REPRESENTATION THEORY OF FINITE GROUPS AND ASSOCIATIVE ALGEBRAS CHARLES W. CURTIS UNIVERSITY OF OREGON IRVING REINER UNIVERSITY OF ILLINOIS INTERSCIENCE PUBLISHERS a division of John Wiley & Sons, New York + Lonpon + SYDNEY1009876 Copyright © 1962 by JOHN WILEY & SONS, INC. ALL RIGHTS RESERVED—Reproduction in whole or in part for any purpose of the United States Government will be per- mitted. Library of Congress Catalog Card Number 62-16994 ISBN 0 470 18975 4Preface Representation theory is the study of concrete realizations of the axiomatic systems of abstract algebra. It originated in the study of permutation groups, and algebras of matrices. The theory of group representations was developed in an astonishingly complete and useful form by Frobenius in the last two decades of the nineteenth century. Both Frobenius and Burnside realized that group repre- sentations were sure to play an important part in the theory of abstract finite groups. The first book to give a systematic account of representation theory appeared in 1911 (Burnside [4]) and contained many results on abstract groups which were proved using group characters. Perhaps the most famous of these is Burnside’s theorem that a finite group whose order has at most two distinct prime divisors must be solvable. Recently, a purely group-theoretic proof of Burnside’s theorem has been obtained by Thompson. The new proof is of course important for the structure theory of groups, but it is at least as complicated as the original proof by group characters. The second stage in the development of representation theory, initiated by E. Noether [1] in 1929, resulted in the absorption of the theory into the study of modules over rings and algebras. The representation theory of rings and algebras has led to new insights in the classical theory of semi-simple rings and to new investigations of rings with minimum condition centering around Nakayama’s theory of Frobenius algebras and quasi-Frobenius rings. Another major development in representation theory is R. Brauer’s work on modular representations of finite groups. Like the original work of Frobenius, Brauer’s theory has many significant applications to the theory of finite groups. At the same time it draws on the representation theory of algebras and suggests new problems on modules and rings with minimum condition. It also emphasizes the fundamental importance of number-theoretical questions in group theory and representation theory. During the past decade there has been increased emphasis on integral representations of groups and rings, motivated to some extent by questions arising from homological algebra. This theory of integral representations has been a fruitful source of problems vvi PREFACE and conjectures both in homological algebra and in the arithmetic of non-commutative rings. The purpose of this book is to give, in as self-contained a manner as possible, an up-to-date account of the representation theory of finite groups and associative rings and algebras. This book is not intended to be encyclopedic in nature, nor is it a historical listing of the entire theory. We have instead concentrated on what seem to us to be the most important and fruitful results and have included as much preliminary material as necessary for their proofs. In addition to the classical work given in Burnside’s book [4], we have paid particular attention to the theory of induced characters and induced representations, quasi-Frobenius rings and Frobenius algebras, integral representations, and the theory of modular rep- resentations. Much of this material has heretofore been available only in research articles. We have concentrated here on general methods and have built the theory solidly on the study of modules over rings with minimum condition. Enough examples and problems have been included, however, to help the research worker who needs to compute explicit representations for particular groups. We have included some applications of group representations to the structure theory of finite groups, but a definitive account of these applications lies outside the scope of this book. In Section 92 we have given a survey of the present literature dealing with these applications and have included in this book all the representation-theoretic prerequisites needed for reading this literature, though not all the purely group- theoretic background which might be necessary. No attempt has been made to orient the reader toward physical applications. For these we may refer the reader to recent books and articles dealing with that part of group theory relevant to physics, and in particular io Wigner [1], Gelfand-Sapiro [1], Lomont (J, and Boerner [1]. It has also been necessary to omit the vast literature on represen- tations of the symmetric group. Fortunately the reader is now able to consult the excellent book on this topic by Robinson [1]. Many of the results on group representations have been generalized to infinite groups and also to infinite-dimensional representations of topological groups. We have felt that these generalizations do not properly fall within the scope of this book and, in fact, would require a lengthy separate presentation. The book has been written in the form of a textbook; a preliminaryPREFACE vii version has been used in several courses. We have assumed that the reader is familiar with the following topics, which are usually treated in a ‘“‘standard’’ first-year graduate course in algebra: ele- mentary group theory, commutative rings, elementary number theory, rudiments of Galois theory, vector spaces, and linear transformations. We are confident that the expert as well as the student will find something of interest in this book. We offer no apology, however, for writing to be understood by a reader unfamiliar with the subject. In keeping with this objective, we have not always presented results in their greatest generality, and we have included details which will sometimes seem tedious to the experienced reader. After serious deliberation, we decided not to introduce the full machinery of homological algebra. Although it would have simplified several sections of the book, we felt that many readers were not likely to be well-grounded in homological algebra, and this book was not intended to be a first course in the subject. The first three chapters are written at the level of a first-year graduate course and include introductory material as well as back- ground for later chapters. Much of this material may be skimmed rapidly or omitted entirely at a first reading, though Sections 9-13 should be read with care. Chapters IV-VII form a unit containing the structure theory of semi-simple rings with minimum condition, and the applications of this theory to group representations and characters. Chapters IV, VIII, IX, and X form a unit on rings with minimum condition and finite-dimensional algebras. Chapter IV develops the theory of the radical and semi-simplicity by the perhaps old-fashioned method of calculations with idempotents, because idempotents furnish the main tool in the study of non-semi-simple rings and algebras in Chapters VIII and IX. Chapters III and XI form a more or less self-contained account of algebraic number theory and integral representations of groups. Some knowledge of earlier chapters is needed, especially in Sections 71-78. Chapter XII is devoted to the theory of modular representations and requires aknowledge of parts of all the preceding chapters. The exact prerequisites for reading Chapter XII are given at the beginning of the chapter. For the reader whose main interest is in representations of finite groups, we may suggest the following sections for a first brief reading: 9-13, 23-27, 30-34, 38-40, 43-46, 49-50, 54-55, 61, 82-92.viii PREFACE These sections are to some extent self-contained, provided that the reader is willing to postpone to the second reading the proofs of some of the results needed from other sections. Exercises are included at the end of almost every section. Some provide easy checks on the reader’s comprehension of the text; others are intended to challenge his abilities. Many are important results in their own right and may occasionally be referred to when needed in later sections. Sections are numbered consecutively throughout the book. A cross reference to (a.b) refers to Section a and to the bth numbered item in that section. There is a fairly large bibliography of works which are either directly relevant to the text or offer supplementary material of interest. An attempt has been made to give credit for some of the major methods and theorems, but we have stopped far short of trying to trace each theorem to its source. We are indebted to many persons and organizations for assisting us with this work. Our students, friends, colleagues, and families have listened to us lecture on these subjects, read portions of the manuscript and proof sheets, made suggestions and corrections, and given us encouragement. We are deeply appreciative of their kind help. Our interest in this subject was stimulated by a seminar conducted at the Institute for Advanced Study in 1954-1955. We are indebted to the participants in that seminar for their help and to the Institute for making possible the preparation of mimeographed seminar notes. It is a pleasure to acknowledge the generous support we have received for the work on this book from the Office of Naval Research. Finally we are grateful to Interscience Publishers for publishing it and giving us their patient and friendly cooperation. Charles W. Curtis Irving Reiner June 1962Contents Notation ........ 0.0. cece cece cece cece eee e ee en eect ee te ence cece ee Xiii I. Background from Group Theory ....................-.e0-0+2 1 1. Permutation Groups and Orbits aL 2. Subgroups and Factor Groups 3 3. Conjugate Classes .. 8 4. Abelian Groups .. 10 5. Solvable and Nilpotent Groups 4 6. Sylow Subgroups...... 17 7. Semi-direct Products .. 21 II. Representations and Modules .. 25 . Linear Transformations .. 26 9. Definitions and Examples of Representations 30 10. Representations of Groups and Algebras 38 “1. Modules ........ 50 12. Tensor Products 59 8. Composition Series 76 14. Indecomposable Modules .. 81 15. Completely Reducible Modules .. 86 Ill. Algebraic Number Theory .. 7 i 91 1. Modules over Principal Ideal ‘Domains 91 17. Algebraic Integers .. 18. Ideals ........ 19. Valuations; Padic Numbers . 20. Norms of Ideals; Ideal Classes .. 21. Cyclotomic Fields ......... 22. Modules over Dedekind Domains... IV. Semi-simple Rings and Group Algebras 157 23. Preliminary Remarks Te 157 24. The Radical of a Ring with Minimum Condition . 159 25. Semi-simple Rings and Completely Reducible Modules 163 26. The Structure of Simple Rings wee ae aA 27. Theorems of Burnside, Frobenius, and Schur : 28. Irreducible Representations of the Symmetric Group .. 190 ix29. Y. Group Characters 31. 36. 37. VI. Induced Characters 38. 39. 40. 41. 42. VII. Induced Representations ...... 43. 44. 45. 46. 47. 48. 49, 50. 51. 52. 53. Vil. 54. 55. 56. CONTENTS Extension of the Ground Field ...................002. 198 . 207 « 207 . 217 . 224 Introduction .. Orthogonality Relations .. Simple Applications of the Orthogonality Relation: Central Idempotents ...... Burnside’s Criterion for Solvable ‘Groups 7 The Frobenius-Wielandt theorem on the Existence of Normal Subgroups in a Group . -. 241 Theorems of Jordan, Burnside, and Schur 0 on , Linear Groups.. - cece eeee tees ee enceeeteeeeees 200 Units in a a Group Ring... Wal leleteleltolptelabelafeltslabeateste stem - 265 - 265 - 279 Introduction ...... Rational Characters .. Brauer’s Theorem on Induced Characters . Applications .. aie . 292 The Generalized Induction Theorem. co - 301 ... 313 Induced Representations and Modules .. 314 The Tensor Product Theorem and the Intertwining Number Theorem .. . Irreducibility and Equivalence of Induced Modules .... 328 Examples: The Tetrahedral and Octahedral Groups .. 329 Applications: Representations of Metacyclic Groups .. 333 A Second Application: Multiplicity-free Representations ..... «+. 340 The Restriction of Trreducible. Modules to. Normal Subgroups .. Imprimitive Modules... Projective Representations Applications . Schur’s Theory of Projective Representations Non-Semi-Simple Rings 5 aA Principal Indecomposable Modules The Classification of the Principal Indecomposable Modules into Blocks .. Projective Modules... 1) 367CONTENTS xi 57. Injective Modules ............ cece cece cece ence ence ee ee 58. Quasi-Frobenius Rings . 59. Modules over Quasi- “Frobenius Rings IX. Frobenius Algebras ....... -. 409 60. Injective Modules for a Finite. Dimensional “Algebra -. 409 61. Frobenius and Quasi-Frobenius Algebras .............. 413 62. Projective and Injective Modules for a Frobenius Algebra .. - 420 63. Relative Projective an ~ 426 64. Group Algebras of Finite Representation Type . 1 65. The Vertex and Source of an Indecomposable Module 435 66. Centralizers of Modules over Symmetric Algebras . 440 67. Irreducible Tensor Representations of GL(V)..... X. Splitting Fields and Separable Algebras ........ wee. 453 68. Splitting Fields for Simple Algebras and Division Algebras .. 69. Separable Extensions “of ‘the ‘Base Field 459 70. The Schur Index.. 71. Separable Algebras.. 72, The Wedderburn-Malcev Theorem .. XI. Integral Representations. . 493 73. Introduction .. - 494 74. The Cyclic Group of Prime Order . 506 75. Modules over Orders.... . 515 76. P-Integral Equivalence .. . 531 77. Projective Modules: Loca! eory . 542 78. Projective Modules: Global Theory 550 79. The Jordan-Zassenhaus Theorem ...... 80. Order Ideals 81. Genus ...... XII. Modular Representations .. BRSRRSE Introduction . Cartan Invariants and Decomposition Numbers Orthogonality Relations .. . 598 Blocks . - 604 The Defect of a - 611 Defect Groups Block Theory for Groups with Normal P-Subgroups .. 627xii CONTENTS 89. Block Distribution of Classes. 90. Miscellaneous Topics...... A. Generalized Decomposition Numbers B. Conjugate Characters C. The Number of Characters Belonging to a Bloc! D. Numerical Bounds .. 91. Examples . 92. Literature on Applications to A. Groups of a Given Order ............ B. Characterizations of Simple Groups me C. Criteria for Existence of Normal Subgroups ...... 654 Bibliography .......... 00... c cece cece cece eee ee tt ee ee eeeeeece ee 655Notation Zz = ring of rational integers Q = rational field a|b : @ divides 6 where a,beZ akb : @ does not divide b, (a,be€Z) PP" \la : p"|a@ but p**' + @ where p is prime and aeZ G.C.D. : greatest common divisor L.C.M. : least common multiple Group theory notation [G:1] = number of elements in G [G: H] = number of distinct left cosets of H in G HAG : H is a normal subgroup of G [x] = cyclic group generated by x Sx = symmetric group on m symbols P(X) = group of all permutations of a set X C(G) =center of the group G C(x) =centralizer of x in G Co(H) centralizer of H in G N(A) =normalizer H in G A(G) = group of automorphisms of G IG) group of inner automorphisms of G G. x G; = direct product of G, and Gz a, : denotes the left multiplication: x— ax G,=G, : G, is isomorphic to G; [G, G] = commutator subgroup of G Notations from field theory char K = characteristic of the field K Irr (a, K) = minimal polynomial of a over K alg. int. {K} =ring of all algebraic integers in K Notations from module theory and linear algebra rR = left regular R-module M+N : external direct sum xiiixiv NOTATION MON : internal direct sum ZOM : direct sum A.C.C. : ascending chain condition D.C.C. : descending chain condition S\N : restriction of f to the subset N of the domain of f Hom,(M, N) \dditive group of R-homomorphisms of M into N Da = ring of all 2 x matrices with entries in D (M: K) = dimension of M over K xX = transpose of the matrix X |X| = det X = determinant of the matrix X 0 I diag {a,, diagonal matrix with diagonal entries ai, ---, a» GL,(K) roup of invertible matrices in K,, K = field GL(M) = group of invertible elements in Homs(M, M), M= vector space over field K M®@rN = tensor product of right R-module M and left R- module N TxXU = Kronecker product of matrices Me Te = induced representationCHAPTER I Background from Group Theory We presuppose a knowledge of elementary group theory, such as that which may be obtained from reading introductory material in any of the following references: M. Hall [2], Kurosh [1], Leder- mann [1], or Speiser [2]. In this chapter, some purely group-theo- retical results are collected which will serve to motivate the later discussion, to suggest problems which the theory of group represen- tations might hope to solve, and to develop concepts and theorems needed for the later chapters. §1. Permutation Groups and Orbits A permutation of a set X is a one-to-one mapping of X onto it- self. As is well known, the set of all permutations of X forms a group P(X), in which the product or of a pair of permutations 0,7 is defined by (or)x = o(cx), xexX, If X contains more than two elements, P(X) is not commutative. Any subgroup of P(X) is called a permutation group on X, or a group of permutations of X. We shall say that the permutations in P(X) act or operate on the elements of X. A permutation group G on X gives rise to a partitioning of X into disjoint subsets. The importance of this simple idea for mathe- matics can scarcely be overstated. We begin by defining an equiva- lence relation in X as follows: We say that x is G-equivalent to y and write x~ y, provided that ox=y for some o¢€G. It is easily verified that G-equivalence is indeed an equivalence relation. The equivalence classes of X under this relation are called the orbits in X relative to G. These orbits are disjoint subsets of X whose union is X. Thus x and y belong to the same orbit if 12 GROUP REPRESENTATIONS §1.1 and only if ox =y for some o€G. If there is only one orbit in X relative to G, we say that G is transitive on X. Clearly, P(X) acts transitively on X; it is easy to see by an example that proper sub- groups of P(X) may also act transitively on X. To get some geometric examples of orbits, the reader may con- sider the set X of all points in the complex plane. If, on the one hand, G is the group of all rotations about the origin, the orbits in X relative to G are the concentric circles about the origin. If, on the other hand, w is a fixed non-zero vector and G is taken to be the set of all translations x>x+am, a real, the orbit containing a complex number x consists of all points on the line through x parallel to uo. Given a permutation group G on X, an equally important concept is that of invariance relative to G. A subset Y of X is called in- variant relative to G if, for eacho €G,o(Y)c Y. Anelement xe X is invariant relative to G if and only if the orbit of x contains only x; an orbit consisting of a single element is called trivial. As a first application of the concept of orbits, consider the symmetric group S, defined as the group of all permutations of the set X={1,2,---,m}. Let [x] denote the cyclic group generated by an element z€S,. We call za cycle if X has only one non-trivial orbit relative to {z]. Each cycle xz cyclically permutes the elements in its non-trivial orbit; hence it may be written as ant, y) where g is the smallest positive integer such that x*=1. Two cycles m, 7: € S, are called disjoint if their non-trivial orbits are disjoint. It is easily seen that disjoint cycles commute with each other. Using this fact, we show m= (y my myer (1.1) Toeorem. Every permutation o € S,,0 #1, is expressible as @ product of disjoint cycles. This expression is unique up to order of occurrence of the factors. Proor. Let Xi,--:, Xm be the distinct orbits of [co]. Define for each i, 1 xa, xeG. Then a—ar, a€G, gives an anti-isomorphism of G onto the sub- group Gr of P(G). We note also that azbe = bear, a,beG. Now let H be a subgroup of G, and let H; and Hp be the sets of left and right multiplications determined by the elements of H. (2.1) Derinition. The orbits of G relative to H; are called right cosets of H in G, those relative to Hr, left cosets. In order to determine the cosets more explicitly, it is convenient4 GROUP REPRESENTATIONS §2.2 to define multiplication of subsets A and B of G by AB = {ab:ae A,be B}. Likewise, define At={a':aeé A}. Now let x €G; the orbit of G relative to Hp containing x is then xH. Similarly, the right coset containing x is Hx. Since cosets are orbits, any two left cosets either are disjoint or coincide. If xH and yH are left cosets, the equation (yx")H = yH shows that xH and yH have the same cardinal number. If G is a finite group, we can decompose G into a union of disjoint left cosets, say, (2.2) G=mHU mH +: UxH. The number 7 of distinct left cosets of H in G is called the index of H in G and is denoted by [G: H]. In keeping with this notation, we use [G:1] to denote the number of elements in G. From (2.2) we deduce Lagrange’s theorem: (2.3) [G:1) = (G: HJ: 1). The one-to-one mapping g—9"', g € G, carries the left coset xH onto the right coset Hx' and effects a one-to-one transformation of the collection of left cosets onto the collection of right cosets. There- fore [G: H] is also the number of distinct right cosets of H in G. More generally, let H and K be a pair of subgroups of G. Because hike = kehz, heH,kek, it follows that H,Kp is a subgroup of P(G). The orbits of G rela- tive to H,Kpr are called the (H, K)-double cosets in G. Being orbits, distinct double cosets are disjoint. The (H, K)-double coset contain- ing x is just HxK. A finite group G also has a decomposition into disjoint double cosets, say, G=HxKvU --- UHnK. However, as we shall see, different double cosets may have different cardinal numbers. For example, let G=S, H=({1,(12}, K={1,(13)}.§2.4 BACKGROUND FROM GROUP THEORY 5 Then the (H, K)-double cosets in G are #H-1-K= 1, (12), (13), (132)},, H- (23) - K = {(23), (123)} . In general, the number of double cosets need not divide [G: 1]. As our next application of orbits and invariance, we shall con- sider the automorphisms of a group G. An automorphism of G is an element a € P(G) such that a(xy) = aXx)aty) , myeG. Thus @ is an isomorphism of G onto G. The set of all automor- phisms forms a subgroup A(G) of P(G). Contained in A(G) are the inner automorphisms {i.,a € G}, defined by igi x axa’, xeG. We have ids in, ta =i, a,beG, which shows that the set J(G) of all inner automorphisms of G isa subgroup of A(G). (2.4) Derinition. A subgroup H of G which is invariant relative to I(G) is called a normal subgroup of G (notation: H A G). In other words; H is normal in G if and only if aHa'cH for all aeG. This assertion is easily seen to be equivalent to the statement aHa'=H, aeG, and this, in turn, to the important relation aH=Ha, aeG. Thus H is normal in G if and only if every right coset is a left coset, and vice versa. If H is a normal subgroup of G, we have (4H)(yH) = xyH, xyeG, and it follows that, relative to set multiplication, the cosets of H in G form a group. This group is called the factor group of G over H and is denoted by G/H. If G/H is a finite group, the number of elements in it is [G: H], the index of HinG. _If G is a finite group we have at once from (2.3)6 GROUP REPRESENTATIONS §2.5 (G/H: 1] = [G: H] = [G: 1/[#: 1). We recall that a homomorphism of a group G into a group G’ is a mapping f:G-—G" such that F (49) =FO)F(Y) xyeG. Because of the manner in which multiplication in a factor group G/H is defined, it is clear that the mapping x— xH of G onto G/H is a homomorphism, called the matural or canonical homomorphism of G onto G/H, and that the normal subgroup H can be characterized as the set of all elements of G mapped onto the identity element of G/H under this homomorphism. The next theorem asserts, among other things, that every homo- morphism arises in this way. (2.5) THzorem (Fundamental Theorem on Homomorphisms). Let f:G—G' be a homomorphism of G onto a group G'. Then H={xeG: f(x) =} is a normal subgroup of G called the kernel of f. The mapping xH— f(x) is an isomorphism of G/Honto G'. There is a one-to-one inclusion-pre- serving correspondence between the set of all subgroups K' of G’ and the subgroups K of G containing H, given by K>f(K)=K', K=f\(K'). Moreover, K A G if and only if K' AG’. If K AG, we have (2.6) GIK = G'/K' = (G/H)(K/H) . We assume that this result is familiar to the reader and omit the proof. (2.7) Dermnition. The center of the group G is the subgroup CG) ={x¢G:xa=ax for all aeG}. As an immediate application of this definition, we may observe that the mapping aia, aecG, is a homomorphism of G onto KG), with kernel C(G). From the fundamental homomorphism theorem, we have GIC(G) = KG).§2.8 BACKGROUND FROM GROUP THEORY 7 Now let H be a normal subgroup of G, and let K be any sub- group of G. Since, for he H and ke K, kh=Wk for some h’' eH, it follows that HK is also a subgroup of G. (Indeed, if also K AG, then HK AG.) The mapping k->kH is a homomorphism of K onto HK/H with kernel Kn H, and so 2.8) KK 0 H) = HKIH. We shall use this basic isomorphism theorem repeatedly. Exercises 1, For a€ A(G) and a€G, prove that aig: a = tga. Deduce from this that I(G) A A(G). In Exercises 2.2 to 2.4, G is a transitive permutation group on the set X = {a1, +++, tn} and H= {g€G; gui =a}. 2, Prove the existence of elements gi, 1sSisn, and such that *,gn€G@ such that gi(a:) = 24, G=giHU -+: UgnH gives a coset decomposition of G. 3. The subgroup H also acts on the set X. Show that the number of orbits of X under H is the same as the number of (H, H)-double cosets in G. [Hint: Suppose that x; and 2, are in the same orbit with respect to H; then 2; = ha, for some hE H and gjthgi:2: > a where the {ge} are as in Exercise 2.2. Then g;!hg:€H, and so HoH = Hg;H. The argument is reversible. Finally, every (H, H)-double coset in G is of the form HgiH for some i.] 4. Assume that G is doubly transitive on X, so that, for each two pairs {xi,aj}(i # 9) and {xx,2.}(k #1), there exists an element g¢€G such that 9%i = Zz, g2)= 2. Prove that H acts transitively on {22, «++, tn}- 5. Let HAG, and set i=[G: H],j =[H:1]. If i and j are relatively prime, show that H is the only subgroup of G of order j. [Hint: Let S be a subgroup of G of order j. Then HS/H is a subgroup of G/H whose order divides i, Also HS/H = S/(S H) implies that the order of HS/H divides j. Therefore HS = H.] 6. A characteristic subgroup H of G is a subgroup for which aH = H8 GROUP REPRESENTATIONS §3.1 for each a€ A(G). Show that every characteristic subgroup of G is a normal subgroup of G. More generally, show that H characteristic in Z and EAG together imply HAG. 7. Let n be fixed, and set S= {x€G:2"=1}. If S isa subgroup of G, then S is a characteristic subgroup of G. § 3. Conjugate Classes Still another important application of the orbit concept is the following, where KG) again denotes the group of inner automorphisms of the group G: (3.1) Dermition. The orbits of G relative to [(G) are called the conjugate classes of G. Elements in the same conjugate class are called conjugates of one another. Being orbits, two conjugate classes either are disjoint or coincide. In order to determine the number of elements in the conjugate class containing x, we first introduce (3.2) Derinirion. The centralizer of x in G is the subgroup C(x) = {a € G: axa"! = x}. Of course x € C(x). Now we note that axa = bxb™' if and only if a ¢ 6- C(x). Thus, the number of distinct conjugates of x is the same as the number of left cosets of C(x) in G. Hence (3.3) If x€G, there are [G: C(x)] elements in the conjugate class containing x. In particular, x coincides with all its conjugates if and only if xe€C(G). Since every element of G lies in exactly one conjugate class, we may write the class equation (3.4) [G: 1] = (CG): 1] + SIG: C~@)] where 3, extends over a set of representatives {x} of the conjugate classes of G having more than one element. Let p be a prime. Call G a p-group if [G:1] is a positive power of p. We may use the class equation to prove (3.5) Toeorem. The center of a p-group contains more than one element. Proor. Let [G:1]=p",n 21. If the conjugate class of x con-§3.6 BACKGROUND FROM GROUP THEORY 9 tains more than one element, then C(x) + G and so p|[G: C(x]. The class equation then implies that p|[C(G): 1]. An example that is of importance to us is the determination of the conjugate classes of S,. If (y oy oty +++ oy) is a cycle in S,, we leave it as an exercise to the reader to show that, for any z € S,, o non = (ny nay) x(o*y) --+ x(o%*y)) is also a cycle in S, of the same length as s. Now letr eS, be written as a product of disjoint cycles (3.6) THO Oy where we put in “cycles” of length 1. Then nen = (non) +++ (eo,n7") gives the analogous decomposition of zrz'. This establishes (3.7) The cycle factorization of xex™' is gotten from that of t by letting x act on the digits in the cycle representation of t. For example, if tv = (123)(45) and x = (12)(34) , then men = (214)(35) . A partition of n is an ordered set of integers {m;:} satisfying m+mt+-::-+m,=n, m2m2---2m,>0. Each r € S, gives rise to a partition of » as follows: write r in the form of (3.6) in which the {o;} are disjoint cycles of lengths m1, Mz, +++, m,, arranged in order of decreasing length. Then (3.7) shows that each conjugate of t yields the same partition of n as t does. Conversely, let yield the same partition of m as r does. Then, for each i, a; and o{ have the same length, say Oo = (Yeo Ymg) OE = (RrB2 ++ my)» For each i, define x on the orbit of [oi] by (y1) = 21, +++, 2( Ym) = Zmge10 GROUP REPRESENTATIONS §3.8 Then z€S,, and arz*=r'. Consequently, (3.8) There is a one-to-one correspondence between conjugate classes in S, and partitions of n. As an analogue of (3.1), we have (3.9) Derinition. Let H be a subgroup of G, and let ae G. We call the subgroup aHa™' a conjugate of H. Define the normalizer of H in G by N(H) = {a € G:aHa™' = H}. This concept of conjugates does not arise from orbits, and so it is possible that distinct conjugates of H will overlap; this is indeed certain to occur since all conjugates of H contain 1. However, we note that the number of distinct conjugates of H is given by the index [G: N(H)]. Further, H AG if and only if G = N(H) and, in particular, H A N(H). It is also convenient to generalize the concept of centralizer as follows: (3.10) Derinition. Let H be a subgroup of G. Define the cen- tralizer of H in G by Ca(H) = {a € G:aha'=h for all he H}. Thus C.(G) is just the center of G. Exercises 1. Let HAG, and let C be a conjugate class in G. Prove that, if CnH# 4%, then CCH. 2. Prove that Co(H) A NH). 3. Let H, K be subgroups of G. The subgroups H and kHk-!, ke K, are called K-conjugates. Prove that the number of distinct K-conjugates of H is exactly [K: Kn N(H)]. §4. Abelian Groups Valuable insight toward most problems in group theory comes by testing these problems on finite abelian groups. This step may be essential for groups which, although they fail to be abelian, are built up in some simple way from abelian groups. In this section we recall some of the fundamental properties of finite abelian groups.§4.1 BACKGROUND FROM GROUP THEORY abe For the most part, these properties are special cases of theorems on Zmodules which we shall prove at the beginning of Chapter III. Let Mi, ---, Hm be a finite collection of groups. By the (external) direct product H, x --- x Hm is meant the group consisting of all m-tuples (M1, ---, hm), ti € Ai, +++, Am € Hm, with multiplication defined by (ay +++ Tim) = ti, ++ +5 Him) = (Uerhi, +++, Itmdiin) « [If the groups are written additively, we refer instead to the (ex- ternal) direct sum H, + ++» + Hn.J Suppose H and K are both subgroups of some group G such that hk =kh for all he Hand ke K. Then, setting HK = {hk:h ¢ H,k €K}, we see that HK is also a subgroup of G. If furthermore Hn K= {1}, it follows that HK=HxK. We shall then call HK the (internal) direct product of H and K in G. For example, suppose H and K are subgroups such that every element of H commutes with every element of K. Suppose also that the orders [H: 1] and [K: 1] are relatively prime. Then Hn K = {1} since the order of any element in Hn K must divide both [H: 1] and [K:1]. By the preceding remarks, we have HK=HxK. We give an immediate application of this fact. Let G be a finite abelian group of order g. For each prime p, let Gp be the set of elements in G whose orders are powers of p. (Of course G, = {1}, if p ¥ 9). Then G, is a subgroup of G, called the p-primary com- ponent of G. We conclude at once that (4.12) G=TILG (direct product) . vig We also note that [G,: 1] = power of p occurring in g. The decom- position (4.1) is absolutely unique in the sense that, if also G=TI®, a direct product of a finite number of subgroups H, whose orders are distinct prime powers, then the {H,} are the p-primary com- ponents of G.12 GROUP REPRESENTATIONS §4.2 For many purposes we require a finer decomposition of finite abelian groups. We first recall that a group G is indecomposable if it admits no non-trivial decomposition G = G, x G;. We now state without proof (4.2) Evementary Divison Tueorem. (i) A finite abelian group is indecomposable if and only if it is cyclic of prime power order. (ii) Every finite abelian group G # {1} is a direct product of indecomposable groups (4.3) G=G,x G:x +++ X Gn where each G; is cyclic of prime power order a;,a;>0. The col- lection of orders { p:"', ---, px™} constitutes the elementary divisors of the decomposition (4.3).t (iii) Any two decompositions of G into indecomposable groups have the same set of elementary divisors. By (iii) we are entitled to speak of the elementary divisors of G. Clearly, two finite abelian groups are isomorphic if and only if they have the same set of elementary divisors. We next require the following lemma: (4.4) Lemma. A direct product H, x --- x H, of cyclic groups, whose orders {hi} are powers of distinct primes, is cyclic. Conversely, any cyclic group is so expressible. Proor. We need only use the fact that, if x and y commute, the order of xy is the L.C. M. (least common multiple) of the order of x and the order of y. We now state the second main result on finite abelian groups. (4.5) Invariant Factor Tueorem. Every finite abelian group G is a direct product (4.6) G=H,x H.x-:: x H, of non-trivial cyclic groups {H:}, whose orders {hi} have the pro- perty that hi\hinn, 1 Sisr—1. If also G= Kix --- x K, where the {K;} are non-trivial cyclic groups of orders {k:} such that k;| Riss, lsiss—l, thnr=s andh=k,1sSisr. We shall prove in this section only that Theorems 4.2 and 4.5 are equivalent. A proof of Theorem 4.5 will be given in § 1! Start with the decomposition (4.3). For each prime p; dividing [G: 1], let pi* be the highest power of ~; which occurs among the t The primes {pi} are not necessarily distinct.§4.6 BACKGROUND FROM GROUP THEORY 13 elementary divisors of G. Then, for each i, some one of the groups on the right-hand side of (4.3) has the order pf, say, the group Gm, Set H, =T[ Ga, . By (4.4), H, is cyclic of order IIpf‘. Apply the same construction to the remaining {G;}, obtaining a cyclic direct factor H,-1 of order h,-1, with h,-1|h,. Continuing in this way, we arrive at the decom- position (4.6). To prove uniqueness, we note that by (4.4) each K; is a direct product of cyclic groups whose orders are powers of distinct primes. By the elementary divisor theorem, the collection of all such prime powers (over all indices i) are the elementary divisors of G. Because ki | Riss, it follows that & is the L.C.M. of the elementary divisors, k,-. is the L.C.M. of the remaining ones after the highest power of each prime is removed, and so on. This proves that Theorem 4.2 implies Theorem 4.5. The proof that Theorem 4.5 implies Theorem 4.2 is straight- forward and will be left to the reader. Exercises 1. Let G(n) be the multiplicative group consisting of those residue classes in Z/nZ whose elements are relatively prime to n. Show that [G(n): 1] = ¢(n) where g is the Euler g-function. Further, if m = [up3¢ where the {pi} are distinct primes, then G(n) = Tl G(p%) (direct product) . For p an odd prime, G(p*) is cyclic, and Gp) = Gi x Ge, [G:1J=p-1, [Gel)=pe'. On the other hand, G(22) is not cyclic for a > 2, and G2) =Gsx Gi, [G1] =2, [Gel] = 22-7, where G, is cyclic. (See Hecke [1].) 2, Let A and B be groups whose orders are relatively prime. Prove that every subgroup of A x B is of the form A, x Bi for some subgroups Ai of A and Bi of B. [Hint: Let H be a subgroup of A x B, and write he H as h=ab. Then (ab)[2:1] € H implies that al®1]¢ H. Since also af4#] € H, we have a€ H. Thus ab € H if and only if both a and b belong to H.]14 GROUP REPRESENTATIONS §5.1 §5. Solvable and Nilpotent Groups For no other general class of groups do we have structure theorems as definitive as those just stated for commutative groups. There are nevertheless important classes of groups more general than the commutative ones, about which our knowledge is more penetrating than in the general case. We begin with (8.1) Derinition. A normal series for a group G is a chain of sub- groups (6.2) G=G,2G,5---3G,={1} in which GiiAG;,,1SisSr—1. The factors of the normal series are the factor groups G,/G:, ---,G,-1/G,. We say that G is solvable if G has a normal series in which all of the factor groups are abelian. Some basic properties of solvable groups are given in thesollow- ing, which we state without proof: (5.3) Toeorem (i) Subgroups of solvable groups are solvable. (ii) Homomorphic images of solvable groups are solvable. Giii) If HAG is such that both H and G/H are solvable, then G is also solvable. For example, the alternating group A, is not solvable when nz25. On the other hand, every abelian group is solvable. Let us show, by induction on the order of G, that every p-group G is solvable. By (3.5), the center C(G) of G is non-trivial. If C(G) =G, then G is abelian and hence solvable. If not, then C(G) and G/C(G) are p-groups of orders less than [G:1] and so are solvable by the induction hypothesis. From (iii) of Theorem 5.3, we conclude that G is also solvable. Burnside proved that if [G:1] = p%q°, where p and q are primes, then Gis solvable. In Chapter Vwe shall give a fairly simple proof of this based on the theory of group characters. A purely group-theore- tical, but more complicated, proof has recently been found by Thomp- son for the case where p and q are both odd. Let HAG; we call H a maximal normal subgroup of G if H is a proper subgroup of G not properly contained in any proper normal subgroup of G. A composition series for G is a series (5.2) in which for 1Si [G, G]. (5.9) Derinition. Let G’ = [G, G], G’’ = [G', G’], and soon. The sequence G2G'DG"5a.-:- is called the derived series of G. (6.10) THeoreM. A group G is solvable if and only if G™ = {1} for some n. Proor. Jf G” = {1}, then G2G'DG">.-- 3G” ={1} is a normal series for G with abelian factors, and so G is solvable. Conversely, let G be solvable, and let G=G,2G,2-::2G,={1} be a normal series with abelian factors. A simple induction argument establishes that G“” c Gis: for each i, whence G’-” = {1}. (6.11) Tueorem. Every nilpotent group is solvable. Proor. Let G be nilpotent, and let (y=McoBAc+:+cZ=G be a central series terminating in G. Since Z,/Z,-1 is the center of G/Z,-1, it follows that G/Z,-1 is abelian and G’ c Z,-,._ An induction argument yields G” c Z,-; for each i, and hence G” = {1}. From (5.10), it follows that G is solvable. The converse of the above theorem is false. For example, the symmetric group S; is solvable, but its central series never reaches past the subgroup {1}. This example also show that the analogue of (iii) of Theorem 5.3 is false for nilpotent groups. Specifically, if HAG is such that H and G/H are both nilpotent, it is false, in general, to conclude that G is nilpotent. However, it is true that subgroups and factor groups of nilpotent groups are nilpotent, though we shall not prove these facts here. Exercises 1. If A,B are normal subgroups of a finite group G such that [A: 1] and§6.1 BACKGROUND FROM GROUP THEORY 17 [B: 1] are relatively prime, show that AN B = {1}. Ifa¢ Aand be B, prove that aba-'b-'€ AN B and hence that ab=ba, a€ A, bE B. Consequently AB=AXB. 2. Let H be a normal subgroup of a finite group G. The set of composi- tion factors of G is the union of the set of composition factors of H with that of G/H. §6. Sylow Subgroups In §4 we saw that, for each prime p dividing the order of a commutative group G, there exists a subgroup G, of G whose index is prime to p. Sylow’s remarkable theorem, proved in 1873, asserts that this result is true for arbitrary finite groups. To prove the theorem, we begin with a preliminary result. (6.1) Let G be an abelian group of order g, and let p|g, p prime. Then G contains an element of order p. Proor. Let G, be the p-primary component of G. Since [G,: 1] is the power of p occurring in g, we conclude that G, contains an element x #1. If the order of x is p", 2 21, then x**"' is an element of order p in G. (6.2) Derinirion. Let G bea finite group and p a prime. A sub- group P of G is a p-Sylow subgroup of G if P is a p-group such that p¢(G:P]. Alternately, P is a p-Sylow subgroup of G if and only if the order of P is the exact power of p which divides [G: 1]. We shall now prove the first of Sylow’s theorems. (6.3) Tazorem. Let G be a finite group and pa prime. Then G contains a p-Sylow subgroup. Proor. Use induction on the order of G, and suppose p* || [G: 1]. If G contains a proper subgroup H whose index is prime to p, then also p*||[H:1]. By the induction hypothesis, H has a subgroup P of order p*, and then P is also a p-Sylow subgroup of G. For the remainder of the proof, we may assume that p|[G: H] for each proper subgroup H of G. Now consider the class equation (3.4) [G: 1] = [CG):1] + YS [G:C@)] oiee where x ranges over the representatives of the conjugate classes having more than one element. Then p divides each summand in18 GROUP REPRESENTATIONS §6.4 the summation and also p|[G:1] (otherwise the result is trivial), so that p|[C(G):1]. By (6.1), it follows that C(G) contains an element x of order p. The cyclic subgroup [x] is normal in G, and, by the induction hypothesis, G/[x] has a p-Sylow subgroup. From (2.5) we conclude that G contains a subgroup P (containing [x]) such that the order of P/[x] is p*"'. But P then has order p* and is a p-Sylow subgroup of G. In order to establish further properties of Sylow groups, we have to look in a more sophisticated way at the ideas used in the develop- ment of the class equation of G. Let P be a p-Sylow subgroup of G where p* ||[G:1],2 21. If His any p-subgroup of G (not necessarily a Sylow subgroup of G), we call the subgroups P and hPh™"',h e H, H-conjugates of one another (see Exercise 3.3). If N(P) denotes, as usual, the normalizer of P in G, there are [H: Hn N(P)] distinct H-conjugates of P. We now show (6.4) Lemma. Hn N(P)=HANP. Proor. Since PC N(P), we have (HN P)C(HnNN(P)). To prove the converse, let Hi = Hn N(P). Since H; is a subgroup of the p-group H, also H, is a p-group. But, on the other hand, PA N(P), and H, is a subgroup of N(P), so that, by (2.8), we have [H.P: P] = [M: H, n P] = power of p, the latter because H; is a p-group. Hence H,P is a p-group contain- ing P, and, since P is a maximal p-subgroup of G, H,P= P. There- fore H, ¢(Hn P), which proves our lemma. Now consider all the conjugate subgroups P, = xPx"' of P,x€G. All of them are p-Sylow subgroups of G, and their number is [G: N(P)] which is prime to p. Let H be a fixed p-subgroup of G and divide the conjugate subgroups {P,} into classes of H-conjugates. The number of subgroups P, which are H-conjugates of P, is just [by Exercise 3.3 and (6.4)] [H: H 0 N(P:)) = [H: Ho Pi). Therefore we obtain a formula (6.5) [G: N(P)] = SH Hn P,] where the sum extends over a set of representatives {P.} of the classes of H-conjugates of P. Since H is a p-group, all the terms on the right-hand side of (6.5) are powers of p, whereas [G: N(P)] is prime to p. We conclude that, for some P.,§6.6 BACKGROUND FROM GROUP THEORY 19 (4: Hn P.)=1, that is, H= Hn P,, and so Hc P,=xPx". This shows (6.6) Every p-subgroup of G lies in some conjugate of the p-Sylow subgroup P of G. As an immediate consequence, we deduce (6.7) Tuzorem. Any two p-Sylow subgroups of a finite group G are conjugate. The total number of distinct p-Sylow subgroups is [G: N(P)], where P is any p-Sylow subgroup of G. If in (6.5) we put H=P, then [H: Hn P,] is a positive power of p except when x=1. This establishes (6.8) The number of distinct p-Sylow subgroups of G is =1 (mod p). From (6.7), it follows readily that P is the only p-Sylow sub- group in its normalizer N(P). Now let x¢€G satisfy xN(P)x' = N(P); then xPx"' is a p-Sylow subgroup of N(P) and hence coincides with P. This shows that xe N(P). Thus (6.9) If P is a p-Sylow subgroup of G, the normalizer of N(P) concides with N(P). For later use we shall need (6.10) Each non-cyclic p-group G has af actor group of type AX A where A is cyclic of order p. Proor. We shall use induction on the order of G and remark first that the result is an immediate consequence of the elementary divisor theorem when G is abelian. Assume thus that the center C of G is different from G, and let yy under the canonical homo- morphism ¢ of G onto G=G/C. If G is non-cyclic, it contains a normal subgroup Af such that G/H = Ax A. Choosing H=¢"'(H), we have [using (2.6)] GIH=GH=Ax A. Finally, if G = [¥] is cyclic, every element of G is expressible as a power of y times an element of C, and it is immediate that G is abelian. The result is known in this case, and (6.10) is proved. In §4 we proved that a finite abelian group is the direct product its p-Sylow subgroups (which are the p-primary components) for the different primes p dividing the group order. Although this statement is not true for arbitrary finite groups, it is true for nilpotent groups.20 GROUP REPRESENTATIONS §6.12 To show this, we need a preliminary result. (6.11) Let H be a proper subgroup of the nilpotent groupG. Then H properly contained in its normalizer N(H). Proor. Let (P=McAc---cZ=G be a central series for G, and let H be a proper subgroup of G. Then Zc H but Z,¢ H, so we may find an index i such that ZcH, ZntH. Since Z;,+:/Z; = center of G/Z;, we have gog'v é Z; for vé Zin, 9 EG. In particular, setting gy = € H and using the fact that Z;c H, we have vh'v' eH for vé Zun,he H. Hence Zin N(H), so that surely H # N(H). We may now prove (6.12) Tuzorem. Let G be a finite nilpotent group. Then each p- Sylow subgroup of G is normal in G, and G is the direct product of its Sylow subgroups. Proor. Let P be a p-Sylow subgroup of G, and let N(P) be its normalizer. By (6.9), N(P) is its own normalizer, and from (6.11) it follows that N(P)=G. This shows that PAG. Now let P, R,---, be the Sylow subgroups of G for the primes b,r, +++, dividing the order of G. Since P, R,---, are normal sub- groups whose orders are relatively prime, it follows from Exercise 5.1 that PR+++=PxRx-::: Since PR --- has the same order as G, we have G=PxXRx-::-. The converse of Theorem 6.12 is also true; that is, a finite group G which is a direct product of its Sylow subgroups is nilpotent. This holds since every p-group is nilpotent and a direct product of nilpotent groups is nilpotent.§7.1 BACKGROUND FROM GROUP THEORY 21 Exercises 1, Let G be a p-group such that G/[G,G] is cyclic. Then G is cyclic. 2. Let H be a subgroup of a p-group G such that [G:H] =p. Prove that HAG. §7. Semi-direct Products When a finite group is constructed in a simple way from some of its subgroups, information about the group can frequently be obtained from a knowledge of properties of the subgroups by the use of some inductive principle to transfer the information to the whole group. As the standard example of this procedure, we cite the theorems which state that, if two groups A and B have a certain property, so does their direct product. A less simple but interesting and useful method of building a new group from a pair of groups is by use of the semi-direct product, which we shall now define. (7.1) Derinition. Let A and B be groups, and suppose there exists a homomorphism b > 6 of B into the group of automorphism of A. The set of all ordered pairs {(a, b): a € A, b € B}can be made into a group if we define products by (a, bY(a1, by) = (@ - (a), bb.) . This group is called the semi-direct product of A and B, relative to the homomorphism 6-6. If 6 =1 for all 6 € B, we obtain just the usual direct product. If we set A'={@,V:a¢ A}, B’={(,b):be BY, then A’ and B’ are subgroups of the semi-direct product G which are isomorphic to A and B, respectively, and will be identified with A and B in what follows. We then have G = AB, An B = {1}, and ba = 5a) +b, acA,beB. These facts show that A AG and that G/A=B. (7.2) Tuzorem. Let G be a group with subgroups A and B, and suppose there is a mapping b — b of B into the set of mappings of A into itself such that (i) G=AB, (ii) ANB={}, (ii) ba = ba) +b, acA,beB.22 GROUP REPRESENTATIONS §7.3 Then b->56 is a homomorphism of B into the group of automorphisms of A, and G is isomorphic to a semi-direct product of A and B (relative to this homomorphism). We shall omit the details of the proof. When the hypotheses of the theorem are satisfied, we shall refer to G as the semi-direct product of its subgroups A and B. As our first illustration of the concept of the semi-direct product, let H be a group, A(H) its group of automorphisms, and H;, the set of left multiplications by elements of H. Both A(H) and Hy are subgroups of the group P(H) of all permutations of H. We call the subgroup K of P(H) generated by A(H) and H;, the holomorph of H. Let us show that H; and A(H) satisfy the conditions of Theorem 7.2. First, (7.3) @:a,=(aa)z-a, ae Hae A(H). Thus, every finite product of elements of H, and A(H) can be written as a single product of an element of Hz and an element of A(H). Hence () is satisfied; that is, K = H,- A(H). It is obvious that H;n A(H) = {1}. Finally, A(H) is already a group of auto- morphisms of H;, provided that we identify H and H;. Thus K is a semi-direct product of Hz and A(H). In some cases, A(H) is known explicitly, and the multiplication table of the holomorph K of H can be given with ease. For example, let H = [a], a finite cyclic group generated by an element a of order m. Then the automorphisms of H are given by the ¢(m) mappings aac, lsism,(i,m=1. The subscripts on the a’s are taken mod m, and we have Bialj = Aj, so that A(H) is isomorphic to the multiplicative group of residue classes modm relatively prime to m. The elements of K are {a"'a;:0 Sr Sm—1,1 Si ) an( 3 onsmt) ja rt fA ae => ( onsets) : ma \ia1 We obtain by comparison T’S = ST, and therefore (8.5) T'=STsS". Conversely, for any invertible matrix S = (o,;) in K,, we may define a new basis of M by (8.4). The matrix of T with respect to the new basis {mi, ---, ms} is then STS™'. We can use the mapping T— T to define the determinant of a linear transformation T. If we set |T|=|T\|=detT, the multiplication theorem for determinants applied to (8.5) guaran- tees that | 7| is independent of the choice of the basis B. The determinant gives us a useful criterion for membership in GL(M), namely, T € GL(M) if and only if | T| #0. Exercises 1. Define the trace of a matrix A = (aij) in Kn by tr. A= Sau. Prove that tr(AB)=tr(BA) for A,B in Kn. Hence prove that, if we define the trace of a linear transformation by (8.6) Tat, the trace function is a well-defined additive map of Homx (M, M) into K such that tr (7172) = tr(T2T1), Ts € Homx (M, M). 2. Let T be the matrix of a linear transformation T in Homx (M, M). Let 4 be a transcendental element over K, and define the characteristic poly- nomial of T to be | Al— T'|, where I is the identity matrix in Kn. Show that the characteristic polynomial does not depend on the choice of a basis of M;30 GROUP REPRESENTATIONS. §9.1 that is, if 7’ is the matrix of T relative to some other basis of M, then jar—T'|=|ar-T|. Also prove that Jal — P| = am — (tr Tyan + e+ (-1 | TI for Té€ Kn. §9. Definitions and Examples of Representations In this section G will denote an arbitrary finite multiplicative group with identity element 1, and K a field. It is understood that all vector spaces considered are finite dimensional over K; for a vector space M, we shall use the notation (M:K) to denote the dimension of M over K. We recall from §8 that GL(M) denotes the group of all invertible linear transformations of a vector space M onto itself, and that GL(z, K) stands for the group of invertible n X n matrices over K. (9.1) Derinition. Let G be a finite group and M a vector space over K. A representation of G with representation space M is a homomorphism T:9— T(g) of G into GL(M). Two representations T and T’ with spaces M and M’ are said to be equivalent if there exists a K-isomorphism S of M onto M’ such that T'@S = STO), geG, that is, T'(g)Sm = ST(g)m for all m € M and g€G. The dimension (M: K) of M over K is called the degree of T and will be denoted by deg T. Similarly, we have the concept of a matrix representation. (9.2) Derinition. A matrix representation of G of degree n is a homomorphism T: 9 > T(g) of G into GL(n, K). Two matrix repre- sentations T and T’ are equivalent if they have the same degree, say n, and if there exists a fixed matrix S in GL(m, K) such that T'g) =ST@S", geG. If T is a representation of G with space M, then from the homomorphism property we have§9.3 REPRESENTATIONS AND MODULES 31 T(@b)=T@T, a,beG, T@*=T@"), TQ) = le where ly denotes the identity mapping on M. The corresponding statements hold, of course, for matrix representations. Let T:G—GL(M) bea representation of G, and let {#, -+-, mn} be a basis of M over K. For each x € G, the matrix T(x) of T(x) with respect to the basis {mm, ---, ma} is in GL(n, K), and T:x> T(x) defines a matrix representation of G called a matrix representation afforded by T (or by M). The representation T affords other matrix representations obtained by selecting other bases of M, but (8.5) together with Definition 9.2 asserts that they are all equivalent. Before introducing further concepts, we pause to lay firm hold on a number of important examples of representations. These are intended to serve as concrete illustrations upon which the reader should test the definitions and theorems to be developed in the rest of the chapter. Example 1. Permutation Representations. Let S, be the sym- metric group on ” symbols, and let M= Km, @--:@ Km, be an n-dimensional vector space over an arbitrary field K. For each o € S,, let P(c) € Home (M, M) be defined by (9.3) Poms = mei) , lsisn. For any o,t € S, we have Plor)mi = mer = Mee) = P(o)P(e)mi for 1 P(e) is an isomorphism32 GROUP REPRESENTATIONS §9.4 of S, onto the subgroup of permutation matrices in GL(n, K). Now let G be an arbitrary group for which there exists a homo- morphism ¢:.G—S, for some m. Then the composite mapping Ti x 9x) > P(P(@)) of G into GL(M) is a representation of G with representation space M, and T: x 9(«) > P(Y(x) is a matrix representation of G by permutation matrices. In particular, let G be a group of order n. By Cayley’s theorem (§2), there is an isomorphism y:G—S, given by x xz. Hence the mapping Rix P(x), xeG, is an isomorphism of G into GL(M). We shall refer to R as the regular representation of G; it is the most important example of a representation of a group. Let us describe this representation as explicitly as possible. Suppose that G = {x,---, xa} is a finite group of order n, and let M be an n-dimensional space with K-basis {m,-++, mn}. In order to determine R(x) for x € G, we proceed as follows: For each i, 1 Sin, there is a unique j, 1 $j T(x) = diag {07', ---, &x'} defines a matrix representation of G of degree m. An easy compu- tation shows, moreover, that (9.5) RG)S = ST(x) where R(x) is the matrix defined in Example 1 and Coe ow sel@o« Ge Then | S| is a Van der Monde determinant, and we have IS[=20---0 TL G-%)- isi Tay Tb? is a homomorphism of G. We list now some pairs of matrices which satisfy (9.7) and hence define representations of G. T@=(° aoe rTo=-( , 1 1 0 0-1 0 0 o 1 1 0 v@=|1 9 © Of ye@af1 o 0 -1 0 0 0 - 0 0 0 1 0 0 1 0 0 0 1 0 The reader may verify that there exists an invertible 4 x 4 matrix S such that SV@s" = [7 0 : 0 T(@). and36 GROUP REPRESENTATIONS §9.8 Tb) 0 svos=[ 0 THI The method used in Example 3 can be used to construct repre- sentations of any group for which we know the generators and a com- plete set of relations which the generators satisfy. The importance of knowing all the relations should be obvious to the reader; in fact, one can easily find matrices X and Y such that X‘= Y*=1 but for which a— X, b-> Y cannot be extended to a representation of the group G of Example 3. Because questions concerning generators and relations can often be troublesome, the most fruitful approach to group representations, as we shall see, seems to be the analysis of the regular representation and representations constructed in some way from it. Example 4. One-Dimensional Representations. Let G be a finite group with identity element 1, and let K be a field. We shall deter- mine all one-dimensional K-representations of G, that is, all maps T:G—K such that (9.8) T(xy) = T(x)T(y), xyeG, TM=1. For x € G, we have at once T(x") = T(x)". Now let G’ be the commutator subgroup of G. Then G’ is generated by the elements {xyx'y"': x, y € G}, and G/G’ is abelian. If T is any one-dimensional representation of G, then Taye ty") = T@TOMXT@IUT) Ft = 1, so that T maps every element of G’ onto the unity element of K. Consequently, T induces a map T: G/G’ > K by means of (9.9) TOG') = TX), xeG. Conversely, starting with a one-dimensional representation T of G/G' in K, equation (9.9) serves to define such a representation T of G. Different T’s yield different T’s, and conversely (see Exercise 9.1). Therefore there is a one-to-one correspondence, given by (9.9), between the distinct one-dimensional K-representations T of G and the distinct one-dimensional K-representations T of G/G’. We are thus faced with the following problem: Given a finite abelian group H of order h and a field K of charac- teristic p, determine all one-dimensional K-representations of H. To solve this problem, we use Theorem 4.2 on the structure§9.10 REPRESENTATIONS AND MODULES 37 of finite abelian groups. We may write H=H,x---X Hn, (Hi: 1] = hi = pi*, pi prime, where each H; is a cyclic group generated by an element x; of order hi; clearly, h = hi-+-hm. Because of (9.8), the representation T is completely determined by the values T(x,),---, T(%m). These values may be chosen independently of one another as long as they satisfy the relations {Tap} = 1. Thus, T is completely determined by an m-tuple {01 +++, Om), o 6K, ot =1, by setting T(x:) = 0; for1 Si GL(M) is the study of those K-subspaces N of M such that Tine N for allxe¢GandneN. Such a subspace will be called 2 G-subspace of M. If we define (10.1) Tix) = THIN, xeG, where T(x) | N denotes the restriction of T(x) to N, then T, is a representation of G with representation space N. By way of illustration, consider the representation space M for the symmetric group S,, where M=Km,@---@ Km, and where for each o € S, we define a linear transformation P(c) by Ploym; = mew , lsisn. (See §9, Example 1). Then N=K(m, + +++ + my) is an S,-subspace of M, since Pom, + +++ + my) = mi t+ +++ +My§10.2 REPRESENTATIONS AND MODULES 39 for each o € S,. In terms of G-subspaces, we can now give some important definitions. (10.2) Derinition. A representation T of G with non-zero re- presentation space M is irreducible if the only G-subspaces of M are {0} and M; otherwise T is called reducible. The representation T is completely reducible if for every G-subspace N of M there exists another G-subspace N’ !such that M= N@N’ (vector space direct sum). A G-subspace N of M is said to be irreducible or completely reducible according as the representation 7, defined in (10.1) is irreducible or completely reducible, respectively. It should be remarked that a reducible representation need not be completely reducible. (See Exercise 10.3.) Let us see what these definitions mean in terms of matrix representations. Let 7: G—>GL(M) be a representation of G, and let N be a G-subspace of M such that N+ {0} or M. Then we may choose bases so that M= Km, ®-::@Km,, N=Km.@0:::OKm, s GL(M/N) defined by (10.4) Tam + N)= Tamt+QNn, xeG, where m+ N ranges over all cosets in the factor space M/N. The mapping 7,(x) is a well-defined K-homomorphism of M/N into itself because N is a G-subspace of M. The cosets {ites + N, +++, m, + N} form a basis for M/N, and clearly T: affords the matrix represen- tation T, relative to this basis. We shall postpone until much later (§73) our discussion of the significance of the matrices V(x), x € G, and shall merely remark that these are certain rectangular matrices. If U is any other matrix representation of G afforded by T,40 GROUP REPRESENTATIONS §10.5 there exists a fixed matrix S ¢ GL(r, K) such that SU(x)S~' has the form (10.3) for all x € G. Now suppose that M is completely reducible. Then we may write M=NON', N=Km.@-:-@Km,, N!' = Kmuwi@---@Km,, and relative to the basis {m,,---,m,}, T affords the matrix repre- sentation 10.5) UH) = ee e ] : xeG, 0 Tix). It is easily verified from (10.4) that the matrix representations T; and 7; are equivalent. Thus, the complete reducibility of T implies that each matrix representation (10.3) is equivalent to one of the form (10.5). With these calculations in mind, it is clear how to define the concepts of irreducible or completely reducible matrix representations. A matrix representation U: G->GLi(r, K) is reducible if U is equi- valent to a representation T given by (10.3); if no such reduction exists, then U is an irreducible matrix representation. The repre- sentation U: G->GL(r, K) is called a completely reducible matrix representation if, whenever U is equivalent to a matrix represen- tation of the form (10.3), U is also equivalent to a representation (10.5). We shall now prove two basic theorems which indicate clearly the direction in which the whole theory is to go, even though we shall not be able to exploit all their consequences until the end of Chapter IV. (10.6) Lemma. A G-subspace of a completely reducible G-space M is completely reducible. Proor. Let N be a G-subspace of M, and let N; be a G-sub- space of N. Then N, is also a G-subspace of M, and there exists a G-subspace L of M such that M=N,@L. Then Li =LoO Nis a G-subspace of N, and we have N= N,@L,, proving the lemma. (10.7) Tueorem. Let T: G—GL(M) be a completely reducible representation of G. Then M is a direct sum of irreducible G-sub- spaces. Proor. Select a G-subspace N + {0} of minimal dimension among all non-zero G-subspaces. Then JN is irreducible, and we can find§10.8 REPRESENTATIONS AND MODULES Al a G-subspace L of M such that M=NQL. If L = {0}, then M itself is irreducible, and there is nothing to prove. If L + {0}, then, by Lemma 10.6, L is completely reducible. Moreover (M: K) > (L: K), so that we may conclude by an induction argument that L is a direct sum of irreducible G-subspaces. But a G-subspace of L is also a G-subspace of M. Therefore M is a direct sum of irre- ducible G-subspaces, and the theorem is proved. (10.8) THeorem. (Maschke [1]). Let T: G-+GL(M) be a repre- sentation of a finite group G by linear transformations on a vector space M over a field K, and assume that pb =char(K) + [G: 1). Then T is completely reducible. Proor. Let N + {0} be a G-subspace of M. We have to produce a G-subspace L of M such that M= N@L. Because M isa vector space, we can at least find a K-subspace R such that M=NOR. With such a decomposition is associated a K-linear map E: M> N defined as follows: For m € M, write m=n+r, neN,reR. Then set Em=n so that E assigns to each m € M its uniquely determined component in N. The linear transformation E is called a projection of M onto N and is characterized by the properties that @) Em=m,meN, Gi) EMCN. In fact, any F ¢ Homg (M,N) satisfying (i) and (ii) gives rise to a decomposition of M into K-subspaces: M=FMQ@(1—-F)M where (1 — F)M={m— Fm:m € M}. We claim now that if (10.9) T(x)F = FT(«), xeG, the subspaces FM and (1 — F)M are G-subspaces of M. For, if Fm is an element of FM and «x is in G, then T(«)Fm = FT(x)m € FM,42 GROUP REPRESENTATIONS §10.10 which shows that FM is a G-subspace of M. A similar proof holds for (l1—F)M. Since FM=N, our theorem will be proved if we can construct from our given projection E a new projection F of M onto N satisfying (10.9). This is done by an “averaging” device; namely, we define F € Hom, (M, M) by F=(G:1)'S T@WET(s)"*, 266 which is meaningful since [G:1] #0 in K. Then for ally ¢ G we have T(y)FT(y)" = (G: WY TO)T@ET ATO) =([G:1)' 3 TiynET yx)" =F, and hence (10.9) holds. = Next we note that, for m € M and x € G, we have T(®)ET(x)"'m € T(Q)EM = T()NCN since N is a G-subspace of M. Therefore Fm =([G:1)" > T(X)ET (x) 'm e N for all m € M, which shows that F ¢ Homg (M,N). For eachne N, we have furthermore ET(x)"'n = T(x)'n, xeG, since E| N = 1; therefore Fn =([G:1)' S T@ET (4) "'n = 1G 1 1G: nan. Thus F is a projection of M onto N for which (10.9) holds, and the theorem is proved. Theorems (10.7) and (10.8) have the following consequence in terms of matrix representations: (10.10) Corotuary. Let T: G->GL(n, K) be a matrix represen- tation of G by matrices with coefficients in a field K whose charac- teristic p}(G:1]. Then there exists a fixed matrix S € GL(n, K) such that, for all x € G, Ti(x) 0 ST(nS" = Tx) 0 Tix) where the {T;} are irreducible matrix representations.§10.11 REPRESENTATIONS AND MODULES 43 The regular representation R of the cyclic group discussed in Example 2 of §9 illustrates some of these ideas. If we view the representations R and T as having coefficients in an algebraically closed field of characteristic not dividing the order of the group, then (9.5) shows the splitting of the regular representation R as a direct sum of irreducible matrix representations. On the other hand, suppose G is cyclic of prime order p, and let K be the field of rational numbers. Then R contains an irreducible K-representation of degree p—1 (see Exercise 10.9), and the formula (9.5) shows that an irreducible K-representation may split up into irreducible K’- representations of lower degree if we view the representation as a K'-representation instead, where K' is an extension field of K. Let T: G-—GL(M) be a representation of G. It is often de- sirable to consider, instead of the linear transformations {7T(x)} themselves, the K-subspace of Homx(M,M) spanned by all the {T(x),x € G}. This space, called the enveloping algebra of T, con- sists of all linear combinations aT (m1) + +++ + aT (er) , meG,aek. Because T(xy) = T(x)T(y), it follows that the enveloping algebra is a subring of Hom, (M, M) as well as a subspace. It is also clear that the G-subspaces of M are identical with those subspaces N such that f(N)c WN for all f belonging to the enveloping algebra of T. These remarks indicate that it is desirable to embed an abstract group G in some sort of a ring such that every element of the ring is a linear combination of the elements of G with coefficients in some field. Our next objective is to give this construction. (10.11) Derinition. An algebra A over a field K is a ring A with an identity element which is at the same time a vector space over K. Moreover the scalar multiplication in the vector space and the ring multiplication are required to satisfy the axiom a(ab) = (@a)b = ala) , a@eK,abea. A subring of A which is also a K-subspace of A is called a sub- algebra of A. Now we come to the construction of the group algebra of an arbitrary finite group G with identity element e. We consider all formal sums Meg, @eK,44 GROUP REPRESENTATIONS §10.11 two such expressions being regarded as equal if and only if they have the same coefficients. (The reader who finds formal sums un- palatable should note that a formal sum is simply a function on G to K, and that the coefficient a, gives the value of the function at the element g in G.) We then define operations on the formal sums by the rules Sag +S Bg = Ses + Bog and (Sy (Soh) = 5 eatioe ie ee ghee = Sr, fra where re = Sa,By-u ie In other words, we are defining addition of functions on G by (fi + F29) = fiulg) + Fl9) and multiplication by convolution: /if2 =f, where Alt) = DADA). see Finally, we define a(S a9) = Daag , aeK, With these definitions, it is easily checked that the set of all formal sums forms an algebra KG, called the group algebra of G over K. The formal sum 1-e, 1 € K, e € G, is the identity element of KG. The formal sums g* = 1-g, g € G, which have all but one coefficient equal to zero, are linearly independent and form a basis of the algebra KG over K. The mapping g— g* is an isomorphism of G into KG, and we shall identify G with its image under this iso- morphism. We can then view G as embedded in KG so that the elements of G form a K-basis for KG. The elements of KG are multiplied according to the distributive law and the multiplication defined for elements of G. The same construction can be carried out with an arbitrary ring R with an identity in place of K and yields the group ring RG of G over R.§10.12 REPRESENTATIONS AND MODULES 45 We have observed in §8 that Homx (M, M) is an algebra over K. This leads to (10.12) Derintrion. Let A be a finite-dimensional algebra over a field K and M a finite-dimensional vector space over K. A repre- sentation of A with representation space M is an algebra homo- morphism T: A> Hom (M,M), that is, a mapping T which satisfies T@+)=T@+T, Te@)=T@T, T(@a)=aTa), Te)=1, abeA,aek, where e is the identity element of A. Two representations T and T', with representation spaces M and M’, are equivalent if there exists a vector space isomorphism S:M=M' such that T'@S = ST(@ for allae A. Now let T be a representation of G with representation space M, where M is a vector space over the field K. Then there is a unique way to extend T to a representation T* of KG with repre- sentation space M, namely (10.13) T*(S a9) = Se Tg) « Conversely, every representation of KG, upon restriction to G, yields a representation of G. There is thus a one-to-one correspondence between K-representations of the group G and representations of the group algebra KG. Moreover, let T: G- GL(M) be a representation of G, and let T*: KG—Homg (M, M) be the corresponding repre- sentation of KG. Let us call a K-subspace N of M a KG-subspace if T*@NCN for all ae KG. It is then clear that a K-subspace N is a KG-subspace if and only if N is a G-subspace. Let us also remark that the set {7 *(a):a¢ KG} forms the enveloping algebra of the linear transformations {7(g): g €G}. Finally, we caution the reader that for a € KG the linear transformation T*(a) need not be invertible.46 GROUP REPRESENTATIONS §10.14 A matrix representation of degree n of a finite-dimensional algebra A is an algebra homomorphism of A into K,. The remarks con- cerning the relation between representations by linear transforma- tions and matrix representations of groups can be carried over verbatim to representations of algebras. The final step in this process of abstraction was taken by Noether [1], who showed that the systematic use of modules over rings and algebras leads to simplifications in the terminology and conceptual framework of the theory of representations. As we have said before, the module-theoretic approach has been adopted in this book, and we proceed now to lay the foundations of this theory. Let G be a group with identity element e, and let T: G-> GL(M) be a representation of G. For each m € M and g € G, we definea new element gm of M according to the rule gm = T(gym. The “product” gm has the following properties: 0.14) {am * ml) = gm + gm! (gg'ym = g(g'm), em=m, for g,9' € G, m,m' € M. These are simply restatements of the formulas TgXm + m') = T(yg)m + T(g)m', T(gq'm = T@)Ty'm, Te)m=m. Thus each representation yields a G-module in the sense of (10.15) Dermirion. Let G be a group and M an additive abelian group. The group M is called a left G-module if, for each g € G and m € M, a product gm € M is defined such that the relations (10.14) hold. Similarly, let 7; A— Homz(M,M) be a representation of an algebra A over K. Then, for each a € A and m € M, we define am = T(a)m and observe that, because of the properties of the representation T, we have for all a,a’ ¢ A, m,m'e¢ M, ae K, aim +m')=am+am', (a+a'\m=am+a'm, (10.16) {cy =ala'm), em=m, (aa)m = a(am) = alam) , where e¢ is the identity element in A.§10.17 REPRESENTATIONS AND MODULES 47 (10.17) Derinition. Let A be an algebra over K, and let M be a vector space over K. We say that M is a left A-module (or a module over A) if for each a € A and m € M,a “product” am € M is defined which satisfies the rules (10.16). In this book all K-spaces which are modules over K-algebras are assumed to be finite-dimensional vector spaces over K. The observation which led to the definition of module works equally well in reverse. For example, let M be a K-space which is a left A-module for an algebra A over K. For each a € A, de- fine a mapping T(@): M- M by setting T@m=am, acA,memM. Then the formulas (10.16) assert that T@(m + m') = Tam + Tan’, T@(am) =aT(am, so that T(a) € Homz (M, M) for each a € A. Moreover, we have T@ + a2) = Tm) + T@), T (aia) = T(as)T(@:) , T(@a) =aT(a), T@=1, which shows that T is indeed a representation of A. Let {m., ---, mn} be a basis of M over K. Then for each ae A we have am; = 5) aji(aym; , lsisn. a Let T(q@) be the matrix (q:,(a)); then our remarks show that T is a matrix representation of A, called a matrix representation afforded by the A-module M. (10.18) Dermition. Let M and M’ be left A-modules where A is an algebra over the field K. The modules M and M’ are said to be A-isomorphic if there exists a vector space isomorphism S of M onto M’ such that for all a ¢ A and m € M we have a(Sm) = S(am) . Our definitions have been set up in such a way that two modules are A-isomorphic if and only if the representations afforded by them are equivalent.48 GROUP REPRESENTATIONS §10.19 The most important example of a left A-module over an algebra A is the vector space A itself, with the module product defined to be the ordinary ring multiplication in A. This module will be denoted by 4A and is called the left regular module of A. (10.19) Derinition. Let M be a left A-module over a K-algebra A where K is a field. A K-subspace N of M is called a sub- module if an € N for alla € A andneN. For example, the submodules of the left regular module ,A are the left ideals of A. Finally, let A = KG be the group algebra of a finite group G over a field K. From the results of this section, we see that there is a one-to-one correspondence between the K-representations of G and the left KG-modules M. For such a module M, the correspond- ing representation T: G-> GL(M) of G is given by T(gm = gm, geG,meM. The KG-submodules of M are precisely the G-subspaces of M intro- duced at the beginning of this section. Two left KG-modules are isomorphic if and only if the corresponding representations of G are equivalent. Exercises 1, Let A be an algebra, and let the center of A be defined as C= {ce € A:ca=ae for alla € A}. Prove that C is a subalgebra of A. 2, Let KG be the group algebra of a finite group. Show that Dzeo belongs to the center of KG. More generally, show that, if © is a fixed con- jugate class in G, then Xeeg 2 belongs to the center of KG. 3, Let K be a field of characteristic p > 0, and let G be the cyclic group [z] of order p. Show that the matrix representation given in (9.6), 1 1 #>(9 1) ; is reducible but not completely reducible. (Thus Theorem 10.8 is not true in general if p|[G:1).) 4. Is the regular representation of a finite group G ever irreducible? 5. Let KG be the group algebra of a finite group G over a field K. Show that the matrix representation of G afforded by the left KG-module xgKG is equivalent to the regular representation of G. The next two problems give another approach to Maschke’s theorem for§10.20 REPRESENTATIONS AND MODULES 49 representations over the fields of real or complex numbers. We begin with some definitions. A complex-valued function f, defined on pairs of vectors (m,n) belonging to a vector space M over the field of complex numbers, is called a positive definite hermitian form if for all vectors m,n,q, F(m + 0,9) = f(m,a) + F(n,a), f(m,n +9) =f(m,n) + f(m, a), (10.20) j f(am,n)=af(m,n), f(m,n) = f(n,m), f(m,m) > 0 for m #0. In this discussion @ denotes the complex conjugate of a. The assumption f(m,n) = Fin, m) implies that f(m,m) is always a real number, so that the last condition makes sense. The standard example of such a form is given by f(m,n) = X ase where {as} and {@;} are the coefficients of m and m with respect to some fixed basis of M. For any form f satisfying (10.20), the set of all K-automorphisms U such that f(Um, Un) = f(m,n), m,n € M, is a subgroup of GL(M) called the unitary group of M with respect to f, and its elements are called unitary transformations. 6. Let T: G-> GL(M) be an arbitrary representation of G by linear trans- formations of a complex vector space M. Prove that there exists a positive definite hermitian form f such that each linear transformation 7(g) is a uni- tary transformation with respect to f. [Hint: Start with any positive definite hermitian form g on M, and define (m,n) = XL o(T(a)m, T(z)n) .] 7. Let T: G->GL(M) be a representation of a finite group G, by linear transformations of a complex vector space M such that each of the transfor- mations {7'(g),g € G} is a unitary transformation with respect to a positive definite hermitian form f. Prove that J is completely reducible by showing that if N is a G-subspace of M, so is N1 = {m € M: f(N, m) = 0}, and that M=NQN?. 8. Let 7: G-> GL(M) be a representation of a possibly infinite group G, where M is a vector space over a field K of characteristic p. Suppose that G contains a subgroup H of finite index such that p{[G:H] and for which T|H (the restriction of T to H) is completely reducible. Prove that T is itself completely reducible. [The proof is almost the same as that of Maschke’s theorem.] 9. Let G =[z] be a cyclic group of prime order p, and let K be the field of rational numbers. Let M be a p-dimentional vector space over K with basis {mo, mi, +++, mp-1}. Define R(x) by mi > mi+1,0 St mo. Then R is the regular representation of G. Let uo = D3-'ms, and let u = mi — m1, 15% p—1. Prove that Ni = Kuo and Ne= DP"! Ku; are irre- ducible G-subspaces of M, and that M= Ni@ Nz. (The proof uses the fact that the polynomial 4?-! +.---+ 4 +1 is irreducible in K{A).)50 GROUP REPRESENTATIONS §11.1 §11. Modules We shall take a somewhat more general point of view than that introduced in §10. In §§ 11-14, R will denote an arbitrary ring with -a unity element 1. (11.1) Dermition. An abelian group M, written additively, is called a left R-module if, for each re R and me M, a product rm & M is defined such that rm +m)=rmt+rm, (ntnma=nm+nm, (rirym = n(rm) , lm=m, for ally in R, min M. A subgroup M, of M is called a submodule if rm, € M,.for all y € R and m, € M,. A one-to-one mapping f of a left R-module M onto a left R-module M’ is called an R-isomor- phism if f(m, + mz) = f(m) + f (mz) and f (ym) = rf(m) for all m € M, vy € R. When there exists an R-isomorphism between M and M’, we say that M and M’ are R-isomorphic and write M=M'. (The group M is called a right R-module if there is a product mr defined such that mr € M and (m, + m)r = mur + mer , mn +72) = mr, + mn, mnre) =(mn)rz, m-l=m, for all ry € R, me M. The concepts of submodule and isomor- phism are defined for right modules in the obvious way.) The left regular module pR of a ring R is the left R-module whose underlying abelian group is the additive group of R, and the module product is given by the ring multiplication rm for ry € R and m€ RR. The submodules of xP are the left ideals of R. The right regular module Rp is defined similarly, and its submodules are the right ideals of R. Included in the concept of module are vector spaces over fields and skewfields, as well as the modules over finite-dimensional algebras considered in the preceding section. In order to check that the pre- sent discussion does indeed include the theory of modules over algebras, let A be an algebra over K with unity element 1* and let M be a left A-module in the sense of Definition 11.1. We show that M is a vector space over K and that the last condition of (10.16) is satisfied. First, we define scalar multiplication on M by the rule a-m=(al*)m, aeK,memM;§11.1 REPRESENTATIONS AND MODULES 51 then it is clear that with this definition, M is a vector space over K and that for allae K,aeA,meM, alam) = (aa)m = alam) . It is also clear that a submodule in the sense of Definition 11.1 is a subspace of the vector space M because of the definition of the scalar multiplication. In the next three subsections, we develop some of the elementary notions of the theory of modules. In this discussion, “R-module” always means “left R-module”; it will be clear, however, that the entire discussion applies equally well to right modules. §11A. Direct sums If M, and M, are submodules of the R-module M, we define the sum of M, and M, by M, + M, ={m. + mz: € Mi, m: € M2}. Then M, + M, is again a submodule of M and is the smallest sub- module which contains both M, and M;. The intersection M,n Mz is the largest submodule contained in both M, and M2. Now let M,,--:, Mz be submodules of the R-module M. We write M=MO-:-OM and call M the (internal) direct sum of Mu, ---, My if @ M=M,+---+M, and Gi) my, +--+ +m,=0, m: € Mi, implies that each m: = 0. It is easily verified that if (i) holds then (ii) is equivalent to either of the following two conditions: di’) Min (M, + +++ + Mir + Miss + +++ + My) =0 for each i; ii’) Every element m € M can be expressed uniquely as a sum m=m, + met +++ +m, me € Mi. More generally, let {M;} be a possibly infinite family of sub- modules of M. The sum 5) M; of the family is the submodule of M whose elements are all possible finite sums of elements from the various submodules M;. We say that M is the (internal) direct sum of the family {Mj} if (a) M=3)M; and (b) Every m € M can be expressed uniquely as a sum ma=mi,t->+mi,, me Mi,. Now let M,, +++, M; be a given set of R-modules, not necessarily52 GROUP REPRESENTATIONS §11.2 submodules of a common R-module. We define their external direct sum M* = M,+.--- + M; to be the set of all A-tuples (mm, ---, mi), m; € Mi, where addition is performed componentwise, and nmi, +++, me) = (rm, +++, 7m), TER, Then M* is an R-module; if we set Mi ={(, ---,0, mi,0,-+-,0), mee Mi, then Mj is a submodule of M*, and Mj = M:. Moreover we have M+---+M@=Mi®---OM. (11.2) Derinition. Let {ma} be a possibly infinite set of elements of an R-module M, where @ ranges over some indexing set. The set {ma} is R-free if whenever (11.3) Ya;Ma, + +++ + 7ajMa, =0, To CR, then necessarily each coefficient rs, = 0. (11.4) Derinirion. A subset {ma} of an R-module M is called a set of generators of M if M= 34 Rma, that is, if every element of M is an R-linear combination of a finite number of the {ma}. A Jinitely-generated module is one having a finite set of generators. A cyclic module is a module with a single generator; thus M is a cyclic if and only if M = Rm for some me M. (11.5) Derinition. A R-free set of generators of M is called an R-basis of M. Thus the set of elements {m.} is an R-basis of M if and only if every element of M can be expressed uniquely as a finite R-linear combination Drama,» Ya, € R, ma, € {ma} Not all modules have bases; an R-module which has an R-basis is called a free (left) R-module. If {ma} is a basis of a free R-module M, each of the modules Rma is isomorphic to the left regular module rR. Conversely, any module which is a direct sum of submodules, each of which is isomorphic to gR, is a free R-module. (11.6) Derinition. An R-module M is R-torsion-free if rm =0, r € R, me M, implies that either r= 0 or m=0. For example, a vector space V over a field K is a torsion-free K-module. On the other hand, an additive abelian group W is torsion-§11.7 REPRESENTATIONS AND MODULES. 53 free as Z-module if and only if W contains no elements of finite order. §11B. Homomorphisms, submodules, and factor modules Let M,N be R-modules. We use the notation Hom, (M, N) to denote the set of all R-homomorphisms of*M into N, that is, the set of all mappings f: M— N such that f(m, +m) =f(m.) + fm), f(rm)=rf(m), meM,reR. The set Homr(M,N) forms a subgroup of Hom(M,N), and Hom, (M, M) forms a subring of Hom (M, M). We call Home (M,M) the ring of R-endomorphisms of M, or sometimes the centralizer of the R-module M, because the elements of Hom, (M, M) are precisely those endomorphisms f of M which commute with all the endo- morphisms of M Trim—>rm determined by the elements of R. If N is a submodule of M, the factor module MIN is the left R-module whose underlying commutative group is the totality of cosets {m+ N} of N in M, and the module composition is defined by (11.7) rm+N)=rm+N; it is well defined because N is a submodule. The mapping v: m—> m+ N, which maps m € M onto the coset containing it, is by (11.7) an R-homomorphism of M onto M/N, called the natural mapping or natural homomorphism of M onto MIN. We state without proof some straightforward analogues of the various homomorphism theorems for groups. The reader who is unfamiliar with modules will profit from working out detailed proofs of these results. (11.8) Let M,N be R-modules, and let f. M—>WN be an R- homomorphism of M onto N. Let M:=f7*(0) be the kernel of f; then M, is a submodule of M, and MIM, = N. More specifically, let » be the natural mapping of M onto M/M,. Then f: v(m) > fn) is the required isomorphism of M/M, onto N.54 GROUP REPRESENTATIONS §11.9 (11.9) Let M, N be submodules of a common R-module. Then (M+N\M=N(MON). (11.10) Let N be a submodule of the R-module M. There exists @ one-to-one inclusion-preserving correspondence between the sub- modules of M which contain N and the submodules of MIN. This correspondence is given as follows: let v: M— M/N be the natural mapping. Then to each submodule L of M such that L> N corresponds the submodule »(L) = L/N of M/N; conversely each sub- module Z of M/N determines a unique submodule L of M such that LON, namely, L=»~(Z). Furthermore, we have (11.11) MIL = (MIN)(LIN) whenever M> LN. (11.12) Let M be an R-module with submodules M,, ---, My such that M=M.9---O®M. For each i, let N; be a submodule of Mi, and set N=M+--+M=MO--- OM. (See Exercise 11.2.) Then MIN = (MIN,) + +++ + hiNz) « A final remark concerns the familiar connection between direct sums and projections. Let M=M,@---@Mi, and for each i, 1sisk, define a mapping 7:: M— M; by the rule mim, + ++ +m) =m. Then all the {z;} are elements of Homg (M, M) and satisfy the con- ditions lamt+e tm, mam, mnjp=mm=0, i#j. Conversely, given an R-module M and any set of R-endomorphisms {n;} satisfying these relations, the sets {z;(M)} are submodules of M, and M is their direct sum. The R-endomorphisms {z;} are called the projections associated with the given direct sum decomposition, and a particular 2; is called the projection of M onto M;. The reader will find it instructive to re-read the proof of Theorem 10.8 with these definitions in mind.§11.13 REPRESENTATIONS AND MODULES 55 §11C. Finiteness conditions For algebras and modules which are finite-dimensional vector spaces over fields, all the conditions discussed in this subsection will automatically be satisfied. The discussion in the general case, however, helps to clarify exactly what hypotheses are needed in order to prove the theorems on composition series and indecomposable modules in §§13 and 14. (11.13) Derinition. Let M be an R-module. The submodules of M are said to satisfy the descending chain condition (D.C.C.) if every chain of submodules of M@ M,> M,> M;>::- terminates, that is, if there exists an index j such that M;= Mj, = +--+. Analogously, the submodules of M are said to satisfy the ascending chain condition (A.C.C.) if every chain of submodules of M M,c Mc Myc::: terminates. If the submodules of M satisfy the A.C.C., we shall call M a noetherian module. A (left) noetherian ring R is one for which the left regular module eR is a noetherian module; in other words, R is a noetherian ring if and only if its left ideals satisfy the A.C.C. (11.14) Tueorem. The following statements concerning the R- module M are equivalent: (i) The submodules of M satisfy the A.C.C. (ii) Every submodule of M is finitely generated. (iii) Every non-empty collection of submodules of M contains a maximal element, that is, a submodule which is not properly contained in any other submodule in the collection. Proor. Statement (i) implies (ii): Suppose (ii) is false, and let M* be a submodule of M which is not finitely generated. Then, by induction, we can construct an infinite sequence of elements a;, a2, --+ in M* such that if M, = Ra, + --- + Ray, then for every k, M; is properly contained in M,,,. This contradicts (i), and the first im- plication is proved. Statement (ii) implies (iii). Suppose (iii) is false. Then there exists a non-empty collection X of submodules which has no maximal element. By mathematical induction, we can find submodules Mi,56 GROUP REPRESENTATIONS §11.15 Mz., +++ in X such that for every k, M, is properly contained in Mess. The union Uz-1.M, = M™* is evidently a submodule of M although not necessarily a member of X. We prove that M“ is not finitely generated. Suppose the contrary, namely, that M* = Rm, + +++ + Rm, for some s. Then each m; € Mi, for some ki, and, choos- ing the largest of the modules M;,, we conclude that M* = Mi, for some index k; and hence that Mi,+: = Mi,. But this contradicts the way the submodules M, were constructed. To sum up, we have shown that if (iii) is false, then (ii) is false, and we can assert there- fore that (ii) implies (iii). Statement (iii) implies (i). Suppose (i) is false. Then there exists an infinite properly ascending chain of submodules, and it is clear that this collection of submodules can have no maximal element. This completes the proof of the theorem.t By the same method of proof, we have (11.15) Toeorem. Let M be an R-module. Then the submodules of M satisfy the D.C.C. if and only if every non-empty collection of submodules of M has a minimal element, that is, a submodule which does not properly contain any other submodule in the collection. We have already discussed the fact that R itself may be regard- ed as a left R-module 2R and that the submodules of xR are the left ideals of R. (Similarly, the submodules of the right module Rr are the right ideals of R.) By Theorem 11.14, we see that all the left ideals of R are finitely generated (as R-modules) if and only if R is a left noetherian ring in the sense of (11.13). We use this result in the course of the proof of the next theorem. (11.16) Tueorem. Let R be a left noetherian ring, and let M be a finitely generated R-module. Then every submodule of M is finitely generated. Proor. Since M is finitely generated, there exist elements m, +++, mm, in M such that M= Rm +---+Rm. We prove the theorem by induction on &, the result being vacuously true if k=0. Thus we may assume & 2 1, and that every R-module which can be generated by fewer than k elements has the property that all its submodules are finitely generated. t The proof of this theorem uses a disguised form of the Axiom of Choice; the theorem can also be proved using the Maximum Principle (§15 A).§11.17 REPRESENTATIONS AND MODULES 57 Let N be a submodule of M. Every element ” in Nan be ex- pressed, possibly in many ways, as an R-linear combination of the generators of M, (11417) n=nm te trem, neR. For n € N, let S(x) be the subset of R consisting of all elements rv, which occur as the coefficient of m, in some expression for m such as (11.17). Let S=U S(x). nen We shall prove that S is a left ideal in R (possibly the zero ideal). Let s,s’ € S; then there exist elements n= sm, + sum, +--+ + sem, n! = s'm, + stm: + +++ + sm in N. Then s—s’ is the coefficient of m, in some expression for n—n', and s—s' eS Further, rn =rsm, + rsymz + +++ + rsh and so vs € S for ally e R. Therefore S is a left ideal in R. By Theorem 11.14 [see also the discussion preceding the state- ment of this theorem], S is a finitely generated R-module, and so there exist elements si, ---, s: in S such that (11.18) S=Rs +++ +Rs. For each i, 1 SiS #, there exists m; ¢ N such that s; € S(m;). Now consider 2 € N given by (11.17). We have 7, € S, so by (11.18) we may write r= mS tos + HS, meER,. Then we have (11.19) n — (any + +++ + Xe) = Yom, + +++ + MME for some elements {yi} in R. Now let us set M’= Rm, +--+ + Rm. Since the left side of (11.19) is in N, the right side is in M’ 1 N, and it follows that (11.20) N=Rn, +--+ +Ru+(M'nN). The submodule M' 1 N is a submodule of M’, and M’ is generated by &—1elements. By our induction hypothesis, M'n N is finitely generated, and hence (11.20) implies that N is finitely generated. This completes the proof. Another proof of this theorem can be obtained by a simple modification of Exercise 11.18 below.58 GROUP REPRESENTATIONS §11.20 Exercises 1. Let Mi,-+-,Mz be submodules of the R-module M such that M= M.:@-+-@®Mz. Show that every element of M is expressible in one and only one way as S'_,mi, mi € Mi. 2. Let the R-module M be the direct sum of its submodules Mi, ---, Me, and let Ni be a submodule of Mi, 1M®x N with the following properties: (a) Every element of M@®x N is a K-linear combination of ele- ments of the form g(m,n), m € M, ne N, and (b) For each vector space V, every bilinear map f: Mx N>V is obtained by first mapping (m,n) into M@x WN by ¢ and then mapping ¢(m, 2) into V by a linear mapping f*: M@x N- V. More compactly, we have Fm, n) = f*(o(m, n)) where f* is a linear transformation. Our point of view is that the theory of tensor products reduces the study of bilinear mappings to the more familiar theory of linear transformations. Now we are ready to study tensor products in general. We shall have to consider mappings that are somewhat weaker than bilinear mappings, and, following Chevalley [3], we call them balanced map- pings in the sense of the following definition: (12.1) Derinition. Let M be a right module and Na left module over an arbitrary ring R with an identity element. Let P be an abelian group, written additively. A balanced map f of the Cartesian product set Mx N into P assigns to each pair (m,n) € Mx N an element f(m,) € P, so that fm + ms, n) = fm, n) + fm, 2), Fm, m + m2) = f(m, m) + fm, m2) , (m,n) = f (mr, n) for ally ¢ R and m: ¢ M, nie N. (12.2) Derinition. Let f: Mx N—-P and g: Mx N->T be balanced maps of M x N into the additive abelian groups P and T respectively. We say that f can be factored through T if there exists a homomorphism f*: T— P such that f=f*e, or more explicitly Sf (m,n) = f*(o(on, n))§12.3 REPRESENTATIONS AND MODULES. 61 for all (m,n) € Mx N. In other words, f can be factored through T if there exists a homomorphism f*: T— P such that the diagram Ve / \ Mx N————P is commutative. (12.3) Tuzorem. Let M and N be right and left R-modules, respec- tively. There exists an abelian group T and a balanced map t: Mx N-T such that (i) The elements t(m,n) generate the additive group T, and in fact every element of T is a sum S\t(mi,ni) where m;€ M, ne N. (ii) Every balanced map of M x N into an arbitrary abelian group P can be factored through T. Proor. We start with the Cartesian product M x N consisting of all ordered pairs (m,n), m € M, n € N. Now form the free Z- module F which has as Z-basis the elements of Mx N, so that F is the additive abelian group which consists of all finite formal sums Di zis(mi, 05) , zweZ,meM, nen. Let H be the subgroup of F generated by the formal sums Gm + mz, 2) — (mi, 2) — (m2, 2) , (12.4) (m, m1 + n2) — (mm, m) — (mM, m2) , (m, rn) — (mr, n) for allme M,neéeN,reR. Let T be the factor group F/H, and define a mapping t Mx N>T by means of tim, n) = (m,n) +H. Since the sums listed in (12.4) all lie in H, we have at once (12.5) tm, + m2, n) — tim, n) — tOmz, n) = 0 (in T), and so on, which shows that ¢ is a balanced map. From the fact that every element of F is a Z-linear combination of ordered pairs,62 GROUP REPRESENTATIONS §12.6 it follows that every element of T is a Z-linear combination of ele- ments of the form t(m,n), m¢M, néN. To show that every element of T is expressible in the form Stim, ni), meM,neN, we need remark only that (12.5) implies t(zm, n) = t(m, zn) = 2t(m, n) for all me M,néN,zeZ. Now let g be any balanced map of Mx N-P. Because the elements (m,) form a Z-basis of F, the mapping ¢ defines a homo- morphism ¢’ of F into P by means of '(D zes(mi, 25) = DY zise(ms, 3) « The fact that g is balanced implies that y'(H)=0. Therefore ¢’ induces a homomorphism ¢* of F/H = T into P such that o* (m,n) + H) = o(m,n), (m,n) @ Mx N. Since t(m, ) = (m,n) + H, we have 9" (Gn, n)) = om, n) , and we have proved that g can be factored through T. This com- pletes the proof of the theorem. (12.6) Derinition. The group J constructed in Theorem 12.3 is called the tensor product of M and N, and will be denoted by MON. The next result shows that the tensor product is uniquely deter- mined up to isomorphism by the properties (i) and (ii) of (12.3). (12.7) Corottary. Let (M@kN, t') be another pair consisting of an abelian group M@®z N and a balanced map t': Mx N>M@zN such that conditions (i) and (ii) of Theorem 12.3 hold. Then there exists a group isomorphism 2 of M@zN onto M@kN such that for all (m,n) € Mx N, we have A(t(m, n)) = t'(m, n) Proor. Applying Theorem 12.3, there exist homomorphisms 4: M@x2N->M@OkN and us M@zkN—> M@zN such that Atm, n)) = tm, 2) and At'(m, n) = tm, n) .§12.8 REPRESENTATIONS AND MODULES 63 Because the elements {t(m,n)} and {t'(m,n)} generate the groups M@zWN and M@zN, respectively, it follows that “A and Ayu are the identity mappings on M@z Nand M@kzN, respectively. There- fore both 4 and y are isomorphisms onto, and the corollary is proved. To recapitulate, we have now defined the tensor product M@z N and have constructed a balanced map Mx N>M@nN. Henceforth we shall write m@®ux for the image of (m,n) under this map. From the proof of Theorem 12.3, we see that the “products” m@n satisfy (m +m) @n=m@n+mOn, (12.8) jm@O(m+m)=m@mt+mOn, mr@n=m@rn, meéeM,neN,reR. These give, in a sense, the basic relations which hold in M@zN. Now let f: Mx N-P be a balanced map; then we know that there exists a homomorphism f*: M@z N-— P such that L*(m @n) =f(m,n). Since every element of M@zN is expressible as 5) mi ® ni, this shows that in fact f* is uniquely determined by f. In other words (12.9) Corotuary. Every balanced map of M x N into an arbitrary abelian group P can be factored in one and only one way through M@znN. (12.10) Tueorem. Let f: MM" and g: N->N' be R-homomor- phisms, where M and M' are right R-modules and N and N' left R- modules. Then there exists a unique homomorphism f@g of M®zN-—>M!' @xN' such that (FB gm Bn) = fm) @ Hn) . Proor. The map (m,n)—>f(m)@g(n) is balanced, and the ex- istence of f@g is immediate from Theorem 12.3. The uniqueness follows from (12.9). If f and g are as in (12.10) and if f’: M’—> M" andg': N’>N" are R-homomorphisms, we have the important formula (12.11) (F' @IVFON=S'FO99. The proof of (12.11) is immediate and will be omitted.64 GROUP REPRESENTATIONS §12.12 It is necessary to interject a word of caution before proceeding to the next result. Let M be a right R-module, N a left R-module, and M, a submodule of M. Then M, is itself a right R-module, and so we may form M;@zN. We thus have two balanced maps t: Mx N>M@nN, t: Mx N>M,@2N. For m, € M, and n € N, the symbol m,@vx is ambiguous since it could signify either ¢(m,m) or t,(m,m”). One might hope to avoid the necessity of distinguishing between these two meanings by em- bedding M:@rN in M@zN in some natural way. Let us show by means of an example that this cannot be accomplished in general. Let us set M = Q (rational field) , M.=Z, N=Z/2Z, all viewed as Z-modules. Then we claim that M@sN=0, M.@sN=N. The former holds since for x € M, né N, x@n=3x@2n=3xQ0=0. The latter isomorphism is given by [see Theorem 12.14] z@Qn-rza, zeM,nen. In view of the above remarks, the fact that the following very useful result holds is remarkable: (12.12) Tuzorem. Let M be a right R-module such that M= M.@® M; where M, and M; are submodules, and let N be a left R- module. Then M®rN=M,@2rN+M@eN. Proor. By the remarks at the end of §11B, the existence of the direct sum decomposition M = M,@ M:z implies the existence of 71, 7 € Homr(M, M) such that (2.13) l=mt+m, mam, m=m, mm=mm=0, mM=M., mM=M,. Set 6;=72;@1, i=1,2; by Theorem 12.10, each 6; is an endo- morphism of M@z N, and, from (12.13) and (12.11), we have 1=(+, G=6,, O=6., 6,02 = 0.0, =0.§12.13 REPRESENTATIONS AND MODULES 65 Setting T: = 0(M@xrN), i=1,2, we conclude that M®8zN=T,0T:. In order to complete the proof, it is sufficient to show, for ex- ample, that 7; = M,:@zN. We shall prove this by showing that T, has the characteristic properties of a tensor product as given in Theorem 12.3. We must find a balanced map g: M, x N-> T, for which the images {g(m, ):m, € M,n € N} generate 7;, and such that every balanced map g: M, x N- P can be factored through 7; by means of ¢. Let us write t: Mx N>M@zN for the mapping determined in Theorem 12.3. Since M, x Nc M «x N, we may set g =t|M, x N, so that (m1, n) = tm, n) , meéeM,,neN. It is clear from the equation M, = 7M and the definition of T, that the image of M, x N under ¢ does indeed generate 7,. Now let g. M, x N-P be a balanced map, where P is any additive abelian group. The bottom line of the following diagram gives a balanced map of Mx WN into P, and so there exists a homomorphism g*: M@r N- P making the diagram commutative. M@zrN 7 \ ett \ N ~“ Mx N—~M, x N——_~P (mx 1) g Let us set 9: =9*|T7;, so that g, is a homomorphism of 7, into P. To complete the proof, we need only verify that geg=gonM,xN. But for m, € M,, n € N, we have gems, n) = gills, n) = g*t(my, n) = 971 X 1m, 2) = yam, n) = ym, 2) « This establishes the result.