Preprint typeset in JHEP style - HYPER VERSION

Chapter 4: Introduction to Representation Theory

Gregory W. Moore

Abstract: BASED MOSTLY ON GTLECT4 FROM 2009. BUT LOTS OF MATERIAL

HAS BEEN IMPORTED AND REARRANGED FROM MATHMETHODS 511, 2014
AND GMP 2010 ON ASSOCIATED BUNDLES. SOMETHING ABOUT SU(2) REPS
AND INDUCED REPS NOW RESTORED. BECAUSE OF IMPORTS THERE IS MUCH
REDUNDANCY. THIS CHAPTER NEEDS A LOT OF WORK. April 27, 2018
-TOC- Contents

1. Symmetries of physical systems 3

2. Basic Definitions 4
2.1 Representation of a group 4
2.2 Matrix Representations 5
2.3 Examples 6
2.3.1 The fundamental representation of a matrix Lie group 6
2.3.2 The determinant representation 6
2.3.3 A representation of the symmetric groups Sn 6
2.3.4 Z and Z2 7
2.3.5 The Heisenberg group 7

3. Unitary Representations 8
3.1 Invariant Integration 9
3.2 Unitarizable Representations 10
3.3 Unitary representations and the Schrödinger equation 11

4. Projective Representations and Central Extensions 12

5. Induced Group Actions On Function Spaces 13

6. The regular representation 14

6.1 Matrix elements as functions on G 16
6.2 RG as a unitary rep: Invariant integration on the group 17
6.3 A More Conceptual Description 17

7. Reducible and Irreducible representations 18

7.1 Definitions 18
7.2 Reducible vs. Completely reducible representations 21

8. Schur’s Lemmas 22

9. Orthogonality relations for matrix elements 23

10. Decomposition of the Regular Representation 25

11. Fourier Analysis as a branch of Representation Theory 31

11.1 The irreducible representations of abelian groups 31
11.2 The character group 31
11.3 Fourier duality 33
11.4 The Poisson summation formula 36

–1–
 11.5 Application: Bloch’s Theorem in Solid State Physics 37
11.6 The Heisenberg group extension of Ŝ × S for an abelian group S 39

12. Induced Representations 41

12.1 Induced Representations of Finite Groups: Frobenius Reciprocity 42

13. Representations Of SU (2) 48

13.0.1 Matrix Reps Of su(2) 51
13.1 Unitary structure 53
13.2 Lie algebra representations and differential operators 54
13.3 Coherent state formalism 54

14. Orbits of the Lorentz group and relativistic wave equations 55

14.1 Orbits, Representations, and Differential Equations 55
14.2 The massless case in 1 + 1 dimensions 57
14.3 The case of d dimensions, d > 2 58

15. Characters and the Decomposition of a Representation to its Irreps 61

15.1 Some basic definitions 61
15.2 Orthogonality relations on the characters 63

16. Some examples of representations of discrete groups and their character

tables 65
16.1 S3 65
16.2 Dihedral groups 68

17. The decomposition of tensor products 69

18. Algebras 71
18.1 Coalgebras, Bialgebras, and Frobenius algebras 72
18.2 When are two algebras equivalent? Introduction to Hochschild cohomology 73

19. The group ring and group algebra 73

19.1 Projection operators in the group algebra 75
19.2 The center of the group algebra and the subalgebra of class functions 76

20. Interlude: 2D Topological Field Theory and Frobenius algebras 77

20.1 Geometric Categories 77
20.2 Some general properties 80
20.3 Two dimensional closed topological field theory and Commutative Frobe-
nius Algebras 82
20.4 Boundary conditions 85

–2–
21. Applications of the Projection Operators 89
21.1 Decomposition of a representation into its isotypical parts 90
21.2 Block diagonalization of Hermitian operators 92
21.2.1 Projecting quantum wavefunctions 92
21.2.2 Finding normal modes in classical mechanics 93

22. Representations of the Symmetric Group 96

22.1 Conjugacy classes in Sn 96
22.2 Young tableaux 96
22.2.1 Example 1: G = S3 98
22.2.2 Example 2: G = S4 99

23. Symmetric groups and tensors: Schur-Weyl duality and the irreps of
GL(d, k) 99
23.1 Free fermions on a circle and Schur functions 103
23.1.1 Schur functions, characters, and Schur-Weyl duality 108
23.2 Bosons and Fermions in 1+1 dimensions 111
23.2.1 Bosonization 111

1. Symmetries of physical systems

♣Much of this is
redundant with
One of the most important applications of group theory in physics is in quantum mechanics. discussion of
Wigner’s theorem
The basic principle is that if G is a symmetry group of a physical system(e.g., rotational which should be
symmetry, translational symmetry, ...) then each element g ∈ G corresponds to a unitary done in general
discussion of
operator acting on the Hilbert space H of physical states, i.e., we have an association: quantum mechanics
in Chapter 2. ♣

∀g ∈ G 7→ U (g) : H → H (1.1)

where U (g) is a unitary operator. Moreover, these unitary operators should “act in the
same way as the physical symmetry.” Mathematically, this means that we have the operator
equation:

U (g1 )U (g2 ) = U (g1 g2 ). (1.2) eq:repi

Moreover, if we have a symmetry of the physical system we should have the same kind
of time-development of two systems related by the symmetry, so

U (g)HU (g)−1 = H (1.3)

where H is the Hamiltonian.

The equation (1.2) is the essential equation defining what is called a representation
of a group, and the above principle is one of the main motivations in physics for studying
representation theory.

–3–
 2. Basic Definitions

2.1 Representation of a group

subsec:ssFD
Let V be a vector space over a field κ and recall that GL(V ) denotes the group of all
nonsingular linear transformations V → V .

Definition 2.1.1 A representation of G with representation space V is a homomorphism

T : g 7→ T (g)
(2.1) eq:repii
G → GL(V )

The “dimension of the representation” is by definition the dimension dimV of the

vector space V . This number can be finite or infinite. For convenience and mathematical
rectitude we will often - but not always - focus on the case dimV < ∞.

Terminology: We will often abbreviate “representation” to “rep.” Moreover we some-

times refer to “the rep (T, V )” or to “the rep T ,” or to “the rep V ,” when the rest of the
data is understood. Some authors call V the “carrier space.”

Definition 2.1.2. A linear transformation A : V → V 0 between representations of G is

called an intertwiner if for all g ∈ G the diagram
A
V → V0
T (g) ↓ ↓ T 0 (g) (2.2)
A
V → V0

commutes. Equivalently,

T 0 (g)A = AT (g) (2.3) eq:intertwin

for all g ∈ G.

Definition 2.1.3. Two representations are equivalent (T, V ) ∼

= (T 0 , V 0 ) if there is an
intertwiner which is an isomorphism. That is,

T 0 (g) = AT (g)A−1 (2.4) eq:wuivrep

for all g ∈ G.
Familiar notions of linear algebra generalize to representations:

1. The direct sum ⊕, tensor product ⊗ etc. of representations. Thus, the direct sum of
(T1 , V1 ) and (T2 , V2 ) is the rep (T1 ⊕ T2 , V1 ⊕ V2 ) where the representation space is
V1 ⊕ V2 and the operators are:



(T1 ⊕ T2 )(g) v1 , v2 := T1 (g)v1 , T2 (g)v2 (2.5)

–4–
 2. Similarly, for the tensor product, the carrier space is V1 ⊗ V2 and the group elements
are represented by:


(T1 ⊗ T2 )(g) v1 ⊗ v2 := (T1 (g)v1 ) ⊗ (T2 (g)v2 ) (2.6)

3. Given a representation of V we get a dual representation on the dual space V ∨ by

demanding that under the pairing between V and V ∨ :

hT ∨ (g)`, T (g)vi = h`, vi, (2.7)

where ` ∈ V ∨ , v ∈ V . Thus, if V, W are representation spaces then so is Hom(V, W ).

4. If V is a complex vector space then the complex conjugate representation sends

g → T̄ (g) ∈ GL(V̄ ). A real representation is one where (T̄ , V̄ ) is equivalent to
(T, V ).

2.2 Matrix Representations

Definition 2.1.3 A matrix representation is a homomorphism:

T : G → GL(n, κ) (2.8)

Given a representation, and an ordered basis {~v1 , . . . , ~vn } for V we can associate to a
representation a matrix representation. The point is, with a basis we can identify GL(V ) ∼ =
GL(n, κ). Specifically, we get the matrix from the linear transformation by:
n
X
T (g)~vk = T (g)jk~vj (2.9)
j=1

so the T (g)jk are the matrix elements of an element of GL(n, κ). We will sometimes denote
this matrix as T (g) if it is understood we are using an ordered basis for V .
Now, if we change basis
n
X
~vi = Sji~vj0 (2.10)
j=1

then the matrices change to

T 0 (g) = ST (g)S −1 (2.11)
and this motivates
Definition Two n-dimensional matrix representations T and T 0 are equivalent, denoted
T ∼
= T 0 if ∃S ∈ GL(n, κ) with
T 0 (g) = ST (g)S −1 (2.12)

Exercise Complex conjugate and transpose-inverse representations

Given a matrix representation of a group g → T (g) show that
a.) g → (T (g))tr,−1 is also a representation.

–5–
 If we choose a basis vi for V then this is the matrix representation for the dual repre-
sentation in the dual basis v̂i .
b.) If T is a complex matrix representation wrt basis vi then the complex conjugate
representation with respect to v̄i is: g → T (g)∗ .
c.) If T is a real representation, then there exists an S ∈ GL(n, C) such that for all
g ∈ G:
T ∗ (g) = ST (g)S −1 (2.13)
Warning: The matrix elements T (g)ij of a real representation can of course fail to be
real numbers!

2.3 Examples
2.3.1 The fundamental representation of a matrix Lie group
GL(n, κ), SL(n, κ), O(n, κ), U (n) are all matrix representations of themselves! In the first
three examples V = κn . In the fourth V = Cn . These are called “the fundamental
representation.” Note that when κ = Cn and n > 2 the fundamental representation is not
equivalent to its complex conjugate 1 so the fundamental representation is not the same as
the “minimal-dimensional nontrivial representation.”

2.3.2 The determinant representation

The general linear group GL(n, κ) always has a family of one-dimensional real representa-
tions detµ , µ ∈ R, given by
T (A) := |detA|µ (2.14) eq:dtrmrep

This is a representation because:

T (AB) = |detAB|µ = |detA|µ |detB|µ = T (A)T (B) (2.15)

2.3.3 A representation of the symmetric groups Sn

The symmetric group Sn has an n-dimensional representation with V = Rn (or Cn ) defined
by choosing the standard basis {ei } and defining

T (σ) · ei := eσ(i) (2.16)

Note that acting on coordinates xi ei this takes xi → xσ−1 (i) , i.e.

   
x1 xσ−1 (1)
 ..   . 
 .  →  ..  (2.17)
xn xσ−1 (n)

and thus the matrix representation has matrix elements which are just 0’s and 1’s with a
single nonzero entry in each row and column.
1
This is easily proven using characters, see below.

–6–
 Explicitly for n = 2 we have, for examples:
!
01
(12) → (2.18) eq:symmrep
10

while for n = 3 we have:

 
0 1 0
(12) → 1 0 0
 
0 0 1
 
1 0 0
(23) → 0 0 1 (2.19)
 
eq:symmrepii
0 1 0
 
0 0 1
(123) → 1 0 0
 
0 1 0

2.3.4 Z and Z2

Sometimes relations come in families. If θ ∈ R/Z then we can define a representation Tθ

of the group Z by Tθ (n) = e2πinθ .
At special values of parameters, special things can happen. For example note that
when θ = 1/2 the subgroup of even integers is represented trivially, and the representation
“factors through” the sign representation of S2 . That is, T1/2 is the composition of the
projection Z → Z2 and the sign representation ε of Z2 .
Similar, but more elaborate and interesting things happen for the representations of
the braid group and its projection Bn → Sn .

2.3.5 The Heisenberg group

An interesting group, the Heisenberg group, can be defined in terms of generators q, U, V

where q is central and U V = qV U :

H = hq, U, V |qU = U q, qV = V q, U V = qV U i (2.20) eq:heisn

If we we add the relations q N = U N = V N = 1 then it is the finite Heisenberg group

extension of ZN × ZN by ZN :

1 → ZN → HeisN → ZN × ZN → 1 (2.21) eq:heisent

described in chapter 1, section 11.3.

Let ω be an N th root of 1.
Let U = Diag{1, ω, ω 2 , . . . , ω N −1 } be the “clock operator.”

–7–
 Let V be the “shift operator:”
 
  0 1 0 ··· 0
! 010  0 0 1 ··· 0 
01  
V = V = 0 0 1 V = · · · ··· ··· ··· ··· (2.22)
   
10  
100 · · · ··· ··· ··· 1 
1 0 ··· ··· 0

It is easy to check that:

U V = qV U (2.23)
with q = ω.
If we let ω range over the N th roots of unity we obtain N inequivalent N × N (irre-
ducible) representations.
If we consider the group without imposing q N = U N = V N = 1 there are infinitely
many inequivalent representations.
These assertions are obvious since different values of the central element q cannot be
conjugated into each other.

3. Unitary Representations
sec:sUR
Of particular importance in physics are the unitary representations. By Wigner’s theorem
described in section **** above we know that symmetry transformations will act as unitary
or anti-unitary transformations. The basic reason for this is that these preserve norms and
hence probability amplitudes.

Definition 3.1. Let V be an inner product space. A unitary representation is a represen-

tation (T, V ) such that ∀g ∈ G, T (g) is a unitary operator on V , i.e.,

hT (g)v, T (g)vi = hv, vi ∀g ∈ G, v ∈ V (3.1)

Definition 3.2. A unitary matrix representation is a homomorphism

T : G → U (n) (3.2)

Exercise
a.) Show that if T (g) is a rep on an inner product space then T (g −1 )† is a rep also.
b.) Suppose T : G → GL(V ) is a unitary rep on an inner product space V . Let {~vi }
be an ON basis for V . Show that the corresponding matrix rep T (g)ij is a unitary matrix
rep.
c.) Show that for a unitary matrix rep the transpose-inverse and complex conjugate
representations are equivalent.

–8–
Definition 3.2. If a rep (T, V ) is equivalent to a unitary rep then such a rep is said to be
unitarizable.

Example. A simple example of non-unitarizable reps are the detµ reps of GL(n, k)
described in section 2.3.4.

3.1 Invariant Integration

When proving facts about unitary representations a very important tool is the notion of
invariant integration. We have already made use of it in our discussion of lattice gauge
theory in Section *** of Chapter 1. Here is a reminder:
************************************

Important Remark: The notion of averaging over the group can be extended to a much
larger class of groups than finite groups. It is given by invariant integration over the group
which is the analog of the operation
1 X
f→ f (g) (3.3)
|G| g

on functions.
In general we replace Z
1 X
f (g) → f (g)dg (3.4)
|G| g G
R
where G f (g)dg should be regarded as a rule such that:
R
1. G f (g)dg is a complex number depending linearly on f ∈ RG .
2. It satisfies the left invariance property:
Z Z
f (hg)dg = f (g)dg (3.5)
G G

for all h ∈ G. This is the generalization of the rearrangement lemma. (We can define
right invariance in a similar way. Compact groups admit integration measures which are
simultaneously left- and right- invariant. )
Examples:

Z Z +2π
dθ
f (g)dg ≡ f (θ)
G=U (1) 0 2π
Z X
f (g)dg ≡ f (n) (3.6) eq:intgroup
G=Z n∈Z
Z Z +∞
f (g)dg ≡ dxf (x)
G=R −∞

The notion of invariant integration also extends to all compact groups. For the impor-
tant case of G = SU (2) we can write it as follows. First, every element of SU (2) can be
written as: ♣Definition of β
here is backwards
from what we use
when we describe
reps using
homogeneous
polynomials below.
♣
–9–
 !
α β
g= (3.7)
−β ∗ α∗
for 2 complex numbers α, β with
|α|2 + |β|2 = 1. (3.8)
In this way we identify the group as a manifold as S 3 . That manifold has no globally
well-defined coordinate chart. The best we can do is define coordinates that cover “most”
of the group but will have singularities are some places. (It is always important to be
careful about those singularities when using explicit coordinates!) One way to do this is to
write
1
α = ei 2 (φ+ψ) cos θ/2
1
(3.9)
β = iei 2 (φ−ψ) sin θ/2

With this definition we can write

! ! !
eiφ/2 0 cos θ/2 i sin θ/2 eiψ/2 0
g=
0 e−iφ/2 i sin θ/2 cos θ/2 0 e−iψ/2 (3.10)
i 21 φσ 3 i 12 θσ 1 i 21 ψσ 3
=e e e

If we take 0 ≤ θ ≤ π, and identify φ ∼ φ + 2π and ψ ∼ ψ + 4π we cover the manifold once,

away from the singular points at θ = 0, π.
The normalized Haar measure for SU (2) in these coordinates is 2
1
[dg] = dψ ∧ dφ ∧ sin θdθ (3.11)
16π 2
For much more about this, see Chapter 5 below.
We give formulae for the Haar measures on the classical compact matrix groups in
Chapter 6 below.
************************************************************

3.2 Unitarizable Representations

Proposition If T is a rep on an inner product space of a compact group G, then V is finite

dimensional and T is unitarizable.

Proof for finite groups: If T is not already unitary with respect to the inner product
(·, ·)1 then we can define a new inner product by:
1 X
hv, wi2 ≡ hT (g)v, T (g)wi1 (3.12)
|G|
g∈G

This proof generalizes using a left- and right- invariant Haar measure, which always exists
for a compact group.
2
We have chosen an orientation so that, with a positive constant, this is Tr2 (g −1 dg)3 ).

– 10 –
 Exercise
a.) Show that
hT (g)v, T (g)wi2 = hv, wi2 (3.13)
and deduce that T (g) is unitary with respect to h·, ·i2 .
b.) Show that if T (g) is a finite-dimensional matrix rep of a finite group then it is
equivalent to a unitary matrix rep.

3.3 Unitary representations and the Schrödinger equation

♣THIS
SUBJECTION IS
Consider the Schrödinger equation for stationary states of energy E: NOW OUT OF
PLACE. ♣

~2
− ∆ψ(~x) + V (~x)ψ(~x) = Eψ(~x) (3.14) eq:shro
2m
and suppose V is rotationally invariant:

V (~x) = V (| ~x |). (3.15)

Now define
VE = {ψ(~x) : ψsolves(3.14) and is normalizable} (3.16)
Normalizable, or square integrable means:
Z
2
kψk = | ψ(~x) |2 d3 x < ∞ (3.17)
R3

so VE ⊂ L2 (R3 ).
Claim: VE is a representation space of O(3): T : O(3) → GL(VE ) given, as usual, by:

[T (g)ψ](~x) = ψ(g −1 · ~x) (3.18)

Must check:
1. ψ solves (3.14) ⇒ T (g)ψ solves (3.14).
2. k ψ k2 < ∞ ⇒k T (g)ψ k2 < ∞
Check of 1: If x0i = (g −1 )ji xj then
3 3
X ∂2 X ∂2
∆0 = = =∆ (3.19)
i=1
∂x0i 2 i=1
∂x2i

and therefore:
∆ T (g)ψ (x) = ∆(ψ(x0 )) = ∆0 (ψ(x0 ))
 
(3.20)
Check of 2: x0 = g −1 x gives
0
dxi
d3 x0 =| det | d3 x = d3 x (3.21)
dxj

– 11 –
 so:
k ψ k2 =k T (g)ψ k2 (3.22)

and hence this is an example of a unitary representation.

cal states of a theory form unitary reps of the symmetry group. Since the group commutes with the Hamiltonian we can diag

Thus, for example, the wavefunctions in a spherically symmetric potential have quan-
tum numbers, |E, j, m, · · · i.

Exercise Helmholtz equation

Consider the Helmholtz equation:

∆u(~x) + k 2 u(~x) = 0
3
X ∂2 (3.23) eq:hlmhltz
∆=
i=1
∂x2i

Let

V = {u(~x) : u satisfies(3.23)} eq:hlmhltziii (3.24) eq:hlmhltziii

Show that V in (3.24) is a representation of the isometry group E(3).

4. Projective Representations and Central Extensions

♣GIVEN THE
EXTENSIVE
A projective representation is a pair (T, V ) where T : G → GL(V ) “almost” is a homomor- MATERIAL ON
CENTRAL
phism, but is allowed to deviate from a homomorphism by a phase: EXTENSIONS IN
CHAPTER ONE
THIS NEEDS TO
BE REWRITTEN
T (g1 )T (g2 ) = c(g1 , g2 )T (g1 g2 ) (4.1) eq:projrep
♣

where c(g1 , g2 ) ∈ C∗ . For unitary representations c(g1 , g2 ) ∈ U (1).

From the associativity of a product of three transformations T (g1 )T (g2 )T (g3 ) we find
that c : G × G → C∗ is a 2-cocycle.
Recalling our discussion of central extensions we see that T should really be regarded
as a representation of the central extension of G by C∗ (or U (1)) constructed from the
cocycle c. Put more formally

T̂ (z, g) := zT (g) (4.2) eq:cntpfjr

defines a true representation of the central extension Ĝ defined by c.

Moreover, if we allow ourselves the freedom to redefine T (g) → T̃ (g) := f (g)T (g)
where f : G → C∗ is a function then c changes by a coboundary.

– 12 –
 Example Returning to our clock and shift operators, we can define a projective repre-
sentation of Zn × Zn by
0
(s̄, s̄0 ) → U s̄ V s̄ (4.3)

but these only satisfy the group law up to a power of ω. In fact, what we have is a
representation of Heis(Zn × Zn ).

5. Induced Group Actions On Function Spaces

:InducedAction ♣NOTE THERE IS
OVERLAP OF
Let X be a G-set and let Y be any set. There are natural left- and right- actions on the THIS MATERIAL
WITH EXAMPLES
function space Map(X, Y ). Given Ψ ∈ Map(X, Y ) and g ∈ G we need to produce a new 5,6 OF FUNCTORS
IN CATEGORY
function φ(g, Ψ) ∈ Map(X, Y ). The rules are as follows: SECTION OF
CHAPTER 1 ♣

1. If G is a left-action on X then

φ(g, Ψ)(x) := Ψ(g · x) right action on Map(X, Y ) (5.1)

2. If G is a left-action on X then

φ(g, Ψ)(x) := Ψ(g −1 · x) left action on Map(X, Y ) (5.2)

3. If G is a right-action on X then

φ(g, Ψ)(x) := Ψ(x · g) left action on Map(X, Y ) (5.3)

4. If G is a right-action on X then

φ(g, Ψ)(x) := Ψ(x · g −1 ) right action on Map(X, Y ) (5.4)

Example: Consider a spacetime S. With suitable analytic restrictions the space of scalar
fields on S is Map(S, κ), where κ = R or C for real or complex scalar fields. If a group
G acts on the spacetime, there is automatically an induced action on the space of scalar
fields. To be even specific, suppose X = M1,d−1 is d-dimensional Minkowski space time, G
is the Poincaré group, and Y = R. Given one scalar field Ψ and a Poincaré transformation
g −1 · x = Λx + v we have (g · Ψ)(x) = Ψ(Λx + v).

Similarly, suppose that X is any set, but now Y is a G-set. Then again there is a
G-action on Map(X, Y ):

(g · Ψ)(x) := g · Ψ(x) or Ψ(x) · g (5.5)

according to whether the G action on Y is a left- or a right-action, respectively. These are

left- or right-actions, respectively.

– 13 –
 We can now combine these two observations and get the general statement: We assume
that both X is a G1 -set and Y is a G2 -set. We can assume, without loss of generality, that
we have left-actions on both X and Y . Then there is a natural G1 ×G2 -action on Map(X, Y )
defined by:
φ((g1 , g2 ), Ψ)(x) := g2 · (Ψ(g1−1 · x)) (5.6) eq:GenAction

note that if one writes instead g2 · (Ψ(g1 · x)) on the RHS then we do not have a well-defined
G1 × G2 -action (if G1 and G2 are both nonabelian). In most applications X and Y both
have a G action for a single group and we write

φ(g, Ψ)(x) := g · (Ψ(g −1 · x)) (5.7)

This is a special case of the general action (5.6), with G1 = G2 = G and specialized to the
diagonal ∆ ⊂ G × G.

Example: Again let X = M1,d−1 be a Minkowski space time. Take G1 = G2 and let
G = ∆ ⊂ G × G be the diagonal subgroup, and take G to be the Poincaré group. Now
let Y = V be a finite-dimensional representation of the Poincaré group. Let us denote the
action of g ∈ G on V by ρ(g). Then a field Ψ ∈ Map(X, Y ) has an action of the Poincaré
group defined by
g · Ψ(x) := ρ(g)Ψ(g −1 x) (5.8)
This is the standard way that fields with nonzero “spin” transform under the Poincaré
group in field theory. As a very concrete related example, consider the transformation of
electron wavefunctions in nonrelativistic quantum mechanics. The electron wavefunction
is governed by a two-component function on R3 :
!
ψ+ (~x)
Ψ(~x) = (5.9)
ψ− (~x)

Then, suppose G = SU (2). Recall there is a surjective homomorphism π : G → SO(3)

defined by π(u) = R where
u~x · ~σ u−1 = (R~x) · ~σ (5.10)
Then the (double-cover) of the rotation group acts to define the transformed electron
wavefunction u · Ψ by !
ψ+ (R−1 ~x)
(u · Ψ)(~x) := u (5.11)
ψ− (R−1 ~x)
In particular, u = −1 acts trivially on ~x but nontrivially on the wavefunction.

6. The regular representation

Now that we have seen several examples of representations we would like to introduce a
“universal example,” of a representation of G, called the regular representation. We will
see that from this representation we can learn about all of the representations of G, at
least when G is compact.

– 14 –
 Recall once again the principle of section 5: If X has G-action then the space FX,Y of
all functions X → Y also has a G-action. In particular, if Y is a vector space then FX,Y is
a vector space and we have a natural source of representations of G. In particular, we can
make the simplest choice Y = C, so the space of complex-valued functions on X, FX,C is
a very natural source of representations of G. ♣THIS IS OLD.
SHOULD MAKE IT
If G acts on nothing else, it certainly acts on itself as a group of transformations of A REPRESENTA-
TION OF G × G
X = G with left or right action. Applying the above general principle we see that the FROM THE
START. ♣
space of complex-valued functions on G:

FG,C := {M aps φ : G → C} (6.1)

forms a natural - “God given” - representation space of G. FG,C is also denoted by RG .

When G is infinite we usually want to put some conditions on the functions in question.
If G admits a left-invariant integration measure then by RG we understand L2 (G) in this
measure.
In fact, RG is a representation space in two ways:
1. Left regular representation (LRR):

φ 7→ L(g) · φ (6.2)

where L(g) · φ has values defined by

(L(g) · φ)(h) ≡ φ(g −1 h) (6.3)

2. Right regular representation (RRR):

φ 7→ R(g) · φ (6.4)

where R(g) · φ has values defined by

(R(g) · φ)(h) ≡ φ(hg) (6.5)

Exercise
Check that
L(g1 )L(g2 ) = L(g1 g2 ) (6.6)
and that
R(g1 )R(g2 ) = R(g1 g2 ) (6.7)

Note too that for all g, h ∈ G

R(g)L(h) = L(h)R(g) (6.8)

so, in fact, RG is a representation of G × G = Gleft × Gright .

– 15 –
 Now let us assume | G |< ∞. Then RG is a finite dimensional vector space. we can
label a function f by its values f (g).

Example. G = Z2 = {1, σ}. Then RZ2 ∼ = C2 as a vector space, since a function on Z2 is

labelled by (a, b) where a = f (1), b = f (σ).
Quite generally, we can introduce a basis of “delta functions” concentrated at g:
(
0 1 f org 0 = g
δg (g ) = (6.9)
0 else
The functions δg ∈ RG span RG : X
f= f (g)δg (6.10)
g

Exercise
If G is a finite group, then show that

dimRG = |G| (6.11)

Exercise
a.) Let δ0 , δ1 , δ2 be a basis of functions in the regular representation of Z3 which are
1 on 1, ω, ω 2 , respectively, and zero elsewhere. Show that ω is represented as
 
010
L(ω) = 0 0 1 (6.12)
 
100
b.) Show that
L(h) · δg = δh·g
(6.13)
R(h) · δg = δg·h−1
and conclude that for the left, or right, regular representation of a finite group the repre-
sentation matrices in the δ-function basis have entries which are simply zeroes or ones in
the natural basis.

6.1 Matrix elements as functions on G

Now we make an important conceptual step forward: ♣Actually, for
current level of
Let g → Tij (g) be any matrix representation of G. earlier chapters this
is pretty obvious to
the reader. ♣

We may regard the matrix elements Tm` as complex-valued functions on G. There-

fore, we may regard Tm` as vectors in the regular representation RG .

– 16 –
6.2 RG as a unitary rep: Invariant integration on the group
When G is a finite group we can turn the vector space RG into an inner product space by
the defining an inner product using averaging over the group:

1 X ∗
hφ1 , φ2 i := φ1 (g)φ2 (g) (6.14) eq:innprod
|G|
g∈G

Theorem. With the inner product (6.14) the representation space RG is unitary for both
the RRR and the LRR.

Proof : Straightforward application of the rearrangement lemma.

6.3 A More Conceptual Description

♣MUCH OF THIS
IS REDUNDANT
As a very nice example of the idea of how group actions on a space induce group actions WITH ABOVE
BUT SOME
on the functions on that space we touch briefly on the Peter-Weyl theorem and the idea of THINGS ARE
SAID MORE
induced representations. NICELY, esp. map
of End(V ) into
We first describe the Peter-Weyl theorem: L2 (G). ♣
Let G be a group. Then there is a left action of G × G on G: (g1 , g2 ) 7→ L(g1 )R(g2−1 ). ♣Notation in this
subsection is at
Now let Y = C. Then Map(G, C) is known as the regular representation of G because the variance with rest
of chapter: Here ρ
induced left-action: is the rep

((g1 , g2 ) · Ψ) (h) := Ψ(g1−1 hg2 )

homomorphism and
(6.15) T is a general
operator... ♣
converts the vector space Map(G → C) of functions Ψ : G → C into a representation space
for G × G.
Suppose, on the other hand that V is a linear representation of G. As mentioned above
this means we have a group homomorphism ρ : G → GL(V ). Then consider the vector
space of linear transformations End(V ) of V to itself. This is also a representation of G×G
because if T ∈ End(V ) then we can define a linear left-action of G × G on End(V ) by:

(g1 , g2 ) · T := ρ(g1 ) ◦ T ◦ ρ(g2 )−1 (6.16)

Now, we have two representations of G × G. How are they related? If V is finite-

dimensional we have a map
ι : End(V ) → Map(G, C) (6.17)
The map ι takes a linear transformation T : V → V to the complex-valued function
ΨT : G → C defined by
ΨT (g) := TrV (T ρ(g −1 )) (6.18) eq:PsiTee

If we choose a basis wµ for V then the operators ρ(g) are represented by matrices:
X
ρ(g) · wν = D(g)µν wµ (6.19)
µ

If we take T = eνµ to be the matrix unit in this basis then ΨT is the function on G given by
the matrix element D(g −1 )µν . So the ΨT ’s are linear combinations of matrix elements of

– 17 –
the representation matrices of G. The advantage of (6.18) is that it is completely canonical
and basis-independent.
Note that ι : T 7→ ΨT “commutes with the G × G action.” What this means is that

(g1 , g2 ) · ΨT = Ψ(g1 ,g2 )·T (6.20) eq:GxG-commute

(The reader should check this carefully.) Such a map is said to be equivariant. Put
differently, denoting by ρEnd(V ) the representation of G × G on End(V ) and ρReg.Rep. the
representation of G × G on Map(G, C) we get a commutative diagram:

End(V )
ι / Map(G, C) (6.21)
ρEnd(V ) ρReg.Rep.
 
End(V )
ι / Map(G, C)

In particular if we have a collection of finite-dimensional representations {Vλ } of G

then we have
⊕λ End(Vλ ) ,→ Map(G, C) (6.22) eq:PW-1

Thanks to the equivariance, the image of (6.22) is a G × G-invariant subspace, i.e. a

subrepresentation of Map(G, C). The very beautiful Peter-Weyl theorem states that, if G
is a compact group, then, as representations of G × G, (6.22) is an isomorphism if the sum
is over the distinct isomorphism classes λ of irreducible representations Vλ of G and we ♣IRREPS NOT
YET EXPLAINED
restrict to the the subspace of Map(G, C) of L2 -normalizable functions with respect to a ♣

left-right-invariant measure on G. 3

7. Reducible and Irreducible representations

7.1 Definitions
Sometimes reps are “too big” and one wants to reduce them to their “essential parts.”
Reducing a representation to smaller representations is closely analogous to diagonalization
of matrices.
For example – we will see that the natural representation of a group - its regular
representation - is in fact highly reducible.

Figure 1: T (g) preserves the subspace W . fig:redrep

3
In order for this to be completely correct we need to assume that G is a compact group. Then we
introduce a left-right-invariant measure on G and replace the LHS by L2 (G).

– 18 –
Definition. Let W ⊂ V be a subspace of a group representation T : G → GL(V ). Then
W is invariant under T if ∀g ∈ G, w ∈ W

T (g)w ∈ W (7.1)

This may be pictured as in 1. Example: Consider the three-dimensional representation

of SO(2) as rotations around the z-axis. Then the vector subspace of the xy plane is an
invariant subspace.
Since T preserves the smaller vector space W , we can define a smaller group represen-
tation (T, W ). This is called the restriction of T to W .
Remarks
•
If T is unitary on V then it is unitary on W .
• Both {~0} and V are always invariant subspaces.

Definition. A representation T is called reducible if there is an invariant subspace W ⊂ V ,

under T , which is nontrivial, i.e., such that W 6= 0, V , If V is not reducible we say V is
irreducible. That is, in an irreducible rep, the only invariant subspaces are {~0} and V . We
often shorten the term “irreducible representation” to “irrep.”

Note:
• Given any nonzero vector v ∈ V , the linear span of {T (g)v}g∈G is an invariant
subspace. In an irrep this will span all of V .
• If W is an subrepresentation of V then T descends to a representation on V /W .

Let us see how this works in terms of matrix representations. Suppose T is reducible.
Then we can choose a basis
{v1 , . . . vk } for W
and
{v1 , . . . vk , vk+1 , . . . vn } for V
In this basis the matrix representation of T looks like:
!
T1 (g) T12 (g)
T (g) = (7.2) eq:redrep
0 T2 (g)
where
T1 (g) ∈ M atk×k
T12 (g) ∈ M atk×n−k
T2 (g) ∈ M at(n−k)×(n−k)
Writing out T (g1 )T (g2 ) = T (g1 g2 ) we see that T1 is the representation on W and T2
is the representation on V /W . T12 transforms in a more complicated way.
Definition. A representation T is called completely reducible if it is isomorphic to

W 1 ⊕ · · · ⊕ Wn (7.3)

– 19 –
where the Wi are irreducible reps. Thus, there is a basis in which the matrices look like:
 
T1 (g) 0 0 ···
 0 T (g) 0
2 ···
T (g) =  (7.4)
 
eq:comredrep
···

 0 0 T3 (g)
··· ··· ··· ···
Examples
• G = Z2
!
10
1→
01
! (7.5)
01
(12) →
10

is a 2-dimensional reducible rep on R2 because W = {(x, x)} ⊂ R2 is a nontrivial invariant

subspace. Indeed, σ 1 is diagonalizable, so this rep is equivalent to
!
10
1→
01
! (7.6)
1 0
(12) →
0 −1

which is a direct sum of two 1 × 1 reps.

• Consider the representation of Sn on Rn . Then the one-dimensional subspace L =

{(x, · · · , x)} is a subrepresentation. Moreover we can take L⊥ = {x1 , · · · xn | xi = 0}.
P

Then L⊥ is an (n − 1)-dimensional representation of Sn and the representation is block

diagonal on L ⊕ L⊥ . ♣THIS IS OLD.
MAYBE START
• Invariant subspaces of the regular representation. Let Tij be any matrix n-dimensional WITH SPAN OF
Tij WHERE BOTH
representation of G. i, j VARY AS REP
OF G × G. WE
Claim: The linear span of functions CAN ALWAYS
THEN RESTRICT
TO G × 1 AND
R := Span{Tij }j=1,...,nµ (7.7) 1 × G. ♣

is an invariant subspace of RG , considered as the RRR.

Proof:
(R(g) · Tij )(h) = Tij (hg)
nµ
X (7.8)
= Tis (h)Tsj (g)
s=1

which is equivalent to the equation on functions:

function on G nµ function on G
z }| { X z}|{
R(g) · Tij = Tsj (g) Tis (7.9) eq:mtxeltinvt
| {z }
s=1
matrix element for RG

– 20 –
 7.2 Reducible vs. Completely reducible representations
subsec:ssRedC
Irreps are the “atoms” out of which all reps are made. Thus we are naturally led to study
the irreducible reps of G. In real life it can and does actually happen that a group G
has representations which are reducible but not completely reducible. Reducible, but not
completely reducible reps are sometimes called indecomposable.

Example . An example of an indecomposible rep which is not completely reducible is the

rep !
1 log|detA|
A→ (7.10)
0 1

of GL(n, R). As we will see, the Poincare group has indecomposable reps.

R_G

W^\perp

Figure 2: The orthogonal complement of an invariant subspace is an invariant subspace. fig:orthcomp

It is useful to have criteria for when this complication cannot occur:

Proposition 7.2.1 Suppose that the representation (T, V ) is a unitary representation on

an inner product space V and W ⊂ V is an invariant subspace then W ⊥ is an invariant
subspace:
Proof : Recall that

y ∈ W ⊥ ⇔ ∀x ∈ W, hy, xi = 0 (7.11)

Let g ∈ G, y ∈ W ⊥ . Compute

T (g)y, x = y, T (g)† x



(7.12)
= y, T (g −1 )x

But, T (g −1 )x ∈ W , since W ⊂ V is an invariant subspace.

therefore: ∀x ∈ W , hT (g)y, xi = 0.

– 21 –
 therefore: T (g)y ∈ W ⊥
therefore: W ⊥ is an invariant subspace. ♠

Corollary: Finite dimensional unitary representations are always completely reducible.

In particular, by section 4:
1. Finite dimensional reps of finite groups are completely reducible.
2. More generally, finite dimensional reps of compact Lie groups are completely re-
ducible.

It follows that for a finite group the regular representation RG is completely re-
ducible. In the next several sections we will show how to decompose RG in terms of
the irreducible representations of G.

If V is a completely decomposable representation (always true for compact groups)

then we can write
∼ ⊕r ⊕aµ T (µ) := ⊕µ aµ T (µ)
V = (7.13)
µ=1 i=1

For a fixed µ let let

a
V (µ) := ⊕i=1
µ
T (µ) (7.14)

It contains the representation µ with the correct degeneracy in V . It is called the isotypical
component belonging to µ. Note that it can be written as

V (µ) = Caµ ⊗ T (µ) (7.15) eq:isotpy

where Caµ is understood to be the trivial representation of G.

We abbreviate Caµ ⊗T (µ) to aµ T µ and with this understood we write the decomposition
into isotypical components as:
V = ⊕aµ T µ (7.16)

8. Schur’s Lemmas
♣This section needs
improvement:
When thinking about irreducible representations it is important to understand how they Schur’s lemma says
that the algebra of
are related to each other. A key technical tool is Schur’s lemma. It is usually stated as intertwiners of an
irrep is a division
two separate statements, each of which is practically a triviality. algebra. So R, C, H.
What is here is
correct, but not the
best viewpoint. ♣

Theorem 1: If A : V → V 0 is an intertwiner between two irreps then it is either an

isomorphism or zero.

– 22 –
 Proof: kerA and Im A are both invariant subspaces. Of course, if V and V 0 are
inequivalent then A = 0. ♠.

Theorem 2: If A is a complex matrix commuting with an irreducible matrix repre-

sentation then A is proportional to the identity matrix.

Proof: Since we are working over the complex field A has a nonzero eigenvector Av =
λv. The eigenspace C = {w : Aw = λw} is therefore not the zero vector space. But it is
also an invariant subspace. Therefore, it must be the entire carrier space. ♠.

Remarks:
• Consider the isotypical decomposition of a completely reducible representation. By
Schur’s lemma, the intertwiners of V (µ) with itself are of the form K ⊗ 1 where K ∈
End(Caµ ) is arbitrary.
• Schur’s lemma is used in quantum mechanics very often. Note that a Hamiltonian
invariant under some symmetry group U (g)HU (g)−1 = H is an intertwiner. Therefore,
if we decompose the Hilbert space of states into irreps of G then H must be scalar on
each irrep. In particular this leads to important selection rules in physics. For example, if
∆H is an invariant operator which commutes with some symmetry G of a physical system
then the matrix elements of ∆H between states in different irreducible representations of
G must vanish.

9. Orthogonality relations for matrix elements

Let G be a finite group. We are interested in finding the reduction of RG to its irreducible
representations. Our next technical tool - very useful in a variety of contexts - are the
orthogonality relations of matrix elements of irreps.
Label the distinct, i.e. inequivalent, irreducible representations by

T (µ) µ = 1, 2, . . . , r (9.1)

and let nµ = dimT (µ) . (We will prove later that r < ∞ and nµ < ∞.) In order not to
overburden the notation we will denote the carrier space and the homomorphism by the
same name T (µ) , relying on context to tell which is intended. Choose bases for T (µ) and
consider these to be matrix irreps.
Let B ∈ M atnµ ×nν , and define a matrix A ∈ M atnµ ×nν by: ♣SHOULD GIVE A
MORE
CONCEPTUAL
1 X (µ) DESCRIPTION OF
A≡ T (g)BT (ν) (g −1 ) (9.2) eq:orthi
WHAT IS GOING
|G| ON HERE USING
g∈G P ∨
g T (g) ⊗ T (g).
♣
We claim that A is an intertwiner, that is, ∀h ∈ G:

T (µ) (h)A = AT (ν) (h) (9.3)

– 23 –
Proof:
1 X (µ)
T (µ) (h)A = T (h)T (µ) (g)BT (ν) (g −1 )
|G|
g∈G
1 X (µ)
= T (hg)BT (ν) (g −1 )
|G|
g∈G
1 X (µ) 0 (9.4)
= T (g )BT (ν) ((h−1 g 0 )−1 )
|G| 0
g ∈G
1 X (µ)
= T (g)BT (ν) ((g)−1 )T (ν) (h)
|G|
g∈G
(ν)
= AT (h)

In the third line let g 0 = hg, and use the rearrangement lemma.
Therefore, by Schur’s lemma:

A = λδµ,ν 1nν ×nν (9.5) eq:orthii

where λ can depend on µ = ν, B.

Now, to extract useful information from this let B = e`m be a matrix unit, 1 ≤ ` ≤ nµ ,
1 ≤ m ≤ nν . Then (9.2)(9.5) imply:

(µ)
X
(ν) −1
Ti` (g)Tms (g ) = |G|λδµν δis (9.6) eq:orthiii
g∈G

where now λ depends on m, `, µ, but does not depend on i, s.

We can determine λ by setting µ = ν, i = s and summing on i:
Then (9.6) becomes:

(µ)
X
Tm` (g −1 g) = |G|λnµ
g∈G

|G|δ`m = |G|λnµ
(9.7)

δ`m
⇒λ=
nµ
Putting it all together we get the main result of this section:
Theorem The orthogonality relations for the group matrix elements are:

1 X (µ) (ν) 1
Ti` (g)T̂sm (g) = δµν δis δ`m (9.8) eq:orthogmatrx
|G| nµ
g∈G

or, for a general compact group:

Z
(µ) (ν) 1
Ti` (g)T̂sm (g)[dg] = δµν δis δ`m (9.9) eq:orthogmatrx
G nµ

– 24 –
 where the Haar measure [dg] is normalized to unit volume. Recall that
(ν) −1,tr
T̂sm (g) = (T (ν) (g))sm (9.10)

is the matrix representation on the dual space.

(µ)
Note: If the matrix reps Ti` are unitary then T (g)tr,−1 = T (g)∗ so we can rewrite
this in a nicer form: Define
(µ) (µ)
φi` ≡ nµ1/2 Ti` (9.11)
then we have:

1 X (µ) ∗
φi` (g)(φ(ν)
sm (g)) = δµν δis δ`m (9.12) eq:onrels
|G|
g∈G

Or, for a general compact group

Z
∗ (µ)
(φ(ν)
sm (g)) φi` (g)[dg] = δµν δis δ`m (9.13) eq:onrels
G

(µ)
Interpretation: The vectors φij form an orthonormal set of vectors in RG .
Below we will apply this to the decomposition of the regular representation into irreps.

10. Decomposition of the Regular Representation

sec:sDRR
(µ)
As before we let Tm` (g) be matrix elements of the distinct irreps of G.
Now recall our important conceptual step forward: We may regard these matrix ele-
(µ)
ments as complex-valued functions on G. That is Tm` are vectors in RG .
Moreover, as we showed in the example surrounding (7.9) the linear span of functions
(µ) (µ)
Ri ≡ Span{Tij }j=1,...,nµ (10.1)

is an invariant subspace of RG , considered as the RRR. Here we are holding µ, i fixed.

Indeed, we have:
function on G function on G
z }| { X nµ z}|{
(µ) (µ) (µ)
R(g) · Tij = Tsj (g) Tis (10.2)
s=1 | {z }
matrix element for RG

We learn two things:

(µ)
1. Ri is an invariant subspace, as claimed.
2. Moreover, the matrix elements are those of the rep T (µ) ! Thus:
(µ) ∼
Ri = T (µ) (10.3)

independent of the value of i!

It follows that RG is highly reducible. Indeed it contains invariant subspaces corre-
sponding to each and every irrep of G. In fact, since the above holds for each i = 1, . . . , nµ
the same irrep occurs nµ times, where nµ = dimT (µ) .

– 25 –
 Let us define

µ n (µ)
W := ⊕µ ⊕i=1 Ri ⊂ RG
z nµ }|times
{ (10.4)
W ∼ ⊕
= µ T µ
⊕ · · · ⊕ T µ

Note:
(µ)
• W = Span{Tij }.
(µ)
• By the orthogonality relations the Tij are linearly independent. Thus we really do
have a direct sum .

Exercise
(µ)
Show that the set of functions Tij with µ, j held fixed forms a copy of the irrep T (µ)
under the left regular rep. More on this below.

At this point the possibility remains that W is a proper subspace of RG . We will now
(µ)
show that in fact W = RG exactly, by showing that Tij span RG .
To do this recall the vector space RG is an inner product space by the rule

1 X ∗
(φ1 , φ2 ) ≡ φ1 (g)φ2 (g) (10.5) eq:innproda
|G|
g∈G

Recall:
1. RG is a unitary representation, wrt this product.
2. The normalized matrix elements φµij are orthonormal wrt this inner product. This
follows from (9.13).
By the proposition 7.2.1 we can use the inner product to decompose

RG ∼
= W ⊕ W⊥ (10.6)

so W ⊥ is also a unitary representation. Moreover, again by 7.2.1 it can be reduced to a

sum of irreducible representations, T (µ) .
Suppose

Span{fi (g)}i=1,...,nµ ⊂ W ⊥ (10.7)

is a subspace transforming in a rep equivalent to T µ . Then, by definition:

– 26 –
 Tsiµ (h)fs (g)
X
(R(h)fi )(g) =
s
⇒
Tsiµ (h)fs (g)
X
fi (gh) =
s
(10.8)
⇒
Tsiµ (g −1 g 0 )fs (g)
X
fi (g 0 ) =
s
µ µ 0
X
= Tsk (g −1 )Tki (g )fs (g)
s,k

where we have set g 0 = gh.

Therefore, we have the equality of functions on G:
µ
X
fi (·) = αk Tki (·)
k
X µ  (10.9)
−1
αk = Tsk (g )fs (g) any g∈G
s

That is, fi is a linear combination of the functions φµij . But this contradicts the
assumption that
{fi } ⊂ W ⊥ (10.10)
so W ⊥ = 0. ♠.
Thus, we have arrived at the decomposition of the RRR into its irreducible pieces:

z nµ }|
times
{
∼ µ
RG = ⊕µ T ⊕ · · · ⊕ T µ
(10.11) eq:rrrep

But we are not quite done. To arrive at the most beautiful statement we should study
RG as a representation under the the left regular rep. So we consider the linear subspaces
where we hold µ, j fixed and consider the span:
(µ) (µ)
Lj ≡ Span{φij }i=1,...,nµ (10.12)

This forms a copy of the dual rep:

(µ) (µ) (µ)
X
(L(g) · φij )(h) = Tis (g −1 )φsj (h)
s
X (µ) (µ)
(10.13)
= T̂si (g)φsj (h)
s

so the span of the matrix elements of the rep µ

(µ)
Span{φij }i,j=1,...,nµ (10.14)

– 27 –
forms the rep T̂ (µ) ⊗ T (µ) of GL × GR . Finally, we can identify

T̂ (µ) ⊗ T (µ) ∼
= End(T (µ) ) (10.15)

This is a representation of GL × GR of dimension n2µ with action

−1
T (gL , gR ) · Φ = T (µ) (gL )ΦT (µ) (gR ) (10.16)

We have finally arrived at the beautiful :

Theorem 10.1 Let G be a finite group. The matrix elements φµij of the distinct irreps of
G decompose RG into irreps of GLeft × GRight :

RG ∼
= ⊕µ T̂ (µ) ⊗ T (µ) ∼
= ⊕µ End(T (µ) ) (10.17)

Moreover, if we use matrix elements of unitary irreps in an ON basis then the φµij form an
ON basis for RG and
RG ∼= ⊕µ T (µ) ⊗ T (µ) (10.18)

Corollary. Comparing dimensions we have the beautiful relation:

X
|G| = n2µ (10.19) eq:grouporder
µ

Remarks:
• This theorem generalizes beautifully to all compact groups and is known as the
Peter-Weyl theorem.
• This leaves the separate question of actually constructing the representations of the
finite group G. Until recently there has been no general algorithm for doing this.
Recently there has been a claim that it can be done. See
Vahid Dabbaghian-Abdoly, ”An Algorithm for Constructing Representations of Finite
Groups.” Journal of Symbolic Computation Volume 39, Issue 6, June 2005, Pages 671-688

Exercise
Show that the number of distinct one-dimensional representations of a finite group G
is the same as the index of the commutator subgroup [G, G] in G. 4

************************************ ***************************************
MATERIAL FROM MATHMETHODS 511 2015:
For now, we just content ourselves with the statement of the theorem for G a finite
group:
4
Hint: A one-dimensional representation is trivial on [G, G] and hence descends to a representation of
the abelianization of G, namely the quotient group G/[G, G].

– 28 –
Theorem: Let G be a finite group, and define an Hermitian inner product on L2 (G) =
Map(G, C) by
1 X ∗
(Ψ1 , Ψ2 ) := Ψ (g)Ψ2 (g) (10.20)
|G| g 1

Then let {Vλ } be a set of representatives of the distinct isomorphism classes of irreducible
(λ)
unitary representations for G. For each representation Vλ choose an ON basis wµ , µ =
λ (g) defined by
1, . . . , nλ := dimC Vλ . Then the matrix elements Dµν
nλ
X
ρ(g)wν(λ) = λ
Dµν (g)wµ(λ) (10.21)
µ=1

form a complete orthogonal set of functions on L2 (G) so that

1 λ1 ,λ2
(Dµλ11 ν1 , Dµλ22 ν2 ) = δ δµ1 ,µ2 δν1 ,ν2 (10.22) eq:PW-2
nλ

Idea of proof : The proof is based on linear algebra and Schur’s lemma. The normalization
constant on the RHS of (10.22) is easily determined by setting λ1 = λ2 and ν1 = ν2 = ν
and summing on ν, and using the hypothesis that these are matrix elements in a unitary
representation. The relation to (6.22) is obtained by noting that the linear transformations
(λ)
T = eνµ , given by matrix units relative to the basis wµ form a basis for End(Vλ ). ♠

Example 1: Let G = Z2 = {1, σ} with σ 2 = 1. Then the general complex valued-function

on G is specified by two complex numbers (ψ+ , ψ− ) ∈ C2 :

Ψ(1) = ψ+ Ψ(σ) = ψ− (10.23)

This identifies Map(G, C) ∼

= C2 as a vector space. There are just two irreducible represen-
tations V± ∼= C with ρ± (σ) = ±1, because for any representation ρ(σ) on a vector space
V we can form orthogonal projection operators P± = 12 (1 ± ρ(σ)) onto direct sums of the
irreps. They are obviously unitary with the standard Euclidean norm on C. The matrix
elements give two functions on the group D± :

D+ (1) = 1 D+ (σ) = 1 (10.24)

D− (1) = 1 D− (σ) = −1 (10.25)

(Here and in the next examples when working with 1×1 matrices we drop the µν subscript!)
The reader can check they are orthonormal, and they are complete because any function
Ψ can be expressed as:
ψ+ + ψ− + ψ+ − ψ− −
Ψ= D + D (10.26)
2 2

Example 2: We can generalize the previous example slightly by taking G = Z/nZ =

hω|ω n = 1i. Let us identify this group with the group of nth roots of unity and choose

– 29 –
a generator ω = exp[2πi/n]. Since G is abelian all the representation matrices can be
simultaneously diagonalized so all the irreps are one-dimensional. They are:
V = C and ρm (ω) = ω m where m is an integer. Note that m ∼ m + n so the set of
irreps is again labeled by Z/nZ and in fact, under tensor product the set of irreps itself
forms a group isomorphic to Z/nZ.
The matrix elements in the irrep (ρm , V ) are
mj
D(m) (ω j ) = ω mj = e2πi n (10.27)

Now we can check that indeed

1 X (m1 )
(D (g))∗ D(m2 ) (g) = δm1 −m2 =0modn (10.28)
|G|
g∈G

The decomposition of a function Ψ on the group G is known as the discrete Fourier trans-
form.

Remark: The theorem applies to all compact Lie groups. For example, when G = U (1) =
{z||z| = 1} then the invariant measure on the group is just −i dz dθ iθ
z = 2π where z = e :
Z 2π
dθ
(Ψ1 , Ψ2 ) = (Ψ1 (θ))∗ Ψ2 (θ) (10.29)
0 2π
Now, again since G is abelian the irreducible representations are 1-dimensional and the
unitary representations are (ρn , Vn ) where n ∈ Z, Vn ∼
= C and

ρn (z) := z n (10.30)

Now, the orthonormality of the matrix elements is the standard orthonormality of einθ and
the Peter-Weyl theorem specializes to Fourier analysis: An L2 -function Ψ(θ) on the circle
can be expanded in terms of the matrix elements of the irreps:
X
Ψ= Ψ̂n D(n) (10.31)
Irreps ρn

When applied to G = SU (2) the matrix elements are known as Wigner functions or
monopole harmonics. They are the matrix elements
j
Dm L ,mR
(g) := hmL |ρj (g)|mR i (10.32)

in the standard ON basis of the unitary spin j representation diagonalizing the diagonal
subgroup of SU (2). So
1 3
j = 0, , 1, , . . . , mL , mR ∈ {−j, −j + 1, . . . , j − 1, j} (10.33)
2 2
Recall that SU (2) ∼= S 3 as a manifold. Using the standard volume form, with unit vol-
ume we can define L2 (SU (2)). The entire theory of spherical harmonics and Legendre
polynomials is easily derived from basic group theory. ♣MAKE GOOD
ON THIS CLAIM
BELOW. ♣

************************************ ***************************************

– 30 –
 11. Fourier Analysis as a branch of Representation Theory

11.1 The irreducible representations of abelian groups

Suppose G is abelian. Let T be an irrep of G. Fix g0 ∈ G. Then

T (g0 )T (g) = T (g0 g) = T (gg0 ) = T (g)T (g0 ) (11.1)

Therefore, by Schur’s lemma:

T (g0 ) = λ(g0 )1 (11.2)
Therefore, all representation matrices of irreducible representations are proportional
to 1 ⇒

All finite-dimensional irreducible complex representations of an abelian group are 1-dimensional.

This is just simultaneous diagonalization of commuting operators. Note that the ad-
jective “complex” is essential here. Consider SO(2) acting on R2 .
As we mentioned before noncompact groups can have indecomposable representations,
e.g. R has the representation !
x 1
x→ (11.3)
0 x
Note that this is not unitary.

11.2 The character group

Notice that the space of 1 dimensional irreps of any group G is itself a group under ⊗
product:
T1 ⊗ T2 (g) = T1 (g)T2 (g) (11.4)
The identity is the trivial representation T0 (g) = 1, and inverses exist

(T )−1 (g) = (T (g))−1 (11.5)

By the same reasoning the space of 1-dimensional unitary irreps of a group is a group.
Note that in this case (T )−1 (g) = T ∗ (g) and T (g) is always represented by a phase.

Definition The space of unitary irreps of a group G is called the unitary dual and is
denoted by Ĝ.
In the case when G is abelian, the unitary dual Ĝ is itself a group. It is sometimes
called the Pontryagin dual group.
Examples
• Finally, let us consider Zn thought of as multiplicative nth roots of unity. For any
integer s we can define a representation by taking

T (s) (ω) = ω s (11.6) eq:srep

where ω is any nth root of unity. These are clearly unitary irreps. Note that

T (s+n) = T (s) (11.7)

– 31 –
and
T (s1 ) ⊗ T (s2 ) ∼
= T (s1 +s2 ) (11.8)
so the dual group is another copy of Zn :

cn ∼
Z = Zn (11.9)

Note that the relation (10.19) is easily verified:

n = 12 + 12 + · · · + 12 (11.10)

• The theory we are discussing in this section extends to locally compact abelian groups
such as G = R. The unitary irreps are specified by a “momentum” k ∈ R.

T (k) : x → eikx (11.11)

This is a rep because T (k) (x + y) = T (k) (x)T (k) (y). It is unitary for k real.
Now notice that
0 0
T (k) ⊗ T (k ) = T (k+k ) (11.12)
so we conclude:
b∼
R =R (11.13)
• G = Z. Unitary irreps of Z are labelled by a real number θ:

T (θ) : n → einθ (11.14)

Note that

T (θ) = T (θ+2π)
0 0 (11.15) eq:unirri
T (θ) ⊗ T (θ ) = T (θ+θ )

So the space of representations is parametrized by θ mod 2π, or, equivalently by

eiθ ∈ U (1). Moreover (11.15) shows that

b∼
Z = U (1) (11.16) eq:dlgrp

• G = R/Z ∼
= U (1). Now the unitary irreps are labelled by n ∈ Z:

T (n) : eiθ → e2πinθ (11.17) eq:unirrii

and we check:

0 0
T (n) ⊗ T (n ) = T (n+n ) (11.18) eq:unirriii

so

[
U (1) ∼
=Z (11.19) eq:dlgrpp

– 32 –
 • G = Λ ⊂ Rd , a lattice. Given a vector ~k ∈ Rd we can define a unitary irrep by:
~ ~
T (k) : ~n → e2πik·~n (11.20)

where we use the standard inner product on Rd .

~ ~0 ~ ~0
As before T (k) ⊗ T (k ) = T (k+k ) . Note however that
~ ~ ~
T (k+b) = T (k) (11.21)

if ~b is in the dual lattice:

Λ∗ = {~b ∈ Rd |∀~n ∈ Λ, ~b · ~n ∈ Z} (11.22)

Thus, the space of unitary irreps of Λ is a compact d-dimensional torus:

b∼
Λ = Rd /Λ∗ (11.23) eq:torirep

thus generalizing Z
b = U (1).
• . G = Rd /Λ is a d-dimensional torus.

Remarks Note that in all the above examples,

G =G (11.24)
bb

This is a general fact for abelian groups.

11.3 Fourier duality

Let us now return to Theorem 10.1 decomposing RG in terms of the irreducible repre-
sentations through the matrix elements of unitary irreps and apply it to finite abelian
groups:
• G = Z2 = {1, σ}. A function f : G → C is specified by its two values

f (1) = f0 f (σ) = f1 (11.25)

The unitary irreps are just:

T (0) (1) = 1 T (0) (σ) = 1

(11.26)
T (1) (1) = 1 T (1) (σ) = −1

Clearly:
f0 + f1 (0) f0 − f1 (1)
f= T + T (11.27)
2 2
Let us generalize this:
•
G = Zn ∼ = {1, ω, ω 2 , . . . , ω n−1 } where ω = e2πi/n . We worked out the unitary irreps
above.

– 33 –
 The orthogonality relations on the irreducible matrix elements (11.6) flow from:

1 X (j) ` 1 X 2πi(j`−k`)/n
T (ω )(T (k) (ω ` ))∗ = e
n n (11.28)
` `
= δj,k

where j, k are understood modulo n.

Now, according to the theorem, we can decompose any function on G = Zn as:

n−1
X
f= fˆj T (j) (11.29)
j=0

and we see that

1X
fˆj = f (ω s )(T (j) (ω s ))∗
n s
(11.30)
1X
= f (ω s )e−2πisj/n
n s

is the discrete Fourier transform.

In fact, our discussion generalizes to all (locally compact5 abelian groups including
infinite groups like R, Z, U (1). The main new ingredient we need is integration over the
1 P
group, which is the analog of |G| g.
For these groups the integration of a function f ∈ F un(G) is the obvious one:

Z Z +2π
dθ
f (g)dg ≡ f (θ)
G=U (1) 0 2π
Z X
f (g)dg ≡ f (n) (11.31) eq:intgroupp
G=Z n∈Z
Z Z +∞
f (g)dg ≡ dxf (x)
G=R −∞

With this in mind the above Theorem 10.1 becomes the following statements:
• Let us begin with G = (R, +) then Ĝ = (R, +). Unitary irreps are labelled by k ∈ R:

1
T (k) (x) = √ e2πikx (11.32)
2π
The orthogonality relations are:
Z
dxT (k) (x)(T (`) (x))∗ = δ(k − `) (11.33)
R

5
A Hausdorff space is called locally compact if every point has a compact neighborhood. All the abelian
groups we will meet in this course are locally compact. **** GIVE AN EXAMPLE OF A non-locally-
compact abelian group ***

– 34 –
 The statement that any L2 function on the group R can be expanded in the matrix
elements of its unitary irreps is that statement that any L2 function f (x) can be expanded:
Z
f (x) = dk fˆ(k)T (k) (x) (11.34)
R
b

So we recognize the usual Fourier expansion.

Of course, since R̂ ∼
= R we have the inverse relation:
b
Z
fˆ(k) = dxf (x)Tb(x) (k) (11.35)
R

Of course, there remarks generalize to Rd .

•
We showed that Z and U (1) are dual groups. Thus, an arbitrary (L2 ) function f (θ)
on G = U (1), can be decomposed into a sum of irreps of U (1):
X
f (θ) = fˆ(n)T (n) (θ) (11.36) eq:foursum
n∈Z
This is the usual Fourier series. and we recognize the Fourier decomposition. The
orthogonality relations on matrix elements read:
Z 2π
dθ (n)
(T (θ))∗ T (m) (θ) = δn,m (11.37) eq:orthogrels
0 2π
Conversely, an arbitrary L2 function fˆ(n) on the abelian group G = Z can be decom-
posed into a sum over the irreps T (θ) of Z, so we have the expansion:
Z 2π
dθ
fˆ(n) = f (θ)T (θ) (n) (11.38) eq:invfour
0 2π
The orthogonality relations read:

X
(T (θ1 ) (n))∗ T (θ2 ) (n) = δperiodic (θ1 − θ2 )
n∈Z
X (11.39) eq:orthogrelsa
≡ δ(θ1 − θ2 − 2πm)
m∈Z

In summary, we have the

Theorem Let G be a locally compact group. Then there is a natural isomorphism

∼
L2 (G)
= L2 (Ĝ) which expands any function on G in terms of the unitary irreps of G:
Z
f (g) := dχχ(g)∗ fˆ(χ) (11.40) eq:svnthm
Ĝ
Moreover, this isomorphism is an isometry:
Z Z
dgf1∗ f2 = dχfˆ1∗ fˆ2 (11.41)
G Ĝ

– 35 –
Remarks
• The functional analysis of L2 (G) is often called “nonabelian harmonic analysis.” For
more about this point of view see,
A.A. Kirillov, Elements of the Theory of Representations, Springer-Verlag, 1976.

11.4 The Poisson summation formula

Let us return to the orthogonality relations for the irreps of Λ and Rd /Λ∗ .
These imply a beautiful formula known as the Poisson summation formula.

Suppose f ∈ S(R) (the Schwarz space of functions of rapid decrese for | x |→ ∞).
Define the Fourier transform:
Z ∞
fˆ(k) = f (y)e−iky dy . (11.42)
−∞

Then:

Theorem [Poisson summation formula].

X X
f (n) = fˆ(2πn) (11.43)
n∈Z n∈Z

Proof: The orthogonality relations tell us that

X X
e2πint = δ(k − t) (11.44) eq:basicpsf
nZ kZ

Now integrate both sides against f (t). ♠

Remarks
• If it possible to give other proofs of this formula. We will content ourselves with its
being a consequence of the orthogonality relations.
• The rapid decrease condition can be relaxed to

const
| f (x) | + | f 0 (x) | + | f 00 (x) | ≤ (11.45)
1 + x2

Remark The above result has a beautiful generalization to arbitrary lattices in Euclidean
space Rd . The generalized Poisson summation formula states that:
X X
e2πi~v·~x = δ (d) (~x − ~l) (11.46)
~v ∈Λ ~l∈Λ∗

or - in terms or functions

– 36 –
 X Z X Z
~ ~t ~
X
−2πim·
f (~n) = e f (t)dt = e+2πim·
~ t
f (t)dt (11.47) eq:poissonsumm
Rd Rd
n∈Z d
~ m∈Z
~ d m∈Z
~ d

That is:
X X
f (~v ) = fˆ(~l) . (11.48)
~v ∈Λ ~l∈Λ∗

Remarks
• Since people have different conventions for the factors of 2π in Fourier transforms it is
hard to remember the factors of 2π in the PSF. The equation (11.44) has no factors of 2π.
One easy way to see this is to integrate both sides from t = −1/2 to t = +1/2.
• One application of this is the x-ray crystallography: The LHS is the sum of scattered
waves. The RHS constitutes the bright peaks measured on a photographic plate.
• Another application is in the theory of elliptic functions and θ-functions described later.

Exercise
a.) Show that
r
X
−πan2 +2πibn 1 X − π(m−b)2
e = e a (11.49) eq:formtwo
a
n∈Z m∈Z

b.) If τ is in the upper half complex plane, and θ, φ, z are complex numbers define the
theta function

2 +2πi(n+θ)(z+φ)
X
ϑθ,φ z = eiπτ (n+θ) (11.50) eq:thefun
n∈Z

(usually, θ, φ are taken to be real numbers, but z is complex).

Using the Poisson summation formula show that

θ −z −1 2 −φ
ϑ[ ]( | ) = (−iτ )1/2 e2πiθφ eiπz /τ ϑ[ ](z|τ ) (11.51) eq:esstmn
φ τ τ θ

11.5 Application: Bloch’s Theorem in Solid State Physics

Consider an electron moving in a crystalline material. Atoms or ions live on a lattice

Λ ⊂ R3 (11.52)

and an electron (in the single-electron approximation) satisfies the Schrödinger equation

~2 2
− ∇ ψ + V (~x)ψ = Eψ. (11.53)
2m

– 37 –
Now, V (~x) will be complicated in general, but we do know that

V (~x + ~a) = V (~x) (11.54)

∀~a ∈ Λ. Therefore the energy eigenspace VE is some representation of the lattice translation
group Λ.
As we have seen, the one dimensional reps of Λ are
~
~a → e2πik·~a (11.55)

and are labeled by ~k ∈ Λ̂ = Rn /Λ∗ . Therefore, VE can be written as a sum of reps labeled
by ~k:

VE = ⊕~k∈Λ̂ VE,~k (11.56) eq:decompvx

where eigenfunctions in VE,~k satisfy:

~
ψ(~x + ~a) = e2πik·~a ψ(~x) (11.57)
~
that is, ψ(~x) = e2πik·~x u(~x) where u is periodic. ~k, the “crystal momentum,” is only defined
modulo Λ∗ . It is properly an element of a torus.
Since we are working with infinite groups the sum on ~k in (11.56) in general must be
interpreted as an integral.

E*

1/a
k1 k2 k k k
3 4

Figure 3: Four irreps occur at energy E∗ in this one-dimensional band structure. For d = 2, 3
dimensional bandstructures there are infinity many k’s forming a surface. The surface at the Fermi
energy is the “Fermi surface.” fig:bandstruct

Remarks

– 38 –
 ~
• If we substitute ψ(~x) = e2πik·~x u(~x) into the Schrödinger equation we obtain an
elliptic eigenvalue problem on a compact space. It will therefore have a tower of discrete
eigenvalues. This leads to the band structure illustrated schematically in 3.
• In solid state physics the dual lattice is referred to as the reciprocal lattice. Physicists
generally choose a fundamental domain for the torus Rd /Λ∗ and refer to it as a Brillouin
zone.
• If the lattice Λ has a symmetry group then it will act on the torus. At fixed points
there is enhanced symmetry and bands will cross.

11.6 The Heisenberg group extension of Ŝ × S for an abelian group S

******
Change notation S → G below. Also need to explain that G × Ĝ → U (1) is a perfect
pairing.
******
Let S be a locally compact abelian group with a measure. Let Ŝ be the dual group.
There is a very natural Heisenberg extension of S × Ŝ defined by the cocycle to choose the
cocycle

1
c (s1 , χ1 ), (s2 , χ2 ) = (11.58) eq:cocyclechoi
χ1 (s2 )
whose antisymmetrization is

χ2 (s1 )
s (s1 , χ1 ), (s2 , χ2 ) = . (11.59) eq:cocyclechoi
χ1 (s2 )
This gives

1 → U (1) → Heis(S × Ŝ) → S × Ŝ → 1 (11.60) eq:heisext

There are two very natural representations of this group. First we consider H = L2 (S).
First of all L2 (S) is a representation of S. After all for s0 ∈ S we can define the
translation operator:

(Ts0 ψ)(s) := ψ(s + s0 ). (11.61) eq:transop

On the other hand, H is also a representation of the Pontrjagin dual group of characters,
denoted Ŝ. If χ ∈ Ŝ is a character on S then we define the multiplication operator Mχ on
H via

(Mχ ψ)(s) := χ(s)ψ(s). (11.62) eq:multop

Note that H is not a representation of S×Ŝ. This simply follows from the easily-verified
relation

Ts0 Mχ = χ(s0 )Mχ Ts0 . (11.63) eq:noncomm

However, H is a representation of Heis(S × Ŝ). It is defined so that

– 39 –
 (z, (s, χ)) → zTs Mχ (11.64) eq:heishom

where z ∈ U (1). Equation (11.64) defines a homomorphism into the group of invertible
operators on L2 (S).
Notice the complete symmetry in the construction between S and Ŝ (since the double-
dual gives S again). Thus, we could also provide a representation from Ĥ = L2 (Ŝ). The
two representations are in fact equivalent under Fourier transform:
Z
ψ̂(χ) := dsχ(s)∗ ψ(s) (11.65) eq:svnthmi

In fact, a theorem, the Stone-von Neumann theorem guarantees the essential unique-
ness of the unitary representations of Heisenberg groups:

Theorem If à is a central extension of a locally compact topological abelian group

A by U (1), then the unitary irreps of à where U (1) acts by scalar multiplication are in
1 − 1 correspondence with the unitary irreps of the center of Ã, where U (1) acts by scalar
multiplication.

Example 1 Consider L2 (Zn ). Regarding Zn as the integers modulo n we have

(T (s̄0 ) · f )(s̄) = f (s̄ + s̄0 ) (11.66)

while the dual group χt̄ has character χt̄ (s̄) = e2πist/n acts by multiplication.

b∼
Example 2 S = R and Ŝ = R. Denote elements q ∈ R and p ∈ R = R.

(T (p) · ψ)(q) = e2πipqθ ψ(q) (11.67)

(T (q0 ) · ψ)(q) = ψ(q + q0 ) (11.68)

Then
(T (q0 )(T (p) · ψ))(q) = e2πiθp(q+q0 ) ψ(q + q0 ) (11.69)
(T (p)(T (q0 ) · ψ))(q) = e2πiθpq ψ(q + q0 ) (11.70)
So T (q0 )T (p) = e2πipq0 T (p)T (q0 ). We recognize standard relations from quantum mechan-
ics with θ playing the role of Planck’s constant.

Example 3 S = Z and Ŝ = U (1) ... This generalizes to the duality between lattices and
tori.

Remarks
• Most free field theories can be interpreted in this framework, with proper regard
for the infinite-dimensional aspects of the problem. There is the interesting exception of
self-dual field theories which are more subtle.

– 40 –
12. Induced Representations
♣MATERIAL
HERE IMPORTED
Induced representations are certain representations defined by taking sections of associated FROM GMP 2010
ASSUMES
bundles. They form a unifying theme for a broad class of representations in physics: KNOWLEDGE OF
BUNDLES. NEED
1. Finite groups TO EDIT THOSE
PARTS AWAY. ♣
2. Compact groups: The Borel-Weil-Bott Theorem
3. Lorentz groups: The Wigner construction.
4. Representations of Heisenberg groups and coherent states
5. Representations of Loop Groups
Suppose H ⊂ G is a subgroup.
Suppose ρ : H → GL(V ) is a representation of H with representation space V .
Note that
H → G
↓π (12.1)
G/H
is a principal H bundle with total space P = G.
Therefore, according to what we have said above, using (ρ, V ) we can form the associ-
ated vector bundle
G ×H V (12.2)
Now, the key point is that Γ(G ×H V ) is also a representation of G. This representation
is called the induced representation and sometimes denoted:

Γ(G ×H V ) = IndG
H (V ) (12.3)

Using the grand tautology (??) we can equivalently define the induced representation
as:
Definition Let (V, ρ) be a representation of H. As a vector space, IndG
H (V ) is the set
of all smooth functions ψ : G → V satisfying the equivariance condition:

ψ(gh) = ρ(h−1 ) · ψ(g) (12.4)

for all h ∈ H, g ∈ G.
Note that, because G is a group we can multiply both on the left, and on the right.
This allows us to define a representation of G in a way that does not spoil the equivariance
condition:
Let us check that this is indeed a representation. If ψ ∈ IndG
H (ρ), and g ∈ G we define
T (g) · ψ as the new function whose values are given in terms of the old one by

(T (g) · ψ)(g 0 ) := ψ(g −1 g 0 ) (12.5) eq:indrepi

The inverse is needed so that we actually have a representation:

(T (g1 )T (g2 ) · ψ)(g 0 ) = (T (g2 ) · ψ)(g1−1 g 0 )

= ψ(g2−1 g1−1 g 0 ) (12.6)
= (T (g1 g2 ) · ψ)(g 0 )

– 41 –
 However, the action (12.5) manifestly does not interfere with the equivariance property
of ψ! In detail: if h ∈ H then

(T (g) · ψ)(g 0 h) = ψ(g −1 g 0 h) = ρ(h−1 )ψ(g −1 g 0 ) = ρ(h−1 )((T (g) · ψ)(g 0 )) (12.7) eq:explicit

So, T (g) · ψ is also equivariant.

Remarks: The associated vector bundle (G ×H V ) → G/H is an example of an

equivariant vector bundle. In general if E → M is a vector bundle over a space M with
a left G-action then E is an equivariant vector bundle, or G-vector bundle if there is an
action of G on the total space of E, linear on the fibers, which covers the G action on M .
Very often in physics (and mathematics) we are in a situation where there is a given G
action on a space M and we must lift the action to make a vector bundle equivariant. This
comes up, for example, in formulating Gauss laws.

12.1 Induced Representations of Finite Groups: Frobenius Reciprocity

The theory of induced representations is already interesting and nontrivial for G, H fi-
nite groups. In this case G → G/H is a finite cover (by H) of a discrete set of points.
Nevertheless, the general geometrical ideas apply.
Let Rep(G) denote the category of finite-dimensional representations of G. Mor-
phisms between W1 , W2 ∈ Rep(G) are linear transformations commuting with G, i.e.
G-intertwiners, and the vector space of all morphisms is denoted HomG (W1 , W2 ). The
induced representation construction defines a functor

Ind : Rep(H) → Rep(G). (12.8)

(We denoted this by IndGH before but H, G will be fixed in what follows so we simplify the
notation.) On the other hand, there is an obvious functor going the other way, since any
G-rep W is a foriori an H-rep, by restriction. Let us denote this “restriction functor”

R : Rep(G) → Rep(H) (12.9)

How are these two maps related? The answer is that they are “adjoints” of each other!
This is the statement of Frobenius reciprocity:

HomG (W, Ind(V )) = HomH (R(W ), V ) (12.10) eq:frobrecip

We can restate the result in another way which is illuminating because it helps to
answer the question: How is IndG
H (V ) decomposed in terms of irreducible representations
of G? Let Wα denote the distinct irreps of G. Then Schur’s lemma tells us that

IndG ∼ G
H (V ) = ⊕α Wα ⊗ HomG (Wα , IndH (V )) (12.11) eq:schurlemma

But now Frobenius reciprocity (12.10) allows us to rewrite this as

– 42 –
 IndG ∼
H (V ) = ⊕α Wα ⊗ HomH (R(Wα ), V ) (12.12) eq:frobrecipii

where the sum runs over the unitary irreps Wα of G, with multiplicity one.
The statement (12.12) can be a very useful simplification of (12.11) if H is “much
smaller” than G. For example, G could be nonabelian, while H is abelian. But the
representation theory for abelian groups is much easier! Similarly, G could be noncompact,
while H is compact. etc.

Proof of Frobenius reciprocity:

In order to prove (12.12) we note that it is equivalent (see the exercise below) to the
statement that the character of IndG
H (V ) is given by

X
χ(g) = χ̂(x−1 gx) (12.13) eq:charind
x∈G/H

where x runs over a set of representatives and χ̂ is the character χV for H when the
argument is in H and zero otherwise.
On the other hand, (12.13) can be understood in a very geometrical way. Think of the
homogeneous vector bundle G ×H V as a collection of points gj H, j = 1, . . . , n with a copy
of V sitting over each point. Now, choose a representative gj ∈ G for each coset. Having
chosen representatives gj for the distinct cosets, we may write:

g · gj = gg·j h(g, j) (12.14) eq:permcos

where j 7→ g · j is just a permutation of the integers 1, . . . , n, or more invariantly, a

permutation of the points in G/H.
Now let us define a basis for the induced representation by introducing a basis va for
the H-rep V and the equivariant functions determined by:

ψi,a (gj ) := va δi,j (12.15)

Geometrically, this is a section whose support is located at the point gi H. The equivariant
function is then given by
ψi,a (gj h) := ρ(h−1 )va δi,j (12.16)
Now let us compute the action of g ∈ G in this basis:

(g · ψi,a )(gj ) = ψi,a (g −1 gj )

= ψi,a (gg−1 ·j h(g −1 , j)) (12.17)
−1 −1
= δi,g−1 ·j ρ(h(g , j) ) · va

Fortunately, we are only interested in the trace of this G-action. The first key point is
that only the fixed points of the g-action on G/H contribute. Note that the RHS above
is supported at j = g · i, but if we are taking the trace we must have i = j. But in

– 43 –
 (23)

g H
1

(123) (123)
(13) (12)

g H (23) g H
3 2

(123)
(12) (13)

Figure 4: The left action of G = S3 on G/H. In fact, this picture should be considered as a
picture of a category, in this case, a groupoid. fig:indrepi

this case ggi = gi h(g, i) and hence g −1 gi = gi h(g, i)−1 so for fixed points we can simplify
h(g −1 , i) = h(g, i)−1 , and hence when we take the trace the contribution of a fixed point
ggi H = gi H is the trace in the H-rep of h(g, i) = gi−1 ggi , as was to be shown ♠

Remark: The Ind map does not extend to a ring homomorphism of representation
rings.
Example A simple example from finite group theory nicely illustrates the general idea.
Let G = S3 be the permutation group. Let H = {1, (12)} ∼ = Z2 be a Z2 subgroup. G/H
consists of 3 points. The left action of G on this space is illustrated in (12.5).
There are two irreducible representations of H, the trivial and the sign representation.
These are both 1-dimensional. Call them V (), with  = ±. Accordingly, we are looking
at a line bundle over G/H and the vector space of sections of G ×H V () is 3-dimensional.
A natural basis for the space of sections is given by the functions which are “δ-functions
supported at each of the three points”:

si (gj H) = δij
g1 H = (13)H = {(13), (123)}
(12.18) eq:basis
g2 H = (23)H = {(23), (132)}
g3 H = (12)H = {1, (12)}

These sections correspond to equivariant functions on the total space. The space of all
functions F : G → R is a six-dimensional vector space. The equivariance condition:

– 44 –
 F (12) = F (1)
F (123) = F (13) (12.19) eq:equicond
F (132) = F (23)

cuts this six-dimensional space down to a three-dimensional space.

We can choose a basis of equivariant functions by choosing 3 representatives g1 , g2 , g3
for the cosets in G/H and setting F i (gj ) = δij . Using such a basis the representation of
the group is easily expressed as a permutation representation.
In our example of G = S3 it is prudent to choose g1 = (13), g2 = (23), g3 = (12) so
that

(12)g1 = g2 (12)
(12)g2 = g1 (12)
(12)g3 = g3 (12)
(13)g1 = g3 (12)
(13)g2 = g2 (12) (12.20) eq:multout
(13)g3 = g1 (12)
(23)g1 = g1 (12)
(23)g2 = g3 (12)
(23)g3 = g2 (12)

From this one easily gets the induced representation

 
0  0
ρind (12) =   0 0 (12.21)
 
00 
 
00 
ρind (132) =   0 0 (12.22)
 
0  0
and so forth.
Now let us look at Frobenius reciprocity. The irreducible representations of G are
W () defined by (ij) → , and W2 defined by the symmetries of the equilateral triangle,
embedded into O(2):

!
−1 0
ρ2 (12) =
0 1
√ ! (12.23) eq:rhotwo
−√12 23
ρ2 (123) =
− 23 − 12

– 45 –
 As H = Z2 representations, we have W () ∼
= V () and

W2 ∼
= V (+1) ⊕ V (−1) (12.24)

Therefore

HomH (W (), V (0 )) = δ,0 R

(12.25)
HomH (W2 , V ()) = R

By Frobenius reciprocity we have

IndSS32 (V ()) = W () ⊕ W2 (12.26) eq:frobexple

Let us check this by computing characters from the geometric perspective.

The induced representation consists of functions on G/H valued in V (). The action
of g acts as a permutation representation on the support of the functions.
Therefore, we can compute the character of g in the induced representation by looking
at its action on fixed points:
X
χρind (g) = χV () (h(g, i)) (12.27)
F ix(g)

To find out how to decompose the representation IndSS32 (V ()) in terms of G = S3 irreps
it suffices to compute the character for g = (12) and g = (123). Now, g = (12) has exactly
one fixed point, namely g3 H and h(g, 3) = (12) for this element. Therefore,

χρind (12) = χV (12) =  (12.28)

On the other hand, g = (123) clearly has no fixed points, and therefore the character is
zero. It follows immediately that we have the decomposition (12.26).
*********************************************
*********************************************
INCORPORATE:
Now let us turn to induced representations:
Let G be a group and H a subgroup. Suppose that ρ : H → End(V ) is a representation
of the subgroup H. Then, as we have seen Map(G, V ) is canonically a G × H-space. To
keep the notation under control we denote a general function in Map(G, V ) by Ψ. Then
the left-action of G × H defined by declaring that for (g, h) ∈ G × H and Ψ ∈ Map(G, V )
the new function φ((g, h), Ψ) ∈ Map(G, V ) is the function G → V defined by:

φ((g, h), Ψ)(g0 ) := ρ(h) · Ψ(g −1 g0 h) (12.29)

for all g0 ∈ G. Now, we can consider the subspace of functions fixed by the action of 1 × H.
That is, we consider the H-equivariant functions which satisfy

Ψ(gh−1 ) = ρ(h)Ψ(g) (12.30) eq:EquivFun

– 46 –
for every g ∈ G and h ∈ H. Put differently: There are two natural left-actions on
Map(G, V ) and we consider the subspace where they are equal. Note that the space of such
functions is a linear subspace of Map(G, V ). We will denote it by IndG H (V ). Moreover, it
is still a representation of G since if Ψ is equivariant so is (g, 1) · Ψ.
The subspace IndG H (V ) ⊂ Map(G, V ) of H-equivariant functions, i.e. functions satis-
fying (12.30) is called the induced representation of G, induced by the representation V of
the subgroup H. This is an important construction with a beautiful underlying geometrical
interpretation. In physics it yields:

1. The irreducible unitary representations of space groups in condensed matter physics.

2. The irreducible unitary representations of the Poincaré group in QFT.

Example: Let us take V = C with the trivial representation of H, i.e. ρ(h) = 1. Then
the induced representation is the vector space of functions on G which are invariant under
right-multiplication by H. This is precisely the vector space of C-valued functions on the
j
homogeneous space G/H. For example, the invariant Wigner functions Dm L ,mR under
j
right-action by the diagonal U (1) subgroup of SU (2) are DmL ,0 (g). These descend to
functions on SU (2)/U (1) ∼= S 2 known (for j integral) as the spherical harmonics. The case
of V a trivial representation generalizes in a beautiful way: When (ρ, V ) is nontrivial the
induced representation is interpreted not as a space of functions on G/H but rather as a
vector space of sections of a homogeneous vector bundle over G/H determined by the data
(ρ, V ). See **** below.

Exercise
Prove (6.20).

Exercise ♣Physics 619, ch.

5, 2002 for more ♣
Let G be the symmetric group on {1, 2, 3} and let H = {1, (12)}. Choose a represen-
tation of H with V ∼ = C and ρ(σ) = +1 or ρ(σ) = −1.
a.) Show that in either case, the induced representation IndG
H (V ) is a three-dimensional
vector space.
b.) Choose a basis for IndG H (V ) and compute the representation matrices of the ele-
ments of S3 explicitly.

Exercise Relation to the Peter-Weyl theorem

– 47 –
 Take H to be the trivial group and V the trivial representation and explain the relation
of (12.12) to the Peter-Weyl theorem.

**************************************************
**************************************************

13. Representations Of SU (2)

♣TAKEN FROM

Let us now consider the space of C∞ sections Γ(Lk ) as a representation of SU (2). The GMP 2010 BUT
ALL MATERIAL
ASSUMING
right-action by U (1) is given by INDUCED REPRE-
3
g · eiχ := ge−iχσ (13.1) SENTATIONS,
BUNDLE
THEORY, AND
where on the RHS we have ordinary matrix multiplication. According to our general BOREL-WEIL-
BOTT IS
principle, the sections of Vk are the same as equivariant functions ψ : SU (2) → C such that SUPPRESSED ♣

3
ψ(ge−iχσ ) = e−ikχ ψ(g) (13.2) eq:eqvfunct

In order to analyze (13.2) we must recall the Peter-Weyl theorem for SU (2):

L2 (SU (2)) ∼
= ⊕j∈ 1 Z+ Dj ⊗ Dj (13.3)
2

as a representation of the Gleft × Gright . An orthonormal basis of functions is given by the

matrix elements of SU (2) in the unitary irreducible representations Dj . Recall these are
labeled by j = 0, 21 , 1, 23 , 2, · · · of dimension n = 2j + 1 = 1, 2, 3, 4, 5, . . . .
Using our parametrization of SU (2) in terms of Euler angles from Chapter 5 we have:

hj, mL | T (j) (g) | j, mR i ≡ Dm

j
L mR
(φ, θ, ψ)
1 3 1 2 1 3
= hj, mL |T (j) (e−i 2 φσ )T (j) (e−i 2 θσ )T (j) (e−i 2 ψσ )|j, mR i (13.4) eq:sutmtrx
−imL φ j
=e Pm L ,mR
(cos θ)eimR ψ

j
Here Pm L ,mR (x) is an associated Legendre polynomial and we have chosen the slightly

unconventional representation of antihermitian generators:

i
J k = − σk [J i , J j ] = ijk J k (13.5)
2
and diagonalized J 3 |j, mi = im|j, mi.
Thus subspace of L2 (SU (2)) satisfying the equivariance condition (13.2) is that spanned
by the D-functions with k = −2mR .
j
Thus, an ON basis of functions satisfying the equivariance condition is Dm L ,−k/2
, as
j, mL range over all possible values. Clearly, we must have j ≥ |k|/2, and for fixed j the
span of functions mL = −j, −j +1, . . . , +j form a representation of SU (2) and hence: From
this description it is clear that the induced representation is infinite dimensional, and is
given by a sum of representations of SU (2),

– 48 –
 SU (2)
IndU (1) (Vk ) ∼
= ⊕j 0 ≥ 1 |k| Dj
0
(13.6) eq:indrepsut
2

where the sum is over j 0 = 12 |k|, 12 |k| + 1, 12 |k| + 2, . . . .

Remarks
• To explain the complex conjugate in (13.6) recall that under the left regular repre-
sentation we have
j
(L(g0 ) · Dm L ,mR
j
)(g) := Dm L ,mR
)(g0−1 g)
= hj, mL |T (j) (g0−1 )T (j) (g)|j, mR i
X j ∗ (13.7)
j
= Dm0 mL (g0 ) Dm 0 m (g)
R
m0

• Let us check the Frobenius reciprocity statement:

0 0
HomSU (2) (Dj , Ind(Vk )) = HomU (1) (R∗ Dj , Vk ) (13.8)

The RHS is one-dimensional exactly for those values of j 0 given by 2j 0 − |k| ∈ 2Z+ .
• In the special case k = 0 we are talking about sections of the trivial line bundle.
But these are just complex valued L2 functions on S 2 . Indeed, in this case mR = 0
and the functions become independent of ψ and hence well-defined functions on S 2 rather
than sections of nontrivial line bundles. In this case the functions are known as spherical
harmonics and usually denoted:
`
Dm,0 = Y`,m (θ, φ) (13.9)

• If we were studying the quantum mechanics of a charged particle of electric charge

+1 moving on a sphere surrounding a magnetic monopole of magnetic charge k then the
SU (2)
Hilbert space of the particle would be Γ(Lk ) = IndU (1) (Vk ). Thus, we will sometimes refer
to vectors in this representation space as “states” or “wavefunctions.”
Now for k ≥ 0 let

Hk := {ψ : ψ(λu, λv) = λk ψ(u, v)} ⊂ C[u, v] (13.10)

be the vector space of homogeneous polynomials in two variables u, v of degree k.

Now, there is a natural basis of functions in Hk , given by by the monomials ψ` (u, v) =
u` v k−` , ` = 0, 1, . . . , `. The matrix elements in this basis can be computed from

!
u
(g · ψ` )(u, v) = ψ` (g −1 )
v
(13.11) eq:mtrxelem
= ψ` ((detg)−1 (g22 u − g12 v, −g21 u + g11 v))
= (detg)−k (g22 u − g12 v)` (−g21 u + g11 v)k−`

By expanding out using the binomial theorem and collecting terms one obtains explicit
matrix elements. We will do this momentarily.

– 49 –
 What we have learned so far is that the holomorphically induced representation from
B to SL(2, C) for Vk is the k + 1-dimensional representation which can canonically be
viewed as the space of homogeneous polynomials of degree k. [*** Redundant??? ***]
GL(2,C)
Now, since SU (2) ⊂ GL(2, C) we can restrict the functions in HolIndB (Vk ) to
∼
SU (2). Since SU (2) ∩ B = T = U (1) they will be functions in the induced representation
SU (2)
IndU (1) (Vk ). The monomial basis functions become

ψ` (g) = α` β k−` ` = 0, 1, . . . , k (13.12) eq:monbasis

SU (2)
Now, let us try to identify this with one of the Wigner functions in IndU (1) (Vk ).
In order to identify (13.12) with Wigner functions let us restrict the GL(2, C) repre-
sentation Hk for k ≥ 0 to SU (2). The representation remains irreducible. If we define j
by k = 2j then it is convenient to rename the basis slightly to

f˜m (u, v) ≡ uj+m v j−m (13.13) eq:hombasp

for m = −j, −j + 1, −j + 2, · · · , j − 1, j. Note that m increases in steps of +1 and

hence j ± m is always an integer even though j, m might be half-integer.
Now, for !
α −β̄
g= |α|2 + |β|2 = 1 (13.14)
β ᾱ

we continue the computation of the matrix elements (13.11) for this representation of
SU (2):

(g · f˜m )(u, v) := f˜m (ᾱu + β̄v, −βu + αv)

= (ᾱu + β̄v)j+m (−βu + αv)j−m
X j (13.15) eq:expbas
:= D̃m0 m (g)f˜m0
m0

Note that for g = exp[− 2i φσ 3 ] = exp[φJ 3 ] (with our unconventional normalization of

J 3 ) we have
g · f˜m = eimφ f˜m (13.16)
j
More generally, we can derive an explicit formulat for the matrix elements D̃m 0 m (g) as

functions on SU (2) by expanding out the two factors in (13.15) using the binomial theorem
and collecting terms:
  
j
X j + m j − m s j−m−t j+m−s
D̃m 0 m (g) = ᾱ α β̄ (−β)t (13.17) eq:xplcttldee
0
s t
s+t=j+m

Warning: This differs from standard Wigner functions by normalization factors. These
are important when dealing with unitary representations. We return to unitarity below.
j j
For the moment we merely remark that D̃m L m is proportional to DmL m .

– 50 –
 SU (2)
Now, we found earlier that IndU (1) (Vk ) is spanned by matrix elements with mR =
−k/2. In equation (13.17) this means we should put m = −j which forces s = 0 and hence
 
j j+m 2j
D̃m,−j = (−1) αj−m β j+m (13.18)
j+m
Up to a normalization constatn this is the explicit basis (13.12) we set out to identify.
Remarks
• If we consider a charged particle confined to a sphere and coupled to a Dirac monopole
of charge k we will see that quantization of the system leads to a Hilbert space
SU (2)
Γ(Lk ) = IndU (1) (Vk ). (13.19)

Since SU (2) is a global symmetry of the quantum system the Hilbert space must be a
representation of SU (2). Above we have given the decomposition of this Hilbert space in
terms of irreps.
• SU (2)/U (1) is also a Kahler manifold and therefore we will be able to introduce the
supersymmetric quantum mechanics with this target space with N = 2 supersymmetry.
If the particle is coupled to a gauge field on a line bundle of first chern class k (i.e. if
the particle is charged, with charge k) then the above representation has the physical
interpretation as the space of BPS states.
************************
***********************

13.0.1 Matrix Reps Of su(2)

As we have mentioned above, given a representation of a Lie group we can recover a
representation of the corresponding Lie algebra. The representations corresponding to Hn
with n = 2j are the following:
Working to first order in  we have:

T (eX ) · f = T (1 + X) · f = (T (1) + T (X)) · f = f + (T (X) · f )

for any X ∈ su(2). Taking X = X3 = 2i σ3 and evaluating the definition of the function on
the LHS evaluated on (u, v) we get:
 
1 1 i ∂ ∂
f (u − i u, v − i v) = f −  u −v f
2 2 2 ∂u ∂v
so we conclude:  
1 ∂ ∂
T (X3 ) · f = −i u −v f
2 ∂u ∂v
and, in particular,
iT (X 3 ) · f˜j,m = mf˜j,m
Similarly we have
i 1
f + T (X 1 ) · f = T (e 2 σ ) · f
i

∂ ∂
 (13.20)
=f − v +u f
2 ∂u ∂v

– 51 –
 i 1
since for g = e 2 σ we have α = 1 + O(2 ), β = − 2i . In just the same way we get:

i 2
f + T (X 2 ) · f = T (e 2 σ ) · f
1

∂ ∂
 (13.21)
=f − v −u f
2 ∂u ∂v

Let us define operators:

J ± ≡ iT (X 1 ) ± T (X 2 )

These act as

∂
J + · f˜j,m = u f˜j,m
∂v
= (j − m)f˜j,m+1
(13.22)
∂
J − · f˜j,m = v f˜j,m
∂u
= (j + m)f˜j,m−1

Or, as matrices:
 
0 1 0 0 ···
 0
 0 2 0 ···
 0 0 0 3 ···
J+ = (13.23)
 
eq:sutwomtrx
· · · ··· ··· ··· ···

 
 0 0 0 ··· 2j 
0 0 0 ··· 0

 
0 0 0 0 ···
 2j
 0 0 0 ···
 0 2j − 1 0 0 ···
J− = (13.24)
 
· · · · · · ··· ··· ···

 
 0 0 0 ··· 0 
0 0 ··· 1 0

Exercise
Check that this is a representation of su(2). It is different from what one finds in most
textbooks on quantum mechanics because it is not a unitary matrix representation. See
the next section.

*************************
*************************

– 52 –
13.1 Unitary structure
In Chapter 7 we observed that we can define an Hermitian metric on the holomorphic line
bundle H k such that the norm-squared of a section is
1
k s k2 = |s(z)|2 (13.25)
(1 + |z|2 )k

in the trivializations on U± . Therefore, we can define a unitary structure on the space of

holomorphic sections by declaring:
Z
hs, si := k s k2 vol (CP 1 ) (13.26)
CP 1

where
i dzdz̄
vol (CP 1 ) = (13.27)
2π (1 + |z|2 )2
is the unit volume form. We claim that the action of SU (2) on H 0 (CP 1 ; H k ) we described
above is unitary with respect to this metric.
To see this let us first note that we can identify the space Hk of homogeneous polyno-
mials in two variables of degree k with the space Pk of polynomials in a single variable of
degree ≤ k as follows:
If ψ(u, v) ∈ Hk then we map it to

s(z) := u−k ψ(u, v) (13.28)

Note that s(z) will be a polynomial in the variable z = v/u of degree ≤ k.

If we require that the map Hk → Pk is an intertwiner of SU (2) representations then
we arrive at the formula for the SU (2) action on Pk : It takes a polynomial s(z) to
 
αz − β
(g · s)(z) := (β̄z + ᾱ)k s (13.29)
β̄z + ᾱ

Now, the map

αz − β
z → z0 = (13.30)
β̄z + ᾱ
preserves the volume form vol (CP 1 ), and a small computation using the identity

1 |β̄z + ᾱ|2
= (13.31)
(1 + |z 0 |2 ) (1 + |z|2 )

shows that
hg · s, g · si = hs, si (13.32)
In this way H 0 (CP 1 ; H k ) becomes a Hilbert space on which SU (2) acts unitarily.
The ON basis corresponding to f˜m is
s
(2j + 1)!
ψj,m = z j−m (13.33)
(j + m)!(j − m)!

– 53 –
 j
and can be taken to be |j, mi in our definition above of the Wigner functions Dm L ,mR .

Exercise
j j
Find the proportionality factor between D̃m L ,mR and DmL ,mR .

************************
***********************
UNITARIZE REP OF J + AND J −
*************************
*************************

13.2 Lie algebra representations and differential operators

TO BE WRITTEN: Explain that
1. Generators are represented by differential operators.
2. The usual differential equations and recursion relations on special functions follow
from this point of view.

13.3 Coherent state formalism

The above discussion of representations of SU (2) is closely related to the subject of coherent
states.
One reference for this material is
A. Perelomov, Generalized Coherent States and Their Applications, Springer Verlag
1986
Another reference is:
K. Fuji, Introduction to coherent states and quantum information theory, arXiv:quant-
ph/0112090
One way to motivate this is to derive the representation of the Lie algebra in terms of
differential operators.
If !
+ 00
J = (13.34)
10
then actiong on Pk we represent
∂
J+ = − (13.35)
∂z
It is then natural to define coherent states:
1
|ζi := exp(ζJ+ )|j, −ji (13.36)
(1 + |ζ|2 )j
One can show that
1. X
|ζi = um (ζ)|j, mi (13.37)
m

– 54 –
with s
(2j)! ζ j+m
um (ζ) = (13.38)
(j + m)!(j − m)! (1 + |ζ|2 )j
******* IS SQRT CORRECT? ********
2. The inner product of two such polynomials is
¯ 2j
(1 + ξη)
hξ|ηi = (13.39)
((1 + |ξ|2 )j (1 + |η|2 )j

In particular |ζi is of unit norm.

3. If we regard polynomials in z of degree ≤ 2j as an irrep of SU (2) as above then we
may identify the state |ζi with the polynomial: ???

(1 + ζz)2j
f (z) = (13.40)
(1 + |ζ|2 )j

Examples:
1. The wavefunction of the oxygen molecule!
2. Wavefunctions of electrons in the presence of a magnetic monopole will take values
in this bundle.
3. Nice application of this coherent state formalism to the QHE in paper by Haldane.
Remarks:
1. The coherent states are good for taking the semiclassical limit of large j. Then the
support of |ζi is sharply peaked at a point on CP 1 . **** SHOW *****
2.
3. Relate infinite dimensional induced rep to the Schwinger representation in terms of
oscillators. *** What is this??? ***

14. Orbits of the Lorentz group and relativistic wave equations

♣OVERLAP HERE
WITH
14.1 Orbits, Representations, and Differential Equations PETER-WEYL
THEOREM IN
Now let us turn to how these orbits are related to some important differential equations in CHAPTER 4 ON
REPRESENTA-
field theory. TIONS. MOVE
THIS? ♣
As we have seen, since O(1, 1) acts on M1,1 , it follows that it acts on the space of fields
Map(M1,1 , κ), where κ = R or C for a real or complex-valued scalar field. For a scalar field
recall the action of A ∈ O(1, 1) on the space of solutions is

(A · Ψ)(x) := Ψ̃(x) := Ψ(A−1 x) (14.1)

Note the A−1 in the argument of Ψ in the second equality. This is necessary to get a
left-action of the group on the space of fields. If we use Ax then we get a right-action. ♣For the case of
complex scalar fields
Now, quite generally, if V is a representation space for G, and O ∈ End(V ) is an when A is in a
component of
invariant linear operator, i.e. an operator which commutes with the action of G, O(1, 1) that reverses
the orientation of
time we have the
ρ(g)O = Oρ(g) (14.2) choice of whether to
include complex
conjugation in the
action. This would
be adding a “charge
conjugation” action.
♣

– 55 –
then any eigenspace, say {v ∈ V |Ov = λv} will be a sub-representation of G.
Consider the operator
∂ µ ∂µ = −∂02 + ∂12 (14.3)
acting on the space of scalar fields. This is an example of an invariant operator, as one
confirms with a simple computation. It can be made manifest by writing

∂ µ ∂µ = −4∂+ ∂− (14.4)

where
∂ 1
∂± := ±
= (∂0 ∓ ∂1 ) (14.5)
∂x 2
The Klein-Gordon equation for a complex or real scalar field Ψ(x0 , x1 ) is

∂ µ ∂µ + m2 Ψ = 0


(14.6)

The space of fields satisfying the KG equation is a representation space of O(1, 1), by our
general remark above.
Now we relate the orbits of O(1, 1) to the representations furnished by solutions of the
KG equation:
For field configurations which are Fourier transformable we can write
Z
0 1
0 1
Ψ(x , x ) = dk0 dk1 ψ̂(k)eik0 x +ik1 x (14.7)

If the field Ψ is on-shell then ψ̂(k) must have support on set

{k : (k0 )2 − (k1 )2 = m2 } (14.8) eq:MassShell

that is:
(k 2 + m2 )ψ̂(k) = 0 (14.9)
Thus, the support in Fourier space is an orbit. In physics the orbit with k0 > 0 is
called the mass-shell.
For example, suppose that m2 > 0. Then on the orbit with k 0 > 0 we have k =
m(cosh θ, sinh θ) and we can write the general complex-valued solution as

ψ̂(k) = δ(k 2 + m2 )ā, (14.10)

where the amplitude ā should be regarded as a complex-valued function on the orbit. Then
a complex-valued solution on the KG equation is generated by such a function on the orbit
by:
Z
Ψ(x) = d2 kδ(k 2 + m2 )āeik·x
Z
dk
= p 1 ā(k1 )eik·x (14.11) eq:MomentumMod
2 2
R 2 k1 + m
Z ∞
1 0 1
= dθā(θ)eim(x cosh θ+x sinh θ)
2 −∞

– 56 –
where we chose a convention designation of the argument of the function ā on the orbit.
Thus, we can identify a space of solutions to the KG equation with a space of functions
on an orbit.
Now recall that the space of functions on an orbit is a special case of an induced
representation. As we will soon see, induced representations are the right concept for
generalization to other representations of the Lorentz group.

Remark 1: The components of the Lorentz group: If we use a particular orbit to generate
representations we only represent the subgroup of the Lorentz group which preserves that
orbit. Thus, for the orbit with m > 0 and k0 > 0 we can include the parity-reversing com-
ponent, but not the time-reversing component. Of course, there is a similar representation
of SO0 (1, 1) q P · SO0 (1, 1) given by the hyperbola with k0 < 0. If we wish to represent the
full O(1, 1) then we must include both orbits. Note that, if we wish to use real solutions of
the KG equation we must include both hyperbolae and there is a reality condition relating
the Fourier modes: (ā(k))∗ = a(−k)

Remark 2: The representation can be made into a unitary representation. See below.

Exercise ♣This, and the

concept of an
Show that an invariant linear operator O ∈ End(V ) on a representation space V of G invariant operator
really goes above in
is a fixed point of the G × G action on End(V ). the general section
on functions on
groups. ♣

14.2 The massless case in 1 + 1 dimensions

Having phrased things this way it is natural to ask what relativistic wave equations corre-
spond to amplitudes supported on the light-like orbits.
In the massless case m2 = 0 the general solution of the KG equation is easily written
as
Ψ = ψL (x+ ) + ψR (x− ) (14.12)
Solutions of the form fR = 0 are called left-moving waves because, as time evolves forward,
the profile of the function fL moves to the left on the x1 -axis. Similarly, solutions of the
form fL = 0 are called right-moving waves.

Remark: In the massless case there is a “more primitive” relativistic wave equation
which is first order, and whose solutions are always solutions of the massless KG equa-
tion. Namely, we can consider the separate equations

∂+ Ψ = 0 (14.13) eq:anti-self-d

∂− Ψ = 0 (14.14) eq:self-dual

Note that these equations are themselves Lorentz invariant, even though the operators
∂± are not invariant differential operators. Solutions to (14.13) are right-moving scalar

– 57 –
fields and solutions to (14.14) are left-moving scalar fields. Such on-shell scalar fields
are also known as chiral scalar fields. Equations (14.13) and (14.14) are notable in part
because they play an important role in string theory and conformal field theory. It is
also interesting to note that it is quite subtle to write an action principle that leads to
such equations of motion. They are also called the anti-self-dual and self-dual equations
of motion, respectively because, if we choose the orientation dx ∧ dt then the Hodge star
operation is ∗dt = dx and ∗dx = dt and hence ∗(dx± ) = ±dx± . Therefore if ∂± Ψ = 0 its
“fieldstrength” F = dΨ satisfies ∗F = ∓F .
The action of P, T ∈ O(1, 1) on a real scalar field is given by:

(P · Ψ)(x, t) := Ψ(−x, t)
(14.15)
(T · Ψ)(x, t) := Ψ∗ (x, −t)

The KG equation is separately P and T invariant, but the (anti)-self-dual equations are
not. Nevertheless, the latter equations are P T invariant. This is a special case of the
famous CPT theorem:
Roughly speaking, mathematical consistency implies that if a physical theory is in-
variant under group transformations in the neighborhood of the identity then it must be
invariant under the transformations in the full connected component of the identity. But
this does not mean the theory is invariant under disconnected components such as the com-
ponents containing P and T . As a matter of fact, Nature chooses exactly that option in
the standard model of the electroweak and strong interactions. However, if we also assume
that there is a certain relation to “analytically continued” equations in Euclidean signature
then, since P T is in the connected component of the identity of O(2), such theories must
in fact be P T invariant. ♣There is much
more to discuss
about the nature of
14.3 The case of d dimensions, d > 2 the analytic
continuation
implied here. In
Now consider Minkowski space M1,d−1 with d > 2. The nature of the orbits is slightly what sense is SO(2)
and “analytic
different. continuation” of
SO(1, 1)? It does
not do simply to
1. For λ2 > 0 we can define make the boost pure
imaginary unless
x± → z, z̄. ♣
O+ (λ) = {x|(x0 )2 − (~x)2 = λ2 & sign(x0 ) = sign(λ)} (14.16)

By the stabilizer-orbit theorem we can identify this with

SO0 (1, d − 1)/SO(d − 1) (14.17)

by considering the isotropy group at (x0 = λ, ~x = 0). See Figure 5(a).

2. For µ2 > 0 we can define

O− (λ2 ) = {x|(x0 )2 − (~x)2 = −µ2 } (14.18)

By the stabilizer-orbit theorem we can identify this with

SO0 (1, d − 1)/SO0 (1, d − 2) (14.19)

– 58 –
Figure 5: Illustrating orbits of the connected component of the identity in O(1, 3). In (a) the top
and bottom hyperboloids are separate orbits, and if we include time-reversing transformations the
orbits are unions of the two hyperboloids. In (b) there are three orbits shown with x0 > 0 x0 < 0
(the future and past, or forward and backward light cones), and the orbit consisting of the single
point. In (c), once x2 has been specified, there is just one orbit, for d > 2. fig:LorentzOrb

by considering the isotropy group at x = (x0 = 0, x1 = 0, . . . , xd−2 = 0, xd−1 = µ).

The sign of µ does not distinguish different orbits for d > 2 because the sphere S d−2
is connected. See Figure 5(c).

3.
O± = {x|x2 = 0 & sign(x0 ) = ±1} (14.20)
Vectors in this orbit are of the form (x0 , |x0 |n̂) where n̂ ∈ S d−2 ⊂ Rd−1 and the sign
of x0 is invariant under the action of the identity component of O(1, 3). (Show this!).
Note that, for d = 2 the sphere S 0 has two disconnected components, leading to
left- and right-movers. But for d > 2 there is only one component. We can think of
n̂ ∈ S d−2 as parametrizing the directions of light-rays. That is, the point where the
light ray hits the celestial sphere. In one spatial dimension, a light ray either moves
left or right, and this is a Lorentz-invariant concept. In d − 1 > 1 spatial dimensions,
we can rotate any direction of light ray into any other. See Figure 5(b). One can
show that these orbits too are homogeneous spaces: 6

O± ∼
= SO0 (1, d − 1)/I (14.21)
6
The isotropy group of a light ray is I ∼ = ISO(d − 2), where ISO(d − 2) is the Euclidean group
on Rd−2 . The easiest way to show this is to use the Lie algebra of so(1, d − 1) and work with light-cone
coordinates. Choosing a direction of the light ray along the xd−1 axis and introducing light-cone coordinates
x± := x0 ± xd−1 , and transverse coordinates xi , i = 1, . . . , d − 2 if the lightray satisfies x− = 0 then we
have unbroken generators M +i and M ij .

– 59 –
 4. The final orbit is of course {x = 0}.

As in 1 + 1 dimensions we can identify a representation space of the Lorentz group

associated with the space of solutions to the KG equation with functions on various orbits.
The formulae are essentially the same. For example for the orbit O+ (m) a function ā :
O+ (m) → C determines a solution:
Z
Ψ(x) = dd kδ(k 2 + m2 )āeik·x
(14.22) eq:MomentumMod
dd−1~k
Z
= p ā(~k)eik·x
Rd−1 2 ~k + m
2 2

However, for scalar fields, there is no analog of the left- and right-chiral boson. There
are analogs involving interesting first order equations such as the Dirac equation and the
(anti-) self-dual equations for fields with spin.
Quite generally, we can define an inner product on the space of complex-valued solu-
tions of the KG equation such that the action of the Lorentz group is unitary. Observe
that, given any two complex-valued solutions Ψ1 , Ψ2 the current

jµ := −i (Ψ∗1 ∂µ Ψ2 − (∂µ Ψ1 )∗ Ψ2 ) (14.23)

Note that, if Ψ1 and Ψ2 both satisfy the KG equation then

∂ µ jµ = 0 (14.24)

is conserved. Therefore, if we choose a spatial slice with normal vector nµ and induced
volume form vol the inner product
Z
(Ψ1 , Ψ2 ) := nµ jµ vol (14.25)
Σ

is independent of the choice. So, fixing a Lorentz frame and taking Σ to be the slace at a
fixed time we have
Z
(Ψ1 , Ψ2 ) := −i (Ψ∗1 ∂0 Ψ2 − (∂0 Ψ1 )∗ Ψ2 ) dd−1 ~x (14.26)
Rd−1

This is clearly not positive definite on the space of all solutions but does become positive
definite when restricted to the space of complex solutions associated with a single orbit.
Indeed, substituting the expansion in momentum space (14.22) we get

dd−1~k
Z
(Ψ1 , Ψ2 ) = p (ā1 (~k))∗ (ā2 (~k)) (14.27)
Rd−1 2 ~k 2 + m2
Having phrased things this way, it is clear that there is an interesting generalization:
We can choose other representations of H = SO(d−1) and consider the induced representa-
tions of the Lorentz group. This indeed leads to the unitary representations, corresponding
to particles with nontrivial spin. ♣Incorporate some
of the following
remarks: ♣

– 60 –
15. Characters and the Decomposition of a Representation to its Irreps

Given a group G there are some fundamental problems one wants to solve:

• Construct the irreducible representations.

• Given a representation, how can you tell if it is reducible?
• Given a reducible representation, what is its reduction to irreducible representations?

I am not aware of any systematic procedure for deriving the list of irreps of any group.
However, if - somehow - the list is known then the answers to 2 and 3 are readily provided
through the use of characters.

15.1 Some basic definitions

Definition 1 Let (T, V ) be a representation of G. The functions on G defined by
X
χ(T,V ) (g) ≡ TrV T (g) ≡ Tii (g) (15.1)
i

are called the characters of the representation T .

Note: χ(T,V ) is independent of basis. Moreover, if (T, V ) ∼
= (T 0 , V 0 ) then

χ(T,V ) = χ(T 0 ,V 0 ) (15.2)

Exercise
Show that for a unitary representation

χ(g −1 ) = χ(g)∗ (15.3)

A second motivation for looking at characters is the following: matrix elements are
complicated. Sometimes one can extract all the information one is seeking by considering
this simpler and more basic set of functions on G associated to a representation.
Recall that for any element of g ∈ G the set

C(g) = {h ∈ G : ∃k, g = khk −1 } (15.4)

is called the conjugacy class of g.

Definition A function on G that only depends on the conjugacy class is called a class
function. In particular, characters are class functions. The space of class functions is a
subspace of RG .

– 61 –
Theorem. The χµ form a basis for the vector space of class functions in RG .

Proof : This is a corollary of the previous section: Any function can be expanded in
X µ µ
F (g) = F̂ij Tij (g) (15.5)

Therefore, it suffices to compute the average of Tijµ (g) over a conjugacy class:

1 X (µ)
(Tijµ )average (g) ≡ Tij (hgh−1 ) (15.6)
|G|
h∈G

Expanding this out and doing the sum on h using the ON relations shows that

(Tijµ )average (g) = nµ χµ (g) (15.7)

On the other hand, if F is a class function then

F average = F (15.8)

so
F̂ijµ nµ χµ (g)
X
F (g) = (15.9)
ij

So that χµ form a basis of class functions. ♠

Now, there is another obvious basis of the space of class functions: Denote the distinct
conjugacy classes by Ci , i = 1, . . . r. For each Ci can define a class function δCi in RG to
be the characteristic function of Ci . That is it is 1 on the class Ci and zero on all other
group elements. Comparing these two ways of finding the dimension we get:

Theorem The number of conjugacy classes of G is the same as the number of irreducible
representations of G.
Since the number of representations and conjugacy classes are the same we can define
a character table:

m 1 C1 m 2 C2 ··· ··· mr Cr
χ1 χ1 (C1 ) ··· ··· ··· χ1 (Cr )
χ2 χ2 (C1 ) ··· ··· ··· χ2 (Cr )
(15.10)
··· ··· ··· ··· ··· ···
··· ··· ··· ··· ··· ···
χr χr (C1 ) ··· ··· ··· χr (Cr )
Here mi denotes the order of Ci .
As we will see, character tables are a very useful summary of important information
about the representation theory of a group.

Exercise

– 62 –
 In a previous exercise we computed the average number of fixed points of a finite group
G acting on a finite set X.
Rederive the same result by interpreting the number of elements of X fixed by g as a
character χ(g) of the representation given by the functions on X.

15.2 Orthogonality relations on the characters

Let χµ be the character of T (µ) . The orthogonality relations for matrix elements implies:

1 X
χν (g −1 )χµ (g) ≡ (χν , χµ ) = δµν (15.11)
|G|
g∈G

To prove this, simply put i = ` and s = m in (9.9) and sum on i and s.

Now, we argued before - abstractly - that any finite dimensional representation (T, V )
of G is completely reducible:

T ∼
= a1 T (1) ⊕ a2 T (2) ⊕ · · · ⊕ ar T (r) = ⊕µ aµ T (µ) (15.12) eq:reducible

The ai are nonnegative integers, they measure the number of times a rep appears and
are referred to as the “multiplicities.”
From the orthogonality relation we get:

aµ = (χ(T,V ) , χµ ) (15.13) eq:multipl

Thus we arrive at a key fact:

Theorem
Two representations are equivalent iff they have the same character.
If you have two reps and you want to see whether they are inequivalent then an efficient
test is to compute the characters.
The orthogonality relations on characters can be stated more beautifully if we use the
fact that characters are class functions:

Theorem Orthogonality Relations for characters Denote the distinct conjugacy

classes Ci ∈ C, and the (distinct) characters of the distinct irreducible unitary representa-
tions χµ . Then

1 X
mi χµ (Ci )χν (Ci )∗ = δµν (15.14) eq:orthogrel
|G|
Ci ∈C

where mi = |Ci | is the order of the conjugacy class Ci . We also have

X |G|
χµ (Ci )∗ χµ (Cj ) = δij (15.15) eq:orthogrelii
µ
mi

– 63 –
 Proof : Let us first prove (15.14). From unitarity we have χµ (g −1 ) = (χµ (g))∗ . Since
χµ are class functions the sum over the elements in the group can be written as a sum over
the conjugacy classes Ci ∈ C.
The equation (15.14) can be interpreted as the statement that the r × r matrix
r
mi
Sµi := χµ (Ci ) µ = 1, . . . , r i = 1, . . . , r (15.16)
|G|

satisfies
r
X
∗
Sµi Sνi = δµν (15.17)
i=1

Therefore, Sµi is a unitary matrix. The left-inverse is the same as the right-inverse, and
hence we obtain (15.15). ♠

Exercise
Show that right-multiplication of the character table by a diagonal matrix produces a
unitary matrix.

Exercise
Using the orthogonality relations on matrix elements, derive the more general relation
on characters:

1 X δµν (ν)
χµ (g)χν (g −1 h) = χ (h) (15.18) eq:convolchar
|G| nµ
g∈G

We will interpret this more conceptually later.

Exercise
A more direct proof of (15.15) goes as follows. Consider the operator L(g1 ) ⊗ R(g2 )
acting on the regular representation. We will compute

TrRG [L(g1 ) ⊗ R(g2 )] (15.19)

in two bases.
First consider the basis φµij of matrix elements. We have:

L(g1 ) ⊗ R(g2 ) · φµij = Tjµ0 j (g2 )Tiµ0 i (g1−1 )φµi0 j 0

X
(15.20)
i0 ,j 0

– 64 –
so the trace in this basis is just:
X
TrRG [L(g1 ) ⊗ R(g2 )] = χµ (g1 )∗ χµ (g2 ) (15.21)
µ

On the other hand, we can use the delta-function basis: δg . Note that

L(g1 ) ⊗ R(g2 ) · δg = δg1 gg−1 (15.22)

2

So in the delta function basis we get a contribution of +1 to the trace iff g = g1 gg2−1 , that
is iff g2 = g −1 g1 g, that is iff g1 and g2 are conjugate, otherwise we get zero.
Now show that

#{g : g2 = g −1 g1 g} = |C(g1 )| = m1 (15.23) eq:weight

by considering Z(g1 ), the centralizer subgroup of g1 in G and identifying the cosets

G/Z(g1 ) with the elements of the conjugacy class C(g1 ).
So we get the second orthogonality relation. ♠

16. Some examples of representations of discrete groups and their char-

acter tables

16.1 S3
The group S3 has three obvious irreducible representations:
1+ . 1 → 1 (ij) → 1 (123) → 1, etc.
1− . 1 → 1 (ij) → −1 (123) → +1, etc. This is the sign representation, or the 
homomorphism to Z2 mentioned in lecture 1.
2. Symmetries of the triangle.
!
10
1→
01
!
−1 0
(23) → (16.1)
0 1
! √ !
cos(2π/3) sin(2π/3) −1/2 3/2
(123) → = √
− sin(2π/3) cos(2π/3) − 3/2 −1/2

These reps are irreducible. Note that by applying group elements in the third repre-
sentation to any vector we obtain a set of vectors which spans R2 .
Indeed we can check:
6 = 12 + 12 + 22 (16.2)
to conclude that there are no other irreps.
It is now straightforward to write out the character table for S3 .

– 65 –
 1

2 3

Figure 6: The symmetries of the equilateral triangle give the 2-dimensional irrep of S3 . fig:triangle

[1] 3[(12)] 2[(123)]

χ1+ : 1 1 1
(16.3)
χ1− : 1 −1 1
χ2 : 2 0 −1
Here is an example of the use of characters to decompose a representation:

Now let us see how we can use the orthogonality relations on characters to find the decom-
position of a reducible representation.

Example 1 Consider the 3 × 3 rep generated by the \ action of the permutation group S3
on R3 . We’ll compute the characters by choosing one representative from each conjugacy
class:
     
100 010 010
1 → 0 1 0 (12) → 1 0 0 (132) → 0 0 1 (16.4)
     
001 001 100

From these representatives the character of V = R3 is easily calculated:

χV (1) = 3 χV ([(12)]) = 1 χV ([(132)] = 0 (16.5)

Using the orthogonality relations we compute

1 3 2
a1+ = (χ1+ , χ) = 3 + 1 + 0 = 1 (16.6)
6 6 6
1 3 2
a1− = (χ1− , χ) = 3 + (−1) · 1 + 0 = 0 (16.7)
6 6 6
1 3 2
a2 = (χ2 , χ) = 3 · 2 + 0 · 1 + (−1) · 0 = 1 (16.8)
6 6 6
Therefore:
χV = χ1+ + χ2 (16.9)

showing the decomposition of R3 into irreps.

– 66 –
Example 2 Consider S3 acting by permuting the various factors in the tensor space
V ⊗ V ⊗ V for any vector space V . Now, if dimV = d then we have

χ([1]) = d3
χ([(ab)]) = d2 (16.10) eq:vcubed
χ[(abc)] = d
So we can compute
1 3 2 1
a1+ = (χ1+ , χ) = d3 + d2 + d = d(d + 1)(d + 2) (16.11)
6 6 6 6
1 3 2 1
a1− = (χ1− , χ) = d3 + (−1) · d2 + d = d(d − 1)(d − 2) (16.12)
6 6 6 6
1 3 3 2 1
a2 = (χ2 , χ) = 2d + 0 · d + (−1) · d = d(d2 − 1)
2
(16.13)
6 6 6 3
Thus, as a representation of S3 , we have
d(d + 1)(d + 2) d(d − 1)(d − 2) d(d + 1)(d − 1)
V ⊗3 ∼
= 1+ ⊕ 1− ⊕ 2 (16.14)
6 6 3
Note that the first two dimensions are those of S 3 V and Λ3 V , respectively, and that the
dimensions add up correctly.
We are going to develop this idea in much more detail when we discuss Schur-Weyl
duality in section ****

Remarks
• Notice that this does not tell us how to block-diagonalize the matrices. We will see
how to do that later.

Exercise
Repeat example 2 to decompose V ⊗2 as a representation of S2 .

Exercise
Write out the unitary matrix Sµi for G = S3 .

Exercise
a.) Suppose we tried to define a representation of S3 by taking (12) → 1 and (23) → −1.
What goes wrong?
b.) Show that for any n there are only two one-dimensional representations of Sn

– 67 –
16.2 Dihedral groups
The irreps of the dihedral group Dn are easily written down.
Recall that Dn is the group generated by x, y subject to the relations:

x2 = 1
yn = 1 (16.15) eq:rels
xyx−1 = y −1
We can define two-dimensional complex representations T (j) by:
!
ωj
T (j) : y →
ω −j
! (16.16) eq:twodee
01
x→
10

where ω = e2πi/n . Clearly, these satisfy the defining relations of Dn .

Note that T (j) = T (j+n) , and moreover, T (j) ∼ = T (n−j) are equivalent representations.
Note that (T (j) )∗ = T (−j) and hence these are all real representations. Indeed, by a
similarity transformation we can make y map to a rotation matrix.
Suppose n is odd. Then the representations T (j) are irreducible and inequivalent for
j = 1, . . . , (n − 1)/2. For j = 0 the representation is reducible to 1+ given by y → 1, x → 1
and 1− given by y → 1, x → −1. Altogether we have
n−1 2
2n = · 2 + 12 + 12 (16.17)
2
so these are the only irreps.
Suppose n is even. Then the representations T (j) are irreducible and inequivalent for
j = 1, . . . , n2 − 1. For j = 0 the representation is reducible to 1++ given by y → 1, x → 1
and 1+− given by y → 1, x → −1. For j = n/2 the representation is reducible to 1−+
given by y → −1, x → 1 and 1−− given by y → −1, x → −1. Altogether we have
n
2n = ( − 1) · 22 + 12 + 12 + 12 + 12 (16.18)
2
so we have all the representations.
The character tables are:
For n = 2k + 1 odd:

[1] 2k[x] 2[y] ··· 2[y k ]

1+ 1 1 1 ··· 1
1− 1 −1 1 ··· 1
2(j) 2 0 ω + ω −j
j ··· ω + ω −jk
jk

– 68 –
 where j = 1, . . . , k.
For n = 2k even we have
MISSING TABLE
where j = 1, . . . , k − 1
**** SHOULD apply these tables to solve some problem ***

17. The decomposition of tensor products

A very common problem one faces is the following:

Suppose we have two representations V1 , V2 . Then the tensor product V1 ⊗ V2 is also
a representation.
Question: If we know how to decompose V1 , V2 into irreps, how does V1 ⊗ V2 fall apart into its irreducible pieces?
An example of this general question arises in the quantum mechanical addition of
angular momenta. For this reason it is sometimes called the “Clebsch-Gordon problem.”
(**** Following should be moved to place where we talk about complete reducibility
****)
If V is a completely decomposable representation (always true for compact groups)
then we can write
aµ
V ∼
= ⊕rµ=1 ⊕i=1 T (µ) := ⊕µ aµ T (µ) (17.1)

For a fixed µ let let

a
V (µ) := ⊕i=1
µ
T (µ) (17.2)

It contains the representation µ with the correct degeneracy in V . It is called the isotypical
component belonging to µ. Note that it can be written as

V (µ) = Caµ ⊗ T (µ) (17.3) eq:isotpyp

where Caµ is understood to be the trivial representation of G. Thus by Schur’s lemma,

the intertwiners of V (µ) with itself are of the form K ⊗ 1 where K ∈ End(Caµ ) is arbitrary.
We abbreviate Caµ ⊗T (µ) to aµ T µ and with this understood we write the decomposition
into isotypical components as:
V = ⊕aµ T µ (17.4)

Now suppose:
V1 = ⊕aµ T µ V2 = ⊕bν T ν (17.5)

then
V1 ⊗ V2 = ⊕µ,ν aµ bν T µ ⊗ T ν (17.6)

so in finding the isotypical decomposition of V1 ⊗ V2 it suffices to find the decomposition

of products of irreps.

Definition Fusion Coefficients The isotypical decomposition of the product of irreps is

of the form:

– 69 –
 µν N λ times
z }| {
T (µ) ⊗ T (ν) (λ)
= ⊕λ T ⊕ · · · ⊕ T (λ)
(17.7) eq:decomp
λ
= ⊕λ Nµν T (λ)

λ are known as fusion coefficients.

The multiplicities Nµν

Exercise
Show that
χT1 ⊗T2 (g) = χT1 (g)χT2 (g) (17.8)

Now let us put this observation to use and solve our problem. Taking the trace of
(17.7) we get:
X
λ
χµ (g)χν (g) = Nµν χλ (g) (17.9) eq:chardec
λ

Now, taking the inner product we get:

λ
Nµν = (χλ , χµ χν ) (17.10) eq:fusion

We can write this out as:

1 X
λ
Nµν = χµ (g)χν (g)χλ (g −1 ) (17.11) eq:fusioni
|G|
g∈G

This can be written in a different way, which we will explain conceptually in the next
section. For simplicity choose unitary irreps, and recall that the orthogonality relations on
characters is equivalent to the statement that
r
mi
Sµi := χµ (Ci ) µ = 1, . . . , r i = 1, . . . , r (17.12)
|G|

is a unitary matrix. Noticeqthat if we take µ = 0 to correspond to the trivial 1-dimensional

mi
representation then S0i = |G| . Thus we can write
X Sµi Sνi S ∗
λ λi
Nµν = (17.13) eq:fusionii
S0i
i

This is a prototype of a celebrated result in conformal field theory known as the

“Verlinde formula.”
Equations (17.10)(17.11)(17.13) give a very handy way to get the numbers Nµν λ . Note
λ are nonnegative integers, although this
that, by their very definition the coefficients Nµν
is hardly obvious from, say, (17.13).

– 70 –
18. Algebras
♣First part with
definitions was
We first need the general abstract notion: Definition An algebra over a field k is a vector copied into chapter
2. No need to
repeat here. ♣
space V over k with a notion of multiplication of two vectors

V ×V →V (18.1)

denoted:
v1 , v2 ∈ V → v1 · v2 ∈ V (18.2)
which has a ring structure compatible with the scalar multiplication by the field. Con-
cretely, this means we have axioms:
i.) (v1 + v2 ) · v3 = v1 · v3 + v2 · v3
ii.) v1 · (v2 + v3 ) = v1 · v2 + v1 · v3
iii.) α(v1 · v2 ) = (αv1 ) · v2 = v1 · (αv2 ), ∀α ∈ k.
The algebra is unital, i.e., it has a unit, if ∃1V ∈ V (not to be confused with the
multiplicative unit 1 ∈ k of the ground field) such that:
iv.) 1V · v = v · 1V = v
If, in addition, the product satisfies:

(v1 · v2 ) · v3 = v1 · (v2 · v3 ) (18.3)

for all v1 , v2 , v3 ∈ V then V is called an associative algebra. In general, a nonassociative

algebra means a not-necessarily associative algebra. In any algebra we can introduce the
associator

[v1 , v2 , v3 ] := (v1 · v2 ) · v3 − v1 · (v2 · v3 ) (18.4) eq:associator

Note that it is trilinear.

In general, if {vi } is a basis for the algebra then the structure constants are defined by
X
vi · vj = ckij vk (18.5)
k

A representation of an algebra V is a vector space W and a map T : V → End(W ) so

that

T (αv + βw) = αT (v) + βT (w)

(18.6) eq:repalgebra
T (v1 · v2 ) = T (v1 ) ◦ T (v2 )

Example. Mn (k) is a vector space over k of dimension n2 . It is also an associative

algebra because matrix multiplication defines an algebraic structure. An important fact is
that the only irreducible representation of the algebra Mn (k) is the natural one acting on
W = k ⊕n . This leads to a concept called “Morita equivalence” of algebras.

– 71 –
 Exercise
a.) Write out a basis and structure constants for the algebra Mn (k).
b.) Show that the set of n × n matrices over an associative algebra, A, denoted Mn (A),
is itself an associative algebra.

Exercise
a.) If A is an algebra, then it is a module over itself, via the left-regular representation
(LRR). a → L(a) where
L(a) · b := ab (18.7)
Show that if we choose a basis ai then the structure constants

ai aj = cij k ak (18.8)

define the matrix elements of the LRR:

k
(L(ai ))j = ckij (18.9)

An algebra is said to be semisimple if these operators are diagonalizable.

b.) If A is an algebra, then it is a bimodule over A ⊗ Ao where Ao is the opposite
algebra.

18.1 Coalgebras, Bialgebras, and Frobenius algebras

Definition A Frobenius algebra A is an associative algebra over a field k with a trace

θ : A → k such that the quadratic form A ⊗ A → k given by a ⊗ b → θ(ab) defines a
nondegenerate bilinear form on A.
Matrix algebras over a field are examples of Frobenius algebras where the nondegen-
erate bilinear form is the trace map. Frobenius algebras have nice properties with their
duals.
The dual space of an algebra V ∗ has a structure known as a coalgebra. The dual of
the multiplication is a comultiplication

∆:V∗ →V∗⊗V∗ (18.10)

and the dual of associativity says (1 ⊗ ∆)∆ = (∆ ⊗ 1)∆.

The dual of the unit is the counit  : V ∗ → k. The dual of the axioms for the unit are

( ⊗ 1)∆ = (1 ⊗ )∆ = 1. (18.11)

– 72 –
 DIAGRAMS.
A bialgebra is an algebra which is simultaneously a coalgebra so that the structures are
all compatible, that is, the comultiplication ∆ and the counit  are algebra homomorphisms.
Finally, a Hopf algebra is a bialgebra with an extra bit of structure S : H → H. It
is a generalization of the algebra of functions on a group and is a starting point for the
theory of quantum groups.

Exercise
a.) In the literature one sometimes sees a Frobenius algebra defined as an algebra A
with a nondegenerate bilinear form

σ :A⊗A→k (18.12) eq:frob

such that σ(xy, z) = σ(y, zx). Show that this is equivalent to our definition.
b.) Show that if (A, θ) is a Frobenius algebra then the dual algebra A∗ is a left A-
module which is isomorphic to A as a left A-module.

18.2 When are two algebras equivalent? Introduction to Hochschild cohomol-

ogy
Deformation - which are trivial? Nontrivial deformations measured by Hochschild coho.
View as a flatness condition.

19. The group ring and group algebra

The group algebra is a useful tool for addressing the fundamental questions at the beginning
of section ****
1. Useful for producing explicit reduction of matrix reps by the method of projection
operators.
2. Useful for producing the representations of the symmetric group Sn .
3. Important and beautiful mathematical structure.
The regular representation is an algebra. Indeed, the space of complex-valued functions
F un(X) on any space X is an algebra under pointwise multiplication:

(f1 · f2 )(x) := f1 (x)f2 (x) (19.1)

Note that this is a commutative algebra. In particular, we can apply this remark to X = G.
Now, the group ring, usually denoted Z[G] is, as an abelian group the free abelian
group generatred by G. That is, it is the space of formal linear combinations
X
Z[G] := {x = xg g : xg ∈ Z} (19.2)
g∈G

– 73 –
The ring structure is defined by taking

δg1 ∗2 δg2 = δg1 ·g2 (19.3)

and extending by linearity. Thus, we have

X  X  group multiplication
X z}|{
x(g)g ∗2 y(g)g := x(g)y(h) g·h
g∈G g∈G g,h∈G
" # (19.4) eq:groupalgmul
X X
= x(g)y(h) k
k∈G g,h∈G:gh=k

Replacing Z → C (more properly, taking the tensor product with C) gives the group
algebra C[G].
As a vector space C[G] is naturally dual to the regular representation: RG ∼
= C[G]∗ .
The natural pairing is X
hf, xi := xg f (g) (19.5)
g∈G

Of course, there is a natural basis for C[G], namely the elements g themselves, and the
dual basis is the basis of delta functions δg . Thus the isomorphism takes f : G → C to
P
g f (g)g.
Now we see that there are two algebra structures on RG and C[G].
First of all, pointwise multiplication defines an algebra structure on C[G] since

f1 ∗1 f2 (g) ≡ f1 (g)f2 (g) ⇒

δg1 ∗1 δg2 = δg1 ifg1 = g2 (19.6) eq:pointwise
=0 otherwise

which implies
(
g1 g1 = g2
g1 ∗1 g2 = (19.7) eq:pointwiseii
0 g1 6= g2

However, the multiplication (19.7) is not what is usually understood when one speaks
of the group algebra, rather what is meant is the multiplication (19.4).
Under the isomorphism RG ∼ = C[G] the product ∗2 must correspond to a second algebra
structure on RG , and indeed there is one - called the convolution product.
X
(f1 ∗2 f2 )(g) := f1 (h)f2 (h−1 g) (19.8) eq:covolution
h∈G

Note that ∗2 is therefore, in general, a noncommutative algebra product, whereas ∗1

is a commutative algebra structure.

– 74 –
Example Z2 = {1, σ}. The group algebra is

RZ2 = {a · 1 + b · σ : a, b ∈ C} (19.9)

Now the algebra structure is:

(a · 1 + b · σ) · (a0 · 1 + b0 · σ) = (aa0 + bb0 )1 + (ba0 + ab0 )σ (19.10)

These are the “double-numbers,” a lesser known cousin of the complex numbers.

Exercise
Which elements of RZ2 are invertible?

19.1 Projection operators in the group algebra

The group algebra is of course a representation of Gleft × Gright acting as a left action:

T (g1 , g2 ) · x = g1 xg2−1 (19.11)

for x ∈ C[G].
We know from the Peter-Weyl theorem that this representation is highly reducible.
We can now construct explicit projection operators which project onto the irreducible
(µ)
components. We assume that the distinct irreps Tij are somehow known.
Let us define:

Pijµ ∈ C[G] 1 ≤ i, j ≤ nµ (19.12) eq:idemps

by

nµ X (µ)
Pijµ := T̂ij (g)g (19.13) eq:projops
|G|
g∈G

Recall that the dual representation is the transpose inverse of the original one so
(µ) (µ)
T̂ij (g) = Tji (g −1 )
These elements satisfy the simple product law:

0
Pijµ Piµ0 j 0 = δµµ0 δji0 Pijµ0 (19.14) eq:prodlaw

They are therefore orthogonal projection operators in the group algebra acting on itself.
Moreover, again using similar manipulations one can show that:

(µ)
g · Pijµ = µ
X
Tsi (g)Psj (19.15) eq:geerep
s

(µ)
Pijµ · g −1 = T̂sj (g)Pisµ
X
(19.16) eq:geerepii
s

– 75 –
 and thus the projection operators Pijµ explicitly decompose C[G] into the irreps T (µ) ⊗
T̂ (µ) of Gleft ×Gright . This will be useful below in showing how to decompose representations
into irreps.

Exercise
Prove (19.14)(19.15) and (19.16) using the orthogonality conditions The manipulations
are very similar to those used to prove theorem 10.1.

19.2 The center of the group algebra and the subalgebra of class functions
As we have noted, the group algebra C[G] is a noncommutative algebra. Correspondingly,
RG is a noncommutative algebra under the convolution product.
A little thought shows that the center of the group algebra Z[C[G]] is spanned by the
P
elements ci = g∈Ci g where we sum over the conjugacy class Ci of g. This simply follows
from X
gci g −1 = ghg −1 = ci (19.17)
h∈Ci

Correspondingly, the center of RG under the convolution product are the class functions.
Indeed, we can now interpret the exercise with equation (15.18) above as saying that
the characters satisfy:
δµν
χµ ∗2 χν = χµ (19.18)
nµ
The product is commutative, and in fact, diagonal in this basis. On the other hand, the
convolution product is nondiagonal in the conjugacy class basis:
X
δCi ∗2 δCj = Mijk δCk (19.19)
k

It is interesting to compare with the pointwise product, where the reverse is true
X
λ
χµ ∗1 χν = Nµν χλ (19.20)
λ

while
δCi ∗1 δCj = δij Ci (19.21)
So the unitary matrix Sµi is a kind of Fourier transform which exchanges the product for
which the characters of representations or characteristic functions of conjugacy classes is
diagonalized: s
X |G|
χµ = Siµ δCi (19.22)
mi
i

Exercise

– 76 –
 Show that the structure constants in the basis δCi in the convolution product can be
written as:
∗ S∗ S
S0i S0j X Siµ jµ kµ
Mijk = (19.23) eq:convolusc
S0k µ nµ

Exercise
An algebra is always canonically a representation of itself in the left-regular represen-
tation. The representation matrices are given by the structure constants, thus

L(χµ )νλ = Nµν

λ
(19.24)

shows that the fusion coefficients are the structure constants of the pointwise multiplication
algebra of class functions.
Note that since the algebra of class functions is commutative the matrices L(χµ ) can
be simultaneously diagonalized.
Show that the unitary matrix Siµ is in fact the matrix which diagonalizes the fusion
coefficients.

20. Interlude: 2D Topological Field Theory and Frobenius algebras

The group algebra illustrates the notions of bialgebra and Frobenius algebra very nicely.
We would like to explain briefly the relation of these algebraic structures to two-dimensional
topological field theory.
The axioms of topological field theory give a caricature of what one wants from a
field theory: Spaces of states and transition amplitudes. By stripping away the many
complications of “real physics” one is left with a very simple structure. Nevertheless, the
resulting structure is elegant, it is related to beautiful algebraic structures which, at least
in two dimensions, which have surprisingly useful consequences. This is one case where
one can truly “solve the theory.”
Of course, we are really interested in more complicated theories. But the basic frame-
work here can be adapted to any field theory. What changes is the geometric category
under consideration.

20.1 Geometric Categories

What are the most primitive things we want from a physical theory in d spacetime dimen-
sions?
In a physical theory one often decomposes spacetime into space and time as in 7. If
space is a (d − 1)-dimensional manifold time Y then, in quantum mechanics, we associate
to it a vector space of states H(Yd ).

– 77 –
 Figure 7: A spacetime Xd = Y × R. Y is (d − 1)-dimensional space. fig:tfti

In a generic physical theory the vector space has a lot of extra structure: It is a Hilbert
space, there are natural operators acting on this Hilbert space such as the Hamiltonian.
The spectrum of the Hamiltonian and other physical observables depends on a great deal
of data. Certainly they depend on the metric on spacetime since a nonzero energy defines
a length scale
~c
L= (20.1)
E
In topological field theory one ignores most of this structure, and focuses on the de-
pendence of H(Y ) on the topology of Y . For simplicity, we will initially assume Y is
compact.
So: We have an association:
(d − 1)-manifolds Y to vector spaces: Y → H(Y ), such that homeomorphic manifolds
map to isomorphic vector spaces.
Now, we also want to incorporate some form of locality, at the most primitive level.
Thus, if we take disjoint unions

H(Y1 q Y2 ) = H(Y1 ) ⊗ H(Y2 ) (20.2) eq:disunion

Note that (20.2) implies that we should assign to H(∅) the field of definition of our
vector space. For simplicity we will take H(∅) = C, although one could use other ground
fields.

Figure 8: Time development in Xd gives a linear map. fig:tftitime

In addition, in physics we want to speak of transition amplitudes. If there is a spacetime

Xd interpolating between two time-slices, then mathematically, we say there is a cobordism
between Y and Y 0 . That is, a cobordism from Y to Y 0 is a d-manifold with boundary and
a partition of its boundary
∂Xd = (∂Xd )in ∪ (∂Xd )out (20.3)

– 78 –
 ∼ Y and (∂Xd )out ∼
so that there is a homeomorphism (∂Xd )in = = Y 0.
Here we will be considering oriented manifolds. Then an oriented cobordism from Y1
to Y2 must have
∂M = Y2 − Y1 (20.4)
where the minus sign indicates the opposite orientation on Y1 . (Recall that an orientation
on M determines one on ∂M . We will adopt the convention “outward normal first” - ONF
- “One Never Forgets.” ).
If Xd is an oriented cobordism from Y to Y 0 then the Feynman path integral assigns
a linear transformation
F (Xd ) : H(Y ) → H(Y 0 ). (20.5)
Again, in the general case, the amplitudes depend on much more than just the topology
of Xd , but in topological field theory they are supposed only to depend on the topology.
More precisely, if Xd ∼
= Xd0 are homeomorphic by a homeomorphism = 1 on the boundary
of the cobordism, then
F (Xd ) = F (Xd0 ) (20.6)
One key aspect of the path integral we want to capture - again a consequence of locality
- is the idea of summing over a complete set of intermediate states. In the path integral
formalism we can formulate the sum over all paths of field configurations from t0 to t2
by composing the amplitude for all paths from t0 to t1 and then from t1 to t2 , where
t0 < t1 < t2 , and then summing over all intermediate field configurations at t1 . We refer
to this property as the “gluing property.” The gluing property is particularly obvious in
the path integral formulation of field theories.

Figure 9: Gluing cobordisms fig:tftii

In topological field theory this is formalized as:

If M is a cobordism from Y1 to Y2 with

∂M = Y2 − Y1 (20.7)

and M 0 is another oriented cobordism from Y2 to Y3

∂M 0 = Y3 − Y2 (20.8)

then we can compose M 0 ◦ M as in 9 to get a cobordism from Y1 to Y3 .

Naturally enough we want the associated linear maps to compose:

F (M ◦ M 0 ) = F (M 0 ) ◦ F (M ) : H(Y1 ) → H(Y3 ) (20.9)

– 79 –
 What we are describing, in mathematical terms is a functor between categories.

Definition A functor between categories C1 , C2 is an association between categories

which preserves structure. Thus there is a map on objects: F : Obj(C1 ) → Obj(C2 ).
Moreover, we have a
a.) Covariant functor if F : Arr(C1 ) → Arr(C2 ) such that F (a → b) is an arrow from
F (a) to F (b) for all a, b ∈ Obj(C1 ) and

F (φ2 ◦ φ1 ) = F (φ2 ) ◦ F (φ1 ) (20.10)

b.)Contravariant functor if F : Arr(C1 ) → Arr(C2 ) such that F (a → b) is an arrow

from F (b) to F (a) for all a, b ∈ Obj(C1 ) and

F (φ2 ◦ φ1 ) = F (φ1 ) ◦ F (φ2 ) (20.11)

The above axioms can be concisely summarized by defining a cobordism category whose
objects are homeomorphism classes of oriented (d − 1)-manifolds and whose morphisms are
oriented cobordisms, two morphisms being identified if they differ by a homeomorphism
which is an identity on the boundary. Let us call this Cob(d).

Definition A d-dimensional topological field theory is a functor from the category Cob(d)
to the category VECT of vector spaces and linear transformations.

Actually, these are both tensor categories and we want a functor of tensor categories.
At this point we begin to see how we can incorporate other field theories. We can
change the geometric category to include manifolds with Riemannian metric, spin structure,
etc.

20.2 Some general properties

Let us deduce some simple general facts following from the above simple remarks.
First note that if M is closed then it can be regarded as a cobordism from ∅ to ∅.
Therefore F (M ) must be a linear map from C to C. But any such linear map is canonically
associated to a complex number. We define the partition function of M , Z(M ) to be this
complex number.
There is one cobordism which is distinguished, namely Y × [0, 1]. This corresponds to
a linear map P : H(Y ) → H(Y ). Indeed, physically it is just the operator

exp[−T H] (20.12)

where H is the Hamiltonian, and T is the Euclidean time interval (in some presumed
metric).
Evidently, by the axioms of topological field theory, P 2 = P and therefore we can
decompose

H(Y ) = P H(Y ) ⊕ (1 − P )H(Y ) (20.13) eq:decompp

– 80 –
 All possible transitions are zero on the second summand since, topologically, we can
always insert such a cylinder. It follows that it is natural to assume that

F (Y × [0, 1]) = IdH(Y ) (20.14) eq:identmap

One can think of this as the statement that the Hamiltonian is zero.

Figure 10: Deforming the cylinder. fig:dfscyl

Now, the cobordism (20.14) is closely related to the cobordism ∅ → Y ∪ −Y thus

defining a map
δ : C → H(Y ) ⊗ H(−Y ) (20.15)
and also to a cobordism Y ∪ −Y → ∅ thus defining a quadratic form:

Q : H(Y ) ⊗ H(−Y ) → C (20.16)

Figure 11: Composing δ ⊗ 1 and 1 ⊗ Q fig:scobord

Let us now compose these cobordisms we get the identity map as in 11. It then follows
from some linear algebra that Q is a nondegenerate pairing, so we have an isomorphism to
the linear dual space:
H(−Y ) ∼
= H(Y )∗ , (20.17)
under which Q is just the dual pairing. Moreover δ(1) has the explicit formula:
X
δ:1→ φµ ⊗ φµ (20.18)

where we can let φµ be any basis for H(Y ), and then define the dual basis by Q(φµ , φν ) =
δµ ν .

Exercise

– 81 –
 αµν φµ ⊗ φν and compute the process in 9.
P
Prove this formula. Write δ(1) =

Figure 12: Composing Q with δ gives the dimension: dimH(Y ) = Z(Y × S 1 ). fig:dimcirc

Now consider the diagram in 12. Geometrically this is a morphism from ∅ to ∅, so

the functor maps this to a linear map from C to C. Such a linear map is completely
determined by a complex number, and this complex number is the composition Qδ. From
our formula for δ(1) we see that the value Z(Y × S 1 ) is just the dimension dimH(Y ). Thus,
in topological field theories, the vector spaces are necessarily finite dimensional.
Note that if we change the category to the category of manifolds with Riemannian
structure and we take the product Riemannian structure on Y × S 1 then

Z(Y × S 1 ) = Tre−βH (20.19) eq:traccrc

where β is the radius of the circle and H is the Hamiltonian. If in addition we enrich
the category to include spin structures then there will be several kinds of traces.

Exercise
Show that a 1-dimensional TFT is completely specified by choosing a single vector
space V and a vector in that space.

Exercise
Show that any cobordism of ∅ to Y defines a state in the space H(Y ). This is a
primitive version of the notion of the “Hartle-Hawking” state in quantum gravity. It is
also related to the state/operator correspondence in conformal field theory.

20.3 Two dimensional closed topological field theory and Commutative Frobe-
nius Algebras
Some beautiful extra structure shows up when we consider the case d = 2, due to the
relatively simple nature of the topology of 1-manifolds and 2-manifolds. To begin, we
restrict attention to closed (d − 1)-manifolds, that is, we consider a theory of closed strings.

– 82 –
 In this case, the spatial (d − 1) manifolds are necessarily of the form S 1 ∪ S 1 ∪ · · · ∪ S 1 ,
i.e. disjoint unions of n S 1 ’s.
If we assign H(S 1 ) = C then

H(S 1 q S 1 q · · · q S 1 ) = C ⊗n (20.20)

Thus, we have a single basic vector space.

Figure 13: Typical Riemann surface amplitude fig:twodeei

Transition amplitudes can be pictured as in 13. We get

F (Σ) : C ⊗n → C ⊗m (20.21)

where Σ is a Riemann surface. There are n ingoing circles and m outgoing circles.
Now, topology dictates that the vector space C in fact must carry some interesting
extra structure.

Figure 14: The sphere with 3 holes fig:product

In 14 we see that the sphere with three holes defines a product

m:C⊗C →C (20.22)

With the other orientation we get a co-product:

∆:C →C⊗C (20.23)

In 15 we see that there is a trace:

θ:C→C (20.24)

In 16 we see that there is a map C → C. This is completely determined by its value

on 1 ∈ C. From the diagram in 17 we see that the image of 1 must be in fact a unit for
the muiltiplication.

– 83 –
 Figure 15: the trace map fig:trace

Figure 16: The disk defines the unit. fig:unit

Figure 17: Showing that 1C really is the unit. fig:uniti

Moreover, from 18 we see the multiplication is associative.

Finally, we can make a diffeomorphism of the disk with 2 holes, holding the outer circle
fixed and rotating the inner two circles. This shows that the product must be commutative.
Any oriented surface can be decomposed into the basic building blocks we have used
above. However, the same surface can be decomposed in many different ways. When we
have different decompositions we get algebraic relations on the basic data m, ∆, θC . At
this point you might well ask: “Can we get more elaborate relations on the algebraic data
by cutting up complicated surfaces in different ways?” The beautiful answer is: “No,
the above relations are the only independent relations.” The algebraic structure we have
discovered is just that of a Frobenius algebra of section *** !

The Sewing Theorem. To give a 2d topological field theory is equivalent to giving

a commutative associative finite dimensional Frobenius algebra.

The proof (which is not difficult) depends on Morse theory and will not be given here.
For a detailed discussion see, for example,

– 84 –
 Figure 18: Associativity fig:associativ

Figure 19: Commutativity. In the supercase we will have graded commutativity: φ1 φ2 =

(−1)deg φ1 deg φ2 φ2 φ1 . fig:commutativ

G. Segal and G. Moore, “D-branes and K-theory in 2D topological field theory” hep-
th/0609042; To appear in Mirror Symmetry II, Clay Mathematics Institute Monograph,
by P. Aspinwall et. al.

20.4 Boundary conditions

Now let us enrich our theory by allowing the time-slices Y to be manifolds with boundary.
There will be a set of boundary conditions B0 , and we will attach an element of B0 to each
boundary component of ∂Y .
A cobordism X from Y0 to Y1 will thus have two kinds of boundaries:

∂X = Y0 ∪ Y1 ∪ ∂cstr X (20.25)

where ∂cstr X is the time-evolution of the spatial boundaries. We will call this the “con-
strained boundary.”

Figure 20: Morphism space for open strings: Oab . fig:openo

In d = 2, in this enlarged geometric category the initial and final state-spaces are
associated with circles, as before, and now also with intervals. The boundary of each
interval carries a label a, b, c, . . . from the set B0 .

– 85 –
 Definition: We denote the space Oab for the space associated to the interval [0, 1]
with label b at 0 and a at 1.
In the theory of D-branes, the intervals are open strings ending on submanifolds of
spacetime. That is why we call the time-evolution of these boundaries the “constrained
boundaries” – because the ends are constrained to live in the D-brane worldvolume.

Figure 21: Basic cobordism of open strings. fig:openi

Figure 22: A disk defining an element 1a ∈ Oaa fig:dsdki

Figure 23: The trace element: θa : Oaa → C. fig:tftiii

As in the closed case, the cobordism [0, 1] × [0, 1] defines Pab : Oab → Oab , and we can
assume WLOG that it is Pab = 1.
Now consider the cobordism in 21. This clearly gives us a bilinear map

Oab × Obc → Oac (20.26)

As in the closed case we see that these maps satisfy an associativity law. Moreover, as in
the closed case, 1a is an identity for the multiplication.
Comparing with the definition of a category we see that we should interpret B0 as the
space of objects in a category B, whose morphism spaces Hom(b, a) = Oab . Note that the
morphism spaces are vector-spaces, so we have a C-linear category. In fact, this category

– 86 –
has a very special property as we learn from considering the S-shaped cobordism (the open
string analog of 11). We learn that θa : Oaa → C defines a nondegenerate inner product:

Qa (ψ1 , ψ2 ) = θa (ψ1 ψ2 ) (20.27)

Thus, the Oaa are Frobenius algebras.

Moreover, using the S-shaped cobordism analogous to 11 we learn that Oab is dual to
Oba . In fact we have

a θ
Oab ⊗ Oba → Oaa → C
(20.28) eq:dualpair
θ
Oba ⊗ Oab → Obb →b C

are perfect pairings with

θa (ψ1 ψ2 ) = θb (ψ2 ψ1 ) (20.29) eq:cyclic

for ψ1 ∈ Oab , ψ2 ∈ Oba .

Definition A Frobenius category is a C-linear category in which there is a perfect
pairing of Hom(a, b) with Hom(b, a) for all a, b ∈ Ob(C) by a pairing which factorizes
through the composition in either order.

Remark: It is important to note that the argument for commutativity fails in the
open case: The algebras Oaa are in general noncommutative.

Figure 24: The open-closed transition maps fig:openclosed

So, to give an open and closed TFT involves giving a Frobenius cateogry. But the
open and closed strings must be related to each other. The essential new information is a
pair of linear maps

ιa : C → Oaa
a
(20.30) eq:linmaps
ι : Oaa → C

corresponding to the open-closed string transitions of 24.

By drawing pictures we can readily discover the following necessary algebraic condi-
tions:
1. ιa is an algebra homomorphism

– 87 –
Figure 25: Factorization of the open string loop on closed string exchange. Also known as the
“Cardy condition.” fig:cardycond

ιa (φ1 φ2 ) = ιa (φ1 )ιa (φ2 ) (20.31) eq:centerh

2. The identity is preserved

ιa (1C ) = 1a (20.32) eq:ident

3. Moreover, ιa is central in the sense that

ιa (φ)ψ = ψιb (φ) (20.33) eq:center

for all φ ∈ C and ψ ∈ Oab

4. ιa and ιa are adjoints:

θC (ιa (ψ)φ) = θa (ψιa (φ)) (20.34) eq:adjoint

for all ψ ∈ Oaa .

5. The “Cardy conditions.”7 Define πb a : Oaa → Obb as follows. Since Oab and Oba
are in duality (using θa or θb ), if we let ψµ be a basis for Oba then there is a dual basis ψ µ
for Oab . Then we define
X
πb a (ψ) = ψµ ψψ µ , (20.35) eq:dblttw
µ

and we have the “Cardy condition”:

a
πb = ιb ◦ ι a . (20.36) eq:cardycon

This is illustrated in 25.

Exercise
7
These are actually generalization of the conditions stated by Cardy. One recovers his conditions by
taking the trace. Of course, the factorization of the double twist diagram in the closed string channel is an
observation going back to the earliest days of string theory.

– 88 –
 Draw pictures associated to the other algebraic conditions given above.

Theorem Open-Closed Sewing Theorem. The above conditions are the complete set
of sewing constraints on the algebraic data.

This is proved in the reference cited above.

Example: Let G be a finite group. In section 8 we saw that the class functions
on G form a commutative Frobenius algebra. Now we take our category of boundary
conditions B to be the category of finite dimensional complex representations V of G, with
OV V = End(V ) and θV : OV V → C given by

1
θV (ψ) = Tr(ψ) (20.37)
|G|

Exercise
Write out the full set of open-closed string data for the example of a finite group and
check the sewing conditions.

21. Applications of the Projection Operators

Let us return to the projection operators in C[G]. In this section we will assume - as we
may - that the irreps T µ are unitary so

(µ) (µ)
T̂ij (g) = (Tij (g))∗ (21.1)

Thus we have

nµ X (µ)
Pijµ = (Tij (g))∗ g (21.2) eq:projopsa
|G|
g∈G

0
Pijµ Piµ0 j 0 = δµµ0 δji0 Pijµ0 (21.3) eq:prodlawa

(µ)
g · Pijµ = µ
X
Tsi (g)Psj (21.4) eq:geerepa
s

(µ)
Pijµ · g −1 = (Tsj (g))∗ Pisµ
X
(21.5) eq:geerepiia
s

– 89 –
21.1 Decomposition of a representation into its isotypical parts
While characters tell us the irreps into which a representation can be decomposed, they do
not tell us how to decompose the representation explicitly. To do this, we need the method
(µ)
of projection operators. We assume that the matrix elements Tij are known.
For simplicity we consider finite groups | G |< ∞, but the same techniques extend to
compact groups.
Now, let T (g) be any representation on V . Extending by linearity we obtain a rep-
resentation of the algebra RG . Define operators on V by applying T to (19.12) to get
8

nµ X (µ)
Pijµ = T (Pijµ ) = (Tij (g))∗ T (g) (21.6)
|G|
gG

Now we imediately get the properties (19.14), (19.15), (19.16) for the operators Pijµ .
Therefore, if Pµlk~v = w
~ l 6= 0 then
n o
~ j } = Pµjk~v
{w (21.7)
j=1,···nµ

span an invariant subspace of V transforming in the rep T (µ) according to the representation
T µ . Using these operators we can in principle decompose a representation into its irreps.

Exercise Calculating projection operators

Now we return to the example of the 3-dimensional reducible representation of S3
acting on R3 in section *** above.
Referring to the three irreducible representations of S3 , called 1+ , 1− , 2 in section ****
show that:

 
111
1
P1+ = 1 1 1 P12+ = P1+

3
111
 
000
P1− = 0 0 0
 
000
  (21.8)
0 0 0
P21,1 = 0 1/2 −1/2 (P21,1 )2 = P21,1
 
0 −1/2 1/2
 √ √ 
0 1/ 3 −1/ 3
√ √ 
P22,1 = 0 −1/2 3 1/2 3 

√ √
0 −1/2 3 1/2 3
8 (µ)
We choose Tij = unitary matrices for simplicity.

– 90 –
 Therefore,    
x 1
x+y+z  
P1+ y  = ( ) 1 (21.9)
 
3
z 1
so:
n o
V1 = Span V~1 = (1, 1, 1) = {(x, x, x) : x ∈ R} (21.10)
Similarly P2 projects onto the orthogonal subspace
       
x 0 x √ 2
1,1   1 2,1   3
P2 y  = (y − z)  1  P2 y  = (y − z) −1 (21.11)
   
2 2
z −1 z −1

So: n o
V2 = Span V~2 = (0, 1, −1)V
~3 = (2, −1, −1) (21.12)
are the two invariant subspaces. Check
 
1 1 1
S = 0 1 −1 (21.13)
 
2 −1 −1

Conjugates the rep into block form. Thus, the decomposition into irreps is:

R3 ∼
= V1+ ⊕ V2 (21.14) eq:expldec

An important shortcut: In general the characters of irreps are much easier to calculate
than the matrix elements. Therefore the following route to decomposition is an important
shortcut:
If
V ∼
= ⊕µ aµ T (µ) (21.15)
then we can easily project onto the isotypical component as follows: Note that for unitary
T (g), (Pµij )∗ = Pµij , so
nµ
Piiµ
X
P µ := (21.16) eq:charporj
i=1
form a system of orthogonal, Hermitian, projection operators:

(Pµ )2 = Pµ
(Pµ )† = Pµ (21.17) eq:projopsaa
Pµ Pν = 0 µ 6= ν

– 91 –
They project onto the invariant subspaces V (µ) .
The point is: from the definition (21.16) we get:

nµ X
Pµ = χµ (g)T (g) (21.18) eq:calcpmu
|G|
g∈G

so, we need only know the characters to compute the Pµ .

Example: In the above example, one easily computes
 
2 −1 −1
1
P2 = −1 2 −1 (21.19)

3
−1 −1 2

21.2 Block diagonalization of Hermitian operators

Suppose H is an N × N Hermitian operator acting on the inner product space CN with

standard innder product. Suppose we have a group representation T of G with CN as
carrier space and that moreover, G is a symmetry of H in the sense that

∀g ∈ G [T (g), H] = 0 (21.20)

Then
[Pijµ , H] = 0 (21.21)

Then, the projection operators Pλ onto the different eigenspaces of H commute with the
different projection operators Pijµ and in particular with P µ .
Therefore, in this situation, by reducing CN to the irreps of G, we have partially block-
diagonalized H.
Of course, this is very useful in quantum mechanics where one wants to diagonalize
Hermitian operators acting on wavefunctions.

21.2.1 Projecting quantum wavefunctions

Suppose G acts on X as a tranformation group, and {ψa } is a collection of functions on X
transforming according to some representation T of G. (For example, energy eigenfunctions
in a Schrödinger problem.)
Then:

nµ X µ
ij
ψa,µ (x) ≡ (Tij (h))∗ ψa (h−1 · x) (21.22)
|G|
hG

for fixed µ, j, a, letting i = 1, . . . , nµ be a collection of functions transforming according to

the irrep T µ : This is a special case of the general statement we made above. Here, again,
is the explicit calculation:

– 92 –
 nµ X µ
ψaij (g −1 · x) = (Tij (h))∗ ψa (h−1 · g −1 · x)
|G|
hG
nµ X µ −1 ∗
= (Tij (g h)) ψa (h−1 · x)
|G|
hG
nµ (21.23)
(Tisµ (g −1 ))∗ ψasj (x)
X
=
s=1
nµ
Tsiµ (g)ψasj (x)
X
= if T µ is unitary.
s=1

Example 2 As a somewhat trivial special case of the above consider Z acting on R x→

−x therefore acts on F un(R). Decompose into irreps:

ψ = function on R

1
ψ 1 (x) = (ψ(x) + ψ(−x))
2 (21.24)
2 1
ψ (x) = (ψ(x) − ψ(−x))
2
transform acccording to the two irreps of Z2 .

21.2.2 Finding normal modes in classical mechanics

Consider a classical mechanical system with degrees of freedom

~q = (q 1 , . . . , q n ) (21.25)

Suppose we have a generalized harmonic oscillator so that the kinetic and potential energies
are:
1 1
T = mij q̇ i q̇ j V = Uij q i q j (21.26)
2 2
where Uij is independent of q i . We can obtain solutions of the classical equations of motion
by taking
q i (t) = Re(γ i eiωt ) (21.27)
where
(−ω 2 mij + Uij )γ j = 0 (21.28)
If a group G acts as (T (g) · q)i = T (g)ij q j and

[T (g), m] = [T (g), U ] = 0 (21.29)

then we can partially diagonalize H = −ω 2 m + U by choosing ~γ to be in irreps of G.

Consider n beads on a ring connected by springs of the same force constant and hence
the same frequency ω as in 26.

– 93 –
 Figure 26: A system of beads and springs. fig:normalmode

The Lagrangian is
1 1 X i
L = m(q̇ i )2 − k (q − q i+1 )2 (21.30)
2 2
i

where we understand the superscript on q i to modulo n: q i+n = q i .

We will consider the system where the angle coordinates q i have been lifted to the
universal cover, i.e. q i ∈ R so that we can consider ~q to be in a vector space. 9
The normal mode problem is

(−mω 2 + kA)γ = 0 (21.31)

where  
2 −1 0 0 ··· ··· −1
−1 2 −1 0 0 0 ···
 
A= 0 −1 2 −1 0 ··· 0  (21.32)
 
 
· · · ··· ··· ··· ··· ··· −1
−1 0 ··· 0 0 −1 2

Thus, we reduce the problem of finding normal modes to the problem of diagonalizing
this matrix.
The problem has an obvious Zn symmetry where a generator T (ω) takes

T (ω) : q 1 e1 + · · · q n en → q 1 e2 + · · · + q n en−1 + q 1 en (21.33)

Here ei is the standard basis for Cn so T (ω ` )ei = ei+` and again the subscript is understood
modulo n.
We know the irreps of Zn : T µ (ω j ) = ω jµ , µ = 0, . . . , n − 1. The projection operators
are thus
n−1
µ 1 X −µ`
P = ω T (ω ` ) (21.34)
n
`=0

9
This alters the problem physically. We should add a constraint which says, roughly speaking that
(q i+1 − q i ) = 2π. We will ignore this constraint in what follows.
P

– 94 –
In terms of matrix units eij we have
1 X −µ`
Pµ = ω ei+`,i (21.35)
n
i,`

On the other hand, X

A=2− (ei,i+1 + ei+1,i ) (21.36)
i
Now we carry out the multiplication:
X
AP µ = 2P µ − ω −µ` (ei+`,i+1 + ei+`,i−1 ) (21.37)
i,`

After some relabeling of the subscripts we easily find

AP µ = (2 − ω −µ − ω µ )P µ
πµ 2 µ (21.38)
= (2 sin ) P
n
Similarly,
 
1 X 1 X X
P µ ~q = ω −µ` q s es+` = ω µs q s  ω −µj ej  (21.39)
n n s
`,s j

So the normal mode frequencies are

k πµ 2
ωµ2 = 4 (2 sin ) (21.40)
m n
and the normal modes are:
X
~q(µ) = Re(αµ eiωµ t ω −µj ej ) (21.41)
j

Note that there is a degeneracy in the mode frequencies

ωµ = ωn−µ . (21.42) eq:funnydegen

It is a general principle that when you have unexplained degeneracies you should search
for a further symmetry in the problem.
Indeed, in this case, we have overlooked a further symmetry:

q i → q n−i (21.43)

This leads to Dn symmetry of the problem which explains the degeneracy (21.42).

Exercise
Using the character tables above for Dn construct the normal modes which are in
representations of Dn .

– 95 –
 22. Representations of the Symmetric Group

22.1 Conjugacy classes in Sn

Recall from chapter one that the conjugacy classes in Sn are labeled by partitions of n: A
cycle decomposition has the form

(1)ν1 (2)ν2 · · · (n)νn (22.1)

where νj is the number of cycles of length j. Clearly since we must account for all n letters
being permuted: X
n = ν1 + 2ν2 + · · · + nνn = jνj (22.2)
To every cycle decomposition we can associate a partition as follows:

λ1 := ν1 + ν2 + · · · + νn
λ2 := ν2 + · · · + νn
λ3 := ν3 + · · · + νn (22.3) eq:lambdaTOnu
······
λn := νn

Note
1. λ1 ≥ λ2 ≥ · · · ≥ λn ≥ 0
P
2. λi = n
3. Given a partition of n we can recover the cycle decomposition (j)νj since the
Q

transformation (22.3) is invertible over the integers: νj = λj − λj+1 .

As we saw when discussing symmetric functions, to a partition of n we can uniquely
associate a picture, called a Young diagram, consisting of rows of λi boxes, for a total of n
boxes.

classes, and hence the irreps of Sn , are in 1-1 correspondence with the partitions of n and hence in 1-1 correspondence with Y

22.2 Young tableaux

Young diagrams can be used to define projection operators onto the irreducible represen-
tations in the group algebra C[Sn ] using the following procedure. We only give the barest
minimal description here. See the references at the end of this section for a full account.
Define a Young tableau to be a Young diagram in which each of the integers 1, 2, . . . , n
has been placed into a box. These integers can be placed in any order. A standard tableau
is one in which the integers are increasing in each row, left to right, and in each column,
top to bottom. For any given partition there are n! different Young tableau.
Let T denote a Young tableau. Then we define two subgroups of Sn , R(T ) and C(T ).
R(T ) consists of permutations in Sn which only permute numbers within each row of T .

– 96 –
 λ = (2,0)

λ = (1,1)

λ = (3,0,0)

λ = (2,1,0)

λ = (5,3,2,1,1,0,0,0,0,0,0,0)

λ = (1,1,1)

Figure 27: Examples of Young diagrams. fig:youngone

Similarly, C(T ) consists of permutations which only permute numbers within each column
of T .
For a Young tableau T define the following elements in the group ring Z[Sn ]:
X
P := p (22.4)
p∈R(T )
X
Q := (q)q (22.5)
q∈C(T )
X
P (T ) := P Q = (q)pq (22.6)
p∈R(T ),q∈C(T )

Then we have the following rather nontrivial statements:

1. P (T )2 = cT P (T ) where cT is an integer.
2. P (T )P (T 0 ) = 0 if T and T 0 correspond to different partitions.
3. If T, T 0 are two different standard tableaux corresponding to the same partition
then P (T )P (T 0 ) = 0.
4. P (T ) projects onto an irreducible representation of Sn . That is C[Sn ] · P (T ) trans-
forms as an irreducible representation R(T ) of Sn in the left-regular representation. All
of the irreducible representations are equivalent to one of these. Representations R(T ) for
different partitions are inequivalent, while two representations for the same partition are
equivalent.
5. The integer cT in P (T )2 = cT P (T ) is given by
n!
cT = (22.7)
dimR(T )

– 97 –
dimR(T ) can also be characterized as the number of standard tableaux corresponding to
the underlying partition.
6. Another formula for dimR(T ) is the hook length formula. For a box in a Young
diagram define its hook length to be the number of squares to the right in its row, and
underneath in its column, counting that box only once. Then
n!
dimR(T ) = Q (22.8)
hooklengths

Figure 28: Four standard Young tableaux for S3 . fig:exmplesthr

22.2.1 Example 1: G = S3
There are 3 partitions, 3 Young diagrams, and 4 standard tableaux shown in 28.
2
P
Clearly P (T1 ) = p∈S3 p and P (T1 ) = 6P (T1 ). Moreover Z[S3 ] · P (T1 ) is one-
dimensional. This is the trivial representation.
P
Similarly, P (T4 ) = (q)q spans the one-dimensional sign representation.
There are two standard tableaux corresponding to λ = (2, 1) with

P (T2 ) = (1 + (12))(1 − (13)) = 1 + (12) − (13) − (132) (22.9) eq:projteetwo

P (T3 ) = (1 + (13))(1 − (12)) = 1 − (12) + (13) − (123) (22.10)

One checks that P (T2 )2 = 3P (T2 ) and P (T3 )2 = 3P (T3 ) and P (T2 )P (T3 ) = 0. These
are consistent with the above statements about cT and the hooklength formula.
Now consider Z[S3 ] · P (T2 ). We compute
(12) · P (T2 ) = P (T2 ) := v1
(13) · P (T2 ) = −1 + (13) − (23) + (123) := v2
(23) · P (T2 ) = −(12) + (23) − (123) + (132) = −v1 − v2 (22.11)
(123) · P (T2 ) = −1 + (13) − (23) + (123) = v2
(132) · P (T2 ) = −(12) + (23) − (123) + (132) = −v1 − v2
Thus the space is two-dimensional.
From this it is easy to compute: (12) · v2 = −v1 − v2 , (13)v2 = v1 and (23)v2 =
(123)P (T2 ) = v2 and so in this basis we have a representation generated by
! ! !
1 −1 01 −1 0
ρ(12) = ρ(13) = ρ(23) = (22.12)
0 −1 10 −1 1

– 98 –
One easily checks that the character is the same as that we computed above for the 2.
Of course, similar statements hold for P (T3 ), which the reader should check as an
exercise.

Figure 29: 10 standard Young tableaux for S4 . fig:exmplesthr

22.2.2 Example 2: G = S4
References
1. For further discussion of the above material see the books by Miller, Hammermesh,
Curtis+Reiner, Fulton+Harris Representation Theory.
2. There is a second, elegant method for constructing the irreps of the symmetric
group using induced representations. See the book by Sternberg for an account.

23. Symmetric groups and tensors: Schur-Weyl duality and the irreps of
GL(d, k)

Consider a vector space V of dimension d.

In previous sections we have seen that

V ⊗n ≡ V ⊗ V ⊗ · · · ⊗ V (23.1)

is a representation of Sn by permuting the factors:

σ · (w1 ⊗ w2 ⊗ · · · ⊗ wn ) = wσ(1) ⊗ wσ(2) ⊗ · · · ⊗ wσ(n) (23.2)

*** ?? should we use σ −1 here ?? ***

Now recall that we also characterized vectors in V ⊗n as tensors. If V is a representation
of a matrix subgroup G of GL(d, R) or GL(d, C) then V ⊗n is also a representation of G
via
ρ(g) · (w1 ⊗ w2 ⊗ · · · ⊗ wn ) = g · w1 ⊗ g · w2 ⊗ · · · ⊗ g · wn (23.3)

– 99 –
 For example, suppose that V = k d is also the fundamental representation of GL(d, R)
or GL(d, C) for k = R or k = C. Given a basis {vi } for V a typical element can be expanded
in the basis as: X
v= ti1 ,i2 ,...,in vi1 ⊗ · · · ⊗ vin (23.4)
i1 ,i2 ,...,in

We will often assume the summation convention where repeated indices are automatically
summed. Under the action of GL(d, k):

g · vi = gji vj (23.5)

we therefore have:
(g · t)i1 ···in = tj1 ···jn gj1 i1 · · · gjn in (23.6) eq:tensor

that is, elements in V ⊗n transform as tensors and V ⊗n is also a representation of

GL(d, R): On the other hand
i ···iσ−1 (n)
(σ · t)i1 ...in = t σ−1 (1) (23.7)

Now, note that the actions of GL(d, k) and Sn commute, so that V ⊗n is a representation
of GL(d, k) × Sn .
Considered as a representation of Sn , we have complete reducibility:

V ⊗n ∼
= ⊕λ Vλ ⊗ Rλ (23.8)

where Rλ is the irrep of Sn corresponding to the partition λ. By Schur’s lemma we can

characterize the multiplicity space Vλ as

Vλ = HomSn (Rλ , V ⊗n ). (23.9)

This is the vector space of Sn -equivariant maps from Rλ to V ⊗n .

Now, because GL(d, k) and Sn commute we know that Vλ will also be a representation
of GL(d, k).
The beautiful result of Schur and Weyl – known as the Schur-Weyl duality theorem –
is

Theorem The representations Vλ are irreducible representations of GL(d, k) and, up to

a power of the determinant representation, all irreducible representations may be obtained
in this way. Moreover, these statements hold for U (d) ⊂ GL(d, C).

Proof For complete proofs see Segal, Miller, Fulton-Harris. We will just sketch the
idea from Segal’s lectures.
We can see fairly easily why all the representations of U (d) must occur if we accept
the Peter-Weyl theorem. Note that the representations which we obtain from V ⊗d have
matrix elements which are algebraic expressions in the matrix elements aij . Indeed, they
lie in C[aij ] (the aij occur individually of course from V itself). But this forms a dense
subalgebra of L2 (U (d)).

– 100 –
 Now let us prove the Vλ is irreducible. Recall that the center of a matrix algebra
End(W ) is just C. Now we note that

EndSn (V ⊗n ) = ((End(V ))⊗n )Sn (23.10)

Since Gl(d) forms a dense open set in End(V ), the operators T (g) acting on V ⊗n generate
an algebra which is all of EndSn (V ⊗n ). However, this means that the operators that
commute with both G and Sn , i.e.

EndG×Sn (V ⊗n ) (23.11)

must lie in the center of EndSn (V ⊗n ). On the other hand,

EndSn (V ⊗n ) = ⊕λ End(Vλ ) (23.12)

and
EndG×Sn (V ⊗n ) = ⊕λ EndG (Vλ (23.13)

Therefore EndG (Vλ must be contained in the center of End(Vλ ), but the latter is just C.
Therefore EndG (Vλ is just C. But, by Schur’s lemma, this must mean that Vλ is irreducible.
♠

We can construct Vλ explicitly by taking a Young diagram and its corresponding

symmetrizer P (T ) (for some tableau T ) and taking the image of P (T ) acting on V ⊗n .
This projects onto tensors of a definite symmetry type. These tensors will transform in
irreducible representations of GL(d, k). As representations of GL(d, k) we have

P (T )V ⊗n ∼
= P (T 0 )V ⊗n (23.14)

if T and T 0 correspond to the same partition, i.e. the same Young diagram.
With some work it can be shown that - as representations of GL(d, k) we have an
orthogonal decomposition
V ⊗n = ⊕T P (T )V ⊗n (23.15)

where T runs over the standard tableaux of Sn . Of course, tableaux with columns of length
≥ d will project to the zero vector space and can be omitted.

Example
• If we take the partition λ = (n, 0, 0, . . . ) then P (λ) projects to totally symmetric
tensors and we get S n (V ). Weyl’s theorem tells us that if V = k d is the fundamental
representation of GL(d, k) then S n (V ) is an irreducible representation. We computed
before that its dimension is  
n n+d−1
dimS (V ) = (23.16)
n
Note in particular that we have infinitely many irreducible representations.

– 101 –
 • If we take the partition λ = (1, 1, . . . , 1) then P (λ)V ⊗n = Λn V is the subspace of
totally antisymmetric tensors of dimension
 
d
dimΛn (V ) = (23.17)
n
Note that this subspace vanishes unless dimV = d ≥ n.
• Let V be of dimension d and consider V ⊗3 as a representation of S3 . Using characters
we showed above that
d(d + 1)(d + 2) d(d − 1)(d − 2) d(d + 1)(d − 1)
V ⊗3 ∼
= 1+ ⊕ 1− ⊕ 2 (23.18)
6 6 3
Clearly, the first and second summands correspond to the totally symmetric and totally
antisymmetric tensors, respectively.
Applying the projector (22.9) to a tensor tijk vi ⊗ vj ⊗ vk produces a tensor with mixed
symmetry:
P (T2 )vi ⊗ vj ⊗ vk = vi ⊗ vj ⊗ vk + vj ⊗ vi ⊗ vk − vk ⊗ vj ⊗ vi − vk ⊗ vi ⊗ vj (23.19)
and therefore the components are of the form:
tijk + tjik − tkji − tjki (23.20)
for arbitrary tijk . This tensor space is the space of tensors t̃ijk which satisfy the identities:

t̃ijk + t̃jki + t̃kij = 0 (23.21) eq:cyclsym

together with

t̃ijk = −t̃kji (23.22) eq:iksymm

The first set of equations (23.21) cuts out a space of dimension 32 d(d2 − 1). The next
set of equations (23.22) cuts it down by half so we get 13 d(d2 − 1) as the dimension of this
space of tensors.
Similarly,
P (T3 )vi ⊗ vj ⊗ vk = vi ⊗ vj ⊗ vk − vj ⊗ vi ⊗ vk + vk ⊗ vj ⊗ vi − vj ⊗ vk ⊗ vi (23.23)
and therefore the components are of the form:
tijk − tjik + tkji − tkij (23.24)
for arbitrary tijk . Thus, this tensor space is the space of tensors t̃ijk which satisfy the
identities:
t̃ijk + t̃jki + t̃kij = 0 (23.25)
t̃ijk = −t̃jik (23.26)

Remarks The algebra of operators A generated by linear combinations of σ acting on

V ⊗n commutes with the algebra of operators B generated by g ∈ GL(d, k) acting on V ⊗n .
In fact these algebras are full commutants of each other: Any operator commuting with A
is in B and vice versa.

– 102 –
23.1 Free fermions on a circle and Schur functions

Figure 30: Fermi sea and particle/hole excitations. fig:Fermisea

Consider a system of N free fermions on a circle z = eiθ , θ ∼ θ + 2π, with Hamiltonian

P d2
H = − i dθ 2 . The one-particle wavefunctions are
i

ψn (z) = z n (23.27) eq:onpart

with energy n2 . Consider a state of N fermions occupying levels ni . By Fermi statistics

these levels must all be distinct and we can assume

nN > nN −1 > · · · > n1 (23.28)

The N -fermion wavefunction is the “Slater determinant”

n
Ψ~n (z1 , . . . , zN ) = det1≤i,j≤N zi j (23.29) eq:slater

and has energy i n2i . States can be visualized as in 30.

P

For simplicity let us assume N is odd. Then there is a unique groundstate obtained
by filling up the states −nF ≤ n ≤ nF with

N −1
nF = (23.30) eq:fermil
2
(If N is even there are two groundstates.) This defines the Fermi sea. The ground
state wavefunction is

Ψgnd (z1 , . . . , zN ) = det1≤i,j≤N zij−1−nF = (z1 . . . zN )−nF ∆0 (z) (23.31) eq:grnd

The ratio of the wavefunction to groundstate wavefunction

n
Ψ~n det1≤i,j≤N zi j
= (23.32) eq:srat
Ψgnd det1≤i,j≤N zij−1−nF
is a totally symmetric function of the zi . If n1 ≥ −nF then it is a symmetric polynomial
known as a Schur function.

– 103 –
 The Schur functions form a linear basis for Z[x1 , . . . , xN ]SN where basis elements are
associated with partitions. To define them mathematically let us return to a monomial
xα = xα1 1 · · · xαNN with the αi ≥ 0 and now try to skew-symmetrize:
X
(σ)σ · xα (23.33)
σ∈SN

Note that this function is totally antisymmetric, and moreover vanishes unless the αi are
all distinct. WLOG assume that

αN > αN −1 > · · · > α1 ≥ 0. (23.34)

The “smallest” such α is

N − 1 > N − 2 > ··· > 1 > 0 ≥ 0 (23.35)

defining a vector δ := (N − 1, N − 2, . . . , 1, 0) i.e. δj = j − 1, 1 ≤ j ≤ N . Using δ we define

λ by
α=λ+δ (23.36)
Note that λj = αj − δj ≥ 0 and moreover

λj + j − 1 > λj−1 + j − 2 ⇒ λj ≥ λj−1 (23.37)

Thus, λ is a partition. It is more convenient to use λ than α so we define:

X
∆λ (x) := (σ)σ · xλ+δ (23.38)
σ∈SN

Note that this can also be written as a determinant:

λ +j−1
∆λ (x) := det(xi j )
· · · xλ1 N +N −1
 λ1 λ2 +1 
x1 x1
 · · · · 
  (23.39) eq:deltsun
= det  · · · ·
 

 
 · · · · 
λ1 λ2 +1 λN +N −1
xN xN · · · xN

Note that ∆0 (x) is the famous Vandermonde determinant:

Y
∆0 (x) := (xi − xj ) (23.40)
i>j

Exercise
Show that
∆λ (σ · x) = (σ)∆λ (x) (23.41)

– 104 –
where (σ) is the sign homomorphism.

Now we define the Schur function to be the ratio:

∆λ (x)
Φλ (x) := (23.42) eq:schurf
∆0 (x)
This is a totally symmetric function. In fact, it is a polynomial in the xi . To see
this consider it as a meromorphic function of a complex variable x1 . Note that it is in
fact an entire function since the potential poles at x1 = xi are cancelled by zeroes of the
numerator. Next note that the growth at infinity is obviously xm 1 for an integer m.

Example. Put N = 2. Then λ2 ≥ λ1 ≥ 0,

∆λ (x) = (x1 x2 )λ1 (xλ2 2 −λ1 +1 − xλ1 2 −λ1 +1 ) (23.43) eq:deltatwo

xλ2 2 −λ1 +1 − xλ1 2 −λ1 +1

Φλ (x1 , x2 ) = (x1 x2 )λ1 = (x1 x2 )λ2 hλ2 −λ1 (x1 , x2 ) (23.44) eq:phitwo
x2 − x1
Returning to the free fermion example, we have

Ψ~n ∆λ (z)
= (23.45) eq:fration
Ψ~0 ∆0 (z)
for
αj = nj + nF
, (23.46)
= λj + j − 1

that is

λj = (nj + nF ) − (j − 1) (23.47) eq:snet

Remarks
• The second main theorem of symmetric polynomials is:

Theorem. The Schur functions Φλ (x) form a linear integral basis for Z[x1 , . . . , xN ]SN .

That is, any symmetric polynomial with integral coefficients can be written as a lin-
ear combination, with integral coefficients, of the Φλ (x). Note that we are not forming
polynomials in the Φλ .
• The module AN of skew-symmetric polynomials in x1 , . . . , xN is isomorphic to ΛN
via multiplication by ∆0 . Therefore the ∆λ form a a Z-basis for AN . This close relation
between completely symmetric and antisymmetric functions comes up in the theory of
matrix models – integrals over space of N × N matrices. It also suggests a relation between
bosons and fermions, at least in 1+1 dimensions. That indeed proves to be the case – there

– 105 –
is a nontrivial isomorphism known as bosonization which is an isomorphism of quantum
field theories of bosons and fermions in 1+1 dimensions.
•
There is an elegant expression for the Schur functions Φλ as a determinant of a matrix
whose entries involve the hk . The expression is

Φλ = det(hλi −i+j ) (23.48) eq:phiach

where we take an N × N matrix for N variables and it is understood that h0 = 1 and

hi = 0 for i < 0. Equivalently

Φλ = det(eλ0i −i+j ) (23.49) eq:phiacha

For a proof see Macdonald, p.41.

Example Consider the partition (λ1 , λ2 , 0, 0, . . . ). Then applying (23.48) we get

 
hλ1 hλ1 +1 hλ1 +2 hλ1 +3 hλ1 +4 ...
h
 λ2 −1 hλ2 hλ2 +1 hλ2 +2 hλ2 +3 ...
Φλ =  0 0 1 h1 h2 ···
 
  (23.50) eq:phitwoa
 0 0 0 1 h1 ···
··· ··· ··· ··· ··· ···
= hλ1 hλ2 − hλ1 +1 hλ2 −1

With a little algebra one can confirm that this is indeed (23.44). Note, however, that
both sides of this equality make sense for N > 2.

Figure 31: A matrix for defining Schur functions. fig:symsq

• Define the Z × N matrix Ξ whose (p, q)th entry is hq−p where h0 = 1 and hk = 0 for
k < 0. Here −∞ < p < ∞ labels the rows, while q = 1, 2, . . . labels the columns. Thus the
matrix looks like 31.
Define the integers sk = k − λk . These eventually become sk = k. Then Φλ is the
determinant of the matrix formed from Ξ by keeping only the rows labeled by sk . This
leads to an upper triangular matrix is the determinant of the finite matrix above.

– 106 –
 • The definition of Φλ only makes sense as a function of N such that λi = 0 for i > N .
However, the relations between Φλ and the hi are universal.

Exercise
Verify that (23.50) and (23.44) are equal for N = 2.

Exercise
a.) Let λ = (n, 0, 0, 0, . . . ). Show that Φλ = hn .
b.) Let λ = (1n , 0, 0, . . . ). Then Φλ = en .

Exercise
Using Cauchy’s determinant identity:
 
1 ∆0 (xi )∆0 (yj )
det =Q (23.51)
1 − x i yj i,j (1 − xi yj )

Show that X  X
1
exp sk (x)sk (y) = Φλ (x)Φλ (y) (23.52)
k
k λ
Hint: Multiply both sides by ∆0 (x)∆0 (y). For the answer see Hammermesh, p. 195

Exercise Laughlin’s wavefunctions

Another natural place where symmetric wavefunctions appear is in the quantum Hall
effect, where the 1-body wavefunctions in the lowest Landau level are
1 1 2
ψn (z) = √ z n e− 2 |z| (23.53)
πn!
Laughlin introduced a fascinating set of approximate eigenfunctions of 2d interacting
electrons in a magnetic field:
Y 1 P 2
Ψ= (zi − zj )2n+1 e− 2 |zi | (23.54)
i<j

Express these in terms of Schur functions.

Many other interesting trial wavefunctions in the FQHE can be generated using theo-
rems about symmetric functions.

– 107 –
23.1.1 Schur functions, characters, and Schur-Weyl duality
Now it is interesting to combine Schur functions with Schur-Weyl duality.
We have seen that irreducible representations of GL(d, C) (and of U (d) ) can be labeled
by Young diagrams with ≤ d rows. We called these Sλ above.
It turns out that the Φλ (x) are characters of the representations Sλ . That is if g ∈
GL(d, C) can be diagonalized to Diag{x1 , . . . , xd } then

TrVλ ρ(g) = Φλ (x1 , . . . xd ) (23.55)

This is known as the Weyl character formula and will be derived later.
The relation between the Φλ and the power functions sk (x) is a very nice application
of the Schur-Weyl duality theorem.
Suppose that a group element σ ∈ SN has cycle decomposition (1)`1 (2)`2 · · · where
P
j`j = N . As we have discussed this determines a conjugacy class C[`] as well as a
partition. Now, suppose g ∈ GL(d, C) is diagonalizable to Diag{x1 , . . . , xd } and let us
evaluate

TrV ⊗N (σg) (23.56) eq:keytrace

where V = Cd is the fundamental representation of GL(d, C).

We will evaluate (23.56) in two ways. On the one hand, it is clear that this should just
be

(Tr(g))`1 (Tr(g 2 ))`2 (Tr(g 3 ))`3 · · · (23.57) eq:prodns

On the other hand, since

V ⊗N ∼
= ⊕λ Sλ ⊗ Rλ (23.58)
the trace must also be X
χλ (σ)Φλ (x) (23.59)
λ
P j
Thus, recalling the symmetric functions which are the power sum functions sj = i xi
we have the following identity due to Frobenius:
Let (`) = (`1 , `2 , . . . , `N ) denote a tuple of nonnegative integers. Then let

s(`) := s`11 s`22 . . . s`NN (23.60)

Then:
X
s(`) = χλ (C[`]))Φλ (23.61) eq:frobenius
{λ}

Returning to the free fermion interpretation we see that the operator

Y
O(`) = (Tr(g j ))`j (23.62)
j

– 108 –
acts on the groundstate wavefunction Ψgnd (x) to produce the quantum states:
X
O(`)Ψgnd (x) = χλ (C[`]))Ψ~n (23.63) eq:transamp
λ

where the ni are the fermion occupation numbers corresponding to the partition asso-
ciated to λ as in (23.47). Thus the characters of the symmetric group can be regarded as
“transition amplitudes” in the free fermion problem!
Indeed, we can now write the character table of the symmetric group SN :
 
χλ (C[`]) = ∆0 (x)s(`) (x) α (23.64) eq:charactsymm
1 ,...,αd

where the subscript means we extract the coefficient of xα1 1 · · · xαd d and we recall that

αj = λj + j − 1 (23.65)

This is the famous formula of Frobenius.

Remark: If we think of Φλ as a character of a representation of U (N ). Then the

Hamiltonian H is closely related to the quadratic Casimir of that representation. That in
turn is related to the solution of two-dimensional Yang-Mills theory. See Cordes et. al. eq.
(4.3)

Exercise Dimensions of irreps of SN

a.) Using the Frobenius formula show that the character of 1, and hence the dimension
of Rλ is
N! Y
dimRλ = (αi − αj ) (23.66)
α1 ! · · · αN !
i>j

b.) Using this formula, derive the hook-length formula.

Warning: This is hard.

Exercise Weyl dimension formula

Show that
Y (λi − λj + j − i)
Φλ (1, 1, . . . , 1) = (23.67) eq:weyldim
(j − i)
1≤i<j≤N

We will see later that this is a special case of the Weyl dimension formula.
Hint: Put xj = ejt and let t → 0.

References:

– 109 –
 There are a lot of beautiful connections here with quantum field theory. In particular
there are close connections to two-dimensional Yang-Mills theory and to the phenomenon
of bosonization in two dimensions. See
For much more about this see the references below.
1. Pressley and Segal, Loop Groups, Chapter 10.
2. M. Stone, “Schur functions, chiral bosons, and the quantum-Hall-effect edge states,”
Phys. Rev. B42 1990)8399 (I have followed this treatment.)
3. S. Cordes, G. Moore, and S. Ramgoolam, “Lectures on 2D Yang-Mills Theory,
Equivariant Cohomology and Topological Field Theories,” Nucl. Phys. B (Proc. Suppl
41) (1995) 184, section 4. Also available at hep-th/9411210.
4. M. Douglas, “Conformal field theory techniques for large N group theory,” hep-
th/9303159.
5. Papers of Jimbo, Miwa, ..... (Kyoto school) on Toda theory.

– 110 –
 *****************************************
FOLLOWING MATERIAL SHOULD BE MOVED TO CHAPTER ON REPRESEN-
TATIONS. IT WOULD MAKE MORE SENSE TO TALK ABOUT REPS OF THE SYM-
METRIC GROUP FIRST.

23.2 Bosons and Fermions in 1+1 dimensions

23.2.1 Bosonization
Finally, we want to describe a truly remarkable phenomenon, that of bosonization in 1+1
dimensions.
Certain quantum field theories of bosons are equivalent to quantum field theories of
fermions in 1+1 dimensions! Early versions of this idea go back to Jordan 10 The subject
became important in the 1970’s. Two important references are
S. Coleman, Phys. Rev. D11(1975)2088
S. Mandelstam, Phys. Rev. D11 (1975)3026
The technique has broad generalizations, and plays an important role in string theory.
To get some rough idea of how this might be so, let us consider the loop group S 1 →
U (1). For winding number zero, loop group elements can be written as:

+∞
X
z → g(z) = exp[i jn z n ] (23.68)
n=−∞

with jn∗ = j−n . This group acts on the one-body wavefunction by

z n → g(z)z n (23.69)

where g(z) is in the loop group of U (1).

Under such a change of one-body wavefunctions the Slater determinant changes by:

∆0 (z) → detg(zi )zij−1

g(z1 ) g(z1 )z1 g(z1 )z12 · · ·
 
2
= det g(z1 ) g(z2 )z2 g(z2 )z2 · · · 
 
.. .. .. (23.70) eq:coherentone
. . . ···
Y
= g(zi )∆0 (z)
X
= exp[i jn sn (z)]∆0 (z)

Note that there are two Fermi levels, and if N is large they are “far apart” meaning
that operators such as i zin for small n will not mix states near the two respective lev-
P

els. Let us therefore imagine taking N large and focussing attention on one of the Fermi
levels. Therefore we adjust our energy level so that the groundstate wavefunctions are
10
P. Jordan, Z. Phys. 93(1935)464.

– 111 –
1, z −1 , z −2 , . . . and we imagine that we have taken N → ∞. Moreover, let us extend the
LU (1) action to LC∗ . Then we can separately consider the action of
0
X
g− (z) = exp[i ϕn z n ] (23.71)
−∞

and
∞
X
g+ (z) = exp[i ϕn z n ] (23.72)
1

on the groundstate Ψ0 . Note if we think of the groundstate wavefunction as a Slater

determinant then acting with g− (z) takes one column to a linear combination of lower
columns and hence does not change the wavefunction. On the other hand, by (23.70)
acting with g+ (z) has a nontrivial action and generates all possible symmetric functions of
the zi .
In this - rather heuristic - sense, the action of the loop group “generates the whole
Hilbert space of fermionic states.” Moreover, by mapping antisymmetric wavefunctions to
symmetric wavefunctions we now view the Hilbert space as the space of polynomials in the
sn . 11
Now we observe the following. The ring of symmetric functions (extending scalars to
C) is the polynomial ring C[s1 , s2 , . . . ]. We now make this space into a Hilbert space. We
introduce the inner product on polynomials in si by
∞
dsk ∧ ds̄k
Z Y P∞ 1 2
hf (s)|g(s)i := f (s)g(s)e− k=1 k |sk | (23.73) eq:innerpf
2πik
k=1

This is the coherent state representation of an infinite system of harmonic oscillators:

[ak , aj ] = kδk+j,0 − ∞ < j, k < ∞ (23.74)

with a−j = a†j . These are realized as follows: a†k is multiplication by sk and ak is k ∂s∂k .
The state
(a†1 )`1 (a†2 )`2 · · · (a†N )`N |0i (23.75)

corresponds to the symmetric function s(`) .

Now one can form the quantum field:
X1
φ := i ak z −k (23.76)
k
k6=0

Then the coherent states I

|φ(z)i = expi φ(z)j(z)|0i (23.77)

give an action of the loopgroup on the bosonic Fock vacuum acting as

11
This statement is rather loose. See the Pressley-Segal book for a mathematically precise treatment.
We are discussing an irreducible representation of a centrally extended group.

– 112 –
 ∞
X
exp[i jn a†n ]|0i (23.78) eq:coherenttwo
n=1
thus producing a nontrivial isomorphism between a bosonic and fermionic Fock space.
To make the isomorphism between bosonic and fermionic Fock spaces more explicit we
+
introduce a second-quantized formalism. B−n creates a fermionic state with wavefunction
n
z , Bn annihilates it, so we introduce:
X
Ψ(θ) = Bn einθ
n∈Z
X (23.79) eq:fermflds
† + −inθ
Ψ (θ) = B−n e .
n∈Z
The filled Fermi sea satisfies the constraints:

Bn |0i = 0, |n| > nF

+
(23.80) eq:fifer
B−n |0i = 0, |n| ≤ nF .
When we have decoupled systems it is appropriate to define two independent sets of
Fermi fields:

1 1
Ψ(θ) = ei(nF + 2 )θ b(θ) + e−i(nF + 2 )θ b̄(θ)
1 1
(23.81) eq:bcsys
Ψ† (θ) = e−i(nF + 2 )θ c(θ) + ei(nF + 2 )θ c̄(θ).
We introduce complex coordinates z = eiθ , and define the mode expansions:

X
b(z) = bn z n
n∈Z+ 12
X
c(z) = cn z n
n∈Z+ 12
X (23.82) eq:bcsysi
b̄(z) = b̄n z̄ n
n∈Z+ 12
X
c̄(z) = c̄n z̄ n
n∈Z+ 12

bn , cn , b̄n , c̄n are only unambiguously defined for | n |<< N . That is, we focus on
operators that do not mix excitations around the two Fermi levels. The peculiar half-
integral moding is chosen to agree with standard conventions in CFT. In terms of the
original nonrelativistic modes we have:

+
cn = B−nF −+n

bn = BnF ++n
(23.83) eq:mapmodes
c̄n = Bn+F +−n
b̄n = B−nF −−n

– 113 –
 where  = 12 , so that

{bn , cm } = δn+m,0 {b̄n , c̄m } = δn+m,0 (23.84) eq:anticomms

and all other anticommutators equal zero.

We may now reinterpret the fields b, c, .... Defining z = eiθ+τ we see that these may be
extended to fields in two-dimensions, and that they are (anti-) chiral, that is, they satisfy
the two-dimensional Euclidean Dirac equation. Upon continuation to Minkowski space we
have z → ei(θ+t) and z̄ → ei(θ−t) . We are thus discussing two relativistic massless Fermi
fields in 1 + 1 dimensions. The b, c fields generate a Fermionic Fock space built on the
product of vacua | 0ibc ⊗ | 0̄ib̄c̄ where bn | 0ibc = cn | 0ibc = 0 for n > 0. Explicitly, the b, c
fields by themselves generate the Fock space:
Y Y 
Hbc = Span bni cmi | 0ibc . (23.85) eq:cftstspc
P H
The space Hbc has a natural grading according to the eigenvalue of n∈Z : bn c−n = bc
(called “bc-number”):

(p)
Hbc = ⊕p∈Z Hbc (23.86) eq:grdspce

and the states obtained by moving a state from the Fermi sea to an excited level
correspond to the subspace of zero bc number:

(0)
Hchiral := Hbc (23.87) eq:chrlspc

NEED TO GIVE HERE THE TRANSLATION BETWEEN a state corresponding to

a Young Diagram and an explicit fermionic Fock state.
The bc system defines a conformal field theory. Indeed we may introduce
∞
X
Ln = (n/2 + m)c−m bm+n (23.88) eq:frmvrop
m=−∞

and compute that the Ln satisfy the Virasoro algebra

c
[Ln , Lm ] = (n − m)Ln+m + n(n2 − 1)δn+m,0 (23.89) eq:virasoroalg
12
with c = 1.
Now let us consider the second quantized operators corresponding to z n . Before taking
the limit these are given by:
X Z
Υn = zin ←→ dθΨ† (θ) einθ Ψ(θ). (23.90) eq:upsop

Now let us take the large N limit and focus on terms that do not mix the two Hilbert
spaces of excitations around the two Fermi levels. Cross terms between barred and unbarred
fields involve operators that mix the two Fermi levels. Since we are only interested in the
case of the decoupled Fermi level excitations we may replace:

– 114 –
 I I
−1−n
Υn → dz z c b(z) + dz̄ z̄ −1+n c̄ b̄(z̄)
X
= cn−m bm + c̄m−n b̄−m (23.91) eq:upsii
m
= αn + ᾱ−n

where we have introduced a field bc = i∂z φ(z) which has expansion

X
∂z φ(z) = i αm z m−1
m∈Z
(23.92) eq:bosalg
[αm , αn ] = [ᾱm , ᾱn ] = mδm+n,0
[αm , ᾱn ] = 0.

In terms of αn , we may introduce a representation of the Virasoro algebra:

1X
Ln = αn−m αm (23.93) eq:virop
2
which satisfy (23.89), again with c = 1. Using the α we can define a vacuum αn | 0i = 0
for n ≥ 0 and a statespace
( )
Y
Hα = Span | ~ki ≡ (α−j ) kj
| 0i . (23.94) eq:achalph

Bosonization states that there is a natural isomorphism:

(0)
Hα ∼
= Hbc (23.95) eq:bsnztin

We will not prove this but it can be made very plausible as follows. The Hilbert space
may be graded by L0 eigenvalue. The first few levels are:

L0 = 1 {b− 1 c− 1 | 0i} {α−1 | 0i}

2 2 (23.96) eq:firstlev
L0 = 2 {b− 1 c− 3 | 0i, b− 3 c− 1 | 0i} {α−2 | 0i, (α−1 )2 | 0i}
2 2 2 2

At level L0 = n, the fermion states are labeled by Young diagrams Y with n boxes.
At level L0 = n, the Bose basis elements are labeled by partitions of n. We will label a
partition of n by a vector ~k = (k1 , k2 , . . . ) which has almost all entries zero, such that
P
j jkj = n. Bosonization states that the two bases are linearly related:

X
|Yi= h~k | Y i | ~ki (23.97) eq:linrel
~k∈Partitions(n)

This last relation can be understood as the relation (23.61).

***************
1. Need to explain this some more.

– 115 –
 2. Look up papers of the Kyoto school, Jimbo et. al. for perhaps helpful ways of
presenting this material.
***************
For much more about this see the references below.
1. Pressley and Segal, Loop Groups, Chapter 10.
2. M. Stone, “Schur functions, chiral bosons, and the quantum-Hall-effect edge states,”
Phys. Rev. B42 1990)8399 (I have followed this treatment.)
3. S. Cordes, G. Moore, and S. Ramgoolam, “Lectures on 2D Yang-Mills Theory,
Equivariant Cohomology and Topological Field Theories,” Nucl. Phys. B (Proc. Suppl
41) (1995) 184, section 4. Also available at hep-th/9411210.
4. M. Douglas, “Conformal field theory techniques for large N group theory,” hep-
th/9303159.

– 116 –
 Next time – include
Verma modules – how the more standard treatment of SU(2) reps fits in.
basis
z j+m
fm (z) = p ↔ | j, m > (23.98)
(j + m)!(j − m)!
for −j ≤ m ≤ j
is orthonormal wrt:
Z
(2j + 1)! 1
(f, g) = d2 z f ∗ g(z) (23.99)
π (1+ | z |2 )2j+1 (z)

– 117 –