28. Last Latexed: April 25, 2017 at 9:45 Joel A.

Shapiro

by the linear independence of the ej ’s. Note the order of the indices now.
Tij can be considered the matrix element Tij (A) = hei | A |ej i.
We can find the composition law for these matrices by noting that
X X  X
ABei = ek Tki (AB) = A(Bei ) = A ej Tji (B) = Tji (B)Aej
Chapter 2 =
X
k

Tji(B)
X
Tkj (A)ek ,
j j

j k

Representations so Tki (AB) =

X
Tkj (A)Tji (B),
j

or T (AB) = T (A)T (B),

In the last chapter we learned something about the structure of the groups where the multiplication is just ordinary matrix multiplication. Thus we
themselves, but we are really interested in how the symmetry groups act on have a homomorphism from the group G onto a group of ℓ × ℓ nonsingular
the eigenstates of a physical Hamiltonian. The set of states corresponding to matrices. This is called an ℓ-dimensional representation of the group G.
a given eigenvalue of degeneracy ℓ forms a vector space of dimension ℓ. The If no two elements of G are mapped into the same matrix, then the
result of applying the symmetry operation to any of these states gives a state homomorphism G → T = {T (A)|A ∈ G} is an isomorphism, G ∼ = T , and we
with the same energy eigenvalue, and thus a state in the same ℓ-dimensional call the representation faithful. An example of an unfaithful representation
vector space. The symmetry acts linearly on the states, for any group is the homomorphism g ∈ G 7→ 1I, in which every element
is mapped into the identity matrix. This trivial representation is called the
S(α |ψ1 i + β |ψ2 i) = αS(|ψ1 i) + βS(|ψ2 i), identity representation of the group.
Note that for any representation
so the operators are linear operators on a vector space, and may be considered
ℓ × ℓ matrices acting on this eigenspace. T (1I)T (A) = T (1IA) = T (A), so T (1I) = 1Iℓ×ℓ ,
Let {ei , i = 1, ..., ℓ} be an orthonormal basis of the vector space. If A ∈ G
is one of the symmetry operators, it will act on ei to give a linear combination as T (A) is nonsingular. Also
of the basis vectors, X T (A−1 )T (A) = T (A−1 A) = T (1I) = 1I, so T (A−1 ) = [T (A)]−1 .
Aei = ej Tji (A).
j If R is an ℓ-dimensional representation of G, and if S is a fixed nonsingular
Notice the order of the indices
P ℓ × ℓ matrix, then
P on T . If ψ = ψi ei is an arbitrary vector in
the eigenspace, and if ψ ′ = ψi′ ei = Aψ, then T (A) := S −1 R(A)S for all A ∈ G,
X X X X
ψ′ = ψi′ ei = A ψi ei = ψi Aei = ψi Tji(A)ej , is also clearly an ℓ-dimensional representations of G. Two representations
ij related by such a fixed similarity transformation are equivalent represen-
or tations.
X
ψj′ = Tjk (A)ψk , Consider an n-dimensional representation T of G on a vector space Ln .
k
Suppose we take a particular vector v and act on it with all the elements T (A)

27
618: Last Latexed: April 25, 2017 at 9:45 29 30. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

and then consider the vector space Lm spanned by these vectors, i.e. the H is clearly hermitian so it can be diagonalized by a unitary U, Hd = U −1 HU,
space of all linear combinations of {T (A)v|A ∈ G}. Surely Lm is contained (Hd )ij = di δij . So
in Ln , but it might be the whole of Ln or it might be a proper subspace1 , XX


of dimension m < n. Clearly Lm is closed under the action di = (Hd )ii = U −1 T (A) ik T † (A)U ki
P of the group, as
for any
P A ∈ G, the action
P of A on an arbitrary
P vector aB T (B)v in Ln is XX
A k

2
A: aB T (B)v 7→ aB T (A)T (B)v = aB T (AB)v, which is included in = †
T (A)U > 0,
ki
the definition of Lm . If we choose a new basis for Ln such that the first m A k
basis vectors span the subspace Lm , we see that
m as U −1 = U † because
 U is unitary, and both T and U are nonsingular. Thus
X 1/2 1/2 1/2
T (A)ei = ej Tji (A) for i ≤ m we can define Hd = di δij and V = UHd will provide the similarity
ij
j=1
transformation from T to an equivalent unitary representation Γ. Let
or Tji (A) = 0 for all i ≤ m, j > m, A ∈ G. So T has the form
−1/2 1/2
 (1)  Γ(A) = V −1 T (A)V = Hd U −1 T (A)UHd .
D (A) X(A)
T (A) = (2.1)
0 D (2) (A) Then
with D (1) an m × m matrix, D (2) an (n − m) × (n − m) matrix, and X an Γ(A)Γ† (A) = Hd
−1/2 1/2
U −1 T (A) UHd Hd U −1 T † (A)UHd
1/2 −1/2
m × (n − m) matrix. | P {z }
H= T (B)T † (B)
If we check how two such matrices multiply we can easily show that X
B
−1/2 −1/2
D (1) (AB) = D (1) (A)D (1) (B) and D (2) (AB) = D (2) (A)D (2) (B), so D (1) and = Hd U −1 T (A)T (B)T †(B)T † (A) UHd
D (2) are m and n − m dimensional representations respectively, and our
|B {z }
original representation T is said to be a reducible representation. The H
subspace Lm is called an invariant subspace. If there is no proper sub- = Hd
−1/2
Hd Hd
−1/2
= 1I,
space of Ln which is closed under action by the group, we say that T is an
irreducible representation. where in the first line we used that Hd P
is hermitean and U is Punitary, and in
We expect symmetry transformations on our hilbert space to be unitary the second that the underbraced sum is B T (AB)T † (AB) = B T (B)T † (B) =
transformations, preserving the norm of the states. Then because we asked H, where the first equality is due to the left invariance of sums over all the
that ei be an orthonormal basis, the matrices T (A) will be unitary matrices, elements in the group.
and the representation is called a unitary representation. We will now This is the first example we have seen, but far from the last, of the
prove that any finite dimensional representation of a finite group is equivalent usefulness of summing over all the elements of a group with equal weight.
to a unitary representation. This will also be true for infinite groups if we can For a Lie group, one with elements indexed by a continuous parameter, we
define finite invariant integration measures, but this we will consider later. will need carefully to give meaning to a sum with equal weights over all
Consider a representation T of a finite group G. Define the matrix elements. We will need to define an invariant integration measure. We will
X then find that for compact Lie groups, for which the total group volume in
H= T (A)T † (A).
A∈G
finite, we can again find a similarity transformation, so that we will have
proven:
1
A subspace of a vector space is a subset closed under addition and scalar multiplication.
A proper subspace is a subspace which is not the whole space and not just the point Theorem:Any finite dimensional representation of a finite group or of a
{0}. compact Lie group is equivalent to a unitary representation.
618: Last Latexed: April 25, 2017 at 9:45 31 32. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

If T was reducible, so is Γ, as there is still an invariant proper subspace. we can also transform our representation Γ by U, defining a new represention
The necessary change of basis to get Γ into the form of eq. 2.1 can be chosen Γ′ (A) = U −1 Γ(A)U, which is still irreducible and unitary, and [Γ′ (A), D] =
unitary, so the new form is also unitary. But then U −1 [Γ(A), H]U = 0. Thus
!

†
D (1) (A−1 ) 0 (Γ′ (A)D)ij = Γ′ (A)ij dj = (DΓ′ (A))ij = Γ′ (A)ij di ⇒ Γ′ (A)ij (dj − di) = 0.
−1 †
Γ(A) = Γ(A ) = † .
X † (A−1 ) D (2) (A−1 ) Now either all the di are equal, in which case D is a multiple of the identity,
or Γ′ (A)ij = 0 for all A, except for i and j in the same block, that is for i
But we know that the lower left block of Γ is zero, so X = 0, and the and j having the same d. But that would mean Γ′ (A) is reducible, and so is
representation is block diagonal, Γ(A), which contradicts the hypothesis. This argument works for both H+
 (1)  and H− , and tells us that both must be multiples of the identity, as must be
D 0
Γ(A) = . M = H+ + iH− .
0 D (2)
This lemma has a useful contrapositive: if a matrix M which is not
A representation which can be successively reduced in this manner to a multiple of the identity matrix commutes with all the representatives of
a block diagonal form of two or more irreducible representations is said to elements in the group, the representation must be reducible.
by fully reducible. We say that Γ is the direct sum of the irreducible As an example of reducibility and the use of Schur’s Lemma, consider
representations. the three dimensional rotation group. This is a continuous (infinite) group.
We will find physics examples after we prove The elements may be parameterized by a 3-vector ~ω , and the group elements
written as
~
g(~ω) = ei~ω ·L ,
2.1 Schur’s First Lemma where Lj are the angular momentum operators2 . [As we shall define later,
these are generators of a Lie algebra, not generators of the group in the
Any matrix which commutes with all the matrices of an irreducible represen-
sense we have used so far, e.g. in D4 = hC, mx i.] As we know from P quantum
tation of a finite or compact Lie group must be a multiple of the identity.
mechanics, the Lj do not commute with each other, but L = j L2j does
2

We can assume we have made the similarity transformation so that our commute with all the Lj , and therefore with all the group elements g ∈
representives Γ(A) are unitary. Whether or not the commuting M is a mul- SO(3).
tiple of the identity will not be affected by this similarity transformation. Thus L2 acts as a constant value on any irreducible representation of the
Then if [M, Γ(A)] = 0 for all A ∈ G, rotation group. That is, all the states in an irreducible representation are
eigenstates of L2 with eigenvalue3 ℓ(ℓ + 1), ℓ = 0, 1, 2, ....
Γ(A)M = MΓ(A), and also M † Γ† (A) = Γ† (A)M † . Now consider a 3-D isotropic harmonic oscillator with H = p~ 2 /2m+ 21 k~r 2 .
The motion in the three cartesian coordinates decouple, so the state of the
We multiply the last equation by Γ(A) on the left and on the right, using
2
the fact that Γ(A) is unitary, we get We are taking
P ~ = 1 here if we wish to associate Lj with the quantum mechanical
operator −i~ kq ǫjkq rk ∂q .
3
Γ(A)M † = M † Γ(A), Systems with fermions can lie in ray representation rather than a true representation
of SO(3). Then the value of ℓ can be half an odd integer. This possibility arises because
the phase of the wavefunction is unphysical, and it is not required that every rotation
so M † also commutes with all Γ(A), and so do the hermitean matrices H+ = which leaves space unchanged, such as rotation by 2π, leave the wavefunction unchanged,
M + M † and H− = i(M † − M). Now either of these hermitean matrices can but only that it leaves |ψ 2 | unchaged. This will be discussed later, when we discuss the
be diagonalized by a unitary matrix U, H = UDU −1 , where Dij = di δij . But simple-connectedness of Lie groups.
618: Last Latexed: April 25, 2017 at 9:45 33 34. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

system can be described by the product of states of one dimensional har- 2.2 Schur’s Second Lemma
monic oscillators, specified by the P excitation levels nj , j = 1, 2, 3. Thus
the energy is given by E = ~ω( j nj + 3/2). The ground state (0, 0, 0) If Γi and Γj are irreducible unitary representations of G, of dimensions ℓi
is nondegenerate, while the first excited state is triply degenerate, with oc- and ℓj respectively, and if
cupation numbers (1, 0, 0), (0, 1, 0), and (0, 0, 1). These form an ℓ = 1 irre-
Γi (A)M = MΓj (A) for all A ∈ G,
ducible representation of the rotation group. The third level has six states
|(2, 0, 0)i , |(0, 2, 0))i , |(0, 0, 2)i , |(1, 1, 0)i , |(1, 0, 1)i and |(0, 1, 1)i. This can- then either M = 0 or Γi ∼
= Γj and M is nonsingular.
not be an irreducible representation. In terms of the raising and lowering op- Proof: Take hermitian conjugate:
erators a† and aj for each direction, the number operator N = nx + ny + nz =
P † j P †
† †
M † Γi (A) = Γj (A)M † .
†
right multiply by M, use Γj (A) = Γj (A−1 )
j aj aj and the angular momentum generators are Li = −i jk ǫijk aj ak , so
M † Γi (A−1 )M = Γj (A−1 )M † M for all A−1 ∈ G, so
| {z }
XX X X X M Γj (A−1 )
L2 = − ǫijk a†j ak ǫiℓm a†ℓ am = − a†j ak a†j ak + a†j ak a†k aj
i jk ℓm jk jk
M † M = λ1I
X 
= −a†j (a†j ak + δjk )ak + a†j (a†k ak + 1)aj by Schur’s first lemma,
P and the fact
P that Γj is irreducible. If λ = 0,
2
jk Tr M † M = 0 = †
ij (M )ij Mji = ij |Mji | , so M = 0. If ℓi = ℓj and
X 2 X  X † λ 6= 0, M is nonsingular, and provides the similarity transformation that
= − a†j a2k − N + aj (N + 3)aj
proves Γi ∼
= Γj . If ℓi 6= ℓj and λ 6= 0, then ℓi > ℓj , because M † cannot be an
j k j
X 2 X  X † onto map from a smaller space to a larger one. But we could have done our
= − a†j a2k + 2N + aj Naj argument slightly differently, premultiplying by M instead of postmultiplying
|{z}
j k j
aj (N −1)
X 2 X  MM † Γi (A−1 ) = MΓj (A−1 )M † = Γi (A−1 )MM †
= − a†j a2k + N 2 + N .
for all A−1 ∈ G, so we also have MM † = λ′ 1I. Then either λ′ = 0 (and as
j k
MM † M = Mλ = λ′ M, so is λ), or MM † has an image which is all of the
i-dimensional space, which implies the j-dimensional space on which M acts
Now N = nx + ny + nz is 2 on all the six states under discussion. In the
is at least of dimension i, in contradiction with what we showed first.
last expression, the first term vanishes on the |(1, 1, 0)i state and its cousins,
and gives −2(|(2, 0, 0)i + |(0, 2, 0)i + |(0, 0, 2)i) when acting on any of the
states |(2, 0, 0)i, |(0, 2, 0)i, or |(0, 0, 2)i. Thus on the state proportional to 2.3 The Great Orthogonality Theorem
(|(2, 0, 0)i + |(0, 2, 0)i + |(0, 0, 2)i) it acts as −6, cancelling the N 2 + N con-
tribution, while on the two states orthogonal to this it gives zero. So we Let Γi and Γj be irreducible unitary representations of a group G, of finite
see that L2 is not a constant on the space of twice-excited states, but divides dimensions ℓi and ℓj respectively. Let X be a fixed ℓi × ℓj matrix, and
it into two subspaces, a singlet with L2 = 0 and a five dimensional space X
with L2 = 6. That is, the second excited state of the isotropic oscillator is M= Γi (A)XΓj (A−1 ).
A∈G
decomposed into an ℓ = 0 irreducible representation and an ℓ = 2 irreducible
R R
representation of the rotation group. [If G is a continuous group withP
an invariant measure, dµ f (BA) = dµ f (A)
R
It is also a very interesting idea to apply this kind of consideration to the for any B and f , replace the A∈G by dµ and the rest of this argument
energy eigenstates of the nonrelativistic hydrogen atom. will go through OK.]
618: Last Latexed: April 25, 2017 at 9:45 35 36. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

Then for any B, In what sense is this orthogonality? Think of the space of all complex-
X valued functions of the
P group elements. Define a norm on this space of
Γi (B)M = Γi (B)Γi (A)XΓj (A−1 )Γj (B −1 )Γj (B) functions by (f, h) = A∈G f ∗
(A) h(A), or for a compact Lie group, (f, h) =
R ∗
A
X dµf (A(µ)) h(A(µ)). Then the great orthogonality theorem tells us that
= Γi (BA)XΓj ((BA)−1 )Γj (B) s
A ℓi i
Γ (A)
= MΓj (B). g ab
form an orthonormal set of vectors in the vector space of all functions on the
Thus by Schur’s second lemma, if Γi is not equivalent to Γj , M = 0 for any
group.
X. In particular, take any b ≤ ℓi and d ≤ ℓj , and let Xef := δeb δf d . Then
If the group G = {A1 , ...Ag } has a finite number of elements, one or-
X thonormal basis for functions on G is the set eα of functions eα (Aβ ) = δαβ ,
Mac = Γiab (A)Γjdc (A−1 ) must vanish
A∈G
for α and β = 1, ..., g. This set clearly spans the space of all functions, as for
X any ψ, X X
so 0 = Γiab (A)Γj∗
cd (A) (for Γi ∼
6= Γj ), (2.2) ψ= ψ(Aα )eα , that is ψ(B) = ψ(Aα )eα (B)
A∈G α α
j
as Γ is a unitary representation. and is linearly independent. Thus the space of functions LG on G is g-
On the other hand, if Γi and Γj are the same representation, then Schur’s dimensional. The orthonormal
P set we got from theP great orthogonality the-
first lemma tells us Mac = cδac . Tracing, orem has dimension i ℓ2i , so it must be true that i ℓ2i ≤ g. We will soon
show it is an equality.
XX X
c ℓi = Γiab (A)Γida (A−1 ) = Γidb (1I) = gδbd , In general, if we have a set of functions ψj (A) on the group, we can define
A∈G a A∈G the operation of the group on these functions by

where g is the order of G, so c = gδbd /ℓi , and (Bψj ) (A) := ψj (AB).

X g The order is important, in order to preserve the group multiplication law,
Γiab (A)Γi∗
cd (A) = δac δbd . (2.3) (CB)ψ = C(Bψ). To show this let both functions act on A.
A∈G
ℓi
((CB)ψ)(A) = ψ(ACB), (C(Bψ))(A) = (Bψ)(AC) = ψ(ACB) good!.
If we understand the index i to run over all inequivalent irreducible repre-
If you had tried to define (Bψ)(A) := ψ(BA), (wrong!), you would not find
sentations, we can write both results (2.2) and (2.3) as
(CB)ψ = C(Bψ).
X g Now the function Bψ is a function of the group elements. Although it is
Γiab (A)Γj∗
cd (A) = δij δac δbd . (2.4)
A∈G
ℓi not the function ψ, it is certainly in the space of all functions on the group.
Thus the space LG of all functions on the group provides a representation of
This is known as the great orthogonality
R theorem. the group, g dimensional. Let us use the basis eα . Then the function Beα
[Note: for the continuum version, if dµ 1 = V , can be Pexpanded with coefficients we call Γreg βα (B) of the basis vectors eβ ,
Z Beα = β Γreg βα (B)eβ . Acting on A γ , we have
V
dµΓiab (A)Γj∗
cd (A) = δij δac δbd .] X reg
ℓi Beα (Aγ ) = Γβα (B) eβ (Aγ ) = eα (Aγ B) = δAγ B, Aα .
| {z }
β
δβγ
618: Last Latexed: April 25, 2017 at 9:45 37 38. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

So we see Multiply this by Γj∗

cd (A) and sum over A ∈ G,
Γreg
γα (B) = δAγ B, Aα . X (i)
X g (j)
This representation is called the regular representation of the group. Γj∗
cd (B) = aab (B) Γiab (A)Γj∗
cd (A) = a (B),
ℓj cd
Included in this space of all functions on G is a set of functions iab
|A {z }
g
ψb (A) = Γiab (A), δij δac δbd
ℓi
where i is any of the irreducible representations of G and a is a fixed index
a ∈ [1, ℓi ]. This set of ℓi functions transform by using the Great Orthogonality Theorem. When we insert this back into (2.5),
X X we have
X ℓi
(Bψb )(A) = ψb (AB) = Γiab (AB) = Γiac (A)Γicb (B) = ψc (A)Γicb (B), δAB = Γiab (A)Γi∗
ab (B), (2.6)
c c
iab
g
P
so Bψb = c ψc Γicb (B), and this set transforms like the ith irreducible repre- which is
sentation. Psort of orthonormality of the g vectors Γ(A) (A is an index here)
in theP ℓ2i dimensional space indexed by (i, a, b). In particular, g vectors
Actually we see that we have ℓi such sets of functions, one set for each in a P i ℓ2i dimensional space can be independent, which they are, only if
a, and they are guaranteed linearly independent by the Great Orthogonality g ≤ i ℓ2i . Perhaps it would be clearer to say this: Using (2.6), any function
Theorem. Thus we see that each irreducible representation Γi appears at on G
least ℓi times
P 2 in the reduction of the regular representation. This again tells
us that X  ℓi 
P 2 i ℓi ≤ g. Very soon, we shall see that it appears exactly ℓi times
X
i∗ i
f (A) = δAB f (B) = Γ (B)Γab (A) f (B)
and i ℓi = g.
B
g ab
iab
Any function on the group is a linear combination of the eα which pro- X  ℓi 
vide the basis of the regular representation. In particular, the function = Γab (B)f (B) Γiab (A),
i∗

fB : G → C given by fB (A) = δA,B is just the (1I, B) element in the regular iab
g
representation Γreg as we have defined it. But the regular representation is P 2
reducible, so it is equivalent to adirect sum is a linear combination of the ℓi functions Γiab (A), so these are a complete
  of irreducible representations: P 2
Γ reg
= S −1
L i
L i set of functions on G, and that space is g-dimensional. So again g ≤ i ℓi .
i,ni Γ S, where i,ni Γ is essentially a block diagonal P 2
But we have already shown g ≥ i ℓi , so we have an important statement
matrix with ni blocks for each irreducible representation Γi , each block an about the full set of irreducible representations and their dimensions:
ℓi × ℓi matrix function on the group. This direct sum matrix thus has two
compound indices, each one a combination of i giving the irreducible repre- X
g= ℓ2i . (2.7)
sentation, n = 1, . . . , ni telling which one of the ni copies, and a or b giving i
which basis vector within that representation. Thus S = S(i,n,b),β and
X

Γreg
αβ = S −1 α,(i,n,a) Γiab S(i,n,b),β , 2.4 Characters
i,n,a,b
P We have seen that two representations are considered equivalent if they are
and if we define aiab (B) = n (S −1 )1I,(i,n,a) S(i,n,b),B we have
related by a similiarity transformation Γ(1) (A) = UΓ(2) (A)U −1 for all A ∈ G.
X (i)
aab (B)Γiab (A) = Γreg To characterize a representation we are not interested in all the equivalent
1I,B (A) = δA,B . (2.5)
iab
forms, so we would like something invariant under similiarity.
618: Last Latexed: April 25, 2017 at 9:45 39 40. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

This is done by defining the character of a representation Γ to be a That is, we know in principle that Γ is reducible to a direct sum of irreducible
function χ : G → C on the group given by representations, but we haven’t done it. Now we will see how to extract the
ai ’s.
χ(A) = Tr Γ(A). The representation Γ has a character χ, so for each possible irreducible
representation Γj which might be included in the direct sum, we can form
Note χ is a complex-valued function on the group, not a matrix-valued func- X X X X
tion as Γ is. Also note it is obviously unaffected by similiarity transforma- χj∗ (A)χ(A) = ai χj∗ (A)χi (A) = ai gδij = gaj , (2.8)
tions. But this has an important consequence on the function χ — namely, A∈G i A∈G i
if A and B are conjugate group elements, i.e. A = C −1 BC for some C ∈ G,
so
then 1 X j∗
aj = χ (A)χ(A). (2.9)

g A∈G
χ(A) = Tr Γ(A) = Tr Γ(C −1 BC) = Tr Γ−1 (C)Γ(B)Γ(C) = Tr Γ(B)
= χ(B). If Γ is in fact irreducible, one a will come out one and the others zero. For
the one j which gives aj = 1, χj = χ, so
Thus the character is actually only a function on the conjugacy classes of the X
group. χ∗ (A)χ(A) = g if χ is irreducible.
Now the great orthogonality theorem (2.4) A∈G

X g In fact, this is an “if and only if” statement, for

Γiab (A)Γj∗
cd (A) = δij δac δbd
ℓi X X X X X
A∈G χ∗ (A)χ(A) = ai aj χi∗ (A)χj (A) = g ai aj δij = g a2i .
A∈G ij A∈G ij
can be traced in ab and in cd, to give
X g X If this is g, exactly one ai is one and the others must be zero.
χi (A)χj∗ (A) = δij δac δac = gδij . Let us return to the regular representation, which is a g × g matrix rep-
ℓ i
A∈G
|ac {z } resentation
ℓi Γreg
αβ (B) = δAα B, Aβ .

Thus two different representations have different characters—in fact they The diagonal elements are Γreg
αα (B) = δAα B, Aα = δB,1I , so
form a set of linearly independent functions on the conjugacy classes of the g
X
group. Thus the number of inequivalent irreducible representations of G χ reg
(1I) = (1I)αα = g, χreg (B) = 0 for B 6= 1I.
must be ≤ the dimension of the space4 of complex functions on the classes α=1
of G, which is just the number of classes of G.
For each irreducible representation Γi of G, χi (1I) = ℓi , so the number of
L Suppose i
we have a representation which might be reducible. Then Γ =
times that representation occurs in the regular representation is
i ai Γ , which means
M M M M 1X i 1
Γi1 Γi1 ... Γi1 Γi2 ... ai = χ (A)χreg (A) = χi (1I)χreg (1I) = ℓi ,
g A g
| {z }
ai1 times P 2
as we already realized from establishing
P 2 that our original inequality g ≥ ℓi
4
Over the field of complex numbers was in fact saturated, g = ℓi . This equality now reflects just that the
618: Last Latexed: April 25, 2017 at 9:45 41 42. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

dimension of the regular representation is the sum of the dimensions of its The equation (2.8) can be thought of as an orthogonality statement for
irreducible components: the characters. As the character of each element in a conjugacy class is the
X X same, if we define ηC to be the number of elements in the class C, we have
dim Γreg = g = ai dim Γi = ℓ2i . X
i i ηCk χi (Ck )χj∗ (Ck ) = gδij .
k
The characters also form a basis for functions on the conjugacy classes.
The fact that any function on classes can be expressed as a linear combination
Let {Ck } be the conjugacy classes5 of G, and
of characters enables us to write
f : {Ck } → C X
δkk′ = ai (Ck′ )χi (Ck )
be an arbitrary function on the classes. Then f˜ : G → C defined by f˜(A) = i

f (C(A)), where C(A) is the conjugacy class containing A, is a map from the Multiply by ηCk χj∗ (Ck ) and sum over k,
group into the complex Pnumbers. As any such function is a linear combination X X
of the Γiab ’s, f˜(A) = iab ciba Γiab (A) for some set of coefficients ciba . Because6 ηCk′ χj∗ (Ck′ ) = ai (Ck′ ) ηCk χi (Ck )χj∗ (Ck ) = gaj (Ck′ ),
f˜(B −1 AB) = f(A), ˜ we have i k

X X X ηCk′ j∗
ciba Γiab (A) = ciba Γiab (B −1 AB) = ciba Γiar (B −1 )Γirs (A)Γisb (B). so aj (Ck′ ) = χ (Ck′ ), and
g
iab iab iabrs
X ηC
But the Γiab are a set of linearly independent functions, so δkk′ = k′
χi∗ (Ck′ )χi (Ck ).
i
g
X
cisr = ciba Γiar (B −1 )Γisb (B), This tells us r
ab ηC i
χ (C)
g
for any B, or in matrix form
is a complete orthonormal basis on the set of conjugacy classes.
ci = Γi (B)ci Γi (B −1 ), or ci Γi (B) = Γi (B)ci .

But then Schur’s first lemma tells us ci ∝ 1I, or ciba = k i δab , and 2.5 Examples
X X X
f˜(A) = ki Γiaa (A) = k i χi (A), Now we will use these results to find the irreducible representations, first
i a i finding their characters, of the group D4 . P
P
or f = i k i χi is a linear combination of the characters, and consequently the D4 has 5 classes and 8 elements. Thus 51 ℓ2i = 8, with each ℓi a positive
number of irreducible representations is the same as the number of conjugacy integer. Except for ordering, the only solution is ℓ1 = ℓ2 = ℓ3 = ℓ4 = 1, ℓ5 =
classes. 2, so there are four one-dimensional representations and one two-dimensional
one.
5
Here I am using k as an index that runs over the conjugacy classes, not the rep- Let us make a table. We take the first representation to be the identity
resentations. We shall show that there are the same number of classes and irreducible
representations, but there is no natural isomorphism between classes and irreducible rep- representation, A 7→ 1 for all A.
resentations. (a) So χ1 (C) = 1 for all C.
6
By the cyclic invariance of a trace, Tr(M N ) = Tr(N M ) with N = B and M = B −1 A. (b) For all five representations, χi ({1I)}) = ℓi .
618: Last Latexed: April 25, 2017 at 9:45 43 44. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

Class σz , and we are only interested in finding the representation up to similar-

❅ ηk {1I} {C, C 3 } {C 2 } {mx , my } {σ+ , σ− }
χ ❅ ity, so we take Γ(5) (mx ) = σz . Γ(5) (my ) = Γ(5) (mx )Γ(5) (C 2 ) = −σz . Now
1 2 1 2 2 Γ(5) (mx Cmx ) = Γ(5) (C 3 ) = Γ(5) (C 2 )Γ(5) (C) = −Γ(5) (C), so Γ(5) (mx ) = σz
χ1 1a 1a 1a 1a 1a anticommutes with Γ(5) (C). Thus Γ(5) (C) = aσx + bσy , with a2 + b2 = −1
χ2 1b −1d 1c −1d 1d because Γ(5) (C 2 ) = −1I, and (a∗ σx + b∗ σy )(aσx + bσy ) = 1I as it needs to
χ3 1b −1d 1c 1d −1d be unitary. This implies |a|2 + |b|2 = 1 and a∗ b real. Then we can take
χ4 1b 1d 1c −1d −1d a = i cos θ, b = i sin θ. Using a similarity transformation with eiθσz permits
χ5 2b 0 −2 0 0 us to choose θ = −π/2, Γ(5) (C) = −iσy . As mx and C generate the group,
we need only multiply to find
(c) The one dimensional representations are the same as their characters,
so χ is a unitary 1 × 1 matrix eiφ . As (C 2 )2 = m2x = σ+ 2
= 1, and C 3 = (C)3 ,
A 1I C C2 C3 mx my σ+ σ−
we have
5
) Γ (A) 1I −iσy −1I iσy σz −σz σx −σx
2
(χi ({C 2 })) = (χi ({mx , my }))2 = (χi ({σ+ , σ− }))2 = 1
for i = 1, ..., 4
(χi ({C}))3 = (χi ({C 3 })) = (χi ({C}))
Note that this forms a group of 2 × 2 matrices, but it is not the group found
2
so the characters of C , mx and σP ± are ±1 and χi (C) = ±1 or 0, with
in the first homework problem of assignment 1, because the σx and σz terms
0 only possible for χ5 . But recall k χ1 (Ck )χj (Ck ) = 0 for j = 2, 3, 4, so have real coefficients here.
1 + χj (C 2 ) + 2(±1 ± 1 ± 1) = 0, which is impossible for χi (C 2 ) = −1. So In general finding the characters and representations may be more in-
χ2 (C 2 ) = χ3 (C 2 ) = χ4 (C 2 ) = 1. volved. The book by Joshi describes a tool based on products of classes, but
I will leave that up to you, should you ever need it.
(d) Then, again for j = 2, 3, 4, as 2 + 2 [χj (C) + χj (mx ) + χj (σ± )] = 0, we
We originally motivated representations by considering spaces of solutions
must have two of them −1 and one +1. We need three solutions and there
of the Schrödinger equation, and the group D4 by considering an electron in
are only three, so except for the order in naming χj , j = 2, 3, 4, the first four
a periodic square lattice with a single impurity. What have we learned that
lines are determined.
might help us in solving this problem? Recall that for a one dimensional
The last row must be orthogonal, with weights ηk , to each of the rows problem, parity reduced the problem from finding a general function of one
above, four linear conditions onPthe four unknowns, and is determined easily. variable to finding either a symmetric or an antisymmetric function of one
In particular, orthogonality to 41 χj gives χ5 (C 2 ), and then to χ1 + χj for variable. On the other hand, a problem with spherical symmetry has a much
j = 2, 3, 4 gives the rest. greater reduction, from a function of three variables to a function of r only.
To find the representations, instead of just the characters, is in this case
(5) 2
For the D4 symmetry we have only a very finite group, so we can’t expect
fairly simple. The one dimensional ones are  just the characters. Γ (C ) is to reduce the space of functions ψ(x, y) on which we start to ones of a single
−1 0
unitary with trace −2, so it can only be . Γ(5) (mx ) is traceless, variable. But the energy eigenstates must transform as a representation of
0 −1
the group, which can be broken into irreducible representations. The wave
unitary, and has square 1. Any 2 × 2 matrix can be written as a + ~b · ~σ in functions ψ(x, y) then need to have appropriate behavior under x → −x
terms of the Pauli matrices7 , and tracelessness gives a = 0, that the square (my ), under y → −y (mx ), and under x ↔ y (σ+ ).
= 1 gives |~b| = 1. The different directions are all equivalent, all similar to If the symmetry group were only the symmetries of a rectangle, hmx , my i ∼
=

0 1
 
0 −i
 
1 0
 Z2 × Z2 , we would just learn in each cartesian coordinate what we learned
7
σx = , σy = , σz = . in the one dimensional case with parity: ψ could be broken into functions
1 0 i 0 0 −1
618: Last Latexed: April 25, 2017 at 9:45 45 46. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

either even or odd in each of the two variables x and y, according to the 2.6 Point, Space, and Crystallographic Groups
representations Γ(mx ) and Γ(my ). Here we know more, however.
Consider first the identity representation. A wave function in this repre- The example we just discussed, D4 , is an example of a crystallographic point
sentation is unchanged under change in sign of x or of y, so it is a function group. There is much interest in condensed matter physics in crystals, which
of x2 and y 2, but it is also unchanged under interchange of x and y, so it is are periodic lattices invariant under translations by
a symmetric function of them,
T~ = n1~a1 + n2~a2 + n3~a3 , nj ∈ Z.
2 2 2 2
ψ(x, y) = S(x , y ) = S(y , x ) for the identity representation.
where the ~aj are three primitive translation vectors which give the edges of
The other one dimensional representations are also easily expressed in terms a unit cell, and the crystal is unchanged by translating an integer number of
of an arbitrary such S: unit cells in any direction. By itself, this symmetry is enough to force the
wave functions to be given in terms of Block waves.
Γ(mx ) Γ(m ) Γ(σ+ ) f (x, y)
y But there may also be symmetries which leave one point invariant, con-
sisting of rotations and reflections. The symmetries of this kind which are
Γ1 1 1 1 S(x2 , y 2 )
also compatible with the lattice are called crystallographic groups. These
Γ2 −1 −1 1 xyS(x2 , y 2)
are a subset of the point groups, groups of rotations and reflections which
Γ3 1 1 −1 (x2 − y 2 )S(x2 , y 2)
leave one point invariant.
Γ4 −1 −1 −1 xy(x2 − y 2 )S(x2 , y 2 )
The combination of a crystallographic group with the translation group
The fifth representation involves two functions, ψ1 and ψ2 , which trans- is called a space group.
form by In three dimensions there are only 32 possible crystallographic groups.
   Point groups such as C5 , rotations though 72◦ about a given axis, or the
−1 0 ψ1 (x, y) = xg(x2 , y)
Γ(my ) = under x ↔ −x, so symmetries of an icosahedron, are incompatible with extension to an infinite
0 1 ψ2 (x, y) = h(x2 , y) lattice, though the latter might be quite useful in discussing excited states
   of a buckyball.
1 0 ψ1 (x, y) = xg ′ (x2 , y 2 )
Γ(mx ) = under y ↔ −y, so
0 −1 ψ2 (x, y) = yh′(x2 , y 2) I am not going to say more about these groups, but they do have many
Finally, we know it transforms as applications in physics.
 
0 1 The book by Tinkham discusses these in detail.
Γ(σ+ ) = under x ↔ y, so g ′ (x2 , y 2) = h′ (y 2, x2 ),
1 0
or 2.7 Direct Products of Representations
ψ1 (x, y) = xk(x2 , y 2) Often a group acts separately on factorizable pieces of the wave function.
ψ2 (x, y) = yk(y 2, x2 ) For example, if we wish to consider the states of two electrons in our lat-
tice, the total
P wave function ψ(x1 , y1 , x2 , y2) could be written as a sum of
The group theory is not able to tell us anything about the function products i ai φi (x1 , y1 )ρi (x2 , y2 ). φi and ρi can be separately decomposed
S(x2 , y 2) except that it is symmetric, or anything about k. For more you into irreducible representations of the group, and the group acts on ψ by the
need to actually consider the Hamiltonian itself. direct product, because it acts together on both ~r1 and ~r2 . The vector space
618: Last Latexed: April 25, 2017 at 9:45 47 48. Last Latexed: April 25, 2017 at 9:45 Joel A. Shapiro

upon which the group is acting here is the tensor product of the space of all If r, s, t are indices in the ℓi dimensional space on which Γi acts and a, b, c
functions of ~r1 = (x1 , y1 ) with the space of all functions of ~r2 . indices for the ℓj dimensional space on which Γj acts,
Perhaps it would be better to consider first a finite-dimensional example. X X
The electron of a hydrogen atom in an ℓ = 1 state is in a three-dimensional Γra,sb (AB) = Γirs (AB)Γjab (AB) = Γirt (A)Γits (B) Γjac (A)Γjcb (B)
space with basis vectors |mi , m = −1, 0, 1. The rotation group acts on this X
t
X
c
space in a three-dimensional representation to be further discussed later. At = Γirt (A)Γjac (A)Γits (B)Γjcb (B) = Γra,tc (A)Γtc,sb (B)
the same time, the electron has a spin degree of freedom in a two-dimensional tc tc
space with basis {|↑i , |↓i}, or |si , s = ± 12 . A full description of the state = (Γ(A)Γ(B))ra,sb .
of the electron lies in a six-dimensional space whose basis is most naturally
written as {|m, si} or The character of the product representation is simple:
! !
|−1, ↑i , |−1, ↓i , |0, ↑i , |0, ↓i , |1, ↑i , |1, ↓i , X X X
χ(A) = Tr Γ(A) = Γ(A)ra,ra = Γirr (A) Γjaa (A) = χi (A)χj (A).
though of course we could still index these basis vectors as e1 , ..., e6 . Now ra r a
some operators, such as the orbital angular momentum L, ~ may act only on
′ ′ Now the product of two irreducible representations need not be irre-
part of the composite index, hm , s | L |m, si = Lm m δs s , while others might
′ ′
ducible, but it must be equivalent to a direct sum of irreducible representa-
only act on the spin, hm′ , s′ | S |m, si = Ss′ s δm′ m , but in general an operator O,
~ S,
~ is a 6×6 matrix O(m′ s′ ),(ms) = hm′ , s′ | O |m, si. The six dimensional tions. For example, for the case of the electron with orbital angular momen-
e.g. L·
tum 1 and also spin, the full rotation group has representations labelled by j,
space on which it acts is the tensor product of the three dimensional space
and our six dimensional space is a sum of a j = 21 and a j = 32 representation,
of orbital angular momentum states and the two dimensional space of spin.
Under a rotation8 A, the state of the electron will be acted on by a six O 1 1
M 3
Γℓ=1 Γs= 2 ∼
= Γj= 2 Γj= 2 .
dimensional representation
s= 1 More generally,
Γm′ s′ ,ms (A) = Γℓ=1
m′ m (A)Γs′ s (A),
2 O M
Γi (A) Γj (A) ∼
= aij k
k Γ (A),
which is the direct product, k
O 1 where each representation may appear aij k times in this sum, called the
Γ(A) = Γℓ=1 (A) Γs= 2 (A).
Clebsch-Gordon series. The non-negative integers aij k are called Clebsch-
More generally we will consider the product of two irreducible representations Gordon coefficients by mathematicians but not by physicists, who mean
of any group, something else, which we will discuss later, by that term.
O From the formula (2.9) for the number of times a given irreducible rep-
Γ(A) = Γi (A) Γj (A).
resentation occurs in an arbitrary one, the coeffients are easily calculated:
This is clearly a representation, because the matrix sum over the tensor
product index is just an independent sum over the indices of each subspace. 1X i
aij
k = χ (A)χj (A)χk∗ (A).
8
g A∈G
I am being sloppy here. The fermion wave function is not, strictly speaking, trans-
forming under the rotation group, because under a rotation by 2π it is not invariant but
changes sign. So what we are really dealing with here are representations of the “covering We reserve our discussion of the physics of this decomposition until after
group” of the rotation group, which is SU(2) rather than SO(3). This will be discussed we have learned to construct irreducible representations of infinite groups.
later, page ∼60.