-F'ublications Mathematical Research Institute 6
Representat Theoryof Flnlte Groups Editedby R.Solomon
deGruyter
Ohio State University Mathematical Research Institute Publications
2 3 4 5
Topology '90, B. Apanasov, W D. Neumann, A. W Reid, L. Siebenmann (Eds.) The Arithmetic of Function Fields, D. Goss, D. R. Ha.ves, M. 1. Rosen (Eds.) Geometric Group Theory, R. Charney, M. Davis, M. Shapiro (Eds.) Groups, Difference Sets, and the Monster, K. T Arasu, 1. F Dillon, K. Harada, S. Sehgal, R. Solomon (Eds.) Convergence in Ergodic Theory and Probability, V Bergelson, P March, 1. Rosenblatt (Eds.)
Representation Theory of Finite Groups Proceedings of a Special Research Quarter at The Ohio State University, Spring 1995
Editor Ronald Solomon
Waiter de Gruyter . Berlin· New York 1997
Editor RONALD SOLOMON
Department of Mathematics, The Ohio State University 231 West 18th Avenue, Co1umbus, OH 43210, USA Series Editors: Gregory R. Baker Department of Mathematics. The Ohio State University, Columbus, Ohio 43210-1174. USA Karl Rubin Department of Mathematics. Stanford University. Stanford. CA 94305-2125. USA Waiter D. Neumann Department of Mathematics. The Uniwrsity of Melbourne. Parkville. VIC 3052. Australia 1991 IvIathematics Suhject Classification: Primary: 20C Secondary: 20C15. 20C20, 20C30. 20C33 Kcnl"(lrds: Finite group, module, representation. modular representation, character. group algebra. module category
@
Prinll'd on acid-free paper \\-hich falls \\-ithill the guidelines of the A1\51 to ensure- permanence and durabilit~
Lihrar\' of' Congn'ss Cata!oging-in-Puhlication Dala
Representation theory of finite groups: proceedings of a special research quarter at the Ohio State University. spring. 1995 / editor. Ronald Solomon. p. cm. - (Ohio State University Mathematical Research Institute publications. ISSN 0942-0363 : 6) Includes bibliographical references. ISBN 3-II-OI5806-X (alk. paper) I. Finite groups - Congresses. 2. Representations of groups - Congresses. r. Solomon. R. C. (Ronald C.) n. Series. QAI77.R46 1997 5I2'.2-dc21 97-35939 CIP
Die Deutsche Bihliothek - CilIa!oging-in-PlIhlication Data
Representation theory of finite groups: proceedings of a special research quarter at The Ohio State University, spring 1995/ ed. Ronald Solomon. - Berlin: New York: de Gruyter. 1997 (Ohio State University, Mathematical Research Institute publications: 6) . ISBN 3-II-OI5806-X
© Copyright 1997 by Waiter de Gruyter & Co .. D-I0785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means. electronic or mechanical. including photocopy. recording. or any information storage and retrieval system. without permission in writing from the publisher. Printed in Germany. Typeset using the authors' T EX files: 1. Zimmermann. Freiburg. Printing: Werner Hildebrand. Berlin. Binding: Liideritz & Bauer GmbH, Berlin. Cover design: Thomas Bonnie. Hamburg.
AVPlfJJ!q lfiflj-tC1X!S S!lf fo UO!SV:J:JO alfJ uo nad JaJZVM 0L
Preface
In March 1896, Richard Dedekind wrote a letter to Georg Frobenius in which he defined his concept of the "group determinant" of a finite group and stated his results concerning the factorization of this polynomial when the group G is abelian. In a second letter in April, he formulated some conjectures concerning the factorization of the group determinant when G is non-abelian. Frobenius attacked the problem immediately and made substantial progress by Summer 1896. This year 1997 represents the centenary of fundamental papers by three mathematicians - Theodor Molien, Georg Frobenius and WilIiam Burnside each developing many of the foundational results of the theory of complex representations of finite groups. It is fitting then that we now publish this set of papers which gives some measure of the distance this theory has advanced in its first century and some clues as to the roads it will follow in its second century. These proceedings record some of the activity which took place during a special research quarter held at the Ohio State University in Spring 1995. This quarter and the concluding conference were supported by the O.S.U. Mathematical Research Institute and by the National Science Foundation. The single most monumental and important achievement of the theory of group characters is the Odd Order Theorem of Waiter Feit and John G. Thompson. This theorem which formed the entire content of Volume 13, no. 3, of the Pacific Journal of Mathematics, Fall 1963, is a remarkable fusion of the "local structure theory" of finite groups, initiated by Ludvig Sylow and the character theory of finite groups, initiated by Frobenius. And it was our great pleasure at this conference to honor Professor WaIter Feit of Yale University on the occasion of his sixty-fifth birthday. It was a pleasure to bring together many of Waiter's mathematical children and grandchildren, friends and admirers, to celebrate his illustrious career in mathematics. In addition to those who have contributed papers to this volume, many other mathematicians joined in the formal and informal discussions at the conference. We thank all of them for their enthusiastic participation. The speakers, their current affiliations and their topics are:
Preface
VlIl
10nathan L. Alperin, University of Chicago, On endo-permutation modules; Matthew K. Bardoe, Imperial College, University of London, Representations, embeddings and geometries; David G. Benson, University of Georgia, Cohomology of modules for a finite group; Ro.bert Boltje, University of Augsburg, Canonical induction formulae and the defect of a character; Michel BroU(~, Denis-Diderot Universite of Paris, The abelian defect group conjecture infinite reductive groups; Everett Dade, University of Illinois, Urbana-Champaign, Counting characters in blocks; Harald Ellers, Northern Illinois University, On Alperin 's weight conjecture and Brauer's First Main Theorem; Karin Erdmann, University of Oxford, Representations ofsymmetric groups and GLn(K); WaIter Feit, Yale University, Schur indices; Guoqiang Huang, Northern Illinois University, On extended block induction and Brauer's Third Main Theorem; Radha Kessar, Yale University, On blocks and source algebrasfor 2Sn and
2A n
;
Lluis Puig, CNRS, Institut de Mathematiques de Jussieu, On the Morita and Rickard equivalences between Brauer blocks; Geoffrey R. Robinson, University of Leicester, Some open conjectures in block theory; Leonard L. Scott, University of Virginia, On the Lusztig conjectures; Sergei Syskin, Reinsurance Group of America, Locally finite varieties and representations offinite groups; 1. P. Zhang, Peking University, Vertices of irreducible modules.
In conclusion it is a pleasure to acknowledge the efforts of Dr. Radha Kessar, who provided much assistance with the organization and correspon-
Preface
IX
dence for the conference. Also my deep appreciation goes to my wife, Myriam Solomon, who was the guiding hand for the conference reception and dinners and who created the lovely design for the conference coffee mugs, as well as of course tolerating me during the entire period. Many thanks also to Lluis and Isabel Puig, who often acted more as hosts than as guests and who in particular organized a lovely evening of music during the conference, graced by the musical talents of Marcus Linckelmann, Richard Lyons and Isabel Puig. We also grateful acknowledge the re-typing work of Mr. Jwalant Vakil and Ms. Terry England, who put this volume into final polished form. Again thanks to all of the participants and to the O.S.U. Mathematical Research Institute and the National Science Foundation for their generous financial support. Particular thanks to Ann Boyle and Andy Earnest for their efficient handling of our proposal and their kind words of encouragement.
Ron Solomon
Table of Contents
Preface
vii
M. K. Bardoe Embeddings, Geometries and Representations: Connections and Computations
1
D. J. Benson Infinite Dimensional Modules for a Finite Group
11
H. I. Blau Degrees and Diagrams of Integral Table Algebras
19
R. Boltje Canonical Induction Formulae and the Defect of a Character
29
E. C. Dade Counting Characters in Blocks, 2.9
45
H. Ellers The Defect Groups of a Clique
61
K. Erdmann Representations of GL n (K) and Symmetric Groups
67
G. Huang On Extended Block Induction and Brauer's Third Main Theorem
85
R. Kessar On Blocks and Source Algebras for the Double Covers of the Symmetric Groups
93
xii
Table of Contents
L. Puig A Survey on the Local Structure of Morita and Rickard Equivalences between Brauer Blocks
101
G. R. Robinson Some Open Conjectures on Representation Theory
127
L. L. Scott Are All Groups Finite?
133
S. A. Syskin Locally Finite Varieties of Groups and Representations of Finite Groups ,
149
Embeddings, Geometries and Representations: Connections and Computations M. K. Bardoe
Abstract. We outline some of the connections between representations and geometries through the computations of embeddings of geometries. We end with a table of many of the computations which have been completed.
Introduction It is the author's belief that the understanding of groups is simplified through the use of geometries on which these groups act. For instance, the geometries for groups of Lie-type, namely buildings, help make the p-Iocal structure of Lie-type groups over fields of characteristic p easily understood. Also, one may see that of the 26 sporadic simple groups, those which are best understood are those with easily understood geometric structures on which the groups act, e.g. M24, COl, and Fi24. Recently, through the work of Quillen, Brown, and Adem and Milgram, e.g. [AM] and others, it has become clear that understanding the geometries associated to a group may also be helpful in computing and understanding the cohomology of that group at a specific prime. Representation theory is also able to reap the benefits of this viewpoint. What we describe here is a two-way street between representation theory and geometry. One way is the method of constructing from a representation of a group a geometry for that group. This direction is well known and easily understood. The other direction is more complicated. It involves inductively creating modules, based on a geometry associated to a particular group, from modules for its subgroups. We give a short outline of what follows. First, we define geometry and give examples of geometries for some familiar groups. Then we outline some of the theory which describes how to construct representations which are described by geometries. We attempt to explain why this process is interesting to both geometers and to representation theorists. Finally we give a partial list computations and constructions which have been completed.
2
M. K. Bardoe
Geometries We take the following to be as our definition of geometry:
Definition 1: Geometry. A geometry of rank n is an ordered sequence
of n pairwise disjoint non-empty sets fi together with a symmetric incidence relation, *, on their union such that if F is any maximal set of pairwise incident elements then IF n fi I = 1 for each i. These first two examples demonstrate how representations have been used to construct geometries. Example 1: The Projective Geometry for an n -dimensional vector space V, denoted PG(V), is the n - 1 sets given by the I-spaces, 2-spaces, ... , n I-spaces with incidence between elements of different sets defined by inclusion. Example 2: Symplectic Geometry for a vector space V. Let V be a 2n -dimensional vector space with a non-degenerate alternating bilinear form. Then S(V) = (isotropic I-spaces, ... , isotropic n-spaces, *)
with incidence defined as inclusion. This is a geometry defined by the subspaces of a module for the group SP2n(k). This example shows a geometry which comes from a setting other than representation theory. Example 3: The M24 2-local geometry, [RS1]. This geometry is based on the Steiner system 5(5,8,24). In the terminology of our definition f = (Octads, Trios, Sextets, *)
An octad is the special 8-sets which form the blocks of the Steiner system, a trio is a set of mutually disjoint octads, and a sextet is a set of six mutually disjoint 4-sets such that any two 4-sets form an octad. An octad is incident with a trio if it is contained in that trio. An octad is incident with a sextet if it is the union of two of the 4-sets of the sextet. A trio is incident with a sextet if each of its octads is the union of two of the 4-sets of the sextet. In what follows it will be important to be able to view a geometry as a simplicial complex on which the automorphism group acts. This may be done
Embeddings, Geometries and Representations
3
in the following way. Define vertices to be the disjoint union of the fi. Then define n-simplices to be n-sets of V such that any two elements are incident, such a set is called a flag. Such a simplicial complex is of dimension one less than the rank of the geometry, and often goes under the name flag complex. The maximal simplices of this complex are termed chambers.
points (I-spaces) lines (2-spaces)
Figure 1: Geometry for SL 3 (2) viewed as simpIicial complex In the first two examples, PG(V) and S(V), the vertices of the simplicial complex are proper subspaces of V, and the larger dimension simplices are chains of subspaces ordered by inclusion. In the case of PG(V) the chambers are maximal chains of subspaces, while for S(V) chambers are maximal chains of isotropic subspaces.
Sheaf Theory From a group theoretic point of view, geometries are useful ways of describing a portion of the subgroup structure of a group. This may be most evident in the case of buildings and p-Iocal geometries for the sporadic groups. Simplistically, we construct a geometry by taking the conjugacy classes of maximal subgroups containing a Sylow p-subgroup to be the objects of our geometry.
4
M. K. Bardoe
Then say that two such subgroups are incident if their intersection contains a common Sylow p-subgroup. Therefore a natural way to exploit geometries in representation theory is to say we understand something about the way a representation restricts to the stabilizers of objects of a p-Iocal geometry. Then what further can we say about representation for the whole group? This work essentially started, in the case of modular representations, with the work Ronan & Smith [RS2]. Inspired by the work of Lusztig, [L] on the representation theory of Lie-type groups over the complex numbers, Ronan & Smith show how to construct a representation for a group by using the geometry to "weave" representations of the various stabilizers of simplices into a coherent representation for the whole group. This is done through the formalism of sheaves. Let k be a field.
Definition 2: Sheaf. A sheaf J' on f assigns to each simplex a E f a representation J'u of Stab(a) C Aut(r). If r is a face of a there is a linear connecting map cPar: : J'a ---* J'r: such that cPpa 0 cPar: = cPpr: whenever this composite map is defined. The J'a and cPar: are required to be Aut(r)-equivariant in the following sense: For each g E Aut(r) there is a mapping g : J' ---* J' such that gh = gft, J'ag = (J')g, and go cPag,r:g = cPar: 0 g. The homology groups of such a sheaf J' are k Aut(f) -modules. In particular, if we assume that the terms of J' are generated by the images of J'c for chambers, what is called chamber generated, then the zero homology of this sheaf is a module that has the extra condition that one can recover J' through the submodule structure of V ~ Ho(J'). Namely, if a, rare simplices of the simplicial complex derived from f, then there exist submodules Va of V upon restriction Stab(a) such that Va ~ ::1a' Also, if rea, then Va C Vr:' Note that the star of a simplex, St(a), is also a simplicial complex, and that a sheaf, ::1, on f defines a sheaf, 9, for St(a). Therefore if one understands the zero homology of sheaves for St(a) for a of dimension S rank(r) - 2, then one may make the following inductive step: Given a module for the stabilizer of a chamber, Mc, and modules for the stabilizers of the faces of the chambers, Mrr;, with connecting maps, cPc,rri' then one may construct a universal sheaf, U, such that the module at any chamber is isomorphic to Mc and Urr; ~ Mrr ;, and the modules at other simplices are defined to be Ho(St(a)).
Embeddings, Geometries and Representations
Example 4: [R82] Let
5
r
be PG(V) where V is a 3 dimensional space over F2. Then Aut(r) ~ SL3(2). Suppose we define a chamber generated sheaf U by assigning to a chamber a fixed I-space. The face of a chamber associated to a point is assigned a fixed I-space, and the face of a chamber associated to a line is assigned a 2-space which contains the three fixed I-spaces of the chambers which contain the line. In the terminology of our definition of a sheaf, Mp is a trivial module for the point and chamber stabilizers and M[ is the 2-dimensional irreducible for the SL2 (2) quotient inflated to the full tine stabilizer, and the connecting maps are inclusion maps. Then Ronan & Smith show that Ho(U) ~ V for the sheaf defined by these conditions.
Another result of Ronan & Smith, [R82], shows that if one can form a sheaf, ~, from the submodule structure of a module, V, then V is a quotient of Ho(~. Therefore from the last example we see that V is the only module satisfying the condition described by that sheaf, as Ho(U) is an irreducible module for SL3 (2). Therefore this result about the form Ho(~ provides a kind of local recognition result for SL3(2) modules. And in general computation of Ho(~ provides a recognition result for modules of Aut(r). The last example should give you an idea as to how to ask the relevant questions about sheaves but tells you little about how to compute. In general computations are ad hoc in nature and not very enlightening.
Embeddings One area in which many computations have been done has been that of embeddings of geometries. Motivated at least partially by the newer machinery of sheaves and a classical geometric question, a new focus has been centered on the question of what projective geometries can an abstract geometry, such as the one in Example 2, be a subgeometry of. In particular, if we restrict attention to a rank 2 geometry where elements of one set are called points and the elements of the other are termed lines, what projective geometries is this geometry a subgeometry of?
Definition 3: Point-Line Embedding. An embedding of a geometry r is an injective incidence preserving map, rr, from r to PG(V), the projective geometry of a vector space over the appropriate field, such that the points are mapped into the I-spaces of V and the lines are mapped into the 2-spaces of V.
6
M. K. Bardoe
Here is an example of a point-line embedding of the geometry of octads, viewed as points, and trios, viewed as lines, coming from Example 3. Example 5: Geometry of octads and trios from the 2-local geometry for M24. Let V be the 11 dimensional irreducible quotient of the binary Golay code. Then the geometry in Example 2 is isomorphic to
r
= (special I-spaces, special 2-spaces, *)
where incidence defined by inclusion, and special indicates a specific orbit of subspaces under the action of M24. In an attempt to find embeddings for a geometry r, we rephrase this question into the language of sheaf theory in the following way: What modules support the sheaf, ~, given by assigning a I-space to the point stabilizer and a 2-space to the line stabilizer? From what we have said above one such module is Ho(~. This module is known to geometers as the universal embedding of a geometry because there cannot exist a larger embedding of the geometry, and any embedding map is factored by the universal embedding map. The situation is particularly nice when we are working with a geometry with 3 points per line. Notice that if our geometry has three points per line, then the most natural projective space to have our geometry embed into is a projective space coming from a vector space over F2. In the case of F2 vector space we have a unique vector in a I-space. Therefore we can define an embedding as a map n' from r to a set of vectors of V the vector space underlying our projective space. Also, the requirement that the 3 vectors assigned to the points of a line span a 2-space is equivalent to the following equation:
vp
+ vq + v, = 0 for l = {p, q, r}
From this we see that we can write a presentation of the universal embedding by starting first with a vector space, with basis indexed by the points of our geometry, and then quotienting out by the subspace spanned by all of the vectors of the type v p + vq + v, = 0 for l = {p, q, r}.
Conclusion Many universal embedding questions have been determined for many of the simple groups and related groups. These computations may be of interest to geometers because embeddings often are helpful in classification and computational problems. They may be of interest to representation theorists because
Embeddings, Geometries and Representations
7
they show which modules are classified by their restrictions to important subgroups of these groups. Below is a list of many of these results and references for them. In this list irreducible modules are denoted by their dimension, and duality is indicated by a overline. Geometry Building Long root Neimir geometry
Aut(r) Lie-type Group of type A, D or E A7
Near-hexagon
M24
Near-hexagon
U4(3)
Near-octagon
12.2
Near-hexagon
3D4(2)
Near-hexagon
U6(2).2
Near-hexagon
07(2)
Univ. Embedding Natural Module Adjoint Module 0
Remarks [RS2] [SV], [VJ [RS3]
11 IE&TI 20
[RS3]
26 IE&I 26 IE&I 20 IE&I 8 0-
[FS]
T
[Y] [FS]
[Y] [Y]
T
Involution geometry
U4(3)
Involution geometry
Suz
Involution geometry 2-local geometry 2-local geometry 2-local geometry Tilde geometry Tilde'geometry Tilde geometry Tilde geometry Petersen geometry Petersen geometry Petersen geometry Petersen geometry Petersen 'geometry Petersen 'geometry Petersen' geometry Petersen geometry Petersen geometry
Co, Co, He Ru M24 3. Sp4(2) He M Ss Aut(M22) 3. Aut(M22) M23 CO2 3:23. CO2 14
BM 4j/l .BM 3
34 1
EB
-,-
34' 1
[B3]
142
[B2]
274 'E&1E&24
[Bl] [Srn] [MS] [MS] [IS2]
24 51 28 11 6EB5 52 0 6 11 12 EB 11 0 23 23 0 0 0
[IS2] [IS2] [IS2] [IS 1] [IS1] [IS 1] [IS I] [IS3] [Sh] [IS2] [IS2]
8
M. K. Bardoe
Bibliography [AM]
A. Adem and R. 1. Milgram, The cohomology of the Mathieu group M22, Topology 34 (1995),389-410.
[B 1]
M. K. Bardoe, The universal embedding for the Co I involution geometry, in preparation, 1995.
[B2]
M. K. Bardoe, The universal embedding for the Suzuki sporadic simple group, Preprint, accepted to J. Algebra, 1995.
[B3]
M. K. Bardoe, The universal embedding for the U4(3) involution geometry, Preprint, accepted to 1. Algebra, 1995.
[FS]
D. Frohardt and S. Smith, Universal embeddings for the 3 D4(2) hexagon and the h near-octagon, Europ. 1. Combin. 13 (1992), 455-472.
[IS 1]
A. A. Ivanov and S. V. Shpectorov, Geometries for sporadic groups related to the Petersen graph. n, Europ. J. Combin. 10 (1989),347-361.
[IS2]
A. A. Ivanov and S. V. Shpectorov, The flag-transitive tilde and Petersen-type geometries are all Known, BuII. Amer. Math. Soc. 31 (1994),173-184.
[IS3]
A. A. Ivanov and S. V. Shpectorov, Natural representations of the P-geometries of C02-type, J. Algebra 164 (1994), 718-749.
[L]
G. Lusztig, The Discrete Series Representations of the General Linear Groups over a Finite Field, Annals of Mathematics Studies 81, Princeton Univ. Press, Princeton, NJ. 1974.
[MS]
G. Mason and S. Smith, Minimal 2-local geometries for the Held and RudvaIis sporadic Groups, J. Algebra 79 (1982), 286-306.
[RS1]
M. A. Ronan and S. D. Smith, 2-Local geometries for some sporadic groups, in: B. Cooperstein and G. Mason, editors, The Santa Cruz Conference on Finite Groups, Proc. Symp. Pure Math. 37, Amer. Math. Soc., Providence RI, 1980, 283-289.
[RS2]
M. A. Ronan and S. D. Smith, Sheaves on buildings and modular representations of ChevaIley Groups, J. Algebra 96 (1985), 319-346.
[RS3]
M. A. Ronan and S. D. Smith, Computation of 2-modular sheaves and representations for L4(2), A7, 356, and M24, Comm. Algebra 17 (1989),1199-1237.
[Sh]
S. V. Shpectorov, Natural representations of some tilde and petersen type geometries, Geom. Dedicata 54 (1995), 87-102.
[Srn]
Stephen D. Smith, Universality of the 24-dimensional embedding of Comm. Algebra 22 (1995), 5159-5166.
[SV]
COl,
S. Smith and H. Vblklein, A geometric presentation for the adjoint module of 5L3(k), 1. Algebra 127 (1989),127-138.
[V]
H. VOlklein, On the geometry of the adjoint representation of a ChevaIley group, 1. Algebra 127 (1989), 139-154.
Embeddings. Geometries and Representations
[Y]
9
S. Yoshiara, Embeddings of flag-transitive classical locally polar geometries of rank 3, Geom. Dedicata 43 (1992), 121-165.
Imperial College of Science and Technology London SW7 287 England Email:
[email protected]
Infinite Dimensional Modules for a Finite Group D. J. Benson
This paper is a transcription of the lecture I gave at the Ohio State University Conference on Representation Theory of Finite Groups. My intention was to talk about the ideas involved in a small corner of my recent joint work with Jon Carlson and Jeremy Rickard [Be2] [Be3] on infinitely generated modules, and try to explain the role of generic points of varieties in this context. As we move into the second century of finite group representation theory, we still find that the vast majority what is being done is concerned with finitely generated modules. It seems to me that the reason for this is largely that we have very few techniques that work in a wider context, say for example the context of arbitrary modules for the group algebra of a finite group over a field. Linear transformations on infinite dimensional spaces don't necessarily have any eigenvalues. There are modules with no indecomposable summands. The Krull-Schmidt theorem fails quite badly, so there are no vertices and sources. There is a module M for Z2 x Z2 in characteristic two, which satisfies M ~ M EBM EBM but not M ~ M EBM. In the light of these pathologies, it is tempting just to give up and return to the relatively safe world of finitely generated modules, especially as there are still many interesting unanswered questions in this context. However, it turns out that even if we are only interested in finitely generated modules, there are recent theorems whose proofs use infinitely generated modules in an essential way; for example, Rickard's (as yet unpublished) classification of the thick subcategories of the stable finitely generated module category for a p-group, and my recent proof [Bel] of the conjectures formulated in [Be4] about finitely generated modules with no cohomology.
1. A Vector Space Lemma The following lemma serves as a replacement for the theory of eigenvalues, and is really the starting point for the recent developments I'm going to discuss. Lemma 1.1. Let k be an algebraically closed field, and k(t) be a simple transcendental extension of k. Let V be a nonzero (and possibly infinite
12
D. J. Benson
dimensional) k-vector space, and let f be linear transformations from V to itself. Thenfor some A in k(t), the linear map
1 0 f - A0 Identity: k(t) 0k V -+ k(t) 0k V is not an isomorphism.
In other words, the reason why there need be no eigenvalues, even over an algebraically closed field, is because the field isn't big enough. Over some extension field, there is always an eigenvalue, if interpreted suitably. The appearance of transcendental extensions in this lemma gives rise to the relevance of "generic points" for infinite dimensional representations, in a way that never becomes relevant for finite dimensional representations. The proof of the lemma is very straightforward. We regard V as a k[x]module in the normal way by letting x act as the linear transformation f. If f - A.Identity is an isomorphism for all A E k, then the action of k[x] extends to an action of the field of fractions k(x). Since k(x) is a field, any nonzero k(x)-module has a summand isomorphic to k(x) itself. But multiplication by 1 0 x - t 0 1 is not an isomorphism on k(t) 0k k(x). The form in which the lemma gets used is the following: if f, g : V -+ W are two maps with the property that after tensoring with k(t), all nontrivial linear combinations of f and g give isomorphisms from k(t) 0k V to k(t) 0k W, then the vector spaces V and Ware both zero. To reduce to the previous form of the lemma, use g to identify V with W.
2. Dade's Lemma The way we use the vector space lemma of the last section is via an infinite dimensional version of Dade's lemma. The original lemma (Dade [Da]) says the following. Let k be a field of characteristic p, and let E = (gl, ... , gr} be an elementary abelian group of order p", Let Xi be the element gi - 1 of the group algebra kE, so that = 0 and Xi is in the Jacobson radical J(kE). Let VE(k) denote the quotient space J(kE)j J 2(kE), and let XI, ... , Xr be the images of X I, ... , X r in VE(k). It is not hard to show that they form a basis for this quotient space. Let YI, ... , Yr be the linear functions VE (k) -+ k given by Yi(Xj) = 1 if i = j and zero otherwise. Then regarding VE(k) as an affine space, its coordinate ring is the polynomial ring k[YI, ... , Yr]. If
Xi
ex = AIXI
+ ... + ArXr E
VE(k)
Infinite Dimensional Modules for a Finite Group
13
is not equal to zero, then we set
an element of order p in the group algebra kE. Theorem 2.1 (Dade's Lemma). Suppose that k is algebraically closed. If M is a finitely generated k E -module such that the restriction of M to (u a ) is free for each point i- a E Vf(k), then M is afree kE -module.
°
The hypothesis that M is finitely generated is certainly necessary here. There are examples of infinitely generated modules for 2/2 x 2/2 which are not projective, but whose restriction to each (u a ) is free (an example is sketched in Section 6). But in some sense, this is because the field is not big enough, because after enlarging the field, one finds that there are values of a for which the restriction to (u a ) is not free. The correct version of Dade's lemma for modules which are not necessarily finitely generated was formulated in [Be3l: Theorem 2.2. Let K be an algebraically closed extension of k of transcendence degreee at least r - 1, and set Vf(K) = 1(KE)/1 2(KE). If M is a kE-module such that (K 0k M) +(u,,} is free for all nonzero a E Vf(K), then M is a free k E -module. The idea of the proof is to reduce to the rank two case, and then choose two linearly independent elements 11, 11' E H 1(E, IFp ) ~ 1(IFpE)/1
so that the Bocksteins f3(11) and of H 2 (E, IFp). They induce maps
2(IF
pE),
f3(11') are algebraically independent elements
with the property that for all A, f-LE K, not both zero, Af
+ ug
~
---2
: ExtKE(K, K 0k M) -+ ExtKE(K, K 0k M)
is an isomorphism. It now follows from the vector space lemma discussed earlier, that Ext~E(k, M) = 0, which implies that M is free.
14
D. J. Benson
3. The Rank Variety For M a finitely generated kE-module, with k algebraically closed, Carlson's definition [Ca] of the rank variety of M is VE(M)
= {OI- ex E
VE(k)
IM
,l.(u a )
is not free} U {O}.
This is a closed homogeneous subvariety of VE(k), and Dade's lemma may be interpreted as saying that M is free if and only if VE(M) = {O}. In fact, the dimension of VE(M) determines the polynomial rate of growth ofthe minimal free resolution of M as a k E -module, which is called the complexity of the module. Thus for example the dimension of VE(M) is equal to one if and only if the minimal resolution of M is periodic (after the first term, which may be too big because of the free summands of M). For infinitely generated modules, the theory is somewhat different, because we must extend the field in order for Dade's lemma to hold. The naive thing to do is just to look at VE(K ®k M), where K is a "large enough" transcendental field extension of k. In general this is not a closed subset of VE(K), so what sorts of subsets occur this way? To answer this question, we next discuss the theory of generic points.
4. Generic Points For this section, we suppose that k is algebraically closed, and that K is an extension of k of transcendence degree at least r. Recall that if V s:;VE(k) is a closed irreducible subvariety, then the set p = {f
E
k[YI, ... , Yr]
I f vanishes on
V}
is a prime ideal in k[YI, ... , Yr], and k[V] = k[YI, . " , Yr]/p
is the coordinate ring of V. It is an integral domain, and its field of fractions key) is the function field of V. It is generated as an extension field of k by the images YI,... ,Yr of the elements YI, ... , Yr. The transcendence degree of key) over k is equal to the Krull dimension of k[V], and is by definition the dimension of the variety V. Since K is an algebraically closed extension field of k of transcendence degree at least as big as that of key), it follows that the inclusion of k into K extends (not by any means uniquely) to an embedding of key) into K. Let
Infinite Dimensional Modules for a Finite Group
15
tl, ... , t, be the images of YI, ... , Yr under such an embedding. The generic point of V is defined to be the element YIXI
+ ... + Yrxr
E
VE(k(V».
Note that this point is well defined, independently of the chosen basis for VE(k). A point which is of the form tlxl
+ ... + t.x,
E
VE(K)
for some embedding of k(V) into K as above, is said to be a generic point of V over K. If tlxl + ... + t.x, is any point in VE(K), set p S; k[YI, , Yr] equal to the ideal consisting of all polynomial relations satisfied by ti, .t, over k. Since K is a field, this ideal is necessarily prime. Let V be the associated subvariety of VE(k). Then the point tlxl + ... + t-x, is a generic point of V over K. So every point defined over K is generic for some uniquely determined closed irreducible subvariety defined over k. If we look at a line through the origin in VE(K), the points in that line may be generic for possibly different subvarieties. However, there is a uniquely determined homogeneous subvariety (i.e., one which is a union of straight lines through the origin) among them. To see this, if tlXI + ... + t.x, is generic for some inhomogeneous subvariety V, then the dimension of V is less than r, so there is an element A E K which is algebraically independent of tl, ... , t.. Then the point Mlxl + ... + Mrx r is generic for the homogeneous hull of V, namely the smallest homogeneous subvariety containing V. Now if M is a kE -module (not necessarily finitely generated), then the question of whether (K 0k M) -J-.(u a ) is free only depends on the line through a in VE(K), and then only on the closed homogeneous irreducible subvariety of VE(k) for which it is generic. So we define VE(M) to be the collection of nonzero closed homogeneous irreducible subvarieties V of VE(k) with the property that if a is the generic point of V then (k(V) @k M) -J-.(u a) is not free. It is easy to see that each of VE(K 0k M) and VE(M) determines the other.
5. Properties of
VB(M)
The following list of properties of VE(M) when M is not necessarily finitely generated parallels the list of properties of VE(M) in the finitely generated case. (i)
VE(M) = 0 if and only if M is projective.
16
D. J. Benson
(ii) More generally, the dimension of V is at most e for all V E V'E(M) if and only if M may be expressed as a filtered colimit of finitely generated
modules of complexity at most e. (iii) V'E(M EBN) = V'E(M) U V'E(N). (iv) V'E(M 0k N) = V'E(M) n V'E(N). (v) Every subset of V'E(k) = Projk[YI, ... , Yr] is equal to V'E(M) for a suitable module M. (vi) If M happens to be finitely generated, then V'E(M) is just the collection of all closed homogeneous irreducible subvarieties of VE(M).
6. An Example Let k be an algebraically closed field of characteristic two, and let E = Zj2 x Zj2 = (gl' g2). Set XI = g, - 1 and X2 = g2 - I. Let M, be the kE-module with generators ml, m i, ... and relations Xlml = 0, and X2mi = X,mi+1 for each i ~ 1. Let M2 be the kE-modulewithgenerators m;, m;, ... and relations X2m; = X I m;+1 for each i ~ 1. The modules M, and M2 are superficially similar, but MI has complexity one, while M2 has complexity two. The set V'E(MI) consists of just a single line through the origin, while
V'E(M2) = V'E(k) \ VE(M,). Let
aX]
+ bX2
and eX,
+ dX2,
where (:
:)
is a nonsingular matrix. This
has the effect of moving the distinguished line through the origin by the corresponding linear transformation on VE(k). If the resulting line corresponds to the point (A : fJ,) E pi (k), let us write MO.:/-l) for the module obtained from M2 in this way. Up to isomorphism, it only depends on the point (A : fJ,) E p' (k). Let
EB (A:/-l)EIP" (k)
MO ..:/-l)·
Infinite Dimensional Modules for a Finite Group
17
Let M be the cokernel of this map. Then M is projective on restriction to (u a ) for every nonzero ex E V~(k), but M itself is not projective. In some sense, what we have done is to take affine 2-space, and remove all the lines through the origin. However, what we haven't done is to remove lines through the origin defined over a larger field. So the generic point of the affine plane is still in the variety. In fact, we have VE(M) = {V~(k)}.
References [Be I]
D. 1. Benson, Cohomology of modules in the principal block of a finite group, New York 1. Math. 1 (1995), 196-205.
[Be2]
D.1. Benson, J. F. Carlson and 1. Rickard, Complexity and varieties for infinitely generated modules, Math. Proc. Cambridge Philos. Soc. 118 (1995), 223-243.
[Be3]
D.1. Benson, 1. F. Carlson and J. Rickard, Complexity and varieties for infinitely generated modules II, Math. Proc. Cambridge Philos. Soc. 120 (1996), 597-615.
[Be4]
D. 1. Benson, 1. F. Carlson and G. R. Robinson, On the vanishing of group cohomology, 1. Algebra 131 (1990),40-73.
[Cal
J. F. Carlson, The varieties and cohomology ring of a module, J. Algebra 85 (1983), 104-143.
[Da]
E. C. Dade, Endo-permutation modules over p-groups II, Annals of Math. 108 (1978), 317-346.
Department of Mathematics University of Georgia Athens, GA 30602 USA Email:
[email protected]
Degrees and Diagrams of Integral Table Algebras Harvey I. Blau
Integral table algebras satisfy fundamental properties abstracted from group algebras and character rings of finite groups, and from the adjacency algebras of commutative association schemes. Implicit in the work of Schur [Se], these structures, or variations of them, have been defined and studied independently under several guises, including" C -algebras" [K, BI], "pseudo groups" [Br], and "hypergroups" (as in [McM]), as well as "table algebras" [A I, AF, Bll l. A connection has been observed recently [B12l between table algebras and certain diagrams, among which the affine diagrams of Lie theory and combinatorial geometry are special cases. We present here an exposition of this link, give some examples, and state some results which show how the structure of an integral table algebra, and of its associated diagram, can be determined in certain situations by information on the "degrees" (defined below) of its distinguished basis elements. The most recent results, Theorems 2 and 3 below, are joint work of Bangteng Xu and the author. Throughout, C denotes the complex numbers, jR the reals, and jR+ the positive reals.
Definition. [AI, AF, A2, BIll Let B = {bl, b2, ... ,bd be a basis of a finite dimensional, associative and commutative algebra A over C, with identity element lA = bl. Then (A, B) is a table algebra (and B is a table basis) if and only if the following hold: (I)
For all i, j, rn, bibj = L~=I f3ijmbm, with f3ijm E jR+ U {O}.
(ll)
There is an algebra automorphism (denoted by -) of A whose order divides 2, such that bi E B implies hi E B. (Then i is defined by bi = hi, and bi E B is called real if i = i.)
(Ill) For all i,j, f3ijl #0 if and only if j
=i.
20
Harvey I. Blau
If (A, B) is a table algebra, then there exists a unique algebra homomorphism f : A -+ C such that f(bi) E lR.+ for all bi E B [AI, Lemma 2.9; Bll, Proposition 2.11]. We fix the notation f for this homomorphism, and call the values f(bi) the degrees of (A, B). A table algebra (A, B) is called integral iff all structure constants {Jijm and all degrees f (bi) are rational integers. Any finite group G yields two examples of integral table algebras: (Z(CG), Cla(G», the center of the group algebra, with table basis the set of sums C of G-conjugacy classes C, with automorphism - extended linearly from inversion in G, and with degrees f(C) = ICI for all C E Cla(G); and (Ch(G), Irr(G», the ring of complex valued class functions on G, with table basis the set of irreducible characters of G, with automorphism - extended linearly from complex conjugation of characters, and with degrees f (X) = X (1) for all X E Irr( G). Another example is the adjacency algebra, or Bose-Mesner algebra, of a commutative association scheme [BI, Section 11.2]. Here, the table basis consists of the adjacency matrices corresponding to the defining relations of the scheme, the automorphism - is matrix transpose, and the degree of each adjacency matrix is its valency (row sum). There is an explicit theory of substructures, quotient structures and homomorphisms for table algebras [Bll], see also [BI, Section 11.5; AF, Section 2]. The theory is important in the proofs of Theorems 2 and 3 stated below, and of related results, but there is no need to discuss it here. We note only that two table algebras (A, B) and (V, V) (or B and V for short) are called exactly isomorphic iff there is an algebra isomorphism between A and V which restricts to a bijection between Band V. In other words, B and V, under suitable orderings, yield the same structure constants. A number of concepts from finite groups have generalizations to an arbitrary table algebra (A, B). An element b E B is called faithful iff U~I SUPPB(b n ) = B [AI]. If G is a finite group and b = C E Cla(G), then b is faithful iff (C) = G. If b = X E Irr( G), then b is faithful iff X is a faithful character in the usual sense. An element b E B is termed linear iff SupPB(b n ) = {l} for some n > 0 (iff SUPPB(bb) = {I}) [AI]. If b = C E Cla(G) then b is linear iff C S; Z(G) iff ICI = 1. If b = X E Irr(G), then b is linear iff X(l) = 1. There are several useful ways to construct new table algebras from old. One of them is called rescaling [A 1, Bll]: given table algebra (A, B), choose AI, A2, ... , Ak E lR.+, subject only to AI = 1 and AI = Ai for all i. Let B' = {Aibi I bi E B}. It is easily seen that (A, B') is another table algebra, with respect to the same automorphism - and homomorphism f. _
A
Degrees and Diagrams of Integral Table Algebras
21
Another easy construction is symmetrization [A2, Section 3], the prototype for which comes from commutative association schemes [BI, p.57]. For all bi E B, define b? := bi if bi is real, b? = br := bi + bi if not. Let BD := {b? I bi E B} and A- := {a E A I a = a}. Then (A-,B o), the symmetrization of (A, B), is a table algebra with all elements real, and is integral if (A, B) is. Let (A, B) be a table algebra with B = {bl = 1, b2, .. " bd. Fix bE B. Then for 1 :::: i :::: k, k
bbi = Ldjibj , j=1
for unique dji
E
lR.+ U {O}.
Definition. [BI2] (See also [BI, p. 114].) The representation graph of B with respect to b (denoted rh (B) ) is the directed graph with vertex set {I, 2, ... , k} (in bijection with B), and where there is a directed edge from vertex i to j (which is labeled by dji ) iff dji > O. It is clear that rb(B) is connected if and only if b is faithful. If b is real, then dij > 0 {} dji > 0, and so rb(B) may be presented as an undirected, labeled graph, where each pair of adjacent vertices and corresponding edges appears as j
and whenever dii > 0, there is a loop:
;0
(d;.).
Suppose that (A, B) is an integral table algebra. Ifthere exists b E B which is faithful, real and has degree f(b) = 2, then rb(B) is the underlying graph of a generalized Cartan matrix (as in [HPR]), on whose index set {I, 2, ... , k} there is an additive function given by the values of f on B [BI2, Proposition 5.8]. The generalized Cartan matrices with an additive function are classified by Vinberg (as in [HPR]), in terms of their graphs. Twenty such graphs exist, 11 of which represent infinite families, and they are called the generaLized EucLidean (or affine) diagrams. A study of the diagrams leads to the following.
22
Harvey I. Blau
Theorem 1 [BI2, Theorem 2]. Exactly 13 of the 20 generalized Euclidean diagrams occur as representation graphs of integral table algebras with a faithful real basis element o[ deEf.ree~. T~e 7Aagr~s which ~o not occur are (in the notation of [HPRJ) An, AI2, B n , L n , BL n , CD n and G2I. Each of the 13 which do occur determines an integral table algebra to exact isomorphism, with the single exception of Dn, for which there are precisely two exact isomorphism classes of integral table algebras.
Remarks. (1) A slightly broader method of assigning labeled graphs to integral table algebras as in Theorem 1 produces 19 of the 20 diagrams, and thereby generalizes the realization of the affine diagrams from the finite subgroups of SL(2, C), as in [M, SI] (see [BI2, Theorem 3].) (2) The integral table algebras with a faithful nonreal element of degree 2 also are classified, under the additional assumption that there are no nontrivial linear elements of degree 2m , for any m :::: 0 [BI2, Theorem I]. We omit here the specific details of the conclusion of Theorem 1, but two examples may be illuminating.
Example 1. Let G be the dihedral group of order 2(2n
+ 1),
n > O. Let B be the subset of Cla(G) which consists of those class sums C such that C S; (x) = Z2n+I, the~nique subgroup of index 2. Let A = (B), b = x + x-I. Then r b (B) is CL n, where all nonzero dj i = I except as noted, and all degrees are as listed:
~1_(_2_'1_)
:
2
~
;
for G = SL(2, 5). If b is an irreducible character of degree 2, then rb(B) is £8, with all nonzero dji = I and all degrees as listed:
Example 2. Let (A, B)
= (Ch(G), Irr(G»
3
]
b 2
3
4
5
6
4
2
Degrees and Diagrams of Integral Table Algebras
23
Question. What can be said about integral table algebras and their diagrams when all faithful elements have degrees larger than 2? No analog of the Vinberg classification seems known in this much more general context, and the complexity ofthe problem, even at degree 3, increases enormously. In order to gain some insight into possible approaches, we have been studying integral table algebras where all nontrivial basis elements have degree 3. Definition. [BXI] A table algebra (A, B) is called homogeneous (of degree A) iff IBI > I and, for some fixed A E ~+, f(b) = A for all b E B\ {l}. Example I, of course, is homogeneous of degree 2. It is not all that restrictive to assume that an integral table algebra is homogeneous of some particular degree, in view of the following observation. Theorem 2 [BX1]. Any integral table algebra has a rescaling which is also integral, and which is homogeneous of some positive integer degree A. This means that a complete classification of homogeneous integral table algebras is rather a tall order. The degree A which results from rescaling an arbitrary integral table algebra, as in Theorem 2, is usually very large. Example 3. [BXI] Fix A E ~ with A > 1. Let a = (A - 1)/2, f3 = (A + 1)/2. For each integer m ~ 0, we define an integral table algebra (A, Tm(A). Let Tm(A):= {l,xO,XI, ... ,Xm } be a basis for an (m+2)dimensional vector space A over C, and define products of these vectors so that I is the multiplicative identity, and with, for i, j ::: m,
°::
XiX) =
I
aXi+j+{3Xi+J+I A·I + axO + (U m aXi+)-m + {3Xi+)-m-1
if if if
i + j < m, i + j = m, i + j > m.
Then (A, T m (A)) is a table algebra, with Xi = Xm-i for all i, and which is homogeneous of degree A. If A ~ 3 is an odd integer, then (A, Tm(A)) is integral. If m is even, then X m j2 is real and faithful, and the representation graph lXm /2 (Tm (A)) is
24
Harvey I. Blau
(13./1) X
(;\
m
(a,a)
(/1,13)
(0.0) Xm./2_1
Im-l
........~
X m -2
X m /2_2
~'
(A,I) X m /2
(Q 'Q)
x,
I m /2+1
(a,a)
({3,{3)
(a.n)
(13.13) Xo
x,
X m !1+2
-- - - ----
We consider briefly the symmetrization (A -, T m(A)O). Order the basis as
{I, Xm/2, XO + Xm , Xm/2-1 + Xm/2+1, XI + Xm-I, Xm/2-2 + Xm/2+2, ... } . The multiplication of these elements by matrix of structure constants
a
1
A a a 2a a a a a a a
a a a a a {3 {3 a a a a a
where 8 = a or {3, as m
X m /2 =
X~/2 yields the tridiagonal
a a a a a a a a a a a a a a a a a {3 a a {3 a a a a a {3
== a or
8
a
a
A-8
A- 8
8
2 (mod 4). Then the graph rxo (Tm(A)O) m(2
is
(>.,1)
(l3,il)
(2a,a)
A
2A
(a.o) 2A
(3. J)
(a,a)
(J,J) 2,\
2A
(A - J,'\ - J) 2A
(J)
2,\
Thus, (A - , T m (A)O), for m even, is a table algebra of P -polynomial type [BI, p. 317], that is, each element of Tm(A)O is a polynomial in Xm/2.
Degrees and Diagrams of Integral Table Algebras
25
The T m (A)Q do not arise from association schemes, as in [Br, Section IILl], since the sequences of subdiagonal entries and of superdiagonal entries are not monotonic (see [Br, Proposition III.I.2l). The existence ofthese P -polynomial table algebras was also observed directly, and independently, by Arad and Muzychuk. The question of which tridiagonal matrices with given nonnegative integer entries yield integral table algebras of P -polynomial type seems highly nontrivial. Example 4. Let H be an abelian group which admits a fixed-point-free action by Zn (cyclic of order n), for some n > O. Let G be the semi-direct product HZ n , let B consist of all C E Cla( G) such that C ~ H, and set A = (B). Then (A, B) is an integral table algebra which is homogeneous of degree n. (Example I is clearly a special case of Example 4.) Remark. The table algebras Tm(A) for m ~ 2 never arise as (Z(CG), Cla(G)), (Ch(G), Irr(G)), or any substructure thereof, for any finite group G. The equation XQXm-1 = CUm-1 + f3iQ holds in T m (A), whereas there existnoelements b, c in Cla(G) or Irr(G) with c =I- b, b =I- b, and SUPPB(bc) = b} [AI, Corollary E'].
rc,
We recently have determined all integral table algebras (A, B) which are homogeneous of degree 3, and for which B contains a faithful, real element [BX2]. The full result is too detailed to give here, so we restrict to the case where B contains no nontriviallinear elements. First, we need to introduce a few more table algebras. Example 5. Three table algebras, with table bases V2, V3, fined by the following products of nontrivial basis elements. := {I, v}, where v 2 = 6·1
V3
:= {I, VI, V2}, where V[ = 3·1
{I, VI,
V2, V3},
resp., are de-
+ v;
V2
V4:=
V4
+ 2V2; VI V2 = V2VI = 2vI + V2, vi = 3· 1 + vI + V2; where V[ = 3·1 + VI + V2 = v5, vi = 3·1 + 2V2, VI V2 = V2VI = VI + 2V3, VI V3 = V3VI = 2V2 + V3, V2V3 = V3V2 = 2vI + V3.
It is easy to check that V2, V3, V4 are indeed table bases for integral table algebras (with trivial automorphism) which are homogeneous of degree 3.
26
Harvey I. Blau
Theorem 3 [BX2]. Let (A, B) be an integral table algebra which is homogeneous of degree 3, and such that B contains afaithful real element and no nontriviallinear elements. Then B is exactly isomorphic to one of V2, V3, V4 or T m (3) for some even integer m :::: o. Arad, Fisman, Muzychuk and Miloslavsky, in recent work [AFM], have obtained some results on homogeneous integral table algebras of degree 3 which have no nontrivial real elements in the table basis. Many instances of Example 4 are of this type, as is Tm (3) for m odd. The main theorem of [AFM] characterizes the algebras of Example 4 when n = 3 and IH I is odd.
References [AI]
Z. Arad and H. L Blau, On table algebras and applications to finite group theory, J. Algebra 138 (1991),137-185.
[A2]
Z. Arad, H. I. Blau, 1. Erez and E. Fisman, Real table algebras and applications to finite groups of extended Camina-Frobenius type, J. Algebra 168 (1994), 615-647. Z. Arad and E. Fisman, On table algebras, C -algebras and applications to finite group theory, Comm. Algebra 19 (1991), 2955-3009.
[AF]
[AFM] Z. Arad, E. Fisman, V. Miloslavsky and M. Muzychuk, On anti symmetric homogeneous integral table algebras of degree 3, Bar-Ilan University preprint. [BI] E. Bannai and T. Ito, Algebraic Combinatorics I: Association Schemes, Benjaminl Cummings, Menlo Park, 1984. [Bll] [BI2]
H. L Blau, Quotient structures in C -algebras, J. Algebra 175 (1995), 24-64. H. L Blau, Integral table algebras, affine diagrams and the analysis of degree two, 1. Algebra 178 (1995), 872-918.
(BX 1] H. I. Blau and B. Xu, On homogeneous integral algebras, 1. Algebra, to appear. [BX2] H. L Blau and B. Xu, Homogeneous integral table algebras of degree 3, Northern Illinois University preprint. R. Brauer, On pseudo groups, 1. Math. Soc. Japan 20 (1968), 13-22. [Br] [HPR] D. Happel, U. Preiser and C. M. Ringel, Binary polyhedral groups and Euclidean diagrams, Manuscripta Math. 31 (1980),317-329. [K] Y. Kawada, Uber den Dualitatssatz der Charaktere nichtcommutativer Gruppen, Proc. Phys. Math. Soc. Japan (3) 24 (1942), 97-109. [McK] 1. McKay, Graphs, singularities and finite groups, in: The Santa Cruz Conference on Finite Groups (B. Cooperstein, G. Mason, eds.), Proc. Sympos. Pure Math. 37, Amer. Math. Soc., Providence, R.L, 1980, 183-186. [McM] R. McMullen, An algebraic theory of hypergroups, Bull. Austral. Math. Soc. 20 (1979), 35-55.
Degrees and Diagrams of Integral Table Algebras
27
[Sc]
I. Schur, Zur Theorie der einfach transitiven Permutations-gruppen, Sitzungsber. Preuss. Akad. Wiss., Phys.-Math. K1, 1933,598-623.
[SI]
p. Slodowy, Simple Singularities and Simple Algebraic Groups, Lecture Notes in Math. 815, Springer-Verlag, Berlin, 1980.
Department of Mathematical Sciences Northern Illinois University DeKalb, IL 60115 U.SA EmaiJ:
[email protected]
Canonical Induction Formulae and the Defect of a Character R. Bo/tje
Abstract. We explain the idea and the machinery of canonical induction formulae with going as little into details as possible and show that a certain version keeps track of the i-defect d(X) of an irreducible character X of a finite group G by mapping X precisely to the d (X)-th layer in the filtration A ~ i-I A ~ i- 2 A ... , for a prime i and a certain free abelian group A.
1. Introduction and Preliminaries Since its publication in 1947 Brauer's induction theorem (see [Br47]) which says that each character of a finite group is expressible as an integral linear combination of induced one-dimensional characters of elementary subgroups has become a cornerstone both in group theory and number theory. The idea that there are distinguished ways of expressing a character according to Brauer's theorem (at least if one allows all subgroups instead of only elementary subgroups to induce from) originated independently with V. Snaith, cf. [S088], and the author, cf. [B089]. To explain what this means we have to introduce some notation.
1.1.
For a finite group G let R(G) denote the character ring of G and R?:.o(G) the additive monoid of characters of G, i.e. R(G) is the free
abelian group on the set Irr(G) of absolutely irreducible characters of G and R?:.o(G) S; R(G) is the set of all non-negative integral linear combinations of elements in Irr(G). Furthermore, let Rt(G) be the free abelian group on G-conjugacy classes [H, lp]e of the poset M(G) of pairs (H, lp) where H is a subgroup of G and lp is a one-dimensional character of H. The poset structure on M(G) is defined by (K, 0/) ::::: (H, lp), if and only if K .:::: Hand 0/ = res (lp), and it is respected by G -conjugation. The group R~(G) has already been introduced by Deligne in [De]. Using methods from topology, Snaith defined in [S088] a function
f
fe: R?:.o(G) ~ R~(G)
30
R. Boltje
which is natural with respect to restrictions. In fact, for H ::: G there is a restriction map
res+~: R~(G) ---+ R~(H),
[K, 1/I]c
f-+
L
[H n gK,
res~~gK (g1/l)]H,
gEH\C/K
where the sum runs over representatives g E G of the double cosets H\ G / K , gK := gKg- 1 and g1/l(gxg- 1) := 1/I(x) for x E K. This restriction map has been defined by Deligne (cf. [De]) and can also be naturally interpreted by viewing R~ (G) as the Grothendieck group of a suitable category of monomial G-modules, cf. [B095, Chapter 5]. Snaith's map lC is a canonical induction formula, where 'canonical' means that lc commutes with restrictions, and 'induction formula' means that b c 0 lC = idR(C), where bc is defined as bc: R~(G) ---+ R(G),
[H,
f-+
ind~(
and also commutes with restrictions. Due to the topological nature of its construction, the map lc is not additive but has a sort of derivation property with respect to addition, cf. [BSS, Theorem 2.16(c)], so that it is not clear how to extend it from R2:o(G) to R(G). Independently of Snaith the author constructed an additive map QC: R(G)
---+ R~(G)
which also is a canonical induction formula, cf. [B090]. This was done by purely algebraic methods. In general, the maps Qc and lc are different, even on Irr(G). But there is also an algebraic description of lc which allows to compare both maps, cf. [BSS]. Moreover, there is a geometric construction of the map QC due to P. Symonds (cf. [BSS, (2.26)] and [Sy]). Here we will only use the canonical induction formula Qc.
1.2.
Without going into details we should try to make clear that the group Rf(G) is a very natural construction. Note that the groups R(H), H ::: G, are tied together by homomorphisms Cg,H: R(H) ---+ R( gH),
res~: R(H) ---+ R(K), ind~: R(K) ---+ R(H), for K ::: H ::: G, g E G, called conjugation, restriction and induction. These homomorphisms satisfy certain compatibility conditions which are the axioms of a Mackey functor on G. For a precise definition see for example
Canonical Induction Formulae and the Defect of a Character
31
[TW]. Let furthermore, Rab(G) S; R(G) denote the span of the set G of all one-dimensional characters. Then the family of groups R ab (H), H ::: G, is stable under conjugation and restriction maps. We will call such a structure a restriction junctor on G. Obviously, there is a forgetful functor from the category Mack(G) of Mackey functors on G to the category Res(G) of restriction functors on G. This functor has a left adjoint functor
-+: Res(G)
~
Mack(G),
which associates to a restriction functor A(H), H ::: G, the family of groups
A+(H) := coinvH(
EB A(K)),
H::: G,
(1.2.1)
K~H
where H acts on EBK
~
Con(G)(A, B),
(1.2.2)
where A and B are restriction functors on G over Q and Con(G) is the category of conjugation junctors, requiring only conjugation maps and no restriction or induction maps. In the case A(H) = Q @ R(H), H ::: G, and B(H) = Q 0 Rab(H), H ::: G, we are left with finding suitable conjugation respecting families of maps (1.2.3) The isomorphism in (1.2.2) is very explicit. For a given morphism of conjugation functors p: Q @ R ~ Q @ R ab as in (1.2.3), the associated morphism of restriction functors
aP:Q@ R ~ Q@Rf is given on the level of a subgroup H of G by
a~(x):= I~I
L
(-l)nIHoi[Ho,resZ~(PHn(reSZn(x)))JH
Ho< .. ·
(1.2.4)
32
R. Boltje
for X E Q 0 R(H), where the sum runs over all chains of subgroups of H. Note that resZ~ (PHn (resZn (X») is an element of Q 0 Rab(Ho) and that for such an element e = Lipdio aipcp E Q 0 Rab(Ho), where aip are rational numbers, we define [Ho, e]H := LipEHO aip[Ho, cp]H. 1.3. For given P as in (1.2.3) there are two natural questions to ask about the associated morphism a P : Q 0 R -+ Q 0 Rf: (i) Is a P an induction formula, i.e. is bH 0 a~ = idiQI0R(H) for all H:::: G? (ii) Is a P integral, i.e. is a~(R(H» ~ Rf(H) for all H :::: G? The best answers we know so far are the following:
1.4. Theorem (cf. [B095, Corollary II.2.4]). Let C be the set of cyclic subgroups of a finite group G and let P E Con(Q 0 R, Q 0 R ab ). If PH = idiQI0R(H) for H E C, then b 0 a P = idiQI0R. Furthermore, for H E C and cP E iI we have a~(cp) = [H, cp]H. Actually Corollary II.2.4 in [B095] is more general. It deals with an arbitrary Mackey functor M on G over Q replacing Q ® R and with a sub-restriction functor A ~ M on G over Q replacing Q 0 R ab such that there is a set C of subgroups of G with A(H) = M(H) and PH = idM(H) for H E C and such that for all H :::: G the elements of M(H) are uniquely determined by their restriction to subgroups K :::: H, K E C. Then the associated morphism a P is a section of b: A+ -+ M and aH(m) is the image of m E M(H) in A+(H) for H E C, where we use the definition of A+(H) in (1.2.1).
1.5. Theorem (cf. [B095, Theorem II.4.S]). Let p: R -+ R ab be amorphism of conjugation functors on G and JT a set of primes such that the following condition holds: (*rr) For each pair of subgroups V :::: V :::: H such that V is normal in V and V/V is a cyclic JT -group, andfor each one-dimensional character ljJ E V which is stable under V -conjugation, the coefficients of ljJ in the elements (res~ opu )(e) and (pv 0 res~)(e) of Rab(V), coincide for e E Irr(V).
Canonical Induction Formulae and the Defect of a Character
Then,for H .:::: G and X be transformed into
E
33
R(H), the explicitformula (1.2.4)for a~(x) can
a~(x) =
where the sum runs over all chains a taken up to H -conjugation in the poset M(H), NH(a) denotes the stabilizer of a in H, (NH(a)j HO)rr the set of Jr-elementsof NH(a)jHo, andforanysubgroup K.:::: H and A E R(K) the expansion PH(A) =
L
m'/J(A)1/I
E
Rab(K)
'/JEf< defines the multiplicities m'/J (A) E Z, 1/1 E
K.
Like Theorem 1.4 also Theorem 1.5 has a more general version (cf. [B095, Theorem 11.4.5]), where Rand R ab are replaced by a Mackey functor M on G and a restriction subfunctor A of M on G such that A has a conjugationstable basis which replaces H, for H .:::: G. Since jGrrl/lGlrr is an integerfor any finite group G (cf. [Hu, V.19.14]), we obtain 1.6. Corollary (cf. [B095, Corollary 11.4.7]). Let p: R -+ R ab be amorphism of conjugation functors on G and let Jr be a set of primes such that condition (*7T) in Theorem 1.5 is satisfied, then IHI7Tla~(x) E Rf:(H) for all H .:::: G and X E R(H). In particular, if Jr is the set of all primes, a P is integral.
2. The Canonical Brauer Induction Formula 2.1. Keeping the notation of the previous section we define amorphism p: R -+ R ab of conjugation functors on G for H :s: G as being induced by taking H' -fixed points of an H -module, thus obtaining an H ab -module, where H' denotes the commutator subgroup and H ab the commutator factor group of H. Note that Rab(H) and R(H ab ) can be naturally identified. In other words, PH is defined as Irr(H)
3X {Xl 0, 1-+
if X E H, otherwise.
(2.1.1)
34
R. Boltje
It is easy to see that the hypotheses of Theorem 1.4 and Theorem 1.5 (cf. the proof of Proposition 4.4) with Jr the set of all primes are satisfied so that the resulting morphism aP: Q 181 R ---+ Q ® Rf of restriction functors gives rise to an integral canonical induction formula aG: R(G) ---+ Rf(G)
which we will call the canonical Brauer induction formula. This map is precisely the one defined in [B090]. 2.2. The advantage of having a canonical Brauer induction formula rather than any expression of a character according to Brauer's induction theorem is best explained at the example of the Adams operations
\lJ~: R(G) ---+ R(G),
X t-+ (g t-+ x(l))
for k E Z, X E R(G), g E G. Note that it is obvious that \lJ~(X) is again a class function. But one has to work a little bit to see that \lJ~ (X) is again an element in R(G). Note also that this is immediate for one-dimensional characters, since \lJ~ (cp) = cpk for cp E G. Finally note that the family \lJ~, H ::: G, is a morphism of restriction functors, i.e. commutes with conjugations and restrictions. Now let
L
aG(X) =
O'[H,cpJG(X)[H, cp]G
(2.2.1)
[H,cpJG
be the expansion of the canonical Brauerinduction formula for X E R(G), thus defining O'[H,cplG (X) as the integral coefficient of the basis element [H, cp]G of Rf(G) in aG(X). Since bG 0 aG = idR(G), we obtain the expression X=
L
O'[H,cpJG(X)ind~(cp)
(2.2.2)
[H,cp lG
for X E R( G) in a way ensured by Brauer's Theorem (except for the restriction that H is elementary). By applying \lJ~ to Equation (2.2.2) we obtain
\lJ~(X) =
L
O'[H,CPlG(X)\lJ~(ind~(cp)).
(2.2.3)
[H,cplG
Now the pity is that Adams operations do not commute with induction, unless k is coprime to the exponent of G. But just for fun we could define an element (2.2.4)
Canonical Induction Formulae and the Defect of a Character
35
in R(G), thus having defined a homomorphism \iI~: R(G) --+ R(G). It turns out that, since the maps a H, H :s G, commute with restrictions, also the maps \iI~, H:s G, commute with restrictions. Furthermore, by Theorem 1.4, one has \iI~ = 'IJ~ for all cyclic subgroups H of G. Hence, if we only assume that 'IJ~ takes values in the group of class functions on G, \iI~ - 'IJ~ is a map from R (G) to the group of class function on G, commuting with restrictions to subgroups and being trivial on cyclic subgroups. Therefore, if X E R(G) and g E G are given, we have -k
k
('lJ G - 'lJG)(X)(g)
-k = ('IJ
k
G
- 'IJ (X)(g»
= 0,
showing that \iI~ = 'IJ~. In particular, 'IJ~ (X) is not only a class function on G but an element of R(G). This means that in the sums in Equations (2.2.3) and (2.2.4) the corresponding summands for themselves may be different, but in the whole sum all the differences cancel due to the naturality of aG. 2.3. The above example shows that the canonical Brauer induction formula can be used to extend maps defined on Rab(G) to maps on R(G). This concept can be used in many situations, yielding a character theoretic definition of the determinant and the Chern class of a character. Moreover, extensions of characters and a correspondence related to the Glauberman correspondence can be defined that way, cf. [B095, Chapter IV]. Also in number theory this method was used to define local root numbers (cf. [B089, Chapter Ill) and conductors of Galois representations (cf. [B089, Chapter 11] and [BeS)). Canonical induction formulae also exist for other representation rings (see [B095, Chapter Ill)), as for example for the Brauer character ring, the trivial source ring, and the linear source ring. Some of the above constructions make sense for one-dimensional representations of those rings, so that we may for example define Chern classes of trivial source modules and other exotic maps whose existence was not at all clear before. 2.4. There are many surprising properties of the map aG. Equation (2.2.2) implies that
L
For example,
a[H,cp]G(X)[G: H] = x(l),
[H,cp]G
but much simpler the degree of X cients a[H.cp]G (X) as
L [H,cp]G
E
R(G) can be extracted from the coeffi-
a[H,cp]G(X) = x(l)·
36
R. Boltje
This follows immediately from Equation (2.2.4) and the equality of \iI~ and 41~ for k = 0 by applying the inner product of R(G) with the trivial character of G on both sides of Equation (2.2.4). For a list of more properties of aG see [B090, Theorem 2.1] and [B095, Proposition III.1.2]. For further applications of canonical induction formulae see [Sn94].
3. Variations of the Canonical Brauer Induction Formula 3.1. In view of Theorem 1.4 we realize that we still obtain a canonical induction formula by defining PH = 0 for non-cyclic subgroups H ::s G and PH = id Q0R (H) for H ::s G cyclic. According to (1.2.4) the corresponding canonical induction formula a P has the explicit description I
P (X) = aH
JHJ
~
(-l)nIHol[Ho,res~ (X)]H
L...J
(3.1.1)
"0
Ho<···
for H ::s G and X E R(H), where the sum runs over all chains of cyclic subgroups of H and [Ho. resZo(X)]H has to be interpreted as explained below (1.2.4). But this formula is not integral. In fact, the smallest nonnegative integer e such that e . a~ (X) E R~ (G) for all X E R (G) turns out to be the Actin exponent of G (see [B090, Theorem 3.17] for a proof of this statement and [CR, § 76] for a definition of the Actin exponent). The resulting expression after applying bG, X=
I
IGI
L
(-l)nJHol ind~o(res~o(X»'
(3.1.2)
Ho< .. ·
for X E R (G) can be transformed by using the Mobius function IJ-(K, H) of the poset of cyclic subgroups of G (see [Ro] for generalities about the Mobius functionofaposet)andtakingintoaccountthat IJ-(K, H) = IJ-(IH/KI), where the latter IJ- denotes the usual number theoretic Mobius function, into X= -
I
IGI
L
. IJ-(IH/KI)IKI md~(res~(x»
(3.1.3)
K
which is precisely the explicit version found by Brauer (cf. [Br51]) of Actin's induction theorem stating that each character can be expressed as a rational linear combination of characters induced from cyclic subgroups.
Canonical Induction Formulae and the Defect of a Character
37
3.2. In view ofBrauer's induction theorem one might think that the morphism p: Q (2J R -+ Q <>9 R ab of conjugation functors, given by PH = for H :s G not elementary, and PH as in (2.1.1) for H :s G elementary, results in an integral canonical induction formula a P . In fact a P is an induction formula, since the hypothesis of Theorem 1.4 is satisfied, but in general a P is not integral. Also the condition ( *rr ) in Theorem 1.5 is not satisfied for the set n of all primes, if V :s U is chosen such that V is elementary and U is not, and if 1/J is the trivial character of V. More generally, let J-C be a set of subgroups of G closed under taking subgroups and conjugates and containing the set of cyclic subgroups of G. Furthermore, let PH = for H rt J-C and let PH be defined as in (2.1.1) for H E J-C. Then the resulting morphism a P is a canonical induction formula by Theorem 1.4, and we know (see [B090, Corollary 3.13] for a proof) precisely in which cases a P is integral, namely if and only if for each pair of subgroups K :s H of G such that K is normal in H, H / K is cyclic, and K lies in J-C, also H lies in J-C. The smallest such set J-C is the set of solvable subgroups of G. This fact about the integrality of a P is related to the fact that the idempotent of the Burnside algebra over Q associated with J-C is integral if and only if the above condition on J-C is satisfied.
°
°
4. Projectification 4.1. Pushing the idea of (3.1) further, we choose a prime [ (the letter P being reserved for maps as in (1.2.3)) and consider the morphism of conjugation functors p: R -+ R ab given by PH = 0, if IH I is divisible by [, and by PH as in (2.1.1), if H is an [' -subgroup of G. By the isomorphism in (1.2.2) we obtain a corresponding morphism at':
Q (2J R
-+
Q (2J R~
of restriction functors on G, which is explicitly given by
a~(x)= I~I
L
(-l)nIHol[Ho,resZ~(PHn(reSZn(x)))JH
(4.1.1)
Ho<···
for H :s G and X E R(H), cf. (1.2.4), where the sum runs over all chains of [' -subgroups of H. It is clear that in general bG 0 a~ cannot be the identity, since for an [-group G and X E R(G) we have only one such chain, and we obtain that bG 0 a~(x) = (X (l)/IGI) . ind? (l) is always a rational
38
R. Boltje
multiple of the regular character of G. We are going to describe the map be 0 a~: Q 0 R(G) ---+ Q 0 R(G) in a different way.
4.2. Let e~ be the characteristic function on G of the subset of I' -elements of G. Obviously e~ is an idempotent class function on G. But it is even an element of Q 0 R(G), since using the expression (3.1.3) for X = 1, multiplying by e~, and using the obvious property res~(e~) = e~ for H ~ G, we are reduced to the case where G is cyclic, and in this case we have e~ = (l/IGI[)· indgl,(l), where Gl' is the I'-partof G. 4.3. Proposition. For a finite group G the map be 0 a~: Q 0 R(G) ---+ Q 0 R(G) is multiplication by e~. In particular, if X E R(G) vanishes on all I-singular elements, then be (a~ (X)) = X.
Proof Let X
E R(G).
We have to show that ['
(be
0
['
ae)(x) = ee . X.
(4.3.1)
By Equation (4.1.1) both (be 0 a~) (X) and e~ . X vanish on I-singular elements. Hence, it suffices to show that
( res e H (be
0
[' ) = res eH ([' ae)(X) ee . X )
(4.3.2)
for all cyclic I' -subgroups H of G. However, since be and a~ commute with restriction and since restriction is multiplicative and maps e~ to e~, Equation (4.3.2) is equivalent to the equation (bH
0
a~)(res~(x)) = e~ . res~(x)
so that we may assume in Equation (4.3.1) that G is a cyclic I' -group. In this case, the right hand side in Equation (4.3.1) equals X and by Equation (4.1.1) the left hand side is equal to IGI
L
(-l)nIHol
ind~o(res~o(X))'
Ho<···
where the sum runs over all chains of subgroups of G. But this sum is equal to X by Equation (3.1.2). Alternatively, we could use l' ae(cp) = [G, cp]e
for cp E
G from Theorem 1,4 to see that
(be
0
a~)(x) = X.
o
Canonical Induction Formulae and the Defect of a Character
39
4.4. Proposition. Forafinite group G and X E R(G) we have IGlt·a~(x) E R~(G). Proof. We will verify the hypothesis of Theorem 1.5 for the set n of all primes different from i. So let V ~ U ~ G be such that V is normal in U and U/ V is a cyclic if -group. Furthermore, let 1/f E V be U -stable and let e E Irr(U). We have to show that the inner product
(1/f, res~ (pu (e»
- PV (res~ (e»)
vanishes. If V is not an if -group, then U isn't either. Hence, Pu and pv are trivial and the above inner product vanishes. So let V be an if -group. Then also U is an if -group, and since U stabilizes 1/f, the cyclic group U / V acts trivially on the cyclic group V / ker 1/f. Hence U / ker 1/f is abelian, and 1/f can be extended to U. If e is one-dimensional, then
res~(pv(e» = res~(e) = pv(res~(e», and we are done. If e is not one-dimensional, then res~(pu(e» = 0, and we have to show that 1/f does not occur as a constituent in pv(res~ (e». However, assuming that 1/f does occur implies that 1/f also occurs in res~ (e). By Frobenius reciprocity we deduce that e occurs in ind~ (1/f), which is the sum of all extensions of 1/f to U. All of them being one-dimensional gives 0 the desired contradiction. Applying bG to the assertion in the above Proposition we obtain the following presumably well-known Corollary.
4.5. Corollary. Let X E R(G), then e~ . X E (l/IGlt) . R(G). In particular
e~
E (l/IGlt) . R(G).
4.6. Remark. Let L be a finite field extension of the field QI of l-adic numbers containig all IG I-th roots of unity, and let <9 be the valuation ring of L. We consider the cyclotomic field of the IG I-th roots of unity over Q both contained in L and in C. The functor L 0(') -: <9G - mod ~ LG - mod induces an isomorphism between the Grothendieck group of finitely generated projective <9G-modules and the subgroup peG) of R(G) consisting of all virtual characters vanishing on i-singular elements (cf. [Se, Theoreme 37]). The groups P(H), H ~ G, form a Mackey subfunctor of R on G. For an if -subgroup H of G we set pab(H) := Rab(H), and for H ~ G with
40
R. Boltje
pab(H) = O. Then pab is a restriction subfunctor of R ab and of P. The morphism p: R -+ R ab from (4.1) can be viewed as amorphism l
I IHI we set
p: R -+ pab
of conjugation functors on G. It is easy to see that P+b(H) is naturally contained in R~(H) for H ::: G, and that ai' can be considered as a morphism (4.6.1) of restriction functors on G.
4.7. Theorem. We keep the notation of Remark 4.6. The restriction of ai' in (4.6.1) to pab is an integral canonical induction formula al': p -+ p+b. Inparticular,
a~(x)
E P+b(H)
~ R~(H) for X
E P(H),
H::: G.
Proof. It follows from Proposition 4.3 that (bHoa~)(x) = X for X E P(H),
H ::: G. We only have to show the integrality part of the statement. Since we have already verified that condition (*rr ) in Theorem 1.5 holds for the morphism p from 4.1 and the set of all primes different from l, we obtain by Theorem 1.5 an explicit formula for a~(x), H ::: G, X E R(H), and it suffices to prove that I(N H (er)/Ho)1' I H INH (er) / Ho I mrpll (res HIl (X))
is an integer for any H ::: G, any chain a = ((Ho, <po) < ... < (Hn ,
Hence, we have to show that INH (er )1,1 . mrpn (resZIl (X)) INH(a)ll'·INH(er)ll
is an integer. It follows from [Hu, Y.19.141 that INH(a)I'I/INH(er)ll' is an integer, and we will show that IN H (er)ll divides mrpn (resZn(X))·
Canonical Induction Formulae and the Defect of a Character
41
Note that 'Pn is the character of a projective CJ Hn -module CJcpn of rank one, and mcpn (resZn(X» is just the number of summands isomorphic to CJcpn in a direct sum decomposition of resZn (M) into indecomposable CJHn -lattices. Let Q be a Sylow I-subgroup of NH(a), H' := QHn (note that Q normalizes a and hence Hn ), and let N be an indecomposable summand of resZ,(M). It suffices to show that IQI divides the multiplicity mcpn(resZ~(e», where is the character of N, and we may assume that this multiplicity is not zero. Since N is a projective CJ H' -module, N is a direct summand of indf'(CJ) ~ indZ~(indfn(CJ». Since H'/Hn ~ Q is an I-group, N is iso-
e
morphic to indZ~ (L) for some indecomposable CJHn-lattice L by Green's indecomposability theorem. Mackey's decomposition formula then yields res H' Hn (N) '" =
CD . d Hn , ('Hn (SL») '" '\J7 In = CD '\J7 Hnn Hn res Hnn sHn sEH'/ HI! sEH'/ HI!
Since CJtn't'1l is a summand of res H ' (N), and since 'Pn is stable under H' < Hn NH(a), we have CJCPI! ~ L, and
hence mcpl!(resZ~(N» = IH'/HIlI =
IQI asrequired.
o
5. Projectification and the Defect of an Irreducible Character The last theorem showed that
ah
ah (X) is integral for X E P (G) and we already
know that (X) is not integral in general. Since we have seen that the worst denominator in the canonical induction formula of (3.1) is the Artin exponent, we may expectthatthe denominators occurring in ah(x) for X E Irr(G) have some significance for X. We already know by Proposition 4.4 that we only have to consider I-power denominators. First we answer the question, what I-power factor is necessary to make (be 0 ah)(x) = e~ . X integral for X E Irr(G). At least the same I-power is necessary to make ah(x) integral, since bcCR~(G» S; R(G). I'm grateful to Burkhard Kiilshammer who pointed out to me that the answer is given in a paper of Brauer and Feit.
42
R. Boltje
5.1. Theorem (cf. [BFD. For X E Irr(G) we have ld(x) . e~ . X E R(G)
ld(x)-I. e~ . X r:t R(G),
and
where d(X) is the l-defect of x, i.e. ld(x) . x(l)[ = IGI[. In particular, for n E No we have In . e~ E R(G)
if and only if In ::::
IGI[ (cf Corollary 4.5).
Proof This theorem is not stated explicitly in [BF]. There, the integers .
aij.= l
d
['
. (Xi,ee' Xj),
1:::::
i,j::::: m,
are studied, where XI, ... ,Xm are the irreducible characters in an l-block of G of defect d. It is easy to see that for X E Irr( G) the irreducible constituents of e~. X lie in the same l-block as X. Hence, for j E {I, ... , m} and k :::: 0, we have: [I l d-k . ee' Xj E R (G) {::::::::} lk
I aij f or i
= 1, ... , m.
(5.1.1)
It follows from (8) and the congruence preceeding (7) in [BF] that the l-part of gcd(aIj, '" ,amj)
is given by the l-height h (Xj) of Xj, which is defined by d (Xj ) + h (Xj) = d. Using (5.1.1) this proves the theorem. 0 The next theorem shows that the map a~ is strict in the sense that it doesn't need higher denominators than the least possible ones. 5.2. Theorem. For X
E
Irr(G) we have
ld(x) . a~(x) E R~(G)
and
ld(x)-I. a~(x) r:t R~(G).
Proof First note that be is injective on a~(R(G)). In fact, if X E R(G)
with be(a~(x)) = 0, then X vanishes on l'-elements by Proposition 4.3. By the explicit formula (4.1. I) for a~ (X) this implies a~ (X) = 0. Since e~ is an idempotent, we have be 0 a~ 0 be 0 a~ = be 0 a~ by Proposition 4.3, and since be is injective on a~(R(G)), it follows that
Now let X E Irr(G). For n E No we have
[' (In . ee [' . X ) = In . ae [' (X ) ae
Canonical Induction Formulae and the Defect of a Character
43
by the above equation applied to Zn . X. Hence, if zn . e~ . X is integral, so is Zn . a~ (X) by Theorem 4.7. Conversely, if Zn . a~ (X) is integral, so is bG a~ (X)) = zn . e~ . X. Therefore, Theorem 5.1 yields the result. 0
un .
References [Bo89] R. Boltje, Canonical and explicit Brauer induction in the character ring of a
finite group and a generalization for Mackey functors, Thesis, University of Augsburg 1989. [Bo90] R. Boltje, A canonical Brauer induction fonnula, Asterisque 181/182 (1990), 31-59. [Bo95] R. Boltje, Mackey functors in representation theory of finite groups and number
theory, Habilitation Thesis, University of Augsburg 1995. [BCS] R. Boltje, G.-M. Cram, V. Snaith, Conductors in the non-separable residue field case, in: Algebraic K-Theory and Algebraic Topology, P. G. Goerrs and 1. F. Jardine (editors), Kluwer 1993, 1-34. [BSS]
R. Boltje, V. Snaith, P. Symonds, Algebraicisation of explicit Brauer induction, 1. Algebra 148 (1992), 504-527.
[Br47] R. Brauer, On Artin's L-series with general group characters, Ann. of Math. 48 (1947),502-514. [Br51] R. Brauer, Beziehungen zwischen Klassenzahlen von Teilkorpern eines Galoisschen Korpers, Math. Nachr. 4 (1951),158-174. [BF]
R. Brauer, W. Feit, On the number of irreducible characters of finite groups in a given block, Proc. Nat. Acad. Sci. U.S.A. 45 (1959), 361-365.
[CR]
C. Curtis, I. Reiner, Methods of representation theory, Vol. 2, 1. Wiley & Sons, New York 1987.
[De]
P. Deligne, Les constantes des equations fonctionelles des fonctions L, Springer Lecture Notes 349 (1974), 501-597.
[Hu]
B. Huppert, Endliche Gruppen I, Springer-Verlag 1967.
[Ro]
G. C. Rota, On the foundations of combinatorial theory, I, Theory of Mobius functions, Z. Wahrsch. verw. Gebiete 2 (1964),340-368.
[Se]
J.-P. Serre, Representation lineaires des groupes finis, 3eme ed., Hennann, Paris 1978.
[Sn88] V. Snaith, Explicit Brauer induction, Invent. Math. 94 (1988), 455-478. [Sn94] V. Snaith, Explicit Brauer induction, Cambridge University Press, Cambridge 1994. [Sy]
P. Symonds, A splitting principle for group representations, Comment. Math. Helv. 66 (1991),169-184.
44 [TW]
R. Boltje J. Thevenaz, P. Webb, The structure of Mackey functors, Trans. Amer. Math. Soc. 347 (1995),1865-1963.
Universitat Augsburg Institut fUr Mathematik 86135 Augsburg Germany Email: [email protected]
Counting Characters in Blocks, 2.9
1
E. C. Dade
In [D2] and [D3] we discussed two forms of a conjecture about the number of irreducible complex characters of a given defect in a given block B of a finite group G. In every case so far considered this conjecture has turned out to be true. There is even some hope of proving it in general. A very complicated form ofthis conjecture, which we shall call the "inductive conjecture," will hold for for all finite groups G if it holds whenever G is simple and non-abelian. So all we have to do is verify this inductive conjecture by direct calculation for each finite simple group. Such a verification may not be beyond the capabilities of present day representation theory. Before we can verify the inductive conjecture for a given group G, we first have to specify that conjecture completely. We're going to sketch that specification in this note, along with that of several other weaker forms of the conjecture. We shall also try to explain why such a complicated form is needed to carry out the inductive proof reducing the conjecture to finite simple groups. The full details of the theory required to understand this form will be the subject of the next paper [D4] in the series we are devoting to the conjecture. So the present note is, as its curious title might suggest, a preliminary announcement of that paper, designed to present the inductive conjecture and its background without getting bogged down in details. It also provides enough information so that anyone seeking to verify that conjecture for a given finite simple group G will know exactly what must be done. Since this note is written for group theorists, we describe projective representations here in terms of group extensions instead of the twisted group algebras used in [D3]. This formulation is not really suitable for explaining the theory behind the subject, and will not be used in [D4]. However, it is much more practical for actual computations using character tables such as those in [Atlas]. 1 Some of this research was supported by NSF Grant DMS 93-02996.
46
E.C.Dade
1. The Basic Conjecture Even describing the basic conjecture [D2, 6.3] requires a considerable amount of notation. We fix a prime p and a local principal ideal domain (Le., a real discrete valuation ring) !R with maximal ideal p, such that the field of fractions J of !R has characteristic zero and the residue class field ~ = !RIp has characteristic p. Let G be a finite group which is totally split over J, in the sense that J is a splitting field for every subgroup H of G. We denote by Irr(H) the set of all irreducible J-characters of any such H. Since J is a splitting field of characteristic zero for H, the degree 4J(l) of any 4J E Irr(H) divides the order IHI of H. The defect d(4J) of 4J is the largest non-negative integer d such that pd divides the integer IHI/4J(l). We know from [D2, §4] that all the usual results about p-blocks of subgroups H of G hold in our present situation. In particular, the p-blocks of any such H correspond one to one to the primitive idempotents of the center Z (!RH) of the group algebra !RH of Hover !R, and also to the epimorphisms of Z(!RH) onto ~ as !R-algebras. We denote by Blk(H) the set of all p-blocks of H, and by d(b) the usual defect of any given b E Blk(H). Any character 4J E Irr(H) lies in a unique p-block of H, a p-block which we denote by B(4J). A p-chain C of G is any strictly increasing chain
C: Po < PI < ... < Pn
(1.1)
of p-subgroups Pi of G. The length ICI of C is the number n :::: 0 ofinclusions in this chain. If K is any group acting (exponentially) as automorphisms of G, then any r E K sends the p-chain C to the p-chain (1.2)
of G. The normalizer of C in K is the subgroup N K (C) of all r E K such that C T = C. In particular, the normalizer N G (C) of C in G is defined in this way using the conjugation action of G on itself. We say that the p-chain C in (1.1) is radical (with respect to G) if its initial subgroup Po is the largest normal p-subgroup Op(G) of G, while its i th subgroup Pi is the similar subgroup Gp (NG (Ci)) of the normalizer of its i th initial subchain Ci: Po < Pj < ... < Pi
for each i = 1,2, ... , n (see [D2, §2]). We denote by ~(G) the family of all radical p-chains of G. The family ~(G) is closed under the conjugation
Counting Characters in Blocks, 2.9
47
action (1.2) of G on its p-chains. We write 'R( G) / G to indicate an arbitrary family of representatives for the G-conjugacy classes in 'R(G). Fix a p-block B of G and an integer d :::: o. Definition 1.3. For any p-chain C of G we denote by k( C, B, d) the number of characters c/J E Irr(Nc(C)) having defect d(c/J) equal to d and belonging to a p-block B(c/J) of Nc(C) such that the p-block B(c/J)c of G induced by B(c/J) (in the sense of Brauer) is equal to B. We should remark that the induced block B(c/J)c in the above definition is always defined by a lemma of Knorr and Robinson [KR, 3.2]. It is easy to see that the non-negative integer k( C, B, d) depends only on the G -conjugacy class ofthe p-chain C. So the sum in the following conjecture is well defined. The Ordinary Conjecture 1.4. If Op(G) = 1 and d(B) > 0, then
L
(_l)IClk(C.B,d) =0.
(1.5)
CE'R(Cl/C
Of course this conjecture is the same as [D2, 6.3]. So it implies Alperin's Weight Conjecture by [D2, 8.3]. Furthermore [D2, 7.1] and [D2, 7.3] show that this conjecture can be false when either of its hypotheses Op(G) = 1 or d(B) > 0 fails to hold.
2. The Invariant Conjecture When you try to reduce the above conjecture to the case where G is simple, you quickly discover weaknesses in its formulation. The conjecture as stated says nothing about what happens to the characters it is counting when G is embedded as a normal subgroup in some larger finite group E. But it is exactly this information which is needed to derive the conjecture for E from that for G. So we have to strengthen the conjecture in order to carry out the reduction to the simple case. We begin by fixing an epimorphism 8: E --# E of finite groups with kernel G. So we have an exact sequence
E:-
I---+G..:::+E---+E---+I
(2.1)
of finite groups and their homomorphisms. We shall assume that E is totally split over J. The conjugation action of E on its normal subgroup G induces an action (1.2) of E on the p-chains C of G. So any such C has a normalizer
48
E.C.Dade
NE (C) in E. The image of this normalizer under the epimorphism s in (2.1) is a subgroup NE(C) = S(NE(C)) which we may call the "normalizer" of C in E. Since the kernel of the epimorphism s: NE(C) - N E(C) is the normal subgroup NG(C) of NE(C), we have an exact sequence <1
E:
1 -+ NG(C) ~ NE(C) -+ N E(C) -+ 1
(2.2)
of finite groups associated with each p-chain C of G. The above group NE (C) acts by conjugation on the set Irr( N G (C)) of all irreducible J' -characters ef> of its normal subgroup N G (C). We denote by NE (C, ef» the stabilizerin NE (C) of any such ef>, and by N E(C, ef» = s(NECC, ef») the image of that stabilizer in E. Since N G (C) is contained in NE (C, ef», the exact sequence (2.2) restricts to an exact sequence
1 -+ NG(C)
<1
~
E:
NE(C, ef» -+ NE(C, ef» -+ 1
(2.3)
associated with any p -chain C of G and any character ef> E Irr (NG (C) ) . In addition to the p-block B of G and the integer d :: 0, we now fix a subgroup F of E. Definition 2.4. For any p-chain C of G we denote by k(C, B, d, number of characters ef> E Irr(NG (C)) satisfying d(ef» = d,
B(ef»
G
= Band
N E(C, ef»
F) the
= F.
It is straightforward to verify that the integer k( C, B, d, F) depends only on the G-conjugacy class of the p-chain C. So the sum in the following conjecture is well defined. The Invariant Conjecture 2.5. If Op(G) = 1 and d(B) > 0, then '" L
(-1) ICI k(C,B,d,F)=O.
(2.6)
CE':R(Gl/G
The above invariant conjecture is the equivalent in our situation of the form of Alperin's Weight Conjecture considered by Robinson and Staszewski in [RS]. It reduces to the Ordinary Conjecture 1.4 when E = G. Furthermore, it implies that conjecture in any case, since the equation (1.5) can be obtained by summing the equation (2.6) over all subgroups F of E.
Counting Characters in Blocks, 2.9
49
3. The Extended Conjecture The Invariant Conjecture 2.5 is reasonably easy to verify or refute for any given group G. But it is not strong enough for an inductive proof. To carry out such a proof we need to calculate, for each p-chain C of G, the number of irreducible J-characters 1/1 of NE(C) with a given defect lying over a fixed irreducible J-character q; of Nc(C). Except in rare cases (which, in fact, are not so rare for simple G), the subgroup N E( C, q;) of E by itself does not determine this number. However, the Clifford extension for q; does determine it. So we reformulate the conjecture in terms of Clifford extensions. We start from the exact sequence (2.2) for a p-chain C of G. To each q; E Irr(Nc(C)) is associated a central extension E[C,q;,J] of the unit group U (J) of the field J by the stabilizer NE (C, q;) of q; in NE (C). Thus we have a new exact sequence
1 -+ U(J)
~
E[C, q;, J]
ll[C,t/J,;JJ
) N E(C, q;) -+ 1
(3.1)
of groups such that U (J) is a central subgroup of E[C, q;, J]. The exact definition of this ClifJord extension E[ C, q;, J], such as in [03, §11], need not bother us here. The important thing is how we can use it to count characters. We denote by Irr( E[C, q;, J]) the set of all irreducible projective J-characters of the extension E[C, q;, J], i.e., of all irreducible J-characters of E [C, q;, J] lying over the natural faithful linear J -character of U (J) given by inclusion in J. Then Clifford theory gives us a bijection of Irr( E[C, q;, J]) onto the set Irr( NE (C) Iq;) of all irreducible J -characters of NE (C) lying over q; (see [03,12.12]). If 1/1' E Irr(E[C,q;,J]) is related to 1/1 E Irr(NE(C) Iq;) in this way, then their degrees 1/1'(1) and 1/1(1) satisfy
1/1(1)
= [NE(C) : N dC, q;) ]q;(1)1/I' (1)
(see [03, 12.17]). It follows that we can compute the defect of 1/1 from the defect of q; and the degree of 1/1'. Thus the number of 1/1 with a given defect is determined by the defect of q; and the central extension E[C, q;, J], without any further knowledge about the structure of N E(C). In fact, we only need to know this central extension to within isomorphisms, i.e., we only need to know the stabilizer N jj;(C, q;) and the element a[C, q;, J] of the second cohomology group H 2 ( NE (C, q;), U (J)) corresponding to the isomorphism class of the central extension E[C, q;, J].
50
E.C.Dade
In addition to the p-block B of G, the integer d 2: 0, and the subgroup ft of E, we now fix an element a in the cohomology group H 2 ( P, U (J)), where P acts trivially on U (J). Definition 3.2. For any p-chain C of G we denote by k(C, B, d, ft, a) the number of characters c/J E Irr(NC (C» satisfying d(c/J)
= d,
B(c/J)c
= B,
N E(C,
c/J)
= ft
and
arC,
c/J, J]
= a.
As usual, the number k(C, B, d, P, a) depends only on the G-conjugacy class of the p-chain C. So the sum in the following conjecture is well defined. The Extended Conjecture 3.3. If Op(G) = 1 and d(B) > O. then '"' ~
- a) = O. (-1) ICI k(C, B, d, F,
(3.4)
CEX(C)/C
This conjecture implies the Invariant Conjecture 2.5 since the equation (2.6) can be obtained by summing (3.4) over all a E H2 (ft, U(J»). If we know that the exact sequence (3.1) must split for all p-chains C of G and all irreducible J-characters c/J of Nc(C), then the Extended Conjecture 3.3 is equivalent to the Invariant Conjecture 2.5. This is notably the case when the group E ~ E / G is cyclic (see [03, 11.20 and 6.5]). It even occurs when the Sylow r -subgroups of E are cyclic for each prime r.
4. Projective Conjectures The next problem to face in trying to construct an inductive proof of the conjecture is that Clifford theory starts from irreducible ordinary characters but ends with irreducible projective ones. Thus it leads us from one situation to a different one, to which we cannot apply the Extended Conjecture 3.3. Howevever, Clifford theory works just as well for projective characters as it does for ordinary ones, and it always gives back projective characters. So we must reformulate our conjectures in terms of projective characters if we wish to prove them inductively using Clifford theory. We begin by replacing the finite group E by some central extension E[J] of U (J) by E. So we fix an exact sequence 1 ---+ U (J)
~
E[J]
~E
---+ 1
(4.1)
of groups such that U (J) is a central subgroup of E[J]. If H is any subgroup of E, then H[J] will denote the inverse image of H in E[J]. SO H[J] is a
Counting Characters in Blocks, 2.9
51
central extension of U (~) by H. We shall assume that ~ is a total splitting field for the extension E[~], in the sense that every irreducible projective ~-character 1J of the subextension H[~] is absolutely irreducible for every subgroup H of E. Then the degree 1J (1) of any such 1J divides the order IH I of the finite group H. So 1J has a defect d (1J), the largest integer d ~ 0 such that pd divides the integer IH I/1J (1). Because the group E[~] is totally split as an extension of U(~), it has a unique subgroup E[91] intersecting U(~) in U(91) and having E as its image (see [D3, 7.8]). Thus the restriction of l][~] is the epimorphism l][91] in an exact sequence I
~ U(91)~ E[91]~
E
~ 1
(4.2)
making E[91] a central extension of U(91) by E. The inverse image H[91] in E[91] of any subgroup H of E plays the same role for H[~] as E[91] plays for E[~]. The twisted group order D[H] of Hover 91 corresponding to the extension H[91] is a suborder spanning the twisted group algebra 2l[H] of H over ~ corresponding to the extension H[~]. So we may define the projective p-blocks of H[~] to be the blocks of this unique 91-order D[H]. We denote by Blk(H[91]) the set of all projective p-blocks of H[~]. Any irreducible projective ~-character 1J of H[~] is an irreducible character of 2l[H], and hence belongs to a unique projective p-block B(1J) of H[~]. The natural conjugation action of H on its twisted group algebra D[H] over 91 has the center Z(D[H]) as its set of fixed points. So we can use this action to define defect groups D :s H for any block b E Blk(H[91]) (see [D3, 9.3]). The defect d(b) of b is then the non-negative integer such that pdCb) is the order of each such D. Brauer induction of projective blocks can be defined just as was Brauer induction of ordinary blocks (see [D3, 10.5]). The equivalent of [KR, 3.2] holds for projective blocks by [D3, 10.14]. Once we change the misprinted "2l[NG(P)] " in the latter proposition to the correct "2l[NG(C)]," it tells us that any projective p-block b of NG (C)[~], for any p-chain C of G, induces a projective p-block bG[J] of G[~]. As usual, we fix a non-negative integer d. The earlier p-block B of G is now replaced by a fixed projective p-block B[~] of the central extension G[~] of U (~) by G. The projective equivalent of Definition 1.3 is Definition 4.3. For any p-chain C of G we define k(C, B[~], d) to be the number of irreducible projective ~ -characters 1J of N G (C) [~] such that d(1J)
=d
and
B(1J)G[J]
= B[~].
52
E.C.Dade
The resulting number k(C, B[J], d) depends only on the G-conjugacy class of C. The projective equivalent of the Ordinary Conjecture 1.4 is
The Projective Conjecture 4.4. If Op(G) = I and d(B[J]) > 0, then
L (- I),c'k(C, B[J], d) = O.
(4.5)
CE::R(GJ/G
This is the conjecture given in [D3, 15.5], but written in a slightly different form. So it implies the Alperin-McKay Conjecture by [D3, 17. IS and 18.5]. Of course it reduces to the Ordinary Conjecture 1.4 when G[J] is a split extension of U (J) by G. If C is any p-chain of G, then its normalizer N E(C) acts naturally by conjugation on NE(C)[JJ, centralizing the subgroup U(J). Since NG(C) is a normal subgroup of NECC), this action leaves NG(C)[J] invariant, and hence permutes among themselves the characters 4> E Irr(NG(C)[JJ). We denote by NE (C, 4» the stabilizer in NE (C) of any such 4>, and by Ni (C, 4» the image Ni(C,
of that stabilizer in
4» =
e(NE(C,
4»)
E. SO C and 4> determine an exact sequence
just as in (2.3). In addition to the above d and B[J] we now fix a subgroup Then the projective equivalent of Definition 2.4 is
F
of
E.
Definition 4.6. For any p-chain C of G we denote by k( C, B[J], d, F) the number of irreducible projective J-characters 4> of NG(C)[J] satisfying d(4)) = d,
B(4))G['5J = B[J]
and
Ni(C,
4» = F.
The number k(C, B[J], d, F) depends only on the G-conjugacy class of C. The projective equivalent of the Invariant Conjecture 2.5 is
The Invariant Projective Conjecture 4.7. If Op(G) = I and d(B[J]) > 0, then
L
(_l)ICik(C, B[J], d,
F) =
O.
(4.8)
CE::R(GJ/G
Of course this reduces to the Projective Conjecture 4.4 when E = G, and to the ordinary Invariant Conjecture 2.5 when E[J] is a split extension of U(J) by E.
Counting Characters in Blocks, 2.9
53
It is well known that Clifford theory works just as well for projective characters as it does for ordinary ones (see [D3, 12.12]). In particular, associated with each p-chain C of G and each character ep E Irr( NG(C)[~]) is a central extension E[C, ep,~] of U(~) by N E(C, ep). Our assumption that E[~] is totally split over ~ implies that E[C, ep,~] is also totally split over ~ (see [D3, 11.20]). There is a one to one correspondence between all characters 1/1 E Irr( N E(C)[~J) lying over ep and all characters 1/1' E Irr( E[C, ep, ~]). Furthermore, the number of such 1/1 with a given defect can be computed from the defect of ep and the cohomology class a [C, ep, ~] E H 2 ( NE (C , ep), U (~) ) corresponding to the extension E[C, ep, ~]. In addition to the above d, B [~] and F, we now fix an element a in the cohomology group H 2 (F, u (~»). The projective equivalent of Definition 3.2 is
Definition 4.9. For any p-chain C of G we denote by k(C, B[~], d, F, a) the number of irreducible projective ~-characters ep of NG(C)[~] such that
deep) = d,
B(cP)G[J]
=
B[~],
N E(C,
ep) = F
and
a[C, cP,~]
= a.
The number k(C, B[~], d, F, a) also depends only on the G-conjugacy class of C. The projective equivalent of the Extended Conjecture 3.3 is The Extended Projective Conjecture 4.10. If Op(G) = I and then 'L"
- a) = O. (-1) ICI k(C, B[~], d, F,
d(B[~])
> 0,
(4.11 )
CE~(G)/G
This reduces to the ordinary Extended Conjecture 3.3 when E[~] is a split extension of U (~) by E. It also implies the Invariant Projective Conjecture 4.7, and reduces to that conjecture whenever each a[C, ep,~] is known to be trivial for all p-chains C in G and all ep E Irr( NG(C)[~J). This last situation occurs in the case where E :::::= E / G has cyclic r -Sylow subgroups for all primes r.
5. The Inductive Conjecture We have still not reached the final modification needed to state a suitable inductive form of the conjecture. The remaining problem can be understood by considering our usual p-chain C of G and character ep E Irr( NG(C)[~J). The associated Clifford extension E[C, cP,~] does determine the degrees of
54
E.C.Dade
the characters 1/1 in Irr( NE(C)(J]) lying over cP, but it does not tell us to which projective p-block of NE( C) [J] any such 1/1 belongs. We need that extra information in order to compute the projective p-block of E[J] induced by B( 1/1). To obtain it we invoke the Clifford theory for blocks developed in [Dl]. Let b be any projective p-block of NG(C)[J]. We denote by NE(C, b) the stabilizer of b under conjugation by elements of NE (C), and by Ni (C, b) the image N i(C, b) = e(NE(C, b))
of that stabilizer in E. In the language of [Dl, 2.17] our present Ni(C, b) would be called Ni(C)h. We denote by Ci(C, b) the normal subgroup of N i(C, b) which would be called N i(C)[b] in the language of[Dl, §2]. This notation is chosen because Jacobinski has noticed that the inverse image CdC, b) = e- J (Ci(C, b))
n NE(C)
of C i (C, b) is precisely the subgroup of all a E NE (C) which centralize to within inner automorphisms the indecomposable direct summand of the 9t-order D[NG(C)] corresponding to b. The Clifford extension for the block b and the exact sequence (2.2) is a central extension E[C, b,~] of U(~) by C i(C, b). So it appears in an exact sequence _
I
-7
<J
_
U (J) ~ E [ C, b, J]
1)[C,h.JJ
) C i (C , b)
-7
1
(5.1)
of groups in which U(~) is a central subgroup of E[C, b, ~]. In the language of[Dl, 2.13] the extension E[C, b,~] would be called N i(C)[b]*. There is a natural conjugation action of the group Ni (C, b) as automorphisms of the central extension E[C, b,~] (see [Dl, 2.19]). Clifford theory for blocks [Dl, 3.7] gives us a one to one correspondence between all projective p-blocks fJ of N E(C, b)[J] lying over b and all N i(c' b)-conjugacy classes of projective blocks fJ' of the central extension E[ C, b,~] of U (~) by C i(C, b), Le., of all blocks fJ' of the twisted group algebra of C i(C, b) over J defined by that extension. Assume that the above block b induces the projective p-block B[J] of G[J]. The latter block has a normalizer Ni(B[J]) andacentralizer Ci(B[J]) in E. It also has a Clifford extension E[B[J]'~] of U(~) by Ci(B[J]), and a conjugation action of N £(B[J]) as automorphisms of that Clifford extension. Since b induces B[J], its normalizer N i(C, b) is a subgroup
Counting Characters in Blocks, 2.9
55
of NE (B [J]). A simple extension of the arguments in [DI, 8.1] shows that CE(B[J]) is a subgroup of CE(C, b), and hence is a normal subgroup of N E(C, b). Furthermore [DI, 8.1] also tells us that the Brauer homomorphism Brc = Br p associated with the final p-subgroup P = Pn in C sends E[B[J],~] monomorphically into E[C, b,~] in such a way that the following diagram commutes U(J)
E[B[J],
1~
l
U(J)
J]
I][B[J]'~l )
src
E[C, b, J]
I][C,b.~J )
C E(B[J])
~
1~ CE(C, b)
(5.2) ~
and the conjugation actions of N E(C, b) on E[B[J],~] and E[C, b,~] are preserved. We have already remarked that J is a total splitting field for the Clifford extension E[C, cP, J] of U(J) by N E(C, cP) associated with any irreducible projective J-character cP of Nc(C)[J]. As in (4.2), this implies the existence of a unique subgroup E[C, cP, 91] of E[C, cP, J] covering N E(C, cP) and intersecting U(J) in U(91). Factoring E[C, cP, 91] by the kernel 1 + P of the natural epimorphism of U(91) onto U(J), we obtain a central extension E[C, cP,~] of U(J) by N E(C, cP). We call E[C, cP, J] the residual Clifford extension for the character cjJ. Suppose that the above character cP lies in the projective p-block b of NcCC)[J]. Then N E(C, cjJ) is a subgroup of N E(C, b). Furthermore, [DI, 13.61 tells us that NE(C, cP) contains CE(C, b). By [DI, 13.101 there is a natural monomorphism /l[C, cP] of E[C, b, Jl into E[C, cP, J] such that the following diagram commutes U(J)
E[C, b,
1=
1
U(J)
E[C,
J]
I][C,b,JJ )
(5.3)
JLlC '4>J
cP, J]
I][C,ep,JJ )
and the conjugation actions of N E(C, cP) on E[C, b, Jl and on E[C, cP, J] are preserved. When the block b containing cP induces B[J], we may compose the monomorphisms in the commutative diagrams (5.2) and (5.3) to obtain a mono-
56
E. C. Dade
morphism p,[B[J], C, cP] = jL[C, cP]
0
Brc: E[B[J], J]
>---;
E[C, cP, J]
of groups such that the following diagram commutes
1 -----+
ry[B[JJ,JJ
U(J)
----=+
1 Br;>], c,
1= 1 -----+ U(J)
E[B[J], J] JL [
>]
Ci(B[J)) -----+ 1
1~
(5.4)
Ni(C,cP)
-----+ 1
ry[C,c/J,JJ
----=+
)
E[C, cP, J]
)
and the conjugation actions of N i(C, cP) on E[B[J], J] and E[C, cP, J] are preserved. If 1/J is an irreducible projective J-character of N E(C) lying over cP, then we can compute the projective p-block B(1/J)E['JJ of E[J] by the following recipe: Let 1/J' be the irreducible projective character of E[C, cP, J] corresponding to 1/J under Clifford theory for cP. Then 1/J' belongs to a projective p-block B(1/J') of E[C, cP, J]. This projective p-block corresponds to a unique projective block B(1/J') of the residual Clifford extension E[C, cP, J]. The latter block lies over a unique N i(C, cP)-conjugacy class of projective blocks fJ of the restriction of E[C, cP, J] to an extension by the normal subgroup CE(B[J)) of NE(C,cP), i.e., to the image of E[B[J],J] under the monomorphism p,[B[J], C, cP] in (5.4). Each such fJ is the image of some unique projective block fJ' of E[B[J], J] under that monomorphism. The resulting fJ' form a single Ni (C, cP) -conjugacy class, which is contained in a unique N i(B[J))-conjugacy class. The projective p-block B(1/J)E[JJ of E[J] corresponds to that N i(B[J))-conjugacy class under Clifford theory for the block B [J]. We assume from now on that our fixed subgroup F of E satisfies
As we have seen, the subgroup Ni (C, cP) satisfies these conditions whenever C is a p-chain of G and cP is an irreducible projective J-character of Nc(C)[J] belonging to a projective p-block B(cP) inducing the fixed projective p-block B[J] of G[J]. Definition 5.5. An augmented central extension of U (J) by F is an ordered pair (F[J], K). The first element in this pair is a totally split central extension
Counting Characters in Blocks, 2.9
57
F[J] of U(J) by F. SO it occurs in"an exact sequence
I ---+ U (J)
~
1][F[~]]_
_
F[J] -----+) F ---+ I ,
in which U (J) is a central subgroup of F[J]. Because this central extension is totally split over J, there is an associated residual central extension F[J] of U (J) by F, with related exact sequence _
I ---+ U(J)
~
_
-
1][F[J]]
F[J]
_
) F ---+ 1.
The second element in our ordered pair is a monomorphism K of the Clifford extension E[B[J], J] for the block B[J] into the central extension F["J]. SO it occurs in a commutative diagram 1][B[J],JJ
----+
U(J)
---=+
E[B[J], J]
1= ~
U(J)
---=+
C t(B[J])
-+
1~
1"
)
(5.6)
1] [F[J]]
F[J]
)
F
-+
with exact rows. Furthermore, this monomorphism K: E[B[J], J] >--+ F[J] carries the conjugation action of F on the former group (the restriction of the conjugation action of N t(B[J]) on that group) into the conjugation action of -
--
F on F[J]. If (F[J]', K') is also an augmented central extension of U (J) by F, then an isomorphism of (F[J], K) onto (F[J]', K') is an isomorphism y of F[J] onto F[J]' as extensions of U (J) by F such that the induced isomorphism y of F["J] onto F[J]' satisfies K'
=
Y0
K :
E[B[J],
J] >--+
F[J],.
From now on we fix, in addition to the earlier d, B[J] and mented central extension (F[J], K) of U(J) by F.
F,
an aug-
Definition 5.7. For any p-chain C of G, we denote by k(C, B[J], d, F[J], K) the number of irreducible projective J-characters rjJ of NG(C)[J] such that d(rjJ) = d,
B(rjJ)G[~] = B[J],
and the augmented central extension (E[C, rjJ, J], Il[B[J], C, rjJ]) of U(J) by Nt(C, rjJ) is isomorphic to the augmented central extension (F[J], K) of U(J) by F (so that, in particular, N t(C, rjJ) is equal to F).
58
E. C. Dade
As always, the number k(C, B[Jl, d, F[J), K) depends only on the G-conjugacy class of C. So the sum in the following conjecture is well defined.
The Inductive Conjecture 5.8. If Op(G) = 1 and d(B[J]) > 0, then '"'" ~
(-1) ICI k(C, B[J], d, F[J], K) = 0.
(5.9)
CE:R(G)/G
Of course this conjecture implies all our previous ones. With a great amount of work it can be shown to hold for all finite groups if it holds whenever G is a non-abelian finite simple group. So it can be proven by checking it for all such G.
6. Simple G The verification of the Inductive Conjecture 5.8 for a given non-abelian finite simple group G requires only a finite calculation. The isomorphism classes of central extensions G[J) of U(J) by G correspond one to one to the elements of the Schur multiplier H 2 (G, U (J»), which is a finite group of known structure for each simple G (see [Atlas]). So there are only a finite number of choices for G[J) once G is given. If G[J) is fixed, there are an infinite number of possible choices for its extension group E[J). However, the inductive conjecture can be shown to hold for all these choices provided it holds for one of them satisfying the extra condition that conjugation by suitable elements of E[J) induces every possible automorphism of the central extension G[J) of U (J). Such an E[J) always exists. So we only need consider one E[J) for any given G[J). Every term in the sum on the left side of (5.9) is zero when the integer d is so big that pd does not divide IG I. Hence the inductive conjecture holds trivially except for a finite number of choices of d. There are, in any case, only a finite number of choices for the other parameters B[J], F, F[J) and K appearing in that conjecture. So the verification of it is a finite calculation for any fixed G. The Inductive Conjecture 5.8 has been verified completely for many finite simple groups G, including all the Mathieu groups, the first three Janko groups, the Suzuki groups Sz(q) and the linear groups PSL2(q). Partial verifications have been made for the two classes of Ree groups, the groups G2 (q), and the linear groups PSL3(Q). In all these cases the inductive conjecture is equivalent to weaker conjectures, such as the Invariant Projective Conjecture 4.7 or the Extended Projective Conjecture 4.10. For example, if every Sylow r -subgroup
Counting Characters in Blocks, 2.9
59
of E is cyclic, for every prime r, then the inductive conjecture is equivalent to the invariant projective conjecture. This is enough to handle the cases where G is a sporadic simple group or an alternating group An with n -=1= 6. It even covers many cases where G has Lie type. The inductive conjecture reduces to the extended projective conjecture when the subgroup CE(B[JD is trivial. Using [D1, O.3bl we can show that this happens for all groups of Lie type when p is their defining characteristic. Another situation where the inductive conjecture reduces to the extended projective one occurs when E is abelian. This is what enables us to handle PSL2(Q) completely. Note added in proof. Since this paper was written, the inductive form of the conjecture has been verified completely for more simple groups, including the McLaughlin group, the Held group, the third Conway group, the Tits group, and the Higman-Sims group. There have also been partial verifications for a few more families of groups of Lie type. So far no counterexample has been found, although there was a brief flurry of interest on the Internet in January and February, 1997, when a counterexample was erroneously thought to exist.
References [Atlas] J. Conway, R. Curtis, S. Norton, R. Parker and R. Wilson, Atlas of Finite Groups, Oxford University Press 1985 [01]
E. Oade, Block Extensions, Ill. J. Math. 17 (1973),198-272.
[02]
E. Oade, Counting Characters in Blocks, I, Invent. Math. 109 (1992), 187-210.
[03]
E. Oade, Counting Characters in Blocks, 11, 1. Reine Angew. Math. 448 (1994), 97-190.
[04]
E. Oade, Counting Characters in Blocks, Ill, in preparation.
[KR]
R. Knorr and G. Robinson, Some Remarks on a Conjecture of Alperin, J. London Math. Soc. (2) 39 (1989), 48-60.
[RS]
G. Robinson and R. Staszewski, More on Alperin's Conjecture, Asterisque 181/182(1990),237-255.
Department of Mathematics The University of Illinois at Urbana-Champaign 1409 W. Green St. Urbana, IL61801, USA Email: [email protected]
The Defect Groups of a Clique H. Ellers
Alperin's Weight Conjecture and Brauer's First Main Theorem on Blocks are strikingly similar. Let k be an algebraically closed field of characteristic p. For any finite group G and any p-subgroup Q of G, let a(G) be the number of irreducible kG -modules and let aQ (G) be the number of irreducible kG-modules with vertex Q; let beG) be the number of blocks of kG and let bQ(G) be the number of blocks with defect group Q. It follows from Brauer's First Main Theorem that beG) =
L bQ(NC(Q», Q
where Q runs through a set of representatives for the conjugacy classes of p-subgroups of G. Alperin's Conjecture suggests that a(G) =
L aQ(NG(Q», Q
where Q runs through the same set. Are there other formulas of this type? This is a report on work that shows that when G is p-solvable, there is a whole family of such formulas, one for each normal subgroup H of G, with Alperin's Conjecture as the case H = I and Brauer's First Main Theorem as the case H = G. Let H be a normal subgroup of G. Let kG H be the centralizer in kG of k H. The group G acts on kG H by conjugation, so there is also a natural conjugation action of G on the set of isomorphism types of irreducible kG H -modules. By Theorem 2.5 in [Ell], there is an analog of Clifford's Theorem for the restriction of an irreducible kG-module V to kG H : the restriction VkGH is semi-simple, the set of all isomorphism types of simple summands of VkGH is a G-conjugacy class, and each isomorphism type of simple summand occurs the same number of times in a decomposition of VkGH. We say irreducible kG -modules V and W are H-equivalent if there is a nonzero submodule of the restriction VkGH that is isomorphic to a submodule of the
62
H. Ellers
restriction WkGH. (It follows from the analog of Clifford's Theorem that this is an equivalence relation. If there is one irreducible submodule of VkGH that is isomorphic to a submodule of WkGH, then every irreducible submodule of VkGH is isomorphic to a submodule of WkGH and every irreducible submodule of WkGH is isomorphic to a submodule of VkGH.) We call the equivalence classes H-cliques. The partition of the set of irreducible kG -modules into H -cliques is similar to the partition into blocks. When H = G, irreducible kG-modules belong to the same H -clique if and only if they belong to the same block. In general, when H is a proper subgroup, the partition into H -cliques is a refinement of the partition into blocks. When H = 1, irreducible kG-modules belong to the same H -clique if and only if they are isomorphic. The papers [Ell], [EI2], and [E13] develop a theory of cliques as similar as possible to Brauer's theory of blocks. In particular, defect groups of a clique are defined in such a way that, for p-solvable groups G, there is a First Main Theorem: for any p-subgroup Q of a p-solvable group G, the number of H -cliques with defect group Q of irreducible kG-modules is equal to the number of N H(Q)-cliques with defect group Q of irreducible kNG (Q) -modules. Now we explain how defect groups are defined. We need the following theorem of Green [Gr]. To state Green's theorem, we need some definitions. Definition 1. A G-algebra is a finite dimensional algebra over k on which the elements of G act as algebra automorphisms. If A is a G-algebra and K is a subgroup of G, then A K is defined to be {a E A lax = a for all x E K}. For any a E A K , Tf (a) = Li a gi , where {gi} is a set of representatives for the right cosets Kgi of K in G. Theorem 2. Let G be a finite group, let A be a G -algebra, and let M be a maximal ideal of A G. Then the the set of all subgroups D of G minimal (with respect to inclusion) with the property Tg (A D) Sf M is a G-conjugacy class of p-subgroups of G. For any maximal ideal M as in Green's theorem, the groups D are called the defect groups of M. Consider the classical case A = kG. For any centrally primitive idempotent e of kG, there is a unique maximal ideal M of kG G with e tt M. The defect groups of M are the same as the defect groups in Brauer's sense of e. If V is an indecomposable kG-module, then Endk(V) is a G-algebra and Endk (V)G is a local ring. The defect groups of the unique maximal ideal of Endk(V)G are the vertices of V.
The Defect Groups of a Clique
63
Let e be an H -clique of irreducible kG-modules. We wish to use Green's Theorem to define defect groups of e. To do this, we need to find an appropriate G-algebra and an appropriate maximal ideal of its subalgebra of G-fixed elements. First, we will examine a G-algebra which is not quite the correct one to use, but which is closely related to the correct one. Note that if V and Ware irreducible kG-modules in the same H -clique, then annkGH (V) = annkGH (W). Definition 3. Let H be a normal subgroup of the finite group G. Let e be an H -clique of irreducible kG-modules. Let Vc be an irreducible kG-module in e. (1)
AG,H(e) = kG H /annkGH(Vc ),
When ambiguity is not possible, we will drop the subscript G, H from the notation and write simply A(e). (2) For any a
E kG H ,
a is its natural image in A(e).
The algebra kG H / J (kG H ) is semi-simple, with one simple summand for each isomorphism type of irreducible kG H -module; the action of G permutes these summands. The G-algebra A(e) is naturally isomorphic to the G-subalgebra of kG H / J (kG H ) consisting of all the simple direct summands corresponding to modules in the single G-orbit of irreducible kG H -modules associated to e. Whenever it is convenient, we will identify A(e) with this G-subalgebra of kG H / J(kG H ). Using the analog of Clifford's Theorem, we see thatkG H n J(kG) = J(kG H ); it follows that the natural map kG H ~ kG/ J(kG) induces an injective map kG H / J(kG H ) ~ kG/ J(kG); therefore A(e)G is contained in the center of A(e). Since A(e) is the direct sum of a single G-orbit of simple k-algebras, we get the following lemma. Lemma 4. Let G, Hand e be as in Definition 3. Then A(e)G ~ k. The algebra A(e) is the most natural G-algebra to associate to e. However, thinking about the classical situation G = H shows that it is not the correct algebra to use to define defect groups. When G = H, A(e) is I-dimensional and the action of G is trivial, so the defect groups of the unique maximal ideal 0 of A(e)G are the Sylow p-subgroups of G, not the defect groups of the block identified with e. We need a larger algebra and an ideal that together encode information about four things: A(e), the action of G on A(e), the action of G on kG, and the natural map kG H ~ A(e). The required algebra is provided by the construction of Definition 5.
64
H. Ellers
Definition 5. Let G be a finite group and let A be a G-algebra. Then A*kG
is the following algebra. As a vector space, A a * x = a ® x. Multiplication is defined by (a]
* x])(az * xz) =
-I
XI
ala z
* kG
is A &h kG with
* x]xz
whenever a 1 and az are elements of A and x] and xz are elements of G, and is extended linearly to all of A * kG. This algebra is a G -algebra via the action given by (a *x)g = (l *g-])(a *x)(l *g). This algebra is sometimes called a skew group ring for G with coefficients in A or a trivial crossed product. (The word trivial is used here because we could also have obtained an associative algebra by introducing a non-trivial factor set I(Xl, xz) on the right hand side of the definition of multiplication above.) The appropriate G-algebra for the definition of the defect groups of e is A(e) * kG. Now we must identify an appropriate maximal ideal of (A(e) * kG)G. First, we need to find the H -fixed elements of A(e). This is similar to the usual determination of the center of a group algebra. Think of elements of the skew group ring as linear combinations with coefficients in A(e) of elements in G. Since the action of H is trivial on the coefficients, the coefficients of any H -fixed element must be constant on H -conjugacy classes in G. Since the class sums for H -conjugacy classes form a basis for kG H , we obtain the following lemma.
* kG)H = A(e) * (kG H ). The appropriate maximal ideal of (A(e) * kG)G is provided by the intersection of (A(e) * kG)G with the kernel of the following map.
Lemma 6. With the above notation, (A(e)
Definition 7. Let G, H, and
e be as in Definition 3
AG,H,e : A(e)
* (kG H ) -+ A(e)
is the map given by AG,H,e(a * b) = ab for all a E A(e) and all b E kG H . Whenever ambiguity is not possible, we will drop the subscript G, H, e from the notation and write simply A. It is easily checked that the map A is a surjective G-algebra homomorphism.
The Defect Groups of a Clique
65
In the special case H = G, the map A is a very familiar object; in this case e is the set of irreducible modules in a block B of kG, A(e) is a one-dimensional G-algebra with trivial action of G, A(e) * kG is the group algebra kG, A(e) * (kG H ) is the center of the group algebra, and A is the central character of kG corresponding to the block B. Since (A(e))G is one-dimensional, and since A is a G-algebra homomorphism, it follows that (A(e) * kG)G n Ker(A) is a maximal ideal of (A(e) * kG)G. Definition 8. Let H be a normal subgroup of the finite group G. Let e be an H -clique of irreducible kG-modules. Let A be the map of Definition 7. The defect groups of e are the defect groups of the maximal ideal (A(e) * kG)G n Ker(A) of (A(e) * kG)G. It is immediate that in the case H = G, this agrees with the usual definition of the defect groups of a block. It is also easy to check that when H = 1 , there is just one irreducible module in the clique e and the defect groups are just the vertices of that module. A warning is needed about the possible dependence of defect groups on H. It is sometimes possible to change H without changing the set of modules e. (For example this happens if H is replaced by a normal subgroup HI of G with HI S; Hand H / HI a group of order prime to p.) However, the algebras and maps in the definition of defect groups depend not only on the set e but also on H. Thus defect groups are, at least in principle, only defined relative to H. It should always be clear from the context which H is intended. Whether changing H can change the defect groups in a case when it does not change e is not known. It is natural to speculate, by analogy with blocks, that the defect groups only depend on the vertices of the modules in the clique; however very little is at present known about this. On the positive side, if D is a defect group of the H -clique e of irreducible kG-modules, and if V is any module in e, then V is D-projective. With this definition of defect groups of a clique, the main theorem of [EI3] is the following. Theorem 9. Let k be an algebraically closed field of characteristic p, let G be a p-solvablefinite group, let H be a normal subgroup of G, and let D be a p-subgroup of G. Then the number of H -cliques with defect group D of irreducible kG-modules is equal to the number of N H(D)-cliques with defect group D of irreducible kNG(D)-modules.
66
H. Ellers
For any normal subgroup H and any p-subgroup Q of a finite group G, let CH (G) be the number of H -cliques of irreducible kG-modules and let CH, Q (G) be the number of H -cliques with defect group Q of irreducible kG-modules. In terms of the notation of the first paragraph, a(G) = Cl (G), aQ(G) = CI,Q(G), beG) = cGCG), and bQ(G) = cG,Q(G). Combining all possible defect groups, we obtain the following corollary. Corollary 10. With the above notation, CH(G)
if
G is p-solvable, then
= LCNH(Q),Q(NGCQ», Q
where Q runs through a set of representatives for the conjugacy classes of p-subgroups of G. There are groups G that are not p-solvable for which the conclusion of Theorem 9 is false, even in the familiar case H = 1. However, the corollary (which is Alperin's Conjecture when H = 1) may turn out to be true for all groups G.
References [AI]
J. L. Alperin, Weights for finite groups, in: The Areata conference on representations of finite groups (P. Fong, ed.), Proc. Symp. Pure Math. 47, Part 1, Providence, R.I., 1986,369-379.
[Gr]
1. A. Green, Some remarks on defect groups, Math. Z. 107 (1968), 133-150.
[Ell]
H. Ellers, Cliques of irreducible representations of p-sol vable groups and a theorem analogous to Alperin's conjecture, Math. Z. 217 (1994), 607-634.
[EI2] [EI3]
H. Ellers, Cliques of irreducible representations, quotient groups, and Brauer's theorems on blocks, Canadian J. Math. 47(5) (1995), 929-945. H. Ellers, The defect groups of a clique, p-solvable groups, and Alperin's Conjecture, J. Reine Angew. Math. 468 (1995), 1-48.
Department of Mathematics Northern Illinois University DeKalb, IL 60115 Email: [email protected]
Representations of GLn(K) and Symmetric Groups Karin Erdmann
Let K be an infinite field of characteristic p ~ O. It is known for some time that there is a close connection between the representation theory of the symmetric groups over K and the theory of polynomial representations of GL n (K). For K = C there is already the work of I. Schur [S], and later one has [CLl. [G], [1] and others. More recently Cline, Parshall and Scott defined quasi-hereditary algebras in order to deal with highest weight categories as they arise in the representation theory of semisimple complex Lie algebras and algebraic groups [CPS 1]. Subsequently, the structure of quasi- hereditary algebras in general was studied in [DR], [R] and by others. This led to the discovery of a new class of modules which are parametrized by highest weights, called canonical modules, or 'tilting modules' [R]. For the case of the highest weight category associated to GLn(K), these modules provide a new connection to the representation theory of the symmetric groups and give new insight to longstanding problems, such as decomposition numbers and dimensions of simple modules. The aim of this paper is to give an exposition of relevant properties of the canonical modules for GL n (K); in particular we shall explain the relationship to symmetric groups and give an overview of some recent results on the problems of decomposition numbers and dimensions of simple modules.
1. Symmetric Groups Sr and Schur Algebras 1.1. We consider modules of the group algebra of the symmetric group Sr over K. For each partition J-L of r, let SI1 denote the Specht module corresponding to J-L; it is defined characteristic-free. If char K = 0 then SI1 is simple and this gives a full set of pairwise non-isomorphic simple modules. Over characteristic p, the simple modules for K Sr are labelled as D A where A. runs through the p-regular partitions of r. If )... is p-regular then the Specht module SA has unique simple quotient D A ; and all other composition factors
68
Karin Erdmann
of SA are of the form DJi- for A < p.,. Here :s denotes the dominance order of partitions. The decomposition numbers can be taken as the composition multiplicities of the Specht modules SJi-; we shall use the notation [SJi- : D A]. We shall need the following sets labelling simple modules. First, A + (n, r) is the set of partitions of r with at most n parts, and A%(n, r) denotes the subset of p-regular partitions.Moreover, we write A +(n) for the set of all partitions with at most n parts, and A%(n) is the subset of p-regular partitions. We also write IAI = r if A is a partition of r. Moreover, we denote by A (n, r) the set of unordered partitions of r with at most n parts. For each A E A(n, r) let M A be the permutation module corresponding to A. If A is a partition then SA is eplicitly defined as a submodule of M A . It has strong uniqueness properties which guarantee that there is a unique indecomposable summand of M A containing SA; and this is by definition the Young module corresponding to A, denoted by fA. The modules fA have filtrations by Specht modules and also by duals of Specht modules. 1.2. [G,M] Let E be a fixed n-dimensional vector space over K. The symmetric group S, of degree r acts on the right of E®' by place permutations. The classical Schur algebra can be defined as the endomorphism ring of this module, S(n, r) := Ends r (£®')
As a module over K S" £®, is isomorphic to the direct sum (J}M A where the sum is taken over A (n, r). In case n ~ r, the Schur functor f from the category of S(n, r)- modules to the category of K S, - modules is an important tool; we recall the definition. Let e : £®, --+ E®' be the projection with image MO') which is zero on M A for A # (1'), then e is an idempotent in S(n, r), and one sets f(M) := eM. The algebra eS(n, r)e is isomorphic to the group algebra of S,. Namely, eS(n, r)e is isomorphic to the endomorphism ring of MO') but MO') is isomorphic to K S,. Hence one can view f M as a left module for K S,. In particular, the above argument also shows that f(£®') ~ KS,.
Representations of GL n (K) and Symmetric Groups
69
2. GL n (K )-Modules and Tilting Modules 2.1.
Let K be an infinite field. Fix an integer n > 1, let G be the group GLn(K). We denote by M r the category of finite- dimensional polynomial representations of G which are homogeneous of degree r. Most important is the natural n-dimensional G-module which we denote by E. Then E belongs to Ml and for any r ~ 1, the r-fold tensor product EQl)r is in M r , and also r-fold symmetric powers and r-fold exterior powers, for r ::::: n. The category M r is equivalent to the category of Sen, r)-modules (see [G, MD. The simple modules are parametrized by highest weights as L(A), for A E A. +(n, r). 2.2. For A E A. +(n, r) let ~(A) be the Weyl module with highest weight A, and let V(A) be its contravariant dual (see e.g [G] or [M] for an explicit construction). For example, the r -fold symmetric power Sr (E) is isomorphic to VCr). Moreover the r-fold exterior power /\r(E) is isomorphic to Vcr) for 1 ::::: r ::::: n. Actually this is simple, ~ L((r)) and isomorphic to its dual. In general, the important property is that ~(A) has a unique simple quotient isomorphic to L(A); and all other composition factors are of the form L(p.,) for p., < A; here ::S is dominance order of partitions. This is analogous to the property of Specht modules SA for p- regular A as mentioned above; except that the order is reversed. The category M r is a highest weight category in the sense of [CPS1] where the weight poset (A.,:::::) is taken as A. +(n, r) with the dominance order of partitions, and where the Weyl modules are the 'standard modules' and their duals the 'costandard modules'. Equivalently, the Schur algebra is quasi-hereditary, with respect to the same data. This was already proved in [0; II (2.2)], then in [Pl. More recently, there is a new proof by 1. A. Green [G2]. 2.3.
A module M has a Weyl filtration if it has submodules 0= Mo C Ml C ... C Mk = M
with Mi/Mi-l ~ ~(p.,i) for some weight p.,i. Let 1'(~) be the category of modules with Weyl filtration. Similarly one defines 1'(V), the category of modules with V -filtrations (or' good filtrations '). For M in 1'(~), the number of times some given ~(p.,) occurs depends only on M and not on the filtration. We denote this multiplicity by [M : ~(p.,)], similarly one defines [M : V(p.,)] for M with a V -filtration.
70
Karin Erdmann
2.4. There is a class of modules discovered recently by Ringel, which also can be parametrized by highest weights and they are important in this context (see § 1 in [D2]). Theorem [R]. For each A, there is a unique indecomposable module T(A) in the intersection of J'(~) and J'(Y') with highest weight A. We call these modules canonical modules. The direct sum of all T(A) is a generalized tilting module as defined in [R], in the usual sense; and with the common terminology any direct sum of modules of the form T (A) is a partial generalized tilting module; the notion 'tilting module' for T (A) is sometimes used, although it is not correct. Write 'J = J'(~) n J'(Y'). It is not hard to see that 'J is closed under direct summands. Furthermore, it was proved that J'(Y') is closed under tensor products. (For a discussion of this, see [Ja;IIA]). It follows that J'(~) and hence 'J are also closed under tensor products. This is very important and it will be used a number of times. 2.5. In the case of GLn(K) (or Schur algebras), there is a more explicit construction which is due to S. Donkin [D2]. Suppose a = (aI, ... ,ak) is a partition, and define the generalized exterior power to be
Then we have Theorem [D2]. (1) The canonical modules for JY(r (or S(n, r») are precisely the indecomposable summands of 1\ a' (E) for a E A +(n, r). Moreover, (2) T(A) is the unique summand of
I\}..' (E)
with highest weight A.
The main ingredient is the result that 'J is closed under tensor products. So for (1) it suffices to show that for 0 :s a :s n the module A a (E) belongs to 'J. But AU (E) is simple and since it is a dual Weyl module it is then also isomorphic to a Weyl module, hence is in 'J. Part (2) follows from weight considerations. 2.5.1. For example, E ~ T(1) ~ Y'(1) ~ ~(1). More generally it is true that T(A) ~ ~(A) if and only if ~(A) ~ L(A), is simple. (If T(A) ~ ~(A) then ~(A) belongs to 'J, hence it must be isomorphic to Y'(A); and then it is simple, by the universal property. The converse holds since L(A) is self-dual.)
Representations of GL n (K) and Symmetric Groups
71
2.6. It is often convenient to work with SLn(K), and for the study of multiplicities of Weyl modules or canonical modules one does not lose anything. Consider the equivalence relation on A + (n) defined by A '" /-t {:} Ai - Ai+1 = /-ti - /-ti+l, (l::::: i ::::: n - I).
Then A '" /-t if and only if the restrictions of M(A) and M(/-t) to SLn(K) are isomorphic in each of the cases when M = L, ~, V or T (see [01 ;3.2.7] and [02;1.3]). The restrictions of A to SL n (K) give precisely the dominant weights, and the restriction of A in terms of the fundamental dominant weights is [AI - A2, ... ,An-I - An]·
We will use that the partition 8 = (n - I, n - 2, ... , 2, I, 0) restricts to p = [I, I, ... , I] which is equal to half the sum of the positive roots. Recall that a dominant weight as above is restricted provided 0 ::::: Ai Ai + I ::::: p - 1 for 1 ::::: i ::::: n - I. 2.7. There is a result on twisted tensor products for canonical modules which is due to Oonkin [02] and which is extremely useful. We consider the case of SLn(K); let F denote the Frobenius twist.
Proposition. Suppose ji = (p - l)p
+ Y where
y is restricted. Then for any dominant weight r, the module T(ji) ® T(r)F belongs to 'I. If it is indecomposable then it is isomorphic to T(ji + pr). This is known to be true when p ::: 2n - 2; in particular always for n = 2.
We shall sketch the first part of the proof. The Weyl module ~«p - I)p) (the Steinberg module) is known to be simple, hence it belongs to 'I (see 2.5.1.) It follows that T (y) ® ~«p - I)p) also belongs to 'I and it has T (ji) as a direct summand since ji is its highest weight. So the first part will follow if one shows that T(y) ® ~«p -I)p) ® T(r)F belongs to 'I, and for this it suffices to establish that ~ «p - 1) p) ® T (r) F has a ~-filtration and a V-filtration. This is now a consequence of a result proved by Jantzen [Ja,II.3.19]. Namely one has
+ p/-t), I)p + p/-t).
V«p - I)p ® V(/-tl ~ V«p - I)p
~«p - I)p) ® ~(/-t)F ~ ~«p -
For the last part and a related conjecture, see the discussion in [02] after 2.1. Example. Assume n = 2 and work with SL2(K). Then using 2.6, the partition A = (A I, A2) corresponds to the weight m = AI - A2.
72
Karin Erdmann
(a) Suppose 0 :::: m :::: p - 1, then T(m) ~ l::1(m) since in this case l::1(m) ~ L(m) ~ V(m); this is well-known (and easy to see by direct calculation, in the m-fold symmetric power.) (b) If m ?: p write m = kp
+i
where 0 :::: i :::: p - l. Then
"-' { T(k)F 0 T(p - 1) T(m) = T(k _ I)F 0 T(p i)
+
i = P- 1 i:::: p - 2
These are special cases of 2.7 (where r = k or k - 1). (c) The proof of 2.7 shows that for any sand 0 :::: i :::: p - 2 the module l::1(s)F 0 T(p+i) belongs to J"(l::1). In fact it is known that it has l::1-quotients l::1(sp j) and l::1((s 1)p i) where i j = p - 2 (see [X]). In particular this gives the l::1-quotients for T(p i).
+
+
+
+
+
3. Symmetric Groups and Schur Algebras 3.1. We have seen that the r-fold tensor product is a direct sum of canonical modules. More precisely, Theorem [D; IV]. There is an isomorphism of Sen, r)-modules E®r ~ tBAdAT(A) where the sum is taken over A;(n, r). Moreover the multiplicity dA is equal to the dimension of the simple module D A of the symmetric group Sr.
We give an outline of the proof, for the details see [EI, 4.2]. Let E®r = tBd),T(A) , where the sum is taken over A +(n, r), we have to find the multiplicities. One reduces to the case n ?: r, then the Schur functor f : M r -+ KS r - mod is defined. One has f(E®r) ~ KS r and f(T(A») ~ yA' 0 K a where yA' is the Young module corresponding to A' and K a is the alternating representation (see [D2;3.8]). It is known that yA' 0 K a is projective if and only if A is p-regular, and if so then it is isomorphic to P(D A). The multiplicity of P(D A) as a direct summand of KS r is dim D A •
3.2.
If A is any quasi-hereditary algebra with respect to (A, ::::) then the endomorphism ring B = EndA(tBT(A» is again quasi-hereditary (with respect to (A, ::::oP», the algebra B is called the 'Ringel dual' of A. It has therefore also canonical modules; if they are labelled as TB (A) then the endomorphism ring EndB(tBTB(A)) is Morita equivalent to A (see [R]).
Representations of GL n (K) and Symmetric Groups
73
Consider the case of S := S (n, r). Let C = ffj T (It) where the sum is taken over It E A+(n,r)\At(n,r). Then by 3.1 we have that T:= E®rffjC isin 'J and every T(A) occurs as a direct summand of T. Hence if S' := Ends(T) then S' is Morita equivalent to the Ringel dual of S. This algebra S' is related to quotients of the group algebra K Sr by the following.
Theorem. Let e
E
S' be the projection onto E®r with kernel C.
(a) We have
eS'e 2:: Ends(E®r) ~ KS r / In where In is the kernel of the action on E®r. Moreover (b) For A E A +(n, r) , the module eLl~(A) is identified with the Specht module SA and eT;(A) with the Young module y A• Moreover eV's'(A) (SA)*. (c) The simple module eLs'(A) corresponds to DA, for A E At(n, r).
(d) Assume that all It with It :::: A in the same block as A are p-regular. Then eLls'(A) ~ Lls,(8) and eTsf(A) ~ TS,(A). The first isomorphism is clear. For the next one, recall that Sen, r) is the endomorphism ring of the K Sr -module E®r, hence there is a canonical map
Pn : KS r --+ Ends(E®r) . It follows from [CP] that Pn is surjective. For the other parts we refer to [El, 4.2]. This may explain the analogy between module structure of Specht modules and of Weyl modules mentioned above; and also the filtration properties of Young modules as compared with those of canonical modules.
3.2.1.
In fact, there is now a generalization. Suppose R is any finite-dimensional algebra, T an R-module and A = EndR(T). In [CPS2J, the relationship between the representation theories of R and A is studied systematically, especially for A quasi-hereditary. This work includes amongst others an abstract 'Specht/Weyl module correspondence'.
3.3.
We shall now see that the above result explains why decomposition numbers for symmetric groups are related to the quasi- hereditary structure of the Schur algebras.
Theorem. Let A, It
E
A +(n, r), then
[T(A) : Ll(It)] = [Ll(It') : L(A')]
74
Karin Erdmann
If in addition A is p-regular then also
The first part is due to Oonkin (in the case when A is p-regular it can also be deduced from the second part.) For the second part, let S' be the Ringel dual of S. By [R] there is an equivalence
which takes the indecomposable injective module Q~(A) to T(A) and also V'~(Il) to 6.(Il). Consequently [T(A) : 6. (II)] = [Q~(A) : V'~(Il)]. By 'Brauer-Humphreys reciprocity' (see [CPSI], the second multiplicity is equal to [6.~(Il) : L~(A)], and by the identification in 3.3. it is equal to [SJ-L : D A ]. Remark. (1) In cases when the decomposition numbers for Sr are known this may be used to find the 6. -filtration of T (A) for A p-regular. The column of D A in the decomposition matrix for partitions of at most n parts gives the 8- quotients of T()'). (2) By 2.6, the filtration multiplicities [T(A) : 6.(Il)] depend only on the equivalence class of A, II with respect to "', that is only on the restriction to SL n (K). In particular it follows that the decomposition numbers [SJ-L : D A] also only depend on the equivalence class. 3.4. Suppose all A E A +(n, r) are p-regular. Then it follows from 3.2 (and using Ringel's equivalence) that the category of K Sr / In -modules which have filtrations by dual Specht modules (SA)* for A E A +(n, r) is contravariantly equivalent to the category 1's(6.) of S-modules with Weyl filtration. Correction. Some statements in the introduction of [E I; p.124] do not agree with what is proved. (1) Equivalence between categories of S(n, r)-modules with 6.-filtration, all 6. having p-regular partitions as weights, and the corresponding categories of modules with Specht filtration. It should have said contravariant equivalence (as it follows from 4.3(b».
(2) The category of modules with Specht filtration. Proofs are given for the category of K Sr / In but not of K Sr; this should have been said in the introduction. The same in 4.4, the second Theorem. Actually, when p > 2 the more general statement with K Sr (instead of the quotient algebra) is true; it is proved in [CPS2]; see also [03, section 4.7]. For p = 2 it is false.
Representations of GL n (K) and Symmetric Groups
75
4. Decomposition Numbers 4.1. One of the basic problems in modular representation theory is that of understanding the decomposition numbers. In the case of symmetric groups these are the composition multiplicities [SJl : D A] of the Specht modules. There is a similar problem for representations of GL n (K), namely the problem of understanding composition multiplicities of the Weyl modules [.6..(J.L) : L(A)]. This is considered to be difficult; and an answer is suggested by the Lusztig conjecture for type A. The following shows that both problems are the same:
Theorem. Let n be fixed. Then the composition multiplicities ([.6..(J.L) : L(A)], A E A +(n)}
are the same as the decomposition numbers
if
r varies.
It was proved by James [13 ] that the decomposition numbers for Sn are composition multiplicites of Weyl modules, but the converse does not follow if one takes n = r. We will explain the proof of the Theorem. First, composition factors of Weyl modules imply decomposition numbers, this follows directly from 3.4 if one specialises to A p-regular. To show that decomposition numbers for Sr imply composition factors of Weyl modules, let 8 be the partition (n - I, n - 2, ... , 2, 1, 0) with n parts. For a partition A of r let teA) = pA + (p - 1)8, this is a partition of t(r) := pr + G)(p -1). We use [Ja, 11.3.19] (see also 2.7), and a special case of 2.6 and have .6..(J.L/ ® .6..((p - 1)8) ~ .6..(t(J.L)), T(A)F 0 .6..((p - 1)8) ~ T(t(A)).
Consequently we get [T(t(A)) : .6..(t(J.L))] = [T(A) : .6..(J.L)]
The important fact is that t (A) is always p-regular. Applying 3.4 shows that [Sr Jl : D rA ] = [.6..(J.L') : L(A')] and the statement is proved.
76
Karin Erdmann
4.2.
One will conclude that the problem of understanding decomposition numbers is in fact hard. Also, bearing in mind the Lusztig conjecture, one may not expect to find easy closed formulae. Nevertheless there may be interesting recursive characterizations.
4.3. For the case when n = 2, G. D. James discovered an algorithm for the decomposition numbers [Jal, 2]. There is a short proof due to Donkin, using the results on twisted tensor products of canonical modules (see [Ell). It is a consequence of the properties we have listed in 2.8; we will now describe a variation of the algorithm. We work with SL2(K) and we shall use 2.8. Recall (see 3.3.1) that for a partition with two parts, [SJl : D A ] = [T(s) : ~(m)] if s = Al - A2 and m = J-LI - J-L2. We define therefore the generating polynomial corresponding to the column for D A in the decomposition matrix by m fs (z) = dmsz
L
m~O
A
where d ms = [SJl : D ], as above. The known properties of Specht modules translate into d ss = I and dms =1= 0 implies m :::: sand m == s (mod 2). Moreover the facts on T (s) listed in 2.8 translate directly in the following.
4.3.1. (a) For 0 :::: s :::: p - 1 we have !v(z) = zs. (b) fkp+(p-1)(z) = Zp-I fk(ZP). Moreover fkp+i (z) = fp+i (z) fk-I (zP) for 0 :::: i :::: p - 2. (c) We have fp+i(Z) = Zp+i
+ Zp-2-i
for 0:::: i :::: p - 2.
By a straightforward induction one deduces the following
'L7
Proposition. Let 0 :::: s with p-adic expansion of theform s = -1 + = I di pt j where 0 :::: tl < ... tk and 1 :::: di :::: p - 1. Then we have the factorization for !v as k-I i fs (z) = ZU fp+d j -I (zpr )
n i=1
where u is such that the total degree of fs becomes s. This is a polynomial in which all coefficients are 0 or 1. The integers m such that dsm = 1 are given in terms of the p-adic expansion of s. If one spells this out explicitly one obtains the formulation given by James.
Representations of GL n (K) and Symmetric Groups
77
Example. Let p = 2. Then only di = 1 occurs and f p +d\-1 = Z2 + 1. For example if s = 8 then fs(z) = Z6(Z2 + 1). Accordingly, there are only two decomposition numbers equal to 1 in the corresponding column.
5. Dimensions of Simple Modules for Symmetric Groups 5.1. Over fields of characteristic zero the dimensions of the simple modules for the symmetric groups are given by the hook formula. On the other hand, for prime characteristic the dimensions of the simple modules are not known in general. We have seen that dim D A is equal to the multiplicity of T (A) as a direct summand of E®r, for A E At(n, r). For such A, the module T(A)0E is then a direct sum of certain T (f-L) for f-L (n, r + 1); and in cases when one has information about the multiplicities of its direct summands then one gets some recursive formulae. This suggests to study generating functions for dim D A , or equivalently the multiplicity of T(A) as a direct summand of E®r , which we denote by {E®r : T (A) }. There are results for n = 2 and arbitrary partitions with two parts [E2]; and also for the class of partitions A with Al - An ::::: p - n, where n < p [Ma]. We will now describe these results.
EAt
5.2. There is a natural way to introduce generating functions for dim D A ; here we work with SLn(K). We fix an integer n, and we consider At(n), the set of p-regular partitions with at most n parts. Consider the equivalence classes under "" as defined in 2.6. Each equivalence class contains a unique partition A with An = O. Take such A, suppose IAI = c + tn where o : : : c ::::: n - 1. Then the class consists precisely of the partitions of the form A+ a multiple of (1 n). We label them as A(k) such that A(k) is a partition of c + kn, then k :::: t. Define the generation function FA (z)
=
L dim
DA(k)
l
k
Then also dim
DA(k)
=
(taking the restriction to SL n (K) ).
{E®(c+k n ) : T(A)}
78
Karin Erdmann
5.3. Assuming n < p, in [Ma], O. Mathieu studies partitions with at most n parts, which belong to the set }In := {A E A;(n) : Al - An :::: p - n}
One main result is a combinatorical description of dim D A for A E }In which is a generalization of the hook formula in characteristic zero. Moreover, he proves
Theorem. Let A E }In, then FA(z) is rational. We will describe the idea of the proof. We work with SL n (K) and assume for A E }In thatn An = O. Let Kb('J) be the Grothendieck ring whose underlying abelian group has basis ([T(A)) : A E A +(n), An = O}
and where the multiplication is induced by the tensor product (see 2.4). Fix some congruence 0 :::: c :::: n - 1, and let V C be the subgroup of the above Grothendieck ring generated by all [T(A)] such that IAI == c(mod n). Tensoring with E0 n induces a linear map m : V C ~ VC. This is not so helpful in general since V C has infinite rank. Let Q be the additive category generated by the T(A) whose dimension is divisible by p; the important fact is that Q Q9 'J S; Q. (This holds more generally for group representations; for the case of finite groups, this is known for some time, see for example [B]). Moreover, for n < p, the module T(A) does not belong to Q if and only if A E }In' In particular the number of such modules is finite. The above Z-linear map m induces therefore a map M : Va ~ Va where Va is the quotient of V C modulo the span ofthe [T(A)) which belong to Q. We write" == (mod Q)" when working in this quotient. If we denote the k-th coefficient of FA by d~c+nk) and write the generating functions FA for A E }J~ as a vector with respect to some fixed order and take M in form of a matrix then we have 00
(FA»..
00
= L(d~c+nk)>..i = L(Mz)k(d~C)>... k=O
k=O
Over C there is an invertible matrix P such that p- I M P has Jordan canonical form, so we write p-I M P = D + N where D is diagonal and N is nilpotent such that N, D commute. We have
Representations of GL n (K) and Symmetric Groups
79
There is some m ::: 1 such that N S = a for s ::: m; and Dz is a diagonal matrix, say with diagonal entries alZ, ... anz. Then the above sum is equal to m-I
L(Nz)S Ds 5=0
where D s is the diagonal matrix with diagonal entries (l - aiz)-(s+1) for 1 ::: i ::: n. Since P is a matrix with entries in C it follows that each FA is rational.
5.3.1. Example. Let n = 3 and p = 5, we consider the congruence c = O. The set ~3 contains precisely two partitions y with Iy I == a (mod 3), namely A
= (2, 1),
JJ-
= 0.
We work over SL3(K), so we write the basis as [T[I, 1]] and [T[O, 0]]. One needs the linear transformation induced by tensoring with E®3. First, one has E®3
== 2T[I,
1] EB T[O, 0] (mod Q)
This follows from 3.1, using that for p = 5 the group algebra of 33 is semi-simple, and dim S(2, I) = 2, dim S(l3) = 1. Next we claim that E®6
== 8T[I,
1] EB 5T[0, 0] (mod Q). 3
The coefficient of T[O, 0] is dim D(2 ). Now the partition (2 3 ) belongs to 3 3 a 5-block of defect zero, so D(2 ) = S(2 ) of dimension 5. Moreover, the coefficient of T[I, 1] is equal to the dimension of D(3,2,1). This belongs to the block of defect I and by well-known properties of such blocks for symmetric groups one finds that the dimension is 8. These two identities show that T[I, 1] 0 T[l, 1] == T[I, 1] EB T[O, 0]. Moreover, T[O,O] ~ K. Hence tensoring with T[ 1, 1] induces the linear transformation with matrix
(
1 1
o'; and it follows that in this case M= (~
~)
on V
One finds that M _ P
-
i)
(aa 0) l
a2
p_ 1
80
Karin Erdmann
where al = 2+.J5 and a2 = 2-.J5.Inparticular N = 0 here. Moreoverwe can take for P a matrix whose columns are eigenvectors of M, for example
(1 +2.J5 I-2.J5) .
P= One calculates
P
((l - ~IZ)-1
o
)
p- 1 =
(l-a2z)-1
1
1-4z-z 2
(1 - z 2z
2z
1 - 3z
)
.
The vector to be used is (~) and we get FA. (z)
= 1_
2z 4z _ z2'
FJi. (z)
=
1 - 3z 1 - 4z _ z2
5.4. We consider now the case when n = 2. The elements of smallest degree in the equivalence classes are ofthe form A = (s, 0) for s ~ O. In the notation of [E2] we had defined d;r) = {E0 r : T(s)} = dim D«s+r)/2,(s-r)/2) and D s (z) = L~o d;r) zr which is almost the same as FA.. For n = 2 the tensor products can be calculated explicitly, using [D 1]. We shall give an ill ustration. Consider an integer s of the form s = kp + i where 0 < i ::: p - 2. Recall that T(l) = E. Using 2.8 we have T(s) ® E ~ T(k -
1/ ® T(p + i) ® T(l).
One shows easily that for 0 < i ::: p - 2 T(p
+ i) ® T(l)
~ T(p
+ i + 1) EB T(p + i
- 1).
Hence if i < p - 2 then it follows that T(s) ® E
~
T(s
+ 1) EB T(s -
1) .
We note that this also holds for characteristic zero and any s; in this case T(s) ~ ~(s) ~ L(s) which is well-known. The situation is complicated by the behaviour for i = P - 2. Then there is a summand T(k - 1)F ® T(p + P - 1) which is by 2.8 isomorphic to T(k - 1)F ® T(l)F ® T(p - I) ~ (T(k - 1) ® T(l))F ® T(p - 1).
One needs an induction and one will get large sums, depending on the p-adic expansion of k - 1.
Representations of GL n (K) and Symmetric Groups
81
We return to the easy case; the identity T(s) 0 E ~ T(s
+ 1) EEl T(s -
1)
which holds for 0 :::: s < p - 2 and translates into recursive formulae for the first few generation functions D s = zD~-1
+ ZDs+I,
0:::: s :::: p - 3
Moreover, one has D p -2 = ZD p -3. This can be written as a system of linear equations
z
-1
-1
z
0 -1
0
o
-1
Z
-1
0
-1 0
o
z
0 -1
-1
Z
Do DI D2
1 0 0
D p -3 D p -2
0 0
One wants to invert this matrix and find an iteration. This can be done with Chebyshev polynomials. One defines the Chebyshev polynomials of the second kind, Vn-I (x) of degree n - 1 to be sin(ne) V I () x =-nsinCe) where x = cos(e). For example, Vo(x) = 1, VI (x) = 2x and V2(X) = 4x 2 - 1 (see for example [RiD. This can also be expressed as a determinant. Namely Vn-I (x) is equal to the determinant of the matrix 2x
-1
-1
0
2x
0 -1
0
-1
2x
-1
0
0
-1 0
2x
0 -1
-1
2x
of size n -1 x n -1. To see this, one checks that the V 's satisfy the polynomial equation Vn(x) - 2XVn_I(X)
(see for example [Ri, Exercise 1.2.15]).
+ V n-2(X) =
0
82
Karin Erdmann
This is why it occurs here. The above matrix has therefore determinant Zp-I Up_1 (l/2z). In [E2l the following result is proved. We write the p-adic expansion of s in the form k S
= -1
+ Ldiptj i=1
where 1 .:::: di .:::: P - 1 and 0.:::: tl < t2 < ... < tk. Then Theorem. We have _ 1
n k
Ds(z) - Z i=1
U(p-dj)p1i -I (l/2z) Upli+l_ 1 (l/2z)
where Um-I (z) is the Chebyshev polynomial of degree m - 1. In particular D s is a rational function.
5.4.1. Example. Consider the case s = O. Then s = -1 t = O. By the above theorem
+ pO,
so d = 1 and
Do (z) = ~ _U-,-(p_-_I)-_I_(l_1_2z_) Z Up_1 (l/2z) (a) Assume first that p = 2, then we get Do(z) = 1. On the other hand, we know that T (0) occurs in E0 r only for r = 0 and then with multiplicity 1. Alternatively the partition (k, k) is 2-singular except when k = O. (b) Now assume p = 3, then we get Do(z) = (l-Z2)-1 which is equal to L~O z2n. The coefficient of z2n is the dimension of D(n,n) and it is known that D(n,n) is the alternating representation for p = 3. For higher degrees, the above form is not so easy to use. In general we have the following result to determine the coefficients. Theorem. Let s be as above. Then the coefficient of zn+s in the generating function D s (z) is zero for n odd, and if n = 2m then it is equal to the coefficient of qm in the power series expansion of
s(q)(l
+ q)2m+s
Representations of GL n (K) and Symmetric Groups
83
where we define
n k
cl>s(q) = (l - q)
i=l
(l _
q(p-di)pfi) HI'
(l-qP'
)
This is proved by using a contour integral. For details we refer to [E2].
References [B]
D. Benson, Modular representation theory: new trends and methods, Lecture Notes in Math. 1081, Springer-Verlag, 1981.
[CL]
R. Carter, G. Lusztig, On the modular representations of the general linear and symmetric groups, Math. Z. 136 (1974), 193-242.
[CPS]
E. Cline, B. Parshall, L. L. Scott, Finite-dimensional algebras and highest weight categories, J. Reine Angew. Math. 391 (1988),85-99.
[CPS2]
E. Cline, B. Parshall, L. L. Scott, Stratifying endomorphism algebras, Mem. Amer. Math. Soc. 59 I, Vol. 124, 1996.
[CP]
C. de Concini, C. Procesi, A characteristic-free approach to invariant theory, Adv. Math. 21 (1976),330-354.
[DR]
Y. Dlab, C. M. Ringel, Quasi-hereditary algebras, Illinois J. Math. 33 (1989), 280-291.
[D I]
S. Donkin, Rational representations of algebraic groups, Lecture Notes in Math. 1140, Springer-Verlag, 1985.
[D2]
S. Donkin, On tilting modules for algebraic groups, Math. Z. 212 (1993), 39-60.
[D3]
S. Donkin, The q -Schur algebra, preprint 1996.
[El]
K. Erdmann, Symmetric groups and quasi-hereditary algebras, in: Finitedimensional algebras and related topics (Y. Dlab and L. L. Scott, eds.), Kluwer, 1994, 123-161.
[E2]
K. Erdmann, Tensor products and dimensions of simple modules for symmetric groups, Manuscripta Math. 88 (1995), 357-386.
[E3]
K. Erdmann, Decomposition numbers for symmetric groups and composition factors ofWeyl modules, J. Algebra 180 (1996), 316-320.
[G]
J. A. Green, Polynomial representations of GL n , Lecture Notes in Math. 830, Springer-Verlag, 1980.
[G2]
J. A. Green, Combinatorics and the Schur algebra, J. Pure Appl. Algebra 88 (1993),89-106.
[JK]
G. D. James, A. Kerber, The representation theory of the symmetric group, Enc. of Math. 16, Addison & Wesley, 1981.
84
Karin Erdmann
[11]
G. D. James, Representations of the symmetric groups over the field of characteristic 2, 1. Algebra 38 (1976), 280-308.
[12]
G. D. James, On the decomposition matrices of the symmetric groups I, J. Algebra 43 (1976),42-44.
[13]
G. D. James, The decomposition of tensors over fields of prime characteristic, Math. Z. 172 (1980),161-178.
[Ja]
1. C. Jantzen, Representations of Algebraic Groups, Academic Press, 1987.
[M]
S. Martin, Schur algebras and representation theory, Cambridge University Press, 1993. O. Mathieu, On the dimension of some modular irreducible representations of the symmetric group, Lett. Math. Phys. 38 (1996), 23-32.
[Ma] [P]
B. Parshall, Finite-dimensional algebras and algebraic groups, Contemp. Math. 82 (1989), 97-114.
[R]
C. M. Ringel, The category of modules with good filtrations over a quasihereditary algebra has almost split sequences, Math. Z. 208 (1990), 209-225. T. J. Rivlin, Chebyshev polynomials: From Approximation Theory to Algebra and Number Theory (2nd ed.), Wiley, 1990. 1. Schur, Ober die rationalen Darstellungen der allgemeinen linearen Gruppe (1927), in: 1. Schur, Gesammelte Abhandlungen III, 68-85, Springer-Verlag, Berlin, 1973. S. Xanthopolous, On a question of Verma about indecomposable representations of algebraic groups and their Lie algebras, PhD Thesis, London, 1992.
[Ri] [S]
[X]
Mathematical Institute 24-29 St. Giles Oxford OX1 3LB England Email: [email protected]
On Extended Block Induction and Brauer's Third Main Theorem G. Huang
1. Introduction As generalizations of Brauer correspondence in the study of the modular representation theory of finite groups, four different definitions of block induction have been proposed and used. They are Brauer induction [Fe], p-regular induction [B11], extended induction [Wh] and Alperin-Burry [AI]. The first three are defined in terms of central characters, while Alperin-Burry induction is defined in terms of module-theoretic properties of block ideals. Although the four definitions are different, they are closely related. Among them, extended block induction is the weakest. We examine whether properties of the other types of induction also hold for extended induction. For instance, we found a class of infinitely many examples to show that the transitivity which holds along blocks under Brauer induction and p-regular induction does not hold under extended induction. We also give some p-Iocal characterizations of extended block induction, discuss Brauer's Third Main Theorem in the extended induction sense, and establish an affirmative result on an aspect of Brauer's Third Main Theorem for p-solvable groups. That is, for p-solvable groups, principal blocks always induce in the extended sense to principal blocks if the induction is defined. Now let us fix our notation and state some definitions. Let G be a finite group. Let p be a prime number and let (F, R, K) be a p-modular system. Assume that F and K are splitting fields for every subgroup of G. Let (9 be R or F and let H be a subgroup of G. The Brauer map Br~ (or simply BrH): (9G -+ (9H is defined by Br~(x) = x if x E Hand 0 if x E G - H for all x E G and is linearly extended to (9G. If A is a G-algebra, let A H denote the set of H -fixed points. Let Gp (resp. G pi) denote the set of p- (resp. pi -) elements of G. For a set S ~ G, let F S denote the subspace spanned by S, and Z F S the intersection of F S with the center ZFG of FG. If b is an (9G-block, where "block" refers to block
86
G. Huang
idempotent, then Ab denotes the central character of FG associated with block b. If e = LrjlEIIT(G) arjl
j)G is defined and equal to b (or j) induces to b in Brauer's sense) {:=> Ab 0 (Br~)zFG = Ab [Fe].
(2)
j)regG is defined and equal to b (or j) p-regularly induces to b) Ab- 0 (BrGH)ZFG p I = (Ab)zFG p I [Bi].
(3)
j)ext G {:=>
(4)
{:=>
is defined and equal to b (or j) induces to b in the extended sense) 0 (Br~)(b') = Ab(b' ). for all blocks b' of ()G [Wh].
Ab
j)(G) is defined and equal to b (or j) induces to b in Alperin-Burry's sense) {:=> there is only one ()G-block b which covers j) [AI]. (We say that an ()G-block b covers j) if j)()H is isomorphic to a direct summand of b()GHxH as ()(H x H)-modules.)
The four definitions are not equivalent, but if any two of the four types of block induction are defined, they are the same. In general we have j)G =b=>j)regG =b=>j)extG =b
and pG)
= b => j)ext G = b.
There are examples showing that it is not true in general to reverse any of the above arrows or get any further implications among these four definitions. [BI2]
2. Strong Covering and its Characterizations In order to investigate extended block induction more closely, we would like to define extended induction in a similar way to the definition of Alperin-Burry induction. Hence we introduce the definition of strong covering of blocks which is indeed a strong case of covering of blocks. (See the remark following Proposition 2.2.) Definition 2.1. Let G be a finite group and H a subgroup of G. Let j) be an () H -block and b a central idempotent of ()G. Then we say b strongly covers j) if Ab 0 (Br~ )(b) =1= O.
On Extended Block Induction
87
By definition, we can see that h induces to a block b of G in the extended sense if and only if b is the unique block which strongly covers h. Also for a given block h of H there is at least a block b of G strongly covering h since Ab 0 (Br~)(1) = 1 -=I O. The following properties are useful when we study strong covering. (See also Prop 1.4 in [Whl)
Proposition 2.2. Let H be a subgroup of G, band b be OG, OH-blocks resp. Then the following are equivalent. (1)
b strongly covers b.
(2)
h(Br~)(b)
f/. J(ZOH).
(3) Let f : bOH ~ (bOG)HxH be the O[H x H]-homomorphismdefined by f(x) = bx and g : (bOG)HxH ~ hOH be defined by g(y) = b Br~(y); then go f is an automorphism of bOH. (4) Let f)
E
Irr(b). Then f)c(1)p = f)b(1)p.
Remark. The equivalence of (1) and (3) shows that strong covering implies general covering. So Alperin-Burry induction implies extended induction. The equivalence of (l) and (4) is useful when we compute examples. H. Ellers, G. Hill and Y. Fan have given characterizations of Alperin-Burry induction in terms of p-local subgroups [El]. The following theorem, part of which is analogous to Fan's theorem, gives some p-Iocal characterizations for extended induction.
Theorem 2.3. Let b be a block of F H with defect group D and b a central .
-
H
-
-
C
ldempotentof FG. Denote B := BrCH(D)(b) = Li bi and B := BrCG(D)(b), where hi'S are blocks in F DCH(D). (By Brauer's First Main Theorem, a block of F N C (D).) Then the following are equivalent.
iJ
is
(i)
b strongly covers h.
(ii)
Br~(b)y = h for some y
(iii)
iJ Brg~~~~ (B)z = iJ forsome
(iv)
B in Nc(D) strongly covers B in NH(D).
(v)
Bin DCc(D) (or Cc(D)) strongly covers hi, for all i, in DCH(D) (or CH(D) resp.).
E
b(FH)H. Z E FCH(D)NH(D).
88
G. Huang
(vi)
B in DCG(D) (or CG(D) ) strongly covers bi,forsome i in DCH(D) (or CH(D) resp.).
(vii) There exists x E FCH(D)CH(D) such that
B Brg~~~~(B)x
=
B.
(viii) For any i, there is a Zi E FCH(D)CH(D) such that hi Brg~~~~(B)Zi = bi.
(ix)
There is an i, such that -
CG(D)
-
bi BrCH(D)(B)Zi = bi for some Zi E FCH(D)CH(D). (x)
r(B) in DCG(D)/ D strongly covers r(hd in DCH(D)/ D for any i (or some i), where r is the homomorphism induced by the canonical map DCG(D) --+ DCGCD)/ D.
In the case where the subgroup is normal, strong covering has a clear relation to general covering.
Proposition 2.4. Let H be a normal subgroup of G. Let Band b be blocks of G and H respectively. Then B strongly covers b if and only if B covers band B is weakly regular with respect to H. As a corollary, we can get a result which is a special case of Corollary 4 in [BI2].
Corollary 2.5. Let H be a normal subgroup of G, and B, b blocks of G, H respectively. Then breg G = B
if and only if bext G
= B.
3. Brauer's Third Main Theorem under Extended Induction For p-regular induction, therefore Brauer induction, the following result is known as Brauer's Third Main Theorem: Theorem. Let H ::: G, B be an R G -block and b an RH-block such that b G = B (or b regG = B). Then B is the principal block of RG if and only if b is the principal block of RH. (See [B] Corollary 2). For Alperin-Burry induction, it is easy to see that the principal block always induces to the principal block if the induction is defined. But the property no longer holds for extended induction (see [W] Example 2.10). Fortunately, for
On Extended Block Induction
89
p-solvable groups, extended induction does have the property stated above for Alperin-Burry induction.
Theorem 3.1. Let G be a p-solvable group, K :::: G. Then the principal block Ba of G strongly covers the principal block bo of K. Hence if b~xt G is defined then b xt G = Ba.
o
Remark. Brauer's Third Main Theorem describes the behavior of principal blocks under the Brauer correspondence of blocks. The theorem above shows that, for p-solvable groups, principal blocks induce to principal blocks in the extended sense if the induction is defined. However, even in p-solvable groups, principal blocks may be induced in the extended sense by some non-principal blocks of subgroups. Example 3.2. Let k generator v. Let
= GF(9). 2
H
= ((
v 0
Let k# be the multiplicative group of k with
0)
v-2 ),
Then H ~ 24 and ISI = 9. Let G = S· H :::: SL(2, 9). Then G is a solvable group of order 36. Let p = 3. Then G has two 3-blocks: Ba and Bl. Let fJ E Irr(H) such that fJ(
(±l 0) 0
±1
) = 1 and fJ(
(v±2 0
-1. Let btJ be the block to which fJ belongs. Then bfJ is not the principal block of H and using Proposition 2.2 we can verify that b~xt G = Ba. This gives an example where a non-principal block induces in the extended sense to the principal block of a solvable group. Therefore extended induction is properly weaker than p-regular induction, even in solvable groups.
4. A Class of Infinitely Many Counterexamples to the Transitivity of Extended Induction It is easy to see by definition that the transitivity property of Brauer block induction and p-regular induction holds along blocks. That is, if T :::: H :::: G is a chain of three groups, and E, band B are blocks of T, H, and G respectively, such that E induces to b, and b induces to B in Brauer's sense (or p -regularly), then the induction of Ein G is defined and equal to B. Ellers
90
G. Huang
showed that Alperin-Burry induction does not have the transitivity property(see [E]). However, general covering obviously has the transitivity property along blocks, therefore, if the induction of h in G is defined, then it must be B. Now we would like to know whether the transitivity property holds under extended block induction. Unfortunately, the answer is negative. Let p = 2, n > 3 be an even integer. Let G = GL(n, 2) and H =
[~
GL(n
~ 1,2)] ..::. GL(n, 2).
Let GF(2n - 1)# = (x). Then under the multiplicative action on the vector space GF(2 n - 1), x can be regarded as an element of H of order 2n - 1 - 1. Let q be a Zsigmondy prime number with respect to (2, n - 1). Namely, q is a divisor of 2 n - 1 - I but not a divisor of any 2' - 1 with r < n - 1. Let T = (z} E Sylq ((x). Then we have a chain of three groups:
T
hexl H
= b;
(ii) b ext G = Ba; and (iii)
hexl G
= B.
Remark. For a general odd prime number p with some modification we have similar results.
References [AI]
J. Alperin, Local Representation Theory, Cambridge University Press, Cam-
bridge, 1986. [Bll]
H. Blau, On block induction, 11, J. Algebra 159 (1993) 477-487.
[Bl2]
H. Blau, On block induction, Ill, J. Algebra 169 (1994) 648-654.
[El]
H. Ellers and G. Hill, A Local characterization of the Alperin-Burry Correspondence, J. Algebra 150 (1992) 427-434.
[Fa]
Y. Fan, Local characterizations of block covers and their applications, J. Algebra 152 (1992) 397-416.
[Fe]
W. Feit, Representation Theory of Finite Groups, North-Holland, Amsterdam, 1982.
On Extended Block Induction
[Wh]
W. Wheeler, Extended block induction, J. London Math. Soc. 49 (1994) 73-82.
Resnet Systems 5451 North East River Road Chicago IL 60656 U.S.A.
91
On Blocks and Source Algebras for the Double Covers of the Symmetric Groups Radha Kessar
This article is a report on some results exhibiting equivalences between block algebras of blocks of finite groups with a common defect group. Most of its contents were presented at the 1995 Ohio State Representation Theory Conference held in honor of Waiter Feit on the occasion of his 65th birthday.
1. Source Algebras, Their Definitions and Some Properties Throughout this article, let p denote a prime number, and let (<9, K, k) be a p-modular system, i.e. <9 is a complete discrete valuation ring with maximal ideal J(<9), k = <9/J(<9) is its residue field with characteristic p that we assume is algebraically closed and K is its field of quotients with characteristic zero. If a finite group G acts on an <9-algebra A (we assume always that A is associative, unitary, finitely generated over <9, and either <9- free or annihilated by J (<9) ), we denote by g a the image ofthe element a of A under the action of the element g of G; and for a subgroup H of G, A H will denote the space of elements of A that are left fixed under the action of H. An <9-algebra A is an interior G-algebra, where G is a finite group, if there is a homomorphism from G into A *, the group of units of A. If A and B are interior G -algebras and if
= ljf(g)f(a) and f(acjJ(g» = f(a)ljf(g) Vg
E
G, a E A.
A block b of a finite group G is a primitive idempotent of the center of the group algebra <9G; the algebra <9Gb is called the block algebra of b.
94
Radha Kessar
Let H be a subgroup of G. The Brauer map of OG with respect to H, denoted Br H is the natural map from (OG) H onto the quotient (OG)H I
L
Tr~ ((OG)R)
+ J(O)(OG)H.
RcH
Here Tr~ ((OG)R) is the ideal in (OG)H which consists of all elements of the type LXER/ H x a, where a is an element of (OG)R. A defect group P of b is maximal among the subgroups of G such that Br p (b) is not zero. A source algebra of b is an algebra of the form WGbi, where i is a primitive idempotent of the algebra (OGb)P for which Brp(i) is not zero. A source algebra iOGbi of b is naturally an interior P-algebra via the homomorphism which maps an element x of P to the element ix of WGi. Actually, the notion of source algebra is not restricted to block algebras of finite groups but exists for any interior G-algebra [PI]. In some sense, the block algebra OGb can be " almost" computed from the knowledge of its source algebra as an interior P -algebra. More precisely, we have the following theorem which is due to Puig. Theorem 1.1 ([PI], [P2]). 1.
The algebras OGb and iOGbi are Morita equivalent, i.e. they have equivalent module categories.
2.
The matrix of generalized decomposition numbers of the block b of G, its local category, the vertices and sources of its indecomposable modules, are all determined by the isomorphism class (as interior P -algebra) of the source algebra iOGbi of b.
In the following example, we explicitly describe the source algebras for certain blocks. This example is taken from [Br]. Example 1.2. Let r be a prime number such that p divides r - 1, (an) be an infinite sequence of integers none of which is divisible by m be an integer such that m < p. We put G n = Glm(r an ) and we by B n the subgroup of G n consisting of the upper triangular matrices We put
and let p. Let denote in G n .
Let P be a Sylow p subgroup of G n . Then P is isomorphic to a direct pr oduct of m copies of the Sylow p subgroup of ('Ill r'll) x. Also, in is a
On Blocks and Source Algebras
95
primitive idempotent of (OGn)p and the algebra inOGi n is a source algebra for the principal block of G n . On the other hand, inOGi n is isomorphic to the algebra O(P x Nc, (P)/CCI (P)).
2. Puig's Conjecture One of the main problems of block theory is to classify the block invariants given the isomorphism type of its defect groups[A]. As is seen from Theorem 1.1 above, the classification of source algebras, given the defect groups of the corresponding blocks, would be quite a step forward in this effort. The following conjecture was announced by Lluis Puig at the Group Theory Conference held at Oberwolfach in 1982. Let P denote a fixed non-identity p group. Conjecture 2.1 (Puig). There are only finitely many isomorphism classes of interior P -a1gebras which are source algebras for blocks of finite groups having P as a defect group. The above conjecture has been verified in its entirety for the family of cyclic p-groups. This result is due to Markus Lincklemann; in fact he gives necessary
and sufficient conditions for two blocks with a common cyclic defect group to have isomorphic source algebras [L]. In a slightly different vein, the conjecture has been verified in some cases if instead of considering blocks of any finite group, only blocks of certain infinite families of groups are considered. For example, if we restrict ourselves to blocks of p-solvable groups, then Puig's conjecture is known to hold; the proof uses the classification of the finite simple groups [H].
2.2 The groups Sn. It is well known [JJ that the ordinary irreducible characters of the symmetric group Sn on n letters are in one to one correspondence with the set of partitions of n and that two irreducible characters of X and rJ of Sn belong to the same p block of Sn if and only if the partitions to which they correspond have the same p-cores. Further, the Murnaghan-Nakayama formula calculates the restriction of an irreducible character X of Sn to a subgroup Srn in terms of removals of hooks from the corresponding partitions. Using this machinery, Joanna Scopes [Se] found a suitable criterion which, if satisfied by two p-blocks , b of Sn and c of Srn, would force the block algebras kSnb and kSrnc to be Morita equivalent. The transitive extension of this criterion subdivides the blocks of the symmetric groups into finitely many families; as a consequence, she proved that Donovan's Conjecture [AJ holds for the blocks of the symmetric groups, namely that there are only finitely
96
Radha Kessar
many Morita equivalence classes of blocks of the symmetric groups that have P as a defect group. Later, Puig showed that the equivalence demonstrated by Scopes could actually be obtained through an isomorphism of the corresponding source algebras [P31 He proved the following theorem.
Theorem 2.2.1. Let band c be blocks of Sn and Sm respectively with defect group P. If band c satisfy Scopes criterion of equivalence then there exists an isomorphism (1.1)
as interior Sm x Sn-m algebras. Further, this isomorphism induces an interior P -algebra isomorphism between a source algebra of b and a source algebra of c.
Corollary 2.2.2. Puig's conjecture holds for the blocks of the symmetric groups. Though the results of Scopes and Puig are essentially the same, the techniques they employ for the proof are quite different. Scopes' proof relies on the bridge between both the ordinary and the modular character theory of the symmetric groups and the theory of partitions whereas in Puig's proof only the correspondence at the level of ordinary irreducible characters is used. It would seem therefore that Puig's method would lend itself more easily to generalization. In fact, Puig believes that his method will yield similar comparisons between blocks in any family of finite groups where the ordinary irreducible characters allow a parameterization similar to that of the symmetric groups. Let us try to outline Puig's general idea. For a finite group G and a block b of G, let Irr( G, b) denote the set of ordinary irreducible characters of G that belong to G. For the purpose of later applying this method to other families of groups, let us use G n to denote Sn. We note that if m < n, Gm may naturally be considered a subgroup of G n and we assume this inclusion. The proof of Theorem 2.2.1 is roughly as follows. I
Given a block of b of G n for n sufficiently large, the group Gm (with C of it are chosen such that there is an integer rand there is a 1-1 correspondence (given by Xn --* Xm) between Irr( G n , b) and Irr( Gm, c) such that for any element Xn of Irr(G n, b), the restriction of Xn to the block c of Gm is equal to r copies of Xm' This easily yields the rank comparison m < n) and a block
rankCJ(cb()Gnbc) = r 2 rankCJ(()Gmc).
On Blocks and Source Algebras
11
97
Since Brp(b) and Brp(c) are respectively blocks of kCGn(P)jZ(P) and kCG m(P)jZ(P)
of defect zero, Brp(c) Brp(b)kCGn(P)jZ(P) Brp(c) and Br p(c)kCG n (P)jZ(P) Br p(c) are simple algebras, and it is easy to analyze the structure of the fixed point algebra (Brp(c) Brp(b)kCGn(P)jZ(P) Brp(c))NGmCP). This information is then "lifted" to give information about the idempotents of the fixed point algebra (cbr:JGnc)G m. It turns out that there is exactly one conjugacy class of primitive idempotents in (cbr:JGnc)G m whose elements are not in the kernel of the Brauer map with respect to P. III Let S be a unitary r:J simple subalgebra of (cbr:JGnc)G m. Then we have a decomposition (cbr:JGnc) :::: S 0CJ CcbCJGnc(S).
The rank comparison in Step I along with a lemma ofPuig [PS, Proposition 3.8] is then used to show that r:JGmc and CcbCJGnc(S) are isomorphic as interior Gm algebras, thus yielding the result. This approach, if modified slightly, can be applied to the covering groups of Sn as we describe below. 2.3 The groups Sn and An. A presentation of the group Sn via generators and relations can be found in [HH]. We have an exact sequence: -
B
1 ---+ (z) ---+ Sn ---+ Sn ---+ 1 where z is a central involution of Sn. The group Sn is a covering group for Sn for n > 3. Let An denote the inverse image, 0- 1 (An), of the alternating group An under 0, An is a double cover of An. Let us assume from now on that p is odd. Let e : r:JS n ---+ r:JS n denote the natural extension of 0 to the group algebra of Sn. Then the image e(b) of a block b of Sn (respectively An) is either 0 in which case b is called a faithful block of Sn (respectively An ) or (b) is a block of Sn (respectively An) in which case b is called a non-faithful block of Sn (respectively An). Also, if b is a faithful block of Sn with a non-trivial defect group, then b is also a block of r:JAn, under the natural inclusion of r:JAn in r:JSn . The faithful ordinary irreducible characters (irreducible characters belonging to faithful blocks) of Sn are parameterized by the set of strict partitions of
e
98
Radha Kessar
n; a strict partition A of n indexes a single character X).. of
Sn
if A is even and A indexes a pair of associate characters X).. and xf if A is odd [8]. If two characters X and 17 of Sn are in the same block of Sn, then the partitions by which they are indexed have the same p -bar core. Conversely, if the partitions indexing X and 17 have the same p-bar core, then either X and 17 are in the same p-block or X and 17 are associate characters and each of them belongs to a block of defect zero [Cl, [H], [0], A. O. Morris [Ml] [M2] has given a formula which relates the restriction of characters of Sn to Srn to the removal of bars from the corresponding partitions. Thus the situation for the double covers is quite similar to that of the symmetric groups. We indicate briefly how Puig's method for proving Theorem 2.2.1 must be modified for these groups. Let the group G n denote now Sn. Since the correspondence between the strict partitions of n and the ordinary faithful irreducible characters of n fails to distinguish between a character and its associate, a correct correspondence between the characters of G n in band the characters of c is obtained only through a finite sequence bl, bz ... b t of blocks (each with P as a defect group) of intermediate groups G nj with n > ni > ni+1 > m for 1 ::: i < t. In other words, in Step I we obtain a rank comparison between rank o (cab<9G nbac) and the algebras ranko (<9G rn c) , where a is the product of the bi 'so The analysis in Step II yields that there are two conjugacy classes of primitive idempotents in (cab<9G n ac)G m whose elements are not in the kernel of the Brauer map with respect to P; hence in step III we work with the algebra I cab<9G n acI where I is an idempotent, suitably chosen, of (cab<9G n ac)G m • In order to get the required result we need also to consider the action of An on (cab<9G nac)G m. This actually points to an analogous result for the groups An, and in fact we obtain the following theorem. Let 'B p denote the set of faithful blocks in the family of groups {Sn k:~o which have a defect group isomorphic to P. For b E 'B p, let Ibl = n if b is a block of Sn.
Theorem 2.3.1. 1.
There exists an integer N p with the following property: For any b in 'B p with n = jbl > N p there exists c in 'B p with m = lel < Ibl such that there is an idempotent I of (<9S n b)Sm for which
On Blocks and Source Algebras
99
as interior Sm aLgebras. In particuLar; there is an interior P -aLgebra isomorphism between a source aLgebra of b (considered as a bLock of Sn) and a source aLgebra of c (considered as a bLock of Sm ). The statem:nt of (I) remains valid on repLacing the group Sn by An and the group Sm by Am.
2.
As an immediate corollary of the above theorem and of Corollary 2.2.2 we get
Corollary 2.3.2. Puig's conjecture hoLds for the p-bLocks of the groups Sn, for every odd prime p. Corollary 2.3.3. Puig's conjecture hoLdsfor the faithfuL p -bLocks ofthe groups An, for every odd prime p. Remark 2.3.4. A result of Puig and Lincklemann on central p extensions proves Corollary 2.3.2 for the case p = 2 as a consequence of Corollary 2.2.2 Remark 2.3.5. For n > 3 and n i= 6, there is another central extension of Sn, usually denoted by Sn. However, the character theory of Sn may be described in exactly the same fashion as that of Sn [C, Section 3.9] and it is easy to see_that th~ results described in this section remain valid if we replace the group Sn by Sn.
References [A]
J. L. Alperin, Local representation theory, in: The Santa Cruz Conference on Finite Groups (B. Cooperstein and G. Mason, eds.), Proc. Symp. Pure Math. 37, Amer. Math. Soc., Providence, R.L, 1980, 369-375.
[B]
M. BroUl~, Theorie locale des blocs, in: Proceedings of the International Congress of Mathematicians, Berkeley, 1986, 360-368.
[C]
M. Cabanes, Local structure of the p-blocks of Sn, Math Z. 198 (1988), 519-543.
[H] [HH]
[J]
J. F. Humphreys, Blocks of projective representations ofthe symmetric groups, J. London Math. Soc. (2) 33 (1986), 441-452. P. N. Hoffman and J. F. Humphreys, Projective Representations of the Symmetric Groups, Oxford Mathematical Monographs, Oxford University Press, New York, 1992. G. D. James, The Representation Theory of the Symmetric Groups, Lecture Notes in Math. 682, Springer-Verlag, Berlin 1978.
100
Radha Kessar
[K]
Radha Kessar, Blocks and source algebras for the double covers of the symmetric and alternating groups, preprint, 1995.
[L]
Markus Lincklemann, The isomorphism problem for cyclic blocks and their source algebras, preprint, 1994.
[M 1]
A. O. Morris, The spin representation of the finite group, Proc. London Math. Soc. (3) 12 (1962), 55-76.
[M2]
A. O. Morris, On Q-functions,1. London Math. Soc. 37 (1962),445-455.
[0]
J. Olsson, On the p-blocks of the symmetric and alternating groups and their covering groups, 1. Algebra 128 (1989),188-213.
[P I]
L. Puig, Pointed groups and construction of characters, Math Z. 176 (1981), 358-369.
[P2]
L. Puig, Local fusions in block source algebras, 1. Algebra (1986),358-369.
[P3]
L. Puig, On Joanna Scopes' criterion of equivalence for blocks of symmetric groups, Algebra Colloq. I: 1 (1994),25-55.
[S]
I. Schur, Uber die Darstellung der symmetrischen und der alternierenden Gruppe durch gebrochene lineare Substitionen, J. Reine Angew. Math. 139 (1911), 155-250.
[Sc]
1. Scopes, Cartan matrices and Morita equivalence for blocks of the symmetric groups, J. Algebra 142 (1991), 441-455.
Department of Mathematics Yale University New Haven eT 06520 U.S.A.
A Survey on the Local Structure of Morita and Rickard Equivalences between Brauer Blocks Lluis Puig
1. Introduction 1.1. Since Jeremy Rickard's thesis - developing a "Morita theory" for the equivalences between the so-called derived categories of the categories of modules over two algebras [15], and exhibiting such an equivalence when the starting two algebras are the block algebras in characteristic p of Brauer blocks with the same cyclic defect gorups and the same inertial quotients - it has appeared an ample belief in the sense that the existence of such equivalences could "explain" the similarities between some pairs of Brauer blocks which, however, are far from being Morita equivalent. Of course, the first example is Rickard's equivalence for blocks with cyclic defect groups mentioned above, lifted by Markus Linckelmann to characteristic zero [6] and explicited by Raphael Rouquier which exhibits in [21] a quite simple two terms bicomplex inducing such an equivalence. Although, at present, there are not so many other examples, the reasonably expected derived equivalences exposed by Michel Broll(~ in [3], [4] and [5] justify, from our point of view, an effort to understand a priori the consequences of such equivalences between the socalled local structures [2], [1] and [8] of the concerned blocks, in particular seeking inductive constructions. A first attempt in that direction is Rickard's result on the so-called splendid equivalences in [18], which we improve here (Section 6). 1.2. But before trying to understand the relationship between the local structures of two blocks having equivalent derived categories of the categories of modules, it is prudent to start by analysing this relationship when they have already equivalent module categories; that is to say, in a widely employed terminology, when the blocks are Morita equivalent. Let us fix some notation; as usual, p is a prime number and eJ is a complete discrete valuation ring having an algebraically closed residue field k = (') / J (eJ) of characteristic p (we allow the possibility eJ = k), and all the modules we consider are finitely
102
L1uis Puig
generated over <9. Let G and G' be finite groups, band b' be respectively blocks of G and G' (i.e. primitive idempotents in Z(<9Gb) and Z(<9G'b'» and finally Py and p y be respectively maximal local pointed groups on A = <9Gb and A' = <9G'b' (i.e. P and P' are respectively defect groups of band b', whereas y and y' are respectively conjugacy classes of primitive idempotents i in A P and i' in (A')P! such that i fJ. A~ and i' fJ. (A')~: for any proper subgroup Q of P and Q' of P'). We denote respectively by AO = <9Gb o and A'o = <9G'b'o the opposite interior G-algebras of A and A', and respectively by Ay and A~! the source algebras of band b'; that is to say, Ay is the <9-algebra i Ai , where i E y, endowed with the group homomorphism P -+ (i Ai)* mapping u E P on ui (see [10], §2 for more detail on the notation). !
1.3. First of all, let recall the current cases we know where band b' are Morita equivalent; they are so in the following three situations:
I.
The blocks band b' are nilpotent and the defect groups P and p' are isomorphic ([9], Main theorem).
2.
The groups G and G' are p-solvable and the groups obtained by iterating Fong's reduction are isomorphic ([11]).
3.
The group G is a Chevalley group over a finite field of characteristic different from p and, for some parabolic subgroup H of G, some block e of H such that eb = e , some Levi complement L in H of the radical of H and some block f of L such that f e = e, formed by cuspidal irreducible characters, we have G' = NG(L, 1), b' = f, P' = P C L ([12], Corollary 5.10).
In all these situations, it happens that P and P' are isomorphic, so that we may assume that P = P' , and that there is an indecomposable <9 P -module N of vertex P such that, setting 5 = Endo (N) and considering it as an interior P -algebra, we have an interior P -algebra embedding Ay -----* 5 ®o A~!
(1.3.4)
(i. e. injective homomorphism with image i"(5 ®o A~/)i" where i" is the image of the unity element) and P stabilizes an <9-basis of 5. Actually, it is not difficult to prove that, conversely, the existence of such embedding forces the symmetric one
A> -----* 50 ®o Ay and implies that band b' are indeed Morita equivalent.
(1.3.5)
Morita and Rickard Equivalences
103
1.4. On the other hand, according to a well-known definition, band b' are Morita equivalent if (and only if) there is an O-free O(G x G')-module M" associated with b 0 (b')O such that, denoting by (M")* the dual O-module Hom(') (M", 0) which is an O-free O(G x G')-module too, associated with bO 0 (b'), we have repectively O(G x G) and O(G' x G')-module isomorphisms (1.4.1) and our first purpose had been to connect these ismorphisms with those embeddings. A first remark in that direction is that these module isomorphisms can be easily modified to obtain algebra isomorphisms; indeed, recall that isomorphisms 1.4.1 imply that, in particular, the restriction of M" to O(l x G') is projective, so that we get an O(G x G)-module isomorphism (1.4.2) but, it is easily checked that this isomorphism ought to be the structural interior G-algebra homomorphism (1.4.3) multiplied by some element in End(') (M")GxG' , which forces homomorphism 1.4.3 to be bijective. 1.5. Another easy consequence of isomorphisms, 1.4.1 is that M" is an indecomposable O( G x G') -module; hence, it has a vertex P" and an 0 P" -source N", and if we are interested in relating to each other the local structures of A and A', it is reasonable to employ the local structure of M" to get it. So, set S" = End(') (N,,) considered as an interior P" -algebra and recall that the induced interior G x G' -algebra Ind~,~G' (S") ([10],2.14) is just
End(')(Ind~,~G'(N")) in that case; now, since M" is a direct summand of the induced O(G x G')-module Ind~,~G' (N") , we get from 1.4.3 an interior G-algebra embedding (1.5.1) and to "compute" the second term, we have been led to introduce the noninjective induction of interior H -algebras.
104
L1uis Puig
2. The Noninjective Induction 2.1. Let Hand H' be finite groups,
(2.1.1)
which is clearly compatible with the action of K by conjugation on B and on C @(lK B; moreover, if a, a' E Band K fixes 1 Q9 a in C Q9(lK B then, for any x E K we get (l Q9 a) . (x . a') = 1 Q9 aa'
(2.1.2)
consequently, the map 2.1.1 restricted to (CQ9(lK B)K induces a new K -stable bilinear map (C Q9(lK B)K
X
(C Q9(lK B) ---+ C Q9(lK B
(2.1.3)
which determines a product in (C Q9(lK B)K (C Q9(lK B)K x (C Q9(lK B)K ---+ (C @(lK B)K
(2.1.4)
mapping (1 @ a, 1 Q9 a') on 1 Q9 aa' for any a, a' E B such that K fixes 1 @ a and 1 Q9 a' in C Q9(lK B. It is easily checked that (C Q9(lK B)K with this product becomes an associative C-algebra and that the structural map H ---+ B* induces a group homomorphism K H = H / K ---+ (C Q9(lK B) .
(2.1.5)
2.2. We are ready to define the induced interior H' -algebra Indrp (B); considering (C @(lK B)K and CH' as C(H x H)-modules, the second via
(x' @ (l Q9 a) Q9 s')(y' @ (l @ a') @ t') =
{
(l Q9 a . Z . a') @ t'
~r
(2.2.2)
according to whether or not there is z E H such that
(2.2.3)
Morita and Rickard Equivalences
105
-1
mapping x' E H' on L y' x' y' 0 (l 0 1 B) 0 y' where y' runs over a set ofrepresentatives for H' j
x' 0 a . (z . n)
I
(x 0(l0a)0s ' )·(y'0n)=
I
~r
(2.3.1)
according to whether or not there is Z E H such that
2.3.2. If N is 0 -free and B ~ End o (N) as interior H -algebra then the action of Indrp(B) on Indrp(N) induces an interior H' -algebra isomorphism Indrp(Endo(N))
~
Endo(Indrp(N)).
2.4. Let HI! be a third finite group and
o 0 o[
(0 00K B)K ----+ 0 00L B
(2.4.1)
which is clearly compatible with the action of L by conjugation and therefore it induces a canonical map (2.4.2) which, unfortunately, needs not to be bijective. It can be proved that this map induces an interior HI! -algebra homomorphism Indrp/(Indrp(B)) ----+ Indrp/orp(B)
(2.4.3)
which is bijective if and only if the map 2.4.2 is so. In particular, this happens when one of the group homomorphisms
106
L1uis Puig
by left and right multiplication, is a direct summand of a permutation (J(L x L)-module.
2.5. Let B ' be an interior H' -algebra; now, we will compare the interior It is not difficult to H'-algebras Indcp(B) 00 B ' and Indcp(B 00 Rescp(B ' see that there is a unique interior H' -algebra homomorphism
».
(2.5.1)
»
mapping (l 0 (l 0 a) 0 I) 0 a ' on I 0 (l 0 (a 0 a' 0 I for any a E B such that K fixes I 0 a in (J 00 K B, and any a ' E B ' ; it makes sense since K fixes 10 (a 0a ' ) in (J 00K (B 00 Rescp(B ' Moreover, if B ' is (J-free we get an interior H' -algebra isomorphism
».
(2.5.2) 2.6. The corresponding version of the so caled "Mackey formula" needs here a pull-back of groups as follows. Let L' be a third finite group and r/ : L' ~ H' a second group homomorphism, and consider the following pull-back diagram of groups
cp H
~
H'
(2.6.1)
L
~
L'
It is clear that, up to suitable identification, cp and 1/1 have the same kernel K; from this fact it is quite clear that we have a canonical interior L' -algebra embedding
(2.6.2) which is an isomorphism if and only if H' = cp(H) . r/(L ' ). More generally, if we consider all the pull-backs determined by the family {(1J')X' }x" where x' runs on a set of representatives for 1J1(L I )\H' jcp(H) in H', then the very Mackey formula states that the image of the unity elements of the corresponding interior L ' -algebra embeddings form a pairwise orthogonal idempotent decomposition of the unity element in Indcp(B).
Morita and Rickard Equivalences
107
2.7. There is no difficulty on inducing any homomorphism between two interior H -algebras to obtain an interior H' -algebra homomorphism between the corresponding induced interior H' -algebras, and on proving that Indip becomes a functor between the categories of interior H - and H' -algebras. Then, all the homomorphisms above are in fact natural maps between suitable functors and, to be complete, it has to be checked all the obvious compatibilities between them, which amounts to prove the commutavity of certain diagrams. 2.8.
Finally, let us consider the relationship between the Brauer section B(Q) of B at any p-subgroup Q of H (i.e. the interior CH(Q)-algebra
k Q9CJ (BQ / B~) where R runs on the set of proper subgroups of Q) and the corresponding Brauer section (Indip(B))(ep(Q)) of Indip(B). It seems hopeless to obtain a precise general answer but if we assume that ep is SUfjective and that, considering the action of H x H on B by left and right muliplication and denoting by L\ (H)) the diagonal subgroup of H x H, B is a direct summand of a permutation <:J«K x K)· L\(H))-module with projective restriction to <:J(K x K) we get the following result:
LR
2.8.1. If Q' is a p-subgroup of H' then the natural homomorphism B K (Q/) ---+ (<:J 0CJK B)K (Q/) determines, for any complement Q of K in ep-l(Q/) an interior ep(CH(Q))-algebra embedding eip.Q(B) : IndipQ(B(Q)) ---+ (Indip(B)(Q/) where epQ : CH (Q) ---+ ep(CH(Q)) is the restriction of ep, and the family {eip. Q (B) (l 0 B rQ (l B))} QEC, where C is a set of representatives for the set of K -orbits on the set of K -complements of ep-l (Q/), is apairwise orthogonal idempotent decomposition of the unity element of (Indip(B))(Q/), which does not depend on the choice of C.
This result depends on the following lemma which generalizes Lemma 4.2 in [18].
Lemma 2.9. Let X be a finite group, Y a normal subgroup of X and M a direct summand of a permutation <:JX -module with projective restriction to <:JY. Then for any p-subgroup Q' of X' = X/Y, the inclusion MY C M induces a kNxf(Q/)-module isomorphism (MY)(Q/) ~
(n
M(Q»)Y
(2.9.1)
Q
where Q runs on the set of complements of Y in the inverse image of Q' in X.
108
L1uis Puig
3. Morita Equivalences between Brauer Blocks 3.1. Let us come back to the end of Section 1; keeping all the notation there, recall that we need to compute the interior G-algebra (Ind~,~G' (S"»I xG'. Denote by Jr : G x G' --+ G and Jr' : G x G' --+ G' the projection maps and set A" = Ind~,~G' (S"); applying successively 2.4.3, 2.5.2 and 2.6.2 (in particularly simple situations ), we have the interior G-algebra isomorphisms
Res~,~G' (A") ~ Ind7l' (Indg:?' (Resg:f (A"»)
~ Ind7l' (A" ®c Indg~f(0» ~ Ind7l' (A" ®c Res7l',(Indf (0»)
(3.1.1)
and the composed isomorphim maps a" E A" on 1 ®Tr:~?' (a" ®(l ® 1® 1»; inparticular,itmaps a" E (A")lxG' on 1®(a"®Tr:~?'(l®I®I», so that it induces an interior G-algebra isomorphism (A,,)lxG' ~ Ind7l' (A" ®c Res7l',(OG'»
(3.1.2)
Hence, denoting respectively by p : p" --+ G and p' : p" --+ G' the restrictions of Jr and Jr', and applying again 2.5.2 and 2.4.3, we get finally
(Ind~,~G' (S,,»lxG' ~ Ind7l'(Ind~,;G'(S") ®c Res7l',(OG'»
~ Ind7l'(Ind~,;G' (S" ®c Resp,(OG'») ~ Indp(S" ®c Resp,(OG'»
(3.1.3)
and setting Ker(p) = 1 X V' for a suitable p-subgroup V' of G', for any family {s;, }X'EG' of elements of S" such that 1 x V' fixes LX'EG's;, ® x' in 0 ®c(1 xV') (S" ®c Resp,(OG'», the composed isomorphism h fulfills
h(Tr~~~:(
L x'EG'
X,-l
® s;, ® I» = 1 ® (l ® (
L
s;/ ® x'» ® 1 (3.1.4)
x'EG'
We are ready to state our first result; to formulate it as a necessary and sufficient condition, we keep just our notation in 1.2.
Theorem 3.2. The blocks band b' are Morita equivalent if and only if, for a suitable p-subgroup P" of G X G' such that Jr(P") = P and Jr'(P") = P' and a suitable indecomposable 0 P" -module N" of vertex P" such that the restriction of N" to P" n (G xl) and to P" n (I x G') are both projective, we have an interior P -algebra embedding
Ay --+ Inda(S" ®c Resa'(A~,)
(3.2.1)
Morita and Rickard Equivalences
109
where a : pll ---+ P and a' : pll ---+ p' are respectively the restrictions of nand n', and S" = End o (Nil). Remark 3.3. In that case, the Morita equivalence is realised by an indecomposable direct summand M" of Ind~;GI (Nil), which can be determined by a suitable multiplicity module ([10], 6.4 and [13], Theorem 3.4), from the multiplicity module of Py on A. Indeed, notice that embedding 3.2.1 determines a local point y of P on Inda(S"00Resal(A~I» and this interior P-algebra is obviously embedded in Inda(S" 00 Respl«(')G'» ;:: (Ind~,~G' (S"» , xG'
(3.3.1)
so in (Res~~g;(A"»'xGI; hence, y is also a point of P x G' on A" and the multiplicity module of Py on A is canonically isomorphic, always via embedding 3.2.1, to an indecomposable projective direct summand of the multiplicity module of (P x G')y on A" which, by a result of Laurence Barker ([1], Theorem 5.1), determines a unique point a" of G x G' on
A" ;:: Endo(Ind~,~GI (Nil»
(3.3.2)
so an indecomposable direct summand of Ind~,~GI(Nil) 3.4. It is now evident that embedding 3.2.1 becomes embedding 1.3.4 whenever a and a' are group isomorphisms, but it may be not yet clear to the reader why S" should have a P -stable (') -basis. It turns out that this is indeed a strong consequence of embedding 1.3.4 itself; precisely, we have the following general result which, as a matter of fact, has been a surprise for us. Recall that the so called local category [8] of the block b is the category where the objects are the local pointed groups on A = (,)Gb and the morphisms are the group exomorphisms induced by composing the suitable restrictions of inner automorphisms of G and the inclusions of pointed groups ([8], Definition 2.1) and that A determines a central k* -extension of the group of automorphisms of any object ([ 10], 6.6). Theorem 3.5. Let Q be a finite p-group, Rand R' interior Q-algebras and N an indecomposable (') Q -module. Assume that Rand R' have Q x Q' -stable (') -bases by left and right multiplication, that R(R) and R' (R) are nonzero symmetric k -algebras for any subgroup R of Q and that, setting S = End o (N), we have an interior Q-algebra embedding
R ---+ S
00
R'.
Then S has a Q -stable (') -basis by conjugation.
(3.5.1)
110
L1uis Puig
Corollary 3.6. With the notation of Theorem 3.2, assume that band b' are Morita equivalent. Then the following conditions are equivalent to each other. 3.6.1. The homomorphism a : p" 3.6.2. The homomorphism a ' : p"
~ ~
P is bijective. pi is bijective.
3.6.3. The prime p does not divide rank o(N"). 3.6.4. The interior p" -algebra S" has a p" -stable ()-basis by conjugation. Moreover, in that case, embedding 3.2.1 induces an equivalence between the local categories of band b' which preserves the corresponding central k* -extensions.
3.7. When band b' are Morha equivalent and all the conditions in Corollary 3.6 hold, we say that the Morita equivalence is basic; in that case, it is now clear that, up to suitable identifications, embedding 3.2.1 becomes indeed embedding 1.3.4; moreover, notice that if N" ~ () then Ay ~ A>, a fact proved independently by Leonard Scott about 1990 [22]. It is remarkable that, when () has characteristic zero, all the known Morita equivalences between Brauer blocks are basic; actually, it can be proved from Theorem 3.2 that if () has characteristic zero and the blocks band b' are nilpotent then all the Morita equivalences between them are basic (Markus Linckelmann proves a stronger result in [7]), and a recent result of Klaus Roggenkamp and Leonard Scott [20] suggests that the same could be true for any pair of blocks whenever G and G' are p-solvable. A stronger consequence of Theorem 3.2 answers in the affirmative the question we raised in [9], 1.8. Theorem 3.8. Assume that () has characteristic zero. If band b' are Morita equivalent and b' is a ni/potent block then b is a ni/potent block too and the defect groups P and pi are isomorphic. 3.9. Actually, the results of this section can be generalized to the so-called Morita stably equivalences between Brauer blocks. We say that band b' are Morita stably equivalent if in 1.4 we have just stable isomorphisms or equivalently, if isomorphisms 1.4.1 are replaced by the ()(G x G)- and ()(G ' x G')-module isomorphisms M" 0oG" (M")* ~ A EB C and (M")* 00G M" ~ A' EB C' (3.9.1)
where C and C' are respectively suitable proective ()(G x G)- and ()(G ' x G ' ) -modules. Then Corollary 3.6 and Theorem 3.8 remain true when replacing "Morita equivalent" by "Morha stably equivalent", and "local category" by a suitable "stable local category" where we exclude the trivial p-subgroup; in
Morita and Rickard Equivalences
111
particular, notice that, since Jeremy Rickard has proved in [16], 5.5 (see also [4], Proposition 5.2) that if band b' are derived equivalent then they are also Morita stably equivalent, if <9 has characteristic zero, band b' are derived equivalent and b' is nilpotent then b is nilpotent too and P ;::; pi.
4. The Differential Structures 4.1. Let H be a finite group; from now on, we have to replace <9 H -modules by complexes of <9 H -modules; fortunately enough, as we will see in the next section, we need only to consider complexes of <9H -modules which are still finitely generated over <9. Recall that a (finitely generated) complex of <9H -modules is just an <9H -module M (always finitely generated over (9) endowed with a gradation, which is nothing but a pairwise orthogonal idempotent decomposition {i~ }zEZ of idM in End(,)H (M), and a differential, which isjust an element d M of End(,)H(M) fulfilling (d M )2=0 and dM(i~(M))Ci~l(M)foranYZEZ
(4.1.1)
notice that, since M is finitely generated, all but a finite set of idempotents i~ are zero, and that the sum
L <9i~ + L <9i: d ZEZ
M
(4.1.2)
ZEZ
is an <9-subalgebra of End(,)H(M).
4.2. In other words, denoting by F the <9-valued functions over Z, by s h : F -----+ F the <9 -algebra automorphism fulfilling (sh(f))(z) = f(z
+ 1) for any Z E
Z and any f E F
(4.2.1)
and by D the <9-algebra containing F and an element d fulfilling d 2 = 0, df = sh(f)d for any f E F, and D = F Efl Fd,
(4.2.2)
M is actually a DH -module; that is to say, an <9-module endowed with a unitary <9 -algebra homomorphism DH -----+ End(') (M).
(4.2.3)
Coherently, we replace interior H -algebras (called also interior <9H -algebras in the sequel) by interior DB -algebras which, mimicking the case of End(') (M), are <9-algebras B endowed with a unitary <9-algebra homomorphism DB -----+ B
(4.2.4)
112
L1uis Puig
As for interior <9 H -algebras, an homomorphism between two interior D H algebras is simultaneously a nonnecessarily unitary <9-algebra homomorphism and a D H -bimodule homomorphism.
4.3.
In order to define the tensor product between either DH -modules or interior D H -algebras, we have to consider the following <9 -subalgebra Do. Let Fo be the <9-subalgebra of F formed by the periodic functions; notice that Fo @o Fo can identified (and we do!) with the <9-algebra of all the <9-valued biperiodic functions over Z x Z (on the contrary, F @o F maps not surjectively into the <9-algebra of all the <9-valued functions over Z x Z); it is quite clear that (4.3.1)
Do = Fo EEl Fod
is an <9-subalgebra of D and the point is that Do admits a Hopf <9-algebra structure, namely with the diagonal <9-algebra homomorphism ~o
: Do ---+ Do @o Do
(4.3.2)
defined by
+ Z') for any z, Z' d @s+I@d
~o(f)(z, Z') = fez
and
~o(d)
=
E Z and any f E Fo,
(4.3.3)
where s E Fo maps z on (-If. Now, for any <9-algebra B (always finitely generated as <9-module!), and any <9-algebra homomorphism g : D ---+ B, it is quite clear that g(Do) = g(D) and it can be proved that for any <9-algebra homomorphism f : D @o D ---+ B, the composed homomorphism 60
Do ---+ Do
@o
Do C D
@o
f
D ---+ B
(4.3.4)
extends to a unique <9-algebra homomorphism ~~(f)
: D ---+ B
(4.3.5)
4.4. Consequently, if Band B ' are interior D H -algebras and we denote by sts : DH ---+ Band sts' : DH ---+ B ' the structural maps, the tensor product of Band B ' is the <9-algebra B @o B ' endowed with the <9-algebra homomorphism DH ---+ B
@o
B'
(4.4.1)
mapping x E H on sts(x)@sts'(x) and g E D on ~~0SI(sts@stsl)(g). When B = Endo(M) and B' = Endo(M'), we get the tensor product of D H -modules; more generally, it is easily checked that if L is a normal
Morita and Rickard Equivalences
113
subgroup of H then the kernel of the canonical map
M Q9C) M' ---+ M Q9C)L M'
(4.4.2)
*
where the right en-module structure of M is given by y m = y-l . m for any y ELand any m EM, is a D H -submodule of M Q9C) M' and therefore M 0C)L M' becomes a D(HjL)-module. The associativity of the tensor product comes from the coassociativity of Doo and the uniqueness of the extension 4.3.5; similarly, the commutativity "up to isomorphism" comes from the equality so
0
Doo = int(to)
0
Doo
(4.4.3)
where So is the automorphism of Do Q9C) Do which exchanges both factors, to E Fo Q9C) Fo maps (z, z') on (-1)zz' and int(to) denotes the corresponding inner automorphism of Do 0C) Do. Moreover we consider the tJ-algebra isomorphism (4.4.4) defined by t(d) = sd and (t(f))(z) which fulfills over Do
=
f( -z) for any z E Z and any f E F
(4.4.5) and allow us to define the opposite interior D H -algebra BO of an interior DH-algebra B and,consequently,the tJ-dual M* of an tJ-free DH-module M. In particular, notice that an interior D H -algebra B has a canonical D(H x H)-module structure obtained from the tJ-algebra homomorphism B Q90 BO ---+ EndC)(B)
(4.4.6)
defined by left and right multiplication. 4.5. Let B be an interior D H -algebra; as when working with D H -modules, we have to consider the so-called cycles and bords of B; actually, for our purpose here, we can restrict ourself to consider only a-cycles and a-bords. It is not difficult to check from the D-module structure of B that the set Co(B) of a-cycles of B is the interior tJ H -subalgebra formed by the elements of B which centralize the image of D by the structural map stB : DH ---+ Band the set Bo(B) of a-bords of B is the ideal of Co(B) formed by the elements d . a + a . d where a runs on the set of elements of degree one of B; as usual, we set
Ho(B) = Co(B)j Bo(B) which is an interior (') H -algebra.
(4.5.1)
114
L1uis Puig
4.6. More generally, if L is a subgroup of H then B L is an interior DCH(L)algebra and we consider also Co(B L )
= CO(B)L, Bo(B L ) C
Bo(B)L and Ho(B L )
= Co(BL)j Bo(B L ).
(4.6.1) Now a point E of L on B is a conjugacy class of primitive idempotents in Co(B L ) = CO(B)L, so it is just an ordinary point of L on the interior (') H -algebra Co(B) ([ 10],2.10) and we say that E is contractile if E C Bo(B); notice that if j E E then j Bj endowed with the map DL -+ j Bj defined by the multiplication by j is an interior D L -algebra, noted BE and obviously we have ([10], 2.13) (4.6.2) As in the interior (')H -algebra case, if B = Endo (M) then BE = End o (j (M)) and j (M) is an indecomposable directsummand ofthe DL-module Resf (M). Coherently, pointed groups on B are nothing but ordinary pointed groups on Co(B) and we define inclusion and localness from those on Co(B) ([10], 2.10); except that there are the contractile pointed groups which we are not interested in. Notice that a pointed group on B contained in a contractile one is itself contractile and it can be proved that a pointed group on B is contractile if and only if its defect pointed groups are so. 4.7. However, localness on interior DH -algebras is not so plain; indeed, if Q is a p-subgroup of H, it is clear that the Brauer section B(Q) of B at Q inherits an interior DCH(Q)-algebra structure from B, so that we can consider the interior (')CH(Q)-algebra Co(B(Q)). On the other hand, according to our definition, the set of local points of Q on B corresponds bijectively with the set of points of the Brauer section (Co(B))(Q) of Co(B) at Q ([10], 2.10.1). It is easily checked that we have a canonical interior CH (Q) -algebra homomorphism (Co(B))(Q) -+ Co(B(Q))
(4.7.1)
but it needs not be either injective or surjective; in particular, a local point of Q on B needs not to determine a point of 1 on B(Q). 4.8. There is a particular situation, which occurs in Rickard equivalence between blocks (see Section 5 below), where homomorphism 4.7.1 is an isomorphism. First of all, recall that a D H -module M is contractile if an only if idM belongs to Bo(Endo(M)) or equivalently, all the points of H on Endo(M) are contractile. On the other hand, any interior H -algebra can be considered
Morita and Rickard Equivalences
lIS
as an interior DH -algebra via the canonical (J-algebra homomorphism (4.8.1) mapping f E F on f (0) and d on zero; in particular, any (J H -module can be considered as a DH -module and we call it a u-restricted DH -module. Now, we call the direct sums of contractile and u -restricted D H -modules a-split D H -modules and it is not difficult to prove that if B is a a-split D Q-module then we have (Co(B))(Q) ~ Co(B(Q))
(4.8.2)
and for any subgroup L of NH(Q) containing Q such that Br~(BL) = B( Q)L, we have also
(4.8.3)
4.9. Finally, it remains to introduce the induction of interior D H -algebras. Let H' be a second finite group,
(4.9.1) and we extend to D H' the structural (J -algebra homomorphism (J H' ----+ Indrp(B) by mapping g E Don Tr:('H)O@00stB(g))01) Itisnotdifficultto check that homomorphisms 2.3., 2.4.1, 2.5.1 and 2.6.2 are interior D-algebra homomorphisms too when we consider them for interior D-algebras. Furthermore, if f3 is a point of H on B and LE a pointed group on B such that
(4.9.2) it is well known ([10], 2.14) that there is a unique exoembedding (Le. a unique conjugacy class of embeddings) of interior (JH -algebras
(4.9.3)
116
L1uis Puig
such that the following diagram of canonical exoembeddings is commutative Resf (h~)
(4.9.4)
r Co(B)e
It can be proved that ii~ extends to a unique exoembedding of interior D Halgebras Bf3 ~ Indf (BE)
(4.9.5)
such that the corresponding extended diagram is commutative too.
s. Rickard Equivalences between Brauer Blocks 5.1. Let us come back to our standard notation in 1.2 and consider respectively A and A' as interior DG- and DG' -algebras, via the structural maps induced by homomorphism 4.8.1; in particular, they become respectively D( G x G)and D(G' x G')-modules 4.4.6. In [16], Jeremy Rickard proves that the socalled derived categories of the categories of A - and A' -modules are equivalent, as triangulated categories, to each other if and only if : 5.1.1. There is an indecomposable D(G x G')-module M" associated with b 0 (b')o such that its restrictions to (9(G x 1) and to (9(1 x G') are both projective and we have respectively D(G x G)- and D(G' x G')-module isomorphisms M" 0CJG ' (M")* ~ A EB C and (M")* 0CJG M" ~ A' EB C'
for suitable contractile modules Cover D( G x G) and C' over D( G' x G').
But notice that Rickard's result does not guarantee that the starting equivalence between those derived categories is induced by M". 5.2. As Rickard points out, in that case M" induces actually an equivalence between the so called homotopic quotients of the categories of DGb- and DG'b' -modules (i.e. the quotients by the homomorphisms which factorize throughout contractile modules) that we denote respectively by ModDGb and ModDGlb " and following Michel Broue [4], we call any equivalence between
Morita and Rickard Equivalences
117
ModDGb and ModDG'b' induced by an indecomposable D(G x G')-module as in condition 5.1.1 Rickard-equivalence between the blocks band b'. 5.3. So, let M" be an indecomposable D(G x G')-module, associated with b 0 (b')o, such that its restriction to tJ(G x 1) and to tJ(l x G') are both projective; it is clear that Endo(M,,)lxG' is an interior DG-algebra and, consequently, that Ho(End o (M")l xG') is an interior G-algebra (cf. 4.5.1) and Endo(M,,)lxG' has a canonical D(G x G')-module structure (cf. 4.4.6). Then, as in 1.4, a first remark is that M" induces a Rickard equivalence between band b' (i.e. we have the isomorphism in 5.1.1) if and only if End o (M")l xG' is a O-split D(G x G)-module and we have an interior tJG-algebra isomorphism A ~ Ho(Endo(M")lxG')
(5.3.1)
moreover, it can be proved that in that case, there is a bijection between the set of pointed groups HfJ on A and the set of noncontractile pointed groups H ~ on End o (M") I X G' such that H fJ and H ~ correspond to each other if and only if isomorphism 5.3.1 induces an interior tJH -algebra isomorphism AfJ ~ Ho«EndO<M")lxG')~)
(5.3.2)
5.4. Since M" is indecomposable, idMII is a primitive idempotent in Co(End o (M")) and thus a" = {idM II } is the unique point of G x G' on Endo(M"); let P;II be a defect pointed group (GxG')a" (either on Endo(M") as interior D(G x G')-algebra or on Co(Endo(M")) as interior tJ(G x G')-algebra); we still call P" a vertex of M" and any j" E y" determines an indecomposable direct summand N" = j" (M") of Res~,~ G' (M"), still called a D P" -source of M". Hence S" = End o (N") is an interior D P" -algebra and we consider respectively the interior D( G x G') - and DG -algebras A" = Ind~,~G' (S") and
A=
(A,,)lxG'.
(5.4.1)
By 4.9.5 we have a canonical exoembedding End o (M") ---+ A"
(5.4.2)
so that a" can be identified with a point of G x G' on A", but notice that, at the same time, it can be considered as a point of G on A; coherently, we identify respectively Endo(M"), S" and Endo(M")lxG' with A~", A~II and
Aall.
The point is that all interior G -algebra isomorphisms we state in
118
L1uis Puig
3.1 to compute A are now compatible with the D-structures, so that, mutatis mutandis, we get the same interior DG-algebra isomorphism (5.4.3) where c:JG ' has the interior DG-algebra structure induced by homomorphism 4.8.1. We are ready to state the generalization of Theorem 3.2; as there, we keep just our notation in 1.2.
Theorem 5.5. The blocks band b' are Rickard equivalent if and only if there are a p-subgroup p" of G X G' such that n(P") = P and n'(P") = pi, an indecomposable D p" -module N" of vertex P" such that its restriction to c:J(P" n (G xl)) and to c:J(P" n (l x G ' )) are both projective and, setting S" = End(') (N") and denoting by (J : P" ~ P and (J' : P" ~ pi the respective retrictions of n and ni, a local point y of P on Inda(S" 0(') Resa,(A~,)) such that Inda(S" 0(') Resa,(A~,))y is a O-split D(P x P)-module and we have an interior P -algebra isomorphism (5.5.1)
Remark 5.6. Actually, as in Remark 3.2, there is a more precise result involving multiplicity modules which gives a necessary and sufficient condition on M" for inducing a Rickard equivalence between band b'. 5.7. From now on, we assume that M" induces a Rickard equivalence between b and hi; then, in 5.4 we may choose P;, in such a way that:
5.7.1. We have n(P") = P, n'(P") = pi and an interior DP-algebra embedding -yl/,y' . ~
hy
. Ay
~
"
I
Ind a (S 0(') Resu,(A y '))
where y is the noncontractile point of P on A such that Ay ~ Ho(A y ) (cf. 5.3.2). In order to give a more precise description of the relationship between the local categories of band b' (cf. 3.4), it is proving useful to consider the following definition (which has not been introduced in Section 3 to avoid too much redundancy). A triple (Qo, Q~I/' Q~,) of local pointed groups Qo on A, Q~" on A" and Q~, on A' is called a local tracing triple on A, A" and A' if we have n (Q") = Q, n ' (Q") = Q' and a canonical interior DQ-algebra exoembedding (cf. Remark 5.2 below) -8" 8'
hg '
~
: Ag
~
"
I
Indr(A o" 0(') Resr,(A lI ,))
(5.7.2)
Morita and Rickard Equivalences
119
where r : Q" ----+ Q and r' : Q" ----+ Q' are respectively the restrictions of rr and rr', and 3 is the noncontractile point of Q on A such that A,s ~ Ho(Ag) (cf. 5.3.2). Notice that ( Py, P:", P~I ) is indeed a local tracing triple on A, A" and A', since A~II ~ 5". Recall that 3 is also a point of Q x G' on A", so on Co(A"), and let us denote respectively by VCO(AII)g(Q~II) and VA'(Q~,) simple modules over Co(A")9" ,s and (A')Q' where 8" and 8' act nontrivially. Theorem 5.8. A triple (Q,s, Q~II' Q~,) of local pointed groups Q,s on A, Q~" on A" and Q~, on A' is a local tracing triple on A, A" and A' if and only if Q~" is a defect pointed group of (Q x G')g on A", we have rr' (Q") = Q' and, considering respectively the structural maps g' and g" from OCC,(Q') to (A')Q' and to Co(A")j", Resgl(VA'(Q~,)) is a quotient of Resg,,(VCo(A")g (Q~II))' Remark 5.9. Notice that neither exoembedding 5.7.2 nor Theorem 5.8 are symmetric in the roles of A and A'; indeed, if ( Q,s, Q~II' Q~,) is a local tracing triple on A, A" and A' then ( Q~/' Q~II' Q(j ) needs not to be a local tracing triple on A', (A")o and A (i.e. with repect to the Rickard equivalence between b' and b induced by (M")*). However, if (Q~" R~II' RE) is a local tracing triple on A', (A")O and A, it can be proved that there is x E G such that (Q(j)X eRE' 5.10. It is clear that G x G' acts by conjugation on the set of local tracing triples on A, A" and A'. Moreover, if ( Q(j, Q~", Q~, ) and ( RE, R;/I' R~, ) are local tracing triples on A, A" and A', we say that (RE' R~II' R~, ) is contained in (Q,s, Q~", Q~,) if we have RE C Q,s on A, R~" C Q~II on A", R~, C Q~I on A' and, for a suitable canonical exoembedding (cf. Remark 5.12 below) of interior D R -algebras, the following diagram is commutative ResW(hr",sI)
(5.10.1)
h:''',f' f
Ag
Indv(A~1I 00 Resv,(A~/))
where v : RI! ----+ R and v' : R" ----+ R' are respectively the restrictions of rr and rr'. Then it is easily proved that the inclusion between local tracing
120
L1uis Puig
triples is transitive and we consider the category where the objects are the local tracing triples on A, A" and A', and the morphisms are the group exomorphisms between the middle terms induced by composing the suitable restrictions of inner automorphisms of G x G' and the inclusions of local tracing triples: we call it the local category of the Rickard equivalence induced by M".
Theorem 5.11. The functor from the local category of the Rickard equivalence induced by M" to the local category of the block b defined by the first projection is full and induces a bijection between the sets of isomorphism classes of objects in both categories.
Remark 5.12. Actually, to give a precise definition and to prove the unique-8" 8'
8" 8'
ness of the canonical exoembeddings h.' in 5.7.2 and gf ,,',f in 5.10.1, we 8 need to introduce the Higman envelope of an interior D H -algebra B, namely the smallest interior DH -algebra G(B) admitting interior DH -algebra exoembeddings I
(f(B) : B ---+ G(B) and hI(! : lnd: (BI(!) ---+ G(B)
(5.12.1)
for any local pointed group TI(! on B, such that the following evident diagram of exoembeddings is commutative ([ 10], 2.13.1 and 2.14.1) Res: ((f(B)) Res: (G(B))
Res: (B)
i·I BI(!
IRes~
0;.)
(5.12.2)
-H d T (BI(!)
Res: (lnd: (BI(!))
6. Basic Rickard Equivalences between Brauer Blocks 6.1. This section is devoted to a special type of Rickard equivalences which generalizes the basic Morita equivalences between blocks (cf. 3.7). To our knowledge, Michel Broll(~ is the first who guessed the existence of some special kind of Rickard equivalences which should appear underlying the so-called Deligne-Lustzig induction, for finite Chevalley groups in characteristic different from p, in some well chosen situations ([3], § 6, [4], § 6C and [5], § 4); according to him, the p-adic cohomology of Deligne-Lusztig varieties should
Morita and Rickard Equivalences
121
provide us with D(G x G')-modules which restricted to any p-subgroup of G x G' would have a stable C)-basis; moreover, the corresponding pairs of blocks foreseen by Broll(~ have isomorphic defect groups and equivalent Brauer categories ([2] § 1). Jeremy Rickard proves in [17] that indeed the proposed p-adic cohomology furnishes the expected kind of D(G x G')-modules and in [18] that, assuming the equivalence of the Brauer categories for the principal blocks, this special kind of Rickard equivalences - which he denominates differently - is inherited locally (Le. between the corresponding blocks in the centralizers of the p-subgroups). 6.2. As a matter of fact, the following weaker hypothesis suffices to get this special situation between Rickard equivalent blocks. We keep all the notation of Section 5 and assume that M" induces a Rickard equivalence between band b'. We say that this equivalence is basic if, considering the action of P" x P" on 5" by left and right multiplication and denoting by Do(P") the diagonal subgroup of P" x P", both (Ker(a) x Ker(a)) . Do(P") and (Ker(a') x Ker(a')) . Do(P") stabilize C)-bases of 5"; obviously, a sufficient condition is that 5" had a (P" x P")-stable C)-basis; but notice that a basic Morita equivalence is a basic Rickard equivalence and, in that case,S" needs not to have a (P" x P")-stable C)-basis. Since we assume that the restriction of M" to c)(G x 1) and to C)(l x G') are both projective, the restriction of N" to Ker(a) = P"
n (l x
G') and Ker(a') = p"
n (G x 1)
(6.2.1)
are both projective too and therefore Ker(a) x Ker(a) and Ker(a') x Ker(a') act freely in their corresponding stable C)-basis; in particular it follows from Weiss' Theorem on permutation modules ([23], Theorem 2; see also [19], Theorem 2) that if C) has characteristic zero, the Rickard equivalence induced by M" is basic if and only if P and P' stabilize respectively C)-bases of Ind a (5") and Indal (5") by conjugation. 6.3. From now on, we assume that M" induces a basic Rickard equivalence between band b'. Unfortunately, here we cannot relate directly the local categories as we do in Theorem 3.6, but only the so-called Brauer categories of band b' (cf. [2], § 1). Recall that one goes from the local to the Brauer categories of b by replacing the local pointed groups Qo on A = c)Gb by the b-Brauer pairs (Q, b(8))), where b(8) is the block of CG(Q) such that we have b(8)BrQ(8) = BrQ(8) in (c)G)(Q) ~ kCG(Q) ([10], 2.9.2), keeping between them all the group exomorphisms in the local category; that is to say, if (Q, f) and (R, g) are two b-Brauer pairs then a group exomorphism from
122
L1uis Puig
R to Q belongs to the Brauer category if and only if it is induced by x E G such that (RE)X C Q 0 for suitable local points E of Rand e5 of Q on A such that beE) = g and b(e5) = f.
6.4. Let Q 0 be a local pointed group on A, (Q 0, Q~I/' Q~, ) a local tracing triple on A, A" and A' and R a normal subgroup of Q, and denote respectively by r : Q" ~ Q and r' : Q" ~ Q' the restrictions of a and a'. It is not difficult to see that our basic hypothesis implies that Q stabilizes an (9-basis of Ag by conjugation and, consequently, that equality 4.8.3 applies; hence the unity element is primitive in Co(Ag(R))Q or, equivalently, Br~ (8) is contained in a point of Q on A(R), which actually is a local point. On the other hand, always from our basic hypothesis, statement 2.8.1 suitably modified replacing (9 by D applies to the interior D Q" -algebra A;I/I8l~ Res,' (A~,) and to the surjective group homomorphism r : Q" ~ Q. It follows quite easily that: 6.4.1. There is a section fl : R ~ r-1(R) of the restriction of r such that the interior DCQ(R)-algebra exoembedding - 01/ 0'
~
h g ' (R): Ag(R)
~
"
(Ind, (Aol/
18l~
I
Res,,(Ao,)))(R)
factorizes via e"R(A;1/ 0~ Resr,(A~,)) in an interior DCQ(R)-algebra exoembedding J-L : Ag(R) ~ " I ' h- g ~ Ind,!, (Aol/(RJ-L) 0k Res,~ (Ao,(RJ-L )))
=
where fl' r ' 0 fl, RJ-L and RJ-L' are respectively the images of fl and fl', and rJ-L ; CQI/(RJ-L) ~ CQ(R) and r~ ; CQI/(RJ-L) ~ CQ,(RJ-L')are respectively the restrictions of rand r'. Notice that fl' is injective since {O} #- A;I/(RJ-L) ~ (A;I/(RJ-L n Ker(r ' ))) (RJ-L) implies RJ-L n Ker(r ') = (1, 1). 6.5. Now, as in 5.7, choose P;I/ in such a way that (Py , P;I/' P;,) is a local tracing triple on A, A" and A'; applying statement 6.4.1 to this triple, with R = P, we get once for ever a section A : P" ~ P of a and then it can be proved with the notation of 6.4 that: 6.5.1 For any x E G such that (Qo)X C Py there is x' E G' such that (Q")(x,x') C P", (Q', b(e5')Y' c (P', bey')) and A(U X) = fl(u)(x,x') for any u ER,
but we know nothing about (e5 ' )(X,x') and (e5')x'. In particular, for any local pointed group Qo on A such that Qo C Py , we can choose a local tracing
Morita and Rickard Equivalences
123
triple(QD,Q~JI,Q~,)on
A,A" and A' such that Q" C pll, (QI,b(O'» C and the restriction of ).. to Q fulfills condition 6.4.1 for R = Q; then applying again statement 6.5.1 to those choices, it is not difficult to prove the following result. We set )..' = a ' 0).. and, for any b-Brauer pair (Q, f) such that (Q, f) c (P, b(y», we denote by QA' the image of Q by )..' in / pi and by fA ' the block of Cc,(QA ) such that (Pi, b(y'»
(QA', fA') C (pi, b(y'».
(6.5.2)
Theorem 6.6. With the hypothesis and the notation above, the group homomorphism )..' : P ----+ pi is bijective and induces an equivalence between the Brauer categories of band b' mapping any b-Brauer pair (Q, f) contained in (P, b(y» on (QA', fA').
6.7.
Finally, we claim that in Theorem 6.6, the blocks of ] of CC(Q) and ]A' of Cc,(QA') over k are basically Rickard equivalent too. Indeed, for any b-Brauer pair, we can choose a G-conjugate (Q, f) such that (Q, f) c (P, b(y» and that Cp(Q) is a defect group of f; then it is not difficult to prove from Theorem 6.6 that C p' (QA ' ) is a defect group of fA'; in particular, if E and EA' are respectively the local points of R = Q . C p (Q) on A and of RA' = QA ' . Cpl(QA') on A' such that RE C Py and R;~, C p~, then
A E (Q) and A' A' (QA ' ) are respectively source algebras of ] and ]A'. On the E other hand, choosing a local tracing triple (RE' R~JI' R~, ) on A, A" and A' and applying statement 6.4.1 to the normal subgroup Q of R, statement 6.5.1 allow us to modify eventually our choices in such a way that we have )"(R) C R" C p" and (RA', b(E A' » C (R', b(E /» C (pi, b(y'» (6.7.1) and an interior DC p (Q) -algebra exoembedding AE(Q) ----+ Indr(A~/I(QA) 0k Resrl(A~I(QA'»)
(6.7.2)
where QA and QA' are respectively the images of Q by ).. and )..1, and r : CR',(QA) ----+ Cp(Q) and r ' : CR',(QA) ----+ CR'(QA') the restrictions of a and a'. 6.8. Actually, since the blocks b( EA') and b( E') are nilpotent ([2], § 1), the inclusion of b-Brauer pairs in 6.7.1 imply I (R A' \A' eRE' c Py'I and CRA,(Q A' ) = CR,(Q A' ) = Cp'(Q A' )(6.8.1) Moreover, although R~/1 needs not be contained in P;/I' there is (x, x') in G x G' such that (R~/I)(X,x') C P;/1 and therefore, it follows from our basic
124
L1uis Puig
hypothesis that A~I/ (QA) is a simple k -algebra, so that A~I/(QA) ~ Endk(M;~,)
(6.8.2)
where M;~, is a simple A~I/(QA)-module, in particular, M;~, becomes a DC R" (QA) - module and although it needs not be indecomposable of vertex CRI/(QA') it can be proved, from exoembedding 6.7.2 and Theorem 5.5, the following result
Theorem 6.9. With the hypothesis and the notation above, there is a subgroup T" of CRI/ (QA) and an indecomposable direct summand N'Q of
Res~~I/(QA)(M;~,)
of vertex T" such that r(T") = Cp(Q),
r'(T")
C p' (QA') and we have an interior DC p (Q) -algebra exoembedding. ~
I
-11
A'
AE(Q) ~ Indp(SQ 0k Resp,(AEA,(Q »)
(6.9.1)
where SQ = Endk(N'Q) and p : T" ~ Cp(Q) and pi: T" ~ Cpl(QA ' ) are respectively the restrictions of rand r ' . In particular, a suitable indecomA'
posable direct summand of Ind~~(Q)XCcl(Q ) (N'Q) induces a basic Rickard equivalence between the blocks j of kCc(Q) and jA ' of kCCI(QA').
Remark 6.10. If N'Q lifts to an <9-free endopermutation <9T" -module N'Q (i.e. such that T" stabilizes an <9-basis of End() (N'Q) too) then Rickard's construction in [18], §5 can be easily generalized to prove that an indecomposable direct summand of A'
I dCc(Q)xCc,(Q )(N" ) n TI/ QA
induces a basic Rickard equivalence between f and fA'.
Remark 6.11. We know that RA' = R' whenever R maps surjectively onto Op(Nc(R E )/ R . Cc(R» ;
otherwise we know nothing about the inclusion R A' C R'.
Morita and Rickard Equivalences
125
References [1]
Laurence Barker, G -algebras, Clifford Theory and the Green Correspondence, 1. Algebra 172 (1995), 335-353.
[2]
Michel Broue, Les £-blocs des groupes GL(n, q) et U(n, q2) et leurs structures locales, Asterisque 133/134 (1986), 159-188.
[3]
Michel Broue, Isometries parfaites, types de blocs, categories derivees, Asterisque 181/182 (1990), 61-92. Michel Broue, Equivalences of blocks of group algebras, Finite dimensional algebras and related topics (V. Dlab et aI., eds.), Proc. Internal. Conference Representations of AIgebras and Related Topics, Ottawa, Canada, 1992, Kluwer, 1994, 1-26. Michel Broue, Rickard Equivalences and Block Theory, Groups '93 GalwaySaint Andrews, London Math. Soc. Lecture Notes 211, Cambridge Univ. Press, 1995 , 58-79. Markus Linckelmann, Derived equivalence for cyclic blocks over a p-adic ring, Math. Z. 207 (1991), 293-304.
[4]
[5]
[6] [7]
Markus Linkelmann, The isomorphism problem for blocks with cyclic defect groups, Invent. Math. 125 (1996), 265-283.
[8]
LIuis Puig, Local Fusions in Block Source AIgebras, J. Algebra 104 (1986), 358-369.
[9]
LIuis Puig, Nilpotent blocks and their source algebras, Invent. Math. 93 (1988), 77-116.
[10]
LIuis Puig, Pointed Groups and Construction of Modules, 1. Algebra 116 (1988), 7-129.
[11]
LIuis Puig, Block source algebras in p-solvable groups, Letter to Morton Harris, 1993.
[12]
LIuis Puig, On Joanna Scopes' Criterion of Equivalences for Blocks of Symmetric Groups, Algebra Colloq.l (1994),25-55. LIuis Puig, On Thevenaz parameterization of interior G-algebras, Math. Z. 215 (1994),325-355. Jeremy Rickard, Morita Theory for derived categories, 1. London Math. Soc. 39 (1991), 37-48.
[13] [14] [15] [16] [17] [18]
Jeremy Rickard, Derived categories and stable equivalences, 1. Pure AppI. Algebra 61 (1989),307-317. Jeremy Rickard, Derived equivalences as derived functors, J. London Math. Soc. 43 (1991), 37-48. Jeremy Rickard, Finite group actions and etale cohomology, Pub!. Math. IHES 80 (1994),81-94. Jeremy Rickard, Splendid equivalences: derived categories and permutation modules, preprint 1994.
126
L1uis Puig
[19]
Klaus Roggenkamp, Subgroup rigidity of p -adic group rings (Weiss arguments revisited), 1. London Math. Soc. 46 (1992), 432-448.
[20]
Klaus Roggenkamp and Leonard Scott, Verbal communication, Luminy, 1988.
[21]
Raphael Rouquier, From stable equivalences to Rickard equivalences for blocks with cyclic defect, Groups '93 Galway-Saint Andrews, London Math. Soc. Lecture Notes 211, Cambridge Univ. Press, 1995.
[22]
Leonard Scott, Unpublished notes.
[23]
Alfred Weiss, Rigidity of p-adic p-torsion, Ann. of Math. 127 (1988), 317332.
CNRS, Institut de Mathematiques de Jussieu 6 Avenue Sizet 94340 Joinville Le Pont France
Some Open Conjectures on Representation Theory G. R. Robinson
Introduction Several of Brauer's questions on block-theoretic invariants (see, for example, [3]) have in recent years been placed in a more conceptual setting. Some still appear unassailable, and some have been modified or extended by later authors. We discuss here some recent developments on some of these questions.
Brauer's Problem 21 Is there a function j : N --+ N such that whenever B is a block with defect group D, and B contains k ordinary irreducible characters, then we have IDI ~ j(k)? See [3] or [6]. This is a precise way of expressing the fact that a block with a large defect group should contain a large number of ordinary irreducible characters (if, as expected, the answer is positive). Little progress has been made on the most general case of this problem, though B. Kulshamrner recently proved that the answer is positive if we restrict attention to blocks of p-solvable groups. Very recently, B. Ktilshammer and I have used E. Zelmanov's solution of the restricted Burnside Problem to prove:
Theorem ([13]). If the p-block B has defect group D and satisfies the equality predicted by the Alperin-Mckay conjecture ([1]), then in terms of k(B).
IDI is bounded
In other words, the Alperin-McKay conjecture imples that Brauer's problem 21 has a positive answer.
128
G. R. Robinson
The" k (B) at Most pd " Conjecture This is Brauer's problem to prove that the number of ordinary irreducible characters in a block B with defect group D is at most ID I. It seems fair to say that no general progress has been made on achieving this precise bound. Brauer and Feit proved that k(B) :::: ~IDe + 1 ([4]), and in general the quadratic nature of the bound has not been improved. Even for blocks of p-solvable groups, the question appears to be very difficult. In that case, the if a pi -group G acts faithfully on problem reduces to the k(GV)-problem: a GF(p)G-module V, does the semi-direct product GV have at most IVI conjugacy classes? Even the p- solvable case of the problem has interesting consequences for blocks of arbitrary groups. For example, Ktilshammer showed in [12] that a positive answer to the Alperin-McKay conjecture, together with a positive answer to this question for blocks of p-solvable groups implies a positive answer to the following conjecture of Olsson: Conjecture (J. B. Olsson [16]). Let B be a block with defect group D. Then ko(B) ::::[D : D '].
R. Knorr ([8], [9]) did fundamental work on the k(GV) problem. Building on that, and influenced by recent work of R. Gow [7], 1. G. Thompson and I proved: Theorem ([20]). Let the finite pi -group G act faithfully and irreducibly on the elementary Abelian p-group V. If there is some v in V such that Resgc(V)(V) has a faithful self-dual submodule (as GF(p)Cc(v)-module), then k(GV) :::: IVI. In my talk at the conference, I expressed the belief that such a v would always exist, thus confirming that the k(GV)-problem would always have a positive answer. Since the conference, there have been further developments. 1. G. Thompson furnished in [21] an example of a pair (G, V) for which there is no such vector v (the prime p is 7). On the positive side, in [20] it is proved that if p is outside a certain finite set of primes, then such a vector v always does exist. This result confirms the p-solvable case of the problem of this section for all but finitely many choices of the prime p. For this work and some extensions, see [19] and [20] Recently, B. Ktilshammer and I have been considering the problem of this section for principal blocks (presumably the most interesting case). We
Some Open Conjectures on RepresentationTheory
129
noted that an easy variant of Nagao's argument in [15] proves that whenever N is a normal subgroup of a finite group G, we have k(B~P)(G» ~ k(B{~P\N»k(GjN), where B{~P) denotes the principal p-block. In [11], Kovacs and I proved that there is a constant c such that when every B is a p-block of defect d of a p-solvable group, then k(B) ~ c d- 1pd. In trying to prove an anologous result for principal blocks of arbitrary groups, Ktilshammer and I have proved that if suffices to consider two configurations for a purported minimal counterexample G: (i)
Op'(G) = 1, E = E(G) i= 1, GjE is a solvable pi_group. For P E Sylp(G) we have that Gj ECG(P) is Abelian. Each component of
E is normal in G. (ii)
OP(F*(G» = 1, and G has a unique minimal normal subgroup V such that,letting X be the full pre-image in G of Op,(GjV), GjX has the structure of case i). Either V ~ (G),or V = F(G). If V ~ (G), then V = X. If V 1:.(G),then V = F*(G) is elementary Abelian.
This work is at a preliminary stage, and we hope to make further progress.
Conjectures of Alperin-Dade Type In [19], we conjecture the following, which is related to a p-Iocal computation of the number of weights (in the sense of Dade [5]) with respect to a normal subgroup. Conjecture. Whenever G is a finite group, N
~
N n Op(Z(G»,
A
[the number of N -projective irreducible characters in B which have defectd, lie over A, but are not Op (N ) -proj ective]= LaE~(N)/G( _l)lal+l [the number of Va -projective irreducible characters of G a which lie over irreducible characters of defect d of Va, lying over A, but not induced from Op(N)]. This generalizes the weight conjecture of Dade [5], which may already be viewed as a far-reaching generalization of the Alperin-McKay conjecture (and Alperin's weight conjecture, [2]) in part by virtue of [10]. Here, 'R(N) denotes the collection of radical p-chains of N. The numbers appearing in the right side are easily computed p-Iocally, using the results of Ktilshammer-Robinson [13] (recall that Va denotes the initial subgroup of the chain a).
130
G. R. Robinson
The case N = G of this conjecture, in this formulation, would have the following consequence, which may be viewed as a complement to Brauer's height 0 conjecture: Conjecture. Let B be a block with defect group D. Then the defect of an irreducible character in B is always the defect of an irreducible character I-t of a radical subgroup U of D with CD(U) ~ U. In particular, if D is non-Abelian, then every irreducible character in B has height less than logp ([D : ZeD)]). Remark. With a little more care it is possible to show that if CJ = Va < VI < ... < Vn is such that B a contains a Va-projective irreducible character lying over the irreducible character I-t of defect d of Va, then we have d ~ ~ [logp (IVn I) + logp (IZ(Vn)l)], and CD(VO) ~ Va.
References [1]
J. L. Alperin, The main problem of block theory, Proc. Conf. Finite Groups, Academic Press, New York, 1976.
[2]
1. L. Alperin, Weight for finite groups, in: Proc. Symp. Pure Math 47, Amer. Math. Soc., Providence, 1987,369-379.
[3]
R. Brauer, Representations of Finite Groups, in Lectures in Mathematics, Vol. 1, Wiley, New York (1963),133-175.
[4]
R. Brauer, W. Feit, On the number of irreducible characters of finite groups in a given block, Proc. Math. Acad. Sci. USA 45 (1959), 361-365.
[5]
E.c. Dade, Counting Characters in Blocks, II, J. Reine Angew. Math. 448 (1994), 97-190.
[6]
W. Feit, The representation theory of finite groups, North Holland, Amsterdam, 1982.
[7]
R. Gow, On the number of characters in a block and the k(GV)-problem self-dual V, J. London Math. Soc. (2) 48 (1993), 441-451.
[8]
R. Knorr, On the number of characters in a p-block of a p-solvable group, Illinois J. Math. 28 (1984),181-210.
[9]
R. Knorr, A remark on Brauer's k(B)-conjecture, J. Algebra 131 (1990),444-450.
for
[10] R. Knorr, G. R. Robinson, Some remarks on a conjecture of Alperin, J. London Math. Soc. (2) 39, (1989),48-60. [11] L. G. Kovacs and G. R. Robinson, On the number of conjugacy classes of a finite group, J. Algebra 160 (1993), 441-460.
Some Open Conjectures on Representation Theory
131
[12] B. Kiilshammer, A remark on conjectures in modular representation theory, Arch. Math. 49 (1987), 366-399.
[13] B. Kulsharnmer, G. R. Robinson, Characters of Relatively Projective Modules II, J. London Math. Soc. (2) 36 (1987), 59-67. [14] B. Kiilshammer and G. R. Robinson, Alperin McKay implies Brauer's Problem 21, to appear in 1. Algebra. [15] H. Nagao, On a conjecture of Brauer for p-solvable groups, 1. Math. Osaka City University 13 (1962), 35-38. [16] J. B. Olsson, Block invariants in An, An and Sn, in: Proc. Symp. Pure Math. 47, Amer. Math. Soc., Providence, 1987,471-474. [17] G. R. Robinson, Some remarks on the k(GV)-problem, 1. Algebra 172 (1995), 159-166. [18] G. R. Robinson, Local structure, vertices, and Alperin's Conjecture, Proc. London Math. Soc. (3) 72 (1996), 312-330. [19] G. R. Robinson, Furtherreductions for the k( G V) -problem, J. Algebra, to appear. [20]
G. R. Robinson and J. G. Thompson, On Brauer's k(B)-problem, J. Algebra 184 (1996), 1143-1160.
[21] J. G. Thompson, private communication.
Department of Mathematics University of Leicester Leicester LE1 7RH England Email: [email protected]
Are All Groups Finite? Leonard L. Scott"
This paper is dedicated to Walter Feit on the occasion of his 65th birthday. Its contents were presented in part at the 1995 Ohio State finite group representation conference organized in celebration of that birthday. Primarily, the paper is a discussion of some classical and recent developments in the modular representation theory of finite groups of Lie type, and the problems which drive that theory. But there is also a philosophical thread ... An old question which arose again at the conference is the following: Are all groups finite? That is, applications and broader issues aside, if we think only of our interest in finite group theory itself, is it possible to safely ignore other groups? My viewpoint is that the answer to this question has two parts: First, in representation theory, at least, we cannot ignore the infinite complex Lie groups and their characteristic p analogs, the algebraic groups over IFp' The second part of my answer is that we can, nevertheless, hope to find understandings within finite group theory and finite dimensional algebra of ideas naturally suggested by these continuous contexts, and take them further. Let me begin by convincing you of the first part of my answer: Suppose one is considering a finite group G (IFq) of Lie type, such as the special linear group S L (n, q) of degree n with coefficients in the field IFq of q elements, q a power of a prime p. The Classification of finite simple groups asserts that almost all of the latter are variations on the finite groups of Lie type together with the alternating groups. Much earlier (1963), Steinberg [34], [35] proved all irreducible representations of G (IFq) with coeffic~ents in a finite field of characteristic p, and, thus, in the algebraic closure IFs- come by restriction from the irreducible representations over lFq of the algebraic group G(lFq ) . The latter group is, of course, quite infinite. It is the analog via the Zariski topology, of the complex analytic Lie group G(e). Moreover, the representations we need are continuous, and even "analytic", in the sense that they are locally defined by polynomial functions. Now, the theory of finite-dimensional
*
The author thanks NSF for its support.
Offprint from: Representation Theory of Finite Groups, Ed.: R. Solomon © by Walter de Gruyter & Co., Berlin' New York 1997
134
Leonard L. Scott
irreducible continuous representations of G(C) has an elegant and powerful formulation, first, in that the irreducible representations are parametrized very completely by the "theory of the highest weight" of Cartan, and, second, that the characters of these representations are known, given by the famous "Weyl character formula". See, for instance, [20] for these theories for complex semisimple Lie algebras. If we had such a parameterization and such a character formula for the finite groups G(IFq ) , not only would we would know their irreducible characters in the describing characteristic p of G (IFq), but we might learn something about the nondescribing case as well, where many analogies with the describing case have been discovered by Dipper and James [15], [16], [14], [21]. Working with the general linear group GL(n, q) they have found families of finitedimensional algebras, the q -Schur algebras, parameterized by a variable q, that control the nondescribing characteristic representation theory of the group GL(n, q) when q is taken to be a prime power (which may also be viewed as a kind of root of unity when the underlying field has positive characteristic), and which control the modular theory in the describing characteristic p when q = 1. If we knew both describing and nondescribing modular theory for the simple or nearly simple groups of Lie type, we could also hope to learn much about the maximal subgroup structure of all other finite groups [30], [2]. Indeed, this is a main organizational theme of the upcoming 1997 Newton Institute program at Cambridge. So, to summarize, it would be highly desirable to have for finite groups G (IFq) of Lie type a parameterization and character formula, as exist for the complex Lie groups G(C). Also, thanks to the work of Steinberg mentioned above, both issues for G (IFq) reduce to the corresponding problem for the algebraic group _G(JFq). Now it is time to tell you that the parameterization problem for G (IFq) was solved even before Steinberg's work by Chevalley, imitating the Lie-theoretic case G(C) mentioned above. Before discussing the character formula issue, let's consider how far we have come in discussing my reply to the philosophical issue, "Are all groups finite?" The first suggestion in my reply is that we cannot ignore G(C) and G (JFq ), and I hope the initial history above of the parameterization for describing characteristic representations of G(IFq ) , through Steinberg and Chevalley, duplicating Cartan's "theory of the highest weight", is convincing evidence of the usefulness of looking at continuous and algebraic groups. The problems are easier for these more richly structured groups, and some have been solved. The second suggestion in my reply, that we can abstract from these continuous contexts, and perhaps go beyond them, is evidenced
Are All Groups Finite?
135
by what has happened to the parameterization theory since that time: First, Curtis and Richen [12], [13], [29] showed it was possible to carry out an analog of the parameterization process, suitably modified, in any finite group with an appropriate split BN pair. Second, Alperin [1] demonstrated with his celebrated conjecture (a main theme of our conference) that it was possible to formulate a version of the parameterization which makes sense for any finite group! Perhaps those involved in the Classification might also say that it was useful to know that_corresponding simple group classifications already existed for G(C) and G(lFq ) , and the uniform theory of groups of Lie type which emerged was (and is) useful in efficiently dealing with many properties of known groups, and in formulating many general concepts. The structure in the Lie-theoretic case was not at all ignored, but, as above, it was only a starting point (together with involutions and the Odd Order Paper!) for a more general (and more elaborate) theory. Let's now go to the issue of a character formula for the describing characteristic representations of G (IFq). We are, I believe, far from a result as complete as the Classification, for irreducible modular representations of G(lFq ) . The few results we have put us at t.!lebeginning of the cycle, where it is still essential to learn from G(C) and G(lFq ) . Nevertheless, it has been part of the point of view of myself and my colleagues, especially in CPS (Ed Cline, Brian Parshall, and myself), to develop a theory as purely algebraic as possible, to both try and attain a more general theory and to allow for elaborations diverging from the cleanest cases. This point of view is also important in our approach to proving the current main conjecture, due to Lusztig. Before describing it, let me describe one of CPS's main algebraic abstractions, which will at least make the Lusztig conjecture easier to explain. The discussion is largely borrowed from my exposition [32].
1.1 Highest weight categories, and examples. Fix a field k, and let e be an abelian k-category (that is, e is abelian, all Hom sets are k-modules, and multiplications of morphisms is k-linear). In all cases we will consider here, e will simply be equivalent to the category of finite dimensional modules over a finite dimensional algebra, but it may not start out looking like that. We suppose the nonisomorphic irreducible objects L(A) to be indexed by the elements A of a poset A, called weights. For simplicity we will assume A is finite here; for a more general notion (requiring only that the intervals of A be finite), the reader is referred to [7]. We will also assume for simplicity that
136
Leonard L. Scott
all objects of e have finite length, that Hom sets between objects are finitedimensional over k, and, moreover, that Endc L(A) ~ k for each A E A. We assume that e has enough projectives, and let peA) denote the projective cover of L(A). We say that e is a highest weight category if there are objects V(A), A E A, such that (1)
has head L(A), and all other composition factors of V(A) have smaller weight than A. V(A)
(2) There is an epimorphism peA) -+ V(A) with kernel filtered by objects V (IL) with IL greater than A. These conditions imply that V (A) is the largest epimorphic image of peA) with A maximal among the weights of its composition factors. Such objects arise naturally in Lie-theoretic contexts. We call V (A) a Weylobject, since it is a Weyl module in our favorite context of characteristic p algebraic group representations. Other good names are Verma object, or simply standard object. Typically, these objects are well understood, and the main object of research is to write the irreducible objects in terms of them in the Grothendieck group (that is, to obtain their "Weyl character formula"). This is precisely what the Lusztig conjecture purports to do, with certain restrictions.
Three examples in Lie theory. (1) The example which first motivated CPS is the following: Let G = G(k) be
a semisimple, simply connected algebraic group over an algebraically closed field k of positive characteristic p (e.g. k = IFq). This is the most relevant example for finite group theory. In discussing it, I will assume some basic terminology from algebraic group theory, but the reader familiar with the basic theory of root systems and Lie algebras as found in [20] should be abl~ to follow much of it. Just keep in mind the basic example G = SL(n, IFq ) . There, T below is the group of invertible diagonal matrices over IFq, B is the group of upper triangular matrices and W is the group of n x n permutation matrices. The Lie algebra of G as a vector space is n x n matrices of trace 0, and its root spaces are just the 1- dimensional spaces with arbitrary entries in the i, j position, for fixed i f. j, and 0 's elsewhere. These are common eigenspaces for the action of T by_conjugation, and the associated homomorphism (character) mapping T to IF; is called a root in the world of algebraic groups, while an arbitrary algebraic group homomorphism from T to JF q is called a weight. Let T be a fixed maximal torus, and denote the root system of T acting on the Lie algebra of G by <1>. We choose a
Are All Groups Finite?
137
set <1>+of positive roots, and let B denote the corresponding Borel subgroup corresponding to the associated set <1>-of negative roots. The set X (T) of characters (weights) on T is partially ordered by the rule: A ~ fJ, {} nO'Ci for non-negative integers n«. We also have an induced fJ, - A = LO'E+ poset structure on the set X (T)+ of dominant weights (relative to <1>+).Fix any finite set Ao of dominant weights, let A be the (finite) set of dominant weights A for which A ~ Ao for some Ao E Ao. Then the category C of finite-dimensional G- modules (in the sense of algebraic groups) which have composition factors each with maximal T - weight in A is a highest weight category with weight poset A. The Weyl modules V(A) are obtained as linear duals of modules induced to G, in the sense of algebraic groups, from dominant weights in X (T) + extended to B. (These induced modules are all finitedimensional!) They may also be obtained by a reduction modulo p process from an irreducible module in characteristic O. As such, their decomposition into weights for T is directly obtainable from the Weyl character formula. Projecting e onto any block of G-modules also gives a highest weight category. If p ~ h, the Coxeter number of the root system, it is well-known [22] that the character formulas for all irreducible modules are deducible from those in the principal block. The weights for the latter are the dominant weights in the orbit Wp.O of 0 under the 'dot' action of the affine Weyl group Wp (defined by w. u. = w(/.l+p)-p, where p is the sum of all the fundamental dominant weights, for W E Wp and fJ, E X (T).) Also, Steinberg's tensor product theorem allows us to restrict attention to restricted weights, those with coefficients less than p when expressed in terms of certain 'fundamental' weights. Let us redefine A as the set of dominant weights which are in the orbit Wp.O and bounded above by a restricted weight in that orbit. Lusztig's conjecture may then be written chL(w.O)
=
L
(-I)f(W)-f(Y)PywO,wwo(l)ch
V(y.O),
y.OEA
for any weight w.O in A. Here y, ware in Wp and Wo denotes the long word in the ordinary Weyl group W. The terms PywO,wwo(l), are values at 1 of Kazhdan-Lusnig polynomials, which are defined in a purely combinatorial way for any pair of elements in a Coxeter group. Finally f.(w) denotes the length of w in the sense of Coxeter groups (the number of fundamental reflections in a minimal expression), and the function ch( -) just assigns an object of e to the associated element in the Grothendieck group of e. The conjectured formula would give character formulas for all irreducible G-modules, so long as p ~ 2h -3, and these would in turn give corresponding
138
Leonard L. Scott
character formulas for any finite group G(lFq ) of Lie type associated to G, with q a power of p. (Actually, so long as p ~ h, the above formula could hold for all restricted weights, and as such would have the same implications for finite groups, for such a p. This stronger version of Lusztig's conjecture was formulated by Kato [23], who apparently originally believed it to be a consequence of the original conjecture, but the arithmetic doesn't work out that way. Let me take this opportunity to mention that the first open cases for the Lusztig conjecture occur for SL(5, 5) and SL(5,7), which I have been examining with the help of an NSF undergraduate REV student. The former case is the first possibility for the Kato and Lusztig versions to diverge.) Lusztig obtained his conjecture by analogy with his conjecture with Kazhdan [24] for complex Lie algebras, which we describe next. (2) Let 9 be a complex semisimple Lie algebra, and fix a Cartan subalgebra fJand Borel subalgebra b containing fJ. Consider the corresponding category (J of BGG. The objects are the g-modules which are h-diagonalizable with finite-dimensional weight spaces (where a 'weight' here is a l-dimensional representation for the Lie algebra h ), and with the set of nonzero weights bounded above by some finite set of weights. We will also restrict attention to the case where all weights are integral; equivalently, they belong (by identification) to the set X (T) of characters for a torus T associated to fJ; these are just the integral linear combinations of the 'fundamental' weights for fJ. It is again true that any block of such modules forms a highest weight category, and all character formulas for irreducible modules are obtainable from the principal block case. The standard objects this time are the Verma modules M A , A E X(T), obtained by tensor induction of A at the universal enveloping algebra level from b to g. We write V(A) = M A , and let L(A) denote the irreducible head of V (A). The weights A indexing irreducible modules in the principal block are just those in the orbit A = W. - 2p. (This is also the orbit of 0 under the 'dot' action, since 0 = WQ. - 2p.) They correspond bijectively to elements of the Weyl group. The Kazhdan-Lusztig conjecture (now a theorem due to Brylinski-Kashiwara [6] and Beilinson-Bernstein [5]) reads chL(w.
- 2p) = L(-l)i'(W)-i'(y)
Py,w(1)ch
V(y. - 2p),
yEW
where again Py,w(1) is the value at I of a Kazhdan-Lusztig polynomial. The similarity of this formula and the previous one is remarkable, and all the more so when one considers that the standard modules in the first case are finite-dimensional, but infinite-dimensional here. The next case, is even more
Are All Groups Finite?
139
remarkable, in that we obtain precisely the same character formula for standard objects which are not modules at all, but complexes of sheaves. (3) A key ingredient in the proof of the Kazhdan-Lusztig conjecture was the Kazhdan-Lusztig formula for the stalk dimensions of the cohomology of perverse sheaves. It can be written as a character formula in the Grothendieck group sense we are using here, and we describe it below. Let X = G/ B denote the flag variety obtained from the simply connected semisimple complex Lie group G associated to the Lie algebra 9 above, and consider the category e of perverse sheaves on X with respect to the Schubert stratification and the middle perversity [4]. (Thus a stratum is a Schubert cell S (w) = B w B / B, w E W, and a perverse sheaf is a complex of sheaves of complex vector spaces with cohomology locally constant (thus constant) and finite-dimensional on Schubert cells, with certain support conditions satisfied.) The poset is W, with its Bruhat-Chevalley order, and the Weyl objects V (w), w E W, are quite easy to describe: Yew) = isCw)!C[f(w)], the extension by o of the constant sheaf, shifted downward as a complex in the derived category by degree few) Every Weyl object Yew) has a unique irreducible quotient L(w), and the axioms for a highest weight category are satisfied [28, §5]. Though unnecessary in our discussion, it is a remarkable fact that L (w) is the downward shift by f( w) of the complex (extended by zero to X) defining Goreski-MacPherson intersection cohomology on the closure of Sew); see [25] and [33]. The Grothendieck group formula of Kazhdan-Lusztig [25] reads ch L(w. - 2p) =
L (_I)£CW)-£CY)Py,w(l) ch V(y.
- 2p),
yEW
which is identical to the form of the Verma module Kazhdan-Lusztig conjecture above. Essentially, the latter conjecture was proved through an equivalence of categories reducing it to the above formula. 1.2 Quasihereditary algebras. Every highest weight category with finite weight poset and all objects of finite length is the category of finite-dimensional modules for a quasihereditary algebra S. Indeed, CPS introduced quasihereditary algebras for this reason, and proved that, conversely, the category of modules for a quasihereditary algebra could be viewed as a highest weight category [31], [281, [7]. We will not reproduce the axioms for a quasihereditary algebra here, but note that examples include hereditary algebras and poset algebras [28], as well as all finite-dimensional algebras of global dimension
140
Leonard L. Scott
two [17]. All quotient algebras of hereditary algebras are quasihereditary. Further Lie-theoretic examples of quasihereditary algebras include Schur algebras and q-Schur algebras, and their generalizations [7], [18]. CPS believes that understanding these algebras (and variations, with various degrees of added structure) will provide a good basis for understanding representations of algebraic groups in characteristic p, and finite groups of Lie type in describing or nondescribing characteristic. A new generalization of the quasihereditary notion, very relevant to the nondescribing characteristic case, is described in the last section of this paper. Every quasihereditary algebra S has finite global dimension. The opposite algebra Sop is quasihereditary with the same weight poset, and the S-module A(A) dual to the Weyl module VOP(A) for Sop has the following remarkable property [7; p. 98, bottom]
{k
if A = ~, otherwise. Here we have assumed, as before, that End L(A) = k to simplify the statement. Using this property, and an Euler characteristic argument of Delorme, it is possible to understand why the character formulas in each of the above three cases have such a remarkably similar appearance. Moreover, by tracking in the abstract setting a version (due to MacPherson, see [33]) of the arguments used to prove the Kazhdan-Lusztig formula in the perverse sheaf case, CPS was able to provide [8] (see also [9] and [10]) the following reductions. The 'length' f(A) of a weight A = w.O below is the number of simple reflections in a reduced expression for w. Extn(V(J,l), A(A» =
o
Theorem (The CPS reductions). In each of the three examples above, the Lusrtig conjecture or its analog (the Kazhdan-Lusztig conjecture, or Kazhdan-: Lusztig formula) is equivalent to each of the following statements: (1) For each A,
u. E A, Ext l (V (J,l), L(A» =j:.0 => f(A) - f(J,l) ==1 (mod
2).
(2) For each A E A, and each weight A' adjacent to A (in the sense that the affine Weyl group or Weyl group element associated to A' is obtained from that associated to A by right multiplication by a simple reflection), we have Ext! (L(A'), L(A») =j:.O. (By a duality principle, one may take here A' < A. or A' > A). (3) For each A, J,l E A, the natural map Ext! (L(J,l), L(A» ---+Ext! (V(J,l), L(A» is surjective.
Are All Groups Finite?
141
The most promising of these reductions is perhaps the second one, though each has its own advantages. When the Lusztig conjecture is true, versions of 1) and 3) hold with the number 1 replaced by n throughout (two replacements in 1) ), and dual versions hold using A(/L), cf. [8]; see also [9]. One may study such conditions purely from the point of view of finite dimensional quasihereditary algebras, though they are far from giving us a description up to isomorphism (or a suitable weaker invariant) of the algebras involved. Until we get that far, we are somewhat in the position of trying to prove deep properties of finite simple groups without knowing what all the simple groups are.
1.3 The Lusztig program. Recently, George Lusztig [27] has formulated an attack on his own conjecture organized around the theory of quantum groups. Work by himself and Kazhdan relates a Lusztig-type conjecture for quantum groups at a root of unity to its validity in a category of 'negative level' representations for affine Kac-Moody Lie algebras. The latter conjecture, at least for simply laced root systems, is settled by work of Kashiwara and Tanisaki (also claimed by Casian, who acknowledges his original proof was in error) by reduction to a category of perverse sheaves on a generalized flag variety. Ignoring the non simply-laced case difficulties", to complete the chain, one requires a reduction from quantum groups at a pth root of unity to algebraic groups in characteristic p. This has been provided for all types by Anderson, Jantzen and Soergel [3] for p sufficiently large, depending on the individual root system and its rank. Unfortunately, no specific bound whatsoever is known for p as of this writing. The problem is that p must stay away from divisors of the index in a maximal order of a certain algebra over ::E,and the algebra is constructed so indirectly that very little information on the index is available. As Jantzen himself reported at the 1994 Banff conference, this situation is simply not acceptable to finite group theory. While CPS thinks highly of the AJS work, we regard the Lusztig conjecture as open and continue to work on it.
1.4 Stratified algebras.
Another aspect of the situation is that CPS wants a theory sufficiently general to be appropriate for nondescribing characteristic, in the spirit of the Dipper-James work on the q-Schur algebra. Already there has been work by Dipper and others (see [19]) dealing with groups other than 1 These difficulties have apparently now been handled by Kashiwara-Tanisaki.
142
Leonard L. Scott
the general linear group. While it may be that quasihereditary algebras are involved in these cases, at least in favorable characteristics, CPS conjectures a role for a slightly more general kind of algebra. This new generalization is called a stratified algebra [11]. I describe first the most basic types of stratified algebras, the algebras with a standard stratification, which are quite close to quasihereditary algebras, and one weaker notion, algebras (whose module category is) equipped with a stratifying system. The description is quite easy to do in both cases if we just think about how we are to relax the corresponding notion (1.1) of a highest weight category: First, we relax the condition that the weights A form a poset, requiring only that they form a quasiposet, so that two weights x and IJ, may satisfy ).."::s:IJ, and IJ, ::s:).."without being equal. The equivalence classes thus obtained do themselves naturally form a poset A, and we let X denote the element of A associated with X E A. Next, in condition 1) for a highest weight category, we require only that all composition factors L(IJ,) of the standard object V()"") satisfy IJ, ::s:)..".(So that L()"") may appear twice or more, along with other composition factors L(IJ,) with X = fl.) The second condition 2) is kept in the "standard" case, and that completes the definition for that case. In the "stratifying system" case the inequality in 2) is relaxed to allow equality, but other relaxations are made as well: We just assume we have a system of objects V ()..,,)and given projective objects P()..,,)mapping onto each of these with kernel filtered by V (IJ,)'s with IJ, ~ )..". We do not require that V ()..,,)have an irreducible head, or that the quasiposet index the irreducible modules. As a replacement for the latter, we do insist that every irreducible module appear in the head of some V ()..,,). Condition 1) is replaced by the requirement that there are no nonzero homomorphisms from P()..,,)to V(IJ,) unless IJ, ::s:)..". In either case, one can prove from these conditions that the underlying algebra A has a sequence of idempotent ideals 0 = Jo C J1 C ... C I n , with n = IAI, such than Ext~/Ji (M, N) = Ext~(M, N) for all left (or right!) AIJ-modules M, N, all integers n ~ 0, and each index i. With mention of A omitted, this is the general CPS notion of a stratified algebra, with a stratification of length n. In the "standard" case, each J, I J, -1 is left projective as an A I J, -1 -module, and this is characteristic of standardly stratified algebras. The ideal need not be right projective, however. Unlike the quasihereditary case, there are algebras which are standardly left stratified but not standardly right stratified (even though the "general" notion of a stratified algebra is leftright symmetric). Apart from that, and the relaxed ordering inside standard modules, the theory of standardly stratified algebras is very close to that of
Are All Groups Finite?
143
quasihereditary algebras and highest weight categories. Note that the standard module V (A) is uniquely determined, in the "standard" case, as the largest quotient of peA) with all composition factors L(fl-) satisfying fl- ::: A. The notion of an algebra whose (left) module category has a stratifying system, and the general notion of a stratified algebra, are both weaker, but more flexible. For instance, an algebra which has a stratification of length n > 1 in the "general" sense also has one of length n - 1, obtained by removing any of the intermediate ideals in the above chain. (There is an analog ofthis statement for algebras with a stratifying system.) Also, as mentioned, the "general" notion is left-right symmetric, though this is not obvious. CPS has taken considerable trouble [11] to be able to recognize when an endomorphism algebra (e.g. a Schur algebra, q-Schur algebra, or future generalization) has a natural structure as a stratified algebra. This includes, of course, the quasihereditary case (easy to check in the presence of a standard stratification), but the generalizations are also interesting, and may be important. The recognition conditions are quite complicated, though simplify very considerably/ in special cases. They involve a kind of generalized "Specht module" theory. Rather than reproduce the conditions here, I will simply give some examples from [11], referring the reader to that paper for additional details, and further examples: Throughout k is an algebraically closed field. (I) A stratification of length 2. Let G be any nontrivial finite group. Consider the direct sum T of the trivial module and the regular permutation module over k. Then A = EndkG(T) is stratified, with IAI = 2 (the length of the stratification). Interesting cases occur already for G the cyclic group of order 2 or the Klein four group. In the first case, A is the wellknown algebra of dimension 5 which has two simple modules a and b, a b. In the second case, A is already not with projective covers band a
a
quasihereditary!. Its projective covers have Loewy layers as indicated by a
the diagrams
a
a a
band
b a
.
2 The simplifications in the preprint "Stratifying endomorphism algebras over Heeke algebras", by Du, Parshall, and Scott, over Z[q, q-l J, might even be described as dramatic.
144
Leonard L. Scott
Thus, unlike the quasihereditary case, the standard object associated to a (the quotient of the first projective cover by the submodule isomorphic to the second) has a appearing with a nontrivial multiplicity. Nevertheless, inside A, the ideal generated by the idempotent associated to the second projective cover is quite nice. It is both idempotent and (left) projective. This example and the next are both standardly stratified. All standardly stratified algebras have such ideals, and their factor algebras are also standardly stratified. (2) A stratification of length 3: the dihedral group of order 8. Let G be the dihedral group of order 8, and let T be the direct sum of the trivial module, the regular module, and the two transitive permutation modules of degree 4 associated to coset spaces of the two conjugacy classes of noncentral subgroups of order 2. We take char k = 2. This time the algebra A = EndkG T is quite difficult to visualize from the given data. The general CPS approach is to try to impose a generalized 'Specht module' filtration on T, and deduce from its properties that A is indeed stratified. Without giving full details, I will at least describe the 'Specht modules' we use. Let a and b denote generators of order two for G, and put A = a - I, B = b - 1. Thus A 2 = 0 = B 2 and ABAB = BABA. We will diagram cyclic modules by indicating where a nontrivial action of A or B occurs, starting from a generator. (No arrow associated to a given node and label indicates a zero multiplication. Note that, if a node was reached by multiplication by A, then that node must be killed by A. A parallel statement holds for B.) Thus, the four transitive permutation modules making up T have diagrams
A
.,
B
A
•
• /
-,
B
•
t
t
•
•
t
•
B
t
-,
/
•
•
A
A A B B
A
• t
•
t,
• t
•
•
B
t
A
t·
B
t
•
•
•
Here the single node. is also representative of the unique irreducible (trivial) module for the group G, and the above diagrams give refined Loewy series pictures. In the CPS set-up, each of these indecomposable components of T has a 'Specht filtration', which turns into the required filtration of projective covers by standard modules for the algebra A = EndkG (T) under the contravariant
Are All Groups Finite?
145
functor HomkG( -, T). (This functor is not exact, but a filtration of T still induces a filtration of A.) Moreover, each component above has a distinguished 'Specht' submodule, though it is possible for two such 'Specht' modules to be isomorphic. For the first and second components above, the Specht submodule is the l-dimensional trivial module, while for the third and fourth component, the Specht submodule is the unique 3-dimensional submodule. The reader will observe that, when the bottom and top trivial modules are eliminated from the second component, the remainder is the direct sum of the Specht modules associated to the third and fourth component. This puts all 'Specht filtrations' in evidence. The reader may consult [11] for some general machinery to check that these filtrations are transformed into the required standard module filtrations for the indecomposable projective components of A (or can attempt a direct verification starting from the filtrations and functor we have given). CPS believes something general is happening here for all Coxeter groups, that a similar construction always leads to a nontrivial stratified algebra. We have conjectured the following:
Conjecture. Let W be a finite Coxeter group with distinguished generating set 5, 151 > 1. For J ~ 5, let denote the permutation module for kW on the cosets {WjW}WEW, Put T = E9j~s T, and A = Endrw T. Then A is stratified with respect to a quasiposet A with IAI ~ 3.
r,
For a more detailed statement of the conjecture, see [~1]. We believe the stratification arises from a stratifying system, and that A may be assumed to have a largest and smallest element, containing only one element each as equivalence classes in A, with these elements associated to Ts (the trivial module) and the sign module submodule of T¢. The stronger form of the conjecture also has as a consequence the existence of certain known resolutions, one of which is the Coxeter complex. As the final version of this paper is being readied for press, it appears that Du, Parshall and Scott will soon prove the stronger conjecture, and a q-analog. 3 CPS also expects (with lie Du) that an enlarged version of A will have a standard stratification related to the filtration of the Tr's by dual left cell modules, in the sense of Kazhdan-Lusztig, of length equal to the number of two-sided cells. The algebra A itself exhibits such a stratification in the order 8 dihedral case above. This is a special case of a Weyl group of type B. In 3 This has now been done, in the preprint described in the previous footnote.
146
Leonard L. Scott
this case lie Du and I are working on another natural enlargement that may be quasihereditary." As mentioned above, it also seems likely that a q-analog of the conjecture holds for Heeke algebras, a possibility which makes the conjecture and stratified algebras quite relevant to nondescribing characteristic theory. This already important area of research will become even more central once the problems posed by the Lusztig conjecture itself are solved.
References [I]
J. Alperin, Cohomology in representation theory, Proc. Symp. Pure Math. 47 (1987).3-11.
[2]
M. Aschbacher and L. Scott, Maximal subgroups of finite groups, J. Algebra 92 (1985), 44-80. H. Andersen, J. Jantzen, and W. Soergel, Representations of quantum groups at a p th root of unity and of semisimple algebraic groups in characteristic p: Independence of p, Asterisque 220 (1994). A. Beilinson, J. Bernstein, and P. Deligne, Analyse et topologie sur les espaces singulares, Asterisque 100 (1982).
[3]
[4]
c. R. Acad.
[5]
A. Beilinson and J. Bernstein, Localisation des 9-modules, Paris 292 (1981),15-18.
[6]
J. Brylinski and M. Kashiwara, Kazhdan-Lusztig conjecture and holonomic systems, Invent. Math. 64 (1981),387-410.
[7]
E. Cline, B. Parshall and L. Scott, Finite dimensional algebras and highest weight categories, 1. Reine Angew. Math. 391 (1988),85-99.
[8]
E. Cline, B. Parshall and L. Scott, Abstract Kazhdan-Lusztig theories, Tohoku Math. 45 (1993), 511-534.
[9]
E. Cline, B. Parshall and L. Scott, Infinitesimal Kazhdan-Lusztig theories, Contemp. Math. 39 (1992), 43-73.
[10]
E. Cline, B. Parshall and L. Scott, Simulating perverse sheaves in modular representation theory, Proc. Symp. Pure Math. 56 (1994), 63-104.
[I1]
E. Cline, B. Parshall and L. Scott, Stratifying endomorphism algebras, Mem. Amer. Math. Soc. 591, Vol. 124,1996.
[12]
C. Curtis, Irreducible representations of finite groups of Lie type, 1. Reine Angew. Math. 219 (1965),180-199.
[13]
C. Curtis, Modular representations of finite groups with a split (B,N) pair, Seminar on algebraic groups and related finite groups (A. Borel. ed.), Lecture Notes in Math. 131, Springer-Verlag, 1970,57-95.
4 This is true. The preprint by Du and Scott is entitled "The q-Schu~ algebra".
Sci.
Are All Groups Finite?
147
[14]
R. Dipper, Polynomial representations of finite general linear groups in nondescribing characteristic, Prog. in Math. 95 (1991), 343-370.
[15]
R. Dipper and G. DJames, The q-Schur algebra, 1. London Math. Soc. 59 (1989), 23-50.
[16]
R. Dipper and G. D. James, q-tensor space and q-Weyl modules, Trans. Amer. Math. Soc. 327 (1991), 251-282.
[17]
V Dlab and C. Ringel, Auslander algebras are quasi-hereditary, 1. London Math. Soc. 39 (1989), 457-466.
[18]
1. Du and L. Scott, Lusztig conjectures, old and new, I, 1. Reine Angew. Math. 455 (1994), 141-182.
[19]
R. Geck and G. Hiss, Modular representations of finite groups of Lie type in non-describing characteristic, in: Finite reductive groups: Related structures and representations (M. Cabanes, ed.), Birkhauser, 1997, 195-249.
[20]
J. Humphreys, Introduction to Lie algebras and representation theory, Grad. Texts in Math. 9, Springer-Verlag, 1970.
[21]
G. D. James, The irreducible representations of the finite general linear groups, Proc. London Math. Soc. 60 (1990), 225-265.
[22]
1. Jantzen, Representations of algebraic groups, Academic Press, 1987.
[23]
S. Kato, On the Kazhdan-Lusztig polynomials for affine Weyl groups, Adv. in Math. 55 (1985), 103-130.
[24]
D. Kazhdan and G. Lusztig, Representations of Coxeter groups and Heeke algebras, Invent. Math.83 (1979), 165-184.
[25]
D. Kazhdan and G. Lusztig, Schubert varieties and Poincare duality, Proc. Symp. Pure Math. 36 (1980), 185-203.
[26]
G. Lusztig, Some problems in the representation theory of finite Chevalley groups, Proc. Symp. Pure Math.37 ( 1980),313-317.
[27]
G. Lusztig, Modular representations and quantum groups, Contemp. Math.82 (1989), 59-77.
[28]
B. Parshall and L. Scott, Derived categories, quasi-hereditary algebras, and algebraic groups, Mathematical Lecture Notes Series 3, Carleton University, 1988,1-105.
[29]
F. Richen, Modular representations of split BN pairs, Trans. Amer. Math. Soc. 140 (1969), 435-460.
[30]
L. Scott, Representations in characteristic p, Proc. Symp. Pure Math. 37 (1980), 319-331.
[31]
L. Scott, Simulating algebraic geometry with algebra, I: Derived categories and Morita theory, Proc. Symp. Pure Math. 152 (1987), 271-281.
[32]
L. Scott, Quasihereditary algebras and Kazhdan-Lusztig theory, Finite dimensional algebras and related topics (V Dlab and L. Scott, eds.), Kluwer, 1994, 293-308.
148
Leonard L. Scott
[33]
T. Springer, Quelques applications de la cohomologie d'intersection, Bourbaki Sem. (1981/1982).
[34]
R. Steinberg, Representations of algebraic groups, Nagoya Math. 1. 22 (1963), 33-56.
[35]
R. Steinberg, Endomorphisms of linear algebraic groups, Mem. Amer. Math. Soc. 80, Amer. Math. Soc., Providence, R.I., 1968.
Department of Mathematics Unversity of Virginia Charlottesville, VA 22903 U.S.A. Email: [email protected]
Locally Finite Varieties of Groups and Representations of Finite Groups Sergei A. Syskin
The purpose of this talk is to attract your attention to a new area of applications of representation theory of finite groups. All definitions and explanations can be found in [2]. Let S be a variety of groups. If G is a group, then S(G) is the smallest among normal subgroups M of G with G j M E S. It is easy to see that S(G) is the subgroup generated by all values W(gl, ...• gn) where w runs over all laws of the variety S and g) • ... gn E G. Such a subgroup is called verbal. Let F be the (absolutely) free group of countable rank. Then F jS(F) is the free group of countable rank for the variety S. We say that S has a finite base of laws if S(F) is finitely generated as a verbal subgroup of F. Every finite group generates a finitely based variety (Oates-Powell Theorem). The first examples of varieties having no finite base of laws have been constructed by A. Ol'shanskii [3]. Those varieties are solvable of exponent 8 pq where p and q are relatively prime odd integers > 1. In particular, they are locally finite. For some important locally finite varieties the problem of the existence of a finite base of laws remains open. For instance (see [1], question 4.48), the following problem is still open. Question 1. Is it true that any locally finite variety of groups with abelian Sylow subgroups is finitely based? Let us consider this problem. We will call such varieties A-varieties. First, two simple facts. Lemma 1. This problem has affirmative solution finitely based.
if each solvable A-variety is
So, we may consider only solvable A -varieties.
150
Sergei A. Syskin
Lemma 2. Let G be a finite group with abelian Sylow subgroups. Then the class of solvability of G is at most the number of prime divisors of IG I.
Let S be any solvable A -variety. Assume that S has no finite base of laws. Standard arguments show that S contains an infinite sequence of groups G I , G2, .. , with the following properties. (1) Every Gi, is a finite solvable group all of whose Sylow subgroups are abelian.
(2) All Gi have a restricted exponent and hence a restricted class of solvability by Lemma 2. (3)
Gi is not isomorphic to any factor of G) for i =j:. j (factor = factor-group of some subgroup).
(4)
G i is a critical group.
By definition, a group is critical if it does not lie in the variety generated by all its proper factors. Under these circumstances (Sylow sUbgroups in S are abelian) this means that every Gi has exactly one minimal normal subgroup Mi. Clearly, Mi is a p-group for some prime number p, and we may assume that p is the same for all Gi. Obviously, Pi = C (Mi) is a Sylow p -subgroup of G i. Hence G i/ Pi is a pi -group, that is of order prime to p. Moreover, we can assume that Pi = Mi' therefore
Thus, we have an ordinary faithful irreducible representation of Gi/ M i on
Mi. We have either to construct such a sequence G i satisfying (1)-(5), or to prove that it does not exist. If we are going to prove the nonexistence, we may assume something more. For example, we may assume that
(2 ') All Gi have the same exponent and the same class of solvability. Let Rand S be solvable finite groups of the same class n. We may define their derived subgroups: R(l) = R, R(i+I) = [R(i), R(i)]. We say that R and S are of the same type if the exponent of R(i) / R(i+I) equals the exponent of S(i) / S(i+I) for all i = I, ... ,n. Clearly, we can assume that (6) All Gi are of the same type.
Locally Finite Varieties
151
Using induction on the type, we can assume that our conjecture holds for the sequence G I Mi. It is a simple matter to prove that the last sequence has an infinite subsequence which is a chain of groups. Hence we may assume that (7)
Gi + I I Mi + I has a subgroup isomorphic to G i/ Mi for all i = 1, 2, ....
Of course, one may assume something else. For example, in the definition of type one can consider the series R(i) instead of RU) where R(i)1 RU+I) is the (unique) largest abelian normal subgroup of RI RU+I). Perhaps, this alternative definition is more helpful since all ordinary representations under consideration are monomial over some extension of G F(p). Thus, we may propose the following
Question 2. Let RI .::: R2 .::: '" be an infinite chain of finite solvable groups with abelian Sylow subgroups. Assume that all Ri have the same class of solvability and the same exponent. Let p be a prime which does not divide this exponent. Let Mi be a fixed faithful irreducible G F (p) Ri -module. Is it true that one of these groups, say Rj, contains a subgroup H isomorphic to Ri, i i= j, such that, for some H -submodule V of Mj, two natural semidirect products Ri Mi and V H are isomorphic? Question 3. Let p be a prime, A p the variety of all abelian groups of exponent p, and let N be a locally finite nilpotent variety such that p does not divide the exponent of N. Is it true that any subvariety of the variety ApN is finitely
based? It is known that any nilpotent variety has a finite base of laws, see [2].
References [1]
Kourovka notebook, A collection of unsolved questions in group theory, 11th edition, Novosibirsk 1990.
[2]
H. Neumann, Varieties of groups, Springer-Verlag, 1967.
[3]
A. Yu. Ol'shanskii, On the problem of a finite basis of identities in groups, Math. USSR-Izvestija 4 (1970),381-389.
Reinsurance Group of America 660 Mason Ridge Center Drive 5t Louis MO 63141
U.5.A.
Contains articles bymanyof theparticipants at theConference on Modular Representation Theory which tookplaceunder theauspices of theMathematical Research lnstitute of theOhioStateUniversity withadditional support fromthe National Foundation.The Science conference celebrated thecareer of Professor V/alter Feitof Yale University ontheoccasion of his65thbirthday. lt alsoroughly coincided withthe centenary of thetheoryof groupcharacters. Theparticipants included mostof the leaders in thefieldof finitegrouprepresentations andthisvolumecontains and expands on manyof thestimulating lectures fromtheconference. Thisvolume is addressed to specialists andstudents in thefieldof grouprepresentation theory, including boththegeneral theory andthespecial theories related to the groupsandthefinitegroupsof Lietype.Latest representations of thesymmetric reportsaregivenon important conjectures including theLusztig conjecture, the Alperin-McKay-Dade conjectures andthekGV)problem. lmportant newresearch avenues areilluminated, including thetheory of infinite-dimensional modules forfinite groupsandthetheoryof Rickard equivalences of modulecategories, Bothexperts andgraduate students willfindthebooka fascinating introduction to a widearrayof problems newtechniques exciting and attheresearch level.
rssN0942-0363 rsBN3-11-015806-X