not be a subalgebra). Given a section s, the map (1.23) ψs : L × L → C, ψs(l1, l2) = [s(l1), s(l2)]K − s([l1, l2]L) turns out to be a 2-cocycle. Moreover, the map K → L ⊕ C, x 7→ f (x) ⊕(cid:0)x − (s ◦ f )(x)(cid:1) = f (x) ⊕ xC , where xC is the C-component of x ∈ K, is an isomorphism from the central exten- sion f : K → L to the central extension E(L, Ker f, ψs). To summarize, modulo some verifications left as an exercise, we have proven the following well-known re- sult. Proposition 1.13. For any 2-cocycle ψ the construction (1.22) is a central ex- tension E(L, ψ) of L and, conversely, every central extension L is isomorphic as central extension to some E(L, ψ). Exercise 1.14. Let ψ : L × L → C be a 2-cocycle and let C′ be a subspace of C satisfying ψ(L, L) := spanF {ψ(l1, l2) : li ∈ L} ⊂ C′. Then E(L, C′, ψ) is also a central extension, and if E(L, C, ψ) is a covering then C = ψ(L, L). Examples 1.15. (a) Any Lie algebra L has many uninteresting central extensions. One can simply take the direct product of L with an abelian Lie algebra, i.e., L × C with product [(l1, c1), (l2, c2)] = ([l1, l2], 0) for li ∈ L, ci ∈ C, and consider the canonical projection prL : L × L → L, which is a central extension (but not a covering, unless L is perfect and C = {0}). Observe that the canonical inclusion inc : L → L × C is a section of prL, not only in the category of vector spaces, but even in the category of Lie algebras. Its associated 2-cocycle ψinc = 0. (b) Let h : L → C be a linear map into some vector space C. Then βh : L × L → C, βh(l1, l2) = h([l1, l2]) is a 2-cocycle, a so-called 2-coboundary. The two examples are related in the following exercise. 12 ERHARD NEHER Exercise 1.16. For a central extension f : K → L of L with C = Ker f the following are equivalent: (i) The extension f : K → L is split in the category of Lie algebras, i.e, there exists a section L → K of f , which is a Lie algebra homomorphism. (ii) For any section s of f the associated 2-cocycle ψs is a 2-coboundary. (iii) There exists a section s of f , for which the associated 2-cocycle ψs is a 2-coboundary. (iv) As central extension, f is isomorphic to the central extension prL : L⊕C → L. If these conditions are fulfilled, one calls f a split extension. Exercise 1.17. Let ψ : L × L → C be a 2-cocycle and let π : C → C′ be a linear map. Show: (a) ψ′ = π ◦ ψ is a 2-cocycle of L and the map E(π) : E(L, C, ψ) → E(L, C′, ψ′), l ⊕ c 7→ l ⊕ π(c) is a homomorphism of central extensions of L: E(L, C, ψ) E(π) E(L, C′, ψ′) $IIIIIIIIII prL ytttttttttt prL L (b) If π is surjective, the map E(π) is a central extension of L′ = E(L, C′, π ◦ ψ), which as central extension of L′ has the form E(L′, C′′, ψ′′) for where γ : C′ → C is a section of π with γ(C′) = C′′. ψ′′(l1 ⊕ c′ 1, l2 ⊕ c′ 2) =(cid:0)(Id −γ ◦ π) ◦ ψ(cid:1)(l1, l2), (c) Conversely, suppose f ′ : L′ → L is a central extension and f : E(L, C, ψ) ։ L′ is a surjective homomorphism of central extensions. Then π = f C maps C onto C′ = Ker f ′ and there exists a unique isomorphism of extensions Φ : L′ → E(L, C′, ψ′), ψ′ = π ◦ ψ such that all triangles in the diagram below commute: E(L, C, ψ) f L′ E(π) 'OOOOOOOOOOO ??????????????????? prL prL Φ ztttttttttt f ′ E(L, C′, ψ′) L Exercise 1.18. Let C = C1 ⊕ C2 be a vector space direct sum and denote by πi : C → Ci the canonical projections. Let ψ : L × L → C be a 2-cocycle with the property that π2 ◦ ψ is a 2-coboundary. Then for ψ1 = π1 ◦ ψ, E(L, C, ψ) ∼= E(L, C1, ψ1) × C2 as central extensions of L (even as central extensions of the Lie algebra E(L, C1, ψ1)). / / $ y / / ' z LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 13 Example 1.19. Let (··) : L × L → F be a symmetric bilinear form, which is invariant, see Ex. 1.6. We denote by SDerF (L) the subalgebra of DerF (L) which consists of all skew-symmetric derivations, where a derivation d ∈ DerF (L) is called skew-symmetric if (d(l1) l2) + (l1 d(l2)) = 0 for all l1, l2 ∈ L. Observe that IDer(L) = {ad l : l ∈ L} ⊳ SDerF (L). Let D be a subspace of SDerF (L) and let D∗ be its dual space. Then the map ψD : L × L → D∗, given by (1.24) for li ∈ L and d ∈ D, is a 2-cocycle of L. ψD(l1, l2)(d) =(cid:0)d(l1) l2) Exercise 1.20. Show: (a) (1.24) defines indeed a 2-cocycle. (b) ψD for D ⊂ IDer(L) is a 2-coboundary. (c) If D is a subspace of D, then the central extension E(L, D∗, ψD) of L factors through the central extension E(L, D∗, ψ D) of L, E(L, D∗, ψD) ։ E(L, D∗, ψ D) ։ L. Definition 1.21 (Graded central extensions). Let Λ be an abelian group, and let L = Lλ∈Λ Lλ be a Λ-graded Lie algebra. We say that f : K → L is a Λ-graded central extension of L if K is a Λ-graded Lie algebra and f is a central extension which is at the same time a homomorphism of Λ-graded algebras: f (K λ) ⊂ Lλ for all λ ∈ Λ. A Λ-graded central extension f : K → L is called a Λ-covering, if f is a covering, i.e., K is perfect. We note that an arbitrary central extension of a graded Lie algebra need not be a graded central extension. A homomorphism of a Λ-graded central extension f : K → L to another Λ- graded central extension f ′ : K ′ → L is a homomorphism g : K → K ′ of Λ-graded Lie algebras satisfying f = f ′ ◦ g, cf. 1.20. To define graded central extensions of a Λ-graded Lie algebra L via a 2-cocycle, we need (obviously) a Λ-graded 2-cocycle, i.e., a 2-cocycle ψ : L × L → C into a Λ-graded vector space C =Lλ∈Λ C λ which is graded of degree 0, ψ(Lλ, Lµ) ⊂ C λ+µ for all λ, µ. For a graded 2-cocycle ψ the Lie algebra K = L ⊕ C of (1.22) is naturally Λ-graded by K λ = Lλ ⊕ C λ and the central extension prL : K → L is a Λ-graded central extension. Conversely, if f : K → L is a Λ-graded central extension, we can choose a section s : L → K of the underlying vector spaces of degree 0, meaning s(Lλ) ⊂ K λ. The 2-cocycle associated to s in (1.23) is then a graded 2-cocycle. Thus, Prop. 1.13 holds in an analogous way for graded central extensions. The following proposition also shows that one does not have to introduce a new object of a "graded universal central extension". Proposition 1.22 ([Ne2, 1.16]). Let L = Lλ∈Λ Lλ be a Λ-graded perfect Lie algebra. Then its universal central extension u : uce(L) → L is Λ-graded, hence a Λ-covering. Moreover, Ker u is a graded subspace of uce(L). Example 1.23. We also have the graded versions of the Example 1.15 (details left to the reader) and the Example 1.19, whose details follow. 14 ERHARD NEHER Let L =Lλ∈Λ Lλ be a Λ-graded Lie algebra and let (··) be an invariant bilinear form on L, which is Λ-graded in the following sense: (Lλ Lµ) = 0 if λ + µ 6= 0. We define the Λ-graded subalgebra of grEndF (L) (1.25) grSDerF (L) = grEndF (L) ∩ SDerF (L) =Lλ∈Λ(cid:0) SDerF (L)(cid:1)λ where (SDerF (L))λ consists of all skew-symmetric derivations of degree λ. If D ⊂ grSDerF (L) is a graded subspace of grSDerF (L), the 2-cocycle ψD of (1.24) is Λ- graded and maps L × L into Dgr∗, thus giving rise to a graded central extension E(L, Dgr∗, ψD) of L. Exercise 1.24. Show that the 2-cocycles ψ of (1.5), (1.16) and (1.17) can be obtained in the form (1.24), i.e., find an invariant bilinear form on L = L(g, σ) resp. L = L(g, σ) and a subspace D ⊂ SDerF (L) such that ψ and ψD yield isomorphic central extensions of L. It is not so surprising that the 2-cocycles we used in sections 1.1 and 1.2 can all be obtained in the form ψD for D ⊂ grSDerF (L). This is a special case of the following general result. Theorem 1.25 ([Ne6]). Let L =Lλ∈Λ Lλ be a Λ-graded Lie algebra, which (i) is perfect and finitely generated as Lie algebra, (ii) has finite homogeneous dimension: dim Lλ < ∞ for all λ ∈ Λ, and (iii) has an invariant nondegenerate Λ-graded symmetric bilinear form. (a) Then DerF (L) = grDerF (L) is Λ-graded and has finite homogeneous dimen- sion, whence the same is true for SDerF (L). (b) The universal central extension uce(L) has finite homogenous dimension with respect to the Λ-grading of 1.22. Moreover, uce(L) ∼= E(L, Dgr∗, ψD) as central extensions of L, where D is any graded subspace of SDerF (L) which complements IDer(L) in SDerF (L), and ψD is the 2-cocycle of (1.24). Remarks 1.26. (a) Th. 1.5 is an application of Th. 1.25, as is Th. 4.13(c). (b) The Exercise 1.18 gives some indication why it is sufficient to take a subspace of SDerF (L) complementing IDer(L) and not an arbitrary subspace of SDerF (L). Exercise 1.27. In the setting of Th. 1.25, every Λ-graded central covering of L is isomorphic as central extension to a central extension E(L, Bgr∗, ψB) for some graded subspace B of D. 2. Extended affine Lie algebras: Definition and first examples Rather than constructing Lie algebras in a concrete way as we have done in Lecture 1, in this chapter we will define extended affine Lie algebras by a set of axioms and give examples. We will see that these examples encompass all the examples of Lecture 1 (with the exception of the choice 2. for D in 1.2). As before we will consider Lie algebras over an arbitrary field F of characteristic 0, but we will no longer assume that F has enough roots of unity (multiloop algebras will not be play a role here), except in §2.4 where F = C). LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 15 2.1. Definition of an extended affine Lie algebra. An extended affine Lie algebra, or EALA for short, is a pair (E, H) consisting of a Lie algebra E over F and subalgebra H satisfying the following axioms (EA1) – (EA6). (EA1): E has an invariant nondegenerate symmetric bilinear form (··). (EA2): H is nontrivial finite-dimensional toral and self-centralizing subalge- bra of E. Before we can state the other four axioms, we need to draw some consequences of the axioms (EA1) and (EA2). But first we give explanations of some of the notions used. The term invariant (= associative) means that (··) satisfies ([e1, e2] e3) = (e1 [e2, e3]) for all ei ∈ E, and (··) is nondegenerate if (e E) = 0 =⇒ e = 0. In the context of above, a toral subalgebra, sometimes also called an ad-diagonalizable subalgebra is a subalgebra H which induces a decomposition of E via the adjoint representation of H: (2.1) E =Lα∈H ∗ Eα, Eα = {e ∈ E : [h, e] = α(h)e for all h ∈ H}. Such a subalgebra is necessarily abelian, whence H ⊂ E0 = {e ∈ E : [h, e] = 0 for all h ∈ H}. That H is also required to be self-centralizing means H = E0. Now to the consequences of (EA1) and (EA2). Because of invariance of the bilinear form (··), we have (2.2) (Eα Eβ) = 0 if α + β 6= 0, in particular the restriction of the bilinear form to E0 = H is nondegenerate. Be- cause of this and finite-dimensionality of H, every linear form α ∈ H ∗ is represented by a unique tα ∈ H, defined by the condition that (tα h) = α(h) holds for all h ∈ H. This allows us to transport the restricted form (··) H × H to a symmetric bilinear form on H ∗, also denoted (··) and defined by (2.3) (α β) = (tα tβ), α, β ∈ H ∗. This transport of bilinear forms is a standard procedure in the theory of semisimple Lie algebras, see for example [Hu, §8]. We can now define (2.4) R = {α ∈ H ∗ : Eα 6= 0} (set of roots of (E, H)), R0 = {α ∈ R : (α α) = 0} (null roots), Ran = {α ∈ R : (α α) 6= 0} (anisotropic roots). We prefer to call R the set of roots of (E, H) and not the "root system" since we want to restrict the latter term for root systems in the usual sense, see 3.3. We point out that by definition 0 is a root, 0 ∈ R0 ⊂ R. This is the customary convention for EALAs and has some notational advantages. We define the core of (E, H) as the subalgebra Ec of E generated by all anisotropic root spaces: We can now state the remaining four axioms. Ec = hSα∈Ran Eα isubalg 16 ERHARD NEHER (EA3): For every α ∈ Ran and xα ∈ Eα, the operator ad xα is locally nilpo- tent on E. (EA4): Ran is connected in the sense that for any decomposition Ran = R1 ∪ R2 with (R1 R2) = 0 we have R1 = ∅ or R2 = ∅. (EA5): The centralizer of the core Ec of E is contained in Ec: {e ∈ E : [e, Ec] = 0} ⊂ Ec. (EA6): The subgroup Λ = spanZ(R0) ⊂ H ∗ generated by R0 in (H ∗, +) is a free abelian group of finite rank. In other words, Λ ∼= Zn for some n ∈ N (including n = 0!). The term locally nilpotent means that for every e ∈ E there exists an n ∈ N, possibly depending on e, such that (ad xα)n(e) = 0. The property (EA5) is called tameness. The condition [e, Ec] = 0 is of course equivalent to [e, Eα] = 0 for all α ∈ Ran. The rationale for this axiom is the following. The subalgebra Ec is in fact an ideal of E (Th. 4.14). Hence we have a representation ρ of E on Ec, given by ρ(e)(xc) = [e, xc] for e ∈ E and xc ∈ Ec. The kernel of the representation ρ is the centralizer of Ec in E. Hence tameness means that Ker ρ ⊂ Ec. The idea here is that the core Ec should control E. We will make this more precise in section 5.4. The rank of the free abelian group Λ in axiom (EA6) is called the nullity of (E, H). It is invariant under isomorphisms. We will describe EALAs of nullity 0 and 1 below. Although the structure of an EALA requires the existence of an invariant non- degenerate symmetric bilinear form (··) in the axiom (EA1), which is then used to define the anisotropic roots, it turns out that this bilinear form is really not so important. Because of this, we have defined an EALA as a pair (E, H) and not as a triple (E, H, (··)) as it is for example done in [AF]. Consequently, an isomorphism from an EALA (E, H) to another EALA (E′, H ′) is a Lie algebra isomorphism f : E → E′ such that f (H) = H ′. It is immediate that any isomorphism induces a bijection f ′ between the set of roots R and R′ of (E, H) and (E′, H ′) respectively. It then follows that f ′ maps Ran onto R′an, whence also R0 onto R′0. One can then show that f ′ preserves the forms on X = spanF (R) and X ′ = spanF (R′) up to scalars. For F = C one can define a special class of EALAs. We call a pair (E, H) a discrete EALA if it satisfies the axioms (EA1) – (EA5) and in addition (DE): R is a discrete subset of H ∗ with respect to the natural topology of the finite-dimensional complex vector space H ∗. It is justified to call a discrete EALA an EALA, since one can show that a discrete EALA also satisfies (EA6). Indeed, this follows from Prop. 3.21 and Th. 3.22. However, not every EALA over C is a discrete EALA (see [Ne5, 6.17]). Some historical comments. Although there were some precursors (papers by Saito and Slodowy for nullity 2), it was in the paper [HT] by the physicists Høegh-Krohn and Torr´esani that the class of discrete extended affine Lie algebras was introduced, however not under this name. Rather, they were called "irreducible quasi-simple Lie algebras" and later ([BGK, BGKN]) "elliptic quasi-simple Lie al- gebras". The stated goal of the paper [HT] was applications in quantum gauge theory. The theory developed there did however not stand up to the scrutiny of mathematicians. The errors of [HT] were corrected in the AMS memoir [AABGP] LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 17 by Allison, Azam, Berman, Gao and Pianzola. There also the name "extended affine Lie algebras" appears for the first time. But not in the sense as defined above. Rather, the authors develop the basic theory of what here are called dis- crete EALAs. Nevertheless, [AABGP] has become the standard reference even for the more general extended affine Lie algebras, since many of the results presented there for discrete extended affine Lie algebras easily extend to the more general setting. The definition of an extended affine Lie algebra given above is due to the author ([Ne4]) and was motivated by the fact that all the examples presented in [AABGP] did make sense over an arbitrary base field F and not just over C only. Before [Ne4] the tameness axiom (EA5) was not part of the definition of an EALA. However, as examples show ([BGK, §3] or [Ne5, 6.10]), it seems impossible to clas- sify EALAs without (EA5). After [Ne4], several generalizations of EALAs have been proposed. They are surveyed in [Ne5]. 2.2. Some elementary properties of extended affine Lie algebras. The fol- lowing chapters will (hopefully) show that extended affine Lie algebras share many properties with familiar Lie algebras, like finite-dimensional split simple Lie alge- bras or affine Kac-Moody Lie algebras. Some of these properties are immediate consequences of the axioms. The following (strongly recommended!) exercise gives an incomplete list of such properties. Exercise 2.1. Let (E, H) be an EALA. We use the notation of above. Show: (a) For α, β ∈ R we have (2.5) [Eα, Eβ] ⊂ Eα+β . Thus the root space decomposition (2.1) is a grading by the abelian group spanZ(R). (b) H is a Cartan subalgebra, defined as a nilpotent subalgebra which is self- normalizing: H = {e ∈ E : [e, H] ⊂ H}. (c) For α, β ∈ R we have (Eα Eβ) = 0 unless α + β = 0. The restriction of the bilinear form (··) to Eα × E−α is nondegenerate, i.e., if xα ∈ Eα satisfies (xα E−α) = 0 then xα = 0. In particular, R = −R. (d) For α ∈ R and xα ∈ Eα and y−α ∈ E−α, (2.6) (2.7) [xα, y−α] = (xα y−α) tα. In particular, [Eα, E−α] = F tα, and if α ∈ Ran then [[Eα, E−α], Eα] = Eα. (e) The core Ec satisfies (2.8) Ec =(cid:16) ⊕α∈Ran Eα(cid:17) ⊕(cid:16)Lα∈R0 (Ec ∩ Eα)(cid:17). We will now show that EALAs are built out of "little" sl2's and Heisenberg's (albeit in a complicated way). Proposition 2.2. Let (E, H) be an extended affine Lie algebra, with anisotropic root Ran and null roots R0. (a) Let α ∈ Ran. Then dim Eα = 1, and for any eα ∈ Eα there exists fα ∈ E−α such that (eα, hα = [eα, fα], fα) ∈ Eα × H × E−α is an sl2-triple: Eα ⊕ [Eα, E−α] ⊕ E−α = F eα ⊕ F hα ⊕ F fα ∼= sl2(F ). 18 ERHARD NEHER (b) Let α ∈ R0. Then for any 0 6= xα ∈ Eα there exists yα ∈ E−α such that [xα, yα] = tα and the 3-dimensional Heisenberg algebra. F xα ⊕ F tα ⊕ F yα ∼= h3, It is not true that dim Eα = 1 if α ∈ R0 (this is already not true in the examples of sections 2.3 and 2.4). But we will show in Th. 4.18 that all root spaces Eα are finite-dimensional in a rather strong way. The following exercise shows that one can "extend" the 3-dimensional Heisenberg subalgebras in (b) above. Exercise 2.3. In the setting and notation of Prop. 2.2(b) show that there exists dα ∈ H such that [dα, xα] = xα and [dα, yα] = −yα. Hence F xα ⊕ F tα ⊕ F dα ⊕ F yα is a 4-dimensional subalgebra. It is 2-step solvable, not nilpotent and isomorphic to the subalgebra of gl3(F ). 0 c d n(cid:16) 0 a b 0 0 0(cid:17) : a, b, c, d ∈ Fo And now an exercise which implies that in an EALA one can produce many so-called elementary automorphisms. Exercise 2.4. Let M be an F -vector space. Once calls an endomorphism f ∈ EndF (M ) locally nilpotent if for every m ∈ M there exists n ∈ N, possibly depend- ing on m, such that f n(m) = 0. (a) Show that the following conditions are equivalent for f ∈ EndF (M ): (i) f is locally nilpotent, (ii) for every finitely spanned subspace N of M there exists a finite- dimensional subspace P of M such that N ⊂ P and f (P ) ⊂ P , (iii) f is nilpotent on every finite-dimensional and f -invariant subspace of M . (b) Let f ∈ EndF (M ) be locally nilpotent and define the exponential exp f of f by for m ∈ M (note that the sum on the right is always finite). Show that exp f is an invertible endomorphism of M with inverse given by (exp f )−1 = exp(−f ). (exp f )(m) =Pn∈N 1 n! f n(m), (c) Let L be a Lie algebra and let d be a locally nilpotent derivation of L. Show that then exp d is an automorphism of L. We will next present some examples of EALAs. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 19 2.3. Extended affine Lie algebras of nullity 0. Let g be a finite-dimensional split simple Lie algebra with splitting Cartan subalgebra h, for example sll(F ) or a finite-dimensional simple Lie algebra over an algebraically closed field. We will show that then (g, h) is an EALA of nullity 0. The facts needed to prove this can be found in [Bou3, VIII, §2] or in [Hu, §8] for F algebraically closed. (EA1) Up to a scalar, there exists only one invariant nondegenerate symmetric bilinear form on g, the Killing form κ. Hence we can (and will) take (··) = κ. (EA2) By definition of a splitting Cartan subalgebra, the Lie algebra g has a root space decomposition g = g0 ⊕(cid:0)Lα∈Φ gα(cid:1), g0 = h, where Φ is the root system of (g, h) (which is a reduced root system in the usual sense, see 3.3) and where the root spaces gα are defined as in 2.1. Hence the set of roots R of (g, h) is (2.9) R = {0} ∪ Φ. It is a basic fact that κ(tα, tα) 6= 0 for tα ∈ h representing α ∈ Φ via κ(tα, h) = α(h) for all h ∈ h. Hence, the anisotropic and null roots are Ran = Φ and R0 = {0}. (EA3) is now obvious: From [gα, gβ] ⊂ gα+β for α ∈ Φ and β ∈ R and finite- dimensionality of g, it is clear that ad xα for xα ∈ gα is not only locally nilpotent but even (globally) nilpotent. (EA4) is another way of saying that Φ is an irreducible root system. This is indeed the case and follows from simplicity of g. (EA5) We first need to determine the core gc of g. By definition, gc is the subalgebra of g generated by Lα∈Φ gα. Since h =Pα∈Φ[gα, g−α] we have gc = g. It is now a tautology that (EA5) holds, i.e., that the centralizer of the core gc is contained in gc = g. Of course, we know even more: The centralizer of the core equals the centre of g, and is therefore {0}. (EA6) We have Λ = hR0i = h{0}i = {0}. We have now shown: (2.10) A finite-dimensional split simple Lie algebra is an EALA of nullity 0. We will see in Prop. 3.24 that the converse of (2.10) is true too. We thus know all the nullity 0 examples of EALAs, and can therefore focus on the higher nullity examples. We will answer the case of nullity 1 in the next section. 2.4. Affine Kac-Moody Lie algebras again. To justify the name extended affine Lie algebra, we will now show that any affine Kac-Moody Lie algebra is an extended affine Lie algebra. To do so, we will need some basic facts about affine Kac-Moody Lie algebras. All of them can be found in Kac's book [Kac]. Since this reference uses C as base field, we will do the same in this section. But everything we say here holds true for arbitrary algebraically closed fields of characteristic 0. Thus we let L =Ln∈Z g¯n ⊗ Ctn L = L(g, σ) = L ⊕ Cc ⊕ Cd 20 ERHARD NEHER be the complex Lie algebra described in (1.9) and (1.10). Recall that g is a finite- dimensional simple Lie algebra over C and σ is a diagram automorphism of g. We let m ∈ {1, 2, 3} be the order of σ, and denote the canonical map Z → Z/mZ by n 7→ ¯n. Recall from (1.1) and (1.2) that σ induces a Z/mZ-grading of g, namely g = g¯0 ⊕ · · · ⊕ gm−1 where g¯n = {x ∈ g : σ(x) = ζ nx} for a primitive mth root of unity ζ. For example, for m = 2 we get a Z/2Z-grading g = g¯0 ⊕ g¯1 with g¯0 = {x ∈ g : σ(x) = x} and g¯1 = {x ∈ g : σ(x) = −x}. We identify g¯0 ≡ g¯0 ⊗ Ct0. We now verify the axioms (EA1) – (EA5) and (DE) which, we recall, implies (EA6). (EA1) We let κ be the Killing form of g and define a bilinear form (··) on L, using the notation of (1.10), (2.11) (cid:0)u¯λ ⊗ tλ ⊕ s1c ⊕ s′ = κ(u¯λ, v¯µ)δλ,−µ + s1s′ 2 + s2s′ 1. 1d v¯µ ⊗ tµ ⊕ s2c ⊕ s′ 2d(cid:1) The form is visibly symmetric. The reader is invited in Exercise 2.6 to show that it is in fact an invariant nondegenerate symmetric bilinear form on L, as required in (EA1). In anticipation of the later developments, we point out that (··) has the following features: • L is an orthogonal sum of L and Cc ⊕ Cd: L = L ⊥ (Cc ⊕ Cd), • Cc ⊕ Cd is a hyperbolic plane, i.e., (c c) = 0 = (d d) while (c d) = 1. • The Laurent polynomial ring C[t±1] has a nondegenerate symmetric bilinear form ǫ given by ǫ(tλ, tµ) = δλ,−µ. It is invariant in the sense that ǫ(pq, r) = ǫ(p, qr) for p, q, r ∈ C[t±1], and is graded in the sense that ǫ(tλ, tµ) = 0 unless λ + µ = 0. For σ = Idg, the bilinear form on the loop algebra g ⊗ C[t±1] is simply the tensor product form κ ⊗ ǫ, and for a general σ the form is obtained by restriction. (EA2) To construct a subalgebra H as required in axiom (EA2) we start with a Cartan subalgebra h of g. Since σ is a diagram automorphism, it leaves h invariant. We let and put h¯0 = h ∩ g¯0 = {h ∈ h : σ(h) = h} H = h¯0 ⊕ Cc ⊕ Cd. One knows that g¯0 is a simple Lie algebra with Cartan subalgebra h¯0 ([Kac, Prop. 7.9]). The grading property implies that [g¯0, g¯n] ⊂ g¯n for n ∈ Z. Hence g¯0 acts on g¯n by the adjoint action. Let ∆¯n be the set of weights of the g¯0-module g¯n with respect to h¯0: g¯n =Lγ∈∆ ¯n g¯n,γ g¯n,γ = {x ∈ g¯n : [h¯0, x] = γ(h¯0)x for all h¯0 ∈ h¯0}. In particular, ∆¯0 \ {0} is the root system of g¯0 with respect to h¯0 and h¯0 = g¯0,0. We extend ∆¯n ⊂ h∗ ¯0 to a linear form on H by zero, i.e., for γ ∈ ∆¯n we put γ(h¯0 ⊕ sc ⊕ s′d) = γ(h¯0) and define a linear form δ on H by δ(h¯0 ⊕ sc ⊕ s′d) = s′. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 21 Then for γ ∈ ∆¯n, n ∈ Z, we have Lγ⊕nδ = {u ∈ L : [h, u] = (γ ⊕ nδ)(h)u for all h ∈ H} (2.12) =(g¯n,γ ⊗ tn, γ ⊕ nδ 6= 0, γ ⊕ nδ = 0, H, whence L = Lα∈R of roots (2.13) Lα has a root space decomposition with respect to H with set R = {γ ⊕ nδ : γ ∈ ∆¯s, ¯n = ¯s, 0 ≤ s < m}. This establishes (EA2). To check the other axioms we first need to determine which of the roots in R are the null respectively anisotropic roots. Following the procedure in §2.1, we consider the restriction of the bilinear form (··) to H. With obvious notation this is 1d h′ ¯0 ⊕ s2c ⊕ s′ 2d) = κ(h¯0, h′ ¯0) + s1s′ 2 + s2s′ 1. (cid:0)h¯0 ⊕ s1c ⊕ s′ Since κh¯0×h¯0 is nondegenerate, this is indeed a nondegenerate symmetric bilinear form on H, as it should be. Let tγ ∈ h¯0 be the element representing γ ∈ h∗ ¯0: κ(tγ, h¯0) = γ(h¯0) for all h¯0 ∈ h¯0. For the canonical extension of γ to a linear form of H, also denoted by γ, we then get (tγ h) = γ(h) for all h ∈ H. Moreover (c h¯0 ⊕ sc ⊕ s′d) = s′ = δ(h¯0 ⊕ sc ⊕ s′d) shows that δ is represented by tδ = c ∈ H. Therefore α = γ ⊕ nδ ∈ R is represented by tγ⊕nδ = tγ ⊕ nc. Now observe (tγ⊕nδ tγ⊕nδ) = (tγ ⊕ nc tγ ⊕ nc) = κ(tγ, tγ). It is of course well- known that κ(tγ, tγ) 6= 0 for 0 6= γ ∈ ∆¯0. But one can (easily) show that this also holds for any 0 6= γ ∈ ∆¯n. We therefore get (2.14) Ran = {γ ⊕ nδ ∈ R : γ 6= 0} and R0 = Zδ, which in the theory of affine Kac-Moody algebras are usually called real and imag- inary roots. We are now set for the verification of the remaining axioms. (EA3) holds in the stronger form: ad Lλ, α ∈ Ran, is nilpotent. (We have already seen the same phenomenon in the Example 2.3 of a finite-dimensional split simple Lie algebra. Perhaps the reader wonders if this is true in general. The answer is yes.) (EA4) The verification of (EA4) is left to the reader. (EA5) The core of L is Lc = (cid:0)Ln∈Z g¯n ⊗ Ctn(cid:1) ⊕ Cc, and therefore equals the derived algebra [ L, L] of L. The centralizer of Lc in L, in fact the centre of L is Cc ⊂ Lc, see Exercise 1.2. (DE) In this example the subgroup Λ = hR0i equals R0 = Zd and is a discrete subset of H ∗. We have now shown one implication of the following result. Theorem 2.5 ([ABGP]). A complex Lie algebra E is a discrete EALA of nullity 1 if and only if E is an affine Kac-Moody Lie algebra. Exercise 2.6. Check the following details of the construction above. (a) (2.11) defines an invariant symmetric bilinear form on L. (b) L has a root space decomposition whose root spaces are given by (2.12) and whose set of roots is (2.13). 22 ERHARD NEHER (c) κ(tγ, tγ) 6= 0 for any 0 6= γ ∈ ∆¯n. (d) (EA4) holds for ( L, H). 2.5. Higher nullity examples. We have seen all examples of EALAs of nullity 0 and 1. In this section we will construct examples of higher nullity. To simplify things we consider untwisted algebras (no non-trivial finite order automorphism are involved). We can therefore go back to our standard setting: g is a split simple Lie algebra over a field F of characteristic 0. As in §1.2 let F [t±1 1 , . . . , t±n n ] be the Laurent polynomial ring in n variables and let L = L(g) = g ⊗ F [t±1 1 , . . . , t±n n ] be the associated untwisted multiloop algebra. We have seen in Exercise 1.4 that L has a 2-cocycle ψ : L × L → F n =: C, given by (1.16): ψ(u ⊗ tλ, v ⊗ tµ) = δλ+µ,0 κ(u, v) λ. We can therefore define the central extension K = L ⊕ C with product (1.22). In (1.19) we have defined degree derivations ∂i, i = 1, . . . , n, of K. Let (2.15) D = spanF {∂1, . . . , ∂n} and define the Lie algebra E as the semidirect product, Let h be a Cartan subalgebra of g and put E =(cid:0)L(g) ⊕ C(cid:1) ⋊ D. We claim that (E, H) is an EALA of nullity n. H = h ⊕ C ⊕ D. (EA1) We will mimic the construction of an invariant nondegenerate symmetric bilinear form in §2.4 and require • (L(g) C ⊕ D) = 0. • C ⊕ D is a hyperbolic space with (C C) = 0 = (D D) and (Pi sici Pi s′ i∂i) =Pi sis′ i, where c1, . . . , cn is the canonical basis of F n. Thus C ⊕ D is the orthogonal sum of the n hyperbolic planes F ci ⊕ F ∂i. • On L(g) the form is the tensor product form of the Killing form κ of g and the natural invariant bilinear form on F [t±1 1 , . . . , t±1 n ]. Putting all these requirements together, we arrive at the global formula which is completely analogous to (2.11): (cid:0)u ⊗ tλ ⊕ Pi sici ⊕Pj s′ j∂j v ⊗ tµ ⊕ Pi tici ⊕Pj t′ i + tis′ i). j∂j(cid:1) = κ(u, v)δλ,−µ + Pi(sit′ (2.16) (EA2) Let h be a splitting Cartan subalgebra and let Φ be the usual root system of (g, h), thus 0 6∈ Φ. We put ∆ = {0} ∪ Φ and then have the root space decompo- sition g = Lγ∈∆ gγ with g0 = h. We embed ∆ ֒→ H ∗ by requiring γ C ⊕ D = 0 for γ ∈ ∆. Also we embed Λ = Zn ֒→ H ∗ by λ(h ⊕ C) = 0 and λ(∂i) = λi for LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 23 λ = (λ1, . . . , λn) ∈ Λ. Then E has the root space decomposition E = Lα∈R Eα with root spaces E0 = H, R0 = Λ. (2.17) (2.18) Eγ⊕λ = gγ ⊗ tλ Ran = Φ × Λ, (γ ⊕ λ 6= 0), It is now not difficult to verify (EA3) – (EA5) and (DE). Thus: Lemma 2.7. The pair (E, H) constructed above is a discrete EALA of nullity n. There is however no analogue of Prop. 2.10 and Th. 2.5: There are many more EALAs of nullity n ≥ 2. We have just seen the "tip of the iceberg"! Other examples can be found in Ch.III of [AABGP], some of them involving heavy-duty nonassociative algebras, like octonion algebras and Jordan algebras over Laurent polynomial rings! Exercise 2.8. Supply the missing details of the proof that (E, H) above is a discrete EALA of nullity n. In particular, prove: (a) (2.16) defines an invariant nondegenerate symmetric bilinear form on E. (b) The root spaces of (E, H) and the anisotropic and null roots are as stated in (2.17) and (2.18). 3. The structure of the roots of an EALA In this chapter we will describe the structure of the set of roots R of an EALA (E, H), defined in (2.4). We have already seen some examples: R can be a finite irreducible reduced root system, (2.9), R can be an affine root system (2.13), i.e., the set of roots of an affine Kac-Moody Lie algebra, or R can be of the form R = S × Zn where S \ {0} is a finite irreducible reduced root system (2.18). Thus any description of the general case has to encompass all these different examples. It turns out that the roots of an EALA form an extended affine root system and that the latter is naturally described as a special case of affine reflection systems. We therefore first introduce the latter, describe their structure and then specialize later to extended affine root systems. Affine reflection systems are themselves special cases of reflection systems, whose theory is developed in [LN]. 3.1. Affine reflection systems: Definition. Throughout this section we work with a triple (R, X, (··)) where • X is a finite-dimensional vector space over a field F of characteristic 0, • (··) is a symmetric bilinear form on X and • R ⊂ X. For any such triple (R, X, (··)) we define X 0 = {x ∈ X : (x X) = 0}, the radical of (··), R0 = {α ∈ R : (αα) = 0}, Ran = {α ∈ R : (αα) 6= 0}, (anisotropic roots) (null roots) hx, α∨i = 2 (xα) (αα) , (x ∈ X and α ∈ Ran) (3.1) sα(x) = x − hx, α∨iα. 24 ERHARD NEHER By definition we therefore have R = R0 ∪ Ran. The map sα : X → X is a reflection in α, i.e., s2 α = IdX and {x ∈ X : sα(x) = −x} = F α. It is also orthogonal with respect to (··): (sα(x) sα(y)) = (x y) for all x, y ∈ X. We call (R, X, (··)), or just R for short, an affine reflection system if (AR1): 0 ∈ R and R spans X, (AR2): sα(R) = R for all α ∈ Ran, (AR3): for every α ∈ Ran the set hR, α∨i is finite and contained in Z, and (AR4): R0 = R ∩ X 0. An affine reflection system is said to be • reduced if for every α ∈ Ran and c ∈ F : cα ∈ Ran ⇐⇒ c = ±1, • connected if for any decomposition Ran = R1 ∪ R2 with (R1 R2) = 0 we have R1 = ∅ or R2 = ∅. The nullity of (R, X) is the rank of the torsion-free abelian group Z[R0] = spanZ(R0) generated by R0 in (X, +). Thus, by definition, nullity of (R, X) = dimQ(Z[R0] ⊗Z Q) = dimF (Z[R0] ⊗Z F ). Since the vector space Z[R0] ⊗Z F maps onto spanF (R0), the nullity of (R, Z) is bounded below by dimF spanF (R0). It is in general not equal to it. But this is of course so for nullity 0: (R, X) has nullity 0 if and only if R0 = {0} ⇐⇒ dimF spanF (R0) = 0. Remarks 3.1. - For a large part of the theory it is not necessary that X be finite-dimensional, see [LN]. But assuming this right from the start, simplifies the presentation. - We need the bilinear form (··) to define R0 and the reflections. But although we will sometimes write (R, X, (··)), we will not consider (··) as part of the structure of an affine reflection system. For example, in the definition of an isomorphism below we will not require that the bilinear forms are preserved. See [LN], where this point of view is emphasized. - The requirement 0 ∈ R is in line with the previous chapter, in which 0 was considered a root of an EALA. This conflicts with the traditional approach to root systems in which 0 is not a root, see for example [Bou2], [Hu] or [Kac]. The question whether 0 is a root or is not a root, has lead to heated debates. In the author's opinion, there are some advantages of considering 0 as a root, which however can only be fully seen when one develops the theory for affine reflection systems. But perhaps the reader can be convinced by the natural) example (R, X) = ({0}, {0}) of an affine reflection system. - The condition hβ, α∨i ∈ Z in axiom (AR3) makes sense since every field of characteristic 0 contains (an isomorphic copy of) the field of rational numbers, which allows us to identify Z ≡ Z1F . - By definition hX 0, α∨i = 0 for all α ∈ Ran. Hence sα(x0) = x0 for x0 ∈ X 0. Also, the inclusion R ∩ X 0 ⊂ R0 in (AR4) is always true. Therefore the axioms (AR2)–(AR4) can be replaced by the following conditions (AR2)′ (AR3)′ (AR4)′ R0 ⊂ X 0. sα(Ran) = Ran for all α ∈ Ran, for every α ∈ Ran the set hRan, α∨i ⊂ Z is finite, This new set of axioms makes it (even more) clear that the conditions on R0 are rather weak: We (may) need R0 to span X from (AR1), we need R0 ⊂ X 0 for LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 25 (AR4)′ and we need 0 ∈ R, which is no condition since one can always add 0 to R0. We will see this phenomena re-appearing in the examples, e.g., in Example 3.2, and in the definition of an extension datum in 3.9. - The definition of a connected affine reflection system is the same as the axiom (EA4) in the definition of an EALA. - The definition of an affine reflection system given in [LN] is not the same as the one given here. The equivalence of two definitions follows from [LN, Prop. 5.4]. An isomorphism from an affine reflection system (R, X, (··)) to another affine reflection system (R′, X ′(··)′) is a vector space isomorphism f : X → X ′ satisfying f (Ran) = R′an and f (R0) = R′0 If such a map exists, (R, X, (··)) and (R′, X, (··)′) are called isomorphic. One can show, as a corollary of the Structure Theorem 3.10, that an isomorphism f also satisfies f ◦ sα = sf (α) ◦ f for all α ∈ Ran, equivalently, hx, α∨i = hf (x), f (α)∨i for all x ∈ X and α ∈ Ran. This is always fulfilled if f is an isometry for (··) and (··)′ respectively. But in general an isomorphism is not necessarily an isometry. For example, one can always multiply the bilinear form (··) by a non-zero scalar without changing hx, α∨i. Since a reflection sα is an isometry, it follows from (AR2) and (AR4) that sα leaves Ran and R0 invariant and is thus an automorphism of (R, X). The subgroup W (R) of the automorphism group of (R, X) generated by all reflections sα, α ∈ Ran, is (obviously) called the Weyl group of (R, X). (It will not play a big role in this chapter.) 3.2. Examples of affine reflection systems. We will now give some immediate examples of affine reflection systems. Example 3.2 (The real part of an affine reflections system). Let (R, X, (··)) be an affine reflection system. Then Re(R) = {0} ∪ Ran, Re(X) = spanF (Ran), (··)Re = (··)Re(X)×Re(X) defines an affine reflection system, called the real part of (R, X), with Re(R)an = Ran, Re(R)0 = {0}, in particular Re(R) has nullity 0. Observe that (··)Re need not be nondegenerate, see Example 3.4 for an example. The fact that one can "throw away" the non-zero null roots and still have an affine reflection system indicates that one has little control over the null roots in a general affine reflection system. This will be made even more evident in the concept of an extension datum 3.9, used in the general Structure Theorem 3.10 for affine reflection systems. It is therefore natural to define subclasses of affine reflection system by imposing conditions on the null roots. For example, we will do so when we define extended affine root systems in 3.4. In [Ne5, 3.6] the author claimed that an affine reflection system of nullity 0 is a finite root system. The example above show that this is far from being true. But what remains true is the converse, also claimed in [Ne5, 3.6]: A finite root system is an affine reflection system of nullity 0, as we will show now. 26 ERHARD NEHER Example 3.3 (Finite root systems). Let Φ be a (finite) root system `a la Bourbaki [Bou2, VI, §1.1]. Recall that this means that Φ is a subset of an F -vector space Y satisfying the axioms (RS1)–(RS3) below. (RS1): Φ is finite, 0 6∈ Φ and Φ spans Y . (RS2): For every α ∈ Φ there exists a linear form α∨ ∈ Y ∗ such that α∨(α) = 2 and sα(Φ) = Φ, where sα is the reflection of Y defined by sα(y) = y − α∨(y)α, (RS3): for every α ∈ Φ the set α∨(Φ) is contained in Z. Observe that the reflection sα defined in (RS2) satisfies sα(α) = −α and sα(y) = y for α∨(y) = 0. It therefore seems to depend on α and the linear form α∨. However, since Φ is finite, there exists at most one reflection s with s(Φ) = Φ and s(α) = −α ([Bou2, VI, §1.1, Lemme 1]). It is therefore not necessary to indicate α∨ in the notation of sα. Note that we do not assume that Φ is reduced. This more general concept of a root system is necessary for the Structure Theorem of affine root systems (3.10). The reader who is only familiar with the theory of reduced finite root systems, as for example developed in [Hu, Ch. III], can perhaps be comforted by the fact that the difference is not very big. Indeed, every finite root system is a direct sum of connected (= irreducible) root systems and there is only one irreducible non-reduced root system of rank l, namely BCl = Bl ∪ Cl = {±εi : 1 ≤ i ≤ l} ∪ {±εi ± εj : 1 ≤ i, j ≤ l} where here and in the following ε1, . . . , εl is the standard basis of F l. (Note 0 ∈ BCl in anticipation of the convention introduced below.) In the context of finite-dimensional Lie algebras, non-reduced root systems arise naturally as the roots of a finite-dimensional semisimple Lie algebra L with respect to a maximal ad-diagonalizable subalgebra H ⊂ L which is not self-centralizing, hence not a Cartan algebra. In particular, non-reduced root systems do not occur over an algebraically closed field. However, they do occur in the context of infinite- dimensional Lie algebras, even over algebraically closed fields, see Ex. 3.6. Given a finite root system (Φ, Y ), define (3.2) S = {0} ∪ Φ and (x y) =Pα∈Φ α∨(x) α∨(y) for x, y ∈ Y . Then (··) is a nondegenerate symmetric bilinear form on Y with re- spect to which all reflections sα are isometric ([Bou2, VI, §1.1, Prop. 3]). Moreover, (α α) is a positive integer for every α ∈ Φ (viewing Q ⊂ F canonically) and hy, α∨i = α∨(y) = 2 (yα) (αα) for all y ∈ Y . Hence sα as defined in (RS2) is also given by the formula (3.1). We have S0 = {0} = X 0 = X 0 ∩ S. Since hΦ, α∨i ⊂ Z we have shown that (S, Y, (··)) as defined in (3.2) is a finite affine reflection system of nullity 0. We will characterize finite root systems within the category of affine reflection systems in Cor. 3.11. In the following we will always assume that a finite root system contains 0. We will usually use the symbol S for a finite root system, and put S× = S \ {0} = Φ. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 27 We will also need the following subsets of roots of a finite root system S: Sdiv is the set of divisible roots, where α ∈ S is called divisible if α/2 ∈ S. In particular 0 ∈ Sdiv. We put S× div = Sdiv ∩ S× = Sdiv \ {0}. Sind = S \ S× div, the subsystem of indivisible roots. We also need the fact that there exists a unique symmetric bilinear form (··)u on Y which is invariant under the Weyl group W (S) and which satisfies 2 ∈ {(αα)u : 0 6= α ∈ C} ⊂ {2, 4, 6, 8} for every connected component C of S. This follows easily from [Bou2, Prop. 7]. Observe that S× div = {α ∈ S : (αα)u = 8}. We use (··)u to define short and long roots: Ssh = {α ∈ S : (αα)u = 2} is the set of short roots. Slg = {α ∈ S : (αα)u ∈ {4, 6}} is the set of long roots in S. Thus Slg = S \ (Ssh ∪ Sdiv). For example, for S = BCl we have BCl,sh = {±εi : 1 ≤ i ≤ l}, BC× l,div = {±2εi : 1 ≤ i ≤ l}, BCl,lg = {±εi ± εj : 1 ≤ i 6= j ≤ l}, in particular BC1,lg = ∅, and if S is simply laced, i.e., S× = Ssh, then Sdiv = {0} and Slg = ∅. Example 3.4 (Untwisted affine reflection systems). Let (S, Y, (··)Y ) be a finite root system. Hence 0 ∈ S and Φ = S \ {0}, as stipulated in Example 3.3. Also, let Z be an n-dimensional F -vector space, say with a basis ε1, . . . , εn. We define X = Y ⊕ Z, Λ = Zε1 ⊕ · · · ⊕ Zεn ⊂ Z, R =Sξ∈S{ξ ⊕ λ : λ ∈ Λ} ⊂ Y ⊕ Z, (x1 x2)X = (y1 y2)Y for xi = yi ⊕ zi with yi ∈ Y and zi ∈ Z. By construction we then have X 0 = Z, R0 = Λ, Ran =Sξ∈Φ ξ ⊕ Λ where of course ξ ⊕ Λ = {ξ ⊕ λ : λ ∈ Λ}. For α = ξ ⊕ λ ∈ Ran with ξ ∈ S and λ ∈ Λ the reflection sα satisfies (3.3) sα(y ⊕ z) = sξ(y) ⊕ (z − hy, ξ∨iλ). We will leave it to the reader to verify that (3.4) (R, X) is an affine reflection system of nullity n. Observe that (R, X) is the set of roots of the EALA constructed in 2.5, see in particular (2.18). Observe that spanF (Ran) = X = Re(X) in case S 6= {0}. This shows that the form (··)Re of the real part ℜ(R) of R need not be nondegenerate. Exercise 3.5. Show the claim in (3.4), and also that (R, X) is reduced resp. connected if and only if (S, Y ) is so. 28 ERHARD NEHER Example 3.6 (Affine root systems). By definition, an affine root system is the set of roots of an affine Kac-Moody Lie algebra, which we studied in §1.1 and then again in §2.4, where we showed that an affine Kac-Moody algebra is an EALA of nullity 1. Our goal here is not surprising. We want to show that (3.5) an affine root system is an affine reflection system of nullity 1. Let us first collect the data necessary to prove this. We use the notation established in 2.4. Thus, L = L(g, σ) is an affine Kac-Moody Lie algebra over C, σ is a diagram automorphism of the simple finite-dimensional Lie algebra g of order m ∈ {1, 2, 3}, and ∆¯s denotes the set of weights of the (g¯0, h¯0)-module g¯s ⊂ g, s = 0, . . . , m − 1. One knows that ∆¯0 is a reduced irreducible root system in h∗ ¯0 =: Y . The roots of L with respect to H = h¯0 ⊕ Cc ⊕ Cd are R = {γ ⊕ nδ : γ ∈ ∆¯s, ¯n = ¯s, 0 ≤ s < m}, see (2.13), hence X = spanC(R) = Y ⊕ Cδ. The bilinear form (··)X used to determine the (an)isotropic roots in R has the form (x1 x2)X = (y1 y2)Y where xi = yi ⊕ aiδ with yi ∈ Y and ai ∈ C, and where (··)Y is the nondegenerate symmetric bilinear form on Y , obtained by transporting the Killing form κ h¯0×h¯0 from h¯0 to Y . It follows that X 0 = Cδ and Ran = {γ ⊕ nδ ∈ R : γ 6= 0}. We can now verify the axioms (AR1)–(AR4). (AR1) holds by definition. (AR2) is a consequence of [Kac, Prop. 3.7(b)]. Con- cerning (AR3), it follows from the structure of (··)X that (3.6) hx, α∨i = hy, γ ∨i for x = y ⊕ aδ ∈ X and α = γ ⊕ nδ ∈ Ran. This implies that hR, α∨i is a finite set since S = ∆¯0 ∪ · · · ∪ ∆m−1 is a finite set (S is actually a finite root system; for m > 1 see the table below). Moreover hR, α∨i ⊂ Z because L is an integrable L-module ([Kac, Lemma 3.5]). Thus (AR3) holds, and (AR4) follows from (3.6) and (γ γ) = 0 ⇔ γ = 0 for γ ∈ S. This proves (3.5). To motivate the definition of extension data in Def. 3.9 and the Structure The- orem 3.10 for affine reflection systems, we will now look at R and S more closely. In the untwisted case, i.e., m = 1, we have of course ∆¯0 = S, R = S × Zδ (m = 1). Thus R is an untwisted affine root system of nullity 1, a special case of the Exam- ple 3.4. For m = 2, 3 the structure of ∆¯s and S is summarized in the table below. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 29 Proofs can be extracted from [Kac, 7.9, 7.8, 8.3]. (g, m) ∆¯0 ∆¯1 S (A2l, 2), l ≥ 1 A1 or Bl(l ≥ 2) ∆¯0 ∪ {±2εi : 1 ≤ i ≤ l} BCl (3.7) (A2l−1, 2), l ≥ 2 (Dl+1, 2), l ≥ 3 (E6, 2) (D4, 3) Cl Bl F4 G2 {0} ∪ Cl,sh {0} ∪ Bl,sh {0} ∪ F4,sh {0} ∪ G2,sh Cl Bl F4 G2 We can now rewrite R. For subsets T ⊂ S and Ξ ⊂ Zδ we put and abbreviate B1 = A1 = {0, ±α}. For m = 2 we get T ⊕ Ξ = {τ ⊕ nδ : τ ∈ T, nδ ∈ Zδ} R = (∆¯0 ⊕ 2Zδ) ∪(cid:0)∆¯1 ⊕ (1 + 2Z)δ(cid:1) =(({0} ⊕ Zδ) ∪(cid:0)(Bl \{0} ⊕ Zδ(cid:1) ∪(cid:0) BC× ({0} ⊕ Zδ) ∪ (Ssh ⊕ Zδ) ∪ (Slg ⊕ 2Zδ), div ⊕(1 + 2Z)δ(cid:1), g = A2l g 6= A2l . For (g, m) = (D4, 3) one knows ∆¯1 = ∆¯2 = {0} ∪ G2,sh, whence R = (∆¯0 ⊕ 3Zδ) ∪(cid:0)∆¯1 ⊕ (1 + 3Z)δ(cid:1) ∪(cid:0)∆¯2 ⊕ (2 + 3Z)δ(cid:1) = ({0} ⊕ Zδ) ∪ (Ssh ⊕ Zδ) ∪ (Slg ⊕ 3Zδ). In all three cases R has a simultaneous description in terms of the root system S and subsets Λsh, Λlg, Λdiv ⊂ Zδ as R = R0 ∪ (Ssh ⊕ Λsh) ∪ (Slg ⊕ Λlg) ∪ (S× div ⊕ Λdiv) Λsh = Zδ = R0, Λdiv = (1 + 2Z)δ, Λlg =  Zδ, g = A2l, m = 2, 2Zδ, g 6= A2l, m = 2, 3Zδ, m = 3. If we define Λξ for ξ ∈ S by Λξ ∈ {Λ0 = R0, Λsh, Λlg, Λdiv} according to ξ belong to the corresponding subset of S, then (3.8) becomes (3.8) where (3.9) (3.10) Note that we also recover [Kac, Th. 5.6(b)]: R =Sξ∈S ξ ⊕ Λξ. R ∩ X 0 = R ∩ Zδ. Example 3.7 (Type A1 generalized ). We consider a final example of an affine reflection system to motivate the definition of an extension datum in 3.9 below. Let Z be a finite-dimensional F -vector space and define the vector space X and a symmetric bilinear form on X by X = F α ⊕ Z, (a1α ⊕ z1 a2α ⊕ z2) = a1a2, where 0 6= α and ai ∈ F . We define R ⊂ X in terms of three non-empty subsets Λ0, Λα, Λ−α ⊂ Z as follows: (3.11) R = Λ0 ∪ (α ⊕ Λα) ∪ (−α ⊕ Λ−α). 30 ERHARD NEHER It is then immediate that X 0 = Z, R0 = Λ0, Ran = (α ⊕ Λα) ∪ (−α ⊕ Λ−α). We will now discuss under which conditions (R, X, (··)) is an affine reflection sys- tem. Let us start with (AR2). For si ∈ {±1}, µ ∈ Λs1α and λ ∈ Λs2α we have hs1α ⊕ µ, (s2α ⊕ λ)∨i = 2 s1s2, ss2α⊕λ(s1α ⊕ µ) = −s1α ⊕ (µ − 2s1s2λ) Hence, all reflections ss2α⊕λ leave R invariant if and only if µ−2s1s2λ ∈ Λ−s1α for µ, λ as above, i.e., in obvious short form Λs1α − 2s1s2Λs2α ⊂ Λ−s1α. In particular, Λα − 2Λα ⊂ Λ−α, Λ−α − 2Λ−α ⊂ Λα Λ−α + 2Λα ⊂ Λα for s1 = −1 = −s2. for s1 = s2, For λ ∈ Λα we therefore get λ − 2λ = −λ ∈ Λ−α, whence −Λα ⊂ Λ−α and, analogously, Λ−α ⊂ −Λα. We therefore obtain (3.12) Λ−α = −Λα, or with the notation of above Λs1α = s1Λα. It is now easy to see that (AR2) is equivalent to the two conditions (3.12) and (3.13) 2Λα − Λα ⊂ Λα. It then follows that R is an affine reflection system if and only if (i) (3.12) and (3.13) hold, (ii) 0 ∈ Λ0, and (iii) Z = spanF (Λ0 ∪ Λα ∪ Λ−α). Observe the similarity with the previous examples: R has the form R =Sξ∈S ξ ⊕ Λξ where S = {0, ±α} is a finite root system and (Λξ : ξ ∈ S) is a family of subsets in X 0. However, in the previous examples the Λξ were subgroups of (Z, +) while here we only have the condition (3.13). Does this imply that Λα is a subgroup? The answer is no! For example, in Z = F the subset Λα = 1 + 2Z ⊂ F satisfies (3.13). A subset A of an abelian group (Z, +) is called a reflection subspace if 2a1−a2 ∈ A for all ai ∈ A (see [L] or [Ne5, 3.3] for a justification for this terminology). Hence, (3.13) just says that Λα is a reflection subspace. The structure of two special types of reflection subspaces is described in Exercise 3.8 below. While in general Λα is far from being a subgroup, one can always "re-coordina- tize" R to at least get 0 ∈ Λα. Namely, for a fixed λ ∈ Λα we have α + Λα = (α + λ) + (Λα − λ). Hence, with α = α + λ and Λ α = Λα − λ, we obtain (3.14) R = Λ0 ∪ (α + Λ α) ∪(cid:0) − (α + Λ α)(cid:1), where now Λ α not only satisfies (3.13) but also 0 ∈ Λ α. In other words, Λ α is a pointed reflection subspace as defined in Lemma 3.8 and therefore also satisfies Λ α = −Λ α. The process of re-coordinatization works well in this example. The reason is that the finite root system S in (3.14) is reduced. Re-coordinatization will not work if S is not reduced, as for example in the case (g, m) = (A2l, 2) of Example 3.6. This "explains" why in the property (ED2) of an extension datum in 3.9 we require 0 ∈ Λξ only for an indivisible root ξ ∈ S. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 31 Exercise 3.8. Let A be a subset of an abelian group (Z, +). As above we put 2A − A = {2a1 − a2 : ai ∈ A}. We denote by Λ = spanZ(A) the Z-span of A in Z. A subset A ⊂ Z is called symmetric if A = −A. (a) The following equivalent conditions characterize symmetric reflection sub- spaces A ⊂ Z: (i) 2A − A ⊂ A and A = −A, (ii) 2λ + a ∈ A for every λ ∈ Λ and a ∈ A, (iii) A is a union of cosets modulo 2Λ, (iv) a1 − 2a2 ∈ A for all ai ∈ A. (b) The following are equivalent for A ⊂ Z: (i) 0 ∈ A and A − 2A ⊂ A, (ii) 0 ∈ A and 2A − A ⊂ A, (iii) 2Z[A] ⊂ A and 2Z[A] − A ⊂ A, (iv) A is a union of cosets modulo 2Z[A], including the trivial coset 2Z[A]. In this case A is called a pointed reflection subspace. (c) Every pointed reflection subspace is symmetric. (d) If A is a symmetric reflection subspace then A + A is a pointed reflection subspace. 3.3. The Structure Theorem of affine reflection systems. After the many examples in 3.2, the following definition should not be too surprising. Definition 3.9. Let S be a finite root system as defined in 3.3. Recall S× = S \{0} and Sind = {0} ∪ {α ∈ S : α/2 6∈ S} = S \ S× div. Also, let Z be a finite-dimensional F -vector space. An extension datum of type (S, Z), sometimes simply called an extension datum, is a family (Λξ : ξ ∈ S) of subset Λξ ⊂ Z satisfying the axioms (ED1)–(ED3) below. (ED1): For η, ξ ∈ S×, µ ∈ Λη and λ ∈ Λξ we have µ − hη, ξ∨iξ ∈ Λsξ(η), in obvious short form Λη − hη, ξ∨iΛξ ⊂ Λsξ(η). (ED2): 0 ∈ Λξ for ξ ∈ Sind, and Λξ 6= ∅ for ξ ∈ S \ Sind = S× div. (ED3): Z = spanF (cid:0)Sξ∈S Λξ(cid:1). The axiom (ED1) is trivially true for η = 0 since hη, ξ∨i = 0 and sξ(0) = 0. Also, if S× div = ∅, then there is no Λξ for ξ ∈ S× div and so the second condition in (ED2) is trivially fulfilled. (ED3) simply serves to determine Z. If it does not hold, one can simply replace Z by spanZ(Sξ∈S Λξ). The definition of an extension datum above is a special case of the notion of an extension datum for a pre-reflection system, introduced in [LN, 4.2]. (The reader will note that the axiom (ED1) in [LN] simplifies since in our setting the subset Sre of [LN] is Sre = San = S \ {0}.) The Structure Theorem 3.10 below is proven in [LN, Th. 4.6] for extensions of pre-reflection systems. Affine reflection systems are special types of such extensions, namely finite-dimensional extensions of finite root systems. The rationale for the concept of an extension datum is the following Structure Theorem for affine reflection systems. 32 ERHARD NEHER Theorem 3.10 (Structure Theorem for affine reflection systems). (a) Let (S, Y, (··)Y ) be a finite root system and let L = (Λξ : ξ ∈ S) be an extension datum of type (S, Z). Define (R, X, (··)X ) by X = Y ⊕ Z (y1 ⊕ z1 y2 ⊕ z2)X = (y1 y2)Y R =Sξ∈S ξ ⊕ Λξ ⊂ Y ⊕ Z = X, for yi ∈ Y and zi ∈ Z. Then (R, X, (··)X ) is an affine reflection system, denoted A(S, L), with For α = ξ ⊕ λ ∈ Ran and x = y ⊕ z ∈ X the reflection sα is given by R0 = Λ0, X 0 = Z and Ran =S06=ξ∈S ξ ⊕ Λξ. sα(x) = sξ(y) ⊕ (z − hy, ξ∨iλ) (b) Conversely, let (R, X, (··)X ) be an affine reflection system. (i) Let f : X → X/X 0 =: Y be the canonical map, put S = f (R) and let (··)Y be the induced bilinear form on Y , that is (f (x1) f (x2))Y = (x1 x2)X . Then (S, Y, (··)Y ) is a finite root system, the so-called quotient root system of (R, X). (ii) There exists a linear map g : Y → X satisfying f ◦g = IdY and g(Sind) ⊂ R. (iii) For g as in (ii) and ξ ∈ S define Λξ ⊂ Ker(f ) =: Z by (3.15) R ∩ f −1(ξ) = g(ξ) ⊕ Λξ. Then L = (Λξ : ξ ∈ S) is an extension datum of type (S, Z). (iv) (R, X) is isomorphic to the affine reflection system A(S, L) constructed in (a). Let us note that it is not reasonable to expect g(S) ⊂ R in (b.ii) above, since R may be reduced while S is not, see for example the case (g, A2l) in 3.6. The quotient root system S is uniquely determined, but not so the extension datum, see [LN, Th. 4.6(c)]. Corollary 3.11. An affine reflection system (R, X, (··)) is nondegenerate in the sense that (··) is nondegenerate if and only if R is a finite root system. Proof. If (R, X, (··)) is an affine reflection system with a nondegenerate form (··), then {0} = X 0 = Ker f , so f is the identity. We have seen the other direction in Example 3.3. (cid:3) Corollary 3.12 ([LN, Cor. 5.5]). Let (R, X, (··)) be an affine reflection system over F = R. Then there exists a positive semidefinite symmetric bilinear form (··)≥ on X such that (R, X, (··)≥) is an affine reflection system with the same (an)isotropic roots and reflections. The morale of the Structure Theorem is affine reflection system = finite root system + extension datum Thus properties of an affine reflection system can be described in terms of properties of its quotient root system and the associated extension datum. Some examples of this philosophy are given in the Proposition 3.13 and the Exercise 3.14 below. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 33 Proposition 3.13 ([LN, Cor. 5.2]). Let R be an affine reflection system, let S be its quotient root system and let (Λξ : ξ ∈ S) be the associated extension datum. We define Then ZΛdiff = Λdiff. Moreover: Λdiff =S06=ξ∈S Λξ − Λξ. (a) R is tame in the sense that R0 ⊂ Ran − Ran if and only if R0 ⊂ Λdiff. (b) All root strings S(β, α) = R ∩ (β + Zα), (β ∈ R, α ∈ Ran) are unbroken, i.e., Z(β, α) = {n ∈ Z : β + nα ∈ R} is either a finite interval in Z or equals Z, if and only if Λdiff ⊂ R0. (c) A tame affine reflection system with unbroken root strings is symmetric. Exercise 3.14. Let R be an affine reflection system, let S be its quotient root system and let (Λξ : ξ ∈ S) be the associated extension datum. We use the notation of Prop. 3.13. Prove: (a) R is reduced if and only if for all 0 6= ξ ∈ S with 2ξ ∈ S we have Λ2ξ ∩ 2Λξ = ∅. In particular, S need not be reduced for R to be reduced! (b) R is connected iff S is connected (= irreducible). (c) R is symmetric, i.e., R = −R, iff Λ0 is symmetric. (d) For all α ∈ Ran and β ∈ R the α-string through β, i.e., S(β, α) has length S(β, α) ≤ 5. (e) Let (α, β) ∈ Ran × R and define d, u ∈ N by put −d = min Z(β, α) and u = max Z(β, α). Then u − d = hβ, α∨i. We will now describe how the examples of affine reflection systems of section 3.2 fit into the general scheme of the Structure Theorem above. Examples 3.15. (a) Let L = (Λξ : ξ ∈ S) be an extension datum of type (S, Z). Observe that the only conditions on Λ0 are 0 ∈ Λ0 from (ED2) and that Λ0 together with the other Λξ's spans Z from (ED3). This is in line with our earlier observation that one has little control over the null roots R0 of an affine reflection system. Following the Example 3.2 we define a new extension datum Re(L) = (Re(Λξ) : ξ ∈ S) of type (S, Re(Z)) by Re(Z) = spanF (cid:0)S06=ξ∈S Λξ(cid:1), Re(Λξ) =({0} for ξ = 0, for ξ 6= 0. Λξ If L is the extension datum associated to the affine reflection system (R, X), then Re(L) is the extension datum associated to the affine reflection system Re(R). (b) All Λξ = {0}, whence Z = {0}, defines a trivial extension datum for any root system S. It is "used" when we view S as an affine reflection system, as done in Example 3.3. (c) Let Λ be a subgroup of a finite-dimensional vector space Z such that spanF (Λ) = Z. Then for any finite root system S the family (Λξ ≡ Λ : ξ ∈ S) is an extension datum of type (S, Z). It is used to construct the untwisted affine reflection systems of Example 3.4. 34 ERHARD NEHER (d) Let R be an affine root system. We have seen that R is an affine reflection system. Its quotient root system S and associated extension datum (Λξ : ξ ∈ S) are described in Example 3.6 using the very same symbols, see the formulas (3.10) and (3.15). (e) The family L = (Λ0, Λ α, Λ− α) in Example 3.7 is an extension datum, but not necessarily L = (Λ0, Λα, Λ−α) since 0 need not lie in Λ±α. In fact, replacing L by L was the rationale for the re-coordinatization in 3.7. To describe the classification of affine reflection systems we need some more properties of the subsets Λξ of an extension datum. They are given in the following exercise (just do it!). Recall from Exercise 3.8 that a reflection subspace A is called symmetric if A = −A and is called pointed if 0 ∈ A. Exercise 3.16. Let (Λη : η ∈ S) be an extension datum of type (S, Z). Show: (a) Every Λξ for 0 6= ξ ∈ S is a symmetric reflection subspace and is even a pointed reflection subspace if ξ ∈ S× ind. (b) For w ∈ W (S), the Weyl group of S, we have (3.16) Λξ = Λw(ξ). In particular, Λξ = Λ−ξ = −Λξ. (c) Whenever 0 6= ξ ∈ S and 2ξ ∈ S, then Λ2ξ ⊂ Λξ. (d) ZΛξ ⊂ Λξ for ξ ∈ S× ind. Let (Λξ : ξ ∈ S) be an extension datum where S is irreducible. Then W (S) acts transitively on the roots of the same length ([Bou2, VI, §1.3, Prop. 11]), i.e., on {0}, Ssh, Slg and Sdiv (some of these sets might be empty). Because of (3.16), there are therefore at most four different subsets Λ0, Λsh, Λlg and Λdiv among the Λξ, defined by Λ0, Λsh, Λlg, Λdiv, ξ = 0; ξ ∈ Ssh; ξ ∈ Slg; ξ ∈ S× div. Λξ =  (3.17) Of course, Λlg or Λdiv only exists if the corresponding subset of roots exits. The assertions below referring to Λlg or Λdiv should be interpreted correspondingly. We have seen in Exercise 3.16 (did you do it?), that the subsets Λsh and Λlg are pointed reflection subspaces and that Λdiv is a symmetric reflection subspace. Assuming only these properties, does however not give an extension datum, since only parts of the axiom (ED1) are fulfilled, namely those with η = ±ξ. We also need to evaluate what happens for η 6= ±ξ with hη, ξ∨i 6= 0. We will do this in the following examples. Examples 3.17. Let S be an irreducible root system. We suppose that we are given a pointed reflection subspace Λsh of a finite-dimensional vector space, and if Slg 6= ∅ or S× div 6= ∅ then also a pointed reflection subspace Λlg and a symmetric reflection subspace Λdiv. We define Λξ, ξ ∈ S, by (3.17) and ask, when is the family Lmin defined in this way an extension datum in Z = span(cid:0)Sξ∈S Λξ(cid:1)? Note that we only have to check (ED1). We will consider some examples of S. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 35 (a) S = A1: In this case there are no further conditions, so Lmin describes all possible extension data for A1 with Λ0 = {0}. (b) S = A2: In this case there exists roots η, ξ with hη, ξ∨i = 1, namely those for which ∠(η, ξ) = π 3 . Evaluating (ED1) for those gives Λsh − Λsh ⊂ Λsh, forcing Λsh to be a subgroup of Z. Thus, Lmin is an extension datum for S = A2 iff Λsh is a subgroup. (c) S simply laced, rank S ≥ 2: The argument in (b) works whenever Ssh contains roots η, ξ with hη, ξ∨i = 1. Since this is the case here, we get that Lmin is an extension datum iff Λsh is a subgroup of Z. (d) S = B2 = {±ε1, ±ε2, ±ε1 ± ε2}: Here we have pointed reflection subspaces Λsh and Λlg. Since non-zero roots of the same length are either proportional or orthogonal, (ED1) is fulfilled for them. Because (ED1) is invariant under sign changes, we are left to evaluate the case of two roots η, ξ of different lengths forming an obtuse angle of 3π 4 , for example η = ε1, ξ = ε2 − ǫ1. If hη, ξ∨i = −2, then η is long, ξ is short and so (ED1) becomes Λlg + 2Λsh ⊂ Λlg. If η is short, ξ is long, we have hη, ξ∨i = −1 and thus get the condition Λsh + Λlg ⊂ Λsh. To summarize: Lmin is an extension datum for S = B2 iff Λsh and Λlg are pointed reflection subspaces satisfying (3.18) Λlg + 2Λsh ⊂ Λlg and Λsh + Λlg ⊂ Λsh. Note that (3.18) implies 2Λsh ⊂ Λlg ⊂ Λsh. (e) S = Bl, l ≥ 3. Recall S = {±ǫi : 1 ≤ i ≤ l} ∪ {±εi ± εj : 1 ≤ i 6= j ≤ l}. Since the short roots in S are either proportional or orthogonal, (ED1) is fulfilled for all short roots η, ξ. But there exist long roots η, ξ ∈ S with ∠(η, ξ) = π 3 , whence hη, ξ∨i = 1 and so (ED1) reads Λlg − Λlg ⊂ Λlg. This forces Λlg to be a subgroup. As for S = B2, (ED1) for roots of different lengths leads to the condition (3.18). It is then easy to check that Lmin is an extension datum for S = Bl, l ≥ 3, iff Λsh is a pointed reflection subspace, Λlg is a subgroup and (3.18) holds. Continuing in this way, one arrives at the following. Theorem 3.18 (Structure of extension data). Let S be an irreducible finite root system and define a family Lmin as in (3.17) with Λ0 = {0}. Then Lmin is an extension datum if and only if Λsh and Λlg are pointed reflection subspaces, Λdiv is a symmetric reflection subspace and the following conditions, depending on S, hold. (i) S is simply laced, rank S ≥ 1 : No further condition for S = A1, but Λsh is a subgroup if rank S ≥ 2. (ii) S = Bl(l ≥ 2), Cl(l ≥ 3), F4 : Λsh and Λlg satisfy Λlg + 2Λsh ⊂ Λlg and Λsh + Λlg ⊂ Λsh. Moreover, • Λlg is a subgroup if S = Bl, l ≥ 3 or S = F4, and • Λsh is a subgroup if S = Cl or S = F4. (iii) S = G2 : Λsh and Λlg are subgroups satisfying Λlg + 3Λsh ⊂ Λlg and Λsh + Λlg ⊂ Λsh. (iv) S = BC1 : Λsh and Λdiv satisfy Λdiv + 4Λsh ⊂ Λdiv and Λsh + Λdiv ⊂ Λsh. (v) S = BCl(l ≥ 2) : Λsh, Λlg and Λdiv satisfy 36 ERHARD NEHER Λlg + 2Λsh ⊂ Λlg, Λdiv + 2Λlg ⊂ Λdiv, Λdiv + 4Λsh ⊂ Λdiv, Λsh + Λlg ⊂ Λsh, Λlg + Λdiv ⊂ Λlg, Λsh + Λdiv ⊂ Λsh. In addition, if l ≥ 3 then Λlg is a subgroup. The inclusions Λdiv + 4Λsh ⊂ Λdiv and Λsh + Λdiv ⊂ Λsh in case (v) above are consequences of the other inclusions. Since 0 lies in Λsh and also in Λlg if it exists, the displayed inclusions in the Structure Theorem above imply (3.19) Λdiv ⊂ Λlg ⊂ Λsh. The details of this theorem are given in [AABGP, II, §2] for the special case of extended affine root systems and then in [Y3] in general (it follows from the Structure Theorem 3.10 that an affine reflection system is the same as a "root system extended by a torsion-free abelian group of finite rank" in the sense of [Y3]). The reference [AABGP] also contains a classification of discrete extension data for extended affine root systems of low nullity. Exercise 3.19. Without looking at [AABGP] or [Y3], work out some of the cases above. 3.4. Extended affine root systems. Let us come back to the beginning of this chapter. Our goal was to describe the structure of the set of roots occurring in an extended affine Lie algebra. After all the preparations in 3.1–3.3, this is now easy. We start with the same setting as in 3.1, i.e., X is a finite-dimensional vector space over a field F of characteristic 0, R is a subset of X and (··) is a symmetric bilinear form on X. As in 3.1 we define R0 = {α ∈ R : (αα) = 0}, Ran = {α ∈ R : (αα) 6= 0} and hx, α∨i = 2 (xα) (αα) , (x ∈ X and α ∈ Ran). Definition 3.20. A triple (R, X, (··)) as above is called an extended affine root system or EARS for short, if the following seven axioms (EARS1)–(EARS7) are fulfilled. (EARS1): 0 ∈ R and R spans X, (EARS2): R has unbroken finite root strings, i.e., for every α ∈ Ran and β ∈ R there exist d, u ∈ N = {0, 1, 2, . . .} such that {β + nα : n ∈ Z} ∩ R = {β − dα, . . . , β + uα} and d − u = hβ, α∨i. (d stands for "down" and u for "up".) (EARS3): R0 = R ∩ X 0. (EARS4): R is reduced as defined in 3.1: for every α ∈ Ran we have F α ∩ Ran = {±α}. (EARS5): R is connected in the sense of 3.1: whenever Ran = R1 ∪ R2 with (R1 R2) = 0, then R1 = ∅ or R2 = ∅. (EARS6): R is tame, i.e., R0 ⊂ Ran + Ran. (EARS7): The abelian group spanZ(R0) is free of finite rank. In analogy with the concept of discrete EALAs we call (R, X, (··)) for F = C or F = R a discrete extended affine root system if (EARS1)–(EARS6) hold and in addition (DE): R is a discrete subset of X, equipped with the natural topology. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 37 As for EALAs, a discrete extended affine root system necessarily satisfies (EARS7), see Proposition 3.21(c) below, so that it is justified to call it an EARS. We will immediately connect EARS to affine reflection systems: Proposition 3.21. (a) A pair (R, X) satisfying (EARS1)–(EARS3) is an affine reflection system. In particular: (i) An extended affine root system is an affine reflection system which is re- duced, connected, symmetric, tame and which has unbroken root strings. (ii) If F = R we can assume that (··) is positive semidefinite. (b) Let (R, X) be an affine reflection system with quotient root system S and extension datum L. Then (R, X) is an extended affine root system if and only if (i) S is irreducible, hence L = (Λ0, Λsh, Λlg, Λdiv), (ii) Λ0 = Λsh + Λsh, Λdiv ∩ 2Λsh = ∅, and (iii) (EARS7) holds. (c) For an extended affine root system (R, X) over F = R or F = C the following are equivalent: (i) R is discrete; (ii) R0 is a discrete subset of X; (iii) spanZ(R0) is a discrete subgroup of X. In this case, all reflection subspaces Λξ of (b) are discrete too, and spanZ(R0) is a free abelian group of finite rank. Proof. (a) Obviously (AR1) = (EARS1) and (AR4) = (EARS3). The axiom (AR2), i.e., sα(R) = R, follows from (EARS2): sα(β) = β + (u − d)α and −d ≤ u − d ≤ u. Also (AR3) is immediate from (EARS2). In any affine reflection system the root strings R ∩ (β + Zα) for (β, α) ∈ R × Ran are finite. This is Exercise 3.14(d) and an immediate consequence of the Structure Theorem 3.10. Also, a tame affine reflection system with unbroken roots strings is necessarily symmetric by Prop. 3.13. The characterization of an EARS in (i) is now clear. (ii) follows from Cor. 3.12. (b) follows from Prop. 3.13 and Exercise 3.14, since for a connected R = irre- ducible S the formula (3.19) implies Λdiff = Λsh + Λsh. (c) (i) ⇒ (ii) is obvious. Suppose (ii) holds. We know from (b) that R0 = Λ0 = Λsh + Λsh. Since Λsh is a pointed reflection subspace, so is Λ0 (Exercise 3.8). Hence 2 spanZ(Λ0) ⊂ Λ0 is discrete. But then so is spanZ(Λ0). Thus (ii) ⇒ (iii). By (3.19) and (b.ii), Λdiv ⊂ Λlg ⊂ Λsh ⊂ Λ0. Hence, if (iii) holds, then all Λξ are discrete subsets of X. But then so is is R, as a finite union of discrete subsets. This shows (iii) → (i). It is well-known fact that every discrete subgroup of a finite-dimensional real (cid:3) vector space is free of finite rank. A quick comparison of [AABGP, Definition 2.1] and our Definition 3.20 together with Prop. 3.21(a) will convince the reader that an extended affine root system in the sense of [AABGP] is the same as a discrete extended affine root system over R in our sense. The reason for the generalization and the change of name is the same as the one justifying our more general notion of extended affine Lie algebras: We are considering EALAs over arbitrary fields of characteristic 0 and the set of roots of an EALA will not be an extended affine root system in the sense of [AABGP]. Finally, here is the result which brings us back to EALAs. 38 ERHARD NEHER Theorem 3.22. Let (E, H) be an EALA over F and let R ⊂ H ∗ be its set of roots. Put X = spanF (R) and let (··)X be the restriction of the bilinear form (2.3) to X. Then (R, X, (··)X ) is an extended affine root system. If F = C, then (E, H) is a discrete EALA if and only if (R, X, (··)X ) is a discrete extended affine root system. Remarks 3.23. (a) For (E, H) a discrete EALA over F = C, the theorem is proven in [AABGP, I, Th. 2.16], using discreteness. The generalization to arbitrary EALAs is due to the author, see [Ne4, Prop. 3]. It has been further generalized to other classes of Lie algebras, the so-called invariant affine reflection algebras, see [Ne5, Th. 6.6 and Th. 6.8]. Special cases have also been proven in [Az1] and [MY]. That R is symmetric, is an easy exercise, namely Exercise 2.1(c). (b) In view of the theorem above, one can ask if every extended affine root system is the set of roots of some extended affine Lie algebra. This is however not the case, see [AG, Th. 6.2] for a detailed discussion of this question. As a first application of this theorem, we can now completely characterize EALAs of nullity 0. Proposition 3.24. The following are equivalent: (i) (E, H) is an EALA of nullity 0, (ii) (E, H) is an EALA with a finite-dimensional E, (iii) E is a finite-dimensional split simple Lie algebra with splitting Cartan sub- algebra H. In this case, E equals its core and the set of roots R coincides with the quotient root system S of R and is an irreducible reduced finite root system. 4. The core and centreless core of an EALA In the previous chapter we have studied affine reflection systems per se. The rationale for doing so became clear only in the end, when we saw in Th. 3.22 that the set of roots R of an EALA (E, H) is an extended affine root system, a special type of an affine reflection system. In this chapter we start by drawing consequences of the Structure Theorem 3.10 of affine reflection systems and the description of extended affine root systems in Prop. 3.21. The examples in §2.4 and §2.5 indicate that the core Ec and centreless core Ecc = Ec/Z(Ec) of an extended affine Lie algebra (E, H) really are the "core" of the matter. We will show in Th. 4.14 and in Cor. 4.16 that both are so-called Lie tori, a new class of Lie algebras which we will introduce in 4.1. We will present some basic properties of Lie tori in 4.2 and describe some examples in 4.4 and 4.5. With some justification, this chapter could therefore also be entitled "On Lie tori". But the reader can be re-assured that we are not getting side-tracked too much: In the next chapter we will see that Lie tori are precisely what is needed to construct EALAs. 4.1. Lie tori: Definition. Lie tori are special objects in the following category of graded Lie algebras. Definition 4.1. Let (S, Y ) be a finite irreducible, but not necessarily reduced root system, as defined in Example 3.3. We denote by Q(S) = spanZ(S) ⊂ Y the root lattice of S. To avoid some degeneracies we will always assume that S 6= {0}. Let Λ be an abelian group. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 39 A (Q(S), Λ)-graded Lie algebra is a Lie algebra L with compatible Q(S)- and Λ-gradings. It is convenient (and helpful) to use subscripts for the Q(S)-grading and superscripts for the Λ-grading. Thus, are Q(S)- and Λ-gradings of L, and compatibility means L =Lq∈Q(S) Lq =Lλ∈Λ Lλ Hence for λ, µ ∈ Λ and p, q ∈ Q(S) L =Lq∈Q(S), λ∈Λ Lλ q for Lλ q = Lq ∩ Lλ. Lλ =Lq∈Q(S) Lλ q , Lq =Lλ∈Λ Lλ q and [Lλ q , Lµ p ] ⊂ Lλ+µ q+p . Thus, L has three gradings, by Q(S), Λ and Q(S)⊕Λ whose interplay will be crucial in the following. Corresponding to these three different gradings are three support sets : suppQ(S) L = {q ∈ Q(S) : Lq 6= 0}, suppΛ L = {λ ∈ L : Lλ 6= 0}, and suppQ(S)⊕Λ L = {(q, λ) ∈ (Q(S), Λ) : Lλ q 6= 0}. Definition 4.2. We keep the notation of the Def. 4.1. A Lie torus of type (S, Λ) is a (Q(S), Λ)-graded Lie algebra L over F , a field of characteristic 0, satisfying the axioms (LT1)–(LT3) below. (LT1): suppQ(S) L ⊂ S, hence L =Lξ∈S Lξ. ξ 6= 0 and ξ 6= 0, then there exist eλ (LT2): If Lλ that ξ ∈ Lλ ξ and f λ ξ ∈ L−λ −ξ such (4.1) (4.2) Lλ ξ = F eλ ξ , L−λ −ξ = F f λ ξ , and for xτ ∈ Lτ we have [[eλ ξ 6= 0 if ξ ∈ S× (LT3): (a) L0 ξ , f λ ξ ], xτ ] = hτ, ξ∨ixτ . ind, i.e., 0 6= ξ ∈ S and ξ/2 6∈ S. (b) As a Lie algebra, L is generated by S06=ξ∈S Lξ. (c) Λ = spanZ(suppΛ L). We will say that L is a Lie torus if L is a Lie torus for some pair (S, Λ). A Lie torus is called invariant, if L has an invariant nondegenerate symmetric bilinear form (··) which is graded in the sense that (4.3) (Lλ ξ Lµ τ ) = 0 if λ + µ 6= 0 or ξ + τ 6= 0. Two Lie tori L and L, both of type (S, Λ), are called graded-isomorphic if there exists a Lie algebra isomorphism f : L → L such that f (Lλ ξ for all (ξ, λ) ∈ S × Λ. Thus, a graded-isomorphism is an isomorphism in the category of graded Lie algebras. But we will use the term "graded-isomorphism" to emphasize that Lie tori are graded algebras. ξ ) = Lλ Remarks 4.3. (a) Let L be a Lie torus. Hence, by (LT1), L = Lξ∈S Lξ = Lξ∈S, λ∈Λ Lλ ξ . We will determine suppQ(S) L in Cor. 4.6 below. The axiom (LT2) implies that dim Lλ ξ = 1 if 0 6= ξ and Lλ ξ 6= 0 (4.4) and that (4.5) (eλ ξ , hλ ξ , f λ ξ ) with hλ ξ = [eλ ξ , f λ ξ ] ∈ L0 0 40 ERHARD NEHER is an sl2-triple. The condition (LT3.a) together with (LT2) ensures that a Lie torus has enough sl2-triples. The other two conditions in (LT3) are not really serious; they just serve to nor- malize things: If (LT3.c) does not hold, one can simply replace Λ by spanZ(suppΛ L). Also, (4.6) (LT3.b) ⇐⇒ Lλ for all λ ∈ Λ. If one has a Lie algebra, for which all axioms except (4.6) hold, one can replace the subspaces Lλ 0 by the right hand side of (4.6) and then gets a Lie torus. Observe that (4.6) for λ = 0 together with (LT2) yields 0 =P06=ξ∈SPµ∈Λ [Lµ ξ , Lλ−µ −ξ ] (4.7) L0 0 =P F hλ ξ where the sum in (4.7) is taken over all pairs (ξ, λ) for which hλ with Lλ ξ 6= 0 and ξ 6= 0. ξ exists, i.e., those (b) A Lie torus is a special type of a so-called division-(S, Λ)-graded Lie alge- bras, or more generally of a root-graded Lie algebra. This and also the different approaches to root-graded Lie algebras are discussed in [Ne5, §5]. Lie tori were first defined by Yoshii in [Y3, Y4], using the notion of a root-graded Lie algebra. The definition above is due to the author [Ne3]. Viewing a Lie torus as a special type of a root-graded Lie algebra is the approach used in the classification of Lie tori. (c) Why was a Lie torus christened a "Lie torus"? The historically correct answer is: Because of pure analogy with already existing names like a quantum torus, defined in 4.25, or an alternative or Jordan torus. All of these are graded algebras, in which every non-zero homogeneous element is invertible. If one interprets the elements e and f of the sl2-triple (4.5) as invertible elements of L, then a Lie torus is a graded Lie algebra in which most of the non-zero homogenous elements are invertible. It is certainly unusual to speak of "invertible elements" in a Lie algebra. But the examples below will provide some justification to that: We will see that the invertible elements of L are given by invertible elements of its coordinate algebra. Besides the analogy with the already existing concepts of "tori" in categories of (non)associative algebras, the fact that a toroidal Lie algebra is a Lie torus, see 4.31, reinforces the choice of the name Lie torus. 4.2. Some basic properties of Lie tori. Throughout this section L is a Lie torus of type (S, Λ). We use the notation of Def. 4.2. We describe some basic properties of Lie tori and prove some of them, in particular those for which there does not yet exist a published proof. We first show that the homogeneous subspaces of the Q(S)-grading of L are weight spaces for the ad-diagonalizable subalgebra ξ : ξ ∈ S× ind}. Lemma 4.4. The subspaces Lτ , τ ∈ S, are given by h = spanF {h0 (4.8) Lτ = {l ∈ L : [h0 ξ, l] = hτ, ξ∨il for all ξ ∈ S× ind}. Proof. The inclusion from left to right holds by (4.2). For the proof of the other ξ, l] = ind if and only if for every α ∈ S we have hα − τ, ξ∨ilα = 0 for ind. Since spanF (Sind) = Y and the bilinear form on Y associated with the inclusion we write l ∈ L as l = Pα lα with lα ∈ Lα. Then l satisfies [h0 hτ, ξ∨il for all ξ ∈ S× all ξ ∈ S× LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 41 root system S is nondegenerate, for every pair (α, τ ) ∈ S2 with α 6= τ there exists ξ ∈ S× ind with hα − τ, ξ∨i 6= 0. Hence, any l belonging to the set on the right hand side of (4.8) has lα = 0 for α 6= τ , proving l ∈ Lτ . (cid:3) Proposition 4.5. For every (ξ, λ) ∈ suppQ(S)⊕Λ L with ξ 6= 0 the map is a well-defined automorphism of the Lie algebra L with the property ξ = exp(cid:0) ad(eλ ξ )(cid:1) exp(cid:0) ad(−f λ ξ )(cid:1) exp(cid:0) ad(eλ ξ )(cid:1) ϕλ (4.9) ϕλ ξ (Lµ τ ) = Lµ−hµ,ξ∨iλ sξ(τ ) . Moreover, for every w ∈ W (S), the Weyl group of S, there exists an automorphism ϕw of the Lie algebra L such that ϕw(Lµ τ ) = Lµ w(τ ) for all τ ∈ S and µ ∈ L. This proposition can be proven in the same way as [AABGP, Prop. 1.27]. Corollary 4.6 ([ABFP2, Lemma 1.10]). The Q(S)-support of L satisfies suppQ(S) L =(S S or Sind if S is reduced, if S is non-reduced. As a consequence of this corollary, suppQ(S) L is always a finite irreducible root system. It is immediate that L is also a Lie torus of type (suppQ(S), Λ). Without loss of generality we can therefore assume that S = suppQ(S) L if this is convenient. Proposition 4.7. For ξ ∈ S define Λξ = {λ ∈ Λ : Lλ ξ 6= 0}, so that suppΛ L = Sξ∈S Λξ. Then the family (Λξ : ξ ∈ S) satisfies the axioms (ED1) and (ED2) of Def. 3.9, (ED1) Λη − hη, ξ∨iΛξ ⊂ Λsξ(η) 0 ∈ Λξ for ξ ∈ Sind and Λξ 6= ∅ for ξ ∈ S× div (ED2) Hence Λξ is a pointed reflection subspace for ξ ∈ S× subspace for ξ ∈ S× div and ind, a symmetric reflection (4.10) Λξ = Λw(ξ) for all w ∈ W (S). Defining Λsh, Λlg and Λdiv as in 3.17, we have (4.11) (4.12) (4.13) (4.14) Λsh ⊃ Λlg ⊃ Λdiv, suppΛ L = Λ0 = Λsh + Λsh, ∅ = 2Λsh ∩ Λdiv, Λ = spanZ(Λsh). The support families (Λξ : ξ ∈ S) are the same for L and L/Z(L). Proof. (ED1) is a consequence of Prop. 4.5 and (ED2) of (LT3.a). It then follows as in section 3 that Λξ, ξ ∈ S×, are pointed respectively symmetric reflection subspaces such that (4.10) and (4.11) hold. (4.12) and (4.13) are proven in [Y3, Th. 5.1] and [ABFP2, Lemma 1.1.12]. (4.14) follows from (4.11) and (4.12). The last claim is also proven in [Y3, Th. 5.1]. (cid:3) 42 ERHARD NEHER Proposition 4.8. Define g = subalgebra generated by {L0 h = spanF {h0 ξ : ξ ∈ S× ind}. ξ : ξ ∈ S× ind}, (a) Then g is a finite-dimensional split simple Lie algebra with splitting Cartan subalgebra h. (b) The root system Sind and the root system of (g, h) are canonically isomorphic. η) = hξ, η∨i ind, such that the map ξ 7→ ξ extends to an isomorphism between the root ind there exists a unique ξ ∈ h∗, defined by ξ(h0 Namely, for every ξ ∈ S× for η ∈ S× system Sind and the root system of (g, h). Proof. This is a special case of a result for arbitrary root-graded Lie algebras, see [Ne1, Remark 2 of §2.1] and [Ne5, Prop. 5.9]. It is essentially a corollary to the Chevalley-Serre presentation of finite-dimensional split simple Lie algebras. For Lie tori it was announced in [Ne3, §3]. The details of the proof are given in [ABFP2, Prop. 1.2.2]. (cid:3) The following Ex. 4.10 lists some more basic properties of Lie tori. You will need Ex. 4.9(a) in part (d) of 4.10. Exercise 4.9. (a) Let K be a perfect Lie algebra. Then K/Z(K) is a perfect and centreless Lie algebra. (b) Let E be a Lie algebra with an invariant nondegenerate symmetric bilinear form (··), and let K be an ideal of E with K = [E, K]. Then {z ∈ K : (z K) = 0} = Z(K). Exercise 4.10. Let L be a Lie torus of type (S, Λ). Show: 0 is given by the formula (4.6). (a) Lλ (b) L is perfect. (c) The centre satisfies Z(L) =Lλ∈Λ Z(L)λ for Z(L)λ = Z(L) ∩ Lλ (d) Let Y =Lλ∈Λ Y λ 0 , be a graded subspace of Z(L). Then L/Y ξ , where for ξ 6= 0 we ξ as vector spaces. In particular, L/Z(L) 0 = Y ∩ Lλ is a Lie torus with respect to the subspaces (L/Y )λ put Y λ is a centreless Lie torus. ξ = {0} and thus have (L/Y )λ ξ = Lλ 0 , Y λ ξ /Y λ ∼= Lλ 0 . ξ (e) For λ, µ ∈ Λξ, ξ ∈ S×, we have hλ (f) L0 = g ⊕ Z(L)0 and L0 0 = h ⊕ Z(L)0. ξ ≡ hµ ξ mod Z(L). (g) Let I be a Λ-graded ideal of L, whence I =Lλ∈Λ I λ for I λ = I ∩ Lλ. Then either I = L or I ⊂ Z(L). In particular, a centreless Lie torus is graded-simple with respect to the Λ-grading of L. Since a Lie torus is perfect by part (b) of the exercise above, it has a universal central extension (Th. 1.10). Theorem 4.11 ([Ne3, §5], [Ne6]). Let u : uce(L) → L be a universal central ξ of type (S, Λ). Then uce(L) is also a Lie extension of a Lie torus L = Lξ,λ Lλ torus of type (S, Λ), say uce(L) =Lξ,λ uce(L)λ Remark 4.12. It follows from this theorem and the exercise above that in order to describe Lie tori up to graded isomorphism, one can proceed in two steps: (A) Classify centreless Lie tori, up to graded isomorphism. We will discuss some examples in 4.4 and 4.5. ξ , and u maps uce(L)λ ξ onto Lλ ξ . LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 43 (B) Describe the universal central extension of the centreless Lie tori from (A). They are unique up to isomorphism. We will not say anything about this here. The reader can find some results in [BGK, BGKN, Ne4, Ne6] for Lie tori arising from EALAs and in [ABG1, ABG2, BS, Ne6] for general root-graded Lie algebras. Once (A) and (B) completed, an arbitrary Lie torus of type (S, Λ) is then obtained as uce(L)/C where L is taken from the list in (A) and where C is a graded subspace of the centre of uce(L). The following results only holds for special types of Lie tori. Theorem 4.13 ([Ne3, Th. 5], proven in [Ne6]). Let L = Lξ∈S, λ∈Λ Lλ torus of type (S, Λ) where Λ is a finitely generated abelian group. (a) Then L is finitely generated as Lie algebra and has bounded homogeneous ξ be a Lie dimension with respect to the Q(S) ⊕ Λ-grading of L. (b) Moreover, the Lie algebra DerF (L) = grDerF (L) where grDerF (L) is natu- rally Q(S) ⊕ Λ-graded and has bounded homogeneous dimension with respect to this grading. (c) If L is invariant, its universal central extension is isomorphic to the cen- tral extension E(L, Dgr∗, ψD) where D is any graded complement of IDer(L) in SDerF (L). Part (c) of this theorem is an immediate corollary of parts (a) and (b) and of Th. 1.25. 4.3. The core of an EALA. We will now connect extended affine Lie algebras and Lie tori, and first introduce some notation. Let (E, H) be an EALA with set of roots R. We have seen in Th. 3.22 that R is an extended affine root system, hence an affine reflection system. We can therefore apply the Structure Theorem 3.10. Recall the following data describing the structure of R: • X = spanF (R) ⊂ H ∗, X 0 = {x ∈ X : (x X) = 0} = {x ∈ X : (x R) = 0}, f : X → X/X 0 = Y the canonical projection, • S = f (R) the quotient root system, a finite irreducible but possibly non- reduced root system, and Sind = {α ∈ S : α/2 6∈ S} ∪ {0}, • g : Y → X a linear map satisfying f ◦ g = IdY and g(Sind) ⊂ R, • (Λξ : ξ ∈ S) the associated extension datum, defined by R ∩ f −1(ξ) = g(ξ) ⊕ Λξ and Λξ ⊂ X 0, • Λ = spanZ(cid:0)Sξ∈S Λξ(cid:1), a free abelian group of finite rank (this is axiom (EARS7)). Hence R =Sξ∈S (cid:0)g(ξ) ⊕ Λξ(cid:1) ⊂ g(Y ) ⊕ X 0, Ran =Sξ∈S×(cid:0)g(ξ) ⊕ Λξ(cid:1), R0 = 0 ⊕ Λ0 = R ∩ X 0. Theorem 4.14 ([AG] for F = C). Let K = Ec be the core of an EALA (E, H). We use the notation of above and define subspaces (4.15) K λ ξ = K ∩ Eg(ξ)⊕λ =(Eg(ξ)⊕λ K ∩ E0⊕λ, ξ 6= 0, ξ = 0. 44 ERHARD NEHER (a) Then K =Lξ, λ K λ finite rank. ξ is a Lie torus of type (S, Λ), where Λ is free abelian of (b) K is a perfect ideal of E. (c) Let (··) be a nondegenerate invariant bilinear form on E, whose existence is guaranteed by the axiom (EA1). Then the radical of the restricted bilinear form (··)K×K equals the centre Z(K), that is (4.16) {z ∈ K : (z K) = 0} = Z(K) =Lλ∈Λ Z(K) ∩ K λ 0 . Remark 4.15. The subspaces K λ ξ in (4.15) and hence the Lie torus structure of K depend on the section g. A different choice of g leads to a so-called isotope of K, see [AF] and [Ne5, Prop. 6.4]. Corollary 4.16. We use the notation of Th. 4.14, and put Ecc = K/Z(K) = L, the centreless core of (E, H). Then L is an invariant centreless Lie torus of type (S, Λ) with respect to the homogeneous subspaces ξ = K λ and the bilinear form (··)L defined by Lλ ξ(cid:14)(cid:0)Z(K) ∩ K λ ξ(cid:1) (¯x ¯y)L = (x y) where x, y ∈ K, ¯x and ¯y are the canonical images in L and (··) is the bilinear form of 4.14(c). Remark 4.17. Yoshii ([Y4]) has shown that any Lie torus of type (S, Λ) with Λ a torsion-free abelian group admits a non-zero graded invariant symmetric bilin- ear form. This implies that the existence of a nondegenerate such form on Ecc. However, Yoshii's proof uses the existence of invariant nondegenerate symmetric bilinear forms on Jordan tori ([NY]) and hence relies on the classification of Jordan tori. We can now show that all root spaces of an EALA are finite-dimensional in a strong form. Proposition 4.18 ([Ne4, Prop. 3]). An EALA has finite bounded dimension. The same is true for its core and centreless core. Proof. Let K = Ec be the core of the EALA (E, H). By Th. 4.14, K is a Lie torus of type (S, Λ), where Λ is a free abelian group of finite rank. Hence, by Th. 4.13, one knows that K has finite bounded dimension with respect to its double grading, say dim K λ ξ ≤ M1 for all pairs (ξ, λ). By the same reference, one also knows that the Lie algebra DerF (K) of all F -linear derivations of K has a double grading by Q(S) and Λ, DerF (K) =Lξ∈S, λ∈Λ(DerF K)λ ξ , where (DerF K)λ ξ+τ , and that DerF (K) has finite bounded dimension with respect to this grading, say dimF (DerF K)λ ξ is the subspace of those derivations mapping K µ ξ ≤ M2 for all pairs (ξ, λ). τ to K λ+µ Since K is an ideal, we have a Lie algebra homomorphism ρ : E → DerF (K), given by ρ(e) = ad eK. It is homogenous of degree 0, i.e., ρ(Eα) ⊂ (DerF K)λ ξ for α = g(ξ) ⊕ λ as in (4.15). Moreover, by the tameness axiom (EA5) for an EALA we know that Ker ρ ⊂ K (whence Ker ρ = Z(K), but we won't need this). It now LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 45 follows that dim Eα = dim Ker(ρEα) + dim ρ(Eα) ≤ dim K λ M1 + M2. ξ + dim(DerF K)λ ξ ≤ (cid:3) 4.4. Lie tori of type Al, l ≥ 3. As explained in Rem. 4.12, in classifying Lie tori one can restrict one's attention to the case of centreless Lie tori, at least modulo the solution of problem (B) in 4.12. In this section we will describe centreless Lie tori of type A. The reader will expect that this will have something to do with trace-0-matrices. This turns out to be correct, but only with the proper interpretation of "trace-0". It will not be sufficient to consider trace-0-matrices over F . We will see in 4.28 that they will only lead to nullity 0-examples. Rather, one must allow matrices with entries from a possibly non-commutative algebra. To avoid some degeneracies, in this section we let N be a natural number with N ≥ 3. We start with an arbitrary associative unital F -algebra A. In particular, A need not be commutative, and hence [A, A] = spanF {a1a2 − a2a1 : a1, a2 ∈ A} is in general non-zero. As usual, glN (A) is the Lie algebra of all N × N matrices with entries in A and Lie algebra product [x, y] = xy − yx, the usual commutator of the matrices x and y. We define the special linear Lie algebra slN (A) as the derived algebra slN (A) = [glN (A), glN (A)]. of glN (A). In particular, slN (A) is an ideal of glN (A). To analyze the structure of slN (A) we use the matrix units Eij , i.e., the N × N matrices with 1 at the position (ij) and 0 at all other positions. They satisfy the basic multiplication rule (4.17) [aEij , bEmn] = δjm ab Ein − δni ba Emj where δ∗ is the usual Kronecker delta. We put EN =PN the Lie algebra slN (A) are listed in the following (very worthwhile) exercise. i=1 Eii. Some properties of Exercise 4.19. (a) slN (A) = {x ∈ glN (A) : tr(x) ∈ [A, A]}. (b) As a vector space, slN (A) decomposes as (4.18) slN (A) = slN (A)0 ⊕(cid:0)Li6=j AEij(cid:1), where slN (A)0 = slN (A) ∩(cid:0)LN i=1 AEii(cid:1) =Pi6=j[AEij , AEji] =Pi6=j spanF {abEii − baEjj : a, b ∈ A} = {cEN : c ∈ [A, A]} ⊕(cid:0)LN −1 i=1 {a(Eii − Ei+1,i+1) : a ∈ A}(cid:1) (c) For a commutative A: slN (A) = {x ∈ glN (A) : tr(x) = 0} = slN (F ) ⊗F A. (d) The centre of slN (A) is Z(slN (A)) = {zEN : z ∈ Z(A) ∩ [A, A]} where Z(A) = {z ∈ A : za = az for all a ∈ A} is the centre of A. (e) For a, b ∈ A and i 6= j, is an sl2-triple if and only if a is invertible and b = a−1. (aEij , Eii − Ejj , bEji) 46 ERHARD NEHER The exercise shows that the structure of a general slN (A) is quite similar to that of slN (F ). In particular, the decomposition (4.18) is a Q(Al)-grading where Al = {εi − εj : 1 ≤ i, j ≤ N }, l = N − 1. is the root system of type Al and slN (A)εi−εj = AEij for i 6= j. In fact, slN (A) is the prototype of an Al-graded Lie algebra (see [BerM]). At this level of generality we are far from the structure of a Lie torus. Most importantly, we are missing a compatible Λ-grading of slN (A). We will use gradings of A, defined as follows. Definition 4.20. Let A = Lλ∈Λ Aλ be a unital associative Λ-graded F -algebra. Then A is called an associative torus of type Λ if it satisfies (AT1)–(AT3) below. (AT1): if every non-zero Aλ contains an invertible element, (AT2): dim Aλ ≤ 1 for all λ ∈ Λ, and (AT3): spanZ(suppΛ A) = Λ. One calls A simply an associative torus if A is an associative torus of type Λ for some abelian group Λ. See 4.23 for a short discussion of associative tori. These definitions are justified by the following exercise, describing when slN (A) is a Lie torus of type (Al, Λ). Exercise 4.21. (a) The Lie algebra slN (A) has a Λ-grading compatible with the In this case, the compatible Q(Al)-grading (4.18) if and only if A is Λ-graded. Λ-grading of slN (A) is given by slN (A) = Lλ∈Λ slN (A)λ where slN (A)λ consists of matrices in slN (A), which have all their entries in Aλ. (b) With respect to the compatible gradings of (a), the Lie algebra slN (A) is a Lie torus of type (AN −1, Λ) if and only if A is an associative torus of type Λ. (c) The Lie torus slN (A) is invariant with respect to the bilinear form (··)sl given by (Pi,j xij Eij Pp,q ypqEpq)sl = Pi,j(xij yji)0 where a0 for a ∈ A denotes the A0-component of a. But we not only have an example of a Lie torus of type (Al, Λ), we actually have all centreless examples. Theorem 4.22. Let l ≥ 3. A Lie algebra L is a centreless Lie torus of type (Al, Λ) if and only if L is graded-isomorphic to sll+1(A) for A an associative torus of type Λ. In this case, L is an invariant Lie torus. Proof. This is a special case of the Coordinatization Theorem of Al-graded Lie algebras ([BerM, Recognition Theorem 0.7]): A centreless Lie algebra L is Al- graded (l = N −1) if and only if L is Q(Al)-graded-isomorphic to slN (A)/Z(slN (A)) for some associative F -algebra A. If L is a Lie torus, it follows as in the Exercise 4.21 above that A is an associative torus. But Z(slN (A)) = {0} for an associative torus ([NY, (3.3.2)] and Exercise 4.19). (cid:3) Besides [BerM], related results are proven in [BGK, Th. 2.65] (see Cor. 4.27 below), [GN, 2.11 and 3.4] and [Y2, Prop. 2.13]. Review 4.23 (Associative tori versus twisted group algebras). In view of Th. 4.22 it is of interest to know more about associative tori. First of all, the identity 1A of an associative torus A satisfies 1A ∈ A0. Hence a−1 ∈ A−λ for every invertible LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 47 a ∈ Aλ. Moreover, since the product of two invertible elements in an associative algebra is again invertible, it follows that suppΛ A is a subgroup of Λ, whence (AT3) is equivalent to (AT3)′: suppΛ A = Λ. Next, choose a family (uλ : λ ∈ Λ) of invertible elements uλ ∈ Aλ. This is then in particular an F -basis of A so that the algebra structure of A is completely determined by the equations (4.19) uλuµ = c(λ, µ)uλ+µ for λ, µ ∈ Λ and suitable non-zero scalars c(λ, µ) ∈ F . It is not necessary that c(λ, µ) = 1, see for example Ex. 4.26. Rather, given an F -vector space with basis (uλ : λ ∈ Λ) one can define a multiplication on A by (4.19), and this multiplication is associative if and only if (4.20) c(λ, µ) c(λ + µ, ν) = c(µ, ν) c(λ, µ + ν) In this case, the algebra is an associative torus of type Λ. It is clear from the construction that, conversely, any associative torus is obtained in this way from a family (c(λ, µ))λ,µ of non-zero scalars. The algebras constructed in this way are called twisted group algebras. The reader with some knowledge in group cohomology will recognize that the families (c(λ, µ)) satisfying (4.20) are precisely the 2-cocycles of Λ with values in F \{0}. One can show that two families define graded-isomorphic tori if and only if their cohomology classes coincide. Example 4.24 (Group algebra). Although, as we have pointed out, the c(λ, ν) need not equal 1 in general, the family for which all c(λ, µ) = 1 satisfies (4.20) and so yields an example of a Λ-torus, called the group algebra of Λ and denoted F [Λ]. In particular, this implies that associative tori exist for all Λ, and hence Lie tori exist for all types (Al, Λ), l ≥ 2. The Lie tori that arise as cores of an EALA have type (S, Λ) where Λ is a free abelian group of finite rank. This condition on Λ follows from the axiom (EA6) or, equivalently from the axiom (EARS7). We therefore discuss this special case now. Definition 4.25. Let q = (qij ) be an n × n matrix such that the entries qij ∈ F satisfy qii = 1 = qijqji for all 1 ≤ i, j ≤ n. The quantum torus associated to q is the associative algebra Fq presented by the generators ti, t−1 , 1 ≤ i ≤ n subject to the relations i tit−1 i = t−1 i ti, and titj = qij tjti for all 1 ≤ i, j ≤ n. For example, if all qij = 1, then Fq = F [t±1 n ] is the Laurent polynomial ring in n variables. Thus, a general Fq is a non-commutative version of F [t±1 1 , . . . , t±1 n ], the coordinate ring of the n-dimensional algebraic torus (F \ {0})n, which explains the name "quantum torus". 1 , . . . , t±1 Exercise 4.26. Let Fq be a quantum torus. Show: (a) Fq =Lλ∈Zn F tλ for tλ = tλ1 1 · · · tλn n . (b) The tλ satisfy the multiplication rule tλtµ = c(λ, µ)tλ+µ with (c) Fq is an associative torus of type Zn. c(λ, µ) =Q1≤j<i≤n qλiµj ij . 48 ERHARD NEHER (d) Every associative torus of type Zn is graded-isomorphic to some quantum torus Fq. (e) The centre Z(Fq) = {z ∈ Fq : [z, Fq] = 0} of the associative algebra Fq is a graded subspace of Fq, namely Z(Fq) = Lγ∈Γ F tγ, where Γ is a subgroup of Zn given by Γ = {γ ∈ Zn : c(γ, µ) = c(µ, γ) for all µ ∈ Zn} Moreover, the following are equivalent: = {γ ∈ Zn :Qn j=1 qγj ij = 1 for 1 ≤ i ≤ n}. (i) All qij are roots of unity. (ii) [Zn : Γ] < ∞. (iii) Fq is finitely generated as a module over its centre Z(Fq). Combining this exercise with Th. 4.22 we obtain the classification of the cores of EALAs of type A: Corollary 4.27 ([BGK, Th. 2.65]). The Lie algebra slN (Fq) for a quantum torus Fq is a centreless Lie torus of type (AN −1, Zn). Conversely, any centreless Lie torus of type (AN −1, Zn) with N ≥ 4 is graded-isomorphic to some slN (Fq). The attentive reader will have noticed that we didn't say anything about Lie tori of type (A1, Λ) and (A2, Λ) in Th. 4.22 and Cor. 4.27. Of course, sl3(A) of an associative Λ-torus will be centreless Lie tori of type (A2, Λ). The limitation N ≥ 4 in Th. 4.22 is justified, since for N = 3 there are more examples: One needs coordinate algebras A which are no longer associative but only alternative. More- over, one needs to replace the matrix algebra sl3(A) by something more general, a Tits-Kantor-Koecher algebra or an abstractly defined Lie algebra, see [BGKN] for details. Analogous remarks apply for the A1-case, in which the coordinates come from certain Jordan algebras (called Jordan tori) and in which sl2 has to be replaced by a Jordan algebra. One now has classification theorems for Lie tori of all types. The references up to 2007 for each type are listed at the beginning of §7 in [AF]. An additional recent reference is [NT] for S = B2. 4.5. Some more easy examples of Lie tori. We describe some more easy exam- ples, where easy means that they do not require some knowledge of non-associative algebras, like Jordan algebras, alternative or structurable algebras. Example 4.28 (Λ = {0}). Let g be a finite-dimensional split simple Lie algebra with splitting Cartan subalgebra h. Then g has a root space decomposition g = system. Since g is simple, S is also irreducible. Using standard properties of finite- Lξ∈S gξ where g0 = h and S is the root system of (g, h), a finite reduced root dimensional split simple Lie algebras, it is easy to check that g =Lξ∈S gξ is a Lie Conversely, if L is a Lie torus of type (S, {0}), then L is a finite-dimensional split torus of type (S, {0}). simple Lie algebra. Indeed, L = g in the notation of Prop. 4.8. Note that this fits nicely the picture of EALAs of nullity 0, which we have characterized in Prop. 3.24 as finite-dimensional split simple Lie algebras. Example 4.29. As in the previous Example 4.28 let g be a finite-dimensional split simple Lie algebra with splitting Cartan subalgebra h and root system S. We would LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 49 like to consider a Lie algebra of the form g ⊗ A where A is an associative algebra. For g of type A we could take non-commutative "coordinates" A to get a Lie torus, see §4.4. However, for g not of type A the algebra A must be commutative in order to get a Lie algebra. Therefore, we let A = Lλ∈Λ Aλ be a commutative associative torus of type Λ and consider g ⊗ A, which becomes a Lie algebra (over F ) by [u1 ⊗ a1, u2 ⊗ a2] = [u1, u2] ⊗ a1a2. It is a centreless Lie torus of type (S, Λ) with respect to the homogeneous subspaces (g ⊗ A)λ ξ = gξ ⊗ Aλ. Note that the support of the Q(S) ⊕ Λ-graded Lie algebra g ⊗ A is the set suppQ(S)⊕Zn g ⊗ A = S × Λ, and that g ⊗ A is an invariant Lie torus with respect to the bilinear form (x ⊗ aλ y ⊗ bµ) = κ(x, y) (aλbµ)0 where κ is the Killing form of g and c0 for c ∈ A is the 0-component of c. This example works for any type of S. But it yields all examples only for special types of S. Theorem 4.30. Any centreless Lie torus of type (S, Λ) for S of type Dl, l ≥ 4 or El, l = 6, 7, 8, is graded-isomorphic to an example as in 4.29 for g of the corresponding type and A a commutative associative torus of type Λ. Proof. The proof is analogous to the proof of Th. 4.22: One applies the Coordina- tization Theorem of [BerM] to get that L has the form g ⊗ A for some commutative associative F -algebra. One then has to discuss when such a Lie algebra is a Lie torus. This is the case exactly when A is a torus. (cid:3) Example 4.31 (Untwisted multiloop algebras). For EALAs it is of interest to describe the centreless Lie tori of type (S, Λ) with Λ a free abelian group of finite rank, say of rank n. Hence Λ ∼= Zn. It is immediate that g ⊗ A is a Lie torus of type (S, Zn) if and only if A is a commutative quantum torus, i.e., a Laurent polynomial ring in several, say n variables. In other words, these are the untwisted multiloop algebra of (1.11), L(g) = g ⊗F F [t±1 1 , . . . , t±1 n ] Hence by Th. 4.11 the universal central extension of L(g), the toroidal Lie algebras of §1.2 are also Lie tori. Finally, Th. 4.30 has the following corollary. Corollary 4.32 ([BGK]). Any centreless Lie torus of type (S, Zn) with S = Dl, l ≥ 4 or S = El, l = 6, 7, 8, is graded-isomorphic to an untwisted multiloop algebra L(g) as in Example 4.31. Perhaps the reader now expects that the next example will be the general multi- loop algebras L(g, σ) defined in (1.13). However, an arbitrary multiloop algebra is in general not a Lie torus, see [ABFP2, Th. 3.3.1] and [Na, Th. 5.1.4] for a charac- terization of centreless Lie tori which are multiloop algebras. But this phenomenon does not occur in nullity 1. Exercise 4.33. Verify that the loop algebra L(g, σ) of (1.3) is an invariant Lie torus of type (S, Z) where S is the root system of Table 3.7. 50 ERHARD NEHER 5. The construction of all EALAs Recall Th. 4.14: If (E, H) is an EALA, its core Ec and its centreless core Ecc are Lie tori, the latter being an invariant Lie torus. Moreover, if (S, Λ) is the type of Ec and Ecc then Λ is a free abelian group of finite rank. Thus: core Ec (Lie torus) EALA (E, H) centreless core Ecc (invariant Lie torus) In this chapter we will reverse the process, starting from an invariant Lie torus we will construct an EALA. To motivate the construction it is useful to look again at the construction in Example 1.1 and 2.4 of an affine Kac-Moody Lie algebra. It can be summarized as follows: We start with a twisted loop algebra L = L(g, σ), which as we now know is an invariant Lie torus (Ex. 4.33). We then take a central extension L, which in this example is the universal central extension and hence by Th. 4.11 again a Lie torus (of course, one can also verify this directly in this example). Finally, we add some (not all) derivations to L to get an affine Kac-Moody Lie algebra, and all affine Kac-Moody Lie algebras are obtained in this way (Kac's Realization Theorem 1.3). To summarize, using the EALA terminology: central extension of L derivations (another Lie torus) EALA (E, H) add invariant Lie torus L To do something like this in general, one faces the following two problems. (A) An invariant Lie torus has in general many central extension. For example, the untwisted multiloop algebra L = g ⊗ F [t±1 n ] is an invariant Lie torus by Ex. 4.31. If n ≥ 2, its universal central extension has an infinite-dimensional centre, a result we already mentioned in §1.2, see in particular Th. 1.5. Hence, there are many possible central extensions. Should we only consider the universal central extension? 1 , . . . , t±1 (B) Which derivations should we add? Already in the affine case did we not add all derivations, as follows for example from Ex. 1.7! It turns out that the two problems are closely related, and we will solve both at the same time. Rather than taking a 2-step approach, we will take one big step, by taking what one may call an affine extension (after all, the result will be an extended affine Lie algebra). In fact, affine extensions are a special case of so-called double extension, see for example [Bor]. central extension of L / EALA (E, H) (5.1) invariant Lie torus L affine extension 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 O O O O / O O 3 3 3 s 3 s 3 s 3 s 3 s 3 s 3 s 3 s 3 s 3 s 3 s 3 s 3 s 3 s LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 51 The key idea is based on the construction of a 2-cocycle in Ex. 1.19: Any subspace D of skew-symmetric derivations will give rise to a 2-cocycle and hence to a central extension. But not only do we get examples of central extensions. By Th. 1.25 and Ex. 1.27, up to isomorphism all central coverings of L are of the form E(L, D, ψD) for some graded subspace D of SDer(L) with D ∩ IDer(L) = {0}. Observe that we can indeed apply this theorem: The invariant Lie torus L (i) is perfect by Ex. 4.10 and is finitely generated as Lie algebra by Th. 4.13 (recall that Λ is free of finite rank, where (S, Λ) is the type of L), (ii) has finite homogeneous dimension, even bounded homogeneous dimension also by Th. 4.13, and (iii) has an invariant nondegenerate Λ-graded symmetric bilinear form, by def- inition of an invariant Lie torus. But we need more than just a central covering. For example, the axiom (EA2) requires that we construct an ad-diagonalizable subalgebra H for which the sub- spaces Lλ ξ , ξ 6= 0, are root spaces, as can be seen from Th. 4.14. By Lemma 4.4 we can realize the subspaces Lξ as root spaces of some natural subalgebra h ⊂ L. But we do not have a result, which describes the subspaces Lλ in a similar fashion, i.e., as root spaces of some toral subalgebra. There is in fact no natural choice of a subalgebra to do so. Rather, we will distinguish these subspaces "externally", i.e., via an action of some non-inner derivation algebra. The required formalism to do this, is described in the next section. This has nothing to do with Lie algebras. Rather, it is a topic in the theory of graded vector spaces, and we will therefore describe it in this setting. 5.1. Degree maps. In this section, V is vector space over a field F , which could be of arbitrary characteristic until Prop. 5.3(b). Also, Λ denotes an arbitrary abelian group. We recall that a Λ-grading of V is simply a direct vector space decomposition of V by a family (V λ : λ ∈ Λ) of subspaces V λ ⊂ V . Our goal in this section is to present a method describing the homogeneous subspaces V λ of a given Λ-grading of V as the joint eigenspaces of a subspace of diagonalizable endomorphisms. To motivate the construction, let us first look at the converse, namely inducing a grading of V via the action of endomorphisms. We will say that a subspace T ⊂ EndF (V ) is a subspace of simultaneously diagonalizable endomorphisms, if (5.2) V =Lλ∈T ∗ V λ for V λ = {v ∈ V : t(v) = λ(t)v for all t ∈ T }. In this case, T obviously consists of pairwise commuting diagonalizable endomor- phisms. Conversely, it is well-known that a finite-dimensional subspace of pairwise commuting diagonalizable endomorphisms is a subspace of simultaneously diagonal- izable endomorphisms (this is no longer true if T is infinite-dimensional). Observe that the decomposition (5.2) is a grading of V by the group spanZ(suppT ∗ V ) where suppT ∗ V = {λ ∈ T ∗ : V λ 6= 0}. Our goal is to realize a given grading of V in this way. To do so, we will use the F -vector space D(Λ) = HomZ(Λ, F ) consisting of all maps θ : Λ → F which are Z-linear: θ(λ1 + λ2) = θ(λ1) + θ(λ2) for all λi ∈ Λ. This is an F -vector space by defining for θ, θi ∈ D(Λ) and s ∈ F the 52 ERHARD NEHER sum θ1 + θ2 and the scalar multiplication sθ by (θ1 + θ2)(λ) = θ1(λ) + θ2(λ) and (sθ)(λ) = s(θ(λ)). Exercise 5.1. (a) Show D(Λ) ∼= HomF (Λ ⊗Z F, F ) = (Λ ⊗F F )∗. Thus D(Λ) is naturally a dual vector space. (b) If Λ is free of rank n, say with Z-basis ε1, . . . , εn, then D(Λ) = F ∂1 ⊕· · ·⊕F ∂n where ∂i ∈ D(Λ) is defined by ∂i(Pj mjεj) = mi. In particular, dimF D(Λ) = n. We now suppose that V =Lλ∈Λ V λ is a Λ-grading of the vector space V . Any θ ∈ D(Λ) defines an endomorphism ∂θ ∈ EndF (V ) by ∂θ(vλ) = θ(λ) vλ for vλ ∈ V λ. We put D(V ) = {∂θ : θ ∈ D(Λ)} and call the elements of D(V ) degree maps. If A = Lλ∈Λ Aλ is a Λ-grading of an algebra A, the maps ∂θ are derivations and D(A) is called the space of degree derivations. The map ∂ : D(Λ) → D(V ) is clearly F -linear and surjective by definition. Its kernel is {θ ∈ D(Λ) : θ(suppΛ V ) = 0}. To make ∂ an isomorphism we will (5.3) from now on assume spanZ(suppΛ V ) = Λ. As we have pointed out at previous occasions, this is not a serious assumptions since one can always replace Λ by spanZ(suppΛ V ) without changing the given grading. Since now ∂ is an isomorphism, we can define a linear form evλ ∈ D(V )∗ for every λ ∈ Λ: The F -linear map evλ(∂θ) = θ(λ) ev : Λ → D(V )∗, λ 7→ evλ is called the evaluation map. By construction, (5.4) since for d = ∂θ and v ∈ V λ we have ∂θ(vλ) = θ(λ)vλ = evλ(∂θ)vλ. V λ ⊂ {v ∈ V : d(v) = evλ(d)v for all d ∈ D(V )} Definition 5.2. In the setting of above, i.e., V =Lλ∈Λ V λ is Λ-graded and (5.3) holds, we will say that a subspace T ⊂ D(V ) induces the Λ-grading of V if V λ = {v ∈ V : t(v) = evλ(t)v for all t ∈ T } holds for all λ ∈ Λ. that (5.3) holds. Proposition 5.3. Let V = Lλ∈Λ V λ be a Λ-grading of the vector space V such (a) A subspace T ⊂ D(V ) induces the Λ-grading of V if the restricted evaluation map is injective. evT : Λ → T ∗, evT (λ) = evλ T (b) Suppose F has characteristic 0 and Λ is torsion-free, i.e., nλ = 0 for some n ∈ Z implies λ = 0. Then Λ embeds into the F -vector space U = Λ ⊗Z F and for every subspace S ⊂ D(Λ) separating the points of Λ in U the corresponding subspace T = ∂(S) ⊂ D(V ) induces the Λ-grading of V . In particular, this holds for D(V ) itself. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 53 5.2. The centroid of Lie algebras, in particular of Lie tori. After the inter- mezzo on how to induce gradings of vector spaces in the previous section 5.1 we now come back to Lie algebras, but not immediately to Lie tori and EALAs. Of course, the topic of this section is motivated by the over-all goal of this chapter: The construction of EALAs from Lie tori using certain subspaces of derivations. The derivations, which in 5.4 will be used in the general construction, are products of degree maps, studied in 5.1, and so-called centroidal transformations, to which this section is devoted. The basic idea of the centroid of a Lie algebra (or of any algebra for that matter) is that it identifies the largest ring over which the given algebra can be considered as an algebra. For example, if one studies the real Lie algebra L which is sln(C) considered as a real Lie algebra by restricting the scalars to R, the centroid will be ∼= C and will thus indicate that L can also be considered as a complex Lie algebra. In general, the centroid will not be a field but only a (commutative) ring. Hence, considering a Lie algebra as algebra over its centroid, necessitates that in the fol- lowing definition and the Lemma 5.6 after it we will deviate from our standard assumption and consider Lie algebras over rings. The definition of a Lie algebra L defined over a ring, say k, is not surprising: L is a k-module with a k-bilinear map [., .] : L × L → L which is alternating, i.e., [l, l] = 0 for all l ∈ L, and satisfies the Jacobi identity. Definition 5.4 ([J, Ch. X]). The centroid Centk(L) of a Lie algebra L defined over a ring k is defined as Centk(L) = {χ ∈ Endk(L) : χ([l1, l2]) = [l1, χ(l2)] for all l1, l2 ∈ L}. Of course, χ ∈ Centk(L) ⇔ χ([l1, l2]) = [χ(l1), l2] for all l1, l2 ∈ L. It is important to indicate k in the notation Centk(L) since the centroid depends on the base ring k. We have k IdL ⊂ Centk(L) for every L. One calls L central if the map k → CentF (L), s 7→ s IdL, is an isomorphism, and one says that L is central-simple if L is just that: central and simple. Let L =Lλ∈Λ Lλ be a Lie algebra graded by an abelian group Λ. We can then also define the Λ-graded centroid as grCentk(L) = grEndk(L) ∩ Centk(L) =Lλ∈Λ Centk(L)λ where Centk(L)λ consists of the centroidal transformations which have degree λ: χ(Lµ) ⊂ Lλ+µ for all µ ∈ Λ. Example 5.5. As an immediate example we calculate the centroid of the Lie algebra L = sl2(C), considered as real Lie algebra by restricting the scalars to R. We let (e, h, f ) be an sl2-triple in sl2(C). Then the relations [h, ce] = 2ce and [h, cf ] = −2cf for c ∈ C show that χ(ce) and χ(cf ) are uniquely determined by χ(h). For example, 2χ(ce) = χ([h, ce]) = [χ(h), ce]. Moreover, χ(h) ∈ Ch because [χ(h), ch] = χ([h, ch]) = 0. Hence dimR CentR(L) ≤ 2. On the other side, C IdL ⊂ CentR(L) is clear, whence C IdL = CentR(L). We leave it to the reader to show CentR(L) = C IdL for L = sln(C) considered as real Lie algebra, without using any of the results mentioned below! The following lemma gives a mathematical meaning to the claims made before the Def. 5.4, and lists the most important properties of the centroid of Lie algebras which are not necessarily Lie tori. 54 ERHARD NEHER Lemma 5.6 (Folklore). Let L be a Lie algebra defined over a ring k. (a) The centroid of L is always a unital associative subalgebra of the endo- morphism algebra Endk(L) of L. Hence Centk(L) is a k-algebra and L becomes a Centk(L)-module by defining the action of Centk(L) on L by χ · l = χ(l) for χ ∈ Centk(L) and l ∈ L. (b) If the centroid of L is commutative, then with respect to the action of Centk(L) on L defined in (a), L is a Lie algebra over the ring Centk(L). Moreover, L is central as a Lie algebra over its centroid. (c) If L is perfect, its centroid is commutative and does not depend on the base ring k: Centk(L) = CentZ(ZL) where ZL is the Lie algebra L with scalars restricted to Z. (d) If L is simple, its centroid is a field and L as a Lie algebra over the field CentF (L) is central-simple. In particular: (i) a finite-dimensional simple Lie algebra over an algebraically closed field F is central-simple, and (ii) the centroid of a simple real Lie algebra L is either ∼= R Id, in which case L is central-simple, or is ∼= C Id, in which case L is a simple complex Lie algebra, considered as a real Lie algebra. (e) Suppose L is Λ-graded. Then grCentk(L) is a Λ-graded subalgebra of the full centroid Centk(L). Moreover, grCentk(L) = Centk(L) if L is finitely generated as an ideal, i.e., there exist l1, . . . , ln ∈ L such that the ideal generated by l1, . . . , ln is all of L. (f) If χ ∈ Centk(L) and d ∈ Derk(L), then χ ◦ d ∈ Derk(L). With respect to this operation, Derk(L) is a Centk(L)-module and IDer(L) is a submodule of the Centk(L)-module Derk(L). The proof of this lemma is a straightforward exercise, which the reader will be asked to do now. The exercise also lists some interesting additional facts on the centroid. Exercise 5.7. (a) For any χ ∈ Centk(L) the kernel Ker χ and the image Im χ are ideals of L satisfying [Ker χ, Im χ] = 0. (b) Prove Lemma 5.6. For part (c) of the Lemma use (a) above. (c) If L is perfect, any χ ∈ Centk(L) is symmetric with respect to any invariant bilinear form on L. Here is the result, which describes the centroid of the Lie algebras of interest in this chapter. We will use the notion of an associative torus, introduced in 4.20 and further discussed in 4.23–4.26. Proposition 5.8 ([BN, Prop. 3.13]). Let L = Lξ∈S, λ∈Λ Lλ torus of type (S, Λ). (a) With respect to the (Q(S) ⊕ Λ)-grading of L we have ξ be a centreless Lie (5.5) In particular, χ(Lξ) ⊂ Lξ for any χ ∈ Centk(L) and ξ ∈ S. CentF (L) =Lλ∈Λ CentF (L)λ 0 = grCentF (L). (b) Moreover, with respect to the decomposition (5.5) the centroid CentF (L) is an associative commutative torus of type Γ, where Γ = suppΛ CentF (L) is a subgroup of Λ, hence a twisted group algebra over Γ. LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 55 (c) In particular, if Λ is free abelian of finite rank n, the centroid CentF (L) is graded-isomorphic to F [Γ], the group algebra of Γ as defined in 4.24, and is thus isomorphic to a Laurent polynomial ring in ν variables, 0 ≤ ν ≤ n. Moreover, L is a free module over its centroid. Proof. Parts (a) and (b) of this proposition are proven in [BN, Prop. 3.13]. Part (c) is ([Ne3, Th. 7]). The first part of (c) follows from (b): A twisted group algebra over a free group is a group algebra. The second part is a special case of a general fact: Any graded module over an associative torus is free. (cid:3) Example 5.9. Let L = slN (A) for A an associative F -algebra, see 4.4, and let Z(A) = {z ∈ A : [z, A] = 0} be the centre of the associative algebra A. Any z ∈ Z(A) induces a centroidal transformation χz defined by mapping x = (xij ) ∈ slN (A) to χz(x) = (zxij ). It is easily seen ([Ne5, 7.9]) that Z(A) → CentF (slN (A)), z 7→ χz is an isomorphism of F -algebras (the only non-obvious part is surjectivity). Let us now specialize to the case of a Lie torus slN (A) of type (AN −1, Zn). Thus, by Cor. 4.27, A = Fq is a quantum torus. A description of the centre Z(Fq) is given in Ex. 4.26(e) (see [BGK, Prop. 2.44] for a proof): Z(Fq) =Lγ∈Γ F tγ where Γ is the subgroup j=1 qγj ij = 1 for 1 ≤ i ≤ n}. Γ = {γ ∈ Zn :Qn of Zn. The centre of Fq is therefore isomorphic to a Laurent polynomial ring in, say, ν variables, as claimed in Prop. 5.8(c). To see that the inequalities 0 ≤ ν ≤ n stated there are sharp, we consider the quantum torus associated to the matrix Specializing the description of Γ above we get q =(cid:20) 1 q−1 q 1(cid:21) . Γ =({0}, mZ ⊕ mZ, q not a root of unity, q an mth root of unity. Hence ν = 0 in the first case and ν = 2 = n in the second case. However, the following result says that this is the only case in which the cen- troidal grading group Γ has smaller rank than Λ. Theorem 5.10 ([Ne3, Th. 7]). Let L be a centreless Lie torus of type (S, Zn) with S not of type A. Then [Zn : Γ] < ∞ and L is a free CentF (L)-module of finite rank. This result, together with the Realization Theorem of [ABFP1] implies that an invariant Lie torus of type (S, Zn), S 6= Al, is graded-isomorphic to a multiloop algebra as defined in (1.13). A characterization of which multiloop algebras are Lie tori is the main result of [ABFP2]. A more general approach to realizing Lie tori as multiloop algebras is developed in [Na]. It is easy to verify Th. 5.10 in case L is a Lie torus of type (S, Zn) and S of type n ]. The D or E. As we have seen in Th. 4.30, in this case L = g ⊗ F [t±1 1 , . . . , t±1 centroids of these types of Lie algebras are described in the next example. Example 5.11. Let g be a finite-dimensional central simple Lie algebra. For example, by [BN, Remark 3.6] any finite-dimensional split simple Lie algebra is 56 ERHARD NEHER central and thus central-simple. (Over algebraically closed fields, this also follows from Lemma 5.6(d).) Also, let A be an associative commutative F -algebra. A straightforward verification shows that for s ∈ F and a ∈ A the map χs,a, defined by u ⊗ b 7→ su ⊗ ab, is a centroidal transformation of the Lie algebra g ⊗ A. It follows from [ABP3, Lemma 2.3(a)] or [Az2, Lemma 1.2] or [BN, Cor. 2.23] that these are all the maps in the centroid of g ⊗ A: F Idg ⊗A ∼= CentF (g ⊗ A), via s ⊗ a 7→ χs,a. Although this will not be needed in the following, we mention that the centroid of an EALA is known too. Proposition 5.12. Let E be an EALA, let K = Ec be its core and put D = E/K. Then K is a central Lie algebra, and CentF (E) = F IdE ⊕ V(K), V(K) = {χ ∈ CentF (E) : χ(K) = 0}. As a vector space, the ideal V(K) of CentF (E) is canonically isomorphic to the D-module homomorphisms D → Z(K): V(K) ∼= HomD(D, Z(K)). This is proven in [BN, Cor. 4.13]. Observe that the reference to [Ne4, Th.6] in the proof of [BN] can now be replaced by the combination of Th. 4.13(c) and Ex. 1.27. 5.3. Centroidal derivations of Lie tori. In this section L is a centreless Lie torus of type (S, Λ). Regarding L as a Λ-graded Lie algebra, the results of section 5.1 apply and provide us with the subspace D = D(L) = {∂θ : θ ∈ D(Λ)} of degree derivations of L. Moreover, we can apply Lemma 5.6(f) and get that χ ◦ ∂θ ≡ χ∂θ is a derivation for any χ ∈ CentF (L). We call the elements of CDerF (L) = CentF (L) D centroidal derivations. (A notion of centroidal derivations for arbitrary Λ-graded Lie algebra is developed in [Ne5, 4.9].) Recall from Prop. 5.8 that CentF (L) = Lγ∈Γ CentF (L)γ is a commutative associative torus of type Γ, where Γ is a sub- group of Λ. Since D consist of degree 0 endomorphisms, CDer(L) is Γ-graded, (5.6) CDerF (L) =Lγ∈Γ CDerF (L)γ CDerF (L)γ = CentF (L)γ D = CentF (L) ∩ EndF (L)γ. for It is then easily seen that CDerF (L) is a Γ-graded subalgebra of DerF (L). For χγ ∈ CentF (L)γ, χδ ∈ CentF (L)δ and θ, ψ ∈ D(Λ) the Lie algebra product of CDerF (L) is given by the formula (5.7) Thus, CDerF (L) is a generalized Witt algebra, see for example [NY, 1.9]. [χγ∂θ, χδ∂ψ] = χγχδ(cid:0)θ(δ) ∂ψ − ψ(γ) ∂θ(cid:1). Suppose now that L is an invariant Lie torus, say with respect to the invariant bilinear from (··). We can then consider the skew centroidal derivations SCDerF (L) = SDerF (L) ∩ CDerF (L), LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 57 defined as the centroidal derivations which are skew-symmetric with respect to (··). This is a Γ-graded subalgebra of CDerF (L) whose homogenous components are given by (5.8) SCDerF (L)γ = {χγ∂θ : χγ ∈ CentF (L)γ, θ(γ) = 0}. In particular, SCDerF (L)0 = D is a toral subalgebra of SCDerF (L) since [∂θ, χδ∂ψ] = θ(δ)∂ψ by (5.7). It is also of interest to point out that [SCDerF (L)γ, SCDerF (L)−γ] = 0, which implies that SCDerF (L) is a semidirect product, SCDerF (L) = D ⋉(cid:0)Lγ6=0 SCDerF (L)γ(cid:1) of the toral subalgebra D and the ideal spanned by the homogeneous subspaces of non-zero degree. For the construction of EALAs, the following theorem is funda- mental. Theorem 5.13 ([Ne3, Th. 9]). Let L be an invariant Lie torus of type (S, Λ) with Λ free of finite rank. Then DerF (L) is a semi-direct product, (5.9) DerF (L) = IDerF (L) ⋊ CDerF (L), hence SDerF (L) = IDerF (L) ⋊ SCDerF (L), where IDerF (L) denotes the ideal of all inner derivations. Some remarks on the proof of this theorem follow. By Prop. 5.8 the centroid of L is a Laurent polynomial ring. Let K be its field of fractions, a field of rational functions. As a CentF (L)-module, L is torsion-free and hence L embeds into the Lie K-algebra L = L ⊗CentF (L) K, the so-called central closure of L. If the CentF (L)-module L is finitely generated, its central closure is a finite-dimensional central-simple Lie algebra. Hence, in this case DerK( L) = IDer( L), from which the theorem easily follows. If however L is not finitely generated as a CentF (L)-module, then we know from Th. 5.10 that L is a Lie torus of type A. More precisely, as a consequence of the results in [Y1] and [NY] for type A1, [BGKN] for type A2 and Cor. 4.27 for type Al, l ≥ 3, such a Lie torus is graded-isomorphic to sln(Fq). But in this case the result follows from [BGK, 2.17, 2.53], [BGKN, Th. 1.40] and [NY, Th. 4.11]. We will discuss the special case L = g ⊗ F [t±1 n ] in Ex. 5.14. To avoid any confusion, we note that the splitting (5.9) is not the one proven in [Be, Th. 3.12] for arbitrary root-graded Lie algebras. 1 , . . . , t±n The importance of the theorem stems from Th. 1.25: It identifies a natural complement of IDer(L) in SDerF (L). Hence, up to graded-isomorphism, any graded covering of L has the form E(L, Dgr∗, ψD) for a graded subspace D ⊂ SCDerF (L). Moreover, since D ⊂ SCDerF (L), we can require that D0 ⊂ D be not too small and use it to distinguish the homogeneous spaces Lλ by applying Prop. 5.3. This will be our approach in section 5.4. But first some examples. Example 5.14. Let L = g ⊗ A where g is a split simple finite-dimensional Lie alge- bra with root system S and where A = F [t±1 n ] is a Laurent polynomial ring 1 , . . . , t±1 58 ERHARD NEHER in n variables. This is an invariant Lie torus of type (S, Zn), see the Examples 4.29 and 4.31. We have seen in (1.18) that DerF (g ⊗ A) = IDer(g ⊗ A) ⊕ (Idg ⊗ DerF (A)). The reader has (or should have) determined DerF (A) in Ex. 1.7: DerF (A) = A D where D = spanZ({∂i : 1 ≤ i ≤ n}) in the notation of the quoted exercise. But by Ex. 5.1, D = D(A) is also the space of degree derivations of A. Since the Λ-grading of L = g ⊗ A is concentrated in the factor A, it follows that Id ⊗D is the space of degree derivations of L, whence, by Example 5.11, we have CDerF (g ⊗ A) = Idg ⊗A D = Idg ⊗ DerF (A). Thus, for the invariant Lie torus g ⊗ A the decomposition (1.18) is the same as the decomposition (5.9)! We have seen in Ex. 4.29 that L is an invariant Lie torus with respect to the tensor product form (··) = κ⊗β where κ is the Killing form of g and where β is the bilinear form on A defined by β(tλ, tµ) = δλ,−µ. It is then easy to identify SCDerF (L) using (5.8). In particular, for n = 1 we see that SCDer(g ⊗ F [t±1]) = F d, where d is the degree derivation of (1.8). In particular, this together with Th. 1.25 gives a new proof of the theorem, mentioned in 1.1, that the Lie algebra L(g, σ) of (1.7) is the universal central extension of the twisted loop algebra L(g, σ). 5.4. The general construction. Finally, we can describe the ingredients (L, D, τ ) of the general construction: ξ is an invariant Lie torus of type (S, Λ) with Λ a free abelian group of finite rank; we put Γ = suppΛ Cent(L), see Prop. 5.8. • L = Lξ∈S,λ∈Λ Lλ • D =Lγ∈Γ Dγ ⊂ SCDerF (L) is a graded subalgebra such that the evalua- tion map (5.10) evD0 : Λ → D0 ∗, λ → evλ D0 is injective. • τ : D × D → Dgr∗ is an affine cocycle, i.e., τ is a bilinear map satisfying for all d, di ∈ D (5.11) (5.12) τ (d, d) = 0 and P(cid:9) d1 · τ (d2, d3) =P(cid:9) τ ([d1, d2], d3), and τ (d1, d2)(d3) = τ (d2, d3)(d1) τ (D0, D) = 0, Recall from Prop. 5.3 that the condition (5.10) implies that D0 induces the Λ-grading of L, i.e., (5.13) For example, (5.10) holds for D = D = SCDerF (L)0 or D any graded subalgebra Lλ = {l ∈ L : d0(l) = evλ(d0)l, for all d0 ∈ D0}. with D0 = D. In (5.11), the symbolP(cid:9) denotes the cyclic sum: P(cid:9) d1 ·τ (d2, d3) = d1 · τ (d2, d3) + d2 · τ (d3, d1) + d3 · τ (d1, d2) and analogously for P(cid:9) τ ([d1, d2], d3). Moreover, d · c for c ∈ Dgr∗ is the contragradient action of D on the graded dual space Dgr∗. The condition (5.11) says that τ is an abelian 2-cocycle, meaning that Dgr∗ ⊕ D is a Lie algebra with respect to the product formula (5.14) for ci ∈ Dgr∗ and di ∈ D. Thus, [c1 ⊕ d1, c2 ⊕ d2] =(cid:0)d1 · c2 − d2 · c1 + τ (d1, d2)(cid:1) ⊕ [d1, d2] 0 / Dgr∗ inc / Dgr∗ ⊕ D prD / / D / 0 / / / LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 59 is an abelian extension: Dgr∗ is an abelian ideal, not necessarily contained in the centre. The conditions in (5.12) will allow us to define a toral subalgebra H and an invariant bilinear form (··) below. We note that an affine cocycle is necessarily graded of degree 0: τ (Dγ , Dδ) ⊂ (Dgr∗)γ+δ for γ, δ ∈ Γ. There do exist non-trivial affine cocycles, see [BGK, Rem. 3.71] and [ERM]. To data (L, D, τ ) as above we associate a Lie algebra E = L ⊕ Dgr∗ ⊕ D with product (li ∈ L, ci ∈ Dgr∗ and di ∈ D) (5.15) [l1 ⊕ c1 ⊕ d1, l2 ⊕ c2 ⊕ d2] =(cid:0)[l1, l2]L + d1(l2) − d2(l1)(cid:1) (cid:0)ψD(l1, l2) + d1 · c2 − d2 · c1 + τ (d1, d2)(cid:1) ⊕ [d1, d2]. Here [., .]L is the Lie algebra product of L, di(lj) is the natural action of D on L, and ψD is the central 2-cocycle of (1.24). It is immediate from the product formula that (i) L ⊕ Dgr∗ is an ideal of E, and the canonical projection L ⊕ Dgr∗ → L is a central extension. (ii) The Lie algebra Dgr∗ ⊕ D of (5.14) is a subalgebra of E. The Lie algebra E has a a subalgebra H = h ⊕ D0 ∗ ⊕ D0 ξ : ξ ∈ S×, λ ∈ Λ} = spanF {h0 where h = spanF {hλ ξ : 0 6= ξ ∈ Sind}. We embed S into the dual space h∗, using the evaluation map of (5.10), and extend ξ ∈ S ⊂ h∗ to a linear form of H by ξ(D0 ∗ ⊕ D0) = 0. We embed Λ ⊂ D0 ∗, using the evaluation map of Prop. 5.3, and then extend λ ∈ Λ ⊂ D0 ∗ to a linear form of H by putting λ(h ⊕ D0 ∗) = 0. Then H is a toral subalgebra of E with root spaces Eξ⊕λ =(Lλ ξ , 0 ⊕ (D−λ)∗ ⊕ Dλ, Lλ ξ 6= 0, ξ = 0. Observe H = E0 since h = L0 defined by 0 by Ex. 4.10. The symmetric bilinear form (··) on E, (cid:0)l1 ⊕ c1 ⊕ d1 l2 ⊕ c2 ⊕ d2(cid:1) = (l1 l2)L + c1(d2) + c2(d1) where (··)L is the given bilinear form of the invariant Lie torus L, is nondegenerate and invariant. With respect to this bilinear form the set of roots of (E, H) is R = R0 ∪ Ran, where R0 = {λ ∈ Λ ⊂ H ∗ : Lλ Ran = {ξ ⊕ λ : ξ 6= 0 and Lλ 0 6= 0} and ξ 6= 0}. We have now indicated that the axioms (EA1) and (EA2) of an extended affine Lie algebra holds for the pair (E, H). The verification of the remaining axioms can be easily be done by the reader, or can be looked up in [Na, Prop. 5.2.4]. This then shows part (a) of the following theorem. 60 ERHARD NEHER Theorem 5.15 ([Ne4, Th. 6]). (a) The pair (E, H) constructed above is an extended affine Lie algebra, denoted E = E(L, D, τ ). Its core is L ⊕ Dgr ∗ and its centreless core is L. (b) Conversely, let (E, H) be an extended affine Lie algebra, and let L = Ec/Z(Ec) be its centreless core, which by Cor. 4.16 is an invariant Lie torus, say of type (S, Λ), with Λ free of finite rank. Then there exists a subalgebra D ⊂ SCDerF (L) and an abelian 2-cocycle τ sat- isfying the conditions (5.10)–(5.12) on (D, τ ) such that E ∼= E(L, D, τ ). We have defined discrete EALAs in 2.1 as a special class of EALAs over the base field F = C. They can now be characterized as follows. Corollary 5.16 ([Ne4, Th. 8]). Let F = C. (a) Let L be an invariant Lie torus of type (S, Λ) with Λ free of finite rank and let D ⊂ SCDerC(L) be a graded subalgebra such that the evaluation map ev : Λ → D0 ∗ is injective with discrete image. Then, for any affine 2-cocycle τ the extended affine Lie algebra E(L, D, τ ) is a discrete EALA. Conversely, any discrete EALA arises in this way. References [AABGP] Allison, B., Azam, S., Berman, S., Gao, Y. and Pianzola, A. Extended affine Lie algebras and their root systems, Mem. Amer. Math. Soc. 126 (1997), no. 603, x+122. [ABG1] Allison, B., Benkart, G., and Gao, Y., Central extensions of Lie algebras graded by finite root systems, Math. Ann. 316 (2000), 499–527. [ABG2] , Lie algebras graded by the root systems BCr, r ≥ 2, Mem. Amer. Math. Soc. 158 (2002), no. 751, x+158. [ABFP1] Allison, B., Berman, S., Faulkner, J., and Pianzola, A., Realization of graded-simple algebras as loop algebras, Forum Math. 20 (2008), 395–432. [ABFP2] , Muliloop realization of extended affine Lie algebras and Lie tori, Trans. Amer. Math. Soc. 361 (2009), 4807–4842. [ABGP] Allison, B., Berman, S., Gao, Y. and Pianzola, A., A characterization of affine Kac- [ABP1] Moody Lie algebras, Comm. Math. Phys. 185 (1997), 671–688. Allison, B., Berman, S., and Pianzola, A., Covering algebras. I. Extended affine Lie algebras, J. Algebra 250 (2002), no. 2, 485–516. [ABP2] , Covering algebras. II. Isomorphism of loop algebras, J. Reine Angew. Math. 571 (2004), 39–71. [ABP3] [AF] [AG] [Az1] [Az2] [Be] [BM] [BN] [BS] [BGK] , Iterated loop algebras, Pacific J. Math. 227 (2006), 1–41. Allison, B. and Faulkner, J., Isotopy for extended affine Lie algebras and Lie tori, arXiv:0709.1181 [math.RA]. B. N. Allison and Y. Gao, The root system and the core of an extended affine Lie algebra, Selecta Math. (N.S.) 7 (2001), 149–212. Azam, S., Generalized reductive Lie algebras: connections with extended affine Lie algebras and Lie tori, Canad. J. Math. 58 (2006), 225–248. Azam, S., Derivations of tensor product algebras, Comm. Algebra 36 (2008), 905–927. Benkart, G., Derivations and invariant forms of Lie algebras graded by finite root systems, Canad. J. Math. 50 (1998), 225–241. Benkart, G. and Moody, R., Derivations, central extensions, and affine Lie algebras, Algebras Groups Geom. 3 (1986), 456–492. Benkart, G. and Neher, E., The centroid of extended affine and root graded Lie algebras, J. Pure Appl. Algebra 205 (2006), 117–145. Benkart, G. and Smirnov, O., Lie algebras graded by the root system BC1, J. Lie Theory 13 (2003), 91–132. Berman, S., Gao, Y. Krylyuk, Y., Quantum tori and the structure of elliptic quasi- simple Lie algebras, J. Funct. Anal. 135 (1996), 339–389. [BGKN] Berman, S., Gao, Y., Krylyuk, Y., and Neher, E., The alternative torus and the struc- ture of elliptic quasi-simple Lie algebras of type A2, Trans. Amer. Math. Soc. 347 (1995), 4315–4363. [BerM] [Bor] [Bou1] [Bou2] [Bou3] [DFP] [D] [ERM] [GN] [G] [He] [HT] [Hu] [J] [Kac] [Kas] [L] [LN] [MP] LECTURES ON EXTENDED AFFINE LIE ALGEBRAS 61 Berman, S. and Moody, R., Lie algebras graded by finite root systems and the inter- section matrix algebras of Slodowy, Invent. Math. 108 (1992), 323–347. Bordemann, M., Nondegenerate invariant bilinear forms on nonassociative algebras, Acta Math. Univ. Comenian. (N.S.), 66 (1997), 151–201. Bourbaki, N., Groupes et Alg`ebres de Lie, Ch. 1, Hermann Paris 1971. , , , Ch. 4–6, Masson 1981. , Ch. 7–8, Hermann Paris 1975. Dimitrov, I., Futorny, V., and Penkov, I., A reduction theorem for highest weight modules over toroidal Lie algebras, Comm. Math. Phys. 250 (2004), 47–63. Dixmier, J., Enveloping algebras, Graduate Studies in Mathematics 11, revised reprint of the 1977 translation, American Mathematical Society, Providence, RI, 1996. Eswara Rao, S., and Moody, R., Vertex representations for n-toroidal Lie algebras and a generalization of the Virasoro algebra, Comm. Math. Phys. 159 (1994), 239–264. Garc´ıa, E., and Neher, E., Tits-Kantor-Koecher superalgebras of Jordan superpairs covered by grids, Comm. Algebra 31 (2003), 3335–3375. Garland, H., The arithmetic theory of loop groups, Inst. Hautes ´Etudes Sci. Publ. Math. 52 (1980), 5–136. Helgason, S., Differential Geometry, Lie Groups, and Symmetric Spaces, Pure and Applied Mathematics 80, Academic Press New York 1978. Høegh-Krohn, R. and Torr´esani, B., Classification and construction of quasisimple Lie algebras, J. Funct. Anal. 89 (1990), 106–136. Humphreys, J. E., Introduction to Lie algebras and Representation Theory, Graduate Texts in Mathematics 9, Springer-Verlag New York, 1972. Jacobson, N., Lie algebras, Interscience Tracts in Pure and Applied Mathematics 10, Interscience Publishers (a division of John Wiley & Sons), New York-London 1962. , Infinite dimensional Lie algebras, third edition, Cambridge University Press, Cambridge 1990. Kassel, C., Kahler differentials and coverings of complex simple Lie algebras extended over a commutative algebra, Proceedings of the Luminy conference on algebraic K- theory (Luminy, 1983), J. Pure Appl. Algebra 34 (1984), 265–275. Loos, O., Spiegelungsraume und homogene symmetrische Raume, Math. Z. 99 (1967), 141–170. Loos, O. and Neher, E., Reflection systems and partial root systems, Jordan Theory Preprint Archives, paper #182, to appear in Forum Math. Moody, R. and Pianzola, A., Lie algebras with triangular decompositions, Can. Math. Soc. series of monographs and advanced texts, John Wiley, 1995. [MRK] Moody, R. V., Rao, S. E. and Yokonuma, T., Toroidal Lie algebras and vertex repre- [MY] [Na] [Ne1] [Ne2] [Ne3] [Ne4] [Ne5] [Ne6] [NT] [NY] sentations, Geom. Dedicata 35 (1990), 283–307. Morita, J. and Yoshii, Y., Locally extended affine Lie algebras, J. Algebra 301 (2006), 59–81. Naoi, K., Multiloop Lie algebras and the construction of extended affine Lie algebras, arXiv:0807.2019. Neher, E.,Lie algebras graded by 3-graded root systems and Jordan pairs covered by a grid, Amer. J. Math 118 (1996), 439–491. , An introduction to universal central extensions of Lie superalgebras, Groups, rings, Lie and Hopf algebras (St. John's, NF, 2001), Math. Appl., vol. 555, Kluwer Acad. Publ., Dordrecht, 2003, pp. 141–166. , Lie tori, C. R. Math. Acad. Sci. Soc. R. Can. 26 (2004), 84–89. , Extended affine Lie algebras, C. R. Math. Acad. Sci. Soc. R. Can. 26 (2004), 90–96. , Extended affine Lie algebras and other generalizations – a survey, in "Trends and developments in infinite dimensional Lie theory", Progress in Mathematics, Birkhauser, to appear in 2009. , Universal central extensions of graded Lie algebras, preprint 2010. Neher, E. and Toc´on, M., Lie tori of type B2 and graded-simple Jordan structures covered by a triangle, preprint 2009. Neher, E. and Yoshii, Y., Derivations and invariant forms of Jordan and alternative tori, Trans. Amer. Math. Soc. 355 (2003), 1079–1108. 62 [RSS] [vdK] [We] [Wi] [Y1] [Y2] [Y3] [Y4] ERHARD NEHER Ramos, E., Sah, C.-H. and Shrock, R. E., Algebras of diffeomorphisms of the N -torus, J. Math. Phys. 31 (1990), 1805–1816. van der Kallen, W., Infinitesimally central extensions of Chevalley groups, Springer- Verlag, Berlin, 1973, Lecture Notes in Mathematics, Vol. 356. C. Weibel, C., An introduction to homological algebra, Cambridge studies in advanced mathematics, vol. 38, Cambridge University Press, 1994. Wilson, R., Euclidean Lie algebras are universal central extensions, Lie algebras and related topics (New Brunswick, N.J., 1981), Lecture Notes in Math., vol. 933, Springer, Berlin, 1982, pp. 210–213. Yoshii, Y., Coordinate Algebras of Extended Affine Lie Algebras of Type A1, J. Algebra 234 (2000), 128–168. , Root-graded Lie algebras with compatible grading, Comm. Algebra 29 (2001), 3365–3391. , Root systems extended by an abelian group and their Lie algebras, J. Lie Theory 14 (2004), 371–394. , Lie tori-a simple characterization of extended affine Lie algebras, Publ. Res. Inst. Math. Sci. 42 (2006), 739–762. Department of Mathematics and Statistics, University of Ottawa, Ottawa, Ontario, K1N 6N5, Canada E-mail address: [email protected],
1810.01636	1	1810	2018-10-03T08:55:02	The classification of $2$-dimensional rigid algebras	[ "math.RA" ]	Using the algebraic classification of all $2$-dimensional algebras, we give the algebraic classification of all $2$-dimensional rigid, conservative and terminal algebras over an algebraically closed field of characteristic 0. We have the geometric classification of the variety of $2$-dimensional terminal algebras, and based on the geometric classification of these algebras we formulate some open problems.	math.RA	math	The classification of 2-dimensional rigid algebras Antonio Jes ´us Calder´ona, Amir Fern´andez Ouaridia, Ivan Kaygorodovb a Universidad de C´adiz. Puerto Real, C´adiz, Espana. b CMCC, Universidade Federal do ABC. Santo Andr´e, Brasil. E-mail addresses: Antonio Jes´us Calder´on ([email protected]), Amir Fern´andez Ouaridi ([email protected]), Ivan Kaygorodov ([email protected]). Abstract. Using the algebraic classification of all 2-dimensional algebras, we give the algebraic classification of all 2-dimensional rigid, conservative and terminal algebras over an algebraically closed field of characteristic 0. We have the geometric classification of the variety of 2-dimensional terminal algebras, and based on the geometric classification of these algebras we formulate some open problems. INTRODUCTION 0.1. Conservative, terminal and rigid algebras. In 1972, Kantor introduced the notion a conservative algebra as a generalization of Jordan algebras [9]. Unlike other classes of non-associative algebras, this class is not defined by a set of identities. Instead they are defined in the following way. Consider an algebra as a vector space V over a field k, together with an element µ of Hom(V ⊗ V, V), so that a · b = µ(a ⊗ b). Given a linear map A : V → V and a bilinear map B : V × V → V, we define the product of a linear map and a bilinear map as the map [A, B] : V × V → V such that [A, B](x, y) = A(B(x, y)) − B(A(x), y) − B(x, A(y)), for all x, y ∈ V. For an algebra (V, P) with a multiplication P and x ∈ V we denote by LP Thus, Kantor defines conservative algebras as follow. x the operator of left multiplication by x. Definition 1. An algebra (V, P), where V is the vector space and P is the multiplication, is called a conservative algebra if there is a new multiplication F : V × V → V such that (1) [LP b , [LP a , P]] = −[LP F (a,b), P], for all a, b ∈ V. Simple calculations take us to the following identity with an additional multiplication F , which must hold for all a, b, x, y ∈ V: (2) b(a(xy) − (ax)y − x(ay)) − a((bx)y) + (a(bx))y + (bx)(ay) − a(x(by)) + (ax)(by) + x(a(by)) = = −F (a, b)(xy) + (F (a, b)x)y + x(F (a, b)y). The class of conservative algebras is very vast [14]. It includes: all associative algebras, all quasi-associative algebras, all Jordan algebras, all Lie algebras, all (left) Leibniz algebras, all (left) Zinbiel algebras, all terminal algebras and many other classes of algebras. There are some properties of conservative algebras which are similar to wonderful properties of Lie algebras. The conservative algebra W (n) plays a similar role in the theory of conservative algebras as the Lie algebra of all n × n matrices gln plays in the theory of Lie algebras. Namely, in [12] Kantor considered the category Sn whose objects are conservative algebras of non-Jacobi dimension n, and proven that the algebra W (n) is the universal attracting object in this category, i.e., for every algebra M of Sn there exists a canonical homomorphism from M into the algebra W (n). In particular, all Jordan algebras of dimension n with unity are contained in the algebra W (n). Some properties of the product in the algebra W (n) were studied in [13, 14, 18]. In 1989, Kantor introduced the class of terminal algebras which is subclass of the class of conservative algebras [10]. To introduce the notion of terminal algebra, we first define the product of two bilinear maps. Given two bilinear maps A : V × V → V and B : V × V → V, we define the operation [A, B] : V × V × V → V such that for all x, y, z ∈ V: [A, B](x, y, z) = A(B(x, y), z) + A(x, B(y, z)) + A(y, B(x, z)) − B(A(x, y), z) − B(x, A(y, z)) − B(y, A(x, z)). Also, for a ∈ V and a bilinear map A : V × V → V, we introduce the operation [A, a](x) = A(a, x). Thus, we define: Definition 2. An algebra (V, P), where V is a vector space and P is a multiplication, is called a terminal algebra if it satisfies, for any a ∈ V: (3) [[[P, a], P], P] = 0. 1 2 The following remark is obtained by straightforward calculations. Remark 3. Any commutative algebra satisfying (3) is a Jordan algebra. The class of terminal algebras includes all Jordan algebras, all Lie algebras, all (left) Leibniz algebras and some other types of algebras. The following characterization of terminal algebras, proved by Kantor [10, Theorem 2], provides a description of this class as a subclass of the class of conservative algebras. Remark 4. An algebra (V, P) is terminal if and only if, for any a, b ∈ V: (4) [LP b , [LP a , P]] = −[LP 2/3P(a,b)+1/3P(b,a), P]. Then Kantor introduced a generalization of conservative algebras [11]: Definition 5. An algebra (V, P), where V is a vector space and P is a multiplication, is called a quasi-conservative algebra if there are a multiplication F : V × V → V and a bilinear form φ : V × V → k, such that (5) [LP b , [LP a , P]] = −[LP F (a,b), P] + φ(a, b)P, for all a, b ∈ V. Let us recall that the structural Lie algebra Str(V, µ) is the subalgebra of the Lie algebra End(V, V)-generated by all operators of left multiplication µa(b) = µ(a ⊗ b), a ∈ V, and denote by R(V, µ) the minimal submodule of the Str(V, µ)-module Hom(V ⊗ V, V), containing µ. Following Kac and Cantarini [5], we can give the following Definition 6. An algebra (V, µ) where V is a vector space and µ is a multiplication, is called a rigid algebra if it satisfies: (6) R(V, µ) = Str(V, µ)µ + kµ. Thus, (6) means a certain rigidity property. Namely, in the case of Jordan algebras, this property means that "small" deformations of the product by the structural group produce an isomorphic algebra. The class of rigid algebras is very vast: it includes all associative algebras, all Jordan algebras, all Lie algebras, all conservative algebras and many other types of algebras. Remark 7. An algebra A is rigid if and only if A is quasi-conservative. 0.2. The classification of 2-dimensional algebras. The study of 2-dimensional algebras has a very big history [4,19, 20]. To give the classification of 2-dimensional algebras we have to introduce some notation used in the latest algebraic classification of 2-dimensional algebras [19]. Let us consider the action of the cyclic group C2 = hρ ρ2i on k defined by the equality ρα = −α for α ∈ k. Now, fix some set of representatives of the orbits under this action and denote it by k≥0. For example, if k = C, then one can take C≥0 = {α ∈ C Re(α) > 0} ∪ {α ∈ C Re(α) = 0, Im(α) ≥ 0}. Let us also consider the action of C2 on k2 defined by the equality ρ(α, β) = (1 − α + β, β) for (α, β) ∈ k2. Let us fix some set of representatives of the orbits under this action and denote it by U . Let us also define T = {(α, β) ∈ k2 α + β = 1}. Given (α, β, γ, δ) ∈ k4, we define D(α, β, γ, δ) = (α + γ)(β + δ) − 1. We define C1(α, β, γ, δ) = (β, δ), C2(α, β, γ, δ) = (γ, α), and C3(α, β, γ, δ) = (cid:16) βγ−(α−1)(δ−1) D(α,β,γ,δ) (cid:17) for (α, β, γ, δ) such that D(α, β, γ, δ) 6= 0. Let us consider the set (cid:8)(cid:0)C1(Γ), C2(Γ), C3(Γ)(cid:1) Γ ∈ k4, D(Γ) 6= 0, C1(Γ), C2(Γ) 6∈ T (cid:9) ⊂ (k2)3. One can show that the symmetric group S3 acts on this set by the equality , αδ−(β−1)(γ−1) D(α,β,γ,δ) σ(cid:0)C1(Γ), C2(Γ), C3(Γ)(cid:1) = (cid:0)Cσ−1(1)(Γ), Cσ−1(2)(Γ), Cσ−1(3)(Γ)(cid:1) for σ ∈ S3. Note that there exists a set of representatives of orbits under this action V such that if (C1, C2, C3) ∈ V and C1 6= C2, then C3 6= C1, C2. Let us fix such V, and define For Γ ∈ V, we also define C(Γ) = {C1(Γ), C2(Γ), C3(Γ)} ⊂ k2. V = {Γ ∈ k4 D(Γ) 6= 0; C1(Γ), C2(Γ) 6∈ T,(cid:0)C1(Γ), C2(Γ), C3(Γ)(cid:1) ∈ V}. Let us consider the action of the cyclic group C2 on k∗ \ {1} defined by the equality ρα = α−1 for α ∈ k∗ \ {1}. >1. For example, if k = C, then >1, we Let us fix some set of representatives of orbits under this action and denote it by k∗ one can take C∗ define >1 = {α ∈ C∗ α > 1} ∪ {α ∈ C∗ α = 1, 0 < arg(α) ≤ π}. For (α, β, γ) ∈ k2 × k∗ Now, from [19] we have the result that classifies all 2-dimensional algebras over an algebraically closed field k. C(α, β, γ) = (cid:26)(cid:0)αγ, (1 − α)γ(cid:1),(cid:18) β γ , 1 − β γ (cid:19)(cid:27) ⊂ k2. 3 Theorem 8. Any non-trivial 2-dimensional k-algebra can be represented by a unique structure from Table 1 in the appendix. In this paper, we consider algebraically closed field k of characteristic zero. 1. THE ALGEBRAIC CLASSIFICATION OF 2-DIMENSIONAL RIGID (QUASI-CONSERVATIVE) AND CONSERVATIVE ALGEBRAS 1.1. The algebraic classification of 2-dimensional rigid algebras. The following result presents this classification. Theorem 9. Let A be a 2-dimensional rigid algebra over an algebraically closed field k of characteristic zero, then A is isomorphic to one of the non-isomorphic algebras presented in Table 2 in the appendix. Remark 10. Let A = (V, P) be a 2-dimensional algebra, with {e1, e2} a basis of A. We can prove that A is rigid by F and φ for the following cases: (1.a) a = b = x = y = e1; (2.a) a = x = y = e1, b = e2; b = x = y = e1, a = e2; (3.a) (4.a) a = b = y = e1, x = e2; (5.a) a = b = x = e1, y = e2; (6.a) x = y = e1, a = b = e2; (7.a) a = x = e1, b = y = e2; b = x = e1, a = y = e2; (8.a) (1.b) a = b = x = y = e2; (2.b) a = x = y = e2, b = e1; b = x = y = e2, a = e1; (3.b) (4.b) a = b = y = e2, x = e1; (5.b) a = b = x = e2, y = e1; (6.b) x = y = e2, a = b = e1; (7.b) a = x = e2, b = y = e1; b = x = e2, a = y = e1. (8.b) Let F : V × V → V be a bilinear map: F (e1, e1) = λ1e1 + λ2e2, F (e1, e2) = µ1e1 + µ2e2, F (e2, e1) = τ1e1 + τ2e2, F (e2, e2) = ν1e1 + ν2e2. Also, let φ : V × V → k be a bilinear form: φ(e1, e1) = φ11 φ(e1, e2) = φ12 φ(e2, e1) = φ21 φ(e2, e2) = φ22. Using the cases described above, we study each family from Table 1 in the appendix. This procedure lead us to a system of equations that can be solved without too much difficulty. Therefore, we just give the complete procedure for the first case, the other cases can be obtained in an analogous way. 1.1.1. Algebra A1(α), α ∈ k. From Remark 10, we obtain the following necessary conditions for the structural constants of F and φ (we have omitted trivial cases): (1.a) λ1 + 2φ11 = 1 and (−2 + α)(2 − α − λ1) + 2φ11 = 0, (3.a) τ1 + φ21 = 0 and (2 − α)τ1 + φ21 = 0, (4.b) α(ν1 + φ22) = 0, (5.b) (1 − α)(ν1 + φ22) = 0, (7.a) α(µ1 + φ12) = 0, (8.a) α(τ1 + φ21) = 0, (2.a) µ1 + φ12 = 0 and (2 − α)µ1 + φ12 = 0, (4.a) 1 − α = (1 − α)(λ1 + φ11), (5.a) α = α(λ1 + φ11), (6.a) (7.b) (8.b) ν1 + φ22 = 0 and (2 − α)ν1 + φ22 = 0, (1 − α)(τ1 + φ21) = 0, (1 − α)(µ1 + φ12) = 0. Solving this system of equations, we conclude that A1(α) is rigid in the following cases: (1) A1(1), where F is given by: F (e1, e1) = e1 + λ2e2, F (e1, e2) = −φ12e1 + µ2e2, F (e2, e1) = −φ21e1 + τ2e2, F (e2, e2) = −φ22e1 + ν2e2. and φ is given by: φ(e1, e2) = φ12, φ(e1, e1) = 0, φ(e2, e1) = φ21, φ(e2, e2) = φ22. 4 (2) A1(2), where F is given by: F (e1, e1) = e1 + λ2e2, F (e1, e2) = µ2e2, F (e2, e1) = τ2e2, F (e2, e2) = ν2e2, and φ = 0. 1.1.2. Algebra A2. The algebra A2 is rigid. The choice of F is given by: F (e1, e1) = −e1 + λ2e2, F (e1, e2) = µ2e2, F (e2, e1) = τ2e2, F (e2, e2) = ν2e2, and φ = 0. 1.1.3. Algebra A3. The algebra A3 is a Leibniz algebra, and obviously, is rigid for any multiplication F and φ = 0. 1.1.4. Algebra A4(α), α ∈ k≥0. From the conditions (4.b), (5.b), (6.a) and (6.b), we conclude that A4(α) is not rigid for any α ∈ k≥0. 1.1.5. Algebra B1(α), α ∈ k. The condition (6.a) shows that B1(α) is not rigid for any α ∈ k. 1.1.6. Algebra B2(α), α ∈ k. The algebra B2(α) is rigid. If α = 1 then F is given by an arbitrary map, otherwise F is given by: and φ = 0. F (e1, e1) = λ2e2, F (e1, e2) = −αe1 + µ2e2, F (e2, e1) = τ2e2, F (e2, e2) = ν2e2 1.1.7. Algebra B3. The algebra B3 is a Lie algebra, and obviously, is rigid for any multiplication F and φ = 0. 1.1.8. Algebra C(α, β), (α, β) ∈ k × k≥0. The algebra C(α, β) is rigid if and only if (α, β) = (1, 0), for any φ, and for F given by: F (e1, e1) = −φ11e2, F (e1, e2) = e1 − φ12e2, F (e2, e1) = e1 − φ21e2, F (e2, e2) = (1 − φ22)e2. 1.1.9. Algebra D1(α, β), (α, β) ∈ U . The algebra D1(α, β) is rigid in the following cases: (1) D1(0, 0). φ = 0 and F is given by: (2) D1(1/2, 0). F is given by: and φ is given by: F (e1, e1) = e1 + λ2e2, F (e1, e2) = µ2e2, F (e2, e1) = τ2e2, F (e2, e2) = ν2e2. F (e1, e1) = 0, F (e1, e2) = − 1 F (e2, e1) = 0, F (e2, e2) = 1 2 e2, 2 e1 + e2, φ(e1, e1) = 1, φ(e1, e2) = 1 2 , φ(e2, e1) = 1 2 , φ(e2, e2) = 0. (3) D1(1, 1). φ is arbitrary and F is given by: F (e1, e1) = (1 − φ11)e1, F (e1, e2) = −φ11e1 + e2, F (e2, e1) = −φ21e1 + e2, F (e2, e2) = −φ22e1 + e2. 1.1.10. Algebra D2(α, β), (α, β) ∈ k2 \ T . The algebra D2(α, β) is rigid for any (α, β) ∈ k2 \ T . φ and F are given by: F (e1, e1) = (1 − φ11)e1 + λ2e2, F (e1, e2) = −φ12e1 + µ2e2, F (e2, e2) = −φ22e1 + ν2e2, F (e2, e1) = −φ21e1 + τ2e2, where βλ2 = 0, β = βµ2, (2 − α)β = βτ2 and βν2 = 0. Now we have next cases: • If β = 0 then φ is arbitrary and F is given by: F (e1, e1) = (1 − φ11)e1 + λ2e2, F (e1, e2) = −φ12e1 + µ2e2, F (e2, e2) = −φ22e1 + ν2e2. F (e2, e1) = −φ21e1 + τ2e2, • If β 6= 0 then φ is arbitrary and F is given by: F (e1, e1) = (1 − φ11)e1, F (e2, e1) = −φ21e1 + (2 − α)e2, F (e2, e2) = −φ22e1. F (e1, e2) = −φ12e1 + e2, 5 1.1.11. Algebra D3(α, β), (α, β) ∈ k2 \ T . From condition (6.b), we conclude that D3(α, β) is not rigid. 1.1.12. Algebra E1(α, β, γ, δ), (α, β, γ, δ) ∈ V. The algebra E1(α, β, γ, δ) is rigid in the following cases: (1) E1(δ′, 1 + δ′, 1 + δ′, δ′), (δ′, 1 + δ′, 1 + δ′, δ′) ∈ V. F is given by: and φ is given by: F (e1, e1) = (2 + δ)e1, F (e2, e1) = e1 + (1 − δ)e2, F (e2, e2) = (2 + δ)e2. F (e1, e2) = (1 − δ)e1 + e2, φ(e1, e1) = −1 − δ, φ(e2, e1) = −1 + δ + 2δ2, φ(e2, e2) = −1 − δ. φ(e1, e2) = −1 + δ + 2δ2, (2) E1(0, β′, γ ′, 0), (0, β′, γ ′, 0) ∈ V. φ is arbitrary and F is given by: F (e1, e1) = β 2γ+φ11−γφ11−1 F (e2, e1) = γ(β−φ21−1)+φ21 e1 + (1−β)(β+φ11) e2, e1 + β(γ−φ21−1)+φ21 −1+βγ e2, F (e2, e2) = (1−γ)(γ+φ22) F (e1, e2) = γ(β−φ12−1)+φ12 −1+βγ −1+βγ −1+βγ −1+βγ −1+βγ e1 + β(γ−φ12−1)+φ12 e2. e1 + β(γ 2−φ22)−1+φ22 −1+βγ −1+βγ e2, 1.1.13. Algebra E2(α, β, γ), (α, β, γ) ∈ k3 \ k × T . The algebra E2(α, β, γ) is rigid for the following parameters: (1) E2(1, 1, γ), (1, 1, γ) ∈ k3 \ k × T . φ is arbitrary and F is given by: F (e1, e1) = (1 − φ11)e1, F (e1, e2) = −φ12e1 + e2, F (e2, e1) = −φ21e1 + e2, F (e2, e2) = (−1 − φ22)e1 + 2e2. (2) E2(1, β, 0), (1, β, 0) ∈ k3 \ k × T. φ is arbitrary and F is given by: F (e1, e1) = (1 + β)e1 + (−β − φ11)e2, F (e1, e2) = e1 − φ12e2, F (e2, e1) = e1 − φ21e2, F (e2, e2) = (1 − φ22)e2. 1.1.14. Algebra E3(α, β, γ), (α, β, γ) ∈ k2 × k∗ >1. The algebra E3(α, β, γ) is rigid for the following parameters: (1) E3(1, γ, γ), γ ∈ k∗ >1. φ is arbitrary and F is given by: F (e1, e1) = (1 − φ11)e1, F (e1, e2) = −φ12e1 + e2, F (e2, e1) = −φ21e1 + e2, F (e2, e2) = (−γ − φ22)e1 + (1 + γ)e2. (2) E3( 1 γ , 1, γ), γ ∈ k∗ >1. φ is arbitrary and F is given by: F (e1, e1) = 1 + γ γ e1 − 1 + γφ11 γ e2, F (e1, e2) = e1 − φ12e2, F (e2, e1) = e1 − φ21e2, F (e2, e2) = (1 − φ22)e2. (3) E3(1, 1, γ), γ ∈ k∗ >1. F is given by: 1 + γ − γ2λ2 γ F (e1, e1) = e1 + λ2e2, F (e1, e2) = (1 + γ − γµ2)e1 + µ2e2, F (e2, e1) = (1 + γ − γτ2)e1 + τ2e2, F (e2, e2) = γ(1 + γ − ν2)e1 + ν2e2, and φ is given by: φ(e1, e1) = − 1 γ φ(e2, e1) = −1, , φ(e1, e2) = −1, φ(e2, e2) = −γ. (4) E3(0, 0, −1). φ = 0 and F is given by: F (e1, e1) = e1, F (e2, e1) = e1 + 2e2, F (e2, e2) = e2. F (e1, e2) = 2e1 + e2, 1.1.15. Algebra E4. The algebra E4 is rigid for F is given by: and φ is given by: F (e1, e1) = 2e1, F (e1, e2) = e1 + e2, F (e2, e1) = e2, F (e2, e2) = e2, φ(e1, e1) = −1, φ(e1, e2) = −1, φ(e2, e1) = 0, φ(e2, e2) = 0. 6 1.1.16. Algebra E5(α), α ∈ k. The algebra E5(α) is rigid for any α ∈ k. φ is arbitrary and F is given by: F (e1, e1) = λ1e1 + (1 − φ11 − λ1)e2, F (e1, e2) = µ1e1 + (1 − φ12 − µ1)e2, F (e2, e1) = τ1e1 + (1 − φ21 − τ1)e2, F (e2, e2) = ν1e1 + (1 − φ22 − ν1)e2. 1.2. The algebraic classification of 2-dimensional conservative algebras. As a corollary of the classification of 2-dimensional rigid algebras, we have the following result. Theorem 11. Let A be a 2-dimensional conservative algebra over an algebraically closed field k of characteristic zero, then A is isomorphic to one of the non-isomorphic algebras presented in Table 3 in the appendix. Proof. In the previous results, choose φ = 0 when possible, to obtain the conservative algebras. (cid:3) 2. THE CLASSIFICATION OF 2-DIMENSIONAL TERMINAL ALGEBRAS The aim of this section is to present the algebraic and geometric classification of the class of the terminal algebras. 2.1. The algebraic classification of 2-dimensional terminal algebras. As a corollary of the classification of 2- dimensional conservative algebras, we have the following result. Theorem 12. Let A be a 2-dimensional terminal algebra over an algebraically closed field k of characteristic zero, then A is isomorphic to one of the non-isomorphic algebras presented in Table 4 in the appendix. Proof. In the previous results, choose F (a, b) = 2/3P(a, b) + 1/3P(b, a) when possible, to obtain the terminal algebras. (cid:3) 2.2. Closures of orbit of 2-dimensional terminal algebras. This subsection is devoted to show the geometric clas- sification of the variety of 2-dimensional terminal algebras. Geometric classification is an interesting subject, which was studied in various papers (see, for example, [3, 6 -- 8, 16]). One of the problems in this direction is to describe all degenerations in a variety of algebras of some fixed dimension satisfying some set of identities. For example, this problem was solved for 2-dimensional pre-Lie algebras in [2], for 2-dimensional Jordan algebras in [1], for 3-dimensional Novikov algebras, for 4-dimensional Lie algebras, for 4-dimensional Zinbiel, and nilpotent Leibniz algebras in [16], for nilpotent 5- and 6-dimensional Lie algebras in [6], and for nilpotent 5-dimensional and 6-dimensional Malcev algebras in [17]. We will denote by T2 the set of all µ ∈ A2 := Hom(V⊗ V, V) such that µ is a representation of a terminal algebra. Consider the Zariski topology on A2, giving it the structure of an affine variety. Since T2 ⊂ A2 is defined by a set of polynomial identities, then T2 is a Zariski-closed of the variety of all two dimensional algebras. Thus, any µ ∈ T2 is determined by the structure constants ck ij ∈ k (i, j, k = 1, 2) such that µ(ei ⊗ ej) = c1 ij e1 + c2 ije2. In the previous subsection, we gave a decomposition of T2 into GL(V)-orbits (acting by conjugation), i.e, a classi- fication, up to isomorphism, of the terminal algebras. In this subsection, we will describe the closures of the orbits of µ ∈ T2, denoted by O(µ), and we will give a geometric classification of T2, by describing it's irreducible components. For that purpose, we will introduce some definitions. Let A and B be two 2-dimensional algebras and let µ, λ ∈ T2 represents A and B respectively. We say that A degenerates to B and write A → B if λ ∈ O(µ). Moreover, we have O(λ) ⊂ O(µ). Hence, the definition of a degeneration does not depend on the choice of the representative µ and λ. We write A 6→ B if λ 6∈ O(µ). Now, let A(∗) := {A(α)}α∈I be a set of 2-dimensional algebras and µ(α) ∈ T2 represent A(α) for α ∈ I. If λ ∈ {O(µ(α))}α∈I , then we write A(∗) → B and say that A(∗) degenerates to B. Again, in the opposite case we write A(∗) 6→ B. Let A(∗), B, µ(α) (α ∈ I) and λ be as above. Let ck e1, e2. If we construct aj a basis of V for any t ∈ k∗, and the structure constants of µ(f (t)) in this basis are ck then A(∗) → B. In this case (a1 parametrized index for A(∗) → B respectively. Note that in the case of I = 1 we only need a parametrized basis. ij (i, j, k = 1, 2) be the structure constants of λ in the basis 2(t)e2 is ij(0) = ck ij , 2(t)e2) and f (t) are called a parametrized basis and a i : k∗ → k (i, j = 1, 2) and f : k∗ → I such that a1 ij (t) ∈ k[t] such that ck 1(t)e2, a1 2(t)e1 + a2 1(t)e1 + a2 1(t)e1 + a2 1(t)e2 and a1 2(t)e1 + a2 The following lemma holds: Lemma 13. Let A → B be a proper degeneration. Then it follows: (1) dim Aut(A) < dim Aut(B). (2) dim [A, A] ≥ dim [B, B]. The following result in [16] gives us a constructive method to prove non-degenerations. 7 Lemma 14. Let B be a Borel subgroup of GL(V) and let R be a closed subset of T2 such that R is stable under the action of B. If A(∗) → B and µ(α), a structure representing of A(α), is in R for all α ∈ I, then there exists a representation λ of B such that λ ∈ R. Constructing a set R under the conditions of the previous result, such that µ(α) ∈ R for any α ∈ I and O(λ)∩R = ∅, gives us the non-degeneration A(∗) 6→ B. In this case, we call R a separating set for A(∗) 6→ B. In this paper, we always choose B as the lower triangular matrices. To prove a non-degeneration, we present the separating set and omit any verification, which can be obtained by straightforward calculations. Theorem 15. The variety of all 2-dimensional terminal algebras has the graph of primary degenerations presented on Figure 1 in the appendix. Proof. This proof is mostly based in results in [19] and in the ideas described there to construct separating sets. All primary degenerations are showed in Table 5 in the appendix and all non-degenerations that do not follow from Lemma 13 are presented in Table 6 in the appendix. (cid:3) The following result gives us the description of the closure of the orbits of the infinite series in T2. For a parametric series of algebras X, we will denote by X(∗), the set of all algebras X(Γ) that are defined and are terminal, i.e, T07(∗) = {T07(α) : α ∈ k \ {1}}, T08(∗) = {T08(α) : α ∈ k \ {2}} and T10(∗) = {T10(α) : α ∈ k}. Theorem 16. A description of the closures of the orbits of the infinite series of algebras in T2 is given in the following table. Set Description O(T07(∗)) T07(∗), T03, T01, T04, T10(1), k2 O(T08(∗)) T08(∗), T03, T02, T05, T10(2), T07(3/2), k2 O(T10(∗)) T10(∗), T06, k2 Proof. Degenerations are proved using the parametrized bases and indexes in Table 7 in the appendix. Also, the non-degenerations can be proven using the separating sets in Table 8 in the appendix. (cid:3) From Theorem 15 and Theorem 16 we obtain the following corollaries. Corollary 17. The lattice of subsets for T2 is presented on Figure 2 in the appendix. Corollary 18. The irreducible components in the variety of 2-dimensional terminal algebras T2 are: O(cid:0)T09(cid:1) = {T09, T07(0), T08(1), T03, k2}, O(cid:0)T07(∗)(cid:1) = {T07(∗), T03, T01, T04, T10(1), k2}, O(cid:0)T08(∗)(cid:1) = {T08(∗), T03, T02, T05, T10(2), T07(3/2), k2}, O(cid:0)T10(∗)(cid:1) = {T10(∗), T06, k2}. Corollary 19. There is only one algebra with open orbit in the variety of 2-dimensional terminal algebras. It is T09. 2.3. Some conjectures. Using the geometric classification of 2-dimensional terminal algebras we can give two con- jectures about the variety of n-dimensional terminal algebras. Let us consider n-dimensional analogues of the algebras T09 and T10(α) : I. the algebra ⊕kei defined by ⊕kei = he1, . . . , eni : e2 i = ei, eiej = 0, (i 6= j); II. the family νn(α) of algebras defined by νn(α) = he, n1, . . . , nn−1i : e2 = e, eni = αni, nie = (1 − α)ni (i = 1, . . . , n − 1; α ∈ k). It is easy to see that algebras ⊕kei and νn(α) are terminal. Conjecture 1. The n-dimensional terminal algebraL kei has an open orbit. Conjecture 2. O(cid:0)νn(α)(cid:1) is an irreducible component of the variety of n-dimensional terminal algebras. Acknowledgment. The authors would like to thank the referee for his exhaustive review of the paper as well as his suggestions which have helped to improve the work. 8 Table 1. Algebraic classification of 2-dimensional algebras 3. APPENDIX: TABLES ≥0 Designation A1(α), α ∈ k A2 A3 A4(α), α ∈ k B1(α), α ∈ k B2(α), α ∈ k B3 C(α, β), (α, β) ∈ k × k D1(α, β), (α, β) ∈ U D2(α, β), (α, β) ∈ k2 \ T D3(α, β), (α, β) ∈ k2 \ T E1(α, β, γ, δ), (α, β, γ, δ) ∈ V E2(α, β, γ), (α, β, γ) ∈ k3 \ k × T E3(α, β, γ), (α, β, γ) ∈ k2 × k∗ E4 E5(α), α ∈ k ≥0 >1 Table 2. Algebraic classification of 2-dimensional rigid algebras Multiplication table e1e1 = e1 + e2, e1e1 = e2, e1e1 = e2, e1e1 = αe1 + e2, e1e1 = 0, e1e1 = 0, e1e1 = 0, e1e1 = e2, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e2 = αe2, e1e2 = e2, e1e2 = 0, e1e2 = e1 + αe2, e1e2 = (1 − α)e1 + e2, e1e2 = (1 − α)e1, e1e2 = e2, e1e2 = (1 − α)e1 + βe2, e1e2 = (1 − α)e1 + βe2, e1e2 = αe2, e1e2 = e1 + αe2, e1e2 = αe1 + βe2, e1e2 = (1 − α)e1 + βe2, e1e2 = (1 − α)γe1 + β e1e2 = e1 + e2, e1e2 = (1 − α)e1 + αe2, γ e2, e2e1 = (1 − α)e2, e2e1 = −e2, e2e1 = 0, e2e1 = −e1, e2e1 = αe1 − e2, e2e1 = αe1, e2e1 = −e2, e2e1 = αe1 − βe2, e2e1 = αe1 − βe2, e2e1 = βe2, e2e1 = −e1 + βe2, e2e1 = γe1 + δe2, e2e1 = αe1 + γe2 , 1−β e2e1 = αγe1 + e2e1 = 0, e2e1 = αe1 + (1 − α)e2, γ e2, R01 R02 R03 R04 R05(α), α ∈ k R06 R07 R08 R09 R10 R11(α, β), (α, β) ∈ k2 \ T R12(α), (α, 1 + α, 1 + α, α) ∈ V R13(α, β), (0, α, β, 0) ∈ V R14(α), α ∈ k∗ R15(α), α ∈ k \ {1} R16(α), α ∈ k∗ R17(α), α ∈ k∗ R18(α), α ∈ k∗ R19 R20 R21(α), α ∈ k >1 >1 >1 A1(1) A1(2) A2 A3 B2(α) B3 C(1, 0) D1(0, 0) e1e1 = e1 + e2, e1e1 = e1 + e2, e1e1 = e2, e1e1 = e2, e1e1 = 0, e1e1 = 0, e1e1 = e2, e1e1 = e1, D1(1/2, 0) D1(1, 1) D2(α, β) E1(α, 1 + α, 1 + α, α) E1(0, α, β, 0) E2(1, 1, α) E2(1, α, 0) E3(1, α, α) E3(1/α, 1, α) E3(1, 1, α) E3(0, 0, −1) E4 E5(α), α ∈ k e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, Table 3. Algebraic classification of 2-dimensional conservative algebras C01 C02 C03 C04 C05(α), α ∈ k C06 C07 C08 C09 C10(α, β), (α, β) ∈ k2 \ T C11(α, β), (0, α, β, 0) ∈ V C12(α), α ∈ k∗ C13(α), α ∈ k \ {1} C14(α), α ∈ k∗ C15(α), α ∈ k∗ C16 C17(α), α ∈ k >1 >1 A1(1) A1(2) A2 A3 B2(α) B3 C(1, 0) D1(0, 0) D1(1, 1) D2(α, β) E1(0, α, β, 0) E2(1, 1, α) E2(1, α, 0) E3(1, α, α) E3(1/α, 1, α) E3(0, 0, −1) E5(α) e1e1 = e1 + e2, e1e1 = e1 + e2, e1e1 = e2, e1e1 = e2, e1e1 = 0, e1e1 = 0, e1e1 = e2, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e2 = e2, e1e2 = 2e2 , e1e2 = e2, e1e2 = 0, e1e2 = (1 − α)e1, e1e2 = e2, e1e2 = 0, e1e2 = e1, e1e2 = 1 e1e2 = e2, e1e2 = αe2, e1e2 = αe1 + (1 + α)e2, e1e2 = αe2, e1e2 = e2, e1e2 = αe2, 2 e1, e1e2 = e2, e1e2 = (α − 1)e1 + 1 e1e2 = 1 α e2, e1e2 = −e1, e1e2 = e1 + e2, e1e2 = (1 − α)e1 + αe2, α e2, e1e2 = e2, e1e2 = 2e2 , e1e2 = e2, e1e2 = 0, e1e2 = (1 − α)e1, e1e2 = e2, e1e2 = 0, e1e2 = e1, e1e2 = e2, e1e2 = αe2, e1e2 = αe2, e1e2 = e2, e1e2 = αe2, e1e2 = e2, e1e2 = (α − 1)e1 + 1 e1e2 = −e1, e1e2 = (1 − α)e1 + αe2, α e2, e2e1 = 0, e2e1 = −e2, e2e1 = −e2, e2e1 = 0, e2e1 = αe1, e2e1 = −e2, e2e1 = e1, e2e1 = 0, e2e1 = 1 e2e1 = e1 − e2, e2e1 = βe2, e2e1 = (1 + α)e1 + αe2, e2e1 = βe1, e2e1 = e1 + αe2, e2e1 = e1, e2e1 = αe1 + 1−α 2 e1, α e2, e2e1 = e1, e2e1 = αe1, e2e1 = −e2, e2e1 = 0, e2e1 = αe1 + (1 − α)e2, e2e1 = 0, e2e1 = −e2, e2e1 = −e2, e2e1 = 0, e2e1 = αe1, e2e1 = −e2, e2e1 = e1, e2e1 = 0, e2e1 = e1 − e2, e2e1 = βe2, e2e1 = βe1, e2e1 = e1 + αe2, e2e1 = e1, e2e1 = αe1 + 1−α α e2, e2e1 = e1, e2e1 = −e2, e2e1 = αe1 + (1 − α)e2, Table 4. Algebraic classification of 2-dimensional terminal algebras T01 T02 T03 T04 T05 T06 T07(α), α ∈ k \ {1} T08(α), α ∈ k \ {2} T09 T10(α), α ∈ k A1(1) A1(2) A3 B2(1) B2(−1) B3 D2(α, 0) D2(α, 3 − 2α 6= 0) E1(0, 0, 0, 0) E5(α) e1e1 = e1 + e2, e1e1 = e1 + e2, e1e1 = e2, e1e1 = 0, e1e1 = 0, e1e1 = 0, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e1 = e1, e1e2 = e2, e1e2 = 2e2 , e1e2 = 0, e1e2 = 0, e1e2 = 2e1 , e1e2 = e2, e1e2 = αe2, e1e2 = αe2, e1e2 = 0, e1e2 = (1 − α)e1 + αe2, e2e1 = 0, e2e1 = −e2, e2e1 = 0, e2e1 = e1, e2e1 = −e1, e2e1 = −e2, e2e1 = 0, e2e1 = (3 − 2α)e2, e2e1 = 0, e2e1 = αe1 + (1 − α)e2, e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = e2 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = e2 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = e2 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = e2 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = 0 e2e2 = e2 e2e2 = e2 9 Parametrized basis Table 5. Degenerations required to prove Theorem 15 Degeneration T01 → T03 T01 → T10(1) T02 → T03 T02 → T10(2) T04 → T03 T05 → T03 T07(α) → T03 T08(α) → T03 T09 → T07(0) T09 → T08(1) Et Et Et Et Et Et Et Et Et Et Et 1 = te1, Et 1 = e1, Et 1 = te1, Et 1 = e1, 1 = e1 + te2, Et 1 = e1 + te2, Et 1 = te1 + te2, Et 1 = te1 + te2, Et Et 1 = e1, Et 1 = e1 + e2, 2 = t2e2 2 = e1 + t−1e2 2 = t2e2 2 = e1 + t−1e2 2 = te1 2 = te1 2 = t2e1 + αt2e2 2 = t2e1 + (3 − α)t2e2 2 = te2 2 = te2 Table 6. Non-degenerations required to prove Theorem 15 Non-degeneration T01 6→ T06, T10(α 6= 1) T02 6→ T06, T10(α 6= 2) T04 T05 6→ T06 6→ T06 T07(α) 6→ T06, T10(γ) T08(α) 6→ T06, T10(γ) T09 6→ T04, T05, T06, T10(α), T07(α 6= 0), T08(α 6= 1) 22 = 0, c2 22 = 0, c1 c1 12 = c1 11, c1 c2 12 2 11 = Separating set 21 = c2 22, c1 12 = 0, c2 , c1 21 = 2c2 22, c1 12 = −c2 21 = − 21 = 0(cid:9) 22, c2 12 = 0, 2c2 22 = c2 22 = c2 12 = 0, c2 12 = 0, c2 22 = c1 22 = 0, c1 12 = αc1 12 = αc1 12 + c2 21, c2 21 = 0, c2 21 = 0, c2 12 = −2c1 11, c1 11, c1 11(c2 11, c1 21 = −c1 11, c1 21 = c1 21 = 0, c1 21 = (3 − 2α)c1 11(c1 22 = 0, c2 11 + c2 12) + c2 22 = 0(cid:9) 22 = 0, c2 11, c1 12 + c2 22 = 0, c1 21 = 0, c1 12 = 0, c2 12 = c2 21, c2 11c2 22 = −c2 21(c1 11 − c2 c2 12 2 (cid:27) 21) = c2 11c1 11c1 21(cid:9) 21 = 0(cid:9) 22 = 0(cid:9) 21)(cid:9) R = (cid:8)µ(cid:12)(cid:12) c1 R = (cid:26)µ(cid:12)(cid:12)(cid:12)(cid:12) R = (cid:8)µ(cid:12)(cid:12) c1 R = (cid:8)µ(cid:12)(cid:12) c1 R = (cid:8)µ(cid:12)(cid:12) c1 R = (cid:8)µ(cid:12)(cid:12) c1 R = (cid:8)µ(cid:12)(cid:12) c1 Table 7. Degenerations required to prove Theorem 16 Parametrized basis Parametrized index Degeneration T07(∗) → T01 T07(∗) → T04 T08(∗) → T02 T08(∗) → T05 T10(∗) → T06 Et 1 = e1 + e2, 1 = e2, 1 = e1 + e2, 1 = e2, 1 = te1, Et Et Et Et Et 2 = te2 2 = te1 2 = te2 2 = te1 2 = e1 − e2 Et Et Et Et f (t) = 1 + t f (t) = 1 + t−1 f (t) = 2 − t f (t) = 2 − t−1 f (t) = t−1 Separating set R = (cid:8)µ(cid:12)(cid:12) c1 R = (cid:8)µ(cid:12)(cid:12) c1 R = (cid:8)µ(cid:12)(cid:12) c1 12 = 0, c1 21 = 0, c2 21 = 0, c1 22 = 0, c2 12 = 0, c1 21 = 0, c2 21 = 3c1 11 − 2c2 22 = 0, c2 11 = 0, c2 22 = c1 12 + c1 21, c1 11 = c2 22 = 0(cid:9) 22 = 0, c2 12, c1 22 = 0(cid:9) 12 + c2 21(cid:9) Table 8. Non-degenerations required to prove Theorem 16 Non-degeneration T07(∗) 6→ T08(∗) 6→ T02, T05, T08(α), T06, T10(α 6= 1) T01, T04, T07(α 6= 3/2), T06, T10(α 6= 2) T10(∗) 6→ T03 10 0 1 2 4 Figure 1. Graph of primary degenerations of the variety of 2-dimensional terminal algebras. 4. APPENDIX: FIGURES T09 α = 0 α = 1 T01 T02 T07(α) T08(α) T04 T05 α = 1 α = 2 T10(α) T3 k2 T6 Figure 2. Lattice of subsets of the variety of 2-dimensional terminal algebras. 4 3 2 0 O(cid:0)T07(∗)(cid:1) O(cid:0)T07(0)(cid:1) O(cid:0)T01(cid:1) O(cid:0)T04(cid:1) O(cid:0)T10(1)(cid:1) O(cid:0)T09(cid:1) O(cid:0)T07(3/2)(cid:1) O(cid:0)T03(cid:1) O(cid:0)T08(∗)(cid:1) O(cid:0)T05(cid:1) O(cid:0)T02(cid:1) O(cid:0)T08(1)(cid:1) O(cid:0)T10(∗)(cid:1) k2 O(cid:0)T10(2)(cid:1) O(cid:0)T06(cid:1) 11 REFERENCES [1] Ancochea Berm ´udez J.M., Campoamor-Stursberg R., Garc´ıa Vergnolle L., S´anchez Hern´andez J., Contractions d´algebres de Jordan en dimen- sion 2, Journal of Algebra, 319 (2008), 6, 2395 -- 2409. [2] Benes T., Burde D., Degenerations of pre-Lie algebras, Journal of Math. Phys., 50 (2009), 11, 112102, 9 pp. [3] Burde D., Steinhoff C., Classification of orbit closures of 4-dimensional complex Lie algebras, Journal of Algebra 214 (1999), 729 -- 739. [4] Calder´on A., Fern´andez Ouaridi A., Kaygorodov I., Classification of bilinear maps with radical of codimension 2, arXiv:1806.07009 [5] Cantarini N., Kac V., Classification of linearly compact simple rigid superalgebras, International Mathematics Research Notices, 17 (2010), 3341 -- 3393. [6] Grunewald F., O'Halloran J., Varieties of nilpotent Lie algebras of dimension less than six, Journal of Algebra, 112 (1988), 315 -- 325. [7] Ismailov N., Kaygorodov I., Volkov Yu., The geometric classification of Leibniz algebras, International Journal of Mathematics, 29 (2018), 5, 1850035. [8] Ismailov N., Kaygorodov I., Volkov Yu., Degenerations of Leibniz and anticommutative algebras, arXiv:1808.00907 [9] Kantor I., Certain generalizations of Jordan algebras [in Russian], Trudy Sem. Vektor. Tenzor. Anal., 16 (1972), 407 -- 499. [10] Kantor I., On an extension of a class of Jordan algebras, Algebra and Logic, 28 (1989), 2, 117 -- 121. [11] Kantor I., Some problems in L-functor theory [in Russian], Trudy Inst. Mat., 16 (1989), Issled. po Teor. Kolets i Algebr, 54 -- 75. [12] Kantor I., A universal conservative algebra, Siberian Mathematical Journal, 31 (1990), 3, 388 -- 395. [13] Kaygorodov I., On the Kantor product, Journal of Algebra and Its Applications, 16 (2017), 5, 1750167. [14] Kaygorodov I., Lopatin A., Popov Yu., Conservative algebras of 2-dimensional algebras, Linear Algebra and its Applications, 486 (2015), 255 -- 274. [15] Kaygorodov I., Popov Yu., Pozhidaev A., The universal conservative superalgebra, preprint [16] Kaygorodov I., Popov Yu., Pozhidaev A., Volkov Yu., Degenerations of Zinbiel and nilpotent Leibniz algebras, Linear and Multilinear algebra, 66 (2018), 4, 704 -- 716. [17] Kaygorodov I., Popov Yu., Volkov Yu., Degenerations of binary-Lie and nilpotent Malcev algebras, Communications in Algebra, 46 (2018), 11, 4929-4941. [18] Kaygorodov I., Volkov Yu., Conservative algebras of 2-dimensional algebras. II, Communications in Algebra, 45 (2017), 8, 3413 -- 3421. [19] Kaygorodov I., Volkov Yu., The variety of 2-dimensional algebras over an algebraically closed field, arXiv:1701.08233. [20] Petersson H., The classification of two-dimensional nonassociative algebras, Results Math., 37 (2000), 1-2, 120 -- 154.
1705.03059	1	1705	2017-05-08T19:43:07	Automatic classification of automorphisms of lower-dimensional Lie algebras	[ "math.RA" ]	We implement two algorithms in MATHEMATICA for classifying automorphisms of lower-dimensional non-commutative Lie algebras. The first algorithm is a brute-force approach whereas the second is an evolutionary strategy. These algorithms are delivered as the MATHEMATICA package cwsAutoClass. In order to facilitate the application of this package to symmetry Lie algebras of differential equations, we also provide a package, cwsLieSymTools, for manipulating finite-dimensional Lie algebras of vector fields. In particular, this package allows the computations of Lie brackets, structure constants, and the visualization of commutator tables. Several examples are provided to illustrate the pertinence of our approach.	math.RA	math	Automatic classification of automorphisms of lower-dimensional Lie algebras 1Department of Mathematics and Statistical Sciences, C. Wafo Soh1 , 2 Jackson State University JSU Box 17610, 1400 JR Lynch Street, Jackson, MS 39217, USA 2DST-NRF Centre of Excellence in Mathematical and Statistical Sciences, School of Computer Science and Applied Mathematics, University of the Witwatersrand, Johannesburg, Wits 2050, South Africa July 31, 2018 Abstract We implement two algorithms in MATHEMATICA for classifying au- tomorphisms of lower-dimensional non-commutative Lie algebras. The first algorithm is a brute-force approach whereas the second is an evolu- tionary strategy. These algorithms are delivered as the MATHEMATICA package cwsAutoClass. In order to facilitate the application of this pack- age to symmetry Lie algebras of differential equations, we also provide a package, cwsLieSymTools, for manipulating finite-dimensional Lie alge- bras of vector fields. In particular, this package allows the computations of Lie brackets, structure constants, and the visualization of commutator tables. Several examples are provided to illustrate the pertinence of our approach. 1 Motivation A symmetry of an object is a transformation that does not affect it. In several applications, it is desirable to compute all the symmetries of the object of in- terest. However, this is a difficult and sometimes an intractable problem. In the case of differential equations, Sophus Lie [1] discovered that the knowledge of its continuous symmetries unravels an algorithmic path for its integration by quadratures. Lie derived an algorithm for the computation of these symme- tries, which in most cases, involved solving an overdetermined (i.e. there are more equations than unknowns) system of linear partial differential equations 1 (PDEs).Until recently, there was not an algorithm for computing all the discrete symmetries of differential equations except for the direct method which leads to equations as difficult to solve as the initial equation itself. That is, the lineariza- tion afforded by continuous symmetries is lost in the search for discrete symme- tries. Beside, the direct method for the search of discrete symmetries does not guarantee that one has found all the discrete symmetries unless one solves the nonlinear determining equations for them. Hydon [2] discovered that, when the continuous symmetry Lie algebra of a differential equation is non-Abelian, in the symmetry group, discrete and continuous symmetries are entangled. However, if one wants to compute all its discrete symmetries, one must first disentangle its two types of symmetries through the classification of automorphisms of its continuous symmetry Lie algebra. This disentanglement produces conditional determining equations for discrete symmetries that are first-order quasilinear PDEs. This conditional determining system is overterdermined and in principle can be solved using Lagrange-Charpit method of characteristics. Once, its solu- tions are obtained, the discrete symmetries are found by constraining these so- lutions to leave invariant the underlying differential equation. This last stage is a mere application of the chain rule for differentiation. However, there are three main roadblocks in the implementation of Hydon's algorithm for the computa- tion of discrete symmetries of a differential equation admitting a non-Abelian symmetry Lie algebras. They are: (1) the solution of a large overtedermined system of quadratic equations, (2) the classification of the automorphisms of the symmetry Lie algebra, and (3) the solution of the conditional determining system. In this work, we focus primarily on the classification of automorphisms of lower-dimensional non-Abelian Lie algebras. Specifically, We implement in MATHEMATICA a brute-force approach together with an evolutionary strat- egy for classifying the automorphisms of a non-commutative finite-dimensional Lie algebra. We have structured this paper as follows. There are four sections including In Section 2, we state the problem to be solved while fix- this introduction. ing notations. Section 3 deals with the implementation of two automorphisms classification algorithms viz. brute-force and genetic strategies. Section 4 is concerned with the test of our implementation. There, we consider several ex- amples that illustrate diverse aspects of our code. The last section i.e. Section 5, summarizes our work and engages in some discussions about it. 2 Statement of the problem Our goal in this section is to present the algorithm for classifying the automor- phisms of a finite-dimensional non-Abelian Lie algebra. We start with some preliminaries on Lie algebras [3]. A Lie algebra L over a field F ∈ {R, C} is a vector space which is equipped with a bilinear operation [., .] (Lie bracket) which enjoys the following two prop- erties: (1) for all x, y ∈ L, [x, y] = −[y, x] (antisymmetry), and (2) for all x, y, and z, [[x, y], z] + [[y, z], x] + [[z, x], y] = 0 (Jacobi's identity). The dimension 2 of a Lie algebra is its dimension when it is treated as a vector space. In the sequel, we shall be dealing solely with finite-dimensional Lie algebras. So sup- pose that L is such Lie algebra with basis {e1, e2, . . . , en}. For all i, j = 1 : n, [ei, ej] ∈ L. So, for all i, j = 1 : n, [ei, ej] = Pn ijek for some constants ck ij ∈ F, k = 1 : n. The ck ij 's are known as the structure constants of L. It can be verified that, thanks to the antisymmetry of the Lie bracket and Jacobi's iden- tity, the structure constants have the following properties: for i, j, k, l = 1 : n, k=1 ck ck ij = −ck ji and n X (cm ij cℓ mk + cm jkcℓ mi + cm kicℓ mj) = 0. (1) m=1 A linear map, f , from the Lie algebra L to itself is called and automorphism if it is a bijection and for all x, y ∈ L, f ([x, y]) = [f (x), f (y)]. If we designate by B = [bj i ] the matrix of f , where i and j are respectively the row and column numbers, then the entries of B are constrained by the two conditions: det(B) n 6= 0, cℓ ijbk ℓ = X ℓ=1 n X ℓ,m=1 ck ℓmbℓ i bm j , i < j = 1 : n. k = 1 : n, (2) (3) Conversely, it can be shown that if a matrix B satisfies the constraints (2)-(3), then it induces an automorphism of the Lie algebra L. In the system of equations (3), there are n2(n − 1)/2 equations for n2 unknown. As n becomes large, it can be a daunting task to solve it. Beside, one may encounter an explosion of cases which also erode considerably our computational budget. Once the system (3) is solved, one must filter the solution set according to the constraint (2). For all j = 1 : n, the mapping x 7→ [x, ej] is a derivation of L whose matrix C(j) is such that its entry at position (i, k) is (C(j))k ij. Each such derivation with matrix C(j), j = 1 : n, induces a one-parameter family of automorphisms of L with matrices {exp(ǫC(j)) : ǫ ∈ F}. Let us introduce the notation A(j, ǫ) = exp(ǫC(j)) for latter convenience. Also, we shall denote the set of all the matrices B satisfying Eqs. (2)-(3) by Aut(L). Thus, for all j = 1 : n and ǫ ∈ F, A(j, ǫ) ∈ Aut(L). i = ck We define on Aut(L) a relation ∼d by: B1 ∼d B2 if and only if there are (ǫ1, ǫ2, . . . , ǫn) ∈ Fn and a permutation σ of {1, 2, . . . , n} such that B1 = A(σ(1), ǫ1) A(σ(2), ǫ2) . . . A(σ(n), ǫn) B2. (4) It can be shown that the relation ∼d is reflexive and symmetric i.e. it is a dependency relation. Our main goal is to determine the dependency classes modulo ∼d. That is, we want to find all the traces of the dependency ∼d. For a given trace, we wish to select its sparsest member as representative. Note that there are three types of A(j, ǫ)′s: (1) those that are equal to the identity matrix In, (2) those that are diagonal matrices distinct from the identity, and (3) those that are non-diagonal. The A(j, ǫ)'s that are identity matrices come from the C(j)'s that are zero matrices. They are generated by ej's that commute with 3 the e′ js that belong to the center of L. They do all the elements of L i.e. not affect the definition of ∼d. So they may be disregarded. Thus, we are left with the nontrivial diagonal and nondiagonal A(j, ǫ)'s. Since a diagonal matrix commute with any matrix of the appropriate size, in the definition of ∼d, after discarding the identities, we may reposition all the diagonal matrices in front and their order does not matter. We may assume without lost of generality that the diagonal A(j, ǫ)'s are numbered from 1 to s, and the nondiagonal ones are numbered from s + 1 to p, with p ≤ n. Now, we may rephrase the definition of ∼d in the following way: B1 ∼d B2 if and only if there are (ǫ1, ǫ2, . . . , ǫp) ∈ Fn and a permutation τ of {s + 1, s + 2, . . . , p} such that B1 = A(1, ǫ1) . . . A(s, ǫs)A(τ (s + 1), ǫs+1) . . . A(τ (p), ǫp) B2. (5) Given B2 ∈ Aut(L), we are after its sparsest dependents B1 which contain less parameters (i.e unspecified entries). We accomplish this by selecting an appro- priate permutation τ and by choosing ǫs+1 to ǫp such that as much as possible entries of B3 = A(τ (s + 1), ǫs+1) . . . A(τ (p), ǫp) B2 are equal to zero. Then, we pick ǫ1 to ǫs such that some rows or columns of A(1, ǫ1) . . . A(s, ǫs)B3 = B1 are appropriately scaled. For a fixed B ∈ Aut(L) and a permutation τ of {s + 1, s + 2, . . . , p}, we may end up with several sparse dependents. We denote by τ (B) the set of such dependents. We define the fitness of τ (B) as the geometric mean of the sparsity of its elements. As τ ranges over permutations of {s + 1, s + 2, . . . , p}, we select the permutation which provides the fittest set of dependents of B. In the scheme for finding dependency classes modulo ∼d that we have just described, there are two main roadblocks: (1) the solution of the system (2)-(3), and (2) the search of optimal representative of dependency classes. Indeed the complexity of both these problems is exponential in the dimension, n, of L. Thus as n increases, it is imperative to adopt strategies for taming this issue. For the first problem, we may partially solve the system (2)-(3) by relying upon the structure of its equation: We may start by solving linear and quadratic equations with fewer terms. As for the second problem, we adopt an evolutionary strategy at the expenses of having sub-optimal representatives of traces. In the remainder of this paper, we work through our implementation of the computation of dependency classes modulo ∼d, test our implementation by treating few examples, and discuss the virtues and limitations of our implemen- tation while suggesting possible future improvements. 3 Implementation This section reifies the algorithm described in the previous section using MATH- EMATICA. We assume that the reader is familiar with MATHEMATICA's syntax. We adopt mostly a functional programming approach. In all the code snippets provided, build-in functions start with a capital letters whereas our variables start with lower cases. We will not provide helper functions since they can be consulted in the accompanying source code. 4 3.1 Solution of the system (2)-(3) The function that solves the system (2) -(3) is named simplifiedB. It consumes five inputs in the following order: (1) the dimension of the Lie algebra L, dim, (2) a function c that allows the calculation of structure, (3) a symbol b that is used to name the entries of the matrix B, (4) the field, dom, of the Lie algebra L, and (4) a list of exigences, constraints, on the entries of the matrix B. It is coded using some helper functions that we do not provide. s i m p l i f i e d B[ dim_ , c_ , b_ , dom_ : Complexes , c o n s t r a i n t s _ :{}]:= s i m p l i f i e d B[ dim ,c ,b , dom , c o n s t r a i n t s ]= Module [{ sol = s o l v e S y s[ dim ,c , c o n s t r a i n t s ][ b ] , bb = s y m b o l i c B[ dim , b ] , bs } , bs = Map [ bb /.#& , sol ]; If [ dom === Reals , Select [ bs , r e a l M a t r i x Q] , bs ]]// S i m p l i f y; 3.2 Determination of a single dependency class Given a solution b of the system (2)-(3) and a permutation sigma of the nondi- agonal A(j, ǫ)'s of the Lie algebra L, the function reducedB finds all its sparsest dependents. The inputs c, dim and dom are as before. The main challenge in implementing the function reducedB is to be able deal with all the cases. Indeed as we zero an entry of B by multiplying it by an A(j, ǫ), we may end up with several possible ǫ's which must be treated separately. The same remark applies when we scale a row or column of B using a non-trivial diagonal A(j, ǫ). r e d u c e B[ c_ , dim_ , perm_ , dom_ : Complexes , c o n s t r a i n t s _ :{}][ b_ ] := r e d u c e B[c , dim , perm , dom , c o n s t r a i n t s ][ b ]= Module [{ f , g , auto = g a t h e r B y D i a g o n a l [ a l l S t r u c t M a t r i x [ dim , c ] ] ,k , sb ={ b }} , $ A s s u m p t i o n s = Fold [ And , F l a t t e n[ Table [ E l e m e n t[ b [i , j ] , dom ] ,{ i ,1 , dim } , {j ,1 , dim }]]]; auto [[1]] = Select [ auto [[1]] , auto [[2]] = P e r m u t e[ auto [[2]] , perm ]; f = J o i n @ @ M a p[ f i n d E p s A n d R e d u c e B [k ,0 , dom , c o n s t r a i n t s ] , # != z e r o M a t r i x[ dim ]&]; m y O u t e r[ Dot , #1 , M a t r i x E x p[ k #2] ]]&; g = J o i n @ @ M a p[ f i n d E p s A n d R e d u c e B [k ,1 , dom , c o n s t r a i n t s ] , m y O u t e r[ Dot , #1 , M a t r i x E x p [ k # 2 ] ] ] & ; { Fold [ g , Fold [f , sb , auto [[2]]] , auto [[1]]] , perm }]// S i m p l i f y 1 2 3 1 2 3 4 5 6 7 8 9 3.3 Determining all the dependency classes Now that we know how to compute the dependents of a solution B of the system (2)-(3) for a given permutation of the non-diagonal A(j, ǫ)'s, we can select the 5 fittest when this permutation ranges over the set of all such permutations. As we mentioned earlier, the fitness of a trace is taken as the geometric mean of the sparsities of its members. 3.3.1 The brute-force approach When in doubt, use brute force. Ken Thompson The function that compute the optimal set of dependents for all the solutions of (2)-(3) is called autoClassificationBruteForce. Its inputs are similar to those of previous functions except for the new input name which is a string that will be used as radical for renaming some variable entries of dependents. The outputs is a pair whose first entry is the set of representatives of dependency classes whereas the second entry comprises restrictions encountered during com- putations. We employed the latter output mainly for backtracking calculations. a u t o C l a s s i f i c a t i o n B r u t e F o r c e [ dim_ ,c_ , dom_ : Complexes , c o n s t r a i n t s _ :{}][ b_ , name_ :\[ Alpha ]]:= Module [{ s , cond , f } , Not [ Normal [ Det [ # ] ] = = = 0 ] & ; f = s = Map [ a u t o C l a s s i f i c a t i o n B r u t e F o r c e O n e B [ dim ,c , dom , c o n s t r a i n t s] , s i m p l i f i e d B [ dim ,c ,b , dom , c o n s t r a i n t s ]]; cond = Fold [ Or , D e l e t e D u p l i c a t e s @ M a p [#[[2]]& , s ]]; s = D e l e t e D u p l i c a t e s [ J o i n @ @ M a p[ r e n a m e V a r i a b l e [ name , F l a t t e n[ s y m b o l i c B[ dim , b ]] ,# , dom ]& , J o i n @ @ M a p [ First , s ]]]; { Select [s , f ] , cond }]; 1 2 3 4 5 6 3.3.2 An evolutionary approach I have called this principle, by which each slight variation, if useful, is preserved, by the term of Natural Selection. Charles Darwin In the function autoClassificationBruteForce, as the number of non- diagonal inner automorphisms increases, our computational budget is quickly depleted. In oder to palliate this situation, we adopt an evolutionary strat- egy through the function autoClassificationGen. Besides the previous type of inputs, it consumes the following ones: sigma, p, popSize, numGen and cond. The argument sigma is the percentage of the optimal fitness we want 6 to achieve, and p is the mutation probability. The initial population size is popSize whereas numGen is the largest number of generations one is willing to go through before stopping if the desired fitness is not realized. In the code below, the function aiReduceB does the heavy lifting of our genetic strategy which encompasses our mating, mutation and selection schemes. For a quick intuition into our evolutionary scheme, consider two parents par1 = (dep1, µ1) and par2 = (dep2, µ2) selected at random with probabilities proportional to their respective fitnesses, where the depi's are lists of dependents and the µi's are the corresponding permutations of non-diagonal inner automorphisms (the so-called A(j, ǫ)'s). In each individual's definition pair, we shall refer to the first entry as its phenotype and the second entry as its DNA. Theses biological analogies are self-explanatory. Note that, given an individual begotten from a solution of (2)- (3) and its DNA, we can always recover its phenotype through computations. Now, the parents par1 and par2 produce six possible types of offspring with DNAs µ−1 2 . After possible mutation of children, the fittest amongst parents and offspring earns the right to belong to the next generation. We accomplish mutation by simply swapping two randomly chosen entries of the underlying DNA. In MATHEMATICA syn- tax, a permutation is represented by a list. For instance a permutation of 1, 2, 3 may be represented as µ = {3, 1, 2}. It means that µ(1) = 3, µ(2) = 1, and µ(2) = 2. Thus, µ′ = {2, 1, 3} is possible mutation of µ which is produce by swapping the nucleotides 3 and 2. 2 , µ1 ◦ µ2, µ2 ◦ µ1, µ1 ◦ µ2 ◦ µ−1 1 , and µ2 ◦ µ1 ◦ µ−1 1 , µ−1 The output of the function autoClassificationGen is formatted according to the boolean input cond. It is a list of three or two elements according to whether cond is true or false. When cond is true, the output comprises the list of dependence classes, the constraints encounter during computations as well as the DNAs of dependency classes. In the even cond is false, the second entry in the output list is omitted. a u t o C l a s s i f i c a t i o n G e n [ sigma_ , p_ , c_ , dim_ , p o p S i z e _ :5 , n u m G e n _:1 , dom_ : Complexes , cond_ : False , c o n s t r a i n t s _ :{}][ b_ , name_ :\[ Alpha ]]:= Module [{ s , perm , cd , f } , f = Not [ Normal [ Det [ # ] ] = = = 0]&; s = Map [ a i R e d u c e O n e B [ sigma ,p ,c , dim , popSize , numGen , dom s i m p l i f i e d B[ dim ,c ,b , dom , c o n s t r a i n t s , c o n s t r a i n t s ] , ]]; = J o i n @ @ M a p[ perm = D e l e t e D u p l i c a t e s [ Map [#[[2]]& , s ]]; s s = Map [ c o l l e c t C o n di tio ns , s ]; {s , cd } = unZip [ s ]; s = D e l e t e D u p l i c a t e s [ J o i n @ @ M a p[ r e n a m e V a r i a b l e [ name , Join [#[[1]]]& , s ]; F l a t t e n[ s y m b o l i c B[ dim , b ]] ,# , dom ]& , s ]]; cd = Fold [ Or , False , Map [ Simplify , D e l e t e D u p l i c a t e s [ F l a t t e n[ Map [ Fold [ And , True , #]& , cd ]]]]]; 1 2 3 4 5 6 7 8 9 10 If [ cond , { Select [s , f ] , cd , perm } , { Select [s , f ] , perm 7 }]]; 4 Tests Here we test our implementation of the classification of automorphisms of non- commutative finite-dimensional Lie algebras. We have encapsulated our algo- rithms in a MATHEMATICA package called cwsAutoClass. Additionally, we provide a package cwsLieSymTools which facilitates some computations per- taining to finite-dimensional Lie algebra represented in terms of vector fields. In particular it allows the calculations of the structure constants and the gen- eration of the commutator table given a basis of the underlying Lie algebra. The examples we shall treat are done in MATHEMATICA version 11 run on a DELL INSPIRON laptop with WINDOWS 10 operating system and the follow- ing additional specifications: INTEL CORE i3-3227U @ 1.90 GHz processor, 3.96 GB of usable RAM and a 64-bit operating system. 4.1 Dependency classes of 3D non-Abelian Lie Algebras There are ten non-commutative three-dimensional Lie algebras [4]. We provide in this section the commands for computing their dependency classes using the package cwsAutoClass. In first line of the code provided below, replace the comment with the ap- propriate information before compilation. The instruction of that line loads the package cwsAutoClass. Get [(* Put the l o c a t i o n of the file c w s A u t o C l a s s . m here e . g . " C :\\ Users \\ C e l e s t i n \\ D e s k t o p \\ T r i p _ t o _ S A \\ c w s A u t o C l a s s . m " ) ]; c1 [1 , 2 , 2] = 1; c1 [2 , 1 , 2] = -1; c1 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c1 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Timing c2 [2 , 3 , 1] = 1; c2 [3 , 2 , 1] = -1; c2 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c2 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Timing c3 [1 , 3 , 1] = 1; c3 [3 , 1 , 1] = -1; c3 [2 , 3 , 1] = 1; c3 [3 , 2 , 1] = -1; 1 2 3 4 5 6 7 8 9 10 11 12 13 8 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 c3 [2 , 3 , 2] = 1; c3 [3 , 2 , 1] = -1; c3 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c3 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Timing c4 [1 , 3 , 1] = 1; c4 [3 , 1 , 1] = -1; c4 [2 , 3 , 2] = 1; c4 [3 , 2 , 2] = -1; c4 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c4 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Timing c5 [1 , 3 , 1] = 1; c5 [3 , 1 , 1] = -1; c5 [2 , 3 , 2] = -1; c5 [3 , 2 , 2] = 1; c5 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c5 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Timing Clear [ a ]; $ A s s u m p t i o n s = ( -1 < a < 1) && a != 0; c6 [1 , 3 , 1] = 1; c6 [3 , 1 , 1] = -1; c6 [2 , 3 , 2] = a ; c6 [3 , 2 , 2] = -a ; c6 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c6 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Timing c7 [1 , 3 , 2] = -1; c7 [3 , 1 , 2] = 1; c7 [2 , 3 , 1] = 1; c7 [3 , 2 , 1] = -1; c7 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c7 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Timing Clear [ a ]; $ A s s u m p t i o n s = a > 0; c8 [1 , 3 , 1] = a ; c8 [3 , 1 , 1] = -a ; c8 [1 , 3 , 2] = -1; c8 [3 , 1 , 2] = 1; c8 [2 , 3 , 1] = 1; c8 [3 , 2 , 1] = -1; c8 [2 , 3 , 2] = a ; c8 [3 , 2 , 2] = -a ; c8 [ i_ , j_ , k_ ] := 0; 9 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 1 2 3 4 5 6 7 8 9 Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c8 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Timing c9 [1 , 2 , 1] = 1; c9 [2 , 1 , 1] = -1; c9 [3 , 2 , 3] = -1; c9 [3 , 1 , 2] = 2; c9 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c9 , Reals ][ b , \[ Theta c9 [2 , 3 , 3] = 1; c9 [1 , 3 , 2] = -2; ] ] [ [ 1 ] ] ] // Quiet // Timing c10 [1 , 2 , 3] = 1; c10 [2 , 1 , 3] = -1; c10 [3 , 1 , 2] = 1; c10 [1 , 3 , 2] = -1; c10 [2 , 3 , 1] = 1; c10 [3 , 2 , 1] = -1; c10 [ i_ , j_ , k_ ] := 0; lst = Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [3 , c10 , Reals ][ b , \[ Theta ] ] [ [ 1 ] ] ] // Quiet // Normal // Timing 4.2 Dependency classes of some 4D non-Abelian Lie alge- bras The code provided below computes the dependency classes of the non-decomposable non-commutative Lie algebras A4,1, A4,4, A1 4,9, and A4,12 [4]. Get [( Put the l o c a t i o n of the file c w s A u t o C l a s s . m here e . g . " C :\\ Users \\ C e l e s t i n \\ D e s k t o p \\ T r i p _ t o _ S A \\ c w s A u t o C l a s s . m " ) ]; c1 [2 , 4 , 1] = 1; c1 [4 , 2 , 1] = -1; c1 [3 , 4 , 2] = 1; c1 [4 , 3 , 2] = -1; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [4 , c1 , Reals ][ b , \[ Mu c1 [ i_ , j_ , k_ ] := 0; ] ] [ [ 1 ] ] ] // Quiet // Timing c2 [1 , 4 , 1] = 1; c2 [4 , 1 , 1] = -1; c2 [2 , 4 , 1] = 1; c2 [4 , 2 , 1] = -1; 10 c2 [2 , 4 , 2] = 1; c2 [4 , 2 , 2] = -1; c2 [3 , 4 , 1] = 1; c2 [4 , 3 , 1] = -1; 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 1 2 3 4 5 c2 [3 , 4 , 3] = 1; c2 [4 , 3 , 3] = -1; c2 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [4 , c2 , Reals ][ b , \[ Mu ] ] [ [ 1 ] ] ] // Quiet // Timing c3 [2 , 3 , 1] = 1; c3 [3 , 2 , 1] = -1; c3 [1 , 4 , 1] = 2; c3 [4 , 1 , 1] = -2; c3 [2 , 4 , 2] = 1; c3 [4 , 2 , 2] = -1; c3 [3 , 4 , 3] = 1; c3 [4 , 3 , 3] = -1; c3 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [4 , c3 , Reals ][ b , \[ Mu ] ] [ [ 1 ] ] ] // Normal // Quiet // Timing c4 [1 , 3 , 1] = 1; c4 [3 , 1 , 1] = -1; c4 [2 , 3 , 2] = 1; c4 [3 , 2 , 1] = -1; c4 [1 , 4 , 2] = -1; c4 [4 , 1 , 2] = 1; c4 [2 , 4 , 1] = 1; c4 [2 , 4 , 1] = -1; c4 [ i_ , j_ , k_ ] := 0; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [4 , c4 , Reals ][ b , \[ Mu ] ] [ [ 1 ] ] ] // Normal // Quiet // Timing 4.3 Dependency classes of symmetry Lie algebras Here, we treat the dependency classes of symmetry Lie algebras of various PDEs. We shall employ the package cwsLieSymTools to facilitate the computation of structure constants and the visualization of commutator tables. We consider in turn the symmetry Lie algebras of the spherical Burgers [5] , Harry-Dym [6], and Black-Scholes [7, 8] equations. Get [( The l o c a t i o n of c w s A u t o C l a s s goes here e . g . " C :\\ Users \\ C e l e s t i n \\ D e s k t o p \\ T r i p _ t o _ S A \\ c w s A u t o C l a s s . m ") ]; Get [( The l o c a t i o n of c w s L i e S y m T o o l s . m goes here e . g . " C :\\ Users \\ C e l e s t i n \\ D e s k t o p \\ T r i p _ t o _ S A \\ c w s L i e S y m T o o l s . m " ) ]; ( S p h e r i c a l Burgers ' e q u a t i o n ) vars = {t , x , u }; X1 = { -2t , -x , u }; X2 = {0 , Log [ t ] , 1/ t }; X3 = {0 , 1 , 0}; l i s t S y m = { X1 , X2 , X3 }; p a r a m L i s t = {}; dim = 11 3; 6 7 8 9 10 11 12 c o m m u t a t o r T a b l e [ vars , listSym , X , paramList , Reals ] // Quiet c1 = s t r u c t u r e C o n s t a n t [ vars , listSym , paramList , Reals ]; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [ dim , c1 , Reals ][b , \[ Alpha ] ] [ [ 1 ] ] ] // Quiet // Timing Map [ MatrixForm , a u t o C l a s s i f i c a t i o n G e n [0.99 , 0.0001 , c1 , dim , 150 , 20 , Reals , True ][ 13 b , \[ Alpha ] ] [ [ 1 ] ] ] // Quiet // Timing 14 15 16 (* Harry - Dym e q u a t i o n ) vars = {t , x , u }; H1 = {0 , 1 , 0}; H2 = {0 , x , u }; H3 = {0 , x ^2 , 17 2 x * u }; H4 = {1 , 0 , 0}; H5 = {t , 0 , -u /3}; l i s t S y m = { H1 , H2 , H3 , 18 H4 , H5 } ; p a r a m L i s t = {}; dim 19 c o m m u t a t o r T a b l e [ vars , listSym , X , paramList , Reals ] // = 5; 20 21 22 23 24 25 Quiet c2 = s t r u c t u r e C o n s t a n t [ vars , listSym , paramList , Reals ]; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [ dim , c2 , Reals ][b , \[ Alpha ] ] [ [ 1 ] ] ] // Quiet // Timing Map [ MatrixForm , a u t o C l a s s i f i c a t i o n G e n [0.3 , 0.0001 , c2 , dim , 150 , 20 , Reals , True ][ 26 b , \[ Alpha ] ] [ [ 1 ] ] ] // Quiet // Timing 27 28 29 (* Black - Sholes e q u a t i o n ) Clear @@ {A , d , c }; vars = {t , x , u }; Y1 = {1 , 0 , 0} ; Y2 = {0 , x , 30 31 32 0} ; Y3 = {2 t , ( Log [ x ] + d * t ) x , 2 c * t * u } ; Y4 = {0 , A ^2* t x , ( Log [ x ] - d t ) * u } ; Y5 = {2* A ^2* t ^2 , 2* A ^2* t * x * Log [ x ] , (( Log [ x ] - d * t ) ^2 + 2* A ^2* c * t ^2 - A * t ^2) * u }; Y6 = {0 , 0 , 33 34 u }; X1 = (1/ A ^2) ( Y1 + d Y2 + 35 c * Y6 ) ; X2 = Y2 ; X3 = Y3 ; X4 = Y4 ; X5 = (1/2) Y5 ; X6 = Y6 ; dim = 6; 36 l i s t S y m = { X1 , X2 , X3 , X4 , X5 , X6 }; p a r a m L i s t = {A , d , 12 c }; 37 38 39 40 41 42 43 44 c o m m u t a t o r T a b l e [ vars , listSym , X , paramList , Reals ] // Quiet c3 = s t r u c t u r e C o n s t a n t [ vars , listSym , paramList , Reals ]; Map [ MatrixForm , a u t o C l a s s i f i c a t i o n B r u t e F o r c e [ dim , c3 , Reals ][b , \[ Alpha ] ] [ [ 1 ] ] ] // Quiet // Timing Map [ MatrixForm , a u t o C l a s s i f i c a t i o n G e n [0.99 , 0.0001 , c3 , dim , 150 , 20 , Reals , True ][b , \[ Alpha ] ] [ [ 1 ] ] ] // Normal // Quiet // Timing 5 Conclusion and discussions We have implemented a MATHEMATICA package, cwsAutoClass, that allows the automatic classification of automorphisms of non-Abelian lower-dimensional Lie algebras. It includes two functions autoClassificationBruteForce and autoClassificationGen which respectively implement brute force and evolu- tionary strategies. We demonstrated this package by considering several ex- amples. Additionally, we have showcased a package, cwsLieSymTools which facilitates the calculation of structure constants and the visualization of com- mutator tables of finite-dimensional Lie algebras of vector fields. The latter package is particularly useful when discussing the automorphisms of symmetry Lie algebras. The package cwsAutoClass has some bottlenecks stemming from the solu- tion of a large system of quadratic equations and the exploration of a large search spaces to find fittest solutions. We address the second problem through a genetic strategy. However, the first issue remains. A possible way to tackle it consists in partially solving the system (2)-(3) followed by our evolutionary algorithm. Then, solve the remaining equations of the system (2)-(3) for each dependency class of the optimal solution. We shall in the future extend our package cwsAtouClass along these lines. Acknowledgments I gratefully acknowledges partial financial support form the National Research Foundation (NRF) of South Africa. I thank Prof. F M Mahomed for facilitating such funding. I shared portions of the code of the package cwsLieSymTools in 2016 with Ms. Nomsa Ledwaba who was then a Masters student at the University of Cape Town. 13 References [1] S. Lie, " Vorlesungen ber Differentialgleichungen mit bekannten infinitesi- malen Transformationen", BG Teubner, 1891. [2] P. E. Hydon, "How to construct the discrete symmetries of partial differ- ential equations", European Journal of Applied Mathematics 11:05 (2000), 515-527. [3] W. de Graaf , "Algorithms for finite -- dimensional Lie algebras", Doctoral dissertation, Technical University of Eindhoven, 1997. [4] J. Patera and P. Winternitz, "Subalgebras of real three- and four- dimen- sional Lie algebras", Journal of Mathematical Physics 18:7 (1977),1449- 1455. [5] J. Doyle and M. J. Englefield, 1990. "Similarity solutions of a generalized Burgers equation", IMA Journal of Applied Mathematics 44:2 (1990),145- 153. [6] W. Hereman, "Review of symbolic software for the computation of Lie symmetries of differential equations", Euromath. Bull. 1:2 (1994), 45-82. [7] R. K. Gazizov and N. H. Ibragimov, "Lie symmetry analysis of differential equations in finance", Nonlinear Dynamics, 17:4(1998), 387-407. [8] G. Silberberg, 2005. "Discrete symmetries of the Black-Scholes equation", Proceedings of 10th International Conference in MOdern GRoup ANalysis 190 (2005), 197. 14
1505.04396	1	1505	2015-05-17T13:40:17	Coding theory applied to KU-algebras	[ "math.RA" ]	The notion of a KU-valued function on a set is introduced and related properties are investigated. Codes generated by KU-valued functions are established. Moreover, we will provide an algorithm which allows us to find a KU-algebra starting from a given binary block code.	math.RA	math	Coding theory applied to KU-algebras By Samy M.Mostafa a , Bayumy.A.Youssefb , Hussein A. Jad c Abstract The notion of a KU-valued function on a set is introduced and related properties are investigated. Codes generated by KU-valued functions are established. Moreover, we will provide an algorithm which allows us to find a KU-algebra starting from a given binary block code. Keywords. KU-valued function, binary block code of KU-valued functions AMS Classification. 06F35 Corresponding Author : Samy M. Mostafa ( [email protected] ) . ---------------------------------------------------------------------------------------------- 1. Introduction BCK-algebras form an important class of logical algebras introduced by Iseki [5,6,7] and were extensively investigated by several researchers. The class of all BCK-algebras is a quasivariety. Iseki posed an interesting problem (solved by Wronski [13]) whether the class of BCK-algebras is a variety. In connection with this problem, Komori [9] introduced a notion of BCC-algebras and Dudek [1] redefined the notion of BCC-algebras by using a dual form of the ordinary definition in the sense of Komori. Dudek and Zhang [2] introduced a new notion of ideals in BCC-algebras and described connections between such ideals and congruences. C.Prabpayak and U.Leerawat ([11], [12]) introduced a new algebraic structure which is called KU-algebra. They gave the concept of homomorphisms of KU - algebras and investigated some related properties. These algebras form an important class of logical algebras and have many applications to various domains of mathematics, such as, group theory, functional analysis, fuzzy sets theory, probability theory, topology, etc. Coding theory is a very young mathematical topic. It started on the basis of transferring information from one place to another. For instance, suppose we are using electronic devices to transfer information (telephone, television, etc.). Here, information is converted into bits of 1’s and 0’s and sent through a channel, for example a cable or via satellite. Afterwards, the 1’s and 0’s are reconverted into information again. Due to technical problems, one can assume that while the bits are sent through the channel, there is a positive probability p that single bits are being changed. Thus the received bits could be wrong. The idea of coding theory is to give a method of how to convert the information into bits, such that there are no mistakes in the received information, or such that at least some of them are corrected. On this account, encoding and decoding algorithms are used to convert and reconvert these bits properly. One of the recent applications of BCK-algebras was given in the Coding theory [3,8 ,12]. In Coding Theory, a block code is an error- correcting code which encodes data in blocks. In the paper [8], the authors introduced the notion of BCK-valued functions and investigate several properties. Moreover,they established block-codes by using the notion of BCK-valued functions. they show that every finite BCK-algebra determines a block- code constructed a finite binary block-codes associated to a finite BCK-algebra. In [3,12] provided an algorithm which allows to find a BCK-algebra starting from a given binary block code. In [12] the authors presented some new connections between BCK- algebras and binary block codes. In this paper, we apply the code theory to KU- algebras and obtain some interesting results. 1 2. Preliminaries Now, we will recall some known concepts related to KU-algebra from the literature which will be helpful in further study of this article. Definition2.1[10,11] Algebra(X, ∗, 0) of type (2, 0) is said to be a KU -algebra, if it satisfies the following axioms: 1ku ) ( 2ku) x  3ku ) x ) 4ku ) zy  y (  0 zy  ( )  zx  zx  ) y z )( x 0 xz )( ( zx  )] zy  ) y ( ( yx  )  [( ( ( 0 , 0 x  )  ,0    ( (  Example 2.2 Let X = {0, 1, 2, 3, 4} in which * is defined by the following table * 0 1 2 3 4 0 0 0 0 0 0 1 1 0 1 0 0 2 2 2 0 2 0 3 3 3 3 0 0 4 4 4 3 2 0 It is easy to show that X is KU-algebra. In a KU-algebra, the following identities are true : If we put in (ku4) y = x = 0 we get (0 * 0) * [ (0 * z) * (0 * z) ] =0 and it follows that : (Ku5) z * z = 0 , if we put y = 0 in (ku4) , we get (p1) z * (x * z ) = 0 . A subset S of KU-algebra X is called sub-algebra of X if x * y  S, whenever x, y  S. A non empty subset A of a KU-algebra X is called a KU-ideal of X if it satisfies the following conditions: (I1) 0  A, (I2) x * (y * z)A , y  A implies x * z  A , for all x , y , zX . Lemma 2.3 [9 ] In a KU-algebra (X, , 0), the following hold: x ≤ y imply y z ≤ x * z . Lemma 2.4 [10 ] If X is KU-algebra then y * [(y * x) * x] = 0. 2 AA :~ is called a KU-valued function (briefly, KU-function) on A . 3. KU-valued functions In what follows let A and X denote a nonempty set and a KU-algebra respectively, unless otherwise specified. Definition 3.1 A mapping X Definition 3.2 A cut function of A~ , for :~ AAq xAAx Obviously, cut of A~ : Example 3.3 Let A ={x, y, z} and let X = {0, a, b, c, d} is a KU-algebra with the following Cayley table: such that qA~ is the characteristic function of the following subset of A , called a cut subset or a q- :)(~ xAq  Xq  is defined to be a mapping 1)(~) )(~: xAAx  )(~ xA 0 . .  }1,0{     q q 0 ( q * 0 a b c d 0 0 0 0 0 0 a a 0 a 0 0 b b b 0 0 b c c b a 0 b d d a d a 0 The function AA :~ X given by A~  A 0  , A a   x  , A b   y ,  A c x a     z y cb AA d ,      x is a KU-function on A , and its cut subsets are Proposition 3.4 Every KU-function  xAXq , that is 1)(    , q AA :~ X )(~: xA Xx )(~ Xq xA , then Xr  , for Ax  , then )(~ xA Xr ,  r Assume that Proof. For any Ax  . Let  1)(~ xAr xAXr  for 1)(~, Since  This completes the proof.   q r on A is represented by the infimum of the set  inf  )(~ xA )(~, xAXq  q 1 .  1)(~ xAso q q  0 and  Xq . , rq * rei 0 .  ,it follows that q  )(~ xA .  q inf  xAXr  1)(~, .  r 3 Proposition 3.5 Let AA :~ X be a KU-function on A . If * pq 0 for all we get A  . p A q Xqp ,  , Proof. Let Xqp  be such that , , * pq 0 and pAx  , then )(~ xA p 0 Using ( 1ku ) and ( 2ku ) ,we have )KU (      2 ))(~()( pq PxA 0  ))(~( xA q  , and so qAx  . Therefore A  . p A q This completes the proof. AA :~ Proposition 3.6 Let be KU-function on A . Then X )(~ )(~)( xAAyx A yA , 1-   )(~ xA )(~)( xAAxXq   A (~ yA 2-   )(    q 0 ( ( ) A )(~ xA A q Proof. (1) The sufficiency is obvious. Assume that A )(~ xA  A (~ yA ) for all Ayx , .Then (2) The necessity follows from Proposition 3.4. Let be such that A q . If q 0 then qAx  . Since 0 , it follows that )(~ xAAx  , yAzAAz  0)(~)(~,  Xq  )(~)(~ xAxA   and ) A (~ yA Ax  A (~ yA ) * A )(~ xA  *  A )(~ xA  )(~ xA Aor A 0 (~ yA 0)(~)(~,  xAzAAz  ) 0     .Thus A )(~ xA so that )(~ xA A q A )(~ xA .This is a contradiction. AA :~ X 0)(~)(~)(   A (~ yA ) A )(~ xA ) . (  yAxAAyx ,  Corollary 3.7 Let be KU-function on A .Then Proof. Straightforward. AA :~ For a KU-function  ,  XqA q ~ A x A x    : X  , consider the following sets: XqA :~  . q Proposition 3.8 Let (  )(  inf X Y AA :~ be KU-function on A . Then X A  inf( Yqq :   XinY ( ) Y  Proof. Let X inf exists there inf)(~   Ax Yqq xA 0    Yqq inf( ) :    x YqA . the This proof. :    q completes : 4 ) q :  YqA  .  that XinY YqqAx  )(~)( xAYr such )0 inf( r (  : .We have )  AxYr r  )( ) ( Corollary 3.9 Let KU-algebra, then AA :~ X S  X , be KU-function on A , where X is a bounded A inf(  qA  q S .  Sqq :  : ) Corollary 3.10 Let , there exists a infimum of Y such that ( AA :~ X be KU-function on A , assume that for any A X ) , we have Yqp    , A p A q Y  X . be a KU-algebra with the following , ,    A set  ,0 and abe Xlet  yx , dcba , The following example shows that the converse of the corollary 3.10 may not true in general. Example 3.11. Let Cayley table: The function 0 0 0 a 0 b 0 c 0 d 0 AA :~ d d a d a 0 a a 0 a 0 0 c c b a 0 b b b b o 0 b X is a KU-function on A , then    And its cut subsets are ,  y Ab  0A     y x  Aa   x , ,    A a A b Note that Ac A X , but inf  x Ad   ba, does exists in X . given by y x A~ ba     0 ~A aA~ bA~ cA~ dA~ x a 0 1 0 1 1 y b 0 0 1 1 0  yx , , 5 Proposition 3.12 Let  A  Proof. Obviously,  XqAq   XqA    q  x AA :~ X be KU-function on A , then     A XqAq  XqA  .Thus   q  A .For every Ax  , let )(~ xA Xq .Therefore the result is valid. .Then qAx  and hence be KU-function on A , then , , (   Proposition 3.13 Let AA :~ X  AxA Ax )( q q AxAx  q     Proof. Note that for any A ) X 1)(~ xA  q From Proposition 3.7 we get the following   AxA q q AA :~ X Let qpXqp A , (    p  (~ )(~ AA xAXq )  Ax    1)(~ xAq some Let for A q    q    ) q q  1)(~   xAA q be KU-function on A and  be a binary operation on X defined by . This completes the proof. A inf A q  . Then  is clearly an equivalence relation on X. and for Xq  , q ](   qxXx *   0 . AA :~ (~ AA ) q ](  X on A , we have )(~ q p xA 0    )(~ xAAx ]   (~ AA ).   0   q ]( Proposition 3.14 For a KU-function  (  qpXqp ,  Proof. We have p ( ] p   (~ AA ) qp A A  q  )(~) xAAx (   )(~ xAAx p (   (~ AA ( )    p ] q ]( This completes the proof. Example 3.15 Let X  ;  3,2,1 ,......, and define a binary operation  on X as follows  X () a i  a j  , where k  and j i ),( is the least common divisor of i and j . Then  na n a ) k 9 j ji ),( ) is a KU-algebra. Its Cayley table is as follows: 6  aa , i j iaX  ,; ( (  1a 2a 3a 4a 5a 6a 7a 8a 9a 1a 1a 1a 1a 1a 1a 1a 1a 1a 1a 2a 2a 1a 2a 1a 2a 1a 2a 1a 2a 3a 3a 3a 1a 3a 3a 1a 3a 3a 1a 4a 4a 2a 4a 1a 4a 2a 4a 1a 4a 5a 5a 5a 5a 5a 1a 5a 5a 5a 5a 6a 6a 3a 2a 3a 6a 1a 6a 3a 2a 7a 7a 7a 7a 7a 7a 7a 1a 7a 7a 8a 8a 4a 8a 2a 8a 4a 8a 1a 8a 9a 9a 9a 3a 9a 9a 3a 9a 9a 1a Let A ~ A  , , ,  edcba , a c a a 7 b a 6    4 and d a 1 X be a KU-function defined by AA :~ e a 2    . Then * 2 4 3 1 ~ aA ~ aA ~ aA ~ aA ~ aA ~ aA ~ aA ~ aA ~ aA 6 9 8 7 5 a 4a 0 0 0 1 0 0 0 1 0 b 6a 0 0 0 0 0 1 0 0 0 c 7a 0 0 0 0 0 0 1 0 0 d 1a 1 1 1 1 1 1 1 1 1 e 2a 0 1 0 1 0 1 0 1 0 ed  , , ~ A a 4 ~ A a 8  eda  , , ,  ~ Aa 6 edb  , , ,  ~ Aa 7 dc  , .  7 and cut sets of Ã are as follows: ~ A a 1  d ~ A a ~ A a ~ Aa ~ A a      , 3 2 5 9 x Ax  ,  is called equivalence class containing x .  x Ax  , we have )(~ xA inf   , that is be a KU- function on A . For every the least element of the  to which it belongs. ; x  X   ; for any AA :~ yxAy  4. Codes generates by KU-functions Let Lemma 4.1 Let )(~ xA Proof. Straightforward.  Let 3,2,1 AA :~ where Such that Let V x Define an order relation x x  i j xx x ..... 1 n j for  yy 1 Ax  , there corresponds a codeword )(~ iA x , V Ai  y ..... ,....., n X A n 2 2 y follows: V x  c x i V y i and X be a finite KU-algebra. Then every KU-function x on A determines a binary block code V of length n in the following way: To every  , V x xx 1 2 ..... x n  1,0 j  and be two code words belonging to a binary block-codeV . . c on the set of code words belonging to a binary block- code V as y …… (4.1) 2,1 ,...., for n i Example 4.2 Let Let ~ A XA :~ 0   0  cba cba X  X ,  cba be a KU-algebra with the following Cayley table: ,0 , 0 * 0 0 0 a 0 b c 0 c c c c 0 b b a 0 b a a 0 0 a be a KU-function on X given by . Then    0 xA~ 0 ~A 1 aA~ 1 bA~ 1 cA~ 1 b 0 0 1 0 c 0 0 0 1 a 0 1 1 0 8 V   1100 , 1110 1001 , , 1000 . See Figure (1) ) ( X , Generally, we have the following theorem. Theorem 4.3 Every finite KU-algebra X determines a block-code V such that X ), ( is isomorphic to ( V ), ( , cX  . ) n ,......,  ia i ;  Proof. Let XA :~ is the desired code under the order X 3,2,1 be a finite KU-algebra in which 1a is the least element and let X  XqAq  be identify KU-function on X .The decomposition of A~ provides a family ~ which Let Xf :  ;~  XqA q  be a function defined by qf )( V x  c x i V y y i for ~ qA  i 2,1 ,...., n for all Xq  .By lemma 4.1, every  class contains exactly one element .So, f is one to one. Let ~ A y A  (by Proposition 3.5), which means that x xeia . 1 ~  . Therefore f is an isomorphism. A x be such that Xyx , Then xy * A y   . y This completes the proof. Example 4.4 Consider a KU-algebra Let XA :~ X a 1 a 1 a 2 a 2 ~ A     X 9 be a KU-function on X given by  na n ,......, 3,2,1 ;   a 3 a 3 a a 4 4 a 5 a 5 a 6 a 6 a 7 a 7 a 8 a 8 a 9 a 9    9 which is considered in example 3.15. Then 3 8 4 1 2 * ~ aA ~ aA ~ aA ~ aA ~ aA ~ aA ~ aA ~ aA ~ aA 9 6 5 7 1a 1 1 1 1 1 1 1 1 1 2a 0 1 0 1 0 1 0 1 0 3a 0 0 1 0 0 1 0 0 1 4a 0 0 0 1 0 0 0 1 0 5a 0 0 0 0 1 0 0 0 0 6a 0 0 0 0 0 1 0 0 0 7a 0 0 0 0 0 0 1 0 0 8a 0 0 0 0 0 0 0 1 0 9a 0 0 0 0 0 0 0 0 1 Thus V= {100000000, 110000000, 101000000, 110100000, 100010000, 11100100,100000100, 110100010, 101000001}. See Figure (2) ( ) ) X , ( V , References [1] W. A. Dudek, The number of subalgebras of finite BCC-algebras, Bull. Inst. Math. Acad. Sinica, 20 (1992), 129-136. [2] W. A. Dudek and X. H. Zhang, On ideals and congruences in BCC-algebras, Czechoslovak Math. J., 48(123) (1998), 21-29. 10 [ 3 ]Cristina FLAUT, BCK-algebras arising from block codes ,arXiv:1408.3598v3 [4] S. M. Hong, Y. B. Jun and M. A.Ozturk, Generalizations of BCK-algebras, Sci. Math. Japo. Online, 8(2003), 549–557 [5] Y.Imai and Iseki K: On axiom systems of Propositional calculi, XIV, Proc. Japan Acad. Ser A, Math Sci., 42(1966),19-22. [6] k.Iseki: An algebra related with a propositional calculi, Proc. Japan Acad. Ser A Math. Sci., 42 (1966), 26-29. [7] K Iseki and Tanaka S: An introduction to theory of BCK-algebras, Math. Japo., 23 (1978) 1-26. [8] Y. B. Jun, S. Z. Song, Codes based on BCK-algebras, Inform. Sci. 181(2011), 5102-5109 [9]Y. Komori, The class of BCC-algebras is not a variety, Math. Japonica, 29 (1984), 391-394. [ 9] S. M. Mostafa – M. A . Abd-Elnaby- M.M. Yousef,“ Fuzzy ideals of KU – algebra”International Mathematical Forum , Vol. 6 , 2011 .no. 63 ,3139 – 3149. [10] C.Prabpayak and U.Leerawat, On ideals and congruence in KU-algebras, scientia Magna in- ternational book series, Vol. 5(2009), No.1, 54-57. [11] C.Prabpayak and U.Leerawat, On isomorphisms of KU-algebras, scientia Magna international book series, 5(2009), no .3, 25-31. [12 ] A. Borumand Saeid, H. Fatemidokht, C. Flaut and M. Kuchaki Rafsanjani On Codes based on BCK-algebras ,arXiv:1412.8395v1. [13] A. Wronski, BCK-algebras do not form a variety, Math. Japonica, 28 (1983), 211-213. Samy M. Mostafa ([email protected]) Department of Mathematics, Faculty of Education, Ain Shams University, Roxy, Cairo, Egypt. Bayumy.A.Youssef ([email protected] ) Informatics Research Institute,City for Scientific Research and Technological Applications, Borg ElArab, Alexandria, Egypt. Hussein ali gad ([email protected] ) Department of Mathematics and Computer Science .Informatics Research Institute(IRI) City of Scientific Research and Technological applications, Alexandria, Egypt 11
1111.1116	2	1111	2012-03-14T08:41:09	Some Remarks on a Generalized Vector Product	[ "math.RA" ]	In this paper we use a generalized vector product to construct an exterior form $\wedge :(\mathbb{R}^{n}) ^{k}\to \mathbb{R}^{\binom{n}{k}}$, where $\binom{n}{k}=\frac{n!}{(n-k)!k!}$, $k\leq n$. Finally, for $n=k-1$ we introduce the reversing operation to study this generalized vector product over palindromic and antipalindromic vectors.	math.RA	math	SOME REMARKS ON A GENERALIZED VECTOR PRODUCT PRIMITIVO B. ACOSTA-HUM ´ANEZ, MOIS´ES ARANDA, AND REINALDO N ´U NEZ In this paper we use a generalized vector product to construct an exterior Abstract. form ∧ : (Rn)k → R(n (n−k)!k! , k ≤ n. Finally, for n = k − 1 we introduce the reversing operation to study this generalized vector product over palindromic and an- tipalindromic vectors. k), where (cid:0)n k(cid:1) = n! MSC 2010. Primary 15A75, Secondary, 15A72 Keywords and Phrases. Alternating multilinear function, antipalindromic vector, exterior product, palindromic vector, reversing, vector product Introduction It is well known that the vector product over R3 is an alternating 2-linear function from R3 × R3 onto R3. Although this vector product is a natural topic to be studied in any course of basic linear algebra, there is a plenty of textbooks on this subject in where it is not considered over Rn. The following definition, with interesting remarks, can be found also in [3, 7, 8]. Let A1 = (a11, a12, . . . , a1n) , . . . , An−1 =(cid:0)a(n−1)1, a(n−1)2, . . . , a(n−1)n(cid:1) be n − 1 vectors in Rn. The vector product over Rn is a function × : (Rn)n−1 → Rn such that × (A1, A2, . . . , An−1) = A1 × A2 × · · · × An−1 = (−1)1+k det (Xk) ek, (1) n Xk=1 where ek is the k−th unity vector of the standard basis of Rn and Xk is the square matrix obtained through the deleting of the k−th column of the (aij)(n−1)×n. Notice that in this case the function is not binary and sends a matrix M of size (n − 1) × n to a vector of its (cid:0) n n−1(cid:1) maximal minors. One aim of this paper is to give an algorithm to construct, using elementary techniques, k) which will be an alternating k-linear a function with domain in (Rn)k and codomain R(n function that obviously generalizes the previous vector product defined over Rn. Using techniques and methods of algebraic geometry we can see that the vector product obtained here, without signs, corresponds to the Plucker coordinates of the matrix M , see [4, 5]. Although this vector product is known and useful to define the concept of Grassmanian variety, see [4], we present an alternative construction, avoiding algebraic geometry, which lead us to known results that can be found as for example in [6]. Another aim of this work, following [1, 2], is the presentation of some original results concerning to the vector product for n = k − 1 in palindromic and antipalindromic vectors by means of reversing operation. The way as is presented this paper can allow to students and teachers of basic linear algebra the implementation of these results on their courses, this is our final aim. 1 2 P. ACOSTA-HUM ´ANEZ, M. ARANDA, AND R. N ´U NEZ 1. A generalized vector product In this section we set some preliminaries, properties and the Cramer´s rule as applica- tion of the generalized vector product. 1.1. Preliminaries. Following [3, 7] we define the generalized vector product over Rn as the function × : (Rn)n−1 → Rn such that for A1 = (a11, a12, . . . , a1n) , . . . , An−1 = (cid:0)a(n−1)1, a(n−1)2, . . . , a(n−1)n(cid:1), n − 1 vectors of Rn, their vector product is given by × (A1, A2, . . . , An−1) = A1 × A2 × · · · × An−1 = (−1)1+k det (Xk) ek, (2) n Xk=1 where ek is the k−th element of the canonical basis for Rn and Xk is the square matrix obtained after the elimination of the k−th column of the matrix (aij )(n−1)×n. The def- inition presented in expression (2) corresponds to a natural generalization of the vector product of two vectors belonging to R3. 1.2. Some properties. Let A1, A2, . . . , An be vectors of Rn. The following statements hold. 1) × (A1, A2, 2) Assume α, β ∈ R, Bi ∈ Rn: . . . , An−1) is an orthogonal vector for the given vectors. A1 × A2 × · · · × (αAi + βBi) × · · · × An−1 = A1 × A2 × · · · × αAi × · · · × An−1 + A1 × A2 × · · · × βBi × · · · × An−1. 3) Let the matrix A given by A = (A1, A2, . . . , An): det A = A1 · (A2 × · · · × An) = (−1)1+j Aj · (A1 × · · · × Aj−1 × Aj+1 × An) . 4) The vectors A1, A2, . . . , An−1 are n − 1 linearly dependent vectors for Rn if and only if A1 × A2 × · · · × An−1 = 0. It is well known that these properties can be proven using the properties of the determi- nant, see for example [3, 7]. 1.3. Cramer's rule. Consider the following system of linear equations a11x1 + a12x2 + · · · + a1nxn = b1 a21x1 + a22x2 + · · · + a2nxn = b2 ... an1x1 + an2x2 + · · · + annxn = bn that can be expressed in vectorial way as x1A1 + x2A2 + · · · + xnAn = B, (3) being Ai = (a1i, a2i, . . . , ani) with i = 1, 2, . . . , n and B = (b1, b2, . . . , bn). Suppose that det (A1, A2, . . . , An) 6= 0. For instance such system has unique solution that can be obtained applying the scalar product between the equation (3) and A2 × A3 × · · · × An, so we obtain (x1A1 + x2A2 + · · · + xnAn) · A2 × A3 × · · · × An = B · A2 × A3 × · · · × An x1A1 · A2 × A3 × · · · × An = B · A2 × A3 × · · · × An since Aj · A2 × A3 × · · · × An = 0 for j = 2, 3, . . . , n. Therefore x1 = B · A2 × A3 × · · · × An A1 · A2 × A3 × · · · × An = det (B, A2, A3, · · · , An) det(A1, A2, A3, · · · , An) . (4) SOME REMARKS ON A GENERALIZED VECTOR PRODUCT 3 In a general way, we can obtain xi = = = B · A1 × A2 × · · · × Ai−1 × Ai+1 × · · · × An Ai · A1 × A2 × · · · Ai−1 × Ai+1 × · · · × An (−1)i+1 det (A1, A2, . . . , Ai−1, B, Ai+1, · · · , An) (−1)i+1 det(A1, A2, A3, · · · , An) det (A1, A2, . . . , Ai−1, B, Ai+1, · · · , An) , det(A1, A2, A3, · · · , An) that is, the well-known Cramer's rule. 2. Didactic way to define ∧: algorithm and properties In this section we propose a didactic way to define the exterior product ∧. To do this, we set an algorithm to the construction of ∧ and as consequence of this construction arise some properties. 2.1. Algorithm to the construction of ∧. Here we present an algorithm and some simple examples to illustrate it. Step 1. Consider n ∈ N and 1 ≤ k ≤ n, being k an integer. We define I = {i1i2 · · · ik : 1 ≤ i1 < i2 < · · · < ik ≤ n} , this means that the elements belonging to I are chains of numbers conformed in agreement with the lexicographic order. Example 1. For n = 5 and k = 3 we have I = {123, 124, 125, 134, 135, 145, 234, 235, 245, 345} . As we can see, #I =(cid:0)n Example 2. For n = 5 and k = 2, we obtain (cid:0)5 3(cid:1) = 10. k(cid:1) =(cid:0)5 I = {12, 13, 14, 15, 23, 24, 25, 34, 35, 45} . 2(cid:1) = 10 and for instance I is given by Step 2. We set that I should be ordered lexicographically. I(1) < I(2) < · · · < I((n k)) In this way, if Is ∈ I, then there exists p (only one) such that Is = I(p). Thus, we can define p as the rank of Is and will be denoted by r (Is) = p. That is, p is the place of Is in I as set of ordered elements lexicographically. In Example 1 we can see that r (234) = 7, r (345) = 10. The same for Example 2, r (25) = 7, r (35) = 9. Step 3. Let u1 = (u11, u12, . . . , u1n) , . . . , uk = (uk1, uk2, . . . , ukn), be k vectors of Rn, with k ≤ n. Consider the matrix U = (uij ) of order k × n conformed by these vectors. Assume i1i2 · · · ik ∈ I and let Ui1i2 ···ik be the matrix of order k, conformed by the k columns i1, i2, · · · , ik of U . From now on, U always will be a matrix of this kind. Example 3. Consider in this case, U123 =  U =  a2 b2 c2 a3 b3 c3 a1 b1 c1 a1 b1 c1 a2 b2 c2 a3 b3 c3 a4 b4 c4   and U245 =  a5 b5 c5 a2 b2 c2   , a4 b4 c4 a5 b5 c5  . Notice that when we choose a particular number of columns of such matrix U exactly corresponds to delete of U the non-selected columns. 4 P. ACOSTA-HUM ´ANEZ, M. ARANDA, AND R. N ´U NEZ Step 4. Consider (Rn)k := Rn × Rn × · · · × Rn . Now we define the function exterior product ∧ : (Rn)k → R(n ∧ (U ) =Xi∈I (−1)(n where e(n basis of R(n k)−r(i)+1 corresponds to the (cid:0)(cid:0)n k). k−times {z k) as follows: } k)−r(i) det (Ui) e(n k(cid:1) − r(i) + 1(cid:1) −th unity vector of the standard k)−r(i)+1, For convenience, we can write ∧ (U ) = ∧ (u1, u2, . . . , uk) = u1 ∧ u2 ∧ . . . ∧ uk. Example 4. Consider the vectors (2, 3, −1, 5) , (4, 7, 2, 0) ∈ R4. The vector (2, 3, −1, 5) ∧ (4, 7, 2, 0) belongs to R(4 2) = R6. In this case I = {12, 13, 14, 23, 24, 34} , 3 −1 7 2 4 3 U = (cid:18) 2 e5 −(cid:12)(cid:12)(cid:12)(cid:12) 2 (cid:12)(cid:12)(cid:12)(cid:12) e6 +(cid:12)(cid:12)(cid:12)(cid:12) 7 (cid:12)(cid:12)(cid:12)(cid:12) 0 (cid:12)(cid:12)(cid:12)(cid:12) 2 −1 4 e1 5 2 4 such that ∧ (U ) = −(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12) 2 4 −1 2 = −2e6 + 8e5 + 20e4 + 13e3 + 35e2 − 10e1 = (−10, 35, 13, 20, 8, −2) . 5 0 (cid:19) , e4 +(cid:12)(cid:12)(cid:12)(cid:12) 0 (cid:12)(cid:12)(cid:12)(cid:12) 5 3 −1 7 2 (cid:12)(cid:12)(cid:12)(cid:12) e3 −(cid:12)(cid:12)(cid:12)(cid:12) 3 7 e2 5 0 (cid:12)(cid:12)(cid:12)(cid:12) Example 5. Consider the canonical basis for R4, that is, e1 = (1, 0, 0, 0), e2 = (0, 1, 0, 0), e3 = (0, 0, 1, 0) and e1 = (0, 0, 0, 1). Thus, the exterior product ei ∧ ej for i < j is given by e1 ∧ e2 = − (0, 0, 0, 0, 0, 1) = −e6 ∈ R6, e1 ∧ e3 = (0, 0, 0, 0, 1, 0) = e5 ∈ R6, e1 ∧ e4 = − (0, 0, 0, 1, 0, 0) = −e4 ∈ R6, e2 ∧ e3 = (0, 0, 1, 0, 0, 0) = e3 ∈ R6, e2 ∧ e4 = − (0, 1, 0, 0, 0, 0) = −e2 ∈ R6, e3 ∧ e4 = (1, 0, 0, 0, 0, 0) = e1 ∈ R6. As we can see, the set B = {e1 ∧ e2, e1 ∧ e3, e1 ∧ e4, e2 ∧ e3, e2 ∧ e4, e3 ∧ e4} ⊂ R6 is one basis for R6. Notice that in a given basis B for Rn, the exterior product of them taken in sets of k-elements without repetition constitutes a basis B ′ for R(n k). 2.2. Some properties of ∧. The following properties are satisfied by ∧: 1) If k = n, then ∧ (U ) = det (U ). 2) If k = n − 1, then ∧ is the generalized vector product. 3) If n is even and k = 1, then U is orthogonal to ∧ (U ). 4) ∧ is k−linear, ∧ (u1, . . . , ui + b, . . . , uk) = ∧ (u1, . . . , ui, . . . , uk) + ∧ (u1, . . . , b, . . . , uk) . 5) If Mp is a permutation of two rows (being fixed the other ones) of M , then ∧ (Mp) = − ∧ (M ). SOME REMARKS ON A GENERALIZED VECTOR PRODUCT 5 6) If u1, . . . , uk are k (≤ n) linear dependent vectors of Rn, then ∧ (u1, . . . , uk) = 0 ∈ R(n k). Proof. We proceed according to each item. 1) Assuming k = n we have n k! = n n! = 1 and r(i) = 1 (due to I has only one element). For instance ∧ (U ) = Xi∈I = det(Ui). (−1)(n k)−r(i) det (Ui) e(n k)−r(i)+1 Trivially we can see that for R, e1 = 1. 2) Assuming k = n−1, we have n Owing to the symmetry of n n − 1! = n, in this way, I has n elements. k! = n k!, the election of n − 1 columns of the matrix U corresponds to the elimination of one column of U (precisely the avoided column in the election). In other words, we can see that Ui = Xn−r(i)+1 where Xn−r(i)+1 corresponds to the matrix that has been obtained throughout U deleting the (n − r(i) + 1)-th column such that ∧ (U ) = Xi∈I = Xi∈I Xj=1 = n (−1)n−r(i) det (Ui) en−r(i)+1 (−1)(n−r(i)+1)+1 det (Ui) en−r(i)+1 (−1)j+1 det (Xj ) ej = u1 × . . . × uk. 3) For n = 2p and k = 1, we have 2p 1! = 2p, thus, the cardinality of I is even and I = {1, 2, . . . , p, p + 1, . . . , 2p} . Furthermore, r (i) = 1. In this way, ∧ (U ) ∈ R2p. On the other hand, considering U = (u1, u2, . . . , u2p) and ∧ (U ) = (v1, v2, . . . , v2p), we obtain ∧ (U ) = Xi∈I Xi=1 = 2p (−1)2p−i det (Ui) e2p−i+1 (−1)i uie2p−i+1 = (u2p, −u2p−1, . . . , u2, −u1) , where it follows that vj = (−1)j+1 u2p−j+1 for j = 1, 2, . . . , 2p. Therefore U · ∧ (U ) = (u1, u2, . . . , u2p−1, u2p) · (u2p, −u2p−1, . . . , u2, −u1) = u1u2p − u2u2p−1 + . . . + u2p−1u2 − u2pu1 = (u1u2p − u2pu1) + . . . + (−1)p+1 (upup+1 − up+1up) = 0. 6 P. ACOSTA-HUM ´ANEZ, M. ARANDA, AND R. N ´U NEZ Items 4), 5) and 6) can be proven using the properties of the determinant in similar way as the previous ones. (cid:3) 3. Reversing operation over ∧ The reversing operation has been applied successfully over rings and vector spaces, see [1, 2]. In this section we apply the reversing operation to obtain some results that involve the exterior product with the palindromic and antipalindromic vectors. The following results correspond to a generalization of some results presented in [2]. Consider the matrix ←− M = (←−mi,j), M = (mi,j) of size m × n. The reversing of M , denoted by ←− where ←−mi,j = mi,n−j+1. We can see that the size of M is m × n too. We denote by Jn = can be proven, see [2]. In the reversing of the identity matrix In of size n. Thus, the following properties ←− M is given by ←− 1. The double reversing: ←−←− M = ( ←−←−mi,j ) = (←−mi,n−j+1) =(cid:0)mi,n−(n−j+1)+1(cid:1) = (mi,j) = M, ←− M = M Jn 2. 3. JnJn = In. The following definitions were introduced in [2]. A matrix M is ←− M = M , in the same way, a matrix M is called called palindromic whether it satisfies ←− antipalindromic whether it satisfies M = −M . In particular, for m = 1, we get palindromic and antipalindromic vectors respectively. As we can see, the palindromic matrix M satisfies that mi,j = mi,n−j+1 and for instance − 1 when n is odd). pair of equal columns whether n is even (as well M has at least This fact lead us to the following result. n 2 n 2 Proposition 6. det(Jn) =( (−1)n/2, n = 2k, k ∈ Z+ 2 , n = 2k − 1, k ∈ Z+ . (−1) n+3 Proof. We proceed by induction over n. Assuming n = 1, we have that In = 1 and Jn = 1, 1+3 thus det (Jn) = 1 = (−1) 2 . Let the proposition be true for n, thus we will prove that is also true for n + 1. We start considering that n is even, so we get det (Jn+1) = 1 (−1)1+(n+1) det (Jn) = (−1)n+2 (−1) n 2 = (−1) n 2 = (−1) (n+1)+3 2 . Now, considering n as an positive odd integer, we have det (Jn+1) = 1 (−1)1+(n+1) det (Jn) = (−1)n+2 (−1) n+3 2 = (−1) (−1) n+3 2 = (−1) n+5 2 = (−1) n+1 2 . (cid:3) Now, we study the relationship between the exterior product ∧ and the reversing op- eration. We start considering k = n − 1, that is, the generalized vector product over Rn. Consider M1 = (m11, m12, . . . , m1n) , . . . , Mn−1 =(cid:0)m(n−1),1, a(n−1),2, . . . , m(n−1),n(cid:1) , n − 1 vectors in Rn. The generalized vector product is given by the equation (1), therefore we obtain × (M1, M2, . . . , Mn−1) = n Xk=1 (−1)1+k det(cid:16)M (k)(cid:17) ek, (5) SOME REMARKS ON A GENERALIZED VECTOR PRODUCT 7 being ek the k-th element of the canonical basis for Rn and M (k) is the square matrix obtained after the deleting of the k-th column of the matrix M = (mij)(n−1)×n. The matrix M (k) is a square matrix of size (n − 1) × (n − 1) and is given by M (k) =(cid:16)m(k) i,j(cid:17) =(cid:26) mi,j , si j < k mi,j+1 si j ≥ k . (6) ←− M (k) = M (n−k+1)Jn−1, for Proposition 7. If we consider M = (mij)(n−1)×n, then 1 ≤ k ≤ n. Proof. We know that On the other hand, ←− M = M Jn, that is, (←−m i,j) = (mi,n−j+1), 1 ≤ j ≤ n. Therefore ←− M (k) = (cid:16)←−m(k) i,j(cid:17) =(cid:26) ←−m i,j, si j < k ←−m i,j+1 si j ≥ k = (cid:26) mi,n−j+1, si j < k mi,n−(j+1)+1 si j ≥ k . Now, we obtain i,j M (n−k+1) =(cid:16)m(n−k+1) M (n−k+1)Jn−1 = (cid:16)m(n−k+1) mi,j+1 si j ≥ n − k + 1 (cid:17) =(cid:26) mi,j , si j < n − k + 1 (cid:17) i,(n−1)−j+1(cid:17) =(cid:16)m(n−k+1) = (cid:26) mi,(n−j), si n − j < n − k + 1 = (cid:26) mi,(n−j), si j > k − 1 = (cid:26) mi,n−j, si j ≥ k mi,(n−j)+1 si j ≤ k − 1 mi,n−j+1 si j < k mi,(n−j)+1 si n − j ≥ n − k + 1 ←− M (k). i,n−j = . (7) (cid:3) The following proposition is a generalization of one result presented in [2], where was analyzed the reversing of the vector product in R3. From now on, for suitability we denote M = (M1, M2, . . . , Mn−1), i.e., M is the matrix that has as rows the vectors M1, M2, . . . , Mn−1, thus we obtain In the same way, for suitability we write Proposition 8. The generalized vector product of ←− M i, being 1 ≤ i ≤ n − 1, satisfies ←− ←− ←− . . . , . . . , M 1, M 1, ←− M 2, ←− M 2, M n−1(cid:17) . M n−1(cid:17) . M =(cid:16)←− M = ×(cid:16)←− 2 (cid:16)←−−−−−−−−−−−−−−−−−−− 2 (cid:16)←−−−−−−−−−−−−−−−−−−− × (M1, M2, × (M1, M2, 3n+1 3n . . . , Mn−1)(cid:17) , n = 2k, . . . , Mn−1) (cid:17) , n = 2k − 1 , M =  being k ∈ Z+. (−1) (−1) Proof. For suitability we denote M = (M1, M2, . . . , Mn−1), i.e., M is the matrix that has as rows the vectors M1, M2, . . . , Mn−1, thus we obtain In the same way, for suitability we write M 1, ←− M =(cid:16)←− M = ×(cid:16)←− M 1, ←− M 2, . . . , ←− M 2, . . . , ←− M n−1(cid:17) . M n−1(cid:17) . ←− 8 P. ACOSTA-HUM ´ANEZ, M. ARANDA, AND R. N ´U NEZ Now, applying the generalized vector product we obtain M = = = = n n n Xk=1 Xk=1 Xk=1 Xk=1 n M (k)(cid:17) ek (−1)k+1 det(cid:16)←− (−1)k+1 det(cid:16)M (n−k+1)Jn−1(cid:17) ek (−1)k+1 det(cid:16)M (n−k+1)Jn−1(cid:17) ek (−1)k+1 det(cid:16)M (n−k+1)(cid:17) det (Jn−1) ek (−1)n−k det(cid:16)M (k)(cid:17) en−k+1 Xk=1 n n = det (Jn−1) × (M1, M2, = (−1)n+1 det (Jn−1) (−1)k+1 det(cid:16)M (k)(cid:17) en−k+1 Xk=1 = (−1)n+1 det (Jn−1) n (−1)k+1 det(cid:16)M (k)(cid:17) ek! Jn Xk=1 = (−1)n+1 det (Jn−1)(cid:16)←−−−−−−−−−−−−−−−−−− . . . , Mn−1)(cid:17) (cid:16)←−−−−−−−−−−−−−−−−−− . . . , Mn−1)(cid:17) , n = 2k 2 (cid:16)←−−−−−−−−−−−−−−−−−− . . . , Mn−1) (cid:17) , n = 2k − 1 . . . , Mn−1)(cid:17) , n = 2k . . . , Mn−1) (cid:17) , n = 2k − 1 2 (cid:16)←−−−−−−−−−−−−−−−−−− 2 (cid:16)←−−−−−−−−−−−−−−−−−− (−1)n+1 (−1) (−1)n+1 (−1) × (M1, M2, × (M1, M2, × (M1, M2, × (M1, M2, (−1) (−1) 3n+1 (n−1)+3 2 n−1 3n . and for instance, M =   =   (cid:3) n 2 − 1 pair of equal If M is a palindromic matrix, then the minors M (k) have at least n − 1 2 columns when n is even and respectively − 1 when n is odd. This implies that for n ≥ 4, the minors have at least one pair of equal columns and for instance det(cid:16)M (k)(cid:17) = 0 for all 1 ≤ k ≤ n and so × (M1, M2, . . . , Mn−1) = 0 ∈ Rn. (8) This means that the generalized vector product of (n − 1) palindromic vectors in Rn is interesting when 1 ≤ n ≤ 3. The same result is obtained when we assume M as an antipalindromic matrix. Final Remarks When we consider the exterior product for k 6= n − 1, the previous results cannot be applied due to in general they are not true. To illustrate it, we present the following example. Example 9. Consider the vectors (2, 3, −1, 5) and (4, 7, 2, 0) in R4. In this case, M =(cid:18) 2 4 3 −1 2 7 5 0 (cid:19) and ←− M =(cid:18) 5 −1 0 2 3 7 2 4 (cid:19) SOME REMARKS ON A GENERALIZED VECTOR PRODUCT 9 As we have seen before, (2, 3, −1, 5) ∧ (4, 7, 2, 0) = (−10, 35, 13, 20, 8, −2) . Therefore (5, −1, 3, 2) ∧ (0, 2, 7, 4) = −(cid:12)(cid:12)(cid:12)(cid:12) +(cid:12)(cid:12)(cid:12)(cid:12) 5 −1 0 −1 2 3 2 (cid:12)(cid:12)(cid:12)(cid:12) 7 (cid:12)(cid:12)(cid:12)(cid:12) e6 +(cid:12)(cid:12)(cid:12)(cid:12) e3 −(cid:12)(cid:12)(cid:12)(cid:12) 5 0 −1 2 5 0 3 2 e5 −(cid:12)(cid:12)(cid:12)(cid:12) 7 (cid:12)(cid:12)(cid:12)(cid:12) e2 +(cid:12)(cid:12)(cid:12)(cid:12) 4 (cid:12)(cid:12)(cid:12)(cid:12) 3 7 e4 + e1 2 2 4 (cid:12)(cid:12)(cid:12)(cid:12) 4 (cid:12)(cid:12)(cid:12)(cid:12) = −(10)e6 + (35)e5 − (20)e4 + (−7 − 6)e3 − (−4 − 4)e2 + (12 − 14)e1 Thus, in general, the exterior product does not satisfies = (−2, 8, −13, −20, 35, −10). ^ U = (−1)p^ ←− U , for some p ∈ Z. Finally, although this paper is presented in a didactic way, there are original results corresponding to the relations between the reversing operation and the generalized vector product. Acknowledgements The first author is partially supported by MICIIN/FEDER grant number MTM2009- 06973 and by Universidad del Norte. The second author is supported by Pontificia Univer- sidad Javeriana. The third author is partially supported by Universidad Sergio Arboleda. The authors thanks to the anonymous referees by their useful comments and suggestions. 10 P. ACOSTA-HUM ´ANEZ, M. ARANDA, AND R. N ´U NEZ References [1] P. Acosta-Hum´anez, A. Chuquen & A. Rodr´ıguez, Pasting and Reversing operations over some rings, Bolet´ın de Matem´aticas, 17, (2010) 143 -- 164 [2] P. Acosta-Hum´anez, A. Chuquen & A. Rodr´ıguez, Pasting and Reversing operations over some vector spaces, Preprint (2011) [3] M. Aranda & R. N´unez, The Cramer's rule via generalized vector product over Rn (Spanish), Universitas Scientorum, 8, Investigaciones Matem´aticas (2003), 13 -- 15. [4] J. Harris, Algebraic Geometry, A First Course, Springer, New York, (1992) [5] W. V. D. Hodge & D. Pedoe, Methods of Algebraic Geometry, vol I, Cambridge Uni- versity Press, (1994) [6] S. Lang, Linear Algebra , Undergraduate Texts in Mathematics, Springer, New York (1987). [7] M. Marmolejo, Vector product over Rn: The Lagrange's general identity (Spanish), Matem´aticas ensenanza universitaria, vol. 3 (1994), 109 -- 117. [8] J. Olivert, Structures of multilinear algebra (Spanish), Universidad de Valen- cia, Valencia, (1996). Primitivo Acosta-Hum´anez Departamento de Matem´aticas y Estad´ıstica Universidad del Norte Barranquilla, Colombia e-mail: [email protected] Mois´es Aranda Departamento de Matem´aticas Pontificia Universidad Javeriana Bogot´a, Colombia e-mail: [email protected] Reinaldo Nunez Escuela de Matem´aticas Universidad Sergio Arboleda Bogot´a, Colombia e-mail: [email protected]
1208.1317	2	1208	2012-11-16T16:46:35	The centre of generic algebras of small PI algebras	[ "math.RA" ]	Verbally prime algebras are important in PI theory. They are well known over a field $K$ of characteristic zero: 0 and $K<T>$ (the trivial ones), $M_n(K)$, $M_n(E)$, $M_{ab}(E)$. Here $K<T>$ is the free associative algebra with free generators $T$, $E$ is the infinite dimensional Grassmann algebra over $K$, $M_n(K)$ and $M_n(E)$ are the $n\times n$ matrices over $K$ and over $E$, respectively. Moreover $M_{ab}(E)$ are certain subalgebras of $M_{a+b}(E)$, defined below. The generic algebras of these algebras have been studied extensively. Procesi gave a very tight description of the generic algebra of $M_n(K)$. The situation is rather unclear for the remaining nontrivial verbally prime algebras. In this paper we study the centre of the generic algebra of $M_{11}(E)$ in two generators. We prove that this centre is a direct sum of the field and a nilpotent ideal (of the generic algebra). We describe the centre of this algebra. As a corollary we obtain that this centre contains nonscalar elements thus we answer a question posed by Berele.	math.RA	math	The centre of generic algebras of small PI algebras∗ Thiago Castilho de Mello†, Plamen Koshlukov‡ Department of Mathematics, IMECC, UNICAMP S´ergio Buarque de Holanda 651 13083-859 Campinas, SP, Brazil e-mail: [email protected], [email protected] Keywords: generic algebras; central elements; polynomial identities; matrices over Grassmann algebras. Abstract Verbally prime algebras are important in PI theory. They are well known over a field K of characteristic zero: 0 and KhT i (the trivial ones), Mn(K), Mn(E), Mab(E). Here KhT i is the free associative algebra with free generators T , E is the infinite dimensional Grassmann algebra over K, Mn(K) and Mn(E) are the n×n matrices over K and over E, respectively. Moreover Mab(E) are certain subalgebras of Ma+b(E), defined below. The generic algebras of these algebras have been studied extensively. Procesi gave a very tight description of the generic algebra of Mn(K). The situa- tion is rather unclear for the remaining nontrivial verbally prime algebras. In this paper we study the centre of the generic algebra of M11(E) in two generators. We prove that this centre is a direct sum of the field and a nilpotent ideal (of the generic algebra). We describe the centre of this algebra. As a corollary we obtain that this centre contains nonscalar elements thus we answer a question posed by Berele. The verbally prime algebras (also called T-prime) play a crucial role in the theory of the ideals of identities (also called T-ideals) of associative algebras. A T-ideal is called T-prime if it is prime in the class of all T-ideals. Let K be a field and denote by KhT i the free associative algebra freely generated by the set T over K. If charK = 0 then the nontrivial T-prime T-ideals are those of the polynomial identities of the following algebras: Mn(K), Mn(E), Mab(E). We denote here by E the infinite dimensional Grassmann (or exterior) algebra over K. The algebra Mab(E) is a subalgebra of Ma+b(E). It consists of the block matrices having blocks a × a and b × b on the main diagonal with entries from E0, and all remaining entries from E1. Here E0 is the centre of E and E1 is the anticommuting part of E. In order to be more precise, assume V is ∗2010 AMS MSC: 16R10, 16R20, 16R99 †Supported by PhD grant from CNPq ‡Partially supported by grants from CNPq (No. 304003/2011-5), and from FAPESP (No. 2010/50347-9) 1 a vector space with a basis e1, e2, . . . , and let E be the Grassmann algebra of V . Then E has a basis consisting of all elements of the type ei1 . . . eik where i1 < · · · < ik, k ≥ 0, and multiplication induced by eiej = −ejei. Hence E0 is the span of all of the above elements with even k while E1 is the span of those with odd k. The above classification of the T-prime algebras was obtained by Kemer, as a part of the theory that led him to the positive solution of the Specht problem, see [13] for an account of Kemer's theory. Although polynomial identities in T-prime algebras have been extensively studied the concrete information is quite scarce. Thus the polynomial identities for Mn(K) are known only for n ≤ 2, see [19, 20, 9] when K is of characteristic 0, and [14, 8], when K = ∞, charK = p > 2. The identities satisfied by the Grassmann algebra E are well known, see [15] when charK = 0, and the references of [12] for the remaining cases for K. The identities of M11(E) were described in characteristic 0 by Popov, [16]. Recall that the paper [16] gives a basis of the identities satisfied by E⊗E but it is well known (see for example [13]) that the latter algebra satisfies the same identities as M11(E) when charK = 0. Our knowledge about the identities even of M3(K) and of M2(E) is quite limited; it should be noted that no working methods are available in order to describe them. Let A be a PI algebra and suppose I = T (A) is its T-ideal in KhT i. The quotient KhT i/I is the relatively free algebra, also called the generic algebra of A. Thus one may want to study the generic algebras of the T-prime algebras. It is worth mentioning that these generic algebras admit quite natural models as matrices over certain algebras. The generic algebra for Mn(K) is called the generic matrix algebra. It is a fundamental object in Invariant theory, and enjoys very many good properties; one associates the study of the generic matrix algebras with Procesi, see for example [17, 18] and also [11]. Concrete models for the generic algebras for Mab(E) and for Mn(E) were described by Berele [5]. Moreover the study of these generic algebras led to descriptions of their trace rings and to many interesting results in Invariant theory, see for example [3, 4, 6, 7]. The detailed knowledge of the generic algebra for Mn(K) led Procesi [18] to the description of the trace identities of this algebra, a result obtained independently by Razmyslov as well, see [20]. The Razmyslov and Procesi's theorem states that the trace identities for Mn(K) all follow from the Cayley -- Hamilton characteristic polynomial. Razmyslov proved an analogue of this assertion for the algebras Mab(E) as well. In [5, Corollary 21] it was proved that the centre of the generic algebra of Mab(E) is a direct sum of the base field and a nilpotent ideal of the centre. Moreover the author of [5] asked whether the centre of that generic algebra contains any non-scalar elements. In this paper we describe completely the centre of the generic algebra in two generators of M11(E). It follows from our description that it is a direct sum of the field and a nilpotent ideal (of the generic algebra). Moreover we obtain a detailed information about that nilpotent ideal. As a corollary we show that there are very many non-scalar elements in the centre. 2 By using this description of the centre we were able to obtain, in character- istic 0, a basis of the polynomial identities satisfied by the generic algebra of M11(E) in two generators. Clearly these differ significantly from the identities of M11(E). This last result requires quite a lot of work, and will be published in a forthcoming paper. 1 Preliminaries We fix an infinite field K of characteristic different from 2. All algebras and vector spaces we consider will be over K. We denote by KhT i the free (unitary) associative algebra freely generated over K by the infinite countable set T = {t1, t2, . . . }. One may conveniently view KhT i as the algebra of polynomials in the non-commuting variables T . If Tk is a finite set with k elements, say Tk = {t1, . . . , tk} then the free algebra in k generators is denoted by KhTki. The polynomial f (t1, . . . , tn) ∈ KhT i is a polynomial identity for the algebra A if f (a1, . . . , an) = 0 for all ai ∈ A. The set of all polynomial identities satisfied by A is denoted by T (A), it is its T-ideal. Here we suppose T (A) ⊆ KhT i. Set Tk(A) = T (A) ∩ KhTki. Furthermore we denote U (A) = KhT i/T (A) and Uk(A) = KhTki/Tk(A) the relatively free algebras of A of infinite rank and of rank k, respectively. With some abuse of notation we shall use the same letters ti for the free generators of KhT i and for their images under the canonical projection on U (A); analogously for the rank k case. The algebra A is 2-graded if A = A0 ⊕ A1, a direct sum of vector sub- spaces such that AiAj ⊆ Ai+j where the latter sum is taken modulo 2. Such algebras are often called superalgebras. A typical example is the Grassmann algebra E = E0 ⊕ E1 as above. We call the elements from A0 ∪ A1 homoge- neous. When a ∈ Ai we denote its homogeneous degree deg a = i, i = 0, 1. If A is 2-graded and moreover ab − (−1)deg a deg bba = 0 for all homogeneous a and b then A is called a supercommutative algebra. Clearly the Grassmann algebra is supercommutative. Next we recall the construction of the free su- percommutative algebra, see for example [5, Lemma 1]. Let X and Y be two sets and form the free associative algebra KhX ∪ Y i. It is 2-graded assuming the elements of X of degree 0 and those of Y of degree 1. Denote by I the ideal generated by all ab − (−1)deg a deg bba where a, b are homogeneous, and put K[X; Y ] = KhX ∪ Y i/I. It is immediate to see that K[X; Y ] ∼= K[X] ⊗K E(Y ). Here K[X] is the polynomial algebra in X and E(Y ) is the Grassmann algebra of the vector space with basis Y . Thus if Y = {y1, y2, . . .} then E(Y ) will have a basis consisting of the products yi1 · · · yik , i1 < · · · < ik, and multiplication induced by yiyj = −yjyi. Below we also recall the construction of the generic algebras for the T-prime algebras. ij }, Y = {yr ij} where 1 ≤ i, j ≤ n, r = 1, 2, . . . ; observe that we use r as an upper index, not as an exponent. Define the matrices Ar = (xr ij ) where z = x whenever 1 ≤ i, j ≤ a or a + 1 ≤ i, j ≤ a + b, and z = y for all remaining possibilities for i and j. Suppose a + b = n, and consider the following subalgebras of Mn(K[X; Y ]). The first Suppose X = {xr ij ), Br = (xr ij + yr ij), Cr = (zr 3 is generated by the generic matrices Ar, K[Ar r ≥ 1]. It is well known it is isomorphic to the relatively free (or universal) algebra U (Mn(K)) of Mn(K). In [5, Theorem 2] it was shown that U (Mn(E)) ∼= K[Br r ≥ 1], and that U (Mab(E)) ∼= K[Cr r ≥ 1]. Moreover the relatively free algebras of finite rank k, denoted by Uk, can be obtained by letting r = 1, . . . , k, that is by taking the first k matrices. We recall another fact from [5] that we shall exploit. It was shown in [5, Theorem 20] that if f is a central polynomial for Mab(E), without constant term, then for some m the polynomial f m is an identity for Mab(E). It follows that the centre of Uk(Mab(E)) must be a direct sum of K and a nilpotent ideal of the centre, see [5, Corollary 21]. We shall need information about the polynomial identities of M11(E). These were described by Popov in characteristic 0, see the main theorem of [16]. As we mentioned above, in [16] it was proved that the T-ideal of E ⊗ E is generated by the two polynomials [[t1, t2]2, t1], [[t1, t2], [t3, t4], t5] (1) where [a, b] = ab − ba is the usual commutator. We consider the commutators left normed that is [a, b, c] = [[a, b], c], and so on in higher degree. The algebra KhT i is multigraded by the degree of its monomials in each variable. We work with the infinite field K therefore every T-ideal is generated by its multihomogeneous elements, see for example [10, Section 4.2]. Thus from now on we shall work with multihomogeneous polynomials only. The algebra E ⊗ E is PI equivalent to M11(E) in characteristic 0, so the polynomials (1) generate the T-ideal of M11(E) as well. This is a result due to Kemer, see [13]. Kemer proved that the tensor product of two T-prime alge- bras (in characteristic 0) is PI equivalent to a T-prime algebra, and described precisely these PI equivalences. Note that if charK = p > 2 then the algebras M11(E) and E ⊗ E are not PI equivalent, see for example [2], or [1]. While the former paper proved directly the non-equivalence the latter proved it by com- puting the GK dimensions of the corresponding relatively free algebras (these turn out to be different). 2 The free supercommutative algebra In this section we denote by F = U2(M11(E)) the relatively free algebra of rank 2 for M11(E). As mentioned before we have that F = K[C1, C2]. From now on we consider the free supercommutative algebra K[X; Y ] where X = {x1, x2, x′ 2}, 1(cid:19), C2 = (cid:18)x2 y2 y′ 2 x′ 1, x′ 2(cid:19). Clearly the Y = {y1, y2, y′ y1 y′ 1 x′ algebra F satisfies all identities of M11(E). 2}, and set C1 = (cid:18)x1 1, y′ The algebra K[X; Y ] is graded by the integers, taking into account the degree with respect to the variables in Y only: K[X; Y ] = ⊕n∈ZK[X; Y ](n). Here K[X; Y ](n) is the span of all monomials of degree n in the variables from Y . It is immediate that the n-th homogeneous component is zero unless 0 ≤ n ≤ 4. 4 The canonical 2-grading on K[X; Y ] and the Z-grading just defined are related as follows: K[X; Y ]0 = K[X; Y ](0) + K[X; Y ](2) + K[X; Y ](4); K[X; Y ]1 = K[X; Y ](1) + K[X; Y ](3). The next facts are quite obvious; we collect them in a lemma for further reference. Lemma 1 Consider the polynomial algebra K[X] ⊆ K[X; Y ]. 1. For every n, 0 ≤ n ≤ 4, K[X; Y ](n) is a free module over K[X], with a basis Bn, where B0 = {1}, and B1 = {y1, y2, y′ B3 = {y1y2y′ 1, y′ 2}, B2 = {y1y2, y1y′ 1, y1y′ 2, y2y′ 1, y1y2y′ 2, y1y′ 1y′ 2, y2y′ 1y′ 2}, B4 = {y1y2y′ 1y′ 2} 1, y2y′ 1y′ 2, y′ 2}. 2. The free supercommutative algebra K[X; Y ] is a free module over K[X] with a basis B = B0 ∪ B1 ∪ B2 ∪ B3 ∪ B4. 3. Every ideal of K[X; Y ] is a K[X]-submodule of K[X; Y ]. In effect one may extend the scalars as follows. Let K(X) be the field of fractions of K[X], and consider the Grassmann algebra E(Y ) on Y over K(X), that is E(Y ) = E(Y ) ⊗K K(X). Lemma 2 The matrices C1 and C2 are not zero divisors in F . Proof . Suppose A = (cid:18)a b y′ 1a + x′ x1a + y1c = 0, c d(cid:19) ∈ F is such that C1A = 0. Then one obtains 1c = 0, x1b + y1d = 0, y′ 1b + x′ 1d = 0. 1c = −y′ 1 and substitute x′ It follows from the second equation that x′ 1a. Now multiply the first equation by x′ 1) = 0. Extending the scalars as above, and noting that the set B from Lemma 1 is a basis of the vector space E(Y ) over K(X), we immediately obtain a = 0 and consequently c = 0. In the same manner but using the last two equations we obtain b = d = 0. In this way C1 is not a left zero divisor. Analogously one ♦ shows it is not a right zero divisor, and the same for C2. 1c by its equal and get a(x1x′ 1 − y1y′ We define an automorphism ′ on K[X; Y ] by setting xi 7→ x′ i, x′ i 7→ xi, yi 7→ y′ i and y′ i 7→ yi. Thus the automorphism ′ is of order two. We also define the following polynomials in K[X]: n qn(x1, x′ 1x′n−i xi 1 ; Qn(x2, x′ 2) = qn(x2, x′ 2); n−1 1) = Xi=0 Xi=0 1) = rn(x′ 1) = rn(x1, x′ sn(x1, x′ (n − i)xn−1−i 1 x′i 1 ; Rn(x2, x′ 2) = rn(x2, x′ 2); 1, x1); Sn(x2, x′ 2) = sn(x2, x′ 2). 5 Lemma 3 The following relations hold among the above polynomials. rn = qn−1 + x1rn−1; qn = xn xn 1 x′ 1 1 + x′ m − xm sn = qn−1 + x′ n + x1qn−1; 1sn−1; (x′ 1qn−1 = x′ 1 n = (x′ 1 x′ 1 − x1)(qnqm−1 − qmqn−1). 1 sn + rn = (n + 1)qn−1; 1 − x1)qn−1 = x′ 1 n − xn 1 ; Proof . The proof consists of an easy induction. ♦ It is immediate to check, once again by induction, that for every m and n, 1rn−1 C n 1 = (cid:18)xn 2 = (cid:18)xm 1 + y1y′ y′ 1qn−1 2 + y2y′ y′ 2Qm−1 Cm 2Rm−1 y1qn−1 n − y1y′ x′ 1 1sn−1(cid:19) ; 2Sm−1(cid:19) . y2Qm−1 x′ 2 m − y2y′ Note that the yi and the y′ the (2, 2)-entries of the above matrices. i anticommute and this produces the minus signs at Therefore for the product C n 1 Cm 2 we have C n 1 Cm 2 = (cid:18) 1 xm xn 2 qn−1 + y′ 1xm y′ 2 + a + d 2x′ 2 nQm−1 + c′ y1x′ 2 mqn−1 + y2xn x′ 1 nx′ 2 m + a′ + d′ 1 Qm−1 + c (cid:19) where a, a′ ∈ K[X; Y ](2), d, d′ ∈ K[X; Y ](4), and c, c′ ∈ K[X; Y ](3). As the elements of the algebra F are linear combinations of products of the above type we obtain immediately the proof of the following lemma. Lemma 4 Let A = (aij) ∈ F , then a22 = a′ 11 and a21 = a′ 12. We shall need the following elements in order to describe the centre of F . 1y′ h1 = y1y2y′ 2; 1(x′ h2 = y1y2(y′ h3 = y′ 1y′ 2(y1(x′ h4 = (y′ 1(x′ 2 − x2) − y′ 2(x′ 2 − x2) − y2(x′ 1 − x1)); 1 − x1)); 1 − x1))(y1(x′ 2 − x2) − y′ 2(x′ 2 − x2) − y2(x′ 1 − x1)). Once again using the fact that y1, y′ 1, y2, y′ 2 anticommute we obtain that the elements h1, h2, h3, h4 satisfy the following relations in K[X; Y ]. 1 = h1y2 = h1y′ 1 − x1)h1; 1 = h3y1 = (x′ 2 = 0; h2y′ h4y2 = (x′ h2y1 = h2y2 = h3y′ 2 = h3y2 = (x′ 2 − x2)h2; 2 − x2)h1; 1 = h3y′ 2 = 0; h1y1 = h1y′ h2y′ h4y1 = (x′ h4y′ 1 = −(x′ 1 − x1)h2; 1 − x1)h3; h4y′ 2 = −(x′ 2 − x2)h3. 6 3 The centre of F2(M11(E)) Take a matrix A = (cid:18)a b c d(cid:19) ∈ F ∼= K[C1, C2]. Assume it is central in F , that is [A, C1] = [A, C2] = 0. Then by′ 2 + cy2 = 0; by′ 1 + cy1 = 0; (a − d)y1 + (x′ (a − d)y′ 1 + (x′ 1 − x1)b = 0; 1 − x1)c = 0; (a − d)y2 + (x′ (a − d)y′ 2 + (x′ 2 − x2)b = 0; 2 − x2)c = 0. (2) We multiply the third equation by (x′ 2 − x2)y1 − (x′ obtaining (d − a)((x′ 2 − x2)y′ equations we get (d − a)((x′ 2−x2), the fourth by (x′ 1−x1) and subtract, 1 − x1)y2) = 0. Similarly from the last two 1 − (x′ 2) = 0. Therefore 1 − x1)y′ d − a ∈ Ann((x′ 2 − x2)y1 − (x′ 1 − x1)y2) ∩ Ann((x′ 2 − x2)y′ 1 − (x′ 1 − x1)y′ 2). Denote by J the intersection of the two annihilators in the right hand side above. Then J is an ideal of K[X; Y ] hence J is a K[X]-submodule as well. Proposition 5 The K[X]-module J is spanned by {h1, h2, h3, h4}. Proof . It is immediate that J contains h1, h2, h3, h4. We shall prove that J is contained in the K[X]-module spanned by {h1, h2, h3, h4}. Let f ∈ J and write f = f0 + f1 + f2 + f3 + f4 where fi ∈ K[X; Y ](i). First note that both 2) lie in K[X; Y ](1) in ((x′ the Z-grading defined above. Thus f annihilates the latter two polynomials if and only if every fi does. Therefore we may and shall assume f is homogeneous in the Z-grading. 1 − x1)y2) and ((x′ 2 − x2)y1 − (x′ 2 − x2)y′ 1 − x1)y′ 1 − (x′ Suppose first f ∈ K[X; Y ](0) ∼= K[X]. As f ∈ J then it follows easily that f = 0. Let f ∈ K[X; Y ](1) ∩ J, then f = α1y1 + α2y2 + α3y′ 1 + α4y′ 2 for some αi ∈ K[X]. But f (y1(x′ 2 − x2) − y2(x′ 1 − x1)) = 0, hence (x′ 1−x1)(−α1y1y2+α3y2y′ 1+α4y2y′ 2)−(x2−x′ 2)(−α2y1y2−α3y1y′ 1−α4y1y′ 2)) = 0. As B2 is a K[X]-basis for the free module K[X; Y ](2) it follows α3 = α4 = 0. In the same way, using the fact that f annihilates y′ 1 − x1), we get α1 = α2 = 0. This implies J ∩ K[X; Y ](1) = 0. 2 − x2) − y′ 2(x′ 1(x′ Let f = α1y1y2+α2y1y′ As above, from f (y1(x′ combination of y1y2y′ 1+α3y1y′ 2 −x2)−y2(x′ 1, y1y2y′ 2, y1y′ 1+α5y2y′ 2 ∈ K[X; Y ](2)∩J. 2+α4y2y′ 1−x1)) = 0 we obtain, in K[X; Y ](3), a linear 1y′ 2, with coefficients respectively 2+α6y′ 2, y2y′ 1y′ 1y′ α2(x′ 1 − x1) + α4(x′ 2 − x2); α3(x′ 1 − x1) + α5(x′ 2 − x2); α6(x′ 2 − x2); −α6(x′ 1 − x1). Thus α6 = 0, α2(x′ 1 − x1) + α4(x′ Analogously, from f (y′ 1 − x1) + α3(x′ 2 − x2) − y′ 2 − x2) = 0, and α4(x′ 1(x′ 2(x′ 1 − x1) + α5(x′ α2(x′ 2 − x2) = 0. 2 − x2) = 0, and α3(x′ 2 − x2) = 0. 1 − x1)) = 0 we obtain that α1 = 0, 1 − x1) + α5(x′ 7 Now the αi are polynomials in K[X]. Consider the field of fractions K(X) of this polynomial ring, and resolve the corresponding linear system of four equations in K(X). One obtains that the solution depends on one parameter β ∈ K(X): α2 = β; α3 = − x′ 1 − x1 x′ 2 − x2 β; α4 = − x′ 1 − x1 x′ 2 − x2 β; α5 = (x′ (x′ 1 − x1)2 2 − x2)2 β, and of course α1 = α6 = 0. Since we are looking for a solution of the system in 2 − x2)2α for some α ∈ K[X], and the solution K[X] then one must have β = (x′ in K[X] will be α2 = (x′ α4 = −(x′ 2 − x2)2α; α3 = −(x′ 1 − x1)(x′ 2 − x2)α; α5 = (x′ 2 − x2)α; 1 − x1)2α. 1 − x1)(x′ When we substitute these in the expression of f we get f = αh4. Let f = α1y1y2y′ 1y′ 2 + α4y2y′ αi ∈ K[X]. Proceeding as above we get the equalities 1 + α2y1y2y′ 2 + α3y1y′ 1y′ 2 ∈ K[X; Y ](3) ∩ J where (α4(x′ (α2(x′ 2 − x2) + α3(x′ 2 − x2) + α1(x′ 1 − x1))y1y′ 1 − x1))y1y′ 1y2y′ 1y2y′ 2 = 0 2 = 0 2 − x2) + α1(x′ therefore α2(x′ 1 − x1) = 0. As above we work first in K(X) and then go back to K[X]. We find that the solution in K[X] is 1 − x1) = 0 and α4(x′ 2 − x2) + α3(x′ α1 = −(x′ 2 − x2)α; α2 = (x′ 1 − x1)α; α3 = −(x′ 2 − x2)β; α4 = (x′ 1 − x1)β where α, β ∈ K[X]. Substituting these values in f we obtain f = αh2 + βh3, and this case is dealt with. Finally it is immediate to see that K[X; Y ](4) = K[X] · h1, and thus we ♦ conclude the proof. Corollary 6 Let the matrix A be as at the beginning of this section. Then A commutes with C1 and C2 if and only if b = f4h2, c = −f4h3, d = a+f1h1+f4h4 for some f1, f4 ∈ K[X]. Proof . We already saw that d−a ∈ J. Thus d−a = f1h1+f2h2+f3h3+f4h4, fi ∈ K[X]. By means of homogeneity we get d − a ∈ K[X; Y ]0, it follows that d − a = f1h1 + f4h4. Now substituting d − a in the system (2) we have b = f4h2 and c = −f4h3. On the other hand if a, b, c, d satisfy the conditions of the statement it is immediate that [A, C1] = [A, C2] = 0. ♦ Remark 1. We just proved that [A, C1] = [A, C2] = 0 if and only if A = aI + f1(cid:18)0 0 h1(cid:19) + f4(cid:18) 0 −h3 h4(cid:19) . h2 0 8 2. Therefore if A is central in F then a12, a21 ∈ K[X; Y ](3). An element a ∈ F will be called strongly central if it is central, and moreover, for every b ∈ F the element ab is central in F (thus ba = ab will be strongly central as well). Let us fix the following matrices in F : A0 = (cid:18)h1 0 0 h1(cid:19) ; A1 = (cid:18)0 0 h1(cid:19) ; A2 = (cid:18) 0 −h3 h4(cid:19) ; A3 = (cid:18)h4 h2 0 0 0 h4(cid:19) . Lemma 7 Let a = α0A0 + α1A1 + α2A2 + α3A3, for some αi ∈ K[X]. If a ∈ F then a is strongly central. Proof . The matrices A0 and A1 are clearly strongly central. Also A2 and A3 are central. One computes A2Ci = (x′ i − xi)(cid:18)h1 0 −h1(cid:19) + x′ i(cid:18) 0 −h3 h4(cid:19) . h2 0 Hence A2Ci is a linear combination (over K[X]) of A0, A1 and A2. Iterating we will have A2Ci1 Ci2 . . . Cir is central for ij = 1, 2 and r = 1, 2 . . . , and A2 is strongly central. One checks in a similar manner that A3Ci = xi(cid:18)h4 0 0 h4(cid:19) + (x′ i − xi)(cid:18) 0 −h3 h4(cid:19) . h2 That is A3Ci is a combination of A2 and A3 and iterating as above we show that A3 is strongly central. ♦ Lemma 8 Let f (t1, t2) = [t1, t2, ti3 , . . . , tik ] be a left normed commutator, ij = 1, 2. Suppose that degt1 f = n, degt2 f = m, n + m = k. Then for every n and m one has f (C1, C2) = (x′ 2 − x2)m−1A(k) where 1 − x1)n−1(x′ A(k) = (cid:18) (−1)k(y′ F (k) 2(x′ 1 − x1) − y′ 1(x′ 2 − x2)) F (k) y1(x′ 2 − x2) − y2(x′ 1 − x1) (cid:19) , F (k) = (y1(x′ 2 − x2) − y2(x′ 1 − x1))y′ i + (−1)kyi(y′ 2(x′ 1 − x1) − y′ 1(x′ 2 − x2)) x′ i − xi . In the last expression we use the shorthand i for ik. Proof . The proof consists of an induction on k. The base of the induction is 1y2. If f = fk is the commutator of the statement ♦ k = 2; then F (2) = y1y′ then one computes [fk, Ci] directly by induction. 2 + y′ Remark It follows from the above lemma that if u is a left normed commutator in t1 and t2 then u(C1, C2) does not depend on the order of the variables starting with the third and up to the last but one. In other words any permutation of the variables in u that preserves the first two and the last one, leaves u invariant. 9 Lemma 9 Let u1(t1, t2) and u2(t1, t2) be two left-normed commutators of de- grees at least two in F , and denote by u = u1u2 their product. Then u is strongly central in F . Proof . Suppose degt1 uj = nj and degt2 uj = mj, j = 1, 2. Using the notation of Lemma 8 we have u1(C1, C2)u2(C1, C2) = (x′ 1 − x1)n1+n2−2(x′ 2 − x2)m1+m2−2A(k1)A(k2) where deg uj = kj. But it is immediate to see that F (k1)F (k2) = αh1, and also F (ki)(y1(x′ F (ki)(y′ 2(x′ 2 − x2) − y2(x′ 1 − x1) − y′ 1(x′ 1 − x1)) = −(−1)kih2, 2 − x2)) = h3 where α ∈ K[X]. Then A(k1)A(k2) = F (k1)F (k2)I +(cid:18) ((−1)k1 + (−1)k2 )h3 −(−1)k1h4 (−1)k2 h4 −((−1)k1 + (−1)k2 )h2 (cid:19) . Therefore u1(C1, C2)u2(C1, C2) = αA0 + (−1)k2 A3 − ((−1)k1 + (−1)k2)A2 is strongly central in F . ♦ Remark Let u1 and u2 be two commutators, deg u1 ≡ deg u2 (mod 2). Then u = u1u2 is central element in F but it is not a scalar multiple of I. In particular [C1, C2]2 = (cid:18)−2h1 + h4 2h3 −2h2 −2h1 − h4(cid:19) is central in F but is not a scalar. This answers Berele's question from [5] for the case of M11(E) and two generators. In this same case we shall give below the precise answer to Berele's question. We consider unitary algebras. Let L(T ) be the free Lie algebra freely gen- erated by T ; suppose further L(T ) ⊆ KhT i. That is we consider the vector space KhT i with the commutator operation [a, b] = ab − ba, and take L(T ) as the Lie subalgebra generated by T . Choose an ordered basis of L(T ) such that the variables from T precede the longer commutators. As KhT i is the universal enveloping algebra of L(T ) one has that a basis of KhT i consists of 1 and all products tn1 uj1 · · · ujm where i1 < · · · < ik, and the uji are commutators of degree at least two. Clearly all this holds for Kht1, t2i and for its homomor- phic image F = K[C1, C2]. Therefore every element of F is a linear combination of products of the type C n 2 u1 · · · ur where the ui are commutators. Moreover we can assume all commutators left normed, and of the type [C1, C2, . . .]. i1 · · · tnk ik 1 Cm Proposition 10 Let f (C1, C2) = C n r ∈ F where the ui are left normed commutators, deg ui ≥ 2. The element f is central in F if and only if k1 + · · · + kr ≥ 2, or else m = n = k1 = · · · = kr = 0. 1 · · · ukr 1 Cm 2 uk1 10 2 and C n Proof . It follows from Lemma 9 that the product of two commutators is strongly central. Thus if k1 + · · · + kr ≥ 2 then f (C1, C2) is central. It remains to prove that C n 2 u are not central where u is a left normed commutator. (In the former we assume n + m > 0.) Take first C n 2 . By the form of the product (computed just before Lemma 4) it follows that the (1, 2) 2 is αy1 + βy2 + γ where (α, β) 6= (0, 0) and γ ∈ K[X; Y ](3). Now entry of C n by the remark following Corollary 6 we have that C n 2 cannot be central as long as n + m > 0. 1 Cm 1 Cm 1 Cm 1 Cm 1 Cm One proceeds in a similar manner when f = C n 2 u where u is a left normed commutator. The (1, 2) entry of f will be xn 1 − x1)) + b for some b ∈ K[X; Y ](3). Once again the remark mentioned above yields that f cannot be central. ♦ 1 Cm 2 (y1(x′ 2 − x2) − y2(x′ 1 xm Proposition 11 Let u1, . . . , ur be left normed commutators, degt1 uj = n, degt2 uj = m for all j. Suppose uj = [t1, t2, tj3 , . . . , tjk ], j = 1, . . . , r. Put f (C1, C2) = Pj αjuj(C1, C2) ∈ F where αj ∈ K[X]. Then the following three conditions are equivalent. (1) The element f (C1, C2) is strongly central in F . (2) The element f (C1, C2) is central in F . (3) The sum α1 + · · · + αr = 0 in K[X; Y ]. Proof . Clearly (1) implies (2). We prove now that (2) implies (3). Suppose f (C1, C2) is central in F . By Lemma 8 the (1, 2) entry of every commutator ui equals (x′ 1 − x1)). Hence the (1, 2) entry of f (C1, C2) equals 2 − x2)m−1(y1(x′ 2 − x2) − y2(x′ 1 − x1)n−1(x′ 2 − x2) − y2(x′ X βi(y1(x′ 1 − x1)) = (y1(x′ 2 − x2)m−1αi. Thus if P αi 6= 0 then the (1, 2) where βi = (x′ entry of f (C1, C2) is a non-zero multiple of y1(x′ 1 − x1) by some element of K[X; Y ]0 and cannot belong to K[X; Y ](3). By the remark following Corollary 6, f (C1, C2) cannot be central. 1 − x1))X βi 2 − x2) − y2(x′ 2 − x2) − y2(x′ 1 − x1)n−1(x′ In order to complete the proof we have to prove that (3) implies (1). Suppose uj have the same rightmost variable then ui(C1, C2) = uj(C1, C2). Thus we divide the commutators uj into two types according to their rightmost variable. P αi = 0. It was observed in the remark preceding Lemma 9 that if ui and Clearly if all of them end with say t1 then P αjuj(C1, C2) = u1(C1, C2)P αj = 0. Hence suppose u1 ends with t1 while u2 ends with t2. Write P αjuj = Then β1 + β2 = P αj = 0 and it suffices to prove u1 − u2 is strongly central. 2 − x2)m−1(F1(k) − F2(k))I where we denote by But u1 − u2 = (x′ Fj(k) the expression F (k) from Lemma 8 obtained by uj, j = 1, 2. Clearly 2 − x2)−1h4 F1(k) − F2(k) = 0 if k is even, and F1(k) − F2(k) = −2(x′ β1u1 + β2u2 where βq is the sum of all αj such that uj ends with tq, q = 1, 2. 1 − x1)n−1(x′ 1 − x1)−1(x′ 11 if k is odd. In this way either u1 − u2 = 0 or u1 − u2 is a multiple of A3. In ♦ both cases it is strongly central in F . Remark We observe that in the previous proposition if we suppose αj ∈ K[X; Y ]0, the statement of the proposition remains valid replacing the condition (3) by the condition (3') The sum α1 + · · · + αr ∈ K[X; Y ](4). Let f (C1, C2) ∈ F = K[C1, C2]. Then f can be written as f (C1, C2) = Xn,m≥0 αnmC n 1 Cm 2 + Xnj ,mj C nj 1 Cmj 2 Xi βijuij + g(C1, C2). Here αnm, βij ∈ K, uij are left normed commutators as in Proposition 11, and 2 u1 · · · uk where ui are left normed commuta- tors. Define Iij as the set of all indices p such that degt1 upj = rij, degt2 upj = sij for some integers rij and sij . moreover g(C1, C2) = Pu γuC n 1 Cm Theorem 12 Using the notation above, f (C1, C2) is central in F if and only if αnm = 0 for all n and m such that n + m ≥ 1, and moreover, for every i and j the equalities Pp βpj = 0 hold where p ∈ Iij . Furthermore f (C1, C2) is strongly central if and only if it is central and α00 = 0. when n + m ≥ 1 and all sums Pp∈Iij Proof . We already proved that f (C1, C2) is central provided that αnm = 0 βpj = 0. Such an element is strongly central if and only if α00 = 0. We shall prove the converse. Clearly g(C1, C2) is strongly central and α00I is central. 1 Cm C nj 1 Cmj So suppose Pn+m≥1 αnmC n 2 +Pnj ,mj computation of C n 1 Cm Pn+m≥1 αnmC n Xn+m≥1 2 done just before Lemma 4 yields that the (1, 2) entry of 2 Pi βijuij is central. The 1 Cm 2 will be equal to αnm(qn−1x′ 2 my1 + Qm−1xn 1 y2) + µ, µ ∈ K[X; Y ](3). C nj 1 Cmj βijxnj 1 xmj 2 (x′ Analogously the (1, 2) entry of Pnj ,mj Xnj ,mj ,i Here rij = degt1 uij, sij = degt2 uij, and ρ ∈ K[X; Y ](3). Since our element is central its (1, 2) entry lies in K[X; Y ](3). Thus we obtain that the sum 2 Pi βij uij is 2 − x2)sij −1(y1(x′ 1 − x1)rij −1(x′ 2 − x2) − y2(x′ 1 − x1)) + ρ. Xn+m≥1 Xnj ,mj ,i αnm(qn−1x′ 2 my1 + Qm−1xn 1 y2) + βijxnj 1 xmj 2 (x′ 1 − x1)rij −1(x′ 2 − x2)sij −1(y1(x′ 2 − x2) − y2(x′ 1 − x1)) 12 must vanish. But the set B1 is a basis of K[X; Y ](1) therefore Xn+m≥1 Xn+m≥1 αnmqn−1x′ 2 αnmQm−1xn m + Xnj ,mj ,i 1 − Xnj ,mj ,i βijxnj 1 xmj 2 (x′ 1 − x1)rij −1(x′ 2 − x2)sij = 0 βijxnj 1 xmj 2 (x′ 1 − x1)rij (x′ 2 − x2)sij −1 = 0. Multiplying the first equation by (x′ up we will obtain that 1 −x1), the second by (x′ 2 −x2) and summing 0 = Xn+m≥1 = Xn+m≥1 = Xn+m≥1 αnm(qn−1x′m 2 (x′ 1 − x1) + Qm−1xn 1 (x′ 2 − x2)) αnm(x′ 2 m(x′ 1 n − xn 1 ) + xn 1 (x′ 2 m − xm 2 )) αnm(x′ 2 mx′ 1 n − xn 1 xm 2 ). Therefore αnm = 0 whenever n + m ≥ 1. So we are left with the sum 2 (x′ βij xnj 1 xmj 1 − x1)rij −1(x′ Xnj ,mj ,i Thus we have that Pnj ,mj ,i βijxnj Pnj ,mj ,i βij xnj 1 − x1)rij (x′ that for each j it holdsPi βij(x′ of the sets Iij we have Pp∈Iij 1 xmj 2 (x′ 2 − x2)sij −1(y1(x′ 2 − x2) − y2(x′ 1 − x1)) = 0. 2 (x′ 1 − x1)rij −1(x′ 1 xmj 2 − x2)sij = 0. Similarly 2 − x2)sij −1 = 0. By homogeneity we deduce 2−x2)sij = 0. Recalling the definition ♦ 1−x1)rij (x′ βpj = 0 and we are done. Corollary 13 For the centre Z(F ) we have Z(F ) = K ⊕I where I is a nilpotent ideal of F (and not only of Z(F )). We observe that the last Corollary, together with Theorem 12 gives a precise answer to the question of Berele, and that I is a nilpotent ideal actually of F , not only of the centre. A further remark is relevant. It is interesting to note that when one deals with 3 generators, say the generic matrices C1, C2, C3, then the element [C1, C2, [C1, C3]] is central in the generic algebra of three generators. But C2[C1, C2, [C1, C3]] is not. Therefore the analogue of the above nilpotent ideal is an ideal of the centre only. References [1] S. M. Alves, P. Koshlukov, Polynomial identities of algebras in positive characteristic, J. Algebra 305 (2) (2006), 1149 -- 1165. [2] S. Azevedo, M. Fidelis, P. Koshlukov, Tensor product theorems in positive characteristic, J. Algebra 276 (2) (2004), 836 -- 845. 13 [3] A. Berele, Trace identities and Z/2Z-graded invariants, Trans. Amer. Math. Soc. 309 (2) (1988), 581 -- 589. [4] A. Berele, Trace rings for verbally prime algebras, Pacific J. Math. 150 (1) (1991), 23 -- 29. [5] A. Berele, Generic Verbally Prime Algebras and their GK-dimensions, Commun. Algebra 21 (5) (1993), 1487 -- 1504. [6] A. Berele, Supertraces and matrices over Grassmann algebras, Adv. Math. 108 (1994), 77 -- 90. [7] A. Berele, Invariant theory and trace identities associated with Lie color algebras, J. Algebra 310 (2007), 194 -- 206. [8] J. Colombo, P. Koshlukov, Central polynomials in the matrix algebra of order two, Linear Algebra Appl. 377 (2004), 53 -- 67. [9] V. Drensky, A minimal basis for the identities of a second-order matrix algebra over a field of characteristic 0 (Russian), Algebra i Logika 20 (3) (1981), 282 -- 290. Translation: Algebra Logic 20 (3) (1981), 188 -- 194. [10] V. Drensky, Free algebras and PI-algebras: Graduate Course in Algebra, Springer, Singapore, 1999. [11] E. Formanek, The Polynomial Identities and Invariants of n × n Matrices, CBMS Regional Conference Series in Mathematics, Vol. 78, AMS, Provi- dence, RI, 1991. [12] A. Giambruno, P. Koshlukov, On the identities of the Grassmann algebras in characteristic p > 0, Israel J. Math. 122 (2001), 305 -- 316. [13] A. R. Kemer, Ideals of Identities of Associative Algebras, Translations Math. Monographs 84, AMS, Providence, RI, 1991. [14] P. Koshlukov, Basis of the identities of the matrix algebra of order two over a field of characteristic p 6= 2, J. Algebra 241 (2001), 410 -- 434. [15] D. Krakowski, A. Regev, The polynomial identities of the Grassmann alge- bra, Trans. Amer. Math. Soc. 191 (1973), 429 -- 438. [16] A. Popov, Identities of the tensor square of a Grassmann algebra, Algebra i Logika 21 (4) (1982) 442 -- 471. Translation: Algebra Logic 21 (4) (1982), 296 -- 316. [17] C. Procesi, Rings With Polynomial Identities, Pure Appl. Math. 17, Marcel Dekker, New York, 1973. [18] C. Procesi, The invariant theory of n × n matrices, Adv. Math. 19 (3) (1976), 306 -- 381. 14 [19] Yu. P. Razmyslov, The existence of a finite basis for the identities of the matrix algebra of order two over a field of characteristic zero (Russian), Algebra i Logika 12 (1973), 83 -- 113, 121; Translation: Algebra Logic 12 (1973), 47 -- 63 (1974). [20] Yu. P. Razmyslov, Identities of algebras and their representations, Transla- tions of Math. Monographs 138, Amer. Math. Soc., Providence, RI, 1994. 15
1910.08984	1	1910	2019-10-20T14:09:05	Relative and unrelative elementary groups, revisited	[ "math.RA", "math.GR" ]	Let $R$ be any associative ring with $1$, $n\ge 3$, and let $A,B$ be two-sided ideals of $R$. In the present paper we show that the mixed commutator subgroup $[E(n,R,A),E(n,R,B)]$ is generated as a group by the elements of the two following forms: 1) $z_{ij}(ab,c)$ and $z_{ij}(ba,c)$, 2) $[t_{ij}(a),t_{ji}(b)]$, where $1\le i\neq j\le n$, $a\in A$, $b\in B$, $c\in R$. Moreover, for the second type of generators, it suffices to fix one pair of indices $(i,j)$. This result is both stronger and more general than the previous results by Roozbeh Hazrat and the authors. In particular, it implies that for all associative rings one has the equality $\big[E(n,R,A),E(n,R,B)\big]=\big[E(n,A),E(n,B)\big]$ and many further corollaries can be derived for rings subject to commutativity conditions.	math.RA	math	COMMUTATORS OF RELATIVE AND UNRELATIVE ELEMENTARY GROUPS, REVISITED N. VAVILOV AND Z. ZHANG Abstract. Let R be any associative ring with 1, n ≥ 3, and let A, B be two-sided ideals of R. In the present paper we show that the mixed commutator subgroup [E(n, R, A), E(n, R, B)] is generated as a group by the elements of the two following forms: 1) zij(ab, c) and zij(ba, c), 2) [tij (a), tji(b)], where 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, c ∈ R. Moreover, for the second type of generators, it suffices to fix one pair of indices (i, j). This result is both stronger and more general than the previous results by Roozbeh Hazrat and the authors. In particular, it implies that for all associative rings one has the equality [E(n, R, A), E(n, R, B)] = [E(n, A), E(n, B)] and many further corollaries can be derived for rings subject to commutativity conditions. To the remarkable St Petersburg algebraist Alexander Generalov 9 1 0 2 t c O 0 2 ] . A R h t a m [ 1 v 4 8 9 8 0 . 0 1 9 1 : v i X r a 1. Introduction In the present note we generalize and strengthen the results by Roozbeh Hazrat and the authors [15, 13, 28] on generation of mutual commutator subgroups of relative and unrelative elementary subgroups in the general linear group. Namely, we both dramatically reduce the sets of generators that occur therein and either seriously weaken, or completely remove commutativity conditions. Let R be an associative ring with 1, and GL(n, R) be the general linear group of degree n ≥ 3 over R. As usual, e denotes the identity matrix, whereas eij denotes a standard matrix unit. For c ∈ R and 1 ≤ i 6= j ≤ n, we denote by tij(c) = e + ceij the corresponding elementary transvection. To an ideal A E R, we assign the elementary subgroup E(n, A) = htij(a), a ∈ A, 1 ≤ i 6= j ≤ ni. The corresponding relative elementary subgroup E(n, R, A) is defined as the normal closure of E(n, A) in the absolute elementary subgroup E(n, R). From the work of Michael Stein, Jacques Tits, and Leonid Vaserstein it is classically known that as a group E(n, R, A) is generated by zij(a, c) = tji(c)tij(a)tji(−c), where 1 ≤ i 6= j ≤ n, a ∈ A, c ∈ R. Key words and phrases. General linear groups, elementary subgroups, congruence subgroups, standard commutator formula, unrelativised commutator formula, elementary generators. The work of the first author was supported by the Russian Science Foundation grant 17-11-01261. 1 2 N. VAVILOV AND Z. ZHANG Further, consider the reduction homomorphism ρI : GL(n, R) −→ GL(n, R/I) modulo I. By definition, the principal congruence subgroup GL(n, I) = GL(n, R, I) is the kernel of ρI. In other words, GL(n, I) consists of all matrices g congruent to e modulo I. A first version of following result was discovered (in a slightly less precise form) by Roozbeh Hazrat and the second author, see [15], Lemma 12. In exactly this form it is stated in our paper [13], Theorem 3A. Theorem A. Let R be a quasi-finite ring with 1, let n ≥ 3, and let A, B be two- sided ideals of R. Then the mixed commutator subgroup [E(n, R, A), E(n, R, B)] is generated as a group by the elements of the form • zij(ab, c) and zij(ba, c), • [tij(a), tji(b)], • [tij(a), zij(b, c)], where 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, c ∈ R. In the present paper, we prove the following result, which is both terribly much stronger, and much more general than Theorem A and which completely solves [13], Problem 1, for the case of GLn. Theorem 1. Let R be any associative ring with 1, let n ≥ 3, and let A, B be two- sided ideals of R. Then the mixed commutator subgroup [E(n, R, A), E(n, R, B)] is generated as a group by the elements of the form • zij(ab, c) and zij(ba, c), • [tij(a), tji(b)], where 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, c ∈ R. Moreover, for the second type of generators, it suffices to fix one pair of indices (i, j). Let us briefly review the sequence of events that led us to this result. It all started a decade ago with our joint papers with Alexei Stepanov and Roozbeh Hazrat [30, 14, 31], which, in particular, gave three completely different proofs of the following birelative standard commutator formula, under various commutativity conditions1. In turn, this formula generalised a great number of preceding results due to Hyman Bass, Alec Mason and Wilson Stothers, Andrei Suslin, Leonid Vaserstein, Zenon Borewicz and the first author, and many others, see, for instance [2, 19, 18, 25, 27, 3], and a complete bibliography of early papers in [6, 32, 10]. Compare also [7, 8, 9, 13] for a detailed description of the recent work in the area. Theorem B. Let R be a quasi-finite ring with 1, let n ≥ 3, and let A, B be two-sided ideals of R. Then the following birelative standard commutator formula holds [E(n, R, A), GL(n, R, B)] = [E(n, R, A), E(n, R, B)]. 1The third of these proofs was essentially reduction to the absolute case via level calculations, as discovered earlier by Hong You [36], of which we were not aware at the time of writing these papers. RELATIVE AND UNRELATIVE ELEMENTARY GROUPS, REVISITED 3 The condition in the above theorem is very general and embraces very broad class of rings, but some commutativity condition is necessary here, since it is known that the standard commutator formula may fail for general associative rings even in the absolute case, see [5]. Last year the first author noticed that for commutative rings an argument from his paper with Alexei Stepanov [24] implies that the standard commutator formula holds also in the following unrelativised form, see [28], Theorem 1. Theorem C. Let R be a commutative ring with 1, let n ≥ 3, and let A, B be two-sided ideals of R. Then the following commutator formula holds [E(n, A), GL(n, R, B)] = [E(n, A), E(n, B)]. In particular, this immediately implies the following striking equality, see [28], Theorem 2. Theorem D. Let R be a commutative ring with 1, let n ≥ 3, and let A, B be two-sided ideals of R. Then one has [E(n, R, A), E(n, R, B)] = [E(n, A), E(n, B)]. Thereupon, the second author immediately suggested that since everything occurs inside the absolute elementary group E(n, R), one should be able to prove Theorem D directly, by looking at the elementary generators in Theorem A and proving that the third type of generators are redundant. Over commutative rings this was essentially accomplished in the more general context of Chevalley groups in our joint paper [33]. Immediately thereafter we started to work on the unitary sequel [34] and discovered that most of the requisite results, apart from the unitary analogue of Theorem A, hold for arbitrary form rings, without any commutativity conditions. This propted us to look closer inside the proofs of [13], Theorems 3A and 3B, and the Main Lemmas of [33, 34]. We discovered that all references to Theorem B or any of its special cases can be easily replaced by elementary calculations that hold over arbitrary associative rings, and only depend on Steinberg relations (so that they can be carried out already in Steinberg groups). Finally, attempting to strengthen the birelative standard commutator formula in the arithmetic case [29], the first author was forced to look for a stronger version of Theorem 1, with demoted set of generators. But a calculation that procures such a reduction was already contained in the papers on bounded generation, see, for instance, [4, 26]. A similar calculation is hidden also in the proof of the main theorem in the recent preprint by Andrei Lavrenov and Sergei Sinchuk [17]. Observe that, as discovered by Wilberd van der Kallen [16], and amply developed by Stepanov [21, 22], one could reduce also the number of requisite zij(a, c)'s. Since both types of generators in Theorem 1 already belong to [E(n, A), E(n, B)], we get the following generalisation of Theorem D, for arbitrary associative rings. Theorem 2. Let R be any associative ring with 1, let n ≥ 3, and let A, B be two-sided ideals of R. Then one has [E(n, R, A), E(n, R, B)] = [E(n, A), E(n, B)]. 4 N. VAVILOV AND Z. ZHANG In turn, together with Theorem B this last result immediately implies the following very broad generalisation of Theorem C. Theorem 3. Let R be a quasi-finite ring with 1, let n ≥ 3, and let A, B be two-sided ideals of R. Then the following commutator formula holds [E(n, A), GL(n, R, B)] = [E(n, A), E(n, B)]. In § 2 we prove Theorem 1, and thus also Theorems 2 and 3. Finally, in § 3 we establish some further corollaries and variations of these results and make some further related observations. In the present paper we describe part of the astounding recent progress in the di- rection of unrelativisation, whose first steps were presented in our talk "Relativisation and unrelativisation" at the Polynomial Computer Algebra 2019 (see http://pca-pdmi.ru/2019/program, April 19). 2. The proof of Theorem 1 Everywhere below the commutators are left-normed so that for two elements x, y of a group G one has [x, y] = xy · y−1 = xyx−1y−1. In the sequel we repeatedly use standard commutator identities such as [xy, z] = x[y, z] · [x, z] or [x, yz] = [x, y] · y[x, z] without any explicit reference. The following lemma is a classical result due to Stein, Tits and Vaserstein, see [27]. Lemma 1. Let R be an associative ring with 1, n ≥ 3, and let A be a two-sided ideal of R. Then as a subgroup E(n, R, A) is generated by zij(a, c), for all 1 ≤ i 6= j ≤ n, a ∈ A, c ∈ R. Since E(n, R, A) is normal in E(n, R) by the very definition, in particular this lemma implies that every elementary conjugate of zij(a, c) is again a product of generators of the same type. The following result is [15], Lemma 12, a detailed proof is also reproduced in [13], Lemma 2A. Lemma 2. Let R be an associative ring with 1, let n ≥ 3, and let A, B be two- sided ideals of R. Then the mixed commutator subgroup [E(n, R, A), E(n, R, B)] is generated as a group by the elements of the form • xzij(ab, c) and xzij(ba, c), • x[tij(a), tji(b)], • x[tij(a), zij(b, c)], where 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, c ∈ R, and x ∈ E(n, R). Both types of generators in the first item belong to E(n, R, AB +BA) and Lemma 1 implies that for them x can be removed, without affecting the subgroup they gener- ate. The first type of generators listed in Theorem 1 still generate the same group E(n, R, AB + BA). An obvious level calculation (see, for instance, [30], Lemma 3 or [13], Lemma 1A) shows the other two types of generators listed in Theorem A still belong to the RELATIVE AND UNRELATIVE ELEMENTARY GROUPS, REVISITED 5 congruence subgroup GL(n, R, AB + BA). Now, in the conditions of Theorem B, an elementary conjugate of these generators is again the same generator, modulo the subgroup E(n, R, AB + BA). However, it is very easy to get rid of any reference to Theorem B here. Let us start with the second type of generators z = [tij(a), tji(b)]. Lemma 3. Let R be an associative ring with 1, n ≥ 3, and let A, B be two-sided ideals of R. Then for any 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, and any x ∈ E(n, R) the conjugate x[tij(a), tji(b)] is congruent to [tij(a), tji(b)] modulo E(n, R, AB + BA). Proof. Clearly, z = [tij(a), tji(b)] resides in the image of the fundamental embedding of E(2, R) into E(n, R) in the i-th and j-th rows and columns, where one has and z = (cid:20)(cid:18)1 a z−1 = (cid:20)(cid:18)1 0 0 1(cid:19) ,(cid:18)1 0 b 1(cid:19) ,(cid:18)1 a b 1(cid:19)(cid:21) = (cid:18)1 + ab + abab −aba 1 − ba(cid:19) 0 1(cid:19)(cid:21) = (cid:18)1 − ab 1 + ba + baba(cid:19) . −bab bab aba Consider the elementary conjugate xz. We argue by induction on the length of x ∈ E(n, R) in elementary generators. Let x = ytkl(c), where y ∈ E(n, R) is shorter than x, whereas 1 ≤ k 6= l ≤ n, c ∈ R. • If k, l 6= i, j, then tkl(c) commutes with z and can be discarded. • On the other hand, for any h 6= i, j the above formulas for z and z−1 immediately imply that [tih(c), z] = tih(−abc − ababc)tjh(−babc), [thi(c), z] = thi(cab)thj(−caba), [tjh(c), z] = tih(abac)tjh(bac), [thj(c), z] = thi(cbab)thj(−cba − cbaba). All factors on the right hand side belong already to E(n, AB + BA) This means that xz ≡ yz (mod E(n, R, AB + BA)) . • Finally, for (k, l) = (i, j), (j, i) we can take an h 6= i, j and rewrite tkl(c) as a commutator tij(c) = [tih(c), thj(1)] or tji(c) = [th(c), thi(1)] and apply the previous item to get the same congruence modulo E(n, R, AB + BA). By induction we get that xz ≡ z (mod E(n, R, AB + BA)) . (cid:3) Thus, to prove the first claim of Theorem 1 it only remains to establish the following lemma. For commutative rings it is essentially the simplest special case of the Main Lemma of [33]. However, there it is expressed in the language of root elements. Even though commutativity is not used in the proof in any material way, it is formally assumed. For completeness, we reproduce the proof of a somewhat stronger fact, in matrix notation. Lemma 4. Let R be an associative ring with 1, n ≥ 3, and let A, B be two-sided ideals of R. Then for any 1 ≤ i 6= j ≤ n, a ∈ A, b ∈ B, c ∈ R and any x ∈ E(n, R) the conjugate x[tij(a), zij(b, c)] of a generator of the third type is congruent to an 6 N. VAVILOV AND Z. ZHANG elementary conjugate of some generator [tkl(a′), tlk(b′)], 1 ≤ k 6= l ≤ n, a′ ∈ A, b′ ∈ B, of the second type, modulo E(n, R, AB + BA). Proof. Indeed, let z = [tij(a), zij(b, c)]. Take any h 6= i, j. Then z = [tij(a), zij(b, c)] = tij(a) · zij(b,c)tij(−a) = tij(a) · zij (b,c)[tih(1), thj(−a)]. Thus, z = tij(a) · [zij (b,c)tih(1), zij(b,c)thj(−a)] = tij(a) · [tih(1 − bc)tjh(−cbc), thi(−acbc)thj(−a(1 − cb))] = tij(a) · [tih(1)u, thj(−a)v], where Thus, u = tjh(−cbc)tih(−bc) ∈ E(n, B), v = thi(−acbc)thj(acb) ∈ E(n, AB). z ≡ tij(a) · [tih(1)u, thj(−a)] (mod E(n, R, AB + BA)) . On the other hand, tij(a) · [tih(1)u, thj(−a)] = tij(a) · tih(1)[u, thj(−a)] · tij(−a), whereas [u, thj(−a)] = tjh(−cbc)tih(bca) · [tjh(−cbc), thj(−a)] ≡ [tjh(−cbc), thj(−a)] (mod E(n, R, AB + BA)) . Summarising the above, we see that xz ≡ xtij (a)tih(1)[tjh(−cbc), thj(−a)] (mod E(n, R, AB + BA)) , where [tjh(−cbc), thj(−a)] is the second type generator, as claimed. (cid:3) At this point we have already established the first claim of Theorem 1 -- and thus also Theorems 2 and 3. The rest is a bonus, that we need for more sophisticated applications. The proof of the final claim of Theorem 1 is in fact a refinement of the proof of [29], Theorem 3. Again, formally commutativity was assumed there, but can be easily circumvented. Lemma 5. Let R be an associative ring with 1, n ≥ 3, and let A, B be two-sided ideals of R. Then for any 1 ≤ i 6= j ≤ n, any 1 ≤ k 6= l ≤ n, and a ∈ A, b ∈ B, the elementary commutator [tij(a), tji(b)] is congruent to [tkl(a), tlk(b)] modulo E(n, R, AB + BA). Proof. Take any h 6= i, j and rewrite the elementary commutator z = [tij(a), tji(b)] as z = tij(a) · tji(b)tij(−a) = tij(a) · tji(b)[tih(a), thj(−1)]. Expanding the conjugation by tji(b), we see that z = tij(a) · [tji(b)tih(a), tji(b)thj(−1)] = tij(a) · [tjh(ba)tih(a), thj(−1)thi(b)]. RELATIVE AND UNRELATIVE ELEMENTARY GROUPS, REVISITED 7 Now, the first factor tjh(ba) of the first argument in this last commutator already belongs to the group E(n, BA) which is contained in E(n, R, AB + BA). Thus, as above, z ≡ tij(a) · [tih(a), thj(−1)thi(b)] (mod E(n, R, AB + BA)) . Using multiplicativity of the commutator w.r.t. the second argument, cancelling the first two factors of the resulting expression, and then applying Lemma 3 we see that z ≡ thj (−1)[tih(a), thi(b)] ≡ [tih(a), thi(b)] (mod E(n, R, AB + BA)) . Similarly, rewriting the commutator z differently, as z = [tij(a), tji(b)] = tij (a)tji(b) · tji(−b) = tij (a)[tjh(b), thi(1)] · tji(−b), we get the congruence z ≡ [thj(a), tjh(b)] (mod E(n, R, AB + BA)) . Obviously, for n ≥ 3 we can pass from any position (i, j), i 6= j, to any other such (cid:3) position (k, l), k 6= l, by a sequence of at most three such elementary moves. This finishes the proof of Theorem 1. 3. Further variations and final remarks The following result is a generalisation of the unrelative normality theorem by Bogdan Nica, see [20], Theorem 2, which pertained to the commutative case. It is an immediate corollary of our Theorem 3. Theorem 4. Let R be a quasi-finite ring with 1, let n ≥ 3, and let A be a two sided ideal of R. Then E(n, A) is normal in GL(n, R, A). Let us mention another amazing corollary of Theorem 1, in the style of stability results without stability conditions by Tony Bak, see [1]. With this end, observe that Lemma 3 implies that the quotient [E(n, A), E(n, B)]/E(n, R, AB + BA) is central in E(n, R)/E(n, R, AB + BA). In other words, the following holds. Lemma 6. Let R be an associative ring with 1, n ≥ 3, and let A, B be two-sided ideals of R. Then [[E(n, A), E(n, B)], E(n, R)] = E(n, R, AB + BA). But now Theorem 1 implies surjective stability of such quotients, which is a gener- alisation of the first half of [13], Lemma 15, to arbitrary associative rings, without any stability conditions, or commutativity conditions. Indeed, in view of Theorem 1 and Lemma 6 as a normal subgroup of E(n, R) the group [E(n, A), E(n, B)] is generated by [E(3, A), E(3, B)]. This can be restated as follows. 8 N. VAVILOV AND Z. ZHANG Theorem 5. Let R be any associative ring with 1, and let A and B be two sided ideals of R. Then for all n ≥ 3 the stability map [E(n, A), E(n, B)]/E(n, R, AB + BA) −→ [E(n + 1, A), E(n + 1, B)]/E(n + 1, R, AB + BA) is surjective. This quotient occurs surprisingly often in seemingly unrelated problems, and is so interesting in itself, that we are now hatching the idea to definitively comprehend its structure. Lemmas 3 and 5 assert that some elementary commutators and their conjugates are congruent modulo E(n, R, AB + BA). We have several further results in the same spirit. For instance, one has • [tij(ac), tji(b)] ≡ [tij(a), tji(cb)] (mod E(n, R, AB + BA)) , • [tij(a1 + a2), tji(b)] ≡ [tij(a1), tji(b)] · [tij(a2), tji(b)] (mod E(n, R, AB + BA)) , • [tij(a), tji(b1 + b2)] ≡ [tij(a), tji(b1)] · [tij(a), tji(b2)] (mod E(n, R, AB + BA)) , • [tij(a), tji(b)]−1 ≡ [tij(−a), tji(b)] ≡ [tij(a), tji(−b)] (mod E(n, R, AB + BA)) , • [tij(a1), tji(b)] ≡ [tij(a2), tji(b)] (mod E(n, R, AB + BA)) , if a1 ≡ a2 (mod AB + BA + A2) • [tij(a), tji(b1)] ≡ [tij(a), tji(b1)] (mod E(n, R, AB + BA)) , if b1 ≡ b2 (mod AB + BA + B2) etc. We have not made any attempt to systematically collect all such congruences in the present article, since they are not directly needed to prove Theorem 1. But they may turn out very useful to control the quotient [E(n, A), E(n, B)]/E(n, R, AB+BA). Observe that by Lemma 6 this quotient is central in E(n, R)/E(n, R, AB + BA), and thus, in particular, it is itself abelian. We intend to list all such properties in a subsequent paper, where we propose to assail the following tantalising problem. Problem 1. Give a presentation of [E(n, A), E(n, B)]/ EE(n, R, AB + BA) by generators and relations. Let us mention yet another corollary of Theorem 1. Let U(n, R) and U −(n, R) be the groups of upper unitriangular and lower unitriangular matrices, respectively. These are unipotent radicals of the standard Borel subgroup, and its opposite Borel subgroup. Further, set U(n, I) = U(n, R) ∩ GL(n, R, I), U −(n, I) = U −(n, R) ∩ GL(n, R, I). In [28] we considered another birelative group EE(n, A, B) = hU(n, A), U −(n, B)i, and established that for commutative rings this group contains [E(n, A), E(n, B)], see [28], Theorem 3. Since in the case EE(n, A, B) contains E(n, R, AB) by [29], Lemma 8, this theorem immediately follows from our Theorem 1. It is natural to expect that an analogue of this result holds over arbitrary associative rings. RELATIVE AND UNRELATIVE ELEMENTARY GROUPS, REVISITED 9 Problem 2. Let R be any associative ring with 1, let n ≥ 3, and let A, B be two-sided ideals of R. Prove that [E(n, A), E(n, B)] ≤ EE(n, A, B). The difficulty now is exactly to prove that E(n, R, AB + BA) ≤ EE(n, A, B). In the non-commutative case the argument used in the proof of [29], Lemma 8, only works in the following form. If i, j 6= n, there exists an h > i, j so that one can express tij(ab) as tij(ab) = [tih(a), thj(b)] ∈ [U(n, A), U −(n, B)], and conclude that zij(ab, c) ∈ EE(n, A, B). Similarly, i, j 6= 1, there exists an h < i, j so that one can express tij(ba) as tij(ba) = [tih(b), thj(a)] ∈ [U −(n, B), U(n, A)], and conclude that zij(ba, c) ∈ EE(n, A, B). In the case of commutative rings (or, more generally, when AB = BA) this implies that E(n, R, AB + BA) ∈ EE(n, A, B), see [16, 21, 22]. But in the general case this would require some additional reasoning. Let us mention an even more challenging related question. Namely, let P be a proper standard parabolic subgroup of GL(n, R). We can define the corresponding subgroup of EE(n, A, B) as follows: EEP (n, A, B) = hUP (A), U − P (B)i, where UP (A) and U − P (B) are the intersections of U(n, A) and U −(n, B) with the unipotent radicals UP and U − P of P and its opposite standard parabolic P −, respec- tively. In the definition of EE(n, A, B) itself P = B(n, R) is the standard Borel subgroup. However, in many cases it is technically much more expedient to work with the maximal standard parabolics instead, see, for instance, the works by Alexei Stepanov [21, 22]. Problem 3. Let R be any associative ring with 1, let n ≥ 3, and let A, B be two-sided ideals of R. Prove that [E(n, A), E(n, B)] ≤ EEP (n, A, B). The next problem proposes to generalise [9], Theorem 8A, and [13], Theorem 5A, from quasi-finite rings, to arbitrary associative rings. In other words, to prove that any multiple commutator of relative or unrelative elementary subgroups is equal to some double such commutator, see [15, 9, 12, 13, 23] Here A ◦ B = AB + BA stands for the symmetrised product of two sided ideals A and B. In general, the symmetrised product is not associative. Thus, when writing something like A ◦ B ◦ C, we have to specify the order in which products are formed. for notation pertaining to multiple commutators. Let G be a group and H1, . . . , Hm ≤ G be its subgroups. There are many ways to form a higher commutator of these groups, depending on where we put the brack- ets. Thus, for three subgroups F, H, K ≤ G one can form two triple commutators 10 N. VAVILOV AND Z. ZHANG [[F, H], K] and [F, [H, K]]. Usually, we write [H1, H2, . . . , Hm] for the left-normed commutator, defined inductively by [H1, . . . , Hm−1, Hm] = [[H1, . . . , Hm−1], Hm]. To stress that here we consider any commutator of these subgroups, with an arbitrary placement of brackets, we write [[H1, H2, . . . , Hm]]. Thus, for instance, [[F, H, K]] refers to any of the two arrangements above. Actually, a specific arrangment of brackets usually does not play major role in our results -- apart from one important attribute. Namely, what will matter a lot is the position of the outermost pairs of inner brackets. Namely, every higher commutator subgroup [[H1, H2, . . . , Hm]] can be uniquely written as [[H1, H2, . . . , Hm]] = [[[H1, . . . , Hh]], [[Hh+1, . . . , Hm]]], for some h = 1, . . . , m − 1. This h will be called the cut point of our multiple commutator. Problem 4. Let R be any associative ring with 1, let n ≥ 3, and let Ai E R, i = 1, . . . , m, be two-sided ideals of R. Consider an arbitrary arrangment of brackets [[. . .]] with the cut point h. Then one has [[E(n, I1), E(n, I2), . . . , E(n, Im)]] = [E(n, I1 ◦ . . . ◦ Ih), E(n, Ih+1 ◦ . . . ◦ Im)], where the bracketing of symmetrised products on the right hand side coincides with the bracketing of the commutators on the left hand side. Observe that Theorem A and its analogues were used by Alexei Stepanov in his remarkable results on bounded width of commutators with respect to elementary generators, see [23], and our survey [8]. Now, it would be natural to refer in these results to our new reduced set of generators from Theorem 1. Analogues of our Theorems 1 and 2 hold for Bak's unitary groups over arbitrary form rings. In particular, this generalises [12], Theorem 9 and [13], Theorem 3B. Also, it solves [13], Problem 1 for the unitary case. These results are now incorporated in our unitary paper [34]. A full analogue of Theorem 1 for Chevalley groups is much more difficult even in the commutative case, and will be published in [35]. The authors thank Roozbeh Hazrat and Alexei Stepanov for ongoing discussion of this circle of ideas, and long-standing cooperation over the last decades. Also, we are grateful to Pavel Gvozdevsky and Sergei Sinchuk for their extremely pertinent questions during the seminar talk, where the first author was reporting [29]. Last, but not least, the first author thanks Nikolai Vasiliev for his insistence and friendly encouragement. References [1] A. Bak, Non-abelian K-theory: The nilpotent class of K1 and general stability, K -- Theory 4 (1991), 363 -- 397. [2] H. Bass, K-theory and stable algebra, Inst. Hautes ´Etudes Sci. Publ. Math. (1964), no. 22, 5 -- 60. [3] Z. I. Borewicz, N. A. Vavilov, The distribution of subgroups in the full linear group over a commutative ring, Proc. Steklov Inst. Math. 3 (1985), 27 -- 46. RELATIVE AND UNRELATIVE ELEMENTARY GROUPS, REVISITED 11 [4] D. Carter, G. E. Keller, Bounded elementary generation of SLn(O), Amer. J. Math. 105 (1983), 673 -- 687. [5] V. N. Gerasimov, Group of units of a free product of rings, Math. U.S.S.R. Sb., 134 (1989), no. 1, 42 -- 65. [6] A. J. Hahn, O. T. O'Meara, The classical groups and K-theory, Springer, Berlin et al., 1989. [7] R. Hazrat, A. Stepanov, N. Vavilov, Zuhong Zhang, The yoga of commutators, J. Math. Sci. 179 (2011), no. 6, 662 -- 678. [8] R. Hazrat, A. Stepanov, N. Vavilov, Zuhong Zhang, Commutator width in Chevalley groups, Note di Matematica 33 (2013), no. 1, 139 -- 170. [9] R. Hazrat, A. Stepanov, N. Vavilov, Zuhong Zhang, The yoga of commutators, further applica- tions, J. Math. Sci. 200 (2014), no. 6, 742 -- 768. [10] R. Hazrat, N. Vavilov, Bak's work on K-theory of rings (with an appendix by Max Karoubi ) J. K-Theory 4 (2009), no. 1, 1 -- 65. [11] R. Hazrat, N. Vavilov, Zuhong Zhang Generation of relative commutator subgroups in Chevalley groups, Proc. Edinburgh Math. Soc., 59, (2016), 393 -- 410. [12] R. Hazrat, N. Vavilov, Z. Zhang, Multiple commutator formulas for unitary groups. Israel J. Math., 219 (2017), 287 -- 330. [13] R. Hazrat, N. Vavilov, Zuhong Zhang The commutators of classical groups, J. Math. Sci., 222 (2017), no. 4, 466 -- 515. [14] R. Hazrat, Zuhong Zhang, Generalized commutator formula, Commun. Algebra, 39 (2011), no. 4, 1441 -- 1454. [15] R. Hazrat, Zuhong Zhang, Multiple commutator formula, Israel J. Math., 195 (2013), 481 -- 505. [16] W. van der Kallen, A group structure on certain orbit sets of unimodular rows, J. Algebra 82 (1983), 363 -- 397. [17] A. Lavrenov, S. Sinchuk A Horrocks-type theorem for even orthogonal K2, arXiv:1909.02637 v1 [math.GR] 5 Sep 2019, pp. 1 -- 23. [18] A. W. Mason, On subgroups of GL(n, A) which are generated by commutators. II, J. reine angew. Math., 322 (1981), 118 -- 135. [19] A. W. Mason, W. W. Stothers, On subgroups of GL(n, A) which are generated by commutators, Invent. Math., 23 (1974), 327 -- 346. [20] B. Nica, A true relative of Suslin's normality theorem, Enseign. Math., 61 (2015), no. 1 -- 2, 151 -- 159. [21] A. Stepanov, Elementary calculus in Chevalley groups over rings, J. Prime Res. Math., 9 (2013), 79 -- 95. [22] A. V. Stepanov, Non-abelian K-theory for Chevalley groups over rings, J. Math. Sci., 209 (2015), no. 4, 645 -- 656. [23] A. Stepanov, Structure of Chevalley groups over rings via universal localization, J. Algebra, 450 (2016), 522 -- 548. [24] A. Stepanov, N. Vavilov, Decomposition of transvections: a theme with variations, K-Theory, 19 (2000), no. 2, 109 -- 153. [25] A. A. Suslin, The structure of the special linear group over polynomial rings, Math. USSR Izv., 11 (1977), no. 2, 235 -- 253. [26] O. I. Tavgen, Bounded generation of Chevalley groups over rings of S-integer algebraic num- bers, Izv. Acad. Sci. USSR, 54 (1990), no.1, 97 -- 122. [27] L. N. Vaserstein, On the normal subgroups of the GLn of a ring, Algebraic K-Theory, Evanston 1980, Lecture Notes in Math., vol. 854, Springer, Berlin et al., 1981, pp. 454 -- 465. [28] N. Vavilov, Unrelativised standard commutator formula, Zapiski Nauchnyh Seminarov POMI. 470 (2018), 38 -- 49. [29] N. Vavilov, Commutators of congruence subgroups in the arithmetic case, Zapiski Nauchnyh Seminarov POMI. 479 (2019), 5 -- 22. 12 N. VAVILOV AND Z. ZHANG [30] N. A. Vavilov, A. V. Stepanov, Standard commutator formula, Vestnik St. Petersburg State Univ., Ser. 1, 41 (2008), no. 1, 5 -- 8. [31] N. A. Vavilov, A. V. Stepanov, Standard commutator formulae, revisited, Vestnik St. Petersburg State Univ., Ser.1, 43 (2010), no. 1, 12 -- 17. [32] N. A. Vavilov, A. V. Stepanov, Linear groups over general rings I. Generalities, J. Math. Sci., 188 (2013), no. 5, 490 -- 550. [33] N. Vavilov, Z. Zhang, Generation of relative commutator subgroups in Chevalley groups. II, Proc. Edinburgh Math. Soc., (2019), 1 -- 16. [34] N. Vavilov, Z. Zhang, Unrelativised commutator formulas for unitary groups, 2019, 1 -- 21. [35] N. Vavilov, Z. Zhang, Commutators of relative and unrelative elementary subgroups, in Cheval- ley groups, 2019, 1 -- 23. [36] Hong You, On subgroups of Chevalley groups which are generated by commutators, J. Northeast Normal Univ., (1992), no. 2, 9 -- 13. Department of Mathematics and Mechanics, St. Petersburg State University, St. Petersburg, Russia E-mail address: [email protected] Department of Mathematics, Beijing Institute of Technology, Beijing, China E-mail address: [email protected]
1210.5061	2	1210	2015-01-06T14:50:55	Determinants for nxn matrices and the symmetric Newton formula in the 3x3 case	[ "math.RA" ]	One of the aims of this paper is to provide a short survey on the Z2-graded, the symmetric and the left (right) generalizations of the classical determinant theory for square matrices with entries in an arbitrary (possibly non-commutative) ring. This will put us in a position to give a motivation for our main results. We use the preadjoint matrix to exhibit a general trace expression for the symmetric determinant. The symmetric version of the classical Newton trace formula is also presented in the 3x3 case.	math.RA	math	Determinants for n × n matrices and the symmetric Newton formula in the 3 × 3 case J. Szigeti and L. van Wyk Abstract. One of the aims of this paper is to provide a short survey on the Z2-graded, the symmetric and the left (right) generalizations of the classical determinant theory for square matrices with entries in an arbitrary (possibly non-commutative) ring. This will put us in a position to give a motivation for our main results. We use the preadjoint matrix to exhibit a general trace ex- pression for the symmetric determinant. The symmetric version of the classical Newton trace formula is also presented in the 3 × 3 case. 1. INTRODUCTION The universal notion of a determinant has a long history. Determinants of ma- trices with entries in non-commutative rings have been considered by many math- ematicians, among them Cayley, Study, Ore, Dieudonn´e and others. An excellent recent survey dealing with almost all existing determinants is [GGRW], where the so called Gelfand-Retakh quasideterminants are used as a main organizing tool. It seems that the symmetric determinant does not fit into the general frame- work presented in [GGRW]. One of the aims of this paper is to provide a short introduction to the symmetric and the corresponding left and right versions of the classical determinant theory. The natural symmetrization of the determinant for- mula and of the adjoint matrix lead to extremely useful concepts. It turns out that these constructions can serve as a starting point of a new symmetric determinant theory for square matrices over an arbitrary ring. The most important feature of this theory is that it can be used to solve systems of left (or right) linear equations and to exhibit left and right Cayley-Hamilton identities for matrices over a Lie nilpotent ring (see [Sz1, Sz2, Sz3, SzT]). 2010 Mathematics Subject Classification. 15A15,15A24,15B33,16S50. Key words and phrases. Z2-graded, symmetric, left (right) determinants and characteristic polynomials, Cayley-Hamilton identities, the symmetric 3 × 3 Newton trace formula. The first author was supported by OTKA K-101515 of Hungary and by the TAMOP-4.2.1.B- 10/2/KONV-2010-0001 project with support by the European Union, co-financed by the European Social Fund. The second author was supported by the National Research Foundation of South Africa under Grant No. UID 72375. Any opinion, findings and conclusions or recommendations expressed in this material are those of the authors and therefore the National Research Foundation does not accept any liability in regard thereto. The authors thank P. N. Anh, L. Marki and J. H. Meyer for fruitful consultations. 1 2 J. SZIGETI AND L. VAN WYK The algebra of n × n matrices over an exterior (Grassmann) algebra E is our basic example. The Lie nilpotent property of E plays a central role in the various applications of the symmetric and the corresponding left and right determinants. The so called (and not widely known) Z2-graded determinant (in [SSz]) is also designed to "attack" matrices over an exterior algebra. Its construction heavily depends on the natural Z2-grading E = E0 ⊕ E1. The Z2-graded determinant (and adjoint) can be used to give an explicit inverse formula and to exhibit left and right Cayley-Hamilton identities for an n × n matrix over E. The "misterious" superdeterminant of a supermatrix due to Kantor and Trishin (in [KT]) is another concept which is closely related to the Z2-grading of the exterior algebra. The treatment in [KT] leads to the solution of certain special systems of left (or right) linear equations and to an invariant Cayley-Hamilton identity for supermatrices. Unfortunately, the lack of complete understanding prevents us from dealing with the KT-superdeterminant. ab map satisfying certain natural rules. Here E× Using the fact that E is a local ring, the Dieudonn´e determinant is a well defined n=1GLn(E) −→ E× ∪∞ ab denotes the Abelianized multiplicative group of units in E (see [Ros]). The Dieudonn´e determinant is an important tool in algebraic K-theory, but we cannot use it to solve systems of linear equations (over a local ring) and to derive Cayley-Hamilton identities. In the rest of this introductory section we try to explain why we restrict our attention to the matrix algebras Mn(E) and Mn,t(E). The Cayley-Hamilton theorem and the corresponding trace identity play a fun- damental role in proving classical results about the polynomial and trace identities of the n × n matrix algebra Mn(K) over a field K (see [Dr, DrF, Row1, Row2]). In case of char(K) = 0, Kemer's pioneering work (see [Ke]) on the T-ideals of asso- ciative algebras revealed the importance of the identities satisfied by the full n × n matrix algebra Mn(E) and by the algebra of (n, t) supermatrices Mn,t(E), where E = K hv1, v2, ..., vi, ... vivj + vjvi = 0 for all 1 ≤ i ≤ ji is the exterior (Grassmann) algebra generated by the infinite sequence of anticom- mutative indeterminates (vi)i≥1. Let K hx1, x2, . . . , xi, . . .i denote the polynomial K-algebra generated by the infinite sequence x1, x2, . . . , xi, . . . of non-commuting indeterminates. The prime T-ideals of this (free associative K-)algebra are exactly the T-ideals of the identities satisfied by Mn(K) for n ≥ 1. The T-prime T-ideals are the prime T-ideals plus the T-ideals of the identities of Mn(E) for n ≥ 1 and of Mn,t(E) for n − 1 ≥ t ≥ 1. Another remarkable result is that for a sufficiently large n ≥ 1, any T-ideal contains the T-ideal of the identities satisfied by Mn(E). Accordingly, the importance of matrices (and supermatrices) over certain non- commutative rings is an evidence in the theory of PI-rings, nevertheless this fact has been obvious for a long time in other branches of algebra (e.g. in the structure theory of semisimple rings). Thus the algebras Mn(E) and Mn,t(E) served as the main motivation for the development of the symmetric and the Z2-graded determinants. In Section 2 we deal with the Study determinant of a quaternionic matrix (see [A, St]) and we show why a similar embedding approach does not work for Mn(E). The main result in Section 2 is based on an embedding of the two generated exterior algebra E(2) into a 2×2 matrix algebra over a commutative ring and gives a Cayley- Hamilton identity of degree 2n in Mn(E(2)). Section 3 is devoted to a simplified DETERMINANTS FOR n × n MATRICES 3 version of the Z2-graded determinant. In Section 4 we present a short introduc- tion to the symmetric theory of determinants, we collect and explain some known results and point out some similarities and differences between the (traditional) commutative base ring case and the general case. We do not intend to give a detailed study of the relationships between the symmetric and any of the already existing determinant notions. Nevertheless, a thorough comparison between the symmetric and the Z2-graded determinant would be essential. For a 2 × 2 matrix over E, the second right (left) determinant is the double of the right (left) Z2-graded determinant. A similar comparison in the 3 × 3 case would probably require computer calculations. The treatment in Section 4 puts us in a position to give a motivation for our new results in Sections 5 and 6. First we prove that sdet(A) = tr(AA∗) = tr(A∗A), where sdet(A) is the symmetric determinant, tr(A) is the sum of the diagonal entries and A∗ is the so called preadjoint matrix of the n × n matrix A ∈ Mn(R). Then we present the following symmetric version of the Newton trace formula for a 3 × 3 matrix A ∈ M3(R): sdet(A) = tr3(A)−tr(A)·tr(A2)−tr(A·tr(A)·A)−tr(A2)·tr(A)+tr(A3)+tr(cid:0)(A⊤)3(cid:1) , where A⊤ denotes the transpose of A. The symmetric characteristic polynomial of this A and the corresponding general Cayley-Hamilton identity are also presented by traces. 2. THE EMBEDDING OF H AND THE STUDY DETERMINANT There are well known embeddings of the complex number field C and of the skew field H of the real quaternions into matrices: a + bi 7−→(cid:20) a −b a (cid:21) , a + bi + cj + dk = a + bi + (c + di)j 7−→(cid:20) a + bi −c + di a − bi (cid:21) . c + di b The above definitions provide injective R-algebra homomorphisms µ : C → M2(R) and ν : H → M2(C). There is a natural µ2 : M2(C) → M2(M2(R)) extension of µ: µ2(cid:18)(cid:20) z1,1 z2,1 z1,2 z2,2 (cid:21)(cid:19) =(cid:20) µ(z1,1) µ(z1,2) µ(z2,1) µ(z2,2) (cid:21) . Since M2(M2(R)) ∼= M4(R), the composition ϑ = µ2 ◦ ν : H −→ M4(R) is the following map: b c a a −d −b −c d −d −c d c a −b b a .   a + bi + cj + dk 7−→  Using the natural extensions νn : Mn(H) −→ Mn(M2(C)) ∼= M2n(C) and ϑn : Mn(H) −→ Mn(M4(R)) ∼= M4n(R), an n × n matrix over H can be viewed as a 2n × 2n matrix over C or as a 4n × 4n matrix over R. Now we can define the Study determinant of a quater- nionic matrix A ∈ Mn(H) as the ordinary determinant S det(A) = detC νn(A) in M2n(C). If we take the absolute value of the complex number detC νn(A), then detC νn(A)2 = detR ϑn(A). The Study determinant has some nice properties and it is used frequently in differential geometry and Lie theory. The Cayley-Hamilton 4 J. SZIGETI AND L. VAN WYK identity for ϑn(A) yields the same identity (with real coefficients) of degree 4n for A itself. A similar approach to get a useful determinant notion and a Cayley-Hamilton identity in Mn(E) is impossible. The reason is that the infinitely generated exterior algebra E cannot be embedded into a full matrix algebra over a commutative ring (E does not satisfy any of the standard identities). On the other hand the embedding approach gives the following: Theorem 2.1. Let A ∈ Mn(E(2)) be an n × n matrix over the two generated exterior algebra Then A satisfies a Cayley-Hamilton identity of the form E(2) = K(cid:10)v1, v2 v1v2 = −v2v1, v2 1 = v2 2 = 0(cid:11) . A2n + c2n−1A2n−1 + · · · + c1A + c0I = 0 where ci ∈ K, 0 ≤ i ≤ 2n − 1. Proof. The assignments 1 7−→(cid:20) 1 0 0 1 (cid:21) , v1 7−→(cid:20) x 0 −x (cid:21) , v2 7−→(cid:20) x y 0 −2y −y (cid:21) define a K-embedding ε : E(2) −→ M2(K [x, y] /(x2, y2)), where (x2, y2) ⊳ K [x, y] is the ideal generated by the monomials x2 and y2. Now consider the induced K-embedding εn : Mn(E(2)) −→ Mn(M2(K [x, y] /(x2, y2))) ∼= M2n(K [x, y] /(x2, y2)). The trace of any 2 × 2 block ε(b0 + b1v1 + b2v2 + b3v1v2) = (cid:20) b0 + b1x + b2y − b3xy + (x2, y2) −2b2y + 2b3xy + (x2, y2) b1x − b3xy + (x2, y2) b0 − b1x − b2y + b3xy + (x2, y2) (cid:21) in εn(A) is of the form 2b0 + (x2, y2). Since the trace of εn(A) is the sum of the traces of the diagonal 2×2 blocks, we have tr(εn(A)) = 2b+(x2, y2) for some b ∈ K. The coefficients of the characteristic polynomial of εn(A) are rational polynomial expressions (with zero constant terms) of the traces tr((εn(A))k) = tr(εn(Ak)), k ≥ 1 (Newton formulae). Thus the Cayley-Hamilton identity for εn(A) is of the form with ci ∈ K, 0 ≤ i ≤ 2n − 1. It follows that (εn(A))2n +(cid:0)c2n−1 + (x2, y2)(cid:1) (εn(A))2n−1 + · · · +(cid:0)c1 + (x2, y2)(cid:1) εn(A) +(cid:0)c0 + (x2, y2)(cid:1) I = 0 εn(A2n + c2n−1A2n−1 + · · · + c1A + c0I) = εn(A2n) + c2n−1εn(A2n−1) + · · · + c1εn(A) + c0I = 0 holds in M2n(K [x, y] /(x2, y2)), and thus the injectivity of εn gives the desired identity. (cid:3) 3. THE Z2-GRADED DETERMINANT A Z2-grading of an (associative) ring R is a pair (R0, R1), where R0 and R1 are additive subgroups of R such that R = R0 ⊕R1 and RiRj ⊆ Ri+j for all i, j ∈ {0, 1} DETERMINANTS FOR n × n MATRICES 5 and i + j is taken modulo 2. The relation R0R0 ⊆ R0 ensures that R0 is a subring of R. It is easy to see that the existence of 1 ∈ R implies that 1 ∈ R0. A Z2-grading (R0, R1) of the ring R is called central if R0 ⊆ Z(R) (here Z(R) denotes the centre of R). The condition R0 ⊆ Z(R) implies the Lie nilpotence (of index 2) of R. The general notion of the Z2-graded determinant (in [SSz]) is defined for an n × n matrix over an arbitrary ring R with a central Z2-grading (R0, R1). In order to present a more natural and understandable treatment of the Z2- graded determinant, we restrict ourselves to the case of the well known central Z2-grading E = E0 ⊕ E1 of the (infinitely generated) exterior algebra. If we add one more (anticommutative) generator w to the infinite sequence (vi)i≥1, we obtain an extended Ew = K(cid:10)w, v1, v2, ..., vi, ... w2 = wvj + vj w = vivj + vj vi = 0 for all 1 ≤ i ≤ j(cid:11) exterior algebra. An n × n matrix A ∈ Mn(E) can be uniquely written as A = A0 + A1 with A0 ∈ Mn(E0) and A1 ∈ Mn(E1). The companion matrix of A is defined as A0 + A1w ∈ Mn(Ew(0)), where the notations Ew(0) = (Ew)0 and Ew(1) = (Ew)1 are used for the even and the odd part of the (central) Z2-grading Ew = (Ew)0 ⊕ (Ew)1 of Ew. Since Ew(0) is commutative, the ordinary determinant and the ordinary adjoint of A0 + A1w are defined and can be written as det(A0 + A1w) = d0 + d1w ∈ Ew(0) and adj(A0 + A1w) = B0 + B1w ∈ Mn(Ew(0)), where d0 ∈ E0, d1 ∈ E1, B0 ∈ Mn(E0), B1 ∈ Mn(E1) and each of these objects is uniquely determined by A. Clearly, d0 = det(A0), B0 = adj(A0) and the elements d1, b(1) i,j (note that A0 = [a(0) i,j ∈ E1 are also polynomial expressions of the entries a(0) i,j and a(1) i,j ], A1 = [a(1) i,j ] and B1 = [b(1) i,j ]). Theorem 3.1. The elements of the product matrices A(B0 + B1) = (A0 + A1)(B0 + B1) and (B0 + B1)A = (B0 + B1)(A0 + A1) are contained in the subring E0[d1] of E generated by d1 and the elements of E0, namely: A(B0 + B1), (B0 + B1)A ∈ Mn(E0[d1]). The containment E0 ⊆ Z(E) implies that the subring E0[d1] ⊆ E is commu- tative (the elements of E0[d1] are polynomials of d1 with coefficients in E0). As a consequence of Theorem 3.1 the determinant and the adjoint of the matrices A(B0 + B1), (B0 + B1)A ∈ Mn(E0[d1]) are defined. We call rgdet(A) = det(A(B0 + B1)) the right Z2-graded determinant and rgadj(A) = (B0 + B1)adj(A(B0 + B1)) the right Z2-graded adjoint (with respect to E = E0 ⊕ E1) of the matrix A ∈ Mn(E). Since A(B0 + B1)adj(A(B0 + B1)) = det(A(B0 + B1))I in Mn(E0[d1]), we immediately obtain (in Mn(E)) that: Argadj(A) = rgdet(A)I. 6 J. SZIGETI AND L. VAN WYK Proposition 3.2. (i) If T ∈ GLn(E0) is an invertible matrix and A ∈ Mn(E), then rgdet(T AT −1) = rgdet(A) and rgadj(T AT −1) = T (rgadj(A))T −1. (ii) If A ∈ Mn(E0), then rgdet(A) = (det(A))n and rgadj(A) = (det(A))n−1adj(A). The polynomial ring E[t] inherits a natural (and central) Z2-grading E[t] = E0[t] ⊕ E1[t] from E = E0 ⊕ E1. We define the right Z2-graded characteristic polynomial of a matrix A ∈ Mn(E) as the right Z2-graded determinant (with respect to E[t] = E0[t] ⊕ E1[t]) of the matrix tI − A ∈ Mn(E[t]), where I is the identity matrix in Mn(E): χA(t) = rgdet(tI − A) = λ0 + λ1t + · · · + λktk ∈ E[t], λ0, λ1, ..., λk ∈ E and λk 6= 0. Since GLn(E0) ⊆ GLn(E0[t]), an immediate consequence of Proposition 3.2 is that χT AT −1(t) = χA(t) for any invertible matrix T ∈ GLn(E0). Proposition 3.3. If χA(t) = λ0 + λ1t + · · · + λktk is the right Z2-graded char- acteristic polynomial of the n × n matrix A ∈ Mn(E), then k = n2 and λn2 = 1, λ0 = rgdet(−A). Theorem 3.4. If χA(t) ∈ E[t] is the right Z2-graded characteristic polynomial of an n × n matrix A ∈ Mn(E) and h(t) ∈ E[t] is arbitrary, then the left substitution of A into the product polynomial χA(t)h(t) = µ0 + µ1t + · · · + µmtm is zero: Iµ0 + Aµ1 + · · · + Amµm = 0. 4. THE SYMMETRIC AND THE RIGHT (LEFT) DETERMINANTS Let Sn denote the symmetric group of all permutations of the set {1, 2, . . . , n}. For an n × n matrix A = [ai,j] over an arbitrary (possibly non-commutative) ring or algebra R with 1, the element sdet(A) = Xτ,ρ∈Sn sgn(ρ)aτ (1),ρ(τ (1)) · · · aτ (t),ρ(τ (t)) · · · aτ (n),ρ(τ (n)) sgn(α)sgn(β)aα(1),β(1) · · · aα(t),β(t) · · · aα(n),β(n) = Xα,β∈Sn of R can be obviously considered as the symmetric determinant of A. The preadjoint matrix A∗ = [a∗ r,s] of an n × n matrix A = [ai,j] (over an arbi- trary ring or algebra R with 1) is defined as the following natural symmetrization of the classical adjoint: sgn(ρ)aτ (1),ρ(τ (1)) · · · aτ (s−1),ρ(τ (s−1))aτ (s+1),ρ(τ (s+1)) · · · aτ (n),ρ(τ (n)) a∗ r,s =Xτ,ρ =Xα,β sgn(α)sgn(β)aα(1),β(1) · · · aα(s−1),β(s−1)aα(s+1),β(s+1) · · · aα(n),β(n) , where the first sum is taken over all τ, ρ ∈ Sn with τ (s) = s and ρ(s) = r (the second sum is taken over all α, β ∈ Sn with α(s) = s and β(s) = r). We note that the (r, s) entry of A∗ is exactly the signed symmetric determinant (−1)r+ssdet(As,r) of the (n − 1) × (n − 1) minor As,r of A arising from the deletion of the s-th row and the r-th column of A. If R is commutative, then A∗ = (n − 1)!adj(A), where adj(A) denotes the ordinary adjoint of A. The right adjoint sequence (Pk)k≥1 of A is defined by the recursion: P1 = A∗ and Pk+1 = (AP1 · · · Pk)∗ for k ≥ 1. Originally the k-th right determinant was DETERMINANTS FOR n × n MATRICES 7 defined as the top left entry of the product matrix AP1 · · · Pk. These definitions were introduced in [Sz1]. The above mentioned k-th right determinant is not invariant with respect to the conjugate action of GLn(Z(R)) on Mn(R). A more appropriate (and invari- ant) definition for the k-th right determinant is the trace of AP1 · · · Pk (see [Do] and [Sz3]): rdet(k)(A) = tr(AP1 · · · Pk). The left adjoint sequence (Qk)k≥1 can be defined analogously: Q1 = A∗ and Qk+1 = (Qk · · · Q1A)∗ for k ≥ 1. The k-th left determinant of A is ldet(k)(A) = tr(Qk · · · Q1A). Note that rdet(k+1)(A) = rdet(k)(AA∗) and ldet(k+1)(A) = ldet(k)(A∗A). The basic properties of these determinants are given in the following theorems. Theorem 4.1. (see [Do], [Sz3]) If T ∈ GLn(Z(R)) is an invertible matrix with entries in the centre Z(R) of R, then tr(T −1AT ) = tr(A), (T −1AT )∗ = T −1A∗T, rdet(k)(T −1AT ) = rdet(k)(A), ldet(k)(T −1AT ) = ldet(k)(A). The next results shed light on the fact that we call radj(k)(A) = nP1 · · · Pk the k-th right adjoint and ladj(k)(A) = nQk · · · Q1 the k-th left adjoint of A. Theorem 4.2. (see [Sz1], [Sz3]) The product matrices Aradj(1)(A) and ladj(1)(A)A in Mn(R) can be written as Aradj(1)(A) = nAA∗ = tr(AA∗)I + C ′ = rdet(1)(A)I + C ′ and ladj(1)(A)A = nA∗A = tr(A∗A)I + C ′′ = ldet(1)(A)I + C ′′ , where I is the identity matrix, tr(C ′) = tr(C ′′) = 0 and all entries of the matrices C ′ and C ′′ are in the additive commutator subgroup [R, R] of R generated by all elements of the form [u, v] = uv − vu, u, v ∈ R. Theorem 4.3. (see [Sz1]) If the ring R satisfies the polynomial identity [[[. . . [[x1, x2], x3], . . .], xk], xk+1] = 0 ( R is Lie nilpotent of index k), then the products Aradj(k)(A) and ladj(k)(A)A are scalar matrices in Mn(R) such that Aradj(k)(A) = nAP1 · · · Pk = rdet(k)(A)I, ladj(k)(A)A = nQk · · · Q1A = ldet(k)(A)I. If R is commutative, then radj(1)(A) = ladj(1)(A) = nA∗ = n!adj(A) and {det(A)}nk−1 rdet(k)(A) = ldet(k)(A) = n {(n − 1)!}1+n+n2+···+nk−1 . If R is Lie nilpotent of index 2 and 1 C ′ ∈ Mn([R, R]) ⊆ Mn(Z(R)) implies that n ∈ R, then in Theorem 4.2 AA∗ = 1 n (tr(AA∗)I + C ′) ∈ Mn(Z(R)[tr(AA∗)]), 8 J. SZIGETI AND L. VAN WYK where the subring Z(R)[tr(AA∗)] generated by Z(R) and tr(AA∗) is commutative. Thus rdet(2)(A) = rdet(1)(AA∗) = n! det(AA∗). Let 1 ≤ t ≤ n − 1 be an integer and R = R0 ⊕ R1 be a Z2-grading of R. Now A ∈ Mn(R) is called an (n, t) supermatrix if ai,j ∈ R0 for all 1 ≤ i, j ≤ t and t + 1 ≤ i, j ≤ n, and ai,j ∈ R1 for all 1 ≤ i ≤ t, t + 1 ≤ j ≤ n and t + 1 ≤ i ≤ n, 1 ≤ j ≤ t. Thus an (n, t) supermatrix can be partitioned into square and rectangular blocks as follows: A =(cid:20) A1,1 A1,2 A2,1 A2,2 (cid:21) , where A1,1 is a t × t and A2,2 is an (n − t) × (n − t) square matrix over R0 and A1,2 is a t × (n − t) and A2,1 is an (n − t) × t rectangular matrix over R1. Clearly, the set of all (n, t) supermatrices Mn,t(R) is a subring (algebra) of Mn(R). Theorem 4.4. (see [Sz2]) If R = R0 ⊕ R1 is a Z2-grading of R and A ∈ Mn,t(R), then A∗ ∈ Mn,t(R) and rdet(k)(A), ldet(k)(A) ∈ R0 for all 1 ≤ k. Let R[z] denote the ring of polynomials of the single commuting indeterminate z, with coefficients in R. The k-th right (left) characteristic polynomial of A is the k-th right (left) determinant of the n × n matrix zI − A in Mn(R[z]): pA,k(z) = rdet(k)(zI − A) and qA,k(z) = ldet(k)(zI − A). Theorem 4.5. (see [Sz2]) If R = R0 ⊕ R1 is a Z2-grading of R and A ∈ Mn,t(R), then pA,k(z), qA,k(z) ∈ R0[z] for all 1 ≤ k. The above characteristic polynomials appear in the following Cayley-Hamilton the- orems. Theorem 4.6. (see [Sz3]) The first right characteristic polynomial pA,1(z) ∈ R[z] of a matrix A ∈ Mn(R) is of the form 0 + λ(1) 1 z + · · · + λ(1) n−1zn−1 + λ(1) pA,1(z) = λ(1) n zn 0 , λ(1) with λ(1) n(zI − A)(zI − A)∗ can be written as 1 , . . . , λ(1) n−1, λ(1) n ∈ R and λ(1) n = n!. The product matrix n(zI − A)(zI − A)∗ = pA,1(z)I + C0 + C1z + · · · + Cnzn, where the matrices Ci ∈ Mn(R) are uniquely determined by A, tr(Ci) = 0 and each entry of Ci is in [R, R], i.e. Ci ∈ Mn([R, R]) for all 0 ≤ i ≤ n. The right (λ(1) 0 I + C0) + A(λ(1) 1 I + C1) + · · · + An−1(λ(1) n−1I + Cn−1) + An(n!I + Cn) = 0 and a similar left (µ(1) 0 I + D0) + (µ(1) 1 I + D1)A + · · · + (µ(1) n−1I + Dn−1)An−1 + (n!I + Dn)An = 0 Cayley-Hamilton identity with right and left matrix coefficients hold for A. DETERMINANTS FOR n × n MATRICES 9 Theorem 4.7. (see [Sz1]) If the ring R satisfies the polynomial identity [[[. . . [[x1, x2], x3], . . .], xk], xk+1] = 0 ( R is Lie nilpotent of index k), then the k-th right characteristic polynomial pA,k(z) ∈ R[z] of a matrix A ∈ Mn(R) is of the form pA,k(x) = λ(k) 1 z + · · · + λ(k) 0 + λ(k) nk ∈ R and λ(k) 0 , λ(k) 1 , . . . , λ(k) nk −1, λ(k) with λ(k) The right nk −1znk−1 + λ(k) nk znk , nk = n {(n − 1)!}1+n+n2+···+nk−1 . (A)pA,k = Iλ(k) 0 + Aλ(k) 1 + · · · + Ank −1λ(k) nk −1 + Ank λ(k) nk = 0 and a similar left qA,k(A) = µ(k) 0 I + µ(k) 1 A + · · · + µ(k) nk−1Ank −1 + µ(k) nk Ank = 0 Cayley-Hamilton identity with right and left scalar coefficients hold for A. We also have (A)u = v(A) = 0, where u(z) = pA,k(z)h(z), v(z) = h(z)qA,k(z) and h(z) ∈ R[z] is arbitrary. Now consider E as a base ring and observe that E is Lie nilpotent of index 2. Thus the above Theorems 4.3 and 4.7 apply to Mn(E). The natural Z2-grading E = E0 ⊕ E1 allows us to apply Theorems 4.4 and 4.5 to Mn,t(E). The most remarkable consequences of these theorems are the following: Mn(E) is integral over E0 of degree 2n2 and Mn,t(E) is integral over E0 of degree n2 (see [Sz1, Sz2]). For a 2 × 2 matrix A = [ai,j ] ∈ M2(E) we have A = A0 + A1 with A0 = [a(0) i,j ] ∈ M2(E0) and A1 = [a(1) i,j ] ∈ M2(E1). Thus adj(A0 + A1w) =" a(0) 2,2 + a(1) 2,1 − a(1) −a(0) 2,2w −a(0) 2,1w a(0) 1,2 − a(1) 1,1 + a(1) 1,1w # = B0 + B1w, 1,2w whence B0 + B1 =" a(0) 2,2 + a(1) 2,1 − a(1) 1,2 − a(1) 2,2 −a(0) a(0) 1,1 + a(1) −a(0) 2,1 1,2 1,1 # =(cid:20) a2,2 −a1,2 a1,1 (cid:21) = A∗ −a2,1 and 2rgdet(A) = 2 det(A(B0 + B1)) = 2 det(AA∗) = rdet(1)(AA∗) = rdet(2)(A) follow. The comparison of rgdet(A) and rdet(2)(A) for a 3 × 3 matrix A ∈ M3(E) is a challenging problem. 5. THE TRACE FORM OF THE SYMMETRIC DETERMINANT If the base ring R is commutative, then tr(AB) = tr(BA) for all A, B ∈ Mn(R). In spite of the fact that this well known trace identity is no longer valid for matrices over a non-commutative ring, the first left and first right determinants of A coincide (it was not recognized in [Sz3]). Theorem 5.1. The traces of the product matrices A∗A and AA∗ are both equal to the symmetric determinant of A: rdet(1)(A) = tr(AA∗) = sdet(A) = tr(A∗A) = ldet(1)(A). 10 J. SZIGETI AND L. VAN WYK Proof. We prove that tr(AA∗) = sdet(A). (The proof of sdet(A) = tr(A∗A) is similar.) The trace of a matrix is the sum of the diagonal entries, hence tr(A∗A) = X1≤r,s≤n a∗ r,sas,r sgn(α)sgn(β)aα(1),β(1) · · · aα(s−1),β(s−1)aα(s+1),β(s+1) · · · aα(n),β(n)aα(s),β(s) = X(α,β,s)∈∆n = Xα′,β ′∈Sn sgn(α′)sgn(β′)aα′(1),β ′(1) · · · aα′(t),β ′(t) · · · aα′(n),β ′(n) = sdet(A), where ∆n = {(α, β, s) α, β ∈ Sn, 1 ≤ s ≤ n, α(s) = s} and the map (α, β, s) 7−→ (α′, β′) with and α(1) α′ =(cid:18) 1 β′ =(cid:18) 1 β(1) is a ∆n −→ Sn × Sn bijection. Since s − 1 . . . . . . α(s − 1) α(s + 1) s . . . n − 1 n . . . α(n) s − 1 . . . . . . β(s − 1) β(s + 1) s . . . n − 1 . . . β(n) s (cid:19) β(s) (cid:19) n sgn(α′) = (−1)n−ssgn(α) , sgn(β′) = (−1)n−ssgn(β) and aα(1),β(1) · · · aα(s−1),β(s−1)aα(s+1),β(s+1) · · · aα(n),β(n)aα(s),β(s) = aα′(1),β ′(1) · · · aα′(t),β ′(t) · · · aα′(n),β ′(n) , the proof is complete. (cid:3) Corollary 5.2. The first right and left characteristic polynomials of a matrix A ∈ Mn(R) coincide: pA,1(z) = qA,1(z). Thus we have λ(1) for all 0 ≤ i ≤ n in the corresponding Cayley-Hamilton identities (see Theorem 4.6). i = µ(1) i In view of Theorem 5.1 and Corollary 5.2, for the above determinants and characteristic polynomials, it is reasonable to use the terminology "symmetric" instead of "first right" and "first left". The following observation for 2 × 2 matrices over the Grassmann algebra is due to Domokos (see [Do]). Proposition 5.3. If A = [ai,j] is in M2(R), then rdet(2)(A) − ldet(2)(A) = S4(a1,1, a1,2, a2,1, a2,2), where S4(x1, x2, x3, x4) =Pσ∈S4 sgn(σ)xσ(1)xσ(2)xσ(3)xσ(4) is the standard polyno- mial of degree four. Proof. Using and the products a1,1 (cid:21) A∗ =(cid:20) a2,2 −a1,2 −a2,1 AA∗ =(cid:20) a1,1a2,2 − a1,2a2,1 −a1,1a1,2 + a1,2a1,1 A∗A =(cid:20) a2,2a1,1 − a1,2a2,1 a2,1a2,2 − a2,2a2,1 −a2,1a1,2 + a2,2a1,1 (cid:21) , −a2,1a1,1 + a1,1a2,1 −a2,1a1,2 + a1,1a2,2 (cid:21) , a2,2a1,2 − a1,2a2,2 DETERMINANTS FOR n × n MATRICES 11 a direct computation shows that rdet(2)(A) − ldet(2)(A) = rdet(1)(AA∗) − ldet(1)(A∗A) = sdet(AA∗) − sdet(A∗A) = (a1,1a2,2 −a1,2a2,1)(−a2,1a1,2+a2,2a1,1)+(−a2,1a1,2+a2,2a1,1)(a1,1a2,2 −a1,2a2,1) −(−a1,1a1,2 +a1,2a1,1)(a2,1a2,2 −a2,2a2,1)−(a2,1a2,2 −a2,2a2,1)(−a1,1a1,2 +a1,2a1,1) −(a2,2a1,1 −a1,2a2,1)(−a2,1a1,2 +a1,1a2,2)−(−a2,1a1,2 +a1,1a2,2)(a2,2a1,1 −a1,2a2,1) +(a2,2a1,2 −a1,2a2,2)(−a2,1a1,1 +a1,1a2,1)+(−a2,1a1,1 +a1,1a2,1)(a2,2a1,2 −a1,2a2,2) = S4(a1,1, a1,2, a2,1, a2,2). (cid:3) Corollary 5.4. If A = [ai,j] is in M2(R), then pA,2(z) − qA,2(z) = rdet(2)(zI − A) − ldet(2)(zI − A) = S4(z − a1,1, −a1,2, −a2,1, z − a2,2) = S4(−a1,1, −a1,2, −a2,1, −a2,2) is a constant polynomial in R[z]. = S4(a1,1, a1,2, a2,1, a2,2) 6. THE SYMMETRIC NEWTON FORMULAE FOR 2×2 AND 3×3 MATRICES If our base ring R is commutative, then the well known Newton trace formulae for 2 × 2 and 3 × 3 matrices are the following: 2 det(A) = tr2(A) − tr(A2), 6 det(A) = tr3(A) − 3tr(A)tr(A2) + 2tr(A3). Proposition 6.1. If R is an arbitrary ring and A ∈ M2(R), then the symmetric analogue sdet(A) = tr2(A) − tr(A2) of the classical 2 × 2 Newton formula holds. Notice that sdet(A) = 2 det(A) in case of a commutative R. Proof. Using A =(cid:20) a b d (cid:21) and A2 =(cid:20) a2 + bc ab + bd cb + d2 (cid:21) , ca + dc c we obtain that tr2(A) − tr(A2) = (a + d)2 − (a2 + bc + cb + d2) = ad + da − bc − cb = sdet(A). (cid:3) Theorem 6.2. symmetric analogue of the classical 3 × 3 Newton formula holds: If R is an arbitrary ring and A ∈ M3(R), then the following Notice that sdet(A) = tr3(A)−tr(A)·tr(A2)−tr(A·tr(A)·A)−tr(A2)·tr(A)+tr(A3)+tr(cid:0)(A⊤)3(cid:1) . sdet(A) = 6 det(A), tr(A)tr(A2) = tr(A·tr(A)·A) = tr(A2)·tr(A), tr(cid:0)(A⊤)3(cid:1) = tr(A3) in case of a commutative R. Proof. Using A =  b e c f a d g h p   , A⊤ =  a d b c g e h f p   12 J. SZIGETI AND L. VAN WYK a2 + bd + cg da + ed + f g ga + hd + pg A2 =  and a2 + bd + cg da + ed + f g ga + hd + pg A3 =  we obtain that ab + be + ch ac + bf + cp db + e2 + f h dc + ef + f p gb + he + ph gc + hf + p2   g h p   gb + he + ph gc + hf + p2   ab + be + ch ac + bf + cp db + e2 + f h dc + ef + f p ·  a d b e c f , and tr(A2) = a2 + bd + cg + db + e2 + f h + gc + hf + p2 tr(A3) = (a2 + bd + cg)a + (ab + be + ch)d + (ac + bf + cp)g +(da + ed + f g)b + (db + e2 + f h)e + (dc + ef + f p)h +(ga + hd + pg)c + (gb + he + ph)f + (gc + hf + p2)p. We obtain a similar expression for tr((A⊤)3). Then the proof can be completed by direct (but annoying) computation. (cid:3) Remark 6.3. For a non-commutative ring R, the identity tr(cid:0)(A⊤)2(cid:1) = tr(A2) holds for any A ∈ Mn(R), but tr(cid:0)(A⊤)3(cid:1) = tr(A3) is not valid even in the 2 × 2 case. Theorem 6.4. If A is a 3×3 matrix over an arbitrary ring R, then the symmetric characteristic polynomial of A in R[z] is pA,1(z) = qA,1(z) = sdet(zI − A) = 6z3 − 6tr(A)z2 + 3(tr2(A) − tr(A2))z − sdet(A). Proof. Using tr(zI − A) = 3z − tr(A), tr3(zI − A) = 27z3 − 27tr(A)z2 + 9tr2(A)z − tr3(A), (zI − A)2 = z2I − zA − Az + A2, tr((zI − A)2) = 3z2 − 2tr(A)z + tr(A2), tr(zI − A) · tr((zI − A)2) = 9z3 − 9tr(A)z2 + 2tr2(A)z + 3tr(A2)z − tr(A)tr(A2), tr((zI − A)2) · tr(zI − A) = 9z3 − 9tr(A)z2 + 2tr2(A)z + 3tr(A2)z − tr(A2)tr(A), tr ((zI − A) · tr(zI − A) · (zI − A)) = = tr(cid:0)3Iz3 − 3Az2 − tr(A)Iz2 − 3Az2 + tr(A)Az + 3A2z + Atr(A)z − Atr(A)A(cid:1) = = 9z3 − 9tr(A)z2 + 2tr2(A)z + 3tr(A2)z − tr(A · tr(A) · A), (zI − A)3 = z3I − z2A − zAz − Az2 + A2z + AzA + zA2 − A3, tr((zI − A)3) = 3z3 − 3tr(A)z2 + 3tr(A2)z − tr(A3), tr((zI − A)⊤)3) = 3z3 − 3tr(A⊤)z2 + 3tr((A⊤)2)z − tr((A⊤)3) = = 3z3 − 3tr(A)z2 + 3tr(A2)z − tr((A⊤)3) and Theorem 6.2, we obtain that sdet(zI − A) = tr3(zI − A) −tr(zI −A)·tr((zI −A)2)−tr ((zI −A) · tr(zI −A) · (zI −A))−tr((zI−A)2)·tr(zI−A) +tr((zI−A)3)+tr(cid:0)((zI − A)⊤)3(cid:1) = 6z3−6tr(A)z2+3(tr2(A)−tr(A2))z−sdet(A). (cid:3) DETERMINANTS FOR n × n MATRICES 13 If A ∈ M3(R), then Theorem 4.6 gives the existence of 3 × 3 Corollary 6.5. matrices Ci, Di (0 ≤ i ≤ 3) with entries in [R, R] such that (−sdet(A)I+C0)+(3(tr2(A)−tr(A2))I+C1)A+(−6tr(A)I+C2)A2+(6I+C3)A3 = 0 and (−sdet(A)I+D0)+A(3(tr2(A)−tr(A2))I+D1)+A2(−6tr(A)I+D2)+A3(6I+D3) = 0. Corollary 6.6. If 1 6 ∈ R and A ∈ M3(R) such that then sdet(A) = 0 and tr(A) = tr(A2) = tr(A3) = tr(cid:0)(A⊤)3(cid:1) = 0, A3 = C0 + C1A + C2A2 + C3A3 = D0 + AD1 + A2D2 + A3D3 for some 3×3 matrices Ci, Di (0 ≤ i ≤ 3) with entries in [R, R]. Thus A3 ∈ M3(T ), where T = R[R, R] ∩ [R, R]R is the intersection of the left and right ideals R[R, R] and [R, R]R of R. We close the paper by the following: Problem 6.7. If R is a commutative ring, then the Newton formula for a 4 × 4 matrix A ∈ M4(R) is 24 det(A) = tr4(A) − 6tr2(A)tr(A2) + 3tr2(A2) + 8tr(A)tr(A3) − 6tr(A4). Find the symmetric analogue of the above formula for sdet(A) over an arbitrary ring R. Acknowledgment. The authors thank the referee for his/her kind help to improve the exposition of the paper. REFERENCES (1) [A] H. Aslaksen, Quaternionic determinants, Math. Intelligencer 18 (3) (1996), 57–65. (2) [Do] M. Domokos, Cayley-Hamilton theorem for 2 × 2 matrices over the Grassmann algebra, J. Pure Appl. Algebra 133 (1998), 69-81. (3) [Dr] V. Drensky, Free Algebras and PI-Algebras, Springer-Verlag, 2000. (4) [DrF] V. Drensky and E. Formanek, Polynomial Identity Rings, Birkhauser- Verlag, 2004. (5) [GGRW] I. Gelfand, S. Gelfand, V. Retakh and R. L. Wilson, Quaside- terminants, Adv. Math. 193 (2005), 56–141. (6) [KT] I. Kantor and I. Trishin, On a concept of determinant in the super- case, Comm. Algebra 22 (10) (1994), 3679-3739. (7) [Ke] A. R. Kemer, Ideals of Identities of Associative Algebras, Translations of Math. Monographs, Vol. 87 (1991), AMS Providence, Rhode Island. (8) [Ros] J. Rosenberg, Algebraic K-Theory and its Applications, GTM Springer, 1994. (9) [Row1] L. H. Rowen, Polynomial Identities in Ring Theory, Academic Press, New York, 1980. (10) [Row2] L. H. Rowen, Ring Theory Vol. I, II, Academic Press, New York, 1988. 14 J. SZIGETI AND L. VAN WYK (11) [SSz] S. Sehgal and J. Szigeti: Matrices over centrally Z2-graded rings, Beitrage Algebra Geom. (Berlin) 43 (2) (2002), 399-406. (12) [St] E. Study, Zur Theorie der linearen Gleichungen, Acta Math. 42 (1920), 1-61. (13) [Sz1] J. Szigeti, New determinants and the Cayley-Hamilton theorem for matrices over Lie nilpotent rings, Proc. Amer. Math. Soc. 125 (1997), 2245-2254. (14) [Sz2] J. Szigeti, On the characteristic polynomial of supermatrices, Israel J. Math. 107 (1998), 229-235. (15) [Sz3] J. Szigeti, Cayley-Hamilton theorem for matrices over an arbitrary ring, Serdica Math. J. 32 (2006), 269-276. (16) [SzT] J. Szigeti and Zs. Tuza: Solving systems of linear equations over Lie-nilpotent rings, Linear and Multilinear Algebra 42 (1997), 43-51. Institute of Mathematics, University of Miskolc, Miskolc, Hungary 3515 E-mail address: [email protected] Department of Mathematical Sciences, Stellenbosch University, P/Bag X1, Matieland 7602, Stellenbosch, South Africa E-mail address: [email protected]
1907.06761	1	1907	2019-07-15T21:31:18	On the Noether Bound for Noncommutative Rings	[ "math.RA" ]	We present two noncommutative algebras over a field of characteristic zero that each posses a family of actions by cyclic groups of order $2n$, represented in $n \times n$ matrices, requiring generators of degree $3n$.	math.RA	math	ON THE NOETHER BOUND FOR NONCOMMUTATIVE RINGS LUIGI FERRARO, ELLEN KIRKMAN, W. FRANK MOORE, AND KEWEN PENG Abstract. We present two noncommutative algebras over a field of charac- teristic zero that each possess a family of actions by cyclic groups of order 2n, represented in 2 × 2 matrices, requiring generators of degree 3n. 1. Background Emmy Noether proved the following theorem that is useful in computing the invariants of a finite group acting linearly on a commutative polynomial ring over a field of characteristic zero (or when the characteristic of k is larger than G). Theorem 1.1 (Noether 1916 [14]). If k is a field of characteristic zero and G is a finite group of invertible n × n matrices acting linearly on A := k[x1, . . . , xn] then the ring of invariants AG can be generated by polynomials of total degree ≤ G. The non-modular case (where characteristic of k does not divide G) was proven independently by Fleischmann (2000) [6], Fogarty (2001) [7], and Derksen and Sidman (2004) [4]. Noether's bound can be sharp in the case G is a cyclic group, and Domokos and Hegedus provided a smaller upper bound on the degrees of generators if G is not cyclic; this result was extended to all characteristics by Sezer. Theorem 1.2 (Domokos and Hegedus 2000 [5], Sezer 2002 [15]). If G is a non- cyclic finite group of invertible n × n matrices acting linearly on A := k[x1, . . . , xn] then the ring of invariants AG can be generated by polynomials of total degree ≤ 3G/4 if G is even, and ≤ 5G/8 if G is odd. The Noether bound does not always hold if the field has characteristic p (see, for example, Example 3.5.5(a) p. 94 [3], where a degree 3 invariant is required to generate the invariants under a group of order 2 acting on polynomials in 6 variables over a field of characteristic 2). For further background on the problem of finding degree bounds for groups acting on A = k[x1, . . . , xn] see the survey of Neusel from 2007 [13]. Symonds proved the following general theorem that is true for a field in any characteristic, showing that there is an upper bound that is a function of both the order of the group and the dimension of the representation of the group (i.e. the number of variables in the polynomial ring). Theorem 1.3 (Symonds 2011 [16]). If G is a finite group of order G > 1 acting linearly on A := k[x1, . . . , xn] with n ≥ 2 then the ring of invariants AG can be generated by polynomials of degree ≤ n(G − 1). The case of group actions on the noncommutative skew polynomial ring A = k−1[x1, . . . , xn] (polynomials where xj xi = −xixj for i 6= j), was considered by 2010 Mathematics Subject Classification. Primary 16W22, 13A50, 16Z05. 1 2 LUIGI FERRARO, ELLEN KIRKMAN, W. FRANK MOORE, AND KEWEN PENG Kirkman, Kuzmanovich, and Zhang ([11]), where results concerning permutation actions on A = k−1[x1, . . . , xn] were proven, and it was noted that the Noether bound does not hold when n = 2 and G is the group of order 2 generated by the transposition of variables, since a generating set of the fixed ring requires a generator of degree 3. Here the difference between the order of the group and the largest degree needed in a set of generators is only 1, but in this paper we show that this difference can be arbitrarily large. Throughout this paper let k be an algebraically closed field of characteristic 0, i 1 2 = −1, and G be the cyclic group of order 2n generated by the matrix g :=(cid:20)0 λ 0(cid:21) where λ = e2πi/n is a primitive nth root of unity, For an algebra A on which g acts, let β(g) denote the minimal degree d such that the ring of invariants AG has a set of algebra generators of degree ≤ d. We compute β(g) explicitly in Section 2 for the skew polynomial ring k−1[u, v], and in Section 3 for the down-up algebra A(0, 1). In both cases, when n is odd β(g) = 3n (Theorems 2.5 and 3.7), the Noether bound does not hold, and β(g) is arbitrarily larger than the order of the group. Moreover, in the case where A = k[x1, . . . , xn] Noether's bound is exceeded in characteristic p by using a representation of G of large dimension. For the noncommutative algebras that we consider here the large values of β(g) are achieved using only a two-dimensional representation. Both of these algebras were considered because they are Artin-Schelter regular algebras (as in [1]), and can be regarded as natural noncommutative generalizations of commutative polynomial rings. On the other hand, F. Gandini [8, Theorem VI.13] has shown that the Noether bound holds for exterior algebras k−1[x1, . . . , xn]/(x2 n). The analog of Noether's bound for noncommutative algebras is not yet evident, but these examples provide data for further research. 1, . . . , x2 Many of the paper's computations were provided by the fourth author in an undergraduate research project in mathematics at Wake Forest University. Com- putations were aided by the NCAlgebra package in Macaulay2 [9]. 2. A cyclic group acting on a skew polynomial ring Let A := k−1[u, v] be the skew polynomial ring with vu = −uv. The group G sum of the monomial uivj, i.e. the sum of the distinct elements in the orbit of uivj under the action of G. The orbit sum of a monomial is equal, up to a constant, to acts on A by g.u = v and g.v = λu. Let O(i, j) := Pk gk.(uivj) denote the orbit the value of the Reynolds operator RG(uivj) =Pg∈G g.(uivj)/G, where the sum is taken over all elements of the group. Proposition 2.1. An element a ∈ AG is a linear combination of elements of the form O(kn − i, i) := ukn−ivi + (−1)(kn−i)iλiuivkn−i, for some k, i, where k ≥ 1 and 0 ≤ i ≤ ⌊kn/2⌋. Proof. Since the elements of G = (g) do not change the total degree of an element of A, an invariant is a linear combination of homogeneous elements of some degree p; these elements are linear combinations of monomials of the form up−ivi for i = 0, . . . , p for some p. Since g2 :=(cid:20)λ 0 0 λ(cid:21), any homogeneous invariant must be a linear combination of invariants of degree a multiple of n, indeed g2.(up−ivi) = λpup−ivi = up−ivi ⇔ p ≡ 0 mod n. ON THE NOETHER BOUND FOR NONCOMMUTATIVE RINGS 3 Hence homogeneous invariants are linear combinations of orbit sums of monomials of the form ukn−ivi for some k, i, and the orbit of ukn−ivi is {ukn−ivi, (−1)(kn−i)iλiuivkn−i} because g2.ukn−ivi = ukn−ivi. (cid:3) Note that when the monomial is an invariant, the two summands in the propo- sition above are identical, and we have defined O(i, j) to be twice the sum of monomials in the orbit (e.g. when n = 2, we define O(2, 2) = 2u2d2). Further, it is possible for O(i, j) = 0 (e.g. when n = 2, we have O(1, 1) = uv + vu = 0). In this section we will show that n 2n 3n if n ≡ 2 mod 4 if n ≡ 0 mod 4 if n ≡ 1 mod 2. β(g) =  Hence the Noether bound does not hold when n is odd. 2.1. Case n even. We begin with the case n ≡ 2 mod 4. Theorem 2.2. If n ≡ 2 mod 4 then β(g) = n (cid:12) G = 2n. Proof. We show that all invariants can be obtained from the invariants of degree n, namely from O(n − i, i) = un−ivi + (−1)iλiuivn−i, for i = 0, . . . , n/2. Inducting on k we show these elements generate all invariants of degree kn where k ≥ 2. Note that O(n/2, n/2)/2 = un/2vn/2 is an invariant of degree n, since n/2 is odd and g.un/2vn/2 = vn/2λn/2un/2 = −λn/2un/2vn/2 = un/2vn/2. It suffices to show that the invariants of the form a = ukn−ivi + (−1)iλiuivkn−i can be generated. If i ≥ n/2, a = un/2vn/2((−1)i+1u(2k−1)n/2−ivi−n/2 + (−1)i(−1)i+1λiui−n/2v(2k−1)n/2−i) = (−1)i+1un/2vn/2(u(2k−1)n/2−ivi−n/2 + (−1)iλiui−n/2v(2k−1)n/2−i) = (−1)i+1un/2vn/2f, where f is the orbit sum of u(2k−1)n/2−ivi−n/2, an invariant of degree (k − 1)n and hence generated by induction. If i < n/2, (un + vn)(u(k−1)n−ivi + (−1)iλiuiv(k−1)n−i) = a + g where g is u(k−1)n−ivn+i + (−1)iλiun+iv(k−1)n−i where n + i ≥ n/2, and f can be generated by the previous case. Hence we have shown β(g) = n. (cid:3) Next we consider the case where n ≡ 0 mod 4. Theorem 2.3. If n ≡ 0 mod 4, then β(g) = 2n. Proof. We first show that β(g) ≤ 2n. Let a ∈ AG of degree kn for k ≥ 2; we show, by induction on k that a can be generated by elements of degree ≤ 2n. Note that unvn ∈ AG is an invariant of degree 2n. Without loss of generality by Proposition 2.1 we can assume that a is of the form a := ukn−ivi + (−1)iλiuivkn−i. If i ≥ n then we can factor the invariant unvn from a, writing a as a = (unvn)b, where b is an invariant of degree (k − 2)n, and b can be generated by invariants of degree ≤ 2n by induction. Therefore it suffices to prove that if i < n then a can be generated by invariants of degree ≤ 2n. 4 LUIGI FERRARO, ELLEN KIRKMAN, W. FRANK MOORE, AND KEWEN PENG If i < n note that O((k − 1)n − i, i)O(n, 0) = O(kn − i, i) + O((k − 1)n − i, n + i), where O((k − 1)n − i, i) can be generated by invariants of degree ≤ 2n by induction and O((k − 1)n − i, n + i) can be generated by invariants of degree ≤ 2n by the case above. So β(g) ≤ 2n. Finally, we show β(g) ≥ 2n, i.e. AG cannot be generated by linear combinations of elements of degree n. Invariants of degree n are linear combinations of elements of the form O(un−ivi) := un−ivi + (−1)iλiuivn−i for i = 0, . . . , n/2 − 1. Note that un/2vn/2 is not invariant because g.un/2vn/2 = λn/2vn/2un/2 = −vn/2un/2 = −un/2vn/2 because n/2 is even. These n/2 invariants are linearly independent. One checks that when n is even these orbit sums satisfy the equation: O(n − i, i)O(n − j, j) = (−1)ij O(2n − i − j, i + j) (2.1) +((−1)i(j+1)λiO(n + i − j, n − i + j) 0 ≤ j ≤ i < n/2 (−1)j(i+1)λj O(n + j − i, n − j + i) 0 ≤ i ≤ j < n/2, so O(n − i, i)O(n − j, j) = O(n − j, j)O(n − i, i). Hence in considering the products of invariants of degree n we may assume that i ≥ j. We will show that one cannot write all orbit sums of degree 2n as linear combinations of products of orbit sums of degree n by showing that the homogeneous system of (cid:0)n/2+1 in n variables (2.2) below has a nontrivial solution. The coefficient matrix of this system is the coefficients of the orbit sums of degree 2n in equation (2.1) above. The existence of this solution shows that the dimension of the space spanned by products of two invariants of degree n is < the dimension of the space spanned by the orbit sums of degree 2n. Hence we next show that the system (cid:1) equations 2 (2.2) 0 = (−1)ijxi+j + (−1)i(j+1)λixn−i+j 0 ≤ j ≤ i < n/2 in the variables xp for p = 0, . . . , n − 1 has a nontrivial solution given by (2.3) xp = (−1)⌊ p+1 2 ⌋λ⌊ p+1 2 ⌋. First we consider the case i = 2ℓ and j = 2k − 1 for ℓ = 0, . . . , n/2 − 1 and k = 1, . . . , n/2, and without loss of generality we can assume i ≥ j. In this case the equation (2.2) becomes 0 = x2(l+k)−1 + λ2lx2(n/2−l+k)−1. Substituting the expres- sion in (2.3) for the variables we get (−1)ℓ+kλℓ+k + λ2ℓ(−1)(n/2−ℓ+k)λ(n/2−ℓ+k) = (−1)ℓ+kλℓ+k + λ2ℓ(−1)(−ℓ+k)λn/2λ(−ℓ+k) = (−1)ℓ+kλℓ+k + (−1)(−ℓ+k)(−1)λ(ℓ+k) = (−1)ℓ+kλℓ+k + (−1)(ℓ+k)(−1)λ(ℓ+k) = 0. The cases i, j both even or both odd are handled similarly. Hence the products of degree n orbit sums cannot determine all the orbit sums O(2n − i, i), and hence invariants of degree n do not generate all invariants of degree 2n. (cid:3) 2.2. Case n odd. When n is odd we prove that the Noether bound does not hold. ON THE NOETHER BOUND FOR NONCOMMUTATIVE RINGS 5 Lemma 2.4. If n is odd then an orbit sum of degree 2n and an orbit sum of degree n multiply as O(2n − i, i)O(n − j, j) = (−1)i(j−1)O(3n − i − j, i + j) +((−1)ijλj O(2n − i + j, n + i − j) (−1)ijλiO(n + i − j, 2n + j − i) i − j < n 2 i − j > n 2 , O(n − j, j)O(2n − i, i) = (−1)ijO(3n − i − j, i + j) +((−1)i(j−1)λjO(2n − i + j, n + i − j) (−1)i(j−1)λiO(n + i − j, 2n + j − i) i − j < n 2 i − j > n 2 . Proof. We calculate O(2n − i, i)O(n − j, j) when i − j < n similar, 2 , the other cases are O(2n − i, i)O(n − j, j) = (u2n−ivi + (−1)iλiuiv2n−i)(un−jvj + λj ujvn−j) = (−1)i(j−1)u3n−i−jvi+j + (−1)i(j−1)λi+j ui+jv3n−i−j + (−1)ijλju2n−i+j + (−1)ijλiun−j+iv2n−i+j. Now it suffices to notice that (−1)i(j−1)u3n−i−jvi+j + (−1)i(j−1)λi+j ui+jv3n−i−j = (−1)i(j−1)O(3n − i − j, i + j), and if i − j < n 2 then 2n − i + j > n + i − j and therefore (−1)ijλju2n−i+j + (−1)ijλiun−j+iv2n−i+j = (−1)ijλj O(2n − i + j, n + i − j). (cid:3) Notice that when i is even the two expressions in Lemma 2.4 coincide; this is expected since in this case O(2n − i, i) is central. When i is odd then the two products in Lemma 2.4 differ by a sign. Theorem 2.5. If n ≡ 1 mod 2 then β(g) = 3n (cid:13) G = 2n. Proof. We first show that β(g) ≤ 3n. It suffices to show by induction on k that for k ≥ 4, an invariant of the form a = O(kn − i, i) can be generated by invariants of degree ≤ 3n. Note that unvn is not invariant since g.unvn = λnvnun = vnun = −unvn because n is odd. First we show that we can generate the invariant u2nv2n. We have the equations: (un + vn)(u3n + v3n) = (u4n + v4n) − (u3nvn − unv3n) (u3n + v3n)(un + vn) = (u4n + v4n) + (u3nvn − unv3n). Adding the two equations above shows that u4n+v4n can be generated by invariants of degree ≤ 3n. Using the equation (u2n + v2n)2 = u4n + v4n + 2u2nv2n we see that u2nv2n can be generated by invariants of degree ≤ 3n. Next we show that we can generate any orbit sum a = O(kn − i, i) for k ≥ 4 if i ≥ 2n. In this case we can factor out u2nv2n and obtain a = u2nv2n(u(k−2)n−ivi−2n + (−1)i(k+1)λiui−2nv(k−2)n−i) = u2nv2nO((k − 2)n − i, i − 2n), 6 LUIGI FERRARO, ELLEN KIRKMAN, W. FRANK MOORE, AND KEWEN PENG and since O((k − 2)n − i, i − 2n) has total degree (k − 4)n it can be generated by invariants of degree ≤ 3n by induction, and therefore so can a. If i < 2n we have the equation (u2n + v2n)O((k − 2)n − i, i) = O(kn − i, i) + O((k − 2)n − i, 2n + i), and O((k − 2)n − i, i) is of degree (k − 2)n, so generated by invariants of degree ≤ 3n by induction. Further O((k − 2)n − i, 2n + i) is generated by invariants of degree ≤ 3n by the first case. Therefore a = O(kn − i, i) is generated by invariants of degree ≤ 3n, and we have shown β(g) ≤ 3n. Next we show β(g) ≥ 3n. We claim that the vector with entries xk = (−1) is a solution to the equations ( 3n−1 2 −k)( 3n−1 2 −k+1) 2 (λ n−1 2 ) 3n−1 2 −k 0 = (−1)ijxi+j +((−1)i(j−1)λjxn+i−j (−1)i(j−1)λix2n+j−i i − j < n 2 i − j > n 2 . We start by checking the first equation in the system in Lemma 2.4 0 = (−1)ij(−1) ( 3n−1 2 −i−j)( 2 3n−1 2 −i−j+1) (λ n−1 2 ) 3n−1 2 −i−j + (−1)i(j−1)λj(−1) ( 3n−1 2 −n−i+j)( 3n−1 2 −n−i+j+1) 2 (λ n−1 2 ) 3n−1 2 −n−i+j. We first check that the signs of the two summands are opposite, we need the following congruence to be true ij + ( 3n−1 2 − i − j)( 3n−1 2 − i − j + 1) 2 ≡ i(j − 1) + ( 3n−1 2 − n − i + j)( 3n−1 2 − n − i + j + 1) 2 + 1 (mod 2), this congruence is equivalent to ( 3n−1 2 − i − j)( 3n−1 2 − i − j + 1) − ( 3n−1 2 − n − i + j)( 3n−1 2 2 − n − i + j + 1) ≡ − i + 1 (mod 2), simplifying the left side yields 2ji − 2jn − in + n2 ≡ −i + 1 (mod 2), which is true since n is odd. Now we check that the power of λ is the same; for that to be true we need ( n − 1 2 )( 3n − 1 2 − i − j) ≡ j + ( n − 1 2 )( 3n − 1 2 − n − i + j) (mod n), since n is odd 2 is invertible, hence multiplying by 4 we get (n − 1)(3n − 1 − 2i − 2j) ≡ 4j + (n − 1)(3n − 1 − 2n − 2i + 2j) (mod n), which is equivalent to 1 + 2i + 2j ≡ 4j + 1 + 2i − 2j (mod n), which is true. The second equation in Lemma 2.4 can be checked with a similar computation. The above computations show that one cannot solve for all generators of degree 3n using lower degree invariants, as the system of equations needed to solve for these orbit sum invariants of degree 3n has rank less than the number of variables, since we have shown that kernel of the coefficient matrix has a nontrivial element. Hence invariants of degree ≤ 2n do not generate all invariants of degree 3n, and β(g) = 3n (cid:13) G = 2n. (cid:3) ON THE NOETHER BOUND FOR NONCOMMUTATIVE RINGS 7 3. A cyclic group acting on a graded down-up algebra The down-up algebras were introduced by Benkart and Roby in [2], as a general- ization of the universal enveloping algebra of some three-dimensional Lie algebras. Let α and β be fixed elements of a field k. The graded down-up algebras A(α, β) are the algebras generated over k generated by two elements u, d with relations d2u = αdud + βud2 and du2 = αudu + βu2d. These algebras have a monomial basis of the form ua(du)bdc, and we define the degree of this monomial to be the total degree in u and d, i.e. a + 2b + c. The down- up algebra A(α, β) is noetherian if and only if it is Artin-Schelter regular if and only if β 6= 0 [12]. Again let λ = e2πi/n be a primitive nth root of unity. Denote by g the automorphism of A(α, β) with g.u = d and g.d = λu, and let G := (g) be the cyclic group of order 2n generated by g. The graded automorphism groups of A(α, β) were computed in [10, Proposition 1.1], and the graded down-up algebras on which g acts are A(0, 1), A(0, −1) and A(2, −1). As in the previous section, any invariant under G must have degree a multiple of n. Let O(ua(du)bdc) :=Pk gk.(ua(du)bdc) denote the orbit sum of the monomial ua(du)bdc. Orbits of monomials will consist of one or two summands; when the monomial is invariant O(ua(du)bdc) will denote twice the orbit sum. Further, we will sometimes denote the orbit sum O(ua(du)bdc) as O(a, b, c). In this section we consider the case α = 0, β = 1, so that the relations in A = A(0, 1) are u2d = du2 and d2u = ud2. We show that β(g) =(2n 3n if n ≡ 0 mod 2 if n ≡ 1 mod 2, and hence in the case that n is odd the Noether bound does not hold. 3.1. Case n even. We begin by proving an upper bound on β(g) when n is even. Proposition 3.1. If n ≡ 0 mod 2 then β(g) ≤ G = 2n. Proof. First, we show that β(g) ≤ 2n. We know that undn is an invariant of degree 2n, because g.undn = dnun = undn. We prove by induction on k that invariants of degree kn for k ≥ 3 can be degenerated by invariants of degree ≤ 2n. Let O(ua(du)bdc) be an orbit sum of degree a + 2b + c = kn for k ≥ 3. Case 1: a ≥ n and c ≥ n. Then O(ua−n(du)bdc−n)O(undn) = O(ua(du)bdc), so that O(ua(du)bdc) can be generated by two invariants of lower degree. Case 2: a ≥ 2n (or by symmetry c ≥ 2n). Then O(ua−n(du)bdc)O(un) = O(ua(du)bdc) + O(ua−n(du)bdc+n) and the second invariant on the right-hand side of the equation belongs to Case 1 because (a − n) ≥ n and (c + n) ≥ n. Hence O(ua(du)bdc) can be generated by two invariants of lower degree. Case 3: b = 0. If 0 < c < n then O(u(k−2)n)O(u2n−cdc) = O(ukn−cdc) + O(u2n−cd(k−2)n+c), and the second invariant on the right-hand side belongs to Case 1. Hence O(ukn−cdc) can be generated by two invariants of lower degree, and by symmetry this is true if 0 < a < n. If b = 0 and c ≥ n, then either a ≥ n and we are done by Case 1 or a < n and we are done by symmetry as noted above. Case 4: When c = 0 (and by symmetry when a = 0). 8 LUIGI FERRARO, ELLEN KIRKMAN, W. FRANK MOORE, AND KEWEN PENG Case 4.1: When c = 0 and n ≤ b. Then O(ukn−2b(du)b−n/2)O((du)n/2) = O(ukn−2b(du)b) + λn/2O(u(k+1)n−2b(du)b−ndn), where the second invariant on the right-hand side belongs to Case 1. Hence O(ukn−2b(du)b) can be generated by two invariants of lower degree. Case 4.2: c = 0 and n/2 ≤ b < n ≤ kn/2. Then O(ukn−2b(du)b−n/2)O((du)n/2) = O(ukn−2b(du)b)+λn/2O(u(k−1)n+1(du)n−b−1d2b−n+1) where the second invariant on the belongs to Case 2, since (kn − n) + 1 ≥ 2n. Then O(ukn−2b(du)b) can be generated by two invariants of lower degree: Case 4.3: c = 0 and b < n/2. Then if b is even: O(u(k−1)n−b(du)b/2)O(un−b(du)b/2) = O(ukn−2b(du)b) + λb/2O(u(k−1)ndn) and if b is odd: O(u(k−1)n−b+1(du) b−1 2 )O(un−b−1(du) b+1 2 ) = O(ukn−2b(du)b)+λ b+1 2 O(ukn−n+1dn−1). In both cases, the second invariant on the right-hand side belongs to Case 2 and O(ukn−2b(du)b) can be generated by two invariants of lower degree. Case 5: a, b, c 6= 0. Case 5.1: b ≥ n. Then : (O(ua(du)b− n O(ua(du)b− n 2 dc)O((du)n/2) = O(ua(du)bdc) + λ n 2 dc)O(u(du) n 2 −1d) = O(ua(du)bdc) + λ n 2 O((ua+n(du)b−ndc+n) if c even 2 O(ua+n(du)b−ndc+n) if c odd. In both cases, the second invariant on the right-hand side belongs to Case 1, and O(ua(du)bdc) can be generated by two invariants of lower degree. Case 5.2: b < n. We know that with both a ≥ n and c ≥ n, the case falls into case (1). Hence we assume a < n, and hence (2b + c) > 2n, so (b + c) > n, and there are 4 subcases as shown below. If b and c are both even: O(ua(du) b 2 db+c−n)O((du)b/2dn−b) = O(ua(du)bdc) + λ− b 2 O(ua+nd2b+c−n). If b is odd and c is even: O(ua(du) 2 dc+b+1−n)O((du) b−1 b+1 2 dn−b−1) = O(ua(du)bdc)+λ− b+1 2 O(ua+n−1d2b+c−n+1). If b is even and c is odd: O(ua(du) b 2 dc+b−n)O(u(du) b 2 −1dn−b+1) = O(ua(du)bdc) + λ− b 2 O(ua+nd2b+c−n). If b and c are both odd (note that c + 2b − n + 1 < 2n − n + 1 and a ≥ 1): O(ua(du) b−1 2 db+c−n+1)O(u(du) b−1 2 dn−b) = O(ua(du)bdc)+λ− b−1 2 O(ua+n−1dc+2b−n+1) In all four cases we are done by Case 3, and by symmetry we are done if c < n, completing the proof of Case 5.2. (cid:3) From now on we will denote the element O(ukn−2i−j (du)idj) as O(kn−2i−j, i, j). We fix a basis for the vector space generated by the orbit sums of degree 2n. The orbit sum O(2n − 2i − j, i, j) is a basis element if n > i + j or if n = i + j and i is even. If an orbit sum is represented by a triple not satisfying the conditions above, we can switch it to one that does using the following formula (3.1) O(2n − 2i − j, i, j) = λi+j  O(j − 1, i + 1, 2n − 2i − j − 1) O(j + 1, i − 1, 2n − 2i − j + 1) O(j, 0, 2n − 2i − j) j odd j even i 6= 0 j even i = 0. ON THE NOETHER BOUND FOR NONCOMMUTATIVE RINGS 9 Proposition 3.2. The product of the orbit sums O(n− 2i − j, i, j)O(n− 2p− q, p, q) is: if j and q are even then the product is O(2n − 2i − j − 2p − q, p + i, q + j)+ λp+q(O(n − j + q + 1, p − i − 1, n + 2i + j − 2p − q + 1) O(n − 2i − j + q + 2p, i − p, n − q + j) p > i p ≤ i. If j is even and q is odd then the product is λp+qO(n − 2i − j + q − 1, i + p + 1, n − 2p − q + j − 1)+ (O(2n − j − 2p − q, p − i, 2i + j + q) O(2n − 2i − j − q + 1, i − p − 1, 2p + j + q + 1) p ≥ i p < i. If j is odd and q is even then the product is λp+qO(n + q − 2i − j, i + p, n + j − 2p − q)+ (O(2n − j − 2p − q + 1, p − i − 1, q + 2i + j + 1) O(2n − 2i − j − q, i − p, 2p + q + j) p > i p ≤ i. If j and q are both odd then the product is O(2n − 2i − j − 2p − q − 1, i + p + 1, q + j − 1)+ λp+q(O(n + q − j, p − i, n + 2i + j − 2p − q) O(n − 2i − j + 2p + q + 1, i − p − 1, n − q + j + 1) p ≥ i p < i. Proof. We show the case j, q even and p ≥ i, the remaining cases are similar. The product un−2i−j(du)idjun−2p−q(du)pdq is equal to u2n−2i−j−2p−q(du)i+pdj+q since the powers of u and d are central. This, together with the product λi+j dn−2i−j(ud)iuj · λp+qdn−2p−q(ud)puq, is equal to O(2n − 2i − j − 2p − q, p + i, q + j). The product un−2i−j(du)idj · λp+qdn−2p−q(ud)puq (3.2) is equal to λp+qu2n−2i−j+q(du)i(ud)pdn−2p−q+j . Since p > i we have (du)i(ud)p = u2id2i(ud)p−i, therefore, since u2i and d2i are central, equation (3.2) reduces to λp+qun−j+q(ud)p−idn+2i+j−2p−q = λp+qun−j+q+1(du)p−i−1dn+2i+j−2p−q+1. This, together with the remaining product, form the desired orbit sum. (cid:3) Next we prove the lower bound on β(g) when n is even. Theorem 3.3. If n is even then β(g) = 2n. Proof. It suffices to show that β(g) ≥ 2n, i.e. the orbit sums of degree 2n cannot be all generated by orbit sums of degree n. In the formulae in Proposition 3.2 we replace the orbit sum O(2n−2l −k, l, k) with the variable xl,k, keeping in mind that the triple may need to be changed using (3.1) if it is not in the desired form. We also change the product O(n − 2i − j, i, j)O(n − 2p − q, p, q) with zero. This gives rises to a linear system of homogeneous equations. To prove the theorem we prove that this system has a nonzero solution. We claim that the vector xl,k = λ−⌊ n−l−k ⌋ is a nonzero solution. We check the case j, q even, p ≥ i, p ≡ i (mod 2) and no changes 2 10 LUIGI FERRARO, ELLEN KIRKMAN, W. FRANK MOORE, AND KEWEN PENG to the triple occurred, the remaining cases are checked similarly. We need to check that xp+i,q+j + λp+qxp−i−1,n+2i+j−2p−q+1 = 0. Under the hypothesis p ≡ i (mod 2) this yields λ− n−p−i−q−j (cid:3) = 0, which is true since λ n 2 = −1. 2 + λ i+j+p+q 2 3.2. Case n odd. We next show that when n is odd β(g) = 3n, and hence the Noether bound does not hold. We fix a basis for the orbit sums of degree 3n as follows: O(3n − 2l − k, l, k) is a basis element if 3n > 2(l + k). If a triple does not have this form, then it can be changed according to the following formula: (3.3) O(3n − 2l − k, l, k) = λl+kO(k, l, 3n − 2l − k). Proposition 3.4. The product of the orbit sums O(n−2i−j, i, j)O(2n−2p−q, p, q) is: if j is even and q is odd then the product is λp+qO(n − 2i − j + q − 1, i + p + 1, 2n − 2p − q + j − 1)+ (O(3n − j − 2p − q, p − i, q + j + 2i) O(3n − 2i − j − q + 1, i − p − 1, j + 2p + q + 1) p ≥ i p < i. If j and q are both odd then the product is O(3n − 2i − j − 2p − q − 1, i + p + 1, j + q − 1)+ λp+q(O(n − j + q, p − i, 2n − 2p − q + j + 2i) O(n − 2i − j + q + 2p + 1, i − p − 1, 2n − q + j + 1) p ≥ i p < i. If j and q are both odd then the product is O(3n − 2i − j − 2p − q, i + p, j + q)+ λp+q(O(n − 2i − j + 1 + 2p, i − p, 2n + j − q) O(n − j + q + 1, p − i − 1, 2n + 2i + j − 2p − q + 1) p ≤ i p > i. If j is odd and q is even then the product is λp+qO(n − 2i − j + q, i + p, 2n − 2p − q + j)+ (O(3n − 2i − j − q, i − p, q + j + 2p) O(3n − 2p − j − q + 1, p − i − 1, j + 2i + q + 1) p ≤ i p > i. The product of the orbit sums O(2n − 2p − q, p, q)O(n − 2i − j, i, j) is: O(n − 2i − j, i, j)O(2n − 2p − q − 1, p + 1, q − 1) if q odd O(n − 2i − j, i, j)O(2n − q, 0, q) if q even and p = 0 O(n − 2i − j, i, j)O(2n − 2p − q + 1, p − 1, q + 1) if q even and p 6= 0   The proof of the previous proposition is similar to the proof of Proposition 3.2 and therefore is omitted. Proposition 3.5. If n is odd then β(g) ≥ 3n. Proof. We need to show that the orbit sums of degree 3n cannot be generated with orbit sums of lower degree. In the formulae in Proposition 3.4 we replace the orbit sum O(3n − 2l − k, l, k) with the variable xl,k, keeping in mind that the triple may need to be changed using (3.3) if not in the desired form. We also change the product O(n − 2i − j, i, j)O(2n − 2p − q, p, q) with zero. This gives rises to a linear ON THE NOETHER BOUND FOR NONCOMMUTATIVE RINGS 11 system of homogeneous equations. To prove the proposition we prove that this system has a nonzero solution. We claim that the vector xl,k = (−1)((l+k)(l+k+n)+l(l+1))/2λ(n+1)(n+1+2l+2k)/4. is a nonzero solution. We check the case j even, q odd, p ≥ i and no changes to the triple occurred, the remaining cases are checked similarly. We need to check that λp+qxi+p+1,2n−2p−q+j−1 + xp−i,q+j+2i = 0. We first prove that (2n + i + j − p − q)(3n + i + j − p − q) + (i + p + 1)(i + p + 2) (p + q + i + j)(p + q + i + j + n) + (p − i)(p − i + 1) 2 2 ≡ + 1 (mod 2). Indeed, computing the left hand side minus the right hand side in Z yields j(2n − 2p − 2q) + 3n2 + n(−3p − 3q + 2i) + p − 2iq + 2i, which is zero modulo 2. It remains to prove that p + q + (n + 1)(n + 1 + 2(i + p + 1) + 2(2n − 2p − q + j − 1) 4 ≡ (n + 1)(n + 1 + 2(p − i) + 2(q + j + 2i)) 4 (mod n). Computing the left hand side minus the right hand side in Z yields −n(−n+p+q−1) which is clearly zero modulo n. (cid:3) Next we prove that when n is odd, then β(g) ≤ 4n. Proposition 3.6. If n ≡ 1 mod 2 then β(g) ≤ 4n. Proof. We show that all invariants can be generated by invariants of degrees ≤ 4n. First of all, we notice that u2nv2n ∈ AG is an invariant of degree 4n, because g.u2nd2n = d2nu2n = u2nd2n. We argue by induction, and show that for a+2b+c = kn, with k ≥ 5, the orbit sum O(a, b, c) can be generated by invariants of smaller degree. Case 1: a ≥ 2n and c ≥ 2n. Since O(a − 2n, b, c − 2n)O(2n, 0, 2n) = 2O(a, b, c), by induction O(a, b, c) can be generated by two invariants of lower degree. Case 2: a ≥ 4n or c ≥ 4n. We may assume a ≥ 4n, and then O(a − 2n, b, c)O(2n, 0, 0) = O(a, b, c) + O(a − 2n, b, c + 2n). The second invariant on the right-hand side belongs to Case 1 because (a−2n) ≥ 2n and (c + 2n) ≥ 2n. Case 3: b ≥ 2n. Then O(a, b − n, c)O(0, n, 0) = O(a, b, c)+O(a + 2n, b − 2n, c + 2n). The second invariant on the right-hand side belongs to Case 1 because (a+2n) ≥ 2n and (c + 2n) ≥ 2n. Case 4: b = 0. We consider O(a, 0, c), where a + c = kn ≥ 5n. Let a ≥ c. Then we have the following cases. Case 4.1: If c ≥ 2n, the case falls into Case 1. Case 4.2: If c ≤ n, the case falls into Case 2 since a ≥ 4n. Case 4.3: If n < c < 2n, where a ≥ 3n, we have: O(a − n, 0, c)O(n, 0, 0) = O(a, 0, c) + O(a − n, 0, c + n) 12 LUIGI FERRARO, ELLEN KIRKMAN, W. FRANK MOORE, AND KEWEN PENG if c is even, while if c is odd we have O(a − n + 1, 0, c − 1)O(n − 1, 0, 1) = O(a, 0, c) + λO(a − n + 2, 0, c + n − 2). In both cases, the second invariant on the right-hand side belongs to Case 1 because c ≥ n + 1 and oth n and c are odd.. Case 5: 0 < b < n 2 then a + c > 4n, so that one cannot have both a < 2n and c < 2n. Without loss of generality we assume that a > 2n and c < 2n, and we write a = 2n + p for 0 < p < 2n and c = 2n − q for 0 < q < 2n. Since a + c > 4n we have p − q > 0. We consider two cases. Case 5.1: a = 2n + p > 3n. Then we can assume a < 4n by Case 1, and then n ≤ c < 2n. Therefore O(2n + p, b, n − q)O(0, 0, n) = O(2n + p, b, 2n − q)+ (O(3n + p − 1, b + 1, n − q − 1) O(3n + p + 1, b − 1, n − q + 1) if q even if q odd. Since 3n + p + 1 > 3n + p − 1 = 2n + p + n − 1 > 3n + n − 1 = 4n − 1, and both cases fall into Case 2. Case 5.2: a = 2n + p ≤ 3n. Then (2b + 2n − q) ≥ 2n. Therefore if q is even: O(2n + p, b − q 2 , 0)O(0, q 2 , 2n − q) = O(a, b, c) + λ− q 2 O(u2n+p(du)b− q 2 (ud) q 2 u2n−q). If q is odd: O(2n + p, b − q+1 2 , 2n − q) = O(a, b, c) + λ−(q+1)/2O(u2n+p(du)b− q+1 2 , 1)O(1, q−1 2 (ud) q−1 2 u2n−qd2). Without complete simplification, it is clear that the degree of u in the second invariant on the right-hand side is always greater or equal to (2n + p + 2n − q) > 4n. This leads the invariant to fall into Case 2, completing Case 5. Case 6: n 2 < b < n. If c is even we have O(a, b − n+1 2 , 0) = O(a, b, c) + λ(n−1)/2O(a + 2b − n, n − b − 1, c + 2b − n + 2). 2 , c + 1)O(1, n−1 If c is odd we have: O(a, b − n−1 2 , c − 1)O(0, n−1 2 1) = O(a, b, c) + λ(n+1)/2O(a + 2b − n + 2, n − b − 1, c + 2b − n). Note that in both cases above, in the second invariant on the right-hand side, the degree of du, which is n − b − 1, is smaller than n 2 . so this case follows from Case 5. Case 7: n ≤ b < 2n. Then in both the cases c even and c odd, we have: O(a, b − n, c)O(0, n, 0) = O(a, b, c)+O(a + 2b − 2n + 1, 2n − b − 1, c + 2b − 2n + 1). Note that the degree of du is now 2n − b − 1, and 0 ≤ (2n − b − 1) < n. Thus the case falls into the union set of Cases 4, 5, and 6. (cid:3) Having shown we can generate the invariants using invariants of degree ≤ 4n we now improve this bound, and together with Proposition 3.5, we compute β(g). Theorem 3.7. If n ≡ 1 mod 2 then β(g) = 3n. Proof. It suffices to show that an invariant (4n − 2b − c, b, c) of degree 4n can be generated by invariants of degree ≤ 3n. We first compute some special cases. We have the following system of equations O(2n, 0, 0)2 = O(4n, 0, 0) + O(2n, 0, 2n) O(n, 0, 0)O(3n, 0, 0) = O(4n, 0, 0) + O(3n − 1, 1, n − 1) ON THE NOETHER BOUND FOR NONCOMMUTATIVE RINGS 13 O(2n, 0, n)O(n, 0, 0) = O(3n − 1, 1, n − 1) + O(2n, 0, 0) that can be solved, so in particular we can generate O(4n, 0, 0) and O(2n, 0, 2n) using invariants of lower degree. We consider the orbit sums O(a, b, c). where a + 2b + c = 4n and show they can be generated by orbit sums of smaller degree. Case 1: b and c have the same parity. Consider the equation: O(2n − 2i − j, i, j)2 = λi+j O(2n, 0, 2n) +(O(4n − 4i − 2j, 2i, 2j) if j even O(4n − 4i − 2j − 1, 2i + 1, 2j − 1) if j odd. Hence we can generate all invariants of degree 4n by lower degree invariants when, in each of the two summands, the du and d exponents have the same parity. Case 2: b and c have opposite parity. As in the proof of Proposition 3.6 we will consider cases according to the value of b. Case 2.1: b = 0 and c is odd. We can assume c < 2n, then O(3n − c + 1, 0, c − 1)O(n − 1, 0, 1) = O(4n − c, 0, c) + λO(3n − c + 2, 0, n + c − 2), since n + c − 2 is even the second summand on the right-hand side of the equation is covered by Case 1. Hence we can generate orbit sums of the form O(4n − c, 0, c) by lower degree invariants. Case 2.2: 0 < b < n/2. Without loss of generality we can assume a < 2n and 2b + c = 4n − a > 2n. Then b + c > 2n − b > 2n − n/2 > n. When b is odd and c is even we have: O(a, (b − 1)/2, b + c + 1 − n)O(1, (b − 1)/2, n − b) = O(a, b, c) + λ−(b+1)/2O(a + n − 1, 0, c + 2b − n + 1), while if b is even and c is odd we have: O(a, b/2, c + b − n)O(0, b/2, n − b) = O(a, b, c) + λ−(b+1)/2O(a + n, 0, c + 2b − n). In either case, the second summand on the right-hand side of the equation is covered by the Case 2.1. Case 2.3: n/2 < b < n. This case follows from the equations in Proposition 3.6 Case 6, for when b is odd and c is even, the second summand on the right-hand side of the equation is O(a + 2b − n, n − b − 1, c + 2b − n + 2), and the second and third components are both odd. Similarly when b is even and c is odd. In both cases the second summand is generated by lower degree invariants by Case 1. Case 2.4: n ≤ b < 2n. This case follows from the equation in Proposition 3.6 Case 7, for the second summand on the right-hand side of the equation is O(a + 2b − 2n + 1, 2n − b − 1, c + 2b − 2n + 1) and 0 ≤ 2n − b − 1 < n with the second and third components of opposite parity, so generated by lower degree invariants by the union of Cases 2.1, 2.2, and 2.3. (cid:3) Question 3.8. It would be interesting to find β(g) for A(0, −1) and A(2, −1). For n = 1, one can show that when g acts on A(0, −1) the invariant (du)2 + (ud)2 of degree 4 is needed to generate AG, and computer calculations suggest that β(g) = 4; for n odd is β(g) = 4n? For n = 1 and g acting on A(2, −1) computer calculations suggest β(g) = 2. For noncommutative algebras how is β(g) related to the order of the group? 14 LUIGI FERRARO, ELLEN KIRKMAN, W. FRANK MOORE, AND KEWEN PENG References [1] M. Artin and W. F. Schelter, Graded algebras of global dimension 3, Adv. in Math. 66 (1987), no. 2, 171 -- 216. [2] G. Benkart and T. Roby, Down-up algebras, J. Algebra 209 (1998), no. 1, 305 -- 344. [3] H. Derksen and G. Kemper, Computational Invariant Theory, Encyclopedia of Mathematical Sciences 130: Invariant Theory and Algebraic Transformation Groups I, Springer-Verlag, Berlin, 2002. [4] H. Derksen and J. Sidman, Castelnuovo-Mumford regularity by approximation, Adv. Math. 188 (2004), no. 1, 104 -- 123. [5] M. Domokos and P. Hegedus, Noether's bound for polynomial invariants of finite groups, Arch. Math. 74 (2000), 161 -- 167. [6] P. Fleischmann, The Noether bound in invariant theory of finite groups, Adv. Math. 156 (2000), no. 1, 23 -- 32. [7] J. Fogarty, On Noether's bound for polynomial invariants of a finite group, Electron. Res. Announc. Amer. Math. Soc. 7 (2001), 5 -- 7. [8] F. Gandini, Ideals of subspace arrangements, Ph.D. Thesis, University of Michigan, Ann Arbor, May 2019, ORCID iD: 0000-0002-2619-3555. [9] Daniel R. Grayson and Michael E. Stillman, Macaulay2, a software system for research in algebraic geometry, Available at http://www.math.uiuc.edu/Macaulay2/ . [10] E. Kirkman and J. Kuzmanovich, Fixed subrings of Noetherian graded regular rings, J. Al- gebra 288 (2005), no. 2, 463 -- 484. [11] E. Kirkman, J. Kuzmanovich, and J. Zhang, Invariants of (-1)-skew polynomial rings under permutation representations, Recent advances in representation theory, quantum groups, al- gebraic geometry, and related topics, 155 -- 192, Contemp. Math., vol. 623, Amer. Math. Soc., Providence, RI, 2014. [12] E. Kirkman, I. M. Musson, and D. S. Passman, Noetherian down-up algebras, Proc. Amer. Math. Soc. 127 (1999), no. 11, 3161 -- 3167. [13] M. Neusel, Degree bounds: An invitation to postmodern invariant theory, Topology and its Applications 154 (2007), no. 4, 792 -- 814. [14] E. Noether, Der endlichkeitssatz der invarianten endlicher gruppen, Math. Ann 77 (1916), 89 -- 92. [15] M. Sezer, Sharpening the generalized Noether bound in the invariant theory of finite groups, J. Algebra 254 (2002), no. 2, 252 -- 263. [16] P. Symonds, On the Castelnuovo-Mumford regularity of rings of polynomial invariants, An- nals of Math. 174 (2011), 499 -- 517. Wake Forest University, Department of Mathematics and Statistics, P. O. Box 7388, Winston-Salem, North Carolina 27109 E-mail address: [email protected] Wake Forest University, Department of Mathematics and Statistics, P. O. Box 7388, Winston-Salem, North Carolina 27109 E-mail address: [email protected] Wake Forest University, Department of Mathematics and Statistics, P. O. Box 7388, Winston-Salem, North Carolina 27109 E-mail address: [email protected] North Carolina State University, Department of Computer Science, Campus Box 8206, 890 Oval Drive, Engineering Building II, Raleigh, NC 27695 E-mail address: [email protected]
1907.06149	1	1907	2019-07-13T23:42:53	On k-Noetherian and k-Artinian Semirings	[ "math.RA" ]	We investigate left k-Noetherian and left k-Artinian semirings. We characterize such semirings using i-injective semimodules. We prove in particular, a partial version of the celebrated Bass-Papp Theorem for semiring. We illustrate our main results by examples and counter examples.	math.RA	math	On k-Noetherian and k-Artinian Semirings* Jawad Abuhlail† Rangga Ganzar Noegraha‡ [email protected] [email protected] Department of Mathematics and Statistics King Fahd University of Petroleum & Minerals 31261 Dhahran, KSA Universitas Pertamina Jl. Teuku Nyak Arief Jakarta 12220, Indonesia July 16, 2019 Abstract We investigate left k-Noetherian and left k-Artinian semirings. We characterize such semirings using i-injective semimodules. We prove in particular, a partial version of the celebrated Bass-Papp Theorem for semiring. We illustrate our main results by examples and counter examples. Introduction Semirings are, roughly, rings not necessarily with subtraction, and generalize both rings and distributive bounded lattices. Semirings, and their semimodules (defined, roughly, as modules not necessarily with subtraction), have many important applications in several aspects of Computer Science and Mathematics, e.g., Automata Theory [HW1998], Tropical Geometry [Gla2002] and Idempotent Analysis [LM2005]. Our main reference for semirings and their semimodules is Golan's book [Gol1999] and for rings and modules Wisbauer's book [Wis1991]. Left (right) Noetherian rings, whose lattices of left (right) ideals satisfy the Ascending Chain Condition, are well studied due to the role it plays in simplifying the ideal structure such rings. On the other hand, left (right) Artinian rings, whose lattices of left (right) ideals satisfy the *MSC2010: Primary 16Y60; Secondary 16P40, 16P20 Key Words: Semirings; Semimodules; Injective Semimodules; Noetherian Semirings; Artinian Semirings The authors would like to acknowledge the support provided by the Deanship of Scientific Research (DSR) at King Fahd University of Petroleum & Minerals (KFUPM) for funding this work through projects No. RG1304-1 & RG1304-2 †Corresponding Author ‡The paper is extracted from his Ph.D. dissertation under the supervision of Prof. Jawad Abuhlail. 1 Descending Chain Condition, generalize simultaneously finite rings and rings that are finite- dimensional vector spaces over fields. Several properties of left (right) modules are valid only over rings with the ACC or the DCC. Some of these properties characterize such rings, e.g. the closure of the class of left (right) injective modules under arbitrary direct sums characterizes left (right) Noetherian rings [Rot2009, 3.39], and a ring R is left (right) Artinian if and only if every finitely generated left (right) R-module is finitely cogenerated [Wis1991, 31.4]. A left (right) ideal I of a semiring S is called a k-ideal, iff I ≤S S is subtractive [Hen1958] (equivalently, I = Ker(S In this paper, we consider the so called left k-Noetherian semirings (left k-Artinian semirings), whose lattice of subtractive left ideals satisfies the ACC (DCC). We generalize several results known for left Noetherian (left Artinian) rings to left k-Noetherian (left k-Artinian) semirings. πI−→ S/I), where πI is the canonical projection). The paper is divided into two sections. In Section 1, we collect the basic definitions, examples and preliminaries used in this paper. In particular, we recall the definitions and basic properties of exact sequences introduced by the first author Abuhlail [Abu2014]. In Section 2, we investigate left k-Noetherian (resp., left k-Artinian) semirings, i.e. semirings In Example 2.10, we show that S := satisfying the ACC (resp., the DCC) on left k-ideals. M2(R+) is left k-Noetherian but not left Noetherian, and is left k-Artinian but not left Artinian. In Theorem 2.13, we show that if every subtractive left ideal of a semiring S is a direct summand, then S is left k-Artinian and left k-Noetherian. In Theorem 2.19, we provide a partial version of the celebrated Bass-Papp Theorem for semirings: we show that if S is a semiring with enough left S-i-injective semimodules and every direct sum of S-i-injective left S-semimodules is S-i- injective, then S is left k-Noetherian. 1 Preliminaries In this section, we provide the basic definitions and preliminaries used in this work. Any notions from the theory of semirings and semimodules that are not defined here can be found in our main reference [Gol1999]. We refer to [Wis1991] for the foundations of the theory of module and rings. Definition 1.1. ([Gol1999]) A semiring is a datum (S, +, 0, ·, 1) consisting of a commutative monoid (S, +, 0) and a monoid (S, ·, 1) such that 0 6= 1 and a · 0 = 0 = 0 · a for all a ∈ S; a(b + c) = ab + ac and (a + b)c = ac + bc for all a, b, c ∈ S. 1.2. [Gol1999] Let S and T be semirings. The categories SSM of left S-semimodules with arrows the S-linear maps, SMT of right S-semimodules with arrows the T -linear maps, and SSMT of (S, T )-bisemimodules are defined in the usual way (as for modules and bimodules over rings). For a left S-semimodule M, we write L ≤S M to indicate that L is an S-subsemimodule of M. 2 Definitions 1.3. ([Gol1999]) Let (S, +, 0, ·, 1) be a semiring. • If the monoid (S, ·, 1) is commutative, we say that S is a commutative semiring. • We say that the semiring S is additively idempotent, iff s + s = s for every s ∈ S. • The set of cancellative elements of a left S-semimodules M is defined as K+(M) = {x ∈ M x + y = x + z =⇒ y = z for any y, z ∈ M}. We say that M is a cancellative semimodule, iff K+(M) = M. Examples 1.4. ([Gol1999]) • Every ring is a cancellative semiring. • Any distributive bounded lattice L = (L, ∨, 1, ∧, 0) is a additively idempotent commuta- tive semiring. • The set (Z+, +, 0, ·, 1) (resp. (Q+, +, 0, ·, 1), (Q+, +, 0, ·, 1)) of non-negative integers (resp. non-negative rational numbers, non-negative real numbers) is a cancellative commutative semiring which is not a ring. • Mn(S), the set of all n × n matrices over a semiring S, is a semiring. • B := {0, 1} with 1 + 1 = 1, is a an additively idempotent commutative semiring called the Boolean semiring. • The max-plus algebra Rmax,+ := (R ∪ {−∞}, max, −∞, +, 0) is a semiring. • The log algebra (R ∪ {−∞, ∞}, ⊕, ∞, +, 0) is a semiring, where x ⊕ y = −ln(e−x + e−y) Example 1.5. ([Gol1999, Example 1.8], [AA1994]) Consider B(n, i) := (B(n, i), ⊕, 0, ⊙, 1), where B(n, i) = {0, 1, 2, · · · , n − 1} and a ⊕ b = a + b if a + b < n; otherwise, a ⊕ b = c is the unique natural number i ≤ c < n satisfying c ≡ a + b mod (n − i); a ⊙ b = ab if ab < n; otherwise, a ⊙ b = c is the unique natural number i ≤ c < n with c ≡ ab mod (n − i). Then B(n, i) is a semiring. Notice that B(n, 0) = Zn (a group) and that B(2, 1) = B (the Boolean Algebra). 3 Example 1.6. ([Gol1999, page 150, 154]) Let S be a semiring, M be a left S-semimodule and L ≤S M. The subtractive closure of L is defined as L := {m ∈ M m + ℓ = ℓ′ for some ℓ, ℓ′ ∈ L}. (1) πL−→ M/L), where πL is the canonical projection. We One can easily check that L = Ker(M say that L is subtractive (or a k-subsemimodule), iff L = L. The left S-semimodule M is a subtractive semimodule, iff every S-subsemimodule L ≤S M is subtractive. Definition 1.7. [Gol1999, page 71] Let S be a semiring. A subtractive left (right) ideal of S is called a left (right) k-ideal [Hen1958]. We say that S is a left subtractive (right subtractive) semiring, iff every left (right) ideal of S is subtractive. We say that S is a subtractive semiring, iff S is both left and right subtractive. Remark 1.8. Whether a left subtractive semiring is necessarily right subtractive was an open problem till a counterexample was given in [KNT2011, Fact 2.1]. Following [BHJK2001], we use the following definitions. 1.9. (cf., [AHS2004]) The category SSM of left semimodules over a semiring S is a variety in the sense of Universal Algebra (closed under homomorphic images, subobjects and arbitrary products). Whence SSM is complete, i.e. has all limits (e.g., direct products, equalizers, kernels, pullbacks, inverse limits) and cocomplete, i.e. has all colimits (e.g., direct coproducts, coequal- izers, cokernels, pushouts, direct colimits). 1.10. An S-semimodule N is a direct summand of an S-semimodule M (i.e. M = N ⊕ N′ for some S-subsemimodule N′ of M) if and only if there exists α ∈ Comp(End(MS)) s.t. α(M) = N where for any semiring T we set Comp(T ) := {t ∈ T ∃et ∈ T with t +et = 1T and tet = 0T =ett}. Indeed, every direct summand of M is a retract of M; the converse is not true in general. Golan [Gol1999, Proposition 16.6] provided characterizations of direct summands. Exact Sequences Throughout, (S, +, 0, ·, 1) is a semiring and, unless otherwise explicitly mentioned, an S-module is a left S-semimodule. Definition 1.11. A morphism of left S-semimodules f : L → M is k-normal, iff whenever f (m) = f (m′) for some m, m′ ∈ M, we have m + k = m′ + k′ for some k, k′ ∈ Ker( f ); i-normal, iff im( f ) = f (L) (:= {m ∈ M m + ℓ ∈ L for some ℓ ∈ L}). normal, iff f is k-normal and i-normal. 4 There are several notions of exactness for sequences of semimodules. In this paper, we use the relatively new notion of exactness introduced by Abuhlail [Abu2014, 2.4] which is stronger than that in the sense of [Tak1982a]. Definition 1.12. ([Abu2014, 2.4]) A sequence f −→ M g −→ N L of left S-semimodules is exact, iff f (L) = Ker(g) and g is k-normal. 1.13. We call a (possibly infinite) sequence of S-semimodules · · · → Mi−1 fi−1→ Mi fi→ Mi+1 fi+1→ Mi+2 → · · · chain complex, iff f j+1 ◦ f j = 0 for every j; (2) (3) exact, iff each partial sequence with three terms M j A short exact sequence (or a Takahashi extension [Tak1982b]) of S-semimodules is an f j→ M j+1 f j+1→ M j+2 is exact. exact sequence of the form 0 −→ L f −→ M g −→ N −→ 0. 2 Noetherian and Artinian Semirings As before, (S, +, 0, ·, 1) is a semiring and, unless otherwise explicitly mentioned, an S-semimodule is a left S-semimodule. Definition 2.1. A left S-semimodule M is Noetherian (resp., k-Noetherian), iff M satisfies the ACC on its S-subsemimodules (resp., subtractive S-subsemimodules). Artinian (resp., k-Artinian), iff M satisfies the DCC on its S-subsemimodules (resp., sub- tractive S-subsemimodules). The corresponding notions for right S-semimodules are defined analogously. Remark 2.2. Every direct summand of an S-semimodule is subtractive. Let M be an S-semimodule and L a direct summand of M. Then there exists N ≤S M such that M = N ⊕ L. Let m ∈ M and ℓ, ℓ′ ∈ L be such that m + ℓ = ℓ′. Write m = en +eℓ for some n ∈ N and eℓ ∈ L, whence m + ℓ = ( n +eℓ) + ℓ = n + (eℓ + ℓ) = ℓ′. Since the sum N + L is direct, n = 0, and thus m =eℓ ∈ L.(cid:4) The following result is an easy observation; however, we highlight it as it will be used fre- quently in the proofs of the main results. Lemma 2.3. Let M be an S-semimodule and N a subtractive S-subsemimodules of M. If M = L ⊕ K for some L ≤S N and K ≤S M, then N = L ⊕ (K ∩ N). 5 Proof. Clearly, L + (K ∩ N) ⊆ N. Let n ∈ N. Since M = L + K, there exist k ∈ K and ℓ ∈ L such that n = ℓ + k. Since ℓ ∈ N and N is subtractive, we have k ∈ N, whence n ∈ L + (K ∩ N). So, N = L + (K ∩ N). Suppose now that ℓ + k = ℓ′ + k′ for some ℓ, ℓ′ ∈ L and k, k′ ∈ K ∩ N. Since the sum L + K is direct, ℓ = ℓ′ and k = k′.(cid:4) Example 2.4. Let S := M2(R+). Consider the left ideals E1 =(cid:26)(cid:20) a 0 and the left ideal b 0 (cid:21) a, b ∈ R+(cid:27) and E2 =(cid:26)(cid:20) 0 c N≥1 :=(cid:26)(cid:20) a c b d (cid:21) a ≤ c, b ≤ d, a, b, c, d ∈ R+(cid:27) . 0 d (cid:21) c, d ∈ R+(cid:27) Then we have N≥1 ∩ (E1 ⊕ E2) = N≥1 ∩ S = N≥1, while N≥1 ∩ E1 = {0} and N≥1 ∩ E2 = E2. So, we have N≥1 ∩ (E1 ⊕ E2) 6= (N≥1 ∩ E1) ⊕ (N≥1 ∩ E2). Notice that N≥1 ≤S S is not subtractive, whence the condition that N is a subtractive subsemi- module of M in Lemma 2.3 cannot be dropped.(cid:4) Definition 2.5. Let S be a semiring, M be a left S-semimodule and N ≤S M. A subtractive left S-subsemimodule L ≤S M is a maximal subtractive subsemimodule of N if L $ N and if L′ is a subtractive subsemimodule of M with L ⊆ L′ ⊆ N, then L = L′ or L′ = N. Lemma 2.6. If M is a k-Noetherian left S-semimodule, then every non-zero subsemimodule of M contains a maximal subtractive S-subsemimodule. Proof. Let N ≤S S be a non-zero subsemimodule and consider I := {L (cid:0)S N L is a subtractive subsemimodule of M}. Notice that L0 := {0M} ∈ I . If L0 is a maximal subtractive subsemimodule of N, then we are done. Otherwise, there exists L1 ∈ I such that L0 $ L1. If L1 is a maximal subtractive subsemimodule of M, we are done. Otherwise, there exists L2 ∈ I such that L1 $ L2. If no such maximal subsemimodule of N exists, we obtain a non-terminating strictly ascending chain L0 $ L1 $ L2 $ · · · $ Lk $ Lk+1 $ · · · of S-subsemimodules of N which are subtractive subsemimodules of M, absurd since M is k- Noetherian.(cid:4) Definition 2.7. The semiring S is left Noetherian (resp., left k-Noetherian), iff SS is Noetherian (resp., left k-Noetherian), equivalently every ascending chain condition of left (resp., subtractive left) ideals of S terminates; left Artinian (resp., left k-Artinian), iff SS is Artinian (resp., left k-Artinian), equivalently every descending chain of left (resp., subtractive left) ideals of S terminates. The right (k-)Noetherian and right (k-)Artinian semirings are defined analogously. A semiring which is both left and right (k-)Noetherian is called (k-)Noetherian, and a semiring which is both left and right (k-)Artinian is called (k-)Artinian. 6 Example 2.8. ([AD1975]) The semiring Z+ is Noetherian but not Artinian. Setting Ik := {0, k, k+ 1, k + 2, · · ·} yields the strictly descending non-terminating chain of ideals of Z+ : I1 % I2 % · · · % Ik % Ik+1 % · · · , i.e. Z+ is not Artinian. Lemma 2.9. The only non-trivial proper subtractive left ideals of S := M2(R+) are E1 = Span(cid:18)(cid:26)(cid:20) 1 0 E2 = Span(cid:26)(cid:20) 0 0 Nr = (cid:26)(cid:20) ra a 0 0 (cid:21)(cid:27)(cid:19) =(cid:26)(cid:20) a 0 0 1 (cid:21)(cid:27) =(cid:26)(cid:20) 0 a b 0 (cid:21) a, b ∈ R+(cid:27) 0 b (cid:21) a, b ∈ R+(cid:27) rb b (cid:21) a, b ∈ R+(cid:27) , r ∈ R+\{0}. Proof. We prove this technical lemma is three steps. Step I: E1, E2 and Nr (r ∈ R+\{0}) are subtractive left ideals of S. E1 ≤ S is a left ideal: for every a, b, c, d, p, q, r, s ∈ R+ we have (cid:20) p q s (cid:21)(cid:20) a 0 b 0 (cid:21) +(cid:20) c 0 d 0 (cid:21) =(cid:20) pa + qb + c 0 ra + sb + d 0 (cid:21) ∈ E1. r Moreover, E1 is subtractive since r s (cid:21) +(cid:20) a 0 b 0 (cid:21) =(cid:20) c 0 d 0 (cid:21) (cid:20) p q s (cid:21) ∈ E1. Similarly, E2 is a subtractive left ideal of S. implies q = 0 = s and(cid:20) p q m n (cid:21)(cid:20) ra a (cid:20) k ℓ r For any nonzero r ∈ R+, Nr is a left ideal since (for all a, b, c, d, k, ℓ, m, n ∈ R+) we have rb b (cid:21) +(cid:20) rc rd d (cid:21) =(cid:20) r(ka + ℓb + c) r(ma + nb + d) ma + nb + d (cid:21) ∈ Nr. ka + ℓb + c c Moreover, Nr ≤ S is subtractive since (cid:20) k m n (cid:21) +(cid:20) ra a rb b (cid:21) =(cid:20) rc rd d (cid:21) , ℓ c whence c = a + k/r = a + ℓ, d = b + m/r = b + n. So, k = rℓ, m = rn, and(cid:20) k m n (cid:21) ∈ Nr. ℓ Step II: E1, E2 and Nr (r ∈ R+\{0}) are subtractive left ideals of S. 7 Let I be a subtractive left ideal of M2(R+) such that E1 $ I. Then there exists(cid:20) p q such that q 6= 0 or s 6= 0, whence(cid:20) 0 q 0 s (cid:21) ∈ I as(cid:20) p 0 r s (cid:21) ∈ I Either way(cid:20) 0 0 0 1 (cid:21) ∈ I, which implies E2 ⊆ I and I = S. Similarly, if I is a subtractive left ideal of M2(R+) such that E2 $ I, then I = S. ℓ Let r ∈ R+\{0} and I be a subtractive left ideal of M2(R+) such that Nr $ I. Then there exists(cid:20) k Then k + p = rℓ for some p ∈ R+\{0}. Thus(cid:20) p 0 m n (cid:21) ∈ I such that k 6= rℓ or m 6= rn. Without loss of generality, assume that k < rℓ. 0 q (cid:21) ∈ I for some q ∈ R+ as Either way we have(cid:20) 1 0 Let I be a proper non-trivial subtractive left ideal of S. Then (cid:20) k Step III: E1, E2 and Nr (r ∈ R+\{0}) are the only subtractive left ideals of S. ℓ m n (cid:21) ∈ I\{0} for some k, ℓ, m, n ∈ R+. If k 6= 0, then If q 6= 0, then If s 6= 0, then or Thus or r 0 (cid:21) ∈ I and s (cid:21) ∈ I 0 s (cid:21) =(cid:20) p q 0 s (cid:21) . 1/q 0 (cid:21)(cid:20) 0 q 0 1/s (cid:21)(cid:20) 0 q 0 s (cid:21) . r 0 (cid:21) +(cid:20) 0 q (cid:20) p 0 0 1 (cid:21) =(cid:20) 0 (cid:20) 0 0 (cid:20) 0 0 0 1 (cid:21) =(cid:20) 0 0 0 r ℓ ℓ ℓ q 0 (cid:21) ∈ I or(cid:20) p 0 rn n (cid:21) ∈ I m m/r (cid:21) ∈ I. q 0 (cid:21) 0 (cid:21)(cid:20) p 0 0 q (cid:21) . 0 (cid:21)(cid:20) p 0 0 0 (cid:21) ∈ I, whence E1 $ I and I = S. m n (cid:21) =(cid:20) rℓ q 0 (cid:21) +(cid:20) k (cid:20) p 0 (cid:20) p 0 0 q (cid:21) +(cid:20) k m n (cid:21) =(cid:20) rℓ 0 0 (cid:21) =(cid:20) 1/p 0 (cid:20) 1 0 0 0 (cid:21) =(cid:20) 1/p 0 (cid:20) 1 0 0 0 ℓ (cid:20) 1/k 0 0 (cid:21)(cid:20) k 0 (cid:21) m n (cid:21) =(cid:20) 1 ℓ/k 0 0 ℓ 8 whence(cid:20) 1 0 Nk/ℓ. If ℓ 6= 0, then whence (cid:20) 0 0 m 6= 0, then 0 0 0 ℓ k/ℓ k/ℓ 1 (cid:21) , 0 m n (cid:21) =(cid:20) 0 0 (cid:21) ∈ I, and it follows that I ∈ {E1, Nk/ℓ, S} as I contains E1 or 0 0 (cid:21) ∈ I, or(cid:20) k/ℓ 1 1/ℓ 0 (cid:21)(cid:20) k (cid:20) 0 ℓ (cid:21) ∈ I, and so I ∈ {E2, Nk/ℓ, S} as I contains E2 or Nk/ℓ. If 0 1 (cid:21) ∈ I or (cid:20) 0 0 (cid:21)(cid:20) k 0 (cid:21) ∈ I, and it follows that I ∈ {E1, Nm/n, S} as I contains E1 or 0 0 (cid:21) ∈ I or(cid:20) m/n 1 0 1/n (cid:21)(cid:20) k (cid:20) 0 m/n ℓ (cid:21) ∈ I. So, I ∈ {E2, Nm/n, S} as I contains E2 or Nm/n.(cid:4) 0 1 (cid:21) ∈ I or(cid:20) 0 0 (cid:21) m n (cid:21) =(cid:20) 1 n/m m n (cid:21) =(cid:20) 0 m/n 1 (cid:21) (cid:20) 0 1/m 0 ℓ ℓ 0 0 0 0 whence(cid:20) 1 0 Nm/n. If n 6= 0, then 0 whence(cid:20) 0 0 We provide an example of a semiring which is left k-Artinian and left k-Noetherian but neither left Artinian (nor left Noetherian): Example 2.10. Let S = M2(R+). By Lemma 2.9, the only subtractive left ideals of S are 0, S, E1, E2 and Nr (r ∈ R+\{0}). Notice that for r 6= s, the left ideals Nr, Ns are not comparable. Thus, the longest ascending (descending) chain of subtractive left ideals of S is 0 $ N $ S (S ' N ' 0) with N = E2 or N = Nr for some r ∈ R+. Whence, S is left k-Artinian and left k-Noetherian. On the other hand, for every r ∈ R+ we have a left ideal of S given by N≥r =(cid:26)(cid:20) a p b q (cid:21) : p ≥ ra, q ≥ rb, a, b, p, q ∈ R+(cid:27) . Thus, we have an infinite strictly descending chain of left ideal that does not terminate N1 % N≥2 % N≥3 % · · · % N≥m % N≥m+1 % · · · , i.e. S is not k-Artinian. On the other hand, we have an infinite ascending chain of left ideals that does not terminate N≥1 $ N≥ 1 2 $ N≥ 1 3 $ · · · $ N≥ 1 m $ N 1 m+1 $ · · · , i.e. S is not k-Noetherian.(cid:4) An additional example of a k-Noetherian semiring that is not Noetherian was communicated to Abuhlail by T. Nam: Example 2.11. The semiring R+[x] is k-Noetherian but not Noetherian. 9 Proof. The semiring B[x], where B is the Boolean semiring, is not Noetherian. The surjective morphism of semirings f : R+ −→ B, r 7→  1, 0, r 6= 0 r = 0 induces a surjective morphism of semirings R+[x] −→ B[x], whence R+[x] is not Noetherian.(cid:4) We do not know whether k-Artinian semirings are k-Noetherian. However, we have the following interesting result. Lemma 2.12. A left S-semimodule M satisfies the ACC on direct summands if and only if M satisfies the DCC on direct summands. Proof. (=⇒) Assume that M satisfies the ACC on direct summands. Let N1 ⊇ N2 ⊇ N3 ⊇ · · · ⊇ Ni ⊇ Ni+1 ⊇ · · · (4) be a descending chain of direct summands of M. For every i ∈ N, there exists a direct sum- mand Li ≤S M such that M = Ni ⊕ Li. Since M = N2 ⊕ L2 and N1 ⊇ N2, we have by (taking into consideration Remark 2.2): N1 Lemma 2.3= N2 ⊕ (N1 ∩ L2) and M = N1 ⊕ L1 = N2 ⊕ (N1 ∩ L2) ⊕ L1. Set K1 := L1 and K2 := (N1 ∩ L2) ⊕ L1, so that N1 ⊕ K1 = M = N2 ⊕ K2 and K1 ⊆ K2. Now, N2 ⊇ N3 and M = N3 ⊕ L3, whence N2 Lemma 2.3= N3 ⊕ (N2 ∩ L3) and so M = N2 ⊕ K2 = N3 ⊕ (N2 ∩ L3) ⊕ K2. Set K3 := (N2 ∩ L3) ⊕ K2, so that M = N3 ⊕ K3 and K2 ⊆ K3. Continuing this way, we obtain an ascending chain K1 ⊆ K2 ⊆ K3 ⊆ · · · ⊆ Ki ⊆ Ki+1 ⊆ · · · (5) of direct summands of SM. By our assumption, the ascending chain (5) terminates, whence there exists t ∈ N such that Ki = Kt for any i ≥ t. For any i ≥ t, we have Nt ⊇ Ni, M = Ni ⊕ Ki and Nt ∩ Kt = 0 and so Lemma 2.3= Ni ⊕ (Nt ∩ Ki) = Ni ⊕ (Nt ∩ Kt) = Ni, Nt thus the descending chain (4) terminates. (⇐=) Assume that M satisfies the DCC on direct summands. Let L1 ⊆ L2 ⊆ L3 ⊆ · · · ⊆ Li ⊆ Li+1 (6) be an ascending chain of direct summands of M. For every i ∈ N, there exists an S-subsemimodule Ni ≤S M such that M = Li ⊕ Ni; in particular M = L1 ⊕ N1. Since L1 ⊆ L2 it follows (taking into consideration Remark 2.2) that L2 Lemma 2.3= L1 ⊕ (L2 ∩ N1), whence M = L2 ⊕ N2 = L1 ⊕ (L2 ∩ N1) ⊕ N2. 10 Since L2 ∩ N1 ⊆ N1 it follows that N1 Lemma 2.3= L2 ∩ N1) ⊕ (N1 ∩ (L1 ⊕ N2)), whence 1 := N1 and N′ Setting N′ Since M = L2 ⊕ N′ M = L1 ⊕ N1 = L1 ⊕ (L2 ∩ N1) ⊕ (N1 ∩ (L1 ⊕ N2)). 2 := N1 ∩ (L1 ⊕ N2), we have L1 ⊕ N′ 1 = M = L2 ⊕ N′ 2 where N′ 1 ⊇ N′ 2. 2 and L2 ⊆ L3, it follows that L3 Lemma 2.3= L2 ⊕ (L3 ∩ N′ 2), whence M = L3 ⊕ N3 = L2 ⊕ (L3 ∩ N′ 2) ⊕ N3. Since L3 ∩ N′ 2 ⊆ N′ 2, we have N′ 2 Lemma 2.3= (L3 ∩ N′ 2) ⊕ (N′ 2 ∩ (L2 ⊕ N3)). Setting N′ 3 := N′ 2 ∩ (L2 ⊕ N3), we have N′ 2 ⊇ N′ 3 and M = L2 ⊕ N′ 2 = L2 ⊕ (L3 ∩ N′ 2) ⊕ N′ 3 = L3 ⊕ N′ 3. Continuing this process, we obtain a descending chain N′ 1 ⊇ N′ 2 ⊇ · · · ⊇ N′ i ⊇ N′ i+1 ⊇ · · · (7) of direct summands of M such that M = Li ⊕ N′ ing chain (7) terminates, i.e. there exists some k ∈ N such that N′ i for every i ∈ N. By our assumption, the descend- i = N′ k for every i ≥ k. Now, for every i ≥ k, we have Lk ⊆ Li, M = Lk ⊕ N′ k and Li ∩ N′ i = 0 and so Li Lemma 2.3= Lk ⊕ (Li ∩ N′ k) = Lk ⊕ (Li ∩ N′ i ) = Lk. Thus the ascending chain (6) terminates.(cid:4) A ring in which every left ideal is a direct summand is left Artinian and left Noetherian [Wis1991, 3.4 and 4.1] (in fact, left semisimple). The following result extends this fact to semir- ings. Theorem 2.13. If every subtractive left ideal of S is a direct summand, then S is left k-Artinian and left k-Noetherian. Proof. Assume that every subtractive left ideal of S is a direct summand. Claim I: S is left k-Artinian. Suppose that I1 % I2 % I3 % · · · % Ii % Ii+1 % · · · (8) is a strictly descending chain of left subtractive ideals of S that does not terminate. For every k ∈ N, there exists, by our assumption, some left ideal Nk ≤S S such that S = Ik ⊕ Nk. The left ideals Ik, Nk are non-zero as the chain does not terminate, and are subtractive by Remark 2.2. 11 Since I1 ⊇ I2 and S = I2 ⊕ N2, we have Lemma 2.3= I2 ⊕ (I1 ∩ N2). I1 Then J1 := I1 ∩ N2 is a subtractive left ideal of S, which is non-zero as I1 % I2, and I1 = I2 ⊕ J1. Since I2 ⊇ I3 and S = I3 ⊕ N3, we have Lemma 2.3= I3 ⊕ (I2 ∩ N3). I2 Then I2 ∩ N3 is a subtractive left ideal of S, which is non-zero as I2 % I3, and I1 = I2 ⊕ J1 = I3 ⊕ J2 ⊕ J1. Continuing this process, we obtain at the kth step, a non-zero subtractive left ideal Jk ≤S S such that Ik = Ik+1 ⊕ Jk and I1 = Ik+1 ⊕ Jk ⊕ · · · ⊕ J1. Setting J′ i := J1 ⊕ · · · ⊕ Ji for each i ∈ N, we have S = J′ i ⊕ Ii+1 ⊕ N1 whence J′ i is subtractive (by Remark 2.2). One can easily show that J := Si∈N J′ i is subtractive. By our assumption, S = J ⊕ N for some left ideal of N ≤S S. Thus 1S = j + n for some j ∈ J i for some i ∈ N, it can be written in a unique way as j = j1 + j2 + ... + ji i+1 + N is direct, and n ∈ N. Since j ∈ J′ for some uniquely determined jk ∈ Jk, k = 1, 2, ..., i. Since J′ whence the sum J1 + J2 + ... + Ji + Ji+1 + N is direct. Setting i+1 ⊆ J, the sum J′ K := J1 ⊕ ... ⊕ Ji ⊕ N, this means that the sum Ji+1 + K is direct. For any si+1 ∈ Ji+1\{0}, we have si+1 = si+11S = si+1( j1 + j2 + ... + ji + n) = si+1 j1 + si+1 j2 + ... + si+1 ji + si+1n where si+1 jk ∈ Jk for k = 1, 2, ..., i and si+1n ∈ N. It follows that si+1 ∈ Ji+1 ∩ K = 0, absurd since si+1 6= 0. So, the descending chain (8) terminates. Claim II: S is left k-Noetherian. Let I1 ⊆ I2 ⊆ I3 ⊆ ... ⊆ Ii ⊆ Ii+1 ⊆ · · · (9) be an ascending chain of subtractive left ideals of S. Since every direct summand of SS is sub- tractive (by Remark 2.2), it follows from the proof of Claim I that SS satisfies DCC on direct summands, whence SS satisfies ACC on direct summands by Lemma 2.12. Since (9) is an ascend- ing chain of subtractive left ideals of S, whence of direct summands of SS (by our assumption), the chain terminates.(cid:4) Example 2.14. Let p be a prime number. Every subtractive ideal of the semiring S = B(p + 1, p) is a direct summand, and S is k-Artinian and k-Noetherian. Proof. S has no non-trivial subtractive ideals, thus every subtractive left ideal of S is a direct summand. Notice that S is k-Artinian and k-Noetherian since it has finitely many elements.(cid:4) 12 Example 2.15. Let S := BN with the canonical structure of a semiring induced by that on B. Then S has a subtractive left ideal which is not a direct summand and S is neither k-Artinian nor k-Noetherian. B is not a direct summand. Notice that neither the ascending Proof. The subtractive left ideal Ln∈N {0} $ B2 × ∏ n≥3 B × ∏ n≥2 chain {0} $ ... $ Bi × ∏ n≥i+1 {0} $ Bi+1 × ∏ {0} $ ... n≥i+2 nor the descending chain {0} × ∏ n≥2 B % {0}2 × ∏ n≥3 B % ... % {0}i × ∏ n≥i+1 B % {0}i+1 × ∏ n≥i+2 B % ... terminates, thus S is neither k-Noetherian nor k-Artinian.(cid:4) 2.16. Let I be a left S-semimodule. For a left S-semimodule M, we say that I is M-injective [Gol1999, page 197], iff for every injective S-linear map f : L → M and any S-linear map g : L → I, there exists an S-linear map h : M → I such that h ◦ f = g; 0 f M h / L g I M-i-injective [Alt2003], iff for every normal monomorphism f : L → M and any S-linear map g : L → I, there exists an S-linear map h : M → I such that h ◦ f = g; 0 / L g uI f (normal) M h M-e-injective [AIKN2018], iff for every short exact sequence 0 −→ L S-semimodules, the following sequence 0 −→ HomS(N, I) 0 of commutative monoids is exact. (g,I) −→ HomS(M, I) f −→ M g −→ N −→ 0 of left ( f ,I) −→ HomS(L, I) −→ We say that I is injective (resp. i-injective, e-injective), iff I is M-injective (resp. M-i- injective, M-e-injective) for every left S-semimodule M. A Bass-Papp Theorem for Semirings The celebrated Bass-Papp Theorem states that a ring R is left (right) Noetherian if and only if every direct sum of left (right) injective R-modules is (R-)injective (e.g., [Rot2009, 3.39], [Gri2007, page 407]). This characterization was extended to semirings: S by Il'in and An e-injective version of this theorem was obtained by Abuhlail et al.: 13 / / / / / / u Proposition 2.17. ([Ili2010, Theorem 3.6] ,[AIKN2018, Theorem 5.5]) The following are equiv- alent for a semiring S : (1) S is a left Noetherian ring; (2) every direct sum of injective left S-semimodules is injective and the left S-semimodule S/V (S) can be embedded in an injective left S-semimodule. (3) every direct sum of e-injective left S-semimodules is e-injective and the left S-semimodule S/V (S) can be embedded in an e-injective left S-semimodule. Notice that the assumptions in Proposition 2.17 force the semiring S to be a ring. In what fol- lows we provide a partial version of the Bass-Papp characterization for left k-Noetherian semir- ings. 2.18. We say that a semiring S has enough S-i-injective left semimodules, iff every left S- semimodule can be embedded into an S-i-injective left S-semimodule. Theorem 2.19. Let S be a semiring with enough left S-i-injective left semimodules. If every direct sum of left S-i-injective S-semimodules is S-i-injective, then S is left k-Noetherian. Proof. Let L1 ⊆ L2 ⊆ L3 ⊆ · · · ⊆ Li ⊆ Li+1 ⊆ · · · be a chain of subtractive left ideals of S and consider the left ideal L =: Sn∈N is subtractive. By our assumption, there exists for every n ∈ N an S-i-injective left S-semimodule Jn and an embedding S/Ln Jn and consider the S-linear map (10) Ln. It is clear that L ιn֒→ Jn. Set J := Ln∈N πi−→ S/Ln ϕn : S ιn֒→ Jn and ϕ : L → J, x 7→ ∞ ∑ k=1 ϕk(x). Notice that ϕ is well defined as each x ∈ L belongs to Ln for some n ∈ N and so ϕk(x) = 0 for all By our assumption, J is S-i-injective and so there exists an S-linear map ψ : S −→ J such that k ≥ n, i.e. ϕ(x) = ϕk(x) = ϕk(x). ∞ ∑ k=1 n−1 ∑ k=1 ψ◦ ι = ϕ. Let ψ(1S) = ∑ tk ∈ J. Then ψ(1S) ∈ ι S ψ / L ϕ J Jk for some m, whence ψ(x) = ψ(x · 1S) = xψ(1S) ∈ 0 m−1Lk=1 m−1Lk=1 Jk for every x ∈ L. In particular, ϕm(x) = (πm ◦ φ)(x) = 0 where πm is the projection on Jm. Thus x ∈ Ln and L = Ln, whence Lk = Ln for all k ≥ n, i.e. the chain terminates. Consequently, S is k-Noetherian.(cid:4) 14 / / / Example 2.20. If S is an additively idempotent semiring such that every direct sum of left S-i- injective S-semimodules is S-i-injective, then S is left k-Noetherian. This results from Theorem 2.19 and the fact that every semimodule over an additively idempotent semiring can be embedded into an e-injective (whence an i-injective) left S-semimodule [AIKN2018, 4.5]. Theorem 2.21. If S is a semiring such that every short exact sequence of left S-semimodules 0 → L → S → N → 0 is left splitting, then S is a left k-Noetherian. Proof. Let N0 $ N1 $ N2 $ · · · $ Nk $ Nk+1 $ · · · be a non-terminating ascending chain of subtractive left ideals of S. Notice that N := Si∈N subtractive ideal of S, whence (by assumption) the following short exact sequence Ni is a 0 → N ι −→ S π −→ S/N → 0 of left S-semimodules is left splitting. Let h : S → N be an S-linear map such that h ◦ ι = idN. Then h(1S) ∈ N, that is h(1S) ∈ Ni for some i ∈ N. If x ∈ Ni+1\Ni, then x = (h ◦ ι)(x) = h(x) = h(x1S) = xh(1S) ∈ Ni a contradiction. Hence S is k-Noetherian.(cid:4) Corollary 2.22. If S is a semiring such that every subtractive left ideal is S-i-injective, then S is a left k-Noetherian semiring. References [AA1994] F. Alarcon and D. Anderson, Commutative semirings and their lattices of ideals, Houston J. Math. 20 (1994), 571-590. 3 [AR] J. Y. Abuhlail and R. G. Noegraha, Injective Semimodules - Revisited, preprint. (available at https://arxiv.org/abs/1904.07708) [Abu2014] J. Abuhlail, Exact sequence of commutative monoids and semimodules, Homology Homotopy Appl. 16 (1) (2014), 199 -- 214. 2, 5 [AD1975] P.J. Allen and L. Dale, Ideal theory in Z+, Publ. Math. Debrecen 22 (1975), 219- 224. 7 [AIKN2018] J. Abuhlail, S. Il'in , Y. Katsov, and T. Nam, Toward homological characterization of semirings by e-injective semimodules, J. Algeb. Appl. 17(4) (2018). 13, 14, 15 [Alt2003] H. Al-Thani, Injective semimodules, J. Inst. Math. Comput. Sci. 16 (3), 143-152 (2003). 13 15 [AHS2004] J. Ad´amek, H. Herrlich and G. E. Strecker, Abstract and Concrete Cate- gories; The Joy of Cats 2004. Dover Publications Edition (2009) (available at: http://katmat.math.uni-bremen.de/acc). 4 [BHJK2001] R. El Bashir, J. Hurt, A. Jancarik, and T. Kepka, Simple commutative semirings, J. Algebra, 236 (2001), 277 - 306. 4 [Gla2002] K. Głazek, A Guide to the Literature on Semirings and their Applications in Mathe- matics and Information Sciences. With Complete Bibliography, Kluwer Academic Publishers, Dordrecht (2002). 1 [Gol1999] J. Golan, Semirings and their Applications, Kluwer Academic Publishers, Dor- drecht (1999). 1, 2, 3, 4, 13 [Gri2007] P. A. Grillet, Abstract Algebra, Second Edition, Springer, New York (2007). 13 [Hen1958] M. Henriksen, Ideals in semirings with commutative addition, Amer. Math. Soc. Notices 6 (1959), 321. 2, 4 [HW1998] U. Hebisch and H. J. Weinert, Semirings: Algebraic Theory and Applications in Computer Science, World Scientific Publishing Co., Inc., River Edge, NJ (1998). 1 [Ili2010] S. N. Il'in, Direct sums of injective semimodules and direct products of projective semimodules over semirings, Russ. Math. 54 27-37 (2010). 14 [KNT2011] Y. Katsov, T. G. Nam, N. X. Tuyen, More on subtractive semirings: simpleness, perfectness, and related problems, Commun. Algebra 39 (2011), 4342-4356. 4 [LM2005] G. L. Litvinov and V. P. Maslov (editors), Idempotent Mathematics and Mathe- matical Physics, Papers from the International Workshop held in Vienna, February 3 -- 10, 2003. Contemporary Mathematics, 377. American Mathematical Society, Providence, RI (2005). 1 [Rot2009] J. Rotman, An Introduction to Homological Algebra, Second Edition, Springer, New York (2009). 2, 13 [Tak1982a] M. Takahashi, On the bordism categories. III. Functors Hom and for semimodules. Math. Sem. Notes Kobe Univ. 10 (2) (1982), pp. 551-562. 5 [Tak1982b] M. Takahashi, Extensions of Semimodules I, Math. Sem. Notes Kobe Univ. 10 (1982), 563 -- 592. 5 [Wis1991] R. Wisbauer, Foundations of Module and Ring Theory, Gordon and Breach, Read- ing (1991). 1, 2, 11 16
1703.08734	2	1703	2017-04-03T07:44:31	Matrix wreath products of algebras and embedding theorems	[ "math.RA" ]	We introduce a new construction of matrix wreath products of algebras that is similar to wreath products of groups. We then use it to prove embedding theorems for Jacobson radical, nil, and primitive algebras. In \S\ref{Section6}, we construct finitely generated nil algebras of arbitrary Gelfand-Kirillov dimension $\geq 8$ over a countable field which answers a question from \cite{8}.	math.RA	math	MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS ADEL ALAHMADIA, HAMED ALSULAMIA, S.K. JAINA , B, EFIM ZELMANOVC Abstract. We introduce a new construction of matrix wreath products of algebras that is similar to wreath products of groups. We then use it to prove embedding theorems for Jacobson radical, nil, and primitive algebras. In §6, we construct finitely generated nil algebras of arbitrary Gelfand-Kirillov dimension ≥ 8 over a countable field which answers a question from [6]. 1. Main Results G. Higman, H. Neumann, and B.H. Neumann [11] proved that every countable group embeds in a finitely generated group. The papers [3], [20], [21], [22], [24] show that some important properties can be inherited by these embeddings. Much of this work relies on wreath products of groups. Following [11], A. I. Malcev [18] showed that every countable di- mensional associative algebra over a field is embeddable in a finitely generated algebra. In §2, 3, we introduce matrix wreath products of algebras and study their basic properties. In §4, we use matrix wreath products to prove embedding theorems for Jacobson radical algebras. S. Amitsur [2] asked if a finitely generated algebra can have a non nil Jacobson radical. The first examples of such algebras were constructed by K. Beidar [7]. J. Bell [4] constructed examples having finite Gelfand- Kirillov dimension. Finally, L. Bartholdi and A. Smoktunowicz [23] 1 MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS2 constructed a finitely generated Jacobson radical non nil algebra of Gelfand-Kirillov dimension 2. Theorem 4.1. An arbitrary countable dimensional Jacobson radical algebra is embeddable in a finitely generated Jacobson radical algebra. Theorem 4.2. An arbitrary countable dimensional Jacobson radical algebra of Gelfand-Kirillov dimension d over a countable field is embed- dable in a finitely generated Jacobson radical algebra of Gelfand-Kirillov dimension ≤ d + 6. We say that a nil algebra A is stable nil (resp. stable algebraic) if all matrix algebras Mn(A) are nil (resp. algebraic). The problem of Koethe ([14], see also [15]) if all nil algebras are stable nil is still open. Theorem 4.3. An arbitrary countable dimensional stable nil algebra A is embeddable in a finitely generated stable nil algebra. If GK dim A = d < ∞ and the ground field is countable, then A is embeddable in a finitely generated nil algebra of Gelfand-Kirillov dimension ≤ d + 6. In §5, we prove embedding theorems for countable dimensional al- gebraic primitive algebras. I. Kaplansky [13] asked if there exists an infinite dimensional finitely generated algebraic primitive algebra, a particular case of the celebrated Kurosh Problem. Such examples were constructed by J. Bell and L. Small in [5]. Then J. Bell, L. Small, and A. Smoktunowicz [6] constructed finitely generated algebraic primitive algebras of finite Gelfand-Kirillov dimension provided that the ground field is countable. Our embedding theorems for algebraic primitive algebras have a spe- cial feature. Let A be an associative algebra over a ground field F . Let X be a countable set. Consider the algebra M∞(A) of X × X matrices over A having finitely many nonzero entries. Clearly, the algebra A is embeddable in M∞(A) in many ways. We say that an algebra A is M∞-embeddable in an algebra B if there exists an embedding ϕ : M∞(A) → B. We say that A is M∞-embeddable in B as a (left, right) ideal if the image of ϕ is a (left, right) ideal of B. Theorem 5.1. An arbitrary countable dimensional stable algebraic primitive algebra is M∞-embeddable as a left ideal in a 2-generated algebraic primitive algebra. MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS3 In particular, this theorem answers the first part of question 7 from [6]. Theorem 5.2. Let F be a countable field. An arbitrary countable dimensional stable algebraic primitive algebra of Gelfand-Kirillov di- mension ≤ d is M∞-embeddable as a left ideal in a finitely generated algebraic primitive algebra of Gelfand-Kirillov dimension ≤ d + 6. In §6, we answer question 1 from [6]. Theorem 6.1. Let F be a countable field. For an arbitrary d ≥ 8, there exists a finitely generated nil F -algebra of Gelfand-Kirillov dimension d. 2. Matrix wreath products of algebras Let F be a field and let A, B be two associative F -algebras. Let Lin(A, B) denote the vector space of all F -linear transformations A → B. We will define multiplication on Lin(B, B ⊗F A). f, g ∈ Lin(B, B ⊗F A). For an arbitrary element b ∈ B, g(b) = Pi bi ⊗ ai, where ai ∈ A, bi ∈ B. Let f (bi) = Pj where aij ∈ A, bij ∈ B. Define Let let bij ⊗ aij, (f g)(b) =Xi,j bij ⊗ aijai. In other words, if µ : A ⊗ A → A is the multiplication on A, then f g = (1 ⊗ µ)(f ⊗ 1)g. Choose an arbitrary basis {bi}i∈I of the algebra B and a linear trans- formation f : B → B ⊗F A. Let f (bj) =Xi bi ⊗ aij. Consider the I × I matrix Af = (aij)I×I. Each column of this matrix contains only finitely many nonzero entries aij. MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS4 so Af = (aij)I×I, Ag = (a′ Let f, g ∈ Lin(B, B ⊗F A). Let f (bj) =Pi (f g)(bj) = (1 ⊗ µ)(f ⊗ 1)(Xi ij)I×I. Then bi ⊗ a′ ij) bi ⊗aij, g(bj) =Pi bi ⊗a′ ij, = (1 ⊗ µ)Xk,i bk ⊗Xi =Xk bk ⊗ aki ⊗ a′ ij akia′ ij, which implies that Af g = Af Ag. Let fMI×I(A) denote the algebra of I × I matrices over A having finitely many nonzero entries in each column. We proved that every basis of the algebra B gives rise to an isomorphism Lin(B, B ⊗F A) ∼= fMI×I(A). Let's define a structure of a B-bimodule on Lin(B, B ⊗F A). For an arbitrary element b ∈ B and a linear transformation f : B → B ⊗F A, we will define linear transformations f b and bf via: (f b)(b′) = f (bb′), b′ ∈ B, and (bf )(b′) = (b ⊗ 1)f (b′). In other words, if f (b′) =Pi bi ⊗ ai, then (bf )(b′) =Pi check that this is indeed a B-bimodule. bbi ⊗ ai. We will Choose arbitrary elements b1, b2 ∈ B. Then ((f b1)b2)(b) = (f b1)(b2b) = f (b1b2b) = (f (b1b2))(b), hence (f b1)b2 = f (b1b2). Also, (b1(b2f ))(b) = (b1 ⊗ 1)(b2f (b)) = (b1 ⊗ 1)(b2 ⊗ 1)f (b) = (b1b2 ⊗ 1)f (b) = ((b1b2)f )b, hence b1(b2f ) = (b1b2)f . Finally, ((b1f )b2)(b) = (b1f )(b2b) = (b1 ⊗ 1)f (b2b). On the other hand, (b1(f b2))(b) = (b1 ⊗ 1)(f b2)(b) = (b1 ⊗ 1)f (b2b). Hence, (b1f )b2 = b1(f b2). MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS5 Now, consider the semidirect sum A ≀ B = B + Lin(B, B ⊗F A) that extends multiplications on B and on Lin(B, B ⊗F A). Theorem. A ≀ B is an associative algebra. Proof. From the isomorphism Lin(B, B ⊗F A) ∼= fMI×I(A), we con- clude that the algebra Lin(B, B ⊗F A) is associative. We checked above that Lin(B, B ⊗F A) is a bimodule over the associative alge- bra B. Hence, it remains to check that for arbitrary elements b′ ∈ B; f, g ∈ Lin(B, B ⊗F A), we have (f b′)g = f (b′g), f (gb′) = (f g)b′, (b′f )g = b′(f g). Indeed, let b ∈ B and let g(b) =Pi (f b′ ⊗ 1)g(b) =X(f b′)(bi) ⊗ ai =X f (b′bi) ⊗ ai. bi ⊗ ai. Then Therefore, Now consider the element (f (b′g))(b). We have f (b′bi) ⊗ ai. b′bi ⊗ ai. ((f b′)g)(b) = (1 ⊗ µ)Xi (b′g)(b) = (b′ ⊗ 1)g(b) =Xi b′bi ⊗ ai) =Xi (f ⊗ 1)(Xi f (b′bi) ⊗ ai. Applying f ⊗ 1, we get Finally, (f (b′g))(b) = (1 ⊗ µ)(f ⊗ 1)(b′g)(b) = (1 ⊗ µ)Xi f (b′bi) ⊗ ai. Hence, (f b′)g = f (b′g). Next, ((f g)b′)(b) = (f g)(b′b) = (1 ⊗ µ)(f ⊗ 1)g(b′b). On the other hand, (f (gb′))(b) = (1 ⊗ µ)(f ⊗ 1)(gb′)(b). Now, g(b′b) = (gb′)(b) shows that (f g)b′ = f (gb′). We will show that (b′f )g = b′(f g). We have ((b′f )g)(b) = (1 ⊗ µ)(b′f ⊗ 1)g(b) = (1 ⊗ µ)(b′ ⊗ 1 ⊗ 1)(f ⊗ 1)g(b), MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS6 whereas (b′(f g))(b) = (b′ ⊗ 1)(1 ⊗ µ)(f ⊗ 1)g(b). Now it remains to notice that (1 ⊗ µ)(b′ ⊗ 1 ⊗ 1) = (b′ ⊗ 1)(1 ⊗ µ), which completes the proof of the proposition. (cid:3) We call A ≀ B = B + Lin(B, B ⊗F A) the matrix wreath product of the algebras A, B. We remark that the above construction was preceded and inspired by (i) constructions of examples in the paper [6] by J. Bell, L. Small, and A. Smoktunowicz, and (ii) a different definition of wreath products by Leavitt path algebras in the paper [1] by A. Alahmadi and H. Alsulami. If BM is a left module over the algebra B, then we can define A ≀M B = B + Lin(M, M ⊗F A). 3. Properties of matrix wreath products Fix an element b ∈ B. For a linear transformation γ : B → A, consider an element cγ ∈ Lin(B, B ⊗F A), cγ(b′) = b ⊗ γ(b′) for an arbitrary element b′ ∈ B. Clearly, ρb = {cγγ ∈ Lin(B, A)} is a right ideal of the algebra Lin(B, B ⊗F A). Consider the algebra S(A, B) =Xb∈B ρb ⊳r Lin(B, B ⊗F A). The algebra S(A, B) consists of linear transformations ϕ : B → B ⊗F A such that there exists a finite dimensional subspace V ⊂ B with ϕ(B) ⊆ V ⊗F A. Once we choose a basis {bi, i ∈ I} of the algebra B and thus define an isomorphism Lin(B, B ⊗F A) ∼= fMI×I(A), the algebra S(A, B) consists of (infinite) I ×I matrices having finitely many nonzero rows. Recall that the algebra M∞(A) consists of I ×I matrices having finitely many nonzero entries. MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS7 In this section, we will study ring theoretic properties of S(A, B) and of subalgebras B + S < A ≀ B, where M∞(A) ⊆ S ⊆ S(A, B). First, we will determine conditions for B + S to be prime. Recall that an algebra is said to be prime if the product of any two nonzero ideals is not equal to zero. In what follows, we assume that the algebra B does not contain nonzero element b such that dimF bB < ∞. Let A denote the unital hull of the algebra A, i.e., A = A if A contains 1, otherwise A = A + F · 1. For an element b ∈ B, let Lb denote the operator of left multiplication Lb : B → B, x → bx. The operator Lb can be viewed as a mapping Lb : B → B ⊗ 1, hence Lb ∈ Lin(B, B ⊗F A). Denote LB = {Lb, b ∈ B} < Lin(B, B ⊗F A). Lemma 3.1. (1) LBS(A, B) + S(A, B)LB ⊆ S(A, B), (2) LB ∩ S(A, B) = (0). Proof. Let ϕ ∈ S(A, B), let V ⊂ B be a finite dimensional subspace such that ϕ(B) ⊆ V ⊗ A. Let b ∈ B. Then (Lbϕ)(B) ⊆ bV ⊗ A, (ϕLb)(B) ⊆ ϕ(B) ⊆ V ⊗ A, which proves (1). If b ∈ B and Lb ∈ S(A, B), then dimF bB < ∞. By our assumption, (cid:3) it implies that b = 0. This completes the proof of the lemma. Let M∞(A) ⊆ S ⊆ S(A, B) be a subalgebra such that BS +SB ⊆ S. Proposition 3.2. The algebra B + S is prime if and only if the algebra A is prime. Proof. Suppose that the algebra B + S is prime. If J1, J2 are nonzero ideals of A, then M∞(J1), M∞(J2) are nonzero left ideals of the algebra B + S. If J1J2 = (0), then M∞(J1)M∞(J2) = (0). It is well known that a product of two nonzero left ideals in a prime algebra is not equal to zero. That contradicts the primeness of B + S. Suppose now that the algebra A is prime. Then the algebra M∞(A) is prime as well. If K1, K2 are nonzero ideal of B + S such that K1K2 = (0), then K1 ∩ M∞(A) = (0) or K2 ∩ M∞(A) = (0). Since MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS8 M∞(A) is a left ideal of B + S, it follows that K1M∞(A) = (0) or K2M∞(A) = (0). Hence, M∞(A) has a nonzero left annihilator in B + S. Let 0 6= b ∈ B, s ∈ S, and suppose that (b + s)M∞(A) = (0). Since the algebra M∞(A) has zero left annihilator in fMI×I(A), it fol- lows that fMI×I(A)(b + s) = (0). For an arbitrary element f ∈ Lin(B, B ⊗F A), we have (f b)(b′) = f (bb′) = (f Lb)(b′). Hence, Lb + s = 0. By Lemma 3.1 (1), b = 0 and it remains to recall again that M∞(A) has zero left (right) an- nihilators in fMI×I(A). This completes the proof of the proposition. (cid:3) Next we will find conditions for B + S to be primitive. Recall that an algebra is said to be (left) primitive if it has a faithful irreducible left module [12]. Lemma 3.3. Let R be a prime algebra with a nonzero left ideal L ⊳e R. Suppose that (1) {ℓ ∈ LLℓ = (0)} = (0), (2) for arbitrary n ≥ 1; arbitrary elements a ∈ R; and ℓ1, · · · , ℓn ∈ L, there exists an element ℓ′ ∈ L such that (a − ℓ′)ℓi = 0, 1 ≤ i ≤ n. Then the algebra R is primitive if and only if the algebra L is primitive. Proof. Let M be a faithful irreducible left module over R. Consider the subspace M ′ = {m ∈ MLm = (0)}. Because of faithfulness of M, we have M ′ (cid:0) M. We will show that the factor space M/M ′ is a faithful irreducible L-module. Indeed, if ℓ ∈ L and ℓ(M/M ′) = (0), then LℓM = (0), which implies that Lℓ = (0). From (1), we conclude that ℓ = 0. We will show that the L-module M/M ′ is irreducible. Let 0 6= m + M ′ ∈ M/M ′. Then Lm = M, which implies L(m + M ′) = M/M ′. Now suppose that the algebra L is primitive and M is a faithful irreducible left module over L. We will define a structure of an R- module on M. Since LM = M, an arbitrary element of M can be represented as ℓimi, ℓi ∈ L, mi ∈ M. For an element a ∈ R, nPi=1 MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS9 we define a( ℓimi) = nPi=1 nPi=1 (aℓi)mi. To check that this action is well defined, we have to show that ℓimi = 0 implies (aℓi)mi = 0. By assumption (2), these exists an element ℓ′ ∈ L such that aℓi = ℓ′ℓi, ℓ′ℓimi = 0. This completes the 1 ≤ i ≤ n. Hence, (aℓi)mi = nPi=1 proof of the lemma. (cid:3) nPi=1 nPi=1 nPi=1 Proposition 3.4. The algebra B + S is primitive if and only if the algebra A is primitive. Proof. Suppose that the algebra B + S is primitive. Since a nonzero two sided ideal of a primitive algebra is primitive, we conclude that the algebra S is primitive and therefore prime. The algebra A is also prime by Proposition 3.2. We will check whether the algebra S and its left ideal M∞(A) satisfy assumptions (1), (2) of Lemma 3.3. Part (1) is trivial. Now we will check assumption (2). Choose elements a ∈ S; a1, · · · , an ∈ M∞(A). Let b1, · · · , bm be elements of the basis of the algebra B such that a1, · · · , an have only nonzero rows that correspond to b1, · · · , bm. In other words, a1, · · · , am ∈ ρbi. mPi=1 Let a′ be the I × I matrix that has the same entries as a in the columns that correspond to b1, · · · , bm and zeros everywhere else. Then a′ ∈ M∞(A) and aai = a′ai, 1 ≤ i ≤ n. By Lemma 3.3, the left ideal M∞(A) of the algebra S is a primitive algebra, which implies primitivity of the algebra A. Now suppose that the algebra A is primitive. By Proposition 3.2, the algebra B + S and S are prime. By Lemma 3.3, the algebra S is primitive. It is easy to see that if a nonzero ideal of a prime algebra is primitive, then the full algebra is primitive as well. This finishes the proof of the proposition. (cid:3) In the rest of this section, we will study growth of some subalgebras in A ≀ B. We will recall some definitions. MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS10 Let R be an F -algebra generated by a finite dimensional subspace V . Let V n = spanF (v1 · · · vkk ≤ n, vi ∈ V, 1 ≤ i ≤ k). Then dimF V n < ∞ and R is the union of the ascending chain V 1 ⊆ V 2 ⊆ · · · . The function g(V, n) = dimF V n is called the growth function of the algebra R that corresponds to the generating subspace V . Given two functions f1, f2 : N → [1, ∞), we say that f1 is asymptot- ically less than or equal to f2 (denote: f1 (cid:22) f2) if there exists c ∈ N such that f1(n) ≤ cf2(cn) for all n. If f1 (cid:22) f2 and f2 (cid:22) f1, then we say that f1 and f2 are asymptotically equivalent (denote: f1 ∼ f2). If V1, V2 are two finite dimensional generating subspaces of R, then g(V1, n) ∼ g(V2, n). We will denote the class of functions that are equivalent to g(V, n) as gR(n). If there exists α > 0 such that gR(n) (cid:22) nα, then we say that growth of R is polynomially bounded. In this case GK dim(R) = inf{α > 0gR(n) (cid:22) nα} is called the Gelfand-Kirillov dimension of R. If R does not have poly- nomially bounded growth, then GK dim(R) = ∞. For a not necessarily finitely generated algebra R, we let GK dim(R) = sup GK dim(R′), where R′ runs over all finitely generated subalgebras of R. Coming back to the algebras A, B, we say that a linear transforma- tion γ : B → A is a generating linear transformation if γ(B) generates A. Now suppose that the algebra B contains 1. Let γ : B → A be a generating linear transformation. As above, we consider the element cγ : b → 1 ⊗ γ(b) ∈ B ⊗F A. If a ∈ A, then we denote acγ = cγ ′, where γ′(b) = aγ(b). Consider the subalgebra C = hB, cγi generated in A ≀ B by B and the element cγ. MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS11 If V is a generating subspace of the algebra B, then U = V + F cγ is a generating subspace of the algebra C. For n ≥ 1, consider the vector space Wn = Xi1+···+ir≤n γ(V i1) · · · γ(V ir ) ⊆ A. Clearly, W1 ⊆ W2 ⊆ · · · , A =Sn≥1 Wn. Lemma 3.5. U n ⊆ Pi+j+k≤n V i(Wjcγ)V k + V n for any n ≥ 1. Proof. For n = 1, the assertion is obvious. We denote the right hand side of the inclusion above as RHS(n). We need to show that U RHS(n − 1) ⊆ RHS(n). Clearly, V RHS(n − 1) ⊆ RHS(n). Let Then cγv = cγ ′, where γ′(b) = γ(vb). We have v ∈ V i. γ′(1) = γ(v) ∈ Wi. Now, cγv(Wjcγ) = c′ γ(Wjcγ) = (γ′(1)Wj)cγ ⊆ Wi+jcγ. Therefore, cγV i(Wjcγ)V k ⊆ (Wi+jcγ)V k ⊆ RHS(n − 1) ⊆ RHS(n). This completes the proof of the lemma. (cid:3) Denote wγ(n) = dimF Wn. Corollary 3.6. gC(n) (cid:22) g2 B(n)wγ(n). If A ∋ 1, then along with the algebra C, we will consider a bigger algebra C ′ = hB, cγ, e11(1)i and its generating subspace U ′ = V + F cγ + F e11(1). Lemma 3.7. U ′n ⊆ Xi+j+k≤n V i(Wjcγ)V k + V n + Xi+j+k≤n V ie11(Wj)V k. Proof. Again, denote the right hand side of the inclusion as RHS(n). V ie11(Wj)V k ⊆ RHS(n) and We need to check that cγ Pi+j+k≤n−1 e11(1) RHS(n − 1) ⊆ RHS(n). The subspace e11(Wj) lies in ρ1. Hence, cγV ie11(Wj) ⊆ e11(γ(V i)Wj) ⊆ e11(Wi+j) MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS12 and therefore e11(Wi+j)V k ⊆ RHS(n − 1). Furthermore, e11(1)V i(Wjcγ)V k = e11(1)V ie11(1)(Wjcγ)V k and it remains to notice that e11(1)V ie11(1) = F e11(1). This completes the proof of the lemma. (cid:3) Corollary 3.8. gC ′(n) (cid:22) g2 B(n)wγ(n). We say that a linear transformation γ : B → A is dense if for ar- bitrary linearly independent elements b1, · · · , bn ∈ B and arbitrary nonzero element a ∈ A, there exists an element b ∈ B such that γ(bib) = 0, 1 ≤ i ≤ n − 1, and aγ(bnb) 6= 0. Lemma 3.9. If γ : B → A is a dense generating linear transformation, then gC(n) ∼ gB(n)2wγ(n). Proof. It is easy to see that V n(Wncγ)V n ⊆ U 3n. We will show that dimF V n(Wncγ)V n = (dimF V n)2w(n). Let b1, · · · , br be a basis of V n and let a1, · · · , at be a basis of Wn. We need to verify that elements bi(ajcγ)bk are linearly independent. For an arbitrary element b ∈ B and arbitrary coefficients γijk ∈ F , we have (cid:16)X γijkbi(ajcγ)bk(cid:17)(b) =X γijkbi ⊗ ajγ(bkb). Since the elements bi are linearly independent, it follows that for every i, Xj,k γijkajγ(bkb) = 0. Let γi0j0k0 6= 0. By density of γ, there exists an element b ∈ B such that γ(bℓb) = 0 for ℓ 6= k0 and (P γi0jk0aj)γ(bk0b) 6= 0, a contradiction. (cid:3) Lemma 3.10. Suppose that the algebra B has a basis b1, b2, · · · that consists of invertible elements. Suppose that A ∋ 1. The basis {bi}i∈I defines the isomorphism Lin(B, B ⊗F A) ∼= fMI×I(A). Let γ : B → A be a generating linear transformation. Then the algebra C ′ = hB, cγ, e11(1)i contains M∞(A). MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS13 Proof. We have eij(1) = bie11(1)b−1 , hence C ′ ⊇ M∞(F ). If γ(bi) = a, then cγbe11(1) = e11(a). Since γ is a generating linear transformation, it follows that C ′ ⊇ e11(A). Now it remains to notice that M∞(F ) and e11(A) generate M∞(A). (cid:3) j 4. Radical Algebras In this section, we will prove embedding theorems 4.1-4.3 for Jacob- son radical algebras. Lemma 4.1. For an arbitrary Jacobson radical algebra A, there exists a Jacobson radical algebra eA and an element u ∈ eA, u3 = 0, such that A is embeddable in the right ideal ueA (resp. left ideal eAu). Proof. Consider the two dimensional nilpotent algebra B with a basis b1 = b, b2 = b2, b3 = 0. Let A be a Jacobson radical algebra. Consider the matrix wreath product A ≀ B = B + M2(A). Clearly, A ≀ B is a Jacobson radical algebra. For 1 ≤ i, j ≤ 2 and an element a ∈ A, we consider the linear transformation eij(a) that maps a basic element bk to δikbj ⊗ a. Then be21(a) = e22(a). Hence e22(A) ⊆ b(A ≀ B), which completes the proof of the lemma. (cid:3) Proof of Theorem 4.1. Let A be a countable dimensional Jacobson rad- ical algebra. By Lemma 4.1, there exists a countable dimensional Ja- Let B be a finitely generated infinite dimensional nil algebra of E. cobson radical algebra eA and an element u ∈ eA, u3 = 0, such that A embeds in eAu. S. Golod [9]. Let B = B + F · 1 be its unital hull. Let γ : B → eA be a generating linear transformation. In the matrix wreath product eA ≀ B, consider the element cγ : B → B ⊗F eA, cγ(b) = 1 ⊗ γ(b). an isomorphism Lin( B, B ⊗F eA) → fMI×I(eA). Consider the element e11(u) ∈ Lin( B, B ⊗F eA) that sends b1 = 1 to 1 ⊗ u and sends other basic elements to zero. Consider the subalgebra C of eA ≀ B generated by B, Cγ, e11(u). Choose a basis {bi}i∈I of the algebra B, b1 = 1. It gives rise to MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS14 Consider also the right ideal ρ1 = {cαα ∈ Lin( B, eA), cα(b) = 1 ⊗ α(b)}. For any α, β ∈ Lin( B, eA), we have cαcβ = α(1)cβ. Hence, the mapping π : ρ1 → eA, π(cα) = α(1) is a homomorphism. The subalgebra hcγ Bi generated by cγ B lies in ρ1. For any basic element bi, we have π(cγbi) = γ(bi). Since γ is a generating linear transformation, it follows that the restriction of π to hcγBi is surjective. Hence C ⊇ hcγ Bie11(u) = e11(eAu) ⊇ e11(A). It remains to show that the subalgebra C is Jacobson radical. We will start by showing that the right ideal F cγ +cγC is Jacobson radical. The right ideal F cγ +cγC is contained in ρ1 and contains hcγ Bi. Hence, the restriction of the homomorphism π to F cγ + cγC is surjective. The kernel of this homomorphism lies in ρ′ 1 = {cαα(1) = 0}, with (ρ′ 1)2 = (0). This proves that the right ideal F cγ + cγC of the algebra C is Jacobson radical. Hence, cγ lies in the Jacobson radical Jac(C) of the algebra C. The ideal generated by e11(u) in the subalgebra hB, e11(u)i lies in MI×I(uF [u]), hence this ideal is nilpotent. Hence e11(u) ∈ Jac(C). Finally, it follows that C/Jac(C) = B + Jac(C)/Jac(C), a nil alge- bra, which implies that C = Jac(C). The algebra C is finitely gener- ated. This completes the proof of Theorem 4.1. (cid:3) Now we turn to Theorem 4.2. Let A be a countable dimensional algebra of Gelfand-Kirillov dimension ≤ d. Let the algebra B be gen- erated by a finite dimensional subspace V . Recall that for a linear transformation γ : B → A, we denote Wn = Xi1+···+ir≤n γ(V i1) · · · γ(V ir ), wγ(n) = dimF Wn. Lemma 4.2. There exists a generating linear transformation γ : B → A such that wγ(n) ≤ nd+ǫn, where ǫn > 0, ǫn → 0 as n → ∞. MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS15 Proof. Let a1, a2, · · · be a basis of the algebra A. Let gk(n) = dimF span(ai1 · · · air ; 1 ≤ r ≤ n; 1 ≤ i1, · · · , ir ≤ k). From GK dim A ≤ d, it follows that there exists an increasing sequence nk, k ≥ 1, such that gk(n) ≤ nd+ 1 k as soon as n ≥ nk. For n ≥ n1, choose k such that nk ≤ n < nk+1. Let ǫn = 1 k . It is clear that ǫn → 0 as n → ∞. Choose a subspace V ′ k ⊂ V nk and an element vk ∈ V nk such that V nk = V nk−1 ⊕ V ′ k ⊕ F vk is a direct sum of subspaces. Then B = V ′ 2 ⊕F v2⊕· · · . Define a linear transformation γ : B → A via γ(V ′ 1 ⊕F v1⊕V ′ i ) = 0, i ≥ 1, γ(vi) = ai. The subspace Wn is spanned by γ(V i1) · · · γ(V ir ), i1 + · · · + ir ≤ n. Hence, γ(V i1) · · · γ(V in) ⊆ spanF (aj1 · · · ajr; 1 ≤ j1, · · · , jr ≤ k). Now we get wγ(n) ≤ gk(n) ≤ nd+ 1 k as n ≥ nk. This completes the proof of the lemma. (cid:3) Proof of Theorem 4.2. Let A be a countable dimensional Jacobson rad- ical algebra of Gelfand-Kirillov dimension ≤ d. Let the ground field F be countable. In [16], T. Lenagan and A. Smoktunowicz constructed a finitely generated nil F -algebra of finite Gelfand-Kirillov dimension. In [17], T. Lenagan, A. Smoktunowicz, and A. Young refined the argument of [16] to construct a finitely generated nil algebra B of Gelfand-Kirillov dimension ≤ 3. Following Lemma 4.2, there exists a generating linear transforma- tion γ : B → eA such that wγ(n) ≤ nd+ǫn, ǫn → 0 as n → ∞. As shown above, the algebra A embeds in a finitely generated algebra C ′ = hB, cγ, e11(u)i. By Corollary 3.8, gc′(n) (cid:22) gB(n)2wγ(u). This implies GK dim C ′ ≤ d + 6. This completes the proof of Theorem 4.2. (cid:3) For the proof of Theorem 4.3, we need to recall more details about the Golod-Shafarevich inequality (see [10]) and Golod's construction [9]. Let F hx1, · · · , xmi be the free associative algebra on m free genera- tors, m ≥ 2. We consider the free algebra without 1, i.e., it consists of formal linear combinations of nonempty words in x1, · · · , xm. Assigning degree 1 to all variables x1, · · · , xm, we make F hx1, · · · , xmi a graded algebra. The degree deg(a) of an arbitrary element a ∈ F hx1, · · · , xmi MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS16 is defined as the minimal degree of a nonzero homogeneous component of a. Let R ⊂ F hx1, · · · , xmi be a subset containing finitely many elements of each degree. Golod-Shafarevich Condition: If there exists a number 0 < t0 < 1 such that Xa∈R tdeg(a) 0 < ∞ and 1 − mt0 +Xa∈R tdeg(a) 0 < 0, then the algebra hx1, · · · , xmR = 0i presented by the set of generators x1, · · · , xm and the set of relations R is infinite dimensional. if lim n→∞ Recall that a function g : N → [1, ∞) is said to be subexponential g(n) eαn = 0 for any α > 0. A finitely generated algebra A has subexponential growth if its growth function gA(n) is subexponential. It is equivalent to gA(n) (cid:22) en. A (not necessarily finitely generated) algebra A is of locally subex- ponential growth if every finitely generated subalgebra of A is of subex- ponential growth. Lemma 4.3. Let F be a countable field and let A be a countable di- mensional F -algebra of locally subexponential growth. Then there exists a subset R ⊂ F hx1, · · · , xmi satisfying the Golod-Shafarevich condition and such that the algebra F hx1, · · · , xmR = 0i ⊗F A is nil. Proof. The algebra F hx1, · · · , xmi ⊗F A is countable. Let F hx1, · · · , xmi ⊗F A = {f1, f2, · · · }. Choose 1 m < t0 < 1 and a sequence ǫ1, ǫ2, · · · > 0 such that ǫi < ∞ ∞Pi=1 and 1 − mt0 + ∞Pi=1 aj ∈ A, Vi = Pj ǫi < 0. Choose i ≥ 1. Let fi ∈ F hx1, · · · , xmi ⊗ aj, F aj. Then for an arbitrary n ≥ 1, we have kPj=1 i . Let gVi(n) = dimF V n f n i ∈ F hx1, · · · , xmin ⊗ V n i . Since the function gVi(n) is subexponential, there exists ni ≥ 1 such that for all n ≥ ni, 0 ≤ ǫi. Let r = gVi(ni), let vi1, · · · , vir be a basis of V ni we have gVi(n)tn and let f ni fij ⊗ vij, deg fij ≥ ni. i i = rPj=1 MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS17 Let R = {fiji ≥ 1, 1 ≤ j ≤ gVi(ni)}. The image of an element fi in F hx1, · · · , xmR = 0i ⊗F A is nilpotent of index ≤ ni. Besides, X gVi(ni)tdeg(fij ) 0 ≤X gVi(ni)tni 0 ≤X ǫi. Hence, R satisfies the Golod-Shafarevich Condition and therefore the algebra F hx1, · · · , xmR = 0i⊗F A is an infinite dimensional nil algebra. This completes the proof of the lemma. (cid:3) Lemma 4.4. Let F be an arbitrary field. There exists an infinite di- mensional finitely generated stable nil F -algebra. Proof. Let F0 be the prime subfield of F . We will apply Lemma 4.3 to the countable dimensional F0-algebra A = F0[ti, i ≥ 1] ⊗F0 M∞(F0) of locally subexponential growth. By Lemma 4.3, there exists a sub- set R ⊂ F0hx1, · · · , xmi satisfying the Golod-Shafarevich condition such that the F0-algebra F0hx1, · · · , xmR = 0i ⊗F0 A is nil. We will show that the F -algebra F hx1, · · · , xmR = 0i is stable nil. In- deed, we need to check only that for arbitrary elements a1, · · · , ak ∈ F0hx1, · · · , xmR = 0i, arbitrary elements α1, · · · , αk ∈ F and arbitrary matrices y1, · · · , yk ∈ Mn(F0), the tensor kPi=1 It follows from the nilpotency of the element kPi=1 ai ⊗ αi ⊗ yi is nilpotent. ai ⊗ ti ⊗ yi of the al- gebra F0hx1, · · · , xmR = 0i ⊗F0 A. This completes the proof of the lemma. (cid:3) Lemma 4.5. Let A be a stable nil (resp. algebraic) algebra. Then for an arbitrary algebra B, the algebra S(A, B) is nil (resp. algebraic). Proof. Recall that S(A, B) consists of I ×I matrices over A with finitely many nonzero rows. For a finite subset T ⊂ I, let ST (A, B) consist of such matrices that for any i ∈ I \ T , the ith row is zero. The algebra S(A, B) is a union of subalgebras ST (A, B). Consider the mapping ϕ ST (A, B) −→ MT ×T (A). For an arbitrary matrix Y ∈ ST (A, B), the matrix ϕ(Y ) is the part of Y at the intersection of rows and columns indexed by T . Clearly, ϕ is a homomorphism and (kerϕ)2 = (0). This implies that the algebra ST (A, B) is nil (resp. algebraic) and completes the proof of the lemma. (cid:3) MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS18 Proof of Theorem 4.3. Let B be an infinite dimensional finitely gener- ated stable nil algebra of Lemma 4.4. If A is a countable dimensional stable nil algebra, then the algebra eA = A ≀ F [uu3 = 0], A ֒→ eAu, is also countable dimensional and stable nil due to Lemma 4.1. In the proof of Theorem 4.1, we embedded the algebra A in a finitely show that for an arbitrary n ≥ 1, the matrix algebra Mn(C) is nil. It generated subalgebra C = hB, Cγ, e11(u)i of eA ≀ B. Now we need to is easy to see that Mn(C) is a subalgebra of Mn(eA) ≀ Mn( B). Moreover, By Lemmas 3.1 (1), 4.5, S(Mn(eA), Mn( B)) is a nil ideal of the algebra Mn(B) + S(Mn(eA), Mn( B)). The algebra Mn(B) is nil because the Mn(C) ⊆ Mn(B) + S(Mn(eA), Mn( B)). algebra B is stable nil. We proved that the algebra C is stable nil. Suppose now that the algebra A is stable nil, the ground field F is countable, and GK dim A ≤ d. Let B be the Lenagan-Smoktunowicz- Young ([16], [17]) nil algebra, GK dim B ≤ 3. Arguing as above, we see that the finitely generated algebra C, in which the algebra A is embedded, is nil. By Corollary 3.6, the growth of C is bounded by n6wγ(n). By Lemma 4.2, a generating linear transformation γ can be chosen so that wγ(n) ≤ nd+ǫn, where ǫn → 0 as n → ∞. This implies that GK dim C ≤ d + 6 and finishes the proof of the theorem. (cid:3) Remark. If we knew that there exists a Lenagan-Smoktunowicz algebra that is stable nil, then we could embed a countable dimensional stable nil algebra of finite Gelfand-Kirillov dimension in a finitely generated stable nil algebra of finite Gelfand-Kirillov dimension. 5. Algebraic Primitive Algebras The purpose of this section is to prove Theorems 5.1, 5.2. Let A be a countable dimensional stable algebraic primitive algebra. Without loss of generality, we will assume that A ∋ 1. Let B be an infinite dimen- sional finitely generated stable nil algebra of Lemma 4.4. Without loss of generality, we will also assume that {b ∈ B dimF bB < ∞} = (0). Consider the matrix wreath product A ≀ B and an element cγ ∈ Lin( B, B ⊗F A), cγ(b) = 1 ⊗ γ(b), where γ : B → A is a gener- ating linear transformation. Choose a basis {bi}i∈N of the algebra B MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS19 that consists of invertible elements. This basis defines an isomorphism Lin( B, B ⊗F A) ∼= fMN ×N (A). As above, algebra C ′ = h B, cγ, e11(1)i. By Lemma 3.10, M∞(A) ⊆ C ′. Since M∞(A) generated consider we the finitely is a left ideal in fMN ×N (A) and BM∞(A) ⊆ M∞(A), it follows that M∞(A) is a left ideal in C ′. Arguing precisely as in the proof of Theorem 4.3 and using Lemma 4.5, we can show that the algebra C ′ is stable algebraic. By Proposition 3.4, the algebra C ′ is primitive. By a theorem of V. T. Markov [19], there exists n ≥ 1 such that the matrix algebra Mn(C ′) is 2-generated. Since Mn(M∞(A)) ∼= M∞(A), the algebra is still M∞-embedded in Mn(C ′) as a left ideal. The algebra Mn(C ′) is still stable algebraic and primitive. This finishes the proof of Theorem 5.1. Assume now that the ground field F is countable, the algebra A is stable algebraic and primitive, and GK dim A ≤ d. For the algebra B, we now take the Lenagan-Smoktunowicz-Young finitely generated nil algebra of Gelfand-Kirillov dimension ≤ 3 (see [16], [17]). Then the algebra C ′ in our construction above is finitely gener- ated and nil (though not necessarily stable nil) and M∞(A) ⊳ℓ C ′. By Lemma 4.2, we can choose a generating linear transformation γ so that wγ(n) ≤ nd+ǫn, ǫn → 0, n → ∞. Then by Corollary 3.6, GK dim C ′ ≤ d + 6, which finishes proof of Theorem 5.2. Remark. If we knew that an infinite dimensional stable nil algebra of finite Gelfand-Kirillov dimension exists, then we could embed an arbi- trary countable dimensional stable algebraic primitive algebra of finite Gelfand-Kirillov dimension in a 2-generated stable algebraic primitive algebra of finite Gelfand-Kirillov dimension, thus answering the second part of question 7 in [6]. MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS20 6. Examples of finitely generated nil algebras of arbitrary Gelfand-Kirillov dimension d ≥ 8 Everywhere in this section, we assume that the ground field F is countable. Let B be an infinite dimensional graded finitely generated Lenagan-Smoktunowicz-Young nil algebra of Gelfand-Kirillov dimen- sion ≤ 3 ([16], [17]). Without loss of generality, we will assume that ℓ(B) = {b ∈ BbB = (0)} = (0). 6.1. For arbitrary elements Lemma b1, · · · , bn ∈ B and arbitrary s ≥ 1, there exists an element b ∈ Bs such that the elements b1b, · · · , bnb are still linearly independent. independent linearly Proof. We will induct on n. For n = 1, the assertion of the lemma means that b1Bs 6= (0), which follows from the assumption on the left annihilator of B. Suppose that the assertion is true for n − 1. Choose an element b ∈ Bs such that the elements b1b, · · · , bn−1b are linearly indepen- Bi = (0), it follows that there exists t ≥ s such that dent. Since Ti≥1 spanF (b1b, · · · , bn−1b) ∩ Bt = (0). Again, by the inductive assumption, we can choose an element b′ ∈ Bt such that b1b′, · · · , bn−1b′ are linearly independent elements. Assuming that the assertion of the lemma is wrong, there exist scalars α1, · · · , αn−1, β1, · · · , βn−1 ∈ F such that bnb = n−1Xi=1 αibib, bnb′ = n−1Xi=1 βibib′. Since the elements b1(b + b′), · · · , bn−1(b + b′) are linearly independent, there exist scalars γ1, · · · , γn−1 ∈ F such that bn(b + b′) = n−1Xi=1 γibi(b + b′). Subtracting the first two equations from the third, we get n−1Xi=1 (γi − αi)bib + n−1Xi=1 (γi − βi)bib′ = 0. It implies that αi = βi = γi, 1 ≤ i ≤ n − 1. MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS21 We will use the following elementary statement from Linear Algebra: Let V be a vector space over an infinite field. Let v1, · · · , vn ∈ V be arbitrary elements and let w1, · · · , wn ∈ V be linearly independent elements. Then there exists a scalar ξ ∈ F such that the elements vi + ξwi, 1 ≤ i ≤ n, are linearly independent. This implies that for an arbitrary element b′′ ∈ Bt, there exists a scalar ξ ∈ F such that the elements bi(b′′ + ξb′), 1 ≤ i ≤ n − 1, are linearly independent. Taking b′′ + ξb′ instead of b′, we get (bn − n−1Xi=1 αibi)(b′′ + ξb′) = 0, and therefore (bn − n−1Xi=1 αibi)b′′ = 0, (bn − n−1Xi=1 αibi)Bt = (0). This contradicts the assumption that the left annihilator of B is zero and completes the proof of the lemma. (cid:3) Since the ground field F is countable, it follows that the algebra B is countable. Let B be the set of all nonempty finite sequences of linearly indepen- dent elements of B, card B = ℵ0. Let u1, u2, · · · be a sequence of elements of B such that each element of B occurs in this sequence infinitely many times. The algebra B is generated by the homogeneous component of degree 1, V = B1, V n = Bi. nPi=1 We will construct an of sequence increasing integers 0 = n0 < n1 < n2 < · · · . Suppose that k ≥ 2 and n0, n1, · · · , nk−1 have already been constructed. Let uk = (b1, · · · , bm) ∈ B, the ele- ments b1, · · · , bm are linearly independent. By Lemma 6.1, there exists an element b ∈ Bnk−1+1 such that the elements b1b, · · · , bmb are lin- early independent. Choose nk such that nk > enk−1, nk > eek, and b1b, · · · , bmb ∈ V nk−1. This completes the construction of the sequence 0 = n0 < n1 < n2 < · · · . MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS22 For an arbitrary α ≥ 2, W. Bohro and H. P. Kraft [8] constructed a graded F -algebra R = Ri generated by two elements x, y ∈ R1, such that for any ǫ > 0 we have ∞Pi=1 nα−ǫ ≤ dimF nXi=1 Ri ≤ nα+ǫ for all sufficiently large n. Let f (n) = dimF Ri. nPi=1 Let J be a graded ideal of the free associative algebra F hx, yi such that F hx, yi/J ∼= R. Now we are ready to introduce a countable dimen- sional Let X = {x1, x2, · · · }, Y = {y1, y2, · · · }. Consider the algebra A presented by the set of generators X ∪ Y and the following set of relations: locally nilpotent algebra A. (1) xixjxk = 0, where i, j, k are arbitrary, distinct integers; (2) J(xi, xj) = (0), i 6= j, where J(xi, xj) is the image of the ideal J under the homomorphism F hx, yi → F hX, Y i, x → xi, y → xj; (3) idF hX,Y i(xi)ni+3 = (0); (4) [X, yi] = [Y, yi] = (0), i ≥ 1; (5) y2 i = 0, i ≥ 1. Let gk(n) = dimF ( Lemma 6.2. Suppose that nk ≤ n < nk+2. Then F xi)µ. nPµ=1 kPi=1 f (n) ≤ gk(n) ≤(cid:18)k 2(cid:19)f (n). Proof. Let eJ be the ideal of the free algebra F hX, Y i generated by (1)-(5). We have hxk−1, xki ∩ eJ ⊆ J(xk−1, xk) + Hence, ∞Xi=nk+2 F hX, Y ii. gk(n) ≥ dimF nXi=1 (hxk−1, xki/J(xk−1, xk))i = f (n). MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS23 On the other hand, relation (3) implies F hx1, · · · , xki ⊆ X1≤i6=j≤k This implies that gk(n) ≤ (cid:0)k lemma. hxi, xji + eJ. 2(cid:1)f (n) and completes the proof of the (cid:3) Now we will define a generating dense linear transformation γ : B → A. Recall that V = B1. We let V 0 = F · 1 and γ(1) = 0. Suppose that γ : V nk−1 → A has already been defined. Let uk = (b1, · · · , bm) ∈ B. Recall that in the course of choosing the num- bers nk, we first chose an element b ∈ Bnk−1+1 such that b1b, · · · , bmb are linearly independent and then choose nk large enough so that b1b, · · · , bmb ∈ V nk−1. Choose an element vk ∈ V nk \ V nk−1 and a subspace Tk ⊂ V nk so that V nk = V nk−1 ⊕ Tk ⊕ F b1b ⊕ · · · ⊕ F bmb ⊕ F vk is a direct sum of subspaces. Define γ(Tk) = 0, γ(b1b) = · · · = γ(bm−1b) = 0, γ(bmb) = yk, γ(vk) = xk. It is clear that γ is a generating linear transformation. We will show that γ is dense. Choose an element u = (b1, · · · , bm) ∈ B and a nonzero element a ∈ A. The linearly independent set u occurs infin- itely many times in the sequence u1, u2, · · · . Choose k ≥ 1 such that uk = (b1, · · · , bm) and yk does not occur in a. Then there exists an element b ∈ B such that γ(b1b) = · · · = γ(bm−1b) = 0, γ(bmb) = yk, ayk 6= 0. As above, consider the element cγ ∈ Lin( B, B ⊗F A), cγ(b) = 1⊗γ(b) for b ∈ B. Let C = hB, cγi. Proof of Theorem 6.1. We proved in §4 that C is a nil algebra. We will show that GK dim C = 2GK dim(B) + α. Indeed, by Lemma 3.9, gC(n) ∼ gB(n)2wγ(n). We will estimate wγ(n). Let nk ≤ n < nk+1. Then Wn = span(γ(V i1) · · · γ(V ir ), i1 + · · · + ir ≤ n). MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS24 Since γ(V ni) ⊆ γ(V ni−1) + F xi + F yi for i ≥ 1, and all i1, · · · , ir are smaller than nk+1, it follows that γ(V i1) + · · · + γ(V ir ) ⊆ kXi=1 F xi + k+1Xi=1 F yi. Then wγ(u) ≤ gk(n)2k+1. By Lemma 6.2, for any ǫ > 0 and a sufficiently large n, we have gk(n) ≤ nα+ǫ(cid:0)k it follows that k < ln(ln n). Hence, for any ǫ′, 0 < ǫ < ǫ′, we have wγ(n) (cid:22) nα+ǫ′. 2(cid:1). Since n ≥ nk > eek, On the other hand, x1, · · · , xk−1 ∈ V nk−1. Hence, wγ(n) ≥ gk−1([ n nk−1 ]). We have nk−1 ≤ [ n nk−1 ] < nk+1. Hence, by Lemma 6.2, gk−1([ n nk−1 ]) ≥ f ([ n nk−1 ]). ] ≥ n From n ≥ nk ≥ enk−1+1, we conclude that nk−1 ≤ ln n − 1 and therefore [ n ln n . Hence, for an arbitrary ǫ > 0 for a sufficiently large n, nk−1 we have wγ(n) ≥ ( n ln n )α−ǫ. This implies that for any ǫ′ > ǫ, we have wγ(n) ≥ nα−ǫ′. This implies that GK dim C = 2GK dim(B) + α and completes the proof of the theorem. (cid:3) Acknowledgement The fourth author gratefully acknowledges the support form the NSF. References 1. Adel Alahmadi and Hamed Alsulami, Wreath products by a Leavitt path algebra and affinizations, Internat. J. Algebra Comput. 24 (2014), no. 5, 707 -- 714. MR 3254719 2. A. S. Amitsur, Algebras over infinite fields, Proc. Amer. Math. Soc. 7 (1956), 35 -- 48. MR 0075933 3. Laurent Bartholdi and Anna Erschler, Imbeddings into groups of intermediate growth, Groups Geom. Dyn. 8 (2014), no. 3, 605 -- 620. MR 3267517 4. Jason P. Bell, Examples in finite Gel′fand-Kirillov dimension, J. Algebra 263 (2003), no. 1, 159 -- 175. MR 1974084 MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS25 5. Jason P. Bell and Lance W. Small, A question of Kaplansky, J. Algebra 258 (2002), no. 1, 386 -- 388, Special issue in celebration of Claudio Procesi's 60th birthday. MR 1958912 6. Jason P. Bell, Lance W. Small, and Agata Smoktunowicz, Primitive algebraic algebras of polynomially bounded growth, New trends in noncommutative al- gebra, Contemp. Math., vol. 562, Amer. Math. Soc., Providence, RI, 2012, pp. 41 -- 52. MR 2905552 7. K. I. Beı dar, Radicals of finitely generated algebras, Uspekhi Mat. Nauk 36 (1981), no. 6(222), 203 -- 204. MR 643075 8. Walter Borho and Hanspeter Kraft, uber die Gelfand-Kirillov-Dimension, Math. Ann. 220 (1976), no. 1, 1 -- 24. MR 0412240 9. E. S. Golod, On nil-algebras and finitely approximable p-groups, Izv. Akad. Nauk SSSR Ser. Mat. 28 (1964), 273 -- 276. MR 0161878 10. E. S. Golod and I. R. Safarevi c, On the class field tower, Izv. Akad. Nauk SSSR Ser. Mat. 28 (1964), 261 -- 272. MR 0161852 11. Graham Higman, B. H. Neumann, and Hanna Neumann, Embedding theorems for groups, J. London Math. Soc. 24 (1949), 247 -- 254. MR 0032641 12. Nathan Jacobson, Structure of rings, American Mathematical Society Collo- quium Publications, Vol. 37. Revised edition, American Mathematical Society, Providence, R.I., 1964. MR 0222106 13. Irving Kaplansky, "Problems in the theory of rings" revisited, Amer. Math. Monthly 77 (1970), 445 -- 454. MR 0258865 14. Gottfried Kothe, Die Struktur der Ringe, deren Restklassenring nach dem ist, Math. Z. 32 (1930), no. 1, 161 -- 186. Radikal vollstandig reduzibel MR 1545158 15. Jan Krempa, Logical connections between some open problems concerning nil rings, Fund. Math. 76 (1972), no. 2, 121 -- 130. MR 0306251 16. T. H. Lenagan and Agata Smoktunowicz, An infinite dimensional affine nil algebra with finite Gelfand-Kirillov dimension, J. Amer. Math. Soc. 20 (2007), no. 4, 989 -- 1001. MR 2328713 17. T. H. Lenagan, Agata Smoktunowicz, and Alexander A. Young, Nil algebras with restricted growth, Proc. Edinb. Math. Soc. (2) 55 (2012), no. 2, 461 -- 475. MR 2928504 18. A. I. Mal′cev, On a representation of nonassociative rings, Uspehi Matem. Nauk (N.S.) 7 (1952), no. 1(47), 181 -- 185. MR 0047028 19. V. T. Markov, Matrix algebras with two generators and the embedding of PI- algebras, Uspekhi Mat. Nauk 47 (1992), no. 4(286), 199 -- 200. MR 1208894 20. B. H. Neumann and Hanna Neumann, Embedding theorems for groups, J. Lon- don Math. Soc. 34 (1959), 465 -- 479. MR 0163968 21. Alexander Yu. Olshanskii and Denis V. Osin, A quasi-isometric embedding the- orem for groups, Duke Math. J. 162 (2013), no. 9, 1621 -- 1648. MR 3079257 22. Richard E. Phillips, Embedding methods for periodic groups, Proc. London Math. Soc. (3) 35 (1977), no. 2, 238 -- 256. MR 0498874 23. Agata Smoktunowicz and Laurent Bartholdi, Jacobson radical non-nil algebras of Gel'fand-Kirillov dimension 2, Israel J. Math. 194 (2013), no. 2, 597 -- 608. MR 3047084 24. John S. Wilson, Embedding theorems for residually finite groups, Math. Z. 174 (1980), no. 2, 149 -- 157. MR 592912 MATRIX WREATH PRODUCTS OF ALGEBRAS AND EMBEDDING THEOREMS26 ADepartment of Mathematics, King Abdulaziz University, Jeddah, SA, E-mail address: [email protected]; [email protected]; BDepartment of Mathematics, Ohio University, Athens, USA, E-mail address: [email protected]; CDepartment of Mathematics, University of California, San Diego, USA E-mail address: [email protected] 1. To whom correspondence should be addressed E-mail: [email protected] Author Contributions: A.A., H.A, S. K. J., E. Z. designed research, performed research, and wrote the paper. The authors declare no conflict of interest.
1604.06671	2	1604	2017-01-05T15:30:21	Eigenvalue placement for regular matrix pencils with rank one perturbations	[ "math.RA", "math.FA" ]	A regular matrix pencil sE-A and its rank one perturbations are considered. We determine the sets in the extended complex plane which are the eigenvalues of the perturbed pencil. We show that the largest Jordan chains at each eigenvalue of sE-A may disappear and the sum of the length of all destroyed Jordan chains is the number of eigenvalues (counted with multiplicities) which can be placed arbitrarily in the extended complex plane. We prove sharp upper and lower bounds of the change of the algebraic and geometric multiplicity of an eigenvalue under rank one perturbations. Finally we apply our results to a pole placement problem for a single-input differential algebraic equation with feedback.	math.RA	math	Eigenvalue placement for regular matrix pencils with rank one perturbations Hannes Gernandt∗ Carsten Trunk†. July 10, 2018 Abstract A regular matrix pencil sE − A and its rank one perturbations are considered. We determine the sets in C ∪ {∞} which are the eigenvalues of the perturbed pencil. We show that the largest Jordan chains at each eigenvalue of sE − A may disappear and the sum of the length of all destroyed Jordan chains is the number of eigenvalues (counted with multiplicities) which can be placed arbitrarily in C ∪ {∞}. We prove sharp upper and lower bounds of the change of the algebraic and geometric multiplicity of an eigenvalue under rank one perturbations. Finally we apply our results to a pole placement problem for a single-input differential algebraic equation with feedback. Keywords: regular matrix pencils, rank one perturbations, spectral perturbation theory MSC 2010: 15A22, 15A18, 47A55 1 Introduction For square matrices E and A in Cn×n we consider the matrix pencil A(s) := sE − A (1) and study its set of eigenvalues σ(A), which is called the spectrum of the matrix pencil. Here λ ∈ C is said to be an eigenvalue if 0 is an eigenvalue of the matrix λE − A and we say that ∞ is an eigenvalue of A(s) if E is not invertible. The spectral theory of matrix pencils is a generalization of the eigenvalue problem for matrices [17, 26, 31, 35]. Recently, there is a growing interest in the spectral behavior under low rank perturbations of matrices [2, 12, 27, 30, 33, 34] and of matrix pencils [1, 13, 14, 15, 29]. For matrices it was shown in [12] that under generic rank one perturbations only the largest Jordan chain at each eigenvalue might be destroyed. Here a generic set of perturbations is a subset of Cn×n which complement is a proper algebraic submanifold. We consider only regular matrix pencils A(s) = sE − A which means that the charac- teristic polynomial det(sE − A) is not zero. Otherwise we call A(s) singular. Jordan chains for regular matrix pencils at λ correspond to Jordan chains of the matrix J at λ, for λ 6= ∞, or to Jordan chains of N at 0, for λ = ∞, in the Weierstrass canonical form [16], S(sE − A)T = s(cid:18)Ir 0 N(cid:19) −(cid:18)J 0 0 0 In−r(cid:19) , r ∈ {0, 1, . . . , n} ∗Institute of Mathematics, TU Ilmenau, Weimarer Strasse 25, 98693 Ilmenau, Germany ([email protected]). †Institute of Mathematics, TU Ilmenau, Weimarer Strasse 25, 98693 Ilmenau, Germany 1 with J ∈ Cr×r and N ∈ C(n−r)×(n−r) in Jordan canonical form, N nilpotent and invertible S, T ∈ Cn×n. In [14] it was shown that under generic rank one perturbations only the largest chain at λ is destroyed. In this paper we investigate the behaviour of the spectrum of a regular matrix pencil A(s) under a rank one perturbation P(s) := sF − G. As we will see in Section 3, the rank one condition allows us to write P(s) in the form P(s) = (su + v)w∗ or P(s) = w(su∗ + v∗) (2) with non-zero vector w and vectors u and v such that at least one of the two is not zero. Rank one perturbations of the above form are considered in design problems for electrical circuits where the entries of E are determined by the capacitances of the circuit. The aim is to improve the frequency behavior by adding additional capacitances between certain nodes. This corresponds, within the model, to a (structured) rank one perturbation of the matrix E, see [4, 7, 19]. Here we follow a more general approach and obtain the following results: (i) We find for unstructured rank one perturbations of the form (2) sharp lower and upper bounds for the dimension of the root subspace Lλ(A + P) of the perturbed pencil at λ in terms of the dimension of Lλ(A), where Lλ(A) denotes the subspace of all Jordan chains of the matrix pencil A(s) at the eigenvalue λ. More precisely, if m1(λ) denotes the length of the longest chain of A(s) at λ and let M (A) be the sum of all m1(λ) over all eigenvalues of A(s), that is, M (A) =Pλ∈σ(A) m1(λ). Then the bounds are dim Lλ(A) − m1(λ) ≤ dim Lλ(A + P) ≤ dim Lλ(A) + M (A) − m1(λ). (3) (ii) We show a statement on the eigenvalue placement for regular matrix pencils which is our main result: M (A) eigenvalues (counted with multiplicities) can be placed arbitrarily in the complex plane. Either by creating new eigenvalues with one chain only or by adding one new (or extending an old) chain at existing eigenvalues of A(s). Here the term new eigenvalue is understood in the sense that this value is an eigenvalue of (A + P)(s) but not of A(s). In addition we obtain the same result for real matrices and real rank one perturbations. Roughly speaking, the behaviour of the spectrum under rank one perturbations described in (i) and (ii) can be summarized in the following way: At each eigenvalue of A(s) the longest chain (or parts of it) may disappear but the remaining chains at that eigenvalue are then Jordan chains of (A + P)(s). Moreover, the sum M (A) of the length of all the longest chains is then the upper bound for the placement of new chains, either at existing eigenvalues or at new eigenvalues. Those new chains have to satisfy two rules. At new eigenvalues there is only one chain with maximal length M (A) and at existing eigenvalues at most one new chain may appear, again with maximal length M (A). For a precise description of this placement result we refer to Theorem 4.4 below. The left hand side of (3) is well-known [1, 14]. In the case of matrices, i.e. E = In in (1) and u = 0 in (2), we refer to [33, 34] and to [3, 20] for operators. Results similar to (i) are known in the literature for generic low-rank perturbations [1, 12, 14, 27, 33]. In the generic case it was shown in [14] that only the largest chain at each eigenvalue is eventually destroyed. In Proposition 4.2 we show a non generic result that gives a bound on the change of the Jordan chains of length k for all k ∈ N \ {0}. Similar bounds were previously obtained for matrices in [33] and for operators in [3]. In special cases the placement problem in (ii) is considered in the literature. For E positive definite and A symmetric the placement problem was studied in [15]. In the matrix case, i.e. E = In and u = 0, the placement problem was solved for symmetric A in [18]. In [22] a related inverse problem was studied: For two given subsets of the complex plane, two 2 matrices were constructed whose set of eigenvalues equal these sets and the matrices differ by rank one. All these eigenvalue placement settings above are special cases of our result in Section 5 below. In Section 6 we investigate the eigenvalue placement under parameter restrictions in in the representation (2) we fix u, v ∈ Cn. This allows us to derive the perturbation, i.e. a sharper bound as in (3). For these restricted placement problems, we obtain simple conditions on the number of eigenvalues that can be assigned arbitrarily. In the final section we present an application. We consider the pole assignment problem under state feedback for single input differential-algebraic equations. This problem is well studied in the literature [8, 23, 24, 25, 28, 32] even for singular matrix pencils, see [11] and the references therein. However, for single input systems we can view this problem as a parameter restricted rank one perturbation problem from Section 6. 2 Eigenvalues and Jordan Chains of Matrix Pencils In this section the notion of eigenvalues and Jordan chains for matrix pencils A(s) = sE − A with E, A ∈ Cn×n is recalled. Furthermore we summarize some basic spectral properties which are implied by the well known Weierstrass canonical form [16]. For fixed λ ∈ C observe that A(λ) is a matrix over C. Hence the spectrum of the matrix pencil A(s) = sE − A is defined as σ(A) := {λ ∈ C 0 is an eigenvalue of A(λ)}, if E is invertible, and σ(A) := {λ ∈ C 0 is an eigenvalue of A(λ)} ∪ {∞}, if E is singular. Obviously the spectrum of a matrix pencil is a subset of the extended complex plane C := C ∪ {∞} and the roots of the characteristic polynomial det(sE − A) are exactly the elements of σ(A)\{∞}. Hence the spectrum of regular matrix pencils consists of finitely many points. For A(s) singular one always has σ(A) = C. We recall the notion for Jordan chains and root subspaces [17, Section 1.4], [26, §11.2]). The set {g0, . . . , gm−1} ⊂ Cn is a Jordan chain of length m at λ ∈ C if g0 6= 0 and (A − λE)g0 = 0, (A − λE)g1 = Eg0, . . . , (A − λE)gm−1 = Egm−2 and we call {g0, . . . , gm−1} ⊂ Cn a Jordan chain of length m at ∞ if g0 6= 0, Eg0 = 0, Eg1 = Ag0, . . . , Egm−1 = Agm−2. Two Jordan chains {g0, . . . , gk} and {h0, . . . , hl} at λ ∈ C are called linearly independent, if the vectors g0, . . . gk, h0, . . . , hl are linearly independent. Furthermore, we say that A(s) has k Jordan chains of length m if there exist k linearly independent Jordan chains of length m at λ ∈ C. We denote for λ ∈ C and l ∈ N \ {0} the subspace of all elements of all Jordan chains up to the length l at λ by Ll λ(A) :=ngj ∈ Cn 0 ≤ j ≤ l − 1, {g0, . . . , gj} is a Jordan chain at λo and the root subspace which consists of all elements of all Jordan chains at λ, Lλ(A) := Ll λ(A). ∞[l=1 3 It is well known that regular pencils A(s) = sE − A can be transformed into the Weierstrass canonical form [16, Chapter XII, §2], i.e. there exist invertible matrices S, T ∈ Cn×n and r ∈ {0, 1, . . . , n} such that S(sE − A)T = s(cid:18)Ir 0 N(cid:19) −(cid:18)J 0 0 In−r(cid:19) , (4) 0 with J ∈ Cr×r and N ∈ C(n−r)×(n−r) in Jordan canonical form and N nilpotent. From the Weierstrass canonical form, we have some well known properties [7, 16]. Proposition 2.1. For a regular matrix pencil A(s) = sE − A with Weierstrass canonical form (4) the following holds. (a) A Jordan chain {g0, . . . , gm−1} of the matrix pencil at λ ∈ C of length m corresponds to a Jordan chain {πrT −1g0, . . . , πrT −1gm−1} ⊂ Cr of J at λ of length m. Here πr denotes the projection of x ∈ Cn onto the first r entries. Vice versa a Jordan chain {h0, . . . , hm−1} of J at λ corresponds to a Jordan chain (cid:26)T(cid:18)h0 0 (cid:19)(cid:27) 0(cid:19) , . . . , T(cid:18)hm−1 of the matrix pencil at λ. (b) A Jordan chain {g0, . . . , gm−1} of the matrix pencil at ∞ of length m corresponds to a Jordan chain {πn−rT −1g0, . . . , πn−rT −1gm−1} ⊂ Cn−r of N at 0 of length m. Here πn−r denotes the projection of x ∈ Cn onto the last n − r entries. Vice versa a Jordan chain {h0, . . . , hm−1} of N at 0 corresponds to a Jordan chain(cid:26)T(cid:18) 0 h0(cid:19) , . . . , T(cid:18) 0 hm−1(cid:19)(cid:27) of the matrix pencil at ∞. (c) A Jordan chain {g0, . . . , gm−1} of the matrix pencil at ∞ of length m corresponds to a Jordan chain {h0, . . . , hm−1} of the dual pencil A′(s) = −sA + E at 0. (d) If r ≥ 1 we have σ(A) \ {∞} = σ(J) and the characteristic polynomial of sE − A is divisible by the minimal polynomial mJ (s) of J with det(sE − A) = (−1)n−r det(ST )−1mJ (s)q(s), (5) where q(s) is a monic polynomial of degree r − deg mJ . The value λ ∈ σ(A) \ {∞} is a root of q(s) if and only if dim ker A(λ) ≥ 2. Moreover the multiplicity of a root λ of det(sE − A) is equal to dim Lλ(A) and we have Xλ∈σ(A) dim Lλ(A) = n. (6) There are dim ker A(λ) linearly independent Jordan chains at λ and, by Proposition 2.1, this corresponds to the number of linearly independent Jordan chains of J at λ 6= ∞ or N at 0 for λ = ∞. Each of these dim ker A(λ) different Jordan chains has a length which we denote by mj(λ), 1 ≤ j ≤ dim ker A(λ). These numbers mj(λ) are not uniquely determined, more precisely, they depend on the chosen Weierstrass canonical form (4) but they are unique up to permutations. In the following, we will choose those numbers in a specific way and we fix this in the following assumption. Assumption 2.2. Given a regular pencil A(s) = sE − A which has Weierstrass canonical form (4) with r ∈ {0, 1, . . . , n} and matrices J ∈ Cr×r and N ∈ C(n−r)×(n−r). Then we assume that for λ ∈ σ(A) the numbers mj(λ), 1 ≤ j ≤ dim ker A(λ), are sorted in a non-decreasing order m1(λ) ≥ . . . ≥ mdim ker A(λ)(λ). (7) 4 Observe that Assumption 2.2 is no restriction for regular pencils. This means that for every regular pencil the matrices S and T in (4) can be chosen in such a way that the Jordan blocks of J satisfy the condition (7), see [16]. Therefore with Assumption 2.2 the minimal polynomial mJ (s) of J can be written as mJ (s) = Yλ∈σ(J) (s − λ)m1(λ) and we introduce mA(s) :=(Qλ∈σ(J)(s − λ)m1(λ), 1, for r ≥ 1, for r = 0. (8) Note that with this definition the equation (5) also holds for r = 0 after replacing mJ (s) by mA(s). 3 The structure of rank one pencils In this section we study pencils of rank one. Recall that the rank of a pencil A(s) is the largest r ∈ N such that A(s), viewed as a matrix with polynomial entries, has minors of size r that are not the zero polynomial [14, 16]. This implies that A(s) has rank equal to n if and only if A(s) is regular. Hence, pencils of rank one are not regular for n ≥ 2, meaning that they cannot be transformed to Weierstrass canonical form. Nevertheless, there is a simple representation given in the following proposition. Proposition 3.1. The pencil P(s) = sF − G with F, G ∈ Cn×n has rank one if and only if there exists u, v, w ∈ Cn with w 6= 0 and (u 6= 0 or v 6= 0) such that P(s) = (su + v)w∗ or P(s) = w(su∗ + v∗). (9) If F, G are real matrices, then u, v, w can be chosen to have real-valued entries. Proof. Given that P(s) has rank one, then all minors of P(s) of size strictly larger than one are the zero polynomial. Since rk (P(λ)) = rk (λF − G) ≤ 1 for all λ ∈ C, (10) for λ = 0 we have rk (G) ≤ 1. Then there exist u, v ∈ Cn with G = uv∗. For λ = 1 we have rk (F − G) ≤ 1, so there exists w, z ∈ Cn with F − G = wz∗. Using the representations above we see 2F − G = 2(F − G) + G = 2wz∗ + uv∗. From (10), for λ = 2 we have that rk (2F − G) ≤ 1. If u and w are linearly independent then z = αv or v = αz for some α ∈ C. Let z = αv (the case v = αz can be proven similarly). Then sF − G = s(uv∗ + wz∗) − uv∗ = (s(u + αw) − u)v∗, therefore P(s) admits a representation as in (9). Now, assume that u and w are linearly dependent. Let u = βw for some β ∈ C (the case w = βu can be proven similarly), then sF − G = s(uv∗ + wz∗) − uv∗ = w(s(βv∗ + z∗) − βv∗) holds, hence (9) is proven. The converse statement is obvious. For F, G ∈ Rn×n the arguments above remain valid after replacing C by R and u∗, v∗, z∗ and w∗ by uT , vT , zT and wT . 5 The following example illustrates that both representations in (9) are necessary. Example 3.2. A short computation shows that the matrix pencils P1(s) :=(cid:18)s + 1 s + 1 P2(s) :=(cid:18)s + 1 1 s + 1 1(cid:19) =(cid:18)1 1 1 (cid:19) =(cid:18)s(cid:18)1 0(cid:19) +(cid:18)1 1(cid:19) (s(1, 0) + (1, 1)) 1(cid:19)(cid:19) (1, 1), admit only one of the representations given in Proposition 3.1. If in (9) the elements u, v ∈ Cn are linearly dependent both representations in (9) coincide and without restriction we can write for non-zero (α, β) ∈ C2 P(s) = (αs − β)uw∗. (11) The next lemma provides a simple criterion for (A + P)(s) to be regular when P(s) is of the form (11). Lemma 3.3. Let A(s) = sE − A be regular. Choose (α, β) ∈ C2 non-zero and let P(s) be given by (11). Then the following holds. (a) Assume α 6= 0. If β/α ∈ σ(A) then β/α ∈ σ(A + P). If β/α /∈ σ(A) then (A + P)(s) is regular. (b) Assume α = 0. If ∞ ∈ σ(A) then ∞ ∈ σ(A + P). If ∞ /∈ σ(A) then (A + P)(s) is regular. Proof. For α 6= 0 we have det(A + P)( β α ) = det(cid:18) β α E − A + (α β α − β)uw∗(cid:19) = det(cid:18) β α E − A(cid:19) = det A( β α ) and (a) follows. For α = 0 we use that ∞ ∈ σ(A) if and only if the leading coefficient E is singular. Since α = 0, the leading coefficient of (A + P)(s) is E for all u, w ∈ Cn, hence (b) is proved. 4 Change of the root subspaces under rank one pertur- bations In this section, we obtain bounds on the number of eigenvalues which can be changed by a rank one perturbation. The following lemma is a special case (r = 1) of [14, Lemma 2.1]. Lemma 4.1. Let A(s) be a regular matrix pencil satisfying Assumption 2.2 and let P(s) be of rank one. Assume that (A + P)(s) is regular and let λ ∈ C be an eigenvalue of A(s). Then (A + P)(s) has at least dim ker A(λ) − 1 linearly independent Jordan chains at λ. For dim ker A(λ) ≥ 2 these chains can be sorted in such a way that for the length of the ith chain mi(λ) the following holds m2(λ) ≥ . . . ≥ mdim ker A(λ)(λ) and mi(λ) ≥ mi(λ), 2 ≤ i ≤ dim ker A(λ). The following result describes the maximal change of the root subspace dimension under rank one perturbations. For matrices, that is, if E = In, this result was obtained in [33], see also [3, 37]. 6 Proposition 4.2. Let A(s) be a regular matrix pencil satisfying Assumption 2.2, then for any rank one pencil P(s) such that (A+P)(s) is regular we have for all λ ∈ C and k ∈ N\{0} dim (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Lk+1 λ Lk (A + P) λ(A + P) − dim Lk+1 λ Lk dim Lk λ(A + P) − dim Lk λ(A) ≤ k. ≤ 1, (A) λ(A) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (12) (13) Proof. We prove the inequality (12). Assume λ 6= ∞ and that for k, l ∈ N \ {0} we have dim Lk+1 λ Lk (A) λ(A) = l ≥ 2. (14) This is equivalent to the fact A(s) has l linearly independent Jordan chains at λ with length at least k + 1 which means m1(λ) ≥ m2(λ) . . . ≥ ml(λ) ≥ k + 1. It follows from Lemma 4.1 that (A + P)(s) has at least l − 1 linearly independent Jordan chains with lengths m2(λ) ≥ . . . ≥ ml(λ) ≥ ml(λ) ≥ k + 1 which leads to dim Lk+1 λ Lk (A + P) λ(A + P) ≥ l − 1. (15) It remains to show that the expression in (15) is less or equal to l + 1. Indeed, assume dim Lk+1 λ Lk (A + P) λ(A + P) ≥ l + 2. If we consider the regular pencil (A + P)(s) and the rank one pencil −P(s) and apply the above arguments, we have that dim Lk+1 λ Lk (A) λ(A) ≥ l + 1, which is a contradiction to (14). Hence, (12) is shown for l ≥ 2. If l = 1, i.e. dim Lk+1 λ Lk (A) λ(A) = 1 (A+P) and assume that dim Lk+1 λ(A+P) ≥ 3. Lemma 4.1 applied to the regular matrix pencil λ Lk (A + P)(s) shows dim Lk+1 λ(A) ≥ 2, a contradiction and (12) follows. λ Lk λ(A) implies L1 λ(A) = ker A(λ). Since A(λ) and (A + P)(λ) are matrices and P(λ) is a matrix of rank at most one (see Proposition 3.1), the estimates Now we show (13). For k = 1 the definition of Li (A) rank (A(λ)) = rank ((A + P)(λ) − P(λ)) ≤ rank ((A + P)(λ)) + rank (P(λ)), rank ((A + P)(λ)) ≤ rank (A(λ)) + 1 7 imply rank ((A + P)(λ)) − rank (A(λ)) ≤ 1 and together with the dimension formula dim ker A(λ) + rk (A(λ)) = n this leads to dim ker A(λ) − dim ker(A + P)(λ) =(cid:12)(cid:12)n − rank (A(λ)) −(cid:0)n − rk ((A + P)(λ))(cid:1)(cid:12)(cid:12) ≤ 1. Therefore (13) holds for k = 1. For k ≥ 2 we have the identity (16) dim Lk λ(A) = dim ker A(λ) + dim Lm+1 λ Lm (A) λ (A) k−1Xm=1 which leads to dim Lk λ(A) − dim Lk λ(A + P) ≤ dim ker A(λ) − dim ker(A + P)(λ) + and (16) together with (12) imply (13). Lm+1 λ Lm (A) λ (A) − dim k−1Xm=1(cid:12)(cid:12)(cid:12)(cid:12)dim Lm+1 λ Lm (A + P) λ (A + P) (cid:12)(cid:12)(cid:12)(cid:12) For λ = ∞ we consider the dual pencil A′(s) = −As + E at λ = 0. Obviously A′(s) and the dual pencil (A + P)′(s) of (A + P)(s) are regular. By Proposition 2.1 (c), it remains to apply (12) and (13) for λ = 0 to A′(s) and (A + P)′(s) to see that (12) and (13) hold for A(s) and (A + P)(s) at λ = ∞. For k = 1 the inequality (13) leads to the following statement. Corollary 4.3. Let A(s) be a regular matrix pencil, then for any rank one pencil P(s) such that (A + P)(s) is regular we have and for every µ ∈ σ(A + P) \ σ(A) nλ ∈ σ(A) (cid:12)(cid:12)(cid:12) dim ker A(λ) ≥ 2o ⊆ σ(A + P) dim ker(A + P)(µ) = 1, i.e., in this case, there is only one Jordan chain of length dim Lµ(A + P). Proposition 4.2 states, roughly speaking, that the largest possible change in the dimen- sions of Lλ(A + P) compared with Lλ(A) is bounded by the length of the largest Jordan chain of A(s) and (A + P)(s). However, the aim of the following Theorem 4.4 is to give bounds for the change of dimension of Lλ(A + P) only in terms of the unperturbed pencil A(s). For this, we use the number m1(λ) which is according to Assumption 2.2 the length of the largest Jordan chain of A(s) at λ and the number M (A) := Xµ∈σ(A) m1(µ). Theorem 4.4. Let A(s) be a regular matrix pencil satisfying Assumption 2.2. Then for any rank one pencil P(s) such that (A + P)(s) is regular we have for λ ∈ σ(A) dim Lλ(A) − m1(λ) ≤ dim Lλ(A + P) ≤ dim Lλ(A) + M (A) − m1(λ), (17) whereas the change in the dimension for λ ∈ C \ σ(A) is bounded by 0 ≤ dim Lλ(A + P) ≤ M (A). (18) 8 Summing up, we obtain the following bounds Xλ∈σ(A) Xλ∈σ(A+P)\σ(A) dim Lλ(A + P) ≥ n − M (A), dim Lλ(A + P) ≤ M (A). (19) Proof. By Assumption 2.2, we have Lλ(A) = Lm1(λ) λ (A). Then (13) implies for λ ∈ σ(A) dim Lλ(A) − m1(λ) = dim Lm1(λ) ≤ dim Lm1(λ) λ λ (A) − m1(λ) (A + P) ≤ dim Lλ(A + P). (20) This is the lower bound in (17). Since (A + P)(s) is regular we can apply (6), (20) and the upper bound for λ ∈ σ(A) follows from dim Lλ(A + P) = n − Xµ∈σ(A+P)\{λ} dim Lµ(A + P) dim Lµ(A) − Xµ∈σ(A)\{λ} ≤ Xµ∈σ(A) = dim Lλ(A) + Xµ∈σ(A)\{λ} ≤ dim Lλ(A) + Xµ∈σ(A)\{λ} dim Lµ(A + P) dim Lµ(A) − Xµ∈σ(A)\{λ} m1(µ) = dim Lλ(A) + M (A) − m1(λ). dim Lµ(A + P) Hence (17) is proved and applying the same estimates for λ ∈ C \ σ(A) proves (18). We continue with the proof of (19). Relation (20) implies Xλ∈σ(A) dim Lλ(A + P) ≥ Xλ∈σ(A) (dim Lλ(A) − m1(λ)) = n − Xλ∈σ(A) m1(λ) = n − M (A) and this yields Xλ∈σ(A+P)\σ(A) dim Lλ(A + P) = n − Xλ∈σ(A) dim Lλ(A + P) ≤ M (A). From the inequality (19) we see that the number of changeable eigenvalues under a rank one perturbation is bounded by M (A). 5 Eigenvalue placement with rank one perturbations In this section we study which sets of eigenvalues can be obtained by rank one perturbations. The following theorem is the main result. It states that for a given set of complex numbers there exists a rank one perturbation P(s) such that the set is included in σ(A+ P), provided the given set has not more than M (A) elements. Theorem 5.1. Let A(s) be a regular matrix pencil satisfying Assumption 2.2 and choose pairwise distinct numbers µ1, . . . , µl ∈ C with l ≤ M (A). Choose multiplicities m1, . . . , ml ∈ i=1 mi = M (A). Then the following statements hold true. N \ {0} withPl 9 (a) There exists a rank one pencil P(s) = (αs − β)uv∗ with α ∈ C \ {0}, β ∈ C and u, v ∈ Cn such that (A + P)(s) is regular, and σ(A + P) = {µ1, . . . , µl} ∪ {λ ∈ σ(A) dim ker A(λ) ≥ 2} (21) with multiplicities dim Lλ(A + P) = dim Lλ(A) − m1(λ) + mi, dim Lλ(A) − m1(λ), mi, 0, for λ = µi ∈ σ(A), for λ ∈ σ(A) \ {µ1, . . . , µl}, for λ = µi /∈ σ(A), for λ /∈ σ(A) ∪ {µ1, . . . , µl}. (22) (b) If E, A are real matrices and {µ1, . . . , µl} is symmetric with respect to the real line with mi = mj if µj = µi and all i, j = 1, . . . , l, there exists α, β ∈ R and u, v ∈ Rn such that P(s) = (αs − β)uvT satisfies (21) and (22). We formulate a special case of Theorem 5.1. Corollary 5.2. In addition to the assumptions of Theorem 5.1, assume that dim ker A(λ) = 1 for all λ ∈ σ(A). Hence m1(λ) = dim Lλ(A) holds for all λ ∈ σ(A) and M (A) = n. Then there exists a rank one pencil P(s) = (αs − β)uv∗ such that (A + P)(s) is regular and the equations (21) and (22) take the following form σ(A + P) = {µ1, . . . , µl} and dim Lλ(A + P) =(mi, 0, for λ = µi, for λ /∈ {µ1, . . . , µl}. Therefore, for each µi ∈ σ(A + P) there is only one Jordan chain of (A + P)(s) of length mi. Combining Theorem 4.4, Theorem 5.1 and Corollary 5.2 we get the following result which solves an inverse problem which was investigated for matrices in [22]. Theorem 5.3. Given pairwise distinct numbers λ1, . . . , λk ∈ C and µ1, . . . , µl ∈ C with k ≤ n, l ≤ n and multiplicities m(λ1), . . . , m(λk), m(µ1), . . . , m(µl) ∈ N \ {0} such that m(λi) = kXi=1 lXi=1 m(µi) = n. Then, there exists a regular matrix pencil A(s) ∈ C[s]n×n and a rank one pencil P(s) ∈ C[s]n×n such that σ(A) = {λ1, . . . , λk}, σ(A + P) = {µ1, . . . , µl}, dim Lλi (A) = m(λi), i = 1, . . . , k, dim Lµi (A + P) = m(µi), i = 1, . . . , l. Lemma 5.4. Let A(s) be a regular matrix pencil satisfying Assumption 2.2. Then for the polynomial mA(s) defined in (8) and for every p(s) ∈ C[s] with there exist u, v ∈ Cn such that deg p ≤ M (A) − 1 If E and A are real matrices and p(s) ∈ R[s], then there exists u, v ∈ Rn satisfying (23). p(s) = v∗mA(s)(sE − A)−1u. (23) 10 (u, v) 7→ v∗mA(s)(sE − A)−1u Proof. We introduce ΘA : Cn × Cn →(cid:8)p(s) ∈ C[s] deg p ≤ M (A) − 1(cid:9), and show the surjectivity of this map. Since the surjectivity of ΘA is invariant under equivalence transformations of the form of sE−A to S(sE−A)T with invertible S, T ∈ Cn×n, we can assume that A(s) is given in Weierstrass canonical form (4) with matrices J and N . If σ(J) = {λ1, . . . , λm} for some complex numbers λ1, . . . , λm, then J and N are given by J = mMi=1 dim ker A(λi)Mj=1 Jmj (λi)(λi), N = dim ker A(∞)Mj=1 Jmj (∞)(0) (24) with Jordan blocks Jk(λ) of size k at λ ∈ C given by λ 1 · · · · · Jk(λ) =  1 λ ∈ Ck×k. This allows us to simplify the resolvent representation with u = (u∗ u0, v0 ∈ Cr and u1, v1 ∈ Cn−r to 0, u∗ 1)∗, v = (v∗ 0 , v∗ 1)∗, v∗mA(s)(sE − A)−1u = v∗ 0mA(s)(sIr − J)−1u0 + v∗ = v∗ 0mA(s) + v∗ 1mA(s) j = 1, . . . , dim ker A(λi ) Mi = 1, . . . , m,   Mj=1,...,dim ker A(∞) (s − λi)−1 . . . . . . 1 mA(s)(sN − In−r)−1u1 (−1)−mj (λi)+1(s − λi)−mj (λi) ... (s − λi)−1  u0 −1 −s . . . −smj (∞)−1 −1 . . . −smj (∞)−2 . . . ... −1  u1. (25) Observe M (A) = deg mA + m1(∞). From (7) and (8) we see that ΘA maps into the set {p(s) ∈ C[s] deg p ≤ M (A) − 1}. Obviously, the right hand side of (25) consists of the sum of products involving two block matrices. Consider the first summand, then the entries of the1 first row of the blocks of the block matrix for j = 1 and i = 1, . . . , m are linearly independent as they are functions with a pole in λi of order from one up to m1(λi). Choosing suitable contours and applying the residue theorem, one sees that {(s − λi)−r i = 1, . . . , m, r = 1, . . . , m1(λi)}. is also linearly independent. After multiplication with mA(s) this set of functions remains linearly independent. Therefore the set P1 := {mA(s)(s − λi)−r i = 1, . . . , m, r = 1, . . . , m1(λi)} is linearly independent and it containsPm i=1 m1(λi) = deg mA elements, each of degree less or equal to deg mA − 1. Moreover, for j = 1, the entries of the first row of the blocks in the 11 block matrix of the second summand on the right hand side of (25) are linearly independent and form the linearly independent set of polynomials P2 := {mA(s)sr r = 0, . . . , m1(∞) − 1} which contains m1(∞) elements of degree between deg mA and deg mA + m1(∞) − 1 = M (A) − 1. Hence P1 ∪ P2 consists of deg mA + m1(∞) = M (A) (26) 0 , v∗ linearly independent elements. Furthermore, one can choose certain entries of v = (v∗ 1) as one and all others as zero such that the multiplication with v∗ 1 in (25) picks exactly the first row of each block with j = 1 for all i = 1, . . . , m. By choosing one entry of u as one and all others as zero we see 0 and v∗ P1 ∪ P2 ⊂ ran ΘA and from the linearity of the map u 7→ ΘA(u, v) with (26) the lemma is proved for matrices E, A ∈ Cn×n. We consider the case where E, A are real matrices. Here we use the Weierstrass canonical form over R, obtained in [16], with transformation matrices S, T ∈ Rn×n. The matrix N is the same as in (24) and J is in real Jordan canonical form, see [21, Section 3.4.1], J = Mλ ∈ σ(J), Im λ > 0 dim ker A(λ)Mj=1 J R mj (λ)(λ) ⊕ Mλ∈σ(J)∩R dim ker A(λ)Mj=1 Jmj (λ)(λ), where Jmj (λ)(λ), λ ∈ σ(A) ∩ R, are Jordan blocks of size mj(λ) and J R l ∈ N \ {0} is a real Jordan block at λ = a + ib with a ∈ R, b > 0, given by l (λ) ∈ R2l×2l for some ∈ R2l×2l, C(a, b) :=(cid:18) a −b a(cid:19) ∈ R2×2. b J R l (λ) :=  C(a, b) I2 · · · · · I2 C(a, b)  Therefore the resolvent of J R l (a, b) is given by (sI2l − J R l (a, b))−1 = (s − C(a, b))−1 . . . . . . (−1)l−1(s − C(a, b))−l ... (s − C(a, b))−1  (27) where the entries are given by (s − C(a, b))−k = ((s − C(a, b))−1)k =(cid:18) 1 (s − a)2 + b2(cid:18)s − a −b b s − a(cid:19)(cid:19)k , k ∈ N \ {0}. Using the expression (27) instead of the blocks occuring in the block matrix in the first summand of the right hand side of (25) for the non-real eigenvalues, one can define again a linearly independent set of polynomials P1 by picking all first row entries. This set consists again of polynomials all of distinct degree, because the factor ((s − a)2 + b2)m1(λ) occurs in the minimal polynomial mA(s). The set P2 remains the same as in the complex valued case. Therefore the same arguments imply the surjectivity of ΘA in this case. 12 Proof of Theorem 5.1. Choose α, β ∈ C such that α 6= 0 and β α /∈ {µ1, . . . , µl} ∪ σ(A) holds. We set γ := mA(β/α) · (β/α − µi)−mi . The condition β/α /∈ {µ1, . . . , µl} ∪ σ(A) implies mA(β/α) 6= 0, hence γ 6= 0. We consider Yµi∈{µ1,...,µl}\{∞} Yµi∈{µ1,...,µl}\{∞} qγ(s) := γ (s − µi)mi . AsPl i=1 mi = M (A), the polynomial qγ(s) satisfies deg qγ ≤ M (A). The degree of mA(s) is M (A) − m1(∞) which is smaller or equal to M (A). From the choice of γ we see that (qγ − mA)(β/α) = 0 holds and therefore qγ(s) − mA(s) αs − β is a polynomial of degree less or equal to M (A) − 1. By Lemma 5.4 there exist u, v ∈ Cn with qγ(s) − mA(s) αs − β = v∗mA(s)(sE − A)−1u. This equation combined with Sylvester's determinant identity leads to qγ(s) mA(s) = 1 + (αs − β)v∗(sE − A)−1u = det(In + (sE − A)−1(αs − β)uv∗). (28) Now, set P(s) = (αs − β)uv∗. Then by Lemma 3.3 (A + P)(s) is regular and from (28) we obtain det(A + P)(s) = det(sE − A + (αs − β)uv∗) = det(sE − A) det(In + (sE − A)−1(αs − β)uv∗) = det(sE − A) qγ(s) mA(s) . Since det(sE − A) is by Proposition 2.1 (d) divisible by mA(s), (21) follows. The equation (22) follows from Proposition 2.1 (d) and Theorem 5.1 (a) is proved. The statements in (b) follow by the same construction as above and by Lemma 5.4 for real matrices. 6 Eigenvalue placement under parameter restrictions In this section we study the eigenvalue placement under perturbations of the form P(s) = (su + v)w∗, w ∈ Cn (29) for fixed u and v in Cn (cf. Proposition 3.1). Special cases of this parameter restricted placement are the pole assignment problem for linear systems under state feedback (cf. Section 7) and also the eigenvalue placement problem for matrices under rank one matrices. Obviously, for the perturbation (29) the bounds on the multiplicities from Theorem 4.4 still hold. But since u and v are now fixed, we obtain tighter bounds which are given below. In 13 the formulation of these bounds we introduce the number mu,v(λ) which basically replaces m1(λ) in the bounds obtained in Theorem 4.4. Assume that the vector-valued function s 7→ (sE − A)−1(su + v) (30) has a pole at λ ∈ σ(A) \ {∞}, i.e. one of the entries has a pole at λ. Then, we denote by mu,v(λ) the order of λ as a pole of (30), which is the maximal order of λ as a pole of one of the entries. For λ = ∞ we define mu,v(∞) as the order of the pole of s 7→ (−sA + E)−1(sv + u) at s = 0. In the case where no pole occurs we set mu,v(λ) = 0. Hence mu,v(λ) is defined for all λ ∈ σ(A). Note that another way of introducing poles and their order is given by the use of the Smith-McMillan form [9], [16, Ch. VI], [17, Ch. S1]. Proposition 6.1. Let A(s) = sE −A be regular and let P(s) be given by (29) with u, v ∈ Cn fixed such that (A + P)(s) is regular. For M (A, u, v) := Xµ∈σ(A) mu,v(µ) and λ ∈ σ(A) we have dim Lλ(A) − mu,v(λ) ≤ dim Lλ(A + P) ≤ dim Lλ(A) + M (A, u, v) − mu,v(λ). (31) For λ ∈ C \ σ(A) we have From this we obtain 0 ≤ dim Lλ(A + P) ≤ M (A, u, v). (32) Xλ∈σ(A) Xλ∈σ(A+P)\σ(A) dim Lλ(A + P) ≥ n − M(A, u, v), dim Lλ(A + P) ≤ M(A, u, v). Proof. From Sylvester's formula we conclude det(A + P)(s) = det A(s)(1 + w∗(sE − A)−1(su + v)). If λ ∈ σ(A)\{∞} then the characteristic polynomial of A(s) has at λ a zero with multiplicity dim Lλ(A). Since the order of λ as a pole of s 7→ 1+w∗(sE −A)−1(su+v) is at most mu,v(λ), the multiplicity of λ as a zero of det(A + P)(s), hence the dimension of Lλ(A + P), can be bounded by dim Lλ(A + P) ≥ dim Lλ(A) − mu,v(λ). This shows the lower bound in (31). The upper bound in (31) for λ ∈ σ(A) \ {∞} and (32) for λ ∈ C \ σ(A) are obtained in the same way as in the proof of Theorem 4.4. The estimates (31) and (32) hold also for the dual pencil A′(s) = −sA + E at λ = 0 and, hence, by Proposition 2.1 (c) the estimates (31) and (32) also hold for A(s) at λ = ∞. Now the remaining assertions follow in the same way as in Theorem 4.4. Within these bounds we investigate the placement of the eigenvalues. We start with a result which can be seen as a generalization of Lemma 3.3. 14 Lemma 6.2. Let A(s) = sE − A be a regular matrix pencil satisfying Assumption 2.2 and let u, v, w ∈ Cn and P(s) = (su + v)w∗, such that (A + P)(s) is regular. Then we have {λ ∈ σ(A) dim ker A(λ) ≥ 2 or mu,v(λ) < m1(λ)} ⊆ σ(A + P). Proof. From Corollary 4.3 we deduce {λ ∈ σ(A) dim ker A(λ) ≥ 2} ⊆ σ(A + P). First, we consider the case λ ∈ σ(A) \ {∞}. The resolvent representation (25) implies that s 7→ (sE − A)−1 has an entry with a pole at λ of order m1(λ). Hence mu,v(λ) ≤ m1(λ). The condition mu,v(λ) < m1(λ), Sylvester's determinant formula, and the representation from Proposition 2.1 (d) lead to det(A + P)(s) = det(sE − A)(1 + w∗(sE − A)−1(su + v)) = (−1)n−r det(ST )−1mA(s)q(s)(1 + w∗(sE − A)−1(su + v)). (33) Hence det(A+P)(s) contains the factor (s−λ), therefore λ ∈ σ(A+P) and the set inclusion is proven for λ ∈ σ(A)\{∞}. In the case λ = ∞ ∈ σ(A), one has to apply the above arguments to the dual pencil A′(s) = −sA + E and (A + P)′(s) = −sA + E + (sv + u)w∗ at λ = 0. Theorem 6.3. Let A(s) = sE − A be a regular matrix pencil satisfying Assumption 2.2 and let u, v ∈ Cn be fixed. Choose µ1, . . . , µl ∈ C with l ≤ M (A, u, v) and multiplicities i=1 mi = M (A, u, v). If u, v are linearly dependent we make the mi ∈ N \ {0} satisfying Pl following further assumptions. (i) If u 6= 0 and v = −µu. Then µ /∈ σ(A) implies µ /∈ {µ1, . . . , µl}. (ii) If u = 0 and v 6= 0. Then ∞ /∈ σ(A) implies ∞ /∈ {µ1, . . . , µl}. Then there exists w ∈ Cn such that for P(s) = (su + v)w∗ the pencil (A + P)(s) is regular and satisfies σ(A + P) = {µ1, . . . , µl} ∪ {λ ∈ σ(A) dim ker A(λ) ≥ 2 or mu,v(λ) < m1(λ)} (34) and the dimensions of the root subspaces are dim Lλ(A + P) = dim Lλ(A) − mu,v(λ) + mi, dim Lλ(A) − mu,v(λ), mi, 0, for λ = µi ∈ σ(A), for λ ∈ σ(A) \ {µ1, . . . , µl}, for λ = µi /∈ σ(A), for λ /∈ σ(A) ∪ {µ1, . . . , µl}. (35) If E, A are real matrices and u, v ∈ Rn then we can find w ∈ Rn such that (34) and (35) hold. Proof. For the proof, we essentially repeat the arguments from the proof of Lemma 5.4. First, let us assume that u and v are linearly dependent with u 6= 0 and, hence, v = −µu and P(s) = (s − µ)uw∗. For S, T ∈ Cn×n such that SA(s)T is in Weierstrass canonical form we write in (33) (1 + (s − µ)w∗(sE − A)−1u) = (1 + (s − µ)w∗T (sSET − SAT )−1Su). If A(s) = sE − A has only one eigenvalue λ 6= ∞ with one Jordan chain at λ of length m1(λ) and mu,v(λ) ≥ 1, one can easily write down the result of the multiplication of (sSET − SAT )−1 with Su = ((Su)j)m1(λ) , cf. (25), j=1 (sSET − SAT )−1Su =(cid:16)Pm1(λ) j=k (−1)j−k(s − λ)−j+k−1(Su)j(cid:17)m1(λ) k=1 . (36) 15 Denote by m ∈ N with 1 ≤ m ≤ m1(λ) the largest index such that (Su)m 6= 0. Such an index exists because of mu,v(λ) ≥ 1 and then (Su)k = 0 for all m < k ≤ m1(λ). Let us now assume that µ /∈ σ(A). The assumption µ /∈ σ(A) implies that mu,v(λ) is not only the order of λ as a pole of s 7→ (s−µ)(sE −A)−1u but also the order of λ as a pole of s 7→ (sE −A)−1u. Therefore m = mu,v(λ) hence (Su)mu,v (λ) 6= 0 and we consider the following functions given by the right hand side of (36) pλ,k(s) := mu,v (λ)Xj=k (−1)j−k(s − λ)−j+k−1(Su)j (37) for 1 ≤ k ≤ mu,v(λ). The summand in each pλ,k(s) with the pole of the highest order has, as (Su)mu,v (λ) 6= 0, a non-zero coefficient so that (37) defines a linearly independent set of functions. If A(s) = sE − A has only the eigenvalue at ∞ with only one Jordan chain of length m1(∞) and mu,v(∞) ≥ 1 then the largest index m ∈ N such that (Su)m 6= 0 holds is m = mu,v(∞) and the right hand side of (36) consists of the functions p∞,k(s) := − mu,v(∞)−1Xj=k sj−k(Su)j+1 for 0 ≤ k ≤ mu,v(∞) − 1. Since (sSET − SAT )−1 has block diagonal structure (25), similar expressions as in (36) occur when A(s) has more than one Jordan chain at λ ∈ σ(A) or more than one eigenvalue. Here there exists for all λ ∈ σ(A) with mu,v(λ) ≥ 1 some block on the right hand side of (25) such that the above construction of the functions pλ,k(s) can be carried out. As in the proof of Lemma 5.4 we multiply with a polynomial em(s) := Yλ∈σ(A)\{∞} (s − λ)mu,v (λ) and conclude that the set {pλ,k(s)em(s) λ ∈ σ(A) \ {∞}, 1 ≤ k ≤ mu,v(λ)} ∪ {p∞,k(s)em(s) ∞ ∈ σ(A), 0 ≤ k ≤ mu,v(∞) − 1} is linearly independent and consists ofPλ∈σ(A) mu,v(λ) = M (A, u, v) polynomials of degree at most M (A, u, v) − 1 since we consider only those λ ∈ σ(A) with mu,v(λ) ≥ 1. Therefore it is a basis in the space of polynomials of degree less or equal to M (A, u, v) − 1. By (i) we have µ /∈ {µ1, . . . , µl} and we consider (38) Yµi ∈{µ1,...,µl}\{∞} Yµi∈{µ1,...,µl}\{∞} γ := em(µ) eqγ(s) := γ (µ − µi)−mi (s − µi)mi . and the polynomial Since T is invertible and from the linear independence of the polynomials there exists w ∈ Cn such that holds. Plugging this into (33) proves (34) and (35) in the case µ /∈ σ(A). eqγ(s) − em(s) s − µ = w∗Tem(s)(sSET − SAT )−1Su (39) 16 Let us now assume that µ ∈ σ(A) and we consider again the case that A(s) = sE − A has only one Jordan chain at λ of length m1(λ) and mu,v(λ) ≥ 1. This means that µ = λ and the definition of mu,v(λ) implies m = mu,v(λ) + 1 and we consider the following functions given by the right hand side of (36) pλ,k(s) := mu,v (λ)+1Xj=k (−1)j−k(s − λ)−j+k−1(Su)j (40) for 1 ≤ k ≤ mu,v(λ)+1. These functions define a linearly independent set. As in the previous k=1 := w∗T sub case, we are looking for a solution w ∈ Cn with n = mu,v(λ) + 1 and (wk)n of the equation Yµi∈{µ1,...,µl}\{∞} (s − µi)mi − em(s) = w∗Tem(s)(s − µ)(sSET − SAT )−1Su (41) = mu,v(λ)+1Xk=1 wkem(s)(s − µ)pλ,k(s). If we set w1 = . . . = wmu,v (λ) = 0 and wmu,v (λ)+1 = −((Su)mu,v (λ)+1)−1 the right hand side of (41) is equal to −em(s). On the other hand, by the linear independence of {em(s)pλ,k(s)}mu,v (λ)+1 k=1 in the space of polynomials of degree at most mu,v(λ) = M (A, u, v), there is a solution w ∈ Cn of the equation Yµi∈{µ1,...,µl}\{∞} (s − µi)mi = mu,v (λ)+1Xk=1 wkem(s)pλ,k(s) This, together with the linearity of equation (41) in w implies the existence of w ∈ Cn such that (41) holds which proves the assertions in this sub case, that there is only one eigenvalue λ 6= ∞. The block diagonal structure of (sSET − SAT )−1 as in (25) implies the existence of w ∈ Cn such that (34) and (35) hold in the case when A(s) has more than one Jordan chain at λ ∈ σ(A) or more than one eigenvalue. One only has to replace the functions pλ,k(s) for λ = µ with those defined in (40). Next, when u and v are linearly dependent with u = 0 then we can further assume that v 6= 0 because v = 0 implies that mu,v(λ) = 0 for all λ ∈ σ(A) and then (34) and (35) hold trivially. This case can be treated by considering the dual pencils A′(s) and (A + P)′(s), because for the dual pencils we are in the case of (i) with µ = 0. Therefore, we have already proven that (34) and (35) hold for the dual pencils under the assumption that 0 /∈ σ(A′) implies 0 /∈ {µ1, . . . , µl}. By Proposition 2.1 (c) this condition follows from the assumption (ii) that ∞ /∈ σ(A) implies ∞ /∈ {µ1, . . . , µl}. Furthermore we see immediately that SA′(s)T is block diagonal and that dim Lλ(A) for λ ∈ σ(A) \ {0} is equal to dim Lλ−1 (A′). This proves (34) and (35) for A(s) and (A + P)(s). This finishes the proof of the theorem given that u and v are linearly dependent. In the case where u and v are linearly independent we have in (33) (1 + w∗(sE − A)−1(su + v)) = (1 + w∗T (sSET − SAT )−1(sSu + Sv)). If A(s) = sE − A has only one eigenvalue λ 6= ∞ with one Jordan chain of length m1(λ) and mu,v(λ) ≥ 1 then the product (sSET − SAT )−1(sSu + Sv) is given by (cid:16)Pm1(λ) j=k (−1)j−k(s − λ)−j+k−1(s(Su)j + (Sv)j )(cid:17)m1(λ) k=1 . (42) 17 This again allows us to define a linearly independent set of polynomials. For this we consider the largest index m ∈ N with 1 ≤ m ≤ m1(λ) such that s(Su)m + (Sv)m 6= 0. This implies that s(Su)k + (Sv)k = 0 for all m < k ≤ m1(λ) and we obtain the following to cases. If (s(Su)m + (Sv)m) and (s − λ) are linearly independent in C[s] then we have m = mu,v(λ) and we consider the following entries of (42) pλ,k(s) := mu,v(λ)Xj=k (−1)j−k(s − λ)−j+k−1(s(Su)j + (Sv)j ) for 1 ≤ k ≤ mu,v(λ). If (s(Su)m + (Sv)m) and (s − λ) are linearly dependent then we have m = mu,v(λ) + 1 and define pλ,k(s) := mu,v(λ)+1Xj=k (−1)j−k(s − λ)−j+k−2(s(Su)j + (Sv)j ) for 1 ≤ k ≤ mu,v(λ)+1. In both cases we have from the condition (s(Su)m+(Sv)m) 6= 0 that the functions pλ,k(s) 6= 0 define a linearly independent set. If A(s) has only the eigenvalue ∞ with one Jordan chain of length m1(∞) we consider in the case (Su)mu,v(∞) 6= 0 p∞,k(s) := − mu,v(∞)−1Xj=k sj−k−1(s(Su)j+1 + (Sv)j+1) for 0 ≤ k ≤ mu,v(∞) − 1 and if (Su)mu,v (∞) = 0 p∞,k(s) := − mu,v(∞)−1Xj=k sj−k(s(Su)j+1 + (Sv)j+1) for 0 ≤ k ≤ mu,v(∞) − 1. If A(s) has more than one eigenvalue and with more than one Jordan chain, then we can use again the block diagonal structure of (sSET − SAT )−1 to define the functions pλ,k(s) for all λ ∈ σ(A). The linear independence of these functions follows from the linear independence of u and v. Now the same linear independency arguments from above can be applied. This implies the existence of w ∈ Cn satisfying the equation Yµi∈{µ1,...,µl}\{∞} (s − µi)mi − em(s) = w∗Tem(s)(sSET − SAT )−1S(su + v). For real matrices E and A one uses the transformation to the real Weierstrass canonical form with S, T ∈ Rn×n such that for λ = a + ib ∈ σ(A) \ R with b > 0 we use the blocks given by (27) to define the polynomials pλ,k(s). These polynomials have to be inserted into (38) which gives a linearly independent set. Since the transformation matrices S, T are real, this proves the statement in the real case. If E = In then σ(A) equals the set σ(A) of all eigenvalues of A and ∞ /∈ σ(A) holds. For the perturbation class given for a fixed v ∈ Cn by P(s) := vw∗, w ∈ Cn, Theorem 6.3 for u = 0 leads to the following eigenvalue placement result for matrices which is a generalization of the placement results obtained in [10, 18, 22, 36]. 18 Corollary 6.4. Let A(s) = sIn − A be a matrix pencil satisfying Assumption 2.2 with minimal polynomial mA(s) of the matrix A ∈ Cn×n and fixed v ∈ Cn. Then for any given values µ1, . . . , µl ∈ C with l ≤ M (A, 0, v) and multiplicities mi ∈ N \ {0} satisfying i=1 mi = M (A, 0, v) there exists w ∈ Cn such that Pl σ(A + vw∗) = {µ1, . . . , µl} ∪ {λ ∈ σ(A) dim ker(λIn − A) ≥ 2, m0,v(λ) < m1(λ)} and (35) holds. If v ∈ Cn is not fixed, then one can always choose v such that m0,v(λ) = m1(λ) holds for all λ ∈ σ(A) implying that M (A, 0, v) = deg mA. For this choice of v there exists w ∈ Cn such that σ(A + vw∗) = {µ1, . . . , µl} ∪ {λ ∈ σ(A) dim ker(λIn − A) ≥ 2}. If E, A are real matrices then u, v can be chosen in Rn. 7 Application to single input differential-algebraic equa- tions with feedback Parameter restricted perturbations of the form (29) occur naturally in the study of differ- ential algebraic equations with a single input given by E, A ∈ Cn×n, b ∈ Cn, x0 ∈ Cn and the equation d dt Ex(t) = Ax(t) + bu(t), t ∈ [0, ∞), x(0) = x0. (43) For this equation we consider a state feedback of the form u(t) = f ∗x(t) with f ∈ Cn. It is well known that the solution of the closed loop-system d dt Ex(t) = (A + bf ∗)x(t), t ∈ [0, ∞), x(0) = x0 can be expressed with the eigenvalues and Jordan chains of the matrix pencil sE − (A+ bf ∗), see [5]. This pencil can be written as a perturbation of sE − A with the rank one pencil P(s) = −bf ∗. By fixing u = 0 and v = −b in (29) we can write P(s) = −bf ∗ = (su + v)f ∗. For E singular we have ∞ ∈ σ(A) and for E invertible we have ∞ /∈ σ(A). In [6] it was shown that a system given by (E, A, b) with sE − A regular is controllable in the behavioural sense if and only if the Hautus condition rk [λE − A, b] = n, for all λ ∈ C holds. A transformation to Weierstrass canonical form (4) reveals that this Hautus condition is equivalent to dim ker A(λ) = 1 for all λ ∈ C ∩ σ(A) and m0,−b(λ) = m1(λ) = dim Lλ(A), for all λ ∈ C ∩ σ(A). Hence Theorem 6.3 implies that all finite eigenvalues can by placed arbitrarily in C. The feedback placement problem for regular matrix pencils was studied in [8, 24, 25, 28]. In the following, as in [25], we do not assume controllability. Then Theorem 6.3 implies the following. Theorem 7.1. Let (E, A, b) be a system given by (43) such that A(s) = sE − A is a regular matrix pencil satisfying Assumption 2.2. Choose pairwise distinct numbers µ1, . . . , µl ∈ i=1 mi = C with l ≤ M (A, 0, −b). Choose multiplicities m1, . . . , ml ∈ N \ {0} with Pl M (A, 0, −b). Then: 19 (a) For E singular there exists a feedback vector f ∈ Cn, such that sE − (A + bf ∗) is regular and σ(sE − (A + bf ∗)) equals to {µ1, . . . , µl} ∪ {λ ∈ σ(A) dim ker A(λ) ≥ 2 or m0,−b(λ) < m1(λ)}. (44) (b) For E invertible and ∞ /∈ {µ1, . . . , µl} there exists a feedback vector f ∈ Cn such that sE − (A + bf ∗) is regular and σ(sE − (A + bf ∗)) equals (44). In both cases (a) and (b) the dimensions of the root subspaces are given by formula (35). If E, A are real matrices and b ∈ Rn then f can be chosen in Rn. Acknowledgments The authors wish to thank the anonymous referees for their careful reading and for valuable comments which improved the quality of the manuscript. The authors also thank A.C.M. Ran for pointing out useful references. References [1] L. Batzke, Generic rank-one perturbations of structured regular matrix pencils, Linear Algebra Appl., 458 (2014), pp. 638–670. [2] L. Batzke, C. Mehl, A. Ran, and L. Rodman, Generic rank-k perturbations of structured matrices, Technical Report no. 1078, DFG Research Center Matheon, Berlin, 2015. [3] J. Behrndt, L. Leben, F. Martínez Pería and C. Trunk, The effect of finite rank perturbations on Jordan chains of linear operators, Linear Algebra Appl., 479 (2015), pp. 118–130. [4] T. Berger, G. Halikias and N. Karcanias, Effects of dynamic and non-dynamic element changes in RC and RL networks, Int. J. Circuit Theory Appl., 43 (2015), pp. 36–59. [5] T. Berger, A. Ilchmann and S. Trenn, The quasi-Weierstrass form for regular matrix pencils, Linear Algebra Appl., 436 (2012), pp. 4052–4069. [6] T. Berger and T. Reis, Controllability of linear differential-algebraic systems – A survey, in Surveys in differential-algebraic equations I, pp. 1–61, Springer, 2013. [7] T. Berger, C. Trunk and H. Winkler, Linear relations and the Kronecker canon- ical form, Linear Algebra Appl., 488 (2016), pp. 13–44. [8] D. Cobb, Feedback and pole-placement in descriptor variable systems, Int. J. Control, 33 (1981), pp. 1135–1146. [9] D. Cobb, A. Perdon, M. Sain and B. Wyman, Poles and zeros of matrices of rational functions, Linear Algebra Appl., 157 (1991), pp. 113–139. [10] J. Ding and G. Yao, Eigenvalues of updated matrices, Appl. Math. Comput., 185 (2007), pp. 415–420. [11] M. Dodig, Pole placement problem for singular systems, Numer. Linear Algebra Appl., 18 (2011), pp. 283–297. 20 [12] F. Dopico and J. Moro, Low rank perturbation of Jordan structure, SIAM J. Matrix Anal. Appl., 25 (2003), pp. 495–506. [13] F. Dopico and J. Moro, Generic change of the partial multiplicities of a regular matrix pencil under low rank perturbations, to appear in SIAM J. Matrix Anal. Appl. [14] F. Dopico, J. Moro and F. De Terán, Low rank perturbation of Weierstrass structure, SIAM J. Matrix Anal. Appl., 30 (2008), pp. 538–547. [15] S. Elhay, G. Golub and Y. Ram, On the spectrum of a modified linear pencil, Comput. Math. Appl., 46 (2003), pp. 1413–1426. [16] F. Gantmacher, Theory of Matrices, Chelsea, New York, 1959. [17] I. Gohberg, P. Lancaster and L. Rodman, Matrix Polynomials, SIAM, Philadel- phia, 2009. [18] G. Golub, Some modified matrix eigenvalue problems, SIAM Rev., 15 (1973), pp. 318–334. [19] E. Hennig, D. Krausse, E. Schäfer, R. Sommer, C. Trunk and H. Winkler, Frequency compensation for a class of DAE's arising in electrical circuits, Proc. Appl. Math. Mech., 11 (2011), pp. 837–838. [20] L. Hörmander and A. Melin, A remark on perturbations of compact operators, Math. Scand., 75 (1994), pp. 255–262. [21] R. Horn and C. Johnson, Matrix Analysis, Cambridge University Press, New York, 2013. [22] M. Krupnik, Changing the spectrum of an operator by perturbation, Linear Algebra Appl., 167 (1992), pp. 113–118. [23] V. Kucera and P. Zagalak, Fundamental theorem of state feedback for singular systems, Automatica, 24 (1988), pp. 653–658. [24] F.L. Lewis, A survey of linear singular systems, Circuits Systems Signal Process, 5 (1986), pp. 3–36. [25] F.L. Lewis and K. Ozcaldiran, On the eigenstructure assigment of singular sys- tems, Proceedings of 24th Conference on Decision and Control (1985), pp. 179–182. [26] A. Markus, Introduction to the Spectral Theory of Operator Polynomials, AMS Trans. Monographs, Providence, RI, 1988. [27] C. Mehl, V. Mehrmann, A. Ran and L. Rodman, Eigenvalue perturbation theory of structured matrices under generic structured rank one perturbations: General results and complex matrices, Linear Algebra Appl., 435 (2011), pp. 687–716. [28] G. Miminis, Deflation in eigenvalue assignment of descriptor systems using state feedack, IEEE Transactions of Automatic control, 38 (1993), pp. 1322–1336. [29] C. Mehl, V. Mehrmann and M. Wojtylak, On the distance to singularity via low rank perturbations, Oper. Matrices, 9 (2015), pp. 733–772. [30] A. Ran and M. Wojtylak, Eigenvalues of rank one perturbations of unstructured matrices, Linear Algebra Appl., 437 (2012), pp. 589–600. 21 [31] L. Rodman, An Introduction to Operator Polynomials, Birkhäuser Verlag, Basel, 1989. [32] H.H. Rosenbrock, State Space and Multivariable Theory, Thomas Nelson and Sons, London, 1970. [33] S.V. Savchenko, Typical changes in spectral properties under perturbations by a rank- one operator, Mat. Zametki, 74 (2003), pp. 590–602. [34] S.V. Savchenko, On the change in the spectral properties of a matrix under a per- turbation of a sufficiently low rank, Funktsional. Anal. i Prilozhen, 38 (2004), pp. 85–88. [35] G. Stewart and J. Sun, Matrix Perturbation Theory, Academic Press Inc., Boston, 1990. [36] R. Thompson, The behavior of eigenvalues and singular values under perturbations of restricted rank, Linear Algebra Appl., 13 (1976), pp. 69–78. [37] R. Thompson, Invariant factors under rank one perturbations, Canad. J. Math, 32 (1980), pp. 240–245. 22
1304.3836	1	1304	2013-04-13T19:44:25	The group of automorphisms of the Lie algebra of derivations of a polynomial algebra	[ "math.RA", "math.AG" ]	We prove that the group of automorphisms of the Lie algebra $\Der_K (P_n)$ of derivations of a polynomial algebra $P_n=K[x_1,..., x_n]$ over a field of characteristic zero is canonically isomorphic to the the group of automorphisms of the polynomial algebra $P_n$.	math.RA	math	The group of automorphisms of the Lie algebra of derivations of a polynomial algebra V. V. Bavula Abstract We prove that the group of automorphisms of the Lie algebra DerK (Pn) of derivations of a polynomial algebra Pn = K[x1, . . . , xn] over a field of characteristic zero is canonically isomorphic to the the group of automorphisms of the polynomial algebra Pn. Key Words: Group of automorphisms, monomorphism, Lie algebra, automorphism, locally nilpotent derivation. Mathematics subject classification 2010: 17B40, 17B20, 17B66, 17B65, 17B30. 1 Introduction In this paper, module means a left module, K is a field of characteristic zero and K ∗ is its group of units, and the following notation is fixed: • Pn := K[x1, . . . , xn] =Lα∈Nn Kxα is a polynomial algebra over K where xα := xα1 • Gn := AutK(Pn) is the group of automorphisms of the polynomial algebra Pn, 1 · · · xαn n , • ∂1 := ∂ ∂x1 , . . . , ∂n := ∂ ∂xn are the partial derivatives (K-linear derivations) of Pn, • Dn := DerK(Pn) = Ln ∂δ − δ∂, i=1 Pn∂i is the Lie algebra of K-derivations of Pn where [∂, δ] := • δ1 := ad(∂1), . . . , δn := ad(∂n) are the inner derivations of the Lie algebra Dn determined by the elements ∂1, . . . , ∂n (where ad(a)(b) := [a, b]), • Gn := AutLie(Dn) is the group of automorphisms of the Lie algebra Dn, i=1 K∂i, • Dn :=Ln • Hn :=Ln • An := Khx1, . . . , xn, ∂1, . . . , ∂ni =Lα,β∈Nn Kxα∂β is the n'th Weyl algebra, i=1 KHi where H1 := x1∂1, . . . , Hn := xn∂n, • for each natural number n ≥ 2, un := K∂1+P1∂2+· · ·+Pn−1∂n is the Lie algebra of triangular polynomial derivations (it is a Lie subalgebra of the Lie algebra Dn) and AutK(un) is its group of automorphisms. The aim of the paper is to prove the following theorem. Theorem 1.1 Gn = Gn. Structure of the proof. (i) Gn ⊆ Gn via the group monomorphism (Lemma 2.3.(3)) Gn → Gn, σ 7→ σ : ∂ 7→ σ(∂) := σ∂σ−1. (ii) Let σ ∈ Gn. Then ∂′ 1 := σ(∂1), . . . , ∂′ n := σ(∂n) are commuting, locally nilpotent deriva- tions of the polynomial algebra Pn (Lemma 2.6.(1)). 1 i=1 kerPn (∂′ i) = K (Lemma 2.6.(2)). (iii) Tn (iv)(crux) There exists a polynomial automorphism τ ∈ Gn such that τ σ ∈ FixGn (∂1, . . . , ∂n) (Corollary 2.9). (v) FixGn (∂1, . . . , ∂n) = Shn (Proposition 2.5.(3)) where Shn := {sλ ∈ Gn sλ(x1) = x1 + λ1, . . . , sλ(xn) = xn + λn} is the shift group of automorphisms of the polynomial algebra Pn and λ = (λ1, . . . , λn) ∈ K n. (vi) By (iv) and (v), σ ∈ Gn, i.e. Gn = Gn. (cid:3) An analogue of the Jacobian Conjecture is true for Dn. The Jacobian Conjecture claims that certain monomorphisms of the polynomial algebra Pn are isomorphisms: Every algebra endo- morphism σ of the polynomial algebra Pn such that J (σ) := det( ∂σ(xi) ) ∈ K ∗ is an automorphism. ∂xj The condition that J (σ) ∈ K ∗ implies that the endomorphism σ is a monomorphism. Conjecture. Every homomorphism of the Lie algebra Dn is an automorphism. Theorem 1.2 [4] Every monomorphism of the Lie algebra un is an automorphism. Remark. Not every epimorphism of the Lie algebra un is an automorphism. Moreover, there are countably many distinct ideals {Iiωn−1 i ≥ 0} such that I0 = {0} ⊂ Iωn−1 ⊂ I2ωn−1 ⊂ · · · ⊂ Iiωn−1 ⊂ · · · and the Lie algebras un/Iiωn−1 and un are isomorphic (Theorem 5.1.(1), [5]). Theorems 1.2 and Conjecture have bearing of the Jacobian Conjecture and the Conjecture of Dixmier [8] for the Weyl algebra An over a field of characteristic zero that claims: every homomorphism of the Weyl algebra is an automorphism. The Weyl algebra An is a simple algebra, so every algebra endomorphism of An is a monomorphism. This conjecture is open since 1968 for all n ≥ 1. It is stably equivalent to the Jacobian Conjecture for the polynomial algebras as was shown by Tsuchimoto [9], Belov-Kanel and Kontsevich [7], (see also [2] for a short proof which is based on the author's new inversion formula for polynomial automorphisms [1]). An analogue of the Conjecture of Dixmier is true for the algebra I1 := Khx, d of polynomial integro-differential operators. dx ,R i Theorem 1.3 (Theorem 1.1, [3]) Each algebra endomorphism of I1 is an automorphism. In contrast to the Weyl algebra A1 = Khx, d dx i, the algebra of polynomial differential operators, the algebra I1 is neither a left/right Noetherian algebra nor a simple algebra. The left localizations, A1,∂ and I1,∂, of the algebras A1 and I1 at the powers of the element ∂ = d dx are isomorphic. For the simple algebra A1,∂ ≃ I1,∂ , there are algebra endomorphisms that are not automorphisms [3]. The group of automorphisms of the Lie algebra un. In [6], the group of automor- phisms AutK(un) of the Lie algebra un of triangular polynomial derivations is found (n ≥ 2), it is isomorphic to an iterated semi-direct product (Theorem 5.3, [6]), Tn ⋉ (UAutK(Pn)n ⋊ (F′ n × En)) where Tn is an algebraic n-dimensional torus, UAutK(Pn)n is an explicit factor group of the group UAutK(Pn) of unitriangular polynomial automorphisms, F′ n and En are explicit groups that are isomorphic respectively to the groups I and Jn−2 where I := (1 + t2K[[t]], ·) ≃ K N 2 and J := (tK[[t]], +) ≃ K N. Comparing the groups Gn and AutK(un) we see that the group (UAutK(Pn)n of polynomial automorphisms is a tiny part of the group AutK(un) but in contrast Gn = AutK(Pn). It is shown that the adjoint group of automorphisms A(un) of the Lie algebra un is equal to the group UAutK(Pn)n (Theorem 7.1, [6]). Recall that the adjoint group A(G) of a i! ∈ AutK(G) where g runs through all the locally nilpotent elements of the Lie algebra G (an element g is a locally nilpotent element if the inner derivation ad(g) := [g, ·] of the Lie algebra G is a locally nilpotent derivation). Lie algebra G is generated by the elements ead(g) :=Pi≥0 ad(g)i 2 Proof of Theorem 1.1 This section can be seen as a proof of Theorem 1.1. The proof is split into several statements that reflect 'Structure of the proof of Theorem 1.1' given in the Introduction. The Lie algebra Dn is Zn-graded. The Lie algebra n Dn = Mα∈Nn Mi=1 Kxα∂i (1) is a Zn-graded Lie algebra Dn = Mβ∈Zn Dn,β where Dn,β = Mα−ei=β Kxα∂i, i.e. [Dn,α, Dn,β] ⊆ Dn,α+β for all α, β ∈ Nn where e1 := (1, 0, . . . , 0), . . . , en := (0, . . . , 0, 1) is the canonical free basis for the free abelian group Zn. This follows from the commutation relations Clearly, for all i, j = 1, . . . , n and α ∈ Nn, [xα∂i, xβ∂j] = βixα+β−ei ∂j − αjxα+β−ej ∂i. [Hj, xα∂i] =(αj xα∂i (αi − 1)xα∂i if j 6= i, if j = i, [∂j, xα∂i] = αjxα−ej ∂i. (2) (3) (4) The support Supp(Dn) := {β ∈ Zn Dn,β 6= 0} is a submonoid of Zn. Let us find the support Supp(Dn), the graded components Dn,β and their dimensions dimK Dn,β. For each i = 1, . . . , n, let Nn,i := {α ∈ Nn αi = 0} and P ∂i n := kerPn (∂i). It follows from the decompositions Pn = P ∂i n ⊕ Pnxi for i = 1, . . . , n that Dn = n Mi=1 Hence, Dn = n Mi=1 (P ∂i n ⊕ Pnxi)∂i = P ∂i n ∂i ⊕ PnHi, n Mi=1 n Mi=1 n ∂i ⊕ Mα∈Nn P ∂i xαHn. Supp(Dn) = xβHn Dn,β =(xα∂i dimK Dn,β =(1 n ai=1 (Nn,i − ei)a Nn. if β = α − ei ∈ Nn,i − ei, if β ∈ Nn. if β = α − ei ∈ Nn,i − ei, n if β ∈ Nn. 3 (5) (6) (7) Let G be a Lie algebra and H be its Lie subalgebra. The centralizer CG(H) := {x ∈ G [x, H] = 0} of H in G is a Lie subalgebra of G. In particular, Z(G) := CG(G) is the centre of the Lie algebra G. The normalizer NG(H) := {x ∈ G [x, H] ⊆ H} of H in G is a Lie subalgebra of G, it is the largest Lie subalgebra of G that contains H as an ideal. Let V be a vector space over K. A K-linear map δ : V → V is called a locally nilpotent map if V = Si≥1 ker(δi) or, equivalently, for every v ∈ V , δi(v) = 0 for all i ≫ 1. When δ is a locally nilpotent map in V we also say that δ acts locally nilpotently on V . Every nilpotent linear map δ, that is δn = 0 for some n ≥ 1, is a locally nilpotent map but not vice versa, in general. Let G be a Lie algebra. Each element a ∈ G determines the derivation of the Lie algebra G by the rule ad(a) : G → G, b 7→ [a, b], which is called the inner derivation associated with a. The set Inn(G) of all the inner derivations of the Lie algebra G is a Lie subalgebra of the Lie algebra (EndK(G), [·, ·]) where [f, g] := f g − gf . There is the short exact sequence of Lie algebras 0 → Z(G) → G ad→ Inn(G) → 0, that is Inn(G) ≃ G/Z(G) where Z(G) is the centre of the Lie algebra G and ad([a, b]) = [ad(a), ad(b)] for all elements a, b ∈ G. An element a ∈ G is called a locally nilpotent element (respectively, a nilpotent element) if so is the inner derivation ad(a) of the Lie algebra G. The Cartan subalgebra Hn of Dn. A nilpotent Lie subalgebra C of a Lie algebra G is called a Cartan subalgebra of G if it coincides with its normalizer. We use often the following obvious observation: An abelian Lie subalgebra that coincides with its centralizer is a maximal abelian Lie subalgebra. Lemma 2.1 1. Hn is a Cartan subalgebra of Dn. 2. Hn = CDn (Hn) is a maximal abelian subalgebra of Dn. Proof. Statements 1 and 2 follows from (6) and (7). (cid:3) Pn is a Dn-module. The polynomial algebra Pn is a (left) Dn-module: Dn × Pn → Pn, i=1 ai∂i where ai ∈ Pn then (∂, p) 7→ ∂ ∗ p. In more detail, if ∂ =Pn ∂ ∗ p = The field K is a Dn-submodule of Pn and ai ∂p ∂xi . n Xi=1 kerPn (∂i) = K. (8) n \i=1 Lemma 2.2 The Dn-module Pn/K is simple with EndDn (Pn/K) = Kid where id is the identity map. Proof. Let M be a nonzero submodule of Pn/K and 0 6= p ∈ M . Using the actions of ∂1, . . . , ∂n on p we obtain an element of M of the form λxi for some λ ∈ K ∗. Hence, xi ∈ M and xα = xα∂i ∗ xi ∈ M for all 0 6= α ∈ Nn. Therefore, M = Pn/K. Let f ∈ EndDn (Pn/K). Then applying f to the equalities ∂i ∗ (x1 + K) = δi1 for i = 1, . . . , n, we obtain the equalities ∂i ∗ f (x1 + K) = δi1 for i = 1, . . . , n. So, f (x1 + K) = λ(x1 + K) and so f = λid, by the simplicity of the Dn-module Pn/K. i=2 kerPn/K(∂i) ∩ kerPn/K(∂2 i ) = (K[x1]/K) ∩ kerPn/K(∂2 i ) = K(x1 + K). Hence, f (x1 + K) ∈ Tn (cid:3) 4 The Gn-module Dn. The Lie algebra Dn is a Gn-module, Gn × Dn → Dn, (σ, ∂) 7→ σ(∂) := σ∂σ−1. Every automorphism σ ∈ Gn is uniquely determined by the elements x′ 1 := σ(x1), . . . , x′ n := σ(xn). Let Mn(Pn) be the algebra of n × n matrices over Pn. The matrix J(σ) := (J(σ)ij ) ∈ Mn(Pn), where J(σ)ij = , is called the Jacobian matrix of the automorphism (endomorphism) σ and its determinant J (σ) := det J(σ) is called the Jacobian of σ. So, the j'th column of J(σ) is the gradient grad x′ )T of the polynomial x′ j. Then the derivations j := ( ∂x′ j ∂xi , . . . , ∂x′ j ∂x1 ∂x′ j ∂xn n := σ∂nσ−1 are the partial derivatives of Pn with respect to the variables x′ 1 := σ∂1σ−1, . . . , ∂′ ∂′ 1, . . . , x′ n, ∂′ 1 = ∂ ∂x′ 1 , . . . , ∂′ n = ∂ ∂x′ n . (9) Every derivation ∂ ∈ Dn is a unique sum ∂ = Pn (∂1, . . . , ∂n)T and ∂′ := (∂′ 1, . . . , ∂′ n)T where T stands for the transposition. Then i=1 ai∂i where ai = ∂ ∗ xi ∈ Pn. Let ∂ := ∂′ = J(σ)−1∂, i.e. ∂′ i = (J(σ)−1)ij∂j for i = 1, . . . , n. (10) n Xj=1 δij = ∂′ i ∗ x′ j = aik ∂x′ j ∂xk Xk=1 In more detail, if ∂′ = A∂ where A = (aij ) ∈ Mn(Pn), i.e. ∂i = Pn i, j = 1, . . . , n, n j=1 aij ∂j. Then for all where δij is the Kronecker delta function. The equalities above can be written in the matrix form as AJ(σ) = 1 where 1 is the identity matrix. Therefore, A = J(σ)−1. Suppose that a group G acts on a set S. For a nonempty subset T of S, StG(T ) := {g ∈ G gT = T } is the stabilizer of the set T in G and FixG(T ) := {g ∈ G gt = t for all t ∈ T } is the fixator of the set T in G. Clearly, FixG(T ) is a normal subgroup of StG(T ). The maximal abelian Lie subalgebra Dn of Dn. Lemma 2.3 1. CDn (Dn) = Dn and so Dn is a maximal abelian Lie subalgebra of Dn. 2. FixGn(Dn) = FixGn(∂1, . . . , ∂n) = Shn. 3. Dn is a faithful Gn-module, i.e. the group homomorphism Gn → Gn, σ 7→ σ : ∂ 7→ σ∂σ−1, is a monomorphism. 4. FixGn(∂1, . . . , ∂n, H1, . . . , Hn) = {e}. Proof. 1. Statement 1 follows from (2). 2. Let σ ∈ FixGn (Dn) and J(σ) = (Jij ). By (10), ∂ = J(σ)∂, and so, for all i, j = 1, . . . , n, δij = ∂i ∗ xj = Jij , i.e. J(σ) = 1, or equivalently, by (8), 1 = x1 + λ1, . . . , x′ x′ n = xn + λn for some scalars λi ∈ K, and so σ ∈ Shn. 3 and 4. Let σ ∈ FixGn = (∂1, . . . , ∂n, H1, . . . , Hn). Then σ ∈ FixGn(∂1, . . . , ∂n) = Shn, by statement 2. So, σ(x1) = x1 + λ1, . . . , σ(xn) = xn + λn where λi ∈ K. Then xi∂i = σ(xi∂i) = (xi + λi)∂i for i = 1, . . . , n, and so λ1 = · · · = λn = 0. This means that σ = e. So, FixGn = (∂1, . . . , ∂n, H1, . . . , Hn) = {e} and Dn is a faithful Gn-module. (cid:3) By Lemma 2.3.(3), we identify the group Gn with its image in Gn. 5 Lemma 2.4 1. Dn is a simple Lie algebra. 2. Z(Dn) = {0}. 3. [Dn, Dn] = Dn. Proof. 1. Let 0 6= a ∈ Dn and a = (a) be the ideal of the Lie algebra Dn generated by the element a. We have to show that a = Dn. Using the inner derivations δ1, . . . , δn we see that ∂i ∈ a for some i. Then a = Dn since xα∂j = (αi + 1)−1[∂i, xα+ei ∂j] ∈ a for all α and j. 2 and 3. Statements 2 and 3 follow from statement 1. (cid:3) Proposition 2.5 1. FixGn (∂1, . . . , ∂n, H1, . . . , Hn) = {e}. 2. Let σ, τ ∈ Gn. Then σ = τ iff σ(∂i) = τ (∂i) and σ(Hi) = τ (Hi) for i = 1, . . . , n. 3. FixGn (∂1, . . . , ∂n) = Shn. Proof. 1. Let σ ∈ F := FixGn (∂1, . . . , ∂n, H1, . . . , Hn). We have to show that σ = e. Since σ ∈ FixGn (H1, . . . , Hn), the automorphism σ respects the weight decomposition of Dn. By (7), σ(xα∂i) = λα,ixα∂i for all α ∈ Nn,i and i = 1, . . . , n where λα,i ∈ K. Clearly, λ0,i = 1 for i = 1, . . . , n. Since σ ∈ FixGn (∂1, . . . , ∂n), by applying σ to the relations αjxα−ej ∂i = [∂j, xα∂i], we get the relations αjλα−ej ,ixα−ej ∂i = [∂j , λα,ixα∂i] = αjλα,ixα−ej ∂i. Hence λα,i = λα−ej ,i provided αj 6= 0. We conclude that all the coefficients λα,i are equal to one of the coefficients λei,j where i, j = 1, . . . , n and i 6= j. The relations ∂j = [∂i, xi∂j] implies the relations ∂j = [∂i, λei,jxi∂j] = λei,j∂j , hence all the coefficients λei,j are equal to 1. So, ⊕n n ∂i ⊆ F := FixDn (σ) := {∂ ∈ Dn σ(∂) = ∂}. To finish the proof of statement 1 it suffices to show that xαHi ∈ F for all α ∈ Nn and i = 1, . . . , n, see (5) and (6). We use induction on α := α1 + · · · + αn. If α = 0 the statement is obvious as σ ∈ F . Suppose that α > 0. Using the commutation relations i=1P ∂i [∂j , xαHi] =(αj xα−ej Hi (αi + 1)xα∂i if j 6= i, if j = i, (11) the induction and the previous case, we see that [∂j, σ(xαHi) − xαHi] = 0 for i = 1, . . . , n. Therefore, σ(xαHi) − xαHi ∈ CDn (Dn) = Dn. Since the automorphism σ respects the weight decomposition of Dn, we must have σ(xαHi) − xαHi ∈ xαHn ∩ Dn = {0}. Hence, xαHi ∈ F , as required. 2. Statement 2 follows from statement 1. 3. Clearly, Shn ⊆ F = FixGn (∂1, . . . , ∂n). Let σ ∈ F and H ′ i := σ(Hi), . . . , H ′ n := σ(Hn). Applying the automorphism σ to the commutation relations [∂i, Hj] = δij∂i gives the relations [∂i, H ′ j − Hj] = 0 for all i and j. Therefore, H ′ j=1 λij ∂j for some elements λij ∈ K. The elements H ′ i = Hi + di for some elements di ∈ CDn (Dn) = Dn (Lemma 2.3.(1)), and so di =Pn j] = δij ∂i. By taking the difference, we see that [∂i, H ′ 1, . . . , H ′ n commute, hence [Hj, ∂i] = [Hi, ∂j] for all i, j, 6 or equivalently, λij ∂j = λji∂i for all i, j. This means that λij = 0 for all i 6= j, i.e. H ′ i = Hi + λii∂i = (xi + λii)∂i = sλ(Hi) where sλ ∈ Shn, sλ(xi) = xi + λii for all i. Then s−1 (statement 2), and so σ = sλ ∈ Shn. (cid:3) λ σ ∈ FixGn (∂1, . . . , ∂n, H1, . . . , Hn) = {e} Lemma 2.6 Let σ ∈ Gn and ∂′ 1 := σ(∂1), . . . , ∂′ n := σ(∂n). Then 1. ∂′ 1, . . . , ∂′ n are commuting, locally nilpotent derivations of Pn. i=1 kerDn (∂′ i) = K. 2. Tn Proof. 1. The derivations ∂′ n commute since ∂1, . . . , ∂n are commute. The inner deriva- tions δ1, . . . , δn of the Lie algebra Dn are commuting and locally nilpotent. Hence, inner derivations 1, . . . , ∂′ δ′ 1 := ad(∂′ 1), . . . , δ′ n := ad(∂′ n) of the Lie algebra Dn are commuting and locally nilpotent. The vector space Pn∂′ the derivations δ′ j since i is closed under δ′ j(Pn∂′ i) = [∂′ j, Pn∂′ i] = (∂′ j ∗ Pn) · ∂′ i ⊆ Pn∂′ i. Therefore, ∂′ 1, . . . , ∂′ n are locally nilpotent derivations of the polynomial algebra Pn. 2. Let λ ∈Tn i=1 kerPn (∂′ i). Then λ∂′ 1 ∈ CDn (∂′ 1, . . . , ∂′ n) = σ(CDn (∂1, . . . , ∂n)) = σ(CDn (Dn)) = σ(Dn) = σ( n Mi=1 K∂i) = K∂′ i, n Mi=1 since CDn (Dn) = Dn, Lemma 2.3.(1). Then λ ∈ K since otherwise the infinite dimensional space 1 would be a subspace of a finite dimensional space σ(Dn). (cid:3) Li≥0 Kλi∂′ The following lemma is well-known and it is easy to prove. Lemma 2.7 Let ∂ be a locally nilpotent derivation of a commutative K-algebra A such that ∂(x) = 1 for some element x ∈ A. Then A = A∂ [x] is a polynomial algebra over the ring A∂ := ker(∂) of constants of the derivation ∂ in the variable x. The next theorem is the most important point in the proof of Theorem 1.1 and, roughly speaking, the main reason why Theorem 1.1 holds. Theorem 2.8 Let ∂′ 1, . . . , ∂′ n be commuting, locally nilpotent derivations of the polynomial algebra i=1 kerPn (∂′ i) = K. Then there exist polynomials x′ 1, . . . , x′ n ∈ Pn such that Pn such that Tn ∂′ i ∗ x′ j = δij. (12) Moreover, the algebra homomorphism σ : Pn → Pn, x1 7→ x′ 1, . . . , xn 7→ x′ n is an automorphism such that ∂′ i = σ∂iσ−1 = ∂ ∂x′ i for i = 1, . . . , n. 7 Proof. Case n = 1: By Lemma 2.6, the derivation ∂′ 1) = K. Hence, ∂′ 1], and so σ : K[x1] → K[x1], x 7→ x′ 1 of the polynomial algebra P1 is a locally 1 ∈ P1. 1, is an automorphism 1 = 1 for some polynomial x′ 1 ∗ x′ i) for i = 1, . . . , n. Clearly, K ⊆ K ′ i. i = K for some i then by the same argument as in the case i] = K[xi], a i = 1, and so Pn = K ′ i ∈ Pn such that ∂′ i ∗ x′ i[x′ nilpotent derivation with K ′ By Lemma 2.7, K[x1] = K ′ such that ∂′ 1 := kerP1 (∂′ 1[x′ 1] = K[x′ σ−1. 1 = d dx′ 1 = σ d dx1 i := kerPn (∂′ Case n ≥ 2. Let K ′ (i) K ′ i 6= K for i = 1, . . . , n: If K ′ n = 1 there exists a polynomial x′ contradiction. 2 ≤ m ≤ n − 1. Changing (if necessary) the order of the derivations ∂′ (ii) Let m be the maximum of card(I) such ∅ 6= I ⊆ {1, . . . , n − 1} and Ti∈I K ′ that A := Tm i 6= K. By (i), n we may assume i 6= K. Then the algebra A is infinite dimensional (since K 6= A ⊆ Pn) and invariant under the action of the derivations ∂′ j for j = m + 1, . . . , n. By the choice of m, 1, . . . , ∂′ i=1 K ′ A∂ ′ j = K ′ j ∩ K ′ i = K for j = m + 1, . . . , n m \i=1 and the derivations ∂′ j = m + 1, . . . , n, there exists an element x′ j acts locally nilpotently on the algebra A∂ ′ j ∈ A such that ∂′ j ∗ x′ j . Therefore, for each index j = 1, and so (Lemma 2.7) A = A∂ ′ j [x′ j ] = K[x′ j] for j = m + 1, . . . , n. (13) (ii)(a) Suppose that m = n − 1, i.e. ∂′ The algebra K ′ n admits the set of commuting, locally nilpotent derivations i ∗ x′ n = δin for all i = 1, . . . , n. By Lemma 2.7, Pn = K ′ n[x′ n]. ∂′′ 1 := ∂′ 1K ′ n, . . . , ∂′′ n−1 := ∂′ n−1K ′ n with Tn−1 i=1 kerK ′ i = K. (ii)(b) Suppose that m < n − 1. By (13), i ) = K ′ i=1 K ′ n (∂′′ n ∩Tn−1 K ∗x′ m+1 + K = K ∗x′ m+2 + K = · · · = K ∗x′ n + K, and so λj := ∂′ n = 0 for j = m + 1, . . . , n − 1. A linear combination of commuting, locally nilpotent derivations is a locally nilpotent derivation (the proof boils down to the case ∂ + δ of two commuting, locally nilpotent n ∈ K for j = m + 1, . . . , n − 1. Hence, (∂′ j − λj∂′ n) ∗ x′ j ∗ x′ derivations, then the result follows from (∂ + δ)m = Pm Using the set of commuting, locally nilpotent derivations ∂′ the set of commuting, locally nilpotent derivations i=0(cid:0)m i(cid:1)∂iδm−i and ∂iδm−i = δm−i∂i). n that satisfy (12) we obtain 1, . . . , ∂′ δ′ 1 := ∂′ 1, . . . , δ′ m := ∂′ m, δ′ m+1 := ∂′ m+1 − λm+1∂′ n, . . . , δ′ n−1 := ∂′ n−1 − λn−1∂′ n, δ′ n := ∂n that satisfy (12) with n = δin for i = 1, . . . , n. Then repeating the arguments of (ii)(a), we see that Pn = K ′ set of commuting, locally nilpotent derivations δ′ i ∗ x′ n[x′ n]. The algebra K ′ n admits the ∂′′ 1 := δ′ 1K ′ n , . . . , ∂′′ n−1 := δ′ n−1K ′ n with n−1 n−1 n−1 n kerK ′ n(∂′′ i ) = K ′ n ∩ kerPn (δ′ i) = K ′ n ∩ kerPn (∂′ i) = K ′ i = K. \i=1 \i=1 \i=1 \i=1 (iii) Using the cases (ii)(a) and (ii)(b) n − 1 more times we find polynomials x′ n and commuting set of locally nilpotent derivations of Pn, say, ∆1, . . . , ∆n that satisfy (12) and such that 1, . . . , x′ (α) ∆i ∗ x′ j = δij for all i, j = 1, . . . , n; 8 (β) the n-tuple of derivations ∆ = (∆1, . . . , ∆n)T is obtained from the n-tuple of derivations n)T by unitriangular (hence invertible) scalar matrix Λ = (λij ) ∈ Mn(K) such that 1, . . . , ∂′ ∂′ = (∂′ ∆ = Λ∂′; and (γ) (where K ′′ 1 := kerPn (∆1), . . . , K ′′ n := kerPn (∆n)) n Pn = K ′′ n[x′ n] = (K ′′ n−1 ∩ K ′′ n)[x′ n−1, x′ n] = · · · = ( K ′′ i )[x′ s, . . . , x′ n] = · · · n = ( \i=1 K ′′ i )[x′ 1, . . . , x′ n] = K[x′ 1, . . . , x′ n]. \i=s (iv) Replacing the row x′ = (x′ Indeed, by (α), Λ · (∂′ 1, . . . , x′ n) by the row x′Λ gives the required elements of the theorem. i ∗ x′ j ) · Λ = 1, as required. n be the set of polynomials as in the theorem. Then σ is an algebra automor- j ) = 1, the identity n × n matrix. Hence, (∂′ i ∗ x′ (v) Let x′ 1, . . . , x′ phism (see (γ) and (iv)) such that ∂′ i = σ∂iσ−1 = ∂ ∂x′ i for i = 1, . . . , n. (cid:3) Corollary 2.9 Let σ ∈ Gn. Then τ σ ∈ FixGn (∂1, . . . , ∂n) for some τ ∈ Gn. Proof. By Lemma 2.6, the elements ∂′ 1 := σ(∂1), . . . , ∂′ 1 := τ −1(∂1), . . . , ∂′ n := σ(∂n) satisfy the assumptions of n := τ −1(∂n) for some τ ∈ Gn. Therefore, Theorem 2.8. By Theorem 2.8, ∂′ τ σ ∈ FixGn (∂1, . . . , ∂n). (cid:3) Proof of Theorem 1.1. Let σ ∈ Gn. By Corollary 2.9, τ σ ∈ FixGn (∂1, . . . , ∂n) = Shn (Proposition 2.5.(3)). Therefore, σ ∈ Gn, i.e. Gn = Gn. (cid:3) The work is partly supported by the Royal Society and EPSRC. Acknowledgements References [1] V. V. Bavula, The inversion formula for automorphisms of the Weyl algebras and polynomial algebras, J. Pure Appl. Algebra 210 (2007), 147-159; arXiv:math.RA/0512215. [2] V. V. Bavula, The Jacobian Conjecture2n implies the Dixmier Problemn, ArXiv:math.RA/0512250. [3] V. V. Bavula, An analogue of the Conjecture of Dixmier is true for the algebra of polynomial integro- differential operators, J. of Algebra 372 (2012) 237-250. Arxiv:math.RA: 1011.3009. [4] V. V. Bavula, Every monomorphism of the Lie algebra of triangular polynomial derivations is an auto- morphism, C. R. Acad. Sci. Paris, Ser. I, 350 (2012) no. 11-12, 553 -- 556. (Arxiv:math.AG:1205.0797). [5] V. V. Bavula, Lie algebras of triangular polynomial derivations and an isomorphism criterion for their Lie factor algebras, Izvestiya: Mathematics, (2013), in print. (Arxiv:math.RA:1204.4908). [6] V. V. Bavula, The groups of automorphisms of the Lie algebras of triangular polynomial derivations, Arxiv:math.AG/1204.4910. [7] A. Belov-Kanel and M. Kontsevich, The Jacobian conjecture is stably equivalent to the Dixmier Conjecture, Mosc. Math. J. 7 (2007), no. 2, 209 -- 218 (arXiv:math. RA/0512171). [8] J. Dixmier, Sur les alg`ebres de Weyl, Bull. Soc. Math. France 96 (1968), 209 -- 242. [9] Y. Tsuchimoto, Endomorphisms of Weyl algebra and p-curvatures. Osaka J. Math. 42 (2005), no. 2, 435 -- 452. Department of Pure Mathematics University of Sheffield Hicks Building Sheffield S3 7RH UK email: [email protected] 9
1212.2124	2	1212	2013-01-14T13:49:23	Semi-Invariant Subrings	[ "math.RA" ]	We say that a subring $R_0$ of a ring $R$ is semi-invariant if $R_0$ is the ring of invariants in $R$ under some set of ring endomorphisms of some ring containing $R$. We show that $R_0$ is semi-invariant if and only if there is a ring $S\supseteq R$ and a set $X\subseteq S$ such that $R_0=\Cent_R(X):={r\in R \suchthat xr=rx \forall x\in X}$; in particular, centralizers of subsets of $R$ are semi-invariant subrings. We prove various properties of semi-invariant subrings and show how they can be used for various applications including: (1) The center of a semiprimary (resp. right perfect) ring is semiprimary (resp. right perfect). (2) If $M$ is a finitely presented module over a "good" semiperfect ring (e.g. an inverse limit of semiprimary rings), then $(M)$ is semiperfect, hence $M$ has a Krull-Schmidt decomposition. (This generalizes results of Bjork and Rowen). (3) If $\rho$ is a representation of a monoid or a ring over a module with a "good" semiperfect endomorphism ring (in the sense of (2)), then $\rho$ has a Krull-Schmidt decomposition. (4) If $S$ is a "good" commutative semiperfect ring and $R$ is an $S$-algebra that is f.p.\ as an $S$-module, then $R$ is semiperfect. (5) Let $R\subseteq S$ be rings and let $M$ be a right $S$-module. If $(M_R)$ is semiprimary (resp. right perfect), then $(M_S)$ is semiprimary (resp. right perfect).	math.RA	math	SEMI-INVARIANT SUBRINGS URIYA A. FIRST Abstract. We say that a subring R0 of a ring R is semi-invariant if R0 is the ring of invariants in R under some set of ring endomorphisms of some ring containing R. We show that R0 is semi-invariant if and only if there is a ring S ⊇ R and a set X ⊆ S such that R0 = CentR(X) := {r ∈ R : xr = rx ∀x ∈ X}; in particular, centralizers of subsets of R are semi-invariant subrings. We prove that a semi-invariant subring R0 of a semiprimary (resp. right per- fect) ring R is again semiprimary (resp. right perfect) and satisfies Jac(R0)n ⊆ Jac(R) for some n ∈ N. This result holds for other families of semiperfect rings, but the semiperfect analogue fails in general. In order to overcome this, we specialize to Hausdorff linearly topologized rings and consider topologically semi-invariant subrings. This enables us to show that any topologically semi- invariant subring (e.g. a centralizer of a subset) of a semiperfect ring that can be endowed with a "good" topology (e.g. an inverse limit of semiprimary rings) is semiperfect. Among the applications: (1) The center of a semiprimary (resp. right per- fect) ring is semiprimary (resp. right perfect). (2) If M is a finitely presented module over a "good" semiperfect ring (e.g. an inverse limit of semiprimary rings), then End(M ) is semiperfect, hence M has a Krull-Schmidt decomposi- tion. (This generalizes results of Bjork and Rowen; see [5], [24], [23].) (3) If ρ is a representation of a monoid or a ring over a module with a "good" semiperfect endomorphism ring (in the sense of (2)), then ρ has a Krull-Schmidt decompo- sition. (4) If S is a "good" commutative semiperfect ring and R is an S-algebra that is f.p. as an S-module, then R is semiperfect. (5) Let R ⊆ S be rings and let M be a right S-module. If End(MR) is semiprimary (resp. right perfect), then End(MS ) is semiprimary (resp. right perfect). 1. Preface Throughout, all rings are assumed to have a unity and ring homomorphisms are required to preserve it. Subrings are assumed to have the same unity as the ring containing them. Given a ring R, denote its set of invertible elements by R×, its Jacobson radical by Jac(R), its set of idempotents by E(R) and its center by Cent(R). We let End(R) (resp. Aut(R)) denote the set of ring homomorphisms (resp. isomorphisms) from R to itself. If X ⊆ R is any set, then its right (left) annihilator in R will be denoted by annr R X (annℓ R X). The subscript R will be dropped when understood from the context. A semisimple ring always means a semisimple artinian ring. For a prime number p, we let Zp, Qp and Zhpi denote the p-adic integers, p-adic numbers and Z localized (but not completed) at pZ, respectively. Let R be a ring and let J = Jac(R). Recall that R is semilocal if R/J is semisimple. If in addition J is idempotent lifting, then R is called semiperfect. For a detailed discussion about semiperfect rings, see [25, §2.7] and [4]. Semiperfect rings play an important role in representation theory and module theory because of the Krull-Schmidt Theorem. Recall that an object A in an additive category A is said to have a Krull-Schmidt decomposition if it is a sum of (non-zero) indecomposable Date: September 23, 2018. This research was partially supported by an Israel-US BSF grant #2010/149. 1 2 URIYA A. FIRST objects and any two such decompositions are the same up to isomorphism and reordering. Theorem 1.1. (Krull-Schmidt, for Categories). Let A be an additive category in which all idempotents split (e.g. an abelian category) and let A ∈ A . If EndA (A) is semiperfect, then A has a Krull-Schmidt decomposition and the endomorphism ring of any indecomposable summand of A is local. Generalizations of this theorem and counterexamples of some natural variations are widely studied (e.g. see [14],[3],[1],[12] and also [11]) and there has been con- siderable interest in finding rings over which all finitely presented modules have a Krull-Schmidt decomposition (e.g. [28, §6],[5],[23],[24],[30]; theorem 8.3(iii) below generalizes all these references except the last). Example 1.2. Semiperfect rings naturally appear upon taking completions: (1) Let R be a semilocal ring and let J = Jac(R). Then the J-adic completion of R, lim ←− {R/J n}n∈N, is well known to be semiperfect. If the natural map R → lim←− {R/J n}n∈N is an isomorphism, then R is called complete semilocal. Such rings (especially noetherian or with Jacobson radical f.g. as a right ideal) appear in various areas (e.g. [19], [16], [28, §6], [24]). (2) Let R be a commutative noetherian domain, let A be an R-algebra that is finitely generated as an R-module and let P ∈ Spec(R). Then the comple- tion of A at P is semiperfect (and noetherian). (See [20, §6]; This assertion can also be shown using the results of this paper.) Let R be any ring and let R0 ⊆ R be a subring. (a) Call R0 a semi-invariant subring if there is a ring S ⊇ R and a set Σ ⊆ End(S) such that R0 = RΣ := {r ∈ R : σ(r) = r ∀σ ∈ Σ} (elements of Σ are not required to be injective nor surjective). The invariant subrings of R are the subrings for which we can choose S = R. (b) Call R0 a semi-centralizer subring if there is a ring S ⊇ R and a set X ⊆ S such that R0 = CentR(X) := {r ∈ R : rx = xr ∀x ∈ X}. If we can choose S = R, then R0 is a centralizer subring. 0 . That is, elements (c) Recall that R0 is rationally closed in R if R×∩ R0 = R× of R0 that are invertible in R are also invertible in R0. Semi-centralizer and semi-invariant subrings are clearly rationally closed. The latter were studied (for semilocal R) in [9] and invariant subrings (w.r.t. an arbitrary set) were considered in [5]. However, the notion of semi-invariant subrings appears to be new. The purpose of this paper is to study semi-invariant subrings of semiperfect rings where our motivation comes from the Krull-Schmidt theorem and the following observations, verified in sections 3: (1) For any ring R, a subring of R is semi-invariant if and only if it is semi- centralizer. In particular, all centralizers of subsets of R are semi-invariant subrings. (2) If R ⊆ S are rings and M is a right S-module, then End(MS) is a semi- (3) If M is a finitely presented right R-module, then End(MR) is a quotient of invariant subring of End(MR). a semi-invariant subring of Mn(R) × Mm(R) for some n, m. While in general semi-invariant subrings of semiperfect rings need not be semiper- fect (see Examples 6.1-6.3 below), we show that this is true for special families of semiperfect rings, e.g. for semiprimary and right perfect rings (Theorem 4.6; see Section 2 for definitions). In addition, if the ring in question is pro-semiprimary, i.e. an inverse limit of semiprimary rings (e.g. the rings of Example 1.2), then its SEMI-INVARIANT SUBRINGS 3 topologically semi-invariant subrings (e.g. centralizer subrings; see Section 5 for def- inition) are semiperfect. This actually holds under milder assumptions regarding whether the ring can be endowed with a "good" topology; see Theorems 5.10 and 5.15. Our results together with the previous observations and the Krull-Schmidt The- orem lead to numerous applications including: (1) The center and any maximal commutative subring of a semiprimary (resp. right perfect, semiperfect and pro-semiprimary) ring is semiprimary (resp. right perfect, etc.). (2) If R is a semiperfect pro-semiprimary ring, then all f.p. modules over R have a semiperfect endomorphism ring and hence admit a Krull-Schmidt decomposition. If moreover R is right noetherian, then the endomorphism ring of a f.g. right R-modules is pro-semiprimary. (This generalizes Swan ([28, §6]), Bjork ([5]) and Rowen ([23], [24]) and also relates to works of V´amos ([30]), Facchini and Herbera ([13]); see Remark 8.4 for more details.) (3) If S is a commutative semiperfect pro-semiprimary ring and R is an S- algebra that is Hausdorff (see Section 8) and f.p. as an S-module, then R is semiperfect. If moreover S is noetherian, then the Hausdorff assumption is superfluous and R is pro-semiprimary, hence the assertions of (2) apply. (The first statement is known to hold under mild assumptions for Henselian rings; see [30, Lm. 12].) (4) If ρ is a representation of a ring or a monoid over a module with a semiper- fect pro-semiprimary endomorphism ring, then ρ has a Krull-Schmidt de- composition. (5) Let R ⊆ S be rings and let M be a right S-module. If End(MR) is semipri- In particular, M has a mary (resp. right perfect), then so is End(MS). Krull-Schmidt decomposition over S. (Compare with [13, Pr. 2.7].) Additional applications concern bilinear forms and getting a "Jordan Decomposi- tion" for endomorphisms of modules with semiperfect pro-semiprimary endomor- phism ring. We also conjecture that (3) holds for non-commutative S under mild assumptions (see Section 10). Other interesting byproducts of our work are the fact that a pro-semiprimary ring is an inverse limit of some of its semiprimary quotients and Theorem 9.6 below. (The former assertion fails once replacing semiprimary with right artinian; see Example 9.11 and the comment before it.) Remark 1.3. It is still open whether all semiperfect pro-semiprimary rings are complete semilocal. However, this is true for noetherian rings; see Section 9. Section 2 recites some definitions and well-known facts. Section 3 presents the basics of semi-invariant subrings; we present five equivalent characterizations of them and show that they naturally appear in various situations. As all our char- acterizations use the existence of some ambient ring, we ask whether there is a definition avoiding this. In Section 4, we prove that various ring properties pass to semi-invariant subrings, e.g. being semiprimary and being right perfect. Section 5 develops the theory of T-semi-invariant subrings. The discussion leads to a proof that several properties, such as being pro-semiprimary and semiperfect, are inher- ited by T-semi-invariant subrings. Section 6 presents counterexamples. We show that semi-invariant subrings of semiperfect rings need not be semiperfect, even when the ambient ring is pro-semiprimary. In addition, we show that in general none of the properties discussed in sections 4 and 5 pass to rationally closed subrings. The latter implies that there are non-semi-invariant rationally closed subrings. In sec- tions 7 and 8 we present applications of our results (most applications were briefly 4 URIYA A. FIRST described above) and in Section 9 we specialize them to strictly pro-right-artinian rings (e.g. noetherian pro-semiprimary rings), which are better behaved. Section 10 describes some issues that are still open. The appendix is concerned with providing conditions implying that the topologies {τ M n=1 defined at Section 8 coincide. n }∞ 2. Preliminaries This section recalls some definitions and known facts that will be used throughout the paper. Some of the less known facts include proofs for sake of completion. If no reference is specified, proofs can be found at [25], [4] or [18]. Let R be a semilocal ring. The ring R is called semiprimary (right perfect) if Jac(R) is nilpotent (right T-nilpotent1). Since any nil ideal is idempotent lifting, right perfect rings are clearly semiperfect. Proposition 2.1. (Bass' Theorem P, partial). Let R be a ring. Then R is right perfect ⇐⇒ every right R-module has a projective cover ⇐⇒ R has DCC on principal left ideals. Proposition 2.2. Let R be a ring. Then R is semiperfect ⇐⇒ every right (left) f.g. R-module has a projective cover ⇐⇒ there are orthogonal idempotents i=1 ei = 1 and eiRei is local for all i. e1, . . . , er ∈ R such thatPr For the next lemma, and also for later usage, let us set some conventions about inverse limits of rings: By saying that {Ri, fij} is an inverse system of rings indexed by I, we mean that: (1) I is a directed set (i.e. a partially ordered set such that for all i, j ∈ I there is k ∈ I with i, j ≤ k), (2) Ri (i ∈ I) are rings and fij : Rj → Ri (i ≤ j) are ring homomorphisms and (3) fii = idRi and fijfjk = fik for all i ≤ j ≤ k in I. When the maps {fij} are obvious or of little interest, we will drop them from the notation, writing only {Ri}i∈I . The inverse limit of {Ri, fij} will be denoted by lim←−{Ri}i∈I . It can be understood as the set of I-tuples (ai)i∈I ∈Qi∈I Ri such that fij(aj) = ai for all i ≤ j in I. Lemma 2.3. Let {Ri, fij} be an inverse system of rings and let R = lim←− {Ri}. Denote by fi the standard map from R to Ri. Then: (i) For all n ∈ N, lim←−{Mn(Ri)}i∈I ∼= Mn(lim←−{Ri}) = Mn(R). (ii) For all e ∈ E(R), lim←− {eiRiei}i∈I ∼= eRe where ei = fi(e). Proof. This is straightforward. (cid:3) Proposition 2.4. (i) Being semiprimary (resp. right perfect, semiperfect, semilo- cal, pro-semiprimary) is preserved under Morita equivalence. (ii) Let R be a ring and e ∈ End(R). Then R is semiprimary (resp. right perfect, semiperfect, semilocal) if and only if eRe and (1 − e)R(1 − e) are. Proof. All statements regarding semiprimary, right perfect, semiperfect and semilo- cal rings are well known. The other statements follow from Lemma 2.3. (cid:3) Part (ii) of Proposition 2.4 does not hold for pro-semiprimary rings. For instance, i=1 Zp} (where Zp are the p-adic integers) and take R = {[ a v let e be the matrix unit e11. 0 b ] a, b ∈ Zp, v ∈L∞ Theorem 2.5. (Levitski). Let R be a right noetherian ring. Then any nil subring of R is nilpotent. 1 An ideal I E R is right T-nilpotent if for any sequence x1, x2, x3, · · · ∈ I, the sequence x1, x2x1, x3x2x1, . . . eventually vanishes. SEMI-INVARIANT SUBRINGS 5 Let R be a ring. An element a ∈ R is called right π-regular (in R) if the right ideal chain aR ⊇ a2R ⊇ a3R ⊇ . . . stabilize.2 If a is both left and right π-regular we will say it is π-regular. A ring that all of its elements are right π-regular is called π-regular. It was shown by Dischinger in [10] that the latter property is actually left-right symmetric. Since π-regularity is not preserved under Morita equivalence (see [26]), it is convenient to introduce the following notion: A ring R is called π∞-regular3 if Mn(R) is π-regular for all n. Proposition 2.6. (i) Let R be a π-regular (π∞-regular) ring and e ∈ E(R). Then eRe is π-regular (π∞-regular). (ii) π∞-regularity is preserved under Morita equivalence. Proof. (i) Assume R is π-regular, let e ∈ R and let a = eae ∈ eRe. By definition, there is b ∈ R and n ∈ N such that an = an+1b. Multiplying by e on the right we get an = an+1ebe, hence an(eRe) = an+1(eRe). Assume R is π∞-regular and let e ∈ R. Let I denote identity matrix in Mn(R). Then (eI)Mn(R)(eI) = Mn(eRe). By the previous argument, the left-hand side is π-regular, hence we are through. (ii) We only need to check that Mn(R) is π∞-regular for all n ∈ N, which is obvious from the definition, and that eRe is π∞-regular, which follows from (i). (cid:3) Proposition 2.7. Let R be a ring and let N denote its prime radical (i.e. the intersection of all prime ideals). Then R is π-regular (π∞-regular) if and only if R/N is. Proof. See [25, §2.7]. (The argument is easily generalized to π∞-regular rings.) (cid:3) Remark 2.8. Any PI semilocal ring with nil Jacobson radical is π∞-regular (see [23, Apx.]). However, there are semilocal rings with nil Jacobson radical that are not π-regular, see [26]. Remark 2.9. We have the following implications: (2.1) =⇒ π∞-regular =⇒ π-regular right artinian =⇒ semiprimary =⇒ left/right perfect However, all these notions coincide for right noetherian rings. Indeed, assume R is π-regular and right noetherian and let J = Jac(R). Then J is nil (see Lemma 4.4(i)), hence Theorem 2.5 implies J n = 0 for some n ∈ N. By Lemma 4.4(ii) below, R is semiperfect and in particular R/J is semisimple. As R is right noetherian, the right R/J-modules {J i−1/J i}n i=1 are f.g., hence their length as right R-modules is finite. It follows that RR has a finite length, so R is right artinian. Throughout, we will use implicitly the next lemma. Notice that it implies that being semiprimary (resp. right perfect, semiperfect, semilocal) passes to quotients. Lemma 2.10. Let R be a semilocal ring. Then any surjective ring homomorphism ϕ : R → S satisfies ϕ(Jac(R)) = Jac(S). Proof. ϕ(Jac(R)) is an ideal of ϕ(R) = S and 1 + ϕ(Jac(R)) = ϕ(1 + Jac(R)) ⊆ ϕ(R×) ⊆ S×, hence ϕ(Jac(R)) ⊆ Jac(S). On the other hand, S/ϕ(Jac(R)) is a quotient of R/ Jac(R) which is semisimple. Therefore, S/ϕ(Jac(R)) is semisimple, implying ϕ(Jac(R)) ⊇ Jac(S). (cid:3) 2 This notion of π-regularity is sometimes called strong π-regularity. 3 This property is sometimes called completely π-regular. 6 URIYA A. FIRST 3. Semi-Invariant Subrings This section presents the basic properties of semi-invariant subrings. We begin by showing that for any ring the semi-invariant subrings are precisely the semi- centralizer subrings. Proposition 3.1. Let R0 ⊆ R be rings. The following are equivalent: (a) There is a ring S ⊇ R and a set Σ ⊆ End(S) such that R0 = RΣ. (b) There is a ring S ⊇ R and a subset X ⊆ S such that R0 = CentR(X). (c) There is a ring S ⊇ R and σ ∈ Aut(S) such that σ2 = id and R0 = R{σ}. (d) There is a ring S ⊇ R and an inner automorphism σ ∈ Aut(S) such that (e) There are rings {Si}i∈I and ring homomorphisms ψ(1) : R → Si such σ2 = id and R0 = R{σ}. , ψ(2) i i (r), ∀i ∈ I}. that R0 = {r ∈ R : ψ(1) i (r) = ψ(2) i Note that condition (e) implies that the family of semi-invariant subrings is closed under intersection. , ψ(2) Proof. We prove (a)=⇒(e)=⇒(c)=⇒(d)=⇒(b)=⇒(a). σ = σ and ψ(2) (a)=⇒(e): Take I = Σ and define Sσ = S, ψ(1) (e)=⇒(c): Let {Si, ψ(1) assume that there is i0 ∈ I such that Si0 = R and ψ(1) S =Q(i,j)∈I×{1,2} Sij where Sij = Si and let Ψ : R → S be given by (r)(cid:17)(i,j)∈I×{1,2} ∈ S . Ψ(r) =(cid:16)ψ(j) i0 = ψ(2) i i i }i∈I be given. Without loss of generality we may i0 = idR. Define σ = idR. P∞ The existence of i0 above implies Ψ is injective. Let σ ∈ Aut(S) be the automor- phism exchanging the (i, 1) and (i, 2) components of S for all i ∈ I. Then one easily checks that σ2 = id and Ψ(R){σ} = Ψ(R0). We finish by identifying R with Ψ(R). (c)=⇒(d): Let S, σ be given and let S′ = S[x; σ] denote the ring of σ-twisted polynomials with (left) coefficients in S. Observe that (x2 − 1) ∈ Cent(S′) (since σ2 = id), hence S′′ := S′/(cid:10)x2 − 1(cid:11) is a free left S-module with basis {1, x} (where a is the image of a ∈ S′ in S′′). Let τ ∈ Aut(S′′) be conjugation by x. Then τ 2 = id and R{τ } = {r ∈ R : xr = rx} = {r ∈ R : σ(r) = r} = R{σ} = R0. (d)=⇒(b): This is a clear. (b)=⇒(a): Let S, X be given. Let S′ = S((t)) be the ring of formal Laurent series n=k antn (k ∈ Z) with coefficients in S. The elements of S′ commuting with X are precisely the elements that commute with t−1 + X (as t−1 is central in S′). However, it is easily seen that all elements in t−1 + X are invertible. For all x ∈ X, let σx ∈ End(S′) be the inner automorphism of S′ given by conjugation with t−1 +x and let Σ = {σx x ∈ X}. Then RΣ = CentR(t−1 + X) = CentR(X) = R0. Corollary 3.2. Let R, W be rings and let ϕ : R → W be a ring homomorphism. Assume W0 ⊆ W is a semi-invariant subring of W . Then ϕ−1(W0) is a semi- invariant subring of R. Proof. By Proposition 3.1(e), there are rings {Si}i∈I and ring homomorphisms ψ(1) (r) ∀i ∈ I}. Define i = ψ(n) ϕ(n) ◦ ϕ and note that ϕ−1(W0) = {r ∈ R : ϕ(r) ∈ W0} = {r ∈ R : i ϕ(r) ∀i ∈ I} = {r ∈ R : ϕ(1) i ϕ(r) = ψ(2) ψ(1) The equivalent conditions of Proposition 3.1 require the existence of some am- : W → Si such that W0 = {r ∈ R : ψ(1) (r) = ϕ(2) (r) ∀i ∈ I}. (r) = ψ(2) , ψ(2) (cid:3) (cid:3) i i i i i i i bient ring. This leads to the following question: Question 1. Is there an intrinsic definition of semi-invariant subrings? SEMI-INVARIANT SUBRINGS 7 Informally, we ask for a definition that would make it easy to show that a given subring is not semi-invariant. The next proposition is useful for producing examples of semi-invariant subrings. Proposition 3.3. Let R ⊆ S be rings and let K be a central subfield of S. Then R ∩ K is a semi-invariant subring of R. Proof. Let S′ = S⊗K S and define ϕ1, ϕ2 : S → S′ by ϕ1(s) = s⊗1 and ϕ2(s) = 1⊗ s. As K is a central subfield, it is easy to check that {s ∈ S : ϕ1(s) = ϕ2(s)} = K, hence {s ∈ R : ϕ1(s) = ϕ2(s)} = R ∩ K. We are done by Proposition 3.1(e). Corollary 3.4. Let K be a field. Then the semi-invariant subrings of K are precisely its subfields. Proof. Any semi-invariant subring R ⊆ K satisfies R× = R ∩ K × = R \ {0}, hence it is a field. The converse follows from the last proposition. (cid:3) (cid:3) Remark 3.5. If K/L is an algebraic field extension, then L is an invariant subring of K if and only if K/L is Galois. We finish this section by introducing two cases where semi-invariant subrings naturally appear. Proposition 3.6. Let R ⊆ S be rings and let M be a right S-module. Then End(MS) is a semi-invariant subring of End(MR).4 Proof. There is a ring homomorphism ϕ : Sop → End(MZ) given by ϕ(sop)(m) = ms for all m ∈ M . It is straightforward to check that End(MS) = CentEnd(MZ)(im ϕ). As End(MS) ⊆ End(MR), it follows that End(MS) = CentEnd(MR)(im ϕ), hence End(MS) is a semi-centralizer subring of End(MR). (cid:3) Proposition 3.7. Let A be an abelian category and let A −→ C → 0 be an exact sequence in A such that for any c ∈ End(C) there are b ∈ End(B) and a ∈ End(A) with cg = gb and bf = f a (e.g.: if both A and B are projective, or if B is projective and f is injective). Then End(C) is isomorphic to a quotient of: −→ B f g (i) A semi-invariant subring of End(A) × End(B). (ii) An invariant and a centralizer subring of End(A ⊕ B), provided f is injec- tive. Proof. Let B0 = im f = ker g. Define R to be the subring of End(B) consisting of maps b ∈ End(B) for which there is a ∈ End(A) with bf = f a. Then for all b ∈ R, b(B0) = b(im f ) = im(f a) ⊆ im f = B0. Therefore, there is unique c ∈ End(C) such that cg = gb. The map sending b to c is easily seen to be a ring homomorphism from R to End(C) and the assumptions imply it is onto. Therefore, End(C) is a quotient of R. Let S = End(A ⊕ B). We represent elements of S as matrices [ x y diagonal matrices in S (i.e. End(A) × End(B)) and let W = CentS((cid:2) 0 f ring homomorphism ϕ : D → End(B) by ϕ((cid:2) x 0 0 y(cid:3)) = y. Then ϕ(CentD((cid:2) 0 f z w ] with x ∈ End(A), y ∈ Hom(B, A), z ∈ Hom(A, B), w ∈ End(B). Let D denote the for a ∈ End(A) and b ∈ End(B), f a = bf if and only if [ a 0 0 b ] ∈ W . Define a ϕ(D ∩ W ) = R. It follows that End(C) is a quotient of R, which is a quotient of D ∩ W , which is a semi-centralizer subring of D = End(A) × End(B). This settles (i). To see (ii), notice that if f is injective, then W consists of upper-triangular matrices, hence ϕ can be extended to W , which is a centralizer and an invariant (cid:3) 0 0(cid:3)). Then 0 0(cid:3))) = subring of S since W = CentS((cid:2) 1 f 0 1(cid:3)) and(cid:2) 1 f 0 1(cid:3) ∈ S×. 4 This proposition is a refinement of [13, Pr. 2.7], which asserts that under the same assumptions End(MS ) is a rationally closed subring of End(MR). 8 URIYA A. FIRST 4. Properties Inherited by Semi-Invariant Subrings In this section we prove that being semiprimary (right perfect, semiperfect and π∞-regular, semiperfect and π-regular) passes to semi-invariant subrings. We also present a supplementary result for algebras. Our first step is introducing an equivalent condition for π-regularity of elements of a ring. Lemma 4.1. Let R be a ring and let a ∈ R be a π-regular element. Define: k=1 akR , k=1 Rak, A =T∞ A′ =T∞ k=1 annr ak, k=1 annℓ ak. B =S∞ B′ =S∞ Then there is e ∈ E(R) such that A = eR, B = f R, A′ = Re and B′ = Rf where f := 1 − e. In particular, RR = A ⊕ B and RR = A′ ⊕ B′. Proof. Let n ∈ N be such that anR = akR and Ran = Rak for all k ≥ n. Notice that this implies annr an = annr ak and annℓ an = annr ak for all k ≥ n. We begin by showing RR = A ⊕ B. That RR = A′ ⊕ B′ follows by symmetry. The argument is similar to the proof of Fitting's Lemma (see [25, §2.9]): Let r ∈ R. Then anr ∈ anR = a2nR, hence there is s ∈ R with anr = a2ns. Observe that an(r − ans) = 0 and ans ∈ anR, so r = ans + (r − ans) ∈ A + B. Now suppose r ∈ A ∩ B. Then r = ans for some s ∈ R. However, r ∈ B = annr an implies s ∈ annr a2n = annr an, so r = ans = 0. Since RR = A ⊕ B there is e ∈ R such that e ∈ A and f := 1 − e ∈ B. It is well known that in this case e2 = e, A = eR and B = f R. This implies B′ = annℓ an = annℓ anR = annℓ eR = Rf and A′ = Ran ⊆ annℓ annr Ran = annℓ annr an = annℓ f R = Re. As RR = A′ ⊕ B′ = Re ⊕ Rf , we must have have A′ = Re. Proposition 4.2. Let R be a ring and a ∈ R. Then a is π-regular ⇐⇒ there is e ∈ E(R) such that (cid:3) (A) a = eae + f af where f := 1 − e. (B) eae is invertible in eRe. (C) f af is nilpotent. In this case, the idempotent e is uniquely determined by a. Proof. Assume a is π-regular and let e, f, A, B, A′, B′, n be as in Lemma 4.1. Then ae ∈ aeR = aA = a(anR) = an+1R = A = eR and af ∈ aB = a annr an ⊆ annr an = B = f R. Therefore, ae = eae and af = f af , hence a = ae + af = eae + f af . This implies ak = (eae)k + (f af )k for all k ∈ N. As an ∈ eR, we have an = ean, hence (eae)n + (f af )n = an = e(eae)n + e(f af )n = (eae)n which implies (f af )n = 0. In particular, for all k ≥ n, ak = (eae)k + (f af )k = (eae)k. Since e ∈ anR, there is x ∈ R such that e = anx = (eae)nx. Multiplying by e on the right yields e = (eae)((eae)n−1xe), hence eae is right invertible in eRe. By symmetry, eae is left also left invertible in eRe, hence we conclude that e satisfies (A) -- (C). Now assume there is e ∈ E(R) satisfying (A) -- (C) and let b be the inverse of a in eRe. Then ak = (eae)k + (f af )k for all k ∈ N. Condition (C) now implies there is n ∈ N such that ak = (eae)k for all k ≥ n. Therefore, for all k ≥ n, an = (eae)n = (eae)kbk−n = akbk−n ∈ akR implying anR = akR. By symmetry, Ran = Rak for all k ≥ n, so a is π-regular. Finally, assume that e, e′ ∈ E(R) satisfy conditions (A) -- (C) and let f = 1 − e, f ′ = 1 − e′. By the previous paragraph a is π-regular, hence Lemma 4.1 implies k=1 annr akR. Let b be the inverse of eae in eRe and let n ∈ N be such that (f af )n = 0. Then e = (eae)kbk = akbk ∈ akR for all k ≥ n, hence e ∈ A, and anf = (eae)nf = 0, hence f ∈ B. Similarly, e′ ∈ A R = A ⊕ B where A =T∞ k=1 akR and B =S∞ SEMI-INVARIANT SUBRINGS 9 and f ′ ∈ B. It follows that e, e′ ∈ A and f, f ′ ∈ B. Since 1 = e + f = e′ + f ′ and R = A ⊕ B, we must have e = e′. (cid:3) Let R, a be as in Proposition 4.2. Henceforth, we call the unique idempotent e satisfying conditions (A) -- (C) the associated idempotent of a (in R). Corollary 4.3. (i) Let R be a ring, R0 ⊆ R a semi-invariant subring and let a ∈ R0 be π-regular in R. Then a is π-regular in R0. (ii) A semi-invariant subring of a π-regular (π∞-regular) ring is π-regular (π∞- regular). Proof. (i) Let S ⊇ R and Σ ⊆ End(S) be such that R0 = RΣ and let a ∈ R0 be π-regular in R. Let e be the associated idempotent of a in R. Then e is clearly the associated idempotent of a in S (hence a is π-regular in S). However, it is straightforward to check that σ(e) satisfies conditions (A) -- (C) (in S) for all σ ∈ Σ (since σ(a) = a), so the uniqueness of e forces e ∈ SΣ ∩ R = RΣ = R0. Therefore, a is π-regular in R0. (ii) The π-regular case follows from (i). The π∞-regular case follows once noting that if R0 is a semi-invariant subring of R, then Mn(R0) is a semi-invariant subring of Mn(R) for all n ∈ N. (cid:3) Lemma 4.4. Let R be a π-regular ring. Then: (i) Jac(R) is nil. (ii) R is semiperfect ⇐⇒ R does not contain an infinite set of orthogonal idempotents. Proof. For a ∈ R, let ea denote the associated idempotent of a and let fa = 1− ea. (i) Let a ∈ Jac(R) and let b be the inverse of eaaea in eaRea. Then ea = b(eaaea) ∈ Jac(R), hence ea = 0, implying a = faafa is nilpotent. (ii) That R is semiperfect clearly implies R does not contain an infinite set of orthogonal idempotents, so assume the converse. Let a ∈ R. Observe that if ea = 0 then a is nilpotent and if ea = 1 then a is invertible. Therefore, if ea ∈ {0, 1} for all a ∈ R, then R is local and in particular, semiperfect. Assume there is a ∈ R with e := ea /∈ {0, 1}. We now apply an inductive argument to deduce that eRe and (1 − e)R(1 − e) are semiperfect, thus proving R is semiperfect (by Proposition 2.4). The induction process must stop because otherwise there is a sequence of idempotents {ek}∞ k=0 ⊆ R such that ek ∈ ek−1Rek−1 and ek /∈ {0, ek−1}. This implies {ek−1 − ek}∞ k=1 is an infinite set of non-zero orthogonal idempotents, which cannot exist by our assumptions. (cid:3) Lemma 4.5. Let R0 ⊆ R be rings. If R is semiperfect and both R0 and R are π-regular, then R0 is semiperfect and Jac(R0)n ⊆ Jac(R) for some n ∈ N. If in addition R is semiprimary (right perfect), then so is R0. Proof. By Lemma 4.4(ii), R does not contain an infinite set of orthogonal idempo- tents. Therefore, this also applies to R0, so the same lemma implies R0 is semiper- fect. Let ϕ denote the standard projection from R to R/ Jac(R). By Lemma 4.4(i), ϕ(Jac(R0)) is nil. Therefore, by Theorem 2.5 (applied to ϕ(R), which is semisim- ple), ϕ(Jac(R0)) is nilpotent, hence there is n ∈ N such that Jac(R0)n ⊆ Jac(R). If moreover R is semiprimary (right perfect), then Jac(R) is nilpotent (right T- nilpotent). The inclusion Jac(R0)n ⊆ Jac(R) then implies Jac(R0) is nilpotent (right T-nilpotent), so R0 is semiprimary (right perfect). (cid:3) Theorem 4.6. Let R be a ring and let R0 be a semi-invariant subring of R. If R is semiprimary (resp. right perfect, semiperfect and π∞-regular, semiperfect and π-regular), then so is R0. In addition, there is n ∈ N such that Jac(R0)n ⊆ Jac(R). 10 URIYA A. FIRST Proof. Recall that being right perfect implies being π-regular by Proposition 2.1. Given that, the theorem follows from Corollary 4.3 and Lemma 4.5. (cid:3) Corollary 4.7. Let R ⊆ S be rings and let M be a right S-module. If End(MR) is semiprimary (resp. right perfect, semiperfect and π∞-regular, semiperfect and π-regular), then so is End(MS) and there exists n ∈ N such that Jac(End(MS))n ⊆ Jac(End(MR)). Proof. This follows from the last theorem and Proposition 3.6. (cid:3) Remark 4.8. Camps and Dicks proved in [9] that a rationally closed subring of a semilocal ring is semilocal, thus implying the semilocal analogues of Theorem 4.6 and Corollary 4.7, excluding the part regarding the Jacobson radical (which indeed fails in this case; see Example 6.7). In fact, the semilocal analogue of Corollary 4.7 was noticed in [13, Pr. 2.7]. However, we cannot use this analogue with the Krull-Schmidt Theorem (as we do in Section 7 with our results) because modules with semilocal endomorphism ring need not have a Krull-Schmidt decomposition, as shown in [14] and [3]. Nevertheless, as there are plenty of weaker Krull-Schmidt theorems for modules that do not require End(MR) to be semiperfect (mainly due to Facchini et al.; e.g. [3], [12]), it might be that if M, R, S are as in Corollary 4.7 and End(MR) is merely semiperfect, then M has a Krull-Schmidt decomposition over S (despite the fact End(MS) need not be semiperfect). To the author's best knowledge, this topic is still open. We finish this section with a supplementary result for algebras. Proposition 4.9. Let R ⊆ S be rings and Σ ⊆ End(S). Assume there is a division ring D ⊆ R such that σ(D) ⊆ D for all σ ∈ Σ. Then dim DΣRΣ ≤ dim DR. Proof. Consider the left D-vector space V = DRΣ. Let {vi}i∈I ⊆ RΣ be a left D-basis for V . We claim that {vi}i∈I is a left DΣ-basis for RΣ. Indeed, let v ∈ RΣ. Then there are unique {di}i∈I ⊆ D (almost all 0) such that v = Pi divi. However, for all σ ∈ Σ, v = σ(v) =Pi σ(di)vi, so σ(di) = di for all i ∈ I, hence v ∈Pi∈I DΣvi. Therefore, dim DΣ RΣ = dim DV ≤ dim DR. (cid:3) Remark 4.10. An invariant subring of a f.d. algebra need not be left nor right artinian, even when invariants are taken w.r.t. to the action of a finite cyclic group. This was demonstrated by Bjork in [5, §2]. In particular, the assumption σ(D) ⊆ D for all σ ∈ Σ in the last proposition is essential. However, Bjork also proved that if Σ is a finite group acting on a f.d. algebra over a perfect field, then the invariant subring (w.r.t. Σ) is artinian; see [5, Th. 2.4]. For a detailed discussion about when a subring of an artinian ring is artinian, see [5] and [6]. 5. Topologically Semi-Invariant Subrings In this section we specialize the notions of semi-invariance and and π-regularity to certain topological rings. As a result we obtain a topological analogue of The- orem 4.6 (Theorem 5.10), which is used to prove that topologically semi-invariant subrings of semiperfect pro-semiprimary rings are semiperfect and pro-semiprimary (Theorem 5.15). Note that once restricted to discrete topological rings, some of the results of this section reduce to results from the previous sections. However, the latter are not superfluous since we will rely on them. For a general reference about topological rings, see [31]. Definition 5.1. A topological ring R is called linearly topologized (abbreviated: LT) if it admits a local basis (i.e. a basis of neighborhoods of 0) consisting of two- sided ideals. In this case the topology on R is called linear. SEMI-INVARIANT SUBRINGS 11 Let us set some general notation: For any topological ring R, let IR denote its set of open ideals. Then R is LT if and only if IR is a local basis. We use Homc (Endc) to denote continuous homomorphisms (endomorphisms). The category of Hausdorff linearly topologized rings will be denoted by LT2, where HomLT2 (A, B) = Homc(A, B) for all A, B ∈ LT2.5 A subring of a topological ring is assumed to have the induced topology. In particular, if R ∈ LT2 then so is any subring of R. The following facts will be used freely throughout the paper. For proofs, see [31, §3]. (1) Let (G, +) be an abelian topological group and let B be a local basis of G. Then for any subset X ⊆ G, X =TU∈B(X + U ). (2) Under the pervious assumptions, G is Hausdorff ⇐⇒ {0} = TU∈B U = {0}. (3) Given a ring R and a filter base of ideals B, there exists a unique ring topology on R with local basis B. This topology makes R into an LT ring. Example 5.2. (i) Any ring assigned with the discrete topology is LT. assigning it the unique ring topology with local basis {Mn(I) I ∈ IR}. (ii) Zp (with the p-adic topology) is LT but Qp is not. (iii) Let R be an LT ring and let n ∈ N. We make Mn(R) into an LT ring by (iv) If R is LT and e ∈ E(R), then eRe is LT w.r.t. the induced topology. (vi) Let R be an inverse limit of LT rings {Ri}i∈I , with the topology induced (vii) If R is LT and J E R, then R/J with the quotient topology is LT. Indeed, {I/J J ⊆ I ∈ IR} is a local basis for that topology. The ring R/J is Hausdorff if and only if J is closed, and discrete if and only if J is open. (v) Let {Ri}i∈I be LT rings. ThenQi∈I Ri is LT w.r.t. the product topology. from the product topology onQi∈I Ri.6 Then by (v) R is LT. The last example implies that LT2 is closed under products, inverse limits and forming matrix rings (with the appropriate topologies). We will say that a property Q of LT rings is preserved under Morita equivalence if whenever R ∈ LT2 has Q, then so does Mn(R) and eRe for e ∈ E(R) s.t. eR is a progenerator.7 Definition 5.3. Let R ∈ LT2. A subring R0 ⊆ R is called a topologically semi- invariant (abbrev.: T-semi-invariant) subring if there is R ⊆ S ∈ LT2 and a set Σ ⊆ Endc(S) such that R0 = RΣ. The subring R0 is called a topologically semi- centralizer (abbrev.: T-semi-centralizer) subring if there is R ⊆ S ∈ LT2 and a set X ⊆ S such that R0 = CentR(X). A T-semi-invariant subring is always closed. In addition, there is an analogue of Proposition 3.1 for T-semi-invariant rings. Proposition 5.4. Let R0 be a subring of R ∈ LT2. The following are equivalent: (a) There is R ⊆ S ∈ LT2 and a set Σ ⊆ Endc(S) such that R0 = RΣ. (b) There is R ⊆ S ∈ LT2 and a subset X ⊆ S such that R0 = CentR(X). (c) There is R ⊆ S ∈ LT2 and σ ∈ Autc(S) with σ2 = id and R0 = R{σ}. (d) There is R ⊆ S ∈ LT2 and an inner automorphism σ ∈ Autc(S) such that (e) There are LT Hausdorff rings {Si}i∈I and continuous ring homomorphisms σ2 = id and R0 = R{σ}. ψ(1) , ψ(2) i i : R → Si such that R0 = {r ∈ R : ψ(1) i (r) = ψ(2) i (r) ∀i ∈ I}. 5 The subscript "2" in LT2 stands for the second separation axiom T2 (i.e. being Hausdorff). 6 With this topology R is indeed the inverse limit of {Ri}i∈I in category of topological rings, i.e. it admits the required universal property. 7 Caution: There is a notion of Morita equivalence for (right) LT rings, but we will not use it in this paper; see [15] and related articles. 12 URIYA A. FIRST Proof. This is essentially the proof of Proposition 3.1, but we need to endow the rings constructed throughout the proof with topologies making them into LT Ha- sudorff rings that contain R as a topological ring. This is briefly done below; the details are left to the reader. (e)=⇒(c): Endow S =Q(i,j)∈I×{1,2} Sij with the product topology. coefficients in J, and give S′′ = S′/(cid:10)x2 − 1(cid:11) the quotient topology. (c)=⇒(d): Observe that B = {I ∩ σ(I) I ∈ IS} is a local basis of S and σ(J) = J for all J ∈ B. Assign S′ = S[x; σ] the unique ring topology with local basis {J[x; σ] J ∈ B}, where J[x; σ] denotes the set of polynomials with (left) (b)=⇒(a): Give S((t)) the unique ring topology with local basis {I((t)) I ∈ IS}, where I((t)) denotes the set of polynomials with coefficients in I. (cid:3) We now generalize the notion of π-regularity for topological rings. Our definition is inspired by Proposition 4.2. Definition 5.5. Let R ∈ LT2 and a ∈ R. The element a is called quasi-π-regular in R if there is an idempotent e ∈ E(R) such that: (A) a = eae + f af where f := 1 − e. (B) eae is invertible in eRe. (C′) (f af )n n→∞−−−−→ 0 (w.r.t. the topology on R). Call R quasi-π-regular if all its elements are quasi-π-regular. so e = e′. Since we only consider LT rings, condition (C′) means that for any I ∈ IR there is n ∈ N such that (f af )n ∈ I. This implies that quasi-π-regularity coincide with π-regularity for discrete topological rings (take I = {0}) and that if a is quasi- π-regular in R then a + I is π-regular in R/I for all I ∈ IR. In particular, if R is quasi-π-regular then R/I is π-regular. We will call the idempotent e satisfying conditions (A),(B) and (C′) the associated idempotent of a. The following lemma shows that it is unique. Lemma 5.6. Let R ∈ LT2 and a ∈ R a quasi-π-regular element. Then the idem- potent e satisfying conditions (A),(B) and (C ′) is uniquely determined by a. Proof. Assume both e and e′ satisfy conditions (A), (B), (C′) and let I ∈ IR. Then e + I and e′ + I are associated idempotents of a + I in R/I, hence e + I = e′ + I, or Remark 5.7. (i) In the assumptions of the previous lemma it is also possible to equivalently e−e′ ∈ I. It follows that e−e′ ∈TI∈IR I = {0} (since R is Hausdorff), show that eR =T∞ (ii) If we do not restrict to LT Hausdorff rings, the associated idempotent need not be unique. For example, in Qp both 0 and 1 are associated idempotents of p. (It is not known if Lemma 5.6 holds under the assumption that R is right LT, i.e. has a local basis of right ideals). n=1 anR and (1 − e)R = {r ∈ R : anr n→∞−−−−→ 0}. (iii) If one assigns a semiperfect ring R with T∞ n=1 Jac(R)n = {0} the Jac(R)- adic topology, then R becomes a Hausdorff LT ring and for any a ∈ R there is an idempotent e satisfying conditions (B) and (C′) (but such e need not be unique even when R is simple). However, condition (A) might be impossible to satisfy for some a. Indeed, the ring R constructed in example 6.1 below, which is isomorphic to M4(Zh3i), is a semiperfect ring having no ring topology making it into a quasi- π-regular Hausdorff LT ring. As Zh3i is quasi-π-regular w.r.t. the 3-adic topology (since it is local), it follows that quasi-π-regularity is not preserved under Marita equivalence. (This also follows from the comment before Proposition 2.6.) (cid:3) It light of the last remark, it is convenient to call an LT Hausdorff ring R quasi- π∞-regular if Mn(R) is quasi-π-regular for all n. This property is preserved under SEMI-INVARIANT SUBRINGS 13 Morita equivalence and turns out to be related with Henselianity (see Section 8). However, to avoid cumbersome notation, we will not mention it in this section. All statements henceforth can be easily seen to hold once replacing (quasi-)π-regular with (quasi-)π∞-regular. Corollary 5.8. (i) Let R ∈ LT2, let R0 be a T-semi-invariant subring of R and let a ∈ R0 be quasi-π-regular in R. Then a is quasi-π-reuglar in R0. (ii) A T-semi-invariant subring of a quasi-π-regular ring is quasi-π-regular. Proof. This is similar to the proof of Corollary 4.3. (cid:3) Lemma 5.9. Let R ∈ LT2 be quasi-π-regular. Then: (i) For all a ∈ Jac(R), an n→∞−−−−→ 0. (That is, Jac(R) is "topologically nil"). (ii) R is semiperfect ⇐⇒ R does not contain an infinite set of orthogonal idempotents. Proof. (i) Let I ∈ IR. Then a + I ∈ (Jac(R) + I)/I ⊆ Jac(R/I), so by Lemma 4.4(i) applied to R/I (which is π-regular), there is n ∈ N such that an ∈ I. (ii) We only show the non-trivial implication. For a ∈ R, let ea denote the associated idempotent of a. Note that ea = 1 implies a ∈ R× and ea = 0 implies an n→∞−−−−→ 0. Assume ea ∈ {0, 1} for all a ∈ R. We claim R is local. This is clear if R = {0}. Otherwise, let a ∈ R and assume by contradiction that ea = e1−a = 0. Let R 6= I ∈ IR (here we need R 6= {0}). Then there is n ∈ N such that an, (1−a)n ∈ I, implying (1− an)n = (1− a)n(1 + a +··· + an−1)n ∈ I. We can write 1 = (1− an)n + anh(a) for some h(x) ∈ Z[x], thus getting 1 ∈ I, in contradiction to the assumption I 6= R. Therefore, one of ea, e1−a is 1, hence one of a, 1 − a is invertible. Now assume there is a ∈ R with e := ea /∈ {0, 1}. Then we can induct on eRe and (1 − e)R(1 − e) as in the proof of Lemma 4.4(ii). However, we need to verify that eRe is quasi-π-regular (w.r.t. the induced topology). Let b ∈ eRe. It enough to show eb ∈ eRe, i.e. eb = eebe. As R ∈ LT2, this is equivalent to eb + I = eebe + I for all I ∈ IR. Indeed, since R/I is π-regular, so is e(R/I)e (by Proposition 2.6(i)), hence b + I has an associated idempotent ε ∈ e(R/I)e. However, it easy to see that ε is also the associated idempotent of b + I in R/I, so necessarily ε = eb + I. As ε = (e + I)ε(e + I), it follows that eb + I = eebe + I. (cid:3) We can now state and prove a T-semi-invariant analogue of Theorem 4.6. Theorem 5.10. Let R0 be a T-semi-invariant subring of a semiperfect and quasi- π-regular ring R ∈ LT2. Then R0 is semiperfect and quasi-π-regular and there is n ∈ N such that Jac(R0)n ⊆ Jac(R). Proof. That R0 is quasi-π-regular and semiperfect follows form Corollary 5.8 and Lemma 5.9(ii). Now let I ∈ IR. Then both R/I and (R0 + I)/I are semiperfect and π-regular (since (R0 + I)/I ∼= R0/(R0 ∩ I) and R0 ∩ I is open in R0). There- fore, by Lemma 4.5, there is nI ∈ N such that Jac((R0 + I)/I)nI ⊆ Jac(R/I)nI . As Jac(R/I) = (Jac(R) + I)/I, this implies Jac(R0)nI ⊆ Jac(R) + I. However, (R/I)/(Jac(R/I)) ∼= R/(Jac(R) + I) is a quotient of R/ Jac(R) which is semisim- ple, hence the index of nilpotence of any of its subsets is bounded (when finite) by length(R/ Jac(R)).8 Therefore, there is n ∈ N such that for all I ∈ IR, Jac(R0)n ⊆ Jac(R) + I or equivalently, Jac(R0)n ⊆TI∈IR(Jac(R) + I) = Jac(R). Thus, we are done by the following lemma. (cid:3) 8 Actually, the index of nilpotence is bounded in any right noetherian ring R. Indeed, the prime radical of R, denoted N , is nilpotent and R/N is a semiprime Goldie ring. Therefore, by Goldie's Theorem, R/N embeds in a semisimple ring and thus has a bounded index of nilpotence. 14 URIYA A. FIRST Lemma 5.11. Let R ∈ LT2 be quasi-π-regular. Then R× and Jac(R) are closed. Proof. Let a ∈ R× and let e be its associated idempotent. Then for any I ∈ IR there is aI ∈ R× such that a − aI ∈ I. Clearly e + I is the associated idempotent of a + I = aI + I in R/I. However, aI + I ∈ (R/I)× and thus 1 + I is its associated idempotent. It follows that e + I = 1 + I for all I ∈ IR, hence e = 1 and a ∈ R×. Now assume a ∈ Jac(R). It is enough to show that for all b ∈ R, 1 + ab ∈ R×. Let I ∈ IR and let aI ∈ Jac(R) be such that a − aI ∈ I. Then 1 + aI b ∈ R× and (1 + ab) − (1 + aI b) ∈ I. Therefore, 1 + ab ∈TI∈IR(R× + I) = R× = R×. Remark 5.12. (i) The assumption that R is quasi-π-regular in the last lemma is essential; see Example 9.2 (take n = 0). In addition, Jac(R)2 need not be closed even when R is quasi-π-regular; see Example 9.11. (cid:3) (ii) If R is quasi-π-regular and semiperfect, then R× and Jac(R) are also open. Indeed, by the previous lemma Jac(R) = Jac(R) = TI∈IR(Jac(R) + I), hence Jac(R) is an intersection of open ideals. Since R/ Jac(R) is artinian, Jac(R) is the intersection of finitely many such ideals, thus open. The set R× is open since it is a union of cosets of Jac(R). In order to apply theorem 5.10 to pro-semiprimary rings, we need to recall some facts about complete topological rings. While the exact definition (see [31, §7-8]) is of little use to us, we will need the following results. Let R ∈ LT2, then: (1) R is complete if and only if R is isomorphic to an inverse limit of an in- verse system of discrete topological rings {Ri}i∈I. In this case, if ϕi is the standard map from R to Ri, then {ker ϕi i ∈ I} is a local basis of R. (2) If R is complete and B is a local basis consisting of ideals, then R ∼= lim←−{R/I}I∈B. (Note that R/I is discrete for all I ∈ B.) We will also use the fact that a closed subring of complete ring is complete. (This can be verified directly for rings in LT2 using the previous facts.) We now specialize the definition of pro-semiprimary rings given in Section 1 to topological rings. For a ring property P, a topological ring R will be called pro-P if R is isomorphic as a topological ring to the inverse limit of an inverse system of discrete rings satisfying P. If in addition the standard map from R to each of these rings is onto9, then R will be called strictly pro-P. Clearly any pro-P ring is complete and lies in LT2. An LT ring R is strictly pro-P if and only if it is complete and admits a local basis of ideals B such that R/I has P for all I ∈ B. Notice that if P is preserved under Morita equivalence, then so does being pro-P and being strictly pro-P (because the isomorphisms of Lemma 2.3 are also topological isomorphisms). Remark 5.13. Any inverse limit of (non-topological) rings satisfying P can be endowed with a linear ring topology making it into a pro-P ring, but this topol- ogy usually depends on the inverse system used to construct the ring. However, when P = semiprimary and the ring is right noetherian, the topology is uniquely determined and always coincide with the Jacobson topology! See Section 9. Once recalling Remark 2.9, the following lemma implies that pro-semiprimary rings are quasi-π-regular. Lemma 5.14. Let {Ri, fij} be an I-indexed inverse system of π-regular rings and let R = lim←−{Ri}i∈I . Then R is quasi-π-regular. 9 This is not trivial since the maps in the inverse system are not assumed to be onto. SEMI-INVARIANT SUBRINGS 15 Proof. We identify R with the set of compatible I-tuples in Qi∈I Ri (i.e. tuples (xi)i∈I satisfying fij(xj) = xi for all i ≤ j in I). Let a = (ai)i∈I ∈ R and let ei ∈ E(Ri) be the associated idempotent of ai in Ri. The uniqueness of ei implies that e = (ei)i∈I is compatible and hence lie in R. We claim that e is the associated idempotent of a in R. Conditions (A) and (C′) are straightforward, so we only check (B): Let bi be the inverse of eiaiei in eiRei. The for all i ≤ j in I, fij (bj) is also an inverse of eiaiei in eiRei, hence fij(bj) = bi. Therefore, b := (bi)i∈I is compatible and lie in R. Clearly b = ebe and b(eae) = (eae)b = e (since this holds in each coordinate), so condition (B) is satisfied. We thus conclude that R is quasi-π-regular. (cid:3) The converse of Lemma 5.9 is almost true; if R ∈ LT2 is quasi-π-regular, then ←−{R/I}I∈IR. The following theorem R is dense in a pro-π-regular ring, namely lim implies that T-semi-invariant subrings of semiperfect pro-semiprimary rings are semiperfect and pro-semiprimary (w.r.t. the induced topology). Theorem 5.15. Assume R = lim←− {Ri}i∈I where each Ri is π-regular. Denote by Ji the kernel of the standard map R → Ri and let R0 be a T-semi-invariant subring of R. Then: (i) R0 is quasi-π-regular and R0 = lim (ii) If R does not contain an infinite set of orthogonal idempotents, then R0 is ←−{R0/(Ji ∩ R0)}i∈I . semiperfect and there is n ∈ N such that Jac(R0)n ⊆ Jac(R). (iii) For all i ∈ I, R0/(Ji ∩ R0) is π-regular. If moreover Ri is semiprimary (right perfect, semiperfect), then so is R0/(Ji ∩ R0). In particular, if R is pro-semiprimary (pro-right-perfect, pro-π-regular-and-semiperfect), then so is R0. Proof. By Lemma 5.14, R is quasi-π-regular, so the first assertion of (i) is Corollary 5.8(ii). As for the second assertion, R is complete and R0 is closed in R, hence R0 is complete. Since {R0 ∩ Ji i ∈ I} is a local basis of R0, R0 = lim←−{R0/(Ji ∩ R0)}i∈I . (ii) follows from Lemma 5.9(ii) and Theorem 5.10. As for (iii), R0/(Ji ∩ R0) is π- regular as a quotient of a quasi-π-regular ring with an open ideal. The rest follows from Lemma 4.5 (applied to R0/(Ji ∩ R0) identified as a subring of Ri). (cid:3) Let P ∈ {semiprimary, right-perfect, π-regular-and-semiperfect, π-regular}. Then Theorem 5.15 implies that pro-P rings are strictly pro-P (take R0 = R). In fact, we can prove an even stronger result: Corollary 5.16. In the previous notation, the inverse limit of a small category of pro-P rings is strictly pro-P. Proof. Let C be a small category of pro-P rings and let {Ri}i∈I be the objects of C can be identified with the set of I-tuples (xi)i∈I ∈ Qi∈I Ri C . Then R = lim←− such that f (xj) = xi for all i, j ∈ I and f ∈ HomC (Rj, Ri). Clearly S :=Qi∈I Ri is pro-P. If we can prove that R is a T-semivariant subring of S, then we are through by Theorem 5.15. Indeed, let πi denote the projection from S to Ri. For all i, j ∈ I and f ∈ HomC (Rj, Ri) define ϕ(1) f = f ◦ πj. Then R = {x ∈ S : ϕ(1) f (x)∀f}, hence R is a T-semi-invariant subring of S by : S → Ri by ϕ(1) f = πi, ϕ(2) f , ϕ(2) f f (x) = ϕ(2) Proposition 5.4(e). (cid:3) In some sense, Corollary 5.16 includes Theorem 4.6 and part of Theorem 5.15 because a T-semi-invariant subring can be understood as the inverse limit of a category with two objects. (Indeed, if R ⊆ S ∈ LT2 and Σ is a submonoid of Endc(S), then take Ob(C ) = {R, S} with EndC (S) = Σ, EndC (R) = {idR}, HomC (S, R) = φ and HomC (R, S) = {i} where i : R → S is the inclusion map). 16 URIYA A. FIRST Corollary 5.17. If R is pro-semiprimary, then for any J ∈ IR there is n ∈ N such that Jac(R)n ⊆ J. In particular,T∞ n=1 Jac(R)n = {0}. Proof. Assume R = lim ←−{Ri}i∈I with each Ri semiprimary and let Ji be as in Theorem 5.15. Since {Ji i ∈ I} is local basis, there is i ∈ I such that Ji ⊆ J. By Theorem 5.15(iii), R/Ji is semiprimary, hence there is n ∈ N such that Jac(R/Ji)n = 0. As Jac(R/Ji) ⊇ (Jac(R) + Ji)/Ji, we get Jac(R)n ⊆ Ji ⊆ J. (cid:3) Remark 5.18. We will show in Proposition 8.6 that Henselian rank-1 valuation rings are quasi-π∞-regular. In particular, non-complete such rings (e.g. the Q- algebraic elements in Zp) are examples of non-complete quasi-π∞-regular rings. In addition, the author suspects that the following are also explicit examples of i=0 aiti ∈ Zp[[t]] with ai → 0 endowed with the t-adic topology (such rings are common in rigid geometry). (2) The ring in the comment after Lemma 2.3 w.r.t. its Jacobson topology. non-complete quasi-π∞-regular rings: (1) The ring of power series P∞ 6. Counterexamples This section consists of counterexamples. In particular, we show that: (1) If R is a semiperfect ring and Σ ⊆ End(R), then RΣ need not be semiper- fect even when Σ is a finite group and even when Σ consists of a single automorphism. Similarly, if X ⊆ R is a set, then CentR(X) need not be semiperfect even when X consists of a single element. (2) The semiperfect analogue of Corollary 4.7 is not true in general. (3) A semi-invariant subring of a semiperfect pro-semiprimary ring need not be semiperfect even when closed (in contrast to T-semi-invariant subrings). (4) Rationally closed subrings of a f.d. algebra need not be semiperfect. particular, Theorem 4.6 do not generalize to rationally closed subrings. In (5) No two of the families of semi-invariant, invariant, centralizer and rationally closed subrings coincide in general. We note that (1) is also true if we replace semiperfect with artinian. This was treated at the end of Section 4. We begin with demonstrating (1). Our examples use Azumaya algebras and we refer the reader to [27] for definition and details. Example 6.1. Let S be a discrete valuation ring with maximal ideal πS, residue field k = S/πS and fraction field F , and let A be an Azumaya algebra over S. Recall that this implies A/πA is a central simple k-algebra and Jac(A) = πA. Assume the following holds: (a) D = F ⊗S A is a division ring. (b) A/πA has zero divisors. In addition, assume there is a set X ⊆ A× generating A as an S-algebra (such X always exists). Note that conditions (a) and (b) imply that A is not semiperfect be- cause A contains no non-trivial idempotents while A/πA = A/ Jac(A) does contain such idempotents, hence Jac(A) is not idempotent lifting. Define R = A ⊗S Aop and let Σ = {σx}x∈X where σx is conjugation by 1 ⊗ xop. Then R is an Azumaya S-algebra which is an S-order inside D ⊗F Dop ∼= Mr(F ). This is well known to imply that R ∼= End(PS) for some faithful finite projective S-module P (in fact, P is free since S is local). Therefore, R is Morita equivalent to S, hence semiperfect. On the other hand, RΣ = CentR({1 ⊗ xop x ∈ X}) = CentR(S ⊗ Aop) = A ⊗ S ∼= A, so RΣ is not semiperfect. An explicit choice for S, A, F, D is S = Zh3i (π = 3), F = Q, D = (−1,−1)Q = Q[ i, j ij = −ji, i2 = j2 = −1] and A = S[i, j]. If we take X = {i, j} then SEMI-INVARIANT SUBRINGS 17 Σ will consist of two inner automorphisms which are easily seen to generate an automorphism group isomorphic to (Z/2) × (Z/2). Example 6.2. Let S, π, A, F, D be as in Example 6.1 and assume there is a cyclic Galois extension K/F such that: (c) K/F is totally ramified at π. (d) K ⊗F D splits (i.e. K ⊗F D ∼= Mt(K)). Write Gal(K/F ) = hσi. Let T denote the integral closure of S in K. Then σ(T ) = T . We claim that T ⊗S A is semiperfect, but (T ⊗S A){σ⊗1} is not. Indeed, T {σ} = T ∩ F = S, so (T ⊗ A){σ⊗1} = S ⊗ A ∼= A which is not semiperfect as explained in the previous example. On the other hand, T ⊗ A is an Azumaya T -algebra and a T -order in K ⊗ D ∼= Mt(K). Again, this implies T ⊗ A is Morita equivalent to T . But T is local because K/F is totally ramified at π, hence T ⊗ A is semiperfect. If we take S, A, F, D as in the previous example, then T = S[√−3], K = Q[√−3] will satisfy (c) and (d). Indeed, dimension constraints imply K ⊗ D is either a division ring or M2(K), but √−3 + i + j + ij ∈ K ⊗ D has reduced norm 0, so the latter option must hold. Example 6.3. Start with a semiperfect ring R and σ ∈ End(R) such that R{σ} is not semiperfect (e.g. those of Example 6.2). Let R′ = R[[t; σ]] be the ring of σ-twisted formal power series with left coefficients in R (i.e. σ(r)t = tr for all r ∈ R) and let x = 1 + t ∈ (R′)×. We claim R′ is semiperfect, but CentR′ (x) is not. Indeed, CentR′ (x) = CentR′ (t) = R{σ}[[t]], so we are done by applying the following proposition for R[[t; σ]] and R{σ}[[t]]. Proposition 6.4. For any ring W and τ ∈ End(W ), W is semiperfect if and only if W [[t; τ ]] is. Proof. Let V = W [[t; τ ]] and let J = Jac(W ) + V t E V . Then V /J ∼= W/ Jac(W ). Since the r.h.s. has zero Jacobson radical, J ⊇ Jac(V ). However, as 1 + J ⊆ V ×, we have J ⊆ Jac(V ), thus Jac(V ) = J. The isomorphism V /J ∼= W/ Jac(W ) now implies that V is semilocal ⇐⇒ W is semilocal. We finish by observing that V t is idempotent lifting (this immediate as V = W ⊕ V t), hence J is idempotent lifting in V ⇐⇒ J/V t is idempotent lifting in V /V t ⇐⇒ Jac(W ) is idempotent lifting in W . (cid:3) We now show (2), relying on the previous examples. Example 6.5. Let R be a semiperfect ring and let X ⊆ R be such that CentR(X) is not semiperfect (the existence of such R and X was shown in previous examples). Let Y = {ya a ∈ X} be a set of formal variables and let S = RhY i be the ring of non-commutative polynomials in Y over R (Y commutes with R). We can make R into a right S-module by considering the standard right action of R onto itself and extending it to S by defining r · ya = ar for all a ∈ X. Let M denote the right S-module obtained thusly. Identify R with End(MR) = End(RR) via r 7→ (m 7→ rm) ∈ End(RR). It is straightforward to check that End(MS) now corresponds to CentR(X). Therefore, End(MR) ∼= R is semiperfect but End(MS) ∼= CentR(X) is not semiperfect. The next example demonstrates (3). Example 6.6. Let p, q be distinct primes. Endow R = Zp×Zq×Q with the product topology (the topology on Q is the discrete topology). Then R is clearly semiperfect and pro-semiprimary. Define K = {(a, a, a) a ∈ Q} and let R0 = R ∩ K. Then R0 is a semi-invariant subring of R by Proposition 3.3 (take S = Qp × Qq × Q) and 18 URIYA A. FIRST it is routine to check R0 is closed. However, R0 is not semiperfect. Indeed, it is isomorphic to T = M −1Z where M = Z \ (pZ ∪ qZ). The ring T is not semiperfect because it has no non-trivial idempotents while T / Jac(T ) ∼= T /pqT ∼= T /pT ×T /qT has such, so Jac(T ) cannot be idempotent lifting. The following example shows that rationally closed subrings of a f.d. algebra need not be semiperfect. As f.d. algebras are semiprimary, this shows that Theorem 4.6 fails for rationally closed subrings. Example 6.7. Let K = Q(x) and R = K × K × K. Define S′ = {f /g f, g ∈ Q[x], g(0) 6= 0, g(1) 6= 0} and observe that S′ is not semiperfect since it is a domain but S′/ Jac(S′) has non-trivial idempotents. (Indeed, S′/ Jac(S′) = S′/ hx(x − 1)i ∼= S′/ hxi × S′/ hx − 1i ∼= Q × Q). Define ϕ : S′ → R to be the Q-algebra homomor- phism obtained by sending x to a := (0, 1, x) ∈ R and let S = im ϕ. It is easy to verify that ϕ is well defined and injective, hence S is not semiperfect. However, S is rationally closed in R. To see this, let q(x) ∈ S′ and assume q(a) ∈ R×. Then q(0), q(1) 6= 0. This implies q(x) ∈ (S′)×, hence q(a) = ϕ(q(x)) ∈ S×. We finish by demonstrating (5). The subring S of the last example cannot be In semi-invariant, for otherwise we would get a contradiction to Theorem 4.6. particular, S is not an invariant subring nor a centralizer subring. Next, let R = Q[ 3√2,√3]. Then the only centralizer subring of R is R itself, the invariant subrings of R are R and Q[ 3√2] (Remark 3.5) and the semi-invariant subrings of R are the four subfields of R (Corollary 3.4). In particular, R admits a semi-invariant non- invariant subring and an invariant non-centralizer subring. 7. Applications This section presents applications of the previous results. In order to avoid cum- bersome phrasing, we introduce the following families of ring-theoretic properties: π∞-regular, π-regular and semiperfect, π-regular Pdisc =(cid:26) semiprimary, right perfect, left perfect, π∞-regular and semiperfect, quasi-Q, quasi-Q and semiperfect (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Q ∈ {π∞-regular, π-regular} ) Ptop =( pro-P, pro-P and semiperfect, (For example, "quasi-π∞-regular and semiperfect" lies in Ptop.) Note that the properties in Pdisc apply to rings while the properties in Ptop apply to LT rings. Nevertheless, we will sometimes address non-topological rings as satisfying one of the properties of Ptop, meaning that they satisfy it w.r.t. some linear ring topology. We also define Pmor (resp. Psp) to be the set of properties in Pdisc ∪ Ptop which are preserved under Morita equivalence (resp. imply that the ring is semiperfect). Recall that a property in Ptop is preserved under Morita equivalence if this holds in the sense of Section 5 (and not in the sense of [15]). For example, "π-regular" and "quasi-π-regular" do not lie in Pmor nor in Psp, "pro-semiprimary and semiperfect" lies in both Psp and Pmor, and "pro-semiprimary" lies in Pmor, but not in Psp. Theorems 4.6, 5.10 and 5.15 and Corollaries 4.3(ii) and 5.8(ii) can be now sum- (cid:27) P ∈ Pdisc marized as follows: Theorem 7.1. Let R be a ring and R0 a subring. (i) If R has P ∈ Pdisc and R0 is semi-invariant, then R0 has P. (ii) If R ∈ LT2 has P ∈ Ptop and R0 is T-semi-invariant, then R0 has P (iii) In both (i) and (ii), if P ∈ Psp, then Jac(R0)n ⊆ Jac(R) for some n ∈ N. Our first application follows from the fact that a centralizer subring is always (w.r.t. the induced topology). (T-)semi-invariant: SEMI-INVARIANT SUBRINGS 19 Corollary 7.2. Let P ∈ Pdisc ∪ Ptop and let R be a ring satisfying P. Then Cent(R) and any maximal commutative subring of R satisfy P. Proof. Cent(R) is the centralizer of R and a maximal commutative subring of R is itself's centralizer. Now apply Theorem 7.1. (cid:3) Surprisingly, the author could not find in the literature results that are similar to the previous corollary, except the fact that the center of a right artinian ring is semiprimary. (This follows from a classical result of Jacobson, stating that the endomorphism ring of any module of finite length is semiprimary, together with the fact that the center of a ring R is isomorphic to End(RRR).) The next applications concern endomorphism rings of finitely presented modules. We will only treat here the non-topological properties (i.e. Pdisc). The topological analogues of the results to follow require additional notation and are thus postponed to the next section. Theorem 7.3. Let R be a ring satisfying P ∈ Pdisc ∩ Pmor and let M be a finitely presented right R-module. Then End(MR) satisfies P. Proof. There is an exact sequence Rn → Rm → M → 0 with n, m ∈ N. Since Rn, Rm are projective, we may apply Proposition 3.7 to deduce that End(M ) is a quotient of a semi-invariant subring of End(Rn)×End(Rm) ∼= Mn(R)×Mm(R). The latter has P because any P ∈ Pdisc ∩ Pmor is preserved under Morita equivalence and under taking finite products. Since all ring properties in Pdisc pass to quotients, we are done by Theorem 7.1. (cid:3) Corollary 7.4. Let ϕ : S → R be a ring homomorphism. Consider R as a right S-module via ϕ and assume it is finitely presented. Then if S satisfies P ∈ Pdisc ∩ Pmor, then so does R. Proof. By Theorem 7.3, End(RS) has P. Therefore, by Corollary 4.7, R ∼= End(RR) has P. Remark 7.5. Theorem 7.3 actually follows from results of Bjork, who proved the semiprimary case and part of the left/right perfect cases ([5, Ths. 4.1-4.2]), and Rowen, who proved the left/right perfect and the semiperfect and π∞-regular cases ([23, Cr. 11 and Th. 8(iii)]). Our approach suggests a single simplified proof to all the cases. Note that we cannot replace "finitely presented" with "finitely generated" in Theorem 7.3; in [6, Ex. 2.1], Bjork presents a right artinian ring with a cyclic left module having a non-semilocal endomorphism ring. See also [13, Ex. 3.5]. (cid:3) By arguing as in the proof of Theorem 7.3 one can also obtain: Theorem 7.6. Let 0 → A → B → C → 0 be an exact sequence in an abelian category A such that B is projective. (i) If End(A) and End(B) has P ∈ Pdisc, then End(C) has P. (ii) If End(A ⊕ B) has P ∈ Psp, then End(C) is semiperfect. Next, we turn to representations over modules with "good" endomorphism ring. By a representation of a monoid (ring) G over a right R-module M , we mean a mononid (ring) homomorphism ρ : G → End(M ) (so G acts on M via ρ). Corollary 7.7. Let R be a ring and let ρ be a representation of a monoid (or a ring) G over a right R-module M . Assume that one of the following holds (i) End(M ) has P ∈ Pdisc ∪ Ptop. (ii) There is a sub-monoid (subring) H ⊆ G such that End(ρH ) has P ∈ Pdisc. (iii) End(M ) is LT and Hausdorff and there is a sub-monoid (or a subring) H ⊆ G such that End(ρH ) has P ∈ Ptop w.r.t. the induced topology. 20 URIYA A. FIRST Then End(ρ) has P. Moreover, if P ∈ Psp, then ρ has a Krull-Schmidt decompo- sition ρ ∼= ρ1 ⊕ ··· ⊕ ρt and End(ρi) is local and has P for all 1 ≤ i ≤ t. Proof. (i) follows from (ii) and (iii) if we take H to be the trivial monoid (basic sub- ring). To see (ii) (resp. (iii)), notice that End(ρ) = CentEnd(ρH )(ρ(G)). Therefore, End(ρ) is a semi-invariant (resp. T-semi-invariant) subring of End(ρH ), hence by Theorem 7.1, End(ρ) has P. If P ∈ Psp, then End(ρ) is semiperfect. The Krull- Schmidt Theorem now implies ρ has a Krull-Schmidt decomposition ρ ∼= ρ1⊕···⊕ρt and End(ρi) is local for all i. Finally, End(ρi) ∼= e End(ρ)e for some e ∈ E(End(ρ)). Since for any ring R and e ∈ E(R), R has P implies eRe has P, we are through. (cid:3) Assume R is a ring and M is a right R-module such that End(MR) is semiperfect and quasi-π-regular (see Theorem 8.3 below for cases when this happens). Then the endomorphisms of M have a "Jordan decomposition" in the following sense: If f ∈ End(MR), then we can consider M as a right R[x]-module by letting x act as f . Clearly End(MR[x]) = CentEnd(MR)(f ), so by Theorem 5.10, End(MR[x]) is semiperfect. Therefore, MR[x] has a Krull-Schmidt decomposition M = M1 ⊕ ···⊕ Mt. (Notice that each Mi is an f -invariant submodule of M .) This decomposition plays the role of a Jordan decomposition for f , since the isomorphism classes of M1, . . . , Mt (as R[x]-modules) determine the conjugation class of f . In particular, studying endomorphisms of M can be done by classifying LE-modules over R[x]. Finally, the results of this paper can be applied in a rather different manner to bilinear forms: Let ∗ be an anti-endomorphism of a ring R (i.e. an additive, unity- preserving map that reverses order of multiplication). Then σ = ∗2 is an endomor- phism of R and ∗ becomes an involution on the invariant subring R{σ}. As some claims on (R,∗) can be reduced to claims on (R{σ},∗R{σ} ), our results become a useful tool for studying the former. Recalling that bilinear (resp. sesquilinear) forms correspond certain anti-endomorphisms and quadratic (resp. hermitian) forms cor- respond to involutions (see [17, Ch. I]), these ideas, taken much further, can be used to reduce the isomorphism problem of bilinear forms to the isomorphism problem of hermitian forms. This was actually done (using other methods) for bilinear forms over fields by Riehm ([22]), who later generalized this with Shrader-Frechette to sesquilinear forms over semisimple algebras ([21]). The author can improve these results for bilinear (sesquilinear) forms over various semiperfect pro-semiprimary rings (e.g. f.g. algebras over Zp). This will be published elsewhere. 8. Modules Over Linearly Topologized Rings In this section we extend Theorem 7.3 and other applications to LT rings. This is done by properly topologizing modules and endomorphisms rings of modules over LT rings. Let R be an LT ring and let M be a right R-module. Then M can be made into a topological R-module by taking {x + M J J ∈ IR} as basis of neighborhoods of x ∈ M . (That M is indeed a topological module follows from [31, Th. 3.6].) Notice that any homomorphism of modules is continuous w.r.t. this topology. Furthermore, End(M ) can be linearly topologized by taking {HomR(M, M J) J ∈ IR} as a local basis.10 We will refer to the topologies just defined on M and End(M ) as In general, that R is Hausdorff does not imply M or their standard topologies. (E.g.: For any distinct primes p, q ∈ Z, the Z-module End(M ) are Hausdorff. Z/q is not Hausdorff w.r.t. the p-adic topology on Z). Observe that {0End(M)} = 10 This topology is the uniform convergence topology (w.r.t. the natural uniform structure of M ). If M is f.g. then this topology coincides with the pointwise convergence topology (i.e. the topology induced from the the product topology on M M ). SEMI-INVARIANT SUBRINGS 21 implies End(M ) is Hausdorff. TJ∈IR Hom(M, M J) = Hom(M,TJ∈IR M J) = Hom(M,{0M}), so M is Hausdorff Now let E• be a be a finite resolution of M , i.e. E• consists of an exact sequence En−1 → ··· → E0 → E−1 = M → 0.11 The maps Ei → Ei−1 will be denoted by di. We say that E• has the lifting property if any f−1 ∈ End(M ) can be extended to a chain complex homomorphism f• : E• → E•. (Recall that f• consists of a sequence {fi}n−1 i=−1 such that fi ∈ End(Ei) and difi = fi−1di for all i.) In other words, E• has the lifting property if and only if the following commutative diagram can be completed for every f−1 ∈ End(M ). En−1 . . . / E1 E0 M fn−1 f1 f0 f−1 En−1 / . . . / E1 / E0 / M • and β• : P ′ For example, any projective resolution has the lifting property. We define a linear ring topology τE on End(M ) as follows: For all J ∈ IR, define B(J, E) to the set of maps f−1 ∈ End(M, M J) that extend to a chain complex homomorphism f• : E• → E• such that im fi ⊆ EiJ for all −1 ≤ i < n. The lifting property implies B(J, E) E End(M ) and it is clear that BE := {B(J, E) J ∈ IR} is a filter base. Therefore, there is a unique ring topology on End(M ), denoted τE, having BE as a local basis. It turns out that if E• is a projective resolution, then τE only depends on the length of E, i.e. the number n. Indeed, if P•, P ′ • are two projective resolutions of length n of M , then the map idM : M → M gives rise to chain complex homomor- phisms α• : P• → P ′ • → P• with α−1 = β−1 = idM . Now, if J ∈ IR and f−1 ∈ B(J, P ), then there is f• : P• → P• such that im fi ⊆ PiJ for all i. Define f ′ • = α•f•β•. Then im f ′ −1 = idM f−1 idM = f−1, so f−1 ∈ B(J, P ′). By symmetry, we get B(J, P ) = B(J, P ′) for all J ∈ IR, hence τP = τP ′ . The topology of End(M ) obtained from a projective resolution of length n will be denoted by τ M n and the closure of the zero ideal in that topology will be denoted by I M is the standard topology on End(M ). (Indeed, if P• : P0 → M → 0 is a projective resolution of length 1, then any f ∈ Hom(M, M J) can be lifted to f0 : P0 → P0J because the map P0J → M J is onto, hence B(P, J) = Hom(M, M J).) More generally, for any resolution E• of M , τE contains the standard topology on End(M ). Therefore, if M is Hausdorff, then τE is Hausdorff. In the appendix we provide sufficient conditions for τ M 2 ⊆ . . . and that that τ M i ⊆ αi(PiJ) ⊆ P ′ i J for all i and f ′ n . Note that τ M 1 ⊆ τ M 1 1 , τ M 2 , . . . to coincide. With this terminology at hand, we can generalize Proposition 3.7: Proposition 8.1. Let R be an LT ring and let E : A → B → C → 0 be an exact sequence of right R-modules satisfying the lifting property (w.r.t. C) and such that A and B are Hausdorff. Assign End(A) and End(B) the natural topology and endow End(C) with τE. Then End(C) is isomorphic as a topological ring to a quotient of a T-semi-invariant subring of End(A) × End(B). Proof. We use the notation of the proof of Proposition 3.7. By that proof, End(C) is isomorphic to a quotient of CentD((cid:2) 0 f 0 0(cid:3)). It is easy to check that the embedding D ֒→ S is a topological embedding, hence End(C) is isomorphic to a quotient of a T-semi-invariant subring of D. That the quotient topology on End(C) is indeed τE is routine. (cid:3) 11 We do not require the map En−1 → En−2 to be injective. / / / / / / / / / / / 22 URIYA A. FIRST We are now in position to generalize previous results. Lemma 8.2. Let R ∈ LT2 and P ∈ Pdisc. Then R is pro-P if any only if R is complete and R/I has P for all I ∈ IR. Proof. If R is complete and R/I has P for all I ∈ IR, then R ∼= lim ←−{R/I}I∈IR, so R is pro-P. On the other hand, if R is pro-P, then it is complete. In addition, it is strictly pro-P (Corollary 5.16), hence there is a local basis of ideals B such that R/I has P for all I ∈ B. Now, let I ∈ IR. Then there is I0 ∈ B contained in I. Now, R/I is a quotient of R/I0. As the latter has P, so does R/I. (cid:3) Recall that a topological ring is first countable if it admits a countable local basis. If R is pro-P, then this is equivalent to saying that R is the inverse limit of countably many discrete rings satisfying P. Theorem 8.3. Let R ∈ LT2 be a ring and let M be a f.p. right R-module. (i) If R is first countable and satisfies P ∈ Ptop ∩ Pmor, then End(M )/I M 2 2 . In particular, if M is Haus- (ii) Assume R is quasi-π∞-regular and let i ∈ {1, 2}. Then End(M )/I M satisfies P when End(M ) is endowed with τ M dorff, then End(M ) has P. quasi-π∞-regular when End(M ) is endowed with τ M i is Hausdorff, then End(M ) is quasi-π∞-regular w.r.t. τ M 1 . is . In particular, if M i (iii) If R is semiperfect and quasi-π∞-regular, then End(M ) is semiperfect. I M 2 Proof. (i) The argument in the proof of Theorem 7.3 shows that End(M ) is a quo- tient of an LT Hausdorff ring satisfying P, which we denote by W (use Proposition 8.1 instead of Proposition 3.7). is a closed ideal of End(M ) and therefore End(M )/I M is a quotient of W by a closed ideal. We finish by claiming that for 2 any closed ideal I E W , W/I satisfies P. We will only check the case P = pro-Q for Q ∈ Pdisc. The other cases are straightforward or follow from the pro-Q case. Indeed, any open ideal of W/I is of the form J/I for some J ∈ IW , hence by Lemma 8.2, (W/I)/(J/I) ∼= W/J satisfies Q. In addition, that R is first countable implies W is first countable, hence by the Birkhoff-Kakutani Theorem, W is metrizable. By [8, p. 163] a Hausdorff quotient of a complete metric ring is complete, hence W/I is complete. Therefore, by Lemma 8.2 (applied to W/I), W/I is pro-Q. (ii) The case i = 2 follows from the argument of (i) since being π-regular passes to quotients by closed ideals (the first countable assumption is not needed). As for i = 1, Since I M is 1 quasi-π∞-regular when M is equipped with τ M 2 . We are done by observing that if a ring is quasi-π∞-regular w.r.t. a given topology, then it is quasi-π∞-regular w.r.t. any linear Hausdorff sub-topology. 2 . Therefore, End(M )/I M 1 1 , it is also closed in τ M is closed in τ M (iii) By (i), End(M ) is a quotient of a semiperfect ring, namely W . (cid:3) Remark 8.4. Part (iii) of Theorem 8.3 was proved in [24] for complete semilocal rings with Jacobson radical f.g. as a right ideal and in [23] for semiperfect π∞-regular rings. Both conditions are included in being semiperfect and quasi-π∞-regular. In addition, V´amos proved in [30, Lms. 13-14] that all finitely generated or torsion- free of finite rank modules rank over a Henselian integral domain12 have semiperfect endomorphism ring. Results of similar flavor were also obtained in [13], where it is shown that the endomorphism ring of a f.p. (resp. f.g.) module over a semilocal (resp. commutative semilocal) ring is semilocal. Corollary 8.5. Let S be a commutative LT ring and let R be an S-algebra s.t. R is f.p. and Hausdorff as an S-module. Then: 12 A commutative ring R is called Henselian if R is local and Hensel's Lemma applies to R. SEMI-INVARIANT SUBRINGS 23 (i) If S is quasi-π∞-regular, then R is quasi-π∞-regular (w.r.t. to some linear ring topology). If moreover S is semiperfect, then so is R. (ii) If S satisfies P ∈ Ptop ∩ Pmor w.r.t. a given topology which is also first countable, then R satisfies P. Proof. We only prove (ii); (i) is similar. By Theorem 7.3, End(RS) satisfies P. For all r ∈ R, define br ∈ End(RS) bybr(x) = xr and observe that CentEnd(RS)({br r ∈ R}) ∼= End(RR) = R, hence R has P by Theorem 7.1. (cid:3) Let C be a commutative local ring. Azumaya proved in [2, Th. 22] that C is Henselian if and only if every commutative C-algebra R with RC f.g. is semiperfect. This was improved by V´amos to non-commutative C-algebras in which all non-units are integral over C; see [30, Lm. 12]. Given the previous corollary, Azumaya and V´amos' results suggest that the notions of Henselian and quasi-π∞-regular might sometimes coincide. This is verified in the following proposition. Proposition 8.6. Let R be a rank-1 valuation ring. Then R is Henselian if and only if R is quasi-π∞-regular w.r.t. the topology induced by the valuation. Proof. Assume R is quasi-π∞-regular. Observe that any free R-module is Hausdorff w.r.t. the standard topology, hence Corollary 8.5(i) implies that any R-algebra A such that AR is free of finite rank is semiperfect. Thus, by [2, Th. 19], R is Henselian. Conversely, assume R is Henselian. Denote by ν the (additive) valuation of R. Since ν is of rank 1, we may assume ν take values in (R, +). For every δ ∈ R, let Iδ = {x ∈ R ν(x) > δ}. Then {Mn(Iδ) δ ∈ [0,∞)} is a local basis for Mn(R). Let a ∈ Mn(R). By the Cayley-Hamilton theorem, a is integral over R, hence R[a] is a f.g. R-module. Let J = Jac(R) · R[a]. Then J E R[a] and it is well known that J ⊆ Jac(R[a]). The ring R[a]/J is artinian, hence a + J has an associated idemptent ε ∈ E(R[a]/J) (i.e. ε satisfies conditions (A) -- (C) of Lemma 4.2). By [2, Th. 22], J is idempotent lifting, hence there is e ∈ E(R[a]) such that e + J = ε. Let f = 1 − e. Then a = eae + f af (since R[a] is commutative). Furthermore, eae + J is invertible in ε(R[a]/J)ε, hence eae is invertible in eR[a]e and in particular in eMn(R)e. Next, (f af )k ∈ J ⊆ Mn(Jac(R)) = Mn(I0) for some k ∈ N. This means (f af )k ∈ Mn(Iδ) for some 0 < δ ∈ R, which implies (f af )m m→∞−−−−→ 0. Thus, e satisfies conditions (A),(B) and (C′) w.r.t. a and we may conclude that Mn(R) is quasi-π-regular for all n ∈ N. (cid:3) Using the ideas in the proof of Theorem 8.3, we can also obtain: Theorem 8.7. Let R be an LT ring and let E : 0 → A → B → C → 0 be an exact sequence of right R-modules such that B is projective and A and B are Hasudorff. Endow End(A) and End(B) with their standard topologies and End(C) with τE and let IE denote the closure of the zero ideal in End(C). Then: does End(C)/IE. (i) If End(A) and End(B) are first countable and satisfy P ∈ Ptop, then so (ii) If End(A) and End(B) are quasi-π-regular, then so does End(C)/IE. (iii) If End(A) and End(B) are quasi-π-regular and semiperfect, then End(C) is semiperfect. In light of the previous results, one might wonder under what conditions all right f.p. modules over an LT ring are Hausdorff. This is treated in the next section and holds, in particular, for right noetherian pro-semiprimary rings and rank-1 complete valuation rings. 24 URIYA A. FIRST We finish this section by noting that we can take a different approach for complete Hausdorff modules. For the following discussion, a right R-module module M is called complete if the natural map M → lim←−{M/M J}J∈IR is an isomorphism.13 Proposition 8.8. (i) Let R be a complete first countable Hausdorff LT ring. Then any Hausdorff f.g. right R-module is complete. (ii) Let P ∈ Pdisc and let R be an LT ring such that R/I has P for all I ∈ IR. Let M be a complete right R-module such that M/JM is f.p. as a right R/J-module for all J ∈ IR (e.g. if M is f.p., or if M is f.g. and R is strictly pro-right-artinian). Then End(M ) is pro-P w.r.t. τ M 1 . If moreover R is semiperfect and M is f.g., then End(M ) is semiperfect. Proof. (i) This is a well-known argument: Let B be a countable local basis of R consisting of ideals. Without loss of generality, we may assume B = {Jn}∞ n=1 with J1 ⊇ J2 ⊇ . . . . Let M be a f.g. Hausdorff R module and let {x1, . . . , xn} be a set of generators of M . Since M is Hausdorff, it is enough to show that any sum j=1 xj rij with i=1 mi converges Indeed, write mi = Pn i=1 rij converges in R for all j, henceP∞ (ii) Throughout, J denotes an open ideal of R. We first note that if M is f.p. then there is an exact sequence Rn → Rm → M → 0 for some n, m ∈ N. Tensoring it with R/J we get (R/J)n → (R/J)m → M/M J → 0, implying M/J is a f.p. R/J-module. Next, if M is f.g. and R is strictly pro-right-artinian, then M/M J is a f.g. module over R/J which is right artinian, hence M/J is f.p. over R/J. i=1 mi with mi ∈ M Ji converges in M . P∞ ri1, . . . , rin ∈ Ji. ThenP∞ toPn j=1 xjrj where rj =P∞ i=1 rij . Now, since M/M J is f.p. over R/J, End(M/M J) satisfies P by Theorem 7.3. There is a natural map End(M ) → End(M/M J) whose kernel is Hom(M, M J). Assign End(M ) the standard topology. Then since M is complete, Hom(M, M ) ∼= lim←−{End(M/M J)}J∈I as topological rings and therefore, End(M ) is pro-P. Finally, assume M is f.g. and R is semiperfect. Then by Proposition 2.2, M admits a projective cover P which is easily seen to be finitely generated. Assume M = M1 ⊕ ··· ⊕ Mt. Then each Mi is f.g. and thus has a projective cover Pi. Necessarily P ∼= P1 ⊕ ··· ⊕ Pt. By Proposition 2.4, End(PR) is semiperfect, hence there is a finite upper bound on the cardinality of sets of orthogonal idempotents in it. This means t is bounded and hence, End(M ) cannot contain an infinite set of orthogonal idempotents. By Lemma 5.9(ii), this implies End(M ) is semiperfect. (cid:3) 9. LT Rings With Hausdorff Finitely Presented Modules In this section we present sufficient conditions on an LT ring guaranteeing all right f.p. modules are Hausdorff (w.r.t. the standard topology). The discussion leads to an interesting consequence about noetherian pro-semiprimary rings. We begin by noting a famous result that solves the problem for many noetherian rings with the Jacobson topology. For proof and details, see [25, Th. 3.5.28]. Theorem 9.1. (Jategaonkar-Schelter-Cauchon). Let R be an almost fully bounded noetherian ring whose primitive images are artinian (e.g. a noetheorian PI ring). Assign R the Jac(R)-adic topology or any stronger linear ring topology. Then any f.g. right R-module is Hausdorff. Example 9.2. The assumption that all powers of Jac(R) are open in the last theorem cannot be dropped: Let R be a Dedekind domain with exactly two prime ideals P and Q. Then R is noetherian, almost fully bounded and any primite image of R is artinian. Let n ∈ N ∪ {0} and let B = {P mQn m ∈ N}. Assign R 13 Completeness can also be defined for non-Hausdorff topological abelian groups; see [31]. SEMI-INVARIANT SUBRINGS 25 the unique topology with local basis B. Clearly Jac(R)k = P kQk is open for all m=1(P n+1Qn+1 + P mQn) = m=1(P min{n+1,m}Qn) = P n+1Qn, so Jac(R)n+1 is not closed. In particular, by (∗) below, R/ Jac(R)n+1 is a f.g. non-Hausdorff R-module. 1 ≤ k ≤ n. However, Jac(R)n+1 = P n+1Qn+1 = T∞ T∞ When considering quasi-π-regular rings, there is actually no point in taking a topology stronger than the Jacobson topology in Theorem 9.1, because for right noetherian rings the latter is the largest topology making the ring quasi-π-regular. Proposition 9.3. Let R be an LT semilocal ring and let τ be the topology on R. Assume R is quasi-π-regular w.r.t. τ and R/I is semiprimary for all I ∈ IR (e.g. if R is right noetherian or pro-semiprimary w.r.t. τ ). Then R is quasi-π-regular w.r.t. the Jacobson topology and the latter contains τ . Proof. We first note that if R is right noetherian, then for all I ∈ IR, R/I is right noetherian and π-regular, hence by Remark 2.9, R/I is semiprimary. If R is pro-semiprimary, then R/I is semiprimary for all I ∈ IR by Lemma 8.2. Let τJac denote the Jacobson topology and let a ∈ R. Then a has an associated idempotent e w.r.t. τ . Let f = 1 − e and observe that Jac(R) is open by Remark 5.12(ii). Then there is n ∈ N such that (f af )n ∈ Jac(R) and it follows that (f af )n n→∞−−−−→ 0 w.r.t. τJac. Therefore, e is the associated idempotent of a w.r.t. τJac, hence R is quasi-π-regular provided we can verify τJac is Hausdorff. This holds since τJac ⊇ τ , by the proof of Corollary 5.17 (which still works under our weaker assumptions). (cid:3) Stronger linear topologies are "better" since they have more Hausdorff modules. Note that the topology of an arbitrary qausi-π∞-regular ring can be stronger than the Jacobson topology. For example, take any non-semiprimary perfect ring R with m = xnxm = 0 ∀n > 2m]) and give Tn∈N Jac(R) = {0} (e.g. R = Q[x1, x2, x3, . . . x2 it the discrete topology. The next result will rely on following observation: (∗) Let R be an LT ring. If M is a right R-module and N is a submodule, then N/N = N /N . In particular, M/N is Hausdorff if and only if N is closed. Indeed, N/N =TJ∈IR(M/N )J =TJ∈IR(M J + M )/N = (TJ∈IR(M J + M ))/N = N /N . We will also need the following theorem. For proof, see [7, §7.4]. Theorem 9.4. Let {Xi, fij} be an I-indexed inverse system of non-empty sets. Assume that for each i ∈ I we are given a family of subsets Ti ⊆ P (Xi) such that for all i ≤ j in I we have: (a) Xi ∈ Ti and Ti is closed under (arbitrary large) intersection. (b) Finite Intersection Property: If L ⊆ Ti is such that the intersection of finitely many of the elements of L is non-empty, thenTA∈L A 6= φ. (c) For all A ∈ Tj, fij(A) ∈ Ti. (d) For all x ∈ Xi, f −1 ij (x) ∈ Tj. Then lim←−{Xi}i∈I is non-empty.14 Lemma 9.5. Let R be a ring and let M be a right R-module. Let {Mi}i∈I be a family of submodules of M and let {xi}i∈I be elements of M . ThenTi∈I (xi + Mi) is either empty or a coset ofTi∈I Mi. Proof. This is straightforward. (cid:3) 14 This can be compared to the following topological fact: An inverse limit of an inverse system of non-empty Hausdorff compact topological spaces is non-empty and compact. 26 URIYA A. FIRST Theorem 9.6. Let R be strictly pro-right-artinian. Then any f.g. submodule of a Hausdorff right R-module is closed. We will show that m ∈ N implies m ∈ N . Let m ∈ N . For every J ∈ B define Proof. Let B be a local basis of ideals such that R/J is right artinian for all J ∈ B. i=1 miR. Assume M is a Hausdorff right R-module, let m1, . . . , mk ∈ M and N =Pk (mi + M J)ai = m + M J) . XJ =((a1, . . . , ak) ∈ (R/J)k : Xi Observe that m ∈ N =TJ∈B(N +M J), hence for all J ∈ B there are b1, . . . , bk ∈ R and z ∈ M J such thatP mibi = m + z, implying XJ 6= φ. For all J ⊆ I in B, let fIJ denote the map from (R/J)k to (R/I)k given by sending (b1 + J, . . . , bk + J) to (b1 + I, . . . , bk + I). Then fIJ (XJ ) ⊆ XI . It easy to check that {XI , fIJXJ} is an inverse system of sets. For all J ∈ B, define TJ to be the set consisting of the empty set together with all cosets of (right) R-submodules of (R/J)k contained in XJ . We claim that conditions (a)-(d) of Theorem 9.4 hold. Indeed, XJ is easily seen to be a coset of a submodule of (R/J)k, thus XJ ∈ TJ . In addition, by Lemma 9.5, TJ is closed under intersection, so (a) holds. Since R/J is right artinian, so is (R/J)k (as a right R-module). Lemma 9.5 then implies that cosets of submodules of (R/J)k satisfy the DCC, hence (b) holds. Conditions (c) and (d) are straightforward. Therefore, we may apply Theorem 9.4 to deduce that lim←− XJ is non-empty. 1 , . . . , a(J) Let x ∈ lim←−{XJ}J∈B. Then x consists of tuples {(a(J) k ) ∈ (R/J)k}J∈B that are compatible with the maps {fIJ}. As R is complete, there are b1, . . . , bk ∈ R i = bi + J for all 1 ≤ i ≤ k and J ∈ B. It follows that m −Pi mibi ∈ such that a(J) TJ∈B M J. As M is Hausdorff, the right-hand side is {0}, so m =Pi mibi ∈ N . (cid:3) Remark 9.7. Theorem 9.6 and its consequences actually hold for the larger class of strictly pro-right-finitely-cogenerated rings. A module M over a ring R is called finitely cogenerated 15 (abbrev.: f.cog.) if its submodules satisfy the Finite Inter- section Property (condition (b) in Theorem 9.4). This is equivalent to soc(M ) being f.g. and essential in M (see [18, Pr. 19.1]). A ring R called right f.cog. if RR is finitely cogenerated. (For example, any right pseudo-Frobeniuos ring is right f.cog.) Among the examples of strictly pro-right-finitely-cogenerated rings are com- plete rank-1 valuation rings. Indeed, if ν : R → R is an (additive) valuation, and R is complete w.r.t. ν, then R = lim ←−{R/{x ∈ R ν(x) > n}}n∈N. For a detailed discussion about f.cog. modules and rings, see [29] and [18, §19]. Notice that a complete semilocal ring is strictly pro-right-artinian if and only if its Jacobson radical is f.g. as a right module. The latter condition is commonly used when studying complete semilocal rings (e.g. [24]). In particular, Hinohara proved Theorem 9.6 for complete semilocal rings satisfying it ([16, Lm. 3]). (Other authors usually assume the ring is right noetherian.) By (∗) we now get: Corollary 9.8. Let R be a strictly pro-right-artinian ring. Then any f.p. right R-module is Hausdorff. We can now prove that under mild assumptions, strictly pro-right-artinian rings are complete semilocal. 15 Other names used in the literature are "co-finitely generated", "finitely embedded" or "essentially artinian". SEMI-INVARIANT SUBRINGS 27 Corollary 9.9. Let R be a strictly pro-right-artinian ring. If J ⊆ Jac(R) is an ideal that is f.g. as a right ideal, then R is complete in the J-adic topology (i.e. R ∼= lim←−{R/J n}n∈N). If moreover R/J is right artinian, then the topology on R is the Jacobson topology! In particular, if Jac(R) is f.g. as a right ideal, then the topology on R is the Jacobson topology and R is complete semilocal. Let ϕ denote the standard map from R to lim Proof. Let B be a local basis of ideals of R such that R/I is right artinian for all I ∈ B. We identify R with its natural copy inQI∈B R/I. Since J is f.g. as a right ideal, then so are its powers. Therefore, by Theorem 9.6, J n is closed for all n ∈ N. ←−{R/J n}n∈N. Define a map ψ : lim←−{R/J n}n∈N → R as follows: Let r ∈ lim←−{R/J n}n∈N and let rn denote the image of r in R/J n. By Corollary 5.17, for all I ∈ B, there is n ∈ N (depending on I) such that J n ⊆ I. Let rI denote the image of rn in R/I. It is easy to check that rI is independent of n and thatbr := (rI )I∈B ∈ R. Define ψ(r) =br. It is straightforward to check that ψ ◦ ϕ = id. Therefore, we are done if we show that ψ is injective. Let y ∈ ker ψ and let yn + J n be the image of y in R/J n. Then for all I ∈ B, J n ⊆ I implies yn ∈ I. This means yn ∈TJ n⊆I∈B I = J n = J n, so yn + J n = 0 + J n for all n ∈ N, hence y = 0. Now assume R/J is right artinian. Then Jac(R)k ⊆ J ⊆ Jac(R) for some k ∈ N, hence the Jacobson topology and the J-adic topology coincide. By Proposition 9.3, the topology on R is contained in the Jacobson topology, so we only need to show the converse. Let n ∈ N. It is enough to show that J n is open. Indeed, since JR is f.g., then so is (J i/J i+1)R (i ≥ 0). As (R/J)R has finite length, (J i/J i+1)R has finite length. Thus, (R/J n)R have finite length as well. Since J n is closed, J n is an intersection of open ideals. As (R/J n)R is of finite length, J n is the intersection of finitely many of those ideals, hence open. (cid:3) Corollary 9.10. Let R be a right noetherian pro-π-regular ring. Then the topology on R is the Jacobson topology, R is strictly pro-right-artinian w.r.t. it and any right ideal of R is closed. In particular, R is semilocal complete. Proof. By Lemma 8.2, R/I is π-regular for all I ∈ IR, hence Remark 2.9 implies R/I is right artinian for all I ∈ IR (since R/I is right noetherian). Therefore, R is pro-right-artinian, with Jac(R)R finitely generated. Now apply the previous corollary. (cid:3) The next example demonstrates that Theorem 9.6 fails for pro-artinian rings (and in particular for pro-semiprimary rings). It also implies that there are pro- artinian rings that are not strictly pro-right-artinian. Example 9.11. Let S = Q(x)[t t3 = 0]. For all n ∈ N define Rn = Q(x2n ) + Q(x)t + Q(x)t2 ⊆ S and In = Q(x2n )t2 ⊆ S. Then Rn is an artinian ring and In E Rn. For n ≤ m define a map fnm : Rm/Im → Rn/In by fnm(x + Im) = x + In. Then {Rn/In, fnm} is an inverse system of artinian rings. Let R = lim←−{Rn/In}n∈N. Then R can be identified with Q + Q(x)t + V t2 where V is the Q-vector space lim←−{Q(x)/Q(x2n )}n∈N (R does not embed in S). Observe that Q(x) is dense in V , but Q(x) 6= V since V is not countable (it contains a copy of all power series P anx2n ∈ Q[[x]]). Therefore, the ideal tR = Qt + Q(x)t2 is not closed in R and by (∗), R/tR is a non-Hausdorff f.p. module. We also note that Jac(R)2 = Q(x)t2 is not closed (but Jac(R) must be closed by Proposition 5.11). We conclude by specializing the results of the previous section to first count- able strictly pro-right-artinian rings. (We are guaranteed that all f.p. modules are 28 URIYA A. FIRST Hausdorff in this case). By Corollary 9.10, this family include all right noether- ian pro-semiprimary rings. More general statements can be obtained by applying Remark 9.7. Corollary 9.12. (i) Let R be a first countable pro-right-artinian ring and let M be a f.p. right R-module. Then End(MR) is pro-semiprimary and first countable (w.r.t. τ M 2 ). If R is semiperfect (e.g. if R is right noetherian), then End(MR) is semiperfect. (ii) Let S be a commutative first countable pro-right-artinian ring and let R be an S-algebra s.t. R is f.p. as an S-module. Then R is pro-semiprimary (w.r.t. some topology). If S is semiperfect (e.g. if S is right noetherian), then R is semiperfect. 10. Further Remarks It is likely that the theory of semi-invariant subrings developed in section 5 can be extended to right linearly topologized rings, i.e. topological rings having a local basis consisting of right ideals. This actually has the following remarkable implication (compare with Corollary 7.4 and Corollary 8.5): Conjecture 10.1. Let S ∈ LT2 be a semiperfect quasi-π∞-regular ring and let ϕ : S → R be a ring homomorphism. Assume that: (a) When considered as a right S-module via ϕ, R is f.p. and Hausdorff. (b) For all r ∈ R and I ∈ IS, there is J ∈ IS such that Rϕ(J)r ⊆ Rϕ(I).16 Then R is semiperfect and quasi-π∞-regular (w.r.t. some topology). The proof should be as follows: For any right S-module M , let W denote the ring of continuous Z-homomorphisms from M to itself. Then W can be made into a right LT ring by taking {B(J) J ∈ IS} as a local basis where B(J) = {f ∈ im f ⊆ M J} (this is the topology of uniform convergence).17 Clearly W W : contains End(MS) as a topological ring (endow End(MS) with τ M 1 ). Now take M = R (where R is viewed as a right S-module via ϕ). Then condition (a) implies End(RS) is semiperfect and quasi-π∞-regular w.r.t. τ R 1 (Theorem 8.3). Condition End(RS), so R is semiperfect and quasi-π∞-regular. and hence lie in W . Since we assume the results of section 5 extend to right LT (b) implies that for all r ∈ R, the map br : x 7→ xr from R to itself is continuous rings, R ∼= End(RR) = CentEnd(RS )({br r ∈ R}) is a T-semi-invariant subring of Examples of rings satisfying conditions (a) and (b) can be produced by taking R to be: (1) a twisted group algebra SαG where G is a finite group and α : G → Autc(G) is a group homomorphism or (2) a "crossed product", i.e. R = CrossProd(S, ψ, G) where S is commutative, G is finite and act on S via continuous automorphisms and ψ ∈ H 2(G, S×). (Further examples can be produced by taking quotients). However, we can show directly that the conjecture holds in these special cases. Indeed, that G is finite implies B = {Tg∈G g(I) I ∈ IR} is a local basis of S and we have RJ ⊆ JR for all J ∈ B. For any right R-module M let W ′ = {f ∈ W : f (M J) ⊆ M J ∀J ∈ B} (with W as in the previous paragraph). Then W ′ is a linearly topologized ring w.r.t. the topology induced from W (as seen by taking the local basis {W ′ ∩ B(J) J ∈ B}). In addition, when M = RS,br of the previous paragraph lies in W ′ (since RJ ⊆ JR for all J ∈ B). Therefore, repeating the argument of the last paragraph with W ′ instead of W , we get that R is semiperfect and quasi-π∞-regular. 16 This equivalent to saying that the topology on R spanned by cosets of the left ideals {Rϕ(I) I ∈ IS} is a ring topology; see [31, §3]. 17 Caution: Not any filter base of right ideals gives rise to a ring topology. By [31, §3], we need to check that for all f ∈ W and I ∈ IS there is J ∈ IS such that f B(J) ⊆ B(I). Indeed, we can take any J with f (M J) ⊆ M I and such J exists since f is continuous. SEMI-INVARIANT SUBRINGS 29 The author could not find nor contradict the existence of the following: (1) A pro-semiprimary ring that is not complete semilocal (i.e. complete w.r.t. its Jacobson topology). (2) A complete semilocal ring, endowed with the Jacobson topology, with a non-Hausdorff f.p. module. 11. Appendix: When Do τ M 1 , τ M 2 , . . . Coincide? This appendix is dedicated to the question of when the topology obtained from a resolution is the standard topology. For that purpose we briefly recall the Artin- Rees property for ideals. For details and proofs, see [25, §3.5D]. Let R be a right noetherian ring. An ideal I E R is said to satisfy the Artin-Rees property (abbreviated: AR-property) if for any right ideal A ≤ R there is n ∈ N such that I n∩A ⊆ AI. This is well known to imply that for any f.g. right R-module M and a submodule N , there is n ∈ N such that M I n ∩ N ⊆ N I. For example, by [25, p. 462, Ex. 19], every polycentral ideal (e.g. an ideal generated by central elements) satisfies the AR-property. In addition, if R is almost bounded (e.g. a PI ring), then all ideals of R are satisfy the AR-property. Now let R be any LT ring. A right R-module M is said to satisfy the topological Artin-Rees property (abbreviated: TAR-property) if for any submodule N ⊆ M and any I ∈ IR there is J ∈ IR such that M J ∩ N ⊆ N I. (Equivalently, the induced topology and the standard topology coincide for any submodule of M .) For example, if R is right noetherian, J E R and R is given the J-adic topology, then all f.g. right R-modules satisfy the TAR-property if and only if J satisfies the AR-property. Proposition 11.1. Let R be an LT ring, let M be a right R-module and let P : Pn−1 → ··· → P0 → P−1 = M → 0 be a projective resolution of M . Assume that that P0, . . . , Pn−1 have the TAR-property. Then τP is the natural topology on End(M ). Proof. Denote by di the map Pi → Pi−1 and let Bi = ker di. We will prove that for all I ∈ IR there is J ∈ IR such that Hom(M, M J) ⊆ B(I, P ). Given I ∈ IR we define a sequence of open ideals In−1, In−2, . . . , I−1 as follows: Let In−1 = I. Given Ii, take Ii−1 to be an open ideal such that Pi−1Ii−1 ∩ Bi−1 ⊆ Bi−1Ii and Ii−1 ⊆ Ii (the existence of Ii−1 follows from the TAR-property). We claim that Hom(M, M I−1) ⊆ B(I, P ). To see this, let f−1 ∈ Hom(M, M I−1) and assume we have constructed maps fi ∈ Hom(Pi, PiIi) for all −1 ≤ i < k such that difi = fi−1di. Then it is enough to show there is fk ∈ Hom(Pk, PkIk) such that dkfk = fk−1dk. The argument to follow is illustrated in the following diagram: Pk dk Bk−1 Pk−1 dk−1 / . . . fk−1 fk−1 fk Pk−1Ik−1 ∩ Bk−1 / Pk−1Ik−1 dk−1 / . . . PkIk dk / Bk−1Ik That dk−1fk−1 = fk−2dk−1 implies fk−1(Bk−1) ⊆ Bk−1. As im fk−1 ⊆ Pk−1Ik−1, we get that fk−1(Bk−1) ⊆ Pk−1Ik−1 ∩ Bk−1 ⊆ Bk−1Ik (by the definition of Ik−1). Since the map PkIk → Bk−1Ik is onto (because im(dk) = Bk−1), we can lift fk−1dk : Pk → Bk−1Ik to module homomorphism fk : Pk → PkIk, as required. (cid:3) / / / / / / / _ / / / / / 30 URIYA A. FIRST Remark 11.2. The proof still works if we replace the assumption that Pn−1 is projective with dn−1 is injective. Corollary 11.3. Let R be an LT right noetherian ring admitting local basis of ideals B such that: (1) all ideals in B have the AR-property (e.g. if all ideals in B are generated by central elements or if R is PI) and (2) all powers of ideals in B are open. Then τ M 2 = . . . for any f.g. right R-module M . 1 = τ M Proof. Let n ∈ N. Since R is right noetherian, any f.g. R-module admits a resolution of length n consisting of f.g. projective modules. The assumptions (1) and (2) are easily seen to imply that any f.g. R-module satisfies the TAR-property. Therefore, by Proposition 11.1, τ M (cid:3) n is the natural topology on End(M ). Example 11.4. Condition (2) in Corollary 11.3 is essential even when all ideals of R have the AR-property: Assign Z the unique topology with local basis B = {2·3nZ n ≥ 0} and let M = Z/4×Z/2. Since M I = 2M for all I ∈ B, the standard topology on End(M ) is obtained from the local basis {Hom(M, 2M )}. (M is not Hausdorff). Let I = 2Z ∈ B and consider the projective resolution P : 4Z × 2Z ֒→ Z × Z → M → 0 . Define f−1 : M → M I = 2M by f (x + 4Z, y + 2Z) = (2y + 4Z, 0). Then any lifting f0 ∈ End(Z × Z) of f−1 must satisfy f0(0, 1) = (4x + 2, 2y) for some x, y ∈ Z. This means that any lifting f1 ∈ End(4Z × 2Z) of f0 (there is only one such lifting) satisfies f1(0, 2) = (8x + 4, 4y) /∈ 8Z × 4Z = (4Z × 2Z)I. Therefore, f1 /∈ Hom(4Z× 2Z, (4Z× 2Z)I), implying f−1 /∈ B(P, I). But this means that B(P, I) ( Hom(M, 2M ), hence τ M 2 6= τ M 1 . Acknowledgements The author thanks his supervisor Uzi Vishne for his help and guidance, to Louis H. Rowen for his useful advice, to Eli Matzri for his help in choosing the appropriate values in Examples 6.1 and 6.2 and to Michael Megrelishvili and Menny Shlonssberg for their help with anything that involves topology. Ofir Gorodetsky, Tomer Schlank and others have contributed to the formulation and proof of Theorem 9.6 and the author is grateful for their help as well. Finally, the author also thanks the referee for his beneficial comments. References [1] Afshin Amini, Babak Amini, and Alberto Facchini. Weak Krull-Schmidt for infinite direct sums of cyclically presented modules over local rings. Rend. Semin. Mat. Univ. Padova, 122:39 -- 54, 2009. [2] Goro Azumaya. On maximally central algebras. Nagoya Math. J., 2:119 -- 150, 1951. [3] Francesco Barioli, Alberto Facchini, Francisco Raggi, and Jos´e R´ıos. Krull-Schmidt theorem and homogeneous semilocal rings. Comm. Algebra, 29(4):1649 -- 1658, 2001. [4] Hyman Bass. Finitistic dimension and a homological generalization of semi-primary rings. Trans. Amer. Math. Soc., 95:466 -- 488, 1960. [5] J.-E. Bjork. Conditions which imply that subrings of semiprimary rings are semiprimary. J. Algebra, 19:384 -- 395, 1971. [6] Jan-Erik Bjork. Conditions which imply that subrings of artinian rings are artinian. J. Reine Angew. Math., 247:123 -- 138, 1971. [7] Nicolas Bourbaki. Elements of mathematics. Theory of sets. Translated from the French. Hermann, Publishers in Arts and Science, Paris, 1968. [8] Nicolas Bourbaki. General topology. Chapters 5 -- 10. Elements of Mathematics (Berlin). Springer-Verlag, Berlin, 1989. Translated from the French, Reprint of the 1966 edition. [9] Rosa Camps and Warren Dicks. On semilocal rings. Israel J. Math., 81(1-2):203 -- 211, 1993. [10] Friedrich Dischinger. Sur les anneaux fortement π-r´eguliers. C. R. Acad. Sci. Paris S´er. A-B, 283(8):Aii, A571 -- A573, 1976. SEMI-INVARIANT SUBRINGS 31 [11] Alberto Facchini. The Krull-Schmidt theorem. In Handbook of algebra, Vol. 3, pages 357 -- 397. North-Holland, Amsterdam, 2003. [12] Alberto Facchini, S¸ule Ecevit, and M. Tamer Ko¸san. Kernels of morphisms between inde- composable injective modules. Glasg. Math. J., 52(A):69 -- 82, 2010. [13] Alberto Facchini and Dolors Herbera. Local morphisms and modules with a semilocal endo- morphism ring. Algebr. Represent. Theory, 9(4):403 -- 422, 2006. [14] Alberto Facchini, Dolors Herbera, Lawrence S. Levy, and Peter V´amos. Krull-Schmidt fails for Artinian modules. Proc. Amer. Math. Soc., 123(12):3587 -- 3592, 1995. [15] E. Gregorio. Generalized Morita equivalence for linearly topologized rings. Rend. Sem. Mat. Univ. Padova, 79:221 -- 246, 1988. [16] Yukitoshi Hinohara. Note on non-commutative semi-local rings. Nagoya Math. J., 17:161 -- 166, 1960. [17] Max-Albert Knus, Alexander Merkurjev, Markus Rost, and Jean-Pierre Tignol. The book of involutions, volume 44 of American Mathematical Society Colloquium Publications. American Mathematical Society, Providence, RI, 1998. With a preface in French by J. Tits. [18] T. Y. Lam. Lectures on modules and rings, volume 189 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1999. [19] Eben Matlis. Injective modules over Noetherian rings. Pacific J. Math., 8:511 -- 528, 1958. [20] I. Reiner. Maximal orders, volume 28 of London Mathematical Society Monographs. New Series. The Clarendon Press Oxford University Press, Oxford, 2003. Corrected reprint of the 1975 original, With a foreword by M. J. Taylor. [21] C. Riehm and M. A. Shrader-Frechette. The equivalence of sesquilinear forms. J. Algebra, 42(2):495 -- 530, 1976. [22] Carl Riehm. The equivalence of bilinear forms. J. Algebra, 31:45 -- 66, 1974. [23] Louis H. Rowen. Finitely presented modules over semiperfect rings. Proc. Amer. Math. Soc., 97(1):1 -- 7, 1986. [24] Louis H. Rowen. Finitely presented modules over semiperfect rings satisfying ACC-∞. J. Algebra, 107(1):284 -- 291, 1987. [25] Louis H. Rowen. Ring theory. Vol. I, volume 127 of Pure and Applied Mathematics. Academic Press Inc., Boston, MA, 1988. [26] Louis Halle Rowen. Examples of semiperfect rings. Israel J. Math., 65(3):273 -- 283, 1989. [27] David J. Saltman. Lectures on division algebras, volume 94 of CBMS Regional Conference Series in Mathematics. Published by American Mathematical Society, Providence, RI, 1999. [28] Richard G. Swan. Induced representations and projective modules. Ann. of Math. (2), 71:552 -- 578, 1960. [29] P. V´amos. The dual of the notion of "finitely generated". J. London Math. Soc., 43:643 -- 646, 1968. [30] Peter V´amos. Decomposition problems for modules over valuation domains. J. London Math. Soc. (2), 41(1):10 -- 26, 1990. [31] Seth Warner. Topological rings, volume 178 of North-Holland Mathematics Studies. North- Holland Publishing Co., Amsterdam, 1993. Department of Mathematics, Bar-Ilan University, Ramat-Gan E-mail address: [email protected]
1703.10641	1	1703	2017-03-30T18:59:07	Formal affine Demazure and Hecke algebras of Kac-Moody root systems	[ "math.RA", "math.AG" ]	We define the formal affine Demazure algebra and formal affine Hecke algebra associated to a Kac-Moody root system. We prove the structure theorems of these algebras, hence, extending several result and construction (presentation in terms of generators and relations, coproduct and product structures, filtration by codimension of Bott-Samelson classes, root polynomials and multiplication formulas) that were previously known for finite root system.	math.RA	math	FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS BAPTISTE CALM`ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG Abstract. We define the formal affine Demazure algebra and formal affine Hecke algebra associated to a Kac-Moody root systems. We prove the structure theorems of these algebras, hence, extending several results and constructions (presentation in terms of generators and relations, coproduct and product structures, filtration by codimension of Bott-Samelson classes, root polyno- mials and multiplication formulas) that were previously known for finite root systems. 1. Introduction The nil-Hecke algebra and 0-Hecke algebra were constructed by Kostant-Kumar in [KK86, KK90] in the study of equivariant singular cohomology and equivari- ant K-theory of Kac-Moody flag varieties. These algebras are generated by the Demazure operators (also known as BGG operators or divided difference opera- tors) satisfying braid relation; they act on their duals by the convolution action of equivariant singular cohomology (resp. equivariant K-theory) on itself. In [HMSZ, CZZ1, CZZ2, CZZ3] these constructions were extended to an ar- bitrary algebraic oriented cohomology theory (corresponding to a one-dimensional commutative formal group law F ) of a usual flag variety, i.e. for finite root systems. The resulting algebra, called the formal affine Demazure and denoted DF , satisfies a twisted braid relation instead of the usual one and its dual is isomorphic to the respective oriented cohomology ring. In particular, the nil-Hecke algebra and the 0- Hecke algebra of Kostant-Kumar correspond to the additive and the multiplicative formal group law F respectively. In the present paper, we extend the previous constructions to an arbitrary Kac- Moody root system (determined by a generalized Cartan matrix), hence, completing the picture. This was originally a question raised by Shrawan Kumar. Note that the key result in this setup is the so-called structure theorem, which describes DF by its polynomial representation. To prove it in the additive and multiplicative case, Kostant-Kumar essentially relied on the study of zeros and poles of such polynomial functions. For a general F , polynomials turn into formal power series, so we no longer can look at zeros or poles. To overcome this difficulty and to prove the 2010 Mathematics Subject Classification. 20C08, 20G44, 14F43, 57T15. B. C. is support by the French Agence Nationale de la Recherche (ANR) under reference ANR-12-BL01-0005. K. Z. was partially supported by the NSERC Discovery Grant RGPIN-2015- 04469. C. Z. was supported by PIMS and NSERC grants of S. Gille and V. Chernousov in 2014, and would like to thank S. Kumar, T. Zhang and G. Zhao for helpful discussions, and to the Max Planck Institute for Mathematics for hospitality during the visit in May, 2014. 1 2 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG generalized version of the structure theorem (Theorem 4.1) we use the language of prime ideals and some general position arguments (see Section 3). Having the structure theorem in hands we extend most of the results and proper- ties concerning DF following the same proofs as in the finite case. For instance, we provide the presentation of DF in terms of generators and relations (Corollary 4.4), hence, generalizing the result of [Le16]. We introduce the coproduct structure on DF and, hence, the product on its dual D∗ F , hence, extending [CZZ1]. We define ∗, and show that the corresponding associated graded filtrations on DF and on DF rings are isomorphic to the nil-Hecke algebra and its dual, respectively (Proposi- tion 5.2). In Section 6 we introduce and study the formal affine Hecke algebra HF , extending definitions of [HMSZ] and [ZZ17]. In the last section, we discuss for- mal root polynomials and various multiplication formulas, hence, extending some results of [LZ16]. 2. Preliminaries 2.1. Generalized Cartan matrices and root data. We introduce notation and list basic properties of the root datum associated to a generalized Cartan matrix following [Ka90, Ku02]. We recall definition of the Demazure lattice due to [Le16]. Let A := (aij )1≤i,j≤n be a generalized Cartan matrix (i.e., aii = 2, −aij ∈ Z+ for all i 6= j, where Z+ is the set of nonnegative integers, and aij = 0 ⇔ aji = 0). Choose a triple (h, Π, Π∨), unique up to isomorphism, where h is a complex vector space of rank 2n − rank A, Π = {α1, . . . , αn} ⊂ h∗, Π∨ = {α∨ 1 , . . . , α∨ n} ⊂ h are linearly independent subsets of simple roots and simple coroots respectively, satisfying aij = hαj, α∨ i i. Such a triple is called the root datum corresponding to the matrix A. Let W be the Weyl group of the root datum, that is, the group generated by fundamental reflections si : λ 7→ λ − hλ, α∨ i iαi, λ ∈ h∗. It is a Coxeter group, and for i 6= j, the order mij of sisj is equal to 2, 3, 4, 6, ∞ when aij aji is 0, 1, 2, 3, ≥ 4, respectively. Let Φ = {w(αi) w ∈ W, 1 ≤ i ≤ n} denote the set of (real) roots. λ − hλ, α∨iα where α∨ = w(α∨ If α = w(αi), we define the reflection sα(λ) = i ). We then have sα = wsiw−1. Let Φ+ = Φ ∩ Pi Z+αi and Φ− = −Φ+ denote the sets of positive and negative (real) roots respectively. We have Φ = Φ+ ∐ Φ−. For each w ∈ W , define Φw = {α ∈ Φ+ w−1(α) ∈ Φ−}. The expression w = si1 . . . sil ∈ W is called reduced if l is minimal possible among all representations of w ∈ W as a product of the si and in this case l = ℓ(w) is called the length of w. We then have (a) For each w ∈ W , αi ∈ Π, we have ℓ(siw) > ℓ(w) ⇔ w−1(αi) ∈ Φ+ and ℓ(siw) < ℓ(w) ⇔ w−1(αi) ∈ Φ−, FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS3 (b) Φsi = {αi}, (c) For a reduced expression w = si1 . . . sil we have Φw = {αi1 , si1 αi2 , . . . , si1 · · · sil−1 αil }, in particular ℓ(w) = Φw. r = Pi Zα∨ Let Λr = Pi Zαi and Λ∨ i denote the root and coroot lattices, re- spectively. Let Λ be a formal Demazure lattice of [Le16, Def. 3.1], i.e., a finitely generated free abelian subgroup of h∗ such that hΛ, Λ∨ r i ⊂ Z and every simple root can be extended to a Z-basis of Λ, i.e., the coordinates of every simple root with respect to some Z-basis of Λ form a unimodular vector. Observe that all these lattices are invariant under the action by W . We refer to [Le16, §3] for their basic properties and examples. 2.2. Formal Demazure operators. We define formal Demazure algebra and De- mazure operators following [CPZ, HMSZ, CZZ1, Le16]. Let F be a one-dimensional commutative formal group law over a commutative ring R. We sometimes write x +F y = F (x, y), and denote by −F x ∈ R[[x]] the formal inverse of x. Examples of formal group laws include the additive formal group law Fa(x, y) = x + y, the multiplicative formal group laws Fm(x, y) = x + y − βxy for β ∈ R, and the universal formal group law Fu over the Lazard ring. Let Λ be the Demazure lattice and let R[[Λ]]F = R[[xλλ ∈ Λ]]/J, where J is the closure of the ideal generated by x0 and xλ+µ − F (xλ, xµ) with λ, µ ∈ Λ. R[[Λ]]F is called the formal group algebra [CPZ]. Example 2.1. (a) If F = Fa, then R[[Λ]]F ∼= S∗ symmetric algebra at the augmentation ideal. R(Λ)∧ is the completion of the (b) If F = Fm, then R[[Λ]]F ∼= R[Λ]∧ is the completion of the group ring. Observe that the construction of R[[Λ]]F is functorial with respect to homomor- phisms of rings R → R′, homomorphisms of lattices Λ → Λ′ and homomorphisms of formal group laws F → F ′. Moreover, the action of W on Λ induces an action of W on R[[Λ]]F by w(xλ) = xw(λ), λ ∈ Λ. There is an augmentation map ǫ : R[[Λ]]F → R, xλ 7→ 0, and let IF denote the kernel. We then have a filtration R[[Λ]]F = I 0 F ⊃ IF ⊃ I 2 F ⊃ · · · ⊃ I i F ⊃ · · · , and define I−i F = R[[Λ]]F for i > 0. The associated graded ring is denoted by GrR[[Λ]]F = M F /I i+1 I i F . i≥0 By [CPZ, Lemma 4.2], we have GrR[[Λ]]F ∼= S∗ that deg q = i and denote by (q)i its image via the map R(Λ). For each q ∈ I i F \I i+1 F , we say F /I i+1 I i F We define (q)j = 0 for j < deg q. ֒→ GrR[[Λ]]F ∼= S∗ R(Λ). Lemma 2.2. Assume that R has characteristic 0. Let α ∈ Φ+. (1) xα is irreducible in R[[Λ]]F for all positive root α. (2) If x ∈ R[[Λ]]F such that xαxβ x for some positive roots α 6= β, then xαx. 4 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG Proof. (1) Applying W we may assume that α = αi is some simple root. Let {ω1, . . . , ωm} be a basis of Λ extending αi. Then the coefficients aij in αi = Pn j=1 aijωj are coprime. Hence, the power series corresponding to xαi is irreducible in R[ω1, . . . , ωm]. (2) Again it suffices to assume that α is a simple root. Expressing β as a linear combination of simple roots, we get that xβ is regular in S/(xα), so the conclusion follows. (cid:3) For each v ∈ W , define cv = Qα∈Φv xα ∈ R[[Λ]]F . Lemma 2.3. If ℓ(sαv) < ℓ(v), then xα cv and x2 α ∤ cv. Proof. If ℓ(sαv) < ℓ(v), then v−1(α) ∈ Φ− by [Ku02, Lemma 1.3.13], hence α ∈ Φv. So xα cv. The second part follows since xα ∤ xβ for any α 6= β ∈ Φ+. (cid:3) Following [CPZ, Corollary 3.4] we define R-linear operators R[[Λ]]F → R[[Λ]]F as ∆α(u) = u − sα(u) xα , Cα(u) = καu − ∆α(u) = u x−α + sα(u) xα , where κα := 1 xα operator, and Cα is called a formal push-pull operator. They satisfy ∈ R[[Λ]]F . The operator ∆α is called a formal Demazure + 1 x−α ∆α ◦ ∆α = κα∆α = ∆ακα, Cα ◦ Cα = καCα ∆α(uv) = ∆α(u)v + sα(u)∆α(v), u, v ∈ R[[Λ]]F . Moreover, they are SWα-linear, where Wα is the subgroup of W generated by sα. For simplicity, we set x±i = x±αi , ∆i = ∆αi and Ci = Cαi . For a sequence I = (i1, . . . , il) we set ∆I = ∆i1 ◦ · · · ◦ ∆il . If E ⊂ [l] = {1, . . . , l}, we denote by IE the sequence consisting of ij with j ∈ E. According to [CZZ1, Lemma 4.8] we have ∆I (uv) = X E1,E2⊆[l] pI E1,E2∆IE1 (u)∆IE2 (v), E1,E2 ∈ IE1+E2−l pI F . 2.3. The formal affine Demazure algebra. We now recall the definition of the formal affine Demazure algebra associated to a generalized Cartan matrix following [Le16, §4]. Let S = R[[Λ]]F and let Q = S[ 1 xα α ∈ Φ+], be the localization of S at all positive real roots. Define the twisted group algebra QW = Q ⋊ R[W ] to be the left Q-module Q ⊗R R[W ] with basis {δw}w∈W , and define the product qδw · q′δw′ = qw(q′)δww′, q, q′ ∈ Q, w, w′ ∈ W. We set δα = δsα and 1 = δe. Define Xα := 1 xα (1 − δα) and Yα = κα − Xα = 1 x−α + 1 xα δα. FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS5 The element Xα ∈ QW is called a formal Demazure element, and Yα is called a formal push-pull element. We then have for any q ∈ Q, α ∈ Φ, w ∈ W Xαq = sα(q)Xα + ∆α(q), X 2 α = καXα = Xακα, δwXαδw−1 = Xw(α), Y 2 α = καYα, δwYαδw−1 = Yw(α). For simplicity, we set δi = δsi , Xi = Xαi and Yi = Yαi . For a sequence I = (i1, . . . , il) we set XI = Xi1 · · · Xil ; set X∅ = 1. Let Iv denote a reduced sequence of v ∈ W , i.e., Iv = (i1, . . . , il) for a reduced expression v = si1 . . . sil . The proofs of the following two lemmas are the same as of [CZZ1, Lemma 5.4] (see also [CZZ2, Lemma 3.2, 3.3]) Lemma 2.4. We have XIv = Pw≤v aX where the sum is taken over all w ≤ v with respect to the Bruhat order. v,wδw for some aX v,v = (−1)ℓ(v) 1 cv v,w ∈ Q, aX , v,wXIw for some bX v,w ∈ S, bX v,e = 1 and bX v,v = Moreover, we have δv = Pw≤v bX (−1)ℓ(v)cv. Lemma 2.5. We have YIv = Pw≤v aY v,wδw for some aY v,w ∈ Q and aY v,v = 1 cv . Moreover, we have δv = Pw≤v bY v,wYIw for some bY v,w ∈ S and bY v,v = cv. Corollary 2.6. The sets {XIw }w∈W and {YIw }w∈W form bases of QW as a left Q-module. Definition 2.7. Let DF be the R-subalgebra of QW generated by elements of S and the Demazure elements {Xα}α∈Φ. We call DF the formal affine Demazure algebra associated to the generalized Cartan matrix A and the Demazure lattice Λ. Observe that DF is generated by elements of S and Xi, i ∈ [n]. Indeed, we have δi = 1 − xiXi, so δw ∈ DF . Since any root α can be written as w(αi) for some w ∈ W and a simple root αi, and Xw(αi) = δwXiδw−1, the conclusion then follows. By definition the algebras QW and DF are functorial with respect to morphisms of rings R → R′, morphisms of formal group laws F → F ′, and morphisms of root data, i.e., morphisms of lattices φ : Λ → Λ′ sending roots (resp. coroots) to roots (resp. coroots), and such that φ∨(φ(α)∨) = α∨. 3. General position arguments and divisibility by positive roots The purpose of the present section is to extend [KK86, Lemma 4.9] (additive case F = Fa) and [KK90, Lemma 2.12] (multiplicative case F = Fm) to an arbitrary formal group law F . This is done in Proposition 3.3 which will be used in the proof of our main theorem (Theorem 4.1). Observe that in the additive (resp. multiplicative) case the ring S is the ring of (resp. Laurent) polynomials (non- complete version of Example 2.1), so it can be viewed as a ring of functions on the lattice Λ. The respective proofs by Kostant-Kumar then deal with analyzing zeros and poles of such functions. For a general F , the ring S becomes the ring of formal power series, so we no longer have functions on Λ as well as zeros and poles. To overcome this difficulty, we use the language of prime ideals instead. 6 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG Suppose R is a domain of characteristic 0. Let K denote its field of fractions. Recall that S = R[[Λ]]F . We set SK = K[[Λ]]F , Sa,K = K[[Λ]]Fa, S′ = S∗ K(Λ), Z = Spec(S), ZK = Spec(SK), Za,K = Spec(Sa,K), Y = Spec(S′) ∼= An K. There are induced actions of W on S, SK, Sa,K and S′, hence, on Z, ZK, Za,K and Y . An element xλ in Sa,K (or in S′) will be simply denoted by λ. There exists an isomorphism h : Fa → F of formal group laws over K, i.e., h(x) ∈ K[[x]] is such that h(x) +F h(y) = h(x + y). It induces an isomorphism Lemma 3.1. The following diagram commutes H : SK → Sa,K, xαi 7→ h(αi). Z w· Z φ φ H H ZK w· ZK ψ ψ Za,K w· Za,K Y / Y w· where w· is the action of w ∈ W on corresponding spectrum, φ is the map induced by the embedding R ֒→ K, and ψ is the map induced by the embedding S′ ֒→ Sa,K (observe that Sa,K is the completion of S′ along the augmentation ideal). Proof. The commutativity of the left and right squares are obvious. Concerning the middle square, we have w(H(xλ)) = w(h(λ)) = h(w(λ)) = H(xw(λ)) = H(w(xλ)). (cid:3) We define the w-invariant subset Z w := {x ∈ Z w(x) = x}, and similarly define a,K, Y w. For a commutative ring R′ and an element r ∈ R′, we denote by K, Z w Z w V (r) the set of prime ideals of R′ containing r. Then we have φ−1(V (xα)) = V (xα), H(V (α)) = H(V (h(α))) = V (xα), ψ−1(V (α)) = V (α), φ−1(Z w) = Z w K, H(Z w a,K) = Z w K, ψ−1(Y w) = Z w a,K. The identity V (h(α)) = V (α) follows from the fact that h(x) can be written as h(x) = xg(x) for some invertible g(x) ∈ K[[x]]. We will use the following general position-type result Lemma 3.2. Suppose R is a domain of characteristic 0. For any root α ∈ Φ, the set Pα := V (xα)\(cid:0)[∪β∈Φ, β6=αV (xβ )] ∪ [∪w6=e,sα Z w](cid:1) is a dense subset of V (xα) ⊂ Z. Proof. It suffices to prove that φ−1(Pα) = V (xα) ⊂ ZK. Indeed, since φ is dominant, which means it maps dense subsets to dense subsets, φ(φ−1(Pα)) = φ(V (xα)) = V (xα) ⊂ Z, and we know φ(φ−1(Pα)) ⊂ Pα, so Pα = V (xα). The map H is an isomorphism, and ψ is flat, so the inverse image via ψ of a dense subset is dense ([Mi80, Corollary I.2.8]). Therefore, it is enough to show that α := V (α)\(cid:0)[∪β6=αV (β)] ∪ [∪w6=e,sα Y w](cid:1) P ′ o o o o / / o o o o / FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS7 is dense in V (α) ⊂ Y . We can assume that K = Q. By functoriality of formal group algebras we may assume that Λ = Λr and α = α1 is a simple root. Then V (β) is the hyperplane in Y = An Q orthogonal to β. Consider the positive half-hyperplane V (α1)+ := {¯a = (0, a2, . . . , an) ∈ Y ai > 0 for i ≥ 2} ⊂ V (α1). We claim that P ′ dense in V (α1). α1 contains V (α1)+ which would immediately imply that P ′ α1 is Indeed, since β = c1α1 + . . . + cnαn where ci's are either all positive or negative, β(¯a) = c2a2 + . . . + cnan 6= 0 unless c2 = . . . = cn = 0 (that is β = α1). Hence, V (α1)+ ∩ V (β) = ∅ for all β 6= α1. Finally, by [Ka90, Proposition 3.12.(a)] the stabilizer W¯a := {w ∈ W w(¯a) = ¯a} is generated by simple reflections it contains, so W¯a = {e, s1}. Hence, V (α1)+ ∩ Y w = ∅ for all w 6= e, s1. (cid:3) Consider an action of the twisted group algebra QW on Q via qδw · q′ = qw(q′), q, q′ ∈ Q, w ∈ W. As before, let XIw denote the product of Demazure elements indexed by a reduced expression for w ∈ W . Recall that (see Corollary 2.6) XIw , w ∈ W form a basis of QW as a left Q-module. We obtain the following generalization of [KK86, Lemma 4.10] Proposition 3.3. Let z = Pw∈W, ℓ(w)≤k pwXIw , pw ∈ S and let α ∈ Φ+. If z · S ⊆ xαS, then xα pw for all w. Proof. We may assume pw 6= 0 for some w with ℓ(w) = k. Set z+ = z + δαz and z− = z − δαz so that 2z = z+ + z−. Since xα x−α is invertible in S, we have (δαz) · S ⊆ δα · (xαS) = x−αS = xαS. By Lemma 2.2 xα is irreducible, so xα 2f ⇒ xα f for any f ∈ S. Hence, it suffices to prove the lemma for z+ and z−, i.e., we may assume δαz = ±z. Moreover, it is enough to show that xα 2pw for all w. Expressing z in terms of the canonical basis {δw} we obtain z = X pwXIw = X ℓ(w)≤k ℓ(w)≤k qwδw, qw ∈ Q. Choose an element w0 of length k such that pw0 6= 0. By Lemma 2.4 qw0 = 1 pw0. cw0 Since δαz = ±z, we have ℓ(sαw0) < ℓ(w0), hence by Lemma 2.3, cw0 = xαc′ where c′ ∈ S is a product of some xβ with β ∈ Φ+\{α}. On the other hand, δαz = ±z implies that qsαw0 = ±sα(qw0 ). Hence, we can write x−αsα(c′) δsαw0 + z′, where z′ := xαc′ δw0 ± sα(pw0 ) z = pw0 X qwδw. ℓ(w)≤k, w6=w0,sαw0 We claim that xα 2pw0. Indeed, it suffices to prove that for any p ∈ V (xα) ( Z = Spec S we have 2pw0 ∈ p. By Lemma 3.2, the set Pα = V (xα)\(cid:0)[∪β6=αV (xβ )] ∪ [∪w6=e,sα Z w](cid:1) is dense in V (xα), so it suffices to prove that for any p ∈ Pα we have 2pw0 ∈ p. 8 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG Let p ∈ Pα. By definition, w(p) 6= p for any w 6= e, sα. Hence, for any w 6= 0 (p) 6= w−1(p) and w−1 w0, sαw0 we have w−1 0 sα(p) 6= w−1(p) which implies that w−1(p)\(cid:0)w−1 0 sα(p)(cid:1) 6= ∅. The latter means that for any w 6= w0, sαw0 0 (p) ∪ w−1 there exists rw ∈ S such that w(rw) ∈ p, w0(rw) 6∈ p and sαw0(rw) 6∈ p. We define r = Y rnw +1 w , qw 6=0 and w6=w0,sαw0 where nw is the order of xα in the denominator of qw. Let S be the localization of S at all xβ with β ∈ Φ+\{α}, then Q = S[ 1 xα ]. sα(c′) ∈ S. Since xβ 6∈ p for any such β, p S := p S is a proper Observe that 1 c′ , prime ideal of S. The assumption z · S ⊆ xαS implies that z · r ∈ (xα) ⊂ p S. 1 (i) Suppose that δαz = −z. We obtain xαz · r = xα(z · r) ∈ p S, and η := [ pw0 c′ δw0 − xαsα(pw0 ) x−αsα(c′) δsαw0 ] · r = pw0 c′ w0(r) − xαsα(pw0 ) x−αsα(c′) sαw0(r) ∈ S. Note that η = xαz ·r − xαz′ ·r, so xαz′ ·r ∈ S. By Lemma 3.4, we have xαz′ ·r ∈ p S, which implies that η ∈ p S. Denote ζ := pw0 w0(r) + 1 is divisible by xα ∈ p, and sα(ζ) − ζ = −xα∆α(ζ) ∈ p, so 2ζ ∈ p S. Since w0(r) 6∈ p, we get 2pw0 ∈ p. sα(ζ) ∈ p S. Note that xα x−α , then η = ζ − xα x−α c′ (ii) Suppose that δαz = z. Similarly, we obtain z · (w−1 0 (xα)r) ∈ p. Moreover, η′ := [ pw0 xαc′ δw0 + sα(pw0 ) x−αsα(c′) δsαw0 ] · (w−1 0 (xα)r) = pw0 c′ w0(r) + sα(pw0 ) sα(c′) sαw0(r) ∈ S. Therefore, z′ · (w−1 z′ · (w−1 0 (xα)r) = z · w−1 0 (xα)r − η′ ∈ S. Similar to Lemma 3.4, we have 0 (xα)r) ∈ p S. Therefore, η′ ∈ p S ⇒ 2pw0 c′ w0(r) ∈ p S ⇒ 2pw0 ∈ p. (cid:3) Lemma 3.4. With the notation of the proof of Lemma 3.3, we have xαz′ · r ∈ p S. Proof. By definition, for w 6= w0, sαw0 we have qwδw · r = qw Y w(rnv +1 v ). v6=w0,sαw0 If xα does not divide the denominator of qw, then qw ∈ S, so qwδw · r ∈ p S since w(rw) ∈ p. If xα divides the denominator of qw, then by definition of r the numerator of qwδw · r belongs to p ⊂ p S. Since Q = S[ 1 xα ], we can rewrite xαz′ · r ∈ S as xαz′ · r = m X j=0 q′ j xj α ∈ S, j ∈ p S for all j, and xα ∤ q′ q′ j for j > 0. If m > 0, then reducing the expression to the common denominator xm tain that xα Pm conclusion follows. α , we ob- m, a contradiction. So m = 0 and the (cid:3) , hence xα q′ jxm−j j=0 q′ α FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS9 4. The formal affine Demazure algebra and its dual In this section we prove the structure theorem (Theorem 4.1) which the key result of this paper. It allows to extend most of the properties and facts concerning the algebra DF and its dual D∗ F from the finite case to the case of an arbitary generalized Cartan matrix. As in the previous section, we consider the QW action on Q. It induces an action of the formal affine Demazure algebra DF on S = R[[Λ]]F via Xi · q = 1 xi (1 − δi) · q = ∆i(q), q ∈ S, so we have DF · S ⊆ S. The following theorem generalizes [KK86, Theorem 4.6], [KK90, Theorem 2.9], and [Le16, Proposition 4.13] to the context of an arbitrary formal group law and an arbitrary Demazure lattice. Theorem 4.1. Let R be a domain of characteristic 0. Let DF be the formal affine Demazure algebra defined for a given generalized Cartan matrix, Demazure lattice and a formal group law over R. Then DF = {z ∈ QW z · S ⊆ S}. Proof. By definition we have DF ⊆ {z ∈ QW z · S ⊆ S}. To prove the opposite inclusion let z ∈ QW be such that z · S ⊆ S. Then by Lemma 2.4, {XIw }w∈W is a p Pℓ(w)≤k pwXIw with pw ∈ S and p is a product basis of QW , so we can write z = 1 of (possibly repeated) xα, α ∈ Φ+. It suffices to assume that p is irreducible, and by Lemma 2.2, we can assume that p = xα. Then Proposition 3.3 implies that xα pw for all w. Therefore, z ∈ DF . (cid:3) As an immediate consequence we obtain Corollary 4.2. The set {XIw }w∈W forms a basis of DF as a left (or right) S- module. Proof. The set {XIw }w∈W spans DF as a left S-module. It follows from Corol- lary 2.6 that this set is linearly independent. (cid:3) Based on Theorem 4.1, the following corollaries hold, whose proofs are exactly the same as in the finite case. First, we obtain a residue construction of the algebra DF for an arbitrary general- ized Cartan matrix and a Demazure lattice. Such construction was first mentioned in [GKV] for elliptic curves and then proved in the finite case in [ZZ17, §4]. Corollary 4.3. We have DF = { X w∈W where Qα = S[ 1 xβ awδw ∈ QW xαaw ∈ Qα and aw + asαw ∈ Qα, ∀α ∈ Φ+} β ∈ Φ+, β 6= α] denotes the respective localization of S. Next, we obtain a uniform description of all relations between generators in DF . Observe that in the finite case explicit relations were given in [HMSZ, Proposition 6.8] and [CZZ1, Lemma 7.1]; for a hyperbolic formal group law and a generalized 10 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG Cartan matrix it was given in [Le16, Example 4.12] based on the explicit presenta- tion of [HMSZ]. Corollary 4.4. If Iw and I ′ w are two reduced expressions of w, then XIw − XI ′ w = X pw,Iv XIv , pw,Iv ∈ S. v<w, ℓ(v)≤ℓ(w)−2 In particular, if (sisj)mij = 1 and w = sisjsi · · · , then } mij times XiXjXi · · · } mij times {z − XjXiXj · · · } mij times {z = v<w, ℓ(v)≤ℓ(w)−2 {z X pvXIv , pv ∈ S. Observe that if F = Fa or Fm, then all pv vanish ([KK86, Proposition 4.2] and [KK90, Proposition 2.4]) and we obtain the usual braid relation. We then identify DF with a subring of R-linear operators on S, that is Corollary 4.5. The map DF → EndR(S), Xi 7→ ∆i is injective. Proof. Let z ∈ DF be in the kernel. Write z = Pℓ(w)≤k pwXIw with pw ∈ S. If z · S = 0, then z · S ⊆ xαS. So by Lemma 3.3, xα pw for any α and any w. Then z′ := 1 . Recursively, we see xα that pw = 0 for all w. (cid:3) z satisfies the same condition as z does, so xα pw xα Example 4.6. Consider the additive case, i.e. F = Fa. In this case, XIw does not depend on the choice of the reduced decomposition Iw of w, so we denote it simply by Xw. Let Qa = S∗ w}w∈W . The map S∗ R(Λ) → R[[Λ]]Fa induces a map between twisted group algebras and, hence, an embedding α α ∈ Φ+] and Qa,W = Qa ⋊ R[W ] with basis {δa R(Λ)[ 1 N ֒→ Da with xw 7→ (−1)ℓ(w)Xw, where N is the nil (affine) Hecke algebra with basis {xw}w∈W of [KK86, §4] and Da denotes DFa . Moreover, we have R[[Λ]]Fa ⊗S ∗ R(Λ) N ≃ Da as R[[Λ]]Fa-modules. Similar to Lemma 2.4, we obtain xw = X aw,vδv, δw = X bw,vxv, with bw,v ∈ S∗ R(Λ), aw,v ∈ Qa. v≤e The coefficient aw,v = (−1)ℓ(w)aX v≤w w,v corresponds to cw−1,v−1 in [KK86, (4.3)]. We now turn to the study of the dual of the algebra DF . As in [CZZ1] we, first, introduce a coproduct structure on QW . We view QW ⊗Q QW as the tensor product of left Q-modules, and define △ : QW → QW ⊗Q QW , qδw 7→ qδw ⊗ δw. The counit is ǫ : QW → Q, qδw 7→ q. We define a product structure on QW ⊗Q QW by (z1 ⊗ z2) · (z′ 1 ⊗ z2z′ 2. 1 ⊗ z′ 2) = z1z′ By [CZZ1, Proposition 8.10] △ is a ring homomorphism, so QW is a cocommutative coalgebra in the category of left Q-modules. FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS11 Remark 4.7. We can also view the counit as ǫ : QW → Q, z 7→ z · 1 where · is the action of QW on Q. In particular, it implies that ǫ(zq) = z · q for z ∈ QW and q ∈ Q. Let Q∗ W be the dual of QW , i.e., Q∗ W = HomQ(QW , Q). By definition, it is a Q-module by (qf )(z) = qf (z), q ∈ Q, z ∈ QW , f ∈ Q∗ W . W can be written as f = P′ Indeed, Q∗ W is equal to the left Q-module Hom(W, Q), the set of maps (of sets) from W to Q. Define {fv}v∈W ⊂ Q∗ W by fw(δv) = δv,w (the Kronecker symbol), w∈W qwfw with qw = f (δw), where P′ then each f ∈ Q∗ denotes a possibly infinite (formal) sum. Since {XIw }w∈W is another basis of QW , we could define {X ∗ (XIv ) = δw,v. Then any f ∈ Q∗ W can be Iw written as f = P′ w∈W f (XIw )X ∗ Iw Lemma 4.8. [CZZ1, Proposition 9.5] For any sequence I = (i1, ..., ik), we have }w∈W such that X ∗ Iw . △(XI ) = X E1,E2⊆[k] pI E1,E2XIE1 ⊗ XIE2 , where pI E1,E2 ∈ Q. Lemma 4.9. [CZZ1, Theorem 9.2] The coproduct on QW induces a coproduct struc- ture on DF , which makes it into a cocommutative coalgebra with counit ǫ : DF → S induced by ǫ : QW → Q. Moreover, ǫ(XIw ) = δe,w. Consider the dual D∗ module. Then the coproduct structure on DF makes D∗ with unit 1 = ǫ ∈ D∗ F , and each f ∈ D∗ D∗ Moreover, as in [CZZ2, Lemma 10.2], we have F . The set of elements {X ∗ Iw F can be uniquely written as f = P′ F = HomS(DF , S) where DF is considered as a left S- F into a commutative ring }w∈W can also be viewed as in with qw ∈ S. w∈W qwX ∗ Iw Lemma 4.10. The algebra D∗ F can be identified with the subset {f ∈ Q∗ W f (DF ) ⊆ S}. In particular, 1 = X ∗ Ie ∈ D∗ F . Proof. The first part follows as in the proof of [CZZ2, Lemma 9.2]. To prove the second part, we have ǫ(XIw ) = XIw · 1 = δw,e ∈ S, so 1(X ∗ ) = δw,e. That is, Iw 1 = X ∗ (cid:3) Ie . Remark 4.11. In the case F = Fa, if f , g ∈ Q∗ W are such that f (XIw ) = 0 and g(XIw ) = 0 for all but finitely many w ∈ W , then f g(XIw ) = 0 for all but finitely many w ∈ W . This implies that the dual of the nil Hecke algebra N (see Example 4.6) has a product structure. For a general F it is unknown whether such finiteness condition holds. Con- sequently, if one restricts to the S-submodule of D∗ F consisting of f such that f (XIw ) = 0 for all but finite many of w ∈ W , then this submodule maybe not be a ring. This is one of the reasons why the definition of Ψ in [KK90, Definition 2.19] (the case F = Fm) does not include the finiteness condition. 12 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG 5. Filtrations on the Demazure algebra and its dual In this section we introduce and study filtrations on DF and its dual D∗ F . The filtration on D∗ F come from the filtration by the codimension of Schubert basis and generalizes the one for N ∗ of [KK90, Definition 2.28]. Our main result (Theo- rem 5.6) generalizes [KK90, Proposition 2.30]. Set D(i) F to be the R-submodule of DF generated by qXI where q ∈ S and deg q − I ≥ i. It gives a filtration · · · D(−i) F ) · · · ) D(0) F ) · · · ) D(i) F · I j F ) · · · . As in [CPZ, Proposition 3.6] we conclude that D(i) {XIw }w∈W is a basis of DF , the submodule D(i) and deg q − ℓ(w) ≥ i. F ⊂ I i+j F . Moreover, since F is spanned by qXIw with q ∈ S Lemma 5.1. (a) For any q ∈ S and a sequence I, we have XIq = X E⊆I φI,E(q)XE, where φI,E(q) ∈ S has degree deg q − I + E. (b) For any i, j we have D(i) F · D(j) F ⊆ D(i+j) F . Proof. (a) The identity follows from [CZZ1, Lemma 9.3]. If I = (i), then Xiq = si(q)Xi + ∆i(q) with deg si(q) = deg q and deg ∆i(q) = deg q − 1. The formula for the degree then follows by induction on I. (b) It suffices to show that pXI qXJ ∈ D(i+j) F if deg p−I ≥ i and deg q −J ≥ j. By part (a) we have XI q = X E⊆I φI,E(q)XE, where deg φI,E(q) = deg q − I + E. So pXI qXJ = PE⊆I pφI,E(q)XEXJ where deg p + deg φI,E(q) − E − J = deg p + deg q − I − J ≥ i + j. Hence, pXI qXJ ∈ D(i+j) F . (cid:3) (i/i+1) F Let GrDF = Li∈Z D , be the associated graded module. By Lemma 5.1 it respects the product structure and, hence, defines a ring. Consider a map ηi : D(i) qXIw 7→ (−1)ℓ(w)(q)i+ℓ(w)xw, q ∈ S, , where D (i+1) (i) F /D F (i/i+1) F = D (5.1) F → N , where N is the nil affine Hecke algebra of [KK86] (cf. Example 4.6). By definition, ηi(qXIw ) = 0 if deg q − ℓ(w) = i + 1. So it factors through ηi : D(i/i+1) → N . F Proposition 5.2. (cf. [CPZ, Proposition 4.8]) The map ⊕ηi : GrDF → N is an isomorphism of GrR[[Λ]]F ∼= S∗ R(Λ)-modules, and it is also an isomorphism of rings. Proof. It follows from Corollary 4.5 that DF is isomorphic to the subalgebra of EndR(S) generated by S and by ∆i, 1 ≤ i ≤ l. As in the proof of [CPZ, Propo- sition 4.8], we have an isomorphism GrDF ≃ N of GrR[[Λ]]F ≃ S∗ R(Λ)-modules. FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS13 Similar to the [CPZ, Proposition 4.4], the action on GrR[[Λ]]F of the class of qXIw in GrDF is the same as the action of (−1)ℓ(w)(q)i+ℓ(w)xw on S∗ R(Λ). In particular, it shows that ηi preserves the product. Therefore, GrDF ≃ N as rings. (cid:3) We define a filtration on D∗ F as follows: for i ≥ 0 we set (D∗ F )(i) = {f ∈ D∗ F f (δw) ∈ I i F for all w ∈ W }. Then D∗ F = (D∗ F )(0) ) · · · ) (D∗ F )(i) ) (D∗ F )(i+1) ) · · · , and since △(δw) = δw ⊗ δw, we have F )(i)(D∗ (D∗ F )(i′) ⊆ (D∗ F )(i+i′). F = Li≥0(D∗ Let GrD∗ associated graded ring. F )(i/i+1), where (D∗ F )(i/i+1) = (D∗ F )(i)/(D∗ F )(i+1), be the Lemma 5.3. (a) For any v = si1 . . . sik , express δv = (1 − xi1 Xi1 ) . . . (1 − xik Xik ) = X pI XI . I⊂{i1,...,ik} Then pI ∈ S and deg pI ≥ I. (b) For any v, express δv = Pw≤v bX Then deg bX v,w ≥ ℓ(w). Moreover, (bX v,wXIw as in Lemma 2.4. v,w)ℓ(w) = (−1)ℓ(w)bv,w ∈ S∗ R(Λ) where bv,w was defined in Example 4.6. Proof. (a) If k = 1, then δi1 = 1 − xi1 Xi1 , so the conclusion holds. Now suppose it holds for k. Then δv = P pI XI with deg pI ≥ I. Consider w = sjv with ℓ(w) > ℓ(v). Then we obtain sj(pI )δj XI = X δw = δsj v = δj X pI XI = X sj(pI )(1 − xjXj)XI I I = X sj(pI )XI − X I I sj(pI )xj X{j}∪I . Since deg sj(pI ) ≥ I and deg[sj(pI )xj ] ≥ I + 1, the conclusion follows. (b) From part (a) we can write δv = PI pI XI with deg pI ≥ I. For each I, we have XI = Pℓ(w)≤I p′ I,wXIw with p′ δv = X I,w ∈ S. Therefore, X I,wXIw , pI p′ w I≥ℓ(w) so bX v,w = PI≥ℓ(w) pI p′ I,w. Since deg pI p′ I,w ≥ deg pI ≥ I ≥ ℓ(w), we have deg bX v,w ≥ ℓ(w). v,wXIw is mapped to δA To prove the last part, consider the map D → N defined in (5.1), then δv = Pw≤v bX v,w)ℓ(w)xw in N because this map preserves product. Since δv = Pw≤v bv,wxw in the twisted group algebra of [KK86] and the set {xw}w∈W is linearly independent, so bv,w = (−1)ℓ(w)(bX v,w)ℓ(w). (cid:3) v = Pw≤v(−1)ℓ(w)(bX (0/−1) F 14 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG Lemma 5.4. (a) The submodule (D∗ qX ∗ Iw (b) The quotient (D∗ such that deg q + ℓ(w) ≥ i. deg q + ℓ(w) = i. F )(i) is (possibly infinitely) spanned by elements F )(i/i+1) is finitely spanned by elements qX ∗ Iw such that Proof. It follows from Lemma 5.3.(2) that qX ∗ Iw have ∈ (D∗ F )(deg q+ℓ(w)). Moreover, we qX ∗ Iw (δw) = qbX F )(j) for j > deg q + ℓ(w). w,w = (−1)ℓ(w)qcw ∈ I deg q+ℓ(w) F So qX ∗ Iw 6∈ (D∗ \I deg q+ℓ(w)+1 F . (cid:3) The following corollary shows that the filtration agrees with the one defined in [CPZ, §7.4]. Corollary 5.5. We have (D∗ F )(i) = {f ∈ D∗ F f (D(−i+1) F ) ⊆ IF }. Proof. By Lemma 5.4 (D∗ By definition, D(−i+1) pX ∗ (qXIv ) = pqδw,v. Iw pX ∗ (qXIv ) ∈ IF if and only if j ≥ i. Iw F )(j) is spanned by elements qX ∗ Iw with deg q + ℓ(w) ≥ j. is spanned by pXIv with deg p − ℓ(v) ≥ −i + 1. We have If w = v, then deg(pq) = deg p + deg ≥ j − i + 1. So (cid:3) F Let Qa,W be the twisted group algebra in Example 4.6 with basis {δa a,W be its dual. Then the S∗ R(Λ)-dual N ∗ of N is contained in Q∗ Q∗ {xw}w∈W is a basis of N , it defines a dual basis {x∗ rings w}, and a,W . Since w} in N ∗. We define a map of φi : (D∗ F )(i) → Q∗ a,W , φi(f )(δa w) = (f (δw))i, If f ∈ (D∗ F )(i+1), then f (δw) ∈ I i+1 F , so (f (δw))i = 0. Therefore, φi induces a map φ′ i : (D∗ F )(i)/(D∗ F )(i+1) → Q∗ a,W . We now ready to state and to prove the main result of this section Theorem 5.6. (cf. isomorphism GrD∗ F [KK90, Proposition 2.30]) The map φ := ⊕i≥0φ′ i is a ring ∼= N ∗. Proof. First, we show that φ is a ring homomorphism. Suppose f ∈ (D∗ and g ∈ (D∗ F )(j+1), then F )(j)/(D∗ F )(i)/(D∗ F )(i+1) φ(f g)(δa w) = (f g(δw))i+j = (f (δw))i(g(δw))j = φ(f )(δa w) · φ(g)(δa w) = [φ(f )φ(g)](δa w). So φ(f g) = φ(f )φ(g). Let f = P′ ∈ (D∗ w∈W qwX ∗ Iw F )(i+1) with qw ∈ S. Then by Lemma 5.4, we know that ℓ(w) = i−deg qw ≤ i, so it is a finite sum. Therefore, it suffices to show that φ′ w for any q ∈ S such that deg q + ℓ(w) = i. Recall u such that Pu av,ubu,w = from Example 4.6 that in N , we have xv = Pu av,uδa ) = (−1)ℓ(w)(q)deg qx∗ F )(i)/(D∗ i(qX ∗ Iw FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS15 δv,w. In DF , by Lemma 5.3 we have δu = Pw′ bX (−1)ℓ(w′)bu,w′. So u,w′XIw′ such that (bX u,w′)ℓ(w′) = φ′ i(qX ∗ Iw )(xv) = φ′ i(qX ∗ Iw )(X av,uδa u) = X av,uφi(qX ∗ Iw )(δa u) u u av,w(qX ∗ = X u = X av,u(qbX u,w)i = X u u = (−1)ℓ(w)(q)deg qδv,w. Iw (δu))i = X av,u(qX ∗ Iw (X bX u,w′XIw′ ))i u w′ (q)deg qav,u(−1)ℓ(w)bu,w Therefore, φ′ phism. i(qX ∗ Iw ) = (−1)ℓ(w)(q)deg qx∗ w and ⊕i≥0φ′ i : GrD∗ F → N ∗ is an isomor- (cid:3) 6. Formal affine Hecke algebras In this section we define the formal affine Hecke algebra for a generalized Cartan matrix and a Demazure lattice following [HMSZ] and [ZZ17]. We state the structure theorem, whose proof is similar to the one for the formal affine Demazure algebra. Fix a free abelian group Γ of rank 1 generated by γ, and let (SF )′ := R[[Γ ⊕ Λ]]F . W )′ be the respective twisted formal W )′ = (QF )′ ⋊R R[W ]. For each root α, define Define (QF )′ = (SF )′[ 1 xα group algebra defined over R, i.e., (QF the element α ∈ Φ], and let (QF T F α := xγXα + δα ∈ (QF W )′. If the formal group law F is clear from the context, we will write S′ = (SF )′, Q′ = (QF )′, Q′ αi for simplicity. We have δwTαiδw−1 = Tw(αi). W )′ and Ti = T F W = (QF For a sequence I = (i1, ..., il) in [n], we denote TI = Ti1...il = Ti1 · ... · Til. For each w, we choose a reduced sequence Iw. Then TIw depends on the choice of Iw unless F is additive or multiplicative. It is straightforward to show that T 2 i = xγκiTi + 1 − xγκi, Tiq − si(q)Ti = xγ ∆i(q), q ∈ Q. Definition 6.1. Define the formal affine Hecke algebra HF to be the R-subalgebra of Q′ W generated by elements of S and Ti, i ∈ [n]. Note that the definition of HF depends on the choice of simple roots, since δi 6∈ HF and Tw(αi) := δwTiδw−1 may not belong to HF . Lemma 6.2. For any reduced sequence Iw we have TIw = X qIw,vδv with qv ∈ Q′, and qw = Y v≤w α∈Φw xα − xγ xα . + xα−xγ Proof. Note that Tα = xγ xα of [CZZ1, Lemma 5.4]. Lemma 6.3. Assume that R is a domain of characteristic 0. Let z = Pℓ(w)≤k pwTIw , pw ∈ S′, be such that z · S′ ⊆ xαS′. Then xαpw for all w. δα, so the conclusion follows similar to the proof (cid:3) xα 16 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG Proof. The proof is similar to that of Lemma 3.3. The only difference is that, if we write qwδw, qw ∈ Q′, z = X pwTIw = X ℓ(w)≤k ℓ(w)≤k in the case ℓ(w) = k, where dw := Qw∈Φw then qw = pw dw (xα − xγ). For any cw root α and w ∈ W , we have V (xα) 6⊂ V (dw), where V (q) denotes set of prime ideals of S′ containing q. So in the proof of Lemma 3.3, we may assume that p ∈ V (xα)\ ∪w∈W V (dw). (cid:3) We then obtain the following analogue of the structure theorem Theorem 6.4. Assume that R is a domain of characteristic 0. Then we have HF = {z = X w qwTw ∈ QW qw ∈ Q′ and z · S′ ⊂ S′}. Moreover, HF is a free S′-module with basis {TIw}w∈W . Proof. It follows from [HMSZ, Proposition 6.8] that each element of HF can be written as a linear combination z = Pw∈W qwTIw with qw ∈ (QF )′. Moreover, since Ti · S′ ⊆ S′, so HF · S′ ⊆ S′. Therefore, HF ⊆ {z = X w∈W qwTIw qw ∈ Q′, z · S′ ⊆ S′}. To prove the opposite inclusion, we replace S by S′ and XIw by TIw in the proof of Theorem 4.1, and use Lemma 6.3. Finally, similar to Lemma 2.4, we show that {TIw }w∈W are linearly independent. Moreover, Lemma 6.3 also shows that if z = Pw qwTIw ∈ HF , then qw ∈ S′. So {TIw} is a S′-basis of HF . (cid:3) The proof of the following corollary is similar to the finite case proven in [Zh15]. Corollary 6.5. The ring homomorphism HF → EndRF (S′) is injective, and the centre of HF is (S′)W . 7. Formal root polynomials and the Pieri-type formula In the present section we generalize the notion of a formal root polynomial introduced in [LZ16] and discuss the multiplication rule in D∗ F . Let Λ be a Demazure lattice and let S and Q be the associated formal group algebra and its localization. Following [LZ16, §2] consider the ring QW [[Λ]] := S ⊗R QW , where the elements of S on the left (denoted by y's) commute with the elements of QW . Given w ∈ W and a reduced word Iw = si1 . . . sil we set RX Iw := l Y k=1 ik (si1 . . . sik−1 (αik )), where hX hX i (λ) = 1 − y−λXi, and RY Iw := l Y k=1 ik (si1 . . . sik−1 (αik )), where hY hY i (λ) = 1 − yλYi. FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS17 Consider the evaluation map ev : QW [[Λ]] → QW induced by yλ 7→ x−λ. Then by the same proof as of [LZ16, Lemma 3.3]) we obtain ev(RX Iw ) = δw and ev(RY Iw ) = θwδw, where θw = Y θ(α), θ(λ) = −x−λ xλ . α∈Σw Moreover, by the same arguments as in [LZ16, §3.2] we obtain that RIw is inde- pendent of a choice of the reduced word Iw iff the formal group law F is hyperbolic (i.e., F (x, y) = x+y−µ1xy 1+µ2xy and R = Z[µ1, µ2]) and 2 is regular in R. Indeed, if the Coxeter exponent mij = 2, 3, 4, 6, we apply the same arguments as in [LZ16, Prop 3.2]. If mij is not from that list, then there is no relation between si and sj which does not affect RIw . The elements RX of QW [[Λ]] will be called formal root polynomials. w = RX Iw w = RY Iw and RY As in [LZ16, §4] the coefficients K X(Iv, w) (resp. K Y (Iv, w)) in the expansions w = X RY K Y (Iv, w)YIw , RX w = X K X(Iv, w)XIw v≤w v≤w satisfy the following Lemma 7.1. (cf. [LZ16, Thm. 4.4]) Suppose that F is a hyperbolic formal group law. We have w,Iv = τ (θwev(K Y (Iv, w))) and bX bY w,Iv = τ (ev(K X (Iv, w))), where τ is an involution xλ 7→ x−λ and bw,Iv are coefficients in the expansion: δw = X w,Iv YIv = X bY bX w,Iv XIv . v≤w v≤w Similar to [LZ16, Prop. 4.7] we obtain the following formula for the multiplica- tion: Lemma 7.2. We have in D∗ F Iu · Y ∗ Y ∗ Iv = X pIw Iu,Iv Y ∗ Iw , w≥u,v where the coefficients pIw Iu,Iv satisfy (7.1) pIw Iu,Iv = 1 bY w,Iw (bY w,IubY w,Iv − X pIt Iu,Iv bY w,It). u,v≤t<w Indeed, pIw Iu,Iv = 0 if either u is not a subword of w or v is not a subword of w. Therefore, we can compute all the coefficients recursively starting with w = u (for v ≤ u, pu u,Iv ) and going up in length of w. Moreover, by Lemma 7.1 one can find the coefficients bY by expanding the formal root polynomial in the hyperbolic case. Iu,Iv = bY u,Iv Observe that the same recurrent formula holds with Y 's replaced by X's. 18 BAPTISTE CALM `ES, KIRILL ZAINOULLINE, AND CHANGLONG ZHONG Remark 7.3. For an additive F there is a closed formula for bY [Bi99, Theorem 4] (see also [LZ16, (47)]) u,v = X bY βj1 . . . βjk , u,v given by Billey 1≤j1<...<jk≤l where Iu = (i1, . . . , il), βj = si1 . . . sij−1 (αij ), k = ℓ(v) and the summation ranges over the integer sequences for which the subword (ij1 , . . . , ijk ) is a reduce word for v. For a multiplicative F there are closed formulas given by Graham-Willems (see [LZ16, §5.2]). Example 7.4. Let F be the hyperbolic formal group law. Using the formula (7.1) we find pe si,si = bY si,si = xi and e,e = 1, psi psi e,e = 1 si,e = bY xi(cid:0)( −xi si,e = −xi x−i )2 − −xi , psi x−i (cid:1) = xi x−i x−i κi ∈ S, where κi = 1 xi + 1 x−i . Combining we obtain Y ∗ e · Y ∗ e = Y ∗ e + X ( xi x−i κi)Y ∗ i + X e,eY ∗ pw Iw and i ℓ(w)≥2 Y ∗ i · Y ∗ e = ( −xi x−i )Y ∗ i + X si,eY ∗ pw Iw . w>si Observe that for a general formal group law Y ∗ e 6= 1 in D∗ F . In the finite case, Y ∗ e is the class of the Bott-Samelson resolution of G/B corresponding to some reduced expression of the element of maximal length (it does not necessarily coincides with the fundamental class). For the usual cohomology (in the finite case) Y ∗ Iw are the Poincar´e duals to the usual Schubert classes. In particular, Y ∗ i corresponds to the class of a divisor. Example 7.5. Consider sjsi with j 6= i. Set xj(i) := xsj (αi). Then bY w,si = − xjxj(i) bY obtain w,w = xj xj(i), . Using the formula (7.1) we w,sj = − xjxj(i) x−j(i) w,e = xj xj(i) x−j x−j(i) and bY , bY x−j si,si bY w,w (cid:0)(bY psj si si,si = 1 bY si,sj = psj si psj si w,w (cid:0)(bY w,si)2 − bY sj ,si = 1 bY w,w sj ,sj bY w,sj )2 − bY w,si(cid:1) = 1 w,sibY bY w,sj(cid:1) = 1 ( xj xj(i) x−j w,sj = xj xj(i) , x−j x−j(i) ( xj xj(i) x−j(i) psj si sj ,sj = 1 bY x−j(i) x−j + xi), + xj). Therefore, we obtain in D∗ F Y ∗ i · Y ∗ i = xiY ∗ i + X xi xj x−j κjY ∗ sj si sisj =sj si + X sisj 6=sj si [xi xi(j) x−i(j) κi(j)Y ∗ sisj + 1 x−j ( xjxj(i) x−j + xi)Y ∗ sj si] + X ℓ(w)≥3 si,siY ∗ pw Iw . In particular, for an additive F we obtain in N ∗ Y ∗ i · Y ∗ i = αiY ∗ i + X (−α∨ j (αi))Y ∗ sj si . Observe that the general formula for a multiplication by Y ∗ i was obtained in [KK86, (4.30)]. in the additive case sisj 6=sj si FORMAL AFFINE DEMAZURE AND HECKE ALGEBRAS OF KAC-MOODY ROOT SYSTEMS19 References [Bi99] [CPZ] S. Billey, Kostant polynomials and the cohomology ring for G/B Duke Mathematical J. 96(1): 205–224, 1999. B. Calm`es, V. Petrov, K. Zainoulline, Invariants, torsion indices and oriented coho- mology of complete flags, Annales scientifiques de l'´Ecole normale sup´erieure (4) 46(3): 405–448, 2013. [CZZ1] B. Calm`es, K. Zainoulline, C. Zhong, A coproduct structure on the formal affine De- mazure algebra, Mathematische Zeitschrift 282(3): 1191–1218, 2016. [CZZ2] B. Calm`es, K. Zainoulline, C. Zhong, Push-pull operators on the formal affine Demazure algebra and its dual, Preprint arXiv:1312.0019. [CZZ3] B. Calm`es, K. Zainoulline, C. Zhong, Equivariant oriented cohomology of flag varieties, Documenta Mathematica Extra Volume: Alexander S. Merkurjev's Sixtieth Birthday: 113–144, 2015. [GKV] V. Ginzburg, M. Kapranov, E. Vasserot, Residue construction of Hecke algebras, Ad- vances in Mathematics 128(1): 1–19, 1997. [HMSZ] A. Hoffnung, J. Malag´on-L´opez, A. Savage, K. Zainoulline, Formal Hecke algebras and algebraic oriented cohomology theories, Selecta Mathematica 20(4): 1213–1245, 2014. [Ka90] V. Kac, Infinite-dimensional Lie algebras (3rd. ed.), Cambridge University Press, Cam- bridge, 1990. xxii+400 pp. [KK86] B. Kostant, S. Kumar, The nil Hecke ring and cohomology of G/P for a Kac-Moody group G∗, Advances in Mathematics 62: 187–237, 1986. [KK90] B. Kostant, S. Kumar, T -equivariant K-theory of generalized flag varieties, J. Differen- [Ku02] tial geometry 32: 549–603, 1990. S. Kumar, Kac-Moody groups, their flag varieties and representation theory, Progress in Mathematics 204, Birkhauser, Boston, MA, 2002. [LZ16] C. Lenart, K. Zainoulline, Towards generalized cohomology Schubert calculus via for- mal root polynomials, to appear in Mathematical Research Letters, 18pp. Preprint arXiv:1408.5952. [Le16] M.-A. Leclerc, The hyperbolic formal affine Demazure algebra, Algebras and Represen- [Mi80] tation Theory, Vol 19, Issue 5, 1043-1057. J. Milne, ´Etale Cohomology, Princeton Mathematical Series 33, Princeton University Press, Princeton, N.J., 1980. xiii+323 pp. [ZZ17] G. Zhao, C. Zhong, Geometric representations of the formal affine Hecke algebra, to appear in Advances in Mathematics, arXiv:1406.1283. [Zh15] C. Zhong, On the formal affine Hecke algebra, J. of the Institute of Mathematics of Jussieu 14(4): 837-855, 2015.
1505.06433	1	1505	2015-05-24T12:03:00	Nuclei and applications to star, semistar, and semiprime operations	[ "math.RA" ]	We show that the theory of quantales and quantic nuclei motivate new results on star operations, semistar operations, semiprime operations, ideal systems, and module systems, and conversely the latter theories motivate new results on quantales and quantic nuclei. Results include representation theorems for precoherent prequantales and multiplicative semilattices; characterizations of the simple prequantales; and a generalization to the setting of precoherent quantales of the construction of the largest finite type semistar operation and the largest stable semistar operation smaller than a given semistar operation.	math.RA	math	NUCLEI AND APPLICATIONS TO STAR, SEMISTAR, AND SEMIPRIME OPERATIONS JESSE ELLIOTT Abstract. We show that the theory of quantales and quantic nuclei moti- vate new results on star operations, semistar operations, semiprime operations, ideal systems, and module systems, and conversely the latter theories motivate new results on quantales and quantic nuclei. Results include representation theorems for precoherent prequantales and multiplicative semilattices; char- acterizations of the simple prequantales; and a generalization to the setting of precoherent quantales of the construction of the largest finite type semis- tar operation and the largest stable semistar operation smaller than a given semistar operation. Keywords: magma; ordered magma; closure operation; quantale; quantic nu- cleus; multiplicative lattice; multiplicative semilattice; ring; integral domain; star operation; semistar operation; semiprime operation; ideal system; module system 1. Introduction Certain classes of ordered algebraic structures -- Boolean algebras, Heyting al- gebras, multiplicative lattices, residuated lattices, locales, and quantales, to name a few -- serve to unify aspects of logic, order theory, algebra, and topology. One such set of unifications occured in the study, known as abstract ideal theory, of the ideal lattice of a commutative ring as a multiplicative lattice [20], initiated by Krull in the mid 1920s. Another set of notions, including Brouwer lattices, Heyting algebras, and frames, led to the study of locales, or "point-free topological spaces" [18]. These in turn led to a common generalization of multiplicative lattices and locales known as quantales, or "quantum locales", introduced in [24] by Mulvey to provide a lattice theoretic setting for the foundations of quantum mechanics and the theory of C∗-algebras and to develop the concept of noncommutative topology introduced in [11] by Giles and Kummer. This paper concerns applications of the theory of quantales to abstract ideal theory and commutative ring theory. We show that the theory of nuclei on ordered magmas [9, Section 3.4.11], and more specifically on suitable generalizations of quantales, provides a noncommutative and nonassociative abstract ideal theoretic setting for the theories of star operations, semistar operations, ideal systems, and module systems, and conversely the latter theories motivate new results on quan- tales. In this introduction we provide some background on the topics mentioned above and then provide a summary, addressing its organization. In Sections 2 through 7 we provide new results on nuclei, as well as on various classes of ordered magmas, including quantales, multiplicative lattices, and multiplicative semilat- tices. Then in Sections 8 through 10 we apply these results to star and semistar operations and ideal and module systems. 1 2 JESSE ELLIOTT A magma is a set M equipped with a binary operation on M , which we write multiplicatively. A magma is unital if it has an identity element. An ordered magma is a magma M equipped with a partial ordering ≤ on M such that x ≤ x′ and y ≤ y′ implies xy ≤ x′y′ for all x, x′, y, y′ ∈ M . For any self-map ⋆ of a set X we write x⋆ = ⋆(x) for all x ∈ X and Y ⋆ = ⋆(Y ) for all Y ⊆ X. If ⋆ is a self-map of a magma M , then the binary operation (x, y) 7−→ x ⋆ y = (xy)⋆ on M will be called ⋆-multiplication. We then consider the set M ⋆ as a magma under ⋆-multiplication restricted to M ⋆. A closure operation on a poset S is a self-map ⋆ of S satisfying the following conditions. (1) ⋆ is expansive: x ≤ x⋆ for all x ∈ S. (2) ⋆ is order-preserving: x ≤ y implies x⋆ ≤ y⋆ for all x, y ∈ S. (3) ⋆ is idempotent: (x⋆)⋆ = x⋆ for all x ∈ S. A closure operation on S is equivalently a self-map ⋆ of S such that x ≤ y⋆ if and only if x⋆ ≤ y⋆ for all x, y ∈ S. A nucleus (resp., strict nucleus) on an ordered magma M is a closure operation ⋆ on M such that x⋆y⋆ ≤ (xy)⋆ (resp., x⋆y⋆ = (xy)⋆) for all x, y ∈ M . Nuclei were first studied in the contexts of ideal lattices and locales and later in the context of quantales as quantic nuclei [25]. Example 1.1. (1) For any commutative ring R, a semiprime operation on R [27] is equivalently a nucleus on the ordered magma I(R) of all ideals of R. For example, the ideal radical operation I 7−→ √I is a semiprime operation on R, and if R is Noetherian of prime characteristic then the operation of tight closure [17] is a semiprime operation on R. (2) By [13, Lemmas 10.1 -- 10.4], for any topological space X the operation reg : U 7−→ int(cl(U )) is a strict nucleus on the locale O(X) of open subsets of X, where cl and int denote the topological closure and interior operations, respectively. (Open sets U such that U = U reg are called regular open sets.) Star, semistar, and semiprime operations are the most prominent examples of nuclei airising in commutative algebra. They are useful, among other reasons, for generalizing results on various classes of rings to much larger classes by relaxing, up to closure, certain ideal-theoretic assumptions, and also for specializing to classes of rings for which all or certain ideals are closed with respect to a given closure operation. Examples are provided by the following equivalences. (1) A domain is a UFD iff every ideal is principal up to t-closure. (2) A domain is Dedekind iff every nonzero ideal is invertible, while a domain is Krull iff every nonzero ideal is invertible up to t-closure. (3) A domain is Prufer iff every nonzero ideal is t-closed. (4) A domain is divisorial iff every nonzero ideal is closed under the divisorial closure operation. (5) A ring is reduced iff every ideal is closed under the radical operation. (6) A Noetherian ring of prime characteristic is weakly F-regular iff every ideal is tightly closed, and is F-rational iff all parameter ideals are tightly closed. The theory of nuclei allows for a unified treatment of closure operations as they naturally arise, not only in commutative algebra, but also in logic, order theory, the theory of quantales, and the theory of C∗-algebras. Further motivation for studying nuclei is provided by the fact that, not only semiprime operations, but PREQUANTALES 3 also star and semistar operations, can be characterized as nuclei. Let D be an integral domain with quotient field F , and let K(D) denote the ordered monoid of all nonzero D-submodules of F under the operation of multiplication. A semistar operation on D is a closure operation ⋆ on the poset K(D) such that (aI)⋆ = aI ⋆ for all nonzero a ∈ F and all I ∈ K(D). Semistar operations were introduced in [26] as a generalization of star operations, which were introduced by Krull in [21, Section 6.43]. The following result is proved in Section 8. Theorem 1.2. Let D be an integral domain with quotient field F . The following conditions are equivalent for any self-map ⋆ of K(D). (1) ⋆ is a semistar operation on D. (2) ⋆ is a nucleus on the ordered monoid K(D). (3) ⋆ is a closure operation on the poset K(D) and ⋆-multiplication on K(D) is (4) ⋆ is a closure operation on the poset K(D) such that (I ⋆J ⋆)⋆ = (IJ)⋆ for all I, J ∈ K(D) (or equivalently such that the map ⋆ : K(D) −→ K(D)⋆ is a magma homomorphism). associative. (5) HJ ⊆ I ⋆ if and only if HJ ⋆ ⊆ I ⋆ for all H, I, J ∈ K(D). (6) (I ⋆ :F J) = (I ⋆ :F J ⋆) for all I, J ∈ K(D). Corollary 1.3. Let D and D′ be integral domains. If the ordered monoids K(D) and K(D′) are isomorphic, then the lattices of all semistar operations on D and D′, respectively, are isomorphic. A D-submodule I of F is said to be a fractional ideal of D if aI ⊆ D for some nonzero element a of F . Let F (D) denote the set of all nonzero fractional ideals of D, which is an ordered submonoid of K(D). A star operation ∗ on D is a closure operation on the poset F (D) such that D∗ = D and (aI)∗ = aI ∗ for all nonzero a ∈ F and all I ∈ F (D). Theorem 8.1 provides a characterization of star operations analogous to that of Theorem 1.2 above. In particular, a star operation on D is equivalently a nucleus ∗ on the ordered monoid F (D) such that D∗ = D. The weak ideal systems and module systems of [14, 15], which generalize semiprime operations and star and semistar operations, respectively, can also be characterized as nuclei: see Section 10. By Theorem 1.2, one can characterize a semistar operation on D as a self-map ⋆ of K(D) satisfying a single axiom, namely, statement (5) (or (6)) of the theorem. The theorem suggests that there is potential overlap among the theory of semistar operations, abstract ideal theory, and the theory of quantales and quantic nuclei. The connections among these theories suggested by Theorem 1.2 and its equivalents for star operations and semiprime operations are the main inspiration for this paper. Quantales provide a natural context in which to study nuclei, as the nuclei on a quantale classify its quotient objects in the category of quantales. A quantale (resp., prequantale) is an ordered semigroup (resp., ordered magma) Q such that the supremum W X of X exists and a(W X) = W(aX) and (W X)a = W(Xa) for any a ∈ Q and any subset X of Q [29, Definition 2.4.2]. A quantale is equivalently an associative prequantale. A multiplicative lattice is a commutative unital quan- tale, and a locale, or frame, is a quantale in which multiplication is the operation ∧ of infimum. A prequantale (resp., quantale, multiplicative lattice) is equiva- lently a magma object (resp., semigroup object, commutative monoid object) in the monoidal category of sup-lattices, that is, the category of complete lattices 4 JESSE ELLIOTT equipped with the tensor product with morphisms as the sup-preserving functions [19, Section II.5]. Example 1.4. (1) The power set 2M of a magma M is a prequantale under the operation (X, Y ) 7−→ XY = {xy : x ∈ X, y ∈ Y }. It is the free prequantale on the magma M . The prequantale 2M is a quantale if and only if M is a semigroup, so 2M is the free quantale on M if M is a semigroup. (2) The poset of all supremum-preserving self-maps of a complete lattice is a unital quantale under the operation of composition. (3) Any totally ordered complete lattice is a frame, as is any complete Boolean algebra. a locale. (4) The complete lattice O(X) of all open subsets of a topological space X is (5) For any algebra A over a ring R, the complete lattice ModR(A) of all (two- sided) R-submodules of A is a quantale under the operation (I, J) 7−→ IJ = {Px∈X,y∈Y xy : X ⊆ I, Y ⊆ J finite}. (6) The poset Max(A) of all closed linear subspaces of a unital C∗-algebra A is a unital quantale with involution [23, Definition 2.4] under the operation (M, N ) 7−→ M N and by [23, Theorem 5.4] is a complete invariant of A. In the definition above of quantales and prequantales, restricting subsets X to be among certain subclasses of subsets of Q yields more general structures with respect to which it is natural to study nuclei. This is required, for example, to accommodate the theory of star operations, since the ordered monoids on which star operations are defined are bounded complete but not complete. We will say that a near prequantale (resp., semiprequantale) is an ordered magma Q such that nonempty subset X (resp., any nonempty finite or bounded subset X) of Q. A near prequantale is equivalently a semiprequantale Q with a largest element, and a W X exists and a(W X) = W(aX) and (W X)a = W(Xa) for any a ∈ Q and any prequantale is equivalently a near prequantale Q with a smallest elementV Q that annihilates every element of Q. We define the terms near quantale, semiquantale, near multiplicative lattice, and semimultiplicative lattice in the obvious way. A theme of this paper is that many known results about quantales and multiplicative lattices generalize to these more general structures. In particular, the attempt is made to state each result in as general a setting as possible. For example, results on star operations are generalized to the context of nuclei on precoherent semimultiplicative lattices. Example 1.5. (1) If a is an idempotent element of a prequantale Q, then the set Q≤a = {x ∈ Q : x ≤ a} is a subprequantale of Q, while the set Q≥a = {x ∈ Q : x ≥ a} is a sub near prequantale of Q. (2) If Q is a prequantale and xy = 0 implies x = 0 or y = 0 for all x, y ∈ Q, where 0 =V Q, then Q−{0} is a sub near prequantale of Q. In particular, the set 2M−{∅} of all nonempty subsets of a magma M is a sub near prequantale of the prequantale 2M , and it is the free near prequantale on the magma M . The remainder of this paper is organized as follows. In Section 2 we introduce In Sections 3 and 4 we study properties of several classes of ordered magmas. PREQUANTALES 5 nuclei and the poset of all nuclei on an ordered magma, with particular attention to the classes of ordered magmas defined in Section 2. In Section 5 we generalize the generalized divisorial closure semistar operations [28, Example 1.8(2)] to the contexts of near prequantales and residuated ordered monoids, and we use this to give a characterization of the simple near prequantales. A closure operation ⋆ on a poset S is said to be finitary if (W ∆)⋆ = W(∆⋆) for all directed subsets ∆ of S for which W ∆ exists. For example, a finite type semistar operation [26, Definition 3] on an integral domain D is equivalently a finitary nucleus on K(D). In Section 6 we examine the poset of all finitary nuclei on an ordered magma. We also generalize the well-known construction of the largest finite type semistar operation ⋆f smaller than a semistar operation ⋆, as follows. An element x of a poset S is said to be compact if whenever x ≤W ∆ for some directed subset ∆ of S for whichW ∆ exists one has x ≤ y for some y ∈ ∆. Generalizing the notion of a precoherent quantale [29, Definition 4.1.1(2)], we say that an ordered magma M is precoherent if every element of M is the supremum of a set of compact elements of M and the set K(M ) of all compact elements of M is a submagma of M . For example, the near multiplicative lattice K(D) is precoherent for any integral domain D with quotient field F since K(K(D)) is the set of all nonzero finitely generated D-submodules of F . The following result is proved in Section 6. Theorem 1.6. If ⋆ is any nucleus on a precoherent semiprequantale Q, then there is a largest finitary nucleus ⋆f on Q that is smaller than ⋆, and one has x⋆f = W{y⋆ : y ∈ K(Q) and y ≤ x} for all x ∈ Q. Also in Section 6, we prove that the association Q 7−→ K(Q) yields an equiva- lence between the category of precoherent near prequantales (resp., precoherent prequantales) and the category of multiplicative semilattices (resp., prequantic semilattices). This result, along with Theorem 1.6 above, provides substantive motivation for the study of precoherent near prequantales, particularly in algebraic settings. In Section 7 we generalize several known results and new results on stable semis- tar operations to the context of precoherent semimultiplicative lattices, where a semistar operation ⋆ on an integral domain D is said to be stable if (I∩J)⋆ = I ⋆∩J ⋆ for all I, J ∈ K(D) [7]. Finally, in Sections 8 through 10 we apply the theory of nuclei to star and semistar operations and ideal and module systems. There, for example, we prove Theorem 1.2 and its equivalents for star operations and ideal and module systems, we give a construction of the smallest semistar operation extending a given star operation, and we provide some new examples of semistar operations induced by complete integral closure, plus closure, and tight closure. The last of these examples leads to two potentially useful definitions of tight closure as a semiprime operation on any commutative ring, not necessarily Noetherian, of prime characteristic. We remark that this paper is a substantial revision of a previous version [6] posted on arxiv.org in 2011 under a different title. 2. Ordered algebraic structures If a proof in this paper is omitted then its reconstruction should be routine. All rings and algebras are assumed unital and all magmas are written multiplicatively. For any subset X of a poset S we write W X =WS X for the supremum of X and 6 JESSE ELLIOTT V X =VS X for the infimum of X if either exists. (Note the cases W∅ =V S and V∅ =W S.) The category of posets has as morphisms the order-preserving functions. A poset in which every pair of elements of has a supremum (resp., infimum) is said to be a join semilattice (resp., meet semilattice). A lattice is a poset that is both a join and meet semilattice. A poset S is complete if every subset of S has a supremum in S, or equivalently if every subset of S has an infimum S. A complete poset is also known as a complete lattice, or sup-lattice. A poset S is bounded complete if every nonempty subset of S that is bounded above has a supremum in S, or equivalently if every nonempty subset of S that is bounded below has an infimum in S. We will say that a poset S is near sup-complete, or a near sup-lattice, if every nonempty subset of S has a supremum in S. A near sup-lattice is equivalently a bounded complete poset with a largest element, or equivalently a poset S such that every subset of S that is bounded below has an infimum in S. A sup-lattice is equivalently a near sup-lattice having a least element. A map f : S −→ T between posets is said to be sup-preserving if f (W X) = W f (X) for every subset X of S for which W X exists. We will say that f : S −→ T is near sup-preserving if f (W X) = W f (X) for every nonempty subset X of S such that W X exists. If f is near sup-preserving, then f is sup-preserving if and only if f (V S) = V T or V S does not exist. The morphisms in the category of sup-lattices (resp., category of near sup-lattices) are the sup-preserving (resp., near sup-preserving) functions. A nonempty subset ∆ of a poset S is said to be directed if every finite subset of ∆ has an upper bound in ∆. A poset S is directed complete, or a dcpo, if each of its directed subsets has a supremum in S. We will say that S is a bdcpo if every directed subset of S that is bounded above has a supremum in S. A subset X of S is said to be downward closed (resp., upward closed) if y ∈ X whenever y ≤ x (resp., y ≥ x) for some x ∈ X. The set X is said to be Scott closed if X is a downward closed subset of S and for any directed subset ∆ of X one has W ∆ ∈ X if W ∆ exists. The set X is Scott open if its complement is Scott closed, or equivalently if X is an upward closed subset of S and X ∩ ∆ 6= ∅ for any directed subset ∆ of S with W ∆ ∈ X. The Scott open subsets of S form a topology on S called the Scott topology. A function f : S −→ T between posets is said to be Scott continuous if f is continuous when S and T are endowed with the Scott topologies. Equivalently, f is Scott continuous if and only if f (W ∆) = W f (∆) for every directed subset ∆ of S for which W ∆ exists. Every near sup-preserving function is Scott continuous, If S and T are posets, then S × T is a poset under the relation ≤ defined by (x, y) ≤ (x′, y′) iff x ≤ x′ and y ≤ y′. The Scott topology on S × T may be strictly finer than the product of the Scott topologies on S and T [10, Exercise II-4.26]. Characterizations of the classes of ordered magmas that will be discussed in this section are given in Table 1, and the various relationships among these classes are summarized in the lattice diagrams of Figures 1, 2, and 3. Within any of these three diagrams the intersection of any two of the classes is the class lying directly above both of them. and every Scott continuous function is order-preserving. A magma is unital (resp., left unital, right unital) if it has an identity element (resp., left identity element, right identity element). A map f : M −→ N of magmas is a homomorphism of magmas if f (xy) = f (x)f (y) for all x, y ∈ M . The morphisms in the category of ordered magmas are the order-preserving magma PREQUANTALES 7 Table 1. Characterizations of ordered magmas (om's) sup-magma near sup-magma dcpo magma bounded complete ordered magma M abbr. W X exists for all X for all X 6= ∅ for all directed X for all X 6= ∅ bounded above for X = M for X = ∅ for all X for all X 6= ∅ for all finite or bounded X 6= ∅ for all finite X bounded above with annihilator prequantale near prequantale semiprequantale s ns d bc b a p np sp ps ms for all finite X 6= ∅ prequantic semilattice multiplicative semilattice Scott topological residuated t r near residuated nr if ∃x, y : X = {z : zy ≤ x} or X = {z : yz ≤ x} if X 6= ∅ and ∃x, y : X = {z : zy ≤ x} or X = {z : yz ≤ x} W(XY ) =W XW Y for X = ∅ or Y = ∅ for all X, Y for all X, Y 6= ∅ for all (finite or) bounded X, Y 6= ∅ for all finite X, Y exist for all X, Y such that for all finite X, Y 6= ∅ for all directed X, Y such that W X, W Y W X, W Y exist for all X, Y 6= ∅ such that W X, W Y exist Figure 1. Relationships among classes of ordered magmas r+d r ③③③③③③ ③③③③③③ &❉❉❉❉❉❉ ❉❉❉❉❉❉ nr ttttt ~ ttttt (❏❏❏❏❏ ❏❏❏❏❏ tttttt } tttttt )❑❑❑❑❑❑❑ ❑❑❑❑❑❑❑ r+ns np+s p )❑❑❑❑❑❑ ❑❑❑❑❑❑ np tttttt ~ tttttt (❏❏❏❏❏❏ ❏❏❏❏❏❏ ssssss } ssssss (❏❏❏❏❏❏ ❏❏❏❏❏❏ tttttt ~ tttttt )❏❏❏❏❏ ❏❏❏❏❏ ttttt } ttttt )❑❑❑❑❑❑ ❑❑❑❑❑❑ rrrrrrr } rrrrrrr d t ttttttt } ttttttt )▲▲▲▲▲▲▲ ▲▲▲▲▲▲▲ om nr+d t+ns (■■■■■ ■■■■■ ✉✉✉✉✉✉ ~ ✉✉✉✉✉✉ (❏❏❏❏❏❏ ❏❏❏❏❏❏ ttttttt } ttttttt t+s ❈❈❈❈❈ ❈❈❈❈❈ ③③③③③③ ③③③③③③ s t+d ns u ! v v y v % v ! u u y ! ! u u ! u 8 JESSE ELLIOTT Figure 2. Further relationships among classes of ordered magmas ps+t ps+s np p ▼▼▼▼▼▼▼▼ ▼▼▼▼▼▼▼▼ qqqqqqq ▼▼▼▼▼▼ qqqqqqq ▼▼▼▼▼▼ qqqqqqq qqqqqqq ms rrrrrrr } rrrrrrr )❑❑❑❑❑ ❑❑❑❑❑ ssssss } ssssss )▲▲▲▲▲▲▲ ▲▲▲▲▲▲▲ rrrrrrr } rrrrrrr qqqqqq qqqqqq ms+t ms+ns t+a a ✉✉✉✉✉ ~ ✉✉✉✉✉ )❏❏❏❏❏❏❏ ttttttt ❏❏❏❏❏❏❏ } ttttttt )▲▲▲▲▲▲▲ ▲▲▲▲▲▲▲ ps t om Figure 3. Further relationships among classes of ordered magmas t+a+b ps+b ms+t+b ms+ns sp a+b t+b ms+b ms+t ms+bc ♣♣♣♣♣♣ { ♣♣♣♣♣♣ +◆◆◆◆◆◆◆ ◆◆◆◆◆◆◆ ♣♣♣♣♣♣♣ { ♣♣♣♣♣♣♣ +❖❖❖❖❖❖❖❖ ❖❖❖❖❖❖❖❖ b ps+t+b ps+s np ♥♥♥♥♥♥♥♥ { ♥♥♥♥♥♥♥♥ +❖❖❖❖❖❖ ♦♦♦♦♦♦♦ ❖❖❖❖❖❖ { ♦♦♦♦♦♦♦ +PPPPPP ♦♦♦♦♦♦♦♦ PPPPPP { ♦♦♦♦♦♦♦♦ ,PPPPPPPPP PPPPPPPPP ♥♥♥♥♥♥♥♥♥ z ♥♥♥♥♥♥♥♥♥ ,◗◗◗◗◗◗◗◗◗◗ ◗◗◗◗◗◗◗◗◗◗ p +❖❖❖❖❖❖❖❖❖ ❖❖❖❖❖❖❖❖❖ ♣♣♣♣♣♣♣♣ +❖❖❖❖❖❖❖ { ♣♣♣♣♣♣♣♣ ❖❖❖❖❖❖❖ +❖❖❖❖❖❖❖ ❖❖❖❖❖❖❖ ♦♦♦♦♦♦♦ { ♦♦♦♦♦♦♦ +❖❖❖❖❖❖❖❖ ❖❖❖❖❖❖❖❖ ♦♦♦♦♦♦♦♦♦ { ♦♦♦♦♦♦♦♦♦ ♥♥♥♥♥♥♥♥ z ♥♥♥♥♥♥♥♥ t om ▼▼▼▼▼▼▼▼ ▼▼▼▼▼▼▼▼ ◆◆◆◆◆◆ ♣♣♣♣♣♣♣ ◆◆◆◆◆◆ ♣♣♣♣♣♣♣ ♣♣♣♣♣♣♣ { ♣♣♣♣♣♣♣ ms homomorphisms. We will say that a sup-magma (resp., near sup-magma, dcpo magma, bdcpo magma) is an ordered magma that is complete (resp., near sup- complete, directed complete, a bdcpo) as a poset. The morphisms in the category of sup-magmas (resp., category of near sup-magmas, category of dcpo magmas, category of bdcpo magmas) are the sup-preserving (resp., near sup-preserving, Scott continuous, Scott continuous) magma homomorphisms. An annihilator of an ordered magma M is a least element 0 = V M of M such that 0x = 0 = x0 for all x ∈ M . If an annihilator of M exists then we say that M is with annihilator. Lemma 2.1. The following are equivalent for any ordered magma M . (1) The map M × M −→ M of multiplication in M is sup-preserving and V M is an annihilator of M if V M exists (resp., multiplication in M is near (2) For all a ∈ M , the left and right multiplication by a maps on M are sup- sup-preserving, multiplication in M is Scott continuous). preserving (resp., near sup-preserving, Scott continuous). u " v ! " u t ! ! u u t ! t s # s # # s s " # # s s # " s t # $ r # s s $ r PREQUANTALES 9 exists. (3) a(W X) = W(aX) and (W X)a = W(Xa) for any a ∈ M and any subset (resp., any nonempty subset, any directed subset) X of M such that W X (4) W(XY ) = W XW Y for any subset (resp., any nonempty subset, any di- rected subset) X and Y of M such that W X and W Y exist. We will say that an ordered magma M is Scott-topological if the multiplication map M × M −→ M is Scott continuous. A Scott-topological ordered magma is equivalently a magma object in the monoidal category of posets equipped with the product × with morphisms as the Scott continuous functions. If M is a topologi- cal magma when endowed with the Scott topology, then M is a Scott-topological magma, but the converse appears to be false. A near prequantale is equivalently a magma object in the monoidal category of near sup-lattices equipped with the product ×. They form a full subcategory of the category of near sup-magmas. A near quantale (resp., near multiplicative lattice) is equivalently a semigroup object (resp., commutative monoid object) in the monoidal category of near sup-lattices. See Example 1.1 for examples. For a further example, note that the poset of all near sup-preserving self-maps of a near sup-lattice is a unital near quantale under the operation of composition. An element a of an ordered magma M is said to be residuated if for all x ∈ M there exists a largest element z = x/a of M such that za ≤ x and a largest element z′ = a\x of M such that az′ ≤ x. (Many authors denote x/a and a\x by x ← a and a → x, respectively.) An ordered magma M is said to be residuated if every element of M is residuated [31]. By [9, Theorem 3.10], if M is residuated, then for all a ∈ M the left and right multiplication by a maps on M are sup-preserving. Example 2.2. (1) By [9, Corollary 3.11] a prequantale is equivalently a complete and residu- ated ordered magma. (2) A Heyting algebra is a bounded lattice that, as an ordered monoid under the operation ∧ of infimum, is residuated [9, Section 1.1.4]. A complete Heyting algebra, also known as a locale, or frame (although the morphisms in their respective categories are diffierent), is equivalently a unital quantale (or prequantale) in which every element is idempotent and whose largest element is the identity element. (3) If S is a nontrivial complete lattice, then S is a near multiplicative lattice under the operation ∨ of supremum, but S is not residuated since one has x ∨W{y ∈ S : x ∨ y ≤ 1} = x > 1 for all x 6= 1 in S. Proposition 2.3. The following are equivalent for any ordered magma M . (1) M is a prequantale. (2) M is complete and residuated. (3) M is a near prequantale with annihilator. (4) M is complete with annihilator and the multiplication map M × M −→ M (5) M is complete and the left and right multiplication by a maps on M are is sup-preserving. sup-preserving for all a ∈ M . morphism. (6) The map 2M −→ M acting by X 7−→W X is a well-defined magma homo- (7) W(XY ) =W XW Y for all subsets X and Y of M . 10 JESSE ELLIOTT We say that an element a of an ordered magma M is near residuated if for all x ∈ M such that za ≤ x (resp., az ≤ x) for some z ∈ M there exists a largest such element z = x/a (resp., z = a\x) of M . We say that M near residuated if every element of M is near residuated. For example, Example 2.2(3) is near residuated. If M is near residuated, then for all a ∈ M the left and right multiplication by a maps on M are near sup-preserving. Although near prequantales are not necessarily residuated, they are near residuated. Proposition 2.4. The following are equivalent for any ordered magma M . (1) M is a near prequantale. (2) M is near sup-complete and near residuated. (3) The ordered magma M0 = M ∐ {0}, where 0x = 0 = x0 and 0 ≤ x for all (4) M is near sup-complete and the multiplication map M × M −→ M is near (5) M is near sup-complete and the left and right multiplication by a maps on x ∈ M0, is a prequantale. sup-preserving. M are near sup-preserving for all a ∈ M . homomorphism. (6) The map 2M−{∅} −→ M acting by X 7−→ W X is a well-defined magma (7) W(XY ) =W XW Y for all nonempty subsets X and Y of M . A multiplicative semilattice is an ordered magma M such that M is a join semi- lattice and a(x ∨ y) = ax ∨ ay and (x ∨ y)a = xa ∨ ya for all a, x, y ∈ M [3, Section XIV.4]. We will say that a prequantic semilattice is a multiplicative semilattice with annihilator. The prequantic semilattices (resp., multiplicative semilattices) form a category, where a morphism is a magma homomorphism f : M −→ M ′ such that f (W X) =W f (X) for all finite subsets (resp., all finite nonempty subsets) X of M . Example 2.5. The set K(2M ) of all finite subsets of a magma M is a sub prequantic semilattice of the prequantale 2M and is the free prequantic semilattice on the magma M , and the set K(2M )−{∅} of all finite nonempty subsets of M is a sub multiplicative semilattice of K(2M ) and is the free multiplicative semilattice on M . Proposition 2.6. The following are equivalent for any ordered magma M . (1) M is a prequantic semilattice (resp., multiplicative semilattice). subset (resp., any finite nonempty subset) X of M . (2) a(W X) = W(aX) and (W X)a = W(Xa) for any a ∈ M and any finite (3) The map K(2M ) −→ M (resp., K(2M )−{∅} −→ M ) acting by X 7−→W X (4) W(XY ) =W XW Y for all finite subsets (resp., all finite nonempty subsets) is a well-defined magma homomorphism. X and Y of M . Proposition 2.7. A prequantale (resp., near prequantale, semiprequantale) is equiv- alently a complete (resp., near sup-complete, bounded complete) Scott-topological prequantic semilattice (resp., multiplicative semilattice, multiplicative semilattice). 3. Nuclei A preclosure on a poset S is a self-map ⋆ of S that is expansive and order- preserving. A closure operation is equivalently an idempotent preclosure. Lemma 3.1. Let ⋆ be a closure operation on a poset S and let X ⊆ S. PREQUANTALES 11 (1) WS⋆ X ⋆ = (WS X ⋆)⋆ = (WS X)⋆ if WS X exists. (2) VS⋆ X ⋆ = (VS X ⋆)⋆ =VS X ⋆ if VS X ⋆ exists. (3) The map ⋆ : S −→ S⋆ is sup-preserving. (4) If S is complete (resp., near sup-complete, bounded complete, directed com- plete, a bdcpo, a join semilattice, a meet semilattice), then so is S⋆. If ⋆ is a nucleus on an ordered magma M , then the set M ⋆ is an ordered magma under ⋆-multiplication, the corestriction M ⋆⋆ : M −→ M ⋆ is a sup-preserving morphism of ordered magmas, and if 1 is an identity element of M then 1⋆ is an identity element of M ⋆. The following elementary results give several characterizations of nuclei. Proposition 3.2. The following conditions are equivalent for any closure operation ⋆ on an ordered magma M . (1) ⋆ is a nucleus on M . (2) xy⋆ ≤ (xy)⋆ and x⋆y ≤ (xy)⋆ for all x, y ∈ M . (3) (x⋆y⋆)⋆ = (xy)⋆ for all x, y ∈ M , or equivalently, the map ⋆ : M −→ M ⋆ is a magma homomorphism. If M is an ordered monoid, then the above conditions are equivalent to the following. (4) ⋆-multiplication on M is associative. Proposition 3.3. The following are equivalent for any self-map ⋆ of a left or right unital ordered magma M . (1) ⋆ is a nucleus on M . (2) xy ≤ z⋆ ⇔ xy⋆ ≤ z⋆ ⇔ x⋆y ≤ z⋆ for all x, y, z ∈ M . (3) x ≤ x⋆, and xy ≤ z⋆ ⇒ x⋆y⋆ ≤ z⋆, for all x, y, z ∈ M . If M is left or right unital and near residuated, then the above conditions are equiv- alent to the following. (4) x⋆/y = x⋆/y⋆ (resp., y\x⋆ = y⋆\x⋆) for all x, y ∈ M such that x⋆/y or x⋆/y⋆ (resp., y\x⋆ or y⋆\x⋆) is defined. We will say that a subset Σ of an ordered magma M is a sup-spanning subset of M if xy =_{ay : a ∈ Σ and a ≤ x} =_{xb : b ∈ Σ and b ≤ y} for all x, y ∈ M . If M is unital and residuated, or more generally if M is left or right unital and for all a ∈ M the left and right multiplication by a maps on M are sup-preserving (resp., near sup-preserving), then Σ is a sup-spanning subset of M if and only if every element of M is the supremum of some subset of Σ (resp., every element of M is the supremum of some subset of Σ andV M is an annihilator of M if V M exists and is not in Σ). For example, the set Prin(D) of all nonzero principal fractional ideals of an integral domain D is a sup-spanning subset of the residuated near quantale K(D) of all nonzero D-submodules of the quotient field of D. Proposition 3.4. Let ⋆ be a closure operation on an ordered magma M and let Σ be any sup-spanning subset of M . Then ⋆ is a nucleus on M if and only if ax⋆ ≤ (ax)⋆ and x⋆a ≤ (xa)⋆ for all a ∈ Σ and all x ∈ M , and in that case Σ⋆ is a sup-spanning subset of M ⋆. 12 JESSE ELLIOTT Proof. Necessity of the given condition is clear. Suppose that ax⋆ ≤ (ax)⋆ and x⋆a ≤ (xa)⋆ for all a ∈ Σ and all x ∈ M . Then xy⋆ = Wa∈Σ: a≤x ay⋆ ≤ (cid:16)Wa∈Σ: a≤x ay(cid:17)⋆ = (xy)⋆, and by symmetry x⋆y ≤ (xy)⋆, for all x, y ∈ M , so that ⋆ is a nucleus. That Σ⋆ is a sup-spanning subset of M ⋆ follows easily from Lemma 3.1(1). (cid:3) For any self-map ⋆ of a magma M we say that a ∈ M is transportable through ⋆ if (ax)⋆ = ax⋆ and (xa)⋆ = x⋆a for all x ∈ M , and we let T⋆(M ) denote the set of all elements of M that are transportable through ⋆. Corollary 3.5. Any closure operation ⋆ on an ordered magma M such that T⋆(M ) is a sup-spanning subset of M is a nucleus on M . For any magma M we let Inv(M ) denote the set of all u ∈ M for which there exists u−1 ∈ M such that the left and right multiplication by u−1 maps are in- verses, respectively, to the left and right multiplication by u maps. For any ordered magma M we let U(M ) denote the set of all u ∈ M such that the left and right multiplication by u maps are poset automorphisms of M . One has Inv(M ) ⊆ U(M ) for any ordered magma M , with equality holding if M is a monoid. Proposition 3.6. One has Inv(M ) ⊆ T⋆(M ) for any nucleus ⋆ on an ordered magma M . Proof. If u ∈ Inv(M ), then x⋆ = u−1(ux⋆) ≤ u−1(ux)⋆ ≤ (u−1(ux))⋆ = x⋆, whence ux⋆ = (ux)⋆, for all x ∈ M . By symmetry one has u ∈ T⋆(M ). Remark 3.7. Let D be an integral domain with quotient field F . By definition, a semistar operation on D is a closure operation ⋆ on K(D) such that (aI)⋆ = aI ⋆ for all I ∈ K(D) and all nonzero a ∈ F , that is, such that Prin(D) ⊆ T⋆(K(D)). Since Prin(D) is a sup-spanning subset of K(D), Corollary 3.5 implies that any semistar operation on D is a nucleus on K(D). Conversely, since Prin(D) ⊆ Inv(K(D)) (and in fact Inv(K(D)) is the set of invertible fractional ideals of D), Proposition 3.6 implies that any nucleus on K(D) is a semistar operation on D. A similar argument shows that a star operation on D is equivalently a nucleus ∗ on F (D) such that D∗ = D. (cid:3) We will say that a U-lattice (resp., near U-lattice, semi-U-lattice) is a sup-magma (resp., near sup-magma, ordered magma that is a bounded complete join semilat- tice) M such that U(M ) is a sup-spanning subset of M . Example 3.8. For any integral domain D with quotient field F , the multiplicative lattice ModD(F ) of all D-submodules of F is a U-lattice, and the near multiplicative lattice K(D) of all nonzero D-submodules of F is a near U-lattice. Proposition 3.9. An ordered magma M is a prequantale (resp., near prequantale, semiprequantale) if and only if M is complete (resp., near sup-complete, a bounded complete join semilattice) and the set of all a ∈ M such that the left and right multiplication by a maps on M are sup-preserving (resp., near sup-preserving, near sup-preserving) is a sup-spanning subset of M . Corollary 3.10. Any U-lattice (resp., near U-lattice, semi-U-lattice) is a prequan- tale (resp., near prequantale, semiprequantale). PREQUANTALES 13 Proposition 3.11. Let ⋆ be a nucleus on an ordered magma M . If M is a prequan- tale (resp., near prequantale, residuated, near residuated, semiprequantale, quantale, near quantale, semiquantale, multiplicative lattice, near multiplicative lattice, semi- multiplicative lattice, U-lattice, near U-lattice, semi-U-lattice, multiplicative semi- lattice, prequantic semilattice), then so is M ⋆. Moreover, if M is near residuated, then (x/y)⋆ = x/y⋆ = x/y (resp., (y\x)⋆ = y⋆\x = y\x) for all x ∈ M ⋆ and all y ∈ M such that x/y (resp., y\x) is defined. Proof. Suppose that M is a prequantale. By Lemma 3.1(4) the partial ordering on M ⋆ is complete. For any a ∈ M ⋆ and X ⊆ M ⋆ we have a ⋆WM ⋆ X = a ⋆ (WM X)⋆ = (a WM X)⋆ = (WM aX)⋆ = (WM (aX)⋆)⋆ = (WM (a ⋆ X))⋆ = WM ⋆ (a ⋆ X), and therefore a ⋆WM ⋆ X =WM ⋆ (a ⋆ X). By symmetry the corresponding equation holds for right multiplication. Thus M ⋆ is a prequantale. The proof of the rest of the first statement of the proposition is similar. The second statement of the proposition follows from the equivalences yz ≤ x⋆ ⇔ y⋆z ≤ x⋆ ⇔ yz⋆ ≤ x⋆ ⇔ y⋆z⋆ ≤ x⋆. (cid:3) near prequantales (resp., prequantales). By following proposition, which is a straightforward generalization of [29, Theo- rem 3.1.1], the nuclei on a near prequantale or prequantale Q classify the quotient objects of Q. Proposition 3.12. Let f : Q → M be a morphism of near sup-magmas (resp., sup-magmas), where Q is a near prequantale (resp., prequantale). (1) If ⋆ is any nucleus on Q, then Q⋆⋆ : Q −→ Q⋆ is a surjective morphism of (2) There exists a unique nucleus ⋆ on Q such that f = (fQ⋆ ) ◦ (Q⋆⋆) and fQ⋆ is injective; moreover, one has x⋆ = W{y ∈ Q : f (y) = f (x)} for all x ∈ Q. (3) fQ⋆ : Q⋆ −→ M is an embedding of near sup-magmas (resp., sup-magmas). (4) fQ⋆ : Q⋆ −→ im f is an isomorphism of ordered magmas. A quantale Q is simple if every nontrivial sup-preserving semigroup homomor- phism from Q is injective. More generally, we will say that a near sup-magma M is simple if every nonconstant near sup-preserving magma homomorphism from M is injective. If M is a sup-magma, then this holds if and only if every nontrivial sup-preserving magma homorphism from M is injective, so our definition is con- sistent with that of a simple quantale. Let d be the identity map on M , and let e : M −→ M be defined by xe =W M for all x ∈ M . Both d and e are nuclei. Corollary 3.13. A near prequantale Q is simple if and only if d and e are the only nuclei on Q. 4. The poset of nuclei Let S be a poset and X a set. The set SX of all functions from X to S is partially ordered, where f ≤ g if f (x) ≤ g(x) for all x ∈ X. If f ≤ g, then we say that f is smaller, or finer, than g, or equivalently g is larger, or coarser, than f . The set C(S) of all closure operations on S inherits a partial ordering from the poset SS. For any ⋆1, ⋆2 ∈ C(S) one has ⋆1 ≤ ⋆2 if and only if S⋆1 ⊇ S⋆2. The identity operation d on S is the smallest closure operation on S. If W S exists then there is a largest closure operation e on S, given by xe = W S for all x ∈ S. If M is an ordered magma, then we let N(M ) denote the subposet of C(M ) consisting of all nuclei on M . In this section we study properties of the poset N(M ). 14 JESSE ELLIOTT The following result characterizes nuclei on near residuated ordered magmas and generalizes [9, Lemma 3.33]. Proposition 4.1. Let C be a subset of a poset S. (1) There exists a closure operation ⋆ on S with C = S⋆ if and only if V{a ∈ C : a ≥ x} exists in S for all x ∈ S. For any such closure operation ⋆ one has x⋆ =V{a ∈ C : a ≥ x} for all x ∈ S, and therefore ⋆ = ⋆C is uniquely determined by C. (2) If S = M is a near residuated ordered magma and ⋆C exists, then ⋆C is a nucleus on M if and only, for all x ∈ C and y ∈ M , one has x/y ∈ C provided x/y exists and y\x ∈ C provided y\x exists. Proof. Statement (1) is well-known and easy to check. We prove (2). If ⋆ is a nucleus, then it follows from Proposition 3.11 that x/y ∈ C and y\x ∈ C for all x, y as in statement (2). Conversely, suppose that this condition on C holds. Let x, y ∈ M . Then we claim that x⋆y = V{a ∈ C : a ≥ x}y is less than or equal to (xy)⋆ = V{b ∈ C : b ≥ xy}. For let b ∈ C with b ≥ xy. Set a = b/y, whence a ≥ x. By hypothesis one has a ∈ C. Therefore x⋆y ≤ ay = (b/y)y ≤ b. Taking the infimum over all such b, we see that x⋆y ≤ V{b ∈ C : b ≥ xy} = (xy)⋆. By symmetry one also has xy⋆ ≤ (xy)⋆. It follows that ⋆ is a nucleus. Corollary 4.2. One has the following. (cid:3) (1) Let C be a nonempty subset of a bounded complete poset S. Then there exists a closure operation ⋆ on S with C = S⋆ if and only if V X ∈ C for all nonempty X ⊆ C bounded below. (2) Let C be a nonempty subset of a bounded complete and near residuated ordered magma Q. Then there exists a nucleus ⋆ on Q with C = Q⋆ if and only if V X ∈ C for all nonempty X ⊆ C bounded below and for all x ∈ C and y ∈ Q one has x/y ∈ C provided x/y exists and y\x ∈ C provided y\x exists. Corollary 4.3. One has the following. (1) Let S be a bounded complete poset. Then C(S) is bounded complete. More- (2) Let Q be a bounded complete and near residuated ordered magma. Then over, one has SW Γ =T⋆∈Γ S⋆ for all bounded subsets Γ of C(S). N(M ) is bounded complete. Moreover, one has QW Γ = T⋆∈Γ Q⋆ for all bounded subsets Γ of N(Q). Proof. One easily checks that the intersection of nonempty subsets C of S that sat- isfy the condition of statement (1) of Corollary 4.2 itself satisfies the same condition and is nonempty given the boundedness condition on Γ. Therefore, since S⋆ ⊆ S⋆′ if and only if ⋆ ≥ ⋆′ for all ⋆, ⋆′ ∈ C(S), statement (1) follows. Statement (2) is proved in a similar fashion. (cid:3) For any self-map ⋆ of a set S we let Fix(⋆) = {x ∈ S : x⋆ = x}. Lemma 4.4. Let S be a bounded complete poset, and let + be a preclosure on S that is bounded above by some closure operation on S. (1) There exists a smallest closure operation ⋆ on S that is larger that +, and one has S⋆ = Fix(+) and x⋆ =V{y ∈ Fix(+) : y ≥ x} for all x ∈ S. or transfinite) ordinals α. One has +α = ⋆ for all α ≫ 0. (2) Define x+0 = x and x+α =W{(x+β )+ : β < α} for all x ∈ S and all (finite PREQUANTALES 15 (3) If S = M is a bounded complete and near residuated ordered magma and xy+ ≤ (xy)+ and x+y ≤ (xy)+ for all x, y ∈ M , then ⋆ is a nucleus on M . Proof. (1) The set {y ∈ Fix(+) : y ≥ x} is nonempty and bounded by the hypothesis on +. Define x⋆ = V{y ∈ Fix(+) : y ≥ x} for all x ∈ S. Clearly ⋆ is a preclosure on S with Fix(+) ⊆ S⋆. For the reverse inclusion note that (x+)⋆ =^{y ∈ Fix(+) : y ≥ x+} =^{y ∈ Fix(+) : y ≥ x} = x⋆ and therefore x ≤ x+ ≤ (x+)⋆ = x⋆, whence S⋆ ⊆ Fix(+). Therefore (x⋆)⋆ =^{y ∈ Fix(+) : y ≥ x⋆} =^{y ∈ Fix(+) : y ≥ x} = x⋆, whence ⋆ is a closure operation on S. It is then clear that ⋆ is the smallest closure operation on S that is larger than +. (2) This follows readily from statement (1). (3) Let a, x ∈ M . Let z be any element of M such that z ≥ ax and z+ = z. Set y = a\z. Then y ≥ x and y+ = (a\z)+ ≤ a\z+ = a\z = y, hence y+ = y. Moreover, we have ay ≤ z. Therefore we have ax⋆ ≤^{ay : y ≥ x, y+ = y} ≤^{z ∈ M : z ≥ ax, z+ = z} = (ax)⋆. By symmetry we also have x⋆a ≤ (xa)⋆, whence ⋆ is a nucleus. (cid:3) Let S be a poset and Γ ⊆ C(S). Define partial self-maps ⊓ Γ and ⊔ Γ of S by x⊓ Γ =^{x⋆ : ⋆ ∈ Γ}, x⊔ Γ =^{y ∈ S : y ≥ x and ∀⋆ ∈ Γ (y⋆ = y)}, respectively, for all x ∈ S such that the respective infima exist. Lemma 4.5. Let S be a near sup-lattice (resp., bounded complete poset). Then C(S) is a complete lattice (resp., bounded complete poset), and one has the following. Γ of C(S). Proof. subsets) Γ of C(S). (1) VC(S) Γ = ⊓ Γ and S⊓ Γ ⊇ S⋆∈Γ S⋆ for all subsets (resp., all nonempty (2) WC(S) Γ = ⊔ Γ and S⊔ Γ =T⋆∈Γ S⋆ for all subsets (resp., bounded subsets) (1) ⊓ Γ is a preclosure on S, and for all x ∈ S one has (x⊓ Γ)⊓ Γ =V⋆∈Γ(x⊓ Γ)⋆ ≤ V⋆∈Γ x⋆ = x⊓ Γ. Therefore ⊓ Γ is a closure operation on S, from which it follows that ⊓ Γ =VC(S) Γ. The given inclusion follows immediately from the definition of ⊓ Γ. (2) Define x+ = W{x⋆ : ⋆ ∈ Γ} for all x ∈ S, which exists since the x⋆ above by WC(S) Γ, and one has x+ = x if and only if x⋆ = x for all ⋆ ∈ Γ. Therefore ⊔ Γ is a closure operation on S by Lemma 4.4(1), and it follows easily that ⊔ Γ = WC(S) Γ. Finally, that S⊔ Γ = T⋆∈Γ S⋆ follows from are bounded above by xWC(S) Γ. Clearly + is a preclosure on S bounded Corollary 4.3(1). (cid:3) 16 JESSE ELLIOTT The following result generalizes [29, Proposition 3.1.3] and [8, Example 1.5]. Proposition 4.6. Let M be a near sup-magma (resp., bounded complete ordered magma). Then N(M ) is a complete lattice (resp., bounded complete poset), and one has the following. subsets) Γ of N(M ). (1) VN(M) Γ = ⊓ Γ and M⊓ Γ ⊇S⋆∈Γ M ⋆ for all subsets (resp., all nonempty (2) WN(M) Γ = ⊔ Γ and M⊔ Γ = T⋆∈Γ M ⋆ for all subsets (resp., all bounded subsets) Γ of N(M ) if M is a near prequantale (resp., bounded complete and near residuated). Proof. (1) By Lemma 4.5 it suffices to show that ⊓ Γ is a nucleus. We have xy⊓ Γ = x ^⋆∈Γ y⋆ ≤ ^⋆∈Γ xy⋆ ≤ ^⋆∈Γ (xy)⋆ = (xy)⊓ Γ, and similarly x⊓ Γy ≤ (xy)⊓ Γ, for all x, y ∈ M , whence ⊓ Γ is a nucleus. (2) By Lemma 4.5 it suffices to show that ⊔ Γ is a nucleus. Since d = ⊔∅ is a nucleus we may assume that Γ is nonempty. Let + be the preclosure on M defined in the proof of Lemma 4.5(2). For all x, y ∈ M we have xy+ = x _⋆∈Γ y⋆ = _⋆∈Γ xy⋆ ≤ _⋆∈Γ (xy)⋆ = (xy)+, and similarly x+y ≤ (xy)+. Therefore ⋆ is a nucleus by Lemma 4.4(3). (cid:3) Note that, if an ordered magma M is a meet semilattice, then N(M ) is also a meet semilattice, and statement (1) of Proposition 4.6 holds for all nonempty finite subsets Γ of M . The proof is similar to that of Proposition 4.6. If N is a submagma of a near sup-magma M , then for any nucleus ⋆ on N we may define indM (⋆) =^N(M){⋆′ ∈ N(M ) : ⋆′N = ⋆}, indM (⋆) =_N(M){⋆′ ∈ N(M ) : ⋆′N = ⋆}, both of which are nuclei on M . We say that a subset X of an ordered magma M is saturated if whenever xy ∈ X for some x, y ∈ M that are not annihilators of M one has x ∈ X and y ∈ X. Proposition 4.7. Let M be a near sup-magma and ⋆ a nucleus on a submagma N of M . (1) Suppose that {⋆′ ∈ N(M ) : ⋆′N = ⋆} is nonempty. Then indM (⋆) is the smallest nucleus on M whose restriction to N is equal to ⋆. If M is bounded complete and near residuated, then indM (⋆) is the largest nucleus on M whose restriction to N is equal to ⋆, and one has ⋆′N = ⋆ if and only if indM (⋆) ≤ ⋆′ ≤ indM (⋆). xindM (⋆) =^{y ∈ M : y ≥ x and ∀z ∈ N (z ≤ y ⇒ z⋆ ≤ y)} for all x ∈ M , and indM (⋆)N = ⋆. (2) If M is a near prequantale and N is a sup-spanning subset of M , then PREQUANTALES 17 (3) If N is a saturated downward closed subset of M and M is bounded above, then xindM (⋆) =(cid:26) x⋆ for all x ∈ M , and indM (⋆)N = ⋆. if x ∈ N W M otherwise Proof. (1) This follows easily from Proposition 4.6. (2) First we define x+ = W{y⋆ : y ∈ N and y ≤ x} for all x ∈ M . As N is a sup-spanning subset of M one has x ≤ x+ for all x ∈ M . Since + is order-preserving it follows that + is a preclosure on M . One has y+ = y if and only if ∀z ∈ N (z ≤ y ⇒ z⋆ ≤ y), whence by Lemma 4.4(1) it follows that the operation ⋆M on M defined by az⋆ ≤ _z∈N : z≤y (az)⋆ ≤ _w∈N : w≤ay x⋆M =^{y ∈ M : y ≥ x and ∀z ∈ N (z ≤ y ⇒ z⋆ ≤ y)} is a closure operation on M . Now, for all a ∈ N and x ∈ M one has ay+ = _z∈N : z≤y Since N is a sup-spanning subset of M , it follows that xy+ ≤ (xy)+, and by symmetry x+y ≤ (xy)+, for all x, y ∈ M . Thus ⋆M is a nucleus by Lemma 4.4(3). Now, + is the smallest preclosure on M whose restriction to N is ⋆, and by Lemma 4.4(1) the operation ⋆M is the smallest closure operation on M that is larger than +, and one has (⋆M )N = ⋆. It follows that ⋆M is the smallest closure operation on M whose restriction to N is ⋆. Therefore, since ⋆M is a nucleus we must have ⋆M = indM (⋆) by statement (1). w⋆ = (ay)+. (3) It suffices to show that the operation defined in the statement is a nucleus. This is easy to check. (cid:3) Remark 4.8. Let ⋆1 and ⋆2 be nuclei on an ordered magma M . Then ⋆1∨⋆2 exists in N(M ) and equals the n-fold composition ⋆ = ··· ◦ ⋆2 ◦ ⋆1 ◦ ⋆2 for some positive integer n if and only if ⋆ is larger than the n-fold composition ··· ◦ ⋆1 ◦ ⋆2 ◦ ⋆1. In particular, ⋆1 ∨ ⋆2 = ⋆1 ◦ ⋆2 for two nuclei ⋆1 and ⋆2 if and only if ⋆1 ◦ ⋆2 ≥ ⋆2 ◦ ⋆1. The operation ◦ on closure operations, though not necessarily closed, is studied in [28, Section 2] and [30]. 5. Divisorial closure operations In this section we generalize the generalized divisorial closure semistar operations [28, Example 1.8(2)], and we use this to derive a characterization of the simple near prequantales that is of a different vein than the characterization in [22, Theorem 2.5] of the simple quantales. If a is an element of an ordered magma M and there exists a largest nucleus ⋆ on M such that a⋆ = a, then we denote it by v(a) = vM (a) and call it divisorial closure on M with respect to a. For example, ifW M exists then v(W M ) exists and equals e =W N(M ). The proof of the following result is straightforward. Proposition 5.1. Let M be an ordered magma. 18 JESSE ELLIOTT a⋆ = a if and only if ⋆ ≤ v(a). exists in N(M ) and av′ (1) Suppose that v(a) exists, where a ∈ M . Then, for all ⋆ ∈ N(M ), one has (2) Let a ∈ M . Then v(a) exists if and only if v′ = W{⋆ ∈ N(M ) : a⋆ = a} (3) If ⋆ is a nucleus on M such that v(a) exists for all a ∈ M ⋆, then ⋆ = = a, in which case v(a) = v′. V{v(a) : a ∈ M ⋆}. The next proposition follows from Propositions 5.1 and 4.6(2). Proposition 5.2. Let Q be a near prequantale. Then v(a) exists and equals W{⋆ ∈ N(Q) : a⋆ = a} for all a ∈ Q, and one has ⋆ =V{v(a) : a ∈ Q⋆} for all ⋆ ∈ N(Q). Corollary 5.3. The following are equivalent for any near prequantale Q. (1) Q is simple. (2) The only nuclei on Q are d and e. (3) v(a) = d for all a <W Q. The divisorial closure operations generalize as follows. If S is a subset of an ordered magma M and there exists a largest nucleus ⋆ on M such that S ⊆ M ⋆, then we denote it by v(S) = vM (S) and call it divisorial closure on M with respect to S. Note that v(a) = v({a}). Proposition 5.1 generalizes as follows. Proposition 5.4. Let S be a subset of an ordered magma M . (1) Suppose that v(S) exists. Then, for all ⋆ ∈ N(M ), one has S ⊆ M ⋆ if and (2) The following conditions are equivalent. only if ⋆ ≤ v(S). (a) v(S) exists. (b) v′ =W{⋆ ∈ N(M ) : S ⊆ M ⋆} exists in N(M ) and S ⊆ M v′ (c) There is a smallest subset C of M containing S such that C = M ⋆ for . some ⋆ ∈ N(M ). Moreover, if the above condtions hold, then v′ = ⋆ = v(S). (3) Suppose that S =Sλ Tλ and v(Tλ) exists for all λ. Then v(S) exists if and only if Vλ v(Tλ) exists, in which case they are equal. (4) ⋆ = v(M ⋆) for any nucleus ⋆ on M . Proposition 5.5. Let Q be a near prequantale. Then v(S) exists for any subset S of Q and equals V{v(a) : a ∈ S}. Proposition 5.6. Let Q be a bounded complete and near residuated ordered magma. If v(S) exists for some subset S of Q, then v(T ) exists for any subset T of Q containing S. If Q is a unital near quantale then we can determine an explicit formula for the divisorial closure operations v(a). Lemma 5.7. Let M be an ordered monoid and a ∈ M . If for any x ∈ M there is a largest element x⋆ of M such that rxs ≤ a implies rx⋆s ≤ a for all r, s ∈ M , then v(a) exists and equals ⋆. Proof. First we show that the map ⋆ defined in the statement of the lemma is a nucleus on M . Let x, y ∈ M . If x ≤ y, then rys ≤ a ⇒ rxs ≤ a ⇒ rx⋆s ≤ a, whence x⋆ ≤ y⋆. Likewise rxs ≤ a ⇒ rxs ≤ a, whence x ≤ x⋆. Moreover, one has rxs ≤ a ⇒ rx⋆s ≤ a ⇒ r(x⋆)⋆s ≤ a, whence (x⋆)⋆ ≤ x⋆, whence equality holds. PREQUANTALES 19 Thus ⋆ is a closure operation on M . Next, note that rxys ≤ a ⇒ rx⋆ys ≤ a ⇒ rx⋆y⋆s ≤ a, whence x⋆y⋆ ≤ (xy)⋆. Thus ⋆ is a nucleus on M . Next we observe that 1a1 ≤ a, which implies 1a⋆1 ≤ a, whence a⋆ = a. Finally, let ⋆′ be any nucleus on M with a⋆′ = a. Then for any x ∈ M we have rxs ≤ a ⇒ rx⋆′ s ≤ a⋆ = a for all r, s ∈ M , whence x⋆′ ≤ x⋆. Thus ⋆′ ≤ ⋆. It follows that ⋆ is the largest nucleus on M such that a⋆ = a, so v(a) exists and equals ⋆. Proposition 5.8. Let Q be a unital near quantale and a ∈ Q. Then one has (cid:3) xv(a) =_{y ∈ Q : ∀r, s ∈ Q (rxs ≤ a ⇒ rys ≤ a)} for all x ∈ Q; alternatively, xv(a) is the largest element of Q such that rxs ≤ a implies rxv(a)s ≤ a for all r, s ∈ Q. Proof. Let x⋆ = W{y ∈ Q : ∀r, s ∈ Q (rxs ≤ a ⇒ rys ≤ a)} for all x ∈ Q. Since Q is a near quantale, one has y ≤ x⋆ if and only if rxs ≤ a implies rys ≤ a for all r, s ∈ Q. Therefore by Lemma 5.7 one has ⋆ = v(a). Corollary 5.9. A unital near quantale Q is simple if and only if, for any x, y, a ∈ Q with a <W Q, one has x ≤ y if and only if rys ≤ a implies rxs ≤ a for all r, s ∈ Q. Corollary 5.10. A multiplicative lattice Q is simple if and only if Q = {0, 1}. Proof. We may assume 0 < 1. Note that r1s ≤ 0 implies rxs = rsx ≤ 0 for all x, r, s ∈ Q, whence by Corollary 5.9 one has x ≤ 1 for all x ∈ Q. Therefore, for any a ∈ Q, one has (a∨ x)(a∨ y) = a2 ∨ ay ∨ xa∨ xy ≤ a∨ xy for all x, y ∈ Q. It follows that the map x 7−→ a ∨ x is a nucleus on Q, whence a ∨ x = x for all x if a < 1. It follows, then, that Q = {0, 1}. (cid:3) (cid:3) Proposition 5.8 and Corollary 5.9 may be generalized to any near prequantale as follows. Let M be any magma. For any r ∈ M , define self-maps Lr and Rr of M by Lr(x) = rx and Rr(x) = xr for all x ∈ M , which we call translations. Let Lin(M ) denote the submonoid of the monoid of all self-maps of M generated by the translations. If M is a monoid then any element of Lin(M ) can be written in the form Lr ◦ Rs = Rs ◦ Lr for some r, s ∈ M . The proofs above readily generalize to yield the following. Proposition 5.11. Let Q be a near prequantale and a ∈ Q. Then one has xv(a) =_{y ∈ Q : ∀f ∈ Lin(Q) (f (x) ≤ a ⇒ f (y) ≤ a)} for all x ∈ Q; alternatively, xv(a) is the largest element of Q such that f (x) ≤ a implies f (xv(a)) ≤ a for all f ∈ Lin(Q). Corollary 5.12. A near prequantale Q is simple if and only if, for any x, y, a ∈ Q with a < W Q, one has x ≤ y if and only if f (y) ≤ a implies f (x) ≤ a for all f ∈ Lin(Q). The following notion of cyclic elements generalizes [29, Definition 3.3.2]. Let M be an ordered magma. We say that a ∈ M is cyclic if xy ≤ a implies yx ≤ a for all x, y ∈ M . If M is associative, then a is cyclic if and only if x1x2x3 ··· xn ≤ a implies x2x3 ··· xnx1 ≤ a for any positive integer n and any x1, x2, . . . , xn ∈ M . If M is a (resp., M R near residuated, then we let M L a ) denote the set of all x ∈ M for which a/x (resp., x\a) is defined and we let M LR a = M L a . In that case a is cyclic if and only if M L . The following result a a and a/x = x\a for all x ∈ M LR a ∩ M R a = M R 20 JESSE ELLIOTT generalizes [9, Lemma 3.35] and the fact that I v(D) = (I −1)−1 for any integral domain D with quotient field F and any I ∈ K(D), where I −1 = (D :F I). Proposition 5.13. Let M be a residuated ordered monoid, or a near residuated ordered monoid such thatW M exists. Let s be any element of M such that s =W M if W M exists. Then v(a) exists for any cyclic element a of M and is given by xv(a) =(cid:26) a/(a/x) =W{y ∈ M LR s a : a/x ≤ a/y} if x ∈ M LR a otherwise a a a a a . Therefore, if x /∈ M LR : a/x ≤ a/y} for x ∈ M LR for all x ∈ M . and x⋆ =W M otherwise. The Proof. For any x ∈ M , let x⋆ = a/(a/x) if x ∈ M LR identity a/(a/x) = W{y ∈ M LR is easy to check. We claim that x⋆ is the largest element y of M such that rxs ≤ a implies rys ≤ a for all r, s ∈ M . By Lemma 5.7 the proposition follows from this claim. First, note that rxs ≤ a implies srx ≤ a, which implies that x ∈ M LR , then W M is the largest element y of M described above. Suppose, on the other hand, . Then rxs ≤ a ⇒ srx ≤ a ⇒ sr ≤ a/x ⇒ x⋆sr = (a/(a/x))sr ≤ that x ∈ M LR a ⇒ rx⋆s ≤ a. Suppose that y′ is any element of M such that rxs ≤ a ⇒ ry′s ≤ a for all r, s ∈ M . Then since (a/x)x1 ≤ a we have (a/x)y′ ≤ a, whence y′(a/x) ≤ a and thus y′ ≤ a/(a/x) = x⋆. This proves our claim. Corollary 5.14. A near multiplicative lattice Q is simple if and only if x/y exists and x/(x/y) = y for all x, y ∈ Q such that x, y <W Q. In particular, a multiplica- tive lattice Q is simple if and only if x/(x/y) = y for all x, y <W Q. Example 5.15. Let G be a bounded complete partially ordered abelian group. Let G[∞] = G ∐ {∞}, where a∞ = ∞a = ∞ and a < ∞ for all a ∈ G. Then G[∞] is a simple near multiplicative lattice, by Corollary 5.14. However, the multiplicative lattice G[±∞] = G[∞] ∐ {−∞}, where −∞ = V G is an annihilator of G[∞], is not simple. For example, if D is a DVR with quotient field F , then K(D) ∼= Z[∞] is simple, but ModD(F ) ∼= Z[±∞] is not simple. Finally we generalize the well-known fact that I v(D) =T{xD : x ∈ F and xD ⊇ I} for any integral domain D with quotient field F and any I ∈ K(D). Proposition 5.16. Let Q be an associative unital semi-U-lattice, and let a ∈ Q. Then v(a) exists and xv(a) =V{uav : u, v ∈ U(Q) and x ≤ uav} for all x ∈ Q. Proof. For all x ∈ Q, let x⋆ =V{uav : u, v ∈ U(Q), x ≤ uav}, which exists since Q for all x ∈ Q and w ∈ U(Q) one has wx⋆ =V{wuav : u, v ∈ U(Q), wx ≤ wuav} = V{u′av : u′, v ∈ U(Q), wx ≤ u′av} = (wx)⋆, and likewise x⋆w = (xw)⋆. Thus ⋆ is a nucleus on Q by Proposition 3.4. Now let ⋆′ ∈ N(Q) with a⋆′ = a, and let x ∈ Q. If x ≤ uav for u, v ∈ U(Q), then x⋆′ ≤ x⋆. Therefore ⋆′ ≤ ⋆. It follows that v(a) exists and equals ⋆. is bounded complete. Clearly ⋆ is a closure operation on Q with a⋆ = a. Moreover, v = uav, whence x⋆′ ≤ ua⋆′ a (cid:3) (cid:3) 6. Finitary nuclei and precoherence A closure operation ⋆ : S −→ S on a poset S may not be Scott continuous, even though the corestriction S⋆⋆ : S −→ S⋆ of ⋆ to S⋆ is always sup-preserving (hence Scott continuous). If ⋆ : S −→ S is Scott continuous then we will say that ⋆ is finitary. (Such closure operations are often said to be algebraic.) Equivalently this PREQUANTALES 21 Table 2. The posets N(M ) and Nf (M ) Nf (M ) is . . . N(M ) is . . . complete bounded complete meet semilattice If M is . . . near sup-magma bounded complete meet semilattice Scott-topological dcpo magma Scott-topological bdcpo magma Scott-topological near sup-magma Scott-topological, bounded complete bounded complete bounded complete algebraic meet semilattice complete bounded complete complete meet semilattice complete meet semilattice operations, ideal systems, and module systems. means that (W ∆)⋆ =W(∆⋆), or W(∆⋆) ∈ S⋆, for any directed subset ∆ of S such thatW ∆ exists. Finitary closure operations generalize finite type star and semistar For any poset S, let Cf (S) denote the poset of all finitary closure operations on S, and for any ordered magma M , let Nf (M ) = N(M ) ∩ Cf (M ) denote the poset of all finitary nuclei on M . In this section we study properties of the poset Nf (M ). A summary of these results and the results for N(M ) from the previous section is provided in Table 2. Recall that an element x of a poset S is said to be compact if, whenever x ≤W ∆ for some directed subset ∆ of S such thatW ∆ exists, one has x ≤ y for some y ∈ ∆. Any minimal element of S, for example, is compact. We let K(S) denote the set of all compact elements of S. Example 6.1. (1) For any set X, the set K(2X ) of compact elements of the complete lattice 2X is equal to the set of all finite subsets of X. (2) For any algebra A over a ring R, the compact elements of the quantale ModR(A) of all R-submodules of A are precisely the finitely generated R- submodules of A. A element x of a poset S is algebraic if x is the supremum of a set of compact elements of S, or equivalently, if one has x = W{y ∈ K(S) : y ≤ x}. A poset is said to be algebraic if all of its elements are algebraic. Note, however, that an algebraic complete lattice is said to be an algebraic lattice. For example, for any set X, the poset 2X is an algebraic lattice, and the poset 2X−{∅} is an algebraic near sup-lattice with K(2X−{∅}) = K(2X )−{∅}. For any set X and any set Γ of self-maps of X, let hΓi denote the submonoid generated by Γ of the monoid of all self-maps of X under composition. Lemma 6.2. Let S be a poset. (1) If S is a dcpo (resp., bdcpo) then Cf (S) is complete (resp., bounded com- plete) and one has WCf (S) Γ =WC(S) Γ and xWC(S) Γ =W{xγ : γ ∈ hΓi} for all x ∈ S and for any subset (resp., any bounded subset) Γ of Cf (S). C(S). (2) If S is a near sup-lattice, then Cf (S) is a sub sup-lattice of the sup-lattice 22 JESSE ELLIOTT (3) If S is an algebraic meet semilattice, then Cf (S) is a sub meet semilattice of the meet semilattice C(S). Proof. (1) Define a preclosure σ of S by xσ = W{xγ : γ ∈ hΓi} for all x ∈ S. The map σ is defined because the set {xγ : γ ∈ hΓi} is directed (and bounded above if Γ is bounded above) for any x ∈ S. Let x ∈ S, and let X = {xγ : γ ∈ hΓi}. Then (W X)⋆ = W{(xγ)⋆ : γ ∈ hΓi} ≤ W X for all ⋆ ∈ Γ, whence (W X)γ = W X, and therefore (xσ)γ = xσ, for all γ ∈ hΓi. Therefore (xσ)σ = xσ, so σ is a closure operation on S, and clearly σ =WC(S) Γ. It remains only to show that σ is finitary, since it will follow that σ =WCf (S) Γ. Let ∆ be a directed subset of S. Then we have (W ∆)σ = W{(W ∆)γ : γ ∈ hΓi} = W{W(∆γ) : γ ∈ hΓi} = W{xγ : x ∈ ∆, γ ∈ hΓi} =W(∆σ). Thus σ is finitary. (2) This follows from (1). (3) Let ⋆, ⋆′ ∈ Cf (S). By Lemma 4.5 it suffices to show that ⋆ ∧ ⋆′ ∈ C(S) is finitary. Let ∆ be a directed subset of S such that x = W ∆ exists, and let z = x⋆∧⋆′ . Let u be any upper bound of ∆⋆∧⋆′ . We claim that u ≥ z. To show this, let t be any . Then t ≤ x⋆ = W(∆⋆) and compact element of S with t ≤ z = x⋆ ∧ x⋆′ = W(∆⋆′ t ≤ x⋆′ for some y, y′ ∈ ∆. Since ∆ is directed, we may choose w ∈ ∆ with y, y′ ≤ w. Then t ≤ w⋆ ∧ w⋆′ = w⋆∧⋆′ ≤ u. Therefore, taking the supremum over all compact t ≤ z, we see that z ≤ u. Therefore (W ∆)⋆∧⋆′ ). Thus ⋆ ∧ ⋆′ is finitary. ), whence t ≤ y⋆ and t ≤ (y′)⋆′ = z =W(∆⋆∧⋆′ . Note first that z is an upper bound of ∆⋆∧⋆′ (cid:3) Proposition 6.3. Let M be an ordered magma. (1) If M is a Scott-topological dcpo magma (resp., Scott-topological bdcpo magma) then Nf (M ) is complete (resp., bounded complete) and one hasWNf (M) Γ = WN(M) Γ and xWN(M ) Γ =W{xγ : γ ∈ hΓi} for all x ∈ M and for any subset (resp., any bounded subset) Γ of Nf (M ). (2) If M is a Scott-topological near sup-magma, then Nf (M ) is a sub sup-lattice of the sup-lattice N(M ). (3) If M is an algebraic meet semilattice, then Nf (M ) is a sub meet semilattice of the meet semilattice N(M ). Proof. By the lemma it suffices to observe that the closure operation x 7−→W{xγ : γ ∈ hΓi}, when defined for Γ ⊆ Cf (M ), is a nucleus if M is Scott-topological. (cid:3) The remainder of this section is devoted to proving Theorem 1.6 of the intro- duction. Lemma 6.4. Let S be a join semilattice. Then x ∈ S is compact if and only if, whenever x ≤W X for some subset X of S such that W X exists, one has x ≤W Y for some finite subset Y of X. Moreover, K(S) is closed under finite suprema. Lemma 6.5. Let ⋆ be a closure operation on a poset S. (1) If x⋆ = x ∨W{y⋆ : y ∈ K(S), y ≤ x} for all x ∈ S, then ⋆ is finitary. (2) If ⋆ is finitary and S is an algebraic join semilattice, then x⋆ =W{y⋆ : y ∈ K(S), y ≤ x} for all x ∈ S, and one has K(S⋆) ⊆ K(S)⋆. PREQUANTALES 23 (3) If ⋆ is finitary and S is a bdcpo, then K(S)⋆ ⊆ K(S⋆). (4) If ⋆ is finitary, and if S is an algebraic bounded complete join semilat- tice (resp., algebraic near sup-lattice, algebraic lattice), then so is S⋆, and K(S⋆) = K(S)⋆. Proof. (1) Let ∆ be a directed subset of S such that x =W ∆ exists. We wish to show that W(∆⋆) exists and equals x⋆. Clearly x⋆ is an upper bound of ∆⋆, and if w is any upper bound of ∆⋆, then we claim that x⋆ ≤ w. Let y ∈ K(S) with y ≤ x. Then y ≤ z for some z ∈ ∆. Therefore y⋆ ≤ z⋆, where z⋆ ∈ ∆⋆. Thus we have y⋆ ≤ w. Taking the supremum over all such y, we see that W{y⋆ : y ∈ K(S), y ≤ x} ≤ w. Moreover, we have x = W ∆ ≤ w, and therefore x⋆ = x ∨W{y⋆ : y ∈ K(S), y ≤ x} ≤ w, as claimed. (2) Let x ∈ S. One has x = W ∆, where ∆ = {y ∈ K(S) : y ≤ x} is directed by Lemma 6.4. Therefore x⋆ = W(∆⋆). Suppose that x ∈ K(S⋆). Then x = W(∆⋆) and ∆⋆ ⊆ S⋆ is directed, so x ≤ y⋆ for some y ∈ ∆, whence x = y⋆. Therefore K(S⋆) ⊆ K(S)⋆. (3) Let x ∈ K(S). Suppose that x⋆ ≤ WS⋆ ∆ for some directed subset ∆ of S⋆ such that WS⋆ ∆ exists. Then ∆ is directed and bounded above, whence WS ∆ exists. It follows that WS⋆ ∆ = (WS ∆)⋆ =WS(∆⋆) =WS ∆. Therefore x ≤ x⋆ ≤WS ∆, so x ≤ y for some y ∈ ∆, whence x⋆ ≤ y⋆ = y. Thus we have x⋆ ∈ K(S⋆). (4) Since K(S)⋆ ⊆ K(S⋆) by (3), it follows from (2) that S⋆ is algebraic and K(S)⋆ = K(S⋆). Since S⋆ is bounded complete (resp., near sup-complete, complete) by Lemma 3.1(4), it follows that S⋆ is an algebraic bounded com- plete join semilattice (resp., algebraic near sup-lattice, algebraic lattice). (cid:3) For any closure operation ⋆ on a bounded complete poset S, let ⋆f denote the operation on S defined by x⋆f =W{y⋆ : y ∈ K(S) and y ≤ x} for all x ∈ S. Proposition 6.6. Let ⋆ be a closure operation on an algebraic bounded complete join semilattice S. Then ⋆f is the largest finitary closure operation on S that is smaller than ⋆, one has x⋆f = x⋆ for all x ∈ K(S), and K(S⋆f ) = K(S)⋆f . Proof. Clearly ⋆f is a preclosure on S. Let x ∈ S. The set {y⋆ : y ∈ K(S), y ≤ x} is directed by Lemma 6.4. Therefore if z ∈ K(S) and z ≤ x⋆f then z ≤ y⋆ for some y ∈ K(S) such that y ≤ x, in which case z⋆ ≤ y⋆. Therefore (x⋆f )⋆f =W{z⋆ : z ∈ K(S), z ≤ x⋆f} ≤ W{y⋆ : y ∈ K(S), y ≤ x} = x⋆f . Thus ⋆ is a closure operation on S, and the rest of the proposition follows from Lemma 6.5. (cid:3) Generalizing [29, Definition 4.1.1], we say that an ordered magma M is preco- herent if M is algebraic and K(M ) is closed under multiplication, and we say that M is coherent if M is precoherent and unital with 1 compact. Example 6.7. (1) For any magma M , the prequantale 2M is precoherent, and if M is unital then 2M is coherent. (2) A K-lattice is a precoherent multiplicative lattice. We say that a near K- lattice is a precoherent near multiplicative lattice. For example, if M is a commutative monoid, then 2M is a K-lattice and 2M−{∅} is a near K-lattice. 24 JESSE ELLIOTT of all nonzero D-submodules of F is a near K-lattice. (3) For any integral domain D with quotient field F , the ordered monoid K(D) (4) Let A be an algebra over a ring R. The quantale ModR(A) of all R- submodules of A is precoherent, and it is coherent if and only if A is finitely generated as an R-module. In particular, if R and A are commutative then ModR(A) is a K-lattice. The following result generalizes the corresponding well-known facts for star and semistar operations and ideal and module systems. Theorem 6.8. If ⋆ is a nucleus on a precoherent semiprequantale (resp., preco- herent near prequantale, precoherent prequantale) Q, then ⋆f is the largest finitary nucleus on Q that is smaller than ⋆, the ordered magma Q⋆f is also a precoherent semiprequantale (resp., precoherent near prequantale, precoherent prequantale), and K(Q⋆f ) = K(Q)⋆f . Proof. By Proposition 6.6, to prove the first claim we need only show that ⋆f is a nucleus. Let x ∈ Q and a ∈ K(Q). We have (ax)⋆f = W{y⋆ : y ∈ K(Q), y ≤ ax} and ax⋆f = W{az⋆ : z ∈ K(Q), z ≤ x}. Let z ∈ K(Q) with z ≤ x, and let y = az. By our hypotheses on Q we have y ∈ K(Q), and y = az ≤ ax. Therefore az⋆ ≤ y⋆ ≤ (ax)⋆f . Taking the supremum over all such z ∈ K(Q) we see that ax⋆f ≤ (ax)⋆f . By symmetry we also have x⋆f a ≤ (xa)⋆f . By Proposition 3.4, then, it follows that ⋆f is a nucleus. To prove the second claim we may assume ⋆ = ⋆f is finitary. Then Q⋆ is a an algebraic semiprequantale (resp., algebraic near prequantale, algebraic prequantale) by Lemma 6.5(3) and Proposition 3.11. Let x, y ∈ K(Q⋆), and suppose that x ⋆ y ≤ WQ⋆ ∆ for some directed subset ∆ of Q⋆ such thatWQ⋆ ∆ exists. ThenWQ ∆ exists, and xy ≤WQ⋆ ∆ = (WQ ∆)⋆ =WQ ∆. Therefore, since xy ∈ K(Q), we have xy ≤ z, whence x ⋆ y ≤ z, for some z ∈ ∆. Thus x ⋆ y ∈ K(Q⋆) and Q⋆ is precoherent. Finally, the third claim follows from Proposition 6.6. (cid:3) Corollary 6.9. Let ⋆ be a nucleus on a precoherent near prequantale Q. Then the nucleus t(a) = v(a)f for any a ∈ Q is the largest finitary nucleus ⋆ on Q such that a⋆ = a, and one has ⋆f = V{t(a) : a ∈ Q⋆f}. More generally, the nucleus t(S) = v(S)f for any subset S of Q is the largest finitary nucleus ⋆ on Q such that S ⊆ Q⋆, and one has ⋆f = t(Q⋆f ). The following result reveals a few subclasses of the precoherent semiprequantales. Proposition 6.10. Let M be an ordered monoid. Then M is a U-lattice (resp., near U-lattice, semi-U-lattice) with 1 compact if and only if M is a coherent pre- quantale (resp., coherent near prequantale, coherent semiprequantale) such that ev- ery compact element of M is the supremum of a subset of U(M ); in that case the compact elements of M are precisely the suprema of the finite subsets of U(M ). Proposition 6.10 follows easily from the following lemma, which generalizes the fact that every invertible fractional ideal of an integral domain is finitely generated. Lemma 6.11. The following are equivalent for any unital ordered magma M . (1) 1 ∈ K(M ). (2) U(M ) ⊆ K(M ). (3) U(M ) ∩ K(M ) 6= ∅. PREQUANTALES 25 In the remainder of this section we apply the theory of nuclei to construct a representation of any precoherent near prequantale (resp., precoherent prequan- tale) as the ideal completion of some multiplicative semilattice (resp., prequantic semilattice), and conversely a representation of any multiplicative semilattice (resp., prequantic semilattice) as the ordered magma of the compact elements of some pre- coherent near prequantale (resp., precoherent prequantale). These representations yield appropriate category equivalences. A subset I of a poset S is said to be an ideal of S if I is a directed downward closed subset of S. For any x ∈ S, the set ↓x = {y ∈ S : y ≤ x} is an ideal of S called the principal ideal generated by x. The ideal completion Idl(S) of S is the set of all ideals of S partially ordered by the subset relation. If S is a join semilattice, then, for any nonempty subset X of S, we let ↓X = {y ∈ S : y ≤_ T for some finite nonempty T ⊆ X}. Lemma 6.12. Let S be a join semilattice. containing X. (1) For any nonempty subset X of S, the set ↓ X is the smallest ideal of S (2) The operation ↓ is a finitary closure operation on the algebraic near sup- (3) If S has a least element, then the operation ↓ extends uniquely to a finitary lattice 2S−{∅} with im↓ = Idl(S). closure operation on the algebraic sup-lattice 2S such that ↓∅ =V S. Proof. It is straightforward to show that ↓X is the smallest ideal of S containing X and ↓ is a closure operation on 2S−{∅}. To show that ↓ is finitary, we show that ↓X = S ↓{T : T ⊆ X finite and nonempty} for any X ∈ 2S−{∅}. Let x ∈↓X. Then x ≤ W T , where T ⊆ X is finite and nonempty. But W T ∈↓T , so x ∈↓T . This proves (1) and (2), and (3) is clear. (cid:3) The following result generalizes [10, Proposition I-4.10]. Proposition 6.13. Let L be an algebraic near sup-lattice and S a join semilattice. inverse acting by x 7−→ (↓x) ∩ K(L). (1) Idl(S) is an algebraic near sup-lattice. (2) K(L) is a sub join semilattice of L. (3) The map S −→ K(Idl(S)) acting by x 7−→↓x is a poset isomorphism. (4) The map Idl(K(L)) −→ L acting by I 7−→W I is a poset isomorphism with Proof. By Lemma 6.12 the map ↓ is a finitary nucleus on the algebraic near sup- lattice 2S−{∅} with im ↓ = Idl(S). By Lemma 6.5, then, it follows that Idl(S) is an algebraic near sup-lattice with K(Idl(S)) =↓(K(2S−{∅})) = {↓x : x ∈ S}. This implies (1) and (3). Next, K(L) is nonempty and therefore a sub join semilattice of L by Lemma 6.4. This proves (2). Finally, the map f : I 7−→ W I in (4) is well-defined because L is a near sup-lattice, the map f is clearly order-preserving, and the map x 7−→ (↓x) ∩ K(L) is an order-preserving inverse to f because L is algebraic and K(L) is a join semilattice. (cid:3) 26 JESSE ELLIOTT Theorems 6.15 and 6.18 below generalize [29, Proposition 4.1.4] and are ana- logues of Proposition 6.13 for precoherent near prequantales and precoherent pre- quantales, respectively. First, we show that the ideal completion of a multiplica- tive semilattice (resp., prequantic semilattice) is a precoherent near prequantale (resp., precoherent prequantale) under the operation of ↓-multiplication, defined by (I, J) 7−→↓(IJ). Lemma 6.14. One has the following. (1) Let M be a multiplicative semilattice. The operation ↓: X 7−→↓ X is a finitary nucleus on the precoherent near prequantale 2M−{∅}, one has Idl(M ) =↓(2M−{∅}), and Idl(M ) is a precoherent near prequantale under ↓-multiplication. (2) Let M be a prequantic semilattice. The operation ↓: X 7−→↓X is a finitary nucleus on the precoherent prequantale 2M , one has Idl(M ) =↓(2M ), and Idl(M ) is a precoherent prequantale under ↓-multiplication. Proof. We prove statement (1). The proof of (2) is similar. Let X, Y ∈ 2M−{∅}, and let z ∈↓X and w ∈↓Y . Then z ≤W S and w ≤W T for some finite nonempty sets S ⊆ X and T ⊆ Y . Therefore zw ≤ W SW T = W(ST ), where ST is a finite nonempty subset of XY , so zw ∈↓(XY ). Thus we have ↓X ↓Y ⊆↓(XY ), so ↓ is a finitary nucleus on 2M−{∅} by Lemma 6.12. By Theorem 6.8, then, Idl(M ) is a precoherent near prequantale. (cid:3) The morphisms in the category of precoherent near prequantales (resp., category of precoherent prequantales) are morphisms f : Q −→ Q′ of near prequantales (resp., prequantales), with Q and Q′ precoherent, such that f (K(Q)) ⊆ K(Q′). Theorem 6.15. Let Q be a precoherent near prequantale and let M be a multi- plicative semilattice. magmas. magmas. (1) Idl(M ) is a precoherent near prequantale under ↓-multiplication. (2) K(Q) is a multiplicative semilattice. (3) The map M −→ K(Idl(M )) acting by x 7−→↓x is an isomorphism of ordered (4) The map Idl(K(Q)) −→ Q acting by I 7−→W I is an isomorphism of ordered (5) If f : Q −→ Q′ is a morphism of precoherent near prequantales, then the map K(f ) : K(Q) −→ K(Q′) given by K(f )(x) = f (x) for all x ∈ K(Q) is a morphism of multiplicative semilattices. (6) If g : M −→ M ′ is a morphism of multiplicative semilattices, then the map Idl(g) : Idl(M ) −→ Idl(M ′) given by Idl(g)(I) =↓(g(I)) for all I ∈ Idl(M ) is a morphism of precoherent near prequantales. (7) The associations K and Idl are functorial and provide an equivalence of categories between the category of precoherent near prequantales and the category of multiplicative semilattices. Proof. Statement (1) follows from Lemma 6.14, and (2) is clear. To prove (3), note that the map M −→ K(Idl(M )) ⊆ Idl(M ) is the composition M −→ 2M−{∅} −→ ↓(2M−{∅}) = Idl(M ) of magma homomorphisms and is therefore a magma homo- morphism. Thus it is an isomorphism of ordered magmas, by Proposition 6.13(3). The map in statement (4) is a magma homomorphism by Proposition 2.4 and there- fore is an isomorphism of ordered magmas by Proposition 6.13(4). Statement (5) PREQUANTALES 27 is clear. To prove (6), first note that g(↓X) ⊆ ↓(g(X)) for any nonempty subset X of M . The map Idl(g) is then easily shown to be a morphism of near prequantales, and since Idl(g)(↓ x) =↓ (g(x)) for all x ∈ Q, it follows from (3) that Idl(g) is a morphism of precoherent near prequantales. Finally, (7) follows from (1) through (6). (cid:3) Corollary 6.16. The following are equivalent for any ordered magma M . (1) M is a precoherent near prequantale. (2) M is isomorphic to the ideal completion Idl(N ) under ↓-multiplication of (3) K(M ) is a sub multiplicative semilattice of M and M is isomorphic to the some multiplicative semilattice N . (4) M is isomorphic to (2N−{∅})⋆ for some multiplicative semilattice N and ordered magma Idl(K(M )) under ↓-multiplication. some finitary nucleus ⋆ on 2N−{∅}. Corollary 6.17. An ordered magma M is a multiplicative semilattice if and only if M is isomorphic to K(Q) for some precoherent near prequantale Q. Similar proofs of the above results yield the following. Theorem 6.18. One has the following. (1) If f : Q −→ Q′ is a morphism of precoherent prequantales, then the map K(f ) : K(Q) −→ K(Q′) given by K(f )(x) = f (x) for all x ∈ K(Q) is a morphism of prequantic semilattices. (2) If g : M −→ M ′ is a morphism of prequantic semilattices, then the map Idl(g) : Idl(M ) −→ Idl(M ′) given by Idl(g)(I) =↓(g(I)) for all I ∈ Idl(M ) is a morphism of precoherent prequantales. (3) The associations K and Idl are functorial and provide an equivalence of categories between the category of precoherent prequantales and the category of prequantic semilattices. Corollary 6.19. The following are equivalent for any ordered magma M . (1) M is a precoherent prequantale. (2) M is isomorphic to the ideal completion Idl(N ) under ↓-multiplication of (3) K(M ) is a sub prequantic semilattice of M and M is isomorphic to the some prequantic semilattice N . (4) M is isomorphic to (2N )⋆ for some prequantic semilattice N and some ordered magma Idl(K(M )) under ↓-multiplication. finitary nucleus ⋆ on 2N . Corollary 6.20. An ordered magma M is a prequantic semilattice if and only if M is isomorphic to K(Q) for some precoherent prequantale Q. The results above also specialize to coherent prequantales and unital prequantic semilattices, coherent near prequantales and unital multiplicative semilattices, and to the corresponding settings where all operations in question are associative and/or commutative. 7. Stable nuclei A semistar operation ⋆ on an integral domain D is said to be stable if (I ∩ J)⋆ = If ⋆ is stable then (I :F J)⋆ = (I ⋆ :F J) for all I ⋆ ∩ J ⋆ for all I, J ∈ K(D). 28 JESSE ELLIOTT I, J ∈ K(D) with J finitely generated, where F is the quotient field of D. For any semistar operation ⋆ on D there exists a largest stable semistar operation ⋆ on D that is smaller than ⋆, given by I ⋆ =[{(I :F J) : J ⊆ D and J ⋆ = D⋆} for all I ∈ K(D). If ⋆ is of finite type then so is ⋆, whence ⋆w = ⋆f is the largest stable semistar operation of finite type that is smaller than ⋆. The w operation w = vw = t is the largest stable semistar operation on D of finite type such that Dw = D. In this section we generalize these results to the context of semimultiplicative lattices. All of the results of this section, excluding only Proposition 7.1, apply to any coherent residuated semimultiplicative lattice that is also a meet semilattice. In particular, they apply to star, semistar, and semiprime operations and ideal and module systems. Unfortunately, we do not know if the results of this section generalize to nonassociative, noncommutative, or nonunital semiprequantales. We will say that a nucleus ⋆ on a near residuated ordered magma M is stable if (V X)⋆ =V(X ⋆) for any finite subset X of M such that V X exists and (x/t)⋆ = x⋆/t (resp., (t\x)⋆ = t\x⋆) for all x, t ∈ M with t compact such that x/t (resp., t\x) exists. In general the first condition above does not necessarily imply the second. (The first condition is automatic if M is a locale.) However, by the following proposition, the implication holds if every compact element of M is the supremum of a finite subset of Inv(M ), which in turn holds for any associative unital semi-U- lattice M , such as the near U-lattice M = K(D) and the semi-U-lattice F (D). Proposition 7.1. Let M be an associative unital semi-U-lattice, or more generally a near residuated ordered magma such that every compact element of M is the supremum of a finite subset of Inv(M ). A nucleus ⋆ on M is stable if and only if (V X)⋆ =V(X ⋆) for any finite subset X of M such that V X exists. Moreover, if M is also a meet semilattice, then x/t and t\x exist in M for all x, t ∈ M with t compact. Proof. Suppose that ⋆ distributes over finite meets. Let x, t ∈ M with t compact, and suppose that x/t exists. We may write t = u1 ∨ u2 ∨ ··· ∨ un with each ui ∈ Inv(M ). Since x/u exists and equals xu−1 and (x/u)⋆ = (xu−1)⋆ = x⋆u−1 = x⋆/u for all u ∈ Inv(M ), it straightforward to check that x/u1 ∧ x/u2 ∧ ··· ∧ x/un exists and equals x/t. Therefore we have (x/t)⋆ = (x/u1)⋆ ∧ ··· ∧ (x/un)⋆ = x⋆/u1 ∧ ··· ∧ x⋆/un = x⋆/t. The proof for t\x is similar, so ⋆ is stable. Also, if M is a meet semilattice and x, t ∈ M with t compact, then, writing t = u1∨ u2∨···∨ un with each ui ∈ Inv(M ), one easily verifies that x/t = x/u1∧x/u2∧···∧x/un exists, and likewise for t\x. (cid:3) If M is an ordered unital magma and ⋆ ∈ N(M ), then, borrowing terminol- ogy from the theory of semistar operations, we will say that z ∈ M is ⋆-Glaz- Vasconcelos, or ⋆-GV, if z ≤ 1 and z⋆ = 1⋆. We let ⋆-GV(M ) denote the set of all ⋆-GV elements of M , which is a submagma of M . If X ⊆ ⋆-GV(M ) is nonempty, then W X ∈ ⋆-GV(M ) if W X exists, and V X ∈ ⋆-GV(M ) if V X exists and X is finite. If ⋆ is a nucleus on a semimultiplicative lattice Q, then we let x⋆ =_{x/z : z ∈ ⋆-GV(Q)} PREQUANTALES 29 for all x ∈ Q, which is well-defined since z is residuated for all z ∈ ⋆-GV(Q). (Indeed, zx ≤ x, and zy ≤ x implies y ≤ 1∗y = z∗y ≤ x∗, so {z ∈ M : zy ≤ x} is nonempty and bounded.) Lemma 7.2. Let ⋆ be a nucleus on a semimultiplicative lattice Q. (1) ⋆ is a preclosure on Q that is smaller than ⋆, and one has xy⋆ ≤ (xy)⋆ and (2) For any x, t ∈ Q with t compact one has t ≤ x⋆ if and only if tz ≤ x for (3) If Q is algebraic then (V X)⋆ = V(X ⋆) for any finite subset X of Q such x⋆y ≤ (xy)⋆ for all x, y ∈ Q. some z ∈ ⋆-GV(Q). that V X exists. (4) If Q is precoherent then ⋆ is a stable nucleus on Q. Proof. (1) One has x⋆ ≤ W{x⋆/z : z ∈ ⋆-GV(Q)} = W{x⋆/z⋆ : z ∈ ⋆-GV(Q)} = x⋆ for all x ∈ Q, whence ⋆ is smaller than ⋆, and the rest of statement (1) is equally trivial to verify. (2) Let x, t ∈ Q with t compact. If tz ≤ x for some z ∈ ⋆-GV(Q), then t ≤ x/z ≤ x⋆. Conversely, if t ≤ x⋆, then we have t ≤ x/z1 ∨ x/z2 ∨ ··· ∨ x/zn ≤ x/(z1z2 ··· zn) for some z1, z2, . . . , zn ∈ ⋆-GV(Q), whence tz ≤ x, where z = z1z2 ··· zn ∈ ⋆-GV(Q). (3) Let X = {x1, x2, . . . , xn} be a finite subset of Q such that a =V X exists. We must show that a⋆ = V(X ⋆). Clearly a⋆ ≤ x⋆ for all x ∈ X. Let b be any lower bound of X ⋆. If t is any compact element of Q such that t ≤ b, then, by statement (2), for each i there exists zi ∈ ⋆-GV(Q) such that tzi ≤ xi, whence tz ≤ V X = a, where z = z1z2 ··· zn ∈ ⋆-GV(Q). Thus t ≤ a⋆. Taking the supremum over all such t we see that b = W{t ∈ K(Q) : t ≤ b} ≤ a⋆. Therefore V(X ⋆) exists and equals a⋆. (4) We first show that ⋆ is idempotent and therefore a nucleus on Q. Let x ∈ Q. Let t be any compact element such that t ≤ (x⋆)⋆. Then tz ≤ x⋆ for some z ∈ ⋆-GV(Q). If u is a compact element of Q such that u ≤ z, then tu ≤ x⋆ and tu is compact, whence tuzu ≤ x for some zu ∈ ⋆-GV(Q). Let z′ =W{uzu : u ∈ K(Q) and u ≤ z}. Then z′ ≤ 1 and (z′)⋆ = (W{uz⋆ u : u ∈ K(Q) and u ≤ z})⋆ = z⋆ = 1⋆, whence z′ ∈ ⋆-GV(Q). Moreover, one has tz′ = W{tuzu : u ∈ K(Q) and u ≤ z} ≤ x. It follows that t ≤ x⋆. Taking the supremum over all t, we see that (x⋆)⋆ ≤ x⋆. Thus ⋆ is a nucleus on Q. To show that ⋆ is stable, let x, t be elements of Q with t compact such that x/t exists. Clearly x⋆/t also exists and (x/t)⋆ ≤ x⋆/t. Let u ∈ K(Q) with u ≤ x⋆/t. Then tu ≤ x⋆ and tu is compact, whence tuz ≤ x for some z ∈ ⋆-GV(Q). Therefore uz ≤ x/t, whence u ≤ (x/t)⋆. Taking the supremum over all u we see that x⋆/t ≤ (x/t)⋆, whence equality holds. Combining this with statement (3), we see that ⋆ is stable. Theorem 7.3. Let Q be a precoherent semimultiplicative lattice such that every compact element of Q is residuated and x ∧ 1 exists for all x ∈ Q, and let ⋆ be a nucleus on Q. For any x ∈ Q let x⋆ = W{x/z : z ∈ ⋆-GV(Q)}. Then ⋆ is the largest stable nucleus on Q that is smaller than ⋆. Moreover, the following conditions on ⋆ are equivalent. (cid:3) 30 JESSE ELLIOTT (1) ⋆ is stable. (2) (x ∧ 1)⋆ = x⋆ ∧ 1⋆ and (x/t)⋆ = x⋆/t for all x, t ∈ Q with t compact. (3) (x/t ∧ 1)⋆ = x⋆/t ∧ 1⋆ for all x, t ∈ Q with t compact. (4) ⋆ = ⋆. Proof. We first show that the four statements of the proposition are equivalent. Clearly we have (1) ⇒ (2) ⇒ (3), and by Lemma 7.2(4) we have (4) ⇒ (1). To show that (3) implies (4), we suppose that (3) holds, and then we need only show that x⋆ ≤ x⋆ for any x ∈ Q. Let t be any compact element of Q with t ≤ x⋆. By the hypothesis on Q the element z = x/t∧ 1 exists in Q. Condition (3) then implies that z⋆ = x⋆/t ∧ 1⋆. Note then that, since 1⋆t ≤ 1⋆x⋆ = x⋆ one has x⋆/t ≥ 1⋆, whence z⋆ = x⋆/t ∧ 1⋆ = 1⋆. Since also z ≤ 1 we have z ∈ ⋆-GV(Q). Therefore, since zt ≤ (x/t)t ≤ x, we have t ≤ x⋆. Since this holds for all compact t ≤ x⋆, we have x⋆ ≤ x⋆, as desired. It remains only to show that ⋆ is the largest stable nucleus on Q that is smaller than ⋆. But by Lemma 7.2 the nucleus ⋆ is itself a stable nucleus that is smaller than ⋆, and if ⋆′ is any other such nucleus, then one has ⋆′ = ⋆′ ≤ ⋆, whence ⋆ is larger than ⋆′. (cid:3) Any precoherent multiplicative lattice satisfies the hypotheses of Theorem 7.3 above. By Propositions 6.10 and 7.1 and Corollary 3.10, so does any commutative associative unital semi-U-lattice Q with 1 compact that is also a meet semilattice, such as the near U-lattice K(D) and the semi-U-lattice F (D) for any integral domain D. Let ⋆w = ⋆f for any nucleus ⋆ on a precoherent semimultiplicative lattice Q. Proposition 7.4. Let ⋆ be a nucleus on a coherent semimultiplicative lattice Q. (1) One has x⋆w = W{x/z : z ∈ ⋆-GV(Q) ∩ K(Q)} for all x ∈ Q, and ⋆w is (2) If every compact element of Q is residuated and x ∧ 1 exists for all x ∈ Q, then ⋆w is the largest stable finitary nucleus on Q that is smaller than ⋆. finitary. In particular, if ⋆ is finitary, then so is ⋆ = ⋆w. Proof. Statement (2) follows from statement (1) and Theorems 7.3 and 6.8. To therefore equality holds since the reverse inequality is obvious. prove (1), let x ∈ Q. One has x⋆w = W{x/y : y ≤ 1 and y⋆f = 1⋆f}. Let t be any compact element of Q with t ≤ x⋆w . Then t ≤ x/y for some y ∈ Q with y ≤ 1 and y⋆f = 1⋆f . Since 1 is compact, the condition y⋆f = 1⋆f implies z⋆ = 1⋆ for some compact z ≤ y. Note then that t ≤ x/z since tz ≤ ty ≤ x. Therefore, since z ∈ ⋆-GV(Q) ∩ K(Q), one has t ≤W{x/z : z ∈ ⋆-GV(Q) ∩ K(Q)}. Since this holds for all compact t ≤ x⋆w , it follows that x⋆w ≤W{x/z : z ∈ ⋆-GV(Q) ∩ K(Q)}, and Suppose now that ⋆ = ⋆f is finitary. To show that ⋆ is also finitary, we let x ∈ Q and verify that x⋆ =W{y⋆ : y ∈ K(Q) and y ≤ x}. Let t be any compact element of Q with t ≤ x⋆. Since ⋆w = ⋆, one has x⋆ =W{x/z : z ∈ ⋆-GV(Q)∩ K(Q)}. Since t ≤ x⋆ is compact and the set {z/x : z ∈ ⋆-GV(Q)∩K(Q)} is directed, it follows that t ≤ x/z for some z ∈ ⋆-GV(Q) ∩ K(Q). Therefore tz ≤ x =W{y ∈ K(Q) : y ≤ x}, whence tz ≤ y for some compact y ≤ x since tz is compact. Thus we have t ≤ y/z, where z ∈ ⋆-GV(Q), whence t ≤ y⋆. It follows that x⋆ ≤W{y⋆ : y ∈ K(Q) and y ≤ x}, and the desired equality follows. (cid:3) The following result follows from Propositions 5.2 and 7.4 and Theorems 6.8 and 7.3. PREQUANTALES 31 Proposition 7.5. Let Q be a precoherent near multiplicative lattice in which every compact element of Q is residuated. Suppose moreover that x ∧ 1 exists for all x ∈ Q. Let ⋆ be a nucleus on Q. (1) The nucleus v(a) = v(a) exists for any a ∈ Q and is the largest stable (2) If 1 ∈ Q is compact, then the nucleus w(a) = t(a) exists for any a ∈ Q and is the largest stable finitary nucleus on Q such that aw(a) = a, and nucleus on Q such that av(a) = a, and ⋆ =V{v(a) : a ∈ Q⋆}. ⋆w =V{w(a) : a ∈ Q⋆w}. any set Γ of stable nuclei on Q. Lemma 7.6. Let Q be a near prequantale. Then V Γ is a stable nucleus on Q for Proof. Let X be a finite subset of Q that is bounded below. Then one has (V X)V Γ = V(cid:8)(V X)⋆ : ⋆ ∈ Γ(cid:9) =V{V(X ⋆) : ⋆ ∈ Γ} = V{x⋆ : x ∈ X, ⋆ ∈ Γ} =V(cid:0)X V Γ(cid:1). Moreover, for any x, t ∈ Q with t compact, if x/t exists, then one has (x/t)∧Γ = V{(x/t)⋆ : ⋆ ∈ Γ} =V{x⋆/t : ⋆ ∈ Γ} = xV Γ/t, and a similar proof holds for t\x if t\x exists. Thus V Γ is stable. Corollary 7.7. Let Q be a precoherent near multiplicative lattice in which every compact element of Q is residuated. Suppose moreover that x ∧ 1 exists for all x ∈ Q. A nucleus ⋆ on Q is stable if and only if ⋆ = V{v(a) : a ∈ X} for some subset X of Q. (cid:3) 8. Star and semistar operations Throughout this section let D denote an integral domain with quotient field F . A Kaplansky fractional ideal of D is a D-submodule of F . Let K(D) denote the ordered monoid of all nonzero Kaplansky fractional ideals of D. Recall that a semistar operation on D is a closure operation ⋆ on the poset K(D) such that (aI)⋆ = aI ⋆ for all nonzero a ∈ F and all I ∈ K(D). The submonoid Prin(D) of K(D) of all nonzero principal D-submodules of F is a sup-spanning subset of K(D) contained in Inv(K(D)). Therefore, by Propositions 3.4 and 3.6, a semistar operation on D is equivalently a nucleus on the ordered monoid K(D). Together with Propositions 3.2 and 3.3, this yields a proof of Theorem 1.2 of the introduction. A semistar operation on D may be defined alternatively as a closure operation on the poset ModD(F ) such that (aI)⋆ = aI ⋆ for all a ∈ F and all I ∈ ModD(F ), or equivalently a nucleus ⋆ on the K-lattice ModD(F ) such that (0)⋆ = (0). This alternative definition is advantageous because K(D) is typically neither complete nor residuated, while every K-lattice is complete and residuated, and the definition thereby eliminates the need for many results, such as [26, Proposition 5, Theorem 20, Lemma 40], to make exception for the zero ideal. The definition also allows for the possibility of relaxing the condition (0)⋆ = (0) by considering all nuclei on ModD(F ). Since K(D) inherits its "near" properties (such as being a near K-lattice and near U-lattice) from corresponding "complete" properties of ModD(F ), all of the results of this paper generalizing results on semistar operations apply as well to all nuclei on ModD(F ). Recall that a D-submodule I of F is said to be a fractional ideal of D if aI ⊆ D for some nonzero element a of F . Although the ordered monoid F (D) of all nonzero fractional ideals of D is not necessarily a near prequantale, it is a semiprequantale (and in fact a semi-U-lattice). Moreover, F (D) is coherent and residuated. A star 32 JESSE ELLIOTT operation ∗ on D is a closure operation on the poset F (D) such that D∗ = D and (aI)∗ = aI ∗ for all nonzero a ∈ F and all I ∈ F (D). The same argument as in the proof of Theorem 1.2 yields the following. Theorem 8.1. Let D be an integral domain with quotient field F . The following conditions are equivalent for any self-map ∗ of F (D) such that D∗ = D. (1) ∗ is a star operation on D. (2) ∗ is a nucleus on the ordered monoid F (D). (3) ∗ is a closure operation on the poset F (D) and ∗-multiplication on F (D) (4) ∗ is a closure operation on the poset F (D) and (I ∗J ∗)∗ = (IJ)∗ for all I, J ∈ F (D) (or equivalently such that the map ∗ : F (D) −→ F (D)∗ is a magma homomorphism). is associative. (5) HJ ⊆ I ∗ if and only if HJ ∗ ⊆ I ∗ for all H, I, J ∈ F (D). (6) (I ∗ : J) = (I ∗ : J ∗) for all I, J ∈ F (D). (7) ∗ = ⋆F (D) for some semistar operation ⋆ on D such that D⋆ = D. The following proposition, which follows from the results of Sections 4 and 6, lists properties of the posets SStar(D), SStarf (D), Star(D), and Starf (D) of all semistar operations, finite type semistar operations, star operations, and finite type star operations, respectively, on D. Proposition 8.2. Let D be an integral domain. One has SStar(D) = N(K(D)), SStarf (D) = Nf (K(D)), Star(D) = N(F (D))≤v, and Starf (D) = Nf (F (D))≤t, and all four of these posets are complete. Let Γ ⊆ SStar(D) and ∆ ⊆ Star(D), and let hΓi (resp., h∆i) denote the submonoid generated by Γ (resp., ∆) of the monoid of all self-maps of K(D) (resp., F (D)). (1) One has I V Γ =\{I ⋆ : ⋆ ∈ Γ}, I W Γ =\{J ∈ K(D) : J ⊇ I and ∀⋆ ∈ Γ (J ⋆ = J)}, for all I ∈ K(D). (2) If Γ ⊆ SStarf (D), then I W Γ =[{I γ : γ ∈ hΓi} for all I ∈ K(D), one has WSStarf (D) Γ = W Γ and VSStarf (D) Γ = (V Γ)f , and if Γ is finite then VSStarf (D) Γ =V Γ. I V ∆ =\{I ∗ : ∗ ∈ ∆}, I W ∆ =\{J ∈ F (D) : J ⊇ I and ∀∗ ∈ ∆ (J ∗ = J)}, (3) One has for all I ∈ F (D). (4) If ∆ ⊆ Starf (D), then I W ∆ =[{I γ : γ ∈ h∆i} for all I ∈ F (D), one has WStarf (D) ∆ = W ∆ and VStarf (D) ∆ = (V ∆)f , and if ∆ is finite then VStarf (D) ∆ =V ∆. PREQUANTALES 33 Next, let J ∈ K(D). For all I ∈ K(D) we let I v(J) = (J :F (J :F I)). The operation v(J) on K(D) is called divisorial closure on D with respect to J [28, Example 1.8(a)]. More generally, for all S ⊆ K(D) we define v(S) =V{v(J) : J ∈ S}, which we call divisorial closure on D with respect to S. We set t(S) = v(S)f , v(S) = v(S), and w(S) = v(S)w = t(S), where for any semistar operation ⋆ one defines ⋆ by I ⋆ =S{(I :F J) : J ⊆ D, J ⋆ = D⋆} for all I ∈ K(D) and ⋆w = ⋆f . A semistar operation ⋆ on D is said to be stable if (I∩J)⋆ = I ⋆∩J ⋆ for all I, J ∈ K(D). The results of Sections 6, 5, and 7 yield the following. Proposition 8.3. Let D be an integral domain and J ∈ K(D). (1) v(J) is the largest semistar operation ⋆ on D such that J ⋆ = J. (2) t(J) is the largest finite type semistar operation ⋆ on D such that J ⋆ = J. (3) v(J) is the largest stable semistar operation ⋆ on D such that J ⋆ = J. (4) w(J) is the largest stable finite type semistar operation ⋆ on D such that J ⋆ = J. More generally, we have the following for any subset S of K(D). (5) v(S) is the largest semistar operation ⋆ on D such that S ⊆ K(D)⋆. (6) t(S) is the largest finite type semistar operation ⋆ on D such that S ⊆ (7) v(S) is the largest stable semistar operation ⋆ on D such that S ⊆ K(D)⋆. (8) w(S) is the largest stable finite type semistar operation ⋆ on D such that K(D)⋆. S ⊆ K(D)⋆. For any semistar operation ⋆ on D, one has the following. (9) ⋆ =V{v(J) : J ∈ K(D)⋆}. (10) ⋆f =V{t(J) : J ∈ K(D)⋆f }. (11) ⋆ =V{v(J) : J ∈ K(D)⋆}. (12) ⋆w =V{w(J) : J ∈ K(D)⋆w}. Moreover, ⋆ is stable if and only if ⋆ = V{v(J) : J ∈ S} for some S ⊆ K(D), in which case ⋆ = v(S). We are unable to characterize all subsets S of K(D) such that V{t(J) : J ∈ S} is of finite type (although it certainly holds if S is finite, by Proposition 6.3(3)). We leave this as an open problem. Propositions 5.8 and 5.16 yield the following. Proposition 8.4. Let D be an integral domain with quotient field F . For all I, J ∈ K(D) one has I v(J) = [{I ′ ∈ K(D) : (J :F I) ⊆ (J :F I ′)} = \{U J : U ∈ K(D) is invertible and I ⊆ U J}. Theorem 7.3 yields the following characterizations of stable semistar operations. Proposition 8.5. Let D be an integral domain with quotient field F , and let ⋆ be a semistar operation on D. The following conditions are equivalent. (1) ⋆ is stable. (2) (I ∩ D)⋆ = I ⋆ ∩ D⋆ and (I :F J)⋆ = (I ⋆ :F J) for all I, J ∈ K(D) with J (3) (I :D J)⋆ = (I ⋆ :D⋆ J) for all I, J ∈ K(D) with J finitely generated. finitely generated. 34 JESSE ELLIOTT Since the ordered monoid F (D) of nonzero fractional ideals of a domain D is a precoherent residuated semimultiplicative lattice, a meet semilattice, and an associative unital semi-U-lattice, and since the ordered monoid I(R) of ideals of a ring R is a quantale, results analogous to Propositions 8.3, 8.4, and 8.5 hold for star operations and for semiprime operations. Proposition 8.3 shows that all semistar operations can be obtained as infima of generalized divisorial closure semistar operations. This has the potential for yielding new results on the number of semistar operations on an integral domain, which is a commonly studied problem in the theory of semistar operations. (Similar comments hold for star operations and for semiprime operations.) For example, combining Corollary 5.14 with [26, Theorem 48], we obtain the following. Proposition 8.6. The following are equivalent for any integral domain D with quotient field F 6= D. (1) D is a DVR. (2) The only semistar operations on D are d and e. (3) SStar(D) = 2. (4) The near multiplicative lattice K(D) is simple. (5) v(J) = d for all J ∈ K(D) with J 6= F . (6) (J :F (J :F I)) = I for all I, J ∈ K(D) with I, J 6= F . Using the divisorial closure operations, one can generalize the equivalence of (1) through (3) in the theorem above as follows. Theorem 8.7 ([5]). Let D be a Dedekind domain with quotient field F and Max(D) the set of all maximal ideals of D. There is a poset embedding of the lattice C(2Max(D)) of all closure operations on 2Max(D) into the lattice SStar(D) of all semistar operations on D. Explicitly, the map C(2Max(D)) −→ SStar(D) acting by ∗ 7−→ v(S), where S = {I ∈ K(D) : {p : IDp = F} is ∗ -closed and {p : Dp ( IDp ( F} is finite} , is a poset embedding. Moreover, the given embedding has an order-preserving left inverse acting by ⋆ 7−→ ∗, where ∗ is the largest closure operation on 2Max(D) such that {p ∈ Max(D) : IDp = F} is ∗-closed for all I ∈ K(D)⋆. Finally, if Max(D) is finite, then the given embedding C(2Max(D)) −→ SStar(D) is an isomorphism. Proof. The theorem follows from [5, Corollary 3.5 and Lemma 4.1]. (cid:3) Note that, if D is a Dedekind domain, then {p : Dp ( IDp ( F} is finite if and only if I is a fractional ideal of some overring of D, if and only if I is a fractional ideal of (I :F I), if and only if ((I :F I) :F I) = (I :F I 2) is nonzero. If D and D′ are Dedekind domains, then by [5, Proposition 3.1(4)] the complete lattices K(D) and K(D′) are isomorphic if Max(D) and Max(D′) have the same cardinality. Combining this with Theorem 8.7 above, Corollary 1.3, [4, Table 1], and results in [1] on the number of closure operations on 2S for finite sets S, we obtain the following. Corollary 8.8 ([5, Theorem 1.3]). Let D and D′ be Dedekind domains. (1) The lattices SStar(D) and SStar(D′) are isomorphic if Max(D) and Max(D′) have the same cardinality. (2) If Max(D) is infinite, then SStar(D) is equal to 22 Max(D) . PREQUANTALES 35 Table 3. Number of semistar operations on a Dedekind domain D Max(D) 1 2 3 4 5 6 7 SStar(D) 2 7 61 2 480 1 385 552 75 973 751 474 14 087 648 235 707 352 472 (3) If Max(D) = n is finite, then 2( n (4) SStar(D) is given for n ≤ 7 as in Table 3. We speculate that it is possible to carry out similar applications of the divisorial [n/2]) ≤ SStar(D) ≤ 22n . closure operations to star operations and semiprime operations. Example 8.9. Let D be a Dedekind domain with quotient field F . Let ID = {p ∈ Max(D) : IDp = F} for any I ∈ K(D). From [5, Proposition 3.1(3)] one can show that I v(J) =(cid:26) K Tp /∈JD for all I, J ∈ K(D). In particular, one has IDp if ID * JD if ID ⊆ JD I v =(cid:26) K if ID 6= ∅ if ID = ∅, I and v(J) = v if and only if JD = ∅. Suppose that D has exactly two maximal ideals p1 and p2. Define I ⋆i = IDpi for i = 1, 2. One has v(J) = ⋆1 if and only if JD = {p2} and v(J) = ⋆2 if and only if JD = {p1}. Also, one has v(J) = e if and only if JD = {p1, p2}. Since ⋆1∧ ⋆2 = d, it follows that SStar(D) = {e, v, ⋆1, ⋆2, ⋆1∧ v, ⋆2 ∧ v, d} ∼= C(2{1,2}) ∼= 2{1,2,3} − {{2}}, with Hasse diagram given below. ⋆1 ⋆1 ∧ v ①①①①①①①①① ②②②②②②②②② ❊❊❊❊❊❊❊❊❊ e v d ❋❋❋❋❋❋❋❋❋ ❊❊❊❊❊❊❊❊❊ ②②②②②②②②② ⋆2 ⋆2 ∧ v For any star operation ∗ on D, there is a unique largest semistar operation l(∗) on D such that I l(∗) = I ∗ for all I ∈ F (D). Indeed, one may extend ∗ to K(D) by defining I l(∗) = F for all I ∈ K(D)−F (D). (See Proposition 4.7(3).) Using Proposition 4.7(2), we may construct the unique smallest semistar operation s(∗) on D such that I s(∗) = I ∗ for all I ∈ F (D), as follows. Proposition 8.10. Let D be an integral domain. 36 JESSE ELLIOTT (1) For any star operation ∗ on D, there exists a unique smallest semistar operation s(∗) on D such that s(∗)F (D) = ∗, and for all I ∈ K(D) one has I s(∗) =\{K ∈ K(D) : K ⊇ I and ∀J ∈ F (D) (J ⊆ K ⇒ J ∗ ⊆ K)}. (2) More generally, for any nucleus γ on a submonoid M of K(D) containing Prin(D), there exists a unique smallest semistar operation ⋆ on D such that ⋆M = γ, and for all I ∈ K(D) one has I ⋆ =\{K ∈ K(D) : K ⊇ I and ∀J ∈ M (J ⊆ K ⇒ J γ ⊆ K)}. Although l(∗) has been studied predominantly in the literature, the semistar operation s(∗) is a useful alternative. Note in particular that s(d) = d on any domain D, and by Proposition 8.11 below ∗ is of finite type if and only if s(∗) is of finite type, where a star or semistar operation ⋆ is said to be of finite type if ⋆ = ⋆f . By contrast, although the star operation d is of finite type, the semistar operation l(d) on D is of finite type if and only if l(d) = d on D, if and only if D is conducive, that is, every overring of D other than F is a fractional ideal of D. Proposition 8.11. Let D be an integral domain. For any semistar operation ⋆ and star operation ∗ on D, one has ⋆F (D) = ∗ if and only if s(∗) ≤ ⋆ ≤ l(∗), in which case s(∗f ) = s(∗)f = ⋆f = l(∗)f . In particular, ∗ is a finite type star operation if and only if s(∗) is a finite type semistar operation, in which case s(∗) is the unique finite type semistar operation ⋆ on D such that ⋆F (D) = ∗. Proof. The first equivalence is clear from the construction of s(∗) and l(∗). Suppose that ⋆F (D) = ∗. For any I ∈ K(D), one has I ⋆f = SJ⊆I f.g. J ∗. It follows that s(∗)f = ⋆f = l(∗)f . It also follows that if ∗ is of finite type, then s(∗)fF (D) = ∗, whence s(∗) ≤ s(∗)f , whence s(∗) is of finite type. In particular, one has s(∗f ) = s(∗f )f ≤ s(∗)f . To prove the reverse inequality, let I ∈ K(D), and note as above that I s(∗)f =SJ⊆I f.g. J ∗. It follows that if K ⊇ I, and if J ⊆ K implies J ∗f ⊆ K for all J ∈ K(D), then I s(∗)f ⊆ K. Therefore I s(∗)f ⊆ I s(∗f ). Thus we have s(∗)f ≤ s(∗f ). The rest of the proposition follows. (cid:3) The star operation v = v(D) is the largest star operation on D, and t = vf is the largest finite type star operation on D. One has l(v) = v and l(t) = t, but one may have s(v) < l(v) and s(t) < l(t). Indeed, let D be any divisorial domain, that is, any domain for which I v = I for all I ∈ F (D). Since v = t = d on D, it follows that s(v) = s(t) = d on D as well. However, if D is non-conducive, say, if D = Z, then d < l(t) ≤ l(v), whence s(v) < l(v) and s(t) < l(t). Finally, we comment on the requirement D∗ = D for a star operation ∗. Let us say that a quasistar operation is a self-map ∗ of F (D) that satisfies all of the star operation axioms except possibly D∗ = D. Equivalently, a quasistar operation is a nucleus on the semimultiplicative lattice F (D). The poset QStar(D) = N(F (D)) of all quasistar operations on D is of course bounded complete, and Star(D) = QStar(D)≤v. An interesting question arises as to when QStar(D) is complete, that is, when there is a largest quasistar operation. We show that the answer is in the of D [12] and F is the quotient field of D. affirmative if and only if (D :F eD) 6= (0), where eD is the complete integral closure An element x of F is said to be almost integral over D if there exists a nonzero element c of D such that cxn ∈ D for all positive integers n, or equivalently, if PREQUANTALES 37 (D :F D[x]) 6= (0). The complete integral closure eD of D is the set of all elements of F that are almost integral over D. The set eD is an overring of D containing the integral closure D of D, and if D is Noetherian then eD = D. One says that D is completely integrally closed if eD = D. In general eD may not be completely integrally closed; but in the next section we will show that there is a smallest completely integrally closed overring D♯ of D, which we call the complete complete integral closure of D. Proposition 8.12. Let D be an integral domain with quotient field F . If ∗ is any quasistar operation on D, then D∗ ⊆ eD. Moreover, the following conditions are equivalent. (1) QStar(D) is a complete lattice. (2) There is a largest quasistar operation e on D. (3) There is a largest overring D′ of D such that (D :F D′) 6= (0). (4) (D :F eD) 6= (0). If the above conditions hold, then one has De = D′ = eD = D♯ and e = v(De). Moreover, the above conditions hold if eD is finitely generated as a D-algebra, hence also if D is completely integrally closed or if D is Noetherian and D is module finite over D. Proof. If ∗ is a quasistar operation on D, then (D :F D∗) 6= (0), which implies that (D :F D[x]) 6= (0) for all x ∈ D∗, and therefore D∗ ⊆ eD. Clearly (1) and (2) are equivalent. If e is a largest quasistar operation on D and E is any overring of D such that (D :F E) 6= (0), then the operation ∗E : I 7−→ IE is a quasistar operation on D, and therefore ∗E ≤ e and so E = D∗E ⊆ De. Therefore (2) implies (3). Suppose that (3) holds. Then for any quasistar operation ∗ one must have D′∗ = D′ and therefore ∗ ≤ v(D′), so that e = v(D′) is the largest quasistar operation on D. Furthermore, if x ∈ eD, then (D :F D[x]) 6= (0), so that D[x] ⊆ D′ and therefore x ∈ D′. Conversely, one has D′ ⊆ eD, so that D′ = eD. Therefore (3) implies both D′ = eD and e = v(De). Since (D :F eD) 6= (0), by [12, Corollary 6] the domain eD is completely integrally closed and therefore eD = D♯. Finally, the last statement Suppose, now, that the four conditions hold. We have seen already that De = (2) and (4). Since also (4) implies (3), all four conditions are equivalent. of the proposition is clear. (cid:3) 9. Integral, complete integral, and tight closure Let us recall the definition of the integral closure semistar operation. Let D If I ∈ K(D), then x ∈ F is in- be an integral domain with quotient field F . tegral over I if there exists a positive integer n and elements ai ∈ I i such that xn + a1xn−1 + ··· + an−1x + an = 0. The semistar integral closure of I is the set I of all elements of F that are integral over I and is given by I = T{IV : V is a valuation overring of D}. It follows that semistar integral closure is a semis- tar operation on D. In the literature of semistar operations it is often called the Indeed, if x ∈ I, say, if b-operation. Semistar integral closure is of finite type. xn + a1xn−1 +··· + an−1x + an = 0 for elements ai =Pj bij1bij2 ··· biji ∈ I i, where each bijk lies in I, then x ∈ J, where J ⊆ I is the ideal generated by the bijk. For any ideal I of D, the ideal I ∩ D is known as the integral closure of I. 38 JESSE ELLIOTT I ♯ =\{J ∈ K(D) : J ⊇ I and J is completely integrally closed}, The operation of complete integral closure can also be used to define a semistar operation. If I ∈ K(D), then x ∈ F is almost integral over I if there exists a nonzero element c of F such that cxn ∈ I n for all positive integers n. We define eI to be the set of all elements of F that are almost integral over I, and we say that I is completely integrally closed if eI = I. Although one may have eD ( eeD, the map I 7−→ eI is at least a preclosure on K(D). We define which we call the semistar complete integral closure of I. Since aeJ = faJ for all nonzero a ∈ F , and therefore IeJ ⊆ fIJ, for all I, J ∈ K(D), by Lemma 4.4 the Proposition 9.1. Let D be an integral domain. The operation ♯ is the unique semistar operation on D such that I ♯ = I for I ∈ K(D) if and only if I is com- pletely integrally closed. Moreover, for all I ∈ K(D) one has I ⊆ eI ⊆ eI ♯ = I ♯, with equalities holding if D is Noetherian. Thus I ♯ is the smallest completely inte- grally closed Kaplansky fractional ideal of D containing I, and ♯ equals the semistar integral closure operation on D if D is Noetherian. operation ♯ is a semistar operation on D. In particular, we have the following. Corollary 9.2. For any integral domain D with quotient field F , one has the following. (1) D♯ is the smallest completely integrally closed overring of D. Sβ<α D♯β for all limit ordinals α. Then D♯ = D♯α for all α ≫ 0. (2) Define D♯0 = D, D♯α+1 = gD♯α for all successor ordinals α + 1, and D♯α = (3) If J ∈ K(D) is completely integrally closed, then the largest overring (J :F J) of D for which J is a Kaplansky fractional ideal is also completely inte- grally closed. Next, let D+ denote the integral closure of D in an algebraic closure of F . For all I ∈ K(D), let I π = ID+ ∩ F , which we call the semistar plus closure of I, and which is easily seen to define a semistar operation π on D. Note that Dπ = D and therefore ID ⊆ I π for all I ∈ K(D). Like semistar integral closure, semistar plus closure is of finite type. Also, for any ideal I of D, the ideal I π ∩ D = ID+ ∩ D is known as the plus closure of I. The operation of tight closure also yields a semistar operation, as follows. Let D be any integral domain (not necessarily Noetherian) of characteristic p > 0 with quotient field F . For any I ∈ K(D) let I T = {x ∈ F : ∃c ∈ F−{0} such that cxq ∈ I [q] for all q = pe, e ≥ 0}, where I [q] for any q = pe denotes the D-submodule of F generated by the image of I under the q-powering Frobenius endomorphism of F . It is clear that T is a preclosure on K(D). Moreover, one has aJ T = (aJ)T for all nonzero a ∈ F , and therefore IJ T ⊆ (IJ)T , for all I, J ∈ K(D). Therefore by Lemma 4.4 the operation τ on K(D) of semistar tight closure, defined by I τ =\{J ∈ K(D) : J ⊇ I and J T = J} for all I ∈ K(D), is a semistar operation on D. Note that by [2, Corollary 4] one has eD ⊆ DT ⊆ eeD, from which it follows that Dτ = D♯. In particular, τ restricts to PREQUANTALES 39 a star operation on D if and only if D♯ = D if and only if D is completely integrally closed. Also, if D is Noetherian then I τ = I T ⊆ I for all I ∈ K(D). Example 9.3. Let k be a finite field of characteristic p > 0, and let D be the subring k[X, XY, XY p, XY p2 , . . .] of k[X, Y ]. Then Dτ = DT = eeD = k[X, Y ], but Y /∈ eD, whence eD ( Dτ . The tight closure I ∗ of an ideal I of D may be defined as the ideal I ∗ = I τ ∩ D. Since τ = T if D is Noetherian, this definition coincides with the well-known definition of tight closure in the Noetherian case. In fact, we may generalize this definition to an arbitrary commutative ring R of prime characteristic p, as follows. For any I ∈ I(R) we let I T = {x ∈ R : ∃c ∈ Ro such that cxq ∈ I [q] for all q = pe ≫ 0}, where Ro is the complement of the union of the minimal primes of R and I [q] is the ideal of R generated by the image of I under the q-powering Frobenius endomorphism of R. We say that I ∈ I(R) is tightly closed if I T = I. It is clear that T is a preclosure on the complete lattice I(R). For any I ∈ I(R) we let I ∗ =\{J ∈ I(R) : J ⊇ I and J is tightly closed}. Since T is a preclosure on I(R) such that IJ T ⊆ (IJ)T for all I, J ∈ I(R), we obtain from Lemma 4.4 the following result. Proposition 9.4. Let R be a commutative ring of prime characteristic p. The operation ∗ is the unique semiprime operation on R such that I ∗ = I for I ∈ I(R) if and only if I is tightly closed. Moreover, for all I ∈ I(R) one has I T ⊆ (I ∗)T = I ∗, with equalities holding if R is Noetherian. In particular, I ∗ for any I ∈ I(R) is the smallest tightly closed ideal of R containing I, and ∗ equals the ordinary tight closure operation if R is Noetherian. Thus ∗ provides a potential definition of tight closure for non-Noetherian com- mutative rings of prime characteristic. The finitary nucleus ∗f on I(R) is another candidate for a definition of tight closure in the non-Noetherian case. Further in- vestigation may determine whether or not either of these generalizations of tight closure is useful. One may seek, for example, a generalization of Tight Closure via Colon-Capturing [17, Theorem 3.1] with respect to some definition of non- Noetherian Cohen-Macaulay rings, such as the notion presented in [16]. 10. Ideal and module systems For any magma M , let M0 denote the magma M ∐ {0}, where 0x = 0 = x0 for all x ∈ M0. A module system on an abelian group G is a closure operation r on the K-lattice 2G0 such that ∅r = {0} and (cX)r = cX r for all c ∈ G0 and all X ∈ 2G0. Theorem 10.1. A module system on an abelian group G is equivalently a nucleus r on the multiplicative lattice 2G0 such that ∅r = {0}. Moreover, the following are equivalent for any self-map r of 2G0 such that ∅r = {0}. (1) r is a module system on G. (2) r is a closure operation on the poset 2G0 and r-multiplication is associative. (3) r is a closure operation on the poset 2G0 and (X rY r)r = (XY )r for all X, Y ∈ 2G0. 40 JESSE ELLIOTT (4) XY ⊆ Z r if and only if XY r ⊆ Z r for all X, Y, Z ⊆ G0. Proof. Let Σ = {{c} : c ∈ G0}. Let r be a module system on G. Since Σ is a sup-spanning subset of 2G0 and Σ ⊆ Tr(2G0), it follows from Corollary 3.5 that r is a nucleus on 2G0. Conversely, let r be a nucleus on 2G0 such that ∅r = {0}. Since Σ−{{0}} ⊆ Inv(2G0), by Proposition 3.4, we have ({c}X)r = {c}X r for all c ∈ G and all X ∈ 2G0. Also, we have ({0}X)r = {0}r = {0} = {0}X r for all X ∈ 2G0. Thus r is a module system on G. (cid:3) An alternative proof of Theorem 1.2 can be given using Theorem 10.1, Proposi- tions 3.2 and 3.3, and the following lemma. Lemma 10.2. Let D be an integral domain, F × the group of nonzero elements of the quotient field F of D, and ⋆ a self-map of K(D). Let ∅r = {0}r = {0} and X r = (DX)⋆ for all subsets X of F containing a nonzero element, where DX denotes the D-submodule of F generated by X. Then ⋆ is a semistar operation on D if and only if r is a module system on F ×. Next, a weak ideal system [14] on a commutative monoid M is a closure operation r on the K-lattice 2M0 such that 0 ∈ ∅r, cM0 ⊆ {c}r, and cX r ⊆ (cX)r for all c ∈ M0 and all X ∈ 2M0 . A weak ideal system on M is said to be an ideal system on M if (cX)r = cX r for all such c and X. We have the following result, whose proof is similar to that of Theorem 10.1. Proposition 10.3. Let M be a commutative monoid. A weak ideal system on M is equivalently a nucleus r on the K-lattice 2M0 such that {0}r = ∅r and {1}r = M0. Moreover, a weak ideal system r on M is an ideal system on M if and only if every singleton in 2M0 is transportable through r. A module system on an abelian group G may be seen equivalently as a nucleus on the K-lattice 2G0−2G. Since 2G0 and 2G0−2G are (coherent) K-lattices and 2G0−2G is a U-lattice, most of the results of this paper, and in particular the results of Sections 3 through 7, apply specifically to module systems. Our results on star and semistar operations generalize to module systems, since the ordered monoids 2G0 − 2G and K(D), where D is any integral domain, share the relevant properties. Finally, as 2M0 is a K-lattice for any commutative monoid M , many of our results apply as well to weak ideal systems. We leave it to the interested reader to carry out the details. References [1] V. B. Alekseev, On the number of families of subsets closed with respect to intersections, Diskretnaya Matematika 1 (2) (1989) 129 -- 136 (in Russian). [2] D. D. Anderson and M. Zafrullah, Pseudo almost integral elements, Comm. Alg. 35 (2007) 1127 -- 1131. [3] G. Birkhoff, Lattice theory, Colloquium Publications, vol. 25, American Mathematical Society, Providence, RI, 1967. [4] P. Colomb, A. Irlande, O. Raynaud, Counting of Moore families for n = 7, in: Formal Concept Analysis, Lecture Notes in Computer Science, Volume 5986, Springer Science+Business Media, Berlin, 2010, pp. 72 -- 87. [5] J. Elliott, Semistar operations on Dedekind domains, Comm. Algebra, Special Issue Dedicated to Marco Fontana, 43 (1) (2015) 236 -- 248. [6] J. Elliott, Prequantales and applications to semistar operations and module systems, arXiv:1101.2462v1 [math.RA], 12 Jan 2011. PREQUANTALES 41 [7] M. Fontana and J. Huckaba, Localizing systems and semistar operations, in: Non Noetherian Commutative Ring Theory, eds. S. Chapman and S. Glaz., Kluwer Academic Publishers, 2000. [8] M. Fontana and M. H. Park, Star operations and pullbacks, J. Alg. 274 (2004) 387 -- 421. [9] N. Galatos, P. Jipsen, T. Kowalski, and H. Ono, Residuated Lattices: An Algebraic Glimpse at Substructural Logics, in: Studies in Logic and the Foundations of Mathematics, vol. 151, Elsevier, New York, 2007. [10] G. Gierz, K. H. Hofmann, K. Keimel, J. D. Lawson, M. Mislove, and D. S. Scott, Continuous Lattices and Domains, Encyclopedia of Mathematics and its Applications, vol. 93, Cambridge University Press, 2003. [11] R. Giles and H. Kummer, A non-commutative generalization of topology, Indiana Univ. Math. J. 21 (1971) 91 -- 102. [12] R. W. Gilmer, Jr. and W. J. Heinzer, On the complete integral closure of an integral domain, J. Austral. Math. Soc. 6 (1966) 351 -- 361. [13] S. Givant and P. Halmos, Introduction to Boolean Algebras, Undergraduate Texts in Mathe- matics, Springer-Verlag, New York, 2009. [14] F. Halter-Koch, Ideal Systems: An Introduction to Multiplicative Ideal Theory, Monographs and Textbooks in Pure and Applied Mathematics 211, Marcel Dekker, New York, 1998. [15] F. Halter-Koch, Localizing systems, module systems, and semistar operations, J. Alg. 238 (2001) 723 -- 761. [16] T. Hamilton and T. Marley, Non-Noetherian Cohen-Macaulay rings, J. Alg. 307 (2007) 343 -- 360. [17] C. Huneke, Tight Closure and its Applications, CBMS Regional Conference Series in Math- ematics, Number 88, Amer. Math. Soc., Providence, RI, 1996. [18] J. R. Isbell, Atomless parts of spaces, Math. Scand. 31 (1972) 5 -- 32. [19] A. Joyal and M. Tierney, An Extension of the Galois Theory of Grothendieck, Memoirs of the American Mathematical Society, vol. 52, no. 309, Providence, RI, 1984. . [20] W. Krull, Axiomatische Begrundung der algemeinen Idealtheorie, Sitzungsberichte der physikalischmedizinischen Societat zu Erlangen 56 (1924) 47- 63. [21] W. Krull, Idealtheorie, Springer-Verlag, Berlin, 1935. [22] D. Kruml and J. Paseka, On simple and semisimple quantales, in: Topology Atlas Invited Contributions, Topology Atlas, Toronto, 2002. [23] D. Kruml, J. W. Pelletier, P. Resende, and J. Rosick´y. On quantales and spectra of C∗- algebras, Appl. Categor. Struct. 11 (2003) 543 -- 560. [24] C. J. Mulvey, &, Rend. Circ. Mat. Palermo 12 (1986) 99 -- 104. [25] S. B. Niefield and K. I. Rosenthal, Constructing locales from quantales, Math. Proc. Camb. Phil. Soc. 104 (2) (1988) 215 -- 234. [26] A. Okabe and R. Matsuda, Semistar-operations on integral domains, Math. J. Toyama Univ. 17 (1994) 1 -- 21. star operations on an integral domain, Math. J. Ibaraki Univ. 46 (2014) 31 -- 36. [27] J. Petro, Some results on the asymptotic completion of an ideal, Proc. Amer. Math. Soc. 15 (1964) 519 -- 524. [28] G. Picozza, Star operations on overrings and semistar operations, Comm. Alg. 33 (2005) 2051 -- 2073. [29] K. Rosenthal, Quantales and their Applications, Pitman Research Notes in Mathematics Series 234, Longman Scientific & Technical, Harlow, Essex, 1990. [30] J. C. Vassilev, Structure on the set of closure operations of a commutative ring, J. Alg. 321 (10) (2009) 2737 -- 2753. [31] M. Ward and R. P. Dilworth, Residuated lattices, Trans. Amer. Math. Soc. 45 (1939) 335 -- 354. California State University, Channel Islands, One University Drive, Camarillo, California 93012 E-mail address: [email protected]
1103.3865	1	1103	2011-03-20T17:24:04	-Clean Rings; Some Clean and Almost Clean Baer -rings and von Neumann Algebras	[ "math.RA" ]	A ring is clean (resp. almost clean) if each of its elements is the sum of a unit (resp. regular element) and an idempotent. In this paper we define the analogous notion for -rings: a -ring is -clean (resp. almost -clean) if its every element is the sum of a unit (resp. regular element) and a projection. Although -clean is a stronger notion than clean, for some -rings we demonstrate that it is more natural to use. The theorem on cleanness of unit-regular rings from [V. P. Camillo, D. Khurana, A Characterization of Unit Regular Rings, Communications in Algebra, 29 (5) (2001) 2293-2295] is modified for -cleanness of -regular rings that are abelian (or reduced or Armendariz). Using this result, it is shown that all finite, type I Baer -rings that satisfy certain axioms (considered in [S. K. Berberian, Baer -rings, Die Grundlehren der mathematischen Wissenschaften 195, Springer-Verlag, Berlin-Heidelberg-New York, 1972] and [L. Vas, Dimension and Torsion Theories for a Class of Baer -Rings, Journal of Algebra, 289 (2) (2005) 614-639]) are almost -clean. In particular, we obtain that all finite type I AW-algebras (thus all finite type I von Neumann algebras as well) are almost -clean. We also prove that for a Baer -ring satisfying the same axioms, the following properties are equivalent: regular, unit-regular, left (right) morphic and left (right) quasi-morphic. If such a ring is finite and type I, it is -clean. Finally, we present some examples related to group von Neumann algebras and list some open problems.	math.RA	math	∗-Clean Rings; Some Clean and Almost Clean Baer ∗-rings and von Neumann Algebras 1 1 0 2 r a M 0 2 Lia Vas Department of Mathematics, Physics and Statistics University of the Sciences in Philadelphia Philadelphia, PA 19104, USA ] . A R h t a m [ 1 v 5 6 8 3 . 3 0 1 1 : v i X r a Abstract A ring is clean (resp. almost clean) if each of its elements is the sum of a unit (resp. regular element) and an idempotent. In this paper we define the analogous notion for ∗-rings: a ∗-ring is ∗-clean (resp. almost ∗-clean) if its every element is the sum of a unit (resp. regular element) and a projection. Although ∗-clean is a stronger notion than clean, for some ∗-rings we demonstrate that it is more natural to use. The theorem on cleanness of unit-regular rings from [4] is modified for ∗-cleanness of ∗-regular rings that are abelian (or reduced or Armendariz). Using this result, it is shown that all finite, type I Baer ∗-rings that satisfy certain axioms (considered in [2] and [26]) are almost ∗-clean. In particular, we obtain that all finite type I AW ∗-algebras (thus all finite type I von Neumann algebras as well) are almost ∗- clean. We also prove that for a Baer ∗-ring satisfying the same axioms, the following properties are equivalent: regular, unit-regular, left (right) morphic and left (right) quasi-morphic. If such a ring is finite and type I, it is ∗-clean. Finally, we present some examples related to group von Neumann algebras and list some open problems. Key words: Clean rings, Baer ∗-rings, ∗-regular rings, Finite von Neumann algebras 1991 MSC: 16U99, 16W10, 16W99 1 Introduction The motivation for this paper came from a question posed by T.Y. Lam at the Conference on Algebra and Its Applications held at Ohio University, Athens, Email address: [email protected] (Lia Vas). Preprint submitted to Journal of Algebra 19 May 2014 OH in March, 2005. Lam asked which von Neumann algebras are clean as rings. A ring is clean if its every element can be written as the sum of a unit and an idempotent. Clean rings were defined by W. K. Nicholson in late 1970s in relation to exchange rings (see [22]) and have been attracting attention ever since. Clean rings are, in a way, an additive analogue of unit-regular rings: in a unit-regular ring, every element can be written as the product of a unit and an idempotent, for clean rings “the product” in this condition changes to “the sum”. Thus, it was not a coincidence that the question of the exact relationship between the classes of unit-regular and clean rings was of particular interest. In [4], the authors characterize unit-regular rings as clean rings in which every element a can be decomposed as the sum of a unit u and an idempotent e with aR ∩ eR = 0 (Camillo-Khurana Theorem). Other examples of clean rings include local rings, semiperfect rings and right artinian rings. A nice overview of commutative clean rings is given in the first section of [20]. A ring is almost clean if its every element can be written as a sum of a regular element (neither a left nor a right zero-divisor) and an idempotent. Up to date, almost cleanness has been considered mostly for commutative rings. The concept was introduced in [21]. In [21], it is shown that a commutative Rickart ring is almost clean. In this paper, we shall exhibit various classes of almost clean rings that are not necessarily commutative. Going back to Lam’s question, let us turn to von Neumann algebras. Since a von Neumann algebra is a ∗-ring (i.e. a ring with involution) in some cases it may be more natural to work with projections, self-adjoint idempotents, rather than with idempotents. Many other examples of different matrix rings and various rings of operators naturally come equipped with an involution. For such rings the properties of being ∗-regular, Baer ∗-ring or Rickart ∗-ring take over the roles of regular, Baer or (right or left) Rickart respectively (see, for example, [2]). Also, as S. K. Berberian points out in [2], “the projections are vastly easier to work with than idempotents” so one should utilize this opportunity when dealing with a ∗-ring. Because of this, we define the following more natural concept of cleanness for ∗-rings. Definition 1 A ∗-ring is ∗-clean (resp. almost ∗-clean) if its every element can be written as the sum of a unit (resp. regular element) and a projection. Clearly, if a ∗-ring is (almost) ∗-clean it is also (almost) clean. This definition is further justified by the following observation: if a unital ring R embeds in a clean (resp. ∗-clean) ring S with the same idempotents (resp. projections), then R is almost clean (resp. almost ∗-clean). Every finite von Neumann algebra A embeds in a unit-regular ring U (the algebra of affiliated operators). Moreover the ring U has the same projections as A. Since unit- 2 regular rings are clean by the Camillo-Khurana Theorem ([4, Theorem 1]), U is clean. Thus, we may hope that the cleanness of U will be preserved in A at least to the extent of almost cleanness. The problem is that the Camillo- Khurana Theorem does not guarantee that one can decompose an element in a unit-regular ∗-ring as the sum of a unit and a projection, just as the sum of a unit and an idempotent. We would need the stronger version to pull this decomposition from U back down to A. However, we shall demonstrate that for a class of von Neumann algebras the ∗-clean decomposition in U is possible and that it yields almost ∗-cleanness of A. The paper is organized as follows. In Section 2, we recall some known con- cepts and results. In Section 3, we prove some preliminary results and basic properties of ∗-clean rings. We also modify the Camillo-Khurana Theorem to fit the ∗-ring setting and prove that abelian (or reduced or Armendariz) ∗- regular rings are ∗-clean (Proposition 6). We also show that every ∗-ring that embeds in a ∗-clean regular ∗-ring with same projections is almost ∗-clean (Proposition 8). In Section 4, we consider a class of Baer ∗-rings for which the involution ex- tends to their maximal right ring of quotients (∗-extendible rings). We recall Baer ∗-ring axioms (A1)–(A7) considered in [2] and [26] and prove a result stating that every type In Baer ∗-ring satisfying (A2) is almost ∗-clean (The- orem 10). Using that, we show that every type I Baer ∗-ring that satisfies (A1)–(A6) is almost ∗-clean (Theorem 12). In particular, this theorem gives us that all finite, type I AW ∗-algebras (thus all finite type I von Neumann al- gebras as well) are almost ∗-clean (Corollary 14). If regularity is also assumed in Theorems 10 and 12 and Corollary 14, we obtain ∗-cleanness. In Section 4, we also prove that for a Baer ∗-ring that satisfies (A1)–(A6), the following properties are equivalent: regular, unit-regular, left (right) morphic and left (right) quasi-morphic (Corollary 13). In Section 5, we turn to AW ∗ and von Neumann algebras. Since every finite AW ∗-algebra of type I is almost ∗-clean (as Corollary 14 shows), we wonder if there are examples of almost ∗-clean AW ∗-algebras that are not type I. We show that a C ∗-sum of finite type I AW ∗-algebras is almost ∗-clean (Proposi- tion 15). Moreover, a C ∗-sum of finite type I regular AW ∗-algebras is ∗-clean. Using this fact, we consider a group von Neumann algebra that is not of type I but is ∗-clean. We also consider other group von Neumann algebras (Example 16) that provide examples of some other possible situations. We conclude the paper with a list of open problems in Section 6. 3 2 Preliminaries An associative unital ring R is a ∗-ring (or ring with involution) if there is an operation ∗ : R → R such that (x + y)∗ = x∗ + y∗, (xy)∗ = y∗x∗, (x∗)∗ = x for all x, y ∈ R. If a ∗-ring R is also an algebra over k with involution ∗, then R is a ∗-algebra if (ax)∗ = a∗x∗ for a ∈ k, x ∈ R. An element p of a ∗-ring R is called a projection if p is a self-adjoint (p∗ = p) idempotent (p2 = p). There is an equivalence relation on the set of projections of a ∗-ring R defined by p ∼ q iff x∗x = p and xx∗ = q for some x ∈ R. Recall that a ring is right Rickart (also called right p.p.-ring) if the right an- nihilator of each element is generated by an idempotent. A ring is said to be Rickart ∗-ring if the right annihilator of each element is generated by a projection. Moreover, it can be shown that such projection is unique. Since annl(x) = (annr(x∗))∗ the property of being Rickart ∗-ring is left/right sym- metric. In every Rickart ∗-ring, the involution is proper (x∗x = 0 implies x = 0, see [2, Proposition 2, p. 13]). The projections in a Rickart ∗-ring form a lattice ([2, Proposition 7, p. 14]) with p ≤ q iff p = pq (equivalently p = qp). Every element x of a Rickart ∗-ring R determines a unique projection p such that xp = x and annr(x) = annr(p) = (1 − p)R and a unique projection q such that qx = x and annl(x) = annl(q) = R(1 − q). In this case, p is called the right projection of x and is denoted by RP(x), and q is the left projection of x and is denoted by LP(x). A ∗-ring R is ∗-regular if any of the following equivalent conditions hold (see [2, Proposition 3, p. 229]): (1) R is regular and Rickart ∗-ring; (2) R is regular and the involution is proper; (3) R is regular and for every x in R there exists a projection p such that xR = pR. A Rickart C ∗-algebra is a C ∗-algebra (complete normed complex algebra with involution such that a∗a = a2) that is also a Rickart ∗-ring. A ring is Baer if the right (equivalently left) annihilator of every nonempty subset is generated by an idempotent. A Baer ∗-ring is a ∗-ring R such that the right annihilator of every nonempty subset is generated by a projection. A ∗-ring is a Baer ∗-ring if and only if it is a Rickart ∗-ring and the lattice of projections is complete ([2, Proposition 1, p. 20]). An AW ∗-algebra is a C ∗-algebra that is a Baer ∗-ring. If H is a Hilbert space and B(H) the algebra of bounded operators on H, then B(H) is an AW ∗- algebra. If A is a unital ∗-subalgebra of B(H) such that A = A′′ where A′ 4 is the commutant of A in B(H), then A is called a von Neumann algebra. Equivalently, A is a von Neumann algebra if it is a unital ∗-subalgebra of B(H) that is closed with respect to weak operator topology. A von Neumann algebra is an AW ∗-algebra ([2, Proposition 9]) while the converse is not true. 3 ∗-clean and almost ∗-clean rings In this section, we turn to ∗-rings. Recall that a ring is abelian if every idem- potent is central. The ∗-version of this concept is as follows. Definition 2 A ∗-ring is ∗-abelian if every projection is central. Clearly, a ∗-ring that is abelian is also ∗-abelian. The following lemma proves that for a Rickart ∗-ring, the conditions of abelian and ∗-abelian are equivalent and gives some further characterization of abelian right Rickart rings. Also, recall that a ring is said to be strongly clean if its every element can be written as the sum of a unit and an idempotent that commute. Clearly, a strongly clean ring is clean. If a ring is abelian, then clean implies strongly clean as well. We shall say that a ring is almost strongly clean if its every element can be written as the sum of a regular element and an idempotent that commute. Let us define a ∗-ring to be (almost) strongly ∗-clean if its every element can be written as the sum of a unit (resp. regular element) and a projection that commute. Lemma 3 (1) If R is a ∗-abelian Rickart ∗-ring, then every idempotent is a projection. (2) If R is a Rickart ∗-ring, then R is abelian iff R is ∗-abelian. (3) Let R be a (∗-)abelian Rickart ∗-ring. Then the following conditions are equivalent: (i) The ring R is ∗-clean. (ii) The ring R is clean. (iii) The ring R is strongly (∗)-clean. (4) If R is an abelian Rickart ∗-ring, then R is almost clean, almost ∗-clean and almost strongly (∗)-clean. PROOF. (1) Let e be an idempotent in a ∗-abelian Rickart ∗-ring R. Since R is Rickart ∗-ring, there is a projection p such that eR = annr(1 − e) = pR. Thus pe = e and ep = p. But since p is central, e = pe = ep = p. Condition (2) follows from (1). 5 (3) Condition (i) always implies (ii). By part (1), the converse holds for Rickart ∗-rings that are abelian (equivalently ∗-abelian by (2)). Condition (iii) always implies (ii). The converse clearly holds if all idempotents are central. (4) By [21, Proposition 16], a commutative Rickart ring is almost clean. The proof uses [6, Lemma 2 and Lemma 3]. These two lemmas ensure that an element in an abelian right Rickart ring is a product of a regular element and an idempotent. The proof of Proposition 16 uses just that the idempotents are central, not that the ring has to be commutative. Thus, we have that an abelian right Rickart ring is almost clean. So, if R is an abelian Rickart ∗-ring, it is almost clean. But since idempotents are projections by (1), R is also almost ∗-clean. Since all the idempotents (projections) are central, R is almost strongly ∗-clean as well. Before we turn to (almost) ∗-cleanness of some specific classes of rings, let us consider the following proposition proving some basic properties of ∗-clean rings. If R is a ∗-ring, the ring Mn(R) of n × n matrices over R has natural involution inherited from R: if A = (aij), A∗ is the transpose of (a∗ ij). So, Mn(R) is also a ∗-ring. Proposition 4 Let R be a ∗-ring. (1) If p is a projection such that pRp and (1 − p)R(1 − p) are ∗-clean, then R is ∗-clean. (2) If p1, p2, . . . , pn are orthogonal projections with 1 = p1 + p2 + . . . + pn, and piRpi ∗-clean for each i, then R is ∗-clean. (3) If R is ∗-clean, then Mn(R) is ∗-clean. PROOF. The proof of (1) follows the proof of [8, Lemma on p. 2590] with idempotent in Pierce decomposition of R changed to projection and every in- stance of idempotent further in the proof changed to projection. The condition (2) follows from (1) by induction, and (3) follows directly from (2). Let us now recall the statement of the Camillo-Khurana Theorem ([4, Theorem 1]). Theorem 5 (Camillo-Khurana) A ring R is unit-regular if and only if ev- ery element a of R can be written as a = u + e where u is a unit and e is an idempotent such that aR ∩ eR = 0. This theorem has the following proposition as a corollary. 6 Proposition 6 Let R be a ∗-ring. If R is ∗-regular and abelian, then every element a of R can be written as a = u + p where u is a unit and p is a projection such that aR ∩ pR = 0. PROOF. The ring R is unit-regular since it is regular and abelian ([7, Corol- lary 4.2, p. 38]). Thus, the condition equivalent to unit-regularity in Theorem 5 holds. Since every ∗-regular ring is Rickart ∗-ring and in every abelian Rickart ∗-ring an idempotent is a projection by Lemma 3, the proposition follows directly from Theorem 5. Remark 7 1. The assumption that R is abelian (let us denote this condition by (1)) in Proposition 6 can be replaced by any of the conditions below. (2) The ring R is semicommutative if ab = 0 implies aRb = 0 for all a, b ∈ R. (3) The ring R is symmetric if rab = 0 implies rba = 0 for all r, a, b ∈ R. (4) The ring R is Armendariz if p(x)q(x) = 0 then piqj = 0 for all p(x) = i=0 pixi and q(x) = Pm j=0 qjxj in R[x]. (5) The ring R is Armendariz of power series type if p(x)q(x) = 0 then piqj = 0 Pn for all p(x) = P∞ In general, (6) ⇒ (3) ⇒ (2) ⇒ (1), (6) ⇒ (5) ⇒ (2), and (5) ⇒ (4) ⇒ (6) The ring R is reduced if it has no nonzero nilpotent elements. i=0 pixi and q(x) = P∞ j=0 qjxj in R[[x]]. (1), and each of the implications is strict (see [10], [18] [17] and [11]). There are analogues of conditions (1)–(6) for R-modules (see [25] and [1]). In particular, a ring R satisfies one of the properties above if and only if the right module RR satisfies the corresponding module theoretic property. [1, Theorem 2.14] proves that the module version of the six conditions are all equivalent for a Rickart module. Thus, conditions (1) – (6) are equivalent for a right Rickart ring. Since regular rings are Rickart, the assumption that R is abelian in Propo- sition 6 can be replaced by any of the conditions (2)–(6). Also, in part (3) of Proposition 8, the condition that the ring is abelian could be replaced by any of the conditions (2)–(6). 2. The projection p and unit u from the statement of Proposition 6 commute since R is abelian so that R is strongly clean as well. 3. If R is a ∗-ring such that every element a of R can be written as a = u + p where u is a unit and p is a projection such that aR ∩ pR = 0, then R is unit-regular by Theorem 5. We will use the following result in the next section. Proposition 8 Let R be a ∗-ring that can be embedded in a ∗-clean regular ∗-ring with the same projections as R. (1) The ring R is almost ∗-clean. (2) If R is regular, then R is ∗-clean. 7 (3) If R is abelian, then R is almost strongly ∗-clean. PROOF. (1) Let Q be a ∗-clean ring in which R is embedded and a ∈ R. Then a = u + p for a unit u in Q and a projection p in Q. But p is in R by assumption. Thus, u = a − p is in R as well. Since u is a unit in Q, 0 = annQ r (u) and the same holds for the left annihilators. Thus u is regular. r (u) ⊇ annR (2) If R is regular, then every element is either invertible or a zero-divisor. Thus, the element u from the proof of (1) is invertible and so R is ∗-clean. (3) An abelian almost ∗-clean ring is almost strongly ∗-clean by Lemma 3. Remark 9 Let us note that parts (1) and (3) of Proposition 8 are not valid if “almost” is deleted. Note that Z (with trivial involution) can be embedded in Q with same projections (0 and 1) but Z is not (strongly) clean. 4 ∗-clean and almost ∗-clean Baer ∗-rings In this section, we introduce a class of ∗-rings that can be embedded in ∗- regular rings with the same projections. Using Proposition 8, we shall prove (almost) ∗-cleanness of a class of Baer ∗-rings. A ∗-ring R is said to be ∗-extendible if its involution can be extended to an involution of its maximal right ring of quotients Qr max(R). It is easy to see that this extension is unique in this case. Also, if R is ∗-extendible, then Qr max(R) is its maximal left ring of quotients (the isomorphism of R to the opposite ring of R extends to an isomorphism of Qr max(R) and its opposite ring). Thus, we can write Qmax for Qr max(R). If R is right nonsingular (thus left nonsingular as well since it is a ∗-ring), then Qmax is regular and left and right self-injective. Thus, Qmax is unit-regular (see [7, Theorem 9.29, p. 105]). max(R) = Ql If ∗ is proper in a nonsingular and ∗-extendible ring R, then the extension of ∗ is proper in Qmax. To see that let x∗x = 0 for some x ∈ Qmax. This implies that (xr)∗xr = r∗x∗xr = 0 and so xr = 0 for all r such that xr ∈ R. Since I = {r ∈ Rxr ∈ R} is a dense right ideal of R, xI = 0 implies that x = 0. The involution in every Rickart ∗-ring is proper and every Rickart ∗-ring is nonsingular. Thus, if R is ∗-extendible Rickart ∗-ring, then Qmax is ∗-regular. In [9], Handelman proved that a ∗-extendible Baer ∗-ring has the same pro- jections as its maximal ring of quotients Qmax. As is the case with many other properties of Baer ∗-rings, sufficient conditions for ∗-extendibility have been 8 given in terms of certain Baer ∗-ring axioms. Let R be a Baer ∗-ring and let us look at the following axioms. (A1) The ring R is finite if x∗x = 1 implies xx∗ = 1 for all x ∈ R. (A2) The ring R satisfies existence of projections (EP)-axiom: for every 0 6= x ∈ R, there exists a self-adjoint y ∈ {x∗x}′′ such that (x∗x)y2 is a nonzero projection; The ring R satisfies the unique positive square root (UPSR)-axiom: for every x ∈ R such that x = x∗ nxn for some n and some x1, x2, . . . , xn ∈ R (such x is called positive), there is a unique y ∈ {x∗x}′′ such that y2 = x and y positive. Such y is denoted by x1/2. 2x2 + . . . + x∗ 1x1 + x∗ (GC) The ring R satisfies the generalized comparability if for every two projections p and q, there is a central projection c such that cp (cid:22) cq and (1 − c)q (cid:22) (1 − c)p. Here p (cid:22) q means that p ∼ r ≤ q for some projection r. (LP∼RP) For every x ∈ R, LP (x) ∼ RP (x). It is easy to see that (A1) is equivalent to the condition that 1 is a finite projection (a projection is said to be finite if it is not equivalent to its proper subprojection). By [9, Proposition 2.10], a Baer ∗-ring with the following properties is ∗- extendible: (1) R is finite; (2) (LP∼ RP) holds; and (3) R has sufficiently many projections (for every x there is y with xy projection or, equivalently, every nonzero right (or left) ideal contains a nonzero projection). These con- ditions are met for finite Baer ∗-rings satisfying (A2) axiom: (LP∼RP) follows from (A2) by [2, Corollary on p. 131] and the existence of sufficiently many projections is guaranteed by (EP) axiom. Thus, every Baer ∗-ring satisfying (A1) and (A2) is ∗-extendible. Let us consider the following additional axioms. (A3) Partial isometries are addable. (A4) The ring R is symmetric: for all x ∈ R, 1 + x∗x is invertible. (A5) There is a central element i ∈ R such that i2 = −1 and i∗ = −i. (A6) The ring R satisfies the unitary spectral (US)-axiom: for a unitary u ∈ R with RP(1−u) = 1, there is an increasingly directed sequence of projections pn ∈ {u}′′ with sup pn = 1 such that (1 − u)pn is invertible in pnRpn for every n. (A7) The ring R satisfies the positive sum (PS)-axiom; if pn is orthogonal se- quence of projections with supremum 1 and an ∈ R such that an is positive (as defined in axiom (A2)) an ≤ pn, then there is a ∈ R such that apn = an for all n. Every finite AW ∗-algebra (so a finite von Neumann algebra as well) satisfies axioms (A1) – (A7) ([2, Remark 1, p. 249]). 9 In [2], Berberian uses (A1)–(A6) to embed a ring R in a regular ring Q: he shows that a Baer ∗-ring R satisfying (A1)– (A6) can be embedded in a regular Baer ∗-ring Q satisfying (A1) – (A6) such that R is ∗-isomorphic to a ∗-subring of Q, all projections, unitaries and partial isometries of Q are in R (see [2, Chapter 8]) and that Q is unique up to ∗-isomorphism (see [2, Proposition 3, p. 235]). The ring Q is ∗-regular as well since it is regular and Baer (thus Rickart as well) ∗-ring. It is called the regular ring of Baer ∗-ring R. The regular ring Q is also the maximal and the classical ring of quotients of R ([26, Proposition 3]). Thus, it is left and right self-injective. Every self-injective (both left and right) and regular ring is unit-regular (see [7, Theorem 9.29, p. 105]) so Q is unit-regular as well. If R satisfies (A1) – (A7), the ring Mn(R) of n × n matrices over R is a Rickart ∗-ring for every n (by [2, Theorem 1, p. 251]). Moreover, Mn(R) is semihereditary ([26, Corollary 5]). In [27] it is shown that Mn(R) is a finite Baer ∗-ring for every n ([27, Theorem 4]). This gave an affirmative answer to the question of Berberian ([2, Exercise 4D, p.253]) whether Mn(R) is a Baer ∗-ring if R satisfies (A1)–(A7). In [2], Berberian uses additional two axioms (called (A8) and (A9) in [26]) to prove that Mn(R) is finite. In [27], it is shown that (A9) follows from (A1)–(A7). Mimicking the type decomposition for von Neumann algebras, the three types of Baer ∗-rings have been considered. A Baer ∗-ring is said to be of: (i) type I if it has faithful abelian projection (a projection p is abelian iff pRp is abelian ring and p is faithful if there are no nontrivial central idempotents e with ep = 0); (ii) type II if it has faithful finite projection but no nontrivial abelian projec- tions; and (iii) type III if there is no nontrivial finite projections. A Baer ∗-ring has a unique decomposition into three Baer ∗-rings of the three different types. Moreover, a finer type decomposition of types I and II is possible. A Baer ∗-ring is said to be of: - type If if it is of type I and finite; - type I∞ if it is of type I and 0 is the only abelian projection; - type II1 if it is finite and 0 is the only finite central projection; - type II∞ if 0 is the only abelian and the only finite central projection. Every Baer ∗-ring has a unique decomposition into five Baer ∗-rings of the following types: If , I∞, II1, II∞ and III. For details on type decompositions see [2, Theorems 2 and 3, p. 94]. 10 Type If rings can be further classified. Let R be of type If such that there is a positive integer n and n equivalent orthogonal abelian projections that add to 1 (called homogeneous partition of 1 of order n). In this case R is said to be of type In. If R is of type If and has (GC), it can be uniquely decomposed into a C ∗-sum of types In ([2, Theorem 2, p. 115]). If R has (GC) and is of type In for some n, then R is ∗-isomorphic to Mn(A) where A is an abelian Baer ∗-ring uniquely determined up to ∗-isomorphism ([2, Proposition 2, p. 112]). This gives us the following result. Theorem 10 If a Baer ∗-ring R is of type In and satisfies (A2), it is almost ∗-clean. If, in addition, R is regular, then it is ∗-clean. PROOF. Note that R satisfies (GC) since (A2) holds ([2, Theorem 1, p. 80]). Thus, R is ∗-isomorphic to Mn(A) where A is an abelian Baer ∗-ring uniquely determined up to ∗-isomorphism ([2, Proposition 2, p. 112]). Moreover, if p1, p2, . . . , pn are orthogonal abelian projections forming a homogeneous par- tition of 1 of order n, then A is ∗-isomorphic to the corner p1Rp1 (and piRpi for any i = 1, . . . , n). A corner pRp of a ∗-ring R where p is a projection, preserves the following properties: Rickart ∗-ring, Baer ∗-ring (see [16, Theorem 4, p. 6, and p. 30]), regular (if x ∈ pRp then x = xyx implies x = xyx = (xp)y(px) = x(pyp)x), ∗- regular (if R has a proper involution then it is easy to see the involution in pRp inherited from R is also proper), and regular and right and left self-injective (see [7, Proposition 13.7, p. 98]). If R is a Baer ∗-ring with Q = Qmax(R) and p is a projection in R, then pQp is the maximal ring of quotients of pRp. This is because R is semiprime (every Baer ∗-ring is semiprime) and nonsingular (every Rickart ∗-ring is nonsingular) so pQp is an essential extension of pRp that is right self-injective and so it is equal to Qmax(pRp) (see [12, Proposition 13.39]). This last result is also in [3, Proposition 0.2, p. 135]. Thus, the maximal ring of quotients of the abelian Baer ∗-ring A ∼= p1Rp1 is p1Qp1. The ring p1Qp1 is an abelian ∗-regular ring so it is ∗-clean by Propo- sition 6. The partition p1, p2, . . . , pn is a homogeneous partition of order n for ∼= p1Qp1 ∼= Qmax(A) is ∗-clean for every i, Q is ∗-clean Q as well. Since piQpi by Proposition 4. The ring R satisfies (A1) and (A2) and so it is ∗-extendible. Thus, the pro- jections of R and Q are the same. Then R is almost ∗-clean by Proposition 8. Now we can prove the almost ∗-cleanness of rings of type If . First we prove the following proposition. 11 Proposition 11 Let R be a Baer ∗-ring that satisfies (A1) – (A6). If there are central orthogonal projections pn with supremum 1 and pnR are either of type In or equal to 0, then R is almost ∗-clean. If R is also regular, then (A6) is not needed, R is isomorphic to a direct product of rings pnR and R is ∗-clean. PROOF. As we noted before, axioms (A1)–(A6) are sufficient to guarantee that R embeds in a ∗-regular ring Q with same projections as R satisfying (A1)–(A6) (for more details see [2, Section 52]). It is easy to see that pnR is also a Baer ∗-ring that satisfies (A1)–(A6) for every n (here it is essential that the projections pn are central). By uniqueness of regular ring of Baer ∗-ring satisfying (A1)–(A6) (see [2, Proposition 3, p. 235]), it is easy to see that pnQ is the regular ring of pnR for all n and thus that pnR and pnQ have the same projections for every n. Since pnR and pnQ have the same projections and pnR is either trivial or of type In, pnQ is also either trivial or of type In as well. Then pnQ is ∗-clean by Theorem 10 for every n. The crucial ingredient in the rest of the proof is the following result given in [2, Theorem 2, p. 237]. (R) If R is a Baer ∗-ring that satisfies (A1) – (A6) and there are central or- thogonal projections pn with supremum 1, then the regular ring Q of R is ∗-isomorphic to the direct product of rings pnQ via x 7→ (pnx). Since a direct product of ∗-clean rings is ∗-clean (easy to see), the regular ring Q of ring R is ∗-clean by (R). Then R is almost ∗-clean by Proposition 8. To prove the last sentence of the statement of this proposition, note that if R is regular and satisfies (A1)–(A5), then (A6) holds as well (see [2, Exercise 4A, p. 247]) and R is equal to its regular ring. Then R is isomorphic to a direct product of rings pnR by (R). In this case, the regular rings pnR are ∗-clean by Theorem 10 and so R, a direct product of ∗-clean rings, is ∗-clean as well. Theorem 12 If a Baer ∗-ring R is of type If and satisfies (A2) – (A6), it is almost ∗-clean. If R is also regular, it is ∗-clean. PROOF. Every Baer ∗-ring of type If that satisfies (A2) has a sequence of orthogonal central projections pn with supremum 1 such that pnR is either 0 or of type In (see [2, Theorem 2, p. 115]). Condition (PC) assumed in [2, Theorem 2, p. 115] follows from (GC) (by [2, Proposition 2, p. 78]) and (GC) follows from (A2) ([2, Theorem 1, p. 80]). Thus every type If Baer ∗-ring that satisfies (A2)–(A6), satisfies the assumptions of Proposition 11. Thus R is almost ∗-clean. If, in addition, R is regular, R is ∗-clean also by Proposition 11. 12 Let us conclude this section with the observation that Baer ∗-rings with (A1)– (A6) are morphic exactly when they are regular (and this happens exactly when the rings are their own regular rings). Recall that a ring is right mor- phic if annr(x) ∼= R/xR for every x ∈ R. The following five conditions are equivalent (see [12, Exercise 19A, p. 270] and [28, Corollary 3.16]): (1) R is unit-regular, (2) R is regular and right morphic, (3) R is regular and left mor- phic, (4) R is right Rickart and left morphic, and (5) R is left Rickart and right morphic. Thus, if R is both left and right Rickart, the conditions of being unit-regular, left morphic and right morphic are equivalent. In [24], it is shown that every right morphic ring is left principally injective (also called left P-injective) meaning that every map Rx → R extends to R ([24, Theorem 24]). If R is both left and right Rickart, the conditions of being regular, left P-injective and right P-injective are equivalent. This follows from [24, Proposition 27] and [28, Corollary 3.15]. In [5], a weaker notion of right morphic ring, called right quasi-morphic, is considered: for every x ∈ R, there is y and z such that xR = annr(y) and annr(x) = zR. It is shown that the results of P-injectivity from [24] hold if “morphic” is replaced by “quasi-morphic”. Corollary 13 Let R be a Baer ∗-ring that satisfies (A1)–(A6). If Q is the regular ring of R, then the following conditions are equivalent: (1) R = Q. (2) R is regular. (3) R is unit-regular. (4) R is left morphic. (5) R is right morphic. (6) R is left quasi-morphic. (7) R is right quasi-morphic. PROOF. (1) ⇔ (2). If R = Q then R is regular since Q is. Conversely, if R is regular, then R = Q by uniqueness of the regular ring ([2, Proposition 3, p. 235]). (2) ⇔ (3). If R is regular, then R = Q implies that R is its own maximal (left and right) ring of quotients (see [26, Proposition 3] and note that the proof uses just (A1)–(A6) and not (A7)). Thus, R is regular and (left and right) self-injective. Then R is unit-regular (see [7, Theorem 9.29, p. 105]). The converse (3) ⇒ (2) always holds. (3) ⇔ (4) ⇔ (5). As observed above using results from [28], this is true for every Rickart ∗-rings. (6) ⇒ (2). As observed above using result from [28], if R is both left and 13 right Rickart, the conditions of being regular, left P-injective and right P- injective are equivalent. Since left quasi-morphic implies right P-injective by [24, Lemma 3], left quasi-morphic implies regular. By symmetry, (7) ⇒ (2) as well. The condition (3) implies (4) and (5) which imply (6) and (7) respectively. Hence, all seven conditions are equivalent. Note that in general the following hold: (2) is weaker than (3), (4) and (5) are weaker than (3), and neither (4) nor (5) implies the other. The examples of rings illustrating these claims can be found in [24]. Also, (6) and (7) are weaker than (3) – (5) and neither (6) nor (7) implies the other. The examples of rings illustrating these properties can be found in [5]. 5 ∗-clean and almost ∗-clean von Neumann algebras In this section, we turn to AW ∗ and von Neumann algebras. First, let us note that all AW ∗-algebras satisfy (A2)–(A7) (see [2, pp. 48, 70, 129, 233 and 244] and [14, p. 327]). In particular, all finite AW ∗-algebras satisfy (A1)–(A7). Thus we have the following. Corollary 14 An AW ∗-algebra of type If (in particular a von Neumann al- gebra of type If ) is almost ∗-clean. If it is also regular, it is ∗-clean. It is interesting to note that one of the key arguments in the proof of Propo- sition 11 is that a direct product of ∗-clean rings is ∗-clean. For C ∗-algebras, however, the concepts of direct product is not suitable concept for “product” since the coordinate-wise norm might fail to be complete. A direct sum of algebras, on the other hand, might fail to be a unitary algebra. Because of this, a more appropriate concept of a product is considered. If {Ai} is a family of C ∗-algebras, its C ∗-sum is defined to be the ∗-subalgebra of the direct product Q Ai containing all elements (ai) such that sup ai is finite. Then the supremum norm is complete and this is a C ∗-algebra (see [2, Section 10]). In [14], this algebra is denoted by P ⊕ Ai. The properties of being Rickart and Baer are preserved by C ∗-sums of C ∗-algebras and so a C ∗-sum of Rickart C ∗ and AW ∗-algebras remain Rickart C ∗ and AW ∗ respectively ([2, Proposition 1, p. 52]). When working with families of central orthogonal projections with supremum 1, C ∗-sums provide an appropriate framework because of the following result ([2, Proposition 2, p. 53]). 14 (P ⊕) If {pi} is an orthogonal family of central projections with supremum 1 in an AW ∗-algebra A, then A is ∗-isomorphic to P ⊕ piA via x 7→ (pix). Proposition 15 If Ai are If type AW ∗-algebras, then P ⊕ Ai is almost ∗- clean. If Ai are also regular, P ⊕ Ai is ∗-clean. PROOF. The proof follows the idea of the proof of Proposition 11. Let A denote P ⊕ Ai. It is easy to see that A is a finite AW ∗-algebra as well so it has the regular ring. Let Qi be the regular ring of Ai and Q the regular ring of A. Since the projections in Q and A are the same, there are central orthogonal projections pi in Q with supremum 1 such that piQ ∼= Qi by uniqueness of the regular ring. So Q is ∗-isomorphic to the C ∗-sum of Qi. On the other hand, since Q is regular, it is ∗-isomorphic to direct product of Qi by result (R) quoted in the proof of Proposition 11. The rings Qi are ∗-clean by Corollary 14 (rings Qi are of type If because rings Ai are of type If and the projections in Qi and Ai are the same). Thus, Q is ∗-clean since it is a direct product of ∗-clean rings. Since the projections in A and Q are the same, A is almost ∗-clean by Proposition 8. If Ai are regular, Ai = Qi and so Ai are ∗-clean. The C ∗-sum P ⊕ Ai is ∗- isomorphic to the direct product so it is ∗-clean as well. We shall exhibit an example of a von Neumann algebra that is ∗-clean and not of type If . An open question which still remains is whether there are such examples of type II1 (and the remaining three types as well). The consider- ation of factors of type II1 would provide an important step in proving (or disproving) the (almost) cleanness of algebras of type II1. To motivate further investigation in this direction, we finish the paper with some examples and some questions. Let us consider the class of group von Neumann algebras. This class provides us with some concrete examples of finite von Neumann algebras. Let G be a group. The complex group ring CG is a pre-Hilbert space with an inner product and an involution given by h X g∈G agg, X h∈G bhh i = X g∈G agbg and  X g∈G ∗ agg  agg−1. = X g∈G The Hilbert space completion of CG is l2(G), the space of square summable complex valued functions over the group G: l2(G) = { X g∈G agg X g∈G ag2 < ∞ }. 15 The involution from CG extends to l2(G). The group von Neumann algebra N G is the space of G-equivariant bounded operators from l2(G) to itself: N G = { f ∈ B(l2(G)) f (xg) = f (x)g for all g ∈ G and x ∈ l2(G) }. The algebra N G is a von Neumann algebra on H = l2(G) since it is the weak closure of CG in B(l2(G)) so it is a ∗-subalgebra of B(l2(G)) which is weakly closed (see [13, Example 9.7] for details). The algebra N G is finite since it has a normal, faithful trace trA(f ) = hf (1), 1i (every finite von Neumann algebra has such a trace and, conversely, a von Neumann algebra with such a trace is finite, for details see [13, Section 9]). The regular ring of N G is the algebra of affiliated operators U(G) (see [13, Section 8]). Since N G is finite, just types If and II1 are possible. Recall that a group is said to be virtually abelian if it has an abelian subgroup of finite index. The types of N G are classified according to the properties of group G as follows: (i) The algebra N G is of type If iff G is virtually abelian. If G is finitely generated, N G is of type II1 iff G is not virtually abelian. (ii) The algebra N G is of type II1 iff Gf has infinite index (Gf is the normal subgroup of G of elements with finitely many elements in the conjugacy class or, equivalently, of elements whose centralizer has finite index). (iii) The algebra N G is a factor (its center is isomorphic to C{1}) iff Gf is trivial. [13, Lemma 9.4, p. 337] contains more details. Moreover, N G is semisimple iff G is finite (see [13, Exercise 9.11, p. 367]). Note also that an AW ∗-algebra is commutative iff it is abelian ([2, Example 2, p. 90]). Thus, G is an abelian group iff N G is an abelian ∗-ring. Now let us consider the following examples. Example 16 (1) Finite abelian groups give rise to commutative and semi- simple group von Neumann algebras. Finite and not abelian groups give rise to non-commutative (nor reduced, nor Armendariz) semisimple group von Neumann algebras. (2) Let G = Z. Then N G is commutative and not semisimple (since G is infinite). Moreover, N G is not regular: N G can be identified with the space of (equivalence classes of) essentially bounded measurable complex function on the unit circle (see [13, Example 1.4]). Its regular ring, the algebra of affiliated operators U(G) of N G, can be identified with the space of (equivalence classes of) all measurable complex functions on a unit circle (see [13, Example 8.11]). Thus N G 6= U(G) and so N G is not regular by Corollary 13. By Corollary 13, N G is not morphic (nor quasi-morphic) since it is not regular. On the other hand, U(G) is strongly ∗-clean by Proposition 16 6 and N G is almost strongly ∗-clean by Proposition 8. Thus, this is an almost strongly clean Baer ∗-ring that is not morphic and not regular. Note also that the group ring CG is not clean by [19, Proposition 2.7] (stating that if R and G are commutative and the group ring RG is a clean ring, then R is clean and G is a torsion group.) Since N G is almost clean, we have the following situation: a non-clean ring is dense (in the topological sense, since N G is the closure of CG in the weak topology) in an almost clean ring which is dense in a clean ring (U(G)). (3) Let G = Z ⊕ D3 where D3 is the dihedral group (group of symmetries) of an equilateral triangle. This is an example of non-regular (so non-morphic as well) and non-abelian ring. The group G is virtually abelian and so N G is of type If . Thus N G is almost ∗-clean by Corollary 14. Note that the algebras in all of the above examples are If type (because all the groups considered are virtually abelian). Now let us consider some non- virtually abelian groups. Let {Gα}α∈Λ be a family of groups. Then N (Y Gα) ∼= X ⊕ N (Gα) ≤ Y N (Gα) The last inclusion could be strict if the family of groups is infinite and the algebras N (Gα) are not regular (e.g. every Gα is Z). This is because Q N (Gα) does not have to be complete and P ⊕ N (Gα) is always complete. The first two algebras are isomorphic since the elements pα = (δαβ1β)β∈Λ where 1α is the identity element in Gα, are central orthogonal projection with supremum 1 in N (Q Gα). Thus, N (Q Gα) is ∗-isomorphic to P ⊕ pαN (Q Gα) = P ⊕ N (Gα) by (P ⊕) ([2, Proposition 2, p. 53]). If all groups Gα are finite, the algebras N (Gα) are regular and so P ⊕N (Gα) = Q N (Gα) by Proposition 11 and by (P ⊕). Note that we have equality and not just isomorphisms because the isomorphism from Proposition 11 and from (P ⊕) that maps x = (xα) onto (pαx) = (xα), is identity in this case. Thus, P ⊕ N (Gα) = Q N (Gα) ∼= N (Q Gα) and N (Q Gα) is ∗-clean. (4) Let {Gn}∞ n=1 be a countably infinite family of finite groups and let G = Q Gn. If Gn are abelian, N G is an example of a type If (I1 in fact) regular ring that is not semisimple. If infinitely many groups Gn are not abelian (for example, we can take Gn = D3 for every n), then N G is a regular and ∗-clean (by Proposition 15) ring that is not of type If (since G is not virtually abelian). This last example shows that there are clean Baer ∗-rings outside of type If . Finally, let us consider some groups of type II1. 17 (5) Let G = Z ∗ Z be the free group on two generators. Then Gf is trivial and has infinite index and so N G is a factor of type II1. Is it (almost) clean? Note also that we can easily construct non-∗-isomorphic factors of type II1. For example, let Π be the group of permutations on a countably infinite set that leave fixed all but finitely many elements ([15, Example 6.7.7]). Then N (Z ∗ Z) and N (Π) are both factors of type II1 but are not ∗- isomorphic (see [15, Theorem 6.7.8]). 6 Questions We conclude the paper with a list of some questions. (1) Find an example of a ∗-ring that is clean but not ∗-clean. By Lemma 3, such example cannot be found in the class of abelian Rickart ∗-rings. (2) Which type If AW ∗-algebras (and Baer ∗-rings with (A1)–(A6)) are clean? We know that regular type If AW ∗-algebras are clean. Are there examples outside of this class? (3) Are AW ∗-algebras (and Baer ∗-rings with (A1)–(A6)) of type II1 (almost) clean? A possible first step in this consideration might be to determine if the group von Neumann algebra of the free group on two generators is almost clean. Then, consideration of other types could follow leading to a complete answer to Lam’s question on cleanness of von Neumann algebras. (4) Do Theorem 10, Theorem 12 and Corollary 14 hold if “almost clean” is replaced by “almost strongly clean”? In case of group von Neumann algebras, are all type If group von Neumann algebras almost strongly clean (i.e. if G is virtually abelian, is N G almost strongly clean)? Also, it would be interesting to see which results that hold for strongly clean rings hold for strongly ∗-clean rings as well. In addition, in [23] Nicholson asked if a unit-regular ring is strongly clean. We ask if a unit- regular and ∗-regular ring is strongly ∗-clean. References [1] N. Agayev, G. Gungoroglu, A. Harmanci and S. Halicioglu, Abelian Modules, Acta Mathematica Universitatis Comenianae, 78 (2009) no. 2, 235–244. [2] S. K. Berberian, Baer ∗-rings, Die Grundlehren der mathematischen Wissenschaften 195, Springer-Verlag, Berlin-Heidelberg-New York, 1972. [3] A. Cailleau, G. Renault, Sur lenveloppe injective des anneaux semi-premiers `a id´eal singulier nul, J . Algebra 15 (1970) 133–141. 18 [4] V. P. Camillo, D. Khurana, A Characterization of Unit Regular Rings, Communications in Algebra, 29 (2001) no 5, 2293–2295. [5] V. P. Camillo, W. K. Nicholson, Quasi-morphic rings, Journal of Algebra and Its Applications, 6 (2007) no. 5, 789–799. [6] S. Endo, Note on p.p. rings (a supplement to Hattoris paper), Nagoya Math. J., 17 (1960) 167–170. [7] K. R. Goodearl, Von Neumann regular rings, Monographs and Studies in Mathematics 4, Pitman, London, 1979. [8] J. Han, W. K. Nicholson, Extensions of clean rings, Communications in Algebra, 29 (2001) no. 6, 2589–2595. [9] D. Handelman, Coordinatization applied to finite Baer ∗-rings, Trans. Amer. Math. Soc., 235 (1978) 1–34. [10] C.Y. Hong, N.K. Kim, T.K. Kwak, Extensions of Generalized Reduced Rings, Algebra Colloquium, 12 (2005) no. 2, 229– 240. [11] C. Huh, Y. Lee, A. Smoktunowicz, Armendariz rings and semicommutative rings, Communications in Algebra, 30 (2002) no. 2, 751–761. [12] T. Y. Lam, Lectures on modules and rings, Graduate Texts in Mathematics, 189, Springer-Verlag, New York, 1999. [13] W. Luck, L2-invariants: Theory and Applications to Geometry and K-theory, Ergebnisse der Mathematik und ihrer Grebzgebiete, Folge 3, 44, Springer- Verlag, Berlin, 2002. [14] R. V. Kadison, J. R. Ringrose: Fundamentals of the theory of operator algebras, volume 1: Elementary theory, Pure and Applied Mathematics Series, 100, Academic Press, London-New York, 1983. [15] R. V. Kadison, J. R. Ringrose, Fundamentals of the theory of operator algebras, volume 2: Advanced theory, Pure and Applied Mathematics Series 100, Academic Press, London-New York, 1986. [16] I. Kaplansky, Rings of Operators, Benjamin, New York, 1968. [17] N. K. Kim, Y. Lee, Armendariz Rings and Reduced Rings, Journal of Algebra, 223 (2000) 477–488. [18] N. K. Kim, K. H. Lee, Y. Lee, Power series rings satisfying a zero divisor property, Comm. Algebra, 34 (2006) 2205–2218. [19] W. Wm. McGovern, A Characterization of Commutative Clean Rings, Int. J. Math. Game Theory Algebra, 15 (2006) no. 4, 403–413. [20] W. Wm. McGovern, Neat rings, Journal of Pure and Applied Algebra, 205 (2006) 243 – 265. [21] W. Wm. McGovern, Clean Semiprime f -Rings with Bounded Inversion, Comm. Algebra, 31 (2003) no. 7, 3295–3304. 19 [22] W. K. Nicholson, Lifting idempotents and exchange rings, Trans. Amer. Math. Soc., 229 (1977) 269–278. [23] W. K. Nicholson, Strongly clean rings and Fittings’ Lemma, Communications in Algebra, 27 (1999) 3583–3592. [24] W.K. Nicholson, E. S´anchez Campos, Rings with the dual of the isomorphism theorem, Journal of Algebra, 271 (2004) 391–406. [25] M. B. Rege, A. M. Buhpang, On reduced modules and rings, Int. Electron. J. Algebra, 3 (2008) 58–74. [26] L. Vas, Dimension and Torsion Theories for a Class of Baer ∗-Rings, Journal of Algebra, 289 (2005) no. 2, 614–639. [27] L. Vas, Class of Baer ∗-rings Defined by a Relaxed Set of Axioms, Journal of Algebra, 297 (2006) no. 2, 470 - 473. [28] H. Zhu, N. Ding, Generalized morphic rings and their applications, Communications in Algebra, 35 (2007) 2820–2837. 20
1206.3726	2	1206	2012-11-05T19:17:42	Global Dimensions of Some Artinian Algebras	[ "math.RA", "math.KT" ]	In this article we obtain lower and upper bounds for global dimensions of a class of artinian algebras in terms of global dimensions of a finite subset of their artinian subalgebras. Finding these bounds for the global dimension of an artinian algebra $A$ is realized via an explicit algorithm we develop. This algorithm is based on a directed graph (not the Auslander-Reiten quiver) we construct, and it allows us to decide whether an artinian algebra has finite global dimension in good number of cases.	math.RA	math	GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS ATABEY KAYGUN AND M UGE KANUNI Introduction The structure of arbitrary associative commutative unital artinian algebras is well-known: they are finite products of associative commutative unital local algebras [6, pg.351, Cor. 23.12]. In the semi-simple case, we have the Artin-Wedderburn Theorem which states that any semi-simple artinian algebra (which is assumed to be associative and unital but not necessarily commutative) is a direct product of matrix algebras over division rings [6, pg.35, Par. 3.5]. Along these lines, we observe a simple classification of artinian algebras and their representations in Proposition 1.3.2 (hereby re- ferred as the Classification Lemma) in terms of a category in which each object has a local artinian endomorphism algebra. This category is constructed using a fixed set of primitive (not necessarily central) idempotents in the underlying algebra. The Classification Lemma is a version of Freyd's Representation Theorem [4, Sect. 5.3]: from an artinian algebra A we create a category CA on finitely many objects, and then the category of A-modules can be realized as a category of functors which admit CA as their domain. This construction can also be thought as a higher dimensional analogue of the semi-trivial extensions of [10] for artinian algebras. We use the Classification Lemma as a starting point to develop a novel effective algorithm to determine finiteness of global dimensions of a large class of artinian algebras. As a first step, in Theorem 2.3.4 (hereby referred as the Source-Sink Theorem) we get explicit lower and upper bounds for the global dimension of artinian algebras in terms of the global dimensions of a pair of artinian subalgebras. Then in the rest of Section 2 we develop our algorithm. This algorithm is based on a directed graph we construct out of a given artinian algebra and it allows us to reduce the question of finiteness of the global dimension of a given artinian algebra to finiteness of the global dimensions of a finite subset of subalgebras. The graph GA we define in Definition 2.2.1 for an artinian algebra A is not the Auslander-Reiten quiver of A and it appears to be a simplified version of the natural quiver of an artin algebra defined in [7]. See Subsection 2.6 and Figure 1 in which we summarize the results and the algorithm we obtain in Section 2. As an application of our algorithm, we obtain a new result in Theorem 3.1.3: we show that the quotient of a free path algebra of a cycle-free graph by a nil ideal has finite global dimension. Using the same algorithm we were also able to reproduce two results from the literature: In Proposition 3.1.2 we prove that free path algebras over cycle-free graphs are hereditary, and in Proposition 3.1.5 we prove that incidence algebras have finite global dimension. See [9, Section 9] and [2, Proposition 1.4] for the classical proofs of these Propositions. 1 2 ATABEY KAYGUN AND M UGE KANUNI At this point, let us make a careful distinction between an artinian algebra (an algebra which satisfies a descending chain condition on left, right or bilateral ideals) and an artin algebra (an algebra which is finitely generated as a bimodule over a unital commutative artinian center.) In this article we assume that our algebras are artinian and we make no assumption on the finiteness of the k-dimensions of the algebras we work with over the base field k. Also, it is common to restrict to the type of a presentation the base algebra has in the relevant literature, such as assuming that the base algebra is a quotient of a free path algebra of a quiver by an ideal contained in the Jacobson radical, or in its square. See for example [1]. But we make no such restrictive assumptions about presentations of our artinian algebras except in Theorem 3.1.3 where we prove finiteness of global dimensions of a class of artinian algebras which contains all incidence algebras. Plan of this article. In Section 1 we either cite or prove some technical results we need in Section 2 in proving our main results. Section 2 contains our algorithm which calculates upper and lower bounds for global dimension of artinian algebras, and all other results needed to develop this algorithm. See Subsection 2.6 and Figure 1 for an overview. In Section 3 we apply the results we obtain in Section 2 to path algebras and incidence algebras, and make explicit calculations in Subsection 3.2. Standing assumptions and notation. Through out the article k is a base field. We do not make any assumption about the characteristic of k. All unadorned tensor products are assumed to be over k. We assume A is a unital associative algebra over k which may or may not be commutative. When we say A is artinian (resp. noetherian) we mean A satisfies descending (resp. ascending) chain condition on both left and right ideals. In this article we heavily use idempotents (elements which satisfy e2 = e), but we make no assumptions on whether these idempotents are in the center. We will use the notation Ax, xA and AxA to denote the left, right and two-sided principal ideals generated by an element x ∈ A, respectively. The multiplicative group of invertible elements in A is denoted by A×. We will use J(A) to denote the Jacobson Radical of A, i.e. the intersection of all maximal left ideals in A. We note that the intersection of all maximal right ideals is also J(A). All A-modules are assumed to be left A-modules unless otherwise explicitly stated. We will denote the category of left A-modules by A-Mod and the category of right A-modules by Mod-A. For a finite set U , we will use U to denote the cardinality of U . Finally, in our graph algebras in forming products of edges we follow the categorical convention: a product of the form f g in a path algebra corresponds to compositions of arrows of the form • f ←−− • g ←−− •. Acknowledgments. We would like to thank both Birge Huisgen-Zimmermann and Gene Abrams for pointing out the mistake in the earlier version of Proposition 1.2.3. 1. The classification lemma 1.1. Technical lemmata. Below we list few lemmata we need, without proofs. Our main references are [5, 6]. Lemma 1.1.1. For every x ∈ A, if x ∈ J(A) then 1 + x ∈ A×. GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 3 Lemma 1.1.2. [6, Theorem 19.1] A is local if and only if A \ A× is a two-sided ideal. Lemma 1.1.3 (Nakayama Lemma). Let M be a finitely generated left A-module. If J(A)M = M then M = 0. Lemma 1.1.4. Let A be an algebra which splits as a direct sum of two (left) ideals A = I ⊕ J. Then there are two orthogonal idempotents such that 1 = e + f . An A-module M is called semi-simple if every submodule of M is a direct summand of M . An algebra A is called semi-simple if A viewed as an A-module is semi-simple. Lemma 1.1.5. Assume A is an artinian k-algebra. Then A is left semi-simple if and only if it is right semi-simple if and only if J(A) = 0. Lemma 1.1.6. Let J be a nil ideal. Then any idempotent in A/J can be lifted to an idempotent in A. In other words, for every idempotent ǫ ∈ A/J there exists an idempotent e ∈ A such that e + J = ǫ. Definition 1.1.7. An idempotent e ∈ A is called primitive when for every pair of orthogonal idem- potents e1, e2 ∈ A if e = e1 + e2 then e1 = 0 or e2 = 0. Lemma 1.1.8. [6, Proposition 21.22] Assume A is an artinian algebra, e ∈ A idempotent, J ⊂ J(A). Then e + J is a primitive idempotent of A/J if and only if e is a primitive idempotent of A. Lemma 1.1.9. [6, Corollary 19.19] Assume A is an artinian k-algebra. Then A is local if and only if the only idempotents of A are 1 and 0. 1.2. Idempotents. Proposition 1.2.1. Assume A is an artinian k-algebra and e ∈ A is an idempotent. Then e ∈ A is a primitive idempotent if and only if eAe is a local ring. Proof. Assume e is a primitive idempotent. Note that, regardless of e being primitive, eAe is a k- algebra with e ∈ eAe being its identity element. Assume eAe is not local. Since eAe is also an artinian algebra on its own right, there exists an idempotent ef e ∈ eAe which is not 0 or e by Lemma 1.1.9. Then e = ef e + (e − ef e) and observe that since ef e(e − ef e) = (e − ef e)ef e = 0, we write e as a sum of two orthogonal idempotents in A which is a contradiction since e is primitive. So, e is primitive implies eAe is a local artinian k-algebra. Conversely, assume eAe is local artinian. Then we know that the only idempotents of eAe are e and 0. Assume that e can be written as a sum of two orthogonal idempotents in A as e = u + v. Then we see that eu = ue = u2 = u and ev = ve = v2 = v and therefore e = u + v = eue + eve in eAe. Since the only idempotents in eAe are 0 and 1 we see that eue = u = 0 or eve = v = 0, i.e. that e is primitive. Thus we conclude that if eAe is local artinian then e is primitive. (cid:3) Proposition 1.2.2. (Compare this with [6, Theorem 23.6]. Note that any artinian ring is semi- perfect.) Let A be an artinian k-algebra. Then the unit 1 ∈ A is a finite sum of pairwise orthogonal primitive idempotents. 4 ATABEY KAYGUN AND M UGE KANUNI idealsLn i=1 Ii. Then by Lemma 1.1.4 there exists pairwise orthogonal idempotents e′ i for i = 1, . . . , n, and 1A′ =Pn Proof. Since A is artinian, so is A′ := A/J(A) and since the Jacobson radical of A′ is trivial we see that A′ is (left) semi-simple by Lemma 1.1.5. So, A′ splits as finite direct sum of minimal (left) i ∈ A′ such that Ii = A′e′ i = f1 + f2 i ⊂ A′f1 ⊕ A′f2. For the converse inclusion, for some orthogonal idempotents f1, f2 of A′. Clearly, A′e′ i = A′f1 ⊕ A′f2. Since, observe that any x = r1f1 + r2f2 ∈ A′f1 ⊕ A′f2, xe′ A′e′ i is a minimal left ideal, we get A′f1 = 0 or A′f2 = 0. So f1 = 0 or f2 = 0. Any idempotent in A′ can now be lifted to an idempotent in A by Lemma 1.1.6, and by Lemma 1.1.8 the lifts can be chosen from primitive idempotents in A. i is primitive, assume e′ i. Now to show that e′ i. Then A′e′ (cid:3) i=1 e′ i = x ∈ A′e′ Proposition 1.2.3. (Compare this with [6, Exercise 21.17]) Assume A is an artinian k-algebra. If E and F are two finite sets of pairwise orthogonal primitive idempotents such that 1 =Pe∈E e =Pf ∈F f then E = F . Proof. We consider the right ideal f A of A for a fixed f ∈ F . One can see that EndA(f A) the k-algebra of right A-module endomorphisms of f A is a quotient of the local artinian k-algebra f Af which acts by multiplication on the left, thus itself is local artinian. We will denote Lx the endomorphism of the the element Lf ef would have been in the unique maximal ideal of f Af for every e ∈ E and Lf = right A-module f A given by any x ∈ f Af . Then the identity morphism Lf is split as Lf =Pe∈E Lf ef since f = Pe∈E f ef . Now, there must be at least one e ∈ E such that Lf ef ∈ (f Af )× otherwise Pe∈E Lf ef would have not been invertible. Consider the sequence of morphisms of right A-modules Lf−−−→ f A. Since Lf e = Lf ef is invertible, we see that f A must be a direct summand f A of eA as a right A-module. However, eAe is also local artinian, and therefore has no idempotents other than 0 and 1 by Lemma 1.1.9. This means eA is indecomposable. Then eA and f A must be isomorphic as right A-modules. Thus get that Le has a two-sided inverse Lf x : eA → f A for some f x which satisfies f = f xe and e = ef x. Then we get f e = f xe2 = f xe = f and ef = ef xe = e2 = e. These equalities imply that for each e ∈ E the idempotent f ∈ F is unique and vice versa. This gives us E = F . Le−−−→ eA (cid:3) 1.3. The classification lemma. Definition 1.3.1. For an artinian k-algebra A we associate a canonical k-linear category CA as follows. The set of objects of CA is a finite set E of orthogonal primitive idempotents which split 1A as 1A =Pe∈E e. This set is nonempty by Proposition 1.2.2. For any e, f ∈ Ob(CA) we let HomCA(e, f ) = f Ae and the composition is defined by the multiplication operation in A. Proposition 1.3.2 (The Classification Lemma). Let Homk(CA, k-Mod) be the category of k-linear functors from CA to k-Mod and their natural transformations. Then the category Homk(CA, k-Mod) is isomorphic to the category A-Mod of left A-modules. GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 5 Proof. Using decomposition of the identity element 1 =Pe∈E e in terms of a set of primitive idempo- tents, we split any left A-module as a direct sum of k-modules M =Le∈E eM and for any x ∈ f Ae, the left action by x defines a k-linear morphism Lx : eM → f M for any e, f ∈ E. In short, every left A-module defines a functor Φ(M ) : CA → k-Mod which is defined by Φ(M )(e) = eM on the level of objects, and for any f xe ∈ HomCA(e, f ) we let Φ(M )(f xe) := Lf xe the k-linear operator defined on M by left action of f xe ∈ f Ae. Conversely, for every functor M : CA → k-Mod we define a left A-module Ψ(M ) :=Le∈E M (e) where the left action of A on Ψ(M ) is defined by using the decompo- sition A =Le,f ∈E f Ae, i.e. f xe · m = 0 unless m ∈ M (e) and then f xe · m is defined as M (f xe)(m), the evaluation of m under the k-linear morphism M (f xe) : M (e) → M (f ). One can easily see that ΨΦ is the identity functor on the category of A-Mod, and conversely ΦΨ is the identity functor on Homk(CA, k-Mod). (cid:3) Corollary 1.3.3. The category CA we defined in Definition 1.3.1 for a given artinian algebra A is independent (up to an isomorphism) of the set of primitive idempotents splitting the identity. Proof. The categories CA one can define for E and for F are both isomorphic to the category of left A-modules, and therefore, isomorphic to each other. (cid:3) Remark 1.3.4. One can view our Classification Lemma 1.3.2 as a generalization of semi-trivial extensions developed in [10] for artinian algebras. The Classification Lemma, in fact, allows one to reduce an artinian algebra to a n × n generalized matrix ring where n = E is the number of primitive idempotents splitting the identity. Semi-trivial extensions are the 2 × 2 examples of this construction (not just for artinian algebras) in which one does not require primitivity of the idempotents. Since we work with artinian algebras, and we reduce our algebras to the submodules eAf using idempotents e and f until they can not be further reduced, (i.e. until idempotents are primitive) we end up with a larger matrix ring. 2. Global dimension of artinian algebras 2.1. Reductions. Remark 2.1.1. Assume M is a module in Mod-A. Define the flat dimension (also known as the weak dimension) of M as fl.dimA(M ) = sup{n ∈ N TorA n (M, N ) 6= 0, N ∈ Ob(A-Mod)} where the quantity of the right side is defined to be ∞ if the subset is unbounded in N. Similarly we define the projective dimension of M as pr.dimA(M ) = sup{n ∈ N Extn A(M, N ) 6= 0, N ∈ Ob(Mod-A)} Now we define the global and the weak global dimension of A as the quantities gl.dim(A) = sup{pr.dimA(M ) ∈ N ∪ {∞} M ∈ Ob(Mod-A)} 6 and ATABEY KAYGUN AND M UGE KANUNI wgl.dim(A) = sup{fl.dimA(M ) ∈ N ∪ {∞} M ∈ Ob(Mod-A)} Since A is not necessarily commutative, technically we should define a left and a right version of these dimensions. But because of Hopkins-Levitzki Theorem, if A is artinian it is also noetherian, and therefore, the left and right global dimensions and the global weak dimensions agree. (cf. [5, Cor. 5.60]) Definition 2.1.2. The two sided bar complex of an algebra A is defined as the differential graded A-bimodule CB∗(A) :=Ln>0 A⊗n+2 together with the differentials (−1)i(a0 ⊗ · · · ⊗ aiai+1 ⊗ · · · ⊗ an+1) dn(a0 ⊗ · · · ⊗ an+1) =Xi=0 We also define another differential graded module CB′ ∗(A) which is defined as a graded module as Ln>0 A⊗n+1 with similar differentials. Remark 2.1.3. The two sided bar complex CB∗(A) gives us a projective resolution of A viewed as an A-bimodule. Thus the modified bar complex CB′ ∗(A) is acyclic. For a right A-module X and a left A-module Y , the differential graded k-module C∗(X; A; Y ) := X ⊗A CB∗(A) ⊗A Y together with the induced differentials give us the Tor-groups TorA a second complex ∗ (X, Y ) in homology. We also define C ′ ∗(X; A; Y ) := X ⊗A CB′ ∗(A) ⊗A Y together with the induced differentials. Since C∗(X; A; A) is a projective resolution of X, we see that C ′ ∗(X; A; A) is acyclic. The same is true for C ′ ∗(A; A; Y ). Proposition 2.1.4. Consider the subcomplex S∗(X; A; Y ) :=Mn∈NMei∈E Xe0 ⊗ e0Ae1 ⊗ · · · ⊗ en−1Aen ⊗ enY of the complex C∗(X; A; Y ). Then the injection S∗(X; A; Y ) → C∗(X; A; Y ) is a quasi-isomorphism. Proof. Using the Pierce decomposition of A asLe,f ∈E eAf we can split C∗(X; A; Y ) as Xe0 ⊗ e1Ae2 ⊗ · · · ⊗ e2p−1Ae2p ⊗ e2p+1Y Mp>0 Mei∈E One can easily see that C∗(X; A; Y ) = S∗(X; A; Y ) ⊕Me6=f C ′ ∗(X; A; Ae) ⊗ C ′ ∗(f A; A; Y ) by counting the number of idempotents e2i 6= e2i+1 in the full decomposition of C∗(X; A; Y ). The result follows from the fact that C ′ (cid:3) ∗(f A; A; Y ) are acyclic. ∗(X; A; Ae) and C ′ 2.2. The directed graph of an artinian algebra. GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 7 Definition 2.2.1. Let us define a directed graph GA using an artinian algebra A and a complete set of primitive idempotents E splitting the identity. The set of vertices of GA is E. Two idempotents e and f are connected with a directed edge f ← e if and only if the k-vector subspace f Ae is a non-zero. Lemma 2.2.2. GA is independent of the choice of any complete set of primitive idempotents splitting identity. Proof. Assume E and F are two sets of primitive idempotents splitting the identity, and let GE A and GF A be the directed graphs defined in Definition 2.2.1 for these sets. By Proposition 1.2.3, we have a bijection ω : E → F which satisfies the identities e · ω(e) = e and ω(e) · e = ω(e) for every e ∈ E. These identities also give us an isomorphism between the right A-modules eA and ω(e)A. Now, for every e, e′ ∈ E we have an edge from e to e′ if and only if e′Ae is non-zero. The k-vector subspace e′Ae is also HomA(eA, e′A) the k-vector space of A-module morphisms from eA to e′A. Thus we obtain that HomA(eA, e′A) is non-zero if and only if HomA(ω(e)A, ω(e′)A) is non-zero. This proves that the graphs GE (cid:3) A we write using E and F are isomorphic. A and GF Remark 2.2.3. For the rest of the article, we will fix an artinian algebra A and a set E of primitive idempotents splitting the identity 1A of A. Let π0(GA) be the set of connected components of GA. Note that since E is finite, π0(GA) is also finite. For every α ∈ π0(GA) we let χα := {e ∈ E the vertex e appears in α}. Note that for each α ∈ π0(GA) eα := Xe∈χα e is a central idempotent in A and also is the identity element of the subalgebra A(χα). Moreover, the set {eα α ∈ π0(GA)} forms a set of pairwise orthogonal central idempotents. Definition 2.2.4. For a right A-module X we define the support of X as The support of a left A-module is defined similarly. EX := {e ∈ E Xe 6= 0} Lemma 2.2.5. Assume X and Y are right A-modules. If there are no paths starting in EY and ending in EX in the directed graph GA then Extn A(X, Y ) = 0 for every n > 1. The analogous result for TorA n (X, N ) is true for every left A-module N . Proof. The proof for the analogous result for TorA Here we give a proof for Extn the homology of the complex n (X, N ) follows immediately from Proposition 2.1.4. A(X, Y ). Use the projective resolution of X given by S∗(X; A; A). Then Mn>0Mei∈E HomA(Xe0 ⊗ e0Ae1 ⊗ · · · en−1Aen ⊗ enA, Y ) this time with the differentials Mn>0Mei∈E n−1Xi=0 8 ATABEY KAYGUN AND M UGE KANUNI A(X, Y ). On the other hand, we have HomA(Ze⊗eA, T ) ∼= with the induced differential, calculates Ext∗ Homk(Ze, T e) for every e ∈ E and for all right A-modules Z and T . So, we can rewrite the complex above as Homk(Xe0 ⊗ e0Ae1 ⊗ · · · en−1Aen, Y en) (dφ)(a0 ⊗ a1 ⊗ · · · ⊗ an) = (−1)iφ(· · · ⊗ aiai+1 ⊗ · · · ) + (−1)nφ(a0 ⊗ · · · ⊗ an−1)an for every φ ∈Lei∈E Homk(Xe0⊗e0Ae1⊗· · · en−1Aen, Y en) and homogeneous tensor a0⊗a1⊗· · ·⊗an ∈ Xe0 ⊗ e0Ae1 ⊗ · · · en−1Aen. The result follows. (cid:3) Proposition 2.2.6. gl.dim(A) = maxα∈π0(GA) gl.dim(A(χα)) Proof. Either we use Lemma 2.2.5, or we observe that one can see that the differential graded k-module S∗(X; A; Y ) splits as a direct sum of differential graded k-modules as S∗(X; A; Y ) = Mα∈π0(GA) S∗(X; A(χα); Y ) and then use the fact that weak global dimension and global dimensions agree when A is artinian. (cid:3) Remark 2.2.7. We will call an artinian algebra A connected if GA is connected, i.e. has only one con- nected component. An immediate corollary of Proposition 2.2.6 is that the problem of calculating the global dimension of an artinian algebra reduces to calculating the global dimensions of the connected subalgebras constructed out of connected components of GA. Therefore, we can assume, without any loss of generality, that A is connected. 2.3. The Source-Sink Theorem. Definition 2.3.1. For any U, V ⊆ E nonempty subset, we define and we also define a new (non-unital) subalgebra A(U ) of A by uAv U AV :=Mu∈UMv∈V A(U ) := U AU = Mu,u′∈U uAu′ Definition 2.3.2. Assume Γ is a directed graph with no loops or multiple edges, and U is an arbitrary subset of vertices. The quotient graph Γ/U is the new graph obtained from Γ by contracting every vertex in U to a single vertex, which is again denoted by U , and then deleting all loops and oriented multiple edges. A vertex is v in Γ is called a sink (resp. a source) if there are no directed edges leaving v (resp. ending at v.) Proposition 2.3.3. Assume that there exists a proper subset of primitive idempotents U ⊂ E with the property that the collapsed vertex U in the quotient graph GA/U is a source or a sink. Then we GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 9 have In particular, if A(U ) has infinite global dimension then so does A. max{gl.dim(A(U )), gl.dim(A(U c))} 6 gl.dim(A) Proof. We will use the fact that A is artinian, and therefore, the global and weak dimensions in both left and right variations all agree. Assume without generality that U is a sink in G/U , and that M is a left A-module. One can easily see that A splits as A = U AU ⊕ U AU c ⊕ U cAU ⊕ U cAU c However, we assumed that U is a sink in G/U . This means U cAU is 0. Therefore, given any free A-module F = A⊕I one has a splitting of the form F = A⊕I = (U AU )⊕I ⊕ (U cA)⊕I Hence any free A-module, restricted to a left A(U )-module becomes a flat left A(U )-module because U cA is a flat left A(U )-module. The same also holds for projective A-modules. This means any A- projective resolution P∗ when restricted to a A(U )-module becomes U P∗ which is a flat A(U )-module. Therefore for every left A-module M we get pr.dimA(M ) > fl.dimA(U )(U M ) which implies gl.dim(A) > wgl.dim(A(U )) = gl.dim(A(U )) If we repeat the same proof for right modules, we get the same inequality for gl.dim(A(U c)). (cid:3) Theorem 2.3.4 (The Source-Sink Theorem). Assume A is an artinian algebra such that there is a proper subset of primitive idempotents U ⊂ E with the property that the collapsed vertex U in the quotient graph GA/U is a source or a sink. Then max{gl.dim(A(U )), gl.dim(A(U c))} 6 gl.dim(A) 6 1 + gl.dim(A(U )) + gl.dim(A(U c)) Moreover, if A is flat over A(U ) and A(U c) then max{gl.dim(A(U )), gl.dim(A(U c))} 6 gl.dim(A) 6 max{1, gl.dim(A(U )), gl.dim(A(U c))} Proof. We proved that the lower bounds hold in Proposition 2.3.3. Now, without loss of generality we assume U is a sink in GA/U . Otherwise we switch U and U c, because U is a sink in GA/U if and only if U c is a source in GA/U c. We define a filtration Lp n =Xq6p Mn0+···+nq=n−q Mβi∈{U,U c} Xβ0⊗A(β0)⊗n0⊗β0Aβ1⊗· · ·⊗A(βq−1)⊗nq−1⊗βq−1Aβq⊗A(βq)⊗nq ⊗βqY We consider the sequence of idempotents in each component Xe0 ⊗ e0Ae1 ⊗ · · · en−1Aen ⊗ enY as an oriented path in GA. The filtration counts the number of times each path enters in an out of U and U c. p+q/Lp−1 When we consider the spectral sequence associated with the filtration we see that E0 p,q = Lp p+q 10 is given by Mq0+···+qp=q Mβi6=βi+1 ATABEY KAYGUN AND M UGE KANUNI Xβ0 ⊗ A(β0)⊗q0 ⊗ β0Aβ1 ⊗ · · · ⊗ A(βp−1)⊗qp−1 ⊗ βp−1Aβp ⊗ A(βp)⊗qp ⊗ βpY Consider the subalgebra B = A(U ) ⊕ A(U c). Then E0 derived category of B-modules p,q represents a (multi-)product in D+(B) the (1) Mβi6=βi+1 Xβ0 ⊗R B β0Aβ1 ⊗R B · · · ⊗R B βp−1Aβp ⊗R B βpY Notice that if A is flat over A(U ) and A(U c) then the columns in the E1-page collapse at q = 0 line for p > 0. In any case, if U were not a source and a sink, we would have had paths where the idempotents on the q = 0 line which would alternate between idempotents in U and U c. The first column E0 0,q, regardless of A is flat over A(U ) and A(U c), is represented by the product in the derived category while the second column E0 1,q is given by XU ⊗R B U Y ⊕ XU c ⊗R B U cY XU ⊗R B U AU c ⊗R B U cY In the first page E1 and the rest of the terms are zero since U is a sink. ∗,∗ the height of the first column (after taking supremum over all X and Y ) is bounded by the maximum of gl.dim(A(U )) and gl.dim(A(U c)). The height of the second column E1 1,q is bounded above by gl.dim(A(U )) + gl.dim(A(U c)). Since this is the second column, its contribution to the homological dimension of A is shifted by 1. In the flat case, we see that E1 p,q = 0 for p > 1 and q > 1. The only non-zero term on q = 0 line is at E1 (cid:3) 1,0 =Pu∈UPv∈U c Xu ⊗A(U ) uAv ⊗A(U c) vY and we have E1 p,0 = 0 for p > 1. Remark 2.3.5. One can obtain a version of Proposition 2.3.3 using Corollary 4.3 of [3], and a version of Theorem 2.3.4 as a consequence of Corollary 3 of [11] for artinian algebras. Remark 2.3.6. One can use Theorem 2.3.4 recursively and reduce the calculation of lower and upper bounds for the global dimension of A into calculations of the global dimensions of certain subalgebras A(U ) and A(U c) provided that the distinguished vertex U in the quotient graph GA/U is a source or a sink at each recursion step. The recursion tree will split the graph GA into full subgraphs and corresponding artinian subalgebras. Since E is finite, the recursion terminates and we obtain a partition U1, . . . , Um of E where every subset U ′ ⊂ Ui is neither source nor a sink in GA(Ui)/U ′ for each i = 1, . . . , m. The partition U1, . . . , Um, or the number of elements in the partition, need not be unique but for any such sequence we have the inequality max{gl.dim(A(U1)), . . . , gl.dim(A(Um))} 6 gl.dim(A) 6 (m−1)+gl.dim(A(U1))+· · ·+gl.dim(A(Um)) From this inequality we deduce that A has finite global dimension if and only if all of these terminal subalgebras have finite global dimension. We will further refine this algorithm in the next section. In the current version and in the best possible case we obtain the following Corollary. GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 11 Corollary 2.3.7. Assume GA has no oriented cycles. Then max{gl.dim(eAe) e ∈ E} 6 gl.dim(A) 6 E − 1 +Xe∈E gl.dim(eAe) If A is flat over A(U ) for every non-empty U ⊂ E then max{gl.dim(eAe) e ∈ E} 6 gl.dim(A) 6 max ({1} ∪ {gl.dim(eAe) e ∈ E}) 2.4. Further reductions. Definition 2.4.1. Fix a non-empty proper subset U ( E and define a new differential graded k- module RU ∗ by letting RU 0 = A and RU 1 =Lu∈U Au ⊗ uA and then n =M Au1 ⊗ u1Au2 ⊗ · · · ⊗ un−1Aun ⊗ unA RU ∗ is a differential graded submodule of CB′ Notice that RU for a right A-module X and a left A-module Y . Now, let us compute the homology Hn(RU consider the brutal truncation RU ∗(A). We define RU ∗ (X) and RU ∗ (Y ) similarly ∗ ). If we ∗>0 Au ⊗ uA Mu∈U d2←− Mu1,u2∈U Au1 ⊗ u1Au2 ⊗ u2A d3←− Mu1,u2,u3∈U we obtain S∗(A; A(U ); A). So, it is clear that Au1 ⊗ u1Au2 ⊗ u2Au3 ⊗ u3A d4←− · · · Hn(RU ∗ ) = TorA(U ) n−1 (A, A) for every n > 2. For lower degrees we must consider A d1←−Mu∈U Au ⊗ uA d2←− Mu1,u2∈U Au1 ⊗ u1Au2 ⊗ u2A Since d1 is really the multiplication morphism, for n = 0 we will get Then ker(d1) = ker(cid:0)Lu∈U Au ⊗ uA → A(cid:1) and H0S∗(A; A(U ); A) =Pu∈U Au ⊗A(U ) uA we get We have similar results for RU ∗ (X) and RU ∗ (Y ), here written only for X as follows: H0(RU ∗ ) = A Pu∈U AuA Au ⊗A(U ) uA → A! ∗ ) = ker Xu∈U X/(Pu∈U XuA) ker(Pu∈U Xu ⊗A(U ) uA → X) n−1 (X, A) TorA(U ) if n = 0 if n = 1 if n > 2 H1(RU HnRU ∗ (X) = Definition 2.4.2. One can define a graded multiplication structure on RU (x0 ⊗ · · · ⊗ xn) ∈ RU n and y = (y0 ⊗ · · · ⊗ ym) ∈ RU m we define x · y ∈ RU n+m as ∗ as follows: for any x = x · y := (x0 ⊗ · · · ⊗ xn−1 ⊗ xny0 ⊗ y1 ⊗ · · · ⊗ ym) 12 ATABEY KAYGUN AND M UGE KANUNI Note that for every (a0u ⊗ ua1) ∈ RU 1 we have (a0u ⊗ ua1) = (a0 ⊗ u)a1 = a0(u ⊗ ua1) and for n > 2 and (a0u1 ⊗ u1a1u2 ⊗ · · · ⊗ un−1an−1un ⊗ unan) ∈ RU n we see (a0u1 ⊗ u1a1u2 ⊗ · · · ⊗ un−1an−1un ⊗ unan) = (a0u1 ⊗ u1) · · · (an−1un ⊗ un)an In other words, RU ∗ is generated (not necessarily freely) by elements of degree 1. Proposition 2.4.3. RU RU RU ∗ (X) is a differential graded right RU ∗ (Y ). ∗ is a differential graded k-algebra. Moreover, if X is a right A-module then ∗ -module. A similar statement holds for a left A-module Y and Proof. Since RU satisfied only for elements of degree 0 and generators of degree 1. We have ∗ is generated by elements of degree 1, we must check whether the Leibniz rule is d((a0u ⊗ u)a1) = a0ua1 = d(a0u ⊗ u)a1 and for any (a0u ⊗ ua1) ∈ RU 1 . For two generators of degree 1 we check d(a0(u ⊗ ua1)) = a0ua1 = a0d(u ⊗ ua1) d((au ⊗ u)(bv ⊗ v)) =d(au ⊗ ubv ⊗ v) = (aubv ⊗ v) − (au ⊗ ubv) On the other hand d(au ⊗ u)(bv ⊗ v) − (au ⊗ u)d(bv ⊗ v) =(aubv ⊗ v) − (au ⊗ ubv) are equal. So, the Leibniz rule d(x · y) = d(x) · y + (−1) deg(x)x · d(y) holds. (cid:3) Proposition 2.4.4. If H0(RU ∗ (X)) = 0 for every A-module X and for every n > 0. In that case A is Morita equivalent to A(U ), and therefore we get gl.dim(A) = gl.dim(A(U )). ∗ ) = A/(Pu∈U AuA) = 0 then A is a flat A(U )-module and Hn(RU X is surjective for every right A-module X, i.e. H0(RU ∗ ) = 0 if and only if A =Pu∈U AuA. In particular, there exists αu, βu ∈ A such that 1 =Pu∈U αuuβu. Since 1 =Pu∈U αuuβu and we see that the action morphism XU ⊗A(U ) U A → ∗ (X)) = 0. Let Pu∈U xuu ⊗ uau be in the kernel of the same morphism, i.e. let 0 =Pu∈U xuuau. Then xuu ⊗ uauαu′u′βu′ =Xu,u′ xuu ⊗ uau =Xu,u′ xuuauαu′u′ ⊗ u′βu′ = 0 Proof. We have H0(RU Xu since the tensor product is over A(U ). H1(RU Q⊗A P ∼= A(U ). Then P ⊗A(U ) Q is isomorphic to A as a A-bimodule via the multiplication morphism. ∗ (X)) = 0. Put P = AU := Pu∈U Au and Q = U A := Pu∈U uA and then we see that In other words, the action morphism is also injective, i.e. GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 13 The result follows that A is Morita equivalent to A(U ). This means U A and AU are projective A(U )- modules. Then A is a flat A(U )-module. This gives us Hn(RU n−1 (X, A) = 0 for n > 1. (cid:3) ∗ (X)) = TorA(U ) Remark 2.4.5. The result we obtain in Proposition 2.4.4 shows obvious similarities with [10] for artinian algebras if one considers Remark 2, Theorem 2 and Remark 3 of ibid. together. However, our proof is specific to artinian algebras, and there are differences in assumptions such as using flatness as opposed to projectivity. Also, we realize the quotient A/(Pu∈U AuA) as the 0-th homology of a differential graded algebra which acts on the (co)homology groups in a crucial way, as opposed to a separate invariant. 2.5. Flatness of A as a bimodule over its subalgebras. Lemma 2.5.1. Assume that there exists a subset U ⊂ E such that A viewed as a (bi-)module over A(U ) is flat. Then gl.dim(A(U )) 6 gl.dim(A) This lemma appears as Lemma 1 of [8]. The proof we present below is different. Proof. Assume M is a left A(U )-module. Then the induced module IndA uM is a left A-module. Let P∗ be a A-projective resolution of the induced A-module IndA length pr.dimA(IndA A(U )(M ) :=Lu∈U Au ⊗A(U ) A(U )(M )) 6 gl.dim(A). Since Lu∈U uA is right A-projective and left A(U )-flat, and since the functorLu∈U uA ⊗A ( · ) sends projective A-modules to flat A(U )-modules, we see that A(U )(M ) of uA ⊗A P∗ is right A(U )-flat resolution of uP∗ =Mu∈U Mu∈U M =Mu∈U Xu′∈U uA ⊗A Au′ ⊗A(U ) u′M Then we easily see that wgl.dim(A(U )) 6 gl.dim(A). The result follows since A(U ) is also artinian. (cid:3) Definition 2.5.2. Assume we have a directed graph G, and a 2-coloring of vertices, i.e. the set of vertices is split as a union of two disjoint subsets. In this set-up an oriented path α is called alternating if any two consecutive vertices have opposite colors. Proposition 2.5.3. Assume there exists a proper subset U ⊂ E with the property that (i) U is not a source or a sink in GA/U , (ii) H0(RU and A(U c), and (iv) the global dimension gl.dimA(U ) and gl.dimA(U c) are both finite. In that case gl.dim(A) is infinite if and only if GA has at least one alternating cycle with respect to the coloring E = U ∪ U c. ∗ ) := A/(Pu∈U AuA) is non-zero, (iii) A is flat over both A(U ) Proof. Assume X = Y = H0(RU filtration we defined in the proof of Theorem 2.3.4. Note that since U is neither a source nor a sink in GA/U , we see that U AU c and U cAU are both non-zero. The first column of the associated spectral sequence in the E1-page yields E1 ∗ ) := A/(cid:0)Pu∈U AuA(cid:1) is non-trivial. Consider S∗(X; A; Y ) and the (X, Y ) and in particular 0,q = TorA(U c) q E1 0,0 = XU c ⊗A(U c) U cY 14 ATABEY KAYGUN AND M UGE KANUNI This is because X = Y are A-modules with the property that action of the subalgebra A(U ) is given by 0. We also see that E1 1,0 is XU ⊗A(U ) U AU c ⊗A(U c) U cY ⊕ XU c ⊗A(U c) U cAU ⊗A(U ) U Y and it is trivial. Moreover, since we assumed A is flat over A(U ) and A(U c), the spectral sequence collapses on two non-zero axis. Since U is not a source or a sink in GA/U , one can write paths of arbitrarily large lengths alternating between U and U c if and only if GA has at least one alternating path. This means on the q = 0 row of the the E1-page all odd dimensional vector spaces E1 2m+1,0 are zero and there are non-trivial term of arbitrarily large even dimension if and only if GA has at least one alternating path. The phenomenon is repeated in homology since the differentials are necessarily 0. This means in the E2-page on q = 0 row we have non-zero terms with arbitrarily large degrees if and only if GA has at least one alternating path. The result follows because the p = 0 column has bounded height due to the fact that A(U c) has finite global dimension. (cid:3) 2.6. The algorithm. Start with U ( E Test another U ( E. NO U is a source or a sink in GA/U NO YES gl.dimA(U ) and gl.dimA(U c) are both finite. YES gl.dimA is finite. A =Pu∈U AuA YES gl.dimA = gl.dimA(U ) NO NO gl.dimA is infinite. NO A is flat over NO A(U ) YES gl.dimA(U ) is finite. YES A is flat over A(U c) NO YES NO YES GA has a U -U c alternating cycle. YES gl.dimA(U c) is finite. Figure 1. An algorithm to estimate the global dimension of A. We summarize results we proved in this section in Figure 1 as an algorithm. This algorithm allows us to decide whether the global dimension of an artinian algebra is finite or infinite for a large class of algebras. However, there are also cases in which our algorithm fails to produce a definitive answer. We start with a proper subset U ( E. If U we chose is a source or a sink in the quotient graph GA/U , we use Theorem 2.3.4 and reduce the calculation of the global dimension of A to computing the global GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 15 dimensions of the subalgebras A(U ) and A(U c). Then we apply the algorithm to the subalgebras A(U ) we observe that A and A(U ) have the same global dimension by Proposition 2.4.4. In that case we replace A by A(U ) and proceed recursively. If it happens that that U is not a source or a sink in the and A(U c) recursively. If the subset U we chose satisfies the condition that A = Pu∈U AuA then quotient graph GA/U , and that A is not equalPu∈U AuA we use a combination of Lemma 2.5.1 and Proposition 2.5.3 to proceed. The algorithm will reduce the calculation of upper and lower bounds for global dimensions of an artinian algebra A at hand until it can no longer reduce A using the graph GA. The type of artinian algebras that can not be further reduced is called non-recursive whose definition we give below. The algorithm we described step by step in this section will determine the finiteness of global dimension of A in terms of the global dimensions of some subset (which is not unique and will depend on the choices made) of possibly non-recursive artinian subalgebras. Definition 2.6.1. An artinian algebra A is called non-recursive if it satisfies the following conditions for every U ( E: (1) U is not a source or a sink in GA/U (2) A 6=Pu∈U AuA (3) A is not flat over A(U ), or A is not flat over A(U c) and gl.dimA(U ) is finite. Otherwise, we will call A as recursive. 3. Incidence algebras and Quotients of free path algebras 3.1. Definitions and basic properties. Assume Q is a finite directed graph. The path algebra kQ of the finite directed graph Q is the vector space spanned by all the paths in Q. The multiplication on the paths is given by the concatenation of the sequences of edges of the paths p and q if source of p is the same as the range of q and zero otherwise. The set of all vertices in Q form a complete set of orthogonal idempotents. Moreover, each vertex in Q, is a primitive idempotent of the path algebra, and vice versa. Hence, the directed graph GkQ of the path algebra kQ in the sense of Definition 2.2.1 has the same vertex set as of Q. For any non-zero path α from vertex e to vertex f , α = f αe is in f (kQ)e which is a non-zero k-vector space of kQ. So there is an edge connecting e → f in GkQ and the directed graph GkQ is an extension of the directed graph Q. If Q does not have any oriented cycles, neither does GkQ. Lemma 3.1.1. Let A := kQ be the free path algebra over a finite directed graph Q. Then A is flat over A(U ) for every subset U of vertices in Q. Proof. Let U be a subset of vertices in Q. Then U A :=Pu∈U uA, the left ideal of paths ending at a vertex in U , has a basis over A(U ). This basis is the set of all paths starting at a vertex in U c, ending at a vertex in U and consist of no cycles within U . It is clear that this is a generating set, let us show that it is a basis. Assume we have distinct paths β1, . . . , βn in this generating set such that there are elements α1, . . . , αn in A(U ) with the property that P αiβi = 0. Then for every path γ appearing in α1 there is a path βj and a path γ′ in A(U ) appearing in αj such that γβ1 = γ′βj. Without loss 16 ATABEY KAYGUN AND M UGE KANUNI of generality, assume the length of β1 is less than or equal to the length of βj. Then, β1 is a part of βj, which forces βj to go through the vertex set U at least twice. This is a contradiction. Then the lengths of β1 and βj are the same, and therefore, they are the same. This is also a contradiction. So, we conclude that the elements β1, . . . , βn are linearly independent which means U A is free over A(U ), and therefore flat. Now, A splits as a direct sum A = U A ⊕ U cA. The action of A(U ) on U cA is by zero making it flat over A(U ). Therefore, A is flat over A(U ). The proof for A viewed as a right module A(U ) is similar. (cid:3) Proposition 3.1.2. [2, Proposition 1.4] Assume Q is a finite directed graph with no oriented cycles. Then the free path algebra kQ of Q is a hereditary k-algebra. Proof. Since kQ is a finite dimensional k-algebra, it is artinian. Moreover, Q ⊂ GkQ has no oriented cycles and e(kQ)e = k for every vertex in Q is of global dimension 0. The result follows using Corollary 2.3.7. (cid:3) Theorem 3.1.3. Let A be the free path algebra of a cycle-free directed graph Q = (Q0, Q1) and I be a nil ideal of A. Then the artinian algebra B := A/I has finite global dimension. idempotent u ∈ A such that π(u) = u by Lemma 1.1.6 and Lemma 1.1.8. Let e =Pu∈EB Proof. Let π : A → B be the canonical quotient morphism. Let EB denote a set of complete primitive idempotents of the artinian algebra B. Since I is nil, for each u ∈ EB we can pick a primitive u the sum of all of these primitive idempotents, and let f = 1A − e. Then f Af is an artinian algebra with unit f , and therefore, has its own set of primitive idempotents splitting its unit f . Thus we completed the lifted set of primitive idempotents {u ∈ A u ∈ EB} to a full set of primitive idempotents EA splitting 1A. This set of primitive idempotents might not be necessarily the set Q0, but one can find a bijection as in Proposition 1.2.2 and the resulting graph is isomorphic to GA by Lemma 2.2.2. Now, assume by way of contradiction that the directed graph GB of B has a cycle α which starts and ends in a primitive idempotent u. Then there exists a lift bα ∈ A of the cycle α which is a linear combination of cycles which start and end in an idempotent u ∈ EA. This is a contradiction since Q, and therefore GA, has no cycles. Now, by Corollary 2.3.7 we get the finite upper bound E − 1 because eBe ∼= k is of global dimension 0 for every e ∈ E. (cid:3) Remark 3.1.4. Note that if I is a nil ideal then it contains no idempotents other than 0. In particular, it contains no elements from the set of primitive idempotents. For any partially ordered set X, define the corresponding incidence algebra I(X) over k as the set of all functions f : X × X → k with f (x, y) = 0 unless x 6 y, together with the operations (f + g)(x, y) =f (x, y) + g(x, y) f g(x, y) = Xx6z6y (rf )(x, y) =rf (x, y) f (x, z)g(z, y) GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 17 for any r ∈ k and f, g ∈ I(X) and x, y ∈ X. However, since the partially ordered sets we consider are all finite, we can identify our incidence algebras as algebras generated by symbols Exy for every x 6 y subject to the relations (1) ExtEyz = 0 for x, y, z, t ∈ X when x 6 t and y 6 z, and finally t 6= y. (2) Exz = ExyEyz for x, y, z ∈ X when x 6 y 6 z. (3) Exy = ExxExy = ExyEyy for x, y ∈ X when x 6 y. A complete set of primitive idempotents of I(X) splitting the identity is {Exx : x ∈ X}. Hence, the set of all vertices of GI(X) is exactly X and there is an edge of the form x ← y in GI(X) if and only if x 6 y in X. Corollary 3.1.5. [9, Section 9] The incidence algebra of any finite poset has finite global dimension. Proof. Any finite poset X determines a directed graph GX = (X, E) where edges are defined by elements which are in relation. Then the incidence algebra I(X) over X is isomorphic to a quotient of the free path algebra kGX of this graph as follows. One can define an algebra epimorphism by mapping each path α in kX to the edge Exy which is an element in I(X) where x is the terminal point (target) of the path α and y is the initial point (source) of α. The kernel of this epimorphism is the subalgebra generated by the difference of paths with the same source and target. Then the result follows from Theorem 3.1.3. (cid:3) 3.2. Calculations. Example 3.2.1. Consider the directed graph Q e• x y * •f and the free path algebra A := kQ generated by this graph over our base field k. Let U = {e} and U c = {f }. We see that A(U ) = Spank(e, yx, (yx)2, (yx)3, . . .) and A(U c) = Spank(f, xy, (xy)2, (xy)3, . . .) are isomorphic to polynomial algebras over one indeterminate over k which are commutative and of global dimension 1. Moreover, A splits as an A(U ) module and as an A(U c)-module A = A(U ) ⊕ xA(U ) ⊕ A(U )y ⊕ xA(U )y ⊕ Spank(f ) A(U c) ⊕ yA(U c) ⊕ A(U c)x ⊕ yA(U c)x ⊕ Spank(e) This means A is flat over A(U ) and A(U c). Moreover, we have H0(RU non-zero. Then by Proposition 2.5.3 we see that gl.dim(kQ) is infinite. ∗ ) ∼= H0(RU c ∗ ) ∼= k both are Example 3.2.2. Consider the same directed graph Q and we define A := kQ/ hxy, yxi. Then A is a finite dimensional algebra A = Spank(e, f, x, y). Let U = {e} and U c = {f }. We see that A(U ) ∼= k ∼= A(U c) which are semi-simple, and therefore, of global dimension 0. This means A is * j j 18 ATABEY KAYGUN AND M UGE KANUNI flat over A(U ) and A(U c). Moreover, we have H0(RU Proposition 2.5.3 we see that gl.dim(A) is infinite. ∗ ) ∼= H0(RU c ∗ ) ∼= k both are non-zero. Then by Example 3.2.3. Consider the same directed graph Q and let A := kQ/ hxyi. We get a finite dimen- sional algebra A = Spank(e, f, x, y, yx) Again, let U = {e} and U c = {f }. We immediately see that A(U c) = Spank(f ) ∼= k is of global dimension 0. On the other hand, A(U ) = Spank(e, yx) of infinite global dimension because it is isomorphic to k[t]/(cid:10)t2(cid:11), and A splits as a A(U )-bimodule as A = A(U ) ⊕ Spank(f, x, y) Then A is flat over A(U ), and therefore, by Lemma 2.5.1 we conclude that A = kQ/ hxyi is of infinite global dimension. Example 3.2.4. Notice that Example 3.6 in [1] considers the path algebra A over Q •f0 •g0 x1 ❈❈❈❈❈❈❈❈ =④④④④④④④④④ y1 •e1 •f1 x2 ④④④④④④④④ ❈❈❈❈❈❈❈❈❈ y2 •g1 x3 ❈❈❈❈❈❈❈❈ =④④④④④④④④④ y3 •e2 subject to the relations x2x1 = 0 = y2y1, x3x2 = y3y2 and computes the gl.dim(A) = 2. Here we use category theoretic composition convention when we multiply elements. With the procedure defined above we get GA as: •g0 •f0 y1 ❈❈❈❈❈❈❈❈ =④④④④④④④④④ x1 •e1 z2 x2 ④④④④④④④④ ❈❈❈❈❈❈❈❈❈ y2 x3 ❈❈❈❈❈❈❈❈ =④④④④④④④④④ y3 / •e2 z1 / •f1 z3 / •g1 Let U = {f0, g0}, then A(U ) = f0Af0 ⊕ g0Ag0 which is isomorphic to k ⊕ k, So gl.dim(A(U )) = 0. Let B = A(U c). Now, pick new U = {e1, f1, g1}, then B(U ) is a free path algebra that is not semi-simple and B(U c) = e2Be2 ∼= k. Hence, gl.dim(A) 6 3 is an upper bound which does not contradict the actual dimension. ! ! = = ! ! = ! ! = ! ! / = = ! ! / = / ! ! = GLOBAL DIMENSIONS OF SOME ARTINIAN ALGEBRAS 19 Example 3.2.5. Consider the directed graph Q, •f0 •e0 x1 ④④④④④④④④ ❈❈❈❈❈❈❈❈❈ y1 •g0 x2 ❈❈❈❈❈❈❈❈ =④④④④④④④④④ y2 •e1 •f1 x3 ④④④④④④④④ ❈❈❈❈❈❈❈❈❈ y3 •g1 x4 ❈❈❈❈❈❈❈❈ =④④④④④④④④④ y4 •e2 · · · • •fn ⑦⑦⑦⑦⑦⑦⑦⑦⑦ ❅❅❅❅❅❅❅❅ •gn x2n ❉❉❉❉❉❉❉❉ =④④④④④④④④ y2n •en and let A be the path algebra over Q subject to the relations x2kx2k−1 = y2ky2k−1 x2k+1x2k = 0 = y2k+1y2k for k = 1, 2, ..., n with the same composition convention as above. Then GA will be: •g0 •e0 z1 y1 ④④④④④④④④ ❇❇❇❇❇❇❇❇❇ x1 •f0 y2 ❈❈❈❈❈❈❈❈ >⑤⑤⑤⑤⑤⑤⑤⑤⑤ x2 / •e1 / •f1 z2 x3 ④④④④④④④④ ❇❇❇❇❇❇❇❇❇ y3 / •g1 x4 ❈❈❈❈❈❈❈❈ >⑤⑤⑤⑤⑤⑤⑤⑤⑤ y4 / •e2 / •g2 · · · • / •f2 •fn zn ⑦⑦⑦⑦⑦⑦⑦⑦⑦ ❄❄❄❄❄❄❄❄ •gn x2n ❉❉❉❉❉❉❉❉ =⑤⑤⑤⑤⑤⑤⑤⑤⑤ y2n / •en Let U = {e0, f0, g0}, then A(U ) = e0Ae0 ⊕ e0Af0 ⊕ e0Ag0 ⊕ f0Af0 ⊕ g0Ag0. Since A(U ) is a free path algebra, it is hereditary. Moreover, A(U ) is algebra isomorphic to  semi-simple, as the submodule   which is not  is not a direct summand. So gl.dim(A(U )) = 1. By k k k 0 0 0 0 0 0 k k k 0 k 0 0 0 k Theorem 2.3.4, gl.dim(A) 6 1 + gl.dim(A(U )) + gl.dim(A(U c)) = 2 + gl.dim(A(U c)) Notice that if we call A = An and Q = Qn, then A(U c) is the path algebra on Qn−1 subject to the relations x2kx2k−1 = y2ky2k−1, x2k+1x2k = 0 = y2k+1y2k for k = 2, 3, ..., n. Use the same procedure for the algebra An−1. Pick U = {e1, f1, g1}. We get, gl.dim(An−1) 6 2 + gl.dim(An−2). If we continue in a similar manner, A0 becomes A0 = e0Ae0 which is isomorphic to k, so gl.dim(A0) = 0. Hence, gl.dim(A) 6 2n. [1] D.J. Anick and E.L. Green. On the homology of quotients of path algebras. Comm. Algebra, 15(1-2):309 -- 341, 1987. References = = ! ! = = ! ! > > ! ! ! ! = ! ! = = = = ! ! / = = ! ! / > > ! ! / / / > / > / = 20 ATABEY KAYGUN AND M UGE KANUNI [2] M. Auslander, I. Reiten, and Smalø S. O. Representation theory of Artin algebras, volume 36 of Cambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 1997. Corrected reprint of the 1995 original. [3] R. M. Fossum, P. A. Griffith, and I. Reiten. Trivial extensions of abelian categories. Lecture Notes in Mathemat- ics, Vol. 456. Springer-Verlag, Berlin, 1975. Homological algebra of trivial extensions of abelian categories with applications to ring theory. [4] P. J. Freyd. Abelian categories. Repr. Theory Appl. Categ., (3):1 -- 190, 2003. [5] T. Y. Lam. Lectures on modules and rings, volume 189 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1999. [6] T.Y. Lam. A first course in noncommutative rings, volume 131 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1991. [7] F. Li. Characterization of left Artinian algebras through pseudo path algebras. J. Aust. Math. Soc., 83(3):385 -- 416, 2007. [8] John C. McConnell. On the global dimension of some rings. Math. Z., 153(3):253 -- 254, 1977. [9] B. Mitchell. Theory of Categories, volume 17 of Pure and Applied Math. Series. Academic Press, 1965. [10] I. Palm´er. The global homological dimension of semi-trivial extensions of rings. Math. Scand., 37(2):223 -- 256, 1975. [11] I. Palm´er and J.-E. Roos. Explicit formulae for the global homological dimensions of trivial extensions of rings. J. Algebra, 27:380 -- 413, 1973. .. Department of Mathematics and Computer Science, Bahc¸es¸ehir University, Bes¸iktas¸ Istanbul, TURKEY E-mail address: [email protected] Department of Mathematics, Bogazic¸i University, Bebek Istanbul, TURKEY E-mail address: [email protected]
1102.2081	1	1102	2011-02-10T11:23:25	Minimal clones with weakly abelian representations	[ "math.RA" ]	We show that a minimal clone has a nontrivial weakly abelian representation iff it has a nontrivial abelian representation, and that in this case all representations are weakly abelian.	math.RA	math	MINIMAL CLONES WITH WEAKLY ABELIAN REPRESENTATIONS TAM ´AS WALDHAUSER Dedicated to B´ela Cs´ak´any on his seventieth birthday Abstract. We show that a minimal clone has a nontrivial weakly abelian representation iff it has a nontrivial abelian representation, and that in this case all representations are weakly abelian. 1. Introduction A concrete clone is a composition-closed collection of operations on some set con- taining all the projections. An abstract clone is a heterogeneous algebra equipped with operations which mimic the composition operations of concrete clones. (For the formal definition see [11] or [6].) A representation of an abstract clone is a homomorphism into the concrete clone of operations on a given set. Usually one obtains a representation by picking a set of generators of the clone and assigning to each of them an operation of the same arity on a set in such a way that this assignment extends to a clone homomorphism. Thus each representation gives an algebra, and these algebras form a variety. (If we choose another set of generators, then we get another variety which is term- equivalent to the previous one.) Conversely, every variety arises in this way from the clone of term functions of the countably generated free algebra in the variety. A clone is minimal if it has exactly two subclones: the clone itself and the clone which consists of projections only. The latter is called a trivial clone, and in this paper we will call an algebra trivial if the clone of its term functions is trivial (even if the algebra has more than one element!). Specially, a groupoid is trivial iff it is a left or right zero semigroup. A nontrivial representation of a minimal clone is also minimal, so if a variety has a minimal clone, then any nontrivial algebra in the variety has a minimal clone. Let us now recall the definition of four variants of abelianness (cf.[2]). For an algebra A let M(A) denote the set of 2 × 2 matrices of the form (cid:16) t(a,c) t(a,d) t(b,c) t(b,d)(cid:17) where t is a polynomial of A of arity n + m and a, b ∈ An, c, d ∈ Am. Definition 1.1. We say that an algebra A is (1) weakly abelian if (cid:0) u u (2) abelian if (cid:0) u u (3) rectangular if (cid:0) u v v w(cid:1) ∈ M(A) implies v = w; u v(cid:1) ∈ M(A) implies u = v; w u(cid:1) ∈ M(A) implies u = v = w; (4) strongly abelian if it is both abelian and rectangular. 2000 Mathematics Subject Classification. 08A40, 20N02. Key words and phrases. clone, minimal clone, (weakly) abelian algebra, groupoid. Research supported by the Hungarian National Foundation for Scientific Research grant no. T 026243 and the Research Group on Artificial Intelligence, HAS-SZTE. 1 2 T. WALDHAUSER All of these properties are inherited by subalgebras and direct products, but not by homomorphic images. If A is a groupoid, and we apply (1) to t(x, y) = xy then we get that whenever in the multiplication table of A we see a configuration like this: · · · ... a · · · ... b ... · · · · · · c ... u · · · ... u · · · ... · · · d ... u · · · ... v ... · · · then we must have u = v. Of course, this is just a necessary condition for A to be weakly abelian. Minimal clones with abelian representations have been described by K. Kearnes in [1]. Here we examine the analogous question for the other three concepts. We will show that if a minimal clone has a nontrivial weakly abelian representation, then it also has a nontrivial abelian representation, and all representations are weakly abelian. From this result we will easily deduce that if a minimal clone has a nontrivial rectangular representation, then it also has a nontrivial strongly abelian representation; moreover, all representations are strongly abelian. 2. Preliminary results Minimal clones are generated by any of their nontrivial elements and it is conve- nient to choose one of minimum arity. Such a generator must be one of five types according to the following theorem of Rosenberg [9] (see also [10]). Theorem 2.1. ([9]). Let f be a nontrivial operation of minimum arity in a minimal clone. Then f satisfies one of the following conditions: (I) f is unary, and f 2(x) = f (x), or f p(x) = x for some prime p; (II) f is a binary idempotent operation, i.e. f (x, x) = x; (III) f is a ternary majority operation, i.e. f (x, x, y) = f (x, y, x) = f (y, x, x) = x; (IV) f (x, y, z) = x + y + z for an elementary abelian 2-group with addition +; (V) f is a semiprojection, i.e. there exists an i (1 ≤ i ≤ n) such that f (x1, x2, . . . , xn) = xi whenever the arguments are not pairwise distinct. A minimal clone cannot contain operations of two different types, therefore we can speak about five types of minimal clones. Any representation of a clone of type (I) is strongly abelian; any nontrivial representation of a clone of type (IV) is abelian, but not rectangular (hence not strongly abelian). A minimal clone of type (III) or (V) cannot have a nontrivial weakly abelian representation. This is shown in Theorem 3.1 in [1]. (This theorem is about abelian representations, but the proof actually shows that there is no weakly abelian representation either.) Thus we have to consider clones of type (II) only. To recall the results of [1], we have to define several clones. By the clone of an affine space we mean the clone of all idempotent term functions of a vector space over some field. This clone is minimal iff the field is a p-element field for some prime number p. If p > 2, then this clone is of type (II): any nontrivial operation of the form λx + (1 − λ)y generates the clone. If p = 2 then the clone is of type (IV): the minority operation x + y + z is a generator of minimum arity. For any prime p, let us define the variety of p-cyclic groupoids by the identities xx = x, x(yz) = xy, (xy)z = (xz)y, (· · · ((xy)y) · · · )y = xyp = x. These groupoids MINIMAL CLONES WITH WEAKLY ABELIAN REPRESENTATIONS 3 have been introduced by P lonka [8]; he also proved that they have minimal clones [7]. Rectangular bands are idempotent semigroups satisfying xyz = xz, and they have minimal clones, too. Now we can describe all minimal clones with a nontrivial abelian representation (Theorem 3.11 in [1]). Theorem 2.2. ([1]). The minimal clones which have a nontrivial abelian repre- sentation are the following: (i) the unary clone generated by an operation f satisfying f (x) = f (y), but not satisfying f (x) = x; (ii) the unary clone generated by an operation f satisfying f 2(x) = f (x), but not satisfying f (x) = f (y) or f (x) = x; (iii) the unary clone generated by an operation f satisfying f p(x) = x for some prime p, but not satisfying f (x) = x; (iv) the clone of any nontrivial rectangular band; (v) the clone of an affine space over a prime field; (vi) the clone of any nontrivial p-cyclic groupoid (or its dual) for some prime p. The following interesting property of abelian representations has also been proved in [1] with the help of absorption identities (see also [6]). Theorem 2.3. ([1]). If a minimal clone has a nontrivial abelian representation, then this representation is faithful. As a special case of this theorem we have that if a variety V has a minimal clone and it contains a nontrivial rectangular band or affine space, then V must be the variety of rectangular bands or a variety of affine spaces. From the proof it is clear that the same is true for p-cyclic groupoids too, although not all of them are abelian, as we will see in the last section. 3. Weak abelianness and distributivity In the theory of groupoids and quasigroups a different notion of 'weak abelian- ness' is defined by the identities (∗) (xx)(yz) = (xy)(xz), (yz)(xx) = (yx)(zx), and a groupoid is called 'abelian' (or medial, or entropic) if (xy)(zu) = (xz)(yu) holds (see [4]). To avoid confusion with the universal algebraic definitions, we will use the word entropic in the latter case. Minimal clones are always idempotent, and in this case the identities (∗) are equivalent to the distributive identities: Left distributivity: Right distributivity: x(yz) = (xy)(xz), (yz)x = (yx)(zx). Any idempotent abelian groupoid is entropic ([1], Theorem 3.2), and one might expect that idempotent weakly abelian groupoids are distributive. We do not know if this is true or not, but for our present purposes the weaker properties stated in the next two lemmas are sufficient. Lemma 3.1. If A is an idempotent weakly abelian groupoid, then uv1 = uv2 = w implies u(v1v2) = w, i.e. {v uv = w} is a subuniverse for any given u, w ∈ A. Proof. Applying the definition of weak abelianness with a = (u, v1, u), b = (u, u, v1), c = v1, d = v2 for t(x1, x2, x3, x4) = (x1x2)(x3x4) we get (cid:18)(uv1)(uv1) (uu)(v1v1) hence u(v1v2) = w. (uv1)(uv2) (uu)(v1v2)(cid:19) =(cid:18)ww uv1 u(v1v2)(cid:19) =(cid:18)w ww w w u(v1v2)(cid:19) ∈ M(A), (cid:3) 4 T. WALDHAUSER Lemma 3.2. Any idempotent weakly abelian groupoid satisfies the following iden- tities: (i) (xy)(xz) = (x(yz))((xy)(xz)); (ii) (yx)(zx) = ((yx)(zx))((yz)x); (iii) (xy)x = x(yx). Proof. Let A be an idempotent weakly abelian groupoid. To prove (i), we will use the 8-ary term ((··)(··))((··)(··)); the underlined letters show the entries occupied by c and d in the definition. We have (cid:18)((xy)(xy))((xx)(zz)) ((xy)(xz))((xy)(xz)) ((xx)(yy))((xx)(zz)) ((xx)(yz))((xy)(xz))(cid:19) =(cid:18)(xy)(xz) (xy)(xz) (x(yz))(xy)(xz)(cid:19) ∈ M(A), (xy)(xz) therefore the equality in (i) holds. Doing the same with the dual hA, yxi of A = hA, xyi, which is of course also weakly abelian, we obtain the second identity. We could derive the third identity in a similar manner, but it is easier to deduce it from the previous ones. If we put z = x in (i) we get (xy)x = (x(yx))((xy)x); writing y = x and z = y in (ii) yields x(yx) = (x(yx))((xy)x); comparing them gives (iii). (cid:3) In light of the last identity we will sometimes omit the parentheses in a product of the form xyx. To make the connection between distributivity and weak abelianness more explicit, we will define a relation ∼ on our groupoid by a ∼ b iff ab = a. Identity (ii) says that A is right distributive 'modulo ∼'. This does not make perfect sense yet, since ∼ may not be an equivalence relation. Our strategy will be to reduce the problem to the case when ∼ is a congruence relation. As a preparation, we first show that assuming that the clone of A is minimal, we can conclude that A satisfies at least one-sided distributivity. Lemma 3.3. A weakly abelian groupoid with a minimal clone must satisfy at least one of the distributive laws. Proof. Suppose that A is a weakly abelian groupoid with a minimal clone, and A is neither left nor right distributive. First we will show that there is a two-element left zero semigroup in V(A). Since A is not right distributive, we can find elements x, y, z such that b = (yz)x 6= (yx)(zx) = a. The second identity of Lemma 3.2 shows that ab = a. If ba = b, then {a, b} is a two-element left zero subsemigroup of A. If ba 6= b, then let c denote the product ba, which is different from a by the weak abelian property (see the figure after Definition 1.1). We have ab = aa = a, so Lemma 3.1 yields that a = a(ba) = ac. With the help of identity (iii) of Lemma 3.2 we can compute cb = (ba)b = b(ab) = ba = c. Thus we have the following part in the multiplication table of A. a b c a a a a b c c c b c If bc = b, then again we have a two-element left zero subsemigroup, {b, c}. Suppose therefore that bc 6= b. Then x(xy) is a nontrivial operation, since a(ab) = aa = a 6= b and b(ba) = bc 6= b. However, the operation x(xy) is trivial on the set {a, c}. The only entry which we need to verify is c(ca) = c. We can get this equality by simply applying the definition of weak abelianness on the following matrix: (cid:18)c(bb) c(ba) c(cb) c(ca)(cid:19) =(cid:18)c c c c(ca)(cid:19) ∈ M(A). MINIMAL CLONES WITH WEAKLY ABELIAN REPRESENTATIONS 5 Therefore any operation in the clone generated by x(xy) is a first projection on {a, c}, and the original multiplication must be in this clone since it was supposed to generate a minimal clone. Thus we have ca = c, that is, {a, c} is a two-element left zero subsemigroup. Passing from A to its dual, which is not left or right distributive (since A itself is not right or left distributive) we see from the fact proved in the preceding paragraph that A also has a two-element right zero subsemigroup. The product of these two is a nontrivial rectangular band in V(A), therefore Theorem 2.3 implies that A itself is a rectangular band. This is a contradiction, since rectangular bands are distributive. (cid:3) With the help of Lemma 3.3 we will be able to handle all cases where ∼ is not a congruence relation, and finally we will arrive at the quotient groupoid A/∼, which will turn out to be distributive. This will be a rather lengthy argument, so we postpone it to the next section. Here we give the characterization of distributive groupoids with a minimal clone, which we will need to analyse A/∼. We will use the classification of entropic groupoids with a minimal clone (cf.[3]). To state this result, we need to define the following varieties. An idempotent semigroup is called a left normal band if it satisfies the identity xyz = xzy; similarly right normal bands are those satisfying the identity xyz = yxz. The variety of normal bands is the join of these two varieties. A groupoid is called a right semilattice if it satisfies the identities xx = x, x(yz) = xy, (xy)z = (xz)y and (xy)y = xy. The dual of a right semilattice is a left semilattice. Now we can describe the entropic groupoids which have a minimal clone. (Note that the statement is slightly different from Theorem 3.20 in [3], because here we formulate the description in terms of concrete clones instead of abstract clones.) Theorem 3.4. ([3]). Let A be an entropic groupoid with a minimal clone. Then A or its dual is an affine space, a rectangular band, a left normal band, a right semilattice or a p-cyclic groupoid. Let us turn to the investigation of distributive groupoids with a minimal clone. It was shown in [5] that every distributive groupoid is trimedial, i.e. any subgroupoid generated by at most three elements is entropic. The next theorem shows that the distributive and entropic properties are equivalent for groupoids with a minimal clone. Theorem 3.5. If A is a distributive groupoid with a minimal clone, then the en- tropic law holds in A. Proof. We know that all three-generated subgroupoids of A are entropic. If they are all trivial, then there must be a left and a right zero semigroup among them (since the clone of A is not trivial), and the product of these gives a nontrivial rectangular band in V(A). Applying Theorem 2.3, we get that A is a rectangular band. If there is a nontrivial 3-generated subalgebra which is an affine space, a rectangular band, or (the dual of) a p-cyclic groupoid, then again by Theorem 2.3 we have that A (or its dual) belongs to one of these varieties. Hence in all these cases A is entropic. So we can assume that every three-generated subgroupoid of A is a left or right semilattice or a normal band. If there is a nontrivial right semilattice among them, then the term x(xy) is the first projection on this subalgebra, hence by the min- imality of the clone we have that A = x(xy) = x. This equation does not hold in a left semilattice or in a normal band, except for a left zero semigroup (which is a right semilattice). Thus we have that every 3-generated subalgebra is a right semilattice. This means that all identities involving at most three variables which 6 T. WALDHAUSER hold in the variety of right semilattices also hold in A. Since right semilattices are axiomatizable by three-variable identities, we conclude that A itself is a right semilattice. The case of left semilattices is similar, so finally we can suppose that we have only normal bands as 3-generated subalgebras, i.e. that A satisfies all 3-variable identities that hold for normal bands. Associativity is such an identity, so our groupoid is a distributive semigroup, hence entropic (cf.[5], Proposition 2.3). (cid:3) Finally let us see which of the varieties mentioned in Theorem 3.4 contain non- trivial weakly abelian algebras. Theorem 3.6. If A is a weakly abelian entropic groupoid with a minimal clone, then A or its dual is a rectangular band, an affine space or a p-cyclic groupoid. Proof. By Theorem 3.4, we only need to show that A cannot be a left or right normal band, or left or right semilattice. A nontrivial semilattice is clearly not weakly abelian. In a nontrivial left normal band one can find elements a, b such that a 6= ab. It is easy to check that {a, ab} is a two-element subsemilattice, contradicting weak abelianness. Similarly, a nontrivial right normal band cannot be weakly abelian either. Finally, let us suppose that A is a right semilattice (the case of a left semilattice is similar). Considering the matrix (xx)(yy) (xx)(xy)(cid:19) =(cid:18)xy xy x(cid:19) ∈ M(A) we see that xy = x holds for all x, y ∈ A, and this contradicts the assumption that A has a minimal clone. (cid:3) (cid:18)(xy)(yy) (xy)(xy) xy 4. Left distributive weakly abelian groupoids with minimal clones Throughout this section A will denote a weakly abelian groupoid with a minimal clone. Lemma 3.3 shows that such a groupoid satisfies at least one of the distributive laws, so we will suppose that A is left distributive. We define a binary relation ∼ on A by a ∼ b iff ab = a. Clearly, this relation is reflexive. In a sequence of lemmas we will prove that if ∼ is not a congruence, then A is a p-cyclic groupoid for some prime p. Lemma 4.1. If ∼ is not symmetric, then A = x(xy) = x. Proof. Suppose that there are elements a, b ∈ A such that a ∼ b but b 6∼ a, that is, ab = a and ba = c 6= b. This situation is the same as in Lemma 3.3, and we will proceed similarly, but this time we go farther. Again, we have c 6= a by the weak abelian property. Let S be the subgroupoid of A generated by a and b. According to Lemma 3.1 {x ax = a} is a subuniverse of A, and it contains a and b. Therefore it contains S, which implies that a is a left zero element in this subgroupoid. Moreover, xy = a implies x = a for x, y ∈ S. This can be seen in the multiplication table of S by weak abelianness. a · · · a a · · · ... ... x ∗ · · · x · · · a · · · ... x · · · y a ... a (Note that we have xx = x by idempotence, and ∗ indicates xa, its value is irrele- vant.) MINIMAL CLONES WITH WEAKLY ABELIAN REPRESENTATIONS 7 Next we show that c is almost a left zero element in S; more precisely, cz = c for all z ∈ S \ {a}. Since z is in the subgroupoid generated by a and b, there is a binary term t such that t(a, b) = z. We prove cz = c by induction on the length of t. If this length is zero, then either t(x, y) = x or t(x, y) = y. The former is impossible because z 6= a. In the latter case we have cb = (ba)b = b(ab) = ba = c. Now for the induction step suppose that z = t(a, b) = uv with u = t1(a, b), v = t2(a, b). Again, u 6= a follows from z 6= a and therefore cu = c by the induction hypothesis. If v is also different from a, then cv = c, so cz = c(uv) = c by Lemma 3.1. If v = a, then we have to prove c(ua) = c. Let us consider the matrix c(ba) c(ua)(cid:19) =(cid:18) cb c(ub) cc c(ua)(cid:19) =(cid:18) c c(ub) c c(ua)(cid:19) ∈ M(A). c(ub) (cid:18)c(bb) of the form (cid:0) c c We know that cu = cb = c, therefore c(ub) = c as before. Therefore our matrix is c c(ua)(cid:1), hence cz = c(ua) = c by weak abelianness. What we just proved means that in the multiplication table of the subgroupoid S, the row of c is constant c except for ca which may be different. In the same way as we proved that xy = a implies x = a, we can show that xy = c implies x = c or y = a, that is, c can appear only in its own row and in the column of a. The knowledge we gathered about the multiplication table is enough to see that the operation x(xy) preserves S \ {c}. Indeed, if x(xy) = c for some x, y ∈ S, then either x = c or xy = a. The latter is impossible since it would force x = a, but then x(xy) = a 6= c. However, the original multiplication does not preserve this set because ab = c. Therefore by the minimality of the clone, x(xy) must be a projection. Since a(ab) = a 6= b, it can only be the first projection, i.e. the identity x(xy) = x holds in A. (cid:3) Lemma 4.2. If ∼ is symmetric but not transitive, then A = x(xy) = x. Proof. Suppose that there are elements a, b, c ∈ A such that a ∼ b ∼ c but a 6∼ c. Then a, b, c must be pairwise different, because ∼ is reflexive by the idempotence of A. A part of the multiplication table looks like this: a b a a a b b c c b c b c It is easy to check that we have the same in the multiplication table of x(xy). But for this operation we can compute the missing two entries, too, with the help of the left distributive identity: a(ac) = (ab)(ac) = a(bc) = ab = a, c(ca) = (cb)(ca) = c(ba) = cb = c. Thus we see that x(xy) is the first projection on the set {a, b, c}, but the original operation xy is not, because a 6∼ c implies ac 6= a. Therefore, by the minimality of the clone of A, x(xy) must be a trivial operation, hence A satisfies x(xy) = x. (cid:3) To finish the investigation of the cases where ∼ is not an equivalence relation, we will show that a weakly abelian groupoid with a minimal clone satisfying x(xy) = x must be a p-cyclic groupoid. This will be the consequence of the following lemma, where we do not assume weak abelianness. Lemma 4.3. Let A be a groupoid with a minimal clone such that A satisfies the identity x(yz) = xy. Then either A is a p-cyclic groupoid, or the identity (xy)y = xy holds in A. 8 T. WALDHAUSER Proof. Suppose that t1, t2 are two terms, and the leftmost variable of t2 is x. Then it can be shown easily by induction on the length of t2, that the identity t1t2=t1x holds in A. This means that any term t of A can be reduced to a left-associated product: t = (· · · ((xy1)y2) · · · )yn. Let us now compute what happens if we multiply a term with its leftmost variable: tx = tt = t because the leftmost variable of the underlined t is also x. Thus we have the same situation as in Claim 3.9 of [3], except that the order of the variables y1, . . . , yn is not irrelevant. However, when we compute binary terms, we do not have to permute them, so every binary term is of the form xyk, and we can proceed as in [3] to show that either (xy)y = xy or xyp = x holds for some prime number p. In the first case we are done, so let us suppose that the latter holds. One can check that the term t(x, y, z) = (((xyp−1)z)y)zp−1 satisfies the identities t(x, x, z) = t(x, y, x) = t(x, y, y) = x, i.e., it is a first semiprojection. Therefore t does not generate any nontrivial binary operation, so it must be trivial: t(x, y, z) = x. Substituting xy for x in this equality and multiplying both sides on the right with z we get the identity t(xy, y, z)z = (xy)z. Computing the left hand side we obtain the identity (xz)y = (xy)z. Thus all the defining identities of the variety of p-cyclic groupoids hold in A. (cid:3) Remark. One might think that in the case A = (xy)y = xy we can conclude that A is a right semilattice, but this is not true. The variety defined by the identities xx = x, x(yz) = xy, (xy)y = xy has a minimal clone. Indeed, any nontrivial term can be written in the form t = (· · · ((xy1)y2) · · · )yn, and identifying all the yis we get xyn = xy. However, these identities do not imply (xy)z = (xz)y, so the variety of right semilattices is a proper subvariety of the above variety. Lemma 4.4. If A is a weakly abelian groupoid with a minimal clone that satisfies the identity x(xy) = x, then A is a p-cyclic groupoid. Proof. We show that weak abelianness and the identity x(xy) = x imply the stronger identity x(yz) = xy. Let t = t(x, y, z) = x(yz), and compute the fol- lowing matrix: t(ty) x(yz) x(yy)(cid:19) =(cid:18)t (cid:18) t(tz) t t xy(cid:19) ∈ M(A). Thus we have x(yz) = xy and we can apply the preceding lemma. The only thing we need to show is that the identity (xy)y = xy cannot hold. We can proceed the same way as we did at the end of the proof of Theorem 3.6 to see that (xy)y = xy would imply xy = x. (cid:3) So far we have proved that if ∼ is not an equivalence relation, then A is a p-cyclic groupoid. From now on we will assume that ∼ is an equivalence relation, and we will force it to be a congruence of A. Using the left distributive identity we can show that ∼ is not very far from being a congruence. Lemma 4.5. For any a, b, c ∈ A, if a ∼ b then the following relations are true: (i) ca ∼ cb, (ii) (ac)(bc) ∼ ac. Proof. To prove (i) we simply apply the left distributive law: (ca)(cb) = c(ab) = ca. For (ii) we substitute x = c, y = a, z = b in the identity (yx)(zx) = ((yx)(zx))((yz)x), which holds in A by Lemma 3.2. We get (ac)(bc) = ((ac)(bc))((ab)c) = ((ac)(bc))(ac) which is just what we had to prove. (cid:3) It would be nice if we had ac ∼ bc in (ii), because then ∼ would be a congruence. With the next lemma we finish the investigation of the case where ∼ is not a congruence. MINIMAL CLONES WITH WEAKLY ABELIAN REPRESENTATIONS 9 Lemma 4.6. If ∼ is not a congruence relation, then A is a p-cyclic groupoid. Proof. We prove first that for any a, b, c ∈ A, if a ∼ b then the subalgebra generated by ac and bc satisfies the identity x(xy) = x. The second part of the previous lemma shows that uv ∼ u holds for u, v ∈ S = {ac, bc}. Next we show that this property is inherited when we pass from S to the subgroupoid generated by S. This can be done using the following two rules: (uw ∼ u, uv ∼ u) ⇒ (uv)w ∼ uv, (wu ∼ w, wv ∼ w) ⇒ w(uv) ∼ w. To check the first one, we calculate u((uv)w) = (u(uv))(uw) = u(uw) = u, which shows that u ∼ (uv)w. We have assumed u ∼ uv therefore by transitivity and sym- metry (uv)w ∼ uv follows. The second one is easier: w(w(uv)) = w((wu)(wv)) = (w(wu))(w(wv)) = ww = w. With these rules one can show by induction on the length of terms that uv ∼ u for all u, v in the subgroupoid generated by S. Hence this subgroupoid satisfies the identity x(xy) = x. If ∼ is not a congruence, then we can find elements a, b, c such that a ∼ b but ac 6∼ bc, that is, (ac)(bc) 6= (ac). If (ac)(bc) = bc, then by the second part of Lemma 4.5 we would have bc ∼ ac, which is impossible since ac 6∼ bc. Thus the subalgebra generated by {ac, bc} is not trivial. Then it has a minimal clone; it is weakly abelian, and satisfies x(xy) = x, therefore by Lemma 4.4 it is a nontrivial p-cyclic groupoid in V(A). With the help of Theorem 2.3 we conclude that V(A) is the variety of p-cyclic groupoids. (cid:3) Let us summarize what we have proved so far in this section. Theorem 4.7. If A is a weakly abelian left distributive groupoid with a minimal clone such that the relation ∼ defined by a ∼ b ⇔ ab = a is not a congruence, then A is a p-cyclic groupoid for some prime p. So finally we can suppose that A is a left distributive weakly abelian groupoid with a minimal clone, and ∼ is a congruence of A. The corresponding factor groupoid A/∼ is distributive; right distributivity follows, because A satisfies identity (ii) from Lemma 3.2. Furthermore, A/∼ has a minimal or trivial clone. Therefore it is entropic by Theorem 3.5, and it must have at least two elements, since A is not trivial. Using the list of entropic groupoids with a minimal clone, we will prove that A is also entropic. The key observation is that by the definition of ∼ we have A/∼ = t1 = t2 ⇔ A = t1t2 = t1. Lemma 4.8. If A/∼ has a two-element left or right zero subsemigroup then A is entropic. It is impossible to have a two-element semilattice among the subgroupoids of A/∼. Proof. First let us suppose that X, Y ∈ A/∼ form a left zero semigroup. Then for any x, y ∈ X ∪ Y we have xy ∼ x. Therefore x(xy) = x holds in X ∪ Y , which is a nontrivial subgroupoid of A, since X and Y are two different congruence classes. By Lemma 4.4 this subgroupoid must be p-cyclic, and by the minimality of the clone of V(A), Theorem 2.3 implies that A itself must also be a p-cyclic groupoid. Now suppose that X, Y ∈ A/∼ form a right zero semigroup. Again, X ∪ Y is a subgroupoid of A, and t1t2 = t1 holds in this subalgebra whenever the rightmost variables of t1 and t2 are the same (i.e., when t1 = t2 holds in right zero semigroups). Using this fact and the weak abelian property, we can compute x(yz) for x, y, z ∈ X ∪ Y as follows: (cid:18)((xy)y)z ((xx)y)z ((xy)z)z ((xx)z)z(cid:19) =(cid:18)(xy)z (xy)z (xy)z xz (cid:19) ∈ M(A), 10 T. WALDHAUSER therefore the identity (xy)z = xz holds in X ∪ Y . Similarly, X ∪ Y = x(yz) = xz can be shown by considering the following matrix: (cid:18)(xz)(zz) (xx)(zz) (xz)(yz) (xx)(yz)(cid:19) =(cid:18)xz xz x(yz)(cid:19) ∈ M(A). xz Thus X ∪Y is a rectangular band, and if it is nontrivial, then A is also a rectangular band by Theorem 2.3, so we are done. If X ∪ Y is trivial, then X and Y must be singletons, because X and Y are left zero subsemigroups. Therefore X ∪ Y is a right zero subsemigroup in A. Forming the direct product of this with any non- singleton congruence class we get a nontrivial rectangular band in V(A), so A is also a rectangular band by Theorem 2.3. If all the ∼-blocks of A are singletons, then A = A/∼ is distributive, hence entropic by Theorem 3.5. Finally, let us suppose that X, Y ∈ A/∼ form a semilattice. Then X ∪ Y satisfies every equation of the form t1t2 = t1 where t1 = t2 is valid in every semilattice. Combining this with identity (iii) from Lemma 3.2 allows us to conclude that the identities (xy)y = ((xy)y)(xy) = (xy)(y(xy)) = xy, (xy)x = ((xy)x)(xy) = (xy)(x(xy)) = xy hold in X ∪ Y . Using these identities we can compute the following matrix: (cid:18)(xy)y (xx)y (xy)x (xx)x(cid:19) =(cid:18)xy xy x(cid:19) ∈ M(A). xy Thus X ∪ Y is a left zero semigroup, contradicting the fact that X and Y are two different congruence classes. (cid:3) Theorem 4.9. If ∼ is a congruence relation of A, then A is entropic. Proof. There are at least two ∼-classes, since otherwise A would be a left zero semigroup. So A/ ∼ has at least two elements, and if it is trivial, then we can apply the previous lemma. If this is not the case, then A/∼ must belong to one of the varieties which have entropic minimal clones. In the case of affine spaces, rectangular bands and p-cyclic groupoids Theorem 2.3 shows that A also belongs to one of these varieties. As we have seen in the proof of Theorem 3.6, a nontrivial left or right normal band always contains a two-element subsemilattice, but Lemma 4.8 shows that this is impossible for A/∼. Finally, let us assume that A/∼ is a nontrivial right semilattice. Then it contains elements a, b such that a 6= ab. Using the defining identities of the variety of right semilattices, one can check that a and ab form a two-element left zero subsemigroup in A/∼, so we can apply Lemma 4.8 again. Similarly, a nontrivial left semilattice must contain a two-element right zero subsemigroup, so Lemma 4.8 applies in this case, too. (cid:3) Putting together Theorems 4.7 and 4.9 with Theorem 3.6 we get the main result of this section. Theorem 4.10. A left distributive weakly abelian groupoid with a minimal clone is either a rectangular band, an affine space or (the dual of ) a p-cyclic groupoid for some prime p. MINIMAL CLONES WITH WEAKLY ABELIAN REPRESENTATIONS 11 5. Summary We have seen that only minimal clones of types (I), (II) and (IV) can have nontrivial weakly abelian representations, and in case of types (I) and (IV) all representations are abelian. A weakly abelian groupoid with a minimal clone is left or right distributive by Lemma 3.3, thus we can apply Theorem 4.10 (after dualizing if necessary) to see that such a groupoid must be a rectangular band, an affine space or (the dual of) a p-cyclic groupoid. This list does not contain any new items compared to Theorem 2.2. Theorem 5.1. If a minimal clone has a nontrivial weakly abelian representation, then it also has a nontrivial abelian representation. Therefore such a clone must be a unary clone, the clone of an affine space, a rectangular band or (the dual of ) a p-cyclic groupoid for some prime p. Unary algebras, rectangular bands and affine spaces are abelian. A p-cyclic groupoid must be weakly abelian, as we shall see in the following lemma. Lemma 5.2. Every p-cyclic groupoid is weakly abelian. Proof. Suppose that A is a p-cyclic groupoid for some prime number p. (Actually, we will not need the fact that p is prime.) Let t be a term of A, with arity n + m, and let a, b ∈ An, c, d ∈ Am be such that the matrix (cid:0) t(a,c) t(a,d) form (cid:0) u u t(b,c) t(b,d)(cid:1) is of the u v(cid:1). As we have seen in the proof of Lemma 4.3, every term of A can be reduced to a left-associated product, so we may assume that t is of the form t = (· · · ((x1x2)x3) · · · )xn+m. Transposing our matrix if necessary, we can suppose that the leftmost variable is occupied by entries belonging to a and b, say a1 and b1. Using the identity (xy)z = (xz)y we can permute the other variables, so that the entries in the first column of the matrix are: t(a, c) = a1a2 · · · anc1c2 · · · cm, and t(b, c) = b1b2 · · · bnc1c2 · · · cm. (Both products are left-associated, we have omitted the parentheses.) Our groupoid is right cancellative, since multiplication by any element on the right is a permutation of order p. Therefore the equation t(a, c) = t(b, c) implies that a1a2 · · · an = b1b2 · · · bn. Multiplying both sides on the right with d1, d2, · · · , dm, we conclude that t(a, d) = t(b, d), that is u = v, so A is weakly abelian. (cid:3) Theorem 5.3. If a minimal clone has a nontrivial weakly abelian representation, then all representations are weakly abelian. As the following example shows, there exist nonabelian p-cyclic groupoids. There- fore the two abelianness concepts differ already for groupoids with minimal clones. Example. For any prime number p let us define the following binary operation on the set Zp × {0, 1}: (a, b) ◦ (c, d) =((a + 1, b) (a, b) if b = 0 and d = 1; otherwise. The algebra A = (Zp ×{0, 1}, ◦) is a p-cyclic groupoid, therefore it is weakly abelian and has a minimal clone. It is not abelian, as we can see from the following matrix. (cid:18)(0, 1) ◦ (0, 0) (0, 0) ◦ (0, 0) (0, 1) ◦ (0, 1) (0, 0) ◦ (0, 1)(cid:19) =(cid:18)(0, 1) (0, 0) (0, 1) (1, 0)(cid:19) ∈ M(A). We conclude with a remark on rectangularity and strong abelianness. A non- trivial affine space or p-cyclic groupoid cannot be rectangular, but unary algebras and rectangular bands are all strongly abelian. Thus these two concepts coincide for concrete minimal clones. 12 T. WALDHAUSER Theorem 5.4. If a minimal clone has a nontrivial rectangular representation, then it also has a nontrivial strongly abelian representation; moreover, all representations are strongly abelian. Such a clone must be unary, or the clone of rectangular bands. References [1] K. A. Kearnes, Minimal clones with abelian representations, Acta Sci. Math. (Szeged) 61 (1995), no. 1-4, 59 -- 76. [2] K. A. Kearnes, E.W. Kiss, Finite algebras of finite complexity, Discrete Math. 207 (1999), no. 1-3, 89 -- 135. [3] K. A. Kearnes, ´A. Szendrei, The classification of commutative minimal clones, Discuss. Math. Algebra Stochastic Methods 19 (1999), no. 1, 147 -- 178. [4] T. Kepka, The structure of weakly abelian quasigroups, Czechoslovak Math. J. 28(103) (1978), no 2, 181 -- 188. [5] T. Kepka, P. Nemec, Notes on distributive groupoids, Math. Nachr. 87 (1979), 93 -- 101. [6] L. L´evai, P. P. P´alfy, On binary minimal clones, Acta Cybernet. 12 (1996), no. 3, 279 -- 294. [7] J. P lonka, On groups in which idempotent reducts form a chain, Colloq. Math. 29 (1974), 87 -- 91. [8] J. P lonka, On k -cyclic groupoids, Math. Japon. 30 (1985), no. 3, 371 -- 382. [9] I. G. Rosenberg, Minimal clones I. The five types, Lectures in Universal Algebra (Szeged, 1983), Colloq. Math. Soc. J´anos Bolyai, 43, North-Holland, Amsterdam, 1986, 405 -- 427. [10] ´A. Szendrei, Clones in Universal Algebra, S´eminaire de Math´ematiques Sup´erieures, 99, Presses de L'Universit´e de Montr´eal, 1986. [11] W. Taylor, Characterizing Mal'cev conditions, Algebra Universalis 3 (1973), 351 -- 397. Bolyai Institute, University of Szeged, Aradi v´ertan´uk tere 1, H6720, Szeged, Hun- gary E-mail address: [email protected]
0812.0140	2	0812	2010-04-26T14:15:11	Homotopy Equivalences induced by Balanced Pairs	[ "math.RA", "math.RT" ]	We introduce the notion of balanced pair of additive subcategories in an abelian category. We give sufficient conditions under which the balanced pair of subcategories gives rise to equivalent homotopy categories of complexes. As an application, we prove that for a left-Gorenstein ring, there exists a triangle-equivalence between the homotopy category of its Gorenstein projective modules and the homotopy category of its Gorenstein injective modules, which restricts to a triangle-equivalence between the homotopy category of projective modules and the homotopy category of injective modules. In the case of commutative Gorenstein rings we prove that up to a natural isomorphism our equivalence extends Iyengar-Krause's equivalence.	math.RA	math	HOMOTOPY EQUIVALENCES INDUCED BY BALANCED PAIRS XIAO-WU CHEN Abstract. We introduce the notion of balanced pair of additive subcate- gories in an abelian category. We give sufficient conditions under which the balanced pair of subcategories gives rise to equivalent homotopy categories of complexes. As an application, we prove that for a left-Gorenstein ring, there exists a triangle-equivalence between the homotopy category of its Gorenstein projective modules and the homotopy category of its Gorenstein injective mod- ules, which restricts to a triangle-equivalence between the homotopy category of projective modules and the homotopy category of injective modules. In the case of commutative Gorenstein rings we prove that up to a natural isomor- phism our equivalence extends Iyengar-Krause's equivalence. 1. Introduction and Main Results Let A be an abelian category. Let X ⊆ A be a full additive subcategory which is closed under taking direct summands. Let M ∈ A. A morphism θ : X → M is called a right X -approximation of M , if X ∈ X and any morphism from an object in X to M factors through θ. The subcategory X is called contravariantly finite (= precovering) if each object in A has a right X -approximation (see [1, p.81] and [11, Definition 1.1]). → X −1 d−1 Recall that for a contravariantly finite subcategory X ⊆ A and an object M ∈ A an X -resolution of M is a complex · · · → X −2 d−2 → X 0 ε→ M → 0 with each X i ∈ X such that it is acyclic by applying the functor HomA(X, −) for each X ∈ X ; this is equivalent to that each induced morphism X −n → Kerd−n+1 is a right X -approximation. Here we identify M with Kerd1 and ε with d0. We denote → X −1 d−1 sometimes the X -resolution by X • ε→ M where X • = · · · → X −2 d−2 → X 0 → 0 is the deleted X -resolution of M . Note that by a version of Comparison Theorem, the X -resolution is unique up to homotopy ([12, p.169, Ex.2]). Recall that the X -resolution dimension X -res.dim M of an object M is defined to be the minimal integer n ≥ 0 such that there is an X -resolution 0 → X −n → · · · → X 0 → M → 0. If there is no such an integer, we set X -res.dim M = ∞. Define the global X -resolution dimension X -res.dim A to be the supreme of the X -resolution dimensions of all the objects in A. Date: Nov. 28, 2009. 1991 Mathematics Subject Classification. 18G25, 18E30,16E65. Key words and phrases. balanced pair, cotorsion triple, homotopy category, relative derived category, left-Gorenstein ring. This project was supported by China Postdoctoral Science Foundation (No.s 20070420125 and 200801230) and by National Natural Science Foundation of China (No.10971206). E-mail: [email protected], URL: http://math.ustc.edu.cn/∼xwchen. 1 2 XIAO-WU CHEN Let Y ⊆ A be another full additive subcategory which is closed under taking direct summands. Dually one has the notion of left Y-approximation and then the notions of covariantly finite subcategory, Y-coresolution and Y-coresolution di- mension Y-cores.dim N of an object N ; furthermore, one has the notion of global Y-coresolution dimension Y-cores.dim A. For details, see [3, Section 2] and [12, 8.4]. Inspired by [12, Definition 8.2.13], we introduce the following notion. Definition 1.1. A pair (X , Y) of additive subcategories in A is called a balanced pair if the following conditions are satisfied: (BP0) the subcategory X is contravariantly finite and Y is covariantly finite; (BP1) for each object M , there is an X -resolution X • → M such that it is acyclic by applying the functors HomA(−, Y ) for all Y ∈ Y; (BP2) for each object N , there is a Y-coresolution N → Y • such that it is acyclic by applying the functors HomA(X, −) for all X ∈ X . Balanced pairs enjoy certain "balanced" property; see Lemma 2.1. As mentioned above, the X -resolution of an object M is unique up to homotopy. Hence the condition (BP1) may be rephrased as: any X -resolution of M is acyclic by applying the functors HomA(−, Y ) for all Y ∈ Y. Similar remarks hold for (BP2). Balanced pairs arise naturally from cotorsion triples; see Proposition 2.6. We say that a contravariantly finite subcategory X ⊆ A is admissible if each right X -approximation is epic. Dually one has the notion of coadmissible covariantly finite subcategory. It turns out that for a balanced pair (X , Y), X is admissible if and only if Y is coadmissible; see Corollary 2.3. In this case, we say that the balanced pair is admissible. Moreover, for an admissible balanced pair (X , Y), X -res.dim A = Y-cores.dim A; see Corollary 2.5. If both the dimensions are finite, we say that the balanced pair is of finite dimension. For an additive category a, denote by K(a) the homotopy category of complexes in a. Our main result is as follows. It gives sufficient conditions under which a balanced pair of subcategories gives rise to equivalent homotopy categories of complexes. Theorem A. Let (X , Y) be a balanced pair of additive subcategories in an abelian category A which is admissible and of finite dimension. Then there is a triangle- equivalence K(X ) ≃ K(Y). The proof of Theorem A makes use of the notion of relative derived category; see Definition 3.1 and compare [26, 5]. In Section 3, we study the relation between homotopy categories and relative derived categories. Our second result is an application of Theorem A to Gorenstein homological algebra. Let R be a ring with identity. Denote by R-Mod the category of (left) R- modules, and by R-Proj (resp. R-Inj, R-mod) the full subcategory consisting of projective (resp. injective, finitely presented) R-modules. Recall from [2, p.400] that a complex P • of projective modules is totally-acyclic if it is acyclic and for any projective module Q the Hom complex HomR(P •, Q) is acyclic (also see [18, 21]). Following [11, 12] a module G is called Gorenstein projective if there is a totally- acyclic complex P • such that the zeroth cocycle Z 0(P •) is isomorphic to G, in which HOMOTOPY EQUIVALENCES INDUCED BY BALANCED PAIRS 3 case the complex P • is said to be a complete resolution of G. Denote by R-GProj the full subcategory of R-Mod consisting of Gorenstein projective modules. Note that R-Proj ⊆ R-GProj. Dually one has the full subcategory R-GInj of R-Mod consisting of Gorenstein injective modules and observes that R-Inj ⊆ R-GInj. Recall that a ring R is Gorenstein if it is two-sided noetherian and the regular module R has finite injective dimension on both sides. Following [3] a ring R is left-Gorenstein provided that any module in R-Mod has finite projective dimension if and only if it has finite injective dimension. Note that Gorenstein rings are left- Gorenstein (by [3, Corollary 6.11] or [12, Chapter 9]), while the converse is not true in general (see [10]). Our second result is as follows, the proof of which makes use of a characterization theorem of left-Gorenstein rings by Beligiannis ([3]; compare a recent work by Enochs, Estrada and Garc´ıa Rozas on cotorsion pairs on Gorenstein categories [9]). Theorem B. Let R be a left-Gorenstein ring. Then we have a triangle-equivalence K(R-GProj) ≃ K(R-GInj), which restricts to a triangle-equivalence K(R-Proj) ≃ K(R-Inj). Theorem B is related to a recent result by Iyengar and Krause ([21]). In their paper, they prove that for a ring R with a dualizing complex, in particular a commu- tative Gorenstein ring, there is a triangle-equivalence K(R-Proj) ≃ K(R-Inj) which is given by tensoring with the dualizing complex; see [21, Theorem 4.2]. We will re- fer to the equivalence as Iyengar-Krause's equivalence. In the case of commutative Gorenstein rings, we compare the equivalences in Theorem B with Iyengar-Krause's equivalence. It turns out that up to a natural isomorphism the first equivalence in Theorem B extends Iyengar-Krause's equivalence; see Proposition 6.2. We draw an immediate consequence of Theorem B. For a triangulated category T with arbitrary coproducts, denote by T c the full subcategory of its compact objects ([27]). For an abelian category A, denote by Db(A) its bounded derived category. We denote by Rop the opposite ring of a ring R. Corollary C. Let R be a left-Gorenstein ring which is left noetherian and right coherent. Then there is a duality Db(Rop-mod) ≃ Db(R-mod) of triangulated cat- egories. Proof. We apply Theorem B to get a triangle-equivalence K(R-Proj) ≃ K(R-Inj). Note that there are natural identifications K(R-Proj)c ≃ Db(Rop-mod)op by Nee- man ([28, Proposition 7.12]; compare Jørgensen [22, Theorem 3.2]), and K(R-Inj)c ≃ Db(R-mod) by Krause ([25, Proposition 2.3(2)]). Finally observe that a triangle- equivalence restricts to a triangle-equivalence between the full subcategories of com- pact objects. (cid:3) Let us remark that the duality above for commutative Gorenstein rings, more generally, for rings with dualizing complexes, is well known (compare [17, Chapter V] and [21, Proposition 3.4(2)]). It is closely related to Grothendieck's duality theory; see [28, Section 2]. We fix some notation. Recall that a complex X • = (X n, dn category a is a sequence X n of objects together with differentials dn such that dn+1 X )n∈Z in an additive X : X n → X n+1 X = 0; a chain map f • : X • → Y • between complexes consists X ◦ dn 4 XIAO-WU CHEN of morphisms f n : X n → Y n which commute with the differentials. Denote by C(a) the category of complexes in a and by K(a) the homotopy category; denote by [1] the shift functor on both C(a) and K(a) which is defined by (X •[1])n = X n+1and dn X . Recall that the mapping cone Cone(f •) of a chain map f • : X • → Y • is a complex such that Cone(f •)n = X n+1 ⊕Y n and dn Cone(f •) = (1 0) → X •[1] Y(cid:19). We have a distinguished triangle X • f • → Y • (0 1) → Cone(f •) X[1] = (−1)dn+1 in K(a) associated to the chain map f •. We also need the degree-shift functor (1) on complexes defined by (X •(1))n = X n+1 and dn X . Denote by (r) the r-th power of the functor (1). For a complex X • in an abelian category, denote by H n(X •) the n-th cohomology. For more on homotopy categories and triangulated categories, we refer to [30, 17, 20, 16, 15, 27]. X(1) = dn+1 (cid:18)−dn+1 X f n+1 0 dn 2. Balanced Pair and Cotorsion Triple In this section, we will study various properties of balanced pairs of subcategories in an abelian category. Balanced pairs arise naturally from cotorsion triples, while the latter are closely related to the notion of cotorsion pair ([19, 9]). Let A be an abelian category. Let us emphasize that in what follows all subcat- egories in A are full additive subcategories closed under taking direct summands. Recall that we have introduced the notion of balanced pair of subcategories in Section 1. The following "balanced" property of a balanced pair justifies the ter- minology. Lemma 2.1. Let (X , Y) be a balanced pair of subcategories in A. Let M, N ∈ A with an X -resolution X • → M and a Y-coresolution N → Y •. Then for each n ≥ 0 there exists a natural isomorphism H n(HomA(X •, N )) ≃ H n(HomA(M, Y •)). Proof. The result can be proven similarly as [12, Theorem 8.2.14]. One can also prove it by considering the two collapsing spectral sequences associated to the Hom bicomplex HomA(X •, Y •) as in the classical homological algebra ([6, Chapter XVI, Section 1]). (cid:3) Let X ⊆ A be a subcategory. Let Z • be a complex in A. We say that Z • is right X -acyclic provided that the Hom complexes HomA(X, Z •) are acyclic for all X ∈ X . Dually we have the notion of left Y-acyclic complex. The following observation is useful. Proposition 2.2. Let A be an abelian category, and let X (resp. Y) be a con- travariantly finite (resp. covariantly finite) subcategory. Then the pair (X , Y) is balanced if and only if the class of right X -acyclic complexes coincides with the class of left Y-acyclic complexes. Proof. Note that an X -resolution is right X -acyclic and the condition (BP1) says that an X -resolution is left Y-acyclic. Dual remarks hold for (BP2). Thus the "if" part follows immediately. To see the "only if" part, assume that the pair (X , Y) is balanced. We only show that right X -acyclic complexes are left Y-acyclic and leave the dual part to the reader. Assume that Z • = (Z n, dn Z )n∈Z is a complex. Consider the induced "short" HOMOTOPY EQUIVALENCES INDUCED BY BALANCED PAIRS 5 Z → Z n → Kerdn+1 complexes 0 → Kerdn Z → 0 for all n ∈ Z. Let us remark that such "short" complexes are left exact sequences, and that they are not necessarily short exact sequences. Since Z • is right X -acyclic, all the induced "short" complexes are right X -acyclic. Observe that if all the induced "short" complexes are left Y- acyclic, then so is Z •. Therefore it suffices to show that a left short exact sequence which is right X -acyclic is necessarily left Y-acyclic. Let 0 → M ′ → M → M ′′ → 0 be a left exact sequence which is right X - acyclic. We will show that it is left Y-acyclic. Choose X -resolutions X ′• → M ′ and X ′′• → M ′′. By a version of Horseshoe Lemma ([12, Lemma 8.2.1]), we have a commutative diagram 0 0 (1 0) / X ′• X • (0 1) X ′′• / M ′ / M / M ′′ / 0 / 0 where the complex X • satisfies that for each n ∈ Z, X n = X ′n ⊕ X ′′n and that the middle column is an X -resolution. Then for each Y ∈ Y we have a commutative diagram of abelian groups 0 0 / HomA(M ′′, Y ) / HomA(M, Y ) / HomA(M ′, Y ) / HomA(X ′′•, Y ) (0 1) / HomA(X •, Y ) (1 0) / HomA(X ′•, Y ) 0 / 0 By (BP1) each column is an acyclic complex. The bottom row is a sequence of complexes, every degree of which is a split short exact sequence. We infer that the upper row is exact by the homology exact sequence. Therefore we deduce that 0 → M ′ → M → M ′′ → 0 is left Y-acyclic, as required. (cid:3) Recall that a contravariantly finite subcategory X of A is admissible provided that each right X -approximation is epic. It is equivalent to that any right X -acyclic complex is indeed acyclic. Similar remarks hold for coadmissible covariantly finite subcategories. Then we observe the following direct consequence of Proposition 2.2. Corollary 2.3. Let (X , Y) be a balanced pair. Then X is admissible if and only if Y is coadmissible. (cid:3) Recall that in the case of the corollary above, we say that the balanced pair (X , Y) is admissible. The following result on resolution dimensions is well known. However it seems that there are no precise references. We include here a proof. Lemma 2.4. Let X ⊆ A be a contravariantly finite subcategory. Let M ∈ A and let n0 ≥ 0. Assume that X is admissible. The following statements are equivalent: (1) X -res.dim M ≤ n0; (2) for each X -resolution X • → M and each object N , H n(HomA(X •, N )) = 0 for all n > n0; / / / / / / / / / / / / / / / / / / / 6 XIAO-WU CHEN (3) for each X -resolution X • → M with X • = (X −n, d−n X )n≥0, the object Kerd−n0+1 X belongs to X . Proof. For "(1) ⇒ (2)", choose an X -resolution X • for n > n0. Then X • Hom complexes HomA(X • H n(HomA(X • 0 = 0 0 and X • are homotopically equivalent, thus so are the 0 , N ) and HomA(X •, N ). Hence for each n we have 0 , N )) ≃ H n(HomA(X •, N )). Then (2) follows directly. 0 → M such that X −n X X X factors through d−n0−1 For "(2) ⇒ (3)", note that H n0+1(HomA(X •, Kerd−n0 X )) = 0 implies that the naturally induced morphism ¯d : X −n0−1 → Kerd−n0 , say X such that ¯d = π ◦ d−n0−1 there is a morphism π : X −n0 → Kerd−n0+1 . Observe = inc◦ ¯d, where "inc" is the inclusion morphism of Kerd−n0 that d−n0−1 X into X −n0. Then we have ¯d = (π ◦ inc) ◦ ¯d. Note that ¯d is a right X -approximation and that X is admissible. Hence ¯d is epic, and then π ◦ inc = IdKerd . Consider the left exact sequence 0 → Kerd−n0 → 0. Since the right side morphism is a right X -approximation, it is necessarily epic and then the sequence is exact. Because the morphism "inc" admits a retraction, the sequence splits and then Kerd−n0+1 is a direct summand of X −n0. Recall that the subcategory X ⊆ A is closed under taking direct summands. Therefore the object Kerd−n0+1 belongs to X . −n0 X inc→ X −n0 → Kerd−n0+1 X X X X X X → (cid:3) The implication "(3) ⇒ (1)" is easy, since the subcomplex 0 → Kerd−n0+1 X −n0+1 → · · · → X 0 → M → 0 is the required X -resolution. We have the following consequence. Corollary 2.5. Let (X , Y) be an admissible balanced pair in an abelian category A. Then we have X -res.dim A = Y-cores.dim A. Proof. We apply Lemma 2.1. Then the result follows directly from Lemma 2.4(2) and its dual for coadmissible covariantly finite subcategories. (cid:3) In what follows we introduce the notion of cotorsion triple, which gives rise naturally to a balanced pair. The notion was suggested by Edgar Enochs in a private communication. A(X, M ) = 0 for all X ∈ X } and ⊥X = {M ∈ A Ext1 Let A be an abelian category. For a subcategory X of A, set X ⊥ = {M ∈ A Ext1 A(M, X) = 0 for all X ∈ X }. A pair (X , Y) of subcategories in A is called a cotorsion pair provided that X = ⊥Y and Y = X ⊥. The cotorsion pair (X , Y) is said to be complete provided that for each M ∈ A there exist short exact sequences 0 → Y → X → M → 0 and 0 → M → Y ′ → X ′ → 0 with X, X ′ ∈ X and Y, Y ′ ∈ Y ([12, Chapter 7] and [19, 9]). Assume that A has enough projective and injective objects. Recall that a sub- category X of A is resolving provided that it contains all projective objects such that for any short exact sequence 0 → X ′ → X → X ′′ → 0 with X ′′ ∈ X in A, X ∈ X if and only if X ′ ∈ X . Dually one has the notion of coresolving subcategory. A cotorsion pair (X , Y) is said to be hereditary provided that X is resolving. It is not hard to see that this is equivalent to that the subcategory Y is coresolving ([13, Theorem 3.4]). HOMOTOPY EQUIVALENCES INDUCED BY BALANCED PAIRS 7 A triple (X , Z, Y) of subcategories in A is called a cotorsion triple provided that both (X , Z) and (Z, Y) are cotorsion pairs; it is complete (resp. hereditary) provided that both of the two cotorsion pairs are complete (resp. hereditary). The following result is essentially due to Enochs, Jenda, Torrecillas and Xu ([13, Theorem 4.1]). The argument resembles the one in [12, Theorem 12.1.4]. For completeness we include a proof. Proposition 2.6. Let A be an abelian category with enough projective and injective objects. Assume that (X , Z, Y) is cotorsion triple which is complete and hereditary. Then the pair (X , Y) is an admissible balanced pair. Proof. Let M ∈ A. Since (X , Z) is complete, we have a short exact sequence f → M → 0 with X ∈ X and Z ∈ Z. Since Z ∈ X ⊥, the sequence ξ : 0 → Z → X ξ is a special right X -approximation ([12, Definition 7.1.6]). In particular, we have that f is a right X -approximation, and then X is contravariantly finite. Dually we have that Y is covariantly finite. Then we get (BP0). Observe that the subcategory X contains all the projective objects. Then right X -approximations are epic, that is, the contravariantly finite subcategory X ⊆ A is admissible. Dually the subcategory Y is coadmissible. To show (BP1), let X • ε→ M be an X -resolution of an object M . Since X ε→ M is acyclic. Since (X , Z) is complete, we is admissible, the sequence X • may assume that all the cocycles of X • (but M ) lie in Z. Then (BP1) follows immediately from the following fact: for a short exact sequence γ : 0 → Z 0 → X 0 → M → 0 with Z 0 ∈ Z and X 0 ∈ X and an object Y ∈ Y, the functor HomA(−, Y ) keeps γ exact. The fact is equivalent to that the induced map HomA(X 0, Y ) → HomA(Z 0, Y ) is surjective. To see this, take a short exact sequence 0 → Z0 → I → Z ′ → 0 with I injective. Since (X , Z) is hereditary, Z is coresolving. Note that Z0, I ∈ Z, and then we have Z ′ ∈ Z. Observe the following commutative diagram 0 0 / Z 0 / X 0 M 0 / Z 0 / I / Z ′ / 0. A(Z ′, Y ) = 0, we deduce that the induced map HomA(I, Y ) → HomA(Z 0, Y ) Since Ext1 is surjective. Note that from the commutative diagram above we infer that the map HomA(I, Y ) → HomA(Z 0, Y ) factors as HomA(I, Y ) → HomA(X 0, Y ) → HomA(Z 0, Y ). Therefore the map HomA(X 0, Y ) → HomA(Z 0, Y ) is surjective. Dually we have (BP2). (cid:3) 3. Relative Derived Category In this section we make preparations to prove Theorem A. We introduce the notion of relative derived category and study its relation with homotopy categories. Let A be an abelian category, and let X ⊆ A be a contravariantly finite sub- category. Recall that the homotopy category K(A) has a canonical triangulated structure. Denoted by X -ac the full triangulated subcategory of K(A) consist- ing of right X -acyclic complexes. A chain map f • : M • → N • is said to be a right X -quasi-isomorphism provided that for each X ∈ X , the resulting chain map / / / / / / / / / / 8 XIAO-WU CHEN HomA(X, f •) : HomA(X, M •) → HomA(X, N •) is a quasi-isomorphism. Denote by ΣX the class of all the right X -quasi-isomorphisms in K(A). Note that the class ΣX is a saturated multiplicative system corresponding to the subcategory X -ac in the sense that a chain map f • : M • → N • is a right X -quasi-isomorphism if and only if its mapping cone Cone(f •) is right X -acyclic (for the correspondence, consult [15, Chapter V, Theorem 1.10.2]). Definition 3.1. The relative derived category DX (A) of A with respect to X is defined to be the Verdier quotient ([30] and [27, Chapter 2]) of K(A) modulo the subcategory X -ac, that is, DX (A) := K(A)/X -ac = Σ−1 X K(A). We denote by Q : K(A) → DX (A) the quotient functor. Remark 3.2. Denote by EX the class of short exact sequences in A on which the functors HomA(X, −) are exact for all X ∈ X . Then (A, EX ) is an exact category in the sense of Quillen ([23, Appendix A]). Observe that if the subcategory X is admissible, then the relative derived category DX (A) coincides with Neeman's derived category of the exact category (A, EX ) ([26, Construction 1.5]; also see [24, Sections 11 and 12]). Note that Buan considers relative derived categories in quite a different setup ([5, Section 2]), and Gorenstein derived categories in the sense of Gao and Zhang are examples of relative derived categories ([14]). (cid:3) In what follows we will study for a complex M • its X -resolution, that is, a right X -quasi-isomorphism X • → M • with each X i lying in X . From now on, X ⊆ A is a contravariantly finite subcategory such that X -res.dim A < ∞. Let M • = M )n∈Z be a complex in A. For each M n, take a finite X -resolution X n,• εn (M n, dn → M n, where X n,• = (X n,−i, dn,−i )i≥0. By a version of Comparison Theorem, there M : M n → M n+1. exists a chain map dn,• Set di,j : X n,• → X n+1,• extending the map dn for all i, j ∈ Z. 1 = (−1)jdi,j 0 v v The following argument resembles the one in [29, Proposition 2.6], while it differs from the proof of [6, Chapter XVII, Proposition 1.2]. It seems that the argument in [6] does not extend to our situation. Consider the bigraded objects X •,•. Note that X i,j 6= 0 only if −(X -res.dim A) ≤ j ≤ 0. The bigraded objects X •,• are endowed with two endomorphisms d0 and d1 of degree (0, 1) and (1, 0), respectively, subject to the relations d0 ◦ d0 = 0 and d0 ◦ d1 + d1 ◦ d0 = 0. Unfortunately, d1 ◦ d1 is not necessarily zero. ◦ dn,• M ◦ dn Consider the chain map dn+1,• : X n,• → X n+2,•, which extends the map 0 = dn+1 M : M n → M n+2. By a version of Comparison Theorem, we infer that the chain map dn+1,• is homotopic to zero. Thus the homotopy maps give rise to an endomorphism d2 on X •,• of degree (2, −1), such that d0◦d2+d1◦d1+d2◦d0 = 0. It is a pleasant exercise to check that d1 ◦ d2 + d2 ◦ d1 commutes with d0, in other words, ◦ dn,• 1 1 1 1 dn+2,•−1 1 ◦ dn,• 2 + dn+1,• 2 ◦ dn,• 1 : X n,• −→ X n+3,•(−1) is a chain map, where (−1) denotes the inverse of the degree-shift functor on com- plexes (see Section 1 for the notation). We need the following easy lemma whose proof is routine. HOMOTOPY EQUIVALENCES INDUCED BY BALANCED PAIRS 9 Lemma 3.3. Let M1, M2 be two objects with X -resolutions X • M2. Let r ≥ 1. Then any chain map f • : X • 2 → 2 (−r) is homotopic to zero. (cid:3) 1 → M1 and X • 1 → X • By the lemma above we deduce that the chain map dn+2,•−1 ◦ dn,• is homotopic to zero. Note that the homotopy maps give rise to an endomorphism d3 of degree (3, −2) such that 2 + dn+1,• ◦ dn,• 1 2 1 d0 ◦ d3 + d1 ◦ d2 + d2 ◦ d1 + d3 ◦ d0 = 0. an endomorphism dl on X •,• of degree (l, −l + 1) such that Pn Iterating this process of finding homotopy maps, we construct for each l ≥ 0, l=0 dl ◦ dn−l = 0 (consult the proof of [29, Proposition 2.6]). We will refer to the bigraded objects X •,• together with such endomorphisms dl as a quasi-bicomplex in A. The "total complex" T • = tot(X •,•) of the quasi-bicomplex X •,• is defined as follows: T n := Li+j=n X i,j (note that this is a finite coproduct), and the differ- T : T n → T n+1 is defined to be Pl≥0 dl (again this is a finite coproduct), . Then we infer from T = 0. There is a natural chain map ε• : T • → M • such that ential dn that is, the restriction of dn above that dn+1 its restriction on X n,0 is εn for each n, and zero elsewhere. T on X i,j is given by Pl≥0 di,j ◦ dn T l We have the following key observation. Proposition 3.4. The chain map ε• : T • → M • is a right X -quasi-isomorphism; moreover, it is a right C(X )-approximation of M • in the category C(A) of com- plexes. Proof. First we introduce a new quasi-bicomplex (C •,•, dl) as follows: C i,j = X i,j, j ≤ 0 and C i,1 = M i, and zero elsewhere; the endomorphisms dl on C i,j are the same as the ones on X •,• for j ≥ 1 or j = 0 and l ≥ 1; di,0 0 = εi, and dl vanishes on C i,1 for all l 6= 1, and di,1 M . One checks that C •,• is a quasi- bicomplex; moreover, it is easy to see that the "total complex" tot(C •,•) of C •,• is the mapping cone of the chain map ε• : T • → M • shifted by minus one. Then for the first statement, it suffices to show that the complex tot(C •,•) is right X -acyclic. Assume that X ∈ X . Consider the complex K • = HomA(X, tot(C •,•)) of abelian groups. Observe that the complex K • is the "total complex" of the quasi-bicomplex HomA(X, C •,•) of abelian groups. As in the case of bicomplexes, we have a de- scending filtration of subcomplexes {F pK •, p ∈ Z} of the "total complex" K • given 1 = −di by F pK n := Li≥p, i+j=n HomA(X, C i,j). This filtration gives rise to a convergent spectral sequence Ep,q → M n is an X -resolution, the complex HomA(X, C n,•) is acyclic for each n. Therefore the spectral sequence van- ishes on E2 (and even on E1), and then we deduce that H n(K •) = 0 for each n. We are done with the first statement. H p+q(K •). Since X n,• εn 2 =⇒ p For the second statement, let f • : X • → M • be a chain map with X • = X )n∈Z ∈ C(X ). Note that the morphism εn : X n,0 → M n is a right X - 0 : X n → X n,0 such X : X n → X n+1,0. Note (X n, dn approximation, hence the map f n factors through it. Take f n that εn ◦ f n 0 = f n. Consider the map dn,0 0 − f n+1 ◦ f n ◦ dn 1 0 10 that XIAO-WU CHEN 1 εn+1 ◦ (dn,0 = dn = dn ◦ f n M ◦ εn ◦ f n M ◦ f n − f n+1 ◦ dn 0 − f n+1 X ) 0 − εn+1 ◦ f n+1 X = 0. ◦ dn 0 0 ◦ dn X Therefore the map dn,0 X n+1,−1 dn+1,−1 X factors through Kerεn+1. Note that 0 → Kerεn+1 is a right X -approximation. Then we have a factorization 0 − f n+1 ◦ f n ◦ dn 1 0 (3.1) dn,0 1 ◦ f n 0 − f n+1 0 ◦ dn X = −dn+1,−1 0 ◦ f n 1 , where f n 1 : X n → X n+1,−1 is some morphism. Rewrite equation (3.1) as d0 ◦ f1 + d1 ◦ f0 = f0 ◦ dX . We will refer to (3.1) as the defining identity for f • 1 . We claim that there exist morphisms (not chain maps) f • i=0 di ◦ fl−i = fl−1 ◦ dX for all l ≥ 1. Assume that the required f1, . . . , fl are chosen. The following computation is similar to the one in the proof of [29, Propositions 2.6 and 2.7]. l : X • → X •+l,−l such that Pl di ◦ fl+1−i − fl ◦ dX ) (d0 ◦ di) ◦ fl+1−i − d0 ◦ fl ◦ dX d0 ◦ ( X1≤i≤l+1 = X1≤i≤l+1 = X1≤i≤l+1 = − X1≤j≤l = − X1≤j≤l = − X0≤j≤l (− X1≤j≤i dj ◦ ( X0≤i≤l−j (dj ◦ fl−j) ◦ dX dj ◦ di−j) ◦ fl+1−i − d0 ◦ fl ◦ dX di ◦ fl+1−j−i) − dl+1 ◦ d0 ◦ f0 − d0 ◦ fl ◦ dX dj ◦ (fl−j ◦ dX ) − d0 ◦ fl ◦ dX = −fl−1 ◦ dX ◦ dX = 0. Note that the second equality uses the identities on the endomorphisms dl's; the fourth one uses the fact d0 ◦ f0 = 0 (note that X n,1 = 0) and the defining identity for f • l . We infer that the morphism l+1−j; the sixth uses the defining identity for f • X1≤i≤l+1 dn+l+1−i,−l−1+i i ◦ f n l+1−i − f n+1 l ◦ dn X : X n −→ X n+1+l,−l factors through Kerdn+1+l,−l and then factors through X n+1+l,−l−1, since the induced map X n+1+l,−l−1 → Kerdn+1+l,−l is a right X -approximation. Take l+1 : X n → X n+1+l,−l−1 to fulfil the factorization. This completes the construc- f n tion of f • l+1's and by induction we construct all the f • l 's. 0 0 Consider the map Pi≥0 f n that this defines a chain map from X • to T •; moreover, this chain map makes f • factor through ε•. This proves that ε• is a right C(X )-approximation of M •. (cid:3) i : X n → T n = Li≥0 X n+i,−i. One checks readily HOMOTOPY EQUIVALENCES INDUCED BY BALANCED PAIRS 11 The following result is a relative version of a well-known result ([20, p.439, Propo- sition 2.12]). Proposition 3.5. Let X ⊆ A be a contravariantly finite subcategory. Assume that X is admissible and X -res.dim A < ∞. Then the natural composite functor K(X ) inc→ K(A) Q → DX (A) is a triangle-equivalence. Proof. The composite functor is clearly a triangle functor. It suffices to show it is an equivalence of categories (see [16, p.4]). By Proposition 3.4 for each complex M •, there is an X -resolution ε• : X • → M •, that is, it is a right X -quasi-isomorphism. Note that ε• becomes an isomorphism in the relative derived category DX (A), in particular, Q(M •) ≃ Q ◦ inc(X •). Therefore the composite functor is dense. We claim that for each X • 0 ∈ K(X ) and each right X -acyclic complex M • ∈ 0 , M •) = 0. This will complete the proof by the following K(A), HomK(A)(X • general fact: for a triangulated category T and a triangulated subcategory N ⊆ T , set ⊥N = {X ∈ T HomT (X, N ) = 0 for all N ∈ N } to be the left perpendicular Q subcategory, then the composite functor ⊥N inc→ T → T /N is fully faithful ([30, 5-3 Proposition]). The claim says precisely that K(X ) ⊆ ⊥(X -ac). By the recalled general fact the composite functor is fully faithful. Note that the functor is dense by above, thus it is an equivalence of categories. To see the claim, take a chain map f • : X • 0 → M •. By Proposition 3.4 we may take an X -resolution ε• : X • → M • which is a right C(X )-approximation. Hence f • factors through ε•. In fact, we will show that X • is null-homotopic, and then ε• and consequently f • is homotopic to zero. Set X -res.dim A = n0. Note that X • = (X n, dn X )n∈Z is right X -acyclic. Consider the canonical factorization X n ∂n X . Recall that the complex X • is null- homotopic if and only if the morphisms ∂n are split epic. Note that the subcomplex · · · → X n−1 → X n ∂n X → 0 can be viewed as a shifted X -resolution. By Lemma 2.4(3) we have that Kerdn−n0+1 belongs to X . Thus all the cocycles Kerdn X of X • lie in X . Since X • is right X -acyclic, the morphism ∂n : X n → Kerdn+1 X is a right X -approximation. In particular, the identity map of Kerdn+1 X factors through ∂n, that is, the morphism ∂n is split epic. We are done. (cid:3) inc→ X n+1 of the differential dn → Kerdn+1 X → Kerdn+1 X Remark 3.6. The composite functor in Proposition 3.5 factors as K(X ) inc−→ ⊥(X -ac) inc−→ K(A) Q −→ DX (A). By the recalled general fact, the composite of the latter two functors is fully faithful. Hence the equivalence in Proposition 3.5 will force the equality K(X ) = ⊥(X -ac), and it also implies that the subcategory X -ac ⊆ K(A) is left admissible and K(X ) ⊆ K(A) is right admissible (= Bousfield ) ([4, Definition 1.2]; compare [27, Chapter 9]). Hence the inclusion functor inc : K(X ) → K(A) has a right adjoint i! : K(A) → K(X ). The functor i! vanishes on X -ac and then factors through the quotient functor Q : K(A) → DX (A) canonically; by abuse of notation we denote the resulting functor by i! : DX (A) → K(X ). This functor is a quasi-inverse of the composite functor in the Proposition 3.5. For later use, let us recall the construction of the quasi-inverse functor i!: for each complex M •, choose a complex i!(M •) ∈ K(X ) and fix a right X -quasi- isomorphism ε• : i!(M •) → M •; for a chain map f • : M • → M ′•, there is a unique, 12 XIAO-WU CHEN up to homotopy, chain map i!(f •) : i!(M •) → i!(M ′•) making the following diagram commute, again up to homotopy i!(M •) ε• M • i!(f •) f • i!(M ′•) ε′• / M ′• One could deduce this by Proposition 3.5, or alternatively, by noting that the cohomological functor HomK(A)(i!(M •), −) vanishes on the mapping cone of ε′•, and then one gets the natural isomorphism HomK(A)(i!(M •), i!(M ′•)) ≃ HomK(A)(i!(M •), M ′•). In this way one defines the functor i! on homotopy categories, which induces the pursued functor i! : DX (A) → K(X ). (cid:3) In this section we prove Theorem A. 4. Proof of Theorem A Let A be an abelian category. Let (X , Y) be an admissible balanced pair in A of finite dimension. For each complex X • ∈ K(X ), choose a complex F (X •) ∈ K(Y) and fix a left Y-quasi-isomorphism X • θ• X→ F (X •) (see Proposition 3.4); for each chain map f • : X • → X ′• there is a unique, up to homotopy, chain map F (f •) : F (X •) → F (X ′•) such that F (f •) ◦ θ• X ′ ◦ f •, again up to homotopy; see Proposition 3.5. This defines a triangle functor F : K(X ) → K(Y). X = θ• Theorem 4.1. Let A be an abelian category. Let (X , Y) be a balanced pair of subcategories in A which is admissible and of finite dimension. Then the above defined triangle functor F : K(X ) ≃ K(Y) is an equivalence. Proof. Note that by Proposition 2.2, the full subcategories X -ac = Y-ac. Here Y-ac means the full subcategory of K(A) consisting of left Y-acyclic complexes. Then we have DX (A) = DY (A). Applying the dual of Proposition 3.5 to Y, we get a ∼−→ DY (A). Composing a quasi-inverse of this natural triangle-equivalence K(Y) equivalence with the equivalence in Proposition 3.5, we get a triangle-equivalence F ′ : K(X ) ∼−→ K(Y). Now observe that the triangle-equivalence F ′ coincides with the functor F just defined above. This follows from the construction in Remark 3.6, while here we need to dualize the argument to construct a quasi-inverse functor of the equivalence K(Y) (cid:3) ∼−→ DY (A). 5. Proof of Theorem B In this section we apply Theorem 4.1 to obtain Theorem B. We will make use of a characterization theorem of left-Gorenstein rings by Beligiannis ([3]). Let R be a ring with identity. Denote by R-Mod the category of left R-modules and by L the full subcategory consisting of modules with finite projective and injective dimension. Following [3] a ring R is called left-Gorenstein provided that any module in R-Mod has finite projective dimension if and only if it has finite injective dimension. In this case, by [3, Theorem 6.9(δ)] there is a uniform upper / / / HOMOTOPY EQUIVALENCES INDUCED BY BALANCED PAIRS 13 bound d such that each module in L has projective and injective dimensions less or equal to d. We will denote by G.dim R the minimal bound. We collect in the following lemma some crucial properties of left-Gorenstein rings. Lemma 5.1. Let R be a left-Gorenstein ring. Then we have the following: (1) the triple (R-GProj, L, R-GInj) is a complete and hereditary cotorsion triple; (2) R-GProj-res.dim R-Mod = G.dim R = R-GInj-cores.dim R-Mod. Proof. We infer (1) by [3, Theorem 6.9(4) and (5)] (which is presented in quite a different terminology). One might deduce (1) also from [9, Theorems 2.25 and 2.26]. Just note that for a left-Gorenstein ring R, R-Mod is a Gorenstein category in the sense of Enochs, Estrada and Garc´ıa Roza ([9, Definition 2.18]). The statement (2) follows from [3, Theorem 6.9(α)]. (cid:3) We are in the position to prove Theorem B. Theorem 5.2. Let R be a left-Gorenstein ring. Then we have a triangle-equivalence K(R-GProj) ≃ K(R-GInj), which restricts to a triangle-equivalence K(R-Proj) ≃ K(R-Inj). Proof. Combining Lemma 5.1(1) and Proposition 2.6 together, we infer that the pair (R-GProj, R-GInj) is an admissible balanced pair in R-Mod. It is of fi- nite dimension by Lemma 5.1(2). By Theorem 4.1 we get a triangle-equivalence ∼−→ K(R-GInj). Denote by F −1 its quasi-inverse. Remind that F : K(R-GProj) the construction of the functors F and F −1 is described before Theorem 4.1 (and its dual). Recall that for a Gorenstein projective module G we have Exti R(G, P ) = 0 for i ≥ 1 and all projective modules P ([7, Lemma 2.2]). Then by the dimension-shift technique we have Exti R(G, M ) = 0 for i ≥ 1 and all modules M of finite projective dimension. Set d = G.dim R. For a projective module P consider its injective resolution 0 → P → I 0 → I 1 → · · · → I d → 0. Write it as P → I •. It is obvious that for a Gorenstein projective module G, HomR(G, I •) has no cohomology in non-zero degrees, for it computes Ext∗ R(G, P ). Thus the injective resolution is right R-GProj-acyclic, and by Proposition 2.2, it is also left R-GInj-acyclic. In particular, it is an R-GInj-coresolution. Take a complex P • in K(R-Proj). Consider the construction of R-GInj-coresolution as in the dual of Proposition 3.4. We find that the R-GInj-coresolution of P • is a complex consisting of injective modules. That is, the essential image of K(R-Proj) under F lies in K(R-Inj). Dually the essential image of K(R-Inj) under F −1 lies in K(R-Proj). Consequently, we have a restricted equivalence K(R-Proj) ≃ K(R-Inj). (cid:3) 6. Comparison of Equivalences In the last section we compare the equivalences in Theorem 5.2 with Iyengar- Krause's equivalence ([21]) in the case of commutative Gorenstein rings. In this case, it turns out that up to a natural isomorphism the first equivalence in Theorem 5.2 extends Iyengar-Krause's equivalence. Let R be a commutative Gorenstein ring of dimension d. Take its injective resolution 0 → R ε→ I 0 → I 1 → · · · → I d → 0. Write it as R ε→ I •. Then the 14 XIAO-WU CHEN complex I • is a dualizing complex ; for details, see [17, Chapter V, §2], [21, Section 3] and [7, Appendix A]. Note that the ring R is noetherian, and then the class of injective modules is closed under coproducts. One infers that for a projective module P and an injective module I the tensor module P ⊗RI is injective. Then we have a well-defined triangle functor − ⊗R I • : K(R-Proj) −→ K(R-Inj). By [21, Theorem 4.2] this is a triangle-equivalence, which we will call Iyengar- Krause's equivalence. We note the following fact. Lemma 6.1. ([8, Corollary 5.7]) Let R be a commutative Gorenstein ring. Then for a Gorenstein projective module G and an injective module I, the tensor module G ⊗R I is Gorenstein injective. (cid:3) By the lemma above we can extend Iyengar-Krause's equivalence to a triangle functor − ⊗R I • : K(R-GProj) −→ K(R-GInj). Recall the construction of the equivalence F : K(R-GProj) ∼−→ K(R-GInj) in Theorem 5.2. For each G• ∈ K(R-GProj) choose an R-GInj-coresolution θ• G : G• → F (G•); for each f • : G• → G′• there is a unique, up to homotopy, chain map F (f •) : F (G•) → F (G′•) such that F (f •) ◦ θ• G′ ◦ f •. This defines the triangle functor F . Consult the construction before Theorem 4.1. G = θ• Note that the mapping cone Cone(θ• G) of θ• G is left R-GInj-acyclic. By the proof of Proposition 3.5 we have HomK(R-Mod)(Cone(θ• G), G• ⊗R I •[n]) = 0, for all n ∈ Z. By applying the cohomological functor HomK(R-Mod)(−, G• ⊗R I •) to the distin- guished triangle associated to θ• G, we deduce a natural isomorphism of abelian groups HomK(R-Mod)(G•, G• ⊗R I •) ≃ HomK(R-Mod)(F (G•), G• ⊗R I •). Note that there is a natural chain map IdG• ⊗R ε : G• → G• ⊗R I •. By the above isomorphism, there exists a unique, up to homotopy, chain map ηG• : F (G•) → G• ⊗ I • such that ηG• ◦ θ• G = IdG• ⊗R ε. It is routine to check that this defines a natural transformation of triangle functors η : F −→ − ⊗R I •. The following result states that in the case of commutative Gorenstein rings the first equivalence in Theorem 5.2 extends Iyengar-Krause's equivalence, up to a natural isomorphism. Proposition 6.2. Use the notation as above. Then for each complex P • ∈ K(R-Proj), the chain map ηP • is an isomorphism in K(R-GInj). Proof. First note that ηG• is an isomorphism if and only if IdG• ⊗R ε : G• → G• ⊗ I • is a left R-GInj-quasi-isomorphism. The "only if" part is clear since θ• G is a coresolution. For the "if" part, assume that IdG• ⊗R ε : G• → G• ⊗ I • is left R-GInj-quasi-isomorphism. Then by a similar argument as above we get a unique HOMOTOPY EQUIVALENCES INDUCED BY BALANCED PAIRS 15 chain map γG• : G• ⊗ I • → F (G•) such that θ• two "uniqueness" imply that η• G and γ• G are inverse to each other. G = γ• G ◦ (IdG• ⊗R ε). Then these Note that IdG• ⊗R ε : G• → G• ⊗R I • is a left R-GInj-quasi-isomorphism if and only if its mapping cone is left R-GInj-acyclic, or equivalently, by Proposition 2.2, its mapping cone is right R-GProj-acyclic. However the mapping cone is given by the tensor complex G• ⊗R Y •; here we denote by Y • the acyclic complex 0 → R ε→ I 0 → I 1 → · · · → I d → 0. So it suffices to show that for each complex P • ∈ K(R-Proj) the tensor complex P • ⊗R Y • is right R-GProj-acyclic. Given any Gorenstein projective module G, we need to show that the Hom complex HomR(G, P • ⊗R Y •) is acyclic. Note that the tensor complex P • ⊗R Y • is the total complex of a bicomplex K •,• such that K i,j = P i ⊗R Y j. Therefore one sees that the complex HomR(G, P • ⊗R Y •) is the total complex of the bicomplex HomR(G, K •,•). Associated to this bicomplex, there exists a convergent spectral sequence Ep,q H p+q(HomR(G, P • ⊗R Y •)). Note that for each i, the column 2 =⇒ p complex K i,• is an injective resolution of P i. By the second paragraph in the proof of Theorem 5.2, we infer that K i,• is right R-GProj-acyclic. Hence the column complex HomR(G, K i,•) is acyclic for each i. Therefore, in the spectral sequence we see that E2 (and even E1) vanishes. Thus we get H n(HomR(G, P • ⊗R Y •)) = 0 for all n ∈ Z. We are done. (cid:3) Acknowledgements The author is indebted to the anonymous referee for numer- ous suggestions which improve the exposition very much. This paper is completed during the author's visit in the University of Paderborn with a support by Alexan- der von Humboldt Stiftung. He would like to thank Prof. Henning Krause and the faculty of Institut fuer Mathematik for their hospitality. The author would like to thank Prof. Edgar Enochs for suggesting him the notion of cotorsion triple. References [1] M. Auslander and S.O. Smalø, Preprojective modules over artin algebras, J. Algebra 66 (1980), 61 -- 122. [2] L.L. Avramov and A. Martsinkovsky, Absolute, relative, and Tate cohomology of modules of finite Gorenstein dimension, Proc. London Math. Soc. (3) 85 (2002), no.2, 393 -- 440 [3] A. Beligiannis, The homological theory of contravariantly finite subcategories: Auslander- Buchweitz contexts, Gorenstein categories and (co)stabilization, Comm. Algera 28 (2000), 4547 -- 4596. [4] A.I. Bondal and M.M. Kapranov, Representable functors, Serre functors, and reconstruc- tions, Izv. Akad. Nauk SSSR Ser. Mat. 53 (1989), 1183 -- 1205. [5] A.B. Buan, Closed subfunctors of the extension functor, J. Algebra 244 (2001), 407 -- 428. [6] E. Cartan and S. Eilenberg, Homological Algebra, Princeton Univ. Press, 1956. [7] L.W. Christensen, A. Frankild and H. Holm, On Gorenstein projective, injective and flat dimensions -- a functorial discription with applications, J. Algebra 302 (2006), 231 -- 279. [8] L.W. Christensen and H. Holm, Ascent properties of Auslander categories, Canad. J. Math. 61 (2009), no.1, 76 -- 108. [9] E.E. Enochs, S. Estrada and J.R. Garc´ıa Rozas, Gorenstein categories and Tate coho- mology on projective schemes, Math. Nach. 281 (4) (2008), 525 -- 540. [10] E.E. Enochs, S. Estrada, J.R. Garc´ıa Rozas and A. Iacob, Gorenstein quivers, Arch. Math. (Basel) 88 (2007), 199 -- 206. [11] E.E. Enochs, O.M.G. Jenda, Gorenstein injective and projective modules, Math. Z. 220 (1995), 611 -- 633. [12] E.E. Enochs, O.M.G. Jenda, Relative Homological Algebra, de Gruyter Expositions in Math. 30, Walter de Gruyter, Berlin New York, 2000. 16 XIAO-WU CHEN [13] E.E. Enochs, O.M.G. Jenda, B. Torrecillas and J.Z. Xu, Torsion theory relative to Ext, Technical Report 98-11, Department of Math., Univ. of Kentucky, (1998). Availabe at http://www.ms.uky.edu/∼math/mareport/98-10.ps. [14] N. Gao, P. Zhang, Gorenstein derived categories, submitted for publication, 2009. [15] S.I. Gelfand, Yu I. Manin, Homological Algebra, Springer-Verlag, Berlin Heidelberg, 1999. [16] D. Happel, Triangulated Categories in the Representation Theory of Finite Dimensional Algebras, London Math. Soc. Lecture Notes Ser. 119, Cambridge Univ. Press, 1988. [17] R. Hartshorne, Residue and Duality, Lecture Notes in Math. 20, Springer-Verlag, 1966. [18] H. Holm, Gorenstein homological dimensions, J. Pure Appl. Algebra 189 (2004), no.1-3, 167 -- 193. [19] M. Hovey, Cotorsion pairs, model category structures, and representation theory, Math. Z. 241(2002), 553 -- 592. [20] B. Iversen, Cohomology of Sheaves, Springer-Verlag, Berlin Heidelberg, 1986. [21] S. Iyengar and H. Krause, Acyclicity versus total acyclicity for complexes over noetherian rings, Documenta Math. 11 (2006), 207 -- 240. [22] P. Jørgensen, The homotopy category of complexes of projective modules, Adv. Math. 193 (2005), 223 -- 232. [23] B. Keller, Chain complexes and stable categories, Manuscripta Math. 67 (1990), 379 -- 417. [24] B. Keller, Derived categories and their uses, Handbook of Algebra 1 (1996), North- Holland, Amsterdam, 671 -- 701. [25] H. Krause, The stable derived category of a noetherian scheme, Compositio Math. 141 (2005), 1128 -- 1162. [26] A. Neeman, The derived category of an exact category, J. Algebra 138 (1990), 388 -- 394. [27] A. Neeman, Triangulated Categories, Annals of Math. Studies 148, Princeton Univ. Press, 2001. [28] A. Neeman, The homotopy category of flat modules, and Grothendieck duality, Invent. Math. 174 (2) (2008), 255 -- 308. [29] J. Rickard, Morita theory for derived categories, J. London Math. Soc. 39 (2) (1989), 436 -- 456. [30] J.L. Verdier, Cat´egories d´eriv´ees, etat 0, Springer Lecture Notes in Math. 569 (1977), 262 -- 311. Xiao-Wu Chen, Department of Mathematics, University of Science and Technology of China, Hefei 230026, P. R. China Current address: Institut fuer Mathematik, Universitaet Paderborn, 33095, Paderborn, Germany
0809.1395	2	0809	2010-05-17T03:40:03	Degeneracy and decomposability in abelian crossed products	[ "math.RA" ]	In this paper we study the relationship between degeneracy and decomposability in abelian crossed products. In particular we construct an indecomposable abelian crossed product division algebra of exponent $p$ and index $p^2$ for $p$ an odd prime. The algebra we construct is generic in the sense of Amitsur and Saltman and has the property that its underlying abelian crossed product is a decomposable division algebra defined by a non-degenerate matrix. This algebra gives an example of an indecomposable generic abelian crossed product which is shown to be indecomposable without using torsion in the Chow group of the corresponding Severi-Brauer variety as was needed in [Karpenko, Codimension 2 cycles on Severi-Brauer varieites (1998)] and [McKinnie, Indecomposable $p$-algebras and Galois subfields in generic abelian crossed products (2008)]. It also gives an example of a Brauer class which is in Tignol's Dec group with respect to one abelian maximal subfield, but not in the Dec group with respect to another.	math.RA	math	Degeneracy and Decomposability in Abelian Crossed Products Kelly McKinnie October 30, 2018 Abstract In this paper we continue the study of the relationship between degeneracy and decomposability in abelian crossed products ([McK08]). In particular we construct an indecomposable abelian crossed product division algebra of exponent p and index p2 for p an odd prime. The algebra we construct is generic in the sense of [AS78] and has the property that its underlying abelian crossed product is a decompos- able division algebra defined by a non-degenerate matrix. This algebra gives an example of an indecomposable generic abelian crossed product which is shown to be indecomposable without using torsion in the Chow group of the corresponding Severi-Brauer variety as was needed in [Kar98] and [McK08]. It also gives an ex- ample of a Brauer class which is in Tignol's Dec group with respect to one abelian maximal subfield, but not in the Dec group with respect to another. 1 Introduction A finite dimensional division algebra D with center a field F is said to be decom- posable if there exists an F -isomorphism D ∼= D1 ⊗F D2 with ind(Di) > 1. By the primary decomposition theorem (see e.g., [Dra83, Cor.11, pg 68]), an indecompos- able division algebra necessarily has prime power index. Recall ind(D) = √dimF D. Moreover, if an F -division algebra D has exponent equal to its index, then D is indecomposable. Therefore, it is only of interest to study indecomposable division algebras, D, of prime power index with ind(D) > exp(D). Prime power index indecomposable division algebras have been studied in many contexts since a construction of one with ind(D) < exp(D) was given by Saltman in [Sal79]. For example, in light of the present paper, we draw the readers attention to [McK08] and [Mou08] where the decomposability of generic abelian crossed product division algebras are studied. In [Mou08, Cor. 3.6] it is shown that a p-power de- gree generic abelian crossed product defined by a non-degenerate matrix (definition given below) is indecomposable. [McK08] and [Mou08] leave open the question of whether any abelian crossed product defined by a non-degenerate matrix is inde- composable. In this paper we answer this question in the negative by constructing a decomposable abelian crossed product ∆FA defined by a non-degenerate matrix (Theorem 2.12/5.2). As mentioned in more detail below, we show ∆FA is defined by a non-degenerate matrix without using the Chow group as was done in [Kar98] and [McK08]. 1 Our example has two additional properties. First, recall that in [McK08, Prop. 3.1.1] it is shown that an abelian crossed product ∆ of exponent p defined by a degenerate matrix has no nontrivial torsion in CH2(X). Here X = SB(∆), the Severi-Brauer variety of ∆. Our abelian crossed product, ∆FA, provides a coun- terexample to the converse of this statement. That is, our example satisfies CH2(X) is torsion free where X = SB(∆FA ) and ∆FA is an exponent p abelian crossed prod- uct defined by a non-degenerate matrix (see remark 2.15 for more details). As a consequence we see that the existence of non-trivial torsion in CH2(X) for X the Severi-Brauer variety of an abelian crossed product is not strictly controlled by degeneracy of the matrix of the abelian crossed product. Secondly, our abelian crossed product ∆FA with center FA gives an example of a finite dimensional division algebra with two abelian maximal subfields, N1 and N2 such that [∆FA] ∈ Dec(N1/FA) and [∆FA] /∈ Dec(N2/FA) (Corollaries 2.11 and 5.4). Here Dec( ) is the decomposition group of Tignol defined in [Tig81]. Recall for any field F and finite abelian extension N/F , Dec(N/F ) is the subgroup of the relative Brauer group Br(N/F ) generated by the subgroups Br(K/F ) where K is a cyclic subfield of N . Let G = Gal(N/F ). Since N/F is an abelian extension there is a basis {σi} such that G ∼= hσ1i × ··· × hσri. In [Tig81, Cor. 1.4] Tignol shows that Dec(N/F ) can also be described as the set of Brauer classes [A] such that the algebra A contains N , has deg(A) = [N : F ] and such that A decomposes into the tensor product of cyclic algebras A ∼= A1 ⊗F ··· ⊗F Ar where for each i, the algebra Ai is a cyclic F -algebra containing Ki as a maximal subfield. Here Ki/F is a cyclic extension with Ki = N Gi and Gi = hσ1i × ··· × N/F ". The example in this paper shows that there exist Brauer classes [D] such that D decomposes with respect to one abelian maximal subfield, but not with respect to another. dhσii × ··· × hσri. If [A] ∈ Dec(N/F ), then A is said to decompose "according to 1.1 Abelian Crossed Products and related definitions Let F be a field. An abelian crossed product is a central simple F -algebra which contains a maximal subfield that is abelian Galois over F . Let ∆ be an abelian crossed product over F (we will write this as ∆/F ) with finite abelian maximal subfield K and G = Gal(K/F ) = hσ1i × . . . × hσri. As detailed in [AS78] or [McK07], for every abelian crossed product there is a matrix u = (uij) ∈ Mr(K ∗) i=1 ∈ (K ∗)r so that ∆ is isomorphic to the following algebra. and a vector b = (bi)r (1.1) Kzi1 1 . . . zir r ∆ ∼= (K/F, G, z, u, b) = M0≤ij ≤nj r 1 Here nj = σj and multiplication in this algebra is given by the conditions zizj = uijzjzi, zni i = bi and zik = σi(k)zi for all k ∈ K. Throughout this paper we will and use multi-index notation: σm = σm1 . . . σmr for m = (m1, . . . , mr) ∈ Nr set zm = zm1 . Moreover, set um,n = zmzn(zm)−1(zn)−1 ∈ K ∗. Since the matrix u determines multiplication in the algebra, properties of the matrix u determine some properties of the abelian crossed product ∆. In [AS78] the notion of a matrix being degenerate was defined and this notion was further studied and extended in [McK07] and [Mou08]. In this paper we use the original definition given in[AS78]. That is, the matrix u is degenerate if there exist elements σm, σn ∈ G and elements a, b ∈ K ∗ so that hσm, σni is noncyclic and um,n = σm(a)a−1σn(b)b−1. The motivating consequence behind this definition is that if u is degenerate, then bzm and a−1zn commute in ∆. . . . zmr r 1 2 Recall that if ∆ is an abelian crossed product then the generic abelian crossed product associated to ∆, which we will denote by A∆, is the abelian crossed product A∆ ∼= (K(x1, . . . , xr)/F (x1, . . . , xr), G, z, u, bx). Here x1, . . . , xr are independent indeterminates and bx = {bixi}r i=1. Let p be a prime. A p-algebra is a central simple algebra over a field of characteristic p with p-power index. As the main result in [AS78], generic abelian crossed products and non-degeneracy were used to establish the existence of non-cyclic p-algebras. 1.2 Related examples As mentioned above, the results in [McK08, Theorem 2.3.1] (in the case char(F ) = p) and [Mou08, Cor. 3.6] show that a generic abelian crossed product of p-power degree defined by a non-degenerate matrix is indecomposable. In [McK08, Section 3.3] examples of such algebras with index pn and exponent p for all p 6= 2 and all n ≥ 2 are given. An example is also given in the case p = 2 and n = 3. In the p 6= 2 example the abelian crossed product ∆ is constructed by generically lowering the exponent of an abelian crossed product with exponent equal to index equal to pn. That is, let ∆′ be an abelian crossed product defined by the group G ∼= (Z/pZ)n, n ≥ 2, with index and exponent pn. Set Y = SB(∆′⊗p) and let F (Y ) be the function field of Y . Then set ∆ = ∆′ ⊗ F (Y ). ∆ has exponent p and index pn by the index reduction theorem of [SVdB92]. ∆ is shown to be defined by a non-degenerate matrix by studying the torsion in CH2(SB(∆)) ([McK08, Prop. 3.1.1]). Since ∆ is defined by a non-degenerate matrix, A∆, the associated generic abelian crossed product, is indecomposable. Because of the method of construction of ∆, by [Kar98, Corollary 5.4], the abelian crossed product ∆ is itself an indecomposable division algebra of exponent p and index pn. In this paper we construct an abelian crossed product, ∆FA, which is decomposable of index p2, exponent p (p 6= 2), and is defined by a non-degenerate , the generic abelian crossed product associated matrix. As mentioned above A∆FA to ∆FA, is therefore indecomposable. The strategy is to make an abelian crossed product decomposable in a generic way and prove that the matrix defining the resulting decomposable abelian crossed product is non-degenerate. The method in this paper does not use the results of [Kar98]. In particular, we avoid using the Chow group of Severi-Brauer varieties, hence providing a more elementary, though still technical, approach to decomposability of abelian crossed products. 1.3 Outline of paper. In section 2 we construct ∆FA, a decomposable abelian crossed product division algebra of index p2 and exponent p (Lemma 2.9 and Corollary 2.19). We study ∆FA for the rest of the paper, with our goal being to show that it is defined by a non- degenerate matrix. The difficulty in proving non-degeneracy of the matrix defining ∆FA lies in the fact that the lattice Mω used in the definition of FA is not H 1- trivial. In section 3 we alleviate this problem by constructing an H 1-trivialization, M , of the lattice Mω and analyzing its structure as a module over a group ring. In section 4 we study the form of elements in M which could possibly make the matrix degenerate. Moreover, it is noted that it suffices to prove the matrix is non-degenerate in the lattice M . Finally in section 5 we prove the main theorem, Theorem 5.2, which states that the matrix defining the abelian crossed product ∆FA is non-degenerate. Acknowledgments The author would like to thank David Saltman and Adrian Wadsworth for help with this project, especially for their help with the homological formulation of the degeneracy condition given in section 5. 3 2 The example The goal of this section is to construct, in a generic way, a decomposable abelian crossed product of index p2 and exponent p. This is done using fields generated by group lattices as in [Sal02], [Sal99, section 12] and [McK08, section 3.2]. We will recall the relevant objects as we need them in this section. For any finite abelian group H of rank r, generated by {σi}r i=1, let I[H] be the augmentation ideal. That is, I[H] is the kernel 0 → I[H] → Z[H] ǫ→ Z → 0 where ǫ(σ) = 1 for all σ ∈ H. Define A2(H) to be the H-lattice which is the kernel 0 −→ A2(H) −→ rMi=1 Z[H]di −→ I[H] −→ 0 (2.1) where di is mapped to σi − 1 for all 1 ≤ i ≤ r. We note here that A2(H) is also known as the relation module for a minimal free presentation of H (see e.g. [Gru76, Prop. 2.3]). Let [cH ] ∈ H 2(H, A2(H)) be the class of the so called "canonical" 2-cocycle. That is, if δ is the co-boundary map we set [c1] = δ([1]) ∈ H 1(H, I[H]) for [1] ∈ H 0(H, Z) ∼= Z. Then, [cH ] = δ([c1]). The module A2(H) and cocycle [cH ] are also used in [Sal99, chapter 12] and [McK08, section 3.2]. For the rest of this paper let p be a prime, p 6= 2, and for i = 1, 2, 3, 4, let Gi = hσii ∼= Cp, the multiplicative cyclic group of order p. Set G = G1×G2×G3×G4 and fix the following notation: for any numbers i, j, k between 1 and 4, let Gij = Gi×Gj, and Gijk = Gi × Gj × Gk, each considered as a subgroup of G . For any subgroup H ≤ G, define H12 = HG34/G34, H3 = HG124/G124 and H4 = HG123/G123 (2.2) so that H12 ≤ G12, H3 ≤ G3 and H4 ≤ G4. Furthermore, to ease the notation slightly, set [c12] = [cG12 ], [c3] = [cG3] and [c4] = [cG4]. These are the cocycle classes we use to build our algebra. Let F be a field. Set L12 = F (A2(G12)) = q(F [A2(G12)]), the field of fractions of the commutative group ring F [A2(G12)] which is a domain. The trivial G12- action on F and the natural G12-action on A2(G12) extend to a G12-action on L12. Since the G12-action on A2(G12) is faithful, L12/LG12 is a G12-Galois extension of 12 fields. For any group H, H-lattice Λ and field K, let e : Λ → K[Λ] denote the canon- ical injection taking the additive group Λ to the multiplicative subgroup of K[Λ] consisting of monomials with coefficient 1. By [Sal99, 12.4(a)], the associated map on cohomology H 2(H, Λ) → H 2(H, K(Λ)∗) is an injection and as such we will not distinguish between cocycle classes in H 2(H, Λ) and their image in H 2(H, K(Λ)∗). For i = 1, 2, 3, 4 set Ni = 1 + σi + . . . + σp−1 i and for i, j ∈ {1, 2}, set bi = Nidi ∈ A2(G12) uij = (σi − 1)dj − (σj − 1)di ∈ A2(G12). 12)2. The matrix Define e(u) = (e(uij)) ∈ M2(L∗ e(u) and the vector e(b) satisfy the conditions in [AS78, Theorem 1.3], and therefore 12) and e(b) = (e(b1), e(b2)) ∈ (L∗ ∆12 = (L12/LG12 12 , zσ, e(u), e(b)) (2.3) is an abelian crossed product. Furthermore, as noted in [McK08, Lemma 3.2.3], there is an isomorphism ∆12 ∼= (L12/LG12 Now we use the groups G3 and G4 to construct two generic degree p cyclic algebras, ∆3 and ∆4. For j = 3, 4, there is an isomorphism A2(Gj) ∼= NjZ. Set Kj = A2(Gj) ⊕ Z[Gj]. Set Lj = F (Kj) where again we take Gj to act trivially on 12 , G12, c12). 4 ∆j =(cid:16)Lj/LGj j , Gj, cj(cid:17) (2.4) F . Since Gj acts faithfully on Kj, Lj/LGj For j = 3, 4 set, j is a cyclic Galois extension of degree p. where cj is a 2-cocycle in the canonical class [cj] ∈ H 2(Gj , L∗ j ). One can show that as a cyclic algebra ∆j = (F (Z[Gj ])(xj )/F (Z[Gj])Gj (xj), xj ) where xj stands for the element Nj ∈ Kj (see the proof of Lemma 2.17). Set Q = A(G12) ⊕ K3 ⊕ K4. Q is a Z[G]-lattice with G-action given as follows. G12 acts in the natural way on A(G12) and trivially on K3 and K4. G3 (resp. G4) acts trivially on A(G12) and K4 (resp. A(G12) and K3) and acts in the natural way on K3 (resp. K4). Set L = F (Q). The action of G on Q extends to L and since G acts faithfully on Q, L/LG is a G-Galois extension. Using the three inclusions LG12 12 , LG3 4 ⊂ LG we have the following three field diagrams: L 3 , LG4 L L LG G12 {{{{{{{{ BBBBBBBB LG34 DDDDDDDD G12 L12 LG12 12 LG124 LG G3 zzzzzzzz CCCCCCCC LG3 3 CCCCCCCC G3 L3 LG G4 zzzzzzzz CCCCCCCC LG4 4 LG123 CCCCCCCC G4 L4 We finally define A to be the central simple LG-algebra A = (∆12 ⊗L G12 12 LG) ⊗LG(cid:16)(∆3 ⊗L G3 3 LG) ⊗LG (∆4 ⊗L G4 4 , (2.5) LG)(cid:17)◦ where ◦ denotes the opposite algebra. Set [ω] = [c12] − [c3] − [c4] ∈ H 2(G, L∗), where [c12] ∈ H 2(G12, L∗ 4) are extended to G by inflation. Lemma 2.6. A ∼= (L/LG, ω), where ω is a 2-cocycle in the class of [ω] ∈ H 2(G, L∗). 3) and [c4] ∈ H 2(G4, L∗ 12), [c3] ∈ H 2(G3, L∗ Proof. By [Rei75, (29.13) & (29.16)] ∆12 ⊗ LG is similar to (L/LG, c′ is a 2-cocycle in the image of the class [c12] under inf : H 2(G12, L∗ Similarly, ∆j ∼ (L/LG, c′ under inf : H 2(Gj, L∗ degree of A equals the degree of (L/LG, ω), this is an isomorphism. 12), where c′ 12 12) → H 2(G, L∗). j is a 2-cocycle in the image of the class [cj] j ) → H 2(G, L∗). Consequently, A ∼ (L/LG, ω). Since the j), where c′ Let FA be the function field of SB(A), the Severi-Brauer variety of A. Since A is a G-crossed product there is an explicit description of FA as follows. By [Sal02, Theorem 0.5], FA ∼= F (Q)ω(I[G])G = Lω(I[G])G. Equivalently, FA can be described by the following construction of the G-lattice Mω. As an abelian group there is an isomorphism Mω ∼= Q ⊕ I[G] and the G-action on Mω is defined by g(x, 0) = (g · x, 0) for x ∈ Q and g(0, g′ − 1) = (ω(g, g′), g(g′ − 1)) for g′ − 1 ∈ I[G]. (2.7) The field FA satisfies the isomorphism FA ∼= F (Mω)G. equal, we have the isomorphism, Since FA is a splitting field for A and the dimensions on both sides of (2.8) are ∆12 ⊗L G12 12 FA ∼= (∆3 ⊗L G3 3 FA) ⊗FA (∆4 ⊗L G4 4 FA). (2.8) 5 In particular, ∆FA := ∆12 ⊗L Lemma 2.9. ∆FA ∼= (F (Mω)G34 /F (Mω)G, G12, c12) is a decomposable abelian crossed product. 12 FA is a decomposable abelian crossed product. G12 Proof. Since we have already noticed that ∆FA is decomposable, we need only show the isomorphism. ∆FA is similar in the Brauer group to (L12FA/FA, H, c′ 12) where 12 is the restriction of c12 to H = Gal(L12FA/FA). However, L12 ∩ F (Mω)G = c′ LG12 and therefore Gal(L12FA/FA) = G12. Thus the given similarity is also an 12 isomorphism. Therefore, L12FA has degree p2 over FA = F (Mω)G and is a maximal subfield of ∆FA. Since both L12 and FA = F (Mω)G are contained in F (Mω)G34 , a degree p2 field extension over F (Mω)G, the composite must satisfy L12FA ∼= F (Mω)G34 . By Lemma 2.9 and equation (2.3), there is an isomorphism ∆FA ∼= (F (Mω)G34 /F (Mω)G, G12, z, e(u), e(b)). (2.10) Corollary 2.11. ∆FA ∼= (F (Mω)G124 /F (Mω)G, x3) ⊗ (F (Mω)G123 /F (Mω)G, x4) and as a consequence, there is a maximal abelian subfield N1 ⊂ ∆FA with N1 ∼= F (Mω)G12 and ∆FA ∈ Dec(N1/F (Mω)G). Proof. A similar argument as in the proof of Lemma 2.9 shows that ∆3 ⊗ FA ∼= (F (Mω)G124 /F (Mω)G, x3) and ∆4 ∼= (F (Mω)G123 /F (Mω)G, x4). Since F (Mω)G124⊗ F (Mω)G123 ∼= F (Mω)G12 we see ∆FA is split by F (Mω)G12 . Since [F (Mω)G12 : F (Mω)G] = p2 = deg(∆FA), there exists a maximal abelian subfield N1 ⊂ ∆FA such that N1 ∼= F (Mω)G12 and ∆FA ∈ Dec(N1/F (Mω)G). We can now state the main theorem of the paper. The proof is given in section 5. Theorem 2.12. Let F be a field with a G-action so that F ∗ is an H 1-trivial G- module. Then, ∆FA ∼=(cid:0)F (Mω)G34/F (Mω)G, G12, z, e(u), e(b)(cid:1) is a decomposable abelian crossed product division algebra defined by a non-degenerate matrix of index p2 and exponent p, p 6= 2. Remark 2.13. By Lemma 2.9 ∆FA is decomposable, hence to prove this theorem there are two things left to show. First we need to show that ∆FA has index p2 and therefore is a division algebra and second that e(u) is non-degenerate in F (Mω)G34 . The difficulty will lie in showing that e(u) is non-degenerate. This is difficult because the way things stand now, there is no "easy to understand" G-action on F (Mω). Recall that a G-lattice Λ is said to be H 1-trivial if H 1(H, Λ) = 0 for all subgroups H ≤ G. By [Sal99, Theorem 12.4(c)], if Λ is H 1-trivial, and K is a field with G-action so that K ∗ is also an H 1-trivial G-module, then K(Λ)∗ ∼=G K ∗ ⊕ Λ ⊕ P , where P is a permutation module and the isomorphism is a G-module isomorphism. In particular, if the lattice Λ is H 1-trivial, the G-action is "easy to understand". In section 3 we show that Mω is not an H 1-trivial G-module. As a consequence in section 3 we also construct an H 1-trivialization of Mω. In Theorem 5.2 we show that e(u) is non-degenerate in the field extension generated by this larger lattice. It is in the proof of the non-degeneracy of e(u) that we use p 6= 2 (see the proof of Lemma 4.7). Remark 2.14. In Theorem 2.12, we can always choose the field F so that it is a trivial G-module. For example, we could choose F to be a field of characteristic p and then, since G is a p-group, F ∗ with trivial G-action is a trivial H 1-module. 6 Remark 2.15. Theorem 2.12 shows that the converse to [McK08, Prop. 3.1.1] is false. That is, [McK08, Prop. 3.1.1] says that if ∆ is an abelian crossed product with exponent p defined by a non-cyclic group G and u is degenerate (p 6= 2), then CH2(SB(∆)) is torsion free. In our case the algebra in Theorem 2.12 has exponent p. Moreover CH2(SB(∆FA)) is torsion free since ∆FA is decomposable (see [Kar96] and [Kar98]). However, the matrix defining ∆FA is non-degenerate. The converse to [McK08, Prop. 3.1.1], therefore, is false. Remark 2.16. Since e(u) is a non-degenerate matrix, ∆FA /∈ Dec(F (Mω)G34 /F (Mω)G) (see e.g., [Sal99, Prop. 7.13]). We also record this result in Corollary 5.4. Index Calculation 2.1 By the Schofield-Van den Bergh index reduction formula [SVdB92], ind(∆12 ⊗ FA) is the minimum of the index of ∆12 ⊗ Ai as i varies and the tensor product is taken over LG. To get a lower bound on this index, we can compute the exponent of this algebra which is precisely the exponent of the cocycle class (i+1)[c12]−i[c3]−i[c4] ∈ H 2(G, Q). It is easy to see that exp(∆12 ⊗ Ai) = exp((i + 1)[c12] − i[c3] − i[c4]) =(cid:26) p p2 p i + 1 p ∤ i + 1 . Therefore, to show that ind(∆12 ⊗ FA) = p2 one need only show when p i + 1, ind(∆12 ⊗ Ai) = pn for some n ≥ 2. This is accomplished by the following lemma. Lemma 2.17. Set Bm = (∆12 ⊗L G12 12 LG)⊗mp ⊗LG (∆3 ⊗L G3 3 LG) ⊗LG (∆4 ⊗L G4 4 LG). Then for all m ∈ Z+, ind(Bm) = pn for some n ≥ 2. Proof. The idea of the proof is to show that the restriction of Bm to LG34, a maxi- mal subfield of ∆12 ⊗ LG, has index p2. Set E = F (A2(G12)⊕ Z[G3]⊕ Z[G4]). E is naturally a subfield of L = F (Q) and in fact L = E(x3, x4) where x3 and x4 are in- dependent indeterminates corresponding to the factor of N3Z and N4Z in the lattice Q. Moreover, x3 and x4 have trivial action by G and therefore, LG = EG(x3, x4). The algebras ∆3 and ∆4 are the cyclic algebras (F (Z[Gj ])(xj )/F (Z[Gj ])Gj (xj ), xj) for j = 3, 4 respectively. We have, G3 3 ∆3 ⊗L ∆4 ⊗L G4 4 LG ∼= (EG124 (x3, x4)/EG(x3, x4), x3), LG ∼= (EG123 (x3, x4)/EG(x3, x4), x4) and Therefore, ∆12 ⊗L G12 12 LG ∼= (EG34 (x3, x4)/EG(x3, x4), c12). Bm ⊗LG EG34(x3, x4) ∼ (EG4 (x3, x4)/EG34(x3, x4), x3) ⊗ (EG3 (x3, x4)/EG34 (x3, x4), x4). (2.18) The algebra in (2.18) is isomorphic to the central localization of the iterated twisted polynomial ring EG34 [t3, t4; σ3, σ4] which is a domain (see e.g., [Sal99, pg. 9]). Therefore, (EG4 (x3, x4)/EG34 (x3, x4), x3) ⊗ (EG3(x3, x4)/EG34(x3, x4), x4) has in- dex p2. This proves the lemma. Corollary 2.19. ∆FA has exponent p and index p2. 7 Proof. By the exponent calculation before Lemma 2.17 we only need to show that p2 ind(∆12 ⊗ Ai) when pi + 1. Assume p i + 1 and set pm = i + 1. Since ∆3 and ∆4 have index p, ∆12 ⊗ Ai ∼ B and hence the corollary follows from Lemma 2.17 and the fact that ind(∆FA) ind(∆12) = p2. 3 An H 1-trivialization of Mω Let Mω be the G-lattice constructed in section 2, equation 2.7. In this section we construct a specific H 1-trivialization of Mω. In particular, we construct an exten- sion of G-lattices 0 → Mω → M → P → 0 such that P is a permutation lattice and M is H 1-trivial. As noted in Remark 2.13, we need this H 1-trivialization because it gives us an easier to understand G-action on the field F (M ). In particular, using this specific H 1-trivialization, M , in Theorem 5.2 we are able to show that e(u) is non-degenerate in F (M )G34 . This implies that e(u) is non-degenerate in F (Mω)G34 because of the natural inclusion F (Mω) ⊂ F (M ). In fact, by [Sal99, Theorem 12.9], F (M )G34 is a rational extension of F (Mω)G34 since M is an extension of Mω by a permutation lattice, however we do not need or use this fact. Remark 3.1. Using the notation and terminology of [Lor05, Sections 2.5-2.7], one [Sal99, 12.5] H 1-trivializations exist. It was pointed out to the author that from [Lor05, 2.7.1(c)] it follows that any two H 1-trivializations are stably permutation could define an H 1-trivialization of a G-lattice Λ to be any G-lattice eΛ such thateΛ is H 1-trivial (or coflasque in [Lor05, 2.5]) and such that Λ ∼fl eΛ, that is, Λ and eΛ are flasque equivalent. In particular, using [Lor05, 2.7.1(c)] we see that eΛ is an H 1- trivialization of Λ if and only if eΛ is H 1-trivial and there is a short exact sequence of G-lattices 0 → Λ → eΛ ⊕ P → Q → 0 with P and Q permutation lattices. By equivalent. That is, if eΛ and Λ′ are both H 1-trivializations of Λ, then there exist permutation lattices P1 and P2 such that eΛ ⊕ P1 ∼= Λ′ ⊕ P2. In particular, an H 1-trivialization is unique up to stable permutation equivalence. In this section we take the time to construct a specific H 1-trivialization because we need to know in Lemma 3.7 the form of the specific 1-cocycles that we split. The specific form of the 1-cocycles allows us to prove Proposition 4.11, giving us a valuable tool for proving that e(u) is non-degenerate in F (M )G34 in Theorem 5.2. Mω is not itself an H 1-trivial G-lattice and in the first lemma we calculate the groups H 1(H, Mω) for all subgroups H ≤ G. Lemma 3.2. Let H be a subgroup of G. Then H 1(H, Mω) ∼= Z/nZ, where H resG H (ω) H (ω) is the order of the class of resG n = H (ω) in H 2(H, Q). where resG Proof. By [McK08, Lemma 3.2.4], A2(G12) is H 1-trivial. Since A2(Gj) ∼= ZNj, a trivial permutation lattice, K3 and K4 are each permutation lattices and therefore they are H 1-trivial. Together this implies that Q = A2(G12)⊕K3⊕K4 is H 1-trivial. From the short exact sequence 0 → Q → Mω → I[G] → 0 we get the long exact sequence of cohomology, . . . → 0 → H 1(H, Mω) → H 1(H, I[G]) → H 2(H, Q) → . . . It is easy to see H 1(H, I[G]) ∼= Z/HZ and this group is generated by the class of the 1-cocycle dH : H → I[G] given by dH (h) = h − 1 for all h ∈ H. By the H (ω)] ∈ H 2(H, Q). Therefore, H 1(H, Mω) is cyclic of definition of Mω, [dH ] 7→ [resG order n where n = H/resG H (ω). 8 Lemma 3.3. Let H be a subgroup of G. Then, resG H (ω) = max {H12,H3,H4} , where H12, H3 and H4 are the quotient subgroups of G12, G3 and G4 (respectively) defined in (2.2). Proof. Let E ≤ G12 be a subgroup. Then it is easy to show resG12 E ([c12]) = E. This follows from looking at a segment of the long exact sequence of cohomology . . . → H 1(E, Z[G12]) → H 1(E, I[G12]) → H 2(E, A(G12)) → . . . Here H 1(E, Z[G12]) = 0 since Z[G12] is a permutation module and by [Sal99, 12.3] all permutation modules are H 1-trivial. The class, resG12 E ([c12]) is the image of the generator of H 1(E, I[G12]), a cyclic group of order E. order of resG G and then restricted to H. In particular, by the above paragraph, resG H12. For j = 3 and 4, we have resG subgroups, we see that resG Let H be a subgroup of G. Since H is a p-group and [ω] = [c12]− [c3]− [c4], the H ([ω]) is the maximum of the orders of [c12], [c3] and [c4], inflated to H ([c12]) = Gj ([cj]) = p. Since Gj has no non-trivial H ([cj]) = Hj. The lemma now follows directly. Combining Lemmas 3.2 and 3.3 we immediately get the following corollary. Corollary 3.4. H 1(H, Mω) ∼= Z/nZ where H n = max{H12,H3,H4} . By [Sal99, 12.5] one can construct an H 1-trivial module containing Mω by split- ting all non-trivial cocycles with permutation modules. We will proceed in a slightly more efficient manner by splitting the non-trivial cocycles from the following subset H of subgroups of G. Let H = {G,{hτ, σ3, σ4iτ ∈ G12}}, a set of subgroups of G. For all H ∈ H, let fH be a 1-cocycle whose class [fH] ∈ H 1(H, Mω) is mapped to resG H (ω)[dH ] ∈ H 1(H, I[G]). In other words, by the proof of Lemma 3.2, [fH ] generates H 1(H, Mω). Define the G module M by the abelian group isomorphism M ∼= Mω ⊕ P (3.5) where P is the permutation lattice LH∈H Z[G/H]. For each H ∈ H fix a set of coset representatives {gi} for G/H and let Z[G/H] be generated over Z by the symbols ugiH. Define the G-action on M by g(x, 0) = (g · x, 0) for x ∈ Mω and g(0, ugiH ) = (gjfH(h), ugj H ), where ggi = gjh with the gi, gj elements of the fixed coset representatives of G/H and h ∈ H. Lemma 3.6. M is H 1-trivial. Proof. From the short exact sequence 0 → Mω → M → P → 0 and the fact that permutation modules are H 1-trivial, there is a surjection H 1(K, Mω) → H 1(K, M ) for all subgroups K of G. Therefore, if we show that the generators of H 1(K, Mω) are split in M , then we will have shown that M is H 1-trivial. By construction of the module M , for subgroups H ∈ H, H 1(H, M ) = 0 because (fH (h), 0) = (h − 1)(0, uH) for all h ∈ H. Moreover, if the generator [fK] ∈ H 1(K, Mω) satisfies K([fH ]) with the generator [fH ] ∈ H 1(H, Mω) for some H ∈ H, then [fK] = resH [fK] is split in M and therefore H 1(K, M ) = 0. We will use these arguments in the cases below. 9 Let K ≤ G a subgroup of G with H 1(K, Mω) ∼= Z/nZ with n 6= 1. By Corollary 3.4 there are no K ≤ G with n = K. We can also assume K 6= G since G ∈ H. We address the three remaining cases separately. K(ω) = p2 and therefore the kernel of Case 1: K = p3 and n = p. In this case resG H 1(K, I[G]) → H 1(K, Q) is generated by p2[dK]. Since restriction commutes with the long exact sequence of cohomology we have the following commutative diagram. 0 0 / H 1(G, Mω) H 1(G, I[G]) H 2(G, Q) resG K resG K resG K / H 1(K, Mω) / H 1(K, I[G]) / H 2(K, Q) K([fG]). Therefore, H 1(K, M ) = 0. K([fG]) 7→ p2[dK]. Therefore, resG K(p2[dG]) = p2[dK], the class K([fG]) = [fK] and since [fG] is split in M , so Since [ω] = p2, [fG] 7→ p2[dG]. Moreover since resG resG is resG Case 2: K = p3 and n = p2. In this case resG K(ω) = p. We will show that K ∈ H. Since n = p2, by Corollary 3.4, K ∩ G34 = p2, and in particular, G34 ≤ K. Then K/G34 is isomorphic to a cyclic p-subgroup of G12, say generated by τ . Then, K = hτ, σ3, σ4i ∈ H and hence H 1(K, M ) = 0. K(ω) = p and therefore p[dK] generates kernel of H 1(K, I[G]) → H 2(K, Q). We will show that K is a subgroup of an H ∈ H − {G} and resH K([fH ]) = [fK]. Since max{K12,K3,K4} = p and K12 = K/K ∩ G34, we see K ∩ G34 ≥ p. Therefore there exist two elements, τ1 ∈ G34 ∩ K and τ2 ∈ K so that K = hτ1, τ2i. Write τ2 = σm1 4 . Then clearly, K ≤ H = hσm1 2 , σ3, σ4i ∈ H − {G}. As in case 2 we have the following commutative diagram. K = p2 and n = p. In this case resG 1 σm2 2 σm3 3 σm4 Case 3: 1 σm2 0 → H 1(H, Mω) H 1(H, I[G]) H 2(H, Q) resH K resH K resH K 0 → H 1(K, Mω) / H 1(K, I[G]) / H 2(K, Q) By Lemma 3.2 resG resH K(p[dH ]) = p[dK], the class resH H (ω) = p. Therefore, [fH ] 7→ p[dH ] ∈ H 1(H, I[G]) and since K([fH ]) = [fK]. K([fH ]) 7→ p[dK]. Therefore, resH vH : H12 → A2(G12) be the 1-cochain defined by vH (¯h1) = P¯h∈H12 c12(¯h1, ¯h) In section 4, Proposition 4.11 we will need an explicit description of a cocycle in the class of [fH] ∈ H 1(H, Mω) for all H ∈ H. In fact Lemma 3.7 is the only result from Section 3 we use in the remainder of the paper. For any H ∈ H let where H12 is considered as a subgroup of G12. Since H12 is a finite group we have H12 [c12] = 0 in H 2(H12, A2(G12)). In fact one can check that the 1-cochain vH satisfies H12 c12 = δvH where δ is the co-boundary map. We will use the cochain vH in the proof of the following lemma. Lemma 3.7. For every H ∈ H there is a 1-cochain zH : H → K3 ⊕ K4 so that the 1-cochain fH : H → Mω defined by fH (h) = (zH (h) − inf H12 H vH (h),resG H (ω)(h − 1)) is a 1-cocycle whose image in H 1(H, I[G]) is resG Proof. If the given cochain is a cocycle, then its image in H 1(H, I[G]) is clearly resG H (ω) [dH ]. Hence we need only show that the given cochain is a cocycle for some cochain zH . H (ω) [dH ]. 10 / / / / / / / / / / / / / / Recall [ω] ∈ H 2(G, Q) is defined by inflation of c12, c3 and c4 from the three sub- groups G12, G3 and G4, viewed as quotient groups of G. Since inflation commutes with restriction ([NSW00, Prop. 1.5.5]), [resG H (ω)] = [inf H12 H (c12)] + [inf H3 H (c3)] + [inf H4 H (c4)] (3.8) Note that for every H ∈ H, max{H12,H3,H4} = H12. Therefore, by Lemma 3.2, resG H (ω) = H12. Moreover, for every H ∈ H, p divides H12. Since [c3] = resG H (ω)c4. Using the fact that inflation commutes with the co-boundary homomorphism ([NSW00, Prop. 1.5.2]) and the 1-cochain vH given above satisfies H12 c12 = δvH , we can explicitly write resG [c4] = p, there are 1-cochains bc3 : H3 → K3 and bc4 : H4 → K4 so that δbc3 = H (ω)c3 and δbc4 = resG H (ω)resG H (ω) as resG H (ω)resG H (ω) = δ(inf H12 H vH + inf H3 Set zH : H → K3 ⊕ K4 to be the 1-cochain zH = −inf H3 δfH(h1, h2) = (δzH (h1, h2) − δinf H12 H vH (h1, h2) + resG = (0, 0). H bc3 + inf H4 H bc4). H bc3 − inf H4 H bc4. Then, H (ω)resG H (ω), 0) The cochain fH is a cocycle since its co-boundary is zero. 4 Elements of M G34 Let M be the H 1-trivialization of the G-lattice Mω defined in (3.5). Let M G34 indicate the elements in M which are fixed by the subgroup G34 ≤ G. In this section we construct a G12-module homomorphism π′ : M G34 → Z/pZ which will be used to distinguish elements of M G34. As a first step we recall some facts about A2(G12) since it naturally sits in M G34 as a G12-submodule. Remark 4.1. The next two lemmas can be gleaned from [Sal99, pg. 49]. Lemma 4.2. A2(G12) is generated over Z[G12] by u12 = (σ2 − 1)d1 − (σ1 − 1)d2, b1 = N1d1 and b2 = N2d2. Lemma 4.3. The relations in A2(G12) are generated over Z[G12] by (σi − 1)bi = 0 for i = 1, 2, (σ2 − 1)b1 = N1u12 and −(σ1 − 1)b2 = N2u12. In particular, let x, y, z ∈ Z[G12] such that xu12 + yb1 + zb2 = 0. Then, x = z′N2 + y′N1, y = −(σ2 − 1)y′ + (σ1 − 1)y′′ and z = (σ1 − 1)z′ + (σ2 − 1)z′′ for some y′, y′′, z′, z′′ ∈ Z[G12]. By Lemma 4.3 A2(G12) ∼= (Z[G12]u12 ⊕ Z[G12]b1 ⊕ Z[G12]b2) /R where R = xu12 + yb1 + zb2(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) x = (z′N2 + y′N1) y = −(σ2 − 1)y′ + (σ1 − 1)y′′ z = (σ1 − 1)z′ + (σ2 − 1)z′′ for some y′, y′′, z′, z′′ ∈ Z[G12]  Let (ǫ, ǫ, ǫ) : Z[G12]u12 ⊕ Z[G12]b1 ⊕ Z[G12]b2 → Z ⊕ Z ⊕ Z be the map induced from the augmentation map ǫ : Z[G12] → Z. Since (ǫ, ǫ, ǫ)(R) ⊂ pZ ⊕ 0 ⊕ 0, there is an induced map A2(G12) → Z/pZ ⊕ Z ⊕ Z. Let π : A2(G12) → Z/pZ be the composition of this map with projection onto the first coordinate. π is a G12- module homomorphism where G12 acts trivially on Z/pZ. We will use the map π to distinguish elements of A2(G12) from one another. Note that if we define A2(G12) to be a G-module by assuming that G34 acts trivially on it, then π : A2(G12) → Z/pZ is also a G-module homomorphism. 11 In the next series of lemmas we prove that there is a G-module homomorphism π′ : M G34 → Z/pZ, extending π : A2(G12) → Z/pZ. In order to prove this, we first need to calculate the value of π at some particular elements of A2(G12). Recall for m = (m1, m2), n = (n1, n2) ∈ N2 we set zm = zm1 2 ∈ ∆12 where ∆12 is the G12-abelian crossed product given in (2.3). Set um,n ∈ A2(G12) to be the unique element such that e(um,n) = zmzn(zm)−1(zn)−1. Lemma 4.4. For all m = (m1, m2), n = (n1, n2) ∈ N2, π(um,n) = m1n2 − m2n1 + pZ. 1 zm2 Proof. Let s, t ∈ N. From the rules of multiplication in ∆12, it is easy to calculate e(u(s,0),(0,t)) = zs 1zt 2(zs 1)−1(zt 2)−1 = s−1Yi=0 t−1Yj=0 1σj σi 2(e(u12)) (4.5) One can also check that zmzn = σm1 fore, 1 (e(u(0,m2),(n1,0)))σn1 1 (e(u(m1,0),(0,n2))). There- um,n = −σm1 1 · n1−1Xj=0 m2−1Xi=0 1σj σi 1  2 u12 + σn1 n2−1Xj=0 m1−1Xi=0 1σj σi 2 u12 . (4.6) Applying π to um,n using formula (4.6) and the fact that π(g · x) = π(x) for all g ∈ G12 and all x ∈ A2(G12) we immediately get π(um,n) = m1n2− m2n1 + pZ. Lemma 4.7. Let H ∈ H and vH : H12 → A2(G12) be the 1-cochain defined before Lemma 3.7. For p 6= 2, π(vH (¯h)) = 0 for all ¯h ∈ H12. Proof. Let H ∈ H and h ∈ H with image ¯h ∈ H12. By definition, vH (¯h) = P¯h′∈H12 c12(¯h, ¯h′). Let ¯h = σm = σm1 2 ∈ H12. By the definition of c12, e(c12(zg, zh)) = zgzh(zgh)−1 for all g, h ∈ G12. Let mi + ni = qip + ri with 0 ≤ ri < p for i = 1, 2. Then, 1 zr2 2 ∈ H12 and let σn = σn1 e(c12(σm, σn)) = zmzn(zr1 1 σm2 1 σn2 2 )−1 = σm1 = σm1 1 (u(0,m2),(n1,0))zm1+n1 1 1 (u(0,m2),(n1,0))bq1 1 σr1 zm2+n2 2 1 (bq2 2 ) (zr1 1 zr2 2 )−1 (4.8) Note that if ¯h = 1, then c12(¯h, ¯h′) = 0 for all ¯h′ ∈ H12. We assume from now on that ¯h 6= 1. We now consider the two cases H = G and H 6= G separately. First, if H 6= G, then H12 = h¯hi with ¯h = σm. For each 0 ≤ i ≤ p − 1 set im1 = qip + ri with 0 ≤ ri < p. Then, π(u(¯h)) = = π(c12(σm, (σm)i)) π(σn1 1 (u(0,m2),(ri,0))) (by (4.8)) p−1Xi=0 p−1Xi=0 = − p(p − 1) 2 m1m2 + pZ. The last line follows from Lemma 4.4 and the fact that ri ≡ imi for all i. For all 2 m1m2 ≡ 0 mod p. Hence π(u(¯h)) ≡ 0 in this case. Similarly, if p 6= 2, − p(p−1) 12 H = G, then H12 = G12 and π(u(¯h)) = = = p−1Xi=0 p−1Xi=0 p−1Xi=0 p−1Xj=0 p−1Xj=0 p−1Xj=0 = −pm2 p(p − 1) 2 π(c12(σm, σi 1σj 2)) π(σm1 (u(0,m2),(i,0))) (by (4.8)) −m2 i (by Lemma 4.4) + pZ = pZ. Therefore π(u(¯h)) ≡ 0 in the case H = G. Lemmas 4.4 and 4.7 will be used to prove that π extends to π′ : M G34 → Z/pZ in Proposition 4.11. In the next lemma we express elements of M as 3-tuples (x, y, z) ∈ order to show the G-morphism π : A2(G12) → Z/pZ extends to M G34, we will need to show that given (x, y, z) ∈ M G34 the I[G] component of the direct sum, y, has a particular form. This is the content of the following lemma. Q ⊕ I[G] ⊕ P ∼= M . Recall Q = A2(G12) ⊕ K3 ⊕ K4 and P =LH∈H Z[G/H]. In Lemma 4.9. Let (x, y, z) ∈ M and let N34 =Pg∈G34 g ∈ Z[G] be the G34 norm. If (x, y, z) ∈ M G34, then y = N34 y1 − py2 where y1 ∈ Z[G12] and y2 ∈ Z[G] with ǫ(y2) = pǫ(y1). Proof. Let g ∈ G34 and ugiH ∈ Z[G/H] for some H ∈ H. Let (0, 0, ugiH ) ∈ M . Since g ∈ H we have gugiH = ugiH . By the definition of the G-action on M and Lemma 3.7 H (ω)ω(gi, g),resG g(0, 0, ugiH ) = (gizH (g) + resG Here we have used the fact that vH (¯g) = 0 since g ∈ G34. Using the fact that presG H (ω) for all H ∈ H, line (4.10) shows for each z ∈ P and each g ∈ G34 there exists xz,g ∈ Q and yz ∈ Z[G] so that g(0, 0, z) = (xz,g, yzp(g − 1), z). Similarly, for each y ∈ I[G] and g ∈ G34 there exists xy,g ∈ Q so that g(0, y, 0) = (xy,g, gy, 0). Finally we can conclude that for g ∈ G34, H (ω)gi(g − 1), ugiH ). (4.10) (g − 1)(x, y, z) = (gx + xy,g + xz,g − x, (yzp + y)(g − 1), 0). Therefore, if (x, y, z) ∈ M G34, then (g − 1)(y + pyz) = 0 for all g ∈ G34. This implies that y = N34 · y1 − pyz for some y1 ∈ Z[G12]. Since y ∈ I[G], ǫ(y) = p2ǫ(y1) − pǫ(yz) = 0. Setting y2 = yz completes the proof. Proposition 4.11. For p 6= 2 there exists a G-module homomorphism π′ : M G34 → Z/pZ extending π : A2(G12) → Z/pZ. Proof. The G12-module homomorphism π : A2(G12) → Z/pZ easily extends to a G- module homomorphism on Q = A2(G12)⊕K3⊕K4 by setting π(x0, x1, x2) = π(x0). This follows because Q is a direct sum as G-modules and G34 acts trivially on A2(G12). Define π′ : M G34 → Z/pZ by π′(x, y, z) = π(x). 13 Since G34 acts trivially both on Z/pZ and M G34, to show that π′ is a G-module homomorphism we need only show that π′ is a G12-module homomorphism. Let j ∈ {1, 2} and let (x, y, z) ∈ M G34. To prove our proposition we need to show π′(σj (0, y, 0)) = 0 and π′(σj(0, 0, z)) = 0. By Lemma 4.9, y = N34 y1 − py2 for some y1 ∈ Z[G12] and y2 ∈ Z[G]. Note that for any α ∈ I[G], π′(0, pα, 0) = pπ′(0, α, 0) = 0. This fact is used many times in the calculations below. Note also that y2 − py1 ∈ I[G] by Lemma 4.9. Therefore, π′(σj (0, y, 0)) = π′(σj (0, y1(N34 − p2) − p(y2 − py1), 0)) = π′(σj (0, y1(N34 − p2), 0)) Set y1 =Pg∈G12 αgg with αg ∈ Z. Then, αg π′ σj (0, Xh∈G34 αg π′ Xh∈G34 ω(σj, gh), Xh∈G34 αg π Xh∈G34 ω(σj, gh)! αg π(cid:0)p2c12(σj, g)(cid:1) = 0 π′(σj (0, y, 0)) = Xg∈G12 = Xg∈G12 = Xg∈G12 = Xg∈G12 (gh − 1) − p2(g − 1), 0)! (gh − 1), 0! (4.12) To show π′(σj (0, 0, z)) = 0 for all j ∈ {1, 2} and all z ∈ P , we show that π′(σj (0, 0, ugiH )) = 0 for all H ∈ H and all coset representatives gi ∈ G/H. Set σj gi = gkh where gk is another fixed coset representative of G/H and h ∈ H. Set resG H (ω) = pδH and note that for all H ∈ H, δH > 0. Then, π′(σj (0, 0, ugiH )) = π′(gkfH (h), ugkH ) = π′(gk(zH (h) − vH (¯h)) + pδH ω(gk, h), pδH gk(h − 1), ugkH ) = π(gkzH (h) − gkvH (¯h) + pδH ω(gk, h)) = π(−vH (¯h)) = 0 The last two equalities follow from the fact that zH(h) ∈ K1 ⊕ K2 (Lemma 3.7) and π(vH (¯h)) = 0 for all ¯h ∈ H12 (Lemma 4.7). 5 e(u) is non-degenerate in F (Mω)G34 Before we prove the main theorem of the paper we give a homological formulation of the degeneracy condition. Let ∆ = (K/F, G, z, u, b) be an abelian crossed product with G any finite abelian group. Any commutator [x, y] in ∆ has reduced norm 1 hence any commutator that lands in K has K/F -norm 1. Therefore, using the K-basis of ∆ given by {zσσ ∈ G} we get a map, ϕ : G × G → H −1(G, K ∗) (σ, τ ) 7→ [zσ, zτ ] In this map we are using the identification H −1(G, K ∗) ∼= {N (K) = 1}/I[G]K ∗ where {N (K) = 1} are the set of elements in K ∗ with K/F -norm equal to 1 and I[G]K ∗ are the elements of the form g(k)/k for k ∈ K ∗ and g ∈ G. Let 14 c(σ, τ ) = zσzτ (zστ )−1 be the 2-cocycle associated to the abelian crossed product ∆. Using the properties of H −1(G, K ∗) and the standard commutator identities for products we can show that ϕ is bimultiplicative. ϕ(στ, γ) = [zστ , zγ] = [c(σ, τ )−1zσzτ , zγ] = [zσzτ , zγ] = σ([zτ , zγ])[zσ, zγ] = ϕ(τ, γ)ϕ(σ, γ) (The identity ϕ(σ, τ γ) = ϕ(σ, τ )ϕ(σ, γ) is done in an identical fashion). Since ϕ is also clearly symplectic, there is an induced map, which we also call ϕ, ϕ : G ∧ G → H −1(G, K ∗) Lemma 5.1. Let ∆ be the abelian crossed product given above. Then u is degenerate if and only if there exists a rank 2 subgroup H ≤ G so that ϕH : H ∧ H → H −1(H, K ∗) is the trivial map. Proof. Assume that u is degenerate. Then by definition there exits σm, σn ∈ G so that H = hσm, σni is a non-cyclic group and um,n = σm(a)a−1σn(b)b−1. In other words, ϕ(σm, σn) = um,n = 0 as an element of H −1(H, K ∗). Now by the bimultiplicativity of ϕ, ϕ((σm)s, (σn)t) = ϕ(σm, σn)st = 0. Therefore, ϕH is the trivial map. Conversely assume there is a rank 2 subgroup H ≤ G so that ϕH is the trivial map. Set H = hσm, σni. Then ϕ(σm, σn) = 0, in other words, um,n ∈ I[H]K ∗. We are done since I[H], the augmentation ideal of Z[H], is generated by (σm − 1) and (σn − 1). In the case under investigation in this paper the group G in the abelian crossed product is isomorphic to Z/pZ× Z/pZ and hence G∧ G ∼= Z/pZ. Therefore to show that u is non-degenerate in this case it suffices to show that ϕ is non-trivial on a sin- gle commutator. In particular, one need only show that u12 6= (σ1)(a)a−1σ2(b)b−1 for all a, b ∈ K ∗. We can now prove the main theorem of the paper which states that the matrix defining the decomposable abelian crossed product ∆FA is non-degenerate. As always we take p a prime, p 6= 2, G = hσ1i × hσ2i × hσ3i × hσ4i the elementary abelian group of order p4, G34 = hσ3i × hσ4i and G12 = hσ1i × hσ2i. Theorem 5.2 (Also listed as Theorem 2.12). Let F be a field with a G-action so that F ∗ is an H 1-trivial G-module. Then, ∆FA ∼=(cid:0)F (Mω)G34/F (Mω)G, G12, z, e(u), e(b)(cid:1) is a decomposable abelian crossed product division algebra defined by a non-degenerate matrix of index p2 and exponent p, p 6= 2. Proof. By Lemma 2.9 and Corollary 2.19 ∆FA is a decomposable abelian crossed product division algebra with isomorphism as stated in the theorem. It is only left to show that e(u) is a non-degenerate matrix. As mentioned above the statement of the theorem, we need only show that there do not exist a, b ∈ F (Mω)G34 such that e(u12) = σ1(a)a−1σ2(b)b−1. Since the G-lattice Mω is a direct summand of M , there is an inclusion of fields F (Mω)G34 ⊂ F (M )G34. Therefore, it is enough to show that there do not exist a, b ∈ F (M )G34 such that e(u12) = σ1(a)a−1σ2(b)b−1. By Lemma 3.6 M is an H 1-trivial G-module. Therefore, by [Sal99, 12.4(c)], we have a G-module isomorphism F (M )∗ ∼= F ∗ ⊕ M ⊕ P ′. Here P ′ is a permutation 15 G-module. Under this isomorphism, e(u12) 7→ u12 ∈ A2(G12) ⊂ M . Hence it is enough to show there do not exist a, b ∈ M G34 so that u12 = (σ1 − 1)a + (σ2 − 1)b. (5.3) By Proposition 4.11 the map π : A2(G12) → Z/pZ extends to a G-module homo- morphism π′ : M G34 → Z/pZ. Taking π′ of both sides of (5.3) we get π′(u12) = 1 = π′((σ1 − 1)a + (σ2 − 1)b) = 0. This is a contradiction. Corollary 5.4. ∆FA /∈ Dec(F (Mω)G34 /F (Mω)G). Proof. By [Sal99, Prop. 7.13 (i) and (iv)], e(u) is degenerate if and only if ∆FA de- composes with respect to the abelian extension F (Mω)G34 ∼= F (Mω)G134⊗F (Mω)G234 . Remark 5.5. Setting N2 = F (Mω)G34 and combining Corollaries 2.11 and 5.4 we see that there exist two maximal abelian subfields N1 and N2 of ∆FA such that and ∆FA ∈ Dec(N1/F (Mω)G) ∆FA /∈ Dec(N2/F (Mω)G). Remark 5.6. Let ∆ = (K/F, G, v, d) be an abelian crossed product defined by the finite abelian group G of rank r, the matrix v ∈ Mr(K ∗) and the vector d = (di) ∈ (K ∗)r. As mentioned in the introduction the generic abelian crossed product associated to ∆ is the abelian crossed product A∆ = (K(x1, . . . , xr)/F (x1, . . . , xr), G, v, dx) where the xi are independent indeterminates and dx = (dixi)r i=1 ∈ (K(x1 . . . , xr)∗)r. A∆ is a division algebra of index G and exponent lcm(exp(G), exp(∆)). Indepen- dently in [McK08, Theorem 2.3.1] (in the case char(F ) = p) and [Mou08, Theorem 3.5] A∆ is shown to be indecomposable if ∆ is defined by a non-degenerate ma- trix. Therefore, for ∆FA as in Theorem 5.2, the generic abelian crossed product is indecomposable of index p2 and exponent p. This example is in contrast A∆FA to the indecomposable generic abelian crossed product examples given in [McK08, section 3.3]. In those examples the generic abelian crossed products A∆ defined by ∆ = (K(x1, . . . , xr)/F (x1, . . . , xr), G, v, dx) have the property that ∆ itself is inde- composable ([McK08, Remark 3.3.3]). Moreover in that case the defining matrix v is shown to be non-degenerate by considering the torsion in CH2(SB(∆)), the chow group of co-dimension 2 cycles of the Severi-Brauer variety of ∆ and applying work of Karpenko [Kar98, Prop. 5.3]. Though the calculations in the proof of Theorem 5.2 may have been tedious, they are elementary is nature, and do not appeal to the torsion in CH2(SB(∆FA)). Remark 5.7. Since the abelian crossed product ∆FA is decomposable and defined by a non-degenerate matrix, decomposability of abelian crossed products is not determined by degeneracy of the matrix defining them (see [McK08, Prop. 3.1.1] and remark 2.15). 16 References [AS78] [Dra83] S. A. Amitsur and D. Saltman. Generic Abelian crossed products and p-algebras. J. Algebra, 51(1):76 -- 87, 1978. P. K. Draxl. Skew fields, volume 81 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 1983. [Gru76] Karl W. Gruenberg. Relation modules of finite groups. American Math- ematical Society, Providence, R.I., 1976. Conference Board of the Math- ematical Sciences Regional Conference Series in Mathematics, No. 25. [Kar96] N. A. Karpenko. On topological filtration for Severi-Brauer varieties. II. In Mathematics in St. Petersburg, volume 174 of Amer. Math. Soc. Transl. Ser. 2, pages 45 -- 48. Amer. Math. Soc., Providence, RI, 1996. [Kar98] N. A. Karpenko. Codimension 2 cycles on Severi-Brauer varieties. K- Theory, 13(4):305 -- 330, 1998. [Lor05] Martin Lorenz. Multiplicative invariant theory, volume 135 of Ency- clopaedia of Mathematical Sciences. Springer-Verlag, Berlin, 2005. In- variant Theory and Algebraic Transformation Groups, VI. [McK07] K. McKinnie. Prime to p extensions of the generic abelian crossed prod- uct. J. Algebra, 317(2):813 -- 832, 2007. [McK08] K. McKinnie. Indecomposable p-algebras and Galois subfields in generic abelian crossed products. J. Algebra, 320(5):1887 -- 1907, 2008. [Mou08] Karim Mounirh. Nondegenerate semiramified valued and graded division algebras. Comm. Algebra, 36(12):4386 -- 4406, 2008. [NSW00] Jurgen Neukirch, Alexander Schmidt, and Kay Wingberg. Coho- mology of number fields, volume 323 of Grundlehren der Mathema- tischen Wissenschaften [Fundamental Principles of Mathematical Sci- ences]. Springer-Verlag, Berlin, 2000. [Rei75] [Sal79] [Sal99] I. Reiner. Maximal orders. Academic Press [A subsidiary of Harcourt Brace Jovanovich, Publishers], London-New York, 1975. London Math- ematical Society Monographs, No. 5. David J. Saltman. Indecomposable division algebras. Comm. Algebra, 7(8):791 -- 817, 1979. David J. Saltman. Lectures on division algebras, volume 94 of CBMS Re- gional Conference Series in Mathematics. Published by American Math- ematical Society, Providence, RI, 1999. [Sal02] David J. Saltman. Invariant fields of symplectic and orthogonal groups. J. Algebra, 258(2):507 -- 534, 2002. [SVdB92] Aidan Schofield and Michel Van den Bergh. The index of a Brauer class on a Brauer-Severi variety. Trans. Amer. Math. Soc., 333(2):729 -- 739, 1992. [Tig81] Jean-Pierre Tignol. Produits crois´es ab´eliens. J. Algebra, 70(2):420 -- 436, 1981. 17
1903.01623	1	1903	2019-03-05T01:31:52	A complete classification of three-dimensional algebras over ${\mathbb R}$ and ${\mathbb C}$	[ "math.RA" ]	We provide a complete classification of three-dimensional associative algebras over the real and complex number fields based on a complete elementary proof. We list up all the multiplication tables of the algebras up to isomorphism. We compare our results with those given by mathematicians in the 19th century.	math.RA	math	A complete classification of three-dimensional algebras over R and C -- OnKoChiShin (visiting old, learn new) Yuji Kobayashi, Kiyoshi Shirayanagi, Sin-Ei Takahasi and Makoto Tsukada ] . A R h t a m [ 1 v 3 2 6 1 0 . 3 0 9 1 : v i X r a Abstract We provide a complete classification of three-dimensional associative algebras over the real and complex number fields based on a complete elementary proof. We list up all the multiplication tables of the alge- bras up to isomorphism. We compare our results with those given by mathematicians in the 19th century. 1 Introduction Since the discovery of the quaternions by Hamilton [1], many mathematicians had been trying to extend number systems to more general systems such as tes- sarines, coquaternions, biquaternions, and algebras introduced by Grassmann and by Clifford etc. (see Kolmogorov and Yushkevich [4] and van der Waerden [10]). These are finite-dimensional associative algebras over the real number field R, which were called "hypercomplex number systems" at that time. It was Peirce who first studied associative algebras systematically ([5]). He enumerated "pure" algebras of low dimension using so called Peirce's decomposition. How- ever, not only was his terminology ambiguous and were his proofs not rigorous, but he also did not give proofs for non-isomorphism between the listed algebras. Hawkes [2] and Taber [9] tried to give rigid reasoning for the methods of Peirce, but their resulting proofs are not satisfactory from the viewpoint of modern mathematics. On the other hand, Scheffers [6] and Study [8] determined unital algebras of low dimension over R and the complex number field C. Keywords: Associative algebras, Unital algebras, Curled algebras, Waved algebras, Straight algebras, Peirce's classification. 2010 Mathematics Subject Classification: Primary 16B99; Secondary 16U99 Research of the first, third and fourth authors was supported in part by JSPS KAKENHI Grant Number JP25400120. Research of the second author was supported in part by JSPS KAKENHI Grant Number JP15K00025. 1 In this paper, we classify (associative but not necessarily unital) algebras of dimension 3 over R and C up to isomorphism. Our classification is complete and the methods are very different from Peirce's. The proofs are rigorous and concrete. We first treat unital algebras in a classical way. Then, we divide non- unital algebras into three types, curled, waved and straight algebras. An algebra A is curled if x and x2 are linearly dependent for any x ∈ A, it is waved if it is not curled but x, x2 and x3 are linearly dependent for any x ∈ A, and it is straight otherwise. We look into subalgebras of A of dimension 2, and consider all combinations of them to list up all possible multiplication tables of algebras of dimension 3. Finally, we make a complete list of algebras of dimension 3 that are not isomorphic to each other. Our methods are elementary and we use ideal theory as little as possible. We appreciate the researches of the abovementioned past mathematicians, in particular, we respect Peirce for his pioneering work. By presenting our new complete classification, we attempt to make the paper as an homage to his work. We believe that the spirit of this paper can be best expressed in the subtitle, con- sisting of four Chinese characters which are pronounced as "OnKoChiShin". It is a word from the Analects of Confucius [11], which is commonly used in Japan as an idiom meaning "visiting old, learn new", or more precisely, discovering new things by studying the past through scrutiny of the old. Our result is as follows. Over C, up to isomorphism, we have 5 unital algebras and among the non-unital algebras, we have 5 curled algebras U 3 0 , U 3 1 , U 3 2 , U 3 3 , U 3 4 , C3 0 , C3 1 , C3 2 , C3 3 , C3 4 , 4 straight algebras and 9 waved algebras S3 1 , S3 2, S3 3 , S3 4 W 3 1 , W 3 2 , W 3 4 , W 3 5 , W 3 6 , W 3 7 , W 3 8 , W 3 9 , W 3 10 and one infinite family (cid:8)W 3 3 (k)(cid:9)k∈H i , C3 of waved algebras, where the symbols U 3 i denote specified multiplication tables and H is the right half-plane {x+yi x > 0 or x = 0, y ≥ 0} of the complex plane. Over R, in addition to the above algebras, we have one unital algebra U 3 3− and one infinite family 3− (k)(cid:9)k≥0 of non-unital waved algebras. The concrete description of the 2−, one non-unital straight algebra S3 above multiplication tables is given in Section 10. i and W 3 (cid:8)W 3 i , S3 The rest of this paper is organized as follows. First, we give a classification of two-dimensional algebras in Section 3, because it will be utilized to achieve the aim of classifying three-dimensional algebras. Next, we determine unital alge- bras in Section 4. We classify curled algebras in Section 5 and straight algebras 2 in Section 6. In the subsequent three sections, we treat the most difficult case where algebras are waved. In this case we focus on two subalgebras having their one-dimensional intersection. We explain the strategy for the classification in Section 7, enumerate all possible waved algebras in Section 8, and check if the algebras enumerated are isomorphic or not in Section 9. In the final section, we summarize the results and give a complete list of algebras modulo isomorphism, and compare it with the lists given by Peirce, Scheffers and Study. Throughout this paper, we assume that an algebra is associative. 2 Preliminaries Let K = R or K = C. In this paper, we express an algebra over K by the mul- tiplication table. Let A be an algebra over K of dimension n where {e1, . . . , en} is a linear bases of A. Then, the structure of A is determined by the table which is an n × n matrix a11 ... an1   ··· . . . ··· a1n ... ann   (1) with aij ∈ A, where aij = eiej for all i ∈ [1, n] and j ∈ [1, n]. That is, for p = Pn i=1 yiei ∈ A with xi, yi ∈ K, the product pq is defined by i=1 xiei, q = Pn pq = Xi,j∈[1,n] xiyjaij. We say that A is the algebra on {e1, . . . , en} defined by (1). We refer to A as a zero algebra if its multiplication table is the zero matrix with aij = 0 for all i ∈ [1, n] and j ∈ [1, n], or equivalently, if pq = 0 for all p, q ∈ A. Related to zero algebras, we call A a zeropotent algebra if p2 = 0 for all p ∈ A. Obviously a zero algebra is zeropotent, but in general the converse is not true. In fact, it is easy to show that A is zeropotent if and only if A is defined by an alternative matrix with aij = −aji for all i ∈ [1, n] and j ∈ [1, n]. For a classification of zeropotent nonassociative algebras, see [3, 7]. Let us describe a criterion for two algebras to be isomorphic. Let A and B be algebras over K of dimension n defined by tables (n×n matrices), (aij ) and (bij ), where {e1, . . . , en} and {f1, . . . , fn} are linear bases of A and B, respectively. Moreover, let eiej = aij = Pn ij ft for all i ∈ [1, n] and j ∈ [1, n]. Suppose that ϕ : A → B is an algebra homomorphism and Mϕ = (mst) with mst ∈ K is the n × n matrix of ϕ as a linear mapping with respect to the linear bases: ijes and fifj = bij = Pn s=1 αs t=1 βt ϕ(es) = n Xt=1 mstft. Let us call Mϕ a transformation matrix from A to B. For each (i, j), the equality ϕ(ei)ϕ(ej) = ϕ(eiej) = ϕ(aij ) holds. Calculating the both sides, we see that A 3 is isomorphic to B over K if and only if there exists a non-singular matrix Mϕ such that Xs αs ij mst = Xk,l βt kl mikmjl for all t, i, j ∈ [1, n]. ········· (∗) Assume that A is an algebra over K of dimension 3. An element e ∈ A is curled (resp. waved) if {e, e2} (resp. {e, e2, e3}) is linearly dependent over K. An algebra A is called curled if every element of A is curled. A is called waved if A is not curled but every element of A is waved. Note that in a waved algebra there is an element e such that {e, e2} is linearly independent but every such element e satisfies that {e, e2, e3} is linearly dependent. We call A straight, if it is not curled nor waved, that is, if {e, e2, e3} forms a linear base of A for some e ∈ A. When B is an algebra of dimension 2, B is curled if every element of B is curled, otherwise B is straight. For two algebras A and B over K, A ⊕ B denotes their direct sum; for p = (x, y), q = (x′, y′) ∈ A ⊕ B their product is defined by pq = (xx′, yy′). 3 Algebras of dimension 2 Before entering the discussion on two-dimensional algebras, let us consider one- dimensional algebras. Let A be an algebra over K of dimension 1 with a linear base {e}. Then, e2 = ke for some k ∈ K. Depending on whether k is 0 or not, it follows that A is the algebra defined by e2 = 0 or is isomorphic to the algebra defined by e2 = e. We denote the former algebra by A1 0 and the latter by A1 1. A1 0 is a zero algebra and A1 1 is a unital algebra with the identity element e. Now let A be an algebra over K of dimension 2 with a linear base {e, f}. First, suppose that A is curled. Then, e2 = ke and f 2 = ℓf (2) for some k, ℓ ∈ K. Here if k 6= 0 (resp. ℓ 6= 0), replacing e (resp. f ) by e/k (resp. f /ℓ), we may assume that k and ℓ are 0 or 1 in (2). We can write ef = ae + bf and f e = ce + df with a, b, c, d ∈ K. We have kae + kbf = kef = e2f = e(ef ) = e(ae + bf ) = ae2 + bef = a(k + b)e + b2f. It follows that Similarly we have ab = 0 and b2 = kb. cd = 0, a2 = ℓa, c2 = ℓc and d2 = kd. (3) (4) 4 We have kce + ade + bdf = e(ce + df ) = ef e = (ae + bf )e = kae + bce + bdf. It follows that Similarly we have k(a − c) = ad − bc. ℓ(d − b) = ad − bc. Because A is curled, xe + yf and (xe + yf )2 = x2e2 + xyef + xyf e + y2f 2 = (kx2 + axy + cxy)e + (ℓy2 + bxy + dxy)f. are linearly dependent over K for any x, y ∈ K. Hence, (cid:12)(cid:12)(cid:12)(cid:12) x kx2 + (a + c)xy y ℓy2 + (b + d)xy(cid:12)(cid:12)(cid:12)(cid:12) It follows that = (cid:0)(b + d − k)x + (ℓ − a − c)y(cid:1)xy = 0. (5) It is easy to solve the equations (3), (4) and (5) for k, ℓ ∈ {0, 1} and a, b, c, d ∈ k = b + d and ℓ = a + c. K. We obtain 7 solutions: (k, ℓ, a, b, c, d) = (0, 0, 0, 0, 0, 0), (0, 1, 0, 0, 1, 0), (0, 1, 1, 0, 0, 0), (1, 0, 0, 0, 0, 1), (1, 0, 0, 1, 0, 0), (1, 1, 0, 1, 1, 0), (1, 1, 1, 0, 0, 1). In correspondence to them we have 7 curled algebras A2 and A′2 6 defined by 0, A2 1, A2 2, A′2 3 , A′2 4 , A′2 5 (cid:18)0 0 0 0(cid:19) ,(cid:18)0 e 0 f(cid:19) ,(cid:18)0 0 f(cid:19) ,(cid:18)e f e 0 0(cid:19) ,(cid:18)e 0 f 0(cid:19) ,(cid:18)e e f f(cid:19) and (cid:18)e f e f(cid:19) , respectively. Using the isomorphism criterion in the previous section, the alge- 1 and A′2 4 are isomorphic via the transformation matrix Mϕ = (cid:18)0 1 1 0(cid:19) bras A2 which is a non-singular matrix satisfying (∗). Similarly, A2 morphic. The algebra A′2 (cid:18)1 1 0 1(cid:19). Similarly, A′2 to see that there is no non-singular matrix satisfying (∗) between A2 Consequently we have three non-isomorphic curled algebras A2 6 is isomorphic to A2 5 is isomorphic to A2 2. On the other hand, it is easy 1 and A2 2. 3 are iso- 1 via the transformation matrix 2 and A′2 0, A2 1, and A2 2. Next, we suppose that A is straight, that is, there is g ∈ A such that {g, g2} is a linear base of A. Then, A is commutative, and we can write g3 = bg2 + cg (6) with b, c ∈ K. 5 If b = c = 0, that is, g3 = 0, then letting e = g2 and f = g, A is defined by the table (cid:18)0 0 0 e(cid:19) . (7) Next, if b 6= 0 and c = 0, then letting f = g/b and e = f 2, A is defined by the table (cid:18)e e e e(cid:19) . Lastly, suppose that c 6= 0. Let e = 1 c (g2 − bg). (8) (9) (10) (11) Then, using (6), we see eg = ge = g, and so e is the identity element of A. Moreover, by (9), {g, e} is a linear base of A and g2 = bg + ce holds. Let D = b2 + 4c, and let h = 2g − be. Then, by (10) we have h2 = 4g2 − 4bg + b2e = De. If K = C and D 6= 0, or K = R and D > 0, let f = 1√D have f 2 = e. Hence, A is defined by h. Then, by (11) we f e(cid:19) . (cid:18)e f If K = R and D < 0, let f = 1√−D h. Then, f 2 = −e by (11). Hence, A is (cid:18)e f −e(cid:19) . defined by (12) (13) f If D = 0, let f = h, then, f 2 = 0 by (11), and A is defined by (cid:18)e f f 0(cid:19) . (14) If A is defined by the table in (8), then by replacing f by f − e, A is defined also by (cid:18)e 0 0 0(cid:19) . (15) If A is defined by the table in (12), then by replacing e by e+f 2 A is defined by and f by e−f 2 , (cid:18)e 0 0 f(cid:19) . (16) We denote the algebras defined by the tables in (7), (15), (12), (13) and (14) by A2 6, respectively. Using the isomorphism criterion in Section 2 it is easy to show that these algebras are not isomorphic to each other. 5− and A2 4, A2 3, A2 5, A2 6 4 Unital algebras In this section, A is a unital algebra over K of dimension 3 with identity element 1. First, suppose that there is an element h of A such that {1, h, h2} forms a linear base of A, in this case we say that A is unitally straight. Then, A is commutative, and we can write h3 = ah2 + bh + c with a, b, c ∈ K. Thus, the algebra A is generated by h and is isomorphic to the residue algebra of the polynomial algebra K[X] modulo the ideal generated by P (X) = X 3 − aX 2 − bX − c. Let α, β and γ be the roots of P in C; P (X) = (X − α)(X − β)(X − γ). (i) Suppose that α, β and γ are different from each other. (i1) Suppose that α, β, γ ∈ K. We have an isomorphism φ : A → K ⊕ K ⊕ K of algebras defined by φ(h) = (α, β, γ). Let e = φ−1(1, 0, 0), f = φ−1(0, 1, 0) and g = φ−1(0, 0, 1). Then, on the base {e, f, g}, A is defined by (17) e 0 0   0 f 0 0 0 g   . (i2) Next, suppose that K = R, and α is real but β and γ are not real. We have an isomorphism φ : A → K ⊕ B of algebras defined by φ(h) = (α, ¯h), where B = K[X]/((X − β)(X − γ)) and ¯h is the image of h by the natural surjection from A to B. By the results in the previous section, B is defined by (cid:18)f′ g′ −f′(cid:19) of B on a suitable base {f′, g′}. Let e = φ−1(1, 0) f = φ−1(0, f′) and g = φ−1(0, g′), then on the base {e, f, g}, A is defined by g′   e 0 0 f 0 0 g g −f   . (18) (ii) Next, suppose that α 6= β = γ, then α, β ∈ R and P (X) = (X − α)(X − β)2. 7 We have an isomorphism φ : A → K ⊕ B defined by φ(h) = (α, ¯h), here B = K[X]/((X − β)2) and ¯h is the natural image of h in B. By the results in the previous section B is defined by (cid:18)f′ 0(cid:19) on the base {f′, g′} with f′ = 1 and g′ = h − β. Thus, in the same way as above A is defined by g′ g′ e 0 0   0 f g 0 g 0   . (19) on a suitable base {e, f, g}. f = h − α and g = f 2. Then, f g = gf = g2 = 0, and so A is defined by (iii) Lastly, suppose that α = β = γ, that is, P (X) = (X − α)3. Let e = 1, e f g   f g 0 g 0 0   . (20) Using the isomorphism criterion it is easy to show that the algebras defined by (17), (19) and (20) are not isomorphic to each other. When K = R, the algebra defined by (18) is not isomorphic to the algebras above either. Next, suppose that A is not unitally straight, that is, there is no h ∈ A such that {1, h, h2} forms a linear base of A. Let {1, f, g} be a linear base of A. Then {1, f} is a linear base of the unital subalgebra B of A generated by f . As discussed in the previous section, B is defined by (12) or (14) when K = C. Hence, by changing f by a suitable element of B, we may suppose that f 2 = 0 or f 2 = 1. When K = R, there is another possibility f 2 = −1, because B may be defined by (13). Similarly, we may suppose that g2 = 0 or g2 = 1 (or g2 = −1 when K = R). f 2 = 0, then we have Let f g = a + bf + cg and gf = a′ + b′f + c′g for a, b, c, a′, b′, c′ ∈ K. If 0 = f 2g = f (a + bf + cg) = af + c(a + bf + cg) = ac + (a + bc)f + c2g. and 0 = gf 2 = (a′ + b′f + c′g)f = a′f + c′(a′ + b′f + c′g) = a′c′ + (a′ + b′c′)f + c′2g. It follows that Similarly, if g2 = 0, we have a = c = a′ = c′ = 0. a = b = a′ = b′ = 0. If f 2 = 1, then (21) (22) g = f 2g = f (a + bf + cg) = af + b + c(a + bf + cg) = ac + b + (a + bc)f + c2g. 8 Thus we have Similarly, we have ac + b = a + bc = c2 − 1 = 0. a′c′ + b′ = a′ + b′c′ = c′2 − 1 = 0. Hence, we have c = ±1, b = ∓a (double-sign corresponds) and c′ = ±1, b′ = ∓a′ (d-s.c.). (23) Similarly, if g2 = 1, we have b = ±1, c = ∓a (d-s.c.) and b′ = ±1, c′ = ∓a′ (d-s.c.). (24) Finally, if K = R and f 2 = −1, then we have −g = f 2g = f (a + bf + cg) = af − b + c(a + bf + cg) = ac − b + (a + bc)f + c2g. Hence, c2 = −1, but this is impossible in R. Similarly, g2 = −1 is impossible. (i) Suppose that f 2 = g2 = 0, then by (21) and (22) we have a = b = c = a′ = b′ = c′ = 0. Hence, A is defined by e f g   f 0 0 g 0 0   (25) on the base {e, f, g} with e = 1. (ii) If f 2 = 0 and g2 = 1, then by (21) and (24) a = c = a′ = c′ = 0, b = ±1 and b′ = ±1. Here, if b = b′ = 1, that is, f g = gf = f , then 1, f + g and (f + g)2 = f 2 + f g + gf + g2 = 2f + 1 are linearly independent. Hence, A is unitally straight. Similarly, the case b = b′ = −1 gives a unitally straight algebra. If b = 1 and b′ = −1, then A is defined by  g f  e e f f 0 g −f   (26) with e = 1. This is not unitally straight, because 1, x + yf + zg and (x + yf + zg)2 = x2 + z2 + 2x(yf + zg) The case b = −1 and b′ = 1 gives the algebra isomorphic to the previous are linearly dependent for any x, y, z ∈ K. algebra defined by (26) via the transformation matrix   1 0 0 −1 0  0 0 . 0 −1 9 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (iii) The case f 2 = 1, g2 = 0 is symmetric to (ii), and yields no new algebra. (iv) Suppose that f 2 = g2 = 1. Because A is not unitally straight, 1, h = x + yf + zg and h2 = x2 + y2 + z2 + 2xyf + 2xzg + yz(f g + gf ) = x2 + y2 + z2 + yz(a + a′) + y(2x + z(b + b′))f + z(2x + y(c + c′))g are linearly dependent for any x, y, z ∈ R. Hence, 1 x x2 + y2 + z2 + yz(a + a′) 0 0 y(2x + z(b + b′)) z(2x + y(c + c′)) y z = yz(y(c + c′) − z(b + b′)) = 0 for any x, y, z ∈ R. It follows that b + b′ = c + c′ = 0. (27) We have 4 solutions in a, b, c, a′, b′, c′ satisfying (23), (24) and (27): (a, b, c, a′, b′, c′) = (−1, 1, 1,−1,−1,−1), (1,−1, 1, 1, 1,−1), (1, 1,−1, 1,−1, 1), (−1,−1,−1,−1, 1, 1). Now, if (a, b, c, a′, b′, c′) = (−1, 1, 1,−1,−1,−1), that is, f g = −1 + f + g and gf = −1 − f − g, then let f′ = f + g and g′ = g. If (a, b, c, a′, b′, c′) = (1,−1, 1, 1, 1,−1), let f′ = f − g and g′ = −g. If (a, b, c, a′, b′, c′) = (1, 1,−1, 1,−1, 1), let f′ = f − g and g′ = g. If (a, b, c, a′, b′, c′) = (−1,−1,−1,−1, 1, 1), let f′ = f + g and g′ = −g. In any case we have f′2 = 0, g′2 = 1, f′g′ = f′ and g′f′ = −f′. Therefore, replacing e, f and g by 1, f′ and g′, respectively, A is defined by (26). The algebras defined by (25) and (26) are not isomorphic, because in the algebra defined by (25), 1 and −1 are the only elements x satisfying x2 = 1, but in the algebra defined by (26), g2 = 1. In summary, over C we have exactly 5 non-isomorphic unital algebras U 3 0 , U 3 4 defined by (25), (26), (17), (19) and (20), respectively. Over R we have exactly 6 non-isomorphic unital algebras, the algebras defined by the same tables as above and another algebra U 3 3 and U 3 1 , U 3 2 , U 3 2− defined by (18). 5 Curled algebras In this section we classify curled algebras over K of dimension 3. Let C3 2 , C3 0 , C3 1 , C3   ,  0 0 0 0 0 0 0 0 0   3 and C3 4 be the algebras on a base {e, f, g} defined by   ,   ,    and     ,  0 0 0 0 0 0 0 0 0 0 0 −e 0 0 f 0 f g e f g 0 0 g 0 0 0 0 e 0 0 e 0 0 0 e (28) 10 respectively. algebras. It is easy to check that these are actually curled (associative) We claim that these algebras are not isomorphic to each other. For an algebra A over K, define the square and the left (and right) annihilator of A by A2 = {Pn i=1 xiyj xi, yj ∈ A, n > 0} , and la(A) = {x ∈ A xA = 0}, ra(A) = {x ∈ A Ax = 0}, respectively. Obviously these are subalgebras of A. Let α(A) = dimK A2, β(A) = dimK la(A), γ(A) = dimK ra(A). Clearly, if A and A′ are isomorphic, then α(A) = α(A′), β(A) = β(A′) and γ(A) = γ(A′). By an easy calculation, we see α(C3 0 ) = 0, α(C3 1 ) = 1 and α(C3 2 ) = α(C3 3 ) = α(C3 4 ) = 3, β(C3 2 ) = 1, β(C3 3 ) = 2 and β(C3 4 ) = 0, and γ(C3 2 ) = 1, γ(C3 3 ) = 0 and γ(C3 4 ) = 2. Thus, our algebras can not be isomorphic to each other, although in this case we need not the information about the values of γ. Next, we shall prove that any curled algebra is isomorphic to one of the algebras listed in (28). We divide cases depending on whether {e, f, ef} or {e, f, f e} forms a base of A or not. (a) Suppose that there are elements e and f of A such that {e, f, ef} is linearly independent. Because A is curled, e2 = ke and f 2 = ℓf (29) for some k, ℓ ∈ K. If k 6= 0 (resp. ℓ 6= 0), by replacing e (resp. f ) by e/k (resp. f /ℓ), we may suppose that k and ℓ are equal to 0 or 1. Set g = ef, then {e, f, g} is a linear base of A. Because A is curled (e + f )2 = ke + ℓf + g + f e = m(e + f ) for some m ∈ K. Hence, f e = (m − k)e + (m − ℓ)f − g. Moreover, we have eg = eef = kef = kg, gf = ef f = ℓef = ℓg, 11 (30) (31) (32) (33) Therefore, ge = ef e = e((m − k)e + (m − ℓ)f − g) = k(m − k)e + (m − k − ℓ)g f g = f ef = ((m − k)e + (m − ℓ)f − g)f = ℓ(m − ℓ)f + (m − k − ℓ)g (34) (35) and g2 = gef = (k(m − k)e + (m − k − ℓ)g)f = ((k + ℓ)(m − k) + ℓ2)g. (36) Because e + g and (e + g)2 = k(m − k + 1)e + ((k + ℓ)(m − k) + m − ℓ2 − ℓ)g are linearly dependent, we have k(m − k + 1) = (k + ℓ)(m − k) + m − ℓ2 − ℓ, that is, (ℓ + 1)(m − k − ℓ) = 0. Because ℓ 6= −1, we have m = k + ℓ. Thus, (31), (34), (35) and (36) become f e = ℓe + kf − g, ge = kℓe, f g = kℓf, and g2 = kℓg. (a1) If k = ℓ = 0, then by (29), (30), (32), (33) and (37) we have (37) ef = g, f e = −g and e2 = f 2 = eg = gf = ge = f g = g2 = 0. Let e′ = −g and g′ = e. Then we have e′2 = e′f = e′g′ = f e′ = g′e = 0, f g′ = e′ and g′f = −e′. Hence, replacing e by e′ and g by g′, A is defined by the second table in (28), and is isomorphic to C3 1 . (a2) If k = 0 and ℓ = 1, then we have f 2 = f, ef = gf = g, f e = e − g and e2 = eg = ge = f g = g2 = 0. Let e′ = e − g, then we have e′2 = (e − g)2 = e2 − eg − ge + g2 = 0, e′f = ef − gf = 0, e′g = eg − g2 = 0 and f e′ = f e − f g = e − g = e′, ge′ = ge − g2 = 0. Hence, replacing e by e′, A is defined by the third table in (28), and is isomorphic to C3 2 . (a3) If k = 1 and ℓ = 0, then we have e2 = e, ef = eg = g, f e = f − g and f 2 = gf = ge = f g = g2 = 0. Let e′ = g, f′ = e, g′ = f − g. Then, we have e′2 = g2 = 0, e′f′ = ge = 0, e′g′ = gf − g2 = 0, f′e′ = eg = g = e′, 12 f′2 = e2 = e = f′, f′g′ = ef−eg = 0, g′e′ = f g−g2 = 0, g′f′ = f e−ge = f−g = g′ and Thus, A is isomorphic to C3 g′2 = f 2 − f g − gf + g2 = 0. 2 again. (a4) Finally, if k = ℓ = 1, then we have e2 = ge = e, f 2 = f g = f, ef = eg = gf = g2 = g and f e = e + f − g. Let e′ = e − g and g′ = g − f . Then, we have e′2 = e′f = e′g′ = f g′ = g′e′ = g′2 = 0, f e′ = e′ and g′f = g′. Therefore, A is isomorphic to C3 2 in this case too. (b) Suppose that {e, f, f e} is linearly independent for some e, f ∈ A. As we get (34) above, we can get ef = k′e + ℓ′f − f e with k′, ℓ′ ∈ K. Thus, {e, f, ef} is also linearly independent. Therefore, this case is included in case (a). (c) Suppose that for any linearly independent elements e and f of A, both {e, f, ef} and {e, f, f e} are linearly dependent. Let {e, f, g} be a linear base of A, and let B be the subalgebra of A generated by {e, f}. Because ef and f e lie in the space spanned by {e, f}, B is spanned by {e, f} and is two-dimensional. As discussed in Section 3, there are exactly three non-isomorphic curled algebras of dimension 2; they are defined by (cid:18)0 0 0 0(cid:19) ,(cid:18)0 e 0 f(cid:19) and (cid:18)0 0 e f(cid:19) . (38) Hence, we may assume that the subalgebra B is defined by one of the tables in (38). Because A is curled, g2 = kg (39) for some k ∈ K. Here we may suppose that k is equal to 0 or 1. (c1) Suppose that B is defined by the first table in (38) on the base {e, f}. Since {e, g, eg} is linearly dependent, eg = ae + bg for some a, b ∈ K. Because e2 = 0, we have 0 = e2g = e(ae + bg) = b(ae + bg) = abe + b2g. Hence, b = 0 and so eg = ae. By this last equality and (39) we have kae = keg = eg2 = aeg = a2e. Hence, a = 0 or a = k. Similarly, we see that ge = a′e, f g = bf, gf = b′f, and a′, b and b′ are equal to 0 or k. Thus, if k = 0, then a = a′ = b = b′ = 0, and hence, eg = ge = f g = gf = g2 = 0. Hence, A is isomorphic to the zero algebra C3 0 . Next, suppose that k = 1. Then, g2 = g and a, a′, b, b′ ∈ {0, 1}. Because A is curled, we have ℓ(e + g) = (e + g)2 = (a + a′)e + g 13 for some ℓ ∈ K. Hence, a + a′ = ℓ = 1. Similarly, we have b + b′ = 1. So we we have four possibilities (a, a′, b, b′) = (0, 1, 0, 1), (0, 1, 1, 0), (1, 0, 0, 1) and (1, 0, 1, 0). In correspondence to them we have the algebras defined by 0 0 e   0 0 f 0 0 g   ,   0 0 e 0 0 0 f 0 g   ,   0 0 0 0 0 f e 0 g   and   0 0 0 0 0 0 e f g   . (40) The algebra defined by the first table in (40) is nothing but C3 3 and the 4 . The algebra defined by the second table algebra defined by the last table is C3 is isomorphic to the algebra C3 2 via the transformation matrix   2 via   1 0 0 0 0 1 0 0 1 0 0 1 0 1 0  . The  . 1 0 0 algebra defined by the third table is also isomorphic to C3 (c2) Suppose that B is defined by the second table in (38). As above we have eg = ae, ge = a′e, a + a′ = k with a, a′ ∈ {0, k}. (41) Because {f, g, f g} is linearly dependent, f g = bf + cg for some b, c ∈ K. We have bf + cg = f g = f 2g = f (bf + cg) = bf + c(bf + cg) = b(c + 1)f + c2g and kbf + kcg = kf g = f g2 = (bf + cg)g = b(bf + cg) + kcg = b2f + (b + k)cg. Hence, Since and bc = c(c − 1) = b(b − k) = 0. 0 = ef g = e(bf + cg) = ace a′e = a′f e = f ge = (bf + cg)e = be + a′ce by (41), we have ac = 0 and b = a′(1 − c). Let gf = b′f + c′g with b′, c′ ∈ K. As above we have b′c′ = c′(c′ − 1) = b′(b′ − k) = 0. and ac′ = 0 and b′ = a′(1 − c′). 14 (42) (43) (44) (45) Because A is curled, for any y, z ∈ K, yf + zg and (yf + zg)2 = y2f + kz2g + yz(bf + cg) + yz(b′f + c′g) = (y + (b + b′)z)yf + (kz + (c + c′)y)zg are linearly dependent. Hence (y + (b + b′)z)yz = (kz + (c + c′)y)yz holds. Because y and z are arbitrary, it follows that b + b′ = k and c + c′ = 1. (46) Now, if k = 0, then by (41) -- (46), we see that a = a′ = b = b′ = 0 and either c = 0, c′ = 1 or c = 1, c′ = 0. In the first case, eg = ge = f g = gg = 0 and gf = g, In the second case, eg = ge = gf = gg = 0 and f g = g. Hence, A is defined by 0 e 0   0 f g 0 0 0   or   0 e 0 0 f 0 0 g 0   . (47) In the first case A is nothing but C3 2 . In the second case, A is isomorphic to C3 3 via   1 0 0 0 0 1 0 1 0  . Next, if k = 1, then again by (41) -- (46), we see that a = 0, a′ = 1 and either (b = c′ = 0, b′ = c = 1) or (b = c′ = 1, b′ = c = 0). Hence, A is defined by 0 e e   0 f f 0 g g   or   0 e e 0 f g 0 f g   . (48) 0 1 0 0 −1 1 1 0 0  . 3 via  2 via  The algebra defined by the first table in (48) is isomorphic to C3 The algebra defined by the second table in (48) is isomorphic to C3  0 0 1 0 . 1 −1 (c3) Suppose that B is defined by the last table. This case is symmetric 1 0 0 with (c2), and from (47) and (48) we have four curled algebras defined by 0 e 0 f 0 0   0 g 0   ,   0 e 0 f 0 g 0 0 0   ,   0 0 0 e f g e f g   and   0 0 0 e f f e g g   . (49) 15 2 via  0 0 0 1 1 0 1 0 0  . 2 via the isomor- The algebra defined by the first table in (49) is isomorphic to C3 The algebra defined by the last table is also isomorphic to C3 phism via   0 0 1 0 1 0 −1 1 0 are isomorphic to C3  . The algebras defined by the second and third tables 4 in the same way as we see above in (c2). (d) By the above case classification, we have proved that there are exactly 4 defined by the tables 3 and C3 0 , C3 1 , C3 2 , C3 five non-isomorphic curled algebras C3 in (28). 6 Straight algebras Let A be a straight algebra over K of dimension 3, and let h be an element of A such that {h, h2, h3} forms a linear base of A. Then, A is commutative, and we can write (50) h4 = ah3 + bh2 + ch with a, b, c ∈ K. If c 6= 0, let e = 1 c (h3 − ah2 − bh), then he = eh = h. Hence, e is the identity element and A is a unital algebra. Because we already have classified unital algebras in Section 4, we suppose that c = 0. Then (50) becomes h4 = ah3 + bh2. (51) Here, if a = b = 0, then h4 = 0. Let e = h, f = h2 and g = h3, then on the base {e, f, g}, A is defined by f g 0   g 0 0 0 0 0   . (52) Next, suppose that a 6= 0 and b = 0. Then h4 = ah3. Let e = h3 a3 , then we see that hn an = e for all n ≥ 3. Let f′ = h a and g′ = f′2 = h2 a2 . Then, we have e2 = ef′ = f′e = g′2 = eg′ = g′e = f′g′ = g′f′ = e. Set f = f′ − e and g = g′ − e. We thus have ef = ef′ − e2 = 0, eg = eg′ − e2 = 0, f e = f′e − e2 = 0, f 2 = f′2 − f′e − ef′ + e2 = g′ − e = g, f g = f′g′ − f′e − eg′ + e2 = 0 16 and ge = g′e − e2 = 0, gf = g′f′ − g′e − ef′ + e2 = 0, g2 = g′2 − g′e − eg′ + e2 = 0. Hence, A is defined by  e 0  0 Finally, suppose that b 6= 0. Let 0 g 0 0 0 0   . g = h3 − ah2 − bh, then hg = gh = 0 by (51), and so Ag = gA = 0. Let e = (a2 + b)h2 − ah3 b2 , then, by (51) and (54) we have eh2 = h2e = h2, e2 = e and (h2)2 = (a2 + 2b)h2 − b2e. (53) (54) (55) Hence, the subspace B = K{h2, h3} spanned by h2 and h3 is a unital straight algebra of dimension 2 over K generated by the element h2 satisfying (55). Let D = a2 + 4b. If K = C and D 6= 0 or K = R and D > 0, then by the discussion in Section 3, with a suitable element f of B, B is defined by (12) on the base {e, f}. Therefore, A is defined by     . e f 0 f e 0 0 0 0 (56) If K = R and D < 0, then B is defined by (13), and so A is defined by f   e 0 f −e 0  .  0 0 0 If D = 0, B is defined by (14) and A is defined by e f 0   f 0 0 0 0 0   . (57) (58) We denote the algebras defined by (52), (53), (56), (57) and (58) by S3 1 , S3 2 , S3 3 , S3 1 has no nonzero idempotent, but the other algebras have one. Their left annihilators are the same I = Kg which 4 , respectively. The algebra S3 3− and S3 17 is a two-sided ideal. The residue algebras of S3 two-dimensional algebras defined by 2 , S3 3 , S3 4 and S3 3− by I is the (cid:18)e 0 0 0(cid:19) , (cid:18)e f f e(cid:19) , (cid:18)e f f 0(cid:19) and (cid:18)e f f −e(cid:19) , (59) respectively. The algebras defined by the first three tables in (59) are not isomorphic over C to each other. Over R, the algebra defined by the last table is not isomorphic to any other algebras. Therefore, S3 4 are non-isomorphic over C, and S3 3 and S3 3− is another non-isomorphic algebra over R. 1 , S3 2 , S3 7 Waved algebras (strategy) In this section and the following two sections we study waved algebras. Let A be a waved algebra over K of dimension 3 which is not unital. Then, there is a non-curled element f ∈ A, such that {f, f 2, f 3} is linearly dependent, that is, {f, f 2} is a linear base of the subalgebra A′ of A generated by f . Take g ∈ A so that {f, f 2, g} forms a linear base of A. Here, if g is curled, that is, g2 = kg for some k ∈ K, then let g′ = ℓf + g with ℓ ∈ K. Write f g + gf = af + bf 2 + cg with a, b, c ∈ K. Choose ℓ so that ℓ 6= 0 and ℓ 6= −b, then g′ and g′2 = ℓ2f 2 + ℓ(f g + gf ) + g2 = ℓ(ℓ + b)f 2 + ℓaf + (ℓc + k)g are linearly independent, that is, g′ is not curled. Hence, from the beginning we may assume that g is not curled and {g, g2} is a linear base of the subalgebra A′′ of A generated by g. The subalgebras A′ and A′′ are straight algebras of dimension 2, and the intersection A′ ∩ A′′ is a subalgebra of A of dimension 1. From the results in Section 3, we have exactly 4 straight non-isomorphic algebras A2 6 of dimension 2 over C defined by 3, A2 4, A2 (a) (cid:18)0 0 5 and A2 e(cid:19) , (b) (cid:18)e 0 0 0 0(cid:19) , (c) (cid:18)e f f e(cid:19) , and (d) (cid:18)e f f 0(cid:19) , respectively. Over R, in addition to the above algebras, we have the algebra A2 5− defined by (r) (cid:18)e f f −e(cid:19) . Note that these algebras are all commutative. As mentioned in Section 3, A2 5 is also defined by (16). The algebra A2 3 has the unique subalgebra of dimension 1, which is the subalgebra Ke generated by e and is isomorphic to A1 4 has the unique subalgebra Kf of dimension 1 isomorphic to A1 0 and the unique subalgebra Ke of dimension 1 isomorphic to A1 5 has the exactly 3 subalgebras Ke, K(e + f ) and K(e− f ) of dimension 1, which are all isomorphic to A1 5, K(e + f ), Ke and Kf are the three 1. If we choose the table in (16) for A2 1. From easy calculations A2 0. A2 18 subalgebras isomorphic to A1 Kf ∼= A1 0 of dimension 1. A2 1. 1 and 1 of dimension 6 has the exactly 2 subalgebras Ke ∼= A1 1. A2 5− has the unique subalgebra Ke ∼= A1 Our strategy for classifying waved algebras is that we check through all possible combinations of the subalgebras A′ and A′′. Lemma 7.1. Let A′ and A′′ be straight subalgebras of dimension 2 of an algebra A of dimension 3. Let {e, f} and {e′, g} be linear bases of A′ and A′′, respec- tively. Suppose that {e, f, g} forms a linear base of A and A′ ∩ A′′ = Ke = Ke′ holds, that is, e = ke′ for some k ∈ K \ {0}. Then we have (1) eg = ge = 0 if e′g = 0. (2) eg = ge = e if e′g = e′. (3) eg = ge = kg if e′g = g. Proof. (1) We have eg = ke′g = 0 if e′g = 0. Moreover, ge = kge′ = ke′g = 0 because A′′ is commutative from the tables (a), (b), (c), (d) and (r). (2) We have eg = ge = ke′g = ke′ = e if e′g = e′. (3) We have eg = ge = ke′g = kg if e′g = g. Lemma 7.2. Suppose the same situation as in Lemma 7.1. (1) If e2 = 0, then e′2 = 0, and ef, f e, eg, ge ∈ Ke. (2) If e2 = e and e′2 = e′, then k = 1 and e = e′. Proof. (1) Because e = ke′ with k 6= 0, we see that e2 = 0 if and only if e′2 = 0. Because ef ∈ A′ we can write ef = ae + bf with a, b ∈ K. If e2 = 0, we have 0 = e2f = e(ae + bf ) = b(ae + bf ) = abe + b2f. It follows that b = 0 and we see ef ∈ Ke. Similarly, f e ∈ Ke and eg, ge ∈ Ke′ = Ke. (2) If e2 = e and e′2 = e′, we have ke′ = e = e2 = kee′ = k2e′2 = k2e′. Because k 6= 0, we get k = 1 and e = e′. Corollary 7.3. If e2 = e, e′2 = e′, ef = f and e′g = g, then A is unital. Proof. Since A′ is commutative, f e = f . By Lemma 7.2, (2) and Lemma 7.1, (3) we have eg = ge = g. These imply that e is the identity element of A. Lemma 7.4. In the same situation as in Lemma 7.1, let f g = ae + bf + cg and gf = a′e + b′f + c′g with a, b, c, a′, b′, c′ ∈ K. (1) If f 2 = 0, then c = c′ = 0, and moreover, if ef 6= 0 or e2 6= 0, then a = a′ = 0. If g2 = 0, then b = b′ = 0, and moreover, if eg 6= 0 or e′2 6= 0, then a = a′ = 0. (2) If f 2 = e and eg ∈ Ke (in particular e2 = 0), then c = c′ = 0 and eg = aef + be = a′ef + b′e. If g2 = e′ and ef ∈ Ke (in particular e2 = 0), then b = b′ = 0 and ef = kaeg + ce = ka′ef + c′e. (3) If f 2 = f , then c = 0 or c = 1, and c′ = 0 or c′ = 1. If g2 = g, then b = 0 or b = 1, and b′ = 0 or b′ = 1. 19 Proof. We have f 2g = f (ae + bf + cg) = af e + bf 2 + cae + cbf + c2g. (60) (1) If f 2 = 0, then (60) becomes 0 = af e + cae + cbf + c2g. Since af e + cae + cbf ∈ A′ and {e, f, g} is linearly independent, we have c = 0 If ef = 0 and e2 6= 0, then and af e = 0. Hence, a = 0 if ef (= f e) 6= 0. 0 = ef g = e(ae + bf ) = ae2 and we see a = 0. Similarly, we have c′ = 0, and a′ = 0 if ef 6= 0 or e2 6= 0. a = a′ = 0. Similarly, if g2 = 0, then b = b′ = 0, and moreover, if eg 6= 0 or e′2 6= 0, then (2) If f 2 = e, then (60) becomes eg = af e + (b + ca)e + cbf + c2g. Here, if eg ∈ Ke (this holds if e2 = 0 by Lemma 7.2, (1)), we see c2g ∈ A′. Hence, c = 0 and eg = aef + be. Similarly, we have c′ = 0 and eg = ge = a′ef + b′e. The case g2 = e′ is similar. (3) If f 2 = f , then by (60) we have ae + bf + cg = f g = f 2g = af e + ace + b(c + 1)f + c2g. (61) Hence, c(c − 1)g ∈ A′ and we have c = 0 or c = 1. Similarly, we have c′ = 0 or c′ = 1. If g2 = g, then we have ae + bf + cg = f g2 = aeg + abe + c(b + 1)g + b2f, (62) and hence b = 0 or b = 1. Lemma 7.5. In the same situation as in Lemma 7.1, if ef = 0 and e′g = g or if e′g = 0 and ef = f , then f g = gf = 0. Proof. If ef = 0, and e′g = g, then by Lemma 7.1, (3) we have 0 = ef g = f eg = kf g and 0 = gef = kgf. Because k 6= 0, we see f g = gf = 0. The case where e′g = 0 and ef = f is similar. Lemma 7.6. Let B be a straight subalgebra of A of dimension 2 and let g be an element of A\ B such that Bg = gB = {0}. Then, A is a straight algebra, if (1) g2 = g, or (2) g2 = 0 and B is isomorphic to A2 5− . 5, A2 6 or A2 Proof. Because B is straight, it is defined by (a), (b), (c), (d) or (r) above. (1) Suppose that g2 = g. Note that there is an element x in B such that x2−x and x3−x are linearly independent. In fact, f , 2e+f , 2f , 2e+f and f are such elements in the algebras defined by (a), (b), (c), (d) and (r), respectively. For such an element x of B, we claim that x + g is not waved. In fact, if a(x + g) + b(x + g)2 + c(x + g)3 = ax + bx2 + cx3 + (a + b + c)g = 0 20 (2) Suppose that g2 = 0 and B isomorphic to A2 for a, b, c ∈ K, then, a+b+c = 0 because g /∈ B. Hence, b(x2−x)+c(x3−x) = 0. Consequently, a = b = c = 0. Hence, A is straight. 5−. Note that B has an element x such that x2 and x3 are linearly independent. In fact, f , e + f and f are such elements in the algebras defined by (c), (d) and (r), respectively. For such an element x of B, x + g, (x + g)2 = x2 and (x + g)3 = x3 are linearly independent because g /∈ B. Hence, A is straight. 6 or A2 5, A2 8 Waved algebras (enumeration) 0 or A1 6; (c) A′ ∼= A2 3, (ab) A′′ ∼= A2 4 and its subdivisions (bb) A′′ ∼= A2 Let A′ and A′′ be straight subalgebras of dimension 2 of a waved algebra A such that A′ ∩ A′′ is one-dimensional, which is isomorphic to either A1 1. We shall advance the discussion by case division as follows: (a) A′ ∼= A2 3 and its 5 and (ad) A′′ ∼= A2 subdivisions (aa) A′′ ∼= A2 6; (b) A′ ∼= A2 5 and (bd) 5 and (cd) A′′ ∼= A2 A′′ ∼= A2 6; (dd) A′ ∼= A′′ ∼= A2 5− and its subdivisions (ra) A′′ ∼= A2 6 and (rr) A′′ ∼= A2 such that e2 = ef = f e = 0 and f 2 = e. 4, (bc) A′′ ∼= A2 6; and finally in the case where K = R (r) A′ ∼= A2 5, (rd) A′′ ∼= A2 4, (ac) A′′ ∼= A2 5 and its subdivisions (cc) A′′ ∼= A2 (aa) Suppose that A′′ are also isomorphic to A2 3. Let {e, f} be a linear base of A′ 3, and let {e′, g} be linear base of A′′ such that e′2 = e′g = ge′ = 0 and g2 = e′. Because Ke and Ke′ are the unique subalgebras of dimension 1 of A′ and A′′, respectively, we see A′ ∩ A′′ = Ke = Ke′. Hence, e′ = ℓe for some ℓ ∈ K \ {0}. (a) Suppose that A′ is isomorphic to A2 3, (rb) A′′ ∼= A2 4, (rc) A′′ ∼= A2 5− . By Lemma 7.1, (1) we have and by Lemma 7.4, (2) we see eg = ge = 0, f g = ae and gf = a′e for some a, a′ ∈ K. Set g′ = af − g, then we have eg′ = aef − eg = 0 and g′e = af e − ge = 0, g′2 = a2f 2 + g2 − a(f g + gf ) = a2e + ℓe − a(a + a′)e = (ℓ − aa′)e, and f g′ = af 2 − f g = ae − ae = 0 g′f = af 2 − gf = (a − a′)e. f g′ = g′f = 0. Thus, replacing g by g′, A is defined by (66) (aa1) If aa′ = ℓ and a = a′, then by (63) -- (66) we have eg′ = g′e = g′2 = (63) (64) (65) 0 0 0   0 0 e 0 0 0   . 21 (67) (aa2) If aa′ = ℓ and a 6= a′, then replacing g by g′/(a − a′), A is defined by 0 0 0   0 0 0 e e 0   . (68) Now suppose that aa′ 6= ℓ. (aa3) If K = C, or K = R and ℓ − aa′ > 0, then letting g′′ = g′/√ℓ − aa′, we have g′′2 = e, eg′′ = g′′e = f g′′ = 0 and g′′f = a − a′ √ℓ − aa′ e. Hence, replacing g by g′′ and letting k = (a − a′)/√ℓ − aa′, A is defined by 0 0 0 e 0 ke   0 0 e   , (69) where k can be an arbitrary element of K. (aa4) If K = R and ℓ − aa′ < 0, then letting g′′ = g′/√aa′ − ℓ, we have g′′2 = −e, eg′′ = g′′e = f g′′ = 0 and g′′f = a − a′ √aa′ − ℓ e. Hence, replacing g by g′′ and letting k = (a − a′)/√aa′ − ℓ, A is defined by   0 e  0 0 0 0  , 0 ke −e (70) (ab) Suppose that A′′ is isomorphic to A2 4 and let {e′, g} a linear base of A′′ By Lemma 7.1, (1), eg = ge = 0, and by Lemma 7.4, (2) we have f g = ae+bf where k ∈ R. such that e′2 = e′g = ge′ = 0 and g2 = g. Then, A′ ∩ A′′ = Ke = Ke′. with a, b ∈ K and 0 = eg = aef + be = be. Hence, b = 0. Moreover, we have ae = f g = f g2 = aeg = 0. Hence, a = 0 and so f g = 0. Similarly, we have gf = 0. (1), and we exclude this algebra (actually, A is isomorphic to S3 Thus, A′g = gA′ = {0} and g2 = g. Hence, A is not waved by Lemma 7.6, (ac) Suppose that A′′ is isomorphic to A2 5. However, this case is impossible 0 but A′′ because A′ has the unique subalgebra of dimension 1 isomorphic to A1 has only subalgebras of dimension 1 isomorphic to A1 1. 2 ). 22 (ad) Suppose that A′′ is isomorphic to A2 6, and let {e′, g} be a linear base of A′′ such that e′2 = 0, e′g = ge′ = e′ and g2 = g. Because Ke′ is the unique subalgebra of A′′ of dimension 1 isomorphic to A1 0, we have Ke = Ke′. By Lemma 7.1, (2) we have eg = ge = e, and by Lemma 7.4, (2) we have f g = ae + bf with a, b ∈ K and e = eg = aef + be = be, Hence, b = 1, that is, f g = ae + f . We have ae + f = f g = f g2 = (ae + f )g = 2ae + f. Hence, a = 0 and we see f g = f . Similarly, we have gf = f. These imply that g is the identity element and A is unital, and we can exclude this algebra. (b) Suppose that A′ is isomorphic to A2 4. (bb) Suppose that A′′ is also isomorphic to A2 4. (bb1) Suppose that A′ ∩ A′′ ∼= A1 1. Let {e, f} and {e′, g} be linear bases of A′ and A′′ such that e2 = e, ef = f e = f 2 = 0 and e′2 = e′, e′g = ge′ = g2 = 0. Then, A′∩A′′ = Ke = Ke′. By Lemma 7.1, (1) we have eg = ge = 0. Moreover, by Lemma 7.4, (1), we see f g = gf = 0. Therefore, A is defined by   (bb2) Suppose that A′ ∩ A′′ ∼= A1 e 0 0 . 0 0 0 0 0 0   (71) 0. Let {e, f} and {e′, g} be linear bases of A′ and A′′ such that e2 = ef = f e = 0, f 2 = f and e′2 = e′g = ge′ = 0, g2 = g. Then, A′ ∩ A′′ = Ke = Ke′. By Lemma 7.1, (1) we have eg = ge = 0. Let f g = ae + bf + cg with a, b, c ∈ K, then by Lemma 7.4, (3) we have c = 0 or c = 1, and b = 0 or b = 1. If c = 0, then by (61), we have ae = af e = 0. Hence a = 0. If c = 1, then by (61) we have bf = −af e = 0. Hence b = 0. Because ae = aeg = 0 by (62), again we have a = 0. Therefore, f g = bf + cg. Similarly, we have gf = b′f + c′g with b′, c′ ∈ K. Thus B = k{f, g} is a subalgebra of A of dimension 2 such that eB = Be = {0} and e2 = 0. Hence, B is curled or isomorphic to A2 4, otherwise A is straight by Lemma 7.6, (2). Because f 2 = f and g2 = g, B cannot be 4, and the only possibility is that B ∼= A2 1 or B ∼= A2 isomorphic to A2 2. Therefore, choosing a suitable base {f, g} of B, A can be defined by 3 nor A2 3 or A2 0 0 0   0 0 f 0 0 g   or   0 0 0 0 0 0 0 f g   . (72) (bc) Suppose that A′′ is isomorphic to A2 5. Let {e, f} be a linear base of A′ such that e2 = e and ef = f e = f 2 = 0. Then, A′ ∩ A′′ = Ke ∼= A1 1. As observed in the previous section, A′′ has exactly three subalgebras (1) Ke, (2) 23 K(e + f ) and (3) K(e− f ) isomorphic to A1 have the following three possiblities. 1 for the table (12). Accordingly we (bc2) Replacing e′ by e+f 2 (bc1) Let {e′, g} be a linear base of A′′ such that e′2 = g2 = e′ and e′g = ge′ = g. Suppose that A′∩A′′ = Ke = Ke′. By Lemma 7.2, (2) and Lemma 7.1, (3), we have e′ = e and eg = ge = g. By Lemma 7.5 we have f g = f g = 0. 5, and f A′′ = A′′f = {0} and Because the subalgebra A′′ is isomorphic to A2 f 2 = 0, A is not waved by Lemma 7.6, (2). and g by e−f 2 , let {e′, g} be a linear base of A′′ such that e′2 = e′, g2 = g and e′g = ge′ = 0 as in (16), and suppose that Ke = Ke′. By Lemma 7.1, (1) we have eg = ge = 0, and by Lemma 7.4, (1) we see f g = bf and gf = b′f with b, b′ ∈ K. Therefore, B = K{f, g} is a subalgebra of dimension 2. Because eB = Be = {0} and e2 = e, B must be curled by Lemma 7.6, (1), because otherwise A would be straight. Because g2 = g, B is isomorphic to either A2 2. Hence, choosing a suitable base {f, g} of B, A is defined by  e 0 0 0  0 f   or   1 or A2 (73) 0 0 g e 0 0 0 0 0 f g 0   (bc3) Similarly to (bc2), let {e′, g} be a linear base of A′′ such that e′2 = e′, g2 = g and e′g = ge′ = 0. Suppose that Ke = Kg. Because the automor- phism of A′′ interchanging e′ and g maps the subalgebra Kg onto the subalgebra Ke′, the situation is essentially the same as (bc2) and we can omit this case. (bd) Suppose that A′′ is isomorphic to A2 6. (bd1) Suppose that A′ ∩ A′′ ∼= A1 0. Let {e, f} be a linear base of A′ such that e2 = ef = f e = 0 and f 2 = f , and {e′, g} be a linear base of A′′ such that e′2 = 0, e′g = ge′ = e′ and g2 = g. Then, A′ ∩ A′′ = Ke = Ke′. By Lemma 7.1, (2) we have eg = ge = e. Let f g = ae + bf + cg with a, b, c ∈ K. By Lemma 7.4, (3) we see b = 0 or b = 1. We have 0 = ef g = e(ae + bf + cg) = ce. Hence, c = 0. We have ae + bf = f g = f 2g = f (ae + bf ) = bf. Hence, a = 0, and we see f g = bf , where b = 0 or b = 1. Similarly, gf = b′f and b′ = 0 or b′ = 1. Because bf = bf 2 = f gf = b′f 2 = b′f, we see b = b′. Therefore, we see f g = gf = 0 or f g = gf = f . Consequently, A is defined by 0 0 0 f e 0   e 0 g   or   0 0 e 0 f f e f g   . 24 (74) (bd2) Suppose that A′ ∩ A′′ ∼= A1 1. Let {e, f} be a linear base of A′ such that e2 = e and ef = f e = f 2 = 0. Let {e′, g} be a linear base of A′′ such that e′2 = e′, e′g = ge′ = g and g2 = 0. Then, Ke = Ke′. Then by Lemma 7.1, (3) and Lemma 7.2, (2) we have eg = ge = g. Moreover, by Lemma 7.5, we see f g = gf = 0. Because the subalgebra A′′ is isomorphic to A2 6, and f A′′ = A′′f = {0} and f 2 = 0, A is not waved by Lemma 7.6, (2). (c) Suppose that A′ is isomorphic to A2 5. (cc) Suppose that A′′ is also isomorphic to A2 5. As mentioned in (bc), A′′ has exactly three subalgebras (1) Ke, (2) K(e+f ) and (3) K(e−f ) isomorphic to A1 1 for (12). Therefore, A′ ∩ A′′ ∼= A1 1 and we have the three possible combinations: (1) and (1), (1) and (2), and (2) and (2), where (3) is similar to (2) and so can be neglected. Accordingly, we have the following three cases. (cc1) Let {e, f} and {e′, g} be linear bases of A′ and A′′, respectively such that e2 = f 2 = e, ef = f e = f , e′2 = g2 = e′ and e′g = ge′ = g. If A′ ∩ A′′ = Ke = Ke′, then A is unital by Corollary 7.3, and we can exclude this case. (cc2) Let {e, f} and {e′, g} be linear bases of A′ and A′′, respectively such that e2 = f 2 = e, ef = f e = f , e′2 = e′, g2 = g and e′g = ge′ = 0. Suppose that Ke = Ke′. By Lemma 7.1, (1) we have eg = ge = 0, and by Lemma 7.5 we see f g = gf = 0. Thus, A is not waved by Lemma 7.6, (1), because the subalgebra A′ is straight, and gA′ = A′g = {0} and g2 = g. (cc3) Let {e, f} a linear base of A′ with e2 = e, f 2 = f and ef = f e = 0, and let {e′, g} be a linear base of A′′ with e′2 = e′, g2 = g and e′g = ge′ = 0. Suppose that A′ ∩ A′′ = Ke = Ke′, By Lemma 7.1, (1) we see eg = ge = 0. Let f g = ae + bf + cg with a, b, c ∈ K. We have 0 = ef g = e(ae + bf + cg) = ae, 1 or A2 (cd) Suppose that A′′ is isomorphic to A2 and hence a = 0, that is, f g = bf + cg. Similarly, gf = b′f + c′g with b′, c′ ∈ K. Hence, B = {f, g} is a subalgebra of A of dimension 2 such that eB = Be = {0}. Because e2 = e, B must be curled by Lemma 7.6, (1). Because f 2 = f , B is isomorphic to A2 2. Thus, choosing a suitable base {f, g} of B, A is defined by one of the tables in (72), and we get no new algebra here. 1. Let {e′, g} be a linear base of A′′ such that e′2 = e′, e′g = ge′ = g and g2 = 0. Because Ke′ 1, we see A′∩A′′ = Ke′. is the unique subalgebra of dimension 1 isomorphic to A1 (cd1) Let {e, f} be a linear base of A′ such that e2 = f 2 = e and ef = f e = f , (cd2) Let {e, f} be a linear base of A′ such that e2 = e, f 2 = f and ef = f e = 0. Suppose that A′∩A′′ = Ke = Ke′. By Lemma 7.1, (3) and Lemma 7.2, (2) we have eg = eg = g, and by Lemma 7.5 we see f g = gf = 0. But A is not waved by Lemma 7.6, (1), because the subalgebra B = K{e, g} is straight, f B = Bf = {0} and f 2 = f . and suppose that Ke = Ke′. Then by Corollary 7.3, A is unital. 6. Then, A′ ∩ A′′ ∼= A1 The case A′ ∩ A′′ = Kf is symmetric, which we can omit. (dd) Suppose that both A′ and A′′ are isomorphic to A2 the previous section, A2 Therefore we have the following two cases. 6 has the exactly 2 subalgebras Ke ∼= A1 6. As observed in 1 and Kf ∼= A1 0. 25 (dd1) Suppose that A′ ∩ A′′ ∼= A1 1. Let {e, f} and {e′, g} be linear bases of A′ and A′′, respectively such that e2 = e, ef = f e = f and f 2 = 0, and e′2 = e′, e′g = ge′ = g and g2 = 0. Then, A′∩A′′ = Ke = Ke′. By Corollary 7.3, A, is unital and we exclude this case. (dd2) Suppose that A′ ∩ A′′ ∼= A1 0. Interchanging e and f , and e′ and g, let {e, f} and {e′, g} be linear bases of A′ and A′′ such that e2 = e′2 = 0, f 2 = f, ef = f e = e, g2 = g and e′g = ge′ = e′. Then, A′ ∩ A′′ = Ke = Ke′. Then, By Lemma 7.1, (2) we have eg = ge = e. Let f g = ae + bf + cg with a, b, c ∈ K. We have e = eg = ef g = e(ae + bf + cg) = (b + c)e. It follows that We have b + c = 1. (75) ae + bf + cg = f g = f f g = f (ae + bf + cg) = ae + bf + c(ae + bf + cg). Hence, Similarly, we have a = a(c + 1), b = b(c + 1) and c2 = c. a = a(b + 1), c = c(b + 1) and b2 = b, (76) (77) By (75) -- (77) we see that a = 0, and b = 1, c = 0 or b = 0, c = 1, that is, f g = f or f g = g. Similarly, we see gf = f or gf = g. Thus, A is defined by e f g e f g e f f e f g 0 e e     ,   0 e e e f g 5− . Let {e, f} be a linear base of A′ such that e2 = −f 2 = e and ef = f e = f . Then Ke is the unique subalgebra of A′ of dimension 1, which is isomorphic to A1 1. 3 has (r) Finally, suppose that K = R and A′ is isomorphic to A2 3. However, this case is impossible because A2   or     ,   0 e e e g g   . e f f e g g (78) 0 e e (ra) Suppose that A′′ ∼= A2 (rb) Suppose that A′′ ∼= A2 the unique subalgebra of dimension 1 isomorphic to A1 0. 4. Let {e′, g} be a linear base of A′′ such that e′2 = e′ and e′g = ge′ = g2 = 0. Then, A′ ∩ A′′ = Ke = Ke′. We have eg = ge = 0, by Lemma 7.1, (1) and f g = gf = 0 by Lemma 7.5. Because gA′ = A′g = {0} and g2 = 0, A is not waved by Lemma 7.6, (2). (rc) Suppose that A′′ ∼= A2 5. (rc1) Let {e′, g} be a linear base of A′′ such that e′2 = g2 = e′ and e′g = ge′ = g. Suppose that A′ ∩ A′′ = Ke′. By Corollary 7.3 A is unital, and we exclude this case. (rc2) Let {e′, g} be a linear base of A′′ such that e′2 = e′, g2 = g and e′g = ge′ = 0. Suppose that A′ ∩ A′′ = Ke = Ke′. We have eg = eg = 0 by Lemma 7.1, (1) and f g = gf = 0 by Lemma 7.5. Since gA′ = A′g′ = {0} and g2 = g, A is not waved by Lemma 7.6, (1) 26 (rd) Suppose that A′′ = A2 6. Let {e′, g} be a linear base of A′′ such that e′2 = e′, e′g = ge′ = g and g2 = 0. Then A′ ∩ A′′ = Ke = Ke′. By Corollary 7.3 A is unital. We exclude this case. 5− . Let {e′, g} be a linear base of A′′ such that e′2 = −g2 = e′ and e′g = ge′ = g. Then, Ke = Ke′. By Lemmas 7.1 and 7.2 we have eg = ge = g. Let f g = ae + bf + cg. We have (rr) Suppose that A′′ is also isomorphic to A2 −g = −eg = f 2g = f (ae + bf + cg) = af − be + c(ae + bf + cg) = (ac − b)e + (bc + a)f + c2g. Hence, c2 = −1, but there is no such c in R. 9 Waved algebras (screening) In the previous section, we have enumerated all possible waved algebras of dimension 3. In this section, we shall investigate whether each two of them are isomorphic or not. (aa1) Let W 3 1 denote the algebra defined by (67). Interchanging f and g in the base {e, f, g}, W 3 1 is defined by 0 0 0   0 0 0 0 e 0   . (aa2) Next, let W 3 2 be the algebra defined by (68). Let f′ = f − g, then we have and ef′ = ef − eg = 0, f′e = f e − ge = 0, f′2 = f 2 − f g − gf + g2 = 0, f′g = f g − g2 = 0, gf′ = gf − g2 = e. Hence, replacing f by f′, W 3 2 is defined by 0 0 0   0 0 0 0 e 0   . (aa3) We denote the algebra defined by (69) by W 3 3 (k) for k ∈ K. First, we 3 (k) via the isomorphism sending e to e, f 3 (−k) is isomorphic to W 3 see that W 3 to f and g to −g. W 3 φ : W 3 Because Ke is the left annihilator of W 3 that is, To prove the converse we may assume that K = C, because if W 3 3 (k) and 3 (ℓ) are not isomorphic over C, they are not isomorphic over R. Assume that 3 (ℓ) is an isomorphism. Let e′ = φ(e), f′ = φ(f ) and g′ = φ(g). 3 (ℓ)), we see Ke = Ke′, 3 (k) (and of W 3 3 (k) → W 3 e′ = me 27 for some m ∈ K \ {0}. Write f′ = ze + xf + yg and g′ = we + uf + vg with x, y, z, u, v, w ∈ K. Then, we have me = e′ = φ(e) = φ(f 2) = f′2 = (x2 + y2 + kxy)e, me = e′ = g′2 = (u2 + v2 + kuv)e 0 = f′g′ = (xu + yv + kyu)e and mℓe = ℓe′ = g′f′ = (xu + yv + kxv)e. Thus replacing x, y, u and v by x/√m, y/√m, u/√m and u/√m, respectively, we have x2 + y2 + kxy = 1, u2 + v2 + kuv = 1, xu + yv + kyu = 0 (79) and xu + yv + kxv = ℓ. (80) Because we have the identity (y2 − u2)v = u2v(x2 + y2 + kxy − 1) + uy(x + ky)(u2 + v2 + kuv − 1) +(y − uvx − u2y − kuvy)(xu + yv + kyu), we find by (79) v(y − u)(y + u) = 0 If v = 0, then by (79) we have u2 = 1, x + ky = 0, y2 = 1 and x2 = k2. Hence, by (80) If y = u, then by (79) ℓ = xu = ±k. (x + v + ku)u = 0. Here, if u = 0, then y = 0 and x2 = v2 = 1 by (79). Hence, by (80) If x + v + ku = 0, then ℓ = kxv = ±k. ℓ = xu + uv + kxv = −(v + ku)(u + kv) + uv = −k(u2 + v2 + kuv) = −k Thus, we have proved that W 3 by (79) and (80). If y = −u, we can show ℓ = ±k in a similar manner. ℓ = ±k. 3 (k) and W 3 3 (ℓ) are isomorphic if and only if 28 We shall show that W 3 (aa4) We denote the algebra defined by (70) by W 3 3− (k) for k ∈ R. For any 3− (k) and W 3 k ∈ R, W 3 3− (−k) are isomorphic via the isomorphism sending e to e, f to f and g to −g. 3 (k′) for any reals k and k′ over R. If they were isomorphic, we would have elements f′, g′ ∈ W 3 3− (k) such that f′2 = g′2 6= 0 and f′g′ = 0. Let f′ = ze + xf + yg and g′ = we + uf + vg with x, y, z, u, v, w ∈ R. Then we have 3− (k) is not isomorphic to W 3 x2 − y2 + kxy = u2 − v2 + kuv 6= 0 and xu − yv + kyu = 0 (81) If y = 0, then by (81), x2 = u2 − v2 + kuv 6= 0 and xu = 0. Hence, u = 0 and (82) x2 = −v2(6= 0) The left-hand side of (82) is positive but the right-hand side is negative, a contradiction. If y 6= 0, then by (81) v = (x+ky)u x2 − y2 + kxy = u2 − (x+ky)2u2 and y y2 + k(x+ky)u2 y2 (x2 − y2 + kxy) 6= 0. y = − u2 3− (k) and W 3 3 (k′) can not be isomorphic. This is again impossible, and W 3 On the other hand, W 3 formation matrix   3− (k) is isomorphic to W 3  1 0 0  . Hence, for distinct positive numbers k and k′, 0 1 0 0 0 −i 3 (ik) over C via the trans- 3− (k) and W 3 3 (ik) and W 3 W 3 are not isomorphic, and hence they are not isomorphic over R, a fortiori. 3−(k′) are not isomorphic over C, because W 3 3 (ik′) (bb1) We denote the algebra defined by the table in (71) by W 3 4 . (bb2) The algebras defined by the first table and the second table in (72) are denoted by W 3 5 and W 3 6 , respectively. (bc2) We denote the algebras defined by the first and the second tables in (73) by W 3 7 and W 3 8 , respectively. (bd1) The algebra defined by the first table in (74) is not waved. In fact, e + f − g, (e + f − g)2 = −2e + f + g and (e + f − g)3 = 3e + f − g are linearly independent. The algebra defined by the second table in (74) is defined by the first one replacing g by g − f . (dd2) The algebra defined by the first (resp. last) table in (78), g (resp. f ) is the identity element. Hence, they are unital. Let W 3 10) be the algebra defined by the second (resp, third) table in (78). In the second table, let g′ = g − f , the we have eg′ = g′e = f g′ = g′2 = 0 and g′f = g′. Thus, replacing g by g′, W 3 9 (resp. W 3 9 is defined by 0 e 0   e f g 0 0 0   . 29 In the third table, let g′ = g − f , then we have eg′ = g′e = g′f = g′2 = 0 and f g′ = g′. Replacing g by g′, W 3 10 is defined by 0 e 0   e f 0 0 g 0   . 4 , W 3 2 , W 3 1 , W 3 6 , W 3 5 , W 3 In summary, any non-unital waved algebra A of dimension 3 over C is isomorphic to W 3 3 (k) with k ∈ H = {x + yi ∈ C x > 0 or x = 0, y ≥ 0}. Over R, A can be isomor- phic to W 3 3−(k) with k ≥ 0 other than above. We shall show these algebras are not isomorphic to each other. As we already showed that W 3 3 (k′) are 3− (k′) are not isomorphic not isomorphic for distinct k, k′ ∈ H, W 3 over R for any distinct nonnegative k and k′. Moreover, W 3 3−(k′) are not isomorphic over R for any nonnegative k and k′. 3−(k) and W 3 3 (k) and W 3 3 (k) and W 3 10 or W 3 7 , W 3 8 , W 3 9 , W 3 We use the same notations as in Section 5 for an algebra A over K; la(A) = {x ∈ A xA = 0}, ra(A) = {x ∈ A Ax = 0}, and α(A) = dimK A2, β(A) = dimK la(A), γ(A) = dimK ra(A). We see α(W 3 1 ) = α(W 3 2 ) = α(W 3 3 (k)) = α(W 3 3− (k)) = α(W 3 4 ) = 1, α(W 3 5 ) = α(W 3 6 ) = 2 and α(W 3 7 ) = α(W 3 8 ) = α(W 3 9 ) = α(W 3 10) = 3. Moreover, β(W 3 1 ) = β(W 3 2 ) = β(W 3 4 ) = β(W 3 5 ) = 2, β(W 3 3 (k)) = β(W 3 3−(k)) = β(W 3 6 ) = β(W 3 7 ) = β(W 3 10) = 1, β(W 3 8 ) = β(W 3 9 ) = 0, γ(W 3 1 ) = γ(W 3 2 ) = γ(W 3 4 ) = γ(W 3 6 ) = 2, and γ(W 3 3 (k)) = γ(W 3 3− (k)) = γ(W 3 5 ) = γ(W 3 8 ) = γ(W 3 9 ) = 1, γ(W 3 7 ) = γ(W 3 10) = 0. These values are summarized in the table shown below. 10 W 3 9 W 3 8 W 3 7 W 3 6 W 3 5 W 3 4 W 3 3 (k) W 3 2 W 3 1 W 3 W 3 3− (k) 1 3 3 3 3 2 2 1 1 1 1 1 1 0 0 1 1 2 2 1 2 2 2 2 1 2 1 2 0 1 1 0 1 α β γ This table implies that W 3 3 (k), W 3 1 , W 3 2 and W 3 any of the others. W 3 other group. W 3 commutative but W 3 2 is not isomorphic to W 3 1 and W 3 2 is not. W 3 5 and W 3 3− (k), W 3 6 are not isomorphic to 4 are not isomorphic to any algebra in the 4 are 4 has a 1 nor to W 3 4 are not isomorphic because W 3 4 because W 3 1 and W 3 30 9 and W 3 10 have two zeropotent elements e and g; e2 = g2 = 0 but W 3 1 has no nonzero idempotent. Finally, W 3 nonzero idempotent e; e2 = e but W 3 7 , 10 are not isomorphic to any algebra in the other group. W 3 W 3 8 , W 3 9 and W 3 7 and W 3 8 have no two zeropotent elements that are linearly independent. Hence, these algebras are not isomorphic to each other. We have proved that all the algebras are not isomorphic to each other. 10 Summary By summarizing the results in Sections 4, 5, 6 and 9, we list up all the multipli- cation tables of three-dimensional algebras. It is easy to see that all the tables satisfy associativity, that is, (eiej)ek = ei(ejek) for all i, j, k ∈ {1, 2, 3} where e1 = e, e2 = f and e3 = g. 2 , U 3 Over C, we have, up to isomorphism, exactly 5 unital algebras U 3 1 , U 3 0 , U 3 3 , U 3 4 respectively, exactly 5 non-unital curled algebras C3 0 , C3 4 defined by defined by e f g   f 0 0 g 0 0   ,   e f f 0 g −f g f e   ,   e 0 0 0 f 0 0 0 g   ,   0 e 0 f 0 g 0 0 0   0 0 0 0 0 0 0 0   ,  0 0 0 e  0 −e 0   ,   0 e 0 0 f g 0 0 0   ,   0 0 e 0 0 f respectively, exactly 4 non-unital straight algebras S3 f g 0   g 0 0 0 0 0   ,   e 0 0 0 g 0 0 0 0   ,   e f 0 f e 0 0 0 0 f g 0 g 0 0   , 0 g 0 e f g   ,   3 , C3 2 , C3 1 , C3   ,   3 , S3 0 0 g 0 0 0 0 e 0 f 0 g   , 4 defined by 2 , S3 1 , S3   ,   1 , W 3 e f 0 f 0 0 0 0 0   , respectively, and exactly 9 non-unital waved algebras W 3 W 3 10 defined by 8 , W 3 2 , W 3 4 , W 3 5 , W 3 6 , 7 , W 3   0 0 0 9 , W 3   ,   0 0 0 0 0 e 0 0 0 0 0 0 0 e 0   ,   0 e 0 0 0 0 0 0 0   ,   0 0 0 0 0 f 0 0 g   ,   0 0 0 0 0 0 f 0 g   , 0 e 0 0 0 g 0 0 0 e f g 0 f g   e 0 0 0 0 0 e 0 0 0 0 f   ,     ,     ,    ,  respectively and one infinite family (cid:8)W 3 3 (k)(cid:9)k∈{x+yi x>0 or x=0,y≥0}   . unital waved algebras defined by 0 0 0 e 0 ke   e f 0 0 g 0 0 e 0 0 0 e of non- 31 Over R, in addition to the above algebras, we have one unital algebra U 3 one non-unital straight algebra S3 non-unital waved algebras defined by 3− and one infinite family (cid:8)W 3 2−, 3− (k)(cid:9)k≥0 of e 0 0   0 0 f g g −f respectively. f   ,    ,  e 0 f −e 0   0 0 0 0 e  0 0 0 0  , 0 ke −e We remark that zeropotent algebras are only C3 0 and C3 1 which are alternative matrices. An algebra is indecomposable, if it is not isomorphic to a direct sum of nontrivial subalgebras. We will not go into the details, but it is possible to show 3 , C3 that the algebras U 3 9 , W 3 10 are indecomposable, and the others are not. 3− (k), W 3 3 (k), W 3 1 , W 3 2 , W 3 1 , C3 0 , U 3 1 , U 3 4 , C3 2 , C3 4 , S3 Peirce [5] listed five families of "pure" algebras of dimension 3. They corre- spond to our U4, S1, W3(k), W2 and C1. The list of unilal algebras of dimension 3 given by Scheffer [6] and Study [8] is in accordance with our list. References [1] W. R. Hamilton, On quaternions; or a new system of imaginaries in algebra, The London, Edinburgh and Dublin Philosophical Magazine and Journal of Science 25 (1844), 489 -- 495. [2] H. E. Hawkes, On hypercomplex number systems, Trans. Amer. Math. Soc., 3 (1902), 312 -- 330. [3] Y. Kobayashi, K. Shirayanagi, S.-E. Takahasi and M. Tsukada, Classifica- tion of three-dimensional zeropotent algebras over an algebraically closed field, Comm. Algebra, Vol. 45, Iss. 12 (2017), 5037 -- 5052. [4] A. N. Kolmogorov and A.P. Yushkevich, Mathematics of the 19th cen- tury, Mathematical Logic, Algebra, Number Theory, Probability Theory, Birkhauser, 1992. [5] B. Peirce, Linear associative algebra, Amer. J. Math., 4 (1881), 97 -- 229. [6] G. Scheffers, Zuruckfuhrung complexer zahlensysteme auf typische formen, Math. Ann. 39 (1891), 293 -- 390. [7] K. Shirayanagi, S.-E. Takahasi, M. Tsukada and Y. Kobayashi: Classifi- cation of three-dimensional zeropotent algebras over the real number field, Comm. Algebra, to appear. [8] E. Study, Uber systeme complexer zahlen und ihre anwendung in der the- orie der transformationsgruppen, Monatsh. Math. u. Phisik 1 (1890), 283 -- 354. 32 [9] H. Taber, On hypercomplex number systems, Trans. Amer. Math. Soc., 5 (1904), 509 -- 548. [10] B. L. van der Waerden, A history of algebra, Springer, 1985. [11] B. Watson, trans. The Analects of Confucius, New York: Columbia Uni- versity Press, 2007. Y. Kobayashi, K. Shirayanagi, and M. Tsukada Department of Information Science, Toho University, Miyama 2-2-1 Funabashi, Chiba 274-8510, Japan [email protected], [email protected], and [email protected] u.ac.jp S.-E. Takahasi Laboratory of Mathematics and Games, Katsushika 2-371 Funabashi, Chiba 273-0032 Japan sin [email protected] 33
1805.07664	2	1805	2018-10-01T15:16:15	Generating adjoint groups	[ "math.RA" ]	We prove two approximations of the open problem of whether the adjoint group of a non-nilpotent nil ring can be finitely generated: We show that the adjoint group of a non-nilpotent Jacobson radical cannot be boundedly generated, and on the other hand construct a finitely generated, infinite dimensional nil algebra whose adjoint group is generated by elements of bounded torsion.	math.RA	math	GENERATING ADJOINT GROUPS BE'ERI GREENFELD Abstract. We prove two approximations of the open problem of whether the adjoint group of a non-nilpotent nil ring can be finitely generated: We show that the adjoint group of a non-nilpotent Jacobson radical cannot be boundedly generated, and on the other hand construct a finitely generated, infinite dimensional nil algebra whose adjoint group is generated by elements of bounded torsion. 1. Introduction Nil rings give rise to multiplicative groups, called adjoint groups. If R is nil (or, more generally Jacobson radical) then its adjoint group R◦ consists of the elements of R as underlying set, with multiplication given by r ◦ s = r + s + rs. We can think of R◦ as {1 + rr ∈ R} with usual multiplication (1 + r)(1 + s) = 1 + r + s + rs. The connections between group-theoretic properties of R◦ and ring-theoretic properties of R were intensively studied (for instance, see [7, 8, 9, 20]). In this way, nil rings give rise to braces (in fact, two-sided braces, namely braces arising from Jacobson radical rings). For more information about braces and their connections with nil rings see [18, 19, 20]. The following is an open question posed by Amber, Kazarin, Sysak [5, 6] and then repeated by Smoktunowicz [19]: Question 1.1. Is there a non-nilpotent, nil algebra whose adjoint group is finitely generated? Note that by [22, Theorem 4.3], if R is Jacobson radical such that R◦ is generated by two elements then R is nilpotent. Bounded generation is a stronger notion of finite generation: it asserts that a group is generated by a finite set {g1, . . . , gn} such that every element can be expressed as gi1 n , namely, all elements have finite width with respect to the generating set. Boundedly generated groups are extensively studied, and for related information and results we refer the reader to [1, 2, 11, 16, 17, 21]. 1 · · · gin It is easy to see that the adjoint group of a non-nilpotent nil algebra cannot be boundedly generated; in Theorem 2.1 we prove that the adjoint of a non-nilpotent Jacobson radical cannot be boundedly generated. Suppose R is a nil algebra of positive characteristic p > 0, then its adjoint group R◦ is residually-p torsion. However, its torsion cannot be bounded, since (unless the algebra is nilpotent) its elements' nilpotency indices are unbounded. In Theorem 3.2 we construct a finitely generated, infinite dimensional nil ring (in fact, a Golod-Shafarevich algebra) whose adjoint group is generated by elements of bounded torsion. Indeed, if the answer to Question 1.1 is affirmative then it The author thanks Prof. Agata Smoktunowicz for helpful discussions. 1 2 BE'ERI GREENFELD evidently yields such example; in this sense, Theorem 3.2 should be seen as an approximation of Question 1.1. Our example also results in an infinite, finitely generated brace whose adjoint group is generated by a set of elements of bounded torsion. Note that the anal- ogous question of Question 1.1 for braces has a positive solution: for any finite, non-degenerate involutive solution (X, r) of the Yang-Baxter equation there is an associated brace G(X, r), which is infinite, both of whose multiplicative and addi- tive groups are finitely generated and also its adjoint group is; note that the additive group of G(X, r) is a free abelian group (our two-sided braces have p-torsion addi- tive groups). For details, see [12, Theorem 4.4]. 2. Boundedly generated adjoint groups Recall that a group is boundedly generated if G = H1 · · · Hn for some cyclic subgroups H1, . . . , Hn ⊆ G. The minimal such n is called the cyclic width of G. Note that the adjoint group of an infinite dimensional nil algebra cannot be boundedly generated, since that would mean the algebra admits a Shirshov base, namely every element would be a linear combination of monomials of the form f i1 1 · · · f in n for some f1, . . . , fn; since the algebra is nil, there are only finitely many such monomials, so the algebra must be finite dimensional. We now extend this result to an arbitrary Jacobson radical. Note that Jacobson radicals need not be nil, so the above argument does not hold any longer. Theorem 2.1. Let R be an algebra which is Jacobson radical. Then R◦ is boundedly generated if and only if R is finite dimensional and nilpotent. Proof. First observe that we only have to deal with the case where the base field F is finite. Indeed, observe by Nakayama's lemma that R/R2 is a non-zero F -vector space. Therefore the additive group of R/R2, being isomorphic to (R/R2)◦ occurs as a homomorphic image of R◦. Hence, if R◦ is finitely generated then so is the additive group of the base field F , from which it follows that F must be finite. We henceforth assume F = Fp. Assume that R is an algebra which is Jacobson radical and R◦ is boundedly generated with cyclic width m. Denote Gn = (Rn+1)◦ ⊳ R◦. Note that Gn has finite index since R/Rn+1 is finite F -dimensional. Given a finite group G denote by exp(G) the minimum power α such that gα = e for all g ∈ G. Observe that exp(R◦/Gn) ≤ p(n + 1) since given f ∈ R, we have that: (1 + f )p⌈logp (n+1)⌉ = 1 + f p⌈logp (n+1)⌉ ∈ Gn (as: f p⌈logp(n+1)⌉ ∈ Rn+1 and n + 1 ≤ p⌈logp(n+1)⌉ ≤ p(n + 1).) Note that if a group G is boundedly generated of cyclic width m then for any finite index normal subgroup H we have that [G : H] ≤ exp(G/H)m and in partic- ular in our case: pdimFp (R/Rn+1) = [R◦ : Gn] ≤ exp(R◦/Gn)m ≤ (p(n + 1))m. It follows that (denoting R0 = R): n X i=0 dimFp(Ri/Ri+1) = dimFp (R/Rn+1) ≤ m(1 + logp(n + 1)) < n GENERATING ADJOINT GROUPS 3 for n ≫ 1, hence for some 0 ≤ i ≤ n we have that R · Ri = Ri+1 = Ri. Since R is a finitely generated algebra which is Jacobson radical it follows by Nakayama's lemma that Ri = 0. Hence R is a finite dimensional, nilpotent algebra. (cid:3) Remark 2.2. Note that in fact, Theorem 2.1 requires a weaker assumption on R◦ than being boundedly generated: if R is a Jacobson radical whose adjoint group is a polynomial index growth (PIG) group, namely group in which the index of any subgroup is polynomially bounded by its relative exponent, then R is nilpotent. For more information and results on PIG groups, see [10, 13, 14, 17]. 3. Adjoint group generated by a uniformly torsion set Let F be a countable field of characteristic p > 0. Consider the free algebra n=0 A(n) is graded by deg(x) = deg(y) = 1. A = F hx, yi. Observe that A = L∞ Denote A+ = L∞ Lemma 3.1. Let a ∈ A+ and write a = a1 + · · · + an a sum of homogeneous elements. For every m ≥ 1 there exist some homogeneous elements h1, . . . , hk and an element b ∈ A+ with v(b) ≥ m such that: n=1 A(n). For a ∈ A, let v(a) = max{da ∈ Li≥d A(i)}. 1 + a + b = (1 + h1) · · · (1 + hk). Proof. Assume on the contrary that b ∈ A+ is such that 1+a+b = (1+h1) · · · (1+hk) for some homogeneous h1, . . . , hk with d = v(b) maximal. Write b = b1 + · · · + bm a sum of homogeneous elements. By maximality of d we get that d = min{deg(bi)1 ≤ i ≤ m}. Note that: (1 + h1) · · · (1 + hk)(1 − b1) · · · (1 − bm) = (1 + a + b)(1 − b1) · · · (1 − bm) = (1 + a + b)(1 − b + c) = 1 + a + (c + ac + bc − ab − b2). with: c = X S⊆{1,...,m};S>1 bi. Y i∈S Observe that v(ab), v(b2), v(c) > d and hence v(c + ac + bc − ab − b2) > d, contra- dicting the maximality of d. (cid:3) Theorem 3.2. There exists a finitely generated, graded, infinite dimensional nil algebra R whose adjoint group R◦ is generated by a set of elements of bounded torsion. Proof. Enumerate A+ = {f1, f2, . . . }. Using Lemma 3.1 pick b1 with v(b1) ≥ 14 such that 1 + f1 + b1 = (1 + h1,1) · · · (1 + h1,k1) for some homogeneous elements h1,1, . . . , h1,k1. Write b1 = b1,1 + · · · + b1,n1 a sum of homogeneous elements with deg(b1,d) = d. Next use Lemma 3.1 to choose b2 with v(b2) ≥ n1+1 and 1+f2+b2 = (1 + h2,1) · · · (1 + h2,k2) for some homogeneous elements h2,1, . . . , h2,k2 . Again, write b2 = b2,n1+1 + · · · + b2,n2 a sum of homogeneous elements with deg(b1,d) = d, and inductively proceed, each time constructing bl+1 with v(bl+1) ≥ nl + 1 such that 1 + fl+1 + bl+1 = (1 + hl+1,1) · · · (1 + hl+1,kl+1 ) for suitable homogeneous elements hl+1,1, . . . , hl+1,kl+1 . Let I ⊳ A+ be the ideal generated by {bi,ji ≥ 1}. Note that I is generated by at most one homogeneous element of each degree, starting from 14. 4 BE'ERI GREENFELD Now let J be the ideal generated by hpα (with α = ⌈logp 7⌉) for all homogeneous elements h ∈ A+. Observe that J is generated by 2d elements of degree pαd ≥ 7d for d ≥ 1. We claim that A+/(I + J) is a Golod-Shafarevich algebra. Indeed, let rn denote the number of degree n generators of I + J. Then for all τ > 0 for which the series are convergent: rnτ n ≤ X n>1 ∞ X d=1 2dτ 7d + X n=3 τ n = 2τ 7 1 − 2τ 7 + τ 14 1 − τ . Setting f (τ ) = 1 − 2τ + Pn>1 rnτ n, we get that: f (0.75) ≈ −0.5 + 0.3642 + 0.071 < 0, hence A+/(I+J) is a Golod-Shafarevich algebra. Note that the semigroup (A+/(I + J))◦ is generated by the set {1 + f f homogeneous}, whose elements are torsion of order ≤ pα. Now by [23] it follows that A+/(I + J) admits a homomorphic image which is an infinite dimensional and nil algebra, say R. Then R satisfies the claimed properties. (cid:3) We conclude with two open questions related to Question 1.1: 4. Further questions Question 4.1. Is there an infinite dimensional, finitely generated nil algebra with polynomial growth whose adjoint group is finitely generated? In [3] it is proved that every elementary amenable subgroup of the adjoint group of a nil ring is locally nilpotent. Question 4.2. Is there an infinite dimensional, finitely generated nil algebra R whose adjoint group R◦ is amenable? Is it possible that R◦ has intermediate growth? References [1] M. Abert, A. Lubotzky, L. Pyber, Bounded generation and linear groups, Int. J. Alg. Comp. 13 4 401-413 (2003). [2] S. I. Adian, J. Mennicke, On bounded generation of SLn(Z), Int. J. Alg. Comp. 2 357 -- 365 (1992). [3] B. Amberg, O. Dickenschied, Ya. P. Sysak, Subgroups of the adjoint group of a radical ring, Canad. J. Math., 50 3-15 (1998). [4] B. Amberg, O. Dickenschied, On the adjoint group of a radical ring, Canad. Math. Bull. 38 262-270 (1995). [5] B. Amberg, L. Kazarin, Nilpotent p-algebras and factorized p-groups, Proceedings of "Groups St. Andrews 2005" , London Math. Society Lecture Note Series 339, Vol 1, Cambridge Univ. Press (2007), 130 -- 147. [6] B. Amberg, Ya. P. Sysak, Radical rings and products of groups, Groups St Andrews 1997 in Bath: Edited by C. M. Campbell, 1 -- 19. [7] B. Amberg, Ya. P. Sysak, Radical rings with Engel conditions, J. Algebra 231 364-373 (2000). [8] B. Amberg, Ya. P. Sysak, Associative rings with metabelian adjoint group, J. Algebra 277 456-473 (2004). [9] B. Amberg, Ya. P. Sysak, Radical rings with soluble adjoint group, J. Algebra 247 692-702 (2002). [10] A. Balog, A. Mann, L. Pyber, Polynomial index growth groups, Int. J. Alg. Comp. 10 773-782 (2000). GENERATING ADJOINT GROUPS 5 [11] D. Carter, G. Keller, Bounded elementary generation of SLn(O), Amer. J. Math. 105 673 -- 687 (1983). [12] F. Ced´o, E. Jespers, J. Okni´nski, Braces and the Yang-Baxter equation, Comm. Math. Pyhs., 327 (1) 101 -- 116 (2014). [13] A. Lubotzky, D. Segal, Subgroup Growth, Progress in Math. 212,Birkhauser, Basel, 2003. [14] A. Mann, D. Segal, Uniform finiteness conditions in residually finite groups, Proc. London Math. Soc. (3) 61 529-545 (1990). [15] A. Muranov, Diagrams with selection and method for constructing boundedly generated and boundedly simple groups, Comm. Algebra, 33 (4) 1217 -- 1258 (2005). [16] N. Nikolov, B. Sury, Bounded generation of wreath products, J. Group Theory 1433-5883 (2015). [17] L. Pyber, D. Segal, Finitely generated groups with polynomial index growth, Journal fur die reine und angewandte Mathematik, Volume 2007, Issue 612, Pages 173 -- 211 (2007). [18] W. Rump, Modules over braces, Algebra Discrete Math., Issue 2, 2006, 127-137. [19] A. Smoktunowicz, A note on set-theoretic solutions of the Yang-Baxter equation, J. Algebra, to appear (2016). [20] A. Smoktunowicz, On Engel groups, nilpotent groups, rings, braces and the Yang-Baxter equation, Trans. AMS, to appear (2017). [21] B. Sury, Bounded generation does not imply finite presentation, Comm. Alg. 25 1673-1683 (1997). [22] Ya. P. Sysak, The adjoint group of radical rings and related questions, In: Ischia Group Theory 2010 (Ischia), 344-365, World Scientific, Singapore 2011. [23] J. Wilson, Finite presentations of pro-p groups and discrete groups, Invent. Math. 105 (1) 177 -- 183 (1991).
0907.1717	2	0907	2010-03-24T00:55:29	Connes-Kreimer quantizations and PBW theorems for pre-Lie algebras	[ "math.RA", "math-ph", "math-ph", "math.QA" ]	The Connes-Kreimer renormalization Hopf algebras are examples of a canonical quantization procedure for pre-Lie algebras. We give a simple construction of this quantization using the universal enveloping algebra for so-called twisted Lie algebras (Lie algebras in the category of symmetric sequences of k-modules). As an application, we obtain a simple proof of the (quantized) PBW theorem for Lie algebras which come from a pre-Lie product (over an arbitrary commutative ring). More generally, we observe that the quantization and the PBW theorem extend to pre-Lie algebras in arbitrary abelian symmetric monoidal categories with limits. We also extend a PBW theorem of Stover for connected twisted Lie algebras to this categorical setting.	math.RA	math	CONNES-KREIMER QUANTIZATIONS AND PBW THEOREMS FOR PRE-LIE ALGEBRAS by Travis Schedler Abstract. -- The Connes-Kreimer renormalization Hopf algebras are examples of a canonical quantization procedure for pre-Lie algebras. We give a simple construction of this quantization using the universal enveloping algebra for so-called twisted Lie algebras (Lie algebras in the cat- egory of symmetric sequences of k-modules). As an application, we obtain a simple proof of the (quantized) PBW theorem for Lie algebras which come from a pre-Lie product (over an arbitrary commutative ring). More generally, we observe that the quantization and the PBW theorem extend to pre-Lie algebras in arbitrary abelian symmetric monoidal categories with limits. We also extend a PBW theorem of Stover for connected twisted Lie algebras to this categorical setting. R´esum´e. -- Les alg`ebres de Hopf de Connes-Kreimer, utlis´ees en renormalisation, sont des exem- ples de proc´ed´es de quantification canoniques pour les alg`ebres pr´e-Lie. On donne une construction simple de cette quantification en utilisant l'alg`ebre enveloppante universelle des "alg`ebres de Lie tordues" (alg`ebres de Lie dans la cat´egorie des modules sym´etriques). Comme application on ob- tient une d´emonstration simple du th´eor`eme PBW (quantifi´e) pour les alg`ebres de Lie issues d'un produit pr´e-Lie (sur un anneau de base commutatif quelconque). Plus g´en´eralement, on observe que la quantification et le th´eor`eme de PBW s'´etendent aux alg`ebres pr´e-Lie dans n'importe quelle cat´egorie symm´etrique monoidale ab´elienne avec limites. On ´etend aussi un th´eor`eme de Stover pour les alg`ebres de Lie tordues connexes dans ce contexte cat´egorique. Contents 1 1. Introduction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2. Theorem 1.1.4 using the graded PBW theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3. Pre-Lie algebras as twisted Lie algebras. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4. Generalizations and full proof of Theorem 1.1.4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 5. Graded PBW theorems in a unified categorical context. . . . . . . . . . . . . . . . . . . . . 20 A. PBW counterexamples and pre-Lie identities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 1. Introduction 1.1. Connes-Kreimer renormalization algebras. -- Connes and Kreimer introduced [CK98] a renormalization Hopf algebra to organize computations involved in certain Feyn- man diagram expansions. The dual of this Hopf algebra is given by (Sym g, ∆, ∗) where g is 2000 Mathematics Subject Classification. -- 17D99, 17B35. Key words and phrases. -- pre-Lie algebras, PBW theorems, renormalization, Hopf algebras, twisted Lie algebras, S-modules. the vector space (or k-module for k an arbitrary commutative ring) freely generated by rooted trees, ∆ is the standard coproduct on Sym g, and ∗ satisfies, for x ∈ Syma g, y ∈ Symb g, (1.1.1) x ∗ y = xy + terms in degrees a, a + 1, . . . , a + b − 1. In particular, if x and y are forests of rooted trees (i.e., monomials in Sym g), then x ∗ y is the sum of all ways of grafting the trees of y to distinct branches of x (or simply adding the trees to the forest without grafting). We will not be concerned further with the specific formula. Connes and Kreimer observed in [CK98] that (Sym g, ∆, ∗) ∼= U g, the universal enveloping algebra of g equipped with the bracket {x, y} = x∗y − y ∗x, as filtered Hopf algebras. Chapoton and Livernet further noted in [CL01] that g is not just a Lie algebra but a (right) pre-Lie algebra (and, in fact, a free pre-Lie algebra). Pre-Lie algebras generalize associative algebras; as defined in [Ger63, Vin63], they consist of a multiplication ◦ on g satisfying (1.1.2) x ◦ (y ◦ z) − (x ◦ y) ◦ z = x ◦ (z ◦ y) − (x ◦ z) ◦ y. As in the associative case, every pre-Lie algebra (g, ◦) as above has an associated Lie bracket, (1.1.3) {x, y} := x ◦ y − y ◦ x. In [OG05, OG08], Oudom and Guin produced from an arbitrary pre-Lie algebra an explicit, interesting multiplication ∗ such that (Sym g, ∆, ∗) ∼= U g, and in [GS08], this construction was used to prove the following theorem (stated dually in [GS08, Proposition 3.5.2]): Theorem 1.1.4. -- [GS08] Let k be any commutative ring. Star products ∗ on Sym g satis- fying the conditions (i) ∗ forms a bialgebra with the usual coproduct ∆, (ii) ∗ is a filtered product whose associated graded is the usual product on Sym g, (iii) ∗ satisfies (1.1.5) (Symm g) ∗ Sym g ⊆ (Symm g)(Sym g) = Sym≥m g, are equivalent to (right) pre-Lie algebra structures ◦ : g ⊗ g → g, under the correspondence (1.1.6) x ∗ y = xy + x ◦ y, ∀x, y ∈ g. Under this equivalence, the dual to the renormalization Hopf algebra, on the nose, is obtained from the much simpler pre-Lie algebra g spanned by rooted trees. (Although this Hopf algebra is, by [CK98], canonically isomorphic to U g, the precise star product and the isomorphism with U g are not defined merely by the Lie bracket on g, but require the pre-Lie structure). 1.2. Star product formulas. -- The difficult part of the proof of the theorem is the con- struction of ∗ from an arbitrary pre-Lie algebra, which was done in [OG05, OG08] using complicated explicit computations, based on the following formulas related to rooted trees: (0) a ◦ 1 = a, (1) a ◦ (bx) = (a ◦ b) ◦ x − a ◦ (b ◦ x), (2) (ab) ◦ c = (a ◦ c′)(b ◦ c′′), (3) a ∗ b = (a ◦ b′)b′′, ∀a, b ∈ Sym g, x ∈ g, To deduce a general formula for ∗, (0), (1), and (2) first extend ◦ to a certain binary operation on all of Sym g (which is not a pre-Lie multiplication) by induction on degree, and then (3) expresses ∗ using this. In parts (2) and (3), we use Sweedler notation ∆(a) = a′ ⊗ a′′, which is shorthand forPi a′ i ⊗ a′′ i , for some a′ i, a′′ i ∈ Sym g. 2 The original goal of this paper was to give an alternative, more conceptual proof of Theorem 1.1.4, avoiding complicated calculations with the above formulas. We succeed in this (in §2) assuming that the graded PBW theorem, Sym g ∼= gr U g, holds for (g, { , }). In full generality, we later prove the theorem as a consequence of the stronger Theorem 4.3.1. The above formulas (0) -- (3) follow immediately from Theorem 1.1.4, as we explain now: Notation 1.2.1. -- Let πn : Sym g → Symn g be the projection to degree n. Proposition 1.2.2. -- Let g be any k-module. Given any star product ∗ on Sym g satisfying (i), (ii), and (iii) from Theorem 1.1.4, define ◦ : Sym g ⊗ Sym g → Sym g by g ◦ h := πn(g ∗ h) for all g ∈ Symn g, h ∈ Sym g, extended linearly. Then, formulas (0) -- (3) above hold. The reader not interested in the following proof can safely skip it. Proof. -- (0) Since a ∗ 1 = a, this follows immediately. (1) If a ∈ Symn g, then a ◦ (bx) = πn(a ∗ (b ∗ x − b ◦ x)) = πn((a ∗ b) ∗ x − a ∗ (b ◦ x)) = (a ◦ b) ◦ x − a ◦ (b ◦ x). (2),(3) We can combine these into the single formula (1.2.3) (ab) ∗ c = (a ◦ c′)(b ◦ c′′)c′′′. This identity is obvious in the case that c ∈ g, so inductively assume it holds for c ∈ Sym≤n g. The inductive step follows since, for c and d of degrees between 1 and n, (ab) ∗ (c ∗ d) = ((ab) ∗ c) ∗ d = ((a ◦ c′)(b ◦ c′′)c′′′) ∗ d = ((a ◦ c′) ◦ d′)((b ◦ c′′) ◦ d′′)(c′′′ ◦ d′′′)d′′′′ = (a ◦ (c′ ∗ d′))(b ◦ (c′′ ∗ d′′))(c′′′ ∗ d′′′). 1.3. PBW theorems. -- Theorem 1.1.4 has the following interesting corollary (which the author did not find mentioned in the literature): Corollary 1.3.1. -- (i) (Pre-Lie graded PBW theorem) If g is the associated Lie algebra of a pre-Lie algebra over an arbitrary commutative ring k, then U g is a filtered Hopf algebra such that Sym g ∼→ gr U g by the canonical map. (ii) (Pre-Lie quantum PBW theorem)(1) The map lifts to a coalgebra isomorphism Sym g ∼→ U g. The pre-Lie assumption has a different flavor from the assumptions of the classical PBW theorems, which impose conditions on the k-module structure of g rather than on its Lie struc- ture. In particular, if g is the associated Lie algebra of an arbitrary associative algebra, then the PBW theorem holds for g, regardless of its k-module structure. This result already seems hard to find in the literature.(2) Note that, in our one-page proof in §2 of Theorem 1.1.4, we actually assume the graded PBW theorem, which is part (i) of the corollary above, or alternatively work under an assumption on the k-module structure of g that implies this, as below. However, we later give a proof which avoids such an assumption, using Theorem 4.3.1, which relies instead on Stover's graded (1)We use here the terminology "quantum PBW" since part (ii) says in particular that U g is a filtered quantization of Sym g (i.e., U g is an associative algebra such that gr U g ∼= Sym g as Poisson algebras). (2)Note that an associative algebra can be viewed as either a right or a left pre-Lie algebra, and the latter corresponds to using the opposite multiplication of ∗, which gives a different explicit quantum PBW isomorphism Sym g → U g. This seems to indicate that this method of proving quantum PBW is not entirely natural for associative algebras. It is tempting to look for an abstract proof, especially since the result extends to associative algebras in general categorical settings (§4.7), but we couldn't find it. 3 PBW theorem [Sto93] in the category of S-modules that imposes no condition on the k-module structure. This is one motivation for the material on S-modules that form the heart of this work: it gives a route (different from [OG05, OG08]) to prove the above PBW theorems even in case of k-modules for which (i) does not necessarily hold, or is not known to hold. The standard contexts in which the PBW theorem is known include (cf. [Hig69]): (1) g is a free k-module [Poi00, Bir37, Wit37] (or, more generally, a direct sum of cyclic modules); (2) k ⊇ Q [Coh63]. Moreover, in these cases, the quantum PBW theorem holds, using an explicit lift Sym g ∼= U g of the graded PBW isomorphism: in case (1), one may obtain a PBW coalgebra isomorphism Sym g ∼= U g using explicit bases for free (or cyclic) k-modules, and in case (2) one has the symmetrization map Sym g ∼→ U g of coalgebras, sending x1 · · · xn to 1 xσ(1) · · · xσ(n). (See §5 for sketches of proofs.) These approaches do not seem to apply to the case of general pre-Lie algebras over an arbitrary commutative ring. However, in these contexts, the proof of §2 of Theorem 1.1.4 suffices (and one obtains a generally different lift of the isomorphism Sym g ∼= gr U g to a coalgebra isomorphism Sym g ∼→ U g than the above). n!Pσ∈Sn Remark 1.3.2. -- There are many other cases where at least the graded PBW theorem holds, although it is no longer clear whether the quantum PBW theorem holds. For example, because the graded PBW theorem holds when k ⊇ Q, it must hold more generally when g is torsion-free over Z, because one can tensor with Q. As another example, if all localizations of g at prime ideals of k are direct sums of cyclic modules (or torsion-free over Z), or direct limits thereof, then the graded PBW theorem must hold. For instance, this happens whenever k is a Dedekind domain (which is a third classical case where the graded PBW theorem holds, attributed to [Laz54, Car58]), or whenever g is a flat k-module. Remark 1.3.3. -- In the appendix, we recall an example from [Coh63] where the graded PBW isomorphism fails (and k is an Fp-algebra). Such examples for p = 2 are even older: see, e.g., [Sir53, Car58]. 1.4. Categorical generalization. -- In §4.7 below, we observe that the above construction makes sense in an arbitrary symmetric monoidal category (which is abelian with arbitrary limits), and the star-product and (quantum) PBW theorems therefore hold in this generality. In particular, the associated Lie algebra of any pre-Lie algebra (or associative algebra) object satisfies the PBW theorem. Moreover, to prove this theorem, we prove a categorical generalization of Stover's graded PBW theorem, replacing k-modules by an arbitrary symmetric monoidal category as above. Hence, connected twisted Lie algebras are replaced by Lie algebras in the category of symmetric sequences of objects of such a category. Although we defer the precise explanations and definitions to that section, it is worth pointing out a simple case where this is nontrivial: Example 1.4.1. -- If we work in the category of Z/2-graded modules over k equipped with the super braiding (x ⊗ y 7→ (−1)xyy ⊗ x), the resulting Lie algebras are commonly called Lie superalgebras. When such a Lie superalgebra g is a free k-module, then the graded PBW isomorphism Sym g ∼→ gr U g holds if and only if {x, x} = 0 for all even x ∈ g and {x, {x, x}} = 0 for all odd x ∈ g (cf. Remark 5.0.12). Note that these conditions do not hold for all Lie superalgebras, and the second condition is not even true for all Lie superalgebras for which 4 {x, y} is the antisymmetrization of a binary operation. However, in a pre-Lie superalgebra, the pre-Lie axiom implies that 2x ◦ (x ◦ x) = 2(x ◦ x) ◦ x for all odd x ∈ g, and hence {x, {x, x}} = 0. 1.5. Twisted algebras. -- To prove Theorem 1.1.4 without assumptions on g and k such as (1) or (2) of §1.3, we exploit a connection between pre-Lie algebras and twisted Lie algebras, which should be interesting in its own right. Here, a twisted Lie algebra (in the sense of, e.g., [Bar78, Joy86])(3) is defined as a Lie algebra object in a certain category which replaces that of k-modules. The category is that of S-modules (otherwise known as symmetric sequences of k- modules, or species). This category is well known: for example, k-linear operads are a different type of monoidal object in this category, and similar categories are used to define various types of spectra in topology. We recall more precisely the definition of this category in §3.2 below, and speak informally in this section for the benefit of the reader not familiar with these notions. Our main observation is that pre-Lie algebra structures on a k-module g are equivalent to twisted Lie algebra structures on the suspension Σg (which places the k-module g in degree one rather than zero), in a certain sense that we will explain. Using this, Theorem 1.1.4 is a "quantization" of a standard type of statement (Proposition 3.4.4) that twisted Poisson algebra structures on the symmetric algebra SymS Σg are equivalent to twisted Lie algebra structures on Σg. More precisely, "quantizing" this equivalence yields a strengthened theorem (4.3.1), which circumvents the need for the graded PBW theorem for g, and proves Theorem 1.1.4 (as well as the pre-Lie PBW theorem) in full generality. The proof uses Stover's graded PBW theorem [Sto93] valid for all connected twisted Lie algebras, which applies to Σg (rather than g). The equivalence between pre-Lie algebra structures on g and twisted Lie algebra structures on Σg, as well as the resulting quantization procedure, generalizes from k-modules g to arbitrary S-modules (§4.5) and even symmetric monoidal categories (§4.7), which implies in particular quantum PBW theorems for pre-Lie algebras in these contexts (as promised in §1.4 above). In the case of S-modules, this requires passing to S-bimodules. We give an alternative ap- proach in this setting that involves taking the suspension Σg in a more careful way, remaining in the realm of S-modules, while still implying the analogue of Theorem 1.1.4 (§4.6). 1.6. Outline of paper. -- The main contributions of this paper are the following: 1. To give a simple proof of Theorem 1.1.4 (§2) using the graded PBW isomorphism (this proof does not require S-modules or twisted algebras); 2. To point out the connection between pre-Lie algebras and twisted Lie algebras (§3), and use this to prove a strengthening of the theorem (Theorem 4.3.1) without any assumptions on g or k; 3. To generalize the above results and observations to the case where g is a twisted pre-Lie algebra, or a pre-Lie algebra in an arbitrary abelian symmetric monoidal category with limits (§§4.5 -- 4.7); 4. To sketch a simple proof of a categorical generalization of Stover's twisted graded PBW theorem as well as the usual graded PBW theorems in a unified context (§5). In the appendix, we recall the PBW counterexamples from [Coh63] and remark that a pre-Lie identity [Tou06] which generalizes a classical p-th power identity of Zassenhaus explains why they do not extend to the pre-Lie setting (in accordance with Corollary 1.3.1.(i)). (3)Twisted Lie algebras are old in topology, and predate these references. 5 1.7. Acknowledgements. -- This work grew out of an attempt to understand and improve [OG05, OG08]. I am grateful to M. Livernet for useful discussions, as well as pointing out the main references including op. cit, and for many helpful corrections and suggestions. I am grateful to M. Van den Bergh for useful discussions, and to my Ph.D. advisor, V. Ginzburg, for his guidance. I also thank M. Ronco for answering questions about [Ron07], and J.-M. Oudom for answering questions about [OG05, OG08] and providing revisions. I am very grateful to the anonymous referee for helpful suggestions, and in particular pointing out the references [Coh63, Rev77, Hig69]. Finally, I would like to thank the participants and organizers of the 2009 CIRM conference on operads for the opportunity to present this work and for their helpful comments and questions. This work was supported by the University of Chicago Mathematics Department's VIGRE grant and a five-year AIM fellowship. 2. Theorem 1.1.4 using the graded PBW theorem This section will not require the notion of S-module or twisted algebras. Here, we prove Theorem 1.1.4 under the assumption that the graded PBW isomorphism Sym g ∼→ gr(U g) holds for a given pre-Lie algebra g (e.g., if g is a free k-module, k ⊇ Q, or k is a Dedekind domain), by inductively constructing a lift to a coalgebra isomorphism Sym g ∼→ U g, that has the needed properties. 2.1. Proof of Theorem 1.1.4. -- It follows immediately from Proposition 1.2.2, specifically formula (1) in §1.2, that, given a star product ∗ as in the theorem, (1.1.6) yields a pre-Lie structure on g. Thus, it remains to show that any pre-Lie algebra (g, ◦) admits a unique star product ∗ on Sym g satisfying (i) -- (iii). By Proposition 1.2.2 again, uniqueness is immediate, so it suffices to show that such a star product ∗ exists. Let (g, ◦) be a pre-Lie algebra. We also let g denote the associated Lie algebra with bracket {x, y} := x ◦ y − y ◦ x. We prove the theorem by constructing a coalgebra isomorphism Φ : Sym g → U g (thereby simultaneously proving Corollary 1.3.1.(ii)). We construct Φ in- ductively on degree, such that gr(Φ) is the graded PBW morphism (which is assumed to be an isomorphism), and such that the induced star-product ∗ on Sym g satisfies (1.1.6) and (1.1.5). (One may also notice that Φ extends uniquely in each degree.) We now begin the inductive construction of Φ. In degree 1, set Φ(x) = x for all x ∈ g. Inductively, begin with a coalgebra morphism Φ≤n−1 : Sym≤n−1 g → U g such that gr Φ≤n−1 is the graded PBW morphism, and such that the product (2.1.1) ∗ : Mi+j≤n−1 Symi g ⊗ Symj g → Sym≤n−1 g defined by Φ(a ∗ b) = Φ(a) · Φ(b) satisfies (1.1.5). (Here · is the product in U g). 6 We will extend Φ≤n−1 to Φ≤n : Sym≤n g → U g satisfying the same conditions. Note that, applying condition (2) of Proposition 1.2.2 repeatedly with c ∈ g,(4) Φ≤n will need to satisfy (2.1.2) Φ≤n−1(x1x2 · · · xn−1)xn = Φ≤n((x1x2 · · · xn−1) ∗ xn) = Φ≤n(x1x2 · · · xn) + Φ≤n−1(cid:16)n−1Xi=1 x1x2 · · · xi−1(xi ◦ xn)xi+1 · · · xn−1(cid:17), ∀x1, x2, . . . , xn ∈ g. By linearity of Φ, setting the LHS to the RHS uniquely extends Φ≤n−1 to Φn. We must check that the formula is well-defined, by showing that the resulting expression for Φ≤n(x1x2 · · · xn) is symmetric in the variables. It is obviously symmetric in x1, x2, . . . , xn−1, so it suffices to check that it is symmetric under permuting xn−1 and xn (this is the main step of the proof). To do this, by the induction hypothesis, (2.1.2) is equivalent to (2.1.3) Φ≤n−1(x1x2 · · · xn−2)xn−1xn = Φ≤n(((x1x2 · · · xn−2) ∗ xn−1) ∗ xn), by further expanding the RHS using the formula (2) of Proposition 1.2.2 with c ∈ g. So, to prove the symmetry of xn−1, xn, it suffices to show that (2.1.4) Φ≤n−1(x1x2 · · · xn−2){xn−1, xn} = Φ≤n(((x1x2 · · · xn−2)∗xn−1)∗xn−((x1x2 · · · xn−2)∗xn)∗xn−1), for all x1, . . . , xn ∈ g. This may be rewritten as the claim: (2.1.5) (x1x2 · · · xn−2)∗{xn−1, xn} = ((x1x2 · · · xn−2)∗xn−1)∗xn −((x1x2 · · · xn−2)∗xn)∗xn−1. By expanding the LHS and RHS using (2.1.2) (i.e., formula (2) of Proposition 1.2.2), and substituting {xn−1, xn} = xn−1 ◦ xn − xn ◦ xn−1, this follows from the pre-Lie identity (1.1.2). Next, we show that Φ≤n is a morphism of coalgebras. This follows from the fact that Φ≤n−1 is a morphism of coalgebras, using (for a, b homogeneous of positive degree such that ab = n) (2.1.6) ∆(Φ≤n(a ∗ b)) = ∆(Φ≤n−1(a)Φ≤n−1(b)) = ∆(Φ≤n−1(a))∆(Φ≤n−1(b)) ≤n−1(∆(b)) = Φ⊗2 ≤n−1(∆(a))Φ⊗2 = Φ⊗2 ≤n(∆(a) ∗ ∆(b)). It remains to show that (2.1.2) defines a product ∗ such that Symi g ∗ Symn−i g ⊂ Sym≥i g. This follows by definition for n − 1 ≤ i ≤ n. For every 1 ≤ i ≤ n − 2, it suffices to show that (2.1.7) (x1x2 · · · xi) ∗ (xi+1 ∗ (xi+2xi+3 · · · xn)) ∈ Sym≥i g, ∀x1, . . . , xn ∈ g. Using reverse induction on i and associativity of ∗, the statement follows immediately. 3. Pre-Lie algebras as twisted Lie algebras In this section, we explain our main observation (Proposition 3.3.3) connecting pre-Lie al- gebras with twisted Lie algebras (whose definition we recall). Along the way, we will use the notion of (twisted) Poisson algebras, although in the end Proposition 3.3.3 does not require it. (4)This is the only part of Proposition 1.2.2 that we need, and it also follows immediately from applying a single coproduct. That is, we will not really need the precise formula for ∗, unlike [OG05, OG08]. (Even the uniqueness of ∗ follows from the existence argument without requiring Proposition 1.2.2, if we are slightly more careful.) 7 3.1. Preliminaries and motivation. -- Let k be an arbitrary commutative ring. All unadorned tensor products, symmetric algebras, tensor algebras, and so on will be assumed to be over k. We will use in this section Roman letters (e.g., V ) for (graded) k-modules, to avoid confusion with the S-modules we will discuss in subsequent sections (which we will denote by Fraktur letters, except for twisted associative or commutative algebras). Observe that a Lie algebra structure on a k-module V is equivalent to a Poisson algebra structure of degree −1 on the commutative algebra Sym V , i.e., a Poisson bracket { , } : Sym V ⊗ Sym V → Sym V satisfying (3.1.1) {V, V } ⊆ V. We may consider also a homogeneous version of the above construction. Roughly, we intro- duce a parameter t of degree one, and define a new bracket {v, w} = t · {v, w}old, for v, w ∈ V . Precisely, let V [t] := V ⊗ k[t] be the free k[t]-module generated by V . By a Lie algebra structure on V [t] over k[t], we mean a Lie algebra structure that is k[t]-linear, i.e., such that {tf, g} = t{f, g} for all f, g. We consider V [t] as graded with t = 1 = V . Next, form the commutative algebra Symk[t](V [t]) ∼= Sym V ⊗ k[t], equipped with the total grading such that t = 1 = V . We consider graded Poisson structures on Symk[t](V [t]) over k[t] (i.e., such that t is central) which satisfy (3.1.2) {V [t], V [t]} ⊆ V [t], i.e., {V, V } ⊆ V t. Such structures are canonically equivalent to graded Lie algebra structures on V [t] over k[t], which are in turn the same as ordinary Lie algebra structures on V . (3.1.3) An alternative and useful construction is to set bV := V ⊕ hti, and notice that In these terms, we are interested in graded Poisson structures on SymkbV such that the second Symk[t](V [t]) ∼= SymkbV . condition of (3.1.2) holds. Remark 3.1.4. -- Without the condition (3.1.2), graded Poisson structures on Symk[t] V [t] over k[t] are the same as filtered Poisson brackets on Sym V of degree ≤ 0. These are determined by their restriction to V ⊗ V → ht2i ⊕ tV ⊕ Sym2 V , yielding a skew-symmetric form on V , a Lie bracket, and a quadratic Poisson bracket on V , satisfying certain compatibility conditions. 3.2. Twisted algebras and S-modules. -- A central observation of this paper is that pre- Lie algebras arise as the twisted version of the above construction. In this subsection we recall the needed preliminaries. Definition 3.2.1. -- An S-module is a N-graded k-module g = Lm≥0 gm together with an action of Sm on gm by k-module automorphisms. Note that an S-module concentrated in degree zero is the same as an ordinary k-module. A "twisted" (commutative, Lie, Poisson) algebra is a (commutative, Lie, Poisson) algebra in the category of S-modules rather than k-modules. To make this precise, one equips the category of S-modules with the structure of a symmetric monoidal category (see, e.g., [JS93]), using the following well known formulas: g ⊗S h :=Mp Mm+n=p IndSp Sm×Sn gm ⊗ hn, β : g ⊗S h ∼→ h ⊗S g, β(g ⊗ h) = (12)h,g(h ⊗ g), (3.2.2) (3.2.3) 8 where g ∈ g and h ∈ h are homogeneous of degrees g and h, respectively, and (12)h,g ∈ Sg+h is the permutation which swaps the two blocks {1, 2, . . . , h} and {h+1, h+2, . . . , g+ h}, i.e., (12)h,g(a) =(a + g, a − h, (3.2.4) if 1 ≤ a ≤ h, if h + 1 ≤ a ≤ g + h. Then, a twisted (commutative, Lie, Poisson) algebra can be defined by first rewriting the usual definition in terms of binary operations g ⊗ g → g satisfying certain diagrams, and then replacing all tensor products by ⊗S. For example, a twisted commutative algebra is an S- module A equipped with a binary operation µ : A ⊗S A → A which is an S-module morphism, is associative, and commutative in the sense that µ = µ ◦ β. Without using symmetric monoidal categories, we may write the necessary definitions explicitly as follows: Notation 3.2.5. -- For brevity, all explicit elements we write of S-modules will be assumed to be homogeneous without saying so. In particular, whenever we write x for x an element of an S-module, x is assumed to be homogeneous (of degree x). Definition 3.2.6. -- A twisted associative algebra is an S-module A = Lm≥0 Am which is also a graded associative algebra such that Am ⊗An → Am+n is a morphism of Sm ×Sn ⊆ Sm+n- modules. A twisted commutative algebra is a twisted associative algebra such that the multiplication satisfies the identity (3.2.7) xy = (12)y,xyx. A twisted Lie algebra is an S-module g equipped with a binary operation { , } : g ⊗ g → g satisfying (3.2.8) (3.2.9) {x, y} = −(12)y,x{y, x}, {x, {y, z}} + (123)y,z,x{y, {z, x}} + (132)z,x,y{z, {x, y}} = 0. A twisted Poisson algebra is an S-module which is equipped with both a twisted commutative and a twisted Lie algebra structure, satisfying the Leibniz rule, (3.2.10) {xy, z} = x{y, z} + (12)y,x,zy{x, z}. In (3.2.9) and (3.2.10) we used the following generalization of (3.2.4): Notation 3.2.11. -- Given any i1, i2, . . . , in ≥ 0 with sum i1 + · · · + in = r, and any per- mutation σ ∈ Sn, define the element σi1,i2,...,in ∈ Sr to be σ applied to the blocks [1, i1], [i1 + 1, i1 + i2], . . . , [i1 + i2 + · · · + in−1 + 1, i1 + i2 + · · · + in = r]. Precisely, σ : [1, r] → [1, r] is the permutation sending each interval [i1 + · · · + ij−1 + 1, i1 + · · · + ij] onto [iσ−1(1) + iσ−1(2) + · · · + iσ−1(σ(j)−1) + 1, iσ−1(1) + iσ−1(2) + · · · + iσ−1(σ(j)−1) + ij], preserving order. In other words, (12)i,j,k = (12)i,j × Idk, (23)i,j,k = Idi ×(23)j,k, (123)i,j,k = (12)i,k,j(23)i,j,k, (132)i,j,k = (23)j,i,k(12)i,j,k, and so forth. 3.3. Twisted Lie algebras and pre-Lie algebras. -- In this subsection we explain the connection between pre-Lie algebras and twisted Lie algebras. Heuristically, pre-Lie algebras are equivalent to "twisted Lie algebras concentrated in degree one." As stated, this doesn't make sense because any graded, let alone twisted, (Lie) algebra concentrated in degree one is trivial; we will fix this by introducing a parameter t as before. 9 Given an S-module g concentrated in degree zero, i.e., just a k-module V := g0, one may form a corresponding S-module concentrated in degree one, Σg, given by (Σg)1 = V and (Σg)m = 0 for m 6= 1. We call this the suspension of g. Given x ∈ g, we abusively denote the corresponding element of Σg also by x. One reason why this suspension is useful is the interesting but simple fact that, if h is an S-module concentrated in degree one, e.g., h = Σg as above, then (3.3.1) SymS h ∼= Tkh, where the notation Tk means that we are taking the tensor algebra in the category of k-modules, using the standard twisted-commutative structure via permutation of components. As we pointed out, Σg cannot admit a nontrivial binary operation. To fix this, we add a parameter t, similarly to §3.1. Namely, rather than considering twisted Lie algebra structures on Σg itself, we consider structures on (3.3.2) Σg[t] := Σg ⊗S k[t] ∼= k[t] ⊗ Σg ⊗ k[t], which is the module over the twisted-commutative algebra k[t] = SymShti freely generated by Σg. By definition, Σg[t] is an S-module with t = 1 = Σg. A twisted Lie algebra structure on Σg[t] over k[t] is, by definition, a twisted Lie bracket that is k[t]-linear, i.e., such that {tf, g} = t{f, g} for all f, g. We may now state our main observa- tion (which makes precise the heuristic equivalence between pre-Lie algebras and "twisted Lie algebras concentrated in degree one"): Proposition 3.3.3. -- Twisted Lie algebra structures on Σg[t] over k[t] are canonically equiv- alent to pre-Lie structures ◦ : g ⊗ g → g on g. The equivalence is given by (3.3.4) {x, y} = (x ◦ y)t − t(y ◦ x), ∀x, y ∈ g. Moreover, such Σg[t] are in fact the associated Lie algebras of twisted pre-Lie algebras. For the final statement, twisted pre-Lie algebras are defined as pre-Lie algebras in the category of S-modules; explicitly, the twisted pre-Lie axiom is (3.3.5) (x ◦ y) ◦ z − x ◦ (y ◦ z) = (23)x,z,y(cid:0)(x ◦ z) ◦ y − x ◦ (z ◦ y)(cid:1). Proof. -- Begin with a twisted Lie algebra structure on Σg[t]. By restricting to Σg, we obtain a map (3.3.6) Σg ⊗ Σg → ((Σg)t ⊕ t(Σg)), {v, w} = (x ◦ y)t − t(y ◦ x). The map ◦ is equivalent to the bracket { , } using the formula (3.3.7) {taxtb, tcytd} = ta(x ◦ y)tb+c+d − ta+b+c(y ◦ x)td. We conclude that twisted Lie structures on Σg[t] over k[t] are the same as binary operations ◦ : g ⊗ g → g such that the bracket defined by (3.3.7) satisfies (3.2.9). It is easy to see that (3.2.9) holds for this bracket if and only if it holds whenever x, y, z ∈ Σg. In this case, the LHS of (3.2.9) lives in (Σg)t2 ⊕ t(Σg)t ⊕ t2(Σg). Each component of the resulting identity is easily seen to be equivalent to the pre-Lie condition (1.1.2). For the final statement, one defines the twisted pre-Lie structure on Σg[t] by (tavtb)◦(tcwtd) = ta(v ◦ w)tb+c+d. It follows from the above analysis that this is a twisted pre-Lie structure. 10 Remark 3.3.8. -- One can also argue slightly differently to prove equivalence of the twisted Jacobi identity for Σg[t] and the pre-Lie identity for g: for (3.3.7) to define a twisted Lie bracket, the commutator {x, y} := x ◦ y − y ◦ x must be a Lie bracket by setting t = 1, and then each component of (3.2.9) becomes the condition that ◦ is a right Lie action of (g, { , }) on itself. As remarked by M. Livernet, this is well known to be equivalent to the pre-Lie condition. 3.4. Twisted Poisson algebras. -- We first make some definitions we will need for the rest of the paper. Let (3.4.1) SymS,k[t] Σg[t] := SymS Σg[t]/(t ⊗ f − tf ) be the twisted-commutative algebra over k[t] generated by the k[t]-module Σg[t]. Here, the quotient is by the twisted-commutative ideal generated by t ⊗ f − tf . Analogously to (3.1.3), and using (3.3.1), we have the formula where SymSbg is equipped with its usual structure of twisted-commutative algebra, and (3.4.3) SymS,k[t] Σg[t] ∼= SymSbg ∼= Tkbg, (3.4.2) again with t = 1. bg := Σg ⊕ hti, Now, we explain a "quasiclassical" analogue of Theorem 4.3.1 of the next section (the result which implies Theorem 1.1.4 in full generality), which can be regarded as a translation of Proposition 3.3.3 to the Poisson (rather than Lie) setting. This will not be needed for the rest of the paper, so the reader can skip it if desired. ing Proposition 3.4.4. -- Twisted Poisson structures on Tkbg = SymS,k[t](Σg[t]) over k[t] satisfy- (3.4.5) {Σg[t], Σg[t]} ⊆ Σg[t], i.e., {Σg, Σg} ⊆ (t(Σg) ⊕ (Σg)t), are equivalent to pre-Lie algebra structures ◦ on g, by (3.3.4). structure on Σg[t]; conversely, there is a unique extension of any twisted Lie algebra structure satisfying (3.4.5) are equivalent to twisted Lie algebra structures on Σg[t]. This follows as in Proof. -- We need to show that twisted Poisson structures on Tkbg = SymS,k[t](Σg[t]) over k[t] the ordinary setting: any twisted Poisson structure on Tkbg restricts to a twisted Lie algebra on Σg[t] satisfying (3.4.5) to a twisted Poisson structure on Tkbg. The latter result is an easy special case of the fact (see [Sch09, Proposition 1.10]) that, for every S-module h, twisted Poisson structures on SymS h are equivalent to binary operations h ⊗S h → SymS h satisfying (3.2.8) and (3.2.9) for x, y, z ∈ h, and similarly when h is over k[t]. The condition (3.4.5) guarantees that the binary operation Σg[t] ⊗ Σg[t] → SymS,k[t](Σg[t]) in fact lands in Σg[t], which is therefore a twisted Lie bracket. 4. Generalizations and full proof of Theorem 1.1.4 The first goal of this section is to state and prove a generalization of Theorem 1.1.4 which resulting Theorem 4.3.1 can be viewed as a noncommutative analogue of Theorem 1.1.4, since yields a twisted coalgebra isomorphism Tkbg = SymS,k[t](Σg[t]) ∼→ US,k[t](Σg[t]) = USbg. The it replaces Symk g (better, Symkbg = Symk g ⊗k k[t]) with Tkbg. Moreover, this theorem is, in a sense, easier to prove than the original one, since the graded PBW isomorphism will be automatic rather than an assumption, and some of the argument simplifies. 11 Then, we explain how to further generalize this to prove twisted analogues of Theorem 1.1.4 and Corollary 1.3.1, applicable to any twisted pre-Lie algebra (Theorem 4.5.1), or more generally, to any pre-Lie algebra in a suitable symmetric monoidal category (Theorem 4.7.3). In the case of twisted pre-Lie algebras, the categorical approach requires working with S- bimodules. This is not entirely satisfactory, and we give a construction entirely in S-modules, by extending the suspension functor to act on arbitrary S-modules, in particular taking twisted pre-Lie algebras to twisted pre-Lie algebras. This culminates in Theorem 4.6.10, which implies all the results about twisted pre-Lie algebras without using S-bimodules. There is also a common generalization of this result and the categorical Theorem 4.7.3: see Remark 4.7.9. Before proving these results, we need to recall Stover's graded PBW theorem for connected twisted Lie algebras. We also recall the notions of twisted coalgebras and bialgebras, relevant to the twisted symmetric and enveloping algebras SymS g and USg. These occupy §§4.1, 4.2. 4.1. Twisted PBW theorem. -- By recasting pre-Lie algebras as twisted Lie algebras, we may apply the twisted PBW theorem. As is well known (see, e.g., [Bar78]), if h is a twisted Lie algebra which is free as a k-module, it satisfies the graded PBW theorem. That is, one may consider the twisted enveloping algebra (4.1.1) USh := TSh/(xy − (12)y,xyx − {x, y})x,y∈h, where TSh is the free twisted associative algebra generated by h, and the quotient is by a twisted algebra ideal (i.e., a usual ideal which is also an S-submodule). Then, USh is again filtered using k = F0(TSh) ⊆ F1(TSh) = k ⊕ h. Under the same assumptions as before (e.g., h is free as a k-module or k ⊇ Q), the canonical epimorphism is an isomorphism (4.1.2) SymS h ∼→ gr USh, preserving all structures. Remarkably, in the case that h is connected, i.e., h is concentrated in positive degrees, Stover noticed [Sto93] that the above isomorphism holds without any hy- potheses on h and k: Theorem 4.1.3. -- [Sto93] Let k be any commutative ring, and let h be any connected twisted Lie algebra over k. Then, the canonical map (4.1.2) is an isomorphism. One rough explanation why this holds is that, for all n, the tensor algebra T n S h is a free Sn- module in each degree (by permutation of the h-factors), where by "free," we mean a module of the form IndSn {1} M = k[Sn] ⊗k M for some M (which is not necessarily free as a k-module). See §5 for a sketch of a simple proof. Caution 4.1.4. -- It is tempting to conclude that Theorem 4.1.3 for connected twisted Lie algebras, together with the connection of Proposition 3.3.3 between pre-Lie algebras g and con- nected twisted Lie algebras Σg[t], yields the pre-Lie graded PBW theorem (Theorem 1.3.1.(i)). However, we do not know how to conclude this. In more detail, if g is a pre-Lie algebra, one deduces from these results that SymS,k[t] Σg[t] ∼→ gr US,k[t]Σg[t]. By taking Sn-coinvariants in each degree n and setting t = 1 (as we will do later to deduce Theorem 1.1.4 from Theorem 4.3.1), one obtains Sym g ∼→ (gr US,k[t]Σg[t])St=1, where here for an S-module M =Ln≥0 Mn, we denote the total coinvariants by MS := Ln≥0(Mn)Sn. However, one would instead like to have an isomorphism with gr U g = gr(cid:0)(US,k[t]Σg[t])S(cid:1)t=1. It is, however, not always true that taking coinvariants commutes with taking associated graded. One could fix this if one had a quantum PBW isomorphism, Sym h ∼→ U h, but Stover only showed that such an isomor- phism exists as a map of k-modules, not as a map of S-modules: in fact, this isomorphism 12 does not exist for all connected twisted Lie algebras, by the remark below (which also shows that (gr USh)S 6∼= gr((USh)S) for general twisted Lie algebras h). This isomorphism does exist, however, in our case h = Σg[t] for g a pre-Lie algebra, but we need Theorem 4.3.1 to show it. Remark 4.1.5. -- As we just mentioned, in the connected twisted case, the graded PBW iso- morphism does not always lift to an isomorphism of S-modules SymS h ∼→ USh (Stover lifted it only to an isomorphism of k-modules), i.e., there is no quantum PBW theorem in this setting. To see this, note that, when the graded PBW isomorphism does lift to an S-module isomorphism, example, given any ordinary graded connected Lie algebra W , we can view it as a connected we can take Sn-coinvariants in each degree to obtain Symk(Lm≥0 hSm) ∼→ Uk(Lm≥0 hSm). For twisted algebra fW with trivial Sm actions in each degree m. Then, if the twisted graded PBW isomorphism for fW lifts to an S-module isomorphism, resp., twisted coalgebra isomor- phism, then the usual graded PBW morphism for W also lifts to a k-module, resp., coalgebra isomorphism, Symk W ∼→ UkW . In particular, the usual graded PBW theorem holds for W , Symk W ∼→ gr UkW . However, there are many examples of connected ordinary Lie algebras W for which the graded PBW theorem doesn't hold, such as the PBW counterexample [Coh63, §5] recalled in the appendix. 4.2. Twisted coalgebras and bialgebras. -- A twisted commutative coalgebra A is a commutative coalgebra in the category of S-modules, which explicitly means the following: A is an S-module equipped with a comultiplication A → A ⊗S A, which is coassociative: (4.2.1) (Id ⊗S∆) ◦ ∆ = (∆ ⊗S Id) ◦ ∆, and cocommutative, (4.2.2) ∆ = β ◦ ∆, where β is given in (3.2.3). We also require a counit, ε : A → k, where k is viewed as an S-module concentrated in degree zero, such that (4.2.3) (ε ⊗S Id) ◦ ∆ = Id = (Id ⊗Sε) ◦ ∆. The twisted symmetric and enveloping algebras SymS h and USh are equipped with a standard twisted cocommutative coproduct, which is the unique S-module extension of (4.2.4) xj, ∀x1, . . . , xm ∈ h, ∆(x1x2 · · · xm) = XI⊔J={1,2,...,m} (σI,J )x1,...,xmYi∈I xi ⊗Yj∈J where σI,J ∈ Sm is the permutation which reorders the indices of the xi's that appear into increasing order (i.e., taking the products over I and J to be in increasing order, this is the appropriate (I, J)-shuffle). The counit ε is the quotient by the augmentation ideal (g). A clearer, although less explicit, way to define the above coproduct is to note that SymS h and USh are in fact twisted bialgebras, which means that ∆ and ε are morphisms of twisted associative algebras, or equivalently that the multiplication and unit are morphisms of twisted coassociative coalgebras. Explicitly, (4.2.5) ∆(xy) = ∆(x) · ∆(y), ε(xy) = ε(x)ε(y), ε(1) = 1, where the multiplication · on A ⊗S A is given by (4.2.6) (u ⊗ v) · (x ⊗ y) = (23)u,x,v,y(ux ⊗ vy), ∀u, v, x, y ∈ A. Then, (4.2.4) is the unique extension of ∆(x) = 1 ⊗ x + x ⊗ 1, x ∈ h, from h to all of SymS h or USh. 13 In the k[t]-module case, one similarly defines twisted-commutative coproducts on SymS,k[t] h and US,k[t]h, yielding twisted bialgebras over k[t]. Being "over k[t]" here refers to the fact that k[t] is a sub-bialgebra (k[t] itself is a twisted bialgebra via k[t] = SymShti, i.e., the usual graded bialgebra k[t] with t = 1 equipped with the trivial Sm-action in each degree m). Thus, bialgebras. Theorem 4.3.1. -- Let k be any commutative ring and g any k-module. Star products ∗ on 4.3. "Noncommutative" generalization of Theorem 1.1.4. -- We will use the filtration specializing to (3.3.2), Tkbg = SymS,k[t] Σg[t] and USbg := US,k[t]Σg[t] are twisted cocommutative on Tkbg generated by the {0, 1}-degree filtration onbg: hti = F0(bg) ⊆ F1(bg) =bg. Tkbg satisfying the conditions (ii) ∗ is a filtered product whose associated graded is the usual product on Tkbg, (iv) f ∗ t = f t and t ∗ f = tf , for all f ∈ Tkbg, are equivalent to (right) pre-Lie algebra structures ◦ : g ⊗ g → g, under the correspondence k (Σg)) ∗ Tkbg ⊆ (T m (i) ∗ forms a twisted bialgebra with the usual coproduct ∆, k (Σg))(Tkbg), (iii) ∗ satisfies (4.3.2) (T m (4.3.3) x ∗ y = xy + (x ◦ y)t, ∀x, y ∈ g. We remark that, using (ii) and (iii), condition (iv) is equivalent to the condition that t is the theorem, if we say instead that star products satisfying (i) -- (iii) are equivalent to twisted twisted-central in (Tkbg, ∗). Furthermore, we need not mention (iv) or the words "pre-Lie" in Lie algebra structures on Σg[t] over k[t] (or twisted Poisson structures on Tkbg over k[t]). Sketch of proof. -- The proof is an adaptation of the proof of Theorem 1.1.4. By a straightfor- ward analogue of Proposition 1.2.2, it again suffices to show that ◦ gives rise to a star product ∗ satisfying (i) -- (iv). (4.3.4) Φ≤n(x1x2 · · · xn−1xn) with 1. First, extend ◦ to an operation onbg such that t ◦ x = 0 = x ◦ t for all x ∈bg. With this, (4.3.3) is valid for x, y ∈bg, using (iv). 2. We inductively construct a twisted coalgebra isomorphism Φ : Tkbg ∼→ USbg, replacing (2.1.2) x1x2 · · · xi−1(xi ◦ xn)xi+1 · · · xn−1(cid:17)t. = Φ≤n−1(x1x2 . . . xn−1)xn − Φ≤n−1(cid:16) nXi=1 3. Since gr Φ≤n is the twisted graded PBW isomorphism (the k[t]-analogue of Theorem 4.1.3 for h = Σg[t], obtained from the graded PBW isomorphism for the latter by imposing it remains to check that this is actually a twisted coalgebra morphism. We may again k[t]-linearity), (4.3.4) certainly gives a well-defined k-linear isomorphism Tkbg ∼→ USbg, and extend the star product to an operation T ≤ibg ⊗ T ≤n−ibg → T ≤nbg such that Φ≤n(a ∗ b) = 4. To check that Φ≤n is an S-module morphism, it is enough by induction to show that Φ≤n Φ≤n(a) · Φ≤n(b). commutes with (n − 1, n), or equivalently, that (similarly to (2.1.5)) (4.3.5) (n−1, n)(x1 · · · xn−2 ∗xn)∗xn−1 = ((x1 · · · xn−2)∗xn−1)∗xn −(x1 · · · xn−2)∗{xn−1, xn}. As before, this follows by expanding ∗ using (4.3.4) and the pre-Lie identities for ◦. 14 5. To check that Φ≤n is a twisted coalgebra morphism, we use (2.1.6). 6. Finally, (4.3.2) is proved by replacing (2.1.7) with (4.3.6) (x1x2 · · · xm) ∗ (xm+1 ∗ (xm+2 · · · xn)) ∈ (T m k (Σg))(T n−m k bg). 4.4. General proof of Theorem 1.1.4. -- Here, we deduce Theorem 1.1.4 from Theorem 4.3.1, thereby proving the former without any hypotheses on g or k. Proof of Theorem 1.1.4. -- We take Sm-coinvariants in each S-module degree m, and set t equal to 1. This transforms SymSbg into Symk g and USbg into Ukg, where g is viewed as an ordinary Lie algebra with bracket {x, y} = x ◦ y − y ◦ x. Thus, any pre-Lie multiplication on g gives rise to a star product satisfying the needed conditions. Uniqueness and the converse follow from Proposition 1.2.2, as explained in §2.(5) 4.5. Twisted generalization of the main results. -- If we had worked originally in the category of S-modules rather than k-modules, everything we did generalizes. In particular, Theorem 1.1.4 generalizes to: Theorem 4.5.1. -- Let g be any S-module. Star products ∗ on SymS g satisfying the conditions (i) ∗ forms a bialgebra with the usual twisted coproduct ∆, (ii) ∗ is a filtered product whose associated graded is the usual product on Sym g, (iii) ∗ satisfies (4.5.2) (Symm S g) ∗ SymS g ⊆ (Symm S g)(SymS g) = Sym≥m S g, are equivalent to (right) twisted pre-Lie algebra structures ◦ : g⊗g → g, under the correspondence (4.5.3) x ∗ y = xy + x ◦ y, ∀x, y ∈ g. In this context, Proposition 1.2.2 goes through in exactly the same way, yielding the formulas (0) a ◦ 1 = a, (1) a ◦ (bx) = (a ◦ b) ◦ x − a ◦ (b ◦ x), (2) (ab) ◦ c = (23)a,c′,b,c′′(a ◦ c′)(b ◦ c′′), (3) a ∗ b = (a ◦ b′)b′′. ∀a, b ∈ Sym g, x ∈ g, As before, we deduce the twisted graded pre-Lie PBW theorem: Corollary 4.5.4. -- If g is a twisted pre-Lie algebra, and USg the universal enveloping algebra of the associated twisted Lie algebra, then 1. The canonical morphism SymS g → gr(USg) is an isomorphism; 2. The canonical morphism lifts to a twisted coalgebra isomorphism SymS g ∼→ USg. ' In fact, the theorem generalizes to the case where g is a pre-Lie algebra in an arbitrary abelian symmetric monoidal category with limits, as we will explain in §4.7. To prove the theorem, we once again need to be given that SymS g → gr(USg) is an iso- morphism: this is Stover's result if g is connected (or more generally if g0 is a free k-module, In the general setting, we can again circumvent the need for this cf. Theorem 5.0.10.(iii)). (5)Alternatively, since we now have the PBW theorem for pre-Lie algebras, one can use a careful version of the proof in §2, which can be modified to avoid the use of Proposition 1.2.2 as pointed out in the footnote there. 15 assumption by proving a k[t]-analogue of the above (generalizing Theorem 4.3.1) and applying it to Σg[t]. However, `a priori, Σg[t] is in the category of S-bimodules, i.e., bigraded vector spaces with actions of Sm × Sn in bidegree (m, n): the second grading comes from the suspension. The twisted graded PBW theorem (Theorem 4.1.3) generalizes to connected S-bimodules. However, these arguments seem to be just as clear when one replaces S-modules by an arbitrary symmetric monoidal category C, and S-bimodules by the category SC of symmetric sequences of objects of C, as we will do in §4.7 below. There is an alternative approach that remains in the category of S-modules: to define the suspension Σ as a functor from S-modules to itself, raising degrees by one. We then obtain an equivalence between twisted pre-Lie algebras g and certain twisted Lie algebra structures on Σg[t]. We carry this through in the next section, which implies a "noncommutative" (better, "suspended") generalization, Theorem 4.6.10, of the above. 4.6. Suspensions of arbitrary twisted pre-Lie algebras. -- Here, we explain how to extend the suspension Σ as an operation on S-modules, which allows us to generalize Proposition 3.3.3 and Theorem 4.3.1, and also gives one way to prove the results of the previous section, remaining in the category of S-modules rather than S-bimodules. The main idea is to interpret Σg, when g is a k-module, as IndS1 V , where the k-module g is viewed also as an S-module concentrated in degree zero. Then, given an arbitrary S-module g, define the S-module Σg by S0×S1 (4.6.1) (Σg)0 = 0, (Σg)m+1 := IndSm+1 Sm×S1 gm. We will use the decomposition as k-modules, (4.6.2) (Σg)m+1 = gm ⊕ Given an element x ∈ gm, let (i, i + 1, . . . , m + 1)gm. mMi=1 (4.6.3) x(0) := x ∈ (Σg)m+1, x(i) := (i, i + 1, . . . , m + 1)x ∈ (Σg)m+1. Heuristically, think of the element x(0) as "xs" where we multiply x by a parameter s, with s = 1. Similarly, x(1) can be thought of as "sx", and x(i) can be thought of as obtained from x by "inserting an s between the (i − 1)-th and i-th Sm-module components," for 2 ≤ i ≤ m. Now, we associate to any binary operation ◦ on g a twisted skew-symmetric bracket on Σg[t] := Σg ⊗S k[t], such that, heuristically, {x(i), y(j)} contains an s and a t in the first and second places where an s appears in x(i) ⊗ y(j), i.e., in the i-th and (x + 1 + j)-th components in the case i, j ≥ 1. We state this precisely for i = j = 0 by (4.6.4) {x(0), y(0)} = (x ◦ y)(x)t − (12)y+1,x+1(y ◦ x)(y)t, which heuristically may be written as (4.6.5) {xs, ys} = (x s ◦ y)t − (12)y+1,x+1(y s ◦ x)t. Another motivation for these formulas is that they describe the quasiclassical limit of star products of the form (4.6.12) we consider below (generalizing (4.3.3)). In the original situation where g is concentrated in degree zero, (4.6.4) reduces to (3.3.4). More generally, Proposition 3.3.3 extends to 16 Proposition 4.6.6. -- The bracket { , } of (4.6.4) defines a twisted Lie algebra structure on Σg[t] over k[t] if and only if ◦ defines a twisted pre-Lie algebra structure on g. In this case, Σg[t] is itself a twisted pre-Lie algebra, under the k[t]-linear operation (4.6.7) x(0) ◦ y(0) := (x ◦ y)(x)t, t ◦ u = 0 = u ◦ t. In the proposition, k[t]-linearity means that (ut) ◦ v = u ◦ (tv), t(u ◦ v) = (tu) ◦ v, and (u ◦ v)t = u ◦ (vt), for all u, v. Proof. -- We need to show that, given a binary operation ◦ on g and a bracket { , } on Σg⊗S k[t] satisfying (4.6.4), the twisted pre-Lie identity (3.3.5) for g is equivalent to the twisted Jacobi identity (3.2.9) for Σg ⊗S k[t]. It suffices to consider the twisted Jacobi identity for elements of the form x(0), y(0), z(0), where x, y, z are homogeneous elements of g. Then, (4.6.8) {x(0), {y(0), z(0)}}+(132)z+1,x+1,y+1{z(0), {x(0), y(0)}}+(123)y+1,z+1,x+1{y(0), {z(0), x(0)}} = (34)x+1,y,z,1,1(cid:0)x ◦ (y ◦ z) − (x ◦ y) ◦ z − (23)x,z,y(x ◦ (z ◦ y) − (x ◦ z) ◦ y)(cid:1)(x)tt + . . . , where . . . are the terms obtained from the given ones by replacing (x, y, z) with (z, x, y) and applying (132)z+1,x+1,y+1, and by performing this operation twice. The equivalence of the proposition follows immediately from this formula. To show that (4.6.7) is a pre-Lie structure is an explicit verification: it suffices to consider the pre-Lie identity for a triple of elements x(0), y(0), z(0) ∈ Σg obtained from x, y, z ∈ g, and then the identity follows from the pre-Lie identity for the triple x, y, z with a little bit of extra bookkeeping. In more detail, x(0) ◦ (y(0) ◦ z(0)) = (34)x+1,y,z,1,1(x ◦ (y ◦ z))(x)tt, and similarly the other three terms of the pre-Lie identity for (x(0), y(0), z(0)) can be obtained from the corresponding terms of the identity for (x, y, z) by the map T 7→ (34)x+1,y,z,1,1T (x)tt. Finally, as before, Theorem 4.5.1 has the following stronger "noncommutative" analogue. Let us again use the notation (3.4.3) and consider the algebras (4.6.9) tions Symm SymSbg = SymS,k[t](Σg[t]), USbg = US,k[t](Σg[t]). (i) ∗ forms a twisted bialgebra with the usual twisted coproduct ∆, Also, let (SymS Σg) ⊂ SymSbg be the twisted commutative subalgebra generated by Σg ⊂bg. Theorem 4.6.10. -- Let g be any S-module. Star products ∗ on SymSbg satisfying the condi- (ii) ∗ is a filtered product whose associated graded is the usual product on SymSbg, (iv) f ∗ t = f t and t ∗ f = tf for all f ∈ SymSbg, are equivalent to (right) twisted pre-Lie algebra structures ◦ : g⊗g → g, under the correspondence S (Σg) ∗ SymSbg ⊆ Symm S (Σg)(SymSbg), (iii) ∗ satisfies (4.6.11) (4.6.12) x(0) ∗ y(0) = x(0)y(0) + (x ◦ y)(x)t, ∀x, y ∈ g. This theorem can be proved using only Theorem 4.1.3 and Proposition 4.6.6 (see the sketch below), and implies all of the other results of Sections 1 through 4. In particular, one may use it to deduce the results of the previous subsection without using S-bimodules. Also, the remarks following Theorem 4.3.1 (allowing us to replace the condition (iv) and the words "pre-Lie"), apply here as well. 17 Sketch of proof. -- The proof is again an adaptation of the proof of Theorem 1.1.4. By a straightforward analogue of Proposition 1.2.2 using (4.6.4), it once again suffices to show that ◦ gives rise to a star product ∗ satisfying (i) -- (iv). 1. Thanks to (iv) and the pre-Lie structure (4.6.7) on Σg[t], (4.6.12) becomes (4.6.13) u ∗ v = uv + u ◦ v, (2.1.2) with valid for all u, v ∈bg. 2. We inductively construct a twisted coalgebra isomorphism Φ : SymSbg ∼→ USbg, replacing σi · x1x2 · · · xi−1(xi ◦ xn)xi+1 · · · xn−1(cid:17), = Φ≤n−1(x1x2 . . . xn−1)xn − Φ≤n−1(cid:16) nXi=1 (4.6.14) Φ≤n(x1x2 · · · xn−1xn) using the permutation (4.6.15) σi := (i + 1, n, n − 1, n − 2, . . . , i)x1,x2,...,xi,xn,xi+1,xi+2,...,xn−1. 3. Since gr Φ≤n is the twisted graded PBW isomorphism (the k[t]-linear quotient of the graded PBW isomorphism of Theorem 4.1.3 for h = Σg[t]), (4.6.14) certainly gives a well- a twisted coalgebra morphism. We may again extend the star product to an operation Sym≤i defined k-linear isomorphism SymSbg ∼→ USbg, and it remains to check that this is actually S bg ⊗ Sym≤n−i S bg such that Φ≤n(a ∗ b) = Φ≤n(a) · Φ≤n(b). 4. To check that Φ≤n is an S-module morphism, it is enough by induction to show that bg → Sym≤n (similarly to (2.1.5)) S (4.6.16) (n − 1, n)x1,...,xn−2,xn,xn−1(x1 · · · xn−2 ∗ xn) ∗ xn−1 = ((x1 · · · xn−2) ∗ xn−1) ∗ xn − (x1 · · · xn−2) ∗ {xn−1, xn}. As before, this follows by expanding ∗ using (4.6.14) and the pre-Lie identities for ◦. 5. To check that Φ≤n is a twisted coalgebra morphism, we use (2.1.6). 6. Finally, (4.5.2) is proved by replacing (2.1.7) with (4.6.17) (x1x2 · · · xm) ∗ (xm+1 ∗ (xm+2 · · · xn)) ∈ (Symm S (Σg))(Symn−m S bg). 4.7. Categorical generalization of the main results. -- In fact, there is nothing in Theorem 4.5.1 that requires the category of S-modules. Take any abelian symmetric monoidal category C (see, e.g., [JS93]) with braiding β and arbitrary limits. Denote the monoidal product by ⊗: we will omit any subscripts of C in this section. Take any Lie algebra (g, { , }) in the category C: that is, an object in C equipped with a morphism { , } : g ⊗ g → g which is skew-symmetric and satisfies the Jacobi identity (using β to express both). One has symmetric and universal enveloping algebras Sym g and U g in the category C, and a natural epimorphism (4.7.1) Sym g ։ gr U g. One similarly has the notion of pre-Lie algebras, and one can consider filtered associative star products ∗ : Sym g ⊗ Sym g → Sym g in C whose associated graded is the morphism (4.7.1). Moreover, Sym g is naturally equipped with the structure of a bialgebra in C: let I be the identity object of C with respect to tensor product. The coproduct (4.7.2) ∆ : Sym g → Sym g ⊗ Sym g 18 is then defined by uniquely extending the canonical morphism g → ((I ⊗ g) ⊕ (g ⊗ I)) so as to obtain a bialgebra. We will be interested in viewing Sym g as a filtered coalgebra, and equipping it with a new associative structure ∗ whose associated graded is the original bialgebra structure on Sym g. In this context, Theorem 4.5.1 goes through without change: Theorem 4.7.3. -- Pre-Lie algebras in C are equivalent to star-product algebra structures on Sym g satisfying (i) ∗ forms a bialgebra in C with the usual coproduct ∆, (ii) ∗ is a filtered product whose associated graded is the usual product on Sym g, (iii) ∗ satisfies (4.7.4) (Symm g) ∗ Sym g ⊆ (Symm g)(Sym g) = Sym≥m g, are equivalent to (right) pre-Lie algebra structures ◦ : g ⊗ g → g, under the correspondence that the restriction of ∗ to a morphism of C, (4.7.5) ∗ : g ⊗ g → Sym g, is the direct sum of the morphisms (4.7.6) · : g ⊗ g → Sym2 g, ◦ : g ⊗ g → g ∼= Sym1 g, the first coming from the standard algebra structure of Sym g, and the second from the pre-Lie structure on g. Moreover, Proposition 1.2.2 has a straightforward categorical generalization as follows. Let µ· and µ∗ denote the usual and star-product multiplications Sym g ⊗ Sym g → Sym g, and µ◦ denote the pre-Lie multiplication on g, which we will extend formally to a binary operation on Sym g (which is not pre-Lie). Let ιr : X ∼→ X ⊗ I be the natural isomorphism which is part of the identity structure of C (for all objects X in C). Then, as in Proposition 1.2.2, the following formulas hold and give an inductive (on degree of Sym g) way to define µ∗ (while simultaneously extending µ◦ to a binary operation on all of Sym g): (0) µ◦ιr = Id, (1) µ◦(Id ⊗µ·)Sym g⊗Sym g⊗g = µ◦(µ◦ ⊗ Id − Id ⊗µ◦)Sym g⊗Sym g⊗g, (2) µ◦(µ· ⊗ Id) = µ·(µ◦ ⊗ µ◦)(Id ⊗β ⊗ Id)(Id ⊗ Id ⊗∆), (3) µ∗ = µ·(µ◦ ⊗ Id)(Id ⊗∆). Corollary 4.7.7. -- If g is a pre-Lie algebra (or associative algebra) in C and U g the universal enveloping algebra of the associated Lie algebra, then 1. The canonical morphism Sym g → gr(U g) is an isomorphism; 2. The canonical morphism lifts to a coalgebra isomorphism Sym g ∼→ U g. Sketch of proof of Theorem 4.7.3. -- The proof of this theorem is a straightforward generaliza- tion of the proof of Theorem 1.1.4 in the case when (4.7.1) is an isomorphism. In more detail, we translate the formulas there into categorical terms, analogously to the way we translated (4.5.3) into (4.7.5) and (4.7.6), and Proposition 1.2.2.(0) -- (3) into the above. To avoid assuming that (4.7.1) is an isomorphism, we can prove a suspended version of the above. This involves considering the category SC, whose objects are of the formLi≥0 Xi, where Xi are objects of C equipped with an action by Si of automorphisms. For example, if C is the category of k-modules, then SC = S-modules, and if C = S-modules, then SC = S-bimodules (as remarked in §4.5). Analogously to the case of S-modules, we can endow SC with the structure 19 of symmetric monoidal category, given by (3.2.2) and (3.2.3) verbatim, provided we understand the operation IndG H(X) for finite groups H < G, and X an object of C equipped with an action of H by automorphisms. One way to construct this is to take X ⊕G, labeling the copies of X by the elements of G, and then quotient by setting the diagonal action of H by automorphisms equal to the action of H by permuting the factors of X according to its action on G. Then, we can form the object Σg[t] categorically. If g is an object of C, we form the object Σg =Li≥0 Xi where X1 = g and all other Xi = 0 (equipped with the trivial action of S1). We can also form Σg[t] =Li≥0 Yi where Yi =Li j=1 I ⊗j−1 ⊗ g ⊗ I ⊗i−j ∼= g⊕i, equipped with the action of Si by permutation of components. Now, Σg[t] is not merely an object of SC, but a module over the algebra P := Sym(ΣI), which replaces the polynomial algebra k[t] (ΣI replaces hti). A straightforward categorical generalization of Proposition 3.3.3 then shows that pre-Lie structures on objects g of C are equivalent to Lie structures on Σg[t] in SC compatible with the P -algebra structure (with P as in the preceding paragraph). The proof, as in the case of Proposition 3.3.3, is just a matter of checking the definitions, and expanding the Jacobi identity in SC into the components of (4.7.8) (Σg[t])3 = (g ⊗ I ⊗ I) ⊕ (I ⊗ g ⊗ I) ⊕ (I ⊗ I ⊗ g). Now, as we will outline in Section 5, Stover's graded PBW theorem for connected Lie algebras ex- tends to the categorical setting of SC. This allows us to prove that SymSC ,P (Σg[t]) ∼→ USC ,P (Σg[t]) via a star-product on the former symmetric algebra. Finally, we can take Sn-coinvariants in each degree n, and perform the identifications I ⊗j−1 ⊗ g ⊗ I ⊗i−j ∼= g which generalize setting t = 1. We conclude that SymC g ∼→ UCg via the star product ∗ asserted in the theorem. Remark 4.7.9. -- There is a common generalization of Theorems 4.7.3 and 4.6.10, which extends the suspension functor to act from SC to itself, taking pre-Lie algebras to pre-Lie algebras, and such that pre-Lie algebra structures on g ∈ SC are equivalent to certain star products on SymSC ,P Σg[t] (with notations as in the proof above). We omit the details. 5. Graded PBW theorems in a unified categorical context In this section, we provide a simple proof of the twisted and non-twisted graded PBW theo- rems in a more general categorical context. We also generalize Stover's graded PBW theorem [Sto93] for connected twisted Lie algebras from the setting of k-modules to that of an arbitrary symmetric monoidal category (Theorem 5.0.10.(iv)). We will use the setup of §4.7. In this section, we address the question: when is (4.7.1) an isomorphism? To answer this, let J ⊂ T g be the kernel of the projection T g ։ U g. For all n, let J≤n be the subobject of T ≤ng given by (5.0.1) (5.0.2) J2 := (Id −β − { , })(T 2g) ⊂ T ≤2g, T ig ⊗ J2 ⊗ T jg, ∀n ≥ 3. J≤n := Xi+j+2≤n Then, it is evident that J = limn→∞ J≤n. The main technical result we need is Lemma 5.0.3. -- The formula (5.0.4) (i, i + 1)·{ ,} := βi,i+1 + { , }i,i+1 20 defines an action of Sn on T ng ⊕ (T ≤n−1g/J≤n−1g) for all n, acting trivially on the second factor, so that taking coinvariants yields T ≤ng/J≤ng. Proof. -- We need to check the following identities: (5.0.5) (5.0.6) (5.0.7) (i, i + 1)2·{ ,} = Id, (i, i + 1)(i + 1, i + 2)(i, i + 1)·{ ,} = (i + 1, i + 2)(i, i + 1)(i + 1, i + 2)·{ ,}, (i, i + 1)(j, j + 1)·{ ,} = (j, j + 1)(i, i + 1) ·{ ,} . The first identity (5.0.5) follows from skew-symmetry of { , }. The braid relation (5.0.6) follows from the Jacobi identity. Finally, (5.0.7) follows by definition. Now, there is a natural exact sequence of Sn-modules (5.0.8) 0 → T ≤n−1g/J≤n−1g → (T ng ⊕ (T ≤n−1g/J≤n−1g), ·{ ,}) → T ng → 0, where, on the left, T ≤n−1g/J≤n−1g is given the trivial action, and on the right, T ng is given the standard Sn-action. We deduce Proposition 5.0.9. -- The graded PBW map (4.7.1) is an isomorphism if and only if, for all n, the sequence (5.0.8) remains exact after taking Sn-coinvariants. For brevity, say that the graded PBW theorem holds for g if (4.7.1) is an isomorphism. Theorem 5.0.10. -- (i) If C is enriched over Q-vector spaces, then the graded PBW theorem holds.(6) (ii) (Usual graded PBW theorem): If C is the category of k-modules, and g is a Lie algebra in C which is free as a k-module, then the graded PBW theorem holds if and only if {x, x} = 0 for all x ∈ g. (iii) (Twisted graded PBW theorem [Sto93]): If C is the category of S-modules, and g is a connected Lie algebra in C, then the graded PBW theorem holds. More generally, if g0 is a free k-module, the graded PBW theorem holds if and only if {x, x} = 0 for all x ∈ g0. (iv) (Categorical version of (iii)): If C is arbitrary and SC is the category of symmetric se- quences of objects of C (cf. §4.7), and g is a connected Lie algebra in SC, then the graded PBW theorem holds. (v) (Pre-Lie PBW theorem): For arbitrary C, if g is the associated Lie algebra of a pre-Lie algebra in C, then the graded PBW theorem holds. Note that (iii) is a generalization of (ii). Also, by a similar argument, one can replace "free" in (ii) and (iii) by the condition of being projective or a direct sum of cyclic modules. Proof. -- (i) If C is enriched over Q-vector spaces, then all Sn actions are actually Q[Sn] actions, and since Q[Sn] is semisimple, taking Sn-coinvariants is exact. (ii) In this case, as an Sn-module, T ng is a direct sum of modules IndSn M , where M is a free k-module of rank one spanned by an element of the form x⊗p1 , for (xi) a fixed k-basis of g, and i1 ≤ i2 ≤ · · · ≤ ir. Thus, there is a splitting of the surjection in (5.0.8) if {x, x} = 0 for all x. Conversely, if {x, x} 6= 0 for some x, then the element {x, x} is already in the kernel of g ։ gr1 U g. i1 ir (gm1 ⊗ · · · ⊗ gmn). Here, mi > 0 for all i. Hence, Sn acts freely on the left cosets Sm/(Sm1 × · · · × Smn). Let Km,n (iii) By definition, for all n, m, T n(g>0)m ∼=Lm1+···+mn=m IndSm Sm1 ×···×Smn Sp1 ×···×Spr ⊗ · · · ⊗ x⊗pr (6)This was noticed in [Fre98, Theorem A.9]; as explained there, it can also be proved in a usual manner. 21 be a set of representatives for these cosets. Thus, T n(g>0)m is a direct sum of induced modules IndSn of the surjection in (5.0.8) in degree m can be obtained from Sm1 × · · · × Smn-module splittings restricted to gm1 × · · · × gmn for all m1, . . . , mn. This proves (iii) in the case that g is connected. The general case is a combination of this argument with that of (ii). {1}(cid:0)σ(gm1 ⊗ · · · ⊗ gmn)(cid:1) for σ ∈ Km,n, and m1 ≤ · · · ≤ mn. Hence, an Sn-module splitting {1}(cid:0)σ(gm1 ⊗ · · · ⊗ gmn)(cid:1) for σ ∈ Km,n, and m1 ≤ · · · ≤ mn, where induction is defined in §4.7. Hence, one obtains Sn-module splittings of the surjection of (5.0.8) in degree m as in (iii). (iv) Just as in (iii), T n(g>0)m is a direct sum of induced modules IndSn (v) This is Corollary 4.7.7.(i). Remark 5.0.11. -- In cases (i), (ii), and (v) of the theorem, in fact the quantum PBW the- orem holds: one may lift the PBW isomorphism Sym g ∼→ gr(U g) to a coalgebra isomorphism Sym g ∼→ U g. This is not always true in cases (iii) and (iv), by Remark 4.1.5. However, note that, as in [Sto93] or the proof of (iii), one can at least lift (4.7.1) to an isomorphism of k-modules, SymS g ∼→ USg, and in the case of (iv), to an isomorphism in the category C, SymSC g ∼→ USC g (just not necessarily in the category SC). Remark 5.0.12. -- In other symmetric monoidal categories, one can always find extra condi- tions on the bracket { , } so that the graded PBW theorem still holds. In general, the graded PBW theorem holds if and only if, whenever a sum of terms of the form (5.0.13) a ⊗ (x ⊗ y − β(x ⊗ y)) ⊗ b is zero, for x, y ∈ g and a, b ∈ T g, then also the corresponding sum of terms (5.0.14) a ⊗ {x, y} ⊗ b is zero. For this to be valid for a symmetric monoidal category where elements of g don't exist, we replace the above by the condition that, for all n ≥ 2, the kernel of the map (g⊗n)⊕(n−1) → g⊗n given by Li(Id −β)i,i+1 injects into the kernel of the map (g⊗n)⊕(n−1) → g⊗(n−1) given by Li{ , }i,i+1. The case n = 2 of this is the condition that ker(β − Id) ⊆ ker({ , }), (5.0.15) in g⊗2. This may be viewed as the generalized alternating condition. This is not enough in some cases, e.g., in the case of Lie superalgebras (see Remark 1.4.1). Remark 5.0.16. -- There are various generalizations of the graded PBW theorem to the case of quotients of T g by ideals that resemble J above, such as quantized enveloping algebras: see, e.g., [Ber92]. These should also have categorical generalizations, which one should be able to prove by modifying the above approach. PBW counterexamples and pre-Lie identities A Here, we recall the example of [Coh63, §5] where the graded PBW theorem does not hold, i.e., Sym g ։ gr U g is not injective, exploiting a classical p-th power identity of Zassenhaus. We remark that the pre-Lie graded PBW theorem (Corollary 1.3.1.(i)) implies that the identity must lift to the pre-Lie setting, which explains such an identity observed in [Tou06].(7) (7)I am grateful to J.-L. Loday, whose question sparked this appendix. 22 Zassenhaus observed in [Zas39] that there exists a Lie polynomial Λp(x, y) such that (A.0.1) (x + y)p − xp − yp ≡ Λp(x, y) (mod p). In [Coh63], this identity was exploited to give an example where the PBW map Sym g ։ gr U g fails to be an isomorphism. Let F be a field of characteristic p > 0, and let k = F[α, β, γ]/(αp, βp, γp). Let L be the k-module presented as L = Span(x, y, z)/(αx = βy + γz). Let g be the free Lie algebra over k generated by L.(8) Then, it is evident that Λp(βy, γz) 6= 0, since Λ is a Lie polynomial of degree p, and so the relations αp = βp = γp = 0 cannot affect the expansion, as iterated brackets of p copies of the same element yields zero. On the other hand, (A.0.1) together with the fact that U g = T L shows that Λp(βy, γz) = 0 in U g, and hence also in gr U g. So g → U g is not injective, and neither is the canonical morphism Sym g → gr U g. Now, if we consider instead of g the free pre-Lie algebra generated by L, which we denote byeg, it follows from the inclusion g →eg that Λp(βy, γz) = 0 in Ueg. Since the graded PBW theorem must hold for all pre-Lie algebras (Corollary 1.3.1.(i)), we deduce that Λp(x, y) must be in the linear span of all compositions of x with itself p times, of y with itself p times, and of x + y with itself p times. Indeed, as observed in [Tou06, (7-8)], one has (A.0.2) (x + y)◦p − x◦p − y◦p ≡ Λp(x, y) (mod p), where y ◦i x := (· · · ((y ◦ x) ◦ x) · · · ◦ x) is the i-th power of the right action of x on y by ◦, and x◦i := x ◦i−1 x. Remark A.0.3. -- The identity (A.0.2) together with the fact that ◦ is a right Lie action (this is equivalent to the definition of a pre-Lie algebra) and the relation ad(xp) ≡ adpx (mod p) for associative algebras in characteristic p yields the identities (where ad(x)y := [x, y]): (A.0.4) (A.0.5) (adp(x + y) − adp(x) − adp(y))z ≡ ad((x + y)◦p − x◦p − y◦p)z z ◦p (x + y) − z ◦p x − z ◦p y ≡ z ◦ (x + y)◦p − z ◦ x◦p − z ◦ y◦p (mod p), (mod p). However, as pointed out in [Tou06, §7] (and is easy to check), the operation x 7→ x◦p fails to yield a restricted Lie algebra structure, i.e., adp(x)z 6≡ ad(x◦p)z (mod p), in general. Similarly, z ◦p x 6≡ z ◦ x◦p (mod p), in general. This leads one to ask: is there an example of a restricted Lie algebra, or more generally, a Lie algebra g over a characteristic-p base ring k with a p-th power operation satisfying (A.0.2), (A.0.4), and (αx)◦p = αpx◦p (for all α ∈ k and x ∈ g), such that the (nonrestricted) PBW morphism Sym g ։ gr U g fails to be injective? References [Bar78] M. G. Barratt, Twisted Lie algebras, Geometric applications of homotopy theory (Evanston, IL, 1977) (Berlin), Lecture Notes in Math., vol. 658, Springer, 1978, pp. 9 -- 15. [Ber92] R. Berger, The quantum Poincar´e-Birkhoff-Witt theorem, Commun. Math. Phys. 143 (1992), 215 -- 234. [Bir37] G. Birkhoff, Representability of Lie algebras and Lie groups by matrices, Ann. of Math. (2) 38 (1937), no. 2, 526 -- 532. [Car58] P. Cartier, Remarques sur le th´eor`eme de Birkhoff-Witt, Ann. Scuola Norm. Sup. Pisa (3) 12 (1958), 1 -- 4. (8)Note that g is in fact graded and connected. 23 [CK98] A. Connes and D. Kreimer, Hopf algebras, renormalization, and noncommutative geometry, Comm. Math. Phys. 199 (1998), no. 1, 203 -- 242, arXiv:hep-th/9808042. [CL01] F. Chapoton and M. Livernet, Pre-Lie algebras and the rooted trees operad, Int. Math. Res. Not. (2001), no. 8, 395 -- 408, arXiv:math/0002069v2. [Coh63] P. M. Cohn, A remark on the Birkhoff-Witt theorem, J. London Math. Soc. 38 (1963), 197 -- 203. [Fre98] B. Fresse, Cogroups in algebras over an operad are free algebras, Comment. Math. Helv. 73 (1998), 637 -- 676. [Ger63] M. Gerstenhaber, The cohomology structure of an associative ring, Ann. of Math. (2) 78 (1963), 267 -- 288. [GS08] W. L. Gan and T. Schedler, The necklace Lie coalgebra and renormalization algebras, J. Non- commut. Geom. 2 (2008), no. 2, 195 -- 214, arXiv:math-ph/0702055. [Hig69] P. J. Higgins, Baer invariants and the Birkhoff-Witt theorem, J. Algebra 11 (1969), 469 -- 482. [Joy86] A. Joyal, Foncteurs analytiques et esp`eces de structures, Combinatoire ´enum´erative (Montreal, Que., 1985/Quebec, Que., 1985) (Berlin), Lecture Notes in Math., vol. 1234, Springer, 1986, pp. 126 -- 159. [JS93] Andr´e Joyal and Ross Street, Braided tensor categories, Adv. Math. 102 (1993), no. 1, 20 -- 78. [Laz54] M. Lazard, Sur les alg`ebres enveloppantes universelles de certaines alg`ebres de Lie, Publ. Sci. Univ. Alger. S´er. A. 1 (1954), 281 -- 294 (1955). [OG05] J.-M. Oudom and D. Guin, Sur l'alg`ebre enveloppante d'une alg`ebre pr´e-Lie, C. R. Math. Acad. Sci. Paris 340 (2005), no. 5, 331 -- 336. [OG08] J.-M. Oudom and D. Guin, On the Lie enveloping algebra of a pre-Lie algebra, J. K-Theory 2 (2008), no. 1, 147 -- 167, arXiv:math.QA/0404457. [Poi00] H. Poincar´e, Sur les groupes continues, Trans. Cambr. Philos. Soc. 18 (1900), 220 -- 225. [Rev77] Ph. Revoy, Alg`ebres enveloppantes des formes altern´ees et des alg`ebres de Lie, J. Algebra 49 (1977), no. 2, 342 -- 356. [Ron07] M. Ronco, Shuffle bialgebras, arXiv:math/0703437v2, 2007. [Sch09] T. Schedler, Poisson algebras and Yang-Baxter equations, Advances in quantum computation (Tyler, TX, 2007) (Providence, RI) (K. Mahdavi and D. Koslover, eds.), Contemp. Math., vol. 482, Amer. Math. Soc., 2009, arXiv:math/0612493, pp. 91 -- 106. [Sir53] A. I. Sirsov, On the representation of Lie rings as associative rings, Uspehi Matem. Nauk (N.S.) 8 (1953), no. 5(57), 173 -- 175. [Sto93] C. R. Stover, The equivalence of certain categories of twisted Lie and Hopf algebras over a commutative ring, J. Pure Appl. Algebra 86 (1993), no. 3, 289 -- 326. [Tou06] Victor Tourtchine, Dyer-Lashof-Cohen operations in Hochschild cohomology, Algebr. Geom. Topol. 6 (2006), 875 -- 894 (electronic), arXiv:math/0504017. [Vin63] E. B. Vinberg, The theory of convex homogeneous cones, Trans. Amer. Math. Soc. 12 (1963), 340 -- 403, translated from Trudy Moskov. Mat. Obshch. 12 (1963), 303 -- 358. [Wit37] E. Witt, Treuer Darstellung Liescher Ringe, J. reine angew. Math. 177 (1937), 152 -- 160. [Zas39] H. Zassenhaus, Uber Liesche Ringe mit Primzahlcharakteristik, Abhandl. Math. Sem. Univ. Hamburg 13 (1939), 1 -- 100. March 23, 2010 Travis Schedler 24
1503.03574	1	1503	2015-03-12T03:48:37	One proof of the original Kemer's theorems (concerning the text of C. Procesi "What happened to PI-theory", arxiv.org/abs/1403.5673)	[ "math.RA" ]	We consider associative algebras over a field of characteristic zero. We give a version of the proof of the Kemer's theorems concerning the Specht problem solution. It is proved that the ideal of graded identities of a finitely generated PI-superalgebra coincides with the ideal of graded identities of some finite dimensional superalgebra. This implies that the ideal of polynomial identities of any (not necessary finitely generated) PI-algebra coincides with the ideal of identities of the Grassmann envelope of a finite dimensional superalgebra, and is finitely generated as a T-ideal.	math.RA	math	One proof of the original Kemer's theorems (concerning the text of C. Procesi "What happened to PI-theory" arxiv.org/abs/1403.5673). I.Sviridova Departamento de Matem´atica, Universidade de Bras´ılia, 70910-900 Bras´ılia, DF, Brazil; [email protected]; sviridova [email protected] March 07, 2015. Abstract We consider associative algebras over a field of characteristic zero. We give a version of the proof of the Kemer's theorems concerning the Specht prob- lem solution [12], [13]-[16]. It is proved that the ideal of graded identities of a finitely generated PI-superalgebra coincides with the ideal of graded identities of some finite dimensional superalgebra. This implies that the ideal of poly- nomial identities of any (not necessary finitely generated) PI-algebra coincides with the ideal of identities of the Grassmann envelope of a finite dimensional superalgebra, and is finitely generated as a T-ideal. MSC: Primary 16R50; Secondary 16R10, 16W50, 16W22 Keywords: Associative algebras, superalgebras, graded identities, PI- algebras. Introduction We present here one of the proofs of the Kemer's theorems about Specht problem solution for associative PI-algebras over a field of characteristic zero [12]. We give principally the proof of the fact that any finitely generated Z/2Z-graded PI-algebra has the same graded identities as some finite dimensional Z/2Z-graded algebra. This result is the crucial and the most difficult step of the whole Kemer's solution of the Specht problem. The proof represented here is the partial case of the proof of a similar result for algebras graded by a finite abelian group given by the author in [23] (the case G = Z/2Z). It is only slightly modified to extend the result from an algebraically closed base field (as it was considered in [23]) to any field of charac- teristic zero. Observe that the proof of generalised results in [23] is independent on the original Kemer's proof. The unique essential reference to the Kemer's results 1 in that paper served to obtain Lemma 1. This reference can be safely exchanged by the Lewin's results [17]. PI-representability of a non-graded finitely generated PI-algebra can be also obtained as a partial case of [23] considering the trivial group G = {e}. The second step of the Kemer's arguments is to prove that any PI-algebra satisfies the same graded polynomial identities as the Grassmann envelope of some finitely generated Z/2Z-graded PI-algebra. We refer the reader to the book [9] for the proof of this fact. The author also can prove this fact independently but assume that the proof represented in [9] is the most elegant, short and clear to understand for a reader. Thus the author thinks that there is no any need to repeat these arguments. These two theorems imply (see also [9]) that any PI-algebra satisfies the same polynomial (non-graded) identities as the Grassmann envelope of some finite dimensional Z/2Z-graded algebra. The positive solution of the Specht problem in the classical case follows from the last classification theorem immediately. Observe that besides the original proof of Aleksandr Kemer [13]-[16], published later also in [12], there are several later versions of various authors including inter- pretations and generalisations. Therefore the author does not pretend on originality or some exclusive properties of this text. And she never would get an idea to rewrite the proof of the original Kemer's results whenever some obvious misunderstandings appeared some time ago. This text is just one of the possible version of the proof as the author understood it generalising these results in [23]. Notice that this text is a little more detailed then the text published in [23]. Even here we omit some details and proofs that seems to us not very difficult to restore for a reader. But the author can assure possible readers that all the statements with omitted arguments were really checked. Moreover, even the text [23] is the restricted version of the original detailed author's text. Throughout the paper we consider only associative algebras over a field of char- acteristic zero (not necessary unitary). Further they will be called algebras. Let F be a field of characteristic zero and F hXi the free associative algebra over F generated by a countable set X = {x1, x2, . . .}. A T-ideal of F hXi is a bilateral ideal invariant under all endomorphisms of F hXi. Let A be an associative algebra over F . A polynomial f = f (x1, . . . , xn) ∈ F hXi is called polynomial identity for A if f (a1, . . . , an) = 0 for any a1, . . . , an ∈ A (f ≡ 0 in A). Let us denote by Id(A) = {f ∈ F hXi f ≡ 0 in A} the ideal of all polynomial identities of A. If A satisfies a non-trivial polynomial identity then A is called P I- algebra. It is well known, for example, that any finite dimensional algebra is PI ([10, 18, 21]). The relation between T-ideals of F hXi and P I-algebras is well understood: for any F -algebra A, Id(A) is a T-ideal of F hXi, and any T- ideal I of F hXi is the ideal of identities of some F -algebra A, in particular, of the relatively free algebra of F hXi/I. An algebra A is called Z/2Z-graded (or super-algebra) if A = A¯0 ⊕ A¯1 is the direct sum of its subspaces A¯0, A¯1 where AθAξ ⊆ Aθξ holds for any θ, ξ ∈ Z/2Z. An element a ∈ Aθ is called homogeneous in the Z/2Z-grading of the graded degree θ, we also write degZ2 a = θ in this case. The homogeneous component A¯0 of the 2 We always assume that the set Nk lexicographical order for any natural k. 0 (N0 = NS{0}) is linearly ordered with the Z/2Z-graded algebra A is called neutral (or even). A homomorphism of Z/2Z-graded algebras A, B ϕ : A → B is graded if ϕ(Aθ) ⊆ Bθ for any θ ∈ Z/2Z. An ideal I ✂ A of a graded algebra A is graded if and only if I is generated by homogeneous in the grading elements. In this case the quotient algebra A/I is also Z/2Z-graded with the grading induced by the grading of A (degθ ¯a = degθ a). Let us denote by A1 × · · · × Aρ the direct product of algebras A1, . . . , Aρ, and by A1 ⊕ · · · ⊕ Aρ ⊆ A the direct sum of subspaces Ai of an algebra A. We also denote by J(A) the Jacobson radical of A. Observe that in general all bases and dimensions of spaces and algebras are defined over the base field F unless otherwise indicated. The most part of notions, definitions, facts and properties of polynomial identi- ties may be found in [5], [7], [8], [9], [12]. 1 Free graded algebra. Let us denote by XZ2 = {xiθi ∈ N, θ ∈ Z/2Z} a countable set of pairwise different elements. The algebra F = F hXZ2 i is the free associative Z/2Z-graded algebra with the grading F = ⊕θ∈Z/2ZFθ, where Fθ = hxi1θ1xi2θ2 · · · xisθsθ = θ1 + θ2 + · · · + θsiF . Since all the variables are pairwise distinct then any multihomogeneous polynomial is also a homogeneous element of F in the sense of the grading. An element xiθ is called graded variables, and f ∈ F a graded polynomial. Let f = f (x1θ1, . . . xnθn) ∈ F hXZ2 i be a non-trivial Z/2Z-graded polynomial. We say that a Z/2Z-graded algebra A satisfies the graded identity f (x1θ1, . . . xnθn) = 0 if f (a1θ1, . . . anθn) = 0 in A for any aiθi ∈ Aθi. Denote by IdZ2(A) ✂ F hXZ2 i the ideal of Z/2Z-graded identities of A. Similar to the case of ordinary (non-graded) identities IdZ2(A) is a two-side graded ideal of F hXZ2 i invariant under all graded endomorphisms of F hXZ2 i. Such ideals are called Z2T-ideals. It is clear that any Z2T-ideal I of F hXZ2 i is the ideal of Z/2Z-graded identities of F hXZ2 i/I. Take a set S ⊆ F hXZ2 i. Denote by Z2T [S] the Z2T-ideal generated by S. Then Z2T [S] contains exactly all graded identities which are consequences of polynomials of the set S. Z/2Z-graded algebras A and B are called Z2PI-equivalent (A ∼Z2 B) if IdZ2(A) = IdZ2(B). Let Γ) be a Z2T-ideal. We write also f = g (mod Γ) for f, g ∈ F hXZ2 i iff f − g ∈ Γ. We consider only graded identities of PI-superalgebras, the case when IdZ2(A) ⊇ Γ for some nonzero ordinary T-ideal Γ ✂ F hXi. This holds iff the neutral component Ae is a PI-algebra (IdZ2(A) ∋ f (x1¯0, . . . , xn¯0) 6= 0) (see [6], and [1]). i.e. Observe that a finitely generated superalgebra always has a finite set of Z/2Z- homogeneous generators. We consider only homogeneous generating sets of super- algebras. Definition 1 Given a finitely generated superalgebra A and a finite homogeneous generating set K of A denote by grk(K) the maximal number of elements of the 3 same graded degree in K. Then the homogeneous rank grk(A) of A is the least grk(K) for all finite homo- geneous generating sets K of A. F hXZ2 ν = {xiθ1 ≤ i ≤ ν; θ ∈ Z/2Z} be a finite set of graded variables and Let XZ2 ν i the free associative Z/2Z-graded algebra of the rank ν, ν ∈ N. Let A be a finitely generated Z/2Z-graded algebra. Then the superalgebra Uν = F hXZ2 ν i) is the relatively free algebra of the rank ν for A. Similarly to case of ordinary identities if ν ≥ grk(A) then IdZ2(A) = IdZ2(Uν ). Moreover the next remark takes place. ν i/(IdZ2(A)T F hXZ2 Given a Z2T-ideal Γ1 ⊆ F hXZ2 i denote by Γ1(F hXZ2 ν i) = {f (h1, . . . , hn)f ∈ Γ1, hi ∈ F hXZ2 ν i the verbal ideal of the free asso- ciative Z/2Z-graded algebra of the rank ν generated by all appropriate evaluations of elements of Γ1. ν i, degZ2 hi = degZ2 xi, ∀i} ✂ F hXZ2 Remark 1 1. f (x1, . . . , xn) ∈ IdZ2(F hXZ2 ν i)) if and only if ν i/(ΓT F hXZ2 f (h1, . . . , hn) ∈ Γ for all homogeneous polynomials of appropriate graded de- grees h1, . . . , hn ∈ F hXZ2 ν i. 2. Let A be a finitely generated associative Z/2Z-graded algebra, Γ2 = IdZ2(A) the ν i) ⊆ Γ2 implies ideal of its graded identities, and ν ≥ grk(A). Then Γ1(F hXZ2 that Γ1 ⊆ Γ2. It's well known due to the linearization process and possibility to identify vari- ables in case of zero characteristic that any system of identities (ordinary or Z/2Z- graded) is equivalent to a system of multilinear identities. Thus in case of zero characteristic it is enough to consider only multilinear identities. 2 Graded algebras. A = A¯0 ⊕ A¯1. Then we have an analog of the Wedderburn-Maltsev decomposition for superalgebras. Let eF be an algebraically closed field. Consider a finite dimensional eF -superalgebra Lemma 1 Let eF be an algebraically closed field. Any finite dimensional eF -superal- gebra A is isomorphic as a superalgebra to an F -superalgebra of the form A′ = (C1 × · · · × Cp) ⊕ J. (1) Where the Jacobson radical J = J(A′) of A′ is a graded ideal, B′ = C1 × · · · × Cp is a maximal graded semisimple subalgebra of A′, p ∈ NS{0}. A Z/2Z-graded simple component Cl is one of the superalgebras Mkl,ml(eF ) (a matrix algebra with an elementary Z/2Z-grading) or Mkl(eF [c]) (a matrix algebra over the group algebra of eF [cc2 = 1] with the grading induced by the natural grading of the group algebra). 4 Moreover A′ = Span eF {D, U }, where D = ∪1≤l≤p Dl, Dl = {Eliljl = εlEliljlεl 1 ≤ il, jl ≤ sl }, sl = kl + ml Dl = {Eliljl = εlEliljlεl, Eliljlc = εl Eliljlc εl 1 ≤ il, jl ≤ sl}, sl = kl, c2 = 1 U = {εl′uθεl′′ l′, l′′ = 1, . . . , p + 1; uθ ∈ Jθ = J ∩ A′ if Cl = Mkl,ml(eF ); if Cl = Mkl(eF [c]); θ, θ ∈ Z/2Z} (2) (3) are the sets of homogeneous in the grading elements. Eliljl is the matrix unit. Elements of D are homogeneous with the grading defined by the next equalities degZ2 Eliljl = ¯0 if 1 ≤ il, jl ≤ kl, or kl + 1 ≤ il, jl ≤ kl + ml, and degZ2 Eliljl = ¯1 oth- erwise in Cl = Mkl,ml(eF ); degZ2 Eliljl = ¯0, degZ2 Eliljl · c = ¯1 for all 1 ≤ il, jl ≤ kl in Cl = Mkl(eF [c]). The element εl is a minimal orthogonal central idempotent of B′, homogeneous in the grading, that corresponds to the unit element of Cl (l = 1, . . . , p). Elements uθ runs on the set of homogeneous basic elements of the Jacobson radical J = J(A) = ⊕p+1 l′,l′′=1εl′Jεl′′, where εp+1 is the adjoint idempotent. Proof. It is a classical result (see, e.g., [9] or [2], [3] for a more general case) that a finite dimensional superalgebra A can be represented as B ⊕ J, where J is the Jacobson radical of A, that is a graded ideal, and B is a maximal Z/2Z-graded semisimple subalgebra of A. It is also well known that B ∼= ×p l=1Cl, where Cl are simple superalgebras. Cl is isomorphic either to a matrix superalgebra Mkl,ml(F ) = Mkl+ml(F ) with an elementary grading (Cl)¯0 =(cid:26) kl ml kl ml 0 (cid:18) ∗ 0 ∗ (cid:19)(cid:27) , (Cl)¯1 =(cid:26) kl ml kl ml ∗ (cid:18) 0 ∗ 0 (cid:19)(cid:27) , with the grading induced by the natural Z/2Z-grading of the group algebra (Cl)¯0 = or to a matrix algebra over the group algebra Mkl(eF [Z/2Z]) ≃ Mkl(eF [cc2 = 1]) Mkl(eF ), (Cl)¯1 = Mkl(eF ) · c. Thus A′ = B′ ⊕ J can be assumed our superalgebra. Any Z/2Z-simple compo- nent Cl of the semisimple graded subalgebra B′ has the unit element of the even degree. It gives the minimal orthogonal central idempotent εl of the algebra B′. D is a homogeneous basis of B′. It is clear that any element r ∈ J can be uniquely l′, l′′ = 1, . . . , p + 1. represented as a sum of elements of graded subspaces εl′Jεl′′, Moreover εla = 0 (and aεl = 0) for any a ∈ εl′Jεl′′ with l 6= l′ (l 6= l′′), l = 1, . . . , p. ✷ The next construction is useful to extend the previous result for any field of zero characteristic in some sense. Take a superalgebra B (not necessarily without unit). We denote by B# = B ⊕ F · 1F the superalgebra with the adjoint unit 1F . 5 Let us take a finite dimensional superalgebra A = B ⊕ J(A) with a maximal Z/2Z-graded semisimple subalgebra B and the Jacobson radical J(A). Consider a q i#, and define on it the Z/2Z-grading by the equalities q i# are q i# q ) the two-sided graded ideal of q ) = Z/2Z-graded subalgebra eB ⊆ B, and a positive integer number q. Consider the free product eB# ∗F F hXZ2 degZ2(u1 · · · us) = (degZ2 u1) + · · · + (degZ2 us), where ui ∈ eB#S F hXZ2 q ) be the graded subalgebra of eB# ∗F F hXZ2 homogeneous elements. Let eB(XZ2 generated by the set eBS F hXZ2 eB(XZ2 q . Particularly, it is clear that eB(XZ2 eB ⊕ (XZ2 Given a Z2T-ideal Γ denote by Γ(eB(XZ2 eB(XZ2 q )) the two-sided graded verbal ideal of q ) generated by results of all appropriate evaluations of polynomials from Γ. Take any s ∈ N, and consider the quotient algebra q ) generated by the set of variables XZ2 q i. Denote by (XZ2 q ). q )) + (XZ2 q )s). (4) Rq,s(eB, Γ) = eB(XZ2 q )/(Γ(eB(XZ2 q )) + (XZ2 q )s), and is nilpotent of degree less or equal to s. q )/(Γ(eB(XZ2 If q ≥ grk(J(A)), and s ≥ nd(A) then IdZ2(Rq,s(A)) = IdZ2(A). Lemma 2 Take any natural numbers q, s and a Z2T-ideal Γ such that Γ ⊆ IdZ2(A). Denote also Rq,s(A) = Rq,s(B, IdZ2(A)) for Γ = IdZ2(A), eB = B. The algebra Rq,s(eB, Γ) is a finite dimensional superalgebra with the ideal of graded identities IdZ2(Rq,s(eB, Γ)) ⊇ Γ. Moreover Rq,s(eB, Γ) = B ⊕ J(Rq,s(eB, Γ)). Here B is a maximal semisimple Z/2Z-graded subalgebra of Rq,s(eB, Γ), and B ∼= eB. The Jacobson radical of Rq,s(eB, Γ) is equal to (XZ2 q ), and eBT I = (0). I is a Proof. It is clear that I = Γ(eB(XZ2 graded ideal of eB(XZ2 q ) → Rq,s(eB, Γ) we obtain B = ψ(eB) ∼= eB. the canonical homomorphism ψ : eB(XZ2 Hence Rq,s(eB, Γ) = B ⊕ ψ((XZ2 nilpotent ideal of the algebra Rq,s(eB, Γ) of degree at most s. It is clear that (XZ2 is Z/2Z-graded and finite dimensional. Then Rq,s(eB, Γ) is also a finite dimensional algebra with the Jacobson radical J(Rq,s(eB, Γ)) = ψ((XZ2 IdZ2(Rq,s(eB, Γ)) for any q, s ∈ N. Let us take Γ = IdZ2 (A), eB = B. Suppose that the Jacobson radical J(A) of ri = Pθ∈Z/2Z riθ, where riθ ∈ J(A)T Aθ (i = 1, . . . , ν). Consider the map ϕ : xiθ = xiθ + I 7→ riθ (i = 1, . . . , ν). Assume that ϕ(b + I) = b for any b ∈ B. If q ≥ ν, and s ≥ nd(A) then ϕ can be extended to a surjective graded homomorphism ϕ : Rq,s(A) → A. Therefore Γ = IdZ2(A) ⊇ IdZ2(Rq,s(A)). ✷ q ). Hence Rq,s(eB, Γ) = eB(XZ2 the algebra A is generated as an algebra by the set {r1, . . . , rν }. Then we have q )/I is the maximal q )/I q )/I is a superalgebra. Then for q )). It is clear that Γ ⊆ q )), where ψ((XZ2 q )) = (XZ2 q )) + (XZ2 q )s ⊆ (XZ2 Definition 2 F -algebra A is called representable if A can be embedded into some algebra C that is finite dimensional over an extension eF ⊇ F of the base field F. 6 Lemma 3 If a Z/2Z-graded F -algebra A is representable then there exists a finite dimensional over F Z/2Z-graded F -algebra U such that IdZ2(A) = IdZ2(U ). Proof. Suppose that A is isomorphic to an F -subalgebra B of a finite dimen- It is clear from the classification of simple superalgebras (see also Lemma 1) sional eF -algebra eB. We can assume that the extension eF ⊇ F is algebraically closed, and B is Z/2Z-graded. Consider the algebra eU = eU¯0 ⊕ eU¯1, where eUθ = (eF Bθ) ⊗ eF eF θ ⊆ eB ⊗ eF eF [Z/2Z]. eU is a finite dimensional Z/2Z-graded eF -algebra. And IdZ2(eU ) = IdZ2(B) = IdZ2(A). Let eCl = eCl¯0 ⊕eCl¯1 be a Z/2Z-simple component in the decomposition (1) of the algebra eU . that eCl contains a finite dimensional over F Z/2Z-graded simple F -subalgebra Cl = Cl¯0 ⊕ Cl¯1 satisfying eClθ = eF Clθ for any θ ∈ Z/2Z (l = 1, . . . , p). More precisely, Cl = Mkl,ml(F ) if eCl = Mkl,ml(eF ), and Cl = Mkl(F [c]) if eCl = Mkl(eF [c]). Moreover the algebras eCl and Cl have the same canonic homogeneous base of type (2) over eF Let us take B = C1 × · · · × Cp, Γ = IdZ2(A), q = dim eF J(eU ), s = nd(eU ). Then F. And IdZ2(U ) = Γ = IdZ2(eU ) (Lemma 2). the F -algebra U = Rq,s(B, Γ) defined by (4) is Z/2Z-graded finite dimensional over ✷ Definition 3 We say that an F -finite dimensional superalgebra A′ has an elemen- tary decomposition if it satisfies the assertion of Lemma 1. and F respectively. It is clear that the direct product of superalgebras with elementary decomposition is the superalgebra with elementary decomposition. Also if F is algebraically closed then any finite dimensional F -superalgebra has an elementary decomposition. Corollary 4 Let F be a field of characteristic zero. Any finite dimensional F - superalgebra is Z2PI-equivalent to a finite dimensional F -superalgebra with elemen- tary decomposition. Proof. A finite dimensional F -superalgebra A can be naturally embedded to the identities are considered over the field F. superalgebra eA = A ⊗F eF preserving graded identities. We assume here that degZ2 a ⊗ α = degG a, for all a ∈ A, α ∈ eF . The superalgebra eA is finite di- mensional over eF . By Lemma 3 there exists a finite dimensional F -superalgebra A′ with elementary decomposition such that IdZ2(A′) = IdZ2(eA) = IdZ2(A). Where all DS U. Particularly, for multilinear graded polynomials it is enough to consider only evaluations by elements of the set DS U. Such evaluation of a multilinear graded ✷ Therefore considering graded identities of a finite dimensional superalgebra A we always can assume that A has a form (1), and a basis of A can be chosen in the set polynomial is called elementary. Elements of the set D are called semisimple, and elements of the set U are radical. Definition 4 Let A = B ⊕ J be a finite dimensional superalgebra, B = ⊕θ∈Z/2ZBθ a maximal semisimple Z/2Z-graded subalgebra of A, and J(A) = J the Jacobson 7 radical of A. We denote by dimsZ2A = (dim B¯0, dim B¯1), and by nd(A) the nilpotency degree of the radical J. Consider as principal the next parameter of A parZ2(A) = (dimsZ2A; nd(A)). The 4-tuple cparZ2(A) = (parZ2(A); dim J(A)) is called complex parameter of A. A finite dimensional superalgebra A is nilpotent if and only if dimsZ2A = (0, 0). Recall that n-tuples of numbers are ordered lexicographically. Then for any nonzero graded two-side ideal I ✂ A we have that cparZ2(A/I) < cparZ2(A). 3 Kemer index. Let f = f (y1, . . . , yk, x1, . . . , xn) ∈ F hXZ2 i be a polynomial linear in all variables of the set Y = {y1, . . . , yk}. The polynomial f is alternating in Y, if f (yσ(1), . . . , yσ(k), x1, . . . , xn) = (−1)σf (y1, . . . , yk, x1, . . . , xn) holds for any permutation σ ∈ Symk. For any polynomial g(y1, . . . , yk, x1, . . . , xn) that is linear in Y = {y1, . . . , yk}, it is possible to construct a polynomial alternating in Y by setting (−1)σg(yσ(1), . . . , yσ(k), x1, . . . , xn). f (y1, . . . , yk, x1, . . . , xn) = AY (g) = Xσ∈Symk tor. Any polynomial f alternating in Y can be decomposed as f =Ps The corresponding mapping AY is a linear transformation, we called it the alterna- i=1 αiAY (ui), where uis are monomials of f, αi ∈ F. The properties of graded alternating polyno- mials are similar to that of non-graded polynomials. We consider only polynomials alternating in homogeneous sets of variables. Given a pair t = (t1, t2) ∈ N2 0 we say that a graded polynomial f ∈ F hXZ2 i has a collection of t-alternating graded variables (f is t-alternating) if f (Y¯0, Y¯1, X) is linear in Y = Y¯0 ∪ Y¯1, and f is alternating in each set Yθ = {y1θ, . . . , ytθθ} ⊆ Xθ, Yθ = tθ, θ ∈ Z/2Z. 0. Suppose that τ1, . . . , τs ∈ N2 Definition 5 Fix any t = (t¯0, t¯1) ∈ N2 0 are some (possibly different) pairs satisfying the conditions τj = (τj¯0, τj¯1) > t = (t¯0, t¯1) for all j = 1, . . . , s. Let f ∈ F hXZ2 i be a multihomogeneous graded polynomial. Suppose that f = f (Y1, . . . , Ys+µ; X) has s collections of τj-alternating variables Yj = Yj¯0∪Yj¯1 (j = 1, . . . , s), and µ collections of t-alternating variables Yj = Yj¯0 ∪ Yj¯1 (j = s + 1, . . . , s + µ). We assume that all these sets are disjoint. Then we say that f is of the type (t; s; µ) = (t1, t2; s; µ). Here Yjθ ⊆ Xθ with Yjθ = τjθ for any j = 1, . . . , s, and Yjθ = tθ for any j = s + 1, . . . , s + µ. Observe that a multihomogeneous polynomial f of a type (t; s; µ) is also of the type (t; s′; µ′) for all s′ ≤ s, and µ′ ≤ µ. Particularly, any nontrivial graded multi- linear polynomial of degree s has the type (0, 0; s; µ) for any µ ∈ N0. 8 Definition 6 Given a Z2T-ideal Γ ✂ F hXZ2 i the parameter β(Γ) = (t¯0, t¯1) is the greatest lexicographic pair t = (t¯0, t¯1) ∈ N2 0 such that for any s ∈ N there exists a graded polynomial f /∈ Γ of the type (t; 0; s). The parameter β(Γ) is well defined for any proper Z2T-ideal Γ ✂ F hXZ2i of a finitely generated Z/2Z-graded PI-algebra. In this case Γ contains the ordinary Capelli polynomial of some order d ([11]). Hence any graded polynomial f of the type (t¯0, t¯1; 0; s) belongs to Γ if t¯0 ≥ d or t¯1 ≥ d. The next parameter is also well defined. Definition 7 Given a nonnegative integer µ let γ(Γ; µ) = s ∈ N be the smallest integer s > 0 such that any graded polynomial of the type (β(Γ); s; µ) belongs to Γ. γ(Γ; µ) is a positive non-increasing function of µ. Let us denote the limit of this function by γ(Γ) = lim µ→∞ γ(Γ; µ) = γ(Γ) for any µ ≥bµ. γ(Γ; µ) ∈ N. Then ω(Γ) is the smallest number bµ such that Definition 8 We call by the Kemer index of a Z2T-ideal Γ the lexicographically ordered collection indZ2(Γ) = (β(Γ); γ(Γ)). Notice that indZ2(Γ) is greater than (0, 0; 1) for any proper Z2T-ideal Γ. We assume also that indZ2(F hXZ2 i) = (0, 0; 1). Let us denote ω(A) = ω(IdZ2(A)), γ(A; µ) = γ(IdZ2(A); µ), indZ2(A) = indZ2 (IdZ2(A)) for a finitely generated PI-superalgebra A. It is clear that A is a nilpotent superalgebra of class s if and only if indZ2(A) = (0, 0; s). Particularly indZ2(A) = parZ2(A) for a nilpotent algebra A. In general case we have Lemma 5 indZ2(A) ≤ parZ2(A) for any finite dimensional superalgebra A. Proof. Let us denote dimsZ2A = (t¯0, t¯1). Suppose that βθ > tθ for some θ = ¯0, ¯1, and a multilinear polynomial f has the type (β¯0, β¯1; 0; nd(A)). Then for any j = 1, . . . , nd(A) all θ-variables of the alternating set Yj θ of f can not be evaluated only by semisimple elements with nonzero result. Therefore f ∈ IdZ2(A), and indZ2(A) ≤ parZ2(A). ✷ Definition 9 Given a Z2T-ideal Γ, and any µ ∈ N0 a multihomogeneous polynomial f ∈ F hXZ2 i is called µ-boundary for Γ if f /∈ Γ, and f has the type (β(Γ); γ(Γ)−1; µ). Let us denote by Sµ(Γ) the set of all µ-boundary polynomials for Γ. Denote also Sµ(A) = Sµ(IdZ2(A)), Kµ(Γ) = Z2T [Sµ(Γ)], Kµ,A = Kµ(IdZ2(A)) = Z2T [Sµ(A)]. Observe that if the Kemer index is well defined for a Z2T-ideal Γ then Γ has multilinear boundary polynomials for all µ ∈ N0. Moreover a polynomial f belongs to Sµ(Γ) if and only if its full multilinearization ef belongs to Sµ(Γ). Definitions 6, 7 immediately imply the next basic properties of the Kemer index and boundary polynomials. 9 Lemma 6 Given Z2T-ideals Γ1, Γ2 admitting the Kemer index if Γ1 ⊆ Γ2 then indZ2 (Γ1) ≥ indZ2(Γ2). Lemma 7 Consider Z2T-ideals Γ, Γ1, . . . , Γρ admitting the Kemer index. Assume that indZ2(Γi) < indZ2(Γ) for all i = 1, . . . , ρ. Then there exists bµ ∈ N0 such that Sµ(Γ) ⊆ ρTi=1 Γi for any µ ≥bµ. Proof. Let us denote indZ2(Γi) = (βi, γi), 1 ≤ i ≤ ρ, that βi < β for i = 1, . . . , ρ′, and βi = β, γi < γ for i = ρ′ + 1, . . . , ρ (0 ≤ ρ′ ≤ ρ). indZ2(Γ) = (β, γ). Assume If β > βi (1 ≤ i ≤ ρ′) then there exists µi such that any polynomial of the type (β; 0; µi) belongs to Γi. If βi = β and γi < γ (i = ρ′ + 1, . . . , ρ) then any polynomial of the type (β; γi; µi) belongs to Γi for all µi ≥ ω(Γi). Thus Sµ(Γ) ⊆ µ ≥ max{µ1, . . . , µρ′, ω(Γρ′+1), . . . , ω(Γρ)}. Lemmas 6, 7 jointly give the next properties. ρTi=1 Γi for any ✷ Lemma 8 Given Z2T-ideals Γ1, Γ2 admitting the Kemer index indZ2(Γ1 ∩ Γ2) = maxi=1,2 indZ2(Γi). Lemma 9 For all finitely generated PI-superalgebras Ai max 1≤i≤ρ indZ2(Ai). indZ2(A1 × · · · × Aρ) = Lemma 10 Given Z2T-ideals Γ1, Γ2 admitting the Kemer index, and satisfying Γ1 ⊆ Γ2 one of the following alternatives takes place: 1. indZ2(Γ1) = indZ2(Γ2), and Sµ(Γ1) ⊇ Sµ(Γ2), Kµ(Γ1) ⊇ Kµ(Γ2) ∀µ ∈ N0; 2. indZ2(Γ1) > indZ2(Γ2), and Sµ(Γ1) ⊆ Γ2 for some µ ∈ N0. Moreover in the case Γ1 ⊆ Γ2 the conditions indZ2(Γ1) > indZ2(Γ2) and Sµ(Γ1) ⊆ Γ2 are equivalent for some µ ∈ N0. Proof. By Lemma 6 we have that indZ2(Γ1) ≥ indZ2(Γ2). The conditions indZ2(Γ1) = indZ2 (Γ2) = (β, γ), Γ1 ⊆ Γ2 imply that Sµ(Γ1) ⊇ Sµ(Γ2), Kµ(Γ1) ⊇ Kµ(Γ2) ∀µ ∈ N0. If indZ2(Γ1) > indZ2(Γ2) then Sµ(Γ1) ⊆ Γ2 holds for some µ ∈ N0 by Lemma 7. And visa versa if Γ1 ⊆ Γ2, and Sµ(Γ1) ⊆ Γ2 for some µ ∈ N0 then Sµ(Γ2) ⊆ Sµ(Γ1) ⊆ Γ2 gives a contradiction. Therefore in this case we obtain that indZ2(Γ1) > indZ2 (Γ2). ✷ Similarly the conditions indZ2(Γ1) = indZ2(Γ2) and Sµ(Γ1) ⊇ Sµ(Γ2) are also equivalent in the case Γ1 ⊆ Γ2. The last lemma has the following corollary. 10 Lemma 11 Given an integer µ ∈ N0 and finitely generated PI-superalgebras A1, . . . , Aρ with the same Kemer index indZ2(Ai) = κ for all i = 1, . . . , ρ it holds Sµ(A1 × · · · × Aρ) = Sµ( IdZ2(Ai)) = Kµ,A1×···×Aρ = Kµ( ρ\i=1 ρXi=1 ρ\i=1 Sµ(Ai), ρ[i=1 IdZ2(Ai)) = Kµ(IdZ2(Ai)) = ρXi=1 Kµ,Ai. Lemma 12 Given a Z2T-ideal Γ, and a non-positive integer µ ∈ N0 we have that indZ2 (Γ) > indZ2(Γ + Kµ(Γ)). Proof. Since Γ ⊆ Γ + Kµ(Γ) and Sµ(Γ) ⊆ Kµ(Γ) ⊆ Γ + Kµ(Γ) then the assertion immediately follows from Lemma 10. ✷ 4 Z2PI-reduced algebras. Definition 10 A finite dimensional superalgebra A with elementary decomposition is Z2PI-reduced if there do not exist finite dimensional superalgebras A1, . . . , A IdZ2(Ai) = IdZ2(A), and cparZ2(Ai) < with elementary decomposition such that cparZ2(A) for all i = 1, . . . , . Ti=1 It is clear that a nilpotent finite dimensional superalgebra A is Z2PI-reduced if and only if it has the minimal dimension among all nilpotent finite dimensional superalgebras satisfying the same graded identities as A. Lemma 13 Any simple finite dimensional superalgebra A with elementary decompo- sition is Z2PI-reduced, and indZ2(A) = parZ2(A) = (t¯0, t¯1; 1) (i.e. β(A) = dimsZ2A, and γ(A) = nd(A) = 1). Proof. Suppose that dimsZ2A = (t¯0, t¯1). Any finite dimensional Z/2Z-graded simple algebra is semisimple, nd(A) = 1, and cparZ2(A) = (dimsZ2A; 1; 0). It follows from Lemma 5 that β(A) ≤ dimsZ2A = (t¯0, t¯1). Hence it is enough to construct a polyno- mial of the type (t¯0, t¯1; 0; s) that is not a graded identity of A for any s ∈ N. Fix any natural number s. Since A has an elementary decomposition then A = Mk,m(F ) or A = Mk(F [c]). Consider the case A = Mk,m(F ) at first. In this case A has a homogeneous in the grading basis of the type {Eiji, j = 1, . . . , k + m}, where Eij are the matrix units. Consider s sets of (k + m)2 distinct graded variables Yd = {yd,(ij) ∈ XZ2 i, j = 1, . . . , k + m}. Here any graded variable yd,(ij) corresponds to the matrix unit Eij, and degZ2 yd,(ij) = degZ2 Eij in the superalgebra Mk,m(F ) (d = 1, . . . , s). Denote by Yd,θ = {y ∈ Yd deg Z2y = θ}. Then for any d we have that Yd,θ = tθ = dim Aθ, θ ∈ Z/2Z. Let us take also the set of graded variables Z = {z(ij) ∈ XZ2 degZ2 z(ij) = degZ2 Eij, i, j = 1, . . . , k + m}. We assume that Z is disjoint with 11 Y = Ss d=1 Yd. We say that the variable z(j1i2) connects the variables yd,(i1j1) and yd,(i2j2). Let us consider for any fixed d the graded monomial wd that is the product of all variables yd,(isjs) connected by variables z(ij) wd = yd,(11)z(11)yd,(12)z(21)yd,(13) · · · yd,(i1j1)z(j1i2)yd,(i2j2)z(j2i3)yd,(i3j3) · · · yd,(k+mk+m−1)z(k+m−1k+m)yd,(k+mk+m), (d = 1, . . . , s); and the monomial W (Y, Z) = z(11) ·(cid:0) sYd=1 (wdz(k+m1))(cid:1)(cid:1). Then the polynomial f (Y, Z) =(cid:16)Qs in any set Yd = Yd,¯0 ∪ Yd,¯1 (d = 1, . . . , s). d=1(AYd,¯0 AYd,¯1)(cid:17) W (Y, Z) is (t¯0, t¯1)-alternating Notice that for any d and θ the set Yd,θ contains at most 1 variable yd,(ij) for the same pair (i, j). Consider the evaluation yd,(ij) = Eij, z(ij) = Eij (∗) of the polynomial f. Since the variables z(ij) fix the positions for indices of the variables y then (∗) gives a nonzero result (5) = E11 6= 0. (6) Similarly, we construct the polynomial f (Y, Z), and the corresponding nonzero evaluation for the case A = Mk(F [c]). In this case A has a homogeneous in the grad- ing basis of the type {Eij, Eijci, j = 1, . . . , k}, where Eij are the matrix units, and c is the central element of A satisfying c2 = 1. Here degZ2 Eij = ¯0, and degZ2 Eijc = ¯1 for all i, j = 1, . . . , k. The set Yd = {yd,¯0,(ij), yd,¯1,(ij) ∈ XZ2 i, j = 1, . . . , k} contains 2k2 graded variables. A graded variable yd,¯0,(ij) corresponds to the basic element Eij, and yd,¯1,(ij) corresponds to Eij · c, degZ2 yd,θ,(ij) = θ (d = 1, . . . , s). Then Yd,θ = {yd,θ,(ij)i, j = 1, . . . , k}, and Yd,θ = tθ = dim Aθ = k2, θ ∈ Z/2Z. All connecting variables z(ij) have even degree. In this case the variable z(j1i2) connects variables yd,θ,(i1j1) and yd,ξ,(i2j2) for any θ, ξ ∈ Z/2Z. The graded monomial wd,θ is the product of all variables yd,θ,(isjs) connected by variables z(ij) for any fixed θ ∈ Z/2Z and d wd,θ = yd,θ,(11)z(11)yd,θ,(12)z(21)yd,θ,(13) · · · yd,θ,(i1j1)z(j1i2)yd,θ,(i2j2)z(j2i3)yd,θ,(i3j3) · · · yd,θ,(kk−1)z(k−1k)yd,θ,(kk), (d = 1, . . . , s, θ = ¯0, ¯1). f(cid:12)(cid:12)(∗) = W(cid:12)(cid:12)(∗) Then the monomial W (Y, Z) = z(11) ·(cid:0) sYd=1 Take the polynomial f (Y, Z) =(cid:16)Qs ternating in any set Yd =Sθ∈Z/2Z Yd,θ (d = 1, . . . , s). d=1 AYd,¯0 · AYd,¯1(cid:17) W (Y, Z). f (Y, Z) is (t¯0, t¯1)-al- For any d and θ the set Yd,θ contains at most 1 variable yd,θ,(ij) for the same pair (i, j) then the evaluation yd,¯0,(ij) = Eij, yd,¯1,(ij) = Eijc, z(ij) = Eij (∗) of f gives a nonzero result (wd¯0z(k1)wd¯1z(k1))(cid:1). (7) f(cid:12)(cid:12)(∗) = W(cid:12)(cid:12)(∗) = E11 · ck2 6= 0. (8) 12 Notice that ck2 = c if k is odd, and ck2 = 1 otherwise. In both of the cases f /∈ IdZ2(A), and this is the desired polynomial of the type (t¯0, t¯1; 0; s). Therefore β(A) = dimsZ2A. By Lemma 5 the conditions dim J(A) = 0, and indZ2(A) = parZ2(A) imply that A is Z2PI-reduced. ✷ Z2PI-reduced algebras exist and possess the next properties. Lemma 14 Given a Z2PI-reduced algebra with the Wedderburn-Malcev decompo- sition (1) A = (C1 × · · · × Cp) ⊕ J we have Cσ(1)JCσ(2)J · · · JCσ(p) 6= 0 for some σ ∈ Symp. pTi=1 Proof. Suppose that Cσ(1)JCσ(2)J · · · JCσ(p) = 0 for any σ ∈ Symp. Consider the Cj) ⊕ J(A) of Z/2Z-graded subalgebras with elementary decomposition Ai = ( Q1≤j≤p j6=i the superalgebra A. Then we have IdZ2(A) = IdZ2(Ai), and dimsZ2Ai < dimsZ2A for any i = 1, . . . , p. This contradicts to the definition of Z2PI-reducible algebra. ✷ Particularly we have nd(A) ≥ p for a Z2PI-reduced algebra A. Lemma 15 Any finite dimensional superalgebra is Z2PI-equivalent to a finite direct product of Z2PI-reduced algebras. Proof. By Lemma 4 any finite dimensional superalgebra is Z2PI-equivalent to a finite dimensional superalgebra A with elementary decomposition. If A is not Z2PI-reduced then it is Z2PI-equivalent to a finite direct product of finite dimen- sional elementary decomposed superalgebras with the complex parameters less than cparZ2(A). We apply this process inductively to all multipliers. The set N4 0 with the lexicographical order satisfies the descending chain condition. Therefore this process of decomposition will stop after a finite number of steps. ✷ Lemma 15 along with Lemmas 10, 11 implies Lemma 16 Any finite dimensional superalgebra A is Z2PI-equivalent to a direct product O(A) × Y(A) of finite dimensional superalgebras with elementary decom- position satisfying indZ2(A) = indZ2(O(A)) > indZ2(Y(A)). Moreover O(A) = A1×· · ·×Aρ, where Ai are Z2PI-reduced superalgebras, and indZ2(Ai) = indZ2(A) for i=1 Sµ(Ai) ⊆ all i = 1, . . . , ρ. There exists bµ ∈ N0 such that Sµ(A) = Sµ(O(A)) =Sρ IdZ2(Y(A)) holds for any µ ≥bµ. Proof. Suppose that A is Z2PI-equivalent to a direct product A1 × · · · × Aρ of Z2PI-reduced superalgebras. By Lemma 9 indZ2(A) = max1≤i≤ρ indZ2(Ai). Assume that the Kemer index has the maximal value for Ai with i = 1, . . . , ρ. Then the superalgebras O(A) = A1 × · · · × Aρ, Y(A) = Aρ+1 × · · · × Aρ satisfy the assertion of the lemma. ✷ Definition 11 O(A) is called the senior part of A, Y(A) is called the junior part of A. The algebras Ai (i = 1, . . . , ρ) are called the senior components of A. bµ(A) is the minimal bµ ∈ N0 satisfying the assertion of Lemma 16. 13 The next lemma shows the relation between the Kemer index and the parameters of a Z2PI-reduced superalgebra. Lemma 17 Given a Z2PI-reduced superalgebra A we have β(A) = dimsZ2A. Proof. For a nilpotent superalgebra the assertion is trivial. Consider any non- nilpotent Z2PI-reduced algebra A with dimsZ2A = (t¯0, t¯1). By Lemma 5 it is enough to find a polynomial of the type (t¯0, t¯1; 0; s) that is not a graded identity of A for any s ∈ N. Assume that A has the elementary decomposition (1). For any l = 1, . . . , p consider the graded monomial Wl(Y(l), Z(l)) of the type (5) or (7) constructing for the simple component Cl (see Lemma 13). Wl(Y(l), Z(l)) depends on the disjoint d=1(Y(l),(d,¯0) ∪ Y(l),(d,¯1)), and Z(l) = {z(l),(ij), i, j = 1, . . . , sl}. Where Y(l) = {y(l),(d,(ij)) ∈ XZ2 1 ≤ i, j ≤ sl; 1 ≤ d ≤ s} for Cl = Mkl,ml(F ), and Y(l) = {y(l),(d,θ,(ij)) ∈ XZ2 θ = ¯0, ¯1; 1 ≤ i, j ≤ sl; 1 ≤ d ≤ l=1 Y(l),(d,θ), and degZ2 z(l),(ij) = degZ2 Elij in Cl. It is clear that Yd,θ = tθ for any d = 1, . . . , s, θ ∈ Z/2Z. sets of graded variables Y(l) =Ss s} for Cl = Mkl(F [c]). We have that Y(l),(d,θ) ⊆ Xθ, Yd,θ = Sp Then the appropriate elementary evaluation of the word Wl(Y(l), Zl) in Cl (see Lemma 13) is equal to the non-zero element El11¯cl, where ¯cl = 1 or ¯cl = c. By Lemma 14 we can assume that A contains an element ε1r1ε2 · · · εp−1rp−1εp 6= 0, where rl ∈ J are some Z/2Z-homogeneous radical elements, and εl = Elilil is slPil=1 the unit of Cl (l = 1, . . . , p). Let us take p nontrivial graded polynomials fl(eX(l), Y(l), Z(l)) = xl,(il)Wl(Y(l), Z(l)) xl,(il,¯cl) (9) slXil=1 depending on the additional set of graded variables eX(l) = {xl,(il), xl,(il,¯cl) ∈ XZ2 1 ≤ xl,(il,¯cl) = degZ2 El1il ¯cl, }. Notice that fl is il ≤ sl; degZ2 xl,(il) = degZ2 Elil1, degZ2 not a multihomogeneous polynomial, although it is linear in Y(l), and Z(l). But fl is homogeneous in the grading of even degree, since any monomial xl,(il)Wl(Y(l), Z(l)) xl,(il,¯cl) has even degree (it has the same Z/2Z-degree as Elilil in Cl). Then the polynomial f (eX, Y, Z) = sYd=1 AYd,¯0 AYd,¯1! (f1x1f2x2 · · · xp−1fp) (10) is linear in Y S Z =(cid:0)Sp l=1 Y(l)(cid:1)S(cid:0)Sp l=1 Z(l)(cid:1) , and alternating in any set Yd,θ ⊆ Xθ, θ ∈ Z/2Z, d = 1, . . . , s. It follows from Lemma 13 that the evaluation y(l),(d,¯0,(iljl)) = εlEliljlεl, z(l),(iljl) = εlEliljlεl, xl,(il,¯cl) = εl(El1il ¯cl)εl, l = 1, . . . , p; q = 1, . . . , p − 1; d = 1, . . . , s, y(l),(d,(iljl)) = εlEliljlεl, y(l),(d,¯1,(iljl)) = εl(Eliljlc)εl, xl,(il) = εlElil1εl, xq = εqrqεq+1, 1 ≤ il, jl ≤ sl, 14 of the polynomial f (eX, Y, Z) is equal to ε1r1ε2 · · · εp−1rp−1εp 6= 0. Therefore we obtain f /∈ IdZ2(A). Hence at least one multihomogeneous com- ponent f of f also is not a graded identity of A. f is alternating in any set Yd,θ (θ ∈ Z/2Z, d = 1, . . . , s). Thus f is the required polynomial. ✷ 5 Exact polynomials. Definition 12 Given a finite dimensional superalgebra A with elementary decom- position an elementary evaluation (a1, . . . , an) ∈ An (namely, ai ∈ DS U ⊆ A (2), (3)) is called incomplete if there exists j = 1, . . . , p such that {a1, . . . , an}\(Cj[{εjuεl, εluεju ∈ U, l = 1, . . . , p + 1}) = ∅. Otherwise (a1, . . . , an) is called complete. Definition 13 An elementary evaluation (a1, . . . , an) ∈ An is called thin if it con- tains less than nd(A) − 1 radical elements (not necessarily distinct). Definition 14 We say that a multilinear graded polynomial f (x1, . . . , xn) ∈ F hXZ2 i is exact for a finite dimensional superalgebra A with elementary decomposition if f (a1, . . . , an) = 0 holds in A for any thin or incomplete evaluation (a1, . . . , an) ∈ An. Lemma 18 If A is a Z2PI-reduced superalgebra then any multilinear polynomial of the type (dimsZ2A; nd(A) − 1; 0) is exact for A. It is clear that a Z2PI-reduced algebra either is not semisimple or is Proof. Z/2Z-simple. For a Z/2Z-graded simple finite dimensional algebra any multilin- ear graded polynomial can be assumed exact. A multilinear polynomial of the type (0, 0; nd(A) − 1; 0) has degree greater or equal to (nd(A) − 1). Hence it is assumed to be exact for a nilpotent algebra A. Suppose that A is not nilpotent, and f is a multilinear polynomial of the type (dimsZ2A; nd(A) − 1; 0). In a thin evaluation at least one of s = nd(A) − 1 collections of τj-alternating variables of f will be completely replaced by semisimple elements. Since τj > dimsZ2A for any j = 1, . . . , s then the result of such evaluation will be zero. Any simple Z/2Z-graded component Cl of A (Lemma 1) has the unit εl ∈ Cl¯0. Hence dimF Cl¯0 > 0 for any l = 1, . . . , p. Therefore an incomplete evaluation can not contain all semisimple elements of the even degree from the base D (2). Taking into account the conditions τj > dimsZ2A for any j = 1, . . . , s we obtain that at least two variables of every collection of τj-alternating variables of f must be substituted by radical elements, otherwise the result of the substitution will be zero. Thus in any case the result of an incomplete evaluation of the polynomial f is zero. ✷ Lemma 19 Any nonzero Z2PI-reduced superalgebra A has an exact polynomial, that is not a graded identity of A. 15 j6=i Cj) ⊕ J(A), Proof. For a nilpotent superalgebra A the assertion follows from Lemma 18. Sup- pose that A is a non-nilpotent superalgebra with the decomposition (1). Con- i = 1, . . . , p. Take q = dimF J(A), sider its subalgebras Ai = ( Q1≤j≤p s = nd(A)−1. Then by Lemma 2 eA = A1 ×· · ·×Ap ×Rq,s(A) is a finite dimensional superalgebra with elementary decomposition satisfying IdZ2(A) ⊆ IdZ2(eA). Let {r1, . . . , rq} ⊆ U be a homogeneous basis of J(A) (3), degZ2 ri = θi (i = 1, . . . , q). Consider the map ϕ defined by the next equalities ϕ(xiθi ) = ri for i = 1, . . . , q, and ϕ(b) = b for any b ∈ B. ϕ can be extended to a surjective graded homomorphism q ) → A. It follows that any multilinear polynomial f ∈ IdZ2(Rq,s(A)) ϕ : B(XZ2 It is also clear that any incom- is turned into zero under any thin substitution. plete substitution in a multilinear graded polynomial f ∈ IdZ2(×p i=1Ai) yields zero. cparZ2(Ai) < cparZ2(A) (1 ≤ i ≤ p), and cparZ2(Rq,s(A)) < cparZ2(A). Since A is Therefore, any multilinear polynomial f ∈ IdZ2(eA) is exact for A. Remark that Z2PI-reduced then IdZ2(A) $ IdZ2(eA). Any multilinear graded polynomial f such that f ∈ IdZ2(eA), and f /∈ IdZ2(A) satisfies the assertion of the lemma. Lemma 20 Let A be a finite dimensional superalgebra with an elementary decom- position, h an exact polynomial for A, and ¯a ∈ An is any complete evaluation of h containing s = nd(A) − 1 radical elements. Then for any µ ∈ N0 there exist a graded polynomial hµ ∈ Z2T [h], and an elementary evaluation ¯u of hµ by elements of A such that: ✷ 2. hµ(¯u) = h(¯a), 1. hµ(eY1, . . . ,eYs+µ, eX,eZ) is τj-alternating in any set eYj with τj > β = dimsZ2A for all j = 1, . . . , s, and is β-alternating in any eYj for j = s + 1, . . . , s + µ (all the sets eYj, eX, eZ are disjoint), 3. all the variables of eXSeZ are replaced by semisimple elements. consequence hµ(eY1, . . . ,eYs+µ) ∈ Z2T [h] that is alternating in all eYj as required (we assume here that eXSeZ = ∅), and replace the variables of the alternating sets eYj by equal elements. Particularly, from conditions nd(A) = 1, p ≥ 2 it follows that h(¯a) = 0. Proof. If h(¯a) = 0 then the assertion of lemma is trivial. It is sufficient to take any Assume that h(¯a) 6= 0. Consider the case nd(A) > 1, p ≥ 2 in the decomposition (1) of the superalgebra A. We can assume for simplicity that the evaluation ¯a has the , b1, . . . , bn−s), form ¯a = (ε1r1εl′′ where 1 ≤ l′ s for any 1 ≤ s ≤ q. Namely, in a complete evaluation all minimal orthogonal idempotents ε1, . . . , εp appear in mixed radical elements (in elements of the type εl′rlεl′′ ∈ U (3) 6= l′′). We assume that they appear in the first q mixed radical elements. with l′ s ≤ p + 1, {1, . . . , p} ⊆ {l′ s 1 ≤ s ≤ q} for some q ≤ s, r2ε2, . . . , εp−1rqεp, εl′ , . . . , εl′ s 6= l′′ l′ rq+1εl′′ s, l′′ rsεl′′ s s , εl′ 2 1 s, l′′ q+1 q+1 16 The last elements n − s of the evaluation ¯a b1, . . . , bn−s ∈ D (2) are supposed to be semisimple, and the first s elements εl′ s ∈ U (3) are radical. srsεl′′ dimsZ2A for any d. d=1(Y(l),(d,¯0) ∪ Y(l),(d,¯1)), l=1(Y(l),(d,¯0) ∪Y(l),(d,¯1)) (d = 1, . . . , s+µ, θ ∈ Z/2Z). consider also a new set of graded variables ys ∈ XZ2 such that degZ2 ys = degZ2 rs, 1 ≤ s ≤ s. Let us Let us take for any l = 1, . . . , p the polynomial fl(eX(l), Y(l), Z(l)) defined by (9) in Lemma 17 assuming s = s+ µ = nd(A)− 1+ µ. Here Y(l) =Ss+µ and Yd,θ =Sp denote eYd,θ = Yd,θS{yd} if degZ2 yd = θ, and eYd,θ = Yd,θ otherwise d = 1, . . . , µ + s). Then we obtain ¯τd = (τd¯0, τd¯1) = (eYd,¯0, eYd,¯1) > (Yd,¯0, Yd,¯1) if d = 1, . . . , s. And ¯τd = (eYd,¯0, eYd,¯1) = (Yd,¯0, Yd,¯1) for any d = s+1, . . . , s+µ. Here (Yd,¯0, Yd,¯1) = Denote by eYd = eYd,¯0 ∪eYd,¯1 (d = 1, . . . , s + µ); eX =(cid:0)Sp l=1 eXl(cid:1)S{x1, . . . , xn−s}; eZ =Sp h′(eY1, . . . ,eYs+µ, eX,eZ) = h(cid:16)f1(cid:0)eX(1), Y(1), Z(1)(cid:1)y1, y2f2(cid:0)eX(2), Y(2), Z(2)(cid:1), . . . , fp−1(cid:0)eX(p−1), Y(p−1), Z(p−1)(cid:1)yqfp(cid:0)eX(p), Y(p), Z(p)(cid:1), yq+1, . . . , ys, x1, . . . , xn−s(cid:17); hµ(eY1, . . . ,eYs+µ, eX,eZ) =(cid:16)s+µYd=1 )(cid:17)h′(eY1, . . . ,eYs+µ, eX,eZ). The polynomial hµ is linear in variables eY =Ss+µ d=1 eYd, and ¯τd-alternating in any eYd (d = 1, . . . , s + µ). Although hµ is not multilinear and is not multihomogeneous in general. l=1 Zl. Consider the polynomials (A eYd,¯0 A eYd,¯1 Consider the following evaluation of the polynomial hµ in the superalgebra A y(l),(d,(iljl)) = εlEliljlεl, y(l),(d,¯1,(iljl)) = εl(Eliljlc)εl, xl,(il) = εlElil1εl, ys = as = εl′ srsεl′′ s ; l = 1, . . . , p; 1 ≤ il, jl ≤ sl, y(l),(d,¯0,(iljl)) = εlEliljlεl, z(l),(iljl) = εlEliljlεl, xl,(jl,¯cl) = εl(El1jl ¯cl)εl, xn′ = an′+s = bn′, (11) d = 1, . . . , s + µ; 1 ≤ s ≤ s; 1 ≤ n′ ≤ n − s. The evaluations of the variables y, z, x, and x are defined as in Lemma 17. It is clear that (11) is an elementary evaluation and satisfies the third claim of the lemma. Due to the evaluation of the variables z, x, and x, the polynomial fl can contain only elements of the simple component Cl or elements εlrsεl, otherwise, we get zero. The second case will give us a thin evaluation of the polynomial h, thus, such summands are also zero. Therefore, the evaluation (11) of the polynomial hµ gives the same result as this evaluation of the polynomial h(f ′ 1 y1, y2f ′ 2, . . . , f ′ p−1 yqf ′ p, yq+1, . . . , ys, x1, . . . , xn−s), where f ′ see that the result of our evaluation of f ′ )fl. Similarly to arguments of Lemmas 17, 13, we can l is equal to εl. Thus (11) for hµ gives the d=1 A eYd,¯0 A eYd,¯1 l = (Qs+µ 17 , εl′ , b1, . . . , bn−s) = result h(ε1r1εl′′ h(a1, . . . , an). Therefore hµ is the desired polynomial. The evaluation (11) can be taken as ¯u. . . . , εp−1rqεp, εl′ , . . . , εl′ rq+1εl′′ rsεl′′ r2ε2, q+1 q+1 s s 1 2 Consider the case nd(A) ≥ 1, and p = 1. If ε1h(a1, . . . , an) = 0 then among s radical elements of ¯a there is a homogeneous in the grading radical element of the type ε2rε1 ∈ ε2J(A)ε1, where ε2 is the adjoint idempotent of A (since the substitution ¯a is complete). We can suppose without loss of generality that a1 = ε2r1ε1, and the first s = (nd(A) − 1) elements of ¯a are radical a2 = εl′ , . . . , as = ∈ J(A), an′+s = bn′ ∈ D, for n′ = 1, . . . , n − s. Then using the same εl′ s ∈ {1, 2}, arguments as in the previous case p ≥ 2, assuming that p = 1, q = 1, we can prove that the evaluation (11) of the polynomial s, l′′ l′ r2εl′′ rsεl′′ s s 2 2 hµ =(cid:16)s+µYd=1 A eYd,¯0 A eYd,¯1(cid:17) h(cid:16)y1 · f1(cid:0)eX(1), Y(1), Z(1)(cid:1), y2, . . . , ys, x1, . . . , xn−s(cid:17) is equal to h(a1, . . . , an). If ε1h(a1, . . . , an) 6= 0 then the similar arguments as in the case p ≥ 2 show that the result of the evaluation (11) of the polynomial hµ =(cid:16)s+µYd=1 (A eYd,¯0 A eYd,¯1 )(cid:17) f1(cid:0)eX(1), Y(1), Z(1)(cid:1) · h(cid:16)y1, y2, . . . , ys, x1, . . . , xn−s(cid:17) is equal to ε1h(a1, . . . , an) = h(a1, . . . , an). Here variables ys can not take places inside f1 with a nonzero result, since it will give a thin evaluation of h. In both of the last cases the polynomial hµ and the corresponding evaluation possess all desired properties. The case p = 0 is trivial. In this case the superalgebra A is nilpotent, and dimsZ2A = (0, 0). Since a multihomogeneous graded polynomial h is exact for A and is not a graded identity of A (h(¯a) 6= 0) then the full degree of h is (nd(A) − 1), and h has the type (0, 0; nd(A) − 1; µ) = (dimsGA; nd(A) − 1; µ) for any µ ∈ N0. Thus in this case hµ = h, ¯u = ¯a. ✷ Lemma 20 has the following important corollaries. Lemma 21 Let A be a Z2PI-reduced algebra then indZ2(A) = parZ2(A). If f is an Proof. By Lemma 19 A has an exact polynomial f /∈ IdZ2(A). f can be nonzero only for a complete evaluation containing exactly (nd(A) − 1) radical elements. exact polynomial for A, and f /∈ IdZ2(A) then Z2T [f ]T Sµ(A) 6= ∅ for any µ ∈ N0. Lemma 20 implies that f has a nontrivial consequence eg /∈ IdZ2(A) of the type (dimsZ2A; nd(A) − 1; µ) for any µ ∈ N0. By Lemma 17 we have β(A) = dimsZ2A. Then Definition 7 implies γ(A) > nd(A) − 1. Taking into account that indZ2(A) ≤ parZ2(A) we obtain γ(A) = nd(A). Moreover by Lemma 20 any exact for A polynomial f such that f /∈ IdZ2(A) has a nontrivial consequence gµ ∈ Z2T [f ] for any µ ∈ N0. Where gµ is µ-boundary polynomial for A. ✷ Lemma 21 along with Lemma 18 immediately implies 18 Lemma 22 Any multilinear µ-boundary polynomial for a Z2PI-reduced algebra A is exact for A. Lemma 23 Given a Z2PI-reduced algebra A, and an integer µ ∈ N0 let SA,µ be any set of graded polynomials of the type (β(A); γ(A) − 1; µ). Then if a multilinear polynomial f ∈ Z2T [SA,µ] + IdZ2(A) then f is exact for A. Proof. A multilinear graded polynomial f ∈ Z2T [SA,µ] + IdZ2(A) has the form f (x1, . . . , xn) = X1≤i≤d1 1≤j≤d2 αij vi · gj(hi1, . . . , hinj ) · wi + g(x1, . . . , xn), where hi1, . . . , hinj , vi, wi ∈ F hXZ2 i are graded monomials, vi, wi are possibly empty; gj(zj1, . . . , zjnj ) ∈ F hXZ2i are full linearizations of some polynomials from the set SA,µ; g is a multilinear graded identity of A; αij ∈ F. It is clear that the multilinear graded polynomials gj(zj1, . . . , zjnj ) have the type (β(A); γ(A) − 1; µ) (∀j = 1, . . . , d2), and the graded monomials vi · hi1 · · · hinj · wi are multilinear and depends on the same variables x1, . . . , xn as f for any i. Particularly, the ) are multilinear, here we have monomials hi1(xδ1, . . . , xδsi1 {δ1, . . . , δsi1 , . . . , λ1, . . . , λsinj ), . . . , hinj (xλ1, . . . , xλsinj } ⊆ {1, . . . , n}. Fix any i = 1, . . . d1, j = 1, . . . , d2. Given an elementary evaluation (a1, . . . , an) of f in A consider homogeneous elements of A, that are the evaluation of the monomials hil semisimple element of A or radical, and ), . . . ,eainj = hinj (aλ1, . . . , aλsinj ). Then eail is either a degZ2 utl = degZ2eail = θl, ∀tl; utl ∈ U (3), ∀tl, Rl = J(A)θl . eai1 = hi1(aδ1, . . . , aδsi1 eail = dim RlXtl=1 α(il),tl utl, utl ∈ D (2), ∀tl, Rl = Bθl, α(il),tl ∈ F, or Then the evaluation (a1, . . . , an) of gj(hi1, . . . , hinj ) yields gj(eai1, . . . ,eainj ) = gj( Xt1,...,tnj α(i1),t1 · · · α(inj ),tnj dim R1Xt1=1 α(i1),t1 ut1, . . . , dim RnjXtnj =1 α(inj ),tnj utnj ) = gj(ut1 , . . . , utnj ). All evaluations (ut1, . . . , utnj of radical elements in (ut1 , . . . , utnj ) of the polynomials gj are elementary. The numbers ) is equal to the number of radical elements in (eai1, . . . ,eainj ) for all t1, . . . , tnj , and does not exceed the number of radical elements in the initial evaluation (a1, . . . , an). Thus if (a1, . . . , an) is a thin evaluation then (ut1 , . . . , utnj ) is also thin for all t1, . . . , tnj and i, j. 19 l6=k l′6=k, l′′6=k Cl) ⊕ ( P1≤l′,l′′≤p+1 εl′J(A)εl′′ ) is a graded subalgebra of A for any k. There- ( Q1≤l≤p fore, if (a1, . . . , an) is incomplete then (eai1, . . . ,eainj ), and (ut1, . . . , utnj incomplete. The last evaluations do not contain elements of CkS{εkrεl, εlrεkr ∈ U, 1 ≤ l ≤ p + 1} if elements of this set does not appear in (a1, . . . , an). A is Z2PI-reduced, and by Lemma 21 indZ2(A) = parZ2(A). Thus gj is a multi- linear graded polynomial of the type (β(A); γ(A) − 1; µ) = (dimsZ2A; nd(A) − 1; µ), and by Lemma 18 gj is exact for A (for any j). ) are also Thus for any thin or incomplete evaluation (a1, . . . , an) we obtain gj(eai1, . . . ,eainj ) = Xt1,...,tnj α(il),t1 · · · α(il),tnj gj(ut1 , . . . , utnj ) = 0. Hence f (a1, . . . , an) = 0, and f is exact for A. ✷ Lemma 24 Let Γ be a proper Z2T-ideal, and A be a Z2PI-reduced algebra such that indZ2 (Γ) = indZ2(A). Suppose that a graded polynomial f satisfies the conditions f /∈ IdZ2(A), and f ∈ Kµ(Γ) + IdZ2(A) for some µ ∈ N0. Then Z2T [f ]T Sµ(A) 6= ∅ holds for any µ ∈ N0. Proof. The full linearization f of some multihomogeneous component of f also satisfies f ∈ Kµ(Γ) + IdZ2(A), f /∈ IdZ2(A). Then by Lemma 23 f is exact for A. And by Lemma 21 we obtain that ∅ 6= (Z2T [ f ]T Sµ(A)) ⊆ (Z2T [f ]T Sµ(A)) for any µ ∈ N0. ✷ 6 Representable graded algebras. Let R be a commutative associative unitary F -algebra. Suppose that a Z/2Z-graded F -algebra A = A¯0 ⊕ A¯1 has a structure of R-algebra satisfying RAθ ⊆ Aθ, ∀θ ∈ Z/2Z. Definition 15 Any R-linear mapping tr : A¯0 → R is called trace on the Z/2Z- graded R-algebra A. Observe that a trace on A is not necessary symmetric (not necessary satisfies tr(ab) = tr(ba)). Denote by S the free associative commutative unitary algebra generated by all symbols tr(u) for nonempty associative noncommutative even monomials u ∈ (F hXZ2 i)¯0 over XZ2. We say that F ShXZ2 i = F hXZ2 i ⊗F S is free Z/2Z-graded algebra with trace. We assume that (f ⊗ s1)s2 = s2(f ⊗ s1) = f ⊗ (s1s2) for all f ∈ F hXi, s1, s2 ∈ S. The Z/2Z-grading on F ShXZ2 i is induced from F hXZ2 i (S is supposed to be trivially graded). Elements of F ShXZ2 i are called graded polynomials with trace. Elements of S are called pure trace polynomials. 20 We identify F hXZ2 i ⊗F 1 with F hXZ2 i. The symbol ⊗ usually is omitted in the notation of graded polynomials with trace. The concept of degrees (homogene- ity in the sense of degree, multilinearity, alternating, etc.) of graded polynomials with trace or pure trace polynomials is defined in similar way to ordinary graded polynomials assuming degx tr(u) = degx u for any x ∈ XZ2 . The S-linear function of trace tr : (F ShXZ2 i)e → S is defined on the S-algebra F ShXZ2 i by the formula tr(Pi uisi) = Pi tr(ui)si, where ui ∈ (F hXZ2 i)e are monomials on XZ2 of the even graded degree, si ∈ S. Let A be a Z/2Z-graded R-algebra with trace, f (x1, . . . , xn) ∈ F ShXZ2 i be a graded polynomial with trace. A satisfies the graded identity with trace f = 0 if f (a1, . . . , an) = 0 holds in A for any a1, . . . , an ∈ A. The ideal of graded identities with trace of A SIdZ2(A) = {f ∈ F ShXZ2 i A satisfies f = 0} is Z/2Z-graded S- ideal of F ShXZ2 i closed under all graded endomorphisms of the algebra F ShXZ2 i, and preserving the trace. SIdZ2(A) also satisfies the condition g · tr(f ) ∈ SIdZ2(A), for any g ∈ F ShXZ2 i, and for any f ∈ (SIdZ2(A))¯0. Ideals of F ShXZ2 i with these generated by a set V ⊆ F ShXZ2 i. properties are called Z2TS-ideals. Given a Z2TS-idealeΓ, and polynomials with trace f, g ∈ F ShXZ2 i we write f = g (mod eΓ) if f − g ∈ eΓ. Z2T S[V] is the Z2TS-ideal LeteΓ ✂F ShXZ2 i be a Z2TS-ideal of F ShXZ2 i. Denote by I = SpanF {tr(f )vf ∈ eΓ¯0, v ∈ S} ✂ S the ideal of S generated by all elements of the form tr(f ) for all polynomials with trace f ∈eΓ of the even graded degree. Let ¯S = S/I be the quotient algebra. Then the quotient algebra F ShXZ2 i = F ShXZ2 i/eΓ is Z/2Z-graded and has the well-defined structure of ¯S-algebra. ¯S-linear function tr : (F ShXZ2 i/eΓ)¯0 → ¯S is naturally defined by the equalities tr(a +eΓ) = tr(a) + I for any a ∈ (F ShXZ2 i)¯0. The ideal of graded identities with trace of F ShXZ2i coincides with eΓ. Moreover F ShXZ2 i/eΓ is the relatively free Z/2Z-graded algebra with trace for the Z2TS-ideal eΓ. ν i/(eΓT F ShXZ2 We also can consider the free associative Z/2Z-graded algebra with trace F ShXZ2 ν i ν i) and the relatively free Z/2Z-graded algebras with trace F ShXZ2 of a finite rank ν ∈ N. Let us take a finite dimensional F -superalgebra A = B ⊕ J with the Jacobson radical J = J(A), and the semisimple part B. Then the even component of the semisimple part B¯0 = B ∩ A¯0 is a finite dimensional subalgebra of A with dimF B¯0 = t¯0, where dimsZ2A = (t¯0, t¯1). Since the left regular representation T : B¯0 → Mt1 (F ) of B¯0 is injective then B¯0 is isomorphic to a subalgebra of the matrix algebra Mt¯0(F ). Therefore the trace tr : A¯0 → F on A is well defined by the rule tr(a) = tr(b + r) = Tr(T(b)), a ∈ A¯0, b ∈ B¯0, r ∈ J¯0, (12) where Tr is the usual trace of a linear operator. Z/2Z-graded identity with trace tr(z)f =Pt¯0 Lemma 25 A finite dimensional F -superalgebra A with the trace (12) satisfies the i=1 f xi:=zxi. Where f (x1, . . . , xt1 , Y ) ∈ F hXZ2 i is any graded polynomial (without trace) of the type (dimsZ2A, nd(A)− 1, 1), alternating in the set {x1, . . . , xt¯0 } ⊆ X¯0. Here t¯0 = dim B¯0, z ∈ X¯0. 21 Proof. A polynomial f of the type (dimsZ2A, nd(A)− 1, 1) is (dimsZ2A)-alternating in at least one set of graded variables. Suppose that {x1, . . . , xt¯0} ⊆ X¯0 is the part of the even graded degree of this set. An elementary evaluation of variables of the set i=1 f xi:=zxi must be semisimple. Moreover the set {x1, . . . , xt¯0} must be exchanged by pairwise different semisimple elements of the basis D (2) of degree ¯0. Otherwise, we get zero. {z, x1, . . . , xt¯0 } of the polynomial g = tr(z)f −Pt¯0 Suppose that z = b ∈ B¯0 in our evaluation. Consider a polynomial f (x1, . . . , xt¯0, . . . ) that is alternating in a homogeneous set of variables of even degree {x1, . . . , xt¯0} ⊆ X¯0. It can be directly checked (see also [[5], Theorem J]) that for an arbitrary linear operator K : B¯0 → B¯0 and for all pairwise distinct basic elements b1, . . . , bt¯0 of B¯0 the equality T r(K) f (b1, . . . , bt¯0 , . . . ) = f (K(b1), . . . , bt¯0 , . . . ) + · · · + f (b1, . . . , K(bt¯0 ), . . . ) holds in A, the replacement of other variables being arbitrary. Applying this ob- servation to the linear operator T(b) of the left multiplication by the element b we complete the proof. ✷ Observe that in Lemma 25 it is enough to consider semisimple or radical evalu- ations of variables (not necessary elementary ones). Lemma 26 Let us take any β ∈ N2 0, γ ∈ N. Consider a graded polynomial without trace f (y1, . . . , yk) ∈ F hXZ2 i of the type (β; γ − 1; 1), and a pure trace polyno- mial s(z1, . . . , zd) ∈ S ({z1, . . . , zd} ⊆ XZ2.) Then there exists a graded polynomial without trace gs(y1, . . . , yk, z1, . . . , zd) ∈ F hXZ2i such that gs ∈ Z2T [f ], and any fi- nite dimensional superalgebra A with the parameters parZ2(A) = (β; γ) satisfies the graded identity with trace s(z1, . . . , zd) · f (y1, . . . , yk) − gs(y1, . . . , yk, z1, . . . , zd) = 0. Proof. If parZ2(A) = (β; γ) then by Lemma 25 A satisfies the Z/2Z-graded trace identity tr(z)f (w1x1, w2x2, . . . , wt¯0 xt¯0, Y ) − (f (zw1x1, w2x2, . . . , wt¯0 xt¯0, Y ) + · · · + f (w1x1, w2x2, . . . , zwt¯0 xt¯0, Y )) = 0. Where wi ∈ (F hXZ2 i#)¯0 are arbitrary (possibly empty) graded monomials (i = 1, . . . , t¯0), and f (x1, . . . , xt¯0 , Y ) ∈ F hXZ2 i is any graded polynomial of the type (β; γ−1; 1), alternating in {x1, . . . , xt¯0 } ⊆ X¯0, z ∈ X¯0. Let us take any graded monomials ui ∈ (F hXZ2 i)¯0 (i = 1, . . . , n), and any (pos- sibly empty) graded monomials vj ∈ (F hXZ2 i#)¯0 (j = 1, . . . , t¯0). Then by induction on the number n ∈ N0 of monomials ui there exist a naturalen, and graded monomials evlj ∈ (F hXZ2 i#)¯0 (possibly empty) such that tr(u1) · · · tr(un)f (v1x1, . . . , vt¯0 xt¯0, Y ) = Pen l=1 f (evl1x1, . . . ,evlt¯0xt¯0, Y ) (mod SIdZ2(A)). Moreover the monomials evlj depend Hence for any pure trace polynomial s(z1, . . . , zd) =P(j) α(j)tr(uj1) · · · tr(ujn) ∈ on the same variables as the monomial u1 · · · unv1 · · · vt¯0. S the algebra A satisfies the identity s(z1, . . . , zd) · f (x1, . . . , xt¯0, Y ) − gs(x1, . . . , xt¯0, Y, z1, . . . , zd) = 0. Where gs ∈ Z2T [f ] is some graded polynomial that does not depend on A. Here uj ∈ (F hz1, . . . , zdi)¯0 are monomials, α(j) ∈ F. ✷ 22 Let A be a finite dimensional F -superalgebra with the semisimple part B = B¯0 ⊕ B¯1, and the Jacobson radical J = J¯0 ⊕ J¯1. Let us denote dim Bθ = tθ, dim Jθ = qθ for any θ ∈ Z/2Z. Given a number ν ∈ N take a set Λν = {λθijθ ∈ Z/2Z, 1 ≤ i ≤ ν, 1 ≤ j ≤ tθ+qθ}. Consider the free commutative associative unitary algebra F [Λν]# generated by Λν, and the associative algebra Pν (A) = F [Λν ]# ⊗F A. The algebra Pν(A) has the The Z/2Z-grading on Pν(A) is induced from A assuming that F [Λν ]# is trivially graded. Define an F [Λν]#-linear map tr : (Pν (A))¯0 → F [Λν ]# by the equalities structure of F [Λν ]#-module defined by a · f = f · a = f · (Pi fi ⊗ ai) =Pi(f fi) ⊗ ai, for any f, fi ∈ F [Λν]#, ai ∈ A, a =Pi fi ⊗ ai ∈ Pν (A). t¯0Xi=1 fj ⊗ rj¯0) = fi Tr(T(bi¯0)). t¯0Xi=1 q¯0Xj=1 tr(a) = tr( fi ⊗ bi¯0 + (13) tr is well defined in Pν (A). Here bi¯0 ∈ B¯0, rje ∈ (J(A))¯0, fi, fj ∈ F [Λν]#. T is the left regular representation of B¯0, and Tr is the usual trace. Lemma 27 Pν(A) satisfies the graded identity with trace (Xt¯0(x) · x)nd(A) = 0. Here Xt¯0(x) is the Cayley-Hamilton polynomial of the degree t¯0, x ∈ X¯0. Proof. The algebra B¯0 with the trace (12) satisfies the identity with trace Xt¯0 (x) · x = 0 ([19], [20]). F [Λν]# is commutative non-nilpotent algebra. F [Λν]# ⊗F (J(A))¯0 is a nilpotent ideal of (Pν (A))¯0 of the degree nd(A). Hence the neutral component (Pν (A))¯0 = (F [Λν ]# ⊗F B¯0) ⊕ (F [Λν ]# ⊗F (J(A))¯0) satisfies the full linearization of the identity (Xt¯0(x) · x)nd(A) = 0. ✷ Let {bθ1, . . . , bθtθ } be a basis of the graded component Bθ (θ ∈ Z/2Z) of the semisimple part B of A, dim Bθ = tθ. Also let {rθ1, . . . , rθqθ } be a basis of the graded component Jθ of the Jacobson radical J = J(A), dim Jθ = qθ. Recall that for a superalgebra with elementary decomposition all these bases may be chosen in the set DS U ((2), (3), Lemma 1). Take elements yθi = λθij ⊗ bθj + λθij+tθ ⊗ rθj ∈ Pν (A), θ ∈ Z/2Z, 1 ≤ i ≤ ν. (14) tθXj=1 qθXj=1 The elements yθi are Z/2Z-homogeneous of degree θ (θ ∈ Z/2Z; 1 ≤ i ≤ ν). Given a positive integer ν consider the F -subalgebra Fν (A) = hyθiθ ∈ Z/2Z, 1 ≤ i ≤ νi of Pν (A) generated by the set Y Z2 ν = {yθiθ ∈ Z/2Z, 1 ≤ i ≤ ν}. Fν(A) is Z/2Z-graded. Any map ϕ of generators to arbitrary homogeneous elementseaθi ∈ Aθ (eαθij ∈ F ) ϕ : yθi 7→eaθi = also inducing the graded homomorphism eϕ : Pν(A) → A defined by the following can be extended to the graded homomorphism of F -superalgebras ϕ : Fν(A) → A, qθXj=1eαθij+tθ rθj ∈ Aθ tθXj=1eαθijbθj + i = 1, . . . , ν) (15) (θ ∈ Z/2Z, 23 equalities eϕ((λθ1i1j1 · · · λθkikjk) ⊗ a) = (eαθ1i1j1 · · ·eαθkikjk ) · a A. The homomorphism eϕ preserves the trace defined by (13) on Pν(A) and by (12) on Elements of Fν (A) are called quasi-polynomials in the variables Y Z2 ν . Products of the algebra Fν (A) are called quasi-monomials. We of the generators yθi ∈ Y Z2 have also IdZ2(Fν (A)) ⊇ IdZ2(Pν (A)) = IdZ2(A) for any ν ∈ N. ν ∀a ∈ A. (16) cj ∈ N, α(i) ∈ F. . . . wck ik i1 i ∈ (Pν (A))¯0 i ) ∈ F [Λν]# (i = 1, . . . , d, Consider the polynomials sil = tr(w2l Fν (A) is a finitely generated PI-algebra. By Shirshov's height theorem [22] Fν (A) has a finite height and a finite Shirshov's basis. Shirshov's basis of an algebra always can be chosen in the set of monomials over the generators ([5], [22]). Thus we can suppose that a Shirshov's basis of Fν (A) consists of homogeneous in the grading elements of Fν(A). More precisely, there exist a positive integer H, and homogeneous in the grading elements w1, . . . , wd ∈ Fν (A) such that any element u ∈ Fν (A) has , where k ≤ H, {i1, . . . , ik} ⊆ {1, . . . , d}, the form u =P(i)=(i1,...,ik) α(i) wc1 Then bF = F [sil 1 ≤ i ≤ d; 1 ≤ l ≤ t¯0 ]# is the associative commutative non-graded Take the Z/2Z-graded bF -subalgebra Tν(A) = bF Fν(A) of Pν(A). Then Fν(A) the restriction of eϕ defined by (16) onto Tν(A) ). i are algebraic of degree nd(A)(t¯0 + 1) over bF . Therefore, by Shirshov's height theorem, Tν(A) is finitely generated bF -module, where bF is is a graded subalgebra of Tν(A). An arbitrary map of type (15) can be uniquely extended to a graded homomorphism from Tν(A) to A preserving the traces (it is F -subalgebra of F [Λν ]# with the unit generated by {sil}, and by the unit of F [Λν ]#. Let V ⊆ F hXZ2i be a set of graded polynomials. We denote by V (Tν(A)) ✂ Tν(A) the verbal ideal generated by results of all appropriate substitutions of Z/2Z- homogeneous elements of Tν(A) to any graded polynomial from V. Noetherian. By theorem of Beidar [4] the algebra Tν(A) is representable. Since for any i = 1, . . . , d we have w2 then it follows from Lemma 27 that all elements w2 l = 1, . . . , t¯0). Lemma 28 Given a set V ⊆ F hXZ2i, and any ν ∈ N the verbal ideal V (Tν(A)) = SpanF { s · v1 f (u1, . . . , un)v2 s ∈ bF ; v1, v2 ∈ Fν(A)# are quasi-monomials, possibly empty; quasi-monomials; f is the full linearization of a multihomogeneous component of any f ∈ V } ui ∈ Fν(A) are is a graded bF -closed ideal of Tν(A). The quotient algebra T ν(A, V ) = Tν(A)/V (Tν(A)) is a representable bF -superalgebra. The ideal of graded identities of T ν(A, V ) satisfies ν i)(cid:1), where eΓ = IdZ2(A) + Z2T [V ]. IdZ2(T ν(A, V )) ⊇eΓ + IdZ2(cid:0)F hXZ2 Proof. It is clear that in case of characteristic zero any verbal ideal of an algebra can be generated by the results of all appropriate substitutions to the full linearizations ν i/(eΓ ∩ F hXZ2 24 have also s· f (c1, . . . , cn) = f (s·c1, . . . , cn) = f (c′ s1i1 · u1i1, . . . ,Pin graded F -subalgebra of Pν(A). Therefore we have IdZ2(T ν(A, V )) ⊇ IdZ2(Tν(A)) ⊇ ui ∈ Fν (A) are quasi-monomials, degZ/2Z ui = degZ/2Z c for all i. If f is a multi- snin · unin) = Any homogeneous element of Tν(A) has the form c =Pi si · ui, where si ∈ bF , linear graded polynomial then f (c1, . . . , cn) = f (Pi1 P(i1,...,in)(cid:0)Qn l=1 slil(cid:1) f (u1i1, . . . , unin) for any homogeneous elements c1, . . . , cn ∈ Tν(A). Here ulil ∈ Fν(A) are quasi-monomials of appropriate graded degrees, slil ∈ bF . We 1, . . . , cn) ∈ V (Tν(A)) for any s ∈ bF . Then the quotient algebra T ν(A, V ) is Z/2Z-graded is an bF -module. T ν(A, V ) is finitely generated over bF , as well as Tν(A), and is also representable. Tν(A) is a IdZ2(Pν(A)) = IdZ2(A). It is clear also that V ⊆ IdZ2(T ν(A, V )), hence eΓ = ν i)(cid:1), and Γ2 = Let us denote the Z2T-ideals Γ1 = IdZ2(cid:0)F hXZ2 ν i) ⊆eΓ ⊆ Γ2. Then for ui ∈ Fν(A) (degG ui = degG xi), and for any si ∈ bF we obtain f (s1u1, . . . , snun) = s1 · · · snf (u1, . . . , un) ∈ bF Γ1(Fν (A)) ⊆ bF Γ2(Fν (A)) ⊆ V (Tν(A)), IdZ2(T ν(A, V )). From the arguments above we have Γ1(F hXZ2 a multilinear graded polynomial f (x1, . . . , xn) ∈ Γ1, for any homogeneous elements ν i/(eΓ ∩ F hXZ2 IdZ2(A) + Z2T [V ] ⊆ IdZ2(T ν(A, V )). of multihomogeneous components of polynomials from the given set. Particularly, V (Tν (A)) is generated by Z/2Z-homogeneous elements, and hence, this is a graded ideal. since grk(Fν (A)) = ν. Therefore f ∈ Γ2, and Γ1 ⊆ Γ2. ✷ Definition 16 We say that a subset V ⊆ F hXZ2 i is multihomogeneous if for any f ∈ V V contains all multihomogeneous components of f. Lemma 29 Let A = A1 × · · · × Aρ be the direct product of arbitrary finite dimen- sional superalgebras A1, . . . , Aρ. Suppose that V ⊆ F hXZ2 i is a multihomogeneous set, and ν is any positive integer. Let us take any f (z1, . . . , zn) ∈ IdZ2(T ν(A, V )), and any homogeneous polynomials h1, . . . , hn ∈ F hXZ2 ν i with degZ2 hl = degZ2 zl. Then the equality f (h1, . . . , hn) = Pj sj · vj1 fj(uj1, . . . , ujn)vj2 (mod SIdZ2(Ai)) holds for any i = 1, . . . , ρ. Here fj are the full linearizations of some polynomials fj ∈ V ; ujl, vjl ∈ F hXZ2 ν i are monomials, vjl may be empty; and sj ∈ S are pure trace graded polynomials in the variables XZ2 ν . Proof. Given a graded polynomial f (z1, . . . , zn) ∈ IdZ2(T ν(A, V )), and arbitrary homogeneous polynomials h1, . . . , hn ∈ F hXZ2 ν i of degrees according to variables of f, we have f (h1, . . . , hn) ∈ V (Tν(A)). Here the quasi-polynomial hi = hi(y1, . . . , y2ν) is obtained by replacement of the variables xl ∈ XZ2 ν by the corresponding elements yl ∈ Y Z2 (14). Since V is a multihomogeneous set then by Lemma 28 in the algebra fj are the full linearizations of polynomials fj ∈ V, ujl = ujl(y1, . . . , ymν ) ∈ Fν(A) Tν(A) we obtain the equality f (h1, . . . , hn) = Pj sj · vj1 fj(uj1, . . . , ujn)vj2, where are quasi-monomials, sj = sj(y1, . . . , ymν) ∈ bF are pure trace quasi-polynomials, vjl = vjl(y1, . . . , ymν ) ∈ Fν (A) are also quasi-monomials, possibly empty. ν 25 An arbitrary graded map ϕ : Y Z2 ν → Ai from the generating set (14) of Fν (A) into any subalgebra Ai ⊆ A (i = 1, . . . , ρ) can be extended to the graded homomorphism eϕ : Tν(A) → Ai preserving the trace. Thus the equality f (h1(a1, . . . , a2ν ), . . . , hn(a1, . . . , a2ν )) −Xj fj(uj1(a1, . . . , a2ν ), . . . , ujn(a1, . . . , a2ν)) vj2(a1, . . . , a2ν ) = 0 sj(a1, . . . , a2ν ) · vj1(a1, . . . , a2ν ) × holds in Ai for any elements a1, . . . , a2ν ∈ Ai of appropriate graded degrees. There- fore f (h1(x1, . . . , x2ν ), . . . , hn(x1, . . . , x2ν )) −Xj fj(uj1(x1, . . . , x2ν ), . . . , ujn(x1, . . . , x2ν)) vj2(x1, . . . , x2ν ) = 0 sj(x1, . . . , x2ν ) · vj1(x1, . . . , x2ν ) × is a graded identity with trace of the algebra Ai, for any i = 1, . . . , ρ. ✷ Lemma 30 Suppose that A1, . . . , Aρ are any Z2PI-reduced algebras with indZ2(Ai) = i=1 Sµ(Ai) (for any µ ≥ 1), and a positive integer ν, there exists an F -finite dimensional superalgebra Cν such that κ for all i = 1, . . . , ρ. Given a subset V ⊆ Sρ IdZ2(Cν) = IdZ2(cid:16)F hXZ2 have Sρ ν i/(cid:0)(Tρ Proof. Let us take A = A1 × · · · × Aρ, and any ν ∈ N. By Lemma 11 we i=1 Sµ(Ai) = Sµ(A) for any µ, hence V ⊆ Sµ(A). Observe that a set of boundary polynomials is always multihomogeneous. Then for any f (z1, . . . , zn) ∈ IdZ2(T ν(A, V )), and any homogeneous polynomials h1, . . . , hn ∈ F hXZ2 ν i of appro- priate graded degrees by Lemma 29 we have ν i(cid:1)(cid:17). i=1 IdZ2(Ai) + Z2T [V ])T F hXZ2 f (h1, . . . , hn) =Xj sjvj1 fj(uj1, . . . , ujn)vj2 (mod SIdZ2(Ai)) ∀ i = 1, . . . , ρ. Where fj ∈ Z2T [V ] are multilinear graded polynomials; ujl, vjl ∈ F hXZ2 monomials, vjl may be empty; sj ∈ S, sj depends on XZ2 ν . ν i are Ai is a Z2PI-reduced superalgebra, hence from Lemmas 21, 9 it follows that parG(Ai) = indG(Ai) = indG(A) = κ = (β; γ) ( ∀i = 1, . . . , ρ). Any polynomial fj has the type (β; γ − 1; µ) (as the multilinearization of the polynomial fj ∈ V ⊆ Sµ(A)). Then by Lemma 26 there exists a graded traceless polynomial gj ∈ Z2T [ fj]∩ ν i such that sj fj(vj1, . . . , vjn) = gj (mod SIdZ2(Ai)) F hXZ2 for any i = 1, . . . , ρ. Therefore ν i ⊆ Z2T [V ] ∩ F hXZ2 f (h1, . . . , hn) =Xj vj1gjvj2 (mod SIdZ2(Ai)) ∀ i = 1, . . . , ρ, SIdZ2(Ai) ∩ F hXZ2 where g =Pj vj1gjvj2 ∈ Z2T [V ]T F hXZ2 (cid:0)IdZ2(A)+Z2T [V ](cid:1)T F hXZ2 ν i = IdZ2(Ai) ∩ F hXZ2 ν i, and f ∈ IdZ2(cid:16)F hXZ2 ν i. Hence we obtain that f (h1, . . . , hn)−g ∈ ν i for all i = 1, . . . , ρ. Thus f (h1, . . . , hn) ∈ ν i(cid:1)(cid:17). ν i/(cid:0)(IdZ2(A)+Z2T [V ])T F hXZ2 26 IdZ2(T ν(A, V )). The superalgebra T ν(A, V ) is representable. Hence by Lemma 3 there exists an F -finite dimensional superalgebra Cν such that By Lemma 28 we have also that IdZ2(cid:16)F hXZ2 IdZ2(Cν) = IdZ2(T ν(A, V )) = IdZ2(cid:16)F hXZ2 ν i(cid:1)(cid:17) ⊆ ν i/(cid:0)(IdZ2(A)+Z2T [V ])T F hXZ2 ν i(cid:1)(cid:17). ν i/(cid:0)(IdZ2(A) + Z2T [V ])\ F hXZ2 ✷ 7 Graded identities of finitely generated PI-algebras. Lemma 31 Let F be a field of characteristic 0. Let Γ be a non-trivial ideal of graded identities of a finitely generated associative PI-superalgebra over F. Then there exists Γ, a finite dimensional associative F -superalgebra eA satisfying the conditions IdZ2(eA) ⊆ indZ2(Γ) = indZ2(eA), and Sµ(O(eA)) ∩ Γ = ∅ for some µ ∈ N0. Proof. Γ contains a non-trivial T-ideal eΓ of ordinary non-graded identities of a finitely generated associative PI-algebra. By [17]eΓ also contains the T-ideal of some finite dimensional non-graded algebra C od. It is clear that A = C od ⊗F F [Z/2Z] is a finite dimensional superalgebra with a Z/2Z-grading induced by the natural grading of the group algebra F [Z/2Z], and IdZ2(A) = Z2T [Id(C od)]. Thus we have that Γ contains the ideal of graded identities IdZ2(A) of some finite dimensional Z/2Z-graded algebra A. We assume by Lemma 16 that A = O(A) × Y(A) with the senior components A1, . . . , Aρ. It is clear that κ = indZ2(Γ) ≤ κ1 = indZ2(A) (Lemma 6). If Γ ⊆ IdZ2(Ai) for some i = 1, . . . , ρ then κ1 = indZ2(Ai) = κ. Thus, the case Γ ⊆ IdZ2(O(A)) is trivial. Assume that Γ * IdZ2(Ai) for all i = 1, . . . , ρ1 (1 ≤ ρ1 ≤ ρ), and Γ ⊆ IdZ2(A′′), i=1Ai. Consider the set V = Sµ(A′) ∩ Γ for µ = i=ρ1+1Ai, A′ = ×ρ1 where A′′ = ×ρ D such that Γ = IdZ2(D). By Lemma 30 there exists a finite dimensional over F su- bµ(O(A)) (Definition 11). Take ν = grk(D) for a finitely generated PI-superalgebra peralgebra Cν such that IdZ2(Cν ) = IdZ2(cid:0)F hXZ2 ν i/(cid:0)(IdZ2(A′)+Z2T [V ])∩ F hXZ2 ν i(cid:1)(cid:1). Then eA = Cν × A′′ × Y(A) is a finite dimensional superalgebra. For any f (x1, . . . , xn) ∈ IdZ2(eA), and for any homogenous polynomials h1, . . . , hn ∈ F hXZ2 ν i of the appropriate graded degrees we have f (h1, . . . , hn) = h+g, where h ∈ IdZ2(A′), g ∈ Γ ∩ IdZ2(Y(A)). Therefore h = f (h1, . . . , hn) − g ∈ IdZ2(A′) ∩ IdZ2(A′′) ∩ IdZ2(Y(A)) = IdZ2(A) ⊆ Γ. Hence f (h1, . . . , hn) = h + g ∈ Γ for any h1, . . . , hn ∈ F hXZ2 Γ = ∅ then in the first case we use Lemmas 11, 16, 10 and conclude for any µ ≥ ν i, and by Remark 1 IdZ2(eA) ⊆ Γ. Particularly indZ2(eA) ≥ κ. 6= 0. Then κ1 = κ = indZ2(eA). Thus either indZ2(Cν ) = κ implying O(eA) = A′′×O(Cν), or indZ2(Cν) < κ implying O(eA) = A′′. Since Sµ(A′′)∩ max{bµ(Cν), µ} we have Sµ(O(eA)) ∩ Γ = Sµ(Cν) ∩ Γ ⊆ Sµ(A′) ∩ Γ ⊆ V ⊆ IdZ2(Cν ). Thus, in both cases Sµ(O(eA)) ∩ Γ = ∅, and eA satisfies the claims of the lemma. If A′′ = 0 then κ ≤ indZ2(eA) = max{indZ2(Cν ), indZ2(Y(A))} ≤ κ1. If indZ2(Cν) = κ1 = indZ2(A′) then O(eA) = O(Cν ), and by analogy with previous case we apply Suppose that A′′ 27 Lemmas 16, 10 for any µ ≥ max{bµ(Cν), µ} and obtain Sµ(O(eA)) ∩ Γ = ∅. It also follows from Lemma 10 that indZ2(eA) = indZ2(Γ), and eA is also the desired algebra. The last case A′′ = 0, indZ2(Cν ) < κ1 gives eA = Cν × Y(A) with IdZ2(eA) ⊆ Γ, and κ ≤ indZ2(eA) < κ1 = indZ2(A). Then by the inductive step on indZ2(A) we can assume that the assertion of the lemma holds in this case. Lemma 31 with Lemma 10 implies the corollary. ✷ Lemma 32 Let F be a field of characteristic 0. Let Γ be a non-trivial ideal of graded identities of a finitely generated associative PI-superalgebra over F. Then Γ contains the ideal of graded identities of a finite dimensional associative F -superalgebra A satisfying indZ2(Γ) = indZ2(A), and Sµ(A) = Sµ(Γ) for some µ ∈ N0. Proof. Let us take a superalgebra A satisfying the claims of Lemma 31. Lemma 10 implies that Sµ(A) ⊇ Sµ(Γ) for any µ ∈ N0. On the other hand indZ2(Γ) = indZ2 (A) = (β; γ). Consider an integer µ = max{bµ(A), µ}, where µ is defined by Lemma 31, and bµ(A) is taken from Definition 11. Then any polynomial f ∈ Sµ(A) has the type (β; γ −1; µ), and satisfies f /∈ Γ (since f ∈ Sµ(O(A))). Thus, f ∈ Sµ(Γ). ✷ Theorem 1 Let F be a field of characteristic zero. For any finitely generated as- sociative Z/2Z-graded PI-algebra D over F there exists a finite dimensional over F associative superalgebra eC such that the Z2T-ideals of Z/2Z-graded identities of D and eC coincide. Proof. Let Γ be the ideal of graded identities of D. We use the induction on the Kemer index indZ2(Γ) = κ = (β; γ) of Γ. Inductive basis. If indZ2(Γ) = (0, 0; γ) then D is a nilpotent finitely generated F -superalgebra. Hence D is finite dimensional. Inductive hypothesis. Lemma 31, and Lemma 32 imply Γ ⊇ IdZ2(A), where A = O(A)+Y(A) is a finite dimensional superalgebra with indZ2(Γ) = indZ2(A) = κ. Moreover, Sµ(Γ) = Sµ(O(A)) = Sµ(A) ⊆ IdZ2(Y(A)) for some µ ∈ N0. Let A1, . . . , Aρ be the senior components of the algebra A. Denote (t¯0, t¯1) = t = t¯0 + t¯1; γ = γ(Γ) = nd(Ai) (for all i = 1, . . . , ρ). Let us β(Γ) = dimsZ2Ai, qi )) + (XZ2 1)(γ + µ), where I = Γi(Bi(XZ2 take for any i = 1, . . . , ρ the algebra eAi = Rqi,s(Ai) defined by (4) for the senior component Ai with qi = dimF Ai, s = (t + 1)(γ + µ). eAi is a finite dimensional superalgebra. Γi = IdZ2(eAi) = IdZ2(Ai), and dimsZ2eAi = dimsZ2Ai = β. The qi )/I of eAi is nilpotent of class at most s = (t + Jacobson radical J(eAi) = (XZ2 as the semisimple part of Ai and of eAi simultaneously (Lemma 2, Lemma 21). Particularly, eAi are superalgebras with elementary decomposition. By Lemma 28, IdZ2(T ν(eA, Γ)), where eA = ×ρ ν i(cid:1). Lemmas 6, 12 imply Let us denote eDν = F hXZ2 that indZ2(eDν ) ≤ indZ2(Γ + Kµ(Γ)) < indZ2(Γ). By the inductive step we obtain i=1eAi, ν = grk(D). ν i/(cid:0)(Γ + Keµ(Γ))T F hXZ2 and Lemma 3 there exists a finite dimensional F -superalgebra C such that IdZ2(C) = qi )s. Here the algebra Bi can be considered 28 IdZ2(eDν) = IdZ2(eU ) for a finite dimensional over F superalgebra eU . Lemma 28 yields Γ ⊆ IdZ2(C ×eU ). Consider a multilinear polynomial f (x1, . . . , xd) ∈ IdZ2(C × eU ). Let us take SΓ + SIdZ2(eAi) for any i = 1, . . . , ρ. Hence h(x1, . . . , xn) ∈ SΓ + SIdZ2(eAi) also holds any multihomogeneous polynomials h1, . . . , hd ∈ F hXZ2 ν i with degZ2 hi = degZ2 xi (i = 1, . . . , d). We have f (h1, . . . , hd) = g + h for some multihomogeneous graded polynomials g ∈ Γ, h ∈ Kµ(Γ). Then by Lemma 29 we obtain h = f (h1, . . . , hd)−g ∈ for the multilinearization h of h. Lemma 23 implies that h is exact for Ai (h ∈ Kµ(Γ)) for any i = 1, . . . , ρ. Let us fix any i = 1, . . . , ρ. Assume that {c1, . . . , cqi} is a basis of Ai of homo- geneous in the grading elements chosen in (2), (3) (Lemma 1), and fix the order of these basic elements. Suppose that ¯a = (r1, . . . , rs, bs+1, . . . , bn) is any elemen- tary complete evaluation of h by elements of the algebra Ai, where rj ∈ J(Ai), A eYd,¯0 A eYd,¯0 A eYd,¯1(cid:1)hµ =Xj (cid:0)s+µYd=1 bj ∈ Bi, s = γ − 1. By Lemma 20 there exist a polynomial hµ(eY1, . . . ,eYs+µ, eX,eZ) ∈ SΓ + SIdZ2(eAi) of the type (β, γ − 1, µ), and an elementary evaluation ¯u in Ai such that hµ(¯u) = h(¯a). Moreover, hµ is alternating in any eYj (j = 1, . . . , s + µ), and all variables from eXSeZ are exchanged by semisimple elements. Then we have α1hµ =(cid:0)s+µYd=1 A eYd,¯1(cid:1)(cid:0)sj gj(cid:1)(mod SIdZ2(eAi)), where eYd = eYd,¯0 ∪eYd,¯1; α1 ∈ F, α1 6= 0; gj ∈ Γ, sj ∈ S. Denote {ζ1, . . . , ζn} the variables eY S eXSeZ of hµ (the first variables are from eY =Ss+µ d=1 eYd). eAi. Here xπ(l)θl ∈ XZ2 (1 ≤ l ≤ n). Consider the following evaluation of hµ(ζ1, . . . , ζn) in the algebra eAi Given an element ul ∈ Ai of the mentioned above evaluation ¯u of the polynomial hµ defined by Lemma 20 we take the element ¯xπ(l)θl = xπ(l)θl + I of the algebra is a graded variable of graded degree θl = degZ2 ul, and π(l) is the ordinal number of the element ul = cπ(l) in our basis of Ai, 1 ≤ π(l) ≤ qi (17) if qi ζl = ul ζl ∈ eY , ul = cπ(l) ∈ εk1Aiεk2; Then we get sj(18) = 0, because the trace of a radical element is zero (12). ζl = εk1 ¯xπ(l)θlεk2 ∈ J(eAi) if ζl ∈ eX[eZ. Suppose that in (17) the pure trace polynomial sj depends essentially on eY . A eYd,¯1(cid:1)(cid:0)sj gj(cid:1) = sj gj, where gj = If sj does not depend on eY then (cid:0)Qs+µ A eYd,¯1(cid:1)gj(cid:1) ∈ Γ. If gj(18) 6= 0 in eAi then one of degree multihomo- (cid:0)(cid:0)Qs+µ geneous components of gj is a µ-boundary polynomial for eAi. And it is not a µ- boundary polynomial for Γ, because it belongs to Γ. This implies that Sµ(A) 6= Sµ(Γ), which contradicts to the properties of A. Therefore, we have gj(18) = 0. d=1 A eYd,¯0 d=1 A eYd,¯0 (18) 29 Thus, in any case hµ(18) = 0 holds in the algebra eAi. Hence the evaluation ζl ∈ eY , ul = cπ(l) ∈ εk1Aiεk2; ζl = vl = εk1xπ(l)θlεk2 ζl = vl = ul if if ζl ∈ eX[eZ qi ) is equal to hµ(v1, . . . , vn) ∈ I = of the polynomial hµ in the algebra Bi(XZ2 Γi(Bi(XZ2 Γi(Bi(XZ2 qi )) + (XZ2 qi )s. Since eY < s, the polynomial hµ is linear in eY , and the vari- ables from eXSeZ are replaced by semisimple elements, then we obtain hµ(v1, . . . , vn) ∈ Consider the map ϕ : xπ(l)θl 7→ cπ(l) (l = 1, . . . , eY ), and ϕ(b) = b for any b ∈ Bi. It is clear that ϕ can be extended to a graded homomorphism ϕ : Bi(XZ2 Then ϕ(hµ(v1, . . . , vn)) = hµ(ϕ(v1), . . . , ϕ(vn)) = hµ(¯u) = h(¯a) ∈ ϕ(Γi(Bi(XZ2 Γi(Ai) = (0). qi ) → Ai. qi ))) = qi )). Therefore h(¯a) = 0 holds in Ai for any elementary complete evaluation ¯a ∈ An containing γ − 1 radical elements. Since h is a multihomogeneous exact polynomial for Ai, and γ = nd(Ai) then h ∈ IdZ2(Ai). Hence h ∈ ∩ρ i=1IdZ2(Ai) = IdZ2(O(A)), and h ∈ IdZ2(O(A) × Y(A)) = IdZ2(A) ⊆ Γ. Then we have f (h1, . . . , hd) = g + h ∈ Γ for any multihomogeneous polynomials h1, . . . , hd ∈ F hXZ2 ν i of appropriate graded degrees. The application of Remark 1 now implies that IdZ2(C ×eU ) ⊆ Γ. Therefore, Γ = IdZ2(C ×eU ). Theorem is proved. ✷ Observe that to be a PI-algebra is an essential condition for a finitely generated superalgebra D in Theorem 1, since a finite dimensional superalgebra is always a PI-algebra. 8 Specht problem. Let E = hei, i ∈ N eiej = −ejei, ∀i, ji be the Grassmann algebra of infinite rank with the canonical Z/2Z-grading E = E¯0 ⊕ E¯1 (E¯0 and E¯1 are the subspaces of E generated by all monomials in the generators of even and odd lengths respectively). Consider for a superalgebra A = A¯0 ⊕ A¯1 the Grassmann envelope E(A) = A¯0 ⊗ E¯0 ⊕A¯1 ⊗E¯1. Observe that the algebra E(A) has the natural Z/2Z-grading E(A)θ = Aθ ⊗ Eθ, θ ∈ Z/2Z. The next remark is obvious. Remark 2 If two PI-superalgebras A and B has the same Z/2Z-graded identities then they have also the same ordinary non-graded polynomial identities. The T-ideal Id(A) is the biggest T-ideal that is contained in IdZ2(A). Lemma 33 (Remark 3.7.6, [9]) If two PI-superalgebras A and B have the same Z/2Z-graded identities then the algebras E(A), and E(B) have the same ordinary non-graded polynomial identities. Proof. This is a simple consequence of the fact that the Z2T-ideals of superalgebras A and E(A) related by some well understood invertible transformation (see, e.g., 30 [12], [16], [9]). This transformation is completely algorithmic. Particularly, if two superalgebras are Z2PI-equivalent then their Grassmann envelopes are also Z2PI- equivalent. By Remark 2 the Grassmann envelopes of these superalgebras also have the same non-graded identities. ✷ Theorem 2 The T-ideal of polynomial identities of an associative PI-algebra over a field of characteristic zero coincides with the T-ideal of identities of the Grassmann envelope of some finitely generated PI-superalgebra. Proof. See Theorem 4.8.2 [9]. ✷ Theorem 1 along with Theorem 2, and Lemma 33 implies the principle Kemer's classification theorem. Theorem 3 The T-ideal of polynomial identities of an associative PI-algebra over a field of characteristic zero coincides with the T-ideal of identities of the Grassmann envelope of some finite dimensional superalgebra. Theorem 3 immediately implies the positive solution of the Specht problem. Theorem 4 (Theorem 2.4, [12]) The T-ideal of polynomial identities of an asso- ciative PI-algebra over a field of characteristic zero is finitely generated as a T-ideal. Proof. Suppose that Γ is a T-ideal that is not finitely based. Then there exists an infinite sequence of multilinear polynomials {fi(x1, . . . , xni)}i∈N ⊆ Γ satisfying the conditions deg fi < deg fj for any i < j and fi /∈ T [f1, . . . , fi−1] for any i ∈ N. Consider the T-ideals Γi generated by all consequences of the polynomial fi of degrees strictly greater then ni = deg fi (i = 1, 2, . . . ), and the T-idealeΓ =Pi∈N Γi. Then we have fi /∈ eΓ for any i ∈ N. By Theorem 3 we obtain eΓ = Id(E(C)) for a Lemma 1 implies E(C) = E(B) ⊕ E(J), where B is the semisimple part of C, J is the Jacobson radical of C. Consider a polynomial fk from our sequence, such that deg fk = nk > nd(C), and consider an evaluation in E(C) of fk of the type finite dimensional superalgebra C. xi = ai = cδi ⊗ gδi, where cδi ∈ BδiS Jδi, gδi ∈ Eδi, δi = ¯0, ¯1 (i = 1, . . . , nk). If for any i = 1, . . . , nk the elements cδi are radical then fk(a1, . . . , ank ) = 0 in E(C), for some l. Then ∈ Bδl since nk > nd(C). Suppose that xl = bδl for any element g0 ∈ E0 we obtain fk(. . . , al, . . . )g0 = fk(. . . , bδl ⊗ (gδl · g0), . . . ) = , where bδl ⊗ gδl fk(. . . , al · (1B ⊗ g0), . . . ) = 0, since fk(x1, . . . , xl · x0, . . . , xnk ) ∈eΓ, x0 ∈ X, 1B is the unit of B. It implies fk(a1, . . . , ank ) = 0. Therefore fk ∈eΓ, that contradicts to the choice of eΓ. ✷ References [1] Bakhturin, Yu., Giambruno, A., Riley, D. (1998). Group-graded algebras with polynomial identities. Israel J. Math. 104: 145-155. 31 [2] Bakhturin, Yu.A., Zaicev, M.V. (2004). Identities of graded algebras and codi- mension growth. Trans. Amer. Math. Soc. 356(10): 3939-3950. [3] Bakhturin, Yu.A., Sehgal, S.K., Zaicev, M.V. (2008). Finite-dimensional simple graded algebras. Sb. Math. 199(7): 965-983. [4] Beidar, K.I. (1986.) On theorems of A.I.Mal'tsev concerning matrix representa- tions of algebras. (Russian) Uspekhi Mat. Nauk 41(5): 161-162; English transl. in Russian Math. Surveys 41(1986). [5] Kanel-Belov, A., Rowen, L.H. (2005). Computational aspects of polynomial identities. A K Peters Ltd., Wellesley, MA. [6] Bergen, J., Cohen, M. (1986). Action of commutative Hopf algebras. Bull. Lond. Math. Soc. 18: 159-164. [7] Drensky, V. (2000). Free algebras and PI-algebras. Springer-Verlag Singapore, Singapore. [8] Drensky, V., Formanek, E. (2004). Polynomial identity rings. Birkhauser Verlag, Basel-Boston-Berlin. [9] Giambruno, A., Zaicev, M. (2005). Polynomial idintities and asymptotic meth- ods. Amer.Math.Soc., Math. Surveys and Monographs 122, Providence, R.I.. [10] Jacobson, N. (1975). PI-algebras: An introductions. Lecture Notes in Math. 441, Springer-Verlag, Berlin-New York. [11] Kemer, A.R. (1980). Capelli identities and the nilpotence of the radical of a finitely generated PI-algebra. Soviet Math. Dokl. 22. [12] Kemer, A.R. (1991). Ideals of identities of associative algebras. Amer.Math.Soc. Translations of Math. Monographs 87, Providence, R.I. [13] Kemer, A.R. (1987). Finite basability of identities of associative algebras. (Rus- sian) Algebra Logika 26(5): 597-641. [14] Kemer, A.R. (1988). Representability of reduced free algebras. Algebra Logic 27(3): 167-184. [15] Kemer, A.R. (1988). Solution of the problem as to whether associative algebras have a finite basis of identities. Soviet Math. Dokl. 37(1): 60-64. [16] Kemer, A.R. (1984) Varieties and Z2-graded algebras. (Russian) Izv. Akad. Nauk SSSR Ser. Mat. 48(5): 1042-1059. [17] Lewin, J. (1974). A matrix representation for associative algebras I, II, Trans. AMS 188: 293-308, 309-317. [18] Procesi, C. (1973). Rings with polynomial identities. Marcel Dekker, New York. 32 [19] Procesi, C. (1976). The invariant theory n × n matrices. Adv.math. 19(3): 306- 381. [20] Razmyslov, Ju.P. (1974). Identities with trace in full matrix algebras over a field of characteristic zero. (Russian) Izv. Akad. Nauk SSSR Ser. Mat. 38: 723-756. [21] Rowen, L.H. (1980). Polynomial identities in ring theory, Academic Press. [22] Shirshov, A.I. identities. (Russian) Mat.Sbornik 43(85): 277-283; English transl. in Amer. Math. Soc. Transl. Ser. 2,119(1983). (1957). On rings with polynomial [23] I.Sviridova, Identities of PI-algebras graded by a finite abelian group, Comm. Algebra, 39(9) (2011), 3462-3490. [24] I.Sviridova, Finitely generated algebras with involution and their identities, J. Algebra, 383(2013), 144-167. 33
1612.01138	1	1612	2016-12-04T16:32:20	Approximations and Mittag-Leffler conditions --- the tools	[ "math.RA" ]	Mittag-Leffler modules occur naturally in algebra, algebraic geometry, and model theory, [18], [12], [17]. If $R$ is a non-right perfect ring, then it is known that in contrast with the classes of all projective and flat modules, the class of all flat Mittag-Leffler modules is not deconstructible [14], and it does not provide for approximations when $R$ has cardinality $\leq \aleph_0$, [6]. We remove the cardinality restriction on $R$ in the latter result. We also prove an extension of the Countable Telescope Conjecture [21]: a cotorsion pair $(\mathcal A,\mathcal B)$ is of countable type whenever the class $\mathcal B$ is closed under direct limits. In order to prove these results, we develop new general tools combining relative Mittag-Leffler conditions with set-theoretic homological algebra. They make it possible to trace the facts above to their ultimate, countable, origins in the properties of Bass modules. These tools have already found a number of applications: e.g., they yield a positive answer to Enochs' problem on module approximations for classes of modules associated with tilting [4], and enable investigation of new classes of flat modules occurring in algebraic geometry [24]. Finally, the ideas from Section 3 have led to the solution of a long-standing problem due to Auslander on the existence of right almost split maps [20].	math.RA	math	APPROXIMATIONS AND MITTAG-LEFFLER CONDITIONS THE TOOLS JAN SAROCH Abstract. Mittag-Leffler modules occur naturally in algebra, algebraic ge- ometry, and model theory, [18], [12], [17]. If R is a non-right perfect ring, then it is known that in contrast with the classes of all projective and flat modules, the class of all flat Mittag-Leffler modules is not deconstructible [14], and it does not provide for approximations when R has cardinality ≤ ℵ0, [6]. We remove the cardinality restriction on R in the latter result. We also prove an extension of the Countable Telescope Conjecture [21]: a cotorsion pair (A, B) is of countable type whenever the class B is closed under direct limits. In order to prove these results, we develop new general tools combining relative Mittag-Leffler conditions with set-theoretic homological algebra. They make it possible to trace the facts above to their ultimate, countable, origins in the properties of Bass modules. These tools have already found a number of applications: e.g., they yield a positive answer to Enochs' problem on module approximations for classes of modules associated with tilting [4], and enable investigation of new classes of flat modules occurring in algebraic geometry [24]. Finally, the ideas from Section 3 have led to the solution of a long-standing problem due to Auslander on the existence of right almost split maps [20]. 1. Introduction The Mittag-Leffler condition for countable inverse systems was introduced by Grothendieck as a sufficient condition for the exactness of the inverse limit func- tor, [12]. Flat Mittag-Leffler modules then played an important role in the proof of the locality of the notion of a vector bundle in [18, Seconde partie]. The renewed interest in Mittag-Leffler conditions stems from their role in the study of roots of the Ext functor, see [1], [2], [5], and [21], as well as their connection to generalized vector bundles [8], [10]. In [14], it was shown that flat Mittag-Leffler modules coin- cide with the ℵ1-projective modules, studied already by Shelah et al. by means of set-theoretic homological algebra, [9]. A connection to model theory was discovered in [19]. The many facets of the notion of a Mittag-Leffler module, as well as its many applications, make it a key notion of contemporary algebra. Though locally projective, flat Mittag-Leffler modules have a very complex global structure in the case when R is not right perfect. Similarly to the classes P0 and F0 of all projective and flat modules, the class FM of all flat Mittag-Leffler modules is closed under transfinite extensions. However, unlike P0 and F0, it is not possible to obtain FM by transfinite extensions from any set of its elements. In other words, FM is not deconstructible, [14]. Moreover, unlike P0 and F0, the class FM does not provide for approximations when R is countable, [6]. Further results of this kind have been proved in [23]. All these results, however, have the drawback of Date: October 15, 2018. Key words and phrases. Mittag-Leffler conditions, approximations of modules, pure-injective module, Bass module, deconstructible class, countable type. Research supported by grant GA CR 14-15479S. 1 2 JAN SAROCH assuming that either the ring or the modules satisfy restrictive conditions on their size or structure. Our goal here is to remove these restrictions. We succeed in doing so in the case of Mittag-Leffler modules by proving that FM is not precovering for an arbitrary non-right perfect ring (Theorem 3.3). However, our key tool, Lemma 3.2, is much more general. It makes it possible to test for non-existence of approximations using certain small, countably presented modules, called the Bass modules. The terminology comes from the prototype example of a Bass module, namely the direct limit of the countable system f0→ R f1→ . . . fn−1→ R fn→ R fn+1→ . . . R where fn is the left multiplication by an in R, and Ra0 ⊇ Ra1a0 ⊇ . . . is a decreasing chain of principal left ideals in R. By a classic result of Bass, the direct limit is projective, if and only if the chain stabilizes. So non-right perfect rings are characterized by the existence of such Bass modules that are not projective. Recently, Lemma 3.2 has been applied to a number of diverse settings: e.g., to locally T -free modules in [4], and locally very flat modules in [24]. There are other general tools developed here that are of independent interest. We prove that every cotorsion pair (A, B) where B is closed under direct limits is a complete cotorsion pair of countable type with B actually being definable (Theorem 6.1). This is a non-hereditary version of the Countable Telescope Conjecture for module categories from [21, Theorem 3.5]. This paper is organized as follows. After some preliminaries in Section 2, we start out in Section 3 by discussing the case of flat Mittag-Leffler modules. Sections 4 and 5 are devoted to relative Mittag-Leffler conditions and their role in connections with vanishing of Ext and pure-injectivity. In Section 6 we consider cotorsion pairs (A, B) where B is closed under direct limits and prove that they are of countable type. 2. Preliminaries Let R be an (associative unital) ring. We denote by Mod-R the category of all (right R-) modules. 2.1. Bass modules. Given a class C of countably presented modules, we call a module M a Bass module over C, provided that M is the direct limit of a countable direct system C0 f0→ C1 f1→ . . . fn−1→ Cn fn→ Cn+1 fn+1→ . . . where Cn ∈ C for each n < ω. All Bass modules over C are countably presented; in fact, they are just the countable direct limits of the modules from C, cf. [23, §5]. 2.2. Pure-injective modules. For a (left, right) module M , we denote by M c = HomZ(M, Q/Z) the character (right, left) module of M . A module M is pure- injective provided that the canonical embedding of M into M cc splits. The pure- injective hull of a module M will be denoted by P E(M ). 2.3. Roots of Ext. For a class of modules C ⊆ Mod-R, we define its left and right Ext-orthogonal classes by ⊥C = Ker Ext1 R(C, −). For C = Mod-R, we have ⊥C = P0 and C ⊥ = I0, the classes of all projective and injective modules, respectively. R(−, C) and C ⊥ = Ker Ext1 APPROXIMATIONS, MITTAG-LEFFLER CONDITIONS -- TOOLS 3 2.4. Deconstructible classes. Given a module M and an ordinal number σ, we call an ascending chain M = (Mα α ≤ σ) of submodules of M a filtration of M , if M0 = 0, Mσ = M , and M is continuous -- that is, Sα<β Mα = Mβ for each limit ordinal β ≤ σ. Moreover, given a class of modules C, we call M a C-filtration of M , provided that each of the consecutive factors Mα+1/Mα (α < σ) is isomorphic to a module from C. A module M is C-filtered, if it admits a C-filtration. Given a class C and a cardinal κ, we use C ≤κ and C<κ to denote the subclass of C consisting of all ≤ κ-presented and < κ-presented modules, respectively. Let κ be an infinite cardinal. A class of modules C is κ-deconstructible pro- vided that each module M ∈ C is C<κ-filtered. For example, the class P0 is ℵ1- deconstructible by a classic theorem of Kaplansky. A class C is deconstructible in case it is κ-deconstructible for some infinite cardinal κ. 2.5. Cotorsion pairs. A pair of classes of modules C = (A, B) is a cotorsion pair provided that A = ⊥B and B = A⊥. The class Ker C = A ∩ B is the kernel of C. Notice that the class A is always closed under arbitrary direct sums and contains P0. Dually, the class B is closed under direct products and it contains I0. If moreover Ext2 R(A, B) = 0 for all A ∈ A and B ∈ B, then C is a hereditary cotorsion pair. The latter is equivalent to A being closed under kernels of epimorphisms. For example, for each n < ω, there exist hereditary cotorsion pairs of the form n ) and (⊥In, In), where Pn and In denote the classes of all modules of (Pn, P ⊥ projective and injective dimension ≤ n, respectively. The cotorsion pair C is said to be generated by a class S provided that B = S ⊥. In this case, if κ is an infinite regular cardinal such that each module in S is < κ- presented, then A is κ-deconstructible. If C is generated by a class S consisting of countably (finitely) generated modules with countably (finitely) presented first syzygies, then C is said to be of countable (finite) type 1. If R is right coherent, then it is equivalent to the statement 'C is generated by a class consisting of countably (finitely) presented modules'. Dually, C is said to be cogenerated by a class C provided that A = ⊥C. 2.6. Approximations. A class C of modules is precovering, if for each module M there exists a morphism f ∈ HomR(C, M ) with C ∈ C, such that each morphism f ′ ∈ HomR(C ′, M ) with C ′ ∈ C factors through f . Such f is called a C-precover of the module M . A precovering class of modules C is called special precovering provided that each module M has a C-precover f : C → M which is surjective and satisfies Ker(f ) ∈ C ⊥. Moreover, C is called covering provided that each module M has a C- precover f : C → M with the following minimality property: g is an automorphism of C, whenever g : C → C is an endomorphism of C with f g = f . Such f is called a C-cover of M . Dually, we define preenveloping, special preenveloping, and enveloping classes of modules. We will also deal with preenveloping and precovering classes in the category of all finitely presented modules -- these classes are usually called covariantly finite and contravariantly finite, respectively. A cotorsion pair C = (A, B) is called complete, provided that A is a special pre- covering class (or, equivalently, B a special preenveloping class). For example, each cotorsion pair generated by a set of modules is complete. Also, every deconstructible class of modules is precovering provided it is closed under transfinite extensions. 1In the literature, the requirement is usually extended from the first one to all syzygies. How- ever, this definition does not seem reasonable for non-hereditary cotorsion pairs, i.e. the ones we want to investigate. Our definition fits nicely into the crucial Theorem 6.1, while it still ensures that finite type of C implies definability of B. 4 JAN SAROCH Further, C is closed, provided that A = lim−→ is a covering class in Mod-R, cf. [11, Part II]. A. If C is closed and complete, then A 2.7. Locally free modules. The following less well known notation from [14] and [23] will be convenient: Definition 2.1. Let R be a ring and λ an infinite regular cardinal. A system S consisting of <λ-presented submodules of a module M satisfying the conditions (1) S is closed under unions of well-ordered ascending chains of length < λ, and (2) each subset X ⊆ M such that X < λ is contained in some N ∈ S, is called a λ-dense system of submodules of M . Of course, M is then the directed union of these submodules. Let F be a set of countably presented modules. Denote by C the class of all modules possessing a countable F -filtration. A module M is locally F -free provided that M contains an ℵ1-dense system of submodules from C. Notice that if M is countably presented, then M is locally F -free, if and only if M ∈ C. We will mainly be interested in the case when F = A≤ω for a cotorsion pair C = (A, B). Then C = A≤ω, and a module is locally A≤ω-free if and only if it admits an ℵ1-dense system of countably presented submodules from A. For a different description of locally A≤ω-free modules, see [4, Theorem 4.4]. We note that by [14], the locally P ≤ω 0 -free modules coincide with the flat Mittag- Leffler modules from Definition 3.1 below. 2.8. Filter-closed classes. For every directed set (I, ≤), there is an associated filter FI on (P(I), ⊆), namely the one with a basis consisting of the upper sets ↑ i = {j ∈ I j ≥ i} for all i ∈ I, that is FI = {X ⊆ I (∃i ∈ I)(↑ i ⊆ X)}. Definition 2.2. Let F be a filter on the power set P(X) for some set X, and let {Mx x ∈ X} be a set of modules. Set M = Qx∈X Mx. Then the F-product ΣFM is the submodule of M such that ΣFM = {m ∈ M z(m) ∈ F} where for an element m = (mx x ∈ X) ∈ M , we denote by z(m) its zero set {x ∈ X mx = 0}. For example, if F = P(X) then ΣFM = M is just the direct product, while if F is the Fr´echet filter on X then ΣFM = Lx∈X Mx is the direct sum. The module M/ΣFM is called an F-reduced product. Note that for a, b ∈ M , we have the equality ¯a = ¯b in the F-reduced product if and only if a and b agree on a set of indices from the filter F. In the case that Mx = My for every pair of elements x, y ∈ X, we speak of an F-power and an F-reduced power (of the module Mx) instead of an F-product and an F-reduced product, respectively. If moreover F is an ultrafilter on X, then the F-reduced power is called an ultrapower of Mx. Finally, a nonempty class of modules G is called filter-closed, if it is closed under arbitrary F-products (for any set X and any arbitrary filter F on P(X)). Notice that a class of modules is filter-closed in case it is closed under direct products and direct limits (of direct systems consisting of monomorphisms), see [17, Lemma 3.3.1]. APPROXIMATIONS, MITTAG-LEFFLER CONDITIONS -- TOOLS 5 3. Flat Mittag-Leffler modules and approximations We start with the notion of a flat Mittag-Leffler module studied since the classic works of Grothendieck [12] and Raynaud-Gruson [18]. From its many facets, we choose the one involving tensor products for a definition: Definition 3.1. Let R be a ring. A module M is flat Mittag-Leffler provided that the functor M ⊗R − : R-Mod → Mod-Z is exact, and the canonical group homomorphism ϕ : M ⊗R Qi∈I Qi → Qi∈I M ⊗R Qi defined by ϕ(m ⊗R (qi)i∈I ) = (m ⊗R qi)i∈I is monic for each family of left R-modules (Qi i ∈ I). The class of all flat Mittag-Leffler modules will be denoted by FM. If R is a right perfect ring, then flat Mittag-Leffler modules coincide with the projective modules, and form a covering class of modules by a classic theorem by Bass. However, if R is not right perfect (e.g., if R is right noetherian, but not right artinian), then the class FM is not deconstructible [14, Corollary 7.3]. In [22, Theorem 1.2(iv)], the Singular Cardinal Hypothesis (SCH) was used to prove that FM is not even precovering when R has cardinality ≤ ℵ0. The assumption of SCH was removed in [6]. Our goal in this section is to remove the cardinality restriction on R as well, and thus to obtain a basic example of a non-precovering class of modules over an arbitrary non-right perfect ring R. The following lemma is formulated for the more general setting of locally F -free modules (see Definition 2.1): Lemma 3.2. Let F be a class of countably presented modules, and L the class of all locally F -free modules. Let N be a Bass module over F . Assume that N is not a direct summand in a module from L. Then N has no L-precover. Proof. As the direct limit, N is an epimorphic image of a direct sum of modules from F . Since any direct sum of modules from F belongs to L (cf. [23, Lemma 4.2]), all the potential L-precovers of N have to be surjective. For the sake of contradiction, let us assume that 0 −→ M m−→ A f −→ N −→ 0 is a short exact sequence where f is an L-precover of N . Let κ be an infinite cardinal such that R ≤ κ and M ≤ 2κ = κω. By [23, Lemma 5.6], there exists a short exact sequence (1) 0 −→ D ⊆ −→ L −→ N (2κ) −→ 0 such that L ∈ L and D is a direct sum of κ modules from F . Applying HomR(−, m) to (1), we obtain the following commutative diagram with exact rows HomR(D, M ) HomR(D,m)y HomR(D, A) −−−−→ Ext1 δ−−−−→ Ext1 Ext1 R(N,m)2 R(N, M )2κ κy R(N, A)2κ −−−−→ Ext1 R(L, M ) Ext1 R(L,m)y −−−−→ Ext1 R(L, A). Since L ∈ L and f is an L-precover, the map Ext1 Ker(Ext1 Im(δ), and so Ext1 HomR(N, f ) being onto. In particular, f splits -- a contradiction. R(L, m) is monic. It follows that ⊆ Im(δ). However, by our assumption, 2κ ≥ HomR(D, M ) ≥ R(N, m) has to be monic as well. The latter is equivalent to (cid:3) R(N, m))2κ Now, we can easily prove our first main result: 6 JAN SAROCH Theorem 3.3. Let R be an arbitrary ring. Then FM is precovering, if and only if R is right perfect. Proof. The if part is well known: FM = P0 whenever R is a right perfect ring. Assume R is not right perfect. By a classic result of Bass, there exists a countably presented flat module (= a Bass module over P <ω ) N such that N /∈ P0. By [14, Corollary 2.10(i)], the class FM coincides with the class of all locally F -free modules for F = P ≤ω . In particular, N /∈ FM. So Lemma 3.2 applies, and shows that N has no FM-precover. (cid:3) 0 0 For further applications of Lemma 3.2, we refer to [4, §4], [24, §3] and [20, §4]. 4. Stationarity and the vanishing of Ext Next, we develop several technical tools involving Mittag-Leffler conditions and pure injectivity. It turns out that these tools are available in a rather general setting, but one has to work with relative Mittag-Leffler conditions rather than the 'absolute' ones. More precisely, we will use B-stationary modules, where B will be a class closed under direct products and direct limits. In the present Section, we review this concept with a special emphasis on its relationship to the vanishing of Ext. In Section 5 we will see that the classes B as above admit a pure-injective cogenerator C, and we will thus focus on C- stationarity. Let us first recall the classic notions related to inverse systems of modules (see [2], [18]): Definition 4.1. Let R be a ring and H = (Hi, hij i ≤ j ∈ I) be an inverse system of modules, with the inverse limit (H, hi i ∈ I). Then H is a Mittag-Leffler inverse system, provided that for each k ∈ I there exists k ≤ j ∈ I such that Im(hkj) = Im(hki) for each j ≤ i ∈ I, that is, the terms of the decreasing chain (Im(hki) k ≤ i ∈ I) of submodules of Hk stabilize. If moreover the stabilized term Im(hkj ) equals Im(hk), then H is called strict Mittag- Leffler. Notice that the two notions coincide when I is countable. Let B be a module and M = (Mi, fji i ≤ j ∈ I) be a direct system of finitely presented modules, with the direct limit (M, fi i ∈ I). Applying the contravariant functor HomR(−, B), we obtain the induced inverse system H = (Hi, hij i ≤ j ∈ I), where Hi = HomR(Mi, B) and hij = HomR(fji, B) for all i ≤ j ∈ I. Let B be a class of modules. A module M is B-stationary (strict B-stationary), provided that M can be expressed as the direct limit of a direct system M of finitely presented modules so that for each B ∈ B, the induced inverse system H is Mittag-Leffler (strict Mittag-Leffler). If B = {B} for a module B, we will use the notation B-stationary instead of {B}-stationary, and similarly for the strict stationarity. Remark 1. Let B be a module. Then the strict B-stationarity of M can equiv- alently be expressed as follows: for each l ∈ I, there exists l ≤ i ∈ I such that Im(HomR(fil, B)) ⊆ Im(HomR(fl, B)), see [2, §8]. Moreover, the notions of a B- stationary and strict B-stationary module coincide in the case when M is countably presented. They also coincide for B (locally) pure-injective by [13, Proposition 1.7]. In fact, if M is B-stationary (strict B-stationary), then the induced inverse system H is Mittag-Leffler (strict Mittag-Leffler) for each presentation of M as the direct limit of a direct system M of finitely presented modules, see [2]. Also, FM coincides with the class of all flat R-stationary modules, cf. [18]. APPROXIMATIONS, MITTAG-LEFFLER CONDITIONS -- TOOLS 7 First, we will deal with the strict B-stationarity of the modules in ⊥B. The fol- lowing result is a mix of Lemmas 2.3 and 2.5 from [21]; for the reader's convenience, we provide a detailed proof here: Lemma 4.2. Let B be a filter-closed class of modules. Then every module M ∈ ⊥B is strict B-stationary. Proof. We will prove the following: if (M, fi i ∈ I) is the direct limit of a direct system M = (Mi, fji i ≤ j ∈ I) consisting of finitely generated modules, then for each l ∈ I there exists l ≤ i ∈ I such that for each B ∈ B and g ∈ HomR(Mi, B), we have gfil ∈ Im(HomR(fl, B)). Suppose that the claim is not true, so there exists l ∈ I such that for each l ≤ i ∈ I there exist Bi ∈ B and gi : Mi → Bi, such that gifil does not factor through fl. For i ∈ I such that l (cid:2) i, we let Bi = 0 and gi = 0. Put B = Qi∈I Bi. We define a homomorphism hji : Mi → Bj for each pair i, j ∈ I in the following way: hji = gjfji if i ≤ j and hji = 0 otherwise. This family of maps gives rise to the canonical homomorphism h : Lk∈I Mk → B. More precisely, if we denote by πj : B → Bj the canonical projection and by νi : Mi → Lk∈I Mk the canonical inclusion, h is the (unique) map such that πj hνi = hji. Note that for every i, j ∈ I such that i ≤ j, the set {k ∈ I hki = hkjfji} is in the associated filter FI since it contains each k ≥ j. Hence, if we denote by ϕ the canonical pure epimorphism Li∈I Mi → M = lim−→i∈I Mi (such that ϕνi = fi for all i ∈ I), then h(Ker(ϕ)) ⊆ ΣFI B. So there is a well-defined homomorphism u from M to the FI - reduced product B/ΣFI B making the following diagram commutative (ρ denotes the canonical projection): B ρ −−−−→ B/ΣFI B −−−−→ 0 hx Li∈I Mi ux ϕ −−−−→ M −−−−→ 0. Since B is filter-closed, ΣFI B ∈ B, whence Ext1 there exists g ∈ HomR(M, B) such that u = ρg. R(M, ΣFI B) = 0. It follows that For each i ∈ I, we have ρgfi = ρgϕνi = ρhνi; so gfi − hνi maps Mi into ΣFI B. Since Mi is finitely generated, there exists i ≤ j ∈ I such that πkgfi = πkhνi for all j ≤ k ∈ I. However, πkhνi = hki = gkfki. In particular, for i = l and k = j, we infer that πj gfl = gjfjl. Thus, gjfjl factors through fl, a contradiction. (cid:3) We will also need the following variant of [21, Proposition 2.7]. Proposition 4.3. Let G be a class of modules and M a countably presented module such that M ∈ ⊥G. Then the following is equivalent: (1) M ∈ ⊥D for each module D isomorphic to a pure submodule of a product of modules from G; (2) M ∈ ⊥D for each module D isomorphic to a countable direct sum of modules from G; (3) M is G-stationary. Proof. Since direct sums are pure in the corresponding direct products, the impli- cation (1) ⇒ (2) is trivial. The implication (2) ⇒ (1) is exactly [21, Proposition 2.7] (for G replaced by its closure under countable direct sums). Its proof in [21] proceeds by showing that (2) ⇒ (4) ⇒ (1), where (4) says that M is D-stationary for any module D isomorphic to a pure submodule of a product of modules from G. However, (4) is equivalent to (3) by [2, Corollary 3.9], whence (2) ⇒ (3) ⇒ (1). (cid:3) 8 JAN SAROCH The next lemma on strict B-stationarity is inspired by [13, Lemma 3.15]. Lemma 4.4. Let B be a module and 0 → N → A → M → 0 a short exact sequence of modules such that M ∈ ⊥(B(I))cc for each set I. Then N is strict B-stationary if so is A. Proof. According to [2, Theorem 8.11] (see also [25]), a necessary and sufficient condition for the strict B-stationarity of N is that for each set I, the canonical map ν : N ⊗R HomZ(B, (Q/Z)I ) → HomZ(HomR(N, B), (Q/Z)I ) defined by ν(n ⊗ f )(g) = f (g(n)) is injective. Since HomZ(B, (Q/Z)I ) ∼= (B(I))c, we have TorR 1 (M, HomZ(B, (Q/Z)I )) = 0 by our assumption on M . From the commutative diagram 0 −−−−→ N ⊗R HomZ(B, (Q/Z)I ) −−−−→ A ⊗R HomZ(B, (Q/Z)I ) νy αy HomZ(HomR(N, B), (Q/Z)I ) −−−−→ HomZ(HomR(A, B), (Q/Z)I ), we infer that ν is injective, because α is injective by our assumption on A. (cid:3) 5. Stationarity and pure-injectivity The classes of pure-injective modules especially relevant here are the Σ-pure in- jectives, and the elementary cogenerators. We will not deal with the model theoretic background of these notions; we just note that, by a classic theorem of Frayne, if two modules M and N are elementarily equivalent, then M is a pure submodule of an ultrapower of N . The pure-injective hull P E(M ) of a module M is elementarily equivalent to M by a theorem of Eklof and Sabbagh, and so is its double dual M cc, cf. [16]. The relation between stationarity and P-pure-injectivity goes back to work by Zimmermann [25]. Lemma 5.1. For a module C, the following conditions are equivalent: (1) C is Σ-pure-injective; (2) all modules are strict C-stationary; (3) all modules are C-stationary. (4) all countably presented modules are C-stationary. Proof. The equivalence of the first two conditions comes from [25, Theorem 3.8], and (2) ⇒ (3) ⇒ (4) are trivial. For the implication (4) ⇒ (3), see [2, Proposition 3.10]. It remains to prove that (3) implies (1). Applying [2, Corollary 3.9], we get that all modules are B-stationary, where B is any pure submodule of a pure-epimorphic image of a direct product of copies of C. In particular, by [17, Lemma 3.3.1], this holds for any pure-injective module B which is elementarily equivalent to C (e.g., for B = P E(C)). Since B is pure- injective, it follows from Remark 1 that all modules are even strict B-stationary, and B is Σ-pure-injective by the above. Thus C is Σ-pure-injective, because C is a pure submodule of B. (cid:3) We now turn to elementary cogenerators. Definition 5.2. A pure-injective module E is called an elementary cogenerator, if every pure-injective direct summand of a module elementarily equivalent to Eℵ0 is a direct summand of a direct product of copies of E. APPROXIMATIONS, MITTAG-LEFFLER CONDITIONS -- TOOLS 9 Notice that by [16, Corollary 9.36], for each module M there exists an elemen- tary cogenerator which is elementarily equivalent to M . This allows to prove the following result which will play an important role in the sequel. Lemma 5.3. Let B be a class of modules closed under direct products and direct limits. Then B contains a pure-injective module C, such that each module B ∈ B is a pure submodule of a direct product of copies of C. Moreover, if B contains a cogenerator, then C is a cogenerator for Mod-R. Proof. First, it is a well-known fact that B = lim−→ direct summands. B yields that B is closed under Now the closure properties of B imply that if B ∈ B is elementarily equivalent to a pure-injective module A, then also A ∈ B. In particular, B is closed under taking pure-injective hulls, and double duals. Thus we can choose among the modules from B a representative set for elemen- tary equivalence, S, consisting of elementary cogenerators. Let C = Q S. Then C ∈ B is pure-injective, and it has the property that every module B ∈ B is iso- morphic to a pure submodule in a direct product of copies of C. Indeed, we have Bcc ∈ B; moreover Bcc is a pure-injective direct summand of (Bcc)ℵ0 which is a module elementarily equivalent to Eℵ0 for some E ∈ S. Hence Bcc is a direct summand in a direct product, D, of copies of C (by Definition 5.2), and B is pure in Bcc, and hence in D. Moreover, if B ∈ B is a cogenerator for Mod-R, then C cogenerates Mod-R as (cid:3) well. Given a pure-injective cogenerator C as above, one can use the following lemma to find in any (strict) C-stationary module a rich supply of countably presented C-stationary submodules. Lemma 5.4. Let C be a pure-injective module which cogenerates Mod-R, and M be a strict C-stationary module. Then there exists an ℵ1-dense system L of strict C-stationary submodules of M such that HomR(M, C) → HomR(N, C) is surjective (†) for every directed union N of modules from L. Proof. Let Q = C c. By [2, Proposition 8.14(2)], a module A is strict C-stationary if and only if it is Q-Mittag-Leffler. Repeatedly using [2, Theorem 5.1(4)] for M , we obtain a ⊆-directed set F of countably presented Q-Mittag-Leffler submodules of M satisfying (2) from Definition 2.1 for λ = ℵ1: observe that the map v in the statement of [2, Theorem 5.1(4)] is monic since the injectivity of v ⊗R Q implies (†) (we use that C is a direct summand in Qc) and C is a cogenerator. We extend F gradually by adding countable directed unions, noticing along the way that each newly added countably presented module N has the property that N ⊆ M stays monic after applying − ⊗R Q (as TorR 1 (−, Q) commutes with direct limits), and it is Q-Mittag-Leffler by [2, Corollary 5.2]. In this way, we eventually arrive at the desired ℵ1-dense system L of strict C-stationary submodules of M . Note that any directed union of modules from L satisfies (†) since (†) is implied by the injectivity of the corresponding tensor map, which, in turn, is a property preserved by taking directed unions. (cid:3) Remark 2. Lemma 5.4 holds with ℵ1 replaced by any regular uncountable cardinal. However, we won't need it in this generality. A useful tool for proving that many classes of the form ⊥B are deconstructible is provided by 10 JAN SAROCH Lemma 5.5. [21, Proposition 2.4] Let B be a filter-closed class of modules such that B cogenerates Mod-R. Then for each uncountable regular cardinal λ and each module M ∈ ⊥B, there is a λ-dense system Cλ of submodules of M such that M/N ∈ ⊥B for all N ∈ Cλ. 6. Countable type Let C = (A, B) be a hereditary cotorsion pair in Mod-R. The Telescope Con- jecture for Module Categories asserts that C is of finite type whenever A and B are closed under direct limits. This statement is known to be true in some special cases [3]. On the other hand, the Countable Telescope Conjecture was proved in full generality in [21, Theorem 3.5]. It states that C is of countable type whenever B is closed under unions of well-ordered chains. As an application of the tools developed above, we prove a version of the Count- able Telescope Conjecture for not necessarily hereditary cotorsion pairs. Theorem 6.1. Let C = (A, B) be a cotorsion pair with lim−→ countable type, and B is definable. B = B. Then C is of Proof. Let C be the pure-injective module constructed for B in Lemma 5.3. Let A0 = A≤ω. By induction on κ, we are going to prove that each κ-presented module M ∈ A is A0-filtered. There is nothing to prove for κ ≤ ℵ0, so let κ be uncountable. In the κ-presented module M ∈ A, we will construct by induction, for each uncountable regular cardinal λ ≤ κ, a λ-dense system Cλ of submodules of M such that M/N ∈ A for each N ∈ Cλ, and (2) Cλ ⊆ A. Our strategy is to start with a Cλ given by Lemma 5.5 for G = B, and then select a suitable subfamily in A. Note that it is enough to ensure that for every N ∈ Cλ there exists L ∈ Cλ, such that N ⊆ L ∈ A. Then the family Cλ ∩ A is the desired one, since each ascending chain in Cλ ∩ A is actually an A-filtration and A is closed under filtrations. Indeed, for each B ∈ B, if N ∈ Cλ and L ∈ A are such that N ⊆ L, then all homomorphisms from N to B extend to M , and hence to L; thus L/N ∈ ⊥{B}. We will distinguish the following four cases: Case 1. λ = ℵ1: Fix a free presentation of M 0 −−−−→ K −−−−→ R(κ) f −−−−→ M −−−−→ 0. Let Cℵ1 be an ℵ1-dense system in M provided by Lemma 5.5. After taking the intersection with the set of all images f (R(X)) where X is a countable subset of κ, we can assume that Cℵ1 is compatible with this presentation, that is, each N ∈ Cℵ1 has the form f (R(XN )) for a countable subset XN of κ. Let K = { Ker(f ↾ R(XN )) N ∈ Cℵ1}. Since B is closed under coproducts and double duals, it follows from Lemma 4.4 that K is strict C-stationary. Thus we can use Lemma 5.4 to obtain another ℵ1-dense system, L, this time consisting of submodules of K. Clearly, the system K ∩ L is ℵ1-dense as well. Our new Cℵ1 is defined as {N ∈ Cℵ1 Ker(f ↾ R(XN )) ∈ L}. Notice that Cℵ1 ⊆ ⊥C. Indeed, given a module N ∈ Cℵ1 , each h : Ker(f ↾ R(XN )) → C can be extended to some h′ : R(XN ) → C by the property (†) from Lemma 5.4 and by the fact that M ∈ ⊥C. Let N ∈ Cℵ1 . Since M/N ∈ A, N is the kernel of the epimorphism M → M/N between two modules from A. By Lemma 4.2, M is strict C-stationary, so using Lemma 4.4 again, we see that each N ∈ Cℵ1 is a countably presented C-stationary APPROXIMATIONS, MITTAG-LEFFLER CONDITIONS -- TOOLS 11 module from ⊥C. Using Proposition 4.3 (for G = {C}) together with the properties of C guaranteed by Lemma 5.3, we conclude that Cℵ1 ⊆ A. Case 2. λ weakly inaccessible: As in the previous step, we start with a family Cλ provided by Lemma 5.5. Since each N0 ∈ Cλ is < µ-presented for some regular uncountable cardinal µ < λ, we simply choose N ′ 0 ∈ Cµ ⊆ A containing N0 as a submodule. Then there is N1 ∈ Cλ containing N ′ 0, etc. The union, N , of the chain N0 ⊆ N ′ 1 ⊆ · · · satisfies N ∈ Cλ. Moreover, N ∈ A since the chain N ′ 0 ⊆ N ′ 0 ⊆ N1 ⊆ N ′ 1 ⊆ · · · is an A-filtration of N . λ = ν+ for a regular cardinal ν. Case 3. λ a successor of a regular cardinal : For each N0 from Cλ (not necessarily satisfying condition (2) above), we easily build a continuous chain N0 = (N 0 α α < ν) of modules from Cν ⊆ A so that N0 ⊆ S N0. Again, the union is in A. We continue by choosing N1 ∈ Cλ containing this union, and a chain N1 = (N 1 α for all α < ν, and N1 ⊆ S N1. We proceed further by taking N2 ∈ Cλ, etc. Clearly, N = Si<ω Ni ∈ Cλ. Furthermore, the ascending chain α α < ν) in Cν such that N 0 α ⊆ N 1 [ N i 0 ⊆ [ N i 1 ⊆ · · · ⊆ [ N i α ⊆ · · · i<ω i<ω i<ω of submodules from Cν forms an A-filtration of N , showing that N ∈ A. λ = ν+ where µ = cf(ν) < ν. Case 4. λ a successor of a singular cardinal : We choose a strictly increasing continuous chain (να α < µ) of infinite cardinals which is cofinal in ν, such that ν0 > µ. Let N0 be arbitrary module from the family Cλ given by Lemma 5.5. We will produce a similar ascending chain as in Case 3, however this time, we have to pick the modules from different classes Cδ, hence we lose continuity. To overcome this problem, we use a well known singular compactness argument: We gradually build the chains Ni = (N i α α < µ), for i < ω, and pick the modules Ni ∈ Cλ in an alternating way, so that the following conditions are satisfied for all i < ω: α ∈ Cν+ α ; (a) N i (b) S Ni ⊆ Ni+1; (c) the generators {ni (d) the generators {ni (e) N i α ⊇ {ni−1 α,β β < να} of the modules N i γ γ < ν} of Ni are fixed; δ,β δ < µ & β < min(νδ, να)} ∪ {ni γ γ < να}. α are fixed; Then Si<ω Ni is in Cλ, and it is equal to the union of the chain H = (Si<ω N i α α < µ). However, this chain is continuous (by the condition (e)), and provides thus an A-filtration of S H. Having constructed the families Cλ (λ ≤ κ), we can use them to build an A- filtration of M consisting of < κ-presented modules. If κ is regular, then it is easy to see that Cκ already contains an A-filtration of M . For κ singular, we apply [21, Lemma 3.2]. By inductive hypothesis, we conclude that M is A0-filtered. It remains to show that the modules in A0 have countably presented first syzy- gies. We know from Lemma 4.4 that their first syzygies are strict B-stationary. In particular, the first syzygies are RRc-stationary. By [2, Proposition 8.14(1)], they are then RR-Mittag-Leffler, and the claim follows from [2, Corollary 5.3]. Finally, B is definable by [11, Theorem 13.41]. (cid:3) We finish this section by examples of cotorsion pairs which satisfy the conditions of Theorem 6.1, but are not hereditary. 12 JAN SAROCH Recall that for n < ω, a module M is an FPn-module provided that M has a projective resolution · · · → Pk+1 → Pk → · · · → P1 → P0 → M → 0 such that Pk is finitely generated for each k ≤ n, [7, §VIII.4]. Let FPn denote the class of all FPn-modules. (Notice that FP0 is the class of all finitely generated modules, and FP1 the class of all finitely presented ones.) The classes (FPn n < ω) form a decreasing chain whose intersection is the class of all strongly finitely presented modules, that is, the modules possessing a projective resolution consisting of finitely generated modules, see [7, Proposition VII.4.5]. Example 6.2. Let n ≥ 2 and let Rn be a ring such that there exists a module M ∈ FPn \FPn+1. Such rings were constructed by Bieri and Stuhler using integral representation theory, see [7, §VIII.5] and the references therein; in fact, Rn can be taken as the integral group ring ZGn for a suitable finitely generated group Gn, and M = Z. (For the particular case of n = 2, a different and simpler example of R2 is constructed in [15, Example 1.4].) Let Cn = (An, Bn) be the cotorsion pair generated by FPn. Since n ≥ 2, Bn is closed under direct limits by [11, Lemma 6.6], so Theorem 6.1 applies. In order to see that Cn is not hereditary, we observe that An is not closed under kernels of epimorphisms: indeed, the syzygy module Ω(M ) does not belong to FPn, and hence Ω(M ) /∈ An. To see the latter fact, notice that An consists of direct summands of FPn-filtered modules by [11, 6.14]. Since Ω(M ) is finitely generated, if Ω(M ) ∈ An then it is a direct summand of a finitely FPn-filtered module, by [11, Theorem 7.10] used for ℵ0. However, FPn is closed under extensions and direct summands, whence Ω(M ) ∈ FPn, a contradiction. References [1] L. Angeleri Hugel, S. Bazzoni, D. Herbera, A solution to the Baer splitting problem, Trans. Amer. Math. Soc. 360 (2008), 2409 -- 2421. [2] L. Angeleri Hugel, D. Herbera, Mittag-Leffler conditions on modules, Indiana Math. J. 57 (2008), 2459 -- 2517. [3] L. Angeleri Hugel, J. Saroch, J. Trlifaj, On the telescope conjecture for module categories, J. Pure Appl. Algebra 212 (2008), 297 -- 310. [4] L. Angeleri Hugel, J. Saroch, J. Trlifaj, Approximations and Mittag-Leffler conditions -- the applications, preprint, arXiv:1612.xxxxx. [5] S. Bazzoni, D. Herbera, One dimensional tilting modules are of finite type, Algebras Repres. Theory 11 (2008), 43 -- 61. [6] S. Bazzoni, J. Stov´ıcek, Flat Mittag-Leffler modules over countable rings, Proc. Amer. Math. Soc. 140 (2012), 1527 -- 1533. [7] K. S. Brown, Cohomology of Groups, Springer Vlg., New York 1982. [8] V. Drinfeld, Infinite -- dimensional vector bundles in algebraic geometry: an introduction, in The Unity of Mathematics, Birkhauser, Boston 2006, 263 -- 304. [9] P. C. Eklof, A. H. Mekler, Almost Free Modules, Revised ed., North -- Holland, New York 2002. [10] S. Estrada, P. Guil Asensio, M. Prest, J. Trlifaj Model category structures arising from Drinfeld vector bundles, Adv. Math. 231 (2012), 1417 -- 1438. [11] R. Gobel, J. Trlifaj, Approximations and Endomorphism Algebras of Modules, de Gruyter Expositions in Mathematics 41, 2nd revised and extended edition, Berlin-Boston 2012. [12] A. Grothendieck, ´El´ements de g´eom´etrie alg´ebrique. III. ´Etude cohomologique des fais- ceaux coh´erents, Inst. Hautes ´Etudes Sci. Publ. Math. No. 11, 1961. [13] D. Herbera, Definable classes and Mittag-Leffler conditions, Ring Theory and Its Appl., Contemporary Math. 609 (2014), 137 -- 166. [14] D. Herbera, J. Trlifaj, Almost free modules and Mittag-Leffler conditions, Advances in Math. 229 (2012) 3436-3467. [15] S. Kabbaj, N. Mahdou, Trivial extensions of local rings and a conjecture of Costa, in Lect. Notes in Pure Appl. Math. 231 (2002), M. Dekker, 301 -- 311. APPROXIMATIONS, MITTAG-LEFFLER CONDITIONS -- TOOLS 13 [16] M. Prest, Model Theory and Modules, London Math. Soc. Lec. Note Ser. 130, Cambridge University Press, Cambridge, 1988. [17] M. Prest, Purity, Spectra and Localisation, Enc. Math. Appl. 121, Cambridge Univ. Press, Cambridge 2009. [18] M. Raynaud, L. Gruson, Crit`eres de platitude et de projectivit´e, Invent. Math. 13 (1971), 1 -- 89. [19] P. Rothmaler, Mittag-Leffler modules and positive atomicity, Habilitationsschrift, Christian -- Albrechts -- Universitat zu Kiel 1994. [20] J. Saroch, On the non-existence of right almost split maps, Invent. Math. (2016), doi:10.1007/s00222-016-0712-2. [21] J. Saroch, J. Stov´ıcek, The countable Telescope Conjecture for module categories, Adv. Math. 219 (2008), 1002 -- 1036. [22] J. Saroch, J. Trlifaj, Kaplansky classes, finite character, and ℵ1-projectivity Forum Math. 24 (2012), 1091 -- 1109. [23] A. Sl´avik, J. Trlifaj, Approximations and locally free modules, Bull. London Math. Soc. 46 (2014), 76 -- 90. [24] A. Sl´avik, J. Trlifaj, Very flat, locally very flat, and contraadjusted modules, J. Pure Appl. Algebra 220 (2016), 3910 -- 3926. [25] W. Zimmermann, Modules with chain conditions for finite matrix subgroups, J. Algebra 190 (1997), 68 -- 87. Charles University, Faculty of Mathematics and Physics, Department of Algebra, Sokolovsk´a 83, 186 75 Praha 8, Czech Republic E-mail address: [email protected]
1708.01201	1	1708	2017-08-03T16:40:35	Logarithms Over a Real Associative Algebra	[ "math.RA" ]	Extending the work of Freese and Cook, which develop the basic theory of calculus and power series over real associative algebras, we examine what can be said about the logarithmic functions over an algebra. In particular, we find that for any multiplicative unital nil algebra the exponential function is injective, and hence the algebra has a unique logarithm on the image of the exponential. We extend this result to show that for a large class of algebras, the logarithms behave incredibly similarly to the logarithms over the real and complex numbers depending on if they are "Type-R" or "Type-C" algebras.	math.RA	math	Logarithms Over a Real Associative Algebra Nathan BeDell [email protected] August 4, 2017 Abstract Extending the work of Freese [4] and Cook [3], which develop the basic theory of calculus and power series over real associative algebras, we examine what can be said about the logarithmic functions over an algebra. In particular, we find that for any multiplicative unital nil algebra the exponential function is injective, and hence the algebra has a unique logarithm on the image of the exponential. We extend this result to show that for a large class of algebras, the logarithms behave incredibly similarly to the logarithms over the real and complex numbers depending on if they are "Type-R" or "Type-C" algebras. 1 Introduction The theory of analysis over commutative unital associative algebras, and more gen- erally the theory of analysis over more general contexts such as for non-commutative or even non-associative algebras is long and complex, with authors often re-inventing the wheel, and a number of different approaches which are not all compatible. Sur- prisingly however, it does not seem that there is any literature available that studies the properties of the logarithm for completely general commutative associative alge- bras of finite dimension. The closest to this might be [5], where the properties of the logarithm are deduced for algebras that take C as their base field. In this paper, we wish to develop the theory of the logarithm in the more general context where we take algebras A with R as a base field, since an algebra over the complex numbers as a base field can simply be viewed as the real algebra C ⊗ A, since C is itself a two dimensional R-algebra. Throughout this paper, unless we explicitly mention it, by an algebra we mean an associative finite dimensional commutative unital algebra over the reals. Definition 1.1. An algebra A is a finite dimensional real vector space together with a bilinear multiplication operation ⋆ : A × A → A satisfying the following properties: 1 1. v ⋆ (w ⋆ z) = (v ⋆ w) ⋆ z for all v, w, z ∈ A 2. There exists an element 1 ∈ A such that 1 ⋆ z = z ⋆ 1 = z for all z ∈ A. If v ⋆ w = w ⋆ v for all v, w ∈ A then we say A is a commutative algebra. A linear map φ : A → B between two algebras is an algebra morphism if it satisfies the homomorphism property φ(z ⋆ w) = φ(z) ⋆ φ(w) and φ(1) = 1. Proposition 1.2. Given an algebra A with basis β = {v1, . . . , vn} and a linear map φ : A → B between two algebras, if φ(vi ⋆ vj) = φ(vi) ⋆ φ(vj) for all basis elements vi, vj and φ(1) = 1, then φ(v ⋆ w) = φ(v) ⋆ φ(w) for all v, w ∈ A. In addition to this, it will be important to note throughout the paper the following as a consequence of the fact that our algebras are finite dimensional vectors spaces: Proposition 1.3. Let A be a commutative algebra, then A is a Noetherian and Artinian ring, In particular, this implies that every ideal I of A is finitely generated. One common way of describing an algebra is by representing it as a suitable subspace of Rn×n, with the usual matrix addition and multiplication representing the multiplication and addition in the algebra. Given a basis β = {v1, . . . , vn} for the algebra, we define a matrix regular representation: Definition 1.4. Given an algebra A the set of all linear transformations T : A → A for which T (x ⋆ y) = T (x) ⋆ y forms the regular representation which we denote RA. Clearly T (x) = T (1 ⋆ x) = T (1) ⋆ x hence the regular representation is formed by left-multiplications of A. Denote the left-multiplication by α ∈ A by Lα(x) = α ⋆ x for each x ∈ A. Given basis β = {v1, . . . , vn}, we define Mβ(α) = [Lα]β and denote the collection of all such matrices by Mβ(A). Since A is unital and finite dimensional it is well-known that A, RA and Mβ(A) are isomorphic as algebras. In particular, we have the identifications: α ↔ Lα ↔ [Lα]β given a choice of basis β. For further discussion and some elementary proofs see [3]. Below we illustrate how to calculate Mβ(z) for an arbitrary element z ∈ A. Example 1.5. With respect to the basis β = {1, j}, let z = x + yj be an arbitrary element of H. (x + yj) ⋆ 1 = x + yj so the first column of Mβ(z) is (x, y). Also, (x + yj) ⋆ j = y + xj, so the second column of Mβ(z) is (y, x). Hence: Mβ(z) = (cid:20)x y y x(cid:21) 2 Example 1.6. Let β = {1, i} and let x = x + iy be an arbitrary element of C. Applying the same method as we did in the preceding example, we find (x + iy) ⋆ 1 = x + iy and (x + iy) ⋆ i = −y + ix. Hence: Mβ(z) = (cid:20)x −y x (cid:21) y Examples 1.5 and 1.6 illustrate an interesting result on the structure of matrix regular representations; given a unital basis {1, v2, . . . , vn} for the algebra A, the ith column of the regular representation is an A-multiple of the first column: Although this paper is not focused on the calculus of logarithms over an alge- bra, to keep the paper self-contained we do need to briefly discuss the notion of A-differentiability, and some of the basic properties of the A-derivative that we will use in Section 2. For a more complete exposition of this we direct the reader to [3]. Definition 1.7. Given a function f : A → A, we say that f is A-differentiable at z0 if the Frechet differential dfz0 exists, and dfzo ∈ RA. This condition implies that dfz0(z) = Lα(z) for some α ∈ A, and hence allows us to define the derivative by setting f ′(z0) = α, where α of course in general depends on z0. We also will use the notation df (z) dz = f ′(z) for the A-derivative. This notion of differentiability satisfies all of the basic properties of the usual real derivative, for example: Proposition 1.8. For A-differentiable functions f, g : A → A, 1. (f (g(z)))′ = f ′(g(z)) ⋆ g′(z). 2. d dz zn = nzn for powers n ∈ N. 3. (f (z) ⋆ g(z))′ = f ′(z) ⋆ g(z) + f (z) ⋆ g′(z) (here we assume A is commutative) 4. if c ∈ A a constant then (cf (z) + g(z))′ = cf ′(z) + g′(z). 1.1 Semisimple and Nil Algebras Another convenient way to describe a finite dimensional, commutative, and unital algebra is as the quotient of some real polynomial ring by an ideal. If we have such an algebra A isomorphic to P = R[x1, . . . , xk]/I for some k ∈ N and I an ideal of R[x1, . . . , xk] we say that P is a presentation of the algebra A. Definition 1.9 (Standard presentations of typical algebras). 1. The n-hyperbolic numbers: Hn := R[j]/hjn − 1i 3 2. The n-complicated numbers: Cn := R[i]/hin + 1i 3. The n-nil numbers: Γn := R[ǫ]/hǫni 4. The total n-nil numbers: Ξn := R[ǫ1, . . . , ǫn]/hǫiǫji, j ∈ {1, 2, . . . , n}i For example, C2 is just the usual complex numbers, denoted simply C, and H2 is just the hyperbolic numbers, denoted simply H. Similarly, we take the convention that Γ by itself denotes Γ2. The nil numbers are a special class of what in our terminology we will call unital nil algebras – that is, an algebra with basis {1, ǫ1, . . . , ǫn−1}, where each ǫk is nilpotent. In other words, for each ǫk there exists m ∈ N such that (ǫk)m = 0. This terminology is inspired by the use of the term nil algebra used by Abian [1] to refer to an algebra in which every element of the algebra is nilpotent. However, we will sometimes say simply "nil algebra" in this paper when we mean "unital nil algebra", since in our context all algebras are unital. A unital nil algebra is in some sense the closest you can get to a nil algebra while still being a unital algebra. Definition 1.10. Given an algebra A, let Nil∗(A) denote the smallest unital sub- algebra of A that contains the nilradical of A, Nil(A) – that is, the ideal formed from all nilpotent elements of A. Proposition 1.11. An algebra A is a unital nil algebra if and only if it is the smallest unital subalgebra of A containing the nilradical of A. In other words, A is a unital nil algebra if and only if A = Nil∗(A). Proof. If A is a unital nil algebra with unital nil basis {1, ǫ1, . . . , ǫn}, then clearly Nil(A) = hǫ1, . . . , ǫni, and A/Nil(A) ∼= R, so Nil(A) is maximal, and hence A is the smallest unital nil algebra containing Nil(A). Conversely, suppose that A = Nil∗(A), and let {ǫ1, ǫ2, . . . , ǫn} be a basis for Nil(A), then clearly {1, ǫ1, . . . , ǫn} is a linearly independent set that spans a subset of Nil∗(A). Also, we argue that this set must span Nil∗(A), since it contains all of Nil(A), and the unit 1, so by the minimality condition, this must be all of Nil∗(A). Thus, A = Nil∗(A) is a multiplicative nil algebra. In addition to these basic families, we should also mention the so called n-complex numbers Cn = C⊗n, where X ⊗n denotes the n-fold tensor product of rings.1 In particular, the analysis of the bicomplex numbers C2 has been studied extensively, for example, by Price [7]. 1For the unfamiliar reader, the tensor product of algebras can be thought of as the algebra which combines the set of generators and relations for a presentation for the algebra. In other words, R[x1, . . . , xn]/I ⊗ R[y1, . . . , ym]/J ∼= R[x1, . . . , xn, y1, . . . , ym]/(I + J). So C2 = C ⊗ C ∼= R[i1, i2]/hi2 2 + 1i for example. 1 + 1, i2 4 If A is a real vector space with basis β = {v1, . . . , vn} then given appropriate ij ∈ R we may define a multiplication on A. In particular, define structure constants ck vi ⋆ vj = n Xk=1 ck ijvk on basis elements, and extend bilinearly to define ⋆ on A. Naturally, the structure constants must be given such that the defined multiplication is associative and unital. That said, we typically begin with a given algebra A and simply use the structure constants with respect to a given basis to study the structure of A. This notion of structure constants will be important in our derivation of the injectivity of the exponential function in unital nil algebras. In some sense, the introduction of nilpotent elements into an algebra introduces complications into our theory. Essentially, this is because otherwise, the classification of algebras becomes rather simple. Definition 1.12. An algebra is called semisimple if it has no nilpotent elements. Usually, one defines a ring to be semisimple if it's Jacobson radical is zero. How- ever, because we are working in the finite dimensional context, our characterization is equivalent to the one involving the Jacobson radical. Theorem 1.13. Every commutative semisimple algebra A is isomorphic to Rn × Cm for some n, m ∈ N. Thus, by a more optimistic characterization, the non-semisimple case, where an algebra has non-trivial nilpotent elements, is where our theory truly departs from the theory of real and complex analysis non-trivially. 2 The Exponential and Logarithm Functions for an Algebra The exponential function over a commutative algebra always exists, and is defined via power series in A: Definition 2.1. Given an algebra A, we define the exponential function, denoted either ez or exp(z) on an algebra by setting ez = ∞ Xk=0 zk k! 5 Convergence of power series in an algebra is studied in [4] where it is shown: Theorem 2.2. For any algebra A, the series expansion for the exponential function has radius of convergence R = ∞, and hence is well-defined on the whole algebra. The exponential function for an algebra obeys all of the familiar algebraic prop- erties of the real exponential, in particular: Proposition 2.3. For all z, w ∈ A: 1. e0 = 1 2. ez+w = ezew 3. e−z = 1 ez 4. ez is a unit 2.1 Logarithms over Semisimple Algebras We now turn our attention to logarithms over a commutative algebra. (From this point on, an "algebra" always means a commutative one) The properties of the real and complex logarithms are of course well-studied. Some authors, such as [6], have even developed the theory of logarithms over algebras such as the hyperbolic numbers, so a natural question, given the results of Freese's series methods that the exponential function exists in any algebra is what can be said about the logarithm in an arbitrary algebra. By a logarithm, we mean an inverse function to ez for an algebra. Thus, for our convenience, we use the notation Ld(A) = Im(exp) to denote the logarithmic domain, or equivalently, the image of the exponential for an algebra. Specifically, since the exponential function may not be injective (for example, as is the case with ez : C → C), given an algebra A, and a connected subset B ⊆ A of the algebra such that ezB is injective, and Im(ezB) = Ld(A) we wish to find a function log : Im(exp) → B such that log(ez) = z for all z ∈ B, and elog(z) = z for all z ∈ Ld(A). We call this a branch of the logarithm for the algebra, borrowing some terminology from complex analysis. From this, we can define logarithms to other bases, which we denote by logb(z), using the standard change of basis formula for logarithms, namely logb(z) = log(z) log(b) . In addition to this, and a fact that is more relevant to our unpublished work concerning ODEs over an algebra, we may also from the existence of a logarithm define arbitrary power functions over an algebra, i.e. by defining ab = eb log(a) for a, b ∈ A. We prove first the existence of a logarithm on at least a sub-domain of the image of the exponential for an arbitrary algebra using series methods: 6 Theorem 2.4. Every algebra A has a function exp−1 : V ⊆ Ld(A) → A which is inverse to the exponential function exp : U ⊆ A → A. Proof. Begin by defining and f (z) = ∞ Xk=1 (−1)k−1 k zk g(z) = ∞ Xk=1 1 k! zk Notice then that g(z) = ez − 1, and to align with the standard series definition of log, we define in our algebra context log(z) = f (z − 1), so that f (z) = log(z + 1). By termwise differentiation of the series (see [4]), we obtain: and f ′(z) = ∞ Xk=1 (−1)k−1zk−1 = 1 1 + z g′(z) = ∞ Xj=1 j j! zj−1 = ∞ Xk=0 zk k! = 1 + g(z) and hence, if we define p(z) = f (g(z)), by the chain rule we obtain p′(z) = 1+g(z)(1 + g(z)) = 1, since for any z ∈ A, ez = 1 + g(z) ∈ U(A), and f ′(g(z))g′(z) = 1 hence Finally, since p(0) = 0, we can conclude that p(z) = f (g(z)) = z for all z ∈ A. 1 1+g(z) is well defined in the algebra. Although this existence theorem is useful for us, there is more to say about the specifics of the possible branches of the logarithm over an algebra, but first we must give some new terminology for algebras: Definition 2.5. Let A be a commutative algebra. If there exists another algebra B such that A ∼= C ⊗ B, then we say A is a type-C algebra. If A is not isomorphic to the direct product of an algebra B and some type-C algebra C, then we say that A is a type-R algebra. Essentially, a type-C algebra is a algebra which can be viewed as an algebra with C as its base field. A type-R algebra then, is an "honest to goodness" algebra over R in the sense that not even part of the algebra can be viewed as being built over C. Our main results in this section revolve around trying to describe the properties of the logarithm in an algebra based on whether it is a type-R, or a type-C algebra. 7 Notice however, that given Definition 2.5, not every algebra is either type-R or ∼= R× C, which has both both a "real piece" type-C. For a basic example, consider H3 and a "complex piece", and thus is neither type-R nor type-C. It follows trivially from the definition and by the finite-dimensional nature of our algebras that this is always the case. In order to prove that the logarithms over an algebra behave as expected over type-R and type-C algebra, we will first recall some preliminary propositions and definitions from ring theory. Definition 2.6. A local (commutative) ring is a ring with exactly one maximal ideal. We say that an algebra is a local is it is local as a ring. Proposition 2.7. Any finite dimensional algebra is the finite product of finite di- mensional local algebras. With which we prove the following classification theorem: Theorem 2.8. Any algebra is the product of a finite number of copies of R, C, and a finite number of different untial nil algebras. Proof. By the preceding proposition, every finite dimensional algebra is the finite product of local rings. Also, if a ring is local, its unique maximal ideal is identical to its Jacobson radical, which in our context is the same thing as Nil(A), and hence, by Proposition 1.11, the local finite dimensional algebras are exactly the same as the unital nil algebras2. Now that we have this result, we are ready to begin our investigation of logarithms over semisimple algebras, which is rather simple with the help of Wedderburn's clas- sification theorem. Lemma 2.9. Let A = A1 × · · · × Ak, and ez : A → A defined as usual, where z = e1x1 + · · · + ekxk, and x1, . . . , xk are elements of the algebras A1, . . . , Ak respectively, then ez = e1ex1 + e2ex2 + · · · + ekexk. Proof. Applying the definition of termwise multiplication in the direct product alge- bra A1 × · · · × Ak we obtain: ez = ee1x1+e2x2+···+ekxk = 1 + (e1x1 + e2x2 + · · · + ekxk) + = 1 + (e1x1 + e2x2 + · · · + ekxk) + (e1x1 + e2x2 + · · · + ekxk)2 (e1x2 1 + e2x2 2! 2 + · · · + ekx2 k) 2! + . . . + . . . 2While we tend to think of unital nil algebras as algebras such as Γ with non-trivial nil radicals, R and C technically also satisfy our definition of unital nil algebra. 8 And hence, since in the direct product algebra 1 = e1 + e2 + · · · + ek, collecting the power series component-wise yields: ez = e1(cid:18)1 + x1 + x2 1 2! + . . .(cid:19) + · · · + ek(cid:18)1 + xk + x2 k 2! + . . .(cid:19) = e1ex1 + · · · + ekexk Thus completing the proof. Theorem 2.10. Let A = A1 × · · · × Ak, then if log1(x1), log2(x2), . . . , logk(xk) de- note branches of the inverses to the exponential functions in the respective alge- bras A1, . . . , Ak where B1, . . . , Bk are the images of said branches for which log1(x1), . . . , logk(xk) are inverses to their respective exponential functions, then the function log(z) = (log1, log2, . . . , logk) : Ld(A1) × Ld(A2) × · · · × Ld(Ak) → B1 × B2 × · · · × Bk is the inverse function of ez : A → A on the branch B1 × B2 × · · · × Bk. Proof. By Lemma 2.9, if z = e1x1 + . . . ekxk, and x1, . . . , xk are elements of the algebras A1, . . . , Ak respectively, then ez = e1ex1 + e2ex2 + · · · + ekexk , and hence by definition (log ◦ exp)(z) = e1 log1(ex1) + · · · + ek logk(exk) = z for all z ∈ B, and (exp ◦ log)(z) = e1elog1(x1) + · · · + ekelogk(xk) = z for all z ∈ Ld(A1) × Ld(A2) × · · · × Ld(Ak). Corollary 2.11. Given a choice of branch cut for each of the complex components, the logarithm for a semisimple algebra A may be induced from the product of func- tions (log, . . . , log, Log1, . . . , Logm) : (R+)n × (C×)m → Rn × Cm under the isomor- phism φ : A → Rn × Cm given by Wedderburn's theorem, where Logi : C× → C denote the chosen branch cut of the complex logarithm.3 This theorem not only gives us a simple way to construct the logarithm for a commutative semisimple algebra from the basic real and complex logarithms we know and love without using series methods, but more importantly allows us to read off the possible domains of the logarithms for an algebra after passing the branches through the isomorphism map to Rn × Cm. 2.2 Nil Exponentials Our goal in this section is to show that the exponential function in a unital nil algebra is always injective. We will then prove that logarithms behave as expected over type- R and type-C algebras, using Lemma 2.9 to extend our result for semisimple algebras to arbitrary algebras once we understand how to deal with the nilpotent pieces. 3Note that usually a branch cut of the complex logarithm is taken with a domain of Cα – that is, C with a ray starting from 0 removed. This is done to ensure that the logarithm is continuous, but sacrifices having an inverse on all of Ld(C) = C×, however this corollary can be modified to better suit either approach to branch cuts of the complex logarithm. 9 Consider the following "generalized divisibility" ordering : Definition 2.12. Let β = {1, v 2, . . . , vn} be a unital basis for A. Set vi (cid:22) vj if and only if there there exists a vk ∈ β such that ck ij denotes the structure constants for A with respect to the basis β. ij 6= 0, where ck Proposition 2.13. (cid:22) is a partial order. Proof. To show reflexivity, since β is a unital basis, for all vi ∈ β we simply take 1 = v1 ∈ β so that v ⋆ v1 = v. In other words, in terms of structure coefficients, ci 1i = 1 is non-zero, hence vi ≺ vi for all i = 1, . . . , n. Now, suppose that vi ≺ vj and vj ≺ vk. By definition, there exists vα ∈ β such jγ 6= 0. Also, by definition of the structure iα 6= 0 and vγ ∈ β such that ck that cj constants we have vi ⋆ (vα ⋆ vγ) = ( v Xk=1 ck iαvk) ⋆ vγ = n Xk=1 ck iαvkvγ = n Xi=1 ck iα( n Xℓ=1 cℓ kγvℓ) = n n Xi=1 Xℓ=1 ck iαcℓ kγvℓ Hence, either ck cannot be the case by the assumption that cj transitive. iρ is non-zero for some ρ or the vi ⋆ (vα ⋆ vγ) must be zero, but this jγ are non-zero. Thus, (cid:22) is iα and ck Using this ordering, we will show that products of power series of indeterminates times one of the nilpotent basis elements v2, . . . , vn are of a very particular form. Lemma 2.14. Given an algebra A with basis β = {1, v2, . . . , vn}, we will consider the indeterminates z1, z2, . . . , zn, and say that zi is the indeterminate associated with the basis element vi for each i, and let p be a polynomial in A with indeterminates z1, . . . , zn. Let P (p) be the proposition that for all i = 1, 2, . . . , n, the coefficient of vi is a polynomial in indeterminates zj such that vj (cid:22) vi. Then we have the following: 1. P holds for all polynomials of the form zivi, which we will call "atomic" poly- nomials. 2. P (p) and P (q) implies P (p + q). 3. If c ∈ R, then P (p) implies P (cp). 4. P (p) and P (q) implies p(p ⋆ q). Proof. Part one of the lemma holds trivially, so we first consider part two. Suppose that F = f1v1 + f2v2 + · · · + fnvn = F and G = g1v1 + g2v2 + · · · + gnvn both satisfy P , where the fi, gj are polynomials satisfying the conditions in the lemma, then F + G = (f1 + g1)v1 + (f2 + g2)v2 + · · · + (gn + fn)vn 10 clearly also satisfies P . Similarly, cF = (cf1)v1 + . . . (cfn)vn also satisfies P . Finally, consider: (fivi)(gjvj) = n Xi,j=1 figj(vivj) n F G = = n Xi,j=1 Xi,j=1 n figj( n Xk=1 ck ijvk) = (ck ijfigj)vk Xi,j,k=1 Then, since by definition ck ij is non-zero if and only if vi (cid:22) vk and vj (cid:22) vk, F G satisfies P . Corollary 2.15. Let A be a unital nil algebra with nil basis β, and let v, w ∈ β and v, w 6= 1. Furthermore, suppose that x, y are two indeterminates associated with v and w respectively, then if p, q are power series in xv and yw respectively, then p ⋆ q is a polynomial satisfying P from Lemma 2.14. Proof. If v, w 6= 1, then by the definition of a unital nil algebra they must be nilpo- tent. Thus, the series p, q both truncate to polynomials. Furthermore, since these polynomials are the linear combination of atomic polynomials, by Lemma 2.14 P holds for both p and q, and hence the product p ⋆ q again by Lemma 2.14 . Unfortunately, there does not seem to be a simple proof that the ordering (cid:22) is always antisymmetric (and thus a poset), since this is necessary for our technique in the following theorem. Thus, until we find a method to prove that (cid:22) is in fact always antisymmetric, we will restrict our results to the class of multiplicative nil algebras. Definition 2.16. Let A be an algebra. We say a basis β = {v1, . . . , vn} of A is multiplicative if for all vi, vj ∈ β we have vi ⋆ vj = cvk for some c ∈ R and vk ∈ β. If an algebra A admits a multiplicative basis, then we say A is a multiplicative algebra. It is easy to show from the definition and the properties of unital nil bases that for such algebras, ≺ is in fact a poset4. Finally, in this context we are able to prove that the exponential function in a unital nil algebra is injective: Theorem 2.17. In any multiplicative commutative unital nil algebra A, the expo- nential function is injective. Proof. After choosing a basis, we identify exp : A → A as a function exp : Rn → Rn, then let (y1, y2, . . . , yn) = f (x1, . . . xn) be an element of the image of exp. We will show that exp is injective by giving an algorithm to construct the unique inverse 4This is what I call, in [2], the nil poset for an algebra with respect to a basis. 11 map exp−1 : Ld(A) → Rn. We will accomplish this by first constructing functions f1, . . . fn such that fi(f (x1, x2, . . . , xn)) = xi, then piecing these together, defining exp−1 = (f1, . . . , fn) so that: exp−1(exp(x1, x2, . . . , xn)) = (f1(x1, . . . , xn), . . . , fn(x1, . . . , xn)) = (x1, x2, . . . , xn) and hence that exp−1 is in fact the inverse function to exp. Recall from Proposition 2.3 that the standard rules of exponents still hold in an algebra, and hence ex1+x2v2+···+xnvn = ex1ex2v2 . . . exnvn. Also note that the co- efficient of the basis element 1 is 1 in the product ex2v2 . . . exnvn (we will denote this by coeff(1) = 1 from now on for brevity), since each factor has the form exivi = 1 + xivi + 1 2(xivi)2 + . . . , and thus the only possible product yielding some constant multiple of 1 is in fact simply 1 · 1 · 1 · · · · 1 = 1. Thus, coeff(1) = ex1 in the product ex1ex2v2 . . . exnvn, and hence in our identification with A as Rn, we may define f1(y1, . . . , yn) = log(y1), which gives us f1(exp(x1, . . . , xn)) = x1. We will now iterate a procedure on the set of minimal elements of the lattice NA − E, in which E is a subset of NA which we initially set E = {1}, in order to define the remaining component functions f2, . . . , fn of exp−1. For the initial step of this procedure, select a minimal element vi of NA − E = NA − {1}. Since the only element below vi in NA is 15, by Corollary 2.15, the coefficient of vi in the product ex2v2 . . . exnvn is a polynomial in the indeterminate xi. Also, notice that the product ex2v2 . . . exnvn when expanded contains the term xivi, and hence the coefficient of vi in this product has the form xi + p(xi) for some polynomial p. Furthermore, p(xi) must be zero, since the only possible product among powers of basis elements v1, v2, . . . , vn yielding a term with coefficient vi will be the given product of terms containing xivi and 1. Hence, we set fi(y1, y2, . . . , yn) := y2 y1 This initial procedure may then be iterated on all other vi minimal in NA, after which we set E := E + {vi1, . . . , vik}, where {vi1, vi2, . . . , vik} is the set of elements minimal in NA − {1}. . The preceding step was the "base case" of our algorithm. We now consider the general step of the algorithm, with E set to {1, vi1, . . . , vik} and fi1, . . . fik correctly defined as we discussed in the outset of the proof. As before, we now consider elements vi minimal in the poset NA − E, and again similarly we argue that the coefficient of vi in the product ex2v2 . . . exnvn must be of the form yi = xi + p(xj1, . . . , xjl), where p is some polynomials in the indeterminates xj1, . . . , xjl associated with the elements va such that va (cid:22) vi. Now, since every element below vi is in E, we have already constructed the functions fj1, fj2, . . . , fjl, so we set fi(y1, . . . , yn) := 1 y1 yi − p(fj1(y1, . . . , yn), fj2(y1, . . . , yn), . . . , fjl(y1, . . . , yn)) 5This is where the antisymmetry condition for (cid:22) is important, as otherwise this may not be true 12 so that fi(ex1+x2v2+···+xnvn) = ex1yi ex1 − p(fj1(ex1+x2v2+···+xnvn), . . . , fjl(ex1+x2v2+···+xnvn)) = yi − p(xj1, xj2, . . . , xjl) = xi + p(xj1, xj2, . . . , xjl) − p(xj1, xj2, . . . , xjl) = xi and finally, to complete the general iterative step, we set E := E ∪ {vi}. Given our setup, we may now summarize the complete algorithm as follows: Procedure. Begin by setting E = {1}, and iterate the procedure as described in paragraph 4 of the proof, constructing the functions fi1, . . . , fik1 and after completing the procedure, setting E := E ∪ {vi1, . . . , vik1}. Next, iterate the procedure described , and setting E := E ∪ {} starting in paragraph 6 of the proof, defining fik1+1, . . . , fik2 at the end of each iteration until CA −E = ∅, at which point the algorithm terminates, yielding the desired functions f1, . . . , fn. Since each step of the algorithm described above decreases the size of the finite set NA − E, the algorithm is clearly productive, and the algorithm halts whenever NA − E = ∅, this algorithm terminates giving us f1, f2, . . . , fn such that exp−1 = (f1, f2, . . . , fn) is the inverse of exp. Example 2.18. In Γ3: ez = ex+ǫy+ǫ2z = exeǫyeǫ2z = ex(cid:18)1 + ǫy + ǫ2y2 2 (cid:19)(cid:0)1 + ǫ2z(cid:1) = ex(cid:18)1 + ǫy + ǫ2(cid:18)z + y2 2 (cid:19)(cid:19) And hence, following the algorithm given in Theorem 2.17, or simply by inspection, we find: LogΓ3(x + ǫy + ǫ2z) = log(x) + ǫ y x + ǫ2(cid:18) z x − y2 2x2(cid:19) In addition to this result, it is easy to see from the structure of the component functions of the exponential as deduced in the proof of Lemma 2.9 that: Proposition 2.19. Given a commutative multiplicative unital nil algebra A with point set identified as Rn, Ld(A) = R+ × Rn−1. Similarly, if we consider the complex- ification C ⊗ A as identified with Cn, then Ld(C ⊗ A) = C× × Cn−1. Since the component functions of the inverse we constructed to the exponential are only undefined where log(x1) is undefined, which is on R− ∪ {0} for the real logarithm, and on {0} for the complex logarithm. 13 2.3 Final Results We now are able to completely characterize the possibilities for logarithms in an arbitrary multiplicative algebra6. Given the results of Theorem 2.17 we state the following result characterizing the logarithms of any multiplicative algebra, extending the results of Corollary 2.11: Proposition 2.20. Given a multiplicative algebra A, then let A ∼= A1 × A2 × · · · × Ak × Rn × Cm be the decomposition of A proven to exist in Theorem 2.8. If A is a type-R algebra, then there exists a unique inverse function to exp on Ld(A), denoted LogA(z). Otherwise, there exists infinitely many logarithms de- termined by the branches of the logarithms defined on the complex portion of the algebra. Thus, at least for the case where our algebra is multiplicative, for a type-R algebra, we see that the exponential function, akin to the real exponential is injective, and hence type-R algebras have unique logarithms defined on all of Ld(A). On the other hand, for a type-C algebra, again assuming our conjecture, we have the following result: Proposition 2.21. For a multiplicative type-C algebra A, Ld(A) = A× Proof. Since (A×B)× = A× ×B×, it suffices for us to prove this result on the possible factors of a type C algebra given in Theorem 2.8. Also, by complex analysis we already know that Ld(A) = C×, so it remains to show proposition 2.21 for complexified unital nil algebras. But for such an algebra A with unital nil basis {1, ǫ1, . . . , ǫn} so that we may identify the algebra with Cn we have Ld(A) = C× × Cn−1, in other words, in terms of the basis, Ld(A) = {a + b1ǫ1 + · · · + bnǫna, b1, . . . , bn ∈ R, a 6= 0} which is precisely the set of units in A.7 And hence, although we lose injectivity of the exponential function, we gain the fact that the logarithm is defined on almost all of the algebra. 6It is easy to see that the direct product factors of a multiplicative algebra must be again multi- plicative. 7This is a consequence of the well-known result from algebra that if a ∈ A× and ǫ, ξ ∈ Nil(A), then a + ǫ ∈ A× and ǫ + ξ ∈ Nil(A) 14 References [1] Alexander Abian. Linear Associtative Algebras. Pergamon Press, 1971. [2] Nathan BeDell. Doing algebra over an associative algebra. 2017. [3] James S. Cook. Introduction to A-calculus. 2012. [4] Daniel Freese and James S. Cook. Theory of series in the A-calculus and the n-pythagorean theorem. 2015. [5] Edgar R. Lorch. The theory of analytic functions in normed abelian vector rings. 1942. [6] Ignazio Lacirasella Luigia Di Terlizzi, Jerzy Julian Konderak. On differentable functions over lorentz numbers and their geometric applications. 2014. [7] G. Baley Price. An Introduction to Multicomplex Spaces and Functions. 1990. 15
1109.2469	1	1109	2011-09-12T13:43:54	Noncommutative identities	[ "math.RA" ]	This is a slightly edited version of my talk on Mathematische Arbeitstagung 2011, Bonn. I present a result relating noncommutative Laurent polynomials with algebraic functions, and show examples of integrability and Laurent phenomenon for free noncommutative variables.	math.RA	math	Noncommutative identities Maxim Kontsevich April 28, 2013 to Don Zagier, on the occasion of his 3·4·5 birthday, with love and admiration 1 “Characteristic polynomial” Let us fix an integer n ≥ 1 and consider the algebra 1 , . . . , X ±1 A := ChX ±1 n i of noncommutative Laurent polynomials with coefficients in C in n invertible variables, i.e. the group ring of the free group Freen on n generators: A = C[Freen ] = (cid:8) Xg∈Freen cg · g cg ∈ C, cg = 0 for almost all g ∈ Freen (cid:9) . Define a linear functional “Tr” on A by taking the constant term, “‘Tr” : A → C , Xg This functional vanishes on commutators, like the trace for matrices. By the analogy with the matrix case, we define the “characteristic polynomial”1 for any a ∈ A as a formal power series in one (central) variable t: tk = 1 + · · · ∈ C[[t]] . Pa = Pa (t) := “det”(1 − ta) := exp − Xk≥1 Theorem 1. For any a ∈ A the series Pa is algebraic, i.e. Pa ∈ C(t) ∩ C[[t]] ⊂ C((t)) . “Tr”(ak ) k cg · g 7→ cid . 1This is an algebraic analogue of so-called Fuglede-Kadison determinant, see [9]. 1 Here are few examples: • the case n = 1 is elementary, follows easily from the residue formula, • for any n ≥ 1 and 1 + · · · + Xn + X −1 a = X1 + X −1 n Pa = one can show that (cid:16) f +1 2 (cid:17)n 2n−1 (cid:17)n−1 , f = f (t) := p1 − 4(2n − 1)t2 = 1 + · · · ∈ Z[[t]] . (cid:16) nf +n−1 • if a = X1 + · · · + Xn + (X1 . . . Xn )−1 then the series Pa is an algebraic hypergeometric function. A sketch of the proof: Let us assume for simplicity that a ∈ Z[Freen ], the general case is just slightly more complicated. Step 1. For a = Pg cg · g ∈ Z[Freen ] the series Pa also has integer coefficients. Indeed, it is easy to see that Pa = Yk≥1 Y(g1 ,...,gk ) (1 − cg1 . . . cgk tk ) where for any k ≥ 1 we take the product over sequences of elements of Freen such that g1 . . . gk = id and the sequence (g1 , . . . , gk ) is strictly smaller than all its cyclic permutations for the lexicographic order on sequences associated with some total ordering of Freen considered as a countable set. Similar argument works if we replace Freen by an arbitrary torsion-free group. Step 2. Consider the following series Fa = Fa (t) := Xk≥1 Then Fa is algebraic. This fact was rediscovered many times, see e.g. [2]. It follows from the theory of noncommutative algebraic series developed by N. Chomsky and M.-P. Schutzenberger in 1963 (see [4]). We recommend to the reader to consult chapter 6 in [10], the algebraicity of the series Fa is the statement of Corollary 6.7.2 in this book. Step 3. Recall the Grothendieck conjecture on algebraicity. It says that for any algebraic vector bundle with flat connection over an algebraic variety “Tr”(ak ) tk ∈ Z[[t]]. 2 defined over a number field, all solutions of the corresponding holonomic system of differential equations are algebraic if and only if the p-curvature vanishes for all sufficiently large primes p ≫ 1. There is a simple sufficient criterion for such a vanishing. Namely, it is enough to assume that there exists a fundamental system of solutions in formal power series at some algebraic point, such that all Taylor coefficients (in some local algebraic coordinate system) of all solutions have in total only finitely many primes in denominators. The Grothendieck conjecture in its full generality is largely inaccessible by now. The only two general results are a theorem by N. Katz on the validity of the Grothendieck conjecture for Gauss-Manin connections, and a theorem of D. Chudnovsky and G. Chudnovsky [5] in the case of line bundles over algebraic curves (see also Exercise 5 to §4 of Chapter VIII, page 160 in [1]). This is exactly our case, by the previous step, and because d dt Pa = − Fa t Pa . 2 Noncommutative integrability, the case of two variables Let now consider the algebra of noncommutative polynomials in two (non- invertible) variables A = ChX, Y i . For any integer d ≥ 1 we consider the variety Md of GLd (C)-equivalence classes (by conjugation) of d-dimensional representations ρ : A → Matd×d (C) of A. More precisely, we are interested only in generic pairs of matrices (ρ(X ), ρ(Y )) and treat variety Md birationally. It has dimension d2 + 1. For generic ρ we consider the “bi-characteristic polynomial” in two com- mutative variables Pρ = Pρ (x, y) := det(1 − xρ(X ) − yρ(Y )) = 1 + · · · ∈ C[x, y ] . The equation Pρ (x, y) = 0 of degree ≤ d defines so-called Vinnikov curve Cρ ⊂ CP 2 . The number of parameters for the polynomial Pρ is (d+1)(d+2) −1, 2 and it is strictly smaller than dim Md for d ≥ 3. The missing parameters correspond to the natural line bundle Lρ on Cρ (well-defined for generic ρ) given by the kernel of operator (1−xρ(X )−yρ(Y )) for (x, y) ∈ Cρ . Bundle Lρ 3 has the same degree as a square root of the canonical class of Cρ , and defines a point in a torsor over the Jacobian Jac(Cρ ). For any given generic curve C ⊂ CP 2 of degree d line bundles on C depend on genus(C ) = (d−1)(d−2) 2 parameters. Now the dimensions match: dim Md = d2 + 1 = (d + 1)(d + 2) 2 − 1 + (d − 1)(d − 2) 2 . The conclusion is that Md is fibered over the space of planar curves of degree d, with the generic fiber being a torsor over an abelian variety. Hence we have one of simplest examples of an integrable system. Any integrable system has a commutative group of discrete symmetries, i.e. birational automorphisms preserving the structure of the fibration, identical on the base, and acting by shifts on the generic fiber. Similarly, one can consider an abelian Lie algebra consisting of rational vertical vector fields which are infinitesimal generators of shifts on fibers. Now I want to consider noncommutative symmetries, i.e. certain univer- sal expressions in free variables (X, Y ) which can be specialized and make sense for any d ≥ 1. An universal discrete symmetry is an automorphism of A (or maybe of some completion of A) which preserves the conjugacy class of any linear combination Z (t) := X + tY , t ∈ C. Indeed, in this case for any d ≥ 1 and any representation, the value of the bi-characteristic polynomial Pρ at any pair of complex numbers (x, y) ∈ C2 is preserved, as it can be written as det(1 − xρ(Z (y/x))). Hence the automorphism under consideration is inner on both variables X and Y : X 7→ R · X · R−1 , Y 7→ R′ · Y · (R′ )−1 . We are interested in automorphisms of A only up to inner automorphisms, therefore we may safely assume that R′ = 1. Thus, the question is reducing to the following one: find R such that for any t ∈ C there exists Rt such that R · X · R−1 + tY = Rt · (X + tY ) · R−1 t First, let us make calculations on the Lie level. Denote by g the Lie algebra of derivations δ of A of the form δ(X ) = [D , X ] for some X ∈ A, δ(Y ) = 0 and such that for any t ∈ C there exists Dt ∈ A such that δ(X + tY ) = [Dt , X + tY ] ⇐⇒ [D , X ] = [Dt , X + tY ] . 4 It is easy to classify such derivations, and one can check that the following elements form a linear basis of g: δn,m (X ) = [cn,m , X ], δn,m (Y ) = 0, n ≥ 0, m ≥ 1 w , C · δn,m . cn−k ,m+k tk . where for any n, m ≥ 0 we define cn,m := X(n+m)! shuffles w n!m! i.e. the sum of all words in X, Y containing n letters X and m letters Y . Elements Dt ∈ A corresponding to the derivation δn,m are given by Dt = X0≤k≤n A direct calculation shows that g is an abelian Lie algebra. Let us go now the completions of algebra A, and of Lie algebra g: bA := ChhX, Y ii, bg := Yn≥0,m≥1 Then the action of bg on bA exponentiates a continuous group action exp ≃ bG ⊂ Aut( bA) . bg For any δ ∈ bg the corresponding one-parameter group of automorphisms acts by exp(τ · δ) : X 7→ R(τ ) · X · R(τ )−1 , Y 7→ Y ∀ τ ∈ C for certain invertible element R(τ ) ∈ bA× . An easy calculation shows that R(τ ) is the unique solution of the differential equation d R(τ ) = δ(R(τ )) + R(τ ) · D , R(0) = 1 dτ where D ∈ bA is such that δ(X ) = [D , X ]. The value R(τ )τ =1 gives exp(δ). Now we can start to look for a class of elements δ ∈ bg such that the corresponding automorphism exp(δ) is sufficiently nice, e.g. if it makes some sense for A without passing to the completion. 5 fn,mxnym ∈ C[[x, y ]] . Let us encode a generic element δ as before by the corresponding gener- ating series in commutative variables x, y : fn,mδn,m ∈ bg eδ := Xn,m δ = Xn,m I suggest the following Ansatz: eδ is the logarithm of a rational function in x, y . Hypothetically, for such δ the corresponding automorphism exp(δ) of bA can be extended to certain “algebraic extension” of A. A good indication is Theorem 2. For any P = P (x, y) = 1 + · · · ∈ C[x, y ] expand log(P ) = Xn,m exp Xn,m (n + m)! n!m! This result is elementary, and I leave it as an exercise to the reader. (Hint: use the residue formula twice.) It implies that the image of R under the abelianization morphism fn,mxnym! fn,mxnym ∈ C[[x, y ]] . Then the series is algebraic. ChhX, Y ii ։ C[[x, y ]], X 7→ x, Y 7→ y is algebraic. Example. Consider the case eδ = log(1 − xy) = − Xk≥1 Then one can show that R = 1 − Y X − C ∈ bA× where C is the unique solution of the equation C = X · (1 − C )−1 · Y . xk yk k . 6 It can be written C = X Y +XX Y Y +XX Y X Y Y +XXX Y Y Y +· · · = ()+(())+(()())+((()))+. . . as the sum of all irreducible bracketings if we replace X by ( and Y by ). The equation for C is equivalent to the generic “quadratic equation” by the substitutions T 2 + AT + B = 0 A := X −1 , B := −X −1Y , T := X −1C . The invertible elements Rt ∈ bA× , t ∈ C are given by Rt := R · (1 − tT 2 ), R · X · R−1 + tY = Rt · (X + tY ) · R−1 t I’ll finish with another example of an integrable system. Few years ago together with S. Duzhin we discovered numerically that the rational map . S−1 : (X, Y ) 7→ (X Y X −1 , (1 + Y −1 ) X −1 ) should be a discrete symmetry of an integrable system, where X, Y are two d × d matrices for d ≥ 1. Recently O. Efimovskaya and Th. Wolf found an explanation (see [7]). Namely, their results suggest that the conjugacy class of the Lax operator which is the matrix L(t) of size 2d × 2d, defined as t (cid:19) L(t) := (cid:18) Y −1 + X tY + Y −1X −1 + X −1 + 1 t X Y + Y −1X −1 + X −1 + 1 Y −1 + 1 does not change under the discrete symmetry S−1 as above, for any t ∈ C. Indeed, one can check directly that S−1 (L(t)) = V (t)L(t)V (t)−1 where2 V (t) := (cid:18)X (1 + X + Y )−1X (1 + Y −1 )−1 X (1 + Y −1 )−1X (1 + X + Y )−1 X Y (1 + X + Y )−1 (cid:19) . tX (1 + X + Y )−1Y 3 Noncommutative integrability for many variables Let M = (Mij )1≤i,j≤3 be a matrix whose entries are 9 = 3 × 3 free indepen- dent noncommutative variables. Let us consider 3 “birational involutions” I1 : M 7→ M −1 I2 : M 7→ M t I3 : Mij 7→ (Mij )−1 ∀i, j . 2 I am grateful to A.Odesskii for help in finding the matrix V (t). 7 The composition I1 ◦ I2 ◦ I3 commutes with the multiplication on the left and on the right by diagonal 3 × 3 matrices. We can factorize it by the action of Diag left × Diagright and get only 4 independent variables, setting e.g. Mij = 1 for i = 3 and/or j = 3. Conjecture 1. The transformation (I1 ◦ I2 ◦ I3 )3 is equal to the identity modulo Diag left × Diagright -action. In other words, there exists two diagonal 3 × 3 matrices DL (M ), DR (M ) whose entries are noncommutative rational functions in 9 variables (Mij ), such that (I1 ◦ I2 ◦ I3 )3 (M ) = DL (M ) · M · DR (M ) . This is a very degenerate case of integrability. Similarly, for 4×4 matrices the transformation I1 ◦ I2 ◦ I3 should give a genuinely nontrivial integrable system. In the simplest case when the entries of this matrix are scalars, the Zariski closure of the generic orbit (modulo the left and the right diagonal actions) is a union of two elliptic curves. Finally, I’ll present a series of hypothetical discrete symmetries of inte- grable systems written as recursions. Fix an odd integer k ≥ 3 and consider sequences (Un )n∈Z (of, say, d × d matrices), satisfying Un = U −1 n−k (1 + Un−1 Un−k+1 ) Un = (1 + Un−k+1 Un−1 ) U −1 n−k for n ∈ 2 Z for n ∈ 2 Z + 1 . Then the map (U1 , . . . , Uk ) 7→ (U3 , . . . , Uk+2 ) is integrable. 4 Noncommutative Laurent phenomenon In the previous example one observes also the noncommutative Laurent phenomenon: Un ∈ ZhU ±1 1 , . . . , U ±1 k i . Also with S. Duzhin we discovered that the noncommutative birational map ∀ n ∈ Z S l : (X, Y ) 7→ (X Y X −1 , (1 + Y l ) X −1 ) for l ≥ 1 satisfies the same Laurent properties, i.e. both components of 2-dimensional vector obtained by an arbitrary number of iterations, belong to the ring ZhX ±1 , Y ±1 i. The case l = 1 is easy, and the case l = 2 was studied by A. Usnich (unpublished) and by Ph. Di Francesco and R. Kedem, see [6]. The Laurent property has now three different proofs for the case l ≥ 3 when the dynamics is non-integrable: 8 • by A. Usnich using triangulated categories, see [11], • an elementary algebraic proof by A. Berenstein and V. Retakh, see [3], • a new combinatorial proof of Kyungyong Lee, which also shows that all the coefficients of noncommutative Laurent polynomials obtained by iterations, belong to {0, 1} ⊂ Z, see [8]. Finally, recently A. Berenstein and V. Retakh found a large class of noncommutative mutations related with triangulated surfaces, and proved the noncommutative Laurent property for them. References [1] Y. Andr´e. G-functions and Geomerty, Aspects of Mathematics, vol. E13, Friedr. Vieweg & Sohn, Braunschweig, 1989. [2] J. Bellisard, S. Garoufalidis, Algebraic G-functions associated to ma- trices over a group-ring, arXiv:0708.4234 . [3] A. Berenstein, V. Retakh, A short proof of Kontsevich cluster conjec- ture, arXiv:1011.0245 . [4] N. Chomsky, M.-P. Schutzenberger, The Algebraic Theory of Context- Free Languages, in Computer Programming and Formal Systems, P. Braffort and D. Hirschberg (eds.), North Holland, 1963, pp. 118-161. [5] D. V. Chudnovsky, G. V. Chudnovsky, Applications of Pad´e approx- imations to diophantine inequalities in values of G-functions, Lect. Notes in Math. 1052, Berlin, Heidelberg, New York, Springer-Verlag (1985), 1-51. [6] Ph. Di Francesco, R. Kedem, Discrete non-commutative integrability: the proof of a conjecture by M. Kontsevich, arXiv:0909.0615 . [7] O. Efimovskaya, Th. Wolf, On integrability of the Kontsevich non- abelian ODE system, arXiv:1108.4208 . [8] Kyungyong Lee, A step towards the cluster positivity conjecture, arXiv:1103.2726 . [9] Wolgang Luck, L2 -Invariants: Theory and Applications to Geometry and K -Theory, Ergebnisse der Mathematik und ihrer Grenzgebiete. 3. 9 Folge. A Series of Modern Surveys in Mathematics, vol. 44, Springer- Verlag, Berlin, 2002. [10] R. Stanley, Enumerative combinatorics, volume 2, second edition, Cambridge University Press, 1997. [11] A. Usnich, Non-commutative Laurent phenomenon for two variables, arXiv:1006.1211 . IHES, 35 route de Chartres, F-91440, France [email protected] 10
1303.0859	1	1303	2013-03-04T21:16:21	New criteria for a ring to have a semisimple left quotient ring	[ "math.RA" ]	Goldie's Theorem (1960), which is one of the most important results in Ring Theory, is a criterion for a ring to have a semisimple left quotient ring. The aim of the paper is to give four new criteria (using a completely different approach and new ideas). The first one is based on the recent fact that for an arbitrary ring $R$ the set $\CM$ of maximal left denominator sets of $R$ is a non-empty set: Theorem (The First Criterion). A ring $R$ has a semisimple left quotient ring $Q$ iff $\CM $ is a finite set, $\bigcap_{S\in \CM} \ass (S) =0$ and, for each $S\in \CM$, the ring $S^{-1}R$ is a simple left Artinian ring. In this case, $Q\simeq \prod_{S\in \CM} S^{-1}R$. The Second Criterion is given via the minimal primes of $R$ and goes further then the First one %and Goldie's Theorem in the sense that it describes explicitly the maximal left denominator sets $S$ via the minimal primes of $R$. The Third Criterion is close to Goldie's Criterion but it is easier to check in applications (basically, it reduces Goldie's Theorem to the prime case). The Fourth Criterion is given via certain left denominator sets.	math.RA	math	New criteria for a ring to have a semisimple left quotient ring V. V. Bavula Abstract Goldie's Theorem (1960), which is one of the most important results in Ring Theory, is a criterion for a ring to have a semisimple left quotient ring. The aim of the paper is to give four new criteria (using a completely different approach and new ideas). The first one is based on the recent fact that for an arbitrary ring R the set M of maximal left denominator sets of R is a non-empty set [2]: Theorem (The First Criterion). A ring R has a semisimple left quotient ring Q iff M is a finite set, TS∈M ass(S) = 0 and, for each S ∈ M, the ring S−1R is a simple left Artinian ring. In this case, Q ≃ QS∈M S−1R. The Second Criterion is given via the minimal primes of R and goes further then the First one in the sense that it describes explicitly the maximal left denominator sets S via the minimal primes of R. The Third Criterion is close to Goldie's Criterion but it is easier to check in applications (basically, it reduces Goldie's Theorem to the prime case). The Fourth Criterion is given via certain left denominator sets. Key Words: Goldie's Theorem, a left Artinian ring, the left quotient ring of a ring, the largest left quotient ring of a ring, a maximal left denominator set, the left localization radical of a ring, a maximal left localization of a ring, a left localization maximal ring. Mathematics subject classification 2010: 15P50, , 16P60, 16P20, 16U20. 1. Introduction. 2. Preliminaries. Contents 3. The First Criterion (via the maximal left denominator sets). 4. The Second Criterion (via the minimal primes). 5. The Third Criterion (in the spirit of Goldie-Lesieur-Croisot). 6. The Fourth Criterion (via certain left denominator sets). 7. Left denominator sets of finite direct products of rings. 8. Criterion for R/lR to have a semisimple left quotient ring. 1 Introduction In this paper, module means a left module, and the following notation is fixed: • R is a ring with 1, n = nR is its prime radical and Min(R) is the set of minimal primes of R; • C = CR is the set of regular elements of the ring R (i.e. C is the set of non-zero-divisors of the ring R); • Q = Ql,cl(R) := C−1R is the left quotient ring (the classical left ring of fractions) of the ring R (if it exists) and Q∗ is the group of units of Q; • Denl(R, a) is the set of left denominator sets S of R with ass(S) = a where a is an ideal of R and ass(S) := {r ∈ R sr = 0 for some s ∈ S}, 1 • max.Denl(R) is the set of maximal left denominator sets of R (it is always a non-empty set, [2]). Four new criteria for a ring to have a semisimple left quotient ring. Goldie's Theorem [9] characterizes left orders in semisimple rings, it is a criterion for a ring to have a semisimple left quotient ring (earlier, characterizations were given, by Goldie [8] and Lesieur and Croisot [12], of left orders in a simple Artinian ring). Theorem 1.1 (Goldie's Theorem, [9]) A ring has a semisimple left quotient ring iff it is a semiprime ring that satisfies the ascending chain condition on left annihilators and does not contain infinite direct sums of nonzero left ideals. In [2] and [1], several new concepts are introduced (and studied): the largest left quotient ring of a ring, the largest regular left Ore set of a ring, a maximal left denominator set of a ring, the left localization radical of a ring, a left localization maximal ring (see Section 2 for details). Their universal nature naturally leads to the present criteria for a ring to have a semisimple left quotient ring. In the paper, several new concepts are introduced that are used in proofs: the core of an Ore set, the sets of left localizable and left non-localizable elements, the set of completely left localizable elements of a ring. The First Criterion is given via the set M := max.Denl(R). • (Theorem 3.1, The First Criterion) A ring R has a semisimple left quotient ring Q iff M is a finite set, TS∈M ass(S) = 0 and, for each S ∈ M, the ring S−1R is a simple left Artinian ring. In this case, Q ≃QS∈M S−1R. The Second Criterion is given via the minimal primes of R and certain explicit multiplicative sets associated with them. On the one hand, the Second Criterion stands between Goldie's Theo- rem and the First Criterion in terms how it is formulated. On the other hand, it goes further then the First Criterion in the sense that it describes explicitly the maximal left denominator sets and the left quotient ring of a ring with a semisimple left quotient ring. • (Theorem 4.1, The Second Criterion) Let R be a ring. The following statements are equivalent. 1. The ring R has a semisimple left quotient ring Q. 2. (a) The ring R is a semiprime ring. (b) The set Min(R) of minimal primes of R is a finite set. (c) For each p ∈ Min(R), the set Sp := {c ∈ R c + p ∈ CR/p} is a left denominator set of the ring R with ass(Sp) = p. (d) For each p ∈ Min(R), the ring S−1 p R is a simple left Artinian ring. If one of the two equivalent conditions holds then max.Denl(R) = {Sp p ∈ Min(R)} and Q ≃Qp∈Min(R) S−1 p R. The Third Criterion (Theorem 5.1) can be seen as a 'weak' version of Goldie's Theorem in the sense that the conditions are 'weaker' than that of Goldie's Theorem. In applications, it could be 'easier' to verify whether a ring satisfies the conditions of Theorem 5.1 comparing with Goldie's Theorem as Theorem 5.1 'reduces' Goldie's Theorem essentially to the prime case and reveals the 'local' nature of Goldie's Theorem. • (Theorem 5.1, The Third Criterion) Let R be a ring. The following statements are equivalent. 1. The ring R has a semisimple left quotient ring Q. 2. The ring R is a semiprime ring with Min(R) < ∞ and, for each p ∈ Min(R), the ring R/p is a left Goldie ring. 2 The condition Min(R) < ∞ can be replaced by any of the four equivalent conditions of Theorem 5.2, e.g., 'the ring R has a.c.c. on annihilator ideals.' As far as applications are concerned, Theorem 5.1 has a useful corollary. • (Theorem 6.2, The Fourth Criterion) Let R be a ring. The following statements are equivalent. 1. The ring R has a semisimple left quotient ring Q. 2. There are left denominator sets S′ 1, . . . , S′ n of the ring R such that the rings Ri := S−1 i R, i = 1, . . . , n, are simple left Artinian rings and the map σ := n Yi=1 σi : R → n Yi=1 Ri, R 7→ ( r 1 , . . . , r 1 ), is an injection where σi : R → Ri, r 7→ r 1 . The paper is organized as follows. In Section 2, necessary concepts are introduced and results are collected that are used in the profs of Theorem 3.1, Theorem 4.1, Theorem 5.1 and Theorem 6.2. In Section 3, the proof of Theorem 3.1 is given. In Section 4, the proof of Theorem 4.1 is given. In Section 5, the proof of Theorem 5.1 is given. In Section 6, the proof of Theorem 6.2 is given. In Section 7, it is shown that the set of maximal left denominators of a finite direct product i=1 Ri of rings is a union of the sets of maximal left denominator sets of the rings Ri (Theorem Qn 2.9). In Section 8, a criterion (Theorem 8.1) is given for the factor ring R/l (where l is the left localization radical of R) to have a semisimple left quotient ring. The criterion is given in terms of the ring R rather than of R/l and is based on four criteria above, it is a long statement. Let us give a flavour. • (Theorem 8.1) The following statements are equivalent. 1. The ring R/l has a semisimple left quotient ring Q. 2. (a) max.Denl(R) < ∞. (b) For every S ∈ max.Denl(R), S−1R is a simple left Artinian ring. So, Theorem 8.1 characterizes precisely the class of rings that have only finitely many maximal left denominators sets and all the left localizations at them are simple left Artinian rings. The proofs of Theorem 3.1 and Theorem 4.1 are based on completely different ideas from existing proofs of Goldie's Theorem and Goldie's Theorem is not used. The key idea of the proof of Theorem 5.1 is to reduce it to Theorem 4.1 and the prime case of Goldie's Theorem. The key idea of the proof of Theorem 6.2 is to reduce it to Theorem 5.1. Old and new criteria for a ring to have a left Artinian left quotient ring. Goldie's Theorem gives an answer to the question: When a ring has a semisimple left quotient ring? The next natural question (that was posed in 50s) and which is a generalization of the previous one: Give a criterion for a ring to have a left Artinian left quotient ring. Small [15, 16], Robson [14], and latter Tachikawa [18] and Hajarnavis [11] have given different criteria for a ring to have a left Artinian left quotient ring. Recently, the author [3] has given three more criteria in the spirit of the present paper. We should mention contribution to the old criteria of Talintyre, Feller and Swokowski [6], [20]. Talintyre [19] and Feller and Swokowski [7] have given conditions which are sufficient for a left Noetherian ring to have a left quotient ring. Further, for a left Noetherian ring which has a left quotient ring, Talintyre [20] has established necessary and sufficient conditions for the left quotient ring to be left Artinian. In the proofs of all these criteria (old and new) Goldie's 3 Theorem is used. Each of the criteria comprises several conditions. The conditions in the criteria of Small, Robson and Hajarnavis are 'strong' and are given in terms of the ring R rather than of its factor ring R = R/n. On the contrary, the conditions of the criteria in [3] are 'weak' and given in terms of the ring R and a finite set of explicit R-modules (they are certain explicit subfactors of the prime radical n of the ring R). 2 Preliminaries In this section, we collect necessary results that are used in the proofs of this paper. More results on localizations of rings (and some of the missed standard definitions) the reader can find in [10], [17] and [13]. In this paper the following notation will remained fixed. Notation: • Orel(R) := {S S is a left Ore set in R}; • Denl(R) := {S S is a left denominator set in R}; • Locl(R) := {S−1R S ∈ Denl(R)}; • Assl(R) := {ass(S) S ∈ Denl(R)} where ass(S) := {r ∈ R sr = 0 for some s = s(r) ∈ S}; • Sa = Sa(R) = Sl,a(R) is the largest element of the poset (Denl(R, a), ⊆) and Qa(R) := a R is the largest left quotient ring associated to a, Sa exists (Theorem 2.1, [2]); Ql,a(R) := S−1 • In particular, S0 = S0(R) = Sl,0(R) is the largest element of the poset (Denl(R, 0), ⊆), i.e. 0 R is the largest left quotient ring of the largest regular left Ore set of R, and Ql(R) := S−1 R [2] ; • Locl(R, a) := {S−1R S ∈ Denl(R, a)}. In [2], we introduce the following new concepts and prove their existence for an arbitrary ring: the largest left quotient ring of a ring, the largest regular left Ore set of a ring, the left localization radical of a ring, a maximal left denominator set, a maximal left quotient ring of a ring, a (left) localization maximal ring. Using an analogy with rings, the counter parts of these concepts for rings would be a left maximal ideal, the Jacobson radical, a simple factor ring. These concepts turned out to be very useful in Localization Theory and Ring Theory. They allowed us to look at old/classical results from a new more general perspective and to give new equivalent statements to the classical results using a new language and a new approach as the present paper, [2], [1], [3], [4] and [5] and several other papers under preparation demonstrate. The largest regular left Ore set and the largest left quotient ring of a ring. Let R be a ring. A multiplicatively closed subset S of R or a multiplicative subset of R (i.e. a multiplicative sub-semigroup of (R, ·) such that 1 ∈ S and 0 6∈ S) is said to be a left Ore set if it satisfies the left Ore condition: for each r ∈ R and s ∈ S, SrT Rs 6= ∅. Let Orel(R) be the set of all left Ore sets of R. For S ∈ Orel(R), ass(S) := {r ∈ R sr = 0 for some s ∈ S} is an ideal of the ring R. A left Ore set S is called a left denominator set of the ring R if rs = 0 for some elements r ∈ R and s ∈ S implies tr = 0 for some element t ∈ S, i.e. r ∈ ass(S). Let Denl(R) be the set of all left denominator sets of R. For S ∈ Denl(R), let S−1R = {s−1r s ∈ S, r ∈ R} be the left localization of the ring R at S (the left quotient ring of R at S). Let us stress that in Ore's method of localization one can localize precisely at left denominator sets. In general, the set C of regular elements of a ring R is neither left nor right Ore set of the ring R and as a result neither left nor right classical quotient ring (Ql,cl(R) := C−1R and Qr,cl(R) := RC−1) exists. Remarkably, there exists the largest regular left Ore set S0 = Sl,0 = Sl,0(R), [2]. This means that the set Sl,0(R) is an Ore set of the ring R that consists of regular elements (i.e., Sl,0(R) ⊆ C) and contains all the left Ore sets in R that consist of regular elements. Also, there exists the largest regular (left and right) Ore set Sl,r,0(R) of the ring R. In general, all the sets 4 C, Sl,0(R), Sr,0(R) and Sl,r,0(R) are distinct, for example, when R = I1 = Khx, ∂,R i is the ring of polynomial integro-differential operators over a field K of characteristic zero, [1]. In [1], these four sets are found for R = I1. Definition, [1], [2]. The ring Ql(R) := Sl,0(R)−1R (respectively, Qr(R) := RSr,0(R)−1 and Q(R) := Sl,r,0(R)−1R ≃ RSl,r,0(R)−1) is called the largest left (respectively, right and two-sided) quotient ring of the ring R. In general, the rings Ql(R), Qr(R) and Q(R) are not isomorphic, for example, when R = I1, [1]. The next theorem gives various properties of the ring Ql(R). In particular, it describes its group of units. Theorem 2.1 [2] 1. S0(Ql(R)) = Ql(R)∗ and S0(Ql(R)) ∩ R = S0(R). 2. Ql(R)∗ = hS0(R), S0(R)−1i, i.e. the group of units of the ring Ql(R) is generated by the sets S0(R) and S0(R)−1 := {s−1 s ∈ S0(R)}. 3. Ql(R)∗ = {s−1t s, t ∈ S0(R)}. 4. Ql(Ql(R)) = Ql(R). The maximal denominator sets and the maximal left localizations of a ring. The set (Denl(R), ⊆) is a poset (partially ordered set). In [2], it is proved that the set max.Denl(R) of its maximal elements is a non-empty set. Definition, [2]. An element S of the set max.Denl(R) is called a maximal left denominator set of the ring R and the ring S−1R is called a maximal left quotient ring of the ring R or a maximal left localization ring of the ring R. The intersection lR := l.lrad(R) := \S∈max.Denl(R) ass(S) is called the left localization radical of the ring R, [2]. For a ring R, there is the canonical exact sequence 0 → lR → R σ→ YS∈max.Denl(R) S−1R, σ := YS∈max.Denl(R) σS, (1) (2) where σS : R → S−1R, r 7→ r 4.1 shows that the left localization radical lR coincides with the prime radical nR of R: 1 . For a ring R with a semisimple left quotient ring, Theorem lR = Tp∈Min(R) p = nR = 0. In general, this is not the case even for left Artinian rings [4]. Definition. The sets Ll(R) := [S∈max.Denl(R) S and N Ll(R) := R\Ll(R) are called the sets of left localizable and left non-localizable elements of R, respectively, and the intersection Cl(R) := \S∈max.Denl(R) S is called the set of completely left localizable elements of R. Clearly, R∗ ⊆ Cl(R). 5 The maximal elements of Assl(R). Let max.Assl(R) be the set of maximal elements of the poset (Assl(R), ⊆) and ass.max.Denl(R) := {ass(S) S ∈ max.Denl(R)}. (3) These two sets are equal (Proposition 2.4), a proof is based on Lemma 2.2 and Corollary 2.3. Recall that for an non-empty set X or R, l.ann(X) := {r ∈ R rX = 0} is the left annihilator of the set X, it is a left ideal of R. Lemma 2.2 [2] Let S ∈ Denl(R, a) and T ∈ Denl(R, b) be such that a ⊆ b. Let ST be the multiplicative semigroup generated by S and T in (R, ·). Then 1. l.ann(ST ) ⊆ b. 2. ST ∈ Denl(R, c) and b ⊆ c. Corollary 2.3 Let R be a ring, S ∈ max.Denl(R) and T ∈ Denl(R). Then 1. T ⊆ S iff ass(T ) ⊆ ass(S). 2. If, in addition, T ∈ max.Denl(R) then S = T iff ass(S) = ass(T ). Proof. 1. (⇒) If T ⊆ S then ass(T ) ⊆ ass(S). (⇐) If ass(T ) ⊆ ass(S). then, by Lemma 2.2, ST ∈ Denl(R) and S ⊆ ST , hence S = ST , by the maximality of S. Then T ⊆ S. 2. Statement 2 follows from statement 1. (cid:3) Proposition 2.4 [2] max.Assl(R) = ass.max.Denl(R) 6= ∅. In particular, the ideals of this set are incomparable (i.e. neither a * b nor a + b). Properties of the maximal left quotient rings of a ring. The next theorem describes various properties of the maximal left quotient rings of a ring. In particular, their groups of units and their largest left quotient rings. The key moment in the proof is to use Theorem 2.1. Theorem 2.5 [2] Let S ∈ max.Denl(R), A = S−1R, A∗ be the group of units of the ring A; a := ass(S), πa : R → R/a, a 7→ a + a, and σa : R → A, r 7→ r 1 . Then 1. S = Sa(R), S = π−1 a (S0(R/a)), πa(S) = S0(R/a) and A = S0(R/a)−1R/a = Ql(R/a). 2. S0(A) = A∗ and S0(A) ∩ (R/a) = S0(R/a). 3. S = σ−1 a (A∗). 4. A∗ = hπa(S), πa(S)−1i, i.e. the group of units of the ring A is generated by the sets πa(S) and π−1 a (S) := {πa(s)−1 s ∈ S}. 5. A∗ = {πa(s)−1πa(t) s, t ∈ S}. 6. Ql(A) = A and Assl(A) = {0}. In particular, if T ∈ Denl(A, 0) then T ⊆ A∗. The left localization maximal rings. These are precisely the rings in which we cannot invert anything on the left (in the sense of Ore). Definition, [2]. A ring A is called a left localization maximal ring if A = Ql(A) and Assl(A) = {0}. A ring A is called a right localization maximal ring if A = Qr(A) and Assr(A) = {0}. A ring A which is a left and right localization maximal ring is called a (left and right) localization maximal ring (i.e. Ql(A) = A = Qr(A) and Assl(A) = Assr(A) = {0}). 6 Example. Let A be a simple ring. Then Ql(A) is a left localization maximal ring and Qr(A) is a right localization maximal ring. In particular, a division ring is a (left and right) localization maximal ring. More generally, a simple left Artinian ring (i.e. the matrix ring over a division ring) is a (left and right) localization maximal ring. Let max.Locl(R) be the set of maximal elements of the poset (Locl(R), →) where A → B for A, B ∈ Locl(R) means that there exist S, T ∈ Denl(R) such that S ⊆ T , A ≃ S−1R and B ≃ T −1R (then there exists a natural ring homomorphism A → B, s−1r 7→ s−1r). Then (see [2]), max.Locl(R) = {S−1R S ∈ max.Denl(R)} = {Ql(R/a) a ∈ ass.max.Denl(R)}. (4) The next theorem is a criterion of when a left quotient ring of a ring is a maximal left quotient ring. Theorem 2.6 [2] Let a ring A be a left localization of a ring R, i.e. A ∈ Locl(R, a) for some a ∈ Assl(R). Then A ∈ max.Locl(R) iff Ql(A) = A and Assl(A) = {0}, i.e. A is a left localization maximal ring. Theorem 2.6 shows that the left localization maximal rings are precisely the localizations of all the rings at their maximal left denominators sets. The core of a left Ore set. The following definition is one of the key concepts that is used in the proof of the First Criterion (Theorem 3.1). Definition, [5]. Let R be a ring and S ∈ Orel(R). The core Sc of the left Ore set S is the set of all the elements s ∈ S such that ker(s·) = ass(S) where s· : R → R, r 7→ sr. Lemma 2.7 If S ∈ Denl(R) and Sc 6= ∅ then 1. SSc ⊆ Sc. 2. For any s ∈ S there exists an element t ∈ S such that ts ∈ Sc. Proof. 1. Trivial. 2. Statement 2 follows directly from the left Ore condition: fix an element sc ∈ Sc, then ts = rsc for some elements t ∈ S and r ∈ R. Since ass(S) ⊇ ker(ts·) = ker(rsc·) ⊇ ker(sc·) = ass(S), i.e. ker(ts·) = ass(S), we have ts ∈ Sc. (cid:3) Theorem 2.8 Suppose that S ∈ Denl(R, a) and Sc 6= ∅. Then 1. Sc ∈ Denl(R, a). 2. The map θ : S−1 c R → S−1R, s−1r 7→ s−1r, is a ring isomorphism. So, S−1 c R ≃ S−1R. Proof. 1. By Lemma 2.7.(1), ScSc ⊆ Sc, and so the set Sc is a multiplicative set. By Lemma 2.7.(2), Sc ∈ Orel(R): for any elements sc ∈ Sc and r ∈ R, there are elements s ∈ S and r′ ∈ R such that sr = r′sc (since S ∈ Orel(R)). By Lemma 2.7.(2), s′ c := ts ∈ Sc for some t ∈ S. Then s′ cr = tr′sc. If rsc = 0 for some elements r ∈ R and sc ∈ Sc then sr = 0 for some element s ∈ S (since cr = 0. Therefore, c := ts ∈ Sc for some element t ∈ S, hence s′ S ∈ Orel(R)). By Lemma 2.7.(2), s′ Sc ∈ Denl(R, a). 2. By statement 1 and the universal property of left Ore localization, the map θ is a well-defined monomorphism. By Lemma 2.7.(2), θ is also a surjection: let s−1r ∈ S−1R, and sc := ts ∈ Sc for some element t ∈ S (Lemma 2.7.(2)). Then s−1r = s−1t−1tr = (ts)−1tr = s−1 c tr. (cid:3) 7 The core of every maximal left denominator set of a semiprime left Goldie ring is a non-empty set and Theorem 3.1.(6) gives its explicit description (via the minimal primes). The maximal left quotient rings of a finite direct product of rings. i=1 Ri be a direct product of rings Ri. Then for each i = 1, . . . , n, the map Theorem 2.9 Let R = Qn is an injection. Moreover, max.Denl(R) =`n max.Denl(Ri) → max.Denl(R), Si 7→ R1 × · · · × Si × · · · × Rn, (5) i=1 max.Denl(Ri) in the sense of (5), i.e. max.Denl(R) = {Si Si ∈ max.Denl(Ri), i = 1, . . . , n}, i R ≃ S−1 S−1 Si in R coincides with the core Si,c of the left denominator set Si in Ri, i.e. i Ri, assR(Si) = R1 × · · · × assRi(Si) × · · · × Rn. The core of the left denominator set (R1 × · · · × Si × · · · × Rn)c = 0 × · · · × Si,c × · · · × 0. The proof of Theorem 2.9 is given in Section 7. A bijection between max.Denl(R) and max.Denl(Ql(R)). Proposition 2.10 Let R be a ring, Sl be the largest regular left Ore set of the ring R, Ql := S−1 l R be the largest left quotient ring of the ring R, and C be the set of regular elements of the ring R. Then 1. Sl ⊆ S for all S ∈ max.Denl(R). In particular, C ⊆ S for all S ∈ max.Denl(R) provided C is a left Ore set. 2. Either max.Denl(R) = {C} or, otherwise, C 6∈ max.Denl(R). 3. The map max.Denl(R) → max.Denl(Ql), S 7→ SQ∗ l = {c−1s c ∈ Sl, s ∈ S}, is a bijection with the inverse T 7→ σ−1(T ) where σ : R → Ql, r 7→ r sub-semigroup of (Ql, ·) generated by the set S and the group Q∗ S−1R = (SQ∗ is the l of units of the ring Ql, and 1 , and SQ∗ l l )−1Ql. 4. If C is a left Ore set then the map max.Denl(R) → max.Denl(Q), S 7→ SQ∗ = {c−1s c ∈ C, s ∈ S}, is a bijection with the inverse T 7→ σ−1(T ) where σ : R → Q, r 7→ r 1 , and SQ∗ is the sub-semigroup of (Q, ·) generated by the set S and the group Q∗ of units of the ring Q, and S−1R = (SQ∗)−1Q. Proof. 1. Let S ∈ max.Denl(R). By Lemma 2.2, Sl ⊆ S. 2. Statement 2 follows from statement 1. 3. Statement 3 follows from statement 1 and Proposition 3.4.(1), [2]. 4. If C ∈ Orel(R) then C = Sl and statement 4 is a particular case of statement 3. (cid:3) 3 The First Criterion (via the maximal left denominator sets) The aim of this section is to give a criterion for a ring R to have a semisimple left quotient ring (Theorem 3.1). The implication (⇐) is the most difficult part of Theorem 3.1. Roughly speaking, it proceeds by establishing properties 1-9 which are interesting on their own and constitute the structure of the proof. These properties show that the relationships between the ring R and its semisimple left quotient ring Q are as natural as possible and are as simple as possible ('simple' in the sense that the connections between the properties of R and Q are strong). 8 Theorem 3.1 (The First Criterion) A ring R have a semisimple left quotient ring Q iff i R is a simple left i=1 ass(Si) = 0 and Ri := S−1 Artinian ring for i = 1, . . . , n. If one of the equivalent conditions hold then the map max.Denl(R) = {S1, . . . , Sn} is a finite set, Tn Yi=1 σi : R 7→ Q′ := Yi=1 σ := n n Ri, r 7→ (r1, . . . , rn), is a ring monomorphism where ri = σi(r) and σi : R → Ri, r 7→ r 1 ; and i=1 Si. 1. C =Tn 2. The map σ′ := Qn 3. max.Denl(Q′) = {S′ of the ring Ri, a′ core S′ i : Q → Q′, c−1r 7→ c−1r. We identify the rings Q and Q′ via σ′. σ′ i=1 σ′ i : Q → Q′, c−1r 7→ (c−1r, . . . , c−1r), is a ring isomorphism where 1, . . . , S′ i := ass(S′ n} where S′ i) = R1 × · · · × 0 × · · · × Rn, S′−1 i := R1 × · · · × R∗ i × · · · × Rn, R∗ i is the group of units i Q′ ≃ Ri for i = 1, . . . , n. The i,c of the left denominator set S′ i is equal to R∗ i = 0 × · · · × 0 × R∗ i × 0 × · · · × 0. 4. The map max.Denl(Q′) → max.Denl(R), S′ i 7→ Si := R ∩ S′ i = σ−1 i (R∗ i ) is a bijection, ai := ass(Si) = R ∩ a′ i. 5. For all i = 1, . . . , n, Si *Sj6=i Sj. Moreover, ∅ 6= Si,c ⊆ Si\Sj6=i Sj = R ∩ (R∗ i is the set of zero divisors of the ring Ri. i := Ri\R∗ where R0 i ×Qj6=i R0 i ) i,c = Si ∩ S′ i 6= ∅ for i = 1, . . . , n j=1 Rj are the natural inclusions r 7→ (0, . . . , 0, r, 0, . . . , 0). For all i 6= j, i ∩Tj6=i aj = (Tj6=i aj)\R0 i,c = Si ∩Tj6=i aj = R∗ 6. Si,c = R ∩ S′ i , R0 where R∗ Si,cSj,c = 0. i ⊆Qn 7. C′ := S1,c + · · · + Sn,c ∈ Denl(R, 0), C′−1R ≃ Q, Q = {Pn for i = 1, . . . , n}, CC′ ⊆ C′ and CSi,c ⊆ Si,c for i = 1, . . . , n. i=1 s−1 i ai si ∈ Si,c, ai ∈ Tj6=i aj 8. Q∗ = {s−1t s, t ∈ C′} = {s−1t s, t ∈ C}. 9. C = {s−1t s, t ∈ C′, s−1t ∈ R}. Proof. (⇒) Suppose that the left quotient ring Q of the ring R is a semisimple ring, i.e. Q ≃ Ri n Yi=1 where Ri are simple left Artinian rings. Every simple left Artinian ring is a left localization maximal ring, hence max.Denl(Ri) = {R∗ i }. Then, by Theorem 2.9, max.Denl(Q) = max.Denl( Yi=1 n Ri) = {S′ 1, . . . , S′ n} where S′ i = R1 × · · · × R∗ i × · · · × Rn, assQ(S′ i) = R1 × · · · × 0 × · · · × Ri and S′−1 i Q ≃ Q/assQ(S′ i) ≃ Ri. The core S′ ring R → Q, r 7→ r Proposition 2.10.(4), the map i,c of the maximal left denominator set S′ i × 0 × · · · × 0. The 1 , is a ring monomorphism. We identify the ring R with its image in Q. By i = 0 × · · · × 0 × R∗ i is R∗ max.Denl(Q) → max.Denl(R), S′ i 7→ Si := R ∩ S′ i, is a bijection and S−1 i R ≃ S′−1 i Q ≃ Ri. The inclusions assR(Si) ⊆ assQ(S′ i) where i = 1, . . . , n imply Tn i=1 assR(Si) ⊆Tn i=1 assQ(S′ i) = 0, and so Tn 9 i=1 assR(Si) = 0. i=1 ass(Si) = 0 and Ri := S−1 i R is a simple left Artinian ring for i = 1, . . . , n. We keep the notation of the theorem. Then σ is a monomorphism i=1 ai = 0 where ai := ass(Si)), and we can identify the ring R with its image in the ring Q′. Repeating word for word the arguments of the proof of the implication (⇒), by simply replacing the letter Q by Q′, we see that statement 3 holds (clearly, S′ i × 0 × · · · × 0). (⇐) Suppose that max.Denl(R) = {S1, . . . , Sn}, Tn (sinceTn The group Q′∗ of units of the direct product Q′ is equal to Qn Step 1: C = RT Q′∗: The inclusion C ⊇ RT Q′∗ is obvious. To prove that the reverse inclusion holds it suffices to show that, for each element c ∈ C, the map i,c = 0 × · · · × 0 × R∗ i=1 R∗ i . ·c : Q′ → Q′, q 7→ qc, 1 r1, . . . , s−1 is an injection (then necessarily, the map ·c is a bijection, then it is an automorphism of the left Artinian Q′-module Q′, its inverse is also an element of the type ·c′ for some element c′ ∈ Q′∗. Then c = (c′)−1 ∈ Q′∗, as required). Suppose that kerQ′ (·c) 6= 0, we seek a contradiction. Fix a nonzero element q = (s−1 n rn) ∈ kerQ′ (·c). Without loss of generality we may assume that s1 = · · · = sn = 1, multiplying several times, if necessary, by well-chosen elements of the set i=1 Si in order to get rid of the denominators. There is a nonzero component of the element q = (r1, . . . , rn), say ri. Then ri = σi(r) for some element r 6∈ ai (otherwise, we would have ri = 0). The equality qc = 0 implies that 0 = riσi(c) = σi(rc). Then siric = 0 for some element si ∈ Si, and so sir = 0 since c ∈ C. This means that 0 = σi(r) = ri, a contradiction. The proof of Step 1 is complete. Sn Step 2: For all i 6= j, S−1 ideal of the ring R, hence S−1 S−1 i aj = 0 or, otherwise, S−1 possible, by Proposition 2.4. Therefore, S−1 i aj = Rj: The ring Ri = S−1 i R is a left Artinian ring and aj is an i aj is an ideal of the simple ring Ri. There are two options: either i aj = Ri. Suppose that S−1 i aj = 0, i.e. aj ⊆ ai, but this is not Step 3: For all i = 1, . . . , n, Si ∩Tj6=i aj 6= ∅: For each pair of distinct indices i 6= j, Ri = S−1 i aj i aj for some elements si ∈ Si and aj ∈ aj. Then si − aj ∈ ai, and (by Step 2), and so Ri ∋ 1 = s−1 so there is an element sij ∈ Si such that sij(si − aj) = 0, and so i aj = Ri. Then tij := sij si = sijaj ∈ Si ∩ aj. ti :=Yj6=i tij ∈ Si ∩ \j6=i aj. 4. We have the commutative diagram of ring homomorphisms: R ❄ ❄ ❄ σ / σi ❄ ❄ ❄ ❄ / Q′ pi Ri ❄ (6) (7) σi = piσ where pi : Q′ = Qn S′−1 i Q′ where Si ∈ max.Denl(R) and S′ and S′ i ). Therefore, i = p−1 (R∗ i i R = Ri ≃ (R∗ i ) i j=1 Rj → Ri, (r1, . . . , rn) 7→ ri. By statement 3, S−1 i ∈ max.Denl(Q′). Then, by Theorem 2.5.(3), Si = σ−1 Si = σ−1 i and so the map (R∗ i ) = (piσ)−1(R∗ i ) = σ−1(p−1 i (R∗ i )) = σ−1(S′ i), max.Denl(Q′) → max.Denl(R), S′ i → Si = σ−1(S′ i) = R ∩ S′ i, is a bijection. It follows from the commutative diagram (7) and Step 3 that ai = ass(Si) = R ∩ (R1 × · · · × 0 × · · · × Rn) = R ∩ a′ i. (8) 10 In more detail, ai ⊆ R ∩ a′ ti(R ∩ a′ i) ⊆ ∩n i=1a′ i = 0). This equality implies the inclusion R ∩ a′ i, by (7). Then ti ∈ Si ∩Tj6=i a′ i,c. The inclusion Si,c ⊇ R ∩ S′ 6. We claim that Si,c = R ∩ S′ j, by (6), and so ti(R ∩ a′ i) = 0 (since i ⊆ ai, and so (8) holds. i,c follows from (8). Suppose that i,c. Then s = (s1, . . . , sn) Si,c\R ∩ S′ with sj 6= 0 for some j such that j 6= i. Let tj be the element from Step 3, see (6), i.e. i,c 6= ∅, we seek a contradiction. Fix an element s ∈ Si,c\R ∩ S′ tj ∈ Sj ∩ \k6=j ak ⊆ ai, since j 6= i. Notice that tj = (0, . . . , 0, tj, 0, . . . , 0) and tj ∈ Sj ⊆ S′ j . On the one hand, stj = 0 since s ∈ Si and tj ∈ ai, and so 0 = σj (stj) = sjtj. On the other hand, sjtj 6= 0 since sj 6= 0 and tj ∈ R∗ i,c. Then Si,cSj,c = 0 for all i 6= j. Then, for all i = 1, . . . , n, j , a contradiction. Therefore, Si,c = R ∩ S′ j, and so tj ∈ R∗ Si,c = R ∩ S′ i,c = Si ∩ S′ i,c = R ∩ S′ a′ j = (R ∩ S′ i) ∩ \j6=i R ∩ a′ j = Si ∩ \j6=i aj ∋ ti, Si,c = S′ i,c ∩ \j6=i aj = R∗ i ∩\j6=i i ∩\j6=i aj = (\j6=i aj)\R0 i 6= ∅. So, statement 6 has been proven. 2 and 7. By statement 6, C′ ⊆ Q′∗ =Qn i , and so C′ ⊆ R ∩ Q′∗ = C, i=1 R∗ by Step 1. Since C ⊆ Q′∗ (Step 1), for each i = 1, . . . , n, CSi,c ⊆ Si,c, by statement 6. Then CC′ ⊆ C′ since C′ =Pn By Theorem 2.8.(2), i=1 Si,c. S−1 i,c R = S−1 i R = Ri for i = 1, . . . , n. Each element q ∈ Q′ can be written as q = (s−1 n rn) for some elements si ∈ Si,c and ri ∈ R. The element ri is unique up to adding an element of the ideal ai. Notice that siai = 0 since si ∈ Si,c, and so si(ri + ai) = siri. Then 1 r1, . . . , s−1 n q = (s−2 1 s1r1, . . . , s−2 n snrn) = s−2 i · siri Xi=1 = (s2 1 + · · · + s2 n)−1(s1r1 + · · · + snrn) = c′−1r 1 + · · · + s2 where c′ = s2 (statement 6). Notice that s2 si ∈ Si,c). Therefore, n ∈ C′ and r = s1r1 + · · · + snrn ∈ R, since Si,cSj,c = 0 for all i 6= j i ∈ Si,c and siri ∈ Tj6=i aj for i = 1, . . . , n (by statement 6, since Xi=1 i ai ti ∈ Si,c, ai ∈ \j6=i aj, i = 1, . . . , n}. t−1 n Q′ = { This equality implies that C′ ∈ Denl(R, 0) and C′−1R = Q′. Since C′ ⊆ C and C ⊆ Q′∗ (Step 1), every element q ∈ Q′ can be written as a left fraction c′−1r where c′ ∈ C′ ⊆ C and r ∈ R. This fact implies that C ∈ Orel(R): for given elements c ∈ C and r ∈ R, Q′ ∋ rc−1 = c′−1r′ for some elements c′ ∈ C′ and r′ ∈ R, or, equivalently, c′r = r′c, and so the left Ore condition holds for C. Since C ∈ Orel(R), C ⊆ Q′∗ and R ⊆ Q′, the map σ′ : Q = C−1R → Q′ is a ring monomorphism which is obviously an epimorphism as Q′ = C′−1R and C′ ⊆ C. Therefore, σ′ is an isomorphism and we can write Q = C−1R = C′−1R = Q′, and statements 2 and 7 hold. 1. By Proposition 2.10.(1) (or by Lemma 2.2), C ⊆ Si for all i = 1, . . . , n, and so C ⊆ n n st. 4= Si R ∩ S′ i = R ∩ n \i=1 S′ i \i=1 = R ∩ R∗ 1 × · · · × R∗ n = R ∩ Q′∗ = C, \i=1 11 by Step 1. Therefore, C =Tn 5. For all i = 1, . . . , n, ∅ 6= Si,c ⊆ Tj6=i aj (statement 4), and so ∅ 6= Si,c ⊆ Si\Sj6=i Sj, by i ), statement 4. Then r ∈ Si iff σi(r) ∈ R∗ statement 6. Let r ∈ R. Recall that Si = σ−1 r 6∈ Si iff σi(r) ∈ Ri\R∗ i=1 Si. i ; and (R∗ i i = R0 i . Therefore, 8. By statement 7, Q = {Pn statement 2. Then q ∈ Q∗ = Q′∗ iff s−1 ai ∈ Si,c for i = 1, . . . , n (since, by statement 6, Si,c = R∗ i ai ∈ R∗ i=1 s−1 R0 j . Sj = R ∩ R∗ i ×Yj6=i Si\[j6=i i ai si ∈ Si,c, ai ∈ Tj6=i aj, i = 1, . . . , n}, and Q = Q′, by i for i = 1, . . . , n iff ai ∈ R∗ i for i = 1, . . . , n iff q = (s1 + · · · + sn)−1(a1 + · · · + an) i ∩Tj6=i aj). Therefore, where s1 + · · · + sn, a1 + · · · + an ∈ C′, i.e. Q∗ = {s−1t s, t ∈ C′}. Since C ⊆ Q∗ (Step 1), the previous equality implies the equality Q∗ = {s−1t s, t ∈ C}. 9. By Step 1, C = R ∩ Q∗. Then, by statement 8, C = {s−1t s, t ∈ C′; s−1t ∈ R}. (cid:3) Corollary 3.2 Suppose that the left quotient ring Q of a ring R is a simple left Artinian ring. Then max.Denl(R) = {C}, and so every left denominator set of the ring R consists of regular elements. Proof. Theorem 3.1.(1). (cid:3) Corollary 3.3 Let R be a ring with a semisimple left quotient ring Q (i.e. R is a semiprime left Goldie ring) and we keep the notation of Theorem 3.1. Then 1. The left localization radical of the ring R is equal to zero and the set of regular elements of R coincides with the set of completely left localizable elements. 2. (a) N Ll(R) = {r ∈ R σi(r) ∈ R0 i for i = 1, . . . , n} = {r ∈ R r + p 6∈ CR/p for all p ∈ Min(R)}. (b) R · N Ll(R) · R ⊆ N Ll(R). (c) N Ll(R) + N Ll(R) ⊆ N Ll(R), i.e. N Ll(R) is an ideal of R, iff N Ll(R) = 0 iff R is a domain. Proof. 1. Theorem 3.1. 2(a). The first equality follows from Theorem 3.1.(4). The second equality follows from Theo- rem 4.1(2c). (b) The inclusion follows from the first equality in (a) and the fact that RiR0 i Ri ⊆ R0 i since Ri is a simple Artinian ring. (c) The statement (c) follows at once from the obvious fact that in a semisimple Artinian ring a sum of two zero divisors is always a zero divisor iff the ring is a division ring. (cid:3) The next corollary (together with Theorem 3.1) provides the necessary conditions of Theorem 4.1. Corollary 3.4 Let R be as in Theorem 3.1 and we keep the notation of Theorem 3.1 and its proof. Then 1. The ring R is a semiprime ring. 2. Min(R) = {a1, . . . , an}. 3. For each i = 1, . . . , n, Si = π−1 i (CRi ) = {c ∈ R c + ai ∈ CRi } where πi : R → Ri := R/ai, r 7→ r + ai, and CRi is the set of regular elements of the ring Ri. 12 4. For each i = 1, . . . , n, CRi r + ai 7→ r 1 . = σ−1 i (R∗ i ) ∈ Orel(Ri) and C−1 Ri Ri ≃ Ri where σi : Ri → Ri, Ri 5. S−1 if i = j, if i 6= j. i aj =(0 Proof. 1. Statement 1 follows from statement 2 and Theorem 3.1: Tp∈Min(R) p =Tn i=1 ai = 0. i ai = 0 for i = 1, . . . , n. Then statement 5 follows from Step 2 in the proof of 5. Clearly, S−1 Theorem 3.1. 2. Statement 2 follows from the following two statements: (i) a1, . . . , an are prime ideals, (ii) Min(R) ⊆ {a1, . . . , an}. Indeed, by (ii), Min(R) = {a1, . . . , am}, up to order. The ideals a1, . . . , an are incomparable, i.e. ai 6⊆ aj for all i 6= j. Then, by (i), we must have m = n (since otherwise we would have (up to order) ai ⊆ am+1 for some i such that 1 ≤ i ≤ m, a contradiction). Proof of (i): For each i = 1, . . . , n, the rings Ri are left Artinian. So, if a is an ideal of i a is an ideal of the ring the ring Ri then S−1 i=1 S−1 i=1 Ri. If ab ⊆ ai for some ideals of the ring R then i a is an ideal of the ring Ri, hence Qa = Qn Q = Q′ =Qn Qa · Qb ⊆ Qab ⊆ Qai = n Yj=1 S−1 j ai = R1 × · · · × 0 × · · · × Rn = a′ i, by statement 5 and Theorem 3.1.(3). The ideal a′ Qa ⊆ a′ ai = R ∩ a′ i. Then, either a ⊆ R ∩ Qa ⊆ R ∩ a′ i or Qb ⊆ a′ i, by Theorem 3.1.(4)), i.e. ai is a prime ideal of the ring R. i of the ring Q is a prime ideal. Therefore, either i = ai (since i = ai or b ⊆ R ∩ Qb ⊆ R ∩ a′ Proof of (ii): Let p be a prime ideal of the ring R. Then ai ⊆ n Yi=1 n \i=1 ai = ker(σ) = {0} ⊆ p (Theorem 3.1), and so ai ⊆ p for some i, and the statement (ii) follows from the statement (i). 3 and 4. There is a commutative diagram of ring homomorphisms R σi / Ri ❄ ❄ ❄ ❄ πi ❄ ❄ ❄ ❄ σi Ri 1 is a monomorphism, πi(Si) ∈ Denl(Ri, 0) and πi(Si)−1Ri ≃ Ri = S−1 where σi : r + ai 7→ r i R (via an obvious extension of σi). The ring Ri is a simple left Artinian ring, hence a left localization maximal ring. By Theorem 2.5.(3), Si = σ−1 . On the other hand, , the map ·c : Ri → Ri, r 7→ rc, is a left Ri-module monomorphism, hence for each element c ∈ CRi as isomorphism, and so c ∈ R∗ i ). Clearly, σ−1 i ) ⊆ CRi i . Therefore, σ−1 . Now, ⊆ R∗ i , i.e. CRi i ) = CRi (R∗ (R∗ (R∗ i i i Si = σ−1(R∗ i ) = (σiπi)−1(R∗ i ) = π−1 i (σ−1 i (R∗ i )) = π−1 i (CRi ). (9) Therefore, CRi = πi(Si) ∈ Orel(Ri) and C−1 Ri Ri ≃ πi(Si)−1Ri ≃ Ri. (cid:3) 4 The Second Criterion (via the minimal primes) The aim of this section is to give another criterion (Theorem 4.1) for a ring R to have a semisimple left quotient ring Q. 13 / O O Theorem 4.1 (The Second Criterion) Let R be a ring. The following statements are equiva- lent. 1. The ring R has a semisimple left quotient ring Q. 2. (a) The ring R is a semiprime ring. (b) The set Min(R) of minimal primes of the ring R is a finite set, say, {a1, . . . , an}. (c) For each i = 1, . . . , n, Si := π−1 i πi : R → Ri := R/ai, r 7→ r + ai. ) = {c ∈ R c + ai ∈ CRi } ∈ Denl(R, ai) where (CRi (d) For each i = 1, . . . , n, the ring Ri := S−1 i R is a simple left Artinian ring. If one of the two equivalent conditions holds then max.Denl(R) = {S1, . . . , Sn} and Q ≃Qn i=1 Ri. Proof. (1 ⇒ 2) Corollary 3.4 and Theorem 3.1. (2 ⇒ 1) It suffices to prove the following claim. Claim: max.Denl(R) = {S1, . . . , Sn}. Since then Tn and as a result the ring R has a semisimple left quotient ring. i=1 ai = 0 (as R is a semiprime ring) and so the assumptions of Theorem 3.1 hold Proof of the Claim: (i) S1, . . . , Sn are distinct since a1 = ass(S1), . . . , an = ass(Sn) are distinct. (ii) max.Denl(R) ⊇ {S1, . . . , Sn}: Every simple left Artinian ring is a left localization maximal i R ≃ πi(Si)−1Ri for i = 1, . . . , n. The ring Ri is a left Artinian ring , πi(Si) ∈ Orel(Ri), ring, for example, Ri = S−1 that contains the ring Ri (via σi). Hence, CRi CRi i and Ri = πi(Si)−1Ri, we see that i . Since πi(Si) ⊆ CRi ⊆ R∗ ⊆ R∗ CRi ∈ Orel(Ri) and C−1 Ri Ri = πi(Si)−1Ri = Ri. Recall that σi : R → Ri, r 7→ r The ring C−1 Ri Proposition 2.10, and CRi max.Denl(Ri) since Ri is a left localization maximal ring. Now, Ri ≃ Ri is a left localization maximal ring. Therefore, max.Denl(Ri) = {CRi i ), by Theorem 2.5.(3). Then, by Theorem 2.5.(3), σ−1 1 , and πi : R → Ri, r 7→ r = r + ai. Clearly, σi = σiπi. }, by i ) ∈ = σ−1 (R∗ (R∗ i i σ−1 i (R∗ i ) = (σiπi)−1(R∗ i ) = π−1 i (σ−1 i (R∗ i )) = π−1 i (CRi ) = Si. Then Si ∈ max.Denl(R for i = 1, . . . , n. (iii) max.Denl(R) ⊆ {S1, . . . , Sn}: We have to show that, for a given left denominator set S ∈ Denl(R, a) of the ring R; S ⊆ Si for some i. We claim that S ∩ ai = ∅ for some i otherwise for each i = 1, . . . , n, we would have chosen an element si ∈ S ∩ ai, then we would have si ∈ S ∩ n Yi=1 n \i=1 ai = S ∩ 0 = ∅, a contradiction. We aim to show that S ⊆ Si. The condition S ∩ ai = ∅ implies that S := πi(S) ∈ Orel(Ri). Let b = assRi (S). (α) We claim that b = 0. The ring Ri is a simple left Artinian ring with Ri ⊆ Ri. In particular, it satisfies the a.c.c. on right annihilators (since Ri is a matrix ring with entries from a division ring). By Lemma 4.2 which is applied in the situation that Ri ⊆ Ri and S ∈ Orel(Ri), the core Sc of the left Ore set S is a non-empty set. Let s ∈ Sc. Then RisRi · b = Risb = 0. The ring Ri is a prime ring and RisRi 6= 0 since 0 6= s ∈ RisRi. Therefore, b = 0. (β) a ⊆ ai: by (α). (γ) SSi ∈ Denl(R), by (β) and Lemma 2.2. Hence S ⊆ SSi = Si, by the maximality of Si. (cid:3) The next result is a useful tool in finding elements of the core of an Ore set in applications. 14 Lemma 4.2 Let R ⊆ R′ be rings and S ∈ Orel(R). Suppose that the ring R′ satisfies the a.c.c. on right annihilators and let MaxR′ (S) = {s ∈ S kerR′ (s·) is a maximal element of the set {kerR′ (s·) s ∈ S} }. Then 1. MaxR′ (S) 6= ∅ and kerR′ (s·) = kerR′ (t·) for all elements s, t ∈ MaxR′ (S). 2. MaxR′ (S) ⊆ Sc. Proof. 1. The ring R′ satisfies the a.c.c. on right annihilators, so MaxR′ (S) 6= ∅. Since S ∈ Orel(R), we must have kerR′ (s·) = kerR′ (t·) for all elements s, t ∈ MaxR′ (S) (the left Ore condition for S implies that s′s = rt for some elements s′ ∈ S and r ∈ R, hence s′s ∈ S and kerR′ (s·), kerR′ (t·) ⊆ kerR′ (ss′·), and so kerR′ (s·) = kerR′ (ss′·) = kerR′ (t·) ). 2. For all s ∈ S, kerR(s·) = R ∩ kerR′ (s·). Therefore, statement 2 follows from statement 1. (cid:3) 5 The Third Criterion (in the spirit of Goldie-Lesieur-Croisot) The aim of this section is to prove Theorem 5.1. This is the Third Criterion for a ring to have a semisimple left quotient ring. It is close to to Goldie's Criterion but in applications it is easier to check its conditions. It reveals the 'local' nature of the fact that a ring has a semisimple left quotient ring. For a semiprime ring R and its ideal I, the left annihilator of I in R is equal to the right annihilator of I in R and is denoted ann(I). A ring is called a left Goldie ring if it has a.c.c. on left annihilators and does not contain infinite direct sums of nonzero left ideals. Theorem 5.1 (The Third Criterion) Let R be a ring. The following statements are equivalent. 1. The ring R has a semisimple left quotient ring. 2. The ring R is a semiprime ring with Min(R) < ∞ and, for each p ∈ Min(R), the ring R/p is a left Goldie ring. Remark. The condition Min(R) < ∞ in Theorem 5.1 can be replaced by any of the equivalent conditions of Theorem 5.2. Theorem 5.2 (Theorem 2.2.15, [13]) The following conditions on a semiprime ring R are equiv- alent. 1. RRR has finite uniform dimension. 2. Min(R) < ∞. 3. R has finitely many annihilator ideals. 4. R has a.c.c. on annihilator ideals. Proof of Theorem 5.1. (1 ⇒ 2) Theorem 4.1 and Goldie's Theorem (since Ri = S−1 i R ≃ C−1 R/ai (R/ai) = Q(R/ai)). (1 ⇐ 2) If the ring R is a prime ring then the result follows from Goldie-Lesieur-Croisot's Theorem in the prime case [8], [12]. So, we can assume that R is not a prime ring, that is n := Min(R) ≥ 2. The idea of the proof of the implication is to show that the conditions (a)-(d) of Theorem 4.1 hold. The conditions (a) and (b) are given. Let min(R) = {a1, . . . , an}, we keep the notation of Theorem 4.1. For each i = 1, . . . , n, the ring Ri = R/ai is a prime left Goldie ring ∈ Orel(Ri) and (by the assumption). By Goldie-Lesieur-Croisot's Theorem in the prime case, CRi Q(Ri) is a simple Artinian ring. The conditions (c) and (d) follows from Proposition 5.3 and the Claim. 15 Claim: ann(ai) ∩ Si 6= ∅ for i = 1, . . . , n. Indeed, by Proposition 5.3 which is applied in the situation R, I = ai, a = 0 and S = CRi , we have Si ∈ Denl(R, ai) and S−1 i R ≃ C−1 Ri Ri = Q(Ri) is a simple Artinian ring, as required. Proof of the Claim. The ring R is semiprime, so ann(ai) + ai = ann(ai) ⊕ ai (since (ann(ai) ∩ ai)2 = 0, we must have ann(ai) ∩ ai = 0, as R is a semiprime ring). Since n ≥ 2, 0 6= Tj6=i ai ⊆ ann(ai) (as ann(ai) ·Tj6=i ai ⊆Tn k=1 ak = 0, the ring R is a semiprime ring). Let πi : R → Ri := R/ai, r 7→ r + ai. Then πi(ann(ai)) is a nonzero ideal of the prime ring Ri. Any nonzero ideal of a prime ring is an essential left (and right) ideal, Lemma 2.2.1.(i), [13]. Hence, π(ann(ai)) ∩ CRi 6= ∅, by Goldie-Lesieur-Croisot's Theorem (in the prime case). Therefore, ann(ai) ∩ Si 6= ∅, as required. (cid:3) Proposition 5.3 Let R be a ring, I be its ideal, π : R → R := R/I, r 7→ r = r+I, S ∈ Denl(R, a), S := π−1(S) and a := π−1(a). If, for each element x ∈ I, there is an element s ∈ S such that sx = 0 then S ∈ Denl(R, a) and S−1R ≃ S R. −1 Proof. By the very definition, S is a multiplicative set. (i) S ∈ Orel(R): Given elements s ∈ S and r ∈ R, we have to find elements s′ ∈ S and r′ ∈ R such that s′r = r′s. Since S ∈ Denl(R, a), s1r = r1s for some elements s1 ∈ S and r1 ∈ R. Then x := s1r − r1s ∈ I, and so 0 = s2x for some element s2 ∈ S, then s2s1r = s2r1s. It suffices to take s′ = s2s1 and r′ 1 = s2r1. (ii) ass(S) = a: Let r ∈ ass(S), i.e. sr = 0 for some element s ∈ S. Then sr = 0, and so r ∈ a. This implies that r ∈ a. So, ass(S) ⊆ a. Conversely, suppose that a ∈ a. We have to find an element s ∈ S such that sa = 0. Since a ∈ a and S ∈ Denl(R, a), s1a = 0 for some element s1 ∈ S. Then s1a ∈ I, and so s2s1a = 0 for some element s2 ∈ S. It suffice to take s = s2s1. (iii) S ∈ Denl(R, a): In view of (i) and (ii), we have to show that if rs = 0 for some r ∈ R and −1 (iv) S−1R ≃ S s ∈ S then r ∈ a. The equality rs = 0 implies that r ∈ a (since S ∈ Denl(R, a)), and so r ∈ a. −1 R: By the universal property of left localization, the map S−1R → S R, s−1r 7→ s−1r, is a well-defined ring homomorphism which is obviously an epimorphism. It suffices to show that the kernel is zero. Given s−1r ∈ S−1R with s−1r = 0. Then s1r = 0 in R, for some element s1 ∈ S. Then s1r ∈ I, and so s2s1r = 0 for some element s2 ∈ S. This means that s−1r = 0 in S−1R. So, the kernel is equal to zero and as the result S−1R ≃ S R. (cid:3) −1 6 The Fourth Criterion (via certain left denominator sets) In this section, a very useful criterion for a ring R to have a semisimple left quotient ring is given (Theorem 6.2). It is a corollary of Theorem 5.1 and the following characterization of the set of minimal primes in a semiprime ring. Proposition 6.1 Let R be a ring and a1, . . . , an ideals of R and n ≥ 2. The following statements are equivalent. 1. The ring R is a semiprime ring with Min(R) = {a1, . . . , an}. 2. The ideals a1, . . . , an are incomparable prime ideals with Tn i=1 ai = 0. Proof. (1 ⇒ 2) Trivial. (1 ⇐ 2) The ideals a1, . . . , an are incomparable ideals hence are distinct. The ideals ai are prime. So, or each i = 1, . . . , n, fix a minimal prime ideal pi ∈ Min(R) such that pi ⊆ ai. We claim j=1 aj = {0} ⊆ pi, imply aj ⊆ pi ⊆ ai for some j, necessarily j = i (since the ideals a1, . . . , an are incomparable). Therefore, pi = ai. So, i=1 ai = 0. Therefore, Min(R) = {a1, . . . , an} (otherwise we would have a minimal prime p ∈ Min(R) distinct from the ideals a1, . . . , an but then that pi = ai for i = 1, . . . , n. The inclusions Qn the ideals a1, . . . , an are minimal primes of R with Tn i=1 ai ⊆Tn Qn i=1 ai = {0} ⊆ p ⇒ ai ⊆ p, for some i, a contradiction). (cid:3) j=1 aj ⊆Tn 16 Theorem 6.2 (The Fourth Criterion) Let R be a ring. The following statements are equiva- lent. 1. The ring R has a semisimple left quotient ring. 2. There are left denominator sets S′ 1, . . . , S′ n of the ring R such that the rings Ri := S′−1 i R, i = 1, . . . , n, are simple left Artinian rings and the map σ := n Yi=1 σi : R → n Yi=1 Ri, R 7→ ( r 1 , . . . , r 1 ), is an injection where σi : R → Ri, r 7→ r 1 . If one of the equivalent conditions holds then the set max.Denl(R) contains precisely the distinct elements of the set {σ−1 (R∗ i ) i = 1, . . . , n}. i Proof. (1 ⇒ 2) By Theorem 3.1 and (2), it suffices to take max.Denl(R) since, by (1), lR = 0. (1 ⇐ 2) If n = 1 the implication (2 ⇒ 1) is obvious: the map σ : R → S′−1 1 R is an injection, 1 ∈ Denl(R, 0). In particular, the set S1 consists of regular elements of the ring R. Since 1 R (vis σ) and the ring S′−1 1 R)∗, and so 1 R is a simple left Artinian ring, S′ 1 ⊆ C ⊆ (S′−1 hence S′ R ⊆ S′−1 C ∈ Orel(R) and Q = S−1 1 R. Suppose that n ≥ 2. It suffices to verify that the conditions of Theorem 5.1.(2) hold. For each i = 1, . . . , n, the ring Ri is a simple left Artinian ring, and so Ri is a left localization maximal ring. By Theorem 2.5.(3), Si = σ−1 i (R∗ i ) ∈ max.Denl(R), S−1 i R ≃ Ri, and obviously S′ i). Up to order, let S1, . . . , Sm be the distinct elements of the collection {S1, . . . , Sn}. By Proposition 2.4, the ideals a1, . . . , am are incomparable. The map σ is an injection, hence i ⊆ Si and ai = ass(Si) = ass(S′ 0 = ker(σ) = ass(S′ i) = n \i=1 n \i=1 ass(Si) = ass(S′ i). m \i=1 −1 i Ri ≃ S−1 For each i = 1, . . . , m, let πi : R → Ri := R/ai, r 7→ ri = r + ai. Then Si := πi(Si) ∈ Denl(Ri, 0) and the ring S i R is a simple left Artinian ring which necessarily coincides with the left quotient ring Q(Ri) of the ring Ri (see the case n = 1). By Goldie-Lesieur-Croisot's Theorem the ring Ri is a prime left Goldie ring. By Proposition 6.1, R is a semiprime ring with Min(R) = {a1, . . . , am}. So, the ring R satisfies the conditions of Theorem 5.1.(2), as required. (cid:3) 7 Left denominator sets of finite direct products of rings i=1 Ri be a direct product of rings Ri and 1 = Pn i=1 Ri be a direct product of rings Ri. The aim of this section is to give a criterion of when a multiplicative subset of R is a left Ore/denominator set (Proposition 7.2) and using it to show that the set of maximal left denominator sets of the ring R is the union of the sets of maximal left denominators sets of the rings Ri (Theorem 2.9). i=1 ei be the corresponding sum of central orthogonal idempotents (e2 i = ei and eiej = 0 for all i 6= j). Let πi : R → Ri, r = (r1, . . . , rn) 7→ ri and νi : Ri → R, ri 7→ (0, . . . , 0, ri, 0, . . . , 0). Clearly, νiπi = idRi, the i=1 ai of ideals Let R = Qn Let R = Qn identity map in Ri. Every ideal a of the the ring R is the direct product a = Qn Lemma 7.1 Let R =Qi 1. If S ∈ Orel(R, a = Qn i=1 ai) then for each i = 1, . . . , n either 0 ∈ Si := πi(S) or otherwise i=1 Rn be a direct product of rings Ri. We keep the notation as above. Si ∈ Orel(Ri, ai) and at least for one i, Si ∈ Orel(Ri, ai). ai = eia of the rings Ri. 17 2. If S ∈ Denl(R, a = Qn Si ∈ Denl(Ri, ai) and at least for one i, Si ∈ Denl(Ri, ai). i=1 ai) then for each i = 1, . . . , n either 0 ∈ Si := πi(S) or otherwise Proof. Straightforward. (cid:3) For each S ∈ Orel(R), let supp(S) := {i πi(S) ∈ Orel(Ri)}. By Lemma 7.1.(1), a = ass(S) = ai, ai =(assRi (πi(S)) Ri if i ∈ supp(S), if i 6∈ supp(S). n Yi=1 (10) The following proposition is a criterion for a multiplicative set of a finite direct product of rings to be a left Ore set or a left denominators set. Proposition 7.2 Let R =Qn and Si := πi(S) for i = 1, . . . , n. Then i=1 Ri be a direct product of rings Ri, S be a multiplicative set of R 1. S ∈ Orel(R) iff for each i = 1, . . . , n either 0 ∈ Si or otherwise Si ∈ Orel(Ri) and at least for one i, Si ∈ Orel(Ri). 2. S ∈ Denl(R) iff for each i = 1, . . . , n either 0 ∈ Si or otherwise Si ∈ Denl(Ri) and at least for one i, Si ∈ Denl(Ri). Proof. 1. (⇒) Lemma 7.1.(1). (⇐) Clearly, supp(S) 6= ∅ and ass(S) is given by (10). Without loss of generality we may assume that supp(S) = {1, . . . , m} where 1 ≤ m ≤ n. We have to show that for given elements s = (si) ∈ S and r = (ri) ∈ R there are elements s′ = (s′ i) ∈ R such that s′r = r′s. Since S1 ∈ Orel(R1), t11r1 = a1s1 for some elements t11 ∈ S1 and a1 ∈ R1 ⊆ R. Fix an element t1 ∈ S such that π1(t1) = t11. Using the same sort of argument but for S2 ∈ Orel(R2), we have t22π2(t1r) = a2s2 for some elements t22 ∈ S2 and a2 ∈ R2 ⊆ R. Fix an element t2 ∈ R such that π2(t2) = t22. Repeating the same trick several times we have the equalities i) ∈ S and r′ = (r′ tiiπi(ti−1 · · · t2t1r) = aisi, i = 1, . . . , m, where tii ∈ Si, ai ∈ Ri and ti ∈ S such that πi(ti) = tii. If m = n then it suffices to take s′ = tn · · · t2t1 and the element r′ = (r′ i) can be easily written using the equalities above 1 = π1(tntn−1 · · · t2)a1, r′ r′ 2 = π2(tntn−1 · · · t3)a2, . . . , r′ n−1 = πn−1(tn)an−1, r′ n = an. If m < n then this case can deduced to the previous one. For, each j = m + 1, . . . , n, choose an element θj = (θji) ∈ S with θjj = 0. Let θ = θm+1 · · · θn. It suffices to take s′ = θtmtm−1 · · · t1 and r′ = θ · (r′′ m, 0, . . . , 0) where 1 , . . . , r′′ 1 = π1(tmtm−1 · · · t2)a1, r′′ r′′ 2 = π2(tmtm−1 · · · t3)a2, . . . , r′′ m−1 = πm−1(tm)am−1, r′′ m = am. {1, . . . , m} where 1 ≤ m ≤ n. By (10), a = Qm Denl(Ri, ai) for i = 1, . . . , m and ker(θ·) ⊇Qn 2. (⇒) Lemma 7.1.(2). (⇐) By statement 1, S ∈ Orel(R). Without loss of generality we may assume that supp(S) = j=m+1 Rj. Using the fact that Si ∈ j=m+1 Rj, we see that S ∈ Denl(R, a): If rs = 0 for some elements r = (ri) ∈ R and s = (si) ∈ S then ri ∈ ai for i = 1, . . . , m since Si ∈ Denl(Ri, ai). Fix an element t1 ∈ S such that π1(t1r) = 0. Since π2(t1r) ∈ a2, fix an element t2 ∈ S such that π2(t2t1r) = 0. Repeating the same argument several times we find elements t1, . . . , tm ∈ S such that i=1 ai ×Qn πi(titi−1 · · · t1r) = 0 for i = 1, . . . , m. Then s′r = 0 where s′ = θtiti−1 · · · t1 ∈ S and the element θ is defined in the proof of statement 1. (cid:3) 18 By Proposition 7.2, for each i = 1, . . . , n, we have the injection max.Denl(Ri) → max.Denl(R = n Yi=1 Ri), Si 7→ R1 × · · · × Si × · · · × Rn. (11) We identify max.Denl(Ri) with its image in max.Denl(R =Qn i=1 Ri). Proof of Theorem 2.9. The theorem follows at once from Proposition 7.2.(2). (cid:3) Corollary 7.3 Let R be a ring and 1 = e1 + · · · + en be a sum of central orthogonal idempotents of the ring R. Then for each S ∈ max.Denl(R) precisely one idempotent belongs to S. Proof. The ring R = Qn from Theorem 2.9. (cid:3) i=1 Ri is a direct product of rings Ri := eiR. Then the result follows 8 Criterion for R/lR to have a semisimple left quotient ring The aim of this section is to give a criterion for the factor ring R/l (where l is the left localization radical of R) to have a semisimple left quotient ring (Theorem 8.1), i.e. R/l is a semiprime left Goldie ring. Its proof is based on four criteria and some results (Lemma 8.2 and Lemma 8.3). Theorem 8.1 Let R be a ring, l = lR be the left localization radical of R, Min(R, l) be the set of minimal primes over the ideal l and πl : R → R/l, r 7→ r = r + l. The following statements are equivalent. 1. The ring R/l has a semisimple left quotient ring, i.e. R/l is a semiprime left Goldie ring. 2. 3. (a) max.Denl(R) < ∞. (b) For every S ∈ max.Denl(R), S−1R is a simple left Artinian ring. (a) l =Tp∈Min(R,l) p. (b) Min(R, l) is a finite set. (c) For all p ∈ Min(R, l), the set Sp := {c ∈ R c + p ∈ CR/p} is a left denominator set of the ring R with ass(Sp) = p. (d) For each p ∈ Min(R, l), S−1 (e) For each p ∈ Min(R, l) and for each l ∈ l, there is an element s ∈ Sp such that sl = 0. p R is a simple left Artinian ring. 4. (a) l = ∩p∈Min(R,l)p. (b) Min(R, l) is a finite set. (c) For each p ∈ Min(R, l), R/p is a left Goldie ring. If one of the equivalent conditions 1 -- 3 holds then (i) The map π′ l : max.Denl(R) → max.Denl(R/l), S 7→ πl(S), is a bijection with the inverse T 7→ π−1 (T ). l (ii) max.Denl(R) = {Sp p ∈ Min(R, l)}. (iii) max.Denl(R/l){Sp/l p ∈ Min(R, p)} where Sp/l := πl(Sp). (iv) For all p ∈ Min(R), Sp = π−1 (Sp/l), ass(Sp/l) = ass(Sp)/l and S−1 p R ≃ S−1 p/lR/l is a l simple left Artinian ring. 19 Theorem 8.1 shows that when the factor ring R/l is a semiprime left Goldie ring there are tight connections between the localization properties of the rings R and R/l. In particular, the map π′ l is an injection. In general, there is no connection between maximal left denominator sets of a ring and its factor ring. l is a bijection. In general situation, Lemma 8.2.(1) shows that the map π′ Before giving the proof of Theorem 8.1 we need some results. Lemma 8.2 Let R be a ring, l = lR be its left localization radical and πl : R → R/l, r 7→ r = r + l. Then 1. The map π′ l : max.Denl(R) → max.Denl(R/l), S 7→ πl(S), is an injection such that S−1R ≃ πl(S)−1(R/l) and assR/l(πl(S)) = assR(S)/l. 2. The map π′′ l : ass.max.Denl(R) → ass.max.Denl(R/l), a 7→ a/l, is an injection. 3. Let T ∈ max.Denl(R/l). Then T ∈ im(π′ l) iff assR/l(T ) = assR(S)/l for some S ∈ max.Denl(R). 4. The map π′ l if a bijection iff the map π′′ l is S ∈ max.Denl(R) such that for each element b ∈ π−1 that sb = 0. l is a bijection iff for every b ∈ max.Denl(R/l) there (b) there is an element s ∈ S such 5. If the map π′′ l is a bijection then its inverse π′−1 l is given by the rule T 7→ π′−1 l (T ). Proof. 1. Since l ⊆ a = ass(S) we have πl(S) ∈ Denl(R/l, a/l) and the map πl,S : S−1R → πl(S)−1R/l, s−1r 7→ π−1 l (s)πl(r), (12) is a ring isomorphism, by Lemma 3.2.(1), [2]. There is a commutative diagram of ring homomor- phisms: R πl σS S−1R πl,S R/l σS/ / πl(S)−1R/l (13) where σS(r) = r . The ring S−1R is a left localization maximal ring. Let G and G′ be the groups of units of the rings S−1R and πl(S)−1R/l respectively. By Theorem 2.5.(3), 1 and σS(r + l) = πl(r) 1 S (G) and S′ := σ−1 By the commutativity of the diagram (13), we have S = σ−1 S (G′) ∈ max.Denl(R/l). Then, by the surjectivity of the map πl, πl(S) = S′ ∈ max.Denl(R/l). So, the map π′ well-defined map which is an injection, by statement 2. l is a S = π−1 l (S′). (14) 2. Every element of ass.max.Denl(R) contains l, hence π′ 3. Statement 3 follows from statement 2 and Corollary 2.3.(2). 4. By Corollary 2.3.(2) and statements 1 and 2, the map π′ l is a bijection/surjection iff the map π′′ is a bijection/surjection. Let LS stands for the statement after the second 'iff' in statement 4 l and let T ∈ max.Denl(R/l) with b = ass(T ). If π′ l is a bijection then LS holds by statement 1 and the inclusion l ⊆ ass(S) for all S ∈ max.Denl(R). If LS holds then clearly π−1 (b) ⊆ ass(S), and so b ⊆ ass(S)/l. Then necessarily b = ass(S)/l, by Proposition 2.4, and so T = πl(S), by statement 1 and Corollary 2.3.(2). This means that the map π′ l is a surjection hence a bijection. l is an injection. l 5. Statement 5 was proved in the proof of statement 1, see (14). (cid:3) 20 / / Lemma 8.3 Let R be a ring, l = lR be its left localization radical and Min(R, l) be the set of minimal primes over l. Suppose that max.Denl(R) = {S1, . . . , Sn} is a finite set and the ideals ai := ass(Si), i = 1, . . . , n, are prime. Then 1. The map max.Denl(R) → Min(R, l), S 7→ ass(S), is a bijection. 2. l =Tp∈Min(R,l) p. Proof. If n = 1 then l = a1 is a prime ideal and so statements 1 and 2 obviously hold. So, let n ≥ 2. The ideals a1, . . . , an are distinct prime ideals with l = ∩n i=1ai. Therefore, the prime ideals a1/l, . . . , an/l of the ring R/l are distinct with zero intersection. By Proposition 6.1, the ring R/l is a semisimple ring with Min(R/l) = {a1/l, . . . , an/l). Hence, Min(R, l) = i=1 an, i.e. statement 2 holds. Now, statement 1 is obvious (by Corollary {a1, . . . , an} with l = Tn 2.3.(2)). (cid:3) Proof of Theorem 8.1. (1 ⇒ 2) This implication follows from Theorem 3.1 and Lemma 8.2.(1). (2 ⇒ 1) By (2), there is a ring monomorphism R/l →QS∈max.Denl(R) S−1R. By Theorem 6.2, the ring R has a semisimple left quotient ring. Then, by Theorem 6.2 and Lemma 8.2.(1, 5), the statement (i) holds. By Goldie's Theorem the ring R is a semiprime left Goldie ring. (1 ⇔ 3) Notice that the map Min(R.l) → Min(R/l), p 7→ p/l, is a bijection. By Theorem 4.1, statement 1 is equivalent to the statements (a)' -- (d)' (we have added 'prime' to distinguish these conditions from the conditions (a) -- (d) of statement 3): (a)' R/l is a semiprime ring, (b)' Min(R/l) < ∞, (c)' For each p/l ∈ Min(R/l), Sp/l := {c + l ∈ R/l c + p ∈ CR/p} ∈ Denl(R/l, p/l), and (d)' For each p/l ∈ Min(R/l), the ring S−1 The conditions (a)' and (b)' are equivalent to the conditions (a) and (b) of statement 3. Now, the implication (1 ⇒ 3) follows: the statement (i) and Lemma 8.2.(4) imply the condition (e); then the conditions (c) and (d) follow from the conditions (c)' and (d)' by using Proposition 5.3. The implication (3 ⇒ 1) follows from Theorem 6.2: the conditions (a) -- (c) imply that there is p/lR/l is a simple left Artinian ring. a ring homomorphism R/l → Yp ∈Min(R,l) S−1 p R. By the condition (d), the rings S−1 ring R/l has a semisimple left quotient ring. p R are simple left Artinian rings. Then, by Theorem 6.2, the (1 ⇒ 4) The conditions (a) and (b) of statement 4 are precisely the conditions (a) and (b) of statement 3. The condition (c) of statement 4 follows from the conditions (c) and (d) of statement 3 and Goldie's Theorem (since Q(R/p) ≃ S−1 p R is a simple left Artinian ring, by statement 3). (4 ⇒ 1) By the condition (c) and Goldie's Theorem, the left quotient rings Q(R/p) are simple left Artinian rings. Then there is a chain of ring homomorphisms R/l = R/ ∩p∈Min(R,p) p → Yp∈Min(R,p) R/p → Yp∈Min(R,p) Q(R/p). By Theorem 6.2, the ring R/l has a semisimple left quotient ring. This finishes the proof of equivalence of statements 1 -- 4. Now, statements (ii) -- (iv) follow from the statement (i) and Lemma 8.2. (cid:3) The work is partly supported by the Royal Society and EPSRC. Acknowledgements 21 References [1] V. V. Bavula, The algebra of integro-differential operators on an affine line and its modules, J. Pure Appl. Algebra, 217 (2013) 495-529. (Arxiv:math.RA: 1011.2997). [2] V. V. Bavula, The largest left quotient ring of a ring, Arxiv:math.RA:1101.5107. [3] V. V. Bavula, Characterizations of left orders in left Artinian rings. Arxiv:math.RA:1212.3529. [4] V. V. Bavula, Left localizations of left Artinian rings. [5] V. V. Bavula, Left localizable rings and their characterizations. [6] E. H. Feller and E. W. Swokowski, Reflective N-prime rings with the ascending chain condition. Trans. Amer. Math. Soc. 99 (1961) 264 -- 271. [7] E. H. Feller and E. W. Swokowski, Reflective rings with the ascending chain condition. Proc. Amer. Math. Soc. 12 (1961) 651 -- 653. [8] A. W. Goldie, The structure of prime rings under ascending chain conditions. Proc. London Math. Soc. (3) 8 (1958) 589 -- 608. [9] A. W. Goldie, Semi-prime rings with maximum condition. Proc. London Math. Soc. (3) 10 (1960) 201 -- 220. [10] A. V. Jategaonkar, Localization in Noetherian Rings, Londom Math. Soc. LMS 98, Cambridge Univ. Press, 1986. [11] C. R. Hajarnavis, On Small's Theorem, J. London Math. Soc. (2) 5 ( 1972) 596 -- 600. [12] L. Lesieur and R. Croisot, Sur les anneaux premiers noeth´eriens `a gauche. Ann. Sci. ´Ecole Norm. Sup. (3) 76 (1959) 161-183. [13] J. C. McConnell and J. C. Robson, Noncommutative Noetherian rings. With the cooperation of L. W. Small. Revised edition. Graduate Studies in Mathematics, 30. American Mathematical Society, Providence, RI, 2001. 636 pp. [14] J. C. Robson, Artinian quotient rings. Proc. London Math. Soc. (3) 17 (1967) 600 -- 616. [15] L. W. Small, Orders in Artinian rings. J. Algebra 4 (1966) 13 -- 41. [16] L. W. Small, Correction and addendum: Orders in Artinian rings. J. Algebra 4 (1966) 505 -- 507. [17] B. Stenstrom, Rings of Quotients, Springer-Verlag, Berlin, Heidelberg, New York, 1975. [18] H. Tachikawa, Localization and Artinian Quotient Rings, Math. Z. 119 (1971) 239 -- 253. [19] T. D. Talintyre, Quotient rings of rings with maximum condition for right ideals, J. London Math. Soc. 38 (1963) 439 -- 450. [20] T. D. Talintyre, Quotient rings with minimum condition on right ideals, J. London Math. Soc. 41 (1966) 141 -- 144. Department of Pure Mathematics University of Sheffield Hicks Building Sheffield S3 7RH UK email: [email protected] 22

Subsets and Splits

No community queries yet

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