Conformal Field Theory

typeset by walker h. stern

The following notes were taken from a course given at Universität Bonn during the
summer semester 2016 by Dr. Hans Jockers
 2

Contents

1 Conformal Field Theories in d dimensions 4

1.1 Conformal group in dimension d (d > 2) 4
1.2 Correlation function of Quantum Fields (d > 2) 8
2 Conformal Field Theories in 2 Dimensions 12
2.1 Conformal Algebra in 2 Dimensions 12
2.2 Correlation functions of (quasi-)primary fields 16
2.3 Radial Quantization 18
2.4 Operator product expansions 23
2.5 Energy-momentum tensor and conformal Ward identities 25
2.6 Examples of CFTs (Free CFTs) 29
2.7 The central charge c 33
2.8 The Hilbert Space of States 40
2.9 Conformal Families and Operator Algebra 42
2.10 Conformal Blacks & Crossing Symmetries 47
3 Minimal Models 50
3.1 Warm-up: Representations of su(2) 50
3.2 Reducible Verma Modules & Singular Vectors 52
3.3 Kac Determinants 55
3.4 Fusion Rules 58
3.5 The Critical Ising Model 60
3.6 Minimal Model Characters 64
4 Modular Invariance 65
4.1 Partition Function 66
4.2 Modular Invariance 68
4.3 Free Boson Parition Function 70
4.4 Interlude: Mode Expansions of Free Fermions 71
4.5 Partition Function of the Free Fermion 75
5 Applications 79
5.1 Affine Kac-Moody algebras and WZW models 79
5.2 Coset Constructions 84
 3

Introduction
Aim: to understand CFT’s in 2d.

Why?

(A) Second order phase transitions in 2d systems (so-called critical

phenomena). For example, the 2d Ising model. Take a 2d lattice of
critical sites (in principal, this is assumed to be infinite):

1
σ spin 2

These spins are equipped with a nearest neighbor interaction. The

energy for the system is given by
X
E({σ}) = − σi σj
adjacent
lattice sites

where σi ∈ ± 12 . There are ground states for the system: all states
having the same spin (ie either all | ↑i or all | ↓i). The partition
function is X
Z= exp (−E({σ})β)
{σ}
1
where β = T is the inverse temperature.
We can then compute the correlation functions:
 
1 X −|i − j|
hσi , σj i − exp (−E({σ})) σi σj ∼ exp
Z ζ(T )
{σ}

where |i − j| is the distance from the site i to the site j.

If T → ∞, then ζ(T ) → 0, so that hσi , σj i → 0. If T → Tc ,
ζ(T ) → ∞, and the system reaches criticality.
At criticality, T = Tc , we have that
1
hσi , σj i ∼ 1
|i − j| 4
This theory is scale-invariant (for scales  a), and can be solved by
use of a 2-dimensional CFT1 1
Onsager won the ’68 Nobel prize in
Chemistry for his solution of the 2d
Ising model at criticality.
 4

(B) String theory provides other examples where CFT’s can be given
meaning. It concerns itself with the time evolution of 1d objects
(”strings“):

2d world sheet

time
t

closed string

This time evolution is described by the Polyakov action, which is

reparametrization invariant and Weyl invariant. Together these give
us conformal invariance, which leads to a 2d CFT description of 2d
worldsheets.

1 Conformal Field Theories in d dimensions

1.1 Conformal group in dimension d (d > 2)

Definition (Local Conformal Transformations). Let M, N be smooth
d-dimensional manifolds with metrics g and h respectively2 . A local 2
Very generally, we can take (M, g)
diffeomorphism of open sets and (N, h) to be semi-Riemannian.
That is, the ‘metrics’ can be taken
to give symmetric, nondegenerate,
φ : U ⊂ M → φ(U ) ⊂ N bilinear pairings on the tangent spaces
(not necessarily positive definite).
is called a local conformal transformation if

φ∗ h = Λ · g

for a smooth scale function

Λ : U → R>0

Remark. (local) conformal transformations preserve angles. Given

two vectors X, Y ∈ Tp M , for p ∈ U , we then have

h(φ∗ X, φ∗ Y )
cos ∠(φ∗ X, φ∗ Y ) =
|φ∗ X|h |φ∗ Y |h
φ∗ h(X, Y )
=
|X|φ∗ H |Y |φ∗ h
Λg(X, Y )
= √ √
Λ|X|g Λ|Y |g
= cos ∠(X, Y )
 5

Remark. • Conformal transformations define ’new‘ metrics on a

space for M = N
g̃ = φ∗ g = Λ · g
for φ : M → M . This implies, with a little computation, that the
Weyl Tensor remains invariant under conformal transformations,
but the Riemann tensor does not.
We will, for now, work with conformally flat space M = N = Rd ,
with the metric

η = diag(−1, . . . , −1, 1, . . . , 1)
| {z } | {z }
n m

where n + m = d. So we can write a conformal transformation

φ : Rd → Rd X µ 7→ φα (X µ ), by requiring the relation

∂φα ∂φβ
ηαβ = Ληµ,ν
∂X µ ∂X ν

• The 2d/3d Weyl tensor always vanishes on any 2d/3d manifold that
is locally conformally flat.

• In 2d, we can see from cartography that we can map a globe minus
a point to a sheet of paper by preserving angles.

We will now consider an infinitesmal transformationX µ 7→ X µ +

µ (X). Which gives us

ds2 → ds2 + (∂µ ν + ∂ν µ )dX µ dX ν

as the transformation of line elements3 3

The line element is, as usual, defined
This transformation is conformal if by
ds2 = ηµν dX µ dX ν

∂µ ν + ∂ν µ = ηµν C

By taking the trace of this relation, we can find that

2
∂µ ν + ∂ν µ = div()ηµν (∗)
d
Furthermore
(∗) 2
∂µ ∂ν (div) = ∂µ ∂ ρ (∂ν ρ + ∂ρ ν − ∂ρ ν ) = ∂µ ∂ν (div) − ∂µ ν
d
Since this expression is symmetric under the interchange of µ and ν,
we can simplify to
2 1
∂µ ∂ν (div) − (∂µ ν + ∂ν µ )
d 2
and, applying (∗) again, we get

(ηµν  + (d − 2)∂µ ∂ν )(div) = 0

 6

Taking the trace, we then get

(div) = 0

Ansatz: For the infinitesmal transformations:

∂µ ∂ν (div) = 0 ⇒ div = A + Bµ X µ

Which implies
µ = aµ + bµν X ν + Cµνρ X ν X ρ
where Cµνρ = Cµρν
From this, we can classify infinitesmal conformal transformations:

(i) µ = aµ , called translations

(ii) µ = bµν X ν . In (∗), we get:

2
bµν + bµν = ηµν (η τ σ bτ σ )
d

a) µ = wµν X ν , where wµν = −wνµ , which we call rotations.

b) µ = λ · ηµν X ν = λXν which we call dilations.

(iii) µ = Cµνρ X ν X ρ . In (∗), we get4 : 4

Where bµ = 1
C
d ρσµ
η ρσ

Cµνρ = ηµρ bν + ηµν bρ − ηνρ bµ

so that
µ = 2(b · X)Xµ − bµ |X|2
which we call special conformal transformations (SCTs).

By integrating these infinitesmal versions, we can obtain expressions

for finite conformal transformations:

(i) Translations:
X̃ µ = X µ + aµ

(ii) Rotations: Poincaré Subgroup

X̃ µ = Mνµ X ν
for Mνµ ∈ SO(n, m).

(iii) Dilations:
X̃ µ = ΛX µ

(iv) SCTs:
X µ − bµ |X|2
X̃ µ =
1 − 2bµ X µ + |b|2 |X|2
 7

As it happens, the conformal group acting on X µ is precisely

SO(n + 1, m + 1)
Interlude: Projective space RPd+1 = Rd+2 \ {0}/R∗ .
Consider the map
d+1
ι : Rd → RP
 
1 1
Xµ 7→ (1 + |X|2 ) : X 1 : · · · : X d : (1 − |X|2 ) =: XP
2 2
Properties:
(i) ι(R) is on the projective light cone:

{0 = ηP (XP , XP )} ⊂ RPd+1

where
ηP = diag(−1, . . . , −1, 1, . . . , 1)
| {z } | {z }
n+1 m+1

d+1
(ii) SO(n + 1, m + 1) acts on RP in the canonical way

XPA = MPA B XPB

where
MPA B ∈ SO(n + 1, m + 1)

(iii) Transformations in SO(n + 1, m + 1) induce the following confor-

mal transformations on Rd

a) Rotations:  
1 0 0
MBA = 0 Mνµ 0
 
0 0 1
with Mνµ ∈ SO(n, m) ⊂ SO(n + 1, m + 1).
Poincaré Subgroup
b) Translations (X µ → X µ + aµ )
 
1 + 12 |a| −aµ − 12 |a|
A
MB =  aµ

1 −aµ 

1
− 2 |a| aµ 1 − 12 |a|

c) Dilations (X µ → rX µ )
1+r 2 1−r 2
 
2r 0 2r
MBA =  0

1 0 

2
1−r 1+r 2
2r 0 2r

X µ −bµ |X|2
d) SCT’s (X µ → 1−2bµ X µ +|b|2 |X|2 )
 
1 + 12 |b| −bµ − 12 |b|
MBA =  −bµ

1 bµ 

1 + 12 |b| −bµ 1 − 12 |b|
 8

Conformal Invariants
For points xi ∈ Rd

• If we require translational and rotational invariance, we can take as

our invariant
|x1 − x2 |

• If we additionally require scale invariance, it suffices to add another

point:
|x1 − x2 |
|x1 − x3 |

• Applying an SCT, we get

− 12
− 21
|x1 −x2 | → 1 − 2bx1 + |b|2 |x1 |2 1 − 2bx2 + |b|2 |x2 |2 |x1 −x2 |

Therefore, conformal invariants are anharmonic ratios 5 : 5

Also called cross ratios

|x1 − x2 ||x3 − x4 |
|x1 − x3 ||x2 − x4 |

N
Remark. There are a total of 2 (N − 3) distinct anharmonic invari-
ants for a set of N points6 . 6
Warning: There can be compli-
cated algebraic relations among such
invariants.
1.2 Correlation function of Quantum Fields (d > 2)
We take the Euclidean signature in d dimensions, so that our confor-
mal group is SO(1, d + 1). We then get the commutation relations

[JAB , JCD ] = i (ηAB JBC + ηBC JAB − ηAC JBD − ηBD JAC )

And JAB = −JBA ,

η = diag(−1, 1, 1, . . . , 1)
| {z }
d+1

and A, B, C, D range through −1, 0, 1, 2, . . . , D. For the Conformal

Generators 7 : 7
µ, ν are assumed to be strictly
positive in the definitions below.
• Lµν = Jµν Rotations
Poincaré Algebra
• Pµ = J−1,µ Translations

• D = J−1,0 Dilations

• Kµ = −J−1,µ + J0,µ SCTs

 9

The Conformal algebra is then the Poincaré algebra plus the com-
mutation relations

[D, Pµ ] = iPµ
[D, Kµ ] = −iK + µ
[D, Lµ,ν ] = 0
[Kµ , Pν ] = 2i(ηµν D − Lµν )

From these algebraic definitions, we can then get conformal trans-

formations, eg:

• exp (−ibµ Kµ ) ∈ SO(1, d + 1) SCT

• exp (−iaµ Pµ ) ∈ SO(1, d + 1) Translations

• exp (−iλD) ∈ SO(1, d + 1) Dilation

• exp − 2i ω µ,ν Lµν ∈ SO(1, d + 1) Rotation

Quantum Fields as ‘∞-dimensional representations’ of

conformal group

Example. We can think of functions {g} as representations of

S O(1, d + 1) via the map

ρ g : S O(1, d + 1) × Maps(R d → R) → Maps(R d → R)

(Λ, g) 7 → g ◦ f Λ −1

where

f Λ : x µ 7 → f ◦ Λ(x µ )
f Λ −1 : x µ 7 → f ◦ Λ −1 (x µ )

We then have the infinitesmal generators of the conformal group in

the representation ρ g

f Λ : x µ 7 → x µ +  µ (x)
f Λ −1 : x µ 7 → x µ −  µ (x)

Example. As an example, we can look at the infinitesmal dilation

lambda Λ λ
f Λ λ : x µ 7 → x µ + λx µ
So we then have
∂ g ! −iλD
ρ g (Λ λ )(g)(x µ ) = g(x µ −  µ (x)) = g(x µ ) − λx µ =e g(x µ )
∂ xµ
which implies D µ = −ix µ d µ
 10

In a similar fashion, we can find that

P µ = −i∂ µ
L µν = i(x µ ∂ µ − x ν ∂ µ )
K µ = −i(2x µ x ν ∂ ν − |x| 2 ∂ µ )

A Quantum Field ψ is then given by φ A such that, for a finite

dimensional vector space V ,

ρ ψ : S O(1, d + 1) × Maps(R d → V ) → Maps(R d → V )

(Λ, ψ A ) 7 → ρ V (Λ) AB ψ B ◦ f Λ −1

Quasi-Primary Fields We consider a field ψ(x) at x = 0 8 . We 8

x = 0 can be thought of as the
then have invariant locus with respect to the
action of the lorentz group, dilations,
and SCTs. See, eg, [1].
Lµν ψ(0) = Sµ,ν ψ(0)
Dψ(0) = −i∆ψ(0)
Kµ ψ(0) = −Kµ ψ(0)

We make the further assumption that we have an irreducible repre-

sentation of the Lorentz group, to get

[∆, Sµν ] = 0

Which, by Schur’s lemma, implies ∆ ∼ 1, ie a number, which we will

call the scaling dimension. We also get

[∆, Kµ ] = −Kµ

so that Kµ = 0
We can also compute the ‘scaling dimension at other locations’
µ
Dψ(xµ ) = De−ix Pµ ψ(0)
h µ
i µ

= D, e−ix Pµ + e−ix Pµ D ψ(0)
[D,Pµ ]=iPµ
 µ µ

= e−ix Pµ (xµ Pµ ) − i∆e−ix Pµ ψ(0)

= (X µ Pµ − i∆)ψ(xµ )

If we assume our field is spinless, ie Sµν = 0, then the field is

characterized entirely by the scaling dimension D. We can then derive
the finite transformation behavior of a spinless field φ(x).
∆/d
Λ ∂ x̃µ
φ(xµ ) −→ φ̃(x̃µ ) = φ(xµ )
∂xν
xµ 7→ x̃µ = fΛ (xµ )

we call such fields Quasi-Primary Scalar Fields.

 11

Correlation Functions of Quasi-Primary Fields

From a CFT, we get the path integral measure

[dΦ] eS[Φ]

Where Φ is the fields in our theory, and S is the action9 functional. 9

Which must be conformally invari-
The correlation function is then ant.
Z
1
hφ1 (x1 ) · · · φN (xN )i = [dΦ]e−S[Φ] Φ1 (x1 ) · · · ΦN (xN )
Z
where Z
Z= [dΦ]e−S[Φ]

Applying a conformal transformation,

Z
1
hφ1 (x̃1 ) · · · φN (x̃N )i = [dΦ̃]e−S[Φ̃] Φ̃1 (x̃1 ) · · · Φ̃N (x̃N )
Z
which implies
−∆1 /d −∆N /d
∂ x̃µ1 ∂ x̃µN
hφ1 (x̃1 ) · · · φN (x̃N )i = ··· hφ1 (x1 ) · · · φN (xN )i
∂ x̃ν1 ∂ x̃νN

2-point Correlation Functions

Take two fields φ1 and φ2 . From the requisite symmetries, we can
deduce some of the form of the correlation function

• Rotation+Translation invariance implies

hφ1 (x1 )φ2 (x2 )i = f (|x1 − x2 |)

• Dilation x̃µ 7→ λxµ invariance implies

hφ1 (x̃1 )φ2 (x̃2 )i

= λ−∆1 −∆2
hφ1 (x1 )φ2 (x2 )i

Together, these computations show us that

C1
hφ1 (x1 )φ2 (x2 )i =
|x1 − x2 |∆1 +∆2

Now applying invariance under SCTs we notice that

∂ x̃k 1
= =: Sk (bµ )
∂xµ (1 − 2bµ xµk + |b|2 |xk |2 )d

and then compute

1
hφ1 (x̃1 )φ2 (x̃2 )i = hφ1 (x1 )φ2 (x2 )i
S1 (bµ )∆1 S2 (bµ )∆2
C12
= ∆1 ∆2
S1 S2 |x1 − x2 |∆1 +∆2
 12

but, on the other hand

C12
hφ1 (x̃1 )φ2 (x̃2 )i =
|x̃1 − x̃2 |∆1 +∆2
C12
= (∆1 +∆2 )/2 (∆1 +∆2 )/2
S1 S2 |x1 − x2 |∆1 −∆2

Hence C12 6= 0 only if

1
(∆1 + ∆2 ) = ∆1 = ∆2
2
or, more precisely

C12

|x1 −x2 |2∆ ∆1 = ∆2 = ∆
hφ1 (x1 )φ2 (x2 )i =
0 else

One can perform similar simplifications for 3 and 4 point correla-

tion functions. It is left as an exercise to see

hφ1 (x1 )φ2 (x2 )φ3 (x3 )i =

C123
|x1 − x2 |∆1 +∆2 −∆3 |x ∆ +∆3 −∆2 |x − x |∆2 +∆3 −∆1
1 − x3 | 1 2 3

and
hφ1 (x1 )φ2 (x2 )φ3 (x3 )φ4 (x4 )i =
Y
f (anh. ratios) |xk − x` |∆/3−∆k −∆`
1≤k<`≤4
P
where ∆ = i ∆i .

2 Conformal Field Theories in 2 Dimensions

2.1 Conformal Algebra in 2 Dimensions

Local conformal transformations in R2 with metric δαβ = ηαβ are
given by differentiable maps

ϕ : U ⊂ R2 → V ⊂ R2

where
ϕ∗ η = Λη, Λ : U R>0
This implies
∂ϕα ∂ϕβ
ηαβ= Ληµν
∂xµ ∂xν
where we assume the transformation is orientation preserving, ie

∂ϕ(x1 , x2 )
>0
∂(x1 , x2 )
 13

We then obtain from η11 and η22

Λ = (∂1 ϕ1 )2 + (∂1 ϕ2 )2 = (∂2 ϕ1 )2 + (∂2 ϕ2 )2

and from η12 and η21

0 = (∂1 ϕ1 )(∂2 ϕ1 ) + (∂1 ϕ2 )(∂2 ϕ2 )

We can combine these two equations into a single complex equation

to get

2
2
(∂1 ϕ1 ) − i(∂2 ϕ1 ) = (∂2 ϕ2 ) + i(∂1 ϕ2 )
or, equivalently

∂1 ϕ1 − i∂2 ϕ1 = ±(∂2 ϕ1 + i∂1 ϕ2 )

We then get solutions

Orientation Preserving Orientation Reversing
(+) (-)
∂1 ϕ = ∂2 ϕ2
1
∂1 ϕ = −∂2 ϕ2
1

∂2 ϕ1 = −∂1 ϕ2 ∂1 ϕ2 = ∂2 ϕ1
∂ϕ(x1 ,x2 ) ∂ϕ(x1 ,x2 )
∂(x1 ,x2 ) >0 ∂(x1 ,x2 ) <0

Metric in homolorphic coordinates

The (+) solution gives precisely the Cauchy-Riemann Differential
Equation, so that local conformal transformations in complex coordi-
nates z = x1 + ix2 are local biholomorphic functions ϕ(z).
In terms of the holomorphic and antiholomorphic coordinates z, z,
we can rewrite the metric

ds2 = (dx1 )2 + (dx2 )2 = dzdz

where dz = dx + idy and dz = dx − idy. Under biholomorphic

mappings, we get
∂ϕ ∂ϕ
dz 7→ dz, dz 7→ dz
∂z ∂z
We then have that the metric transforms as
2
φ ∂ϕ
ds2 = dzdz 7→ dzdz
∂z
so that
2
∂ϕ
∆(z, z) =
∂z
We thereby see that local conformal transformations are local holo-
morphic coordinate changes.

Remark. • Often we regard z and z as independent coordinates, and

enhance (x1 , x2 ) ∈ R2 to complex coordinates (x1 , x2 ) ∈ C2 .

• ‘Physical Condition’. We impose z = z ∗ , the complex conjugate.

 14

Conformal Algebra
If we take an infinitesmal conformal transformation z 7→ z̃ =
z + (z), and take a Laurent expansion
∞
X
(z) = cn z n+1
n=−∞

where the cn are assumed to be infinitesmal, we can compute the

action on functions ϕ(z, z) ∈ Map(C → R).

ecn `n +cn `n ϕ(z, z) = ϕ(z − (z), z − (z))

which is equivalent to

(1 + cn `n + cn `n )ϕ(z, z) = (1 − (z)∂z − (z)∂z )ϕ(z, z)

So the generators of the conformal algebra (in ‘function representa-

tion’) are
`n = −z n+1 ∂z , `n = −z n+1 ∂z
The commutation relations these generators satisfy (Witt Algebra
relations)

[`n , `m ] = (n − m)`n+m
[`n , `m ] = (n − m)`n+m

Remark. • Treating z and z as independent variables yields two

copies of the Witt Algebra A ⊕ A.

• Imposing the ‘physical condition’ leaves us with the subalgebra of

A ⊕ A generated by `n + `n and i(`n − `n for all n

• For a Quantum Theory, we need a central extension of the Witt

Algebra, which is the so-called Virasoro Algebra.

Global Conformal Transformations

On S 2 = C∪{∞}, global holomorphic transformations are generated
by global vector fields.
We take as an Ansatz
∞
X
v(z) = − an `n
n=−1
∞
X
= an z n+1 ∂z
n=−1
∞
z→w= z1 X
7→2 − an w−n+1 ∂w
∂z →w ∂w
n=−1

With a global vector field

v(z) = −(a−1 `−1 + a0 `0 + a1 `1 )

We then have generators (applying the physical condition)

 15

• `−1 + `−1 and i(`−1 − `−1 ) translations

• `0 + `0 dilation

• i`0 − i`0 rotation

• `1 − `1 and i(`1 − i`1 ) SCT

Finite Global Conformal Transformations

For S 2 = C ∪ {∞}, a, b, c, d ∈ C, ad − bc = 1 we have
" #
a b
∈ SL(2, C)/Z2 = SO+ (1, 3)
c d

which is the conformal group in d = 2 dimensions. From this, we have

the group of Möbius transformations of the sphere
az + b
z 7→
cz + d
Here, our generators are

• Translations " #
1 ã
1

• Dilations " #
λ1/2
λ−1/2

• Rotations " #
θ
ei 2
θ
e−i 2

• SCT " #
1 0
b̃ 1

Remark. • For four points z1 , . . . , z4 , we can define invariant 4-point

cross ratios
(z1 − z2 )(z3 − z4 )
η=
(z1 − z3 )(z2 − z4 )
There are relations among these cross ratios. In fact, for four
points, there is only one independent cross ratio. This can be seen
by noting that there is always a transformation in SL(2, C)/Z2
which maps    
z1 ∞
z  1
 2
  7→  
 
z3  η
z4 0
 16

Physical States are characterized by the eigenvalues of the dilation

operator `0 + `0 and the rotation operator i`0 − i`0 . These can be
written in terms of the quantities h, h10 10
Sometimes known as the conformal
weights of the state |ψi.
`0 |ψi = h|ψi
`0 |ψi = h|ψi

The scaling dimension is the ∆ = h + h (the dilation operator eigen-

value) and the spin11 is s = h − h (the ‘rotation operator eigenvalue’). 11
As one might expect, for bosonic
states, we have s ∈ Z. For Fermionic
states, s ∈ Z + 12 . We could also take,
2.2 Correlation functions of (quasi-)primary fields more generally, a parafermionic state
s ∈ Q.
Definition. A field φ(z, z) of conformal weight (h, h) is a primary
field if it transforms under local conformal transformations

z 7→ φ(z) = z̃

locally as
 −h  −h
∂φ ∂φ
φ(z, z) 7→ φ̃(z̃, z̃) = φ(z, z) (∗)
∂z ∂z

Remark. • A Field is called a quasi-primary field if (∗) holds for

global conformal transformations12 12
Sometimes also called SL(2, C)-
primaries
• A field wtih a different transformation behavior is called a sec-
ondary field.

If we take an infinitesmal variation of a primary field

z̃ = z + (z)
z̃ = z + (z)

We can compute

δ, φ(z, z) = φ̃(z, z) − φ(z, z)

= φ̃(z̃ − (z), z̃ − (z)) − φ(z, z)

Taylor expanding to first order we get

= φ̃(z̃, z̃) − (z)∂z φ̃ − (z)∂z φ̃ − φ(z, z)

h i
= (1 + ∂z ((z))−h (1 + ∂z (z))−h − (z)∂z − (z)∂z − 1 φ(z, z)

which then becomes

 
δ, φ(z, z) = −(h(∂z (z)) + (z)∂z ) − (h(∂z − (z)) + (z)∂z ) × φ(z, z)
(∗∗)
 17

2-point correlation function of quasi-primary fields

We consider a quasi-primary field as above13 . We can then compute 13
The equation (∗∗) holds for quasi-
primary fields where  is infinitesmal
0 = δ, hφ1 (z1 , z 1 )φ2 (z2 , z 2 )i and a generator of a global conformal
transformation.
= hδ, φ1 (z1 , z 1 )φ2 (z2 , z 2 )i + hφ1 (z1 , z 1 )δ, φ2 (z2 , z 2 )i

We can write our infinitesmal conformal transformations as

 = c−1 + c0 z + c1 z 2
 = c−1 + c0 z + c1 z 2

for these  and , (∗∗) holds for any quasi-primary field. This implies
that

0 = [h1 (∂z1 (z1 )) + (z1 )∂z1 + h2 (∂z2 (z2 )) + (z2 )∂z2 + c] × hφ1 φ2 i

We can then use the coefficients of

(z) = c−1 + c0 z + c1 z 2

c−1 : We have (∂z1 + ∂z2 )hφ1 φ2 i = 0 So

hφ1 φ2 i = c(z1 − z2 )

c0 : We have
(h1 + h2 + z1 ∂z1 + z2 ∂z2 )hφ1 φ2 i = 0
so, substituting our first result, we get

(h1 + h2 )c(z1 − z2 ) + (z1 − z2 )c0 (z1 − z2 ) = 0

and solving this differential equation:

C12
c(z1 − z2 ) =
(z1 − z2 )h1 +h2

c1 : We have

(2h1 z1 + 2h2 z2 + z12 ∂z1 + z22 ∂z2 )hφ1 φ2 i = 0

substituting in
C12 C12
(2h1 z1 +2h2 z2 ) −(h1 +h2 )(z12 −z22 ) =0
(z1 − z2 )h1 +h2 (z1 − z2 )h1 +h2 +1
so
C12 (h1 − h2 )
=0
(z1 − z2 )h1 +h2 +1
which means that either C12 = 0 or (h1 = h2 ).
 18

If we add in complex conjugation, we see that


C12

h +h h +h
h1 = h2 , h1 = h2
hφ1 (z1 , z 1 )φ2 (z2 z 2 )i = (z1 −z2 ) 1 2 (z1 −z2 ) 1 2
0 else

Remark. • We have a single-values correlator for s = h − h ∈ Z or in

Z + 12 . This is owing to the face that

(z1 − z2 )−2h (z 1 − z 2 )−2h = |z1 − z2 |−2h (z 1 − z 2 )2s

If we let s ∈ Q, we get a multiple-valued correlation function for

parafermions.

• If we let spin equal zero, ie h = h, we recover our previous d-

dimensional result.

More generally, we can apply the same techniques to compute 3-

and 4- point correlation functions, finding, when we do, that
C123
hφ1 (z1 , z 1 )φ2 (z2 , z 2 )φ3 (z3 , z 3 )i =
(z1 − z2 )h1 +h2 −h3 (z h +h3 −h2 (z − z )h2 +h3 −h1
1 − z3 ) 1 1 2
1
×
(z 1 − z 2 )h1 +h2 −h3 (z 1 − z 3 )h1 +h3 −h2 (z 2 − z 3 )h2 +h3 −h1

and

hφ1 (z1 , z 1 )φ2 (z2 , z 2 )φ3 (z3 , z 3 )φ4 (z4 , z 4 )i

Y
= f (η, η) (zk − z` )h/3−hk −h` (z k − z ` )h/3−hk −h`
1≤≤`≤4

where
(z1 − z2 )(z3 − z4 )
η=
(z1 − z3 )(z2 − z4 )
and
4
X
h= hi
i=1
4
X
h= hi
i=1

2.3 Radial Quantization

First Example
We now return to the context of string theory, with 2-dimensional
Minkowski space. Suppose we have a conformally flat world-sheet
 19

2d world sheet

time
x
t

closed string

where we have coordinates t ∈ R (‘time’) and x ∈ S 1 = R/(L · Z) where

L is the length of the string.
In 2d Minkowski space we can take the light cone coordinates t ± x,
and apply a Wick Rotation
τ := it
So that we can define complex coordinates

ζ = τ + ix

ζ = τ − ix
with the identification ζ ∼ ζ + iL
We then have a conformal map to the punctured conformal plane
C∗
−∞ ∞

τ1 τ2

ζ 7→ exp( 2πζ
L )

τ1 τ2

τ = −∞ τ =∞
z=0

In Quantization we have

• A Hilbert state space at each fixed spatial slice (ie, fixed time τ )

• Dynamics given by a propagator amounst such slices in time.

 20

On the conformal plane, we have

• Hilbert state space defined on circles about the origin

• Propagation of states in the radial direction

i) Dilation operator is the Hamiltonian for the string

ii) Rotation operator is a spatial translation along the string.

For such strings, a radial quantization scheme is very natural.

Second Example
We again consider the Euclidean space of statistical mechanics. The
standard quantization is comprised of:
• Hilbert Spaces of states along 1d slices (for example in a 2d lattice).

• Transfer matrices describing a correlation orthogonal to the quan-

tized slices.
In the limit where the lattice spacing goes to zero, and at criticality,
we have a conformal symmetry. So taking radial slices of the plane
and a radial propagator gives the same result14 . Radial quantization is 14
We could also use something
stranger, like, for example
a convenient choice because of the use of the radial operator product
expansion.

In and Out States

We make the assumptions:
i) There is a vacuum state |0i from which we can construct the
Hilbert space of states in terms of creation operators (positive
but in practice, the radial and eu-
frequency modes)
clidean schemes prove easiest to work
with.
ii) As τ → ±∞, that is, as (z, z) → ∞, 0, the Hilbert space of states
looks like the Hilbert space of states for a theory of free fields.
We can define an in-state

|φin i = lim φ(z, z)|0i

z,z→0

This equality is known as the state-field correspondence

 21

Hermitian Product
In string theory, Hermition conjugation has no effect on the 2d
Minkowski space

• Hermitian conjugation on Euclidean time:

τ = it : τ 7→ −τ

• Hermitian conjugation on radial coordinates

1
z 7→
z∗

On quasi primary fields, the Hermitian conjugation is given by:

1 1
φ(z, z)† = z −2h z −2h φ( , )
z z

We can then define an out-state by

hφout | := |φin i

that is
hφout | = lim h0|φ(z, z)†
z,z→0

This defines a good Hermitian product for quasi-primary fields.

As a first example, we can compute:

hφout |φin i = lim h0|φ(w, w)† φ(z, z)|0i

z,z,w,w→0
1 1
= lim w−2h w−2h h0|(φ( , )φ(z, z)|0i
 w
w 
? −2h −2h 1 1
= lim w w h0|R φ( , )φ(z, z) |0i
w w
1 1
= lim w−2h w−2h h(φ( , )φ(z, z)i
w w
C
= lim
2h 1
2h
2h 2h 1
w w w −z w −z

So that
hφout |φin i = C

More generally, if we consider multiple fields, we find


C h1 = h2 , h1 = h2
12
hφ1,out |φ2,in i =
0 else

In our computation above, one step went unexplained: The radial

ordering (?). We now delve into its meaning
 22

Recall that in QFT, the N -point correlator of quantum fields is

given by a time-ordering prescription.

hφ1 (x1 , t1 ) · · · φN (xN , tN )i = h0|T (φ1 (x1 , t1 ) · · · φN (xN , tN )) |0i

where

T (φ1 (x1 , t1 ) · · · φN (xN , tN )) = ±φσ(1) (xσ(1) , tσ(1) ) · · · φσ(N ) (xσ(N ) , tσ(N ) )

The sign is due to Bose/Fermi statistics, and the permutation σ is

such that
tσ(1) > tσ(2) > · · · > tσ(N )
We can perform a similar ordering for radial quantization: the
radial ordering.

φ (z, z)φ (w, w)
1 2 |z| > |w|
R (φ1 (z, z)φ2 (w, w)) =
±φ2 (w, w)φ1 (z, z) |z| < |w|

The positive sign comes in the case where at least one of the fields in
question is bosonic, and the minus sign in the case where both fields
are fermionic.
With the radial ordering, we have the operator-state correspondence

hφ1 (z1 , z 1 ) · · · φN (zN , z N )i = h0|R (φ1 (z1 , z 1 ) · · · φN (zN , z N )) |0i

Similarly to a Fourier expansion, we have the Mode expansion for

quasi-primary fields of dimension (h, h).
X
φ(z, z) = z −m−h z −n−h φm,n
m,n∈Z

where the φm,n are operators.

Using the calculus of residues, we can compute the φm,n
I I
dz m+h−1 dz n+h−1
φm,n = z z φ(z, z)
2πi 2πi
which then gives us the operator product expansion for the Hermitian
conjugate
1 1 X
φ(z, z)† = z −2h z −2h φ( , ) = z −m−h z −n−h φ−m,−n
z z n,m

Looking at X
φ(z, z)† = z −m−h z −n−h φ†m,n
m,n
we see
φ†m,n = φ−m,−n
 23

A well-defined in-state implies that

lim φ(z, z)|0i

z,z→0

must be finite. This, in turn, requires that we have

φm,n |0i = 0 for m > −h, n > −h

Remark. The mode expansion gives rise to string Fourier modes on

the cylinder
   
2πi 2πi
z = exp (t + x) z = exp (t − x)
L L
giving the expansion
   
X 2πi 2πi
φ(z, z) = φm,n exp − (∆ + m + n)t exp − (s + m + n)x
L L
m,n∈Z

where s = h − h is the spin, ∆ + m + n is the energy eigenstate of

the mode, and s + m − n is the wave propagation along the spatial
direction of the string.

2.4 Operator product expansions

Correlators of 2 or more fields typically exhibit singularities as their
insertion points coincide. For example, for quasi-primaries

hφk (z1 , z2 )φ` (z2 , z 2 )i = h0|R(φk (z1 , z 1 )φ` (z2 , z 2 )|0i

δ`k
=
(z1 − z2 )2hk (z 1 − z 2 )2hk
the last expression is called the canonical norm for quasi-primaries.
The operator product expansion (OPE): describes the behavior
of radially ordered quantum fields as the coincide. For example, for
A(z, z) and B(z, z)
N M
X X{A, B}n,m (w, w)
R(A(z, z)B(w, w)) =
n=−∞ m=−∞
(z − w)n (z − w)m

with N, M ≥ 0. In this case, we get a finite number of singular terms.

These singular terms play a special role:
N X M
X {A, B}n,m (w, w)
R(A(z, z)B(w, w)) ∼ = A(z, z)B(w, w)
n=1 m=1
(z − w)n (z − w)m

The degree zero term, known as the normal order product, is writ-
ten
dz R(A(z, z)B(w, w))
I I
dz
: A(w, w)B(w, w) : = {A, B}0,0 (w, w) =
w 2πi w 2πi (w − z)(w − z)
 24

Thus
singular terms
normal order product
R(A(z, z)B(w, W )) = A(z, z)B(w, w) + : A(w, w)B(w, w) : +O(z−w, z−w)

Remark. • Fields φ(z) that depend only on z are called chiral fields.
Fields φ(z) that depend only on z are called anti-chiral fields

• OPEs of chiral fields will play an important role.

• Notation: In OPEs the radial order symbol R is often dropped

R(A(z, z)B(w, w)) = A(z, z)B(w, w)

OPEs and commutators

Let A(z) and B(z) be bosonic chiral fields. We can deform the
contour of a small circle about w as seen below
Im(z)

Re(z)

Im(z)


w
Re(z)

So we can rearrange the contour integral of the radially ordered prod-

uct:
I I I
dz dz dz
R(A(z), B(w)) = A(z)B(w) − B(w)A(z)
w 2πi |w|+ 2πi |w|− 2πi
= QA B(w) − B(w)QA
= [QA , B(w)]
 25

H dz
where QA = 2πi A(z).
The result is that contour integrals encode (anti-)commutators at
equal time15 I
15
Whether we get a commutator or
dz an anti-commutator is dependent on
R(A(z)B(w) = [QA , B(w)]± the Bose/Fermi statistics. If at least
χ 2πi
one of the operators is bosonic, we
More generally, we can take have
I I [−, −]+ = [−, −]
dw dz
R(A(z), B(w)) = [QA , QB ]± If both are Fermionic, we have
2πi 2πi
[−, −]− = {−, −}
with QA and QB as above.

2.5 Energy-momentum tensor and conformal Ward identities

Interlude: Conserved charges and infinitesmal transformations of

fields.
If we have a (d + 1) dimensional QFT with a conserved Noether
current j µ , (ie, ∂µ j µ = 0) then we get a conserved charge (the Noether
Charge) Z
Q= dxd [j0 (x)]

For infinitesmal symmetry transformations acting on a quantum

field φ(x),
δ φ(x) = φ̃(x) − φ(x) = −[Q, φ(x)]
where the last term is the equal time commutator.
In the context of a 2d CFT with radial quantization, the conserved
charge Q of the radial component of a conserved chiral current yields
I
dz
δ φ(w, w) = −[Q, φ(w, w) = − R(A, φ)
w 2πi

Energy-Momentum tensor
The energy-momentum tensor Tµ,ν generates local coordinate trans-
formations16 . For Lorentz-invariant theories it is conserved 16
In Lagrangian theories, it is the
response of the Lagrangian to the
∂ µ Tµν = 0 variation of space-time.

so we get a conserved charge

Z
Pµ = dxd Tµ,nu

which are the momentum operators, and are translationally symmetric.

By rotational invariance, Tµν can also be chosen to be symmetric

Tµν = Tνµ

In conformal theories, we call the conserved current

jµD = xν Tνµ
 26

the dilation current because the associated charge D is the dilation

operator.
Scale invariance (generated by the dilation operator) leads to

0 = ∂ µ jµD = Tµµ

so that the energy-momentum tensor is traceless in CFTs.

Back in the 2d setting, we can take holomorphic coordinates

z = x1 + ix2 z = x1 − ix2
1 1
∂z = (∂1 − i∂2 ) ∂z = (∂1 + i∂2 )
2 2
We can then compute the energy-momentum tensor

∂x1 ∂x1 ∂x1 ∂x2 ∂x2 ∂x2

Tz,z = T1,1 + T2,2 + T2,2
∂z ∂z ∂z ∂z ∂z ∂z
1
= (T1,1 − 2iT1,2 − T2,2 )
4
1
Tz,z = (T1,1 + 2iT1,2 − T2,2 )
4
∂x1 ∂x1
 1 2
∂x2 ∂x1 ∂x2 ∂x2

∂x ∂x
Tz,z = Tz,z = T1,1 + + T1,2 + T2,2
∂z ∂z ∂z ∂z ∂z ∂z ∂z ∂z
1
= (T1,1 + T2,2 )
4
From the metric

gz,z = gz,z = η(∂z , ∂z ) = η(∂z , ∂z ) = 0

1
gz,z = gz,z = η(∂z , ∂z ) =
2
we can compute conservation laws:

i) Translation invariance gives

∂z Tz,z + ∂z Tz,z = 0
∂z Tz,z + ∂z Tz,z = 0

ii) Because of scale invariance

Tz,z = Tz,z = 0

so that

∂z Tz,z = 0
∂z Tz,z = 0

Implying that Tz,z and Tz,z are chiral and anti-chiral currents17 , 17
Notation: the chiral energy-
respectively. momentum current is
T (z) := −2πTz,z (z)
and the anti-chiral is
T (z) := −2πTz,z (z)
 27

Conformal Transformations of primary fields

Let φ(w, w) be a chiral field of conformal weight (h, h). Taking a
local infinitesmal coordinate transformation

z 7→ z̃ = z + (z)

we get

δ φ(w, w) = −[Q , φ(w, w)]

I
dz
=− (z)R(T (z), φ(w, w))
w 2πi
= − (h(∂w (w))φ(w, w) + (w)∂w φ(w, w)) (∗)

We notice that
I
dz (z)φ(w, w)
= ∂w ((w))φ(w, w)
2πi (z − w)2
Iw
dz (z)φ(w, w)
= (w)∂w φ(w, w)
w 2πi (z − w)

so that, with (∗) we find that a primary field φ(w, w) is characterized

by its OPE with T (z) and T (z).

h 1
R(T (z)φ(w, w)) ∼ φ(w, w) + ∂w φ(w, w)
(z − w)2 z−w
h 1
R(T (z)φ(w, w)) ∼ φ(w, w) + ∂w φ(w, w)
(z − w)2 z−w

Remark. Theres OPEs of φ(z, z) with T (z) and T (z) are often used
as the definition of a primary field.

Conformal Ward identities

Let φk (wk , wk ) be any primary fields located at |wk | < R, k =
1, . . . , N . For z 7→ z + (z), we can use the contour

Im(z)

R `R
w1
w2 Re(z)

w3
 28

to compute
I 
dz
(z)T (z)φ1 (w1 , w1 ) · · · φN (wN , wN )
`R 2πi
N  I  
X dz
= φ1 (w1 , w1 ) · · · (z)T (z)φk (wk , wk ) · · · φN (wN , wN )
k=1 wk 2πi

which implies

hT (z)φ1 (w1 , w1 ) · · · φN (wN , wN )i

N  
X hk 1 ∂
∼ + hφ1 (w1 , w1 ) · · · φN (wN , wN )i
(z − wk )2 z − wk ∂wk
k=1

which is known as the conformal Ward identity for primary fields.

The conformal Ward identities imply
I
dz
δ hφ1 · · · φN i = − (z)hT (z)φ1 (w1 , w1 ) · · · φN (wN , wN )i
` 2πi
N I  
X dz hk 1 ∂
(z) + hφ1 · · · φN i
` 2πi (z − wk )2 z − wk ∂wk
k=1

in particular
δ hφ1 · · · φN i = 0

for an infinitesmal global conformal transformation

(z) = c−1 + c0 z + c1 z 2

allows us to recover differential equations for correlators as discussed

before.
There are some technical assumptions that were needed to derive
the Ward identities:

• T (z) must be regular in the conformal plane18 . 18

It is worth recalling that T (z)
should not be thought of as a func-
• Poles must only arise when the fields coincide. tion, but rather as a holomorphic
section of a bundle specified by the
• T (z) is an energy density, that is scaling dimension. As a result, we
can require more strongly that T (z)
be regular on P1 without necessarily
Tµν ∼ (length)−2 imposing the condition that T (z) be
constant.
From which we can calculate the scaling dimension
∆
=2
2
which implies T (z) has conformal weight h = 2, and T (z) has
conformal weight h = 2.
 29

If we take the coordinate transformation

1
z 7→ w = −
z
we see that
 −2
dw
T (z) 7→ T̃ (w) = T (z) = z 4 T (z)
dz

which tells us that T (z) must decay as z −4 if regularity is to hold at

z = ∞.

2.6 Examples of CFTs (Free CFTs)

Recall from QFT:

(A) Noether’s Theorem (See, eg, di Francesco et al): For a QFT

with an action Z
S[φ ] = dd xL[φA ]
A

and an infinitesmal symmetry transformations

δxµ
x̃µ = xµ +
δwa

δF A [φA ]
φ̃A (x̃µ ) + wa
δwa
there is a conserved current

∂ µ jµa = 0

given by

δxν ∂L δF A
 
∂L
jµa = ∂ν φA − ηµν L +
∂(∂ µ φA ) δwa ∂(∂ µ φA ) δwa

(B) Wick’s Theorem: A time-ordered product19 of fields equals 19

For CFT, we replace the time-
the sum of normal ordered products with all possible contrac- ordered product with the radially
ordered product, ie, τ (· · · ) with
tions, eg R(· · · ).

τ (φ1 , φ2 , φ3 , φ4 ) = : φ1 φ2 φ3 φ4 : + : φ1 φ2 φ3 φ4 : + · · · + : φ1 φ2 φ3 φ4 :

+ : φ1 φ2 φ3 φ4 : + · · · + : φ1 φ2 φ3 φ4 :

where the contractions are given by20 20

Here I use the notational conven-
tion that φ̂i means omit the entry
: φ1 · · · φk1 · · · φk2 · · · φn : = (±) : φ1 · · · φ̂k1 · · · φ̂k2 · · · φN : × hφk1 φk2 i φi
 30

Example I: The Free Boson

Take the free, real field φ : R2 → R with
Z
g
S[φ] = d2 x∂µ φ(x)∂ µ φ(y)
2
Z Z
1
= d2 x d2 yφ(x)A(x, y)φ(y)
2
as the action21 where 21
Notice that this is a scalar field
whose action is obviously invariant
A(x, y) = −gδ(xµ − y µ )∂µ ∂ µ with respect to translations, rota-
tions, and scale trasformations. A
The two point correlation function is based upon little work shows that it is also SCT
invariant.
R(φ(x)φ(y)) = GF (x, y)
the Feynman propagator, which is defined to be
Gf (x, y) = A−1 (x, y)
in the sense that22 22
That is, in a very loose sense,
inverse under the inner product on
Z
d2 yA(x, y)GF (y, z) = δ(xµ − z µ ) distributions of R2 .

This implies that

∂ ∂
δ(xµ − z µ ) = −g Gf (x, z) (∗)
∂xµ ∂xµ
As we saw previously, rotational and translational invariance means
GF (x, y) = GF (|x − y|) = GF (r)
and, expressing the equation (∗) in polar coordinates and integrating,
we get Z   
1
1 = 2π dr r −g ∂r r∂r GF (r)
r
which, when solved, yields
1
G0F (r) = −
2πgr
so that
1
GF (x, y) = − log |x − y| + const
2πg
In complex coordinates23 , we can rewrite the action as 23
Where
Z i
d2 z = dzdz = dx2
S[φ] = 2g d2 z∂z φ(z, z)∂z φ(z, z) 2

so that
1
R(φ(z, z)φ(w, w)) = − (log(z − w) + log(z − w) + const)
4πg

For a chiral (holomorphic) field, we have ∂z φ(z, z) = 0, so24 24

Note that, for free fields there is
only a single singular term in the
1 1 OPE.
R(∂z φ(z, z)∂w φ(w, w)) ∼ −
4πg (z − w)2
 31

Energy-Momentum tensor
We take an infinitesmal symmetry transformation

x̃µ = xµ + µ

acting on S[φ].
φ̃(x̃µ ) = φ(xµ )
Then we get
∂L
Tµν = ∂ν φ − ηµν L
∂(∂ µ φ)
g
= g∂µ φ∂ν φ − ηµν ∂ρ φ∂ ρ φ
2
so the tensor is symmetric and traceless (as expected from a scale
invariant theory).
In complex coordinates, we can compute (cf section 2.5)
1
T (z) = −2π (T1,1 − 2iT1,2 − T2,2 )
4
= −2πg∂z φ(z, z)∂z φ(z, z)

From quantum theory, we have

h0 | T (z) | 0i = 0

so that
T (z) = −2πg : ∂z φ(z, z)∂z φ(z, z :

OPEs

(A) We compute the radial ordering

R(T (z)∂w φ(w, w)) = −2πg : ∂z φ∂z φ : ∂w φ

 
∼ −2πg : ∂z φ∂z φ : ∂w φ+ : ∂z φ∂z φ : ∂w φ
∂z φ(z, z)
∼
(z − w)2
∂z φ(w, w) ∂z φ(w, w)
∼ +
(z − w)2 (z − w)

Which tells us that ∂z φ(z, z) is a chiral primary field with con-

formal weight h = 1. Similarly, we have that ∂z φ(z, z) is an
anti-chiral primary with conformal weight h = 1.
 32

(B) We now work on the energy-momentum tensor

R(T (w)T (w)) ∼ 4π 2 g 2 (: ∂z φ∂z φ : : ∂w φ∂w φ : )

!
∼ 4πg 2( : ∂z φ∂z φ∂w φ∂w φ : ) + 4 : ∂z φ∂z φ∂w φ∂w φ :

1/2 : ∂z φ∂z φ :
∼ 4
− 4πg
(z − w) (z − w)2
2 φ
1/2 2(−2πg : ∂w φ∂w φ : ) : ∂w φ∂w :
∼ 4
+ 2
− 4πg
(z − w) (z − w) (z − w)
Where the last step comes by Fourier expanding. In conclusion,
we then have

c/2 2T (w) ∂w T (w)

R(T (z)T (z)) ∼ + +
(z − w)4 (z − w)2 (z − w)

where c = 1.

Remark. Note that T (z) is not primary because of the circled term.
It is ‘almost’ a chiral primary field of conformal weight h = 0

c is called the central charge. For the free boson, as we have seen,
c = 1.

Example II: The Free Fermion CFT We take the Euclidean

action of a real (Majorana) fermion25 25
Where the γ µ are Pauli matrices.
Z In principle, we could take any, by
g here we take
S[ψ] = d2 xΨ† γ 2 γ µ ∂µ Ψ
2
 
0 −i
γ1 =
i 0
 
Z ∂z
= g d2 xΨ†  ddots Ψ and
   
0 1
γ2 =
∂z 1 0
so that
Z
= g d2 z ψ∂z ψ + ψ∂z ψ


{γ µ , γ ν } = 2η µν

where Ψ = (ψ, ψ).

The equations of motion are then

∂z ψ = 0 ⇒ ψ is a chiral (holomorphic) field

∂z ψ = 0 ⇒ ψ is an anti-chiral field

We can then calculate the Feynman Propagator26 26

As before, the idea is to write the
Z Z action in terms of a kernel which we
1 then invert
S[Ψ] = d2 x d2 yΨ† (x)A(x, y)Ψ(y)
2
A(x, y) = gδ(xµ − y µ )γ 2 γ µ ∂µ
 33

We then obtain the Feynman propagators27 27

Notice that the fermi statistics is
encoded within the propagator.
1 1
R(ψ(z, z)ψ(w, w)) ∼
2πg z − w
1 1
R(ψ(z, z)ψ(w, w)) ∼
2πg z − w
R(ψ(z, z)ψ(w, w)) = 0

We can also compute the energy-momentum tensor using Noether’s

theorem

T (z) = −πg : ψ∂z ψ :

T (z) = −πg : ψ∂z ψ :

OPEs

(A) We have two possible contractions to worry about in computing

the radial ordering

R(T (z)ψ(w, w) ∼ πg : ψ∂z ψ : ψ(w, w)

1 ∂z ψ(z, z) 1 1
∼ − ψ(z, z)∂z
2 (z − w) 2 z−w
| {z } | {z }
red orange

1 ∂w ψ(w, w) ∂w ψ(w, w)
∼ +
2 (z − w)2 (z − w)

which implies that ψ(z, z) is a chiral primary with conformal

weight h = 12 . Similarly, ψ(z, z) can be shown to be an anti-chiral
primary with conformal weight h = 21

(B) We also have, for the energy-momentum tensor

1/2
2 2T (w) ∂w T (w)
R(T (z)T (w)) ∼ + +
(z − w)4 (z − w) 2 (z − w)

which tells us that the free Majorana fermion has central charge
c = 21 .

2.7 The central charge c

Observation:
c
2 2T (w) ∂w T (w)
R(T (z)T (w)) ∼ + +
(z − w)4 (z − w)2 (z − w)

We call c the central charge. As we have already calculated, the free

boson has c = 1, and the free Majorana fermion has c = 21 .
 34

Properties

• For dilations z 7→ λz, we have

T (z) 7→ λ−2 T (z)

which, in turn, implies

R(T (z)T (w)) 7→ λ−4 R(T (z)T (w))

so the central charge is consistent with the expected scaling rela-

tion.

• For a sum of decoupled CFTs CFT1 , . . . , CFTN , the energy-

momentum tensor is
N
X
Ttotal (z) = T1 (z)
i=1

And, since the theories are decoupled, this means

N
X
R(Ttotal (z), Ttotal (w)) = R(Tk (z)Tk (w))
k=1

In turn, this means that we get

ctotal = c1 + · · · + cN

ie, the central charge is an extensive quantity.

• Transformation behavior under infinitesmal conformal transforma-

tions
xµ 7→ x̃µ = xµ + µ (x)
yields
I
dz
δ T (w) = −[Q , T (w)] = − (z)R(T (z)T (w))
2πi
(z) 2c
I  
dz z(z)T (w) (z)∂w T (w)
= + +
2πi (z − w)4 (z − w)2 z−w
c 3


=− ∂ (w) − 2(∂w (w))T (w) − (w)∂w T (w)
12 w

A finite local conformal transformation takes the form

z 7→ z̃ = w(z)

so that  −2 
dw c 
T̃ (z̃) = T (z) − {w(z); z}
dz 12
with the Schwarzian derivative 28 28
Interestingly, while the invariance
properties proved below related
the Schwarzian derivative to the
complex plane, it also appears to
bear some relation to solutions of the
hypergeometric equation.
 35

2
w000 (z) 3 w00 (z)

{w(z); z} = 0 −
w (z) 2 w0 (z)

For z 7→ u(z) and u 7→ w(u) we have the following ‘chain rule’ for
the Schwarzian derivative
du
{w(u(z)); z} = {u(z); z} + {w(u); u} (∗)
dz
Now, moving on to global conformal transformations in P SL(2, C),
we have
az + b
z 7→ z̃ =
cz + d
Such transformations are generated by

(i) z 7→ λz, which has Schwarzian derivative

{λz; z} = 0

(ii) z 7→ z + c, which has Schwarzian derivative

{z + c; z} = 0

(iii) z 7→ − z1 , which has Schwarzian derivative

2
−3!z −4 −2z −3

1 3
{ ; z} = −2
−
z −z 2 −z −2
= 6z −2 − 6z −2 = 0

As a result, we see that {w(z); z} = 0 for any P SL(2, C) transfor-

mation. This implies that for such transformations
 −2
dw
T̃ (w) = T (z)
dz

and thus, T (z) is a chiral quasi-primary field with conformal weight

h = 2.
 36

Conformal anomaly and central extensions

Symmetry group
G on some space
lifts to action
lifts to a
on phase
unitary
space?
rep?

Quantum Mechanics Classical Mechanics

Want to impose G Poisson Algebra act-
action is unitary ing on phase space

Anomaly
Obstruction to lifts

In some cases of anomalies, we can find a group extension Ĝ of G

by some group H 29 such that a representation can be constructed 29
That is, a Ĝ fitting into a short
with respect to ĝ exact sequence
0 → H → Ĝ → G → 0
Example. In quantum mechanics in three dimensions, consider the
symmetry group SO(3), and the state space of spin 21 states

1 1
,±
2 2
The action of
 
1 1 1 1 1
e2πiJ3 ,± = e2πi(± 2 ) ,±
2 2 2 2

1 1
=− ,±
2 2
does not quite agree with identites.
However, there is an extension of SO(3) by Z/2, ie

SU (2)/(Z/2) ∼
= SO(3)

Which implies that, taking Ĝ = SU (2), and considereding

" #
−1
∈ SU (2)
−1

we get the desired result.

Moreover, this extension is central. That is, Z/2 is in the center of
SU (2).
 37

Example (Classical Mechanics). We can find an explicit central

extension of the Galileo group, see [2] for details.

Central Extensions of the deWitt Algebra

We consider the OPE of T (z), the infinitesmal generator for local
conformal transformations.
c/2 zT (w) ∂w T (w)
R(T (z), T (w)) ∼ 4
+ 2
+
(z − w) (z − w) z−w
Recall that T (w) has the expansion
X
T (z) = z −n−2 Ln
n∈Z
I
dz n+1
Ln = z T (z)
2πi
which gives us the commutation relations
I I
dw dz m+1 n+1
[Ln , Lm ] = w z R(T (z)T (w))
0 2πi w 2πi
From these relations, we retrieve the Virasoro Algebra Vir ⊕ Vir30 30
This algebra is completely charac-
terized by the commutation relations
Claim. Vir is a central extension of the deWitt algebra31 A by C. c
[Ln , Lm ] = (n − m)Ln+m + n(n2 − 1)δn+m,0
That is, there is a short exact sequence of Lie algebra homomorphisms 12
c
f g [Ln , Lm ] = (n − m)Ln+m + n(n2 − 1)δn+m,0
0 → C → Vir → A → 0 12
[Ln , Lm ] = 0
where C is generated by c, and there includes into the center of Vir, 31
Whose commutation relations are of
and f, g respect the Lie bracket of Lie algebras. the form

Remark. Vir is the unique non-trivial32 central extension of A. [`n , `m ] = (n − m)`n+m

32
Non-trivial meaning the sequence
To see why this is the case, we first comment that non-trivial cen- does not split, that is, is not equiva-
tral extensions are classified by the cohomology group lent to the sequence

Z 2 (g, a) 0→a→a⊕g→g→0
H 2 (g, a) =
B 2 (g, a)
Where
(a) Θ ∈ Z 2 (g, a) means that

Θ:g⊗g→a
Θ(X, Y ) = −Θ(Y, X)
Θ(X, [Y, X]) + Θ(Z, [X, Y ]) + Θ(Y, [Z, X]) = 0

we call such Θ cocycles

(b) Θ ∈ B 2 (G, a), the exact cocycles, are characterized by the prop-
erty that there exists
µ:g→a
such that
Θ(X, Y ) = µ([X, Y ])
 38

Given a representative of an equivalence class [Θ] ∈ H 2 (g, a), we see

that Θ defines a Lie bracket for h (The direct sum of g and a as vector
spaces) by the formula33 33
More generally, the second Lie Al-
gebra Cohomology classifies extensions
[L ⊕ Θ, L0 ⊕ Θ]h = [L, L0 ] + Θ([L, L0 ]) of a Lie algebra g by a module over
g. See, for example [3] pp. 234 or [4]
pp. 153 for a purely mathematical
Since the Θ is exact, we have treatment.

L̃ = L ⊕µ L

and hence
[L̃, L̃0 ]h = [L̃, L̃0 ]g
Now, returning to the case of the de Witt algebra A, one can com-
pute that
H 2 (A, C) ∼
=C∼ = hΘi
where
c
Θ(Ln , Lm ) = n(n2 − 1)δn+m,0
12
This tells us that Vir is unique as a non-trivial central extension34 . 34
For a more thorough treatment, see
[5] ch. 4 & 5.

The physical interpretation of c

(1) Casimir Energy35 : Previously we mapped the cylindrical world- 35

The Casimir effect is a phenomenon
sheet of a string to the (punctured) conformal plan. If we now do whereby a force is observed between
two plates positioned extremely close
the reverse, we get a map together.

L
z 7→ ζ = log(z)
2π
where L is the circumference of the cylinder. In these new coordi-
nates,
 
 2 Schwarz. Deriv.
2π c z }| {
Tcyl (ζ) = z 2 T (z) − {log(z), z} 
 
L 12 | {z }
1 −2
2z
| {z }
dζ 2
( dz )

So that
 2 
2π c 
Tcyl (z) = z 2 T (z) −
L 24

Assuming that the expectation values satisfy

hT (z)i = T (z) = 0

then
1

hT0,0 i = − hTcyl (ζ)i + T cyl (ζ)
2π
π(c + c)
=
12L2
 39

This allows us to compute an energy for the cylinder

Z L
π(c + c)
Ecyl = − hT0,0 i = −
0 12L
which is the Casimir energy.

Ecyl

Notice that, as expected, if L → ∞, we have Ecyl → 0

Casimir Energy of the Free Scalar Field (Heuristic deriva-
tion):
We can compute the ground state energy of a sum of harmonic
oscillators X 1
E|0i = ωp
2
p∈{Harm. Osc.}

which yields36 36
We can think of our harmonic
∞   oscillators as merely being waves on a
X 1 2π
E|0i = 2 ·k τ -slice. The factor of 2 appearing in
2 L the equation can then be said to arise
k=1
as a result of left- and right-moving
To attempt to remove this divergence, we introduce the UV regu- waves (sin and cos respectively).
lator
L
= 1
ΛU V
and add an exponential damping term
∞  
X 1 2π

E|0i = 2 · k e−k
2 L
k=1
∞
!
2π d X −k
= e
L d
k=0
 
2π 1 1
= − + O()
L 2 12

We then define the regularized energy to be

 
reg  2π π
E|0i = lim E|0i − 2 = −
→0 L 6L

And, since for the scalar field c = c = 1,

reg π
E|0i =− (c + c)
12L
 40

(2) Trace anomaly on a closed oriented Riemann Surface Cg

of genus g:
On the classical level, Tµµ = 0.
On the quantum level, we get

hTµµ iCg ∼ c · R

where R is the Ricci scalar37 . 37

The term on the left hand side
is called the trace anomaly. Notice
The proportionality constant can be computed from the free that R is a local quantity which
fermionic CFT. See [6] for more details. This yields depends locally on the metric and has
scaling dimension 2. This dependence
c on the metric is acceptable since a
T µµ = R conformal transformation is also a
24π metric transformation.

2.8 The Hilbert Space of States

As before we consider the OPE
c/2 zT (w) ∂w T (w)
T (z)T (w) ∼ + +
(z − w)4 (z − w)2 z−w

with X
T (z) = z −n−2 Ln
n

and the commutation relations

c
[Ln , Lm ] = (n − m)Ln+m + n(n2 − 1)
12
In this system the vacuum, T (z)|0i for z → 0 must be well defined,
which implies that38 38
This follows from the fact that
Ln |0i = Ln |0i = 0 T (z)|0i must not have singular terms.

for n > −2.

In particular, the vacuum |0i invariant under global conformal
transformations, which are generated by L±1 , L0 , L±1 , and L0 .

Raising and Lowering operators of primaries

We consider a primary of conformal weights (h, h)
X
φ(w, w) = w−m−h w−k−h φm,k
k,m

And the commutation relations39 39

Following from the OPE
hT (w) ∂w T (w)
I
dw T (z)φ(w, w) ∼ +
[Ln , φ(w, w)] = R(T (z)φ(w, w))z n+1 (z − w)2 (z − w)
w 2πi
− h(n + 1)wn φ(w, w) + wn+1 ∂w φ(w, w)

Which, in turn, implies40 40

Because φ−m,−k |0i vanishes for
m < h, k < h.
 41

[Ln , φm,k ] = (n(h − 1) − m)φn+m,k

We define an asymptotic state

|h, hi := φ(0, 0)|0i = φ−h,−h |0i

And get the action of L0

L0 |h, hi = [L0 , φ(0, 0)]|0i = h|h, hi

analogously, we see that

L0 |h, hi = h|h, hi

We then see that |h, hi is the energy eigenstate of the Hamiltonian

H = L0 + L0

The commutation relations

[L0 , φm,k ] = −mφm,k

tell us that φm,k are raising operators for m < 0 and lowering opera-
tors for m > 0.

Verma Module
The Virasoro raising operators L−m (m > 0) acting on |h, hi yield

[L0 , L−m ] = mL−m

⇒ L0 L−m |h, hi = (m + h)|h, hi

By similar arguments, we see that Lm for m > 0 are lowering

operators

Lm |h, hi = Lm φ−h,−h |0i

= (m(h + 1) + h)φ−h+m,−h |0i
=0

where the last equality holds because φ−m,−h |0i = 0 for m < h.

Definition. A descendant state is given by acting with a chain of

raising operators L−m1 , . . . , L−kn on the asymptotic state |h, hi. Any
descendant state can be written in the form

L−k1 · · · L−kn |h, hi

with the ordering convention 1 ≤ k1 ≤ k2 · · · ≤ kn

The Hilbert space of states arising from |h, hi is called a Verma

modules. It forms a representation of the Virasoro algebra41 41
ie a module over the Virasoro
algebra.
 42

2.9 Conformal Families and Operator Algebra

Recall that a descendant state42 42
Recall:
X
L−n |h, hi = [L−n , φ(0, 0)]|i T (z) = z −n−2 Ln
n
dz R(T (z)φ(0, 0)
I
= |0i
0 2πi z n−1
for n ≥ 1 yields a definition of a descendant field (secondary field):
dz R(T (z)φ(w, w))
I
φ{u} (w, w) =
w 2πi (z − w)n−1
for n ≥ 1. Recursively, we have descendant states:
dz R(T (z)φ{ks−1 ,...,k1 } (w, w)
I
{ks ,ks−1 ,...,k1 }
φ (w, w) =
w 2πi (z − w)ks−1
where ki ≥ 1
Remark.
• We use the short-hand notation
~
φ{k} (w, w) = φ{ks ,ks−1 ,...,k1 } (w, w)

• If we include anti-holomorphic descendants43 , we get states 43

Using the operator L−n instead of
~ L−n .
~
φ{k},{k} (w, w)

We can also compute correlation functions of descendants, for in-

stance:
I
{k} dz 1
hφ0 (w0 )φ1 (w1 ) · · · φN (wN )i = k−1
hT (z)φ0 (w0 ) · · · φN (wN )i
w 2πi (z − w0)
I 0
dz 1
= k−1
w0 2πi (z − w0)
N   !
X 1 hn
× ∂w + hφ0 (w0 ) · · · φN (wN )i
n=0
(z − wn ) n (z − wn )2
where the second equality follows from the Ward identity (cf section
2.5).
The correllators
hT (z)φ0 (w0 ) · · · φN (wN )i
can have poles only at z → wk . This implies that
N I
{k}
X dz 1
hφ0 (w0 )φ1 (w1 ) · · · φN (wN )i = − k−1
2πi (z − w0)
`=1 w`
N   !
X 1 hn
× ∂w + hφ0 · · · φN i
n=0
z − wn n (z − wn )2
N  
X ∂w` he ll
=− + (−(n − 1))
(w` − w0 )k−1 (w` − w0 )k
`=1

× hφ0 · · · φN i
 43

So that

hφ{k} (w0 ) · · · φN (wN )i = L−k hφ0 (w0 ) · · · φN (wN )i

where
N  
X (k − 1)h` 1
L−k = − ∂`
(w` − w0 )k (w` − w0 )k−1
`=1

In general, any correlator

{k~0 } {k~ }
hφ0 · · · φN N i

can be rewritten in terms of differential operators acting on

hφ0 · · · φN i

with the help of the Ward identity. Therefore,

hφ0 · · · φN i

determines all descendant correlators.

Definition. The set of a primary field φ together with all its descen-
dant fields is called the conformal family of φ, which we denote by
[φ].

Conformal families have the following properties:

• Members of the conformal family [φ] transform under (local) confor-

mal transformations among themselves. That is, OPEs of T (z) with
any field of [φ] will involve only fields of [φ].

• We have the operator state correspondence

1:1
{Conformal families} ↔ {Verma Modules}

The Operator Algebra

For a given CFT, the OPEs among all its primaries (including
regular terms) form the so-called operator algebra. The knowledge of
the operator algebra determines all correlators of the CFT (it ‘solves’
the CFT).
Given a CFT (with countably many primaries), we normalize pri-
maries such that
δk`
hφk (w, w)φ` (z, z)i =
(z − w)2hk (z − w)2h`

Then44 , 44
See, for instance, the definitions of
in and out states in section 2.3.
lim w2hk w2hk hφk (w, w)φ` (z, z)i = hhk , hk |h` , h` i = δk`
(w,w)→∞
(z,z)→0
 44

The orthogonality of primaries therefore implies the orthogonality of

entire Verma modules. To see this, we can use the Virasora algebra for
descendant states. For example:

L−n |h` , h` and L−n2 L−n3 |hk , hk i (ni > 0)

gives us

hhk , hk |L†−n3 L†−n2 |h` , h` i = hhk , hk |Ln3 Ln2 L−n1 |h` , h` i

= hhk , hk |Ln3 [Ln2 , L−n1 ]|h` , h` i
= hhk , hk |Ln3 ((n2 + n1 )Ln2 −n1 )
c
+ n22 (n22 − 1)δn2 −n1 ,0 |h` , h` i
12
non-vanishing
= (n2 + n1 )hhk , hk |Ln3 Ln2 −n1 |h` , h` i for n1 >n2
n1 >n2
= (n2 + n1 )hhk , hk |[Ln3 , Ln2 −n1 ]|h` , h` i
= (n2 + n1 )hhk , hk |(n3 + n1 − n2 )Ln2 −n3 −n1
c
+ n3 (n23 − 1)δn2 +n3 −n1 ,0 |h` , h` i
12
= (n2 + n1 )δn2 +n3 ,n1 ((n3 + n1 − n2 )h`
c
+ n3 (n23 − 1))hhk , hk |h` , h` i
12
=0

Scale invariance determines the structure of the operator alge-

bra:

X X X  s{~k}{~k} ~
φk (z, z)φ` (0, 0) = Ck` z hs −hk −h` +|k|
[φs ] {~
k} {~
k}

~ ~ ~
×z hs −hk −h` +|k φ{s k}{k} (0, 0)

where X
|~k| = kn
n

and
~k = (k1 , k2 , . . .) k1 ≤ k2 ≤ k3 · · ·

s{~
k}{~
k}
The constants Ck` are the structure constants of the operator
algebra.

Three-Point Correlator
 45

We can compute the correlator

hφs |φk (z, z)|φ` i = lim w2hs w2hs hφs (w, w)φk (z, z)φ` (0, 0)i
w,w→∞

w2hs
= lim Csk`
w,w→∞ z hk +h` −hs whs +h` −hk (w − z)hs +hk −h`
anti-holomorphic


× part
Csk`
=
z hk +h` −hs z hk +h` −hs
or, alternately
XXX
hφs |φk (z, z)|φ` i = hφs | [. . .] |0i
[φ` ] {~
k} {~
k}
r{}{}
X
= Ck` hφs |φr iz hs −hk −h` z hs −hk −h`
r
r{}{}
δsr Ck`
=
z hk +h` −hs z hk +h` −hs
so that
s{}{} s
Ck` = Ck` = Csk`
Since all correlators of descendants arise from correlators of pri-
maries, one can show that:

s{~
k}{~
k} s k} s{~
s{~ k}
Ck` = Ck` βk` β k`

where
s{~
k} s{~
k}
βk` = βk` (hs , hk , h` , c)
s{}
βk` = 1

Example. Consider two chiral primaries φk (z) and φ` (z) with h =

hk = h` (for simplicity), then
X
s hs −2h
φk (z)φ` (0) = Ck` z Xs (z)
s

where
∞ X
s{~
k}
X
Xs (z) = z N βk` L−{~k} φs (0)
N =0 {~k}
|~
k|=N

Hence
∞
X
Xs (z)|0i = z N |N ; hs i
N =0

where |H; hs i is the level N state descending from

|φs i = |hs i
 46

We can then compute:

Ln φk (z)φ` (0) = [L − n, φk (z)]|h` i

= z n+1 ∂z + (n + 1)hz n φk (z)|h` i

so that
∞
X
Ln (Xs (z)|0i) = z N Ln |N ; hs i
N =0
∞
!
X
(N + hs − 2h)z n+n + (n + 1)hz N +n |N ; hs i


=
N =0

Which gives us that

Ln |N + n; hs i = (hs + (n − 1)h, +N )|N ; hs i (∗)

For low N , we can then begin to examine what descendant states

are produced:

Level 1 There is one descendant state

s{1}
|1; hs i = βk` L−1 |hs i

(∗)
(i) L1 |1; hs i = hs |0; hs i = hs |hs i
s{1} s{1}
(ii) L1 |1; hs i = βk` [L1 , L−1 ] |hs i = 2βk` h2 |h2 i
| {z }
2L0
Hence,
s{1} 1
βkl =
2
Level 2 There are 2 descendant states
s{2} s{1,1}
|2, hs i = βk` L−2 |hs i + βk` L−1 L−1 |hs i

(i) We have
(∗) 1
L1 |2; hs i = (hs + 1)|1; hs i = (hs + 1)L−1 |hs i
2
(∗)
Ls |2; hs i = (hs + h)|hs i

From (i) and (ii), we find a pair of linearly independent equations

s{1,1} s{2}
for βk` and βk` :

s{1,1} c − 12h − 4hs + chs + 2h2s

βk` =
4(c − 10hs + 2chs + 16h2s )
s,{2} 2h − hs + 4hhs + h2s
βk` =
c − 10hs + 2chs + 16h2s
 47

Remark.
s{~
k}
• All of the coefficients βk` (hk , h` , hs , c) can, in principle, be recur-
sively determined. As a result, the operator algebra of a CFT is
determined completely by

a) conformal families [φs ]

b) three-point correlators of the primaries, ie Csk`

• Csk` must be obtained separately (eg, by dynamic input or crossing

symmetries).

2.10 Conformal Blacks & Crossing Symmetries

Having now treated three-point correlation functions, we an in a posi-
tion to revisit four-point and higher correlation functions, ie

hφk (z1 , z 1 )φ` (z2 , z 2 )φm (z3 , z 3 )φn (z4 , z 4 )i

We apply a conformal transformationwhich sends

z1 → ∞, z2 → 1, z3 → η, z4 → 0

where, η is, as usual, the cross ratio

(z1 − z2 )(z3 − z4 )
η=
(z1 − z3 )(z2 − z4 )

We can then write down a correlation function matrix element45

G`k
mn (η, η) = lim z
2hk 2hk
z hφk (z, z)φ` (1, 1)φm (η, η)φn (0, 0)i 45
Where we use the indexing con-
z,z→∞
vention that the indices are read
= hhk , hk |φ` (1, 1)φm (η, η)|hn , hn i counterclockwise around the symbol,
as in
If we then insert the operator product expansion of φm (η, η) and G`k
mn
φn (0, 0), we get
*
p{~
k}{~
k} hp −hm −hn −|~
XX
`k k|
Gmn (η, η) = hk , h)k | φ` (1, 1) Cmn η
[φp ] ~
k,~
k

hp −hm −hn −|~
k|
×eta × L−{~k} L | hp , hp
−{~
k}

We can then compute

X `k
G`k
mn (η, η) =
p
Cmn `k
Cpk` Fmn (p|η)F mn (p|η)
[φp ]
 48

where
X ~ ~ hhi |φj (1)L−{~k} |hp i
`k
Fmn (p|η) = η hp −hm −hn η |k| βmn
p{k}
hhi |φj (1)|hp i
{~
k}

× L−{~k} L |hp , hp i
−{~
k}

The
`k
Fmn (p|η)
are called conformal blocks of the conformal family [φp ].
`k
Remark. • Fmn (p|η)η hn +hm −hp are regular holomorphic functions at
η = 0.
`k
• The coefficients of the expansion of Fmn (p|η) depend only on
c, hp , h` , hk , hm , hn
that is, there is no dependence on the 3-pt structure constants.
`k
• For explicit expressions for Fmn (p|η), see, eg, [7]

Crossing Symmetries
The Gij
k` depend on a particular order of primaries, which results
in one particular OPE. We could, however, think of reordering the
primaries to get a different expression.
We can diagrammatically represent the conformal blocks in the
following way, permuting the indices each time46 : 46
Only the index circled in orange
must be fixed under this permutation.
1) This is because it determines both the
out state and the η hi ··· regularization
(1) j i (∞) factor in the limit.

j i
Fk` (p|x) = p

(η) k ` (0)
which, in turn, implies that we can write the s-channel expres-
sion47 47
From quantum field theory one
   would also expect a T-channel and a
j i j i U-channel, which, as we will shortly
   see, also exist.
p p
X p   
ji
Gk` (x) = Cij Cpk` 
  
 
p
  
(η) k ` (η) k `
  

Here, we have
(∞ − 1)(η − 0)
x= =η
(∞ − η)(1 − 0)
 49

2)
(1) j i (∞)
ji
p
Fk` (p|x) =
(η) k ` (0)
giving the t-channel
  
X p p
G`i
kj (1 − η, 1 − η) = Ci`p Cpkj   
p η η

Here we have
(∞ − 0)(η − 1)
x= =1−η
(∞ − η)(0 − 1)
3)
(1) j i (∞)

F`jki (p|x) = p
(η) k ` (0)
giving the u-channel
  
 
1 1 X p
G`i
kj , = Cik Cp`j  p
 p

η−1 η−1 p η η

here,
(∞ − η)(0 − 1) 1
x= =
(∞ − 0)(η − 1) η−1
We would expect these different interpretations to agree, that is, we
impose the crossing symmetries
 
ji `i ki 1 1
Gk` (η, η) = Gkj (1 − η, 1 − η) = G`j ,
| {z } | {z } η−1 η−1
s-channel t-channel | {z }
u-channel

Bootstrap approach: Specify the dynamics of a CFT by consis-

tency using crossing symmetries.
Given a CFT with central charge c and N conformal families we
have
N 3 + |{z}
|{z} N
Ck`m h`

parameters, for k, `, m ∈ {1, . . . , N }. The crossing symmetries provide

N 4 constraints, so naive counting suggests that crossing symmetries
fix all unknown parameters. This is true in some cases, eg for minimal
models.
 50

Remark. Higher point correlators can be calculated analogously to

the 4 point case.

3 Minimal Models
Definition. A minimal model conformal field theory is a conformal
field theory with a finite numbe rof conformal families.

Example. The 2d Ising model at criticality.

Remark. For the purposes of this lecture, we will focus on unitary

minimal models.

3.1 Warm-up: Representations of su(2)

The Lie algebra48 su(2) has generators J3 and J± 49 , satisfying 48
More commonly studied are rep-
resentations of the Lie group SU (2),
[J3 , J± ] = ±J± but for our purposes, we need the
corresponding Lie algebra.
[J+ , J− ] = 2J3 49
It also admits a Casimir Operator
1
J 2 := J3 + {J+ , J− }
To find representations of su(2), we start with a highest weight 2
state |ji with

J3 |ji = j|ji
J+ |1i = 0

k
We can then define a decendant state J− , where
k k
J3 (J− |ji) = (j − k)|J− |ji
k−1
X
k k−`−1 `
J+ (J− |ji) = 2 J− J3 J− |ji
`=0
k−1
!
X
k−1
=2 (j − `) J− |ji
`=0
k−1
= k(2j + 1 − k)J− |ji

This implies that, for j ∈ 2N0 , the value k = 2j + 1 yields a highest

weight state, ie  
2j+1
J+ J− |ji = 0
The state
2
|χ2j+1 i := J− j + 1|ji
is called a singular vector. |χ2j+1 i generates a subrepresentation of the
highest weight representation generated by |ji.

Hilbert Space of States & Unitary Representations

 51

Let |ψi be a state with conjugate hψ|,

J3 = J3†
†
J± = J∓

and take the normalization

hj|ji

(i) For 2j ∈ N0 , if we have

k1 k2
hj|J+ J− |ji =
6 0

then k1 = k2 . Additionally

hχ2j+1 |χ2j+1 i = 0

and
k k
hχ2j+1 |J+ J− |χ2j+1 i = 0
so that the subrepresentation generated by |χ2j+1 i yields only
null states.
We can therefore take our Hilber space of states to be the quo-
tient of our representation by the relations
k
|ψi ∼ |ψi + αJ− |χ2j+1 i

That is
 k
H = J− |ji k=0,...,2j
= h|jii / h|χ2j+1 i]i

So we get a finite spin representation of su(2)50 50

Which is, in addition, unitary.

(ii) Let j > 0, 2j ∈

/ N0 : We then get a non-finite representation of
su(2). Pick k = d2j + 1e, then
k k k−1 k−1
hj|J+ J− |ji = hj|J+ J− |ji ×k (2j + 1 − k)
| {z } | {z }
>0 <0

We easily see that we get a negative norm state at level d2j + 1e,
giving us a non-unitary representation of su(2).

Observation. • Singular vectors give rise to unitary (finite dimen-

sional) representations of su(2).

• Highest weight representations without singular vectors yield (infi-

nite dimensional) non-unitary representations of su(2).
 52

3.2 Reducible Verma Modules & Singular Vectors

We explore highest weight representations of the Virasora algebra with
central charge c. Let |hi be a highest weight state:

Ln |hi = 0 for n > 0

L0 |hi = h|hi

The space of states will then be

L−{~k} |hi

with inner product.

Definition. A singular vector is a descendant state |χi with Ln |χi =

0 for all n > 0, that this, |χi is a highest weight state.

Observation. A singular vector and its descendants are orthogonal to

any other state.

Proof. Let |χi be a singular vector at level N , so that it has an expan-

sion as X
|χi = c~k L−{~k} |hi
~
k
|~
k|=N

Then the state

hψ|L−{~`} |χi

can only be non-vanishing if ψ is a state at level N + |~`|.

Let
|ψi = L−{m}~ |hi

~ = N + |~`|. Then
where |m|

hψ|L{vec`} |χi = hh|L{m}

~ L−{~`} |χi
h i
= hh| L{m}
~ , L−{~ `} |χi = 0

because h i X
L{m}
~ , L−{`} =
~ D~s L~s
~
s

~ − |~`| = N > 0.
where |~s| = |m|

Reducible Verma Module

Definition. A Verma module V (c, h) is reducibleif there is a singular

vector at some level N , ie
X
|χi = c~k L−{~k} |hi
~
k
|~
k|=N
 53

such that, for all n > 0, Ln |χi = 0.

In this case, |χi is a highest weight state generating a Verma sub-
module Vχ of V (c, h).
Remark. • |χi is a primary state of conformal weight h + N , and it
is also a descendant state of |hi

• Vχ is orthogonal to V (c, h) with respect to the defined inner prod-

uct.

• Let V (c, h) be a Verma Module with singular vectors |χi i, then we

can construct irreducible 51 Verma module M (c, h) by quotienting 51
See previous note.
out all Verma submodules

M (c, h) = V (c, h)/ ∼

where
|φi ∼ |φi + |ψi
L
for all |φi ∈ V (c, h) and for all |ψi ∈ i Vχi .

• M (c, h) are the building blocks of minimal models.

Singular Vectors & Negative Norm States at Low

Levels

Level 0 hh|hi = 1: normalization condition.

Level 1 L−1 |hi has norm

hh|L1 L−1 |hi = 2hhh|hi = 2h

so that the Gram matrix at level 1 is

M (1) = (hh|L1 L−1 |hi) = (2h)

That means we get a null state for h = 052 , and a necessary condi- 52
That is, L−1 |0i is a singular vector.
tion for unitary representations is that This null state corresponds to the
vacuum.
h≥0

Level 2 We now have two possible states

|ψ1,1 i = L2−1 |hi

|ψ2 i − L−2 |hi

We can compute the 2 × 2 Gram matrix at level 2,

" #
(2) hψ1,1 |ψ1,1 i hψ1,1 |ψ2 i
M (c, h) =
hψ2 |ψ1,1 i hψ2 |ψ2 i
" #
4h(2h + 1) 6h
=
6h 4h + 21 c
 54

We compute that the trace of this matrix is

1
trM (2) (c, h) = 8h(h + 1) + c
2
and the determinant53 53
Known as the Kac determinant.

det M (2) (c, h) = 32(h − h1,1 )(h − h1,2 )(h − h2,1 )

Each zero of this determinant will correspond to a singular vector.

the first,
h1,1, = 0
corresponds to the singular vector we have already found. The
other two
1  p 
h1,2 = 5 − c − (1 − c)(25 − c)
16
1  p 
h2,1 = 5 − c + (1 − c)(25 − c)
16
are new

Exercise. There are two singular vectors at level 2, |χ1,2 i and |χ2,1 i
associated to h = h1,2 and h = h2,1 respectively.

A First Glance at Unitary Representations

We have singular vectors h = h1,1 = 0, and
  
4 2 4 c 3
h = h1,2 , h2,1 ⇔ 1 = h+ − h− +
3 3 3 6 2

h c=1

5
h= 8
Unitary repr. still
possible from anal-
generic ysis up to level 2,
points non- BUT: we expect
1 unitary further constraints.
h= 4

manifest non-unitary rep-

resentation (level 1)
 55

3.3 Kac Determinants

The Kac determinants are the determinants of the Gram Matrices,
and display a number of useful properties:

(i) A singular vector |χi at level K yields null states at level N >
K:
L−{~k} |χi

with |~k| = N − K. Each singular vector at level K therefore

yields p(N − K)54 null states at level N . 54
Where p(m) is the number of
partitions of m. More precisely, it is
(ii) The order of the entries of the Gram determinants M (N ) (c, h) at the number of distinct non-decreasing
sequences of natural numbers which
level N in h is given by add up to m.

• Since each element in L† = L{~k} ‘commutes to L0 ’,

−{~
k}
 
ordh hh|L† L−{~k} |hi = length(~k)
−{~
k}

• For ~k 0 6= ~k
 
ordh hh|L† L−{~k} |hi < length(~k)
−{~
k0 }

Since now not all generators in L† ~ = L{~k} do not commute

−{k}
to L0 .

This tells us that the diagonal terms in the Gram matrix at level
N give rise to the leading contribution in h55 : 55
The second step of this calculation
  is a purely number theoretic identity,
X
ordh det M (N ) (c, h) = length(~k) not a physical principle.

~
k
|~
k=N
X
= p(N − r · s)
r,s∈Z
1≤r,s
r·s≤N

(iii) At level N we have

# ((r, s) | 1 ≤ r, s and r · s = N )

new singular vectors, labelled by hr,s

We can compute a formula for the Kac determinants

p(N −r·s)
Y
det M (N ) (c, h) = αN (h − hr,s (c))
r,s≥1
r·s≤N

where αN is some numerical (non-vanishing) constant independent of

h and c.
 56

Conjecture (Kac). 56 The values of hr,s and c for minimal models 56

Later proved by Feign and Fuchs.
are given by
((m + 1)r − ms)2 − 1
hr,s =
4m(m + 1)
6
c(m) = 1 −
m(m + 1)
Note. There are two possible solutions for the second equation
r
1 25 − c
m1/2 = − ±
2 1−c
However,
hr,s (m1 (c)) = hs,r (m2 (c))
We take the Convention that, for m real, we choose m ∈ R≥0

Unitary representations of Vir

We begin with a few observations

(i) For level 1, if

det M (1) (c, h) ≥ 0
then h ≥ 0

(ii) hh|L †−n L −n |hi = 2nh + c

12 n(n
2
− 1) > 0
This must hold for all n, which implies that c ≥ 0

(iii) If we consider the Kac determinants 1 < c < 25, we get that
m 1/2 ∈/ R, which implies that h r,s ∈
/ R for r 6 = s. If r = s, then
h r,r < 0 for r >. Similarly, for c ≥ 25, h r,s < 0, so there are no
positive solutions for h.
Taken all together, this implies that det M (k) is non-vanishing
and positive definite for c > 1 and h ≥ 0 (see exercises).

(iv) In the region 0 ≤ c ≤ 1, h ≥ 0 we have the Kac determinant

√ √
2
96h r,s + 4(1 − c) = 1 − c(r + s) ± 25 − c(r − s) ≥ 0

a) near c = 1, take c = 1 − 6. For r 6 = s

1 1 √
h r,s () = (r − s) 2 ± (r 2 − s 2 )  + O()
4 4
For r = s
1−c 2
h r,r = (r − 1)
24
Diagrammatically, we can represent what we know so far about uni-
tary representations and minimal models in a somewhat finer version
of the graph from last section:
 57

h
h4,2
h3,1

35
24
h4,3

min. model
m=4
h1,4

1 1
h3,2

non-unitary unitary
representations representations

5 h2,1
8
min. model
m=3

h1,3
1
3
h3,3
1
4
h2,4
1 h2,2
8
h3,4
h1,2 h2,3
0 c
h1,1 1 7 1
min. model

2 10
m=2

non-unitary
representations

Returning to the region 0 ≤ c < 1, h ≥ 0,

• A generic point in this region gives rise to non-unitary representa-

tions only.

• Frienan, Qiu, Shenkar, in [8] found unitary representations on ‘first

intersections’. The idea is that the resulting singular vectors at
‘first intersections’ are sufficient to make the associated Verma
module Unitary. These ‘first intersections’ are given by precisely
the formulae from Kac’s conjecture above, where 1 ≤ r < m and
1≤s<r

Note. • h r,s (m) = h m−r,m+1−s (m) for m ∈ Z ≥2

• We denote the associated conformal families

[φ r,s ] ≡ [φ m−r,m+1−s ]
 58

We can codify the distinct conformal families with the use of

conformal diagrams. For example, when m = 2, we have c = 0
h 1,2 = h 1,2 = 057 or, diagrammatically 57
That is, this minimal model gives
the vacuum CFT.
r

φ1,1

φ1,2
s

1
In a more complicated case, we can take m = 3, so that c = 2.
Looking at the diagram
r

φ2,1 φ2,2 φ2,3

φ1,1 φ1,2 φ1,3

we see that

[φ1,1 ] = [φ2,3 ], h1,1 = 0

1
[φ2,1 ] = [φ1,3 ], h2,1 =
2
1
[φ2,2 ] = [φ1,2 ], h2,2 =
16
The conformal diagrams list unitary represenations for a given
central charge c = c(m).

Goal: Construct, for a given c(m), unitary CFTs constructed frm

the conformal families associated to unitary representations of the
Virasora algebra for the given c(m). The resulting CFT’s are unitary
minimal models.

3.4 Fusion Rules

Singular vectors give rise to selection rules for non-vanishing three
point correlators.
 59

Example. We work at level 2 (cf Exercise 8.3). We have a singular

vector  
3
L−2 − L2−1 |h2,1 i = |χ2,1 i
2(2h2,1 + 1)
with the null field
{2} 3 {1,1}
χ2,1 (z) = φ2,1 (z) − φ2,1
2(2h2,1 + 1)

where58 58
This holds because
φ1,1 ∂z2 φ2,1 (z) dz
I
2,1 (z) = {1}
φ2,1 (w) = R(T (z)φ2,1 (w))
w 2πi
Since |χ2,1 i is singular, we have = ∂w φ2,1 (w)

0 = hχ2,1 (z)φ1 (z1 )φ2 (z2 )i

 z C(z,z1 ,z2 )
3 }| {
= L−2 − ∂z2 hφ2,1 (z)φ1 (z1 )φ2 (z2 )i
2(2h2,1 + 1)

where

C(z, z1 , z2 ) = C(z − z1 )h2 −h1 −h2,1 (z1 − z2 )h2,1 −h1 −h2 (z2 − z)h1 −h2,1 −h2

(cf section 2.9). We then have the differential equation for C(z, z1 , z2 ),
given by
 
h1 ∂z1 h2 ∂z2 3 2
0= − + − − ∂
(z1 − z)2 z1 − z (z2 − z)2 (z2 − z) 2(2h2,1 + 1) z
× C(z, z1 , z2 )

This gives us a constraint for C 6= 0:

2(h2,1 + 1)(h2,1 + 2h2 − h1 ) = 3(h2,1 − h1 + h2 )(h2,1 − h1 + h2 + 1)

Inserting h2,1 = h2,1 (m), h1 = hr,s (m) and h2 = hr0 ,s0 (m) yields the
solution
r 0 = r ± 1 s0 = s
So the selection rule is

6 0 ⇒ r0 = r ± 1,
hφ2,1 φr,s φr0 ,s0 i = s0 = s
6 0 ⇒ r0 = r,
hφ1,2 φr,s φr0 ,s0 i = s0 = s ± 1

Fusion rules, on the other hand, summarize the conformal families

appearing in the OPE. For a CFT with states [φp ], we can define the
fusion product: X
`
[φp ] × [φj ] = Nkj [φ` ]
`
where Nkj ∈ Z≥0 . For minimal models in particular, we can go fur-
ther, and say
`
Nij ∈ {0, 1}
 60

The three point selection rules at level 2 then tell us

(r 0 ,s0 )
N(2,1),(r,s) = 0 for (r0 , s0 ) 6= (r ± 1, s)

and
(r 0 ,s0 )
N(1,2),(r,s) = 0 for (r0 , s0 ) 6= (r, s ± 1)
A dynamical analysis of minimal models yields59 for a unitary 59
See, for example, [6] or [9].
minimal model c = c(m)

0 0
X −1
k=r+r X −1
`=s+s
 
[φ(r,s) ] × [φ(r0 ,s0 ) ] = φ(k,`)
k=1+|r−r 0 | `=1+|s−s0 |
k+r+r 0 ≡1 mod 2 `+s+s0 ≡1 mod 2

3.5 The Critical Ising Model

Recall the Ising model from the first lecture:

We have N × N lattice of critical sites:

1
σ spin 2

These spins are equipped with a nearest neighbor interaction. The

energy for the system is given by
X
E({σ}) = − σi σj
adjacent
lattice sites

where σi ∈ ± 12 . There are ground states for the system: all states
having the same spin (ie either all | ↑i or all | ↓i). The partition func-
tion is X
Z= exp (−E({σ})β)
{σ}

1
where β = T is the inverse temperature. The energy at a site is
 X
k = σi σk
q n o
i∈ nearest neigh.
of k
 61

We can take a continuum limit, a → 0, where N a = const. In this

case, we get

σ(z, z) spin field

(z, z) energy density field

The correlation functions for these fields will be given by

X
hσ(z, z))σ(0, 0)i = lim Z −1 σk σ0 e−βE({σ})
a→0
N a=const {σ}
X
−1
h(z, z))(0, 0)i = lim Z k 0 e−βE({σ})
a→0
N a=const {σ}

The Ising model has a critical temperature Tcrit

h0i = ±1 hσi ' 0

T
0 Tcrit

and the correlator goes as

hσ(z, z)σ(0, 0)i ∼ e−|z|/ζ(T )

for T > Tcrit . Where ζ goes as:

Correlation
Length

T
Tcrit

At Tcrit , Onsager showed local conformal invariance, and computed

the correlators
1
hσ(z, z)σ(0, 0)i =
|z|1/4
1
h(z, z)(0, 0)i = 2
|z|

with c = c = 12 .
He also computed the field content of the Ising model60 60
It is a spinless theory, ie h = h.
 62

 
1 1
σ(z, z) : (h, h) = , spin operator
16 16
 
1 1
(z, z) : (h, h) = , energy operator
2 2
1: (h, h) = (0, 0) (in any CFT)

The m = 3 minimal model has c = 21 , and

1
[φ2,2 ] = [φ1,2 ] : h2,2 =
16
[φ1,1 ] = [φ2,3 ] : h1,1 = 0
1
[φ2,1 ] = [φ1,2 ] : h2,1 =
2
The physical ‘Ising minimal model’ includes the anti-holomorphic
sector to match with the operators

σ(z, z) = φ2,2 (z)φ2,2 (z)

(z, z) = φ2,1 (z)φ2,1 (z)

The fusion rules for the Ising model can be determined using the
order-disorder symmetry and the symmetry

σ 7→ −σ

This yields

[σ] × [σ] = [1] + []

[σ] × [] = [σ]
[] × [][1]

We would like to determine the 3 point structure constants. From the

above, we have
Cσσ1 = C1 = 1
We need to determine Cσσ by dynamics.
To do this, we first examine the leading terms in the Operator
Product
1
σ(z)σ(0) = 1 (1 + · · · + |z|Cσσ (0, 0) + · · · ) (∗)
|z| 4

The m = 3 minimal model has a singular vector at level 2

4, 3 2
 
|χ1,2 i = L−2 − |h2,2 i
L −1
(|χ2,2 i = |χ1,2 i)

Which implies that

 
4 2
L−2 − L−1 hσ(z, z)φ1 (z1 , z 1 ) · · · φN (zN , z N )i = 0 (∗∗)
3
 63

We would like to determine the four-point correlation function

1
Gσσ
σσ = lim |z| hσ(z, z)σ(1, 1)σ(η, η)σ(0, 0)i
4
z,z→∞

the idea is to take (∗∗) with φ1 = φ2 = φ3 = σ and derive a differential

equation in η. We obtain the equation

d2
 
1 d 1
η(1 − η) 2 + ( − η) + fk (η) = 0
dη 2 dη 16

for k = 1, 261 . The matrix element can be expressed in terms of these 61

That is, since this is a second order
fk 62 as: differential equation (with singular
2 points at 0,1, and ∞) we have two
1 X linearly independent solutions.
Gσσ
σσ (η, η) = 1 ck` fk (η)f ` (η) 62
|η(1 − η)| 4 k,`=1 And the f k , which turn out to
satisfy the same differential equations.
The solutions fk are given by
 p  12
f1/2 = 1 ± 1 − η

However, there are multiple branch cuts involved in this definition.

Since we want Gσσ σσ to be single valued, there is only one choice we can
make for the constants ck` , which gives us
C  p p 
Gσσ
σσ (η, η) = 1 |1 + 1 − η| + |1 − 1 − η|
|η(1 − η)| 4
Expanding around η = 0, we get (to leading order)
 
C 1
Gσσ
σσ = 1 2 + |η| + · · · (∗ ∗ ∗)
|η| 4 2

We can also obtain an expression for Gσσ 63

σσ using the OPE , If we
63
Corresponding to the diagram
apply (∗), we find that
(1) σ σ (z)
1
Gσσ
σσ = lim |z| hσ(z, z)σ(1, 1)σ(η, η)σ(0, 0)i
4
z,z→∞
P =1·
1
hσ(z,z)σ(1,1)i
|z| 4
= lim 1 hσ(z, zσ(1, 1)1i+ (η) σ σ (0)
z,z→∞ |η| 4

+|η|Cσσ hσ(z, z)σ(1, 1)(0, 0)i + · · ·
1
2
 
|z| 4 1 Cσσ
= lim 1 1 + |η| 1 1 1 1
z,z→∞ |η| 4 |z − 1| 4 |z − 1| 8 + 8 −1 |z| 8 +1− 8
1 2


= 1 1 + |η|Cσσ
|η| 4
Comparing this result with (∗ ∗ ∗), we get that
1 1
c=
Cσσ =
2 2
As a result, we have solved the Ising model by consistency, and the
CFT we have found completely determines the dynamics of the sys-
tem.
 64

Remark. Remarkably enough, we did not at any point need a La-

grangian of our theory!

3.6 Minimal Model Characters

The character of a (potentially reducible) Verma module V (c, h) is
defined to be64 64
Where #(N ) is the number of
∞ ∞
states at level N .
c c
X Y
χV (c,h) = q h− 24 #(N )q N = q h− 24 (1 − q N )−1
N =0 N =1

This last equality means that we can represent the character in

terms of the Euler function
∞
Y
φ(q) = (1 − q N )
n=1

as c
q h− 24
χV (c,h) =
φ(q)
Turning our attention to the characters of a unitary minimal model
representation M (c, h), we first note that

hr,−s (m) − hr,s (m) = h−r,s (m) − hr,s (m) = r · s

Additionaly, we have a ‘symmetry of indices’ so that

hr,s (m) = h−r,−s (m) = hr+m,s+(m+1) (m)

We can use these to find out more about our singular vectors from
first intersections
Singular vectors Level Conf. Weight
|χr,s i r·s hr,s + r · s = h−r,s = hm+r,m+1−s
hr,s + (m − r)(m + 1 − s) = hm−r,s−m−1
|χm−r,m+1−s i (m − r)(m + 1 − s)
= hr,2(m+1)−s
The irreducible Verma module is therefore given by


M (c(m), hr,s (m)) = Vr,s / Vr+m,−s+m+1 ∪ Vr,2(m+1)−s

To compute the character, we want to subtract contributions from

each submodule, but this leads to the possibility of double counting
sub-submodules, and so on. We therefore need to look at the structure
of submodules:

Vr,s = V−r,−s = Vr+m,s+m+1

Vr+m,m+1−s = V−r−m,s−m−1 = Vm−r,s+(m+1)
Vr,2(m+1)−s = V−r,s−2(m+1) = V2m−r,s

so we have
 65

Vr+m,m+1−s ⊃ Vr+2m,s ∪ V2m−r,−s

≡
Vr,2(m+1)−s ⊃ Vr+m,s−(m+1) ∪ V3m−r,(m+1)−s

These are the only common submodules, but they might have non-
trivial intersections.
Considering higher submodule chain structure gives us65 65
Where each arrow in the diagram
goes from a module to a submodule.

Vr+m,m+1−s Vr+2m,s ··· V 1−(−1)k

r+km,(−1)k s+ 2 (m+1)

Vr,s

Vr,2(m+1)−s Vr,2(m+1)+s ··· V 1−(−1)k

r,k(m+1)+(−1)k s+ 2 (m+1)

The character of M (c(m), hr,s (m)) thus becomes

∞
q −c/2π
 h
1−(−1)k
X
(m) hr,s k k
χr,s (q) = q + (−1) · q r+km,(−1) s+ 2
φ(q)
k=1
!!
k 1−(−1)k
+ q r,k(m+1)+(−1) s+ 2 (m+1)

Or, if we define the functions

(m) q −1/24 X (2m(m+1)k+r(m+1)−ms)2 /(4m(m+1))

Kr,s = q
φ(q)
k∈Z

then
(m)
χ(m) (m)
r,s = Kr,s − Kr,−s
(m)
Remark. χr,s are the generating functions for the (physical) states
at level N 66 . 66
Shifted by
q hr,s −c/24
4 Modular Invariance
Up to this point, we have studied CFTs on the conformal plane. We
now aim to study CFT’s on the 2-torus.
Motivation:
• Consistent 2d CFTs describing critical phenomena should be lo-
cally independent of the 2d geometry. In particular, it should be
independent of the quantization scheme (on the torus T 2 ).

• In String Theory, CFTs on Riemann surfaces form part of perturba-

tive string theory. T 2 67 gives rise to a 1 loop correction term. 67
A genus 1 Riemann surface.
 66

4.1 Partition Function

We can consider our map on the punctured plane from section 2.3,
and extend it to the torus:

L
z 7→ ζ = 2π log(z)

identifying
boundaries

On T 2 , our local symmetries are still the Virasora generators Ln .

Our global symmetries, however, are generated by L0 and L0 68 . 68
So that the group of symmetries is
U (1) × U (1), the isometries of T 2 with
the flat metric.
Space-time Structure of the Torus
We can obtain the torus as a quotient of the complex plane by a
lattice generated by a pair of vectors ω1 and ω2 .
Diagrammatically, we take the quotient of:
 67

Imz

ω1

ω2

ω2
ω1
Rez
P

to get:

Translations along the cycle are given by

 
a
exp − (H · Imω2 − iP Reω2 )
|ω2 |
Recall (cf. section 2.7) that we calculated the energy momentum
tensor for the cylinder:
 2 X !
2π −2πn/Lζ c
Tcyl (ζ) = Ln e −
L 24
n∈Z

This allows us to compute the operators H and P :

Z  2 
1 2π c  2π  c 
H= L0 + L0 − = L0 + L0 −
2π ω1 L 12 L 12
Z  2
1 2π
2π

P = L0 + L0 = L0 + L0
2π ω1 L L
The partition function on T 2 is then given by

Z(ω1 , ω2 ) = Tr (exp (− (HImω2 − iP Reω2 )))

Using the modular parameter τ = τ1 + iτ2 where

Reω2 Reω2
τ1 = =
ω1 L
Imω2 Imω2
τ2 = =
ω1 L
 68

the partition function takes the form

   c   c 
Z(τ ) = Tr exp 2πi τ L0 − − τ L0 −
24 24
Or  
c c
Z(τ ) = Tr q L0 − 24 q L0 − 24

where q = exp(2πiτ ).

Observation. The partition function only depends on the modular

parameter τ (that is, on the shape of the torus/the twisted gluing
condition). No dependence on the size of the torus exists, which is
consistent with the conformal property of CFTs.

4.2 Modular Invariance

So far, we have singled out a particular choice of lattice (1-cycles) on
T 2 , however, we can think somewhat more generally. Consider the
picture:

Imz

ω1 + ω2

H ω2
ω1 ω1

ω2
ω1 + ω2
ω2
ω1
Rez
P

It is not hard to see that the lattice Λ generated by ω1 and ω2 will

also be generated by ω1 + ω2 and ω1 , and will be generated by ω1 + ω2
and ω2 . Consequently, many different choices of initial cycles will give
us the same torus.
In general, if (ω2 , ω1 ) defines the lattice Λ ⊂ C, then
" # " #" #
ω20 a b ω2
=
ω10 c d ω1

for " #
a b
∈ SL(2, Z)
c d
describes the same lattice Λ.
 69

Under such a transformation, the modular parameter transforms as

aτ + b
τ0 =
cτ + d

where69 " # 69
The group
a b P SL(2, Z) = SL(2, Z)/(Z/2)
∈ SL(2, Z)/(Z/2)
c d
is the modular group of the torus.
Partition functions of consistent CFTs must be modular invariant
(ie, independent of the quantization scheme) due to conformal invari-
ance. Therefore we require that
2 2
T 0 T
ZCF T (τ ) = ZCF T (τ )

" #
0 aτ +b a b
for τ = cτ +d where ∈ P SL(2, Z).
c d

Remarks. (i) P SL(2, Z) is generated by the transformations

1
T : τ 7→ τ + 1 S : τ 7→ −
τ

So Z(τ ) is modular invariant if and only if

Z(τ + 1) = Z(τ )
 
1
Z − = Z(τ )
τ

(ii) The shape of a torus is specified by a point in the upper half-

plane τ ∈ H = {C | Imτ > 0}

Imz

Rez
1

tori with the ‘same shape’ are therefore identified by the action
of P SL(2, Z) on τ . As a result the moduli space of ‘shapes of
tori’70 is given by 70
More explicitly, this is the moduli
Mcs ' H/P SL(2, Z) space of complex structures on the
surface of genus 1.
 70

To describe distinct points on Mcs , we can look at the fundamental

domain F0 ⊂ H, which is given explicitly by

Imτ > 0, 1 ≤ Reτ ≤ 0, |τ | ≥ 1
2
F0 =
Imτ > 0, 0 < Reτ < 1 , |τ | > 1
2

or, diagrammatically:
F0

T F0

SF0

4.3 Free Boson Parition Function

Recall (Exercise 5.1). The Free Boson CFT has a continuum of pri-
maries
Vα = : eiαφ(z,z) :
for α ∈ R. These satisfy the relations71 71
Where g = 1
4π
is the coupling
constant.
2
α
L0 |Vα i =
|Vα i
2
α2
L0 |Vα i = |Vα i
2
The free Boson CFT also has central charge

cBoson = 1

We now compute the partition function of the free Boson72 72

Where, as before,
q = e2πiτ
 c c

ZBoson (τ ) = Tr q L0 − 24 q L0 − 24
 
√ Z ∞  X α2 ~ 1 
= 2 dα 
 q 2 +|k|− 24 

|{z} −∞
conv. | {z } {~
k}
normal-
ization int. k1 <k2 <···
over | {z }
primaries hol. descendants
 
 
α2
+|~ 1 
X
k|− 24

×
 q 2


{~
 
k}
k1 <k2 <···
| {z }
anti-hol. desc.
 71

√
Z ∞ 
2 1 1
= 2 dαe−2piIm(τ )α q − 24 q − 24
−∞
| {z }
1
(Im(τ )) 2
∞ ∞
Y
−1 Y
−1
× 1 − qN × 1 − qN
N =1 N =1

So that, with some simplification73 ,we get 73

Notably, using Dedekind’s eta
function
1 ∞
ZBoson (τ ) = p 1 Y
η(τ ) = q 24 (1 − q N )
Im(τ )|η(τ )|2 N =1

to ease writing.
To see modular invariance, we consider the modular properties of
η(τ ).
2πi
T : η(τ + 1) = e 24 η(τ ).

√
S: η − τ1 = −iτ η(τ ) (This can be computed by Poisson resumma-
tion).

Returning to ZBoson , we see that

T : |η(τ )| 7→ |η(τ )|

and since Im(τ + 1) = Im(τ ), we see

ZBoson (τ + 1) = ZBoson (τ )

Similarly, we note
  2
1
S: η − = |τ ||η(τ )|2
τ

and    
1 τ Im(τ )
Im − = Im − 2 =
τ |τ | |τ |2
so that  
1
ZBoson − = ZBoson (τ )
τ
Taken together, this shows that the free Boson CFT is modular
invariant.

4.4 Interlude: Mode Expansions of Free Fermions

The mode expansion of the Boson and the Fermion CFT is given by
X
i∂z φ(z, z) = αn z −n−1
n∈Z
 72

The radial ordering is given by

1
(R (∂z φ(z, z)∂w φ(w, w))) ∼ −
(z − w)2

And
1
X
ψ(z) = ψn z −n− 2
n
1
where h = 2 and

n ∈ Z + 1
2 periodic boundary cond.
n ∈ Z anti-periodic boundary cond.

We call the periodic boundary conditions the NS (Nevea-Schwartz)-

sector and the anti-periodic boundary conditions the R (Raymond)
sector. We can think of these two different sectors in terms of mon-
odromy about the origin:

In the NS sector, we get ψ(e2πi z) = ψ(z), whereas in the R sector, we

get ψ(e2πi z) = −ψ(z). In more explicit geometric terms, we can think
of ψ as sections of a spin bundle74 on the (punctured) complex plane. 74
As it turns out, there are precisely
The commutation relations for the modes are given by two spin bundles on the annulus
C∗ ' S 1 up to isomorphism, corre-
I I  sponding to the NS and R sectors.
sz n dw m
[αn , αm ] = i2 z ∂z , w ∂w φ = nδn+m,0 The structure group for spin bundles
2πi 2πi is Spin(n), defined by the short exact
sequence
and, taking the anticommutator because of Fermi statistics.
0 → Z2 → Spin(n) → SO(n) → 0

{ψn , ψm } = δn+m,0 For S 1 , we consider SO(1) ∼ = 0, so

that Spin(1) ∼ = Z2 . The two principal
Z2 -bundles over S 1 (H 1 (C∗ , Z2 ) ∼
=
We can also consider twist fields, which change the boundary condi-
H 1 (S 1 , ZZ ) ∼
= Z2 ) are then given by:
tions of fermions from periodic to anti-periodic. 1 The map
As an application, let us look at the spin operator of the Ising S 1 × Z2 → S 1
model: (eiφ , σ) 7→ eiφ

7→
Ising Model c = 1 1: 1 1
2 Free Fermion CFT c = 2

2 The map
[σ] × [σ] = [1] + []
S1 → S1
(z, z) = φ(2,1) (z)φ(2,1) (z) i : ψψ :
eiφ 7→ e2iφ
h = h = 12

7→
 73

From [] × [σ] = [σ] we infer

σ(w, z)
(z, z)σ(w, w) ∼ 1 1
2(z − w) 2 (z − w) 2

because Cσσ = 12 . This implies that75 75

Where we denote by µ the disorder
operator of the Ising model, ie the
µ(w, w) dual field to σ.
ψ(z)σ(w, w) ∼ √ 1
2(z − w) 2
and, dually,
σ(w, w)
ψ(z)µ(w, w) ∼ √ 1
2(z − w) 2

Important for us: The spin operator σ(w, w) changes the sign of
ψ(z).
In the NS and R sectors, we can then compute 2-point correlators.
In the NS sector:
∞ ∞
1 1
X X
hψ(z)ψ(w)iN S = h0| ψn z −n− 2 ψm w−m− 2 |0i
n= 12 m=− 21
∞
1 1
X
{ψn ,ψm }=δn+m,0 = z −n− 2 wn− 2
n= 21
∞
1 X −k k 1
= z w =
z (z − w)
k=0

In the R sector (from the twist field):

hψ(z)ψ(w)iR = h0|σ(∞)ψ(z)ψ(w)σ(0)|0i
= hσ|ψ(z)ψ(w)|σi
∞ X
−∞
1 1
X
= hσ|ψn ψm |σiz −n− 2 w−m− 2
n=0 m=0
∞
1 1 X 1 1
= √ hσ| {ψ0 , ψ0 } |σi + z −n− 2 w−n− 2
zw 2 | {z } n=1
ψ0 2
1
pz pw

2 w+ z
=
z−w
Remarks. • The short distance behavior is unchanged between the
sectors76 76
Does this correspond to the local
triviality of (spin) bundles?
• The boundary conditions modify the behavior at 0 and ∞

• Similarly, we can introduce µ(z, z) as a twist field!

If we now examine the Fermionic zero mode ψ0 in the R-sector, we

have
(−1)F : fermionic mode counting operator
 74

which satisfies the anti-commutation relation

(−1)F , ψn = 0


for all n. This implies that (−1)F has eigenvalues ±1 for even/odd
numbers of fermionic creation and annihilation operators.
In particular:

(−1)F , ψ0 = 0

n√ √ o
2ψ0 , 2ψ0 = 2

(−1)F , (−1)F = 2


furnish a 2 dimensional Clifford algebra:

{Γµ , Γν } = 2g µν

where " #
µν 1
g =
1
Furthermore
[L0 , ψ0 ] = [L0 , (−1)F ] = 0
so the R-ground state must be a representation of the 2 dimensional
Clifford algebra. The smallest representation is |±iR with

(−1)± |±i = ±|±iR

and
√
( 2ψ0 ) = |±iR

Remarks. • We identify:

|+iR = σ(0, 0)|0i

|−iR = µ(0, 0)|0i

• Considering ψ(z) and ψ(z) together, there are two operators (−1)FL
and (−1)FR .

We consider: (−1)F := (−1)FL +FR and a two state subset of vac-

uum states (the non-chiral subsector ) |±iR with

(−1)FL +FR |±iR = ±|±iR

1
with h = h = 16

If instead of the punctured plane we work on the cylinder, we must

recall that primaries transform as
 −h  −h
∂φ ∂φ
φ(z, z) 7→ φ(z, z)
∂z ∂z
 75

The mapping to the cylinder is given by

l
z 7→ ζ(z) = log(z)
2π
∂ζ L 1 2π
so ∂z = 2π z and e L ζ So, for Bosons

2π X 2πζ
(i∂φ)cyl (ζ) = αn e L n
L
n∈Z

and for Fermions

r
2π X 2πζ
ψcyl (ζ) = ψn e− L n
L
ψn

where 
Z + 1
2 NS sector
n∈
Z R sector

We can summarize the boundary conditions for fermions77 77

(L0 )cyl = L0 − c
24
.

Conformal Plane Cylinder

Periodic bound-
NS sector: Anti-periodic bound-
ary conditions.
half-integral ary conditions.
L0 |0i = 0 1
modes (L0 )cyl |0i = − 48 |0i
(−1)± |0i = 0
Anti-periodic
boundary Periodic bound-
R sector: ψ0
conditions. ary conditions.
zero mode 1 1
L0 |±iR = 16 |±iR (L0 )cyl |±iR = 24 |±iR
±
(−1) |±iR = ±|±iR

4.5 Partition Function of the Free Fermion

For a torus of shape τ

Imz

Rez
1

we have 4 choices of boundary conditions:

 76

P A R sector
along spa-
P P tial slices

P A NS sector
along spa-
A A tial slices

Observation. Except for P ,P , boundary conditions are not modular

invariant

P T A

A A

Our expectation, therefore, is that a modular invariant partition func-

tion with anti-periodic boundary conditions must involve all sectors
involving anti-periodic boundary conditions.

In the operator formalism, we look at hψ(z)Xi where X is some

product of fields inserted at various positions.
In this case hψ(z)Xi 6= 0 only if X is
fermionic. We can then transport ψ(z) past
all operators X. In anti-periodic boundary
operators
spatial conditions
slice
ψ(z + τ ) = −ψ(z)
ψ(z)
In periodic boundary conditions, the insertion of (−1)F operator yields
an additional minus sign

hψ(z)(−1)F Xi =
6 0

for X bosonic.
ψ(z + τ ) = ψ(z)

Thus, in the holomorphic sector, we get, in the spatial R sector

1 1
P = √ q − 48 trR ((−1)F q L0 )
2
P

1 1
A = √ q − 48 trR q L0
2
P
 77

And, in the spatial NS sector

1
P = q − 48 trN S ((−1)F q L0 )
A

1
A = q − 48 trN S q L0
A
Where the √12 in the R sector is a convenient normalization factor
related to picking only two out of 4 ground state values.
In the NS sector: [0] + [h = 12 ] ' [φ1,1 ] + [φ2,1 ] conformal families.

L0 eigenvalue state
0 |0i primary of conformal weight h = 0
1
2 ψ− 12 |0i primary of conformal weight h = 12
3
2 ψ− 32 |0i desc. of [ 21 ] level 1
2 ψ− 23 ψ− 12 |0i desc. of [0] level 2
5
2 ψ− 52 |0i desc. of [ 21 ] level 2
.. .. ..
. . .
4 ψ− 25 ψ− 32 |0i desc. of [0] level 4 (two states)

so we have
3 5 7
trN S q L0 = 1 + q + q 2 + q 2 + q 2 + q 2 + 2q 4 + · · ·
∞  
1
Y
= 1 + q n+ 2
n=0

and
1 3 5 7
trN S (−1)F q L0 = 1 − q 2 − q 2 + q 2 − q 2 + q 3 − q 2 + 2q 4 + · · ·

∞  
1
Y
= 1 − q n+ 2
n=0

Remarks. • The structure of states is consistent with the singular

vector structure:

[φ1,1 ]: state at level 1 singular vector

[φ2,1 ]: singular vector at level 2, so there is only one (rather than
two) state at level 2.

• Projections to m = 3 minimal model characters

 
1
− 48 1 f L0
χ[φ1,1 ] = q trN S (1 + (−1) )q
2
 
1
− 48 1 F L0
χ[φ2,1 ] = q trN S (1 − (−1) )q
2
 78

In the RR sector, we have [σ] and [µ] dual conformal families

[φ1,2 ] ' [φ2,2 ]

L0 eigenvalue State
1
16 + 0 |±iR
1
16 + 1 ψ−1 |±iR
1
16 + 2 ψ−2 |±iR

so we have
∞
Y
trR q L0 =
n=0

= 2 + 2q + · · ·
∞
Y
trR (−1)q L0 = (1 − q n ) = 0
n=0

because of the fermionic zero mode.

So, in summary, we find78 : 78
Where θi are the Jacobi theta
functions:
s
∞ 1 2
1 θ1 (τ )
X
1 θ2 (τ ) = q (n+ 2 ) /2
Y
P = √ q − 24 (1 − q n ) = =0
2 n=0
η(τ ) n∈Z
2
P
X
θ3 (τ ) = qn /2

∞
s n∈Z
1 1
Y θ2 (τ ) 2
= √ q − 24 (1 + q n ) =
X
A θ4 (τ ) = (−1)n q n /2
2 n=0
η(τ ) n∈Z
P Which display the modular proper-
s
∞ ties
1
− 48
Y θ3 (τ )
P =q (1 − q n ) = θ2 (τ + 1) = eiπ/4 θ2 (τ )
η(τ )
n= 12 1 √
A θ2 (− ) = −iτ θ4 (τ )
s τ
∞
1
− 48
Y θ4 (τ ) θ3 (τ + 1) = θ4 (τ )
A =q (1 + q n ) = √
η(τ ) 1
θ3 (− ) = −iτ θ3 (τ )
n= 12
A τ
θ4 (τ + 1) = θ3 (τ )
We can then write down the modular invariant partition function of 1 √
θ4 (− ) = −iτ θ2 (τ )
the fermion: τ

2 2 2 2

Zf erm (τ ) = P + A + P + A
P P A A

Or
exchange wrt S

θ2 (τ ) θ3 (τ ) θ4 (τ )
Zf erm (τ ) = + +
η(τ ) η(τ ) η(τ )
wrt T

is modular invariant.
 79

Remarks. • The normalization projection onto ‘physical’ zero mode

ψ0 + ψ 0 in RR rector means we choose 2 out of 4 values, so as to
arrive at the √12 normalization factor.

• Ising model partition function has only [σ] family and no [µ] family,
which implies
Zf erm (τ ) = 2ZIsing (τ )
so the fermion description induces the ‘dual’ Ising model descrip-
tion.

5 Applications

5.1 Affine Kac-Moody algebras and WZW models

Let g be a simple Lie algebra

a) A Lie algebra g is a vector space equipped with an anti-symmetric

bilinear pairing
[·, ·] : g × g → g
called the Lie bracket satisfying the Jacobi identity. That is, such
that for any x, y, z ∈ g

[[x, y], z] + [[z, x], y] + [[y, z], x] = 0

b) a simple Lie algebra g is one such that there are no proper subsets
S ⊂ g such that
[S, g] ⊂ S
These have be classified, and are precisely the Lie algebras An ,Bn ,
Cn , Dn ,E6 , E7 , E8 ,F4 , and G2 .79 79
The dual Coexter numbers as-
sociated with these algebras are,
c) Let J a be generators of g, then respectively:
X g
[J a , J b ] = if abc J c An n+1
c Bn 2n − 1
Cn n+1
where the f abc are the real structure constants of g. The dimension Dn 2n − 2
of g is then given by E6 12
E7 18
E8 30
dim g = #(generators) F4 9
G2 4
d) The (non-degenerate) Killing Form for g is given by
1
K(x, y) = Tr (ad(x) · ad(y))
2g
Where g is the dual Coexter number and

ad(x) : y 7→ [x, y]
 80

From g, we can consider Affine Kac-Moody algebras:

i) The loop algebra of g is the vector space of maps

Map(S 1 → g)

Since these functions are periodic, we can consider the Fourier

decomposition of J(θ) ∈ Map(S 1 → g)
X
J(θ) = einθ Xn , Xn ∈ g
n

Identifying eiθ with t, we get an isomorphism given by the Fourier

decomposition

Map(S 1 → g) ↔ g ⊗ C[[t, t−1 ]]

The algebra on the right has generators

J a ⊗ tn =: Jna

for n ∈ Z. Multiplying two periodic functions according to the Lie

bracket yields another periodic function

[J a ⊗ tn , J b ⊗ tm ] = if abc J c tn+m
[Jna , Jm
b
] = if abc Jn+m
c

ii) There is a unique central extension of the loop algebra

X
[Jna , Jm
b
]= if abc Jn+m
c
+ K̂n K(J a , J b ) δn+m,0
c
| {z }
δ a,b

which satisfies the commutation relations

X
[Jna , Jm
b
]= if abc Jn+m
c
+ K̂nδ a,b δn+m,0
c

[Jna , K̂] = 0

This is the affine Kac-Moody algebra of g.

Associated to an affine Kac-Moody algebra, we have a current

algebra
kδ ab X J c (w)
R(J a (z)J b (w)) ∼ 2
+ if abc
(z − w) c
z−w
where X
J a (z) = z −n−1 Jna
n∈Z

In modes, we recover
X
[Jna , Jm
b
] := if abc Jn+m
c
+ K̂nδa,b δn+m,0
c
 81

Suyawara Energy-Momentum Tensor:

J a (z) are primary fields with respect to the energy-momentum
tensor X
T (z) = γ : J a Ja : (z)
a

With coupling constant γ, which is fixed by the OPE

c/2 2T (w) ∂T (w)
R(T (z)T (w)) ∼ + +
(z − w)4 (z − w)2 (z − w)
So one obtains

1 k · dim g
γ= ; c(gk ) =
2(k + g) k+g

Moreover:
 
b a
du X  J (u)Jb (z)J (w) 
I
R(T (z)J a (w)) = γ
2πi u−z
 
z b

J a (w) ∂w J a (w)
 
= · · · = 2γ(k + g) 2
+
(z − w) z−w

Thus, J a (w) are chiral primary fields of conformal weight h = 1.

The mode expansion yields
X
T (z) = Ln z −n−2
n

where80 80
We define
1 XX
a (
Ln = : Jm Ja,n−m : Jna Ja` n<`
2(k + g) a : Jna Ja,` : =
n∈Z J`a Ja,n m≥`
So we see that
a a
[Ln , Jm ] = −mJn+m
In particular
[L0 , Jna ] = −nJna
We can then consider primary states. The representation theory of
our affine Kac-Moody algebra yields affine heighest weight states |λi:

• For a simple Lie algebra g, we have generators E ±α 81 and H i 82 for 81

The Raising and lowering genera-
i = 1, . . . , r where r = rank(g). They satisfy commutation relations tors.
82
The Cartan generators.

[H i , H j ] = 0 [H i , E ±α ] = ±αE ±α

where α ∈ ∆+ is a positive root.

As an example, we can consider SU (2), which has Cartan generator
J 3 and raising/lowering generators J ± satisfying

[J3 , J± ] = ±J± , [J+ , J− ] = +2J3

 82

Recall that unitary representations are labeled by j, mi, 2j ∈ Z≥0 ,

and m ∈ {−j, −j + 1, . . . , j − 1, j}. The highest weight state is then
|j, ji.

• Affine Kac-Moody algebras have generators En±α and Hni . For a

Wess–Zumino–Witten (WZW) highest weight state 83 83
Also known as a WZW primary.

En±α |λi = Hni |λi = E0α |λi = 0

Note that |λi will also be a Virasora primary state, since, for n ≥ 0
X
a
Ln |λi = : Jm Ja,n−m : |λi = 0
m

That said, a Virasora primary need not be a WZW primary.

Example (SU (2)k ). We have generators Jn± and Jn3 with

± ±
[Jn3 , Jm ] = ±Jn+m
[Jn3 , Jm
3
] = knδn+m,0
−
[Jn+ , Jm 3
] = 2Jn+m + knδn+m,0
−
Observe that J˜+ := J+1
+
, J˜− := J−1 and J˜3 := J03 + k
2 form a
SU (2)-subalgebra. The highest weight states is

|j, ji

which has WZW descendants

|j, j − 1i, . . . , |j, −ji

The norm of the final descendant is

+ −
hj, −j| J+1 J−1 |j, −ji = (−2j + k)hj, −j|j, −ji
| {z }
2J˜3

So we derive a necessary condition for our representation to be

unitary:
k
k≤ , k ∈ Z≥0
2
which must hold lest we have negative norm states.

To compute the conformal weight of the highest weight state, we

note first that g(SU (2)) = 2, so that
1  ~ 2 
L0 = (J0 ) + . . .
k+2
and hence
j(j + 1)
h|j,ji =
h+2
 83

and
3
c=
k+2
According to our constraint derived in the example, the unitary
representations of SU (2)k can be labeled by [j]W ZW for j = 1, 21 , . . . , k2

Example (SU (2)1 ). We have two unitary representations: [0]W ZW

and [ 21 ]W ZW ], both with c = 184 . All states are generated by J˜− and 84
For those familiar with string
J0+ . theory, this can be associated with the
Boson on the self-dual circle. It arises
from T-duality on the bosonic string.

m = −2 m = −1 m=0 m=1 SU (2)

[0]W ZW
m̃ = − 32 m̃ = − 21 m̃ = 21 m̃ = 23 rep.

L0

0 |0i spin0
1
in 2
sp ˜−
^
J ^1 + spin
^3
spin 2 2
^1
spin 2

1 |−i | + −i | + +−i spin1

J0+

0 | − + + −i spin1 +
2 | − +−i | + + − +−i
sing. vect | + − + −i spin0

| − + + − + −i
| − − + +−i | + − − + + −i
| + − − + + −i 2spin1 +
3
| − + − +−i | + − + − + −i
| + − + − + −i spin0

Virasora Conformal Families

Note. The commutation relation

[Ln , J03 ] = 0

ensures that the Virasora generators do not change the m-quantum

numbers.85 85
I have not yet included the
[ 21 ]W ZW chart.
Remarks. • WZW models are unitary, which can be proven by:

a) Pure representation theoretic arguments on gk .

 84

b) WZW models admit a non-linear σ-model description with

manifest unitary action86 86
For g : S 2 → G, the compact Lie
Z group associated to G.
k
d2 x K(g −1 ∂ µ g, g∂µ g) + 2πkS W Z (g)
 
SW ZW = −
8π S 2

where, for an extension of g g̃ : B 3 → G87 , the correction term is 87

Such an extension always exists
given by since π2 (G) vanishes for G a compact,
simply-connected Lie group.
Z
1
S W Z (g) = − K g̃ −1 ∂µ g̃, g̃ −1 ∂ν g̃, g̃∂ρ g̃ dxµ ∧dxν ∧dxρ
 

2
48π B 3

• A WZW model has a finite number of conformal WZW families,

but an infinite number of Virasora conformal families.

• Roughly speaking, a CFT is called a Rational Conformal Field The-

ory (RCFT) if it has a finite number of conformal families with
respect to some current algebra (eg, minimal models, WZW mod-
els).

5.2 Coset Constructions

If we consider the affine Lie algebra

gk ⊕ g`

we find that gk+` is a diagonal affine Lie subalgebra, that is

D
Jna = Jn(1) a + Jn(2) a
D a D b
Jn , Jm = if abc D Jn+m
c
+ (k + `)nδ ab δn+m,0

This allows us to construct the diagonal coset CFT.

ĝk ⊕ ĝ` /ĝk+`

is generated by the Virasora generators

Lcoset
n := Lgnk + Lgn` − Lgnk+`

which implies that the central charge is

 
k ` k+`
ccoset − dim g + −
k+g k+g k+`+g

by construction, such coset CFTs are unitary because the gk and g`

WZW models are unitary.
Minimal model CFTs arise naturally in this framework, via

SU (2)k ⊕ SU (2)`
SU (2)k+`
 85

which has
 
k 1 k+1
c=3 + −
k+` 3 k+3
6
=1−
(k + 1)(k + 3)

We can then identify these with m = 2, 3, . . . minimal models via

m = k + 2.
 86

References
[1] G. Mack and A. Salam, “Finite component field representations
of the conformal group,” Annals Phys., vol. 53, pp. 174–202,
1969.

[2] S. Garcia, “Hidden invariance of the free classical particle,” Sub-

mitted to: Amer. J. Phys., 1993.

[3] C. A. Weibel, An introduction to homological algebra, vol. 38

of Cambridge Studies in Advanced Mathematics. Cambridge
University Press, Cambridge, 1994.

[4] J. Hilgert and K.-H. Neeb, Lie-Gruppen und Lie-Algebren (Ger-

man Edition). Vieweg+Teubner Verlag, 1991 ed., 1 1991.

[5] M. Schottenloher, “A mathematical introduction to conformal

field theory,” Lect. Notes Phys., vol. 759, pp. 1–237, 2008.

[6] P. Di Francesco, P. Mathieu, and D. Senechal, Conformal Field

Theory. Graduate Texts in Contemporary Physics, New York:
Springer-Verlag, 1997.

[7] E. Perlmutter, “Virasoro conformal blocks in closed form,” JHEP,

vol. 08, p. 088, 2015.

[8] D. Friedan, S. H. Shenker, and Z.-a. Qiu, “Details of the Nonuni-

tarity Proof for Highest Weight Representations of the Virasoro
Algebra,” Commun. Math. Phys., vol. 107, p. 535, 1986.

[9] P. H. Ginsparg, “Applied Conformal Field Theory,” in Les

Houches Summer School in Theoretical Physics: Fields, Strings,
Critical Phenomena Les Houches, France, June 28-August 5,
1988, 1988.

[10] W. Lerche, D. J. Smit, and N. P. Warner, “Differential equations

for periods and flat coordinates in two-dimensional topological
matter theories,” Nucl. Phys., vol. B372, pp. 87–112, 1992.

[11] D. Friedan, Z. A. Qiu, and S. H. Shenker, “Conformal Invariance

and Critical Exponents in Two Dimensions,” 1986.

[12] R. Blumenhagen and E. Plauschinn, “Introduction to conformal

field theory,” Lect. Notes Phys., vol. 779, pp. 1–256, 2009.

[13] J. Fuchs, “Lectures on conformal field theory and Kac-Moody

algebras,” 1997. [Lect. Notes Phys.498,1(1997)].