Canadian Mathematical Society Societe mathematique du Canada Editors-in-Chief Ridaaeurs-en-chef Jonathan M. Borwein Peter Borwein
Springer New York Berlin Heidelberg Barcelona Hong Kong London Milan Paris Singapore Tokyo
CMS Books in Mathematics Ouvfages de mathematiques de 10 SMC 1 2 3 4 5 6
HERMAN/KuCERAI$IMSA Equations and Inequalities ARNOLD Abelian Groups and Representations of Finite Partially Ordered Sets BORWEIN/LEWIS ConvexAnalysis and Nonlinear Optimization LEVIN/LuBINSKY Orthogonal Polynomials for Exponential Weights KANE Reflection Groups and Invariant Theory PHILLIPS Two Millennia of Mathematics: From Archimedes to Gauss
David M. Arnold
Abelian Groups and Representations of Finite Partially Ordered Sets
Springer
David M. Arnold Department of Mathematics Baylor University Waco, TX 76798-7328 USA
Editors-in-Chief Redacteurs-en-chef Jonathan M. Borwein Peter Borwein Centre for Experimental and Constructive Mathematics Department of Mathematics and Statistics Simon Fraser University Burnaby, British Columbia V5A IS6 Canada
Mathematics Subject Classification (2000) : 16G20, 20Kxx Library of Congress Cataloging-in-Publication Data Arnold, David M. Abelian groups and representations of finite partially ordered sets I David M. Arnold . p. em. - (CMS books in mathematics) Includes bibliographical references and indexes . ISBN 0-387-98982-X (hardcover : alk. paper) 1. Abelian groups . 2. Partially ordered sets. I. Title. II. Series. QAI80 .A76 2000 512'.2-dc21 99-087081 Printed on acid-free paper. © 2000 Springer-Verlag New York, Inc. All rights reserved . This work may not be translated or copied in whole or in part without the written permission of the publisher (Springer-Verlag New York. Inc., 175 Fifth Avenue, New York, NY 10010, USA), except for brief excerpts in connection with reviews or scholarly analysis . Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use of general descriptive names, trade names. trademarks. etc., in this publication . even if the former are not especially identified. is not to be taken as a sign that such names, as understood by the Trade Marks and Merchandise Marks Act. may accordingly be used freely by anyone. Production managed by Michael Koy; manufacturing supervised by Jeffrey Taub. Typeset by Techlsooks, Fairfax. VA. Printed and bound by Edwards Brothers, Inc., Ann Arbor. MI. Printed in the United States of America. 9 8 765 432 I ISBN 0-387-98982-X Springer -Verlag New York Berlin Heidelberg
SPIN 10751459
To my wife, Betty, for her deep affection, strong encouragement, extraordinary patience with this project, and repeated reminders ofthe lessons ofPoseidon.
Preface
A recurring theme in a traditional introductory graduate algebra course is the existence and consequences of relationships among different algebraic structures. This is also the theme of this book, an exposition of connections between abelian groups and representations of finite partially ordered sets. Emphasis is placed throughout on classification, a description of the objects up to isomorphism, and computation of representation type, a measure of when classification is feasible. The subject of representations of finite partially ordered sets over a field is deeply rooted in classical linear algebra . Questions such as similarity and equivalence of matrices and simultaneous similarity and equivalence of pairs of matrices can be interpreted as questions about classification of representations up to isomorphism. Fundamental theorems about representation type were originally proved in the early 1970s by matrix arguments. More recently, the subject of representations of partially ordered sets over fields has been subsumed into the more general, and more sophisticated, subject of finitely generated modules over finite-dimensional algebras . Abelian group theory has a rich tradition of classification results. In the torsion-free case, finite direct sums of subgroups of the additive group of rational numbers were characterized by 1937. After such a promising start, by the early 1960s it became clear that torsion-free abelian groups were ju st too complicated to admit any reasonable classification scheme . As Irving Kaplansky wrote about torsion-free abelian groups of finite rank, "In this strange part of the subject anything that can conceivably happen actually does happen" [Kaplansky 69]. A breakthrough occurred in 1968. An equivalence between quasi-homomorphism categories of certain torsion-free abelian groups of finite rank, now called Butler groups, and representations of finite partially ordered sets over the field of
viii
Preface
rational numbers was published by M.C.R . Butler. A survey paper by the same author in 1987 contained a suggestion that the well-developed theory of representations over fields be applied to abelian group theory. Implementation of this suggestion and its natural extensions has introduced some order into the previously chaotic subject of torsion- free abelian groups of finite rank. This book is partially motivated by the scarcity of published books and monographs on representations of partially ordered sets over fields. [Ringel 84] and [Simson 92] are refreshing exceptions but not introductory in the sense that the focus is on a more general setting. A second motivation is the absence of an expository introduction to finite rank Butler groups . Standard texts for abelian group theory were published before the development of much of the theory. Chapter 1 is an elementary introduction to fundamental techniques for, and properties of, representations of finite partially ordered sets over a field. Basic notions of countable torsion-free abelian groups and relationships between finite rank Butler groups and representations of finite partially ordered sets over a field constitute Chapters 2 and 3, respectively . Recently discovered relationships between Butler groups and representations of finite partially ordered sets over discrete valuation rings are included in Chapter 4. In contrast to representations over fields, very little of a theoretical nature is known about representations over discrete valuation rings . Nonetheless, representation type can be computed, primarily by the use of traditional matrix techniques. Chapter 5 is dedicated to almost completely decomposable groups, a special case of finite rank Butler groups that is currently receiving a great deal of attention in the literature. Computations of representation type of representations of partially ordered sets over discrete valuation rings lead to explicit procedures for constructing indecomposable almost completely decomposable groups of arbitrarily large finite rank with types in a fixed finite set of types. Such constructions are quite difficult without the assistance of techniques from representation theory. Included in Chapters 6 and 7 is a brief introductory development of Coxeter correspondences and almost split sequences within the context of representations and finite rank Butler groups, a realization theorem for endomorphism rings of finite rank Butler groups, and classification results for a special class of finite rank Butler groups called bracket groups. Chapter 8 is devoted to applications to finite rank torsion-free modules over a discrete valuation ring and finite valuated p-groups. These applications are intrigu ing in that representations bridge the gap between the apparently disparate subjects of finite rank Butler groups, finite valuated groups, and p-Iocal torsion-free abelian groups of finite rank. No attempt has been made to be comprehensive or encyclopedic, due to the breadth of the topics discussed and the voluminous literature. Topics are selected to support the general theme of this book. A notable omission is any mention of uncountable groups, an active area of research requiring set-theoretic techniques beyond the scope of this book. Notes at the end of each chapter provide a brief guide to the published literature. In the interest of space and clarity,
Preface
ix
some long , highly technical proofs are not included , but in all cases a complete proof is referenced for the interested reader. Numerous open questions arise and some of these are mentioned at the end of the appropriate section. Explicit constructions and examples are emphasized throughout ; pure existence proofs are rare. Although the reader will witness repeated interaction between such general topics as linear algebra, rings, modules, categories, abelian groups both torsion and torsion-free, and combinatorics, the development is intended to be relatively elementary. A beginning graduate algebra text, such as [Hungerford 74], should be sufficient background. In a few instances, ancillary material is summarized and referenced. A graduate-level course taught by the author at the University of Essen , Germany, in 1996 provided the impetus for this book. Class notes from the lectures were taken by the students . These notes provided the foundation for the first three chapters , an examination of relationships between quasi-homomorphism categories of countable Butler groups and representations of finite partially ordered sets over fields. Other graduate-level courses are possible from this book, such as "Introduction to Representations of Finite Posets ," Chapters 1, 4.1-4.2, and 6, and "Introduction to Torsion-Free Abelian Groups of Countable Rank," Chapters 2.1-2.4, 3.1, 3.2, 3.4,5 .1,5.4, and 7.
Acknowledgments Support and facilities during the preparation of this manuscript were provided, in part, by the Ralph and Jean Storm Endowment Fund ; Baylor University College of Arts and Sciences, Department of Mathematics, and Summer Sabbatical Program; German-Israeli Foundation; University of Essen, Germany ; and University of Connecticut, Storrs. Special thanks to Professor Ed Oxford for his encouragement and Professors Rudiger Gobel and Charles Vinsonhaler for their encouragement and kind hospitality. Much of the work on the interaction of representations of partially ordered sets over discrete valuation rings and abelian groups has been done in conjunction with Professor Manfred Dugas . His insight and computational skills have been an inspiration. Motivation for an introductory book on representations of partially ordered sets was provided by the experience offormer Baylor University students Rebekah Hahn, Amy Maddox, and Mary Alice Mouser, in the preparation of their master's theses . Thanks to Rebekah Hahn for permission to include some diagrams and observations from her thesis. I am grateful to Professor EL Lady for the opportunity to read a preliminary manuscript on Butler modules over Dedekind domains. My appreciation to Professors S. Brenner, M.C.R. Butler, L. Fuchs, R. Gobel, H. Krause, A. Mader, K. Rangaswamy, F. Richman, C. Ringel, D. Simson, A. Skowronski, C. Vinsonhaler, and B. Zimmermann-Huisgen, among others , for illuminating conversations about abelian groups and representations of partially ordered sets.
x
Preface
A number of people have patiently listened to lectures about the contents of this book. Many thanks to Professors S. Files, S. Glaz, R. Gobel, J. Hurley, A. Paras, S. Pabst, J. Reid , E. Spiegel, C. Vinsonhaler, and W. Wickless for their helpful comments. My gratitude to A. Elder, G. Hennecke, A. Opendhovel, and L. Striingmann for their assistance with the lecture notes, 1. Fang and M. Kanuni for reading a preliminary version, and Roxie Ray for her skillful assistance with the preparation of this manuscript.
Contents
Preface
vii
1 Representations of Posets over a Field 1.1 Vector Spaces with Distinguished Subspaces 1.2 Representations of Posets and Matrix Problems 1.3 Finite Representation Type . . . . . 1.4 Tame and Wild Representation Type 1.5 Generic Representations . . . . . . .
1 I 10 19 31
2 Torsion-Free Abelian Groups 2.1 Quasi-isomorphism and Isomorphism at p . 2.2 Near-isomorphism of Finite Rank Groups . 2.3 Stable Range Conditions for Finite Rank Groups 2.4 Self-Small Groups and Endomorphism Rings 3 Butler Groups 3.1 Types and Completely Decomposable Groups . 3.2 Characterizations of Finite Rank Groups . . . . 3.3 Quasi-isomorphism and Q>-Representations of Posets . 3.4 Countable Groups . . .. 3.5 Quasi-Generic Groups . . . . . . . . . . . . . 4 Representations over a Discrete Valuation Ring 4.1 Finite and Rank-Finite Representation Type . 4.2 Wild Modulo p Representation Type . . . . . 4.3 Finite Rank Butler Groups and Isomorphism at p
39
47 47 56 62
69
76 76
88 100 105 III
126 126 135 140
xii
Contents
5 Almost Completely Decomposable Groups 5.1 Characterizations and Properties . . . . . . 5.2 Isomorphism at p and Representation Type 5.3 Uniform Groups . . . . . . . . . . . . . . . 5.4 Primary Regulating Quotient Groups . . . .
. . . .
. . . .
. . . .
. . . .
6 Representations over Fields and Exact Sequences 6.1 Projectives, Injectives, and Exact Sequences . . 6.2 Coxeter Correspondences . . . . . . . . . . . . 6.3 Almost Split Sequences . . . . . . . . . . . . . 6.4 A Torsion Theory and Localizations . . . . . .
. . . .
. . . .
. . . .
7
. . . .
Finite Rank Butler Groups
. . . .
. . . .
. . . .
. . . .
. . . .
. . .. .. ..
144 144 154 164 169 173 173 178 187 191 197
7.1 Projectives, Injectives, and Exact Sequences. . . . . . . . . . .. 197 7.2 Endomorphism Rings . . . . . . . . . . . . . . . .. 202 7.3 Bracket Groups. . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
8 Applications of Representations and Butler Groups 8.1 Torsion-Free Modules over Discrete Valuation Rings 8.2 Finite Valuated Groups
211 211 218
References List of Symbols Index of Terms
223 235 241
1 Representations of Posets over a Field
1.1
Vector Spaces with Distinguished Subspaces
Finite-dimensional vector spaces with finite sets of distingui shed subspaces are illustrative examples of representations of finite posets . This provides a natural setting for equivalence and similarity of matrices, as demonstrated in the exercises . Letk be a field and Sn denote the set {I , 2, . . . , n} of integers. A k-representation of S; isan (n+ Ij-tuple V = (Vo, VI, ... , Vn)consisting ofa finite-dimensional kvector space U« with n distinguished subspaces VI, . . . , V n • A morphism I : V ~ V ' of representations is a k-linear transformation j : U« ~ V~ with !(Vi ) 5; V; for each i . Two representations V and V' are isomorphic if there is a representation morphism ! : V ~ V' such that I : Vo ~ V~ is a vector space isomorphism. Equivalently, there is a representation morphism g : V' ~ V with I g = I tr , the identity morphism of V ' , and gl = Iv. In this case, I is called an isomorphism. If ! is an isomorphism from V to V ', then I : Vi ~ V; is a vector space isomorphism for each i. The set of representation morphisms V ~ V , called endomorphisms of V, is denoted by End V . It follows immediately from the definition of a representation morphism that End V is a ring under the operations of addition and composition of morphisms. In particular, End V is a subring of the ring End Vo of k-linear transformations of Vo. The field k is a subring of End V, since scalar multiplication on Vo by an element of k is an endomorphism of V. An element I of End V is idempotent if 12 = I . If I is idempotent, then I - I is also idempotent, since (l - Ii = 2 I - 21 + 1 = I - I.
2
I. Representations of Posets over a Field
The direct sum of two repre sent ation s V = (Vo, Vi : 1 .::::: i .::::: n ) and V ' i .::::: n) is the representation
( V~, V( : I .:::::
=
V EB V ' = (Vo EB V~ ' VI EB V ;, ... , U; EB V~ ) ,
where V i EB V ( de notes the vector space direct sum of V i and V(. A represe ntation V is decomposable if there is a vector space decomposition V o = Vo EB Wo with V i = (Vo n V i ) EB ( Wo n Vi ) for each i and both Vo and Wo nonzero . In this case, V = (Vo, V n Vi : 1 .::::: i .::::: n ) and W = (Wo, W n Vi : I i .::::: n ) are called direct summa nds of V. A representation V is indecomposable if = (0, 0, .. . , 0) and V are the only direct summands of V . Since Ue has finite dimension , any repre sentation V can be written as the direct sum of finitely many indecomposable representations.
s
°
Lemma 1.1.1 A k-representation V of S; is indecomposable and 1 are the only idemp otents of End V .
if and only if
°
PROOF. Let V = (Vo, VI , .. . , V n) be a representation and f 2 = f E End V. Then V = (f (Vo), f(V I ),· . . , f(Un )) EB «(1 - f) (Vo), (1 - f) (Vd ,··· , (1 - f)( V n)) is a direct sum of repre sentations. To confirm this assertion, two conditions must be met. The first condition is that Vo = f (Vo) EB (1 - f)( V o) as vector spaces . This condition follow s from the observations that uo = f (uo) + ( 1 - f)( uo) for each Uo E V o and f (Vo) n (1- f) (Vo) = 0, noticing that if x = f (u o) = (1- f) (vo) E f( vo)n(1 - f)( Vo), then x = f 2(uo) = f (x ) = f (1 - f)(vo) = ( f - f 2)(vO) = O. The second condition is that for each i, [( Ui) = [ (Uo) n U, and (1 - f)( Ui) = (I - f)( Vo) n Vi ' Thi s condition is satisfied for the idempotent f , since f( V i) ~ f( Vo) n V i and if x = f (uo) E f (Vo)nvi , then x = f (u o) = f 2(uO) = f (x ) E f( Vi) . Since 1- f is also idempotent, a similar argument shows that (I - f)( Vi ) = ( I - f)( Vo) n V i , as desired. If f is neither nor 1, then V is decomposable. Consequently, if V is indecomposable, then and 1 are the only idempotents of End V. Conversely, assume that V = V EB W is a direct sum of representations with both Vo and Wo nonzero. Then U» = VoEB Wo with each Vi = (VonUi ) EB(WOnV i) . Let f be a vector space projection Uo ~ Vo defined by f(vo, wo) = Vo · Then f is a representation morphism of V , because f (Vi) ~ Vo n Vi for each i , Furthermore, f is clearly an idempotent, and f is neither nor 1, since the kernel of f is Wo =1= V o and the image of f is Vo =1= Vo. It follows that if and 1 are the onl y idempotents of End V , then V must be indecomposable.
°°
°
°
°
Representations of the form (Vo, 0 , . . . , 0) with Uo =1= are called trivial. Trivial represent ations can be viewed as vector spaces. For inst ance , two trivial representations V and V are isomorphic as representations if and only if Uo and Vo have the same dimen sion as vector spaces. A trivial represent ation is a finite direct sum
1.1
Vector Spaces with Distinguished Subspaces
3
of indecomposable trivial representations, since if k" denotes an n-dimensional vector space, then
(e, 0, . .. , 0) = (k, 0, . . . , 0) 61 .. . 61 (k, 0, . .. ,0) is a direct sum of representations. Hence , up to isomorphism, (k, 0, . . . , 0) is the only indecomposable trivial representation. (Uo, UI, ... , Un) be a representation. Then Wo UI + ... + Un, Let U the subspace of Uo generated by UI. . . . , Un, is a vector space summand of Uo, say Uo Vo 61 Woo Hence, U is the direct sum of the trivial representation (Vo , 0, . . . , 0) and the representation (Wo, U I, . . . , Un)' In particular, if U has no trivial summands, then Uo = U I + ...+ Un' The set of isomorphism classes of indecomposable representations of Sn over a field k is denoted by Ind(n , k). For small n, the indecomposable representations of Sn are easily described.
=
=
=
Example 1.1.2 Elements oflnd(l , k) are U = (k , k) and U = (k , 0). In each case, End U = k. PROOF. As noted above, (k, 0) is the only trivial indecomposable representation
of SI' If U = (Uo, UI) has no trivial summands, then Uo = UI is a k-vector space, say of dimension m. Then U is isomorphic to (k, k)m, the direct sum of m copies of the representation (k , k). Hence , U is indecomposable if and only if m = 1. If U = (k, k) or (k, 0), then End U = k, since k S; End U and k-endomorphisms of k are just multiplication by elements of k. But and 1 are the only idempotents of the field k, whence U is indecomposable by Lemma 1.1.1.
°
Example 1.1.3 Elements of Ind(2, k) are (k, 0, 0), (k, k, 0), (k, 0, k), and (k , k, k). In each case, the endomorphism ring is k. PROOF. The representation (k, 0, 0) is the only trivial indecomposable represen-
tation of S2. Assume that U is a nontrivial indecomposable representation of S2 . Then U = (Uo, UI , U2) with Uo = UI + U2. If UI n U2 = 0, then Uo = UI 61 U2 and V = (VI, VI, 0) 61 (V2, 0, V2). Since U is indecomposable, U is isomorphic to either (k , k , 0) or (k, 0, k). Next, assume that UI n U2 f and write Uo = (VI n U2) 61 V for some subspace V of U«. Since UI n U2 S; Ui, U, = (VI n U2) 61 (Ui n V) for i = 1,2. Consequently,
°
°
As U is indecomposable and UI n U2 f 0, it follows that V = and U is isomorphic to (k, k, k). The argument that each of the U 's listed has endomorphism ring k, hence is indecomposable, is the same as that for Example 1.1.2.
4
1. Representations of Posets over a Field
Thefirstthreeindecomposablerepresentations of S2listedin Example 1.1.3have at least one subspaceequal to 0. Deletinga zero subspace givesan indecomposable representation of SI. Thus, (k, k , k) is the only indecomposablerepresentation of S2 that does not arise from a representation of SI. This is an illustrationof the next proposition. Proposition 1.1.4 Given positive integers m ::: n, there is a correspondence F from k-representations of Sm to k-representations of Sn suchthat if V and V' are k-representations of Sm, then End V = End F(V) and V is isomorphic to V' if and only if F(V) is isomorphic to F(V'). Moreover, F : Ind(m, k) ~ Ind(n, k) is a one-to-one function. PROOF.
The correspondence F is defined by V
= (Vo, VJ, .. . , Vm) ~
F(V)
= (Vo, VI,. ' "
Vm, 0, . . . ,0).
It is easy to confirm that End V = EndF( V) and if V and V' are k-representations of Sm, then V is isomorphic to V' if and only if F(V) is isomorphic to F(V'). As an application of Lemma 1.1.1, V is indecomposable if and only if F(V) is indecomposable. Thus, F: Ind(m, k) ~ Ind(n, k) is a well-defined one-to-one function. Elements of Ind(3, k) can also be listed, but the proof of completeness of the list is slightly more complicated than for the case n ::: 2. Two arguments are given. The first argument, Example 1.1.5, is technical but does illustrate the solution of a matrix problem, while the second argument, a reduction to Ind(2, k), is Corollary 1.2.4. Givena finite-dimensional k-vectorspace V, let (l + 1)V = {(x, x): x E V}, the image of the diagonal embedding of V in V $ V . Write V $ = {(x, 0) : x E V} and 0$ V = {(O, x) : x E V} for the coordinate spaces of V EB V. Then (1 + I)V , V $0, and 0$ V aresubspaces of V $ V and (V $ V, V EBO, 0$ V, (1 + I)V) is a k-representation of S3'
°
Example 1.1.5 Elements oflnd(3. k) are
°
(k, 0, 0, 0), (k, k, 0, 0), (k, 0, k , 0), (k, 0, 0, k), (k, k , k, 0), (k, 0, k, k) , (k, k , 0, k) , (k, k, k, k), (k $ k, k $ 0, $ k, (1 + l)k) .
In each case, the endomorphism ring is isomorphic to k. Each of the representationslisted has endomorphismring k, hence is indecomposable by Lemma 1.1.1. In particular, for each of the first eight representations, U« = k, so that End V = k. For the last case, let V = (k $ k, k $ 0, 0$ k , (1+l)k) and f E End V. Since f preservesthe twocoordinatespaces, f = (fl, !z) with each f i a k-endomorphismof k. But f also preservesthediagonalembedding,
PROOF.
1.1
Vector Spaceswith Distinguished Subspaces
5
whence II = fz E k. It follows that k = End V , recalling that k is contained in End V via scalar multiplication. The remainder of the proofconsists in confirming that the list is complete. In view of Proposition 1.1.4, Ind(2, k) can be embedded in Ind(3, k) for each two-element subset 52 of 53. This, together with Example 1.1.3, accounts for the first 7 representations in the list, a complete list of those indecomposables V = (Va, VI , V 2, V3) with some Vi = O. Suppose V = (Va, VI, V2, V3) is an indecomposable representation of 53 with each Vi =1= O. Then Va = VI + V2 + V3' since V is nontrivial and indecomposable. First, assume that W = VI n V 2 n V 3 =1= O. Write Va = WEB V for some V so that Vi = WEB (V n Vi) for each i, since W ~ Vi . Consequently,
V
= (W, W, W, W) EB (V, V n VI , V n V2, V n V3)
is a direct sum of representations. Since V is indecomposable with W =1= 0, it follows that V = 0 and V is isomorphic to (k , k , k , k). Next, assume that VI n V2 n V3 = O. Then Vi n U, = 0 for each 1 ~ i =1= j ~ 3. To see this, suppose, for example, that VI n V 2 =1= O. Then Va = (VI n V 2) EB V for some V ;2 V 3, using the assumption that VI n V 2 n V3 = O. Hence, Vi = (VI n V 2 ) EB (Vi n V) for i = 1,2, and so
But V is indecomposable with VI n V 2 =1= O. Thus , 0 = V ;2 V3, a contradiction to the assumption that V 3 =1= O. The arguments for the remaining cases are similar. Atthis stage, the assumptions are that V = (Va, VI, V 2, V3) is indecomposable, U« = VI + V2 + V3 with each Vi =1= 0, and Vi n U, = 0 for 1 ~ i =1= j ~ 3. Write V3 = (V 3 n (VI EB V 2 )) EB V for some vector space V and observe that Va = VI + V2 + V3 = (VI EB V 2) EB V. Then
Since V is indecomposable with each Vi =1= 0, it follows that V = 0 and VI EB V2 =
Va;2 V3.
The only case remaining is the case that V = (Va, VI, V2, V3) is indecomposable, Va = VI EB V2, each Vi =1= 0, and Vi n V3 = 0 for i = I , 2. This case can be interpreted as a matrix problem. Let B I = {XI , , x r } be a k-basis for VI, B2 = {YI, . . . , Ys} a k-basis for V2 , and B3 = {ZI, , Zt } a k-basis for V3, a subspace of Va = VI EB V 2. For each i ,
In particular, V3 is the row space of a tx(r M
+ s) k-matrix
= (A I B),
6
1.Representations of Posets overa Field
where A = (aU) is a t x r k-matrix and B = (bi j ) is a t x s k-matrix. Columns of A are labeled by B I , columns of B are labeled by B2, and rows of M are labeled by B3. The matrices A and B are referred to as block matrices. Obviously, M depends on a choice of bases for the Vi'S, but V does not. The following invertible matrix operations on M do not change V : (a) Elementary column operations within A (a basis change for VI), (b) Elementary column operations within B (a basis change for V2), (c) Elementary row operations on M (a basis change for V3). Write M :::::: N if the matrix N can be obtained from M by a sequence of operations (a), (b), (c). Using (a) and (c), the block matrix A can be reduced to echelon form
with I denoting an identity matrix. This echelon matrix must actually be of the form (I 0) because the block matrix A cannot have a row of zeros. Otherwise, there is some row in M having nonzero entries only in the columns indexed by a basis of V2. This row would then denote a oasis element of V3 that is also an element of V2 n V3 = 0, a contradiction. In fact, A also cannot have a column of zeros. To see this, suppose that the jth column of A is all zeros. Then VI = kx, EB VI for some VI with V 3 ~ VI EB V 2 , and so
Since U is indecomposable and kx] #- 0, U is isomorphic to (k , k , 0, 0), a contradiction to the assumption that each Vi #- O. Consequently, M:::::: (I I B)
for some B with I a t x t identity matrix, and so t operation E on (I I B) yields
= r , An elementary row
M:::::: (EIEB). However, using (a), (E I EB):::::: (I
= EE-II EB).
Hence, M:::::: (I I EB) for any elementary row operation E on M . Now use (b) and (c) to reduce B to echelon form I, noting that B cannot have a row or column of zeros for the same reasons that A cannot. As a consequence of
1.1 Vector Spaces with Distinguished Subspaces
7
the preceding remarks, it follows that
M
~
(Ill).
In particular, t = r = sand V3 is isomorphic to EEl{k(xj + Yj): 1 ~ j ~ t}, corresponding to the row space of (I I l). Thus, V is isomorphic to EEl{(kxj EEl ky], kx], ky], k(xj + Yj»: 1 ~ j ~ t}, a direct sum of representations. Since V is indecomposable, V must be isomorphic to (k EEl k , k EEl 0, 0 EEl k, (1 + l)k). This completes the proof that an indecomposable representation V of 53 is isomorphic to one of the nine representations listed . The preceding examples confirm that Ind(n, k) is finite and the endomorphism ring of each indecomposable k-representation of 53 is isomorphic to k for n ~ 3. The following two examples demonstrate that neither of these properties holds for n ~ 4. Let k[x] denote the polynomial ring with coefficients in k .
Example 1.1.6 (a) Ind(n, k) is infinite for n ~ 4. (b) For each irreducible polynomial g(x) in k[x] and each integer e ~ 1, there is an indecomposable k-representation V of 54 with End V = k[x]/(g(xY) . PROOF. Let A be an m x m k-matrix and define
where (1 + A)k m = {(x, Ax) :x E k m} . Then V A is a k-representation of 54. Given another m x m k-matrix B, VA is isomorphic to VB if and only if A and B are similar, i.e., there is an invertible k-matrix M with M AM- 1 = B. To see this, let f : VA -+ VB be an isomorphism. Since f preserves the first three subspaces, it follows that f = (h, h) with h a k-automorphism of k'": Moreover,
f«(1
+ A)k m) = {(h(x) , h(Ax»: x
E
k m} = {(y , By) :Y
E
k m}
= (1 + B)k m,
whence f(Ax) = h(Ax) = Bh(x) = Bf(x) for each x in k'" , Letting M be the matrix of h relative to the standard basis for k" gives an invertible matrix M with M A = B M . Conversely, the existence of such an M gives rise to a representation isomorphism f = (h, h) from VA to VB. Let Matm(k) denote the ring of all m x m k-matrices . An argument like that of the preceding paragraph shows that End V A is isomorphic to the ring C(A) = {M : M E Matm(k), M A = AM} , called the centralizer of the matrix A. The remainder of the argument is an exercise in linear algebra [Hungerford, 74] . Given an m x m k-matrix A , define a k[x]-module structure on VA = k" by xy = Ay for each Y E VA . Since k[x] is a principal ideal domain, VA is an indecomposable k[x]-module if and only if VA is isomorphic to k[x]/(g(x)'} for
8
1. Representations of Posetsovera Field
some irreducible polynomial g(x) E k[x] and integer e 2: 1. In this case, m is the degree of g(xY , and g(xY is the minimal polynomial of A. Now assume that VA is an indecomposable k[x [-module. As x acts on VA by A, C(A) is isomorphic to Endk1xj(k[x]/(g(xY)). But k[x]/(g(xY) is isomorphic to Endk[xl(k[x]/(g(xY)) via b 1-+ multiplication by b. It follows that End VA is isomorphic to k[x]/ (g(x Y). Hence , V A is an indecomposable representation by Lemma 1.1.1, since 0 and 1 are the only idempotents of k[x]/(g(xY). (a) In view of Proposition 1.1.4, it is sufficient to assume that n = 4. Since m 2: 1 is arbitrary, there are infinitely many indecomposable k-representations VA of 54 with A an m x m k-matrix. In particular, for each A E k and m 2: I, there is an m x m Jordan block matrix A(m, A) with minimal polynomial (x - A)m, where A(m, A) is the matrix with A'S on the diagonal, I's on the superdiagonal, and O's elsewhere. Notice that VA(m ,A) is an indecomposable k[x]-module and A(m, A) and A(m', A') are similar if and only if m = m' and A = A'. This shows that
(VA : A = A(m, A) , m 2: 1, A E k} is an infinite subset of Ind(4, k). (b) Suppose g(x) is an irreducible polynomial and V = k[x]/(g(xY) has dimension m as a k-vector space. Pick a basis for V and let A be the m x m k-matrix determined by multiplication by x relative to this basis. As above, End V A is isomorphic to k[x]/(g(xY) , and so VA is an indecomposable representation. Even though there are infinitely many elements in Ind(4, k) , they have all been classified; see Example 6.2.7. In fact, each V in Ind(4, k) has endomorphism ring isomorphic to either k or k[x)/(g(xY) for some irreducible g(x) in k[x) . The situation is quite different for n 2: 5. Given a field k, a ring R is a k-algebra if k = klR is a subring of CR = {x E R: x y = yx for all y E R}, a subring of R called the center of R. A k-algebra R is a k-vector space and is said to be ajinite-dimensional k-algebra if R has finite k-dimension. Example 1.1.7 If n 2: 5 and R is a finite-dimensional k-algebra, then there is a k-representation V of 5n with End V isomorphic to R. PROOF. Via Proposition 1.1.4, it is sufficent to assume that n = 5. Let A and B be two m x m k-matrices. Define the centralizer of A and B, C(A, B), to be {M: ME Matm(k), M A = AM, and M B = BM} , a subring of Matm(k). Define
V A.S
= (km EB v: k m EB 0, 0 EB r: (1 + l)k m , (l + A)k m , (l + B)k m ) ,
a k-representation of 55' Then End V A ,S is isomorphic to C(A, B). To see this, first observe that if f E End VA,S, then f = (g, g) for some endomorphism g of k", since f preserves the first three subspaces of k'" EB k" . Preservation of the last two subspaces yields g(Ax) = Ag(x) and g(Bx) = Bg(x) for each x E k" , Choose M to be the matrix of g relative to the standard basis for k" so that M A = AM and M B = BM .
1.1 Vector Spaces with Distinguished Subspaces
9
It now follows that the correspondence f H- M is an isomorphism from End VA,B to qA, B), noticing that if ME qA, B), then M induces a endomorphism of VA,B. Let R be a finite-dimensional k-algebra and {Xl, ••• , XI} a k-basis for R with XI = 1. Define (t + 2) x (t + 2) R-matrices
0 1 0 0 0 1
0 0 0 0
0 0 0 0 0 0
0 1 0 0
B=
A=
0 1 1 0 0 0
0 0 1 X2
0 0 0 1
0 0 0 0
0 0 0 0
0 0 0 0
0 0
0 0
XI
0 0 1 0
It suffices to prove that R is isomorphic to qA, B) . In this case, V A,B is a krepresentation of 55 with End V A,B isomorphic to R, as desired . Suppose M = (aij) E qA, B) is a (t + 2) x (t + 2) matrix with each aij a txt k-matrix representing a k-endomorphism of the t-dimensional vector space R. Since M A = AM, it follows that M is of the form
o
a3
at
az
a l+2 al+1
o o
0 0
0 0
a2
al
a2
al
Since M B = BM, it follows that M = a I for some a E End, R with ax, = x,a for each i. Because the Xi'S are a k-basis for R, ar = ra for each r E R. In particular, a is an element of EndR(R), the R-endomorphism ring of R regarded as a left R-module. This shows that qA, B) is contained in EndR(R) . Conversely, let f E EndR(R) and let a denote the txt k-matrix of f relative to the k-basis {XI," " XI} of R . Then M = aI E qA , B), and so qA, B) = EndR(R) . Finally, R is isomorphic to EndR(R), the isomorphism given by r H- f,L" where f,Lr is right multiplication by r . The proof is now complete.
EXERCISES 1. Two m x m k-matrices A and B are equivalent if there are invertible m x m k-matrices M and N with MAN = B. Define a k-representation WA of S3 by
(a) Prove that A and B are equivalent if and only if the representations WA and WBare isomorphic. (b) Use Example 1.1.5 to prove that a square k-matrix is equivalent to a diagonal matrix with entries either 0 or I.
10
I. Representations of Posets over a Field
2. Two pairs of m x m k-matrices (A , B ) and (A'. B' ) are simultaneously equivalent if there are invertible k-matrices P and Q with PA = A' Q and P B = B' Q. Define a k-representation WA,B of 54 by
(a) Prove that (A. B) and (A' . B ' ) are simultaneously equivalent if and only if W A,B and WN ,B' are isomorphic as representations. (b) All elements of Ind(54. k ) are listed in Example 6.2.8. What are the consequences of this classification for pairs of square k-matrices up to simultaneous equivalence?
3. Two pairs (A . B ) and (A'. B' ) of m x m k-m atrices are simultaneously similar if there is an invertiblem x m k-matrix P with P A p- 1 = A' and P B p- 1 = B'. Let UA ,B be the k-representation of 55 defined in the proof of Example 1.1.7. (a) Prove that UA,B and U N ,B' are isomorphic as representations if and only if (A. B) and (A'. B ') are simultaneously similar. (b) What are the consequences of Example 1.1.7, Example 1.4.1, and Proposition 1.4.2 for the problem of classifying pairs of matrices up to simultaneous similarity?
1.2 Representations of Posets and Matrix Problems A representation of a finite partially ordered set (poset) over a field can be written uniquely up to isomorphism and order as a finite direct sum of indecomposable representations (Theorem 1.2.2), In particular, classification of representations up to isomorphism reduc es to cla ssification of indecomp osa ble representations. With each poset , there is an ass ocia ted matrix problem as illustrated in Example I. 1.5 for 53. Interpretation of representations as matrices gives a procedure for computing endomorphism rings and constructing indecomposable representations (Theorem 1.2.6 and Example 1.2.7 ). A partially ordered set is a set S together with a binary relation-e on S that is reflexive, s :::: s for each s E S ; antisymmetric, if s :::: t and t :::: s , then s = t ; and transitive, if s :::: t and t :::: u , then s :::: u. An anticha in is a partially ordered set such that no two elements are related, For example, Sn is an antichain. Fundamental properties of representations of posets are conveniently expressed in a categorical setting, Following is a brief summary of standard categorical terminology that will be used henceforth. A category is a class X of objects together with a set of morphisms Hom(A, B) for each pair A, B of objects in X and a composition Hom(B , C) x Hom(A , B) H- Hom(A , C) , written (g, f) H- gf , for each triple A , B, C of objects of X such that (i) composition is associative and (ii) for each object A there is l A in Hom(A, A ) with flA = f and l Ag f E Hom(A , B ), g E Hom(C, A), and objects Band C ofX.
=
g for
The notation A E X stands for "A is an object of X," a mild abuse of notation, since X need not be a set. A category X is additive if, in add ition ,
1.2 Representations of Posets and Matrix Problems
11
(iii) for each A, B E X, there is a binary operation + such that (Hom(A, B) , +) is an abelian group with g(fl + h) = gil + gh and (fl + h)h = II h + hh for I i E Hom(A, B) , g E Hom(B, C), h E Hom(D, A), and C, D E X and (iv) given A I, ... , An E X there is an object A and morphisms ij E Hom(A j , A) such that if h E Hom(A i- B), then there is a unique I E Hom(A, B) with Ii j = h for each I .:s j .:s n . The object A of(iv) is called a direct sum of AI, . . . , An, denoted upto isomorphism by Al EB · . . EB An. The ij 's are called injections. If X is an additive category and A is an object of X, then it follows from (iii) that Hom(A , A) is a ring with identity lA, denoted by End A . As an example, the class of finite-dimensional vector spaces over a field k is an additive category; morphisms are k-linear transformations. In this case, A = A I EB . . . EB An is the usual vector space direct sum with injections i j : A j -+ A given by ij(aj) = (0, ... ,0, aj, 0, . . . ,0). Let k be a field and S a finite poset. Define rep(S, k) to be the category with objects U = (Uo, U, : i E S), where Uo is a finite-dimensional k-vector space, each U, is a subspace of Uo, and if i .:s j in S, then U, is contained in U] . Morphisms in this category, called representation morphisms, are k-linear transformations I : UO -+ U6 with I(Ui) a subset of U; for each i in S. If U is an objectofrep(S, k), then End U is a k-algebra, since endomorphisms of U commute with multiplication by elements of k . Objects of rep(Sn, k) are precisely the k-representations of Sn, as defined in Section 1.1. Let R be a ring with identity 1R. An element r of R is a unit of R if there is an s E R such that rs = I R = sr . The ring R is local if r + s is a nonunit for each pair rand s of nonunits of R.
Lemma 1.2.1 Suppose k is afield and S is a finite poset. (a) The category rep(S, k) is an additive category. Furthermore, if U E rep(S, k) and e is an idempotent in End U, then U = e(U) EB (1 - e)(U) in rep(S, k). (b) A representation U in rep(S, k) is indecomposable if and only if End U is a local ring. PROOF. (a) A routine verification shows that rep(S, k) is an additive category with
a direct sum of U
= (Uo, Ui :i E S) and V = (Vo, Vi : i E S) given by U EB V = (Uo EB Vo, U, EB Vi : i E S) .
This follows directly from the fact that the category of finite-dimensional vector spaces is an additive category. Injections for this direct sum are representation morphisms i u : U -+ U EB V defined by iu(uo) = Uo EB for each Uo E Uo and iv(vo) = EB Vo for each Vo E Vo . Moreover, if U is in rep(S, k) and e2 = e in End(U), then
°
U = e(U) EB (1- e)(U)
°
= (e(Uo), e(Ui): i E S) EB «(1- e)(Uo), (1- e)(Ui) : i E S)
12
1. Representations of Posetsover a Field
is a direct sum decomposition of V in rep( S, k). The proof is just as in Lemma 1.1.1 for the case that S = Sn' (b) A representation V is indecomposable if and only if and 1 are the only idempotents of End V, a consequence of (a). The ring End V is a finite-dimensional k-algebra, since Vo is a finite-dimensional k-space and End V is contained in End, Ue; Hence, End V is a local ring if and only if and 1 are the only idempotents of End V (Exercise 1.2.3). The proof of (b) is now complete.
°
°
An E rep(k, S) with each A j = (A jO, A js : s E S). Then = (EBjAjo ,EBjAjs :s E S) = A I EB " 'EBA n is a direct sum in rep(k, S) with injections i j : A j --+ V defined by ij(ajo) = (0, .. . , 0, ajO , 0, ... , 0) for each 1 :::: j :::: n. Define Pj : Vo --+ A jo by pj(alO , .. . , ano) = ajO for each j. Each P! is a representation morphism such that pji, = if Suppose
V
=
A I , . .. ,
(Vo,Vs :s
E S)
°
the identity on A j, ej = i j Pj is an idempotent of End V, and 1u = el +.. ·+en· The p/s are called projectionsfor the direct sum V = AI EB ·· · EB An. Conversely, if V, AI, ... , An E rep(k , S) with morphisms ij : A j --+ V and Pj :V --+ A j for each 1 :::: j :::: n such that pjit = if j =f. t, pji j is the identity of A j , ej = ijpj is an idempotent of End V, and I u = el + ... + en, then V is isomorphic to Al EB··· EB An (Lemma 1.2.I(a) and Exercise 1.2.6(b» .
j
=f. t, Pj i j
°
Theorem 1.2.2 Let S be afinite poset and k afield, (a) Each V E rep (S, k) is afinite direct sum of indecomposable representa-
tions. (b) A decomposition of V into indecomposable representations is unique up to
isomorphism and order ofsummands.
(a) The proof is by induction on dim V, the k-dimension of U«. If dim V = 1, then End V = k, and so V is indecomposable. Now assume that dim V > I and that V is not indecomposable. Write V = V EB W as a direct sum of nonzero representations. Then dim V < dim V and dim W < dim V. By induction, both V and Ware finite direct sums of indecomposable representations. Since dim V is finite, the proof is complete. (b) Suppose V = AI EB · . ' EB Am = B I EB· . ·EBBn, a direct sum of representations with each A j = (A jO, A js : s E S) and B, = (B iO' Bi, : s E S) indecomposable representations. Since V = A I EB ... EB Am, there are injections i j : A j --+ V and projections Pj : V --+ A j as defined above with Pj it = if j =f. t, Pj i j the identity of A l» ej = i j Pj an idempotent of End V, and I u = el +...+ em ' Similarly, there are injections ij : B, --+ V and projections pj : V --+ B, for each I :::: j :::: n with Pji; = if j =f. t , pjij the identity on Bj, ej = ij pj an idempotent of End V, and I u = e; + ...+ e~. Notice that the identity of AI is equal to PI(1u)il = PI(e; + ... + e~)il = EtPI(i;p;)i l = E,(P li;)(p;il). Since End Al is a local ring , (Pli;)(p;id = u is a unit of End At for some t . Assume, for notational convenience, that t = I . This involves a permutation of the subscripts of the B/s. Thenf = Pli; : B I --+ Al PROOF.
°
°
1.2 Representations of Posets and Matrix Problems
13
and g = piilu-I : Al -+ BI with fg = uu:" = 1. Now, e = gf E End BI with eZ = fgfg = fg = e. By Lemma l.2.l(a), B I = e(A I) E& (1 - e)(Ar). As BI is indecomposable, B I = e(A I) and eg : Al -+ B I is an isomorphism. It follows that V = Al E& Bz E& . .. E& Bn , where the injection for Al is iIU- I, the projection for A 1 is f p~ = (PI i;) pi = PI (ii pi) = PI e~ , the injection for each B, is ij, and the projection for each B, is pj. Since V = Al E& Az E& .•. E& Am, A z E& . . . E& Am is isomorphic to Bz E& . . . E& Bn . The proof is completed by an induction on m. The point of Theorem 1.2.2 is that all representations in rep(S, k) can be determined uniquely up to isomorphism from the indecomposable ones. Write Ind(S, k) for the set of isomorphism classes of indecomposables in rep(S, k). If S = Sn, then Ind(S, k) = Ind(n , k), as defined in Section 1.1. The next theorem shows that classification of nontrivial elements of Ind(Sn, k) can be interpreted as a matrix problem of the kind discussed in Section 1.1. Define /),.(Sn, k) to be the full subcategory of rep(Sn+l, k) with objects of the form
V = (Vo = E&iVi, Vi, V* : i
E
Sn)
such that V i n V* = 0 for each i E Sn' The term full subcategory means that objects of /),.(Sn, k) are objects of rep(Sn+l, k) and morphism sets of objects in /),.(Sn , k) are exactly as in rep(Sn+l, k) .
Theorem 1.2.3 Let k be afield. (a) There is a correspondence H : /),.(Sn, k) -+ rep(Sn, k) with Hom(V , V') isomorphic to Hom(H(V) , H(V'»for each V, V' in /),.(Sn, k). The image of H consists ofall representations of Sn with no trivial summands. (b) The correspondence H induces an injection Ind(/),,(Sn, k» -+ Ind(Sn, k) with nontrivial indecomposable representations as the image. PROOF.
(a) Define H: /),.(Sn , k) -+ rep(Sn , k) by
H(V)
= (Vo, Vi:i
E
S) withVo = Vo/V* andVi
= (Vi + V*)/V*,
where V = (Vo = E&iVi , Vi , V*: i E Sn) E /),.(Sn, k). If f: V -+ V' is a morphism in /),.(Sn, k), then f induces a representation morphism HU): H(V) -+ H(V'), since f(V*) ~ V; and f(V i) ~ V[ for each i in The correspondence H induces a group homomorphism H : Hom(V , V ') -+ Hom(H(V) , H(V'» . This homomorphism is one-to-one, since if HU) = 0, then f(Vo) ~ V; and f(V i) ~ V[ n V; = 0 for each i. ToconfirmthatH : Hom(V, V') -+ Hom(H(V), H(V'» is onto, letg : H(V) -+ H(V'). Then g(Vi) ~ V;' with Vi = (Vi + V*)/ V* isomorphic to Vi, since Vi n V* = O. In particular, g induces a k-linear transformation f : E&i Vi -+ E&i V[ with f(V i) ~ V [ and n' f = git , where tt : E& Vi -+ Vo is the k-linear transformation with ker n = {x E E&Vi I rr(x) = O} = V*. Therefore, f: V* = ker zr -+
s..
14
v; =
I. Representations of Posets over a Field
kerzr', from which it follows that I :V -+ V ' is a representation morphism
with H(f) = g. To find the image of H, suppose V = (Vo, Vi: i E S) E rep(Sn, k) with no trivial summands. Then Vo = ~i Vi, and so inclusion of each Vi in Vo induces an epimorphism 1T: EBi Vi -+ Vo with ker 1T = V* and Vi n V* = 0 for each i , Consequently, (EBi Vi, Vi , V*) E !:J.(Sn , k) with H (EBi Vi, Vi, V*) = V, as desired. (b) By (a), H : Hom(V, V) -+ Hom(H(V) , H(V» is a group isomorphism for each V and V in !:J.(Sn , k). In particular, V is isomorphic to V if and only if H(V) is isomorphicto H(V). Notice that H(fg) = H(f)H(g) for representation morphisms I and g. Thus, e is an idempotent of End V if and only if H(e) is an idempotent of End H(V), and if I is an idempotentof End H(V) , then there is an indempotent e in End V with H (e) = I .In viewof Lemma I, V is indecomposable if and only if 0 and I are the only idempotents of End V. Since H(O) = 0 and H (I u) = I H(U), the correspondence sending the isomorphism class of V to the isomorphism class of H(V) induces an injection Ind(!:J.(Sn , k» -+ Ind(Sn, k) . Finally, apply (a) to see that the image consists of the isomorphism classes of nontrivial indecomposables in rep(Sn, k) . In Example 1.1.5, classificationof indecomposable representationsin rep(S3, k) is reduced to classificationof indecomposable representations in !:J.(S2, k) . These indecomposables were then described by solving a matrix problem. As an alternative proof, the correspondence H of Theorem 1.2.3 can be used to translate indecomposables in !:J.(S2, k) into indecomposables in rep(S2, k) , which are then easily classified as in Example 1.1.3. Corollary 1.2.4 IIV = (VI EB V2, VI, V2, V3) E rep(S3 , k) is indecomposable with each Vi =j:. 0 and U, n V3 = V2 n V3 = 0, then V is isomorphic to (k EB k, k EB 0, 0 EB k, (1 + I)k).
PROOF. Notice that V E !:J.(S2, k). By Theorem 1.2.3, H(V) = (Vo, VI, V2) E rep (S2, k) is indecomposable with Vi isomorphic to Vi =j:. 0 for i 1,2 and Vo = (VIEBV2)/ V3. As a consequenceof Example 1.1.3, H(V) is isomorphicto (k , k , k). However, H(k EB k , k EB 0, 0 EB k, (I + l)k) =
=
((k EB k)/(1
+ l)k, (k EB 0)/(1 + I)k , (0 EB k)/(I + I)k)
is isomorphic to (k , k, k). Apply Theorem 1.2.3 to conclude that V must be isomorphic to (k EB k ; k EB 0, 0 EB k , (1 + I)k). In view of Theorem 1.2.3, indecomposable representations in rep(Sn, k) are determined by indecomposable representations in !:J.(Sn, k). Each representation V in !:J.( Sn, k) can be interpretedas a partitionedmatrix M U • Let V = (Vo, Vi, V * : i E Sn) E !:J.(Sn, k), recalling that Vo = EBiVi and Vi n V* = 0 for each i , Pick a basis B, for each Vi and a basis B of V*. Write each element of B as a k-linear combination of elements of the Bi'«, and use the resulting coefficientsto represent
1.2 Representations of Posets and Matrix Problems
15
V* as the row space of a k-matrix Mu
= (.. ·IAiI ···)
with one block matrix Ai for each i E Sn. The columns of each Ai are labeled by Bi , and the rows of Mu are labeled by B, with the convention that an Ai-block is empty if Vi = O. The condition that Vi n V * = 0 for each i means that each row of M u must have nonzero entries in at least two of the blocks . Moreover, if V has no l-dimensional summands, then each column of M u must be nonzero . A detailed example of the construction and properties of M u is given in the proof of Example 1.1.5, wherein
n
= 3.
The representation V remains unchanged by the following invertible matrix operations on Mu: (a) Elementary column operations within each block Ai for each i E S; (a basis change of Vi ), (b) Elementary row operations on Mu (a basis change of V*). Representations V and V in ti(Sn, k) are isomorphic if M u can be reduced to M» by a series of matrix operations (a) and (b). This is because V and V are isomorphic ifand only if there is an isomorphism f : V o -+ Vo with f(V i) = Vi for each i. Hence, classification of indecomposables in ti(Sn , k) up to isomorphism amounts to the solution of a matrix problem, i.e., find canonical forms for the matrices M u subject to the matrix operations (a) and (b). The following example demonstrates how to interpret the representation V A of S4 given in Example 1.1.6 as a matrix . A square matrix A is indecomposable if the vector space VA, as defined in Example 1.1.6, is an indecomposable k[x]-module.
Example 1.2.5 Let A be an indecomposable n x n k-matrix. If V = (Vo = k" EEl k" EEl k" EEl k" , VI, V 2, V 3, V4, V*) E ti(S4 , k), where Vi is the ith coordinate space of Vofor 1 ::: i ::: 4 and V* is the row space of the matrix
then H(V) is isomorphicto (k" EEl k", k" EEl 0,0 EEl k" , (1 + I)k n , (l + A)k n ) = VA' Vo = VI EEl . . . EEl V4 and that Vi n V* = 0 for each i, since no linear combination of the rows of M u will result in a vector with a nonzero entry in exactly one block. The dimension of V o is 4n, the dimension of each Vi is n, and the dimension of V* is 2n. With H as defined in Theorem 1.2.3, H(V) = (Vo, VI, V2 , V3 , V4), where Vo = VI EEl V2 = «VI + V*)/V*) EEl «V2 + V*)/ V*) = k" EEl k", The first row of M u gives V3 = (U3 + V*)/ V* = (l + I)k n , and the second row yields V4 = (U4 + V*)/ V* = (l + A)k n .
PROOF. Notice that
16
I. Representations of Posets over a Field
There is a generalization of the notion of a matrix problem for rep(Sn, k) to an arbitrary finite poset S, but the construction is more complicated than it is for antichains. Let U = tU«; U, : i E S) E rep(S, k) and assume that U has no trivial summands. For each i in S, U, = ut EI1 E{Uj : j < i} for some subspace ut of U, with ut isomorphic to U;/ E {Uj : j < i}. In general , the subspace ut is unique only up to isomorphism. Since U has no trivial summands, Uo = E{Ui : i E S}, whence Uo = E{Ut: i E S} and U, = E{Ut: j :s i} for each i in S. Let U* denote the kernel of the k-linear transformation EI1{ U;* : i E S} ---+ Uo induced by inclusion of each U;* in Ui: Then U;* n U* = 0 for each i. To find a matrix associated with U, pick a basis B, for each U;* and a basis B of U* . Write each element of B as a k-linear combination of elements of the Bi's and use the resulting coefficients to represent U* as the row space of a matrix
Mu
= (... lAd· ..)
with one block matrix Ai for each i E S. The columns of Ai are labeled by B i , the rows of M u are labeled by B, and the block Ai is empty if U] = o. If S = Sn , then U;* = U, for each i so that this construction agrees with the previous construction for antichains. Theorem 1.2.3 can be generalized to arbitrary finite posets (Exercise 1.2.7). As part of this generalization, the next theorem shows how the endomorphism ring of a representation U can be computed from the ring of endomorphisms Ru of the partitioned matrix M u- This, together with Lemma 1.2.1, gives a computational procedure to determine whether or not U is an indecomposable representation. Let rs( M u ) denote the row space of the matrix M tj , Theorem 1.2.6 Let S be a finite poset, k afield, and U = (Uo, U, : i E S) E rep(S, k) with no trivial summands. For each i E S, choose a summand U;* of U, with U;* isomorphic to U;/{EU j : j < i}. Then End U is isomorphic to the ring R u consisting of all f = EI1{f; : i E S} E n {End U;* : i E S} with
f(rs(M u)) £ rs(Mu)·
Recall that U* is the kernel of the onto linear transformation EI1{ U;* : i E S} ---+ Uo with U, = E{ Uj : j :s i in S} for each i in Sand U* = rs(Mu) . Each g in End U induces ¢(g) = EI1{gi : i E S} with gi : U;/(E {Uj : j < i}) ---+ U;/(E{U j: j < i}) defined by gi(Ui + Ej
Indecomposable matrices can be used to construct representations V that can be shown to be indecomposable by computing their endomorphism rings via Ri],
1.2 Representations of Posets and Matrix Problems
17
Example 1.2.7 Let S be the poset (l < 2,3 < 4,5 < 6). Given an n x n k-matrix A, define (Uo, U I , Uz, U3, U4, Us, U6) E rep(S, k) by Uo = k n $k n $k n , UI = 0$ U3 = (l Us
°
n
$ k £ Uz = (l
+ l)k n $
°£
= 0$ (l + l)k n £
Then
Mu = (
+ A)k n $ r ,
= kn $ n U6 = k $
U4
n
k $ 0, and (l
+
n
l)k •
° °°/ /°/) °/ , 0 /
A
/00/
/0
End U is isomorphic to C(A), and U is indecomposable if A is an indecomposable matrix. direct calculation shows that candidates for the (ut)'s with U, = ut $ = O$O$kn, Ui = (I + A)kn $0, U; = (1 + l)k n $0, = = 0$ (l + 1)kn , and U;, = k" $ $ 0. Recall that the row space of Mu is the kernel U* of ${ut : i E S} -+ Uo. Notice that U;" U;, and ut are the respective coordinate spaces of k" $ k" $ k", The first row of M u arises from Ui = (l + A)kn $ 0, the second row from U; = (l + 1)kn $ 0, and the third row from U; = 0$ (l + l)k n . By Theorem 1.2.6, End U is isomorphic to Rij, To compute Rij, let I = II EB ... $ 16 with l(rs(M u)) contained in rs(M u). View each Ii as an n x n k-matrix. Then h = 14 = 16 (from the second row of M u ), II = 14 = Is (from the third row of M u ), and [z = 16 with 14A = AI4 (from the first row of Mu). Thus, I = (h, . .. , h) with h E C(A), i.e., hA = Ah. Conversely, each h in C(A) determines an element 1= (h , . . . , h) E Ri], This shows that Ru may be identified with C(A), whence End U is isomorphic to C(A) by Theorem 1.2.6. Finally, if A is an indecomposablematrix, then C(A) has only and 1 as idempotents, recalling from Section 1.1 that C(A) is isomorphic to k[x]/(g(xn with g(x) an irreduciblepolynomialand g(x Y the minimalpolynomialof A. This shows that U is indecomposable.
PROOF. A
:E{Uj U;
u;
° :
j < i} are $ k" $ 0, U;
°
°
The matrix Mu is defined in terms of a choice of the U;*'s. The following invertible matrix operations on Mu = (... IAi I...) preserve U, recalling that the columns of Ai are labeled by a basis of ut and the rows of M u by a basis of U*. (a) Elementary column operations within each block Ai (a basis change of the subspace (b) Elementary row operations on M u (a basis change of the subspace U*), (c) Elementarycolumn operations from block A i to block A j if i < j in S (a different choice of an embedding of the subspace ut into Uj ) .
Un,
18
1. Representations of Posets over a Field
Two representations V and V are isomorphic if M u can be reduced to M v by a series of matrix operations (a), (b), and (c). Classification of indecomposables in rep(S, k) amounts to finding a canonical form for matrices Mu subject to the matrix operations (a), (b), and (c). This procedure is practical only for posets that are relatively uncomplicated.
Example 1.2.8 Suppose S Ind(S, k) are (k, 0, .. . , 0),
= {I
(k,O,
< 2 < . . . < n} is a poset. The elements of
,0, k),
(k,O,k ,
,k),
(k, 0, .. . , 0, k, k), .. . ,
(k,k, .. . ,k ,k).
In each case, the endomorphism ring is k. PROOF. Let V = (Vo, V t S; . . . S; Vn) E rep(S, k) be an indecomposable representation and write M u = (At I· .. IAn)· First assume that V t is nonzero. Use elementary row and column operations (a) and (b) to reduce Al to a matrix of the form (~ ~ . Next use (c) to see that
° °°
I I ... I 0) determines a representation sumfor some k-matrices e.. But (l mand V = (VO , Vi) of V with =1= Vt , and V;* = for each i ::: 2. Since S is linearly ordered with 1 as the least element, vt = VI = V2 = '" = Vn . But V is indecomposable and VI =1= 0, so that V must be isomorphic to (k, k , .. . , k). Now assume that VI = 0. Then Al is empty, and the proof is completed by an induction on n.
°
EXERCISES
1. Solve a matrixproblemto find all elementsoflnd(S, k), where S is the poset {I, 2 < 3}. 2. Let A and B be m x m k-matrices. (a) Find a representation W isomorphic to the representation given in Example 1.2.7 with Mw =
I
1 A)
0 0 0 0 1 0 0 1 1
(o
,
0 1 101
= (km EEl k'" , km EEl 0,0 EEl k'" , (l + l)k m , (l + A)k m , (1 + B)k m ) rep(Ss, k) . Show that Mu = (~ I~ IgI~ I~) and End U = C(A, B).
(b) Define U
E
1.3 Finite Representation Type
19
I I IbI
(c) Define a matrix M = (~ ~ ~ ~) . Find a V E rep(Ss, k) with Mv = M and show that End V = C(A , B) . (d) Expla in why the representations V and V are not isomorphic . 3. Let R be a finite-dimen sional k-algebra, k a field. Prove that R is a local ring if and only if 0 and 1 are the only idempotents of R. 4. Letk be a field of characteristic 0 and define V = (Vo, VI, V2, V3) E rep(S3, k), where Vo =k 4 , VI =k EElk EEl 0 EEl 0, V2 = oEEl oEEl k EElk, and V3 = (1,1 ,1, I)k EEl (2,3,3, 3)k. Find Mu and solve a matrix problem to write Mu as a direct sum of indecomposable representations. Describe these indecomposable representations. What changes if V3 is replaced by (1, 1, 1, I)k EEl (2,3,3, 3)k EEl (l , 0, 1, O)k? 5. Let S be a finite poset. Define a poset T = S u {*) with s < * for each s in S. Prove that there is a one-to-one correspondence between nontrivial indecomposables in Ind(T , k) and nontrivial indecomposables in Ind(S , k) . 6. Let X be an additive category. Idempotents split in X if whenever e E End A is an idempotent for some object A of X, then there is an object B in X, q E Hom(B , A), and p E Hom(A , B) with qp = e and pq = l s . (a) Prove that idempotents split in rep(S, k) . (b) Assume that idempotents split in X and that V, A I, ... , An are objects of X. Prove that V = A I EEl . . . EEl An if and only if there are morphisms i j : A j -)0 V and p j : V -)0 A j for each 1 .::: j .::: n such that p j i, = 0 if j =1= t, Pjij is the identity of A j, ej = i j p j is an idempotent element of End V, and 1u = el + ...+ en. (c) ([Bass 68] or see [Arnold 82]) Suppose X is an additive category such that idempotents split and A = A I EEl . . . EEl An is a direct sum of objects Ai of X with each End Ai a local ring. Generalize the proof of Theorem 1.2.2 to show that if A = BI EEl .. . EEl B, with each B, an indecomposable object of X, then t = n and there is a permutation a of {I, 2, . . . , n} such that Ai is isomorphic to Bo(i)for each i . 7. Given a finite poset S, define A(S , k) and a correspondence H: A(S, k) -)0 rep(S , k) generalizing Theorem 1.2.3 such that RH(U ) = End V for each V E A(S , k).
1.3 Finite Representation Type Finite posets having only finitely many isomorphism classes of indecomposable k-representations can be described (Theorem 1.3.6) . This description depends only on the poset and is independent of the field k. In this case there is a systematic procedure for finding all indecomposable k-representations up to isomorphism (Theorem 1.3.7 and references). A finite poset S has finite representation type over a field k if lnd( S, k) is finite and has infinite representation type if lnd(S, k) is infinite. For example, Sn has finite representation type for n ::s 3 (Example 1.1.5) and infinite representation type for n 2: 4 (Example 1.1.6(a)). A subposet of a poset S is a subset T of S equipped with the ordering on S; specifically, if i, JET, then i ::: j in S if and only if i ::: j in T . Part (a) of
20
1. Representations of Posets over a Field
the following proposition extends the correspondence F of Proposition 1.1.4 to arbitrary finite posets S; part (b) is an alternative correspondence. Proposition 1.3.1 Suppose T is a subposet ofa finite poset S. (a) There is a correspondence F+: rep(T, k) --+ rep(S, k) with Hom(V, V) isomorphic to Hom(F+V, F+V)for each V, V in rep(T, k). (b) There is a correspondence F- :rep(T, k) --+ rep(S, k) with Hom(V, V) isomorphic to Hom(F-V, F-V)for each V, V in rep(T , k). (c) If S has finite representation type, then so does T . (d) If T has infinite representation type, then so does S. PROOF. (a) The correspondence is given by F+(V) = W = (Wo, Wi: i E S) E rep(S, k) for V = (Vo, Vi: i E T), with Wo = U» , Wi = Vi if i E T, Wi = :E{V j: JET, j < i} if i E S\T and there is JET with j < i, and Wi = 0 otherwise. If f: V --+ V is a morphism in rep(T, k), define F+(f) = f, a morphism in rep(S, k). Then F+: Hom(V, V) --+ Hom(F+(V), F+(V)) is an isomorphism for each pair V, V of objects of rep(T, k). (b) The correspondence is given by F-(V) = W = (Wo, Wi: i E S) E rep(S, k) for V = (Vo, Vi: i E T) with Wo = Ue; Wi = Vi if i E T, Wi = n{Vj :j E T, j > i} ifi E S\T and there is JET with j > i, and Vi = Vo otherwise. If f: V --+ V is a morphism in rep(T, k), define F-(f) = f, a morphism in rep(S, k). Then F- : Hom(V, V) --+ Hom(F-(V), F-(V)) is an isomorphism for each pair V, V of objects of rep(T , k). (c), (d) Just as in Proposition 1.1.4, both F+ and F- induce an injection Ind(T, k) ~ Ind(S, k). The width of a poset S, denoted by w(S), is the largest number of pairwise incomparable elements of S. Corollary 1.3.2 If Sis a finite poset with w(S) ::: 4, then rep(S, k) has infinite representation type. PROOF. Apply Proposition 1.3.1 and the fact that rep(S4, k) has infinite representation type (Example 1.1.6(a)). A finite poset S is a chain if S is linearly ordered. Lemma 1.3.3 Suppose S is a finite poset with w(S)
= 1.
(a) Then S is a chain and rep(S, k) has finite representation type. Elements of Ind(S, k) are (k, 0, ... , 0),
(k, 0,
(k,O,k,
, 0, k), , k),
(k, 0, ... , 0, k, k), .. . ,
(k,k, ... ,k ,k).
1.3 Finite Representation Type
21
If U = (Uo, Ui : i
E S) and V = (VO, Vi :i E S) E rep(S , k) with Uo £ Vo and U, = UonV;joreachi E S,thenV = WE9UforsomeW E rep(S,k).
(b)
PROOF. Since w(S) = 1, any two elements of S are comparable. Thus, S is linearly ordered, say S = {I < 2 < ... < m}. (a) This is proved in Example 1.2.8 using a matrix argument. Following is an alternative proof. Let U = (Uo, UI £ ... £ Um) be an indecomposable representation of S and write Uo = UmE9 Vo for some Vo . There is a representation direct sum
°
U is indecomposable, either Um = and U is isomorphic to (k,O, . .. ,O), or Vo=O and U=(Um,UI "",Um _I,Um)' Next write Um= Since Um -
I
E9 Vm so that
°
Since U is indecomposable, either Um-I
=
°
and U is isomorphic to (k, 0, . . . ,
0, k), or else Vm = and U = (Um-I, UI , .. . , Um-2 , Um-l, Um-I). Repeating this
process completes the list of indecomposables as given. (b) Let X = (Vol Uo , (VI + Uo)1 U» . ... , (Vm + Uo)1 Uo) E rep(S , k). First assume that X is indecomposable. Then X is isomorphic to one of the representations listed in (a). There is some t and x E VI\Uo such that VolUo = k(x + Uo), (Vi + Uo)/Uo = for i < t, and (Vi + Uo)1 Uo = k(x + Uo) for i ::: i . Hence , Vo = kx E9 Uo with kx £ Vi for i ::: t . Then Vi = Vi n Vo = kx E9 (Uo n Vj ) = kx E9 U, for each i :::t and Vi = Uo n Vj = U, for each i < t. This shows that
°
V
= (Vo, VI ,"
" Vm) = (kx, 0, ... ,0, kx, . . . , kx) E9 (Uo , UI, . . . , Um)
= WE9U is a direct sum of representations, as desired . For the general case, write X as a direct sum of n indecomposable representations in rep(S , k). Use the above argument on an indecomposable summand of X and induction on n to see that U is a summand of V. Given U = (Uo, U, : i E S) E rep(S, k), define the dimension of U to be the k-dimension of Uo.
Lemma 1.3.4 If S is a finite poset with w(S) = 2, then rep(S, k) has finite representation type. Moreover, each indecomposable U E rep(S, k) has dimension 1. PROOF. Let a be a minimal element of S and partition S into three sets A
{a} , B
= {i E
S:i > a}, and C
= {i
E
S :i
t
=
a}. Observe that C is a chain,
22
I. Representations of Posets over a Field
since w(5) = 2 and the minimality of a in 5 implies that any two elements of C are comparable. The proof is by induction on the cardinality 15 1of 5. If 151 = 2, then 5 = 52. By Example 1.1.3, 52 has finite representation type and each indecomposable representation has dimension 1. Now assume that 151 > 2, rep(T , k) has finite representation type, and each indecomposable U E rep(T , k) has dimension 1 for each proper subposet T of 5. Let U = (Uo, U, : i E 5) be an indecomposable representation of 5. If U; = 0, then T = B U C is a proper subposet of 5, and U is in the image of F+: rep(T, k) -+ rep(5, k) given in Proposition l.3.I(b). By the induction hypothesis, it follows that there are, up to isomorphism, only finitely many U E Ind(5, k) with U« = O. Next assume that U = (Uo , U, : i E 5) is indecomposable with Ua ::j:. O. Then W = (Ua , Ua n Uj : j E C) and Uc = (Uo , Uj : j E C) are in rep(C, k). Since C is a chain and Ua ::s Uo , it follows from Lemma 1.3.3(b) that Uc = W EB V for some V = (Vo , Vj : j E C) . Specifically, write C = {I < 2 < . . . < m} so that
Uo =
u, EB Yo,
o, = u), n U j ) EB Vj for 1 s j s m, VI S; ... S; Vm S; Yo· The next step is to verify that U
= X EB Y, where X = (Xo, Xi : i E 5) is defined
by
Xo
= Us, Xi = Ua for i ::: a,
and Y = (Yo , Yi : i
Yo
E
and Xi
5) is defined by
= Vo, Ya = 0, Yi = Vo n U, for i > a,
Notice that X, Y
E
= Ua n U, for i E C and Yi
= Vi for i E C.
rep(S, k). From the above decomposition of Uc,
Uo = U; EB Vo = Xo EB Yo , U, = (Ua n Ui) EB Vi = Xi EB Yi for i
u, = U; EB 0 = x, EB Ya.
E C,
Finally, assume that i ::: a. Then
Ui
o,
= Ua EB Vo n u, = Xi EB Yi ,
u,
since S; o, S; Uo and Uo = EB Yo. Since U is indecomposable and U« ::j:. 0, it must be the case that U = X. Therefore, U is in the image of F- : rep(C, k) -+ rep(S, k) given in Proposition 1.3.1(b). As a consequence of Lemma 1.3.3(a), there are, up to isomorphism, only finitely many indecomposable U's with Ua ::j:. 0, and each such U has dimension 1. This completes the proof.
1.3 FiniteRepresentation Type
23
There is a more general version of Lemma 1.3.4. A splitting decomposition of a finite poset 5 is a partition of 5 into three subsets A, B, C such that C is either a chain or empty and a < b for each a E A, b E B. For example, if A and B are finite posets and 5 is the disjoint union of A and B subject to the additional relations a < b for each a in A and b in B, then the partition A U B is a splitting decomposition of 5 with C empty.
Proposition 1.3.5 Ifa finite poset 5 has a splitting decomposition A, B , C, then each U in Ind(5, k) is in the image of either Ind(A U C, k) or Ind(B U C, k). The proof is essentially the same as the proof of Lemma 1.3.4 and is left as an exercise (Exercise 1.3.2).
PROOF.
The only remaining case for determination of finite representation type is for posets of width 3. This case is significantly more difficult. There is a standard notation for posets. A chain with n elements is denoted by (n). If T) , .. . , Tm are posets, then the disjoint union of the T; 's is written as (T) , . .. , Tm ) . As an example, (1, 1, 1, 1, 1) is an alternative notation for the antichain 55, and (2,2,2) denotes the poset {1 < 2,3 < 4,5 < 6}. Define N to be the poset{I < 2 > 3 < 4}. The following theorem shows that finite representation type for rep(5, k) does not depend on the field k.
Theorem 1.3.6 [Kleiner 75A] Let 5 be a finite poset. Then 5 does not contain 54, (2, 2, 2), (1,3 ,3), (1,2,5), or (N, 4) as a subposet has finite representation type.
if and only if rep(5, k)
In view of Proposition 1.3.1, it suffices to show that if 5 is one of 54, (2, 2, 2), (1,3 ,3), (1,2,5), or (N, 4), then 5 has infinite representation type. These five posets are called critical posets. Following is a list of representations for each critical poset 5, demonstrating that rep( 5, k) has infinite representation type. The idea is to show that if A is an indecomposable n x n k-matrix, then there is a representation U E rep(S', k) with dim U ~ n and End U = C(A) , the centralizer of A . Since Jordan block matrices provide examples of n x n indecomposable matrices, ranks of indecomposable representations in rep(5, k) are unbounded. This shows that rep(5, k) has infinite representation type. The only difficulty is, given the list, to show that each of these representations has endomorphism ring C(A). The case that 5 = 54 is proved in Example 1.1.6 by computing End U directly. The case that 5 = (2, 2, 2) is Example 1.2.7, confirmed by computing Ru from the matrix Mu- The remainder of the computations are given as exercises. PROOF. (~)
24
I. Representations of Posets over a Field
(i) 5
= 54
V = (Vo, VI , V z , V3, V 4) Vo = k" EB k", Vz=OEBk n,
VI=knEBO,
V3 = (l + l)k n,
V 4=(l+A )kn.
246
I I I
= (2, 2, 2) = 1 3 5 V = (Vo, VI ~ o; V3 ~ V 4, u, ~ V6), Vo = k n EB k n EB k" , n VI = EB EB k" , U: = (l + A)k EB k" , V3 = (l + l)k n EB 0, V 4 = k" EB k" EB 0,
(ii) 5
Vs =
°° °
n EB (l + l)k ,
n V6 = k" EB (l + l)k .
4
(iii)
7
I I 136 I I S = (l, 3, 3) = 2 5 u=
(Vo, VI, V2 S; V3
S;
V4, u, S; V6 S; V7),
Vo = k" E9 k" E9 k" E9 k" , VI = (l V2 =
+ 1)k n E9 (l + I )k n ,
°E9
(l
°
n V3 = E9 k" ffi k E9 0,
+ l)e E9 0,
°
V4 = ffi k" ffi k" E9 k" ,
8
I
7 1
3
~
(iv) 5=(N ,4)= 2
4
5
NI
V = (Vo, Vo
V 4 ~ V3, o, ~ V6
~
V7
~
Vs),
= k" EB k" EB k" E9 k" E9 k",
Uz = V4
o, ~ VI 2
(1 + l)k n EB (l
+ l)k n EB 0,
= °E9 (l + l)k n EB °E9 0,
Vs =
°EB °E9 °EB °EB k" ,
V 7 = k" EB
°° EB
EB k" EB k" ,
VI = k n EB k n EB k" EB k" EB 0,
°
= EB k" EB k" E9 (l + l)k n , V6 = k" EB °EB EB °EB k", Ue = k" EB (l + A )kn EB k" EB k", V3
°
1.3 FiniteRepresentation Type
25
8
I I 3 6 I I 2 5 I 7
= (1,2,5) = 4 v = (Vo, VI , o, ~ V3 , V4 ~ Vs ~ V6 ~ V7 ~ Vg), Vo = kn E9 kn E9 kn E9 kn E9 kn E9 k" , n n VI = (0 + 1 + 1 + 0 + 0 + O)k + (0 + 0 + 1 + 0 + 0 + l)k
(v) S
+ (0 + 0 + 0 + 1 + 1 + O)k n , n n Uz = (1 + l)k E9 (1 + l)k E9 0 E9 0, V4 = 0 E9 0 E9 0 E9 0 E9 0 E9 k", V6
= k" E9 0 E9 0 E9 0 E9 k" E9 k",
Vg = k" E9 (l
V3 = k" E9 k" E9 k" E9 k" E9 0 E9 0,
= 0 E9 0 E9 0 E9 0 E9 k" E9 k" , V 7 = k" E9 0 E9 0 E9 k" E9 k" E9 k" , Vs
+ A)k n E9 k" E9 k" E9 k" ,
(=» Suppose S has infinite representation type . Then w (S ) ::: 3 by Lemmas 1.3.3 and 1.3.4. Moreover, any poset S containing S4 as a subposet has infinite representation type by Corollary 1.3.2. It now suffices to assume that w(S) = 3 and prove that S must contain (2,2,2), (1,3,3), (1,2 ,5), or (N,4) as a subposet. There are several proofs of this result. All are lengthy and complicated and as such are beyond the scope of this book. A proof in [Ringel 84, Section 2.6] uses the theory of integral quadratic forms. A more traditional proof is given, for example, in [Simson 92 , Theorem 10.1]. This proof uses the notion of a Zavadiskij derivative of a poset , as defined below. An elementary outline of this proof is given in [Hahn 97] .
In view of Proposition 1.3.1, elements of Ind(S, k) include those elements of Ind(T, k) for each proper subposet T of S. Proposition 1.3.5 illustrates that for certain finite posets S, all elements ofInd(S, k) are ofthis form. These observations motivate the notion of sincere representations and posets. Given V = (Vo, Vi : i E S) E rep(S, k), define cdn V = (uo, u, : i E S), where Uo is the k-dimension of U«; u, is the k-dimension of Vi / Vr, and Vr is the subspace of Vi generated by {V j : j < i in S}. Call V a sincere representation if each U i # O. If V is not sincere, say u, = 0 for some minimal i, then T = S\{i} is a proper subposet of S with V in the image of F+: rep(T , k) ---+ rep(S , k) , as given in Proposition 1.3.1. Moreover, ifrep(S, k) has finite representation type , then so does rep(T, k). A finite poset S is sincere if there is a sincere indecomposable representation V in rep(S , k ).
26
1. Representations of Posets overa Field
Theorem 1.3.7 [Kleiner 75B] A finite poset S with finite representation type is sincere if and only if S is one ofthe posets (1), (1, 1), (1, 1, 1), (1, 1,2), (1,2,2), (1,2,3), (1,2,4), (N, 2), (N, 3),
1\ 1
3
6
3
2
5
1
4
1;\1
I
I
PROOF. [Simson
7,
8
I I
1\/5
7
1
6,
4
92, Theorem 10.2] or [Ringel 84, Section 2.6].
A complete list of all of the indecomposable sincere representations for each sincere poset, a total of 41 representations, is given in [Kleiner 75B] . There are several minor typographical errors in this list, repeated in [Arnold 89] for Butler groups. A corrected list can be found in [Ringel 84, Section 2.6], [Simson 92, Section 10.7], or [Arnold, Richman 92]. In the latter paper, an algorithm was implemented confirming that each of these representations is in fact indecomposable. Here is an example of the representations in this list: Example 1.3.8 (a) If S = (1,1,2), then the only sincere element ofInd(S, k) is
(k EB k, k EB 0, (1 + l)k, 0 EB k , k EB k) . (b)
If S = (1, 2, 2), then the sincere elements ofInd(S, k) are (k EBk EDk , (l
° °ED ED ED ° EDO,k ED kED 0, ° EDO °EDk ED kED 0, ED ° ED kED
+ 1 + l)k, k ED OEDO, k ED k ED 0, OED
EDk,
(k ED k EDk , (l +0+ l)k + (l + 1 +O)k ,k (k ED k, (l
+ l)k,
k
EDk,
k
k,
k,
k),
k),
k).
The following corollary is striking. It can be proved by examining the list of 41 sincere indecomposable representations, together with the observation that if U is indecomposable in rep(S, k), then U may be regarded as a representation of T for some sincere subposet T of S. There is a more theoretical proof in [Ringel 84] using integral quadratic forms. Corollary 1.3.9 If S is a finite poset, k is a field, and rep(S, k) has fin ite representation type, then each indecomposable U in rep(S, k) has rank less than or equal to 6 and End U = k.
1.3 Finite Representation Type
27
All the representations given for critical posets in the proof of Theorem 1.3.6( ¢::) can be constructed from the fundamental representation UA of S4 defined in Example 1.1.6. These constructions employ the Zavadskij derivative on a finite poset relative to a suitable pair of elements of S. This derivative is also used in the proof of Theorems 1.3.6 and 1.4.4 given in [Simson 92]. Although the proofs are not included, definitions and some examples are given in the remainder of this section. An ordered pair of elements (a, b) in S is called suitable if a is not less than or = S\(a<J U b" of S is either empty or a chain equal to b, and the subposet, C = {cl < Cz < . . . < Cm }, where a <J is the set of all elements in S greater than or equal to a, and b" is the set of all elements in S less than or equal to b. Define two copies of C,
sg
C-
= {cl < Cz <
.. . < c;;;} and C+
= {ci
< ct < .. . < c;}
if sg is not empty . Otherwise, let C- and C+ be empty. The derivativeof S with respect to the suitable pair (a, b), Sf = O(a,b)S, is the disjoint union of posets a <J U b" U c- U C+ with additional order relations given below. 1. If x , Y E a<J U b" , then x ::: y in Sf if and only if x ::: y in S. 2. < ci for all i. 3. If x E br>\{b}, then x < ci in Sf if and only if x < c, in S. 4. If x E a<J\{a}, then x > ci in Sf if and only if x> Cj in S. 5. a < b, a < ci , and c;;; < b. 6. If the above relations result in x < y and y < x for some x, y E Sf, then set x = y.
c;
The following example demonstrates, in terms of derivatives, an interrelationship among the five critical posets 0, 1, 1, 1), (2,2,2),0 ,3,3), (N, 4), and (1,2,5) listed in Theorem 1.3.6.
Example 1.3.10 2
4 6
I I I (a) If S = (2,2,2) = 1 3 5, then (5,2) is suitable and 2
5 contains S4 = {4- , 1, 3+ , 6} as a subposet.
4 7
I I I I 2 5, then (5, 4) is suitable and O(5,4)S = 3 6
(b)
If S
= (1,3,3) =
28
1. Representations of Posets over a Field
7
~ 1~ ~
3j 't- ~6 2
contains (2, 2, 2) 8
= {2 < 3, 1- <
1+,6 < 7} as a subposet.
I
7
I
1 3 6 (c)IfS
N I
= (N ,4)=2
4
5, then (5, I) is suitable and 0(5.1)S =
8
2
4 subposet.
contains (1, 3, 3) = {l,4 < 3' < 3+,6 < 7 < 8} as a
8
I
7
I
3 6
I I I 4, then (4, 3) is suitable and
2 5 (d)
If S = (1,2,5)
=
O(4,3)S
=
8
21- 4 contains(N,4)={2<3>1-
S1
1. 1 < 2,5 < 6;
1.3 Finite Representation Type
29
2.3- < 3+,4- < 4+; 5. 5 < 2,5 < 3+,4- < 2. The numbering corresponds to that found in the definition of O(a,b)S, and 3, 4, and 6 of the definition add no new relations. Thus, 0(5 ,2)(2, 2, 2) is as pictured in (a). A derivative O(a ,b) on a finite poset S with suitable pair (a, b) induces a correspondence O(a,b) : rep(S, k) --+ rep(S' k) . Let S be a finite poset with suitable pair (a, b) and = {cl < ... < cn }. Given V = (Vo, Vi : i E S) E rep(S, k), define O(a ,b)V = W E rep(S', k) as follows . Choose a subspace V of U« = Vb with
si
o, + o, = V EB o, and let W = (Wo, Wi: i E S') E rep(S', k), where
Wo
= Vo/V, Wc(j)- = ((Vb n VC(j») + V)/V for each j s n,
WC(j)+ = (Va + VC(j»)/V foreachj
= (VI + V)/V for each t E a<J U b" ,
Theorem 1.3.11 Suppose S is a finite poset with suitable pair (a, b), Si {cl < .. . < cn }, and S' = O(a,b)S,
=
(a) The correspondenceO(a,b) : rep(S, k) --+ rep(S', k) yields an onto correspondence O(a,b): Ind(S, k) --+ Ind(S' , k). (b) IInd(S, k)1 = IInd(S', k)1 + 1 + n. In particular, rep(S, k) has finite representation type if and only if rep( S', k) has finite representationtype. PROOF. [Simson 92, Theorem 9.10] . It appears from the definition that the derivative of a poset is more complicated than the poset. However, as a consequence of Theorem 1.3.11(b), if S is a finite poset, then S has finite representation type if and only if a repeated application of the derivative eventually results in a l-element poset. There is another "derivative" of a finite poset with this same property that can be used to classify those finite posets S such that rep(S, k) has finite representation type [Gabriel 73A] . Corollary 1.3.12 There are embeddings Ind(S4, k) ~ Ind«2, 2, 2), k) ~ Ind«(l, 3, 3), k) ~ Ind«N, 4), k) ~ Ind«(l, 2, 5), k). PROOF. Theorem 1.3.1 1(a), Example 1.3.10, and Proposition -1.3.1. As an illustration, the last example of this section shows how the representation for S = (2, 2, 2) in Theorem 1.3.6(ii) can be constructed from the one for (l, 3, 3)
30
I. Representations of Posets over a Field
in Theorem 1.3.6(iii) and vice versa. The inverse of the correspondence Theorem 1.3.11 is spelled out in [Simson 92, Section 9.3].
O(a.b)
of
4 7
I I I I 2 5 with suitable pair (5,4) and 3 6
Example 1.3.13 Let S = (l , 3, 3) = V E rep(S , k) as given in Theorem l.3 .6(iii). Then o(S.4)S ;2 (2,2,2) and 0(S,4)V contains the representation/or (2,2,2) listed in Theorem 1.3.6(ii). PROOF. The representation V E rep«l, 3, 3), k) in Theorem 1.3.6(iii) is
= k n $ k n $ k n $ k", n V2 = $ (l + l)k $ 0, n n V4 = 0$ k $ k $ r,
V3
= 0$ kn $
Us
= k" $
V6 = k" $
V7
= k" $
Vo
VI = (l
°
°° $
$ k" ,
+ l)k n $
(l
+ l)k n ,
kn $ 0 ,
°° $
(l
$ 0,
+ A)k n $
k",
Notice that V s + V 4 = V s $ V4 , so choose V = V s in the definition of 0(S,4 )V, Then O(S,4)V is given by WI- =«U4 n VI ) + V)/ V =OEB (1 + IW,
n
Wo = Vo/V =k EB k" EB k" , WI+ =(Vs + VI)/V =k EB(1 n
+ l)k
n
W3 = (V3 + V)/ V = k" EB k" EB 0, Ws = (Vs
+ V)/ V
W2=(V2 + V)/V =(1
+ I)k n EB 0,
W4 = (U4 + V)/ V = k" EB k" EB k" , W6 = (U6 + V )/ V = 0 EB 0 EB k" ,
= 0 EB 0 EB 0,
W7 =(U7 + V)/V =(1 +A)k EBk n
,
n
•
By Example 1.3.1O(b), (2,2,2) = {2 < 3,1- < 1+,6 < 7} is a subposet of 0(S,4)S, Then (Wo, W6 S; W7, W 2 S; W3, W I- S; WI +) is the representation of
(2,2,2) given in Theorem 1.3.6(ii).
EXERCISES
1. Let S = (1,2,2). (a) Find all subposets of S. (b) List all elements ofInd(S, k). 2. Prove Proposition 1.3.5. 3. Show that End V and (v).
= C(A) for each of the representations V in Theorem 1.3.6, (iii), (iv),
4. Confirm the assertions of Example 1.3.10.
1.4 Tameand WildRepresentation Type
31
5. Prove that if U = (Uo, U, : i E S) E rep(S, k) is indecomposable and i E S with 0=1= U, =1= Uo, then there is a 3-element antichain T that is a subposet of S with i E T . 6. Prove, without reference to Theorem 1.3.7, that if S is a sincereposet with w(S) = 3, then S has 3 maximal elements.
1.4 Tame and Wild Representation Type Infinite representation type is typically divided into two kinds. Heuristically, if rep(S, k) has tame representation type, then there is a potential classification scheme for indecomposables in rep(S, k). On the other hand, if rep(S, k) has wild representation type, then it is unrealistic to attempt a complete classification of indecomposable representations. Representations of finite posets over a field can be generalized to an arbitrary ring R. It is assumed herein that a ring has an identity and that R-modules are left unitary R-modules unless otherwise specified. Given a finite poset S, define rep(S, R) to be the category with objects U = (Uo, Ut : i E S), where Uo is a finitely generated free R -module, each Ut is a finitely generated free R-summand of Uo, and if i :::: j in S, then U, 5; U] . Morphisms in this category, called representation morphisms, are R-module homomorphisms f: Uo ~ U with f(U j ) 5; U; for each i in S. Then rep(S, R) is an additive category just as in Lemma 1.2.1, since R-modules with R-homomorphisms form an additive category. Let X and X' be additive categories. An (additive)functor F : X ~ X' is a correspondence F from the objects of X to the objects of X' and a group homomorphism F: Hom(U, V) ~ Hom(F(U) , F(V» for each pair of objects U, V of X such that F(fg) = F(f)F(g) for each g E Hom(U, V), f E Hom(V, W), and FOu) = IF(u). For example, the correspondences H of Theorem 1.2.3 and F+, F- of Proposition 1.3.1 are functors. If F is a functor, then F(U $ V) is isomorphic to F( U) $ F( V) as a consequence of the definition of a direct sum in an additive category. A functor F preserves indecomposables if F(U) is indecomposable whenever U is indecomposable and reflects isomorphisms if U is isomorphic to V whenever F(U) is isomorphic to F(V). Moreover, F is a faithful functor if F :Hom(U, V) ~ Hom(F(U), F(V» is one-to-one for each pair U, V of objects of X and a full functor if F :Hom(U, V) ~ Hom(F(U), F(V» is onto for each pair U, V of objects of X. Notice that a fully faithful functor preserves indecomposables, since F(e) is an idempotent if and only if e is an idempotent, and reflects isomorphisms, since F(fg) = F(f) F(g) = 1 if and only if f g = 1. Let k(x , y) denote the k-algebra of polynomials in noncommuting variables x and y with coefficients in a field k, and let modk(x, y) denote the category of k(x, y)-modules that have finite k-dimension. The endomorphism ring of N E rep(S, k < x, is denoted by Endk(x ,y)N. Write EndkN for the endomorphism ring of N regarded as a k-representation, albeit one with possibly infinite dimension .
o
y»
32
1. Representations of Posets over a Field
The category rep(S, k) has wild representation type if there is an N E rep(S,k(x,y) such that the functor FN:mod k(x,y) ~ rep(S ,k) given by FN(M) = W 0k(x.y) M preserves indecomposables and reflects isomorphisms and Endk(x.y)N = EndkN. The category rep(S ,k) has fully wild representation type if FN is a fully faithful functor for some N . If rep(S, k) has fully wild representation type, then rep(S, k) has wild representation type, since a fully faithful functor preserves indecomposables and reflects isomorphisms.
Example 1.4.1 The category rep(S5, k) has fully wild representation type. PROOF. Define N E rep(S5, R) for R
N
= (R €I'
= k(x, y) by
R, R €I' 0, 0 €I' R, (1 + l)R, (l
+ x)R, (l + y)BJ
As in Example 1.1.7, EndkN = {r E EndkR :xr = rx, yr = ry}. Then EndkN is isomorphic to R = EndRN, the endomorphism ring of R as an R-mOOule. If ME mod R, then
N 0R M
= «R €I'
R) 0R M, (R €I' 0) 0R M, (0 €I' R) €I'R M,
+ l)R 0R M, (l +x)R 0R M, (l + y)R 0R M) (M €I' M, M €I' 0, 0 €I' M, (l + l)M, (l +x)M, (l + y)M). (l
=
Hence, N 0R M E rep(S, k), since M has finite k-dimension. ForM,M' E modk(x, y),define¢ : Hom(M, M') ~ Hom(N®RM , N0 RM') by ¢(f) = 1 0 f . Then ¢ is a one-to-one group homomorphism. Furthermore, ¢ is onto because a representation morphism g : N 0 R M ~ N 0 R M' is of the form g = (f, f), f a k-homomorphism from M ~ M' with fx = xf and fy = yf. It follows that f is an R-homomorphism from M to M' with 1 0 f = g. A salient property of rep(S5, R) confirmed in Example 1.1.7 is that each finitedimensional k-algebra can be realized as the endomorphism ring of some U E rep( S5, k). As an immediate consequence of the following proposition, if rep( S, k) has wild representation type, then rep(S, k) contains a copy of each finitedimensional indecomposable R-module for each finite-dimensional k-algebra R. This is why it is unreasonable to expect an explicit classification of all indecomposable representations in a category with wild representation type . Let Ind(mod R) denote the set of isomorphism classes of indecomposable modules in mod R.
Proposition 1.4.2 [Brenner 74A] Let R be afinite-dimensional k-algebra. (a) There is a fully faithfulfunctor F: mod R ~ mod k(x, y). (b) If rep(S, k) has wild representation type, then there is an embedding of Ind(mod R) into Ind(S, k).
1.4 Tameand Wild Representation Type
33
PROOF. Let {Xl, " " x n } be a k-basis for R with Xl = 1 and let A and B be (n + 2) x (n + 2) R-matrices as defined in Example 1.1.7. Specifically,
0 1 0 0 0 1
0 0 0 0
0 0 0 0 0 0
0 1 0 0
B=
A=
0 1 1 0
0 0
0 0 1 Xl
0 0 0 1
0 0 0 0
0 0 0 0
0 0 0 0
0 0
0 0
1 Xn
0 0 1 0
Suppose M is an R-module with finite k-dimension. Define F(M) = M n +2 , a k{x, y)-module in mod k{x , y) via the multiplication xm = Am, and ym = Bm for each min M n+l . If f : M -+ M' is an R-homomorphism, define F(f) = f I , I an (n + 2) x (n + 2) identity matrix . It is routine to verify that F : HomR(M , M') -+ Homk(x ,y)(F(M), F(M'» is a one-to-one group homomorphism. To see that F is onto, let g : F(M) -+ F(M') be a k(x , y)-homomorphism. A matrix computation, just as in Example 1.1.7, shows that g = (f, . . . , f) with f E Homk(M , M') and f Xi = Xi f for each 1 SiS n. Thus, g = f I with f E HomR(M , M') as desired . Call rep( S, k) endowild if for each finite-dimensional k-algebra R, there is V E rep(S, k) with End V isomorphic to R .
Corollary 1.4.3 Let S be a finite poset and k afield. Ifrep(S, k) has fully wild representation type, then rep(S , k) is endowild.
PROOF. Let R be a finite-dimensional k-algebra. Then R E mod R with EndR(R) = R. By Proposition 1.4.2, F(R) = M E mod k(x , y) with Endk(x,y)(M) = R, because F is fully faithful. Since rep(S, k) has fully wild representation type, U = N ®k(x,y) M E rep(S , k) with End V isomorphic to R.
Theorem 1.4.4 [Nazarova 75) Let S be afinite posetandk afield. Thenrep(S , k) has wild representation type if and only if S contains Ss, 0 ,1,1,2), (2, 2, 3), 0,3 ,4), (N , 5), or 0,2,6) as a subposet. PROOF. (<=) By Proposition 1.3.1, if T is a subposet of S, then there is a fully faithful functor F+ : rep(T, k) -+ rep(S, k). Since the composition of fully faithful functors is fully faithful, it is sufficent to show that if T is one of Ss, 0, 1, 1,2), (2,2,3),0,3 ,4), (N, 5), or 0,2,6), then rep(T, k) has fully wild representation type. In particular, rep(S, k) would then have wild representation type. The case that T = Ss is presented in Example 1.4.1. Following is a list of representations NT = (Vo, Vi : i E T) with F NT : mod k{x, y ) -+ rep(S, k) a fully faithful functor. The proof is a computation like that of Example 1.4.1 to show that EndkNT = R = EndR(R) and Homk(x,y}(M, M') is isomorphic to
34
1. Representations of Posets over a Field
Hom(NT ~k(x,y) M, NT ~k(x ,y) M') for each M , M' in mod k(x, y). Write R = k(x, y).
5 (i) T=(Ss)orT=(1,1 ,1,1,2)=1
I
2 3 4
= R$R$O$O,
Uo=R$R$R$R,
UI
Uz = 0$0$ R $ R,
U3 = (1 + 0 + 1 + O)R $ (0 + 1 + 0 + 1)R,
U4 = (1 +O+x + 1)R ,
Us = (1 + 0 + x + 1)R $ (0 + 1 + y + O)R.
5
I I I I
246 (ii) T
= (2,2, 3) = 1
3 7
Uo = R $ R $ R $ R $ R $ R ,
= (y + x + 0 + 1 + 0 + l)R $ (0 + 1 + 1 + 0 + 1 + O)R, Ui = (1 + 0 + 1 + 0 + 0 + O)R $ (0 + 1 + 0 + 1 + 0 + O)R UI
$(y+x+O+ 1 +0+ 1)R$(0+ I + 1 +0+ 1 +O)R,
= 0 $ 0 $ 0 $ 0 $ R $ R, U7 = R$O$O$O$O$O, U3
U4 = 0 E9 0 $ R E9 R E9 R $ R,
U6
=R$
Us = R $ R $ R $ R $ 0 $ O.
8
I I I 3 6 I I (iii) T = (1,3,4) = 1 2 5 4 7
Uo = R $ R$R $ R $ R $ R$ R $ R,
UI
= (0+ 1 +
1 +0+ 1 +O+O+O)R
$~+x+0+1+0+1+0+mR
$ (1 + 0 + 1 + 0 + 0 + 0 + 1 + O)R $(0+ 1 +0+ 1 +0+0+0+ 1)R, Uz = 0 $ 0 $ 0 $ 0 $ 0 $ 0 $ R $ R,
R $ 0 $ 0 $ 0 $ 0,
1.4 Tame and Wild Representation Type
= 0 $ 0 $ 0 $ 0 $ R $ R $ R $ R, U4 = 0 $ 0 $ R $ R $ R $ R $ R $ R, U5 = R $ 0 $ 0 $ 0 $ 0 $ 0 $ 0 $ 0, U6 = R $ R $ R $ 0 $ 0 $ 0 $ 0 $ 0, U3
U? = R $ R $ R $ R $ R $ 0 $ 0 $ 0,
Ug
= R$ R$ R$ R$ R$ R$ R$
O.
9
I I 247 NI 136 I (iv) T = (N , 5)= 5 8
Uo UI
= R $ R $ R $ R $ R $ R $ R $ R $ R $ R, = 0 $ 0 $ 0 $ 0 $ 0 $ 0 $ R $ R $ R $ R,
~=O$O$R$R$O$O$R$R$R$R
$~+0+0+0+1+0+0+0+1+mR
$(0+0+0+0+0+ 1 +0+0+0+ l)R , U3 = (0 + 0 + 0 + 0 + 1 + 0 + 0 + 0 + 1 +
O)R
$(0+0+0+0+0+ 1 +0+0+0+ l)R,
U4 = (0 + 1 + 1 + 0 + 1 + 0 + 0 + 0 + 0 + O)R $0+x+0+l+0+l+0+0+0+~R
$ (1 + 0 + 1 + 0 + 0 + 0 + 1 + 0 + 0 + O)R $~+1+0+1+0+0+0+1+0+mR $~+0+0+0+1+0+0+0+1+mR
$(0+0+0+0+0+ 1 +0+0+0+ 1)R ,
= R $ 0 $ 0 $ 0 $ 0 $ 0 $ 0 $ 0 $ 0 $ 0, U6 = R $ R $ R $ 0 $ 0 $ 0 $ 0 $ 0 $ 0 $ 0, U? = R $ R $ R $ R $ R $ 0 $ 0 $ 0 $ 0 $ 0,
U5
35
36
1. Representations of Posets over a Field ~=RmRmRmRmRmRmRmomom~
U9 = R m R m R m R m R m R m R m R mom o. 9
I I 3 7 I I 2 6 I 5 I T = (1,2 ,6) = 4 8
(v)
u« = Rm Rm RmRmRm RmRmRmRm RmRm R,
UI
= (0 + 0 + 0 +
1 + 0 + 0 + 0 + 0 + 0 + 1 + 0 + y)R
m(o+o+ 1 +0+0+0+0+0+ 1 +0+ 1 +x)R m~+1+0+0+0+1+0+0+0+1+1+~R
m(1 +0+0+0+ 1 +0+0+0+ 1 +0+0+ I)R m~+0+0+0+0+0+0+1+0+0+I+mR
EB(O+O+O+O+O+O+ 1 +0+0+0+0+ I)R, Uz = 0 EB 0 EB 0 EB 0 EB 0 EB 0 EB 0 EB 0 EB REB R EB R EB R,
= 0 EB 0 EB 0 EB 0 EB REB R EB R EB R EB R EB R EB R EB R, U4 = R EB R EB 0 m 0 EB 0 m 0 EB 0 EB 0 EB 0 EB 0 EB 0 EB 0, U3
U5 = R EB R EB R EB 0 EB 0 EB 0 EB 0 EB 0 EB 0 EB 0 EB 0 EB 0,
= R EB R ED R m R m 0 ED 0 m 0 ED 0 ED 0 ED 0 ED 0 m 0, U7 = R ED R ED R ED R ED R ED REDO ED 0 ED 0 ED 0 ED 0 ED 0, U6
~=REDREDREDREDRmREDREDRmOEDOmOED~
U9
= R ED R ED R m R ED R ED R ED R ED R ED R ED REDO ED O.
(=>) This argument is long and technical and as such is omitted. A complete proof can be found in [Simson 92, Theorem 15.3].
Define Rep(S, k) to be the category with objects U = tU«, Ut : i E S), where Uo is a k-vector space with countable dimension, each U, is a subspace of Uo, and if
1.4
Tame and WildRepresentation Type
37
i ::: j in S, then Vi 5; U]. Morphisms in this category are k-line ar transformations f: Vo -+ V~ with f( V i) 5; V ; for each i in S. Then rep(S , k ) is the full subcategory of Rep(S , k ) with objects those V E Rep(S , k ) such that V o has finite k-dim ension. For a ring R , Mod R den otes the category of all count ably generated R-modules. A category rep (S , k ) has strong ly wild representation type if there is an N in rep(S, k (x , y}) such that Endk(x,y)N = EndkN and the functo r FN : M odk (x , y} -+ Rep(S , k ) preserves indecomposables and reflects isomorphisms. Clearly, if rep(S , k ) has strongly wild representation type , then it has wild representation type, since mod k( x , y } is a full subcategory of Mod k (x , y }. All of the variations on the definitions of wild repre sentation type are equivalent for rep(S , k ).
Proposition 1.4.5 Let S be afinite poset and k afield. The fo llowing statements are equivalent:
(a) rep(S, k ) has fully wild representation type ; (b) rep(S, k) has wild representation type ; (c) rep(S , k) has strongly wild representation type. In this case, rep(S , k ) is endowild. PROOF. (a) ~ (b) is a consequence of the definition of fully wild representat ion type and the proof of Theorem 1.4.4. (c) ::::} (b) follows from the definition of strongly wild repre sentation type . (b) ::::} (c) Assume that rep(S , k ) has wild representation type, T is one of the posets Sj,(1, 1, 1, 2), (2, 2,3), (1 , 3, 4), (N ,5), or (1, 2,6), and let N T be the associated representation given in (i)-(v) of Theorem 1.4.4. The same proof as that of Theorem 1.4.4 shows that Homk(x.y)(M , M ' ) is isomorphic to Hom(N T I8lk(x,y) M , NT I8lk(x ,y) M ' ) for each M , M' in Mod k( x , y }. This shows that rep(T , k ) has strongl y wild representation type for each such T . Finally, if T is a subposet of S, then there is a fully faithful functor F + : Rep(T , k ) -+ Rep(S , k ) defined just as in Proposition 1.3.1 for rep(S , k ), The conclusion is that rep(S , k ) has strongly wild repre sentation type. For the last statement, apply Corollary 1.4.3.
As a consequence of Theorems 1.4.4 and 1.3.6, rep(S, k) has exactly one type from among finite representation type , wild representation type, and infinite but not wild representation type. The category rep(S , k) has tam e representation type, provided that it has infinite repre sentat ion type and for each w = (wo, Wi : i E S) there are finitely many N I , . . . , N m E rep(S , k[x]) with EndkNi = EndklxjNi for each i such that if V E rep(S , k ) is indecomposable with cdn V = w, then V is isomorphic to N, I8lklxj B for some i and indecomposable cyclic k [x ]-module B. In other words, if rep(S , k ) has tame representation type, then the elements of Ind(S , k) are characterized up to isomorphism by cdn V , finitely many k[x]representations of S, and indecomposable k[x ]-modules of the form k[ x] / (g(x for some irreducible polynomial g(x).
n
38
1. Representations of Posets overa Field
Theorem 1.4.6 Let S be a finite poset and k a field. The following statements are equivalent:
(a) The category rep(S , k) has tame representation type; (b) The category rep(S , k) has infinite representation type but does not have wild representation type; (c) TheposetS contains Si, (2,2,2),0,3 ,3),0,2,5), or(N , 4) as a subposet but does not contain Ss, 0, I, 1,2), (2,2,3),0, 3,4), (N , 5), or 0,2,6) as a subposet. PROOF. (b) ¢::} (c) Theorems 1.4.4 and 1.3.6.
(a) ¢::} (b) is a complicated argument, proved in [Simson 92, Theorems 15.54 and 15.57] by induction and the Zavadskij derivative. The next lemma relates representations of a finite poset S = (S,:::) and the opposite poset sop = (S, :::). A contravariant functor F : X ~ X' between two additive categories X and X' is a correspondence that satisfies all of the properties of a functor except that F(fg) = F(g)F(f) for each I and g, i.e., the order of composition is interchanged by F. For example, if X is the category of finite-dimensional vector spaces, then F: X ~ X, defined by F(V) = Hom(V , k), is a contravariant functor. In particular, if I : V ~ W is a k-linear transformation, then F(f): Hom(W, k) ~ Hom(V, k) is defined by F(f)(a) = af for each a E Hom(W, k) . Moreover, for each finite-dimensional vector space V, there is an isomorphism ¢v : V ~ F 2(V) = Hom(Hom(V, k), k) given by 1>v(v)(a) = a(v) for each v E V and a E Hom(V, k). This isomorphism is a natural isomorphism from the identity functor to F 2 in the sense that if I: V ~ V'is a k-linear transformation, then F 2(f)¢v = ¢v' I. A contravariant functor F with a natural isomorphism from the identity functor to F 2 is called a contravariant duality.
Proposition 1.4.7 Let S be afinite poset and k afield. There is a contravariant duality a: rep(S , k) ~ rep(SOP , k). Moreover, a : Ind(S , k) ~ Ind(SOP, k) is one-to-one and onto, and rep(S, k) has finite representation type if and only if rep( SOP, k) has finite representation type.
= (Vo , Vi) E rep(S, k), define a(V) = V , where Vo = Homk(VO, k) and Vi = (f: Uo ~ k : I(Vi) = OJ. The proof is an exercise in vector space duality (Exercise 1.4.4).
PROOF. Given V
The converse to the last conclusion of Proposition 1.4.5 remains unresolved .
Open Question: If S is a finite poset, k a field, and rep(S, k) is endowild, does rep(S, k) have wild representation type?
1.5 Generic Representations
39
EXERCISES 1. Let (S(n) , m)
=
{aom, aIm , ... , a nm} be a copy of the poset{ao, al , . . . , an} consisting of an antichain {aI , . .. , an} with ao > ai for each 1:::: i :::: n . Given positive integers n and j , define S(n, j) to be the poset consisting of a disjoint union of the posets (S(n) , 0), . . . , (S(n), j - 1) with the additional relation ai.m :::: ai.m+1 for each 0 :::: i :::: n, 0 :::: m :::: j - 2. Determine, in terms of nand i . exactly when rep(S(n, j), k), k a field, has finite, tame, or wild representation type.
2. Given positive integers nand i , define pen, j) to be the poset lao > bl > .. . > b j_l,al , . .. ,an} with {aI, . .. , a n } an antichain and ao > ai for each 1 :::: i :::: n. Determine, in terms of nand i, exactly when rep(S(n, j), k), k a field, has finite, tame, or wild representation type. 3. Provide the missing computations for the proof of Theorem IAA( {=) . 4. Complete the proof of Proposition 104.7. 5. Let S = (1,2,2). (a) Describe SoP . (b) Find a sincere indecomposable in rep(S, k) that is decomposable when viewed as an element of rep(SOP, k) . (c) Use the duality a to compare the indecomposable representations of rep(S, k) and rep(SOP, k) (see Exercise 1.3.1). 6. Let T be a subposet of a finite poset S and let F+ and F- be the functors defined in Section 1.3. Find a functor G : rep(S, k) --+ rep(T, k) such that both G F+(U) and G F-(U) are isomorphic to U for each U E rep(T, k) .
1.5 Generic Representations There are infinite-dimensional k-representations of a finite poset S, called generic representations, that determine the representation type of rep(S, k) (Theorems 1.5.3 and 1.5.4). Generic representations give rise to, and arise from, finite-dimensional representations. For a ring R and a finite poset S, define a category of R-representations Rep(S, R) with objects U (Uo, U, : i E S), where U» is a countably generated free R-module, each U, is a countably generated free R-summand of Uo, and U, S; Uj if i ::: j in S. A morphism f : U ~ V is an R-homomorphism f : Uo ~ Vo with f(U i) S; Vi for each i in S. An R-module M is simple if 0 and M are the only submodules of M. The Rmodule M has a composition series if there is a finite ascending chain Mo 0C M I C . , . C M; M of R-submodules of M such that each Mil Mi - I is simple. In this case, M is said to have R-length n. The R-Iength of M is independent of the composition series for M as a consequence of the classical Jordan-Holder theorem [Anderson, Fuller 74]. If U tU«, U, : i E S) E Rep(S, R), then Uo is a left module over End U, the endomorphism ring of U , defined by fu feu) for u E Uo and f E End U. This
=
=
=
=
=
40
I. Representations of Posetsover a Field
operation is well-defined, since End V is a subring of the R-endomorphism ring of Vo. The length of Vo as an End V -module is called the endolength of V. Given a field k, a representation V = (Vo, Vi : i E S) E Rep(S, k) is a generic representation if V is indecomposable, Vo has countably infinite k-dimension, and Vo has finite length as an End V -module. If V is indecomposable and Vo has finite k-dimension, then V satisfies all of the conditions for a generic representation except for having infinite k-dimension. A number of examples of generic representations are given below. Generic representations have special endomorphism rings . Following is a brief review ofsome ring-theoretic notions and properties that are used; see, for example, [Anderson, Fuller 74] or [Hungerford 74]. The Jacobson radical JR of a ring R is the intersection of the maximal right ideals of R. Then JR is a two-sided ideal of R with J(RjJR) = and if x EJR and r E R, then 1 - rx is a unit of R. Moreover, R is a local ring if and only if R j JR is a division ring. A ring R is semisimple if JR = and left A rtinian if it has the descending chain condition on left ideals. An ideal I is nilpotent if I", the ideal of R generated by {XIX2 ... Xn : Xi E I}, is equal to zero for some positive integer n. If R is left Artinian, then JR is nilpotent. A semisimple left Artinian ring R is a finite product of full matrix rings over division rings. In this case, R is also right Artinian. Idempotents lift modulo JR if whenever e is an idempotent in R/JR, then there is an idempotent fER with f +JR = e. If JR is nilpotent, then idempotents lift modulo JR . Finally, a ring R is semiperject if R j JR is left Artinian and idempotents lift modulo JR. For example, a left Artinian ring is semiperfect. If M is a finitely generated left R-module and (JR)M = 0, then M = 0, a result commonly referred to as Nakayama's lemma [Anderson, Fuller 74].
°
°
Lemma 1.5.1 Assume V = (Vo, Vi: i E S) E rep(S, k) is a generic representation. Then D = End V j lEnd V is a division k-algebra, (lEndV)m = 0 for some positive integer m, and V o has the structure ofa right D-module with D-dimension Vo = endolength V. PROOF. Assume that V is a generic representation. Then (lEnd V)Vo ;2 (JEnd V)2Vo ;2 . . . is a descending chain of End V -submodules of Uo. Since Vo has finite length as an End V-module, (JEnd v)mvo = (JEnd v)m+lVo for some positive integer m. Also, Vo is finitely generated as an End V-module. By Nakayama's lemma, (lEndV)mVo = 0. Hence, (lEndV)m = 0, since (lEndV)m ~ EndV. Since V is indecomposable, the only idempotents of End V j JEnd V are and 1. But End V j JEnd V is left Artinian, because V has finite endolength. It follows that the semisimple left Artinian ring End V j JEnd V = D is a division k-algebra. In fact, D is a unique simple End V module. This is because D is a division algebra and if M is a simple End V-module, then (lEnd V)M = as an application of Nakayama's lemma. Consequently, M is isomorphic to D . Next, let
°
°
1.5 Generic Representations
°
41
M« = C M I C .. . C M; = V o be an End V composition series for V o with n = endolength V . Each M;/Mi _ 1 is a simple module, hence isomorphic to D. Write M;/ Mi - I = (Xi + Mi-dD for some Xi E Mi. Then U« has the structure of a D-module with U« = XI DEB · .. EB X n D. In particular, the D-dimension of Vo is n, the endolength of V . Define k(x) = (f(x)/ g(x) : f(x) , g(x) E k[x], of the polynomial ring k[x].
° =1=
g(x)} , the field of quotients
Example 1.5.2 Given afield k, there is a generic k -representation V of 5s with End V noncommutative, lEndV =1= 0, End V / lEndV = k(x), (lEnd V)z = 0, and endolength V = 4. Let R be the ring of2 x 2 matrices of the form (3 D for A, (J E k(x). Then R/lR = k(x),JR =1= 0, and (JR)z = 0, noting thatJR consists of those matrices in R having all zeros on the diagonal. Moreover, R is a k(x )-vector space of dimension 2. Define V = (Vo, VI, o: V3 , V 4, Vs), where
PROOF.
°
Vo = REB R , VI = REB 0, Ui = EB R , V 3 = 0 V4 = 0 +x)R, V s = 0 + c)R,
+ I)R,
(8 6)
and c = is an element of R . To see that End V is isomorphic to R, notice that R embeds in End V via r t-+ (JLr , JLr), where JLr is left multiplication by r. Conversely, if f E End V, then f = (g , g) with g E Endk(R) , since VI, Vz, and V3 are preserved by f. But gx = xg, since f(U4) S; V4, and gc = cg; since f(V s) S; U« . It follows that g E EndR(R); hence g is multiplication by some r in R, as desired. Finally, End V / lEnd V = k(x), so that V is indecomposable. Moreover, Vo has countably infinite k-dimension, and endolength V = 4 is the k(x)-dimension of Vo = REB R. This completes the proof. The category rep(5, k) is generically trivial if there are no generic representations in rep(5, k). Theorem 1.5.3 Let 5 be a finite poset and k afield. Then rep(5 , k) has finite representation type if and only ifrep(5, k) is generically trivial. Assume that rep(5, k) has finite representation type. If V is an indecomposable in Rep(5, k), then Vo has finite k-dimension [Ringel , Tachikawa 75]. Thus, rep(5, k) has no generic representations. Conversely, assume that rep(5, k) has infinite representation type. As a consequence of Theorem 1.3.6, it suffices to show that if 5 is one of the critical posets , namely 54, (2, 2, 2),0,3 ,3),0 ,2,5), or (N , 4), then there is a generic representation in rep(5 , k). For each of the critical posets 5, a generic representation V is listed below. Each representation is derived from the finite-dimensional representation of 5 PROOF.
42
1. Representations of Posets over a Field
given in the proof of Theorem 1.3.6 with k" replaced by the field k(x). In each case, it can be shown,just as in the proof of Theorem 1.3.6, that End V = k(x) . In particular, V is an indecomposable k-representation with infinitek-dimension. Since Vo has finitek(x)-dimension with k(x) = End V , V E rep(S, k) is a generic representation. (i) S = S4
= k(x) EB k(x) , V 2 = °EB k(x),
VI = k(x) EB 0,
Vo
V 3 = (l
+ l)k(x) ,
V 4 = (l
+ x)k(x).
246
(ii) S
I 3I 5I
= (2, 2, 2) = I
= k(x) EB k(x) EB k( x), V2 = (l + x)k(x) EB k(x) , Vo
V 4 = k(x) EB k(x) EB 0,
V6 = k(x) EB (l
VI =OEBOEBk(x),
V3 = (l
+ l)k(x) EB 0, V 5 = EB (l + l)k(x),
+ l)k(x).
°
4 7
I I
3 6
I I
(iii) S=(1,3,3)=
2 5
Vo = k(x) EB k(x) EB k(x) EB k(x),
° =°
VI = (l
+ l)k(x) EB (l + l)k(x),
°EB
V2 =
EB (1 + l)k(x) EB 0,
U3 =
V4
EB k(x) EB k(x) EB k(x),
V5 = k(x) EB
V6 = k(x) EB
°°
EB EB k(x) ,
V7
8
I
7
1 3
!
NI
(iv) S = (N, 4) = 2
4 5
Uo = k(x) EB k(x) EB k(x) EB k(x) EB k(x),
V2 = (l
+ l)k(x) EB (1 + l)k(x) EB 0,
k(x) EB k(x) EB 0,
°EB °EB 0,
= k(x) EB (1 + x)k(x) EB k(x).
1.5 Generic Representations
= 0 EB (1 +
43
VI = k(x) EB k(x) EB k(x) EB k(x) EB 0,
V4
V3 = 0 EB k(x) EB k(x) EB (1 + l)k(x),
Vs = 0 EB 0 EB 0 EB 0 EB k(x),
V6 = k(x) EB 0 EB 0 EB 0 EB k(x) ,
V7 = k(x) EB 0 EB 0 EB k(x) EB k(x),
1)k(x) EB 0 EB 0,
Vs = k(x) EB (1 + x)k(x) EB k(x) EB k(x) . 8
I I 3 6 I I 2 5 I S = (1,2,5) = 4 7
(v)
Vo
= k(x) EB k(x) EB k(x) EB k(x) EB k(x) EB k(x),
VI = (0 + 1 + 1 + 0 + 0 + O)k(x) + (0 + 0 + 1 + 0 + 0 + 1)k(x) + (0+0+0+ 1 + 1 +O)k(x),
V2 = (1 + l)k(x) EB (1 + 1)k(x) EB 0 EB 0, V3 = k(x) EB k(x) EB k(x) EB k(x) EB 0 EB 0, V4
= OEBOEBOEBOEBOEBk(x),
V s = OEBOEBOEBOEBk(x)EBk(x),
V6 = k(x) EB 0 EB 0 EB 0 EB k(x) EB k(x) , V7 = k(x) EB 0 EB 0 EB k(x) EB k(x) EB k(x) ,
V s = k(x) EB (1 + x)k(x) EB k(x) EB k(x) EB k(x). Call rep(S, k) generically wild if there is a generic representation V such that End V / lEnd V contains a copy of k(x , y).
Theorem1.5.4 Let S be afinite poset and k afield. For the following conditions, (a)
=> (b) => (c).
(a) rep(S, k) has wild representation type; (b) rep(S, k) is generically wild; (c) rep(S , k) is endowild. Moreover, that n ~ 5.
if S = Sn, then (a), (b), and (c) are all equivalent to the condition
PROOF. (a) => (b) Let N = (No, N; : i E S) E rep(S, k(x, y) be as given in the definition of wild representation type. There is a division k-a1gebra D with
44
I. Representationsof Posets over a Field
countable k-dimension containing k (x, y) by Lemma 3.5.7. Define V = N Q9k(x .y}D = (Vo, Vi : i E S) = (No Q9k(x, y) D, N, Q9k(x ,y}D : i E S) E rep(S, D). Then Vo has countably infinite k-dimension and finite D-dimension, since No is a finitely generated k{x, y)-module. Moreover, V is indecomposable in Rep(S, k), because rep(S, k) has strongly wild representation type by Propos ition 1.4.5. Since D is contained in End V and Vo is a finite-dimensional D-space, it follows that V has finite endolength. Hence, V is a generic representation with k (x, y) £ End V / lEnd V . This proves that rep(S, k) is generically wild. (b)::::} (c) Let V = (Vo, Vi : i E S) be a generic representation for rep(S, k) with End V / lEnd V = D, a division k-algebra containing k{x, y) . As a consequence of Lemma 1.5.1, V = WD for some W = (Wo , Wi: i E S) E rep(S, k{x, y)) . Suppose A and B are two nonzero m x m k-matrices with minim al polynomials f(xi and g(y)j, respectively, and f(x) and g(y) irreducible polynomials. Let R = k(x,y)/K, where K = (xY,f(xi,g(y)j) is an ideal of k(x,y). Then V = WQ9k (x ,y}R = (Wo/KWo, (Wi + KWo)/KWo) E rep(S ,k).Moreover, C(A , B) is contained in Endk(x,y} V , a k-subalgebra of End V, observing that x acts on Wo/K Wo by A and y on Wo/ K Wo by B. In particular, C(A , B) is a ksubalgebra ofEnd V. As noted in Section 1.1, for each finite-dimensional k -algebra A, there are matrices A and B with A isomorphic to C(A, B). Thi s shows that rep(S, k) is endowild. As for the final statement, suppose that rep(Sn, k) does not have wild representation type . Then n .::: 4 by Theorem 1.4.6. If V is an indecomposable in rep(S4, k), then End V is a factor k-algebra of k[x] (Example 6.2.8). Thus, rep(S, k) is not endowild. This proves (c) ::::} (a) for the case that S = S; is an antichain. Notice that (c) ::::} (a) of Theorem 1.5.4 is the open question given in Section 1.4. The final example of this section demonstrates that generic representations in rep(S, k) , necessarily of infinite dimension, can be used to construct indecomposable finite-dimensional representations in rep(S, k) . The proof is a routine computation, just as in Theorem 1.5.3. Example 1.5.5 Let V be one ofthe generic representations for rep(S, k) listed in the proofofTheorem 1.5.3 and g(x) an irreducible polynomial in k[x]. Define V /g(xYV = (Vo/g(xYVo, (Vi + g(x)eVo)/g(xYVo: i E S) E rep(S , k). Then
V /g(xYV is indecomposable with endomorphism ring k[x]/(g(xY).
The following questions are motivated by analogous results for finitedimensional algebras over algebraically closed fields [Crawley-Boevey 91] . Open Questions:
1. Is there an existence condition on generic representations that is equivalent to rep(S, k) having tame representation type?
1.5 GenericRepresentations
45
2. Is there a condition on the endomorphism ring of generic representations that is equivalent to rep(S, k) having tame representation type? EXERCISE
1. Do the computations to complete the proof of Theorem 1.5.3 and confirm Example 1.5.5. NOTES ON CHAPTER 1
The invention of a derivative of a finite partially ordered set by Nazarova and Roiter in the late 1960s or early 1970s was a seminal event in the subject of representations of finite partially ordered sets (see [Simson 92]). This derivative was used by Kleiner in his classification of posets of finiterepresentation type and sincereposets (Theorems 1.3.6 and 1.3.7). Partitioned matrix representations of a field k over a partially ordered set, herein called Mu for a representation U, are used in [Nazarova, Roiter 75] for the investigation of modules over finite-dimensional k-algebras. The category of S-spaces over k, denoted in this manuscript by rep(S, k), is investigated in [Gabriel 72] as a tool for the computation of representation type of representations of quiversand categoriesof modules. Various relationships between representations of finitepartially ordered sets and finitely generated modulesover finite-dimensional k-algebrasare givenin [Ringel84], [Ringel91, Section4], [Simson 92], and, from a historical perspective, [Gustafson 82]. In the literature on modules over k-algebras, it is commonly assumed that k is an algebraically closed field. Applications to abelian group theory frequently involveeither the field of rational numbers or finite fields, which are not algebraically closed. Hence, from the perspective of abelian group theory, results that do not depend on the field k, such as the representation type of rep(S, k), are most useful. The elementaryintroductorymaterial in Sections 1.1, 1.2, and the beginningof Section 1.3 is folklore. Included in [Simson 92] is a comprehensive treatment of / -spaces for a finitepartiallyorderedset l . There are a numberof proofsof Kleiner's theorems,Theorems 1.3.6 and 1.3.7. The proofs of these theoremsin [Ringel84] employ the theory of integral quadratic forms,while those in [Simson92] use the Zavadiskij derivative; a refined version of the original Nazarova-Roiterderivative. An alternative proof of Theorem 1.3.6 is given in [Kerner81]. Matrix problems for representations of partially ordered sets have been put in more general contexts such as vector space categories and bocses; see, for example, [CrawleyBoevey 88], [Ringel 84], [Rojter 80], and [Simson 92]. This is, in part, an attempt to find categorical settings for the combinatorial classification, via the Nazarova-Roiter or Zavadskij derivative,of indecomposable representations for finiteposets with finiteor tame representation type. The notion of wild representation type for group representations, in the sense of Proposition 1.4.2(b), is observed in [Krugljak 64]. Tame representation type, in the sense that the representation type is infinitebut the indecomposables are classifiable, is demonstrated for group representations in [Heller, Reiner 61]. Tamerepresentation type for rep(S, k) has been subdivided into variouskinds, e.g., domestic,finitegrowth, linear growth,polynomial growth, and one parameter. Characterizations for these subdivisions of the representation
46
1. Representations of Posetsover a Field
type ofrep(S , k) in termsof the finite poset S are known; see [Nazarova 81], [Simson 92], and references therein. Section 1.5 is motivated by moregeneralresultson genericmodules in [Crawley-Boevey 91, 92]. The deepest of these results uses the theory of bocses, which is not available for fields that are not algebraically closed. Progress for the theory of generic representations overfields thatare notalgebraically closedawaits, among otherthings, the resolution of the problemof whether or not endowild is equivalent to wild representation type, as indicated in the open questions for Section 1.5.
2 Torsion-Free Abelian Groups
2.1 Quasi-isomorphism and Isomorphism at p There are two equivalence relations on torsion-free abelian groups that are weaker than group isomorphism, namely quasi-isomorphism and isomorphism at a prime p. Properties of these equivalence relations are conveniently expressed in a categorical setting. Abelian groups are written additively. An abelian group G is torsion-free if na i= 0 for each nonzero integer n and nonzero a E G. Hence , if n is a nonzero integer and a and b are elements of a torsion-free abelian group G with na = nb, then a = b. If G is a torsion-free abelian group, then G is isomorphic to a subgroup of a Q- vector space Q l8lz G = QG, where Q is the field of rational numbers. The group G is identified with its image in QG, so that QG = {qa: a E G}. Let A denote the category with torsion-free abelian groups as objects and group homomorphisms Hom(G , H) for G, H in A as morphisms. It is not difficult to verify that A is an additive category. A group homomorphism is a monomorphism if it is one-to-one and an epimorphism if it is onto. The group endomorphism ring of G is denoted by End G. The rank of a torsion-free abelian group G is the cardinality of a maximal Z-independent subset of G, equivalently the Q-dimension ofQG . Define a category AQ with objects those of A but with morphism sets QHom(G, H) for G, H in A. Composition is defined by (qf)(rg) = (qr)(fg) for q, r E Q and f, g E Hom( G, H). Composition is associative, and 1G is an identity in the AQ-endomorphism ring QEnd G of G. Recall that QEnd G is semiperfect if QEnd G / J QEnd G is left Artinian and idempotents lift modulo JQEnd G.
48
2. Torsion-Free Abelian Groups
Proposition 2.1.1 (a) The category AQ is an additive category. IfG = H EB Lin AQ, then there is an idempotent e E Q End G with H = e(G) in AQ. Conversely, ifG is in A Q and e is an idempotent in Q End G, then G = e(G) EB (l - e)(G ) in AQ. (b) If G is indecomp osable in A Q and Q End G is semipe rfect, then Q End(G ) is a local ring. In pa rticular, Q End G is a local ring if G has finite rank. PR OOF. (a) Not ice that Q Hom(G , H ) is an abelian group for each G , H in AQ with (f + g)(x) = f (x )+ g(x )for each x E G and that composition is distributi ve with respect to +.As for finite dire ct sums, let G i be in AQ and K = G 1 EB G 2 EB · . ·EBGn the group direct sum with injections ij E Hom(G l : K ) and projections Pj : K -+ G i- Then K is a direct sum in AQ with the same injections and projections . To see this, let /j E Q Hom(G j , H ). Choose a nonzerointegerm withm/j E Hom (G I» H) for each j. Since K is a direct sum in A, there is a unique f E Hom(K, H) with fi j = m], for each j. Then (llm)f is a unique element ofQHom(K , H) with (l I m )f i j = f j for each j, as required. Notice that ej = i j Pj is an idempotent in Q End K with I = el + ...+ en and ej em = 0 if j =P m . Con versel y, let e = (mln )f be an idempotent in QEnd G for some min in Q and f E End G . Then G is the direct sum of e(G) and (l - e)(G) in A Q with inclusions as injections. (b) If G is indec omposable in AQ, then Q End G has no idempotents other than o or I by (a) . Hence, Q End G I l QEnd G has no nontrivial idempotents, since idempotents lift modulo l QEnd G . Moreover, Q End GI1 QEnd Gis sernisimple Art inian , whence Q End G I l QEnd G is a division ring. Consequently, Q End G is a local ring. Finall y, if G has finite rank , then Q End G is a finite-dimensional Q -algebra. In this case, Q End Q is Artin ian , hence semiperfect.
Categorical notion s of isomorphism and direct sums in AQ have group-theoretic interpretations that in fact, precede the categorical concepts [Jons son 57, 59] . An abelia n group G is bounded if there is a nonzero integer n with n G = O.
Corollary 2.1.2 Let G and H be objects of AQ. (a) The group G is isomorphic to H inAQ ifandonlyifthereis f E Hom(G, H), g E Hom(H, G), and a nonzero integer n with fg = nIH and gf = n IG . (b) If f E Hom(G, H) , then f is an isomorphism in A Q if and only if f is a monomorphism and Hlf(G) is bounded. Moreover, if f E End G is a monomorphism and G has finite rank, then f is an isomorphism in A Q• (c) The group G is a direct sum Gl EB··· EB G; in AQ if and only if there is a monomorphism f from the group direct sum G 1 EB . . . EB G; into G such that G /image f is bounded. (d) The group G is indecomp osable in AQ ifand only ifwhenever G I(H EB K ) is bounded, then H = 0 or K = O. (e) The group H is asummand of G inA Q ifandonlyifthere is f E Hom(G , H ), g E Hom (H , G ), and a nonzero integer n with f g = niH .
2.1 Quasi-isomorphism and Isomorphism at p
49
(f) lfQEnd Gis semiperfect, then G is a finite direct sum of indecomposables in AQ. Such a decomposition is unique up to isomorphism in AQ and the order of the summands. JfG has finite rank, then QEnd Gis semiperfect.
PROOF. (a) First suppose G and H are isomorphic in AQ.Then there are rfs, min E Q, f E Hom(G, H), and g E Hom(H, G) with (rls)f(mln)g = lH and (mln)g (rls)f = lG. Hence, (rf)(mg) = nslH and (mg)(rf) = nslG, with rf E Hom(G, H) and mg E Hom(H, G), as desired. Conversely, if f E Hom(G, H), g E Hom(H, G), and n is a nonzero integer with fg = nIH and gf = nlG , then (lln)f E QHom(G , H) with (lln)fg = lH and g(lln)f = lG. This shows that (1 In) f is an isomorphism in AQ. (b) Assume that f : G -+ H is an isomorphism in AQ. By (a), there is g E Hom(H, G) and a nonzero integer n with fg = nIH and gf = nlG' Thus, f is a monomorphism and Hlf(G) is bounded , since nH = fg(H) £ f(G) £ H. Conversely, suppose f is a monomorphism and n is a nonzero integer with nH £ f(G) £ H . Then g = (j-I)n E Hom(H, G) with fg =nlH and gf =nlG ' Now apply (a). Next, suppose rank G is finite and f E End G is a monomorphism. Then QEnd G;2 f(QEnd G);2 . .. is a descending chain of right ideals of QEndG. Since G has finite rank, QEnd G is right Artinian. Choose n with r(QEnd G) = r+I(QEnd G). Then there is g E End G and alb E Q with [" = r +l(alb)g. So r(b - afg) = 0 and b = afg , since f is a monomorphism. Hence , bG = afg(G) £ f(G) , Glf(G) is bounded, and f is an isomorphism in AQ by the first part of the proof. Assertions (c), (d), and (e) follow from (a), (b), and the definitions of direct sums and indecomposability in AQ. (f) Since QEnd G 11QEnd Gis Artinian and idempotents lift modulo lQEnd G, it follows that G can be written as the direct sum of finitely many groups G I EEl • . . EEl G n with each G, indecomposable in AQ [Lambek 66]. This uses the fact that for each such sum, 1 = el + ... + en with each ei idempotent and eiej = 0 if i =1= j, as noted in the proof of Proposition 2.I.1(a). Each QEnd G, is a local ring by Proposition 2.1.1(b). Now suppose G = HI EEl • • • EEl Hm in AQ with each H] indecomposable in AQ. Then m = n, and there is a permutation a of {I, ... , n} such that G i is isomorphic to Ha(i) for each I ::: i ::: n (Exercise 1.2.6). The proof is essentially the same as that of Theorem 1.2.2, using the assumption that each Gi has a local endomorphism ring in AQ.The last statement follows from Proposition 2.I.1(b). Standard terminology in abelian group theory for each of the conditions of Corollary 2.1.2 is as follows : (a) G is quasi-isomorphic to H. (b) f is a quasi -isomorphism from G to H or a quasi-automorphism of G if H=G . (c) G is a quasi-direct sum of G I , · · · , G n.
50
2. Torsion-Free Abelian Groups
(d) G is strongly indecomposable. (e) H is a quasi-summand of G. (f) G has the Krull-Schmidt property in A Q. The full subcategory of A Q consisting of finite rank groups is a Krull-Schmidt category.
Much of the structure of a group in A is lost by passing to the category A Q. There are many examples of indecomposable groups in A that are not indecomposable in A Q, for instance any indecomposable almost completely decomposable group (Chapter 5). This problem can be remedied to some extent by localizing a homomorphism group at a prime p instead of tensoring the homomorphism group with Q, as is the case with A Q. Let p be a prime and define a category A p with objects those of A but with morphism sets Hom(G, H)p = Zp@zHom(G, H) for G, H in A where Zp = {m/ n E Q : gcd {n, p} = I} is a subring of Q called the localization of Z at a prime p . Composition is defined by (q @ f)(r @ g) = qr @ fg . It is routine to verify that composition is associative and IG is an identity in (End G)p, the A pendomorphism ring of G. It is important to observe that the homomorphism groups, not the groups themselves, are localized for the definition of the category A p . Localization commutes with homomorphisms for finitely generated torsion-free groups but not for finite rank torsion-free groups in general. For example, if H is the subgroup of Q generated by {1/ p : p a prime}, then Z p is isomorphic to Hp for each prime p . However, Hom(H , Z) = 0, so that H is not isomorphic to Z in A p for any p. Proposition 2.1.3 (a) The category A p is an additive category. JfG is in A p and e is an idempotent in QEnd G, then G = e(G) EB (l - e)(G) in A p. (b) IfG is indecomposable in A p and (End G)p is semiperfect, then (End G)p is a local ring.
PROOF. The proof is analogous to that of Proposition 2.1.1, the only difference being that elements of Z p are those elements m/ n of Q with m , n E Z subject to the additional condition that n is prime to p. Corollary 2.1.4 Let G and H be objects of A p.
(a) The group G is isomorphic to H in A p ifand only ifthere is f E Hom(G , H) , g E Hom(H , G), and a nonzero integer n prime to p with fg = nIH and gf = nlG' (b) If f E Hom(G, H) , then f is an isomorphism in A p if and only if f is a monomorphism and H / f(G) is bounded by an integer prime to p. (c) The group G is a direct sum G 1 EB . . . EB G n in A p if and only if there is a monomorphism f from the group direct sum Gl EB ... EB G n into G such that G / image f is bounded by an integer prime to p.
2.1 Quasi-isomorphism and Isomorphism at p
51
(d) The group G is indecomposable in Ap ifand only ifwhenever G /(H $ K) is bounded by an integer prime to p, then H = 0 or K = O. (e) The group H is a summand ofG inA p ifandonlyifthere is f E Hom(G, H),
g E Hom(H, G), and a nonzero integer n prime to p with fg = nIH' (f) If(End G)p is semiperfect, then G is afinite direct sum ofindecomposables in A p. Moreover, such a decomposition is unique up to isomorphism in A p and the order ofthe summands.
PROOF.
Just as in Corollary 2.1.2.
Terminologyfor each of the conditions of Corollary 2.1.4 is as follows: (a) (b) (c) (d) (e) (f)
G is isomorphic at p to H .
f is an isomorphism at p. G is a direct sum of G I, . .. , Gnat p. G is indecomposable at p . H is a summand ofG at p. G has the Krull-Schmidt property in A p •
Despite the obvious similarities between the categories AQ and A p , there is one important distinction, namely that (End G)p need not be a local ring for each finite rank G that is indecomposable in A p • Moreover, the subcategory of Ap with objectsthose torsion-free abeliangroups of finiterank need notbe a Krull-Schmidt category by Example 2.1.11. Example 2.1.5 There is afinite rank torsion-free abelian group G and a prime p such that G is indecomposable in Ap but (End G)p is not a local ring.
Let Z[i] denote the ring of Gaussian integers with i 2 = -I. Then 5Z[i] = lvli, where It = (I + 2i)Z[i] and lz = (l - 2i)Z[i] are distinct maximal ideals of Z[i]. Now, Z[i]5 is a subring of Q[i], 5Z[ils is the Jacobson radical of Z[i]5, and Z[i]5/5Z[ils = Z[i]/5Z[i] is isomorphic to Z[i]/ It EEl Z[i]/ /z. In particular, Z[ils is not a local ring. By Theorem2.4.6, there is a finiterank torsion-freeabelian group G with End G isomorphicto Z[i]. Then (End G)s = Z[ils is not a local ring, but G is indecomposable in A5, since Z[ils has no idempotentsother than 0 or 1.
PROOF.
The next example demonstratesthat isomorphismis strongerthan isomorphism at p, in fact strongerthan isomorphismat p for each prime p . A torsion-freeabelian group G is completely decomposable if G is a finite direct sum of rank-1 groups and almost completely decomposable if G is quasi-isomorphic to a completely decomposable group. Equivalently, G contains a completelydecomposablegroup C as a subgroup with G/ C bounded. Example 2.1.6 There are rank-2 indecomposable, almost completely decomposable groups G and H such that G and H are isomorphic at p for each
52
2. Torsion-Free Abelian Groups
prime p but G and H are not isomorphic as groups. Moreover, G and Hare decomposable in A p for all but one prime p.
if],
if
Let X = Z[ the subring of Q generated by and Y = Z[ir]. Define G = X $ Y +Z(t, t) and H = X $ Y +Z(t, ~), subgroups ofQ$Q. Then both G and H are almost completely decomposable groups of rank 2. Since 11X = X and 31 Y = Y, it follows that Hom(X, Y) = Hom(Y , X) = O. Consequently, if f E End G, then f : X -+ X and f : Y -+ Y. Since X is a subring of Q, End X = X, where an element x of X is regarded as an endomorphism of X by multiplication. Similarly, End Y = Y. Restriction induces an isomorphism from End G to the ring Ro = {(x, y) E X x Y: x == y(mod5)} . To see this, let f E End G. Then f = (x , y) : X $ Y -+ X $ Y for some (x, y) in X $ Y. Moreover, (x , y)=5f(l, 1)=z(l , 1) + 5(x ', y') for some integer z, x' in X , and y' in Y . This shows that x == y(mod 5). Conversely, if (x, y) E X x Y with x == y(mod 5), then (x, y) induces, a well-defined endomorphism of G . It follows that G is indecomposable as a group, since 0 = (0, 0) and 1 = (l, 1) are the only idempotents of Ro. A similar argument shows that End H is isomorphic to R H = {(x, y) E X x Y: x == 2y(mod 5)}, whence H is also indecomposable. Since5G ~ X$Y ~ Gand5H ~ X$Y ~ H,both G and H are isomorphic at p to X $ Y for each prime p i= 5. As for p = 5, define f : G -+ H by f(x, y) = (x , 2y) and fUs '~) = (~, If) for x in X, yin Y, and z in Z. Then f is a well-defined group homomorphism with 2H ~ f(G) ~ H, and so G is isomorphic to H at p = 5 as well. Finally, assume that f : G -+ H is a group isomorphism. Then fx = f : X -+ X and Iv = f : Y -+ Y are automorphisms, hence units of X and Y, respectively. In particular, fx = 11; or -11 ; for some integer i and fy = 31 j or - 31 j for some integer j . It follows that f x and fy are congruent mod 5 to either 1 or -1 , since 11 == 31 == 1 (mod 5). But f is onto, so that (t,~) = f(x + ~,y + ~) for some x in X, Y in Y, and z in Z. Hence , 5x + z E X and 5 y + z E Y with fx(5x + z) = 5fx(x) + fx(z) = 1 and fy(5y + z) = 5fy(y) + fy(z) = 2. Therefore, zfx(l) == l(mod 5) and zfy(l) == 2(mod 5). However, there is no such integer z. since both fx and fy are congruent to either 1 or -1 mod 5. The conclusion is that G and H are not isomorphic . PROOF.
A subgroup H of a torsion-free abelian group G is a pure subgroup of G if nG n H = nH for each nonzero integer n. Equivalently, G/ H is a torsionfree group . It is easy to confirm that a summand of a torsion-free abelian group is a pure subgroup . A subgroup H of a torsion-free abelian group G is a ppure subgroup of G for a prime p if pG n H = p H , and G is p-reduced if pW(G) = nip; G : i ::: O} = o. Lemma 2.1.7 Let G and H be torsion-free abelian groups offinite rank.
(a) IfG and H are isomorphic in A p , then there are f E Hom(G, H) and g E Hom(H , G) such that fg - IH E pEnd Hand gf - 10 E pEnd G.
2.1 Quasi-isomorphism and Isomorphism at p
53
(b) IfG and Hare p-reduced andthere is f E Hom(G , H) and g E Hom(H, G) such that fg - I H E pEnd Hand gf - 10 E pEnd G, then G and Hare isomorphic in A p • PROOF. (a) Assume that G and H are isomorphic in A p . By Corollary 2.1.4(a), there are f E Hom(G, H), g E Hom(H, G), and a nonzero integer n prime to p with fg = nIH and gf = nlG . Write 1 = rn + sp for some r, s in Z. Then (rg)f = rn lG = (l-sp)lG and f(rg) = rnlH = (l-sp)lH . In particular, rg E Hom(H, G) with f(rg) - lH E pEnd Hand (rg)f - lG E pEnd G , as desired. (b) Suppose x E K = ker fg S; H. Then -x = fg(x) - x = (fg - 1)(x) = ph(x) for some h E End H . Thus, K S; pH n K = pK , since K is a pure subgroup of H . But H is p-reduced, so that K = kerfg = O. Hence, fg is a monomorphism. Moreover, image f g is p-pure in H . To see this , let x E H with px = f g(y) = y + ph(y) for some h E End Hand y E H. Then y = pa E pH, for a = x - hey) E H , and so px = fg(pa) = pfg(a). Consequently, x = fg(a) E image f g, because H is torsion-free. Since f g E End H is a monomorphism with p-pure image and H has finite rank , there is some integer n prime to p with nH S; f g(H) S; H , as an application of Corollary 2.1.2(b). Consequently, nH S; f(G) S; H . A similar argument shows that gf, hence f, is a monomorphism. By Corollary 2.1.4(b), G is isomorphic to
HinA p • Let frA denote the subcategory of A consisting of torsion-free abelian groups of finite rank and frA p the full subcategory of A p with objects those of frA. The criterion for isomorphism in fr A p given in Lemma 2.1.7 can be placed in a categorical setting. Define a category frAj p with objects p-reduced torsion-free groups of finite rank and morphism sets Hom(G, H)/ pHom(G , H). It can be readily verified that frA/ p is an additive category with direct sums in frAj p induced by direct sums in frA . In view of Lemma 2.1.7, if G and H are p -reduced, then G and H are isomorphic in frAp if and only ifthey are isomorphic in frA/ p . Notice that if G is in frAj p, then the endomorphism ring of Gin frAj p, End G j pEnd G, is a finite-dimensional algebra over the field Zj pZ. Thus, in contrast to frA p (Example 2.1.5), G is indecomposable in frAj p if and only if its endomorphism ring in frAj p is a local ring. In view of the preceding observations, frAj p would be a Krull-Schmidt category if idempotents split in frAj p, i.e., if e is an idempotent in End G j pEnd G, then there exist some f E Hom(G, H)j pHom(G, H) and g E Hom(H , G)j p Hom (H, G) with gf = e and fg = 1 E End HjpEnd H (see Exercise 1.2.6). In this case, H is a summand of G in frAj p that can be identified with e(G). However, idempotents need not split in frAj p, as demonstrated in Example 2.1.5. In that example, G is 5-reduced, (End G)sj5(End G)s has nontrivial idempotents, but G is indecomposable in frAs, hence in frAj5 by Lemma 2.1.7 . The following theorem shows how to embed frAj p in a Krull-Schmidt category obtained by inserting the "missing" summands resulting from the fact that idempotents need not split.
54
2. Torsion-Free AbelianGroups
Theorem 2.1.8 [Lady 75] There is a Krull-Schmidt category (frAj p)* and a fully faithful functor F : frAj p -+ (frAj p)*. The proof is outlined; details are left as an exercise. Define (frAj p)* to be the category with objects (e, G) such that e is an idempotent in End G j pEnd G. Morphisms are a :(e, G) -+ (j, H), where a E Hom(G, H) with a = fae . Routine arguments show that (frAj p)* is an additive category with finite direct sums given by (e, G) E9 (j, H) = (e E9 f, G E9 H). Moreover, idempotents split in (frAjp)*, so that (e, G) is indecomposable if and only if the endomorphism ring of (e , G) in (frAj p)* is a local ring . Then (frAj p)* is a Krull-Schmidt category. Finally, define F : frAj p -+ (frAj p)* by F(G) = (1, G) and conclude that F is fully faithful. In particular, F(G) is isomorphic to F(H) if and only if G is isomorphic to H. PROOF.
Even though frAj p is not a Krull-Schmidt category, as a consequence of Example 2.1.11 and Corollary 2.1.9 there are some uniqueness properties with regard to direct sums derived from the embedding of Theorem 2.1.8. Corollary 2.1.9 [Lady 75] Assume that G, H, and K arefinite rank, p-reduced, torsion -free abelian groups . (a) If G E9 K is isomorphic at p to H E9 K , then G is isomorphic to H at p. (b) If G" is isomorphic at p to H" for some positive integer n, then G is isomorphic to H at p. (a) The group G E9 K is isomorphic to H E9 K in frAj p by Lemma 2.1.7, hence in (frAj p)* by Theorem 2.1.8. Since the latter category is a Krull-Schmidt category, G is isomorphic to H in (frAj p)*, hence in frAj p and frA p. Assertion (b) is proved by an argument similar to that of (a), using the fact that (frAj p)* is a Krull-Schmidt category. PROOF.
There are some conditions on the endomorphism ring of a group G in frA sufficient to construct nonunique direct sum decompositions in frA p. Proposition 2.1.10 Suppose G is a strongly indecomposable torsion-free abelian group offinite rank, p is a prime, and there are right ideals II , .. . , In of End G with (End G)p = (It)p + ... + (In)p. Iffor each i there is a subgroup G i of G with Ii G ~ G i such that G is not isomorphic at p to G i r then H = G 1 E9 ... E9G n has a non unique direct sum decomposition in frA p. PROOF. To see that G is a summand of H at p , choose ri E I, and m.] k, E Zp with mi, k, E Z such that 1 = (mil kdrl + ... + (mnj kn)rn E (End G)p. Let k = k l . . . k n, an integerrelatively prime to p . Then k =m;rl + .. . +m~rn, where
2.1 Quasi-isomorphism and Isomorphism at p
55
m;=km;/kjEZ. Define f E Hom(G ,H) by f(x)=(m;rlx •. . . , m ~ rn x ) and g E Hom(H . G) by g(XI, . . . , x n) =XI + ... +Xn, observing that each m'r,x E IjG 5; Gi . Then gf =kl G , and so G is a summand of Hat p. The group G is strongly indecomposable, hence indecomposable at p. Moreover, G is not isomorphic to a summand at p of any G;, since rank G; ~ rank G is finite and G is not isomorphic at p to G i - This shows that H has a nonunique direct sum decomposition in frA p.
Example 2.1.11 (a) Let p ::: 5 be a prime and R
= Z[x]/(h(x»). where hex) = x(x + l)(x + 2) (x + 3) + p(x z + 2). Then R is a subring ofan algebraic numberfield, and there are ideals II and lz of R with R p = (It)p + (/z)p . (b) Choose afinite rank torsion-free abelian group G with End G = R. There are subgroups G; ofG with I;G 5; G; such that G is not isomorphic at p to G; for each i. (c) The group H = G] $ Gz has a nonunique decomposition into indecomposables in frA p.
PROOF. (a) Multiplying polynomials yields
hex) =x 4 + 6x 3 + (11 + p)x z + 6x +
2p, an irreducible polynomial in Q[x] by Eisenstein's criterion [Hungerford 74] . Thus, R is a subring of the algebraic number field QR = Q[x]/(h(x») with R/ pR a 4-dimensional Z/ pZ-vector space. Let a = x + (h(x») E R. Now, M] = a R + pR, Mz = (a + I)R + p R , M3 = (a +2)R + pR , and M4 = (a + 3)R + pR are distinct ideals of R with each (M;)p a maximal ideal of R p containing pR p. This is so because each Rp/(Mj)p = Zp/pZp = Z/pZ is a field. Define It = M]Mz and l: = M3M4 • Then (/])p + (/z)p = R p, as a consequence of the maximality of the (M;)p's. (b) By Theorem 2.4.6 , there is a finite rank torsion -free group G with End G = R. Let a = at + az, where a] E M]G\(MzG)p and az E MzG\(M] G)p. Such a choice is possible, since (MI)p and (Mz)p are distinct maximal ideals in R p. Consequently, na ¢. M] G U MzG for each integer n prime to p. Recall that It = M]Mz and Iz = M3 M4 . Define GI = It G + Za, so that It G ~ G] ~ G. Assume, by way of contradiction , that G is isomorphic at p to G b say f E End G = R with nG I ~ f(G) ~ G I forsomeintegern relatively prime to p. Then f ¢. M I UMz;otherwise,na E M I GU MzG, which is impossible by the choice of a . On the other hand, (/])p 5; Rpf. This is so because j'v'nr.t G) 5; G ,whencef-lnIt 5;EndG = RandnII 5; Rf with gcd{n, p} = 1. Since (M])p and (Mz)p are the only maximal ideals of R p containing (It)p, f is a unit in R p and Rpf = R p. Thus , (G])p = G p. But Z/ pZ is isomorphic to (G])p/(/]G)p = Gp/(ItG)p and Gp/(ItG)p is isomorphic to G p/(M] G)p $ G p/(MzG)p, a contradiction. Similarly, using the ideals M3 and M4 , there is a subgroup G z of G containing lzG such that Gz is not isomorphic at pto G.
Assertion (c) is a consequence of Proposition 2.1.10.
56
2. Torsion-Free Abelian Groups
EXERCISES I. Assume that G is a torsion-free abelian group of finite rank. Prove that if H is a subgroup with G/ H bounded , then G/ H is finite. 2. Assume that G and H are torsion -free abelian groups of finite rank and p is a prime . Prove that G is isomorphic at p to H if and only if G is a summand of H at p and H is a summand of Gat p . 3. A torsion-free abelian group G is p-local for a prime p if qG = G for each prime q =1= p . (a) Show that if G is a torsion-free abelian group , then G p is a p-local group. Prove that G is p-Iocal if and only if G is a torsion -free Zp-module. Assume that G and Hare p-reduced p-local groups of finite rank. (b) Show that if rank G = I , then G is isomorphic to Q or 7lp . (c) Prove that G and H are isomorphic if and only if they are isomorphic at p. (d) Prove that G is isomorphic to H if either G" is isomorphic to H" for some n ~ I or else there is a finite rank torsion-free abelian group K with G ED K isomorphic to
H EDK . (e) Find a p -local group of finite rank that does not have the Krull-Schmidt property. 4. Complete the details of the proof of Theorem 2.1.8.
if 1 t-l
!) .
5. Let G = 7l[ ED 7l[ + 7l(! , Use the techniques of the proof of Example 2.1.6 to find all groups H that are isomorphic at 5 to G and all groups H isomorphic to G ED zr that have a completely decomposable subgroup 7l[
if 1 t-].
2.2 Near-isomorphism of Finite Rank Groups The category frA p is still not completely satisfactory from a group-theoretic point of view, since indecomposables in frA need not be indecomposable in frA p (Example 2.1.6). Also, in order to fully exploit the category frA/ p there is an annoying assumption that groups in frA p be p-reduced. Both of these difficulties can be overcome by introducing an equivalence relation called nearisomorphism, for which isomorphism at p is a local version. In fact, a torsionfree abelian group G of finite rank is indecomposable if and only if it is indecomposable with respect to near-isomorphism (Corollary 2.2.11). Consequently, G is indecomposable if and only if it is indecomposable at p for each prime p. Two torsion-free abelian groups G and H are nearly isomorphic if G and H are isomorphic at p for each prime p. Moreover, G is a near summand of H if G is a summand of H at p for each prime p . A torsion-free abelian group G is divisible if nG = G for each nonzero integer n. A divisible group is a Ql-vector space, hence, as a group, a direct sum of copies of Ql. A torsion-free abelian group G has a unique maximal divisible subgroup d(G) = n{nG : 1 ::: n E Z} . Then d(G) is a summand of G, being injective as an abelian group. If G = d( G) $ H, then H is reduced, that is, d(H) = O.
2.2 Near-isomorphism of Finite Rank Groups
57
Lemma 2.2.1 If G is a reduced torsion-free abelian group offinite rank, then there is a nonzero square-free integer k such that k Y i= Y for each nonzero subgroup Y ofG. PROOF. Let Il = {PI, , Pn, . ..} be an indexing of the set of primes . Then pi(G) ;2 pi(piG) ;2 is a descending chain of pure subgroups of G. Since G is reduced with finite rank, there is some i with pi piG· .. pfG = O. Let k = PI' " Pi. IfkY = Y, then Y £; pipiG .. · pfG = O. The next theorem is a globalization of the definition of near-isomorphism. Theorem 2.2.2 Suppose G and H are reduced torsion-free abelian groups of finite rank. (a) The groups G and H are nearly isomorphic if and only iffor each nonzero integer n, there is f E Hom(G, H), g E Hom(H, G), and an integer m relatively prime to n with fg = mlH and gf = mIG· (b) The group G is a near summand of H if and only if for each nonzero integer n' , there exist some f E Hom(G, H), g E Hom(G, H), and integer m relatively prime to n' with gf = mIG.
PROOF. It is sufficient to prove (b). In fact, if G and H are nearly isomorphic, then G is a near summand of Hand H is a near summand of G. Conversely, if G and H have finite rank, G is a near summand of H, and H is a near summand of G, then G and H are nearly isomorphic. This follows from the observation that if G is a summand of Hat p and H is a summand of Gat p, then G and H are isomorphic at p (Exercise 2.1.2). (<=) To see that G is a near summand of H, let p be a prime . Choosing n' = p shows that G is a summand of H at p, whence G is a near summand of H. (::::}) Assume that G is a near summand of H and let p be a prime. Since G is a summand of Hat p, there exist f E Hom(G, H), g E Hom(H, G), and an integer m prime to p with gf = mIG by Corollary 2.1.4(e) . As in Lemma 2. 1.7(a), there are fp E Hom(G, H) and gp E Hom(H, G) with gpfp - IG E pEnd G. More generally, for each square-free integer n, there exist fn E Hom(G, H) and gn E Hom(G, H) with s-J« - I G E nEnd G. The proof is by induction on n. The case that n is prime is confirmed in the previous paragraph. Now assume that n is not prime and write n = mk for relatively prime square-free integers m < nand k < n. By the induction hypothesis, there are fb fm E Hom(G, H) and gb gm E Hom(H, G) with gmfm - IG EmEnd G and gdk - IG E kEnd G. Write 1 = rm + sk for integers rand s and define fn = skfm + rmfi. and B« = skg m -s-rmgi, A straightforward computation shows that gnfn -IG E nEnd G, as desired. By Lemma 2.2.1, there is a square-free integer k such that if Y is a subgroup of H with k Y = Y, then k = O. To complete the proof, let n' be an integer and n the square-free integer consisting of the product of the distinct prime divisors of
58
2. Torsion-Free Abelian Groups
n'k. Notice that k, being square-free, divides n. By the preceding paragraph, there exist some fn E Hom(G, H) and S« E Hom(H, G) with gnfn - IG E nEnd G. Then n(ker gnfn) = ker gnfn, since (gnfn - l)(x) = gnfn(x) - x = nh(x) E nG for some h E End G and ker gn fn is a pure subgroup of G. Since k divides n, k(ker gnfn) = ker gnfn, so that ker gnfn = 0 by the choice of k. Thus, gnfn is a monomorphism. Also, image gnfn is q-pure for each prime q dividing n, again since gnfn(x) - x EnG. Since gn fn is a monomorphism and G has finite rank, there is some integer m prime to n with mG S; gnfn(G) S; G by Corollary 2.1.2 . Finally, h; = (gnfn)-lm E Hom(G, G), hng n E Hom(H, G), fn E Hom(G, H), and (hngn)fn = m lG. But m is also relatively prime to n', by the choice of n, so that the proof is complete.
Corollary 2.2.3 Let G and H be reduced torsion-free abelian groups offinite rank. Then G is nearly isomorphic to H frAj p for each p.
if and only if G is isomorphic to
H in
PROOF. Apply Lemma 2. 1.7(a) and the proof of Theorem 2.2.2, wherein it is deduced that G is nearly isomorphic to H from the assumption that for each prime p, there are fp E Hom(G, H) and gp E Hom(H, G) with gpfp - lG E pEnd G and fpgp - lH E pEnd H .
Corollary 2.2.4 Suppose G, H, and K are finite rank torsion-free abelian groups. (a) IfG E9 K is nearly isomorphic to H E9 K, then G is nearly isomorphic to H . (b) If G" is nearly isomorphic to H" for some positive integer n, then G is nearly isomorphic to H . PROOF. It is sufficient to assume that G, H, and K are reduced groups, since if A and B are torsion-free abelian groups of finite rank, then A and B are nearly isomorphic if and only if Ajd(A) and Bjd(B) are nearly isomorphic (Exercise 2.2.5). Now apply Corollary 2.2.3 and Theorem 2.1.8. A torsion-free abelian group is semilocal if pG = G for all but a finite number of primes p. For semi local groups, isomorphism and near-isomorphism are equivalent :
Corollary 2.2.5 Let G, H, and K be finite rank torsion-free abelian groups. If G and Hare semilocal, then: (a) G is nearly isomorphic to H if and only if G is isomorphic to H. (b) G is a near summand of H if and only if G is a summand of H. (c) G has no nontrivial near summands if and only if G is indecomposable as a group.
2.2 Near-isomorphism of Finite RankGroups
59
(d) IfG EB K is isomorph ic to H EB K , then G is isomorphic to H . (e) lf G" is isomorphic to H" for som e positive integer n, then G is isomorphic to H. It is sufficient to prove (b). Then (a) follows, since if G is a summand of Hand H is a summand of G , then G and H are isomorph ic. Moreover, (c) is an immediate consequence of (b), and (d) and (e) follow from (a) and Corollary 2.2.4. (b) Assume that G is a near summand of H and let n be the product of those primes p such that pG # G. Since G is sernilocal, n is an integer. By Theorem 2.2.2(b), there is an integer m prime to n , f E Hom(G , H ), and g E Hom(H , G) with gf = mIG ' Then m is a unit of the endomorphism ring of G by the choice of n. Hence , m:' g E Hom (H , G) with (m- 1g)f = IG. This shows that G is a summand of H . Conversely, a summand is clearly a near summand.
PROOF.
The next corollary provides some ju stification for the term near summand. Corollary 2.2.6 Suppose G , H , and K are torsion-free abelian groups offinite rank. If G is a near summand of both Hand K, then G is a group summand of H EB K. Let p be a prime and choose, via Theorem 2.2.2(b), an integer m prime to p and f p E Hom(G , H ), gp E Hom(H , G ) with gpfp = mIG . Another application of Theorem 2.2.2(b) yields an integer n prime to m and f m E Hom(G , K ), gm E Hom (K , G ) with gmfm = nl G. Write 1 = rm + snfor some integersrands.Then f EHom(G , HEBK ), where f (x ) = (fp(x), fm(x )),and g EHom(HEBK , G)with g(x , y) = rg p(x )+sgm(y ). A simple computation shows that gf = IG,asdesired.
PROO F.
Let G be a torsion-free abelian group of finite rank. As noted in Section 2.1, QEnd G is a finite-dimensional Q -algebra, l QEnd G is nilpotent, and QEnd G/ l QEnd G is a sernisimple Artinian Q-algebra. Define NEnd G = (lQEnd G) n End G. Then NEnd G is a nilpotent ideal of End G , and End G / NEnd G is contained in QEnd G/1QEnd G = K, x ... x K n , where each K, = Matn(D j ) is a full matrix ring over a division Q-algebra Di, Following is a brief summary of basic definitions and properties of orders . Further details and proofs can be found in [Arnold 82] or [Reiner 75] . Let K = Matn(D), D a finite-dimensional division algebra over Q , and let F be the center of K . Then F is the center of D , hence an algebraic number field. Let S be a subring of F . A subring R of K with Q R = K is called an S-order in K if R is an S-algebra that is finitely generated as an S-module. Then R is a maximal S-order if R = R' for each S-order R' with R ~ R' C K . For example , if S is a principal ideal doma in with quotient field F, then Matn(S) is a maximal S-order in Matn(F ) [Arnold 82, Corollary 10.3]. Lemma 2.2.7 Assume that G is a torsion-free abelian group offinite rank and QEnd H/1 QEnd H = K ] x . . . x K; with each K, = Matn(D j ) afull matrix
60
2. Torsion-Free Abelian Groups
ring over a division Q -algebra D;. There is a smallest positive integer n( G) with n(G)R* ~ End G j NEnd G ~ R* = R 1 X • . • x Rn and each R; a maximal Si-order in K; for some Dedekind subring S; ofthe center of D;. PROOF. [Arnold 82, Corollary 10.14] .
For some groups G, the integer n(G) is easy to compute. For example, if X is a rank-I group and G = X" is the direct sum of n copies of X, then End G = Matn(End X) with End X a subring of Q, QEnd G Matn(Q), and JQEnd G = NEnd G = O. In this case, n(G) = 1, since S = End X is a principal ideal domain by Exercise 2.2.6. More generally, it is proved in Example 5.1.7 that if G is a torsion-free abelian group of finite rank and C is a fully invariant completely decomposable subgroup of G with nG ~ C ~ G for some least positive integer n , then n = n(G). With the notation of Lemma 2.2.7, R = End G j NEnd G is a subring of R* and n(G)R* ~ R ~ R*. An R-module M is an R-lattice if M is a finitely generated R-module with additive group a torsion-free abelian group, necessarily of finite rank. Examples of R-Iattices include finitely generated projective R-modules. Two R-Iattices M and N are in the same genus class if M p and N p are isomorphic R p modules for each prime p of Z, recalling that M p Z p (8)z M .
=
=
Lemma 2.2.8 Assume that G is a torsion-free abelian group offinite rank and n( G)R* ~ R = End G j NEnd G ~ R* as given in Lemma 2.2.7. (a) If M is an R-lattice, then R*M is projective as an R*-module. (b) Two R-lattices M and N are in the same genus class if and only if either n(G) = 1 and QM is isomorphic to QN as QR-modules or else n(G) > 1 and M p is isomorphic to N p as R p -modules for each prime divisor p ofn( G). PROOF. Assertion (a) follows from the observation that R* is a finite product of
maximal orders R;, R;M is an R;-Iattice, and lattices over maximal orders are projective [Arnold 82, Theorem 11.3]. (b) [Arnold 82, Corollary 12.4]. The integer n(G) provides a single integer to test for near isomorphism. Proofs of the next corollary and proposition use the functorial correspondence K ~ Hom(G, K)jHom(G, K)NEnd G from the category of summands of finite direct sums of copies of G to the category of finitely generated projective modules over End G j NEnd G given in Theorem 2.4.3. Briefly, this correspondence identifies near isomorphism classes of groups with genus classes of R-Iattices.
Corollary 2.2.9 Let G be a torsion-free group offinite rank. A torsion-free abelian group K of finite rank is nearly isomorphic to G if and only if K is a summand of G E9 G and either n(G) = 1 and G is quasi-isomorphic to K or else n(G) > 1 and G is isomorphic at p to K for each prime p dividing n(G).
2.2 Near-isomorphism of Finite RankGroups
61
PROOF. Assume that K is nearly isomorphic to G . Then K is quasi-isomorphic to G and isomorphic at p to G for each p. Moreover, K is a summand of G ED G by Corollary 2.2.6. Conversely, suppose either n(G) = 1 or else n(G) > I and G is isomorphic to K at p for each prime p dividing n. Let R = End G/ N End G. Then Rand M = Hom( G, K)/Hom( G, K)NEnd G are finitely generated projective R -modules. By Theorem 2.4.3(a) and (c), either n(G) = 1 and QM is QR-isomorphic to QR or else n(G) > 1 and M pis Rp-isomorphic to R p for each prime divisor p of n(G). Hence, Rand M are in the same genus class as an application of Lemma 2.2.8(b). By Theorem 2.4.3(d), G and K are nearly isomorphic. The next technical proposition is the key to confirming in Corollary 2.2.11(b) that a finite rank group H is indecomposable if and only if it is indecomposable at p for each prime p.
Proposition 2.2.10 Let G be a finite rank torsion-free group and n(G) as given in Lemma 2.2.7.
(a) If K is a near summand of G and g E Hom(G, K), then there is h E Hom(image g, G) with gh = n(G)limage g . (b) If X and K are near summands of G and there is an integer m prime to n(G) with mK ~ X ~ K, then X is nearly isomorphic to K. PROOF. (a) Suppose K is a near summand of G , a summand of G ED G by Corollary 2.2.6, and that g E Hom(G, K). Let R = End G / NEnd G and M = Hom(G, K)/ Hom(G, K)NEnd G . Then Rand M are finitely generated projective R- modules. Since K ~ M is a functor , there is an R-homomorphism g': R -+ M induced by g'(f) = fg for each f E Hom(G,K) (Theorem 2.4.3). But n(G)R* ~ R ~ R* and R* is torsion-free, so that g' extends to an R*- homomorphism g* = O/n(G))g' :R* -+ R*M . Now, image g* is an R*-lattice, hence a projective R*-module by Lemma 2.2.8(a) . Thus , there is an R*-homomorphism h*: image g* -+ R* with g*h* = limage g" Moreover, h' = n(G)h* : image g' -+ R is an R-homomorphism with g'h' = n(G)limage s' : It follows from Theorem 2.4.3 that there is a homomorphism h : image g -+ G with gh = n(G)l image s (b) In view of Corollary 2.2.6, both X and K are summands of G ED G. Then M = Hom(G , X)/Hom(G, X)NEnd G and N = Hom(G, K)/Hom(G, K)N End G are finitely generated projective R-modules, where R = End G/NEnd G. As a consequence of the hypothesis, there is an integer m prime to n(G) with mN ~ M ~ N. Hence, mNp = M p = N p for each prime p dividing n(G). By Lemma 2.2.8(b), M and N are in the same genus class. Consequently, X and K are nearly isomorphic by Theorem 2.4.3(d).
Corollary 2.2.11 Assume that G and H are torsion-free abelian groups offinite rank.
(a) The group H is a near summand of G ifand only if G groups X and X' with X nearly isomorphic to H .
= X ED X' for some
62
2. Torsion-Free Abelian Groups
(b) The group G is indecomposable summands of G.
if and
only
if 0 and G
are the only near
PROOF. (a) Assume that H is a near summand of G and let neG) be as given in Lemma 2.2.7. There are f E Hom(H , G) , g E Hom(G, H) , and an integer m prime to neG) with gf = mlH' Write I = rm + sn (G ) for integers rand sand let X = image g 5; H. By Proposition 2.2.l0(a), there is h E Hom(X, G) with gh = n(G)l x. Define ¢ E Hom(X, G) by ¢(x) = r f(x) + sh (x ). Then g¢ = rgf+sgh=rm+sn(G) = lx, whence X is a summand of G. Furthermore, mH = gf(H) 5; g(G) = X 5; H, so that X is nearly isomorphic to H by Proposition 2.2.1O(b). Assertion (b) is a consequence of (a).
EXERCISES The groups in Exercises 1-5 are assumed to be torsion-free abelian of finite rank.
= G I EEl Gz with G, nearly isomorphic to Hi . 2. Show thatif G is nearly isomorphic to H. then G is indecomposable if and only if H is indecomposable. 3. Prove that if G is nearly isomorphic to H. then G EEl G' is isomorphic to H EEl H for some G' nearly isomorphic to G. 4. Prove that if G is nearly isomorphic to H" , then G is isomorphic to H n - 1 EEl G' for some G' nearly isomorphic to H . 5. (a) Prove that d(GEEl H ) = d(G) EEl d(H ). (b) Prove that G is nearly isomorphic to H if andonly if G/ d (G ) is nearly isomorphic to H/d(H). 6. Show thatif R is a subring of therationals Q, then R is a principal ideal domain. 1. Prove thatif G is nearly isomorphic to HI EEl Hz, then G
2.3 Stable Range Conditions for Finite Rank Groups The endomorphism ring of a torsion-free abelian group of finite rank has 2 in the stable range (Theorem 2.3.6). As a consequence, two torsion-free abelian groups G and H of finiterank are nearly isomorphic if and only if for some positive integer n, the direct sum of n copies of G is isomorphic to the direct sum of n copies of H (Theorem 2.3.12). There is a deceptively easy, but false, argument that if A, H , and K are abelian groups with AEEl H isomorphic to AEEl K, then H is isomorphic to K . The argument is that H is isomorphic to (AEElH)/ A, (AEElK)/ A is isomorphic to K, (AEElH)/ A is isomorphic to (A EEl K)/ A, and so H is isomorphic to K. The flaw in this argument, which is correct if AEEl H is actually equal to AEEl K , is a confusion between equality and isomorphism. This distinction motivates the notion of substitution.
2.3 StableRangeConditions for Finite Rank Groups
63
An abelian group A has the substitution property if given G = A I $ H = A 2 $ K with A isomorphic to both AI and A2, then there is a subgroup C of G with G = C $ H = C $ K. In this case, Hand K are isomorphic. The group A has I-substitution ifwheneverG = A61B and f : G ~ A, g : A ~ G with fg = lA , then there is ¢: A ~ G with f¢ = I A and G = ¢(A) $ B. A ring R has I in the stable range iffor ftgl + hg2 = I with f;, gi E R, there is some hER with fl + hh a unit of R. Proposition 2.3.1 [Warfield 80], [Fuchs 70B] The following statements are
equivalent for an abelian group A: (a) A has I-substitution ; (b) End A has I in the stable range; (c) A has the substitution property.
(a):::} (b) Assume I.s, + hg2 = I with f;, gi E End A. Define G = A $ A, f = (fJ , h) : G ~ A, and g = (gl, g2): A ~ G, where f and g are defined by f(al ' a2) = ft (aJ) + h(a2) and g(a) = (gJ(a), g2(a» for a, a, E A. By (a), there is ¢ = (¢I, (2): A ~ G with ¢i E End A such that f¢ = fl¢l + h¢2 = I A and G = ¢(A) $ A. Since G = A $ A, there is an isomorphism hi : A ~ A with l I A = (h h 0)( ¢h (2) = hi ¢I. Hence, ¢I is an isomorphism and fl + h( ¢2¢i ) = J ¢i is a unit of End A, as desired. (b):::} (c) Suppose G = A61H = A 2 $ K with y : A2 ~ A an isomorphism. Let f: G ~ A be a projection of G onto A2 (with kernel K) followed by y and write f = (fl,ex). Also, letg = y-I :A ~ A2 ~ G = A $ Hand writeg = (gJ ,f3). Then I A = yy-I = fg = flgl + (exf3)I A . By (b), there is a unit u of End A and h E End A with flU + (exf3)h = I A and e = (u, f3h) : A ~ G = A $ H with I A = flU + exf3h = [B, Therefore, G = e(A) $ kerf = e(A) $ K, recalling that ker f = K . Moreover, ¢ = (u- I , 0) : G = AffiH ~ A with ker e = Hand lA = ¢e = (u- I , O)(u , f3h). Hence, if e(A) = C , then G = C $ ker ¢ = C ffi H. Since
PROOF.
G = C $ K, the proof is complete. (c):::} (a) Given G = A $ B, f : G ~ A, and g : A ~ G with fg = lA, then G = g(A) $ ker f with g(A) isomorphic to A. By (c), G = C $ B = C61 ker f for some subgroup C of G. Hence, there is an isomorphism ¢ : A ~ C. Now, f¢(A) = f(C) = fCC $ ker f) = f(G) = A, so that f¢ is a unit of End A. Then, ¢' = ¢(f¢)-l: A ~ G with f¢' = I A and G = C $ B = ¢(A)61 B = ¢'(A) $ B, as desired. Example 2.3.2 (a) [Bass 68 ] If R is a ring and R/JR is Artinian, then I is in the stable range
ofR. (b) I is not in the stable range of'll. PROOF.
Ii, St
E
(a) It is sufficient to assume JR = O. To see this, let fl gl + hg2 = I with R. Then I.s, + hg2 == I (mod JR), i.e., the coset flgl + hg2 + JR is
64
2. Torsion-Free AbelianGroups
equal to I +JR, and J(Rj J(R» = 0. If there are h, u E R with fl + [zh == u(mod JR ) and u a unit modulo JR , then fl + hh = u + x for some x in JR , and there is v E R with I - (u + x)v E JR. Hence, (u + x)v = I - (l - (u + x)v) is a unit of R. Thus, u + x is a unit of R with fl + hh = u + x , as desired. Now assume that R is Artinian with JR = 0, i.e., R is a semisimple Artini an ring. Then the right ideal fI R is a summand of R [Hungerford 74], and so R = fl REB I for some right ideal I ~ hR . Define ¢: R -+ fIR by ¢(r) = Iv . Since R is semisimple Artinian, ker ¢ is a summand of R, say 0: R -+ ker ¢ with O(x) = x for each x E ker ¢. Hence, R fl REB I is isomorphic to fl REB ker ¢ . By the classical Krull-Schmidt theorem for decompositions of finitely generated modules over semisimple Artinian rings [Anderson, Fuller 74], there is an isomorphism a: ker ¢ -+ I . Now,(¢,O) :R -+ fIREBker¢ , (l ,a):fIREBker¢ -+ R flREBI ,andthe composite u = (I, a)(¢ , 0) is an R-automorphism of R as aright R-module. Thus, u is multiplication by some unit u(l) = (fl, x) = fl + x of R with x E I ~ hR. Write x = hh for some hER so that fl + hh = u(l) is a unit of R. This shows that I is in the stable range of R. (b) Simply observe that 3.5 + 7( -2) = I but 3 + 7h is not a unit of Z for any h in Z.
=
=
Lemma 2.3.3 If A is a tors ion-free abelian group offinite rank and f lgl + hs: is a monomorphism for Ii, gi E End A, then there is some h E End A with fl + hh a monomorphism. PROOF. As a consequence of Corollary 2.1.2(b), f E End A is a monomorphism if and only if f is a unit in QEnd A. Since QEnd A is Artinian, this is equi valent to f being a unit modulo JQEnd A. Moreover, f is a unit modulo J QEnd A if and only if f is not a left zero divisor modulo JQEnd A, since QEnd Aj J QEnd A is semisimple Artinian. It is now sufficient to assume that JQEnd A = and Iss, + hs: is a unit of QEnd A and prove that fl + hh is not a left zero divisor in QEnd A. As in Example QEnd A EB I for some right ideal I ~ hQEnd A. Write 2.3.2 , QEnd A I = fl + hh' for some h' = (ljm)h E QEndA with f. mE Z and h E End A . Then m = m], + [zh - However, fl + hh is not a left zero divisor ofQEnd A. This is so because if (fl + hh)r = 0, then = fl r + hhr E fI QEndA EB I = QEndA. Hence, fIr = hhr = 0, and so mr = (mfl + hh)r = 0. Since Of. m, r = 0. Consequently, fl + hh is not a left zero divisor.
°
= t.
°
°
Theorem 2.3.1 provides conditions under which a group A can be canceled from a direct sum. There is a weaker condition that allows a partial cancellation. A group A has the 2-substitution property if given G = A EB A EB Band f : G -+ A , g : A -+ G with fg = lA , then there is ¢: A -+ G with f¢ = I A and G = ¢(A) EB C EB B for some C ~ A EB A. In this case, A EB A is isomorphic to ¢(A) EB C with ¢(A) isomorphic to A.
2.3 Stable Range Conditions for Finite Rank Groups
65
Lemma 2.3.4 If an abelian group A has the 2-substitution property and A EB A EB X is isomorphic to A EB Y for abelian groups X and Y, then Y is isomorphic to A' EB X for some A' with A EB A' isomorphic to A EB A. Let G = A) EB A 2 EB X = A EB Y with A isomorphicto Ai for each i and let f: G ~ A be a projectionwithker f = Y. Since A has the 2-substitutionproperty, there is ¢: A ~ G with f¢ = lA and G = A) EB A 2 EB X = ¢(A) EB A' EB X for some A' ~ A EB A. Also, G = ¢(A) EB kerf = ¢(A) EB Y as f¢ = lAo Thus, Y can be shown to be isomorphic to A' EB X, by canceling ¢(A). Furthermore, G = A I EB A 2 EB X = ¢(A) EB A' EB X, so that A I EB A2 is isomorphicto ¢(A) EB A' , shown by canceling X. But ¢ is a monomorphism, so that ¢(A) is isomorphic to A . Since A is isomorphic to Ai for i = 1,2, A EB A is isomorphicto A EB A' with Y isomorphic to A' EB X, as desired.
PROOF.
Given a ring R , 2 is in the stable range of R if fIg) + hg2 + hg3 = I with fi , gi E R implies there is hi, k i E R with I = (ft + hh)k) + (12 + f3h 2)k2. Theorem 2.3.5 [Warfield 80] An abelian group A has the 2-substitution property ifand only if2 is in the stable range of End A. A has the 2-substitution property and let Iss, + hg2 + hg3 = lA with fi, gi E End A. Define G = A EB A EB A, f = (ft, h , 13): G ~ A,and g = (g),g2 ,g3):A ~ G, so that fg = ftg) + hg2 + hg3 = lA' Since A has the 2-substitution property, there is ¢ = (h), h 2, h3): A ~ G with lA = f¢ = f)h) + hh 2 + f3h 3 and G = ¢(A)EBC EBA for some C ~ A EBA EB O. Canceling A from G givesB = (k) , k2, 0): G ~ AwithB¢ = k)h)+k 2h2 = lA. Therefore,
PROOF. (::::}) Assume that
(f)
+ h(h 3kd)h) + (12 + h(h 3 k2»h 2 = fth) + hh2 + hh3(k)h) + k2h2) = f)h) + [zha + hh 3 = lA'
This shows that 2 is in the stable range of End A. ({::) Let G = A EB A EB B,f = (f) , h , Ol) : G ~ A, and g = (gl, g2, ,B) : A ~ G with lA = fg = ftg) + hg2 + (Ol,B)lA. Since 2 is in the stable range of End A, lA = (f) + (Ol,B)h)k) + (12 + (0l,B)h 2)k2 for some hi, ki E End A. Define ¢ = (k), k2, ,B(h)k) + h 2k2»: A ~ G, so that f¢ = f)k) + hk2 + Ol,B(h)k) + h2k2) = (f) + (Ol,B)h)k) + (12 + (0l,B)h2)k2 = lA. Then B = (f) + (Ol,B)h) , 12 + (0l,B)h 2, 0) : G = A EB A EB B ~ A with B¢ = (f) + (Ol,B)h»k) + (12 + (0l,B)h 2)k2 = lAoIn particular, G = ¢(A) EB ker B. But G = A EB A EB Band B ~ ker B, so that ker B = B EB (ker B) n (A EB A) with C = (ker B) n (A EB A) ~ A EB A. Thus, G = ¢(A) EB B EB C with C ~ A EB A . Theorem 2.3.6 [Warfield 80] IfG is a torsion-free abelian group offinite rank, then 2 is in the stable range of End G.
66
2. Torsion-FreeAbelian Groups
PROOF. Assume fl gl + hs: + hg3 = I with Ii, gj E End G. It is sufficient to assume that fl is a unitofQEnd G. To see this, there is, by Lemma 2.3.3 , anh in End G such that !I' = fl + (hg2 + hg3)h is a unit of QEnd G. Now,
iI'gl + h(g2 -
g2hg)) + h(g3 - g3hg))
+ (hg2 + hg3)h)gl + h(g2 = flgl + hs: + hg3 = I. = (fl
Hence, if (!I' + hhdk l + (12 + hh 2)k2 substituting for fl' gives an equation IG
=
g2hgl) + h(g3 - g3hg))
I for some hj , k, E End G, then
= (fl + (hg2 + hg3)h + hhl)k l + (12 + hh 2)k2 = (fl
+ h(g3 h + hi
- h 2g2h))k l
+ (12 + hh 2)(g2hkl + k2)
of the desired form to show that 2 is in the stable range of End G. So, assume that fI is a unit in QEnd G. There is a positive integer n with n EndG ~ flEndG ~EndG.ByExample2.3 .2, I isinthestablerangeofEndG/ nEnd G, since End G/nEnd G is finite. Thus, hg2 + (flgl + hg3)1 == l(mod nEnd G) yields an h E End G such that 12 + (flgl + hg3)h == u(mod nEnd G) with , u a unit modulo nEnd G. Since nEnd G S;; fIEnd Gs ]z + hg3h == u(mod fIEnd G). Because u is a unit modulo nEnd G, there is v E End G with uv - I E nEnd G. Then (12 + hg3h)V == I(mod fIEnd G), since nEnd G S;; fIEnd G . Write 1 = [ik: + (12 + hg3~)V for some k l E End G . This yields (fl + hO)k l + (12 + h(g3h))V = 1, whence 2 is in the stable range of EndG.
Corollary 2.3.7 Suppose G is a torsion-free abelian group offinite rank. (a) The group G has the 2-substitution property. (b) If G EB G EB X is isomorphic to G EB Y for finite rank torsion-free abelian groups X and Y, then Y is isomorphic to G' EB X for some G' with G EB G' isomorphic to G EB G. PROOF. Apply Lemma 2.3.4 and Theorems 2.3.5 and 2.3.6.
It is known that if G EB G' is isomorphic to G EB G, then G is isomorphic to G' for "most" finite rank torsion-free abelian groups G . A finite-dimensional simple Q-algebra K is a totally definite quaternion algebra if K is a division algebra with center F, K has F -dimension 4, and for each archimedean valuation v of F, the vcompletion F; of F is the reals and Fv@F K is the ring of Hamiltonian quatemions.
Proposition 2.3.8 Suppose G is a torsion-free abelian group offinite rank and write QEnd G/JQEnd G = K I X . •• x K n , with each K, a matrix ring over a division algebra Di. Ifno K, is a totally definite quaternion algebra and G EB G' is isomorphic to G EB G, then G is isomorphic to G' .
2.3
Stable Range Conditions for Finite Rank Groups
67
PROOF. [Arnold 82] .
It follows from Theorem 3.1.8 that if a torsion-free abelian group G is quasiisomorphic to a finite direct sum of rank-l groups, then G satisfies the hypotheses of the next example. Example 2.3.9 Let G be a finite rank torsion -free abelian group and write Q End G / lQEnd G = Matn(l)(Dd x . .. x Matn(m )(D m) with each D; a division algebra and, for each i, either n(i) > 1 or else n(i) = I and D; is a field. If G EB G EB X is isomorphic to G EB Y for finite rank torsion-fre e abelian groups X and Y, then Y is isomorphic to G EB X. PROOF. A consequence of Corollary 2.3.7(b) and Proposition 2.3.8.
Lemma 2.3.10 JfG is a torsion-free abelian group offinite rank, then there are only finitely many isomorphism classes of summands of G. PROOF. A consequence of the Jordan-Zassenhaus theorem for Z-orders ([Lady 74A] or see [Arnold 82]).
Lemma 2.3.11 Assume that G is a torsion-free abelian group offin ite rank. There is a positive integer n such that if G is nearly isomorphic to H, then G" EB C is isomorphic to H" EB C with C a summand of G'" for some m ?: 1. PROOF. Define the Grothendieck group of G to be Ko(G) = F/ R , where F is the free abelian group with a basis consisting of isomorphism classes (A ) of groups A such that A is a summand of for some m ?: I and R the subgroup of F generated by elements of the form (A EB B) - (A) - ( B) . Define [A] to be the coset (A) + R. Then [A EB B] = [A] EB [B] and n[A] = [An] for a positive integer n. It follows that each element of Ko(G) is of the form x = [A] - [B] and that x = 0 if and only if there is some C with A EB C isomorphic to B EB C and C a summand of G'" for some m. The latter condition can be seen by writing (A) (B) as a linear combination of the generators of R, rearranging this equation so that only positive coefficients are on each side of the equation, and using the fact that F is a free abelian group. Let U be the set of elements of Ko(G) of the form [H] - [G] such that H is nearly isomorphic to G . To see that U is a subgroup, it suffices to show that U is closed under sums and additive inverses. If [H] - [G] and [K] - [G] are elements of U, then by Corollary 2.2.6, G EB L = H EB K for some group L. But L is nearly isomorphic to G by Corollary 2.2.4 , since G is nearly isomorphic to both Hand K. Hence, ([H] - [G]) + ([K] - [G]) = [L] - [G] E U, since [H] + [K] = [H EB K] = [G EB L] = [G] + [L] and Ko(G) is a group. By Corollaries 2.2.6 and 2.2.4, H EB M = G EB G for some group M nearly isomorphic to G . Thus, [M] - [G] = -([H] - [G]) E U.
em
68
2. Torsion-Free Abelian Groups
If[H] - [G] E U, then H is a summand of G EB G. By Lemma 2.3.10, there are only finitely many isomorphism classes of summands of G EB G . In particular, U must be finite. Choose a positive integer n with nU = O. If H is nearly isomorphic to G, then [H] - [G] E U, and so n([H] - [G]) = O. Thus, n[G] = [G n ] = [H n ] = nH E Ko(G) and G" EB C is isomorphic to H" EB C with C a summand of G'" for some m ::: I.
Theorem 2.3.12 [Warfield 80] Two torsion-free abelian groups G and H of finite rank are nearly isomorphic ifand only if there is some integer n with G" isomorphic to H". is Corollary 2.2.4(b). (=}) By Lemma 2.3.11, there is some positive integer m with G" = K EB K' and positive integer n with G" EB K isomorphic to H" EB K. Then, Gn +m is isomorphic to G" EB K EB K', which is isomorphic to H" EB K EB K' = H" EB G'" , Hence, G n +m is isomorphic to H" EB G'", First assume m ::: 2. By Corollary 2.3.7(b), H" EB G m- I is isomorphic to G' EB m G +n- 2 with G' EB G isomorphic to G 2 • Thus, H" EB G m - I is isomorphic to Gn+(m-l ). This reduces m by 1. It now suffices to assume that m = I, i.e., Gn + 1 is isomorphic to H" EB G. Next, suppose n ::: 2. Then G EB G EB G n - I is isomorphic to G EB H" , Again, by Corollary 2.3.7(b), H" is isomorphic to G' EB Gn - I for some G' with G EB G' isomorphic to G 2 . Hence, H" is isomorphic to (G' EB G) EB Gn - 2 and so to G" , as desired. The only remaining case is m = n = 1, i.e., G 2 is isomorphic to G EB H. Since G is nearly isomorphic to H , G EB G' is isomorphic to H 2 for some G'. Then H 3 = H 2 EB H is isomorphic to, sequentially, G EB G' EB H , G EB H EB G' , G EB G EB G' , G EB H EB H, and G 2 EB H. By Corollary 2.3.7, H has the 2-substitution property and G 2 is isomorphic to H EB H' for some H' with H EB H ' isomorphic to H 2 . Thus, G 2 is isomorphic to H 2, as desired. PROOF. (<==)
EXERCISES 1. [Goodearl 76] Suppose S isa subring of R with nR 5; S forsome nonzero integer nand R/nR Artinian. Prove that if 1 is in the stable range of R, then I is in thestable range
of S. 2. Suppose G is a reduced, semilocal, torsion-free abelian group of finite rank. Let n be theproduct of theprimes p with pG :I: G. Show thatnEnd G 5; lEnd G and conclude that 1is in thestable range of EndG. 3. A torsion-free abelian group G of finite rank has self-cancellation if G €a G isomorphic to G €a G' implies G is isomorphic to G'. Show G has self-cancellation in thefollowing cases: (a) If EndG €a M is isomorphic to EndG €a End G as right EndG-modules, then M is isomorphic to EndG.
2.4 Self-SmallGroups and EndomorphismRings
69
(b) G is isomorphic to H 2 for some H . (c) G = H ffi K and both Hand K have self-cancellation. 4. Assume that G, H , and K are torsion-free abelian groups of finite rank with G ffi K isomorphicto H ffi K . Prove that G is isomorphicto H in the following cases: (a) Either G or K is semilocal. (b) K is isomorphic to a summand of G and K has self-cancellation. (c) For some n > I, G has self cancellation and G" = Y ffi X for some X nearly isomorphicto K.
5. [O'Meara, Vinsonhaler99] Let G and H be torsion-free abelian groups of finite rank. Prove that G" is isomorphic to H" for each integer n ::: 2 if and only if G ffi G, G ffi H , and H ffi H are all isomorphic.
2.4 Self-Small Groups and Endomorphism Rings There is a category equivalence from the category of summands of direct sums of a self-small abelian group A to the category of projective End A-modules (Theorem 2.4 .1) . In particular, finite rank torsion-free abelian groups can be investigated in the context of finitely generated projective modules over subrings of semisimple finite-dimensional Q-algebras (Theorem 2.4.3). Moreover, each reduced subring of a finite-dimensional Q-algebra may be realized as the endomorphism ring of a finite rank torsion-free abelian group (Theorem 2.4 .6). Write ffi/A for the direct sum of copies of an abelian group A indexed by the set I. An abelian group A is self-small if for each 1 E Hom(A , ffi/A) , there is a finite subset J of I with I(A) ~ ffiJA . For example, if A is a torsion-free abelian group with finite rank, then A is self-small. To see this, let 1 E Hom(A , ffi/A) . Since A has finite rank, there is a finitely generated free subgroup B of A with rank B = rank A and AlBa torsion group. Then I(B) ~ ffiJA for some finite subset J of I . But ffiJA is a pure subgroup of ffi/A, being a summand. Hence, I(A) ~ C ~ ffiJA, where C is the pure subgroup of ffi/A generated by I(B). For an abelian group A, P(A) is the category with summands of direct sums of copies of A as objects and group homomorphisms as morphism sets. For a ring R , P(R) denotes the category of projective right R -modules with right R -module homomorphisms as morphism sets. Both P(A) and P(R) are additive categories. Suppose F , G: X -+ X' are functors with X and X' add itive categories. A natural equivalence of F and G is an isomorphism
=
Theorem 2.4.1 If A is a self-small abelian group. then there is a category equivalence H A : P(A) -+ P(EndA) given by HA(G) = Hom(A , G) .
70
2. Torsion-Free Abelian Groups
PROOF. The group A is a left End A-module with [ a = I(a) for I E End A and a E A. Then Hom(A, G) is a right End A-module, where fr E Hom(A, G) is the composite of I E Hom(A, G) and r E End A . If I E Hom(G, K), then HA(f) : Hom(A, G) --+ Hom(A, K) is defined by HA(f)(g) = Ig for each g E Hom(A, G). Thus, H A is a functor from P(A) to the category of right End Amodules, observing that HA(fI') = HA(f)HA(f') and HA(lG) = 1. Moreover, HA(G EB K) = HA(G) EB HA(K), noticing that if G EB K has injections tc and iK, then HA(iG), HA(iK) are injections for HA(G) EB HA(K). Since A is self-small, H A (tB/ A) = EB/ H A (A) . To see this, for each j E I, leti j be an injection and 1fj a projection for the direct sum EB/ A. If I E Hom(A, EB/ A), then 1= (1fjlij)/ E EB/Hom(A , A), since 1filij = 0 for j E IV, where J is a finite subset of I with I (A) S; EB J A . It now follows that H A : P (A) --+ P (End A) is a functor. Define a functor TA : P(EndA) --+ P(A) by TA(M) = M ®End AA and TA(f) = I ® 1 : TA(M) --+ TA(N) for I E HomR(M, N). There is a natural homomorphism OM: TAHA(M)--+ M given by 0M(f®a) = I(a) for IE Hom(A, M) and a E A. Since OA: TAHA(A) --+ A is an isomorphism, 0: TAHA(EB/A) --+ EB/A is an isomorphism. Hence, if G EB H = EB/A, then 0G : TAHA(G) --+ G is an isomorphism. This shows that 0 : TAH A --+ I PtA) is a natural equivalence from the functor TAH A to the identity functor on P(A). A similar argument shows that
Corollary 2.4.2 II A is a self-small abelian group , then each group in P(A) is isomorphic to a direct sum 01 copies 01 A if and only if each projective right End A -module is free . PROOF. A consequence of Theorem 2.4.1.
For an abelian group A, Pf(A) is the category with summands of finite direct sums of copies of A as objects and group homomorphisms as morphism sets. For a ring R, Pf(R) denotes the category of finitely generated projective right R-modules with right R-module homomorphisms as morphism sets. Modules in Pf(R) are precisely summands of finitely generated free R-modules. If R = End A / NEnd A with A a torsion-free abelian group of finite rank, then each M in Pf(R) is an R-lattice as defined in Section 2.2. This follows from the observation that the additive group of R is torsion-free with finite rank, since NEnd A = End A n J QEnd A is a pure subgroup of the finite rank torsion-free group End A. Consequently, a summand ofa finite direct sum ofcopies of R is a finitely generated R-module with finite rank torsion-free additive group . The next proposition demonstrates the intimate connection between torsionfree abelian groups of finite rank and lattices over subrings of finite-dimensional semisimple Q-algebras.
2.4 Self-Small Groups and Endomorphism Rings
71
Theorem 2.4.3 Assume that A is a torsion-free abelian groupoffinite rankand NEnd A = End A n lQEnd A. There is a functor H : Pf(A) ~ Pf(End AI' NEnd A) given by H(G) = HA(G)I HA(G)NEnd A such that ifG and K are in peA), then G and K are: (a) quasi-isomorphicifand only ifQH(G) andQH(K) are isomorphicas right
QEnd AI1QEnd A-modules; (b) isomorphic if and only if H(G) is isomorphic to H(K); (c) isomorphicat a prime p ifand only if H(G)p is isomorphicto H(K)p as a right (End AI NEnd A)p-module; (d) nearly isomorphic if and only if the End AI NEnd A-lattices H(G) and
H(K) are in the same genus class. PROOF. Notice that H is a functor, being the composite of the functor H A of Theorem 2.4.2 and the functor P(End A) ~ P(End AI NEnd A) given by M t-+ MIMNEndA. In particular, if f E Hom(G , K), then HA(f) : HA(G) ~ HA(K) induces H(f) : H(G) = HA(G)IHA(G)NEnd A ~ H(K) = HA(K)I HA(K) NEnd A. Assume that G, K E peA), f E Hom(G, K), g E Hom(K, G), and m is a nonzero integer with fg = mIx and gf = mIG· Then H(f)H(g) = mIH(x) and H(g)H(f) = mIH(G), since H is a functor . Conversely, assume that a': H(G) ~ H(K) and (3': H(K) ~ H(G) are End AINEnd A-homorphisms and m is a nonzero integer with a'B ' = mt ni«, and (3' a' = m IH(G). Since HA(G)andHA(K) are projective End A-modules, there exist a : HA(G) ~ HA(K) and (3 : HA(K) ~ HA(G) with a{3(HA(K» + HA(K)NEnd A = mHA(K) + HA(K)NEnd A. Now, NEnd A s;:; lEnd A and HA(G) is finitely generated as an End A-module, so, by Nakayama's lemma, a{3(HA(K» = mHA(K). Therefore, y = m- 1a{3 : HA(K) ~ HA(K) must be an automorphism. By Theorem 2.4.2, there are f : G ~ K and g : K ~ G with H A (f) = a and HA(g) = (3y-l . Hence, HA(gf) = HA(g)HA(f) = ({3y -l)a = (3{3-1a- 1 ma = m = HA(mIx) , and HA(fg) = HA(f)HA(g)=a({3y-l) = a{3{3-1a- 1 m = HA(mI G ) . But HA is a faithful functor, whence fg = mIx and gf = mIG. The proof is completed by choosing various values of m. For (a), any nonzero m will do, noticing that Q(End AI NEnd A) = QEnd AI lQEnd A. For (b), let m = 1, and for (c), let m be an integer prime to p . Finally, (d) is a consequence of (c) and the definitions of near-isomorphism and genus class of R-Iattices given in Section 2.2. Let R be an integral domain and M a torsion-free R-module, that is, if rm = 0 with r E Rand m E M , then either r = 0 or m = O. If M is a torsion-free Rmodule and N is a submodule of M, then N is a pure submodule if r M n N = r N for each r E R; equivalently, MIN is a torsion-free R -module, If N is a submodule of M, then {m EM: rm E N for some 0 =F r E R} is the pure submodule of M generated by N. The rank of M is the cardinality of a maximal R-independent subset of M .
72
2. Torsion-Free Abelian Groups
Theorem 2.4.4 Suppose R is a principal ideal domain and M is a countably generated torsion-free R-module. Then M is afree R-module if and only if each pure finite rank submodule of Mis afinitely generated free module. PROOF. (=}) Assume that M is a free R-module and N is a finite rank pure Rsubmodule of M. Then N is contained in, hence a pure submodule of, a finitely generated free summand of M . Thus, it is sufficient to assume that M is a finitely generated free R-module. In this case, M / N is a finitely generated torsion-free R-module, hence a free R-module (Exercise 2.4 .2). Because free modules are projective, N is a summand of M. Since M is a finitely gener ated free module and R is a principal ideal domain, N is a finitely generated free R-module. ( ~) Let {Xl , • •. , X n , .. . } be a countable set of generators for M . Define M; to be the pure submodule of M generated by {XI , . . . , x n }. Then MI ~ . . . ~ M; ~ . .. ~ M is an ascending chain of pure finite rank submodules of M with M = U{M n : n ~ I}. By assumption, each M; is a free R-module. For each n, Mn+iI M; is a finitely generated torsion-free R-module, hence free by Exercise 2.4.2. Since free modules are projective, Mn+l = Nn+l EB M; for some free R-module N; and n ~ 1. Define Ni = M, and observe that for each n , M; = N, EB . . . EB N n - l • Then M = EB{Nn : n ~ I} is a free R-module, since M = U{M n: n ~ I} and each N; is a free R-module.
Corollary 2.4.5 [Pontryagin 34] Assume that G is a countable torsion-free abelian group offinite rank. Then G is afree group if and only if each purefinite rank subgroup is afree group. PROOF. Apply Theorem 2.4.4 with R = Z.
Theorem 2.4.6 [Comer 63] Suppose R is a ring with identity and that the additive group ofR is countable, reduced, and torsion-free. There is a countable torsion-free abelian group G with End G isomorphic to R. Moreover. if the additive group of R has rank n, then G may be chosen with rank G = 2n. PROOF. Only an outline of this proof is given. See, for example, [Fuchs 73] for more detailed arguments. Suppose that the additive group of R is torsion-free with rank n. The Z-adic topology is the topology on R with open sets r + nR such that r E Rand 0 #- n E Z. Since R is reduced, R is Hausdorff in this topology. Let R* be the completion of R in the Z-adic topology . Then R* is a commutative Z*-algebra with additive group uncountable and torsion-free and R ~ R*. In fact, Z* = DpZ; with Z; the p-adie integers, an uncountable integral domain, and the transcendence degree of the quotient field of Z; over the algebraic closure of Q is uncountable. First, assume that the additive group of R has finite rank and let {I = rl , ... , r n } be a maximal Z-linearly independent subset of R. Since QR is countable, there is a
2.4 Self-Small Groups andEndomorphism Rings
73
subset {ai , . . . , an} of Z* that is algebraicallyindependent over QR , i.e., for each nonzero polynomial h(XI, . . . ,xn) with coefficients in QR , h(aj , . .. ,an) 1= 0. Define G to be the pure subgroup of R* generated by the group R + RfJ, where fJ = rial + ... + rnan E R*. Then rank G = 2n, and R is isomorphic to a subring of End G, via r t-+ left multiplication by r . On the other hand, let f E End G. Then f extends uniquely to a Z*-endomorphism g of R*. For some nonzero integer m, mf(fJ) = a + bfJ E R + RfJ and mf(ri) = ai + b,fJ E R + RfJ for each i. Hence, mf(fJ) = mg(fJ) = mg(rl)al
+ ...+ mg(rn)an = mf(rl)aj + ... + mf(rn)an = a + bfJ.
Substitution gives
Since the a/s are algebraically independent over QR, each b, equals 0, a = 0, and br, = ai for each i. Hence, mf(ri) = br, for each i, In fact, mf(l) = b, since rl = 1. This shows that mf(x) = bx for each x E R + RfJ with b E R. Since R is pure in R* and mf(l) = b E R, it follows that f is left multiplication by f(l) E R . Consequently, R is isomorphic to End G. For the general case, assume that the additive group of R has countably infinite rank. Define G to be the pure subgroup of R* generated by {R, RfJr : r E R}, where fJr = Pr lR + a.r and {Pr, ar : r E R} is a countable subset of Z* that is algebraically independent over QR. Then G is a countable group, and calculations like those above show that End G is isomorphic to R.
Theorem 2.4.7 [Beaumont,Pierce 61A] If R is a subring ofafinite-dimensional Q-algebra, then the additive group of R is quasi-isomorphic to the group NR EI7 R/NR. PROOF. See, for example, [Arnold 82].
EXERCISES
1. Prove thata functor F : X ~ X' is a category equivalence of additive categories if and only ifthere isa functor G : X' ~ Xwith F G naturally equivalent totheidentity functor lx' and GF naturally equivalent to theidentity functor Ix-
2. Show that if R is a principal ideal domain and M is a finitely generated torsion-free R-module, then M is a free R-module. 3. Assume that A is a torsion-free abelian group of finite rank and End A is a principal ideal domain. (a) Prove thatif G is a subgroup of$1 A with Hom(A , G)A = G, then G is isomorphic to $ J A for some subset J of I . (b) Prove that if G is a pure subgroup of $1 A with I finite, then G is a summand of $1 A. Conclude thatG is a finite direct sumof copies of A.
74
2. Torsion-Free Abelian Groups
NOTES ON CHAPTER
2
Chapter 2 is a brief introduction to some fundamental techniques for countable torsion-free abelian groups. Further results and historical background, especially for groups of infinite rank, may be found in [Fuchs 73], a comprehensive exposition of the status of the subject up to 1973. Much of Chapter 2 is a revised and extended version of portions of [Amold 82], lecture notes focusing on torsion-free abelian groups of finite rank and their relationship to modules over subrings of finite-dimensional Ql-algebras. Included in [Arnold 82] are complete arguments for properties of orders and lattices over orders referenced in Chapter 2. The notion of quasi-isomorphism of torsion-free abelian groups of finite rank is due to [J6nsson 57, 59]. The categorical version here is as given in [Walker 64]. In [Jonsson 59], it is proved directly that frA
2.4 Self-Small Groupsand Endomorphism Rings
75
Theorem 2.4.3 is the crucial result for an interpretation of torsion-free abelian groups of finite rank as finitely generated projective modules over subrings of finite-dimensional Ql-algebras. Suchan interpretation provides justification for thecomplexity of thesegroups as well as for positive results for groups with restricted endomorphism rings. This point of view, as well as some far-reaching generalizations of Theorem 2.4.1, is exploited in [Albrecht 83, 89, 91], [Albrecht, Goeters 89, 94], and references therein. Comer's theorem, Theorem 2.4.6, can be used to construct pathological direct sum decompositions [Comer69]. This theorem has also provided motivation for a largenumber of papers on realization of endomorphism rings; see, for example, [Brenner 67], [Dugas, Gobel 82,97], [Comer, Gobel85], [Gobel, Shelah 95], and references therein.
3 Butler Groups
3.1 Types and Completely Decomposable Groups Finite direct sums of torsion-free abelian groups of rank I, called completely decomposable groups, can be classified in terms of types . Included in this section is a compilation of some of the fundamental properties of types and fully invariant subgroups determined by types . The isomorphism class of a rank-I torsion-free abelian group A is denoted by type A. Types of rank-l groups are represented by isomorphism clas ses of subgroups of Q . If X and Y are subgroups of Q and f : X --+ Y is a nonzero homomorphism, then f extends to a unique Q-endomorphism g ofQ, with g(qx) = qf(x)foreach q E Q and x E X. Since Q-endomorphisms of Q'are multiplication by elements of Q, Hom(X, Y) is isomorphic to a subgroup of Q, each 0 =f=. f E Hom(X, Y) is a monomorphism, and End X is isomorphic to a subring of Q. Lemma 3.1.1 (a) If X and Yare torsion-free abelian groups of rank 1, then the following statements are equivalent: (i) Hom(X , Y) and Hom(Y, X) are both nonzero; (ii) X and Yare quasi-isomorphic; (iii) X and Yare isomorphic. (b) The set oftypes form a distributive lattice, where for subgroups X and Yof Q, (i) type X s type Y ifand only ifHom(X, Y) =f=. 0; (ii) the meet, type X n type Y, of type X and type Y is type X n Y; and (iii) the join, type X U type Y, of type X and type Y is type X + Y.
3.1 Types and Completely Decomposable Groups
77
PROOF. Exercise 3.1.1. A consequence of Lemma 3.1.1(a) is that if X and Yare nonzero subgroups of = type Y if and only if (X + Y)/X and (X + Y)/ Yare finite. Let G be a torsion -free abelian group. Recall from Section 2.1 that a subgroup K of G is a pure subgroup of G if K n n G = n K for each nonzero integer n, equivalently, G/ K is torsion-free. It is routine to confirm that if K is a pure subgroup of Hand H is a pure subgroup of G, then K is a pure subgroup of G. Let H be a subset of G and define (H) to be the subgroup of G generated by H and define (H)* = {x E G : nx E (H)} for some 0 =1= n E Z}, the pure subgroup of G generated by H. In particular, (H)* is a pure subgroup of G . If x is a nonzero element of a torsion-free abelian group G and X is the pure rank-1 subgroup of G generated by x , then type x is the isomorphism class of X. Define typeset G = {type x : 0 =1= x E G}. If K is a pure subgroup of G and x is a nonzero element of K, then x has the same type in K as it does in G . Pure rank-I subgroups of a torsion-free group G are like simple modules in the sense that if X and Yare pure rank-I subgroups of G, then either X = Y or X n Y = O. Types can be viewed as equivalence classes of height sequences h = (hp)n, indexed by the set of primes IT, where h p is a nonnegative integer or 00 for each prime p . The terminology arises from the notion of p-heights. Let G be a torsionfree abelian group and x a nonzero element of G . Define h p(x), the p-height ofx in G, to be 00 if x E pWG and n if x E pnG\pn+JG . Then hex) = (hp(x»n is a height sequence. Let X be the pure subgroup of G generated by x . Since X is a pure subgroup of G, the p-height of x in G is exactly the same as the p-height of x in X for each prime. Hence, rank-l groups give rise to height sequence s. There is an equivalence relation on height sequences that determ ines precisely when associated rank-I groups are isomorph ic. Two height sequences h and h' are equivalent if they differ in at most finitely many finite entries. For example , (1,1 , . . . ,1, .. .) is equivalent to (0,0,0,1,1, .. .) but not to (0,00,1,1 , . . .). A routine verification shows that this is an equivalence relation . Observe that if m and n are nonzero integers and 0 =1= x E G, then h(mx) is equivalent to h(nx). It follows that if G is a torsion-free group, =1= x E G, and X = (x)*, then hey) is equivalent to h(x) for each =1= y EX .
Q, then type X
°
°
Theorem 3.1.2 There is a one-to-one correspondencefrom the set of types to the set of equivalence classes ofheight sequences. PROOF. The correspondence is given by ¢(type X) = [hex)], where [hex)] is the equivalence class of the height sequence of 0 =1= x EX. To see that ¢ is welldefined, assume that X and Y are rank-l groups, 0 =1= x EX, 0 =1= y E Y , and f : X -+ Y is an isomorphism. Then hex) = h(f(x». Since Y = (f(x»)* = (y)*, h(f(x» is equivalent to hey), and so hex) is equivalent to hey). As for one-to-one, suppose that hex) and hey) are equivalent. Then there are nonzero integers m and n with h(mx) = h(ny) . Define f: Qmx -+ Qny by
78
3. Butler Groups
f(qmx) = qny for each q E Q. Then f: X ~ Y is a well-defined isomorphism, since h(mx) = h(ny), X = (mx)*, and Y = (nYk Finally, let h = (hp)ll be a height sequence and define X to be the subgroup of Q generated by {l / pi : i ::: h p , p E TI}. Then lQ E X, h(lQ) = h, and ¢(type X) = [h] . This completes the proof. The type of an element of a direct sum can be computed from the types of its components. Suppose G = A $ B is a direct sum of torsion-free abelian groups and x = (a, b) is a nonzero element of G. Then type x = (type a) n (type b), the meet of type a and type b. This is because hp(x) = min{hp(a) , hp(b)} for each prime p and the type of an element is determined by the equivalence class of its height sequence. More generally, if G = A + B, not necessarily a direct sum , and x = a + b with a in A and bin B, then type x ~ (type a) n (type b), since hp(x) ~ min{hp(a), hp(b)} for each prime p.
Lemma 3.1.3 If G is a torsion-free abelian group with finite typeset, then the typeset of G is closed under meets. PROOF. If rank G = l , then each nonzero element of G has the same type. Assume that rank G ~ 2 and let x and y be Z-independent elements of G. Then {x + my: m nonzero integer} is an infinite subset of G . Since typeset G is finite, there are nonzero integers m and m' with m i= m' and T = typex + my = typex + m'y E typeset G. Now, T ~ typex n type my = typex n typey . On the other my) - (x m'y), so that type y = type (m - m')y ~ hand, (m - m')y = (x (type x + my) n (type x + m' y) = T. Similarly, type x ::: T, since (m' - m)x = m'(x + my) - m(x + m'y). Hence, type x n type y ~ T, and so type x n type y = T E typeset G.
+
Let
T
+
be a type and G a torsion-free abelian group. Define G(T)
= {x: type x 2: T},
called the t-socle of G, a pure subgroup of G. Define G*(T)
= ({x E G : type x >
T})
to be the subgroup of G generated by {x E G : type x > T} and G#(T)
= (G*(T))* ,
the pure subgroup of G generated by G*(T). For example, if X is a rank-l group, then X (T) = X if type X ~ T, and X(T) = 0 otherwise; X*(T) = X if type X > T, and X*(T) = 0 otherwise; and X#(T) = X*(T) for each T.
Lemma 3.1.4 Suppose G and H are torsion-free abelian groups and are types.
(1
and T
3.1 Types and Completely Decomposable Groups
79
(a) If t ~ a, then G(a) £; G(r). (b) G(r)(a) = G(r U a) = G(r) n G(a). (c) If H is a pure subgroup of G, then H(r) = H n G(r). (d) If f E Hom(G , H) then f(G(r» £; H(r) , f(G*(r» £; H*(r), and f(G#(r» £; H#(r). (e) (G $ H)(r) = G(r) $ H(r) , (G $ H)*(r) = G*(r) $ H*(r), and (G $ H)#(r) = G#(r) $ H#(r). PROOF. Assertions (a), (b), and (c) are consequences of the definitions , while (d)
follows from the fact that if X is a pure rank-I subgroup of G with f(X) =f:. 0, then type X ~ type f(X) , since hp(x) ~ hpf(x) for each x E X and prime p . To confirm (e), apply (d) to injections and projections for the direct sum G $ H . Recall that a completely decomposable group is a finite direct sum of rank-I torsion-free abelian groups . The r-socle of a completely decomposable group C is completely determined by the rank-I summands of C. Example 3.1.5 Suppose C = X I $ ... $ X; is a completely decomposable group with rank Xi = 1 for each i.
lf:
is a type, then CCr) = ${X i : type Xi ::: r}, C*(r) = ${Xi: type Xi > r ] = C#(r), and C(r)/C#(r) is isomorphic to ${X i: type Xi = r}, (b) The typeset of C is the meet closure of {type X i: 1 ~ i ~ n}. In particular, typeset C is finite. (a)
If x = Xl $ . . . $ Xn is a nonzero element of C with Xi in X i, then type x = n{typexi : Xi =f:. O] and type Xi = typer, if Xi is nonzero. This observation can be used to confirm both (a) and (b).
PROOF.
The critical typeset of a torsion-free abelian group G is Tcr(G) = {r : G(r)/ G#(r) =f:. O} . In view of Example 3.1.5(a), the critical typeset of a completely decomposable group records the types of rank-I summands of G . Specifically, if C = X 1$ · .. $ X; with rank Xi = 1 for each i , then Ter(C) = {type (Xi): 1 ~ i ~ n} . A torsion-free abelian group G is homogeneous if there is a type r with type X = r for each nonzero element X of G. In this case, G is called t-homogeneous . A rank-I group X is homogeneous. It follows from Example 3.1.5 that if G is homogeneous completely decomposable, then G is isomorphic to x n for some rank-I group X with type X = r , Homogeneous completely decomposable groups have some special properties. Lemma 3.1.6 [Baer 37] Assume that C is a r -homogeneous completely decomposable group.
(a)
If C = G / K for some torsion-free abelian group then K is a summand of G.
G and G( r ) + K
= G,
80
3. Butler Groups
(b) Each pure subgroup of C is a summand. (c) Each summand of C is r -homogeneous completelydecomposable. PROOF. (a) It is sufficient to assume that C has rank I and type r . For the general case project onto a rank-I summand of C and induct on the number of rank-I summands of C. Let JT : G -+ C = G I K denote the canonical epimorphism with ker zr = K. Define I = {JTh: h E Hom(C , G)}, a right ideal of End C. Then IC = C . To see this, let 0 :f:. x E e. Since G(r) + K = G and JT is onto with ker zr = K, there is y E G(r) with x = JT(y). Because type y ::: r = typex and x E C with rank C = 1, there is h E Hom(C, G) with y E h(C). Thus, x = JT(y) E JTh(C) 5:; l C. Since End C is a principal ideal domain, I = fEnd C for some f E End C. Then f C = I C = C, whence f E I is an automorphism of C, and so I = End C . But Ie E I, so that there is h E Hom(C , G) with JTh = Ie . This shows that K = ker JT is a summand of G. Both (b) and (c) are consequences of Exercise 2.4.3. Following is a more direct argument. (b) It is sufficient to assume that C = n for some subgroup X of Q of type r with Z 5:; X, since a homogeneous completely decomposable group C is a direct sum XI EB· . .EB K; with each Xi a rank-I group isomorphic to a common subgroup X of Q containing Z. Write C = X" = X (Z") with C IZn a torsion group. If H is a pure subgroup of C , then H nzn is a pure subgroup, hence a summand, of Z", This is because zn I H nzn is free, being finitely generated and torsion-free. Write Hn EB D . Then H = X(H nzn), since H I(H nzn) is torsion and X(H nzn) is a pure subgroup of C. Thus, C = x(zn) = X(HnZn)EBX D = H EBX D, and H is asummandofC. (c) Assume that H is a summand of C. As in (b), C = H EB H' , where zn = (H n zn)EBD with X(HnZn)=H and H'=XD . Then H is r-homogeneous completely decomposable, since H n is a finitely generated free group, being a summand of the free group zn, and X is a rank-I group of type r ,
x
zn =
zn
zn
Theorem 3.1.7 [Baer 37] (a) Summands of completelydecomposable groups are completelydecompos-
able. (b) Let C = XI EB .. . EB X m and D = YI EB ... EB Yn be two completely decomposable groupssuch that Xi and Yj are rank-l groups. Thefollowing statements are equivalent: (i) C and D are isomorphic; (ii) C and D are quasi-isomorphic; (iii) m = n and there is a permutation a of {l , 2, .. . , n} with type Xi = type Yu(i)/or each i; (iv) rank C(r)/C#(r) = rank D(r)1 D#(r)for each type r .
PROOF. (a) Let G be a summand of a completely decomposable group C, say C = G EB H. If r E Ter(C), then C(r) = C r EB C*(r) with C r a r-homogeneous
3.1 Types and Completely Decomposable Groups
81
completely decomposable group by Example 3.1.5(a) . Moreover, C = EB{ C r : r E Tcr(C)}. The proof is an induction on n, the cardinality of Tcr(C). If n = I, then C is homogeneous and G is completely decomposable by Lemma 3.1.6(c). Assume that n » I and that r is a maximal element of Tcr(C). Then C r = C(r) is homogeneous completely decomposable and a summand of C. But C(r) = G(r) EB H(r), by Lemma 3.1.4(e) . Thus, G(r) and H(r) are homogeneous completely decomposable summands of C by Lemma 3.1.6, hence of G and H, respectively. Write G = G(r) EB G' and H = H(r) EB H' for some groups G' and H'. Then EB{Cu:a E Tcr(C)} = C = G EB H = C r EB G' EB H', whence G' EB H ' is isomorphic to EB{Cu : a E Tcr(C), a =f:. r ]. By induction on n, G' is completely decomposable, as is G = G(r) EB G' . (b) (ii) ~ (iii) Assume that C and D are quasi-isomorphic and let r be the type of some C, or D]. By Example 3.1.5(a), C(r) = C, EBC*(r) and D(r) = Dr EB D*(r) with C r and Dr r-homogeneous completely decomposable groups . Then C(r) is quasi-isomorphic to D(r), C*(r) is quasi-isomorphic to D*( r ), and so C, is quasiisomorphic to Dr. In particular, rank C, = rank Dr for each r E critical typeset C. Thus , C, and Dr are isomorphic, both being isomorphic to X" for some rankI group X of type r. It follows that C = EB{C r : r E Tcr(C)} is isomorphic to D = EB{D r : r E Tcr(D)}. The implications (iii) =} (iv) =} (i) =} (ii) follow immediately from the definitions and Example 3.1.5(a) . Recall from Chapter 2.2 that if G is a torsion-free abelian group of finite rank, then NEnd G = (JiQEnd G) n End G is a nilpotent ideal containing each nilpotent ideal of End G.
Proposition 3.1.8 If C is a completely decomposable group, then End C/ NEnd C is a finite product of matrix rings over subrings ofiQ. Let C = X I EB . .. EB X n be a completely decomposable group with rank Xi = I for each i. By grouping together those Xi 'S that are isomorphic, write C = Al EB · ·· EB Am, where each Ai is ri-homogeneous completely decomposable and the ri's are distinct types . Observe that HomrA}, A j) =f:. 0 if and only ri :::: rj. If Horn/A, , A j) =f:. 0 and Hom(A l> A i) =f:. 0, then i = j, since the ri's are assumed to be distinct. Since End C = EB{Hom(A j , A j ) : 1 :::: i, j :::: m} , it is sufficient to prove that if I = EB{Hom(A i, A j ) : I :::: i =f:. j :::: m}, then I is a nilpotent ideal of End C. In this case, I ~ NEnd C and (End C)/ I = End Al x ... x End An is isomorphic to Matn(l)(End X t> x -.. x Matn(m)(End X m ) with each End Xi asubring ofiQ and rank Ai = nU) for each i, Consequently, N«End C)/ l) = 0 and NEnd C ~ I, since NEnd C is a nilpotent ideal of End C. Then NEnd C = I and End C/ NEnd C is a finite product of matrix rings over subrings of iQ. To see that I is an ideal, let 0 =f:. f E Horn/A}, A j ) and 0 =f:. g E Hom(A n As) ~ I with r =f:. s . Then t; < rs . If 0 =f:. fg E Homt.A,, A j ) , then t, < r s :::: ri :::: rj, whence r =f:. j . Thus, f gEl, and I is a left ideal of End C. A similar argument shows that I is also a right ideal of End C. PROOF.
82
3. Butler Groups
If n is a positive integer and x E In, then x is a finite sum of a composition of homomorphisms Iu E Hom(A i, A j) ~ I with i i= j. Since m, the number of Ai'S, is finite, there is a sufficiently large n such that any such composition has some subscript i repeated. Each such composition is 0; otherwise, there is some j i= i with both Hom(Ai, A j ) and Hom(A j , Ai) nonzero, a contradiction. Consequently, In = O.
The typeset of a torsion-free abelian group G records the isomorphismclasses of pure rank-I subgroups of G. There is also a set of types that records the isomorphism classes of rank-l torsion-free homomorphic images of G. Up to isomorphism, a rank-l torsion-freehomomorphic image of G can be written as f(G) for some 0 i= f : G ~ Q. For a finite rank torsion-free abelian group G, define cotypeset G to be {type f(G): 0 i= f: G ~ Q}. Equivalently, cotypeset G is the set of all types of rank-l homomorphic images of G. For a type r, define G[r] = n{kemelf: f
s r},
E
Hom(G, Q) , type f(G)
E
Hom(G, Q), type f(G) < r}.
called the r -radical of G and let G*[r]
= n{kemelf : f
° °
Notice that G*[r] = n{G[a] :a < r}, G[r] is a subgroup of G*[r], and if i= x E G, then there is a pure subgroup K of G with rank G/ K = 1 and x + K i= 0. Hence, G[r] = G if and only if a 1:. r for each a E cotypeset G, and G[r] == if and only if r ::: a for each a E cotypeset G. For example, if X is arank-l group and r is a type, then X[r] = X if type X 1:. r and if type X :::: r , Moreover, X*[r] = X if type X -{:. rand 0 if type X < r , In particular, if r = type X, then X[r] = 0 and X*[r] = X. The next proposition shows that properties of r-socles have analogues for rradicals.
°
Lemma 3.1.9 Suppose G and Hare torsion-free abelian groups offinite rank.
(b) (c)
(d) (e)
s
=
=
r, then G[r] ~ G[a]. Moreover, (G/ G[ rDl r] 0, and if H[ r] 0, then each nonzero element of H has type less than or equal to r . /f g E Hom(G, H) and Tis a type, then g(G[r]) ~ H[r], and g(G*[ rD ~ H*[r]. /f H is a pure subgroup ofG[r], then (G/ H)[r] G[r]/ H. For each type r, (G E9 H)[ r] G[r] E9 H[ r] , and (G E9 H)*[ r] G*[ r] E9 H*[r]. If cotypeset G is finite, then it is closed under joins.
(a) Ifa
=
=
=
(a) Assume that a :::: r. Then G[r] = n{kerh: h : G ~ Q, typeh(G) :::: G[a] = n{kerh : h: G ~ Q, typeh(G) :::: a}, since if h: G ~ Q with type h(G):::: a, then typeh(G):::: r .
PROOF.
r]
~
3.1 Types and Completely Decomposable Groups
83
Next, let g : G ~ G j G [T] be the canonical epimorphism defined by g(a) = a + G[ T] . If x + G[ T] E (G jG[T])[T] = n(ker h : h : G jG[T] ~ Q, type h(G) :::: T}, then x E n(kerhg : hg : G ~ Q, typeh(G):::: T}. On the other hand , if f: G ~ Q with type f(G) :::: T, then f(G[T]) = 0 and f = hg for some h: GjG[T] ~ Q with typeh(G):::: T. Thus ,x E G[T] = n{kerf : f :G ~ Q ,typef(G):::: T} and so (G j G[ T])[T] = O. Finally, let X be a pure rank-I subgroup of H . If type Xi T, then f (X) = 0 for each f E Hom(H, Q) with type f(H) :::: To Hence, X ~ H[T] = 0, a contradiction. (b) If h : H ~ Q with type h(H) :::: T, then hg: G ~ Q with type hg(G) :::: typeh(H) :::: To Let x E G[T] . Then hg(x) = 0 and g(x) E kerh for each h: H ~ Q with type h(H) :::: T, i.e., g( x) E H[T]. A similar argument, using thoseh:H ~ Q with type h(H) < T,showsthatg(G*[T]) ~ H*[T] . (c) Let K be the subgroupofG with K j H = (G j H)[T] .As a consequence of(b), G[T]jH is contained in (GjH)[T], and so G[T] is contained in K . Conversely, if h: G ~ Q with typeh(G) :::: T, then h(H) = 0, since H ~ G[T]. Thus, h : GjH ~ Q with type h(GjH):::: T, and h(K) = h(KjH) = h«GjH)[T]) = O. Therefore, K ~ n{kerh : h: G ~ Q , typeh(G) :::: T} = G[T]. (d) This follows from (b) using projections and injections of G E& H . (e) Let 0 i= f , g E Hom(G , Q) with type f(G) i= typeg(G). For each integer n , n; = f+ng: G ~ Qisanonzerohomomorphismwithhn(G) ~ f(G)+g(G) . In particular, type hn(G) :::: type f(G) U type g(G) for each n. Since cotypeset G is finite, there are integers n i= m with type hn(G) = typehm(G). Now, b; h m = (n - m)g, and -mh n + nh m = (n - m)f with n - m i= O. Hence , (n - m)(f(G) + g(G)) ~ hn(G) + hm(G), and so type f(G)Utypeg(G) :::: typehn(G) U type hm(G) = typehn(G). Consequently, type f(G) U typeg(G) = typehn(G) E cotypeset G, as desired . The critical cotypesetofa torsion-free abelian group G is CTcr(G) = {T : G*[T]j G[ T] i= OJ. The critical cotypeset is, in fact, a subset of the cotypeset of G, since if G*[T]jG[T] i= 0, there must be h : G ~ Q with type h(G) = T. Just as for typesets, the critical cotypeset records the types of rank-I summands of a completely decomposable group . Example 3.1.10 (a) IfC = XIE&· "E&X n isa completely decomposable group with rank Xi for each i and T is a type, then C[ T]
= E&{X i : type X;f. T},
C*[T]
=1
= E&{Xi : type XiI T},
C j C[ T] is isomorphic to E&{X i : type Xi :::: T}, and C*[T]j C[T] is isomorph ic to E&{X i : type Xi T}. Moreover, cotypeset C is the join closure of {type Xi : 1 :::: i :::: n}, ajinite set. (b) If C and D are completely decomposable groups, then C and D are isomorphic if and only ifrank C*[T]jC[T] = rank D*[T]j D[T]for each T.
=
84
3. Butler Groups
PROOF. (a) The first statement follows from Lemma 3.1.9(d) and the computations of X[ r] and X*[r] for a rank-l group X. As for the latter statement, let r E cotypeset C and 0 =f- f : C ~ iQ with type f(C) = r. Then f(C) = :E{f(X;) : f(X;) =f- 0, type f(X;) ::: r}, whence r is the join of {type X; : f(X;) =fOJ. Conversely, cotypeset C is closed under finite joins by Lemma 3.1.9(e) . Assertion (b) is a consequence of (a) and Lemma 3.1.9(b).
Afinite Boolean algebra T is a finite complemented distributive lattice (T, :::) with a least element 0 and a greatest element 1. An atom of T is an element t of T with 0 < t such that if sET and 0 < s ::: t, then s = t. A finite Boolean algebra T is lattice isomorphic to the Boolean algebra peS) of all subsets of a set S with n elements. In Pt S], 0 is the empty set, 1 is the set S, and the atoms are precisely the l-element subsets of S. For an example of a Boolean algebra of types, let S3 = {type Z[l/ p], typeZ[I/ q], type Z[l/r]}, where p, q , and r are distinct primes and Z[1/p] = {m/ pj : m E Z, j ::: OJ, the subring of Q generated by lip. The sublattice T3 of the lattice of types generated by S3 is a Boolean algebra with S3 as the set of atoms. Example 3.1.11 Let T3 be afinite Boolean algebra oftypes with exactly 3 atoms rl, r2, r3, and X; a subgroup ofiQ with type X; = r; for i = 1,2,3. Define G = XI E17X2+X3(l+ I) f iQEI7iQ, a torsion-free abelian group ofrank 2. Then: (i) (ii) (iii) (iv)
typeset G = {ro, 1'1, r2, 1'3}, where 1'0 is the least element of Ti. cotypeset G = {rl U r2, r2 U r3, 1'1 U r3, rl U r2 U r3}' G(1'o) = G, G(1';) = X; for i = 1,2, and G(1'3) = X3(l + 1). G*[1'1 U r2 U r3] = G[rl U r2 U 1'3] = 0, G[r; U rj] = x, with ii, j, k} = {I, 2, 3}, and G*[1'; U 1'j] = G whenever i =1= j.
PROOF. Exercise 3.1.2.
As noted in Exercise 3.1.3, the lattice of types is not a complete lattice. Specifically, infinite meets and infinite joins need not exist. Nevertheless, there are types that bound all of the elements of the typeset and, respectively, the cotypeset of a finite rank torsion-free abelian group G even if the typeset of G, respectively the cotypeset of G, is infinite. Let S = {XI ," " x n } be a maximal Z-independent subset of G. Define the inner type of G to be IT(G) = n{typex; : 1 ::: i ::: n}. The outer type of G is OT(G) = U{typeG/Y;: 1 ::: i ::: n}, where Y; is the pure subgroup of G generated by {x j : 1 ::: i =f- j ::: n}, noticing that rank Y; = n - 1 and rank G/ Y; = 1 for each i . Lemma 3.1.12 [Warfield 68] Let G be a torsion -free abelian group offinite rank. Then (a) IT(G) and OT(G) do not depend on the maximal Z-independent set S;
3.1 Types and Completely Decomposable Groups
85
(b) IT(G ) ::: r for each r E typeset G and OT(G) ::: a for each a E cotypeset
G; (c) iftypeset G isfinite, then !T(G ) E typeset G, and ifcotypeset G isfinite, then OnG ) E cotypeset G; and (d) IT(G ) ::: OT(G ). x n} and 5' = {x ; , . . . , x~ } be two maximal Zindependent subsets of G . For each i , there is a nonzero integer m with mx; = k\ x\ + ... + knx n for some k, E Z. Then type x; = type m x ; ::: n{type k.x , : k, =I O} ::: n {typexi ; 1 ::: i ::: n} . Hence, n{type x;; I ::: i n} ::: n{typexi: 1 i n}. The reverse inclusion is proved by a similar argument. The fact that OnG) does not depend on 5 follows from the proof of (b). (b) If r E typeset G, then r = typex for some 0 =I x E G. Extend x to a maximal independent subset 5, so that type x ::: n{type s: s E 5} = !T(G). Next, let X be a rank-I group with OT(G) = U{type G I Y i : 1 ::: i ::: n} = type X. There is a monomorphism h : G ~ G I Y\ EEl • • . EEl G I Yn defined by h(x) = (x + Y I , . •• , x + Yn). Since type G I Yi ::: type X, there is a monomorphism g. : GIYi ~ X for each i . The composite g = (g l, . .. , gn)h : G ~ K" is a monomorphism with rank G = n = rank X", Let a E cotypeset G and f: G ~ Q with type f (G ) = a . Define C to be the pure subgroup of x n generated by g( ker f) . Then rank C = n - 1, and so X n I C is a rank-1 group isomorphic to X by Lemm a 3. 1.6(b) and (c). Thus, g induces anonzerohomomorphismG/ker f ~ Xnl C ~ X. By Lemma 3.1.1(b), a = type f( G) ::: type X = OT(G ), as desired. Assertion (c) follows from (b) and Lemmas 3.1.3 and 3.1.9(e). As for (d), let o =I x\ E G, 5 = {XI , X2, . • • , x n } a maximal Z-independent subset of G, and Yi the pure subgroup of G generated by { Xj : I ::: i =I j ::: n}. Then type Xi ::: type G I Yi ::: OT(G ), since 0 =I X i + Yi E G I Yi . It follows that IT(G ) ::: OT(G ). PROOF. (a) Let
5 =
{XI , . .. ,
s
s s
There are relation ships between inner types and r -socles and between outer types and r -radicals, Proposition 3.1.13 [Lady 79] IfG is a torsion-free abelian group of fi nite rank, then: (a) G(r) = G if and only if Yi'(G'y > t ; (b) G[r] 0 ifand only ifOT(G)::: t ; and (c) the followin g statements are equivalent: (i) r = !T(G ) = OT(G ); (ii) Gis t-homogeneous completely decomposable; (iii) G(r ) = G and G[r] = O.
=
PROOF. Statements (a) and (b) are consequences of the definitions.
(c) (i) ~ (ii) Assume that IT(G ) = OT(G ) = r. Let 0 =I f E Hom(G , Q). Then G(r ) = G , since InG ) = r , and type f (G ) = r , since OT(G ) = r = !T(G ). By Lemma 3. 1.6(a), G = K EElY with K = ker f and Y isomorphic to
86
3. Butler Groups
f(G) . Then IT(K) = r = OT(K) so, by induction on the rank of G, K is rhomogeneous completely decomposable. Since Y is a rank-l group with type r , G is r-homogeneous completely decomposable. (ii) => (iii) If G is r-homogeneous completely decomposable, then G(r) = G by Example 3.1.5(a) and G[r] = 0 by Example 3.1.10(a). (iii) => (i) By (a) and (b), IT(G) ::: OT(G). Since IT(G) ~ OT(G) by Lemma 3.1.12(d), the conclusion is that IT(G) = OT(G).
Proposition 3.1.14 [Warfield 68] Suppose G and H are torsion-free abelian groups offinite rank. Then rank Hom( G , H) = (rank G)(rank H) if and only if IT(H) ::: OT(G). PROOF. Observe that Hom( G, H) is contained in Hom(iQG, iQH) , a iQ-vector space
with dimension equal to the product of the dimension of iQG and the dimension of iQH. Then Hom(G, H) is a torsion-free abelian group with rank ~ (rank G) (rank H) . Assume IT(H) ::: OT(G). Let S (x), . . . , x n } be a maximal Z-independent subset of H with IT(H) = n{type Xi r l ~ i ~ n}, Xi the pure subgroup of H generated by Xi, and T = {Yl, " " Ym} a maximal Z-independent subset of G with OT(G) = U{type G j Yj : } ~ j ~ n}, where Yj is the pure subgroup of G generated by {Yi :} ~ i #- j ~ n} . Then type Xi ::: IT(H) ::: OT(G) ::: type G[Y], Hence, there are nonzero homomorphisms Ii i : G ~ Gf Y, ~ Xi. Now, Xl $ . .. $ K; S; H, so that {fji:} ~ j ~ m, } ~ i ~ n} is a Z-linearly independent subset ofHom(G, H). It now follows that rank Hom(G , H) mn (rank G)(rank H). Conversely, assume rank Hom(G, H) = (rank G)(rank H). Then iQHom (G , H) = Hom(iQG, iQH), since iQHom(G , H) is a subspace of Hom(iQG, iQH) with the same dimension. It suffices to prove that if G j K is a rank-I quotient of G and C is a pure rank-I subgroup of H , then Hom( G j K, C) #- O. The composite iQG ~ iQ(Gj K) ~ iQC S; iQH is an element f E Hom(iQG, iQH) = iQHom(G, H) with iQK = ker f . Choose 0 #- k E Z with kf E Hom(G, H) . Then kf(G) S; iQC n H = C, since C is a pure subgroup of H, and so kf induces a nonzero homomorphism G j K ~ C , as desired.
=
=
=
Conditions on the r -radical and r -socle influence the structure of subgroups and torsion-free quotients of a finite rank torsion -free abelian group G. Notice that (c) of the next proposition is a partial "dual" to Lemma 3.}.6(a). Proposition 3.1.15 [Lady 79] Let G be a torsion-free abelian group offinite rank, r a type , and K a pure subgroup of G. (a) If(GjK)(r) = GjK andG[r] = 0, then GjK isa x-homogeneous completely decomposable group. (b) If K(r) K and G[r] 0, then K is a x-homogeneous completely decomposable summand of G.
=
=
3.1 Typesand CompletelyDecomposable Groups
87
(c) If H is a t-homogeneous completely decomposable subgroup of G with H ED G[r] a pure subgroup of G, then G = H ED Lfor some subgroup L of G with G[r] ~ L. PROOF. (a) By Proposition 3.1.13(a) and (b), IT(G/K) 2: r 2: OT(G). Now, OT(G) 2: OT(G / K) 2: IT(G/ K) , observing that a rank-l quotient of G / K is also a rank-l quotient of G . Then OT(G/K) = r = IT(G/K). By Proposition 3.1.13(c), G / K is r -homogeneous completely decomposable. (b) Let p be a prime. By Proposition 3.1.13(a) and (b), IT(K) 2: r 2: OT(G). Either K p is divisible, or else G p is a free Zp-module. To see this, let G / K be a torsion-free rank -l quotient of G and X a pure rank-l subgroup of K. Since type G/K ~ OT(G) ~ IT(K) ~ type X, there is a nonzero f :G/K -+ X , hence a nonzero homomorphism fp : (G / K)p -+ X P ' If (G / K)p is not a free Zp -module, then (G/K)p is isomorphic to Q. This is because a rank-I p-Iocal group is isomorphic to either Z p or Q by Exercise 2.1.3. Hence, X p is isomorphic to Q, and so either G p is a free Zp-module or K p is divisible. Now, Hom(G , K)p is a p-pure subgroup of Hom(G p, K p), and as a consequence of Proposition 3.1.14 and the assumption that OT(G) ~ IT(K) , rank Hom(G , K) = rank Hom(G , K)p = (rank G)(rank K) = (rank Gp)(rank K p) . Thus , Hom(G, K)p = Hom(G p, K p). Since K p is divisible or G p is a free Zpmodule, the restriction homomorphism ¢p : Hom(G p, K p) -+ Hom(K p, K p) is onto . But Hom(G, K) = n{Hom(G , K)p : p E TI} and Hom(K, K) = n{Hom(K, K)p : p ETI} , by Exercise 3.1 .9, so that the restriction homomorphism ¢ : Hom(G , K)-+Hom(K, K), is onto . Choosing g E Hom(G , K) with ¢(g) = I K shows that K is a summand of G . Finally, write G = K ED H . Then 0 = G[r] = K[r] ED H[r] by Lemma 3.1.9(d), whence K[r] = O. Since K(r) = K, K is r -homogeneous completely decomposable by Proposition 3.1.13(c). (c) In this case, K = (H ED G[r])/G[r] is a pure r-homogeneous completely decomposable subgroup of G/G[r] with K(r) = K and (G/G[r])[r] = O. By (b), K is a summand of G / G [r], say f : G/ G [r] -+ K is a projection. Define g = f3fex : G -+ H , where ex : G -+ G/G[r] is the quotient map and f3 : K -+ H is the natural isomorphism. Then G[r] is contained in ker g and g(x) = x for each x E H. Therefore, G = H ED ker g with G[r] ~ ker g , as desired.
EXERCISES 1. [Fuchs 73] or [Arnold82] Providea proof for Lemma 3.1.1. 2. Verify Example 3.1.11. 3. (a) Show that the lattice of types is not a completelattice. (b) Prove that there are infinitechains in the lattice of all types with no least element and no greatest element. (c) Prove that the lattice of types is uncountable.
88
3. Butler Groups
4. Let T be a finite Boolean algebra of types with n atoms . Prove that if {PI, . .. , PnI is a set of n distinct primes, then T is lattice isomorphic to the lattice of types generated by (type Z[I/ pd , . . . , typeZ[1/ Pnll. 5. Let G be a torsion-free abelian group of finite rank and r a type. Prove that if H is a pure subgroup of G with H(r) = Hand H EB G[r] is a pure subgroup of G, then H is a summand of G and G/ H is homogeneous completely decomposable. Give examples to show that the hypothesis that H EB G[ r] is a pure subgroup of G is necessary for this exercise and Proposition 3.1.15(c). 6. (a) Give an example of a rank-2 torsion-free abelian group G with infinite typeset. (b) Give an example of a rank-2 torsion-free abelian group G such that IT(G) fI. typeset G. (c) Give an example of a rank-2 torsion-free abelian group G such that OT(G) fI. cotypeset G. 7. Let X be a subgroup of lQ,0 =1= x EX, and h(x) the height sequence of x in X. (a) Give necessary and sufficient conditions on h(x) for X to be a subring of lQ. (b) Let Y be another subgroup oflQ and 0 =1= y E Y. Show that Hom(X , Y) is a torsionfree abelian group of rank I and compute a height sequence for Hom(X, Y) from h(x) and h(y). 8. Let G be a torsion-free abelian group of finite rank and F a free subgroup of G with rank F = rank G. Define the Richman type of G , RT(G) , to be the quasi-isomorphism class of the torsion group G/ F, where two torsion groups T and T' are quasi-isomorphic if there is a homomorphism f : T --+ T' with ker f and T' /image f finite groups . (a) Prove that RT(G) does not depend on the choice of F . (b) Compute the Richman type ofa rank-I group G in terms ofa height sequence h(x) for G. (c) Write G/ F = EB{Tp : P E ill, Tp = Z(p i(p.I»EB · . ' EB Z(pi(p,n» a direct sum of pgroups with 0::: i(p, I) ::: . . . ::: i(p, n) ::: 00 . Prove that IT(G) is the equivalence class of the height sequence (i(p, I» and OT(G) is the equivalence class of the height sequence (i(p,
n» .
9. Prove that if G is a torsion-free abelian group of finite rank and Il is the set of primes of Z, then G = n{G p : p E ill . Conclude that if H is another torsion-free abelian group of finite rank, then Hom(G, H) = n{Hom(G, H)p : p E ill.
3.2 Characterizations of Finite Rank Groups The class of finite rank Butler groups arises naturally . For instance, it is the smallest class of torsion -free abelian groups that contains all rank-I groups and is closed under isomorphism, finite direct sums, pure subgroups, and torsion-free homomorphic images. Furthermore, the class of Butler groups is precisely the class of finite rank torsion-free abelian groups that arise as representatives of Q-representations of finite posets, as demonstrated in Section 3.3. The first example illustrates that pure subgroups or homomorphic images of completely decomposable groups need not be completely decomposable.
3.2 Characterizations of FiniteRank Groups
89
Example 3.2.1 Let PI , P2, and P3 be three distinct primes, Xi the localization of Z at Pi, and B the pure subgroup of the completely decomposable group Xl EI3 X2 EI3 X3 generated by (1, I , 0) and (0, 1, 1). Then B is a strongly indecompo sable group of rank 2 and a homomorphic image of the completely decomposable group (XI n X 2) EI3 (X 2 n X3) EI3 (XI n X 3). PROOF. Clearly, rank B = 2. Now, type (1, 1,0) = type XI n X2, type (0, 1, I) = type X2 n X 3 , and type (1,0, -I) = type (XI nX 3) . Since (1,0, -I) = (1,1,0)(0, I, I) E B , the typeset of B contains three pairwise incomparable elements. Consequently, B is strongly indecomposable; otherwise, B is quasi-isomorphic to a completely decomposable group of rank 2, which can have at most two pairwise incomparable types by Example 3.1.5. A routine calculation shows that B = «(1, 1,0))* + «0, I , 1))* + «(1 ,0, -1))*, whence B is a homomorphic image of (XI n X2) EI3 (X2 n X 3) EI3 (XI n X3)·
Theorem 3.2.2 [Butler 65] Suppose G is a torsion-free abelian group offinite rank. Then G is a pure subgroup of a completely decomposable group if and only ifG is a homomorphic image of a completely decomposable group. PROOF. A consequence of the more general Lemma 3.2.3. A group epimorphism f : G ---+ H of torsion-free abelian groups is balanced if
f (G (r)) = H (r ) for each type r. In this case, typeset H is contained in typeset G, and f induce s an onto homomorphism G(r)/G#(r) ---+ H(r)/ H#(r) for each
type r . As an example, the homomorphism in Example 3.2.1, (XI n X2) EI3 (X2 n X3) EI3 (XI n X 3) ---+ B, is balanced. A monomorphism f : G ---+ His cobalanced if f(G) is a pure subgroup of H and f induces a monomorphism fr : G/G[r] ---+ H/ H[r] with pure image for each type r , In this case, cotypeset G is contained in cotypeset H . Notice that if f : G ---+ H is a cobalanced monomorphism, then fr(G*[rJ/G[r]) is a pure subgroup of H* [r J/ H [r] for each r , The proof of Lemma 3.2.3 is constructive in the sense that if G is a pure subgroup of XI EI3 ... EI3 X; with rank Xi = I, then there are rank-I groups Yj , constructed from the X/s, such that G is a homomorphic image of YI EI3 .•. EI3 Ym .
Lemma 3.2.3 Let G be a torsion-free abelian group offinite rank. (a) IfG is a pure subgroup ofa completely decomposable group, then there is a completely decomposable group D and a balanced epimorphism f : D ---+ G.
(b) If G is the homomorphic image of a completely decomposable group, then there is a completely decomposable group D and a cobalanced monomorphism f : G ---+ D.
PROOF. (a) Assume that G is a pure subgroup of C = XI EI3 ..• EI3 X; with each Xi a rank-l group . Then typeset G is finite, as a consequence of Example 3.1.5(b),
90
3. BUller Groups
and closed under meets by Lemma 3.1.3. Given x = Xl E9 ... E9 X n in C, let support X = {i : Xi # O}, support G = {support x : 0 # X E G} ordered by inclusion , and S the set of minimal elements in support G . Then S is a finite set, and for each I in S, there is a nonzero X/ in G with support X/ = I. Let X/ be the pure rank-I subgroup of G generated by X/ and notice that support y = I for each nonzero y in XI. Define D = E9{X/ : I E S} and f : D ~ G by f(E9x/) = :Ex/ for X/ E XI' It is sufficient to show that f(D(r» = G(r) for each type r , in which case f(D) = f(D(r» = G(r) = G for r = IT(G), the inner type of G . If r is a type, then D(r) = E9(X/ : type X/ ?: r}, by Example 3.1.5(a) , and H = f(D(r» = :E{X/ : type X/ ?: r] ~ G(r). The proof is concluded by showing that G(r )p is contained in H p for each prime p. In this case, via Exercise 3.1.9, G(r) = n{G(r)p : p prime} ~ H = n{Hp : p prime}, and so H = G(r). Let X = ql X\ + .. . + qnxn be a nonzero element of G(r)p with qi in Zp and Xi in Xi and choose I in S with I ~ support x . Next choose a nonzero X/ = k\x\ + ... + knxn in X/ with each k, in Z, support X/ = I, and k j prime to p for some j in I, recalling that X/ is a pure subgroup of G. Then type xi = n{type Xi: i E I} ?: n{type Xi : qi # O} ?: r, since I ~ support x andx E G(r)p. This shows that X/ E G(r)p. Furthermore, y = X - (qj/kj)x/ E G(r)p with support y ~ support x\ {j } C support x . By induction on the cardinality of support x, Y E Hp, observing that if support X has one element, then X = xi E H . Therefore, X = (qj/ kj)x/ + y E Hp and G(r)p ~ H p as desired. (b) Assume that n : C ~ G is an epimorphism for C = X\ E9 . . . E9 Xn a completely decomposable group with rank Xi = I for each i, Given a nonzero f : G ~ Q, define cosupport f = Ii : frr(X i ) = OJ, cosupport G = {cosupport f : 0 # f E Hom(G, Q)} ordered by inclusion , and S the set of maximal elements of cosupport G. For each I E cosupport G, choose j) E Hom(G, Q) with cosupport !J = I. Define D = E9{!J(G) : I E S} and ¢ : G ~ D by ¢(x) = E9{!J(x): I E S}. If r is a type, then D/ D[r] = E9{f/(G): type !J(G) ::: r ] by Example 3.1.10(a). For each type r , ¢ induces a homomorphism ¢r : G/ G[ r] ~ D / D[ r] with ¢r(x + G[r]) = ¢(x) + D[r] . It suffices to prove that ¢r is a monomorphism with pure image, in which case ¢ is a cobalanced monomorphism. In particular, cotypeset G has a largest element r = OT(G) by Lemma 3.1.9(e), since cotypeset G ~ cotypeset C is finite by Example 3.1.10(a) . By Proposition 3.1.13(b) , G[r] = O. Hence, G/G[r] = G, and ¢ = ¢r is a monomorphism with pure image. To see that ¢r is a monomorphism, it is sufficient to prove that if 0 # X + G [r] E G/G[r], then ¢r(x) # O. Since x tf. G[r], there is some f: G ~ Q with type f(G) ::: rand f(x) # O. Choose I E S with cosupport f ~ I . Then type !J(G) = U{type Xi : i tf. I} ::: U{type Xi: i tf. cosupportf} = type f(G) ::: r. If !J(x) # 0, then ¢r(x+D[r]) # 0 by the definition of¢r, and the proofis complete. Now suppose that !J(x) = O. Since !J # 0, there is 1 ::: i ::: n with i tf. I ; hence i tf. cosupport [ . Choose 0 # q E Q such that 0 # h = q!J - f : G ~ Q with i E cosupport h, i.e., hrr(X i) = O. Then type h(G) ::: (type q!J(G»U(type f(G» ::: r with cosupport h C cosupport f, since i E cosupport h \ cosupport f. Furthermore,
3.2 Characterizations of Finite Rank Groups
91
=
h(x) # 0, since f(x) # 0 and h(x) O. By induction on the cardinality of cosupport f, there is some I' E S with fl'(x) # O. Hence, ¢r(x) # 0, as desired. It remains to prove that image ¢r is a pure subgroup of D / D[ r]. Replace G[r] with G/G[r] and apply Lemma 3.1.9(a) to safely assume that G[r] 0 and ¢ ¢r. It suffices to prove that if p is a prime and x E G\pG, then ¢(x) ¢: pD . There is 0 # f: G 4- iQ with f(x) ¢: pf(G) and f(G)p = 7l pf(x) = f(G)p. This holds because G p is a finite direct sum of copies of tl p and iQ, being an epimorphic image of a direct sum C p = (X\)p $ ... $ (Xn)p of rank-l p-Iocal groups, and if X = (x)*, then X p = ZpX is a pure submodule, hence a summand of, G p. As above, there is I E S with cosupport f S; I and type h(G) type f(G) r . If h(x) ¢: ph(G), then ¢(x) ¢: pD by the definition of ¢ . Now assume 7lpf(x), and that h(x) E ph(G). Observe that h(x) E ph(G)p S; f(G)p if i ¢: I, then f(Jr Xi)p # O. Thus, there is some i ¢: I and integers a and b such that 0 # h = ah - bf:G 4- Q,h(JrX i) = 0, the p -height of bf(x) is 0 in bf(G), and the p-height of ah(x) is not equal to 0 in ah(G). Hence, h(x) ¢: ph(G), h(G)p = 7l ph(x) , and cosupport h C cosupport f. By induction on the cardinality of the cosupport of f, h '(x) ¢: pfl'(G) for some I' E S. Thus, ¢(x) ¢: pD, as desired.
=
:s
=
:s
=
A finite rank Butler group is a pure subgroup of a completely decomposable group, equivalently, a torsion-free homomorphic image of a completely decomposable group. Corollary 3.2.4 [Butler 65] The class of finite rank Butler groups is closed underfinite direct sums, pure subgroups, and torsion-free homomorphic images. Moreover, if G is a torsion-free abelian group offinite rank and H is a finite rank Butler group quasi-isomorphic to G, then G is a Butler group. PROOF. Suppose G is quasi-isomorphic to H. Then there is a monomorphism
f: H 4- G with G/f(H) finite, say G = H + 7lXl + .. . + 7lx n. Since H is a Butler group, there is a completely decomposable group C and an epimorphism g : C 4- H. There is also an epimorphism h : tl n 4- 7lxI + ... + 7lx n defined by hia», ... , an) a\x\ + ...+ anx n. Then (fg, h) : C $ tl n 4- G is an epimorphism and C $ tl n is a completely decomposable group. The remainder of the corollary follows from Theorem 3.2.2.
=
The next theorem, a characterization of Butler groups among the class of torsionfree groups of finite rank, is one of the most useful tools in the subject. This is so because the finiteness conditions of this theorem are amenable to induction arguments. Theorem 3.2.5 Let G be a torsion-free abelian group offinite rank. The following statements are equivalent:
(a) The group G is a Butler group;
92
3. Butler Groups
(b) [Butler 65] (i) The typeset of G is finite and (ii) for each type r in typeset G, there is a decomposition G(r) = G, EB G#(r) with G, a t-homogeneous completely decomposable group and G#(r)jG*(r) bounded; (c) (i) ThecotypesetofG is finite and (ii) for each type r E cotypeset G, G*[r]jG[r] is a t-homogeneous completely decomposable summand ofGjG[r] and if H is the image of G jG*[r] in EB {G j G[a] : a < r ], then (H) *j H is bounded. PROOF. (a) :::} (b)
(i) If G is a pure subgroup of a completely decompo sable group, then typeset G is finite by Example 3.1.5(b) . (ii) By Lemma 3.2.3, there exist C = XI EB· ·· EB X n , with each Xi ofrank 1, and a balanced epimorphism f: C ---+ G . As a consequence of Example 3.1.5(a), C(r) = Cr EB C*(r) with Cr = EB{X i : type Xi = r ] a r-homogeneous completely decomposable group . Then Dr = f(C r ) is r-hornogeneous completely decomposable as a consequence of Lemma 3.1.6. Moreover, Dr n G#(r) is a pure subgroup, hence a summand of Dr. Write Dr = G« EB(D r nG#(r» with G r r-homogeneous completely decomposable . Then G(r) = f(C(r» = Dr + f(C*(r» = Dr + G*(r ) = Dr + G#(r) = G; EB G#(r). Furthermore, G#(r) is a Butler group , being a pure subgroup of the Butler group G. Hence, G#(r ) = Y1 + . . . + Ym +G* (r), with each Yi a pure rank-I subgroup of G of type r. Then G#(r)jG*(r) is torsion , since G#(r ) = (G*(r ») *. Now, Yi n G*(r ), being a pure subgroup of G*(r ), is a rank-l group of type r = type Yi . Consequently, each Yi] Yi n G*(r) is bounded , from which it follows that G#(r)jG*(r) is bounded. (b) :::} (a) The proof is an induction on the length of a chain from an element of typeset G to a maximal element of typeset G. By (i), such a chain must have finite length. Let r be a maximal type in typeset G. Then G(r) = G, is r-homogeneous completely decomposable, in particular a Butler group. Let r E typeset G and suppose, by way of induction, that G(a) is a Butler group for each a in typeset G with a > r . Then G*(r) is a Butler group, being a homomorphic image of the Butler group EB{G(a): a > r} . Moreover, G(r)j(G r EB G*(r» is bounded by (ii). As a consequence of Corollary 3.2.4, G( r ) is a Butler group for each r in typeset G. Let r = IT(G) E typeset G. Then G(r) = G is a Butler group. (c) :::} (a) Let r be a minimal element of cotypeset G . Then G*[r] = G, and by (ii), GjG[r] = G*[r]jG[r] is a r-homogeneous completely decomposable group , hence a Butler group. Now assume that r E cotypeset G and, by way of induction, that GjG[a] is a Butler group for each a E cotypeset G with a < r. By (ii), GjG*[r] is quasi-isomorphic to a pure subgroup of the Butler group EB{GjG[a]: r > a E cotypeset G}. Thus, GjG*[r] is a Butler group as a consequence of Corollary 3.2.4. Moreover, G*[r]jG[r] is r-hornogeneous completely decomposable and GjG[r] is isomorphic to G*[r]jG[r] EB GjG*[r] by (ii). This shows that GjG[r] is a Butler group for each r E cotypeset G. Let r = OT(G) E cotypeset G . Then G[r] = 0, and G = GjG[r] is a Butler group.
3.2 Characterizations of FiniteRankGroups
93
(a) => (c)
(i) Since G is a Butler group, there is a completely decomposable group C and an epimorphism f : C ~ G. Then cotypeset G S; cotypeset C, a finite set by Example 3.1.10(a). (ii) In view of Lemma 3.2.3, there is a completely decomposable group C such that G is a pure cobalanced subgroup of C. It follows that if T is a type, then G*[T]/G[T] is a pure subgroup of C*[T]/C[T]. But C*[T]/C[T] is a Thomogeneous completely decomposable group by Example 3.1.10(a) , so that G* [T]I G [T] is a r-homogeneous completely decomposable group by Lemma 3.l .6(b)and (c). Moreover, G*[T]/G[T] is a pure subgroup ofG I G[ T], since G*[T] is a pure subgroupofG . Now(G*[ T]I G[ T])(T) = G*[T]I G[ T] and (G I G[ T])[ T] = as a consequence of Lemma 3.1.9(a). Thus, G*[T]/G[T] is a r-homogeneous completely decomposable summand of GIG[T] by Proposition 3.1.15(b). As for the final statement, let H be the image of the monomorphism fa : G I G*[T] ~ EB{ G I G[a] : a < T} defined by fa(x+G*[ T]) = EB{x+G[a] : a < T}, recalling that G*[T] = n{G[a]:a < T}. Since G is a cobalanced subgroup of C, GIG[a] is a pure subgroup of CIC[a] for each type a . Hence, there is a monomorphismh : EB{GIG[a]:a < T} ~ EB{C/C[a] :a < T},inducedby hex + G[a]) = x + C[a], with pure image. Define g : GIG*[T] ~ CIC*[T] by g(x + G*[T]) = X + C*[T]. Then bfo = fcg , where Ic : CIC*[T] ~ EB{CIC[a] :a < T} is a monomorphism. By Example 3.1.10(a) , the image of fc is a pure subgroupofEB{CIC[a] :a < T}. It now suffices to show that if B is the image of g , then (B)*I B is bounded. In this case, since H = image fa , (H) *I H would also be bounded, as desired. There is a commutative diagram with exact rows and columns:
o
o o~
o~ o~
0 -l-
-l-
G*[T]/G[T]
~
-l-
C*[T]/C[T]
~
CIC[T]
-l-
g
-l-
-l-
o
~
-l-
GIG[T] GIG*[T]
0 -l-
~
CIC*[T]
-l0
A
~
0
~
0
~
0
-l-
~
K
-l-
~
L
-l0
with A a r-homogeneous completely decomposable group . Now, K[ T] = 0, since (C I C[ T])[T] = 0 by Lemma 3.1.9(a), and if A' is the pure subgroup of K generated by A, then A'(T) = A', since A(T) = A. By Proposition 3.1.15(b), A' is Thomogeneous completely decomposable. It follows that A'lA is bounded, whence the torsion subgroup of L is bounded, and since B = image g, (B)* I B must also be bounded, as desired. Corollary 3.2.6 Assume that G is a fin ite rank Butler group and T is the finite lattice of types generated by the typeset of G. (a) The typeset ofG is the meet closure ofthe critical typeset ofG.
94
3. ButlerGroups (b) The cotypeset of G is the join closure ofthe critical cotypeset of G. (c) The lattice T is the join closure ofthe typeset of G and the meet closure of the cotypeset of G.
PROOF. (a) The typeset of G is closed under finite meets by Lemma 3.1.3. Con-
versely, first assume that r is a maximal type in typeset G. Then r E critical typeset G . Next assume that r E typeset G, r ~ critical typeset G, and for each t < a E typeset G there is a subset Sa of the critical typeset of G with a = n{8 : 8 E Sa}. Since r ~ critical typeset G , G(r) = G#(r), whence G(r)jG*(r) is bounded. Consequently, there is x E G with type x = r and x = 1: {xu :a > r] for some Xu E G with Xu = 0 or type Xu = a. Then r = typex ~ n{typexu:a > r ] ~ n{a : a > r} ~ r , By induction on the length of a chain from r to a maximal type in typeset G, r = n{8 : 8 E S} for S = U{Su : a > r ], a subset of the critical typeset of G. Thus, typeset G is contained in the meet closure of the critical typeset ofG. (b) The proof is analogous to that of (a); just use induction on the length of a chain from an element of the cotypeset of G to a minimal element of the cotypeset ofG . (c) The join closure of the typeset of G and the meet closure of the cotypeset of G is contained in T . Conversely, the lattice of all types is a distributive lattice. It follows that elements of T are joins of meets of elements of the typeset of G . But the typeset of G is closed under meets by Lemma 3.1.3. Hence, T is the join closure of the typeset of G . Analogously, T is the meet closure of the cotypeset of G, since the cotypeset of G is closed under joins by Lemma 3.1.9(e). As a consequence of the following corollary, homogeneous Butler groups are completely decomposable. Corollary 3.2.7 Assume that Gis afinite rank Butler group.
= critical typeset G and G is completely decomposable. (b) If the critical cotypeset of G is linearly ordered, then cotypeset G = critical cotypeset G and G is completely decomposable. (c) Homogeneous Butler groups are completely decomposable. (d) If all rank-l torsion-free quotients of G are isomorphic. then G is a homogeneous completely decomposable group. (a)
If the critical typeset of G is linearly ordered, then typeset G
PROOF. (a) If the critical typeset is linearly ordered, then the critical typeset is equal
to the typeset by Corollary 3.2.6(a). Hence, the typeset of G is linearly ordered. Write typeset G = {ro < ... < rn}. Then G = G(ro) = Go EB G#(ro) with Go ro-homogeneous completely decomposable by Theorem 3.2.5(b) . Moreover, G#(ro) = G(rl), and typeset G(rl) = {rl < . . . < tn }. An induction on n shows that G(rl) is completely decomposable, noting that G(rn) is completely decomposable. Then G must be completely decomposable.
3.2 Characterizations of Finite Rank Groups
95
(b) The proof is analogous to that of (a) using Corollary 3.2.6(b) and Theorem 3.2.5(c). Assertions (c) and (d) are consequences of (a) and (b), respectively. Radicals and socles are intimately related for finite rank Butler groups . Proposition 3.2.8 [Lady 79] IfG is afinite rank Butler group and rand 0 are types, then (a) (b) (c) (d)
G[r] = (G(a) : a1:. r), andG*[r] = (G(a): aj:. r)* ; G(r) = n{G[a] : at r} and G#(r) = n{G[a]: air}; G#(r) £; G[r] and G(r) £; G*[r]; and G[r no] = (G[r] + G[a])* .
PROOF. (a) Define H = (G(a) :a1:.r)* . Then H £; G[r], since if f:G -+ Q with type f(G) :::: r , and a 1:. r , then f(G(a)) = 0; otherwise, a = typex :::: type f(x) :::: r , By Lemma 3.2.3(a), there is a completely decomposable group C and a balanced epimorphism g: C -+ G. Now, g(C/C[r]) £; G/G[r]' C[r] = ~{C(a) : af.:. r} by Examples 3.1.5 and 3.1.10, and g : C/C[r] -+ G/ H is onto, noting that g(C[ r]) £; H , since g(C(a)) £; G(a) for each a . Then OT(G / H) .s OT(C/C[r]) :::: r by Example 3.1.10(a). Hence, (G/H)[r] = 0 by Proposition 3.1.13(b). But G[ r]/ H £; (G / H)[r] , as a consequence of Lemma 3.1.9(b), and so H = G[r]. A similar argument shows that G*[r] = (G(a): a f:. r)*. (b) Observe that G(r) £; n{G[a] :a t r} , This holds because if x E G(r) and f : G -+ Q with type f(G) = at r, then f(x) = 0, otherwise, a = type f( x) ::: typex ::: r . Conversely, G is a pure cobalanced subgroup of a completely decomposable group C = XI E9 . . . E9 K; with rank Xi = 1 for each i by Lemma 3.2.3(b) . Thus, n{G[a] :a t r} £; n{C[a]:a t r} = C(r), as a consequence of Example 3.I.I0(a). Since n{G[a] : at r} ~ G n C(r) = G(r), the proof is complete. Analogously, G*(r) = n{G[a] : air}. Assertion (c) follows from (a) and (b), and (d) follows from (a) and Lemma 3.1.4(b). Corollary 3.2.9 Assume that Gis afinite rank Butler group. JfG has no rank-l summands, then: (a) the critical typeset of G has neither a least nor a greatest element; and (b) the critical cotypeset ofG has neither a least nor a greatest element.
PROOF. (a) Suppose r E critical typeset G with r :::: a for each a E critical typeset G . By Theorem 3.2.5(b), G = G(r) = G r E9 G#(r) with G; a nonzero r-homogeneous completely decomposable group . This is a contradiction to the assumption that G has no rank-I summands. Next suppose r E critical typeset G with r ::: a for each a in critical typeset G. In view of Corollary 3.2.6(a), Proposition 3.1.13(b), and Proposition 3.2.8(a), G[ r] = o and G*[r] =1= O. Then G*[r]/G[r] = G*[r] is a nonzero r-homogeneous
96
3. Butler Groups
completely decomposable summand of GIG[r] a contradiction. (b) The proof is analogous to that of (a).
= G, by Theorem 3.2.5(c), again
Socles and radicals can be used to identify completely decomposable groups among finite rank Butler groups. If G is a finite rank Butler group and r is a type , then G(r)/G#(r) and G#[r]IG[r] are both r-homogeneous completely decomposable. By Proposition 3.2.8(c), inclusion induces a homomorphism
(a) G isacompletelydecomposablegroupifandonlyifG*[r] = G(r)+G[r] for each type t ; and (b) G is quasi-isomorphic to a completely decomposable group if and only if G*[r]/(G(r) + G[r]) is boundedfor each type r . PROOF. If C is completely decomposable, then C* [r 1 = C( r) + C[ r] for each type r by Examples 3.1.5 and 3.1.10. Moreover, if C is a subgroup of G with GIC bounded, then G(r)/C(r), G*(r)/C*(r), G[r]/C[r], and G*[r]/C*[r] are all bounded. It now follows that G*[r]/(G(r) + G[r]) is bounded. Conversely, assume that G*[r]/(G(r) + G[r]) is bounded for each type r . In view of the hypotheses and Theorem 3.2.5, if r is a minimal element of critical cotypeset G, then G*[r] = G and G/(G(r) + G[rJ) = G/(G r $ G#(r) + G[r)) = G/(G r $ G[r]) is bounded for some nonzero r-homogeneous completely decomposable group Gr . If H = G[r], then rank H < rank G and H*[a]/(H(a)+ H[a]) is also bounded for each type a as a consequence of Proposition 3.2.8. By induction on the rank of G , G[ r] is quasi-i somorphic to a completely decomposable group . Hence, G is also quasi-isomorphic to a completely decomposable group. This proves (b). As for (a), the same argument shows that if, in addition, G*[r]/(G(r)+ G[ r]) = 0, then G is completely decomposable. Corollary 3.2.11 (a) Given a type r, there is afunctor F; from the category offinite rank Butler groupsto the categoryofabeliangroupsdefinedby Fr( G) = G*[r ]/( G( r)+ G[r]). (b) A finite rank Butler group G is completely decomposable if and only if Fr(G) = for each type r . Moreover, G is quasi-isomorphicto a completely decomposable group ifand only if Fr(G) is a bounded groupfor each r. (c) For each type r and finite rank Butler group G, image
°
maximal rank.
3.2 Characterizations of Finite Rank Groups
97
(d) A finite rank Butler group G has no rank-l quasi-summands if and only
if
G(r) is contained in G[r]for each type T. PROOF.
Exercise 3.2.4.
Let G be a finite rank Butler group. By Theorem 3.2.5(b), G(r) = G, ED G#(T) with Gr a homogeneous completely decomposable group of type T for each r in the critical typeset of G. Define a regulating subgroup of G to be B = 1:{Gr : r E critical typeset G}. Notice that B need not be unique, since the Gr 's are unique only up to isomorphism. The followingpropositionisdue to A. Maderand KJ. Krapf;see [Mutzbauer93]. Proposition 3.2.12 Let G be afinite rank Butler group.
(a) If B is a regulating subgroup ofG, then G/ B is bounded. (b) There are only finitely many regulatingsubgroups of G.
(a) Write B = 1:{Gr : T E critical typeset G}. The proof is by induction on the length of a chain from a type to a maximal type in the typeset of G. If r is maximal in typeset G, then G(r) = Gr. So assume that r is not maximal and G(a) / 1:{G8 : 8 ~ a} is bounded for each T < a E typeset G. Then G(r) = G; ED G#(T), G#(r)/G*(r) is bounded, and G*(T)/1:{G 8 : 8 > T} is bounded by induction. Hence, G(T)/1:{G u:a ~ T} is finite for each r E typeset G. Let T = IT(G). Then G = G(T) and G/ B is finite. (b) For each Tin typeset G, choose n(r) with n(T)(G#(T)/ G*(T» = 0. It suffices to show that if e = lcmlnt r): r E typeset G} , d is the length of a longest chain in typeset G, and B = 1:{Gr : r E critical typeset G} is a regulating subgroup of G, then ed G £ B £ G. In this case, there are only finitely many regulating subgroups, since G/ ed G is bounded, hence finite by Exercise 2.1.1. For =f:. x in G, let d(x) denote the length of a longest chain in typeset G from type x to a maximal type. It suffices to prove that ed(x)x E B. If d(x) = 0, then type x is maximal, G#(typex) = 0, and x E G, 5; B for T = typex. Now assume that d(x) > and assume, by way of induction, that ed(Y)y E B for each y in G with type y > typex = T. Then G(T) = G, ED G#(T) and x = c + y for some c in G r and y E G#(T). If type y > typex , then d(y) < d(x) and ed(x)x = ed(x)c + ed(x)y E B. Finally,assume T = typex = typeyandd(y) = d(x).Inviewofthechoiceofe , ey = 1:{yu . type y, = a> T} E G*(r).Byinduction,ed(y)y = :E{ed(y)-IYu:a > T} E B, since d(yu) < d(y) for each a. But d(y) = d(x) , so that ed(x)x = ed(x)c + ed(Y)y E B, as desired.
PROOF.
°
°
Givena Butler group G,definethe regulatorsubgroupR(G) to be theintersection of all regulatingsubgroupsof G. A finitepartially orderedset S is an invertedforest if for each s in S, {T E S: T ~ s in S} is linearly ordered. Part (c) of the next
98
3. Butler Groups
corollary is an unpublished result of O. Mutzbauer, wherein an inverted forest is called a V-tree.
Corollary 3.2.13 Let G be afinite rank Butler group. (a) The group G/R(G) isfinite. (b) If f : G ~ H is an isomorph ism, then f(R(G)) = R(H) and G / R(G) is isomorphic to H / R(H). (c) If the critical typeset ofG is an invertedforest, then R(G) ~{G(r) : r E crit ical typeset G} is a unique regulating subgroup of G, and G(r) is completely decomposable with linearly ordered typesetfor each r in the critical typeset of G.
=
PROOF. Assertion (a) follows from Proposition 3.2.12, since G has only finite many regulating subgroups B and each G / B is bounded, hence finite . (b) Notice that f(B) is a regulating subgroup of H ifand only if B is a regulating subgroup of G. This follows from the fact that if f is an isomorphism, then f(G(r)) f(G r) EB f(G#(r)) f(G r) EB H#(r). Hence, f(R(G)) R(H), since the regulator subgroup is the intersection of all the regulating subgroups. (c) Let B = ~ { G r : r E critical typeset G} be a regulating subgroup of G . Then the typeset of G (r) is a chain for each r in the critical typeset by hypothesis. Hence, G(r) is completely decomposable by Corollary 3.2.7(a), and G(r) = EB{G u : a ~ r ] by Example 3.1.5(a). Thus, G(r) S; B for each r in the critical typeset of G. Since B S; B' = ~{G(r) : r E critical typeset G}, B = B' must be a unique regulating subgroup of G.
=
=
=
The following proposition is a "dual" of the definition of a regulating subgroup and Proposition 3.2 .12(a).
Proposition 3.2.14 Let G be a finite rank Butler group. For each type r in the critical cotypeset of G, choose an H with G/G[r] = (G*[r]/G[r]) EB H and let fr : G/G[r] ~ G*[r]/G[r] be a projection with ker fr = H . Then f = EB fr : G ~ EB{ G*[ r]/ G[ r] : r E critical cotypeset G} is a monomorphism with (f(G)}*/f(G) bounded, hence finite . PROOF. Exercise 3.2.3. This section concludes with another characterization of finite rank Butler groups that will be used in Theorem 3.4 .6. Let G be a torsion-free abelian group of finite rank. Given a nonempty subset S of Il , the set of primes of Z, define G s n{G p : pES} . Observe that G n{G p ; pEn} is a subgroup of Gs .
=
=
Proposition 3.2.15 [Bican 70] Let G be a torsion-free abelian group offinite rank. Then G is a finite rank Butler group if and only if there is a partition Il = U{S(i) : 1 ::: i ::: n} such that each G S(i ) is a completely decomposable group with linearly ordered typeset.
3.2 Characterizations of Finite Rank Groups
99
PROOF. Assume that G is a finite rank Butler group and write G = XI +...+ X; for pure rank-I subgroups Xi of G . Choose 0 #- Xi E X i ' Given a permutation a of n+ = {I, 2, . .. , n}, define 5(0') to be the set of primes p with p-height xO'(I) ::: . . . ::: p-height xO'(n). Then Il = U{5(O'): a permutation of n+}, since for each prime p there is a permutation a of n+ with p-height xO'(I) ::: .. . ::: p-height xO'(n). Now, GS(O') = (X ])S(O')+...+ (Xn)s(O') has a linearly ordered typeset, since type XO'(I) ::: . . . ::: typexO'(n). But G S(O') is a finite rank Butler group , so that G S(O') is completely decomposable by Corollary 3.2.7(a). To complete the proof, choose subsets 5'(0') of 5(0') such thatthe 5'(O')'s partition Il, and observe that each Gs'(O') is completely decomposable with linearly ordered typeset for the same reason that GS(O') is. Conversely, assume that Il = U{5(i) : I ~ i ~ n} is a partition such that each G S(i) is a completely decomposable group with linearly ordered typeset. Define a homomorphism f: G ~ EB{G S(i) : I ~ i ~ n} by f(x) = (x , .. . , x). Since Il = U{5(i) : I ~ i ~ n}, G = n{GS(i) : I ~ i ~ n} is a pure subgroup of the completely decomposable group EB{ G S(i) : I ~ i ~ n} . EXERCISES
1. Suppose G is a finite rank Butler group, C = A I ED . . . ED An with rank Ai = 1 for each i , and f : C --+ G is an epimorphi sm with f(A i) i 0 for each i, For each nonempty subset S of (I , 2, . . . , n), define is = n{type Ai : i E S}. Prove that if X is a pure rank-l subgroup of G, then type X = U{rs : X n (~(f(Ai) : i E S}) i O} . 2. Prove that if G is a finite rank Butler group , then G is completely decomposable if and only if G[r] = ~{G (a) : aj r} for each r E cotypeset G. 3. Give a proof for Proposition 3.2.14. 4. Give a proof for Corollary 3.2.11. 5. Prove that if G is a r-homogeneous completely decomposable group and H is a ahomogeneous completely decomposable group, then G ~z H is a 8-homogeneous completely decomposable group , where if r has height sequence (m p) and a has height sequence (n p) , then 8 has height sequence (m p + n p) , agreeing that n + 00 = 00 for each 0 ~ n ~ 00. 6. A torsion-free abelian group G is locally completely decomposable if G p is completely decomposable for each prime p. Prove that a finite rank Butler group is locally completely decomposable. Give an example of a locally completely decomposable group of finite rank that is not a Butler group . 7. [Koehler 65] Let G be a torsion-free abelian group of finite rank with finite typeset. (a) Prove that if T is a nonempty subset of typeset G and a = n{r : rET), then there is a maximal independent subset B of G(a) such that each b E B has type a in G. (b) Show that there is a finite rank Butler group H such that H is a subgroup of G, G(r)j H(r) is torsion for each type r , and typeset H = typeset G. Conclude that H is unique up to quasi-equality. 8. Compute the cotypeset of the group B given in Example 3.2.1. Is B --+ X I ED X2 ED X3 a cobalanced monomorphism?
100
3. ButlerGroups
3.3 Quasi-isomorphism and Q-Representations of Posets There are equivalences between categories of finite rank Butler groups and repre sentations of finite posets over the field Q of rational numbers. These equivalences can be used to transfer properties of representations developed in Chapter I to the quasi-isomorphism category of finite rank Butler groups. If G is a finite rank Butler group, then typeset G is finite. Hence, the finite sublattice of the lattice of all types generated by typeset G is finite. Given a finite lattice T of types, define B(T) to be the category of finite rank Butler groups G with typeset G contained in T. Clearly, each finite rank Butler group is in B(T) for some T. As a consequence of the next lemma, B(T) is the smallest category containing all rank-l groups with types in T that is closed under isomorphism, finite direct sums, pure subgroups, and torsion-free homomorphic images.
Lemma 3.3.1 [Butler 65] Let T be a finite lattice of all types and G a finite rank Butler group. The following statements are equivalent: (a) G is in B(T); (b) G is a pure (cobalanced) subgroup ofa completely decomposable group in B(T); (c) G is a (balanced) homomorphic image ofa completely decomposable group in B(T).
(a) =? (c) By Lemma 3.2.3(a) , there is a balanced epimorphism C -+ G, where C is a direct sum of rank-I groups XI with type XI E typeset G £ T . Then typeset C £ T by Example 3.1.5(b) and the fact that T is closed under finite meets. This shows that C E B(T). (c) =? (b) In this case, G is a pure cobalanced subgroup of a finite direct sum C of rank-I quotients YI of G by Lemma 3.2.3(b). Then type YI E T, since T is closed under finite joins. Hence, typeset C £ T and C E B(T). (b) =? (a) follows from the fact that if G is a pure subgroup of C E B(T), then typeset G S; typeset C S; T .
PROOF.
Let T be a finite lattice of types with least element .0 and B(T)Q the quasiisomorphism category of B(T). Define JI(T) to be the poset of join irreducible elements of T , i.e., those. > .0 in T such that if . = a U 8, then either a = r or 8 = r. Notice that if.o < r E T, then a :s r for some a E lI(T), and if r E T\JI(T), then. = Uta E JI(T) :a < r} , Define ST to be JI(T)0P, the poset with elements those of lI(T) but with reverse ordering. In particular, if a, r EST, then a :s r in ST if and only if a ::: r as types .
Theorem 3.3.2 [Butler 68] Let T be afinite lattice oftypes. There is a category equivalence H: B(T)Q -+ rep(ST, Q) given by H(G) = (QG, QG(r) : r E ST)'
3.3 Quasi-isomorphism and Q-Representations of Posets
101
PROOF. The strategy is to define a fully faithful functor H : B(T)Q -+ rep(YOP, Q) by H(G) (QG, QG(r) : rET) and then prove that the image of H coincides with the image of the functor F-: rep(Sr , Q) -+ rep(YOP , Q) given in Proposition 1.3.I(b). Observe that H (G) E rep(YOP, Q), since if a and r are elements of T with a :::: r as types, then r :::: a as elements of TOP and G(r) S; G(a) by Lemma 3.1.4(a). For j E Hom(G, K) define H(f) : QG -+ QK to be the unique extension of j. Then H(f)(QG(r)) S; QH(r) for each r E Sr, since j(G(r» S; H(r) by Lemma 3.1.4(d). Hence, H(f) is a representation morphism. It follows that H is a faithful functor. To see that H is a full functor, let g : QG -+ QK be a representation morphism from H(G) to H(K). Since G is a finite rank Butler group , there are finitely many pure rank-l subgroups XI, . . . , X; of G with G = Xl + ...+ X n • Since g : QG(r) -+ QK(r)foreach type r , there is a nonzero integer m, with mig : Xi -+ K for each I :::: i :::: n . Let m mi '" m.: Then mg E Hom(G , K) with H(m!) = g , as desired. In order to confirm that image H is contained in the image of F- , let ro ::j:. rET . Then r = U{a E JI(T) : a :::: r in T}. As a consequence of Lemma 3.1.4(b), QG(r) = n{QG(a):a E JI(T),a :::: r in T}. Hence, (QG ,QG(a) :a E Sr) E rep(Sr, Q) with H(G) F-(QG, QG(a): a E Sr). It remains to prove that if U (Uo , U, : r E TOP) E rep(YOP , Q) is in the image of F-, then H(G) = U for some G E B(T). Observe that U is in the image of F- if and only if U r Uu = n for each a, rET . Choose a free subgroup Vo of Uo with QVo = Uo and let Vr = Vo n U; for Ur ' each rET . Each Vr is a pure subgroup, hence a summand of Yo, and Q Vr For each r E Sr , let X r be a subgroup of Q with type X r = r and let X0 be a subgroup of Q with type Xo roo Define
=
=
=
=
u, u,
=
=
G = Xo Vo
=
+ E{X r Vr : rEST},
=
a subgroup of Q Vo Uo with QG U«. There exist a completely decomposable group C , a finite direct sum of groups isomorphic to X o or X r for rEST S; T, and an epimorphism n : C -+ G as a consequence of the definition of G . By Lemma 3.3.1 , G E B(T). The proof is concluded by showing that QG(r) = U; for each rET. Now, tj, = QX r Vr S; QG(r) for each r E Sr, since type X r = r , Vr is a free group, and X r Vr is r-homogeneous completely decomposable. Because U is in the image of F- and the elements of Sr are the join irreducible elements of T, Uris contained in QG(r) for each rET. As for the reverse inclusion, let X be a pure rank-I subgroup of G with type X = rET. Then rr-I(X) is a Butler group, being a pure subgroup of C. Write rr-I(X) = YI +...+ Yn with each Yi a pure rank-I subgroup ofC and rr(Yi)::j:. O. Now, n restricts to an epimorphism YI EB · .. EB Yn -+ X and r = type X = a (l) U .. . U a(n), where aU) = type Yi E T . Then QX = Qrr(Yi) S; QG(ai) S; UU(i) for each i . Consequently, QX ~ U, = n{Uu(i) : I .::: i .::: n}, since U is in
102
3. Butler Groups
the image of F-. Therefore, QG(T) is contained in V T for each type T E T , as desired. Theorem 3.3.2 gives an explicit method of constructing groups in B(T) from representations in rep(ST , Q) as illustrated by the following example: Example 3.3.3 If T is a finite lattice, and ST = {type X I , type X 2, type X3} is an antichain with each Xi a subgroup ofQ and Xo = Xi n X j for each i =f. j, then a strongly indecomposable group in B(T) is quasi-isomorphic to one of the following rank-l groups with types in T, Xo, Xl , X 2 , X 3, XI
+ X 2 , X 2 + X 3, XI + X 3, XI + X 2 + X3 ,
or a rank-2 group
In each case, the endomorphism ring is isomorphic to a subring ofQ. PROOF. By Example 1.1.5, the indecomposable representations in rep(S3, Q) are (Q, 0, 0 , 0), (Q , Q, 0, 0), (Q, 0, Q, 0), (Q, 0, 0, Q), (Q, Q, Q , 0), (Q, 0 , Q , Q) , (Q , Q , 0, Q), (Q , Q, Q, Q), and (Q E9 Q, Q E9 0, E9 Q, (1 + l)Q). In each case, the representation endomorphism ring is isomorphic to Q. Now apply Theorem 3.3.2. For example, if V = (Vo, VI, V2 , V3) is an indecomposable representation with Q-dimension Vo = 1, then G = Xo + I:{Xi: Vi =f. O} is a rank-I group with type G = U{type Xi : Vi =f. OJ.
°
The category B(T)Q has finite, tame, or wild representation type as, respectively, rep(ST, Q) has finite, tame, or wild representation type . Since the functo r of Theorem 3.3.2 is a category equivalence, the representation type of B(T)Q is well-defined and is determined by the structure of ST.
Corollary 3.3.4 Suppose T is a finite lattice oftypes.
If B(T)Q has infinite representation type, then there are strongly indecomposable groups in B(T) ofarbitrarily large rank. (b) If B(T)Q has wild representation type, then for each finite-dimensional Qalgebra A, there is a G in B(T) with QEnd G isomorphic to A. (a)
PROOF. Assertion (a) is a consequence of Theorem 3.3.2, while (b) follows from Theorem 3.3.2, Proposition 1.4.5, and Corollary 1.4.3.
Corollary 3.3.5 Let T be afinite lattice oftypes. Then B(T)Q has: (a) finite representation type if and only if ST does not contain S4, (2, 2, 2), (1, 3, 3), (1, 2, 5), or (N, 4) as a subposet;
(b) wild representation type if and only if ST contains S5, (1, 1, 1,2), (2,2,3), (1,3,4), (N, 5), or (1, 2, 6) as a subposet;
3.3 Quasi-isomorphism and Q-Representations of Posets
103
if and only if ST contains S4, (2, 2, 2), (1, 3, 3), (1,2,5), or (N, 4) but does not contain S5, (1, 1, 1,2), (2,2,3), (1,3,4), (N, 5), or (1,2,6) as a subposet.
(c) tame representation type
PROOF. Apply Theorems 3.3.2, 1.3.6, 1.4.4, and 1.4.6. The tools of representation theory can be used to provide some striking results for finite rank Butler groups . Corollary 3.3.6 Suppose T is a finite lattice of types. Then B(T)Q has finite representation type if and only if End G is a subring of Q for each strongly indecomposable group G in B(T). In this case, each strongly indecomposable group G in B(T) has rank less than or equal to 6 and G is determined uniquely up to quasi-isomorphism by the vector (rank G, rank G(r) : r E JI(T)). PROOF. A consequence of Theorem 3.3.2, Corollary 1.3.9, and Theorem 6.2.7. The next corollary provides analogues of sincere representations and sincere posets. Corollary 3.3.7 Let T be afinite lattice oftypes. If B(T)Q hasfinite representation type, then there is a strongly indecomposable group G in B(T)Q with rank G(r)/G#(r) =I Ofor each rEST if and only if ST is one of
(1), (1,1), (1,1,1), (1,1 ,2), (1,2,2), (1,2,3), (1,2,4), (N,2), (N,3) , 4
3
6
8
3
~I I !X
5
4
7,
4
I
7
I
6,
6
8
~\I/I I
5
7
2
I 1
4
I
3
6
8
f/J~ 1
PROOF. Apply Theorems 3.3.2 and 1.3.7. Given a poset S with finite representation type, all ofthe indecomposable representations in rep(S, Q) can be found by listing the indecomposable representations for each sincere subposet of S. In view of Theorem 3.3.2, this same process carries over verbatim to B(T)Q.
Example 3.3.8 Let T be afinite lattice oftypes with ST = (l, 1,2) = {type XI, type X2, type X3 > type X 4 }, each Xi a subgroup of Q.
If G is a strongly
104
3. Butler Groups
indecomposable group in B(T), then G is quasi-isomorphic to a rank-I group with type in T or one ofthe rank-2 groups
+ X4 ) EB X 3 ) + (I + I)X 2 , (XI EB X 2 ) + (I + I)X 3 , (XI EB X2) + (I + l)X 4 .
«XI
In each case, End G is a subring ofQ. PROOF. A consequence of Theorems 3.3 .2 and 1.3.7 and Examples 1.3.8 and U.5 . The rank-2 groups arise from a sincere 2-dimensional representation of (I , 1,2) and Example 3.3.3 , noticing that S3 can be embedded as a subposet of (I, I, 2) in two different ways.
Example 3.3.9 Let 5
j \ 3 4 j \ j
T=
I 2 \ j
o
be a lattice of types. Then Sr = {I , 4 < 2}. Moreover, the rank-I groups with type in T are the only strongly indecomposable groups in B(T). PROOF. Exercise 3.3.3.
Example 3.3.10 Assume that T is a finite lattice of types and Sr = S4 = {type X I, type X2, type X 3 , type X 4 }. There are strongly indecomposable groups ofarbitrarily large finite rank in B(T). ln fact.for each irrreducible polynomial g(x) E Q[x] and integer e :::: I, there is a strongly indecomposable group G in B(T) with QEnd G isomorphic to Q[x]j(g(x)").
I. 1.6. Specifically, given an n x n Z-matrix A that is indecomposable as a Q- matrix,
PROOF. Apply Theorem 3.3 .2 and Example
G = X~ EB X~
+ (I + I)X 3+ (I + A)X~
is a strongly indecomposable group in B(T) with QEnd G isomorphic to Q[x]j (g(x)"), where g(x)" is the minimal polynomial of A.
3.4 Countable Groups
105
EXERCISES
1. Let T be a finite lattice of types. Prove that W(ST) ::::: 2 if and only ifeach group in B(T ) is quasi-isomorphic to a completely decomposable group. 2. State and prove Theorem 3.3.2 for modules overan arbitrary principal ideal domain. 3. Confirm Examples 3.3.8 and3.3.9. 4. Prove that if S is a finite partially ordered set, then there is a finite lattice of types T with ST = S. Show that the elements of T may bechosento be isomorphismclasses of subringsof Q.
3.4 Countable Groups A countable torsion-free abelian group G is a Butler group if each pure finite rank subgroup of G is a Butler group. Such groups are also called finit ely Butl er groups [Fuchs, Metelli 92]. An equivalent characterization of countable Butler groups in terms of balanced exact sequences is given in Theorem 3.4.6. A countable torsion-free abelian group G is completely decomposable if G is a direct sum of rank-I groups. Proposition 3.4.1 Assum e that G is a countable torsion -free abelian group. (a) If G is a countable pu re subgroup of a completely decomp osable group, then G is a Butl er group. (b) The group G is a Butler group ifand only ifG is the union of an ascending chain 0 S; B, S; . . . S; Bn S; B n+l S; . . . ofpure subgroups such that each B; is a finite rank Butler group. The Bn's may be chosen with rank Bn+d B; = 1. (c) If G is a hom ogeneous Butler group, then G is completely decomposabl e.
PROOF. (a) Suppose G is a pure subgroup of a completely decomposable group C and H a pure finite rank subgroup of G. Since H has finite rank, H is a pure subgroup of a finite rank completely decomposable summand of C, hence a finite rank Butler group. (b) Assume that G is a Butler group and {Xl ,"" xn, ... j is a set of generators of G. Define B; to be the pure subgroup of G generated by {Xl , ... , Xn j. Then B, S; .. . S; B; S; . . . is an ascending chain of pure finite rank subgroups of G with G = U{Bn : n ::: I}. Since G is a Butler group, each Bn is a Butler group. Moreover, Bn+ll B; S; Q( Xn+ l + Bn ), so that rank Bn+ll B; = 1. Conversely, suppose B, S; ... S; Bn ~ • . . is an ascending chain of pure subgroups of G with G = U{ B; : n ::: l} and each B; a finite rank Butler group. If H is a pure finite rank subgroup of G, then H is a pure subgroup of some Bn • To see this, let {Xl , . . . , x m } be a maximal Z-independent subset of H . Choose n
106
3. ButlerGroups
sufficiently large with each X i E Bn . Then H = (Xl, ••• , x n )* is a pure subgroup of Bn , whence H is a Butler group by Corollary 3.2.4. (c) By (b), there is an ascending chain BI ~ ... ~ Bn ~ • . • of pure subgroups of G with G = U{Bn : n ::: I} and B; a finite rank Butler group. Since G is homogeneous, B; is homogeneous, hence completely decomposable by Corollary 3.2.7(a) . Each Bn is a summand of Bn+1 by Lemma 3.1.6(a), say Bn+1 = Bn EB Cn+ 1 for each n ::: I. Define C I = B I . For each i , C, is a finite rank homogeneous completely decomposable group, by Lemma 3.1.6(c), and so G = EB{Ci : i ::: I} is a homogeneous completely decomposable group. In contrast to finite rank groups, the converse of Proposition 3.4.1(a) does not hold for countable groups.
Example 3.4.2 There is a countable Butler group that is not isomorphic to a pure subgroup ofa completelydecomposable group. PROOF. Let S be the set of words on the alphabet {0,1}. The empty word is denoted by 0 . For each s E S, choose a proper subgroup X s of Q such that if t = s I and u = sO, then X s = XI n Xu. In other words, the Xs's for s E S form a tree with root X 0 . In addition, choose the Xs's such that if r is a type and for each n there is s E S with length s ::: nand r ::: type Xs , then r = type Q. To see that such a choice is possible, let Jro = {TI}, Jrl = {Po , PI}, Jrz = {Poo , POI, PIO, PII }, . . . , x; = {P s : s E S , length s = n}, . .. be successive partitions of the set TI of all primes such that for each s E S, P, is infinite, P, is the disjoint union of Pso and PsI. and if s(1), . . . , s(n), . . . is a collection of elements of S with length s(i) = i, then n{ps(i) : I ~ i} is empty . For s E S, let X s = n{Zp : pEPs}. These Xs's satisfy the above conditions. Given a nonnegative integer n, define G; = EB{X s : s E S, length s = n}. There is a homomorphism fn: G« ~ Gn+1 with pure image defined by f(x s ) = (x s , x s ) E Xso EB X s), since X, = X so n X sI' Define G to be the direct limit of {Gn, fn : n ::: I}. Then G is a countable Butler group by Proposition 3.4.1(b) and is reduced, since each X s is a proper subgroup of Q. Assume, by way of contradiction, that G is a pure subgroup of a completely decomposable group C = CI EB Cz EB .. . EB Cn EB . . . with rank Cn = 1 for each n. Since G is reduced, G is a pure subgroup of C/d(C), so that it is sufficient to assume that each C; is reduced. Then X0 is a pure subgroup of C l EB · . . EB Cm for some fixed m. Let a i= X E X 0 and n ::: I. Then X = EB{xs : length s = n} E G; with each x, i= a by the choice of the Xs's. Moreover, X s = c(1, s) EB . .. EB c(k, s) with c(i , s) E C, for some k ::: m. Now, type x, = n{typec(i,s): 1 ~ i
~
k}
~
n{typec(i ,s) : 1 ~ i
~
m}
3.4 CountableGroups
107
and x = EI7{xs : length s = n} = EI7{I:{c(i, s): length s = n}: 1
=:: i =:: k}
E C I EI7 .. . EI7
Cm'
Since 0 "# x, there is some I =:: j =:: m with c(j, s) "# O. In summary, for each n ::: 1 there exist s E S with length s = n and some I =:: j =:: m with type Xs =:: type C]. By the choice of the Xs's , type C, = type Q, a contradiction. The next theorem gives a method of constructing countable Butler groups of infinite rank with finite typesets. The proof is a modification of the original proof that a torsion-free homomorphic image of a finite rank completely decomposable group is a Butler group given in [Butler 65], wherein R = Z. Let R be a ring with free additive group such that QR is a division ring. Then Z ~ R and a torsion-free R-module is also torsion-free as an abelian group. For a torsion-free R-module M, R-rank(M) is the cardinality of a maximal Rindependent subset of M, equivalently the QR-dimension ofQM. Let R p = ZpR, the localization of R at a prime p of Z. If R/ pR has no zero divisors, then each element of R p is of the form p iu for u a unit of R p' To see this, let 0 "# x E R p and write x = pi r for some r E R p\pR p. This is possible, since the additive group of R is free, hence preduced. Since QR p = Q R is a division ring, there are s E R p and j ::: 0 with r(s/pj) = I = (s/p j)r . Then rs = p j = sr. But R/pR = Rp/pR p has no zero divisors, so that s = pj u for some u in R p. Hence, ru = 1 = ur, and r is a unit of R p' as desired.
Theorem 3.4.3 Suppose R is a countable ring with free additive group, QR is a division ring, and R/ pR has no zero divisors for each prime p ofZ. lfG is a torsion-free R-module with G = RYI + ... + RYnfor rank-I subgroups Yi of G. then G is a countable Butler group with finite typeset. PROOF. It is sufficient, via Proposition 3.4.l(a), to prove that G is a pure subgroup of a countable completely decomposable group with finite typeset. The proof is by induction on the R -rank of G . If R-rank G = I, then G = RYI is a countable direct sum of copies of YI , since R is a countable free abelian group. Hence, G is a countable homogeneous completely decomposable group with type G = type YI. Next assume that R-rank G > 1 and let N, = (Q RYi ) n G, a pure subgroup of G containing R Yi • Define an R -homomorphism f : G -+ H = G / N I EI7 .. . EI7 G / Nn by f(x) = (x + Nv , . .. , x + N n ) . Since N, is a pure subgroup of G, G/Ni is a torsion-free abelian group that is generated by the rank-I groups (Y j + Ni ) / N, for j "# i as an R-module. Moreover, ker f = N I n·· · n N; = 0, since R-rank G > 1, so that at least two of the R Yi's are R -independent.
108
3. Butler Groups
The next step in the proof is to prove, by contradiction, that the image of f is a pure subgroup of H. If not, there is a prime p, Y E G, and Yi E G such that P(YI + NI,"" Yn + Nn) = (y + NI,···, Y + Nn) and Y - PYi = Xi E Ni\pNi for each i. Now, (Ni)p = RpXi since Xi (j. p N] , Yi is a rank-1 group, and QR is a division algebra. Thus, G pi pG p = (Rpl pRp)(Y + pG p) is cyclic, since + (Nn)p = RpXl + ... + Rpxn, and Y - Xi E pG for each i. G p = (Nd p + Let 5 = {Xl, , x m } be a maximal Rp-independent subset of {XI, ... , x n} with m = Rp-rank G p = R-rank G > 1 and Y = XI. Given m + 1 .::: i .::: n, there is some r in Rp with r x, in K = RpXI ED·· · ED Rpxm , since {Xlt ... , x m } is a maximal Rp-independent subset of {Xl, .. . , Xn}, a set of Rp-generators for G r: Write r = pj u for some unit u of Rp. Then p! UX; E K ~ G p, whence p! Xi E K , since u is a unit. Consequently, there is some e with p' Xi E K for each i, and so peG p = pe(RpXI + ... + Rpxn) ~ K. Thus, K = G p, since GplpeGp is generated by Y + peG p and Y = Xl E K. In particular, Kl p K = GplpGp is cyclic, which contradicts K I pK = (Rpl pRp)xI ED·· 'ED(RplpRp)xmwithm > 1. By induction on the R -rank of G, each GIN; is a pure subgroup of a countable completely decomposable group with finite typeset. This shows that H, and consequently G, is a pure subgroup of a countable completely decomposable group with finite typeset. Let T be a finite lattice of types with 5 r = {Tl, T2} = 52. As a consequence of Theorem 3.3.2 and Example 1.1.3, each strongly indecomposable group in B(T) has rank 1. Hence, each group in B(T) is quasi-isomorphic to a completely decomposable group. The next example demonstrates that countable Butler groups with typeset contained in T are much more complicated. Example 3.4.4 Assume that T2 = {type XI n X 2, type XI , type X 2}, Xi is a subgroup of Q containing 1 with type X I and type X2 incomparable, p is a prime with lip (j. Xi , and R = End X, for i = 1,2. There is a strongly indecomposable Butler group G with countably infinite rank such that the typeset of G is contained in T2 and End G is isomorphic to R. PROOF. The first step is to construct V = (Vo, VI , V2) E Rep(52, R) with Vo a countable free R-module, each Vi a pure submodule of Vo, Vo/(VI ED V2) a p-group, and End V = (f: V o --+ U«: f(V i) ~ V;} isomorphic to R. A slight modification of the proof of Theorem 2.4.6 can be used to find a torsionfree R -module M of countably infinite rank containing a free R -submodule L with MIL a p-group, End M = R, and Hom(M, R) = O. Specifically, let {Jri : i 2: l} be a countably infinite subset of the p-adic integers R* that are algebraically independent over Q, L a free R-submodule of R* with basis {I, Jri: i 2: I}, and M the pure subgroup of R* generated by L. Then M is a torsion-free Rmodule with countably infinite rank, and MIL is a p-group. Since each f E End M extends uniquely to multiplication by an element of R* and {zr, : i 2: I} is an algebraically independent set over Q, it follows that End M = Rand Hom(M, R) = O.
°
3.4 Countable Groups
°
109
Let 4- VI 4- Vo 4- M 4- be a countable free R-resolution of M . Then Vo is a countably free R-module, and VI is a pure free R-submodule of Vo, since M is torsion-free, Vo is a free module, and R is a principal ideal domain . Let VI $ L' be the preimage of Lin Vowith L' isomorphic to the free R-module L , {e; : i ::: I} a basis of L', and ((n(i) , u(i» : i ::: I} an indexing of the countable set N x VI, N the set of natural numbers. Define Vz to be the pure R -submodule of Vo generated by {pn(i)e; +u(i): i ::: I}. Then VI nvz = 0, since L'nV 1 = 0, and Vo/(V I $Vz) is a p-group, since MIL is a p-group. To see that End V is isomorphic to R, let f E End V . There is r E R with g = f-r : U» 4- VI.Thisissobecausef : U» 4- V o, f (V I ) S; VI, M = VoIVJ, and End M = R. Assume, by way of contradiction, that g =f:. 0. Then g(Vd =f:. 0, since V, is a free R-module and Hom(M , R) = 0. But Vo is p -reduced so that =f:. g(pn(; )e; + u(i» E g(Vz) for some i. Hence, g(Vz) S; Uz n VI = 0, a contradiction. It follows that each endomorphism of V is multiplication by an element of R, whence End V is isomorphic to R. Define G = X I V , $ XzUz + U» . an R -submodule of QVo = QV I $ QVz. A routine computation, using the assumptions that V; is pure in Vo and lip fj. X;, shows that if r, = typeX}, then G(r;) = X iV;. Moreover, G is a Butler group, since if H is a pure finite rank subgroup of G, then H = X 1 VI $ X Z Vz + Vo, with each V; a finite rank free R-summand of V;. Hence, H is a Butler group by Corollary 3.2.4. Finally, since R = End X, = End x-, it follow s that R S; End G S; End V = R.
°
In the previous example, G I(G(rl) $ G(rz» is an unbounded group. However, if G is indecomposable with G/(G(r,) $ G(rz» bounded, then rank G ~ 2 as a consequence of the next theorem, the proof of which is not included. The finite rank case is proved in Corollary 5.1.12. Theorem 3.4.5 [Files , Gobel 98] If G is a countable Butler group with typeset G S; Tz = fro, rl , rz} and Gj(G(rd $ G(rz» is a bounded group , then G is a direct sum ofgroups of rank less than or equal to 2.
°
°
An exact sequence 4- K 4- H ~ G 4- of abelian groups with G torsionfree is balanced if rank-l torsion-free abelian groups are projective with respect to this sequence, i.e., if X is a rank-l torsion-free abelian group and g: X 4- G, there is h: X 4- G with fh = g. If H is also torsion-free, then this definition of balanced coincides with that of Section 3.2 for exact sequences of finite rank torsion-free abelian groups (Exercise 3.4.3).
°
Theorem 3.4.6 [Bican, Salce 83] IfG is a countable torsion-free abelian group, then G is a Butler group if and only if each balanced exact sequence 4- T 4H 4- G 4- of abelian groups with T a torsion group is split exact.
°
PROOF.
[Mines, Vinsonhaler 92] Assume that G is a Butler group and
04-T4-H~G4-0
110
3. Butler Groups
is a balanced exact sequence with T a torsion group. By Proposition 3.4.1(b), G is the union of an ascending chain 0 ~ G I ~ . . . ~ G n ~ G n+ I ~ . .. of pure subgroups such that each G, is a finite rank Butler group with rank(G i+l / G i) = 1. Assume, by way of induction, that f-I(G n) = T EB An. Let K = G n+l . By Proposition 3.2.15 , there is a partition IT = U{ S(i) : 1 :::: i :::: m} such that each KS(i) is a completely decomposable group with linearly ordered typeset and K is a pure subgroup of EB{Ks(i): 1 :::: i :::: mI. Since rank K/G n = 1, the partition may be refined to assume that KS(i) = (Gnh(i) EB DS(i) for some rank-I group D isomorphic to K/G n. The sequence 0 -+ TS(i) -+ U-I(K))s(i) -+ KS(i) = (Gn)s(i) EB DS(i) -+ 0 is balanced exact, hence split exact since KS(i) is completely decomposable (see Exercise 3.4.3) . Since T is a torsion group and the S(i)'s partition IT , T = EB{Ts(i) : I :::: i :::: m}. Use the composite of the homomorphisms (K) -+ EB{f-I(K)s(i) : 1 :::: i :::: m} -+ EB{TS(i): 1 :::: i :::: m} -+ T to see that f-I(K) = f-I(G n+d = TEBA n+1forsomeA n+1 2 An. By transfinite induction, H = TEBA for some A isomorphic to G. Conversely, assume that G is a countable torsion-free abelian group and each balanced exact sequence 0 -+ T -+ B -+ G -+ 0 with T a torsion group is split exact. Let {XI, . . . , Xi , .. .} be a maximal Z-independent subset of G . For each n, define F; = ZXI EB EB Zxn. Enumerate all pure rank-I subgroups X of G with X n Fn =f:. 0 by AnI , , Ani, ... by listing all linear combinations al Xl +- . -+anx n with ai E Z and gcd{al, . . . , an} = 1 and taking the pure subgroup of G generated by each such linear combination. Define F = E9{ZXi : I :::: i), M n •i = F; EB An.1 EB · .. EBAn.i and M = U{Mn.i : 1 :::: i, n}. Then M = FE9(E9{A n.i : 1 :::: i, n}). There is an exact sequence 0 -+ K -+ M -+ G -+ 0 with K = ker f and f : M -+ G induced by inclusion. Since each pure rank-I subgroup X of G with X n Fn =f:. 0 is equal to A n•i for some nand i, f is onto and, the sequence is balanced. For each nand i , there is an induced exact sequence 0-+ K n Mn.i -+ Mn.i -+ G n.i -+ O. Notice that each Gn.i is a finite rank Butler group, since Mn .i is a finite rank completely decomposable group . Furthermore, G I ~ ... ~ G; ~ .. . is an ascending chain of pure subgroups of G , where G n = U{ G n.i : i :::: 1), and Gn.i = Gn •i - l + A n.i = An.1 + ...+ An.i. The next step in the proof is to define a set of elements p(n, i) that are either 1 or a prime and aset of positive exponents e(n , i) . If Gn,;-I = G n.i , define pen, i) = 1 = e(n , i) . If Gn.i-I =f:. Gn,i, choose a prime p = pen, i) with (Gn,i- I)p =f:. (Gn.i)p and a least positive integer e = e(n, i) with pe(Gn,i_d ~ (Fn)p + d«G n.i-d p) ~ (Gn,i)p, where (Fn)p is a free Zp-moduleandd«Gn,i_l) p) is the divisible subgroup of (Gn.i-d p. Such a choice is possible because (An,j)p is isomorphic to either Zp or Q with (An.j)p n (Fn)p =f:. 0 for each j :::: i, and G n.i = An.1 + ...+ An,;. It is now sufficient to prove that given n, P; = {pen, i) : pen, ne(n.i) =f:. l} is finite . In this case, Gn.i-I =f:. G n.i for at most finitely many i, and so G; = Gn,i is a finite rank Butler group for some sufficiently large i. An application of Proposition 3.4.1(b) shows that G is a Butler group. For each i, n :::: 1, define mn,i = lcm{p(n , iie(n.j) : 1 :::: j :::: i} and L = ~{mn .i(K n Mn.i) : 1 s i, n}. Then K/ L is torsion, since K = U{K n Mn ,i : 1 :::: i, n} , and the induced sequence 0 -+ K / L -+ M/ L -+ G -+ 0 is a balanced exact
r:'
3.5
Quasi-Generic Group s
III
sequence. By hypothesis, this sequence is split exact, so choo se g : G --+ M / L with fg = IG· Because fg = IG, g(Xi) - (Xi + L ) E ker f = K/L for each i . Since KIL is torsion and {XI , " " x n } is a basis of Fn , there is a nonzero integer m with m (g(y ) - (y + L )) = 0 E K/L for each y E Fn. In this case, h c(my ) = h M/ dm y + L ), where hG(m y ) denote s the p-height of my in G . In view of the fact that L = b {mn,i(K n Mn,i ): I s i, n} and mn,i = lcm{p(n , i)ZE(n,j) : 1 ~ j s i }, it suffices to assume that all prime divisors of m are in Pn . It remai ns to show that each prime p E Pn divides m. In this case , P; must be finite, as desired . Let 1 =f. P E Pn. Then (Gn,i-d p =f. (Gn,i)p so there exist X E A n.i with px E G n.i - 1 and ph+l x = y + d E ( Fn)p + d «Gn ,i- d p) for some y E ( Fn)p and d E d «Gn,i-l) p) with h ~ e(n , i), h = hG(n ,i- l)(y ), and hG(n,i)(Xy ) ~ h + 1. Then hM/dm y + L ) = hG(n ,i)(my) ~ h'l (m ) + h + 1. Choo se a = al EB a2 E L with at E b{mn,j(K n Mn,j ) : j ~ i-I}, a2 E b{mn ,j(K n Mn,j) : j ~ i}, and hG(my + a) ~ h'l(m) + e(n, i) = h'l(m) + h + 1. By the definition of the mn,i's, h M(a2) ~ 2e(n, i ) > e(n, i) + h . It follows that peen , i ) divides m, as desired ,
EXERCISES
I. Let G and H be count able completely decompo sable groups. Prove that G and H are isomorphic if and only if rank G(r)/ G#(r) = rank H (r)/ H #(r ) for each type r . 2. Prove that if G is a countable completely decompo sable group and H is a summand of G , then H is a completely decomposable group. 3. Let 0 -l- K -l- H ~ G -l- 0 be an exact sequence of torsion-free abelian groups. Prove that f: H (r) -l- G (r) is onto for each type r if and only if for each rank-I torsion-free abelian group X and g : X -l- G, there is h : X -l- H with f h = g.
3.5 Quasi-Generic Groups There are countable Butler group s of infinite rank , called quasi-generic groups , that determine the representation type of quasi-homomorphism categories of finite rank Butler groups (Theorems 3.5.5 and 3.5.11). In fact, forthe tame representation type case, all strongly indecomposable finite rank Butler groups can be constructed from quasi-generic groups, Quasi-generic groups are related to generic representations (Section 1.5). However, the connection between countable Butler groups and count able Q- representations of finite posets is more complicated than the connec tion between finite rank Butler group s and finite-dimensional Q -representations given in Section 3.3. The first lemma con sists of examples of strongly indecomposable torsion-free abelian groups of countably infinite rank with finite typesets. For each critical poset ST and associated group G , in Lemma 3.5.1, (Q(x) G, Q(x) G( r): rEST ) is the repre sentation of ST given in Theorem 1.5.3 with k(x ) repla ced by Q [x ].
112
3. Butler Groups
Lemma 3.5.1 Let T be a finite lattice oftypes and ST the opposite ofthe poset ofjoin irreducible elements ofT. (a) For each of the following posets ST, there is a strongly indecomposable torsion-free abelian group G with countably infinite rank, typeset G ~ T,
and QEnd G = Q[x]. Write R for Z[x] and Ai for a subgroup of Q with 1 E Ai and type Ai = i. (i) ST = S4 G
= (AI R $ 0) + (0$ A2R) + «(1 + 1)A3R) + «I +x)A4R) ~ Q[xf. 246
I I I
(ii) ST = (2,2,2) = 1 3
5
+ «I + X)A2R $ A 2R) + «I + 1)A3R $ A4R $ 0) + (0 $ (1 + l)A sR) (1 + 1)A6R) ~ Q[x]3.
G = (0 $ 0 $ AIR)
+ (A 4R $ +(A6R $
0)
4 7
I I
3 6
I I
(iii) ST=(1,3,3)= G = «I
2
+ I)A I R $
+ (0 $
5
(1 + I)AI R) + (0 $ (l
+ I)A2R
$ 0)
A3R $ A 3R $ 0) + (0 $ A 4R $ A4R $ A4R)
+ (AsR E9 0 $ 0 $ 0) + (A6R $ 0 E9 0 $ A6R) + (A 7R E9 (l
+ x)A 7R $
A7R) ~ Q[X]4 .
8
I
7
1 3 (iv) ST G
~
NI
= (N, 4) = 2
4
5
= «(1 + 1)A2R $ (1 + l)A2R $O)+(AIR $ + (0 $ + (0 $
(l
+ I)A 4R $
0 $ 0)
+ (0 $
AIR $ AIR $ AIR $0)
A 3R $ A3R E9 (I
+ I)A3R)
0 $ 0 $ 0 $ AsR) + (A6R $ 0 $ 0 $ 0 $ A6R)
+ (A7R $ 0 $ 0 $ A7R $ A7R) + (AsR $ (I
+ x)AsR $
AsR $ AsR) ~ Q[x]s.
3.5 Quasi-Generic Groups
113
8
I
7
I
3 6
I I
2 5 (v) Sr=(l,2,5)=
I
4,
G = (0 + 1 + 1 + 0 + 0 + O)AI R + (0 + 0 + 1 + 0 + 0 + l)A,R
+ (0 + 0 + 0 + 1 + 1 + O)A, R + ((1 + l)A2R $ (1 + l)A2R $ 0 $ 0) + (A 3R $ A 3R $ A 3R $ A 3R $ 0 $ 0) + (0 $ 0 $ 0 $0 $ 0 $ A4R)+ (0 $ 0 $ 0$0$ AsR $ A sR)+(A6R $ 0 $ 0$ 0 $ A6R $ A6R) + (A 7R $0 $0$ A7R $ A7R $ A7R) + (AsR $ (1 (b)
+ x)AsR $
AsR $ AsR $ AsR) £; Q[xt
If n
:::: 2 is a positive integer, G is one of the groups constructed in (a) with typeset G £; T, and H = Gf Gx", then H is a strongly indecomposable group in B(T) with QEnd(H) = Q[x]j(x n }.
PROOF. (a) In each case, G is a torsion-free abelian group with countably infinite rank and Z[x] £; End G. It suffices to prove that End G is contained in Q[x], in which case G is strongly indecomposable, since QEnd G = Q[x] has no nontrivial idempotents. The proof is analogous to that of Theorem 1.5.3. As an illustration, let G be as defined in (i). A brief argument shows that if rj = type A j, then G(rd = A,Z[x] , G(r2) = A 2Z[x], G(r3) = A 3Z[x](1 , 1), and G(r4) = A4Z[x](1, x). Let f E End G. Since f : G(rj) -+ G(rj) for 1 s i s 3, f = (g, g) with g E End Q[x]. On the other hand, f : G(r4) -+ G(r4), so that g commutes with x . Thus, g E EndQ[xjQ[x] = Q[x] , as desired. (b) Since H is generated as a group by finitely many copies of the Aj's and type A j E Sr £; T , H E B(T) . The same argument as that for (a), shows that QEnd H = Q[x]j(x n } , and so H is strongly indecomposable. As an illustration, let Sr = S4 and R; = Z[x]j(x n }. In this case,
A torsion-free abelian group G is quasi -generic if G is strongly indecomposable with countably infinite rank and QG has finite length as a Q End G-module. This definition parallels that of a generic representation given in Section 1.5. The groups in Lemma 1 are not quasi-generic, since in each case, Q End G = Q[x] and QG has infinite length as a Q[x ]-module. This problem can be remedied
114
3. ButlerGroups
by replacing Z[x] with a principal ideal domain A such that QA = Q(x), a field. The ring A also satisfies the hypotheses of Theorem 3.4.3 and thereby establishes a method for constructing countable Butler groups with finite typesets that are quasi-generic. If f(x) E Z[x], the ring of polynomials with coefficients in Z , then the content of f, denoted by c(f), is the greatest common divisor of the coefficients of f . The polynomial f(x) is called a primitive polynomial if c(f) = 1. Since c(fg) = c(f)c(g), by Gauss's lemma [Hungerford, 74], the set S of primitive polynomials in Z[x] is a multiplicatively closed set. Define A = Z[x]s , the localization of Z[x] at S, a subring of the field of quotients Q(x) of Z[x] . Elements of A are of the form f(x)/g(x) with f(x), g(x) E Z[x] and g(x) a primitive polynomial. Lemma 3.5.2 The ring A has a countablefree additive group and is a principal ideal domain containing Z[x] withQA = Q(x) and 11./ pA afieldforeachprime pofZ. PROOF. (a) Observe that Z[x] is a subring of A and A is a subring of Q(x) with
QA contained in Q(x). On the other hand, if f(x)/g(x) E Q(x) with f(x), g(x) E Z[x], then g(x) = ah(x) with c(g) = a and hex) a primitive polynomial. Thus, f(x)/g(x) = (l/a)(f(x)/ hex»~ E QA. The additive group of A is countable and torsion-free. To prove that A is free as a group, it suffices, by Corollary 2.4.5, to show that each pure finite rank subgroup of A is free. Let H be a pure finite rank subgroup of A and suppose H is the pure subgroup of A generated by a finite number fl (x)/g, (x) , . .. , fn(x)/gn(x) of elements of A with each fi(X), gi(X) E Z[x] and gi(X) a primitive polynomial. Then g(x) = g, (x)· .. gn(x) is a primitive polynomial in Z[x] , since each gi(X) is a primitive polynomial, and c(g) = c(g,) . . . c(gn). Consequently, the subgroup of A generated by the fi(X)/gi(X)'S for 1 ::: i ::: n is a subgroup of B = (l/g(x»Z[x] . Observe that B is a free abelian group , being isomorphic to Z[x] . In fact, B is a pure subgroup of A. To see this, leta(x)/b(x) E A and n a nonzero integer with n(a(x)/b(x» = f(x)/g(x) E B. Then na(x)g(x) = b(x)f(x) E Z[x]. Taking the content of both sides yields nc(a) = c(f) E Z, recalling that c(g) = 1 = c(b), since g(x) and b(x) are assumed to be primitive polynomials. Thus, f(x) = np(x) for some p(x) E Z[x], and so a(x)/b(x) = p(x)/g(x) E B . Since B is a pure subgroup of A and each fi( X)/gi(X) E B, it follows that H is a subgroup of B. Because H is a subgroup of the free group B, H is also a free group, as desired. To see that A is a principal ideal domain , suppose that y = f(x)/g(x) is an element of A with f(x), g(x) E Z[x] and g(x) a primitive polynomial. Write f(x) = nh(x) for some primitive polynomial hex), where n = c(f). Then y = n(h(x)/g(x» with h(x)/g(x) a unit of A. It now follows that if I is a nonzero ideal of A, then I = nil., where n is the least positive integer such that nu E I for some unit u of A. Moreover, if p is a prime of Z , then 11./ p A is a field, since each yEA is an integer multiple of a unit of A.
3.5 Quasi-Generic Groups
115
Lemma 3.5.3 Let T be a finite lattice of types, G the group associated with ST as defined in Lemma 3.5.1, and A the ring given in Lemma 3.5.2. Then G' = GA is a quasi-generic group with End G' = A, QEnd G' = Q(x), and typeset G' 5; T. PROOF. In each case , G' is a countable Butler group with typeset contained in T, via Lemma 3.5.2 and Theorem 3.4.3. As in Lemma 3.5.1, A = End G' . Then QEnd G' = QA = Q(x) , and so G' is strongly indecomposable. Moreover, QG' is a finite-dimensional Q(x)-vector space, whence Q' has finite length as a QEnd G'-module and G' is a quasi-generic group. The quasi -generic groups G' given in Lemma 3.5.3 have some additional properties. Call a countable torsion-free abelian group G special if there are finitely many pure rank-l subgroups XI , " " X; of G such that Gf[(End G)X 1 + ... + (End G)X n ] is a bounded group. If G is a countable torsion-free abelian group with typeset G contained in a finite lattice T, then End UG = {f: QG ~ QG: f(QG("r» £ QG(r), rEST} is the representation endomorphism ring of UG = (QG, QG(r): rEST) E Rep(ST , Q) . Since each G(r) is fully invariant in G, QEnd G £ End UG • A special group G is central-special if there are finitely many pure rank-l subgroups Xi of G such that G/(CGX I + .. . + CGXn) is bounded as a group, where CG = (CEnd UG) n (End G) and CEnd UG is the center of End UG. Notice that CG is contained in CEnd (G), the center of End G. Each finite rank Butler group is central-special, being generated as a group by finitely many pure rank-l subgroups. For each countable Butler group G' in Lemma 3.5.3, End UG' = Q(x). Hence, G' is central-special, since CG' = Q(x) n End G' = End G' = A and G' is generated as a A-module by finitely many pure rank-l subgroups. Lemma 3.5.4 If G is a central-special group, then Q End G = End UG and CG = CEnd G. In particular; G is strongly indecomposable if and only if UG is indecomposable in Rep(ST, Q). PROOF. Since G is central-special, there are pure rank-l subgroups Xi of G such that G/(CGX I+ " , +CGXn)isboundedasagroup,saytG £ CGX I+" , +CGXn for some nonzero integer t. If f E End UG, then f(X i) 5; QG(ri), where type Xi = ri . Hence, there is a nonzero integer s with sf : Xi ~ G (ri) for each i. Now let x E G. Then tx = CIXI + . . . + Cn Xn for some c, in CG and Xi E Xi. Hence, tsf(x) = sf(tx) = sf(clx» + ... + sf(cnxn) = cl(sf(xl» + ... + cn(sf(xn E G, since sf E End UG and each c, E CG 5; CEnd UG. This shows that tsf E EndG, whence f E QEnd G and so End UG = QEnd G . Consequently, QEnd G has no nontrivial idempotents if and only if End UG does. In particular, G is a strongly indecomposable group if and only if UG is an indecomposable representation in Rep(ST , Q). Finally, CG = CEnd UG n End G= CQEnd G n End G = CEnd G .
»
116
3. ButlerGroups
For a finite lattice T of types, a torsion-free abelian group G is called a T-group if G is a countable Butler group with typeset G contained in T. The next theorem is the Butler group analogue of Theorem 1.5.3. Theorem 3.5.5 Let T be a finite lattice oftypes. The following statements are equivalent: (a) The category B(T)Q has finite representation type; (b) There are no central-special quasi-generic T -groups; (c) QEnd H= Qfor each strongly indecomposable H in B(T)Q. PROOF. (a) ~ (b) Suppose G is a central-special quasi-generic T-group. Then QEnd G = EndUe by Lemma 3.5.4. Now, UG = (QG,Q(r) :r E Sr) E Rep (Sr, Q) and rep(Sr, Q) has finite representation type in view of (a) and Theorem 3.3.2. By Theorem 1.5.3, UG is the direct sum of finite-dimensional Q-representations of Sr. This contradicts the fact that a quasi-generic group is strongly indecomposable with countably infinite rank. (b) ~ (a) Assume that B(T)Q has infinite representation type. By Corollary 3.3.5(a), ST contains one of the critical posets S4, (2, 2, 2), (I, 3, 3), (N, 4), or (1, 2, 5) as a subposet. For each of these critical posets Sr, there is a central-special quasi-generic T-group given by Lemma 3.5.3. This contradicts (b). (a) ~ (c) This follows from the category equivalence of B(T)Q with rep(Sr, Q) , Theorem 3.3.2, and the corresponding result for rep(Sr, Q), Corollary 1.3.9. (c) ~ (a) Suppose that B(T)Q has infinite representation type. By Corollary 3.3.5(a), it suffices to show that if 5r is one of the critical posets 54, (2, 2, 2), (1, 3, 3), (1, 2, 5), or (N, 4), then there is a strongly indecomposable H in B(T)Q with QEnd H not isomorph ic to Q. An application of Lemma 3.5.I(b) completes the proof.
The next two lemmas establish the framework for existence conditions for quasigeneric T -groups if B(T)Q has wild representation type. The groups listed in the next lemma are obtained from the list of Q-representations of the poset Sr given in Theorem 1.4.4. In each case, UG = (QG, QG(r) : E Sr) is the corresponding representation of ST . Lemma 3.5.6 Let T be a finite lattice oftypes. For each ofthe following posets Sr, there is a strongly indecomposable torsion-free T-group G with countably infinite rank and QEnd G = Q( x, y}. Let R = Z (x, y) and Ai a subgroup ofQ with I E Ai and type Ai = i.
5 (i) Sr
= S5 and(1,I,I, 1,2) = 1
2
3
I
4
3.5 Quasi-Generic Groups
117
G = (AIR EB AIR EBOEB 0) + (OEBOEB AzR EB AzR) +(1 +0+ 1 +0)A3REB(0+ 1 +0+ l)A3R) +(1 +O+x
+ I)A 4R
+«(1 + 0 + x + I)A sR EB (0 + 1 + Y + O)AsR). 5
I
246
I
(ii)
I
ST = (2,2, 3) = 1 3
G
I
7,
= (y+x +0+ 1 +0+ l)A IR EB(O+ 1 +
1 +0+ 1 +O)AIR
+ ((l + 0 + 1 + 0 + 0 + O)AzR EB (0 + 1 + 0 + 1 + 0 + O)AzR EB (y + x + 0 + 1 + 0 + l)AzR EB (0 + 1 + 1 + 0 + 1 + O)AzR) + (0 EB 0 EB 0 EB 0 EB A3R EB A 3R) + (0 EB 0 EB A4R EB A 4 R EB A4R EB A4R) + (A7R EBOEBOEBOEBOEBO) + (A6R EB A6R EB 0 EB 0 EB 0 EB 0) + (AsR EB AsR EB AsR EB AsR EB 0 EB 0). 8
I
4 7
I I
3 6 (iii)
ST=(l ,3 ,4)=1
I I
25,
G = «0 + 1 + 1 + 0 + 1 + 0 + 0 + O)AI R EB (y + x + 0 + 1 + 0 + 1 + 0 + O)AI R EB (1 + 0 + 1 + 0 + 0 + 0 + 1 + O)AI R EB(O+ 1+0+ 1+0+0+0+ l)A IR) + (0 EB 0 EB 0 EB 0 EB 0 EB 0 EB AzR EB AzR) + (0 EB 0 EB 0 EB 0 EB A 3R EB A3R EB A3R EB A3R) + (0 EB 0 EB A 4R EB A4R EB A4R EB A4R EB A4R EB A 4R)
+ (AsR EB 0 EB 0 EB 0 EB 0 EB 0 EB 0 EB 0) +
(A6R EB A6R EB A6R
EBO EB 0 EB 0 EB 0 EB 0) + (A 7R EB A7R EB A7R EB A7R EB A7R EB 0 EB 0 EB 0) + (AgR EB AgR EB AgR EB AgR EB AgR EB AgR EB AgR EB 0).
118
3. Butler Groups
9
I I 247 NI 1 3 6 I (iv) T = (N, 5) = 5 8
G
= (0$0$0$0$0$0$A1R$A1R$A1R$A1R) + «0 $ 0 $ A 2R $ A 2R $ 0 $ 0 $ A2R $ A2R $ A2R $ A2R) $ (0 + 0 + 0 + 0 + 1 + 0 + 0 + 0 + 1 + 0)A 2R $(0+0+0+0+0+ 1 +0+0+0+ l)A2R) +«0+0+0+0+ 1 +0+0+0+ 1 +0)A3R $(0+0+0+0+0+ 1 +0+0+0+ 1)A3R) + «0 + 1 + 1 + 0 + 1 + 0 + 0 + 0 + 0 + 0)A 4 R $(Y +x + 0+ 1 + 0 + 1 + 0+0+0+ 0)A4R
$(1 +0+ 1 +0+0+0+ 1 +0+0+0)A 4R $(0+ 1 +0+ 1 +0+0+0+ 1 +0+0)A4R $(0 + 0 + 0 + 0+ 1 + 0 + 0+0+ 1 + 0)A 4 R $(0+0+0+0+0+ 1 +0+0+0+ 1)A4R) +(A 5R $ 0 $ 0$0$ 0$ 0$0 $ 0 $ 0$ 0) + (A6R $ A6R $ A6R $ 0 $ 0 $ 0 $ 0 $ 0 $ 0 $ 0) +(A7R $ A7R $ A7R $ A7R $ A7R $0$0$0$0$0) +(AsR $ AsR $ AsR $ AsR $ AsR $ AsR $ AsR $ 0$0$ 0) +(A 9R $ A9R $ A9R $ A9R $ A 9R $ A 9R $ A9R $ A9R $0$0). 9
I
8
I
3 7
I I
126
I
5 (v) Sr
= (l, 2, 6) =
I
4
3.5 Quasi-Generic Groups G
119
= «0+0+0+ 1 +0+0+0+0+0+ 1 +0+ y)A)R $(0+0+ 1 +0+0+0+0+0+ I +0+ 1 +x)A)R $ (0 + 1 + 0 + 0 + 0 + 1 + 0 + 0 + 0 + I + 1 + O)A) R $(1 +0+0+0+ 1 +0+0+0+ I +0+0+ l)A)R $(0+0+0+0+0+0+0+ I +0+0+ 1 +O)A)R $ (0 + 0 + 0 + 0 + 0 + 0 + I + 0 + 0 + 0 + 0 + l)A) R) +(0 $ 0$ 0 $ 0$0 $0$ 0 $ 0 $ AzR $ AzR $ AzR $ AzR)
+(0$0$0$0$A3R $ A3R $ A3R $ A3R $ A3R $ A3R $A 3R $ A 3R)+(A 4R $ A4R $ 0$ 0$ 0 $ 0$ 0$ 0$ 0$ 0 $ 0 EB 0) + (AsR EB AsR $ AsR EB 0 $ 0 EB 0 $ 0 EB 0 EB 0 $ 0 $ 0 $ 0) + (A6R $ A6R $ A 6R $ A6R EB 0 $ 0 $ 0 $ 0 $ 0 EB 0 $ 0
EBO)+(A7R EB A7REB A7R EB A7R $ A7R $ A7R EB 0 $ 0 EB 0 EB 0 EB 0 $ 0) + (AgR EB AgR EB AgR EB AgR EB AgR $ AgR EB AgR $ AgR EB 0 $ 0 EB 0 $ 0) + (A 9R $ A9R EB A9R
$ A9R EB A9R EB A9R $ A9R $ A9R EB A9R EB A9R EB 0 EB 0). PROOF. In each case, G is a countable torsion -free group with typeset G S; T. It is sufficient to show that End G S; Q(x, y}, whence G is strongly indecomposable, since Q(x, y} = QEnd G has no nontrivial idempotents. The computations to show that End G S; Q(x, y} are as in Theorem 1.4.4. The idea is, using the invariance of the G(r;)'s and the fact that QlA;R = Ql(x, y) , to show that if f E EndG , then f E EndQQ(x, y} with fx = xf and yf = fy. In this case,
f
E
EndQ(x ,y)Q(x, y} = Q(x, y}.
A noncommutative version of Lemma 3.5.2 is needed to construct quasi-generic groups from the groups in Lemma 3.5.6 via Theorem 3.4.3.
Lemma 3.5.7 There is a ring l:1 with countablefree additive group containing Z( x , y} such that Ql:1 is a division algebra and l:1/ p S. has no zero divisorsfor each prime p ofZ. PROOF. Let R = Z[x, t, 2] be a skew polynomial ring over Z in two indeterminates x and t subject to the condition that tx = xt z . Induction arguments show that for positive integers m and n,
t nx=xt 2n , where rr(n)
= 2n - 1 +
tnxm=xmtUforu=2mn,
and
(xtt=x nt1f(nl,
2n - 2 + .. . + 2 2 + 2 + 1 if n ::: 2, rr(O) = 0, and rr(l)
= 1.
120
3. Butler Groups
It follows that a monomial xe(I)(xt)!(I) ... xe(k)(xt)!(k) is equal to Xe(I)+ !(I)+ ·+e(k)+!(klrlJ(f(l),e(2),...,e(k),j(k»,
where f3(f(1))
= JT(f(1)) = 2!(1)-1 + 2!(I)-2 + ... + 2 + 1
and
f3(f(1), e(2), . . . , e(k), f(k)) = 2!(k)+e(k)f3(f(l) , e(2) , . . . , e(k - I), f(k - 1)) + JT(f(k))
is defined recursively for k 2: 2. For example, with k = 2, f3(f(1), e(2), f(2)) = 2e(2)+!(2)JT(f(1)) + JT(f(2)).
Notice that f3(f(1), e(2) , . .. , e(k), f(k)) is a sum of powersof2. The next step is to show that Z(x , y) is isomorphic to the subring Z(x, xt) of R . To see this, view f3(f(1), e(2), ... , e(k), f(k)) as a 2-adic expansion of a natural number, i.e.,sumsofpowersof2 withcoefficients eitherO or I. Asa consequenceof the definition of f3(f(1), e(2), . . . , e(k), f(k)), this expansion has exactly k blocks of consecutive 1's as coefficients. Since 2-adic expansions of natural numbers are unique, k can be recapturedfrom f3(f(1) , e(2), ... , e(k) , f(k)). Now suppose that two monomialsxe(l)(xt)!(I) . . . xe(k)(xt)!(k) andx e'(I)(xt)f'(I) ... xe'(k')(xt)f'(k') are equal. From the precedingparagraph, f3(f(l) , e(2), ... , e(k), f(k)) = f3(f'(1), e'(2) , . . . , e'(k), f '(k))
and e(1)
+ f(1) + ...+ e(k) + f(k) = e'(1) + f'(1) + ...+ e'(k) + f '(k).
= k', e(i) = e'(i), and f(i) = f '(i) for each i, This shows that f : Z(x , y) -+ Z(x , xt) induced by f(x) = x and f(y) = xt is a ring iso-
Hence, k
morphism, as desired. Now, R is a right Ore domain, i.e., aR n bR #- (/) for each pair a, b of nonzero elements of R. This can be easily seen by observing that Z[t] is a right Ore domain and g : Z[t] -+ Z[t] induced by f(t) = t 2 is a one-to-one ring endomorphism. Hence, R has a right quotient ring D , a division algebra with each element of D of the form f(x , xt)g(x, xt)-I for f(x, xt), g(x , xt) E Rand g(x , xt) #- O. Note that !Q>(x, y) is isomorphicto !Q>(x, xt), a subringof D. Given g(x,xt) in R, define c(g(x,xt)) to be the gcd of the coefficients of g(x, xt). It is not difficult to confirmthat c(g(x, xt)f(x, xt)) = c(g(x, xt))c(f(x, xt)),just as in Gauss's theoremfor polynomials in onevariable. Let T = (g(x, xt) E R : c(g(x, xt)) = I}. Then T is a multiplicatively closedset. Hence, I:!,. = RT- 1 = (f(x, xt)g(x , xt)-I : f(x, xt) E R, g(x, xt) E T} is a subringof D . An argument
3.5 Quasi-Generic Groups
121
like that of Lemma 3.5.2 shows that QD. = D , the additive group of D., is a countablefree abeliangroup,and D. / PD. has no zerodivisorsforeachprime p of Z. The category B(T)Q is generically wild if there is a special quasi-generic Tgroup G such that QEnd G/ JQEnd G contains a copy of Q(x, y). Such a G cannot be central-special, since the center of Q(x , y) is Q and QG has infinite Q-dimension. The next theorem is a Butler group version of Theorem 1.5.4 for representations. Theorem 3.5.8 For the following statements, (a)
~
(b)
~
(c):
(a) B(T)Q has wild representation type ; (b) B(T)Q is generically wild; (c) B(T)Q is endowild.
If St n
~
= Sn, then
(a), (b), and (c) are all equivalent to the condition that
5.
(a) ~ (b) Assume that B(T)Q has wild representation type. By Corollary 3.3.5(b), s- contains one of the posets S5 , (1, I, 1,2), (2,2,3), (1, 3,4), (1,2,6), or (N , 5) as a subposet. By Lemma 3.5.7 there is a ring D. with countable free additive group such that QD. is a division algebra containing Q(x , y ) and D. / pI). has no zero divisors for each prime p of Z . Givena group G in Lemma3.5.6, define H = GD. . Then H is a countableButler groupof infiniterank withtypesetcontainedin T by Theorem3.4.3. Moreover, H is a special group from the definitionof G. Just as in Lemma 3.5.6,QEnd H = QD. is a division algebra containing Q(x, y). Thus, H is strongly indecomposable. Since H has finite dimension over QEnd H, H is a quasi-generic T-group. (b) ~ (c) Assume that B(T)Q is generically wild and let G be a special quasigeneric T -group withQEnd G/ J QEnd G a divisionQ-algebracontainingQ(x , y). Then Q(x, y) = QZ(x , y) is contained in QEnd G, so that Z(x, y) ~ End G. Since G is special, there are pure rank-l subgroups Xj of G with G/«End G)X) + . .. + (End G)X n ) bounded. Define H = Z(x, y)X) + ... + Z(x, y)X n • Then H is a countable torsion-free abelian group with typeset H contained in T and y ) ~ EndH. As a consequence of Example 1.1.7, for each finite-dimensional Q-algebra X, there are matrices A and B with X isomorphic to C(A , B). Suppose that A and B are two nonzero m x m Q-matrices with minimal polynomials f(xl and g(y)i, respectively, for irreducible polynomials f(x) E Q[x) and g(y) E Q[y). Let L = H/ K H, where K = (xy , f(xl, g(y) i) n Z(x , y) is an ideal of Z(x, y) . Then L E B(T)Q, since the Q-dimension ofQL is finite and H is generatedas a Z(x, y)-module by finitely manypure rank-I subgroupswithtypesin T. Moreover, C(A , B) is contained in QEn~(x . y)L. This is so because x acts on QH /QK H by A and y acts on QH /QK H by B . In particular, C(A, B) is a Q-subalgebra of QEnd L. It follows that B(T)Q is endowild.
PROOF.
z«.
122
3. Butler Groups
If ST = Sn and B(T)Q does not have wild representation type, then n ~ 4 by Corollary 3.3.5. Also , if G is a strongly indecomposable group in B(T)Q, then QlEnd G = End VG for VG an indecomposable in rep(Sn, Ql) by Theorem 3.3.2. However, each such End VG is a factor algebra of Ql[x] by Example 6.2.7. This contradicts (c). Hence , (c) => (a) if ST = S«. There is an interpretation of tame representation type of B(T)Q in terms of central-special quasi-generic T -groups . Suppose R is an integral domain with QlR = F, a field. A finite rank torsion-free R-module M is said to be afinite rank R-Butler module if there are finitely many pure rank-l subgroups Xi of M such that M/(RX I + ...+ RX n ) is bounded as a group. A finite rank R-Butler module M is strongly indecomposable if FEndRM / JFEndRM is a division F-algebra. These definitions directly generalize the definition of a finite rank Butler group. Quasi-generic Butler groups can be constructed from finite rank R-Butler modules via Theorem 3.4.3. An additional condition on the endomorphism ring yields a central-special quasi-generic group .
Lemma 3.5.9 Let R be an integral domain with countablyfree additive group of infinite rank such that p R is a prime ideal of R for each prime p of Z and QlR = F is a field. If G is a finite rank strongly indecomposable R-Butler module with R S; CG, then G is a central-special quasi-generic Butler group with End G = EndRG and QlEnd G/JQlEnd G a division F-algebra offinite F-dimension. By Theorem 3.4.3, G is a Butler group with countably infinite rank . Then G is a central-special group, since G is an R-Butler module and R S; C G . Hence, CG = CEnd G by Lemma 3.5.4 . Since G is a finite rank strongly indecomposable R-Butler module, QlEndRG / JQlEndRG = FEndRG/ J FEndRG is a division Falgebra. But QlEnd G = QlEndRG , since R S; CEnd G, and so QlEndG/ JQlEnd G is a division Q-algebra. Hence , G is strongly indecomposable, and QG has finite length as a QlEnd G-module. Since G has countably infinite rank, G is quasigeneric. PROOF.
For H E B(T)Q' define cdn H = (rank H, rank H(r) : rEST). Then cdn H = cdn V H, where V H = (Q H, QlH (r ) : rEST). The next lemma generalizes the construction of central-special quasi-generic groups given in Lemma 3.5.3.
Lemma 3.5.10 Suppose N = (No, N, : i E ST) is an indecomposable in rep(ST , Ql[x]) with EndQN = EndQ[x]N. There is a central-special quasigeneric T -group G such that V G = NA E rep(ST,Q(X)), End G = EndAG, and QEnd G/ JQEnd G is afinite-dimensional division Q(x)-algebra. Let A denote the principal ideal domain given in Lemma 3.5.2. Then EST) E rep(ST,Ql(X)) as QA = Q[x]A = Q(x) .Infact, N A is an indecomposable Ql(x)-representation, since N is an indecomposable
PROOF.
NA
= (NoA ,NiA:i
3.5 Quasi-Generic Groups
123
Q[x]-representation. This holds because EndQ(x)(N A) = Q(x)End!Q[xIN since Q(x) is the quotient field of Q[x] and N is finitely generated as a Q[x ]-module. Now, NA = Ua = (QG,QG(r) :r EST) for some finite rank A-Butler module G with EndQ(x)N A isomorphic to QEnd AG (Exercise 3.3.2), noting that QG = Q(x)G and Q(x) is the quotient field of the principal ideal domain A. Since N A is an indecomposable Q(x )-representation, QEnd AG has no nontrivial idempotents. This shows that G is a strongly indecomposable A-Butler module . By hypothesis, EndQN = EndQ[xIN, so that Q(x) £ CEndQN A = CEndQ Ua and A £ Ca = (CEndQUa)n(EndG).ItfollowsfromLemma3.5.9andTheorem 3.4.3 that G is a central-special quasi-generic T-group with End G = End-, G and QEnd G/JQEnd G a finite-dimensional division Q(x)-algebra. Given a finite lattice of types T , a countable torsion-free abelian group H is pregeneric if H is a torsion-free Z[x ]-module, H /(Z[x]X I + ... + Z[x ]Xn ) is bounded for pure rank-1 subgroups X; of H with type X; E T, and End H = Endz1xlH. A pregeneric group H is special , but a strongly indecomposable pregeneric group need not be quasi-generic. For instance, all the groups listed in Lemma 3.5.1 are pregeneric groups. As a consequence of the next theorem, if B(T)Q has tame representation type, then each strongly indecomposable group in B(T)Q can be constructed as a factor group of a pregeneric subgroup of a central-special quasi-generic T-group.
Theorem 3.5.11 The category B(T)Q has tame representation type ifand only iffor each sequence of nonnegative integers w = (wo, w r : tEST) there are finitely many strongly indecomposablepregenericgroups HI, .. . , Hn such that: (a) For each i, G; = H;A is a central-special quasi-generic T-group with EndAG; = End G;. (b) If H E B(T)Q is strongly indecomposablewith cdn H = w, then H is quasiisomorphic to H;A for some A = Z[x]/(f(xn and f(x) an irreducible polynomial in Z[x]. First assume that (a) and (b) hold. Fix wand define N; = (QH;, QH;(r) : rEST) E rep(ST, Q[x D. If U E rep(ST, Q) is an indecomposable representation with cdn U = w, then U = (QH, QH(r) : rEST) for some strongly indecomposable H in B(T)Q and cdn H = w by Theorem 3.3.2. By (b), H is quasi-isomorphic to K = H;A for some A = Z[x]/(f(xn, f(x) an irreducible polynomial in Z[x]. Hence , QA = Q[x]/(f(xy} with f(x) irreducible in Q[x]. Then U = (QH,QH(r) :r EST) is isomorphic to (QK ,QK(r) :r EST) = (QH;A, H;(r)Q[x]QA: rEST) = N;QA, noting that K(r) = H;(r)Z[x]A for each type r. This proves that B(T)Q has tame representation type. Conversely, assume that B(T)Q has tame representation type. Fix wand choose indecomposable N I , . .. , Nm E rep(ST, Q[xD with Q[x] £ CEnd N; such that if U E rep(ST, k) is indecomposable with cdn U = w, then U is isomorphic to N;B for some indecomposable cyclic Q[x]-module B. By Lemma 3.5.10, there is a central-special quasi-generic T -group G; such that (QG ;, QG; (r ) : rEST) =
PROOF.
124
3. Butler Group s
NjA E rep(ST, Q(x» , End G j = End"G j , and QE(Gj)/lQE(G j) is a finitedimensional Q(x)-algebra. In fact, G j is a finite rank strongly indecomposable A-Butler module, so there are pure rank-l subgroups X" ... , X m of G, such that G;/(AX, + ... + AXm ) is bounded. Hence, G, is a central-special quasi-generic T -group by Lemma 3.5.9. Define H, = Z [x ]X , +.. ·+Z[x]X n , a Z[x]-modulewithEnd H, = Endz[xlHj, since End G j = End" G j . Then H, is a pregeneric group with G i = H, ® Z[x I A. Moreover, H, is strongly indecomposable, since G, is strongly indecomposable. This proves (a). As for (b), let H be a strongly indecomposable in B(T)Q. Then UH = (QH, QH (r ) : r E ST) is an indecomposable in rep(ST, Q), and so UH = Nj®Q[xIB for some B = Q[x]/(g(x)') by the definition of tame representation type. Moreover, K = HjA is quasi-isomorphic to H for some A = Z[x]/(f(xY) with QA = B. This is so because UK = (QK,QK(r:):r: EST) = (QHjA, QHj(r:)A: r EST) = NjB = UH , whence K is quasi -isomorphic to H by Theorem 3.3.2. The proof of (b) is now complete. Open Question: Are there existence conditions on central-special quasi-gen-
eric T -groups, or their endomorphism rings, equivalent to tame representation type for B(T)Q? This question is the analogue of the open questions in Section 1.5 for k-representationsofa finite poset.
NOTES ON CHAPTER
3
Section 3.1 is an exposition of the traditional tools of the theory of torsion-free abelian groups of finite rank used in this book. Additional properties and example s of such groups are in [Fuchs 73] and [Arnold 82]. The notion of r -socle dates back to [Baer 37], but the first systematic development of r-radicals seems to be due to [Lady 79]. Finite rank Butler groups are called quasi-essential groups in [Koehler 65], purely finitely generated groups in [Bican 70, 78], R-groups in [Butler 65], and rr-diagrammatic groups in [Butler 87]. One ofthe attracti ve features of this category of groups is the number and variety of characterizations of these groups within the category of torsion-free abelian groups of finite rank. The work of [Koehler 65], as acknowledged by [Butler 87], is fundamental to Corollary 3.2.4, Theorem 3.2.5, and Theorem 3.3.2. Some other characterizations are given in [Arnold 81]. Results of [Bican 70, 78], including Proposition 3.2.15, give rise to a characterization of finite rank Butler groups in terms of balanced extensions, Theorem 3.4.6. This characterization provides both a natural definition for Butler groups of arbitrary rank and a host of associated problems for torsion-free abelian groups. The resolution of these problems involves sophisticated set theory and logic; see [Fuchs 94]. Included in [Fuchs, Metelli 92] is a development of properties of countable Butler groups using techniques potentially useful for Butler groups of arbitrary rank. An error in an argument of [Bican , Salce 83], repeated in [Arnold 86] as observed in [Mines, Vinsonhaler 92], is corrected in [Bican, Rangaswamy 95]. The corrected argument can be used to provide an alternative proof of Theorem 3.4.6. In another direction , the properties listed in Theorem 3.2.5(b) characterizing finite rank Butler groups have been considered for torsion-free abelian groups of infinite rank in [Mader, Mutzbauer, Rangaswamy 94].
3.5
Quasi-Generic Groups
125
In a series of papers, summarized in [Lady 83], the results of Section 3.2 are extended to a class of finite rank torsion -free modules over Dedekind domains called Butler modules . The setting is different from that of this manuscript, since the quasi-homomorphism category of finite rank Butler modules is subdivided into subcategories determined by a generalization of the notion of a splitting field. These subcategories are then related to categories of finitely generated modules over finite-dimensional algebras . Some partial results on representation type of these subcategories are obtained from classifications of representations of species [Ringel 76]. Finite rank Butler modules over l-dimensional Noetherian domains are considered in [Goeters 99]. Other generalizations include valuated vector spaces , [Richman 84], and groups with Murley groups playing the role of rank-I groups [Albrecht, Goeters 98]. Quasi-generic groups arose out of the notion of generic representations given in Section 1.5. There are functorial correspondences from categorie s of infinite-dimensional representations of posets to categories of countable Butler groups , [Duga s, Thome 91] and [Rangaswamy, Vinsonhaler 94], but the morphisms for groups are "local quasihomomorphisms," not quasi-homomorphisms as defined herein . This provides an obstacle to the determination of a Butler group analogue of generic representations. The point of Theorem 3.5.11 is that in the tame representation type case, strongly indecomposable Butler groups of finite rank are determined by quasi-generic Butler groups of infinite rank.
4 Representations over a Discrete Valuation Ring
4.1 Finite and Rank-Finite Representation Type There are several choices for extending the definition of representations of finite posets over fields to representations over discrete valuation rings [Plahotnik 76]. The choice made herein arises naturally from the connection between repre sentations of finite posets over discrete valuation rings and isomorphism at p categories of finite rank Butler groups; Section 4.3. Posets with finite representation type are characterized in Corollary 4.1.7. A discrete valuation ring is a principal ideal domain R with a unique prime ideal p R. If R is a discrete valuation ring, then R is a local ring, the Jacobson radical of R is p R , and R / p R is a field. An important example from abelian group theory is the discrete valuation ring Zp, p a prime of Z. Specifically, a torsion-free abelian group is a p -local group if and only if it is a torsion-free Zp-module. Observe that pZp is the unique prime ideal of Zp and Zp/ pZp = Z/ pZ is a field. Throughout this chapter, R denotes a discrete valuation ring. If M is a finitely generated free R-module and N is a pure submodule ofM, i.e., N n pM = pN, then M / N is a finitely generated torsion-free R-module. Since R is a discrete valuation ring, in particular a principal ideal domain , M / N is a finitely generated free R-module, and so N is a summand of M . Consequently, a submodule of a finitely generated free R-module is a summand if and only if it is a pure submodule. Let S be a finite poset and define rep(S, R) to be the category with objects V = (Vo, Vi : i E S) such that Vois a finitely generated free R-module, each Vi is a summand of Vo,and Vi is contained in U], provided that i :5 j in S. A morphism from V = (Vo, Vi : i E S) to V = (VO, Vi : i E S) is an R-homomorphism
4.1 Finite and Rank-Finite Representation Type
127
f: V o -+ Vo with f(V i) ~ Vi for each i in S. Since R is a discrete valuation ring, each Vi is a finitely generated free R-module, being a summand of a free R-module Vo. The rank of a representation V = (Vo, Vi : i E S) E rep(S , R) is the rank of Vo as a free R-module. Rank-I representations are easily described. Up to isomorphism, they are of the form V = tU«; Vi : i E S) with V o = R and Vi = or R, subject to the condition that if i ~ j in S and Vi = R, then U, = R. Given an integer j ~ 0, let rep(S, R, j) denote the full subcategory ofrep(S, R) with objects (Vo, Vi : i E S) such that pj V o ~ ~ {Vi : i E S} ~ Ue: In particular, rep(S, R , 0) consists ofrepresentations V = (Vo, Vi: i E S) with Vo = ~{Vi : i E S} and each Vi a summand of Vo. The category rep(S, R , 0) is a direct generalization of the category of elements V in rep(S, k) with no trivial summands, k a field. Define Ind(S, R, j) to be the set of isomorphism classes of indecomposable representations ofrep(S, R, j).
°
°
Example 4.1.1 If S is a finite poset with w(S) = 1 and j ~ is an integer, then S is a chain and rep(S, R, j) has finite representation type. Elements of Ind(S, R , j) are rank-l representations, and each indecomposable representation has endomorphism ring R. PROOF.
The argument of Lemma 1.3.3(a) for fields carries over to this setting.
For posets that are not chains, there are more indecomposable representations for discrete valuation rings than there are for fields . Example 4.1.2 The category rep(S2, R, j) has finite representation type for each j ~ 0. The elements of Ind(S2, R, j) are: (i) rank-l representations (R, 0, 0), (R, R, R), (R , R, 0), and (R, 0, R) with endomorphism ring R. (b) rank-2 representations (R EB R + (l + I)R(lj pn), R EB 0, EB R) with endomorphism ring {(r, s) E R x R : r - s E p" R} for each 1 ~ n ~ j .
°
Let V = (Vo, VI , V 2) E rep(S2, R, j) be an indecomposable representation with rank Vo > 1. Notice that each Vi is nonzero and VI n V 2 = 0, since otherwise V is a rank-I representation as in Example 1.1.3. Hence, p" Vo ~ VI EB V2 ~ Vo for some smallest positive integer n ~ i, observing that if n = 0, then U« = VI EB V2, and V would again be a rank-l representation. Now, V = (VI EB V2, VI, V2, pnvo) is an R-representation of S3 with pnvo a free R-module, being isomorphic to Us, and pnVo n Vi = pnVi, since Vi is a pure submodule of Ui; However, p" Vo need not a summand of VI EB V2. There is an isomorphism End V -+ End V given by sending f to the restriction of f to VI EB V2, recalling that End V is the ring of R-endomorphisms f of VI EB V2 with PROOF.
128
4. Representations over a DiscreteValuation Ring
f(Ui) £ u, and f(pnUo) £ ti; Then V is an indecomposable representation with rank greater than 1. The remainder of the proof is a matrix argument generalizing that of Example 1.1.5 for representations of S3 over a field. The setting is that U = (Uo, UI , U2) is indecomposable and V = CUI EB U2, UI, U2, pnUo) is an indecomposable representation with End V isomorphic to End U. Let BI = {XI, , x,} be an R-basis of UI, B2 = {YI, ... , Ys} an R-basis of U2, and B3 = {ZI, , z.l an R-basis of p'U«. For each i,
In particular, pnUo can be interpreted as the row space of a t x (r Mv
+ s) R-matrix
= (A I B),
where the rows of Mv are labeled by B3, A = (aij) is a t x r R-matrix with columns labeled by BJ, and B = (bij) is a t x s R-matrix with columns labeled by B2. The following invertible R-matrix operations on M» do not change V: (a) Elementary column operations within A (a basis change for VI); (b) Elementary column operations within B (a basis change for V2); (c) Elementary row operations on M (a basis change for pnvo). Write M v:::::: N if the matrix N can be obtained from Mv by a sequence of operations (a), (b), and (c). Since R is a principal ideal domain, the block matrix A can be reduced by a series of elementary invertible row and column operations to a diagonal matrix [Hungerford 74]. Since n is minimal and R is a discrete valuation ring, this diagonal matrix can be chosen to be
(01
0)
pD
with D a diagonal R-matrix and I an identity matrix, agreeing that a 0 x 0 matrix is empty. If X = UI EB U2 E pn Vo with u, E Vi, and p-height X :::: n, then p-height UI = p-height U2, since VI and V 2 are pure submodules of Vo. It follows that M» is equivalent to a matrix of the form
(:
0 pI 0
Bll pB21
B12 pB22
B2n ) pB"
pnl pnBnl
p nBn2
pnB nn
0 0
with each row of (l-I Bi!, pi-I Bi2, ... , pi-I Bin) containing an element of
4.1 Finite and Rank-Finite Representation Type
129
p-height i - 1. If E is an operation as in (c), then (l 0 . . . 0 I BII B12
Bin)
.. .
(E 0 . .. 01 EB II EB n . .. EBln) ~ (l = EE- I 0 .. . 0 I EB II EB I2 ... EB ln)
~
as an application of (c) and (a). Applying operations (b) and (c) to diagonalize the matrix (B II BI2 . .. Bin) yields M» ~ N =
(:
I pB 21
0 pI
0 0
0
pnl
pnBn1 p" Bn2
Now use (c) followed by (a) to see that N
0 pI
C
0
0 p B22
0 0
I 0
p" I 0
P~m
)
pnBnn
~
0 p C22
p nCn2
P~m
)
pnCnn
Since n is chosen to be minimal, I must be nonempty. As in Example 1.1.5, V = (Rx i EI1 RYt , RXI, RYI, Rtx, + Yt)), since V = (VI EI1 V2, VI, V 2, pnVo) is indecomposable. Then V = (Rx, EI1RYI + R(xI + Yd/ p", RXI, RYd is isomorphic to W = (R EI1 R + R(I + 1)/ p", R EI1 0, 0 EI1 R). A straightforward calculation shows that f E End W if and only if f = (r, s) E R x R with r - s E p" R. Hence, End V is isomorphic to {(r,s) E R x R : r - s E pnR}. For a finite poset S, define S* to be the disjoint union of S and an element * unrelated to any element of S. For example, If S = Sn is an antichain, then S* = Sn+I' Given n ::: 2, let Cn denote a chain with n elements. Then C~ is the poset (n, I). For example, q = (l , I) = S2 and q = (l, 2).
Proposition 4.1.3 Let n ::: 2 be an integer. (a) There is a fully faithful functor Fj : rep(C~_l' R, j) --+ rep(C~, R , 0) for each j ::: O. (b) If V E rep( C~ , R, 0) is indecomposable with rank> 1, then there is some integer j ::: 0 and indecomposable V E rep(C~_I ' R , j) with Fj(V) = V. (a) Write Cn- I = {I < 2 < ... < n - I} and let V = (V o, VI s:; ... s:; V n- I, V*) E rep(C~_I' R, j) . Define Fj(V) = (Vo, VI s:; ... s:; Vn-I s:; Vo, V*) E rep( C~, R, 0). It is routine to verify that with Fj(f) = f, Fj : rep( C~_l' R , j) --+
PROOF.
rep( C~ , R, 0) is a fully faithful functor. (b) Let V = (Vo, VI s:; .. . s:; Vn , V*) be an indecomposable representation in rep(C~ , R , 0) with rank> 1. Then Vo = Vn + V*, and so V* = WI EI1 (V* n Vn)
130
4. Representationsover a Discrete Valuation Ring
for some WI. This holds because V* n Vn is a pure submodule, hence a summand, of V*, since Vn is a pure submodule of Vo . Therefore,
and
is a representation direct sum. But WI is a free R-module. Since V is indecomposable with rank> I, it follows that WI = 0, Vo = Vn contains V*, and V =
(Vn, VI, "" Vn- I, Vn, V*) . Furthermore, p! Vn is contained in Vn- 1 + V* for some j. To see this, let W be the pure submodule of Vn generated by Vn-l + V* . Then Vn = W' ffi W for some
W' and
V =(W',O, .. . ,O, W',O)ffi(W , VI, . .. , Vn-I, W, V*). Again, W' = 0, since V is indecomposable of rank greater than 1, and so Vn W . Thus, there is some j with p! Vn contained in Vn-I + V*, since Vn is a finitely generated free R-module with rank Vn rank Vn- I + V* . Hence, V = (Vn, VI, .. " Vn-I, V*) E rep(C~_l' R , j) with V = Fj(V). As a consequence of (a), V must be indecomposable.
=
=
=
Corollary 4.1.4 If S (1,2), then rep(S, R , 0) has infinite representation type but each indecomposable has rank less than or equal to 2. Rank-2 elements of Ind(S, R, 0) are (Vo, 0 ffi R, R ffi 0 5; Vo)for j ~ 1, where Vo = R ffi R + R(1 + l)jp j.
=
=
=
PROOF. As mentioned above, q (1,2) and q (1,1) S2. By Theorem 4 .1.3, each indecomposable in rep«l, 2), R, 0) is of the form Fj(V) for some indecomposable V E rep(S2, R, j). Now apply Example 4.1.2.
Lemma4.1.5 If Sis a finite poset, S' is a subposet of S, and j' ::5 j, then there is a fully fa ithful functor F- : rep( S', R, j') ~ rep( S, R, j).
=
=
PROOF. Given V (Vo, U, : i E S') E rep(S', R, l'). define F-(V) (Vo, Vj : E S) E rep(S, 'lp, j) by Vo V o, Vj U, if i E S', Vj n{Vj: j E S', j > i} if i E S\S' and there is some j E S' with j > i, and Vj Vo otherwise. Each Vj is a pure submodule, hence a summand, of Vo, and pj Vo 5; :E j Vj, since j' ::5 j and pj Uo 5; :E j Ui. Then F- is a fully faithful functor, just as in Proposition 1.3.1 (b).
i
=
=
= =
The functor F+, as defined in Proposition 1.3.1 (a), is not directly relevant for representations over discrete valuation rings, since if U, and U, are pure submodules of Vo, then U, + U, need not be a pure submodule of Us;
4.1
Finite andRank-Finite Representation Type
131
Proposition 4.1.6 Let S be afinite poset ofwidth 2 that does not contain (1, 2) as a subposet. For each j :::: 0, rep(S, R, j) has finite representation type, and each indecomposable has rank ~ 2.
PROOF. The proof is by induction on lSI, the cardinality of S. If lSI = 2, then S = Sz, rep(S, R, j) has finite representation type, and each indecomposable has rank ~ 2 by Example 4.1.2. Now assume lSI:::: 3 and let a and b be incomparable elements of S. Since w(S) = 2 and S does not contain (1, 2) as a subposet, it follows that if s E S with s =f:. a and s =f:. b, then either s < a and s < b or else s > a and s > b. Hence, there is a partition S = AU B U C of S with C = Is E S : s < a, s < b}, A = Ia, b}, and B = {s E S: s > a, s > b}. If the pair (a, b) of incomparable elements of S is chosen to be minimal in the sense that w( C) = I, then C is a chain. In this case, the partition S = AU B U C is a splitting decomposition, as defined in Section 1.3, since s < t for each sEA and t E B. Use the proof of Proposition 1.3.5 to see that each indecomposable U in rep(S, R, j) is isomorphic to F-(V) for some indecomposable representation V in either rep(A U B , R, j) orrep(B UC, R, j), with F- defined as in Lemma 4.1.5. Both A U Band B U C are posets of width 2 that do not contain (1, 2) as a subposet, being subposets of S. By induction on lSI, AU Band B U C have finite represention type, and each indecomposable in rep( A U B, R, j) or rep( B U C, R, j) has rank less than or equal to 2. Consequently, rep(S, R, j) has finite representation type and each indecomposable has rank less than or equal to 2. A finite poset S is a garland if w(S) = 2 and S does not contain (1, 2) as a subposet. Garlands also arise for representations of finite posets over fields [Simson 92]. The terminology is motivated by the fact that if S is a garland, then S is a subposet of the poset Gn,adisjointunionoftwochains Cn = {l < 2 < .. . < n} and C~ = {I' < 2' < .. . < n'} subject to the additional relations that i < (i + 1)' and i' < i + 1 for each I ~ i ~ n - 1. Corollary 4.1.7 Let S be a finite poset. Then rep(S, R, j) has finite representation type if and only if w(S) = 1 or S is a garland.
PROOF. If w(S) = I or S is a garland, then rep(S , R, j) has finite representation type by Example 4.1.1 and Proposition 4.1.6. The converse follows from Corollary 4.1.4 and Lemma 4.1.5 in the case w(S) = 2 and from Example 4.2.4 and Lemma 4.1.5 if w(S) :::: 3. An additive category has rank-finite representation type if there is a bound for ranks of indecomposables in the category and rank-infinite representation type if there is no such bound. Finite representation type implies rank-finite representation type, but Corollary 4.1.4 illustrates that the converse is not true in general. The next proposition relates rep(Sn+z, R/pR) to rep(S, R , 2n + 1) for S = (1, 2). This provides a systematic construction of indecomposable representations
132
4. Representations over a Discrete Valuation Ring
in rep(S, R, j) for sufficiently large j from indecomposable representations of S; overthe field k = R] p R, If M is an R-matrix,then M(mod p) is the Rj pR-matrix obtained by reducing each element of M modulo p. Define E(n , k) to be the subcategory of rep(Sn+2, k) , consisting of representations V = (Vo, Vi: I ::: i ::: n + 2) such that Vo = VI E9 . .. E9 Vn , there is a fixed s with s = k-dimension Vi for each i ::: n, Vn+ 1 E9 Vn+2 S; Yo, and Vn + 1 = (1 + I + ... + l)k S is the image of the diagonal embedding of kS into Yo. Proposition 4.1.8 Let S = (1,2), n ::: 0, and k = Rj p R, There is a correspondence 4J : E(n, k) ~ rep(S, R, 2n + 1) such that if V E E(n, k) is an indecomposable representation, then 4J(V) is an indecomposable representation. PROOF.
Let k
= RjpR and V = (Vo, Vi)
E
E(n, k). Then Vo
Vn+1 E9 Vn+2 )j K, where K is the row space of a k-matrix
M'
= ( MI,I II, M2 I'"
= (VI
E9 .. . E9
I I I
... MI,n 0I 0I )
with each M: an s x s k-matrix, s the k-dimension of Vi for each i ::: n. The first row of M' gives the relations for Vn+ l , and the second row for those of Vn+2 . Choose s x s R-matrices M, with Mi(mod p) = M: for I ::: i ::: n. Let Wand X be free R -moduleswithrank 2s, and Y and Z free R-rnodules with rank(n + I)s . Define 4J(V) = V = (Vo, V), V 2 S; V 3) E rep(S, R, 2n + I) by
= W E9 X E9 Y ffiZ+ (1/p2n+I)(1 + 1 + 0+ O)R 2s , VI = W E9 Z, V3 = X E9 Y E9 Z, and V2 the row space of the matrix Mu = (E I, E2 , l) = Uo
pM I p 2M2
p 2I p 31
pnM n pn+1 I
pn+II 0
p 3I 0 0 pSI
0 0
0 0
0
0 p2n+1 I 0
0 0
I 0 0 I
0 0 0 0
0 0 0 p2n+31 0 0
I 0 0 I
a submodule of V3 = X E9 Y E9 Z. The columns of M u are indexed by X, Y , and Z, respectively, the l's stand for s x s identity matrices, the O's stand for s x s matrices of all zeros, and (1 + I + 0 + O)R 2s = {x E9 x E9 0 E9 0 E Vo : x E R2s }. Notice that V2 is a summand of V3. The strategy is to show that if f E End V with R-matrix M f relative to an R-basis of Vo, then Mf(mod p) = aI + N for some nilpotent k-matrix Nand some k-matrix a representing an endomorphism of V . In this case, it is verified that if f is idempotent,then f = 0 or I as a consequence of the assumption that V is indecomposable. Hence, 4J(V) = V must be an indecomposable representation of rank (6 + 2n)s.
4.1 Finite and Rank-Finite Representation Type
If I E End V, then I : VI = W E9 Z ~ VI = W E9 Z, I : V3 = X E9 Y E9 Z V 3 = X E9 Y E9 Z, I : Z ~ Z as VI n V3 = Z, I : V z ~ Uz, and I: (l
+ 1+ 0 + O)R 2s(mod pZn+l) ~
(l
133 ~
+ I + 0 + O)R 2s(mod pZn+I).
Then I can be represented as an endomorphism of W E9 X E9 YE9 Z with matrix
Mf
=
(
~I 0
o
o
0
hz hz
h3 133
o
0
where III : W ~ W, 114 : W ~ Z, [n : X ~ X, h3 : X ~ Y, h4 : X ~ Z, 132 : Y ~ X, 133 : Y ~ Y, 134: Y ~ Z, I44:Z ~ Z, and I: V z ~ UiSince I: (l + I + 0 + O)R zS ~ (l + 1+ 0 + O)R 2s + pZn+l Vo, it follows that III == hz(mod pZn+l), !J4 == h3 == h4 == 0 (mod pZn+I), and
o III hz
o
Also, I : U: ~ Ui. and so MUMf (Edll
0 0 133
0
== (0, E I, E z, l)Mf(mod pzn+l)
+ Ezhzl Ed33 I Ed34 + 144) == (g E I I g Ezi g l)(mod
pZn+l)
for some g. Equating block matrices yields
+ 144/ == g/(mod p2n+l), Edll + Ezhz == gEl == (Ed34 + I44l)E 1(mod p2n+l), Ed33 == g E; == (Ed34 + I44l)Ez(mod pZn+I). Ed34
Write matrices 133 = (aij)(n+l)x(n+l) representing an endomorphism of Y = (w)n+1 and 144 = (bij)(n+l)x(n+1) representing an endomorphism ofZ = (w)n+1 with each aij and bij an s x s R-matrix representingan endomorphismof W. Then Ed33
== (pZi+l aij)(mod p2n+l),
I44Ez
== (pzj+lbij)(mod pZn+I),
Ed33
==
Ezh4Ez
+ I44Ez(mod In+l).
Write 134 = (xij) with each xij an s x s R-matrix. Combine equations to see that pZi+l aij == pZj+2+Zi Xij + pZj+1bij(mod pZn+l) for 1 ::: i, j ::: n + 1. Then pZi+l aij == pZj+lbij(mod pmax(Zi+l ,Zj+I». If i < j, then aij == O(mod p), and if i > i. then bij == O(mod p). Consequently, h3(mod p) is lower triangular, I44(mod p) is upper triangular, and aii == bii(mod p) for each 1 ::: i ::: n + 1.
134
4. Representations over a Discrete Valuation Ring
Write the matrix fll = (Cij hX2 representing an endomorphism of X = (R s )2 with Cij an s x s R-matrixrepresentingan endomorphismof W. Recallfrom above that g E, == Edll + Ezf32 == (Ezf34 + f44l)E I(mod p2n+I). From thelast row of the latter equation, CII
== bn+1 ,n+l(mod p) and CI2 == O(mod p2).
The first row yields CII
== bll(mod
p)
== c22(mod p) .
From the first column, Micll == b.,Mj(mod p) for each i :::: n. Finally, the second column of this equation gives C22
== bii(mod p) for each i
:::: n.
Let a = clI(mod p), an s x s k-matrix. By the results of the preceding two paragraphs, a == c22(mod p) == bjj(mod p) == aii(mod p) for each i :::: n + 1, and a M, == M ja(mod p) for each i :::: n. In particular, aM' = M'a, recalling the definition of M' and the fact that M; = Mi(mod p) . Consequently, al(n+2)sx(n+2)s is a k-matrix representing an endomorphism of V E E(n, k) . This is so because a represents an endomorphism of VI E9 .. . E9 Vn+2 with a(K) ~ K, the row space ofM'. In summary, Mf(mod p) = al + N , where NI
N-
(
0 0
o
o
N2
132
o
0 0 N3
0
and N, = f ii(mod p) - al is of the form
A straightforwardcalculation shows that N is nilpotent. Finally, assume that V is indecomposable and f2 = f E End V . Then (al + N)2 = a 2I + aN + N a + N 2 = a I + N. Since N is a nilpotent matrix with zeros on the diagonal, it follows that a 2 = a. Thus, al(n+2)s x(n+2)s represents an idempotentendomorphism of V. But V is indecomposable, so a = 0 or 1.1f a = 0, then N is both nilpotent and idempotent. In this case, f E pEnd V ~ EndR(Vo) with f2 = f. Since Vo is a free R-module, f = O. On the other hand if a = 1, then 1 - f = 0 and f = 1. This shows that ¢(V) = V is an indecomposable representation. Corollary 4.1.9 If S tion type for j :::: 5.
= (1,2),
then rep(S, R, j) has rank-infinite representa-
4.2 Wild Modulo p Representation Type
135
PROOF. Let n = 2. Given a positive integer s, there is an indecomposable representation
in rep(S4, k) with A an indecomposable s x s k-matrix. In the notation of Proposition 4.1.8,
In fact, V E E(2, k), since Vo = VI EB V2, k-dimension VI = k-dimension V2 = s, and V3 EB V4 £; VI EB V2. By Proposition 4.1.8, ¢(V) is an indecomposable in rep(S, R, j). Since there are indecomposable k-matrices A withs arbitrarily large, it follows that rep(S, R, j) contains indecomposable representations of arbitrarily large finite rank for j 2': 5 = 2 . 2 + 1. Example 4.1.10 If S = (1,2), then rep(S, R, 4) has rank-infinite representation type.
PROOF. Exercise 4.1.1. Open Question : For S = (1, 2) and 1 ::: j ::: 3, does rep(S, R, j) have rankfinite or rank-infinite representation type? In this case, can indecomposable representations in rep(S, R, j) be classified up to isomorphism ?
EXERCISE
1. Find a matrix to show directly that if S = (1,2), then rep(S. R, 4) has rank-infinite representation type (Proposition 4.1.8 does notapply in this case).
4.2 Wild Modulo p Representation Type Includedin thissectionis a computationof therepresentation typeof rep(S, R, j) in terms of Sand j (Corollary4.2.5). The definition of wild modulo p representation typeis motivated by the definition of endowildforrep(S, k), k = R/ p R a field. The categoryrep(S, R, j) has wild modulo p representation type if for each R -algebra r that is finitely generatedand free as an R-module, there is U E rep(S, R, j) such that r / p r is a ring-homomorphic image of End U . It remains an open question as to whetherifrep(S, R, j) has wild modulo p representation type, then for each finite-dimensional k-algebra A there is U E rep(S, R, j) with A a homomorphic image of End U (see Open Question 2). If rep(S, R, j) has wild modulo p representation type, then it has infinite representation type. Moreover, in view of the diversity of endomorphism rings in
136
4. Representations overa Discrete Valuation Ring
this case (Exercise 4.2.1), it is not reasonable to expect to be able to classify the indecomposable representations. The first lemma gives a computational tool for determining wild modulo p representation type. The idea is that for a given I', there are R -matrices A and B with qA, B) = rand qA(mod p), B(mod p» = r/pr . The matrices A and B are then used to construct representations U with qA , B) S; U and a homomorphism End U --+ qA(mod p), B(mod p». It follows that there is an onto ring homomorphism End U --+ r / p r. The matrices A and B are somewhat special , since qA, B) --+ qA(mod p) , B(mod p» need not be onto for arbitrary matrices A and B (Exercise 4.2.3). Lemma 4.2.1 Suppose r is an R-algebra that is finitely generated andfree as an R-module. There are R-matrices A and B such that (a) qA, B) = r andqA(mod p), B(mod p» = r/pr, and (b) the ring homomorphism qA, B) --+ qA(mod p), B(mod p» given by M r+ M(mod p) is onto. (a) If {I, Y2, . .. , Yt} is an R-basis of I', then {I, X2, . . . , x t } is a k-basis for r / pi', where k = R/pR is a field and Xi = Yi + pr o Define (t + 2) x (t + 2) r-matrices
PROOF.
A=
0 0
1 0
0 1
0 0
0 0
0 0
0 0
0 0
0 0
I
, B=
0
0 1 Yl 0
0 0 I Y2
0
0
0 0 0 I
0 0 0 0 Yt
0 0 0 0 0
Then A and B are R-matrices with qA, B) isomorphic to Endj-Il") = r. The proof is exactly the same as for Example 1.1.7. Similarly, r / p r is isomorphic to qA(mod p) , B(mod p», as desired . Assertion (b) follows from (a). Corollary 4.2.2 If S = (1,2), then rep(S, R, j) has wild modulo p representation type for j ::: 7. PROOF. Let r be an R-algebra that is finitely generated and free as an R-module and, via Lemma 4.2.1, let A and B be s x s R-matrices with r = qA, B) and r / pr = qA(mod p), B(mod p» . Define V E rep(S3, k) by
V = (kS t1J k S t1J k", k S t1J 0 t1J 0, 0 t1J k' t1J 0, 0 t1J 0 t1J k S ,
(1 + I
+ 1)kS, (l + A' + B')kS)
for s x s k-matrices A' = A(mod p) and B' = B(mod p). Then End V = qA', B'), by Exercise 1.2.2(c), whence End V is isomorphic to r / pro It follows from the definitions that V E £(3 , k), as defined in Section 4.1. In the notation of
4.2 Wild Modulo p Representation Type
137
Proposition 4 .1.8,
M'
= ( ~ I ~, I ~, I~ I~ ) . =
Moreover, ¢(V ) = U = (Uo, UI, U2 £; U3) E rep(S, R , 7) is given by Uo WEB X EB Y EB Z + (1/ p7)(1 + 1 + 0 + O)R 2s , UI = WEB Z, U3 = X EB Y EB Z, and U2 , a submodule of X EB Y EB Z, the free R-module with basis the rows of M u = (E I , E2, I) =
p2I p3I p4 I
o
o o
p3I 0 0
p7 I
o
0
0 0 0 p9I
I
0
0 0 0
0 0
I
o
0)
o
I
I
0 0
.
Notice that C(A, B) £; EndU, since if b E C(A , B), then bMu = Mub, and so bI E EndU. As in Proposition 4.1.8 , if f E EndU, then Mf(modp) = aI + N for some unique aI E End V = C(A', B') and nilpotent k-matrix N . It follow s that there is a ring homomorphism x : End U ~ End V I' / pf given by x (f) = a I . Since C(A, B) £; End U, n is onto by Lemma 4.2.1. This shows that rep(S , R , 7) has wild modulo representation type. Now apply Lemma 4.1.5 to see that rep (S , R , j) has wild modulo p representation type for j ::: 7.
=
Submodules of finitel y generated free R-modules need not be summands, unless they are pure . Another category of representations of finite posets over discrete valuation ring s is defined to reflect this distinction. Let rep f (S , R) be the category of representations U = (Uo , U, : i E S) such that Uo is a finitely generated free R-module, each U, is a R-submodule of Ui ; U, £; U, if i ::::: j in S, and U« = b {U i : i E S}. The morphisms in this category are representation morphisms. Assertion (b) of the next lemma is a generalization of Theorem 1.2.3, while statement (a) for the special case that n = 2 was used in the proof of Example 4.1 .2. For an antichain Sn, define fl(Sn , R) to be the full subcategory of rep(Sn+1, R , 0) with objects of the form U (Uo, Ui, U* : i E S) such that Uo EB{ U, : i E S}, U, n U* = 0, and U, + U* is pure in Uo for each i E S« . Recall that the poset S* = S U {*} is the disjoint union of S and a point *.
=
=
Lemma 4.2.3 Suppose S is a finite poset. (a) There is afully faithful functor H' : rep(S, R , j) ~ rep f(S*, R). The image of H' consists ofall V (Vo, Vi, V* : i E S) with Vo b{Vi : i E S}, each Vi a pure submodule of Vo, and V* n Vi pi Vi for each i in S. (b) There is a category equivalence H: fl(Sn , R) ~ rep(Sn, R, 0), where Sn is
=
=
=
an antichain. PROOF. (a) Define H'(U) = V = (b {Ui : i E S}, Ui , pi Uo), where U = (Uo, U, : i E S ), and define H'(f) to be the restriction of f to b{Ui : i E S}. Then H' is
138
4. Representations over a Discrete Valuation Ring
a functor. For each i E S, pj Vo n Vi = p! Vi and Vi is a pure submodule of E i Vi, since Vi is a pure submodule of Vo. Each Vi is a summand of E {Vi: i E S}, but pj Vo need not be, which is why the image of H' is in rep reS, R) and not necessarily in rep(S, R) . The functor H' is fully faithful, since if g: H '(V) ~ H '(V') is a representation morphism, then g extends uniquely to f: Vo ~ V~ with H'(f) = g . To see that the image of H' is as described, let V = (Vo , Vi, V*: i E S) E rep [(S* , R) with Vo = E {Vi: i E S}, each Vi a pure submodule of Va, and V* n Vi = p! Vi for each i. Then pj Vo = E{pj Vi : i E S} S; V*. Define V = (Vo, Vi : i E S) with Vo = Vo + (11 pj)V* . Now, Vo is a finitely generated torsionfree, hence free, R-module, and pj Vo is contained in Va. To see that Vi is pure in Uo, suppose x E Vo with px E Vi. Then p! x E V* n p Vi = pHI Vi, and so x E Vi, as desired. This shows that V E rep(S , R , j) with H'(V) = V . (b) Given V = (Vo, Vi, V* : i E Sn) E D.(Sn, R) with Vo = EB{Vi : i E Sn}, define H: D.(Sn ,k) ~ rep(Sn, R,O) by H(V) = (Vo, Vi :i E Sn), where Vo = Vol V* and Vi = (Vi + V*)I V* for each i in S. Then Vo is a finitely generated free R-module, since V* is a pure submodule of the finitely generated free R-module Vo, and Vo = L{Vi : i E Sn} . Because Vi + V* is pure in Vo, Vo/(Vi + V*) is a torsion-free module, and so Vi is pure in Va . Thus, H(V) E rep(Sn, R, 0). The proof that H is a category equivalence is the same as that of Theorem 1.2.3. Example 4.2.4 The category rep(S3, R, 0) has wild modulo p representation type. Let r be an R -algebra that is finitely generated and free as an R -module. By Lemma 4.2.1, there are n x n R -matrices A and B with r = qA , B) and rip r = qA(mod p), B(mod p)) . Define V = (Vo, VI, V2, V3 , V*) E rep(S4, R, 0) by
PROOF.
Uo
= R7n EB R6n EB R5n, VI = R7n EB 0 EB 0, V2 = 0 EB R6n EB 0, V3 = 0 EB 0 EB R5n,
and V* the free R-module with the rows of the following matrix as a basis:
1000000 0100000 0010000 0001000 o 0 0 0 p2 1 0 0 o 0 0 0 0 p6I 0 o 0 0 0 0 0 p7I
o
000000010 0 0 0 0 I 0 0 0 p2I o p6I 0 0 0 0 I 0 0 p5B o 0 0 0 0 0 0 I 0 p6I o 010010000 o 001001000 0000/00100
p2I
A routine computation shows that V E D.(S3, R). If f E End V, then M [(mod p) = al + N for some nilpotent k-matrix N with zeros on the diagonal and al E qA(mod p), B(mod p)) (Exercise 4.2 .2). Then 1T : End V ~ qA(mod p), B(mod p)) = I defined by 1T(f) = al is a ring
r pr
4.2 WildModulo p Representation Type
139
homomorphism. Since C(A, B) embeds in End U (Exercise 4.2.2), n is onto by Lemma 4.2.1(b). This shows that rep(S3, R, 0) has wild modulo p representation type. Recall that if S is a finite poset with w(S) only if S contains (I, 2) as a subposet.
= 2, then S is not a garland if and
Corollary 4.2.5 The category rep(S, R, j) has: (a) wild modulo p representation type if either w(S) ::: 3 or w(S) = 2, S is not a garland, j ::: 7; (b) rank-infinite representation type ifw(S) = 2, S is not a garland, j ::: 4; (c) finite representation type if and only if either w(S) = 1 or S is a garland. PROOF. Statement (a) is a consequence of Corollary 4.2.2, Example 4.2.4, and Lemma 4.1.5 . Assertion (b) follows from Corollary 4.1.9, Example 4.1.10 and Lemma 4.1.5. Statement (c) is Corollary 4.1.7.
For antichains the characterizations are complete; there is only finite or wild modulo p representation type.
Corollary 4.2.6 The category rep( s.. R, j) has (a) wild modulo p representation type ifand only ifn ::: 3, and (b) finite representation type if and only if n ~ 2.
Open Questions 1. Suppose S is a finite poset with w(S) = 2 and S is not a garland. If 6 ::: j ::: 4, does rep(S, R, j) have wild modulo p representation type? In this case, can indecomposable representations in rep(S, R, j) be classified up to isomorphism? 2. For which finite-dimensional algebras I" over a field k = R / p R does there exist an R-algebra r that isfinitely generated and free as an R-module with r/pr = r '? EXERCISES
1. Let G be a finite group,not necessarily abelian, and kG the groupalgebra of G overthe field k = R/ p R. Provethat the group algebra RG of Gover R is an R-algebrathat is finitely generated and free as an R-module with RG/pRG = kG. 2. Complete the proofof Example 4.2.4.
?).
3. Let R be a discrete valuation ring, A = (A f) and B = (A Show that the natural homomorphism C(A, B) ~ C(A(mod p), B(mod p» is not onto. 4. Let rep/(S, R, j) be the full subcategory of rep/(S, R) with objects V = (Vo, Ut : i E S) such that pj Vo £; Ej Ui. Use the results of Sections4.1 and 4.2 to find bounds for the representation type of rep/(S, R, j) in terms of Sand j .
140
4. Representations over a Discrete Valuation Ring
4.3 Finite Rank Butler Groups and Isomorphism at p Computations of representation type for categories of representations of posets over the discrete valuation ring Zp can be transferred directly to isomorphism at p categories of Butler groups (Corollary 4.3.3). Recall that for a finite lattice T of types, B(T) is the category of finite rank Butler groups G with typeset G ~ T, and Sr is the oppos ite of the poset of join irreducible elements of T. For an integer j :::: 0, let B(T, j) denote the full subcategory of B(T) consisting of those groups G with p! G contained in :E{G(r) : r E Sr} . The isomorphism at p category of B(T, j) is denoted by B(T, j)p. Notice that indecomposable groups in B(T, j)p are indecomposable in B(T, j), i.e., indecomposable as groups . Up to rank-I summands and isomorphism at p, there is no loss of generality in considering B(T, j): Proposition 4.3.1 If G E B(T) with no rank-I summands and p is a prime, then G is isomorphic at p to a group in B(T, j)for some nonnegative integer j . Let IT(G) denote the inner type of G. Since G is a Butler group, IT(G) E typeset G by Lemma 3.1.12 and Theorem 3.2.5. Then :E{G(a):a E STl = G*(IT(G» . This holds because if x E G with type x = r > IT(G) and r E T\Sr is not join irreducible, then r > a for some join irreducible, a E Sr and x E G(r) ~ G(a). Since G is a Butler group, G = G(IT(G» = Grr(G) $ G#(IT(G» with Grr(G) an IT(G)-homogeneous completely decomposable group and G#(IT(G»jG*(IT(G» finite (Theorem 3.2.5). But G has no rank-I summands, so that Grr(G) = and GjG*(IT(G» is a finite group. Choose a subgroup H of G with H jG*(IT(G» the subgroup of G jG*(IT(G» consisting of elements with order a power of p. Then G j H is a finite abelian group with order prime to p. Hence, G is isomorphic to H at p. Moreover, for some i . pj(H jG*(IT(G» = 0, so that pj H ~ G*(IT(G» ~ H*(IT(G» ~ Hand H E B(T, j). PROOF.
°
Given a prime p, a finite lattice T of types is p-locally free if X p is isomorphic to Zp for each rank -I group X with type X E T . Restriction to the case that T is p -locally free amounts to the p-reduced case , in that if type X E T and X p is not isomorphic to Zp, then X p is isomorphic to Q (Exercise 2.1.3). A torsion-free abelian group G is p-locally free if G p is a free Zp-module. Corollary 4.3.2 [Richman 95] 1fT is a p-Iocally free finite lattice oftypes and j is a nonnegative integer, then there is a category equivalence F : B(T, j)p ~ rep(Sr, Zp, j) given by F(G) = (G p » G(r)p : r E Sr). PROOF. The proof is analogous to that of Theorem 3.3.2 , bearing in mind that integers prime to p are units in Zp.
4.3 FiniteRank ButlerGroups and Isomorphism at p
141
Corollary 4.3.3 If T is a finite p-locally free finite lattice of types, then B(T,
n, has:
(a) wild modulo p representation type if either W(ST) ~ 3 or W(ST ) = 2, ST is not a garland, j ~ 7; (b) rank-infinite representation type ifw(ST) = 2, ST is not a garland, j ~ 4; and (c) finite representation type ifand only ifeither W(ST ) = lor ST is a garland. PRO OF.
A consequence of Corollaries 4.3.2 and 4.2.5.
Following are illustrations of how representations over discrete valuation rings can be used to construct finite rank Butler groups . If G is a finite rank Butler group with linearly ordered typeset , then G is a completely decomposable group by Corollary 3.2.7(a) . This is, in the isomorphism at p category, a manifestation of the fact that representations of chains over a discrete valuation ring are direct sums of rank-l representations. Observe that a finite lattice of types T is a chain if and only if ST is a chain .
Example 4.3.4 If T is a finite lattice of types and ST is a chain, then rank- I groups are the only indecomposable groups in B(T, j) p. PROOF.
Corollary 4.3.2 and Example 4.1.1
Example 4.3.5 Let T be afinite lattice oftypes with ST = {type X I, type X z} = Szfor subgroups Xi ofQ with 1 E Xi and 1/ p rt. X;/or each i . Indecomposable groups in B(T , j)p are: (i) rank-I groups G = Xl n Xz, Xl, X Z, Xl + X z with (End G)p = Z p, and (ii) rank-2 groups G = Xl $ X z + (1 + I)Z(1/ pi ) with (End G)p = fer, s) E z p x z p: r - s E pi Z p} for each 1 ~ i ~ j .
Apply Corollary 4.3 .2 and Example 4.1.2, noticing that if G is as given in (ii), then (G p- G(r\)p , G(rz)p) = (Zp $ Zp + (1 + I)Zp(l/ pi), Zp $0, 0$ Zp) E rep(Sz , Zp, j). PROOF.
Example 4.3.6 Let p be a prime, 5
/ \
3
T
4
/ \ /
=1
2
\ /
°
142
4. Representations over a Discrete Valuation Ring
a p-Iocally free finite latticeoftypes. and Xi a subgroup ofG. with 1 E Xi, type Xi = i, and 1/ p r/. Xi. Then B(T, O)p has infinite representation type, but each indecomposable has rank less than or equal to 2. Indecomposable groups in B(T, O)p ofrank 2 areoftheform G = X4 EeX3 + X z(1, 1)/ pi for some j :::: 1. Moreover, B(T, j) p has indecomposable groups ofarbitrarily large finite rank for j :::: 4. PROOF.
Notice that Sr
= C~ = 2 I
4 Apply Corollaries 4.3.2 and 4.1.4 to see that indecomposable groups have rank less than orequalto 2. If G is as definedabove, then (G p' G(4)p, G(2)p, G(1)p) = (V o, Zp Ee 0 f Vo,O Ee Zp) with Vo = Zp Ee Zp + Zp(1, I)/pi, as required by Corollary 4.1.4. The last statement of the theoremis a consequenceof Corollaries4.3.2 and4.1.9 and Example 4.1.10. In particular, for r = 5,
G = Xf Ee Xf Ee X~s Ee X~s
+ X4M + (1/ p5)(1 + 1 + 0 + O)Z2s ,
where M is the row space of the matrix pZ[
p3[
p3[
0
0 p5[
000 and A is an s x s Z-matrix such that A(mod p) is an indecomposable Z/pZ-matrix. Notice that Tcr(G) = {type Xi; 1 ::: i ::: 4}. In view of Corollary 4.3.2, the following open question is a group-theoretic version of a special case of open questions in Sections 4.1 and 4.2. Open Question: What is the representation type of B(T, j)p if Sr = (1,2) and 1 ::: j ::: 3 or 6 :::: j :::: 4? Can the indecomposable groupsbe classified in these cases?
EXERCISES
J. Let T be a finite lattice of types with Sr = {type XI, type Xz, type X3}. an antichain, for subgroups Xi of Q with 1/ p ¢ Xi for each i. Provethat the following construction demonstrates that B(T3, 2)p has wild modulo p representation type. Given two n x n Z-matrices A and B, define a subgroup G of Q'1 $ Q'1 $ Q'1 $ Q'1 by G
= Xr $
Xr + (I + 0 + 0 + I)X3+ (0 + 1 + 0 + I)X~
+ (I/pz)(l +0 + A + l)zn + (l/p)(O+ I + B +O)zn.
2. Provide a proof for Corollary 4.3.2.
4.3 NarES ON CHAPTER
Finite Rank Butler Groups and Isomorphism at p
143
4
There is very little literature on representations of posets over discrete valuation rings . The contents of this chapter are distilled from [Arnold, Dugas 97, 99]. The methods , albeit ad hoc, are sufficient to determ ine the representation type of most of the isomorphism at p categories of finite rank Butler groups. In addition , indecomposable representations give rise to groups that are indecomposable in the isomorphism at p category, hence indecomposable as groups . The representation types of some more general categorie s of representations of posets over discrete valuation rings and their factor rings are given in [Plahotnik 76] and [Simson 96].
5 Almost Completely Decomposable Groups
5.1 Characterizations and Properties An almost completely decompo sable group is a torsion-free abelian group G of finite rank quasi-isomorphic to a completely decomposable group. An almost completely decomposable group is a Butler group, by Coroll ary 3.2.4, but any strongly indecomposable Butler group with finite rank greater than I is not almost completely decomposable. Almost completely decomposable groups are quite complicated, as a consequence of the variety of ways that a completely decomposable group can be embedded as a subgroup of bounded index in an almost completely decomposable group. Chapter 5 is a brief introducti on to this technic al subject. Properties of finite rank Butler groups given in Chapter 3 are assumed. In particular, if C is a subgroup of a finite rank Butler group G with G/ C bounded , then G/C is finite by Exercise 2.1.1. Moreover, Tcr(G) = [r : G(r)/G#(r) =f:. O} is the critical typeset of G, and a regulating subgroup of G is B = ~ {Gr : r E Tcr(G)} with G(r) = G r EB G#(r) and G, a r-homogeneous completely decomposable group. The first characterization of almost completely decomposable groups is in terms of the ranks of the G r 'so
Proposition 5.1.1 Afinite rank Butler group G is an almost completely decomposable group if and only if rank G = ~{rank G(r )/G#(r ): r E Tcr(G)}. In this case, each regulating subgroup of G is a completely decomposable group isomorphic to EB{G(r)/G#(r ) : r E Tcr(G)}.
5.1
Characterizations and Properties
145
PROOF. Let G be an almost completely decomposable group and C a completely decomposable subgroup of G with G I C finite. Then rank G = rank C and Ter(G) = Ter(C) . Moreover, G(r)/C(r) and G#(r)/C#(r) are finite, whence rank G(r)/G#(r) = rankC(r)/C#(r) for each type r E Ter(G). But rank C = E{rank C(r)/C#(r) : r E Ter(C)} by Example 3.1.5,so that rank G = E{rankG(r)1 G#(r): r E Ter(G)} . Conversely, assume that G is a finite rank Butler group with rank G = E {rank G(r)/G#(r) : r E Ter(G)}, G(r) = G, E9 G#(r), and B = E{G, : r E Ter(G)} a regulating subgroup of G . Then G I B is finite by Propo sition 3.2.12. Hence, Ter(G) = Ter(B), rank G, = rank G(r)/G#(r) = rank B(r)1 B#(r) for each type r E Ter(G), and E{rank B(r)1 B#(r): r E Ter(B)} = rank B. It follows that B = E9{ G, : r E Ter( G)} is completely decomposable. Since G I B is finite, G is almost completely decomposable. There is a criterion for deciding whether or not a completely decomposable subgroup B of an almost completely decomposable group G is a regulating subgroup of G. Define a positive integer i (G) to be the least element of {I G I CI : C completely decomposable, G I C finite}.
Proposition 5.1.2 [Lady 74B] Let G be an almost completely decomposable group and B a completely decomposable subgroup of G. (a) The group B is a regulating subgroup of G ifand only iflG I BI = i(G) . (b) IfC is a completely decomposable subgroup ofG with GIC finite, then
i(G) divides IG/CI. (c) The group G is completely decomposable ifand only ifi(G) = 1.
(a), (b) The first step is to show that if Band C are distinct regulating subgroups of G, then IGI BI = IGIC I. Write B = ${B, : r E Tcr(G)} and C = ${C, : r E Ter(G)} with G(r) = B, E9 G#(r) = C, E9 G#(r) and B, and C, r-homogeneous completely decomposable groups for each r E Tcr(G) . The proof that IG I BI = IGI CI is an induction on the cardinality of {a E Ter(G) : B" =I- C,,}. Since Band C are distinct, there is some a E Tcr(G) with B" =I- C". Define D = C" E9 (${B, :a =I- r E Ter(G)}) = E9{D, : r E Ter(G)}, another regulating subgroup of G with I{a E Tcr(G) : B" =I- D,,}I < I{a E Ter(G): B" =I- C,,}I. By induction, it is now sufficient to prove that IGIBI = IGIDI . To this end, define E = B + C" , Then E = B + E(a) = D + E(a), since B" and C" are subgroups of E(a). Consequently, lEI BI = jE(a)1 B n E(a)1 = IE(a)1 B(a)1 and, similarly, lEI DI = IE(a)1 D(a)l. Notice that B(a) = E9{B, : r ::: a} and D(a) = C" E9 (E9{B, : r > a}) are regulating subgroups of E(a), and B#(a) = D#(a) = E9{B, : r > a} is a regulating subgroup of E#(a) . Now,
PROOF.
IE(a)1 B(a)1 = I(B" E9 E#(a))/(B" E9 B#(a))1
= IE#(a)1 B#(a)1
146
5. Almost Completely Decomposable Groups
and
= I(Cu ED E#(a»/(Cu ED D#(a»1 = IE#(a)/ D#(a)1 with IE#(a)/ B#(a)1 = IE#(a)/ D#(a)l . Thus, IE(a)/ B(a)1 = IE(a)/ D(a)1 and IE(a)/ D(a)1
IG/BI
= IG/EIIE/BI = IG/EIIE(a)/B(a) 1 = IG/ E I IE(a )/ D(a )1= IG/EIIE/DI = IG/DI ,
as desired. It remains to prove that if B is a completely decomposable subgroup of G and B is not a regulating subgroup of G, then there is a completely decomposable subgroup C of G such that IG/C1 is a proper divisor of IG/ BI. In this case, if B is a completely decomposable subgroup of G with IG/ B I = i (G), then B must be a regulating subgroup of G. Conversely, if B is a regulating subgroup, then B = i(G). Otherwise, there is a completely decomposable subgroup C of G with IG/C1 = i(G) < IG/BI. Thus, C is a regulating subgroup of G with IG/CI:/; IG/BI ,acontradiction. Now assume that B is a completely decomposable subgroup of G and B is not a regulating subgroup of G. Choose a maximal in Tcr(G) such that B(a) = BuEDB#(a)butG(a):/; BuEDG#(a).DefineC = GuED(ED{Br :a:/; r E Tcr(G)}), where G(a) = Gu ED G#(a) and B(r) = B; ED B#(r) for each type r , Then C is a completely decomposable subgroup of G. By the choice of a, B#(a) = C#(a) is a regulating subgroup of G#(a). Define E = C + Bu = B + Gu. Then E = C + D(a) = B + D(a), and as above, IE/CI = IE(a)/C(a)1 and lEI BI = IE(a)1 B(a)l . Now,
IG(a)/C(a)1
= IG(a)/ E(a)IIE(a)/C(a)1
and IG(a)/B(a)1 = IG(a)/E(a)IIE(a)/B(a)l ·
However, IG(a)/CCa)1 = I(Gu ED G#(a»/(G u ED C#(a»1 = I(Gu ED G#(a»/ tG; ED B#(a»1 is a proper divisor of IG(a)/B(a)1 = I(Gu ED G#(a»/(Bu ED B#(a»1 by the choice of a. Hence, IE/CI = IE(a)/C(a)1 is a proper divisor of IE(a)/B(a)1 = IE/BI , and so IG/C1 = IG/EIIE/C1 is a proper divisor of IG/BI
= IG/EIIE/BI·
Assertion (c) is an immediateconsequence of (a).
If Band B' are tworegulating subgroupsof an almostcompletely decomposable
group G, then IG/BI = IG/B'I = i(G) by Proposition 5.I.2(a). However, the following example shows that G/ B need not be isomorphicto G/ B' . Example 5.1.3 [Lady 74B] Let P be a prime and choose rank-l subgroups X), X2, X3, X4 of Q such that 1 E Xi, lip ¢ Xi, type X), typeX 2, and type X3
5.1 Characterizations and Properties
147
are pairwise incomparable, and type X4 = type XI n X2 1: type X3. Define G = Xl EEl X2 EEl X3 EEl X4 + Z(1, 1,0, O)lp + Z(O, 0,1 ,1)1 p. Then (a) B = XIEElX 2EElX3EElX4 isa regulating subgroup ofG with G I B isomorphic to ZI pZ EEl ZI pZ, and (b) if Y = X 4(1, 1,0, p)1p, then C = XI EEl X2 EEl X3 EEl Y is also a regulating subgroup of G with G I C isomorphicto ZI p2Z. PROOF. (a) Observe that G is an almost completelydecomposablegroup. Let ri = type Xi . Then r3 and r4 are incomparable, since r4 1: r3 by assumption and r3 ~ r4 ~ rl is a contradiction. Hence, G(ri) = Xi with G#(ri) = for i = 1,2,3. Since r4 = rl n r2, G(r4) = X4 EEl G#(r4) with G#(r4) = (XI EEl X2 ) + Z(1, 1,0,0)1 p. This shows that B is a regulating subgroup of G. Clearly, G IBis isomorphic to ZI pZ EEl ZI pZ . (b) On the other hand,
°
G(r4) = (X 4 EEl XI EEl X2) + Z(1, 1,0, O)lp = YEEl«X I EEl X2) + Z (1, 1,0,0)lp)= YEElG#(r4),
observing that (1, 1,0, p)lp = (1, 1,0,0)lp + (0,0,0, 1) E G. Hence, C is a regulating subgroup of G. Finally, G I C = ZI p2Z, since G I C is generated by x = (0, 0, I, 1)1 p + C with order x = p2. This holds because px = (0,0, 0, 1) + C = -(1, 1,0 ,0)1 p + C with (0,0,0,1) rt C, and p2x = p«O, 0, 0, 1) + C) = (O,O ,O,p)EC. Corollary 5.1.4 [Lady 74B] Suppose G and H are almost completelydecom-
posable groups. (a) IfG is nearly isomorphicto H, then i(G) = i(H). (b) IfG = HEll K , then i(G) = i(H)i(K).
PROOF. (a) Choose a monomorphism f: G -+ H with IHlf(G)1 relativelyprime to i(H). By Proposition 5.1.2(b), i(H) divides HI f(B), for a regulating subgroup B ofG, since f(B) is a completelydecomposablesubgroupof H. But IHlf(B)1 = IHlf(G)llf(G)lf(B)I, so i(H) divides If(G)lf(B)1 = IGIBI = i(G). A symmetric argument shows that i(G) divides i(H). (b) If C = EEl{C r : r E Tcr(H)} is a regulating subgroup of Hand D = EEl{D r : r E Tcr(K)} is a regulating subgroup of K , then B = C EEl D is a regulating subgroup of G. This holds because G(r) = H(r) EEl K(r) = Cr EEl Dr EEl H#(r) EEl K#(r) = C, EEl Dr EEl G#(r) for each type r. By (a), i(G) = IGI BI = IHI CIIK I DI = i(H)i(K). The exponent of a bounded abelian group A, written e = exp A, is the least positive integer e with e A = 0. If G is a finite rank Butler group, then G I R( G) is finiteby Corollary 3.2.13(a), where R(G) is the intersectionof all of the regulating subgroups of G.
148
5. Almost Completely Decomposable Groups
Define eG = exp G/ R(G) and br(G) = exp G#(r)/ R(G#(r» for each type r, adoptingthe convention that s.t G) = 1if G#(r ) = O. Noticethat G#(r)/R(G#(r» is finite for each type r , since G#(r ) is a Butler group, beinga pure subgroupof the Butlergroup G. Moreover, br(G) is the least commonmultipleof{exp G#(r)/ D : D regulating subgroup of G#(r)}, since R(G#(r» = n{D : D regulating subgroup of G#(r)}. The br(G)'s are called the Burkhardt invariants of G. The following lemma gives a procedure for constructing R(G) from the Burkhardt invariants and a given regulating subgroup of an almost completely decomposable group G. Lemma 5.1.5 [Burkhardt 84] Let G be an almost completely decomposable group and B = EI7{B r : r E Tcr(G)} a regulating subgroupofG. Then R(G) = EI7{b r(G)Br : r E Tcr(G)}. Write br for br(G) and let C = EI7{ Cr : r E Tcr(G)} be an arbitraryregulating subgroup of G. Then G(r) = B, EI7 G#(r) = C, EI7 G#(r) for each r E Tcr(G). If r E Tcr(G), then EI7{CO': r < a} is a regulating subgroup of G#(r), since G(a) = G(r)(a) if r < a . Then b.B; ~ brG(r) = brC r EI7 brG#(r) ~ b.C, EI7 R(G#(r» ~ Cr EI7 (EI7{Ca : r < a}) = C(r) ~ C, using the fact that brG#(r) ~ R(G#(r» . Since C is an arbitrary regulating subgroup of G, EI7{b rBr : r E Tcr(G)} ~ R(G). Conversely, let x E R(G)\(EI7{b rBr : r E Tcr(G)}). Since R(G) ~ B, it suffices to assume that x = EI7{x r E B, \b r B; : rES} for some subset S of Tcr(G). In particular, b, =I- 1 for each rES. Let r be a maximal element of Sand write B r = AI EI7 ... EI7 An with each Ai a rank-I group of type t . Then X r = al EI7 . .. EI7 an for some a; E A i. Because x, rt brBr, there is some i, say i = 1, with a, rt b.B«. Since b, = exp G#(r)/ R(G#(r», there is some a E G#(r) with order (a + R(G#(r» = br =I- 1 in G#(r)/ R(G#(r» . Let D = EI7{D r : r E Tcr(G#(r»} be an arbitrary regulating subgroup of G#(r) and define C = EI7{CO' : a E Tcr(G)}, where Cr = A~ EI7 A2 EB··· EB An with A ~ the pure rank-I subgroup of G generated by al + a , CO' = DO' if a > r, and CO' = BO' if a t r. Then C is a regulating subgroup of G, so that x E R(G) ~ C, say x = EB{cO' E C u : a E Tcr(G)}. Now,O = x - x = I:{xu - CO' : a E Tcr(G)} , so that Xr - Cr = I:{cu - Xu : a =Ir] E G(r). Since r is maximal in S, X a = 0 if a > rand X r - Cr = I:{cu : a > r] E G#(r). Also, X r - Cr = (al EB · . . EBa n ) - (a~ El7a~ EI7 . .. EB a~) E Br EI7 Cr with a; E A 2 ~ B; for 2 ~ i anda; = ql(al+a)forsomeql E Q. Thenxr-cr+qla = (al -qla»EI7(EB{aj -a;:2 ~ i ~ n}) = I:{cu :a > r}+q la E B; nG#(r) = 0, since (al - qlal) EB (EI7{aj - a;: 2 ~ i ~ n}) E B; and a, cuG#(r). Consequently, al = qlaJ, ql = I, and a = I:{-cu :a > r} E D , recalling that CO' = DO' if a > r . Since D is an arbitrary regulating subgroup of G#(r), a E R(G#(r», a contradiction to the choice of a. PROOF.
Theorem 5.1.6 (a) [Burkhardt 84] If B = EB{G r : r E Tcr(G)} is a regulating subgroup of an almost completelydecomposable group G, then R(G) = I:{br(G)G(r):
5.1 Characterizations and Properties
149
r E Ter(G)} is a completely decomposable fully invariant subgroup
ofG. (b) [Mader, Vinsonhaler 94] If G and H are nearly isomorphic almost completelydecomposablegroups, then R( G) is isomorphic to R(H) and G/ R( G)
is isomorphic to H/ R(H). PROOF. (a) By Lemma 5.1.5, R(G) = ${brB r : r E Ter(G)}. Hence, R(G) is a completely decomposable subgroup of G contained in 'E{brG(r) : r E Ter(G)}. Conversely, in view of the definition of b, and Lemma 5.1.5, if r E Ter(G), then brG(r) = b.B; $ brG#(r) S; b.B; $ R(G#(r» S; R(G) as a consequence of Exercise 5.1.12(b). Consequently, 'E{brG(r): r E Ter(G)} S; R(G), as desired. If f E End G, then f(R(G» S; R(G) , since R(G) = 'E{brG(r): r E Ter(G)}, b r is an integer, and f(G(r» S; G(r) for each r. This shows that R(G) is a fully invariant subgroup of G. (b) Since G/ R(G) is finite, G is quasi-isomorphic to R(G). But G is nearly isomorphic to H, so that R(G) is quasi-isomorphic to R(H) and Ter(G) = Ter(H). By (a), R(G) and R(H) are completely decomposable groups, whence R(G) is isomorphic to R(H) by Theorem 3.1.7(b). In view of Theorem 2.2.2(a), it is sufficient to assume that there is an integer m relatively prime to ea = exp G/ R( G) with m H S; G S; H. Then m H (r) S; G(r) S; H(r) for each type rand br(G) = br(H) for each r E Ter(G). Hence, mR(H) S; R(G) = 'E{br(G)G(r) : r E Ter(G)} S; R(H) = 'E{br(H)H(r): r E Ter(G)}. Since eG = exp G/ R(G) is relatively prime to m, G/ R(G) is isomorphic to H/R(H).
If Hand K are almost completely decomposable groups, then H $ K is an almost completely decomposable group with R(H $ K) S; R(H) $ R(K) (Exercise 5.1.12(a» . However, the group constructed in Example 5.1.3 is an example of an almost completely decomposable group H$K with R(H$K) # R(H)$R(K). Example 5.1.3 (continued) Let G be as defined in Example 5.1.3. Then (c) G = H$K, where H = (X I$X2)+ Z(l, 1,0, O)/pand K =(X3 $ X 4) + Z(O, 0, I , I)/p; (d) R(H) = X I $ X 2 and R(K) = X 3 $ X4; and (e) R(G) = XI $ X2 $ X3 $ pX 4 • Statement (c) is immediate from the definition of G . As for (d), the critical typeset of H is {rl, r2} with rl and r2 incomparable and H(ri) = Xi. Then R(H) = XI $ X 2 . Alternatively, R#(H(ri» = 0, so that the Burkhardt invariants for H are b, = I for r = rio Hence , R(H) = XI $ X2 by Theorem 5.1.6(a). Similarly, R(K) = X 3 $ X 4 • (e) In this case, there is not a unique regulating subgroup of G , by Example 5.1.3(a) and (b). However, the Burkhardt invariants b, of G can still be computed. Now, Ter(G) = {rl , r2, r3, r4} with G#(ri) = for i = 1,2,3 and
PROOF.
°
150
5. Almost Completely Decomposable Groups
G#(r4) = (XI EB X 2 ) + Z(l , 1,0, O)/p. Hence, b r = 1 for r = ti, 1 ~ i ~ 3, and b r = p for r = r4. By Theorem 5.1.6(a), R(G) = I:{brG(r): r E Tcr(G)} =
XI EB X2 EB X3 EB pX4 •
The next examplecompletesthe computation of the integer n( G) for an almost completeIy decomposable group G mentioned just after the proofof Lemma2.2.7. Example5.1.7 If G is an almost completely decomposable group and eG is the exponent of G/ R(G), then eG = neG). PROOF. Since C = R(G) is a fully invariant completely decomposable subgroup of G, restriction induces a ring embedding ec End C £; End G £; End C by Theorem 5.1.6(a). Then eG(End C/ NEnd C) £; End G/ NEnd G £; End C/ NEnd C. By Proposition 3.1.8, End C/NEnd C is isomorphic to Matn(l)(R 1) x . .. x Matn(m)(R m) with each R, a subring of Q. Since a subring of Q is a principal ideal domain, each Matn(i)( R i ) is a maximalorder over Ri, as noted in Section2.2. Apply the definition of neG) to complete the proof. Parts (b), (c), and (d) of the nextcorollaryare consequences of Corollary2.2.11, the proof of which uses the fact that maximal orders are hereditary. A grouptheoretic proof for the special case of almost completely decomposable groups is a consequence of Exercise 5.104 and the proof of Corollary 2.2.11 . Corollary5.1.8 Suppose G and H are almost completely decomposable groups. (a) The groups G and H are nearly isomorphic ifand only ifthere is a completely decomposable group C with G EB C isomorphic to H EB C. In this case, C
may be chosen to be isomorphic to both R(G) and R(H). (b) The group H is a near summand of G if and only if G X EB X' for some X and X' with X nearly isomorphic to H . In particular, G is indecomposable if and only if and G are the only near summands of G. (c) IfG EB K is nearly isomorphic to H EB K for some torsion -free abelian group K offinite rank, then G is nearly isomorphic to H. (d) JfG is nearly isomorphic to GI EB · ·· EB G«. then G = HI EB·· · EB H; with each Gi nearly isomorphic to Hi.
°
=
PROOF. (a) SupposeG and H are nearlyisomorphic. Then C = R( G) is completely decomposable, C is isomorphic to R(H) by Theorem 5.1.6(b), and eG £; C with e = exp G/ R( G). By Theorem 2.2.2(a), there is f E Hom(G, H) and g E Hom(H, G) and a nonzero integer m prime to e such that gf = mIG and fg = mlH. Write 1 = rm + se for integers rand s and define ¢ : G ~ HEBe by ¢(x) = (fx, ex) and a: H EB C ~ G by a(x , y) = rgx + sy . Then a¢ = IG, so that H EB C is isomorphicto G EB D for some D. Since G and H are nearlyisomorphic, C is nearlyisomorphicto D by Corollary 2.2A(a). Thus, D is isomorphic to a summand of the completely decomposable
5.1 Characterizations and Properties
151
group C E9 C , by Corollary 2.2.6, hence completely decomposable by Theorem 3.1.7(a). Therefore, C is isomorphic to D, since quasi-isomorphic completely decomposable groups are isomorphic by Theorem 3.1.7(b). Conversely, if G E9 C is isomorphic to H E9 C , then G is nearly isomorphic to H by Corollary 2.2.4(a). Statement (b) is a consequence of Corollary 2.2.11, (c) is a special case of Corollary 2.2.4, and (d) follows from (b) and (c). There are examples of almost completely decomposable groups G and H with no rank-I summands such that G E9 H has a rank-I summand (Exercise 5.1.7). However, the next lemma demonstrates that an almost completely decomposable group contains a maximal completely decomposable summand that is unique up to isomorphism. Proposition 5.1.9 [Lady 74B] Assume that G is an almost completely decomposable group. (a) Then G = C E9 H , where C is completely decomposable and H has no
rank-l summands. (b) If G = C' E9 H ' is another such decomposition of G, then C is isomorphic to C' and H is nearly isomorphic to H '.
(a) Since G has finite rank, it follows immediately that G has such a decomposition. (b) First assume that C and C' are r-homogeneous completely decomposable. Then H n C' is a pure subgroup, hence a r-homogeneous completely decomposable summand, of C' by Lemma 3.1.6. Write C' = (H n C') E9 D . Then G = C' E9 H ' = (H n C') E9 D E9 H' , from which it follows that H n C' is a summand of H . But H has no rank-I summands, whence H n C' = o. Let f : G ~ C be a projection with ker f = H. Since H n C' = 0, f : C' ~ C must be a monomorphism. Similarly, there is a monomorphism g: C ~ C'. This shows that rank C = rank C'. Since C and C' are r-homogeneous completely decomposable groups of the same rank, they must be isomorphic. Moreover, H is nearly isomorphic to H ' by Corollary 5.1.8(c) . For the general case, let r be a maximal type in the union of the typeset of C and the typeset of C'. Then C(r) and C'(r) are r-homogeneous completely decomposable groups by the choice of r. Write C = C(r)E9 D and C' = C'(r) E9 D'. Then G = C(r) E9 (D E9 H) = C'(r) E9 (D' E9 H'), and the groups D E9 Hand D' E9 H' have no rank-I summands of type r. The argument of the preceding paragraph shows that C(r) is isomorphic to C'(r) and D E9 H is nearly isomorphic to D' E9 H' with D and D' completely decomposable. In view of Corollary 5.l.8(d), it suffices to assume that D E9 H = D' E9 H'. By induction on the rank of G, D is isomorphic to D' and H is nearly isomorphic to H ' . Thus , C = C(r) E9 D is isomorphic to C' = C'(r ) E9 D',
PROOF.
152
5. Almost Completely Decomposable Groups
There is a characterization,in terms of thejoin irreducibleelements ST, of those finite lattices T such that each finite rank Butler group with typeset contained in T is almost completely decomposable. Corollary 5.1.10 Let T be a finite lattice of types. Each group in B(T) is an almost completely decomposable group if and only if W(ST) :::: 2.
If W(ST) ::: 3, then there is a strongly indecomposable group G in B(T) with rank G = 2 by Example 3.3.3. In this case, G cannot be quasi-isomorphic to a completely decomposable group. Conversely, if W(ST) :::: 2, then each strongly indecomposable group in B(T) has rank 1, as a consequence of Theorem 3.3.2 and Example 1.1.3. Hence, each G in B(T) is quasi-isomorphic .to a completely decomposable group. PROOF.
The remainderof this sectionis devotedto thecase that T is a finitelatticeof types with ST = S2 = {r, a}. If G E B(T), then G is almost completely decomposable by Corollary5.1.10. If, in addition, G has no rank-l summandsand G(r)nG(a) = 0, then Tcr(G) = S2 and R(G) = G(r) $ G(a) is a unique regulating subgroup of G. It followsfromTheorem 3.4.5 that G = GI $ . . . $ Gn is a finitedirect sum of indecomposablegroups of rank 2. Moreover, R(G) = G(r) $ G(o;) = (GI(r) $ GI(a»$ '" $(Gn(r) $ Gn(a» = R(G I)$ '" $R(G n) with GjR(G) = Gti R(Gr) $ . .. $ Gnj R(G n) and each G;/ R(Gi) a finite cyclic group. The following proposition is the converse in that a decomposition of the finite group G j R( G) intoadirectsum of cyclicgroupsgivesrise to acorrespondingdirect sumdecomposition of G intoindecomposable groupsof rank lessthan or equalto 2. Proposition 5.1.11 [Mader, Vinsonhaler95] Suppose T is afinite latticeoftypes with ST = [r, a} . LetG E B(T) with no rank-l summandsandG(r)nG(a) = o and write G j R(G) = CI E9 .. . $ Cn with each C, a cyclic group. Then
G = G I E9 . . . $ G; for some indecomposable almost completely decomposable group G, of rank 2 with G i j R(G i) = C, for each 1 :::: i :::: n.
PROOF. Assume that C, is a cyclic group of order ai and let X and Y be subgroups of Q containing 1 with type X = r and type Y = a. Choose Xi E G(r) and Yi E G(a) with z, = (Xi + Yi)jai E G, z. + R(G) a generator of Ct. {XI, X2, ... ,X n, YI, , Yn} a Z-independent subset of G, and C = (XXI $ . .. $ Xx n) $ (YYI $ $ YYn) a pure subgroup, hence a summand, of R(G) = G(r) $ G(a). This can be done inductively, observing that G(r) and G(a) are non-zero homogeneous completely decomposable groups with G(r) n G(a) = 0 and by recalling that G has no rank-l summands. Define Gi = (XXi $ YYi) + ZZi, a subgroup of G. Then G;/ R(G i) = C, and (G I $ . . '$G n)+ R(G) = Gas a consequenceofthefactthat each Zi is a generator of Ci' Now, C = (XXI $ . .. $ Xx n) $ (YYI $ . . . $ YYn) = C(r) $ C(a) is a summand of R(G) = G(r) $ G(a), say R(G) = C $ D. Since GjC = Gj R(G), it follows that D is a summand of G. But G has no rank-l summands, whence
5.1
Characterizations and Properties
153
R(G) = C ~ G I EB . .. EB Gn . Thus, G = G I EB .. . EB G; with rank G, = 2 and G;/ R(G j ) = Ci, as desired.
Corollary 5.1.12 Suppose G is afinite rank Butler group with typeset ~ Tz = {r n a , r , a} and r and a incomparable. Then G is a finite direct sum of indecomposable groups with rank less than or equal to 2. PROOF. By Proposition 5.1.9, it is sufficient to assume that G has no rank-1 summands. In view of the hypotheses, G(r U a) = G(r) n G(a) = O. Since the finite group G I R(G) can be written as a direct sum of cyclic groups, Proposition 5.1.11 applies. Proposition 5.1.11 can be used to demonstrate that direct sum decompositions of almost completely decomposable groups are not unique up to near-isomorphism even if the critical typeset has 2 elements. As an illustration, suppose G satisfies the hypotheses of Proposition 5.1.11 with G I R(G) = Z/30Z . Then G I R(G) = Z/2Z EB Z/3Z EB Z/6Z = Z/6Z EB Z/5Z = Z/lOZ EB Z/3Z = Z/15Z EB Z/2Z. Each of these decompositions of G I R( G) into cyclic groups induces a direct sum decomposition of G into indecomposable groups of rank 2. These decompositions are not unique up to near-isomorphism by Theorem 5.1.6(b), since if X and X' are nearly isomorphic summands of G, then X I R(X) is isomorphic to X'I R(X'). On the other hand, if G I R(G) is a p-group for some prime p, necessarily finite, then any two decompositions of G I R( G) into cyclic groups are unique up to isomorphism and order. This uniqueness carries over to G , since it is shown in Corollary 5.4.3 that if G I R(G) is a p-group, then direct sum decompositions of an almost completely decomposable group G into indecomposable groups are unique up to isomorphism at p.
EXERCISES
1. Prove that if G is an indecomposable finite rank Butler group with critical typeset G ~ Tz = Iri . rz} and rank G = 2, then G is isomorphic to XI ffi Xz + Z(I, m)/n, where m and n are relatively prime integers with n > I and X, is a subgroup of iQ containing 1 with type Xj = rj. 2. Find a complete set of near isomorphism invariants for finite rank Butler groups G with typeset G ~ h 3. [Cruddis 70], [Amold, Dugas 93B] Let G be a finite rank Butler group with Tcr(G) ~ T3 = {rl, rz, r3} and rl + rz + r3 = type iQ. Prove that G is the direct sum of groups of rank less than or equal to 2. Find a complete set of near-isomorphism invariants for G. 4. Let G be an almost completely decomposable group with eGG ~ R(G) ~ G and K a near summand of G. Prove that if g: G --+ K, then there is h: g(G) --+ K with ghg = eag , Conclude that X = g(G) is a summand of G nearly isomorphic to K . 5. Let G be an almost completely decomposable group with eGG ~ R(G) ~ G and D the unique maximal subgroup of G with eGD = D. Write R(G) = ffi{A r : r E
154
5. Almost Completely Decomposable Groups Tcr(G)}, A r a r-homogeneous completely decomposable group . Prove that: (a) D E9{A, : r E Tcr(G), eGA, A,}; and
= =
=
D E9 E , where E is the pure subgroup of G generated by E9{A, : r E Tcr(G), eGA, =1= A r }.
(b) G
6. Assume that G is an almost completely decomposable group with Tcr(G) an inverted forest. Compute the Burkhardt invariants of G .
=
7. (a) [Jonsson 57] Give an example of an almost completely decomposable group G C E9 H = C ' E9 H ' such that C and C ' are completely decomposable groups , H and H' have no rank-I summands, but H and H ' are not isomorphic. (b) [Jonsson 57] Give an example of two almost completely decomposable groups G and H with no rank-I summands such that G E9 H has a rank-I summand. 8. Find almost completely decomposable groups G and Hand that f(R(G» is not contained in R(H).
f
E Hom(G , H) such
=
9. [Mader 95] Let G be an almost completely decomposable group and define G(O) G, G(I) = :E{G#(r) : r E Tcr(G)}, and, inductively, G(n+l) = (G(n»)O). (a) Prove that G(n) 0 for some n and that G(O) 2 G(1) 2 . .. 2 G(n) 0 is a descending chain of purely invariant subgroups of G. (b) Prove that G(IT(G»/G(1) is isomorphic to E9{G(r)/G#(r) :r is minimal in
=
=
Tcr(G)} .
(c) Prove that i(G) = IG/G(IT(G))i IG(1) /G(1)(IT(G(1»I · · · IG(n)/G(n)(IT(G(n»I. 10. [Burkhardt 84] Let G be an almost completely decomposable group and C a completely decomposable subgroup of G with G/ C finite. Prove that C is a regulating subgroup of G if and only if there is a decomposition C = E9{ C, : r E Tcr (G)} such that each for each r, C, 5; (expG#(r)/C#(r»G(r) 5; C(r). II . [Burkhardt 84] Let G be an almost completely decomposable group . Prove that if G 7l/ p n71 and G/ C (71/ p71)n, has regulating subgroups Band C such that G/ B then for each finite abelian group A with IAI = pn there is a regulating subgroup D ofG with G/D A.
=
=
=
12. Let G and H be almost completely decomposable group s. (a) Prove that R(G E9 H) 5; R(G) E9 R(H). (b) Prove that if r E Tcr(G), then R(G#(r» 5; R(G) .
5.2
Isomorphism at p and Representation Type
The representation type of the category of almost completely decomposable groups in B(T, j) is computed in Corollaries 5.2.4 and 5.2.9 by determining the representation type of a subcategory of rep(ST , Zp, j). Indecomposable representations in rep(ST, Zp, j) can be used to construct indecomposable almost completely decomposable groups in B(T, j). Let T be a finite lattice of types, j a nonnegative integer, and C(T, j) the full subcategory of B(T, j) consisting of almost completely decomposable groups. Specifically, the groups in CCT, j) are almost completely decomposable groups G with typeset G 5;; T and p! G 5;; b {G( r ) : rEST} 5;; G . Since ST consists
5.2 Isomorphism at p and Representation Type
155
of the join-irreducible elements of T, :E{G(r): r E Sr} = G *(IT(G». Henceforth , it is assumed that j ::: 1, since C(T, 0) consists only of almost completely decomposable groups (Exercise 5.2.2). For almost completely decomposable groups, there is little loss of generality, up to isomorphism at p, in assuming that T is a p -locally free lattice of types . By Proposition 5.1.9, an almost completely decomposable group G is the direct sum of a completely decomposable group C and an almost completely decomposable group H with no rank-l summands. Moreover, H is unique up to near isomorphism. As a consequence of Proposition 4.3 .1, if G is an almost completely decomposable group with no rank-l summands, the typeset of G is contained in T , and p is a prime, then G is isomorphic at p to some H E C(T, j) with en = exp H j R(H) a power of p . If D is the maximal p-divisible subgroup of H , then D is completely decomposable and H = DEBE for some E with typeset E a p-locally free set of types (Exercise 5.1.5). Hence, if H is indecomposable at p with rank H > 1, then H E C(T' , j) for some finite p-locally free lattice T ' . Recall from Corollary 4.3.2 that if T is a p -locally free finite lattice of types and j ::: 1 an integer, then there is a category equivalence Fr : B(T, j)p ~ rep(Sr, Zp, j) given by Fr(G) = (G p: G(r)p: r E Sr). Define Crep(Sr , Zp , j) to be the full subcategory ofrep(S, Zp, j) with objects of the form Fr(G) , G E C(T, j). Clearly, the categories C(T, j)p and Crep(Sr, Zp, j) are equivalent categories.
Lemma 5.2.1 1fT is afinite lattice, T' is a sublattice ofT, and j' :::: j, then there is a fully faithful functor F: Crep(Sr', R , j') ~ Crep(Sr , R, j). PROOF. A consequence of the definition of Crep(Sr, Zp , j) and the observation that C(T', j) is contained in C(T, j). Specifically, F = Fr(FT'r l • It is not all that easy to identify those representations in rep (Sr, Zp , j ) that are in Crep(Sr , Zp, j); see Exercise 5.2 .1. By Proposition 5.1.1 , G E B(T) is an almost completely decomposable group if and only if rank G = :E{rank G(r)j G#(r) : r E Tcr ( G)}. However, Tcr ( G) need not be contained in Sr, the join irreducible elements of T . For example, let T be a Boolean algebra with exactly 3 atoms. Then Sr consists of these three atoms . If G is an almost completely decomposable group with critical typeset consisting of three pairwise incomparable elements of T that are not atoms, then the critical typeset of G is clearly not contained in Sr . Define Ccrit(T, j) to be the category consisting of all almost completely decomposable groups in C(T, j) with critical typeset contained in Sr , the set of join irreducible elements of T. The image of the functor Fr on groups in Ccrit(T, j), denoted by CrePcrit(Sr , Zp, j), consists of those representations V = (Vo, Vi : i E Sr) E Crep(Sr, Zp, j) such that rank Vo = :E{rank Ui] Vr : i E Sr}, where Vr is the purification of :E{Vs : s < i in Sr} in Vi , as a consequence of Proposition 5.1 .1 and the definition of Fr. For each such representation, Vo is a finitely generated free Zp-module and each Vi is a summand of Vo.
156
5. Almost Completely Decomposable Groups
Notice that if Crepcrit(Sr , 7lp , j) has infinite or wild modulo p representation type, then Crep(Sr, 7lp , j) has, respectively, infinite or wild modulo p representation type by Lemma 5.2.1. This is because CrePcrit(Sr, 7lp , j) embeds as a full subcategory of Crep(Sr, 7lp , j). The representations constructed in the following examples are in CrePcrit(Sr, 7lp , j), hence automatically in Crep(Sr, 7lp , j).
Example 5.2.2 Let T be a p-locallyfree finite lattice oftypes. If either w(Sr) = 5, j ::: 1; w(Sr)::: 4 and j ::: 2; or w(Sr) = 3 and j ::: 3, then Crep(Sr, 7lp , j) has wild modulo p representation type. Let r be a 7lp-algebra that is finitely generated and free as an 7lp-module and A and B two n x n 7lp-matrices with C(A, B) = I', guaranteed by Lemma 4.2.I(a). (i) w(Sr) = 5 and j ::: 1. By Lemma 5.2.1, it is sufficientto prove that CrepeSs, 7lp , 1)haswildmoduloprepresentationtype.DefineU=(Uo, UI, UZ, U3, U4, Us) E Crep(Ss, 7lp , 1) by PROOF.
Uo = 7l; EI7 7l; EI7 7l; EI7 7l; EI7 7l;
+ (II p)(1 + 0 + 1 + A + B)71; ,
+ (llp)(O + 1 + 1 + 1 + 1)71;, UI = 7l; EI7 0 EI7 0 EI7 0 EI7 0, U2 = 0 EI7 7l; EI7 0 EI7 EI7 0, U3 = EI7 0 EI7 7l; EI7 EI7 0, U4 = EI7 EI7 EI7 7l; EI7 0,
°
°
Us = 0 EI7 0 EI7 0 EI7 0 EI7 7l; .
°°°
°
As a consequence of Lemma 4.2.3(a), End U is isomorphic to End V, with V = (Vo , UI, Uz, U3, U4, Us, pUo) E rep/(S6, 7lp ) and Vo = UI EI7 U2 EI7 U3 EI7 U4 EI7 Us = 7l~n . Moreover, pll« = pVo + W , where W is the free 7lp-module with the rows of the 7lp-matrix M = (~I~I~I~I~) as a basis and I an n x n identity matrix. To see that there is an onto ring homomorphism End V ~ C(A(mod p), B(modp)), let f E End V . Then f can be written as (f1,h,h,f4,/s) with each fi a 7lp-endomorphism of Ui, and f(pUo) S; pll«. In matrix terms, fM == M g(mod p) for some g. This equation implies that h == h == f4 == fs(mod p) (from the second row) and fl == h(mod p) with itA == Af4(mod p) and fIB == Bfs(modp) from the first row). Hence, f(modp) = (a,a,a,a,a), with a E C(A(mod p), B(mod p)). Then ¢ : End V ~ C(A(mod p) , B(mod p)), given by ¢(f) = a, is a ring homomorphism. Now, C(A, B) embeds in End V, viaa H- (a, a, a, a, a), whence ¢ is onto by Lemma 4.2.1. Since End U is isomorphic to End V, CrepeSS, 7lp' 1) has wild modulo p representation type. (ii) w(Sr) ::: 4 and j ::: 2. Onceagain,it is sufficient toprovethat Crep(S4, 7lp , 2) has wild modulo p representation type. Define U = (Uo, UI, U2, U3, U4) E
5.2 Isomorphism at p and Representation Type
V o = Z; ED Z; ED Z; ED Z;
157
+ (1/p2)(1 + I + A + B)Z;
+ (1/p)(O + I + I + 1)Z;, VI
= Z; ED 0 ED 0 ED 0, V 2 = 0 ED Z; ED 0 ED 0,
V 3 = 0 ED 0 ED Z; ED 0, and V 4 = 0 ED 0 ED 0 ED Z;. By Lemma 4.2.3(a), End V is isomorphic to End V, with V = (Vo, VI, V 2, V3, V 4, p2Vo) E repJ(Ss, Zp) and Vo = VI ED V 2 ED V3 ED V 4 = z~n. Then p2Vo = p2Vo + W, where W is the free Zp-module with the rows of the Zp-matrix
M_ I I pIAI pIB) - (I0 I pI as a basis. If I E End V , then I = (fl, h. h. 14) with each Ii a Zp-endomorphism of Vi and 1M == Mg(mod p2) for some g. This equation implies, from the first row of M, that II == h(mod p), f3A == A!I(mod p), and 14B == BII(mod p) . From the second row, h == 13 == 14(mod p). Hence, I(mod p) = (a, a, a, a) with a E C(A(mod p), B(mod p». Then ¢: End V --+ C(A(mod p), B(mod p) given by ¢(f) = a is a ring homomorphism and C(A, B) embeds in End V, via a r-+ (a , a , a , a). Thus, ¢ is onto by Lemma 4.2.1. Since End V is isomorphic to End V, rep(S4 , Z p, 2) has wild modulo p representation type, as desired. (iii) W(ST) = 3, j :::: 3. Define V E Crep(S3, Zp, 3) by Uo = z~n ED z~n ED z~n VI = z~n
+ (1/ p3)V*,
ED 0 ED 0, V2 = 0 ED z~n ED 0, V3 = 0 ED 0 ED z~n ,
and V* the free Zp-module with the rows ofthe Zp-matrix
M
I
=
0
(
o o
0 pI 0 0
0 pI 21 _p 0 pI o 0 o
Then End V is isomorphic to End V, where V = (Vo, VI, V2, V3, p3Vo) and Vo = VI ED V2 ED V3· If I E End V, then 1= (fl , h. 13) with Ii: Vi --+ Vi and 1M == Mg(mod p3) for some g. Write !I = (aij), h = (bij), and 13 = (cij) as 3 x 3 matrices with each aij, bij, and Cij an n x n Zp-matrix. The equation, 1M == M g(mod p3), yields, from the first row of M, all == b ll == c22(mod p) and a!2, an , C2h C23 all == 0 modulo p. Continuing with each row of M leads to the conclusion that a., == bu == cii(mod p) = a for each 1 ~ i ~ 3 with
158
a
E
5. Almost Completely Decomposable Groups
C(A(mod p), B(mod p)) and f(mod p)
N=
0 0 ( ail
= al + N with
0 0 0 b~2 0000 ai2 0 0 0
b~3 0 0 0) 0000 0 0 0 0
a nilpotent 7l/p71-matrix. Complete details of this computation are given in [Arnold, Dugas 97, Appendix] . The ring homomorphism if> : End V -;. C(A(mod p) , B(mod p)), defined by if>(f) = a, is onto by Lemma 4.2.1, since C(A , B) embeds in End V . Since End U is isomorphic to End V, Crep(S3 , 7lp , 3) has wild modulo p representation type. Finally, apply Lemma 5.2.1 to see that Crep(ST , 7lp , j) has wild modulo p representation type for W(ST) = 3 and j ::: 3.
Example 5.2.3 Let T be a p-locally free finite lattice of types. If W(ST) = 4 and j ::: 1 or w(Sr) ::: 3 and j ::: 2, then Crep(ST, 7lp , j) has rank-infinite representation type. Let A be an n x n 7lp-matrix with A(mod p) an indecomposable Z/ p71matrix. (i) W(ST) = 4 and j ::: 1. Define U = (Uo , UI, U2 , U3 , U4) E Crep(S4, 7lp , 1) by
PROOF.
Uo
UI
= 7l; EB 7l; EB 7l; EB71; + (1/p)(1 + 0+ 1+ l)71; + (1/p)(O + 1+ 1 + A)71; , = Z; EB 0 EB EB 0, U2 = EB 7l; EB 0 EB 0, U3 = 0 EB 0 EB Z; EB 0,
°
°
U4 = OEBOEBOEB71; .
As a consequence of Lemma 4.2.3(a) , End U is isomorphic to End V with V = (Vo , UI, U2, U3, U4, pUo) and Vo = UI EB U2 EB U3 EB U4. Then pUo = pVo + W, where W is the free R-module with the rows of the R-matrix M = (~I~I~I~) as a basis. An argument like that of Example 5.2.2 shows that if f E End V, then f(mod p) = aI, with a E C(A(mod p)). Iff is an idempotent, then a is an idempotent. Since A(mod p) is an indecomposable matrix, a = 0 or 1. If a = 0, then f2 = f E pEnd V . Since Vo is a finitely generated free Zp-module, f = O. On the other hand, if a = 1, then the same arguments show that 1 - f = 0 and f = 1. Hence, V, and consequently U is an indecomposable representation. Since there are indecomposable 7l/p71-matrices for each n, Crep(S4, 7l p , 1) has rank-infinite representation type. Now apply Lemma 5.2.1. (ii) W(ST) ::: 3 and j ::: 2. Define U = (Uo, UI, U2, U3) E Crep(S3, t.; 2) by
Uo = Z; EB Z; EB Z; + (1/p2)(1 + 1 + A)Z; + (1/p)(O + 1 + l)71;,
= Z; EB 0 EB 0, U2 = 0 EB 7l; EB 0, U3 = 0 EB 0 EB 7l;. Then End U is isomorphic to End V, with V = (VO, UI, U2 , U3, p2Uo) and Vo = UI
5.2 Isomorphism at p and Representation Type
159
VI ED V2 ED V3. In particular, p2Vo = p2Vo + W, where W is the free Zp-module with the rows of the Zp-matrix
M_(IIIIA) 0 pI pI as a basis . A by now familiar argument shows that if f E End V, then f(mod p) = al , with a E C(A(mod p» . As in the proof of (i), Crep(S3 , Zp , 2) has rank-infinite representation type . Again, apply Lemma 5.2.1. As a consequence of the preceding constructions, the representation type for C(T, j)p can be computed. The case that W(ST) = 3 and j = 1 is addressed in Corollary 5.2.9.
Corollary 5.2.4 Let T be a p-locally free finite lattice oftypes. Then C(T, j)p has: (a) wild modulo p representation type if either (i) WeST) ~ 5, j ~ 1; (ii) WeST) = 4, j ~ 2; (iii) W(ST) = 3, j ~ 3; or (iv) W(ST) = 2, ST is not a garland, and j ~ 7. (b) rank-infinite representation type if either (i) WeST) ~ 4, j ~ I; (ii) WeST) = 3, j ~ 2; or (iii) W(ST) = 2, ST is not a garland, and j ~ 4. (c) finite representation type if either (i) W(ST) = 1, j ~ 1 or (ii) W(ST) = 2, j ~ I, and ST is a garland. PROOF. As a consequence of Corollary 5.1. 10, C(T, j) = B(T, j) if W(ST) ~ 2. Thus, (a)(iv) , (b)(iii), and (c) follow from Corollary 4.3.3 . The remainder of the corollary is a consequence of the definition of Crep(ST , Zp , j) and Examples 5.2.2 and 5.2.3, wherein the representations constructed are in Crepcrit(ST, Zp, j).
An important consequence of Corollary 5.2.4 is the following corollary, demonstrating circumstances under which there are indecomposable almost completely decomposable groups of arbitrarily large finite rank . Representations constructed in this section can be used to give explicit descriptions of such groups ; see Exercise 5.2.3.
Corollary 5.2.5 Assume that T is a p-locallyfree finite lattice oftypes for some prime p. There are indecomposable almost completely decomposable groups ofarbitrarily large finite rank with typeset contained in T if either W(ST) ~ 4, j ~ 1; W(ST) = 3, j ~ 2; or W(ST) = 2, ST is not a garland, and j ~ 4.
160
5. Almost Completely Decomposable Groups
Use Corollary 5 .2.4 and the observation that if a group is indecomposable at p , then it is indecomposable.
PROOF.
=
=
Suppose X n {rl ' . .. , r n } is a p-locally free set of types with rj n rj ro for each 1 ~ i =P j ~ n. Define C(n) to be the category of almost completely U {roJ . Given j :::: 0, C(n , j) is decomposable groups with typeset G 5; the subcategory of groups G in CCn) with p jG 5; G*(ro) 5; G. In this case, G*(ro) = G(r))EB' . . EBG(rn ) = R(G). The category C(n , j)p is the isomorphism at p category of C(n, j). For example, X; could be the set ofatoms for a p-Iocally free Boolean algebra Tn. Then C(n, j) Ccrit(Tn , j), those almost completely decomposable groups in B(Tn , j) with critical typeset contained in X n , the join irreducible elements of
x,
=
r;
Corollary 5.2.6 The category C(n, j)p has: (a) wild modulo p representation type if either n :::: 5, j :::: 1; n = 4, j :::: 2; or n = 3, j :::: 3. (b) rank-infinite representation type if either n = 4, j :::: 1 or n = 3, j = 2. (c) finite representation type if and only if either n :s 2, j :::: lor n = 3, j = 1. A special case of Corollary 5.2.4, where T is the sublattice of the lattice of all types generated by X n, except for the case n = 3, j = 1. It is a consequence of Theorem 5.2.S(c) and Example 1.1.5 that CC3, l)p has finite representation type . PROOF.
Remark Let (S(n), j -1) be the posetdefined in Exercise 1.4.1. Observe that by comparison, the representation type of CCn, j)p is exactly that of rep«S(n), j - 1), k) , k a field, with the possible exception of n = 3, j = 2. A direct proof ofthis observation is not known .
:s
It remains to compute the representation type of CCT, j)p with W(ST) 3 and j = 1. Oddly enough, this apparently simpler case seems to require an additional hypothesis on ST. A finite poset S is a forest if it is a disjoint union of trees, equivalently, {s : s ~ i in S} is a chain for each i in S. For example, Sn is a forest. If ST is a forest, then CrePcrit(ST, 'Zp, j) has some special properties as given in (c) of the next lemma, the representation analogue of Theorem 3.2.5(b) for almost completely decomposable groups. The module EB{ V;* : i E ST} is analogous to a regulating subgroup for almost completely decomposable groups. If G is an almost completely decomposable group with critical typeset that is an inverted forest, then there is a unique regulating subgroup R(G) (Corollary 3.2.13). This property is reflected in CrePcrit(ST, 'Zp, j) in the case ST is a forest, remembering that ST is the poset of join irreducible elements of T with opposite order.
Lemma 5.2.7 Suppose T is a p-locally free finite lattice of types and V (Vo, U, : i E ST) E Crepcrit(ST, Zp, j).
=
5.2 Isomorphism at p and Representation Type
161
(a) For each i in ST, U, = V;* E9 Vr, where Vr is the pure submodule of U, generated by {Vi : j < i in ST}. (b) The Z p-module E9{U;*:i E Srl is a submodule of ~{Vj :i EST} with ~{Vj : i E Srl/(E9{V;* : i E ST}) and Ui] E9 {Vs* : s S i} bounded Zp modules for each i E ST. (c) /fST is aforest, then ~{Vj : i EST} = E9{V;* : i EST} and U, = E9{Vs* :s S i} for each i EST' (a) The pure submodule Vr is a summand of the finitely generated free Zp · module U: (b) If i is minimal in ST, then Vr = 0 and U, = V;*. Next assume that i is not minimal in ST and Vi/~{us*:s S j} is bounded for each j < i in ST. Then Vr 1~{Vi : j < i} is a bounded module, since Vr is a finitely generated free Zp. module and rank Vr = rank ~{Vi: j < i}. By induction, Ui] E9 {Vs* : s S i} is a bounded module for each i in ST. Hence, ~ {Vj : i E ST}I ~ {V;* : i E ST} must be a bounded module. Finally, ~ {V;* : i E Srl = E9{ V;* : i E Srl, since rank ~j V;* = rank ~ j Vi = rank V o = ~j rank Ui] Vr and Ui] Vr is isomorphic to V;* by (a). (c) In view of (b), it is sufficient to show that U, = ~ {V s* : s S i} for each i in ST. If i is minimal in ST, then U, = V;*. Now suppose i is not minimal in ST. Then C; = {s : s S i} is a chain, since ST is a forest. Thus, Vj# = VI for t the maximal element of Cj\{i}. By induction on the length of Ci, VI = ~{V;:s S t}. This shows that U, = ~{Vs* : s S i}.
PROO F.
There is a connection between CrePcrit(ST , Z p, j) and representations of S; over the finite ring Z I pi Z, recalling that S; is the disjoint union of ST and a point * unrelated to any element of ST. For a finite poset ST, let ~rep(ST , Z I pi Z) be the category of representations V = (Vo, U, : i E S;) such that Vo = ~ {Vi : i E ST} with each Vi a free Z I pi Z-module, Vi <; Vi if i S j in ST, and V* n U, = 0 for each i E S. Morphism sets are Zl piZ-homomorphisms f :Vo -+ V~ with f(V i) <; V; for each i E ST. Elements V of Mep(ST, Z I piZ) with Vo = E9{ Vi : i E ST} can be analyzed by solving matrix problems, as illustrated in Section 4.1 for rep(S, Zp, j). An additive functor F : X -+ X' between additive categories is a representation embedding if F preserves indecomposables and reflects isomorphisms. The functor F is a representation equivalence if F is a representation embedding that is dense, that is, for each Y in X' there is X in X with F(X) isomorphic to Y. A representation embedding, respectively equivalence, induces an injection, respectively a bijection, from Ind X to Ind X'. In the next theorem, (a) relates categories of Z p-representations of ST to categories of Z I pi Z representations and corresponding matrix problems, (b) generalizes the category equivalence of Lemma 4.2.3(b) for ST = Sn, and (c) and (d) are applications of (a) and (b) that relate the representation type of categories of almost completely decomposable groups to categories of representations of finite posets
162
5. Almost Completely Decomposable Groups
over the field k = Z/ pZ. A nonfunctorial partial generalization of (a), without the hypothesis that ST is a forest, is given in Theorem 5.4.2.
Theorem 5.2.8 Assume that T is a p-locallyfree finite lattice oftypes, ST is a forest, and j ::: 1. (a) There is afullfunctor F: CrePcrit(ST, Zp, j) ~ ~rep(ST, Z/pjZ) that is
a representation equivalence.
(b) There is a category equivalence H : Mep(ST, Z/pZ) ~ rep(Sr , Z/pZ). (c) The category of almost completely decomposable groups Ccrit(T, I)p has
finite representation type if and only if rep(ST, Z/ pZ) has finite representation type. (d) /frep(ST,Z/pZ) has wild representation type, then Ccrit(T, I) has wild modulo p representation type.
(a) For V = tU«, Vi : i E ST) E Crepcrit(ST, Zp, j), let hU = h {Vi: i E S} ~ Ue. Define F(V) = (Vo = hU/pjhU, Vi = (Vi + pjhU)/pjhU, V. = pjVO/pjhU : i E ST)and FU): F(V) ~ F(V), the morphism induced by amorphism f : V ~ V. Then F is a functor with F( V) E Mep(Sr, Z/ pj Z), noticing that f(hU) ~ hv . In particular, for each i, Vi is a pure submodule of Ui; Hence, Vi = (Vi + pj Vo)/ p! Vo is isomorphic to the free Z/ pjZ-module Vii p! Vi and
PROOF.
Vi
n V. = O.
To see that F is a full functor, let g: F(V) ~ F(V) be a representation morphism. By Lemma 5.2.7, Vi V;* ED V1, hU ED{V;*: i E ST}, and Vi = ED {V; : s ~ i} for each i E ST. It follows by induction that g lifts to an homomorphism f: hU ~ hV with f(Vi) ~ Vi for each i EST and f(pj Vo) ~ pjvo. Then f extends to h: Vo ~ Vo with F(h) = g, as desired. To confirm that F is a dense functorlet V = (Vo, Vi, V. : i E ST) E Mep (ST, Z/ p! Z) . Each Vi = vt ED Vi# , where Vi# is the injective envelope of h {VS : s < i} in Vi. This is so because Vi is a free Z/ pj Z-module and Z/ pj Z is self-injective. Then V/isafreeZ/pjZ-module,andjustasinLemma5.2.7, h{Vi :i EST} = ED{Vti E ST} and Vi = ED{Vs' :s ~ i} for each i EST. Choose a finitely generated free Zp-module Wo = ED{ V;* : i E ST} with Wo/pjwo = ED{Vt: i EST} , (V;* + pjwo)/pjwo = vt for each i EST, and a submodule K of Wo with K/ pj Wo = V•. Define Vi = ED{ Vs' : s ~ i) for each i in ST and Vo = Wo+(l/pj)K.Theneach Vi is pure in Vo,since KnVi ~ pjWo.Furthermore, (Vi + pjwo)/pjwo = Vi , pjvo/pjwo = K/pjWo = V., Wo = hU, and rank Vo = rank Wo = h{rank V;* : i E ST)' Then V = (Vo, Vi : i E ST) E Crepcrit(ST, Zp, j) with F(V) = (Vo, Vi, V* : i EST) = V, as desired. If F(V) is isomorphic to F(V), then, since F is a full functor, there are representation morphisms f: V ~ V and g: V ~ V with Iv - fg E pjEndV and I u - gf E pjEnd V. Since Vo is a free R-module and pZp = lZp, the Jacobson radical of Zp , it follows that pEnd V ~ lEnd V . Then gf = I - (I - gf) is a unit of End V, and similarly, f g = 1 - (1 - f g) is a unit of End V. Consequently, V is isomorphic to V.
=
=
5.2 Isomorphism at p and Representation Type
163
Finally, assume that F(V) = V' EEl V" is a representation direct sum. Since F is a dense functor, there are V' , V" E CrePcrit(Sr, 'lL p , j) with F(V') = V' and F(V") = V". Then F(V' EEl V") = F(V') EEl F(V") = F(V). But F reflects isomorphisms, so that V is isomorphic to V' EEl V" . If V is indecomposable, then V' or V" = O. Hence , V' or V" is 0, and so F(V) is indecomposable. This shows that F preserves indecomposables. (b) Let k = 'ILl p'IL, V = (Va, Vi, V*: i E Sr) E !:Hep(Sr , k), and define H: L\rep(Sr, k) --+ rep(Sr, k) by
H(V)
= «EEl{Vi: i E Sr})IV* , (Vi + V*)IV*: i E Sr).
Then H is a fully faithful functor . To see that H is dense, let V = (VA , Vi : i E Sr) E rep(Sr,k) with no trivial summands. Since Sr is a forest, Vi = V/ EEl Vi# and Vi = EEl{V/: j ~ i in Sr} for each i E Sr as a consequence of Lemma 5.2.7. Write Va = :E{Vi: i E Srl = (EEl{Vi : i E Sr})1 V* for some V*. Define Va = EEl{Vi:i E Sr}) and Vi = EEl{Vj:j ~ i}, where (V;* EEl V*)IV* = V/, for each i E Sr . Then V = (Va, Vi , V* : i E Sr) E !:Hep(Sr, k) with H(V) = V , as desired. (c) Observe that Ccrit(T, I)p has finite representation type ifand only ifrep(Sr, k) has finite representation type, by (a) and (b) and the fact that Ccrit(T, l)p is category equivalent to CrePcrit(Sr, 'lL p , 1). (d) Let I' be an 'lLp-algebra that is finitely generated and free as a 'lLp-module. Since rep(Sr, k) has wild representation type, rep(Sr, k) is endowild by Corollary 1.4.3. In view of the category equivalence (b), there is some V in !:Hep(Sr , 'ILl p'IL) with End V = I' I pi', Since F is dense and full, there is V E CrePcrit(Sr , 'lL p , 1) with F(V) = V and End V --+ End V = I' Ipf onto. This proves that CrePcrit(Sr, 'lL p , 1) has wild modulo p representation type. The next corollary completes the computation of the representation type of
C(T , l)p with w(Sr) = 3 for the case that Sr is a forest. Corollary 5.2.9 Suppose T is a p-locallyfree finite lattice oftypes and Sr is a forest with w(Sr) = 3. Then C(T , l)p has: (a) rank-infinite representation type if Sr contains (2, 2, 2), (1,3, 3), or (1,2 ,5)
as a subposet; and (b) wildmodulo p representation type ifSr contains (2, 2, 3), (1,3,4), or(l, 2, 6)
as a subposet. (a) By Theorem 5.2.8, Ccrit(T, l), has rank-infinite representation type if and only if rep(Sr, k) has infinite representation type. Now apply Theorem 1.3.6, bearing in mind that Sr is a forest with w(Sr) = 3, and that Ccrit(T, j)p is a full subcategory of C(T, j)p. Assertion (b) follows from Theorems 5.2.8 and 1.4.4.
PROOF.
164
5. Almost Completely Decomposable Groups
Open Questions 1. Suppose T is a finite p-locally free lattice oftypes with w(Sr) = 2 and Sr is not a garland. 1f6 :::: j :::: 4, does C(T , j)p have wild modulo p representation type? Ifl :s j :s 3, does C(T, j)p have rank-finite or rank-infinite representation type? In either case, can the indecomposable groups in C(T, j)p be classified? 2. What is the representation type of C(T, I)p if w(Sr) = 3 and Sr is not a forest ? In particular, what if Sr = (N , 4)? 3. Do the categories C(T, j)p and Ccrit(T , j)p have the same representation type? Ifw(Sr) = 3 and j = 2, does C(T, j)p have wild modulo p representation type? Does C(3 , 2)p have wild modulo p representation type? Can the indecomposable groups be classified ? 4. Is there a definition oftame representation type such that (i) if C(T, j)p has tame representation type, then indecomposables can be classified, and (ii) each C(T, j)p has exactly one offinite, tame, or wild modulo p representation type?
EXERCISES
1. Give a description ofrepresentations in Crep(Sr , Zp, j) . 2. Prove that each group in C(T, 0) is completely decomposable. 3. Give explicit constructions of indecomposable almo st completely decomposable group s corresponding to the indecomposable representations in Crep (Sr . 7lp • j ) given in Examples 5.2.2 and 5.2.3 .
5.3 Uniform Groups An almost completely decomposable group G is called uniform if there is a prime p such that G IG*(lT(G» is a free ZI pjZ-module, where IT(G) is the inner type of G. Define V(T, j) to be the subcategory of C(T, j) consisting of the uniform almost completely decomposable groups in C(T, j) and define Vcrit(T, j) to be the full subcategory of all uniform almost completely decomposable groups in Ccrit(T , j). Uniform almost completely decomposable groups arise naturally and are intimately related to more general finite rank Butler groups .
Example 5.3.1 Let T be a finite lattice oftypes. (a) Then V(T , I) = C(T, I) and Vcrit(T, I ) = Ccrit(T , I). (b) Let G be afinite rank Butler group in B(T) and write G = CI K for some completely decomposable group C E B(T). Then H = C + (11 p j)K is a uniform almost completely decomposable group in V(T , j).
5.3 Uniform Groups
165
(c) Conversely, suppose H = C + (11 pi)K is a uniform almost completely decomposable group in VeT, j) with C a completely decomposable group and K a pure subgroup of C. Then CI K is a finite rank Butler group in B(T). (a) Each 'I.I p'I.-module is free. (b) The group H is almost completely decomposable, pi H S; C, and HI C is isomorphic to pi HI pi C = (p i C + K)I pi C . But (p i C + K)I pi C is isomorphic to the free 'I.I pi 'I.-module K I pi K , since K is a pure subgroup of C. Furthermore, typeset H = typeset C S; T . (c) Since K is a pure subgroup of C , G = C I K is a finite rank Butler group. Moreover, G E B(T) by Lemma 3.3.1, since typeset C = typeset H S; T and T is a lattice.
PROOF.
The following three examples demonstrate that representation type changes from C(T, j)p to VeT, j)p; compare Corollary 5.2.4.
Example 5.3.2 The category Vcrit(T , j)p has finite representation type if ST = S3 and j :::: 2. In this case, each indecomposable in Vcrit(T, j)p has rank less than or equal to 9. [Arnold, Dugas 93A]. A complete and explicit list of indecomposables is given in [Mouser 93] .
PROOF.
Example 5.3.3 The category Vcrit(T, j)p hasfinite representation type if ST = (1, 2). Each indecomposable has rank less than or equal to 3. If G E Vcrit(T, j)p, then Fr(G) = (G p, G(T)p : TEST) = (Vo, VI, V2 ~ V3) = V is an indecomposable in Crep(ST, 'I.p, 2) with S = (1,2) and (End G)p
PROOF.
isomorphic to End V by Corollary 4.3.2. Then End V is isomorphic to End V for V = (VI EEl V3, VI, V2 S; V3, p2Vo) by Lemma 4.2.3(a). In particular, V is indecomposable. Assume that rank V > 1 and let A = (M I I M2 I M3) be a 'I.p-matrix with rows a basis of the free 'I.p-module p2Vo and the columns labeled by bases of VI, V2, and respectively. As demonstrated in Section 4.1, allowable operations that do not change the isomorphism class of V are invertible elementary 'I.p-column operations in M, for 1 :::: i :::: 3, invertible 'I.p-column operations from M2 to M3 (but not from M 3 to M2), and invertible 'I.p-row operations on M. Diagonalizing M I by invertible row and column operations yields A ~
V; ,
(LIMIN)= (I
o
I
o Mil pD M21
M12INII N 12) M22 N21 Nn
with p D a diagonal matrix having diagonal entries pi, i 2: 1. Since V is uniform,
166
5. AlmostCompletely Decomposable Groups
each linear combination of the rows of the matrix (pD I MZ1 Mzzi NZI Nn) must contain a unit in the (M zl M zz) block and a unit in the (N zI Nzz) block. Furthermore, (L I MIN) ~ (l 0 I Mz I M3). This can be seen by first diagonalizing (M Z1 Mzz) with elementary row and column operations. The result is a matrix (0 EpD I I 0 I NZI Nzz) with E a product of elementary row operations . Using column operations, this matrix may be assumed to be of the form (0 pD 1 I I 0 I NZI Nzz). Next, apply column operations from Mz to M3 to clear the M3 block. This yields the second row block of A equivalent to (0 pD I I 0 I 0), contradicting the assumption that V is uniform. In particular, the second-row block must be empty. A similar argument shows that A = (l 0 I Mzl M3) is equivalent to ( I
o
0II I 0
0 0 1 0) pD 0 I
with D a diagonal matrix. Since V is indecomposable with rank V > I, either rank V = 3 and, from the second-row block, V = (Zpx EB Zpy EB Zpz, Zpx, Zpy, Zpy EB Zpz, (I + pi + l)Zp), i < j, or else, from the first row block, rank V = 2 and V = (Rx EB Ry, Rx, Ry, Ry, (I + l)R) . Example 5.3.4 The category V(T, j)p has wild modulo p representation type if (a) W(ST) = 4, j ::: 2, or (b) W(ST) = 3, j ::: 3. (a) Suppose W(ST) = 4, j ::: 2. Let r be a Zp-algebra that is finitely generated and free as a Zp-module and let A, B be n x n Z p-matrices with r = C(A, B). Define V = (Vo, VI , V z, V 3, V 4) E rep(S4, Zp , 2) by
PROOF.
Vo = VI EB Ui EB V 3 EB V4
VI
= Z~ EB 0 EB 0 EB 0,
U3 = 0 EB 0 EB Z~ EB 0,
+ (II pZ)Vs, Ui = 0 EB z;n EB 0 EB 0, and
V4 = 0 EB 0 EB 0 EB
z;n,
where Us is the free R -rnodule with the rows of the following matrix as a basis:
I
o (o o
0 I 0 0
0 0 lOA 0 0 0 I 0 I 0 I 0 pI 0 I 0 I 0
0)
pB
I pI
'
for I an n x n identity matrix and 0 an n x n matrix ofo's. Notice that V = FdG) for some G E V(T', 2) ~ V(T, j) with ST' = S4, since V E CrePcrit(S4, Zp , j) with Vo/(VI + Ui + V3 + V4) a free ZI pZZ-module. Thus, (End G)p = End V. Now, End V is isomorphic to End V, where V = (VI EB U: EB V3 EB V 4, VI, Us. V 3, V 4, pZVo). A by now familiar computation yields an epimorphism End V ---+ C(A(modp), B(modp» and C(A, B) embeds in End V. It follows from
5.3 Uniform Groups
167
Lemma 4.2.1 that VeT , j)p has wild modulo p representation type in the case WeST) 2: 4, j 2: 2. (b) The case that w(S) = 3, j 2: 3 follows from Corollary 4.3 .2 and Example 5.2.2, since in that example VO/(UI + V z + V 3 ), being isomorphic to V*/ p 3V*, is a free Z/p3Z- module. Example 5.3.5 The category VeT , j)p has: (a) rank-infinite representation type if W(ST) = 2, j 2: 6, and ST contains either (1,3) or (2,2) as a subposet, and (b) wild modulo p representation type if W(ST) = 2, j 2: 11, and ST contains either (1,3) or (2,2) as a subposet. PROOF.
The associated representations are Exercises 5.3.3 and 5.3.4.
Corollary 5.3.6 Let T be a finite p-localiy free lattice oftypes and j a positive integer. Then VeT, j)p has: (a) wild modulo p representation type if either (i) W(ST) = 5, j 2: 1; (ii) W(ST) = 4, j 2: 2; (iii) W(ST) = 3, j 2: 3; or (iv) W(ST) = 2, j 2: 11, and ST contains (1,3) or (2, 2) as a subposet. (b) rank-infinite representation type if either (i) weST> = 4, j 2: 1; or (ii) W(ST) = 2, j 2: 6, and ST contains (1,3) or (2,2) as a subposet. (c) finite representation type if either W(ST) = 1; or W(ST) = 2, j 2: 1, and ST is a garland.
Moreover, Ucrit(T, j) has finite representation type if W(ST) = 2 and ST is a forest that does not contain (1, 3) or (2, 2) as a subposet. PROOF. Statements (a) and (b) follow from Examples 5.3.4 and 5.3.5 and Corollary 5.2.4, noticing that U(T , 1) = C(T, 1) by Example 5.3.1(a). As for (c), apply Corollary 4.3.3( c). For the last statement, if ST is a forest of width 2 that does not contain (2, 2) or 0 , 3), then ST = (1,2). Now apply Example 5.3.3.
Descriptions of representation type for V (T, R, j) p remain incomplete for those T with W(ST) = 2 or W(ST) = 3 and j :::: 2. This deficiency already arose in Section 5.2 for C(T, l)p with W(ST) = 3 and ST not a garland. Define V (n, j) to be the full subcategory of uniform almost completely decomposable groups in C(n, j) and Utn, j)p the isomorphism at p category of Utn , j). Corollary 5.3.7 The category V(n, j)pfor j 2: 1 has: (a) wild modulo p representation type j ?:. 3.
if either n 2: 5; n
= 4, j
2: 2; or n
= 3,
168
5. Almost Completely Decomposable Groups
(b) rank-infinite representation type ifn = 4, j = 1; and (c) finite representation type ifn ~ 2 or n = 3, j ~ 2; Corollary 5.3 .6 and Example 5.3 .2.
PROOF.
Open Questions 1. What is the representation type for the missing cases of Corollary 5.3.6 ? 2. Are there representation embeddings or equivalences from representations over fields that can explain the representation type of U(T , j)p? 3. Give a theoretical explanation of why Ccrit(S3, 2)p has rank-infinite representation type but Ucrit(S3, 2)p has finite representation type.
EXERCISES
1. Prove that there are indecomposable groups G in B(T3) of arbitrarily large finite rank with G = G*(lT(G)). 2. Prove that if ST = (1, 2) , then there is a bound on the ranks of indecomposable groups in B(T) with G = G*(IT(G)).
3. Assume that ST = (l, 3).
(a) Use the following representation to show that U(T, j)p, j ~ 11, has wild modulo p representation type: U = (Uo, UI, U2 ~ U3 ~ U4) with Uo = z~n 61 z~n 61 z~n 61 z~n + (l/pll)U* , UI = z~n 61 0 61 0 61 0, U2 = 0 61 z~n 61 0 61 061, U3 = 0 61 z~n 61 z~n 61 0, U4 = 0 61 z~n 61 z~n 61 z~n, and U* the row space of the matrix contained in UI 61 U2 61 Uj 61 U: : I
(
o o o o
0 I 0 0 0
0 0 0 0 0 0 I 0 0 0 I 0 0 0 I
p5 I p61 P7 I p8 I
0
p 61
I 0 P7 A 0 p 2 1 p8BOO
0
0 plO I 0
0 0
0 0
p 41
0 0
0 0 0
p 61
0
~ 0~ ~ 0~ 0~)
o o
I
0 0 I 0 p81 0 0 0 I
.
(b) Use the following representation to show that U(T, j)p , j ~ 6, has infinite representation type: U = (Uo, UI, U2 ~ U3 ~ U4) with Uo = z~n 61 z~n 61 z~n 61 z~n + (l/p6)U* , UI = z~n 610610610, U2 = 061Z;: 61061061, U3 = 061z~n 61 z~n 61061, U4 = 061z~n 61z~n 61Z;:, and U* the row space of the matrixcontained in UI 61 U2 61 U; 61 Ur I
(
o o
0 0 p 31 p4A I 0 I 0 p 41 p5 I 0 p2[ 0 [ p5 I 0 0 0
0 0
0 0) [ 0 p4 I 0 [ .
4. Assumethat ST = (2, 2). (a) Use the following representation to show that U(T, j)p , j ~ 10, has wild modulo p representation type: U = (Uo, UI ~ U2, U3 ~ U4) ~ with Uo = z~n 61 z~n 61 Z;: 61 z~n + (l / p IO)U*, U I = z~n 61 0 61 0 61 0, U2 = z~n 61 z~n 61 0 61 0,
5.4 Primary Regulating QuotientGroups
169
U3 = 0 $ 0 $ z~n $ 0, U4 = 0 $ 0 $ z~n $ z~n , and U* the row space of the matrixcontainedin U, $ Ui $ U3 $ U; :
o o
I 0 o 0 o 0 p 5I 0
0 I 0 0 0
0 0 I 0 0
0 0 0 I 0
0 0 0 0 I
4I p p5I p6I p7I 0
p5I I 0 p6A 0 I p 7 BOO 0 0 0 p9I 0 0
0 0 I 0 0
0 0 0 I 0
0) 0 0 . 0 I
(b) Use the following representation to show that U(T, })p, } :::: 6, has infiniterepresentation type: U = (Uo , U\ ~ U2 , U3 ~ U4) with Uo = z~n $ z~n $ z~n $
z~n +(I/p6)U* , U\ = z~n $0$0$0, U2 = z~n $z~n $0$0$0, U3 = 0$0$ z~n $ 0, U4 = 0 $ 0 $ z~n $ z~n , and U* the row space of the matrixcontained
in U, $ Ui $ U3 $ U; : I 0 0 p2I 0 I 0 p3I p 5I 0 0 I p4I
o o
p3A I 0 0 ) p4I 0 I 0 . 0 0 0 I
5.4 Primary Regulating Quotient Groups An almost completely decomposable group G is a p-primary regulating quotient group if p is a prime and G has a regulating subgroup B such that G j B is a pgroup, equivalently, i(G) is a power of p by Proposition 5.1.2. It follows that an almost completely decomposable group is a p-primary regulating quotient group if and only if eo = exp G j R(G) is a power of p. Notice that the class of pprimary regulating quotient groups is closed under finite direct sums , since if B is a regulating subgroup of Hand C is a regulating subgroup of K with IH j B I and IK j C I both powers of p , then D = B EB C is a regulating subgroup of G = H EB K with IGjDI a power of p . The p-primary regulating quotient groups are generalizations of groups in qT, j)p in the following sense . If Tcr(G) is an inverted forest and G is an almost completely decomposable group with no rank-I summands, then R( G) = G*(IT(G» is a unique regulating subgroup of G by Corollary 3.2.13(c). If, in addition, typeset G is contained in a finite lattice T of types, then G E qT, j)p if and only if G is a p-primary regulating quotient group . For p-primary regulating quotient groups, near-isomorphism is the same as isomorphism at p:
Lemma 5.4.1 Let G and H be p-primary regulating quotient groups. Then G and H are nearly isomorphic if and only if G and H are isomorphic at p. Moreover, G is indecomposable if and only if G is indecomposable at p. Since near-isomorphism is isomorphism at p for each prime p, it suffices to prove that if G and H are isomorphic at p , then they are nearly isomorphic. If G
PROOF.
170
5. Almost Completely Decomposable Groups
and H are isomorphic at p, then there is a monomorphism f : G ~ H such that HI f(G) is finite and relatively prime to p. Consequently, R(G) is isomorphic to R(H) as in Theorem 5.1.6(b). Since eG is a power of p , G is isomorphic to R(G) at q for each prime q -:;6 p . Similarly, H is isomorphic to R(H) at q for each prime q -:;6 p . Thus, G is isomorphic to H at q for each prime q , and so G is nearly isomorphic to H . The last statement follows from Corollary 5.1.8(b). The next theorem is a generalization of Theorem 5.2.8(a), wherein ST is assumed to be a forest. However, in contrast to that theorem , this construction need not be functorial , since there are almost completely decomposable groups G and H and a homomorphism f : G ~ H such that f(R(G)) is not contained in R(H); Exercise 5.1.8 . Theorem 5.4.2 [Mader, Vinsonhaler 97] Suppose G is a p-primary regulating quotient group with eG = pl . Define a 7l,lpj7l,-representation F(G) = (Uo, U T , U* : r E Tcr(G))by Uo = R(G)I p j R(G), U T = (R(G)(r)) + p! R(G))I p j R(G), and U* = pl Gf p! R(G). (a) There is a ring isomorphism ¢ : End G I pjEnd R(G) ~ End F(G). If f is an idempotent ofEnd F( G), then there is an idempotent g ofEnd G with ¢(g) = f . (c) The group G is indecomposable ifand only if(End G)p is a local ring.
(b)
PROOF. (a) Let ¢(/)EEndF(G) be the endomorphism induced by f, which is well-defined, since R(G) is a fully invariant subgroup of G by Theorem 5.1.6(a). Then ¢(/) = 0 if and only if f : R( G) ~ p j R( G). Equivalently, f E pj EndR(G) ~ EndG , since pjG ~ R(G). The proof that ¢ is onto is a direct analogue of the proof of Theorem 5.2.8(a). The crucial observations are that R(G) is a completely decomposable subgroup of G with p! G ~ R(G), and R(G) is a fully invariant subgroup of G . (b) Suppose f is an idempotent of End F(G) . By (a), there is h E End G with ¢(/) = hand h 2 - h E p jEnd R(G). Now End R(G)INEnd R(G) is a product of matrix rings over subrings of Q by Proposition 3.1.8. In fact, h + pjEnd R(G) lifts to an idempotentg of End R(G). This is so because idempotents of AIp! A lift to idempotents of A for A a matrix ring over a subring of Q, hence for A = End R(G)INEndR(G) . But idempotents of End R(G)I NEnd R(G) lift to idempotents of End R(G), since NEnd R(G) is nilpotent. Since R(G) is a fully invariant subgroup of G with G I R(G) finite, there is an idempotent g E End G with ¢(g) = f. (c) If (End G)p is a local ring, then 0 and 1 are the only idempotents of (End G)p. Hence, G is indecomposable at p and so must also be indecomposable as a group. Conversely, suppose that G is indecomposable. Then End G has only 0 and 1 as idempotents, and so End F(G) has only 0 and 1 as idempotents by (b). Since End F(G) is finite, it must be a local ring. By (a), (EndG)plp j(End R(G))p is a local ring , whence (End G) p is also a local ring , since p E J (End G) p -
5.4 PrimaryRegulating Quotient Groups
171
The near-isomorphism category of torsion-free abelian groups of finite rank has nonunique direct sum decompositions into indecomposables by Example 2.1.11. The near-isomorphism category of almost completely decomposable groups also has nonunique direct sum decompositions by Proposition 5.1.11. However, as a consequence of the next theorem, the near-isomorphism category of p -locally free almost completely decomposable groups G with ea a power of p is a KrullSchmidt category.
Corollary 5.4.3 [Faticoni, Schultz 96] Assume that G and Hare p-primary regulating quotient groups. If G = Xl ED . . . ED X n is nearly isomorphic to Y1 ED · ·· ED Yn with each Xi and Yi an indecomposable group, then m = nand there is a permutation C1 of {I, 2, . . . , n} with Xi nearly isomorph ic to YU(i) for each I ~ i ~ n. By Corollary 5.1.8(d), it suffices to assume that G = H. It is also sufficient to assume that both G and H are p-reduced, hence p -locally free; Exercise 5.1.5. Now apply Theorem 5.4.2(c) and Exercise 1.2.6 to see that there is a permutation C1 of {l , 2, . .. ,n} with Xi isomorphic at p to Yu (i) for each 1 ~ i ~ n. Since eo and en are powers of p, it follows from Lemma 5.4 .1 that Xi and Yu(i) are nearly isomorphic. PROOF.
There are examples of almost completely decomposable groups G and H with no rank-I summands such that G ED H has a rank-I summand; Exercise 5.1.7(b) . This cannot happen if G and H are p -primary regulating quotient groups:
Corollary 5.4.4 [Lady 74B] Suppose G and Hare p-primary regulating quotient groups. (a) IfG and H have no rank-Y summands, then GEDH has no rank-s summands. (b) Any two direct sum decompositions ofG into indecomposable groups have the same number of rank-t summands oftype r for each r.
This corollary is a special case of Corollary 5.4.3 and the observation that nearly isomorphic completely decomposable groups are isomorphic.
PROOF.
EXERCISE
I. Assume that G = A E9 H with A a completely decomposable group. Prove that if B is a regulating subgroupof G, then B E9 H is a regulating subgroup of Hand G/ B is isomorphic to H / H n B. Conclude that if G is a p-primaryregulating quotient group, then so is H . What can be said if A is not completely decomposable? NOTES ON CHAPTER
5
Finite rank completely decomposable groups are classified in [Baer 37]. Examples of pathological direct sum decompositions of almost completely decomposable groups were
172
5. Almost Completely Decomposable Groups
constructed in [Levi 17] and [Jonsson 57, 59]. For many years , almost completely decomposable groups were considered primarily as a good source of counterexamples; see [Fuchs 73]. Some positive results for almost completely decomposable groups , most of which are given in Section 5.1, first appeared in [Lady 74B]. An equivalence condition on almost completely decomposable groups G and H, G ffi C ::::::: H ffi C for some completely decomposable group C , led to the notion of near-isomorphism for torsion-free abelian groups of finite rank. Most of Section 5.2 is derived from [Arnold, Dugas 98]. As demonstrated in that section, representations of finite posets can be used to easily construct examples of indecomposable almost completely decomposable groups of arbitrarily large finite rank with types in a fixed typeset. The fact that categories of almost completely decomposable groups with typesets contained in a relatively small lattice of types can have wild modulo p representation type vividly demonstrates the complexity of almost completely decomposable groups. Almost completely decomposable groups are viewed as extensions of finite torsion groups by completely decomposable groups in [Mader 95]. Near-isomorphism is rediscovered in the sense that an equivalence, called type isomorphism, which arose naturally in this context , is the same as near-isomorphism. Mader 's approach, together with the results and techniques of [Burkhardt 84], [Krapf, Mutzbauer 84], and [Schultz 87], has rekindled interest in the subject of almost completely decomposable groups ; see [Mutzbauer 99], [Mader, Nonxga 99], [Ould-Beddi, Striingmann 99], and [Reid 99]. [Mader 00] includes a comprehensive, detailed exposition of the subject of almost completely decomposable groups and an extensive bibliography. Several classes of almost completely decomposable groups have been classified by invariants up to near isomorph ism: uniform block rigid groups [Dugas, Oxford 93]; groups divisible by all but three primes [Cruddis 70]; groups with critical typeset of two elements; and cyclic regulating quotient groups [Mader, Mutzbauer 93], [Mader, Vinsonhaler 95], [Burkhardt, Mutzbauer 96], and [Dittmann, Mader, Mutzbauer 97]. Examples given in these papers demonstrate that classification of almost completely decomposable groups up to isomorphism involves algebraic number-theoretic difficulties that can be avoided by classification up to near isomorphism. There is a substantial amount of literature on almost completely decomposable groups with small typeset; see [Arnold , Dugas 99]. Results of [Cruddis 70] are extended in [Arnold 73]. In the latter paper, some of the arguments are invalid. Corrected proofs and related results can be found in [Arnold, Dugas 95A], [Lewis 93], and [Mader, Vinsonhaler 94]. Finite rank Butler groups G such that each pure subgroup of G or each torsion-free quotient of G is almost completely decomposable are characterized in terms of the critical typeset of G in [Nonxga, Vinsonhaler 95]. Invariants for almost completely decomposable groups are discussed in [Arnold , Vinsonhaler 93] and [Vinsonhaler 95].
6 Representations over Fields and Exact Sequences
6.1 Projectives, Injectives, and Exact Sequences The terms exact sequence, kernel, and cokernel are defined for arbitrary categories [Mitchell 65]. In order to avoid lengthy arguments, these terms are defined herein in terms of their descriptions in rep(S, k) rather than derived. Let V = (Vo, Vi : i E S), V = (VO, Vi : i E S), and W = (Wo, Wi : i E S) be objects ofrep(S, k), I E Hom(V , V), and g E Hom(V, W). The sequence f
g
O-+V-+V-+W-+O
is an exact sequence if the induced sequence
o-+ Vi -+f Vi -+g Wi -+ 0 of vector spaces is exact for each i in S U {O}, i.e., for each i , I: U, -+ Vi is one-to-one, g : Vi -+ Wi is onto, and ker g = image I . A representation morphism I : V -+ V is a monomorphism if I :VO -+ Vo is l -to-l and an epimorphism if the k-linear transformation I : Vi -+ Wi is onto for each i E S U {O} . The kernel of I is Ker 1= (ker I, (ker f)
n Vi : i E S) E rep(S, k),
where ker I is the kernel of the k-linear transformation I: VO -+ Vo. If I is an epimorphism, then 0 -+ Ker I -+ V -+ V -+ 0 is an exact sequence of representations, since 0 -+ (ker f) n Vi -+ Vi -+ Vi -+ 0 is an exact sequence of vector spaces for each i E S U {O} .
174
6. Representations over Fieldsand ExactSequences
A representation morphism I: V -+ V is apure morphism and I(V) = (/(Vo), I(V;): i E S) is a pure subrepresentation of V if I(V;) = I(Vo) n V; for each i E S. For example, the kernel of a representation morphism is a pure subrepresentation. The terminology arises from the abelian group setting . A pure morphism is called a proper morphism in [Simson 92]. If I : V -+ V is a pure morphism, then the cokernel 01 I is Coker 1= (Voll(Vo), (V; + I(Vo»II(Vo): i E S) E rep(S, k) . The sequence 0 -+ V -+ V -+ Coker 1-+ 0 in rep(S , k) is exact if I is a pure monomorphism, since 0 -+ I(Vo)n V; -+ V; -+ (V; + I(Vo))//(Vo) -+ Ois an exact sequence of vector spaces for each i, A representation V E rep(S, k) is projective if whenever I : V -+ W is a representation epimorphism and g : V -+ W is a representation morphism with V, W E rep(S, k), then there is a morphism h: V -+ V with Ih = g. Dually, a representation V is injective if whenever I : V -+ W is a pure monomorphism and g : V -+ V with V, WE rep(S, k), then there is a morphism h: W -+ V with hi = g. Recall that V = (Vo, V; : i E S) E rep(S, k) is a trivial representation if Vo =P 0 but V; = 0 for each i in S. Up to isomorphism, the only trivial indecomposable representation is P(O) = (k, P(O); : i E S) with P(O); = 0 for each i in S. Call V E rep(S, k) cotrivial if V; = Vo for each i. The only cotrival indecomposable representation, up to isomorphism, is 1(0) = (k , 1(0); : i E S) with 1(0); = k for each i in S. Lemma 6.1.1 Let S be a finite poset, k afield, and a: rep(S , k) -+ rep(SOP, k) the contravariant duality given in Proposition 1.4.7. Then: (a) a sequence 0 -+ V -+ V -+ W -+ 0 is exact in rep(S, k) if and only if 0 -+ a(W) -+ a(V) -+ a(V) -+ 0 is exact in rep(SOP, k); (b) a representation V is projective in rep(S, k) if and only if a(V) is injective in rep(SOP, k); and (c) V is a trivial representation if and only if a(V) is a eotrivial representation. PROOF.
Exercise 6.1.1.
Let S be a finite poset and k a field. For t E S, define pet) = (k, pet); : i E S) E rep(S, k),
= k for each i>: tin Sand P(t)j = 0 if j t. tin S. Dually, define l(t) = (k , I(t);: i E S), where I(t)j = k for each j 1: tin Sand I(t)j = 0 if j ~ tin S. Observe that if pet) E rep(S, k), then a(P(t» = let) E rep(SOP, k) . For example, let S = {I,2 < 3}. Then P(O) = (k, 0, 0, 0), P(l) = (k , k, 0, 0), P(2) = (k, 0, k, k), and P(3) = (k, 0, 0, k). Hence, sop = {I,3 < 2} and a(P(O» = where P(t)j
6.1 Projectives, Injectives, and Exact Sequences
=
=
=
(k, k , k , k) 1(0), a(P(I» (k , 0, k , k) 1(1), a (P(2» and a(P(3» = (k , k , 0, k) = 1(3) E rep(SOP, k).
Lemma 6.1.2 Let S be a finite poset, k a field, and V rep (S , k ). Then:
175
= (k, k , 0, 0) = 1(2), = (Vo, Vi : i E S) E
(a) V is projective in rep(S , k ) if and only if V is isomorph ic to a finite direct sum of representations P (O) and P(t)for tin S; and (b) V is injective in rep (S, k) if and only if V is isomorphic to a finite direct sum of representations I (0) and I (t) for t in S. PROOF. The first step is to prove that P(t) is projective for each t E S U {O} . To this end, assume that f : V --+ W is an epimorphism and g : P(t) --+ W . Since P(t)o = k and f is onto, there is a k-linear transformation h : P(t)o --+ Vo with f h = g. Then h is a representation morphism, recalling that if t E S, then P(t)j = k if j ~ t and P(t)j = 0 otherwise. This shows that each P(t) is projective. It follows that finite direct sums of P(O) and P(t) are projective. Conversely, let V = (Vo , Vi : i E S) E rep(S, k) be a projective representation. There is an epimorphism f : P --+ V , where P is a finite direct sum of representations P(t ), t in S U {O}. This holds because for a given t , there is a finite direct sum V (t ) = (Vo, Vi : I E S) of copies of pet) and a representation morphism V (t ) --+ V such that V, is contained in the image of V,. Taking the direct sum of all such V (t)'s for t E S U {O} gives the desired P. Since 1u : V --+ V and V is projective, there is a morph ism h : V --+ P with f h = 1u- In particular, V is isomorphic to a representation summand of P . Since P is a finite direct sum of P(t)'s for t E S U {O} and each P(t ) is indecomposable, V is isomorphic to a finite direct sum of cop ies of P (O ) and P (t ) for t E S by Theorem 1.2.2 . (b) Apply (a), Lemma 6.1.1 (b), and the observation that a (P (t» = I (t) .
Example 6.1.3 Suppose S is a chain with n elements and k is a field. If V
E
rep(S , k), then V is both projective and injective.
PROOF . The representation V is a finite direct sum of indecomposable representations, and each indecomposable has dimension I by Lemma 1.3.3(a). If V E rep(S, k) has dimension 1, then V = P(s) = I(t) for some t < s E S or V = P(O) = I (n) . By Lemma 6.1.2, V is both projective and injective, as desired.
An epimorphism f: P --+ U in rep (S, k) , with P a projective representation, is a projecti ve cover of V if given g : P' --+ V an epimorphism with P' projective, then there is an epimorphism h : P' --+ P with f h = g. In this case , P is isomorphic to a summ and of P' , since P is projective. Notice that a representation U is projective if and only if the identity homomorphism U --+ V is a projective cover of V. A pure monomorphism f: V --+ I in rep(S, k) with I an injective representation is an injective envelope of V if for each pure monomorphism g : V --+ I' with I '
176
6. Representations over Fieldsand Exact Sequences
injective, there is a pure monomorphism h : I -+ I' with hf = g. In particular, I is a summand of I' . Moreover, V is injective if and only if the identity is an injective envelope of V. Projective covers and injective envelopes are unique up to isomorphism (Exercise 6.1.10) . Projective covers and injective envelopes of V E rep(S, k) can be constructed explicitly. Assume that V E rep(k, S) has no trivial summands. For each i in S, let Vr = 'E{V j : j < i in S}. Define P(V)
= (P(V)o , P(V)i) with P(V)o = E9{ Vi/Vr : i E sj and P(V)i =E9{Vj/Vrj~i}
for each i in S. If V is trivial, define P(V) = V. Then P(V) is a finite direct sum of copies of P(t), t E S U {O}, whence, by Lemma 6.1.2(a), P(V) is a projective representation in rep(S , k). If V E rep(S, k) has no cotrivial summands, let Vi = n{ U, : j > i in S} agreeing that Vi = Vo if there is no j in S with j > i , Define I(V)
= (/(V)o, I(V)i) with I(V)o = E9{V;j Vi : i E S} and I(V)i = E9{Vj/ u.. i i j}.
If V is cotrivial , define I(V) = V . By Lemma 6.1.2(b), I(V) is an injective representation in rep(S, k) for each V . Lemma 6.1.4 Let V
E
rep(S, k)for afinite poset S andfield k.
(a) There is a projective cover pu : P(V) -+ V and (b) an injective envelope qu : V -+ I(V).
PROOF. (a) It suffices to assume that V has no trivial summands. For each i, there is a k-linear transformation hi : Vi -+ Ui] Vr given by h(x) = x + Vr Since ker hi = Vr is a vector space summand of Vi, there is a k-linear transformation gi : Ui] Vr -+ Vi S; Vo with h ig, = 1. Define a k-linear transformation p : P(V)o = E9{ Ui] Vr: i E S} -+ Vo by p = 'E{gi: i E S}. Then p is a representation morphism, since P(P(V)i) = p(E9{Vj/ Vr j s i}) = 'L, {g/V j/ VJ): j ~ i} S; 'L, {Vi : j s i} S; Vi for each i E S. An induction argument, beginning with a minimal element of S, shows that P(P(V)i) = Vi for each i in S. In particular, if i is a minimal element of S, then Vr = 0, P(V)i = Vi, and P(P(V)i) = Vi . Now assume that i is not a minimal elementofS and p(P(V)j) = Vjforeachj
E
rep(S , k) is both projective and injective, then w(S)
=
= I.
6. Confirm the assertions of Lemma 6.1.4(b) and Example 6.1.5.
°
7. The Cartan matrix of a finite poset S is an lSI x lSI matrix C(S) = (cij), where Cij = 1 if i ~ j in Sand Cij = otherwise. Show that the ith column of C(S) is the transpose of the vector cdn P(i) and that the ith row of C(S) is cdn I(i). 8. Define the extended Cartan matrix to be an (lSI + 1) x (lSI + 1) matrix C*(S) consisting of C(S) together with last row (0, 0, . . . , 0, 1) and last column the transpose of (-1, -1, . . . , -1 ,1 ). Label the elements of S by {I, 2, . . . , n}. Given variables XI , , Xn+l , define the quadratic form as of S by qS(Xl , ... , Xn+l ) = (Xl , , Xn+1)C*(S )(XI, , xn+d ' . (a) Show that qS(XI, ,X n+() = L{x1:l~i~n+l}-L{xiXj:i<jinS}+ Xn+l (L{Xi : 1 s i s n}). (b) [Ringel 84] In each example below, confirm that qs (z ) = 0, where z = (zj , .. . , zn+() and S = {I, 2, . . . , n}.
178
6. Representations over Fields and Exact Sequences (i)z=(1 ,1 ,1 ,1,2),
(ii)
z = (1, 1, 1, 1, 1, 1, 3),
S=l
2
3
S= 2
4
6
I
(iii)
z = (2, 1, 1, 1, 1, 1, 1, 4),
I
3
5
S= 1
4
7
3
I
(iv) z = (3,2,2, 1, 1, 1, 1, 1, 6),
I
I
I
S= 1
I
6
I
2
5
3
8
I
2
4
I
7
I
6
I
5
I
4 (v) Z
= (1,2,2,1,1 ,1 ,1,1,5),
S=2
4
I\ I
1
3
8
I
7
I
6
I
5 (c) Provethat if qs(z) > 0 for each 0 =/: z = (ZI, .. . , Zn) with Zi a nonnegative integer for each i, thenS has finiterepresentation type(theconverseis alsotrue; [Drozd74] or [Simson92]).
9. Provethat U = (Uo, U, : i E S) E rep(S, k) is a sincere representation if and only if the projectivecover P(U) of U contains a copy of P(i) for each i E S. 10. Prove that projective covers and injective envelopes in rep(S, k) are unique up to isomorphism.
6.2 Coxeter Correspondences The construction of matrices to find indecomposable representations in rep(S , k) , as described in Section 1.2, is impractical for a general finite poset. In this section, correspondences are defined that can be used to generate indecomposables from a given indecomposable. In fact, ifrep(S, k) has finite representation type, then all indecomposables in rep(S, k) can be found by applying these correspondences to l-dimensional projective or injective representations (Theorem 6.2.6). Let V = (Vo, Vi : i E S) E rep(S, k) and PU : P(V) ~ V be a projective cover of Vas defined in Lemma 6.1.4(a) with P(V)o = tI7{V;/ Vr: i E S} and P(V)i =
6.2 Coxeter Correspondences
${ Vjl V J: j ~ i}. Define representations W (V ) = (Wo, Wi : i Wo = P(V)o
E
179
S) , with
= e] Vi/V!: i E S} , Wi = ${Vi/V;: i i
j}
and L (V ) = (L (V )o, L(V)i : i E S), with L(V)o = Vo, L(V)i
= 'E{Vj:i ij} .
Then
c+(V) = (ker pi), (ker Pu) n Wi: i
E
S)
is a kernel of the representation morphism pu : W (V) --+ L( V). Hence, 0--+ C+(V) --+ W(V) --+ L(V) --+ 0 is an exact sequence of representations with W(V) an injective representation by Lemma 6.1.2(b) . Let qu : V --+ I (V) be an injective envelope of V as defined in Lemma 6.1.4(b). Define representations
with
and K(U ) = (K (U )o, K (U )i ), with K(U)o = qu(Uo), K (U) i = qu(Uo) n W'(U)i. Then
is a cokernel of the pure monomorphism K(V) --+ W'(V). Hence, 0--+ K(V) --+ W'(V) --+ C- (V) --+ 0 is an exact sequence of representations with W' (V) a projective representation. Proposition 6.2.1 Let S be a finite poset and k a field. Then C+ and C- are correspondences from rep( S, k) to rep( S, k). If U E rep( S, k), then C+ V = 0 if and only if V is a projective representation, and C- V = 0 if and only if V is an injective representation. Moreov er, C+ and C- preserve finite direct sums.
180
6. Representations over Fields and Exact Sequences
The representation U is projective if and only if P(U) = U, equivalently, the kernel of pi): P( U) -+ U is O. Thus , C+ U = 0 if and only if U is projective. A similar argument shows that U is injective if and only if C- U = O. Both P(U)$P(V) and PCU$V) are projective covers of U$V . Hence P(U)$P(V) is isomorphic to P(U $ V) by Exercise 6.1.10 . Now, C+(U $ V) is constructed in terms of P (U $ V) , while C+ U $ C+ V is constructed from P(U) $ P(V). It follows that C+(U $ V) is isomorphic to C+ U $ C+ V. A similar argument, using injective envelopes, shows that C-CU $ V) is isomorphic to PROOF.
C-U $ C-V.
Example 6.1.5 (continued) (c) If U has no trivial summands, then
W(U)
= (U I $
.. . $ Un, Uz $ . . . $ Un '
UI $ U3 $ L(U) = (Uo , Uz +
$ Un,·· ·, UI $
+ Un "
' " UI +
$ Un-I),
+ Un -I>,
and C+U
= (ker pi) , (ker Pu) n CUz $
.. . $ Un) , .. . ,
(ker Pu) n CUI $ .. . $ Un-I» ·
(d)
If U has no cotrivial summands, then W'(V) = (Vo/V I E9 • . . E9 VO/V n, Vo/V 1 ,
••• ,
Vo/Vn),
K(U) = (qu(Uo), qu(Uo) n (Uo/ VI> , . . . , qu(Uo) n (Vo/ Un», and
C- U
= «Uo/ UI> $
. . . $ (Uo/ Un»/qu(Uo),
«Uo/Ui)+qu(Uo»/qu(Uo) : 1 ~i
~n).
=
(e) Suppose cdn U (uo, UI,"" un)' (i) If U is not a projective representation, then cdn C+U
= (UI + UI
(ii)
+
+ Un + Un -I
Uo , Uz +
... + Un -
Uo, . .. ,
- uo)·
If U is not an injective representation, then cdnC-U = «n -l)uo - UI - . . . - Un, Uo - UI ,.· · , Uo - Un)'
cv
For a positive integer i, define to be the composition of C+ repeated i times and define to be the composition of C- repeated i times. Define C+OU = C-oU = U. Call U in rep(S, k) preprojective if U is isomorphic to P for
c:'
c:'
6.2 CoxeterCorrespondences
181
some projective P and i ::: 0, and preinjective if U is isomorphic to C+ i I for some injective I and i ::: 0. Each projective is preprojective and each injective is preinjective. Example 6.2.2 Following is a listingofthe indecomposable preprojectives and preinjectivesfor rep(S3, k): (a) indecomposable preprojectives: P(O), P(1), P(2), P(3) ,
°
C- P(O) = (k EfJ k, k EfJ 0, EfJ k, (1 + l)k),
= 1(0), C- P(1) = 1(1), C- P(2) = 1(2), C- P(3) = 1(3).
C- 2 P(O)
(b) indecomposable preinjectives: 1(0),/(1),/(2),/(3)
C+ 1(0) = C- P(O), C+2 / (O) = P(O), C+ 1(1) = P(1) , C+ 1(2) = P(2), C+ 1(3) = P(3). Recall that P(O) = (k , 0, 0, 0), P(1) = (k, k, 0, 0), P(2) = (k, 0, k, 0), and P(3) = (k , 0, 0 , k) are the projectives for rep(S3, k) . Then cdn P(O) = (1,0,0,0), cdn P(1) = (I, 1,0,0), cdn P(2) = (1,0,1,0), and cdn P(3) = (1,0,0, I) . The injectives for rep(S3, k) are 1(0) = (k, k, k, k) , 1(1) = (k, 0, k, k) , 1(2) (k , k, 0 , k), 1(3) = (k , k, k , 0) with cdn 1(0) = (1,1 ,1,1), cdn 1(1) = (1,0,1 , I), cdn 1(2) = (1, 1,0, I), and cdn 1(3) = (I, I, 1,0). (a) By Example 6.1.5(e), cdn C- P(o) = (2, 1, I , I), cdn C- 2 P(o) = (l, I, I, I) cdnC-P(1)=(l,O, I, I), cdnC-P(2)=(1, 1,0, 1), and cdnC-P(3)=(l,I,
PROOF.
=
1, 0). Since l-dimensional representations are determined up to isomorphism by their coordinate vector, it remains only to determine C- P(O). By Example 6.1.5(d), W'(P(O» = (k EfJ k EfJ k, k EfJ EfJ 0 , EfJ k EfJ 0, EfJ EfJ k), K(P(O» = «1 + 1 + l)k , 0, 0, 0), and C- P(o) = (Vo, VI> V2 , V3) with
° °
°°
+ 1 + I)k, VI = «k EfJ EfJ 0) + (1 + I + I)k)/(1 + 1 + I)k, V2 = «0 EfJ k EfJ 0) + (l + I + I)k)/(1 + 1 + I)k, V3 = «0 EfJ EfJ k) + (1 + 1 + l)k)/(1 + 1 + l)k. It follows that C- P(O) is isomorphic to (k EfJ k, k EfJ 0, EfJ k, (1 + l)k) . Vo = (k EfJ k EfJ k)/(l
° °
°
(bj By Example S.l.Xej.cdn C't J'(O) = (2,1,1 , 1),cdnC+2/(O) = (1,0,0,0), cdn C+ 1(1) = (1, 1,0,0), cdn C+ 1(2) = (1,0, 1, 0) and cdn C+ 1(3) =(1,0,0, 1). Once again, only C+ I (0) needs to be computed. By Example 6.1.5(c),
182
6. Representations over Fields and Exact Sequences
W(l(O» = (k EB k EB k , 0 EB k EB k, k EB 0 EB k , 0 EB 0 EB k), L(l(O» and C+ 1(0) = (k EB k, k EB 0, 0 EB k, (l + 1)k), as desired.
= (k, k , k, k),
There are several intriguing observations about the list in Example 6.2 .2. The first is that C+ and C- applied to an indecomposable injective representation of 53, respectively an indecomposable projective representation, resulted in an indecomposable representation. This property is, in fact, true for each finite poset and each indecomposable representation by Corollary 6.2.5(d) . Secondly, each indecomposable in rep(53, k) is both preprojective and preinjective. This property holds for each 5 such that rep(5, k) has finite representation type by Theorem 6.2.6. Finally, if V E rep(53 , k) and V is indecomposable, then V is isomorphic to C- C+ V if C+ V =1= 0 and V is isomorphic to C+ C- V if C- V =1= O. This property holds for each finite poset 5 by Corollary 6.2.4(c) and (d). Fundamental properties of the correspondences C+ and C- are derived in Corollary 6.2.4 by showing that the duality a of Lemma 6.1.1 is a factor of both C+ and
C-.
Proposition 6.2.3 If V E rep(5, k), 5 a finite poset, and k afield, then (a) aC+(V)
=
C-a(V); = a C+ and V has no projective summands, then p2(V) is naturally isomorphic to V ; (c) 0 ~ C+V ~ W(V) is an injective envelope and (d) W' (V) ~ C- V ~ 0 is a projective cover of U, (b)
if p
«c-o, c-
PROOF. (a) There is a projective cover 0 ~ M (V) ~ P( V) ~ V ~ 0 of U in
rep(5, k) with P(U)o = EB{Ui/Ur i E 5}, and P(U)j = EB IUd UJ: j ::::: i}. Then o ~ C+(V) ~ W(V) ~ L(V) ~ 0 is an exact sequence of representations with
and L(V)o
= Vo, L(V)j = E{V j : i i
j).
Hence, 0 ~ a L(V) ~ aW(V) ~ aC+(V) ~ 0 is an exact sequence of representations with aW(U)o
= a P(V)o = EB{a(Vi/Vr) : i E 5}, aW(V)j = EB{a(Vi/V7): j::: q,
and a L(U)o
= aVo, a L(U)i = a(E{Vj : i i
j}).
On the other hand, 0 ~ a V ~ a P(V) ~ a M(V) ~ 0 is an injective envelope of a V . Then 0 ~ K (a V) ~ W'(a V) ~ C-(a V) ~ 0 is an exact sequence of
6.2 Coxeter Correspondences
183
representations with
= a P(U)o = EB{a(U;/Ur): i E S}, W'(aU); = EB{a(UdUJ): t s i},
W'(aU)o
and
= aUo , K(U); = a(~{Uj: i i j}) . Thus, aW(U) = W'(aU) and aL(U); = K(U); for each i E S. This shows that aC+(U) = C-aU, as desired. K(aU)o
(b) There is an exact sequence of representations
0---+ C+(U) ---+ W(U) ---+ L(U) ---+ 0,
recalling that W(U) is injective with W(U)o = P(U)o and L(U)o Applying a yields an exact sequence of representations 0---+ a L(U) ---+ a W(U) ---+ p(U)
= Uo.
= aC+(U) ---+ 0
with a W(U) projective. Now, P(p(U» is a projective cover of p(U), so that aW(U) = P(p(U» EB P for some projectiverepresentation P. The definition of C+(p(U» yields an exact sequence 0---+ C+(p(U» EB / ---+ W(p( U» EB / ---+ L(p(U» ---+ 0
for someinjective / with W(p(U»o EB /0 = P(p(U»o EB Po = a W(U)o Hence, 0---+ a L(p(U» ---+ a(W(p(U» EB I) ---+ aC+ p(U) EB a(l)
= a P(U)o.
= p2(U) EB a(l) ---+ 0
is an exact sequence of representations with a(W(p(U» EB I) projective and a(W(p(U» EB 1)0 a 2P(U)o. Thus, a 2 : P(U)o ---+ a(W(p(U»EB 1)0 is a vector space isomorphism that induces a representation isomorphismex : U ---+ p2(U) EB a(l) with a(l) projective. Since U has no projectivesummands, a(l) = 0, whence / = 0 and P = O. Hence, U is isomorphicto p2(U). It is straightforward to see that this isomorphismis nat-
=
ural. (c) As in the proof of (b), a(W(U» ---+ aC+U is a projective cover of aC+U . By Lemma 6.1.1, C+(U) ---+ W(U) must be an injective envelopeof C+ U. (d) Since, C-aU = aC+U, it follows that W'(U) ---+ C- U is a projectivecover ofC-U .
Corollary 6.2.4 Let p
= aC+ and U E rep(S, k). Then:
(a) ifU has no trivial summands, then C+U is isomorphic to ap(U); (b) ifU has no cotrivial summands, then C-U is isomorphic to pa(U); (c) if U has no projective summands, then C- C+ U is isomorphic to U ; (d) if U has no injective summands, then C+C- U is isomorphic to U;
184
6. Representations over Fields and ExactSequences
(e) V is preprojective if and only if C+i V = 0 for some i ~ I.. and (f) V is preinjective if and only if V = 0 for some i ~ I.
c:'
PROOF. (a) As a consequence of Lemma 6.1.1, a p( V) = a 2 C+V is isomorphic to C+V for each V E rep(S, k) . As for (b), pa(V) = aC+a(V) = C-(a 2(V» by Proposition 6.2.3(a), and C-(a 2(V) is isomorphic to C-(V) by Lemma 6.1.1. (c), (d) Notice that C-C+V = (C-a)(aC+)(V) = p2(V) is isomorphic to V as a consequence of Lemma 6.1.1 and Proposition 6.2.3(b) . Assertions (e) and (f) are consequences of Proposition 6.2.1 and (c) and (d), respectively.
Corollary 6.2.5 Suppose V, V
E
rep(S, k), Sis a finite poset, and k is afield.
(a) If V has no injective summands and V has no projective summands, then Hom(V, C+V) is isomorphic to Hom(C-V, V). (b) If V has no projective summands, then End V is isomorphic to End C+ V. (c) If V has no injective summands, then End V is isomorphic to End C- V. (d) If V is indecomposable, then C+ V and C- V are indecomposable. (e) If V is either preprojective or preinjective, then End V is isomorphic to k. PROOF. (a) It is sufficient to prove that Hom(V , V) is isomorphic to Hom(p(V), p(V». To see this, observe that Hom(V, C+V) = Hom(V , ap(V» is isomorphic to Hom(p(V), a(V» by Lemma 6.1.1. Then Hom(p(V) , a(V» is isomorphic to Hom(pa(U), p2(V» = Hom(C -U, V) via Proposition 6.2.3. Let Pu : P( U) -+ U -+ 0 be the projective cover of U given in Lemma 6.1.4 and f: U -+ V. By the construction of P(U), W(U),and L(U),f inducesg: W(U)-+ W(V) and h: L(U)-+ L(V) with hpu = Pv S» where pu : W(U)-+ L(U). Since C+ U = Ker Pu, f induces f': C+ U -+ C+ V. This yields a homomorphism a:Hom(V,V)-+Hom(C+U,C+V) given by a(f)=f'. By Lemma 6.1.1, a :Hom(C+V,C+V)-+Hom(aC+V,aC+V) is an isomorphism. Since p = o C", a a : Hom(V, V) -+ Hom(p(V), p(V» is a homomorphism. Similarly, there is a homomorphism Hom(p(V), p( V» -+ Hom(p2V, p2V). It follows from Proposition 6.2.3(b) that Hom(V, V) is isomorphic to Hom(p(V) , p(V» . (b), (c) Now, Hom(V, V) is isomorphic to Hom(V, C+C- V) by Corollary 6.2.4(d), and by (a), Hom(V, C+C-V) is isomorphic to Hom(C-V, C-V). On the other hand, Hom(V, V) is isomorphic to Hom(C-C+ V, V), by Corollary 6.2.4(c), which is isomorphic to Hom(C+V, C+V) by (a). Statement (d) is an immediate consequence of (b) and (c), since a representation V is indecomposable if and only if 0 and I are the only idempotents of End V . (e) Assume that V is an indecomposable preprojective representation. Then V = c:' pet) for some t E S U {O} . By (a), End U = End C+i c:' pet) = End pet). But End pet) = k, since pet) is a I-dimensional representation. Similarly, if V is an indecomposable preinjective representation, then End U = End I (t) = k for some tESU{O}.
6.2 CoxeterCorrespondences
185
Theorem 6.2.6 Let S be afinite poset and k afield. Thefollowing areequivalent: (a) rep(S, k) has finite representation type; (b) If U is an indecomposable in rep(S, k), then U is preinjective; (c) If U is an indecomposable in rep(S , k), then U is preprojective; (d) End U is isomorphic to kfor each indecomposable U in rep(S, k).
In this case, ifU is an indecomposable in rep(S , k), then U is determined up to isomorphism by cdn U. PROOF. (a)::::} (b) [Ringel, 80].
(b) ::::} (d) follows from Corollary 6.2.5(e) . (d) ::::} (a) If rep(S , k) has infinite represention type, then S contains one of the critical posets T = S4 , (2,2,2),0,3 ,3),0,2,5), or (N, 4) as a subposet by Theorem 1.3.6. Choose an indecomposable U in rep(T, k) with End U not isomorphic to k, as given in the proof of Theorem 1.3.6. By Proposition 1.3.1, U gives rise to an indecomposable V in rep(S, k) with End V not isomorphic to k. (c)::::} (d) is a consequence of Corollary 6.2.5(e). (a)::::} (c) follows from (a)::::} (b), Corollary 6.2.4, and Proposition 1.4.7.
Example 6.2.7 The categoryteot S«, k) has tame representation type. Following is a complete list ofelements of Ind(S4, k): (a) indecomposable preprojectives U; all have endomorphism ring k and are classified by cdn U: (i) cdn c:"P(O) = (2n + 1, n, n, n, n),
U(XI , " " x n)
= (0, XI ," "
V(XI , •• • ,xn)
= (XI, ...
X n, X n , ..• , XI)
, X n, X n , • • • , XI,
and
0).
(ii) cdn C- 2n PO) = (2n - 1, n, n - 1, n - 1, n - 1),
c- 2n PO) =
(kn E9 k n -
I,
k n E9 0, 0 E9 k n -
I,
u'(k n -
I ),
v'(k n -
I
», where
= (XI, . . . ,Xn-I, XI, X2, ••• , Xn-I , 0) and V'(XI , • •• , xn -d = (XI, . •. , Xn-I , 0, XI ,···, xn-d .
U'(XI , .•• , Xn-I)
(iii) cdn C- 2n P(2) = (2n - 1, n - 1, n, n - 1, n - 1),
(iv) cdnC- 2n P(3) = (2n -1 ,n - 1, n -1,n ,n -1),
C-2n P(3) = (kn E9 r:', 0 E9 r:', u'(k n -
I
),
r
E9 0, v'(k n -
I
».
186
6. Representations over Fields and Exact Sequences
(v) cdn C- 2n P(4) = (2n - 1, n - 1, n - 1, n - I, n), C- 2n P(4)
= (k n $ r:' , 0 $ r:', u'(kn- t ), v'(k n- t ), k n $
0).
(vi) cdn c-(2n-l) P(l) = (2n, n - 1, n, n , n),
C-2n- 1 P(l)
= (k"
$ k" , w(k n- I), k" $ 0,0 $ k", (l
W(XI,·· ·, Xn-l)
(vii) cdn c-(2n-l) P(2)
+ I)(k n»
,
= (0, Xl, ... , Xn- l , Xl, .··, Xn-l , 0).
= (2n , n, n -
C-(2n-l) P(2) = (k n $
1, n, n),
r, k n $
0, w(k n- I ) , 0,
r, (l + I)(kn».
= (2n, n, n, n - 1, n), C-(2n -l) P(3) = (k n $ k", k" $ 0, 0 $ k" , w(k n- l), (l + I)(k n».
(viii) cdn c-(2n-l) P(3)
(ix) cdn c-(2n-l) P(4) = (2n, n, n, n , n C-(2n-l) P(4) = (k n $
r, k n $
0,
0, 0 $
v, (l + 1)(k n) , w(k n- ». 1
(b) indecomposable preinjectives V; all have endomorphism ring k and are classified by cdn V. (i)-(ix) Apply a to the representations in (a).
(c) cdn VA
= (2n , n , n , n, n)
VA = (k n tIJkn,k n tIJ 0, 0 tIJ k n, (1 + l)kn,(l+A)kn),EndVA :::::k[x]j (g(x)") , A an n x n k-matrix with minimal polynomial g(xY, g(x) irreducible.
(d) cdn V = (2n
+ 1, n , n + 1, n + 1, n)
V = (k n $kn+l,k n $O,O$kn+l ,a(kn+l),b(kn»),EndV:::::k[x]/(xn), where a(xl, . . . , xn+t> = (Xl, . . . ,xn, Xl, ... , Xn+l) and btx«, . .. , x n) (Xl, ... , Xn, 0, Xl> . . . , Xn) andfive other similar representations obtained by permuting the subspaces, VI, V2, V3, V 4 corresponding to permutations ofcdn V.
=
PROOF . [Gelfand, Ponomarev 70] or [Brenner 74B].
EXERCISES
1. Prove that if S is a forest, then C+ , C" : rep(S , k) -? rep(S, k) are functors. What is the situation if S is not a forest?
2. Confirm the assertions of Example 6.1.5(c), (d), and (e). 3. Explicitly construct the representations in Example 6.2.7(b) and (d).
6.3 Almost Split Sequences
187
4. Suppose U E rep(S, k ) with cdn U = (uo, u; : i E S). (a) Prove that if cdn a (U ) = (vo, Vi : i E S), then Vo = uo and 'E{u j : j ~ i} + 'E{Vj :j > i} = Uo for each i E S. (b) Prove that if cdn p (U ) = (wo, Wi : i E S), then Wo = 'E{(Ui - uo) : i E S}and ui ; = ui for each i E S. 5. Let X, Y E rep(S, k) be l-dimensional representations. Prove that for each positive integer n : (a) C+n X is isomorphic to C+n Y if and only if cdn C+n X = cdn C+n Y. (b) C- n X is isomorphic to c » Y if and only if cdn c » X = cdn C- n Y.
6.3 Almost Split Sequences Almost split sequences are one of the most important tools for classification of finitely generated modules over finite-dimensional k-algebras with tame representation type. The most important theorem about almost split sequences in rep(S, k) is that they exist (Theorem 6.3.4). Even more, the left-hand term of an almost split sequence in rep(S, k) determines the right-hand term, and the right-hand term determines the left-hand term. Almost split sequences in rep(S, k) can be constructed explicitly (Corollary 6.3.5 ). A representation morphism f: V -+ V in rep(S, k) is a splitting monomorphism if there is g : V -+ V with gf = I u and a splitting epimorphism if there is g : V -+ V with f g = I v. An exact sequence
£ :0-+ V ~V";'W-+O of representations in rep (S, k) is split exact if there is h: V -+ V with hf = I u , the identity on V. Equivalently, there is h : W -+ V with gh = I w- Then E is an almost split sequence if: (i) E is not a split exact sequence; (ii) if h : Y -+ W is not a splitting epimorphism, then there is a : Y -+ V with ga = h; and (iii) if h : V -+ Y is not a splitting monomorphism, then there is a : V -+ Y with af =h. The sequence E is left almost split if (i) and (iii) hold and right almost split if (i) and (ii) hold. Given exact sequences
£ :0-+ V~V";'W -+ 0, J'
g'
£':0-+ V '-+ V '-+ W -+ 0, call (a , (3, y): E -+ E' a homomorphism of exact sequences if a: V -+ V' , {3 : V -+ V ' , and y : W -+ W ' are representation morphisms with {3f = f'a, and yg = g'{3 . The homomorphism (a , (3, y) is an isomorphism if a, {3, and y are representation isomorphisms.
188
6. Representations over Fields and Exact Sequences
As a consequence of the following lemma, almost split sequences are unique up to isomorphism. Lemma 6.3.1 Suppose Sis a finite poset, k is afield, and f
g
E :O~V~V~W~O
is an almost split sequence in rep(S, k) (a) Both End V and End Wa re local rings. In particular, V and Ware inde-
composable representations.
(b)
If
f'
g'
E':O~ V'~V'~W~O
is a right almost split sequence with V' indecomposable, then there is an isomorphism (a , a, 1) : E' ~ E.
(c)
If
f'
e'
E' :O~ V~V '~W'~O
is a left almost split sequence with W' indecomposable, then there is an isomorphism (1, a, a): E' ~ E.
(a) Suppose hand h' are nonunits in End W. Since End W is a finitedimensional k-algebra, h and h' are not splitting epimorphisms. Choose f3 , f3' : W ~ V with gf3 = hand gf3' = h'. Then g(f3 + f3') = h + h' = u is not a unit of End W ; otherwise, u- I g(f3 + f3') = 1, contradicting the assumption that E is not split exact. (b) Since E' is not split exact, g' is not a splitting epimorphism. Choose a : V' ~ V with ga = g'. This choice is possible, since E is almost split. But E' is right almost split, so that there is f3 : V ~ V' with g' f3 = g. Since g' f3a = g'l w. there is a homomorphism (y, f3a, 1) : E' ~ E' with Ba]' = I'r . If y is a unit of End V', then a: f'(V ') ~ f'(V') is a unit, since V ' is indecomposable. In this case, (a, a , 1): E' ~ E is the desired isomorphism. Now assume that y is not a unit of End V'. Since V'is indecomposable, y is nilpotent, say yn = 0 for some n. Then g'(f3a)n = g', an epimorphism from V' to W, and (f3a)n induces h': W ~ V', since yn = O. As g'(f3a)n = g'l w' and g' is onto , it follows that g'h' is a unit of End W, a contradiction to the fact that g' is not a splitting epimorphism. (c) The proof is dual to that of (b). PROOF.
Almost split sequences are minimal with respect to being a nonsplit exact sequence, provided that the left- and right-hand terms of the sequence are known.
6.3 Almost Split Sequences
189
Lemma 6.3.2 Assume that S is a finite poset, k is a field, and that there is an almost split sequence
Let
be a nonsplit exact sequence in rep( S, k) . The following statements are equivalent :
(a) E ' is a left almost split sequence. (b) If V' is a nonzero pure subrepresentation of V , then 0 --+ V / V ' --+ V' / !'(V') --+ W --+ 0 is a split exact sequence. (c) E' is an almost split sequence. (d) If W' # W is a pure subrepresentation of W , then the sequence 0 --+ V --+ (g')-I(W') --+ W' --+ 0 is split exact. (e) E ' is a right almost split sequence. PROOF . (a) =} (b) Since V' is nonzero, h : V --+ V / V ' is not a splitting monomorphism . By (a), there is a: V' --+ V / V ' with a]' = h. Now, af'(V') ~ ker h = V', so a : V' / f'(V') --+ V / V' with a]' = 1, as desired. (b) =} (c) Notice that g' : V' --+ W is not a splitting epimorphi sm, since E' is not a split exact sequence. Since E is almost split, there is a : V ' --+ V with ga = s'. If a' = f-Ia!' is a unit of End V, then (a ' , a, 1) : E' --+ E is an isomorphism, whence E' is an almost split sequence . If a ' is not a unit, then ker e ' is a pure nonzero subrepresentation of V . By (b), there is a split exact sequence 0--+ V /ker e" --+ V' / f '(kercx) --+ W --+ O. Since a sends V' /ker o' isomorphically to a subrepresentation of V and ga = s'. this split exact sequence induces a splitting of E', a contradiction. (c) =} (a) is clear. The equivalence of (c), (d), and (e) is proved by a dual argument.
Corollary 6.3.3 Suppose E: 0 --+ V
~V
s; W --+ 0
is an almost split sequence in rep(S, k) and either End V or End W is a division algebra. If I'
g'
E' : 0 --+ V --+ V' --+ W --+ 0 is an exact sequence that is not split exact, then E' is an almost split sequence. PROOF . Assume that End W is a division algebra. Then!, : V --+ V ' is not a splitting monomorphism, since E' is not a split exact sequence . Since E is almost
190
6. Representations over Fields and Exact Sequences
split, there is a: V ~ V' with af = f '. Then a induces f3 : W ~ W with f3g = g'«. Moreover, 0 =1= f3, because E is not split exact. Since End W is a division algebra, f3 is an automorphism of W. Thus, 0, a, f3) : E ~ E' is an isomorphism, and E' is an almost split sequence. The proof for the case that End V is a division algebra is dual. The proof of the following theorem is beyond the scope of this book .
Theorem 6.3.4 [Bautista, Martinez 79] Let S be a finite poset and k afield.
If W is an indecomposable in rep(S, k) that is not projective, then there is an almost split sequence 0 ~ C+ W ~ V ~ W ~ O. (b) If V is an indecomposable in rep(S, k) that is not injective, then there is an almost split sequence 0 ~ V ~ V ~ C- V ~ O. (a)
Given that almost split sequences exist, they can be constructed as follows :
Corollary 6.3.5 Suppose S is a finite poset, k is a field, and V E rep( S , k) is a nonprojective indecomposable representation. Let Pu : P(V) ~ V be a
projective cover of V and
o ~ C+V ~
W(V)~ L(V) ~ 0
the resulting injective envelope «c-u. (a) There is a morphism i : V ~ L(V) induced by inclusion of Vo in L(V)o. (b) Define B = (Bo, B, E S) with Bo = {(x, y) E L(V) EB W(V) :ip u(x) = iCY)} and B, = Bo n ((C+V) i EB l-t) for each i E S. The sequence 0 ~ C+V ~ B ~ V ~ 0 is an almost split sequence in rep(S, k) . PROOF. Exercise 6.3.4.
Let (XI, " " X n ) be an n-tuple of l-dimensional representations in rep(S , k). Choose a representation embedding f i : Xi ~ I (0) and define V (X I, .. . , X n) E rep(S, k) to be the kernel ofEBfi : EB{X i : i E S}~ 1(0). Representations V(X i , ... , X n ) are the representation analogue of bracket groups, as defined in Chapter 7.3 .
Example 6.3.6 Suppose X E rep(S, k) is nonprojectivewith dimension 1. Then C+ X = V(l(i): Xi =1= O)for I-dimensional injective representations I(i), and
the sequence
o~
-4 VeX, I(i) : Xi =1= 0) ~ X ~ 0 = (0, XI , • •. , X n ) , g(x , XI, •• • ,Xn ) = X is an almost split
V(l(i): Xi
with f(xI, . .. ,Xn ) sequence.
=1=
0)
6.4 A Torsion Theory andLocalizations PROOF. Up to isomorphism, it suffices to assume X cover PX : P(X)
= $(P(i) : X; =f. O} -+
~
191
/(0). There is a projective
X with Px($;Y;)
= ~;y;.
Then 0 -+ C+ X -+ W(X) -+ L(X) -+ 0 is an injective envelope of C+ U with W(X) = ${I(i) : X; =f. O} an injective representation and X ~ L(X) a l-dimensional representation. Hence, C+ X = U(l(i) : (X); =f. 0), being the representation kernel of PX : W(X) -+ L(X). Notice that U(X, /(i): (X); =f. 0) = B, as defined in Corollary 6.3.5.
EXERCISES I. A representation morphism f : U -+ V in rep(S, k) is irreducible if f is neither a split monomorphism nor a split epimorphism and if f = gh for some h :U -+ Wand g : W -+ V, then either h is a splitting monomorphism or g is a splitting epimorphism.
Prove that if 0-+
U~V~W-+O
is an almost split sequence in rep(S, k) and V = VI ED . . . ED Vn with each V; an indecomposable representation, injections i i - and projections tti - then it j f and gij are irreducible morphisms foreach j . 2. Prove thatif f : U -+ V is irreducible, then f =f. L,j h j g j whenever Zj is an indecomposable representation and g j : U -+ Zi - h j : Zj -+ V arenotisomorphisms. 3. Prove Lemma 6.3.I(c) andtheequivalence of (c), (d), and(e) in Lemma 6.3.2. 4. Provide a proofforCorollary 6.3.5.
6.4 A Torsion Theory and Localizations Given a finite poset S with unique maximal element 00 and discrete valuation ring R, there is an associated category of representations of a quiver over R and a torsion theory for which the poset representations of S are the torsion-free elements (Corollary 6.4.6). A poset S can be viewed as a quiver, a directed graph. In this case, vertices are elements of S, and there is a directed edge i -+ j if and only if i :::: j in S. Representations of quivers and finitely generated modules over finite-dimensional k-algebras, k a field, are investigated thoroughly in [Ringel 84] and [Simson 92]. This chapter is devoted to a brief discussion of the category of quiver representations of S over R with relations induced by the poset S and its connection with rep(S, R). Define a category Qrep(S, R) with objects M = (M; : fij : i :::: j) where each M; is a finitely generated free R-module, fij : M; -+ M, is an R-homomorphism with pure image whenever i :::: j in S, [u is the identity of M; for each i E S, and
192
6. Representations over Fields and Exact Sequences
/ik fij = hk whenever i ~ j ~ k in S. A morphism from M to M' is g = {gi : i E S} with each gi : M, ~ M; an R-homomorphism and gihj = fijgi whenever i ~ j in S. Alternatively, Qrep(S, R) can be thought of as a category of modules over the R-incidence algebra of S that are finitely generated and free as R-modules; Exercise 6.4.1.
Proposition 6.4.1 Let S be afinite poset with unique maximal element 00 and R a discrete valuation ring. Then Qrep(S, R) is an additive category closed under direct sums, idempotents split in Qrep(S, R), andrep(S, R) is afull subcategory of Qrep(S, R). PROOF . It is straightforward to confirm that Qrep(S, R) is a category and the set ofmorphisms from M to M' is an abelian group for each M, M' in Qrep(S, R). If M = (M j, f ij: i ~ j) and M' = (M;, f: j : i ~ j) arein Qrep(S, R), then M E9 M' = tM, E9 M;, fij E9 f: j : i ~ j) is a direct sum of M and M' in Qrep(S, R). If e = {ej: i E S} is an idempotent in End M , the endomorphism ring of M in Qrep(S, R), then el = e, for each i. Hence, M = K E9 L with K = (ei(Mi), ej h jei: i ~ j) and K E Qrep(S, R), since ei!ijei has pure image in ej(Mj) and ekf·ke· J J Je.·e· IJ I = ekJ"ke I I' whenever i -< J' < - kin S. Let V = (V oo , Ui : i E S) E rep(S, R). Then M = (V ;, hj : i ~ j) E Qrep(S, R), where fij : Vi ~ U, is inclusion for i ~ j in S. Each hi has pure image, since Vi is pure in V oo , hence in U], whenever i ~ j, and /ikhj = hk whenever i ~ j ~k in S. ThecorrespondencefromHom(V, V') in rep(S, R) to Hom(M , M') in Qrep(S, R) given by f ~ {h = f: i E S} is an isomorphism for each V, V ' E rep(S, R) . It follows that rep(S, R) is a full subcategory of Qrep(S, R) .
If N = tN], gij) and (M i, fij) are in Qrep(S, R), then N is a subrepresentation of M if N, is a submodule of M, for each i and each gij is the restriction of hj to Ni. Given M = (M i, fij : i ~ j) E Qrep(S, R), define T(M) = (T(M)i, gij : i ~ j) by T(M)oo = 0, T(M)i = ker hoo if i E S\{oo}, and gij : T(M)i ~ T(M)j the restriction of hj to T(M)i. Then gij is well-defined, and the image of each gij is pure, since /ioohi = f ioo and the image of f ioo is pure. Notice that T(M) is a subrepresentation of M and each T(M)i is a free R-module, being a submodule of the free R-module M) . Moreover, M;/T(M)j is a finitely generated torsion-free R -module, hence free .
Proposition 6.4.2 Let S be a finite poset with unique maximal element 00 , R a discrete valuation ring, and M , N E Qrep(S, R). Then T is a functor from Qrep(S, R) to Qrep(S, R) such that: (a) (b) (c) (d) (e)
T(M) is a subrepresentation of M that preserves monomorphisms; T(M E9 N) = T(M) E9 T(N); T(T(M» = T(M); T(M) = 0 ifand only if hoo is a monomorphism for each i E S; and T(M/T(M» = O.
6.4 A Torsion Theory and Localizations
193
If f: M -+ N, then T(f) = f : T(M) -+ T(N). Hence, T: Qrep(S, R) -+ Qrep(S, R) is a functor and T(f) is a monomorphism if f is a monomor-
PROOF. (a)
phism. (b) Apply the definitions of T(M) and direct sums . (c) By definition, T(M) = (T(M)j , gjj : i ~ j) with T(M)oo = 0, T(M)j = ker f;oo if i E S\{oo}, and gjj: T(M)j -+ T(M)j, the restriction of fij to T(M)j. Then 0 = gjoo : T(M)j = ker fjoo -+ T(M)oo = 0, whence ker gjoo = T(M)j for each i, This shows that T(T(M» = T(M). Statement (d) follows immediately from the definition of T(M). (e) Write M IT(M) = (M;/T(M)j , hij), where hij : M;/T(M)j -+ MjIT(M)j is induced by f;j . Since hjoo: M;/T(M)j -+ MooIT(M)oo = Moo is a monomorphism, ker h joo = 0 for each i, Now apply (d). Call M E Qrep(S, R) torsion if T(M) = M and torsion-free if T(M) = O. Then M = (M j, f;j) is torsion if and only if fjoo = 0 for each i E S. A nonempty subset P of S is an up-set in S if given i E P and i < j in S, then j E P. An up-set is called a filter by some authors. Let M = (Mj , fij) E Qrep(S, R) and define M p = «Mp)j,gij), where (Mp)j = Moo if i E P, (Mp)j = M, if i 1. P, gjj = 1· if i ~ j in S with i E P, and gjj = f;j if i ~ j in S with j 1. P. Since P is an up-set, M» is a well-defined element of Qrep(S, R) . Call M = (M j, fij) E Qrep(S, R) cotrivial if each M, = Moo and fij is the identity for each i ~ j in S. If S has a unique minimal element and M is cotrivial, then M is both injective and projective. The constructions of the next two propositions resemble localizations of modules.
Proposition 6.4.3 Suppose S is a finite poset with unique maximal element, R is a discrete valuation ring, M E Qrep(S, R), and P is an up-set in S. Then (a) M ~ M», (Mp)p = M», and T(Mp) = T(M)p; (b) if M is torsion-free, then M p is torsion-free, and if M is torsion, then Mp
is torsion; and (c) if P = S, then M p is cotrivial. PROOF . Immediate consequences of the definitions.
There is a localization "dual" to that of M r- If P is an up-set in Sand M E QRep(S, R), define M P = «MP)j, gij), where (MP)j = M, if i E P; (MP)j = 0 ifi 1. P;gjj = fijifi ~ j,iEP;andg jj =Oifi ~ j,j 1. P .SincePisan up-set, M P is a well-defined element of QRep(S, R).
Proposition 6.4.4 Suppose S is a finite poset with unique maximal element, R is a discrete valuation ring, M = (M j, f ij) E QRep(S, R), and P is an up-set in S. Then (a) M P ~ M, (MP)P
= M P, and T(M P) = T(M)P;
(b) (Mp)P = (MP)p = (Nj, gij) with N, = Moo for i E P, N, = gij = 1 for each i;
°
ifi 1.P, and
194 (c)
6. Representations overFields andExact Sequences
If M is torsion-free, then M P is torsion-free, and if M is torsion, then M P is torsion;
(d) M = I:{M P : P up-set}; and (e) there is a subset X ofthe set of up-sets in 5 with M = E9{M P: P E X}for
each torsion M in Qrep(5, R) ifand only if 5\{oo} is the disjoint union of {P\{oo}: P EX}. PROOF. Statements (a), (b), and (c) are immediate consequences of the definitions. As for (d), if i E 5, then i is an element of the up-set P, = {s E 5 : s ::: i}. Hence, if M = (M], fij) E Qrep(5, R), then M = I:{M P : P up-set}, since M, = (MP)j for P =Pj • Assertion (e) follows from the arguments in (d), observing that if M is torsion, then Moo = O. The next corollary is a reduction step in a characterization of those finite posets 5 with unique maximal element such that T(M) is a summand of M for each M E Qrep(5, R). The complete characterization is left as an exercise (Exercise 6.4.2).
Corollary 6.4.5 Assume that 5 is afinite poset with unique maximal element 00, R isa discretevaluation ring, and thereisa subset X ofthe set ofup-sets in 5 with 5\{00} the disjoint union of{P\{oo}: P E X}. If ME Qrep(5 , R), then T(M) is a summand of M ifand only ifT(M P) is a summand of M P for each P EX. PROOF. If T(M) is a summand of M and P is an up-set, then T(M)P is a summand of M P. But T(Ml = T(M P) by Proposition 6.4.4(a). Conversely, suppose that fP : M P ~ T(Ml is a splitting of the inclusion of T(M P) = T(Ml in M P for each P EX . By Proposition 6.4.4(e), T(M) = E9{T(M)P : P E X} and M, = E9{(M P)j : P E X} for each i E 5\{00}. Hence, /; = E9p M, ~ T(M) j is a splitting for each i E 5. Define foo = O. Then f = {lj : i E 5}: M ~ T(M) is a splitting of the inclusion of T(M) in M, since T(M)oo = O.
r:
For s E 5, pes) = (P(s)j , /;j) E Qrep(5, R) is defined by P(S)j = R if s ~ i, P (s), = 0 if s is not less than or equal to i, /;j = 1 if s ~ i, and /;j = 0 otherwise. Define P(O) = (P(O)j , /;j) E Qrep(5 , R) by letting P(O)j = Rand /;j = 1 for each i E 5 and i ~ j in 5. Observe that each pes) is torsion-free. The following result is given in [Butler 87], [Bautista, Martinez 79], and [Buenermannn 81, 83] for the case that R is a field.
Corollary 6.4.6 (a) Projectivesin Qrep(5 , R) are, up to isomorphism, summands ofdirectsums of representations ofthe form P(s) . (b) Torsion-free elements of Qrep(5, R) are, up to isomorphism, the elements of rep(5, R).
6.4 A Torsion Theory andLocalizations
195
(c) An ME Qrep(S, R) is torsion-free if and only if M is a subrepresentation ofa projective representation of Qrep(S, R). PROOF. (a) Each pes), hence the direct sum of P(s)'s, is readily seen to be projective. Moreover, if ME Qrep(S, R), then there is N E Qrep(S, R) with N isomorphic to a direct sum of P(s)'s and an onto morphism N -r M. Specifically, N = EB{N(s) :s E S}, where N(s) = (N(s);, f ij) with N(S)i = M, ifi :::: s, N (s), = 0 if its, fij = 1 if s ::: i ::: j, and fij = 0 otherwise. Notice that each N(s) is isomorphic to a direct sum of copies of pes). If M is projective, then M is a summand of N. (b) If V E rep(S , R), then fioo :Vi -r Voo is inclusion, so that T(V)i = ker fioo = 0 for each i. Thus, T(V) = O. Conversely, if M = (M] , f ij :i ::: j) E Qrep(S , R) is torsion-free , then each fioo is a monomorphism by Proposition 6.4.2(d). But hoofij = /;00 if i ::: j in S, whence each Iu is a monomorphism. Hence, M is isomorphic to V = (Moo, /; oo(Mi): i E S\{oo}) E rep(S , R), observing that each fioo has pure image in Moo . (c) In view of (a), Proposition 6.4.2(b), and the fact that each pes) is torsion-free, a projective representation is torsion-free . Since T preserves monomorphisms, a subrepresentation of a projective representation is also torsion-free . Conversely, if M = (Mo, Mi) E rep(S , R), then, by Proposition 6.4.3, M is a subrepresentation of a cotrivial projective representation.
EXERCISES I. Let S be a finite poset and R a discrete valuation ring. Define the incidence algebra RS to be the set of matrices M = (aij) such that aij E R if i ::::: j in Sand aij = 0 otherwise. (a) Prove that RS is a ring withidentity. (b) Prove thatthere is a fully faithful embedding Qrep(S, R) -+ mod RS, thecategory of finitely generated RS-modules. Describe the image of thisembedding. 2. Let S be a finite posetwith unique maximal element 00 and R a discrete valuation ring. Prove that T(M) is a summand of M for each M E Qrep(S , R) if andonly if S\{oo} is a forest.
NOTES ON CHAPTER 6 Descriptions of projectives, injectives, projective covers, andinjective envelopes in rep(S, k) aregiven in [Gabriel 73A). Coxeter correspondences on rep(S, k) arecommonly defined by regarding rep(S, k) as a subcategory of the category modkS of finitely generated modules over the incidence algebra kS of S and using the Auslander-Reiten correspondences DTr and TrD for these modules [Bautista, Martinez 79], [Drozd 74], and [Btinermann 81 , 83]. In this chapter, the definitionand development of Coxeter correspondences, in the spiritof [Gelfand, Ponomarev 70]forthecasethatS is an antichain, areentirely within thecategory
196
6. Representations over Fields and Exact Sequences
rep(S , k) and without reference to any module category. Hence, they are immediately available for quasi-homomorphism categories of finite rank Butler groups , as briefly discussed in Section 7.1. The existence of almost split sequences in rep(S, k) is derived in [Auslander, Smale 81], as well as [Bautista, Martinez 79] and [Biinermann 81, 83] using the torsion theory of Section 6.4, from the existence of almost split sequences in mod kS [Auslander, Reiten 75]. There does not seem to be a proof of their existence entirely within the category rep(S , k) . All indecomposable representations for the critical posets S4, (2, 2, 2), 0, 3, 3), (N, 4), and 0 , 2, 5) are given in [Ringel 84] via a description of the Auslander-Reiten quiver. The list is not quite as explicit as that of Example 6.2.8 for S4. An intriguing aspect that is not discussed herein is the characterization of the representation type of rep(S, k) in terms of an integral quadratic form qS, see Exercise 6.1.8. If rep(S, k) has finite representation type, then there is a one-to-one correspondence from the vectors cdn U, U E 1nd(S, k), to the roots of qS [Drozd 74]. Moreover, the critical posets can be associated with certain Dynkin diagrams . See [Ringel 84] and [Simson 92] for comprehensive discussions of the role of quadratic forms in representations of partially ordered sets and quivers and characterizations of the representation type of rep(S , k) in terms of properties of q s . Techniques from the theory of torsion-free abelian groups, particularly the analogue of rank-I groups and balanced exact sequences as discussed in Chapter 7, have been carried over to rep(S , k) for a finite poset S and field k in [Arnold , Vinsonhaler 92] and [Nonxga , Vinsonhaler 97]. The representation type of the category Qrep(S, k) for a finite poset S and a field k is completely described in [Loupias 75].
7 Finite Rank Butler Groups
7.1 Projectives, Injectives, and Exact Sequences Free groups are the projectives and divisible groups are the injectives for the category of torsion -free abelian group s [Hungerford 74]. Changing the category and the class of defining homomorphisms can change the projecti ves and injectives. Properties of Coxeter correspondences and almost split sequences are applied to categories of finite rank Butler groups in this section. A group G in B(T ) is projective in B (T ) if whenever f : H --+ K is an epimorphism of groups and g : G --+ K , then there is h : G --+ H with f h = g. The group G is injecti ve in B(T ) if for f: H --+ K a monomorphism of groups in B (T) with pure image and g : H --+ G , there is h : K --+ G with hf = g . Proposition 7.1.1 Let T be afinite lattic e of types. (a) The projectives in B(T) are .(D)-homogeneous completely decomposable groups, .(D) the small est element ofT. (b) The injectives in B(T) are .(oo)-homogeneous completely decomposable groups , .(00) the largest element ofT.
f: H --+ K is an epimorphism of group s in B(T) , X is a rank-l group with type .(D) , and D f. g : X --+ K. Let 0 f. x E X and 0 f. y E H with f( y ) = g(x). If Y is the pure rank-l subgroup of H generated by y , then Y E B(T), and so .(0) ~ type Y. Since type X = .(0), there is some h : X --+ H with f h = g. A standard argument using injections and projections for finite direct sums [Hungerford 74] shows that a . (D)-homogeneous completely decomposable group is projective in B(T).
PROOF. (a) Suppose
198
7. Finite Rank Butler Groups
Conversely, let G be projective in B(T). Since G E B(T), there is an epimorphism f : C -+ G , with C a completely decomposable group in B(T), by Lemma 3.3.1. Because G is projective, there is h : G -+ C with hf = IG . Therefore, G is a summand of C, hence completely decomposable by Theorem 3.1.7(a). It remains to show that if X is a rank-I summand of C with type X > r(O), then X is not projective in B(T). To see this, let Y be a rank-I group in B(T) with type Y = r(O) < type X. Choose 0 =I x E X and 0 =I y E Y. Since Y is reduced, there is an integer n such that na = x has no solution a E Y EI1 X. Define G to be the pure subgroup of Y EI1 X generated by ny + x . Then type G = type Y and type (Y EI1 X)/G = type X = r. Now assume, by way of contradiction, that X is projective in B(T). Since (Y EI1 X)/G is isomorphic to X, G is a summand of H = Y EI1 X, say H = G EI1 D. Then type D = r , from which it follows that H(r) = X EI1 D . But y tf. X EI1 D , since na = x has no solution a E Y EI1 X, a contradiction. (b) Let X bearank-l group with type X = r(oo), G and HE B(T), f : G -+ H a monomorphism with pure image, and 0 =I g : G -+ X. Then QX = Q @z X is an injective abelian group, so there is h : H -+ QX with hf = g. Now, type h(H) E cotypeset H ~ T, so that type h(H) ~ type X, the largest element of T . Consequently, there is h: H -+ Xwith hf = g . This shows that X, hence a r(oo)-homogeneous completely decomposable group, is injective in B(T). Conversely, suppose G E B(T) is injective in B(T). Then G is a pure subgroup of a completely decomposable group C in B(T) by Lemma 3.3.1. Since G is injective in B(T), G is a summand of C, hence completely decomposable by Theorem 3.1.7. It now suffices to show that if Y is a rank-l summand of G with type Y < type X = r(oo), then Y is not injective in B(T) . In this case, G must be a r(oo)homogeneous completely decomposable group. A construction like that of (a) shows that Y EI1 X has a pure rank-I subgroup with type equal to type Y that is not a summand. Hence, Y cannot be injective in B(T). A finite rank Butler group G is balanced projective if whenever f: H -+ K is a balanced epimorphism of finite rank Butler groups Hand K and g : G -+ K, then there is h : G -+ H with f h = g . The group G is cobalanced injective if for Hand K finite rank Butler groups , f : H -+ K a cobalanced monomorphism, and g : H -+ G , there is h : K -+ G with hf = g.
Proposition 7.1.2 For finite rank Butler groups, completely decomposable groups are the balanced projective groups and the cobalanced injective groups. PROOF. Suppose f
: H -+ K is a balanced epimorphism, X is a rank-I group with type X = r, and 0 i= g: X -+ K. There is a pure rank-I subgroup Y of H with type Y ~ rand f(Y) = g(X). But type Y ~ type f(Y) = type X = r , whence type Y = r = type X. It follows that there is an isomorphism h : X -+ H with f h = g, and so X is balanced projective. Consequently, completely decomposable groups are balanced projective.
7.1
Projectives, Injectives, andExactSequences
199
Conversely, if G is a balanced projective finite rank Butler group, then there is a completely decomposable group C and a balanced epimorphism f : C ~ G, by Lemma 3.2.3(a). Since G is balanced projective, G is isomorphic to a summand of C, hence completely decomposable. This shows that completely decomposable groups are the balanced projectives. To see that a completely decomposable group is cobalanced injective , assume that f : H ~ K is a cobalanced monomorphism and X is a rank-l group with type X = r. It suffices to prove that the induced homomorphism Hom(K, X) ~ Hom(H, X), given by h 1-+ hf, is onto. In this case, X, hence a completely decomposable group, is cobalanced injective . It is also sufficient to assume H [r] = K [r] = O. This is so because Hom( H, X) = Hom(HIH[r] , X) by the definition of G[r] and (HIH[r])[r] = 0 by Lemma 3.1.9(a). Similarly, Hom(K, X) = Hom(K I K[r], X) and (KI K[r])[r] = O. To complete the proof that X is cobalanced injective, let g : H ~ X and Y = g(H) with type Y = a ::: r . Then g: HI H[a] ~ Y is onto and f : HI H[a] ~ K I K [a] is a pure monomorphism, since f: H ~ K is cobalanced. Define G = (Y $ (KIK[a]))IM, M the pure subgroup of Y $ (KIK[a]) generated by {(g(a),-f(a)) :a E HIH[a]) . Then KIK[a] ~ G is onto, from which it follows that G[a] = 0, since (KIK[a])[a] = 0 by Lemma 3.1.9(a) . As Y is a pure subgroup of G with Yea) = Y and G[a] = 0, Y is a summand of G by Proposition 3.1.15(b) . Since Y is a summand of G, there is h': G ~ Y with the restriction of h' to Y the identity endomorphism of Y. Define h : K ~ Y by h(k) = h'«k + K[a]) + M) E h'(G) = Y. If a E H, then hf(a) = h'«f(a) + K[a]) + M) = h'(g(a) + M) = g(a), and so hf = g, as desired. Conversely, assume that G is cobalanced injective. Then G is a cobalanced subgroup of a completely decomposable group C by Lemma 3.2.3(b). Hence, G is a summand of C, so that G is completely decomposable, as desired. A sequence 0 ~ G ~ L ~ K ~ 0 of groups in B(T)Q is exact if f is a monomorphism, (ker g + image f)1(ker g n imagef) is finite, and (image g + K)/(image g) n K is finite. In this case, 0 ~ QG ~ QL ~ QK ~ 0 is an exact sequence of Q-vector spaces . Proposition 7.1.3 Let H : B(T)Q ~ rep(ST, Q) be the category equivalence (QG, QG(r) : rEST). Then 0 ~ G ~ given in Theorem 3.3.2 with H(G) L ~ K ~ 0 is an exact sequence of groups in B(T)Q if and only if 0 ~ H(G) ~ H(L) ~ H(K) ~ 0 is an exactsequence in rep(S, k).
=
PROOF. Assume that 0 ~ G ~ L ~ K ~ 0 is an exact sequence in B(T)Q. Then 0 ~ QG ~ QL ~ QK ~ 0 is an exact sequence of vector spaces. It is sufficent to prove that if rEST, then QG(r) ~ QK(r) ~ 0 is onto. In this case, QG(r) = QG n QL(r) and 0 ~ QG(r) ~ QL(r) ~ QK(r) ~ 0 is an exact sequence for each rEST , as desired . Let X be a pure rank-l subgroup of K with type X :::: rEST and A = g-l(X) the preimage of X in L . Then A E B(T), being quasi-isomorphic to a pure subgroup
200
7. Finite Rank ButlerGroups
of L E B(T). Hence, A = Y1 + ...+ Yn for pure rank-l subgroups Yi of A with g(Yi):f. 0, type Yi = a., and type X = U{ai : 1 ::: i ::: n} ::: r. Since the lattice of all types is distributive, r = U{ r nai : 1 ::: i ::: n}. But r is join irreducible, whence r r n a, and a, ::: r for some i, Thus, QX Qg(Yi) with Yi E L( r ), as desired. As for the converse, since H is a category equivalence , there is a category equivalence H' :rep(Sr, k) ~ B(T)Q with H H' naturally equivalent to the identity functor on rep(Sr, k) and H' H naturally equivalent to the identity functor on B(T)Q. It follows that if 0 ~ H(G) ~ H(L) ~ H(K) ~ Oisanexactsequence in rep(S, k), then 0 ~ G ~ L ~ K ~ 0 is an exact sequence in B(T)Q.
=
=
An element r of a finite lattice of types T is meet-irreducible if whenever r = 8 n a with 8, a E T , then r = 8 or r = a .
Corollary 7.1.4 Let T be afinite lattice oftypes. (a) Projectives in B(T)Q are direct sums of rank-l groups X with type X a join-irreducible type in T. (b) lnjectives in B(T)Q are direct sums of rank-l groups X with type X a meet-irreducible type in T . (c) Each G E B(T) has an projective cover PG : peG) ~ G in B(T)Q, where peG) = ${G(r)/G#(r) :r E SrI. (d) Each G E B(T)Q has an injective envelope iG : G ~ leG) in B(T)Q, where leG) = ${G*[r]/G[r] : r E Sr}.
PROOF. Statements (a) and (b) follow from Proposition 7.1.3 and Lemma 6.1.2, while (c) and (d) are consequences of Lemma 6.1.4 and Propo sition 3.2.8. The duality a of Lemma 6.1.1 induces a duality for categories of Butler groups .
Theorem 7.1.5 Let T' and T' befinite lattices oftypes and a : T'
~
T a lattice
anti-isomorphism. Then: (a) there is a contravariant category equivalence Da : B(T')Q ~ B(T)Q such
that Da-I D a is naturally equivalent to the identity functor on B(T')Q; (b) Da(G(r» is quasi-isomorphic to Da(G)/ Da(G)[a(r)]for each rET'; (c) Da sends balanced sequences in B(T')Q to cobalanced sequences in B(T)Q;and (d) Da sends cobalanced sequences in B(T')Q to balanced sequences in B(T)Q. PROOF. (a) Notice that a induces an anti-isomorphism a : Sr' ~ S", Sr the opposite of the poset of meet irreducible elements of T . Now, H : B(T')Q ~ rep( Sr', k) with H (G) = (QG , QG (r) : r E Sr) is an exact category equivalence by Proposition 7.1.3, and a: rep(Sr, k) ~ rep(Sr', k) is an exact contravariant category equivalence by Lemma 6.1.1.
7.1
Projectives, Injectives, and Exact Sequences
201
Define a correspondence F : B(T)Q -+ rep(ST, k) by F(G) = (QG, QG[T] : TEST). Then F is a functor, as a consequence of Lemma 3.1.9(b). Now, QG[T] = :E{QG(a) : a;fT} and QG(T) = n{QG[a] : atT} for each type T by Proposition 3.2.8. Since H is a category equivalence, a representation U in rep(ST, Q) is completely determined by (QG, QG(T): T E T). Hence, U is also completely determined by (QG , QG[T]: T E T). In fact, (QG, QG[T]: T E T) is completely determined by (QG , QG[T] : TEST), since QG[a n T] = QG[a] + QG[T] by Proposition 3.2.8(d). It now follows that F is an exact category equivalence, in particular, Hand F are manifestations of the same functor from B(T)Q to rep(T , Q). Define D cx by F-1a H: B(T')Q -+ B(T)Q . Then D cx is an exact contravariant equivalence. Since Hand F are manifestations of the same functor, D cx - 1 = H-1a F, whence D cx - 1Ds = (H-1a F)(F-1a H) = H- 1a 2 H is naturally equivalent to the identity functor. Assertions (b), (c), and (d) are left as exercises (Exercise 7.1.1). Corollary 7.1.6 Suppose Tis afinite lattice oftypes and G, H E B(T)Q. There are correspondences C+, C- : B(T)Q -+ B(T)Q such that: (a) C+ G = 0 if and only if G is projective in B(T)Q and C- G = 0 if and only if G is injective in B(T)Q. (b) If G has no injective summands and H has no projective summands, then QHom(G , C+ H) is isomorphic to QHom(C-G, H) . (c) IfG has no projective summands, then QEnd G is isomorphic toQEnd C+G , and if G has no injective summands, then QEnd G is isomorphic to QEnd C-G . (d) If G E B(T)Q is a strongly indecomposable group, then C+ G and C- G are strongly indecomposable groups. PROOF. Apply Proposition 7.1.3 and Corollary 6.2.5. Call G E B(T)Q preprojective if C+i G = 0 for some i ~ 1 and preinjective if C-iG = 0 for some i ~ 1. Let cdn G = {rank G, rank G(T): TEST}. Corollary 7.1.7 Let T be afinite lattice of types. Thefollowing statements are equivalent: (a) B(T)Q has finite representation type. (b) If G is a strongly indecomposable group in B(T)Q' then G is preinjective. (c) IfG is a strongly indecomposable group in B(T)Q, then G is preprojective. (d) The quasi-endomorphism ring of G is isomorphic to Q for each strongly indecomposable G E B(T)Q. In this case, if G is an strongly indecomposable group in B(T)Q, then G is determined up to isomorphism by cdn G.
202
7. Finite Rank Butler Groups
PROOF .
Theorem 6.2.6 and Proposition 7.1.3.
Corollary 7.1.8 Let T be a finite lattice oftypes. (a) If G E B(T)Q is a strongly indecomposable group in B(T)Q that is not projective, then there is an almost split sequence 0 --+ C+G --+ B --+ G --+
oin B(T)Q.
(b) IfG is a strongly indecomposablegroup in B(T)Q that is not injective, then there is an almost split sequence 0 --+ G --+ B --+ C-G --+ 0 in B(T)Q. PROOF .
Proposition 7.1.3 and Theorem 6.3.4
Open Question: Given a finite lattice T of types, is there a category X of abelian groupscontaining B(T)Qand a categoryequivalenceX --+ Qrep(Sr , Q) that restrictsto the categoryequivalence B(T) Q --+ rep(S, Q) of Theorem3.3.2 ? What is the situation for B(T)p and Qrep(Sr, Zp)?
EXERCISES
I. Complete the proof of Theorem 7.1.5.
2. Give explicit constructions for C+ G and C- G, G E B(T)Q' Give examples of preprojective groups that are not projective and preinjective groups that are not injective. 3. Given X a rank - I group in B(T k r X. (a) Compute C+ X and (b) Find an almost split sequence 0 -+ C+ X -+ B -+ X -+ O. (c) Find an almost split sequence 0 -+ X -+ B -+ C- X -+ O.
c:
4. For G E B(T)Q and M a subset of T , define G(M) to be the pure subgroup of G generated by {G(r): rEM} and G[M] = n{G[r] : rEM}. Let 0 -+ G -+ H -+ K -+ 0 be an almost split sequence in B(T)Q' Prove the following : (a) If G(M) =1= G, then 0 -+ G(M) -+ H(M) -+ K(M) -+ 0 is a split exact sequence in B(T)Q. (b) If G[M] =1= 0, then 0 -+ G/ G[M] -+ H/ H[M] -+ K/ K[M] -+ 0 is a split exact sequence in B(T)Q.
7.2
Endomorphism Rings
If R is a ring with additive group a reduced torsion-free abelian group, then R is the endomorphism ring of G for some reduced torsion-free abelian group G of finite rank by Theorem 2.4.6. However, the G constructed in that theorem is not, in general, a Butler group. Endomorphism rings of finite rank Butler groups are described in this section. Given a ring R, R+ denotes the additive group of R.
7.2 Endomorphism Rings
203
Proposition 7.2.1 If G is a finite rank Butler group, then (End G)+ is a finite rank Butler group. PROOF . By Corollary 3.2.4, there are completely decomposable groups C and
D such that G is a pure subgroup of D and an epimorphic image of C, say f: C -)0 G . Then 4> : End G -)0 Hom (C, D), defined by 4>(g) = gf, is a monomorphism with pure image. To see that 4> has pure image, suppose p is a prime, h E Hom (C, D) , and ph = gf for some g E End G. If x E C, then ph(x) = gf(x) E pD n G = pG . It follows that (g/p) E End G and h = is! p)f. But Hom(C, D) is a completely decomposable group, since C and D are completely decomposable groups. This shows that (End G)+ is a finite rank Butler group .
Proposition 7.2.2 Assume that R+ is a reduced torsion-free abelian group of finite rank. Then (a) R+ is a Butler group ifand only if(NR)+ is a Butler group and (R/NR)+ is
almost completely decomposable; and (b) if (R/NR)+ is a Butler group and QR/ JQR is a simple algebra, then (R/NR)+ is homogeneous completely decomposable. PROOF . (a) The ring R is a subring of QR, a finite-dimensional Q-algebra. Recall thatNR = R n JQR, JQR the Jacobson radical ofQR. First, assume that R+ is a Butler group . Then (R/NR)+ , a homomorphic image of R+, and (NR)+, a pure subgroup of R , are Butler groups by Corollary 3.2.4. As in Lemma 2.2.7, there is a nonzero integer n with nR* 5; R/NR 5; R* = R) x . .. x R m , each R; a finitely generated free S;-module, S; a Dedekind domain, is homogeneous, so that and QR; a simple Q-algebra. Since S; is a domain, Rt is homogeneous. But (R*)+, hence each Rt, is a Butler group, since (R*)+ is quasi-isomorphic to the Butler group (R/NR)+ . Furthermore, each (R;)+ is homogeneous coinpletely decomposable, because homogeneous Butler groups are completely decomposable by Corollary 3.2.7(a). This shows that (R/NR)+ is almost completely decomposable. Conversely, assume that (NR)+ is a Butler group and (R/NR)+ is almost completely decomposable. Then R+ is quasi-isomorphic to (NR)+ ED (R/NR)+ by Theorem 2.4.7. Thus , R+ is a Butler group by Corollary 3.2.4. (b) In this case , as in (a), (R/NR)+ is quasi-isomorphic to a homogeneous completely decomposable group, hence homogeneous completely decomposable.
st
Not every R with R+ a Butler group is the endomorphism ring of a Butler group. For example, if R is the ring of Hamiltonian quaternions over Zp, then R+ is a free Zp-module that is q-div isible for each prime q #- p, and QR is not a field, as required by (a) of the next corollary. Corollary 7.2.3 [Arnold, Vinsonhaler 89] Suppose QR is a division algebra and R is the endomorphism ring of a finite rank Butler group.
204
7. FiniteRank ButlerGroups
(a) If R+ is p-reducedfor at most 4 primes p of 7l. then QR is afield isomorphic
to Q[x]/ (g(x Y) for some irreducible g(x) E Q[x]. (b) If R+ is p-reducedfor at most 3 primes p of Z, then R is a subring of Q.
(a) Since R is a subring of a division algebra, NR = 0 and R+ is homogeneous completely decomposable. Hence, R+ is not p-reduced if and only if R+ is p-divisible. Thus, typeset R+ S; T4, the finite Boolean algebra of types generated by {type Zp : R is p-reduced}, since R+ is p-reduced for at most 4 primes , say PI,
PROOF.
P2, P3, P4·
Let G be a finite rank Butler group with End G = R. Then G is strongly indecomposable, since QR is a division algebra. Moreover, typeset G S; T4 • This is so because if q is a prime with q # Pi for each i , then q R = R, and so qG = G. In particular, G E B(T4 ) . By Theorem 3.3.2, B(T4 )Q is category equivalent to rep(S4, Q) . Moreover, each indecomposable in rep(S4, Q) has endomorphism ring Q[x]/(g(xY} for some irreducible g(x) E Q[x] by Example 6.2.7. Thus, QR = QEnd G must also be of this form . (b) In this case, just as in (a), if R = End G for some finite rank Butler group G, then G E B(T3) and G is strongly indecomposable. Thus, QR = QEnd G = Q by Example 3.3.3. Theorem 7.2.4 [Arnold, Vinsonhaler 87] Assume that R is a ring with additive group afinite rank Butler group. If R is p-reducedfor at leastfive primes p, then R is the endomorphism ring ofa finite rank Butler group. Let PI, . . . , PS be five primes for which R is p -reduced. For each 1 :::: i :::: 5, define Xi to be the subring of Q generated by {l/pjlj i= i} . By Example 1.1.7, there is
PROOF.
for some m x m Q-matrices A and B with QR = C(A , B) = End(U). Let H = (XIR+)m $ ... $ (XsR+)m, a finite rank Butler group, with QH = (QR)m $ . . . $ (QR)m. Define G = V n H, V the row space of the Q-matrix
Then G is a Butler group, being a pure subgroup of H . A routine matrix calculation shows that {(r, r): r E R} S; End G S; {(f, f) : f E C(A, B» = ({(r, r) :r E QR}. Notice that each XiR+ is a fully invariant subgroup of H, since R is Pireduced for each 1 :::: i :::: 5. It follows that if(r, r) E End G, then r E n{X i R : 1 :::: i :::: 5} = R. Hence, End G is isomorphic to R, as desired. Theorem 7.2.4 can be used to construct Butler groups with pathological direct sum decompositions just as for arbitrary torsion-free abelian groups of finite rank. For instance:
7.3 Bracket Groups
205
Example 7.2.5 (a) There is afinite rank Butlergroup G and a primep such that G is indecomposable at p but (End G)p is not a local ring. (b) If P ~ 5 is a prime, then there is a finite rank Butler group H with a nonunique decomposition into indecomposables in the isomorphism at p categoryoffinite rank Butler groups. (c) There is afinite rank stronglyindecomposable Butler group G with a subgroup H such that G is nearly isomorphic to H, H is a summandof G tB G, but G is not isomorphic to H. PROOF. (a) By Theorem 7.2.4, there is a finite rank Butler group G with End G = Z[i], the Gaussian integers, since Z[i] is a free abelian group of rank 2. The remainder of the proof is as in Example 2.1.5. (b) A consequence of Theorem 7.2.4 and Example 2.1.11, in which case R is also a ring with finitely generated free additive group. (c) By Theorem 7.2.4, there is a finite rank Butler group with End G = Z[(_5)1 /2] = R, since R+ is a free group of rank 2. Then R contains a nonprincipal ideal I with R/I finite, for example, I = (3, 1 + (_5)1 /2). Since R is a Dedekind domain, I and R are in the same genus class, as in Lemma 2.2.8. Hence, H = I G is a subgroup of G with G/ H finite and H nearly isomorphic to G by Theorem 2.4.3.
7.3 Bracket Groups This section consists of a brief introduction to a well-studied class of Butler groups. A bracketgroup, also called a B(I)-group [Fuchs, Metelli 91] or a Richman-Butler group [Hill, Megibben 93], is a group G = C/ X with C a finite rank completely decomposable group and X a pure rank-I subgroup of C. In this case, G is a finite rank Butler group with rank G = rank C - 1. Example 3.2.1 is an example of a bracket group, as is C+ Y for a rank-l group Y in B(T); Exercise 7.1.3. In the latter case, C+ Y = C/ X, where C is an injective group in B(T)Q and X is a rank-l group, as a consequence of Example 6.3.6 and Proposition 7.1.3. Bracket groups are intimately related to those almost completely decomposable groups containing a completely decomposable group of cyclic index. This relationship has yet to be investigated in any depth. Proposition 7.3.1 (a) If G = C/ X is a bracket group and n a nonzero integer, then H = C + (l / n)X is an almost completely decomposable group with H/ C a cyclic
group isomorphic to X/nX. (b) If G is an almost completely decomposable group and C is a completely decomposable subgroup of G with G/ C cyclic, then G is a bracketgroup. (c) If G = C/ X is a bracketgroup, then G is almost completely decomposable if and only if Tcr(C) has a uniqueminimal element.
206
7. FiniteRank ButlerGroups
PROOF . (a) The group H is almost completely decomposable. Moreover, H / C is isomorphic to nHInC = (nC + X)/ X = X/(X n nC) = X/nX, since X is a pure subgroup of C. (b) Let C = A I E9 . . . E9 An be a completely decomposable subgroup of G with rank Ai = I for each i and G = C + Zx. Write x = (I/m)(al , ... , an) with a; E Ai and m a nonzero integer. Then G = (C E9Zx)/ Y is a bracket group, where Y is the pure rank-I subgroup of G generated by (mx, -aI, .. . , -an)' (c) Exerc ise 7.3.2 .
Let A = (A I , , An) be an n-tuple of subgroups of iQl with I E Ai for each i E n+ = {I, 2, , n}. Define G[A] = (AI E9 . . . E9 An)/ X, where X is the pure rank-I subgroup of A I E9 . . . E9 An generated by (1, I, I , ... , I). For example, G[AI, A z] = Al + Az, since 0 -+ AI n Az -+ AI E9 A z -+ AI + A z -+ 0 is an exact sequence with I E A] n Az. Notice that type X = n{type Ai : I ~ i ~ n}, and if G[A] is an almost completely decomposable group, then G must be completely decomposable. Call A cotrimmed if each (Ai + X)/ X is a pure subgroup of G[A]. For example, G[A] , Az] is not cotrimmed unless AI = Az.
Lemma 7.3.2 Let G = G[A]for some A = (AI ," " An)' (a) Then A is cotrimmed if and only ifn{A j
(b)
= Ai +n{A j :j =/: G[A] = G[B].
If B,
i}, then B
:
j
=/: i} s:;; Ai for each i.
= (BI, . . . ,Bn) iscotrimmedwith
PROOF. Exercise 7.3.3.
Let A = (AI, . .. , An), E a subset of n+, and define rE = type n {A j : j E E} . Given a partition P = {P(1), . . . , P(m)} of n+, let r(P) = n{rp(i) U rp(j) : I ~ i < j ~ m}. For a type r , two elements i and j of n+ are t -equivalent if either i = j or there is a sequence i(1), ... , i(n) E n+ with i(1) = i, i(n) = j, and r 1:. type Ai(k) + Ai(k+l) for each I ~ k ~ m - 1. It can be readily continned that r-equivalence is an equivalence relation on n+ For instance, if r = type Ai for some i , then {i} is a r -equivalence class of n+, since type Ai ~ type Ai + A j for each j E n" .
Lemma 7.3.3 Let G be a bracket group. Then: (a) G is quasi-isomorphic to G[A] E9 D for some completely decomposable group D and cotrimmed A = (AI, . . . , An); (b) typeset G[A] = {r(P): P partition ofn+}; (c) G[A] is quasi-isomorphic to G[B(1)] E9 . .. E9 G[B(m)], where each B(i)
is a subtuple of Ai and (d) ifG = G[A] and r E typeset G, then G(r) is quasi-isomorphic to G[B], where B = (B I, . . . , Br ) , B, is a rank-Y group with type B, = rpi for a
7.3 Bracket Groups partition Pi
= {E(i) , E(i)C},
207
and E(1), ... , E(r) are the t-equivalence
classes of n" , In particular; rank G(r)
=r
-
1.
= C I Y, where C = XI EEl .. . EEl X; with rank Xi = I for each i and Y is a pure rank-I subgroup of G. It is sufficient to assume that o =1= Y = (XI, . . . , x n) E y with each Xi =1= O. Let Ai be a subgroup of Q containing I and Ii : Xi ~ Ai an isomorphism. Then f = fl EEl .. . EEl fn : C ~ C' = A I EEl . .. EEl An is an isomorphism inducing an isomorphism f : G ~ C' I f(y) · If 0 =1= y = (ql , .. . qn) E f(y), and X is the pure subgroup of C' generated by (1, I, . .. , I), then each qi =1= 0 and (ql-I, . .. , q; -I)f : G ~ G[A] is a quasi-isomorphism. By Lemma 7 .3.2, A may be assumed to be cotrimmed. (b) Let n : C A ~ G[A] be the canonical epimorphism with C A = Al EEl . .. EEl An and ker n = X = «I, .. . , I)) • . Recall from Section 3.2 that cosupport G[A] = {cosupportf: 0 =1= f E Hom(G[A], Q)}, where cosupport f = {i : frr(A i) = O}. It follows that the maximal elements of cosupport G[A] are the subsets I of n+ with III = n - 2. For each such I write n+\J = {i, There is an epimorphism !J : G[A] ~ G[Ai , Aj] = Ai + A j ~ Q induced by the projection of C A onto Ai EEl Aj. In particular, !J«al , '" ,an) + X) ai - aj ' There is a cobalanced monomorphism ¢: G[A] ~ EEl{!J(G): I E 5} EEl{A i + A j : I ::; i < j ::; n} given by ¢«al,'" ,an) + X) EEl{ai - aj : I::; i < j ::; n} (Lemma 3.2.3(b» . In particular, since the image of ¢ is pure, type (al, '" ,an) + X n{typeai -a j : I::; i < n} . Let 0 =1= y = (aI , . . . , an) + X with a, E Ai. There is an equivalence relation on n+ defined by i y -equivalentto j if a, = ai - Let P = {P(1), . .. , P(m)} be the partition of n+ consisting of y-equivalence classes. It follows from the definition of ¢ that type y type ¢(y) rep), as desired. (c) The proof is an induction on n. Let r = type A I and E(1) = {I}, E(2), .. . , E(r) be the r -equivalence classes of n+ . It suffices, by relabeling, to assume that E(2) = {2, 3, . .. , mI. Define B = (AI, A z, · · · , Am)' Now, G[B] is a summand of G[A]. To see this, write G[A] = CAl X with C A = Al EEl··· EEl An' X = «I , I, . .. , I))., and G[B] = CnlY with C n = AI EEl· ·· EEl Am and Y = «(1 , I, . . . , I)) • . A projection f: C A ~ Ce induces an onto homomorphism f : G[A] ~ G[B] . Next, define g : G[B] ~ G[A] by g«al, .. . ,am) + Y) = (aI, ... , am, aJ, ... , at) + X. Since f g is the identity on G[B], G[B] is a quasi-summand of G[A], provided that g is a quasi-homomorphism. To this end, let X E G[B]. By (b), type X = rep), where P = {P(1), ... , P(t)} is the partition of m+ given by the x-equivalence classes. The g(x)-equivalence classes in n+ are P' = {P '(l) , P(2) , . .. , pen)}, where P'(l) is the union of P(l) and {i: m + I ::; i ::; n} . If I ::; i ::; m and m + I ::; j ::; n, then type AI ::; type Ai + A j , since r = type A I and Ai and A j are in different r -equivalence classes. It follows that type x = rep) ::; type g(x) = reP '). Consequently, g is a quasi-homomorphism from G[B] to G[A] by Theorem 3.3.2 . In particular, G[A] is quasi-isomorphic to G[B] EEl ker f . However, ker f = G[C], where C = (Am+J, . . . , An)' An induction on n completes the proof. PROOF. (a) Write G
n.
=
ts
=
=
=
= =
208
7. Finite Rank Butler Groups
(d) Given Bi , define f; : B, -+ G[A] by f;(1) = (aJ, ... , an) + X, where = 0 if j ~ E(i) and aj = 1 if j E E(i). Then B, -+ G[A](r) is a quasitype B, :::: r . homomorphism as a consequence of (b), observing that type fi(1) Now, f fl EB .. • EB f,: BI EB ... EB B, -+ G[A](r), and f(l, 1, ... ,1) O. Hence, f induces f: G[B] -+ G[A](r). If y E G[A](r), then type y :::: r. It follows from (b) that y E image f, so that f is a quasi-epimorphism. Finally, image f has rank r - 1, since E(1), ... , E(r) is a partition of n+.
t. .
aj
=
=
=
Theorem 7.3.4 Let T be afinite lattice oftypes and G = G[A]for some n-tuple A = (AI , . . . , An) ofsubgroups ofQ. Thefollowing statements are equivalent: (a) G is strongly indecomposable; (b) rank G( r) 1 for each r typeA i ; (c) QEnd G Q.
= =
=
PROOF. (a) => (b) Ifn = 2, then rank G = 1, and so rank G(r) = 1 for each r = type Aj.Now suppose n :::: 3, r type Ai, and rank G(r) > l.ByLemma7.3.3(d), {r}, E(2), andE(3). there are at least three r-equivalence classes of n+, E(l) In view of Lemma 7.3 .3(c) , G[B] is a proper quasi-summand of G, where B (A }: i E E(1) U E(2». This is a contradiction to the assumption that G is strongly indecomposable. type Ai. Since (b) => (c) If f E End G, then f(G(ri)) ~ G(ri) for each ri rank G(ri) = 1, f: G(ri) -+ G(ri) is multiplication by some qi E Q . However, if 0:1= x = (al ,'" , an) E X , thenqlal + .. ·+qnan = f(x+X ) = f(O ) = 0 E G, and so qlal ED · · · ED qnQn = q(Q, ED··· ED an) E C for some q E Q . Thus, q = qi for each i and f = q E Q , as desired. (c) => (a) is a consequence of Proposition 2.1.1(a), since Q End G = Q has no nontrivial idempotents.
=
=
=
=
If G[A] and G[B] are strongly indecomposable, then they are isomorphic if and only if they are nearly isomorphic. This is a consequence of Theorem 2.4.3, since End G[A] is a subring of Q by Theorem 7.3.4, hence a principal ideal domain. If A (AI,"" An) and r E typeset G[A] with rank G(r) 1, then, by Lemma 7.3.3(d) and the distributivity ofthe lattice of all types, there is a subset E of n+ with r = type Dr, where Dr = n{A i: i E E} + n{A j : i ~ E}. Define 8(A) = (Dr: rank G(r) 1).
=
=
=
Theorem 7.3.5 Assume that T is afinite lattice oftypes, A = (A I, . . . , An) and B = (B I , •• • , Bm ) are tuples ofsubgroups ofQ, and G = G[A] and H = G[B] are strongly indecomposable groups in B(T). Then
=
(a) G and H are quasi-isomorphic if and only if rank E{G(r) :r E M} rank E {G (r) : rEM} for each nonempty subset M of T ; and (b) G and H are isomorphic if and only if they are quasi-isomorphic and there is 0:1= q E Q with q8(A) = 8(B).
7.3
Bracket Groups
209
PROOF. See [Arnold, Vinsonhaler 93] and references therein.
Theorem 7.3.6 [Goeters, Ullery, Vinsonhaler 94] Assume that Tis a.finite latticeoftypes, A = (A I,"" An) and B = (B I , .. . , Bm)are tuples ofsubgroups of Q, and G = G[A] and H = G[B] are strongly indecomposable groups in B(T). Then G and H are quasi-isomorphic ifandonly ifrank G(r) = rank H(r) for each rET.
EXERCISES
1. [Goeters, Ullery 95] Let A = (AI, ... , An) for n ~ 3. (a) Prove that G[A] is strongly indecomposable if and only if for each i E n+ and n rF. partition {{i}, E, F} of n'". type A; (b) Prove that iffor each 3-.element subset {i, i . k} of n +, type A; + A j, type A j + Ak, and type A j + Ak are pairwise incomparable, then G[A] is strongly indecomposable.
irE
2. [Fuchs, Metelli 91] Prove Proposition 7.3.1(c). 3. Prove Lemma 7.3.2. 4. Prove that if A = (A I , .. . , An) is an n-tuple of subgroups of Q, G = G[A] , T is a finite lattice of types, and D : B(T)Q -+ B(TOP)Q is the duality given in Theorem 7.1.5, then D (G[A]) = G(A ), the kernel of the homomorph ism Al E9 . . . E9 An -+ Q given by inclusion of the A;'s in Q.
5. State and prove the duals of the results of Section 7.3 for groups of the form G(A ), as defined in the previous exercise.
NOTES ON CHAPTER 7
Projectives and injectives in B(T) are characterized in [Butler 68] and [Lady 79]. Coxeter correspondences and almost split sequences for B(T)Q have yet to find significant application. An alternative development of Coxeter correspondences is given in [Lady 83] for Butler modules over Dedekind domains . There are scattered results on endomorphism rings of Butler groups of infinite rank; see [Arnold, Dugas 96], [Dugas, Gabel 97], and references therein. An interesting class of finite rank Butler groups, called Kravchenko groups and defined in terms of balanced projective resolution s, are investigated in [Nonxga, Vinsonhaler 96]. These classes of groups form a descending chain of subclasses of the class of finite rank Butler groups with intersection the class of finite rank completely decomposable groups . Representation analogues of Kravchenko groups are investigated in [Nonxga, Vinsonhaler 97]. There is an extensive body of literature on bracket groups, beginning with [Richman 83]. Although this is a restricted class of groups , it is rare in the theory of torsion-free abelian groups of finite rank to be able to classify groups by numerical invariants. A survey of this literature up to 1991, including a discussion of invariants and classification, can be found in [Arnold, Vinsonhaler 93]. More recent work includes [Hafting 93],
210
7. Finite Rank Butler Groups
[Goeters, Ullery 91, 92, 94], [Goeters, Ullery, Vinsonhaler 94], [DeVivo, Metelli 96, 99], and [Yom 94, 98]. Cyclic regulating quotient groups, those almost completely decompos able groups G containing a regulating subgroup B with G/ B cyclic, have been characterized , as noted in Notes on Chapter 5. In view of Proposition 7.3.1, cyclic regulating quotient groups are related to bracket groups. Implications of these characterizations of cyclic regulating quotient groups for the corresponding bracket groups , given in Proposition 7.3.1(a), remain unexplored.
8 Applications of Representations and Butler Groups
8.1 Torsion-Free Modules over Discrete Valuation Rings The theory of Butler groups has, a priori, little relevance for p-Iocal torsion-free groups. This is so because p-Iocal Butler groups are completely decomposable. Nevertheless, isomorphism at p classes of finite rank Butler groups can be interpreted as torsion-free Zp-modules of finite rank (Corollary 8.1.9). Rank-l groups correspond to purely indecomposable p-Iocal groups , and bracket groups correspond to copurely indecomposable p-Iocal groups . Let R be a discrete valuation ring with prime p and quotient field Q, R* the completion of R in the p-adic topology, and k = Rj pR a field. Then R* is a discrete valuation ring with prime p and R*jpR* = RjpR [Fuchs 70, 73]. For example, if R = Zp, then R* is the ring of p-adic number s with quotient field Q*, the field of p-adic rational numbers, and k = Z] pZ. A torsion-free R-module M is divisible if pM = M and reduced if M has no nonzero divisible submodules. There is a unique divisible submodule d(M) of M with M = d(M) ED Nand N a reduced module . The p-rank of M is the Rj pRdimension of M j pM, and M* is the completion of M in the p-adic topology. A pure submodule B of M is a basic submodule of M if B is a finitely generated free R-module with M / B divisible. Fundamental properties of completions [Fuchs 73] are listed in the following lemma. Lemma 8.1.1
If M is a reduced torsion-free R-module offinite rank, then:
(a) M is a pure R-submodule of M*, M* is a free R*-module with R*-rank M* = p-rank M, and M*j M is a divisible R-module;
212
8. Applications of Representations andButler Groups
(b) each R-endomorphism of M extends uniquely to an R*-endomorphism of
M*; (c) if B is a basic submodule of M, then B* = M*; and (d) if N is a reduced torsion-free R-module of finite rank, then each f HomR(M, N) extends to a unique R*-homomorphism f* : M* -+ N*.
E
PROOF. Exercise 8.1.1.
Lemma 8.1.2 Suppose M is a torsion-free R-module offinite rank. (a) If M is not divisible, then M has a basic submodule B with rank B = prank M s rank M. (b) If p-rank M = n and rank M = n + k, then M is free if and only if k = 0 and divisible if and only if n = O. (c) If 0 -+ K -+ M -+ N -+ Oisanexactsequenceoftorsion-free R-modules, then p-rank M = p-rank K + p-rank N. PROOF. (a) Notice that M/ pM is a nonzero finite-dimensional k-vector space. Let {x! + pM, . . . ,Xn+ pM} be a k-basis of M/ pM and define B = Rx, E9 ... E9 Rx n, a free submodule of M. Then M/ B is divisible as B + pM = M. To see that B is pure in M, let x E M and px = rlx! + ... + rnxn. Since the (Xj + pM)'s are k-linearly independent, p divides r, for each i . Thus, x E B, as desired. (b) Observe that M is free if and only if B = M, since if M is free, then M/ B is both free and divisible, hence zero. Thus, by (a), M is free if and only if p-rank M = rank M. Apply (a) to see that M is divisible if and only if M does not have a basic submodule, equivalently, p-rank M = O. Statement (c)follows from thefactthat 0 -+ K/ p K -+ M/ pM -+ N/ p N -+ 0 must be an exact sequence, since the image of K is a pure submodule of M. Let M be a finite rank torsion-free R-module. If M is reduced with p-rank M = I, then M is apurely indecomposable module. In this case, M* is isomorphic to R*, and so M is isomorphic to a pure submodule of R*. If p-rank M = I, then M = d(M) E9 N with N a purely indecomposable module. For an example of a purely indecomposable module, choose t E R*\R and let R, be the pure submodule of R* generated by {I, t }. Then R is a basic submodule of R R; = R*, p-rank R, = 1, and rank R, = 2. " Purely indecomposable R-modules have properties reminiscent of torsion-free abelian groups of rank 1 (Lemma 3.1.1):
Proposition 8.1.3 Suppose M and N are twopurelyindecomposable R-modules. (a) Thefollowing statements are equivalent:
(i) M and N are isomorphic. (ii) M is quasi-isomorphic to N. (iii) HomR(M, N) f. 0 and HomR(N, M) f. O. (iv) HomR(M, N) f. 0 and rank M = rank N.
8.1 Torsion-FreeModules over Discrete ValuationRings
213
(b) The endomorphism ring of M is a discrete valuation ring isomorphic to a
pure subring of R*. M and N are pure submodules of R*. The implications (i) =} (ii) =} (iii) are routine. (iii) =} (iv) Let 0 -:f= f E HomR(M, N) . Then f extends to an R*-endomorphism f* of R*, and so f is multiplication by some a E R*. Hence, f is a monomorphism, and rank M = rank f(M) ::: rank N . Similarly, rank N ::: rank M , since HomR(N, M) -:f= O. (iv) =} (i) Let 0 -:f= f E HomR(M, N). Since rank M = rank Nand f is a monomorphism, N /f(M) is bounded, say rN 5; f(M) with 0 -:f= r E R. Write r = pi u with p-height u = O. Then u is a unit of R, and pi N 5; f(M) 5; N. Since p -rank N = 1, N / piN is a cyclic R-module. It now follows that f(M) = pj N for some j , whence M and N are isomorphic. (b) Exercise 8.1.1.
PROOF. (a) It suffices to assume that
The next proposition gives analogues of the pure fully invariant subgroups G( r), for G a torsion-free abelian group and r a type. Here, an element t of R* plays the role of a type . Let M be a torsion-free module of finite rank with a basic submodule B and assume B 5; M 5; M*, a free R*-rnodule with an R -basis of B as an R*-basis of M*. Let t E R* and define M(t) = {m EM :tm EM}. Notice that if t E R , then M(t) = M .
Proposition 8.1.4 Assume that M and N are reduced torsion-free R-modules offinite rank and t E R*. Then: (a) M(t) is a pure submodule ofM , and iff : M --+ N is an Rshomomorphism, then f(M(t)) 5; N(t); (b) RrM(t) = Hom(R r, M)R r; (c) M(t) n M(s) 5; M(t + s); and (d) ifs E R*, then M(t)(s) = M(st) n M(t).
m E M and pm E M(t) . Then ptm E pM* n M = pM, so that tm E M. Thus, m E M(t) and M(t) is pure in M . The R-homomorphism f : M --+ N extends to an R*-hornomorphism f* : M* --+ N*. Hence, if m E M with tm E M , then f(tm) = f*(tm) = tf*(m) = tf(m) E N. This shows that f(M(t)) 5; N(t). (b) If r E R , then M(t) = M, R, = R, and RM = Hom(R, M)R. Now assume t E R*\R and let m E M(t). Then Rm 6' Rtm S; M , noting that Rm n Rtm = 0, sincet rf. R.Hence,Rrm 5; M,sinceR r = (R6'Rt)*andMispureinM* .Define f: R, --+ M by fey) = ym, an R-homomorphism with m = f(1) E image f. Thus, M(t) 5; Hom(R r, M)R r, and so RrM(t) S; Hom(R r, M)R r. Conversely, let 0 -:f= f : R, --+ M. Then f is a monomorphism, since R, is purely indecomposable. Hence, image f = Rsm 5; M for m = f(l) E M. In particular, m E M(t), and so image f 5; RrM(t), as desired.
PROOF. (a) Let
214
8. Applications of Representations and ButlerGroups
(c) If m EM and tm, sm are in M, then (t + s)m is in M. (d) Let m E M(t)(s). Then m E M(t) with sm E M(t), and so tsm E M. This shows that M(t)(s) s:; M(st) n M(t). Conversely, if m E M(st) n M(t), then st m E M and tm EM. Hence, m E M(t)(s). Let TF R denote the category of reduced torsion-free R-modules of finite rank . As a consequence of the next two theorems, there is a variety of embeddings of rep(Sn, R, j) into TF R • Theorem 8.1.5 Let A = {tJ, . . . , tn} be a subset of R*\R such that {I, tj , t;tj: I ::s j ::s n} is a Q-independent set for each i, (a) Foreach integer j 2: 0, there is afully faithfulfunctor FA : rep(Sn, R, j) -+ TF R • (b) The image of FA consists ofall those M E TF R such that (i) M has a basic submodule B with Mj(B + R" M(t,) + ... + R,.M(tn)) torsion; (ii) B is afully invariant submodule of M; and (iii) p! B s:; M(tl) + ...+ M(tn) s:; B. PROOF. Let V = (V o, V; : i E Sn) E rep(Sn, R , j) and define M = FA(V) to be the pure submodule of Va generated by V o + Uit, + ... + Vntn. Given a representation morphism f: V -+ V, define FA(f) = [": FA(V) -+ FA(V), noticing that !*(u;t;) = f(u; )t; E Viti for each i. Then FA is a faithful functor. Observe that Vo is a basic submodule of M , since Vo is a free R-module, Vo is a pure submodule of Va' hence of M, and M j U» is divisible , since M is a pure submodule of Va and j Vo is divisible. Morever, M (t;) = V; for each i, To see this, first observe that V; is contained in M(t;) . Conversely, let m E M(t;). Then m = rouo + rltlu, + + rntnun for some r, E Q and U; E V;. Moreover, tim = rot;uo + rlt;t,ul + + rnt;tnun = qovo + q,vltl + ...+ qnvntn E M for some qi E Q, V; E V;. Fix a basis of Vo, express each U; and V; as a linear combination of the elements of this basis , equate coefficients, and use the hypothesis that {I. tj. t;tj : I ::s j ::s n} is Q-independent for each i to see that m E V;. Now, M j(Vo + RtlM(td+' " + R,kM(tn)) is torsion, since Viti s:; R,;M(t;) = R,;Vi. Moreover, pjvo s:; M(td + ... + M(tn) s:; Vo, since V; = M(t;) and V E rep(Sn. R, j) . Each M(t;) is fully invariant in M by Proposition 8.1.4(a) . Thus, if f E End M , then pj f( V o) s:; Vo n pj M = pj V o, since Vo is a pure submodule of M . This shows that Vo is a fully invariant basic submodule of M. To see that F is a full functor, let g : FA(V) -+ FA(V) and f the restriction of g to Vo, a basic submodule of FA(V). Now, f: Vo -+ Vo with !(V;) = !(FA(V)(t;)) s:; FA(V)(t;) = V; for each i. Thus, f : V -+ V is a representation morphism with FA(f) = g, as desired. Finally, let M be in TF R satisfying the above conditions (i)-(iii). Then V = (B, M(t;) : i E Sn) E rep(Sn, R, j), since M(t;) is a pure submodule, hence a
o;
8.1 Torsion-FreeModules over Discrete Valuation Rings
summand, of the finitely generated free R-module Band p! B S; M(td M (tn) S; B. Moreover,
215
+ .. . +
As a consequence of (b) of the next corollary, FA sends rank-I representations to purely indecomposable modules. In view of (a), not all modules in TF R are in the image of FA . For example, if p-rank M = n ~ 2 and rank M = n + 1, then M is not in the image of FA. Corollary8.1.6 Let FA: rep(Sn, R, j)
~
TF R be thefunctorofTheorem8.1.5.
If FA(V) = M, with V = (Vo, U, : i
E Sn), then p-rank M = rank Vo and rank M = rank Vo + rank VI + ...+ rank U; ~ 2(p-rank M). (b) If V = (Vo, Ui : i E Sn) with rank Vo = I, then M = FA(V) is purely indecomposable and End M = R.
(a)
PROOF. Assertion (a) followsfrom the definitionof FA (V), since Vo+ Viti +...+ Vktk = Vo $ Viti $ . .. $ Vktk by the independencecondition on A and p-rank M = rank Vo S; rank VI + ...+ rank Uv, Statement(b) is a consequenceof (a) and the fact that R = End V is isomorphic
to End FA(V), since FA is a fully faithful functor.
Recall from Section 4.2 that !!..(Sn, R) is the full subcategoryof rep(Sn+l, R , 0) with objects V = (Vo, VI, .. . , Vn, V*) such that Vo = VI $ . . . $ Vn, V* is a pure , tn) to be submodule of Yo, and V* n Vj = 0 for each 1 :5 i :5 n. Define V*(tl , the set of elements (tl VI , . . . , tnvn) E V; = $ .. . $ Vn*, with (VI , , vn) E V* S; Vo = VI $ . . . $ Vk . Let M E TF R with basic submodule B = B I $ $ Bn. Then M is a pure submoduleof B* = Bj $ .. . $ B~ . Given (t) = {tl' , tn}, a subset of R*, define M(I) = {(VI , .. . , Vk) EM: (tl vI , . . . , tkVk) EM} . There is another uncountable family of embeddings ofrep(S, R, 0) into TF R •
vt
Theorem 8.1.7 Let A = {tl, " " tn} be a subset of R*\R such that {I, tj , tjtj : 1 :5 j :5 n} is a Q-independent set for each i. There is a faithful embedding G A :rep(Sn, R, 0)
~
TF R
that is full for all those V E rep(Sn , R, 0) with no rank-Y summands. The image of G A consists ofall those M such that:
»
+ M(1)(tI ' . . . , tn is torsion, where B = B I $ .. . $ Bn is a fully invariant basic submodule of M with each B, fully invariant in M,. and (ii) M(1) is contained in B. (i) M I(B
PROOF. In view of Lemma 4.2.3(b), it suffices to define G A: !!..(Sn , R) ~ TF R • = (Vo, Vj, V* : i E Sn) E !!..(Sn , R), recalling that Vo = VI $ . . . $ Vn
Let V
216
8. Applications of Representations and ButlerGroups
«», o
and V* is a summand of Vo. Define M = G A(V) = (Vo + V*(tl.·.· , ~ Va. Given a representation morphism f : V ~ V' , extend f: Vo ~ V to an R*-homomorphism f* : Vo* --+ VQ* , and let G A (f) : G A(V) ~ G A (V') be the restriction of f* to G A (V). It follows that G A is a faithful functor. Now, Vo is a basic submodule of M . since Vo is free, Vo is pure in Vo*' hence in M, and M/ Vo is divisible, since M is pure in Va and Va/ Vo is divisible. Moreover, M(t) ~ B. as a consequence of the independence conditions on A and the fact that
Va =
vt $
... $ Vn* ·
To see that G A is full, assume that V and V' have no proper rank-l summands and let g : M = G A(V) --+ M' = G A(V') be an R-homomorphism. Extend g to Then g : M(t ) ~ M(t): an R*-homomorphism g* : M* = Va ~ (M ')* = (V with M(t) = V*. Since V has no proper rank-l summands, each V;/rri(V*) must be torsion; otherwise. there is a cyclic summand of Vi that gives rise to a representation summand of V . It now follows that g : Vi ~ V;' for each i, since g : V* ~ V~. Now. Vo = VI $ .. . $ Vb so letting f = g gives a representation morphism V ~ V' with G A (f) = g. as desired. Finally, let M be as described in the statement of the theorem. Then V = (B, Bit M(t) : i E Sn) E /::}.(Sn, R). noting that M(t) is pure in Band M, n B, = 0 for each i , Hence. G A(V) = (B + M(t )(tl, .. . , tn))* = M. as desired.
o)*·
A torsion-free R-module of finite rank with no free summands is copurely indecomposable if rank M = p-rank M + 1. Observe that if M is a finite rank torsion-free module with rank M = p-rank M + 1, then M = L $ F, where F is a free R-module and L is copurely indecomposable.
Corollary8.1.8 LetG A :rep(Sn, R .O) ~ TF R be thef unctor of Theorem 8.1.7. = G A (U), then p-rank M = rank UI + ... + rank U; and rank M = 2 (p-rank M) - rank Uo ~ 2(p-rank M). (b) LetU = (Uo.Ui:i E Sn) E rep(Sn,R.O) with n :::: 2, Uo = R n- I, UI = R $ 0 $ . . . $ 0, . . . , Un-I = 0 $ 0 $ .. . $ R, and Un = R(l, 1•...• 1). Then M = G A (U) is copurely indecomposable with End M = R.
(a)
If M
Assertion (a) is a consequence of the definition of GA. (b) Apply the functor of Lemma 4.2.3(b) to see that U ~ V, where V = (R" = R$ · ··$R. R$O$ · ··$0, . . . • 0$ ·· ·$R, R(I, ...• 1)) E ~(Sn, R). By (a), M = GA(V) = tR", R(tl, . . . , tn))* has p-rankn, rankn + 1, and endomorphism ring R. In particular, M is copurely indecomposable. PROOF.
Corollary 8.1.9 Let T be a finite lattice of types with Sr = Sn and A = {tl' ... , td a subset of 7l; \71 p such that {I, tj . tits- : 1 ~ j ~ k} is a Q-independent set for each i. (a) There is a fully faithful embedding FA : B(T. j)p
FA(G) = (Gp + G(rl)ptl
~
TF given by
+ ...+ G(rk)ptk)* 5; G; .
8.1
Torsion-Free Modules over Discrete Valuation Rings
217
(b) There is a faithful embedding G A : B(T, 0) p ~ TF that is full for groups H with no proper rank-I summands given by GA(H)
= (H(rt>p E9 ~ H(rt>;
E9
E9 H(rk)p
+ H*p(t) , ... , tk»)*
E9 H(rk);,
where H* is the kernel of H(rl) E9 . . . E9 H(rd
~
H.
PROOF. A consequence of Corollary 4.3.2, Theorem 8.1.7, and Corollary 8.1.8. Indecomposable modules in the image of FA and G A have wild modulo p representation type if IAI ~ 3, since rep(S3, R, 0) has wild modulo p representation type by Example 4.2.4. Following is a complete list of indecomposable modules in the image of FA and the image of G A for the case that IA I = 2. This list arises from a complete list of indecomposables in rep(Sz, Zp, j); Example 4.1.2.
= {r), rz} = Si and A = {t), tz} a subset ofZ; \Zp such that {I, tl, tztI. ti, tll is a Q-independent set for each i. Corollary 8.1.10 Let T be afinite lattice of types with Sr
(a) IfG to: (i) (ii) (iii) (iv)
E
B(T, j) and M
= FA(G) is indecomposable, then M is isomorphic
tl p if M has p-rank I and rank I ; (71 p, 7l pt))* or (71 p, 7lptz)* if M has p-rank 1 and rank 2; (Zp , Zpt), 7lptz)* ~ Z; if M has p -rank 1 and rank 3; or (B, Zpt) E9 0, 0 E9 Zptz)* ~ B* if M has p-rank 2 and rank 4, where E9 + 7lp (l , 1)/ pj 5; Q EEl Q . B =
z, z,
(b) If H E B(T, 0) and M = G A(H) is indecomposable, then M is isomorphic to: (i) tl p if M has p-rank I and rank 1; or (ii) (71 p E9 + 7lp(t) , tz»)* ~ 7l; E9 7l; if M has p-rank 2 and rank 3.
z,
PROOF. Exercise 8.1.2. The next corollary demonstrates a relationship between rank-1 groups in B(T) and purely indecomposable p-local groups and between bracket groups and copurely indecomposable p-local groups .
Corollary 8.1.11 Assume that T is a finite lattice of types with Sr = S; = {r), .. . ,rn } . (a) IfG E B(T, j) has rank 1, then FA(G) is purely indecomposable. = G[X), . . . , X n ] isa bracket group, where each Xi isa rank-I group with type Xi = Ti, then G A (H) is copurely indecomposable.
(b) If H
218
8. Applications of Representations and Butler Groups
PROOF. Exercise 8.1.2.
OpenQuestion: Canthe resultsofthissectionbe extendedtofunctorial embeddings ofrep(5, R , j)for an arbitraryfinite poset 5 ? Can each reduced p-local torsion-freegroup offinite rank be so represented? EXERCISES I. Prove Lemma 8. I.l and Proposition 8.1.3(b).
2. Confirm Corollaries 8.1.10 and 8.1.11. 3. (a) Show thatif M is a purelyindecomposable R-moduleof rank2, then M is isomorphic to M, for some t E R*. (b) Compute the endomorphism ring of Mr . (c) [Richman 68] Find necessary and sufficient conditions in terms of sand t for M, and M, to be isomorphic. 4. [Arnold 72] Let M be a copurely indecomposable module with p-rank n and rank n + I. (a) Prove that QEnd M is a subfield of the p-adic completion of the quotient field Q of R. (b) Give an example to demonstrate that quasi-isomorphic copurely indecomposable modules need not be isomorphic. Why must such an example have rank greater than or equal to 3? (c) Prove that if N is another copurely indecomposable module, then M is quasiisomorphic to N if and only if the nth exterior power of M is isomorphic to the nth exterior power of N .
8.2 Finite Valuated Groups There are correspondences between finite valuated p -groups and representations of finite po sets over the field Z/ pZ that can be used to determine the representation type of some categories of finite valuated groups. These correspondences also give rise to correspondences between indecomposable finite rank Butler groups and indecomposable finite valuated p-groups with torsion-free groups of rank 1 corresponding to cyclic valuated p-groups. Let p be a prime. An abelian p-group G is a valuated p-group if there is a mapping v from G to the ordinals together with 00 such that v(O) 00 , v(px) > v(x) , v(rx) = v(x) if r is an integer prime to p , and v(x + y) ::: min{v(x) , v(y)} for each x , y E G . Here, it is agreed that 00 > 00. As an example of a valuated p-group, let G be a subgroup of a p-group Hand define v(x) to be the p -height of x in H for each x E G. Then G is a valuated p-group, as a consequence of properties of the p-height function on H . In fact, this is the only example, since if G is a valuated p-group with valuation v, there is a p-group H such that G is a subgroup of Hand v(x) is the p-height of x in H for each x E G [Richman, Walker 79].
=
8.2 Finite Valuated Groups
219
The class of finite valuated p-groups forms a category. Morphisms, called valuated homomorphisms, are group homomorphisms f: G -4 H with v(x) ::: v(f(x» for each x E G. This category is an additive category; the direct sum of G and H is the group direct sum G EB H with v(x, y ) = min{ v(x) , v( y)} for x E G, y E H [Richman, Walker 79]. As will be seen , indecomposable valuated groups need not be indecomposable as groups. Let G be a finite valuated p-group and x E G. The set T(x , G) = {y E G : p" y = x , some n ~ O} is a finite poset, where y ::: y' if y = r" y' for some m ~ O. Then T(x , G) is a tree with unique minimal element x , called a root of T . If y , yl E T(x, G) with y < v', then V(yl) < v(y). Hence, T(x , G) is an example of afinite valuated tree, a rooted tree T with each vertex y of T labeled by v(y), an ordinal, or 00, such that if y , y' E T with y < v', then v( y') < v(y). Valuated trees play the same role for valuated p-groups as height sequences do for torsion-free groups. The point is that divisibility by integers is unique for a torsion-free group G, so that there is no need to record those y for which ny = x . However, divisibility by p is not unique for p-groups. Thus, vertices of the finite valuated tree T(x, G) record the elements y of G for which p" y = x ; in addition, these vertices are labeled by their values . For finite valuated trees Sand T, define S ::: T if there is a function f : S -4 T such that if y < y' in S, then f( y) < f( y') in T and v( y) ::: v(f (y)) for each y E S. Two finite valuated trees Sand T are equivalent if S ::: T and T ::: S. If S is a finite valuated tree and y E S, then length y is the length of a chain of elements from the root of S to s.
Lemma 8.2.1 The set of equivalence classes [T] offinite valuated trees forms a lattice. Specifically: (a) [S] ::: [T] if there is an SI equivalent to S and a T ' equivalent to T with SI s T ' . (b) [S] n [T) = [U], where U = {(s, t) E S x T: length(s) = length(t)} isafinite valuated tree with root U = (root S, root T) and v(s, t) = min{ v(s), v(t)}. (c) [S] U [T) = [V] , where V, the disjoint union of Sand T with the roots of S and T identified, is afinite valuated tree with v(root V) = max{v(root S) , v(root T)} and the values ofall other elements unchanged. PROOF. Exercise
8.2.1.
If x is an element of a finite valuated group G, then u-height x is the equivalence class of the valuated tree T(x, G). For an equivalence class [T] of valuated trees, define G(T) = {x E G : v-height x ::: [T]}.
Lemma 8.2.2 Suppose G and Hare finite valuated p-groups. (a) If f: G -4 H is a valuated homomorphism, then f(G(T)) ~ H(T) for each finite valuated tree T. (b) If[S] and [T] are equivalence classes offinite valuated trees with [S] ::: [T], then G(T) ~ G(S).
220
8. Applications of Representations and Butler Groups
PROOF. Statement (a) follows from the fact that v(x) :::: v(f(x» for each x and (b) is clear.
E
G,
The machinery is now in place for the finite valuated p-group analogue of the category of finite rank Butler groups with typeset contained in a finite lattice of types. Let L be a finite sublattice of the lattice of equivalence classes of valuated trees. An L-group is a finite valuated p-group G with v-height x E L for each x E G . Each finite valuated p-group G is an L -group for L the finite lattice generated by {v-heightx : x E G} . Define Ji. to be the poset ofjoin irreducible elements of L with reverse order. For a p-group G, and integer j ::: 1, G[pj) = {x E G: p j x = OJ. Notice that if f : G ~ H is a group homomorphism, then f(G[p j)) is contained in H[pj) .
Theorem 8.2.3 Let L be afinite sublattice ofthe lattice of equivalence classes offinite valuated trees and j ::: 1 an integer.
Z/ pjZ) given by Fj(G) (G[p j) , G(T)[pj) : [T) E h). (b) If U = (U o, U, : i E h) E rep(h, Z/ pZ), then there is an L-group G with Ft(G) = U . (a) There isafunctor Fj from thecategoryofL-groups torep(h,
=
(a) For L -groups G and Hand f : G ~ H a valuated homomorphism, define Fj(f) : G[pj] ~ H[p j). Then Fj(f) is a representation morphism by Lemma 8.2.2(b). It now follows that Fj is a functor. (b) Let n = lSI and m the Z] pZ-dimension of Us; Define G to be the direct sum of m copies of Z/pnz with Uo = G[p) . Write S = {I, 2, ... , n}, so that U (Uo, UI> . . . , Un)' There is a valuation v of G given by v(O) 00 , v(x) 2(n - k) + 1 if order x pk with k ::: 1 and pk-I x E Ui, and v(x) 2(n - k) if order x pk with k ::: 1 and pk-I x fj; Ui, To see that v is a valuation, notice that if x , y E G with order y < order x , then v(y) > v(x) . Thus, v(px) > v(x), since order px < order x. Moreover, v(x + y) ::: min{ v(x) , v(y)}, since order x+ y :::: max{order x , order y}. If order x + y order x order y k, then v(x) , v(y), and v(x + y) are either 2(n - k) or 2(n - k) + 1. If v(x) = v(y) = 2(n - k), then pk-I x and pk-I y are 2(n - k) . The other elements of Ui; whence pk-I(x + y) E Uk and v(x + y) cases are routine. Finally, if gcd(r, p) = I, then v(rx) = v(x), since pk-I x E U« if and only if p'
. . . > tn} with v(t;) = 2i - 2 if Uk n Un-i+1 0 and v(t;) 2i - I if Uk n Un-i+I =1= O. It remains to prove that U, G(Tk)[p) for each k, in which case F(G) U . First let 0 =1= x E Ui, Then order x p so that v(x) 2n - 2 if x fj; U I and v(x) = 2n - 1 if x E UI. Thus, v(tn) :::: v(x) . Similarly, T(x , G) ;2 Ti, so that v-height x ::: Tk and x E G(Tk)[p). Therefore, Uk ~ G(Tk)[P) . Conversely, let x E G(Tk)[p) . Since T(x , G) ;2 Ti; there is y E G with pk-I y = x and v(y) ::: v(tn-k+)). Then x E Ui; otherwise, v(y) = 2(n - k) < PROOF.
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
=
8.2 Finite Valuated Groups
221
V(tn-k+I) = 2(n - k + I) - I = 2(n - k)+ I, since Uk n Un-(n-k+I)+1 = Uk # 0 . This shows that G(Tk([p]) is contained in Ui , as desired.
If U is an indecomposable representation and G is an L-group with Fj(G) = U, then G must be an indecomposable as a valuated group. This observation allows an explicit construction of indecomposable valuated L-groups from a given representation . Example 8.2.4 Assume J: = S3. The only indecomposable L -groups constructed in Theorem 8.2.3(b) are: (i) G = (Zl p3Z)x with (v(x) ,v(px),v(p2x» = (0,2,4), (1,2,4), (0,3,4), (0,2,5), (1,3 ,4), (1,2,5), (0,3 ,5), or (1 , 3,5); and (ii) G (Zl p 3Z)a El7 (Zl p3Z)b with (v(a), v(pa), v(p2 a»= (1,2,4), (v(b), v(pb), v(p 2b» (0,3,4) and (v(a + b), v(p(a + b), v(p2(a + b» = (0,2,5).
=
=
k), l-dimensional representations and a 2-dimensional representation. The groups in (i) correspond to the l-dimensional representations (k, 0, 0, 0), (k, k, 0, 0), (k , 0, k, 0), (k, 0 , 0, k),(k, k, 0), (k , k, 0, k), (k , 0, k, k), and (k, k, k, k), respectively, Similarly, the group in (ii) corresponds to the 2-dimensional representation (k El7 k, k El7 0, El7 k, (1 + I)k).
PROOF. Via Example 1.1.5, the only indecomposable representations in rep(S3,
k=
'Ill pZ, are the
°
=
If G is an L-group with FI (G) = U (U«, U, : i E S), as constructed in Example 8.2.4, then p" G 0, where n is the number of elements of S. Moreover, v(x) ::: 2n + 1 for each x E G .
=
Example 8.2.5 There are infinitely many isomo rphism classes of indecomposable p4-boundedfinite valuated groups G with v(x) ::: 9 for each x E G. PROOF. Apply Theorem 8.2.3 and the fact that rep(S4,
'Ill pZ)
has infinite repre-
sentation type; Example 1.1.6.
Example 8.2.6 The category of p5-bounded finite valuated groups G with v(x) ::: 11 for each x E G has wild representation type. PROOF. A consequence of Theorem 8.2.3 and the fact that rep(S5,
'Ill pZ) has wild
representation type ; Example 1.4.1 . The connection between finite rank Butler groups and finite valuated p-groups is intriguing. A group G E B(T)p is sent to UG (G p» G(r)p : r E Sr) E rep(Sr , Zp). If L is a finite lattice of equivalence classes of finite valuated trees Sr as posets, then G' (Gf p G , (G(r) + pG)lpG: r E Sr) E with Jt.
=
=
=
222
8. Applications of Representations and Butler Groups
rep(h, Z/pZ) . Moreover, there is an L-group H with Ft(H) = G' by Theorem 8.2.3. Consequently, G -* H is a correspondence from B(T) to L-groups such that if G is indecomposable at p, then H is indecomposable as a valuated group . Moreover, rank G = 1 if and only if H is a cyclic group.
Open Questions: 1. Do the functors of Theorem 8.2.3 and Proposition 8.2.4 reflect isomorphisms?preserve indecomposables? 2. Is Theorem 8.2.3(b)truefor all j ::: I?
EXERCISES 1. Prove Lemma 8.2.1. 2. Prove that there are infinitely many isomorphism classes of finite valuated p-groups G that are p3-bounded with v(x) :::: 13 for each x E G.
NOTES ON CHAPTER 8 There is a duality on the quasi -homomorphism category of finite rank torsion-free modules over a discrete valuation ring that sends purely indecomposable modules to copurely indecomposable modules [Arnold 72]. A more abstract version of this duality is given in [Lady 83] for Butler modules over a Dedekind domain R. The representation type for subcategories of TF R, defined in terms of the Q-dimension of a splitting field of a finite rank torsion-free R-module, is partially computed in [Lady 77B, 80A]. These computations employ represent ations of species [Ringel 76] instead of representations of partially ordered sets. Other papers on torsion -free modules of finite rank over discrete valuation rings include [Turgi 77] and [Franzen 83]. There is an extensive and evolving body of literature on torsion-free modules over valuation domains ; see [Fuchs, Salce 85] and references therein . Section 8.2 is a very elementary introduction to valuated groups . The seminal paper on the subject is [Richman, Walker 79]. Other embeddings from categories of representations of finite posets over 'Ill pZ into the category of finite valuated p-groups are given in [Arnold, Richman, Vinsonhaler 91]. These embeddings are functorial and send rank-l groups to simply presented valuated groups , but the functors are not additive functors . It is evident in [Beers, Hunter, Walker 83] that classification of finite valuated p-groups is intimately related to the solution of matrix problems for representations of finite partially ordered sets over 'Ill pi Z. A classification of valuated p-groups with values less than or equal to 4 is given in [Richman , Walker 99]. Each such group is the valuated direct sum of a short list of indecomposable finite valuated p-groups. This parallels the fact that if rep(S , k) has finite representation type, then a representation of arbitrary dimension is a direct sum of finite-dimensional indecomposable representations. On the other hand, there are arbitrarily large valuated p-groups with values less than or equal to 5. The question of whether or not the category of finite valuated p-groups with values less than or equal to 5 has wild representation type remains unresolved . Just as for representations of finite posets over discrete valuation rings, a suitable definition of tame representation type remains elusive.
References
[Albrecht 83] Albrecht, U. Endomorphism rings and A-projective torsion-free abelian groups , Abelian Group Theory, Lect. Notes in Math. 1006, Springer-Verlag , New York, 1983,209-252. [Albrecht 89] Albrecht , U. Endomorph ism rings and a generalization of torsion-freeness and purity, Comm. in Alg. 17 (1989), 1101-1135. [Albrecht 91] Albrecht , U. On the quas i-splitting of exact sequence s, J. Alg . 144 (1991), 344-358. [Albrecht , Goeters 89] Albrecht, U. and Goeters, P. A dual to Baer's lemma , Proc. Amer. Math. Soc. lOS (1989), 817-826. [Albrecht , Goeters 94] Albrecht , U. and Goeters, P. Pure subgroups of A-projective groups, Acta. Math. Hungar . 65 (1994 ), 217-227. [Albrecht, Goeters , Faticoni, Wickles s 97] Albrecht, U; Goeter s, P.; Faticoni, T.; and Subalgebras of rational matrix algebras, Acta Math Hungar 74 (1997). Wickless,
w.,
[Albrecht, Goeters 98] Albrecht , U. and Goeter s, P. Butler theory over Murley groups, J. Alg . 200 (1998),118-133 . [Albrecht, Hill 87] Albrecht U. and Hill, P. Butler groups of infinite rank and axiom 3, Czech. Math. J. 37 (1987), 293-309. [Anderson , Fuller 74] Anderson, EW. and Fuller, K.R. Rings and Categories of Modul es, Graduate Texts in Mathematics, Springer-Verlag, New York, 1974. [Ara, Goodearl , O'Meara, Pardo 98] Ara, P., Goodearl , K., and O'Meara, K. Separative cancellation for projective modules over exchange rings, Israel J. Math . 105 (1998), 105-1 37.
224
References
[Arnold 72] Arnold, D.M. A duality for torsion free modules of finite rank over a discrete valuation ring, Proc. London Math. Soc. (3) 24 (1972), 204-216. [Arnold 73] Arnold, D.M. A class of pure subgroups of completely decomposable abelian groups, Proc. Amer. Math. Soc., 41 (1973),37-44. [Arnold 77] Arnold, D.M. Genera and decompositions of torsion free modules, Abelian Groups, Lect. Notes in Math. 616, Springer-Verlag, New York, 1977, 197-218. [Arnold 81] Arnold, D.M. Pure subgroups of finite rank completely decomposable groups , Abelian Group Theory, Lect. Notes in Math. 874, Springer-Verlag, New York, 1981, 1-31. [Arnold 82] Arnold, D.M. Finite Rank Torsion-FreeAbelian Groups and Rings, Lect. Notes in Math . 931, Springer-Verlag, New York, 1982. [Arnold 86] Arnold, D.M. Notes on Butler groups and balanced extensions, Bull. U.M.I. (6) 5-A (1986),175-184. [Arnold 89] Arnold, D.M. Representations of partially ordered sets and abelian groups, Abelian Group Theory, Contemporary Math. 87 (1989), 91-109. [Arnold 95] Arnold, D.M. Near isomorphism of Butler groups and representations of finite posets, Abelian Groupsand Modules, Kluwer, Boston, 1995,29-40. [Arnold, Dugas 93A] Arnold D.M. and Dugas, M. Block rigid almost completely decomposable groups and lattices over multiple pullback rings, J. Pure and Appl. Alg. 87 (1993), 105-121. [Arnold, Dugas 93B] Arnold D.M. and Dugas, M. Butler groups with finite typesets and free groups with distinguished subgroups , Comm. Alg. 21 (1993), 1947-1982. [Arnold, Dugas 95A] Arnold D.M. and Dugas, M. Locally free finite rank Butler groups and near isomorphism, Abelian Groupsand Modules, Kluwer, Boston, 1995,41-48. [Arnold, Dugas 95B] Arnold, D.M. and Dugas M. Representations of finite posets and near-isomorphism of finite rank Butler groups , Rocky Mountain J. Math. 25 (1995), 591-609. [Arnold, Dugas 96] Arnold, D.M. and Dugas, M. A survey of Butler groups and the role of representations, Abelian Groupsand Modules, Lect. Notes in Pure and Appl. Math. 182, Marcel Dekker, New York, 1996, 1-13. [Arnold, Dugas 97] Arnold, D.M. and Dugas, M. Representation type of finite rank Butler groups , Colloquium Math. 74 (1997), 299-320. [Arnold, Dugas 98] Arnold D.M. and Dugas, M. Representation type of posets and finite rank almost completely decomposable groups, Forum Math. 10 (1998), 729-749. [Arnold, Dugas 99] Arnold, D.M. and Dugas, M. Finite rank Butler groups with small typeset, Abelian Groupsand Modules, Birkhauser, Boston, 1999, 107-120. [Arnold, Dugas, Rangaswamy 98] Arnold, D.M., Dugas, M., and Rangaswamy, K.M. Torsion-free abelian groups with semi-perfect quasi-endomorphism rings, preprint. [Arnold, Lady 75] Arnold, D.M., and Lady, L. Endomorphism rings and direct sums of torsion-free abelian groups , Trans. Amer. Math. Soc., 211 (1975),225-237.
References
225
[Arnold, Hunter, Richman 80] Arnold,D.M., Hunter,R., and Richman, F.Global Azumaya theorems in additivecategories,J. Pure and Appl. Algebra, 16 (1980), 223-242. [Arnold, Murley 75] Arnold, D.M., and Murley, C.E. Abelian groups A, such that Hom(A, -) preserves direct sums of copies of A, Pac. 1. Math., 56 (1975),7-20. [Arnold, Richman 92] Arnold D. M. and Richman, F. Field independentrepresentationsof partially ordered sets, Forum Math. 4 (1992), 349-357. [Arnold,Richman,Vinsonhaler91]Arnold, D.M., Richman,F.,andVinsonhaler, C. General p-valuations of abelian groups, Comm. in Alg. 19 (11) (1991), 3075-3088. [Arnold, Vinsonhaler 81] Arnold, D.M. and Vinsonhaler, C. Pure subgroups of finite rank completely decomposable groups II, Abelian Group Theory , Lect. Notes in Math. 1006, Springer Verlag, 1981,97-143. [Arnold, Vinsonhaler 87] Arnold D. and Vinsonhaler, C. Endomorphism rings of Butler groups, J. Austral. Math. Soc. 42 (1987), 322-329. [Arnold, Vinsonhaler89] Arnold, D. and Vinsonhaler, C. Quasi-endomorphism rings for a class of Butlergroups,Abelian Group Theory , Contemporary Math. 87 (1989),91-110 . [Arnold, Vinsonhaler92] Arnold, D.M. and Vinsonhaler, C. Invariantsfor classes of indecomposable representations of finite posets, J. Alg . 147 (1992), 245-264. [Arnold, Vinsonhaler 93] Arnold D.M. and Vinsonhaler, C. Finite rank Butler groups: A survey of recent results, Abelian Groups, Lect. Notes in Pure and Applied Math. 146, Marcel Dekker,New York, 1993, 17-41. [Auslander 74] Auslander, M. Representation theory of Artin algebras II, Comm. in Alg 1 (1974),269 - 310. [Auslander,Reiten75] Auslander, M. and Reiten, I. Representation theoryof Artin algebras III: Almost split sequences, Comm. in Alg. 3 (1975), 239-294. [Auslander, Reiten 77A] Auslander, M. and Reiten, I. Representation theory of Artin algebras IV: Invariants given by almost split sequences, Comm. in A/g. 5 (1977),443518. [Auslander, Reiten 77B] Auslander, M. and Reiten, I. Representation theory of Artin algebras V: Methods for computing almost split sequences and irrreduciblemorphisms, Comm . in Alg . 5 (1977),519-554. [Auslander, Smale 81] Auslander, M. and Smale, S.O. Almost split sequences in subcategories, J. Alg. 69 (1981),426-454. [Baer 35] Baer, R. Typesof elements and characteristicsubgroupsof abelian groups, Proc. London Math . Soc. 39 (1935),481-514. [Baer 37] Baer, R. Abelian groups withoutelements of finiteorder, Duke Math . J. 3 (1937), 68-122 . [Bass, 68] Bass, H. Algebraic K-Theory, Benjamin, New York, 1968. [Bautista, Martinez 79] Bautista, R. and Martinez, R. Representations of partially ordered sets and I-Gorenstein Artin algebras, Proc. of 1978 Antwerp Conf on Ring Theory, Marcel Dekker, New York, 1979,385-433.
226
References
[Beaumont, Pierce 61A] Beaumont, R.A., and Pierce, R.S. Torsion-free rings, Illinois. J. Math. 5 (1961), 61-98. [Beaumont, Pierce61B] Beaumont, R.A., and Pierce,RS . Torsion freegroupsofrank two, Mem. Amer. Math. Soc. 38 (1961). [Beaumont, Pierce 61C] Beaumont, RA., and Pierce, RS . Subrings of algebraic number fields, Acta Sci. Math. (Szeged) 22 (1961), 202-216. [Beers, Hunter, Walker 83] Beers,D., Hunter, R, and Walker E.A. Finitevaluated p-groups, Abelian GroupTheory, Lect. Notesin Math. 1006,Springer-Verlag, 1983. [Benabdallah, Mader98] Benabdallah, K. and Mader, A. Almostcompletely decomposable groups and integrallinear algebra, J. Alg. 204 (1998),440-482. [Bican 70] Bican, L. Purely finitely generated abelian groups, Comm. Math. Univ. Carol. II (1970), 1-8. [Bican 78] Bican, L. Splitting in abeliangroups, CzechMath. J. 28 (103) (1978),356-364. [Bican, Rangaswamy 95] Bican L. and Rangaswamy K.M. A result on 81 -groups, Rend. Sem. Univ. Padova 94 (1995), 95-98 . [Bican, Salce 83] Bican L. and Salce, L. Butler groups of infinite rank, Abelian Group Theory, Lect. Notes in Math. 1006, Springer-Verlag, New York, 1983. [Blazhenov 96] Blazhenov, A.Y. The genera and cancellation of torsion-free modules of finite rank, St. Petersburg Math. J. 7 (1996), 891-924. [Brenner67] Brenner, S. Endomorphism algebrasof vectorspaces with distinguished sets of subspaces, J. Alg. 6 (1967), 100-114. [Brenner74A] Brenner, S. Decomposition properties of some small diagrams of modules, Symposia Mathematica XIII, Academic Press, New York, 1974, 127-141. [Brenner74B] Brenner, S. On foursubspaces of a vectorspace,J. Alg., 29 (1974), 587-599. [Biinermann 81] Biinermann, D. Auslander-Reiten quiversof exactone-parameter partially ordered sets, Representations ofAlgebras, Lect. Notesin Math. 903, Springer-Verlag, New York, 1981,55-61. [Biinermann 83] Biinermann, D. Hereditary torsion theories and Auslander-Reiten sequences, Arch. der Math. 41 (1983),304-309. [Burkhardt 84] Burkhardt, R On a special classof almostcompletely decomposable groups I, Abelian Groups and Modules, CISM Courses and Lectures 287, Springer-Verlag, New York, 1984, 141-150. [Burkhardt, Mutzbauer96] Burkhardt, Rand MutzbauerO. Almostcompletely decomposable groupswith primarycyclicregulating quotient, Rend. Sem. Mat. Univ. Padova 95 (1996), 81-93. [Butler65] Butler,M.C.R A classof torsion-freeabeliangroupsof finite rank,Proc. London Math. Soc. (3) 15 (1965), 680-698. [Butler 68] Butler, M.C.R. Torsion-free modules and diagrams of vector spaces, Proc. London Math. Soc. (3) 18 (1968), 635-652.
References
227
[Butler 87] Butler, M.C.R.Some almostsplitsequencesin torsion-free abeliangrouptheory, Abelian Group Theory, Gordon and Breach, New York, 1987,291-302. [Charles 68] Charles,B. Sous-groupes fonctoriels et topologies, Studies on Abelian Groups , Springer-Verlag Dunod, Paris, 1968, 75-92. [Comer 61] Comer, A.L.S. A note on rankand directdecompositions of torsion-free abelian groups, Proc. Cambridge Philos. Soc. 57 (1961),230-233. [Comer 63] Comer, A.L.S. Every countablereduced torsion-free ring is an endomorphism ring, Proc. London Math . Soc. 13 (1963), 687-710. [Comer 69] Comer,A.L.S. A noteon rankand directdecompositions of torsion-free abelian groups, Proc. Cambridge Philos . Soc. 66 (1969), 239-240. [Comer, Gobel 85] Comer, ALS. and Gobel, R. Prescribing endomorphism algebras: a unified treatment, Proc. London Math. Soc. (3) 50 (1985), 447-479. [Crawley-Boevey 88] Crawley-Boevey, W. On tame algebras and bocses, Proc. London Math . Soc. (3) 56 (1988), 451-483.
[Crawley-Boevey 91] Crawley-Boevey, W. Tamealgebrasand generic modules, Proc. London Math. Soc. (3) 63 (1991), 241-265.
[Crawley-Boevey 92] Crawley-Boevey, W. Modules of finite length over their endomorphism rings, London Math . Soc. Lecture Notes 168, CambridgeUniversity Press, 1992, 127-184.
[Cruddis 70] Cruddis, T.B. On a class of torsion free abelian groups, Proc. London Math. Soc. 21 (1970),243-276.
[DeVivo, Metelli 96] DeVivo C. and Metelli, C. B(I)-groups: Some counterexamples, Abelian Groups and Modules, Lect.NotesinPureandAppl.Math. 182, MarcelDekker, New York, 1996, 227-232. [DeVivo, Metelli 99] Devivo, C. and Metelli, C. Admissible matrices as base changes of B(ll.groups, Abelian Groups and Modules , Birkhauser, Boston, 1999, 135-148. [Dittmann, Mader, Mutzbauer 97] Dittmann, U., Mader, A., and Mutzbauer, O. Almost completelydecomposablegroups with a cyclic regulating quotient, Comm. in Alg. 25 (1997), 769-784. [Dlab , Ringel 76] Dlab, V. and Ringel, C. Indecomposable representations of graphs and algebras, Mem . Amer. Math . Soc. 173 (1976).
[Drozd 72] Drozd, Y. Representations of commutative algebras, Funct . Anal. and Appl. 6 (1972),286-288.
[Drozd 74] Drozd, Y. Coxeter transformations and representations of partiallyorderedsets, Funct . Anal. and App. 8 (1974), 219-225.
[Dugas,Gobel 82] Dugas, M. and Gobel, R. Everycotorsion-free ring is an endomorphism ring, Proc. London Math. Soc. 45 (1982),319-336. [Dugas,Gobel 97] Dugas, M. and Gobel, R. Endomorphism rings of B2-groups of infinite rank, Israel J. Math . 101 (1997), 141-156.
228
References
[Dugas, Oxford 93] Dugas, M. and Oxford , E. Near isomorphism invariants for a class of almost completely decomposable groups, Abelian Groups, Lect. Notes in Pure and Applied Math 146, Marcel Dekker, New York, 1993, 129-150. [Dugas, Thome 91] Dugas, M. and Thome B. Countable Butler groups and vector spaces with four distinguished subspaces, J. Alg. 138 (1991), 249-272. [Faticoni 87] Faticoni, T. Each countable reduced torsion-free commutative ring is a pure subring of an E-ring, Comm. in Alg. 15 (12), 1987, 2545-2564. [Faticoni , Schultz 96] Faticoni, T. and Schultz, P. Direct decompositions of almost completely decomposable groups with primary regulating index, Abelian Groups and Module s, Marcel Dekker, New York, 1996,233-242. [Files, Gobel 98] Files, S. and Gobel , R. Gau,B ' theorem for two submodules, Math. Zeit. 228 (1998), 511-536. [Franzen 83] Franzen, B. Exterior powers and torsion-free modules over almost maximal valuation domains , Abelian Group Theory, Lect. Notes in Math. 1006, Springer- Verlag, New York, 1983,599-606. [Fuchs 58] Fuchs, L. Abelian Groups., Publ. House of the Hungar. Acad . Sci., Budapest, 1958. [Fuchs 70A] Fuchs , L.Infinite Abelian Groups, Vol. I, Academic Press, 1970. [Fuchs, 70B] Fuchs, L. The cancellation property for modules, Rings and Modules, Lect. Notes in Math. 246, Springer-Verlag, New York, 1970, 192-213. [Fuchs 71] Fuchs, L. Note on direct decompositions of torsion-free abelian groups, Comment. Math. Helv. 46 (1971), 87-91. [Fuchs 73] Fuchs, L. Infinite Abelian Groups, Vol. II, Academic Press, New York, 1973. [Fuchs 94] Fuchs, L. A survey of Butler groups of infinite rank, Abelian Group Theory, Contemporary Math. 171 (1994),121-139. [Fuchs, Grabe 75] Fuchs , L. and Grabe , P. Number of indecomposable summands in direct decompositions of torsion free abelian groups, Rend. Sem. Mat. Univ. Padova 53 (1975),135-148 . [Fuchs, Loonstra 70] Fuchs , L., and Loonstra, E On direct decompositions of torsion-free abelian groups of finite rank, Rend. Sem. Mat. Univ. Padova 44 (1970), 75-83. [Fuchs, Loonstra 71] Fuchs, L. and Loonstra E , On the cancellation of modules in direct sums over Dedekind domains , Indag. Math. 33 (1971), 163-169. [Fuchs , Metelli 91] Fuchs, L. and Metell i, C. On a class of Butler groups, Manuscripta Math. 71 (1991), 1-28. [Fuchs, Metelli 92] Fuchs, L. and Metelli, C. Countable Butler groups, Abelian Groups and Non-commutative Rings , Contemporary Math. 130 (1992), 133-143. [Fuchs , Salce 85] Fuchs, L. and Salce, L. Modules over valuation domains, Lect. Notes in Pure and Appl. Math. 97, Marcel Dekker, New York, 1985. [Gabriel 72] Gabriel , P. Unzerlegbare Darstellung I, Manuscripta Math. 6 (1972), 71-
103.
References
229
[Gabriel 73A] Gabriel , P. Representations indecomposable des ensembles ordonnes, Seminaire P. Dubriel, Paris, 1972-1973, 1301-1304. [Gabriel 73B] Gabriel, P. Indecomposable representations II, Symposia Mathematica XI, Academic Press , Orlando, 1973,81-104. [Gelfand , Ponomarev 70] Gelfand, I. and Ponomarev, I. Problems of linear algebra and classification of quadruples in a finite dimensional vector space, Colloq. Math. Soc. Bolyai, Tihany, 1970, 163-237. [Gobel , Shelah 95] Gobel. R. and Shelah, S. On the existence of rigid aleph-I free abelian groups of cardinality aleph-l , Abelian Groups and Modules , Kluwer, Boston, 1995, 227-237. [Goeters 99] Goeters, H.P. Butler modules over I-dimensional Noetherian domains ,Abelian Groups and Modules , Birkhauser, Boston , 1999, 149-166. [Goeters, Ullery 90] Goeters, H.P. and Ullery W. Butler groups and lattices of types, Comm. Math. Univ. Carol. 31 (1990) ,613-619. [Goeters, Ullery 91] Goeters, H.P. and Ullery W. Homomorphic images of completely decomposable finite rank torsion-free groups , 1. Alg. 140 (1991) , I-II. [Goeters, Ullery 92] Goeters, H.P. and Ullery W. Almost completely decomposable torsionfree groups, Rocky Mtn. J. Math. 22 (1992) , 593-600. [Goeters , Ullery 94] Goeters, H.P. and Ullery W. The Arnold-Vinsonhaler invariants, Houston J. Math. 20 (1994) ,17-25. [Goeters, Ullery 95] Goeters, H.P. and Ullery W. On quasi-decompositions of B0l-groups, Comm. in Alg. 23 (1995) ,131-137. [Goeters, Ullery, Vinsonhaler 94] Goeters , H.P., Ullery w., and Vinsonhaler, C. Numerical invariants for a class of Butler groups, Abelian Group Theory, Contemporary Math . 171 (1994), 159-172. [Goodearl 76] Goodearl, K.R. Power cancellation of groups and modules , Pacific J. Math., 64 (1976) , 387-411. [Griffith 67] Griffith, P. Purely indecomposable torsion-free groups, Proc. Amer. Math Soc. 18 (1967), 738-742. [Gustafson 82] Gustafson, W. The history of algebra and their representations, Representations ofAlgebras, Lect. Notes in Math . 944, Springer-Verlag, New York, 1982, 128. [Hahn 97] Hahn , R. Representations of Partially Ordered Sets , Master's Thesis, Baylor University, 1997. [Heller, Reiner 61] Heller, A. and Reiner, I. Indecomposable representations, Ill. J. Math. 5 (1961) , 314-323. [Hill, Megibben 93] Hill, P. and Megibben, C. The classification of certain Butler groups, J. Alg. 160 (1993), 524-551. [Holling 93] Holling, B. On direct summands of B(I)-groups, Comm. in Alg. 21 (1993) , 1849-1856.
230
References
[Hungerford, 74] Hungerford, T. Algebra, GraduateTextsin Mathematics, Springer-Verlag, New York, 1974. [Hunter, Richman, Walker 84] Hunter, R., Richman, E, and Walker, E.A. Subgroups of bounded abelian groups, Abelian Groups and Modules, CISM Courses and Lectures 287, Springer-Verlag, New York, 1984, 17-36. [Jacobinski 68] Jacobinski, H. Genera and decompositions of lattices over orders, Acta . Math . 121 (1968),1-29. [Jonsson 57] Jonsson, B. On direct decompositions of torsion-free abelian groups, Math. Scand. 5 (1957),230-235 . [Jonsson 59] Jonsson, B. On direct decomposition of torsion-free abelian groups, Math. Scand. 7, (1959) 361-371. [Kaplansky 69] Kaplansky, I. Infinite Abelian Groups, University of Michigan Press, Ann Arbor, Michigan, 1954, 1969. [Kerner 81] Kerner, O. Partiallyordered sets of finite representation type, Comm. in Alg. 9 (1981),783-809. [Kleiner 75A] Kleiner, M. Partially ordered sets of finite type, J. Soviet Math. 23 (1975), 607-615. [Kleiner75B] Kleiner, M. On exactrepresentations of finitetype, J. Soviet Math . 23 (1975), 616-628. [Koehler 64] Koehler, J. Some torsion-free rank two groups, Acta . Sci. Math . Szeged 25 (1964),186-190. [Koehler65] Koehler, J. The type set of a torsion-free group of finite rank, Illinois 1. Math. 9 (1965),66-86. [Krapf, Mutzbauer84] Krapf, K.J. and Mutzbauer, O. Classification of almost completely decomposable groups,Abelian Groups and Modules, CISMCoursesand Lectures287, Springer-Verlag, New York, 1984, 151-162. [Krugljak 64] Representations of the (p , p) group over a field of characteristic p , Soviet Math. Dokl. 4 (1964),1809-1813. [Lady 74A] Lady, E.L. Summands of finite rank torsion free abelian groups, J. Alg. 32 (1974),51-52. [Lady 74B] Lady, E.L. Almostcompletelydecomposable torsionfree abeliangroups, Proc. A.M.S. 45 (1974),41-47 . [Lady 75] Lady, E.L. Nearly isomorphic torsion free abelian groups, J. Alg. 35 (1975), 235-238. [Lady 77A] Lady, L. On classifying torsion free modules over discrete valuation rings, Abelian Group Theory, LectureNotesin Math. 616, Springer-Verlag, NewYork, 1977, 168-172. [Lady 77B] Lady, L. Splitting fields for torsion-free modulesover discrete valuation rings I, J. Alg. 49 (1977),261-275.
References
231
[Lady 79] Lady, L. Extension of scalars for torsion-free modules over Dedekind domains, Symposia Math. XXIII, Academic Press, New York, 1979,287-305. [Lady 80A] Lady, L. Splitting fields for torsion-free modules over discrete valuation rings II, J. Alg. 66 (1980), 281-306. [Lady 80B] Lady, L. Splitting fields for torsion-free modules over discrete valuation rings III, J. Alg. 66 (1980) , 307-320. [Lady 83] Lady, L. A seminar on splitting rings for torsion-free modules over Dedekind domains, Abelian Groups, Lect. Notes in Math. 1006, Springer-Verlag, New York, 1983,1-48. [Lambek 66] Lambek, 1. Lectures on Rings and Modules , Blaisdell , Waitham, 1966. [Lee 89] Lee, W.Y. Co-diagonal Butler groups, Chinese J. Math. 17 (1989), 259-271 . [Levi 17] Levi, FW. Abeische Gruppen mit abzahleren Elementen, Habilitationschrift (Leipzig, 1917). [Lewis 93] Lewis , W.S. Almost completely decomposable groups with two critical types, Comm. in Alg. 21 (1993),607-614. [Loup ias 75] Loupias , M. Indecomposable representations of finite ordered sets, Representations of Algebras , Lect. Notes in Math . 488, Springer-Verlag, New York, 1975, 201-209. [Mader 93] Mader , A. Friedrich Wilhelm Levi, 1888-1966, Abelian Groups, Lect. Notes in Pure and App!. Math. 146, Marcel Dekker, New York, 1993, 1-14. [Mader 95] Mader, A. Almost completely decomposable groups, Abelian Groups and Module s, Kluwer, Boston , 1995,343-366. [Mader 00] Mader, A. Almost Completely Decomposable Groups, Gordon and Breach , 2000 . [Mader, Mutzbauer 93] Mader, A. and Mutzbauer O. Almost completely decomposable groups with cyclic regulating quotient , Abelian Groups, Lect. Notes in Pure and App!. Math. 146, Marcel Dekker, 1993, 193-200. [Mader, Mutzbauer, Rangaswamy 94] Mader, A., Mutzbauer, 0 ., and Rangaswamy , K.M., A generalization of Butler groups, Abelian Group Theory, Contemporary Mathematics 171 (1994), 257-275. [Mader, Nonxga 99] Mader, A. and Nonxga, L. Completely decomposable summands of almost completely decomposable groups, Abelian Groups and Modules , Birkhauser, Boston, 1999,167-190. [Mader, Vinsonhaler 94] Mader, A. and Vinsonhaler, C. Classifying almost completely decomposable groups, J. Alg. 170 (1994), 754-780. [Mader, Vinsonhaler 95] Mader, A. and Vinsonhaler, C. Almost completely decomposable groups with a cyclic regulating quotient, J. Alg. 177 (1995), 463-492. [Mader, Vinsonhaler 97] Mader, A. and Vinsonhaler, C. The idempotent lifting property for almost completely decomposable groups , Colloq. Math. 72 (1997), 305-317.
232
References
[Mines,Vinsonhaler 92] Mines, R. and Vinsonhaler, C. Butlergroups and Bext:a constructive approach, Contemporary Math . 130 (1992), 289-299. [Mitchell 65] Mitchell,B. Theory of Categories, AcademicPress, New York, 1965. [Mouser 93] Mouser, M.A. On Zlp2Z- modules with distinguished submodules, master's thesis, Baylor University, 1993. [Mutzbauer 93] Mutzbauer, O. Regulating subgroups of Butler groups, Abelian Groups, Lect. Notes in Pure and App!. Math. 146, Marcel Dekker,New York, 1993,209-217. [Mutzbauer99] Mutzbauer, O. Normalforms of matriceswith applications to almostcompletelydecomposable groups,Abelian Groups and Modules, Birkhauser, Boston, 1999, 121-134. [Nazarova 75] Nazarova, L.A. Partiallyordered sets of infinitetype, Math . USSR /zvestija 9 (1975), 911-938. [Nazarova 81] Nazarova, L.A. Poset representations, Integral Representations and Appli cations , Lect. Notes in Math. 882, Springer-Verlag, New York, 1981,345-356. [Nazarova, Roiter 75] Nazarova, L.A. and Roiter, A.V. Representations of partiallyordered sets, J. Soviet Math. 23 (1975),585-607. [Nonxga,Vinsonhaler 95] Nonxga,L. and Vinsonhaler, C. Completelydecomposable subgroups and factor groups of finite rank completely decomposable groups, Abelian Groups and Modules, Kluwer, Boston, 1995,385-394. [Nonxga, Vinsonhaler 96] Nonxga, L. and Vinsonhaler, C. Balanced Butler groups, J. Alg. 180 (1996), 546-570. [Nonxga, Vinsonhaler 97] Nonxga, L. and Vinsonhaler, C. Balanced and cobalanced representations of posets, Comm. in Alg. 25 (12) (1997), 3735-3749. [O'Meara, Vinsonhaler 99] O'Meara, K. and Vinsonhaler, C. Separativecancellation and multipleisomorphismin torsion-free abelian groups, J. Alg . 221 (1999),536-550. [Ould-Beddi, Strungrnann 99] Ould-Betti, M.A., and Strungrnann, L. Stacked bases for a pairof homogeneous completelydecomposable groupswithboundedquotient,Abelian Groups and Modules, Birkhauser, Boston, 1999, 199-210. [Pierce 60] Pierce,R.S. Subringsof simplealgebras,Michigan Math . 1., 7 (1960),241-243. [Plahotnik 76] Plahotnik, V.V. Representations of partially ordered sets over commutative rings, Math . USSR Isvestija 10 (1976), 497-514. [Pontryagin 34] The theory of topological commutative groups, Ann . Math. 35 (1934), 361-368. [Rangaswamy, Vinsonhaler 94] Rangaswamy, K.M. and Vinsonhaler, C. Butlergroupsand representations of infinite rank,). Alg. 167 (1994), 758-777. [Reid 62] Reid, 1. On quasi-decompositions of torsion-free abelian groups, Proc. Amer. Math. Soc. 13 (1962),550-554. [Reid 63] Reid, 1. On the ring of quasi-endomorphisms of a torsion-free group, Topics in Abelian Groups, Scott-Foresman, Chicago, 1963,51-68.
References
233
[Reid81] Reid,J. Abeliangroupsfinitely generatedovertheir endomorphism rings, Abelian Group Theory, Lect. Notes in Math. 874, Springer-Verlag, New York, 1981,41-52. [Reid 83] Reid, 1. Abelian groups cylic over their endomorphism rings, Abelian Group Theory , Lect. Notes in Math. 1006, Springer-Verlag, 1983, 190-203. [Reid 99] Reid, 1. Some matrix rings associated with acd-groups, Abelian Groups and Modules, Birkhauser, Boston, 1999, 191-198. [Reiner 75] Reiner, 1. Maximal Orders, AcademicPress, New York, 1975. [Reiten80] Reiten, 1. The use of almostsplit sequencesin the representation theoryof Artin algebras, Representations ofAlgebras, Lect. Notesin Math. 944, Springer-Verlag, New York, 1980, 163-237. [Richman68] Richman,F. A classof rank-2torsion-freegroups, Studies on Abelian Groups, Springer-Verlag Dunod, Paris, 1968,327-333. [Richman 77] Richman, F. A guide to valuated groups, Abelian Groups, Lect. Notes in Math. 616, Springer, 1977,73-86. [Richman83] Richman,F. An extensionof the theoryof completely decomposable torsionfree abelian groups, Trans. A.M.S. 279 (1983), 175-185. [Richman84] Richman, F. Butler groups, valuated vector spaces, and duality, Rend. Sem. Mat. Univ. Padova 72 (1984),13-19. [Richman95] Richman, F. Isomorphism of Butlergroupsat a prime, Abelian Group Theory, Contemporary Math. 171 (1995),333-337. [Richman, Walker79] Valuated groups,1. Alg. 56 (1979), 145-167. [Richman, Walker98] Richman, F. and Walker, E. Filtered modules over discrete valuation domains, J. Alg. 199 (1998), 618-645. [Richman,Walker99] Richman,F. andWalker, E. Subgroupsof p5-boundedgroups,Abelian Groups and Modules , Birkhauser, Boston, 1999,55-74. [Ringel76] Ringel,C. Realizations ofK-species andbimodules, J. Alg . 41 (1976),269-302. [Ringel 78] Ringel, C. The spectrumof a finite dimensional algebra, Proceedings of 1978 Antwerp Conference on Ring Theory, Marcel Dekker,New York, 1979,535-597 . [Ringel 80] Ringel, C. Report on the Brauer-Thrall conjectures, Representation Theory I, Lect. Notes in Math. 831, Springer-Verlag, New York, 1980, 104-136. [Ringel84] Ringel,C. Tame Algebras and Integral Quadratic Forms, Lect. Notes in Math. 1099,Springer-Verlag, New York, 1984. [Ringel 91] Ringel, C. Recent advances in the representation theory of finite dimensional algebras, Progress in Math. 95 (1991),141-191. [Ringel, Tachikawa 75] Ringel, C. and Tachikawa, H. QF-3 rings, 1. Reine Angew. Math . 272 (1975), 49-72 . [Rojter 80] Rojter, A.V. Matrix problems and representations of bocses, Representation Theory I, Lect. Notes in Math. 831, Springer-Verlag, New York, 1980,288-324.
234
References
[Schultz 87) Schultz, P. Finite extensions of torsion-free groups, Abelian Group Theory, Gordonand Breach, New York, 1987,333-350. [Simson 74) Simson, D. Functor categories in whichevery finiteobject is projective, Bull. Acad. Polon. Sci. 22 (1974), 375-380. [Simson 92) Simson,D. Linear Representations ofPartially OrderedSets and Vector Space Categories, Gordon and Breach, Philadelphia, 1992. [Simson 93) Simson, D. On representation types of modulesubcategories and orders, Bull. Acad. Pol. Sci. Math. 41 (1993),77-93. [Simson 96) Simson, D. Socle projective representations of partially ordered sets and Tits quadraticformswith applications to latticesoverorders,AbelianGroupsandModules, Marcel Dekker, New York, 1996, 73-112. [Stelzer 85) Stelzer,1. A cancellationcriterionfor finite rank torsion-free abelian groups of finiterank, Proc. Amer. Math. Soc. 94 (1985), 363-368. [Stelzer 87] Stelzer, 1. Ring theoretical criteria for cancellation, Abelian Group Theory, Gordon and Breach, New York, 1987, 175-196. [Swan 62] Swan, RG. Projective modules over group rings and maximal orders, Ann. of Math. (2) 76 (1962), 60-69. [Turgi 77] Turgi,M. A sheaf-theoretic interpretation of the Kuroschtheorem,AbelianGroup Theory, Lect. Notes in Math. 616, Springer-Verlag, New York, 1977, 168-196. [Turnbull, Aitken61] Turnbull,H. andAitken,A.An Introduction to theTheoryofCanonical Matrices, Dover, New York, 1961. [Vinsonhaler 95] Vinsonhaler, C. Invariants for almost completely decomposable groups, Abelian Groups and Modules, Kluwer, Boston, 1995,485-490. [Walker 64] Walker, E.A. Quotient categories and quasi-isomorphisms of abelian groups, Proc. Colloq. Abelian Groups, Budapest, 1964, 147-162. [Warfield 68] Warfield, RB. Jr. Homomorphisms and dualityfor torsion-free groups,Math. Zeit. 107 (1968), 189-200. [Warfield 72] Warfield, R.B. Jr. Exchangeringsanddecompositions of modules. Math.Ann. 199 (1972), 31-36.
[Warfield 80] Warfield, RB. Jr. Cancellation for modulesand stablerangeof endomorphism rings, Pac. J. Math. 91 (1980),457-485. [Yakolev 91] Yakolev, A. Torsion-free abeliangroups of finiterank and their direct decompositions, J. Soviet Math. 57 (1991), 3524-3533. [Yom 94] Yom, P. A characterization of a class of Butlergroups II, Abelian GroupTheory, Contemporary Math. 171 (1974),419-432. [Yom 98] Yom. P. A characterization of a class of Butler groups, preprint.
List of Symbols
A=(A1, . . ·,A n )
a<1
A(m, A)
Ap
AQ
b'> bf(G)
B(T) B(T, j) B(T, j)p B(T)Q C(A) C(A, B) cdn H cdn U
CG
CCn
-
C(n)
-
coker f Coker f == (mod p)
-
n-tuple of subgroups of Q subset of a poset S with a E S Jordan block matrix isomorphism at p category of torsion -free abelian groups quasi-homomorphism category of torsion-free abelian groups subset of poset S with b E S Burkhardt invariant of group G category of finite rank Butler groups, T a lattice of types subcategory of B(T), j ::: 0 isomorphism at p category of B(T, j) quasi-homomorphism category of B(T) centralizer of matrix A centralizer of two matrices A and B coordinate vector of a torsion-free abelian group H coordinate vector of representation U subring of the center of the center of End UG, UG E rep(ST , Q) Coxeter correspondence chain with n elements a category of almost completely decomposable groups with typeset contained in an antichain the cokemel of an R-module homomorphism the cokemel of a representation morphism congruence of integers modulo a prime p
236
List of Symbols
c(f)
C+ CR Crep(Sr,
CTcr(G)
CCT, j) CCT, n.
c-sr. j)
8(A) 8(a,b)S
d(G)
z; j)
-
content of polynomial I(x) E 'll[x] Coxeter correspondence center of ring R the subcategory of rep( Sr , 7l p » j) corresponding to
CCT, n,
- the subcategory of rep( Sr , 7lp » j) corresponding to Ccrit(T, j) - critical cotypeset of G - almost completely decomposable groups in B(T, j) - isomorphism at p category of CCT, j) - almost completely decomposable groups in B(T, j) with critical typeset S;; Sr - n-tuple of subgroups of Q - derivative of poset S with respect to suitable pair (a, b) - maximal divisible subgroup of torsion-free abelian group
G d(M)
dimU
EB eG E
~ EndU EndR(M) E(n, k)
Ig
r:
frA frAlp
frA p frAQ
G[A]
gcd GIH
Gn
Gp G[pi]
- maximal divisible submodule of torsion-free module over a discrete valuation ring - a subcategory of rep(Sr, 711 pi'll) of representations in matrix problem form - a subcategory of rep(Sn+l, R) of representations in matrix problem form over a ring R - dimension of a representation U - direct sum - exponent of G I R( G) for a finite rank Butler group G - element of - not element of - ring of endomorphisms of representation U - R-endomorphism ring of R-module M - a subcategory ofrep(Sn+2, k) - composition of morph isms I and g defined by Ig(x) = I(g(x» - composition of a morphism I repeated n times - category of torsion- free abelian groups of finite rank - mod p category of torsion-free abelian groups of finite rank - isomorphism at p category of torsion-free abelian groups of finite rank - quasi-homomorphism category of frA - bracket group - greatest common divisor - group of cosets of a subgroup H of an abelian group G - direct sum of n copies of abelian group G - localization of torsion-free abelian group at a prime p - pi -socle of an abelian group
Listof Symbols
G(T)
- localizationof a torsion-free abelian group at a set of primes S - socle of a valuated abelian group for a tree T
Gr
-
Gs
G[r] G*[r] G(r) G*(r) G#(r)
hex) hp(x)
H+K
Hom(A , B)
Hom(G, H) Hom(G, H)p HomR(M, N) (i)
i(G) 1M
In lnd(mod R) lnd(n, k)
Ind(S, k) Ind(S, R , j) I (t)
IT(G) iu I(U)
n
J/(T)
it
JR
ker f Ker f kn
k[x] k(x) k(x, y) k(U)
237
- r-radical
subgroup containing r -radical r -socle of torsion-free abelian group G subgroup of G generated by {x : type X > r ] pure subgroup generated by G*( r ) r-homogeneous completely decomposablesubgroup of Butler group G with G(r) = G r E9 G#(r) - height sequence of an element x in an abelian group - p-height of an element x in an abelian group - subgroup of group G generated by Hand K - morphism group of objects A and B in additivecategory - group of homomorphisms of abelian groups G and H - localization of Hom(G, H) at a prime p - group of R-homomorphisms of R-modules M and N - chain of length i - regulating index of an almost completely decomposable group G - submoduleof R-module M generated by {xm: x E I , m E M} for ideal I of R - nth power of ideal I - isomorphismclasses of indecomposables in mod R - isomorphismclasses of indecomposable k-representations of an antichain - isomorphismclasses of indecomposable representations of poset S - isomorphismclasses of indecomposables in rep(S, R , j) - l-dimensional injective in rep(S, k) - inner type of G - injection for direct sum - injectiveenvelopeof U E rep(S, k) - intersectionof sets, greatest lower bound of types - join-irreducible elements of a poset T - opposite of the poset of a lattice of v-heights - Jacobson radical of ring R - the kernel of an R -homomorphism f - the kernel of a representation morphism f - n-dimensional k-vector space for field k - polynomial ring in indeterminate x with coefficients in k - quotient field of k [x] - polynomial ring in two noncommuting variables x and y - a representation determined by U E rep(S, k)
238
List of Symbols
km L(U) Matm(k) M~N
M- 1
MIN modR
Mp
M(t) MOl
JLr
Mu
=(Ail, . · IA n )
n(G)
n+
NR OT(G)
n
P(A)
Pf(R) P(n, j)
pWG
peR) P(t) Pu
P(U)
o Q
qG
iQG Qrep(S, R) qu
R
R* rep(S, k)
-
least common multiple representation determined by U E rep(S, K) ring of all m x m matrices equivalence of matrices M and N under row operations and block column operations - inverse of square matrix M - module of cosets of submodule N of module M - category of finitely generated R-modules - a localization of M E Qrep(S, R) at an up-set P containing M - a localization of M E Qrep(S, R) at an up-set P contained in M - a module over R, t E R* - a module over R, (t) a subset of R* - right multiplication by an element r of a ring - matrix of representation U with block matrices Ai , - an integer determined by the endomorphism ring of a torsion-free abelian group of finite rank - {1,2, . .. ,n} - nil radical of ring R with finite rank torsion-free additive group - outer type of G - set of primes of the integers - category of summands of direct sums of copies of an abelian group A - category of summands of finite direct sums of copies of an abelian group A - category of finitely generated projective right R -modules - a poset - p-divisible subgroup of abelian group G - category of projective right R -modules - I-dimensional projective in rep(S, k), in Qrep(S, R) - projective cover P(U) ~ U E rep(S , k) - projective representation determined by U E rep(S , k) - field of rational numbers - quotient field of discrete valuation ring - {qx: x E G} for q a rational and G a torsion-free abelian group - {qx: q E iQ, x E G} for torsion-free abelian group G - category of quiver representations of a finite poset S over a principal ideal domain R - injective envelope U ~ leU) E rep(S, k) - ring, associative, with identity - completion of a ring R - category of finite-dimensional k-representations of S
List of Symbols rep(S, R)
rep(S, R , j) Rep(S, k) Rep(S, R)
R1
X
R(G)
R2
R+
RT(G)
Ru lSI
S* a
(SI," ·, Sn) Sn S(n , j) sop
I:{Vi:iES} I:{G i :iES} Sr (S) (S)*
T(M) r(P) T(x , G)
VI +V2 V = (Vo, Vi: Vcrit(T, j)
239
- category of finite rank free representations of poset S over ring R, over discrete valuation ring - category of finitely generated representations containing rep(S, R) - subcategory ofrep(S, R) for R a discrete valuation ring - category of countable representations of poset S over field k - category of countable representations of poset S over ring R - ring product of rings R1 and R2 - regulator subgroup of a finite rank Butler group G - additive group of a ring R - Richman type of G - endomorphism ring of matrix M u - cardinality of set S - disjoint union of a poset S and a point - dualityfromrep(S ,k)torep(SOP ,k) - disjo int union of posets S, - antichain with n elements - poset - oppos ite of poset S - subspace of Vo generated by subspaces {Vi : i E S} of Vo E rep(S, k) - subgroup of G generated by subgroups G, of G - opposite of poset of join irreducible elements of T - subgroup of an abelian group generated by a subset S - pure subgroup of an abelian group generated by a subset S - critical typeset of G - tensor product - meet of a set of types with indices in E - category of torsion-free modules of finite rank over a discrete valuation ring R - torsion-subrepresentation of M E Qrep(S, R) - type of a partition of n+ - finite valuated tree for x E G, a valuated abelian p-group - subspace of Vo generated by VI and V2 i E S) - representation of poset S - subcategory of Ccrit(T, j) of uniform almost completely decomposable groups - subspace of Vi generated by {V j : j < i} - n{Vj : j > i} for V = tU«, Vi) E rep(S, k) - direct sum of n copies of representation V - union of sets, least upper bound of types
*
240
List of Symbols
VeT, j) V
VA w(S) W(V) W'(V)
Z Z[ljn]
- subcategory of qT, j) of uniform almost completely decomposable groups - valuation of an abelian p-group - k[x]-module defined by matrix A - width of poset S - injective envelope of C+V for V E rep(S, k) - projective cover of C- V for V E rep(S, k) - ring of integers - subring of rationals generated by 1 and I j n for nonzero integer n - 'Zj i'Z if i integer and 'Zp/'Z if i = 00 - localization of integers at a prime p
Index of Terms
abelian group almost completely decomposable, 51 balanced projective , 198 B(I) -,205 bounded ,48 bracket, 205 Butler, 91, 105 central special, 115 cobalanced injective, 198 completely decomposable, 51, 105 divisible, 56 exponent of finite, 147 finitely Butler, 105 homogeneous , 79 indecompo sable at p , 51 injective in B(T), 197 L- ,220 locally completely decomposable, 99 p-local,56 p-primary regulating quotient, 169 pregeneric, 123 preprojective , 20 I preinjective , 201 p-reduced, 52 projective in B(T), 197 quasi-generic, 113 reduced ,56 Richman-Butler, 205 self-cancellation, 68 self-small, 69 semilocal, 58 special, 115
strongly indecompo sable, 50
T- , 116 r-homogeneous,79 torsion -free, 47 uniform almost completely decomposable, 164 valuated p-, 218 algebraically independent elements , 73 antichain, 10 Boolean algebra, 84 Burkhardt invariants, 148 cancellation self,68 Cartan matrix, 177 category additive, 10 definition , 10 equivalence , 69 categories of abelian groups almost completely decomposable, 154, 155, 160 Butler, 100, 140 generically wild, 121 isomorphi sm at p , 51 Krull-Schmidt, 50 modulo p , 53 quasi-homomorphism, 47 torsion-free , 47 uniform almost completely decomposable, 164
242
Index of Terms
categories of representations of posets genericalIy trival,41 genericalIy wild, 43 over a discretevaluation ring, 126, 127 over a field, 1, 11,36 over a ring, 31 centralizer matrix,7 pair of matrices, 8 chain, 20 completion p-adic,211 l-adic,72 composition series of a module, 39 coordinate vector Butlergroup, 122 representation, 25 Comer's theorem,72 cosupport homomorphism, 90 Butlergroup, 90 cotrimmed n-tuple of groups,206 cotypeset critical, 83 definition, 82 Coxetercorrespondences, 179,201 dimension of representation, 21 direct sums abelian groupsat p, 51 objects in a category, 11 representations, 2, 12 quasi-,49 duality, 38 endolength of a module, 40 endomorphism ring representation of antichain, 1 representation of poser, 11 endowild, 33 equivalence relationson n+ r-equivalence,206 y-equivalence, 207 exact sequences abelian groupsin quasi-homomorphism category, 199 almostsplit, 187 balanced, 109 homomorphism, 187 isomorphism, 187 left almostsplit, 187 representations, 173 right almostsplit, 187 filter, 193 functor additive, 31 contravariant, 38 dense,69 faithful; 31
full,31 naturalequivalence, 69 preserves indecomposables, 31 reflects isomorphisms, 31 genusclass of a module, 60 Grothendieck group,67 height sequence, 77 equivalence, 77 p-,77 idempotents, I lift in a ring, 40 split in a category, 19 injective envelope of representation, 175 isomorphism of abelian groups at p,51 near-,56 quasi-,47 join irreducible elementsof a lattice, 100 k-algebra finite-dimensional, 8 Kleiner'sTheorem,23 KruII-Schmidt-Azumaya theorem for additive categories, 19 Krull-Schmidtcategory, 50 lengthof a module, 39 localization group at a set of primes,98 integers at p , 50 representation, 193 matrix centralizer, 7 equivalence, 9 indecomposable, 15 problems, 15, 128 similarity, 10 simultaneous equivalence of pairs, 10 simultaneous similarityof pairs, 10 meet-irreducible elementsof a latttice,200 module Butler, 122 copurelyindecomposable, 216 divisible, 211 endolength, 40 length,39 purelyindecomposable, 212 rank,71 reduced, 211 R-Iattice,60 simple,39 stronglyindecomposable, 122 torsion-free, 71
Index of Terms morphisms of abelian groups balanced epi-, 89 cobalanced mono-, 89 endo -,47 epi-,47 iso- at p, 51 modulo p, 53 mono-,41 quasi iso-, 49 valuated ,219 morphisms of representations definition , I, 11 endo-, I epi-, 173 injection s for direct sum, 12 irreducible, 191 iso-, 173 mono-, 173 projections for direct sum, 12 proper, 174 pure, 174 splitting epi-, 187 splitting rnono-, 187 Nakayama's lemma , 40 Nazarova's theorem, 33 nilpotent ideal, 40 p-Iocally free groups , 140 lattice of types, 140 polynomials content, 114 primitive, 114 Pontryagin 's lemma, 72 poset chain, 20 definition, 10 derivative, 27 forest, 160 garland, 131 opposite of, 38 sincere, 25 splitting decomposition, 23 sub,19 suitable pair, 27 tree, 219 V-free, 98 width,20 p-rank of module, 211 projective cover of representation, 175 quadratic form of poset, 177 quiver, 191 quotient field of polynomial ring, 41 radical Jacobson, 40 nil,59 T-,82
243
rank module, 71 representation, 127 representation of poset cotrivial, 174 definition, I , 11,31,36, 127 generic, 40 indecomposable, 2, II injective, 174 preinjective , 181 preprojective, 180 projective, 174 sincere, 25 trivial,2 representation of quiver, 191 representation type endowild,33 finite, 19 fully wild, 32 infinite, 19 rank-finite, 131 rank-infinite , 131 strongly wild, 37 tame, 37 wild,32 wild modulo p, 135 ring artinian, 40 discrete valuation, 126 incidence algebra of poset, 195 local, 11 maximal S-order, 59 one in the stable range, 63 Ore domain, 120 polynomi al in two noncommuting variables , 31 S-order,59 semiperfect , 40 semisimple, 40 totally definite quaternion, 66 two in the stable range, 65 root of tree, 219 socle T-,78 subcategory full,13 subgroup of an abelian group pure, 52 regulating, 97 regulator, 97 T -radical , 82 T-socle,78 submodule basic, 211 pure, 71, 126 substitution property definition , 63 one-,63 two-, 64
244
Index of Terms
summandof an abelian group at p,51 near, 56 quasi-,50 support element,90 Butler group, 90
inner,84 lattice, 76 outer,84 Richman, 88 typeset critical, 79 definition, 77
topology p-adic, 211 Z-adic,72 tree, 219 type definition, 76
up-set, 113 valuated tree definition, 219 equivalence, 219 v-height, 219