Introduction to representation theory

Pavel Etingof, Oleg Golberg, Sebastian Hensel,

Tiankai Liu, Alex Schwendner, Dmitry Vaintrob, and Elena Yudovina
January 10, 2011
Contents
1 Basic notions of representation theory 5
1.1 What is representation theory? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5 Quotients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Algebras dened by generators and relations . . . . . . . . . . . . . . . . . . . . . . . 11
1.7 Examples of algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.8 Quivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.9 Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.10 Tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.11 The tensor algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.12 Hilberts third problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.13 Tensor products and duals of representations of Lie algebras . . . . . . . . . . . . . . 20
1.14 Representations of sl(2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.15 Problems on Lie algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2 General results of representation theory 23
2.1 Subrepresentations in semisimple representations . . . . . . . . . . . . . . . . . . . . 23
2.2 The density theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3 Representations of direct sums of matrix algebras . . . . . . . . . . . . . . . . . . . . 24
2.4 Filtrations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.5 Finite dimensional algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1
2.6 Characters of representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.7 The Jordan-Holder theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.8 The Krull-Schmidt theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.10 Representations of tensor products . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3 Representations of nite groups: basic results 33
3.1 Maschkes Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.2 Characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4 Duals and tensor products of representations . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Orthogonality of characters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.6 Unitary representations. Another proof of Maschkes theorem for complex represen-
tations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.7 Orthogonality of matrix elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.8 Character tables, examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.9 Computing tensor product multiplicities using character tables . . . . . . . . . . . . 42
3.10 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4 Representations of nite groups: further results 47
4.1 Frobenius-Schur indicator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 Frobenius determinant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.3 Algebraic numbers and algebraic integers . . . . . . . . . . . . . . . . . . . . . . . . 49
4.4 Frobenius divisibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5 Burnsides Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.6 Representations of products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.7 Virtual representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.8 Induced Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.9 The Mackey formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.10 Frobenius reciprocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.11 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.12 Representations of S
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.13 Proof of Theorem 4.36 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
4.14 Induced representations for S
n
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2
4.15 The Frobenius character formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.16 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.17 The hook length formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.18 Schur-Weyl duality for gl(V ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.19 Schur-Weyl duality for GL(V ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.20 Schur polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.21 The characters of L

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.22 Polynomial representations of GL(V ) . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.23 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.24 Representations of GL
2
(F
q
) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.24.1 Conjugacy classes in GL
2
(F
q
) . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.24.2 1-dimensional representations . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.24.3 Principal series representations . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.24.4 Complementary series representations . . . . . . . . . . . . . . . . . . . . . . 73
4.25 Artins theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.26 Representations of semidirect products . . . . . . . . . . . . . . . . . . . . . . . . . . 76
5 Quiver Representations 78
5.1 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.2 Indecomposable representations of the quivers A
1
, A
2
, A
3
. . . . . . . . . . . . . . . . 81
5.3 Indecomposable representations of the quiver D
4
. . . . . . . . . . . . . . . . . . . . 83
5.4 Roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.5 Gabriels theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
5.6 Reection Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5.7 Coxeter elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.8 Proof of Gabriels theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
6 Introduction to categories 98
6.1 The denition of a category . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6.2 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.3 Morphisms of functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.4 Equivalence of categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
6.5 Representable functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3
6.6 Adjoint functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.7 Abelian categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.8 Exact functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7 Structure of nite dimensional algebras 106
7.1 Projective modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.2 Lifting of idempotents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.3 Projective covers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
INTRODUCTION
Very roughly speaking, representation theory studies symmetry in linear spaces. It is a beautiful
mathematical subject which has many applications, ranging from number theory and combinatorics
to geometry, probability theory, quantum mechanics and quantum eld theory.
Representation theory was born in 1896 in the work of the German mathematician F. G.
Frobenius. This work was triggered by a letter to Frobenius by R. Dedekind. In this letter Dedekind
made the following observation: take the multiplication table of a nite group G and turn it into a
matrix X
G
by replacing every entry g of this table by a variable x
g
. Then the determinant of X
G
factors into a product of irreducible polynomials in x
g
, each of which occurs with multiplicity
equal to its degree. Dedekind checked this surprising fact in a few special cases, but could not prove
it in general. So he gave this problem to Frobenius. In order to nd a solution of this problem
(which we will explain below), Frobenius created representation theory of nite groups.
1
The present lecture notes arose from a representation theory course given by the rst author to
the remaining six authors in March 2004 within the framework of the Clay Mathematics Institute
Research Academy for high school students, and its extended version given by the rst author to
MIT undergraduate math students in the Fall of 2008. The lectures are supplemented by many
problems and exercises, which contain a lot of additional material; the more dicult exercises are
provided with hints.
The notes cover a number of standard topics in representation theory of groups, Lie algebras, and
quivers. We mostly follow [FH], with the exception of the sections discussing quivers, which follow
[BGP]. We also recommend the comprehensive textbook [CR]. The notes should be accessible to
students with a strong background in linear algebra and a basic knowledge of abstract algebra.
Acknowledgements. The authors are grateful to the Clay Mathematics Institute for hosting
the rst version of this course. The rst author is very indebted to Victor Ostrik for helping him
prepare this course, and thanks Josh Nichols-Barrer and Thomas Lam for helping run the course
in 2004 and for useful comments. He is also very grateful to Darij Grinberg for very careful reading
of the text, for many useful comments and corrections, and for suggesting the Exercises in Sections
1.10, 2.3, 3.5, 4.9, 4.26, and 6.8.
1
For more on the history of representation theory, see [Cu].
4
1 Basic notions of representation theory
1.1 What is representation theory?
In technical terms, representation theory studies representations of associative algebras. Its general
content can be very briey summarized as follows.
An associative algebra over a eld k is a vector space A over k equipped with an associative
bilinear multiplication a, b ab, a, b A. We will always consider associative algebras with unit,
i.e., with an element 1 such that 1 a = a 1 = a for all a A. A basic example of an associative
algebra is the algebra EndV of linear operators from a vector space V to itself. Other important
examples include algebras dened by generators and relations, such as group algebras and universal
enveloping algebras of Lie algebras.
A representation of an associative algebra A (also called a left A-module) is a vector space
V equipped with a homomorphism : A EndV , i.e., a linear map preserving the multiplication
and unit.
A subrepresentation of a representation V is a subspace U V which is invariant under all
operators (a), a A. Also, if V
1
, V
2
are two representations of A then the direct sum V
1
V
2
has an obvious structure of a representation of A.
A nonzero representation V of A is said to be irreducible if its only subrepresentations are
0 and V itself, and indecomposable if it cannot be written as a direct sum of two nonzero
subrepresentations. Obviously, irreducible implies indecomposable, but not vice versa.
Typical problems of representation theory are as follows:
1. Classify irreducible representations of a given algebra A.
2. Classify indecomposable representations of A.
3. Do 1 and 2 restricting to nite dimensional representations.
As mentioned above, the algebra A is often given to us by generators and relations. For
example, the universal enveloping algebra U of the Lie algebra sl(2) is generated by h, e, f with
dening relations
he eh = 2e, hf fh = 2f, ef fe = h. (1)
This means that the problem of nding, say, N-dimensional representations of A reduces to solving
a bunch of nonlinear algebraic equations with respect to a bunch of unknown N by N matrices,
for example system (1) with respect to unknown matrices h, e, f.
It is really striking that such, at rst glance hopelessly complicated, systems of equations can
in fact be solved completely by methods of representation theory! For example, we will prove the
following theorem.
Theorem 1.1. Let k = C be the eld of complex numbers. Then:
(i) The algebra U has exactly one irreducible representation V
d
of each dimension, up to equiv-
alence; this representation is realized in the space of homogeneous polynomials of two variables x, y
of degree d 1, and dened by the formulas
(h) = x

x
y

y
, (e) = x

y
, (f) = y

x
.
(ii) Any indecomposable nite dimensional representation of U is irreducible. That is, any nite
5
dimensional representation of U is a direct sum of irreducible representations.
As another example consider the representation theory of quivers.
A quiver is a nite oriented graph Q. A representation of Q over a eld k is an assignment
of a k-vector space V
i
to every vertex i of Q, and of a linear operator A
h
: V
i
V
j
to every directed
edge h going from i to j (loops and multiple edges are allowed). We will show that a representation
of a quiver Q is the same thing as a representation of a certain algebra P
Q
called the path algebra
of Q. Thus one may ask: what are the indecomposable nite dimensional representations of Q?
More specically, let us say that Q is of nite type if it has nitely many indecomposable
representations.
We will prove the following striking theorem, proved by P. Gabriel about 35 years ago:
Theorem 1.2. The nite type property of Q does not depend on the orientation of edges. The
connected graphs that yield quivers of nite type are given by the following list:
A
n
:

D
n
:

[

E
6
:

[

E
7
:

[

E
8
:

The graphs listed in the theorem are called (simply laced) Dynkin diagrams. These graphs
arise in a multitude of classication problems in mathematics, such as classication of simple Lie
algebras, singularities, platonic solids, reection groups, etc. In fact, if we needed to make contact
with an alien civilization and show them how sophisticated our civilization is, perhaps showing
them Dynkin diagrams would be the best choice!
As a nal example consider the representation theory of nite groups, which is one of the most
fascinating chapters of representation theory. In this theory, one considers representations of the
group algebra A = C[G] of a nite group G the algebra with basis a
g
, g G and multiplication
law a
g
a
h
= a
gh
. We will show that any nite dimensional representation of A is a direct sum of
irreducible representations, i.e., the notions of an irreducible and indecomposable representation
are the same for A (Maschkes theorem). Another striking result discussed below is the Frobenius
divisibility theorem: the dimension of any irreducible representation of A divides the order of G.
Finally, we will show how to use representation theory of nite groups to prove Burnsides theorem:
any nite group of order p
a
q
b
, where p, q are primes, is solvable. Note that this theorem does not
mention representations, which are used only in its proof; a purely group-theoretical proof of this
theorem (not using representations) exists but is much more dicult!
6
1.2 Algebras
Let us now begin a systematic discussion of representation theory.
Let k be a eld. Unless stated otherwise, we will always assume that k is algebraically closed,
i.e., any nonconstant polynomial with coecients in k has a root in k. The main example is the
eld of complex numbers C, but we will also consider elds of characteristic p, such as the algebraic
closure F
p
of the nite eld F
p
of p elements.
Denition 1.3. An associative algebra over k is a vector space A over k together with a bilinear
map A A A, (a, b) ab, such that (ab)c = a(bc).
Denition 1.4. A unit in an associative algebra A is an element 1 A such that 1a = a1 = a.
Proposition 1.5. If a unit exists, it is unique.
Proof. Let 1, 1

be two units. Then 1 = 11

= 1

.
From now on, by an algebra A we will mean an associative algebra with a unit. We will also
assume that A ,= 0.
Example 1.6. Here are some examples of algebras over k:
1. A = k.
2. A = k[x
1
, ..., x
n
] the algebra of polynomials in variables x
1
, ..., x
n
.
3. A = EndV the algebra of endomorphisms of a vector space V over k (i.e., linear maps, or
operators, from V to itself). The multiplication is given by composition of operators.
4. The free algebra A = kx
1
, ..., x
n
). A basis of this algebra consists of words in letters
x
1
, ..., x
n
, and multiplication in this basis is simply concatenation of words.
5. The group algebra A = k[G] of a group G. Its basis is a
g
, g G, with multiplication law
a
g
a
h
= a
gh
.
Denition 1.7. An algebra A is commutative if ab = ba for all a, b A.
For instance, in the above examples, A is commutative in cases 1 and 2, but not commutative in
cases 3 (if dimV > 1), and 4 (if n > 1). In case 5, A is commutative if and only if G is commutative.
Denition 1.8. A homomorphism of algebras f : A B is a linear map such that f(xy) =
f(x)f(y) for all x, y A, and f(1) = 1.
1.3 Representations
Denition 1.9. A representation of an algebra A (also called a left A-module) is a vector space
V together with a homomorphism of algebras : A EndV .
Similarly, a right A-module is a space V equipped with an antihomomorphism : A EndV ;
i.e., satises (ab) = (b)(a) and (1) = 1.
The usual abbreviated notation for (a)v is av for a left module and va for the right module.
Then the property that is an (anti)homomorphism can be written as a kind of associativity law:
(ab)v = a(bv) for left modules, and (va)b = v(ab) for right modules.
Here are some examples of representations.
7
Example 1.10. 1. V = 0.
2. V = A, and : A EndA is dened as follows: (a) is the operator of left multiplication by
a, so that (a)b = ab (the usual product). This representation is called the regular representation
of A. Similarly, one can equip A with a structure of a right A-module by setting (a)b := ba.
3. A = k. Then a representation of A is simply a vector space over k.
4. A = kx
1
, ..., x
n
). Then a representation of A is just a vector space V over k with a collection
of arbitrary linear operators (x
1
), ..., (x
n
) : V V (explain why!).
Denition 1.11. A subrepresentation of a representation V of an algebra A is a subspace W V
which is invariant under all the operators (a) : V V , a A.
For instance, 0 and V are always subrepresentations.
Denition 1.12. A representation V ,= 0 of A is irreducible (or simple) if the only subrepresenta-
tions of V are 0 and V .
Denition 1.13. Let V
1
, V
2
be two representations of an algebra A. A homomorphism (or in-
tertwining operator) : V
1
V
2
is a linear operator which commutes with the action of A, i.e.,
(av) = a(v) for any v V
1
. A homomorphism is said to be an isomorphism of representations
if it is an isomorphism of vector spaces. The set (space) of all homomorphisms of representations
V
1
V
2
is denoted by Hom
A
(V
1
, V
2
).
Note that if a linear operator : V
1
V
2
is an isomorphism of representations then so is the
linear operator
1
: V
2
V
1
(check it!).
Two representations between which there exists an isomorphism are said to be isomorphic. For
practical purposes, two isomorphic representations may be regarded as the same, although there
could be subtleties related to the fact that an isomorphism between two representations, when it
exists, is not unique.
Denition 1.14. Let V
1
, V
2
be representations of an algebra A. Then the space V
1
V
2
has an
obvious structure of a representation of A, given by a(v
1
v
2
) = av
1
av
2
.
Denition 1.15. A nonzero representation V of an algebra A is said to be indecomposable if it is
not isomorphic to a direct sum of two nonzero representations.
It is obvious that an irreducible representation is indecomposable. On the other hand, we will
see below that the converse statement is false in general.
One of the main problems of representation theory is to classify irreducible and indecomposable
representations of a given algebra up to isomorphism. This problem is usually hard and often can
be solved only partially (say, for nite dimensional representations). Below we will see a number
of examples in which this problem is partially or fully solved for specic algebras.
We will now prove our rst result Schurs lemma. Although it is very easy to prove, it is
fundamental in the whole subject of representation theory.
Proposition 1.16. (Schurs lemma) Let V
1
, V
2
be representations of an algebra A over any eld
F (which need not be algebraically closed). Let : V
1
V
2
be a nonzero homomorphism of
representations. Then:
(i) If V
1
is irreducible, is injective;
8
(ii) If V
2
is irreducible, is surjective.
Thus, if both V
1
and V
2
are irreducible, is an isomorphism.
Proof. (i) The kernel K of is a subrepresentation of V
1
. Since ,= 0, this subrepresentation
cannot be V
1
. So by irreducibility of V
1
we have K = 0.
(ii) The image I of is a subrepresentation of V
2
. Since ,= 0, this subrepresentation cannot
be 0. So by irreducibility of V
2
we have I = V
2
.
Corollary 1.17. (Schurs lemma for algebraically closed elds) Let V be a nite dimensional
irreducible representation of an algebra A over an algebraically closed eld k, and : V V is an
intertwining operator. Then = Id for some k (a scalar operator).
Remark. Note that this Corollary is false over the eld of real numbers: it suces to take
A = C (regarded as an R-algebra), and V = A.
Proof. Let be an eigenvalue of (a root of the characteristic polynomial of ). It exists since k is
an algebraically closed eld. Then the operator Id is an intertwining operator V V , which
is not an isomorphism (since its determinant is zero). Thus by Proposition 1.16 this operator is
zero, hence the result.
Corollary 1.18. Let A be a commutative algebra. Then every irreducible nite dimensional rep-
resentation V of A is 1-dimensional.
Remark. Note that a 1-dimensional representation of any algebra is automatically irreducible.
Proof. Let V be irreducible. For any element a A, the operator (a) : V V is an intertwining
operator. Indeed,
(a)(b)v = (ab)v = (ba)v = (b)(a)v
(the second equality is true since the algebra is commutative). Thus, by Schurs lemma, (a) is
a scalar operator for any a A. Hence every subspace of V is a subrepresentation. But V is
irreducible, so 0 and V are the only subspaces of V . This means that dimV = 1 (since V ,= 0).
Example 1.19. 1. A = k. Since representations of A are simply vector spaces, V = A is the only
irreducible and the only indecomposable representation.
2. A = k[x]. Since this algebra is commutative, the irreducible representations of A are its
1-dimensional representations. As we discussed above, they are dened by a single operator (x).
In the 1-dimensional case, this is just a number from k. So all the irreducible representations of A
are V

= k, k, in which the action of A dened by (x) = . Clearly, these representations are

pairwise non-isomorphic.
The classication of indecomposable representations of k[x] is more interesting. To obtain it,
recall that any linear operator on a nite dimensional vector space V can be brought to Jordan
normal form. More specically, recall that the Jordan block J
,n
is the operator on k
n
which in
the standard basis is given by the formulas J
,n
e
i
= e
i
+ e
i1
for i > 1, and J
,n
e
1
= e
1
. Then
for any linear operator B : V V there exists a basis of V such that the matrix of B in this basis
is a direct sum of Jordan blocks. This implies that all the indecomposable representations of A are
V
,n
= k
n
, k, with (x) = J
,n
. The fact that these representations are indecomposable and
pairwise non-isomorphic follows from the Jordan normal form theorem (which in particular says
that the Jordan normal form of an operator is unique up to permutation of blocks).
9
This example shows that an indecomposable representation of an algebra need not be irreducible.
3. The group algebra A = k[G], where G is a group. A representation of A is the same thing as
a representation of G, i.e., a vector space V together with a group homomorphism : G Aut(V ),
whre Aut(V ) = GL(V ) denotes the group of invertible linear maps from the space V to itself.
Problem 1.20. Let V be a nonzero nite dimensional representation of an algebra A. Show that
it has an irreducible subrepresentation. Then show by example that this does not always hold for
innite dimensional representations.
Problem 1.21. Let A be an algebra over a eld k. The center Z(A) of A is the set of all elements
z A which commute with all elements of A. For example, if A is commutative then Z(A) = A.
(a) Show that if V is an irreducible nite dimensional representation of A then any element
z Z(A) acts in V by multiplication by some scalar
V
(z). Show that
V
: Z(A) k is a
homomorphism. It is called the central character of V .
(b) Show that if V is an indecomposable nite dimensional representation of A then for any
z Z(A), the operator (z) by which z acts in V has only one eigenvalue
V
(z), equal to the
scalar by which z acts on some irreducible subrepresentation of V . Thus
V
: Z(A) k is a
homomorphism, which is again called the central character of V .
(c) Does (z) in (b) have to be a scalar operator?
Problem 1.22. Let A be an associative algebra, and V a representation of A. By End
A
(V ) one
denotes the algebra of all homomorphisms of representations V V . Show that End
A
(A) = A
op
,
the algebra A with opposite multiplication.
Problem 1.23. Prove the following Innite dimensional Schurs lemma (due to Dixmier): Let
A be an algebra over C and V be an irreducible representation of A with at most countable basis.
Then any homomorphism of representations : V V is a scalar operator.
Hint. By the usual Schurs lemma, the algebra D := End
A
(V ) is an algebra with division.
Show that D is at most countably dimensional. Suppose is not a scalar, and consider the subeld
C() D. Show that C() is a transcendental extension of C. Derive from this that C() is
uncountably dimensional and obtain a contradiction.
1.4 Ideals
A left ideal of an algebra A is a subspace I A such that aI I for all a A. Similarly, a right
ideal of an algebra A is a subspace I A such that Ia I for all a A. A two-sided ideal is a
subspace that is both a left and a right ideal.
Left ideals are the same as subrepresentations of the regular representation A. Right ideals are
the same as subrepresentations of the regular representation of the opposite algebra A
op
.
Below are some examples of ideals:
If A is any algebra, 0 and A are two-sided ideals. An algebra A is called simple if 0 and A
are its only two-sided ideals.
If : A B is a homomorphism of algebras, then ker is a two-sided ideal of A.
If S is any subset of an algebra A, then the two-sided ideal generated by S is denoted S) and
is the span of elements of the form asb, where a, b A and s S. Similarly we can dene
S)

= spanas and S)
r
= spansb, the left, respectively right, ideal generated by S.
10
1.5 Quotients
Let A be an algebra and I a two-sided ideal in A. Then A/I is the set of (additive) cosets of I.
Let : A A/I be the quotient map. We can dene multiplication in A/I by (a) (b) := (ab).
This is well dened because if (a) = (a

) then
(a

b) = (ab + (a

a)b) = (ab) +((a

a)b) = (ab)
because (a

a)b Ib I = ker , as I is a right ideal; similarly, if (b) = (b

) then
(ab

) = (ab +a(b

b)) = (ab) +(a(b

b)) = (ab)
because a(b

b) aI I = ker , as I is also a left ideal. Thus, A/I is an algebra.

Similarly, if V is a representation of A, and W V is a subrepresentation, then V/W is also a
representation. Indeed, let : V V/W be the quotient map, and set
V/W
(a)(x) := (
V
(a)x).
Above we noted that left ideals of A are subrepresentations of the regular representation of A,
and vice versa. Thus, if I is a left ideal in A, then A/I is a representation of A.
Problem 1.24. Let A = k[x
1
, ..., x
n
] and I ,= A be any ideal in A containing all homogeneous
polynomials of degree N. Show that A/I is an indecomposable representation of A.
Problem 1.25. Let V ,= 0 be a representation of A. We say that a vector v V is cyclic if it
generates V , i.e., Av = V . A representation admitting a cyclic vector is said to be cyclic. Show
that
(a) V is irreducible if and only if all nonzero vectors of V are cyclic.
(b) V is cyclic if and only if it is isomorphic to A/I, where I is a left ideal in A.
(c) Give an example of an indecomposable representation which is not cyclic.
Hint. Let A = C[x, y]/I
2
, where I
2
is the ideal spanned by homogeneous polynomials of degree
2 (so A has a basis 1, x, y). Let V = A

be the space of linear functionals on A, with the action

of A given by ((a)f)(b) = f(ba). Show that V provides such an example.
1.6 Algebras dened by generators and relations
If f
1
, . . . , f
m
are elements of the free algebra kx
1
, . . . , x
n
), we say that the algebra
A := kx
1
, . . . , x
n
)/f
1
, . . . , f
m
) is generated by x
1
, . . . , x
n
with dening relations f
1
= 0, . . . , f
m
=
0.
1.7 Examples of algebras
1. The Weyl algebra, kx, y)/yx xy 1).
2. The q-Weyl algebra, generated by x, x
1
, y, y
1
with dening relations yx = qxy and xx
1
=
x
1
x = yy
1
= y
1
y = 1.
Proposition. (i) A basis for the Weyl algebra A is x
i
y
j
, i, j 0.
(ii) A basis for the q-Weyl algebra A
q
is x
i
y
j
, i, j Z.
11
Proof. (i) First let us show that the elements x
i
y
j
are a spanning set for A. To do this, note that
any word in x, y can be ordered to have all the x on the left of the y, at the cost of interchanging
some x and y. Since yx xy = 1, this will lead to error terms, but these terms will be sums of
monomials that have a smaller number of letters x, y than the original word. Therefore, continuing
this process, we can order everything and represent any word as a linear combination of x
i
y
j
.
The proof that x
i
y
j
are linearly independent is based on representation theory. Namely, let a be
a variable, and E = t
a
k[a][t, t
1
] (here t
a
is just a formal symbol, so really E = k[a][t, t
1
]). Then E
is a representation of A with action given by xf = tf and yf =
df
dt
(where
d(t
a+n
)
dt
:= (a+n)t
a+n1
).
Suppose now that we have a nontrivial linear relation

c
ij
x
i
y
j
= 0. Then the operator
L =

c
ij
t
i
_
d
dt
_
j
acts by zero in E. Let us write L as
L =
r

j=0
Q
j
(t)
_
d
dt
_
j
,
where Q
r
,= 0. Then we have
Lt
a
=
r

j=0
Q
j
(t)a(a 1)...(a j + 1)t
aj
.
This must be zero, so we have

r
j=0
Q
j
(t)a(a 1)...(a j + 1)t
j
= 0 in k[a][t, t
1
]. Taking the
leading term in a, we get Q
r
(t) = 0, a contradiction.
(ii) Any word in x, y, x
1
, y
1
can be ordered at the cost of multiplying it by a power of q. This
easily implies both the spanning property and the linear independence.
Remark. The proof of (i) shows that the Weyl algebra A can be viewed as the algebra of
polynomial dierential operators in one variable t.
The proof of (i) also brings up the notion of a faithful representation.
Denition. A representation : A End V is faithful if is injective.
For example, k[t] is a faithful representation of the Weyl algebra, if k has characteristic zero
(check it!), but not in characteristic p, where (d/dt)
p
Q = 0 for any polynomial Q. However, the
representation E = t
a
k[a][t, t
1
], as weve seen, is faithful in any characteristic.
Problem 1.26. Let A be the Weyl algebra, generated by two elements x, y with the relation
yx xy 1 = 0.
(a) If chark = 0, what are the nite dimensional representations of A? What are the two-sided
ideals in A?
Hint. For the rst question, use the fact that for two square matrices B, C, Tr(BC) = Tr(CB).
For the second question, show that any nonzero two-sided ideal in A contains a nonzero polynomial
in x, and use this to characterize this ideal.
Suppose for the rest of the problem that chark = p.
(b) What is the center of A?
12
Hint. Show that x
p
and y
p
are central elements.
(c) Find all irreducible nite dimensional representations of A.
Hint. Let V be an irreducible nite dimensional representation of A, and v be an eigenvector
of y in V . Show that v, xv, x
2
v, ..., x
p1
v is a basis of V .
Problem 1.27. Let q be a nonzero complex number, and A be the q-Weyl algebra over C generated
by x
1
and y
1
with dening relations xx
1
= x
1
x = 1, yy
1
= y
1
y = 1, and xy = qyx.
(a) What is the center of A for dierent q? If q is not a root of unity, what are the two-sided
ideals in A?
(b) For which q does this algebra have nite dimensional representations?
Hint. Use determinants.
(c) Find all nite dimensional irreducible representations of A for such q.
Hint. This is similar to part (c) of the previous problem.
1.8 Quivers
Denition 1.28. A quiver Q is a directed graph, possibly with self-loops and/or multiple edges
between two vertices.
Example 1.29.

//

oo

OO
We denote the set of vertices of the quiver Q as I, and the set of edges as E. For an edge h E,
let h

, h

denote the source and target of h, respectively:

h
h
//

Denition 1.30. A representation of a quiver Q is an assignment to each vertex i I of a vector

space V
i
and to each edge h E of a linear map x
h
: V
h
V
h
.
It turns out that the theory of representations of quivers is a part of the theory of representations
of algebras in the sense that for each quiver Q, there exists a certain algebra P
Q
, called the path
algebra of Q, such that a representation of the quiver Q is the same as a representation of the
algebra P
Q
. We shall rst dene the path algebra of a quiver and then justify our claim that
representations of these two objects are the same.
Denition 1.31. The path algebra P
Q
of a quiver Q is the algebra whose basis is formed by
oriented paths in Q, including the trivial paths p
i
, i I, corresponding to the vertices of Q, and
multiplication is concatenation of paths: ab is the path obtained by rst tracing b and then a. If
two paths cannot be concatenated, the product is dened to be zero.
Remark 1.32. It is easy to see that for a nite quiver

iI
p
i
= 1, so P
Q
is an algebra with unit.
Problem 1.33. Show that the algebra P
Q
is generated by p
i
for i I and a
h
for h E with the
dening relations:
13
1. p
2
i
= p
i
, p
i
p
j
= 0 for i ,= j
2. a
h
p
h
= a
h
, a
h
p
j
= 0 for j ,= h

3. p
h
a
h
= a
h
, p
i
a
h
= 0 for i ,= h

We now justify our statement that a representation of a quiver is the same thing as a represen-
tation of the path algebra of a quiver.
Let V be a representation of the path algebra P
Q
. From this representation, we can construct a
representation of Q as follows: let V
i
= p
i
V, and for any edge h, let x
h
= a
h
[
p
h
V
: p
h
V p
h
V
be the operator corresponding to the one-edge path h.
Similarly, let (V
i
, x
h
) be a representation of a quiver Q. From this representation, we can
construct a representation of the path algebra P
Q
: let V =

i
V
i
, let p
i
: V V
i
V be the
projection onto V
i
, and for any path p = h
1
...h
m
let a
p
= x
h
1
...x
hm
: V
h

m
V
h

1
be the composition
of the operators corresponding to the edges occurring in p (and the action of this operator on the
other V
i
is zero).
It is clear that the above assignments V (p
i
V) and (V
i
)

i
V
i
are inverses of each other.
Thus, we have a bijection between isomorphism classes of representations of the algebra P
Q
and of
the quiver Q.
Remark 1.34. In practice, it is generally easier to consider a representation of a quiver as in
Denition 1.30.
We lastly dene several previous concepts in the context of quivers representations.
Denition 1.35. A subrepresentation of a representation (V
i
, x
h
) of a quiver Q is a representation
(W
i
, x

h
) where W
i
V
i
for all i I and where x
h
(W
h
) W
h
and x

h
= x
h
[
W
h

: W
h
W
h
for
all h E.
Denition 1.36. The direct sum of two representations (V
i
, x
h
) and (W
i
, y
h
) is the representation
(V
i
W
i
, x
h
y
h
).
As with representations of algebras, a nonzero representation (V
i
) of a quiver Q is said to be
irreducible if its only subrepresentations are (0) and (V
i
) itself, and indecomposable if it is not
isomorphic to a direct sum of two nonzero representations.
Denition 1.37. Let (V
i
, x
h
) and (W
i
, y
h
) be representations of the quiver Q. A homomorphism
: (V
i
) (W
i
) of quiver representations is a collection of maps
i
: V
i
W
i
such that
y
h

h
=
h
x
h
for all h E.
Problem 1.38. Let A be a Z
+
-graded algebra, i.e., A =
n0
A[n], and A[n] A[m] A[n + m].
If A[n] is nite dimensional, it is useful to consider the Hilbert series h
A
(t) =

dimA[n]t
n
(the
generating function of dimensions of A[n]). Often this series converges to a rational function, and
the answer is written in the form of such function. For example, if A = k[x] and deg(x
n
) = n then
h
A
(t) = 1 +t +t
2
+... +t
n
+... =
1
1 t
Find the Hilbert series of:
(a) A = k[x
1
, ..., x
m
] (where the grading is by degree of polynomials);
14
(b) A = k < x
1
, ..., x
m
> (the grading is by length of words);
(c) A is the exterior (=Grassmann) algebra
k
[x
1
, ..., x
m
], generated over some eld k by
x
1
, ..., x
m
with the dening relations x
i
x
j
+ x
j
x
i
= 0 and x
2
i
= 0 for all i, j (the grading is by
degree).
(d) A is the path algebra P
Q
of a quiver Q (the grading is dened by deg(p
i
) = 0, deg(a
h
) = 1).
Hint. The closed answer is written in terms of the adjacency matrix M
Q
of Q.
1.9 Lie algebras
Let g be a vector space over a eld k, and let [ , ] : g g g be a skew-symmetric bilinear map.
(That is, [a, a] = 0, and hence [a, b] = [b, a]).
Denition 1.39. (g, [ , ]) is a Lie algebra if [ , ] satises the Jacobi identity
_
[a, b] , c

+
_
[b, c] , a

+
_
[c, a] , b

= 0. (2)
Example 1.40. Some examples of Lie algebras are:
1. Any space g with [ , ] = 0 (abelian Lie algebra).
2. Any associative algebra A with [a, b] = ab ba .
3. Any subspace U of an associative algebra A such that [a, b] U for all a, b U.
4. The space Der(A) of derivations of an algebra A, i.e. linear maps D : A A which satisfy
the Leibniz rule:
D(ab) = D(a)b +aD(b).
Remark 1.41. Derivations are important because they are the innitesimal version of automor-
phisms (i.e., isomorphisms onto itself). For example, assume that g(t) is a dierentiable family of
automorphisms of a nite dimensional algebra A over R or C parametrized by t (, ) such that
g(0) = Id. Then D := g

(0) : A A is a derivation (check it!). Conversely, if D : A A is a

derivation, then e
tD
is a 1-parameter family of automorphisms (give a proof!).
This provides a motivation for the notion of a Lie algebra. Namely, we see that Lie algebras
arise as spaces of innitesimal automorphisms (=derivations) of associative algebras. In fact, they
similarly arise as spaces of derivations of any kind of linear algebraic structures, such as Lie algebras,
Hopf algebras, etc., and for this reason play a very important role in algebra.
Here are a few more concrete examples of Lie algebras:
1. R
3
with [u, v] = u v, the cross-product of u and v.
2. sl(n), the set of n n matrices with trace 0.
For example, sl(2) has the basis
e =
_
0 1
0 0
_
f =
_
0 0
1 0
_
h =
_
1 0
0 1
_
with relations
[h, e] = 2e, [h, f] = 2f, [e, f] = h.
15
3. The Heisenberg Lie algebra H of matrices
_
0
0 0
0 0 0
_
It has the basis
x =
_
_
0 0 0
0 0 1
0 0 0
_
_
y =
_
_
0 1 0
0 0 0
0 0 0
_
_
c =
_
_
0 0 1
0 0 0
0 0 0
_
_
with relations [y, x] = c and [y, c] = [x, c] = 0.
4. The algebra a(1) of matrices (

0 0
)
Its basis consists of X = (
1 0
0 0
) and Y = (
0 1
0 0
), with [X, Y ] = Y .
5. so(n), the space of skew-symmetric n n matrices, with [a, b] = ab ba.
Exercise. Show that Example 1 is a special case of Example 5 (for n = 3).
Denition 1.42. Let g
1
, g
2
be Lie algebras. A homomorphism : g
1
g
2
of Lie algebras is a
linear map such that ([a, b]) = [(a), (b)].
Denition 1.43. A representation of a Lie algebra g is a vector space V with a homomorphism
of Lie algebras : g EndV .
Example 1.44. Some examples of representations of Lie algebras are:
1. V = 0.
2. Any vector space V with = 0 (the trivial representation).
3. The adjoint representation V = g with (a)(b) := [a, b]. That this is a representation follows
from Equation (2). Thus, the meaning of the Jacobi identity is that it is equivalent to the
existence of the adjoint representation.
It turns out that a representation of a Lie algebra g is the same thing as a representation of a
certain associative algebra |(g). Thus, as with quivers, we can view the theory of representations
of Lie algebras as a part of the theory of representations of associative algebras.
Denition 1.45. Let g be a Lie algebra with basis x
i
and [ , ] dened by [x
i
, x
j
] =

k
c
k
ij
x
k
. The
universal enveloping algebra |(g) is the associative algebra generated by the x
i
s with the
dening relations x
i
x
j
x
j
x
i
=

k
c
k
ij
x
k
.
Remark. This is not a very good denition since it depends on the choice of a basis. Later we
will give an equivalent denition which will be basis-independent.
Exercise. Explain why a representation of a Lie algebra is the same thing as a representation
of its universal enveloping algebra.
Example 1.46. The associative algebra |(sl(2)) is the algebra generated by e, f, h with relations
he eh = 2e hf fh = 2f ef fe = h.
Example 1.47. The algebra |(H), where H is the Heisenberg Lie algebra, is the algebra generated
by x, y, c with the relations
yx xy = c yc cy = 0 xc cx = 0.
Note that the Weyl algebra is the quotient of |(H) by the relation c = 1.
16
1.10 Tensor products
In this subsection we recall the notion of tensor product of vector spaces, which will be extensively
used below.
Denition 1.48. The tensor product V W of vector spaces V and W over a eld k is the quotient
of the space V W whose basis is given by formal symbols v w, v V , w W, by the subspace
spanned by the elements
(v
1
+v
2
) wv
1
wv
2
w, v (w
1
+w
2
) v w
1
v w
2
, av wa(v w), v awa(v w),
where v V, w W, a k.
Exercise. Show that V W can be equivalently dened as the quotient of the free abelian
group V W generated by v w, v V, w W by the subgroup generated by
(v
1
+v
2
) w v
1
w v
2
w, v (w
1
+w
2
) v w
1
v w
2
, av w v aw,
where v V, w W, a k.
The elements v w V W, for v V, w W are called pure tensors. Note that in general,
there are elements of V W which are not pure tensors.
This allows one to dene the tensor product of any number of vector spaces, V
1
... V
n
. Note
that this tensor product is associative, in the sense that (V
1
V
2
) V
3
can be naturally identied
with V
1
(V
2
V
3
).
In particular, people often consider tensor products of the form V
n
= V ... V (n times) for
a given vector space V , and, more generally, E := V
n
(V

)
m
. This space is called the space of
tensors of type (m, n) on V . For instance, tensors of type (0, 1) are vectors, of type (1, 0) - linear
functionals (covectors), of type (1, 1) - linear operators, of type (2, 0) - bilinear forms, of type (2, 1)
- algebra structures, etc.
If V is nite dimensional with basis e
i
, i = 1, ..., N, and e
i
is the dual basis of V

, then a basis
of E is the set of vectors
e
i
1
... e
in
e
j
1
... e
jm
,
and a typical element of E is
N

i
1
,...,in,j
1
,...,jm=1
T
i
1
...in
j
1
...jm
e
i
1
... e
in
e
j
1
... e
jm
,
where T is a multidimensional table of numbers.
Physicists dene a tensor as a collection of such multidimensional tables T
B
attached to every
basis B in V , which change according to a certain rule when the basis B is changed. Here it is
important to distinguish upper and lower indices, since lower indices of T correspond to V and
upper ones to V

. The physicists dont write the sum sign, but remember that one should sum
over indices that repeat twice - once as an upper index and once as lower. This convention is
called the Einstein summation, and it also stipulates that if an index appears once, then there is
no summation over it, while no index is supposed to appear more than once as an upper index or
more than once as a lower index.
One can also dene the tensor product of linear maps. Namely, if A : V V

and B : W W

are linear maps, then one can dene the linear map AB : V W V

given by the formula

(A B)(v w) = Av Bw (check that this is well dened!)
17
The most important properties of tensor products are summarized in the following problem.
Problem 1.49. (a) Let U be any k-vector space. Construct a natural bijection between bilinear
maps V W U and linear maps V W U.
(b) Show that if v
i
is a basis of V and w
j
is a basis of W then v
i
w
j
is a basis of
V W.
(c) Construct a natural isomorphism V

W Hom(V, W) in the case when V is nite

dimensional (natural means that the isomorphism is dened without choosing bases).
(d) Let V be a vector space over a eld k. Let S
n
V be the quotient of V
n
(n-fold tensor product
of V ) by the subspace spanned by the tensors T s(T) where T V
n
, and s is some transposition.
Also let
n
V be the quotient of V
n
by the subspace spanned by the tensors T such that s(T) = T
for some transposition s. These spaces are called the n-th symmetric, respectively exterior, power
of V . If v
i
is a basis of V , can you construct a basis of S
n
V,
n
V ? If dimV = m, what are their
dimensions?
(e) If k has characteristic zero, nd a natural identication of S
n
V with the space of T V
n
such that T = sT for all transpositions s, and of
n
V with the space of T V
n
such that T = sT
for all transpositions s.
(f ) Let A : V W be a linear operator. Then we have an operator A
n
: V
n
W
n
, and
its symmetric and exterior powers S
n
A : S
n
V S
n
W,
n
A :
n
V
n
W which are dened in
an obvious way. Suppose V = W and has dimension N, and assume that the eigenvalues of A are

1
, ...,
N
. Find Tr(S
n
A), Tr(
n
A).
(g) Show that
N
A = det(A)Id, and use this equality to give a one-line proof of the fact that
det(AB) = det(A) det(B).
Remark. Note that a similar denition to the above can be used to dene the tensor product
V
A
W, where A is any ring, V is a right A-module, and W is a left A-module. Namely, V
A
W
is the abelian group which is the quotient of the group V W freely generated by formal symbols
v w, v V , w W, modulo the relations
(v
1
+v
2
) w v
1
w v
2
w, v (w
1
+w
2
) v w
1
v w
2
, va w v aw, a A.
Exercise. Throughout this exercise, we let k be an arbitrary eld (not necessarily of charac-
teristic zero, and not necessarily algebraically closed).
If A and B are two k-algebras, then an (A, B)-bimodule will mean a k-vector space V with
both a left A-module structure and a right B-module structure which satisfy (av) b = a (vb) for
any v V , a A and b B. Note that both the notions of left A-module and right A-
module are particular cases of the notion of bimodules; namely, a left A-module is the same as an
(A, k)-bimodule, and a right A-module is the same as a (k, A)-bimodule.
Let B be a k-algebra, W a left B-module and V a right B-module. We denote by V
B
W the
k-vector space (V
k
W) / vb w v bw [ v V, w W, b B). We denote the projection of
a pure tensor v w (with v V and w W) onto the space V
B
W by v
B
w. (Note that this
tensor product V
B
W is the one dened in the Remark after Problem1.49.)
If, additionally, A is another k-algebra, and if the right B-module structure on V is part of an
(A, B)-bimodule structure, then V
B
W becomes a left A-module by a (v
B
w) = av
B
w for
any a A, v V and w W.
18
Similarly, if C is another k-algebra, and if the left B-module structure on W is part of a (B, C)-
bimodule structure, then V
B
W becomes a right C-module by (v
B
w) c = v
B
wc for any
c C, v V and w W.
If V is an (A, B)-bimodule and W is a (B, C)-bimodule, then these two structures on V
B
W
can be combined into one (A, C)-bimodule structure on V
B
W.
(a) Let A, B, C, D be four algebras. Let V be an (A, B)-bimodule, W be a (B, C)-bimodule,
and X a (C, D)-bimodule. Prove that (V
B
W)
C
X

= V
B
(W
C
X) as (A, D)-bimodules.
The isomorphism (from left to right) is given by (v
B
w)
C
x v
B
(w
C
x) for all v V ,
w W and x X.
(b) If A, B, C are three algebras, and if V is an (A, B)-bimodule and W an (A, C)-bimodule,
then the vector space Hom
A
(V, W) (the space of all left A-linear homomorphisms from V to W)
canonically becomes a (B, C)-bimodule by setting (bf) (v) = f (vb) for all b B, f Hom
A
(V, W)
and v V and (fc) (v) = f (v) c for all c C, f Hom
A
(V, W) and v V .
Let A, B, C, D be four algebras. Let V be a (B, A)-bimodule, W be a (C, B)-bimodule, and X a
(C, D)-bimodule. Prove that Hom
B
(V, Hom
C
(W, X))

= Hom
C
(W
B
V, X) as (A, D)-bimodules.
The isomorphism (from left to right) is given by f (w
B
v f (v) w) for all v V , w W
and f Hom
B
(V, Hom
C
(W, X)).
1.11 The tensor algebra
The notion of tensor product allows us to give more conceptual (i.e., coordinate free) denitions
of the free algebra, polynomial algebra, exterior algebra, and universal enveloping algebra of a Lie
algebra.
Namely, given a vector space V , dene its tensor algebra TV over a eld k to be TV =
n0
V
n
,
with multiplication dened by a b := a b, a V
n
, b V
m
. Observe that a choice of a basis
x
1
, ..., x
N
in V denes an isomorphism of TV with the free algebra k < x
1
, ..., x
n
>.
Also, one can make the following denition.
Denition 1.50. (i) The symmetric algebra SV of V is the quotient of TV by the ideal generated
by v w w v, v, w V .
(ii) The exterior algebra V of V is the quotient of TV by the ideal generated by v v, v V .
(iii) If V is a Lie algebra, the universal enveloping algebra U(V ) of V is the quotient of TV by
the ideal generated by v w w v [v, w], v, w V .
It is easy to see that a choice of a basis x
1
, ..., x
N
in V identies SV with the polynomial algebra
k[x
1
, ..., x
N
], V with the exterior algebra
k
(x
1
, ..., x
N
), and the universal enveloping algebra U(V )
with one dened previously.
Also, it is easy to see that we have decompositions SV =
n0
S
n
V , V =
n0

n
V .
1.12 Hilberts third problem
Problem 1.51. It is known that if A and B are two polygons of the same area then A can be cut
by nitely many straight cuts into pieces from which one can make B. David Hilbert asked in 1900
whether it is true for polyhedra in 3 dimensions. In particular, is it true for a cube and a regular
tetrahedron of the same volume?
19
The answer is no, as was found by Dehn in 1901. The proof is very beautiful. Namely, to
any polyhedron A let us attach its Dehn invariant D(A) in V = R (R/Q) (the tensor product
of Q-vector spaces). Namely,
D(A) =

a
l(a)
(a)

,
where a runs over edges of A, and l(a), (a) are the length of a and the angle at a.
(a) Show that if you cut A into B and C by a straight cut, then D(A) = D(B) +D(C).
(b) Show that = arccos(1/3)/ is not a rational number.
Hint. Assume that = 2m/n, for integers m, n. Deduce that roots of the equation x+x
1
= 2/3
are roots of unity of degree n. Conclude that x
k
+x
k
has denominator 3
k
and get a contradiction.
(c) Using (a) and (b), show that the answer to Hilberts question is negative. (Compute the
Dehn invariant of the regular tetrahedron and the cube).
1.13 Tensor products and duals of representations of Lie algebras
Denition 1.52. The tensor product of two representations V, W of a Lie algebra g is the space
V W with
V W
(x) =
V
(x) Id +Id
W
(x).
Denition 1.53. The dual representation V

to a representation V of a Lie algebra g is the dual

space V

to V with
V
(x) =
V
(x)

.
It is easy to check that these are indeed representations.
Problem 1.54. Let V, W, U be nite dimensional representations of a Lie algebra g. Show that
the space Hom
g
(V W, U) is isomorphic to Hom
g
(V, U W

). (Here Hom
g
:= Hom
|(g)
).
1.14 Representations of sl(2)
This subsection is devoted to the representation theory of sl(2), which is of central importance in
many areas of mathematics. It is useful to study this topic by solving the following sequence of
exercises, which every mathematician should do, in one form or another.
Problem 1.55. According to the above, a representation of sl(2) is just a vector space V with a
triple of operators E, F, H such that HE EH = 2E, HF FH = 2F, EF FE = H (the
corresponding map is given by (e) = E, (f) = F, (h) = H).
Let V be a nite dimensional representation of sl(2) (the ground eld in this problem is C).
(a) Take eigenvalues of H and pick one with the biggest real part. Call it . Let

V () be the
generalized eigenspace corresponding to . Show that E[
V ()
= 0.
(b) Let W be any representation of sl(2) and w W be a nonzero vector such that Ew = 0.
For any k > 0 nd a polynomial P
k
(x) of degree k such that E
k
F
k
w = P
k
(H)w. (First compute
EF
k
w, then use induction in k).
(c) Let v

V () be a generalized eigenvector of H with eigenvalue . Show that there exists
N > 0 such that F
N
v = 0.
(d) Show that H is diagonalizable on

V (). (Take N to be such that F
N
= 0 on

V (), and
compute E
N
F
N
v, v

V (), by (b). Use the fact that P
k
(x) does not have multiple roots).
20
(e) Let N
v
be the smallest N satisfying (c). Show that = N
v
1.
(f ) Show that for each N > 0, there exists a unique up to isomorphism irreducible representation
of sl(2) of dimension N. Compute the matrices E, F, H in this representation using a convenient
basis. (For V nite dimensional irreducible take as in (a) and v V () an eigenvector of H.
Show that v, Fv, ..., F

v is a basis of V , and compute the matrices of the operators E, F, H in this

basis.)
Denote the + 1-dimensional irreducible representation from (f ) by V

. Below you will show

that any nite dimensional representation is a direct sum of V

.
(g) Show that the operator C = EF + FE + H
2
/2 (the so-called Casimir operator) commutes
with E, F, H and equals
(+2)
2
Id on V

.
Now it will be easy to prove the direct sum decomposition. Namely, assume the contrary, and
let V be a reducible representation of the smallest dimension, which is not a direct sum of smaller
representations.
(h) Show that C has only one eigenvalue on V , namely
(+2)
2
for some nonnegative integer .
(use that the generalized eigenspace decomposition of C must be a decomposition of representations).
(i) Show that V has a subrepresentation W = V

such that V/W = nV

for some n (use (h)

and the fact that V is the smallest which cannot be decomposed).
(j) Deduce from (i) that the eigenspace V () of H is n + 1-dimensional. If v
1
, ..., v
n+1
is its
basis, show that F
j
v
i
, 1 i n +1, 0 j are linearly independent and therefore form a basis
of V (establish that if Fx = 0 and Hx = x then Cx =
(2)
2
x and hence = ).
(k) Dene W
i
= span(v
i
, Fv
i
, ..., F

v
i
). Show that V
i
are subrepresentations of V and derive a
contradiction with the fact that V cannot be decomposed.
(l) (Jacobson-Morozov Lemma) Let V be a nite dimensional complex vector space and A : V
V a nilpotent operator. Show that there exists a unique, up to an isomorphism, representation of
sl(2) on V such that E = A. (Use the classication of the representations and the Jordan normal
form theorem)
(m) (Clebsch-Gordan decomposition) Find the decomposition into irreducibles of the represen-
tation V

of sl(2).
Hint. For a nite dimensional representation V of sl(2) it is useful to introduce the character

V
(x) = Tr(e
xH
), x C. Show that
V W
(x) =
V
(x) +
W
(x) and
V W
(x) =
V
(x)
W
(x).
Then compute the character of V

and of V

and derive the decomposition. This decomposition

is of fundamental importance in quantum mechanics.
(n) Let V = C
M
C
N
, and A = J
M
(0) Id
N
+Id
M
J
N
(0), where J
n
(0) is the Jordan block
of size n with eigenvalue zero (i.e., J
n
(0)e
i
= e
i1
, i = 2, ..., n, and J
n
(0)e
1
= 0). Find the Jordan
normal form of A using (l),(m).
1.15 Problems on Lie algebras
Problem 1.56. (Lies Theorem) The commutant K(g) of a Lie algebra g is the linear span
of elements [x, y], x, y g. This is an ideal in g (i.e., it is a subrepresentation of the adjoint
representation). A nite dimensional Lie algebra g over a eld k is said to be solvable if there
exists n such that K
n
(g) = 0. Prove the Lie theorem: if k = C and V is a nite dimensional
irreducible representation of a solvable Lie algebra g then V is 1-dimensional.
21
Hint. Prove the result by induction in dimension. By the induction assumption, K(g) has a
common eigenvector v in V , that is there is a linear function : K(g) C such that av = (a)v
for any a K(g). Show that g preserves common eigenspaces of K(g) (for this you will need to
show that ([x, a]) = 0 for x g and a K(g). To prove this, consider the smallest vector subspace
U containing v and invariant under x. This subspace is invariant under K(g) and any a K(g)
acts with trace dim(U)(a) in this subspace. In particular 0 = Tr([x, a]) = dim(U)([x, a]).).
Problem 1.57. Classify irreducible nite dimensional representations of the two dimensional Lie
algebra with basis X, Y and commutation relation [X, Y ] = Y . Consider the cases of zero and
positive characteristic. Is the Lie theorem true in positive characteristic?
Problem 1.58. (hard!) For any element x of a Lie algebra g let ad(x) denote the operator g
g, y [x, y]. Consider the Lie algebra g
n
generated by two elements x, y with the dening relations
ad(x)
2
(y) = ad(y)
n+1
(x) = 0.
(a) Show that the Lie algebras g
1
, g
2
, g
3
are nite dimensional and nd their dimensions.
(b) (harder!) Show that the Lie algebra g
4
has innite dimension. Construct explicitly a basis
of this algebra.
22
2 General results of representation theory
2.1 Subrepresentations in semisimple representations
Let A be an algebra.
Denition 2.1. A semisimple (or completely reducible) representation of A is a direct sum of
irreducible representations.
Example. Let V be an irreducible representation of A of dimension n. Then Y = End(V ),
with action of A by left multiplication, is a semisimple representation of A, isomorphic to nV (the
direct sum of n copies of V ). Indeed, any basis v
1
, ..., v
n
of V gives rise to an isomorphism of
representations End(V ) nV , given by x (xv
1
, ..., xv
n
).
Remark. Note that by Schurs lemma, any semisimple representation V of A is canonically
identied with
X
Hom
A
(X, V )X, where X runs over all irreducible representations of A. Indeed,
we have a natural map f :
X
Hom(X, V )X V , given by gx g(x), x X, g Hom(X, V ),
and it is easy to verify that this map is an isomorphism.
Well see now how Schurs lemma allows us to classify subrepresentations in nite dimensional
semisimple representations.
Proposition 2.2. Let V
i
, 1 i m be irreducible nite dimensional pairwise nonisomorphic
representations of A, and W be a subrepresentation of V =
m
i=1
n
i
V
i
. Then W is isomorphic to

m
i=1
r
i
V
i
, r
i
n
i
, and the inclusion : W V is a direct sum of inclusions
i
: r
i
V
i
n
i
V
i
given
by multiplication of a row vector of elements of V
i
(of length r
i
) by a certain r
i
-by-n
i
matrix X
i
with linearly independent rows: (v
1
, ..., v
r
i
) = (v
1
, ..., v
r
i
)X
i
.
Proof. The proof is by induction in n :=

m
i=1
n
i
. The base of induction (n = 1) is clear. To perform
the induction step, let us assume that W is nonzero, and x an irreducible subrepresentation
P W. Such P exists (Problem 1.20).
2
Now, by Schurs lemma, P is isomorphic to V
i
for some i,
and the inclusion [
P
: P V factors through n
i
V
i
, and upon identication of P with V
i
is given
by the formula v (vq
1
, ..., vq
n
i
), where q
l
k are not all zero.
Now note that the group G
i
= GL
n
i
(k) of invertible n
i
-by-n
i
matrices over k acts on n
i
V
i
by (v
1
, ..., v
n
i
) (v
1
, ..., v
n
i
)g
i
(and by the identity on n
j
V
j
, j ,= i), and therefore acts on the
set of subrepresentations of V , preserving the property we need to establish: namely, under the
action of g
i
, the matrix X
i
goes to X
i
g
i
, while X
j
, j ,= i dont change. Take g
i
G
i
such that
(q
1
, ..., q
n
i
)g
i
= (1, 0, ..., 0). Then Wg
i
contains the rst summand V
i
of n
i
V
i
(namely, it is Pg
i
),
hence Wg
i
= V
i
W

, where W

n
1
V
1
... (n
i
1)V
i
... n
m
V
m
is the kernel of the projection
of Wg
i
to the rst summand V
i
along the other summands. Thus the required statement follows
from the induction assumption.
Remark 2.3. In Proposition 2.2, it is not important that k is algebraically closed, nor it matters
that V is nite dimensional. If these assumptions are dropped, the only change needed is that the
entries of the matrix X
i
are no longer in k but in D
i
= End
A
(V
i
), which is, as we know, a division
algebra. The proof of this generalized version of Proposition 2.2 is the same as before (check it!).
2
Another proof of the existence of P, which does not use the nite dimensionality of V , is by induction in n.
Namely, if W itself is not irreducible, let K be the kernel of the projection of W to the rst summand V1. Then
K is a subrepresentation of (n1 1)V1 ... nmVm, which is nonzero since W is not irreducible, so K contains an
irreducible subrepresentation by the induction assumption.
23
2.2 The density theorem
Let A be an algebra over an algebraically closed eld k.
Corollary 2.4. Let V be an irreducible nite dimensional representation of A, and v
1
, ..., v
n
V
be any linearly independent vectors. Then for any w
1
, ..., w
n
V there exists an element a A
such that av
i
= w
i
.
Proof. Assume the contrary. Then the image of the map A nV given by a (av
1
, ..., av
n
) is a
proper subrepresentation, so by Proposition 2.2 it corresponds to an r-by-n matrix X, r < n. Thus,
taking a = 1, we see that there exist vectors u
1
, ..., u
r
V such that (u
1
, ..., u
r
)X = (v
1
, ..., v
n
). Let
(q
1
, ..., q
n
) be a nonzero vector such that X(q
1
, ..., q
n
)
T
= 0 (it exists because r < n). Then

q
i
v
i
=
(u
1
, ..., u
r
)X(q
1
, ..., q
n
)
T
= 0, i.e.

q
i
v
i
= 0 - a contradiction with the linear independence of
v
i
.
Theorem 2.5. (the Density Theorem). (i) Let V be an irreducible nite dimensional representation
of A. Then the map : A EndV is surjective.
(ii) Let V = V
1
... V
r
, where V
i
are irreducible pairwise nonisomorphic nite dimensional
representations of A. Then the map
r
i=1

i
: A
r
i=1
End(V
i
) is surjective.
Proof. (i) Let B be the image of A in End(V ). We want to show that B = End(V ). Let c End(V ),
v
1
, ..., v
n
be a basis of V , and w
i
= cv
i
. By Corollary 2.4, there exists a A such that av
i
= w
i
.
Then a maps to c, so c B, and we are done.
(ii) Let B
i
be the image of A in End(V
i
), and B be the image of A in
r
i=1
End(V
i
). Recall that as
a representation of A,
r
i=1
End(V
i
) is semisimple: it is isomorphic to
r
i=1
d
i
V
i
, where d
i
= dimV
i
.
Then by Proposition 2.2, B =
i
B
i
. On the other hand, (i) implies that B
i
= End(V
i
). Thus (ii)
follows.
2.3 Representations of direct sums of matrix algebras
In this section we consider representations of algebras A =

i
Mat
d
i
(k) for any eld k.
Theorem 2.6. Let A =

r
i=1
Mat
d
i
(k). Then the irreducible representations of A are V
1
=
k
d
1
, . . . , V
r
= k
dr
, and any nite dimensional representation of A is a direct sum of copies of
V
1
, . . . , V
r
.
In order to prove Theorem 2.6, we shall need the notion of a dual representation.
Denition 2.7. (Dual representation) Let V be a representation of any algebra A. Then the
dual representation V

is the representation of the opposite algebra A
op
(or, equivalently, right
A-module) with the action
(f a)(v) := f(av).
Proof of Theorem 2.6. First, the given representations are clearly irreducible, as for any v ,= 0, w
V
i
, there exists a A such that av = w. Next, let X be an n-dimensional representation of
A. Then, X

is an n-dimensional representation of A
op
. But (Mat
d
i
(k))
op

= Mat
d
i
(k) with
isomorphism (X) = X
T
, as (BC)
T
= C
T
B
T
. Thus, A

= A
op
and X

may be viewed as an
n-dimensional representation of A. Dene
: A A
. .
n copies
X

24
by
(a
1
, . . . , a
n
) = a
1
y
1
+ +a
n
y
n
where y
i
is a basis of X

. is clearly surjective, as k A. Thus, the dual map

: X A
n
is injective. But A
n

= A
n
as representations of A (check it!). Hence, Im

= X is a subrepresen-
tation of A
n
. Next, Mat
d
i
(k) = d
i
V
i
, so A =
r
i=1
d
i
V
i
, A
n
=
r
i=1
nd
i
V
i
, as a representation of A.
Hence by Proposition 2.2, X =
r
i=1
m
i
V
i
, as desired.
Exercise. The goal of this exercise is to give an alternative proof of Theorem 2.6, not using
any of the previous results of Chapter 2.
Let A
1
, A
2
, ..., A
n
be n algebras with units 1
1
, 1
2
, ..., 1
n
, respectively. Let A = A
1
A
2
...A
n
.
Clearly, 1
i
1
j
=
ij
1
i
, and the unit of A is 1 = 1
1
+ 1
2
+... + 1
n
.
For every representation V of A, it is easy to see that 1
i
V is a representation of A
i
for every
i 1, 2, ..., n. Conversely, if V
1
, V
2
, ..., V
n
are representations of A
1
, A
2
, ..., A
n
, respectively,
then V
1
V
2
... V
n
canonically becomes a representation of A (with (a
1
, a
2
, ..., a
n
) A acting
on V
1
V
2
... V
n
as (v
1
, v
2
, ..., v
n
) (a
1
v
1
, a
2
v
2
, ..., a
n
v
n
)).
(a) Show that a representation V of A is irreducible if and only if 1
i
V is an irreducible repre-
sentation of A
i
for exactly one i 1, 2, ..., n, while 1
i
V = 0 for all the other i. Thus, classify the
irreducible representations of A in terms of those of A
1
, A
2
, ..., A
n
.
(b) Let d N. Show that the only irreducible representation of Mat
d
(k) is k
d
, and every nite
dimensional representation of Mat
d
(k) is a direct sum of copies of k
d
.
Hint: For every (i, j) 1, 2, ..., d
2
, let E
ij
Mat
d
(k) be the matrix with 1 in the ith row of the
jth column and 0s everywhere else. Let V be a nite dimensional representation of Mat
d
(k). Show
that V = E
11
V E
22
V ... E
dd
V , and that
i
: E
11
V E
ii
V , v E
i1
v is an isomorphism for
every i 1, 2, ..., d. For every v E
11
V , denote S (v) = E
11
v, E
21
v, ..., E
d1
v). Prove that S (v)
is a subrepresentation of V isomorphic to k
d
(as a representation of Mat
d
(k)), and that v S (v).
Conclude that V = S (v
1
) S (v
2
) ... S (v
k
), where v
1
, v
2
, ..., v
k
is a basis of E
11
V .
(c) Conclude Theorem 2.6.
2.4 Filtrations
Let A be an algebra. Let V be a representation of A. A (nite) ltration of V is a sequence of
subrepresentations 0 = V
0
V
1
... V
n
= V .
Lemma 2.8. Any nite dimensional representation V of an algebra A admits a nite ltration
0 = V
0
V
1
... V
n
= V such that the successive quotients V
i
/V
i1
are irreducible.
Proof. The proof is by induction in dim(V ). The base is clear, and only the induction step needs
to be justied. Pick an irreducible subrepresentation V
1
V , and consider the representation
U = V/V
1
. Then by the induction assumption U has a ltration 0 = U
0
U
1
... U
n1
= U
such that U
i
/U
i1
are irreducible. Dene V
i
for i 2 to be the preimages of U
i1
under the
tautological projection V V/V
1
= U. Then 0 = V
0
V
1
V
2
... V
n
= V is a ltration of V
with the desired property.
25
2.5 Finite dimensional algebras
Denition 2.9. The radical of a nite dimensional algebra A is the set of all elements of A which
act by 0 in all irreducible representations of A. It is denoted Rad(A).
Proposition 2.10. Rad(A) is a two-sided ideal.
Proof. Easy.
Proposition 2.11. Let A be a nite dimensional algebra.
(i) Let I be a nilpotent two-sided ideal in A, i.e., I
n
= 0 for some n. Then I Rad(A).
(ii) Rad(A) is a nilpotent ideal. Thus, Rad(A) is the largest nilpotent two-sided ideal in A.
Proof. (i) Let V be an irreducible representation of A. Let v V . Then Iv V is a subrepresen-
tation. If Iv ,= 0 then Iv = V so there is x I such that xv = v. Then x
n
,= 0, a contradiction.
Thus Iv = 0, so I acts by 0 in V and hence I Rad(A).
(ii) Let 0 = A
0
A
1
... A
n
= A be a ltration of the regular representation of A by
subrepresentations such that A
i+1
/A
i
are irreducible. It exists by Lemma 2.8. Let x Rad(A).
Then x acts on A
i+1
/A
i
by zero, so x maps A
i+1
to A
i
. This implies that Rad(A)
n
= 0, as
desired.
Theorem 2.12. A nite dimensional algebra A has only nitely many irreducible representations
V
i
up to isomorphism, these representations are nite dimensional, and
A/Rad(A)

i
EndV
i
.
Proof. First, for any irreducible representation V of A, and for any nonzero v V , Av V is a
nite dimensional subrepresentation of V . (It is nite dimensional as A is nite dimensional.) As
V is irreducible and Av ,= 0, V = Av and V is nite dimensional.
Next, suppose we have non-isomorphic irreducible representations V
1
, V
2
, . . . , V
r
. By Theorem
2.5, the homomorphism

i
: A

i
EndV
i
is surjective. So r

i
dimEndV
i
dimA. Thus, A has only nitely many non-isomorphic
irreducible representations (at most dimA).
Now, let V
1
, V
2
, . . . , V
r
be all non-isomorphic irreducible nite dimensional representations of
A. By Theorem 2.5, the homomorphism

i
: A

i
EndV
i
is surjective. The kernel of this map, by denition, is exactly Rad(A).
Corollary 2.13.

i
(dimV
i
)
2
dimA, where the V
i
s are the irreducible representations of A.
Proof. As dimEndV
i
= (dimV
i
)
2
, Theorem 2.12 implies that dimAdimRad(A) =

i
dimEndV
i
=

i
(dimV
i
)
2
. As dimRad(A) 0,

i
(dimV
i
)
2
dimA.
26
Example 2.14. 1. Let A = k[x]/(x
n
). This algebra has a unique irreducible representation, which
is a 1-dimensional space k, in which x acts by zero. So the radical Rad(A) is the ideal (x).
2. Let A be the algebra of upper triangular n by n matrices. It is easy to check that the
irreducible representations of A are V
i
, i = 1, ..., n, which are 1-dimensional, and any matrix x acts
by x
ii
. So the radical Rad(A) is the ideal of strictly upper triangular matrices (as it is a nilpotent
ideal and contains the radical). A similar result holds for block-triangular matrices.
Denition 2.15. A nite dimensional algebra A is said to be semisimple if Rad(A) = 0.
Proposition 2.16. For a nite dimensional algebra A, the following are equivalent:
1. A is semisimple.
2.

i
(dimV
i
)
2
= dimA, where the V
i
s are the irreducible representations of A.
3. A

i
Mat
d
i
(k) for some d
i
.
4. Any nite dimensional representation of A is completely reducible (that is, isomorphic to a
direct sum of irreducible representations).
5. A is a completely reducible representation of A.
Proof. As dimAdimRad(A) =

i
(dimV
i
)
2
, clearly dimA =

i
(dimV
i
)
2
if and only if Rad(A) =
0. Thus, (1) (2).
Next, by Theorem 2.12, if Rad(A) = 0, then clearly A

=

i
Mat
d
i
(k) for d
i
= dimV
i
. Thus,
(1) (3). Conversely, if A

i
Mat
d
i
(k), then by Theorem 2.6, Rad(A) = 0, so A is semisimple.
Thus (3) (1).
Next, (3) (4) by Theorem 2.6. Clearly (4) (5). To see that (5) (3), let A =

i
n
i
V
i
.
Consider End
A
(A) (endomorphisms of A as a representation of A). As the V
i
s are pairwise non-
isomorphic, by Schurs lemma, no copy of V
i
in A can be mapped to a distinct V
j
. Also, again by
Schurs lemma, End
A
(V
i
) = k. Thus, End
A
(A)

=

i
Mat
n
i
(k). But End
A
(A)

= A
op
by Problem
1.22, so A
op
=

i
Mat
n
i
(k). Thus, A

= (

i
Mat
n
i
(k))
op
=

i
Mat
n
i
(k), as desired.
2.6 Characters of representations
Let A be an algebra and V a nite-dimensional representation of A with action . Then the
character of V is the linear function
V
: A k given by

V
(a) = tr[
V
((a)).
If [A, A] is the span of commutators [x, y] := xy yx over all x, y A, then [A, A] ker
V
. Thus,
we may view the character as a mapping
V
: A/[A, A] k.
Exercise. Show that if W V are nite dimensional representations of A, then
V
=
W
+

V/W
.
Theorem 2.17. (i) Characters of (distinct) irreducible nite-dimensional representations of A are
linearly independent.
(ii) If A is a nite-dimensional semisimple algebra, then these characters form a basis of
(A/[A, A])

.
27
Proof. (i) If V
1
, . . . , V
r
are nonisomorphic irreducible nite-dimensional representations of A, then

V
1

Vr
: A End V
1
End V
r
is surjective by the density theorem, so
V
1
, . . . ,
Vr
are
linearly independent. (Indeed, if

V
i
(a) = 0 for all a A, then

i
Tr(M
i
) = 0 for all M
i

End
k
V
i
. But each tr(M
i
) can range independently over k, so it must be that
1
= =
r
= 0.)
(ii) First we prove that [Mat
d
(k), Mat
d
(k)] = sl
d
(k), the set of all matrices with trace 0. It is
clear that [Mat
d
(k), Mat
d
(k)] sl
d
(k). If we denote by E
ij
the matrix with 1 in the ith row of the
jth column and 0s everywhere else, we have [E
ij
, E
jm
] = E
im
for i ,= m, and [E
i,i+1
, E
i+1,i
] = E
ii

E
i+1,i+1
. Now E
im
E
ii
E
i+1,i+1
forms a basis in sl
d
(k), so indeed [Mat
d
(k), Mat
d
(k)] = sl
d
(k),
as claimed.
By semisimplicity, we can write A = Mat
d
1
(k) Mat
dr
(k). Then [A, A] = sl
d
1
(k)
sl
dr
(k), and A/[A, A]

= k
r
. By Theorem 2.6, there are exactly r irreducible representations of A
(isomorphic to k
d
1
, . . . , k
dr
, respectively), and therefore r linearly independent characters on the
r-dimensional vector space A/[A, A]. Thus, the characters form a basis.
2.7 The Jordan-Holder theorem
We will now state and prove two important theorems about representations of nite dimensional
algebras - the Jordan-Holder theorem and the Krull-Schmidt theorem.
Theorem 2.18. (Jordan-H older theorem). Let V be a nite dimensional representation of A,
and 0 = V
0
V
1
... V
n
= V , 0 = V

0
... V

m
= V be ltrations of V , such that the
representations W
i
:= V
i
/V
i1
and W

i
:= V

i
/V

i1
are irreducible for all i. Then n = m, and there
exists a permutation of 1, ..., n such that W
(i)
is isomorphic to W

i
.
Proof. First proof (for k of characteristic zero). The character of V obviously equals the sum
of characters of W
i
, and also the sum of characters of W

i
. But by Theorem 2.17, the charac-
ters of irreducible representations are linearly independent, so the multiplicity of every irreducible
representation W of A among W
i
and among W

i
are the same. This implies the theorem.
3
Second proof (general). The proof is by induction on dimV . The base of induction is clear,
so let us prove the induction step. If W
1
= W

1
(as subspaces), we are done, since by the induction
assumption the theorem holds for V/W
1
. So assume W
1
,= W

1
. In this case W
1
W

1
= 0 (as
W
1
, W

1
are irreducible), so we have an embedding f : W
1
W

1
V . Let U = V/(W
1
W

1
), and
0 = U
0
U
1
... U
p
= U be a ltration of U with simple quotients Z
i
= U
i
/U
i1
(it exists by
Lemma 2.8). Then we see that:
1) V/W
1
has a ltration with successive quotients W

1
, Z
1
, ..., Z
p
, and another ltration with
successive quotients W
2
, ...., W
n
.
2) V/W

1
has a ltration with successive quotients W
1
, Z
1
, ..., Z
p
, and another ltration with
successive quotients W

2
, ...., W

n
.
By the induction assumption, this means that the collection of irreducible representations with
multiplicities W
1
, W

1
, Z
1
, ..., Z
p
coincides on one hand with W
1
, ..., W
n
, and on the other hand, with
W

1
, ..., W

m
. We are done.
The Jordan-Holder theorem shows that the number n of terms in a ltration of V with irre-
ducible successive quotients does not depend on the choice of a ltration, and depends only on
3
This proof does not work in characteristic p because it only implies that the multiplicities of Wi and W

i
are the
same modulo p, which is not sucient. In fact, the character of the representation pV , where V is any representation,
is zero.
28
V . This number is called the length of V . It is easy to see that n is also the maximal length of a
ltration of V in which all the inclusions are strict.
The sequence of the irreducible representations W
1
, ..., W
n
enumerated in the order they appear
from some ltration of V as successive quoteints is called a Jordan-Holder series of V .
2.8 The Krull-Schmidt theorem
Theorem 2.19. (Krull-Schmidt theorem) Any nite dimensional representation of A can be uniquely
(up to an isomorphism and order of summands) decomposed into a direct sum of indecomposable
representations.
Proof. It is clear that a decomposition of V into a direct sum of indecomposable representations
exists, so we just need to prove uniqueness. We will prove it by induction on dimV . Let V =
V
1
... V
m
= V

1
... V

n
. Let i
s
: V
s
V , i

s
: V

s
V , p
s
: V V
s
, p

s
: V V

s
be the natural
maps associated to these decompositions. Let
s
= p
1
i

s
p

s
i
1
: V
1
V
1
. We have

n
s=1

s
= 1. Now
we need the following lemma.
Lemma 2.20. Let W be a nite dimensional indecomposable representation of A. Then
(i) Any homomorphism : W W is either an isomorphism or nilpotent;
(ii) If
s
: W W, s = 1, ..., n are nilpotent homomorphisms, then so is :=
1
+... +
n
.
Proof. (i) Generalized eigenspaces of are subrepresentations of W, and W is their direct sum.
Thus, can have only one eigenvalue . If is zero, is nilpotent, otherwise it is an isomorphism.
(ii) The proof is by induction in n. The base is clear. To make the induction step (n 1 to n),
assume that is not nilpotent. Then by (i) is an isomorphism, so

n
i=1

i
= 1. The morphisms

i
are not isomorphisms, so they are nilpotent. Thus 1
1

n
=
1

1
+ ... +
1

n1
is an
isomorphism, which is a contradiction with the induction assumption.
By the lemma, we nd that for some s,
s
must be an isomorphism; we may assume that
s = 1. In this case, V

1
= Im(p

1
i
1
) Ker(p
1
i

1
), so since V

1
is indecomposable, we get that
f := p

1
i
1
: V
1
V

1
and g := p
1
i

1
: V

1
V
1
are isomorphisms.
Let B =
j>1
V
j
, B

=
j>1
V

j
; then we have V = V
1
B = V

1
B

. Consider the map

h : B B

dened as a composition of the natural maps B V B

attached to these
decompositions. We claim that h is an isomorphism. To show this, it suces to show that Kerh = 0
(as h is a map between spaces of the same dimension). Assume that v Kerh B. Then v V

1
.
On the other hand, the projection of v to V
1
is zero, so gv = 0. Since g is an isomorphism, we get
v = 0, as desired.
Now by the induction assumption, m = n, and V
j

= V

(j)
for some permutation of 2, ..., n.
The theorem is proved.
Exercise. Let A be the algebra of real-valued continuous functions on R which are periodic
with period 1. Let M be the A-module of continuous functions f on R which are antiperiodic with
period 1, i.e., f(x + 1) = f(x).
(i) Show that A and M are indecomposable A-modules.
(ii) Show that A is not isomorphic to M but AA is isomorphic to M M.
29
Remark. Thus, we see that in general, the Krull-Schmidt theorem fails for innite dimensional
modules. However, it still holds for modules of nite length, i.e., modules M such that any ltration
of M has length bounded above by a certain constant l = l(M).
2.9 Problems
Problem 2.21. Extensions of representations. Let A be an algebra, and V, W be a pair of
representations of A. We would like to classify representations U of A such that V is a subrepre-
sentation of U, and U/V = W. Of course, there is an obvious example U = V W, but are there
any others?
Suppose we have a representation U as above. As a vector space, it can be (non-uniquely)
identied with V W, so that for any a A the corresponding operator
U
(a) has block triangular
form

U
(a) =
_

V
(a) f(a)
0
W
(a)
_
,
where f : A Hom
k
(W, V ) is a linear map.
(a) What is the necessary and sucient condition on f(a) under which
U
(a) is a repre-
sentation? Maps f satisfying this condition are called (1-)cocycles (of A with coecients in
Hom
k
(W, V )). They form a vector space denoted Z
1
(W, V ).
(b) Let X : W V be a linear map. The coboundary of X, dX, is dened to be the function A
Hom
k
(W, V ) given by dX(a) =
V
(a)XX
W
(a). Show that dX is a cocycle, which vanishes if and
only if X is a homomorphism of representations. Thus coboundaries form a subspace B
1
(W, V )
Z
1
(W, V ), which is isomorphic to Hom
k
(W, V )/Hom
A
(W, V ). The quotient Z
1
(W, V )/B
1
(W, V ) is
denoted Ext
1
(W, V ).
(c) Show that if f, f

Z
1
(W, V ) and f f

B
1
(W, V ) then the corresponding extensions
U, U

are isomorphic representations of A. Conversely, if : U U

is an isomorphism such that

(a) =
_
1
V

0 1
W
_
then f f

B
1
(V, W). Thus, the space Ext
1
(W, V ) classies extensions of W by V .
(d) Assume that W, V are nite dimensional irreducible representations of A. For any f
Ext
1
(W, V ), let U
f
be the corresponding extension. Show that U
f
is isomorphic to U
f
as repre-
sentations if and only if f and f

are proportional. Thus isomorphism classes (as representations)

of nontrivial extensions of W by V (i.e., those not isomorphic to W V ) are parametrized by the
projective space PExt
1
(W, V ). In particular, every extension is trivial if and only if Ext
1
(W, V ) = 0.
Problem 2.22. (a) Let A = C[x
1
, ..., x
n
], and V
a
, V
b
be one-dimensional representations in which
x
i
act by a
i
and b
i
, respectively (a
i
, b
i
C). Find Ext
1
(V
a
, V
b
) and classify 2-dimensional repre-
sentations of A.
(b) Let B be the algebra over C generated by x
1
, ..., x
n
with the dening relations x
i
x
j
= 0 for
all i, j. Show that for n > 1 the algebra B has innitely many non-isomorphic indecomposable
representations.
Problem 2.23. Let Q be a quiver without oriented cycles, and P
Q
the path algebra of Q. Find
irreducible representations of P
Q
and compute Ext
1
between them. Classify 2-dimensional repre-
sentations of P
Q
.
30
Problem 2.24. Let A be an algebra, and V a representation of A. Let : A EndV . A formal
deformation of V is a formal series
=
0
+t
1
+... +t
n

n
+...,
where
i
: A End(V ) are linear maps,
0
= , and (ab) = (a) (b).
If b(t) = 1 +b
1
t +b
2
t
2
+..., where b
i
End(V ), and is a formal deformation of , then b b
1
is also a deformation of , which is said to be isomorphic to .
(a) Show that if Ext
1
(V, V ) = 0, then any deformation of is trivial, i.e., isomorphic to .
(b) Is the converse to (a) true? (consider the algebra of dual numbers A = k[x]/x
2
).
Problem 2.25. The Cliord algebra. Let V be a nite dimensional complex vector space
equipped with a symmetric bilinear form (, ). The Cliord algebra Cl(V ) is the quotient of the
tensor algebra TV by the ideal generated by the elements v v (v, v)1, v V . More explicitly, if
x
i
, 1 i N is a basis of V and (x
i
, x
j
) = a
ij
then Cl(V ) is generated by x
i
with dening relations
x
i
x
j
+x
j
x
i
= 2a
ij
, x
2
i
= a
ii
.
Thus, if (, ) = 0, Cl(V ) = V .
(i) Show that if (, ) is nondegenerate then Cl(V ) is semisimple, and has one irreducible repre-
sentation of dimension 2
n
if dimV = 2n (so in this case Cl(V ) is a matrix algebra), and two such
representations if dim(V ) = 2n+1 (i.e., in this case Cl(V ) is a direct sum of two matrix algebras).
Hint. In the even case, pick a basis a
1
, ..., a
n
, b
1
, ..., b
n
of V in which (a
i
, a
j
) = (b
i
, b
j
) = 0,
(a
i
, b
j
) =
ij
/2, and construct a representation of Cl(V ) on S := (a
1
, ..., a
n
) in which b
i
acts as
dierentiation with respect to a
i
. Show that S is irreducible. In the odd case the situation is
similar, except there should be an additional basis vector c such that (c, a
i
) = (c, b
i
) = 0, (c, c) =
1, and the action of c on S may be dened either by (1)
degree
or by (1)
degree+1
, giving two
representations S
+
, S

(why are they non-isomorphic?). Show that there is no other irreducible

representations by nding a spanning set of Cl(V ) with 2
dimV
elements.
(ii) Show that Cl(V ) is semisimple if and only if (, ) is nondegenerate. If (, ) is degenerate, what
is Cl(V )/Rad(Cl(V ))?
2.10 Representations of tensor products
Let A, B be algebras. Then A B is also an algebra, with multiplication (a
1
b
1
)(a
2
b
2
) =
a
1
a
2
b
1
b
2
.
Exercise. Show that Mat
m
(k) Mat
n
(k)

= Mat
mn
(k).
The following theorem describes irreducible nite dimensional representations of AB in terms
of irreducible nite dimensional representations of A and those of B.
Theorem 2.26. (i) Let V be an irreducible nite dimensional representation of A and W an
irreducible nite dimensional representation of B. Then V W is an irreducible representation of
AB.
(ii) Any irreducible nite dimensional representation M of A B has the form (i) for unique
V and W.
Remark 2.27. Part (ii) of the theorem typically fails for innite dimensional representations;
e.g. it fails when A is the Weyl algebra in characteristic zero. Part (i) also may fail. E.g. let
A = B = V = W = C(x). Then (i) fails, as AB is not a eld.
31
Proof. (i) By the density theorem, the maps A EndV and B End W are surjective. Therefore,
the map AB EndV EndW = End(V W) is surjective. Thus, V W is irreducible.
(ii) First we show the existence of V and W. Let A

, B

be the images of A, B in EndM. Then

A

, B

are nite dimensional algebras, and M is a representation of A

, so we may assume
without loss of generality that A and B are nite dimensional.
In this case, we claim that Rad(AB) = Rad(A) B+ARad(B). Indeed, denote the latter
by J. Then J is a nilpotent ideal in A B, as Rad(A) and Rad(B) are nilpotent. On the other
hand, (A B)/J = (A/Rad(A)) (B/Rad(B)), which is a product of two semisimple algebras,
hence semisimple. This implies J Rad(A B). Altogether, by Proposition 2.11, we see that
J = Rad(A B), proving the claim.
Thus, we see that
(AB)/Rad(AB) = A/Rad(A) B/Rad(B).
Now, M is an irreducible representation of (A B)/Rad(A B), so it is clearly of the form
M = V W, where V is an irreducible representation of A/Rad(A) and W is an irreducible
representation of B/Rad(B), and V, W are uniquely determined by M (as all of the algebras
involved are direct sums of matrix algebras).
32
3 Representations of nite groups: basic results
Recall that a representation of a group G over a eld k is a k-vector space V together with a
group homomorphism : G GL(V ). As we have explained above, a representation of a group G
over k is the same thing as a representation of its group algebra k[G].
In this section, we begin a systematic development of representation theory of nite groups.
3.1 Maschkes Theorem
Theorem 3.1. (Maschke) Let G be a nite group and k a eld whose characteristic does not divide
[G[. Then:
(i) The algebra k[G] is semisimple.
(ii) There is an isomorphism of algebras : k[G]
i
EndV
i
dened by g
i
g[
V
i
, where V
i
are the irreducible representations of G. In particular, this is an isomorphism of representations
of G (where G acts on both sides by left multiplication). Hence, the regular representation k[G]
decomposes into irreducibles as
i
dim(V
i
)V
i
, and one has
[G[ =

i
dim(V
i
)
2
.
(the sum of squares formula).
Proof. By Proposition 2.16, (i) implies (ii), and to prove (i), it is sucient to show that if V is
a nite-dimensional representation of G and W V is any subrepresentation, then there exists a
subrepresentation W

V such that V = W W

as representations.
Choose any complement

W of W in V . (Thus V = W

W as vector spaces, but not necessarily
as representations.) Let P be the projection along

W onto W, i.e., the operator on V dened by
P[
W
= Id and P[

W
= 0. Let
P :=
1
[G[

gG
(g)P(g
1
),
where (g) is the action of g on V , and let
W

= ker P.
Now P[
W
= Id and P(V ) W, so P
2
= P, so P is a projection along W

. Thus, V = W W

as
vector spaces.
Moreover, for any h G and any y W

,
P(h)y =
1
[G[

gG
(g)P(g
1
h)y =
1
[G[

G
(h)P(
1
)y = (h)Py = 0,
so (h)y ker P = W

. Thus, W

is invariant under the action of G and is therefore a subrepre-

sentation of V . Thus, V = W W

is the desired decomposition into subrepresentations.

The converse to Theorem 3.1(i) also holds.
Proposition 3.2. If k[G] is semisimple, then the characteristic of k does not divide [G[.
33
Proof. Write k[G] =

r
i=1
End V
i
, where the V
i
are irreducible representations and V
1
= k is the
trivial one-dimensional representation. Then
k[G] = k
r

i=2
End V
i
= k
r

i=2
d
i
V
i
,
where d
i
= dimV
i
. By Schurs Lemma,
Hom
k[G]
(k, k[G]) = k
Hom
k[G]
(k[G], k) = k,
for nonzero homomorphisms of representations : k[G] k and : k k[G] unique up to scaling.
We can take such that (g) = 1 for all g G, and such that (1) =

gG
g. Then
(1) =
_

gG
g
_
=

gG
1 = [G[.
If [G[ = 0, then has no left inverse, as (a) (1) = 0 for any a k. This is a contradiction.
Example 3.3. If G = Z/pZ and k has characteristic p, then every irreducible representation of G
over k is trivial (so k[Z/pZ] indeed is not semisimple). Indeed, an irreducible representation of this
group is a 1-dimensional space, on which the generator acts by a p-th root of unity, and every p-th
root of unity in k equals 1, as x
p
1 = (x 1)
p
over k.
Problem 3.4. Let G be a group of order p
n
. Show that every irreducible representation of G over
a eld k of characteristic p is trivial.
3.2 Characters
If V is a nite-dimensional representation of a nite group G, then its character
V
: G k
is dened by the formula
V
(g) = tr[
V
((g)). Obviously,
V
(g) is simply the restriction of the
character
V
(a) of V as a representation of the algebra A = k[G] to the basis G A, so it carries
exactly the same information. The character is a central or class function:
V
(g) depends only on
the conjugacy class of g; i.e.,
V
(hgh
1
) =
V
(g).
Theorem 3.5. If the characteristic of k does not divide [G[, characters of irreducible representa-
tions of G form a basis in the space F
c
(G, k) of class functions on G.
Proof. By the Maschke theorem, k[G] is semisimple, so by Theorem 2.17, the characters are linearly
independent and are a basis of (A/[A, A])

, where A = k[G]. It suces to note that, as vector

spaces over k,
(A/[A, A])

= Hom
k
(k[G], k) [ gh hg ker g, h G

= f Fun(G, k) [ f(gh) = f(hg) g, h G,

which is precisely F
c
(G, k).
Corollary 3.6. The number of isomorphism classes of irreducible representations of G equals the
number of conjugacy classes of G (if [G[ , = 0 in k).
34
Exercise. Show that if [G[ = 0 in k then the number of isomorphism classes of irreducible
representations of G over k is strictly less than the number of conjugacy classes in G.
Hint. Let P =

gG
g k[G]. Then P
2
= 0. So P has zero trace in every nite dimensional
representation of G over k.
Corollary 3.7. Any representation of G is determined by its character if k has characteristic 0;
namely,
V
=
W
implies V

= W.
3.3 Examples
The following are examples of representations of nite groups over C.
1. Finite abelian groups G = Z
n
1
Z
n
k
. Let G

be the set of irreducible representations

of G. Every element of G forms a conjugacy class, so [G

[ = [G[. Recall that all irreducible

representations over C (and algebraically closed elds in general) of commutative algebras and
groups are one-dimensional. Thus, G

is an abelian group: if
1
,
2
: G C

are irreducible
representations then so are
1
(g)
2
(g) and
1
(g)
1
. G

is called the dual or character group

of G.
For given n 1, dene : Z
n
C

by (m) = e
2im/n
. Then Z

n
=
k
: k = 0, . . . , n 1,
so Z

n

= Z
n
. In general,
(G
1
G
2
G
n
)

= G

1
G

2
G

n
,
so G

= G for any nite abelian group G. This isomorphism is, however, noncanonical:
the particular decomposition of G as Z
n
1
Z
n
k
is not unique as far as which elements
of G correspond to Z
n
1
, etc. is concerned. On the other hand, G

= (G

is a canonical
isomorphism, given by : G (G

, where (g)() = (g).

2. The symmetric group S
3
. In S
n
, conjugacy classes are determined by cycle decomposition
sizes: two permutations are conjugate if and only if they have the same number of cycles
of each length. For S
3
, there are 3 conjugacy classes, so there are 3 dierent irreducible
representations over C. If their dimensions are d
1
, d
2
, d
3
, then d
2
1
+d
2
2
+d
2
3
= 6, so S
3
must have
two 1-dimensional and one 2-dimensional representations. The 1-dimensional representations
are the trivial representation C
+
given by () = 1 and the sign representation C

given by
() = (1)

.
The 2-dimensional representation can be visualized as representing the symmetries of the
equilateral triangle with vertices 1, 2, 3 at the points (cos 120

, sin 120

), (cos 240

, sin 240

),
(1, 0) of the coordinate plane, respectively. Thus, for example,
((12)) =
_
1 0
0 1
_
, ((123)) =
_
cos 120

sin120

sin 120

cos 120

_
.
To show that this representation is irreducible, consider any subrepresentation V . V must be
the span of a subset of the eigenvectors of ((12)), which are the nonzero multiples of (1, 0)
and (0, 1). V must also be the span of a subset of the eigenvectors of ((123)), which are
dierent vectors. Thus, V must be either C
2
or 0.
3. The quaternion group Q
8
= 1, i, j, k, with dening relations
i = jk = kj, j = ki = ik, k = ij = ji, 1 = i
2
= j
2
= k
2
.
35
The 5 conjugacy classes are 1, 1, i, j, k, so there are 5 dierent irreducible
representations, the sum of the squares of whose dimensions is 8, so their dimensions must
be 1, 1, 1, 1, and 2.
The center Z(Q
8
) is 1, and Q
8
/Z(Q
8
)

= Z
2
Z
2
. The four 1-dimensional irreducible
representations of Z
2
Z
2
can be pulled back to Q
8
. That is, if q : Q
8
Q
8
/Z(Q
8
) is the
quotient map, and any representation of Q
8
/Z(Q
8
), then q gives a representation of Q
8
.
The 2-dimensional representation is V = C
2
, given by (1) = Id and
(i) =
_
0 1
1 0
_
, (j) =
_
1 0
0

1
_
, (k) =
_
0

1 0
_
. (3)
These are the Pauli matrices, which arise in quantum mechanics.
Exercise. Show that the 2-dimensional irreducible representation of Q
8
can be realized in
the space of functions f : Q
8
C such that f(gi) =

1f(g) (the action of G is by right

multiplication, g f(x) = f(xg)).
4. The symmetric group S
4
. The order of S
4
is 24, and there are 5 conjugacy classes:
e, (12), (123), (1234), (12)(34). Thus the sum of the squares of the dimensions of 5 irreducible
representations is 24. As with S
3
, there are two of dimension 1: the trivial and sign repre-
sentations, C
+
and C

. The other three must then have dimensions 2, 3, and 3. Because

S
3

= S
4
/Z
2
Z
2
, where Z
2
Z
2
is e, (12)(34), (13)(24), (14)(23), the 2-dimensional repre-
sentation of S
3
can be pulled back to the 2-dimensional representation of S
4
, which we will
call C
2
.
We can consider S
4
as the group of rotations of a cube acting by permuting the interior
diagonals (or, equivalently, on a regular octahedron permuting pairs of opposite faces); this
gives the 3-dimensional representation C
3
+
.
The last 3-dimensional representation is C
3

, the product of C
3
+
with the sign representation.
C
3
+
and C
3

are dierent, for if g is a transposition, det g[

C
3
+
= 1 while det g[
C
3

= (1)
3
= 1.
Note that another realization of C
3

is by action of S
4
by symmetries (not necessarily rotations)
of the regular tetrahedron. Yet another realization of this representation is the space of
functions on the set of 4 elements (on which S
4
acts by permutations) with zero sum of
values.
3.4 Duals and tensor products of representations
If V is a representation of a group G, then V

is also a representation, via

V
(g) = (
V
(g)

)
1
= (
V
(g)
1
)

=
V
(g
1
)

.
The character is
V
(g) =
V
(g
1
).
We have
V
(g) =

i
, where the
i
are the eigenvalues of g in V . These eigenvalues must be
roots of unity because (g)
[G[
= (g
[G[
) = (e) = Id. Thus for complex representations

V
(g) =
V
(g
1
) =

1
i
=

i
=

i
=
V
(g).
In particular, V

= V

as representations (not just as vector spaces) if and only if

V
(g) R for all
g G.
36
If V, W are representations of G, then V W is also a representation, via

V W
(g) =
V
(g)
W
(g).
Therefore,
V W
(g) =
V
(g)
W
(g).
An interesting problem discussed below is to decompose V W (for irreducible V, W) into the
direct sum of irreducible representations.
3.5 Orthogonality of characters
We dene a positive denite Hermitian inner product on F
c
(G, C) (the space of central functions)
by
(f
1
, f
2
) =
1
[G[

gG
f
1
(g)f
2
(g).
The following theorem says that characters of irreducible representations of G form an orthonormal
basis of F
c
(G, C) under this inner product.
Theorem 3.8. For any representations V, W
(
V
,
W
) = dimHom
G
(W, V ),
and
(
V
,
W
) =
_
1, if V

= W,
0, if V W
if V, W are irreducible.
Proof. By the denition
(
V
,
W
) =
1
[G[

gG

V
(g)
W
(g) =
1
[G[

gG

V
(g)
W
(g)
=
1
[G[

gG

V W
(g) = Tr [
V W
(P),
where P =
1
[G[

gG
g Z(C[G]). (Here Z(C[G]) denotes the center of C[G]). If X is an irreducible
representation of G then
P[
X
=
_
Id, if X = C,
0, X ,= C.
Therefore, for any representation X the operator P[
X
is the G-invariant projector onto the subspace
X
G
of G-invariants in X. Thus,
Tr [
V W
(P) = dimHom
G
(C, V W

)
= dim(V W

)
G
= dimHom
G
(W, V ).
37
Theorem 3.8 gives a powerful method of checking if a given complex representation V of a nite
group G is irreducible. Indeed, it implies that V is irreducible if and only if (
V
,
V
) = 1.
Exercise. Let G be a nite group. Let V
i
be the irreducible complex representations of G.
For every i, let

i
=
dimV
i
[G[

gG

V
i
(g) g
1
C[G] .
(i) Prove that
i
acts on V
j
as the identity if j = i, and as the null map if j ,= i.
(ii) Prove that
i
are idempotents, i.e.,
2
i
=
i
for any i, and
i

j
= 0 for any i ,= j.
Hint: In (i), notice that
i
commutes with any element of k [G], and thus acts on V
j
as an
intertwining operator. Corollary 1.17 thus yields that
i
acts on V
j
as a scalar. Compute this
scalar by taking its trace in V
j
.
Here is another orthogonality formula for characters, in which summation is taken over irre-
ducible representations rather than group elements.
Theorem 3.9. Let g, h G, and let Z
g
denote the centralizer of g in G. Then

V
(g)
V
(h) =
_
[Z
g
[ if g is conjugate to h
0, otherwise
where the summation is taken over all irreducible representations of G.
Proof. As noted above,
V
(h) =
V
(h), so the left hand side equals (using Maschkes theorem):

V
(g)
V
(h) = Tr[

V
V V
(g (h

)
1
) =
Tr[

V
EndV
(x gxh
1
) = Tr[
C[G]
(x gxh
1
).
If g and h are not conjugate, this trace is clearly zero, since the matrix of the operator x gxh
1
in the basis of group elements has zero diagonal entries. On the other hand, if g and h are in the
same conjugacy class, the trace is equal to the number of elements x such that x = gxh
1
, i.e., the
order of the centralizer Z
g
of g. We are done.
Remark. Another proof of this result is as follows. Consider the matrix U whose rows are
labeled by irreducible representations of G and columns by conjugacy classes, with entries U
V,g
=

V
(g)/
_
[Z
g
[. Note that the conjugacy class of g is G/Z
g
, thus [G[/[Z
g
[ is the number of elements
conjugate to G. Thus, by Theorem 3.8, the rows of the matrix U are orthonormal. This means
that U is unitary and hence its columns are also orthonormal, which implies the statement.
3.6 Unitary representations. Another proof of Maschkes theorem for complex
representations
Denition 3.10. A unitary nite dimensional representation of a group G is a representation of G
on a complex nite dimensional vector space V over C equipped with a G-invariant positive denite
Hermitian form
4
(, ), i.e., such that
V
(g) are unitary operators: (
V
(g)v,
V
(g)w) = (v, w).
4
We agree that Hermitian forms are linear in the rst argument and antilinear in the second one.
38
Theorem 3.11. If G is nite, then any nite dimensional representation of G has a unitary
structure. If the representation is irreducible, this structure is unique up to scaling by a positive
real number.
Proof. Take any positive denite form B on V and dene another form B as follows:
B(v, w) =

gG
B(
V
(g)v,
V
(g)w)
Then B is a positive denite Hermitian form on V, and
V
(g) are unitary operators. If V is
an irreducible representation and B
1
, B
2
are two positive denite Hermitian forms on V, then
B
1
(v, w) = B
2
(Av, w) for some homomorphism A : V V (since any positive denite Hermitian
form is nondegenerate). By Schurs lemma, A = Id, and clearly > 0.
Theorem 3.11 implies that if V is a nite dimensional representation of a nite group G, then
the complex conjugate representation V (i.e., the same space V with the same addition and the same
action of G, but complex conjugate action of scalars) is isomorphic to the dual representation V

.
Indeed, a homomorphism of representations V V

is obviously the same thing as an invariant

sesquilinear form on V (i.e. a form additive on both arguments which is linear on the rst one and
antilinear on the second one), and an isomorphism is the same thing as a nondegenerate invariant
sesquilinear form. So one can use a unitary structure on V to dene an isomorphism V V

.
Theorem 3.12. A nite dimensional unitary representation V of any group G is completely re-
ducible.
Proof. Let W be a subrepresentation of V . Let W

be the orthogonal complement of W in V

under the Hermitian inner product. Then W

is a subrepresentation of W, and V = W W

.
This implies that V is completely reducible.
Theorems 3.11 and 3.12 imply Maschkes theorem for complex representations (Theorem 3.1).
Thus, we have obtained a new proof of this theorem over the eld of complex numbers.
Remark 3.13. Theorem 3.12 shows that for innite groups G, a nite dimensional representation
may fail to admit a unitary structure (as there exist nite dimensional representations, e.g. for
G = Z, which are indecomposable but not irreducible).
3.7 Orthogonality of matrix elements
Let V be an irreducible representation of a nite group G, and v
1
, v
2
, . . . , v
n
be an orthonormal
basis of V under the invariant Hermitian form. The matrix elements of V are t
V
ij
(x) = (
V
(x)v
i
, v
j
).
Proposition 3.14. (i) Matrix elements of nonisomorphic irreducible representations are orthog-
onal in Fun(G, C) under the form (f, g) =
1
[G[

xG
f(x)g(x).
(ii) (t
V
ij
, t
V
i

) =
ii

jj

1
dimV
Thus, matrix elements of irreducible representations of G form an orthogonal basis of Fun(G, C).
Proof. Let V and W be two irreducible representations of G. Take v
i
to be an orthonormal basis
of V and w
i
to be an orthonormal basis of W under their positive denite invariant Hermitian
forms. Let w

i
W

be the linear function on W dened by taking the inner product with

39
w
i
: w

i
(u) = (u, w
i
). Then for x G we have (xw

i
, w

j
) = (xw
i
, w
j
). Therefore, putting P =
1
[G[

xG
x, we have
(t
V
ij
, t
W
i

j
) = [G[
1

xG
(xv
i
, v
j
)(xw
i
, w
j
) = [G[
1

xG
(xv
i
, v
j
)(xw

i
, w

j
) = (P(v
i
w

i
), v
j
w

j
)
If V ,= W, this is zero, since P projects to the trivial representation, which does not occur in
V W

. If V = W, we need to consider (P(v

i
v

), v
j
v

). We have a G-invariant decomposition

V V

= C L
C = span(

v
k
v

k
)
L = span
a:
P
k
a
kk
=0
(

k,l
a
kl
v
k
v

l
),
and P projects to the rst summand along the second one. The projection of v
i
v

to C CL
is thus

ii

dimV

v
k
v

k
This shows that
(P(v
i
v

i
), v
j
v

j
) =

ii

jj

dimV
which nishes the proof of (i) and (ii). The last statement follows immediately from the sum of
squares formula.
3.8 Character tables, examples
The characters of all the irreducible representations of a nite group can be arranged into a char-
acter table, with conjugacy classes of elements as the columns, and characters as the rows. More
specically, the rst row in a character table lists representatives of conjugacy classes, the second
one the numbers of elements in the conjugacy classes, and the other rows list the values of the
characters on the conjugacy classes. Due to Theorems 3.8 and 3.9 the rows and columns of a
character table are orthonormal with respect to the appropriate inner products.
Note that in any character table, the row corresponding to the trivial representation consists
of ones, and the column corresponding to the neutral element consists of the dimensions of the
representations.
Here is, for example, the character table of S
3
:
S
3
Id (12) (123)
# 1 3 2
C
+
1 1 1
C

1 -1 1
C
2
2 0 -1
It is obtained by explicitly computing traces in the irreducible representations.
For another example consider A
4
, the group of even permutations of 4 items. There are three
one-dimensional representations (as A
4
has a normal subgroup Z
2
Z
2
, and A
4
/Z
2
Z
2
= Z
3
).
Since there are four conjugacy classes in total, there is one more irreducible representation of
dimension 3. Finally, the character table is
40
A
4
Id (123) (132) (12)(34)
# 1 4 4 3
C 1 1 1 1
C

1
2
1
C

2 1
2
1
C
3
3 0 0 1
where = exp(
2i
3
).
The last row can be computed using the orthogonality of rows. Another way to compute the
last row is to note that C
3
is the representation of A
4
by rotations of the regular tetrahedron: in
this case (123), (132) are the rotations by 120
0
and 240
0
around a perpendicular to a face of the
tetrahedron, while (12)(34) is the rotation by 180
0
around an axis perpendicular to two opposite
edges.
Example 3.15. The following three character tables are of Q
8
, S
4
, and A
5
respectively.
Q
8
1 -1 i j k
# 1 1 2 2 2
C
++
1 1 1 1 1
C
+
1 1 1 -1 -1
C
+
1 1 -1 1 -1
C

1 1 -1 -1 1
C
2
2 -2 0 0 0
S
4
Id (12) (12)(34) (123) (1234)
# 1 6 3 8 6
C
+
1 1 1 1 1
C

1 -1 1 1 -1
C
2
2 0 2 -1 0
C
3
+
3 -1 -1 0 1
C
3

3 1 -1 0 -1
A
5
Id (123) (12)(34) (12345) (13245)
# 1 20 15 12 12
C 1 1 1 1 1
C
3
+
3 0 -1
1+

5
2
1

5
2
C
3

3 0 -1
1

5
2
1+

5
2
C
4
4 1 0 -1 -1
C
5
5 -1 1 0 0
Indeed, the computation of the characters of the 1-dimensional representations is straightfor-
ward.
The character of the 2-dimensional representation of Q
8
is obtained from the explicit formula
(3) for this representation, or by using the orthogonality.
For S
4
, the 2-dimensional irreducible representation is obtained from the 2-dimensional irre-
ducible representation of S
3
via the surjective homomorphism S
4
S
3
, which allows to obtain its
character from the character table of S
3
.
The character of the 3-dimensional representation C
3
+
is computed from its geometric realization
by rotations of the cube. Namely, by rotating the cube, S
4
permutes the main diagonals. Thus
(12) is the rotation by 180
0
around an axis that is perpendicular to two opposite edges, (12)(34)
41
is the rotation by 180
0
around an axis that is perpendicular to two opposite faces, (123) is the
rotation around a main diagonal by 120
0
, and (1234) is the rotation by 90
0
around an axis that is
perpendicular to two opposite faces; this allows us to compute the traces easily, using the fact that
the trace of a rotation by the angle in R
3
is 1 + 2 cos . Now the character of C
3

is found by
multiplying the character of C
3
+
by the character of the sign representation.
Finally, we explain how to obtain the character table of A
5
(even permutations of 5 items). The
group A
5
is the group of rotations of the regular icosahedron. Thus it has a 3-dimensional rotation
representation C
3
+
, in which (12)(34) is the rotation by 180
0
around an axis perpendicular to two
opposite edges, (123) is the rotation by 120
0
around an axis perpendicular to two opposite faces,
and (12345), (13254) are the rotations by 72
0
, respectively 144
0
, around axes going through two
opposite vertices. The character of this representation is computed from this description in a
straightforward way.
Another representation of A
5
, which is also 3-dimensional, is C
3
+
twisted by the automorphism
of A
5
given by conjugation by (12) inside S
5
. This representation is denoted by C
3

. It has the
same character as C
3
+
, except that the conjugacy classes (12345) and (13245) are interchanged.
There are two remaining irreducible representations, and by the sum of squares formula their
dimensions are 4 and 5. So we call them C
4
and C
5
.
The representation C
4
is realized on the space of functions on the set 1, 2, 3, 4, 5 with zero
sum of values, where A
5
acts by permutations (check that it is irreducible!). The character of
this representation is equal to the character of the 5-dimensional permutation representation minus
the character of the 1-dimensional trivial representation (constant functions). The former at an
element g equals to the number of items among 1,2,3,4,5 which are xed by g.
The representation C
5
is realized on the space of functions on pairs of opposite vertices of the
icosahedron which has zero sum of values (check that it is irreducible!). The character of this
representation is computed similarly to the character of C
4
, or from the orthogonality formula.
3.9 Computing tensor product multiplicities using character tables
Character tables allow us to compute the tensor product multiplicities N
k
ij
using
V
i
V
j
=

N
k
ij
V
k
, N
k
ij
= (
i

j
,
k
)
Example 3.16. The following tables represent computed tensor product multiplicities of irre-
ducible representations of S
3
, S
4
, and A
5
respectively.
S
3
C
+
C

C
2
C
+
C
+
C

C
2
C

C
+
C
2
C
2
C
+
C

C
2
S
4
C
+
C

C
2
C
3
+
C
3

C
+
C
+
C

C
2
C
3
+
C
3

C
+
C
2
C
3

C
3
+
C
2
C
+
C

C
2
C
3
+
C
3

C
3
+
C
3

C
3
+
C
+
C
2
C
3
+
C
3

C
2
C
3
+
C
3

C
3

C
+
C
2
C
3
+
C
3

42
A
5
C C
3
+
C
3

C
4
C
5
C C C
+
3
C
3

C
4
C
5
C
3
+
CC
5
C
3
+
C
4
C
5
C
3

C
4
C
5
C
3
+
C
3

C
4
C
5
C
3

C C
5
C
3
+
C
3
+
C
4
C
5
C
3
+
C
3

C
4
C
5
C
4
C
3
+
C
3

C C
4
C
5
C
3
+
C
3

2C
5
C
4
C
5
C C
3
+
C
3

2C
4
2C
5
3.10 Problems
Problem 3.17. Let G be the group of symmetries of a regular N-gon (it has 2N elements).
(a) Describe all irreducible complex representations of this group (consider the cases of odd and
even N)
(b) Let V be the 2-dimensional complex representation of G obtained by complexication of the
standard representation on the real plane (the plane of the polygon). Find the decomposition of
V V in a direct sum of irreducible representations.
Problem 3.18. Let G be the group of 3 by 3 matrices over F
p
which are upper triangular and have
ones on the diagonal, under multiplication (its order is p
3
). It is called the Heisenberg group. For
any complex number z such that z
p
= 1 we dene a representation of G on the space V of complex
functions on F
p
, by
(
_
_
1 1 0
0 1 0
0 0 1
_
_
f)(x) = f(x 1),
(
_
_
1 0 0
0 1 1
0 0 1
_
_
f)(x) = z
x
f(x).
(note that z
x
makes sense since z
p
= 1).
(a) Show that such a representation exists and is unique, and compute (g) for all g G.
(b) Denote this representation by R
z
. Show that R
z
is irreducible if and only if z ,= 1.
(c) Classify all 1-dimensional representations of G. Show that R
1
decomposes into a direct sum
of 1-dimensional representations, where each of them occurs exactly once.
(d) Use (a)-(c) and the sum of squares formula to classify all irreducible representations of
G.
Problem 3.19. Let V be a nite dimensional complex vector space, and GL(V ) be the group of
invertible linear transformations of V . Then S
n
V and
m
V (m dim(V )) are representations of
GL(V ) in a natural way. Show that they are irreducible representations.
Hint: Choose a basis e
i
in V . Find a diagonal element H of GL(V ) such that (H) has
distinct eigenvalues. (where is one of the above representations). This shows that if W is a
subrepresentation, then it is spanned by a subset S of a basis of eigenvectors of (H). Use the
invariance of W under the operators (1+E
ij
) (where E
ij
is dened by E
ij
e
k
=
jk
e
i
) for all i ,= j
to show that if the subset S is nonempty, it is necessarily the entire basis.
Problem 3.20. Recall that the adjacency matrix of a graph (without multiple edges) is the matrix
in which the ij-th entry is 1 if the vertices i and j are connected with an edge, and zero otherwise.
Let be a nite graph whose automorphism group is nonabelian. Show that the adjacency matrix
of must have repeated eigenvalues.
43
Problem 3.21. Let I be the set of vertices of a regular icosahedron ([I[ = 12). Let Fun(I) be the
space of complex functions on I. Recall that the group G = A
5
of even permutations of 5 items
acts on the icosahedron, so we have a 12-dimensional representation of G on Fun(I).
(a) Decompose this representation in a direct sum of irreducible representations (i.e., nd the
multiplicities of occurrence of all irreducible representations).
(b) Do the same for the representation of G on the space of functions on the set of faces and
the set of edges of the icosahedron.
Problem 3.22. Let F
q
be a nite eld with q elements, and G be the group of nonconstant inho-
mogeneous linear transformations, x ax + b, over F
q
(i.e., a F

q
, b F
q
). Find all irreducible
complex representations of G, and compute their characters. Compute the tensor products of irre-
ducible representations.
Hint. Let V be the representation of G on the space of functions on F
q
with sum of all values
equal to zero. Show that V is an irreducible representation of G.
Problem 3.23. Let G = SU(2) (unitary 2 by 2 matrices with determinant 1), and V = C
2
the
standard 2-dimensional representation of SU(2). We consider V as a real representation, so it is
4-dimensional.
(a) Show that V is irreducible (as a real representation).
(b) Let H be the subspace of End
R
(V ) consisting of endomorphisms of V as a real representation.
Show that H is 4-dimensional and closed under multiplication. Show that every nonzero element in
H is invertible, i.e., H is an algebra with division.
(c) Find a basis 1, i, j, k of H such that 1 is the unit and i
2
= j
2
= k
2
= 1, ij = ji = k, jk =
kj = i, ki = ik = j. Thus we have that Q
8
is a subgroup of the group H

of invertible elements
of H under multiplication.
The algebra H is called the quaternion algebra.
(d) For q = a+bi+cj+dk, a, b, c, d R, let q = abicjdk, and [[q[[
2
= q q = a
2
+b
2
+c
2
+d
2
.
Show that q
1
q
2
= q
2
q
1
, and [[q
1
q
2
[[ = [[q
1
[[ [[q
2
[[.
(e) Let G be the group of quaternions of norm 1. Show that this group is isomorphic to SU(2).
(Thus geometrically SU(2) is the 3-dimensional sphere).
(f ) Consider the action of G on the space V H spanned by i, j, k, by x qxq
1
, q G,
x V . Since this action preserves the norm on V , we have a homomorphism h : SU(2) SO(3),
where SO(3) is the group of rotations of the three-dimensional Euclidean space. Show that this
homomorphism is surjective and that its kernel is 1, 1.
Problem 3.24. It is known that the classication of nite subgroups of SO(3) is as follows:
1) the cyclic group Z/nZ, n 1, generated by a rotation by 2/n around an axis;
2) the dihedral group D
n
of order 2n, n 2 (the group of rotational symmetries in 3-space of
a plane containing a regular n-gon
5
;
3) the group of rotations of the regular tetrahedron (A
4
).
4) the group of rotations of the cube or regular octahedron (S
4
).
5) the group of rotations of a regular dodecahedron or icosahedron (A
5
).
5
A regular 2-gon is just a line segment.
44
(a) Derive this classication.
Hint. Let G be a nite subgroup of SO(3). Consider the action of G on the unit sphere. A
point of the sphere preserved by some nontrivial element of G is called a pole. Show that every
nontrivial element of G xes a unique pair of opposite poles, and that the subgroup of G xing a
particular pole P is cyclic, of some order m (called the order of P). Thus the orbit of P has n/m
elements, where n = [G[. Now let P
1
, ..., P
k
be the poles representing all the orbits of G on the set
of poles, and m
1
, ..., m
k
be their orders. By counting nontrivial elements of G, show that
2(1
1
n
) =

i
(1
1
m
i
).
Then nd all possible m
i
and n that can satisfy this equation and classify the corresponding groups.
(b) Using this classication, classify nite subgroups of SU(2) (use the homomorphism SU(2)
SO(3)).
Problem 3.25. Find the characters and tensor products of irreducible complex representations of
the Heisenberg group from Problem 3.18.
Problem 3.26. Let G be a nite group, and V a complex representation of G which is faithful,
i.e., the corresponding map G GL(V ) is injective. Show that any irreducible representation of
G occurs inside S
n
V (and hence inside V
n
) for some n.
Hint. Show that there exists a vector u V

whose stabilizer in G is 1. Now dene the map

SV Fun(G, C) sending a polynomial f on V

to the function f
u
on G given by f
u
(g) = f(gu).
Show that this map is surjective and use this to deduce the desired result.
Problem 3.27. This problem is about an application of representation theory to physics (elasticity
theory). We rst describe the physical motivation and then state the mathematical problem.
Imagine a material which occupies a certain region U in the physical space V = R
3
(a space
with a positive denite inner product). Suppose the material is deformed. This means, we have
applied a dieomorphism (=change of coordinates) g : U U

. The question in elasticity theory

is how much stress in the material this deformation will cause.
For every point P U, let A
P
: V V be dened by A
P
= dg(P). A
P
is nondegenerate,
so it has a polar decomposition A
P
= D
P
O
P
, where O
P
is orthogonal and D
P
is symmetric. The
matrix O
P
characterizes the rotation part of A
P
(which clearly produces no stress), and D
P
is
the distortion part, which actually causes stress. If the deformation is small, D
P
is close to 1, so
D
P
= 1+d
P
, where d
P
is a small symmetric matrix, i.e., an element of S
2
V . This matrix is called
the deformation tensor at P.
Now we dene the stress tensor, which characterizes stress. Let v be a small nonzero vector in
V , and a small disk perpendicular to v centered at P of area [[v[[. Let F
v
be the force with which
the part of the material on the v-side of acts on the part on the opposite side. It is easy to deduce
from Newtons laws that F
v
is linear in v, so there exists a linear operator S
P
: V V such that
F
v
= S
P
v. It is called the stress tensor.
An elasticity law is an equation S
P
= f(d
P
), where f is a function. The simplest such law is a
linear law (Hookes law): f : S
2
V End(V ) is a linear function. In general, such a function is
dened by 9 6 = 54 parameters, but we will show there are actually only two essential ones the
compression modulus K and the shearing modulus . For this purpose we will use representation
theory.
45
Recall that the group SO(3) of rotations acts on V , so S
2
V , End(V ) are representations of this
group. The laws of physics must be invariant under this group (Galileo transformations), so f must
be a homomorphism of representations.
(a) Show that End(V ) admits a decomposition RV W, where R is the trivial representation,
V is the standard 3-dimensional representation, and W is a 5-dimensional representation of SO(3).
Show that S
2
V = R W
(b) Show that V and W are irreducible, even after complexication. Deduce using Schurs
lemma that S
P
is always symmetric, and for x R, y W one has f(x + y) = Kx + y for some
real numbers K, .
In fact, it is clear from physics that K, are positive. Physically, the compression modulus K
characterises resistance of the material to compression or dilation, while the shearing modulus
characterizes its resistance to changing the shape of the object without changing its volume. For
instance, clay (used for sculpting) has a large compression modulus but a small shearing modulus.
46
4 Representations of nite groups: further results
4.1 Frobenius-Schur indicator
Suppose that G is a nite group and V is an irreducible representation of G over C. We say that
V is
- of complex type, if V V

,
- of real type, if V has a nondegenerate symmetric form invariant under G,
- of quaternionic type, if V has a nondegenerate skew form invariant under G.
Problem 4.1. (a) Show that End
R[G]
V is C for V of complex type, Mat
2
(R) for V of real type,
and H for V of quaternionic type, which motivates the names above.
Hint. Show that the complexication V
C
of V decomposes as V V

. Use this to compute the

dimension of End
R[G]
V in all three cases. Using the fact that C End
R[G]
V , prove the result
in the complex case. In the remaining two cases, let B be the invariant bilinear form on V , and
(, ) the invariant positive Hermitian form (they are dened up to a nonzero complex scalar and a
positive real scalar, respectively), and dene the operator j : V V such that B(v, w) = (v, jw).
Show that j is complex antilinear (ji = ij), and j
2
= Id, where is a real number, positive in
the real case and negative in the quaternionic case (if B is renormalized, j multiplies by a nonzero
complex number z and j
2
by z z, as j is antilinear). Thus j can be normalized so that j
2
= 1 for
the real case, and j
2
= 1 in the quaternionic case. Deduce the claim from this.
(b) Show that V is of real type if and only if V is the complexication of a representation V
R
over the eld of real numbers.
Example 4.2. For Z/nZ all irreducible representations are of complex type, except the trivial one
and, if n is even, the sign representation, m (1)
m
, which are of real type. For S
3
all three
irreducible representations C
+
, C

, C
2
are of real type. For S
4
there are ve irreducible representa-
tions C
+
, C

, C
2
, C
3
+
, C
3

, which are all of real type. Similarly, all ve irreducible representations

of A
5
C, C
3
+
, C
3

, C
4
, C
5
are of real type. As for Q
8
, its one-dimensional representations are of
real type, and the two-dimensional one is of quaternionic type.
Denition 4.3. The Frobenius-Schur indicator FS(V ) of an irreducible representation V is 0 if it
is of complex type, 1 if it is of real type, and 1 if it is of quaternionic type.
Theorem 4.4. (Frobenius-Schur) The number of involutions (=elements of order 2) in G is
equal to

V
dim(V )FS(V ), i.e., the sum of dimensions of all representations of G of real type
minus the sum of dimensions of its representations of quaternionic type.
Proof. Let A : V V have eigenvalues
1
,
2
, . . . ,
n
. We have
Tr[
S
2
V
(AA) =

ij

j
Tr[

2
V
(AA) =

i<j

j
Thus,
Tr[
S
2
V
(A A) Tr[

2
V
(A A) =

1in

2
i
= Tr(A
2
).
47
Thus for g G we have

V
(g
2
) =
S
2
V
(g)

2
V
(g)
Therefore,
[G[
1

V
(

gG
g
2
) =
S
2
V
(P)

2
V
(P) = dim(S
2
V )
G
dim(
2
V )
G
=
_
_
_
1, if V is of real type
1, if V is of quaternionic type
0, if V is of complex type
Finally, the number of involutions in G equals
1
[G[

V
dimV
V
(

gG
g
2
) =

real V
dimV

quat V
dimV.
Corollary 4.5. Assume that all representations of a nite group G are dened over real numbers
(i.e., all complex representations of G are obtained by complexifying real representations). Then
the sum of dimensions of irreducible representations of G equals the number of involutions in G.
Exercise. Show that any nontrivial nite group of odd order has an irreducible representation
which is not dened over R (i.e., not realizable by real matrices).
4.2 Frobenius determinant
Enumerate the elements of a nite group G as follows: g
1
, g
2
, . . . , g
n
. Introduce n variables indexed
with the elements of G :
x
g
1
, x
g
2
, . . . , x
gn
.
Denition 4.6. Consider the matrix X
G
with entries a
ij
= x
g
i
g
j
. The determinant of X
G
is some
polynomial of degree n of x
g
1
, x
g
2
, . . . , x
gn
that is called the Frobenius determinant.
The following theorem, discovered by Dedekind and proved by Frobenius, became the starting
point for creation of representation theory (see [Cu]).
Theorem 4.7.
det X
G
=
r

j=1
P
j
(x)
deg P
j
for some pairwise non-proportional irreducible polynomials P
j
(x), where r is the number of conju-
gacy classes of G.
We will need the following simple lemma.
Lemma 4.8. Let Y be an n n matrix with entries y
ij
. Then det Y is an irreducible polynomial
of y
ij
.
Proof. Let Y = tId+

n
i=1
x
i
E
i,i+1
, where i+1 is computed modulo n, and E
i,j
are the elementary
matrices. Then det(Y ) = t
n
(1)
n
x
1
...x
n
, which is obviously irreducible. Hence det(Y ) is
irreducible (since factors of a homogeneous polynomial are homogeneous).
Now we are ready to proceed to the proof of Theorem 4.7.
48
Proof. Let V = C[G] be the regular representation of G. Consider the operator-valued polynomial
L(x) =

gG
x
g
(g),
where (g) EndV is induced by g. The action of L(x) on an element h G is
L(x)h =

gG
x
g
(g)h =

gG
x
g
gh =

zG
x
zh
1z
So the matrix of the linear operator L(x) in the basis g
1
, g
2
, . . . , g
n
is X
G
with permuted columns
and hence has the same determinant up to sign.
Further, by Maschkes theorem, we have
det
V
L(x) =
r

i=1
(det
V
i
L(x))
dimV
i
,
where V
i
are the irreducible representations of G. We set P
i
= det
V
i
L(x). Let e
im
be bases of V
i
and E
i,jk
EndV
i
be the matrix units in these bases. Then E
i,jk
is a basis of C[G] and
L(x)[
V
i
=

j,k
y
i,jk
E
i,jk
,
where y
i,jk
are new coordinates on C[G] related to x
g
by a linear transformation. Then
P
i
(x) = det [
V
i
L(x) = det(y
i,jk
)
Hence, P
i
are irreducible (by Lemma 4.8) and not proportional to each other (as they depend on
dierent collections of variables y
i,jk
). The theorem is proved.
4.3 Algebraic numbers and algebraic integers
We are now passing to deeper results in representation theory of nite groups. These results require
the theory of algebraic numbers, which we will now briey review.
Denition 4.9. z C is an algebraic number (respectively, an algebraic integer), if z is a
root of a monic polynomial with rational (respectively, integer) coecients.
Denition 4.10. z C is an algebraic number, (respectively, an algebraic integer), if z is an
eigenvalue of a matrix with rational (respectively, integer) entries.
Proposition 4.11. Denitions (4.9) and (4.10) are equivalent.
Proof. To show (4.10) (4.9), notice that z is a root of the characteristic polynomial of the matrix
(a monic polynomial with rational, respectively integer, coecients).
To show (4.9) (4.10), suppose z is a root of
p(x) = x
n
+a
1
x
n1
+. . . +a
n1
x +a
n
.
Then the characteristic polynomial of the following matrix (called the companion matrix) is
p(x):
49
_
_
_
_
_
_
_
0 0 0 . . . 0 a
n
1 0 0 . . . 0 a
n1
0 1 0 . . . 0 a
n2
.
.
.
0 0 0 . . . 1 a
1
_
_
_
_
_
_
_
.
Since z is a root of the characteristic polynomial of this matrix, it is an eigenvalue of this matrix.
The set of algebraic numbers is denoted by Q, and the set of algebraic integers by A.
Proposition 4.12. (i) A is a ring.
(ii) Q is a eld. Namely, it is an algebraic closure of the eld of rational numbers.
Proof. We will be using denition (4.10). Let be an eigenvalue of
/ Mat
n
(C)
with eigenvector v, let be an eigenvalue of
B Mat
m
(C)
with eigenvector w. Then is an eigenvalue of
/Id
m
Id
n
B,
and is an eigenvalue of
/B.
The corresponding eigenvector is in both cases v w. This shows that both A and Q are rings.
To show that the latter is a eld, it suces to note that if ,= 0 is a root of a polynomial p(x) of
degree d, then
1
is a root of x
d
p(1/x). The last statement is easy, since a number is algebraic
if and only if it denes a nite extension of Q.
Proposition 4.13. A Q = Z.
Proof. We will be using denition (4.9). Let z be a root of
p(x) = x
n
+a
1
x
n1
+. . . +a
n1
x +a
n
,
and suppose
z =
p
q
Q, gcd(p, q) = 1.
Notice that the leading term of p(x) will have q
n
in the denominator, whereas all the other terms
will have a lower power of q there. Thus, if q ,= 1, then p(z) / Z, a contradiction. Thus,
z A Q z Z. The reverse inclusion follows because n Z is a root of x n.
Every algebraic number has a minimal polynomial p(x), which is the monic polynomial
with rational coecients of the smallest degree such that p() = 0. Any other polynomial q(x) with
rational coecients such that q() = 0 is divisible by p(x). Roots of p(x) are called the algebraic
conjugates of ; they are roots of any polynomial q with rational coecients such that q() = 0.
50
Note that any algebraic conjugate of an algebraic integer is obviously also an algebraic inte-
ger. Therefore, by the Vieta theorem, the minimal polynomial of an algebraic integer has integer
coecients.
Below we will need the following lemma:
Lemma 4.14. If
1
, ...,
m
are algebraic numbers, then all algebraic conjugates to
1
+ ... +
m
are of the form

1
+... +

m
, where

i
are some algebraic conjugates of
i
.
Proof. It suces to prove this for two summands. If
i
are eigenvalues of rational matrices A
i
of
smallest size (i.e., their characteristic polynomials are the minimal polynomials of
i
), then
1
+
2
is an eigenvalue of A := A
1
Id + Id A
2
. Therefore, so is any algebraic conjugate to
1
+
2
.
But all eigenvalues of A are of the form

1
+

2
, so we are done.
Problem 4.15. (a) Show that for any nite group G there exists a nite Galois extension K C
of Q such that any nite dimensional complex representation of G has a basis in which the matrices
of the group elements have entries in K.
Hint. Consider the representations of G over the eld Q of algebraic numbers.
(b) Show that if V is an irreducible complex representation of a nite group G of dimension
> 1 then there exists g G such that
V
(g) = 0.
Hint: Assume the contrary. Use orthonormality of characters to show that the arithmetic mean
of the numbers [
V
(g)[
2
for g ,= 1 is < 1. Deduce that their product satises 0 < < 1.
Show that all conjugates of satisfy the same inequalities (consider the Galois conjugates of the
representation V , i.e. representations obtained from V by the action of the Galois group of K over
Q on the matrices of group elements in the basis from part (a)). Then derive a contradiction.
Remark. Here is a modication of this argument, which does not use (a). Let N = [G[. For
any 0 < j < N coprime to N, show that the map g g
j
is a bijection G G. Deduce that

g,=1
[
V
(g
j
)[
2
= . Then show that K := Q(), = e
2i/N
, and does not change under the
automorphism of K given by
j
. Deduce that is an integer, and derive a contradiction.
4.4 Frobenius divisibility
Theorem 4.16. Let G be a nite group, and let V be an irreducible representation of G over C.
Then
dimV divides [G[.
Proof. Let C
1
, C
2
, . . . , C
n
be the conjugacy classes of G. Set

i
=
V
(g
C
i
)
[C
i
[
dimV
,
where g
C
i
is a representative of C
i
.
Proposition 4.17. The numbers
i
are algebraic integers for all i.
Proof. Let C be a conjugacy class in G, and P =

hC
h. Then P is a central element of Z[G], so it
acts on V by some scalar , which is an algebraic integer (indeed, since Z[G] is a nitely generated
Z-module, any element of Z[G] is integral over Z, i.e., satises a monic polynomial equation with
integer coecients). On the other hand, taking the trace of P in V , we get [C[
V
(g) = dimV ,
g C, so =
[C[
V
(g)
dimV
.
51
Now, consider

V
(g
C
i
).
This is an algebraic integer, since:
(i)
i
are algebraic integers by Proposition 4.17,
(ii)
V
(g
C
i
) is a sum of roots of unity (it is the sum of eigenvalues of the matrix of (g
C
i
), and
since g
[G[
C
i
= e in G, the eigenvalues of (g
C
i
) are roots of unity), and
(iii) A is a ring (Proposition 4.12).
On the other hand, from the denition of
i
,

C
i

V
(g
C
i
) =

i
[C
i
[
V
(g
C
i
)
V
(g
C
i
)
dimV
.
Recalling that
V
is a class function, this is equal to

gG

V
(g)
V
(g)
dimV
=
[G[(
V
,
V
)
dimV
.
Since V is an irreducible representation, (
V
,
V
) = 1, so

C
i

V
(g
C
i
) =
[G[
dimV
.
Since
[G[
dimV
Q and

C
i

V
(g
C
i
) A, by Proposition 4.13
[G[
dimV
Z.
4.5 Burnsides Theorem
Denition 4.18. A group G is called solvable if there exists a series of nested normal subgroups
e = G
1
G
2
. . . G
n
= G
where G
i+1
/G
i
is abelian for all 1 i n 1.
Remark 4.19. Such groups are called solvable because they rst arose as Galois groups of poly-
nomial equations which are solvable in radicals.
Theorem 4.20 (Burnside). Any group G of order p
a
q
b
, where p and q are prime and a, b 0, is
solvable.
This famous result in group theory was proved by the British mathematician William Burnside
in the early 20-th century, using representation theory (see [Cu]). Here is this proof, presented in
modern language.
Before proving Burnsides theorem we will prove several other results which are of independent
interest.
Theorem 4.21. Let V be an irreducible representation of a nite group G and let C be a conjugacy
class of G with gcd([C[, dim(V )) = 1. Then for any g C, either
V
(g) = 0 or g acts as a scalar
on V .
52
The proof will be based on the following lemma.
Lemma 4.22. If
1
,
2
. . .
n
are roots of unity such that
1
n
(
1
+
2
+ . . . +
n
) is an algebraic
integer, then either
1
= . . . =
n
or
1
+. . . +
n
= 0.
Proof. Let a =
1
n
(
1
+. . . +
n
). If not all
i
are equal, then [a[ < 1. Moreover, since any algebraic
conjugate of a root of unity is also a root of unity, [a

[ 1 for any algebraic conjugate a

of a. But
the product of all algebraic conjugates of a is an integer. Since it has absolute value < 1, it must
equal zero. Therefore, a = 0.
Proof of theorem 4.21.
Let dimV = n. Let
1
,
2
, . . .
n
be the eigenvalues of
V
(g). They are roots of unity, so

V
(g) is an algebraic integer. Also, by Proposition 4.17,
1
n
[C[
V
(g) is an algebraic integer. Since
gcd(n, [C[) = 1, there exist integers a, b such that a[C[ +bn = 1. This implies that

V
(g)
n
=
1
n
(
1
+. . . +
n
).
is an algebraic integer. Thus, by Lemma 4.22, we get that either
1
= . . . =
n
or
1
+ . . . +
n
=

V
(g) = 0. In the rst case, since
V
(g) is diagonalizable, it must be scalar. In the second case,

V
(g) = 0. The theorem is proved.
Theorem 4.23. Let G be a nite group, and let C be a conjugacy class in G of order p
k
where p
is prime and k > 0. Then G has a proper nontrivial normal subgroup (i.e., G is not simple).
Proof. Choose an element g C. Since g ,= e, by orthogonality of columns of the character table,

V IrrG
dimV
V
(g) = 0. (4)
We can divide IrrG into three parts:
1. the trivial representation,
2. D, the set of irreducible representations whose dimension is divisible by p, and
3. N, the set of non-trivial irreducible representations whose dimension is not divisible by p.
Lemma 4.24. There exists V N such that
V
(g) ,= 0.
Proof. If V D, the number
1
p
dim(V )
V
(g) is an algebraic integer, so
a =

V D
1
p
dim(V )
V
(g)
is an algebraic integer.
Now, by (4), we have
0 =
C
(g) +

V D
dimV
V
(g) +

V N
dimV
V
(g) = 1 +pa +

V N
dimV
V
(g).
This means that the last summand is nonzero.
53
Now pick V N such that
V
(g) ,= 0; it exists by Lemma 4.24. Theorem 4.21 implies that g
(and hence any element of C) acts by a scalar in V . Now let H be the subgroup of G generated
by elements ab
1
, a, b C. It is normal and acts trivially in V , so H ,= G, as V is nontrivial. Also
H ,= 1, since [C[ > 1.
Proof of Burnsides theorem.
Assume Burnsides theorem is false. Then there exists a nonsolvable group G of order p
a
q
b
. Let
G be the smallest such group. Then G is simple, and by Theorem 4.23, it cannot have a conjugacy
class of order p
k
or q
k
, k 1. So the order of any conjugacy class in G is either 1 or is divisible
by pq. Adding the orders of conjugacy classes and equating the sum to p
a
q
b
, we see that there has
to be more than one conjugacy class consisting just of one element. So G has a nontrivial center,
which gives a contradiction.
4.6 Representations of products
Theorem 4.25. Let G, H be nite groups, V
i
be the irreducible representations of G over a
eld k (of any characteristic), and W
j
be the irreducible representations of H over k. Then the
irreducible representations of GH over k are V
i
W
j
.
Proof. This follows from Theorem 2.26.
4.7 Virtual representations
Denition 4.26. A virtual representation of a nite group G is an integer linear combination of
irreducible representations of G, V =

n
i
V
i
, n
i
Z (i.e., n
i
are not assumed to be nonnegative).
The character of V is
V
:=

n
i

V
i
.
The following lemma is often very useful (and will be used several times below).
Lemma 4.27. Let V be a virtual representation with character
V
. If (
V
,
V
) = 1 and
V
(1) > 0
then
V
is a character of an irreducible representation of G.
Proof. Let V
1
, V
2
, . . . , V
m
be the irreducible representations of G, and V =

n
i
V
i
. Then by
orthonormality of characters, (
V
,
V
) =

i
n
2
i
. So

i
n
2
i
= 1, meaning that n
i
= 1 for exactly
one i, and n
j
= 0 for j ,= i. But
V
(1) > 0, so n
i
= +1 and we are done.
4.8 Induced Representations
Given a representation V of a group G and a subgroup H G, there is a natural way to construct
a representation of H. The restricted representation of V to H, Res
G
H
V is the representation given
by the vector space V and the action
Res
G
H
V
=
V
[
H
.
There is also a natural, but more complicated way to construct a representation of a group G
given a representation V of its subgroup H.
Denition 4.28. If G is a group, H G, and V is a representation of H, then the induced
representation Ind
G
H
V is the representation of G with
Ind
G
H
V = f : G V [f(hx) =
V
(h)f(x)x G, h H
54
and the action g(f)(x) = f(xg) g G.
Remark 4.29. In fact, Ind
G
H
V is naturally isomorphic to Hom
H
(k[G], V ).
Let us check that Ind
G
H
V is indeed a representation:
g(f)(hx) = f(hxg) =
V
(h)f(xg) =
V
(h)g(f)(x), and g(g

(f))(x) = g

(f)(xg) = f(xgg

) =
(gg

)(f)(x) for any g, g

, x G and h H.
Remark 4.30. Notice that if we choose a representative x

from every right H-coset of G, then

any f Ind
G
H
V is uniquely determined by f(x

).
Because of this,
dim(Ind
G
H
V ) = dimV
[G[
[H[
.
Problem 4.31. Check that if K H G are groups and V a representation of K then Ind
G
H
Ind
H
K
V
is isomorphic to Ind
G
K
V .
Exercise. Let K G be nite groups, and : K C

be a homomorphism. Let C

be the
corresponding 1-dimensional representation of K. Let
e

=
1
[K[

gK
(g)
1
g C[K]
be the idempotent corresponding to . Show that the G-representation Ind
G
K
C

is naturally iso-
morphic to C[G]e

(with G acting by left multiplication).

4.9 The Mackey formula
Let us now compute the character of Ind
G
H
V . In each right coset HG, choose a representative
x

.
Theorem 4.32. (The Mackey formula) One has
(g) =

H\G:xgx
1

V
(x

gx
1

).
Remark. If the characteristic of the ground eld k is relatively prime to [H[, then this formula
can be written as
(g) =
1
[H[

xG:xgx
1
H

V
(xgx
1
).
Proof. For a right H-coset of G, let us dene
V

= f Ind
G
H
V [f(g) = 0 g , .
Then one has
Ind
G
H
V =

,
and so
(g) =

(g),
55
where

(g) is the trace of the diagonal block of (g) corresponding to V

.
Since g() = g is a right H-coset for any right H-coset ,

(g) = 0 if ,= g.
Now assume that = g. Then x

g = hx

where h = x

gx
1

H. Consider the vector space

homomorphism : V

V with (f) = f(x

). Since f V

is uniquely determined by f(x

),
is an isomorphism. We have
(gf) = g(f)(x

) = f(x

g) = f(hx

) =
V
(h)f(x

) = h(f),
and gf =
1
h(f). This means that

(g) =
V
(h). Therefore
(g) =

H\G,g=

V
(x

gx
1

).
4.10 Frobenius reciprocity
A very important result about induced representations is the Frobenius Reciprocity Theorem which
connects the operations Ind and Res.
Theorem 4.33. (Frobenius Reciprocity)
Let H G be groups, V be a representation of G and W a representation of H. Then
Hom
G
(V, Ind
G
H
W) is naturally isomorphic to Hom
H
(Res
G
H
V, W).
Proof. Let E = Hom
G
(V, Ind
G
H
W) and E

= Hom
H
(Res
G
H
V, W). Dene F : E E

and F

: E

E as follows: F()v = (v)(e) for any E and (F

()v)(x) = (xv) for any E

.
In order to check that F and F

are well dened and inverse to each other, we need to check

the following ve statements.
Let E, E

, v V , and x, g G.
(a) F() is an H-homomorphism, i.e., F()hv = hF()v.
Indeed, F()hv = (hv)(e) = (hv)(e) = (v)(he) = (v)(eh) = h (v)(e) = hF()v.
(b) F

()v Ind
G
H
W, i.e., (F

()v)(hx) = h(F

()v)(x).
Indeed, (F

()v)(hx) = (hxv) = h(xv) = h(F

()v)(x).
(c) F

() is a G-homomorphism, i.e. F

()gv = g(F

()v).
Indeed, (F

()gv)(x) = (xgv) = (F

()v)(xg) = (g(F

()v))(x).
(d) F F

= Id
E
.
This holds since F(F

())v = (F

()v)(e) = (v).
(e) F

F = Id
E
, i.e., (F

(F())v)(x) = (v)(x).
Indeed, (F

(F())v)(x) = F(xv) = (xv)(e) = (xv)(e) = (v)(x), and we are done.

Exercise. The purpose of this exercise is to understand the notions of restricted and induced
representations as part of a more advanced framework. This framework is the notion of tensor
products over k-algebras (which generalizes the tensor product over k which we dened in Denition
56
1.48). In particular, this understanding will lead us to a new proof of the Frobenius reciprocity
and to some analogies between induction and restriction.
Throughout this exercise, we will use the notation and results of the Exercise in Section 1.10.
Let G be a nite group and H G a subgroup. We consider k [G] as a (k [H] , k [G])-bimodule
(both module structures are given by multiplication inside k [G]). We denote this bimodule by
k [G]
1
. On the other hand, we can also consider k [G] as a (k [G] , k [H])-bimodule (again, both
module structures are given by multiplication). We denote this bimodule by k [G]
2
.
(a) Let V be a representation of G. Then, V is a left k [G]-module, thus a (k [G] , k)-bimodule.
Thus, the tensor product k [G]
1

k[G]
V is a (k [H] , k)-bimodule, i. e., a left k [H]-module. Prove
that this tensor product is isomorphic to Res
G
H
V as a left k [H]-module. The isomorphism Res
G
H
V
k [G]
1

k[G]
V is given by v 1
k[G]
v for every v Res
G
H
V .
(b) Let W be a representation of H. Then, W is a left k [H]-module, thus a (k [H] , k)-
bimodule. Then, Ind
G
H
W

= Hom
H
(k [G] , W), according to Remark 4.30. In other words, Ind
G
H
W

=
Hom
k[H]
(k [G]
1
, W). Now, use part (b) of the Exercise in Section 1.10 to conclude Theorem 4.33.
(c) Let V be a representation of G. Then, V is a left k [G]-module, thus a (k [G] , k)-bimodule.
Prove that not only k [G]
1

k[G]
V , but also Hom
k[G]
(k [G]
2
, V ) is isomorphic to Res
G
H
V as a left
k [H]-module. The isomorphism Hom
k[G]
(k [G]
2
, V ) Res
G
H
V is given by f f (1) for every
f Hom
k[G]
(k [G]
2
, V ).
(d) Let W be a representation of H. Then, W is a left k [H]-module, thus a (k [H] , k)-
bimodule. Show that Ind
G
H
W is not only isomorphic to Hom
k[H]
(k [G]
1
, W), but also isomorphic to
k [G]
2

k[H]
W. The isomorphism Hom
k[H]
(k [G]
1
, W) k [G]
2

k[H]
W is given by f

gP
g
1

k[H]
f (g) for every f Hom
k[H]
(k [G]
1
, W), where P is a set of distinct representatives for the right
H-cosets in G. (This isomorphism is independent of the choice of representatives.)
(e) Let V be a representation of G and W a representation of H. Use (b) to prove that
Hom
G
_
Ind
G
H
W, V
_
is naturally isomorphic to Hom
H
_
W, Res
G
H
V
_
.
(f) Let V be a representation of H. Prove that Ind
G
H
(V

)

=
_
Ind
G
H
V
_

as representations of
G. [Hint: Write Ind
G
H
V as k [G]
2

k[H]
V and write Ind
G
H
(V

) as Hom
k[H]
(k [G]
1
, V

). Prove
that the map Hom
k[H]
(k [G]
1
, V

)
_
Ind
G
H
(V

)
_
k given by
_
f,
_
x
k[H]
v
__
(f (Sx)) (v) is
a nondegenerate G-invariant bilinear form, where S : k [G] k [G] is the linear map dened by
Sg = g
1
for every g G.]
4.11 Examples
Here are some examples of induced representations (we use the notation for representations from
the character tables).
1. Let G = S
3
, H = Z
2
. Using the Frobenius reciprocity, we obtain: Ind
G
H
C
+
= C
2
C
+
,
Ind
G
H
C

= C
2
C

.
2. Let G = S
3
, H = Z
3
. Then we obtain Ind
G
H
C
+
= C
+
C

, Ind
G
H
C

= Ind
G
H
C

2 = C
2
.
3. Let G = S
4
, H = S
3
. Then Ind
G
H
C
+
= C
+
C
3

, Ind
G
H
C

= C

C
3
+
, Ind
G
H
C
2
= C
2
C
3

C
3
+
.
Problem 4.34. Compute the decomposition into irreducibles of all the representations of A
5
in-
duced from all the irreducible representations of
57
(a) Z
2
(b) Z
3
(c) Z
5
(d) A
4
(e) Z
2
Z
2
4.12 Representations of S
n
In this subsection we give a description of the representations of the symmetric group S
n
for any
n.
Denition 4.35. A partition of n is a representation of n in the form n =
1
+
2
+ ... +
p
,
where
i
are positive integers, and
i

i+1
.
To such we will attach a Young diagramY

, which is the union of rectangles i y i+1,

0 x
i
in the coordinate plane, for i = 1, ..., p. Clearly, Y

is a collection of n unit squares. A

Young tableau corresponding to Y

is the result of lling the numbers 1, ..., n into the squares of

Y

in some way (without repetitions). For example, we will consider the Young tableau T

obtained
by lling in the numbers in the increasing order, left to right, top to bottom.
We can dene two subgroups of S
n
corresponding to T

:
1. The row subgroup P

: the subgroup which maps every element of 1, ..., n into an element

standing in the same row in T

.
2. The column subgroup Q

: the subgroup which maps every element of 1, ..., n into an

element standing in the same column in T

.
Clearly, P

= 1.
Dene the Young projectors:
a

:=
1
[P

gP

g,
b

:=
1
[Q

gQ

(1)
g
g,
where (1)
g
denotes the sign of the permutation g. Set c

= a

. Since P

= 1, this
element is nonzero.
The irreducible representations of S
n
are described by the following theorem.
Theorem 4.36. The subspace V

:= C[S
n
]c

of C[S
n
] is an irreducible representation of S
n
under
left multiplication. Every irreducible representation of S
n
is isomorphic to V

for a unique .
The modules V

are called the Specht modules.

The proof of this theorem is given in the next subsection.
Example 4.37.
For the partition = (n), P

= S
n
, Q

= 1, so c

is the symmetrizer, and hence V

is the trivial
representation.
58
For the partition = (1, ..., 1), Q

= S
n
, P

= 1, so c

is the antisymmetrizer, and hence V

is
the sign representation.
n = 3. For = (2, 1), V

= C
2
.
n = 4. For = (2, 2), V

= C
2
; for = (3, 1), V

= C
3

; for = (2, 1, 1), V

= C
3
+
.
Corollary 4.38. All irreducible representations of S
n
can be given by matrices with rational entries.
Problem 4.39. Find the sum of dimensions of all irreducible representations of the symmetric
group S
n
.
Hint. Show that all irreducible representations of S
n
are real, i.e., admit a nondegenerate
invariant symmetric form. Then use the Frobenius-Schur theorem.
4.13 Proof of Theorem 4.36
Lemma 4.40. Let x C[S
n
]. Then a

xb

(x)c

, where

is a linear function.
Proof. If g P

, then g has a unique representation as pq, p P

, q Q

, so a

gb

= (1)
q
c

.
Thus, to prove the required statement, we need to show that if g is a permutation which is not in
P

then a

gb

= 0.
To show this, it is sucient to nd a transposition t such that t P

and g
1
tg Q

; then
a

gb

= a

tgb

= a

g(g
1
tg)b

= a

gb

,
so a

gb

= 0. In other words, we have to nd two elements i, j standing in the same row in the
tableau T = T

, and in the same column in the tableau T

= gT (where gT is the tableau of the

same shape as T obtained by permuting the entries of T by the permutation g). Thus, it suces to
show that if such a pair does not exist, then g P

, i.e., there exists p P

, q

:= gQ

g
1
such that pT = q

(so that g = pq
1
, q = g
1
q

g Q

).
Any two elements in the rst row of T must be in dierent columns of T

, so there exists q

1
Q

which moves all these elements to the rst row. So there is p

1
P

such that p
1
T and q

1
T

have
the same rst row. Now do the same procedure with the second row, nding elements p
2
, q

2
such
that p
2
p
1
T and q

2
q

1
T

have the same rst two rows. Continuing so, we will construct the desired
elements p, q

. The lemma is proved.

Let us introduce the lexicographic ordering on partitions: > if the rst nonvanishing

i
is positive.
Lemma 4.41. If > then a

C[S
n
]b

= 0.
Proof. Similarly to the previous lemma, it suces to show that for any g S
n
there exists a
transposition t P

such that g
1
tg Q

. Let T = T

and T

= gT

. We claim that there are

two integers which are in the same row of T and the same column of T

. Indeed, if
1
>
1
, this is
clear by the pigeonhole principle (already for the rst row). Otherwise, if
1
=
1
, like in the proof
of the previous lemma, we can nd elements p
1
P

, q

1
gQ

g
1
such that p
1
T and q

1
T

have the
same rst row, and repeat the argument for the second row, and so on. Eventually, having done
i 1 such steps, well have
i
>
i
, which means that some two elements of the i-th row of the rst
tableau are in the same column of the second tableau, completing the proof.
59
Lemma 4.42. c

is proportional to an idempotent. Namely, c

2

=
n!
dimV

.
Proof. Lemma 4.40 implies that c
2

is proportional to c

. Also, it is easy to see that the trace of

c

in the regular representation is n! (as the coecient of the identity element in c

is 1). This
implies the statement.
Lemma 4.43. Let A be an algebra and e be an idempotent in A. Then for any left A-module M,
one has Hom
A
(Ae, M)

= eM (namely, x eM corresponds to f
x
: Ae M given by f
x
(a) = ax,
a Ae).
Proof. Note that 1 e is also an idempotent in A. Thus the statement immediately follows from
the fact that Hom
A
(A, M)

= M and the decomposition A = Ae A(1 e).

Now we are ready to prove Theorem 4.36. Let . Then by Lemmas 4.42, 4.43
Hom
Sn
(V

, V

) = Hom
Sn
(C[S
n
]c

, C[S
n
]c

) = c

C[S
n
]c

.
The latter space is zero for > by Lemma 4.41, and 1-dimensional if = by Lemmas 4.40
and 4.42. Therefore, V

are irreducible, and V

is not isomorphic to V

if ,= . Since the number

of partitions equals the number of conjugacy classes in S
n
, the representations V

exhaust all the

irreducible representations of S
n
. The theorem is proved.
4.14 Induced representations for S
n
Denote by U

the representation Ind

Sn
P

C. It is easy to see that U

can be alternatively dened as

U

= C[S
n
]a

.
Proposition 4.44. Hom(U

, V

) = 0 for < , and dimHom(U

, V

) = 1. Thus, U

, where K

are nonnegative integers and K

= 1.
Denition 4.45. The integers K

are called the Kostka numbers.

Proof. By Lemmas 4.42 and 4.43,
Hom(U

, V

) = Hom(C[S
n
]a

, C[S
n
]a

) = a

C[S
n
]a

,
and the result follows from Lemmas 4.40 and 4.41.
Now let us compute the character of U

. Let C
i
be the conjugacy class in S
n
having i
l
cycles
of length l for all l 1 (here i is a shorthand notation for (i
1
, ..., i
l
, ...)). Also let x
1
, ..., x
N
be
variables, and let
H
m
(x) =

i
x
m
i
be the power sum polynomials.
Theorem 4.46. Let N p (where p is the number of parts of ). Then
U

(C
i
) is the coecient
6
of x

:=

j
j
in the polynomial

m1
H
m
(x)
im
.
6
If j > p, we dene j to be zero.
60
Proof. The proof is obtained easily from the Mackey formula. Namely,
U

(C
i
) is the number of
elements x S
n
such that xgx
1
P

(for a representative g C
i
), divided by [P

[. The order of
P

is

i
!, and the number of elements x such that xgx
1
P

is the number of elements in P

conjugate to g (i.e. [C
i
P

[) times the order of the centralizer Z

g
of g (which is n!/[C
i
[). Thus,

(C
i
) =
[Z
g
[

j

j
!
[C
i
P

[.
Now, it is easy to see that the centralizer Z
g
of g is isomorphic to

m
S
im
(Z/mZ)
im
, so
[Z
g
[ =

m
m
im
i
m
!,
and we get

(C
i
) =

m
m
im
i
m
!

j

j
!
[C
i
P

[.
Now, since P

j
S

j
, we have
[C
i
P

[ =

j1

j
!

m1
m
r
jm
r
jm
!
,
where r = (r
jm
) runs over all collections of nonnegative integers such that

m
mr
jm
=
j
,

j
r
jm
= i
m
.
Indeed, an element of C
i
that is in P

would dene an ordered partition of each

j
into parts
(namely, cycle lengths), with m occuring r
jm
times, such that the total (over all j) number of times
each part m occurs is i
m
. Thus we get

(C
i
) =

m
i
m
!

j
r
jm
!
But this is exactly the coecient of x

in

m1
(x
m
1
+... +x
m
N
)
im
(r
jm
is the number of times we take x
m
j
).
4.15 The Frobenius character formula
Let (x) =

1i<jN
(x
i
x
j
). Let = (N 1, N 2, ..., 0) C
N
. The following theorem, due
to Frobenius, gives a character formula for the Specht modules V

.
Theorem 4.47. Let N p. Then
V

(C
i
) is the coecient of x
+
:=

j
+Nj
j
in the polyno-
mial
(x)

m1
H
m
(x)
im
.
Remark. Here is an equivalent formulation of Theorem 4.47:
V

(C
i
) is the coecient of x

in the (Laurent) polynomial

i<j
_
1
x
j
x
i
_

m1
H
m
(x)
im
.
61
Proof. Denote
V

shortly by

. Let us denote the class function dened in the theorem by

. We
claim that this function has the property

, where L

are integers and L

= 1.
Indeed, from Theorem 4.46 we have

S
N
(1)

U
+()
,
where if the vector + () has a negative entry, the corresponding term is dropped, and if
it has nonnegative entries which fail to be nonincreasing, then the entries should be reordered in
the nonincreasing order, making a partition that well denote + ()) (i.e., we agree that
U
+()
:= U
+())
). Now note that = + ()) is obtained from by adding vectors
of the form e
i
e
j
, i < j, which implies that > or = , and the case = arises only if
= 1, as desired.
Therefore, to show that

, by Lemma 4.27, it suces to show that (

) = 1.
We have
(

) =
1
n!

i
[C
i
[

(C
i
)
2
.
Using that
[C
i
[ =
n!

m
m
im
i
m
!
,
we conclude that (

) is the coecient of x
+
y
+
in the series R(x, y) = (x)(y)S(x, y),
where
S(x, y) =

m
(

j
x
m
j
)
im
(

k
y
m
k
)
im
m
im
i
m
!
=

m
(

j,k
x
m
j
y
m
k
/m)
im
i
m
!
.
Summing over i and m, we get
S(x, y) =

m
exp(

j,k
x
m
j
y
m
k
/m) = exp(

j,k
log(1 x
j
y
k
)) =

j,k
(1 x
j
y
k
)
1
Thus,
R(x, y) =

i<j
(x
i
x
j
)(y
i
y
j
)

i,j
(1 x
i
y
j
)
.
Now we need the following lemma.
Lemma 4.48.
i<j
(z
j
z
i
)(y
i
y
j
)

i,j
(z
i
y
j
)
= det(
1
z
i
y
j
).
Proof. Multiply both sides by

i,j
(z
i
y
j
). Then the right hand side must vanish on the hyperplanes
z
i
= z
j
and y
i
= y
j
(i.e., be divisible by (z)(y)), and is a homogeneous polynomial of degree
N(N 1). This implies that the right hand side and the left hand side are proportional. The
proportionality coecient (which is equal to 1) is found by induction by multiplying both sides by
z
N
y
N
and then setting z
N
= y
N
.
Now setting in the lemma z
i
= 1/x
i
, we get
Corollary 4.49. (Cauchy identity)
R(x, y) = det(
1
1 x
i
y
j
) =

S
N
1

N
j=1
(1 x
j
y
(j)
)
.
62
Corollary 4.49 easily implies that the coecient of x
+
y
+
is 1. Indeed, if ,= 1 is a permu-
tation in S
N
, the coecient of this monomial in
1
Q
(1x
j
y
(j)
)
is obviously zero.
Remark. For partitions and of n, let us say that _ or _ if is a sum of
vectors of the form e
i
e
j
, i < j (called positive roots). This is a partial order, and _ implies
. It follows from Theorem 4.47 and its proof that

U
.
This implies that the Kostka numbers K

vanish unless _ .
4.16 Problems
In the following problems, we do not make a distinction between Young diagrams and partitions.
Problem 4.50. For a Young diagram , let A() be the set of Young diagrams obtained by adding
a square to , and R() be the set of Young diagrams obtained by removing a square from .
(a) Show that Res
Sn
S
n1
V

=
R()
V

.
(b) Show that Ind
Sn
S
n1
V

=
A()
V

.
Problem 4.51. The content c() of a Young diagram is the sum

j
i=1
(i j). Let C =

i<j
(ij) C[S
n
] be the sum of all transpositions. Show that C acts on the Specht module V

by
multiplication by c().
Problem 4.52. (a) Let V be any nite dimensional representation of S
n
. Show that the element
E := (12) +... + (1n) is diagonalizable and has integer eigenvalues on V , which are between 1 n
and n 1.
Represent E as C
n
C
n1
, where C
n
= C is the element from Problem 4.51.
(b) Show that the element (12)+... +(1n) acts on V

by a scalar if and only if is a rectangular

Young diagram, and compute this scalar.
4.17 The hook length formula
Let us use the Frobenius character formula to compute the dimension of V

. According to the
character formula, dimV

is the coecient of x
+
in (x)(x
1
+ ... + x
N
)
n
. Let l
j
=
j
+ N j.
Then, using the determinant formula for (x) and expanding the determinant as a sum over
permutations, we get
dimV

sS
N
:l
j
Ns(j)
(1)
s
n!

j
(l
j
N +s(j))!
=
n!
l
1
!...l
N
!

sS
N
(1)
s

j
l
j
(l
j
1)...(l
j
N+s(j)+1) =
n!

j
l
j
!
det(l
j
(l
j
1)...(l
j
N +i + 1)).
Using column reduction and the Vandermonde determinant formula, we see from this expression
that
dimV

=
n!

j
l
j
!
det(l
Ni
j
) =
n!

j
l
j
!

1i<jN
(l
i
l
j
) (5)
63
(where N p).
In this formula, there are many cancelations. After making some of these cancelations, we
obtain the hook length formula. Namely, for a square (i, j) in a Young diagram (i, j 1, i
j
),
dene the hook of (i, j) to be the set of all squares (i

, j

) in with i

i, j

= j or i

= i, j

j.
Let h(i, j) be the length of the hook of i, j, i.e., the number of squares in it.
Theorem 4.53. (The hook length formula) One has
dimV

=
n!

i
j
h(i, j)
.
Proof. The formula follows from formula (5). Namely, note that
l
1
!

1<jN
(l
1
l
j
)
=

1kl
1
,k,=l
1
l
j
k.
It is easy to see that the factors in this product are exactly the hooklengths h(i, 1). Now delete the
rst row of the diagram and proceed by induction.
4.18 Schur-Weyl duality for gl(V )
We start with a simple result which is called the Double Centralizer Theorem.
Theorem 4.54. Let A, B be two subalgebras of the algebra EndE of endomorphisms of a nite
dimensional vector space E, such that A is semisimple, and B = End
A
E. Then:
(i) A = End
B
E (i.e., the centralizer of the centralizer of A is A);
(ii) B is semisimple;
(iii) as a representation of A B, E decomposes as E =
iI
V
i
W
i
, where V
i
are all the
irreducible representations of A, and W
i
are all the irreducible representations of B. In particular,
we have a natural bijection between irreducible representations of A and B.
Proof. Since A is semisimple, we have a natural decomposition E =
iI
V
i
W
i
, where W
i
:=
Hom
A
(V
i
, E), and A =
i
EndV
i
. Therefore, by Schurs lemma, B = End
A
(E) is naturally identi-
ed with
i
End(W
i
). This implies all the statements of the theorem.
We will now apply Theorem 4.54 to the following situation: E = V
n
, where V is a nite
dimensional vector space over a eld of characteristic zero, and A is the image of C[S
n
] in EndE.
Let us now characterize the algebra B. Let gl(V ) be EndV regarded as a Lie algebra with operation
ab ba.
Theorem 4.55. The algebra B = End
A
E is the image of the universal enveloping algebra U(gl(V ))
under its natural action on E. In other words, B is generated by elements of the form

n
(b) := b 1 ... 1 + 1 b ... 1 +... + 1 1 ... b,
b gl(V ).
Proof. Clearly, the image of U(gl(V )) is contained in B, so we just need to show that any element
of B is contained in the image of U(gl(V )). By denition, B = S
n
EndV , so the result follows from
part (ii) of the following lemma.
64
Lemma 4.56. Let k be a eld of characteristic zero.
(i) For any nite dimensional vector space U over k, the space S
n
U is spanned by elements of
the form u ... u, u U.
(ii) For any algebra A over k, the algebra S
n
A is generated by elements
n
(a), a A.
Proof. (i) The space S
n
U is an irreducible representation of GL(U) (Problem 3.19). The subspace
spanned by u ... u is a nonzero subrepresentation, so it must be everything.
(ii) By the fundamental theorem on symmetric functions, there exists a polynomial P with
rational coecients such that P(H
1
(x), ..., H
n
(x)) = x
1
...x
n
(where x = (x
1
, ..., x
n
)). Then
P(
n
(a),
n
(a
2
), ...,
n
(a
n
)) = a ... a.
The rest follows from (i).
Now, the algebra A is semisimple by Maschkes theorem, so the double centralizer theorem
applies, and we get the following result, which goes under the name Schur-Weyl duality.
Theorem 4.57. (i) The image A of C[S
n
] and the image B of U(gl(V )) in End(V
n
) are central-
izers of each other.
(ii) Both A and B are semisimple. In particular, V
n
is a semisimple gl(V )-module.
(iii) We have a decomposition of A B-modules V
n
=

, where the summation

is taken over partitions of n, V

are Specht modules for S

n
, and L

are some distinct irreducible

representations of gl(V ) or zero.
4.19 Schur-Weyl duality for GL(V )
The Schur-Weyl duality for the Lie algebra gl(V ) implies a similar statement for the group GL(V ).
Proposition 4.58. The image of GL(V ) in End(V
n
) spans B.
Proof. Denote the span of g
n
, g GL(V ), by B

. Let b EndV be any element.

We claim that B

contains b
n
. Indeed, for all values of t but nitely many, t Id+b is invertible,
so (t Id + b)
n
belongs to B

. This implies that this is true for all t, in particular for t = 0, since
(t Id +b)
n
is a polynomial in t.
The rest follows from Lemma 4.56.
Corollary 4.59. As a representation of S
n
GL(V ), V
n
decomposes as

, where
L

= Hom
Sn
(V

, V
n
) are distinct irreducible representations of GL(V ) or zero.
Example 4.60. If = (n) then L

= S
n
V , and if = (1
n
) (n copies of 1) then L

=
n
V . It was
shown in Problem 3.19 that these representations are indeed irreducible (except that
n
V is zero
if n > dimV ).
65
4.20 Schur polynomials
Let = (
1
, ...,
p
) be a partition of n, and N p. Let
D

(x) =

sS
N
(1)
s
N

j=1
x

j
+Nj
s(j)
= det(x

j
+Nj
i
).
Dene the polynomials
S

(x) :=
D

(x)
D
0
(x)
(clearly D
0
(x) is just (x)). It is easy to see that these are indeed polynomials, as D

is an-
tisymmetric and therefore must be divisible by . The polynomials S

are called the Schur

polynomials.
Proposition 4.61.

m
(x
m
1
+... +x
m
N
)
im
=

:pN

(C
i
)S

(x).
Proof. The identity follows from the Frobenius character formula and the antisymmetry of
(x)

m
(x
m
1
+... +x
m
N
)
im
.
Certain special values of Schur polynomials are of importance. Namely, we have
Proposition 4.62.
S

(1, z, z
2
, ..., z
N1
) =

1i<jN
z

i
i
z

j
j
z
i
z
j
Therefore,
S

(1, ..., 1) =

1i<jN

j
+j i
j i
Proof. The rst identity is obtained from the denition using the Vandermonde determinant. The
second identity follows from the rst one by setting z = 1.
4.21 The characters of L

Proposition 4.61 allows us to calculate the characters of the representations L

.
Namely, let dimV = N, g GL(V ), and x
1
, ..., x
N
be the eigenvalues of g on V . To compute
the character
L

(g), let us calculate Tr

V
n(g
n
s), where s S
n
. If s C
i
, we easily get that this
trace equals

m
Tr(g
m
)
im
=

m
H
m
(x)
im
.
On the other hand, by the Schur-Weyl duality
Tr
V
n(g
n
s) =

(C
i
)Tr
L

(g).
Comparing this to Proposition 4.61 and using linear independence of columns of the character table
of S
n
, we obtain
66
Theorem 4.63. (Weyl character formula) The representation L

is zero if and only if N < p,

where p is the number of parts of . If N p, the character of L

is the Schur polynomial S

(x).
Therefore, the dimension of L

is given by the formula

dimL

1i<jN

j
+j i
j i
This shows that irreducible representations of GL(V ) which occur in V
n
for some n are labeled
by Young diagrams with any number of squares but at most N = dimV rows.
Proposition 4.64. The representation L
+1
N (where 1
N
= (1, 1, ..., 1) Z
N
) is isomorphic to
L

N
V .
Proof. Indeed, L

N
V V
n

N
V V
n+N
, and the only component of V
n+N
that has
the same character as L

N
V is L
+1
N. This implies the statement.
4.22 Polynomial representations of GL(V )
Denition 4.65. We say that a nite dimensional representation Y of GL(V ) is polynomial (or
algebraic, or rational) if its matrix elements are polynomial functions of the entries of g, g
1
,
g GL(V ) (i.e., belong to k[g
ij
][1/ det(g)]).
For example, V
n
and hence all L

are polynomial. Also dene L

r1
N := L

(
N
V

)
r
(this
denition makes sense by Proposition 4.64). This is also a polynomial representation. Thus we
have attached a unique irreducible polynomial representation L

of GL(V ) = GL
N
to any sequence
(
1
, ...,
N
) of integers (not necessarily positive) such that
1
...
N
. This sequence is called
the highest weight of L

.
Theorem 4.66. (i) Every nite dimensional polynomial representation of GL(V ) is completely
reducible, and decomposes into summands of the form L

(which are pairwise non-isomorphic).

(ii) (the Peter-Weyl theorem for GL(V )). Let R be the algebra of polynomial functions on
GL(V ). Then as a representation of GL(V ) GL(V ) (with action ((g, h))(x) = (g
1
xh),
g, h, x GL(V ), R), R decomposes as
R =

,
where the summation runs over all .
Proof. (i) Let Y be a polynomial representation of GL(V ). We have an embedding : Y Y R
given by (u, (v))(g) := u(gv), u V

. It is easy to see that is a homomorphism of representations

(where the action of GL(V ) on the rst component of Y R is trivial). Thus, it suces to prove
the theorem for a subrepresentation Y R
m
. Now, every element of R is a polynomial of g
ij
times a nonpositive power of det(g). Thus, R is a quotient of a direct sum of representations of the
form S
r
(V V

) (
N
V

)
s
. So we may assume that Y is contained in a quotient of a (nite)
direct sum of such representations. As V

=
N1
V
N
V

, Y is contained in a direct sum of
representations of the form V
n
(
N
V

)
s
, and we are done.
(ii) Let Y be a polynomial representation of GL(V ), and let us regard R as a representation
of GL(V ) via ((h))(x) = (xh). Then Hom
GL(V )
(Y, R) is the space of polynomial functions
on GL(V ) with values in Y

, which are GL(V )-equivariant. This space is naturally identied

67
with Y

. Taking into account the proof of (i), we deduce that R has the required decomposition,
which is compatible with the second action of GL(V ) (by left multiplications). This implies the
statement.
Note that the Peter-Weyl theorem generalizes Maschkes theorem for nite group, one of whose
forms states that the space of complex functions Fun(G, C) on a nite group G as a representation
of GG decomposes as
V Irrep(G)
V

V .
Remark 4.67. Since the Lie algebra sl(V ) of traceless operators on V is a quotient of gl(V ) by
scalars, the above results extend in a straightforward manner to representations of the Lie algebra
sl(V ). Similarly, the results for GL(V ) extend to the case of the group SL(V ) of operators with
determinant 1. The only dierence is that in this case the representations L

and L
+1
N are
isomorphic, so the irreducible representations are parametrized by integer sequences
1
...
N
up to a simultaneous shift by a constant.
In particular, one can show that any nite dimensional representation of sl(V ) is completely
reducible, and any irreducible one is of the form L

(we will not do this here). For dimV = 2 one

then recovers the representation theory of sl(2) studied in Problem 1.55.
4.23 Problems
Problem 4.68. (a) Show that the S
n
-representation V

:= C[S
n
]b

is isomorphic to V

.
Hint. Dene S
n
-homomorphisms f : V

and g : V

by the formulas f(x) = xa

and
g(y) = yb

, and show that they are inverse to each other up to a nonzero scalar.
(b) Let : C[S
n
] C[S
n
] be the automorphism sending s to (1)
s
s for any permutation s.
Show that maps any representation V of S
n
to V C

. Show also that (C[S

n
]a) = C[S
n
](a),
for a C[S
n
]. Use (a) to deduce that V

= V

, where

is the conjugate partition to ,

obtained by reecting the Young diagram of .
Problem 4.69. Let R
k,N
be the algebra of polynomials on the space of k-tuples of complex N by N
matrices X
1
, ..., X
k
, invariant under simultaneous conjugation. An example of an element of R
k,N
is the function T
w
:= Tr(w(X
1
, ..., X
k
)), where w is any nite word on a k-letter alphabet. Show
that R
k,N
is generated by the elements T
w
.
Hint. Consider invariant functions that are of degree d
i
in each X
i
, and realize this space as
a tensor product
i
S
d
i
(V V

). Then embed this tensor product into (V V

)
N
= End(V )
n
,
and use the Schur-Weyl duality to get the result.
4.24 Representations of GL
2
(F
q
)
4.24.1 Conjugacy classes in GL
2
(F
q
)
Let F
q
be a nite eld of size q of characteristic other than 2, and G = GL
2
(F
q
). Then
[G[ = (q
2
1)(q
2
q),
since the rst column of an invertible 2 by 2 matrix must be non-zero and the second column may
not be a multiple of the rst one. Factoring,
[GL
2
(F
q
)[ = q(q + 1)(q 1)
2
.
68
The goal of this section is to describe the irreducible representations of G.
To begin, let us nd the conjugacy classes in GL
2
(F
q
).
Representatives
Number of elements in a conjugacy
class
Number of classes
Scalar
_
x 0
0 x
_
1 (this is a central element)
q 1 (one for every non-
zero x)
Parabolic
_
x 1
0 x
_
q
2
1 (elements that commute with
this one are of the form
_
t u
0 t
_
, t ,=
0)
q 1 (one for every non-
zero x)
Hyperbolic
_
x 0
0 y
_
, y ,= x
q
2
+q (elements that commute with
this one are of the form
_
t 0
0 u
_
, t, u ,=
0)
1
2
(q 1)(q 2) (x, y ,= 0
and x ,= y)
Elliptic
_
x y
y x
_
, x F
q
, y
F

q
, F
q
F
2
q
(characteris-
tic polynomial over F
q
is irre-
ducible)
q
2
q (the reason will be described
below)
1
2
q(q 1) (matrices with
y and y are conjugate)
More on the conjugacy class of elliptic matrices: these are the matrices whose characteristic
polynomial is irreducible over F
q
and which therefore dont have eigenvalues in F
q
. Let A be such
a matrix, and consider a quadratic extension of F
q
,
F
q
(

), F
q
F
2
q
.
Over this eld, A will have eigenvalues
=
1
+

2
and
=
1

2
,
with corresponding eigenvectors
v, v (Av = v, Av = v).
Choose a basis
e
1
= v +v, e
2
=

(v v).
In this basis, the matrix A will have the form
_

1

2

2

1
_
,
justifying the description of representative elements of this conjugacy class.
In the basis v, v, matrices that commute with A will have the form
_
0
0
_
,
for all
F

q
2
,
so the number of such matrices is q
2
1.
69
4.24.2 1-dimensional representations
First, we describe the 1-dimensional representations of G.
Proposition 4.70. [G, G] = SL
2
(F
q
).
Proof. Clearly,
det(xyx
1
y
1
) = 1,
so
[G, G] SL
2
(F
q
).
To show the converse, it suces to show that the matrices
_
1 1
0 1
_
,
_
a 0
0 a
1
_
,
_
1 0
1 1
_
are commutators (as such matrices generate SL
2
(F
q
).) Clearly, by using transposition, it suces
to show that only the rst two matrices are commutators. But it is easy to see that the matrix
_
1 1
0 1
_
is the commutator of the matrices
A =
_
1 1/2
0 1
_
, B =
_
1 0
0 1
_
,
while the matrix
_
a 0
0 a
1
_
is the commutator of the matrices
A =
_
a 0
0 1
_
, B =
_
0 1
1 0
_
,
This completes the proof.
Therefore,
G/[G, G]

= F

q
via g det(g).
The one-dimensional representations of G thus have the form
(g) =
_
det(g)
_
,
where is a homomorphism
: F

q
C

;
so there are q 1 such representations, denoted C

.
70
4.24.3 Principal series representations
Let
B G, B =
_

0
_

(the set of upper triangular matrices); then

[B[ = (q 1)
2
q,
[B, B] = U =
_
1
0 1
_
,
and
B/[B, B]

= F

q
F

q
(the isomorphism maps an element of B to its two diagonal entries).
Let
: B C

be a homomorphism dened by

_
a b
0 c
_
=
1
(a)
2
(c),for some pair of homomorphisms
1
,
2
: F

q
C

.
Dene
V

1
,
2
= Ind
G
B
C

,
where C

is the 1-dimensional representation of B in which B acts by . We have

dim(V

1
,
2
) =
[G[
[B[
= q + 1.
Theorem 4.71. 1.
1
,=
2
V

1
,
2
is irreducible.
2.
1
=
2
= V

1
,
2
= C

, where W

is a q-dimensional irreducible representation of

G.
3. W

= W

if and only if = ; V

1
,
2

= V

1
,

2
if and only if
1
,
2
=

1
,

2
(in the
second case,
1
,=
2
,

1
,=

2
).
Proof. From the Mackey formula, we have
tr
V

1
,
2
(g) =
1
[B[

aG, aga
1
B
(aga
1
).
If
g =
_
x 0
0 x
_
,
the expression on the right evaluates to
(g)
[G[
[B[
=
1
(x)
2
(x)
_
q + 1
_
.
If
g =
_
x 1
0 x
_
,
71
the expression evaluates to
(g) 1,
since here
aga
1
B a B.
If
g =
_
x 0
0 y
_
,
the expression evaluates to
_

1
(x)
2
(y) +
1
(y)
2
(x)
_
1,
since here
aga
1
B a B or a is an element of B multiplied by the transposition matrix.
If
g =
_
x y
y x
_
, x ,= y
the expression on the right evaluates to 0 because matrices of this type dont have eigenvalues over
F
q
(and thus cannot be conjugated into B). From the denition,
i
(x)(i = 1, 2) is a root of unity,
so
[G[
V

1
,
2
,
V

1
,
2
) = (q + 1)
2
(q 1) + (q
2
1)(q 1)
+ 2(q
2
+q)
(q 1)(q 2)
2
+ (q
2
+q)

x,=y

1
(x)
2
(y)
1
(y)
2
(x).
The last two summands come from the expansion
[a +b[
2
= [a[
2
+[b[
2
+ab +ab.
If

1
=
2
= ,
the last term is equal to
(q
2
+q)(q 2)(q 1),
and the total in this case is
(q + 1)(q 1)[(q + 1) + (q 1) + 2q(q 2)] = (q + 1)(q 1)2q(q 1) = 2[G[,
so

1
,
2
,
V

1
,
2
) = 2.
Clearly,
C

Ind
G
B
C
,
,
since
Hom
G
(C

, Ind
G
B
C
,
) = Hom
B
(C

, C

) = C (Theorem 4.33).
Therefore, Ind
G
B
C
,
= C

; W

is irreducible; and the character of W

is dierent for distinct

values of , proving that W

are distinct.
72
If
1
,=
2
, let z = xy
1
, then the last term of the summation is
(q
2
+q)

x,=y

1
(z)
2
(z) = (q
2
+q)

x;z,=1

2
(z) = (q
2
+q)(q 1)

z,=1

2
(z).
Since

zF

2
(z) = 0,
because the sum of all roots of unity of a given order m > 1 is zero, the last term becomes
(q
2
+q)(q 1)

z,=1

2
(1) = (q
2
+q)(q 1).
The dierence between this case and the case of
1
=
2
is equal to
(q
2
+q)[(q 2)(q 1) + (q 1)] = [G[,
so this is an irreducible representation by Lemma 4.27.
To prove the third assertion of the theorem, we look at the characters on hyperbolic elements
and note that the function

1
(x)
2
(y) +
1
(y)
2
(x)
determines
1
,
2
up to permutation.
4.24.4 Complementary series representations
Let F
q
2 F
q
be a quadratic extension F
q
(

), F
q
F
2
q
. We regard this as a 2-dimensional vector
space over F
q
; then G is the group of linear transformations of F
q
2 over F
q
. Let K G be the cyclic
group of multiplications by elements of F

q
2
,
K =
_
x y
y x
_
, [K[ = q
2
1.
For : K C

a homomorphism, let
Y

= Ind
G
K
C

.
This representation, of course, is very reducible. Let us compute its character, using the Mackey
formula. We get

_
x 0
0 x
_
= q(q 1)(x);
(A) = 0 for A parabolic or hyperbolic;

_
x y
y x
_
=
_
x y
y x
_
+
_
x y
y x
_
q
.
The last assertion holds because if we regard the matrix as an element of F
q
2, conjugation is an
automorphism of F
q
2 over F
q
, but the only nontrivial automorphism of F
q
2 over F
q
is the q
th
power
map.
73
We thus have
Ind
G
K
C

q

= Ind
G
K
C

because they have the same character. Therefore, for

q
,= we get
1
2
q(q 1) representations.
Next, we look at the following tensor product:
W
1
V
,1
,
where 1 is the trivial character and W
1
is dened as in the previous section. The character of this
representation is

_
x 0
0 x
_
= q(q + 1)(x);
(A) = 0 for A parabolic or elliptic;

_
x 0
0 y
_
= (x) +(y).
Thus the virtual representation
W
1
V
,1
V
,1
Ind
G
K
C

,
where is the restriction of to scalars, has character

_
x 0
0 x
_
= (q 1)(x);

_
x 1
0 x
_
= (x);

_
x 0
0 y
_
= 0;

_
x y
y x
_
=
_
x y
y x
_

q
_
x y
y x
_
.
In all that follows, we will have
q
,= .
The following two lemmas will establish that the inner product of this character with itself is
equal to 1, that its value at 1 is positive. As we know from Lemma 4.27, these two properties imply
that it is the character of an irreducible representation of G.
Lemma 4.72. Let be the character of the virtual representation dened above. Then
, ) = 1
and
(1) > 0.
Proof.
(1) = q(q + 1) (q + 1) q(q 1) = q 1 > 0.
We now compute the inner product , ). Since is a root of unity, this will be equal to
1
(q 1)
2
q(q + 1)
_
(q1)(q1)
2
1+(q1)1(q
2
1)+
q(q 1)
2

elliptic
(()+
q
())(() +
q
())

74
Because is also a root of unity, the last term of the expression evaluates to

elliptic
(2 +
q1
() +
1q
()).
Lets evaluate the last summand.
Since F

q
2
is cyclic and
q
,= ,

q
2

q1
() =

q
2

1q
() = 0.
Therefore,

elliptic
(
q1
() +
1q
()) =

q
(
q1
() +
1q
()) = 2(q 1) =
since F

q
is cyclic of order q 1. Therefore,
, ) =
1
(q 1)
2
q(q + 1)
_
(q1)(q1)
2
1+(q1)1(q
2
1)+
q(q 1)
2
(2(q
2
q)2(q1))
_
= 1.
We have now shown that for any with
q
,= the representation Y

with the same character

as
W
1
V
,1
V
,1
Ind
G
K
C

exists and is irreducible. These characters are distinct for distinct pairs (, ) (up to switch

q
), so there are
q(q1)
2
such representations, each of dimension q 1.
We have thus found q 1 1-dimensional representations of G,
q(q1)
2
principal series repre-
sentations, and
q(q1)
2
complementary series representations, for a total of q
2
1 representations,
i.e., the number of conjugacy classes in G. This implies that we have in fact found all irreducible
representations of GL
2
(F
q
).
4.25 Artins theorem
Theorem 4.73. Let X be a conjugation-invariant system of subgroups of a nite group G. Then
two conditions are equivalent:
(i) Any element of G belongs to a subgroup H X.
(ii) The character of any irreducible representation of G belongs to the Q-span of characters of
induced representations Ind
G
H
V , where H X and V is an irreducible representation of H.
Remark. Statement (ii) of Theorem 4.73 is equivalent to the the same statement with Q-span
replaced by C-span. Indeed, consider the matrix whose columns consist of the coecients of the
decomposition of Ind
G
H
V (for various H, V ) with respect to the irreducible representations of G.
Then both statements are equivalent to the condition that the rows of this matrix are linearly
independent.
75
Proof. Proof that (ii) implies (i). Assume that g G does not belong to any of the subgroups
H X. Then, since X is conjugation invariant, it cannot be conjugated into such a subgroup.
Hence by the Mackey formula,
Ind
G
H
(V )
(g) = 0 for all H X and V . So by (ii), for any irreducible
representation W of G,
W
(g) = 0. But irreducible characters span the space of class functions, so
any class function vanishes on g, which is a contradiction.
Proof that (i) implies (ii). Let U be a virtual representation of G over C (i.e., a linear combina-
tion of irreducible representations with nonzero integer coecients) such that (
U
,
Ind
G
H
V
) = 0 for
all H, V . So by Frobenius reciprocity, (
U[
H
,
V
) = 0. This means that
U
vanishes on H for any
H X. Hence by (i),
U
is identically zero. This implies (ii) (because of the above remark).
Corollary 4.74. Any irreducible character of a nite group is a rational linear combination of
induced characters from its cyclic subgroups.
4.26 Representations of semidirect products
Let G, A be groups and : G Aut(A) be a homomorphism. For a A, denote (g)a by g(a).
The semidirect product GA is dened to be the product A G with multiplication law
(a
1
, g
1
)(a
2
, g
2
) = (a
1
g
1
(a
2
), g
1
g
2
).
Clearly, G and A are subgroups of GA in a natural way.
We would like to study irreducible complex representations of GA. For simplicity, let us do
it when A is abelian.
In this case, irreducible representations of A are 1-dimensional and form the character group
A

, which carries an action of G. Let O be an orbit of this action, x O a chosen element,

and G
x
the stabilizer of x in G. Let U be an irreducible representation of G
x
. Then we dene a
representation V
(O,U)
of GA as follows.
As a representation of G, we set
V
(O,x,U)
= Ind
G
Gx
U = f : G U[f(hg) = hf(g), h G
x
.
Next, we introduce an additional action of A on this space by (af)(g) = x(g(a))f(g). Then its
easy to check that these two actions combine into an action of G A. Also, it is clear that this
representation does not really depend on the choice of x, in the following sense. Let x, y O,
and g G be such that gx = y, and let g(U) be the representation of G
y
obtained from the
representation U of G
x
by the action of g. Then V
(O,x,U)
is (naturally) isomorphic to V
(O,y,g(U))
.
Thus we will denote V
(O,x,U)
by V
(O,U)
(remembering, however, that x has been xed).
Theorem 4.75. (i) The representations V
(O,U)
are irreducible.
(ii) They are pairwise nonisomorphic.
(iii) They form a complete set of irreducible representations of GA.
(iv) The character of V = V
(O,U)
is given by the Mackey-type formula

V
(a, g) =
1
[G
x
[

hG:hgh
1
Gx
x(h(a))
U
(hgh
1
).
76
Proof. (i) Let us decompose V = V
(O,U)
as an A-module. Then we get
V =
yO
V
y
,
where V
y
= v V
(O,U)
[av = (y, a)v, a A. (Equivalently, V
y
= v V
(O,U)
[v(g) = 0 unless gy =
x). So if W V is a subrepresentation, then W =
yO
W
y
, where W
y
V
y
. Now, V
y
is a
representation of G
y
, which goes to U under any isomorphism G
y
G
x
determined by g G
mapping x to y. Hence, V
y
is irreducible over G
y
, so W
y
= 0 or W
y
= V
y
for each y. Also, if hy = z
then hW
y
= W
z
, so either W
y
= 0 for all y or W
y
= V
y
for all y, as desired.
(ii) The orbit O is determined by the A-module structure of V , and the representation U by
the structure of V
x
as a G
x
-module.
(iii) We have

U,O
dimV
2
(U,O)
=

U,O
[O[
2
(dimU)
2
=

O
[O[
2
[G
x
[ =

O
[O[[G/G
x
[[G
x
[ = [G[

O
[O[ = [G[[A

[ = [GA[.
(iv) The proof is essentially the same as that of the Mackey formula.
Exercise. Redo Problems 3.17(a), 3.18, 3.22 using Theorem 4.75.
Exercise. Deduce parts (i)-(iii) of Theorem 4.75 from part (iv).
77
5 Quiver Representations
5.1 Problems
Problem 5.1. Field embeddings. Recall that k(y
1
, ..., y
m
) denotes the eld of rational functions
of y
1
, ..., y
m
over a eld k. Let f : k[x
1
, ..., x
n
] k(y
1
, ..., y
m
) be an injective k-algebra homomor-
phism. Show that m n. (Look at the growth of dimensions of the spaces W
N
of polynomials of
degree N in x
i
and their images under f as N ). Deduce that if f : k(x
1
, ..., x
n
) k(y
1
, ..., y
m
)
is a eld embedding, then m n.
Problem 5.2. Some algebraic geometry.
Let k be an algebraically closed eld, and G = GL
n
(k). Let V be a polynomial representation
of G. Show that if G has nitely many orbits on V then dim(V ) n
2
. Namely:
(a) Let x
1
, ..., x
N
be linear coordinates on V . Let us say that a subset X of V is Zariski dense
if any polynomial f(x
1
, ..., x
N
) which vanishes on X is zero (coecientwise). Show that if G has
nitely many orbits on V then G has at least one Zariski dense orbit on V .
(b) Use (a) to construct a eld embedding k(x
1
, ..., x
N
) k(g
pq
), then use Problem 5.1.
(c) generalize the result of this problem to the case when G = GL
n
1
(k) ... GL
nm
(k).
Problem 5.3. Dynkin diagrams.
Let be a graph, i.e., a nite set of points (vertices) connected with a certain number of edges
(we allow multiple edges). We assume that is connected (any vertex can be connected to any
other by a path of edges) and has no self-loops (edges from a vertex to itself ). Suppose the vertices
of are labeled by integers 1, ..., N. Then one can assign to an N N matrix R

= (r
ij
), where
r
ij
is the number of edges connecting vertices i and j. This matrix is obviously symmetric, and is
called the adjacency matrix. Dene the matrix A

= 2I R

, where I is the identity matrix.

Main denition: is said to be a Dynkin diagram if the quadratic from on R
N
with matrix
A

is positive denite.
Dynkin diagrams appear in many areas of mathematics (singularity theory, Lie algebras, rep-
resentation theory, algebraic geometry, mathematical physics, etc.) In this problem you will get a
complete classication of Dynkin diagrams. Namely, you will prove
Theorem. is a Dynkin diagram if and only if it is one on the following graphs:
A
n
:

D
n
:

[

E
6
:

[

78
E
7
:

[

E
8
:

(a) Compute the determinant of A

where = A
N
, D
N
. (Use the row decomposition rule, and
write down a recursive equation for it). Deduce by Sylvester criterion
7
that A
N
, D
N
are Dynkin
diagrams.
8
(b) Compute the determinants of A

for E
6
, E
7
, E
8
(use row decomposition and reduce to (a)).
Show they are Dynkin diagrams.
(c) Show that if is a Dynkin diagram, it cannot have cycles. For this, show that det(A

) = 0
for a graph below
9
1 1 1
1
1
(show that the sum of rows is 0). Thus has to be a tree.
(d) Show that if is a Dynkin diagram, it cannot have vertices with 4 or more incoming edges,
and that can have no more than one vertex with 3 incoming edges. For this, show that det(A

) = 0
for a graph below:
1
1
1
1
2 2
(e) Show that det(A

) = 0 for all graphs below:

1 1 2
3 2
2
1
4 3 2 1
2
1 2 3
7
Recall the Sylvester criterion: a symmetric real matrix is positive denite if and only if all its upper left corner
principal minors are positive.
8
The Sylvester criterion says that a symmetric bilinear form (, ) on R
N
is positive denite if and only if for any
k N, det
1i,jk
(ei, ej) > 0.
9
Please ignore the numerical labels; they will be relevant for Problem 5.5 below.
79
1 2 3 4 5 6 4 2
3
(f ) Deduce from (a)-(e) the classication theorem for Dynkin diagrams.
(g) A (simply laced) ane Dynkin diagram is a connected graph without self-loops such that the
quadratic form dened by A

is positive semidenite. Classify ane Dynkin diagrams. (Show that

they are exactly the forbidden diagrams from (c)-(e)).
Problem 5.4. Let Q be a quiver with set of vertices D. We say that Q is of nite type if it
has nitely many indecomposable representations. Let b
ij
be the number of edges from i to j in Q
(i, j D).
There is the following remarkable theorem, proved by P. Gabriel in early seventies.
Theorem. A connected quiver Q is of nite type if and only if the corresponding unoriented
graph (i.e., with directions of arrows forgotten) is a Dynkin diagram.
In this problem you will prove the only if direction of this theorem (i.e., why other quivers
are NOT of nite type).
(a) Show that if Q is of nite type then for any rational numbers x
i
0 which are not simul-
taneously zero, one has q(x
1
, ..., x
m
) > 0, where
q(x
1
, ..., x
m
) :=

iD
x
2
i

1
2

i,jD
b
ij
x
i
x
j
.
Hint. It suces to check the result for integers: x
i
= n
i
. First assume that n
i
0, and consider
the space W of representations V of Q such that dimV
i
= n
i
. Show that the group

i
GL
n
i
(k) acts
with nitely many orbits on W k, and use Problem 5.2 to derive the inequality. Then deduce the
result in the case when n
i
are arbitrary integers.
(b) Deduce that q is a positive denite quadratic form.
Hint. Use the fact that Q is dense in R.
(c) Show that a quiver of nite type can have no self-loops. Then, using Problem 5.3, deduce
the theorem.
Problem 5.5. Let G ,= 1 be a nite subgroup of SU(2), and V be the 2-dimensional representation
of G coming from its embedding into SU(2). Let V
i
, i I, be all the irreducible representations of
G. Let r
ij
be the multiplicity of V
i
in V V
j
.
(a) Show that r
ij
= r
ji
.
(b) The McKay graph of G, M(G), is the graph whose vertices are labeled by i I, and i is
connected to j by r
ij
edges. Show that M(G) is connected. (Use Problem 3.26)
(c) Show that M(G) is an ane Dynkin graph (one of the forbidden graphs in Problem 5.3).
For this, show that the matrix a
ij
= 2
ij
r
ij
is positive semidenite but not denite, and use
Problem 5.3.
Hint. Let f =

x
i

V
i
, where
V
i
be the characters of V
i
. Show directly that ((2
V
)f, f) 0.
When is it equal to 0? Next, show that M(G) has no self-loops, by using that if G is not cyclic
then G contains the central element Id SU(2).
80
(d) Which groups from Problem 3.24 correspond to which diagrams?
(e) Using the McKay graph, nd the dimensions of irreducible representations of all nite
G SU(2) (namely, show that they are the numbers labeling the vertices of the ane Dynkin
diagrams on our pictures). Compare with the results on subgroups of SO(3) we obtained in
Problem 3.24.
5.2 Indecomposable representations of the quivers A
1
, A
2
, A
3
We have seen that a central question about representations of quivers is whether a certain quiver
has only nitely many indecomposable representations. In the previous subsection it is shown that
only those quivers whose underlying undirected graph is a Dynkin diagram may have this property.
To see if they actually do have this property, we rst explicitly decompose representations of certain
easy quivers.
Remark 5.6. By an object of the type 1
//
0 we mean a map from a one-dimensional vector
space to the zero space. Similarly, an object of the type 0
//
1 is a map from the zero space into
a one-dimensional space. The object 1
//
1 means an isomorphism from a one-dimensional to
another one-dimensional space. The numbers in such diagrams always mean the dimension of the
attached spaces and the maps are the canonical maps (unless specied otherwise)
Example 5.7 (A
1
). The quiver A
1
consists of a single vertex and has no edges. Since a repre-
sentation of this quiver is just a single vector space, the only indecomposable representation is the
ground eld (=a one-dimensional space).
Example 5.8 (A
2
). The quiver A
2
consists of two vertices connected by a single edge.

//

A representation of this quiver consists of two vector spaces V, W and an operator A : V W.

V
A
//

W
To decompose this representation, we rst let V

be a complement to the kernel of A in V and
let W

be a complement to the image of A in W. Then we can decompose the representation as

follows

V
A
//

W
=
ker A
0
//

0

V

//
A

ImA

0
0
//

The rst summand is a multiple of the object 1

//
0 , the second a multiple of 1
//
1 , the
third of 0
//
1 . We see that the quiver A
2
has three indecomposable representations, namely
1
//
0 , 1
//
1 and 0
//
1 .
Example 5.9 (A
3
). The quiver A
3
consists of three vertices and two connections between them.
So we have to choose between two possible orientations.

//

//
or
//

oo
1. We rst look at the orientation

//

//
.
81
Then a representation of this quiver looks like

V
A
//

W
B
//

Y
.
Like in Example 5.8 we rst split away

ker A
0
//

0
0
//

0
.
This object is a multiple of 1
//
0
//
0 . Next, let Y

be a complement of ImB in Y .
Then we can also split away

0
0
//

0
0
//

which is a multiple of the object 0

//
0
//
1 . This results in a situation where the map
A is injective and the map B is surjective (we rename the spaces to simplify notation):


A
//

W
B
// //

Y
.
Next, let X = ker(B A) and let X

be a complement of X in V . Let W

be a complement
of A(X) in W such that A(X

) W

. Then we get


A
//

W
B
// //

Y
=
X
A
//

A(X)
B
//

0

X


A
//

B
// //

Y
The rst of these summands is a multiple of 1
//
1
//
0 . Looking at the second summand,
we now have a situation where A is injective, B is surjective and furthermore ker(BA) = 0.
To simplify notation, we redene
V = X

, W = W

.
Next we let X = Im(B A) and let X

be a complement of X in Y . Furthermore, let

W

= B
1
(X

). Then W

is a complement of A(V ) in W. This yields the decomposition


A
//

W
B
// //

Y
=
V

A
//

A(V )

B
//

X

0
//

B
// //

Here, the rst summand is a multiple of 1

//
1
//
1 . By splitting away the kernel of B,
the second summand can be decomposed into multiples of 0
//
1
//
1 and 0
//
1
//
0 .
So, on the whole, this quiver has six indecomposable representations:
1
//
0
//
0 , 0
//
0
//
1 , 1
//
1
//
0 ,
1
//
1
//
1 , 0
//
1
//
1 , 0
//
1
//
0
2. Now we look at the orientation

//

oo
.
Very similarly to the other orientation, we can split away objects of the type
1
//
0 0
oo
, 0
//
0 1
oo
which results in a situation where both A and B are injective:


A
//

W
oo
B
?
_

Y
.
82
By identifying V and Y as subspaces of W, this leads to the problem of classifying pairs of
subspaces of a given space W up to isomorphism (the pair of subspaces problem). To do
so, we rst choose a complement W

of V Y in W, and set V

= W

V , Y

= W

Y .
Then we can decompose the representation as follows:


//

W
oo ?
_

Y
=
V


//

oo ?
_

V Y
//

V Y

oo
V Y
.
The second summand is a multiple of the object 1
//
1 1
oo
. We go on decomposing the
rst summand. Again, to simplify notation, we let
V = V

, W = W

, Y = Y

.
We can now assume that V Y = 0. Next, let W

be a complement of V Y in W. Then
we get


//

W
oo ?
_

Y
=
V


//

V Y
oo ?
_

Y

0
//

0
oo
The second of these summands is a multiple of the indecomposable object 0
//
1 0
oo
.
The rst summand can be further decomposed as follows:


//

V Y
oo ?
_

Y
=
V
//

V

0
oo

0
//

Y
oo
These summands are multiples of
1
//
1 0
oo
, 0
//
1 1
oo
So - like in the other orientation - we get 6 indecomposable representations of A
3
:
1
//
0 0
oo
, 0
//
0 1
oo
, 1

//
1 1

oo
,
0
//
1 0
oo
, 1
//
1 0
oo
, 0
//
1 1
oo
5.3 Indecomposable representations of the quiver D
4
As a last - slightly more complicated - example we consider the quiver D
4
.
Example 5.10 (D
4
). We restrict ourselves to the orientation

//

oo

OO
.
So a representation of this quiver looks like

V
1
A
1
//

V
3
A
3
oo

V
2
A
2
OO
83
The rst thing we can do is - as usual - split away the kernels of the maps A
1
, A
2
, A
3
. More
precisely, we split away the representations

ker A
1
0
//

0
oo

0
OO

0
//

0
oo

ker A
2
0
OO

0
//

ker A
3
0
oo

0
OO
These representations are multiples of the indecomposable objects

1
0
//

0
oo

0
OO

0
//

0
oo

1
0
OO

0
//

1
0
oo

0
OO
So we get to a situation where all of the maps A
1
, A
2
, A
3
are injective.

V
1

 A
1
//

V
3
?
_
A
3
oo

V
2

?
A
2
OO
As in 2, we can then identify the spaces V
1
, V
2
, V
3
with subspaces of V . So we get to the triple of
subspaces problem of classifying a triple of subspaces of a given space V .
The next step is to split away a multiple of

0
//

0
oo

0
OO
to reach a situation where
V
1
+V
2
+V
3
= V.
By letting Y = V
1
V
2
V
3
, choosing a complement V

of Y in V , and setting V

i
= V

V
i
,
i = 1, 2, 3, we can decompose this representation into


//

3
?
_
oo

?
OO

//

oo

Y
OO
O
The last summand is a multiple of the indecomposable representation

//

oo

1
OO
O
84
So - considering the rst summand and renaming the spaces to simplify notation - we are in a
situation where
V = V
1
+V
2
+V
3
, V
1
V
2
V
3
= 0.
As a next step, we let Y = V
1
V
2
and we choose a complement V

of Y in V such that V
3
V

,
and set V

1
= V

V
1
, V

2
= V

V
2
. This yields the decomposition

V
1


//

V
3
?
_
oo

V
2

?
OO
=


//

V
3
?
_
oo

?
OO

//

0
oo

Y
OO
O
The second summand is a multiple of the indecomposable object

//

0
oo

1
OO
O
.
In the resulting situation we have V
1
V
2
= 0. Similarly we can split away multiples of

//

oo

0
OO
and

0
//

oo

1
OO
O
to reach a situation where the spaces V
1
, V
2
, V
3
do not intersect pairwise
V
1
V
2
= V
1
V
3
= V
2
V
3
= 0.
If V
1
V
2
V
3
we let Y = V
1
(V
2
V
3
). We let V

1
be a complement of Y in V
1
. Since then
V

1
(V
2
V
3
) = 0, we can select a complement V

of V

1
in V which contains V
2
V
3
. This gives
us the decomposition

V
1


//

V
3
?
_
oo

V
2

?
OO
=

//

0
oo

0
OO


//

V
3
?
_
oo

V
2

?
OO
The rst of these summands is a multiple of

//

0
oo

0
OO
By splitting these away we get to a situation where V
1
V
2
V
3
. Similarly, we can split away
objects of the type

0
//

0
oo

1
OO
O
and

0
//

oo

0
OO
to reach a situation in which the following conditions hold
85
1. V
1
+V
2
+V
3
= V.
2. V
1
V
2
= 0, V
1
V
3
= 0, V
2
V
3
= 0.
3. V
1
V
2
V
3
, V
2
V
1
V
3
, V
3
V
1
V
2
.
But this implies that
V
1
V
2
= V
1
V
3
= V
2
V
3
= V.
So we get
dimV
1
= dimV
2
= dimV
3
= n
and
dimV = 2n.
Since V
3
V
1
V
2
we can write every element of V
3
in the form
x V
3
, x = (x
1
, x
2
), x
1
V
1
, x
2
V
2
.
We then can dene the projections
B
1
: V
3
V
1
, (x
1
, x
2
) x
1
,
B
2
: V
3
V
2
, (x
1
, x
2
) x
2
.
Since V
3
V
1
= 0, V
3
V
2
= 0, these maps have to be injective and therefore are isomorphisms. We
then dene the isomorphism
A = B
2
B
1
1
: V
1
V
2
.
Let e
1
, . . . , e
n
be a basis for V
1
. Then we get
V
1
= Ce
1
Ce
2
Ce
n
V
2
= CAe
1
CAe
2
CAe
n
V
3
= C(e
1
+Ae
1
) C(e
2
+Ae
2
) C(e
n
+Ae
n
).
So we can think of V
3
as the graph of an isomorphism A : V
1
V
2
. From this we obtain the
decomposition

V
1


//

V
3
?
_
oo

V
2

?
OO
=
n

j=1

C(1, 0)


//

C
2

C(1, 1)
?
_
oo

C(0, 1)

?
OO
These correspond to the indecomposable object

1
//

1
oo

1
OO
Thus the quiver D
4
with the selected orientation has 12 indecomposable objects. If one were to
explicitly decompose representations for the other possible orientations, one would also nd 12
indecomposable objects.
It appears as if the number of indecomposable representations does not depend on the orienta-
tion of the edges, and indeed - Gabriels theorem will generalize this observation.
86
5.4 Roots
From now on, let be a xed graph of type A
n
, D
n
, E
6
, E
7
, E
8
. We denote the adjacency matrix
of by R

.
Denition 5.11 (Cartan Matrix). We dene the Cartan matrix as
A

= 2Id R

.
On the lattice Z
n
(or the space R
n
) we then dene an inner product
B(x, y) = x
T
A

y
corresponding to the graph .
Lemma 5.12. 1. B is positive denite.
2. B(x, x) takes only even values for x Z
n
.
Proof. 1. This follows by denition, since is a Dynkin diagram.
2. By the denition of the Cartan matrix we get
B(x, x) = x
T
A

x =

i,j
x
i
a
ij
x
j
= 2

i
x
2
i
+

i,j, i,=j
x
i
a
ij
x
j
= 2

i
x
2
i
+ 2

i<j
a
ij
x
i
x
j
which is even.
Denition 5.13. A root with respect to a certain positive inner product is a shortest (with respect
to this inner product), nonzero vector in Z
n
.
So for the inner product B, a root is a nonzero vector x Z
n
such that
B(x, x) = 2.
Remark 5.14. There can be only nitely many roots, since all of them have to lie in some ball.
Denition 5.15. We call vectors of the form

i
= (0, . . . ,
ith
..
1 , . . . , 0)
simple roots.
The
i
naturally form a basis of the lattice Z
n
.
Lemma 5.16. Let be a root, =

n
i=1
k
i

i
. Then either k
i
0 for all i or k
i
0 for all i.
Proof. Assume the contrary, i.e., k
i
> 0, k
j
< 0. Without loss of generality, we can also assume
that k
s
= 0 for all s between i and j. We can identify the indices i, j with vertices of the graph .

i

87
Next, let be the edge connecting i with the next vertex towards j and i

be the vertex on the other

end of . We then let
1
,
2
be the graphs obtained from by removing . Since is supposed
to be a Dynkin diagram - and therefore has no cycles or loops - both
1
and
2
will be connected
graphs, which are not connected to each other.

i

1

j

2
Then we have i
1
, j
2
. We dene
=

m
1
k
m

m
, =

m
2
k
m

m
.
With this choice we get
= +.
Since k
i
> 0, k
j
< 0 we know that ,= 0, ,= 0 and therefore
B(, ) 2, B(, ) 2.
Furthermore,
B(, ) = k
i
k
i
,
since
1
,
2
are only connected at . But this has to be a nonnegative number, since k
i
> 0 and
k
i
0. This yields
B(, ) = B( +, +) = B(, )
. .
2
+2 B(, )
. .
0
+B(, )
. .
2
4.
But this is a contradiction, since was assumed to be a root.
Denition 5.17. We call a root =

i
k
i

i
a positive root if all k
i
0. A root for which k
i
0
for all i is called a negative root.
Remark 5.18. Lemma 5.16 states that every root is either positive or negative.
Example 5.19. 1. Let be of the type A
N1
. Then the lattice L = Z
N1
can be realized as
a subgroup of the lattice Z
N
by letting L Z
N
be the subgroup of all vectors (x
1
, . . . , x
N
)
such that

i
x
i
= 0.
The vectors

1
= (1, 1, 0, . . . , 0)

2
= (0, 1, 1, 0, . . . , 0)
.
.
.

N1
= (0, . . . , 0, 1, 1)
naturally form a basis of L. Furthermore, the standard inner product
(x, y) =

x
i
y
i
88
on Z
N
restricts to the inner product B given by on L, since it takes the same values on the
basis vectors:
(
i
,
i
) = 2
(
i
,
j
) =
_
1 i, j adjacent
0 otherwise
This means that vectors of the form
(0, . . . , 0, 1, 0, . . . , 0, 1, 0, . . . , 0) =
i
+
i+1
+ +
j1
and
(0, . . . , 0, 1, 0, . . . , 0, 1, 0, . . . , 0) = (
i
+
i+1
+ +
j1
)
are the roots of L. Therefore the number of positive roots in L equals
N(N 1)
2
.
2. As a fact we also state the number of positive roots in the other Dynkin diagrams:
D
N
N(N 1)
E
6
36 roots
E
7
63 roots
E
8
120 roots
Denition 5.20. Let Z
n
be a positive root. The reection s

is dened by the formula

s

(v) = v B(v, ).
We denote s

i
by s
i
and call these simple reections.
Remark 5.21. As a linear operator of R
n
, s

xes any vector orthogonal to and

s

() =
Therefore s

is the reection at the hyperplane orthogonal to , and in particular xes B. The

s
i
generate a subgroup W O(R
n
), which is called the Weyl group of . Since for every w W,
w(
i
) is a root, and since there are only nitely many roots, W has to be nite.
5.5 Gabriels theorem
Denition 5.22. Let Q be a quiver with any labeling 1, . . . , n of the vertices. Let V = (V
1
, . . . , V
n
)
be a representation of Q. We then call
d(V ) = (dimV
1
, . . . , dimV
n
)
the dimension vector of this representation.
We are now able to formulate Gabriels theorem using roots.
Theorem 5.23 (Gabriels theorem). Let Q be a quiver of type A
n
, D
n
, E
6
, E
7
, E
8
. Then Q has
nitely many indecomposable representations. Namely, the dimension vector of any indecomposable
representation is a positive root (with respect to B

) and for any positive root there is exactly

one indecomposable representation with dimension vector .
89
5.6 Reection Functors
Denition 5.24. Let Q be any quiver. We call a vertex i Q a sink if all edges connected to i
point towards i.
//

i
oo
OO
We call a vertex i Q a source if all edges connected to i point away from i.

i
oo //

Denition 5.25. Let Q be any quiver and i Q be a sink (a source). Then we let Q
i
be the
quiver obtained from Q by reversing all arrows pointing into (pointing out of) i.
We are now able to dene the reection functors (also called Coxeter functors).
Denition 5.26. Let Q be a quiver, i Q be a sink. Let V be a representation of Q. Then we
dene the reection functor
F
+
i
: RepQ RepQ
i
by the rule
F
+
i
(V )
k
= V
k
if k ,= i
F
+
i
(V )
i
= ker
_
_
:

ji
V
j
V
i
_
_
.
Also, all maps stay the same but those now pointing out of i; these are replaced by compositions
of the inclusion of ker into V
j
with the projections V
j
V
k
.
Denition 5.27. Let Q be a quiver, i Q be a source. Let V be a representation of Q. Let be
the canonical map
: V
i

ij
V
j
.
Then we dene the reection functor
F

i
: RepQ RepQ
i
by the rule
F

i
(V )
k
= V
k
if k ,= i
F

i
(V )
i
= Coker () =
_
_

ij
V
j
_
_
/Im.
Again, all maps stay the same but those now pointing into i; these are replaced by the compositions
of the inclusions V
k

ij
V
j
with the natural map V
j
V
j
/Im.
Proposition 5.28. Let Q be a quiver, V an indecomposable representation of Q.
90
1. Let i Q be a sink. Then either dimV
i
= 1, dimV
j
= 0 for j ,= i or
:

ji
V
j
V
i
is surjective.
2. Let i Q be a source. Then either dimV
i
= 1, dimV
j
= 0 for j ,= i or
: V
i

ij
V
j
is injective.
Proof. 1. Choose a complement W of Im. Then we get
V =

0
//

0
oo

0
OO
V

Since V is indecomposable, one of these summands has to be zero. If the rst summand is
zero, then has to be surjective. If the second summand is zero, then the rst one has to be
of the desired form, because else we could write it as a direct sum of several objects of the
type

0
//

0
oo

0
OO
which is impossible, since V was supposed to be indecomposable.
2. Follows similarly by splitting away the kernel of .
Proposition 5.29. Let Q be a quiver, V be a representation of Q.
1. If
:

ji
V
j
V
i
is surjective, then
F

i
F
+
i
V = V.
2. If
: V
i

ij
V
j
is injective, then
F
+
i
F

i
V = V.
Proof. In the following proof, we will always mean by i j that i points into j in the original
quiver Q. We only establish the rst statement and we also restrict ourselves to showing that the
spaces of V and F

i
F
+
i
V are the same. It is enough to do so for the i-th space. Let
:

ji
V
j
V
i
91
be surjective and let
K = ker .
When applying F
+
i
, the space V
i
gets replaced by K. Furthermore, let
: K

ji
V
j
.
After applying F

i
, K gets replaced by
K

=
_
_

ji
V
j
_
_
/(Im).
But
Im = K
and therefore
K

=
_
_

ji
V
j
_
_
/
_
_
ker( :

ji
V
j
V
i
)
_
_
= Im( :

ji
V
j
V
i
)
by the homomorphism theorem. Since was assumed to be surjective, we get
K

= V
i
.
Proposition 5.30. Let Q be a quiver, and V be an indecomposable representation of Q. Then
F
+
i
V and F

i
V (whenever dened) are either indecomposable or 0.
Proof. We prove the proposition for F
+
i
V - the case F

i
V follows similarly. By Proposition 5.28 it
follows that either
:

ji
V
j
V
i
is surjective or dimV
i
= 1, dimV
j
= 0, j ,= i. In the last case
F
+
i
V = 0.
So we can assume that is surjective. In this case, assume that F
+
i
V is decomposable as
F
+
i
V = X Y
with X, Y ,= 0. But F
+
i
V is injective at i, since the maps are canonical projections, whose direct
sum is the tautological embedding. Therefore X and Y also have to be injective at i and hence (by
5.29)
F
+
i
F

i
X = X, F
+
i
F

i
Y = Y
In particular
F

i
X ,= 0, F

i
Y ,= 0.
Therefore
V = F

i
F
+
i
V = F

i
X F

i
Y
which is a contradiction, since V was assumed to be indecomposable. So we can infer that
F
+
i
V
is indecomposable.
92
Proposition 5.31. Let Q be a quiver and V a representation of Q.
1. Let i Q be a sink and let V be surjective at i. Then
d(F
+
i
V ) = s
i
(d(V )).
2. Let i Q be a source and let V be injective at i. Then
d(F

i
V ) = s
i
(d(V )).
Proof. We only prove the rst statement, the second one follows similarly. Let i Q be a sink and
let
:

ji
V
j
V
i
be surjective. Let K = ker . Then
dimK =

ji
dimV
j
dimV
i
.
Therefore we get
_
d(F
+
i
V ) d(V )
_
i
=

ji
dimV
j
2 dimV
i
= B(d(V ),
i
)
and
_
d(F
+
i
V ) d(V )
_
j
= 0, j ,= i.
This implies
d(F
+
i
V ) d(V ) = B(d(V ),
i
)
i
d(F
+
i
V ) = d(V ) B(d(V ),
i
)
i
= s
i
(d(V )) .
5.7 Coxeter elements
Denition 5.32. Let Q be a quiver and let be the underlying graph. Fix any labeling 1, . . . , n
of the vertices of . Then the Coxeter element c of Q corresponding to this labeling is dened as
c = s
1
s
2
. . . s
n
.
Lemma 5.33. Let
=

i
k
i

i
with k
i
0 for all i but not all k
i
= 0. Then there is N N, such that
c
N

has at least one strictly negative coecient.

93
Proof. c belongs to a nite group W. So there is M N, such that
c
M
= 1.
We claim that
1 +c +c
2
+ +c
M1
= 0
as operators on R
n
. This implies what we need, since has at least one strictly positive coecient,
so one of the elements
c, c
2
, . . . , c
M1

must have at least one strictly negative one. Furthermore, it is enough to show that 1 is not an
eigenvalue for c, since
(1 +c +c
2
+ +c
M1
)v = w ,= 0
cw = c
_
1 +c +c
2
+ +c
M1
_
v = (c +c
2
+c
3
+ +c
M1
+ 1)v = w.
Assume the contrary, i.e., 1 is a eigenvalue of c and let v be a corresponding eigenvector.
cv = v s
1
. . . s
n
v = v
s
2
. . . s
n
v = s
1
v.
But since s
i
only changes the i-th coordinate of v, we get
s
1
v = v and s
2
. . . s
n
v = v.
Repeating the same procedure, we get
s
i
v = v
for all i. But this means
B(v,
i
) = 0.
for all i, and since B is nondegenerate, we get v = 0. But this is a contradiction, since v is an
eigenvector.
5.8 Proof of Gabriels theorem
Let V be an indecomposable representation of Q. We introduce a xed labeling 1, . . . n on Q, such
that i < j if one can reach j from i. This is possible, since we can assign the highest label to any
sink, remove this sink from the quiver, assign the next highest label to a sink of the remaining
quiver and so on. This way we create a labeling of the desired kind.
We now consider the sequence
V
(0)
= V, V
(1)
= F
+
n
V, V
(2)
= F
+
n1
F
+
n
V, . . .
This sequence is well dened because of the selected labeling: n has to be a sink of Q, n 1 has
to be a sink of Q
n
(where Q
n
is obtained from Q by reversing all the arrows at the vertex r) and
so on. Furthermore, we note that V
(n)
is a representation of Q again, since every arrow has been
reversed twice (since we applied a reection functor to every vertex). This implies that we can
dene
V
(n+1)
= F
+
n
V
(n)
, . . .
and continue the sequence to innity.
94
Theorem 5.34. There is m N, such that
d
_
V
(m)
_
=
p
for some p.
Proof. If V
(i)
is surjective at the appropriate vertex k, then
d
_
V
(i+1)
_
= d
_
F
+
k
V
(i)
_
= s
k
d
_
V
(i)
_
.
This implies, that if V
(0)
, . . . , V
(i1)
are surjective at the appropriate vertices, then
d
_
V
(i)
_
= . . . s
n1
s
n
d(V ).
By Lemma 5.33 this cannot continue indenitely - since d
_
V
(i)
_
may not have any negative entries.
Let i be smallest number such that V
(i)
is not surjective at the appropriate vertex. By Proposition
5.30 it is indecomposable. So, by Proposition 5.28, we get
d(V
(i)
) =
p
for some p.
We are now able to prove Gabriels theorem. Namely, we get the following corollaries.
Corollary 5.35. Let Q be a quiver, V be any indecomposable representation. Then d(V ) is a
positive root.
Proof. By Theorem 5.34
s
i
1
. . . s
im
(d(V )) =
p
.
Since the s
i
preserve B, we get
B(d(V ), d(V )) = B(
p
,
p
) = 2.
Corollary 5.36. Let V, V

be indecomposable representations of Q such that d(V ) = d(V

). Then
V and V

are isomorphic.
Proof. Let i be such that
d
_
V
(i)
_
=
p
.
Then we also get d
_
V
(i)
_
=
p
. So
V
(i)
= V
(i)
=: V
i
.
Furthermore we have
V
(i)
= F
+
k
. . . F
+
n1
F
+
n
V
(0)
V
(i)
= F
+
k
. . . F
+
n1
F
+
n
V
(0)
.
But both V
(i1)
, . . . , V
(0)
and V
(i1)
, . . . , V
(0)
have to be surjective at the appropriate vertices.
This implies
F

n
F

n1
. . . F

k
V
i
=
_
F

n
F

n1
. . . F

k
F
+
k
. . . F
+
n1
F
+
n
V
(0)
= V
(0)
= V
F

n
F

n1
. . . F

k
F
+
k
. . . F
+
n1
F
+
n
V
(0)
= V
(0)
= V

95
These two corollaries show that there are only nitely many indecomposable representations
(since there are only nitely many roots) and that the dimension vector of each of them is a positive
root. The last statement of Gabriels theorem follows from
Corollary 5.37. For every positive root , there is an indecomposable representation V with
d(V ) = .
Proof. Consider the sequence
s
n
, s
n1
s
n
, . . .
Consider the rst element of this sequence which is a negative root (this has to happen by Lemma
5.33) and look at one step before that, calling this element . So is a positive root and s
i
is a
negative root for some i. But since the s
i
only change one coordinate, we get
=
i
and
(s
q
. . . s
n1
s
n
) =
i
.
We let C
(i)
be the representation having dimension vector
i
. Then we dene
V = F

n
F

n1
. . . F

q
C
(i)
.
This is an indecomposable representation and
d(V ) = .
Example 5.38. Let us demonstrate by example how reection functors work. Consider the quiver
D
4
with the orientation of all arrows towards the node (which is labeled by 4). Start with the
1-dimensional representation V

4
sitting at the 4-th vertex. Apply to V

4
the functor F

3
F

2
F

1
.
This yields
F

1
F

2
F

3
V

4
= V

1
+
2
+
3
+
4
.
Now applying F

4
we get
F

4
F

1
F

2
F

3
V

4
= V

1
+
2
+
3
+2
4
.
Note that this is exactly the inclusion of 3 lines into the plane, which is the most complicated
indecomposable representation of the D
4
quiver.
5.9 Problems
Problem 5.39. Let Q
n
be the cyclic quiver of length n, i.e., n vertices connected by n oriented edges
forming a cycle. Obviously, the classication of indecomposable representations of Q
1
is given by
the Jordan normal form theorem. Obtain a similar classication of indecomposable representations
of Q
2
. In other words, classify pairs of linear operators A : V W and B : W V up to
isomorphism. Namely:
(a) Consider the following pairs (for n 1):
1) E
n,
: V = W = C
n
, A is the Jordan block of size n with eigenvalue , B = 1 ( C).
2) E
n,
: is obtained from E
n,0
by exchanging V with W and A with B.
96
3) H
n
: V = C
n
with basis v
i
, W = C
n1
with basis w
i
, Av
i
= w
i
, Bw
i
= v
i+1
for i < n, and
Av
n
= 0.
4) K
n
is obtained from H
n
by exchanging V with W and A with B.
Show that these are indecomposable and pairwise nonisomorphic.
(b) Show that if E is a representation of Q
2
such that AB is not nilpotent, then E = E

,
where E

= E
n,
for some ,= 0.
(c) Consider the case when AB is nilpotent, and consider the operator X on V W given
by X(v, w) = (Bw, Av). Show that X is nilpotent, and admits a basis consisting of chains (i.e.,
sequences u, Xu, X
2
u, ...X
l1
u where X
l
u = 0) which are compatible with the direct sum decompo-
sition (i.e., for every chain u V or u W). Deduce that (1)-(4) are the only indecomposable
representations of Q
2
.
(d)(harder!) generalize this classication to the Kronecker quiver, which has two vertices 1 and
2 and two edges both going from 1 to 2.
(e)(still harder!) can you generalize this classication to Q
n
, n > 2, with any orientation?
Problem 5.40. Let L
1
2
Z
8
be the lattice of vectors where the coordinates are either all integers
or all half-integers (but not integers), and the sum of all coordinates is an even integer.
(a) Let
i
= e
i
e
i+1
, i = 1, ..., 6,
7
= e
6
+ e
7
,
8
= 1/2

8
i=1
e
i
. Show that
i
are a basis
of L (over Z).
(b) Show that roots in L (under the usual inner product) form a root system of type E
8
(compute
the inner products of
i
).
(c) Show that the E
7
and E
6
lattices can be obtained as the sets of vectors in the E
8
lattice L
where the rst two, respectively three, coordinates (in the basis e
i
) are equal.
(d) Show that E
6
, E
7
, E
8
have 72,126,240 roots, respectively (enumerate types of roots in terms
of the presentations in the basis e
i
, and count the roots of each type).
Problem 5.41. Let V

be the indecomposable representation of a Dynkin quiver Q which corre-

sponds to a positive root . For instance, if
i
is a simple root, then V

i
has a 1-dimensional space
at i and 0 everywhere else.
(a) Show that if i is a source then Ext
1
(V, V

i
) = 0 for any representation V of Q, and if i is
a sink, then Ext
1
(V

i
, V ) = 0.
(b) Given an orientation of the quiver, nd a Jordan-Holder series of V

for that orientation.

97
6 Introduction to categories
6.1 The denition of a category
We have now seen many examples of representation theories and of operations with representations
(direct sum, tensor product, induction, restriction, reection functors, etc.) A context in which one
can systematically talk about this is provided by Category Theory.
Category theory was founded by Saunders MacLane and Samuel Eilenberg around 1940. It is a
fairly abstract theory which seemingly has no content, for which reason it was christened abstract
nonsense. Nevertheless, it is a very exible and powerful language, which has become totally
indispensable in many areas of mathematics, such as algebraic geometry, topology, representation
theory, and many others.
We will now give a very short introduction to Category theory, highlighting its relevance to the
topics in representation theory we have discussed. For a serious acquaintance with category theory,
the reader should use the classical book [McL].
Denition 6.1. A category ( is the following data:
(i) a class of objects Ob(();
(ii) for every objects X, Y Ob((), the class Hom
(
(X, Y ) = Hom(X, Y ) of morphisms (or
arrows) from X, Y (for f Hom(X, Y ), one may write f : X Y );
(iii) For any objects X, Y, Z Ob((), a composition map Hom(Y, Z)Hom(X, Y ) Hom(X, Z),
(f, g) f g,
which satisfy the following axioms:
1. The composition is associative, i.e., (f g) h = f (g h);
2. For each X Ob((), there is a morphism 1
X
Hom(X, X), called the unit morphism, such
that 1
X
f = f and g 1
X
= g for any f, g for which compositions make sense.
Remark. We will write X ( instead of X Ob(().
Example 6.2. 1. The category Sets of sets (morphisms are arbitrary maps).
2. The categories Groups, Rings (morphisms are homomorphisms).
3. The category Vect
k
of vector spaces over a eld k (morphisms are linear maps).
4. The category Rep(A) of representations of an algebra A (morphisms are homomorphisms of
representations).
5. The category of topological spaces (morphisms are continuous maps).
6. The homotopy category of topological spaces (morphisms are homotopy classes of continuous
maps).
Important remark. Unfortunately, one cannot simplify this denition by replacing the word
class by the much more familiar word set. Indeed, this would rule out the important Example 1,
as it is well known that there is no set of all sets, and working with such a set leads to contradictions.
The precise denition of a class and the precise distinction between a class and a set is the subject
of set theory, and cannot be discussed here. Luckily, for most practical purposes (in particular, in
these notes), this distinction is not essential.
98
We also mention that in many examples, including examples 1-6, the word class in (ii) can
be replaced by set. Categories with this property (that Hom(X, Y ) is a set for any X, Y ) are
called locally small; many categories that we encounter are of this kind.
Sometimes the collection Hom(X, Y ) of morphisms from X to Y in a given locally small category
( is not just a set but has some additional structure (say, the structure of an abelian group, or
a vector space over some eld). In this case one says that ( is enriched over another category
T (which is a monoidal category, i.e., has a product operation and a neutral object under this
product, e.g. the category of abelian groups or vector spaces). This means that for each X, Y (,
Hom(X, Y ) is an object of T, and the composition Hom(Y, Z) Hom(X, Y ) Hom(X, Z) is a
morphism in T. For a more detailed discussion of this, we refer the reader to [McL].
Example. The category Rep(A) of representations of a k-algebra A is enriched over the
category of k-vector spaces.
Denition 6.3. A full subcategory of a category ( is a category (

whose objects are a subclass

of objects of (, and Hom
(
(X, Y ) = Hom
(
(X, Y ).
Example. The category AbelianGroups is a full subcategory of the category Groups.
6.2 Functors
We would like to dene arrows between categories. Such arrows are called functors.
Denition 6.4. A functor F : ( T between categories ( and T is
(i) a map F : Ob(() Ob(T);
(ii) for each X, Y (, a map F = F
X,Y
: Hom(X, Y ) Hom(F(X), F(Y )) which preserves
compositions and identity morphisms.
Note that functors can be composed in an obvious way. Also, any category has the identity
functor.
Example 6.5. 1. A (locally small) category ( with one object X is the same thing as a monoid.
A functor between such categories is a homomorphism of monoids.
2. Forgetful functors Groups Sets, Rings AbelianGroups.
3. The opposite category of a given category is the same category with the order of arrows and
compositions reversed. Then V V

is a functor Vect
k
Vect
op
k
.
4. The Hom functors: If ( is a locally small category then we have the functor ( Sets given
by Y Hom(X, Y ) and (
op
Sets given by Y Hom(Y, X).
5. The assignment X Fun(X, Z) is a functor Sets Rings
op
.
6. Let Q be a quiver. Consider the category ((Q) whose objects are the vertices and morphisms
are oriented paths between them. Then functors from ((Q) to Vect
k
are representations of Q over
k.
7. Let K G be groups. Then we have the induction functor Ind
G
K
: Rep(K) Rep(G), and
Res
G
K
: Rep(G) Rep(K).
8. We have an obvious notion of the Cartesian product of categories (obtained by taking the
Cartesian products of the classes of objects and morphisms of the factors). The functors of direct
99
sum and tensor product are then functors Vect
k
Vect
k
Vect
k
. Also the operations V V
n
,
V S
n
V , V
n
V are functors on Vect
k
. More generally, if is a representation of S
n
, we
have functors V Hom
Sn
(, V
n
). Such functors (for irreducible ) are called the Schur functors.
They are labeled by Young diagrams.
9. The reection functors F

i
: Rep(Q) Rep(

Q
i
) are functors between representation cate-
gories of quivers.
6.3 Morphisms of functors
One of the important features of functors between categories which distinguishes them from usual
maps or functions is that the functors between two given categories themselves form a category,
i.e., one can dene a nontrivial notion of a morphism between two functors.
Denition 6.6. Let (, T be categories and F, G : ( T be functors between them. A morphism
a : F G (also called a natural transformation or a functorial morphism) is a collection of
morphisms a
X
: F(X) G(X) labeled by the objects X of (, which is functorial in X, i.e., for
any morphism f : X Y (for X, Y () one has a
Y
F(f) = G(f) a
X
.
A morphism a : F G is an isomorphism if there is another morphism a
1
: G F such that
aa
1
and a
1
a are the identities. The set of morphisms from F to G is denoted by Hom(F, G).
Example 6.7. 1. Let FVect
k
be the category of nite dimensional vector spaces over k. Then the
functors id and on this category are isomorphic. The isomorphism is dened by the standard
maps a
V
: V V

given by a
V
(u)(f) = f(u), u V , f V

. But these two functors are not
isomorphic on the category of all vector spaces Vect
k
, since for an innite dimensional vector space
V , V is not isomorphic to V

.
2. Let FVect

k
be the category of nite dimensional k-vector spaces, where the morphisms
are the isomorphisms. We have a functor F from this category to itself sending any space V to
V

and any morphism a to (a

)
1
. This functor satises the property that V is isomorphic to
F(V ) for any V , but it is not isomorphic to the identity functor. This is because the isomorphism
V F(V ) = V

cannot be chosen to be compatible with the action of GL(V ), as V is not
isomorphic to V

as a representation of GL(V ).
3. Let A be an algebra over a eld k, and F : A mod Vect
k
be the forgetful functor.
Then as follows from Problem 1.22, EndF = Hom(F, F) = A.
4. The set of endomorphisms of the identity functor on the category A mod is the center of
A (check it!).
6.4 Equivalence of categories
When two algebraic or geometric objects are isomorphic, it is usually not a good idea to say that
they are equal (i.e., literally the same). The reason is that such objects are usually equal in many
dierent ways, i.e., there are many ways to pick an isomorphism, but by saying that the objects are
equal we are misleading the reader or listener into thinking that we are providing a certain choice
of the identication, which we actually do not do. A vivid example of this is a nite dimensional
vector space V and its dual space V

.
For this reason in category theory, one most of the time tries to avoid saying that two objects
or two functors are equal. In particular, this applies to the denition of isomorphism of categories.
100
Namely, the naive notion of isomorphism of categories is dened in the obvious way: a functor
F : ( T is an isomorphism if there exists F
1
: T ( such that F F
1
and F
1
F are equal
to the identity functors. But this denition is not very useful. We might suspect so since we have
used the word equal for objects of a category (namely, functors) which we are not supposed to
do. And in fact here is an example of two categories which are the same for all practical purposes
but are not isomorphic; it demonstrates the deciency of our denition.
Namely, let (
1
be the simplest possible category: Ob((
1
) consists of one object X, with
Hom(X, X) = 1
X
. Also, let (
2
have two objects X, Y and 4 morphisms: 1
X
, 1
Y
, a : X Y
and b : Y X. So we must have a b = 1
Y
, b a = 1
X
.
It is easy to check that for any category T, there is a natural bijection between the collections
of isomorphism classes of functors (
1
T and (
2
T (both are identied with the collection of
isomorphism classes of objects of T). This is what we mean by saying that (
1
and (
2
are the same
for all practical purposes. Nevertheless they are not isomorphic, since (
1
has one object, and (
2
has two objects (even though these two objects are isomorphic to each other).
This shows that we should adopt a more exible and less restrictive notion of isomorphism of
categories. This is accomplished by the denition of an equivalence of categories.
Denition 6.8. A functor F : ( T is an equivalence of categories if there exists F

: T (
such that F F

and F

F are isomorphic to the identity functors.

In this situation, F

is said to be a quasi-inverse to F.
In particular, the above categories (
1
and (
2
are equivalent (check it!).
Also, the category FSet of nite sets is equivalent to the category whose objects are nonneg-
ative integers, and morphisms are given by Hom(m, n) = Maps(1, ..., m, 1, ..., n). Are these
categories isomorphic? The answer to this question depends on whether you believe that there
is only one nite set with a given number of elements, or that there are many of those. It seems
better to think that there are many (without asking how many), so that isomorphic sets need not
be literally equal, but this is really a matter of choice. In any case, this is not really a reasonable
question; the answer to this question is irrelevant for any practical purpose, and thinking about it
will give you nothing but a headache.
6.5 Representable functors
A fundamental notion in category theory is that of a representable functor. Namely, let ( be a
(locally small) category, and F : ( Sets be a functor. We say that F is representable if there
exists an object X ( such that F is isomorphic to the functor Hom(X, ?). More precisely, if we
are given such an object X, together with an isomorphism : F

= Hom(X, ?), we say that the
functor F is represented by X (using ).
In a similar way, one can talk about representable functors from (
op
to Sets. Namely, one
calls such a functor representable if it is of the form Hom(?, X) for some object X (, up to an
isomorphism.
Not every functor is representable, but if a representing object X exists, then it is unique.
Namely, we have the following lemma.
Lemma 6.9. (The Yoneda Lemma) If a functor F is represented by an object X, then X is unique
up to a unique isomorphism. I.e., if X, Y are two objects in (, then for any isomorphism of functors
: Hom(X, ?) Hom(Y, ?) there is a unique isomorphism a

: X Y inducing .
101
Proof. (Sketch) One sets a

=
1
Y
(1
Y
), and shows that it is invertible by constructing the inverse,
which is a
1

=
X
(1
X
). It remains to show that the composition both ways is the identity, which
we will omit here. This establishes the existence of a

. Its uniqueness is veried in a straightforward

manner.
Remark. In a similar way, if a category ( is enriched over another category T (say, the category
of abelian groups or vector spaces), one can dene the notion of a representable functor from ( to
T.
Example 6.10. Let A be an algebra. Then the forgetful functor to vector spaces on the category
of left A-modules is representable, and the representing object is the free rank 1 module (=the
regular representation) M = A. But if A is innite dimensional, and we restrict attention to the
category of nite dimensional modules, then the forgetful functor, in general, is not representable
(this is so, for example, if A is the algebra of complex functions on Z which are zero at all points
but nitely many).
6.6 Adjoint functors
Another fundamental notion in category theory is the notion of adjoint functors.
Denition 6.11. Functors F : ( T and G : T ( are said to be a pair of adjoint functors if for
any X (, Y T we are given an isomorphism
XY
: Hom
(
(F(X), Y ) Hom
T
(X, G(Y )) which is
functorial in X and Y ; in other words, if we are given an isomorphism of functors Hom(F(?), ?)
Hom(?, G(?)) (( T Sets). In this situation, we say that F is left adjoint to G and G is right
adjoint to F.
Not every functor has a left or right adjoint, but if it does, it is unique and can be constructed
canonically (i.e., if we somehow found two such functors, then there is a canonical isomorphism
between them). This follows easily from the Yoneda lemma, as if F, G are a pair of adjoint functors
then F(X) represents the functor Y Hom(X, G(Y )), and G(Y ) represents the functor X
Hom(F(X), Y ).
Remark 6.12. The terminology left and right adjoint functors is motivated by the analogy
between categories and inner product spaces. More specically, we have the following useful dic-
tionary between category theory and linear algebra, which helps understand better many notions
of category theory.
Dictionary between category theory and linear algebra
Category ( Vector space V with a nondegenerate inner product
The set of morphisms Hom(X, Y ) Inner product (x, y) on V (maybe nonsymmetric)
Opposite category (
op
Same space V with reversed inner product
The category Sets The ground eld k
Full subcategory in ( Nondegenerate subspace in V
Functor F : ( T Linear operator f : V W
Functor F : ( Sets Linear functional f V

= Hom(V, k)
Representable functor Linear functional f V

given by f(v) = (u, v), u V

Yoneda lemma Nondegeneracy of the inner product (on both sides)
Not all functors are representable If dimV = , not f V

, f(v) = (u, v)
Left and right adjoint functors Left and right adjoint operators
Adjoint functors dont always exist Adjoint operators may not exist if dimV =
If they do, they are unique If they do, they are unique
Left and right adjoints may not coincide The inner product may be nonsymmetric
102
Example 6.13. 1. Let V be a nite dimensional representation of a group G or a Lie algebra g.
Then the left and right adjoint to the functor V on the category of representations of G is the
functor V

.
2. The functor Res
G
K
is left adjoint to Ind
G
K
. This is nothing but the statement of the Frobenius
reciprocity.
3. Let Assoc
k
be the category of associative unital algebras, and Lie
k
the category of Lie
algebras over some eld k. We have a functor L : Assoc
k
Lie
k
, which attaches to an associative
algebra the same space regarded as a Lie algebra, with bracket [a, b] = ab ba. Then the functor L
has a left adjoint, which is the functor U of taking the universal enveloping algebra of a Lie algebra.
4. We have the functor GL
1
: Assoc
k
Groups, given by A GL
1
(A) = A

. This functor
has a left adjoint, which is the functor G k[G], the group algebra of G.
5. The left adjoint to the forgetful functor Assoc
k
Vect
k
is the functor of tensor algebra:
V TV . Also, if we denote by Comm
k
the category of commutative algebras, then the left adjoint
to the forgetful functor Comm
k
Vect
k
is the functor of the symmetric algebra: V SV .
One can give many more examples, spanning many elds. These examples show that adjoint
functors are ubiquitous in mathematics.
6.7 Abelian categories
The type of categories that most often appears in representation theory is abelian categories.
The standard denition of an abelian category is rather long, so we will not give it here, referring
the reader to the textbook [Fr]; rather, we will use as the denition what is really the statement of
the Freyd-Mitchell theorem:
Denition 6.14. An abelian category is a category (enriched over the category of abelian groups),
which is equivalent to a full subcategory ( of the category A-mod of left modules over a ring A,
closed under taking nite direct sums, as well as kernels, cokernels, and images of morphisms.
We see from this denition that in an abelian category, Hom(X, Y ) is an abelian group for each
X, Y , compositions are group homomorphisms with respect to each argument, there is the zero ob-
ject, the notion of an injective morphism (monomorphism) and surjective morphism (epimorphism),
and every morphism has a kernel, a cokernel, and an image.
Example 6.15. The category of modules over an algebra A and the category of nite dimensional
modules over A are abelian categories.
Remark 6.16. The good thing about Denition 6.14 is that it allows us to visualize objects,
morphisms, kernels, and cokernels in terms of classical algebra. But the denition also has a big
drawback, which is that even if ( is the whole category A-mod, the ring A is not determined by (.
In particular, two dierent rings can have equivalent categories of modules (such rings are called
Morita equivalent). Actually, it is worse than that: for many important abelian categories there
is no natural (or even manageable) ring A at all. This is why people prefer to use the standard
denition, which is free from this drawback, even though it is more abstract.
We say that an abelian category ( is k-linear if the groups Hom
(
(X, Y ) are equipped with
a structure of a vector space over k, and composition maps are k-linear in each argument. In
particular, the categories in Example 6.15 are k-linear.
103
6.8 Exact functors
Denition 6.17. A sequence of objects and morphisms
X
0
X
1
... X
n+1
in an abelian category is said to be a complex if the composition of any two consecutive arrows
is zero. The cohomology of this complex is H
i
= Ker (d
i
)/Im(d
i1
), where d
i
: X
i
X
i+1
(thus
the cohomology is dened for 1 i n). The complex is said to be exact in the i-th term if
H
i
= 0, and is said to be an exact sequence if it is exact in all terms. A short exact sequence
is an exact sequence of the form
0 X Y Z 0.
Clearly, 0 X Y Z 0 is a short exact sequence if and only if X Y is injective,
Y Z is surjective, and the induced map Y/X Z is an isomorphism.
Denition 6.18. A functor F between two abelian categories is additive if it induces homomor-
phisms on Hom groups. Also, for k-linear categories one says that F is k-linear if it induces k-linear
maps between Hom spaces.
It is easy to show that if F is an additive functor, then F(X Y ) is canonically isomorphic to
F(X) F(Y ).
Example 6.19. The functors Ind
G
K
, Res
G
K
, Hom
G
(V, ?) in the theory of group representations over
a eld k are additive and k-linear.
Denition 6.20. An additive functor F : ( T between abelian categories is left exact if for
any exact sequence
0 X Y Z,
the sequence
0 F(X) F(Y ) F(Z)
is exact. F is right exact if for any exact sequence
X Y Z 0,
the sequence
F(X) F(Y ) F(Z) 0
is exact. F is exact if it is both left and right exact.
Denition 6.21. An abelian category ( is semisimple if any short exact sequence in this category
splits, i.e., is isomorphic to a sequence
0 X X Y Y 0
(where the maps are obvious).
Example 6.22. The category of representations of a nite group G over a eld of characteristic
not dividing [G[ (or 0) is semisimple.
Note that in a semisimple category, any additive functor is automatically exact on both sides.
104
Example 6.23. (i) The functors Ind
G
K
, Res
G
K
are exact.
(ii) The functor Hom(X, ?) is left exact, but not necessarily right exact. To see that it need not
be right exact, it suces to consider the exact sequence
0 Z Z Z/2Z 0,
and apply the functor Hom(Z/2Z, ?).
(iii) The functor X
A
for a right A-module X (on the category of left A-modules) is right exact,
but not necessarily left exact. To see this, it suces to tensor multiply the above exact sequence
by Z/2Z.
Exercise. Show that if (F, G) is a pair of adjoint additive functors between abelian categories,
then F is right exact and G is left exact.
Exercise. (a) Let Q be a quiver and i Q a source. Let V be a representation of Q, and W a
representation of Q
i
(the quiver obtained from Q by reversing arrows at the vertex i). Prove that
there is a natural isomorphism between Hom
_
F

i
V, W
_
and Hom
_
V, F
+
i
W
_
. In other words, the
functor F
+
i
is right adjoint to F

i
.
(b) Deduce that the functor F
+
i
is left exact, and F

i
is right exact.
105
7 Structure of nite dimensional algebras
In this section we return to studying the structure of nite dimensional algebras. Throughout the
section, we work over an algebraically closed eld k (of any characteristic).
7.1 Projective modules
Let A be an algebra, and P be a left A-module.
Theorem 7.1. The following properties of P are equivalent:
(i) If : M N is a surjective morphism, and : P N any morphism, then there exists a
morphism : P M such that = .
(ii) Any surjective morphism : M P splits, i.e., there exists : P M such that = id.
(iii) There exists another A-module Q such that P Q is a free A-module, i.e., a direct sum of
copies of A.
(iv) The functor Hom
A
(P, ?) on the category of A-modules is exact.
Proof. To prove that (i) implies (ii), take N = P. To prove that (ii) implies (iii), take M to be free
(this can always be done since any module is a quotient of a free module). To prove that (iii) implies
(iv), note that the functor Hom
A
(P, ?) is exact if P is free (as Hom
A
(A, N) = N), so the statement
follows, as if the direct sum of two complexes is exact, then each of them is exact. To prove that
(iv) implies (i), let K be the kernel of the map , and apply the exact functor Hom
A
(P, ?) to the
exact sequence
0 K M N 0.
Denition 7.2. A module satisfying any of the conditions (i)-(iv) of Theorem 7.1 is said to be
projective.
7.2 Lifting of idempotents
Let A be a ring, and I A a nilpotent ideal.
Proposition 7.3. Let e
0
A/I be an idempotent, i.e., e
2
0
= e
0
. There exists an idempotent e A
which is a lift of e
0
(i.e., it projects to e
0
under the reduction modulo I). This idempotent is unique
up to conjugation by an element of 1 +I.
Proof. Let us rst establish the statement in the case when I
2
= 0. Note that in this case I is a
left and right module over A/I. Let e

be any lift of e
0
to A. Then e
2

= a I, and e
0
a = ae
0
.
We look for e in the form e = e

+b, b I. The equation for b is e

0
b +be
0
b = a.
Set b = (2e
0
1)a. Then
e
0
b +be
0
b = 2e
0
a (2e
0
1)a = a,
so e is an idempotent. To classify other solutions, set e

= e + c. For e

to be an idempotent, we
must have ec + ce c = 0. This is equivalent to saying that ece = 0 and (1 e)c(1 e) = 0, so
c = ec(1 e) + (1 e)ce = [e, [e, c]]. Hence e

= (1 + [c, e])e(1 + [c, e])

1
.
106
Now, in the general case, we prove by induction in k that there exists a lift e
k
of e
0
to A/I
k+1
,
and it is unique up to conjugation by an element of 1 + I
k
(this is sucient as I is nilpotent).
Assume it is true for k = m 1, and let us prove it for k = m. So we have an idempotent
e
m1
A/I
m
, and we have to lift it to A/I
m+1
. But (I
m
)
2
= 0 in A/I
m+1
, so we are done.
Denition 7.4. A complete system of orthogonal idempotents in a unital algebra B is a collection
of elements e
1
, ..., e
n
B such that e
i
e
j
=
ij
e
i
, and

n
i=1
e
i
= 1.
Corollary 7.5. Let e
01
, ..., e
0m
be a complete system of orthogonal idempotents in A/I. Then there
exists a complete system of orthogonal idempotents e
1
, ..., e
m
(e
i
e
j
=
ij
e
i
,

e
i
= 1) in A which
lifts e
01
, ..., e
0m
.
Proof. The proof is by induction in m. For m = 2 this follows from Proposition 7.3. For m > 2,
we lift e
01
to e
1
using Proposition 7.3, and then apply the induction assumption to the algebra
(1 e
1
)A(1 e
1
).
7.3 Projective covers
Obviously, every nitely generated projective module over a nite dimensional algebra A is a direct
sum of indecomposable projective modules, so to understand nitely generated projective modules
over A, it suces to classify indecomposable ones.
Let A be a nite dimensional algebra, with simple modules M
1
, ..., M
n
.
Theorem 7.6. (i) For each i = 1, ..., n there exists a unique indecomposable nitely generated
projective module P
i
such that dimHom(P
i
, M
j
) =
ij
.
(ii) A =
n
i=1
(dimM
i
)P
i
.
(iii) any indecomposable nitely generated projective module over A is isomorphic to P
i
for
some i.
Proof. Recall that A/Rad(A) =
n
i=1
End(M
i
), and Rad(A) is a nilpotent ideal. Pick a basis of
M
i
, and let e
0
ij
= E
i
jj
, the rank 1 projectors projecting to the basis vectors of this basis (j =
1, ..., dimM
i
). Then e
0
ij
are orthogonal idempotents in A/Rad(A). So by Corollary 7.5 we can lift
them to orthogonal idempotents e
ij
in A. Now dene P
ij
= Ae
ij
. Then A =
i

dimM
i
j=1
P
ij
, so P
ij
are projective. Also, we have Hom(P
ij
, M
k
) = e
ij
M
k
, so dimHom(P
ij
, M
k
) =
ik
. Finally, P
ij
is
independent of j up to an isomorphism, as e
ij
for xed i are conjugate under A

by Proposition
7.3; thus we will denote P
ij
by P
i
.
We claim that P
i
is indecomposable. Indeed, if P
i
= Q
1
Q
2
, then Hom(Q
l
, M
j
) = 0 for all j
either for l = 1 or for l = 2, so either Q
1
= 0 or Q
2
= 0.
Also, there can be no other indecomposable nitely generated projective modules, since any
such module has to occur in the decomposition of A. The theorem is proved.
References
[BGP] J. Bernstein, I. Gelfand, V. Ponomarev, Coxeter functors and Gabriels theorem, Russian
Math. Surveys 28 (1973), no. 2, 1732.
107
[Cu] C. Curtis, Pioneers of Representation Theory: Frobenius, Burnside, Schur, and Brauer, AMS,
1999.
[CR] C. Curtis and I. Reiner, Representation Theory of Finite Groups and Associative Algebras,
AMS, 2006.
[FH] W. Fulton and J. Harris, Representation Theory, A rst course, Springer, New York, 1991.
[Fr] Peter J. Freyd, Abelian Categories, an Introduction to the Theory of Functors. Harper and
Row (1964).
[McL] S. MacLane, Categories for a working Mathematician: 2nd Ed., Graduate Texts in Mathe-
matics 5, Springer, 1998.
108