Representations of finite groups

Abstract
A micro-introduction to the theory of representations of finite groups.

1 Representations
Let G be a finite group. A linear representation of G is a homomorphism
ρ : G → GL(V ) where GL(V ) is the group of invertable linear transformations
of the vector space V . We shall restrict ourselves to finite-dimensional V . The
dimension dim V = n is called the degree of the representation. In order to make
life easy, we only consider vector spaces over C, the field of complex numbers.
(The theory is easy for finite groups because we can average over the group
to get something that is invariant for the group action. In the averaging process
we divide by the order of the group, and the theory (of modular representations)
is more difficult when the characteristic of the field divides the order of G. For
Schur’s Lemma we need an eigenvalue, and life is a bit easier for algebraically
closed fields.)
Two representations ρ1 : G → GL(V1 ) and ρ2 : G → GL(V2 ) are called
equivalent when they are not really different: V1 and V2 have the same dimen-
sion, and for a suitable choice of bases in V1 and V2 the matrices of ρ1 (g) and
ρ2 (g) are the same, for all g ∈ G. (Equivalently, ρ1 and ρ2 are equivalent when
there is a linear isomorphism f : V1 → V2 such that f ρ1 (g) = ρ2 (g)f for all
g ∈ G.)
A subspace W of V is called ρ(G)-invariant if ρ(g)W ⊆ W for all g ∈ G.
The first example of averaging is to get a ρ(G)-invariant complement of a ρ(G)-
invariant subspace.

Theorem 1.1 (Maschke) Let G be a finite group, let V be a vector space over
the field F , and let ρ : G → GL(V ) be a linear representation of G on V . If
the subspace W of V is ρ(G)-invariant and |G| is nonzero in F , then there is a
ρ(G)-invariant subspace U of V such that V = U ⊕ W .

Proof: Let U0 be a complement of W in V , so that V = U0 ⊕ W . Let P0

be the projection onto W along U0 , that is, P0 v = w when v = u0 + w with
u0 ∈ U0 and w ∈ W . Put
1 X
P = ρ(g)P0 ρ(g)−1 .
|G|
g∈G

Then P ρ(g) = ρ(g)P for all g ∈ G, and P is a projection onto W . Now the
kernel U of P is ρ(G)-invariant. 2

1
1.1 Direct sum
Given two representations ρ1 : G → GL(V1 ) and ρ2 : G → GL(V2 ), their direct
sum ρ1 ⊕ ρ2 is the representation ρ : G → GL(V ), where V = V1 ⊕ V2 , defined
by ρ(g)v = ρ1 (g)v1 + ρ2 (g)v2 for v = v1 + v2 with v1 ∈ V1 and v2 ∈ V2 . In
matrix form, this says that the matrix R(g) of ρ(g) is given by
 
R1 (g) 0
R(g) = .
0 R2 (g)
The degree of a direct sum is the sum of the degrees of the summands. A
representation ρ is called irreducible if it is not the direct sum ρ1 ⊕ ρ2 of two
representations of nonzero degree.
By induction on dim V we see that each representation is the direct sum
ρ1 ⊕ · · · ⊕ ρm of (zero or more) irreducible representations.

1.2 Tensor product

Given two representations ρ1 : G → GL(V1 ) and ρ2 : G → GL(V2 ), their tensor
product ρ1 ⊗ ρ2 is the representation ρ : G → GL(V ), where V = V1 ⊗ V2 ,
defined by ρ(g)v = ρ1 (g)v1 ⊗ ρ2 (g)v2 for v = v1 ⊗ v2 with v1 ∈ V1 and v2 ∈ V2 .
In matrix form, this says that the matrix R(g) of ρ(g) is R(g) = R1 (g) ⊗ R2 (g).
The degree of a tensor product is the product of the degrees of the factors.

1.3 Schur’s Lemma

Theorem 1.2 (Schur) Let ρ1 : G → GL(V1 ) and ρ2 : G → GL(V2 ) be irre-
ducible representations. Let f : V1 → V2 be a linear map with ρ2 (g)f = f ρ1 (g)
for all g ∈ G.
(i) If ρ1 and ρ2 are not equivalent, then f = 0.
(ii) If V1 = V2 = V and ρ1 = ρ2 , then f = λI, where I is the identity on V .

Proof: (i) We may suppose f 6= 0. The subspace W1 = ker f of V1 is ρ1 (G)-

invariant, so W1 = 0 or W1 = V1 , but f 6= 0, so W1 = 0. The subspace
W2 = im f of V2 is ρ2 (G)-invariant, so W2 = 0 or W2 = V2 , but f 6= 0, so
W2 = V2 . Now f is an isomorphism, and ρ1 and ρ2 are equivalent.
(ii) Let λ be an eigenvalue of f (there is one, since C is algebraically closed).
Now apply the previous argument to f − λI instead of f . Since f − λI is not
an isomorphism, is must be 0. 2
By averaging over G we can turn a linear map f into one that satisfies the
hypothesis of Theorem 1.2. This yields:

Corollary 1.3 Let ρ1 : G → GL(V1 ) and ρ2 : G → GL(V P 2 ) be irreducible

representations. Let f : V1 → V2 be linear and put f˜ = |G|
1
g∈G ρ2 (g)
−1
f ρ1 (g).
˜
(i) If ρ1 and ρ2 are not equivalent, then f = 0.
(ii) If V1 = V2 = V and ρ1 = ρ2 , then f˜ = λI, where λ = n1 tr f .

Proof: In case (ii), tr f˜ = tr f and tr I = n. 2

Now choose bases, and let f = Ejk be the linear map with a matrix that is
zero everywhere except at the (j, k)-entry where it is 1. We obtain orthogonality
relations for the matrix entries of representations.

2
Corollary 1.4 Let ρ1 : G → GL(V1 ) and ρ2 : G → GL(V2 ) be irreducible
representations.
(i) If ρ1 and ρ2 are not equivalent, then
1 X
R2 (g)−1
ij R1 (g)kl = 0 for all i, j, k, l
|G|
g∈G

where Ri (g) denotes the matrix of ρi (g) for i = 1, 2 and g ∈ G.

(ii) If V1 = V2 = V and ρ1 = ρ2 = ρ, then

1 X 1/n if i = l and j = k
R(g)−1
ij R(g)kl =
|G| 0 otherwise
g∈G

where R(g) denotes the matrix of ρ(g) for g ∈ G.

Proof: Apply the previous corollary, and take the (i, l)-matrix entry of f˜
where f = Ejk . In case (ii), tr f equals 0 if j 6= k and 1 if j = k. 2

2 Characters
Given a representation ρ : G → GL(V ), let its character be the map χ : G → C
defined by χ(g) = tr ρ(g). It will turn out that ρ is determined up to equivalence
by its character χ.
Lemma 2.1 Let χ = χρ denote the character of ρ. Then
(i) χρ1 ⊕ρ2 = χρ1 + χρ2 ,
(ii) χρ1 ⊗ρ2 = χρ1 χρ2 ,
(iii) χρ (1) = n if ρ has degree n,
(iv) χ(g −1 ) = χ(g) for all g ∈ G,
(v) χ(h−1 gh) = χ(g) for all g, h ∈ G.
Proof: Only part (iv) requires comment. Since G is finite, g has finite order, so
ρ(g) has finite order, and its eigenvalues are roots of unity. If ρ(g) has eigenvalue
ζ, then ρ(g −1 ) has eigenvalue ζ −1 = ζ. And the trace of ρ(g −1 ) is the sum of
its eigenvalues. 2

2.1 Inner product

For functions φ, ψ : G → C, put
1 X
hφ, ψi = φ(g) ψ(g).
|G|
g∈G

Theorem 2.2 If the representation ρ : G → GL(V ) is irreducible, then its

character χ satisfies hχ, χi = 1. If the irreducible representations ρ, ρ0 are in-
equivalent, then their characters χ, χ0 satisfy hχ, χ0 i = 0.
Proof: Since χ(g) = tr ρ(g) = i R(g)ii and χ(g) = χ(g −1 ), we have
P

1 X 1 XX
hχ, χ0 i = χ(g −1 )χ0 (g) = R(g)−1 0
ii R (g)jj .
|G| |G| i,j
g∈G g∈G

Now apply Corollary 1.4. 2

3
2.2 The character determines the representation
Theorem 2.3 Let σ : G → GL(V ) be a representation with character φ, and
let σ = σ1 ⊕ · · · ⊕ σm be a decomposition of σ into irreducible representations.
Let ρ : G → GL(W ) be an irreducible representation with character χ. Then
the number of σi equivalent to ρ equals hφ, χi.

Proof: Let σi have character φi , so that φ = φ1 + · · · + φm . Then hφ, χi =

hφ1 , χi + · · · + hφm , χi. Now apply Theorem 2.2. 2
It follows that although the decomposition into irreducible subspaces is not
unique in general, the number of subspaces of given type is determined. (In
fact, also the sum of all subspaces of given type is determined uniquely.)

Corollary 2.4 Two representations are equivalent if and only if they have the
same character. 2

We saw that if ρ is irreducible, then hχ, χi = 1.PBut the opposite is also true: if
hχ, χi = 1 then ρ is irreducible. Indeed, if χ = i mi χi where the χi are distinct
irreducible
Pcharacters (that is, characters of irreducible representations), then
hχ, χi = i m2i , and this equals 1 only when there is only one summand and
m1 = 1.

2.3 The regular representation

So far, we have not constructed any representations. Let G be a finite group.
The regular representation of G is the representation ρ on the vector space CG
(with basis G) defined by ρ(g)h = gh for g, h ∈ G. This representation has
degree n = |G|, and character χ satisfying

|G| if g = 1,
χ(g) = tr ρ(g) =
0 otherwise.

If χi is any irreducible character, of degree ni , then

1 X
hχ, χi i = χ(g)χi (g) = χi (1) = ni ,
|G|
g∈G

so that χ = ni χi and |G| = i n2i .

P P
It follows that there are only finitely many distinct irreducible characters,
all found in the character of the regular representation.

2.4 Class functions

For g ∈ G, the conjugacy class C(g) is the set {h−1 gh | h ∈ G}. Lemma
2.1(v) says that a character is a class function, that is, is constant on conjugacy
classes. We shall see that conversely any class function is a linear combination
of characters.
1
P
For a class function φ and a representation ρ, let fφ,ρ = |G| g∈G φ(g)ρ(g),
a linear transformation of the vector space V , weighted average of the ρ(g).

Lemma 2.5 If ρ is irreducible, then fφ,ρ = λI, where nλ = tr fφ,ρ = hφ, χi.

4
Proof: We have
1 X 1 X
ρ(g)−1 fφ,ρ ρ(g) = φ(h)ρ(g −1 hg) = φ(ghg −1 )ρ(h) = fφ,ρ .
|G| |G|
h∈G h∈G

Now by Schur’s Lemma (Theorem 1.2) fφ,ρ = λI for some constant λ, and λ is
found by taking traces. 2

Theorem 2.6 The irreducible characters form an orthonormal basis for the
vector space of class functions. In particular, the number of irreducible charac-
ters equals the number of conjugacy classes of the group G.

Proof: The ‘orthonormal’ part is the content of Theorem 2.2. Remains ‘basis’.
If φ is a class function orthogonal to all irreducible characters χi , then consider
the linear transformation fφ,ρ for various ρ. The above lemma says that fφ,ρ = 0
when ρ is irreducible. For arbitrary ρ the function fφ,ρ is a direct sum of the
functions fφ,ρj for the irreducible constituents ρj of ρ, hence fφ,ρ = 0 for all ρ.
Now let ρ be the regular representation and compute the image of fφ,ρ on the
1
P
basis vector 1. Since ρ(g)1 = g, we find 0 = fφ,ρ 1 = |G| g∈G φ(g)g, so that all
coefficients φ(g) vanish, and φ = 0. 2

2.5 Character table

The square matrix X with rows indexed by the irreducible characters χi and
columns by the conjugacy classes of G and entries Xχ,C = χ(g) for g ∈ C, is
called the character table of G.
Let D be the diagonal matrix with rows and columns indexed by the conju-
|C|
gacy classes of G, where DCC = |G| , so that tr D = 1.
>
The fact that different characters are orthogonal is expressed by XDX = I.
>
But if AB = I then also BA = I, so it follows that X X = D−1 . This shows
that also the columns of X are orthogonal, and that the sizes of the conjugacy
classes can be seen from X.

3 Example: Sym(5)
Let G be the symmetric group Sym(5). Its character table is

1 (12)(34) (123) (12345) (12) (1234) (12)(345)

χ1 1 1 1 1 1 1 1
χ2 1 1 1 1 −1 −1 −1
χ3 4 0 1 −1 2 0 −1
χ4 4 0 1 −1 −2 0 1
χ5 5 1 −1 0 1 −1 1
χ6 5 1 −1 0 −1 1 −1
χ7 6 −2 0 1 0 0 0

The table is square, with 7 rows and columns. The columns are labeled by
representatives of the conjugacy classes. The conjugacy classes have sizes 1, 15,

5
20, 24, 10, 30, 20, respectively. The sum of the squares of the character degrees
12 + 12 + 42 + 42 + 52 + 52 + 62 = 120 equals |G|.
How was this table constructed? By finding some easy characters and de-
composing those into irreducibles.
1. The first is the trivial character, the character of the trivial representation
that maps every g ∈ G to the identity I = (1) of order 1. This gives χ1 .
2. The second is the sign character. A permutation can be even or odd, and
the sign character is 1 on even and −1 on odd permutations. This gives χ2 .
3. The third construction is that of a permutation character. If G acts as a
group of permutations on a set Ω, we find a representation in CΩ . (The regular
representation is the example where G acts on itself.) Now χ(g) = tr ρ(g) is the
number of fixpoints of g.
The group Sym(5) has an obvious action on the set S = {1, 2, 3, 4, 5}, and
the permutation character is π = (5, 1, 2, 0, 3, 1, 0) with entries in the order of
the columns of the table. Now hπ, πi = 2, so this is the sum of two irreducible
characters. And hπ, χ1 i = 1, so χ1 is one of them. Then χ3 = π − χ1 must be
the other. This gives χ3 .
The group Sym(5) also has an action on the ten pairs from S. The corre-
sponding permutation character is π2 = (10, 2, 1, 0, 4, 0, 1). Now hπ2 , π2 i = 3,
and hπ2 , χ1 i = 1, and hπ2 , χ3 i = 1, so π2 decomposes into three irreducibles,
namely χ1 and χ3 and χ5 = π2 − χ1 − χ3 . This gives χ5 .
4. The fourth construction is that of taking tensor products. We find irre-
ducible characters χ4 = χ2 χ3 and χ6 = χ2 χ5 . Now only χ7 is left, and we can
write it down using the orthogonality relations of the columns. But we can also
compute the product χ23 and find that it decomposes as χ23 = χ1 + χ3 + χ5 + χ7 .
That completes the table.

3.1 Alt(5)
The even permutations in Sym(5) form the alternating group Alt(5). It has
character table

1 (12)(34) (123) (12345) (12354)

χ1 1 1 1 1 1
χ2 3 −1 0 s t
χ3 3 −1 0 t s
χ4 4 0 1 −1 −1
χ5 5 1 −1 0 0
√ √
where s = (1 − 5)/2 and t = (1 + 5)/2.
We lose the classes of odd permutations, but the class of 5-cycles now splits
into two, since (12345) and (12354) are no longer conjugate. (In Sym(5) one had
(12354) = (45)(12345)(45), but (45) is odd.) The restrictions of the characters
of Sym(5) to Alt(5) give characters again, and we find χ1 , χ4 , χ5 and χ2 + χ3 .
1
(The formula for the inner product of two characters involves a factor |G| , so
0
if the irreducible character χ of Sym(5) vanishes outside Alt(5), and χ is the
restriction of χ to Alt(5), then hχ0 , χ0 i = 2.) If x is an element of Sym(5)\Alt(5),
and χ a character of Alt(5), then also χ0 defined by χ0 (g) = χ(x−1 gx) is a
character of Alt(5). The characters χ2 and χ3 must be related this way, so have
the same value on the classes that do not split, and all that remains is to find s

6
and t. The orthogonality relations give s2 + t2 = 3 and st = −1 (and we already
knew s + t = 1), and this determines s, t. That completes the table.
Exercise Construct matrices for a representation with character χ2 .

4 Some additional material

4.1 Frobenius reciprocity
Let G be a group and H a subgroup and let φ : H → C be a class function on
H. The function φG obtained by inducing φ up to G is by definition φG (g) =
1 −1
P
|H| x∈G φ̂(x gx), where φ̂ : G → C is defined by φ̂(h) = φ(h) for h ∈ H,
and φ̂(x) = 0 for x ∈ G \ H. Now φG is a class function on G. Conversely, if
ψ : G → C is a class function on G, then the restriction ψ|H of ψ to H is a class
function on H.

Proposition 4.1 Let φ : H → C be a class function on H and ψ : G → C a

class function on G. Then hφG , ψi = hφ, ψ|H i. In particular, if φ is a character
of H, then φG is a character of G.

Proof:

hψ, φG i = 1
ψ(g)φ̂(x−1 gx) = 1
ψ(x−1 gx)φ̂(g)
P P
|G|.|H| g,x∈G |G|.|H| g,x∈G
1
P
= |H| g∈H ψ(g)φ(g) = hψ|H , φi.

Necessary and sufficient in order to be a character is that the inner product

with all irreducibles is a nonnegative integer. 2

Another useful identity is (φ.ψ|H )G = φG .ψ.

4.2 Permutation characters

Recall that if G acts as a group of permutations on a set Ω, then the corre-
sponding permutation character is the function π : G → N defined by π(g) =
#{ω ∈ Ω | gω = ω}, the number of fixpoints of g in this action.
If G acts transitively on Ω, then the rank of the action is the number of
orbits of a point stabilizer, or, what is the same, the number of orbits of G in
the natural action on Ω × Ω.
For a group G, let 1G be the trivial character on G (that is identically 1).

Proposition 4.2 Let π be the permutation character of a permutation repre-

sentation of the group G on the set Ω. Let G have orbits Ω1 , . . . , Ωm . Let, for
1 ≤P i ≤ m, the group Hi be the stabilizer in G of some element in Ωi . Then
m
π = i=1 (1Hi )G . In particular,
(i) The number of orbits m of G equals h1, πi.
(ii) If G is transitive, then it has rank hπ, πi.
(iii) For m = 2, the number of orbits of G on Ω1 × Ω2 (which equals the
number of orbits of H1 on Ω2 and the number of orbits of H2 on Ω1 ) equals
hπ1 , π2 i.

7
Proof: The permutation representation on Ω is the direct sum of the repre-
sentations on the Ωi , so we may assume that G is transitive, i.e., m = 1. Put
H = H1 . Now Ω can be identified with the set of left cosets gH of H, with
G acting by left multiplication. A left coset xH is fixed by multiplication by g
when gxH = xH, i.e., when x−1 gx ∈ H (and each left coset xH has |H| repre-
sentatives x). Now by definition (1H )G (g) = |H|1
#{x ∈ G | x−1 gx ∈ H} = π(g),
so that π = (1H )G .
For (i), h1G , πi = h1H , 1H i = 1.
For (ii): the action of G on Ω × Ω (via g(a, b) = (ga, gb)) has character π 2 .
Now the rank is h1, π 2 i = hπ, πi.
For (iii): the action of G on Ω1 × Ω2 (via g(a, b) = (ga, gb)) has character
π1 π2 . Now the number of orbits is h1, π1 π2 i = hπ1 , π2 i. 2