Course 424

Group Representations II
Dr Timothy Murphy
EELT3

Tuesday, 13 April 1999

16:0017:30

Answer as many questions as you can; all carry the same number
of marks.
All representations are finite-dimensional over C.
1. What is meant by a measure on a compact space X? What is meant
by saying that a measure on a compact group G is invariant? Sketch
the proof that every compact group G carries such a measure. To what
extent is this measure unique?
Answer: A measure on X is a continuous linear functional
: C(X) C,
where C(X) = C(X, R) is the space of real-valued continuous functions
on X with norm kf k = sup |f (x)|.
The compact group G acts on C(G) by
(gf )(x) = f (g 1 x).
The measure is said to be invariant under G if
(gf )(f )
for all g G, f C(G).
By an average F of f C(G) we mean a function of the form
F = 1 g1 f + 2 g2 f + + r gr f,
P
where 0 i 1,
i = 1 and g1 , g2 , . . . , gr G.
If F is an average of f then
(a) inf f inf F sup F supf ;
(b) If is an invariant measure then (F ) = (f );
(c) An average of F is an average of f .

Continued overleaf

If we set
var(f ) = sup f inf f
then
var(F ) var(f )
for any average F of f . We shall establish a sequence of averages F0 =
f, F1 , F2 , . . . (each an average of its predecessor) such that var(Fi ) 0.
It follows that
Fi c R,
ie Fi (g) c for each g G.
Suppose f C(G). It is not hard to find an average F of f with
var(F ) < var(f ). Let
1
V = {g G : f (g) < (sup f + inf f ),
2
ie V is the set of points where f is below average. Since G is compact,
we can find g1 , . . . , gr such that
G = g1 V gr V.
Consider the average
F =

1
(g1 f + + gr f ) .
r

Suppose x G. Then x gi V for some i, ie

gi1 x V.
Hence

1
(gi f )(x) < (sup f + inf f ),
2

and so
1
r1
sup f + (sup f + inf f )
r
2r
1
= sup f sup f inf f .
2r

F (x) <

Hence sup F < supf and so

var(F ) < var(f ).
This allows us to construct a sequence of averages F0 = f, F1 , F2 , . . .
such that
var(f ) = var(F )0 > var(F )1 > var(F )2 > .

But that is not sufficient to show that var(F )i 0. For that we must
use the fact that any f C(G) is uniformly continuous.
[I would accept this last remark as sufficient in the exam, and would
not insist on the detailed argument that follows.]
In other words, given  > 0 we can find an open set U 3 e such that
x1 y U = |f (x) f (y)| < .
Since
(g 1 x)1 (g 1 y) = x1 y,
the same result also holds for the function gf . Hence the result holds
for any average F of f .
Let V be an open neighbourhood of e such that
V V U,

V 1 = V.

(If V satisfies the first condition, then V V 1 satisfies both conditions.)

Then
xV yV 6= = |f (x) f (y)| < .
For if xv = yv 0 then
x1 y = vv 0

U.

Since G is compact we can find g1 , . . . , gr such that

G = g1 V gr V.
Suppose f attains its minimum inf f at x0 gi V ; and suppose x gj V .
Then
gi1 x0 , gj1 x V.
Hence
gj1 x

1

1
x0 U,
gi1 x0 = gi gj1 x

and so
|f (gi gj1 x) f (x0 )| < .
In particular,
(gj gi1 f )(x) < inf f + .
Let F be the average
F =
Then
sup F <

1 X
gj gi1 f.
r2 i,j

1
r2 1
sup f + 2 (inf f + ),
2
r
r

and so
r2 1
1
var(f ) + 2 
2
r
r
2
r 1/2
var(f ),
<
r2

var(F ) <

if  < var(f )/2.

Moreover this result also holds for any average of f in place of f . It
follows that a succession of averages of this kind
F0 = f, F1 , . . . , Fs
will bring us to
1
var(f ).
2
Now repeating the same argument with Fs , and so on, we will obtain a
sequence of successive averages F0 = f, F1 , . . . with
var(F )s <

var(F )i 0.
It follows that
Fi c
(the constant function with value c).
It remains to show that this limit value c is unique. For this we introduce right averages
X
H(x) =
j f (xhj )
j

where 0 j 1,
j = 1. (Note that a right average of f is
in effect a left average of f, where f(x) = f (x1 ). In particular the
results we have established for left averages will hold equally well for
right averages.)
Given a left average and a right average of f , say
X
X
F (x) =
i f (gi1 x), H(x) =
j f (xhj ),
we can form the joint average
X
J(x) =
i j f (gi1 xhj ).
i,j

It is easy to see that

inf F inf J sup J sup H,
sup F sup J inf J inf H.

But if now H0 = f, H1 , . . . is a succession of right averages with Hi d

then it follows that
c = d.
In particular, any two convergent sequences of successive left averages
must tend to the same limit. We can therefore set
(f ) = c.
Thus (f ) is well-defined; and it is invariant since f and gf have the
same set of averages. Finally, if f = 1 then var(f ) = 0, and f, f, f, . . .
converges to 1, so that
(1) = 1.
The invariant measure on G is unique up to a scalar multiple. In other
words, it is unique if we normalise the measure by specifying that
(1) = 1
(where 1 on the left denotes the constant function 1).
2. Prove that every simple representation of a compact abelian group is
1-dimensional and unitary.
Determine the simple representations of SO(2).
Determine also the simple representations of O(2).
Answer: Suppose is a simple representation of the compact abelian
group G in V .
Suppose g G. Let be an eigenvalue of g, and let E = E be the
corresponding eigenspace. We claim that E is stable under G. For
suppose h G. Then
e E = g(he) = h(ge) = he = he E.
Since is simple, it follows that E = V , ie gv = v for all v, or
g = I.
Since this is true for all g G, it follows that every subspace of V is
stable under G. Since is simple, this implies that dim V = 1, ie is
of degree 1.
Thus a simple representation of G is a homomorphism : G C .
We must show that
|(g)| = 1
for all g G.
If |(g)| > 1 then
|(g n )| = (|g|)n .

This is a contradiction, since im C is compact and so bounded.

On the other hand, if |(g)| < 1 then |(g 1 )| > 1. Hence |(g)| = 1
for all g, ie is unitary.
We can identify SO(2) with
U(1) = {z C : |z| = 1}.
From above, a representation of U(1) is a homomorphism
: U(1) U(1).
For each n Z the map
E(n) : z z n
defines such a homomorphism. We claim that every representation of
U(1) is of this form.
3. Determine the conjugacy classes in SU(2); and prove that this group
has just one simple representation of each dimension.
Find the character of the representation D(j) of dimensions 2j + 1
(where j = 0, 12 , 1, 23 , . . . ).
Determine the representation-ring of SU(2), ie express each product
D(i)D(j) as a sum of simple representations D(k).
Answer: We know that
(a) if U SU(2) then U has eigenvalues
ei ( R).
(b) if X, Y GL(n, k) then
X Y = X, Y have the same eigenvalues.
A fortiori, if U V SU(2) then U, V have the same eigenvalues.
We shall show that the converse of the last result is also true, that is:
U V in SU(2) if and only if U, V have the same eigenvalues ei ,
This is equivalent to proving that
 i

e
0
U U () =
,
0 ei
ie we can find V SU(2) such that
V 1 U V = U ().

To see this, let v be an ei -eigenvalue of U . Normalise v, so that v v =

1; and let w be a unit vector orthogonal to v, ie w w = 1, v w = 0.
Then the matrix
V = (vw) Mat(2, C)
is unitary; and
V


UV =

ei x
0 ei

But in a unitary matrix, the squares of the absolute values of each row
and column sum to 1. It follows that
|ei |2 + |x|2 = 1 = x = 0,
ie
V 1 U V = U ().
We only know that V U(2), not that V SU(2). However
V U(2) = | det V | = 1 = det V = ei .
Thus
V 0 = ei/2 V SU(2)
and still
(V 0 )1 U V = U ().
To summarise: Since U () U () (by interchange of coordinates),
we have show that if
C() = {U SU(2) : U has eigenvalues ei }
then the conjugacy classes in SU(2) are
C()

(0 ).

Now suppose m N, Let V (m) denote the space of homogeneous polynomials P (z, w) in z, w. Thus V (m) is a vector space over C of dimension m + 1, with basis z m , z m1 w, . . . , wm .
Suppose U SU(2). Then U acts on z, w by
 
 
 0
z
z
z
=U
.
7
w
w0
w
This action in turn defines an action of SU(2) on V (m):
P (z, w) 7 P (z 0 , w0 ).

We claim that the corresponding representation of SU(2) which we

denote by Dm/2 is simple, and that these are the only simple (finitedimensional) representations of SU(2) over C.
To prove this, let
U(1) SU(2)
be the subgroup formed by the diagonal matrices U (). The action of
SU(2) on z, w restricts to the action
(z, w) 7 (ei z, ei w)
of U(1). Thus in the action of U(1) on V (m),
z mr wr 7 e(m2r)i z mr wr ,
It follows that the restriction of Dm/1 to U (1) is the representation
Dm/2 |U(1) = E(m) + E(m 2) + + E(m)
where E(m) is the representation
ei 7 emi
of U(1).
In particular, the character of Dm/2 is given by
m/2 (U ) = emi + e(m2 i + + emi
if U has eigenvalues ei .
Now suppose Dm/2 is not simple, say
Dm/2 = + .
(We know that Dm/2 is semisimple, since SU(2) is compact.) Let a
corresponding split of the representation space be
V (m) = W1 W2 .
Since the simple parts of Dm/2 |U(1) are distinct, the expression of
V (m) as a direct sum of U(1)-spaces,
V (m) = hz m i hz m1 wi hwm i
is unique. It follows that W1 must be the direct sum of some of these
spaces, and W2 the direct sum of the others. In particular z m W1 or
z n W2 , say z m W1 . Let


1 1 1
U=
SU(2).
2 1 1

Then

 


1
z
z+w
7
w
2 z + w

under U . Hence
z m 7 2m/2 (z + w)m .
Since this contains non-zero components in each subspace hz mr wr i, it
follows that
W1 = V (m),
ie the representation Dm/2 of SU(2) in V (m) is simple.
To see that every simple (finite-dimensional) representation of SU(2) is
of this form, suppose is such a representation. Consider its restriction
to U(1). Suppose
|U(1) = er E(r)+er1 E(r1)+ +er E(r)

(er , er1 , . . . , er N).

Then has character

(U ) = () = er eri + er1 e(r1)i + + er eri
if U has eigenvalues ei .
Since U () U () it follows that
() = (),
and so
ei = ei ,
ie
() = er (eri + eri ) + er1 (e(r1)i + e(r1)i ) + .
It is easy to see that this is expressible as a sum of the j () with integer
(possibly negative) coefficients:
() = a0 0 () + a1/2 1/2 () + + as s ()

(a0 , a1/2 , . . . , as Z).

Using the intertwining number,

I(, ) = a20 + a21/2 + + a2s
(since I(Dj , Dk ) = 0). Since is simple,
I(, ) = 1.

It follows that one of the coefficients aj is 1 and the rest are 0, ie

() = j ()
for some half-integer j. But
() = j () = I(, Dj ) = I(Dj , Dj ) = 1,
which is impossible. Hence
() = j (),
and so (since a representation is determined up to equivalence by its
character)
= Dj .
Finally, we show that
Dj Dk = Dj+k + Dj+k1 + + D|jk| .
It is sufficient to prove the corresponding result for the characters
j ()k () = j+k () + j+k1 () + + |jk| ().
We may suppose that j k. We prove the result by induction on k.
If k = 0 the result is trivial, since 0 () = 1. If k = 1/2 then




j ()1/2 () = e2ji + e2(j1)i + e2ji ei + ei




= e(2j+1)i + e(2j1)i + e(2j1)i + e(2j+1)i
= j+1/2 () + j1/2 (),
as required.
Suppose k 1. Then
k () = k1 () + (eki + eki ).
Thus applying our inductive hypothesis,
j ()k () = j+k1 () + + jk+1 + j ()(eki + eki ).
But
j ()(eki + eki ) = e2ji + e2(j1)i + e2ji
= j+k () + j k(),

eki + eki

giving the required result

j ()k () = j+k1 () + + jk+1 + j+k () + j k()
= j+k () + + jk .

4. Show that there exists a surjective homomorphism

: SU(2) SO(3)
with finite kernel.
Hence or otherwise determine all simple representations of SO(3).
Determine also all simple representations of O(3).
Answer: The set of skew-hermitian 2 2 matrices


ia
b + ic
S=
(a, b, c, d R)
b + ic
id
forms a 4-dimensional real vector space U . The group SU(2) acts on
this space by
(U, S) 7 U 1 SU = U SU,
since
(U SU ) = U S U = U SU.
The 3-dimensional subspace W U formed by trace-free skew-hermitian
matrices


ix
y + iz
T =
(x, y, z R)
y + iz
ix
is stable under SU(2) since
tr(U T U ) = tr(U 1 T U ) = tr T = 0.
Thus W carries a representation of SU(2) of degree 3, corresponding
to a homomorphism
: SU(2) GL(3, R).
Moreover, this homomorphism preserves the positive-definite quadratic
form
det T = x2 + y 2 + z 2
on W since
det(U T U ) = det(U 1 T U ) = det T.
Hence
im O(3).
Finally, SU(2)
= S 3 is connected; and so therefore is its image. But
SO(3) is an open subgroup of O(3). Hence
im SO(3).

Thus our homomorphism takes the form

: SU(2) SO(3).
It remains to show that has a finite kernel, and is surjective.
If
U ker
then
U 1 T U = T
for all T W . Each S U can be expressed in the form
S = T + I,
where T W and = tr S/2. It follows that
U 1 SU = S
for all skew-hermitian S U .
Hence
U 1 HU = H
for all hermitian H, since H is hermitian if and only if S = iH is
skew-hermitian.
It follows from this that
U 1 XU = X
for all X Mat(2, C), since every X is expressible in the form
X = H + S,
with H = (X + X )/2 hermitian and S = (X X )/2 skew-hermitian.
But it is a simple matter to see that the only such U are U = I. Thus
ker = {I}.
To see that is surjective, we note that if
 i

e
0
U () =
0 ei
then
1

U ()




ix
y + iz
ix
e2i (y + iz)
U () = 2i
,
y + iz
ix
e (y + iz)
ix

ie

1
0
0
U () = 0 cos 2 sin 2 = R(Ox, 2),
0 sin 2 cos 2

rotation about Ox through angle 2. In particular, im contains all

rotations about Ox.
Now let

1
U=
2


1 1
.
1 1

Then U SU(2) and





ix
y + iz
iz
y ix
1
U
U=
.
y + iz
ix
y ix
iz
Thus

0 0 1
(U ) = 0 1 0 = R(Oy, /2).
1 0 0

Writing P = R(Oy, /2),

P 1 R(Ox, )P = R(Oz, ).
Thus im contains all rotations about Oz as well as Ox. But is is
easy to see that every rotation R SO(3) is expressible as a product of
rotations about Ox and Oz. Hence
im = SO(3),
ie is surjective.
Thus
SO(3) = SU(2)/{i}.
It follows that the representations of SO(3) are just the representations
of SO(2) such that
(I) = I.
In particular, the simple representations of SO(3) are those simple representations Dj of SU(2) such that Dj (I) = I. But Dj is defined by
the action of SU(2) on the polynomials
P (z, w) = c0 z 2j + c1 z 2j1 w + + c2j w2j .
It is clear that
P (z, w) = P (z, w)
for all P of degree 2j if and only if 2j is even, ie j is an integer.

Thus the simple representations of SO(3) are D0 , D1 , D2 , . . . of degrees

1, 3, 5, . . . .
Since
O(3) = SO(3) C2 ,
where C2 = {I}, the simple representations of O(3) are of the form
, where is a simple representation of SO(3), and is a simple
representation of C2 . Thus the simple representations of O(3) are Dj
1 and Dj , where j N and  is the representation I 1 of C2 .
5. Explain the division of simple representations of a finite or compact
group G over C into real, essentially complex and quaternionic. Give
an example of each (justifying your answers).
Show that if is a simple representation with character then the
value of
Z
(g 2 ) dg
G

determines which of these three types falls into.

Answer: Suppose is a simple representation of G over C. Then
is said to be real if
= C
for some representation of G over R. If this is so then the character
(g) = (g)
is real. We say that is quaternionic if its character is real, but it is
not real. Finally, we say that is essentially complex if its character
is not real.
The trivial character 1 of any group is real, since it is the complexification of the trivial character over R.
The 1-dimensional character of the cyclic group C3 = hgi given by
: g 7 = e2/3
is essentially complex, since its character is not real.
Consider the quaternion group
Q8 = {1, i, j, k}.
We can regard the quaternions H as a 2-dimensional vector space over
C = h1, ii. The action of Q8 on H by multiplication on the left defines
a 2-dimensional representation of D8 . We assert that this is a simple
quaternionic representation.

It is certainly simple, since otherwise H would have a 1-dimensional

subspace hqi stable under D8 , and therefore under H, since D8 spans
H. But that is impossible since
x = (xq 1 )q
for any x H. The
representations of D4 must have dimensions
P simple
1, 1, 1, 1, 2 (since
dim2i = 8). It follows that
=
since there is only 1 2-dimensional simple representation. Hence is
real.
It remains to show that is not real. Consider the 4-dimensional representation of D8 over R, defined by the same action of D4 on H.
This is easily seen to be simple, by the argument above. It follows that
C is either simple, or splits into 2 simple representations over C of
dimension 2. The only possibility is that
C = 2.
Now if were real, say
= C
we would deduce that = 2 which is impossible, since is simple.
Now suppose is a simple representation of G in V . Then ( )2 is
the representation arising from the action of G on the space of bilinear
forms on V .
But
= is real.
Thus
(
1 if is real or quaternionic
I(1, ( )2 ) = I(, ) =
.
0 if is essentially complex
In other words, there is just 1 invariant bilinear form (up to a scalar
multiple) if is real or quaternionic, and no such form if is essentially complex.
Now the space of bilinear forms splits into the direct sum of symmetric (or quadratic) and skew-symmetric forms, since each bilinear form
B(u, v) can be expressed as
B(u, v) =

1
1
(B(u, v) + B(v, u)) + (B(u, v) B(v, u)) ,
2
2

where the first form is symmetric and the second skew-symmetric.

It follows that
( )2 = + ,
where is the representation of G in the space of symmetric forms,
and the representation in the space of skew-symmetric forms.
If is essentially complex, there is no invariant symmetric or skewsymmetric form. But if is real or quaternionic, there must be just 1
invariant form, either symmetric or skew-symmetric. We shall see that
in fact there is an invariant symmetric form if and only if is real.
Certainly if is real, say = C , where is a representation in the
real vector space U , then we know that there is an invariant positivedefinite form on U , and this will give an invariant quadratic form on
V = UC .
Conversely, suppose is a quaternionic simple representation on V .
Then = R is simple. For
(R )C = +
for any representation over C. Thus if = + 0 then (with
quaternionic)
2 = C + C0 ,
and it will follow that
= C = C0 ,
so that is real.
Since is simple, there is a unique invariant quadratic form P on
VR , and this form is positive-definite. But if there were an invariant
quadratic form Q on V this would give an invariant quadratic form on
VR , which would not be positive-definite, since we would have
Q(iu, iu) = Q(u).
Thus if is quaternionic, then there is no invariant quadratic form on
V , and therefore there is an invariant skew-symmetric form.
It follows that we can determine which class falls into by computing
I(1, ) and I(1, ).
To this end we compute the characters of and .
Suppose g G. Then we can diagonalise g, ie we can find a basis
e1 , . . . , en of V consisting of eivenvectors, say
gei = i ei .
The space of quadratic forms is spanned by the n(n + 1)/2 forms
xi xj

(i j),

where x1 , . . . , xn are the coordinates with respect to the basis e1 , . . . , en .

It follows that
X
X
(g) =
2i +
i j .
i<j

Now
(g) =

i , (g 2 ) =

2i .

It follows that
p hi(g) =

1
(g)2 + a lpha(g 2 ) .
2

We deduce from this that

I(1, ) =

1 X
(g)2 + (g 2 ) .
2kGk gG

Since
I(1, ) + I(1, ) = I(1, ( )2 )
1 X
=
(g 1 )2
kGk g
1 X
=
(g)2 ,
kGk g
it follows that
I(1, ) =

1 X
(g)2 (g 2 ) .
2kGk gG

Putting all this together, we conclude that

1 X
(g 2 ) = I(1, ) I(1, )
kGk gG

if is real,
1
= 1 if is quaternionic,

0
if is essentially complex.