Text
                    T •
T H
413
Representations of Algebraic
Groups, Quantum Groups,
and Lie Algebras
AMS-IMS-SIAM Joint Summer Research Conference
July 11-15,2004
Snowbird Resort, Snowbird, Utah
Georgia Benkart
Jens C. Jantzen
Zongzhu Lin
Daniel K. Nakano
Brian J. Parshall
Editors
American Mathematical Society


Representations of Algebraic Groups, Quantum Groups, and Lie Algebras
r\ .V- .. 1-4 £V - Conference Group Photo Snowbird Resort July 2004
Contemporary Mathematics 413 Representations of Algebraic Groups, Quantum Groups, and Lie Algebras AMS-IMS-SIAM Joint Summer Research Conference July 11-15,2004 Snowbird Resort, Snowbird, Utah Georgia Benkart Jens C. Jantzen Zongzhu Lin Daniel K. Nakano Brian J. Parshall Editors American Mathematical Society Providence, Rhode Island
Editorial Board Dennis DeTurck, managing editor George Andrews Carlos Berenstein Andreas Blass Abel Klein This volume contains the proceedings of an AMS-IMS-SIAM Joint Summer Research Conference on Representations of Algebraic Groups, Quantum Groups, and Lie Algebras, held at the Snowbird Resort, Snowbird, UT, from July 11-15, 2004, with support from the National Science Foundation, grant DMS-9973450. 2000 Mathematics Subject Classification. Primary 05E10, 14L17, 16G20, 17Bxx, 20C08, 20Gxx. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. Library of Congress Cataloging-in-Publication Data AMS-IMS-SIAM Joint Summer Research Conference, Representations of Algebraic Groups, Quantum Groups, and Lie Algebras (2004 : Snowbird, Utah) Representations of algebraic groups, quantum groups, and Lie algebras : AMS-IMS-SIAM Joint Summer Research Conference, Representations of Algebraic Groups, Quantum Groups, and Lie Algebras, July 11-15, 2004, Snowbird, Utah / Georgia M. Benkart... [et al.], editors, p. cm. — (Contemporary mathematics, ISSN 0271-4132 ; v. 413) ISBN 0-8218-3924-1 (alk. paper) 1. Representations of groups—Congresses. 2. Affine algebraic groups—Congresses. 3. Quantum groups—Congresses. 4. Lie algebras—Congresses. I. Benkart, Georgia, 1949- II. Title. III. Series: Contemporary mathematics (American Mathematical Society) ; v. 413. QA176.A47 2004 512/.22—dc22 2006045952 Copying and reprinting. Material in this book may be reproduced by any means for educational and scientific purposes without fee or permission with the exception of reproduction by services that collect fees for delivery of documents and provided that the customary acknowledgment of the source is given. This consent does not extend to other kinds of copying for general distribution, for advertising or promotional purposes, or for resale. Requests for permission for commercial use of material should be addressed to the Acquisitions Department, American Mathematical Society, 201 Charles Street, Providence, Rhode Island 02904-2294, USA. Requests can also be made by e-mail to reprint-permissionQams. org. Excluded from these provisions is material in articles for which the author holds copyright. In such cases, requests for permission to use or reprint should be addressed directly to the author(s). (Copyright ownership is indicated in the notice in the lower right-hand corner of the first page of each article.) © 2006 by the American Mathematical Society. All rights reserved. The American Mathematical Society retains all rights except those granted to the United States Government. Copyright of individual articles may revert to the public domain 28 years after publication. Contact the AMS for copyright status of individual articles. Printed in the United States of America. @ The paper used in this book is acid-free and falls within the guidelines established to ensure permanence and durability. Visit the AMS home page at http://www.ams.org/ 10 9 8 7 6 5 4 3 2 1 11 10 09 08 07 06
Contents Preface vii List of Talks ix List of Participants xi Extensions for finite groups of Lie type II: Filtering the truncated induction functor Christopher P. Bendel, Daniel K. Nakano, and Cornelius Pillen 1 Algebras, representations and their derived categories over finite fields Bangming Deng and Jie Du 25 On localization of D-modules Yoshitake Hashimoto, Masaharu Kaneda, and Dmitriy Rumynin 43 Representations of reduced enveloping algebras and cells in the affine Weyl group J. E. Humphreys 63 Nakajima's monomials and crystal bases Seok-Jin Kang, Jeong-Ah Kim, and Dong-Uy Shin 73 A new Lie bialgebra structure on s/(2,1) Gizem Karaali 101 The Steinberg tensor product theorem for GL(m\n) Jonathan Kujawa 123 Cyclotomic g-Schur algebras and Schur-Weyl duality Zongzhu Lin and Hebing Rui 133 Geometric crystals and affine crystals Toshiki Nakashima 157 Self-extensions for finite symplectic groups via algebraic groups Cornelius Pillen 173
vi CONTENTS Classification of finite dimensional simple Lie algebras in prime characteristics Alexander Premet and Helmut Strade 185 Prom quantum groups to unitary modular tensor categories Eric C. Rowell 215 A trip from representations of the Kronecker quiver to canonical bases of quantum affine algebras Jie Xiao and Guanglian Zhang 231
Preface Representation theory has played a central role in mathematics through its rich interplay with, and applications to, many other fields. The 2004 AMS- IMS-SIAM Joint Summer Research Conference, Representations of Algebraic Groups, Quantum Groups, and Lie Algebras, focused on the geometric and combinatorial aspects of the subject. New developments involving quiver representations were presented in connection with important constructions for quantum groups. Another major theme was use of methods from algebraic geometry, via derived categories, to study the representation theory of algebraic groups and Lie algebras, including Kac-Moody Lie algebras, modular restricted Lie algebras (or, more generally, finite group schemes), and Lie superalgebras. Each morning session featured principal speakers on the designated major themes. Each afternoon, two parallel sessions allowed attendees to present talks on current research, providing a forum for junior mathematicians to communicate new developments in the area, followed by ample time for informal discussions and interaction. The present volume brings together papers from the principal speakers and other participants on a wide variety of topics in modern representation theory. Several contributions are surveys that aim to introduce the topics to a wider audience of researchers. All of the papers were carefully refereed, and the editors express their gratitude to the anonymous referees for the high standards employed in preparing their reports. During the conference, a banquet was held to celebrate the achievements of James E. Humphreys on the occasion of his 65th birthday. Over the last 40 years, Jim's contributions have inspired many deep insights and new developments in the representations of algebraic groups and finite groups of Lie type. In addition, his well-known books in the area have brought vast, intertwined research topics together in a concise and coherent manner. Jim has also encouraged many of us by taking a genuine interest in our work. Several months before the conference, he formally retired from the University of Massachusetts, Amherst, to devote himself to research and to book writing. We are delighted to include here one of his recent articles, which poses an interesting conjecture relating irreducible representations of semisimple Lie algebras in positive characteristics to left cells in affine Weyl groups. Financial support for the conference was provided by a grant from the National Science Foundation, and the staff of the American Mathematical Society provided considerable logistical support. In particular, the organizers acknowledge Wayne Drady for his professional dedication to managing the conference and Christine M. Thivierge for her patience and help in editing this volume. We also thank the participants for making the conference a success: the speakers during the conference vii
viii PREFACE and at the banquet and the afternoon session chairs for their work in keeping the conference on schedule. Special thanks go to Leonard Scott, whose toastmastery during the banquet provided many humorous and wonderful memories. Georgia Benkart Jens C. Jantzen Zongzhu Lin Daniel K. Nakano Brian J. Parshall January 2006
List of Talks Talks by Principal Speakers Henning H. Andersen, Cohomology of line bundles Jie Du, Strong monomial basis property and canonical basis for a cyclic quiver Eric M. Priedlander, 7r-points for finite group schemes Seok-Jin Kang, Crystal bases for quantum affine algebras and combinatorics of Young walls Alexander Kleshchev, On the structure of finite W-algebras of type A Ivan Mirkovic, Beilinson-Bernstein localization for quantum groups at roots of unity Hiraku Nakajima, Instanton counting Alexander Premet, Minimal nilpotent representations, quantizations of Slodowy slices, and the Joseph ideal Eric Vasserot, Representations of double affine Hecke algebras Jie Xiao, Representations of tame quivers and affine canonical bases Contributed Talks Susumu Ariki, Representation type of Hecke algebras and the Poincare polynomial Christopher P. Bendel, Cohomology for Frobenius kernels Brian D. Boe, Varieties of nilpotent matrixes for simple Lie algebras: Restricted null- cones and support varieties Jon F. Carlson, Endotrivial modules for finite groups of Lie type Joseph Chuang, Derived equivalence between blocks of GL(n) Stephen Doty, Generators and relations for generalized q- Schur algebras ix
LIST OF TALKS David J. Hemmer, Fixed point functors for symmetric groups and Schur algebras Terrell L. Hodge, Nilpotent orbits in restricted symmetric spaces James E. Humphreys, Representations of reduced enveloping algebras and cells in the affine Weyl group Dijana Jakelic, Crystal and tensor products in category O Joel Kamnitzer, Mirkovic- Vilonen cycles and polytopes Masaharu Kaneda, Localization of D-modules in positive characteristic Gizem Karaali, How to construct an r-matrix on a Lie superalgebra Sergei Krutelevich, Exceptional groups, Jordan algebras, and higher composition laws Jonathan Kujawa, Crystal structures arising from representations of GL(m\n) Yiqiang Li, Affine quivers of type An and canonical bases George J. McNinch, Optimal SL(2)-homomorphisms Kailash C. Misra, Affine Lie algebra representations and multisum identities of Rogers-Ramanujan type Toshiki Nakashima, Geometric crystals and crystal bases Alison Parker, Higher extensions for SL^fc) Aaron Phillips, On 2-modular representations of the symmetric groups Cornelius Pillen, Extensions for finite groups of Lie type and the truncated induction functor Eric C. Rowell, Towards a classification of modular tensor categories Travis Schedler, Quantization of necklace Lie algebras Toshiyuki Tanisaki, The Beilinson-correspondence for quantized enveloping algebras Monica Vazirani, Vanishing integrals of Macdonald polynomials Weiqiang Wang, A super duality and Kazhdan-Lusztig polynomials
List of Participants Henning H. Andersen, Aarhus University, DENMARK Susumu Ariki, Kyoto University, JAPAN Christopher P. Bendel, University of Wisconsin-Stout, USA Georgia Benkart, University of Wisconsin Madison, USA Matthew Beswick, Kansas State University, USA Brian D. Boe, University of Georgia, USA Jonathan Brundan, University of Oregon, USA Jon F. Carlson, University of Georgia, USA Joseph Chuang, University of Bristol, UNITED KINGDOM Wesley Cramer, University of Virginia, USA Stephen Doty, Loyola University Chicago, USA Jie Du, University of New South Wales, AUSTRALIA Eric M. Priedlander, Northwestern University, USA Fredrick Goodman, University of Iowa, USA Holly Hauschild, University of Iowa, USA Xuhua He, Massachusetts Institute of Technology, USA David J. Hemmer, University of Toledo, USA Anthony Henderson, University of Sydney, AUSTRALIA Terrell L. Hodge, Western Michigan University, USA James E. Humphreys, University of Massachusetts-Amherst, USA Dijana Jakelic, University of California-Riverside, USA Jens C. Jantzen, Aarhus University, DENMARK Joel Kamnitzer, University of California-Berkeley, USA Masaharu Kaneda, Osaka City University, JAPAN Seok-Jin Kang, Korea Institute for Advanced Study, SOUTH KOREA Gizem Karaali, University of California-Berkeley, USA Jeon-Ah Kim, Korea Institute of Advanced Study, SOUTH KOREA Alexander Kleshchev, University of Oregon, USA
Xll LIST OF PARTICIPANTS Sergei Krutelevich, University of Ottawa, CANADA Jonathan Kujawa, University of Georgia, USA Weiping Li, Walsh University, USA Yiqiang Li, Kansas State University, USA Zongzhu Lin, Kansas State University, USA Jill E. McCarthy, University of Virginia, USA Kevin McGerty, Institute for Advanced Study, USA George J. McNinch, Tufts University, USA Ivan Mirkovic, University of Massachusetts-Amherst, USA Kailash C. Misra, North Carolina State University, USA Hiraku Nakajima, Kyoto University, JAPAN Daniel K. Nakano, University of Georgia, USA Toshiki Nakashima, Sophia University, JAPAN Alison Parker, University of Sydney, A USTRALIA Brian J. Parshall, University of Virginia, USA Julia Pevtsova, University of Oregon, USA Aaron M. Phillips, University of Virginia, USA Alexander Premet, University of Manchester, UNITED KINGDOM Zhenbo Qin, University of Missouri, USA Eric Rowell, Indiana University, USA Oliver Ruff, University of Oregon, USA Yoshihisa Saito, University of Tokyo, JAPAN Travis Schedler, University of Chicago, USA Leonard Scott, University of Virginia, USA Dong-Uy Shin, Korea Institute for Advanced Study, SOUTH KOREA Eric Sommers, University of Massachusetts-Amherst, USA Anna Stokke, University of Winnipeg, CANADA Toshiyuki Tanisaki, Osaka City University, JAPAN Nathaniel Thiem, University of Wisconsin, USA Michela Varagnolo, University of Cergy-Pontoise, FRANCE Eric Vasserot, University of Cergy-Pontoise, FRANCE Monica Vazirani, University of California at Davis, USA Weiqiang Wang, University of Virginia, USA Jie Xiao, Tsinghua University, CHINA Cornelius Pillen, University of South Alabama, USA
Contemporary Mathematics Volume 413, 2006 Extensions for finite groups of Lie type II: Filtering the truncated induction functor Christopher P. Bendel, Daniel K. Nakano, and Cornelius Pillen Dedicated to James E. Humphreys on the occasion of his 65th birthday Abstract. In [BNP5] the authors relate the extensions between two simple modules for a finite group of Lie type Ga($q) (where q = pr) to certain extensions for the corresponding reductive group and its Probenius kernels. Several of these results require the characteristic p of the underlying field to be sufficiently large (p > 3(h — 1), with h being the Coxeter number of the root system). In this paper we will generalize these results to all primes p assuming instead lower bounds on the prime powers pr (approximately of the order of 1. Introduction 1.1. Let G be a connected reductive algebraic group scheme defined over ¥p and let F : G —> G be the Probenius map. Let Gr be the r-th Probenius kernel which is the scheme theoretic kernel of Fr (F composed with itself r times) and let G{¥q) be the fixed points under Fr. We will assume that k is an algebraically closed field of characteristic p > 0. The finite groups G(¥q) are called the finite Chevalley groups. There has been much effort in the last thirty years aimed at understanding the interrelationships between the representation theory of these three algebraic objects. For a comprehesive treatment of this subject we refer the reader to Humphreys' book [Hum2]. In a series of papers [BNP1, BNP2, BNP3, BNP5] the authors investigated the deep connections between the cohomology theories of G, Gr and G(¥q). The philosophy behind our approach involved using certain truncated categories of rational G-modules which approximate the categories of Gr and G(¥q)-modules. These truncated categories are highest weight categories and contain enough projective modules so one can directly compare these categories to the categories of Gr and 2000 Mathematics Subject Classification. Primary 20C, 20G; Secondary 20J06, 20G10. Research of the first author was supported in part by NSF grant DMS-0400558. Research of the second author was supported in part by NSF grant DMS-0400548. ©2006 American Mathematical Society
2 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN G(Fg)-modules through the use of Grothendieck spectral sequences. In the construction of these spectral sequences, we study certain families of finite dimensional submodules of the infinite dimensional induced module indG^F ^(N) where N G mod(kG(¥q)). These modules can be described as the maximal submodules of ind^ )(JV) whose highest weights are contained in specified finite saturated sets of weights. An important example occurs when we let N be the trivial module. For large primes (p > 3(h — 1)) and truncation at approximately twice the Steinberg weight, it was shown that this module (when N = k) is completely reducible ([BNP1, BNP5]). This fact led to explicit formulas describing extensions of simple modules over G(¥q) via extensions of modules for G [BNP2, BNP3, BNP5]. As an application, we were able to use our formulation to answer many of the questions posed in Humphreys' 1985 article on self-extensions [Huml]. In this paper, we will use the same setup as in [BNP5] and consider the more general family of finite groups of Lie type. We denote these groups by Ga(¥q) where a is the corresponding automorphism of G. For small primes these truncated induction functors are no longer semisimple. Our goal is to study the resulting modules for small primes p, but large prime powers pr. The precise definitions and some basic properties of these truncated categories and associated functors are given in Section 2. Then some useful cohomology facts will be noted in Section 3. Section 4 is devoted to demonstrating that under suitable conditions on pr these truncated induced modules admit a filtration with sections of the form H°(-w0(Tfi) ® #°(/x)(r) (Theorem 4.7). In Section 5, we apply this filtration to make some cohomological computations. For example, the existence of this filtration allows us to show that for r > 2 and sufficiently large q = pr the finite group Ga(¥q) does not allow self-extensions between simple modules (Theorem 5.4). Tiep and Zalesskii [TZ, Prop. 1.4] have shown that the existence of self-extensions are an important factor in the ability to lift irreducible representations from characteristic p to characteristic zero. Finally, in Section 5.6, it is shown that for all primes but r > 3 and q sufficiently large the group of extensions between two simple Ga(¥q)-modules is isomorphic to the G-extensions between a suitable pair of g-restricted simple G-modules. Roughly speaking one can say that Ext^(F ^ for pairs of simple Ga(¥q)-modules mirrors the theory of Ext^ between g-restricted simple modules, provided that r > 3 and q is at least of the order of the Coxeter number squared. No restriction on the prime is necessary. 1.2. Notation. Let G be a connected simply connected almost simple algebraic group defined and split over the finite field ¥p with p elements and k be the algebraic closure of Fp. We will also consider G as an algebraic group scheme over Fp. Let $ be a root system associated to the pair (G, T) where T is a maximal split torus. Moreover, let 3>+ (resp. $~) be the positive (resp. negative) roots and A be a base consisting of simple roots. Let X(T) be the integral weight lattice obtained from <£ contained in the Euclidean space E with the inner product denoted by ( , ). The set X(T) has a partial ordering given by A > \i if and only if A — \i e ]Ca€A^a f°r ^A* e X(T). The set of dominant integral weights is denoted by X(T)+ and the set of pr-restricted weights by Xr(T).
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 3 Let W be the Weyl group. The group W acts on X(T) via the "dot action" given by w • A = w(X + p) — p where p is the half sum of the positive roots. Let av = 2a/(a,a) be the coroot corresponding to a G <£. The longest element in W is w0 and the Coxeter number for $ is h = (p, 0$) + 1 where ao is the maximal short root. Let B be a Borel subgroup containing T corresponding to the negative roots. For A G X(T)+ set H°(X) = ind#A. The simple G-module corresponding to A is denoted by L(A) and the Weyl module is V(A). The injective hull of L(A) as a G- module will be denoted /(A). For more details about the definitions and properties of these objects we refer the reader to [Janl]. Let F : G —> G be the Frobenius map and Fr the composition of the Frobenius map with itself r-times. Now suppose that a is an automorphism of the Dynkin diagram of <£. The automorphism a can be extended to the weight lattice X(T) and under this extension a permutes the fundamental weights and preserves the inner product ( , ) as well as the partial order on X(T). Moreover, cr(ao) — ao- The graph automorphism a also induces an automorphism on G which will also be denoted by a. The automorphism a commutes with F and is compatible with the action of a on X(T). Set Ga(¥q) as the group of fixed points of Fr o a = a o Fr, where q = pr. The groups Ga(¥q) can be either (i) untwisted (Chevalley) groups, (ii) Steinberg groups, or (in) Suzuki-Ree groups. For more information about these groups see [Car] [GLS]. For simplicity we will exclude the Suzuki-Ree groups from our discussion. With some exceptions for the Ree groups of type F4, the extensions for these groups are known due to [Sinl, Sin2, Sin3]. Thoughout this paper, for v G X(T), set v = —w0(ru. 2. Induction and Truncation. 2.1. Induction. For a finite dimensional Ga(¥q)-module M and a finite dimensional Gr-module N, we define G{M) = md%A¥q)(M) and H(N) = indgr(JV). In particular for the trivial module k we set Q(k) = ind§CT(Fg)(fc) and H(k) = indGr(k). If M and N are G-modules, the tensor identity implies G(M) = M® md%A¥q)(k) and H(N) = N® indgr(fc). Our first result shows that Q(k) (resp. H(k)) is injective upon restriction to Gr (resp. Ga(Fq)). Proposition . (i) H(k) is injective as a Ga(¥q)-module. (ii) Q(k) is injective as a Gr-module. Proof, (i) It is well-known that H(k) * k[G/Gr] ^ k[G]{r) * 0 (/(^WjdimLM vex(T)+ as a G-module where I(v) is the injective hull of the simple module L(v). Moreover, I(y) = I(y)(r) as a G<j(Fg)-module and is injective since G/G(T(¥q) is affine. (ii) As a Ga(¥q)-module, the Steinberg module Str is both projective and injective. Furthermore, Str = St^.r^ as G(T(¥q)-modules. The functor Q sends injective
4 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN Ga(¥q)-modules to injective G-modules. Therefore, from the tensor identity we obtain the following sequence of isomorphisms of injective G-modules: £(str) ^ g{stir)) * st^.r) ® g(k). Restriction from G-modules to Gr-modules sends injectives to injectives (because G/Gr is affine) and the r-th Probenius twist of Str, viewed as a module for G>, is a direct sum of trivial modules. Hence, as G>-modules, (?(Str) = Q(k)dimStr and the assertion follows. □ 2.2. Saturated sets of weights. For any finite set of weights ir C X(T)+ we define Q^M) (resp. H^(N)) to be the maximal G-submodule of Q{M) (resp. H{N)) having composition factors with highest weights in 7r. The following three sets of weights will play an important role in the upcoming results: r = {A6I(T)+|(A,^)<2(ft-l)} n = {\eX(T)+\{\,al>)<2pr(h-l)} A = {A 6 X(T)+ | (A, ctf) < 3(pr - l)(h - 1)}. Our goal is to understand the structure of the module Gn(k) for arbitrary (especially small) primes p and large pr. We begin by constructing an ascending chain of submodules for Gn(k). Fix an order Ai, A2, A3,..., An of the elements in F such that i < j whenever (pr — woa)Xi < (pr — wo&)Xj. Notice that i < j whenever A; < Xj. Clearly Ai = 0. Then we define subsets of T as follows, for i = 1,2,..., n, set Ti = {Xj e r I j < i} and r0 = 0. Let 70,71 e T. Then (71,^) < 2(h~l)-l and <7o+Pr7i,^> < 2pr(h- 1) -pr + 2(h — 1). If pr > 2(h — 1), then 70 +pTn/i G O. For the remainder of this section we assume that pr > 2(h— 1). We define subsets of ft: Qi = {7 e X(T)+ I 7 < (pr - w0a)u for some v e Ti}. The subsets Ti and Cti together with their W-conjugates are saturated, Fq = Oo = 0, and ri = Oi = {0}. We have the following series of inclusions (2.2.1) k = gQo(k) cgUl(k) cgQ2(k) c-c^tjcgQ(k). Later we will show that Gnn(k) = Gn(k) for sufficiently large pr. Remark . In [BNP2, BNP5] the notation Q{k) is used for a truncated submod- ule of ind§ /F \(k). Here Q{k) will always denote the infinite-dimensional module ind§CT(Fg)(fc) itself. Any finite-dimensional truncated submodule will be denoted by G7r(k) with 7r being the corresponding finite set of weights. 2.3. Injectives and projectives in the truncated categories. Let ir be a finite set of dominant weights such that n together with its ^-conjugates is saturated. Let Mod(7r) denote the full subcategory of Mod(G) with objects having composition factors whose highest weights lie in n. Such a truncated category has both injective and projective modules. For a weight A G 7r, we denote the injective hull and the projective cover of the simple module L(A) by 7^ (A) and Ptt(A), respectively. The module 1^ (A) can be described as the maximal G-submodule of the injective hull /(A) of L(A) in Mod(G) whose composition factors have weights in 7r. In particular, the module /^(A) is finite dimensional and has a good filtration.
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 5 Moreover, for any 7 e n, the multiplicity of the factor H0^) in a good nitration of /tt(A), denoted by [/^(A) : H°(j)]g, equals the multiplicity of the the simple module L(A) as a composition factor in H°(j), denoted by [-ff°(7) : L(7)]g- For a general treatment of truncated categories we refer to [Donl, Don2] or [Janl, II.A]. Next consider the automorphism a on G and Ga(¥q). One obtains for G- modules M and N (2.3.1) Ext2G(M, N) ^ Ext2G(MCT_1, AT-1) for i > 0. The Frobenius morphism Fr is also an automorphism on Ga(¥q) with Frocr = croFr being the identity. Hence (2.3.2) Ext^(Fg)(M, AT) - Ext^^M*-1,^-1) - ExfGff(rJ(tfW,JVW) for i > 0. Moreover, from [Jan2, 1.3] one concludes for A G X(T)+ that L(A)CT_1 ^ L(aX) as a G-module, and L(A)CT_1 ^ L(crA) ^ L(A)(r) as a G^F^)-module. It follows that (2.3.3) P.(A)^"1 * PffW(aA) and [PaM(aX) : L(a7)]c = [P*(A) : L(7)]o. Also (2.3.4) P7r(A)(r) ^ ^(A)*7"1 ^ Pct(tt)(^A) as a Gff(F,)-module. Finally, note that the projective module Ptt(A) is isomorphic to the dual module of I-wo^i-WoX). 3. Cohomological Facts In this section, we record several cohomological results which will be used a number of times later in the paper. 3.1. The following gives a condition under which homomorphisms over Gr may be identified with those over G. Proposition . Assume A,/x e Xr(T) and M is a finite dimensional rational G-module such that all its weights v satisfy (v,oiq) < pr. Then HomGr (L(A), L(/x) ® M) = HomG(L(A), L(/x) ® M). Proof. Without loss of generality (by dualizing if necessary), we may assume that (/x,olq) < (A,Qq). All G-composition factors of HomGr(L(A),L(/x) ® M) are Gr-trivial so must be of the form L("y)^r\ For such a factor, A + £>r7 is a weight of L(/x) <g) M and hence (A + pr7, ctf) < </x + 1/, c#> < (A, 0%) + (i/, ajf) for a weight 1/ of M. Hence, pr(7, Oq ) < (1/, 0$) < pr and so we must have 7 = 0. Therefore Hom<3r (£(A), L(/x) ® M) has a trivial G-structure and the claim follows since HomG(L(A),L(/x) ® M) = HomG/Gr(/c,HomGr(L(A),L(/x) ® M)). D
6 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN 3.2. Ext1 for modules with small highest weights. Here we give an upper bound on the size of of the weights of a G-module M to insure H^G.M). Lemma . Assume that the root system $ of G is not of type A\. Let M be a finite dimensional rational G-module whose highest weights A satisfy (A,ao) < (a) Ifp^2or$ is not of type Cn, then YLl{Gr, M) ^ H^G, M). (b) If p = 2, <£ is of type Cn, and k is not a composition factor of M, then K1(Gr,M)^K1(G,M). (c) If p = 2, $ is of type Cn, and k is a composition factor of M, then T3}{Gr,M) ^ B}(G,M) 0 (0jL(u;i))(r\ where l > 0 and ui denotes the first fundamental weight of $. Proof. Consider the Lyndon-Hochschild-Serre (LHS) spectral sequence E%j = I?(G/Gr,H.j(Gr,M) =» Hi+J'(G,M). If M has a composition factor of the form L("y)^r\ the assumption that (pr7, (*$) < pr implies that 7 = 0. Hence, Home, (A;, M) = Horned, M). Therefore, E1/ = YLl(G/Gr, HomGr (fc, M)) ^ ftl{G/Gr, k) ® HomG(fc, M) = 0 and so tffCM) = E1 * E°2A = Homc/c^H^M)). Thus the isomorphisms in parts (a) and (b) hold if the highest weight of H1(Gr, M) is zero. Since the weight u\ is not contained in the root lattice, part (c) follows if the highest weights of H1(Gr, M) are zero or u)\. By induction on a composition series for M it now suffices to prove the assertion for a simple module L(X). If A = 0, then it follows from [Andl] that xii/^ / \ / L(^i)(r) if P = 2 and * is of type cn H<G" *> = {(, else. For A ^ 0 define the quotient Q via the short exact sequence 0 -> L(A) -> #°(A) -> Q -> 0 and consider a portion of the associated long exact sequence HomGr(fc, L(A)) -+ HomGr(fc, tf°(A)) -+ HomGr(fc, Q) ->H1(Gr,L(A))^H1(Gr,i7°(A)). The size of A forces all composition factors of Q and H°(X) to be ^-restricted. This implies that all Hom<3r in the above sequence can be replaced by Home giving HomG(fc, L(A)) -+ HomG(fc, ff°(A)) -> HomG(fc, Q) ^^(G^LiX^^R'iG^H^X)). The first map is an isomorphism and Horned, Q) = H1(G, L(A)) by [Janl, II.2.14]. It is therefore sufficient to show that H1(Gr,i7°(A)) = 0. It follows from [BNP4, 3.2] that H1(Gr,i7°(A)) = 0 unless A = prv -pla with v e X(T), a e A, and 0 < i < r. Since A is dominant and not zero, we have (u,a^) > 1. For all root systems other than A\ and C2, one has {ol,olq) < 1. Hence, (A,0$) < pr — pr~1 implies H1(Gr,iif0(A)) = 0. If $ is of type C2, the above argument fails in the
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 7 case A = prv — pr 1a with a being the long simple root. However this case is not of interest because A being dominant forces (i/, 0$) > 2 and (pru — pr_1a,a^) > 2pr-2pr-1>pr-1(p-l). □ Remark . Direct computation shows that the Proposition also holds for type A\ and p = 2. For type A\ and odd primes, one obtains H1(Gr,M) = H1(G, M) for all finite dimensional G-modules with highest weights A satisfying (A, 0$) < f-l{p-2). 3.3. Vanishing of certain Ext^-groups. It is well-known that a finite- dimensional G-module admits a good nitration if and only if ExtJ?(Vr(/x), M) = 0 for all ji G X(T)+. Our goal in Section 4 is to show that certain modules have nitrations with factors of the form H°(j20) <S> H°(fii)^ where /x0, pi € T. In order to establish these results, we need the following proposition. Here we set 1^ = —woa(Ti). Proposition . Let pr~1(p — 1) > 4ft —6 and 70,11,^0,^1 £ IV Assume that the root system <£ of G is not of type A\. If p = 2 and $ 25 of type Cn, then we assume in addition that 70 — 71 and fiQ — fii are contained in the root lattice. Then the following hold: (i) ExtG(V(£0) ® VWr), tf°(7o) ® ff°(7i)(r)) = 0. (ii) ExtG(L(£0) ® £(Mi)(r), ^(7o) ® /r<(7i)(r)) = 0. (iii) ExtG(PPi(£o) ® Pr,(Mi)(r), L(%) ® L(7i)(r)) = 0. Proof. We apply the LHS spectral sequence (3.3.1) E? = EXtjJ/Gr(VO*i)(r),EXtir(V(/io),^0(7o))®H0(7l)(r)) (3.3.2) =» Ext^'(y(Mo)®^(Mi)(r),^0(7o)®ff°(7i)(r))- All weights involved are pr restricted. Therefore, EomGr(V(iL0),H°(%)) S HomG(y(Mo),^°(7o)) is either the trivial module or zero. It follows from [Janl, II.4.13] that the E2y -term vanishes. For any composition factor L(X) of V(/zo)* ® i/°(7o), we have (A, Oq) < 2(ft — 1) - 1 + 2(ft - 1) - 1 = 4ft - 6. Hence, by Lemma 3.2, Ext^r(VX/io),#0(7o)) = Ext^(Vr(/xo)5 H°(lo)) unless p = 2, $ is of type Cn, and A; is a composition factor of V(/Io)* ® -ff°(7o)- If we exclude this case, it follows from [Janl, II.4.13] that the E2 -term also vanishes. If p = 2, <£ is of type Cn, and fc is a composition factor of V(/Io)* ®#°(7o) then /jio—wolo is in the root lattice. Therefore, /xi—^o7i = (^o7o_woli) + (no —^o7o)_ (/xo — /xi) is also in the root lattice. Prom Lemma 3.2, Ext^ (V(/Io)j#0(7o)) — Exto(Vr(/Io), #o(7o))e(0zL(wi)(r)). By [Janl, II.4.13], the first summand vanishes and so E°/ S* eiHomG/G^V^OW.L^OW ® H°(7i)(r)). Furthermore, this vanishes because HomG/Gr(y(Mi)(r),L(a;i)W ® #°(7i)(r)) S HomG(L(-™0u;i),tf0(-™oMi) ® #°(7i)) and all weights of H°(—wofJ-i)^ ® i/°(7i) are contained in the root lattice. The assertion (i) follows. Statements (ii) and (iii) follow along the same lines. n
8 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN Remark . We assume that the conditions of Proposition 3.3 are satisfied and have a closer look at (3.3.1). It follows from [Janl, II.4.13] and Lemma 3.2 that ££° = 0 for i > 0 and E^1 = 0 for i > 0. Hence E2 ^ £7°'2. If /i0 = 7o = 7i = 0, then E%2 = HomG/G!r(Vr(/xi)r, Ext Qr(k, k)). The group Ext^r(fc, A;) does not vanish [BNP6] and for appropriate choices of /xi neither does the i?2-term. We conclude that neither Ext&VCpo) ® V(pn)<r\ H°(%) ® ^°(7i)(r)) nor Ext2G(L(flo) ® L(Mi)(r),/Fi(7o) ® /ri(7i)(r)) vanish in general. 3.4. Important submodules for injectives. It follows from the Remark in Section 3.3 that /p. (70) ® ii\(7i)^ is not injective for the full subcategory of modules whose composition factors are of the form L(Vq + prvi) with i/q, ^1 € IV Perhaps this is not surprising since the set of weights {Vq + prv\ \ uq, v\ <E I\} together with its H^-conjugates is in general not saturated. However, under the conditions of Proposition 3.3, the module /p. (70) <g) ii\(7i)^ can be characterized as the maximal submodule of /(Pq + Pr7i) whose composition factors are of the form L{pQ + prv\) with v$, 1/1 G IV This will be shown in the following lemma. Assume that pr > 2(h — 1) and 70,71 € T. Then all composition factors of ir(7o) have ^-restricted highest weight. Therefore, soc<3r(ir(7o)) — £(70) and socGr(/F(7o)®ir(7i)(r)) = £(70)® ir(7i)(r). It follows from [Janl, 11.3.16(2)] that socG(/F(7o) ® /r(7i)(r)) = £(70) ® socG(/F(7i)(r)) = L(7o +Pr7i). Therefore, (3.4.1) /r(7o)^/r(7i)(r) ^ i(7o+Pr7i), and Pn(7o+pr7i) -* i¥(7o)®iM7i)(r). Lemma . Assume that the root system $ of G is not of type A\. Let pr~1(p — 1) > Ah — 6 and Ze£ M 6e a finite dimensional rational G-module whose composition factors are of the form £(70 +pr7i) with 70,7i E IV If P — 2 and $ 25 0/ fype Cn, we assume in addition that 70 — 71 is contained in the root lattice. Then [M : L(7o +pr7i)]c = dim HomG (M, if. (70) ® ii\(7i)(r)) = dimHomG(Ppi(7o)®Pri(7i)(r),M). Proof. The second equality follows easily by using duality. We will proceed to prove the first equality. Clearly, [M : I/(7o+pr7i)]c = dimHomc(M, i(7o+pr7i))- It suffices to prove that dimHomG(M, I(% + pr7i)) = dim HomG (Af, ip.(70) ® iri(7i)(r))- Clearly, for p = 2 and $ of type Cn both Horn-groups vanish unless 70 — 71 is contained in the root lattice. We use induction on the number of composition factors of M. The module ip.(7o) ® ii\(7i)^ is a submodule of /(70 + prJi) (by (3.4.1)) and their socles are simple. The assertion holds therefore for simple modules. Next assume that L(V0 + prvi) is a simple quotient of M. This implies for the case p = 2 and $ of type Cn that vq — v\ is in the root lattice. Define S via
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 9 the exact sequence 0 —> S —> M —> L(u0 + prv{) —> 0. One obtains the long exact sequences 0 -> HomG(L(P0 + prvi)J(% + Pr7i)) -> HomG(M,/(70 + pr7i)) -> HomG(S, /(7o + Pr7i)) -> Ext^(L(P0 + pr*>i), I(% + Pr7i)) and 0 - HomG(L(P0 +Pr^i),/Pi(7o) ® /r,(7i)(r)) - HomG(M,/fi(70) ® /ri(7i)(r)) -> HamcOS, JPi(7o) ® /ri(7i)(r)) - Ext^(L(P0 +Pr^),/Pi(7o) ® /r,(7i)(r)). Obviously, Ext^(L(P0 +^r^)5^(7o + Pr7i)) = 0- Moreover, Proposition 3.3 shows that Ext^(L(Po + Prv),If . (7o) ® ^i\(7i)^) = 0- From the induction hypotheses one concludes that dimRomG(L(V0+pri/), I(%+pr7i)) = dim Home (L(V0+prv), 7p. (70)® 7^(71 )(r)) and dimHomG(S', /(70 +Pr7i)) = dimHomG(S', 7p.(7o) ® /ri(7i)(r))- Hence, dimHomcCM, /(7o+pr7i)) = dimHomcCM, /f .(7o)®^ri(7i)^) as claimed. □ 4. A Filtration of Gn(k) The tensor identity says that Q(Str) = Str <8>G(k). Once we pass to truncated categories, such an identity no longer holds in general. However, for pr > 4(ft—1) we will show that (?A(Str) = Str <8>Gn(k) (Proposition 4.4). Furthermore, a complete description of the module £/A(Str) will be given (Theorem 4.3). This will allow us to determine the character of Gn(k). The ultimate goal of this section is to study the nitration of Gnn(k) given in (2.2.1) in order to identify the factors and show that Gnn(k) = Gn(k) for sufficiently large pr (Theorem 4.7). 4.1. Composition factors of Gn(k). Since — wq and a permute the fundamental weights, any weight 7 G X(T)+ can be expressed uniquely in the form 7 = — woa^o + pr7i = 70 + £>r7i with 70 G Xr(T) and 71 G X(T)+. One can now use the methods in [BNP5, Prop. 2.5] to prove the following result. Proposition . IfL(%+prji) is a composition factor ofGn(k) then^o^li £ r. 4.2. The following results will help us relate Gn(k) to (?A(Str) and understand (?A(Str). Proposition (A). Let pr > 4(ft - 1) and i/0, v\ e r. (a) Str ®L(i>b) ® L(^i)(r) G Mod(A); (b) Str(g)L(i/0)(8)/rK)(r) GMod(A); (c) Str ®Gn(k) is a submodule o/£/A(Str). Proof, (a) From i^eT one concludes that (i'i, 0$) < 2(ft— 1) — 1. Therefore, <(pr-l)/»+H)+pri/i,a£> < (pr -l)(h-l) + 2(h-l) +pr2(h-l)- pr = 3(pr-l)(h-l) + 4(h-l)-pr < 3(pr-l)(h-l).
10 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN Part (b) follows immediately from above. Proposition 4.1 and part (a) imply that Str <g> Gn(k) e Mod(A). Part (c) now follows from Str <g> GQ(k) C (Str <g> G(k))A = f?A(Str). □ Proposition (B). Let pr > 4(ft - 1) and i/0, v\ e I\ Then (a) Str (8)L(i/q) ® Ir(vi)^ is an injective module in Mod(A). (b) socG(Str (g)L(i/0) ® /r(^i)(r)) = socG(Str <g>L(i/0)) ® £K)(r) Proof, (a) Let 7 = 70 + prji be a weight in A. It suffices to show that Ext^(L(7), Str ® L(i/0) ® /r(^i)(r)) = 0 for all such 7. Consider the LHS spectral sequence E? = ExtG/Gr(L(7i)(r), ExtGr(L(7o), Str ® L{vQ)) ® /r(^i)(r)) =► Ext*t'(L(7o) ® L(7i)(r), Str ® L{v0) ® IT{yi){r)). All i?*'-'-terms with j > 0 vanish because Str is injective as a Gr-module. Thus ^i ^ si,o ^ ExtG/Gr(L(7i)(r),HomGr(L(7o), Str ® Lfo)) ® Jr>i)(r)) s ExtG/Gr(L(7i)(r), /r(^i)(r)) ® HomG(L(7o), Str ® Lfo)). The last isomorphism is a consequence of Proposition 3.1. For Home (L(70), Str <g) L(i/o)) to be non-zero, it is necessary that (70 — wo^o,&o) ^ 0?r ~~ 1)(^ ~~ !)• This forces (70, c#) > (pr - l)(ft - 1) - 2(ft - 1) and (pr - l)(fc - 1) - 2(^-1) + pr(7i,c#> < <7lc#> < 3(pr - l)(fc-l). One concludes that pr(7i,a:o) < 2pr(h — 1) and 71 e I\ Hence, by the injectivity of /r(i/i), Ext^/Gr(L(7i)(r),/r(^i)(r)) vanishes. (b) Let 7 = 70 + pr7i be a weight in A. Prom Proposition 3.1, one concludes that HomG(L(7), Str ® L(v0) ® /r(^i)(r)) ^ HomG/Gr(L(7i)(r), HomGr(L(7o), Str ® L(i/0)) ® Jr(i>i)(r)) ^ HomG/Gr(L(7i)(r), /r(^i)(r)) ® HomG(L(7o), Str ® L(v0))- The assertion follows. □ 4.3. We can now provide a description of £/A(Str) for pr sufficiently large. Theorem . Let pr > 4(ft - 1). Then gA(Str) ^0Str®L(P)®/F(i/)(r). Proof. Both modules are injective in Mod(A). (The left-side since Str is injective over G(¥q) and the right-side by Proposition 4.2(B) part (a).) It suffices therefore to show that both modules have the same G-socle. Let 7 = 70 -f prji be a weight in A. dimHomG(L(7), SA(Str)) = dim Homo (L(7o) ® £(7i)(r), Str ® G(k)) = dimHomG(Fg)(L(7o) ® L(aji), Str) (by adjointness) = [L(7o) ® L((Tji) : Str]G(F,) A*€A-(T) +
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 11 where the last equality follows from [Jan2, Satz 1.5]. The above expression is zero unless (4.3.1) <7o +7i,<#> > (Pr ~ !)(&- 1) + (Pr ~ l)(M,«oV)- Moreover, 3(pr-l)(fe-l) > (7o+pr7i,a0V) > (pr-l)(ft-l) + br-l)(M,a0v) + (p'--l)(7i,a0v). The last inequality implies that (71,0$) < 2(ft — 1). Hence, any weight 7 that appears in the socle of (?A(Str) has 71 contained in I\ It follows now from (4.3.1) that 2(ft - 1) > (pr - l)(/i,c#). Together with pr > 4(ft — 1) this forces \i — 0. One concludes that, if L(j) is a composition factor of £A(Str), then dimHomG(L(7), £A(Str)) = [L(7o) ® L(a7l) : Str]G. Using Proposition 3.1 and the fact that Str is injective as a Gr-module, one observes that dimHomG(L(7o) ® L(<ryi) : Str) < [L(j0) ® £(<ryi) • Str]G < [L(7o)0L(a7i):Str]Gr = dimHomGr(L(7o) ® L(aji) : Str) = dim Home(L(7o) ® L(aji) : Str). Therefore, dimHomG(L(7),^A(Str)) = dim Home(L(70), Str ® L(-w0a^i)) = dimHomG(L(7o), Str ® L(7i)) and socG(£A(Str)) * 0 HomG(L(7o), Str ® Lfyi)) ® L(7o) ® £(7i)(r) 7o€Xr(T),7i€r ^ 0 socG(Str ® L(fi)) ® L(7i)(r) 71 er = 0socG(Str ® L(P) ® /i»(r)) by Proposition 4.2(B) part (b). □ 4.4. A "tensor identity" and the character of Qn(k). Proposition . Letpr > 4(ft- 1). Then GA(Str) * Str®gn(k) and chgQ(k) = ]Tch(tf°(A) ® H°(X)^). xer Proof. Prom Proposition 4.2(A), we know that Str®(/n(fc) C <?A(Str). We will show here that equality holds. Prom Theorem 4.3, we know that the formal character of Qn(k) is a summand of ch(0I/€F L(p) <g) Ir(v)^) and, by Proposition 4.1, all composition factors of Gn(k) are of the form L(7b)®L(7i)^ with 70,71 G T.
12 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN Note that since cr(ao) — ao and — Wo(oto) = ao, o-(T) = T = — wq(T). We have [Gn(k):L(f0+pr^i)]G = dim HomG (Pq (70 + Prli),Gn{k)) > dimHomG(Pr(fo)®Pr('n)ir\Gn(k)) (by 3.4.1) = dimHomG<T(Fg)(Pr(fo) ® Pr(7i)(r\ *) (by adjointness) = dimHomGCT(Fg)(Pr(7o) ® ft(^71), *) (by 2.3.4) = dimHomG(ft(7b) ® ft(^7i), &) (by [CPSvdK, Thm. 7.4]) = dimHomG(ft(cr7i), 7r(^"7o)) (by duality) = dimHomG(PF(7i), /r(7o)) (by 2.3.3) = [M7o) • L(li)h = [(L(fo) ® M7o)(r)) : L(fo) ® L(7i)(r)]c = [(©i(5) ® /rW(r)) : L(fo +Pr7i)]o- It follows that £A(Str) ^ Str (g) Gn(k). Finally, ch(0L(P)®Jr(i/)(r)) - ]Tch(L(P)®/r(i/)(r)) = £ Dw: #°(A)] ch(m ® fl°(A)W) = £ £#° w: L^ ch(L(p) ® #°(A)(r)) = ^ch(i7°(A)0i7°(A)^). D We immediately get the following where f is the ordering on X(T) as given in [Janl, II.6.4]. Corollary . Let pr > 4(ft — 1). // L(70 + pr7i) 25 a composition factor of Gn{k), then there exists AeT such that 70 T ^ and 7i T A. Remark . An easy computation shows for pr > 4(ft — 1) that Hn(k) = ©i,€X(T)+(/r(^)(r))dimL(l/) and WA(Str) ^ Str ® Wn(Jfc). One concludes from Theorem 4.3 and Proposition 4.4 that dimftA(Str) = dimSa(Str) and dimHn(k) = dimGQ(k). 4.5. Good £>r-filtrations and Donkin's conjecture. Let M be a finite dimensional G-module. We say that M has a good pr-filtration if and only if there exists a sequence of submodules 0 = Mo C Mi C M2 C • • • C Ms = M such that Mi/Mi+1 * L{i4) ® fr°0*J)<r> where ^ = $ + pr»] e X{T)+ with /x? e Xr(T) for all i The definition for the case when r = 1 was first introduced by Donkin in 1990 (i.e. the notion of a good ^-filtration). Donkin conjectured that if M is a finite dimensional rational G-module, then M has a good ^-filtration if and only if M <g) Sti has a good nitration. Andersen [And2, Cor. 2.8] proved one direction of
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 13 a generalization of this conjecture for large primes (p > 2 (ft — 1)), namely, given a finite-dimensional rational G-module M with a good pr-filtration then M ® Str has a good nitration. In the same paper, it is also shown that the other direction holds for G = SL2(k). Since Str is injective over Ga(¥q) and A is saturated, £/A(Str) has a good nitration. For pr > 4(ft — 1), from Proposition 4.4, it follows that Str <g) Gn(k) has a good nitration. The validity of the generalized version of Donkin's conjecture would imply that Gn(k) has a good pr-filtration. One can see some indication of this in Proposition 4.4. Indeed, this provides the motivation for formulating and proving Theorem 4.7 which demonstrates in the case when M = Gn(k) and pr~1(p — 1) > 4ft — 6 that the generalized version of Donkin's conjecture holds. 4.6. The G-socle of Gn^k). We now determine the socle of each Gn^k) for pr sufficiently large. Proposition . Let pr > 2(ft - 1) and i/0, v\ G Xr(T). Then u /T/z-NoT// \(r) r n,\\~jk if vo = "i and vie Tu Proof. Set N = V(V0) ® V(vi)M. Then HomG(N,gni(k)) C HomG(iV,indgCT(Fg)(A;)) = HomG<T(Fg)(A/r, k) (by adjointness) = HomGa(Fq)(Vr(Po) ® ^(tn/i), k) = HomG<T(Fg)(Vr(ai/i), iJ°(<n/0)) = HomG(V(wi), H°(ai/0)) (since pr > 2(ft - 1)) ■! A; if i/o = ^i 0 else. If 1/0 = 1/1, then a non-trival homomorphism will exist precisely if (pr — wqg)u\ = Po + prv\ G f^i. That occurs if and only if i/q = v\ G T^. □ Corollary . Le£pr > 2(ft - 1). Tften socGGni(k)=@L(V + pri/). ^gi\ Proof. Any simple G-module L(v + pri/) in the socle of £7^ (A;) gives rise to a non-trivial homomorphism from V(j£) <S> V(i/)^ to (7^ (A;). The result now follows from the proposition. □ Remark . Prom Proposition 4.4, Corollary 4.6 (for pr > 4) and our earlier results in [BNP2] (for p = 3) we conclude the following: (4.6.1) If G is of type Ax and pr > 2, then Gn(k) ^ k 0 L(pr + 1).
14 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN 4.7. Filtrations of GnAk) and Gn(k). In this section, we show that Gn(k) admits a natural nitration. Theorem . Let pr~1(p - 1) > 4ft - 6. Then Gn(k) has a filtration 0 = Gn0(k) c gQl(k) c GnM Q • • • c Snn(*) = Sn(*) with factors Gnj(k)/Gnj_1(k) = H°(\j) ® H°(Xj)(r\ each Xj G T appearing exactly once. Proof. The submodules GnAk) were denned in (2.2.1). Using induction on i, we will show that GnAk)/^n^Ak) - H°fti) ® H°(\i)(r\ For i = 1 one has nx = Ti = {0} and GnAk) ~k = H°lo)®H°(0)(r\ Assume i > 1. By Proposition 4.1 and Corollary 4.4, all composition factors of GnAk) C Gn(k) are of the form L(Vo+pri/i) with i/o5 ^i £ T and v§ — v\ in the root lattice. We first show inductively on i that in fact i/q, v\ G IV Note that this clearly holds for i = 0. The Theorem holds for type A\ by (4.6.1). For the remainder of the proof we may therefore assume that G is not of type A\. We apply Proposition 3.3 to conclude that Ext^VXPo) ® V[yx)^,H°(\j) ® H°(Xj)^) = 0 for i/0,i/i, A, G T. The induction hypothesis implies that Ext^V^Po) ® V^ijW,^.^!!;)) = 0. Therefore the short exact sequence 0 - Gn^Ak) - 0n<(*) - 0n,(*)/Sn,-i(*) - 0 gives rise to the exact sequence 0 -> HomG(V(P0) ® V(,t>1)<-r\gni_l(k)) - HomG(y(P0) ® V(i>i)(r),0n,(*)) -> HomG(V(P0) ® V(i^)W, Sni(*)/Sni_1(*)) -» 0. One concludes from Proposition 4.6 that HomG(F(P0) ® lWr^n,(*0/Sn,-i(*)) = The module V(Ai) ® V(Ai)^ has simple G-head L(A; + prA;) and one concludes that the G-socle of Gni(k)/GQi_1(k) is isomorphic to L(A; + prA;). Next we embed GnAk)/^i-i(k) m the injective hull /(Ai + prA;). The module /(A; +prA;) has a good nitration with factors H°(^y) with (pr — iuo0")Ai f 7. Here f is the ordering on X(T) as given in [Janl, II.6.4]. Clearly the only such 7 that is contained in f^ is (pr — w0o-)Xi. One obtains an embedding GnAk)/^^i-i(k) ^ H°((pr — wo<r)Xi). Therefore all weights 7 in GnAk)/^i-i(k) satisfy 7 | (pr — wo0")A;. From above, the multiplicity [^(A;) : L(Xi+prXi)]G is one. We apply Lemma 3.4 and conclude that Gni(k)/Gni_1(k) also embeds in If(Xi)<S>Ir(Xi)(r\ The module fy(Xi) ® 7r(Ai)(r) has a filtration with modules of the form H°(%) ® #°(7i)(r) such that Ai f 70,71. Each of these factors has simple socle with highest weight 7o + Pr7i > (pr — WQ(r)Xi. The only such weight in Fli is (pr — woo~)Xi. It follows that (4.7.1) GnAk)/Gn^Ak) ^ H°(%) ® H°(X^rK From the induction hypothesis we conclude that all composition factors of GnAk) are of tne form ^(7o + Pr7i) with 70,71 G IV I A; if uq = v\ 10 else. Ai,
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 15 For the remainder of the proof we set P = Pp (70) ® Pr^Ti)^- The module P has a nitration P = Pm 2 Pm-i 2 • • • 2 Pi 2 Po = 0 with factors of the form V(V0) ® V(i'i)^. Consider the LHS spectral sequence: El2'j =E^h/Gr(V(u1)^,ExtiJr(y(p0),V(jio))<8> V(/i!)W) =* Extjt'(V(%) ® ^(^i)(r), V(/io) ® V(/n)(r)). Note that £,2' =0 unless uq < //o> and from Lemma 3.2, i?2' =0 unless v\ < fi\. Hence Ext^V^Po) ® V(ui)^\ V(j20) ® V(/Jii)(r)) = 0 unless i/0 < /i0 and vx < /ii. We can rearrange the above nitration such that, for a certain 1 < I < m, Pi G Mod(f^) while the kernel S of the projection P -» Pi has all its factors V(90) <g> V(ui)^ outside of Mod(f^). Observe that HomGCT(Fg)(F(Po)0^i)(r),A;) HomG<T(Fg)(Vr(P0) ® ^(tn/i), &) HomGCT(Fg)(F(P0),^°(Pi)) HomG(y(P0),ff°(Pi)) = 0, unless i/o — ^i G IV This implies that P0 + Prv\ = (pr — wqg)v\ G fV One concludes that Rome(S, Qn^k)) <-+ HomoiS, indgCT(Fg) fc) = 0. Prom the two long exact sequences for Hoihg(-,Gsii(k)) and Homc(—,indGff^F ) k) associated to 0 —> 5 -> P -> P^ -> 0, this forces (4.7.2) HomG(P,^W) ^ Hom^, &,,(*)) (4.7.3) ^ HomG(P,, indgCT(Fg) *) ^ HomG(P, indgCT(Fg) *). Let L(7o +pr7i) be an arbitrary composition factor of Qo,i{k). Recall from above that 70,7i G I\ and 70 — 71 is in the root lattice. The preceding observations HomG(F(P0) ® m)(r\ mdgCT(Fg) *) ^ rs^
16 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN allow us to argue as follows: [an,(*):£(70+Pr7i)]G = dimHomG(P, Ga^k)) (by Lemma 3.4) = dimRomG(P,md%A¥q)(k)) (by 4.7.2-4.7.3) = dim Hom^ (f )(P, k) (by adjointness) = dimHomGCT(Fij)(Pp.(7o), Jf.(7i)) (by duality and 2.3.4) = dimHomG(PPi(70), 1^(70) (by [CPSvdK, Thm 7.4]) = [/rv(7i):£(7o))]G (by 2.3.3) = £ ([#» : L(7l)]G • [#» : L(7o)]g) (see Section 2.3) = £ ([tf»(r) : L(7i)(r)]G • \H\u) : L(%)]G) = ]T [tf°(P) ® fl»M : L(7o + pr7i)]G- For the last equality one makes use of the fact that all composition factors of H°(v) are ^-restricted. One concludes that chGntik) = ]T ch(H°(V)®H°(v)^) and chgnt(k)/Gnt_Ak) = M&fo) ® ^(A,)^). The embedding (4.7.1) is therefore an isomorphism and the assertion follows. Finally, Proposition 4.4 implies that Qn(k) = Gnn(k). □ 5. Ga(¥q)-Extensions In this section, we apply the nitration of Qn(k) obtained in Theorem 4.7 to make computations about extensions between simple GCT(Fg)-modules. 5.1. The following theorem involves a minor modification of results proved in [BNP2, Thm. 2.2] and [BNP5, Thm. 2.3]. The proof can be easily adapted to our situation. Theorem . Let A,/xe Xr(T). Then Ext^(Fg)(L(A), L(fj)) * Ext^(L(A), L(/x) ® gn(k)). 5.2. Theorem 4.7 implies that Qn(k) has a nitration with factors of the form H°(V) <S)H°(u)^ for v e I\ In order to apply Theorem 5.1 and obtain information about Ext^/F n(L(A),L(/x)), we investigate the Ext-groups Ext^(L(A), L(/x) ® H°(rj) ® ^°(^)(r)) = Ext^(L(A) ® F(-™077)(r), L(/x) ® H°(f})) * Ext^(L(A) ® V{v){r\ L(/x) ® tf0^)), where i/ = —worj. The following lemma says that for these groups to be non-zero, we must in fact have v e Th = {y e X(T)+ | (i/, o%) < h}.
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 17 Lemma . Let A,/i e Xr(T) and v e X(T)+ with {v,o%) < pr. If p = 2, assume further that pr > 4. // Ext^(L(A) <g> V{v)^r\ L(/x) <g> H°(au)) ^ 0, then (^, &o) < (h — 1). (In the excluded cases for p = 2, one can replace h — 1 by h.) Proof. Consider the LHS spectral seqeuence E? = Exth/Gr(V(^r\Extir(L(X),L(/i) ® H°{av))) => Ext^+J'(L(A) ® V{v)(r\ L(ji) ® H°{av)). Notice that, by Proposition 3.1, E\fi = ExtG/Gr(l»(r), Jfc) ® HomG(L(A), L(p) ® H°(av)). Hence, E^'0 = 0 by [Janl, II.4.13] and so E2 S E"'1 S HomG/Gr(y(i/)W,ExtGr(L(A),L(M) ® ff°(<n/))). Let pr7 be a weight of ExtGr(L(A),L(/x) ® H°(a(v))). It follows from the argument in [BNP4, 5.2]) that (5.2.1) pr(1, c%) < (A, c%) + (n, a0v) + {au, a%) + 3pr~\ Consider the short exact sequence 0 —> L(/x) —> Str (g) L((pr — l)p + wo/x) —► -R —> 0. Using the long exact sequence in cohomology and the fact that Str is injective as a Gr-module one obtains a surjection HomGr(L(A), R ® H°(o~(v)) -» Ext^r(L(A), L(/x) 0 ff°(cr(i/))). Hence, any weight pr^ of Ext^r (L(A), L(/x) (g i7°(cr(i/))) also satisfies (5.2.2) pr(j, a0v) < 2(pr - l)(h - 1) - (A, <#> - (M, ^) + <<«/, <#>. Adding equations (5.2.1) and (5.2.2) and dividing by two yields (5.2.3) pr(7,a0v) < (pr-l)(ft-l) + (<«',c#> + fpr~1 (5.2.4) = (au, a0v) + (pr - l)h + 1 - pr (l - A) . Replacing 7 by 1/ results in 0/ - 1)<„, c#> < (pr -l)h+l-pr(l-^)< (pr - l)h and the assertion follows. □ 5.3. We can now use Theorem 5.1 to show that Ext^F )(L(A), L(/x)) embeds as a submodule of a certain direct sum of G-extensions. The conclusion is not as strong as in [BNP2, 2.5,3.2] (where equality was shown to hold), however the assumption on the prime is much less restrictive here. Moreover, this result is sufficient to obtain several nice applications which will follow. Let M be a G-module with a filtration 0 = M0 C Mi C M2 C • • • C Mi = M. Then one can argue inductively that for all N e Mod(G) and i > 0 1 (5.3.1) dimExt2G(iV,M) < ^dimExt^A^Mi/M^i). 2=1
18 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN Proposition . Let pr~1(p - 1) > 4ft - 6, A, \i e Xr{T), and T'h = Th- {0}. (a) Assume r > 1, then Ext^(Fq)(L(A),L(/i)) - ExtG(L(A),L(M)) efl, w/iere i? ^ 0 Ext^(L(A) ® V{v){r\ L(/x) ® ff0(cri/)). * 0 HomG/Gr(FH(r),Ext^(L(A),L(/x) ® H°(av))). (b) Assume r > 2 and Ze£ s = [§]. Assume further that ps > h. Set A = Ao +ps\i and ji = no + psH\ with Ao,/xo £ ATS(T) ana? Ai,//i G Xr_s(T). T/ien we ma?/ reidentify R as 0 (Ext^(L(Ai) ® F(i/)(r"s), L(^i)) ® HomG(L(A0), L(/x0) ® #°(<n/))) = 0 (HomG(F(i/)(r-s), Ext^r_a (L(Ai), L(^))) ® HomG(L(A0), L(/x0) ® H°(cri/))). Proof. Prom Theorem 4.7, (fo(A;) has a nitration with factors of the form H°(V) ® H°(u)^ where each v G T appears exactly once. By Theorem 5.1, Lemma 5.2 (and the remarks), and (5.3.1), we have dimExt^(Fg)(L(A),L(/x)) = dimExt^(L(A),L(/x) ® gn(k)) < J2 dimExt^(L(A)®y(i/)W,L(^)®fr°(cr(i/))) »erh = dim Exto(L( A), L(/i)) + ]T dimExt^(L(A) ® F(i/)(r), L(/x) ® H°(a(v))). ^€i\-{o} Since all modules involved are trivial as G-modules, the claimed embedding follows. Consider the LHS spectral sequence B? = Exth/Gr(V(v)(r\Ext^r(L(\),Lfa) ® H°(au))) =» Ext2ctj(L(A) ® F(i/)^,L(/x) ® H°(av)). As in the proof of Lemma 5.2, it follows that E^° = 0 and hence E2 = E^'1, which gives part (a). Now assume r > 2. Without loss of generality (by dualizing if necessary), we may assume that (/xi, a^) < (Ai, Oq ). We use the LHS spectral sequence E? = Extb/c.^CAOW ® V»(r), ExtGs(L(A0), L(Mo) ® HX*))) ® £(/*i)w) =► Ext^'(L(A) ® y(t/)(r),L(/x) ® fl°(<r(i/))). It follows from (5.2.4) that any weight psj of ExtGa(L(A0),L(/io) ® H°(a{y))) satisfies p'fr.Oo) < (p» - \)h + 1 - £ + <i/,c#> < pr - fp» + 1 + <i/,<#) < pr - 1 + (i/,a^>. On the other hand, if i/ ^ 0, the module L(Xi)^ ® F(t/)<r) has simple head with weight ps\i +prv. Comparison of weights forces E0'1 = 0.
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 19 Therefore, by Proposition 3.1, Ext^(L(A) ® V{v){r\ L(/x) ® H°(av)) ** E1/ = Ext^/Ga(L(Ai)W ® y(i/)W,HomGa(L(A0),L(M) ® ^°H) ® £(^i)W) ^ Ext^/Gs (L(Ai)W ® K(i/)<r\ L(/xi)(s)) ® HomG(L(A0), L(w) ® # V*)) ^ Ext^(L(Ai) ® F(i/)(r"s), L(w))W ® HomG(L(A0), L(/x0) ® ff°(<">)), which gives the first reidentification. To investigate ExtG(L(Ai) <g) Vr(i/)^r~s\ L(/xi)), consider the LHS spectral sequence E? = Extj,/Gr_.(V(i/)('-),Ext^_j(L(A1),L(/i1)) =*- Ext^'(L(Ai) ® V(i/)(r->,L(Aii)). As before, using Proposition 3.1 (with M = fc), we see that E2y = 0 which gives the result. □ Remark . Note that the condition on the prime given in part (b) is almost always stronger than the initial assumption of the proposition. If r is odd and greater than one, then the assumption that ps > h implies pr~1(p — 1) > 4ft — 6. Indeed, we have 4ft - 6 < 4ft - 6 + (ft - 2)2 = ft2 - 2 < ft2 < (ps)2 = p1"-1 < pr-\p - 1). For r even, it is a bit more involved but straightforward exercise to show that the condition ps > ft implies pr~1(p — 1) > 4ft — 6 as long as p > 3. If p = 3, the implication fails only when r = 2 and ft = 3. If p = 2, the implication fails only when ft = 2,3,4,5, or 6. Corollary . Assume r > 2 and Ze£ s = [§]. Assume that ps > ft and Pr_1(p "~ 1) > 4ft — 6. Given A, /x G ATr(T), Ze£ A = Ao + psX\ and /x = /xo + ps/^i W2*£ft Ao,/xo £ XS(T) and Ai,/xi G Xr_s(T). // either of the following conditions hold: (a) Ext^_s(L(A1),L(/x1))=0 (b) HomG(L(A0), L(/x0) ® H°(av)) = 0, then Ext^(Fg)(L(A), L(/x)) ^ Ext^(L(A), L(/x)). Proof. If either condition holds, then it follows from part (b) of the proposition that the remainder term R is zero and hence ExtG/F n(L(A),L(/x)) embeds in ExtG(L(A), L(fi)). On the other hand, from [CPSvdK], we know that the restriction map ExtG(L(A),L(/x)) —> ExtG(F )(L(A),L(/x)) is an embedding. Hence it must be an isomorphism. □ 5.4. Self-extensions for small primes. The following theorem improves on [BNP2, Thm. 3.4] because it can be applied to primes which are smaller than 3(ft — 1). First, we need an observation to deal with the special case of type Cn when p = 2. Lemma . Letp = 2, $ be of type Cn, and A G Xr(T). Then Ext^r(L(A),L(A)) is either zero or isomorphic to N^, where N is a G-module whose weights are not contained in the root lattice.
20 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN Proof. Ext^r(L(A),L(A)) embeds in Ext^r(L(A),#°(A)). By [Janl, II.12.8], Ext;^(L(A), H°(X)) ^ indg(Ext^r (L(A), A)). Assume that Ext^r (L(A), A)) + 0. It follows from [Andl] (see also [Janl, II.12.1 -12.5]) that the J3-socle of Ext£r(L(A), A) is isomorphic to \otn where an denotes the unique last simple root of <£. Hence, the weights of Ext^r(L(A), L(X)) are not in the root lattice. □ Theorem . Assume r > 2 and let s = [§]. Assume ps > h and pr~1(p — 1) > 4/i-6. Then Ext^(Fg)(L(A),L(A)) = 0 forallXeXr(T). Proof. Let A = A0 +psAi with A0 G XS(T) and Ai G Xr-S(T). If p + 2 or $ is not of type Cn, since ExtGr_s(£(Ai), £(Ai)) = 0, it follows from Corollary 5.3 that Ext^(Fg)(L(A),L(A)) ^ Ext^(L(A),L(A)) = 0 as claimed. lip = 2 and $ is of type Cn, by Proposition 5.3, it suffices to show that for all v e Th - {0} HomG(V(v)(r-s\ Ext^r_s (L(Ai), L(Ai))) ® HomG(L(A0), L(A0) ® H°(av)) = 0. HomG(-t/(Ao),L(Ao) ® H°(au))) ^ 0 implies that cri/ and hence i/ are contained in the root lattice. On the other hand, HomG(Vr(i/)(r-s),Ext^rs(L(Ai),L(Ai)) ^ 0 forces v to be outside the root lattice by the above lemma. □ 5.5. Cohomology for small primes. The following shows that the first GCT(Fg)-cohomology with coefficients in a simple module can be described in terms of the first G-cohomology with coefficients in a simple ^-restricted module, provided r is sufficiently large. No condition on the prime is necessary. Theorem . Assume r > 2 and let s = [§]. Assume ps~1(p — 1) > h. Given X G Xr(T), let X = A0 +psAi with A0 G XS(T) and X1 G Xr-S(T). Define X = aXi+pr-sX0. Then H(G.(Fg),L(A)) = |Hl(^L(A)) dse Proof. Note that if ps~1(p - 1) > h, then ps > h and pr~1(p - 1) > Ah - 6 so the results in 5.3 may be applied. The Probenius map is an automorphism on GCT(Fg) and L(A)(r-^ s L(A) as a GCT(F,)-module. Therefore, H1(GCT(Fg),L(A)) ^ H1(Gff(F,),L(A)<r-*>) ^H1(GCT(Fg),L(A)). Since tfidiA)) ^ H1(GCT(Fg),L(A)) and tffdiA)) ^ H^G^F,),!^)), one concludes that H1(GCT(Fg),L(A)) = 0 implies H^G^A)) = H^G^A)) = 0. Assume that H1(G,L(A)) ^ 0. If A0 £ Th - {0}, then HomG(fc,L(A0) ® H°(au)) S Homo(V(i/),L(Ao)) = 0, for all i/ e I\ - {0}. Corollary 5.3 (part (b)) now implies that H1(G<T(Fg),L(A))^H1(G,L(A)). If A0 e Tft - {0}, then we apply Proposition 5.3 to H1(G„(F9),L(A)). Notice that r — s>s and the same argument as in the proof of Proposition 5.3 yields that H^G^F,), L(A)) — H^G, L(A)) e A
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 21 where R= 0 RomG(V(v)(s\ R\GS, L(A0))) ® HomG(fc, L(crAi) ® # °M))- i^erh-{o} It follows from Lemma 3.2 that H^G^L^o)) = H^G,!,^)). Hence, HomGlVMf'U1!^,^)) = 0 for all i/ e I\ - {0} and H^G* (Fg), L(A)) =* H^G, L(A)), as claimed. n 5.6. Extensions between simple modules for small primes. The following theorem generalizes [BNP5, Thm 3.2(a)] to arbitrary primes but large prime powers. Theorem . Assume r > 3 and let s = f11^]. Assume ps > h. Given A,/x e Xr(T) such that let X = X^i=o P%^i and I1 — E[=o P%^ w^ ^/^ ^ -^lCO- Then there exists an integer 0 < n < r such that Ext^(Fq)(L(A),L(M)) <* ExtJ,(L(A), £(£)), where n—l 1—1 A = ^pV(Ai+r_n)+^A_n6lr(T), i=0 i=n n—l i—1 2=0 i=n Proof. Note that that ps > h implies pr~1(p — 1) > 4/i — 6. Indeed, we have 4/i - 6 < 4/i - 6 + (h - 2)2 = h2 - 2 < h2 < (ps)2 = p2s < p*-1 < pr~1{p - 1). Thus the results in 5.3 and 5.4 may be applied. If A = \i the claim follows from Theorem 5.4. Assume A ^ /x. Then there exists 0 < I < r with Xi ^ in. If I < s we set n = s — I and for I > s we set n = r + s — Z. As before we note that L(A)(n) = L{X) and L(/x)(n) = L(/x) as GCT(Fg)-modules. Therefore, Ext^(Fg)(L(A),L(/x)) ^ Ext^(Fg)(L(A)W,L(/x)W) ^ Ext^(Fg)(L(A),L(/x)). Moreover Xs = Xt ^ & = Jis. Set A' = YZoP^ A" = EZl+if-8-1^, and // = ££oVft> //' = A = A' +paA« + ps+1A" and Ji = // + pafJLl +ps+V". Since s = l11^] implies r — s — 1 > 5, we can use the same arguments as in 5.3 to conclude that Ext^(F )(L(A),L(//)) ^-> Ext^(L(A),L(/x)) ® i?, where i? is isomorphic to 0 (Ext^(L(A,,)0F(i/)(r"s-1\ L(/x,,))0HomG(L(A,+^AO, L(//+p>)®#°M)) with T^ = I\ — {0}. Prom Proposition 3.1, one obtains HomG(L(A' + psXt), L(// + p'^) ® H°(au))) <= HomG/Gs(L(AO(s),HomGs(L(A,),L(/x,) ®H°(av)) ® L(/xO(s)) <= HomG(L(A0, L(/xO) 0 HomG(L(A'), L(//) 0 ff°(cri/)). Now A; ^ /x* forces i? = 0 and the assertion follows. □
22 CHRISTOPHER P. BENDEL, DANIEL K. NAKANO, AND CORNELIUS PILLEN As a corollary we can show that for sufficiently large r the dimension of Ga(¥q)- extensions between simple modules is bounded by the dimension of G-extensions between simple ^-restricted modules. Corollary . Assume r > 3 and let s = i11^]. Assume ps > h. Then max{dimfcExt^(Fg)(L(A),L(/x)) | A,/x e Xr(T)} = max{dimfcExt^(L(A),L(/x)) | A,/x e Xr(T)}. References [Andl] H.H. Andersen, Extensions of modules for algebraic groups, Amer. J. Math., 106, (1984), 498-504. [And2] H.H. Andersen, p-Filtrations and the Steinberg module, J. Algebra, 244, (2001), 664- 683. [BNP1] C.P. Bendel, D.K. Nakano, C. Pillen, On comparing the cohomology of algebraic groups, finite Chevalley groups, and Frobenius kernels, J. Pure & Appl. Algebra, 163, no. 2, (2001), 119-146. [BNP2] C.P. Bendel, D.K. Nakano, C. Pillen, Extensions for finite Chevalley groups I, Adv. Math., 183, (2004), 380-408. [BNP3] C.P. Bendel, D.K. Nakano, C. Pillen, Extensions for finite Chevalley groups II, Trans. Amer. Math. Soc, 354, no. 11, (2002), 4421-4454. [BNP4] C.P. Bendel, D.K. Nakano, C. Pillen, Extensions for Frobenius kernels, J. Algebra, 272, (2004), 476-511. [BNP5] C.P. Bendel, D.K. Nakano, C. Pillen, Extensions for finite groups of Lie type: twisted groups, in Finite Groups 2003, eds. Ho et al, de Gruyter, New York, 2004, pp. 29-46. [BNP6] C.P. Bendel, D.K. Nakano, C. Pillen, Second cohomology groups for Frobenius kernels and related structures, preprint, 2004. [Car] R. W. Carter, Finite Groups of Lie Type: Conjugacy Classes and Complex Characters, John Wiley and Sons, New York, 1985. [CPSvdK] E. Cline, B. Parshall, L. Scott, W. van der Kallen, Rational and generic cohomology, Invent. Math., 39, (1977), 143-163. [Donl] S. Donkin, On Schur algebras and related algebras I, J. Algebra, 104, (1986), 310-328. [Don2] S. Donkin, On Schur algebras and related algebras II, J. Algebra, 111, (1987), 354- 364. [GLS] D. Gorenstein, R. Lyons, R. Solomon, The Classification of the Finite Simple Groups, Mathematical Surveys and Monographs, 40, No. 3, AMS, Providence, RI, 1998. [Huml] J.E. Humphreys, Non-zero Ext1 for Chevalley groups (via algebraic groups), J. London Math. Soc, 31, (1985), 463-467. [Hum2] J.E. Humphreys, Modular representations of finite groups of Lie type, London Math. Soc. Lecture Note Series, 326, Cambridge Univ. Press, 2006. [Janl] J. C. Jantzen, Representations of Algebraic Groups, Second Edition, Mathematical Surveys and Monographs, 107, AMS, Providence, RI, 2003. [Jan2] J. C. Jantzen, Zur Reduktion modulo p der Charaktere von Deligne und Lusztig, J. Algebra, 70, (1981), 452-474. [Sinl] P. Sin, Extensions of simple modules for Sp4(2n) and Suz(2m), Bull. London Math. Soc. , 24, (1992), 159 -164. [Sin2] P. Sin, Extensions of simple modules for G2(Sn) and 2G2(3m), Proc. London Math. Soc. (3), 66, (1993), 327 - 357. [Sin3] P. Sin, The cohomology in degree 1 of the group F4 in characteristic 2 with coefficients in a simple module, J. Algebra , 164, (1994), 695 - 717. [TZ] P.H. Tiep, A.E. Zalesskii, Mod p reducibility of unramified representations of finite groups of Lie type, Proc. London Math. Soc, 84, (2002), 439-472.
EXTENSIONS FOR FINITE GROUPS OF LIE TYPE II 23 Department of Mathematics, Statistics and Computer Science, University of Wisconsin- Stout, Menomonie, WI 54751, USA E-mail address: bendelcQuwstout.edu Department of Mathematics, University of Georgia, Athens, GA 30602, USA E-mail address: nakainoQmath.uga.edu Department of Mathematics and Statistics, University of South Alabama, Mobile, AL 36688, USA E-mail address: pillenQjaguarl.usouthal.edu
This page intentionally left blank
Contemporary Mathematics Volume 413, 2006 Algebras, representations and their derived categories over finite fields Bangming Deng and Jie Du Abstract. We present a survey on the latest development in the representation theory of algebras with Frobenius morphisms. This includes a generalization of the graph and root system folding relation to folding relations at the representation level, at the category level, and further at the derived category level. As an application, we use it to approach Kac's theory for Kac-Moody algebras with symmetrizable Cartan matrices. Introduction Let ¥q be the finite field of q elements and let k — ¥q be its algebraic closure. A Frobenius map on a vector space over k is an abelian automorphism F : V —► V satisfying (Fl) F(Xv) = \<*F(v) for all v e V and A e fc; (F2) for any v eV, Fn(v) = v for some n > 0. In case V is finite dimensional, by Lang-Steinberg's theorem [29, 10.1], (F2) follows from (Fl) ( see, e.g., [7, Lemma 2.2] for a proof). Note that the existence of a Frobenius map F on V implies that VF = {v e V \ F(v) = v} is an F^-structure of V, i.e., V = VF ®Fg fc, and vice versa (see [9, 3.5]). This special case of Galois Descent Theory plays a fundamental role in our consideration. A Frobenius morphism on a fc-algebra A (with 1) is a Frobenius map F = Fa on the underlying vector space satisfying F(ab) = F(a)F(b) for all a,b e A. If M is an A-module, then we call a Frobenius map Fm on the space M a module Frobenius map (relative to Fa) if FM(am) = F(a)FM(m) for all a e A and m e M. In this case, the fixed point space AF = {a e A \ F(a) = a} is an Fg-algebra; while MFm is naturally an AF-module. The notion of Frobenious morphisms on fc-algebras is a natural extension of Fronbenius morphisms on algebraic varieties/groups over k. The latter is fundamental in the theory of algebraic groups and their representations. In [25], G. 2000 Mathematics Subject Classification. 16G10, 16G70, 18E30. Key words and phrases, finite dimensional algebra, representation, Frobenius morphism, quiver with automorphism, A us lander- Reiten quiver, derived category. Supported partially by the NSF of China, the Doctoral Program of Higher Education, and the Australian Research Council. ©2006 American Mathematical Society 25
26 BANGMING DENG AND JIE DU Lusztig investigated Probenius maps on representation varieties of a quiver with automorphism in connection with the geometric construction of quantized enveloping algebras of symmetrizable Kac-Moody algebras. Inspired by this work, we obtain in [6, 7] a general theory which relates representation theories of fc-algebras with Probenius morphisms and their fixed point algebras. One of the highlights of the theory is to generalize the folding relation between quivers and valued quivers, which induces a folding relation on their root systems, to a "folding" relation between the representations of quivers and Fg-species. As further applications of the theory and partially built on the work of Hubery [20, 21], connections between Kac's theories including Kac's polynomials and Kac's theorem for symmetric Kac- Moody algebras and symmetrizable ones have been established. Moreover, such a folding relation has also been extended nicely to the derived/root category level in [7]- This is a going-down approach in which the study of representations of a k- algebra A with a Probenius morphism F completely determines the representations of the fixed-point Fq-algebra AF. Since every F^-algebra is isomorphic to such a fixed-point algebra, this approach becomes a powerful approach to representations of Fq-algebras. Moreover, going-down from A to AF = B and from (F-stable) >l-module M to >lF-module MF\ one sees easily which properties or theories that hold for A continue to hold for B (see [7, §9]). It should be pointed out that Probenius maps on representations of quivers over finite fields have often been used in the context of Galois group actions; see [22, §3] and [24, §5]. In this approach, one starts with an F^-algebra B and considers A = B <S>Yq K where K is a finite Galois extension of ¥q. Then one studies those ^-modules X<S>^q K arising from J5-modules X. This has been a standard approach used in the literature (see for example [5, 24, 20, 21]). We call it the going-up approach. In this approach, there is a natural Probenius morphism F on A = B<g>K taking b <S> A i—> b <S> Xq. Then B = AF and for each J5-module X, the A-module X (g)Fg K admits a natural Probenius map F on X (&^q K taking m^Anm^A9 such that (X (g) K)F = X. Thus, results similar to 1.5, 2.1 and 3.2 have been obtained in this going-up approach. This paper presents a brief account of the main achievements on representations of algebras with Probenius morphisms. In §1, we define the Probenius (twist) functor in two versions. We then give a criterion for the existence of a module Probenius map on an A-module and embed naturally the category of >lF-modules as the subcategory of F-stable ^-modules. In §2, we apply the theory to quivers with automorphisms and prove that every Fq-species is isomorphic to some AF where A is the path algebra of a quiver Q and F is the Probenius morphism on A induced from an automorphism of Q. We further show that we may realize every finite dimensional basic fc-algebra in terms of a quiver with automorphism and certain relations. Several important topics in representations of algebras are discussed in §3. They include Morita equivalence, representation type, hereditary algebras and the Auslander-Reiten theory. In §4, we look at some applications to Lie Theory, especially to Kac's theory. Thus, we prove that counting number of representations or indecomposable ones of any F^-species results in certain polynomials. We also present a generalization of Kac's theorem. Finally, we discuss Probenius functors on the associated homotopy, derived and root categories and establish a triangulated
ALGEBRAS, REPRESENTATIONS AND THEIR DERIVED CATEGORIES 27 category equivalence between the bounded derived categories S>b{AF) of AF-mod and @b(A)F of F-stable objects in @b(A). Our theory has provided a simple and convenient approach to representations of finite dimensional algebras over finite fields, and has been used in three recent PhD theses [3, 31, 32] to study a certain elliptic Lie algebra and the structure of Hall algebras of valued quivers; see §5 for more details. Throughout we assume that all k-algebras A and A-modules are finite dimensional, and all modules are left modules. 1. Frobenius (twist) functors We first define the Frobenius twist of an A-module. The definition has two versions: the absolute version and the relative version. • The absolute version Let f : fc —► fc be the field automorphism sending A to A9. For each fc-space V and r > 1, let V^ be the new vector space obtained from V by base change via f: V^ = Vr®rfc. Thus, for v G V and A G fc, we have Xv <g) 1 = v <g) Xq . In other words, putting y(r) = y (g) l5 we have (u + u)(r) = u^r) + */r), (Au)(r) = A«Vr>. Note that V^ may be identified as V with a twisted scalar multiplication A. v = qy/\v. Further, for a fc-linear map <\> : U —► V, the map (f)^ := 0 <g) 1 : [/(r) —► V^r) is again a fc-linear map. In this way, we obtain an exact additive functor ( )(r) from the category of fc-vector spaces onto itself (see [12]). Let 7v,r : V —► V^ be the Fq-linear isomorphism sending v to v^ and let rv = Ty,\. If A is a fc-algebra, then A^ is also a fc-algebra, and ta • -A —► -A^) becomes an Fq-algebra isomorphism. Clearly, a map F : A —► >1 is a Frobenius morphism if and only if F o r^1 : A^1) —► >1 is a fc-algebra isomorphism. Definition 1.1. Let A be a fc-algebra with Frobenius morphism F and let M be an Amodule defined by the fc-algebra homomorphism 7r : A —► EikU(M). This gives a fc-algebra homomorphism 7r^^ : A^1) —► Endjt(M)^1^. Thus, the composition of the following maps A £^ A I±> A<X> Z^ Endk(M)W s End^M*1)) defines an Amodule structure on M^ with the following new action (1.1.1) a . (m(1)) = (F-^aJmjW, VoeAmeM. We denote this module by M^ and call it the Frobenius twist of M. If 7rW denotes the corresponding representation of M'1', then (1.1.2) 7r[1](a) = rM oTr^-^a)) o r^1 for all a G A
28 BANGMING DENG AND JIE DU If / : M —> N is an A-module homomorphism, then the ^-linear map f^ : j\/f(i) _> jy(i) becomes an A-module homomorphism AfW —► JVM which is denoted by /M in the sequel. Inductively, we can define the 5-fold Probenius twist AfW := (M^'1^ of M and /W = (/[s_1I)[11 for O 1, where M™ = M and /M = / by convention. Thus, the corresponding representations 7r[s] : A —► Endfc(M[s]) of M^ is given by (1.1.3) <K[s]{a) = rM,s o 7r(F"s(a)) o r^)s for all o G A Further, we can define M'-1! to be the ^-module N such that M = iVM and similarly for /I"1!. Thus, AfW and /I8! are well-defined for all sgZ. • The relative version We now define a Probenius twist relative to a given Probenius map. Let M be an A-module and let Fm • M —► M be any given Probenius map (not necessarily a module Probenius map). We define M^Fm^ to be the A-module such that mIFm1 = M as a vector space with F-twisted action (1.1.4) a*m := F^(F'^aJF^^m)) for all a e A,m e M. In other words, if n : A -> Endfc(M) and tt^I : ,4 -+ End^M^l) denote the corresponding representations, then (1.1.5) ^Fm\o) = FM o7r(F_1(a)) oF^1 for all a e A. The A-module mIFm1 is called the FM-twist of M. Similarly, for each sgZ, we define the 5-fold FM-twist 71-^' : A —► End^M^I) by (1.1.6) tt™(o) = FSM o 7r(F"s(a)) o F^s for all aeA Lemma 1.2. For eac/i s eZ, we have JlfW = M^I. /n particular, the s-fold Fm-twist mIfm] of M is independent of the selection of Fm, up to isomorphism. Proof. By (1.1.3) and (1.1.6), the fc-linear isomorphism ipM = tm,s ° F^s : MIfmI _> MM satisfies 7rls] = <^>M o -jtIfm] o <Pm, that is, <pm is an A-module homomorphism. D Note that, if Fm happens to be a module Probenius map, then a*m = am for all a e A and m e M. Thus, in this case, M\Fm\ = M as ^-modules. Let M and N be two A-modules with Probenius maps Fm and F/v, respectively. If / : M —► A/" is an A-module homomorphism, then (1.2.1) f^ := Fiv o / o F^1 : M^ —♦ iV^ is also an A-module homomorphism. • Module Frobenius maps and F-periods Recall that a module Probenius map on an ^-module M is a Probenius map Fm on M satisfying Fm{o>'^) = F(o)Fm(™)' Not every A-module M admits a module Probenius map, as seen from the following criterion (see [7, Prop. 2.8]).
ALGEBRAS, REPRESENTATIONS AND THEIR DERIVED CATEGORIES 29 Lemma 1.3. Let M be an A-module. Then M = AfM if and only if there exists a Frobenius map Fm on M such that FrM(am) = Fr{a)FrM{m), \/aeA,meM (or equivalently, M^I — M as A-modules by (1.1.6)). Moreover, M admits a module Frobenius map (relative to F) if and only if M = AfW. Proof. By (1.1.3) and (1.1.6), we observe that a Frobenius map Fm on M satisfies F^am) = Fr(a)F^(m) for all a e A, m e M if and only if F^ o r^r : is an A-module isomorphism. This clearly implies the sufficiency of the first assertion. For the necessity, take an A-module isomorphism <j> : AfM ^ M. Then the composition F' := <j> o tm,t is a Frobenius map on M with respect to qr or ¥qr. By choosing a basis for the ¥qr-structure MF of M, we may define a Frobenius map Fm : M —> M such that Fjj^ = F' = 4>otm,d that is, FJ^or^ = <j> is an A-module isomorphism AfM —► M. D Let p(M) = Pf(M) be the minimal number r satisfying M = M'rL We call it the F-period of M and call M F-stable if pf(M) = 1. The lemma above shows that if pF(M) > 1 then M does not admit a module Frobenius map relative to F. However, we have the following. Corollary 1.4. Let M be an A-module with F-period r. Then M := M 0 AfW 0 • • • 0 M^-1! admits a module Frobenius map Fm defined by , X\, . . . , Xj—\ ) = (Fjvf (xr_i), Fm(#o)5 • • • 5 i^Af(av-2))j w/iere Fm is a Frobenius map on M satisfying M^F^\ = M as A-modules.. • The Frobenius functor ( )Uod Let 4-mod denote the category of finite dimensional (left) A-modules. Then, Frobenius twisting induces a functor, the Frobenius functor (1.4.1) ( )W = ( )Wod : .4-mod -+ A-mod. This functor will be called the Frobenius (twist) functor on A-mod. Clearly, it is a category equivalence. The Frobenius functor determines a new category AmodF whose objects are F-stable A-modules M with a fixed isomorphism 4>m • M'1' ^ M and whose morphisms are compatible with the isomorphisms 4>m, i.e., HomA.modF(M, N) = {fe HomA(M, N) \ <j)N o /M = f o </>M}. Clearly, a selection of different isomorphisms 4>m results in an equivalent category. Some version of the following category equivalence has already been obtained by Hubery [20, Prop. 17] in the going-up approach. Theorem 1.5. There is a category equivalence AF-mod ^ A-modF. Proof. The base change functor sending an AF-module N to Nk = N<g> k has an "inverse" which takes an F-stable module (M,Fm) to its fixed point module MFm. D
30 BANGMING DENG AND JIE DU 2. Representations of quivers with automorphisms Let Q = (Qo,Qi) be a finite quiver, where Q0 (resp. Q\) denotes the set of vertices (resp. arrows) of Q. For each arrow p in Qi, we denote by pf -^ p" to indicate the tail p' and the head p" of p. Let a be an automorphism of Q, that is, a is a permutation on the vertices of Q and on the arrows of Q satisfying the compatibility conditions: cr(p') = &(p)f and &{p") = &(p)ff for any p G Q\. Let >1 := &Q be the path algebra of Q over fc = ¥q which has the identity 1 = YlieQo ei wnere ei is the idempotent (as a length zero path) corresponding to the vertex i. Then a induces a Frobenius morphism (2.0.1) FQ^ = FQ^q :A->A; ^xsPs ^ JTx^fo), where ^s #sPs is a fc-linear combination of paths ps, and o~(ps) = &(pt)' * * &(Pi) if Ps = Pt - ' Pi for arrows pi,..., pt in Qi. We now construct an F^-species1 (see [14, 27, 10, 11]) from a quiver Q with automorphism a. First, associated to (Q, cr), there is a valued quiver T = T(Q, a) = (r0, Ti) whose vertex set r0 (resp. arrow set Ti) is the set of cr-orbits in Q0 (resp. in Qi), and whose valuation is denned as follows: we associate to each k e r0 the number e^ = #k, and to each arrow p : p' —> p" in V the pair (df^dfp) of the numbers denned by (2.0.2) ep = #p, d"p = ePlepn and d'p = ep/ep>. Second, using the Frobenius morphism F = Fq^ on A denned above, we can attach naturally to T an ¥q-modulation as follows: for each vertex i e T0 and each arrow p e I\, we fix i0 G i, po G P-> and consider the F-stable subspaces of A £i-l ep-\ Ai = 0 kei = 0 ke*3(io) and AP = 0 kP = 0 ***(P0). zGi s=0 pep t=0 Then we have (2.0.3) ei-l eP-l Af = {]P x^e^s^) | x G k,xq£i = x} and ^ = { ^ x9V*(p0) I £ € /c,xq£p = x}. Further, the algebra structure of A induces an ^,,-^,-bimodule structure on Ap for each arrow p : p' —> p" in P. Thus, we obtain an F^-modulation M = M(Q, a) := ({A[}{, {A^}p) over the valued quiver I\ The F^-species (r, M) denned above will be denoted by DJIq^ = 9Kq,<7# = (I\M). The following result given in [7, 9.3] shows that every F^-species may arise in this way. Thus, by Theorem 1.5, we may regard representations of Fq-species as F-stable representations of the corresponding quiver. Theorem 2.1. Let (Q, a) be a finite quiver with automorphism a, and let OJIq^ be the ¥q-species associated to (Q,cr). (1) If A = kQ is the path algebra of Q, and F = Fq^ is the Frobenius morphism on A induced by a, then the fixed point algebra AF is isomorphic to the tensor (or path) algebra T(DJIqi(T) ofDJlQi(T. We used the term ¥q-modulated quivers [1] for Fq-species in [6, 7].
ALGEBRAS, REPRESENTATIONS AND THEIR DERIVED CATEGORIES 31 (2) For any given ¥q-species DJl, there is a quiver Q and an automorphism a of Q such that the tensor algebra T(97t) of$Jl is isomorphic to (kQ)FQ>a. Remark 2.2. If an F^-species SOT involves only natural bimodules Fgr (Fqn)FgS in the sense that the bimodule structure is induced from the subfield structure, where r,s,n ^ 1, r\n and s\n, then the corresponding quiver Q and its automorphism a have been constructed in [20, Lemma 21] (see also [18]). In fact, Hubery further observed that the Probenius map on KQ, where K is a finite extension of ¥q, considered in the going-up approach (§1) decomposes into the product of a quiver automorphism and a field automorphism (see the discussion prior to [20, Prop. 22]). This decomposition is similar to Steinberg's decomposition for an automorphism of a finite Chevalley group (see [2, 12.5.1]). Our next result deals with a more general case and shows that every Probenius morphism on a basic k-algebra is induced from the path algebra of a quiver with automorphism. Theorem 2.3 ([7, 9.5]). If A is a finite dimensional basic k-algebra with a Probenius morphism F, then there are a quiver Q with automorphism a and an algebra epimorphism <p : kQ —> A such that the following diagram is commutative kQ - A FQ,o V kQ In particular, every finite dimensional basic ¥q-algebra B is isomorphic to (kQ/I)FQ'a for some quiver Q with automorphism a and some Fq^-stable admissible ideal I of kQ. 3. Folding representations with Probenius morphisms We now look at several important topics in the representation theory of finite dimensional algebras. • Morita equivalence Up to Morita equivalence, the study of representations of algebras may be reduced to that of basic algebras. By Theorem 2.3, a Probenius morphism on a basic algebra arises from a quiver with automorphism. Thus, it would be interesting to know if a Morita equivalence is invariant under a Probenius morphism. The following result is given in [7, 9.5(2)]. Theorem 3.1. If A is a finite dimensional k-algebra with a Probenius morphism F. Then there exists a basic algebra A' with a Probenius morphism F' such that both pairs (A, A') and (AF,Af ) are Morita equivalent. • Representation type
32 BANGMING DENG AND JIE DU A (finite dimensional) algebra A is said to be of finite representation type if, up to isomorphism, it has only finitefy many indecomposable representations. Other representation types including tame type and wild type can be also denned. The following result has been proved in [6, 5.1]; compare [22, Lemma 3.4], [24, 5.3] and [20, Prop. 17]. Theorem 3.2. Let M be an indecomposable A-module with F-period r. Then there exists a Frobenius map Fm on M such that the pair (M,Fm) defined in 1.4 is indecomposable in A-modF and EndAF(MF)/R&d(EndAF(MF)) ^ ¥qr (F = FM). Moreover, every indecomposable AF-module is isomorphic to a module of the form MF for some indecomposable A-module M and Frobenius map Fm- Corollary 3.3. Let A be a k-algebra with a Frobenius morphism F. Then A is of finite representation type if and only if so is AF. It is natural to expect that this result continues to hold for tame and wild types. • Finite dimensional hereditary algebras A quiver automorphism a is called admissible if there are no arrows connecting vertices in the same cr-orbit. Call a quiver with an admissible automorphism an ad- quiver (see [6, 3.5]). The following characterization of finite dimensional hereditary algebras is an easy consequence of Theorem 2.1. Theorem 3.4. An algebra B is a finite dimensional hereditary basic algebra overFq if and only if B is isomorphic to (kQ)FQ^ for some ad-quiver (Q, a) without oriented cycles. • The Auslander-Reiten quiver (or AR-quiver) of a fc-algebra Since the algebra A is denned over the algebraically close field k = Fg, we may regard the AR-quiver Q = Qa of A as an ordinary quiver. We first observe that Q admits an admissible automorphism s. For each vertex [M] e Q, s([M]) is defined to be [MM]. If M and N are indecomposable A-modules, then there are nst arrows iff from [AfW] to [N^] in Q, where 0 < s < p(M) - 1, 0 < t < p(N) - 1, nst = dimfclrr^AfM, N^) and 1 ^; m < nst. Note that nst = ns+i^+i for all s,£, where subscripts are considered as integers modulo the F-periods p(M) and p(N), respectively. We now define 8(-yff) = 7i+i,t+i for all 0 < 5 < p(M) - 1 and 0 < * < p(N) - 1. Clearly, s is an admissible quiver automorphism and (Q,s) is an ad-quiver. Associated to (Q,s), we may define an Fq-species 971q,s as in Section 3: let A = kQ denote the path algebra of Q and F = Fq,s be the Probenius morphism on A induced by the automorphism s. For each vertex i(M) (i.e., the s-orbit of [M])
ALGEBRAS, REPRESENTATIONS AND THEIR DERIVED CATEGORIES 33 and each arrow p (i.e., an s-orbit of arrows in Q) in T(Q, s), we define subspaces p(M)-l *4i(M) = @ *e[Af[-i] and ,4p = @/cp, s=o pep of *4, which are obviously F-stable. By definition, the ¥q-modulation M(Q,s) is given by (*4i(M))F and (*4P)F for all vertices i(M) and arrows p in T(Q,s). Recall from [6, 6.2] the definition of isomorphisms for Fq-species. Two species are isomorphic if their associated "simple" ones (obtained by summing up the valuations and bimodules over parallel edges) are isomorphic. We now can state the following result. Theorem 3.5. The ¥q-species WIq,s associated to the AR-quiver (Q,s) of A defined above is isomorphic to the AR-quiver Q^f of AF. Moreovery the Auslander- Reiten translation of A naturally induces that of the fixed-point algebra AF. • Other topics Properties like global dimension, self-injectivity and preprojectivity of a k- algebra A are also F-invariant. In other words, A has such a property if and only if so does AF. However, not every property which A possesses is F-invariant. For example, the number of irreducible modules is not F-invariant. It seems true that quasi-heredity is not F-invariant. 4. Applications to Lie theory The characterization of hereditary algebras (Theorem 3.4) in terms of quivers with automorphisms establishes direct links between representations of hereditary algebras and Kac-Moody Lie algebras. We now look at them at several levels. • Symmetrizable generalized Cartan matrices and their root systems Let Q be a finite quiver without oriented cycles. Then Q defines a symmetric generalized Cartan matrix Cq = (aij)ijeQo ky a■■- I 2 if *=jf ^ I — |{arrows between i and j}\ if i ^ j. If Q is equipped with an admissible automorphism cr, then the associated valued quiver T (§3) defines a symmetrizable generalized Cartan matrix C? = (fry)ijG/ by ifi=j 6°"\ -Ep£p/£i if i#J where the sum is taken over all arrows p between i and j. Since the definition is regardless of the orientation, it is easy to see that all symmetrizable generalized Cartan matrices can be obtained in this way.' Let A(Q) c NQo (resp. A(T) c NTq) be the root system associated with the quiver Q (resp. the valued quiver T), or equivalently, the root system of the Kac-Moody algebra associated with the Cartan matrix Cq (resp. Cr) (see [22] or [23] for its definition).
34 BANGMING DENG AND JIE DU The quiver automorphism a extends linearly to a group automorphism a on ZQo denned by 0-(]P aii) = ]P a,i<r(i). ieQo ieQo Let (ZQo)a denote the subset of cr-fixed points in ZQo- This set can be identified with the group Z/ via the canonical bijection a : (ZQo)*7 — Z/; ]T M ^ ]T a'^ ieQo iei where a\ := bi = bj for all i,j G i. For (3 G A(Q), let t ^ 1 be the minimal integer satisfying cr*(/3) = /?. We call t the a-period of /?, denoted by p(/3) = p<r((3). We have the folding relation between the root systems of Q and T (see [30, Prop. 2] and [20, Prop. 4]). Proposition 4.1. Let (3 e A(Q) and set 0 := /? + *(/?) + • • • + cr^W) G (ZQoV, w/iere £ = pCT(/3). T/ien /? i—> <r(/3) defines a surjective map A(Q) —► A(r). Moreover, 2/ <r(/3) 25 reaZ, £/ien /? 25 reaZ and is unique up to a-orbit. Thus, to an ad-quiver (Q, a) (without oriented cycles), we may associate a finite dimensional hereditary ¥q-algebra (see 3.4) and a Kac-Moody algebra with the generalized Cartan matrix Cp. The connection between the two algebraic structures has been investigated since 1970s. Gabriel's theorem [13], Kac's theorem [22] and Ringel-Hall algebra approach to quantum groups [28] are the milestone contributions to this investigation. • Counting F-stable representations Let Q be a finite quiver with an admissible automorphism cr, and let WIq^ be the associated Fg-species with underlying valued quiver T = T(Q, a). Put / = r0. Given a matrix x = (xij) G fcmxn and an integer r > 0, we define XM = (xfj) e kmxn. For each /? = Y,ieQ0 ^ e (NQ'o)CT5 let V% = kbi for each i G Qo- We consider the affine variety R(P) = R(Q,0) = n Homfc(fcV,fcV') * ]J fcV'*V. peQi peQi Then a point x = (xp)p of R((3) determines a representation V(x) = (Vi,xp) of Q. The algebraic group <?(/?)= Yl GLbi(k) C GL(V) ieQo acts on R((3) by conjugation (9i)i ' (xP)p = (9p"Xpg~})pi and the G(/3)-orbits Ox in R((3) correspond bijectively to the isoclasses [V(x)] of representations of Q with dimension vector /?.
ALGEBRAS, REPRESENTATIONS AND THEIR DERIVED CATEGORIES 35 Further, we define a Probenius map F on the Qo-graded vector space V = ®ieQ0Vi sucn tnat5 f°r v e Vi, vW := F(v) e V„(i) for all i e Qo- Then the Probenius map F on V induces a Probenius map on the variety R{@) such that, for x = (xp) e R(/3), F(x) = (yp) is denned by yp(F(v)) = F{xa-i{p){v)) for all peQi,ve Va-i(pf), and a Probenius map on the group G((3) given by F(gv) = F(g)(F(v)) for all g e G(/3), v e V. The action of G{f3) on R(0) restricts to an action of G{f3)F on R{f3)F. Then, the G(/?)F-orbits in R((3)F correspond bijectively to the isoclasses of F-stable representations of Q with dimension vector /?, or equivalently, to the isoclasses of T(97lQ,CT)-modules with dimension vector <r(/3) =: a. Now let Mg)<T(a, q) = # of isoclasses of T(97lQ,CT)-modules of dimension vector a, /Q,CT(a, ^) = # of isoclasses of indecomposable T(SDtQ)CT)-modules of dimension vector a. The following result is proved in [6, 9.1-2]. See [18] for some natural Fg-species in which the Fgr-Fgs-bimodule structure on ¥qm in the modulation is given naturally as subfields. Theorem 4.2. Both MQi(T(a,q) and lQ^a(a,q) are polynomials in q with rational coefficients and are independent of the a-admissible orientation of Q. The proof uses some standard counting formulas for GL(n,q) given in [26, p.272] together with Burnside's counting formula: M geG 9ecc\(G) ^ where G = G(/3)F, X = R((3)F, and ccl(G) is a set of representatives of conjugacy classes of G. Remark 4.3. In [20], the so-called isomorphically invariant representations (ii-representations for short) of Q have been studied. This generalizes some results in [30]. Given a representation V = (Vi,Va) of Q over fc, we define a new representation aV = (Wi,Wa) of Q by Wi = Va-i^) and Wa = V^-i(Q) for all i e Qo and a e Q\. The representation V is called isomorphically invariant if V = aV. It is said to be ii-indecomposable if it is not a direct sum of two non-zero ii-representations. In case k is an algebraically closed field of characteristic not dividing the order of cr, the dimension vectors of ii-indecomposable representations of Q over k have been described (see [20, Thm 1]). Further, the polynomials for counting the number of isoclasses of ii-indecomposable and absolutely ii-indecomposable representations of Q over finite fields have been investigated. In the affine case these polynomials are explicitly calculated. We refer to [19, Chapter 4] for details. • A generalization of Kac's theorem
36 BANGMING DENG AND JIE DU Let Q be a finite quiver without oriented cycles and a an admissible automorphism of Q. Let 971q,ct be the associated Fg-species with the underlying valued quiver V and vertex set / = To. The following theorem is known as Kac's theorem when a = 1 and is proved in [6, 10.3]. Theorem 4.4. (1) The polynomial lQ,a(a, q) is non-zero «<=> a G A(r)+. (2) a G A(r)+ is real => IQl(T(a, q) = 1. Note that our proof requires Kac's theorem together with a result of Hubery [20, Thml]. It should be interesting to find an independent proof of the theorem in the species case. See the conjecture [6, 10.5]. Remarks 4.5. (a) This theorem is a generalization of Kac's theorem for quivers over a finite field to a result for all Fg-species. This result has been proved by Hua and Hubery for the natural Fg-species. (b) By using Ringel's Hall algebra approach, Deng and Xiao proved in [8] that, for any prime power q and dimension vector a, the number Iq CT(a, q) =fi 0 4=> a G A(iy. • Kac conjectures For each a G A(r)+ and each r ^ 1, let lQ,a(a, q'lr) — # °f isoclasses of indecomposable T(9JIqi(T)-modules of dimension vector a arising from indecomposable fcQ-modules of F-period r. Clearly, lQ,a(a, q) = Ylr>i ^Q^(a^ #5 r)> and lQ,a{&, <751) is the number of isoclasses of absolutely indecomposable representations of Q over ¥q. For a = 1, we put Iq(a,q) = IQ^(a,q) and IQ(a,q;r) = IQ^(a,q,r). Thus, since the constant term of 7g(a,^;r), r > 2, is zero (see [17, 2.3]), it follows that the constant terms of Iq(ol, q) and Iq(&, q; 1) coincide. Kac proves that Iq(ol, q\ 1) G Z\q] and makes the following two conjectures: Kac Conjecture for Positivity: Iq(a,q; 1) G N[#]. Kac Conjecture for Multiplicity: /q(q:,0; 1) = multfl(a). Here g is the Kac-Moody algebra associated to Cq and multfl (a) = dim ga. These two conjectures have been proved by Crawley-Boevey and Van den Bergh [4] for indivisible roots a, but are still open in general. It would be interesting to study the polynomials lQ,a(a,q;r) for an arbitrary a. • Some other applications Applications of this theory to Hall algebras, Lie algebras and quantum groups can be found in [3, 31, 32]. More precisely, in [3], by studying the Probenius morphism of the tubular algebra of type T(3,3,3) and the root category of its fixed point algebra, the author obtains a realization of the elliptic Lie algebra of type F4 ' ;. In [31], based on the folding relation between quivers and valued quivers in §3, the author studies representations of (arbitrary) Fg-species with oriented
ALGEBRAS, REPRESENTATIONS AND THEIR DERIVED CATEGORIES 37 cycles and their associated Ringel-Hall algebras. Finally, in [32], by describing instable representations of affine quivers with automorphisms, the PBW type bases of composition algebras of affine valued quivers (thus of the corresponding quantum groups) have been constructed. Moreover, the minimal generating system for the Ringel-Hall algebra of an affine valued quiver is obtained. The author also presents a ^-analogue of the Weyl-Kac denominator identity for affine valued quivers of type Bn, DDn, F4i, and G2i- 5. Folding derived/root categories with Probenius functors The Probenius functor ( )"Jod : A-mod —► A-mod and the category equivalence AF-mod —> A-modF can be naturally lifted to the derived and root category levels. • The derived category level Let ^(A) := ^(A-mod) denote the category of (cochain) complexes of A- modules M = (Mi,di) = ► M*-1 ^ Mi -^ Mi+1 *—> • • • where d2 = 0. Let Jff(A) := J(f(A-mod) be the homotopy category defined by , n / Ob(jr(A)) = Ob(V(A)), and for M,N e Ob(JT(A)), (5'U-ij \ Homjr(A)(M,N) = Homv(A)(M,N)/Ht(M,N), where Ht(M, N) denotes the subspace of Hom<^(^)(M, N) consisting of morphisms homotopic to zero. Further we let 3>(A) be the derived category of A, i.e., $>{A) is the localization of J(f(A) at the class S of all quasi-isomorphisms. (A morphism / : M —► N in Jff{A) is called a quasi-isomorphism if Hl(f) is an isomorphism for each ieZ, where H% : J(f(A) —► A-mod is the 2-th cohomological functor.) We shall denote the full subcategory of C(A), where C = <€, <# or ^, consisting of bounded complexes (resp. complexes bounded below, complexes bounded above) by Cb(A) (resp. C+(A), C~{A), etc.). Applying the Probenius functor to each M% in (5.0.1), we obtain a new complex MW = •••-> {M^p ^T (^)W {dX] (M*+i)W ^ .... This will be called the Probenius twist of M. Thus, the Probenius functor on A-mod defined in (1.4.1) induces a functor ()[11 = ()SU:*(^*m which we still call the Frobenius (twist) functor (on complexes). Since a morphism / : M —► N is homotopic to zero (resp. a quasi-isomorphism) if and only if so is /M, the Probenius functor ( )W on ^(A) induces functors {)W = {)[c\A)--C{A)^C{A){C = X,&), which clearly preserve distinguished triangles. Thus, the Frobenius functors on J(f(A) and ${A) are equivalences of triangulated categories. For C e {*£, Jf, &}, let Cb(A)F be the category whose objects consist of rn ^ bounded complexes M satisfying ML J = M and whose morphisms are compatible
38 BANGMING DENG AND JIE DU with these isomorphisms: Homcw(M,N) = {/ e Homcb(A)(M,N) | ^ o/M = /°<^J- Clearly, a different selection of the isomorphisms results in an equivalent category. The following results are proved in [7, 4.1-2, 5.2-4] (cf. 1.5). Theorem 5.1. The embedding tfb(AF) -+ tfb(A) sending X to Xk = X <g> k induces a category equivalence ffb(AF) = ^b(A)F and faithful functors $c : Cb(AF) —> Cb(A) for C = Jf, $>. Moreover, these embeddings result in triangulated category equivalences Xb{AF)^Jfb{A)F and @b(AF) * @b(A)F. • The root category level Following [15, 5.1], the root category &(A) of A is the quotient (or orbit) category @b(A)/(T2) of @b(A) by the automorphism T2, where T is the shift functor on @b(A) induced from the one on ^(A)2 Thus, by definition, the objects m&(A) are T2-orbits of objects in ®b(A), i. e., Om = {T2iM \ i e Z}, M G @b(A). A morphism / = (fji) : Ojvi —> 0^ is given by morphisms fji : T2lM —> T2jN in 3>b(A) satisfying (1) Tifji) = /i+i,i+i for all ij e Z, (2) For each fixed iGZ, all but finitely many fji are zero. The composition of the morphisms / : O^l —* ^N and g • On ~~> ®L ls the morphism h = (hji) = gh with hji = Ylsezdjsfsilt is easy to check that the Frobenius functor ( )J(A\ commutes with the shift functor T and induces a functor Note that (CM)[1] = ^mI1' for each M in @b(A) and /[l1 = (/ji1) for a morphism / = (fr) in #(4). Let 8£(A)F be the category consisting of Mi ^M Objects: Om such that C?£[ ^ Om in ^(A), Morphisms: Hom^(A)F (M, N) = {£ G Hom^(A)(M, N) | ^ o f M = £ o 0fj. Theorem 5.2. Le£ ,4 be of finite global dimension. Then §® : @b(AF) —► @b(A) induces a faithful functor &(AF) -^> &(A) and a category equivalence &(AF)^&(A)F. Let >1 be a hereditary (basic) /^-algebra with a Frobenius map F. Then >1 and AF are related by an ad-quiver (Q, <j) (Thm 3.4). So >1 can be identified as the path algebra of Q and AF as the path (or tensor) algebra of the associated F^-species (r, M) of Q via a. Now, using the notation introduced in the previous section, the theorem above together with Theorem 4.4 and [15, 4.7] implies immediately the following. 2The shift functor on <*f is defined by (TM)* = Mi+1, d^M = -d1^ and T(/)* = fi+1 if / is a morphism in ^(A).
ALGEBRAS, REPRESENTATIONS AND THEIR DERIVED CATEGORIES 39 Corollary 5.3. Suppose that A is hereditary (and basic). If O^ is an indecomposable object in &(A), then <j(dim(9^) = dimOx, where X G S>b{AF) with M = Xk in @b(A), and the root system A(r) = {(j(dimO^) | M indecomposable in @b(A)}. We remark that it is known that the results given in [15, 4.7], [16] are over an algebraically closed field and continue to hold for a species over a finite field. However, the argument in the latter case is not given there. Using Probenius mor- phisms, the argument to the theory is now complete and elegant. • Frobenius maps on complexes Let M = (M\ dl) be a complex in <tf(A) and let 7 := {F* : Mi -> Mi \ i G Z} be a family of Probenius maps. We shall call 5F a Frobenius map on M. For each i G Z, let (AP)[F*] denote the F.-twist of M\ Then each d{ : Mi -> Mi+1 gives an A-module homomorphism (see (1.2.1) for the notation) d^ = Fm o dlM o F'1 : (M*)W -+ (Afi+1)[F*+1l. Thus we obtain a complex ((M2)^, cHFJ), which is called the CF-twist of M and is denoted by There is a complex version of 1.3. Lemma 5.4. Let M = (M2,d2) be a complex in ^(A). Then M = MM if and only if there exists a Frobenius map 5F = {Fi : M2 —> M% \ i G Z} on M s^c/i £/m£ = M as complexes of A-modules. If M S M[1] then M is called an F-stable complex. The method of constructing F-stable modules from F-periodic modules given at the end of §2 can be generalized to complexes. Let M be an F-periodic complex in ff(A) with F-period r, i.e., M = Mr' in ^(A) with r minimal. By Lemma 5.4, there is a Probenius map 5 = {Fi : M2 —> M2 | i G Z} such that M = M'^ as complexes. For each i, let at = m2 e (M2)[Fi] e • • • e (Af*)^*"1' and define a Probenius map F; : M% —> Af2 by (5.4.1) Fi(x0, xi,..., av_i) = (Fi(xr_i), Fi(x0),..., Fi(xr_2)). Further, let & = diag(d2, (d2)[F],..., (d*)1^"1') : M2' -> Mi+1. Then we obtain an F-stable complex M = (Af2, d2) satisfying ML = M, where 7 = {F{ \ i G Z}. This gives a complex M in ^(AF). Since every bounded complex has a finite F-period, this construction applies to every object in ^b(A). Moreover, every complex is isomorphic to one containing no non-zero contractible summands in the homotopy category J(f(A), the same construction applies to every object in J(fb(A) which contains no non-zero contractible summands. Thus the first part of the following theorem generalizes Theorem 3.2 (see [6, 5.1] and [7, 4.4]), while the second part is given in [7, 5.1,5.6]. Theorem 5.5. Maintain the notation above. Let M be an indecomposable complex in Cb(A) with Fq -period r, where C G {^, Jf, &}.
40 BANGMING DENG AND JIE DU (1) IfC = CS> or Jff, M is indecomposable in Cb(AF) and Endcb(AF)(M )/Rad(Endcb(AF)(M' )) ^ Endcb(A)F(M)/Rad(Endcb(A)F(M)) ^¥qr. Moreover, every indecomposable complex in Cb(AF) is isomorphic to a complex of the form M for some F-periodic indecomposable complex M in Cb(A). (2) Suppose in addition that A has a finite global dimension. Then there exists X e @b(AF) such that M^Xk in @b(A) and End^b(AF)(X)/Rad(End^b(AF)(X)) ^ End^b(A)F(M)/Rad(End^(A)F(M)) ^ ¥qr. Hence X is indecomposable. Moreover, every indecomposable object in 3>b{AF) can be obtained in this way. Acknowledgment We are grateful to the referee for a careful reading and some helpful comments, especially for pointing out the explicit relation (1.1.2) in a comparison with (1.1.5) and comments on the existence of the going-up approach. References [1] D. Benson, Representations and Cohomology, Vol I, Cambridge Studies in Advanced Mathematics: 30. Cambridge University Press, Cambridge, 1995. [2] R.W. Carter, Simple groups of Lie type, A Wiley-Interscience Publication. John Wiley &; Sons, Inc., New York, 1989. [3] Z. Chen, Hall algebras of tubular algebras and Lie algebras, Ph.D. thesis, Xiamen University, 2005. [4] W. Crawley-Boevey and M. Van den Bergh, Absolutely indecomposable representations and Kac-Moody Lie algebras, Invent. Math. 155 (2004), 537-559. [5] C.W. Curtis and I. Reiner, Methods of representation theory. With applications to finite groups and orders, Vol. I. A Wiley-Interscience Publication, New York, 1981. [6] B. Deng and J. Du, Frobenius morphisms and representations of algebras, Trans. Amer. Math. Soc, in press. [7] B. Deng and J. Du, Folding derived categories with Frobenius functors, to appear. [8] B. Deng and J. Xiao, A new approach to Kac's theorem on representations of valued quivers, Math. Z. 245 (2003), 183-199. [9] F. Digne and J. Michel, Representations of finite groups of Lie type, London Math. Soc. Student Texts, 21. Cambridge University Press, Cambridge, 1991. [10] V. Dlab and CM. Ringel, On algebras of finite representation type, J. Algebra 33, (1975), 306-394. [11] V. Dlab and CM. Ringel, Indecomposable representations of graphs and algebras, Memoirs Amer. Math. Soc. 173, 1976. [12] E.M. Friedlander and A. Suslin, Cohomology of finite group schemes over a field, Invent. Math. 127 (1997), 209-270. [13] P. Gabriel, Unzerlegbare Darstellungen I, Manuscripta Math. 6 (1972), 71-103. [14] P. Gabriel, Indecomposable representations II, Istit. Naz. Atta Mat., Symp. Math. XI (1973), 81-104. [15] D. Happel, On the derived category of a finite-dimensional algebra, Comment. Math. Helvetici 62 (1987), 339-389. [16] D. Happel, Triangulated Categories in the Representation Theory of Finite-dimensional Algebras, London Math. Soc. Lecture Note Series, 119. Cambridge University Press, Cambridge, 1988. [17] J. Hua, Representations of quivers over finite fields, Ph.D. thesis, University of New South Wales, 1998.
ALGEBRAS, REPRESENTATIONS AND THEIR DERIVED CATEGORIES 41 [18] J. Hua, Numbers of representations of valued quivers over finite fields, preprint, Universitat Bielefeld, 2000 (www.mathematik.uni-bielefeld.de/~sfbll/vquiver.ps). [19] A. Hubery, Representations of quivers respecting a quiver automorphism and a theorem of Kac, Ph.D. thesis, University of Leeds, August 2002. [20] A. Hubery, Quiver representations respecting a quiver automorphism: a generalisation of a theorem of Kac, J. London Math. Soc. (2) 69(2004), 79-96. [21] A. Hubery, Representations of a quiver with automorphism: generalising a theorem of Kac, Fields Inst. Commun., 45, Amer. Math. Soc, Providence, RI, 2005, 187-200. [22] V. Kac, Infinite root systems, representations of graphs and invariant theory, Invent. Math. 56 (1980), 57-92. [23] V. Kac, Infinite dimensional Lie algebras, Third edition, Cambridge University Press, 1990. [24] H. Kraft and C. Riedtmann, Geometry of representations of quivers, London Mathematical Society Lecture Note Series 116 (ed. P. Webb, Cambridge University Press, Cambridge, 1986) 109-145. [25] G. Lusztig, Canonical bases and Hall algebras, Representation theories and algebraic geometry (Montreal, PQ, 1997), NATO Adv. Sci. Inst. Ser. C. Math. Phys. Sci., 514, Kluwer Acad. Publ., Dordrecht, 1998, 365-399. [26] I.G. Macdonald, Symmetric functions and Hall polynomials, 2nd edition, Clarendon Press, Oxford, 1995. [27] C. M. Ringel, Representations of K-species and bimodules, J. Algebra 41 (1976), 269-302. [28] C. M. Ringel, Hall algebras and quantum groups, Invent. Math. 101 (1990), 583-592. [29] R. Steinberg, Endomorphisms of linear algebraic groups, Memoirs of the American Mathematical Society, 80, 1968. [30] T. Tanisaki, Foldings of root systems and Gabriel's theorem, Tsukuba J. Math. 4(1980), 89-97. [31] Y. Wang, Representations of valued graphs with cycles and double Ringel-Hall algebras, Ph.D. thesis, Tsinghua University, 2005. [32] A. Wufu, Ringel-Hall algebras of affine valued quivers, Ph.D. thesis, Beijing Normal University, 2005. School of Mathematical Sciences, Beijing Normal University, Beijing 100875, China. E-mail address: dengbmQQbnu.edu.cn School of Mathematics, University of New South Wales, Sydney 2052, Australia. Web-page: http: //www. maths. unsw. edu. au/~j ied E-mail address: j . duQQunsw. edu. au
This page intentionally left blank
Contemporary Mathematics Volume 413, 2006 On Localization of D-modules Yoshitake Hashimoto, Masaharu Kaneda, and Dmitriy Rumynin To Professor Jim Humphreys on the occasion of his 65th birthday The localization theorem of A. Beilinson and J. Bernstein [BB] has played a prominent role in the representation theory of complex semisimple Lie algebras, leading to the solution of the Kazhdan-Lusztig conjecture [KL] by J. L. Brylinski and Kashiwara M. [BK] and Beilinson and Bernstein [loc. cit.]. The theorem consists of two parts; the first gives an isomorphism between a central reduction U° of the universal enveloping algebra U of the Lie algebra and the algebra Diff of global differential operators on the corresponding flag variety, and the second, called the D-affinity, establishes an equivalence of the category of Diff'-modules of finite type and the category of coherent modules over the sheaf Viff of differential operators on the flag variety. In trying to carry over the theorem to the representation theory of simple algebraic group G in positive characteristic, the first part was found false by S. P. Smith [Sm] in SL2 replacing U by the algebra of distributions on G. For the second part, after B. Haastert [Haa87] showed the D-affinity of the projective spaces and of the flag variety in case G = SL3, Kashiwara and N. Lauritzen [KLa] found its failure in SL5. In positive characteristic Viff admits, in addition to the filtration by order, another filtration by V^ = Mode {rn+1)(Ox,Ox), rn e N, where X^m+1^ is the (m + l)-st Probenius twist of X. Each V^ is a central reduction of Berthelot's ring £>(m) of arithmetic differential operators of level m [B96]. In particular, V^ used to be called the sheaf of PD-differential operators [B74] and is without divided powers, more like the universal enveloping algebra U than like the algebra of distributions Dist(G) [BB93]. Replacing Viff by V^\ R. Bezrukavnikov, I. Mirkovic and D. Rumynin have now successfully recovered the localization theorem with the second part surviving as a derived equivalence, where V^ is called the sheaf of crystalline differential operators. The Lie algebra q of G in positive characteristic p admits a p-th power map x \-> x^\ x e Q, where x^ is the p-th power of x in Dist(G) which stays in g. The subalgebra 3pt of U generated by xp — x^\ x G g, is central in U, called 2000 Mathematics Subject Classification. 14F10, 16S32, 17B50, 20G10. supported in part by JSPS Grant in Aid for Scientific Research. ©2006 American Mathematical Society 43
44 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN the Probenius center of U. If m is a maximal ideal of 3pt generated by xp — x&\ x e g, the derived equivalence of Bezrukavnikov, Mirkovic and Rumynin refines to an equivalence between the bounded derived categories of coherent V^-modules that are annihilated by a power of m and of U°-modules of finite type that are likewise annihilated by a power of m. Now V^ is the central reduction of V^ by m while the central reduction U (g)3Fr (3pr/tn) of U is Dist(Gi) the algebra of distributions of the Probenius kernel G\ of G. The representation theory of G\ is intimately related to, and is often more accessible than that of G. We thus start in this paper an investigation of the localization theorem for p(m) on the flag variety B of G. Unfortunately, the natural k-algebra homomorphism Dist(Gi) —> T(B,V^) is, due to Bezrukavnikov, not surjective in general, which we will explain in §2. On the positive side we will find in §§4 and 5 that the derived equivalence holds for P(m)-modules on the projective space Fn iff n < pm+1, and also for V^ on the flag variety in SL3 iff p > 3. We will show in §3 that the derived equivalence for p(m) on smooth projective variety X follows from Beilinson's lemma [Be], [Ba] if the dual (F™+1Ox)y °f the direct image of the structure sheaf of X under the (m + l)-st Probenius endomorphism Fm+1 on X is tilting. On the projective space the direct image of an invertible sheaf under the Probenius endomorphism splits into a direct sum of invertible sheaves as found by Hartshorne [HASV]. We will describe in §4 the multiplicity of each direct summand to find when (F™+10x)v is tilting. On the flag variety of SL3 a close analysis of the Humphreys-Verma module associated to {F™+lOxY wil1 yield in §5 that (F*Ofi)v is tilting iff p > 3. Notations: If C is a category, C(A, B) will denote the set of morphisms in C from object A to object B. If X is a variety, Modx (resp. coh(X) ) will denote the category of (resp. coherent) Ox-modules; if R is a ring, Rmod (resp. mo&R) will denote the category of left (resp. right) R-modules of finite type. For all the other unexplained notations refer to [J]. Acknowledgements: The project was started while the second author visited the third in February of 2004; the second author wishes to thank University of Warwick for the hospitality. He is also grateful to Asashiba H. for consultation in ring theory, to Gyoja A. for encouragement, and to American Mathematical Society for financial assistance during the stay of the Humphreys conference. We are much indebted to Roman Bezrukavnikov for allowing us to include a result of his yet to be published. 1° Arithmetic differential operators Let X be a smooth variety over an algebraically closed field k of characteristic p > 0, and Viff = Viffx = Viffx/k the sheaf of k-algebras of differential operators on X. In positive characteristic, in addition to the filtration by orders, there is another filtration on Viff p(°) < pt1) < ... < pW = Modoxirn+1) (Ox, Ox)<"-< Viff = hmf>^\ m
ON LOCALIZATION OF D-MODULES 45 called the p-filtration, where if Fx (resp. F™) is the ra-th absolute Probenius endomorphism on X (resp. Spec(k)), structure Spec(k) structure Spec(k). Each V^ admits a lift £><m\ which is denned over Z(p) p J( b} when X is denned over Z(p) [B96]: £>(m) {f GQ I a,be Z with TO// with Viff-limV^l £>(m) Berthelot calls £>(m) the sheaf of arithmetic differential operators of level m on X/k. While p(m) is by definition Morita equivalent to 0X(m+i), Berthelot's specific local description of £>(m) yields that p(m) is a central reduction of V^m\ that 2)(m) is Azumaya [BMR]/[K], and allows us to write down a simple presentation; Berthelot's construction [B96, 2.2] of £>(m) is very intricate and elaborate. In fact, Bezrukavnikov et al. [BMR] define their sheaf of crystalline differential operators by the following presentation for m = 0 and prove directly that it is Azumaya; their work is entirely independent of Berthelot's construction. Proposition: IfViff71, n e N, is the sheaf of differential operators of order < n and ifrT^{Viff2prn~1) is the tensor algebra ofViff2?™"1 overk, V{m) „ TiCDiff2*7"-1)/^ - Xl0x ,5®5'-5'®5-[5,5'],5®5" - 55" \ A e k, 5" e Vifr™-1; 5,5' e Viff*m). Proof: By [B96, 2.2.3.2] £>(m) ■Biff 2>£2-i-------2X//2*m-1, where V^l _ i is the sheaf of differential operators of level m and of order < 2pm — 1. If 'p(™) = Tk(X>i//2Pm-i)/(A - \l0x, 6 ®6' -6' ®6-[6,6'}, 6 ® 5" - 66" | A e k,6" e Viffrm-l;8,6' e Vifpm), one therefore obtains, as [vim)Mm)] < !>£>_!
46 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN Vr, s by [B96, 2.2.4(iv)], a commutative diagram Tfc(2X//»--i) q"°tient > 'P<"> I "-^ ITT "^ ^ Y Viff2Pm-K ^ £>(m). To see that ir is an isomorphism, the question being local, we may assume X admits a coordinate. By [B96, 2.2.4.1 and 2.2.5.1], as Viff2^'1 ~ V^i_^ any element of ;£>(m) can be written as an Ox-linear combination of dimX m—1 n«n (9f]rj)(dr])ai}, ^ e m,* e n. 2=1 J=0 But [B96, 2.2.4.1, 2.2.5.1 and 2.2.3] together assert that those elements form an Ox-linear basis of V^m\ It follows that 7r is invertible. 2° Localization on the flag variety (2.1) Let X = G/B be a flag variet}/-, which we will denote by S, with G a simply connected simple algebraic group over k and B a Borel subgroup of G. In characteristic 0 a basic result in the representation theory of G is the localization theorem of Beilinson and Bernstein [BB]: there is a natural isomorphism of k-algebras, k being momentarily of characteristic 0, (Ll) \J°~Diff:=r(B,Viff), where U° = U ®3HC ko is a central reduction of the universal enveloping algebra U of the Lie algebra of G, and there is a categorical equivalence r(B,?) (L2) coh(Viff) - Diffmod T>iff®Diff ? such that the functors T(S, ?) and Viff<S)Diff ? are quasi-inverse to each other. Back to positive characteristic, however, Kashiwara and Lauritzen [KLa] have crushed a hope of carrying the localization theorem over to positive characteristic by showing in SL& that IT(B, Viff) ^ 0 for some i > 1. It follows, in particular, that H2(S, 2>(m)) ^ 0 for some m. Nevertheless, for m = 0, if gr(£>(°)) is the graded algebra of 2>(°) with respect to the order filtration, IT (B, gr(£>(0))) ~ ff(B, £(S(g/b))) with g (resp. b) the Lie algebra of G (resp. J5) = 0 \/i > 1 by [Haa87, 4.1.1], so that IP(S,£>(0)) = 0 Vi > 1, and Bezrukavnikov, Mirkovic and Rumynin [BMR] show for p = ch (k) > h the Coxeter number of G that there is a natural isomorphism of k-algebras U°-D<°>:=r(B,P<°>),
ON LOCALIZATION OF D-MODULES 47 and that (L2) survives as a derived equivalence Kr(#,?) D6(coh(P(°))) *~ D6(L>(°)mod), ^®D(0))' which further induces a derived equivalence D6(coh0(P(0))) ~ D6(U°mod0), where coh0(£>(0)) = {M e coh(P(°>) | mn.M = 0 3n e N}, m = (x*-xM | x e g) < 3pt = k[xP -xW\xe g], and U°mod0 = {M e U°mod | mnM = 0 3n e N}. Consider now the central reductions Z>(°> (g)z(p(o)) (Z(P(°))/mZ(I>(0))) ~ p(°) and U (g)3Fr (3pr/m) ~ Dist(Gi) with Gx = ker(FG/k : G -> G*1*) the Probenius kernel of G. If Dist(Gi)0 is the central reduction of Dist(Gi) with respect to the Harish-Chandra center corresponding to U°, the determination of the irreducible Dist(G?i)°-T characters, T a maximal torus of J3, will solve Lusztig's conjecture for the irreducible characters of G. Could the Bezrukavnikov-Mirkovic-Rumynin (BMR for short) localization theorem be further cut down to a derived equivalence D6(coh(2M°))) ~ D6(Dist(Gi)°mod)? The following result of Bezrukavnikov answers negatively on (LI), while (L2) may have a chance as we demonstrate in §§4 and 5. (2.2) Theorem (Bezrukavnikov): Assume rkG > 2 and p > 2(h — 1). Then Vra G N, the natural homomorphism of k-algebras is not surjective. Proof: Vra e N+, let F = F^/k : B -> S(m) be the ra-th Probenius morphism relative to k, and consider the commutative diagram G/B = B >- B^m\ G/GmB By [J, Rmk.I.5.19] there is an isomorphism of G-equivariant 0G/GrnB-modules q+0B*CG/GmB(wd%"Bk). As F is invertible, one obtains an action of Dist(Gm) on £<3/<3m#(ind£mBk)) via Dist(Gm) ^ Mod (m) (Ob, Ob) I l o Y ModG/GmB(£G/GmB(ind^Bk),£G/GmB(indg-Bk)) ^— ModG/GmB(q*OB,q*0B). If x0 = GmB in G/GmB, there is an isomorphism of GmJ5-modules eve : CG/GrnB(md<j>™Bk)(x0) - ind^Bk,
48 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN and hence going to the residue field at x0 yields a commutative diagram Dist(Gm) ^ T(B, V^-V) Y Y Modk(indg-Bk,ind^Bk)^ ModG/GmB(£G/GmB(ind^Bk),£G/GmB(ind^Bk)). Just suppose the homomorphism Dist(Gm) —> r(S, p(m-1)) is surjective. Recall that the Mehta-Ramanan-Ramanathan Probenius splitting of B yields a decomposition of 0<3/<3mB-modules [K95, 3.2] £G/GmB(ind<rBk) ~ 0G/GmB e£G/GmB((ind|'"Bk)/k). Let £ e ModG/GmB(£G/GmB(ind^mBk),£G/GmB(indf'"Bk)) be the idempotent inducing the projection onto Oo/GmB- Suppose \x € Dist(Gm) is mapped to e. Then 1 = rk(e(so)) = rk(MlindG™Bk) = rk Hndg~k) as (indg-Bk)|Gra ~ indg-k by [J, II.9.1.3] > [ind^k : k] as p\k = ^\oG/GrnB(Xo) = idoG/GmB(*o)- Now let A be the character group of maximal torus T, and for each A G A let Lm(X) be the simple GmT-module of highest weight A. We choose a positive system of roots i?+ such that the roots of B are negative. If ao is the highest short root, then by [J, II.9.1.3 and 4] [indg-k : k]Gm > [ind^k : k]GmT + [indg-?k : Lm(-p™a0)]cmT = l + [indg-?k:Lm(-pma0)]OmT. Moreover, HKk : Wi(-Pm+1ao)]Gm+1r = Eiind^:?k : ^m(pm»7)]GmT[ind|^(r?) : Li(-pao)]Glr rjeA by [DS, 3.2] > [indg-Jk : k]GmT[ind%£k : Li(-pa0)]GlT = [ind^k : Li(-JK*o)]giT. Now let *4 denote the set of alcoves and for each A G A let Oa be the image of 0 G A in A under the •-action of the affine Weyl group of G. For each v G pA let n~ = {A G A | (A + p, aV) G [(i/, aV) -p, (i/, aV)] Vz G /} with {a, | i G /} the set of simple roots, and let Wv be the Weyl group around the special point v — p. Then by [Ye] {AeA\ [ind£g(0,0 : ii(0AlpQo)]GlT ? 0} = {.4 e ^ | ^ c n:pQO+pp}w_pao+pp, where At.pao is the translation of the bottom dominant alcove A+ by — pao- Thus [indggk : ii(-pa0)]GlT = [indf,g(<U+) : Li(0A+ )]GlT ^ 0
ON LOCALIZATION OF D-MODULES 49 if thre is w G W-pOi0+pp such that A+w C II_pao+p/9, i.e., w x • 0 G II_pao+p/9, leading to a contradiction. To obtain such w, we have only to find w e W such that Vz G /, (1) [-p(a0,a2v),-p(a0,a2v) + p] 3 p((a0 - p, w'1^) + 1 - (a0, a2v)) + (p, w-1a2v). If In = {i e I \ (ao5 aV) = n}, n G N, one checks by inspection that I = I0 U I\. Then (1) reads as vn . / -l v\ r- J °>P if * e/0 [ -p,0 lfzG/i, p((a0-p,w 1a2v) + l-(a0, i.e., Vz G /, p(ao — p5 w-1^) + (p, iu-1^) g] —p, 0[. But one can find by inspection w eW such that Vz G /, either (ao—p, wa^) = — 1 with wo:; > 0 or (ao — P, wctf) = 0 with woti < 0, as desired. (2.3) Remarks: (i) In case G = SX2, due to [Sm], one has a commutative diagram -^. /^v,n not surjective „,—. ^ „„x Dist(G) * » r(B, Pi//) limDist(GTO+i) r(£,limP<ro)) limr(B,P(m)). ra As lim is exact, there must be some m G N such that Dist(C?m+i) A ^(S, f>^). Now recall the short exact sequence (1) 0 -> /C0 -> £>(0) -> P(0) -> 0. As 2M°) = Viffp-i in 5L2, one has a commutative diagram of Og-modules P(°) A ■p<°>. p. (0) ■p-1 It follows that the sequence (1) splits over Ob- As W(B,V^) = 0 Vi > 1, we obtain Hi(B,/Co) = 0 Vt>l, and hence a commutative diagram (2) r(B,x>(°>)—-r(B,p(°>) u Dist(Gi).
50 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN Consequently, the surjectivity of Dist(Gi) —► T(B,V^) follows from that of U —> r(£,X>(°)). We claim: Ifp = 2, Dist(Gi) > r(£,Z>(°)), hence abo U > T(B,V^). On the other hand, ifp > 3, Dist(Gi) -» T(B,f>^). To see that, assume first p = 2. By [Haa86, 5.12.3] if t is a local coordinate of U+B/B and if d{i} = % with d=^-, ' i\ dt' p—l i+1 p (3) r(B,f>^) = kie]]L{(]]Lktjd^)e ]J k(tkd^- i=l j=0 fc=i+2 /c-(i+2) fy - fc + i + 1 + f\ f-p + fc - 1 = kl 0 kd 0 k£<9 0 k£2<9. ^=0 ^ % ' ^ If *(a,6,c), (a,6,c) e {0,1}3, is the image of e(a)rj/(c) in r(B,P<°>) with e,h,f the standard basis of g, one has from [Haa86, 5.12.4] (4) <w*»)=(-1)^(7) ("2(;+c)) (T>n+c-a- Consequently, $(o,o,o) = 1; *(o,o,i) (*n) = ntn+1, so *(o,o,i) = ^ $(o,i,o) (tn) = -2ntn = 0, so $(o,i,o) = 0; *(i,o,o)(*n) = -n*n"1, so *(1,0,o) = d. It follows that the image of Dist(Gi) in T(B, P(0)) is kl 0 kd 0 kt2d < T(B, P(0)). Assume next p>S_>h. Then U -» r(S,£>(0)) by [BMR, 3.3.1.U], and hence also Dist(Gi) -» r(B,P(°)) by (2). (ii) Back to the general set up, let W be the Weyl group of G. For each w eW we have constructed in [K04]/[K] a T-equivariant f>^ -module Zw^m) to yield an isomorphism of Gm+iT-modules r(6,£ty,(m)) ~ Dist(Gm+i) <8>Dist(™Bm+i) kw;#0-(p^+1-l)(p+«;p)5 which is a Humphreys-Verma module of character / „x tt l-e(-pm+1a) ±x 1 —e(—a) While T(S, ZW)(m)) is not irreducible as Gm+iT-module, Zw^m^ is as P(m)-module. 3° Tilting sheaves Let X be a smooth projective variety over k.
ON LOCALIZATION OF D-MODULES 51 (3.1) As X is projective, any M G coh(X) admits a resolution by locally free sheaves of finite rank, and of length < dimX as X is smooth. In particular, if C -» M for some locally free C of finite rank, Modx(.M, M) < Modx(£, M) ~ T(X, £v ®x M), so that lS/lodx(M, M) is finite dimensional over k by Serre's theorem. Then by a result of Auslander [Rot, 9.23] the left and the right global dimensions of ring ModxOM, M) coincide, which we will denote by gldimModx(.M,M). After [Ba] we say T G coh(X) is tilting iff (Tl) Exfx(T,T) = 0 Vi > 1; (T2) T Karoubian (K- for short) generates D6(coh(X)), i.e., D6(coh(X)) coincides with the smallest full triangulated subcategory of D6(coh(X)) containing T which is closed under taking direct summands; (T3) gldim Modx (T, T) < oo. (3.2) Beilinson's lemma [Be], [Ba]: Let T G coh(X) and T = Mod* (T,T). The following are equivalent: (i) T is tilting. (ii) There is a triangulated equivalence MModx(T,?) D6(coh(X)) c D6(modT), whereT acts onModx(T,M), M G coh(X), viaf-b = /(6-?), / eModx(T,M), beT. (iii) T/iere is a triangulated equivalence $ : D6(coh(X)) —> D6(modT) s^c/i that $(T) ~ T. (3.3) Variations on (T3): (i) [Ba, 3.2.1] (T3j ma?/ 6e replaced by the existence of some tilting sheaf on X. (ii) [Bo] Let Mi,...,Mn G coh(X) with Modx(Mi,Mi) ~ k Vi. Define a graph by letting the Mi be the vertices and assigning an arrow from Mi to Mj, i ¥" h iffModx(Mi, Mj) ^ 0. If the graph does not contain a circuit, then gldimModx([[Mi,]J[Mi) < n. i i For completeness let us sketch a proof of each statement: (i) Let T be a tilting sheaf on X and put C = MLodx(Jr, T). By Beilinson's lemma MModx(.F,?) : D6(coh(X)) -> D6(modC) gives a triangulated equivalence. Put $ = RModx(^, ?) and M = $(T). Recall from [II, Cor. II.2.2.1] that D6(coh(X)) ~ D*oh(X) the full subcategory of the bounded derived category Db(X) of Ox-modules consisting of the complexes whose cohomologies are all coherent. As C is finite dimensional over k, D6(modC) ~ D|n(ModC) the full subcategory of D6(ModC) consisting of the complexes whose cohomologies are all of finite type over C [Mi, 10.14]. Then
52 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN \/ieZ, Bb(ModC)(M,M\i]) ~ Db(modC)(M,M\i]) ~ Db (coh(X)) (T,T\i]) ^D6(I)(r,TH)^E4(T,T) _ f T if i = 0, [0 otherwise by (Tl). Also M K-generates D6(modC) by (T2). But D6(modC) ~ K6(projC) as gldimC < oo [Hap, p.29]. It follows that M forms a tilting complex over C [KZ, Def. 1.3.2.1]. Then D6(ModC) ~ D6(ModT°P) by Rickard [KZ, Th. 1.3.2.1], consequently oo > gldimTop = gldimT [Hap, Lem., p. 101], and hence (T3). The following argument for (ii) is provided by Asashiba H. Put M = ]\i Mi, E = Modx(M,M), and e* G E the idempotent such that eiM = Mi \/i. As Modx(Mi, Mi) ~ k, each Mi is indecomposable. Thus 1 = e\ + • • • + en is a decomposition into primitive idempotents. Mi ^ j, by the hypothesis either eiEej ~ Modx(ejM,eiM) ~ Modx(Mj,Mi) = 0 or ejEei ~ Modx(Mi,Mj) = 0. It follows that Eej qk Eei as E'-modules. Thus E forms a basic finite dimensional k-algebra. After [ARS] define the quiver Q of E by enumerating the set of vertices as 1,... ,n and inserting as many as dime~j(J/J2)e~i arrows from i to j for e* = ei + JeE/J. Vi^j, dimej(J'/J2)e~i < e^Eei = dimModx(Mi,Mj) while dime~i(J/J2)ei < dimeiJe; with ti Jei = eiEei n J by [NT, 1.3.9] = ke* D J as eiEei ~ Modx(Mi, Mi) ~ k = 0 as ei £ J. It follows from the hypothesis that Q has neither any circuits nor any cycles. We now claim pd(£'ei) < n — 1 Vz, which will force gldimE1 < n — 1. Start with an exact sequence of E'-modules 0 —► Jei —> Ee^ —> l^e* —> 0. If Je* ^ 0, there is A ^ [1, ri] and aij G N+, j e h, to form an exact sequence 0 ^ II (Jej)eaii -► II (£e*)0ai'" -> ^/J2e, -+ 0. Then by NAK there is ifi < \ljeIl(Jej)®aij to yield an exact sequence j€/i If i^i ^ 0, there is 72 S [1? ft] and a^- G N+, j G ^ to form an exact sequence 0 -> II (^)ea2j -> U (Eej)®0** -> /fi/J/fi -> 0. Then there will be K2 < Hjei2(Jej)®a2j to yield an exact sequence 0 -> /f2 -> ]J (Ee,-)0"1' -> /fi -> 0. jeh
ON LOCALIZATION OF D-MODULES 53 Repeat to obtain a projective resolution ]J iEe,)®*™* -> > H {Eej)®a^ -> Eei -> £e; -> 0. j€/m jeh By construction for each j G 7m there is k G /m-i such that 0 ^ ElSA.o&{Eej,Jek/J2ek) — ej(J/J2)ek, and hence there is an arrow from fc to j in Q. It follows that Q has a sequence of arrows jm <- jfm-1 < <~ jl <~ * With jfc G Ik- As Q does not have a circuit, however, we must have m + 1 < n, and hence pd(2£ei) < n — 1, as desired. (3.4) Proposition: Let m G N. T/ie following are equivalent: (i) Afodx(^xtlox,£M is a ftftm^ sfcea/. (ii) The derived localization theorem of Beilinson and Bernstein Rr(x,?) D6(coh(P(m))) c D6(£>(m)mod) ^(m)®c(m)7 fco/cb and W(X, f>^) = 0 for alii > 1. Proof: We have by definition a Morita equivalence, called the Cartier-Chase-Smith categorical equivalence, Modx(rn+1)(Ox,Ox(rn+1))<g)t>(rn) ? (1) coh(P(m)) *" coh(X(m+1)). Put F = FX,T= {F^OxY = ModxiF^OxiOx), and T = Modx(T,T). Assume first T is a tilting sheaf, so by Beilinson's lemma there is an equivalence KModx(T,?) (2) D6(coh(X)) *" D6(modT). On the other hand, as F™+1Ox is locally free of finite rank over Ox, one has natural isomorphisms of rings T°p ~ ModxiF^OxiF^Ox) ^ Modx(m+1)(Ox,Ox) = D^\
54 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN which induces an equivalence of categories modT ~ D^mod. Thus together with (1) and (2) one has a derived equivalence °* ®°x(m+l) V ®T (^X/t1. i V®k(FZfc\Ox)v 1 V <8>jT , \ \ r oxy Db(coh(P(m))) \ Db(coh(X(m+1>)) V Db(coh(X)) \ Db(modT) Db(D(m>mod) V I Xodx(m+i)(Ox,C)x(m+l)) (S)^(m) V I A^odx(F„m+1Ox,Ox)(8)^(m) V I KModx(T,T(g>^(m) V), where (F™/+1^Ox)v = M^i^D^'.Ox.Ojf^+i)). We must now check that RModx(T,T ®e(m) V) ~ KT(X,V) and Ox ®ox(m+1) (V 8% (F^+\Ox)v) ~ ^(m) ®^(ra) V- As D6(cohpf(TO+1>)) ~ D*oh(X(m+1)), it follows from the Morita equivalence of V^ and 0X(m+i) that D6(coh(p(TO))) ~ D*oh(P(TO)) and that W € D6(coh(P<TO))), if V -► £ is an injective resolution in Dfc(2?(m)), T ®f,im) V -> T ®f)(m) £ remains an injective resolution in Db(X). Then RModx(T,T®p(m) V) ~ Modx(T,T ®f,(m) S) ~ Modx(0x,^odx(Fr+1Ox,Fr+1Ox) ®C(m) 5) ~ Modx(0x, *T+1£) * r(X, Ff+15) ~ r(X, S) ~ RT(X, V) as £ is flasque [KS, 2.4.6(vii), p. 99]. In turn, Ox®ox(m+1) (V ®fc (FZfcOxY) c Ox ®ox(m+1) mfcOxY «fe(ra) V) * {Ox ®ox(m+1) {F^OxY) ®hD(m) V * Modx^MF^+ktOx,F^+klOx) ®fc(ra) V ~ P(m) ®fc(m) V. One also checks the compatibility of the T- and the £>(m)-actions. For each i > 1 0 = Extx (T,T) ~ Extx(Ox, (F™+1Ox) ®x T) ~ Extx(Ox,Modx(F?+1Ox,F™+1Ox)) ~ Extx(Ox, F,m+1P<m)) ~ H*(X, F™+1P(m>) ~ rP(X, X)(TO)) by the degeneracy of the Leray spectral sequence . Conversely, if the derived localization theorem holds, then with T = Modx(m+1)((FZ+\Oxy, (FZ+iOxV)
ON LOCALIZATION OF D-MODULES 55 there will be an equivalence Ox ®ox(rn+1) M D6(coh(P(m))) Rr(X>?) > D^ODWmod) M D6(coh(X(m+1))) O D6(modT) Dfc(modT'), such that *&*xj*\OxY) = mX,Ox 0ox(m+1) {F^\OxY) ~ RTiX^Modx.m^iF^Ox.Ox)) ~ RT(X,f>^) ~ T(X,V^) as H^(X,V^) = 0 for all t > 1 = D^m\ The structure of right T'-module on D^ is given by 5.b' = (bf)v • S = (bf)v oS, Se D^\b' e T'. One then has an isomorphism of right T'-modules D^ —> T' via S i—► Sw. It will now follow from (3.2) that (F%j£OxY must be tilting on X<m+1), so therefore should be {F™+1OxY on X. 4° Projective spaces Let X = P£ and F = Fx the absolute Probenius endomorphim on X. We will abbreviate Ox(i) as 0(i), i G Z. Although the localization theorem holds for Vi//-modules on the projective spaces by Haastert [Haa87], it is not automatic that it carries oyer to P(m)-modules: 0(-pm+1) ~ 0®0im+i) (0(m+1)(-l)) admits a structure of 2Mm)-module with no global section. (4.1) Proposition: Let m G N+ = N \ 0, r G Z, and write r = vq + pmri witt ro G [0,pm[ and r\ G Z. T/ien we /mue a decomposition of O-modules n Fr{0{r)) = ]]L0{r1-i)®6> with 0i = {je[0,p7n[n+1\\j\=ro+prni}, 2=0 where \j\ = jo + ji + — - + jn- In particular, all 9i > 0 iffpm > n + ro. Proof: Proof: As F™(0(r)) ~ (F*m(0(^o))) ®x O(ri) by the projection formula, we may assume r = ro G [0,pm[. Let A = k[xo, x\,..., xn] be the polynomial Ik-algebra in xq, #i, ..., xn graded such that each Xk has degree 1, so Proj(A) = FQ. Using an equivalence of categories between the category of quasi-finitely generated graded A-modules and that of coherent Ox-modules [H, Ex. II.5.9], we will argue with graded A-modules. The
56 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN graded >l-module associated to F™(0(r)) is r.(FT(0(r))) = Hr(x,(FT(o(r)))(e)) £ez ~ JJ T(X, F™(C*(r + pm£))) by the projection formula again £ez = H T(X, 0{r + pmi)) = JJ T(X, 0(r + pm£)) eez £eN ~ Y^Ar+prnt, £ where At denotes the ^-th homogeneous part of A and the structure of graded >l-module on ]J^ Ar+pm£ is given by m Xk - a = x\ a Vfc G [0, n\. On the other hand, Wi G N, the graded >l-module associated to 0(—i) is r.(O(-0) = ]Jr(x,(0H))W) = ]Jr(x,o(*-o) * II ^-* £ez £ez £eN with the natural structure of graded >l-module: Xk • a = x^a. Now put q = pm and let V% = TT kxj for each i G N. We have then an \j\=r+qi isomorphism of k-linear spaces Ar+q£ - JJ(V* 0k A-i) via abq <-\ a ® 6 Va G V*, 6 G At-i. It follows that there is an isomorphism of graded ^-modules II^+^-II^®^!!^^)}, ££N i€N £eN where the structure of graded A-module on the RHS is given by the one on ]J^ Ai-i. Thus 0i = dim V*. Finally, (q — l)(n + 1) > qn + r iff g > n + 1 + r iff q > n + r, and hence the assertion holds. (4.2) Theorem: Vra G N+, (F™0)v 25 a taftm# sheaf on P£ iffn<pm. Proof: While 0(-pm) - Fm*(C7(-l)) admits a structure of P^"1)-module, if n > pm, Rr(X, 0(— p171)) = 0, so that there cannot be a derived equivalence Rr(x,?) D^con^™"1))) c D^D^-^mod). ^(m"1)®5f>(m-.i) ? Thus (F™0)v is not tilting for n>pm. Now put T = (FJJnO)v. In the case of the projective space for T to be tilting, (Tl) is immediate from (4.1) and the condition (T3) is superfluous by (3.3.i); lir=o°W is a tiltinS sheaf [Be]/[Ba, 4.1.1]. Assume that if n < pm. Then by (4.1) with r = 0, all 0(—i), 2 G [0,n], appear as direct summands of F™0. Let
ON LOCALIZATION OF D-MODULES 57 (T) be the full triangulated subcategory of D6(coh(X)) generated by T. For each M e coh(X) there is k e N such that O(-k) <g)k T(X,M) -» M. It thus suffices to show that each 0(—k) e (T), k e N+. Recall the Koszul resolution of X: if V = T(X, 0(1)), which is of dimension n + 1, (1) 0 -► 0(-n - 1) <g)k A£+1F -► O(-n) <g)k A£F -► ... -► O(-l) 0k Aj^ -► Ox -► 0. As the sequence locally splits, taking the dual and tensoring with 0(—k) yields another exact sequence (2) 0 <- 0(-k + n + 1) ®k A£+1V <- 0(-fc + n) <g)k A£F <- ... <- 0(-k + 1) ®k a£V <- O(-fc) <- 0. It follows by induction that O(-k) e (T). 5° The flag variety in SL3 In case G is of type A2 (resp. B2) we know by Haastert [Haa87] (resp. [AKOO]) that IT(B,P(o)) = 0 Vt>l. We will verify in this section that the BMR localization theorem carries over to P<°> inSL3 iff p > 3. (5.1) Let A be the character group of the maximal torus T and A+ the set of dominant weights such that the roots of B are negative. If A G A+ we let A (A) (resp. L(A)) denote the Weyl (resp. simple) G-module of highest weight A, and put V(A) = r(S, £(A)) with £(A) the invertible 0#-module induced by the J9-module A. Let A e A+. Recall from [J, II.8.5] that there is a morphism j\: B —► P(V(A)) such that £(A) ~ ,7'£(0p(v(A))(l))- Therefore, putting n + 1 = dim V(A), tensoring with 0(n + 1) and taking ft, the exact sequence (4.2.1) yields an exact sequence (1) o - Ob - C(Xf(n^) - £(2A)<-i) - • • • - £(fcA)<#-*) - ► C(nX)<^) -+ C((n + 1)A) -+ 0. Now let cji and u>2 be the fundamental weights for our SL3 corresponding to the simple roots ai and a2. Put p = u)\ + u2. Taking A = Ui, i e {1,2}, the exact sequence (1) reads as (2) 0 -► Ob -► £(^)03 -+ £(2c^)03 -► >C(3^) -+ 0. If S is a collection of the objects of coh(S), let (S) be the subcategory of D6(coh(S)) K-generated by S. Lemma: (0B,£(a;i),£(2a;i)i Ap) I * = 1,2) = D6(coh(S)). Proof: Put S = {Ob, C{uoi), £(2^), C{p) \ i = 1,2}. If M € coh(B), Ob ®b r(B, A4 <8>s £(np)) -» A4 ®s £(np) Vn » 0,
58 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN and hence C(—np) <8>b T(B, M ®& £(np)) -» M. It is therefore enough to show that £(A) G (S) VA G A. For that we have with the help of the exact sequence (2) only to show (3) C(M + ku2)e(S) Vj,ke[0,2]. As Ob, C(uj{), C(2uji) G (5), one obtains using (2) all C{pw{) G (S), n G Z. By symmetry C(nu2) G (5) Vn G Z. One has (S) 3 £(p)(g)kV(wi) ~ £(p®kV(wi)). As chV(u;i) = e(uj\)+e{—uj\+(jo2)+e{—uj2), £(p)®kV(u;i) admits a nitration with the subquotients £(pH-u;i) = £(2u;i + uj2),C(p — w\ +^2) = C(2u2),C(p — uj2) = C(ui). It follows that C(2ui +(jo2) G (5). By symmetry C(u)\ + 2a;2) G (5). Then using (2) again yields C(nui + cj2), C(ui + nw2) G (5) Vn G Z. Likewise from C(2ui +w2) ®k V^) one obtains C(2p) G (5). Then C(2ui +nuj2) G (5) Vn G Z by (2), and (3) holds. (5.2) Before we go on, let us take care of the case p = 2. If F&/k : B —► B^ is the relative Probenius morphism on S, C(—2uj{) ~ F^k(C(—uJi)^) is equipped with a structure of p(°)-module while H*(S,£(-2wi)) = 0, and hence RT(B,?) cannot be an equivalence. (5.3) Assume from now on throughout the rest of the paper that p > 3. Let F = F& : B —> B be the absolute Probenius morphism on S. Lemma: (F*06)v K-generatesDb(coh(B)). Proof: We will show that (F*OsY decomposes in Mod^ as (1) {F*0B)y ~ Ob 0 {C(p) 0k L((p - 2)p)} 0 {£(0;!) 0k L(wi + (p - 2)w2))} e {£(w2) ®k ^((p - 2)wi + w2))} e c(p) 0 {£(Mi) 0k L((p - 3)a;i)} 0 {£(M2) 0k L((p - 3)w2)}, where M\ and M2 are J3-modules fitting into nonsplit exact sequences of J3-modules 0 —> cj2 —> Mi —> 2u>i —► 0, 0 —> cji —> M2 —> 2cj2 —> 0. Then the assertion will follow from (5.1). Recall a commutative diagram of schemes Fb „ structure G/GiB ^ gU) ^ Spec(k), where 0 is invertible as morphism of schemes. We will often identify G/G\B with B through 0. Thus for each /x G A if L(/x) is the simple GiJ3-module of highest weight p, and if /x = /x° + p/x1 with /x° G A+ such that (/x°, a;) < p\/i, (L(fi))^C(fi1)®kL(fi°).
ON LOCALIZATION OF D-MODULES 59 Recall also isomorphisms of Oq/Gi b-modules {F*OsY ~ {q*0By ~ £G/GlB(V(k))v ~ £G/GlB((V(k))*) -£G/GlB(V(2(p-l)p)), where V = ind^1 is the Humphreys-Verma induction functor. For p > 3 we know from [K89, 4.5.i and 4.15] and [AK89] the GiT-socle series soc7 of V(2(p — l)p): if soc^ = soc^/soc7-1, suppressing k from (gfc, (2) soc = soci = L(2(p — l)p), 2 soc2 = L(pp) 0 JJ{L((p - 3)ui) <g> GiMod(L((p - 3)^), soc2)} 2=1 0 L((p + l)ui + (p - 2)u2) 0 L((p - 2)ui + (p + l)w2), soc3 = L((p + l)ui + (p - 2)u2) 0 L((p - 2)ui + (p + l)w2), SOC4 = k, each of which is equipped with a structure of G\B-module and the decomposition is the one as GiJ3-modules; GiMod(L((p — 3)^),soc2) fit into exact sequences of J3-modules (3) 0 -> pun ^GiMod(L((p - 3)^2), soc2) -> 2pw2 -► 0, (4) 0 -► pu2 ^GiMod(L((p - 3)a;i), soc2) -> 2p^i -► 0. Put T = (F*0e)v. The Probenius splitting of B splits off Ob ~ CG/GlB{k) from T to yield a decomposition of Og-modules T~Ob(BCg/GiB(soc3). Also the inclusion £g/Gib(soc) ^ £g/Gib(soc3) splits in ModG/dB to yield £g/Gib(soc3) ^ CG/GlB(soc)eCG/GlB(soc3/soc) - {£(p) ® L((p - 2)p)} 0 £G/GlB(soc3/soc); using Kempf's vanishing theorem one checks ExtG/G1B(£G/G1B(sOC3/sOc),£G/GlB(sOc)) = 0. Likewise the exact sequence 0 -► CG/GlB(soc2) -> £G/GlB(soc3/soc) -> CG/GlB(soc3) -► 0 splits over Og/GiB. Consider finally CG/GlB(soc2). By (2) one has a decomposition 2 2=1 0 {£(u>i) <8> L(wi + (p - 2)u2)} 0 {C{uj2) <8> L((p - 2)ui + u2)}. Note that Ext1B(C(2u1), C{uo2)) ^ H1^, £(w2 - 2wi)) ~ r(B, Ob) + 0 by [J, II.5.5] while Extg(T,T) ~ H^S, V^) = 0. We must therefore have exact sequences (3) and (4) both nonsplit; take M2- = GiMod(L((p - 3)a;i),soc2)'~1', i = 1,2.
60 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN (5.4) At this point we have verified conditions (Tl) and (T2) for T = (F*Ob)v. In order to see (T3) also holding, we may instead show by (3.3.i) that 2 2 T = 0B 0 C(p) 0 H C(ui) 0 ]J £(GiMod(L((p - 3)o;i), soc2)I-1]) 2=1 2=1 forms a tilting sheaf on B. As Tf is a direct summand of T, (Tl) holds on T while (T2) follows from (5.2). To check (T3) holding, we have only to show that Ob, £(p), £(wi), £(w2), >C(GiMod(L((p-3)u;1),soc2)[-1]), and £(GiMod(L((p- 3)a;2),soc2)[-1^) form a strong exceptional collection, i.e., verifying the condition (3.3.ii). Put Mi = GiMod(L((p - 3)w;),soc2)[~1], i = 1,2. One has Mods(00,00) ~ ModB(£(wi),£(a;i)) ~ ModB(£(w2),£(w2)) ~ ModB(£(p), £(p)) ~ r(B, Ob) c^ k. As Extg (Tf,Tf) = 0, one has an exact diagram Mods(£(2u;i),£(u;2)) MocIb(£(2u;i),£(2u;i)) 0 >■ Modfi(£(Afi), £(w2)) >■ Modfi(£(Afi), £(Afi)) ^ Modfi(£(Afi), £(2wi)) 5- 0 ModB(£(u2), C(u2)) Ext^(£(2u;i),>C(u;2)) Mods(£(w2),£(2u;i)) with ModB(£(2u;i), C(u2)) ^ r(B, £(w2 - 2wi)) = 0, Mod6(£(2u;i), £(2wi)) ~ T(S, Ob) - k, ModB(C(u2), C(u2)) ^ r(B, Ob) ~ k, ModB(£(^2), £(2wi)) - r(B, £(2wi - w2)) = 0, Ext^(£(2a;i),£(cja)) - H1^,^ - 2wi)) ^r(B,0B) by[J,II.5.5] It follows that ModB(£(Mi),£(Mi)) - k, and hence ModB (C(M2),C(M2)) ^ k as well by symmetry.
ON LOCALIZATION OF D-MODULES 61 Likewise, noting that Extg(T',T') = 0, one finds the graph for the direct summands of Tf to be (1) £(wi) C{u2) having no circuit, completing the verification of (T3) for T'. We have thus obtained (5.5) Proposition: If G — SL3, the BMR derived localization theorem for f>^ holds on the flag variety iff p > 3, induced by the the tilting sheaf (Fb*0&)w = Aiodj3(Fj3*Oj3,Oj3). In particular, for p > 3 the global dimension of D^ = Mode> {1){Ob,Ob) is finite. A proof of Jantzen conjecture, Advances in Soviet References [AK89] Andersen, H.H. and Kaneda M., Loewy series of modules for the first Probenius kernel in a reductive algebraic group, Proc. LMS (3) 59 (1989), 74-98 [AK00] Andersen, H.H. and Kaneda M., On the D-affinity of the flag variety in type B<2, Manuscripta Math. 103 (2000), no. 3, 393-399 [ARS] Auslander, M., Reiten, I. and Smal0, S., Representation Theory of Artin Algebras, Cambridge studies in adv. math. 36, Cambridge U. Press 1995 [Ba] Baer, D., Tilting sheaves in representation theory of algebras, Manus. Math. 60 (1988), 323-347 [Be] Beilinson, A. A., Coherent sheaves on Pn and problems of linear algebra, Punc. Anal. Appl. 12 (1979), 214-216. [BB] Beilinson, A. and Bernstein, J., Localisation de g-modules, C. R. Acad. Sci. Paris 292 (1981), 15-18 [BB93] Beilinson, A. and Bernstein, J., Math. 16(part 1)(1993), 1-49. [B74] Berthelot, P., Cohomologie Cristalline des Schemas de Caracteristique p > 0, Lecture Notes in Math. 407, Berline/Heidelberg 1974 (Springer-Verlag) [B96] Berthelot, P., V-modules arithmetiques I. Operateurs differentiels de niveau fini, Ann. scient. Ec. Norm. Sup. 29 (1996), 185-272. [BMR] Bezrukavnikov, R., Mirkovic, I. and Rumynin, D., Localization of modules for a semisimple Lie algebra in prime characteristic, to appear in Ann. Math. [Bo] Bondal, A. I., Representation of associative algebras and coherent sheaves, Math. USSR Izv. 34(1) (1990), 23-42 [BA] Bourbaki, N., Algebre X, Hermann 1980 [BK] Brylinsky, J. L. and Kashiwara M., Kazhdan-Lusztig conjecture and holonomic systems, Inv. Math. 64 (1981), 387-410. [DS] Doty, S. and Sullivan, J.B., Filtration patterns for representations of algebraic groups and their Probenius kernels, Math. Z. 195 (1987), 391-407. [EGAI] Grothendieck, A. and Dieudonne, J., Elements de Geometrie Algebrique I, Springer- Verlag, Berlin 1971
62 YOSHITAKE HASHIMOTO, MASAHARU KANEDA, AND DMITRIY RUMYNIN [Haa86] Haastert, B., Uber Differentialoperatoren und D)-Moduln in positiver Charakteristik, Dissertation, Univ. Hamburg, 1986 [Haa87] Haastert, B., Uber Differentialoperatoren und D)-Moduln in positiver Charakteristik, Manusc. Math. 58 (1987), 385-415 [Hap] Happel, R., Triangulated Categories in the Representations Theory of Finite Dimensional Algebras, LMS Lecture Notes 119, Cambridge U. Press 1988 [H] Hartshorne, R., Algebraic Geometry, Springer-Verlag, New York 1977 [HASV] Hartshorne, R., Ample Subvarieties of Algebraic Varieties, Springer LNM, 156 (1970). [II] Illusie, B., Existence de resolutions globales, SGA6, LNM 225, Springer, Berlin 1971 [J] Jantzen, J. C, Representations of algebraic groups, Orlando 1987 (Academic Press) [K89] Kaneda M., Extensions of modules for infinitesimal algebraic groups, J. Alg. 122 (1989), 188-210. [K95] Kaneda M., The Frobenius morphism of Schubert schemes, J. Alg. 174 (1995), 473- 488. [K04] Kaneda M., On Kashiwara's equivalence in positive characteristic, Manusc. Math. 114 (2004), 457-468 [K] Kaneda M., Direct images of T>-modules in prime characteristic, RIMS kokyuroku 1382 (2004), 154-170 [KLa] Kashiwara M. and Lauritzen, N., Local cohomology and V-affinity in positive characteristic, C. R. Acad. Sci. Paris, Ser I 335 (2002), 993-996. [KS] Kashiwara M. and Schapira, P., Sheaves on Manifolds, Springer-Verlag, Berlin 1990. [KL] Kazhdan, D. and Lusztig, G., Representations of Coxeter groups and Hecke algebras, Inv. Math. 53 (1979), 165-184 [KZ] Konig, S. and Zimmermann, A., Derived equivalences for Group Rings, Springer LNM 1685, Berlin 1998 [L80] Lusztig, G., Hecke algebras and Jantzen's generic decomposition patterns, Adv. Math. 37 (1980), 121-164 [Mat] Matsumura H., Commutative Ring Theory, Cambridge Univ. Press, Cambridge 1986 [Mi] Miyachi J., Derived Categories with Applications to Representations of Algebras, Seminar Note at Chiba Univ., 2000 [NT] Nagao H. and Tsushima Y., Representations of finite groups, Orlando 1989 (Academic Press) [Rot] Rotman, J. J., An Introduction to Homological Algebra, Academic Press, Orlando 1979 [Sm] Smith, S. P., Differential operators on the affine and projective lines in characteristic p > 0, in Seminaire d'algebre Paul Dubreil et Marie-Paule Malliavin, 37eme annee (Paris, 1985), 157-177, Lecture Notes in Math., 1220, Springer, Berlin-New York, 1986 [Ye] Ye J.-c, Filtrations of principal indecomposable modules of Frobenius kernels of reductive groups, Math. Z. 189 (1985), 515-527 558-8585 Osaka City University Department of Mathematics, Japan E-mail address: hashimotQsci.osaka-cu.ac.jp 558-8585 Osaka City University Department of Mathematics, Japan E-mail address: kanedaQsci.osaka-cu.ac.jp Mathematics Institute, University of Warwick, Coventry, CV4 7AL, England E-mail address: rumyninQmaths.warwick.ac.uk
Contemporary Mathematics Volume 413, 2006 Representations of reduced enveloping algebras and cells in the affine Weyl group J.E. Humphreys Abstract. Let G be a semisimple algebraic group over an algebraically closed field of characteristic p > 0, and let g be its Lie algebra. The crucial Lie algebra representations to understand are those associated with the reduced enveloping algebra Ux(g) for a "nilpotent" x £ 0*- We conjecture that there is a natural assignment of simple modules in a regular block to left cells in the affine Weyl group Wa (for the dual root system) lying in the two-sided cell which corresponds to the orbit of x m Lusztig's bijection. This should respect the action of the component group of Cq{x) and fit naturally into Lusztig's enriched bijection involving the characters of Cq(x)- Some evidence will be described in special cases. In order to explain the conjecture, we have to review some facts about three logically independent topics: (A) cells in affine Weyl groups, (B) nilpotent orbits, (C) Lie algebra representations in characteristic p > 0. Subtle connections between (A) and (B) have been discovered by Lusztig, while connections between (B) and (C) have emerged over several decades (notably in the work of Kac-Weisfeiler, Priedlander-Parshall, Premet, and others cited below). We hope to build further links between (A) and (C), with the goal of finding a representation-theoretic model for Lusztig's formal conjecture in [22, §10]. Notation varies considerably in the literature (and sometimes clashes). Our conventions here start with a simple, simply connected algebraic group G over an algebraically closed field K of characteristic p > 0. Let T be a maximal torus and W the Weyl group. Denote by $ the root system, with positive system 3>+ relative to a simple system A. The character group X = X(T) is the full weight lattice for <£. Let Q = Z$ be the root lattice. 1. Cells in affine Weyl groups 1.1. First we recall some basic results about cells. These arise in the work of Kazhdan and Lusztig on arbitrary Coxeter groups and their Hecke algebras, but 2000 Mathematics Subject Classification. Primary 17B05; Secondary 20F55 20G05. For helpful advice I am grateful to Roman Bezrukavnikov, Paul Gunnells, Jens C. Jantzen, Victor Ostrik, Jian-yi Shi, and Eric Sommers. ©2006 Americzm Mathematical Society 63
64 J.E. HUMPHREYS here we focus just on the case of affine Weyl groups. (See Lusztig's papers [18]-[22] as well as Shi [31], Xi [35, 36].) _ Define Wa := W k Q (the affine Weyl group) and Wa := W k X (the extended affine Weyl group). The latter is not usually a Coxeter group, but is important for Lusztig's p-adic group program; here we focus just on Wa. We say Wa is of type Xn if $ is of type Xn. It is important to note that Wa is a dual version of the usual affine Weyl group constructed by Bourbaki via the coroot lattice; this reflects the influence of Langlands duality in Lusztig's program. As in the case of an arbitrary Coxeter group, the group Wa is partitioned into two-sided cells (here denoted 0). Each of these is in turn partitioned into left cells (here denoted T) or equally well into right cells, each of which is the set of inverses of elements in some left cell. These partitions arise (together with Kazhdan-Lusztig polynomials) from comparison of the Kazhdan-Lusztig basis for the Hecke algebra with the standard basis. The definition of cells yields a natural partial ordering on the collection of two- sided cells. The highest cell in this ordering contains just the identity element 1 of Wa. Since Wa acts simply transitively on the alcoves in the affine space E := M<g)zX, the various cells can be identified with sets of alcoves. In this picture W labels the family of alcoves around the special point 0. (Conventions differ in the literature; for example, some authors work with right actions rather than left actions in this context.) 1.2. Beyond these generalities, Lusztig develops more special features of cells for Wa. Generalizing the case of a Weyl group, he defines in [19] an a-invariant a(w) for each w e Wa, constant on each two-sided cell and denoted a(Q). This is an integer between 0 and N := |$+|, defined combinatorially in terms of the Hecke algebra. The a-invariant respects (inversely) the partial ordering of two-sided cells. For example, a(O) = 0 precisely when O = {1}. At the other extreme, it turns out that there is a unique cell Q with a(Q) = N; this is the lowest two-sided cell [32]. 1.3. With the help of the a-invariant, Lusztig shows that Wa has only finitely many two-sided cells, each partitioned into finitely many left (or right) cells. It is then natural to ask how many two-sided and one-sided cells there are. These questions are extremely difficult to approach in a purely combinatorial way, though they have been answered for type An by Shi [31] and in some isolated low rank cases. To formulate and prove general conjectures, some connection with the geometry of the nilpotent variety and flag variety seems to be essential. 1.4. The lowest two-sided cell Q has been explored thoroughly by Shi [32]. It contains \W\ left cells, each obtained by intersecting Q with a Weyl chamber. The entire antidominant chamber is one left cell. On the other hand, the intersection T of O with the dominant chamber is a shifted version of this chamber: Taking 0 as the origin in E, consider the special point p (the sum of fundamental weights) at which translates of all root hyperplanes meet. Then T consists of the alcoves lying on the positive sides of all these hyperplanes. 1.5. In [20], Lusztig defines a set V of distinguished involutions in Wa, as follows. For w e Wa, let £(w) be its length and let S(w) be the degree of the Kazhdan-Lusztig polynomial PiiW(q). Then w G V iff a(w) = £(w) — 25(w), in
REDUCED ENVELOPING ALGEBRAS AND CELLS IN THE AFFINE WEYL GROUP 65 which case w is shown to be an involution. Each left cell contains a unique distinguished involution. For example, in the dominant left cell of the lowest two-sided cell D, described above, the distinguished involution belongs to the lowest of the \W\ alcoves around the special point 2p. The distinguished involution in the an- tidominant left cell of Q is the longest element of W (if 1 corresponds to the lowest dominant alcove). 1.6. Lusztig and Xi [28] show that each two-sided cell of Wa contains a canonical left cell, whose corresponding alcoves all lie in the dominant Weyl chamber C C E. In this way, C is partitioned into canonical left cells belonging to the two-sided cells. Chmutova-Ostrik [9] develop an algorithm to compute the distinguished involutions in all canonical left cells, with explicit tables given in low ranks. But it seems to be more difficult to locate these involutions in arbitrary left cells. 1.7. Pictures of the cells for the affine Weyl groups of types A2,B2, G2 are given by Lusztig [19, §11]. Paul Gunnells has used computer graphics to investigate all three-dimensional cells as well. Jian-yi Shi [31] has worked out the combinatorics in considerable detail for type An, while developing general tools such as "sign types" for the study of cells. Other affine Weyl groups of low rank have been studied in a similar spirit by him and a number of other people, including Robert Bedard, Cheng Dong Chen, Jie Du, Gregory Lawton, Feng Li, Jia Chun Liu, He Bing Rui, Nanhua Xi, Xin Fa Zhang. 2. Nilpotent orbits and cells 2.1. Denote by AT the set of nilpotent elements in q := LieG. This is the nilpotent variety (or nullcone). It consists of finitely many orbits under the adjoint action of G, partially ordered by inclusion of one orbit in the closure of another. The orbits range from {0} to the regular orbit, which is dense in Af and therefore has dimension 2N = |3>|. Whenever p is a "good" prime (as in 3.1 below), there is a G- equivariant isomorphism between Af and the unipotent variety of G. Moreover, the partially ordered set of G-orbits in Af is isomorphic to the corresponding set for the Lie algebra over C of the same type. Although Lusztig's use of unipotent classes is based in characteristic 0, the ideas therefore transfer readily to our situation. (Jantzen [17] gives a helpful account with emphasis on characteristic p.) Various other varieties and groups are associated with Af. The flag variety B of G may be identified with the collection of Borel subalgebras of g. If e e Af, the set of Borel subalgebras containing e is denoted by Be. It plays an essential role in the Springer resolution of singularities of Af, where it is referred to as a Springer fiber. Let Cc(e) be the centralizer of e in G, and denote by A(e) the finite component group Cg{^)/Cg{^)°'. The cohomology of Be with suitable coefficients (complex or Z-adic) vanishes in odd degrees and has commuting actions by the finite groups W and A(e). We write simply Hl(Be). This is the framework for the Springer Correspondence (see for example [17, §13]). 2.2. Soon after the Kazhdan-Lusztig theory was developed, Lusztig [18, 3.6] conjectured the existence of a bijection between the collection of two-sided cells of Wa (based as above on the root lattice rather than coroot lattice) and the collection
66 J.E. HUMPHREYS of unipotent classes in G (or equivalently, the collection of nilpotent orbits in g). This bijection should respect the natural partial orderings, with the cell {1} corresponding to the regular nilpotent orbit and the lowest two-sided cell corresponding to the zero orbit. (His ideas were formulated in characteristic 0 but adapt to our setting when p is good.) By combining a number of deep techniques, Lusztig was able to construct a suitable bijection in [22]. Under his bijection, if the two-sided cell Q corresponds to the orbit of some e G A/*, then a(Q) = dimSe. But the order-preserving property remained elusive except in low ranks. This was later proved combinatorially for type An by Shi, while the general case follows from recent work of Bezrukavnikov [5, Thm. 4]. 2.3. In [18, 3.6], Lusztig formulated a further conjecture on left cells in terms of the fixed points of A(e) on the cohomology of Be: (LC) The number of left cells in the two-sided cell corresponding to a nilpotent e should be equal to J2i(~l)2dimiiP(Se)A(e). Due to the vanishing of cohomology in odd degrees, the contributions here are all nonnegative. While (LC) has not yet been proved in general, it agrees with direct calculations in low ranks and with the results of Shi for type An [31, 14.4.5,15.1,17.4]. Here all component groups are trivial, while on the other hand the representation of W on the cohomology is known to be induced from the trivial character of a parabolic subgroup Wi generated by reflections relative to a set / of simple roots. (See the discussion in [17, p. 203]). When translated into the language of partitions, the number |W|/| W/| agrees with the number of left cells found by Shi for a corresponding two-sided cell. 2.4. As part of his more refined study of the "asymptotic Hecke algebra" in connection with p-adic representations, Lusztig [22, §10] formulated more detailed conjectures relating the cells with geometry. Fix a two-sided cell Q corresponding in his bijection to the orbit of e G N, and let Tq be its canonical left cell. Denote by F a maximal reductive subgroup of Cc(e), so F/F° = A(e). Write F for the set of isomorphism classes of irreducible representations of F. Lusztig's conjectural set-up involves a finite set F, acted on by A(e), with cardinality equal to the Euler characteristic of Be. The orbits of A(e) in Y should be in bijection with the left cells in O, with a singleton orbit expected to correspond to the canonical left cell. In general, the isotropy group in A(e) of an element y EY corresponds to an intermediate subgroup F D Fy D F°. The representations of F or Fy enter via a notion of "F-vector bundle" on Y or Y x Y. This formalism is then subject to several requirements in [22, 10.5]. For example, the representation of A(e) on H*(Be) should be equivalent to the permutation representation of A(e) on Y. (This recovers the statement (LC) above.) As a consequence, one should have a natural bijection between T^ nT^1 and F. For an arbitrary left cell V corresponding to the orbit of y G Y, the group F should be replaced by the group Fy. Out of this abstract framework emerges a conjectural bijection between pairs (Oe, <p) and X+, where Oe is a nilpotent orbit and <p an irreducible representation of Co(e). (Such a bijection was conjectured independently by Vogan.) Note that when we work with Wa rather than Wa, the root lattice Q replaces X.
REDUCED ENVELOPING ALGEBRAS AND CELLS IN THE AFFINE WEYL GROUP 67 Bezrukavnikov has found suitable bijections in [4] and [3]; these are shown in [5, Remark 6] to coincide. For other work related to Lusztig's conjectures (especially in this last formulation), see the individual and joint papers by Achar and Sommers [33, 2, 1], Bezrukavnikov and Ostrik [30, 7], Lusztig [23], Xi [34, 35, 36]. 3. Lie algebra representations in characteristic p > 0 3.1. The representation theory of g has been studied over a long period of time: for surveys of earlier work, see [10] and [11]. In a series of papers, Jantzen [13]-[16] has extended the theory considerably. Here we focus on just the simple modules for the universal enveloping algebra U(q). These all occur as modules for reduced enveloping algebras Ux(g), which are finite-dimensional quotients of U(q) parametrized by \ G g*. Those Ux(g) for \ in a single orbit under the coadjoint action of G are isomorphic, so one looks for a well-chosen orbit representative \. In order to obtain uniform results, Jantzen imposes several relatively weak hypotheses (H1)-(H3) on g and p, which we also assume. For a simply connected group, he requires the prime p to be good for $, which eliminates some root systems when p = 2,3,5. Moreover, the algebras s[(n, K) with p\n should be omitted (or replaced by the Lie algebras of corresponding general linear groups). Then there is always a G-equivariant isomorphism between g and g*, which transports the Jordan decomposition in g to g*. Earlier work of Kac-Weisfeiler shows that the crucial case to study is that of a nilpotent \ G g* (corresponding to some nilpotent e G g). Here one begins to make connections with the results on nilpotent orbits summarized above and with related conjectures arising in Lusztig's work [24, 25, 26, 27]. Prom now on we consider only the nilpotent case, subject to the above restrictions on p and $. 3.2. The blocks of Ux(g) have been determined by Brown and Gordon [8]. As summarized by Jantzen [16, C.5], there is a natural bijection between the blocks and the "central characters", which in turn are parametrized by the W-orbits in X/pX under the dot action w • A := w(X + p) — p. This is a Lie algebra version of the Linkage Principle. If e is the nilpotent element corresponding to x-> the component group A(e) permutes the simple modules in a block. This action is understood only in some special cases. In general the simple modules in a given block are not easy to parametrize by weights, though each can be obtained as a quotient of one or more "baby Verma modules": these are induced from one-dimensional modules for a Borel subalgebra b satisfying x(&) = 0. The choice of b affects this construction when x 7^ 0 if #e has more than one irreducible component. 3.3. To make contact with the geometry of A/*, we look only at regular blocks: those for which the weight parameters attached to simple modules lie inside alcoves. This requires p > h (where h is the Coxeter number). Jantzen's translation functors then furnish information about other blocks. For a regular block of Ux(g), the work of Bezrukavnikov, Mirkovic, and Rumynin provides a geometric interpretation. Under the assumption that p > h, they prove that the number of nonisomorphic simple modules in the block is equal to the Euler characteristic of the Springer fiber Be: see [6, 5.4.3, 7.1.1].
68 J.E. HUMPHREYS 3.4. The best understood case involves a nilpotent orbit in g* containing some X in standard Levi form, which means that the corresponding nilpotent element e is regular in some Levi subalgebra of a parabolic subalgebra p/ of g (determined by a set / of simple roots). All nilpotent orbits satisfy this condition for q = ${(n, K), but in general things get more complicated. (See [13, §10], [15, §2], [16, D.l].) Jantzen has studied simple Ux(g)-modules (and their projective covers) in considerable detail when \ has standard Levi form. In particular, each simple module can be labelled as LX(X) for one or more X e X. Here LX(X) = Lx(p) if and only if fi G Wi - X + pX, where Wj is the subgroup of W generated by simple reflections for a e I and w • A := w(X + p) — p. This can be pictured in terms of the alcove geometry of Wa, with the origin of the affine space E taken to be — p and the translations all multiplied by p. Jantzen calls the group Wp in this setting. Fixing a weight A inside the lowest dominant alcove, the orbit Wp • A under the natural dot action contains (with periodic repetitions) all weights needed to parametrize the simple modules in a single regular block. In fact, it suffices to work with the \W\ alcoves surrounding a single special point such as —p. Then the induced action of Wi on these alcoves identifies those which correspond to the same simple module. 3.5. In [14, 16], Jantzen has also studied in depth the case of a subregular x- its G-orbit has dimension 2N — 2, where N = |$+|. Only in types An and Bn does such an orbit have a representative in standard Levi form. But the simple modules in a regular block of Ux(fl) can be correlated closely with the irreducible components of Be (here a Dynkin curve), which helps to bypass the problem of labelling by weights. 4. Simple modules and left cells 4.1. Here we suggest closer connections between the representation theory discussed in §3 and the cells in Wp. While our ideas are speculative, they have some support from computations in special cases (including unpublished work of Jantzen as well as [12]). Fix a regular block of Ux(g), with \ nilpotent, and denote by S a complete set of nonisomorphic simple modules in this block. As suggested by Bezrukavnikov, this is a candidate for the finite set Y in Lusztig's formulation discussed in §2. If x corresponds to e e g, denote by C the collection of left cells of the two-sided cell Ct corresponding in Lusztig's bijection to the orbit of e. In case the component group A(e) is trivial, the cardinalities of S and C are expected to be the same: compare the theorem of [6] cited in §3 with the conjecture (LC) in §2. Conjecture. Fix notation as above. (a) There is a natural map <p from S onto C, whose fibers are the orbits of A(e) in S. (b) A simple module fixed by A(e) maps under <p to the canonical left cell F in ft. (We call this module (icanonical'\) 4.2. The meaning of "natural" in part (a) of the conjecture has to be clarified. What we have in mind is a simple recipe for assigning modules to left cells, but it has only been made rigorous in special cases. Consider for example the case when X has standard Levi form, so the modules in S can be parametrized by weights in a Wp-orbit which lie in alcoves surrounding any given special point v G E. Suppose
REDUCED ENVELOPING ALGEBRAS AND CELLS IN THE AFFINE WEYL GROUP 69 v can be chosen inside the dominant Weyl chamber in such a way that weights in those surrounding alcoves which lie in the canonical left cell F suffice to parametrize S. If w e Wp = Wa labels one of these alcoves, assign the corresponding simple module to the left cell in Q, containing the alcove labelled by w~l. (It would still have to be shown that this assignment is independent of the choice of the special point.) In particular, when w labels the distinguished involution in T, then w = w~1', so the simple module in S corresponding to this alcove is assigned to V. That this "canonical" simple module should be fixed by A(e) is suggested by the parallel discussion in [22, 10.7]. In rank 2 cases all of this can be observed directly. But in general there are serious combinatorial difficulties in working with the geometry of the cells even in the good case when \ has standard Levi form. The first problem is to locate a suitable special point v. One might look at the alcove containing the distinguished involution in the canonical left cell T: this will be the lowest alcove in T attached to some special point v. Do the surrounding alcoves which lie in F suffice to account for all simple modules in 5? In rank 3, where Gunnells has constructed pictures of the cells, the evidence about the number of available alcoves is encouraging. (But there is one nilpotent orbit of type C3 which seems to require an alternate choice of special point. This orbit has an element in standard Levi form, while the component group A(e) has order 2.) 4.3. The highest two-sided cell corresponds to the regular nilpotent orbit. Here the related representation theory is quite transparent, since a regular block has only one simple module (of dimension pN). At the other extreme, one can say quite a bit about the lowest two-sided cell fj, which corresponds to the zero orbit. Here the canonical left cell is just a shifted version of the dominant chamber, whose geometry is transparent. The associated representation theory comes from the group G, with simple modules parametrized in the usual way by highest weights. Using suggestions of Shi, we can argue as follows. Start with a special point for Wp lying in Q such as v = 2(p — l)p; the surrounding \W\ alcoves lie inside the canonical left cell I\ If we write v = x • (—p) (with x a translation from pQ), these alcoves are obtained by applying x to the alcoves around — p labelled by the elements w e W, and thus are labelled by elements xw. Now x~l • p lies inside the antidominant chamber, which is a single left cell of O. Since W acts simply transitively on the Weyl chambers, we see that the alcoves labelled by the various (xw)-1 = w~1x~1 all lie in distinct Weyl chambers and thus in distinct left cells of n. It is easy to see that the resulting bijection between S and C is independent of the choice of v, since the role of W is independent of translations by elements of pQ. 4.4. Jantzen's study of the subregular case makes it possible to say something, even though \ can be chosen to have standard Levi form only for root systems of type An and Bn. In a regular block there is always an isolated simple module, denoted L0 in [14, D.6] and associated with the longest element wq of W. This module is characterized in terms of its "/s-invariant" and has a projective cover of smallest possible dimension.
70 J.E. HUMPHREYS The dominant alcove A obtained by reflecting the lowest alcove across its upper wall H contains the distinguished involution in the canonical left cell T; it is the lowest alcove in T among those sharing the vertex obtained by reflecting — p in the hyperplane H. In our framework it is natural to assign the simple module Lq to A and thus to the left cell T. (This is motivated in part by the approach to computing dimensions in [12], where H plays a key role.) Low rank evidence indicates that the translate of A attached to the special point — p is in the same W/-orbit as the alcove labelled by wq in types An and Bn. Here / is the set of simple roots involved in Jantzen's choice of subregular nilpotent element. For type G2, there are five simple modules in a regular block, three of equal dimension being permuted by A(e) = S3. Here the two-sided cell is finite, with three left cells: the canonical left cell (to which Lq should be assigned) has 8 elements, while the others have respectively 8 and 7. Comparison with Lusztig's model, as developed by Xi [35, 11.2], shows that the triple of simple modules should be assigned to the cell with 7 elements: here the isotropy group in S'3 has order 2. However, it is unclear for root systems other than An,Bn, G2 how to assign the simple modules other than Lq to left cells. 4.5. For a fixed nilpotent orbit, our broader hope is to model Lusztig's conjectural set-up in full detail. Besides taking for the finite set Y the set S above, one needs to bring in the action of Cc(e). Still missing is a construction (presumably based on Be) of suitable modules which carry compatible actions of q and F. But there is a reasonable prototype in the case \ = 0. Here one starts with Weyl modules V(X) with A G X+. Their duals are realized as spaces of global sections of line bundles on B (the Springer fiber in this case). With these modules one has a Kazhdan-Lusztig theory, conjectured by Lusztig (for p not too small) to determine simple modules L(A) via an alternating sum formalism with coefficients depending on Kazhdan-Lusztig polynomials for Wp. In turn L(A) factors (by Steinberg's theorem) into a tensor product of a simple Ux(g)-module and the Frobenius twist of a simple module for G (which looks like the characteristic 0 version if A is suitably bounded relative to p). One would like to find a similar construction for all \. A geometric construction of g-modules using the Springer fiber has been proposed by Mirkovic-Rumynin [29], but without the additional features indicated above. 4.6. When \ is fixed, motivation for correlating simple modules with left cells comes indirectly from the experimental calculations reported in [12]. These are reinforced by Jantzen's unpublished calculations in higher rank cases. The idea here is that the geometry of lower boundaries of canonical left cells, together with the placement of weights in alcoves, should play a key role in predicting the dimensions (and formal characters) of simple modules. The experimental evidence also reinforces the suggestion above about the existence of a tensor product decomposition of Steinberg type. 4.7. Lusztig's conjectural framework works with a fixed nilpotent orbit or two- sided cell. But there is additional motivation for assigning simple modules to left cells when we compare one orbit with an orbit in its closure. When \j) is in the closure of the G-orbit of x-> one expects that a simple Ux(g)-module will "deform" into a not necessarily simple [/^(gji-module.
REDUCED ENVELOPING ALGEBRAS AND CELLS IN THE AFFINE WEYL GROUP 71 On the level of Grothendieck groups, this would imply a recipe for writing the dimension of the given simple Ux(g)-module as a sum of dimensions of simple L^(g)-modules. In all known cases these dimension formulas are given uniformly by polynomials in p and the weight coordinates (compare [6, §6]). Experimentation in low ranks by Jantzen and the author suggests that such decompositions may be possible in a unique way. Ostrik proposes that deformation should be studied in the context of projective covers of simple modules. He suggests an interpretation of the process in terms of comparison of Lusztig's equivariant if-theory bases for the two Springer fibers: these bases may be comparable even when the Springer fibers themselves are not. Using this viewpoint he recovers for example our dimension comparisons in the case of type G2. In low ranks, the cell pictures related to our hypothetical assignment of simple modules to left cells show an intriguing correlation with the computed degeneration in dimension formulas. But all of this remains to be placed in a rigorous theoretical setting, beginning with the process of deformation. References 1. P.N. Achar, On the equivariant K-theory of the nil-potent cone in the general linear group, Represent. Theory 8 (2004), 180-211. 2. P.N. Achar and E.N. Sommers, Local systems of nilpotent orbits and weighted Dynkin diagrams, Represent. Theory 6 (2002), 190-201. 3. R. Bezrukavnikov, On tensor categories attached to cells in affine Weyl groups, Representation Theory of Algebraic Groups and Quantum Groups, 69-90, Adv. Stud. Pure Math., 40, Math. Soc. Japan, Tokyo, 2004. 4. , Quasi-exceptional sets and equivariant coherent sheaves on the nilpotent cone, Represent. Theory 7 (2003), 1-18. 5. , Perverse sheaves on affine flags and nilpotent cone of the Langlands dual group, arXiv:math.RT/0201256. 6. R. Bezrukavnikov, I. Mirkovic, D. Rumynin, Localization of modules for a semisimple Lie algebra in prime characteristic, arXiv:math.RT/0205144, to appear in Ann. of Math. 7. R. Bezrukavnikov and V. Ostrik, On tensor categories attached to cells in affine Weyl groups II, Representation Theory of Algebraic Groups and Quantum Groups, 101-119, Adv. Stud. Pure Math., 40, Math. Soc. Japan, Tokyo, 2004. 8. K.A. Brown and I. Gordon, The ramification of centres: Lie algebras in positive characteristic and quantised enveloping algebras, Math. Z. 238 (2001), 733-779. 9. T. Chmutova and V. Ostrik, Calculating canonical distinguished involutions in the affine Weyl groups, Experiment. Math. 11 (2002), 99-117. 10. I. Gordon, Representations of semisimple Lie algebras in positive characteristic and quantum groups at roots of unity, pp. 149-167, Quantum Groups and Lie Theory, ed. A. Press- ley, Proc. Durham 1999, London Math. Soc. Lecture Note Ser., 290, Cambridge Univ. Press, Cambridge, 2001. 11. J.E. Humphreys, Modular representations of simple Lie algebras, Bull. Amer. Math. Soc. (N.S.) 35 (1998), 105-122. 12. , Analogues of WeyVs formula for reduced enveloping algebras, Experiment. Math. 11 (2002), 567-573. 13. J.C. Jantzen, Representations of Lie algebras in prime characteristic, Notes by Iain Gordon, pp. 185-235, NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., 514, Representation theories and algebraic geometry (Montreal, 1997), Kluwer Acad. Publ., Dordrecht, 1998. 14. , Subregular nilpotent representations of Lie algebras in prime characteristic, Represent. Theory 3 (1999), 153-222. 15. , Modular representations of reductive Lie algebras, J. Pure Appl. Algebra 152 (2000), 133-185.
72 J.E. HUMPHREYS 16. , Representations of Lie algebras in positive characteristic, Representation Theory of Algebraic Groups and Quantum Groups, 175-218, Adv. Stud. Pure Math., 40, Math. Soc. Japan, Tokyo, 2004. 17. , Nilpotent orbits in representation theory, pp. 1-211, Lie Theory, ed. J.-P. Anker and B. Orsted, Progr. Math., vol. 228, Birkhauser, Boston, 2004. 18. G. Lusztig, Some examples of square integrable representations of semisimple p-adic groups, Trans. Amer. Math. Soc. 277 (1983), 623-653. 19. , Cells in affine Weyl groups, Algebraic Groups and Related Topics (Ky- oto/Nagoya, 1983), 255-287, Adv. Stud. Pure Math., 6, Math. Soc. Japan, Tokyo, 1985. 20. , Cells in affine Weyl groups II, J. Algebra 109 (1987), 536-548. 21. , Cells in affine Weyl groups HI, J. Fac. Sci. Univ. Tokyo Sect IA Math. 34 (1987), 223-243. 22. , Cells in affine Weyl groups IV, J. Fac. Sci. Univ. Tokyo Sect IA Math. 36 (1989), 297-328. 23. , Cells in affine Weyl groups and tensor categories, Adv. Math. 129 (1997), 85-98. 24. , Periodic W-graphs, Represent. Theory 1 (1997), 207-279. 25. , Bases in equivariant K-theory, Represent. Theory 2 (1998), 298-369. 26. , Subregular nilpotent elements and bases in K-theory, Canad. J. Math. 51 (1999), 1194-1225. 27. — , Bases in equivariant K-theory, II, Represent. Theory 3 (1999), 281-353. 28. G. Lusztig and N. Xi, Canonical left cells in affine Weyl groups, Adv. Math. 72 (1988), 284-288. 29. I. Mirkovic and D. Rumynin, Geometric representation theory of restricted Lie algebras, Transform. Groups 6 (2001), 175-191. 30. V. Ostrik, On the equivariant K-theory of the nilpotent cone, Represent. Theory 4 (2000), 296-305. 31. Jian Yi Shi, The Kazhdan-Lusztig cells in certain affine Weyl groups, Lect. Notes in Math. 1179, Springer-Verlag, Berlin, 1986. 32. ^ , A two-sided cell in an affine Weyl group, II, J. London Math. Soc. (2) 37 (1988), 253-264. 33. E. Sommers, Lusztig's canonical quotient and generalized duality, J. Algebra 243 (2001), 790-812. 34. N. Xi, The based ring of the lowest two-sided cell of an affine Weyl group. II, Ann. Sci. Ecole Norm. Sup. (4) 27 (1994), 47-61. 35. , Representations of affine Hecke algebras, Lect. Notes in Math. 1587, Springer- Verlag, Berlin, 1994. 36. , The based ring of two-sided cells of affine Weyl groups of type An-\, Mem. Amer. Math. Soc. 157 (2002), no. 749. Dept. of Mathematics k Statistics, U. Massachusetts, Amherst, MA 01003 E-mail address: jehOmath.umass.edu
Contemporary Mathematics Volume 413, 2006 Nakajima's monomials and Crystal bases Seok-Jin Kang, Jeong-Ah Kim, and Dong-Uy Shin Abstract. In this paper, we give a survey of the recent results for the realize tion of crystal bases using Nakajima's monomials. In particular, we introduce a new realization of crystal bases for quantum classical algebras which gives a natural bijection between the set of Nakajima's monomials and the set of Kashiwara and Nakashima tableaux. Introduction The crystal basis theory introduced by Kashiwara yields a nice combinatorial tool to understand the structure of integrable modules over quantum groups [8]. Hence one of the most important problems in crystal basis theory is to realize the crystal bases explicitly. For this, there are several well-known descriptions, e.g., Young tableaux realization for classical Lie algebras [11, 14], path realization using perfect crystals for quantum affine algebras [2, 3, 4], Young wall realization for quantum affine algebras [1], Littlemann's path realization for symmetrizable Kac-Moody algebras [15, 16], and polyhedral realization [17, 18]. Let q be a finite simple Lie algebra of type ADE, and let Uq(Lo) be its quantum loop algebra. The Grothendieck group Rep t/9(Lg) of the category of finite dimensional representations of Uq(Lg) has two bases, the set of simple modules L(P) and the set of standard modules M(P), introduced by Nakajima in [19], where P is the Drinfel'd polynomial. In [19, 21], Nakajima defined a polynomial ZpQ(t), and showed that the multiplicity of simple module L(P) in the standard module M(Q) is given by ZpQ(t) at t = 1. These polynomials ZpQ(t) are the Poincare 1991 Mathematics Subject Classification. 17B37, 81R50. Key words and phrases. Monomial, crystal bases. The first author was supported in part by KRF Grant # 2005-070-C00004 and Seoul National University Grant 2004. The second and third authors were supported in part by KOSEF Grant # R01-2003-000- 10012-0. ©2006 American Mathematical Society 73
74 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN polynomials of intersection cohomology of certain graded quiver varieties and are determined by a bar involution on the ^-analogue of Rep Uq(Lq). In [21], Nakajima introduced the t-analogues of g-characters Xq,t-> and he computed this bar involution using Xq,t- Moreover, he discovered that the set of monomials appearing in ^-analogues of g-character of standard module has a crystal structure [20]. In [10, 22], motivated by this work, Kashiwara and Nakajima independently denned a crystal structure on the set Ai of all Nakajima monomials, where the action of Kashiwara operators is interpreted as multiplication by certain monomials. Moreover, they showed that the connected component M.(M) containing a maximal vector M of a dominant integral weight A is isomorphic to the irreducible highest weight crystal B(X). The explicit description of this connected component M(M) was given for Uq(o) (q = An,Bn,Cn,Dn,G2 and An') by Kang-Kim-Shin, Shin, and Kim [5, 6, 12, 23]. They also gave a crystal isomorphism between the monomial realization and the tableaux realization or the path realization. Recently, in [7], Kang, Kim and Shin gave a realization of the crystal J5(oo) for symmmetrizable Kac-Moody algebras. In their work, they introduced the notion of modified Nakajima monomials by adding a new variable 1 and define a crystal structure on the set of all modified Nakajima monomials. Moreover, they showed that the connected component M(M) containing a maximal vector M of an integral weight A is isomorphic to the crystal J5(oo) <g) T\. For the type An' and An, they gave an explicit description of the connected component Ai(M) and constructed a natural isomorphism between the path realization and monomial realization. However, it is still an open problem to give explicit characterizations of connected components of the Nakajima monomials and modified Nakajima monomials for general symmetrizable Kac-Moody algebras. In this paper, we give a survey of the recent results in [5, 6, 7, 12] and we obtain characterization of another connected component M(M) for Uq(q) (q = An,Cn,Bn,Dn). Moreover, we construct a natural crystal isomorphism between the monomial realization and the tableau realization of B(X) given by Kashiwara and Nakashima. 1. Quantum groups and Crystal bases In this section, we recall the basic notion of the quantum groups and crystal bases. Let / be a finite index set and let A = (a^)^/ be a generalized Cartan matrix. A Cartan datum (of A) consists of • generalized Cartan matrix A = (a^-)^/, • dual weight lattice Pv = (0i€/ Zh{) e (®fl^nkA ZdX • weight lattice P = {A e J)* | A(PV) C Z}, where J) = Q <g>z Pv, • the set of simple coroots IIv = {hi \ i G /},
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 75 • the set of simple roots U= {cti\i e I}. Here, the simple roots are linearly independent and satisfy otj{hi) = a,ij for all i,j el. We denote by P+ = {A G P | X(hi) > 0 for all i G 1} the set of dominant integral weights. For instance, the fundamental weights A* (i G /) denned by Ai(hj) = 5ij, Ai(dj) = 0 are dominant integral weights. In this paper, we assume that A is symmetrizable, i.e., there is a diagonal matrix D = diag(si G Z>o | i G /) such that DA is symmetric and Si are relatively prime. Let qi = qSi and define [k]i = Qki~Q^ Qi-Qi -l ' [n]i\ nw«> k=i [n]i\ [m-n]iV Definition 1.1. The quantum group Uq(o) associated with a Cartan datum (A, PV,P, IIV,II) is the associative algebra with 1 over Q(q) generated by e^, /; (i G /) and qh (h G Pv) with the following denning relations: <Z° = 1, qhqh' =qh+h' (h,hf G Pv), ^e^"^ = q"*™*, qhfiq~h = q~aiW fi (h ePw,ie /), 1-CLij fc=0 l-ai:7- fc=0 1 — a>i Qi-Qi l — dij—k k n 1 — a< '*7 DJ^2 /l —a^j — k r rfc i JjJi 0 (W), (^i). The category Oint consists of C/g(g)-modules M satisfying the following properties: • M = 0A€P MA, where Mx = {v G M \ qhv = qx^v for all h G Pv} is finite dimensional, • there exist finitely many elements Ai,...,As G P such that wt(M) C U*=i(Ai -Q+)i where wt(M) = {A G P|MA ^ 0} and Q+ = ZieI Z>o^, • ei and /^ (i G /) are locally nilpotent on M. It is known that the category Oint is semisimple and every simple object in Oint is isomorphic to the irreducible highest weight module V(X) with a dominant integral highest weight A G P+. Fix an index i G / and set ejn) = e?/[n]i!, /^n) = /f/[n]*!. Let M be a C/g(g)-module in Oint. Then every weight vector v G MA can be written uniquely as «= £/,<%, k>0
76 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN with Vk G ker e^ D M\+fcai. The Kashiwara operators ii and fi on M are denned by fc>l fc>0 Let A0 = {f/g G Q(q) \f,ge Q[q], g(0) ^ 0} be the localization of Q(q) at (9). Definition 1.2. A crystal basis of M is a pair (L, J5) such that (i) L is a free Ao-submodule of M such that M = Q(q) <S>a0 L, (ii) B is a Q-basis of L/qL = Q (g)Ao L, (iii) L = 0A€PLA, where Lx = LnMx, (iv) J5 = \JxeP B*> where J5A = J5 n (Lx/qLx), (v) e;L C L, /iL C L for all 2 G /, (vi) eiB CBU {0}, frB cBU {0} for all i G /, (vii) for all b,bf G B and i G /, fib = bf if and only if 6 = e$. Prom the above conditions, we get a colored oriented graph with the arrows denned by b —^-> b' if and only if fib = b'. We call B the crystal graph of M. Let V(A) be the integrable highest weight module with the highest weight A G P+ and the highest weight vector v\. In [8], Kashiwara proved that there is a unique crystal basis (L(A), B(X)) of V(A), where L(A) is the free Ao-submodule of V(X) spanned by the vectors of the form fh --firv\ (ik el,re Z>0) and B(X) = {fh ■ ■ ■ firvx + qL(X) e L(X)/qL(X)} \ {0}. Fix i e I. For any P G U~(q), there are uniquely determined Q,R G U~(q) such that e*P - Pe. fc ~ % * We define the endomorphisms e^, e" : f7g (g) —> Uq (q) by eUP)=il, eJ'(P) = Q. Then every u G f/~ (q) can be written uniquely as u = 22 fi uk, where e^Uk = 0 for all k > 0, k>0 and we define the Kashiwara operators ii, fi on U~(g) by e> = ^ fik~1}uk, fiU = ]T /i*+1 W k>l k>0
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 77 Definition 1.3. A crystal basis of U~(g) is a pair (L, J5), where (i) L is a free Ao-submodule of U~(q) such that U~(g) = Q(q) <8>a0 L, (ii) B is a Q-basis of L/qL = Q (g)Ao £, (iii) iiL C L, /;L C L for all 2 G /, (iv) iiB CBU {0}, fiB CBU {0} for all 2 G /, (v) for all b,bf e B and 2 G /, fib = bf if and only if b = e$. In [8], Kashiwara also showed that there is a unique crystal basis (L(oo), J5(oo)) of U~(q), where L(oo) is the free Ao-submodule of U~(g) spanned by the vectors of the form h • • • fir -1 (*fc e J, r e Z>0) and £(oo) = {4 • • • fir • 1 + gL(oo) G L(oo)/<?L(oo)} \ {0}. 2. Abstract crystals Let (A, Pv, P, IIV, II) be a Cartan datum and f7g(g) be the associated quantum group. Definition 2.1. An abstract crystal for Uq(g) or a Uq(o)-crystal is a set J5 together with the maps wt : B —> P, £*, ^ : B —> Z U {—oo}, e*, /i : J5 —> J5 U {0} (i G /) such that for alH G / and b G J5, (i)^(6) = ei(6) + (/ii,wt(6)), (ii) wt(e;6) = wt(6) + a*, wt(/;6) = wt(b) - a;, (iii) £;(e;6) = €i(b) - 1, £,(£6) = e*(6) + 1, (iv) <£>;(e;6) = ^(6) + 1, <pi(fib) = <pi(b) - 1, (v) /^ = 6; if and only if e;6' = b for 6,6' G J5, (vi) e;6 = fib = 0 if £;(6) = -oo. Definition 2.2. (a) A morphism \j) : B\ —> J52 of crystals is a map ^ : J5i U {0} —> J?2 U {0} satisfying the following conditions: (i) tf(0) = 0, (ii) wt(V>(&)) = wt(6), £i{i>{b)) = ei(b) and <Pi(tp(b)) = tpi(b) if b e JBi and iW) e JB2) (iii) /^(6) = V(/«fc) if b, fib e Bx, V(&), W<&) e B2. (b) A morphism is said to be strict if it commutes with the Kashiwara operators e*,/i (i e I). (c) An injective crystal morphism from J5i U {0} to B2 U {0} is called an embedding. Definition 2.3. The tensor product J5i <g> B2 of £/g(g)-crystals J5i and J52 is again a C/g(g)-crystal such that
78 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN (i) wt(6i <g> 62) = wt(6i) + wt(62), (ii) ^(61 <g>62) = max(ei(bi),ei(b2) - (fti,wt(6i))), (iii) (fi(bi <g> b2) = max(<^(62), ¥>i(&i) ~ {K, wt(62))), /. x - /. ^ . x J g<fti ® 62 if <^(&i) > Ci(62), (iv) ei(6i<g>62) = < [61 0 iib2 if ^i(6i) < £i(62), /a //1. ^^ //<6i®62 if^(6i)>ei(62), (v) /i(6i <8)62) = < (M/A if ^i(fti) <ei(62). Example 2.4. (a) For AeP+, the crystal graph B(X) of the irreducible highest weight module V(A) is a C/g(g)-crystal, called the irreducible highest weight crystal with 6i(b) = max{k > 0 | e\b ^ 0} and ^(6) = max{fc > 0 | fj°b ^ 0} for b G J3(A). (b) The crystal graph J5(oo) of U~(q) is a C/g(g)-crystal with 6i(b) = max{fc > 0 | e\b ^ 0} and <^(6) = €i(b) + (hu wt(6)) for 6 G J5(oo). (c) The singleton TA = {£A} (A G P) is a C/g(g)-crystal with wt(£A) = A, £;(£A) = <fi(t\) = -00, e^A = /itA = 0 (ie /). (d) For each i G /, we define the elementary crystal Bi = {bi(n) \ n G Z} by wt (6i(n)) = no:*, £i(6i(n)) = -n, (fi(bi(n)) = n, e»(6i(n)) = bi(n + 1), fi(bi(n)) = bi(n ~ 1)> Sj(bi(n)) = <Pj(bi(n)) = -00, ej(bi(n)) = fj(bi(n)) =0 (j^ i). Proposition 2.5. [9] (a) For AeP+, there is a crystal embedding B(X) <-^ J3(oo) <g) TA sending u\ to u^ <g) tA. (b) For eac/i i e I, there exists a unique strict embedding of crystals $i : J5(oo) <-^ J5(oo) <g) ^ sending u^ to u^ <g) &i(0). Remark 2.6. Proposition 2.5 (b) yields a strict embedding of crystals J5(oo) ^-> •••(g) Bik (g) • • • <g) J3;2 (g) J5^, where (&fc)j&i is a sequence in / such that every i e I appears infinitely many times. For a subset J of /, we denote by Uq(gj) be the quantum group associated with the generalized Cartan matrix Aj = (a^)^/. Similarly, we denote by J3j(oo) the crystal graph of U~(qj). For a Uq(g)-crystal J3, we define ^j(B) to be the Uq(gj)- crystal obtained from B by forgetting the maps e%, (fi, e^, fi for i £ J. Thus the crystal graph ^fj(B) is obtained by removing all the 2-arrows for i £ J, The following recognition theorems play a crucial role in proving the main results. Theorem 2.7. [13, Theorem 2.1, Theorem 2.2] (a) Suppose that a Uq(g)- crystal B satisfies the following condition: for any subset J of I with \J\ < 2,
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 79 every connected component of^j(B) containing a maximal vector b is isomorphic to the crystal associated with an integrable highest weight Uq(gj)-module. Then every connected component of B containing a maximal vector b is isomorphic to B(wt(b)). (b) Suppose that a Uq(o)-crystal B satisfies the following condition: for any subset J of I with \J\ < 2, every connected component of ^fj(B) containing a maximal vector b is isomorphic to Bj(oo)<S>Twt^. Then every connected component of B containing a maximal vector b is isomorphic to J5(oo) (g) Twt(6)- 3. Perfect crystal and Paths Let (A, Pv, P, IIV, II) be an affine Cartan datum and Uq(g) be the associated quantum group. The subalgebra Uq(g) of Uq(g) generated by ei, /i, Kf1 (i G I) is also called the quantum affine algebra. Let P = Zho 0 Zh\ 0 • • • 0 Z/in, J) = C ®z P and P = ZA0 0 ZAi 0 • • • 0 ZAn. The elements of P are called the classical weights. Let P = {A G P\ X(hi) > 0 for alH G /} be the set of classical dominant integral weights, and Pt = {A G P \{c, A) = /}, where c is the canonical central element. The algebra U'q(g) can be regarded as the quantum affine algebra associated with the classical Cartan datum (A, II, IIV, P,P ). We now define the notion of perfect crystals. Let B be a classical crystal. For b G J3, we define <*>) = E£i^Ai> v(b) = J2 ^At and wtw = vw - £(6)- i i Definition 3.1. For I G Z>0, a finite classical crystal B is called a perfect crystal of level I for Uq(o) if (i) there is a finite dimensional Ufq(g)-module whose crystal graph is isomorphic to B, (ii) B <g) B is connected, (iii) there exists some Aq G P such that wt(B) c A0 + ^ J2 Z<0ai, | BAo |= 1, where do is the coefficient of ao in the null root 5, (iv) for any b G B, we have (c,e(b)) > /, (v) for each A G Pt , there exist unique bx, b\ G B such that e(bx) = A, (p(b\) = A. Example 3.2. Let Uq(g) be the quantum affine algebra of type An and let B = {b = (xo, xi,..., xn) | a?i G Z>0, Si ^t = '}• For b = (xo, a?i, • • •, xn) G B, define (fi(b) = Xi, 6i(b) = Xi+i and wt(6) = ^(^»(6) - e»(6))Ai,
80 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN and (xo, • • •, %i - 1, #2+i + 1, £i+2,..., xn) for i ^ n, (x0 + 1, xi,..., xn - 1) for i = n, (xq, ...,Xi + 1, a?i+i - 1, £i+2, • • •, xn) for z ^ n, (x0 - 1, xi,..., xn + 1) for i = n. Then B is a perfect crystal of level I for Uq(An '). For a dominant integral weight A = aoAo H h anAn of level I, we have bX = K, clq,..., an_i), 6a = («o, ai, • • •, «n)- Fix a positive integer I > 0 and let B be a perfect crystal of level /. By definition, for any classical dominant integral weight A € Pt , there exists a unique element b\ € B such that p(b\) = A. Set /i = A — wt(&A) = £(&a), and denote by Up the highest weight vector of the crystal graph J5(/i). Then, using the fact that B is perfect, we have Theorem 3.3. [3] Let B be a perfect crystal of level I > 0. Then for any dominant integral weight A € Pt , there exists a crystal isomorphism # : B(X) -Z+ B(e(bx)) ® B given by ux >—+ ue{bx) ® 6A, wftere b\ is the unique vector in B such that (p(b\) = A. Set A0 = A, Afc+i = e(&Afc)j and b0 = bXl bk+i = b\k+1. By taking the composition of crystal isomorphism \I> in Theorem 3.3 repeatedly, we get a crystal isomorphism Vk(k>l):B(\)-^B(\k)®B®k given by u\ i—y u\k <g) 6fc_i (g) • • • (g) 6i (g) 6o- Since B is perfect, there is a positive integer N > 0 such that Aj+w = A^, bj+N = bj for all j = 0,1,... ,fc. Definition 3.4. (a) The sequence pA = (MfcLo = * * -®&JH-i®&fc®- * -®&i®&o is called the ground-state path of weight A. (b) A X-path in B is a sequence p = (pfc)^=o = * * * ® P/c ® * * * ® Pi ® Po such that p/c = 6^ for all k > 0. Let P(A) = P(A,B) be the set of all A-paths in B. We define a £^(g>crystal structure on V(X) as follows: Let p = (Pk)^=o be a A-path in V{\) and let N > 0 /i&= iib =
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 81 be a positive integer such that pk = bk for all k > N. For each i e /, we define N-l wt(p) = e*p = fiP = e»(p) = Vt(p) = : XN + ^ 1 fc=0 = ' ' ' <g> PN+1 : "-^PiV+l : max(ei(p/) *t(Pfc)> ®h{PN ® - <8>fi(PN®" -(Pi(bN),0), = y>i(p/)+max(y>i(6jV) -< •®Po), •®Po), £;(p'),0), where wt denote the classical weight and p' = Pn-i ® • • • ® Pi ® Po- Then we have the path realization of the classical crystal B(X). Theorem 3.5. [3] There exists an isomorphism of classical crystals V : J5(A) -^ V(X) given by ux '—► pA. Let {Bi}i>i be a family of perfect crystals B\ of level I and jBjmin = {b e Bi\{c,e{b) = I}}. We take the index set J = {(l,b)\l € N,6 € J3™n}. Definition 3.6. (a) A classical crystal B^ with an element b^ is called a /zm^ of {B/}/>iif (i) wt^) = 0, £(&oo) = ^(6oo) = 0, (ii) for any (I, b) € J, there is an embedding of crystals /(i,6) : rc(6) (g) Si (g) T_^(6) -+ Boc (iii) B00 = |J(/,6)€JIm/(W- (b) If a limit exists for the family {i?i}i>i, {J3j}j>i is called a coherent family of perfect crystals. Remark 3.7. A limit (J3oo,&oo) of a coherent family {Bi} is unique up to isomorphism. By the properties of the perfect crystal Bi, the following is immediate. Proposition 3.8. [2] (a) B^ ® B^ is connected. (b) For any b e B^, (c,e(b)) > 0. (c) The maps e and <p give bisections from B™in = {b G J3oo|(c, e(b) = 0)} to
82 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN Example 3.9. Let Uq(g) be the quantum affine algebra of type An\ B^ = {b = (60,61,..., bn) e Z-+1 I Zto bi = °}> and 600 = (0,..., 0). Define wt(6) = (b0 - 61)A0 + (61 - 62)Ai + • • • + (bn - b0)K, 6i(b) = bi+i (i^n), en(b) = b0, <p»(6) = 6*, lb = < ^°' ■ ■ •'ft* " 1' ft*+1 + 1' *'''bn^ i ^ n' I (60 + l,6i,...,6n-1), i = n, ~, J (b0,...,bi + l,b;+i - l,...,6n), ^ rc, e^o = < [ (b0- l,6i,...,6n + l), z = n, then J5oo is a limit of {BJ/>i for Uq(A^). Theorem 3.10. [2] There is a crystal isomorphism J5(oo) ^ J5(oo) <g> Boo sending u^ to u^ (g) 600. By applying Theorem 3.10 repeatedly, we obtain a crystal isomorphism fa (k > 1) : J5(oo) -^> J5(oo) ® B®* given by (8)600 <8) ••• 0 6oo- Definition 3.11. (a) The sequence poo = (600)^=0 = * * * ® &<*> ® • • • 0 600 0 600 is called the ground-state path. (b) A pa#i in JBqo is a sequence p = (pk)^=o = • • • ®Pfc 0 • • • ®Pi <8>Po such that Pfc = 6oo for all k > 0. We denote by P(oo) be the set of all paths in B^. Then we have Theorem 3.12. [2] There is a crystal isomorphism from B(oo) to V(cx>) given by Uoo 1—► poo. 4. Nakajima monomials and crystals In this section, we recall the crystal structures on the sets of monomials discovered by Nakajima [19, 20] and modified monomials given by Kang, Kim and Shin [7]- Let M be the set of monomials in the variables Yi(n) for i e I and n e Z. Choose a set C = (cij)i^j of integers such that Cij + Cji = 1, and define Mn) = Yi(n)Yi(n + 1) J] Yj(n + Cji)ai^\ Then we define a C/g(g)-crystal structure on Ai as follows:
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 83 For a monomial M = YlieI neZ Yi{n)yi^n\ wt(M) = £(5>(«))Ai, iei nez (fi(M) = max { ^T yi(k) |neZ}, k<n (4.1) 6i(M) = max { - 53 Vi(k) | n € Z} /«(M) = £<(M) = /c>n 0 i£ <pi(M) = 0, Aiinf^M ify>«(M)>0, 0 if£i(Af) = 0, Ai(ne)M itei(M)>0. Here, nf = min{n | <fi(M) = ^ Vi(k)} = min{n | 6i(M) = - ]P S/»(fe)}, /c<n /c>n ne = max{n | ^(M) = ^ Jfc(fc)} = max{n | 6i(M) = - ^ jfc(fc)}. /c<n /c>n Theorem 4.1. [10, 20] Let M be a monomial of weight A such that eiM = 0 for all iei, and let Ai(M) be the connected component of Ai containing M. Then there exists an isomorphism of crystal M(M)-^->B(\) given by M \—► v\. Now, in order to give a new realization of the crystal J5(oo), we introduce a new variable 1 which commutes with all Fi(n)'s, and choose a set C = (cij)i^j of nonnegative integers such that Cij + Cji = 1. Let M. be the set of all monomials of the form (4.2) M = JJJJyi(n)^n>l iein>0 such that yi(n) G Z and yi(n) = 0 for all but finitely many n's. We will call the monomials in M the modified Nakajima monomials.
84 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN For a modified Nakajima monomial M of the form (4.2), we define wt(M) = ^£yi(n))Ai, iel n>0 <Pi(M) = max{ ^T yi(k) \ n > 0}, 0<k<n (4.3) Si(M) = ip^M) - (hi, wt(M)>, - ^ fo if £i(M) = 0, [^(riejM if ^(M)>0, fiM = Aifaf) M, where nf = nf(M) = min{n > 0 | ^(M) = Eo<k<nVi(k)}^ ne = ne(M) = max{n > 0 | ^(M) = £o<fc<n &(*)}• Then it is straightforward to verify that the set M. of all modified Nakajima monomials forms a C/g(g)-crystal with the maps wt, 6i,<pi, ei, fi (iel). Moreover, we have Theorem 4.2. [8, Theorem 3.1] For any maximal vector M e M, the connected component Ai(M) of Ai containing M is isomorphic to the Uq(g)-crystal J3(oo) ®Twt(M). In particular, we have A4(l) —* J5(oo). 5. Characterization: An type Let I = {1,..., n} and let A = (aij)ijei be the generalized Cartan matrix of type An. We take the set C = (cij)i^j to be Cij = 0 if i > j, 1 if i < j and set yb(^)±1 = ^n+i(^)±1 = 1 for all m e Z. Then for i € / and m e Z, we have i4i(m) = ^(ro^ro + l)y<_i(m + l)"1^^™)-1. For a monomial M = Ylt ^at(mt)_1^6t(nt) with a>t+™<t = bt+nt, define M(fc)+ (k = -n,..., -1) and M(fc)~ (k = -n + 1,..., 0) by M(k)+= JJ yflt(mt)-1nt(nt) = nirat(mt)-1nt(fc)> t:nt=k t M(k)~= H Yat(mt)-1Ybt(nt) = l[Yat(ky1Ybt(nt)- t:mt=k t For M(fc)+ = ]lt^K)"1^(*) and M(fc)~ = lit 1^(*)-%("*), we set the sequence A+(M(fc)) = (6^,6^,... ,6ir), and A~(M(fc)) = (ah,aj2,... ,aja), where &ii < &i2 < * * * < bir < n + 1 and a^ < aj2 < • • • < aJs < n + 1. Theorem 5.1. Le£ A = aiAi + • • • + anAn be a dominant integral weight and let M0 = Fi(-l)air2(-2)a2 • • • Yn(-n)a«
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 85 be a maximal vector of weight A in M.. The connected component A4a{Mq) in Ai containing Mo is characterized as the set of monomials of the form X]Yat{mt)-lYbt{nt) t with at + mt = bt + nt satisfying the following conditions: for\+(M(k)) = (bil,bi2,...,bir) and\-(M(k)) = (ah , aj2,..., ajs) with k — -n + l,...,-l, (i) r-s = ak, (ii) bip < ajp for all p = 1,..., s. Proof. Our claim can be proved by the same argument in the proof of Theorem 2.8 in [5]. □ Example 5.2. Let M be a monomial M = Y.ior^i-lfYsi-iy^i^Ysi^)-1. It can be expressed as M = (Fo(o)-1y2(-2))(Fi(o)-1y2(-i))2(y3(-i)-1i4(-2))(i3(-2)-1y4(-3)). Then we have M(-3)+ = y3(-2)-1F4(-3), m(-2)+ = (y0(o)-1i2(-2))(F3(-i)-1n(-2)), M(-i)+ = y!(o)-2y2(-i)2, and m(-2)- = y3(-2)-1yr4(-3), M(-l)- = ^(-l)-1^^), M(0)" = ^(O)-1^^)^^)-2^^!)2. Moreover, the sequences A+(M(-3)) = (4), A+(M(-2)) = (2,4), A+(M(-1)) = (2,2), A"(M(-2)) = (3), A"(M(-1)) = (3), and A"(M(0)) = (0,1,1). Therefore, M belongs to M(Ai + A2 + A3). For i € I and m € Z, we introduce new variables Xi(m) = Yi-i(m + l)_1yi(m). Then Ai(m) = Xi(m)Xi+\(m)~l and we have: Corollary 5.3. Let A = aiAi + • • • + a„A„. Then the connected component M.{\) containing the maximal vector M0 = Y1(-l)air2(-2)a2 •••Fn(-n)a" = X1(-l)^(X1(-l)X2(-2)r ■ • • (*!(-!) • • • Xn(-n))a»
86 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN is characterized as the set of monomials M = XtlA (-1) • • • XKai (-1) • • • XtnA {-n) ■ ■ ■ Xtn<an (-n) satisfying the following conditions : (i) ak = ak H (- an for k = 1,..., n, (ii) tkli ^ tfc,2 ^ • • • ^ tk^k fork = l,...,n, (iii) tj^ -< tj+^/c /or eac/i j = 1,..., n — 1 and k = 1,..., ctj+i. Let A be a dominant integral weight and let T(A) be the set of all semistan- dard tableaux of shape A with entries from {1,2,..., n + 1}. Recall that T(A) has a C/g(>ln)-crystal structure which is isomorphic to the crystal graph of the finite dimensional irreducible module V^A) [11]. Theorem 5.4. Let A = aiAi H hanAn be a dominant integral weight. Then there is a Uq(An)-crystal isomorphism iJ)a '• jM^Mq) —> T^(A). Proof. Let M be a monomial in A4(A), which can be expressed as M = XtlA (-1) • • • Xtliai (-1) • • • XtnA (-„) • • • Xtnian (-n). We define ^(M) to be a tableau with entries t^i,..., tk^k in the A>th row (from top to bottom) for k = 1,..., n. By the characterization of A4(A), it is clear that ^(M) is a semistandard tableau. Conversely, let S be a tableau in T(A), then we define ^X(T) to be the monomial Y\Xi(—p), where Xi(— p) corresponds to the entry i in the p-th row of T. Then it is easy to see that i/ja and i/j^1 are inverses to each other and that i/ja is a crystal isomorphism. □ We close this section with the characterization of the C/g(>ln)-crystal J5(oo) in terms of modified Nakajima monomials. Theorem 5.5. [8, Corollary 4.4] The Uq(An)-crystal J5(oo) can be characterized as the crystal consisting of the modified Nakajima monomials of the form n m = Y[Y[Xi(k)a^i k>0i=0 satisfying the following conditions: (i) ai(k) = 0 fork^>0, (ii) E?=oai(k) = °forallk>°> (iii) a0(k) > 0 and Y7i=i ai(k) < 0 for all p = 1,..., n.
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 87 6. Characterization: Cn type Let / and C = (cij)i^j be the sets given in section 5. Let A = (aij)ijej be the generalized Cart an matrix of type Cn. For i e I and m e Z, we have A(m) = i yn-i(m + l)-2yn(m)yn(m+l) for % = n, \ y»_i(m + l^Yi^Yiim + 1)^+1 (ra)"1 otherwise. Let B = {1,2,..., n, n,..., 1} and define a total ordering on B by 1 ^ 2 ^ <n -<n -< • • • -< T. For ie/ and ra G Z, we introduce new variables Xi(m)=yi_i(m + l)-1yi(m), Xjim) = yi_i(m + (n - t + l))y*(m + (n - 2 + l))'1. Then it is straightforward to verify that for i = 1,..., n — 1 (6.1) Xi(p)XM = Xi+i(p)Xj+T(<z) ifp-9 = n-t. Proposition 6.1. Lei A = aiAi H 1- anAn. 77ien the connected component M(X) containing the maximal vector M0 = Y1(-l)aiF2(-2)a' • • -Yn(-n)a" = X.i-ir (X1(-l)X2(-2)r ■ ■ ■ (*i(-l) • • -Xn(-n)r- is characterized as the set of monomials (6.2) M = Xtlil (-1) • • • Xtl,ai (-1) • • • XtnA (-n) • • • Xtn^ (-n) satisfying the following conditions : (i) otk = ak H h an /or fc = 1,..., n, (ii) *fc,i ^ tk,2 ^ • • • ^ *fc,afe /or fc = 1,..., n, (iii) /or eac/i j = 1,..., n — 1 and k = 1,..., aj+i, £j,fc -< £j+i,/c- Proof. Our claim can be proved by the same argument in the proof of Proposition 2.4 in [6]. □ Let M be a monomial of M(M0) given in Proposition 6.1. Then M can be expressed as (6.2) satisfying the conditions (i)-(iii). Unfortunately, by (6.1), this expression is not unique. However, one can find a canonical expression as is described in the following. Step 1. Given an expression M = Xtlil (-1) • • • Xtliai (-1) • • • XtnA (-n) • • • Xtn>Qn {-n), we associate a tableau T(M) with entries ^,i, • • • ^/c,afe in the fc-th row (from top to bottom) for k = 1,..., n. By the characterization of A4(A), the entries of T(M)
88 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN in each column are strictly increasing. For simplicity, we will say that there is an i(p) if there exists an entry i lying in the p-th row of T(M) from top. Step 2. We define the following equivalence relations on the set of tableaux T(M): (al-1) For each a = 1,..., n — 1, if there is a pair (a(p), a(q)) such that q — p=± n — a, and a(p) and a(q) lie in the same column or a(q) lies in the left hand side of o(p), then replace (a(p), a(q)) with (a + l(p), a + 1(g)). If there are several such pairs, then we carry out this process for the pair (a(p),a(q)) consisting of the rightmost a(p) and the leftmost a(q) and continue as is shown below. We will apply this rule from a = 1 to a = n — 1. (al-2) For each b = 2,..., n, if there is a pair (b(p),b(q)) such that q—p = n—6+1, and b(q) lies in the right hand side of b(p), then we replace (b(p), b(q)) with (b — l(p),b — l(q)). If there are several pairs (b(p),b(q)), then we carry out this process for the pair (b(p), b(q)) consisting of the leftmost b(p) and the rightmost b(q) and continue as is shown below. We will apply this rule from b = n to b — 2. From now on, we will denote by [T(M)] the tableau obtained from T(M) by applying (al-1) and (al-2). The corresponding monomial [M] will be called the canonical expression of M. Example 6.2. Let A = Ai + A2 + A3 for C3 and let Then it can be expressed as M = X2(-l)2X3(-l)X1(-2)XT(-2)Xr(-S) and hence M € M(A\ + A2 + A3). Moreover, we have T(M) = 2 2 T 2 1 3] (al-1) 2 3 1 3 1 3] Therefore, [M] = X2(-1)X3(-1)2X^2)XT(-2)XT(-S).
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 89 From the above algorithm, we have Theorem 6.3. Let A = aiAi + • • • + anAn. Then the connected component Aic(Mo) containing the maximal vector M0 = ri(-l)aiy2(-2)a2 • • -F„(-n)a" = X1(-l)a^(X1(-l)X2(-2))^ • • • (Xi(-l) • • • Xn(-n))a» is characterized as the set of monomials M = XtlA(-l).-.Xtuai(-l)-.-Xtn,1(-n)...Xtn,aJ-n) satisfying the following conditions : (i) otk = ak H (- an for k = 1,..., n, (ii) *fc,i ^ tka -< ''' -< tky0ck fork = l,...,n, (iii) for each j = 1,..., n — 1 and k = 1,..., a^+i, tjtk ■< ^+i,fc; (iv) there is no pair {Xtpk(—p),Xtql{—q)) with k>l,p<q such that tp,k = Q>, tq,i — o and q — p = n — a, (v) there is no pair {Xtpk(—p),Xtql{—q)) with k <l, p < q such that tp,k = Q>, tq,i = fl and q — p = n — a + 1. Let Tc(A) be the set of Cn-tableaux of shape A given by Kashiwara and Nakashima in [11]. Then we have THEOREM 6.4. Let A = aiAi -f • • • + anAn be a dominant integral weight and let Mo be the monomial given in Theorem 6.3. Then there is a crystal isomorphism i>c:Mc(Mo)^Tc(\). Proof. Let M be a monomial in jM(A), which can be expressed as M = XtlA (-1) • • • Xtltai (-1) • • • Xtn<1(-n) ■ ■ ■ Xtn,an (-n). Consider the tableau T([M]) associated with [M]. If there is no pair (a(p),a(q)) such that (6.3) q —p < n — a, and a(p) and a(q) lie in the same column or a(q) lies in the left-hand side of a(p), we define \j)c{M) to be the semistandard tableau T(M) itself. Secondly, suppose that there is a pair (a(p),a(q)) satisfying the condition (6.3) in T([M]). Then we define ij){M) to be the tableau obtained from T([M]) by replacing a with a + 1. If there are several such pairs (a(p), a(q)) such that q—p < n—a (a = 1,..., n— 1), then ^c(M) is defined by applying the above rule repeatedly from 1 to n — 1. Conversely, let S be a tableau in T(A). If there is no pair (a(p), a(q)) satisfying the condition (6.3), we define ^X(T) to be the monomial Y[Xi(—p), where Xi(— p)
90 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN corresponds to the entry i in the p-th row of T. Suppose that there is a pair (a(p),a(q)) satisfying the condition (6.3). Then we have a tableau X" obtained from T by replacing a in the p-th row and a in the q-th. row with a — 1 and a — 1. If there are several such pairs, then T' is denned by applying the above rule repeatedly from a = n to a = 2. Now, we define ^^(T) to be the monomial associated to X". Then we can see that %j)c and i\)^} are inverses to each other and that \j)c is a crystal isomorphism. □ Example 6.5. Let A = Ai + A3 for C4 and let M = y^-l^l)"^^), which can be expressed as [M]=X1(-l)2X2(-2)XT(-3) and T([M}) Since there is a pair (1(-1), l(-3)) in T([M}) such that p - q = -1 - (-3) < 4 — 1 = n — a, they corresponds to (2(—1),2(—3)). Here, 1(—1) is just the one in the second column from left. Moreover, this changed 2(—3) and 2(—2) also satisfies p — q = —2 — (—3) <4 — 2 = n — a, which implies that they corresponds to (3(-2),3(-3)). Therefore, ^c(M) = 7. Characterization: J5n, Dn types Let / and C = (cij)i^j be the sets given in section 5. Let A = (aij)ijei be the generalized Cart an matrix of type Bn. For i e I and m G Z, we have A(m) = i ^-2(m+l)-1Fn_1(m)Fn_1(m+l)Fn(m)-2 fort = n-l, i[m) \ yi_i(m+l)-1yi(m)yi(m + l)yi+i(m)-1 otherwise. Let B = {1,2,..., n, 0, n,..., 1}, and define a total ordering on B by 1^:2^; <n^0^n^ <T. For m e Z, we introduce new variables X(m) = i ''*-1(m+1) Yi(m) for i = l,...,n-l, Xi(m) , „ ,„ , ,„,,_ , 1W2 yi_i(m+(n-i + l))yj(m+(n-i + l))_1 for i = l,...,n-1, Fn_i(m+l)y„(m+l)-2 fori = n, X0(m) = yn(m)Fn(m + l)-1. Then Xi(m) (i e J, m € Z) satisfy the relations in [7, Lemma 2.8].
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 91 Any dominant integral weight A can be expressed as A = aiui H h anLOn + 6An, where Ui = A; if i =fi n, 2An if i = n and a; G Z>0 (2 = 1,..., n), 6 = 0 or 1. We first consider the case An for Bn. Proposition 7.1. Let Mq = Yn(—n) be a maximal vector of weight An. Then the connected component Mb {Mq) of M. containing Mq = Yn(—n) is characterized as the set of monomials of the form satisfying the following conditions: (i) 1 ■< i\ -< %2 -< - - - -< in d 1 and ik ^ 0 for all k = 1,..., n, (ii) there is no pair (ip = a, iq = a) for all a,p,q = 1,..., n with p < q. Proof. Our claim can be proved by the same argument in the proof of Proposition 2.11 in [6]. □ For general case, we know that the expression of M in terms of Xi(m) is not unique. But, by a similar algorithm as in the case of Cn, we have a canonical expression of M. (Compare with [7, Theorem 2.23]). Theorem 7.2. Let\ = aiuj\-\ \-anujn (resp. a\u)\-\ hano;n+An). Then the connected component .Mb(Mo) containing the maximal vector M0 = Fi(-l)ai • • • Yn{-n)2a" = Xi(-l)ai • • • (Xi(-l) • • • Xn(-ri))a« (resp. Yi(-l)ai •••Fn(-n)2a"+1 = X1(-ir^-(X1(-l)^-Xn(-n)r^X1(-l)-'Xn(-n)) is characterized as the set of monomials M = Xtl<1 (-1) • • • Xtliai (-1) • • • XtnA (-n) • • • Xtnian (-n) (resp. M = Xtlll(-!)••• Xtliai (-1) • • • XtnA(-n) • • • XtB>OB(-n)y/Xtl(-l)-~X.n(-n) ) satisfying the following conditions : (i) ak = ak H h an for k = 1,..., n, (ii) for each k = 1,..., n, tk,i d U,2 d -• d tk,aj (resp. sk ^ tkA -< tk,2 d —1 *k,(Xk)j (iii) for each j = 2,..., n and k = l,...,aj, tj-itk ~< tj,k or tj-i,k = tj,k = 0 (resp. tj-i^k -< tjik or tj-\ik = tjik = 0 and Sj-\ -< Sj, and Sj (j = 1,..., n) satisfy the conditions of Proposition 1.1), (iv) there is no pair (Xtpj(-p),Xtpj+1(-p)) with tpj = tpj+i = 0, (v) there is no pair (Xtp k(— p),Xtq z(—q)) with k > I, p < q such that
92 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN tp,k = a> tq,i — fl and q — p = n — a, (vi) there is no pair (Xtp fc(— p),Xtqa(—q)) with k <l, p < q such that tp,k = &, tqj = a and q — p = n — a + 1. Now, consider the Dn type. Let A = (aij)ijej be the generalized Cartan matrix of type Dn. For i e I and m G Z, Ai(m) is Vn_3(m + l)-1yn_2(m)yn_2(m + l)Fn_1(m)-1Fn(m)-1 for i = n - 2, Fn_2(ra+ l)~1Fn_i(ra)Fn_i(ra+ 1) for z = n- 1, 1Fn_2(m + l)-1Fn(m)yn(m + 1) for i = n, Fi_i(ra + l)-1yi(m)Fi(m+ l)yi+i(m)~1 otherwise. Let B = {1,2,..., n, n,..., 1}, and define a total ordering on B by 1 -< 2 -< • • • -< n - 1 -< n,n -< n- 1 -< < T. For m G Z, we introduce new variables Fi_i(ra + l)_1Fi(m) for i = 1,... , n — 2,n, *i(ro) = X?(m) = ^ | Fn_2(m + 1) 1Fn_i(m)Fn(m) for i = n - 1, f Yi_i(m+ (n — z))Y;(ra + (n — i))_1 for z = 1,... , n — 2, Fn_2(m + l)yn_i(m + l)"1^™ + 1)_1 for i■ = n - 1, [ Fn_i(ra)Fn(ra + 1)_1 for i = n. Then Xi(m) (i G /, m e Z) satisfy the relations in [7, Lemma 2.16]. Set Ai for 2 = 1,... ,n — 2, An_i+An for z = n-l, 2An for i = n, ^2An._i for ii = n + 1. Then any dominant integral weight A can be expressed as one of the following: • aiUJi H h anwn, • aiui H h an_iu;n_i + an+iu;n+i, • aia;i H h ancjn + An, • ai^i H h an_iu;n_i + an+iu;n+i + An_i, where a^ G Z>o for all i = 1, • • • , n + 1. Now, consider the case of u;n, cjn+i, An and An_i for Z)n. Proposition 7.3. LetM0 = Yn(-n)Yn(-n+l) (resp. Fn_i(-n)Fn_i(-n+l)) be a maximal vector of weight uon = 2An (resp. u>n+i = 2An_i). Then the connected component M.D (Mo) of Ai containing Mq is characterized as the set of monomials of the form
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 93 ^(-l^HO.-JU-n) satisfying the following conditions : (i) ij -< ij+i, or (ij,ij+i) = (n,n) or (n,n), (ii) ik = n implies n — k is even (resp. odd) and ik =n implies n — k is odd (resp. even), (iii) ip = a and iq = a implies that q — p is neither n — a nor (n— 1) — a. Proof. Our claim can be proved by the same argument in the proof of Proposition 2.20 in [6]. □ For a monomial M — ^Yi(k)Yi(k + 1), we denote by \M\ the monomial Yi(k), which has the same weight as that of M. Then we have Proposition 7.4. Let Mo = Yn(—n) (resp. Yn-i(—n)) be a maximal vector of weight An (resp. An_i). Then the connected component A4d(Mq) of Ai containing Mq is characterized as the set of monomials of the form \^Xh(-l)Xi2(-2)---Xin(-n)\ satisfying the following conditions : (i) 1 ^ zi -< i2 -< < in di T, (ii) there is no pair (ip = a, iq = a) for all p,q = 1,..., n, (iii) ik = n implies n — k is even (resp. odd), (iv) ik = n implies n — k is odd (resp. even). Proof. Our claim can be proved by the same argument in the proof of Proposition 2.21 in [6]. □ For general case, we know that the expression of M in terms of Xi(m) is not unique. But, by the similar algorithm as in the case of Cn, we have a canonical expression of M. (Compare with [7, Theorem 2.25]). Theorem 7.5. (a) Let A = a\u\ -\ 1- anuon (resp. a\U\ -\ h an-iujn-\ + cbn+\wn+\). Then the connected component A4d(Mo) containing the maximal vector M0 = yi(-l)ai •••(Fn_i(-n)yn(-n))a"-1(F„(-n)y„(-n + l))a- = X1(-ir-..(X1(-l)---X„(-n))a" (resp. M0 = yi(-l)ai---(yn-i(-n)yn(-n))a"-1(lrn-i(-n)y„-i(-n + l))a"+1 = *! (-l)«i ...(*! (-1) •• • Xw(-n))a^ ) is characterized as the set of monomials M = Xtlil (1) • • • Xtltai (1) • • • XtnA (n) • • • Xtnian (n) satisfying the following conditions :
94 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN (i) ak = ak H \-anfork = l,...,n, (ii) for each k = 1,..., n, tk,i •< tk$ d * * * d U,afe, (iii) /or eac/i j = 2,..., n and k = 1,..., ctj, tj-i,k -< *j,fc, or (tj-iik,tj,k) = (n,n) or (n,n), (iv) there is no pair (Xtp k(—p), Xtql(—q)) with k > I, p < q such that tPyk — a, tqj = a and q — p = (n — 1) — a, (v) there is no pair (Xtpk(—p),Xtql(—q)) with k <l, p < q such that tp,k — a, tq,i — fl o>nd q — p = n — a, (vi) for each j, if every Uj (i = 1,..., n) exists in M, then tij (j = 1,..., n) satisfy the conditions of Proposition 7.3. (b) Let A = aiui H h anujn + An {resp. aiui H h an_iu;n_i + an+iujn+i + An_i). Then the connected component M.d(Mq) containing the maximal vector M0 = FiC-ir • • • (Yn_1(-n)Yn(-n))a^(Yn(-n)Yn(-n + l))a»Yn(-n) X1(-l)a*---(X1{-l)---Xn{-n))a" y/X1(-l)--Xn(-n) (resp. M0 = yi(-l)ai • • • {Yn^{~n)Yn{-n))a^ x {Yn^{-n)Yn^{-n + l^^l^-n) = X1(-l)ai-..(X1(-l)...X?r(-n))a^ is characterized as the set of monomials ) A(-i)--M-n) y/Xtl(-l)~-X.n(-n) M = Xtlil (-1) • • • Xtl<ai (-1) • • • Xtntl (-n) • • • Xtn,Qn (-n) satisfying the following conditions : (i) ak = ak H h an /or A; = 1,..., rc, (ii) /or eac/i A; = 1,..., n, sk ^ tk,i •< tk,2 di * * * d £/c,afe, (iii) /or eac/i j = 2,...,n and k = 1,..., a^, tj_i,fc ^ tj,fc, or (tj-iik,tjik) = (n,n) or (n,n), (iv) £/iere zs no pair (Xtpik(—p)^Xtqa(—q)) with k >l, p < q such that tp,k = aj tq,i — a cmd q — p = (n — 1) — a, (v) there is no pair {Xtpk(—p),Xtql(—q)) with k <l, p < q such that tp,k = ^, tq,i — o o>nd q — p = n — a, (vi) for each j, if every Uj (i = 1,..., n) exists in M, then Uj (j = l satisfy the condition of Proposition 7.3, (vii) Sj (j = 1,..., n) satisfy the conditions of Proposition 7.4. ,n) Let Tb(A) and To (A) be the sets of J3n-tableaux and Z}n-tableaux of shape A given by Kashiwara and Nakashima, respectively [11]. Then we have
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 95 Theorem 7.6. Let X be a dominant integral weight and let Mq be the monomial given in Theorem 7.2 and Theorem 7.5. Then there are crystal isomorphisms \J)b • Mb(M0) -► TB(X) and </>D : MD(M0) -► TD(X). Proof. Combining Proposition 3.6 and Proposition 3.9 in [6], and the argument in Theorem 6.4, we get the desired results. □ Remark 7.7. In [5, 6], for a dominant integral weight A, we gave the characterization of M(X) containing a maximal vector of weight A which is different from M0 given in this paper (for example, M0 = Yi(0)ai •••Fn(0)an for Uq(An)) and discussed the connection of the tableau realization 5(A) given by Kim and Shin in [13]. Moreover, using insertion scheme introduced [14], we obtained a crystal isomorphism between M(X) and the tableau realization T(A). 8. Characterization: An' type In this section, we give the characterization of the irreducible highest weight crystal B(X) and the crystal J5(oo) over Uq(An '). Let / = {0,1,..., n) be the index set and identify / with Z/(n + 1)Z so that —l = n<Q = n + l. Set C = (cij)i^j be the set given in section 5. Then Ai(m) can be written as (8.1) Mm) = Yi_1(m+l)-1Yi(m)Yi(m+l)Yi+1(m)-1 for i G I = Z/(n + 1)Z, m G Z>0. Definition 8.1. For a monomial M expressed as s rp(r)nnt(m«+i)-1nt(mt) with r G Z>o and 0 = ra0 < rrti < • • • < ms < ms+i = r, we say p satisfies the ground-state condition for A^ if the condition r = ro mod (n +1) implies p = k — r^. For a monomial M = YP1 (ni)YP2(n2) • • • YPl (m) l\t Yat (mt)_176t (nt) with at + mt = bt + nt (mod n + 1), we define M(k)+ (k e Z>0) and M(k)~ (k G N) by M(k)+= n Ybt(nt)=l[Ybt(k), t:nt=k t M(k)~= J! Yatimtr^IlY^k)-1. t:mt=k t Theorem 8.2. [15, Theorem 4.7] Let X = Ah -\ hAi£ (k < i2 < • • * < U) be a dominant integral weight of level I and let Mq = Y^ (0) • • • Y^ (0) be the monomial of weight X such that e^Mo = 0 for all i G /. Then the connected component Aifi(Mo) in M. containing Mq is characterized as the set of monomials of the form M = YP1 (ni)YP2 (n2) ■■■YPI (m) fj Yat {mt)-xYbt (nt) t
96 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN with ni>0 (1 < i < 1) satisfying the following conditions : (i) at -f mt = bt + nt (mod n + 1) and 0 < nt < mt for each £, (ii) deg(M(0)+) = I and deg(M(fc)+ • M(Jfc)-) = 0 for jfc € / \ {0}. (iii) for each ik (k = 1,..., Z), the element pk satisfies the ground-state condition for Aik. Let i = (zo, 2i,..., in) be an (n + l)-tuple of elements in Z>o with Y^=o ^ = '• For each A; € Z>o, we define new variables Xi(k) = l[[xt(ky\ t=0 where Xt(k) = Yt-i(k + l)~1Yt(k). Then Theorem 8.2 can be rewritten as follows. Corollary 8.3. [15, Theorem 4.9] Let A = Ah + • • • + Aiz (h < z? < • • • < 2*) fre a dominant integral weight of level I and let Mo = ^(O) • • -^(0) be the monomial of weight A such that e^Mo = 0 for all i € /. Then the connected component M.(X) of Ai containing Mq is the set of monomials of the form r-l YP1(r)YP2(r)---Ypl(r)l[Xik(k) k=0 with r € Z>o and pk (k = 1,..., I) satisfying the ground-state condition for A^fe. Let B be a level I perfect crystal and let ^(A) be the set of all A-paths in B. Then we have Theorem 8.4. [15, Theorem 5.1] Let A = A^ H hA;, (n < i2 < • • • < U) be a dominant integral weight of level I and let Mo be the monomial given in Corollary 8.3. Then there is a U'q{An ^-crystal isomorphism i/j£ : M^(Mo) —» V{\). Let M = nr=o FIn>o Yi(n)Vi^l be a modified Nakajima monomial. For each k > 0, we define M^\ (resp. M7ks) to be the product of all positive powers (resp. negative powers) of Fi(fc)'s appearing in M, and set M^) — MX,\M7ky The modified Nakajima monomials in A4(l) are characterized in the following theorem. Theorem 8.5. [8, Theorem 4.1] (i) munt > 0, mt ^ nt, {«-n M(l) = { M-M^K)-1^^)! (ii) at + mt = bt + n* (mod n + 1), (iii) deg M(fc) = 0 for all k>0 Example 8.6. Let g = A^ and m = F0(i)>o(2)-1yi.(o)'2y2(o)y2(i)-1vr2(2)y3(o) 1.
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 97 Then M can be expressed as M = (yo(l)^i(0)-1)(r3(2m(0)-1)(Fo(2)-1F2(0))(F2(l)-1i3(0)) 1, which satisfies the conditions (i)-(iii). Hence M € M(l). We have another expression of A4(1) using the variables Xi(m). Corollary 8.7. [8, Corollary 4.3] (i) Oi(k)=0 for fc>0, (ii) ]Tai(fc) = 0 for all fc>0 J (i) Oi(fc) = 0 for fc>0, (ii) ^diik) = 0 for all k > 0 J ' M(i)=iM=nn^(fc)ai(fe) fc>o;=o I fc>Oi=0 ai(fc)-ai+i(fc-l) z=0 In the following theorem, using the characterization given in Corollary 8.7, we will construct a natural crystal isomorphism between M(l) and P(oo). Theorem 8.8. [8, Theorem 5.1] There exists a U'q(An)-crystal isomorphism $ : M(l) —► P(oo) defined by M = Y[Y[Xi(k)ai{k)l fc>Oi=0 OO >-* P = (p(*))2io = ®M*), ai(*), • • •, «»(*))• fc=0 Example 8.9. (1) Let g = A^ and m = y0(o)-2y0(i)-1Ki(o)2Yi(2)y2(i)y2(2)-1 i be the modified Nakajima monomial given in Example 8.6. Then M can be expressed as M = Xo(l)X2(l)-1Xo(0)-2X1(0)21, which is mapped onto $(M) = (..., (0,0,0), (1,0, -1), (-2,2,0)) € P(oo). (2) Conversely, if p = (..., (0,0,0), (-1,1,0), (0, -1,1), (1,0, -1)) e 7>(oo), then we have ^(p) = Xo(2)-1X1(2)X1(l)-1X2(l)X0(0)X2(0)-11 = Y0(Q)Y0(3)-1Y2(0)-1Y2(3)1.
98 SEOK-JIN KANG, JEONG-AH KIM, AND DONG-UY SHIN References [I] S.-J. Kang, Crystal bases for quantum affine algebras and combinatorics of Young walls, Proc. London Math. Soc. 86 (2003), 29-69. [2] S.-J. Kang, M. Kashiwara, K. C. Misra, Crystal bases of Verma modules for the quantum affine Lie algebras, Composito Math. 92 (1994), 299-325. [3] S.-J. Kang, M. Kashiwara, K. C. Misra, T. Miwa, T. Nakashima, A. Nakayashiki, Affine crystals and vertex models, Int. J. Mod. Phys. A. Suppl. 1A (1992), 449-484. [4] S.-J. Kang, M. Kashiwara, K. C. Misra, T. Miwa, T. Nakashima, A. Nakayashiki, Perfect crystals of quantum affine Lie algebras, Duke Math. J. 68 (1992), 499-607. [5] S.-J. Kang, J.-A. Kim, D.-U. Shin, Monomial realization of crystal bases for special linear Lie algebras, J. Algebra 274 (2004), 629-642. [6] S.-J. Kang, J.-A. Kim, D.-U. Shin, Crystal bases for quantum classical algebras and Nakajima's monomials, Publ. Res. Inst. Math. Sci. 40 (2004), 758-791 [7] S.-J. Kang, J.-A. Kim, D.-U. Shin, Modified Nakajima Monomials and the Crystal B(oo), submitted [8] M. Kashiwara, On crystal bases of the q-analogue of universal enveloping algebras, Duke Math. J. 63 (1991), 465-516. [9] M. Kashiwara, The crystal base and Littlemann's refined Demazure character formula, Duke Math. J. 71 (1993), 839-858. [10] M. Kashiwara, Realizations of crystals, in Combinatorial and Geometric Representation Theory (Seoul, 2001), S.-J. Kang, K.-H. Lee (eds.), Contemp. Math. 325 (2003), 133-139, Amer. Math. Soc. [II] M. Kashiwara, T. Nakashima, Crystal graphs for representations of the q-analogue of classical Lie algebras, J. Algebra 165 (1994), 295-345. [12] J.-A. Kim, Monomial realization of crystal graphs for Uq(A\l)), Math. Ann. 332 (2005), 17-35 [13] J.-A. Kim, D.-U. Shin, Insertion scheme for the classical Lie algebras, Comm. Algebra 32 (2004), 3139-3167 [14] J.-A. Kim, D.-U. Shin, Correspondence between Young walls and Young tableaux and its application, J. Algebra 282 (2004), 728-757 [15] P. Littlemann, A Littlewood-Rechardson rule for symmetrizable Kac-Moody Lie algebras, Invent. Math. 116 (1994), 329-346. [16] P. Littlemann, Paths and root operators in representation theory, Ann. of Math. 142 (1995), 499-525. [17] T. Nakashima, Polyhedral realizations of crystal bases for integrable highest weight modules, J. Algebra 219 (1999), 571-597. [18] T. Nakashima, A. Zelevinsky Polyhedral realizations of crystal bases for quantized Kac-Moody algebras, Adv. Math. 131 (1997), 253-278. [19] H. Nakajima, Quiver varieties and finite dimensional representations of quantumn affine algebras, J. Amer. Math. Soc. 14 (2001), 145-238. [20] H. Nakajima, Quiver varieties and tensor products, Invent. Math. 146 (2001), 399-449. [21] H. Nakajima, t-analogue of the q-characters of finite dimensional representations of quantum affine algebras, in "Physics and Combinatorics", Proceedings of the Nagoya 2000 International Workshop, World Scientific, (2001), 195-218. [22] H. Nakajima, t-analogs of q-characters of quantum affine algebras of type An, Dn, Contemp. Math. 325 (2003), Amer. Math. Soc, 141-160.
NAKAJIMA'S MONOMIALS AND CRYSTAL BASES 99 [23] D.-U. Shin, Crystal Bases and Monomials for Uq(G 2)-modules, Comm. Algebra 34 (2006), 129-142 * Department of Mathematical Sciences and Research Institute of Mathematics, Seoul National University, Seoul 151-747, Korea E-mail address: sjkang@math.snu.ac.kr ^Department of Mathematics, University of Seoul, Seoul 130-743, Korea E-mail address: jakimfluos. ac.kr * School of Mathematics, Korea Institute for Advanced Study, Seoul 130-722, Korea E-mail address: shindongQkias.re.kr
This page intentionally left blank
Contemporary Mathematics Volume 413, 2006 A New Lie Bialgebra Structure on s/(2,1) Gizem Karaali Abstract. Unlike in the Lie algebra case, there are normalized non-skew- symmetric r—matrices on simple Lie super algebras with non-degenerate Killing forms that cannot be obtained using Belavin-Drinfeld type data. We explicitly construct such an r—matrix on the Lie superalgebra 5/(2,1). We describe the Lie bialgebra structure related to this r—matrix and show that it makes 5/(2,1) into the Drinfeld double of a four-dimensional subalgebra. It is well-known that non-skew-symmetric r—matrices (describing quasitriangu- lar Lie bialgebra structures) on simple Lie algebras are classified by Belavin-Drinfeld triples, (the original references are [BD1, BD2], more pedagogical presentations providing ample background can be found in [CP, ES]). A similar construction using Belavin-Drinfeld type triples is possible for simple Lie superalgebras with non-degenerate Killing forms, see [Kar]. Surprisingly, though, in the super setting, there are certain non-skew-symmetric r—matrices that do not fit such a description. The purpose of this note is to study the super Lie bialgebra structure associated to such an r—matrix on the simple Lie superalgebra 5/(2,1). We begin in Section 1 with a short overview of the Yang-Baxter equations. In Section 2, we summarize the Belavin-Drinfeld result for simple Lie algebras. In Section 3, we recall some basic definitions and results about simple Lie superalgebras, and after developing the necessary ingredients we state the main theorem of [Kar] in Section 4. This theorem is very much in the spirit of the Belavin-Drinfeld result. It tells us that given a Belavin-Drinfeld type triple, one can construct a normalized non-skew- symmetric r—matrix in a way similar to the construction in the Lie algebra case. However, unlike in the Lie algebra case, this is not a complete classification result; in Section 5, we construct a normalized non-skew-symmetric r—matrix that cannot be obtained by this theorem. After a short interlude (Section 6) providing some background on super Lie bialgebra structures and some basic constructions related to them, we describe explicitly in Section 7 the super Lie bialgebra structure on 5/(2,1) determined by the r—matrix of Section 5; this structure makes 5/(2,1) into the Drinfeld double of a four-dimensional subalgebra. A comparison with 2000 Mathematics Subject Classification. Primary 17B62, 17B20. Key words and phrases. Classical r-matrices, Lie superalgebras, super Lie bialgebras, Drinfeld double. ©2006 American Mathematical Society 101
102 GIZEM KARAALI the standard super Lie bialgebra structure is also provided in this section. We end in Section 8 with a brief discussion of the results and further directions for investigation. Acknowledgments. The author thanks N. Reshetikhin, V. Serganova and M. Yakimov for their comments and suggestions. Also it is a pleasure to thank the organizers B. Parshall, G. Benkart, J. Jantzen, Z. Lin, and D. Nakano, of the 2004 AMS-IMS-SIAM Summer Research Conference on Representations of Algebraic Groups, Quantum Groups and Lie Algebras, where the author had the opportunity to present her results in [Kar]. 1. The Yang-Baxter Equations 1.1. Historical Background. Among all the equations named after Yang and Baxter, the earliest to be studied was the one which we now call the quantum Yang-Baxter equation. In its early incarnations, the current form of the quantum Yang-Baxter equation showed up mostly as a consistency condition under various names, (e.g. the star-triangle equation, the triangle equation, the factorization equation). The main starting points were the commuting transfer matrices in statistical mechanics and factorizable S'-matrices in field theory. More specifically, each solution of the quantum Yang-Baxter equation can be treated, on the one hand, as a vertex weight matrix of an exactly solvable statistical model on a plane lattice, and on the other hand, as an exact factorized S-matrix in some (1 + l)-dimensional field theory. The equation also arises as the consistency condition for the Bethe Ansatz solution of quantum field models and those of one-dimensional magnetics. For a survey of these early results, along with more on the quantum Yang-Baxter equation see [Jl]. [J2] collects reprints of several papers related to the topic, and includes many of the pioneering papers in the area. These early works, along with the development of soliton theory, helped the Leningrad school develop the quantum inverse scattering method (QISM) to bring together classical and quantum integrable systems. The classical Yang-Baxter equation came into the scene as a result of the correspondence principle in QISM. More specifically, one can obtain the classical Yang-Baxter equation via a formal limiting operation (taking the "semiclassical limit") applied to the quantum Yang-Baxter equation. A complete classification of the non-skew-symmetric solutions of the classical Yang-Baxter equation exists in the case when the underlying Lie algebra is simple; see [BD1] and [BD2] for the original proofs, and [ES] for a more pedagogical exposition. For a geometric interpretation of the solutions of the classical Yang- Baxter equation (called r-matrices), see [Dl], where Drinfeld relates r-matrices to Lie bialgebra structures on the associated Lie algebra and subsequently to Poisson- Lie structures on the corresponding Lie group. [STS] provides a different geometric interpretation of the solutions. During the development of QISM, it became clear that under quantization of a classical system, certain structures (e.g. the Poisson-Lie structure involved) undergo quantum deformations. While quantizing certain solutions of the classical Yang-Baxter equation, Kulish and Reshetikhin in [KR] came up with a deformation of the universal enveloping algebra of 5/(2). This example was later generalized to
A NEW LIE BIALGEBRA STRUCTURE ON sl(2y 1) 103 arbitrary simple and affine Lie algebras, and came to be known as a quantized enveloping algebra. Eventually these objects came to be known as quantum groups. In [D2] Drinfeld showed that this new concept could be viewed in the framework of Hopf algebras. The language of Hopf algebras also turned out to be the appropriate one for QISM. See [D3] for an overview of this approach. [FRT] provides another approach to quantum groups that is in spirit closer to QISM. 1.2. The Classical Yang-Baxter Equation. In the literature, the term classical Yang-Baxter equation is mainly used for the following general functional equation: [r12(m - u2), r13(m - u3)] + [r12(ui - u2), r23(u2 - u3)] + [r13(ui - u3), r23(u2 - u3)] = 0, where r(u) is a meromorphic function taking values in g(gg, where g is a Lie algebra. When we assume that r(u) = r e g <S> g, we get the simpler constant coefficient equation: [r12jr13] + [r12jr23] + [r13jf.23j=0! This is the equation we will be interested in in the present note. We explain the notation above: If r = ]T\ a{<S>b{ G g<8> g, then: r12 = ^2 ai ® bi ® 1, r13 = ^2 ai ® 1 ® bu r23 = ^ 1 <g> a{ (g) bu iii and: [r125r13] = ^T[ai,aj]®bi®bj, i>3 [r12, r23] = ^ o» (g) [6», aj] (g) bj [r13, r23] = £] a» <g> a,- <g> [6», ^-] *»j A solution r(u) to the classical Yang-Baxter equation is called a classical r- matrix (or simply an r-matrix). r is called non-skew-symmetric if it satisfies: r + T(r) + 0, where T : g <8> g -^ g® g is the standard permutation map on g <g) g mapping any element a (g) 6 to 6 (g) a, for a, 6 G g. Lemmas by Whitehad (see, for instance, [W]) imply that in order to study non-skew-symmetric r—matrices on simple Lie algebras, it suffices to consider only the normalized non-skew-symmetric r—matrices, i.e. those r—matrices that satisfy: r + T(r) = ft, where Q stands for the element of (g <S> g)9 that corresponds to the quadratic Casimir element in the universal enveloping algebra iig of g. In [BD1] and [BD2] Belavin and Drinfeld classified such r—matrices. Their classification is given by a discrete parameter called an admissible (or a Belavin-Drinfeld) triple, and a continuous parameter ro which satisfies certain relations depending on the given admissible triple. We will describe this result in more detail in the next section.
104 GIZEM KARAALI 2. Classification Theorem for Lie Algebras Here we present the main result of [BD1] and [BD2] for non-skew-symmetric i—matrices on Lie algebras. Let g be a simple Lie algebra with non-degenerate Killing form (•,•). Denote by Ct the element of (g <E> g)0 that corresponds to the quadratic Casimir element in the universal enveloping algebra iigoig. Fix a positive Borel subalgebra b+ and a Cartan subalgebra J) C b+. Let T = {ai, 0:2> * * * > otr} be the set of simple roots of g. An admissible triple is a triple (Fi, T2, r) where T^ C T and r : Ti -^ T2 is a bijection such that (1) for any a,/3e Tu (r(a),r(/3)) = (a,/?); (2) for any a G Ti there exists afceN such that rk(a) 0 IY Fix a system of Weyl-Chevalley generators Xa, Ya, Ha for a G T. Recall that these elements generate the Lie algebra g with the denning relations: [Xai, Yaj] = 5ijHaj, [Hai,Xa.] = aijXaj and [Hai,Ya.] = -a^Y^. for all a*, a, G T, (where a*j — aj(H(*i) — (q'^) )? al°ng w^h the well-known Serre relations. Denote by g^ the subalgebra of g generated by the elements Xa, Fa, i7a for all a eTi. We define a map <p by: for all a G Ti. This can then be extended uniquely to an isomorphism <p : gi —► g2 because the relations between Xa,lra, Ha for a G Ti will be the same as the relations between XT^,YT^, HT^ for a G Ti, (r is an isometry). Next extend r to a bijection r : Ti —► r2, where T^ is the set of those roots which can be written as a nonnegative integral linear combination of the elements of T^. In each root space ga, choose an element ea such that (ea, e_a) = 1 for any a and y>(ea) = e^a) for all a G Fi Finally, define a partial order on the set of all positive roots: a -< (3 if and only if there exists afceN such that /? = rk(a). Note that if a -< /?, then necessarily a G Ti, /? G r2. Now we can state the Belavin-Drinfeld theorem ([BD2]; also see [ES]): Theorem 2.1. (%) //r0Gf|®[) satisfies ro + T(ro) = n0, (2.1) (r(a) (8) l)(r0) + (1 (8) a)(r0) = 0 /or all a G Ti, (2.2) where Oq G J) 0 J) is tte I)—component of ft, then the element r of g ® g defined by: a>0 a,/3>0,a^/3 25 a solution to the system: r + T(r) = ft, (2.3) [f.l2j r13j + [f.l2j r23] + [rl3j r23] = Q {2A) (2) Any solution to this system can be obtained as above from some admissible triple (Fi,r2,r) and some ro G J)<g>J) that satisfies Equations (2.1) and (2.2), &?/ choosing a suitable triangular decomposition of g and a set of Weyl-Chevalley generators.
A NEW LIE BIALGEBRA STRUCTURE ON 5/(2,1) 105 3. Basic Facts About Lie Super algebras We now wish to study the analogous super structures, and develop a similar theory for non-skew-symmetric r—matrices on Lie super algebras. A full superization of [BD1, BD2] was first attempted in [LSe]. However, no proofs were provided there, and furthermore, there were certain mistakes in the paper due to a gap in the arguments regarding the classification of trigonometric solutions of the classical Yang-Baxter equation, and we wish to start afresh. This section collects together some background information on Lie superalge- bras. In the next section we will start describing our theory for non-skew-symmetric r—matrices on Lie superalgebras. 3.1. Definitions and Basic Examples. A super vector space is a Z/2Z- graded vector space V = Vq 0 Vy. Elements of V that lie completely in Vq, (respectively in Vy) are called even or homogeneous of parity 0, (respectively odd or homogeneous of parity 1). A superalgebra is a Z/2Z-graded algebra A = Aq 0 Aj where the algebra operation satisfies: Xa-yp € Aa+p for all xa e Aa, and yp e Ap, where a, (3 e Z/2Z. A Lie superalgebra is a superalgebra q = g$ 0 Qj where the algebra operation is called a super bracket, denoted by [•,•], and satisfies the following conditions: (1) Graded skew-symmetry: [51,52] = -(-l)l9lllS2l[<72,si]; (2) Graded Jacobi identity: (-1)|91||S3|[[<?1,<?2],<?3] + (-l)MMll92,93},9i] + (-l)MM[[93,9i],92] = 0; where each gi is homogeneous of parity \gi\. The super bracket extends linearly to non-homogeneous elements of g. Note that we have: [0a, 0/3] Cga+/3 a,/?eZ/2Z. The above definition for a Lie superalgebra g = Qq 0 Qj clearly implies that Qq is a Lie algebra with the restriction of the super bracket and that Qj is a ^q—module. For more on Lie superalgebras, one can look at [Kac] and [Sc]. We will also need the super twist map Ts : V <8>V —► V <8>V defined on the homogeneous elements of a given super vector space V = Vq 0 Vy as T5(a®&) = (-l)|a||6|&<g)a. In other words, Ts is the permutation map in the category of super vector spaces.1 We can also remark here that in this categorical language, Lie superalgebras are merely the Lie algebra objects in the category of super vector spaces. We will not be emphasizing this point of view much though.
106 GIZEM KARAALI Example 3.1. If V = Vq 0 Vy is a super vector space, then the algebra of endomorphisms of V has a natural Z/2Z-grading, and with the super bracket [/,<?]= /<7-(-l)l/llffl<7/, it becomes a Lie superalgebra, denoted by gl(m,n) where m = dim(Vo) and n = dim(Vj). Clearly, the even elements of gl(m,n) preserve the parity of any given homogeneous vector in V, while the odd elements change it. Note also that if we fix a homogeneous basis for V, we can view gl(m, n) as the space of (m + n) x (m + n) matrices. Example 3.2. Consider the following subsuperalgebra of gl(m, n): sl(m,n) = {Ae gl(m,n) \ str(A) = 0}. Here the supertrace str of any A e gl(m, n) is denned by str(A) =tr(a)-tr(d), where a and d are, respectively, the upper m x m and lower n x n even diagonal blocks of A written in any homogeneous basis for V. Clearly the supertrace is independent of the choice of a homogeneous basis for V. sl(m,n) is a simple Lie superalgebra for m =fi n; in other words, it has no nontrivial graded ideals. For the rest of this paper, unless otherwise stated, let q be a simple (nonabelian) Lie superalgebra with non-degenerate Killing form. In fact, most of our results can be extended to the whole class of classical Lie superalgebras because most of the statements involving the Killing form may be asserted more generally for a non- degenerate invariant form. 3.2. The Quadratic Casimir Element: Let {/a} be a homogeneous basis for q and denote by {/a*} the dual basis of q with respect to the non-degenerate (Killing) form. Thus we have: (la, 1(3*) = <W- Denote the parity of a homogeneous element x G q by \x\; then |/a| = |/a*|, since the Killing form is consistent. Hence the quadratic Casimir element of q is 11 = £(-1)""l|/a*l/° ® la* = £(-l)|/alIa ® I«*. a a For a definition of the Casimir element, one can look at [Kac, Sc]. Example 3.3. Let g = gl(m,n). Fix the basis {e^|l < i, j < m + n}, where \eij | = 0 if and only if 1 < i, j < m or m + 1 < i, j < m + n. The dual basis is: where M = J° {f 3 - m' m \l if j>m, and ( , ) is the supertrace form. Then this gives us: $7 = £(-l)l'«lla ® Ia* = ^(-i)l^ley ® (-l)Wei4 = ^(-1)^^ ® e>i.
A NEW LIE BIALGEBRA STRUCTURE ON 5/(2,1) 107 3.3. Borel subsuperalgebras and Dynkin diagrams: Let J) C Q be a Cartan subalgebra. By definition, J) C Qq is a Cartan subalgebra of the even part of g. Let A = Aq + Ay be the set of all roots of q associated with the Cartan subalgebra J), where Aq and Ay are the even and odd roots respectively. We recall here that a Lie subsuperalgebra b of a Lie superalgebra g is a Borel subsuperalgebra if there is some Cartan subsuperalgebra J) of q and some base T for A, such that b = f>0 0 fla, aEA+ where A+ is the set of all positive roots. In the Lie algebra case, subalgebras given by this definition are all maximally solvable, and all maximally solvable subalgebras of a simple Lie algebra are of this type. Therefore, this definition agrees with the usual definition of a Borel subalgebra as a maximally solvable subalgebra. However, Borel subsuperalgebras as denned above are not necessarily maximally solvable. For instance if a is a positive isotropic root of the simple Lie superalgebra g, and if b is the sum of all the positive root spaces, then b is a Borel subsuperalgebra, but it is not maximally solvable. The (parabolic) subsuperalgebra p = b 0 g_a is also solvable. In fact, maximally solvable subsuperalgebras may be more complicated than merely parabolic. (See [Sh] for maximally solvable subsuperalgebras of gl(m,n) and sl(m,n).) We also note here that different Borel subsuperalgebras may correspond to different Dynkin diagrams and Cartan matrices. Let us then fix some Borel subsuperalgebra b, or equivalently some set of simple roots F and the associated Dynkin diagram D. 4. The Construction Theorem for Lie Superalgebras 4.1. The Data for the Theorem: In this setup, let ri,T2 C T be two subsets and r : I\ —► T2 be a bijection. The triple (ri,r2,r) will be called admissible if: (1) for any a,/3e I\, (r(a),r(/3)) = (a,/?); (2) for any a e I\ there exists & k eN such that rk(a) 0 Ti; (3) r preserves the grading of the root space. Given an admissible triple (ri,r2,r), let I\ for i = 1,2 be the set of those roots that are nonnegative integral linear combinations of the elements of IV Then r extends linearly to a bijection f : F\ —► T2, so we can define a partial order on A+: a -< (3 if and only if there exists afcGN such that /? = rk(a). For any a e T, pick a nonzero ea GgQ. Since each ga is one dimensional, and the Killing form is a non-degenerate pairing of ga with g_a, one can uniquely pick e_a e Q-a such that (ea,e_a) = 1. Therefore we have: [6q,6_Q;J ^Cq, 6 — Q; J/l/Q ,
108 GIZEM KARAALI where ha G J) is defined by (ha, h) = a(h) for all h G J). Since the set {/ia|^ £ T} is a basis for J), we can write Oo, the J)—part of 0, as follows: f^o := 2J ^a ® ^a*' aGr where the set {/ia*|a £ T} is the basis in J) dual to {ha\a G T}. Next, for each a G A+\F, choose a nonzero ea G ga- This will uniquely determine e_a G g_a satisfying (ecne_a) = 1. Then the duals with respect to the standard (Killing) form will be: p * — e for all positive roots a, where |a| is the parity of the root a. Therefore the quadratic Casimir element of g will be: i aGr a€A = ft0 + 5Z (-l)Hea®e_a+ 5Z e-«^ea. aGA+ aGA+ Example 4.1. Once again, let g = gl(m, n). We can think of g as the space of (m + n) x (ra + n) matrices. Let J) and b be the (ra + n) x (m + n) diagonal matrices and the upper triangular matrices, respectively. Then the positive root spaces are spanned by {eij\i < j}. If for each positive root a, we let ea be the unique e^ G ga, then i < j and e_a = (-l)^ej;. We have: p * _ p. .* _ / in[*]p.. _ p e_Q* = (-1)I*V = (-l)[il(-l)[ileii = (-l)|Q|ea, and the above formula for Q agrees with the Casimir element found earlier in Example 3.3. 4.2. Statement of the Theorem: At this point, we have developed enough notation and terminology to be able to state a result partially analogous to Theorem 2.1. The following is the main theorem of [Kar] and we refer the reader there for its proof: Theorem 4.2. Let r0 G J) ® J) satisfy: r0 + T5(r0) = fto, (4.1) (r(a) <g> l)(r0) + (1 <8> a)(r0) = 0 /or a// a G I\. (4.2) T/ien £/ie element r of g ® g defined by: r = r0 + ^e_a ®ea + ^ (e_a ® e/3 - (-l)1"^ ® e_a) (*) a>0 a,/3>0,a^/3 25 a solution to the system: r + Ts(r) = Q, (4.3) [r12j r13] + [r12j r23] + [rl3j r23] = Q (4>4)
A NEW LIE BIALGEBRA STRUCTURE ON 5/(2,1) 109 Remark 4.3. If g is a simple Lie algebra, then (*) reduces to the corresponding equation in Theorem 2.1. This result provides us with a method to construct many non-skew-symmetric solutions of the classical Yang-Baxter equation for simple Lie superalgebras with non-degenerate Killing forms. It clearly tells us that, given a Belavin-Drinfeld type (admissible) triple, and a continuous parameter ro G J) 0 J), one can construct a non- skew-symmetric i—matrix in a way similar to the construction in the Lie algebra case. Using this construction we can obtain the standard r—matrices and some nonstandard ones. However, we definitely do not have a full analogue of Theorem 2.1. In fact, we will see shortly (in Section 5) that Belavin-Drinfeld type triples are not sufficient to classify all non-skew-symmetric solutions. 4.3. A Note about the Proof. The following lemma is a basic step in the proof of Theorem 4.2, (see [Kar]): Lemma 4.4. Let g be a simple Lie superalgebra with non-degenerate Killing form. Fix a homogeneous basis {Ia} for g and denote by {/a*} the dual basis of g with respect to the non-degenerate (Killing) form. Let f : g —► g be an even linear map, and set r = (f ® l)fi. Then the system of equations: r + Ts(r) = n, (4.3) [ri25 ri3] + [ri2j r23] + [f.i3j r23] = 05 (44) is equivalent to the system: / + r = i, (4.5) (/ - l)[f(x), f(y)} = /([(/ - IX*), (/ - l)(y)]), (4.6) where /* stands for the adjoint of f with respect to the standard from ( , ). This lemma allows us to translate the conditions on r to conditions on the associated linear map / : g —► g. Similarly one can translate the conditions on the continuous parameter ro in the main theorem to conditions on some linear map /o : *) —► J). Thus we can restate our problem as follows: Given an admissible triple (ri,r2,r) with a linear map /o : J) —► J) satisfying those particular conditions, construct a linear map / : g —► g satisfying Equations (4.5) and (4.6). Prom this point on, the proof involves linear maps, their kernels and their images. 5. Defining the r—matrix r(f) Now we introduce our main example. This will be a (normalized) non-skew- symmetric r—matrix on g = 5/(2,1) which does not correspond to any Belavin- Drinfeld type classification. In particular, we will define a 2—tensor in g ® g satisfying Equations (4.3) and (4.4) which cannot be constructed via Theorem 4.2. Recalling Lemma 4.4, we start by defining a map f on the standard basis for g: f(En -f £33) = 0, /(^22 + ^33) = ^22 + Es3, /(£?2i) = 0, f(E12) = E12, f(E2s) = 0, f(E13) = £7i3, f(E3i) = —E13, f(Es2) = E23 + £"32-
110 GIZEM KARAALI We can then extend / to a linear map on g. It is easy to check that for any x and y in g, this function satisfies: (/ - l)[/(x), f(y)] = /([(/ - l)(a;), (/ - 1)0,)]), (4.6) which according to Lemma 4.4 is equivalent to the associated 2—tensor being an r—matrix. We write the quadratic Casimir element of g or equivalently the invariant tensor in g <g> g: Q = (En + E33) ® (-E22 - E33) + (-E22 - E33) ® (En + E33) + (£7i2 ® £?2i + £?2i <g> £?i2) + (-£?i3 <8> £?3i + £31 ® #13) + (-E23 <8> #32 + #32 ® £?23). Denning r(/) to be the 2—tensor (/ ® l)fi, we get: K/) = ro + E12 ® £72i - £?i3 ® £73i + ^32 ® ^23 ~ ^13 (8) £13 + #23 ® ^23 where ro = (—E22 — £33) ® (En + E33). It is easy to see that r(f) satisfies: r + Ts(r) = n. (4.3) Recall here that Ts is the permutation map in the category of super vector spaces. r(f) does not allow a straightforward Belavin-Drinfeld type description. In fact, we can prove that the two subsuperalgebras Im(/) and Im(/ — 1) will never be simultaneously isomorphic to root subsuperalgebras. In other words, whenever we use the root space decomposition of q to write one as a direct sum of root subspaces of g, it will no longer be possible to present the other one in a similar form. On the other hand, it can be shown that the corresponding subsuperalgebras for functions constructible by Belavin-Drinfeld type data can always be simultaneously presented as root subsuperalgebras. To see this, one would only need to review the details of the proof of Theorem 4.2 and follow how the functions / leading to r-matrices of the form (*) are explicitly constructed. All r—matrices on a simple Lie algebra satisfying Eqn.(4.3) (which in this case is equivalent to Eqn.(2.3) as the grading is trivial) are constructible by Belavin- Drinfeld type data. Thus, the existence in the super case of an 1—matrix satisfying this equation but not allowing a Belavin-Drinfeld type description provides us with yet another example when the graded case is more involved than the non-graded case. 6. Lie Bialgebra Structures on Lie Superalgebras We will study the r—matrix r(f) from Section 5 in more detail shortly. In particular we will explicitly describe the Lie bialgebra structure that it gives rise to. However, for the sake of completeness, we first present a brief summary of facts about Lie bialgebra structures on Lie superalgebras. Not much of this material is new, although the presentation (in particular in Subsection 6.3) diverges somewhat from the standard references.
A NEW LIE BIALGEBRA STRUCTURE ON sl(2, 1) 111 6.1. Cohomology of Lie super algebras. The cohomology theory of Lie super algebras is more complicated than that of Lie algebras. Even for simple Lie super algebras and for low dimensions, it is not yet completed. Here we summarize certain basic facts that we will use. For more on the cohomology theory of Lie superalgebras one can look at [F, ScZ]. Recall that if g is a Lie algebra, then an n—cochain taking values in a g—module M is an alternating n—linear map f(x\,X2, • • • , xn) of n variables in g. We can view each such n—cochain as a linear map / : /\n g —> M. In this case, the coboundary df of an n—cochain / is the (n + 1)—cochain defined by: n+l df(xU • • • , Xn+i) = ^(-l)2+1Xi/(Xi, • • • , £;, • • • , Xn+1) i=l i / ^ V -LJ J\[xii xj\i xli ' ' ' ixii''' ixji''' ixn-\~l) i<i<j<n+i If g is a Lie super algebra, then the space of n—cochains with values in a g—module M = Mq 0 My is itself a graded space. Denoting this space by Cn(g, M), we have: Cn(g,M)= 0 Homage® S'biiM). i+j=n The even part of Cn{g, M) is: i+j=n while the odd part is given by: C?(fl,M)= 0 ^(^85®^,%!). i+j=n Equivalently we can view Cn(g, M), for n > 1, as the Z/2Z-graded vector space of all super alternating n—linear maps f of gn=gxgx-'Xg into M, i.e. maps / satisfying: We set C°(jj,M) = M. The differential d is then defined as follows: For an n—cochain /, the coboundary df is an (n + 1)—cochain given by: n+l df(X\,-- ,£n+l) = 1^2(Ti(iJ)xif(Xi,'- ,Xi,'" >^n+l) i=l + ^2 a2(hJ)f([XiiXj]iXli'- ,^i,-" i^V 5^n+l) l<i<.7<n+l where the signs in the above sums are as follows: ai(i,j) = (-l)i+1(-i)I^KI/l+lxil+lx2|+-+|xi_i|)5 We note that this formula for d agrees with that of [ScZ] when we use the super alternating property of /.
112 GIZEM KARAALI If M = g ® g, then g acts on M on the left by the following extension of the adjoint representation: g . (a ® b) = (g • a) ® b + (-l)|p||a|a ® (#•&) = [#, a] ® & + (-l)^l|a|a ® [#, 6]. In this setup, a 0—cochain is a linear map /o : C —► g ® g. Therefore it is determined uniquely by /o(l) £ £ 0 £ and hence can be identified with an element r of g ® g. The coboundary dr of r is a 1—cochain defined by: dr(a) = a* r = [a ® 1 + 1 ® a, r]. A 1—cochain is a linear map / : g --> g ® g. It is a 1—cocycle if df = 0, or in other words: 0 = df(a,b) = (-iya^a-f(b)-(-iyb^+^b-f(a)-f([a,b}) = (_i)l«H/l[a <8) 1 + 1 ® a, /(6)] _(_1)IH(I/I+M)[6®l + l®6, /(a)] - /([a, 6]) which we can rewrite as: f([a, b}) = (-l)laH/l[a®l + l®a,/(6)] _(_l)l*l(l/l+l«l)[6 ®i + i®6, /(a)] = [/(a), b ® 1 + 1 ® 6] + (-l)laH/l [a ® 1 + 1 ® a, /(i)] = (-l)WI/la • /(6) - (-l)l6ll/(a)l& • /(a). We will call the resulting formula the super cocycle condition: f([a, b}) = (-lp^a ■ f(b) - (-1)I»H/WI6 • /(a). 6.2. Super Lie bialgebras. A super Lie bialgebra is a triple (g, [• ,-],S) such that: (1) g is a Lie superalgebra with the super bracket [• , •]; (2) S : g —► g ® g is a skew-symmetric linear map such that the associated dual map 5* : g* ® g* —► g* defines a Lie superalgebra structure on g*; (3) J and [• , •] are compatible in the following sense: <J([a, b]) = [6(a), 6®H-l®6] + [a®H-l®a, 5(b)], We will denote such a super Lie bialgebra by (g, S) if the super bracket [• , •] is unambiguous. Note that the last condition is equivalent to S being a 1—cocycle on g with values in g® g, for the cohomology theory of Lie superalgebras as summarized in Subsection 6.1. Since 5* is a super bracket, S is even, and the super cocycle condition above coincides with the non-graded version. The Jacobi identity for J* is equivalent to the following coJacobi identity for S which holds for any x G g: Alt8{&®Id)-&{x) = 0. Here Alts :g®g®g—>g®g®gis defined on homogeneous basis vectors by: Alts(a ®b®c) = a®b®c+ (-l)H(lfeMcD& ®c®a+ (-i)N(lal+l6l)c ® a ® b.
A NEW LIE BIALGEBRA STRUCTURE ON sl(2, 1) 113 A (finite dimensional) super Manin triple2 is a triple (g,g+,g_) of (finite dimensional) Lie superalgebras such that: (1) q is equipped with a non-degenerate super-symmetric invariant bilinear form (•,•); (2) g+ and g_ are Lie subsuperalgebras of q and q = g+ 0 g_ as vector spaces; (3) g+ and g_ are isotropic with respect to (•,•). Since the bilinear form is non-degenerate, g+ and g_ are in fact maximal isotropic or Lagrangian subsuperalgebras. These two notions (i.e. super Lie bialgebras and super Manin triples) are related to one another in a way similar to the Lie algebra case: Proposition 1. Let (p, [, ], S) be a super Lie bialgebra. Set g+ = p and g_ = p*. Define q = g+ 0 g_. Then (g,g+,g_) is a super Manin triple. Conversely, any finite dimensional super Manin triple (g,g+,g_) gives rise to a super Lie bialgebra structure on g+. Remark 6.1. This is Proposition 1 of [A] where it was proved modulo certain calculations left to the reader. 6.3. The Drinfeld double construction. Another related construction is that of the Drinfeld double. Here we will use a direct analogue of the non-graded version, in the spirit of [GZB]. Before explicitly presenting this approach, we should also mention that other superizations of the double construction do exist. See for instance [V] for a more geometrically motivated development of the double. Let (g, [• , -]flj^g) be a finite dimensional super Lie bialgebra. Then clearly g* is also a super Lie bialgebra with the associated structures defined by: hV = (*«)*. *•• = ([•>•]«)*• Let us fix a homogeneous basis {e;} for g and define the structure constants Cfc7, E*k °f tne relevant structures on g as follows: k i,j where we use the notation: aAb = a®b— (—l)laH6l&0a for any two homogeneous elements a, b. Prom these we can determine the structure constants of g*; if we let {e*} be the homogeneous basis for g* dual to {ei}, then we have: [e*, e*]r = £ Cije*k, Sg. (ej) = £ D%e* A e*, k i,j where: (_1)I*IWZ>« i?j, ((-l)le'Ne<IC$ i?j, ID* i = j. ^ ( -2C& i = j. <% = < J>a We will be assuming finite dimensionality, as this will be sufficient for our purposes. Infinite dimensional analogues will be more technically involved; for instance one needs to take into account the topology on vector spaces. Since we do not need it here, the infinite-dimensional case will not be discussed any further.
114 GIZEM KARAALI These will follow directly from the definitions of linear duality: ([x*,y*]f,z) = (x*®y*,5i(z)) (8f(z*),x ® y) = (z*,[x,y}&) where we assume x,y,z e q and x*,y*,z* G g* are homogeneous. Clearly C^ = D™ = 0 unless e^ (and hence e|) is odd. In this setup, the opposite super Lie bialgebra structure on g* can be denned as follows, (to compare with [GZB], note that Ts(a A b) = —(a A b).): [e*>ej](0*)°*> = [e^eJV ti(9*)op(e*k) = SQ*(e*k) or equivalent ly: K '}(9*)op = (~<y * S(9*)op = V Note that we are only taking the opposite in terms of the Lie superalgebra structure. The Drinfeld double D of g will be denned as the super Lie bialgebra with the underlying graded vector space identified with g0g* = g0 (g*)op. In order to define a Lie superalgebra structure on D, we first define a non-degenerate inner product (•, •) on d by asserting super-symmetry: <s*,y> = (-l)|x*llwl <».**> and the isotropy of the subspaces g and g*: <0,fl) = <fl*,fl*> = O. This choice of notation is intentional, and is meant to agree with that for the duality. We will require invariance of this form, which in terms of the bracket on d translates to: ([x*,y]i>,z) = (x*,[y,2:]d) (s*,[yV]t>> = {[x*,y*]x»z). Then the condition that [•, -]5 restricts to [•, -]0 and [•, -](0*)°p, respectively, on g and g* yields the following description of [•, •]*> in terms of the structure constants of g and g*: k k k k The super Lie bialgebra structure on d is denned to make the natural injections g —► d and (g*)op —► d embeddings of super Lie bialgebras, and hence is given by: fa = S9 + S(9*)op = SQ + V ' With the given structures, it can be shown (see [GZB] for details) that d is a quasitriangular super Lie bialgebra, with the r—matrix: i Although the superization of such concepts as Lie bialgebras, Manin triples, Drinfeld doubles may seem straightforward, it can be shown that several unexpected situations come up during the process. We refer the interested reader to [LSh].
A NEW LIE BIALGEBRA STRUCTURE ON sl(2,1) 115 7. The Super Lie Bialgebra Structure Associated to r(f) In the rest of this note, we will concentrate on the Lie bialgebra structure on g = «s/(2,1) associated to r(f) = (/ ® l)fi, where / is the linear map introduced in Section 5 above. After explicitly describing this super Lie bialgebra structure we will compare it with the standard structure. 7.1. The cocommutator Sf. We first describe this structure in terms of a cocommutator Sf, the coboundary d(r(f)) of the r—matrix r(f). We have: r(f) = (-E22 - £33) ® (En + £33) + E12 ® £21 - Ei3 ® £31 + -£32 ® -£23 — E13 ® E\s + -£"23 ® -£23- To compute Sf we use £/(</) = d(r(f))(g) = g • r(f) = [g ® 1 + 1 ® #, r(/)], and this gives us: Sf(Eu + £33) = —£23 A E23, Sf(E22 + £33) — Eis A £?i3, Sf(E2i) = E21 A (.£11 + .Ess) — -£"23 A (Eis H- -£31)5 5f(Ei2) = E12 A (—E22 — £33) — (—^13) A (£23 + £32)5 */(£?23) = 0, Sf(Ei3)=0, Sf(Esi) = (Eis + £31) A (£11 + £33) + E21 A £235 5f(Ez2) = (E23 + £32) A (—E22 — Ess) + (—^12) A Eis, where we once again use the notation: a A b = a (g) b — (—l)laH6l& (g) a for any two homogeneous elements a, b. 7.2. Two subalgebras of g. Consider the following subspaces denned by /: Si =Im(/ — 1) = {En + Ess,E2i,E2s,Eis -f £31), #2 = Im(/) = {E22 + Ess, E12, E13, E2S + Es2) • The fact that r(f) is an r—matrix satisfying Eqn.(4.3) implies that these image subspaces are indeed Lie subsuperalgebras of g (see Lemma 4 of [Kar]). In fact it is not difficult to see that both Si are isomorphic as Lie superalgebras to a four-dimensional Lie superalgebra: 5 = %©5T; Sq = {H,x} sT= (2/1,2/2) with the following relations: [h, x] = -x, [h, 2/1] = -2/1, [s, 2/2] = 2/1, [2/1,2/2] =s, [2/2,2/2] = 2ft, (any other commutator will be equal to zero). Therefore we can write: where the direct sum is the direct sum of graded vector spaces. Note that s is solvable.
116 GIZEM KARAALI Next we compute the restriction of Sf to the S;. This is straightforward; on S\ we get: Sf(Eu + £33) = — E23 A E23, Sf(E2i) = E21 A (En + -£"33) — -^23 A (E13 + E31), 6f(E23) = 0, 5f(Ei3 + E31) = (Eis + .E^i) A (£11 + -£"33) + E21 A £23, and the restriction to S2 is given by: Sf(E22 + ^33) = Eis A £?i3, Sf(Ei2) = E12 A (—E22 — ^33) — (—^13) A (E23 + £"32)5 */(£?13)=0, Sf(E23 + -£"32) = (-£"23 + £'32) A (—E22 — Ess) + (—-E12) A £13. In fact we can see that this gives Lie bialgebra structures to the S{. Hence (Si, <J/|sJ are actually Lie subbialgebras of g. Let us now compute the Lie brackets defined on S* by 5f\si- For simplicity, we will work with the isomorphic super Lie bialgebras on s. Denote the associated cocommutators on $ by Sf, in other words, define Si and S2 so that (£1, #/1 s^) is isomorphic to the super Lie bialgebra (s,Si) and (S2,Sf\s2) is isomorphic to the super Lie bialgebra (s,S2). Clearly we get: Si(h) = -yi A 2/1, 62(h) = 2/1 A 2/1, <Ji(x) = x A ft - 2/1 A 2/2, <fe(a;) = -(a; Ah-yiA y2), Si(yi) = 0, S2(yi) = 0, Si(y2) = 2/2 Ah + xAyi, S2(y2) = -(2/2 A h + x A 2/1), and we have <Ji = —#2- In particular, we see that (S^^/ls-J is (isomorphic to) the opposite super Lie bialgebra of (Si, Sf\sx)- Recall that the Lie bracket [• , ]i on the dual s* associated to Si can uniquely be determined by the following: For any two elements a, (3 G s* and any element s G s, we have: ([a,0\1,8) = (a®!3Ms)), where (a, s) = a(s) is the pairing of s with its dual s*. For instance we have: M, Vi\i,h) = (yi ® yi, 61(h)) = - (yt ® yi, 2/1 A yi) = -2 (yj ® yj, yx ® j/i) = -2(-l)l«"H'i*l(yI(y1))2 = 2 and ([2/i,2/i]i,s) = 0 for any other basis vectors of s. Therefore we get: [yi,yi]i = 2h\ Similarly we have: ([vi.ySh.*} = (vi®V2>8i(x)) = (yi®yZ,xAh-yir\y2) = - (yi ® y^vi ® V2) = -(-l)\y^\yi(yi)y*2(y2) = l and ([2/1,2/2)1, 5) = 0 for any other basis vectors of s. Therefore we get: [2/1,2/2)1 = **•
A NEW LIE BIALGEBRA STRUCTURE ON sf(2,1) 117 Likewise, we compute the other brackets on s*. The nonzero brackets are: [h*,x*]i = -x\ [h\y*2]i = -y$, \x\vV\x =y*2, [yi,V2]i=**, b/i,yi]i = 2/i*. At this point, it is easy to notice that this is actually isomorphic to the Lie super- algebra $ itself (via the map: h*^h, x* ^ x, yl^V2, 2/2 ■—^ 2/i)- Similar computations on (£2, <J/|s2) show that the super Lie bialgebra (s, 82) is also self-dual. In particular, the Lie bracket [• , -^ on the dual s* associated to 82 is given by: [h*,x*]2 = x*, [h\y2h = yh [x*,ylh = -yS. [yl,y2]2 = -s*, [yi,yi]2 = -2ft*, and this is isomorphic to the Lie superalgebra s itself (via the map: h* i-> -ft, x* h^ x, y*^ 2/2, 2/2 |-> -yi)- We have seen earlier that (S2,6f\s2) is (isomorphic to) the opposite super Lie bialgebra of (S'l, 8f \ sx). Therefore: ^se(s*)op where the direct sum is that of graded vector spaces. Here we may assume that s is equipped with the super Lie bialgebra structure given by S\ or 82, as both are self-dual. 7.3. The Drinfeld double of s. Let d be the Drinfeld double of (s, <J2), (we can carry out the following using (s,Si) instead, our choice is in fact arbitrary). The computations in Subsection 7.2 can be used to conclude that d = g as Lie super algebras. Explicitly, the map i\ : s —> g given on the generators of s as: h i-> E22 + £33, x^E12, yi ^ ^13, y2 "-> ^23 + £32, and the map %2 : (s*)op -^ g given on the generators as: 2/J ^ -£7i3 - £73i, y£ »-► ^23, are both super Lie bialgebra homomorphisms. We also note that Im(zi) = S'2 and Im(z2) = Si. The inner product defined on d is given by: (si + ai,s2 + a2) =de/ ai(52) + (-l)|a2||si|o2(*i)- Clearly Si and S2 are both isotropic with respect to this form. We only need to consider (0:1,52) and (51,0:2) where Si G Im(2i) and O; G Im(22)- Now, (aus2) = 01 (*2), (si,o2) = (-l)|a2||si|o2(si), and this form is in fact the super trace form on g. For example we can compute: (i?23j ^32) = (^235 (^23 + ^32) — ^23) = (^23 5 (^23 + ^32)) = <*2(y2),*i(y2)) = y2(y2) = i-
118 GIZEM KARAALI (£l3, E3i) — (£13, — (—E13 — £31) — Ei3) — (E13, —( — E13 — E31)) = (ii(vi),-i2(vl)) = -(-i)MWivl(yi) = 1. We know the Lie superalgebra structure when restricted to Si and £2. From the invariance of the form we can find the mixed brackets. In other words we use: ([ai, si], s2) = (ai, [si, s2]), ([si, ai], a2) = (si, [ai, a2]), where a; G Im(i2) = Si and s$ G Im(zi) = S'2. Some more computation shows that indeed the Lie superalgebra structure on d is the usual one, in other words, we find that d = g as Lie super algebras. The super Lie bialgebra structure on d is given by (J|im(s) + #|im(s*)°p or equiv- alently by (J|im(s) + <J|im(s*)- But this is equal to S/\s1 + Sf\s2 = £/• Hence, g = 5/(2,1) with the super Lie bialgebra structure Sf is the Drinfeld double of the four dimensional solvable Lie superalgebra s. 7.4. The standard structure. It may be interesting to compare the above with the standard Lie bialgebra structure on q. Note that rs = (-E22 - E33) <g> (En + E33) + Eu <g> E2i - Ei3 <g> E31 + E32 <g> E23 is a 2—tensor satisfying Eqn.(4.3) and is in fact a standard r—matrix constructible by Belavin-Drinfeld type data (involving a trivial admissible triple, i.e. the isom- etry r involved is trivial). (4.3) can be obtained by Using 8S = d(rs) to find the associated cocommutator Ss, we see that: 5a(E11 + E33) = 0, 6S(E22 + E33)=0, Ss(E2i) = E21 A (Eu + £33) — E23 A E3i, 5s(Ei2) = E12 A (—E22 — E33) — (—Ei3) A E32, Ss(E2s) = 0, 5s(Ei3)=0, 6s(E3i) = E3iA(Eii + E33), 5S(E32) = E32 A (—E22 — E33). Restrictions of Ss to Si would not give us well-defined maps on Si. Instead, we will consider its restrictions to two other solvable subalgebras of q. Define Ti = (Eu + E33, E21, E23, E3i), T2 = (E22 + E33, E12, E13, E32). Then on Ti we get: 6s(Eu + E33) = 0, Ss(E2i) = E21 A (En + £33) — E23 A E31, 6S(E23)=0, Ss(E3i) = E3i A (Eu + E33),
A NEW LIE BIALGEBRA STRUCTURE ON sl(2,1) 119 and the restriction of Ss to T2 is given by: 6S(E22 + E33) = 0, ^(^12) = E12 A (—E22 — E33) — (—E13) A £32, 6S(E13)=0, 83(^32) — -£"32 A {—E22 — E33). It is easy to see that (Ti, Ss\Ti) is a Lie subbialgebra of (g, Ss) for each i. It is clear that Ti = t where t is the four dimensional solvable Lie superalgebra t = %etT; %=(M) %= (2/1,3/2) with the following relations: [ft, x] = -x, [ft, 2/1] = -2/1, [2/1,2/2] = a:, (any other commutator will be equal to zero). Therefore, we can write: u^tet where the direct sum is that of graded vector spaces. Let us now compute the Lie brackets denned on T* by 6s\Ti- For simplicity, we will work with the isomorphic super Lie bialgebras on t. Denote the associated cocommutators on t by Jsi; in other words, define <Jsl and Ss2 so that (Ti,5s\ti) is isomorphic to the super Lie bialgebra (t,<Jsl) and (T2,Ss\t2) 1S isomorphic to the super Lie bialgebra (t, <Js2)- Clearly we get: <W0 = O, 5s2(h) = 0, Ssl(x) =xAh-y1Ay2, Ss2(x) = -(x A ft - 2/1 A 2/2), *s 1(2/1) = 0, ^2(2/1) = 0, *« 1(2/2) = 2/2 A ft, ^2(2/2) = -2/2 A ft, and we have <Jsl = — Js2- In particular, we see that (T2,5s\t2) is (isomorphic to) the opposite super Lie bialgebra of (Ti, (Js|ti)- At this point, the Lie bracket [• ,-]i on the dual t* associated to Ssl can be easily determined. The nonzero brackets are: [ft*,x*]i = -x\ [ft*, 2/2)1 = -vh [2/1,2/2)1 = x*- Notice that this is actually isomorphic to the Lie superalgebra t itself (via the map: ft* k+ ft, x* ^ x, 2/* k+ 2/2, 2/2 ^ 2/1 )• As (T2,Js|t2) is (isomorphic to) the opposite super Lie bialgebra of (Ti,^!^), we get: where the direct sum is that of graded vector spaces, and we are considering t with the super Lie bialgebra structure given by <Jsl. Of course, we can show that (t, Ss2) is self-dual as well, and hence the above identity would still hold if we assumed that the super Lie bialgebra structure on t is the one associated to Js2-
120 GIZEM KARAALI Now arguments similar to those in Subsections 7.2 and 7.3 can be used to conclude that q is isomorphic to the double of t as a Lie superalgebra. Explicitly, the map isl : t —► q given on the generators of t as: h i-> E22 + £33, xv-*E12, 2/ii->£?i3, V2^E32, and the map is2 : (t*)op —► g given on the generators as: ^^-(En + Eas), x*^E21, 2/I«->-£?3i, y2^E23 are both super Lie bialgebra homomorphisms. We also notice that Im(2sl) = T2 and Im(2s2) — ^i- The inner product denned on g by this double structure is given by: <5i + ai,s2 + a2> =dc/ ai(52) + (-l)MMa2(Sl). Clearly T\ and T2 are both isotropic with respect to this form. We only need to consider (ai,£2) and (£1,0:2) where U G Im(zsl) and a^ G Im(zs2), but (*i,s2) = ai(52), <5i,a2> = (-l)|a2||si|a2(5l), and we see that this form also coincides with the super trace form on g. We know the super Lie algebra structure when restricted to T\ and T2. Prom the invariance of the form we can find the mixed brackets. In other words we use: ([ai,*i],*2> = (otu[tut2]), ([ti,ai],a2) = (tu [ai,a2]), where a^ G lm(is2) = T\ and U G Im(isl) = T2. Some more computation shows that indeed the Lie superalgebra structure on q is the usual one. The super Lie bialgebra structure on q coming from this double construction is given by (J|im(t) + £|im(t*)°p or equivalently by (J|im(t) + ^|im(t*)- But this is equal to Ssl^ + Ss\t2 = Ss. Therefore, g = s/(2,1) with the standard super Lie bialgebra structure is the Drinfeld double of the four dimensional solvable Lie superalgebra t (equipped with the super Lie bialgebra structure given either by <Jsl or Ss2). 8. Conclusion Unlike in the non-graded case, there are super r—matrices satisfying Eqn.(4.3) (in other words, normalized non-skew-symmetric r—matrices) which cannot be obtained via a simple modification of the Belavin-Drinfeld construction. In this note, we have studied the Lie bialgebra structure associated to one such r—matrix on si (2,1), and we have shown that it has a nice description as the double of a four dimensional subalgebra. In the non-graded case, such structures only arise from twists of the standard r—matrix. Our r—matrix is not of this form3, but shows similarities to such. These similarities may lead to an understanding of these special types of non-skew-symmetric r—matrices that do not fit a Belavin-Drinfeld type description. One way to see this would be to note that the subalgebras s and t are not isomorphic; they have clearly distinct Jordan-Holder decompositions.
A NEW LIE BIALGEBRA STRUCTURE ON sl(2,1) 121 Although our ultimate aim is a full classification result, there are still various intermediate questions to be answered, and some of them look more tractable than others. For instance we would like to explicitly determine how r—matrices obtained from nonisomorphic Dynkin diagrams are related to one another. This should be reasonably straight-forward, clearly an answer should involve Serganova's odd reflections ([Se]). There are certain cohomological concerns that need to be addressed as well, for instance the fact that Whitehead lemmas no longer hold for simple Lie super algebras; these will most likely prove much more difficult. We hope to pursue these questions in the near future. References [A] Andruskiewitsch, N.; "Lie superbialgebras and Poisson-Lie supergroups11, Abh. Math. Sem. Univ. Hamburg 63 (1993), pp.147-163. [BD1] Belavin, A. A., Drinfeld, V. G.; "Solutions of the classical Yang-Baxter equation and simple Lie algebras", Punct. Anal. Appl. 16 (1982), pp.159-180. [BD2] Belavin, A. A., Drinfeld, V. G.; "Triangle equation and simple Lie algebras", Soviet Scientific Reviews Sect. C 4 (1984), pp.93-165. [CP] Chari, V., Pressley, A.; A Guide to Quantum Groups, Cambridge University Press, 1995. [Dl] Drinfeld, V.G.; "Hamiltonian Structures on Lie Groups, Lie Bialgebras and the Geometric Meaning of the Classical Yang-Baxter Equations", Soviet Math. Dokl. 27 (1983), no. 1, pp.68-71. [D2] Drinfeld, V. G.; "Hopf Algebras and the Quantum Yang-Baxter Equation", Soviet Math. Dokl. 32 (1985), pp.254-258. [D3] Drinfeld, V. G.; "Quantum Groups", Proceedings of the International Congress of Mathematicians, Berkeley, 1987, pp.798-820. [ES] Etingof, P., Schiffmann, O.; Lectures on Quantum Groups, International Press, 1998. [FRT] Fadeev, L., Reshetikhin, N. Yu., Takhtajan, L.; "Quantization of Lie Groups and Lie Algebras", in: Algebraic Analysis, M. Kashiwara and T. Kawai eds., Academic Press, 1989, pp. 129-139. [F] Fuks, D. B.; Cohomology of Infinite-Dimensional Lie Algebras, (Contemporary Soviet Mathematics), 1986. [GZB] Gould, M. D., Zhang, R. B., Bracken, A. J.; "Lie bi-superalgebras and the graded classical Yang-Baxter equation", Rev. Math. Phys. 3 (1991), no. 2, pp.223-240. Lie no. 1, [Jl] Jimbo, M.; "Introduction to the Yang-Baxter Equation", International J. Mod. Phys. A 4 (1989), pp.3759-3777. [J2] Yang-Baxter Equation in Integrable Systems, M. Jimbo ed., Advanced Series in Math. Phys., vol. 10, World Scientific, 1990. [Kac] Kac, V. G.; "Lie Superalgebras", Advances in Mathematics, 26 (1977) pp.8-96. [Kar] Karaali, G.; "Constructing r-matrices on simple Lie superalgebras", J. Algebra 282 (2004), no.l, pp.83-102. [KR] Kulish, P. P., Reshetikhin, N. Yu.; "Quantum Linear Problem for the sine-Gordon Equation and Higher Representations", J. Soviet Math. 23 (1983), pp.2435-2441. [LSe] Leites, D., Serganova, V.; " Solutions of the classical Yang-Baxter equation for simple superalgebras", Theoret. and Math. Phys. 58 (1984), no. 1, pp.16-24. [LSh] Leites, D., Shapovalov, A.; "Manin-Olshansky triples for Lie superalgebras", J. Nonlinear Math. Phys. 7 (2000), no. 2, pp.120-125. [Sc] Scheunert, M.; The Theory of Lie Superalgebras: An Introduction, Lecture Notes in Mathematics 716, Springer-Verlag, 1979. [ScZ] Scheunert, M., Zhang, R. B.; "Cohomology of Lie superalgebras and their generalizations", J. Math. Phys. 39 (1998), pp.5024-5061. [Sh] Shchepochkina, I. M.; "Maximal Solvable Subalgebras of Lie Superalgebras gl(m\n) and sl(m\n)", Funct. Anal. Appl. 28 (1994), no. 2, pp.147-149 [STS] Semenov-Tian-Shansky, M. A.; "What is a Classical r-matrix?", Funct. Anal. Appl. 17 (1984), pp.259-272.
122 GIZEM KARAALI [Se] Serganova, V.; "Oii Generalizations of Root Systems", Comm. Algebra 24 (1996), no. 13, pp.4281-4299. [V] Voronov, T.; " Graded manifolds and Drinfeld doubles for Lie bialgebroids", In: Quantization, Poisson Brackets and Beyond, Contemp. Math. 315, Amer. Math. Soc, Providence, RI, 2002, pp. 131-168, arXiv:math.DG/0105237. [W] Weibel, C. A.; An Introduction to Homological Algebra, Cambridge University Press, 1997. Department of Mathematics, University of California, Santa Barbara, Ca 93106 E-mail address: gizemSmath.ucsb.edu
Contemporary Mathematics Volume 413, 2006 The Steinberg Tensor Product Theorem for GL(m\n) Jonathan Kujawa Dedicated to James E. Humphreys on the occasion of his 65th birthday. Abstract. We formulate and prove a version of the Steinberg Tensor Product Theorem for the supergroup GL(m\ri). 1. Introduction The Steinberg Tensor Product Theorem is a fundamental result in the modular representation theory of algebraic groups. The purpose of the present article is to formulate and prove the analogous theorem for the supergroup GL(m\n). This result was first mentioned without proof in [2]. We emphasize that our approach closely parallels the analogous result for the supergroup Q(n) proven by Brundan and Kleshchev [1], which in turn follows the approach of Cline, Parshall, and Scott [3]. Similar arguments are also used in [10] in the setting of Hopf algebras. The preliminaries are outlined in section 2. They are an abbreviated form of what can be found in [2] and [7]. Sections 3 and 4 contain the new results of the present article with the main theorem being the following version of the Steinberg Tensor Product Theorem. Before stating the result, we require some notation. We direct the reader to section 2 for precise statements of definitions. Throughout, let A; be a fixed ground field of characteristic p > 0 which is algebraically closed. All objects under discussion are denned over k. Let T be the maximal torus of GL(m\n) consisting of diagonal matrices. We identify the character group X(T) = Hom(T, Gm) with the free abelian group on generators £i,..., £m+n, where e% picks out the 2th entry of a diagonal matrix. We call the set {m+n ^ A = Y, A^ e X(T) : Ai > • • • > Am and Am+1 > > \m+n \ , the set of dominant weights. The irreducible GL(m|n)-supermodules are parameterized by highest weight by the set X+(T) and we write L(A) for the irreducible supermodule of highest weight A e X+(T). A weight is p-restricted if it is dominant 2000 Mathematics Subject Classification. 20G05. Key words and phrases. Modular representation theory, algebraic groups. Research supported in part by NSF grant DMS-0402916. ©2006 American Mathematical Society 123
124 JONATHAN KUJAWA and Xi — A^+i < p for i = 1,..., m — 1 and i = ra+l,...,ra + n — 1. Denote the set of p-restricted weights by Ar+(T). Let F : GL(m\n) —► GL(m) x GL(n) be the Frobenius morphism given by raising entries to the pth power. Given a GL(m) x GL(n)-supermodule M we can view it as a GL(m\n)-supermodule via inflation through F. We call this the Frobenius twist of M and denote by F*M. Theorem 1.1 (Steinberg Tensor Product Theorem). For A G X+(T) and /x G X+{T), L(\+pfi)^L(\)®F*L'(fi), where L'{ji) denotes the irreducible GL(m) x GL(n)-supermodule of highest weight /i. Acknowledgements. This work was done as part of the author's PhD thesis at the University of Oregon [8]. The author is grateful for the guidence and patience of his advisor, Jonathan Brundan. 2. Definitions and Basic Results In this section we outline the basic definitions and results we require. For an account of the basic language of super algebras and supergroups adopted here, we refer the reader to [1], [2], and [7]; see also [4], [5], [9, ch.I] and [11, ch.3, §§1-2, ch.4, §1]. 2.1. The supergroup GL(m\n). We use the language of supergroup schemes to define GL{m\n). Our approach parallels that of [4]. Throughout, let k be an algebraically closed field of characteristic p > 0. All objects (superalgebras, supergroups, ...) will be defined over k. A super space is a Z2-graded k- vector space. If V is a superspace and v G V is a homogeneous vector, then we write v G Z2 for the degree of v. A commutative superalgebra is a Z2-graded associative algebra A = Aq 0 A\ with ab = (—l)abba for all homogeneous a, b G A. If p = 2 we also assume that a2 = 0 for all a G A\. A morphism of superalgebras is a homomor- phism of graded algebras; that is, it is an algebra homomorphism which preserves the Z2-grading. The supergroup G = GL(m\n) is the functor from the category of commutative superalgebras to the category of groups defined on a commutative superalgebra A by letting G(A) be the group of all invertible (m + n) x (m + n) matrices of the form ' w Y X ~ Z where W is an m x m matrix with entries in Aq, X is an m x n matrix with entries in Ai, Y is an n x m with entries in A\, and Z is an n x n matrix with entries in Aq. If / : A —► B is a superalgebra homomorphism, then G(f) : G(A) —► G(B) is the group homomorphism defined by applying / to the matrix entries. Let Mat be the affine superscheme with Mat (A) consisting of all (not necessarily invertible) (m + n)x (m-\-n) matrices of the above form. For 1 < 2, j < m + n, let Tij be the morphism defined by having Tij : Mat (A) —► A map a matrix to its ij-entry. Then the coordinate ring k[Mat] is the free commutative superalgebra on the generators Tij (1 < i,j < m + n) with Tij having parity i + j, where we write i = 0 for i = 1,..., m and i = I for i = m + 1,..., m + n. By [9, 1.7.2], a
THE STEINBERG TENSOR PRODUCT THEOREM FOR GL(m\n) 125 matrix g G Mat (A) of the form (2.1) is invertible if and only if det W det Z e Ax, where here det denotes the usual matrix determinant. Hence, G is the principal open subset of Mat defined by the function det : g i—► det W det Z. In particular, the coordinate ring k[G] is the localization of k[Mat] at det. Just as for group schemes [4, 1.2.3], the coordinate ring k[G] has the naturally induced structure of a Hopf superalgebra. Explicitly, the comultiplication and counit are the unique superalgebra maps satisfying ra+n (2-2) A(3ij)= Y,Tith®Thtj, (2.3) e(Tij) = 6ij for all 1 < 2, j < m + n. By definition a representation of G means a natural transformation p : G —► GL(M) for some vector superspace M, where GL(M) is the supergroup with GL(M)(A) being equal to the group of all even automorphisms of the A-supermodule M0-A, for each commutative superalgebra A. Equivalently, as with group schemes [4, 1.2.8], M is a right fc[G]-cosupermodule. That is, there is a Z2-grading preserving structure map rj : M —► M 0 k[G] satisfying the usual comodule axioms. We will usually refer to such an M as a G-supermodule. If p : G —> GL(M) and p' : G —> GL{M') are two representations of G, then a morphism of representations is a linear map / : M —> M' such that for any commutative superalgebra A we have pf(g)(f(m)) = f(p(g)(m)) for all g G G(A) and all m G M 0 A. In the language of fc[G]-cosupermodules, if rj : M —> M 0 fc[G] and r/ : M' —► M' 0 fc[G] are the cosupermodule structure maps, then / : M —> M' is a morphism if / 0 1 o 77 = 7/ o /. We denote by G-moD the category of all G-supermodules. We emphasize that we allow all morphisms and not just graded (i.e. even) morphisms. However, note that for superspaces M and M' the space Horn*; (M,Mf) is naturally Z2-graded by declaring / G Homfc(M,Mf)r if f(Ms) C Mfs+r for all 5 G Z2. This gives a Z2-grading on YiomG(M,M') C Homfc(M,Mr). We remark that G-moD is not an abelian category. However, the underlying even category of G-moU, consisting of the same objects as G-moU but only the even morphisms, is an abelian category. This, along with the parity change functor n, which, roughly speaking, interchanges the Z2-grading of a supermodule, allows one to make use of the tools of homological algebra. The underlying purely even group Gev of G is by definition the functor from superalgebras to groups given by Gev(A) = G(Aq). Thus, Gev(A) consists of all invertible matrices of the form (2.1) with X = Y = 0, so Gev — GL(m) x GL(n). Let T be the usual maximal torus of Gev consisting of diagonal matrices. The character group X(T) = Hom(T, Gm) as defined in [4, 1.2.4] can then be identified with the free abelian group on generators £i,... ,£m+n, where ei is the function which picks out the ith diagonal entry of a diagonal matrix. Let B denote the subgroup of G given by letting B(A) equal the set of all of all upper triangular invertible matrices of the form (2.1). We call this the standard Borel subgroup. Note that the underlying purely even subgroup, Bev, is given by the upper triangular matrices in Gev- The root system of G is the set $ = {£; — e3■ : 1 < i, j < m + n, i ^ j}. There are even and odd roots, the parity of the root ei — Sj being i + j. Our choice of
126 JONATHAN KUJAWA Borel subgroup, B, defines a set, (2.4) $+ = {ei - 6j : 1 < i < j < m + n}, of positive roots. The simple roots then are e% — 6i+i where i = l,...,m + n — 1. The corresponding dominance order on X(T) is denoted <, defined by A < \i if H — A can be written as the sum of positive roots. 2.2. The Superalgebra of Distributions. Just as for algebraic groups [4, 1.7.7] one can abstractly define the superalgebra of distributions Dist(G) of G. We sketch how this is done. Let X be the kernel of the counit e : k[G] —► fc, a superideal oik[G\. For r > 0, let Distr(G) = {xe k[G]* : x(Jr+1) = 0} ^ (fc[G]/Jr+1)*, Dist(G) = (J Distr(G). r>0 There is a multiplication on k[G]* dual to the comultiplication on k[G\, defined by (xv)(f) — (x®y)(A(f)) for x,t/G fc[G]* and / G fc[G]. Note here (and elsewhere) we are implicitly using the superalgebra rule of signs: (x<g>y)(f ® g) = (—l)^^x(f)y(g) where y and / are assumed to be homogeneous. The general case is obtained via linearity. In fact, Dist(G) is a subsuperalgebra of k[G\* (see [2]). In the case when G = GL(m\n), however, we can describe Dist(G) explicitly as the reduction modulo p of the universal enveloping superalgebra of the Lie superalgebra g[(m|n, C). We now describe how this can be done. Recall that g[(ra|n,C) is the Lie superalgebra given by letting g[(ra|n,C) be the set of all (m + n) x (m + n) matricies over C. If for 1 <i,j < m + n we write eij for the ij matrix unit, then the aj provide a homogeous basis with the degree of eij defined to be i + j. The bracket is given by (2.5) . [eij,ekii] = S^e^ - (-if+W^S^j By the PBW theorem for Lie superalgebras (see [5]) we have that the universal enveloping superalgebra of g[(ra|n,C), C/c, nas basis consisting of all monomials n ^ n <? l^ij^rn+n l<ij<.m-{-n z+J=0 i+j=l where a;j e Z>o, dij G {0,1}, and the product is taken in any fixed order. We shall write hi = e^i for short. Define the Kostant Z-form Uz to be the Z-subalgebra of Uc generated by elements eij (1 < i,j < m + n,i + j = I), efj (1 < 2,j < m + n,i =fi j,i + j = 0,r > 1), and (h;) (1 < i < m + n,r > 1). Here, e\rJ := e^/(r!) and (h;) := hi(hi — 1) • • • (hi — r + 1)/(H). Following the proof of [12, Th.2], one verifies the following lemma. Lemma 2.1. The superalgebra Uz is a free Z-module with basis given by the set of all monomials of the form n #j) n C;) n ♦ l^J^m+n l<i<m+n ^ ^ l<i,j<m+n *^',*+J=0 1+3 = 1 for all aij,ri G Z>o and dij G {0,1}, where the product is taken in any fixed order.
THE STEINBERG TENSOR PRODUCT THEOREM FOR GL(m\n) 127 The enveloping superalgebra U<c is a Hopf superalgebra in a canonical way, hence Uz is a Hopf superalgebra over Z. Consequently, k <g>z Uz is naturally a Hopf superalgebra over k. It is known, for example by [2, Thm. 3.2], that Dist(G) ^k®zUz as Hopf superalgebras. We identify these Hopf super algebras and will abuse notation by using the same symbols e\j, (^), etc. for the canonical images of these elements of Uz in Dist(G). Note that the monomials given in Lemma 2.1 form a homogeneous basis of Dist(G). It is also easy to describe the superalgebras of distributions of our various natural subgroups of G as subalgebras of Dist(G). For example, Dist(T) is the subalgebra generated by all (^) (1 < i < m + n, r > 1), Dist(Bev) is the subalgebra generated by Dist(T) and all e\j (1 < i,j < m + n,i < j,i + j = 0,r > 1), and Dist(B) is the subalgebra generated by Dist(Bev) and all e^j (1 < i,j < m + n,i + j = l,i < j). Let us describe the category of Dist(G)-supermodules. The objects are all left Dist(G)-modules which are Z2-graded: that is, fc-superspaces, M, satisfying Dist(G)rMs C Mr+S for r,s e Z2. A morphism of Dist(G)-supermodules is a linear map / : M —> M' satisfying f(xm) = (—l)fxxf(m) for all m e M and all x G Dist(G). Note that this definition makes sense as stated only for homogeneous elements; it should be interpreted via linearity in the general case. We emphasize that morphisms are not necessarily even. However, the Horn-spaces are naturally Z2-graded and our remarks about the category G-mod made in the previous subsection apply here as well. For A = X^Htn ^iei e X{T) and a Dist(G)-supermodule M, define the X-weight space of M to be (2.6) Mx = lmeM : (hi\rn= ()m for all 1 < i < m + n,r > l|. We call a Dist(G)-supermodule M integrable if it is locally finite over Dist(G) and satisfies M = Yl\ex(T) ^* If M is a G-supermodule then we can view M as a Dist(G)-supermodule as follows. Given a G-supermodule M with structure map 77: M —> M <S> k[G], we can view M as a Dist(G)-supermodule by xm = (l(§)x o rj)(m). In fact, in this way we obtain a functor from G-mod to the category of Dist(G)-supermodules. Moreover, the notion of weight space denned above for Dist(G)-supermodules coincides with the usual notion of weight space of M with respect to the torus T. It is then straightforward to verify that the G-supermodule M is integrable when viewed as a Dist(G)-supermodule. We prove the following theorem in [2, Corollary 3.5]. Theorem 2.2. The category G-moX) is isomorphic to the full subcategory of integrable Dist(G)-supermodules via the aforementioned functor. In view of this result, we will not distinguish between G-supermodules and integrable Dist(G)-supermodules in what follows. 2.3. Classification of irreducible GL(ra|n)-supermodules. Now we describe the classification of the irreducible representations of G by their highest weights. It seems to be more convenient to work first in the category Op : the full subcategory of all Dist(G)-supermodules M such that M = (&\eX(T) ^a and
128 JONATHAN KUJAWA M is locally finite over Dist(J3). This is an analogue of Bernstein, Gelfand and Gelfand's category O in classical Lie theory. We remark that Theorem 2.2 implies that G-mod can be viewed as a full subcategory of Op. Prom now on we will assume all Dist (G)-supermodules under discussion are objects in Op. For A G X(T), we have the Verma supermodule M(X) := Dist(G) ®Dist(B) *a, where k\ denotes k viewed as a Dist(B)-supermodule of weight A concentrated in degree 0. Note that by Lemma 2.1 it follows that M(A) is an object in Op. We say that a homogeneous vector v in a Dist (G)-supermodule M is a primitive vector of weight A if Dist(B)v = k\ as a Dist(B)-supermodule. Familiar arguments show that M(A) is universal among all supermodules of Op which are generated by a primitive vector of weight A and M(A) has a unique maximal subsupermodule, hence an irreducible quotient which we denote by L(A). Taken together these imply that {L(A) : A G X(T)} gives a complete set of pairwise non-isomorphic irreducibles in Op. In this way, we get a parametrization of the irreducible objects in Op by their highest weights with respect to the ordering <. Now we pass from Op to G-moX). Recall that {ra+n A= Y,Xi6ieX(T) : A!>...>Am,Am+1>..->Am+: 2=1 denotes the set of all dominant integral weights. The proof of the following lemma is due to Kac [6] (see also [2]). Lemma 2.3. Given any A G X(T), L(A) is finite dimensional if and only if A G X+(T). In particular, the supermodules {L(X)}\eX+(T) form a complete set of pairwise non-isomorphic irreducible supermodules in G-raod. 3. Probenius Kernels For r > 1, we define the Frobenius morphism Fr : G —> Gev by having Fr : G(A) —> Gev(A) raise each matrix entry to the prth power for any commutative superalgebra A. Note that for a G Aj, ap = 0 so the morphism makes sense. Let Gr denote the kernel of Fr, the rth Frobenius kernel, a normal subgroup of G. Similarly, let Gev,r denote the kernel of Fr|<3ev, Br denote the kernel of Ft\b, etc. Lemma 3.1. Fr : G —> Gev is a quotient of G by Gr in the category of super- schemes. That is, for any morphism f : G —> S of superschemes which is constant on Gr(A)-cosets of G(A) for all commutative superalgebras A, there is a unique morphism f : Gev —> S such that f = f o Fr. Proof. Let -k : G —> Gev be the superscheme morphism denned by projection. That is, if g G G(A) is as in (2.1), then n acts as the identity on the entries of W and Z, and sends the entries of X and Y to zero. Let / : G —> S be a morphism of superschemes which is constant on Gr(>l)-cosets. For any element g G G(A) written as in (2.1) we have / im xz-^y1 (w x\ _ (w o\ \YW~l In J \Y Z) \0 ZJ> where Ik denotes the k x k identity matrix. That is, hg = n(g) for some h € Gr{A). Thus / = /|Gev ° ""• However from the purely even theory (see [4, 1.9.5]), Fr\cev
THE STEINBERG TENSOR PRODUCT THEOREM FOR GL(m\n) 129 is a quotient of Gev by Gev,r- Consequently, since /|g6V ^s constant on Gev,r-cosets of Gev, there is a unique morphism / : Gev —► S such that /|g6V — / ° Fr\Gev- Therefore / = f\Gev on = fo Fr\Gev on = foFr. D Observe that k[Gr] = k[G]/Ir where Ir is the ideal generated by {Tf-, T%k — 1 : 1 < i,j, k < m+n, i ^ j}. Consequently, a basis for k[Gr] is given by the monomials inT^jJ for 1 < i,j < ra+n, where aij G {0,1,... ,pr—1} if i+j = 0 and aij G {0,1} if i + j = 1, with the product taken in any fixed order. In particular, the dimension of k[Gr] is finite so by definition Gr is a finite algebraic supergroup. Moreover the pr-th power of any element of X := Ker(e : k[Gr] —> k) lies in Ir so X is nilpotent. That is, Gr is infinitesimal and, consequently, Dist(G>) can be identified with the Hopf superalgebra dual of k[Gr]. It follows as in [4,1.8.1-6] that the category of Gr- supermodules is isomorphic to the category of Dist(G>)-supermodules. Also, under this identification we can take as our basis for Dist(G>) C Dist(G) the ordered PBW monomials p-i) n 4?" n m n >;,0 TT fhk\ TT e(a*.i) l<ij<m+n l<k<m+n ^ ' l<ij<m+n i<j i<j where a,ij,dk G {0,... ,pr —1} for 1 < z,j, k < m+n when i+j = 0, and a*j G {0,1} when i + j = T. Similarly we can describe bases for Dist(Br), etc. Prom this we observe the following lemma. Lemma 3.2. Dist(G>) is a free right Dist(Br)-supermodule with basis given by the ordered monomials n l<i,j<m-{-n i<j where aij G {0,... ,pr — 1} when i + j = 0, and aij G {0,1} when i + j = T. Having identified the representations of Gr and Br with Dist(G>)-supermodules and Dist(i?r)-supermodules, respectively, we have the induction functor given by ind^ M = Dist(Gr) (8)Dist(Br) M- Prom Lemma 3.2 we see this is an exact functor which is left adjoint to restriction. Given A G X(T), let k\ denote k viewed as a Tr-supermodule of weight A concentrated in degree 0. The classical theory [4, II.3.7] gives the following lemma. Lemma 3.3. The set {k\ : A G X(T)} is a complete family of irreducible Tr- supermodules. Moreover, k\ = k^ if and only if A — \i G prX(T). Furthermore, {k\ : A G X(T)} provides a complete set of irreducible Br- supermodules via inflation. For A G X(T), define Zr(X) = ind^ /cA. Let Lr(A) denote the Gr-head of Zr(X). Proposition 3.4. {Lr(A) : A g X(T)} is a complete set of irreducible Gr- supermodules. Furthermore, Lr(X) = Lr(fi) if and only if A — /x G prX(T).
130 JONATHAN KUJAWA Proof. Let U~ denote the unipotent radical of the lower Borel; that is, U~(A) is the subgroup of G(A) given by lower triangular matricies with ones along the diagional. Then by definition U~ is the kernel of Fr restricted to U~. Observe that by Lemma 3.2 we have that Zr(X) = Dist(U~) as £7,7-supermodules. Consequently we have dim^ Hom^- (Zr(A), k) = dim^ Hom^- (Dist(f7rT), k) = 1. Thus Zr(X) has an irreducible f/~-head and it then follows that Zr(X) has an irreducible G>-head. That is, Lr(A) is irreducible. Now if L is an irreducible Gr-supermodule then we can choose A e X(T) so that Hom£r (k\, L) 7^ 0. By Probenius reciprocity L is isomorphic to a quotient of Zr(\), hence L = Lr(A). Finally, from the classification of the irreducible supermodules of Br we see that Lr(A) = Lr(/x) if and only if A - \i e prX(T). D 4. The Steinberg Tensor Product Theorem We are now able to prove the Steinberg Tensor Product Theorem for GL(m\n). Lemma 4.1. Let L be an irreducible G-supermodule. Then L is completely reducible as a G\-supermodule. Proof. Let Li be an irreducible supermodule in the Gi-socle of L. Since G\ is a normal subgroup of G each translate, gL\, by an element g G G(k) is an irreducible Gi-subsupermodule of L. Thus M:= ]T gLx geG(k) is a completely reducible Gi-subsupermodule of L. It suffices, then, to prove M = L. Since L is irreducible it suffices to to show M is Dist(G)-stable. Clearly M is G(fc)-stable. Since G(k) = Gev(k) is dense in Gev, M is necessarily a Gev- supermodule by [4, 1.6.16,1.2.12(5)]. That is, M is Dist(Gi) and Dist(Gev)-stable. However Dist(G) is generated by Dist(Gi) and Dist(Gev), so M is Dist(G)-stable. □ Lemma 4.2. Let A e X+(T). Then Dist(Gi)L(A)A is a Gi-subsupermodule of L(X) isomorphic to Li(A). Proof. As a B-supermodule L(\)\ = k\, so they are isomorphic as B\- supermodules as well. Thus there is a Bi-supermodule homomorphism k\ —> L(A) with image L{\)\. By Probenius reciprocity we have a nonzero G\-supermodule homomorphism Z\(A) —> L(A) with image Dist(Gi)L(A)A. By Lemma 4.1 Dist(Gi)L(A)A is completely reducible as a Gi-supermodule while Z\(X) has irreducible Gi(A)- head, Li(A). Consequently Dist(Gi)L(A)A is an irreducible G\-supermodule isomorphic to Li(A). D Recall that we say a weight A G X+(T) is p-restricted if it is dominant and A^ — Ai+i < p for i = 1,..., m — 1 and z = ra+l,...,ra + n— 1 and that we denote the set of p-restricted weights by X+(T). Lemma 4.3. For A e X+{T), the irreducible G-supermodule L(X) is irreducible as a Gi -supermodule and L(A) = Li(A) as G\ -supermodules.
THE STEINBERG TENSOR PRODUCT THEOREM FOR GL(m\n) 131 Proof. Throughout the proof we write ea for eij where e% — £j = a is a root. Given a monomial of the PBW basis eii • • • eafc for roots ai,..., a^, we define the total degree of the monomial to be the nonnegative integer s\ -\ + Sk> Let M = Dist(Gi)L(A)A. By Lemma 4.2 M is isomorphic to Li(A). Consequently it suffices to show M = L(X). We do this by showing M is Dist(G) invariant—hence equal to L(A) by irreducibility. First we make several reductions. Note that Dist(G) is generated by Dist(Gi) and Dist(Gev) and M is clearly Dist(Gi)-stable, so it suffices to check that it is Dist(Gev)-stable. Since Dist(Gev) is generated by Dist(J5ev) and A := < e_a : r G Z>o, ot an even simple root >, it suffices to show M is invariant under the action of Dist(Bev) and the elements of A. However, Bew normalizes G\ and L(\)\ is a .Bev-subsupermodule of L(A) so M is Dist(i?ev)-stable. Therefore we have reduced the problem to proving that M is invariant under the action of the elements of A. Fix 0 7^ v\ G L(\)\. By Lemma 3.2 M is spanned by vectors of the form Xv\ where X is a monomial in the e_Vs for /3 a positive root and s G {0,1,... ,p — 1} if /3 is even and s G {0,1} if /3 is odd. Consequently, it suffices to prove e_aXv\ G M for e_a G A and such monomials X. We prove this by inducting on the total degree of e^X. The base case when the total degree is zero is immediate. Now assume the total degree of e_aX is greater than zero. If the total degree of X is zero, then we have e_av\. If r < p then e_a G Dist(Gi) by (3.1) and the result is immediate. Now say a = Si — €i+i and say r > Xi — A^+i, then e_av\ = 0 by SL(2) theory. Since A G X+(T), our two cases cover all possibilities. Thus the result always holds. Now assume the total degree of X is greater than zero. We can then write X = e_lY where /3 is a positive root and s G {1,... ,p — 1} if /3 is even and s = 1 if /3 is odd, and Y is a monomial of total degree strictly less than the total degree of X. If a + /3 is not a root, then e_ae_p = e_pe_a and the result holds by induction. If a + /3 is a root, then using (2.5) we have e(r) (s) _V-_ (b) (c) (d) where the sum is over all b,c,d G Z>o with ra + 5/3 = bf3 + ca + d(a + /3) for some integral coefficients a^,c,d (cf. [12, Lemma 8]). Observe that b + d = s so s>b, d which implies e_L e_}.^ G Dist(Gi). Also observe that c + d = r < r + s so by the inductive assumption e_ae_}a+^Yv\ G M. Therefore all terms of the sum lie in M, proving the desired result. □ Given a Gev-supermodule, M, we can inflate M to a G-supermodule through the Frobenius morphism F = F1 : G -^ Gev. We denote the resulting G-supermodule by F*M and call it the Frobenius twist of M. This defines a functor from the category of Gev-supermodules to the category of G-supermodules. For example, if we let Lev(n) be the irreducible Gev-supermodule of highest weight /x, which is simply the irreducible Gev-module viewed as a supermodule concentrated in degree 0, we have the G-supermodule F*Lev{ji). Conversely, if N is a G-supermodule, then there is
132 JONATHAN KUJAWA an induced Gev structure on the fixed point space NGl. Namely, the representation G —> GL(NGl) is constant on GL-cosets so factors through to give a representation Gev —> GL(NGl) by Lemma 3.1. Therefore by taking Gi-fixed points we have a functor from G-supermodules to Gev-supermodules which is right adjoint to F*. We are now prepared to prove the main result. Theorem 4.4. For X e X+(T) and \i e X+(T), L(X + pfi) ^ L(X) ® F*Lev(/x), where Lev(n) denotes the irreducible Gev-supermodule of highest weight /jl. Proof. For A e X+(T), L(X) is irreducible as a Gi-supermodule by Lemma 4.3. By Lemma 4.2 and Proposition 3.4 we know H := HomGl(L(A),L(A+p/i))o ± 0. We view H as a G-supermodule by conjugation: the action of u G Dist(G) is given by (uf)(x) = J2i Uif{a{vi)x) for / G H and x G L(A), where A(u) = £\u>i®Vi and A and a are the comultiplication and antipode of Dist(G), respectively. Checking directly one can verify that the map H <S> L(X) —► L(X + p/x) given by / 0 x \-> f(x) is an even G-supermodule homomorphism. Since H is nonzero, the map must be nonzero hence, by the irreducibility of L(X + p/x), surjective. On the other hand by the complete reducibility of L(X + p/x) by Lemma 4.1 and the super version of Schur's Lemma, dimfc (H <g> L(A)) = dim* (HomGl (L(A), L(X + p/jl))q (g) L(A)) < (dimfc L(X + p/jl)/ dim/c L(X)) • dim^ L(A) = dimfcL(A + p/x), so our map must be an isomorphism. Finally, since the action G\ on H is trivial, we have H = F*M for some Gev-supermodule M. Since L(X +p/x) is irreducible, M must be irreducible. Since H has highest weight p/x, M = Lev(/i). □ References [1] J. Brundan and A. Kleshchev, Modular representations of the supergroup Q(n), I, J. Algebra 260 (2003). [2] J. Brundan and J. Kujawa, A new proof of the Mullineux conjecture, J. Alg. Comb. 18 (2003). [3] E. Cline, B. Parshall, and L. Scott, On the tensor product theorem for algebraic groups, J. Algebra, 63 (1980). [4] J. C. Jantzen, Representations of algebraic groups, second edition, Math. Surveys 107, American Math. Soc, 2003. [5] V. Kac, Lie superalgebras, Advances Math. 26 (1977), 8-96. [6] V. Kac, Representations of classical Lie superalgebras, in: Lecture Notes in Mathematics, vol. 676, 597-626, Springer, 1978. [7] J. Kujawa, Crystal structures arising from representations of GL(m\n), Rep. Theory, (2006). [8] J. Kujawa, The representation theory of the supergroup GL(m\n), University of Oregon, 2003. [9] D.A. Leites, Introduction to the theory of supermanifolds, Russian Math. Surveys 35 (1980). [10] Z. Lin, Induced representations of Hopf algebras: Applications to quantum groups at roots of 1, J. Algebra, 154 (1993). [11] Yu I. Manin, Gauge field theory and complex geometry, Grundlehren der mathematischen Wissenschaften 289, second edition, Springer, 1997. [12] R. Steinberg, Lectures on Chevalley groups, Yale University, 1967. Dept. of Mathematics, University of Georgia, Athens, GA 30602 E-mail address: kujawaQmath.uga.edu
Contemporary Mathematics Volume 413, 2006 Cyclotomic q-Schur algebras and Schur-Weyl duality Zongzhu Lin and Hebing Rui Dedicated to James E. Humphreys on the occasion of his 65th birthday 1. Introduction The representation theory of Hecke algebras plays an important part towards understanding (ordinary or modular) representation theories of finite groups of Lie type. Schur algebras, as endomorphism algebras, connect the representation theory of general linear groups and the representations of symmetric groups via Schur-Weyl duality. The quantum version of Schur-Weyl duality was established by Jimbo [J]. Since then, the representation theories of Hecke algebras and g-Schur algebras have played important roles in relating the representation theories of quantum groups and finite groups. Another important feature of Hecke algebras is their role in decomposing representations of algebraic groups in positive characteristics and quantum groups at roots of unity, via Lusztig's conjectures. In [VV], Varagnolo and Vasserot reformulated Lusztig's conjecture about decomposing the Weyl modules for quantum groups at roots of unity using the geometric description of the representations of affine Hecke algebras and affine g-Schur algebras. They proved that the Lusztig conjecture is equivalent to the equality certain canonical bases in the Fock spaces constructed using representations of Hecke algebras. This formulation enables Schiffmann [Schl] to give a different proof of the Lusztig conjecture for quantum sin at roots of unity. However this celebrating approach has been so far limited to type A as the symmetric groups appear as the Weyl groups on one hand and the permutation group acting on the tensor factors on the other hand. In this paper, we study the relation between the cyclotomic Hecke algebras and the corresponding q-Schur algebras in searching for "right" tensor spaces in this case so that the corresponding representations should reflect the representations of quantum groups. The representation theory of cyclotomic Hecke algebras i7m,r has been studied (see, e.g. [AK, DJM, DR1, DR2, GL], etc.) mostly along the line of representations of symmetric groups. One of the important features of g-Schur algebras of type 2000 Mathematics Subject Classification. Primary 20G; Secondary 17B50. Research of the first author was supported in part by NSF grant DMS-0200673. Research of the second author was supported in part by a grant of NSFC and a Foundation of Minister of Education in China. ©2006 American Mathematical Society 133
134 ZONGZHU LIN AND HEBING RUI A is that they are quasi-hereditary [CPS1] and have many important applications as studied by Cline-Parshall-Scott and many others. Dipper, James, and Mathas [DJM, Ml] have denned a general version of cyclotomic g-Schur algebra and studied those which are quasi-hereditary. We are interested in fitting the representations of Hecke algebras into the picture of [VV] and in making connections with representations of affine quantum groups. In this paper, we define a different version of cyclotomic g-Schur algebra Sm(n,r) in Section 5. Although it may not be quasi- hereditary, it is a finite dimensional quotient of the affine quantum group Uq(g[n). The definition depends on choosing a suitable tensor space on which both Uq(g[n) and i/m?r act and the two actions commute to each other. The paper is organized as follows. We discuss the multi-compositions and the standard setting in Section 2. In Section 3, we follow the setup of [DJM], and many others to discuss several cellular bases of i/m,r and other lemmas which will be needed later on. In Section 4 we discuss quasi-hereditary cyclotomic g-Schur algebras in the setting of [DJM] corresponding to each saturated set I\ They are all quasi-hereditary and then we prove a double centralizer property over certain commutative rings. This generalizes [DPS1, 6.2]. When ujr is in V (then Hm^r acts faithfully on the "tensor space"), the double centralizer property was proved by Mathas in [M2]. In our proof, we had to appeal to fact that the cyclotomic g-Schur algebras satisfy the base change property and then we can follow the argument of [DPS1]. In Section 5, we construct a cyclotomic g-tensor space and define a special cyclotomic g-Schur algebra Sm(n,r), which contains a usual g-Schur algebra as a subalgebra and at the same time is a quotient of the affine g-Schur algebra. Using results of Ginzburg-Vasserot, Lusztig, and Varagnolo-Vasserot in [GV, Lu, VV], we establish a quantum Schur-Weyl reciprocity between Uq(gln) and the semi-simple cyclotomic Hecke algebras. The result of the paper was presented at the Workshop on Finite Dimensional Algebras, Algebraic Groups and Lie Theory at Fields Institute in 2002. Since then there have been new results on double centralizer properties of cyclotomic Hecke algebras with quantum groups Uq(gin) by Hu and Stoll [HS] as well as by Sawada and Shoji [SS]. In [HS], based on Ariki's construction of the tensor space V®r with a graded structure on V in [A3], they proved a double centralizer theorem of a special type of cyclotomic Hecke algebra with the quantum group U(gi). Our construction here uses a quotient of the affine tensor spaces and establishes the double centralizer property of a special cyclotomic Hecke algebra with the affine quantum group U(gln), at least in the generic situation. Thus the special cyclotomic g-Schur algebra Sm(n, r) is a quotient of the quantum group U(gln) and its the representations are finite dimensional representations of U(gin). This establishes the relations between the cyclotomic g-Schur algebra and affine quantum groups in the same way as the classical Schur algebras characterize the polynomial representations of GLn. Acknowledgement. The research was conducted during the second author's visits to Kansas State University during Spring of 2002 and Spring of 2003 and he wishes to thank the host department for support. 2. Young tableaux and symmetric groups 2.1. Suppose that r is a positive integer. A composition A of r is a sequence of nonnegative integers (Ai, A2, • • • , An, • • •) with |A| = ]T\ A^ = r. If the sequence is weakly decreasing, A is called a partition. Let A(r) (resp. A+(r)) be the set of
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 135 all compositions (resp. partitions) of r. The above definitions also makes sense for r = 0 as we will need this special case in the following definition. Let m be a positive integer. An m-composition A of r is an m-tuple of compositions (A*1*, • • • , A<m)) such that J2?=i |A(i)| = r, where A« = (a[°, a£\ • • • ,) is composition of |AW|. If each AW is a partition, then A is called an m-partition of r. We denote by Am(r) (resp. A+(r)) the set of all m-compositions (resp. m-partitions) of r. It is known that A(r) is a poset with dominance order < such that A < /x for A, /x e A(r) if X^7=i ^' ^ 1^=1 A^' f°r a^ * • The dominance order on Ai(r) = A(r) extends to a partial order < on Am(r) as follows. For two compositions A = (AW,--. ,A(m)) with AW = (A^,A^,... ,) and /x = (/x^, • • • ,/x<m)) with ^(2) = (/4 , /4 , • • • ,), define A < /x if, for any 1 < k < m and /, Ei^i + E^^Ei^i + E^- Let A(m, r) = {(ai,..., am) e Nm | Yliai = r}- I*1 this paper, N is the set of all nonnegative integers. Each element in A(ra, r) can be thought as a composition of r with ith components being zero for all i > m. Similarly, A+ (m, r) is the set of all partitions with at most m-parts. Each m-composition A = (A^,..., A^m^) defines a composition [A] = flA^I,..., |A^|) e A(m, r). It follows from the definition that A < /x implies [A] < [/x]. For a given m-composition A = (X^\..., A^m^), let A (A) be the set of all m-compositions obtained by permuting the entries within each component AW. In A(A) there is a unique maximal m-composition A+ under the dominance order. A+ is the unique m-partition in the set A(A). 2.2. A Young diagram Y(X) of a composition A of r is the set {(z, j)| i > 1, 1 < j < Ai} in N2 (rotated 90 degrees clockwise). We will think of the Young diagram of A as r boxes placed at the matrix entries {(i, j)| 1 < j < A2}. Thus we can talk of rows and columns of a Young diagram. If A2 = 0, then there is no box in the 2th row. Thus each box is determined by its coordinates (i,j). A A-tableau s is a bijective map from the set of boxes in F(A) to {1,2,..., r}. Thus a A-tableau can be thought as colored Young diagram Y(X) by placing the integers 1, 2, • • • , r into the boxes of the Young diagram without repetition. A A-tableau s is called row- standard if the entries in s are increasing from left to right in each row. When A is a partition, a A-tableau s is said to be standard if s is row-standard and the entries of s are increasing downward in each column. The Young diagram Y(X) of an m- composition A is an m-tuple of Young diagrams (Y(X^), Y(X^), • • • , F(A^m^)). A A-tableau t is a bijective map from the set of all boxes in Y(X) to {1,2,..., r}. Thus t(fc, i,j) is the number at the (i,j)-entry of the k-th component with 1 < j < X\ \ When the A-tableau t is given, for 1 < i < r, we write (comp(2'),row(2),col(2)) for the box that contains 2 in t. A A-tableau t has m components ti,... ,tm. Note that each t^ is a A^-tableau with entries in a subset of {1,2, ...,r} instead of {1,..., |A^^|}. However, the meaning of t^ being row-standard or standard will be clear. If each t^ is row-standard, we call t row-standard. If A is an m-partition, t is said to be standard if each component is standard. Let TS(X) be the set of all standard A-tableaux. We emphasize that standard tableaux are denned for m-partitions A only.
136 ZONGZHU LIN AND HEBING RUI 2.3. Given a composition A, the dual composition A' is denned by A^ = 10 | <\? > *}l- Note that A' is necessarily a partition. If A is a partition, then y(A') is matrix transpose of Y(X). In particular, for each nonnegative integer /, we have £a; = $>-/). i>l j=l If A = (\(x\..., A(m)) is an m-composition, its dual m-composition is A' = ((A^m^);,..., (A^)'). Note also that A' is necessarily an m-partition of r. It follows from the definition that (A')' = A if and only if A is an m-partition. It is well known that for two partitions A, /x of r, A <! /x if and only if A' > /x'. The following Lemma is a generalization of this fact to m-partitions. Lemma 2.4. For two m-partitions A and /x of r, A < /x if and only if A' > /x'. Proof. We need only to prove A > /x implies /x' > A' since A" = A for any m-partition A. By the definition of dual m-partition, (A')^ = (A^™-*^1))'. Thus for any given pair (fc, /) we have fc — 1 I m—k E i(A')(i)i+E(A')5fc)=»•-[£ iA(i)i+E(A(ro~fe+1))i] 2=1 7=1 2=1 j>l (2-4.1) , fc+n , V ' m-k (A(m-fc+i)); = r-E|A«|+ E (Af-^-O]- 2=1 j = l If A' ^ /x', there exist integers fc > 0 and / > 0 such that (2.4.2) e k*')wi+E(A')f > E i(^')(i)i+X>o?>• 2 = 1 J = l 2=1 j = l We can choose fc and / such that (2.4.3) e ka')wi+E(A')f) < E imwi+i>')?'\ 2=1 J=l 2=1 j = l for any pairs (fc',/') with either fc' < fc or fc' = fc but /' < /. In particular (A(m-fc+i))/ = (A/)p) > ^/j(fc) = (^(m-fc+i))/# By (2.4.1) and (2.4.2), we have m-fc (M(m_ fe+1)M r-[£iA(2)i+ E (*Jm-fc+1) - oi »=i j=i m-k (AC"-**1))} >--[EiaWi+ E (ASm_fc+1)-o] 2=1 j = l m-fc (M(m_ fe+1))! >r-[j>wi + E (^"^-o] 2=1 j = l which implies m-fc (M(m-fe+1))! m-fc (M(m-fe+1))! E iA(i)i + E A5m"/C+1) < E i/^i + E /4m_;c+1) 2=1 j = l 2=1 .7=1
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 137 contradicting A > \i. D 2.5. Let &r be the symmetric group on the set {1,2, ...,r}. For a given m-composition A, the group &r acts on the set of all A-tableaux (from right) by permuting its entries. Given a A-tableau t, the row subgroup of t is denned to be St = {w G &r | (t(k,ij))w = t(k,i,f) for some 1 < f < \\k}}. Let tA be the A-tableau obtained by placing {1,2,..., r} in the order of the (total) lexicographic order on the set of boxes {(fc, i,j)\l<j<\\\i>l,l<k< m} in F(A). The row subgroup St\ is called the Young subgroup of A and denoted by ©a- Let D\ be the set of distinguished right coset representatives of &\\&r (of minimal length). It is known [DJ1] that the map w h+ txw is a bijection from D\ to the set of row-standard A-tableaux. Let Z?a,a' = D\C\ D^1, where D^1 is the distinguished left coset representatives in &r/&\>. We remark that the Young subgroup &\ denned for an m-composition A is the same as 6^ defined in [DR2], where A is the composition of r obtained from A by concatenation of the nonzero entries of components of A. We will also occasionally use the notion <5\ for &\. If A is an m-partition and t is a A-tableau, the transpose of t is the A'-tableau t' denned by t'(fc, z, j) = t(ra — fc+1, j, i). Define t^ to be the A-tableau obtained from F(A) by putting the integers 1, 2, • • • , r downward in the first column of Y(X^), followed by the second column of F(A^m^) and so on and then followed the same way for the columns of Y^™"1)), columns of Y(X^) etc. Note that tA = (tA')'. For each A G Am(r), each A-tableau t defines an d\(t) G &r such that t = txd(t). Then d\ defines a bijection between the set of all A-tableaux and 6r. Set w\ = ^a^a)- We omit the subscript A in d\ if A is clear from the context. If A is an m-partition we have w\> = w^1 since (tw)f = t'w for all w e &r and (t;); = t. 2.6. For a sequence a = (0 = ao < ai < a2 < • • • < am = r), an element of wa G &r is denned in [DR1, 1.6] by (2.6.1) (o>i-i + l)wa = r — a,i + I for all i with a^_i < a^, 1 < / < a^ — a^_i. For example, if a = (0 < 4 < 8 < 9), then _ A 2345678 9\ Wfl"\6 7 8 9 2 3 4 5 1/' For any A = (Ai, A2, • • • , Am) G A(ra, r), we define a sequence a = (0 < a^ < a2 < • • • < «m = r) by setting a^ = Y?j=i \^j\- We will write w\ for wa in this case. Each composition A G A(m, r) also defines an m-partition A with each component having only one part. Then we have w\ = w^. This fact will be used in Cor 2.8. Now let A be an m-partition and recall that [A] = (|A^^|,..., |A^|) G A(m, r). Let V be the 2-th component of the tableau tA and define w(i) G ©r so that it is identity on all entries of t-7 for j ^ i while it permutes the entries of V such that Vw{i) is the tableau by rearranging the entries of V in increasing order in the first column down and then second column and so on. Then w(l),-— , w(m) are pairwise commutative and (2.6.2) w\ = w(l)w(2) - - - w(m)w[x\. Let Si = (i,i + 1) be the basic transposition. Then S = {si, • • • , sr_i} is the set of Coxeter generators of the symmetric group &r as a Coxeter group. We will
138 ZONGZHU LIN AND HEBING RUI use the standard notion of reduced expressions of elements in &r and the length function l(w) for elements w G ©r. Denote by >wk the weak Bruhat order on 6r, i.e., x >wk y if there is a reduced expression y = s^s^ - — s^ of y such that x = s^s^ -"Sik for some k < I. Note that the weak Bruhat order should not be confused with the usual Bruhat order. In fact, x >wk y implies that x < y under the usual Bruhat order >. (We keep the terminology from [DJ1].) The following result was proved by Dipper, James, and Murphy in [DJ1] and [DJMu] form =1,2. Proposition 2.7. Let A be an m-partition ofr. (a) d\ defines a bijection from TS(X) to {w G ©r | w >wk w\}. (b) For any t G Ts(\), dA(t)dv(t,)_1 = wx and l(dx(t)) + l(dx,(t')) = l(wx). Proof, (a) Note that the map w i—► txw defines a one-to-one correspondence from ©r to the set of all A-tableaux. We only show that txw is standard if and only if w >wk w\. Note that ii w = w\, then txwx = t^, which is standard by the definition. Suppose t = txw ^ t^ is a standard A-tableau. Suppose that w >wk w\. Then there is an Si G S such that w ^>wk wsi ^_wk w\. By induction, we assume that txwsi = t is a standard A-tableau. Since l(wsi) = l(w) + 1, we have j = (i + l^ius;)-1 < (^(wsi)"1 = I. If i and i + 1 are in a same row (or same column) of a component of t, then col^') > col(/) (or row^') > row(/) ) in tA which is impossible. Thus i and 2 + 1 have to be in different rows and columns or in different components and tsi remains a standard A-tableau exchanging i and 2 + 1 in t. We claim w >wk w\ if there is a component t; of t which does not contain all entries of (t\)i. We take i maximal with this property. Thus numbers in the set A = {|A(z+1)I H (- |A(m)| + 1,..., r} will be the entries of the components of ti,..., t;. There exists k > \X^| + (- |A^| which appears in t^ while there is a number / G A such that / < k which is an entry of tj for some j < i. By choosing such / maximal, we have / in tj for some j < i and / + 1 in t^. Since tA = tiu_1, it follows from (Z)^"1 G t$ and (/ + l)^"1 G t£ that (Z)^"1 < (/ + l)^"1. Hence l(wsi) = l(w) + 1 and w >wk wsi. Since / and / + 1 are not in the same row or column of t, the A-tableau tsi is standard. By induction assumption, wsi >wk w\, and hence the claim follows. Now suppose that U and (t\)i have the same entry set for each i; = 1,..., m. If t 7^ t^, say tj 7^ (t\)j for a j, then there exist k < I such that col(fc) > col(/) in tj. Since tj is standard, it is necessary that row(fc) < row(/). We can choose k < I with this property such that I — k is the smallest. If / — k > 1, we note that k + 1 is an entry of tj and col(fc) < col(k + 1). Then k + 1 < / is also such a pair. Thus / = k + 1. Thus tsk is standard and (A:)^-1 < (k + l)iu_1. The latter is equivalent to say l(wsk) = l(w) + 1. By induction assumption, wsk >wk ^a5 and hence w >wk w\. (b). It follows from definition that for any A-tableau t txdx(t) = (t')' = (tA'dv(t'))' = tA<Mt') = t Wv(t') and w\ = dxfydxtft)-1. If t G TS(A), then (a) implies dx(t) >w wx and (b) follows. □ Corollary 2.8. (a) For any composition A = (Ai,--- ,Am) of r (with Xi = 0 for all i > m), A = ((Ax, 0,...), (A2,0, ...),•• • , (Am, 0,...)) 2's an
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 139 m-partition. Then w^ is the longest element in Dx = D^ such that for any x G Dx, x >wk w^, i.e., there is a y e &r such that w^ = xy and l(xy) = l(x)+l(y). (b) For any m-partition X, the element W[X] is the longest element in £*[a],[A']- Proof. There is a bijection between the set of row-standard A-tableaux and the set of standard A-tableaux. So, (a) follows from Proposition 2.7. In order to prove (b), we have to verify W[X] G D7~J, for any m-partition A, which has been proved in [DR1, 1.8]. □ 3. Cyclotomic Hecke algebras of type G(m, 1, r) 3.1. Let Rbe a, commutative ring with identity 1. We fix q, ixi, • • • , um G R such that q is a unit in R. In [AK], Ariki and Koike denned the cyclotomic Hecke algebra i/m,r of type G(m, l,r) as an associative algebra over R generated by Ti,0 <i <r — 1 satisfying the following conditions: ( (T0 - Ui)(To - U2) • • • (To - Urn) = 0, (Ti - q2)(Ti + 1) = 0, for any 1 < i < r - 1, < Toll Toll = TiToTiTo, TiTi+iTi = Ti+1TOi+u for 1 < 2 < r - 2, { TiTj = TjTu if \j - i\ > 1. Following [AK], write ti = T0 and U = <?~2T;_i£;_iT;_i if i > 1. Then U and tj commute with each other. 3.2. Let Si = (z, i +1) be a basic transposition. Set TSi = Ti. Ifw = Si1—- Sik is a reduced expression of w G 6r, we define Tw := T^T^ -Tik. It is known that Tw is independent of the reduced expression of w. For any m-composition A of r, recall from 2.5 that A is a composition and &x is the corresponding Young subgroup of &r. We set xx= "£TW and y-x = £ (-q*)-l{w)Tw. Write A = (A*1),--- ,A<m>). Recall from 2.6, a{ = Y?j=i\*{j)\- Following [DJM] and [DR1], we define 7r[A] = Yl™^1 7rai(ui+1) and 7r[A] = Yl™^1 7rai(um-i), where 7Ta(x) = n?=ife "~ x) f°r any x ^ R and any positive integer a. Let x\ = tt[X]Xx and y\ = tt[a]2/a- There is an R-linear anti-involution * on Hmr such that T* = Ti,0 <i <r — 1. For any standard A-tableaux s, t G TS(A), set (3.2.1) x*t = T^s)xxTd{t) and y*t = T^s)yxTd{t). Note that T^ = Tw-i and t* = U. Thus x\ = xx, y\ = yx. By [DR1, Cor. 2.7], xx and yx commute with n[X] and 7T[a] . Therefore we have *a = xx, y\ = yx, «t)* = *t,s> (i£t)* = 2/tV Let $ : i? -^ i? be a ring automorphism, for example., take $ to be the identity map. Set q = $(g)_1 and Ui = $(um_i+i) for i = 1,2, ...,m. Let Hm^r = Hm^r(q, ui,..., Um) be the cyclotomic Hecke algebra over R corresponding to the parameters q, &i,..., um. We will add ~ to any symbols representing objects related to i/m,r to indicate that they are related to Hmr. We now extend $ to a
140 ZONGZHU LIN AND HEBING RUI ring isomorphism $ : Hmr —► Hm,r by $(T;) = (—q2)~lrTi for 1 < i < r — 1 and $(T0) = T0. It follows from the definition, one has $(TW) = (-q2)~l{w)fw and ®(U) — U- We have *(*a) = #A> *(W) = *A> *(*>]) = *"[A], *(*[A]) = *[A]- Therefore *(xa) = y\ and $(?/A) = xx. Furthermore, *(*2i) = (-4»)-('«-))+'«*»)^ and *(£) = (-^J-CW-M+iWtMJjA . Proposition 3.3. Suppose A e A+(r). (a) [DR2, 2.10] The right Hm^-module Sx = x\TWxy\'Hm,r, called the Specht module with respect to the m-partition X, is a non-zero free R-module with a basis {xxTWxyyTd(t) | t e TS(X')}. (b) The left Hmr-module Hin,rx\TWxy\> is free over R and has an R-basis {T^t)xxTWxyy\teTs(X)). Proof. The result (a) is proved in [DR2, 2.10] and thus holds for #m,r. Applying the ring isomorphism 3>_1 : Hm^ -+ Hm,r to the Specht module Sx for Hm,r, we have <f>-\Sx') = ^-\xxtfWxlyxH^r) = yX'TWx,xxHmtr. Applying the anti-involution * : Hmr —► Hm^ to both sides of the above equality and using w^,1 = w\, we have *o$-1(Sx') = HmtrxxTwxyx: Now, the result (b) follows from (a) immediately. □ 3.4. For an m-partition A, an m-tableau S of shape A is a map S : Y(X) —> {1,..., m} x N. This is equivalent to coloring the boxes of Y(X) by ordered pairs of numbers (i,j) such that 1 < i < m and j > 1. The type of S is the m-composition /x = (/x(1),..., /x(m)) such that /x|*° = IS-1^, l)\. In [DJM] S is called a A-tableau of shape /x. We call S an m-tableau of shape A and type /x to distinguish with the A-tableaux t defined in 2.2. When m = 1, they are all called A-tableaux although they are still different from the A-tableaux t. If S is an m-tableau of shape A and type /x, then S'(fc, i,j) = S(ra — k + 1,^', i) defines an m-tableau of shape A' and type /x. The lexicographic order on the set {1,..., m} x N is a total order. For an m-partition A, we call an m-tableau S of shape A is semi-standard if (a) S(k,i,j) > S(k,i,jf) if j > f, i.e, weakly increasing in each row; (b) S(k,i,j) > S(k,i',j) if i > i', i.e., strictly increasing in each column; (c) If S(k,i,j) = (p, Z), thenp> k. For A e A+(r) and /x e Am(r), let TSS(A, /x) be the set of all semi-standard m-tableaux of shape A and type /x. Note that TSS(A, /x) ^ 0 implies A > /x. An m-tableau T is called column semi-standard [Ml] if T' is semi-standard. Let TCS(A,/x) be the set of column semi-standard A-tableaux of type /x. Then Tcs(A,/x) ^ 0 implies A' > /x.
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 141 Let uj = ((0),..., (0), (lr)) G A+(r). For any A G A+(r), and any A-tableau t we can define an m-tableau T(t) by T(t)(k,i,j) = (m,t(k,i,j)) of shape A and type u. We have T(t) G Tss(X,u) if t G Ts(\). 3.5. Suppose A G A+ (r) and /x G Am(r). For any t G TS(A), let /x(t) be the ra- tableau of shape A and type /x denned by /x(t)(fc, i,j) = (p, I) if t(fc, z,^) = tM(p, /,/) for some 1 < f < /x^ . For any m-tableau S of shape A and type /x, let /x-1(S) be the set of all t G TS(X) such that /x(t) = S. It is known (see [Ml]) that there is a unique standard A-tableau S G /x_1(S) such that d(S) >wk d(t) for any t G /x_1(S) [Ml]. The following lemma follows immediately. Lemma 3.6. Suppose S G Tss(A,/x). (a) s G /x_1(S) if and only if its conjugate s' G /x~1(S/) C Ts(Xf). (b) J(d((S)')) > /(d(t)) for any t G /x"1^') with t ^ (S)'. Proof, (a) follows from the definition of the map \i in 3.5. By Proposition 2.7, we have ^t^t')"1 = wx with l(d(t)d(t')) = l(d(t)) + /(d(t')-1) for any standard A-tableau t. Now, (b) follows from the fact d(S) >wk d(t) for any t G /x_1(S). D The following result will be used repeatedly, we quote it here for convenience. Lemma 3.7. (a) [Ml] For A,/x G Am(r), x^Hm^yx ^ 0 if and only if there is a v G A+(r) such that Tcs(v,X) ^ 0 and Tss(i/,/x) ^ 0. in particular, Xf > n if X e A+(r). (b) //A,/x G A+(r) and^vtfm,r^svt ^ 0/orsomes,t G f(A'),u,ve Ts(/x), tten n < A. (c) //A,/x G A+(r) an^svttfm,r:r£v ^ 0/orsomes,t G Ts(A'),u,v G Ts(/x), tten /x < A. Proof, (a) is due to Mathas in [Ml, 6.8]. Under the assumption of (b), x^Hjn^yy =fi 0. By (a), there is a v G A+ (r) such that i/ > A' and v > /x. Using Lemma 2.4 we get A > i/. Now (c) follows by applying the anti-involution * to (b). □ Lemma 3.8. Suppose X G A+ (r). (a) For any w G &r with l(w) < l(w\), we have x\Twy\> =fi 0 only when w = w\. (b) For any x,y G &r with l(x) + l{y) < l(w\), we have r.r„v - {j; „ r rp n ) x\TWxy\>, ifxy = wx, x\±xlyy\> = \„ ^u otherwise. Proof, (a) By writing w = w\W2Ws with w\ G &\ and w% G ©a' and i^ G A\,A'5 we can assume that w G £>a,a' with /(iu) < l(w\). Suppose x\Twy\> ^ 0. By writing w = W2W1 with w\ G D[a] and W2 G ©[a]5 we have x\Twy\> =xx7r[A]Ttl,2rtl,17r[v]2/v- Note that T^2 commute with 7T[a] since 1^2 G ©[a]- Therefore, x\Twy\> ^ 0 implies TTfAjT^Tqv] ^ 0. By the proof of [DR1, 3.1], ir[X]TWlit[x'] = 0 for w1 G £>[a][A'] unless wi = W[x\- Note that W[X] is the longest element in -D[a][a;] (Corollary 2.7(b)). Now,
142 ZONGZHU LIN AND HEBING RUI write W2 = X\X2 • • 'Xm such that Xi G ©|a*| and ®|A*| IS the symmetric group on the set {Epi Ia0)I H-1, • • • , £j=i |AW|}. By the definition of yx,, we have ?/A(1)/VA(2)/V---VA(^)/^[A] = ^[A]^"'' where the composition A^1) V A^2) V • • • V A^m) is obtained by concatenating the components A<x> ,A<2> ,••• ,A<m> . We have x\Tw2n[\]TW[x]7r[x<]yx< = I JJ^a*^.^' J ^M^^IA'] 7^ 0. We can similarly assume that Xi is a (minimal) double coset representative of 6A(o\©|A(0|/©A(i)/- If Kw) ^ K^a), there is an i such that l(xi) < l(w(i)), the latter is denned in Section 2.5. Thus, xX{i)TXiyX{i)> ^ 0. This is a contradiction since [DJ1, 4.1] implies that w(i) is the unique element in DX(i) A(*>/ such that ^0^(1)^(0' ¥" 0- This completes the proof of (a). (b) For any x,y G 6r, X^X^ is a linear combination of the element Tz with /(z) < l(x) + /(?/). By (a), we can assume that l(z) = l(wx). This happens only when z = xy. Now, the result follows from (a) immediately. □ 3.9. For any S G Tss(\, /1) and T G TSS(A, 1/), following [DJM] set ™<st = Yl Xst' sC:M-1(S),t€i/-1(T) Set TA = A(tA). Then A"X(TA) = {tA}. Therefore, for any T G Tss(A,/x) (3.9.1) mTXT= ]T x\Tm. tG/x-MT) Lemma 3.10. For am/ A G A+(r) and /x G Am(r), suppose T, Ti G Tcs(A',/x) wiffc /(d(ti)) < /(d(t)) and t G TS(A). Then (3.10.1) mTAT/^'it = iT^^^A'^t). Proof. By Lemma 3.6 and Proposition 2.7, we have (a) l(d(s)) + /(d(t)-1) < Z(wA) for any s G /i-1^') with equality holds only if d^T)"1 = wA, i.e., s = (T)'. (b) l(d(s)) + Z(d(f i)"1) < l(wx) and ^(sj^t)-1 ^ wx for any s G /x_1(T'i). Using (3.9.1), (3.10.1) follows from Lemma 3.8 immediately. □ 3.11. (Graham-Lehrer [GL]) An algebra A over a ring # together with the poset (A, <) is called a cellular algebra if A is free as an R-module of finite rank and for each A G A, there is an index set /(A) such that (a) Ua<ea£A is an .R-basis of A, where Bx = {c^ \ i, j G /(A)}; (b) There is an anti-involution a on A such that cr(c^) = cA^; (c) For each a e A, one has a • c^ = Ylkei(X) /*,a(g> ^)ckj (m°d A>x), where fi,\(a, k) G R is independent of j, and >1>A is the free .R-submodule of A spanned by Um>a#m. Fix an index j G /(A). Let A(A, j) be the free .R-submodule of A/A>x generated by {c^ = c\- + ,4>A I i G /(A)}. Then A(A,j) is a left ^-module such that A(A,j) = A(A,j') for any f G /(A). This module, denoted by A(A), is called a cell module with respect to A.
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 143 Theorem 3.12. (a) [DJM] The set {x*t | A g A+(r),s,t g Ts(X)} is a cellular basis of i/m,r with respect to the dominance order on A+(r). (b) [DR2] The set {y£t j A G A+(r),s,t G TS(X)} is another cellular basis of Hm,r with respect to the dominance order on A+(r). Note that applying the isomorphism $_1 : Hmr —► Hmr (cf. 3.2) to the basis in (a) for the algebra #m,r, one can get the basis in (b) for Hm^ up to invertible elements of R. For any subset T C Am(r), let r+ = {A G A+(r) | A > /x for some /x G T}. We say that T is saturated [Ml] if T+ C T, i.e., T+ = T D A+ (r) is a coideal in A+ (r) under the dominance order. Proposition 3.13. IfTC Am(r) is saturated, then the R-submodule j(r+) = £ %AtlM^s(A')} A€A+(r)\r+ 25 a two-sided ideal of Hm^. Proof. Note that T is saturated implies that T+ is a coideal in A+ (r). By Lemma 2.4, (r+)' is an ideal and A+(r) \ (r+)' is a coideal. Note that 7(r+) = SAeA+(r)\(r+)' ^{^/st I s5* ^ Ts(\)}, which is a two sided ideal following Theorem 3.12(b) and the definition of a cellular basis. □ For any saturated subset F C A+ (r), set ff(r) = Hm,r/I(T+). Corollary 3.14. IfTC. Am(r) is saturated, then H(T) is a cellular algebra with cellular basis 0&|Ae(r+)')S,te:r(A)}. 4. Quasi-hereditary cyclotomic g-Schur algebras Definition 4.1. [M2] For a finite subset r of Am(r), set Tr = 0A<Er£A#m,r, which is a right i7m,r-module and S'(r) = End#mr(Tr). Tr is an S(T)-Hm^r- bimodule. Proposition 4.2. [DJM, 6.3] Let /x and v be two m-compositions. Then ^^Hm,r H Hjn^Xy is a free R-module with a basis {?tist | S G TSS(A,/x),T G Tss{\,v), for some A G A+(r)}. For any T C Am(r), define f = {A G A+(r) | Tss(A,/x) ^ 0 for some /x G T}. Note that r = T+ if T is saturated. But they are different when T is not saturated (see Remark 5.10). For any /x, v G Am(r), A G A+(r), S G Tss(A,/x) and T G Tss(A,i/), define $£T G Hom#mr(^#m,r,:rM#m,r) by (4.2.1) $st(^) = ^st^. $gT can be regarded as an element in S'(r) by letting $gT(xM/i) = ^vym^^h. Theorem 4.3. [DJM, 6.6, 6.18] Assume T C Am(r) is finite. (a) S'(r) 25 a cellular algebra over R with a cellular basis {$£T | s g rss(A,/x), t g rss(A, i/), /x, v g r, a g f}.
144 ZONGZHU LIN AND HEBING RUI (b) When F is saturated and R is a field, the cyclotomic q-Schur algebra S(F) is a quasi-hereditary algebra in the sense of [CPS1], with standard modules A(A) (X G T+) being the cell modules with R-bases {$jgT I ^ ^ TSS(X, /i), /i G T} for a fixed T G TSS(A, v) and a fixed v G T. 4.4. Suppose T is saturated. For each A G T+ C T, by 3.9, mT\T\ = x\ and $£,ata £ S'(r) is the identity map in Hom#m r(x\Hmir,x\Hmir). Thus $st$t*t* = $st for any s e T88(i/,fj) and T G Tss(v,X). This implies A(A) = S(r)$*XTX for any A G T+. Lemma 4.5. For any X G T+ C T and t G TS(X), there is an S(T)-module homomorphism ft : A (A) —► Tp s^c/i £/m£ /t(*r*T*) = x\TWxyx>Td(t). Proof. We need to show that for any <j> G S'(r) one has cf)(x\TWxy\'Td(t)) = 0 whenever </>$*ATA = 0. Note that $st^t*t* = ® an(* ^ST^A^^A'T^t)) = 0 if T G Tss(v,fii) with /xi ^ A. By Theorem 4.3(a), we can assume SGTss(i/,^),TGTss(i/,A) Since 4>^xTx G £(r)>A, we have i/> A for all i/ with rgT ^ 0 (see 3.11). For i/>A, Lemma 3.7 and 3.9 imply $sT(x\TWxy\iTd(t)) = msTTWxyx>Td(t) = 0 for all S,T. This proves our claim. □ Theorem 4.6. Suppose T is saturated. For any X G T+, Homs(r)(A(A),Tr) is a free R-module and the map x\TWxy\>Td(t) •—> ft ft G TS(X)) defines an isomorphism of right Hmr -modules SA^Hom5(r)(A(A),Tr). Proof. For any / g Homs(r)(A(A),Tr), we compute /($AATA) g Tr. By [Ml, 5.9], x^H^r is free over R with a basis {x^y^ | S G rcs(i/;, /i), t G Ts(i/), ^ G A+(r), i/ > /x}. Therefore, /(*TVT>) = £ r&*M0fc S<ETcs(i/,^),t€Ts(i/) where ^ G # . NotinS that $t*t*(xm) = <Wa, we have ^ata/(^ata) = /(*t*t*) and i$t = ° for any S G ^i^' aO with A* ^ A* Thus we have (4.6.1) /(*i*T*)= E ^>^It- A<i/€r+ S<=Tcs(i/,A),t€Ts(i/) We claim r| = 0 for any v > A and S G Tcs(v', A). Otherwise, take i/o to be maximal among all v G T+ with i/ D> A and r£. ^ 0 for some S and t. Also, choose So such that /(d(So)) is maximal among all S 6 Tcs(vf0,X) with r^° ^ 0 for some
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 145 t. Since i/0 G T+ C T, we have $t°,oS, G S(r)- Hence ^tW (*t*t*) = ° b^ Theorem 4.3(a). On the other hand, Lemma 3.10 implies t€T-(i/J) Now, Proposition 3.3(a) shows that r^° = 0 for any t G Ts(v'0), which contradicts the choice of v$ and So- Note that xa = mTATA. Applying Lemma 3.10 again we have /(*t*t*)= Yl rtxxTWxyx>Tdit), teT3(\f) where rt G R. Therefore, by Lemma 4.5, t€Ts(A') Using Proposition 3.3 again, one will see immediately that {ft | t € Ts(Xf)} is an .R-basis of Homs(r)(A(A),Tr) which is therefore a free .R-module. Obviously, the ^-linear isomorphism sending ft to x\TWxy\>Td(t) is a right i7m,r-module isomorphism. D Proposition 4.7. For any T C Am(r) (W£ necessarily saturated), the left S(T)-module Tr has a A-filtration such that the multiplicity of A(X) is #TS(X) for any X G T. Proof. By [DJM, 4.12], Tr is a free .R-module with a basis {mST(t) I S G Tss(A,/x),t G Ts(A),/x G T, A G T}. Note that T(t) G Tss(A,o;) denned in 3.4. For notational simplicity, we will simply write t for T(t). Its meaning will be clear from the context. Write T = {Ai, A2 • • • , } such that Xj < Xi implies j > i. Let Mi be the .R-submodule of Tr generated by {mst I S G TM(Aj,/i),t G Ts{Xj),ii G T and j < %). For any A G T,/x, 1/ G T, and S G Tss(A,/x),T G Tss(A,i/), we will show that $ST(mSi t) G Mi for any Si G Tss(Xj,fii) and t G TS(A^) (j < i) for some /xi G I\ By (4.2.1), mSlt = ^sit(^)- If /j,! z£ v, then $gT o $>SJ t = 0. Therefore, we can assume ji\ = v. Since $st ° *sl,t e HomHmir(iw5m)r,iMffm>r), we consider the obvious embedding S(T) C S(r U a;), and use cellular basis in Theorem 4.3(a) (for S(T U {u})) to (4.7.1) *aT°*£,t= E /5a,.*S S2,s^S2,s* Aj<ry€ru{o;} S2€Tss(r/,M),s€Ts(r/) Here /g2S G #. In the summation, Tss(rj, /x) ^ 0 implies 7/ G f. Thus 7/ = Aj > Xj for some I < j < i. Acting on x^ we get $|2 s(xa;) G M*. Therefore $sT(mSit) £ A^ and Mi is an S'(r)-submodule of Tp. In the summation of (4.7.1), If 77 = A^, then s = t (again by the cellular basis property for S^r U {u}) as in Theorem 4.3(a)). Thus for each t G TS(A;), the .R-submodule M^ of M;/M;_i generated by {m^ |
146 ZONGZHU LIN AND HEBING RUI S G Tss(A;,/x),/x G T} is an S(r)-submodule and free over R. Fix T0 G Tss(A;,/x) for some /x G T. Using Theorem 4.3(a) again, one can verify similarly as above, that the .R-linear map / : M^ -+ A(Ai) with /(m^) = $St0 is a left S(r)-module isomorphism. Thus Mi/M.-i = A(A;)#TS(Ai) as 5(r)-modules. □ In [DIR] extended the concept of quasi-hereditary algebra to quasi-hereditary rings. More generally, in [CPS2], the concept of quasi-hereditary algebras over commutative rings are also characterized in terms of stratified algebras. In this case, the class of modules A-modules and V-modules are described. A module is called tilting in this setting is it has a A-filtration and a V-filtration. Lemma 4.8. Suppose that ixi, v,2, • • • , um are invertible in R = Z[q, qr1]. // T is saturated, then (a) Tr is a tilting S(T) -module. (b) The R-algebra Ends(r)(Tr) is a free R-module. Proof. Since 1x1,1x2,-•• ->um are invertible, by [Ml, 5.13], there is a non- degenerated bilinear from { ,)\ : X\Hm^ x x\Hm^ —> R. It induces a non- degenerated bilinear from (,) on Tr with IU h \ - ) ^1, ^A' if hl' h<2 e XA#m,r, A G T, JO, otherwise. The bilinear form (,) is associative in the sense (xh, y) = (x, y(h*)) for all h G i/m,r. Thus Tr is self-dual as a right i7m>r-module. Since * : i/m,r —> Hm,r is an anti- automorphism, [DJ2, (1.5), (1.6)] implies that Tr is also self-dual as a left S(F)- module. By Proposition 4.7, Tr has a A-filtration. This implies that Tr also has a V-filtration as S'(r)-module. Since S'(r) is a quasi-hereditary algebra over any field k and the cellular basis in Theorem 4.3(a) is independent of k, then A(A)^ = A<S)Rk and, by [DPS1, 4.2(b)], Ext25(r)(A(A),TF) = 0 for all i > 0. Theorem 4.6 now implies that Ends(F)(Tr) has a S^-filtration of right i7m,r-module. This implies that Ends(r)(Tr) is a free .R-module. □ To extend the result to other base rings, we need the following Lemma 4.9. Suppose u\,U2, — ' ,um are invertible elements in R = Z[g_1,g]. For any commutative R-algebra A and any saturated subset T C Am(r), let S(T)a = S(T) <S)r A and (Tt)a = ®\erX\Hm,rA. Then End5(r)Crr) ®R A ** End5(r)®HA((Tr)A). In particular, if A is a field, then dimAEnds{r)0RA((Tr)A) = £ #T*(A)2. Proof. In the proof of Lemma 4.8, we have proved that Tp is a tilting S(T)- module. By using [DPS1, 4.2c], we see that the base change property holds for Ends(r)(®A€r#A#m,r)- Now, the formula about the dimension follows from Theorem 4.6, Proposition 3.3(a), and Proposition 4.7. □ Theorem 4.10. IfT is a finite saturated subset of A+(r), and R is any commutative Z[q, q~1]-algebra containing invertible elements u\,..., um, then H(T) = End5(r)(eA<ErZA#m,r) and s(r) - Endtf(r)(eA<ErZA#m,r).
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 147 Proof. The natural map Hmr —> Ends(F) (Tr) by right action defines a homo- morphism of R-algebras. For any \i G A+(r)\r+, by Lemma 3.7(b), y£t annihilates Tr. Thus the map Hm,r -> End5(r)(Tr) factors through <j>: H(T) -> Ends(r)(Tr). We claim that the map <\> is injective. Suppose we have ^( Yl rs,t2/s,t) = °> and not a11 rs,t are zero. A€r+,s,t€Ts(A;) We can choose v G T+ maximal among all A G T+ with r£t =fi 0 for some s, t G TS(A'). For such a */, choose so G Ts(v') such that Z(d(so)) is maximal among all s G Ts(i/) with Tg t ^ 0 for some t. By Proposition 2.7, there is a, y e &r with i<V = d(so)y and Z(i<v) = /(d(so)) + l(y). Acting on xvT* G xI/i7m>r, we have (4.10.1) 0= ]T rsvt^r;^'t = ]T ^x^t^^t^). AGr+ ,s,t€Ts (A') A€r+ ,s,t€T* (A') Lemma 3.7(b) implies that XvTyT^yx ^ 0 only if v < A. By the choice of */, the only possible nonzero terms on the right hand side of (4.10.1) are when A = v. Now, we deal with terms x^T^Tt^/T^t) with s ^ s0. Since Z(d(so)) is maximal, Lemma 3.8(b) implies that xJT*T^,^yviT^^ = 0 for all s ^ s0. If s = s0, then xuTyT^B)yv'Td(t) = xvTWuyv,Tm (using wv, = w'1), which is a basis element of the Specht module S" (see Proposition 3.3(a)). This forces rs0,t — 0, contradicting the choice of so and v. Hence, all r£t are zero and <\> is injective. Note that H(T) is a free .R-module of finite rank. The proof of the injectivity of (f> works for arbitrary commutative ring R. To show the surjectivity, we use the assumption on R, which makes R an Z[g, g_1]-algebra. For any field A and any ring homomorphism R —> A, A is automatically a Z[q, g_1]-algebra. Now we can apply Proposition 3.13 and Lemmas 4.8-4.9, to get dimA(H(T)®RA)= J2 #T°(\')2= Y, *TS(X)2 = dimA(End5(r)(eA€rZA#m,r) ®R A)- Thus <j> <S>r 1 : H(T) <S>r A —> Ends(r)(®AGr#A#m,r) <8>r A is an isomorphism for any field S. Now, Nakayama lemma implies that <j> is an isomorphism over R. □ 4.11. In [M2, 5.3], Mathas proved the double centralizer property for finite saturated set T such that u> = ((0),..., (0), (lr)) G T, by using the cyclotomic Schur functor. In this case, X(jj — 1 and Tr is thus a faithful Hm r-module. In our case, Tr is not faithful. We remark that the double centralizer property for the Hecke algebras of type A plays an important role in the proof of quantum Schur-Weyl reciprocity in [DPS1]. However, we do not know how to realize the permutation modules Tr as a representation of certain quantum groups. In the next section, we will construct a special tensor space Tr on which the affine quantum group Uq(gln) acts and the cyclotomic q-Schur algebra will be a quotient of the affine quantum group Uq(Qln). This enables us to setup the quantum Schur-Weyl duality in this case.
148 ZONGZHU LIN AND HEBING RUI 4.12. For any quasi-hereditary algebra iona poset A. the isomorphism classes of indecomposable tilting modules are indexed by A [R]. Ringel [R] calls an A-module M full tilting if it is a tilting module M and every indecomposable tilting module is isomorphic to a direct summand of M. If M is a full tilting module, the endomorphism algebra End^(M) is called a Ringel dual of A. The following result generalizes [DPS1, 8.4]. Using the Morita equivalent theorem for i7m,r [DM], we can focus on the case Ui = qCi, i = 1, 2, • • • , m where c; are some integers. This will enable us to use [A2, 4.3]. Corollary 4.13. Suppose T c Am(r) is finite and saturated, R is a field, and Ui = qCi. Then H{T) is the Ringel dual of S{T) if and only if all partitions in {A' | A G T+} are Kleshchev m-partitions. Proof. By Lemma 4.8, Tr is a tilting S'(r)-module. Tr is a full tilting S(T)- module if and only if the non-isomorphic indecomposable S'(r)-summands of Tr are indexed by T+. By [CPS2, 1.1], Tr is full tilting for S(T) if and only if the isomorphism classes of irreducible Ends(r)(Tr)-modules are indexed by T+. Now using Theorem 4.10, we further get that Tr is a full tilting S'(r)-module if and only if the isomorphism classes of irreducible iif(r)-modules are indexed by T+. Therefore H(T) is a Ringel dual of S'(r) if and only if the irreducible H(T)-modules are indexed by T+. However, by Corollary 3.14, H(F) is a cellular algebra on the poset (r+)', which is an ideal of A+ (r). By [A2, 4.3], the isomorphism classes of irreducible i7(r)-modules are indexed by the set {A | A e (r+)' is a Kleshchev m-partition}. (In fact, In order to use [A2, 4.3], we need the fact $(a£t) = (-q2)-(i(d(s))+i(d(t)))~\^ where $ : i/m,r —> Hm^r = Hm^r(q, tti,..., um) is a ring isomorphism given in the proof of Proposition 3.3). By taking $ to be the identity over R, we have q = q_1 and Ui = Ujn-i+i and H satisfies the same conditions as i/m,r does. Therefore, H(T) is a Ringel dual of S'(r) if and only if all m-partitions in the set {A; | A e T+} are Kleshchev m-partitions. □ 5. Quantum Schur-Weyl duality 5.1. Let A = Z[q, q_1] be the ring of Laurent polynomials in indeterminate q. The extended affine Hecke algebra Hr of type Ar-\ is a unital associative algebra with generators T^X^ subject to the following conditions: (a) (Ti + l)(Ti -q2) = 0, 1 < i < r - 1, (b) TiTj=TjTu |j-t|>2, (c) TiTjTi=TjTiTj, it\i-j\ = l, (a) XiXj = XjXi, XiX^ = 1, (e) TiXj=XjTi, ifjVM + 1, (/) TiXiTi = q2Xi+1. Let p = X\T\ - - -Tr_i. Then pTip~x = TJ+i, where the lower indices should be read module r. Denote by T0 = pTr_ip_1. Then T0,Ti,--- ,Tr_i generate a subalgebra Hr, which is isomorphic to the affine Hecke algebra of type Ar-i. The subalgebra Hr generated by Ti,T2,--- ,Tr_i is isomorphic to the Hecke algebra associated to the symmetric group ©r. Take invertible elements ui,--- ,um in Z[q, q~1]. Consider the polynomial / = IIIliP^i ~ui)- Then the cyclotomic Hecke
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 149 algebra is the quotient algebra i/m,r = Hr/I, where / is the two-sided ideal of Hr generated by the polynomial /. We mention that Cherednik [Ch] studied finite dimensional quotient algebras of affine Hecke algebras. 5.2. Let V be a vector space over the field C(q) with dimension n. All tensor product is over the field C(q). The tensor space V®r has a basis {v\ | i G [l,n]r}, where v\ = v^ <g) Vi2 <g) • • • <g) vir with i = (h, ^ * * * ,v) £ [1> nY'• The symmetric group 6r acts on the set [l,n]r by iiu = (^i)w,i(2)iu> • • •jfc(r)w) f°r &U * £ [ljn]r and w e &r. &r also acts on V®r by permuting the basis elements vf = v-lw. It is known that the tensor space is a right i7r-module [J] such that the following conditions hold: (5.2.1) ViTk = { qvu Xik = ik+i, qv*k, ififc>Zfc+i, [qv?k + (q2 - l)vu if ik<ik+1. Recall A(n, r) = {A = (Ai,..., An) G Nn | £"=1 Aj = r) an<* A(n, r)+ is the set of partitions of r with at most n parts. Recall that &\ is the Young subgroup with respect to the composition A and X\ = Ylwe&x ^wi where TSi := T; for the basic transposition Si = (i,i + 1). It is known that there is a right i7r-module isomorphism AGA(n,r) Let x = (xi,..., xr) be indeterminates. The affine g-tensor space is defined to be the C(g)-vector space f(n, r) = V®r <8>c(q) C(g)[xf, • • • , xf] which has a basis {vix* | i G [l,n]r,a G Zr}, where x& = x^'-x^ with a = (ai,...,ar). For notational simplicity, we have dropped the tensor product <g) in the basis element v\ <g) x&. The affine tensor space T(n, r) is a right i7r-module such that the action of Xi is to multiply x~x and the action of Tk is given by VixaTk = ViTkx^ + (1 - q-2)ViXi{xl ~Xa) X{ 3?i+l The above action can be found in [GV]. In fact, it comes from the formula on the multiplication of Tk and Xi in Hr. Lemma 5.3. There is a right Hr-module isomorphism (B\eA(n,r)x\Hr = V®r ® C(g)[xf,...,x±]. Proof. The weight of i G [l,n]r is A G A(n, r) defined by \k = \{j G [1, r] | Zj = fc}|. Note that iw and i have the same weight A. For a given A G A(n, r), set i(A) = (l,...,l,2,...,2,...,n,...,n) Ai times A2 times An times which is an element of [1, n]r of weight A. For any i G [1, n]r of weight A, there is a unique w G D\ C 6r such that i = (i(A))iu. This defines a bijection between D\ and i(A)6r. Thus we have v\ = vf(X)' Then the linear map V®r —> ©agA^^^a^ defined by ^i »—► x^T^ is the right iJr-module isomorphism mentioned earlier. Now, the linear map f{n,r) -> 0AGA(n,r)^A^r
150 ZONGZHU LIN AND HEBING RUI defined by ^xa |—> xxTwXa if i = i(X)w has weight A and w G D\, is a bijection. Also, by Bernstein formula (5.3.1) X&Tk = TkX? + (1 - g-2)*<+'(X» ~/a) -*2+l "~ -A 2 one will check easily that the map is a right ^-module homomorphism. □ 5.4. We now define the cyclotomic g-tensor space Tm(n,r) = (^r0C(g)[xf,...,x±])^ri7m,r where i/m,r is the cyclotomic Hecke algebra of type G(ra, l,r) mentioned above. For each A G A(n, r), we associate an ra-composition X^ with rath component being A and all other components being 0. Let A(n, r)Iml C Am(r) be the subset of all such ra-compositions A'ml with A G A(n, r). Then for A G A(n, r), the image of x\ denned in 5.2 under the quotient map Hr —> i/m,r is xA[m] as denned in 3.2. Tensoring i7m,r over Hr to the isomorphism in Lemma 5.3, we have the following isomorphism of right i7m,r-modules (5.4.1) Tm(n,r)^ 0 xxHm,r A€A(n,r)[H In [AK], it is proved that Hmr is a left i7r-module with basis {th | b G Z£j. Here Zm = {0,l,...,ra— 1}. The following result can be proved easily. Lemma 5.5. The tensor space Tm(n, r) = V®r ® C(g)[xf, • • • , x±] ®£ #, ra,r /ms a C(q)-basis {v{th | i G [l,n]r, b G Z£j. 5.6. We recall the definition of quantum afnne Uq(Qln) introduced by Drinfeld in [D], called Drinfeld's new realization. Uq(Qin) is an associative algebra over C(q) with generators C,C-\Ka,K~\Ha(^,ieZ\{0} ^ndEa(j),Fa(j)J eZ,a = l,2r-^n-l with relations, which can be found in [GV, Sec.3]. The algebra Uq(Qln) can be denned over the ring C[q, q_1] and then q can specialized to non-zero elements of C, which will still denote by q. It is proved in [GV, Lu, VV] that, if q is not a root of unity, there is a surjective algebra homomorphism [Sch2, 7.2, 7.3] (5.6.1) E/,(fl\)-*S(n,r)=EndAr( 0 xxHr). A€A(n,r) The latter is called afnne g-Schur algebra denned in [Gr]. We now define 5m(n,r) =EndHmir (Tm(n,r)). By setting V = A(n, r)^m\ we have Sm(n,r) = S(T) denned in Section 4. But A(n, r)[ml is not saturated. For any </> G S(n,r), <j> = </> <g) 1 : Tm(n, r) —> Tm(n, r) is a homomorphism of i7m,r-module. Thus we have a natural algebra homomorphism £(71, r) —> Sm(n, r). Proposition 5.7. (a,) T/ie natural algebra homomorphism S(n,r) -» Sm(n,r) is surjective. (b) If q is not a root of unity, then there is a epimorphism Uq(Qln) -» Sm(n, r).
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 151 Proof, (a) For A G A(n, r), we simply drop the superscript [m] from X^ for notational simplicity. For any A, \i G A(n, r) and 1/ G A+(r), S G Tss(v, A) and T G Tss(v, /x), $gT G Hom#m r (x\Hmir, x^Hm^) is determined uniquely by ^stC^a) = ^st £ x^Hjn^r D Hm,rx\. Since 7r: iJr —> i7m,r is a surjective algebra homomorphism, one can find rfisT € -Hr obtained from msT by replacing ^ with Xi for all 2. By the proof of [DJM, 6.3] (which follows from [DJM, 4.10]), mST e x^Hr D Hrx\. So one can define <j> G Hom^ (xxHr,x^Hr) with </>(£a) = msT- By Theorem 4.3, the natural algebra homomorphism S(n,r) —> Sm(n,r) is surjective. (b) follows from (a) and (5.6.1). □ 5.8. Recall (cf. 4.3) that for T C Am(r), r = {A G A+(r) | TSS(A, /x) ^ 0, for some /x G T}. Theorem 5.9. Suppose i7m,r 25 semisimple over a field R. (a) For any T C Am(r), tte algebra homomorphism i7m,r —> Ends(r)(Tr) 25 onto. (b) Suppose R = C(q). Then Uq(gln) acts on 0A€A(n,r)M ^A-Hm.r Ton the left) commuting with the action of Hm,r and the images of Uq(Qln) and Hm,r in End#(®A€A/n wm] x\Hm,r) are centralizers each other, i.e., both maps (5.9.1) #m,r->End^(fl-jj( ^ xxHm,r) and A€A(n,r)M C/g(g\) - EndHm,r( 0 xAi7m,r) A€A(n,r)[H are surjective. Proof. Let / C iifm,r be the annihilator of the right i7m,r-module Tp. The natural map Hm,r —> Ends(r)(Tr) factors through the quotient Hm,r/I. Moreover, the induced map is injective. We claim that (5.9.2) dim#m,r//= dim Ends(r)(Tr). In fact, by Proposition 4.7 and Schur's lemma (using the semi-simplicity of Hmr, thus the semi-simplicity of S(T)) we have (5.9.3) dim End5(r)(Tr) = £*T'(X)2. xeT Now, we compute the dimension of the quotient algebra Hmtr/I. For A G T, there exist /i G T such that Tss(A,/x) ^ 0. Also note that TCS(A, A') ^ 0 (cf. 3.4). Since y^t = Tt,y\'Td(t), and TJ, v and Td(t) are invertible, Lemma 3.7(a) shows that xpHjn^T^^yx'T^ =fi 0 and the image of y^t in Ends(r)(Tr) is not zero. We claim that the image of the set (5.9.4) {y^lAer.s.tGT'CA')}
152 ZONGZHU LIN AND HEBING RUI in Ends(r)(Tr) is linearly independent. The claim implies that dimiifm,r// > EA€r(#TSW)2andhence dinLffm,r/J = dimEnds(r)(® sMflm,r)- Suppose s,teTa(\') where /*t G R. If not all /*t are zero, take v G f maximal among all A G f with f£ ^ 0 for some s,t G TS(A'). Then there exists /x G T such that Tss (*/,//) ^ 0. By [DJM, 4.14], the set {m^u | S G T"(A, /x), u G T*(A), A G A+(r)} is a basis of xMiifm,r, where (5.9.5) m^u= ]T 4eTr. vG/x-MS) Since i/ is maximal among all A, Lemma 3.7(b) implies rasu2/st ¥" 0 only if ^ ^ A. Thus, (5.9.6) 0 = (mSu) £ /i^= E /s't'msXt- A€r s,t€Ts(i/;) s,t€Ta(A') In order to compute msuVst^ by (5.9.5), we need to calculate x^uy^t. First, take So G Ts{v') such that Z(d(so)) is maximal among all s with f£ot ^ 0 for some t G Ts(uf). By Proposition 2.7(b), d(so)dK)-1 = W"1 where s^ G Ts(i/). Take u = s'0. Then for any s G Ts(i/) with /£' ^ 0, we have J(d(s)) + l(d((s0Y)) < Kwv)- BY Lemma 3.8, «„)?&' = 0 unless s = s0. Thus (ras,u)2£t = ° for all s ^ So and (5.9.6) becomes (5.9.7) 0= 53 /s^su^ot = £ f^T^x^y^y t6Ts(i/;) v€M~1(S),t€Ts(i//) Take to G Ts(v') such that /(d(to)) is maximal among all t with f£ot ^ 0. By Proposition 2.7(b) again, we have d(to)d(to)-1 = w^. By the anti-involution * and Lemma 3.8, yv'Td(t)T^tiQ)Xv = 0, unless t = t0. Letting T^,t,^xvTWuyv> G i/m,r act (on the right) on both sides of (5.9.7), we have 0= 2^ fs0t0Td(v)xvTWuyl/'TWu/xl/TWuyl/'. vG/x-MS) Using [DR2, 2.2], we can show that the i?-module x^Hrn^y^ is of rank one with a basis xvTw„yv>. Thus xvTWvyviTw^xvTw„yv> = g{q)xvTWuyv> for some g(q) G i?, and £ fsoto9(q)TdM^Tw„yv>=Q- u6Ts(i/) By Proposition 3.3(b), (5-9.8) f£tog(q) = 0.
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 153 Suppose that #m,r is semi-simple. Then J2 = J. Moreover, by [DR2, 2.2], In particular, xvTw^yv,TWij,xvTWuyv> ^ 0. Thus g(q) ^ 0, forcing f£to = 0 by (5.9.8), a contradiction. This completes the proof of (a). In (b), we take T = A(n,r)Im' and apply (a) to get the surjectivity of the map, Hmir -> End5m(n)r)(0M6A(nr)[m] x^Hrns). Now (b) follows from Proposition 5.7. D 5.10. We remark that (1) if n > r, then (lr, 0, • • • , 0) € A(n, r). In this case the annihilator / is zero and hence ifm,r = Endsm(n)r)(®A6A(nr)[m] x\Hm,r). This result was proved by Mathas in [M2, 5.3]. However, (lr, 0, • • • ,0) 0 A(n, r) if n < r. (2) The set f defined in definition 5.8 is not a co-ideal. For example, let r = m = 2 and n = 1. Obviously, A = ((11), 0) > (0, (2)) = \i. But A £ f since T3S(\,\i) = 0 and [x e T. Therefore, Theorem 5.9 are not included in [M2, 5.3] and Theorem 4.10. 5.11. Relation with g-Schur algebras. Following the settings in 5.2, we take T{n,r) = V®r for an n-dimensional C(g)-vector space V. On T(n,r) the Hecke algebra Hr acts from right and Sfo^r) = End#r (T(n,r)) is the g-Schur algebra. Hr is a subalgebra of i/m,r and Tm(n,r)^T(n,r)®HrHmir as iifm)r-modules. Now for any i7r-module map (j) : T{n,r) —► T{n,r), the map <t>m = 0 ® 1 : Tm(n,r) —> T^n^r) is a homomorphism of i7m)T.-module. Since Hm,r is free (thus faithfully flat) over Hr, we can regard S(n, r) as a subalgebra of Sm(n,r). Note that Hr has several different bases such as Kazhdan-Lusztig basis. Then tensoring over Hr with i/m,r, one can get a basis of i/m,r. Much of the setting in 5.2 can be done over R = TL\q, q~x\. Thus many results in [DPS2] can be extended to algebras Smfo, r), which carries a nice stratification although Sm{n, r) is not necessarily quasi-hereditary in general. More detailed information on the relations between S^n^r) and S(n,r) will be pursued later elsewhere. Finally we remark that the surjectivity of the map in (5.6.1) over C[g, q~l] was proved by Schiffmann in [Sch3]. Thus Proposition 5.7 holds over C[q,q~l). Therefore Sm(n,r) is a finite dimensional quotient of Uq(gln). In a following up paper we will study the detailed relation between the representations of Smin^r) and the finite dimensional representations of Uq(gin). References [Al] S. Ariki, Representations of Quantum algebras of type A\,_i and Combinatorics of Young Tableaux, Univ. Lect. Notes 26, American Mathematical Society, Providence, RI, 2002. [A2] S. Ariki, On the classification of simple modules for cyclotomic Hecke algebras of type <3(ra, 1, n) and Kleshchev multi-partitions. Osaka J. Math. 38 (2001), no. 4, 827-837. [A3] S. Ariki, Cyclotomic g-Schur algebras as quotients of quantum algebras. J. Reine Angew. Math. 513 (1999), 53-69. [AK] S. Ariki and K. Koike, A Hecke algebra of (Z/mZ) I <5r and the construction of its irreducible representations Adv. Math. 106 (1994) 216-243. [AM] S. Ariki and A. Mathas, The number of simple modules of the Hecke algebras of type <3(r,l,n). Math. Z. 233 (2000), no. 3, 601-623.
154 ZONGZHU LIN AND HEBING RUI [Ch] I.V. Cherednik, A new realization of Gelfand-Tzetlin bases, Duke. Math. J. 54 (1987), 563-577. [CPS1] E. Cline, B. Parshall and L. Scott, Finite-dimensional algebras and highest weight categories. J. Reine Angew. Math. 391 (1988), 85-99. [CPS2] E. Cline, B. Parshall and L. Scott, Stratifying endomorphism algebras, Mem. Amer. Math. Soc. 124 (1996), No.591. [D] V. Drinfeld, A new realization of Yangians and of quantum affine algebras. Dokl. Akad. Nauk SSSR 296, no. 1, 13-17 (1987); translation in Soviet Math. Dokl. 36, no 2, 212-216 (1988). [DJM] R. Dipper, G. James and A. Mathas, The cyclotomic g-Schur algebra, Math. Zeit. 229 (1998), 385-416. [DJMu] R. Dipper, G. James and G. Murphy, Hecke algebras of type Bn at roots of unity. Proc. London Math. Soc. 70 (1995), 505-528. [DJ1] R. Dipper, G. James, Representations of the Hecke algebra of the general linear groups, Proc. London Math. Soc. 52 (1986), 20-52. [DJ2] R. Dipper, G. James, ^-tensor spaces and g-Weyl modules, Trans. Amer. Math. Soc. 327 (1991), 251-282. [DM] R. Dipper, and A. Mathas, Morita equivalence of Ariki-Koike algebras, Math. Zeit, 240 (2002), no. 3, 579-610. [DIR] V. Dlab, and C. Ringel, Quasi-hereditary algebras, Illinois J. Math. 33 (1989), no. 2, 280-291. [DPS1] J. Du, B. Parshall and L. Scott, Quantum Weyl reciprocity and tilting modules. Comm. Math. Phys. 195 (1998), no. 2, 321-352. [DPS2] J. Du, B. Parshall and L. Scott, Stratifying endomorphism algebras associated to Hecke algebras. J. Algebra. 203 (1998), 169-201. [DR1] J. Du and H. Rui, Ariki-Koike algebras with semisimple bottoms, Math. Zeit. 234 (2000), no. 4, 807-830. [DR2] J. Du and H. Rui, Specht modules for Ariki-Koike algebras, Comm. Algebra 29 (2001), 4710-4719. [GV] V. Ginzburg, E. Vasserot, Langlands reciprocity for affine quantum groups of type An, Intemat. Math. Res. Notices, (1993), 67-85. [GL] J. Graham and G. Lehrer, Cellular algebras, Invent. Math. 123, (1996), 1-34. [Gr] R.M. Green, The affine g-Schur algebra. J. Algebra 215 (1999), no. 2, 379-411. [HS] J. Hu and F. Stoll, On double centralizer properties between quantum groups and Ariki-Koike algebras. J. Algebra 275 (2004), no. 1, 397-418. [J] M. Jimbo, A ^-analog of C/(0(n+i), Hecke algebra, and the Yang-Baxter equation, Lett, in Math. Phys., 11 (1986), 247-252. [Lu] G. Lusztig, Transfer maps for quantum affine $ln, Representations and quantizations (Shanghai, 1998), 341-356, China High Educ. Press, Beijing, (2000), [Ml] A. Mathas, Tilting modules for cyclotomic Schur algebras, J. Reine Angew. Math. 562 (2003) 137-169. [M2] A. Mathas, The representation theory of the Ariki-Koike and Cyclotomic g-Schur algebras, Representation theory of algebraic groups and quantum groups, 261-320, Adv. Stud. Pure Math., 40, Math. Soc. Japan, Tokyo, 2004. [R] C. Ringel, The category of modules with good filtrations over a quasi-hereditary algebra has almost split sequences. Math. Z. 208 (1991), no. 2, 209-223. [SS] N. Sawada and T. Shoji, Modified Ariki-Koike algebras and cyclotomic g-Schur algebras. Math. Z. 249 (2005), no. 4, 829-867. [Schl] O. Schiffmann, The Hall algebra of a cyclic quiver and canonical bases of Fock spaces. Intemat. Math. Res. Notices 8 (2000), 413-440. [Sch2] O. Schiffmann, Quivers of type A, flag varieties and representation theory, Representations of finite dimensional algebras and related topics in Lie theory and geometry, 453-479, Fields Inst. Commun., 40, Amer. Math. Soc, Providence, RI, 2004. [Sch3] O. Schiffmann, Algebres affines quantiques aux racines de l'unite et K-theorie equivariante, C.R. Acad. Sci. Paris Ser. I Math. 327 (1998) 433-438. [W] M. Varagnolo, E. Vasserot, On decomposition matrices of the quantized Schur algebra, Duke Math. J., 100 (1999), 267-297.
CYCLOTOMIC g-SCHUR ALGEBRAS AND SCHUR-WEYL DUALITY 155 Department of Mathematics, Kansas State University,, Manhattan, KS 66506, USA E-mail address: zlinQmath.ksu.edu Department of Mathematics, East China Normal University, Shanghai,200062 China E-mail address: hbruiQmath.ecnu.edu.cn
This page intentionally left blank
Contemporary Mathematics Volume 413, 2006 Geometric Crystals and Affine Crystals Toshiki Nakashima 1. Introduction Theory of crystal base is initiated by Kashiwara, which is now widely applied to many areas in mathematics and mathematical physics. The notion of geometric crystals and unipotent crystals has been introduced by Berenstein and Kazhdan([l]) for reductive algebraic groups and it is extended to Kac-Moody setting in [12]. It seems to be a geometric lifting of the Kashiwara's crystal base theory. They are related to each other by "tropicalization/ultra-discretization" procedures. Schubert varieties/cells associated with Kac-Moody groups have a canonical geometric/unipotent crystal structures ([12]). We shall see that the geometric crystal on Schubert varieties/cells and tensor product of some crystals are related by tropicalization/ultra-discretization procedures. One of the most significant results of geometric crystal theory is that if (X, 7, {e;}) is a geometric crystal, there exists a rational Weyl group action on X. As a result, we obtain a rational Weyl group action on Schubert variety/cell (see 5.2). Perfect crystals([6],[7]) are defined for quantum affine algebras, which play an important role in studying vertex type solvable lattice models. Certain limit of perfect crystals (denoted by J3oo) have been treated in [5]. Some affine geometric crystal can be realized in fundamental representation of affine Lie algebra ([8]). We shall see that for the affine A type, the crystal -B<x> is obtained from the above affine geometric crystal by the ultra-discretization procedure. In this article, first we review the basic definitions and properties of geometric crystals following [12] and in the last section we see the explicit relation between the affine geometric crystal in the fundamental representation V{w\) and the crystal BooforsU+ias]). 2. Kac-Moody groups and Ind-varieties We review Kac-Moody groups and ind-varieties following [10],[11],[13]. 2.1. Kac-Moody algebras and Kac-Moody groups. Fix a symmetriz- able generalized Cartan matrix A = (aij)ijej, where / is a finite index set. Let (t, {ai}iei, {hi}iei) be the associated root data, where t is the vector space over 1991 Mathematics Subject Classification. 17B37, 17B67,46E25, 20C20. Key words and phrases. Geometric crystal, crystal, tropicalizat ion, ultra-discretization. The author was supported in part by JSPS Grants in Aid for Scientific Research #16540039. ©2006 American Mathematical Society 157
158 TOSHIKI NAKASHIMA C with dimension |/|+ corank(^l), and {(*i}iei C t* and {hi}iej C t are linearly independent indexed sets satisfying ctj(hi) = a^. The Kac-Moody Lie algebra q = q(A) associated with A is the Lie algebra over C generated by t, the Chevalley generators e^ and fi (i G /) with the usual denning relations ([10],[11]). There is the root space decomposition q = 0a€t* &*• Denote the set of roots by A := {a G t*|a ^ 0, ga ^ (0)}. Set Q = J^Za*, Q+ = ]T\ Z>oai and A+ := A D Q+. An element of A+ is called a positive root. Define simple reflections Si G Aut(t) (i G /) by Si(h) := h — ai(h)hi, which generate the Weyl group W. We also define the action of W on t* by Si(\) := A — \(hi)ai. Set Are := {w(ai)\w G W, 2 G /}, whose element is called a real root. Let gf be the derived Lie algebra of q and G the Kac-Moody group associated with g'([ll]). Let Ua := expga (a G Are) be an one-parameter subgroup of G. The group G is generated by Ua (a G Are). Let U*1 be the subgroups generated by U±a (a G Ar+e = Are n Q+), i.e., U± := (U±a\a G Ar+e). For any i G /, there exists a unique homomorphism; fa : SL2(C) —> G such that fa 11 0 j J J = exptei, fa ( ( t i )) = exp^/i (t G C). Set Xi(t) := exptei, yi(t) := exptfi, Ti := </>;({diag(£,£_1)|£ G C}) and Ni := NGi(Ti). Let T (resp. N) be the subgroup of G generated by T; (resp. Ni), which is called a maximal torus in G and B± = U±T be the Borel subgroup of G. We have the isomorphism <j>: W-^N/T defined by (/>(si) = NiT/T. An element S; := Xi(-l)2fc(l)xi(-l) is in NG(T), which is a representative of s; G W = NG(T)/T. Define R(w) for w € W by R(w) := {(il,22,"" ,*«) e /'|w = S^Sij,- "SiJ, where I is the length of w. We associate to each w G W its standard representative w G Nq(T) by w = s^Siz "-sin for any (21,22, ••• >*0 € #(w). 2.2. Ind-variety and Ind-group. Let us recall the notion of ind-varieties and ind-groups. (see [9]). Definition 2.1. Let k be an algebraically closed field. (i) A set X is an ind-variety over k if there exists a filtration Xq C X\ C X2 C • • • such that (a)Un>0*™=*- (b) Each Xn is a finite-dimensional variety over fc such that the inclusion Xn <-+ Xn+\ is a closed embedding. (ii) A Zariski topology on an ind-variety X is defined as follows; a set U C X is open if and only if U D Xn is open in Xn for any n > 0. (iii) Let X and F be two ind-varieties with nitrations {Xn} and {Yn} respectively. A map / : X —> Y is a morphism if for any n > 0, there exists m such that f(Xn) C Fm and /xn : Xn —> Fm is a morphism. A morphism / : X —> y is said to be an isomorphism if / is bijective and /_1 : F —> X is also a morphism. (iv) Let X and F be two ind-varieties. A rational morphism f : X —> V is an equivalence class of morphisms /f/ : £/ —> F where U is an open dense subset of X, and two morphisms fu:U—>Y and fy : V —> F are equivalent if they coincide on 17 D V.
GEOMETRIC CRYSTALS AND AFFINE CRYSTALS 159 Definition 2.2. An ind-variety H is called an ind {algebraic)-group if the underlying set H is a group and the maps HxH —► H H —► H (x,y) i—> xy x i—> x_1 are morphisms of ind-varieties. We have the following facts: (i) A finite dimensional variety over k holds canonically an ind-variety structure, (ii) If X and Y are ind-varieties, then IxFis canonically an ind-variety by taking the nitration (X x Y)n := Xn x Yn. (iii) Let G be a Kac-Moody group and U±, B± be its subgroups as above. Then G is an ind-group and U±, B± are its closed ind-subgroups. (iv) The multiplication maps TxU —> B U~ xT —> B~ (t,u) i—► tu (vit) •—> vt are isomorphisms of ind-varieties. 3. Crystals In this section we review the theory of crystals, which is the notion obtained by abstracting the combinatorial properties of crystal bases. Definition 3.1. A crystal B is a set endowed with the following maps: wt : B —>P, Si'.B—>ZU{-oo}, (fiiB—>ZU{-oo} for iel, ^ : B U {0} —► B U {0}, /; : B U {0} —► B U {0} for i € /, e;(0) = fi(0) = 0. those maps satisfy the following axioms: for all 6, &i, b2 G B, we have ^t(6)=et(6) + (fei,^(6)>, wt(eib) = wt(b) + a; if iib G J5, w*(/i6) = wt(b) - oti if fib G B, e{b2 = bi<=> fibx = b2 (61, b2 G B), Si(b) = -00 =» ei6 = /;6 = 0. The following tensor product structure is one of the most crucial properties of crystals. Theorem 3.2. Let Bi and B2 be crystals. Set Bi <g) B2 := {bi <g> b2\ bj G Bj (j = 1,2)}. Then we have (i) J5i <g) B2 is a crystal. (ii) For b\ G B\ and b2 G B2, we have f(h <*h\-l fibl ® b2 if ^bl"> > Si^2^
160 TOSHIKI NAKASHIMA { eibi ® b2 if (pi(bi) > ei(b2), Definition 3.3. Let B\ and B2 be crystals. A strict morphism of crystals ^ : J5i —> B2 is a map ^ : J5i U {0} —► J52 U {0} satisfying: ^(0) = 0, if) commutes with all e; and fi and if b € B\ and ^(6) G B2, then Wt(^(b)) = Wt(b), SiWb)) = Si{b), ^(6)) = <pi(b). In particular, a bijective strict morphism is called an isomorphism of crystals. Example 3.4. // (L, B) is a crystal base, then B is a crystal. Hence, for the crystal base (L(oo), J5(oo)) of the nilpotent subalgebra U~(g) of the quantum algebra Uq(g), J5(oo) is a crystal. Example 3.5. For i e I, the crystal Bi := {(x)i : x € Z} is defined by: wt((x)i) = xoti, Si{{x)i) = -x, <Pi((x)i) = x, ej((x)i) = -°°> Vj((x)i) = -OO for j ^ 2, €j(x)i = Sij(x + l)i, fj(x)i = 6ij(x - 1);. Note that as a set Bi is identified with the set of integers Z. Example 3.6. For A € P, set T\ := {£A}- We define a crystal structure on T\ by ei(t\) = fi{t\) = 0, ei(t\) = <pi(*A) = -00, wt(t\) = A. 4. Geometric Crystals and Unipotent Crystals In this section, we define geometric crystals and unipotent crystals associated with Kac-Moody groups, which is just a generalization of [1] to the Kac-Moody setting ([12]). 4.1. Geometric Crystals. Let (aij)ijei be a symmetrizable generalized Car- tan matrix and G be the associated Kac-Moody group with the maximal torus T. An element in Hom(T,Cx) (resp. Hom(Cx,T)) is called a character (resp. co-character) of T. We define a simple co-root otf € Hom(Cx,T) (i € /) by &i(t) := T{. We have a pairing (aV', otj) = a^. Definition 4.1. (i) Let X be an ind-variety over C, 7 : X —► T be a rational morphism and {e^}^/ be a family of rational C-actions e* : Cx x X^X (i€ J); ei : Cx x X —> X (c,x) *-+ ef(x). The triplet x = P^7> {ei}ie/) is a geometric pre-crystal if it satisfies {1} x X C dom(ei), e1(x) = x and (4.1) 7(e?(x))=aY(c)7(x). (ii) Let (X,7x, {e^}iG/) and (V,7y, {ef }zg/) be geometric pre-crystals. A rational morphism / : X —> V is a morphism of geometric pre-crystals if / satisfies that foef= e( o /, 7x = 7y ° /• In particular, if a morphism / is a birational isomorphism of ind-varieties, it is called an isomorphism of geometric pre-crystals.
GEOMETRIC CRYSTALS AND AFFINE CRYSTALS 161 Let x = (-X'j 7> {^i}iei) be a geometric pre-crystal. For a word i = (n, 22, • • • , k) £ R(w) (w e W), set a(z) := ain a^_1) := ^(a^J, •••, a(1) := sit •••5i2(ail). Now for a word i = (h,i2,"' ,h) € i?(iu) we define a rational morphism ei : Definition 4.2. (i) A geometric pre-crystal \ is called a geometric crystal if for any iu € W, and any i, i' € ,R(it;) we have (4.2) Ci = Ci/. (ii) Let (X, 7x, {e*}^/) and (V,7y, {ef }i£/) be geometric crystals. A rational morphism / : X —► Y is called a morphism (resp. an isomorphism) of geometric crystals if it is a morphism (resp. an isomorphism) of geometric pre-crystals. The following lemma is a direct result from [1] [Lemma 2.1] and the fact that the Weyl group of any Kac-Moody Lie algebra is a Coxeter group [2] [Proposition 3.13]. LEMMA 4.3. The relations (4.2) are equivalent to the following relations: e^ef =efe? ifaij=aji = 0, e?efc*e? = efe^ef if ^ = aji = -1, e?efc *e?*e? = ef^ef^ if ^ = -2, aji = -1, e?ef*ef*ef*e?<»e? = e?e?c>efch?C2efC2e? if ^ = -3, aj{ = -1, Remark. If a^a^ > 4, there is no relation between e; and ej. 4.2. Unipotent Crystals. In the sequel, we denote the unipotent subgroup [/+ by U. We define unipotent crystals (see [1],[12]) associated to Kac-Moody groups. DEFINITION 4.4. Let X be an ind-variety over C and a : [/ x I -^ I be a rational [/-action such that a is denned on {e} x X. Then, the pair X = (X, a) is called a U-variety. For [/-varieties X = (X, ax) and Y = (V, ay), a rational morphism / : X —► Y is called a U-morphism if it commutes with the action of U. Now, we define a [/-variety structure on B~ = U~T. As in Sect.2, B~ is an ind-subgroup of G and hence an ind-variety over C. The multiplication map in G induces the open embedding; B~ xU «-* G, which is a birational isomorphism. Let us denote the inverse birational isomorphism by g\ g:G —> B" x U. Then we define the rational morphisms n~ : G —► J5~ and 7r : G —► U by 7r~ := projB- og and 7r := proj^ og. Now we define the rational [/-action aB- on J5~ by aB- := 7r~ o m : U x J5~ —> J5~, where m is the multiplication map in G. Then we get [/-variety B~ = (J5~, aB-). Definition 4.5. (i) Let X = (X, a) be a [/-variety and / : X —> B~ a [/-morphism. The pair (X, /) is called a unipotent G-crystal or, for short, unipotent crystal.
162 TOSHIKI NAKASHIMA (ii) Let (X, fx) and (Y, /y) be unipotent crystals. A [/-morphism g : X —► V is called a morphism of unipotent crystals if fx = fy° 9- In particular, if # is a birational isomorphism of ind-varieties, it is called an isomorphism of unipotent crystals. We define a product of unipotent crystals following [1]. For unipotent crystals (X, fx), (Y, /y), define a morphism axxY :UxXxY^XxYby (4.3) <*xxY(u,x,y) := (ax(u,x),aY(ir(u> fx(x)),y)). If there is no confusion, we use abbreviated notation u(x,y) for axxy(w,x,y). Theorem 4.6 ([1]). (i) The morphism ctxxY defined above is a rational U-morphism on X xY. (ii) Let m: B~ x B~ —> B~ be a multiplication morphism and f = fxxY • X x Y —► B~ be the rational morphism defined by fxxY :=mo(fx x /y). Then fxxY is a U-morphism and (X x Y, /xxy) is a unipotent crystal, which we call a product of unipotent crystals (X, fx) and (Y, /y). (iii) Product of unipotent crystals is associative. 4.3. From unipotent crystals to geometric crystals. For i e I, set Uf" := £/* H SiU+s-1 and Ul± := £7* n s^s"1. Indeed, Uf = U±Cii. Set Y±ai := (x±ai(t)Uax±ai(-t)\t e C, a e Ar±e \ {±a*}>- We have the unique decomposition; U~ = Ur. Y±ai = U-ai ■ UL. By using this decomposition, we get the canonical projection & : U~ —> I7_a.. Now, we define the function on U" by Xi:=y2"1o^:C/--^C/_aiJ^C, and extend this to the function on B~ by %i(^ * t) := Xi(u) f°r u £ ^~ and t ET. For a unipotent G-crystal (X, fx), we define a function Si := ef : X —► C by ^ :=Xi°fx, and a rational morphism 7x : X — > T by (4.4) lx := projT o fx : X -► J5" -> T, where projT is the canonical projection. Remark. Note that the function £; is denoted by <pi in [1],[12]. Suppose that the function e% is not identically zero on X. We define a morphism e* : Cx x X -> X by (4.5) e?(*):=*«(^)(aO. Theorem 4.7 ([1]). For a unipotent G-crystal (X, fx), suppose that the function Si is not identically zero for any i e I. Then the rational morphisms ^x • X —> T and e* : Cx x X —► X as a&oue de/me a geometric G-crystal (X,jx,{ei}iei), which is called the induced geometric G-crystals by unipotent G-crystal (X, fx)-
GEOMETRIC CRYSTALS AND AFFINE CRYSTALS 163 Note that in [1], the cases Si = 0 for some i e I are treated by considering Levi subgroups of G. But here we do not treat such cases. The following product structure on geometric crystals are most important results in the sense of comparison with the tensor product theorem in Kashiwara's crystal theory (Theorem 3.2). Proposition 4.8. For unipotent G-crystals (X, fx) and (Y, fy), set the product (Z, fz) := (X, fx) x (Y, fy), where Z = XxY. Let (Z, jz, {^i}) be the induced geometric G-crystal from (Z, fz)- Then we obtain; (i) jz = mo(7x x-yy). (ii) For each i e I, (x, y) e Z, (4.6) ef(x,y)=e?(x) + ^{y) OLi{lx{x))' (iii) For any i e I, the action e; : Cx x Z —> Z is given by: e^(x,y) (ef 0*0,<2(y)), where /.7x „ cai(^x(x))ef(x)+eY(y) <Xi(lx{x))e? (x) + ej (y) (4.7) ci = —-— x —yTT' C2 = ai(jx(x))ef(x) + el(y) ' ai(7x(x))ef (x) + c^ej(y) Here note that c\C2 = c. The formula c\ and c2 in [1] seem to be different from ours. 5. Geometric crystal structure on Schubert varieties 5.1. Highest weight modules and Schubert varieties. As in Sect.2, let G be a Kac-Moody group, B± = U±T (resp. U±)be the Borel (resp. unipotent) subgroups in G and W be the associated Weyl group. Here, we have the following Bruhat decomposition and Birkhoff decomposition; Proposition 5.1 ([9],[11],[13]). We have (5.1) G= [j B+wB+ = [j U+wB+ (Bruhat decomposition), wew wew (5.2) G = [j B~wB+ = \J U~wB+ (Birkhoff decomposition). wew wew Let J C / be a subset of the index set / and Wj := (si\i G J) be the subgroup of W associated with J. Set Pj := B+WjB+ and call it a (standard) parabolic subgroup of G associated with J C I. We denote the set of the minimal coset representatives of W/Wj in W by WJ. There exist the following parabolic Bruhat/Birkhoff decompositions: Proposition 5.2 ([9],[11],[13]). Let J be a subset of I and, Wj and WJ be as above. Then we have G= [j U+w*Pj, G= [j U~w*Pj. w*ewJ w*ewJ
164 TOSHIKI NAKASHIMA 5.2. Unipotent crystal structure on Schubert variety. For AeP+ (P+ is the set of dominant integral weights), let us denote an integral highest weight simple module with the highest weight A by L(A)([2]) and its projective space by P(A) := (L(A) \ {0})/Cx. Let vA € P(A) be the point corresponding to the line containing the highest weight vector of L(A) and set X(A):;=G-^ACP(A). Set Ja := {i G I\(hi,A) = 0}. By Proposition 5.2 and the fact that PjA is the stabilizer of v\, we have the isomorphism between X(A) and the flag variety G/PjA: Proposition 5.3 ([11],[13]). There is the following isomorphism and the decomposition; P'-G/PjA=UweWjAU±wPjA/PjA ^ X(A) 9-Pja >-* 9-va Definition 5.4. We denote the image p(U+wPjA/PjA) (resp.p(U~wPjA/PjA)) by X(A)W (resp. X(A)W) and call it a finite (resp. co-finite) Schubert cell and its Zariski closure in P(A) by X(A)W (resp. X(A)W) and call it & finite (resp. co-finite) Schubert variety. The names "finite" and "co-finite" come from the fact dimX(A)™ = l(w), codimx(A)X(A)™ = l(w). Indeed, X(A)W = Cl(w\ There exist the following closure relations; (5.3) X(A)W = [J X(A)y, X(\r= [J X(A)v. y<w,y€;WJ& y>w,yeWJ& Indeed, by [9, 7.1,7.3], (5.4) X(A)W and X(A)W are ind-varieties. Let us associate a unipotent crystal structure with X(A)W. Since by the definition of X(A)W and Proposition 5.3, we have X(A)W = U+w • v\ and Lemma 5.5. Schubert cell X(A)W is a U-variety. Next, let us construct a [/-morphism X(A)W —► B~. For that purpose, we consider the following: let w = s^ Si2 • • • s;fe be a reduced expression and set Uw = U D wU'w-1 and Uw = U D wUw"1. Define /?i = a^, 02 = Sti(ai2)> *•• iPk = s^Siz •••5ife_1(aiJ, then we have Uw := Upx - Up2 --Upk. This is a closed subgroup of U and we have an isomorphism of ind (algebraic- varieties ([13]) (5.5) Uw * Up, x U& x • • • x U0k * C\ by (5.6) Uw-w = UailSi! • C/ai2si2 • • • • f/aife5ife-^>Ck Xi^ai)^ 'Xi2(a2)si2 Xik(a>k)Sik h+ (ai,a2,-** ,flfc).
GEOMETRIC CRYSTALS AND AFFINE CRYSTALS 165 Lemma 5.6 ([13, 2.2]). For any w G WJa (A G P+), there exists an isomorphism of ind (algebraic)-varieties S:UW ^ X(A)W U I—► U ' V\ Define an isomorphism of ind (algebraic)-varieties C: X(A)W -^ Uww v \-> £(v) := S"1^)™, where w G WJa and A G P+. Since X(A)W is [/-orbit of p(w • Pja/Pja), U acts rationally on X(A)W. We denote the action of x G U on v G X(A)W by x(v). Lemma 5.7. The isomorphism £ : X(A)W —> Uww is a U-morphism. Define a rational morphism fw : X(A)W —> J5~ by /^ = 7r~ o £. The following is one of the main results of this article. Theorem 5.8. For A e P+ and w e WJa, let X(A)W be a finite Schubert cell and fw : X(A)W —► B~ be as defined above. Then the pair (X(A)W, fw) is a unipotent G-crystal. In the sense of Definition 4.5(ii), £ is an isomorphism of unipotent crystals on X(A)W and Uww. Since X(A)W <-+ X(A)W is an open embedding, they are birationally equivalent. Let u) : X(A)W —> X(A)W be the inverse birational isomorphism. Thus, fw := fwou :X(A) w —> B is a [/-morphism. Then we have Corollary 5.9. Let X(A)W be a finite Schubert variety and fw be defined as above. Then the pair (X(A)W, fw) is a unipotent G-crystal. Remark. Note that for all w < wf, we have the closed embedding X(A)W <-+ X(A)W' ([13]), and the isomorphism X(A) -^ lim X(A)W. wewJA Nevertheless, in general, we do not obtain a unipotent crystal structure on X(A) by using this direct limit since for y < w, the rational morphism fw : X(A)W —► B~ is not denned on X(A)y. 5.3. Geometric Crystal structure on X(A)W. As we have seen in 3.3, we can associate geometric crystal structure with the finite Schubert cell (resp. variety) X(A)W (resp. X(A)W) since we have seen that they are unipotent G-crystals. By Theorem 4.7, we have Theorem 5.10. For w e W, suppose that I = I(w). We can associate the geometric G-crystal structure with the finite Schubert cell X(A)W (resp. variety X(A)W ) by setting (see (4.4) and (4.5)) jw := projT o fw (resp.%, := projT o /w), e^x) = x{ ( -—r J (x), where projT :B~ = U~T^>T. We denote this induced geometric crystal by (X(A)w,jw, {e;}i€/) (resp. (X(A)w,yw,{ei}ieI)). This geometric/unipotent crystal (X(A)w^w,{ei}ieI) is realized in B~ in the following sense.
166 TOSHIKI NAKASHIMA Proposition 5.11. For w — Si1 ''' Sik, define B- := {Yw(ci, • • • , cfc) := Y^a) • • • Yik(ck) G B~\a G Cx}. where Yi(c) = yi(^)a^(c) and U-actions on B~ by u(Yw(cu • • • , ck)) :=■ tt~(u • yw(ci, • • • , ck)) (u G U). ThenX(A)w andB~ are birationally equivalent via fw and isomorphic as unipotent crystals. Moreover, they are isomorphic as induced geometric crystals. 6. Tropicalization of Crystals and Schubert Varieties 6.1. Positive structure and Ultra-discretizations/Tropicalizations. Let us recall the notion of "positive structure" ([1],[12] ). The setting below is simpler than the ones in ([1],[12] ), since it is sufficient for our purpose. Let T = (Cx)1 be an algebraic torus over C and X*(T) = l) (resp. X*(T) = Zz) be the lattice of characters (resp. co-characters) of T. Set R := C(c) and define v: R\{0} —► Z /(c) » deg(/(c)). Here note that for /i, /2 € R \ {0}, we have (6.1) v(hf2) = v(h) + v(f2), v(^=v(h)-v(f2) Let / = (/i, • • • ) fn) '■ T —► T" be a rational morphism between two algebraic tori T = (Cx)m and T' = (Cx)n. We define a map / : X*(T) -» X^T) by (/(0)(c):=(c^(«W),...,c^^)))), where ^ G X*(T). Since v satisfies (6.1), the map / is an additive group ho- momorphism. Identifying X*(T) (resp. X*(r/))with Zm (resp. Zn) by {(c) = (cZl,--- ,cz™) 4-> (h,-- ,lm) eZm, we write /(Jl,-, im) := (V(/1«(C))), ■ ■ • , V(fn(£(c)))). A rational function /(c) G C(c) (/ ^ 0) is positive if / can be expressed as a ratio of polynomials with positive coefficients. Remark. A rational function /(c) G C(c) is positive if and only if /(a) > 0 for any a>0 If /i, f2 € R are positive, then we have (6.1) and (6.2) v(h + /2) = max(t;(/i), v(/2)). Definition 6.1 ([8]). (i) A non-zero rational function on an algebraic torus T is called positive if it is written as g/h where g and h are a positive linear combination of characters of T. (ii) Let /: T —> X" be a rational morphism between two algebraic tori T and X". We say that / is positive, if \ o / is positive for any character x- T" —> C. Denote by Mor+(T, T') the set of positive rational morphisms from T to T'. Lemma 6.2 ([1]). For any positive rational morphisms f G Mor+(Ti,T2) and g G Mor+(T2, T3), tfie composition g o f is in Mor+(Ti, T3). By Lemma 6.2, we can define a category T+ whose objects are algebraic tori over C and arrows are positive rational morphisms.
GEOMETRIC CRYSTALS AND AFFINE CRYSTALS 167 Lemma 6.3 ([1]). For any algebraic tori T\, T2, T3, and positive rational mor- phisms f e Mor+(Tl5T2), g e Mor+(T2,T3), we have go f — go J. By this lemma, we obtain a functor UD: T+ —* 6et T ^ X*(T) (f:T^V) ^ (/: X*(T) - X*(T'))) Definition 6.4 ([1]). Let \ = P^7> {ei}iei) be a geometric crystal, V be an algebraic torus and 6 : V —► X be a birational isomorphism. The isomorphism 9 is called positive structure on \ if it satisfies (i) the rational morphism 7 o 6 : V —► T is positive. (ii) For any i G I, the rational morphism e^ : Cx x T" —> X" denned by eito(c, t) := 0_1 o e\ o 0(£) is positive. Let 0 : T —► X be a positive structure on a geometric crystal \ = (X, 7, {^i}i€/})- Applying the functor UD to positive rational morphisms e^g : Cx x T' —> T' and 7 o 0 : T" —> T (the notations are as above), we obtain Ci := W%^):ZxI,(T)^I*(T) 7 := UD{1oe):X*{T')^X*{T). Now, for given positive structure 6 : X" —► X on a geometric pre-crystal % = (X, 7, {ei}i€/), we associate the triplet (X*(X"),7, {e;}i€/) with a free pre-crystal structure (see [1, 2.2]) and denote it by UDe^T'ix)- By Lemma 4.3, we have the following theorem: Theorem 6.5. For any geometric crystal \ = (X, 7, {e;};€/) and positive structure 0 : T' —► X, the associated pre-crystal UDq^'{x) = (-^*(Tr/), -7, {e~i}iei) is a free W-crystal (see [1, 2.2]) We call the functor UD uultra-discretization" instead of "tropicahzation" unlike in [1]. And for an object B in 6et, if there exists a geometric crystal x, an algebraic torus T in T+ and a positive structure 6 on \ such that UDe^ix) — B as crystals, we call x a tropicahzation of J5. Now, we define certain positive structure on geometric crystal B~ (I = I(w), and w e WJa) and see that it turns out to be a tropicahzation of (Langlands dual of) some Kashiwara's crystal. Let Bi (i e I) be the crystal defined in Example 3.5. For w = Si1Si2 • • • s;fe e W and i = (ii, 22, * * * , h) £ -R(w), we define the morphism 6\ : (Cx)fc —► J5~ by (6.3) <9i(ci,c2,--- ,ck):=Yil(c1)---Yik(ck) = yil(—)<(ci) •• •?/;*(—)<*£(<*) Proposition 6.6. (i) For an?/ i e R(w) (w e W, I(w) = I), the morphism 6\ defined in (6.3) is a positive structure on the geometric crystal B~. (ii) Geometric crystal B~ is a tropicalization of the Langlands dual of the crystal Bix ®Bi2 <g) • • '<8>Bik with respect to the positive structure #i(ci, c2, ••, ck), or equivalently UD(B~) = Langlands dual(Bi1 ® • • • <g) Bik) as crystals. Indeed, we have 7(yii(—X(ci)---yife(—X(cfc)) =a2vl(ci)---a2vfe(cfc),
168 TOSHIKI NAKASHIMA and the explicit action of e\ on Yw(ci, • • • , c^): e?(Yw(cir- ,cfc)) =Xi ( /v , rr j (K;(ci,--- ,cfc))) — ^(Ci,-" A), \^t(>w(Cl,--- >c/c))/ where (6.4) Cj:=Cj ^ /."*!•* ...rairn-1,ir *-* rail,i ...raim'1,ir l<m<j,im=i cl cm-l c™ j<m<fc,im=i cl Wn-1 c™ 2L< ai-i «<m-i.* l<m<j,im=i Cl Cm-1 cm j<m<ife,im=i cl cm-l c™ Furthermore, we describe the action of e\ on Bix (g> • • • <g> £?ife. Take &i = (fci)^ 0 • • • (g) (6fc)ifc (i = (ii, • • • , 2/c)> bj G Z). Since the action of e* on tensor products is described explicitly in [3], we obtain e2c(6i) = (/?i)i10---0(Ak5 where / /3j = bj + max V (6.5) —max max (c-bm-^bidi^), max (-bm - ]P btai4l) 1<™<7, /<m j<m<fc, j<m i<m=i i<m=i j ( \ max (c-bm-^T M;,;*), max (-6m - ]P Mmi) l<m<j, j<m j<m<k, i<m Now, we know that (6.4) and (6.5) are related to each other by the tropicaliza- tion/ ultra-discretization operations: _ ultra—discretization tropicahzation Cj -< >. ft^ X • y *e >- X + y y •< *» x-y x + y ■< >■ max(x, y) CLi^j ■< ' ' >■ Qij,i Langlands dual 6.2. Rational Weyl group action on Schubert variety. Let (X, {ei},7) be a geometric crystal. Due to Lemma 4.3, Si(x) — e?^7^" (x) (x e X) defines a rational Weyl group action on X. Applying this to £?~, the rational Weyl group action on Schubert cell/variety is given by: Si(Yw(ci, • • • , Cfc)) =: Yw(du '" ,dk)
GEOMETRIC CRYSTALS AND AFFINE CRYSTALS 169 where ^ raii'i...raim-1'ir ^ ^ cl cra—1 rn l<m<j, j<m<k, *«m+l.« ...„a*fe^ (Lj '•— C j « V 1 + V c caim+1,'...caifc,i 1^ ^ ! *"Cm-l cm 7. Afflne geometric crystal for s[n+i In this subsection, we see an application of ultra-discretization of affine geometric crystal of type An (see [8]). 7.1. Perfect crystals and their limit. Perfect crystals are defined for quantum affine algebras and they play an important role in studying solvable lattice models([6],[7]). In [5], certain limit of perfect crystals are introduced, which is denoted by B^. Let q be an affine Lie algebra and Pci be a classical weight lattice and set (Pd)t := {A G Pci\(c,\) = I, (hi, A) > 0} (Z G Z>0) where c is a canonical central element of q. Definition 7.1. A crystal B is a perfect of level I if (i) B <g> B is connected. (ii) There exists \q G Pci such that wt(B) C A0 + ]T Z<0au tt^Ao = 1 (iii) There exists a finite dimensional U'q(%)-module V with a crystal pseudo-base Bp3 such that B ^ Bps/±1 (iv) We have e, <p : Bmin := {b G B\(c,e(b)) = l}-^(P^)i (bijective). Now let us define the limit of perfect crystals. Let {Bi}i>\ be a family of perfect crystals of level Z and set J := {(/, b)\l > 0, b e B^™}. Definition 7.2. A crystal B^ with an element b^ is called a limit of {Bi}i>i if (i) wt(bao) = eiboo) = tpiboo) = 0. (ii) For any (Z, b) G J, there exists an embedding of crystals: ^e(fe) ® b ® t-^b) h+ b^ (iii) Boo = U0>6)€JIm/(W- (As for the crystal T\, see Example 3.6.) If a limit exists for a family {£*}, we say that {Bi} is a coherent family of perfect crystals. Let jB(oo) be the crystal of the subalgebra U~(q). Then we have the isomorphism of crystals: 5(oo)AB(oo)0 5oo.
170 TOSHIKI NAKASHIMA In the case q = sln+ij the crystal B^ is given as follows ([5]): Boc := {^ = K ^2, • • • , vn)Wi e Z}(= Zn) e0H = (i/i - l,i/2, •••)> ^(^) = (•••, I/* + 1, ^i+i - 1, • • •) (i = 1, • • • , n), /; = e~-\ I Si{v) = Vi+i (i/n+i := -(i/i + • • • + i/n)) [wt(i/) = (-i/i + 2/„+i)A0 + ]Cr=i(2/* ~ ^i+i)Ai5 where A; is a fundamental weight. 7.2. Geometric Crystal in Fundamental Representation. Let w\ = Ai — Ao be a fundamental weight of level 0 for sln+i and W(tui) be a fundamental representation of sln+i with a generator iz^ (see [4]). For w* = si • • • sn_isn, set K,* = {v(xir- ,Xn) ^yi^iJ-'-yn^nJlX^ G W(tx7i) |xi • • • , Xn G CX}. On Vu;*, we can define the following An -geometric crystal structure: e^v{x) = v(- - • , cXi, • • •) if i ^ 0, ec0v(x) = v(—, —,-•-, —) if i = 0, c c c e0(v(x)) = xi, 6i(v(x)) = -^1 (1 < i < n),en(v(a;)) = —, X{ Xji 1 X2 7o(v(s)) = —. 7iM*)) = -1 (7t(v(x)) = ay(7(v(x)))), £l#n %2 X2 ~2 7i(v(*)) = ! (K * < n), in(v(x)) = Now, we consider the following positive structure on Vw*: 6: (Cx)n — Vw* (h, hr" > 'n) ^ v(Zi, Zi/2, * * * ,h • • • in). Through this 6, on (Cx)n we obtain ei,e(' "•>'*» '<+i» •••) = (■■•> c'<> c'^t+i, •■• )(* = !>■••, 7i - 1), 6i(l) = Z»+i (0 < 2 < n), en(Z) = Zn+i, 70(0 = ¥1, 7i(0 = 7^-(l<<<n), 7n(0 = 7^-- *1 H+l ln+l where /n+i = , 1, . On the other hand, on B^ we have feg(l/i, I/2, * * * , Vn) = {y\ ~ C, l/2, ' ' ' , Vn) e* (• • • , i/», i/»+i, ••■) = (••• ,v% + c, i/i+i - c, • • •) (i = 1, • • • , n - 1), ,en(l/i,--- ,l/n_i,l/n) = (i/i,-"' ,Vn-l,Vn + c) £i(i/) = i/i+i (0 < i < n), ^o(^) = -v\ + ^n+i, ^(i/) = I/* - i/i+i (1 < i < n), wtn(i/) = vn - i/n+i,
GEOMETRIC CRYSTALS AND AFFINE CRYSTALS m where wti(v) = (hi,wt(u)). Therefore, we have (7.1) W2?e(V^) = *oo. References [1] Berenstein A. and Kazhdan D., Geometric crystals and Unipotent crystals, GAFA 2000 (Tel Aviv,1999), Geom. Funct. Anal. 2000, Special Volume, Part I, 188-236. Kac V.G., Infinite dimensional Lie algebras 3rd ed., Cambridge University Press. Kashiwara M., Crystal base and Littelmann's refined Demazure character formula. Duke Math. J. 71 (3), 839-858 (1993). Kashiwara M., On level-zero representations of quantized affine algebras, Duke Math.J., 112(2002), 117-175. Kang S-J., Kashiwara M. and Misra K.C., Crystal bases of Verma modules for quantum affine Lie algebras, Compositio Mathematica 92 (1994), 299-345. Kang S-J., Kashiwara M., Misra K.C., Miwa T., Nakashima T. and Nakayashiki A., Affine crystals and vertex models, Int.J.Mod.Phys.,A7 Suppl.lA (1992), 449-484. Kang S-J., Kashiwara M., Misra K.C., Miwa T., Nakashima T. and Nakayashiki A., Perfect crystals of quantum affine Lie algebras, Duke Math. J., 68(3), (1992), 499-607. Kashiwara M., Nakashima T. and Okado M., Affine geometric crystals and limit of perfect crystals, math.QA/0512657. Kumar S., Kac-Moody groups, their Flag varieties and Representation Theory, Progress in Mathematics 204, Birkhauser Boston, 2002. Kac V.G. and Peterson D.H., Defining relations of certain infinite-dimensional groups; in "Arithmetic and Geometry"(Artin M.,Tate J.,eds), 141-166, Birkhauser, Boston-Basel- Stuttgart, (1983). Peterson D.H., and Kac V.G., Infinite flag varieties and conjugacy theorems, Proc.Nat.Acad.Sci.USA, 80, 1778-1782, (1983). Nakashima T., Geometric crystals on Schubert varieties, Journal of Geometry and Physics, 53,197-225 (2005). Slodowy P., On the geometry of Schubert varieties attached to Kac-Moody Lie algebras, Can.Math.Soc.Conf.Proc. on 'Algebraic geometry' (Vancouver) 6, 405-442, (1986). Department of Mathematics, Sophia University, Kioicho 7-1, Chiyoda-ku, Tokyo 102-8554, Japan E-mail address: toshikiQmm.sophia.ac.jp
This page intentionally left blank
Contemporary Mathematics Volume 413, 2006 Self-extensions for finite symplectic groups via algebraic groups Cornelius Pillen Dedicated to James E. Humphreys on the occasion of his 65th birthday Abstract. For large primes it was proved in [BNP1, BNP3] that a finite group of Lie type does not admit self-extensions, i.e. non-trivial extensions of a simple module with itself, unless the group is one of the symplectic groups Sp2n(^p), n > 1. In this paper it is shown that self-extensions indeed exist for these groups for all ranks and odd primes. The method of proof is based on ideas due to James Humphreys and Henning Andersen. Some of the results in this paper assume the Lusztig Conjecture. 1. Introduction 1.1. Let G be a connected simply connected almost simple algebraic group denned and split over the field ¥p with p elements, and k be the algebraic closure of Fp. Let G(¥q) be the finite Chevalley group consisting of the Fq-rational points of G where q = pr for a positive integer r. Moreover, let Gr be the rth Probenius kernel. In [Huml] J.E. Humphreys constructed examples of self-extensions, i.e. non- trivial extensions of a simple module with itself, for the finite symplectic groups Sp4(¥p) with p odd. In the same paper Humphreys conjectured that the root systems of type Cn, n > 1, might be exceptional for the existence of self-extensions. Humphreys' conjecture was motivated by a theorem of H.H. Andersen [Andl] that says that Probenius kernels do not admit self-extensions of simple modules unless the underlying root system is of type Cn (n > 1) and the prime is two. It is well-known that the algebraic group does not admit self-extensions. In [BNP2] the following generalization of Andersen's result was given: Given a pair of simple G-modules with p-restricted weights A and \i that are "close", i.e. (A - /x,av) < p/3 for any root a, then Ext^1(L(A),L(/x)) = Ext^(L(A),L(/x)), unless G is of type Cn. However, the existence of "close" pairs of weights with ExtQl(L(X),L(fi)) ^ Ext^(L(A),L(/x)) for type Cn has not been established for n> 2. 2000 Mathematics Subject Classification. Primary 20C, 20G; Secondary 20J06, 20G10. ©2006 American Mathematical Society 173
174 CORNELIUS PILLEN In [Huml] such pairs of weights also appear in the construction of self-extensions for Sp4(¥p). The results in [BNP1, BNP3] imply that such pairs of weights with non-vanishing G\-extensions are necessary in order to find self-extensions for finite groups of Lie type. In particular, for large primes it was shown that self- extensions can only exist for the finite symplectic groups Sp2n (Fp), thus confirming Humphreys' conjecture. The purpose of this paper is to prove the existence of self-extensions for symplectic groups of arbitrary rank. The construction involves extensions between certain pairs of simple modules for an algebraic group G of type Cn of arbitrary rank. One of these modules has restricted highest weight while the other one is non-restricted. The restrictions of these extensions to the finite group G(¥p) then contain the desired self-extensions as submodules, while the restrictions to the first Probenius kernels result in the aforementioned "close" Gi-extensions (Proposition 2.4 and Corollary 5.1). One family of self-extensions described in this paper (Corollary 4.3(A)) was discovered by Tiep and Zalesskii in [TZ] via reduction modulo p of certain ordinary representations for Sp2n (Fp). Producing these extensions via the algebraic group Sp2n(k) yields additional new examples of self-extensions for Sp2n(Fp), especially for large primes (Corollary 4.3(B)). The nicest and most comprehensive results are obtained when assuming the Lusztig Conjecture. In Section 5 it is shown that all simple Sp2n (Fp)-modules whose highest weights are p-regular and adjacent to the hyperplane HaniP/2 = {xGRn | (x + p, 0%) = p/2} admit self-extensions. Recall that Andersen's theorem [Andl] says that ExtG1(L(X), L(X)) = 0, unless p = 2, G of type Cn, and A e #an,p/2- 1.2. Notation: G will always denote a connected simply connected almost simple algebraic group that is denned and split over the field Fp with p elements. k denotes the algebraic closure of Fp and Gr is the rth Probenius kernel of G. The conventions in the paper will follow the ones used in [Janl]. Let T be a maximal torus in G and $ the associated root system. The positive roots are denoted by $+ and the negative roots by $~. Let B be a Borel subgroup containing T and corresponding to the negative roots. X(T) denotes the weight lattice, X{T)+ the dominant weights, and Xr(T) the pr-restricted weights. For a weight 7 e X(T)+, iif0(7), V(j), and L(j) denote the induced module, the Weyl module, and the simple module, respectively. With the exception of Sections 2.1 and 3.1 we will always assume that G is of type Cn. We follow [Bou, p.254] and denote the short simple roots by a* = e* — Ci+i, 1 < i < n, while an = 2en is the unique long simple root. The fundamental weights are uji = Y?k=i e^' with oj\ being the unique minuscule weight. For convenience we will frequently switch between the {ei}, {u>i}, and {o^} bases. The highest short root is ao = €1 + £2 and the longest element of the Weyl group W is — 1. The simple modules are therefore self-dual and H°(j) and V(j) are dual to each other. The following orders appear: A < /x, if \i — A is a sum of positive roots, \i <q A, if A — /jl is a linear combination of positive roots with non-negative rational coefficients, and A | /x if there exist sequences of weights //i,//25 •••^m and reflections si, 52,..., sm+i such that A < si • A = \i\ < s2 • /xi = /X2 < ••• < sm • /xm_i = /xm < Sm+i -Hm=V>- (see [Janl, 11.6.4.(1)]).
SELF-EXTENSIONS FOR FINITE SYMPLECTIC GROUPS 175 2. Special G-extensions for algebraic groups of type Cn It is well-known that Ext^(Vr(/x),i7°(A)) = 0 for any pair of weights A,/x. It follows from [Andl] that Ext^l(Vr(A),i7°(A)) = 0 unless G is of type Cn, p = 2, and A is contained in the the hyperplane #an,p/2 = {# G Mn | (x + p, a^) = |}. In [BNP2, Prop. 5.2(a)] a generalization of Andersen's result for odd primes was found. It was shown that Ext^ (V(//),-ff°(A)) vanishes for a pair of restricted weights A and /jl that are "close", i.e. (/x — A, av) < p/3 for any root a, unless G is of type Cn and the weights are reflections of each other across the hyperplane HaniP/2. In this section we show that extensions for such "close" pairs of weights indeed exist for type Cn. 2.1. The following lemma is well-known. It is included for the benefit of the reader. We will make repeated use of it in later arguments. Lemma . Let i > 0 be a positive integer, a a simple root, V a rational G- module, and 7 G I(T) with —p< (7, av) < -1. Then Proof. Here P{ot) denotes the parabolic subgroup corresponding to the root a. We apply the spectral sequence [Janl, 1.4.5] Ext^V, R> ind£(Q) 7) =* Ex^j(V, 7). If (7)av) = -1 then i# ind£(a) 7 = 0 for all j > 0 [Janl, 11.5.2(b)], which forces Ext,B(1l/,7) = 0forain>0. Otherwise it follows from [Janl, 11.5.2(d)] that ExtB(V, 7) <* ExtJT^V, R1 ind£(a) 7). Now — p < (7, av) < —2 implies that 0 < (sa • 7, av) < p — 2. It follows from [Janl, 11.5.3(b)] that R1 ind£(a) 7 ^ ind£(a)(sa • 7). Finally, [Janl, 11.4.7(1)] yields Ext*B(V, 7) = Ext^^V, ind^(a)(5a • 7)) = Extjf ^ sa • 7). □ 2.2. The G-module L(uoi) = H°(uji) is multiplicity free with dimension 2n. The weights are expressed most conveniently in the form ±Ci with i = 1,..., n. Lemma . Let G be of type Cn, p odd, and 7 G X\{T) such that (7,a%) = (p — 2 ± l)/2 and san • (7 — pen) G Xi(T). Then the following hold: (a) Ext1B(V(san • (7 — pen)), 7 — pe^) 25 isomorphic to Ext^(L(5an • (7 - pen)), 7 - pe*) ^ j Jj ^= n (b) If i < n and /x G -X"(T)+ W2#i // < 7, ^en '* 2/(7,0 <P-2^-l<(7,<) + (7,<+i>, and fi = sai • (7-pCi), 0 e/se. (c) Ifi<n, then ExtB(y(sQ„ • (7 - pen)), 7 - pe*) = 0. Ext^(L(/x),7-p€i)^^
176 CORNELIUS PILLEN (d) Ifi<n, then Ext2B(L(san • (7 - pen)), 7 - pei) [Ext^(L(5an • (7 - pen)), H°(sai • (7 - pei))) if (7, aV> < p - 2, and p-l<<7,^} + <7,<i>, 0 e/se. = < Proof. Let V be a homomorphic image of V(/i) with // e X(T)+ and // < 7. Since 7 is restricted one has — p < (7 — pe*, c^) < —1 and it follows from Lemma 2.1 that ^HomB(F, 5ai • (7 - pe*)) else. Since V is a highest weight module, the homomorphism group vanishes unless ^ = s^ ' (7 — P^i) and fi is dominant. Part (a) follows. Next, assume i < n and // = sQi • (7 — pe*). Then (sa. • (7 — pei)),aj) = {l-pei + (p-1 - (7, <#))<**> a#) > 0 for all j^i + 1, while (sQi • (7-pe;)), <*i+i) = (7-^i+(p-l-(7,aV))ai,<1) = (7,0 + (7,aV+1) + l-p.If (7,aV) + (7,aV+1) < p — 1 then sai - (7 — pe^ fails to be dominant and the expression vanishes. This implies part (b). From Lemma 2.1 one obtains \Extk(V,sai (7-^i)) else. We have seen that sai • (7—pe*) is dominant if and only if (7, 0%) + (7, c*/+1) > p — 1. In this case one obtains Ext#(V, sa. • (7 — pei)) ^ Ext#(V, H°(sai • (7 — pe*))). If V = V(san • (7 — pen)) the expression vanishes due to [Janl, II.4.13]. Finally assume (7, a/) -f (7, a^+i) < p — 1 and /x = sotn • (7—pen). Then Lemma 2.1 yields Ext^(V, sai • (7 - pe^) ^ HomB(V; sQi+1sai • (7 - pei)) if san < (7 - pen) = sai+1sai • (7 - pei) else. {: Recall that san • (7 - pen) = 7 ± en. If sQn • (7 - pen) = sat+1sai • (7 - P**) then 5a.+1-(7±en) = sai'(lf-p€i). This implies that 7±en-((7±en,aV+1) + l)ai+i = 7 — pei -h (p — 1 — (7, a^))ai, which forces n~1 n±l (p - 1 - (7, o%))ai + ((7 ± en, a,v+1) + l)ai+i = pe* ± en = p(]T afc) + —^<*n Comparing the coefficients for oti shows that this is impossible. Parts (c) and (d) follow. □ 2.3. Here it is shown that exceptions to [BNP2, Prop. 5.2] indeed exist for groups of type Cn. Proposition . Let G be of type Cn, p odd, and A e Xi(T) with (\,o%) = (p - l)/2, then (a) Ext^(F(A - |an), H°(X) ® L^)^) * fc,
SELF-EXTENSIONS FOR FINITE SYMPLECTIC GROUPS 177 (b) //Ext^(L(A — \an), A — pti) = 0 for all i < n, then Ext^(L(A - \an), H°(X) ® L^)^) * k. Proof. Let V e {V(\-\an), L(X-\an)}. The G-module L{u\) is multiplicity free with weight spaces ie*, i = 1,..., n. Let S denote the J3-submodule consisting of the weight spaces — ci,..., — en, R be the J3-submodule consisting of the weight spaces — ei,..., —en_i, and Q the J3-quotient L(ui)/S. The weight spaces of Q are ^li •••5 ^n* The short exact sequence 0 —> 5 —> L(tc>i) —> Q —► 0 yields the exact sequence H6mB(V;A(8)QW) -> Ext^(V, A® 5(1)) -> Ext^(Vi A ® LK)(1)) -> Ext^(V, A ® Q*1*). The first term in this sequence is zero because A — \oin is not a weight of A ® Q^\ The last term vanishes because the height of a weight of A ® Q^ is greater than the height of a weight of V [Janl, 11.4.10(b)] . Therefore Extk(V,A ® i(wi)(1)) = Ext^(V, A ® S(1)). Next we use the short exact sequence 0 —► R —> 5 —> —en —> 0 to obtain Extk(V,A<8>#(1)) -> Ext^(V,A®5(1)) -> Ext^(V, A - pen) -> Ext|(V, A ® #(1)) Recall that sotn • (A — pen) = A — en = A — \an. Lemma 2.2(a) implies that the first term in the above sequence is zero and the last term vanishes by Lemma 2.2(c) and the assumption in part (b). One concludes from Lemma 2.2(a) and [Janl, 11.4.7(1)] that Ext^(V, H°(X) ® L(wi)(1)) = Ext^(V, A ® L(wi)(1)) ^ Ext^(V, A - pen) ^ k. D Corollary . Let G be of type Cn, p odd, and A G Xi(T) with (\,a„) = (p - l)/2, then HomG(Ext^ (V(X - ±an), ff°(A)), L(u>i)W) <* k. Proof. Consider the Lyndon-Hochschild-Serre spectral sequence E? =ExfG/Gi(Ext^(F(A- ±o*),H°(\)W<<>i)ll)) =» Ext%j(V(\ - \an), H°(X) ® L^)^). Since HomGl(L(A - ±an),H°(\)) = 0, we have E^° = E%° = 0 and from the corresponding five-term sequence E1 = E%yl. n 2.4. We establish the existence of certain non-trivial G-extensions between simple G-modules, one with restricted highest weight and the other non-restricted. These will later yield the self-extensions for the finite symplectic groups. Proposition . Let G be of type Cn, p odd, and X € Xi(T) with (A,c#) = (p — l)/2. In addition, assume that (i) (A,aV) + <A,aty+1> < p - 1, for 1 < i < n - 1, (ii) H°(X) and H°(X — \an) have only p-restricted composition factors, then Ext^(L(A - \an), L(X) ® L(vi)W) = k.
178 CORNELIUS PILLEN Proof. It follows from Lemma 2.2(d) and Proposition 2.3(b) that Ext^(L(A- |an),i7°(A) <g) L(ui)^) = k. Therefore there exists a composition factor L(/x) of H°(X) such that Ext^(L(A - \an), L(/x) <g> L{u{)^) ^ 0. We make use of the short exact sequence 0 —> R —> F(A — |an) —> L(A — ^an) —> 0 to obtain the exact sequence HomG(fl, L(/x) ® L^1)) - Ext^(L(A - \an)), L(/x) 0 L^1)) - Ext^(F(A-ian)^(/^)^^i)(1))- The first term vanishes because \i and the composition factors of R are restricted. The last term is isomorphic to Ext^(L(/x), H°(X-±an)®L(uJi)W) ^ Ext^(L(/i), A- |an (g) L^i)^1^). This implies that Ext#(L(/x),A — \an ± pe^ =fi 0 for some i. Hence Ext#(L(/x),A — ^an + pei) =fi 0 is not possible by height comparison. If Ext#(L(/x), A — \an —pei) ^ 0, Lemma 2.2(b) yields i = n and 2.2(a) forces \i = A (recall that sQn • (A — \an) = A). □ 3. Constructing self-extensions via the algebraic group Here we use methods due to Andersen to generalize Humphreys' construction of self-extensions. 3.1. In this section we allow for algebraic groups G other than type Cn. The associated finite groups of Lie type obtained as fixed points of the r-th Probenius morphism twisted by an automorphism a coming from an automorphism of the Dynkin diagram are denoted by Ga(¥q), where q = pr (see [Jan2, 1.3]). The following is a generalization of work by H.H. Andersen [And2, Prop 2.7]. If p > 2(/i — 1) the set 7T\ can be replaced by the set of pr-bounded weights. Proposition . Let A,/x e Xr(T) with A jtq /jl. Set tt\ = {v = u0 +prvx \ u0 e Xr(T) and vx G X(T)+ such that z/0 - A ^ prrj for any rj e X(T)+}. Then 0 HomG(L(/x), L(u0) ® L(pv{)) ® Ext^(L(i/), L(A)) -> Ext^(Fg)(L(/x), L(A)). i/€tta Proof. Set A = (pr — l)p + wqX. If prj is a dominant weight of HomGr (L(i/0), Str ® L(X)) S HomGr (L(i/0) ® L((pr - l)p - A), Str), then one of the weights appearing in L(v0) ® L((pr — l)p — A) is (pr — l)p + pr7. This forces i/0 — A > pr7, which is not allowed. Hence (3.1.1) HomGr(L(i/0), Str ® L(A)) = 0. Moreover (3.1.2) HomG(L(i/),Str(g)L(A)) ^HomG(L(i/i)(r\HomGr(L(i/0),Str(g)L(A))) =0. The Steinberg module is injective as a Gr-module. It follows from the five-term- exact sequence of the Lyndon-Hochschild-Serre spectral sequence and (3.1.1) that (3.1.3) Ext^(L(i/), StP®L(A)) ^ Ext^/Gr(L(i/i)(r), RomGr(L(v0), StP®L(A)) = 0. Define Q via the exact sequence of G-modules (3.1.4) 0 -» L(X) -* StP ® 1(A) -» Q -» 0.
SELF-EXTENSIONS FOR FINITE SYMPLECTIC GROUPS 179 Prom the corresponding long exact sequence as well as (3.1.2) and (3.1.3) one obtains that (3.1.5) Ext^(L(i/), L(A)) * HomG(L(i/), Q). Next, we use the injectivity of Str as a Ga(¥q)-module and weight comparison to argue, by [Jan2, Satz 1.5]), dimHomGCT(Fg)(L(/x), Str ® L(A)) = [L(jjl) ® L((pr - \)p - A) : Str]GCT(Fg) = Yl [^)^^((Pr-l)p-A)0L(a7):Str0L(7)(r)]G. 7€X(T) + The last expression vanishes unless \i > A + (pr — a)j > A. Now A ytq \i forces 7 = 0 and /x = A. One concludes that (3.1.6) dimHomG<T(Fg)(L(/x),Str(g)L(A) ^ The long exact sequence arising from (3.1.4) together with (3.1.6) and the injectivity of Str as a GCT(Fg)-module now imply (3.1.7) Ext^(Fg)(L(M),L(A)) S RomGAFq)(L(»),Q). Restriction from G to Ga(¥q) induces an embedding of (3.1.8) HomG(L(/x), L(v0) <g> L(<n/i)) -+ HomG<T(Fg)(L(/x), L(u0) <g> L(^i)) ^ HomGff(Fg)(L(/i),LK) ® L(^i)(r)) = HomGff(Fg)(L(/i),L(.)). The module ®„€7r £(^) 0 Hoiiig(£(^), Q) is a G-submodule of the G-socle of Q. It is also a GCT(Fg)-submodule of Q. Therefore 0 HomG(L(/x), L(i/0) ® L(<n/i)) 0 Ext^(L(i/), L(A)) ^ 0 HomG(L(/x),L(i/o)0L(ai/1))0HomG(L(i/),Q) (by (3.1.5)) — 0 HamG„(F,)(L(/i),L(i/))®HomG(L(i/),Q) (by (3.1.8)) - HomGCT(Fg)(L(/x),Q) - Ext^(Fg)(L(/x),L(A)) (by (3.1.7)). D Setting 7 = A = /x and applying Proposition 3.1 yields: Corollary . Le£ 7, u0 e Xr(T) and vx e X(T)+ such that v0 jt 7. // (i) Home(1/(7), L(i/0) ® L(avi)) =fi 0, and (ii) Ext^(L(7),L(i/0) ^ L(i/i)(r)) 7^ 0, ^enExt^(Fg)(L(7),L(7))^0. 4. Self-extensions for Sp2n(¥p) and odd primes. Here we use the algebraic group to construct families of self-extensions for Sp2n (¥p) including ones discovered by Tiep and Zalesskii (see [TZ, 3.18]). The examples exist for all odd primes. I fc if \i = A 10 else.
180 CORNELIUS PILLEN 4.1. In order to make use of Proposition 3.1 one needs to show that L(X) appears in the G-socle of L(X — \otn) <g) L(ui). Lemma . Let p be odd, G be of type Cn, and an be the unique long simple root. Assume that X € Xi(T) with (A, o^) = (p - l)/2. For all 1 < i < n - 1 with (A, a/) > 0 we assume in addition that (X + p, (a; + ... + an_i)v) ^ 0 (modp). Then HomG(L(A), L(X - \an) <g> L(u>i)) ^ 0. Proof. The tensor product H0(X-^an)<8>H°(uji) = H°(X-±an)<8>L(uji) has a nitration with factors -ff°(7) where 7 e S = {X±ei-en | 1 < i < n}nX(T)+. Notice that for 1 < i < n-2, A-(e;+en) = A-(a; + ...+an) € S if and only if (A, a() > 0. Assume that A—(ai+...+an) is dominant and that A—(a;+...+an) | A. This implies that a hyperplane of the form i/ai+...+arMmp = {x + p € Mn | (x, (a* + ... + an)v) = mp} lies between the two weights. One concludes that (A + p, (a^ + ... + an)v) = 1 (modp), which yields (A + p, (a* +... +an_i)v) = 0 (modp). Such weights A have been excluded. If (A, a^-i) = 0 then (^ + P-> (an-i + an)v) = P + 2. Moreover, (A + p, o^) = ^^ < p. Therefore neither A — (an_i + an) nor A — an is strongly linked to A. One concludes that A is minimal in the "j"-order among the weights in S. As a consequence of [Janl, II.4.18 and II.6.13] the module L(A) C H°(X) is a submodule of H°(X — |an) (g) L(o;i). At this point of the proof we know that L(X) is a submodule of H°(X — ^an) <g) L(cji) and we want to prove that it is actually in L(X — ^an) <g) L(o;i). It is clear that HomG(L(A),L(7) (g) L(c^i)) 7^ 0 for some composition factor £,(7) of H°(X — ^an)5 so we have to show that 7 = A — \an = A — en. For such 7 wehavedimHomG(Vr(7),i7°(A)(g)i70(cc;i)) = dimHomG(Vr(A),#°(7)(g)#Vi)) > dim Home(L(A),L(7) (g) L(c^i)) 7^ 0. This together with the fact that 7 | A — en forces 7 to be either equal to A — en or dominant of the form A — 6i = X — \an — (oti +... + an-i) for some 1 < i < n — 1. The latter implies that (A — \an + p, (a^ + ... + an_i)v) = 1 (modp), which forces (A + p, (a; + ... + an_i)v) = 0 (modp). Now the premises of the lemma says that (A, 0%) = 0. Hence 7 = A — e; is not dominant. This leaves 7 = A — en, as claimed. □ 4.2. Here we give sufficient conditions for the existence of self-extensions. Their usefulness will become apparent in the following sections. Proposition . Letp be odd, G be of type Cn, and an be the unique long simple root Assume that X € Xi(T) with (A, a^) = (p — l)/2. In addition, assume that (i) (A,aV) + (A,aV+1) <p- 1, for 1 < i < n - 1, (ii) H°(X) and H°(X— \cxn) have only p-restricted composition factors, (iii) For 1 < i < n - 1, (A, a/) > 0 implies that (X + p, (a; + ... + an_i)v) ^ 0 (modp), then Ext^2n(Fp)(L(A),L(A)) ^ 0 and Ext^n(Fp)(L(A - \an),L{X - \an)) + 0. Proof. Proposition 2.4 implies that Ext^(L(A - \an), L(X) ® L^1)) * Ext^(L(A), L(X - \an) ® L{u{)<U) * k. Lemma 4.1 yields HomG(L(A - \an), L(X) ® L(wi)) * HomG(L(A), L(A - \an) 0 L(wi)) ^ *. The assertion follows from Proposition 3.1. □
SELF-EXTENSIONS FOR FINITE SYMPLECTIC GROUPS 181 4.3. The weight ^y^n satisfies the conditions (i) through (iii) of Proposition 4.2 because (£=-^a;n, Oq) = p—1. This yields a family of self-extensions discovered by Tiep and Zalesskii (see [TZ, 3.18]). Their method of proof is quite different. They show that in order for certain irreducible p-modular representations to be lifted to characteristic zero the representation has to admit self-extensions. Corollary A follows now from a result due to Zalesskii and Suprunenko [ZS] that says that L(2^LOn — \ocn) and L(2^LOn) are reduction modulo p of irreducible complex Weil representations for Sp2n(^p)- Using the algebraic group to construct these extensions, has the advantage that additional families of self-extensions arise. Corollary (A). Letp be odd, G be of type Cn, an be the unique long simple root, and un the corresponding fundamental weight. Then (i) Ext^2n(Fp)(L(^u;n),L(^n)) t^O and (ii) Ext^n(Fp)(L(V^n - \*n),L{^un - \an)) ± 0. Assume that p > 2n and that A is p-regular, i. e. (A + p, av) ^ 0 (modp) for any root a. Clearly any p-regular weight satisfies condition (iii) of Proposition 4.1. Moreover, the weight A— \an is a reflection of A across the hyperplane {x+p, a^) = | that bisects the alcove containing A. It follows that A — \an is a p-regular weight inside the same alcove and the translation principle allows for the following simplification of condition (ii) in Proposition 4.2. Later we will see that conditions (i) and (ii) can be dropped if one assumes the Lusztig conjecture. Corollary (B). Let p >2n, G of type Cn. Assume that A e Xi(T) is a p-regular weight with (A,a^) = (p— l)/2, where an denotes the unique long simple root in the root system. Assume further that (i) (A, a{) + (A, aV+1) < p - 1 for all 1 < i < n - 1, and (ii) H°(X) has only p-restricted composition factors, then Ext^2n(Fp)(L(A),L(A)) + 0 and Ext^n(Fp)(L(A - \an),L{\ - \an)) + 0. 5. Self-extensions for Sp2n(^p) via the Lusztig conjecture 5.1. Lusztig conjecture and equivalences. Throughout this section we assume that G is of type Cn and p > h = 2n. Set T = {7 e X(T)+ | (7 + p, 0%) < p(p-h + 2)}. If A e Xi(T) with (A + p, av) <p(p-h + l) then the non-restricted weights A + pu\ and A — \an + pu\ are also contained in T. Furthermore, we assume throughout this section that the Lusztig conjecture, as stated in [Janl, II.8.22], holds for G. For any p-regular weight 7 e X(T) and any a e 3>+ there exists a unique integer na with nap < (7-i-p, av) < (na + l)p. As in [Janl, II.6.6] we set ^(7) = Ylae^+ n<*' A result due to Cline, Parshall, and Scott says that the Lusztig conjecture is equivalent to each of the following statements [CPS, 5.4]. (5.1.1) For 7,77 E r,Ext2G(L(7),#°(77)) ^ 0 implies d(7) - d(rj) = i (mod2). (5 1 2) For 7,r? G r' the natural maP E^gC^M'^)) -* Exth(L(^)^°(r])) is surjective. This will allow us to eliminate conditions (i) and (ii) of Proposition 2.4.
182 CORNELIUS PILLEN Lemma . Let G be of type Cn and p > 2n. Assume that A e X\(T) is a p- regular weight with (A, a^) = (p — l)/2, where an denotes the unique long simple root. Assume further that (A + p, a^) < p(p — fc + 1) and £/m£ tte Lusztig Conjecture holds for G. T/ienExt^(L(A-^an),L(A)(g)L(a;i)(1))^Ext^(L(A),L(A-^an)(g)L(cc;i)(1))^0. Proof. Proposition 2.3 and (5.1.2) reduce the assertion to Ext^(L(A — -an), A — pci) = 0 for all i < n. By Lemma 2.2(d) it is sufficient to show that Ext^(L(A— \ctn\ H° (sai- (X—pei))) = 0. We will show that d(X — \cxn) — d(sai • (A — pe^)) is even. The assertion then follows from the Lusztig Conjecture via (5.1.1). Observe that sai • (A — pti) = sai • A — pe^+i. The reflection sai permutes all the positive roots other than a^. Since A is p-restricted, one concludes that d(sai • A) = d(X) — 1. Next we compare d(sai • A) to d(sai • A — pei+\). For 1 <l <i one has (ei+i, (e* - ei+i)v) = -1 and (ei+i, (e* + e*+i)v) = 1. For i + 1 < I < n one has (e^+i,^ — ej)v) = 1 and (e^+i,^ + ej)v) = 1. All other short roots are perpendicular to ei+i. For the long roots one obtains (e;+i, (2e*)v) = 8i+\,i. It follows that d{s(Xi'\)—d{s(Xi'\—pei+\) = 2(n— (i+l)) — 1. Recall that A and A — \an are contained in the same alcove. One concludes that d(\— \ocn)—d(soti • (A—pe;)) = d(\) - d(sai • A) + d(sai • A) - d(sa. • A - pe^-i) = 2(n - (i + 1)) is even. n Using the same argument as in the proof of Corollary 2.3 one can show that exceptions to [BNP2, Thm 5.3(A) part (a)] indeed exist. Corollary . Let G be of type Cn and p > 2n. Assume that A e X\(T) is a p-regular weight with (A,a^) = (p— l)/2, where an denotes the unique long simple root. Assume further that (A + p, 0%) < p(p — h + l) and that the Lusztig Conjecture holds for G. Then HomG(Ext^(L(A - ±an),L(\)),L(ui)M) ^ Jfc. 5.2. With the help of Corollary 3.1, Lemma 4.1, and Lemma 5.1 we can now generalize Humphreys' Sp4(¥p) example to higher ranks, at least for p-regular weights. Proposition (A). Let G be of type Cn and p > 2n. Assume that A e Xi(T) is a p-regular weight with (A,a^) = (p — l)/2, where an denotes the unique long simple root. Assume further that (A + p, a^ ) < p(p — h + 1) and that the Lusztig Conjecture holds for G. ThenExt1Sp2n{¥p)(L(X),L(X)) + 0 andExt^n(Fp)(L(A-ian),L(A-ian)) ± 0. Not all p-singular weights adjacent to the hyperplane Han^p/2 will admit self- extensions. This can already be observed in the case of Sfp4(Fp), where self- extensions do not occur when A is contained in a ai-wall (see [And2, Note on p. 402]). Our methods show that self-extensions exist for all p-singular weights that are adjacent to Han^p/2, as long as they are not contained in any (a^ + ... + an_i)- wall, with 1 < i < n — 1. Proposition (B). Let G be of type Cn and p > 2n. Assume that A e Xi(T) is a p-singular weight with (A + p, (a^ + ... + an_i)v) ^ 0 (modp), for 1 < i < n — 1, and (A, o^) = (p — l)/2. Assume further that (A + p, Oq) < p(p — h + 1) and that the Lusztig Conjecture holds for G. ThenExt1Sp2n{¥p)(L(X),L(X)) ^ 0 andExt\p2n{¥p){L{X-\an),L{X-\an)) + 0.
SELF-EXTENSIONS FOR FINITE SYMPLECTIC GROUPS 183 Proof. For each a e $+ there exists a unique non-negative integer na such that nap < (A + p, av) < (na + l)p. Then CA = {7 G X(T)+ | nap < (7 + p, av) < (na + l)p for all a e 3>+} describes the alcove that contains A in its upper closure. Set Rx = {cx G $+ I (A + p, av) = (na + l)p}. Notice that the conditions on A imply that no root of the form a; +... + an_i is contained in Rx. At the same time one observes that the weight A — \an is contained in the upper closure of Cx if and only if Rx contains no elements of the form a* + ... + an-i with 1 < i < n — 1. Hence both A and A — \an lie in the upper closure of the same alcove. The alcove Cx is bisected by the hyperplane Han^p/2 = {x G Mn | (x + p, 0%) = p/2}. Since p > h there exists a p-regular weight p inside Cx- We denote its reflection across the hyperplane Han^p/2 by p. We choose a pair (p, p) with minimal distance p — p. This implies that (p + p, a^) = (p + l)/2. Lemma 5.1 implies that Ext^(L(/x),L(/i) <g> L(ui)W) = Exto(L(/i),L(/i +po;i)) ^ 0. We define EM via a non-split sequence 0 —> L(/i + po;i) —> ^ —> L(/x) —> 0. Clearly ^M embeds in H°(p + puj\). We set Ex = T^E^, where T* denotes the translation functor as defined in [Janl, II.7.6]. The exactness of the translation functor [Janl, II.7.7.6] yields an embedding of Ex in T*H°(p + pui) = H°(X - \an + pui). Moreover, since the weight A — ^an + puo\ is in the upper closure of the alcove containing p H-po;i, one concludes that E\ has exactly two composition factors, namely L(A) and L(A — \an +pwi). Therefore, Ext^(L(A - \an), L{\) 0 L^)^) * Ext^(L(A), L(X - ^an) 0 L^)^) ^ 0. The assertions follows from Lemma 4.1 and Corollary 3.1. □ References [Andl] H.H. Andersen, Extensions of modules for algebraic groups, Amer. J. Math., 106, (1984), 498-504. [And2] H.H. Andersen, Extensions of simple modules for finite Chevalley groups, J. Algebra, 111, (1987), 388-403. [Bou] N. Bourbaki, Groupes at algebres de Lie, Chaps 4-6, Hermann, Paris, 1968. [BNP1] C.P. Bendel, D.K. Nakano, C. Pillen, Extensions for finite Chevalley groups I., Adv. Math., 183, (2004), 380-408. [BNP2] C.P. Bendel, D.K. Nakano, C. Pillen, Extensions for finite Chevalley groups II., Trans. AMS, 354, (2002), 4421 -4454. [BNP3] C.P. Bendel, D.K. Nakano, C. Pillen, Extensions for finite groups of Lie type: twisted case, in Finite Groups 2003: Proceedings of the Gainesville Conference on Finite Groups, March 6-12, 2003, (2004), 29-46. [CPS] E. Cline, B. Parshall, L. Scott, Abstract Kazhdan-Lusztig theories. Tohoku Math. J. (2) 45, (1993), no. 4, 511-534. [Huml] J.E. Humphreys, Non-zero Ext1 for Chevalley groups (via algebraic groups), J. London Math. Soc, 31, (1985), 463-467. [Hum2] J.E. Humphreys, Generic Cartan invariants for Frobenius kernels and Chevalley groups, J. Algebra, 122, (1989), 345-352. [Janl] J. C. Jantzen, Representations of Algebraic Groups, Second edition, Mathematical Surveys and Monographs, 107 AMS, Providence, RI, 2003. [Jan2] J. C. Jantzen, Zur Reduktion modulo p der Charaktere von Deligne und Lusztig, J. Algebra, 70, (1981), 452-474. [TZ] P.H. Tiep, A.E. Zalesskii, Mod p reducibility of unramified representations of finite groups of Lie type, Proc. London Math. Soc, 84, (2002), 439-472. [ZS] A. E. Zalesskii, I. D. Suprunenko, Representations of dimensions (pn ± l)/2 of the symplectic group of degree 2n over a field of characteristic p, Vestsi Acad. Navuk BSSR Ser. Flz.-Mat., no. 6, (1987), 9-15.
184 CORNELIUS PILLEN Department of Mathematics and Statistics, University of South Alabama, Mobile, AL 36688, USA E-mail address: pillenQjaguarl.usouthal.edu
Contemporary Mathematics Volume 413, 2006 Classification of finite dimensional simple Lie algebras in prime characteristics Alexander Premet and Helmut Strade Abstract. We give a comprehensive survey of the theory of finite dimensional Lie algebras over an algebraically closed field of prime characteristic and announce that the classification of all finite dimensional simple Lie algebras over an algebraically closed field of characteristic p > 3 is now complete. Any such Lie algebra is up to isomorphism either classical or a filtered Lie algebra of Cartan type or a Melikian algebra of characteristic 5. Unless otherwise specified, all Lie algebras in this survey are assumed to be finite dimensional. In the first two sections, we review some basics of modular Lie theory including absolute toral rank, generalized Winter exponentials, sandwich elements, and standard nitrations. In Section 3, we give a systematic description of all known simple Lie algebras of characteristic p > 3 with emphasis on graded and filtered Cartan type Lie algebras. We also discuss the Melikian algebras of characteristic 5 and their analogues in characteristics 3 and 2. Our main result (Theorem 7) is stated in Section 4 which also contains formulations of several important theorems frequently used in the course of classifying simple Lie algebras. The main principles of our proof of Theorem 7, with emphasis on the rank two case, are outlined in Section 5. As suggested by the referee, we mention in Section 6 some interesting open problems related to the subject. We would like to thank the referee for careful reading and valuable comments. 1. The beginnings The theory of Lie algebras over a field F of characteristic p > 0 was initiated by Jacobson, Witt and Zassenhaus. In [J 37], Jacobson investigated purely inseparable field extensions E/F of the form E = F(ci,..., cn) where c? e F for all i < n. Although such field extensions do not possess nontrivial F-automorphisms, Jacobson developed for them a version of Galois theory. The role of Galois automorphisms in his theory was played by F-derivations. The set Der^ E of all F-derivations of E carries the following three structures: • a natural structure of a vector space over E, • a natural p-structure given by the pth. power map D i—► Dp, 1991 Mathematics Subject Classification. Primary 17B20, 17B50. Key words and phrases, finite dimensional simple Lie algebras. ©2006 American Mathematical Society 185
186 ALEXANDER PREMET AND HELMUT STRADE • a Lie algebra structure given by the commutator product. Let # denote the set of all subfields of E containing F and £ the set of all E- subspaces of Der^ E stable under the pth power map and Lie bracket in Der^ E. Both sets # and £ are partially ordered by inclusion. Given a subset X in Der^ E we let Ex denote the subfield of E consisting of all a G E satisfying x(a) = 0 for all x G X. Theorem 1 ([J 37]). The map £^Lh EL e $ is an order-reversing bisection between £ and #. Jacobson singled out the p-structure above as being of major importance for Lie theory. Definition 1 ([J 37]). A Lie algebra L over F is called restrictable if for any x G L the derivation (ad x)p of L is inner. Any restrictable Lie algebra L carries a p-mapping a; ^ x'p' which enjoys the three following properties: 1. (Ax)W =\px\p\ 2. (adx)p-adx^, = x^ + yM + Y%=i si(xi 2/)> where Si(x, y) G L are such that p-i ^isifay)?-1 = (ad(te + y))p"1(x) 2=1 (here x, y G L, A G F, and tisa variable). Such ap-mapping is uniquely determined up to a p-linear map from L into its center 3(L). It is therefore unique for any restrictable Lie algebra L with j(L) = (0). Once the mapping [p] is fixed, the pair (L, [p]) is called a restricted Lie algebra. If / is a restricted ideal of L, that is an ideal of L such that 1^ C /, then the quotient Lie algebra L/I carries a natural p-mapping given by (x + 1)^ = x^l + I for all x G L. We mention for completeness that the Lie algebras of linear algebraic groups over F are all equipped with canonical p-mappings, hence carry canonical restricted Lie algebra structures. ^From now on we assume that F is algebraically closed. Some time before 1939 Witt discovered (for any p > 3) a p-dimensional simple Lie algebra with no finite dimensional analogues in characteristic 0. The Witt algebra W(l;l) has basis {e_i, eo, ei,..., ep_2} over F and the Lie product in W(l;l) is given by \p pl = l 0"-0et+i if -l<^+J<P-2, L 2' jl \ 0 otherwise. As Witt himself never published his example, we have only indirect information about his discovery. Zassenhaus generalized Witt's example by considering a subgroup G of order pn in the additive group of F and by giving a pn-dimensional vector space Wq := 0.eG Feg a Lie algebra structure via [eg, eh] '= (h — g)eg+h for all ^, h G G. Such Lie algebras are often referred to as Zassenhaus algebras. In [Z 39], Zassenhaus investigated irreducible representations of nilpotent Lie algebras over fields of prime characteristics. This paper is the starting point of the modular representation theory of Lie algebras. In [Cha 41], Chang described all irreducible representations of the Witt algebra W(l;l). According to [Cha 41], Witt used the following realization of the Lie algebra W(l;l): Let 0(1; 1) denote the truncated polynomial algebra F[X]/(XP),
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 187 and let x be the image of X in 0(1; I). Give 0(1; 1) an algebra structure by setting {/>#} :— f(dg/dx) — g(df/dx) for all /,g G 0(1; 1). It is readily seen that the map e; i—> x%+1 extends to an algebra isomorphism W(l;l) -^ (0(1;I),{-, •}). For i G Fp set Ui — (1 +x)2+1. Then {v,i,Uj} = (j — i)ui+j for all i,j G Fp. This shows that W(l;l) is isomorphic to the Zassenhaus algebra associated with the additive subgroup Fp C F. 2. Some basics This section is a short introduction into the general theory of modular Lie algebras with emphasis on results and techniques used in Classification Theory. Most of the results discussed here are valid for any prime p. 2.1. Maximal tori in restricted Lie algebras. Let g be a restricted Lie algebra over F. An element x G g is called semisimple (respectively, nilpotent) if x lies in the restricted subalgebra of g generated by x^ (respectively, if x^e = 0 for e ^> 0). For any x G g there exist unique commuting xs and xn in g such that xs is semisimple, xn is nilpotent, and x = xs + xn. We denote by g% the set of all y G g such that (adx)dlm0(y) = 0, and define rk(g) := min{dimg° \x e g}. If dim g% — rk(g) then g% is a Cartan subalgebra of g (this is a standard fact of Lie theory). An element t G g is called toral if t^ = t. A restricted subalgebra t of g is called foraZ (or a fonzs of g) if the p-mapping is invertible on t. Any toral subalgebra of g is abelian and admits a basis consisting of toral elements. Set MT(g) := max {dim 111 is a torus in g}. A torus t of g is called maximal if the inclusion tct' with t' toral implies t = t'. The centralizer c0(t) of any maximal torus in g is a Cartan subalgebra of g and, conversely, the semisimple elements of any Cartan subalgebra of g lie in its center and form a maximal torus in g. The reader should be warned, however, that maximal tori (and their centralizers) in a restricted Lie algebra may have different dimensions (see [St 77]). In other words, there may exist maximal tori in g of dimension less that MT(g). Let t be a maximal torus of g, J) = c0(t), and let V be a finite dimensional restricted g-module (this means that py(x^) = pv(x)p fc>r any x G g where py denotes the corresponding representation). Since py(t) is abelian and consists of semisimple elements, V decomposes into weight spaces relative to t: V = 0 Va, Vx = {ve V\t.v = X(t)v V*Gt}. AGt* The set of t-weights {A G t* | V\ ^ 0} of V will be denoted by TW(V, t). It is worth mentioning that if t is a toral element of t then A(£) G ¥p for any A G TW(V, t). Set r(V, t) - rw(V, t) \ {0}. For Fp-independent linear functions /n,..., pk G r(V, t) define VX/ii,...,/!*) := 0 Vi1/il+...+ifc/ifc. (ii,...,*fc)€Fj The subspace V(p\,..., pk) is called a k-section of V. If V is an algebra over F (not necessarily associative or Lie) and g acts on V as derivations then V(p\,... ,/x/c) is a subalgebra of V. If V = g, the adjoint
188 ALEXANDER PREMET AND HELMUT STRADE 0-module, then r = r(g, t) is nothing but the set of roots of $ relative to t, and aer is the root space decomposition. A Cartan subalgebra f) of g is called regular if [) = c0(t) where t is a torus of maximal dimension in 0. By the main result of [P 86b], all regular Cartan subalgebras of g have dimension equal to rk($). Let [) = c0(t) be a regular Cartan subalgebra of q. In [Win 69], Winter proved that for any x G 07 satisfying x^ = 0 the exponential operator expadx G GL($) maps the root space decomposition of g relative to [) onto that of another regular Cartan subalgebra, denoted f)x. To appreciate this result one should keep in mind that in characteristic p the condition x^ = 0 does not always guarantee that expadx is an automorphism of g (for example, consider the case where q = W(l;l) and x = e_i). In [Wil 83], Wilson assigned a generalized exponential operator to any root vector x G 07 such that x^ G t. Inspired by Wilson's construction, the first author assigned generalized exponential operators to all root vectors in $7; see [P 86b]. Generalized exponential operators and resulting switchings of regular Cartan subalgebras in $ play an important role in Classification Theory. Let £ G Homirp (F, F) be such that £p — £ = Id.p. As F is algebraically closed, it is straightforward to see that £ : F —► F exists and is uniquely determined up to a linear map from F to Fp. Given x G 07, where 7 G T, we denote by m = m(x) the least positive integer k with x^ G t (such an integer exists because t is a maximal torus in g). Set q(x) - f ZTJi1 xW for m>l, q[X) ~ \ 0 for m = 1. Note that q(x) G i). Define the generalized Winter exponential Ex£ G GL($) by setting p—1 p—1 Ex,dv) = -E II ((a^r))+j)Idg - ad q(x)) (ad x)*(y) i=0 j=i+l for all y G 0a, where a G T U {0}, and extending to $ by linearity (our convention here is that Qo = *))• Notice that if x^ = 0 then 2?x>£ = expadx. In general, Ex£ is a polynomial in ad x; see [P 89]. According to [P 86b], f)x = Ex^) is a regular Cartan subalgebra of g and 9 = •)*© Y, ExA&cx) is the root space decomposition of g relative to f)x. For t G t set tx := t-7(t)(x + g(x)). The subspace tx = {tx \ t G t} coincides with the unique maximal torus in [)x; see [P 86b]. The set of roots r($,tx) of q relative to tx has the form r($,tx) = {ax?£ I a G T} C t* where <xx,z(tx) = a(t) - C(a(xWm)) 7(0 (Vtx G tx). If f)7 = j^ft) for some y G U7er £7? we sav that f)7 is obtained from f) by an elementary switching. By [P 89], any two regular Cartan subalgebras of Q can be
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 189 obtained each from another by a finite sequence of elementary switchings. This result has the following important consequence: PROPOSITION 2 ([P-St 99]). Let ti andt2 be two tori of maximal dimension in $, V a finite dimensional restricted ^-module, A$ = Fw(V,ti), and Qi the ¥p-span of Ai in t*, where i = 1,2. Then there exists a linear isomorphism of¥p-spaces ip : Qi —► Q2 such that ip(Ai) = A2 and dim V^ = dim V^^ for all \i G Ai. As a consequence one obtains that F* Si C Ai for some Si G i\ if and only if F* S2 C A2 for some S2 G t£. Also, 0 G Ai if and only if 0 G A2. 2.2. Absolute toral rank. It is often useful to view a Lie algebra as a sub- algebra of a restricted Lie algebra. Definition 2 ([St-F 88]). Let L be a Lie algebra. A triple (£,[p],i) where & is a restricted Lie algebra with p-mapping \p] : £ —► £ and i : L ^> L is an infective Lie algebra homomorphism, is called ap-envelope of L if the restricted Lie subalgebra of £ generated by i(L) coincides with £. The Lie algebra L is often identified with i(L) C £. We list below a few basic properties of p-envelopes. All proofs can be found in [St-F 88, St 04]. 2.2.1. Let (£, \p],i) and (£/, [p]',*') be two p-envelopes of L. Then there exists an isomorphism of restricted Lie algebras ip : L>/i{L>) —> £'/i{£j') such that iponoi = n'oi' where n and n' denote the canonical homomorphisms of restricted Lie algebras & -» £/j(£) and £/ -^ L'/i(L'). 2.2.2. Ap-envelope (£, [p],i) of L is called minimal if j(£) is contained inj(i(L)). Any L admits a minimal p-envelope, and any two minimal p-envelopes of L are isomorphic as ordinary Lie algebras. 2.2.3. Suppose L is semisimple. Then L has one "obvious" minimal p-envelope, namely, the restricted Lie subalgebra of Der L generated by ad L. This p-envelope is semisimple. Any two semisimple p-envelopes of L are isomorphic as restricted Lie algebras. Definition 3. Let (£, \p],i) be a p-envelope of L. The absolute toral rank of L, denoted TR{L), is the maximal dimension of tori in the restricted Lie algebra £/j(£). In other words, TR(L) := Mr(£/3(£)). In view of (2.2.1), this definition is independent of the choice of a p-envelope of L. For L semisimple, TR(L) = MT(LP) where Lp stands for the restricted Lie subalgebra of DerL generated by adL (see (2.2.3)). We shall need a few basic properties of TR(L) all of which can be found in [St 04]: 2.2.4. L is nilpotent if and only if TR(L) = 0. 2.2.5. If / is an ideal of L then TR(L/I) + TR{I) < TR(L). 2.2.6. Let T be a torus of maximal dimension in a finite dimensional p-envelope of L and let 71,..., 7^ be Fp-independent roots in r(L, T). Then T/J(L(7i,...,7fc))<fc. In particular, TR(L(a)) < 1 for any a G T(L,T) and TR(L(a,p)) < 2 for any two a, (3 G T(L, T).
190 ALEXANDER PREMET AND HELMUT STRADE 2.3. Sandwich elements. Given an arbitrary Lie algebra L over a field we define S(L) := {s G L | (ads)2 = 0}. The set S(L) plays a crucial role in Kostrikin's work on the restricted Burnside problem (see [Ko 90]). If 2L = L and s G S(L), then (ad s) (ad x) (ad s) = 0 for any x G L. Because of this property the elements of S(L) are often referred to as sandwich elements (the term is due to Kostrikin). As an example, S(W(1;I)) = ©2i>» ^ei- ^n general> S(L) is not closed under vector addition however. If 2L = L, then S(L) is closed under Lie multiplication (see [Ko 90] for more detail). Assume until the end of this subsection that char F = p > 2 and let L be finite dimensional over F. Let c G S(L) and x G L. Since (adc)(adx)(adc) = 0 we have ((adc)(adx))2 = 0. This implies that tr(adc)(adx) =0. As a consequence, S(L) is contained in the radical of the Killing form of L. The Lie algebras over F containing nonzero sandwich elements are called strongly degenerate (the term is due to Kostrikin). It follows from the preceding remark that the Killing form of any strongly degenerate simple Lie algebra over F is identically zero. By the Engel-Jacobson theorem, the linear span (S) of S = S(L) is a nilpotent Lie subalgebra of L. Since S(L) is invariant under all automorphisms of L the same is true for the normalizer of (S) in L. As a consequence, every strongly degenerate simple Lie algebra L contains a proper nonzero subalgebra invariant under all automorphisms of L. (This remark also shows that in characteristic 0 the equality S(L) = {0} is equivalent to the semisimplicity of L.) For p > 3, the Lie algebras L over F with S(L) = {0} are closely related to the Lie algebras of semisimple algebraic groups over F\ see the discussion in (3.1) for more detail. In [Ko-S 66], Kostrikin and Shafarevich conjectured that for p > 5 the nor- maliser of (S) in any strongly degenerate simple Lie algebra L is a maximal subalgebra of L. In his PhD thesis and a subsequent series of preprints, S. A. Kirillov verified this conjecture for all known finite dimensional simple Lie algebras of characteristic p > 3. Unfortunately, all attempts to find an a priori proof of the conjecture failed. 2.4. Standard filtrations. Let L be a simple Lie algebra over F and L(0) a maximal subalgebra of L. Let £(-i) be a subspace of L such that L(0) C L(_i) and [L(o),L(_i)] C £(-i), and assume further that L(_!)/L(0) is an irreducible L(0)-module. Following Weisfeiler [We 68] we define the standard filtration of L associated with the pair (L(0),Z/(_i)) by setting £(»+i) = {x£ L{i) | [x, L(_i)] C L(i)}, i > 0, Z/(_;_i) = [£(_»),£(_!)] + £(_»), i > 0. Since L(0) is a maximal subalgebra of L this filtration is exhaustive. Since L is simple, the filtration is separating. So there are si > 0 and 52 > 0 such that L = £(_5l) D ... I) I/(o) D ...D £(*a+i) = (°)- By construction, all subspaces L^ of L are invariant under the action of the restricted subalgebra of DerL generated by adL(o). A standard filtration is called long if L(i) ^ (0). Now let G = ®i€Z Gi be a graded Lie algebra, that is [Gi,Gj] C Gi+j for all i,j G Z. The following four conditions occur very frequently in Classification Theory: (gl) G-i is an irreducible and faithful Go-module; (g2) G.i = [G_i+i,G_i] for all i > 1;
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 191 (g3) if x G Gi, i > 0, and [x, G_i] = (0), then x = 0; (g4) if x G G-i, i > 0, and [x, Gfc] = (0) for all k> 0, then x = 0. The graded Lie algebra grL = ®*=_5 gr^Z/, where g^L = L(i)/L(i+1), corresponding to the standard filtration above satisfies the conditions (gl), (g2), (g3). The quotient of grL by its largest ideal contained in ^2i<_1 gr^ L satisfies all four conditions (gl) - (g4). 3. Classes of simple Lie algebras The main conjecture on the structure of finite dimensional simple Lie algebras over algebraically closed fields of characteristic p is known as the generalized Kostrikin—Shafarevich conjecture. It states the following: For p > b, any finite dimensional simple Lie algebra over F is either classical or isomorphic to one of the filtered Lie algebras of Cartan type. This conjecture is due to Kac [Kac 71, Kac 74] who formulated it for p > 3 (see also [Ko 71]). Our next goal is to give a detailed description of the Lie algebras mentioned in the generalized Kostrikin-Shafarevich conjecture. 3.1. Classical Lie algebras. Let $ be a simple Lie algebra over C, f) a Cartan subalgebra of $, $ = $($, [)) the corresponding root system, and A = {ai,..., on} a basis of simple roots in <I>. For q;,/3g $ set (/?, av) = 2(/J|a)/(a|a), where, as usual, (• I •) denotes a scalar product on the M-span of $ invariant under the Weyl group of $. Theorem 3 ([Che 56]). The Lie algebra g has a basis S = {ea I a G $} U {hi | 1 < i < 1} such that the following conditions hold: (1) [hi,hj}=0, 1 < t, J < /. (2) [fti,c/3] = (/3,aV)c/3, 1<*</, /?€*. (3) [ea,e-a] = ha is a Z-linear combination of hi,... ,hi. (4) Let a,(3 E &, (3 ^ ±a, and let {(3 — qa,..., /3 + ra} be the a-string through (3. Then [ea,ep] = 0 ifa+(3 & $ and [ea,ep] = ±(q+l)ea+p ifa+(3 G $. Moreover, q G {0,1,2} if a + (3 G $. The Z-span g% of B is a Z-form in g closed under taking Lie brackets. Therefore, 0f '•= Qz <8>z F is a Lie algebra over F with basis B(g)l and structure constants obtained from those for $% by reducing modulo p. For p > 3, the Lie algebra Qf fails to be simple if and only if the root system $ = $($, fj) has type A\ where I = mp—1 for some m G N. If $ has type ^4mp_i then Qf — sl(mp) has a one-dimensional center (consisting of scalar matrices) and the Lie algebra Qf/$(&f) — psl(mp) is simple. The simple Lie algebras over F thus obtained are called classical. All classical Lie algebras are restricted withpth power map given by (ea(g)l)^ = 0 and (hi ® 1)^ = hi ® 1 for all a G $ and 1 < i < I. As in characteristic 0, they are parametrized by Dynkin diagrams of types An, Bn, Cn, Dn, G2, F4, E$, E?, Eg. We stress that, by abuse of characteristic 0 notation, the classical simple Lie algebras over F include the Lie algebras of simple algebraic F-groups of exceptional types. All classical simple Lie algebras are closely related to simple algebraic groups over F.
192 ALEXANDER PREMET AND HELMUT STRADE A Lie algebra L of characteristic p > 3 is called almost classical if adgcLc Derq where $ is a direct sum of classical simple Lie algebras. One of the examples of such algebras is the Lie algebra p$l(n) := #l(n)/FIn. When p does not divide n, we have that p$l(n) = si(n) as Lie algebras. However, pQl(mp) ^ sl(rap), because for p > 2 the Lie algebra si(mp) is perfect with a 1-dimensional center, while the Lie algebra pgi(mp) is centerless and [p&l(mp),pQl(mp)] = psi(mp) is an ideal of codimension 1 in pQl(mp). It is easy to see that the Lie algebra p&l(mp) is almost classical. All almost classical Lie algebras are semisimple, but the case of pgi(mp) shows that they are not always direct sums of classical simple Lie algebras. Kostrikin conjectured in [Ko 63, Ko 71] that for p > 5 a Lie algebra L over F is almost classical if and only if S(L) = {0} (a closely related conjecture can be found in the last section of [Ko-S 66]). Kostrikin's conjecture was proved in [P 86a] for p > 5 and in [P 86c] for p = 5. 3.2. Graded Lie algebras of Cartan type. In [Ko-S 69], Kostrikin and Shafarevich gave a unified description of a large class of nonclassical simple Lie algebras over F. Their construction was motivated by classical work of E. Cartan [C 09] on infinite dimensional, simple transitive pseudogroups of transformations. To define finite dimensional modular analogues of complex Cartan type Lie algebras Kostrikin and Shafarevich replaced formal power series algebras over C by divided power algebras over F. Let Nq1 denote the additive monoid of all m-tuples of nonnegative integers. For a,/3 G N^ define («) = (««) • • • («£>) and a\ = UT=i «(0«- For 1 < i < m set £i = (tin, • • •, Sim) and 1 = ei + ... + em. Give the polynomial algebra F[Xi,..., Xm] its standard coalgebra structure (with all Xi being primitive) and denote by 0(ra) the graded dual of F[Xi,..., Xm], a commutative associative algebra over F. It is well-known (and easily seen) that 0(ra) has basis {xa \ a G N™} and the product in 0(ra) is given by i)xa+/3 for all a,/?GN£\ We write Xi for x€i G 0(m), 1 < i < m. For each m-tuple n G Nm we denote by 0(ra; n) the F-span of all xa with 0 < a(i) < pUi for i < m. This is a subalgebra of 0(ra) of dimension p'-' where \n\ =n\-\ h nm. Note that 0(ra;l) is isomorphic to the truncated polynomial algebra F[X\,..., Xm]/(Xf,..., Xm). Assigning degree |a|=a(l) + --- + a(m) to each xa G 0(m) gives rise to a grading of the algebra 0(ra), called standard. Each 0(ra;n) is a graded subalgebra of 0(ra). The A:th graded component of 0(ra) is denoted by 0(ra)fc. The subspaces 0(m)(£) := ©i>fc 0(m)i form a decreasing filtration of 0(m), called the standard filtration. The completion of 0(ra) relative to its standard filtration is denoted by 0((ra)). The elements of 0((ra)) are the infinite formal sums of the form ^2a ^<* X<X with Xa G F. The algebra 0((ra)) is linearly compact and 0(m) is canonically embedded into 0((m)). The subspaces 0((m))^ := {52\a\>k Aaxa | Xa G F} and 0((m))fc := 0(m)k induce a decreasing filtration and topological grading of 0((m)), respectively. These are, again, called standard. x xH
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 193 There is a family of continuous maps {y i—> y^ \s G No} from 0((m))^ into 0((ra)), called divided power maps, such that x(0) = 1 for all xe 0((m))(i); (x")(s) = ((5a)!/(a!)s5l)xsa for all a^(0,...,0); (Ax)(s) = Asx(s) for all A G F, x G 0((m))(1); s (x + y)W = ^VV^ for all x,y G 0((m))(1). 2 = 0 A continuous automorphism <f> (respectively, derivation D) of the topological algebra 0((ra)) is called admissible (respectively, special) if (f){x^) = (fix)^ (respectively, D(x^) = x^-^Dx) for all x G 0((ra))(i) and all 5 G N0. For 1 < i < ra, the ith partial derivative <92 of 0((ra)) is defined as the special derivation of 0((ra)) with the property that di(xa) = xa~ei if a(i) > 0 and 0 otherwise. Each admissible automorphism of 0((ra)) respects the standard filtration of 0((ra)). Each finite dimensional subalgebra 0(ra; n) is stable under the partial derivatives <9i,..., <9m. The set W((m)) of all special derivations of 0((ra)) is an infinite dimensional Lie subalgebra of DerO((ra)) and an 0((m))-module, via (fD)(x) = fDx for all / G 0((ra)) and D G W((m)). Since each D G W((m)) is uniquely determined by its values Dx\,..., £>xm, the Lie algebra W((m)) is a free 0((m))-module with basis di,..., 9m. The subspaces ra ra W((m))fc := ©0((m))k+i9i and W((m))(k) := ©0((m))(H1)ft 2=1 2=1 for A: > —1 form a topological grading and decreasing filtration of W((m)), respectively. Needless to say, both are called standard. Note that [W((m)){i), W((m)){j)] C W((m)){i+j) for all t > -1, j > 0. The group Autc0((ra)) of all admissible automorphisms acts on W((m)) by the rule D i-> D^ := ^~1D(f), where 0 G Autc0((ra)) and L> G W((m)), and respects the standard filtration of W((ra)). The general Cartan type Lie algebra W{m\n) is the O(ra;n)-submodule of W((m)) generated by the partial derivatives di,... ,9m. The Lie algebra W{m\n) is a subalgebra DerO(ra;n). When n = I, it is isomorphic to the full derivation algebra of F[Xi,..., Xrn]/(X^1..., X£J, a truncated polynomial ring in ra variables. In the literature, W(m;n) is often referred to as a Lie algebra of Witt type. Since W(m;n) is obviously a free O(ra;n)-module of rank ra, we have that dim W(m\n) = rap'-L The Lie algebra W{m\n) is simple unless (p,ra) = (2,1). If XL 7^ 1 and nr ^ 1 then df ^ 0 on 0(ra; n). Since d£ is not a special derivation of 0((ra)) it follows that W(m; n) is restrict able if and only if ra = L Give the 0((ra))-module ^((ra)) := HomO((m))(w((m)),0((m))) a W((m))-modu\e structure by setting {Da){Df) := D(a(D')) - a{[D,D'\) for all D,D' G W(M) and a G ^((ra)), and define d: 0((ra)) —► ^(M) by the rule (d/)(£>) = £>/ for all D G W((ra)) and / G 0((ra)). Notice that d is
194 ALEXANDER PREMET AND HELMUT STRADE a homomorphism of W((ra))-modules and n1((m)) is a free O((ra))-module with basis dx\,..., dxm. Let n((m))= 0 nk((m)) 0<k<m be the exterior algebra, over 0((ra)), on Q1((m)). Then fi°((ra)) = 0((m)) and each graded component fifc((ra)), A: > 1, is a free O((ra))-module with basis {dx^ A ... A dxik 11 < i\ < ... < %k < m}. The elements of fi((ra)) are called differential forms on 0((ra)). The map d extends (uniquely) to a zero-square linear operator of degree 1 on Q((m)) such that d(fu) = (df) Aw + fd(u), d{u)X A u)2) = d{ux) A u2 + (-l)^"1^ A d(u2) for all / G 0((ra)) and all homogeneous a;,a;i,a;2 G fi((ra)). For L> G W((m)), we have that D(fcu) = (Df)u + fD(ou). It follows that each £> G W((m)) extends to a derivation of the F-algebra fi((m)). All such derivations commute with d. The group Autc 0((ra)) acts on n1((m)) by the rule (<fiu>)(D) := </>M^)) for all <£ G Autc 0((m)), u G ^((m)), £> G W((m)). Moreover, 4>{f<jo) = c/)(f)(f)(Lj) and (f)o d = do (f) for all (j) G AutcO((ra)), cj G fi((ra)), / G 0((m)). It follows that the action of AutcO((ra)) on n1((m)) extends to an embedding AutcO((ra)) *-> Autjrfi((m)). It can be shown that D*{«>) = ct>-\D{ct>{u)) for all D G W((m)), <£ G Autc 0((m)), cj G ft((ra)). Each m-tuple r of nonnegative integers induces a grading of the algebra 0(ra) defined by assigning deg(arQ!) = r(l)a(l) + ••• + r(m)a(m) to each monomial xa G 0(m). Such a grading, in turn, induces (topological) gradings and decreasing filtrations of the algebras 0(ra; n), 0((m)), W(m; n), and W((m)). It also induces a topological grading of the algebra fi((ra)) which extends that of 0((ra)) = fi°((m)). The differential d of fi((ra)) preserves all components of this grading. The gradings and filtrations thus obtained are all said to be of type r. In this new terminology, the standard gradings and filtrations defined above are all of type I. As in the characteristic 0 case, the three differential forms below are of particular interest: lus := dx\ A ... A dxm, m > 3, uh := Dl=i dxi A dxi+r, rn = 2r > 2, uK := dx2r+i + J21i=i(xi+rdxi-Xidxi+r), rn = 2r + 1 > 3. These forms give rise to the following Lie algebras: 5((m)) := {DeW((m))\D(us) = 0}, special Lie algebra, H((m)) := {D€W((m))\D(u>H)=0}, Hamiltonian Lie algebra, K((m)) := {D€W((m))\D(uK)€0((m))LJK}, contact Lie algebra.
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 195 Define Lie algebras CS((m)) and CH((m)) by setting CS((m)) := {DeW((m)\D(Lus)eFLUs}, CH((m)) := {D G W((m)) \ D{uH) G FuH}. Obviously, CX((m))^ C X((m)) for X G {S,H}. For X G {W,S,CS,#,C#}, set r_x = ei H h em = L For X = X, set rx = e\ H h em_i + 2em = 1 + em. For X G {W, 5, CS, H, CH, K} and n G Nm, define X(ra;n) = X((m)) Cl W(m; n). Each X(m;n) is a graded subalgebra of the Lie algebra X((m)) regarded with its grading of type r_x. The graded components of X(m; n) are denoted by X{m\ n)i, i G Z. Note that X(m;n)i = (0) for i < -2 if X ^ X. Also, dim K{m\n)-2 = 1 and K(m\n)i = (0) for z < —3. Suppose p > 3. In [Ko-S 69], it was shown that the Lie algebras 5(m; n)^\ H(m\n)^ and K(m;n)^ are simple for m > 3 and that so is i/(2;n)(2). Moreover, X(m;n) = X(ra;n)(1) unless p \ (ra + 3). For X G {W, 5, CS, #, C#, K} any rx-graded Lie subalgebra of X(m;n) containing X(ra;n)(°°) is called a finite dimensional graded Lie algebra of Cartan type. According to [Ko-S 66] the Lie algebra X{m\ n)^°°^ is restrictable if and only if n = 1. The original Kostrikin-Shafarevich conjecture [Ko-S 66] of 1966 states the following: For p > b, any finite dimensional restrictable simple Lie algebra over F is either classical or isomorphic to one of the Lie algebras W(ra;I), m > 1, 5(m;l)^1\ m > 3, H(m; 1)(2>, m > 2, K(m\V){1), m > 3. 3.3. Filtered Lie algebras of Cartan type. In order to give a unified description of all known finite dimensional simple Lie algebras of characteristic p > 5 Kac [Kac 74] and Wilson [Wil 69, Wil 76] introduced certain filtered deformations of finite dimensional graded Lie algebras of Cartan type. A streamlined treatment of these algebras is given in [St 04]. We first outline Wilson's original approach. Let X G {W, 5, if, if}, n G N™, and let $ be an admissible automorphism of 0((ra)). For X = K assume further that $ respects the rx-filtration of 0((ra)) (if X ^ K this assumption is fulfilled automatically). Define X(m;n;$) := ($"1 o X((m)) o $) n W(m;n). It is clear from the definition that X(ra;n;Id) = X(m;n) and W{m\n\^) = W(m\n). Definition 4 ([Wil 76]). The Lie algebra X(m;n;$)(oo) is called a filtered Lie algebra of Cartan type if X(m;n;$) satisfies the following two conditions: 1. X(m;n;$) nW(m;n)i2+6x>K),x ^ (0); 2. X(m;n;$) + ($oX((m))o$-1)nW(m;n)(1+(5xK)jX = $oX((m))o$-1. Here W(m;ri)(k),x denotes the kt\i component of the rx-filtration ofW{m\n). The embedding of a filtered Cartan type Lie algebra X(ra; n; $)(°°) into the Lie algebra W{m\n) regarded with its filtration of type rx, induces a natural filtration of X(ra; n; <J>)(°°). The corresponding graded algebra is isomorphic to a graded Cartan type Lie algebra (possibly of type CS or CH) containing X(m; r&)(°°) as a
196 ALEXANDER PREMET AND HELMUT STRADE minimal ideal. The subalgebra L(0) = X(m;n;$)(°°} H W(m;n)(o) is called the standard maximal subalgebra of L = X(m\ n; <J>)(°°). For p > 3, this subalgebra can be characterized as the unique proper subalgebra of maximal dimension in L; see [Kr 71], [Sk 95], [St 04]. As a consequence, L(0) is stable under all automorphisms of L. For p > 3, each Cartan type Lie algebra L is simple ([Wil 76]). The following important abstract characterization of filtered Cartan type Lie algebras is due to Wilson [Wil 76]. Let £ be a Lie algebra over F and let £o be a subalgebra of £. Then we have a natural representation p : £q —> 0t(*C//Lo) °f the Lie algebra £0 given by (p(x))(y + /L0) = [z, y] + £o for all xg£0, y G £. Theorem 4 (Wilson's Theorem). Let & be a simple Lie algebra over F and suppose that char F = p > 3. Then ,C is isomorphic to a finite dimensional filtered Cartan type Lie algebra if and only if £ is strongly degenerate and contains a maximal subalgebra &o su°h that either £q has codimension 1 in & or else p(£o) contains a linear transformation Y of rank 1 such that [Y, [Y,p(£0)]] ¥" (0). Kac's approach [Kac 74] to filtered Cartan type Lie algebras pushed further by Skryabin in [Sk 86, Sk 90, Sk 91, Sk 93, Sk 95] involved more general differential forms in fi((ra)). Combined with Wilson's theorem it eventually led to a complete classification of filtered Lie algebras of Cartan type. Recall that the algebra 0((ra)) is linearly compact. Given a unital associative subalgebra B of 0((ra)) we let W(B) and W(B)^ denote the normalizers of B and B n 0((ra))(o) in W((m)), respectively. We denote by B* the group of invertible elements of B. Following [Sk 91] we say that B is an admissible subalgebra of 0((ra)) if B is closed in 0((ra)) and W(B)^ has codimension m in W(B). This definition is inspired by a crucial definition in [Kac 74]. Any finite dimensional subalgebra of 0((ra)) of the form </>(0(ra;n)) with </> G AutcO((ra)) and n G Nm is admissible. Conversely, given a finite dimensional admissible subalgebra B C 0((ra)) there are an automorphism <j> G Autc 0((ra)) and a tuple n G Nm such that B = </>(0(ra;n)); see [Sk 91]. To ease notation we set Q = fi((m)), Qk = Qk((m)) for 0 < k < m, and put neven := 0i>o n22. Observe that fieven is a commutative algebra over F and W((m)) acts on f}even as derivations. The subspace n*ven := O((m))(1)00i>1 Q2i is a maximal ideal of Qeven which intersects trivially with (fieven)^«m)) = ~p\. It is well-known that the first cohomology group Hl(W((m)), neven) vanishes; see [Sk 91, Theorem 7.5] for example. According to [Sk 91, Proposition 1.2] this implies that there exists unique system of divided powers u i—> u^s\ s > 0, on Qeven yftfa respect to which W((m)) acts on fleven as special derivations. It has the property that a/*) G Q2is whenever u G Q2z and s > 1. Recall that a differential form uj G ft is called closed if du = 0. Following [Kac 74] we say that u G fJm is nondegenerate if m > 2 and u = tp dx\ A ... A dXm for some tp G 0((m))*. We call u G fi2 nondegenerate if m = 2r > 2, a; is closed, and the form a;^r^ G nm is nondegenerate (if m = 2 this is consistent with the previous definition). Finally, we say that u G ft1 is nondegenerate if m = 2r +1 > 3 and (du)(r) Au G fim is nondegenerate. Given a finite dimensional admissible subalgebra B of 0((ra)) we let Q(B) = 0£LO nfc(^) denote the ^-subalgebra of Q generated (over B) by dB. For / G ^((m))(i) we set exp / := J2i>o /^> an element in 0((m))*. Let s(^) (respectively,
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 197 h(B)) denote the set of all nondegenerate forms u G fim (respectively, u G Q2) such that u = (expu)a/ for some a/ G ii(B) and u G 0((ra))(!) satisfying du G Q}{B). Let k(i?) denote the set of all nondegenerate forms in £ll(B). For a; G s(B) define the Lie algebras S(B, u) := {D G W^B) | Do; = 0}; CS(£; a;) := {D G W^B) | Da; G Fa;}. For a; G h(B) define the Lie algebras H{B\ u) := {£> G W(B) | Do; = 0}; CH(B; u) := {L> G W(£) | Do; G Fa;}. For u G k(B) define the Lie algebra K(B\u>) := {L> G W(£) | Do; G Bo;}. It is proved in [Kac 74, Sk 93, Sk 95] that except for two cases in characteristic 2 the Lie algebras W(B\ S(B-lu)^\ H(B\u>)(2\ K(B\u)M are simple. Dimensions and explicit bases of the Lie algebras X(B;u), X{B\u)^ and X(B-,u)W are found in [Sk 95, Kir 89, Kir 90] (see also [B-K-K 95]). The most accessible reference, by far, is [St 04, Sect. 6]. Any simple Lie algebra L = X(B',lu)^°°^ is naturally filtered and grL, the corresponding graded algebra, is isomorphic to a graded Lie algebra of Cartan type. In view of Wilson's theorem this implies that for p > 3 each X(B',lj)^°°^ is isomorphic to a filtered Cartan type Lie algebra. The converse is also true: for p > 3 any filtered Cartan type Lie algebra -X"(m;n;$)(°°) is isomorphic to one of X(B',lj)(°°) where B = <£(0(ra;n)) for some (f) G AutcO((ra)) (see [Kac 74], [Ku 89], [Sk 91]). The p-structure of filtered Cartan type Lie algebras is described by the following theorem. Theorem 5 ([Kac 74, Sk 95]). Let B = </>(0(ra;n)) where <\> G AutcO((ra)). (1) The Lie algebras W(B), CS(B;u), CH(B\u), K(B;u) and K(B',u)W are restrictable if and only if n=l. (2) The Lie algebras S(B; u) and H(B; u) are restrictable if and only ifn = \ and Lo G Q(B). (3) The Lie algebras S(B;u)^\ H(B;v)W and H(B\lj)W are restrictable if and only if n = 1 and u G dQ(B). It follows from Theorem 5 that for p > 3 the Lie algebra X(m;n;$)^°°^ is restrictable if and only if it is isomorphic to one of W(ra;l), 5(m;l)(1\ if (ra;l)(2\ K(ma){1). The realizations of filtered Cartan type Lie algebras just described are very useful in view of Kac's Isomorphism Theorem which was later refined by Skryabin; see [Kac 74, Sk 91, Sk 95]. Let B (respectively, B') be an admissible subalgebra of 0((m)) (respectively, 0((ra'))), and X,X' G {W,S,H,K}. Slightly abusing notation we set W{B\u) = W(B) and likewise for W{B'). We call a linear map a : X(B;u)^ —> X'(B'\u)')^
198 ALEXANDER PREMET AND HELMUT STRADE standard if a(D) ^^oDof1 for all D G X(B',lj)(°°\ where ip : 0((ra)) -^> 0((m/)) is a continuous isomorphism of divided power algebras satisfying ij){B) = B' and ip{u) = Cu' with C e F* for u G s(B) U h(B) and C G B'* for cj G k(B). Clearly, any standard map is a Lie algebra isomorphism. Also, if a : X{B\u)^°°^ —► X'(B'',lu')(°°} is a standard map then necessarily m = ra' and X = X'. Theorem 6 (Isomorphism Theorem). Let B,B' and X,X' be as above. Then with eight exceptions in characteristic 2 and three exceptions in characteristic 3 any isomorphism between the Lie algebras X{B\u)^°°^ and X'(B'',lu')(°°} is standard. In our further discussion of the Lie algebras X(B\u) we shall assume (without loss of generality) that B = 0(ra;n). In this special case, X(B\u) is denoted by X(ra;n;a;). The corresponding set x(£) of nondegenerate forms will be denoted by x(ra;n). We shall also assume (as we may) that n is a partition of |n|, that is n\ > ... > nm. Let G{m\n) denote the set-wise stabilizer of 0(ra;n) in AutcO((ra)). This is a connected algebraic group with a large unipotent radical; see [Wil 71] for more detail. It follows from Theorem 5 that with a few exceptions in characteristics 2 and 3 the Lie algebras X(ra; n; lu)^°°^ and ^'(ra';??/;^/)^00) are isomorphic if and only if ra = ra', n = n' and gu = Clo' for some g G G(ra;n), where C G F* for X G {S, H} and C G 0(m;n)* for X = K. Let e(n) denote the group of all permutations n of {1,2,..., ra} such that n^i = rii for all i. The orbits of G(ra;n) on s(ra;n) are described in [T 78] and [Wil 80]. Let I(n) denote the subset of {1,2,... ,ra} consisting of 1 and all A: with n& < n^-i. Set SR = (pni - 1,..., Pnm - 1). According to [T 78, Wil 80], each u G s(ra;n) is conjugate under G(ra; n) to a nonzero scalar multiple of precisely one form in the set {(expxi) lus | it I(n)} U {lus, (1 - x6*-) lus}- As a consequence, for p > 2 and n G Nm fixed, there are only finitely many filtered Lie algebras of type 5(ra; n; $)(°°) up to isomorphism. The orbits of G{m\n) on k(ra;n) are described in [K-K 86b] in the simplest case n = 1 and in [Sk 86] for any n. Let D^ denote the set of all decompositions of {1,2,..., ra} into a disjoint union of the form I = {t0}U{ti,ii}U...U{tr,*;}, **<**, (different orderings of the subsets within the union are not distinguished). The group 6(n) acts on the set Dk- Given I G Dk define r wkj := dxio + ^xikdxi'k, k=l an element in k(ra;n). It is proved in [Sk 86] that for p > 2 each u G k(ra;n) is conjugate under G{m\n) to fcux,! for some / G 0(ra;n)* and I G 2)k- Moreover, the orbit of fcuK,i under G{m\n) intersects with 0(ra;n)*a;K,i/ for I' G Dk if and only if there is a 7r G 6(n) such that 7r(I) = F. Thus for p > 2 any filtered Cartan type Lie algebra K(m\ n; lu)^°°^ is isomorphic to a graded Cartan type Lie algebra K{m\r]!)^ (here |n7| = \n\ but in general n7 need not be a partition of |n|). It follows that for p > 2 and n G Nm fixed, there are only finitely many Cartan type Lie algebras K{m\ n; $)(°°) up to isomorphism. The orbit set h(ra; n)/G(ra; n) is studied in [Kac 74, K-K 86a, B-G-O-S-W, Sk 86, Sk 90]. In the simplest case n = 1 it is described in [K-K 86a]. For an
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 199 arbitrary n, a reasonably small set of representatives for each G(ra; n)-orbit in h(ra;n) is found in [B-G-O-S-W]. A complete description of h(ra;n)/G(ra;n) is given in [Sk 86, Sk 90]. Set hi(ra;n) := h(m;n)n(l(0(m;n)) and h2(ra;n) := h(ra;n) \ hi(ra;n). Both hi(m,n) and h2(ra;n) are G(m;n)-stable. The orbit sets h2(ra;n)/G(ra;n) and k(m + 1; n)/(G(m +1; n) ix 0(ra +1; n)*) are somewhat similar to each other. Let Dh denote the set of all decompositions I = {i1,i'1}U...U{ir,ifr}, ik <i ki of {1,2,..., m} into a disjoint union of pairs (different orderings of the pairs within the union are not distinguished). The group &(n) acts on the set Dh- Given i G {1,2,..., m} and I G Dh define r UH,i,i := dfexpx* ^ x^rfx^J, fc=i an element in h2(m;n). It is proved in [Sk 86, Sk 90] that for p > 2 each u G h2(ra; n) is conjugate under G(ra; n) to wh,i,i for some I G 2)h and z G {1,2,..., m}. Moreover, the G(m; n)-orbit of uiH,i,i intersects with F*ljh,j,v for F G V^ and j G {l,2,...,ra} if and only if there is a n G 6(n) such that 7r(I) = I7 and ni = j. As a consequence, for p > 2 and n G Nm fixed, there are only finitely many isomorphism classes of Lie algebras of the form H(m\ n; u)^°°^ with u G h2(m;n). The orbit set hi (m;n)/G(m;n) is much more complicated. It is no longer discrete, for m > 4, and this allows one to exhibit multiparameter families of pairwise nonisomorphic simple Lie algebras of dimensions p'-' — 2 and p'-' — 1. This phenomenon was first discovered by Kac who disproved an earlier conjecture of Kostrikin stating that no such families could exist for p > 3 (see [Ko 71]). Let Ji (A) denote the Jordan block of order / with eigenvalue A G F. Let 0\ (respectively, E{) denote the zero (respectively, identity) matrix of order /. Let Cr = 0 1 0 0 1 0 a monomial matrix of order /. Given d, s G N and A G F define Os Es .. Os Cdt8(X) = os os Js(X) Os Es Os a block-monomial matrix of order ds. Let JCm denote the set of all pairs of block- diagonal, skew-symmetric matrices (A, B) = (diag^i,..., Ak), diag(£i,..., £fc)) of order m = 2r = 2r\ + ... + 2rk > 2 such that Ai Ori Eri —Er. Or.
200 ALEXANDER PREMET AND HELMUT STRADE and Bi is one of 0Ti Jrt(fi) -Jri(0) Ori Ori Cdi,ai(X) -c*,.,(A) ori On Cri —Cri Ori where A ^ 0 and r* = diS{. To each (A, B) = ((a^), (6^)) G IHm one associates a differential form lua,b £ ^((m)) by setting va,b = ^2 [aij+bijxi l~ xj 3 JdxiAdxj. It is straightforward to see that ua,b £ hi(ra;n). One of the main results in [Sk 86] (see also [Sk 90]) says that any uj G hi (ra;n) is conjugate under G(ra;n) to one of ua,b with (^4, B) G IHm (this holds in all prime characteristics). Skryabin also found a necessary and sufficient condition for two forms ua,b and wa',b' to be conjugate under G{m\n). It involves an equivalence relation on the set of all pairs of sequences of natural numbers, finite of equal length or periodic; see [Sk 90] for more detail. 3.4. Melikian algebras and their relatives. In this subsection we assume that p G {2,3,5}. Around 1980, Melikian (a PhD student of Kostrikin at the time) discovered a new series of finite dimensional simple Lie algebras M(ra,n) of characteristic 5 depending on two parameters m, n G N. Suppose charF = 5. In [M 80, M 82], the algebra M(ra,n) is described as a graded Lie algebra L = 0^>_2 ^i of dimension 5m+n+1 whose graded subalgebra L-2 ©L-i is isomorphic to a five dimensional Heisenberg Lie algebra and Lq ' = W(l;l) as Lie algebras. Moreover, L0 = Lq 0 3(Lo), 3(^0) = Fz, (^Z)\L = k • IdLfc for all k G Z, and L_i S* 0(1; 1)/F as W{1\ l)-modules. It is shown in [M 82] that each Melikian algebra is strongly degenerate and the only restrictable algebra in the family is M(l, 1) (see also [Ku 90] and [St 04] where all derivations of M(ra, n) are determined). It is stated in [M 80] that M(l, 1) is neither a classical Lie algebra nor a Lie algebra of Cartan type. In [M 82], Melikian outlines a proof of this statement relying on properties of Z-gradings in the contact Lie algebra if (3;I). An alternative proof will be given below. Melikian's work showed that the assumption that p > 5 in the generalized Kostrikin-Shafarevich conjecture could not be relaxed. A few years later Ermolaev observed that $ = M(m, n) admits a more natural Z-grading q = ©i>_3 &(i) that satisfies the conditions (gl), (g2), (g3) of (2.4) and has the property that 0f<1 (j(i), regarded as a local Lie algebra, is isomorphic to the local Lie algebra associated with a depth 3 grading of a Lie algebra £ of type G2- In particular, the nonpositive part 0i<o fl(i) of q is isomorphic to a maximal parabolic subalgebra of £. This observation enabled Kuznetsov to give in [Ku 91] an explicit description of M(m,n). Set n := (m, n) and define Go := 0 fl(»), Gl ••= 0 fl(0, G-2 := 0 0(i). i=0(mod3) i=l(mod3) i=2(mod3) Then g = G§ 0 G\ 0 G5 is a (Z/3Z)-grading of g. According to [Ku 91], Go0GT0G2 = W(2;n)0O(2;n)0W(2;n)
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 201 as vector spaces. Moreover, Gq is identified with W(2;n) as Lie algebras, G\ is identified with 0(2; n) as vector spaces, and G^ is identified with W(2',n) — {D | D G W(2;n)}, a vector space copy of W(2',n). The Lie product in $ is given by [£>, E] = \D^E] + 2div(£>) E, [A/I = £>(/)-2div(L>)/, [/l#l + /2#2, 9ldl + ^2^2] = /l#2 - /201, [/J] = fE, [f,g] = 2(fVg-tfDf), Vh = 8^)82-d2(h)du for all D,E G W(2;n), /,0,ft,/i,# G 0(2; n). Here div : W(2;n) -> 0(2, n) is the linear map taking fid\ -\-f2d2 to di(fi) + d2(f2)> It follows from the above formulae that the Lie subalgebra of M(ra,n) generated by the graded components $(±1) is isomorphic to a classical Lie algebra of type G2. Assume for a contradiction that M(l, 1) is either classical or of Cartan type. Since M(l, 1) is strongly degenerate, simple, and restrictable it must be isomorphic to one of W(ra;l), S(ro;D(1), #(ra;l)(2), K(m;V)W. Since dimM(l,l) = 125, there is only one option, namely, M(l, 1) = K(3;l). Using the above multiplication table one can observe that to := F(l + Xi)d\ 0 F(l + #2)^2 is a torus in M(l, 1) whose centralizer \) is a five dimensional Cartan subalgebra of M(l, 1) with the property that [I), [f),f)]] = to (see [P 94] for more detail). However, all Cartan subalgebras in if(3;l) are abelian, as can be deduced from [Dem 72] and [St 04, (7.5)]. Thus M(l,l) ^ #(3;1), and so M(l, 1) is neither classical nor of Cartan type. Although the Melikian algebras have sporadic nature and can survive as Lie algebras only at characteristic 5, they have some relatives in characteristics 3 and 2. This was discovered by Skryabin [Sk 92] and Brown [Br 95]. Suppose char F = 3. Each Skryabin algebra g is equipped with a Z-grading 0 — 0i>-4 8* satisfying the conditions (gl), (g2), (g3) of (2.4) and one of the three conditions below: 1) & = (0) for i < -3 and g0 = &K&-i)i 2) & = (0) for i < -3 and $0 = *K(J-i)> 3) 9i = (0) for i < -5 and g0 = 0l(8-i)> dim 8-4 = 3. Moreover, dim $_i = 3 and $_2 = A2$_i in all cases, and $-3 = A3$_i in case 3). In cases 1) and 2), each Skryabin algebra admits a natural (Z/2Z)-grading 0 = Gq 0 Gi such that Go is either W(3;n) or 5(3;n;u)W with u G s(3;n) and Gi is a nice irreducible Go-module. In case 3), each Skryabin algebra admits a natural (Z/4Z)-grading q = Gq 0 G\ 0 G5 0 G3 such that Gq = W(3; n) and each G^ with i ^ 0 is a nice Go-module. In all cases, the Lie bracket in $ is given by explicit formulae involving classical operations with differential forms (see [Sk 92] for more detail). Now suppose char F = 2. In [Br 95], Brown constructed three series of simple Lie algebras over F one of which relates closely with the Melikian series. Following [Br 95] consider the (Z/3Z)-graded algebra £ = £q ® ^1 ® ^2 sucn that £q = W(2;n), £2 = 0(2; n), and £T = {fu\f G 0(2; n)}, a second vector space copy of 0(2; n). The multiplication function [•,•]: £ x £ —> £ satisfies the
202 ALEXANDER PREMET AND HELMUT STRADE identity [x,x] = 0, agrees with the Lie bracket of W(2;n), and has the following properties: [DJu] = div(/D)ti, [D,f\ = D(f), [/tilSti] = 0, \fu,g] = fVg, [f,g) = <Dg{f)u (here f,g G 0(2; n), D G W(2;n), and D^ has the same meaning as before). It is shown in [Br 95] that £ is a Lie algebra carrying a natural Z-grading £ = ©i>-4 ^i sucn tnat £-4 = j(£). The Lie algebra g := (Jd/i(Jd))^ is denoted by G2(2;n). It is simple, has dimension 2'-l+2 — 2, and inherits from £ a natural Z- grading $ = 0i>_3 0i satisfying the conditions (gl), (g2), (g3) of (2.4). Moreover, 0o = flKfl-i)» dim fl-i =z dim 0-3 = 2, and g_2 = A2fl_i. The 14-dimensional Lie algebra G2(2;I) is not restrict able but can be obtained by reducing modulo 2 a nonstandard Z-form of a complex Lie algebra of type G2 (see [Br 95] for more detail). 4. Classification theorems One of the main goals of this survey is to announce the following theorem which, in particular, confirms the original Kostrikin-Shafarevich conjecture in full generality; see [P-St 06]. Theorem 7 (Classification Theorem). Let L be a finite dimensional simple Lie algebra over an algebraically closed field of characteristic p > 3. Then L is either a classical Lie algebra or a filtered Lie algebra of Cartan type or one of the Melikian algebras. Our proof of Theorem 7 relies on several earlier classification results which we are going to formulate. Prom now on we assume that charF = p > 3. The following useful characterization of classical Lie algebras is due to Seligman and Mills: Theorem 8 ([M-Se 57]). A Lie algebra L over F is a direct sum of classical simple Lie algebras if and only if the following conditions hold: (1) L is perfect and j(L) = (0); (2) L contains an abelian Cartan subalgebra H such that (a) L = H 0 Y^a^o L<* where La = {x G L \ [h,x] = a(h)x (Vft G H)}; (b) if La ^ (0), then dim [La,L_a] = 1; (c) if La ^ (0) and Lp ^ (0), then La+kp = (0) for some k G Fp. A short proof of the Seligman-Mills theorem based on the Kac-Moody theory can be found in [S 80]. The following important theorem allows one to recognize certain filtered simple Lie algebras: Theorem 9 (Recognition Theorem). Let L be a finite dimensional simple Lie algebra over an algebraically closed field of characteristic p > 3. Let L = L(_s/) D ... D Z/(o) D ...D L(8) D (0), [L(i),L(j)] C L(i+J-), be a filtration of L satisfying the following conditions: (a) 5, s' > 1 and s' < s;
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 203 (b) L(0)/Z/(i) is a direct sum of ideals each of which is either classical simple or %i{n), sl(n), pQl(n) with p\n or abelian; (c) L(_i)/Z/(o) is an irreducible L^-module; (d) for all j < 0, if x G L^ and [x,L^] C L^+2), then x G £(j+i); (e) for all j > 0, if x G Ly) and [x, £(-i)] ^ ^{j)y then x G Ly+i). T/ien L is either classical or is isomorphic as a filtered algebra to a Lie algebra of Cartan type or a Melikian algebra regarded with their natural filtrations. The Recognition Theorem incorporates Wilson's theorem [Wil 76] and earlier results of Kostrikin and Shafarevich [Ko-S 69]. Kac was the first to formulate a version of this theorem for graded Lie algebras, and he made in [Kac 70] many deep and important observations towards its proof. One of Kac's original assumption on the pair (L(_!), £(o)) was relaxed by Benkart-Gregory in [B-G 89]. The first complete proof of the Recognition Theorem for graded Lie algebras was obtained only very recently by Benkart-Gregory-Premet; see [B-G-P]. Theorem 9 is a consequence of this result; see [St 04, Section 5] for more detail. Theorems 8 and 9 are fundamental, and most of the classification proofs rely on them at some stage. Given a nilpotent Lie subalgebra H of L we denote by H£or the unique maximal torus in the p-envelope of H in Der L. We say that H is triangulable if ad h is a nilpotent linear operator for any h G H^ (this is the same as to say that ad if stabilizes a flag of subspaces in L). We list below a few other classification results which are invoked frequently. All of them share the assumption that L is a finite dimensional simple Lie algebra over F. 4.1. Kaplansky [Kap 58]: If p > 3 and L contains a one dimensional Cartan subalgebra Ft with ad£ toral, then L is either sl(2) or W(l;l). 4.2. Demushkin [Dem 70, Dem 72], Strade [St 04, (7.5)]: If L is a restricted Lie algebra of Cartan type, then all maximal tori of L have the same dimension and split into finitely many conjugacy classes under the action of Aut L. 4.3. Kuznetsov [Ku 76], Weisfeiler [We 84], Skryabin [Sk 97], Strade [St 04]: If p > 3 and L contains a solvable maximal subalgebra, then either L^si(2) or L^W(l',n). 4.4. Wilson [Wil 77], Premet [P 94]: If if is a nontriangulable Cartan subalgebra of L, then p = 5 and there exist Fp-independent a,/3 G r(L, HpOT) and an ideal R(a, fi) of the 2-section L(a, /?) such that L(a,^)/%,^)^M(1,1). 4.5. Wilson [Wil 78], Premet [P 94]: If p > 3 and L contains a Cartan subalgebra H with dim Hlov = 1, then L is one of «[(2), W(l;n), H(2; n; $)<2). 4.6. Block-Wilson [B-W 82], Wilson [Wil 83]: Suppose L is restrictable and p > 7. If L contains a toral Cartan subalgebra, then either L is classical orL^ W(n;l). 4.7. Benkart-Osborn [B-O 84]: If L contains a one dimensional Cartan subalgebra and p > 7, then L is either sl(2) or W(l\n) or L = H(2;n;$)<2) and dim L = p'-'.
204 ALEXANDER PREMET AND HELMUT STRADE The following result of Block-Wilson marked the first real breakthrough in solving the classification problem for p > 7. Theorem 10 ([B-W 88]). The original Kostrikin-Shafarevich conjecture is true for p > 7. Relying heavily on an important intermediate result of [B-W 88] and the classification techniques of Block-Wilson the second author was able to generalize Theorem 10, with some support of R.L. Wilson (see [St 89b, St 91, St 92, St 93, B-O-St 94, St 94, St 98]). Theorem 11 (Strade 1998). The generalized Kostrikin-Shafarevich conjecture is true for p > 7. Large parts of the proof of Theorem 11 go through for p > 3 and are incorporated into our proof of Theorem 7. 5. Principles of the classification Let L be a simple Lie algebra over F (recall that charF = p > 3). As in the characteristic 0 case we hope to get more insight into the structure of L by looking at the root space decomposition of L relative to its Cartan subalgebra [). However, most of the classical results are no longer valid in our situation. For example, a (2m + l)-dimensional Heisenberg Lie algebra over F admits irreducible representations of dimension pm. This implies that Lie's theorem on solvable Lie algebras fails in characteristic p. The Killing form of any strongly degenerate simple Lie algebra over F vanishes (see (2.3)). Since all finite dimensional Cartan type Lie algebras over F are strongly degenerate, Cartan's criterion is no longer valid in characteristic p either. Cartan subalgebras of L need not be conjugate under the automorphism group AutL and, in fact, may have different dimensions (see our discussion in (2.1)). In characteristic 5, one can even expect L to possess nontriangulable Cartan subalgebras (see (4.4) and our discussion in (3.4)). In general, a nonrestrictable Lie algebra does not possess a Jordan-Chevalley decomposition. To fix that we embed L = ad L into its semisimple p-envelope £ (see (2.2.3)). The Lie algebra £ C DerL is restricted, hence admits a Jordan- Chevalley decomposition. By construction, £(*) C L (and £ = L if and only if L is restrictable). We choose a torus T of maximal dimension in £ and take a close look the root space decomposition L = He ^2 La aer(L,T) of L relative to T. Although the subalgebra H = {x G L \ [t,x] = 0 Vt G T} is nilpotent it is not always a Cartan subalgebra of L (if L is nonrestrictable, it may even happen that H = (0)). We wish to gather as much information as we can on the structure of 1- and 2-sections of L relative to T. In characteristic 0, such information eventually allows one to determine the global structure of L. In characteristic p, the local analysis is much more involved. There are a number of reasons for that. To mention just a few, the 1-sections of L relative to T are no longer "reductive" and their irreducible representations are hard to describe. Some tori of maximal dimension in £ are unsuitable for our purposes, and a lot of effort is spent on optimizing a randomly chosen T by using generalized Winter exponentials; see (2.1). In the course of the proof one has to make various
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 205 sophisticated choices of maximal subalgebras, carry out detailed computations in Lie algebras of small rank, and study central extensions of such algebras and their irreducible representations. For any a 6 r(L, T) the semisimple quotient L[a] of the 1-section L(a) is either zero or sl(2) or W(l;l) or the inclusion tf(2;l)(2) cL[a]c J?(2;U holds (this follows from (4.5)). Accordingly we call a solvable, classical, Witt or Hamiltonian. It is not difficult to show that the radical of L(a) is T-stable. So T acts as derivations on L[a] and L[a]^2\ Following Block-Wilson we say that a is a proper root if either L[a] G {(0),sl(2)} or L[a] is of Cartan type and the standard maximal subalgebra of L[a](2) is T-invariant. If a is not a proper root we say that a is improper. The main intermediate result of [B-W 88] is a classification of all simple Lie algebras of absolute toral rank 2 for p > 7. Combining this classification with a version of (4.4) for p > 7, Block and Wilson succeeded to describe the semisimple quotients of all 2-sections in a restricted simple Lie algebra. Having achieved that they proceed as follows: The description of the quotients L[a] mentioned above implies that each 1- section L(a) contains a unique subalgebra Q(a) with H C Q{o<) and dim Q(a) = dim L(a) — e(a), where {0 if a is solvable or classical, 1 if a is Witt, 2 if a is Hamiltonian. The subalgebra Q(a) is solvable if a is solvable or Witt, and Q(a)/rad Q(a) = sl(2) if a is classical or Hamiltonian. In all cases, Q(a) is T-invariant if and only if a is a proper root of L. Generalized Winter exponentials are now used to "optimize" T. A torus T C & is called optimal if dim T = MT(L>) and the number of proper roots in T(L, T) is maximal possible. Using their description of the semisimple quotients L(a,/3)/rad L(a,/3) Block and Wilson prove that in the restricted case all roots of L relative to an optimal torus T C £ are proper. They then look again at the 2-sections of L relative to T to prove that the T-invariant subspace 0 = 0(L,r):= Yl £(<*) aer(L,T) is a Lie subalgebra of L. The rest of the proof is straightforward. If Q = L, Block and Wilson show that the Seligman-Mills theorem applies to L. So L is classical in this case. If Q ^ L, they show that Q can be embedded into a maximal subalgebra satisfying the conditions of the Recognition Theorem. For an arbitrary simple L, the second author used the Block-Wilson classification of simple Lie algebras of rank 2 to obtain a list of all possible T-semisimple quotients of the 2-sections of L (this list is longer than in the restricted case). He then succeeded to optimize T in & and in the joint work with Benkart and Osborn [B-O-St 94] constructed a large Lie subalgebra Q = Q(L,T) of L. However, the final parts of the proof in the general case are much more involved; see [St 91, St 93, St 94, St 98]. Essentially, this is due to the fact that £ is no longer simple. Since optimal tori may lie outside L some 3-sections have to be thoroughly investigated.
206 ALEXANDER PREMET AND HELMUT STRADE It turned out that if all regular Cartan subalgebras of £ are triangulable, then the final parts of the second author's classification go through for p > 3 after a proper modification. This modification is carried out in [P-St 04, P-St 06], thus settling the remaining case p = 7 of the generalized Kostrikin-Shafarevich conjecture. If & contains a nontriangulable regular Cartan subalgebra, then [P-St 04, Theorem A] and (4.4) imply that p = 5 and one of the semisimple quotients L(a,/3)/radL(a,/3) is isomorphic to M(l,l). This situation is investigated in [P-St 07], the last paper of the series. The main result of [P-St 07] states that L is then isomorphic to a Melikian algebra M(m,n). The hardest part of our proof of Theorem 7 is the classification of the simple Lie algebras of absolute toral rank 2 and the description of the 2-sections of L relative to T. The former is obtained in [P-St 97, P-St 99, P-St 01] while the latter is carried out in [P-St 04]. Below we outline our arguments in the rank 2 case. (A) Prom now on we assume that L is a nonclassical simple Lie algebra of absolute toral rank 2 and £ is the semisimple p-envelope of L; see (2.2.3) and Definition 3 in (2.2). In view of (4.4) and (4.5) we may assume that for any maximal torus Tc£ the centralizer H = Cl(T) is triangulable and has the property that dim i/*or = 2 (in particular, it can be assumed that all maximal tori in £ are two dimensional). Finally, we may assume that all simple Lie algebras g with TR(g) = 2 and dim g < dim L are known. Our ultimate goal is to prove that L admits a filtration satisfying the conditions of the Recognition Theorem. However, at the beginning of the investigation any long filtration invariant under a two dimensional torus in £ would do. Thus we have to address the following Problem. Find a long standard filtration in L stable under the action of a maximal torus in £. This problem is solved in [P-St 97] by producing a root sandwich in L, that is a nonzero sandwich element c G L such that [T, c] C Fc for some torus T of maximal dimension in £. The set of all such sandwiches is denoted by S(L, T). Adopting the method used in [P 86a, P 86c] for proving Kostrikin's conjecture we first show that under some mild assumptions on a 1-section of L there exists a nonzero x eHu (J L7 7^r(L,T) such that (adx)3 = 0. Then we use some techniques from [B 77, Ko 67] and the theory of finite dimensional Jordan algebras to find a root sandwich c. More precisely, we prove Theorem 12 ([P-St 97]). Let g be a simple Lie algebra of absolute toral rank 2 over F. Then either g is classical or g = f/(2;l;<I>)(2) with dimg = p2 — 1 or there exists a two dimensional torus t in the semisimple p-envelope of g such that S(u,t)^0. Having found a root sandwich c G L we now observe that any maximal sub- algebra L(o) of L containing H + cl(c) gives rise to a long T-invariant filtration of L. Indeed, let £(-i) be any L(0)-stable subspace of L such that £(-i) 2 -k(o) and L(_!)/L(0) is an irreducible L(0)-module. The L(0)-module £(-i) is if-stable, hence T-stable (for T = H^v). Therefore, so are all components of the standard
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 207 filtration associated with the pair (£(-i), £(o))*> see (2.4) for more detail. Since [!/(_!), c] C Cl(c) C L(o) we have that O^cG ^(i)- I*1 [P-St 97] we use our assumption that H is triangulable to show that maximal subalgebras of L containing H + cl(c) exist. (B) Next we investigate the graded Lie algebra G := gr L. Let M(G) denote the largest ideal of G contained in Xli<-i ^i» an(^ & :~ G/M(G). By a theorem of Weisfeiler [We 78], the Lie algebra G is semisimple and has a unique minimal ideal, denoted A(G). Furthermore, G inherits a natural grading from G which satisfies the conditions (gl) - (g4). Note that finite dimensional semisimple Lie algebras over F need not be direct sums of simple ideals (in fact, simple ideals may not exist at all). The structure of semisimple modular Lie algebras was determined by Block in [B 69]. The following important theorem describes the structure of a semisimple Lie algebra with a unique minimal ideal: Theorem 13 (Block's Theorem). Let q be a finite dimensional semisimple Lie algebra over an algebraically closed field of characteristic p > 0 and suppose that $ contains a unique minimal ideal, I say. Then there exist an r G No and a simple Lie algebra s such that I = s® 0(r; 1) as Lie algebras. Moreover, $ = ad/ $ and (ads) ® 0(r; 1) Cad/gC (Ders) ® 0(r; 1) x Id, ® W(r;l). In a sense, the above-mentioned theorem of Weisfeiler can be regarded as a graded version of Block's theorem; see [St 04, (3.5)] for more detail. In [P-St 99], we show that our maximal subalgebra L(0) can be chosen such that either G2 ^ (0) or [[G_i,Gi],Gi] ^ (0). In this case, Weisfeiler's theorem says that A(G) = Q)iA(G)i where A(G)i = A(G) fl Gi and there exist a graded simple Lie algebra S = 0i Si and an integer m > 0 such that i4(G)^S®0(m;l) = 0 (S; ® 0(ra; 1)) as graded Lie algebras. In [P-St 99] we show that m < 1. Moreover, we prove that if m = 1, then the absolute toral rank of S drops. In view of (2.2.4) and (4.5) the equality ra = 1 implies that S is one of sl(2), W(l; 1), H(2\ V){2)• We first consider the case where ra = 1. By Block's theorem, we then have an embedding G <-► (Der5)®0(l;l) xld5®^(l;l). In view of a conjugacy theorem proved in [P-St 99] along comes an induced embedding of tori T ^ T0 ® 1 + Id5 ® Fzd, z e{x,l + x}, where Tq is a one dimensional torus in Der S. We then show that T and L(0) can be chosen such that S = H(2;V)^ as graded Lie algebras, where if(2;l)(2) is regarded with its grading of type 1; see (3.2). In particular, So = si(2) and S-k = (0) for k > 2. We also show that M(G) = (0). This information enables us to conclude, eventually, that p = 5 and L = M(l, 1). (C) Prom now on we may assume that ra = 0. Using the inequality TR(G) < TR(L), proved in [Sk 98], we show that TR(S) = 2. We now wonder whether S is listed in the Classification Theorem.
208 ALEXANDER PREMET AND HELMUT STRADE First we observe that the root sandwich c G L(i) gives rise to a nonzero sandwich element of G contained in the graded component Gi for some / > 1. Since M(G) C ©$<_! Gi, it follows that the Lie algebra G must be strongly degenerate. Since 5 C G C Der 5, it follows that 5 is not a classical Lie algebra. Next we observe that the quotient space M := M(G)/M(G)2 is a G-module, hence an 5-module. Let Sp denote the p-envelope of 5 in Der 5. We show in [P-St 01] that any composition factor V of the 5-module M can be viewed in a natural way as a restricted 5p-module and T can be identified with a two dimensional torus in 5p. Since H C L(0) it must be that Ogrw(V,T). On the other hand, we show in [P-St 01] that if 5 is isomorphic to one of 5(3; I)*1), ff(4;l)(2\ tf(3;l), M(l,l), #(2; (2,1))<2>, then T has weight 0 on any finite dimensional restricted 5p-module. This implies that M{G) = (0) if 5 is one of these Lie algebras. A slight modification of the argument shows that M(G) = (0) if 5 is one of W(2] 1), W(l;2), #(2; 1;$)<2). Since TR(S) = 2 we deduce the following: if 5 is known, then M(G) = (0). Suppose 5 is known. Then G == G = gr L, hence L is a filtered deformation of G c Der 5. So there exists a Lie algebra £ over the polynomial ring F[t] such that £/(t - A)£ £* L if A ^ 0, and £/t£ ^ G D S. Suppose 5 is a Melikian algebra. Since TR(S) =2we then have 5 = M(l, 1). By [Ku 90], all derivations of M(l, 1) are inner (see also [St 04, (7.1)]). So it must be that G == 5. We already mentioned in (3.4) that M(l, 1) contains a two dimensional torus to whose centralizer f) is a nontriangulable Cartan subalgebra of M(l, 1). As TR(S) = 2, the Cartan subalgebra f) is regular in 5. As all regular Cartan subalgebras of a finite dimensional restricted Lie algebra have the same dimension (see (2.1)) we can lift f) to a nontriangulable Cartan subalgebra of minimal dimension in £®F[t] F(t). We then use a deformation argument to show that L contains a nontriangulable Cartan subalgebra as well. Using (4.4) we finally conclude that L S* M(l, 1). Suppose 5 = X(m; n)^ where X G {W, 5, H, K}. Any grading of a Lie algebra g is induced by the action of a one dimensional torus of the algebraic group Autg. Each such torus is contained in a maximal torus of Aut g. The conjugacy theorem for maximal tori of algebraic groups enables us to prove that any grading of 5 is obtained by assigning certain integral weights to the elements of a generating set of the divided power algebra 0(m;n). This procedure also describes the gradings of Der 5 and provides valuable information on gradings of G (for G can be regarded as a graded subalgebra of Der 5). It turns out that very few gradings of G can satisfy the conditions (gl), (g2), (g3). Taking graded Cartan type Lie algebras of rank 2 one at a time we show that our choice of L(0) (and T) forces the grading of 5 = X(ra;n)(2) to be standard. At this point Wilson's theorem enables us to conclude that L is a filtered Lie algebra of Cartan type. If 5 is a filtered Cartan type Lie algebra not considered before, that is one of type H(2; 1; $)(2\ then 5 is nonrestrictable of dimension p2 — 1 or p2. In this case, Sp is known to possess a two dimensional toral Cartan subalgebra t with the property that dim 57 = 1 for all 7 G T(5, t). This information and an intermediate
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 209 result of [B-W 82] (applicable for p > 3 in view of (4.5)) allow us to show that L too is of Cart an type. (D) It remains to consider the case where S == G = gr L is a minimal counterexample to our theorem. At this stage we may also assume that passing from L to G always produces unknown simple graded Lie algebras (subject to certain conditions on T and L(o))- We use this as a technical tool for improving L(0) and obtaining more information on the structure of r(5^i<0 Gi,T). Given a G T(G, T) we set Ka := {x e Ga |a([x, G_Q]) = 0} and denote by K'(G,T,a) the Lie subalgebra of G generated by all Kia with i e F*. It follows from the main result of [P-St 99] that K'(G, T, a) is a triangulable subalgebra of G. The most important task for us now is to determine the graded component Go- Prom [Sk 97] we know that the radical of Go is abelian, while (4.3) entails that Go is nonsolvable. Thus if radGo ^ (0), then Go := Go/radGo has absolute toral rank 1. Moreover, it follows from (4.5) that Go is either ${(2) or W(l;l) or the inclusion if(2;I)(2) C Go C ff(2;l) holds. Combining some representation theory with the fact that K'(G, T, a) is triangulable (see [P-St 99]), we show after a detailed analysis that either Go = W(l; I) k 0(1; I) (a natural semidirect product) or the radical of Go is one dimensional and central, and the extension 0 -> rad Go -► G0 -► G0 -► 0 splits. If Go is semisimple with a unique minimal ideal J, then Block's theorem says that J = $ ® 0(r;I) for some simple Lie algebra s. If r > 0 we prove that $ has absolute toral rank 1 and there are a vector space V over F and a linear isomorphism G_i -^+ V®0(fc;£) such that 0(fc,{) = 0(r;I) as algebras and the action of Go on G_i is induced by a Lie algebra embedding Go <-* Ql(V) (8) 0(fc;{) >i Idv (8) W(k;Q. Moreover, 7r(Go), the image of Go under the canonical projection tt: fllOO®0(fc;{) >i Idv ®W(k;Q —> W(fe;i), is transitive, that is has the property that 7r(Go) + W(fc;{)(0) = W(k;l). Using the simplicity of G and Cartan prolongation techniques inspired by earlier work of Kuznetsov (see e.g. [Ku 76]) we show that 7r(Go) is an O(fc;[)-submodule of W(fc;{). The transitivity of 7r(Go) now forces 7r(Go) = W(fc;{), while toral rank considerations yield fc = 1, [ = 1. This enables us to prove that Go 3 8 (8) 0(1; 1) xi Idfi 0 W(l; 1), where $ is either s((2) or W(l;l). As a consequence, we obtain that Go belongs to a short list of known linear Lie algebras. Considering algebras from this list one at a time we show that T can be chosen such that all roots in T(G,T) are proper. This allows us to obtain much better estimates for dim Gin with i < 0 and 7 G T(G,T). We use this new information to show that either Go is a classical Lie algebra of rank 2 or the p-envelope 9o of Go in Der G is isomorphic to gl(2) as restricted Lie algebras. Let G' denote the Lie subalgebra of G generated by G±\ and M{G') the maximal graded ideal of G' contained in X)i<o G*. If M(G') ^ (0) we combine the Recognition Theorem with some representation theory of Cartan type Lie algebras
210 ALEXANDER PREMET AND HELMUT STRADE to show that 90 = fll(2) and G'/M(G') is classical of type A2, C2 or G2- We then use the representation theory of algebraic groups to show that this cannot happen. As a result, M(G') = (0). Then the Recognition Theorem applies to G itself, showing that G is known. This contradiction proves the Classification Theorem in the rank 2 case. 6. Some open problems The classification problem in characteristics 2 and 3 is wide open. Since our knowledge of finite dimensional simple Lie algebras over algebraically closed fields of characteristics 2 and 3 is very limited, it is not clear at present whether a complete classification of such algebras can ever be achieved. As indicated in our discussion at the end of (3.3) the classification of Hamiltonian forms in hi (m,n) was reduced by Skryabin to a certain problem of linear algebra. Luckily, the problem turned out to be tame. But if it turned out to be wild, we would never have a complete classification in characteristic p > 3. The first three items will address issues in characteristics 2 and 3. Conjecture 1. The automorphism group of any finite dimensional simple Lie algebra over an algebraically closed field of characteristic p > 0 is infinite. For p > 3, one can easily deduce Conjecture 1 from the results in [P 86a, P 86c] or, alternatively, from Theorem 7. However, the conjecture remains wide open for Pe{2,3}. In [Sk 98], Skryabin proved that any finite dimensional simple Lie algebra of absolute toral rank one over an algebraically closed field of characteristic 3 is isomorphic to either sl(2) = W(l; 1) or psl(3) = H(2;1)W. He also proved in loc. cit. that no finite dimensional simple Lie algebras of absolute toral rank 1 exist in characteristic 2. Problem 1. Classify all finite dimensional simple Lie algebras of absolute toral rank two over algebraically closed fields of characteristics 2 and 3. In characteristic 2, strong results closely related to Problem 1 are obtained by A. Grishkov and the first author (work in progress). We are unaware of any ongoing work on the characteristic 3 case of Problem 1. As mentioned in [B-G-P], it would be very useful to have a version of the Recognition Theorem for graded Lie algebras of characteristics 2 and 3. Problem 2. Classify all finite dimensional graded Lie algebras G = ©iGZ Gi over algebraically closed fields of characteristics 2 and 3 that satisfy the conditions (gl) - (94) °f (%-4) and have the property that Gq is isomorphic to the Lie algebra of a reductive group. The last four items will deal, mainly, with the case where p > 3. Problem 3. Determine the absolute toral rank of all finite dimensional simple Lie algebras over algebraically closed fields of characteristic p > 3. One should stress here that the value of TR(L) is known for many simple Lie algebras L. In particular, it is known for all restricted Lie algebras of Cartan type. Some results related to Problem 3 can be found in [B-K-K 95]. The most interesting open case of Problem 3 is the case where L = H(m;n]UA,B)^ and ua,b £ hi (ra; n) is such that det .8 = 0.
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 211 Problem 4. Determine the automorphism groups of all finite dimensional simple Lie algebras over algebraically closed fields of characteristic p > 3. In particular, is it true that any finite dimensional simple Lie algebra L admits a Z-grading L = 0iGZ Li with Lq ^ L? Equivalently, is it true that the connected component of the algebraic group AutL is not unipotent? There are many examples of simple Lie algebras with solvable automorphism groups; in fact, such algebras occur in all four Cartan series. Probably, the most interesting open case of Problem 4 is the case where L = H(m\n;,UA,B)^ and wa,b £ hi(ra;n) is such that det B ^ 0. The next problem was suggested to the first author by R. Guralnick. Question 1 (cf. [G-K-P-S, Question 2.3]). Is it true that any finite dimensional simple Lie algebra over an algebraically closed field of characteristic p > 0 can be generated by two elements? For p G {2,3} Question 1 is out of reach at the moment. However, for p > 3, finite dimensional simple Lie algebras are likely to enjoy a much stronger property which is nowadays referred to as "one and a half generation". Problem 5 (cf. [G-K-P-S, Question 2.4]). Let L be a finite dimensional simple Lie algebra over an algebraically closed field of characteristic p > 3. Use Theorem 7 to prove that for any nonzero x G L there is y G L such that L = (x, y). The simplicity assumption on L in Problem 5 is crucial. Indeed, analyzing the semidirect products £(8,m) := (ldfl ® V) x (fl®0(m;l)), where $ is a finite dimensional simple Lie algebra over F and D is the commutative subalgebra of W(m; 1) spanned by di,..., 9m, one can observe that for any natural number n there exists a finite dimensional semisimple Lie algebra L over F such that the set Sn(L) := {x G L | (x, 2/1,..., yn) is solvable for all yu ..., yn G L} is nonzero. More precisely, it is not hard to see that for the semisimple Lie algebra L = £($, n + 1) one has This is in sharp contrast with the situation for finite groups; see [G-K-P-S, Theorem 1.1] for more detail. References [B 77] G.M. Benkart, On inner ideals and ad -nilpotent elements of Lie algebras, Trans. Amer. Math. Soc. 232 (1977), 61 - 81. [B-G 89] G.M. Benkart and T.B. Gregory, Graded Lie algebras with classical reductive null component, Math. Ann. 285 (1989), 85 - 98. [B-G-O-S-W] G.M. Benkart, T.B. Gregory, J.M. Osborn, H. Strade and R.L. Wilson, Isomorphism classes of Hamiltonian Lie algebras, Lect. Notes in Math., Vol. 1373, Springer-Verlag, Berlin/New York, 1989, pp. 42 - 57. [B-G-P] G.M. Benkart, T. Gregory and A. Premet, Recognition theorem for graded Lie algebras in prime characteristic, preprint arXiv:math./RA0508373v2, 154 pp. [B-K-K 95] G.M. Benkart, A.I. Kostrikin and M.I. Kuznetsov, Finite-dimensional simple Lie algebras with a nonsingular derivation, J. Algebra 171 (1995), 894 - 916. [B-O 84] G.M. Benkart and J.M. Osborn, Rank one Lie algebras, Ann. of Math. 119 (1984), 437 - 463.
212 ALEXANDER PREMET AND HELMUT STRADE [B-O-St 94] G.M. Benkart, J.M. Osborn and H. Strade, Contributions to the classification of simple modular Lie algebras, Trans. Amer. Math. Soc. 341 (1994), 227 - 252. [B 69] R.E. Block, Determination of the differentiably simple rings with a minimal ideal, Ann. of Math. 90 (1969), 433 - 459. [B-W 82] R.E. Block and R.L. Wilson, The simple Lie p-algebras of rank two, Ann. of Math. 115 (1982), 93 - 186. [B-W 88] — , Classification of the restricted simple Lie algebras, J. Algebra 114 (1988), 115 - 259. [Br 95] G. Brown, Families of simple Lie algebras of characteristic two, Comm. Algebra 23 (1995), 941 - 954. [C 09] E. Cartan, Les groupes de transformations continus, infinis, simples, Ann. Sci. Ecole Norm. Sup. 26 (1909), 93 - 161; Oeuvres completes, Tome 2, Partie II, Gauthier-Villars, Paris, 1953, pp. 857 - 925. [Cha 41] H.J. Chang, Uber Wittsche Lie-Ringe, Abh. Math. Sem. Univ. Hamburg 14 (1941), 151 - 184. [Che 56] C. Chevalley, Sur certains groupes simples, Tohoku Math. J. 7 (1956), 14 - 66. [Dem 70] S.P. Demushkin, Cartan subalgebras of the simple Lie p-algebras Wn and Sn, Sibirsk. Mat. Zh. 11 (1970), 310 - 325 [Russian]; Siberian Math. J. 11 (1970) 233 - 245. [English transl.] [Dem 72] — , Cartan subalgebras of simple Lie p-algebras, Izv. Akad. Nauk SSSR Ser. Mat. 36 (1972), 915 - 932 [Russian]; Math. USSR-Izv. 6 (1972), 905 - 924. [English transl.] [G-K-P-S] R. Guralnick, B. Kunyavskii, E. Plotkin, A. Shalev, Thompson-like characterization of the solvable radical, J. Algebra (to appear). [J 37] N. Jacobson, Abstract derivation and Lie algebras, Trans. Amer. Math. Soc. 42 (1937), 206 -224. [J 41] — , Classes of restricted Lie algebras of characteristic p, Amer. J. Math. 63 (1941), 481 - 515. [Kac 70] V.G. Kac, The classification of the simple Lie algebras over a field with nonzero characteristic, Izv. Akad. Nauk SSSR Ser. Mat. 34 (1970), 385 - 408 [Russian]; Math. USSR-Izv. 4 (1970), 391 - 413. [English transl.] [Kac 71] — , Global Cartan pseudogroups and simple Lie algebras of characteristic p, Uspekhi Mat. Nauk 26 (1971), 199 - 200. [Russian] [Kac 74] — , Description of filtered Lie algebras with which graded Lie algebras of Cartan type are associated, Akad. Nauk SSSR Ser. Mat. 38 (1974), 800 - 834; Errata, 40 (1976), 1415 [Russian]; Math USSR-Izv. 8 (1974), 801 - 835; Errata, 10 (1976), 1339. [English transl.] [Kap 58] I. Kaplansky, Lie algebras of characteristic p, Trans. Amer. Math. Soc. 89 (1958), 149 - 183. [Kir 89] S.A. Kirillov, Special Lie algebras of Cartan type, Preprint 247, IPF Acad. Nauk SSSR (1989). [Russian] [Kir 90] — , Hamiltonian Lie algebras of Cartan type, Preprint 257, IPF Acad. Nauk SSSR (1990). [Russian] [Ko 63] A. I. Kostrikin, Lie algebras and finite groups, Proc. Intern. Cogress Math., Stockholm, 1963, pp. 264 - 269 [Russian] [Ko 67] — , Squares of adjoint endomorphisms in simple Lie p-algebras, Izv. Acad. Nauk SSSR Ser. Mat. 31 (1967), 445 - 487 [Russian]; Math. USSR-Izv. 1 (1967) 435 - 473 [English transl.] [Ko 71] — , Variations modulaires sur un theme de Cartan, Actes Congres Intern. Math., Tome 1, Nice, 1971, pp. 285 - 292. [Ko 90] — , Around Bumside, Springer-Verlag, Berlin/Heidelberg, 1990. [Ko-S 66] A.I. Kostrikin and I.R. Shafarevich, Cartan pseudogroups and Lie p-algebras, Dokl. Akad. Nauk SSSR 168 (1966), 740 - 742 [Russian]; Soviet Math. Dokl. 7 (1966), 715 - 718. [English transl.] [Ko-S 69] — , Graded Lie algebras of finite characteristic, Izv. Akad. Nauk SSSR Ser. Mat. 33 (1969), 251 - 322 [Russian]; Math. USSR-Izv. 3 (1969), 237 - 304. [English transl.] [Kr 71] V.A. Kreknin, The existance of a maximal invariant subalgebra in simple Lie algebras of Cartan type, Mat. Zametki 9 (1971), 211- 222 [Russian]; Math. Notes 9 (1971), 124 - 130. [English transl.] [Ku 76] M.I. Kuznetsov, Simple modular Lie algebras with a solvable maximal subalgebra, Mat Sb. 101 (1976), 77 - 86 [Russian]; Math. USSR-Sb. 30 (1976) 68 - 76. [English transl.]
CLASSIFICATION OF FINITE DIMENSIONAL SIMPLE LIE ALGEBRAS 213 [Ku 89] —, Truncated induced modules over transitive Lie algebras of characteristic p, Izv. Acad. Nauk SSSR Ser. Mat. 53 (1989), 557 - 589 [Russian]; Math. USSR-Izv. 34 (1990), 575-608. [English transl.] [Ku 90] — , On Lie algebras of contact type, Comm. Algebra 18 (1990), 2943 - 3013. [Ku 91] — , The Melikyan algebras as Lie algebras of the type G2, Comm. Algebra 19 (1991), 1281 - 1312. [K-K 86a] M.I. Kuznetsov and S.A. Kirillov, Hamiltonian differential forms over an algebra of truncated polynomials, Uspekhi Mat. Nauk 41 (1986), 197 - 198 [Russian]; Russian Math. Surveys 41 (1986), 205 - 206. [English transl.] [K-K 86b] — , Contact forms over truncated polynomial algebras, Preprint 151, IPF Acad. Nauk SSSR (1986). [Russian] [M 80] G.M. Melikian, On simple Lie algebras of characteristic 5, Uspekhi Mat. Nauk 35 (1980), 203 - 204. [Russian] [M 82] — , Simple irreducible 2-graded Lie algebras with the component £0 — W\ 0 k, Deposit in VINITI No. 1688-82, 1982. [Russian] [M-Se 57] W.H. Mills and G.B. Seligman, Lie algebras of classical type, J. Math. Mech. 6 (1957), 519 - 548 [P 86a] A. Premet, Lie algebras without strong degeneration, Mat. Sb. 171 (1986), 140 - 153 [Russian]; Math. USSR-Sb. 57 (1987), 151 - 164. [English transl.] [P 86b] — , On Cartan subalgebras of Lie p-algebras, Izv. Acad. Nauk SSSR Ser. Mat. (1986) 50, 788 - 800 [Russian]; Math. USSR-Izv. 29 (1987), 145 - 157. [English transl.] [P 86c] — , Inner ideals of modular Lie algebras, Vesti Acad. Navuk BSSR Ser. Fiz.-Mat. Navuk (1986) No. 5, 11 - 15. [Russian] [P 89] — , Regular Cartan subalgebras and nilpotent elements in restricted Lie algebras, Mat. Sb. (1989) 180, 542 - 557 [Russian]; Math. USSR-Sb. (1990) 66, 555 - 570. [English transl.] [P 94] — , A generalization of Wilson's theorem on Cartan subalgebras of simple Lie algebras, J. Algebra 167 (1994), 641 - 703. [P-St 97] A. Premet and H. Strade, Simple Lie algebras of small characteristic I. Sandwich elements, J. Algebra 189 (1997), 419 - 480. [P-St 99] — , Simple Lie algebras of small characteristic II. Exceptional roots, J. Algebra 216 (1999), 190 - 301. [P-St 01] — , Simple Lie algebras of small characteristic III. The toral rank 2 case, J. Algebra 242 (2001), 236 - 337. [P-St 04] — , Simple Lie algebras of small characteristic IV. Solvable and classical roots, J. Algebra 278 (2004), 766 - 833. [P-St 06] — , Simple Lie algebras of small characteristic V. The non-Melikian case, preprint arXiv:math./RA0601323vl, 33 pp. [P-St 07] — , Simple Lie algebras of small characteristic VI. Completion of the classification (in preparation). [S 80] S. Serconek, Finite Cartan matrices and Lie algebras of classical type, J. Algebra 66 (1980), 594-599. [Sk 86] S.M. Skryabin, Canonical form of Hamiltonian and contact forms over divided power algebras, Deposit in VINITI No. 8594-B86, 1986. [Russian] [Sk 90] — Classification of Hamiltonian forms over divided power algebras, Mat. Sb. 181 (1990), 114-133 [Russian]; Math. USSR-Sb. 69 (1991), 121 - 141. [English transl.] [Sk 91] — Modular Lie algebras of Cartan type over algebraically non-closed fields. I, Comm. Algebra 19 (1991), 1629 -1741. [Sk 92] — New series of simple Lie algebras of characteristic 3, Ross. Akad. Nauk Mat. Sb. 183 (1992), 3 - 22 [Russian]; Russian Acad. Sci. Sb. Math. 76 (1993), 389 - 406. [English transl.] [Sk 93] — An algebraic approach to the Lie algebras of Cartan type, Comm. Algebra 21, 1229 - 1336. [Sk 95] — Modular Lie algebras of Cartan type over algebraically non-closed fields. II, Comm. Algebra 23 (1995), 1403-1453. [Sk 97] — , On the structure of the graded Lie algebra associated with a noncontractible filtration, J. Algebra 197 (1997), 178 - 230. [Sk 98] — , Toral rank one simple Lie algebras of low characteristics, J. Algebra 200 (1998), 650 - 700.
214 ALEXANDER PREMET AND HELMUT STRADE [St 77] H. Strade, Cartanalgebren in modularen Lie-Algebren, Comm. Algebra 5 (1977), 1335 - 1359. [St 89a] —, The absolute toral rank of a Lie algebra, Lect. Notes in Math., Vol. 1373, Springer- Verlag, Berlin/New York, 1989, pp. 1 - 28. [St 89b] — , The classification of the simple modular Lie algebras: I. Determination of the two- sections, Ann. of Math. 130 (1989), 643 - 677. [St 91] — , The classification of the simple modular Lie algebras: III. Solution of the classical case, Ann. of Math. 133 (1991), 577 - 604. [St 92] — , The classification of the simple modular Lie algebras: II. The toral structure, J. Algebra 151 (1992), 425 - 475. [St 93] — , The classification of the simple modular Lie algebras: IV. Determining the associated graded algebra, Ann. of Math. 138 (1993), 1 - 59. [St 94] — , The classification of the simple modular Lie algebras: V. Algebras with Hamiltonian two-sections, Abh. Math. Sem. Univ. Hamburg 64 (1994), 167 - 202. [St 98] — , The classification of the simple modular Lie algebras: VI. Solving the final case, Trans. Amer. Math. Soc. 350 (1998), 2553 - 2628. [St 04] — , Simple Lie Algebras over Fields of Positive Characteristic, Volume I: Structure Theory, DeGruyter Expositions in Math., Vol. 38, Berlin, 2004. [St-F 88] H. Strade and R. Farnsteiner, Modular Lie Algebras and their Representations, Marcel Dekker Textbooks and Monographs, Vol. 116, Dekker, New York, 1988. [St-W 91] H. Strade and R.L. Wilson, Classification of simple Lie algebras over algebraically closed fields of prime characteristic, Bull. Amer. Math. Soc. 24 (1991), 357 - 362. [T 78] S.A. Tyurin, The classification of deformations of special Lie algebras of Cartan type, Mat. Zametki 24 (1978), 847 - 857 [Russian]; Math. Notes 24 (1978), 948 - 954. [English transl.] [We 68] B. Ju. Weisfeiler, Infinite dimensional filtered Lie algebras and their connection with graded Lie algebras, Funktsional. Anal, i Prilozhen. 2 (1968), 94 - 95. [Russian] [We 78] — , On the structure of the minimal ideal of some graded Lie algebras of characteristic p > 0, J. Algebra 53 (1978), 34 - 361. [We 84] — , On subalgebras of simple Lie algebras of characteristic p > 0, Trans. Amer. Math. Soc. 286 (1984), 471 - 503. [Wil 69] R.L. Wilson, Nonclassical simple Lie algebras, Bull. Amer. Math. Soc. 75 (1969), 987 - 991. [Wil 71] — , Classification of generalized Witt algebras over algebraically closed fields, Trans. Amer. Math. Soc. 153 (1971), 191 - 120. [Wil 76] — , A structural characterization of the simple Lie algebras of generalized Cartan type over fields of prime characteristic, J. Algebra 40 (1976), 418 - 465. [Wil 77] — , Cartan subalgebras of simple Lie algebras, Trans. Amer. Math. Soc. 234 (1977), 435 - 446; Correction, 305 (1988), 851 - 855. [Wil 78] — , Simple Lie algebras of toral rank one, Trans. Amer. Math. Soc. 236 (1978), 287 -295. [Wil 80] — , Simple Lie algebras of type S, J. Algebra 62 (1980), 292 - 298. [Wil 83] — , Classification of the restricted simple Lie algebras with toral Cartan subalgebras, J. Algebra 83 (1983), 531 - 570. [Win 69] D. J. Winter, On the toral structure of Lie p-algebras, Acta Math. 123 (1969), 69 - 81. [Z 39] H. Zassenhaus, Uber Liesche Ringe mit Primzahlcharakteristik, Abh. Math. Sem. Hamburg 13, (1939) 1 - 100. School of Mathematics, The University of Manchester, Oxford Road, M13 9PL, United Kingdom E-mail address: sashapQmaths.man.ac.uk Fachbereich Mathematik, Universitat Hamburg, Bundesstrasse 55, 20146 Hamburg, Germany E-mail address: stradeQmath.uni-hamburg.de
Contemporary Mathematics Volume 413, 2006 From Quantum Groups to Unitary Modular Tensor Categories Eric C. Rowell ABSTRACT. Modular tensor categories are generalizations of the representation categories of quantum groups at roots of unity axiomatizing the properties necessary to produce 3-dimensional TQFTs. Although other constructions have since been found, quantum groups remain the most prolific source. Recently proposed applications to quantum computing have provided an impetus to understand and describe these examples as explicitly as possible, especially those that are "physically feasible." We survey the current status of the problem of producing unitary modular tensor categories from quantum groups, emphasizing explicit computations. 1. Introduction We outline the development of the theory of modular tensor categories from quantum groups with an eye towards new applications to quantum computing that motivate our point of view. In this article, we take quantum group to mean the "classical" g-deformation of the universal enveloping algebra of a simple complex finite dimensional Lie algebra as in the book by Lusztig [L], rather than the broader class of Hopf algebras the term sometimes describes. 1.1. Background. The representation theory of quantum groups has proven to be a useful tool and a fruitful source of examples in many areas of mathematics. The general definition of a quantum group (as a Hopf algebra) was given around 1985 by Drinfeld [D] and independently Jimbo [Ji] as a method for finding solutions to the quantum Yang-Baxter equation. These solutions led to new representations of Artin's braid group Bn and connections to link invariants. In fact, specializations of the famous polynomial invariants of Jones [J], the six-authored paper [HOMFLY] and Kauffman [Kf] have been obtained in this way. Reshetikhin and Turaev [RT] used this connection to derive invariants of 3-manifolds from modular Hopf algebras, examples of which can be found among quantum groups at roots of unity (see [RT] and [TWl], much simplified by constructions in [A]). When Witten [Wi] introduced the notion of a topological quantum field theory (TQFT) 2000 Mathematics Subject Classification. Primary 20G42; Secondary 20F46, 57R56. Key words and phrases, ribbon category, modular tensor category, quantum groups at roots of unity. ©2006 American Mathematical Society 215
216 ERIC C. ROWELL relating ideas from quantum field theory to manifold invariants, non-trivial examples were immediately available from the constructions in [RT] (after reconciling notation). Modular Hopf algebras were replaced by the more general framework of modular tensor categories (MTCs) by Turaev [Tl] (building on definitions in [Mc] and [JS]), axiomatizing the conditions necessary (and sufficient, see [T2], Introduction) to construct 3-dimensional TQFTs. Aside from the quantum group approach to MTCs, there are several other general constructions. Representation categories of Hopf algebra doubles of finite group algebras are examples of MTCs that are often included with quantum groups in the more general discussion of Hopf algebra constructions. A geometric construction using link invariants and tangle categories was introduced in [T2], advanced by Turaev and Wenzl in [TW2] and somewhat simplified by Blanchet and Beliakova in [BB]. However, all examples that have been carried out lead to MTCs also obtainable from quantum groups. Yet another construction of MTCs from representations of vertex operator algebras has recently appeared in a paper by Huang [Hu]. See Subsection 3.2 for further discussion of these approaches. Although it is expected that there are non-trivial examples of MTCs that do not arise from Hopf algebras (e. g. quantum groups and finite group algebras), none have been rigorously produced. This is probably due to the highly advanced state of the theory of representations of quantum groups at roots of unity provided by the pioneering work of many including Lusztig ([L]) and Andersen and his co-authors ([A], [AP] [APW]). The description of the MTCs derived from quantum groups can be understood with little more than a firm grasp on the theory of representations of simple finite-dimensional Lie algebras found in Humphrey's book [Hm] or any other introductory text. The purpose of this paper is two-fold: to survey what is known about the modularity and unitarity of categories arising from quantum groups at roots of unity, and to give useful combinatorial tools for explicit computations in these categories. For more in-depth developments the reader is directed to two references: 1) Bakalov and Kirillov's text ([BK], Sections 1.3 and 3.3) contains concise constructions and examples of quantum group MTCs, and 2) Sawin's paper ([S2]) gives a thorough treatment of the representation theoretic details necessary to construct MTCs from quantum groups. The modularity results described below partially overlap with Section 6 of [S2]. 1.2. Motivation. There are two fairly well-known motivations for studying MTCs. They are applications to low-dimensional topology (see [T2]), and confor- mal field theory (see [Hu] and references therein). Recently, an application of unitary MTCs to quantum computing has been proposed by Freedman and Kitaev and advanced in the series of papers ([FKW], [FKLW], [FLW] and [FNSWW]). Their topological model for quantum computing has a major advantage over the "classical" qubit model in that errors are corrected on the physical level and so has a higher error threshold. For a very readable introduction to topological quantum computing see [FKW]. In this model unitary MTCs play the role of the software, while the hardware is implemented via a quantum physical system and the interface between them is achieved by a 3-dimensional TQFT. The MTCs encode the symmetries of the corresponding physical systems (called topological states, see [FNSWW]), and must be unitary by physical considerations.
FROM QUANTUM GROUPS TO UNITARY MODULAR TENSOR CATEGORIES 217 Aside from the problem of constructing unitary MTCs, there are several open problems currently being studied related to the quantum computing applications. One question is whether the images of the irreducible unitary braid representations (see Remark 2.1) afforded by a unitary MTC are dense in the unitary group. This is related to a sine qua non of quantum computation known as universality. Progress towards answering this question has been made in [FLW] and was extended by Larsen, Wang, and the author in [LRW]. Another problem is to prove the conjecture of Z. Wang: There are finitely many MTCs of a fixed rank (see Subsection 2.2), This has been verified for ranks 1,2,3 and 4: see [Ol] and [02] for ranks 2 and 3 respectively, and [BRSW] for both ranks 3 and 4. It is with this conjecture in mind that we provide generating functions for ranks of categories in Subsection 4.7. Acknowledgements. The author wishes to thank the referees for especially careful readings of previous versions of this article and for comments leading to a much-improved exposition. Special thanks also to Z. Wang for many useful discussions on topological quantum computation. 2. General Definitions We give the basic categorical definitions for modular tensor categories, remark on some consequences and describe the crucial condition of unitarity. 2.1. Axioms. In this subsection we outline the axioms for the categories we are interested in. We follow the paper [Tl], and refer to that paper or the books by Turaev [T2] or Kassel [K] for a complete treatment. Let O be a category defined over a subfield fe C C. A modular tensor category is a semisimple ribbon ytfr-category O with finitely many isomorphism classes of simple objects satisfying a non-degeneracy condition. We unravel these adjectives with the following definitions. (1) A monoidal category is a category with a tensor product ® and an identity object 1 satisfying axioms that guarantee that the tensor product is associative (at least up to isomorphism) and that 1 <8> X * X <8> 1 = X for any object X. See [Mc] for details. (2) A monoidal category has duality if there is a dual object X* for each object X and morphisms bx : 1 -> X (8) X*, dx : X* (8) X -> 1 satisfying (Idx<8)dx)(&x<8)Idx) = Idx, (dx®Idx*)(ldx*®bx) = Idx*. The duality allows us to define duals of morphisms too: for any / G Hom(X,F) we define /* G Hom(F*,X*) by: /* = (dY (8) Idx*)(I<*Y* <8> / <8> Idx*)(Idy* (8) &x).
218 ERIC C. ROWELL (3) A braiding in a monoidal category is a natural family of isomorphisms cx,y :X®Y-+Y®X satisfying cx,y®z = (Idy ® cx,z)(cx,y ® Idz), CX07,Z = (CX,Z ® Idy )(Idx ® Cy,z). (4) A twist in a braided monoidal category is a natural family of isomorphisms 9X : X -+ X satisfying: Ox®y = cy,xCx,y{0x®0y)- (5) In the presence of a braiding, a twist and duality these structures are compatible if 9x- = (9xT. A braided monoidal category with a twist and a compatible duality is a ribbon category. (6) An ^46-category is a monoidal category in which all morphism spaces are A;-vector spaces and the composition and tensor product of morphisms are bilinear. (7) An ^46-category is semisimple if it has the property that every object X is isomorphic to a finite direct sum of simple objects-that is, objects Xi with End(Xi) = k satisfying the conclusion of Schur's Lemma: Hom(Xi,Xj)=0 for i^j. Turaev [T2] gives a weaker condition for semisimplicity avoiding direct sums, but we omit it for brevity's sake. (8) In a ribbon ytfr-category one may define a fc-linear trace of endomor- phisms. Let / G End(X) for some object X. Set: tr(f) = dxcx,x*(0xf ® Idx*)bx where the right hand side is an element of End(l) = k. The value of tr(Idx) is called the categorical dimension of X and denoted dim(X). (9) A semisimple ribbon ytfr-category is called a modular tensor category if it has finitely many isomorphism classes of simple objects enumerated as {Xq = 1, Xl, ..., Xn-i} and the so called S-matrix with entries Sij :=tr(cXj,Xi °cXi,xj) is invertible. Observe that S is a symmetric matrix. 2.2. Notation and Remarks. In a semisimple ribbon ytfr-category O with finitely many simple classes the set of simple classes generates a semiring over k under ® and 0. This ring is called the Grothendieck semiring and denoted Gr(0). If {X0 = 1, Xl, ..., Xn_i} is a set of representatives of these isomorphism classes, the rank of O is n. The axioms guarantee that we have (using Kirillov's notation [Ki]): (2.1) XitoXj^Y*1*****
FROM QUANTUM GROUPS TO UNITARY MODULAR TENSOR CATEGORIES 219 for some N^- G N. These structure coefficients of Gr{0) are called the fusion coefficients of O and (2.1) is sometimes called a fusion rule. Having fixed an ordering of simple objects as above, the fusion coefficients give us a representation of Gr(0) via Xi —► Ni where A^, (Ni)kj = (N*-) is called the fusion matrix associated to Xi> If we denote by z* the index of the simple object X*, the braiding and associativity constraints give us: It also follows from associativity that the fusion matrices pairwise commute, so that full fusion rules may sometimes be computed just from a single fusion matrix (i.e. using a Grobner basis algorithm). The first column (and row) of the 5-matrix consists of the categorical dimensions of the simple objects, i.e. Si$ = dim(Xi). We denote these dimensions by di. We also have that Sij = Sj^ = Si*j*. Since the twists 6x G End(X) for any object X, 6xi is a scalar map (as Xi is simple). We denote this scalar by 0^. Standard arguments show that the entries of the 5-matrix are determined by the categorical dimensions, the fusion rules and the twists on these simple classes, giving the following extremely useful formula (see [BK]): (2-2) 5^ = iEAr",A^- Provided O is modular the 5-matrix determines the fusion rules via the Verlinde formula (see [BK], and [Hu]). To express the formula we must introduce the quantity D2 = J2id2- Tnen: k _ ST" Sj,tSj,tSk*,t (2.3) Kkd = J2 D2So, This formula corresponds to the following fact: the columns of the 5-matrix are simultaneous eigenvectors for the fusion matrices A^, and the categorical dimensions are eigenvalues. Remark 2.1. The braiding morphisms cx,x induce a representation of the braid group Bn on End(X®n) for any object X via the operators Ri = Idf*-1 0 cx,x 0 Idf71"^1 G End(X®n) and the generators &i of Bn act by left composition by Ri. Remark 2.2. The term "modular" comes from the following fact: if we set T = (Sij6i)ij then the map: o -i\ sJi i 1 0 ) ' \0 defines a projective representation of the modular group SL(2,Z). In fact, by re- normalizing S and T one gets an honest representation of SL(2,Z).
220 ERIC C. ROWELL 2.3. Unitarity. A Hermitian ribbon ytfr-category has a conjugation: t:Hom(X,r)-+Hom(y,X) such that (/t)t = /, (f®g)i = /t(g)^t and (/o#)t = ^Q/t. On k c C, f must also act as the usual conjugation. Furthermore, f must be compatible with the other structures present i.e. (cx,y)f = (cx,y)_1, (Ox)* = (0x)-\ (6x)f = dxcx,x*(0x®ldx*), (dx)* = (Idx*®^1)^*,*)"1^. For Hermitian ribbon ytfr-categories the categorical dimensions di are always real numbers. If in addition the Hermitian form (f,g) = tr(fg^) is positive definite on Hom(X,Y) for any two objects X,Y £ O, the category is called unitary, and the categorical dimensions are positive real numbers. If O is unitary, then the morphism spaces End(X) are Hilbert spaces with the above form, and the representations Bn -+ End(X®n) described in Remark 2.1 are unitary. 3. Constructions MTCs have been derived in varying degrees of detail from several sources. A very general approach is through representations of quantum groups at roots of unity. We give a very broad outline of how these are obtained and mention a few other sources and constructions. 3.1. MTCs from Quantum Groups. The following construction is now standard, and can be found in more detail in the books by Turaev [T2] or Bakalov and Kirillov Jr. [BK] (both of which include examples). The procedure is a culmination of the work of many, but the major contributions following those of Drinfeld and Jimbo were from Lusztig (see [L]), Andersen and his collaborators ([APW],[A] and [AP]) and Turaev with Reshetikhin ([RT]) and Wenzl ([TW1]). Let $ be a Lie algebra from one of the infinite families ABCD or an exceptional Lie algebra of type E, F or G and q a complex number such that q2 is a primitive -fth root of unity, where £ is greater than or equal to the dual Coxeter number of $. Let U = Uq(&) be Lusztig's [L] "modified form" of the Drinfeld-Jimbo quantum group specialized at q and denote by T Andersen's [A] subcategory of tilting modules over U. A module V is called tilting if both V and its dual, V*, admit Weyl filtrations: i.e. sequences {0} = V0 C V1 C • • • C Vn = V with each Vi/Vi-i a Weyl module. The ratio of the square lengths of a long root to a short root will play an important role in the sequel, so we denote it by the letter m. Observe that m = 1 for Lie types ADE, m = 2 for Lie types BC and F, and m = 3 for Lie type G. It can be shown that T is a (non-semisimple) ribbon ytfr-category (see [A] and [TWl]). The ribbon structure on T comes from the (ribbon) Hopf algebra structure on U (see [ChP]), i.e. the antipode, comultiplication, .R-matrix, quantum Casimir etc. The set of indecomposable tilting modules with dim(X) = 0 (categorical dimension) generates a tensor ideal X C T, and semisimplicity is
FROM QUANTUM GROUPS TO UNITARY MODULAR TENSOR CATEGORIES 221 recovered by taking the quotient category F = T/X. Moreover, the category J7 has only finitely many isomorphism classes of simple objects, labelled by the subset of dominant weights (denoted P+) in the fundamental alcove: c( v f{A€P+:<A + p,tf0><<} ifm|* i[9)' |{AGP+:(A + p,^)<f} if mf* where $o is the highest root and $1 is the highest short root. Here the form (, ) is normalized so that (a, a) = 2 for short roots. While T is always a semisimple Hermitian ribbon ytfr-category with finitely many isomorphism classes of simple objects, the further properties (modularity and unitarity) of T depend on 0, the divisibility of £ by ro, and the specific choice of q. We denote the category T by C(g,£, q) to emphasize this dependence. The 5-matrices for these categories are well-known. For A,/i G Cg(g) we have: where p is the half sum of the positive roots and e(w) denotes the sign of the Weyl group element w. Remark 3.1. In practice, Formula (2.2) is often more useful than Formula (3.1) for computing the entries of the 5-matrix, as computing the twists 6\, g-dimensions d\ (see below) and fusion coefficients N% (via the quantum Racah formula, see [AP] and [S2]) is more straightforward than summing over the Weyl group. The twist coefficients for simple objects are also well known: 8\ = q(x>x+2ti, as are the categorical g-dimensions: where [n] = q ~q_t and <£+ is the set of positive roots. We note that the fusion coefficients of C(g,^, q) only depend on $ and L A complete description of the braiding and associativity maps is quite difficult in general; fortunately one is usually content to know they exist, relying on the S- matrix, fusion matrices and twists for most calculations. Remark 3.2. An issue has recently come to light regarding the explicit fusion rules for these categories. While Andersen-Paradowski [AP] showed that for many cases the fusion rules for the truncated tensor product in the category T are determined from the classical multiplicities by an anti-symmetrization over the affine Weyl group, their proof appeared in a paper that restricted attention to the root lattice. Evidently the first general proof of this "quantum Racah" formula is in the preprint [S2]. 3.2. Other Constructions. The most direct construction of MTCs comes from the representation category of the semidirect product D(G) := k[G] k ^(G) of the group algebra of a finite group with its (Hopf algebra) dual and can be found in the book [BK]. For example, the representation category of the Hopf algebra D(Ss) is a rank 8 MTC that does not arise from a quantum group construction as outlined above. These MTCs always have integer g-dimensions.
222 ERIC C. ROWELL The geometric construction of MTCs alluded to in the introduction is summarized as follows. One starts with a link invariant satisfying a number of mild (but technical) conditions and produces a new category from the category of tangles via an idempotent completion of quotients of endomorphism spaces. This produces a semisimple braided category, and if there is explicit information available for the link invariant one can sometimes verify the remaining axioms. This has been carried out for the Jones polynomial (Chapter XII of [T2]) and the Kauffman polynomial [TW2]. Blanchet and Beliakova [BB] gave a complete analysis of the modularity and modularizability of these categories corresponding to BMW algebras-the algebras supporting the Kauffman polynomial. Although the work in [BB] eliminated the need to appeal to quantum group characters as in [TW2], these constructions give rise to essentially the same MTCs as those obtained from quantum groups of types B, C and D at roots of unity. An advantage of this geometric approach is that the braid representations are more transparent than in the quantum group construction, although one pays for this convenience by having a less natural description of objects. As we noted in the introduction, MTCs have also been constructed from representation categories of certain vertex operator algebras (VOAs) by Huang [Hu]. Rigidity and modularity are the most difficult to verify, while the monoidal structure was previously obtained. The allure of this approach is that it includes a proof of a very general form of the Verlinde conjecture from conformal field theory. Although this VOA construction of MTCs is more difficult than other approaches, it gives credence to the thesis that MTCs describe symmetries in quantum physical systems. There are two indirect constructions that should be mentioned. One is the quantum double technique of Miiger [Mg] (inspired by the double of a Hopf algebra) by which an MTC is constructed by "doubling" a monoidal category with some further technical properties. An example of this approach is the finite group algebra construction mentioned above. Bruguieres [Br] describes conditions under which one may modularize a category that satisfies all of the axioms of an MTC except the invertibility of the 5-matrix (called a pre-modular category). This corresponds essentially to taking a quotient or sub-category that does satisfy the modularity axiom. 4. Modularity, Unitarity and Ranks for Quantum Groups There remains a fair amount of work to be done to have a complete theory of abstract unitary modular tensor categories; however, for quantum groups much is known. The condition of modularity has been settled for nearly all of the categories C($,^, g), as well as the question of unitarity. The modularity condition is often difficult to verify. Recently a modularity criterion was proved that sometimes simplifies the work (see [Br]): Theorem 4.1 (Bruguieres). Suppose O is a pre-modular category, and let {Xq = l,Xi,... ,Xn_i} be a set of representatives of the simple isomorphism classes. Then O is modular if and only if N := {Xi : Sij = didj for all X5} = {1}.
FROM QUANTUM GROUPS TO UNITARY MODULAR TENSOR CATEGORIES 223 si.r d Ar r+1 Br, r odd 2 Dr, r even 2 Dr, r odd 4 Eq 3 #7 2 Table 1. Values of d for Lie algebra types with d ^ 1 Observe that one has Soj = dodj = dj. The non-trivial elements of Af are the obstructions to modularity, i.e. the objects for which the corresponding columns in the 5-matrix are scalar multiples of the first column. In the following subsections we describe the modularity and unitarity of the categories C($,£, q), first for the cases that can be handled uniformly, and then for those that must be considered individually as well as a few subcategories of interest. Subsection 4.7 concerns the ranks of the categories C(g,£,q) and can be safely skipped by those readers not interested in this issue. 4.1. Uniform Cases m | £. For the categories C($,£, q) the cases where £ is divisible by m have been mainly studied in the literature. The invertibility of S for Lie types A and C with q = e*1^ was shown in [TW1] using the work of Kac and Peterson [KP], and a complete treatment (for all Lie types with q = e*1/1) is found in [Ki]. The invertibility can be extended to other values of q by the following Galois argument, which is found in [TW2] in a different form. By Formula (3.1) we see that the entries of the 5-matrix: Sx» = (const.) J2 e(w)q2^x+^+ri wew are polynomials in ql/d where d G N is minimal so that d(\ ji) G Z for all weights A,/i. Thus det(5) is non-zero for any Galois conjugate of e7™/^, i.e. for any q = eZ7Tl^ with gcd(z,cW) = 1. Table 1 lists the values of d for all Lie types for which d ^ 1. Notice that there are sub-cases for types B and D. When d = 1 and m\£ the uniform case covers all possibilities, since then the condition gcd(z, d£) = 1 is equivalent to the original assumption that q2 is a primitive ^th root of unity. So when m\£, the cases Br with r even, Cr, Eg, F4, and G2 do not require further attention. If m = 1 and gcd(^, d) ^ 1 the condition gcd(z, d£) = 1 also degenerates to the original assumption that q2 is a primitive ^th root of unity so we need not consider Dr with £ even, Eq with $\£ or E7 with £ even. Following a conjecture of Kirillov Jr. [Ki], Wenzl [W] showed that the Hermit- ian form on C(q,£, q) is positive definite for the uniform cases for certain values of q, and Xu [X] independently showed some of the cases covered by Wenzl. Their results are summarized in: Theorem 4.2 (Wenzl/Xu). The categories C(g,£, q) are unitary whenm\£ and 4.2. Type A. For Lie type Ar corresponding to q = 5[r+i we have m = \ and d = r +1. Bruguieres [Br] shows that one has modularity for q = eZ7ri/e if and only if gcd(z, (r + 1)£) = 1. Moreover, Masbaum and Wenzl [MW] show that when gcd(£,r + 1) = 1 the subcategory of C(5[r+i,^, q) generated by the simple objects labelled by integer weights is a modular subcategory whose rank is l/(r + 1) times the rank of the full category. There are a number of other proofs of this fact, see e.g. [Br] Section 5. Denote this subcategory by Z(Ar), and see Subsection 5.1.
224 ERIC C. ROWELL 4.3. Type B, £ odd. The category C($02r+i,£,Q) with £ odd has been considered to some extent by several authors including Sawin [SI], [S2] and Le-Turaev [LT]. It is shown in ([TW2], Theorem 9.9) that if £ is odd, the subcategory of C(s02r+i»^j<z) generated by simple objects labelled by integer weights is modular and has rank exactly half of that of C(so2r+iA-> o)- Combining the computations in [Rl] and the modularity criterion of [Br] one has: Theorem 4.3. The category C(s02r+i» ^» o) w^ t °dd is modular if and only if q£ = — 1 and r is odd. Proof. By the modularity criterion we wish to show that there are obstructions to modularity (i.e. non-trivial objects in the set A/*, see Theorem 4.1) if and only if the conditions of the theorem are not satisfied. By the modularity of the subcategory generated by simple objects labelled by integer weights, any obstructing object must be labelled by a half-integer weight. In [Rl] the object X7 labelled by the (half-integer) weight that is furthest from the 0 weight in the fundamental alcove is shown to induce an involution of the fundamental alcove (by tensoring with Xy) that preserves g-dimension up to a sign. This implies that Xy is the only potential obstruction to modularity. In [Rl] (Scholium 4.11) the signs of the g-dimensions are analyzed, and the theorem then follows from the explicit computations of N" x, d\ and 6\ (also found in [Rl]) together with Formula (2.2) and the obstruction equation 57?a = dyd\. □ The subject of the author's thesis [R2] (the results of which can be found in [Rl]) is the question of unitarizability of the family of categories C(so2r+i,^#) with £ odd. Using an analysis of the characters of the Grothendieck semirings it is shown that no member of this family of categories is unitary. In fact, there is a much stronger statement, for which we need the following definition: Definition 4.4. A pre-modular category O is called unitarizable if O is tensor equivalent to a unitary pre-modular category O'. By tensor equivalent we mean there exists a functor preserving the monoidal structure that is bijective on morphisms and such that every object in the target category is isomorphic to an object in the image of the functor. Using a structure theorem of Tuba and Wenzl [TbW] it is shown in [R2] that: Theorem 4.5. Fix q with q2 a primitive £th root of unity, £ odd, and r satisfying 2(2r + 1) < £. Then no braided tensor category having the same Grothendieck semiring as C(s02r+i>^<z) is unitarizable. Remark 4.6. When £ < 2(2r + 1) the rank of C(so2r+i, ^, q) is relatively small and the fusion rules of the category may coincide with those of another category that is known to be unitarizable. For example C(s05,7,g) has the same Grothendieck semiring as C(sl2> 7, q) which is unitary for q = e7™/7. 4.4. Type C, £ odd. For type C one has m = 2, so it remains to analyze the cases with £ odd. For this, we resort to the "rank-level duality" result of [Rl] (Corollary 6.6) showing that the categories C(so2r+i > A q) and C($pi_2r-i, £, q) are tensor equivalent. Theorem 4.5 immediately implies these categories are not unitarizable for £ odd if 2(2r -f 1) < L Moreover, the technique in the proof of Theorem 4.3 can be applied to this case using the explicit values of d\ and 6\ and
FROM QUANTUM GROUPS TO UNITARY MODULAR TENSOR CATEGORIES 225 the image of the object Xy under the tensor equivalence afforded by this rank-level duality. We then have: Theorem 4.7. If£ is odd, the categories C($p2r,£,q) are not modular and if in addition 2(2r + 1) < £ they are not unitarizable. 4.5. Remaining Types Z), Eq and E7 Cases. The only remaining question for the sub-cases not covered by the uniform case is whether the condition gcd(z, d£) = 1 is necessary for modularity. For Lie types D and E? the sub-cases correspond to £ odd, and for Lie type Eq the sub-cases correspond to 3 \ £. In our opinion this question is still open, of limited interest and one probably does not get modularity. 4.6. Types F4 with £ odd, and G2 with 3 \ £. To our knowledge both the question of modularity and unitarizability are still open for F4 with £ odd and G2 with 3 \ £. In light of the results in the Lie types B for £ odd (see Theorems 4.3 and 4.5), one might expect to find that these categories are not unitarizable (except possibly for small £), but sometimes modular. 4.7. Generating Functions for |Cg(fl)|. For applications it is useful to know the ranks of the categories C($, £, q). We define an auxiliary label £m — 0 if m \ £ and £m = 1 if m \ £ for notational convenience. We reduce the problem of determining the cardinalities of the labeling sets Ct(g) of simple objects to counting partitions of n with parts in a fixed (finite) multiset <S($, £m) that depends only on the rank and Lie type of q and the divisibility of £ by m. Fix a simple Lie algebra q of rank r and a positive integer £. Let A = Y2iai^i De a dominant weight of g written as an N-linear combination of fundamental weights A*. To determine if A G C^($), we compute: r i where j = 0 or 1 depending on if m \ £ or not. Setting L)?' = (Af,t?j) we see that the condition that A G Ct{%) becomes: k i Since a^, Vf' G N we have: Lemma 4.8. The cardinality ofCt(g) is the number of partitions of all natural numbers n, 0 < n < £ — (p, $7) — 1 into parts from the size r = rank(g) multiset $b,tm) = [L?)ri. So it remains only to compute the numbers (p,$j) and L^ (with j — 0,1) for each Lie algebra q and integer £ > (p,t?j) and to apply standard combinatorics to count the number of partitions into parts in S(g,£m). The first task is easily accomplished with the help of the book [Bo]. Table 2 lists the results of these computations, where £q := min{^ : £ > (p, fij) -f 1} is the minimal non-degenerate value of L Let P-r(n) denote the number of partitions of n into parts in a multiset T, and Pr[s] = Y2n=o ^(n) tne number of partitions of all integers 0 < n < s into parts
226 ERIC C. ROWELL Table 2. C(g,q,£) Data xr J\f Br, £ odd Br, £ even Cr, £ odd Cr, £ even Dr Eq E7 Eg F4, £ even F4, I odd G2, 3 | * G2, 3{^ 5(0,O [1,...,1] [1,2,...,2] [2,2,4,...,4] [1,2,...,2] [2,...,2] [1,1,1,2,...,2] [1,1,2,2,2,3] [1,2,2,2,3,3,4] [2,2,3,3,4,4,5,6] [2,4,4,6] [2,2,3,4] [3,6] [2,3] £0 r + 1 2r + l 4r-2 2r + l 2r + 2 2r-2 12 18 30 18 13 12 7 from the multiset T. Any standard reference on generating functions (see e.g. [Sn]) will provide enough details about generating functions to prove the following: Lemma 4.9. The number Pr(w<) of partitions of n into parts from the multiset T has generating function: 1 00 I1t^ = E^w^ teT n=0 while the number Pr [s] of partitions of all n G N with 0 < n < s into parts from the multiset T has generating function: teT s=o Applying this lemma to the sets 5(g,fm) given in Table 2 we obtain: Theorem 4.10. Define 1 rr 1 *i,U*) n X AX 1 keS(9,£m) Then the rank \Ct(&)\ of the pre-modular category C($,q,£) is the coefficient of x .t-to+tm in the Taylor series expansion of Fg^rn(x). Proof. It is clear from Lemma 4.9 that the coefficients of generating function Fgjm(x) counts the appropriate partitions. The coefficient of x that gives the rank for a specific £ is shifted by the minimal non-degenerate £0, which corresponds to the x° = 1 term if m \ £ and to the x1 = x term of m \ £, hence the correction by xirn. With this normalization only the coefficients of those powers of x divisible (resp. indivisible) by m give ranks corresponding to £ divisible (resp. indivisible) by m. □ We illustrate the application of this theorem with some examples.
FROM QUANTUM GROUPS TO UNITARY MODULAR TENSOR CATEGORIES 227 Example 4.11. Let $ be of type G2. (a) Let I = 27. Then £m = 0 and £0 = 12. So the rank of C(fl(G2),g,27) is given by the (27 - 12 + 0) = 15th coefficient of 1 = (l + x + x2)(l + 2x3 + 4x6 + 6x9 + 9x12 + 12x15 + • • •) (l-x)(l-x3)(l-x6) so|C27(8(G2))| = 12. (b) Let £ = 14. Then im = 1 and £0 = 7. So |Ci4(0(G2))| is the (14 - 7 + l)th coefficient of 1 = l + x + 2x2 + 3x3 + 4x4 + 5x5 + 7x6 + 8x7 + 10x8 • • • (l-x)(l-x2)(l-x3) so the rank of C(g(G2), q, 14) is 10. 5. Examples We provide examples of two pre-modular categories, one of which is modular and unitary, while the other is not modular but has a (non-unitary) modular subcategory. We only give enough information to discuss the modularity and unitarity of the category. 5.1. TypeZ(^i)at^ = 5. The following MTC is obtained from C(sl2,5, e^/5) by taking the subcategory of modules with integer highest weights. There are two simple objects l,and X\ satisfying fusion rules: X\®X\ = 10Xl and l®Xi = Xi. 1+v^ 2 and the twists: 60 = 1, 6X = e4™/5. It is clear that det(5) ^ 0, and it follows from [W] that the category is unitary (notice that the categorical dimensions are both positive). 5.2. Type i?2 at 9th Roots of Unity. Consider the pre-modular categories C(505,9, eJ7™/9) with gcd(18, j) = 1. There are 12 inequivalent isomorphism classes of simple objects. The simple iso-classes of objects are labelled by (Ai, A2) G ^(N2) with Ai > A2. The twist coefficients for X\ is g(A+2^'A) where the form is twice the usual Euclidean form. The obstruction to modularity mentioned in the proof of Theorem 4.3 is labelled by 7 := |(5,5) The categorical dimension function is: [2(Ai + A2 + 2)] [2(Ai - A2 + 1)] [2Ai + 3] [2A2 + 1] A: [4][3][2][1] One checks that the simple object X1 is indeed the cause of the singularity of the 5-matrix, that is, S1^\ = d7d\ for all A. Thus this category is not modular by Bruguieres' criterion, Theorem 4.1. Now let us consider the subcategory of C(505,9,eJ7n/9) with gcd(18,j) = 1 generated by the simple objects labelled by integer weights: {(0,0), (1,0), (2,0), (1,1), (2,1), (2,2)}. The braiding and twists from the full category restrict, so the entries of the S- matrix are computed from Formula (2.2). Taking the ordering of simple objects
228 ERIC C. ROWELL above, we denote the categorical dimensions by di 0 < i < 5. The fusion matrix corresponding to (1,0) is: 0 0\ 0 0 1 0 10' 1 1 1 V It is not hard to show that Ni determines the other five fusion matrices by observing that Ni has six distinct eigenvalues and the fusion matrices commute. There are a total of six categories corresponding to the six possible values of q. To describe the 5-matrices we let a be a primitive 18th root of unity, and set 7*1 = —a — a2 + a5, r*2 = ol + ol2 — a4 and r3 = a4 — a5. Then we get the following 5-matrices (for the 6 choices of a): / 1 r2 r3 1 -1 n \ r2 1 1 ri -r3 1 r3 1 1 r2 -ri 1 1 n r2 1 -1 r3 -1 -r3 -n -1 1 -r2 I \n 1 1 r3 -r2 1 / One checks that det(5) 7^ 0 for any a, so these categories are modular. A bit of Galois theory shows that there are only three distinct S for the six choices of a. Notice that it is already clear that the first column of S is never positive, since both 1 and —1 appear regardless of the choice of a. So none of these categories is unitary. References [A] H.H. Andersen, Tensor products of quantized tilting modules, Comm. Math. Phys. 149 (1991), 149-159. [AP] H.H. Andersen and J. Paradowski, Fusion categories arising from semisimple Lie algebras, Comm. Math. Phys. 169 (1995), 563-588. [APW] H. H. Andersen, P. Polo, K. X. Wen, Representations of quantum algebras. Invent. Math. 104 (1991) no. 1, 1-59. [BK] B. Bakalov and A. Kirillov Jr., Lectures on tensor categories and modular functors. Amer. Math. Soc, Providence, 2001. [BB] A. Beliakova and C. Blanchet, Modular categories of type B, C and D, Comment. Math. Helv. 76 (2001), 467-500. [BRSW] S. Belinschi, E. Rowell, R. Stong and Z. Wang, Classification of Modular Tensor Categories I: Low-rank cases, in preparation. [Bo] N. Bourbaki, Groupes et algebres de Lie, Chap. 4-6 (1968), Hermann, Paris. [Br] A. Bruguieres, Categories premodulaires, modularisations et invariants des varietes de dimension 3, Math. Ann. 316 (2000), 215-236 (French). [ChP] V. Chari and A. Pressley, A Guide to Quantum Groups. Cambridge University Press, Cambridge, 1994. [D] V. G. Drinfeld, Quantum groups, in: Proceedings of the International Congr. Math. Berkeley, 1986. [FKLW] M. Freedman, A. Kitaev, M. Larsen, Z Wang, Topological quantum computation, Mathematical challenges of the 21st century. Bull. Amer. Math. Soc. (N.S.) 40 (2003) no. 1, 31-38. [FKW] M. Freedman, A. Kitaev, and Z. Wang, Simulation of topological field theories by quantum computers, Comm. Math. Phys. 227 (2002) no. 3, 605-622. /o 1 0 0 0 V> 1 0 1 1 0 0 0 1 0 0 1 0 0 1 0 1 1 0
FROM QUANTUM GROUPS TO UNITARY MODULAR TENSOR CATEGORIES 229 [FLW] M. Freedman, M. Larsen and Z. Wang, The two-eigenvalue problem and density of Jones representation of braid groups, Comm. Math. Phys. 228 (2002) no. 1, 177-199. [FNSWW] M. Freedman, C. Nayak, K. Shtengel, K. Walker and Z. Wang, A class of P,T- invariant topological phases of interacting electrons, Ann. Physics 310 (2004) no. 2, 428-492. [HOMFLY] P. Freyd, D. Yetter, J. Hoste, W. Lickorish, K. Millett, A. Ocneanu, A new polynomial invariant of knots and links. Bull. Amer. Math. Soc. (N.S.) 12 (1985) no. 2, 239-246. [Hu] Y.-Z. Huang, Vertex operator algebras, the Verlinde conjecture and modular tensor categories, Proc. Natl. Acad. Sci. USA 102 (2005) no. 15, 5352-5356. [Hm] J. E. Humphreys, Introduction to Lie algebras and representation theory. Graduate Texts in Mathematics, Vol. 9. Springer-Verlag, New York-Berlin, 1972. [Ji] M. Jimbo, A q-difference analogue of U(g) and the Yang-Baxter equation, Lett. Math. Phys. 10 (1985) no. 1, 63-69. [J] V. F. R. Jones, A polynomial invariant of knots via von Neumann algebras, Bull. Amer. Math. Soc. 12 (1985), 103-112. [JS] A. Joyal and R. Street, The geometry of tensor calculus. I. Adv. Math. 88 (1991) no. 1, 55-112. [KP] V. Kac and D. Peterson, Infinite Dimensional Lie Algebras, theta functions and modular forms, Adv. Math. 53 (1984), 125-264. [K] C. Kassel, Quantum Groups. Graduate Texts in Mathematics, Vol. 155. Springer-Verlag, New York, 1995. [Kf] L. H. Kaufmann, An invariant of regular isotopy, Trans. Amer. Math. Soc. 318 (1990) no. 2, 417-471. [Ki] A. Kirillov Jr., On an inner product in modular categories, J. of Amer. Math. Soc. 9 (1996) no. 4, 1135-1169. [LRW] M. Larsen, E. Rowell and Z. Wang, The N-eigenvalue problems and two applications, Int. Math. Res. Not. 2005 (2005) no. 64, 3987-4018. [LT] T. Le and V. Turaev, Quantum groups and ribbon G-categories, J. Pure Appl. Algebra 178 (2003), 169-185. [L] G. Lusztig, Introduction to quantum groups. Progress in Mathematics, Vol. 110. Birkhauser Boston, Inc., Boston, MA, 1993. [Mc] S. MacLane, Categories for the working mathematician, Graduate Texts in Math. Vol. 5. Springer-Verlag, 1971. [MW] G. Masbaum and H. Wenzl, Integral modular categories and integrality of quantum invariants at roots of unity of prime order. J. Reine Angew. Math. 505 (1998), 209-235. [Mg] M. Miiger, From subfactors to topology II. The quantum double of tensor categories and subfactors, J. Pure Appl. Algebra, 180 (2003) no. 1-2, 159-219. [Ol] V. Ostrik, Fusion categories of rank 2, Math. Res. Lett. 10 (2003), 177-183. [02] V. Ostrik, Pre-modular categories of rank 3, preprint arXiv, math.QA/0503564. [RT] N. Reshetikhin and V. Turaev, Invariants of ^-manifolds via link polynomials and quantum groups, Invent. Math. 103 (1991), no. 3, 547-597. [Rl] E. C. Rowell, On a family of non-unitarizable ribbon categories, Math. Z. 250 (2005) no. 4, 745-774. [R2] E. C. Rowell Tensor categories arising from quantum groups and BMW-algebras at odd roots of unity, thesis, U.C. San Diego, 2003. [SI] S. Sawin, Jones-Witten invariants for nonsimply connected Lie groups and the geometry of the Weyl alcove, Adv. Math. 165 (2002), no. 1, 1-34. [S2] S. Sawin, Quantum groups at roots of unity and modularity, preprint arXiv, math.Q A/0308281. [Sn] R. Stanley, Enumerative combinatorics, Vol. 2, Cambridge University Press, Cambridge, 1999. [Tl] V. Turaev, Modular categories and 3-manifold invariants, Int. J. of Modern Phys. B, 6 (1992) no. 11-12, 1807-1824. [T2] V. Turaev, Quantum invariants of knots and 3-manifolds, de Gruyter Studies in Math., Vol. 18. Walter de Gruyter & Co., Berlin, 1994. [TWl] V. Turaev and H. Wenzl, Quantum invariants of 3-manifolds associated with classical simple Lie algebras, Int. J. of Modern Math. 4 (1993), 323-358. [TW2] V. Turaev, H. Wenzl, Semisimple and modular categories from link invariants. Math. Ann. 309 (1997) no. 3, 411-461.
230 ERIC C. ROWELL [TbW] I. Tuba and H. Wenzl, On braided tensor categories of type BCD, J. Reine Angew. Math. 581 (2005), 31-69. [W] H. Wenzl, C* tensor categories from quantum groups. J. Amer. Math. Soc. 11 (1998) no. 2, 261-282. [Wi] E. Witten, Topological quantum field theory, Comm. Math. Phys. 117 (1988), 353-386. [X] F. Xu, Standard X-lattices from quantum groups. Invent. Math. 134 (1998) no. 3, 455-487. Department of Mathematics, Indiana University, Bloomington, IN 47405 E-mail address: errowellQindiana.edu
Contemporary Mathematics Volume 413, 2006 A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER TO CANONICAL BASES OF QUANTUM AFFINE ALGEBRAS JIE XIAO AND GUANGLIAN ZHANG 0. Introduction In [Rl] C.M.Ringel defined the Hall algebras for the representation categories of hereditary algebras of Dynkin type and pointed out a close relation with the quantized enveloping algebras of semisimple Lie algebras in the sense of Drinfeld [Drl] and Jimbo [J]. This opened a new direction for representation theory. Now the Hall algebra in the sense of Ringel and its relation with Lie algebras and quantum groups have been intensively studied. In [DX], a survey for an overview of this subject was given but it emphasized on the explicit connection between the structure of the whole double Ringel-Hall algebras and that of the quantized enveloping algebras of generalized Kac-Moody algebras in the sense of Borcherds [B](see[Ka]). In the present notes, we will demonstrate that the Ringel-Hall algebra approach can provide an explicit correspondence between representations of tame quivers and base elements of affine Lie algebras and quantum affine algebras, in particular, elements in the canonical bases which was defined by Lusztig in [LI]. We use the Kronecker quiver as a model for easier understanding and the main results presented here were taken from [LXZ], [DDX] and others. However we should remark here that an algebraic construction of the canonical bases for a quantum affine algebra has been given in [BCP] and [BN] (see [N] for a recent survey). Finally we should mention that Kapronov's work [Kap] on the Hall algebra of coherent sheaves over the projective line and its very important developments by Schiffmann in [S2] and by Lin and Peng in [LP], in particular, the canonical bases given in [S4] and [S5]. In the present notes we don't follow this line but in Section 2 we point out the similar category structure between the representations of Kronecker quiver and the coherent sheaves over the projective line. 1. Representations of the Kronecker quiver 1.1 A quiver Q is just a directed graph. The Kronecker quiver K is a directed graph with two vertices 1, 2 and two arrows a, (3 from 2 to 1, i.e, ex. (3 Supported in part by the NSF of China (10471071) and by the 973 project of the Ministry of Science and Technology of China. 2000 Mathematics Subject Classification. Primary 16G10, 17B37; Secondary 16G20, 17B67. ©2006 American Mathematical Society 231
232 JIE XIAO AND GUANGLIAN ZHANG Let A; be a field ( C or finite field). A representation of K is defined as a collection of A;-vector spaces Vi for each vertex i G {1,2} and a linear map (fh : V2 —► V\ for each arrow h G {a, /?} : When we take Vi = &m, V2 = &n, the representation has the following form: nXm A morphism from (Vi,(fh) to (V/, ^) is defined to be a collection of linear maps fi for each vertex i e {1,2} such that fi<pa = <Paf2ifi<Pp — ^Jg/2, that is, the following diagram commutes The same definition holds for representations of a general quiver Q. We can thus form the abelian category rep(Q, k) whose objects are the representations of the quiver Q and whose morphisms are as just described. 1.2 We also use K to denote the Kronecker algebra, that is, the path algebra of the Kronecker quiver. It is well known that the category of finite dimensional modules of if, denoted by mod-if, is equivalent to rep(if, k). For each if-module M, we denote by dimM the image of M in the Grothendieck group Kq(K) of the Kronecker algebra K, which is called the dimension vector of M. Cast in terms of rep (if, &), the dimension vector of representation V is given by dimF = (dimVi,dimF2) G Z|0. The classification of the indecomp>osable if-modules goes back to Kronecker and Weierstrass [Kr]. Using the language of representations of quivers, the classification of the indecomposable if-modules can be explained via the Auslander-Reiten quiver. By [R6], the Auslander-Reiten quiver of if consists of one preprojective component, one preinjective component and regular components. All indecomposable representations in the preprojective component, called indecomposable preprojective representations, have the dimension vectors (n+1, n) and all indecomposable preinjective representations have the dimension vectors (n, n +1). All other components, called regular components, consist of one family of separating homogeneous tubes T(p) = ZAqo/1, for p G F1(A;), where we may understand F1(k) as spec(&[x]) if k is the finite field ¥q (see the next section). All indecomposable regular if-modules have the dimension vectors (n, n) for n G N.
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 233 The indecomposable preprojective representations of dimension vector (n +1, n) are isomorphic to the representation given by two maps from kn to &n+1 in matrices ( ™ 1 and ( T 1 , and the indecomposable preinjective representations of dimension vector (n, n + 1) are isomorphic to the representation given by two maps from &n+1 to kn in matrices ( In 0 ) and ( 0 In ). Each homogeneous tube T(p) is a serial category. The indecomposable object in T(p) has a unique chain of subobjects belonging to T(p) with all composition factors isomorphic to the quasi-simple object V = (Vi,V^;<T,r), where V\ = V2 = ¥q[x]/(p(x)) for an irreducible polynomial p(x) in ¥q[x] and a is the identity map and r is given by the multiplication by x, except for p = (0,1) G P1(fc) where the quasi-simple object is V = (¥q,¥q; 0,1). The Auslander-Reiten quiver of the Kronecker quiver looks as follows. PR I Figurel The Auslander-Reiten quiver provides us much information about the module category, for instance, the following directing property and the total order we will give in Section 6 and Section 8 for the dimension vectors of indecomposable modules. If P, R and I are, respectively, preprojective, regular and preinjective modules, then we have following properties (RomA(R, P) = Honu(J, P) = HomA(I, R) = 0 Ext\(P, R) = Ext\(P, I) = Ext\(R, I) = 0. The dimension vectors of the indecomposable if-modules are given by: dimM = (n + l,n), or (n,n + 1) (real roots), for nonnegative integers n and dimM = (n,n) (imaginary roots), for positive integers n. Obviously this set of dimension vectors of indecomposable if-modules naturally identifies with the set of positive roots of the affine Lie algebra 5/2 (type A[*). 1.3 This observation is not occasional in the representation theory of quivers. Let Q be a finite quiver. For the path algebra A of Q, we have the following result : {A is a finite representation type <=> Q is a Dynkin quiver A is a tame representation type <=> Q is an affine quiver All other cases are wild type. Moreover, the dimension vectors of indecomposable A-modules are exactly the positive roots of the corresponding Kac-Moody algebras; and the number or parameter
234 JIE XIAO AND GUANGLIAN ZHANG of the isomorphism classes of indecomposable ^4-modules for a fixed dimension vector (over finite field and C, respectively) have some deep meaning in geometry and Lie theory. These results are known due to Gabriel [Ga] for finite type, Dlab-Ringel [DR] for tame type and Kac [Ka2] for general cases. We may ask the following question. Question 1 : Could we use the above observation to find a better way to connect representations of quivers with Kac-Moody algebras more directly? 2. Coherent sheaves over the projective line F1(k) 2.1 Let X be a projective curve over k (C or finite field). A vector bundle T over X is a coherent Ox-module, locally free of finite rank. A line bundle C over X is a vector bundle of rank 1. If X is a projective curve, a coherent Ox -module Q over X is called a torsion sheaf if the rank of Qx is 0 for any x E X. We only consider the projective line in this section. Let us consider the homogeneous coordinates (t : u) on F1(k). The two affine open subsets {Uf = (t,u)\t ^ 0} and {U" = (t,u)\u ^ 0} cover F1(k). That is, the formulae z = u/t and z~x = t/u define coordinates on U' and [/", respectively. The rings k[z] and A;[z_1] are the rings of regular functions on U' and /7", respectively. If A is a commutative domain, M an ^4-module and z £ A, then Mz denotes the localized ^4-module obtained from M by inverting z. An object of the category Coh(P1 (k)) of coherent sheaves on P1 (k) can ba interpreted as a triple (M', M", (/?), where M' is a finitely generated k[z]-module, M" is a finitely generated &[z-1]- module, and ip : M'z —► M"_i is an isomorphism of k[z, z~^-modules. A morphism in Coh(P1(A;)) from {M'^M",^) to {Nf,N",ip) is a pair of maps (/',/"), where /' : M' —► M" is a A;[z]-linear map and /" : N' —► N" is a k[z~^-linear map such that V/^ = /"-!</>• 2.2 By Grothendieck [Gr] and Beilinson [Be] we have for M G Coh(P1(A;)), where T and T are the vector bundle and the torsion parts of M, respectively. A more explicit decomposition is r i=\ and T^ 0 % and Tx s O^ ® ■ ■ ■ © £>£'. x£supp(T) In the case of P1 (A;), for any n G Z, the sheaf 0(n) = (M7, M7/, tp) defined by letting M' = k[z], M" = k[z_1] and (p : k[z,z~x) —► k[z, z~l] being the multiplication by z~n, is indecomposable vector bundle (line bundle) and any vector bundle is isomorphic to a direct sum of finitely many 0{n)'s. Let x G F1(k) be a closed point given by an irreducible homogeneous polynomial P G k[X, Y] and O™ = (M',M",<p), where M' = k[z]/(P(l,z)n), M" = k[z]/(P(z, l)n) and ip is induced
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 235 by the identity of k[z, z~1]. The sheaf O™ is an indecomposable torsion sheaf. Note that supp(T) is always finite for T G Fl(k). By some standard results of algebraic geometry, we have (1) Hom(T, F) = Ext1(^r, T) = 0 if T is a locally free sheaf and T is a torsion sheaf. (2) Hom(<?i,<?2) = Ext1 (^1,^2) = 0 if Q\ and Q2 are torsion sheaves with disjoint supports. (3) Ext2(—, —) = 0 identically, so that the category CohP^A;) is hereditary. Since the canonical sheaf on F1(k) is 0(—2), the Serre duality theorem has the following form Hom(JT £) ^ D(Extl(g,F®0(-2))) where D = Hom(—,&). In fact, it is known that there exist Auslander-Reiten sequences in the category Coh(P1(A;)) (see [MZ] and [Jo] for more information). Prom the above Serre duality, we can see that the functor ®0(—2) plays a role in Coh(P1(A;)) similar to that played by the Auslander-Reiten translate r in the module category of a path algebra. So, the Auslander-Reiten quiver of Coh(P1(A;)) looks like: 2) 2.3 There exist the following fundamental exact sequences in Coh(P1(A;)). (1) 0 -+ 0(m) -A 0(p) 0 0(q) -^ 0(n) -+ 0 for m < min(p, q), max(p, q) < n and p + q = m + n. (2) 0 -+ 0(m) 1+ 0(n) A 0*=1 Orx\ -+ 0 for m < n in Z. We define the rank and the degree of indecomposable sheaf by rk(0(n)) = l,deg(0(n)) = n,rk(0£) = 0,deg(^) = ndeg(x), where deg(x) is the degree of the irreducible homogenous polynomial P G k[X, Y]. For.FeCohP1^), define dim(^r) = (rank?7 + deg.77, deg.77), hence for any indecomposable T we have (n+ l,n), n G dim(JT) = J (n,n), n G Z+. Thus the vectors dim(^r), .T7 indecomposable, identify with another half part of the root system of the affine Lie algebra 5/2 (type A['). The close relation between our first two examples is also not accidental, because of the equivalence of derived categories: Dh(TepK)^Db(CohP1(k)) a special case of the theorem of Geigle-Lenzing [GL].
236 JIE XIAO AND GUANGLIAN ZHANG 3. Convolution multiplication, a geometric setting of Ringel-Hall algebras 3.1 Let Q = (Qo, Qi, s, t) be a quiver, where Qo and Q\ are the sets of vertices and arrows, respectively, and s,t : Q\ —> Qo are maps such that any arrow h starts at s(h) and ends at t(h). We consider representations of Q over C. Let J be the index set of isomorphism classes of simple representations of Q, then we can identify J with Qq. Let M(Q) = {isomorphism classes of representations of Q}, T(Q) = {isomorphism classes of indecomposable representations of Q} Ma(Q) = {[Af]|[Af] G M(Q)AimM = a)} Xa(Q) = {[M]\[M] G J(Q),dimM = a}. For any dimension vector a = Y2ieia^ ^ ^M> we ^x a ^_graded space C^ = (Ca0*€/-Then Ea= 0 Homc(Ca^>,Caw) h:s(h)-+t(h) is an affine space. Set Ga=nie/GL(ai,C). For any (xh) G Ea and g = (#) G Ga, we define the action g- (xh) = (9t(h)Xh9j(h))> The orbit space is Ea/Ga. There is a natural bijection between Ma(Q) and Ea/Ga. So we may identify them, and regard Xa(Q) as a topological subspace of Ma(Q)- For any Q-representation M with dimM = a, let Om C Ea be the Ga-orbit of M. For an algebraic variety X over C, a subset A of X is said to be constructible if it is a finite union of locally closed subsets. A function / : X —► C is constructible if it is a finite C-linear combination of characteristic functions \q for constructible subsets O. The space of constructible functions over Ma(Q) is exactly the space of constructible Ga-invariant functions over Ea. Let us look at the example in which Q is the Kronecker quiver K. It is known that Ia(K) = I pl one point •, if a is a real root l(&), if a is an imaginary root. In fact, the topological space Ta(Q) is clearly known for any tame quiver Q (see [FMV]). 3.2 We define Ca(Q) to be the space of constructible Ga-invariant functions Ea —► C, and let C(Q) = 0a€N/ Ca(Q). Let mdEa(Q) to be the topological subspace of Ea consisting of all points x which correspond to indecomposable Q-representations, and let ind£a(<2) to be the space of constructible G^-invariant functions over indEa. We may regard as ind£a(<2) = {/ G Ca(Q)\suppf C indEa}, and ind£(<2) = (&aeR+mdCa(Q), where, by Kac theorem, R+ is the positive root system of the Kac-Moody Lie algebra corresponding to Q. Now we consider the following famous diagram due to Lusztig [LI]: E^ T? P1 "I?' p2± T?" P3, "I? a X &(3 <~ & ► ^ ► &a+(3, where E" = {{x,W)\x G Ea+/3, W is /-graded subspace of Ca+P such that x(W) C
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 237 W, dhnW = a } and E' = {(x,wy,r")\(x,W) G E" and isomorphisms r' : Ca -+ W and r" : C^ -+ Ca+P/W} and the morphisms p2 : E' —► E" by p2(x, W, r', r") = (x, W) and p3 : E" —► Ea+^ by p3(x, W) = x and p\ : E' —► Ea x E^ by pi (x, W, r', r") = (x', x") where (x', x") is uniquely determined by the equation xhfs{h) = rt{h)xh and xh,rs(h) = rt(h)Xh for all h G Q\. We can define the convolution multiplication as follows. fa*fp = (P3)\(f>2)\>(pi)*(g) for fa G Ca(Q) and fp G Cp{Q), where g(xux2) = fa(^i)f 13(^2) and the symbols ()i, ()b and ()* are famous Grothendieck operations (see [BBD]). We may compare this convolution operation with the multiplication of Ringel- Hall algebras for quivers over finite fields (see Section 5) as follows. If we denote 1qm the characteristic function of the orbit Om, then where T^N = {V subrepresentation of P\[V] = [N] and [P/V] = [M]}, and x(X) denotes Euler characteristic of the topological space X. This is a geometric way to define Ringel-Hall algebras. We have the following result. Theorem 3.1 (Ringel) Let Q be a quiver and k = C. (1) The space C(Q) is an associative N[I] -graded algebra under the convolution multiplication *. (2) The subspace mdC{Q) is R+-graded Lie algebra with the bracket operation [foe, ffi] = fa* f(3~ f(3* fa where R+ is the positive root system of the Kac-Moody Lie algebra corresponding toQ. One may prove further that ind£(<2) is the positive part of a generalized Kac- Moody Lie algebra and C(Q) contains the enveloping algebra of it (see [DXX] for a detail). 4. The Affine Kac-Moody Algebra of type A[ ^ 4.1 Let Q be a quiver. The Euler form (—, —) is defined on Ko(repQ) = Z[I] by (aiP) = 5Z aibi ~ ^2 a*(h)bt(h) ieQo heQi for a = (di)i€Q0, (3 = (bi)i£Q0 in Z[J]. This is a bilinear form on Z[J]. The formula (a, p) = dim fcHom(M, N) - dim ^Ext^M, N) for a = dimM, /3 = dim AT and M, N G repQ provides a homological interpretation. Now we consider an affine quiver Q. We denote by e(a,/3) = (—l)<a'^; it is called the Euler cocycle. By using a 2-cocycle as a building block of the structure constants, one can have a combinatorial way to get the integral form of the affine
238 JIE XIAO AND GUANGLIAN ZHANG Lie algebra (see [FK]). Let S be the minimal positive imaginary root. We define a .R+-graded space as follows. \rt(n\ — / ^e^' a a rea,l ro°t C[I]/C5, a = nS an imaginary root. For h G CI, we denote by h(n) G N^siQ) ^ne image of h under the natural projection CI -+ CI/CS. The Lie bracket on Me{Q) is given by {e(a, f3)ea+p, if a + /3 G R+ a real root e(a, /3)a(l), if a + (3 = 15 an imaginary root 0, if a+ /?£#+; [ft(n),ea] = -[ea,ft(n)] = e(rcJ,a)(/i,a)ea+n$; [/i(n),/i(ra)]=0 where (/i, a) = (h, a) + (a, h) is the symmetric Euler form. Then by Frenkel-Kac [FK], we have Proposition 4.1 The R+-graded space Me{Q) = @aen+ Na(Q) under the Lie bracket defined above is the positive part A/*+ of the corresponding affine Kac-Moody Lie algebra. For the Kronecker quiver K, the corresponding Lie algebra is sfe (type A[*). We can write down precisely the basis of Afe(K) and the Lie bracket as follows. The following ql\ is the simple root corresponding to the dimension vector (1,0). Example 4.2 The case A\l\ We have [en,n+i,em?m+i] = 0, [en+ijn,em+ijm] = 0, [ai(n),ai(ra)] = 0; [ai(n),em+i>m] = 2(—l)n em+n+i>m+n, [ai(n),em>m+i] = 2(—l)nem+n?m_l_n_l_i; [en,n+i,em+i,m] = (-l)m+nai(m + n + l). On the other hand, we may consider the category repQ and the Lie algebra ind£(<2) with the Lie bracket induced by the convolution multiplication *. For the Kronecker quiver K, it is also not difficult to write down the precise formulae. Example 4.3 Consider indC(K) with K being Kronecker quiver. For a G R+ being a real root, let Ea be the characteristic function of indEa which corresponds to the indecomposable representation with dimension vector a, and E^n^ be the constant function equal to 1 on indE(n?n), here indE(n?n)/G(n?n) = P1(C). We then have #(0,1) * ^Wi-ljIindE^j - E(n,n), ^(0,1) * E{n,n) lindE(nin+1) ~ 2£(n,n+l);
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 239 £(„,„-!) * £(0,l)lindE(n,n) = °> E{n,n) * ^(0,l)lindE(n,n+1) = °5 #(i,o) * ^(n-i.njIindE^) = °' EW * JW)lindE(n+1,n) = °5 #(n-l,n) * #(1,0) lindE(ri)ri) = ^W), #(n,n) * £?(l,0)lindE(n+1,n) = 2£7(n+l,n). The elements 25(0jl), £(1,0), -E(n+i,n), #(n-i,n), J^n) sPan a Lie subalgebra A/**(X) of ind£(if). It is very interesting to compare these two families of formulae. Obviously the Lie bracket induced from the formulae in Example 4.3 is almost the same as that in Example 4.2. Precisely we have Proposition 4.4 Let K be the Kronecker quiver. The linear map <p : Afe(K) —► J\f*(K) given by (f(ea) = Ea for a G R+ being a real root and ip(ai(n)) = (—l)^n+1^£'(n?n) for ot\(n) an imaginary root induces an isomorphism of Lie algebras. In fact, this is only one example of the result in Prenkel-Malkin-Vybornov [FMV], which has established a canonical isomorphism Af*(Q) = Ne(Q) when Q is a quiver of affine type A, D, E. This work generalizes RingePs result in the finite type to the affine type case. However, the method used here only gave a realization of the nilpotent part of a Kac-Moody Lie algebra. One may wonder whether it is possible to use representations of quivers to recover the whole Lie algebra. In [PX], it has been proved that the derived categories of hereditary algebras can provide successful models to realize all symmetrizable Kac-Moody Lie algebras. Actually, in [LP], the precise formulae for an integral basis of the whole part of an affine algebra were given. Here we consider representations of Q over the complex numbers C. We may ask the following question. Question 2: What will happen for the structure of C(Q) if we replace k = C by a finite field ¥q? 5. Ringel-Hall algebras 5.1 Let Q be a quiver, A = ¥qQ the path algebra of Q over ¥q : the finite field with q elements. Set V = {isoclasses of representations of Q}. For any aePwe choose Va to be a representative in the class a. Given three classes A, a, (3 G V, let g*p = %{w < V\\W ^ V/3, Vx/W ^ Va}. By taking v = y/q and the integral domain Q(v), we have the following Definition 5.1 The Ringel-Hall algebra H(A) of A is a free Q(v)-module with the basis {ux\X G V} whose multiplication is given by Ua'Uf3=^2 9a(3uX for al1 OL,f3eV. xev The twisted Ringel-Hall algebra H*{A) is a free Q(v)-module with basis {ux\X G V}, and multiplication is given by ua*up = v<a>0) J2 9a?u\ f°r al1 ^PeV. xev
240 JIE XIAO AND GUANGLIAN ZHANG 5.2 We may consider the subalgebra of H*(A) generated by Ui = uai, for i G /(= Qo) and where oti G V is the isoclass of simple ^4-module at vertex i. The subalgebra is called the composition algebra and it is denoted by C*(A). On the other hand, for A = ¥qQ, or more generally, for A any finite dimensional hereditary algebra, (J, (—, —)) is a Cartan datum, where J is the index set of simple ^4-modules and (—, —) is the symmetric Euler form of A. We note here that the Cartan datum of any type can be realized from a finite dimensional hereditary algebra in this way (see [G]). For a Cartan datum (J, (—, —)), the quantized enveloping algebra Uq defined by Drinfeld[Drl] and Jimbo[J] is associated with it. The positive part U+ is generated by Ei, i E I with subject to the quantum Serre relations: l-2(ei,e:,)/(ei,ei) Ej^(»j£ ■^(l-2(ei,ej)/(ei,ei)-s) _ q % 3 % 3=0 for t^-el, where e\s) = £?/[*],!, [*]< = (vf - «r')/(t>i - vf1), [a]i! = IE=iM« and Vi = v(eiiei)/2. The following fundamental result is due to Ringel and Green (see [G] and [Ob]). Theorem 5.2 (Ringel-Green) There exists a canonical isomorphism ip : C*(A) —► U+ by sending Ui to Ei for i G /, if the Cartan datum ofliq is given by the index set I of simple A-modules and the symmetric Euler form of A, where A is a finite dimensional hereditary algebra. There is a usual way to define the generic form C*(Q) of the composition algebra C*(A) by considering the representations of Q over infinitely many finite fields. Then C*(Q) is a Q(v)-algebra where v becomes a transcendental element over Q. Note that q for C*(A) is merely a power of a prime number. Put t-4 = -^Vj for i e I and n G N and let C*{Q)z be the integral form of C*(Q), which is generated by uf , i G /, n G N over the integral domain Z = Z[v,v-1]. Also the quantum group U+ has the integral form Z/J, which is generated by E\n\ i G /, n G N over Z. Then the above theorem can be strengthened into the following. Theorem 5.3 (Ringel-Green) The map i\) : C*(Q)z —► 11% by sending uj to E\ for i € I and n E N is a Z-algebra isomorphism, if the two algebras share a common Cartan datum. There are bar involutions ( ) : U+ —► U+ and U% —► U% defined by v = v_1, Ei = Ei and E^ = E^ . Of course it can be carried over C* (Q) via the canonical isomorphism. In [LI] and [L2] Lusztig defined the canonical basis of ZY+ to be the set B = Uagnj ba which satisfies the conditions: (1) BAH(-BA) = 0, (2) BAn(-a(BA)) = 0, (3) cr(BA) = BA and (4) BA is a basis of U+ over Z also a basis of U+ over Q(v), where A G NJ and a : U+ —► U+ is the anti-automorphism given by a(Ei) = Ei, i G J. In fact Lusztig proved that the basis B can be characterized by the following property: B = {x eU% \x = x and (x,x) = 1 + v_1Z[v_1]}
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 241 where (—, —) is the usual bilinear inner product on U (see [L5]). 6. Canonical bases for U+(sl2). 6.1 We now go back to Kronecker quiver K. By Theorem 5.3, we have the canonical isomorphism i/>:C*(K)z*u£(?h). Since the positive root system R+ of sl2 is : i?+ = {(Z + l,Z),(m,ra),(n,n+l)|Z > 0,ra > l,n > 0}. By the structure of the Auslander-Reiten quiver of K, we can arrange the root system R+ in the following order (1,0) -< < (ra + 1, ra) -< (ra + 2, ra + 1) -< < (A;, k) -< (A; + 1, k + 1) -< < (n + 1,n + 2) -< (n, n + 1) -< < (0,1). For a preprojective or a preinjective if-module P, let (P) = ^-dimp+aimli/nap^^ We define the root vectors in C*(K)z as follow: -E(m+l,m) = V~ mW(m+i,m) = (U(m+l,m))> -E(n,n+1) = v~ n,w(n,n+l) = (^(n,n+l))- c*ev regular dimvQ=m<5 In particular, we let E\ = 2£(i,o) and #2 = ^(0,1) • For an n-partition w = (w\,W2, • • • ,wm) G P(n), let JE^J = EWlS * ^^2(5 * • • • * EwmS- We have the following result due to P.Zhang (see [Z]) and an improvement due to X.Chen (see [C]). Proposition 6.1 The set {(P) * Ews * (T) ||P G V preprojective, w G P(n), T G V preinjective, n G N} 25 an integral basis ofCg. Remark 6.2 (1) It is not difficult to see that the root vectors provided here exactly correspond to the root vectors of Uq(sl2) provided by Damiani in [Da] and by Beck in [Be]. A function c : R+ —► N is called support-finite if c(a) ^ 0 only for finitely many a G /2+. The set of all support-finite functions c : R+ —► N is denoted by N^ . For ceN*+,if {a G R+\c(a) ^ 0} = {ft -< ft < • • • < ft}, we set EC = E(*c(Pl)) ^ E(*C(P2)) ^ . . . ^ E(*c(Pk)) where E{£m) is the divided power if ft is a real root and £?J*c(/3fc)) = £^(/?fe) if ft = ra£. Then Proposition 6.1 says that the set {Ec\c G NR } is a Z-basis of C*.
242 JIE XIAO AND GUANGLIAN ZHANG For d = (di, d2) G N2, we set E(d) = E^ * E[*dl). So, if {a G R+\c(a) ^ 0} = {ft -< (32 -< • • • < ft}, we set £(c) = £(c(ft)ft) * £(c(ft)ft) * • •' * E(c(0k)f3k). It is a monomial on the Chevalley generators 2?i and #2 in the form of divided powers. Therefore E(c) EC*. For any cGN$ we assume that Ec = (P)*Eu)s * (T), where P is a preprojective module and T is a preinjective module. Let V^s be a module such that dimCV^ = max{dim Ov\Ew5 = EWl5 * EW2s * ••• * EWrn5 = ^2av(V),av 7^ 0}- v We choose VC = P® V^s © /, Oc = Op©Vw6©j. Let <p : NR+ -+ N2 be defined by <p(c) = T,aeR+ c(a)a- Then for any d G n2> </?-1(d) is a finite set. We define a (geometric) order in </?-1(d) as follows: d •< c if and only if c' = c or c' ^ c but dim(9c/ < dimOc. We have the following results (see [LXZ]). Proposition 6.3 For any c G N^ , let d = <p(c). Then which satisfies that (1) hcc, G2, (2) hcc = 1, f5j i//i£/ 7^ 0 and c' ^ c, £/ien c' -< c, (4) E(c) is bar-invariant We can set W= E wc'#c' for any cGN*+, where a;£, G2, since the set {£c|ceNfi+} is a Z-basis of C*. Then we have Proposition 6.4 For any c G N^ , wcc = 1 and, if wcd ^ 0 and c' ^ c £/ien c' -< c. Consider the bar involution ( ) : C* —► C*, we have i£c = Ec for any cGN$ . Hence c' d ,c" Since {£?c|c G N*+} is a Z-basis of C*, if c" = c, we have £c, a£^' = 1; if c" ^ c, we have $^c, o;£,a;£,/ = 0; that is the orthogonal relation c' Therefore, according to a linear algebra method introduced by Lusztig[L6], the system of equations c'-sc"-sc
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 243 with the unknowns Q, G Z[v 1], c' ■< c and c' , c G tp 1(d) has unique solution such that Q = l and Qev-^iv-1] for all c'^ c. For any c G </?-1(d) and d G N2, we set ec = J2 &EC' and J = {£c\cep-1(d),deN2}. Then Proposition 6.5 The set J provides the canonical basis of C*z. Remark 6.6 Our method used in this section is very close to the method of Lusztig in [LI] for the canonical bases of finite type. 7. Nilpotent representations of cyclic quiver and canonical bases of type Sin 7.1 All results presented here are taken from [DDX]. Let A = A(n) be the cyclic quiver with vertex set Aq = Z/nZ = {1,2.,-•• ,n} and arrow set Ai = {i —► i + lli GZ/nZ}. We consider the category T = T(n) of finite dimensional nilpotent representations of A(n) over Fq. Because of the shape of its Auslander-Reiten quiver, T(n) is called a tube of rank n. Let Si, i G A0, be the irreducible objects in T(n) and Si[l] the (unique up to isomorphism) indecomposable object in T(n) with top Si and length /. Again in this section, we let V be the set of isomorphism classes of objects in T(n) and H the Ringel-Hall algebra of T(n), H* the twisted Ringel-Hall algebra of T(n). Since the Hall polynomials always exist in this case, we may regard the algebras H and W* in their generic form. So they all are defined generically over Q(v), where v is transcendental. Note that, throughout this section, all properties we obtain are generic and independent of the base field ¥q. 7.2 Let II be the set of n-tuples of partitions. Each element defines an object 7r=(7r(l),7r(2),...,7r(n))Gn M(n) = 0 Si\nf] i£A0 3>l
244 JIE XIAO AND GUANGLIAN ZHANG in •) is the partition dual to 7rW. In this way we obtain a bijection between II and the set V. So we simply denote by un, n G II, the element u[M(n)] in W*. An n-tuple n = {^l\^2\ • • -n^) of partitions in II is called aperiodic (in the sense of Lusztig [L3]), or separated (in the sense of Ringel [R2]), if for each / > 1 there is some i = i(l) G Aq such that itj 7^ / for all j > 1. By IP we denote the set of aperiodic n-tuples of partitions. An object M in T is called aperiodic if M ~ M(n) for some n G IIa. For any dimension vector a G Nn(= NJ), we let Ua = {A G n|dimM(A) = a} and n^nanna. Given any two modules M, N in T, there exists a unique (up to isomorphism) extension L of M by N with minimal dimEnd(L). This extension L is called the generic extension of M by N and we denote it by L = M o N. If we define the operation in V by [M] o [N] = [M o AT], then (V,o) is a monoid with identity [0]. Let Q be the set of all words on the alphabet Aq. For each w = i\%2 • • • im € ^> we set M(w) = Sh oSi2o---oSim. Then there is a unique n G II such that M(w) ~ M(w), we define p(w) = n. It has been proved in [DDX] that n = p(w) G IP and p induces a surjection p : Q -» IIa. We have a partial order in V, or equivalently in II, as follows: for /i, A G II, /i ^ A if and only if Om(h) £ Om(A)» or equivalently, dimHom(M, M(A)) < dimHom(M, M(/x)) for all modules M in T. For each module M in T and integer 5 > 1, we write sM as the direct sum of s copies of M. For w = Jl1 ffi '" JT ^ ^ with jr_i ^ jr for all r, this is the tight form of w, and A G II, we take /ir G II such that M{jir) = erSjr. We have the Hall polynomial gM\ )...m(u ) anc* smiPly denote it by g*. A word w is called distinguished if the Hall polynomial gS, = 1. This means that M(p(w)) has a unique reduced filtration of type w, that is, M(p(w)) has a unique filtration of the form: M(p(w)) = M0 D Mi D • • • D Mt-i D Mt = 0, which satisfies Mr-\/Mr ~ erSjr for all r. According to [DDX], we have Proposition 7.1 For any n G na, there exists a distinguished word ^=jrj2e2---itetennp-1(7r) in tight form, that is, M(ir) has a unique reduced filtration of type w^. In H\ let *4*m) = ^i*m) = Q,» G A0,ra > 1. The Z-subalgebra C*z of W* generated by u\*m\ i G Ao, m > 1 is the composition algebra of T. 7.3 For each w = j^j^2 • • • Jt* G ^ witn ir-i ¥" jr for all r, let m<™>=i^ci)*...*£(.*ct) in C*. For each a G Nn and tt G I1J, we now fix a distinguished word wn G Q fl p-1(7r). In this way, we obtain a set V = {wn\w G na}, which is called in [DDX] a section of distinguished words.
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 245 Let L0 — ej1Sj1,L\ — £j1Sj1 oej2Sj2,L2 — L\ oej35j3, • • • ,Lt-\ — Lt-2<>ejtSjt. Since Li is the generic extension of Li-\ by £ji+1Sji+1 and thus dimEnd(Li) is minimal, we have M(w) ~ Lt-\ . Since 1 = 9ln = 9e]YS6x,eHSi29h\,eizS6z ' " 9Lt-2,eJtSJti we obtain that ofi 0 Q = 1,1 < i < t — 2. Recall that (M) = ^-dimM+dimEnd(M)^[M], we have (Li-1)*(eji+1Sji+1) = (Li)+ J2 a^X) X,dimOx <dimOhi with ax G Z. So m<^> = <M(7r)) + X;&<M(A)>, where f *w G Z and of course if f *w ^ 0 then dimM(A) = dimM(7r) = a. Then we can define the vector En inductively by the relation E7r=m(w^- \^ v-dimM(7r)+dimEndM(7r)+dimM(A)-dimEndM(A) A (v2\£ A^7r,A€ll£ Therefore we have the relation E, = (M(tt)) + Y, Vx(M(X)) A€nQ\n^,A^7r with 77 J G Z. Proposition 7.2 Let V = {w^ln G IP} be a section of distinguished words of Q over IP. Then each of the following sets forms a Z-basis ofCz. (1) {m^)|7TGlP} (2) {Ev\ic G IP} Moreover we have relations m(^) = E„+ sr^ v-dimM(7r)+dimEndM(Tr)+dimM(\)-dimEndM(\) A tv2\jg for each n G II£, that is, the transition matrix between the two bases is triangular with diagonal entries equal to 1. 7.4 The definition of the basis {En\7r G IP} is given according to the choice of the section V of distinguished words, but eventually it has been proved in [DDX] that it is independent of the choice of the sections of distinguished words. We may regard {m^^Tr G IP} as a monomial Z-basis of C£ and {En\w G IP} as a "PBW"-basis of Cz. Since the triangular relation between the two bases, we can follow Lusztig [LI] by using a standard linear algebra method, as we did in Section 4, to obtain the canonical basis {S^\k G IP} of C*z £n= J2 P*«E*> f°r n e n-' <* G N"' A^7r,A€lI^ with p^ = 1 and p\^ G v~1Z[v~l] for A -< n.
246 JIE XIAO AND GUANGLIAN ZHANG Theorem 7.3 [DDX] The set {E^ G IP} is the canonical basis of C*z. 8. Afflne canonical bases 8.1 By [R6], the Auslander-Reiten quiver of an affine quiver Q looks as follows. * * • * * * it * * * Figure 2: Now mod A for A = ¥q(Q) has a preprojective component, a preinjective component and a stable separating tubular P1(Fg)-family, called regular components. 8.1.1 We first consider the integral bases arising from preprojective and preinjective components. The situation we meet in this subsection is essentially the same as in the case of finite type. Let Q = (QoiQi) be a tame quiver and (ZJ, (—, —)) the corresponding Cartan datum. Let li = liq be the quantized affine enveloping algebra associated to it, with the Chevalley generators: E^Fi and Ki . Lusztig in [L5] has introduced the symmetries T"^ : U —► U for i G /, given by T'i[\{Ei) = —FiKi, T"tl(Fi) = —KiEi Tl'^Ej) r+s= — a,i. r+s= — di. n[1{K0)=KSi (#> where a^ = (i,j) for i,j G /, and (3 G Z/ and Sj(/3) = 0 — (P,i)i. For each i € I, one may define U+[i\ = {x eU+\Tl[x{x) e«+}. Then T-'tl : U+[i\ —> U+[i] is an automorphism. Moreover, let U^[i] = U% nl4+[i], then T(y.U+\i]-+U+[i]. Let GiQ be the quiver obtained from Q by reversing the direction of all arrows connected to the vertex i. For i a sink of Q, we have the BGP reflection functor (see[BGP] or[DR]): r+ • mod A modai^L where A = ¥q(Q) and a\A = ¥q(aiQ) the path algebras of Q and oiQ respectively. Therefore we have the homomorphism: satisfying that ai:W*(i4)[t]—*W*(aii4)[t] (Ti(ua) = ucr+(a) for any V<* £ modi4[i],
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 247 where mod^4[z] is the subcategory of all representations which do not have Si as a direct summand and W*(A)[i] is the subalgebra of H*(A) generated by ua with Va G modi4[i]. We remark that the action of Oi can be restricted on C*(A). It induces the action on C*{Q)z- Notice here that C*(Q)z is canonically isomorphic to C*(aiQ)z by fixing the Chevalley generators which correspond to the simple representations of Q and o~iQ respectively. We can identify the two algebras and define C*(Q)z[i] = {x e C*(Q)z\°i{x) G C*(Q)2}. We can regard that the functor af induces the isomorphism: <Ji:C*(Q)z[i\—+ C*(Q)z\i], It is known that Oi = T[\ under the identification C*(Q) = ZY+ (for example, see[XY]) . Dually, we have the similar results for a source i of Q. 8.1.2 We denote by Prep and Prei the isomorphism classes of indecomposable preprojective and preinjective ^4-modules respectively. The set {(v>sm)\M is indecomposable in Prep or Prei and s > 1} lies in C*z. Let imi'" ih be an admissible sink sequence of Q, that is, vyh is a sink of Q and for any 1 < t < m, the vertex it is a sink for the quiver (Tit+1 • • crirnQ. Let M e Prei. There exists an admissible sink sequence of Q such that ^ = <"-<,($m+1), where Sim+1 is a simple representation in moda^ • • • a^A. Lemma 8.1 Let M be an indecomposable preinjective representation. Then (uM)=n'ul---n'm^Eim+x), where M = a^- • • a* (Sim+1), for an admissible sink sequence im, • • • , i\ of Q. This means that in H*(A) the vectors provided by indecomposable preinjective modules can be obtained by applying the Lusztig symmetries on the Chevalley generators in an admissible order. However the vectors provided by indecomposable regular modules can not be obtained in this way. There is a dual statement for indecomposable preprojective M. 8.1.3 Since Prei is representation-directed, we may give a total ordering of Prei as follows. Let {•••,/?3,/?2,/?l} be all positive real roots appearing in Prei, and be all indecomposables in Prei with dimM(/3i) = fa. We require that a total ordering ■< in Prei satisfies the following Hom(M(ft), M(Pj)) ± 0 implies ft -< fy and i > j. Then such an ordering has the property (ft, (3j) > 0 implies ft -< (3j and i > j
248 JIE XIAO AND GUANGLIAN ZHANG and (fa, (3j) < 0 implies (3j -< fa and i < j and Ext^MC^MC^)) = 0 for i > j. Therefore $ •< (3j if and only if i > j. There is no harm to denote by Prei = {•••,&, ft, A}. 8.1.4 Similarly, we can give a total ordering of Prep as follows. Let {ai,a2,a3,---} be all positive real roots appearing in Prep, and {M(a1),M(a2),M(a3)r-} be all indecomposables in Prep with dimM(a».) = a*. We require that a total ordering -< in Prep satisfies the following Hom(M(ai),M(aj)) ^ 0 implies on -< otj and i < j. Then such an ordering satisfies that (ai,aj) > 0 implies a* ■< ctj and i < j and (a*, otj) < 0 implies ay -< a* and j < i and Ext 1(M(ai), M^)) = 0 for i < j. There is no harm to denote Prep by Prep = {a1? a2,#3, • • • }. Let NPre2 be the set of all support-finite functions b : Prei —► N. Then M(b)= 0 b(A)M(A) ft € Prei is a preinjective representation and any preinjective representation is isomorphic to one of this form. We denote by <M(b)> = (uM{b)) = „-dimM(b)+dimEnd(M(b))UM(b)_ We have Lemma 8.2 For any b G NPre% (M(b)> = (b(Am)M(Am)> * • • • * (M&JM(A,)), w/iere {/3im -< /^ -< < ftj = {/? G Prei | b(/3) ^ 0}. So (Af (b)) G C*z for all b G NPrei. Therefore we are ready to define C*(Prei) to be the Z-submodule of Cg spanned by {(M(b))|bGNprei}. We have Lemma 8.3 The Z-submodule C*(Prei) is a subalgebra of Cg and {(M(b))|b G NPrei} is a Z-basis of C*(Prei). A similar result holds for Prep.
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 249 Lemma 8.4 (1) For any a G NPrep,M(a) = ®aiePrepa(ai)M(ai), then (M(a)) = (aK)MK)) * • • • * <a(aim)M(airJ), where {a^ -< on2 -< • • • -< a^m} = {aG Prep | a(a) 7^ 0}. (2) Let C* (Prep) be the Z-submodule of C*z spanned by {(M(a))|aeNPrep}. Then C*(Prep) is a subalgebra of C*z and {(M(a))|a G NPrep} is a Z-basis of C*(Prep). 8.3 Let {Si,S2,'' * ,Sn} be a complete set of non-isomorphic simple modules of mod A with an admissible ordering: Ext1(Si,Sj) = 0iovi >j. Any module M with dimension vector d = (di, cfo, • • • , dn) has a unique filtration M = M0 2 Mi D • • • D Mn = 0 with the factors Mi-i/Mi isomorphic to d^, since Ext 1(Si, Sj) = 0 for i > j. This shows that the Hall polynomial g%[si—dnsn = 1- Then in H* USi * US2 * *uSn ~~ V / ,"M(a)$M(t)$M(b); where M(a) is preprojective, M(t) regular and M(b) preinjective such that dimM(a)+ dimAf(t) + dimM(b) = (di, • • • , dn) = d. 8.4 We now come to the construction of the integral bases for the generic composition algebras. 8.4.1. We first consider the embedding of the representation category of the Kro- necker quiver into the representation category of a general tame quiver. Let Q be a tame quiver, e be an extended vertex of Q and A = ¥qQ the path algebra of Q over ¥q. Let P = P(e) be the corresponding indecomposable projective module, that is, top(P) = S is the simple module corresponding to the vertex e, and p = dimP(e). It is known that (p,p) = 1 = (p,6) and there exists a unique indecomposable preprojective module L with dimL = p + S. Moreover, we have Honu (L, P) = 0 and Ext ^(L,P) = 0. This means that (P,L) is an exceptional pair. Let <£(P, L) be the smallest full subcategory of mod A which contains P and L and is closed under extensions, kernels of epimorphisms and cokernels of monomor- phisms. Also we have dimFqHom^(P, L) = 2, therefore <£(P, L) is equivalent to the module category of the Kronecker quiver over Fq. Thus it induces an exact embedding F : mod K <-^> mod A, where K is the path algebra of the Kronecker quiver over ¥q. We note here that the embedding functor F is essentially independent of the field ¥q. This gives rise to an injection of algebras, still denoted by F : H*(K) <-> H*(A). In H*(K) we have defined the element Em6 for m > 1. We may still denote by Ems for its image F(Ems). Since Ems is in C*(K), so Ems is in C*(A), in fact in C*(A)z. Let /C be the subalgebra of C*(A) generated by Ems for m € N, it is a polynomial ring of infinitely many variables {Ems\m > 1}, and its integral form is the polynomial ring of variables {Ems\m > 1} over Z. 8.4.2 We may list all non-homogeneous tubes 7^,7^,-•• ,7^ in mod A (in fact, s < 3). For each 7^, let r* = r(%) be the period of 7^, i.e., the number of quasi- simple modules in 7i, then r^ > 1. We have the generic composition algebra C*(%) of % and its integral form C*(Ti)z- For each % we have the set II^1 of aperiodic
250 JIE XIAO AND GUANGLIAN ZHANG rvtuples of partitions such that for any n G Uf, M{tt) is an aperiodic module in %. We have constructed in Section 7.3 the element E* = (M(tt)) + Yl nl{M{\)) A€ni\n^,A^7r such that {En\n G 11°} is a Z-basis of C*(7^)z. To produce a PBW basis of the generic composition algebra, a natural idea is to put them together with the vectors we obtained from indecomposable preprojective and preinjective modules. However the following known fact ( for example see [CB]) tells us that there still is one dimension missing. The missing part will be filled up through the embedding of the module category of Kronecker quiver into that of the tame quiver. Lemma 8.5 (a) The equation Xli=i(ri — 1) = 1^1 — 2 holds, (b) The dimension of the root space of m5 in the corresponding Kac-Moody algebra equals \I\ — 1, where \I\ is the number of vertices of Q. Now we define a set M by the following rule. Any c G M is given by the data: (1) a support-finite function ac : Prep —► N, (2) a support-finite function bc : Prei —► N, (3) an element 7TiC G Ilf for each 7^, 1 < i < s, (4) a partition wc = (u>i,u>2> • • • ,wt) for some t > 1, where wi < W2 < • — < wt areinN\{0}. Then for each c E M. we may define Ec = (M(ac)) * EVlo * EV2o * • • • * EVao * EWcS * (M(bc)), where (M(ac)) and (M(bc)) are defined in Section 8.1, Enic for 1 < i < s are defined in Section 7.3 and EWcs is defined in Section 6.1. We see that {Ec\c G M} lies in C*(Q), in fact in C*(Q)z, and are linearly independent over Q(v), since the corresponding elements in Ringel-Hall algebras are linearly independent. Proposition 8.6 The set {Ec\c G M} is a Q(v)-basis ofC*(Q). Its proof is a hard part in the work of [LXZ]. Based on this result, the following can be obtained by applying the representations of quivers. Theorem 8.7 The set {Ec\c G M} is a A-basis ofC*(Q)A, where A = Q[v,v-1]. As a consequence, the canonical mapping <p : C*(Prep) ®A C*(7i) ®A--®A C*{TS) ®A K ®A C*{Prei) -. C*{Q)A is an isomorphism of free A-modules. This answers the main question in [Z]. We remark here that a similar result are obtained in [H], too. 8.5 Our varieties still define over C. Let AcEQ and B C E^ be subvarieties, we define the extension set A • B of A by B as follows. A • B = {z G E^+^l there exists an exact sequence 0 -+ M(x) -+ M(z) -+ M(y) -+ 0 with x G B, y G A}. It is known that if both A and B are irreducible algebraic varieties and are stable under the action of Ga and G^ respectively, then A • B is irreducible and stable under the action of Ga+^.
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 251 For any c G M we define the variety Oc = 0M(8lc) * °Mnic * ®Mn2c • • • • • 0Mnsc *Mm * ^M(bc) where Mm = Mm *' " *Mm if ^c = (^i, u>2, • • • >w*) and Mm are *ne uni°n of orbits of regular modules of £(P, L) with dimension vector WiS. Proposition 8.8 For any c G .M, £/iere exists a monomialmc on the divided powers of ust, i G /, s^c/i £/m£ mc = Ec + J2 hc'EC'> c'eM, dimOc,<dimOc where h%, G Q[v,v-1]. It is also a hard part of [LXZ] to find a monomial basis {mc||c G M} on the divided powers of the Chevalley generators such that the transition matrix between {mc||c e M} and {Ec\c G M} is triangular with diagonal entries equal to 1. In fact, a precise way was given in [LXZ] to construct the basis {mc||c G M} with respect to the geometric order induced by dim(9c. Once we get it, we may use the standard linear algebra method by Lusztig to obtain the relation: w = J2 u^E° forany ceM c'eM with u°r G A such that u^ = 1 and if luc, ^ 0 and c ^ c then dimCV < dim0c. Solving the system of equations c< = E wc"c^ dim oct <dim oc„ <dim oc to get a unique solution such that Cc = 1 and <£ ^ ^_1Qk_1] if dim0C' < dimOc. Let Ec= J2 &EC' for any c G M. &eM Note that this is a finite sum. Then the main result in [LXZ] is the following. Theorem 8.9 The set {£c\c G M.} provides the canonical bases o/C*(Q)^, which is characterized by the two properties: (a) £c = £c for all c G M. (b) 7r(£c) = 7r(Ec), where n : C*(Q)a —> C* {Q) a/ v-1^ {Q) A is the canonical projection. Finally we still have the following question. Question 3: Does the basis {£c|c e M} exactly equal to the canonical base B defined by Lusztig in [L5]? Acknowledgement: We are very grateful to the referee for his (her) helpful suggestions for the improvement of the manuscript, both on mathematics and on English writing.
252 JIE XIAO AND GUANGLIAN ZHANG References [BBD] A.A.Beilinson, J.Bernstein and P.Deligne, Faisceaux pervers, Asterisque 100 (1982), Soc. Math. Prance. [BK] P.Baumann and C.Kassel, The Hall algebra of the category of coherent sheaves on the projective line, J.reine angew.Math. Vol. 533 (2001), 207-233. [Be] J.Beck, Braid group action and quantum affine algebras, Comm.Math.Phys. 165 (1994), 555-568. [BCP] J.Beck, V.Chari and A.Pressley, An algebraic characterization of the affine canonical basis, Duke Math. J.99 (1999),455-487. [BN] J.Beck and H.Nakajima, Crystal bases and two sided cells of quantum affine algebras, Duke Math. J. 123 (2004), 335-402. [BGP] I.N.Bernstein, I.M.Gelfand and V.A.Ponomarev, Coxeter functors and Gabriel theorem, Uspechi Mat. Nauk 28 (1973); Russian Math. Surv. 28 (1973), 17-32. [Bei] A. Beilinson, Coherent sheaves on P™ and problems of linear algebras, Punct. Anal. Appl. 12 (1978), No.3, 68-69. [B] R.E.Borcherds, Generalized Kac-Moody algebras, J.Algebra, 115 (1988), 501-512. [C] X.Chen, Root vectors of the composition algebra of Kronecker algebras, Preprint 2003. [CX] X.Chen and J.Xiao, Exceptional sequences in Hall algebras and quantum groups, Compositio Mathematica 117(2) (1999),165-191. [CB] W.Crawley-Boevey, Lectures on Representations of Quivers, Mathematical Institute, Oxford University, 1992. [Da] I.Damiani, A basis of type Poincare-Birkhoff-Witt for the quantum algebra of sfo, J.Algebra,161(1993),291-310. [DDX] B.Deng, J.Du and J.Xiao, Generic extensions and canonical bases for cyclic quivers, preprint (2004). [DX] B.Deng and J.Xiao, On Ringel-Hall algebras, in Representations of finite dimensional algebras and related topics in Lie theory and geometry, 319-348 Fields Institute Communications 40, Amer. Math. Soc, Providence, RI,(2004). [DR] V. Dlab and C. M. Ringel, Indecomposable representations of graphs and algebras, Memoirs Amer. Math. Soc. 173, 1976. [Drl] V.G.Drinfeld, Hopf algebras and the quantum Yang-Baxter equation, Dokl. Akad. Nauk SSSR 283 (1985), 1060-1064.(=Soviet Math. Dokl.32 (1985),254-258). [Dr2] V.G.Drinfeld, A new realization of Yangians and of quantum affine algebras, Soviet Math.Dokl.36 (1988), 212-216. [DXX] M.Ding, J.Xiao and F.Xu, Realizing enveloping algebras via varieties of modules, preprint (2006). [FK] I.Frenkel and V.Kac, Basic representations of affine algebras and dual resonance models, Invent, math. 62(1980), 23-66. [FMV] I. Frenkel, A. Malkin and M. Vybornov, Affine Lie algebras and tame quivers, Selecta Math. (N.S) 7 (2001),1-56. [Ga] P. Gabriel, Unzerlegbare Darstellungen /, Manuscripta Math. 6 (1972), 71-103. [GL] W.Geigle and H.Lenzing, A class of weighted projective curves arising in the representation theory of finite-dimensional algebras, Singularities, representation of algebras and vector bundles, Springer LNM 1273, 265-297, (1987). [G] J. A. Green, Hall algebras, hereditary algebras and quantum groups, Invent. Math. 120(1995), 361-377. [Gr] A.Grothendieck, Sur la classification des fibres holomorphes sur la sphere de Riemann, Amer. J. Math. 79 (1957), 127-138. [H] A.Hubery, The composition algebra of an affine quiver, Preprint, arXiv:math.RT/0403206. [J] M. Jimbo, A q-difference analogue ofU(g) and the Yang-Baxter equation, Lett. Math. Phys. 10 (1985),63-69. [JKK] K.Jeong, S.-J.Kang and M.Kashiwara, Crystal basis for quantum generalized Kac-Moody algebras, arxiv:math.QA/0305390, May 2003. [Jo] P. Jorgensen, Auslander-Reiten theory over topological spaces, Comment. Math. Helv. 79 (2004), no. 1, 160-182. [Ka] S.-J.Kang, Quantum deformations of generalized Kac-Moody algebras and their modules, J. Algebra, 175 (1995), 1041-1066.
A TRIP FROM REPRESENTATIONS OF THE KRONECKER QUIVER 253 [K] M.Kashiwara, On crystal bases of the q-analogue of universal enveloping algebras, Duke Math.J. 63(1991),465-516. [Kal] V.Kac, Infinite dimensional Lie algebras, 3rd ed., Cambridge Univ. Press, Cambridge, UK,1990. [Ka2] V.Kac, Infinite root systems, representations of graphs and invariant theory I, Invent. math.56(1980), 57-92. [Ka3] V.Kac, Root systems, representations of quivers and invariant theory, in Springer Lee. Notes 996 (1983), 74-108. [Kap] M.M.Kapranov, Eisenstein series and quantum affine algebras, Journal of Mathematical Science 84(5) (1997), 1311-1360. [Kr] L.Kronecker, Algebraische Reduktion der Scharen bilinearer Formen, Sitzungsber. A kad. Berlin (1890),1225-1237. [LP] Y.Lin and L.Peng, Elliptic Lie algebras and tubular algebras, to appear in Adv.Math. [Lin] Z.Lin, Lusztig's geometric approach to Hall algebras, Representations of finite dimensional algebras and related topics in Lie theory and geometry, 349-364, Fields Inst. Commun., 40, Amer. Math. Soc, Providence, RI, 2004. [LXZ] Z.Lin, J.Xiao and G.Zhang, Representations of tame quivers and affine canonical bases, preprint, 2004. [LI] G. Lusztig, Canonical bases arising from quantized enveloping algebras, J. Amer. Math. Soc. 3 (1990), 447-498. [L2] G. Lusztig, Quivers, perverse sheaves, and the quantized enveloping algebras, J. Amer. Math. Soc. 4 (1991), 366-421. [L3] G. Lusztig, Affine quivers and canonical bases, Inst. Hautes Etudes Sci. Publ. Math. 76 (1992), 111-163. [L4] G. Lusztig, Canonical bases and Hall algebras, A.Broer and A.Daigneault(eds.), Representation Theories and Algebraic Geometry, 365-399, 1998 Kluwer Academic Publishers. [L5] G. Lusztig, Introduction to Quantum Groups, Birkhauser, Boston, 1993. [L6] G. Lusztig, Left cells in Well groups, in Lie group representations, I, 99-111, Lecture Notes in Math., 1024, Springer-verlag, 1983. [MZ] R. Martnez-Villa and D.Zacharia, Approximations with modules having linear resolutions, J. Algebra 266 (2003), no. 2, 671-697. [N] H. Nakajima, Crystal, canonical and PBW bases of quantum affine algebras, preprint (2005). [Ob] Abdukader Obul, The Serre relations in Ringel-Hall algebras, Chinese Ann. Math. Ser. B 23(2002), no. 3, 349-360. [PX] L.Peng and J.Xiao, Triangulated categories and Kac-Moody algebras, Invent, math. 140 (2000), 563-603. [Re] M.Reineke, The monoid of familes of quiver representations, Proc. London, Math. Soc,3(84) (2002) ,663-685. [Rl] C. M. Ringel, Hall algebras and quantum groups, Invent, math. 101 (1990), 583-592. [R2] C. M. Ringel, The composition algebra of a cyclic quiver, Proc. London Math. Soc. 66 (1993), 507-537. [R3] C. M. Ringel, PBW-bases of quantum groups, J.reine angew.Math. Vol. 470 (1996), 51-88. [R4] C. M. Ringel, The Hall algebra approach to quantum groups, Aportaciones Matematicas Comunicaciones 15 (1995), 85-114. [R5] C. M. Ringel, Hall algebras, In: Topics in Algebra, Banach Center Publ. 26 (1990), 433-447. [R6] C. M. Ringel, Tame algebras and integral quadratic forms, Lect.Notes Math. 66 (1099), Springer-Verlag,Berlin-New York 1984. [R7] C. M. Ringel, Hall algebras revisited, Israel Mathematical Conference Proceedings, Vol. 7 (1993), 171-176. [Sa] Y. Saito, PBW basis of quantized universal enveloping algebras, Publ.RIMS,30 (1994), 209- 232. [Si] O. Schiffmann, The Hall algebra of a cyclic quiver and canonical bases of Fock spaces, Internat. Math. Res. Notices (2000), 413-440. [S2] O. Schiffmann, Noncommutative smooth projective curves and quantum loop algebras, Duke Math. J. 121 (2004), 113-168. [S3] O. Schiffmann, Quivers of type A, flag varieties and representation theory, in Representations of finite dimensional algebras and related topics in Lie theory and geometry, 453-479, Fields Institute Communications 40, Amer. Math. Soc, Providence, RI,(2004).
254 JIE XIAO AND GUANGLIAN ZHANG [S4] O. Schiffmann, Canonical bases of loop algebras via quot schemes I, preprint math.QA/0404032. [S5] O. Schiffmann, Canonical bases of loop algebras via quot schemes II, preprint math.QA/0408401. [SV] B.Sevenhant and M.Van den Bergh, A relation between a conjecture of Kac and the structure of the Hall algebra, J. Pure and Appl. Algebra 160(2001), 319-332. [VV] M. Varagnolo and E. Vasserot, On the decomposition matrices of the quantized Schur algebra, Duke Math. J. 100 (1999), 267-297. [XY] J. Xiao and S. Yang, BGP-reflection functors and Lusztig's symmetries: a Ringel-Hall algebra approach to quantum groups, J.Algebra 241 (2001),204-246. [Z] P.Zhang, PBW-bases of the compostion algebra of the Kronecker algebra, J.reine angew.Math. Vol. 527 (2000), 97-116, Department of Mathematics, Tsinghua University, Beijing 100875,P.R.China E-mail address: jxiaoQmath.tsinghua.edu.cn Department of Mathematics, Tsinghua University, Beijing 100875,P.R.China E-mail address: zhangguanglianQmails.tsinghua.edu. en
Titles in This Series 414 Deguang Han, Palle E. T. Jorgensen, and David Royal Larson, Editors, Operator theory, operator algebras, and applications, 2006 413 Georgia Benkart, Jens C. Jantzen, Zongzhu Lin, Daniel K. Nakano, and Brian J. Parshall, Editors, Representations of algebraic groups, quantum groups and Lie algebras, 2006 412 Nikolai Chernov, Yulia Karpeshina, Ian W. Knowles, Roger T. Lewis, and Rudi Weikard, Editors, Recent advances in differential equations and mathematical physics, 2006 411 J. Marshall Ash and Roger L. Jones, Editors, Harmonic analysis: Calderon-Zygmund and beyond, 2006 410 Abba Gumel, Carlos Castillo-Chavez, Ronald E. Mickens, and Dominic P. Clemence, Editors, Mathematical studies on human disease dynamics: Emerging paradigms and challenges, 2006 409 Juan Luis Vazquez, Xavier Cabre, and Jose Antonio Carrillo, Editors, Recent trends in partial differential equations, 2006 408 Habib Aramari and Hyeonbae Kang, Editors, Inverse problems, multi-scale analysis and effective medium theory, 2006 407 Alejandro Adem, Jesus Gonzalez, and Guillermo Pastor, Editors, Recent developments in algebraic topology, 2006 406 Jose A. de la Pefia and Raymundo Bautista, Editors, Trends in representation theory of algebras and related topics, 2006 405 Andrew Markoe and Eric Todd Quinto, Editors, Integral geometry and tomography, 2006 404 Alexander Borichev, Hakan Hedenmalm, and Kehe Zhu, Editors, Bergman spaces and related topics in complex analysis, 2006 403 Tyler J. Jarvis, Takashi Kimura, and Arkady Vaintrob, Editors, Gromov-Witten theory of spin curves and orbifolds, 2006 402 Zvi Arad, Mariagrazia Bianchi, Wolfgang Herfort, Patrizia Longobardi, Mercede Maj, and Carlo Scoppola, Editors, Ischia group theory 2004, 2006 401 Katrin Becker, Melanie Becker, Aaron Bertram, Paul S. Green, and Benjamin McKay, Editors, Snowbird lectures on string geometry, 2006 400 Shiferaw Berhanu, Hua Chen, Jorge Hounie, Xiaojun Huang, Sheng-Li Tan, and Stephen S.-T. Yau, Editors, Recent progress on some problems in several complex variables and partial differential equations, 2006 399 Dominique Arlettaz and Kathryn Hess, Editors, An Alpine anthology of homotopy theory, 2006 398 Jay Jorgenson and Lynne Walling, Editors, The ubiquitous heat kernel, 2006 397 Jose M. Mufioz Porras, Sorin Popescu, and Rubi E. Rodriguez, Editors, The geometry of Riemann surfaces and Abelian varieties, 2006 396 Robert L. Devaney and Linda Keen, Editors, Complex dynamics: Twenty-five years after the appearance of the Mandelbrot set, 2006 395 Gary R. Jensen and Steven G. Krantz, Editors, 150 Years of Mathematics at Washington University in St. Louis, 2006 394 Rostislav Grigorchuk, Michael Mihalik, Mark Sapir, and Zoran Sunik, Editors, Topological and asymptotic aspects of group theory, 2006 393 Alec L. Matheson, Michael I. Stessin, and Richard M. Timoney, Editors, Recent advances in operator-related function theory, 2006 392 Stephen Berman, Brian Parshall, Leonard Scott, and Weiqiang Wang, Editors, Infinite-dimensional aspects of representation theory and applications, 2005
TITLES IN THIS SERIES 391 Jiirgen Fuchs, Jouko Mickelsson, Grigori Rozenblioum, Alexander Stolin, and Anders Westerberg, Editors, Noncommutative geometry and representation theory in mathematical physics, 2005 390 Sudhir Ghorpade, Hema Srinivasan, and Jugal Verma, Editors, Commutative algebra and algebraic geometry, 2005 389 James Eells, Etienne Ghys, Mikhail Lyubich, Jacob Palis, and Jos£ Seade, Editors, Geometry and dynamics, 2005 388 Ravi Vakil, Editor, Snowbird lectures in algebraic geometry, 2005 387 Michael Entov, Yehuda Pinchover, and Michah Sageev, Editors, Geometry, spectral theory, groups, and dynamics, 2005 386 Yasuyuki Kachi, S. B. Mulay, and Pavlos Tzermias, Editors, Recent progress in arithmetic and algebraic geometry, 2005 385 Sergiy Kolyada, Yuri Manin, and Thomas Ward, Editors, Algebraic and topological dynamics, 2005 384 B. Diarra, A. Escassut, A. K. Katsaras, and L. Narici, Editors, Ultrametric functional analysis, 2005 383 Z.-C. Shi, Z. Chen, T. Tang, and D. Yu, Editors, Recent advances in adaptive computation, 2005 382 Mark Agranovsky, Lavi Karp, and David Shoikhet, Editors, Complex analysis and dynamical systems II, 2005 381 David Evans, Jeffrey J. Holt, Chris Jones, Karen Klintworth, Brian Par shall, Olivier Pfister, and Harold N. Ward, Editors, Coding theory and quantum computing, 2005 380 Andreas Blass and Yi Zhang, Editors, Logic and its applications, 2005 379 Dominic P. Clemence and Guoqing Tang, Editors, Mathematical studies in nonlinear wave propagation, 2005 378 Alexandre V. Borovik, Editor, Groups, languages, algorithms, 2005 377 G. L. Litvinov and V. P. Maslov, Editors, Idempotent mathematics and mathematical physics, 2005 376 Jose" A. de la Pefia, Ernesto Vallejo, and Natig Atakishiyev, Editors, Algebraic structures and their representations, 2005 375 Joseph Lipman, Suresh Nayak, and Pramathanath Sastry, Variance and duality for cousin complexes on formal schemes, 2005 374 Alexander Barvinok, Matthias Beck, Christian Haase, Bruce Reznick, and Volkmar Welker, Editors, Integer points in polyhedra—geometry, number theory, algebra, optimization, 2005 373 O. Costin, M. D. Kruskal, and A. Macintyre, Editors, Analyzable functions and applications, 2005 372 Jose1 Burillo, Sean Cleary, Murray Elder, Jennifer Taback, and Enric Ventura, Editors, Geometric methods in group theory, 2005 371 Gui-Qiang Chen, George Gasper, and Joseph Jerome, Editors, Nonlinear partial differential equations and related analysis, 2005 370 Pietro Poggi-Corradini, Editor, The p-harmonic equation and recent advances in analysis, 2005 369 Jaime Gutierrez, Vladimir Shpilrain, and Jie-Tai Yu, Editors, Affine algebraic geometry, 2005 For a complete list of titles in this series, visit the AMS Bookstore at www.ams.org/bookstore/.
This book contains several well-written, accessible survey papers in many interrelated areas of current research. These areas cover various aspects of the representation theory of Lie algebras, finite groups of Lie types, Hecke algebras, and Lie superalgebras. Geometric methods have been instrumental in representation theory, and these proceedings include surveys on geometric as well as combinatorial constructions of the crystal basis for representations of quantum groups. Humphreys' paper outlines intricate connections among irreducible representations of certain blocks of reduced enveloping algebras of semi-simple Lie algebras in positive characteristic, left cells in two-sided cells of affine Weyl groups, and the geometry of the nilpotent orbits. All of these papers provide the reader with a broad picture of the interaction of many different research areas and should be helpful to those who want to have a glimpse of current research involving representation theory. ISBN O-flElfl-B^-l 9l'780821«'839249l CONM/413 AMS on the Web www.ams.org