3 Conformal field theories in 2D

The Polyakov formalism of string theory is based on a dynamical metric coupled to a Weyl
invariant field theory on the two dimensional worldsheet. As we will see in the next chapter,
upon gauge fixing, the worldsheet theory can be described as a conformal field theory on a
Riemann surface with zero central charge and a BRST symmetry. In this chapter we will
introduce the defining properties and basic examples of 2D conformal field theories that will
serve as building blocks for the worldsheet theory.

3.1 Correlation functions and operator product expansion

We begin by considering a Poincaré invariant quantum field theory in d-dimensional Minkowskian

spacetime, parameterized by the coordinates x0 , · · · , xd 1 . Among the local field operators,
there is a stress-energy tensor operator Tab (x) that is a symmetric, conserved tensor, namely

Tab = Tba , @a T a b = 0, (3.1)

such that the corresponding conserved Noether charge

Z
a
P = T 0a dx1 ^ · · · ^ dxd 1 , (3.2)
S

defined on any spatial slice S of the spacetime, is the energy-momentum vector. In particular
P 0 ⌘ H is the Hamiltonian. It also follows from (3.1) that

@a (xb T ac xc T ab ) = 0, (3.3)

and the conserved charges associated with the current Jabc ⌘ xb Ta c xc Ta b are the angular
momentum and Lorentz boost charges M bc . Together, the energy-momentum P a and angular
momentum/boost charges M ab generate the Poincaré symmetry.
We will assume that there is a Poincaré invariant vacuum state |0i, and write the vacuum
correlation functions of local operators as

hO1 (x1 ) · · · On (xn )i ⌘ h0|O(x1 ) · · · O(xn )|0i. (3.4)

We will also assume microcausality, meaning that any pair of local (bosonic) operators Oi (x)
and Oj (y) at spacetime points x = (x0 , ~x), y = (y 0 , ~y ) commute provided that x and y are
spacelike separated, namely

[Oi (x), Oj (y)] = 0, (x y)2 ⌘ (~x ~y )2 (x0 y 0 )2 > 0. (3.5)

20
In particular, if all of the xi ’s appearing in (3.4) are spacelike separated, then the correlation
function is independent of the ordering of the operators. Starting from such a configuration,
we can analytically continue the time coordinates x0i ’s to imaginary values i⌧i , and obtain
the Euclidean correlation function
⌦ ↵
O1 (xE E
1 ) · · · On (xn ) , xE xi ) 2 R d .
i ⌘ (⌧i , ~ (3.6)

Such an analytic continuation is possible because we can write (3.4) as

X
hO1 (x1 ) · · · On (xn )i = h0|O(x1 ) · · · On 1 (xn 1 )|↵ih↵|On (xn )|0i
↵
X (3.7)
iP↵ ·xn
= h0|O(x1 ) · · · On 1 (xn 1 )|↵ie h↵|On (0)|0i
↵

where |↵i is a complete orthonormal basis of the Hilbert space of states, with energy-
momentum vector P↵ = (E↵ , P~↵ ). Provided that the energy E↵ is bounded from below, the
sum over states converges and is analytic in x0n on the upper half complex x0n -plane. If the
xi ’s are spacelike separated to begin with, we can also analytically continue x0n to the lower
half complex plane, by moving On (xn ) to the left of the vacuum correlation function, and
repeating the same argument by inserting a complete basis of states between h0|On (xn ) and
O1 (x1 ) · · · On 1 (xn 1 )|0i. Therefore, we can analytically continue the correlation function
in x0n to the complex plane, away from the real axis, without encountering any singularities.
We can repeat the same procedure for the other xi ’s as well.
The resulting Euclidean correlation function (3.6) will be a real analytic function in
the xE i ’s provided that the latter are separated points in R . Starting from the Euclidean
d

correlation function, we can also go back to the Lorentzian correlator by analytic continuation
in ⌧i . We expect singularities when a pair of points xi and xj become null-separated, and
the analytic continuation must be performed in a way that avoids the singularities. By
inspecting for instance
X
h· · · Oi (xi )Oj (xj ) · · · i = h· · · Oi (0)|↵ieiP↵ ·(xi xj ) h↵|Oj (0) · · · i, (3.8)
↵

we see that analyticity would be preserved if Im(x0i x0j ) < 0 is maintained in the an-
alytic continuation. Thus, we can recover the Lorentzian correlator by analytic contin-
uation, maintaining separation of the points in imaginary (or Euclidean) time, so that
(xi xj )2 = (~xi ~xj )2 (x0i x0j )2 is never zero. If xi and xj are to become timelike
separated, i.e. (xi xj )2 goes from a positive real value to a negative real value, the operator
ordering between Oi and Oj in the resulting Lorentzian correlator is specified by whether
(xi xj )2 goes around zero from above or below the real axis in the analytic continuation
procedure.

21
 In the Euclidean signature, rather than working with constant time slices, we may con-
sider any (d 1)-dimensional oriented hypersurface S in Rd , and consider a certain space
HS of states on S. In a path integral formulation, such states may be specified by wave
functionals on S. Generally, there isn’t a natural inner product defined on HS , and so HS is
not quite a Hilbert space (unless S is time-reflection symmetric). Rather, we have a bilinear
pairing between states in HS and those in HS , where S is S with the opposite orientation.
For instance, we can take S to be a (d 1)-dimensional sphere centered at the origin.
One way to create a state in HS is to consider the wave functional produced by the path
integral on the ball B enclosed by S, with a set of local operators Oi (xi ) inserted within B.
Q
We will denote this state by | i Oi iS . Likewise, a set of operators Oj0 (x0j ) in the complement
Q Q Q
of B (denoted by B 0 ) defines a state S h j Oj0 | 2 HS . The pairing S h j Oj0 | i Oi iS is given
by the Euclidean correlation function of the Oi ’s and Oj0 ’s. HS should be complete with
respect to the pairing in the following sense: if a sequence of states | n i 2 HS , n 1, has
Q
the property that the sequence S h j Oj0 | n i converges for any set of operators Oj0 in B 0 , then
Q Q
| n i converges to a state | 1 i 2 HS such that limn!1 S h j Oj0 | n i = S h j Oj0 | 1 i.
Let H" be the space of states in HS created by a set of local operators in a small ball
B" (0) of radius " centered at the origin. We will now argue that H" is in fact dense in HS .
Let us pick a point y 2 B 0 , and consider the ball B" (y) of radius " centered at y. A set
of operators in B" (0) can be related to a set of operators in B" (y) by mirror reflection with
respect to a hyperplane in between 0 and y. Given a basis of states | ↵ i 2 H" , let | ↵ i 2 HS
be the states created by the mirror reflection of those that create | ↵ i. A ↵ ⌘ h | ↵ i is a
positive definite matrix by reflection positivity. Now for any state | i 2 HS ,
X
| 0i ⌘ | i h | i(A 1 )↵ | ↵ i (3.9)
↵,

has vanishing pairing with all | ↵ i. In other words, | i has vanishing correlator with any
set of local operators in B" (y). By analytic continuation of correlators, | 0 i has vanishing
pairing with all states in HS and therefore must be the zero state vector. This shows that
| i is a limit of vectors in H" .
By taking " arbitrarily small, we conclude that the space of states spanned by |Oi (0)iS
for all local operators Oi at the origin should be dense in HS . In particular, given a pair of
local operators Oi (xi ) and Oj (xj ) inside B, the state |Oi (xi )Oj (xj )iS should be expressible
as X
|Oi (xi )Oj (xj )iS = Cijk (xi , xj )|Ok (0)iS , (3.10)
k

where Cijk (xi , xj )

are real analytic functions in xi and xj , away from the locus xi = xj . This
is known as the operator product expansion (OPE). (3.10) can be viewed as an operator

22
equation X
Oi (xi )Oj (xj ) = Cijk (xi , xj )Ok (0), (3.11)
k

in the sense that (3.11) is expected to hold when inserted into a (Euclidean) correlation
function, with a convergent sum over correlators involving Ok on the RHS, provided that all
other local operators in the correlation function lie outside of B.
We can of course also shift the origin of the ball B, and write the OPE of a pair of
operators around any given point y 2 Rd ,
X
Oi (xi )Oj (xj ) = Cijk (xi y, xj y)Ok (y), (3.12)
k

provided that the minimal sphere centered at y that encloses xi and xj does not enclose
other local operators in a given correlation function.

3.2 Conformal symmetry

The trace of the stress-energy tensor, T a a , is a scalar operator. Typically, one expresses it
as a linear combination of a basis of scalar operators OI ,
X
T aa = I OI , (3.13)
I

where the coefficients I are known as beta functions.

A conformal field theory (CFT) is a quantum field theory whose stress-energy tensor Tab
is traceless, namely
T a a = 0. (3.14)
It follows from (3.14) that there are new conservation laws

@a (xb T a b ) = 0, @a (T ab x2 2T a c xb xc ) = 0. (3.15)

The symmetries generated by a conserved current of the form Ja (x) = Tab (x)v b (x) obey
the same algebra as those of the vector fields v a (x)@a , which generate the infinitesimal
di↵eomorphism
xa 7! x0a = xa + ✏v a (x). (3.16)
In this sense, we can organize the symmetries generated by the conserved currents appearing
in (3.1), (3.3), (3.15) according to the di↵eomorphisms generated by the vector fields

@a , xa @ b xb @ a , xa @ a , x2 @ a 2xa xb @b . (3.17)

23
(3.17) generate the d-dimensional conformal group, which is isomorphic to SO(d, 2) for
conformal symmetry in d-dimensional Minkowskian spacetime, or SO(d + 1, 1) for conformal
symmetry in Euclidean spacetime.
In the rest of this chapter we will focus on conformal field theories in 2D (having in
mind application to the worldsheet theory of the string), where the conformal group is
automatically enhanced to an infinite dimensional one. This is most easily seen by working
in lightcone coordinates ( + , ), in which the traceless stress-energy tensor obeys

@ T++ = @+ T = 0, T+ = T + = 0. (3.18)

Consequently, we have an infinite set of conservation laws

+
@ (v( )T++ ) = 0, @+ (e
v( )T ) = 0, (3.19)

for any functions v( + ) and ve( ). In Euclidean signature, ( + , ) are replaced by complex
coordinates (z, z̄). The non-vanishing components of the stress-energy tensor are

T (z) ⌘ Tzz (z), Te(z̄) ⌘ Tz̄z̄ (z̄), (3.20)

where we have indicated the conservation laws by declaring that T (z) is holomorphic and
Te(z̄) is anti-holomorphic.
Let us note that the holomorphy of T (z), like the conservation equations for any symmetry
currents in a quantum field theory, should be understood as an operator equation, in the
¯ vanishes,
sense that any correlation function of local operators that involve @T
⌦ ↵
¯ (z)O1 (z1 , z̄1 ) · · · On (zn , z̄n ) = 0,
@T (3.21)

provided that the points z1 , · · · , zn are separated from z. This means that the correlation
function
hT (z)O1 (z1 , z̄1 ) · · · On (zn , z̄n )i (3.22)
is a holomorphic function in z, away from z1 , · · · , zn . Generally, (3.22) will have singularities
when z coincides with one of the zi ’s.
As a simple example, a linear conformal transformation amounts to a combination of
translation, rotation, and dilation, which may be characterized by the linear holomorphic
di↵eomorphism
z 7! z 0 = z + u, z̄ 7! z̄ 0 = ¯ z̄ + ū, (3.23)
where , u are complex constants, with nonzero. If we set u = 0, then (3.23) would fix the
origin, and the associated dilation/rotation maps an operator at the origin, O(0), to another

24
operator at the origin O0 (0). Typically, it is possible to find a basis of operators O on which
dilation and rotation act diagonally, namely
h¯ e
O0 (0) = h
O(0), (3.24)

for a pair of constants h and e h. We will refer to h and eh as the holomorphic and anti-
holomorphic (or left and right) conformal weights of O. For nonzero u, the linear conformal
transformation takes O to O0 , such that
h¯ e
O0 (u) = h
O(0) (3.25)

There is a conformal transformation that maps the punctured complex plane C⇥ =

C {0} to the Euclidean cylinder, parameterized by w = + i⌧ , ⇠ + 2⇡, via the
holomorphic di↵eomorphism
z = e iw . (3.26)
Heuristically, we then expect that a CFT on the plane, possibly with a singularity at the
origin, to be equivalent to the same CFT on the Euclidean cylinder. The singularity at the
origin can be characterized by an insertion of a local operator on the plane. The Euclidean
cylinder is related to the Lorentzian cylinder by analytic continuation in ⌧ , and characterizes
time evolution of the CFT defined on a spacial circle.
This leads to the state/operator correspondence, that is, there is a canonical 1-1 corre-
spondence between the space of local operators in a CFT at a given point (here taken to
be the origin of the complex plane) to the space H of states of the CFT on a circle of unit
radius. Given a local operator O(0), we will write the corresponding state as |Oi.
The dilation and rotation symmetry (3.23) on the plane corresponds to time and space
translation on the cylinder under the state/operator map. Thus, we expect that an operator
O(0) of conformal weight (h, eh) to map to a state |Oi on the cylinder with energy h + e
h and
e
spatial momentum h h. We will see in section 3.5 that this is correct up to a shift of the
ground state energy-momentum that is fixed by the conformal symmetry.
We may equivalently think of the state |Oi as defined by a wave functional on the
unit circle S : |z| = 1 in the plane, obtained by performing the path integral on the disc
|z|  1 with the insertion of O(0), as described in section 3.1. The conformal symmetry
corresponding to the inversion map z 0 = 1/z gives a natural identification between HS and
HS , known as BPZ conjugation, thereby defining a bilinear inner product on HS . The
Hermitian inner product on HS is related by a further complex conjugation on one of the
two states in the bilinear pairing.
The operator product expansion in a CFT can be organized according to the conformal
weights. Namely, given a basis of local operators Oi (z, z̄) of weight (hi , e
hi ), we can write the

25
OPE as X
hi hj e
hk e
hi e
Oi (z, z̄)Oj (0) = Cijk z hk z̄ hj
Ok (0), (3.27)
k

where Cijk are constants. When inserted into a correlation function, the RHS of (3.27) as
a power series in z, z̄ is expected to have a radius of convergence at least R if there are no
other local operators within the disc centered at 0 of radius R. In writing (3.27) we have
only made use of the dilation and rotation symmetry. We will see in section 3.8 that the
full conformal symmetry leads to a much more refined version of the OPE, with far stronger
convergence properties.
The conserved currents appearing in (3.19) can be written in the Euclidean signature as
holomorphic and anti-holomorphic currents

j(z) = v(z)T (z), j(z̄) = ve(z̄)Te(z̄).

e (3.28)

They generate the infinitesimal conformal transformations that correspond to the infinitesi-
mal holomorphic di↵eomorphisms

z 7! z 0 = z + ✏v(z), z̄ 7! z̄ 0 = z̄ + ✏v̄(z̄), (3.29)

by acting on a local operator O(z, z̄) with the conserved charge

I 
b dw dw̄ e
Q= j(w) j(w̄) , (3.30)
C 2⇡i 2⇡i

where C is a counterclockwise contour that encircles z. That is to say, the infinitesimal

conformal transformation maps the operator O(z, z̄) to

O0 (z, z̄) = O(z, z̄) + O(z, z̄),

I 
dw dw̄ (3.31)
b
O(z, z̄) = ✏QO(z, z̄) = ✏ v(w)T (w) ve(w̄)Te(w̄) O(z, z̄).
C 2⇡i 2⇡i

The contour integral on the RHS can also be expressed in terms of the singular terms in the
OPE of the stress-energy tensor with O, namely
h i
O(z, z̄) = ✏ Resw!z v(w)T (w)O(z, z̄) + Resw̄!z̄ ve(w̄)Te(w̄)O(z, z̄) (3.32)

If O transforms under an infinitesimal linear conformal transformation v(w) = ↵w + ,

¯ w̄ + ¯ with definite conformal weights (h, e
ve(w̄) = ↵ h), as in (3.25), then we know that

O(z, z̄) = ✏(h↵ + e

h¯↵ + @ + ¯@)O(z,
¯ z̄). (3.33)

26
It follows that the OPE of T, Te with O must take the form (after a overall translation and
relabeling of coordinates)
h 1
T (z)O(0) = · · · + O(0) + @O(0) + · · · ,
z2 z
e (3.34)
h 1¯
Te(z̄)O(0) = · · · + 2 O(0) + @O(0) + · · · .
z̄ z̄
On the RHS of the first line, the · · · to the left represents terms in the Laurent expansion
in z of higher order poles than z12 , whereas the · · · to the right represents terms that are
regular at z = 0. The omitted terms in the second line are similar, with z replaced by z̄.
To determine the transformation of O under a more general conformal transformation with
nonlinear v(z), ve(z̄), would require knowing the higher order pole terms in the OPE (3.34).
A finite conformal transformation can be characterized by a general holomorphic di↵eo-
morphism
z 7! z 0 = z 0 (z), z̄ 7! z̄ 0 = z̄ 0 (z̄). (3.35)
Under such a symmetry, a local operator O should transform into another operator O0 , such
that O0 (z 0 , z̄ 0 ) behaves like a local operator at position z. If (3.35) is obtained by composing
infinitely many infinitesimal transformations of the form (3.29), then we can determine O0
by composing the infinitesimal transformations (3.32) on the operator O.
A special class of operators, called conformal primary operators or simply primaries, obey
OPEs with the stress-energy tensor of the form (3.34) without higher than second order poles
in z or z̄, namely
h 1
T (z) (0) = 2 (0) + @ (0) + · · · ,
z z
e (3.36)
h 1
Te(z̄) (0) = 2 (0) + @¯ (0) + · · · .
z̄ z̄
Under a general infinitesimal conformal transformation (3.29), transforms by

= ✏(h@v + v@ + e ¯v + ve@)
h@e ¯ . (3.37)
0
The finite form of the conformal transformation ! under (3.35) can be expressed as
e
0
(z 0 , z̄ 0 ) = (@z z 0 ) h
(@¯z̄ z̄ 0 ) h
(z, z̄). (3.38)

3.3 Free bosons

As a first example, let us consider the theory of a free massless scalar field in 2D, described
by the action Z Z
1 2 a 1 ¯
S= d @a X@ X = d2 z@X @X. (3.39)
4⇡↵0 2⇡↵0

27
Here we adopt a normalization convention of the field X so as to agree with one of the scalar
fields in the Polyakov action in conformal gauge.
We will write w = + i⌧ for the complex coordinate on the Euclidean cylinder, related
to the coordinate z of the complex plane by the exponential map z = e iw . We will denote
by t the time coordinate on the Lorentzian cylinder, which is related to the Euclidean time
⌧ by t = i⌧ .
The Hamiltonian on the cylinder is
Z 2⇡ 
1
H= d ⇡↵0 ⇧2 + (@ X)2 , (3.40)
0 4⇡↵0
1
where ⇧ = 2⇡↵ 0 @t X is the canonical momentum density. At time t = 0, we can decompose

X( , t = 0) and ⇧( , t = 0) into their Fourier modes,

r
↵ 0 X ↵n ↵ e n in
X( , t = 0) = x̂ + i e ,
2 n6=0 n
(3.41)
k̂ 1 X
in
⇧( , t = 0) = + p e n )e ,
(↵n + ↵
2⇡ 2⇡ 2↵0 n6=0

en turn into operators acting on the Hilbert space of the

where the coefficients x̂, k̂, ↵n , ↵
free boson theory on the unit circle. It follows from the equal time canonical commutation
relation

[X( , t), ⇧( 0 , t)] = i ( 0

), [X( , t), X( 0 , t)] = [⇧( , t), ⇧( 0 , t)] = 0 (3.42)

en are
that the non-vanishing commutators among x̂, k̂, ↵n , ↵

[x̂, k̂] = i, [↵n , ↵m ] = [e em ] = n

↵n , ↵ n, m . (3.43)
p
en ) with negative n may be viewed as |n| times the creation operator
The oscillator ↵n (or ↵
that creates a massless left (or right) moving particle on the circle that carries |n| units of
p
momentum. The oscillator ↵n (or ↵ en ) with positive n is n times the annihilation operator
of the corresponding mode. The Hilbert space of the free boson CFT on the circle was
already constructed in (2.38). In the oscillator notation, we can write a basis of states as

|k; {ni , n
ei }i 1 i = ↵ n1 ↵ n2 ···↵
e e ne2
e1 ↵
n · · · |k; 0i, (3.44)

where ni and nei are a set of positive integers. The oscillator vacuum |k; 0i has eigenvalue k
R 2⇡
with respect to k̂ = 0 ⇧d , and is annihilated by all of ↵n , ↵ en with positive n.

28
 What operators are (3.44) mapped to under the state/operator correspondence? A
naively guess is
@ n1 X@ n2 X · · · @¯ne1 X @¯ne2 X · · · eikX , (3.45)
which has the correct transformation property under the global symmetry that shifts X,
ei dependence in its conformal weight. However, as a product of field
and the expected ni , n
operators at coincident points, (3.45) is not well defined as a local operator, unless a suitable
normal ordering or regularization prescription is adopted.
Indeed, any correlation function that involves a pair of free boson field operators X(z1 , z̄1 )
and X(z2 , z̄2 ) has a divergence in the z1 ! z2 limit, of the form

hX(z1 , z̄1 )X(z2 , z̄2 ) · · · i = ↵0 ln |z12 |h· · · i + (regular terms in z1 ! z2 limit), (3.46)

due to the Wick contraction of X(z1 , z̄1 ) with X(z2 , z̄2 ). If we subtract o↵ the Green function
↵0 ln |z12 | from X(z1 , z̄1 )X(z2 , z̄2 ) to define the normal ordered product

: X(z1 , z̄1 )X(z2 , z̄2 ) :⌘ X(z1 , z̄1 )X(z2 , z̄2 ) + ↵0 ln |z1 z2 |, (3.47)

then
: X 2 (z2 , z̄2 ) :⌘ lim : X(z1 , z̄1 )X(z2 , z̄2 ) : (3.48)
z1 !z2

is a well defined local operator. We can also phrase this in terms of the OPE

X(z, z̄)X(0) = ↵0 ln |z|+ : X(z, z̄)X(0) :

X z n z̄ m (3.49)
= ↵0 ln |z| + : X@ n @¯m X(0) : .
n,m 0
n!m!

The second line is organized as an expansion in powers of z, z̄, with operators of increasing
conformal weights.
The normal ordering can be generalized to
⇢ Z
1
: F[X] : ⌘ exp d2 z1 d2 z2 ↵0 ln |z12 | F[X], (3.50)
2 X(z1 , z̄1 ) X(z2 , z̄2 )

where the functional F[X] may be taken to be product of the free boson field X and its
derivatives, and the functional derivative / X is normalized such that

X(z 0 , z̄ 0 ) = 2
(z z 0 ). (3.51)
X(z, z̄)

From now on, when we write the normal ordered product of free fields and their derivatives
at the same position, the normal ordering sign will be omitted.

29
 Now consider the OPE of X(z, z̄) with a general operator O(0) at the origin. The free
¯
boson equation of motion, @ @X(z, z̄) = 0, holds away from z = 0. In particular, @X(z) is
meromorphic with a Laurent expansion in z, and likewise @X(z̄) ¯ is anti-meromorphic. Thus
we can write
" r ✓ ◆#
i↵ 0
↵ 0 X 1 ↵n ↵ en
X(z, z̄)O(0) = x̂ k̂ ln |z|2 + i n
+ n O(0), (3.52)
2 2 n6=0 n z z̄

en are regarded as operators acting on O(0). Under the state/operator map,

where x̂, k̂, ↵n , ↵
iw
z = e , (3.52) is equivalent to the action of the operator X inserted at w on the Euclidean
cylinder, acting on the state |Oi, and x̂, k̂, ↵n , ↵
en coincide with those defined by the canonical
quantization of (3.41).
When it is understood that an operator is inserted at the origin, one often writes (3.52)
by omitting O(0) on both sides of the equation. We can also express k̂, ↵n , ↵
en as
I I
dz 2i dz̄ 2i ¯
k̂ = 0
@X = @X,
2⇡i ↵ 2⇡i ↵0
r I r I (3.53)
2 dz n 2 dz̄ n ¯
↵n = i z @X, ↵en = i z̄ @X, n 6= 0,
↵0 2⇡i ↵0 2⇡i
where the contour is always taken to be counterclockwise
q around the origin unless otherwise
↵0
e0 =
noted. Sometimes one also defines ↵0 = ↵ 2
k̂, so that the second line of (3.53) holds
for all integer n.
The state/operator correspondence maps the identity operator 1 to the ground state
|0; 0i, as both are annihilated by ↵n , ↵
en for n 1, as well as by b
k. From the first line of
ikX(0)
(3.53) and using the OPE of @X(z) with e , we have
I 
ikX(0) dz k ikX(0) 2i
k̂e = e + 0 : @X(z)eikX(0) : = keikX(0) , (3.54)
2⇡i z ↵

we see that the operator eikX(0) should be identified with the oscillator ground state |k; 0i.
By repeatedly acting on @ n1 X · · · @¯ne1 X · · · eikX(0) with ↵n or ↵ en with negative n, we see that
the Fock state (3.44) corresponds to the operator
" p #2 p 3
Y i 2/↵0 Y i 2/↵0 ne
@ nj X 4 @¯ ej X 5 eikX(0) . (3.55)
j
(nj 1)! (enej 1)!
e
j

Classically, the stress-energy tensor T (z) and Te(z̄) of the free boson theory are given by
1 1 ¯ ¯
T = @X@X, Te = @X @X. (3.56)
↵0 ↵0
30
In the quantum CFT, (3.56) must be related by the normal ordered expressions,
1 1 ¯ ¯
T = @X@X, Te = @X @X. (3.57)
↵0 ↵0
As usual, the product of fields is understood to be normal ordered. Note that there is no
room for the ambiguity of a normal ordering constant, as T and Te are expected to have
definite conformal weights (2, 0) and (0, 2) respectively.
By explicitly computing the OPE with T (z) and Te(z̄), one can see that eikX(0) , for
instance, is a primary of weight h = e
0 2
h = ↵ 4k . An important result is the OPE of T (z) with
itself, which can be put to the form
1 2 1
T (z)T (0) ⇠ 4
+ 2 T (0) + @T (0). (3.58)
2z z z
Here by ⇠ we mean the two sides of the equation are equal up to terms that are regular at
z = 0. This is a special case of the OPE algebra of the stress-energy tensor of a general 2D
CFT (3.92), with central charge c = 1.
A generalization of the free boson CFT, called the linear dilaton theory, is the following.
Starting with the free boson field X(z, z̄) of the same OPE relation as (3.49), we consider
the modified stress-energy tensor
1 Q 1 ¯ ¯ Q
T = 0
@X@X + p @ 2 X, Te = 0
@X @X + p @¯2 X, (3.59)
↵ ↵0 ↵ ↵0
for a constant Q known as the background charge. While X transforms as a weight (0, 0)
primary in the free boson CFT, its conformal transformation is modified in the linear dilaton
theory. Under the infinitesimal conformal transformation (3.29), X transforms by
" p #
↵ 0Q
X = ✏ (v@ + ve@)X¯ + ¯v ) .
(@v + @e (3.60)
2

The finite form of the conformal transformation, with respect to (3.35), is

p
X 0 (z 0 , z̄ 0 ) = X(z, z̄) ↵0 Q ln |@z z 0 | . (3.61)

In particular, in the conformal mapping from the cylinder to the plane (3.26), the current
@w X (w) that generates the shift symmetry of X on the cylinder is related to the current
@z X (z) on the plane by
" p #
0
↵Q
@w X (w) (w, w̄) = @w z @z X (z) (z, z̄) + . (3.62)
2z

31
By comparing the charge with respect to this current, we conclude that the oscillator ground
state |k, 0i should be mapped to the operator
✓ ◆
Q
exp p + ik X(0) . (3.63)
↵0
The T T OPE of the linear dilaton theory obeys the relation (3.92) with central charge
c = 1 + 6Q2 . The operator (3.63) has conformal weight h = e h = 14 (Q2 + ↵0 k 2 ). Note that,
for nonzero real background charge Q, the closure of OPE would demand that we include
states with complex values of k; as a result the linear dilaton theory is generally non-unitary.4
q
It is sometimes useful to speak of the “holomorphic part” of the free boson ↵20 X(z, z̄)
as a “chiral boson” field (z), that obeys the OPE

(z) (0) ⇠ ln z, (3.64)

and “holomorphic operators” such as e of weight h = 12 2 (and e h = 0), as building blocks

of more complicated theories. The OPEs of such “chiral” or “holomorphic” operators are
generally not single-valued functions of the coordinates, and their definitions require choice
of branch cuts. We will see in section 6.2 that they are a special case of “defect operators”
that are attached with topological defect lines.

3.4 Free fermions

The 1+1 dimensional free massless fermion theory in a general background metric gab can
be described by the covariant action
Z
1 p
S= d2 g a rspin
a . (3.65)
4⇡
is a Grassmannian 2-component spinor field, whose components are denoted ↵ , ↵ = ±.
The spinor index can be raised or lowered according to the convention ↵ = ✏↵ , ✏+ =
✏+ = 1. Unless otherwise noted, the contraction of spinor indices is from upper left to lower
right, e.g. ( ) ⌘ ↵ ↵ , ( a ) ⌘ ↵ ( a )↵ , etc. The matrices a are dependent on
and obey { a , b } = 2g ab . They are related to the standard 2D gamma matrices ˆ i by
a
= ei a ˆ i , (3.66)

where ei a is the inverse frame field, related to the metric by

j
gab = ⌘ij ei a ej b , ei a e j a = i. (3.67)
4
However, we will see in section 6.8 that the linear dilaton theory with real Q can be deformed in a certain
sense to a unitarity interacting CFT, known as Liouville CFT.

32
Our convention for ˆ i is such that in light cone coordinates ±
= ± t, the non-vanishing
components of ˆ ± are ( ˆ + )++ = ( ˆ ) = 2.
The spin connection rspin
a is defined as
1
rspin
a ⌘ @a + !a ij ˆ ij , (3.68)
4
where !a ij is defined by the property that the spin covariant derivative of the frame field ei a
is zero, namely
ra ei b + !a i j ej b = 0, (3.69)
where ra is the Levi-Civita connection, ra ei b = @a ei b c i
ab e c .

Note that in 2D, since ˆ ij ⌘ 0, the spin connection term involving !a ij does not
contribute to the action (3.65), and so we can in fact replace rspin
a by @a in (3.65). However,
ij
!a does enter the equation of motion, which is given by
a
rspin
a = 0. (3.70)

The classical stress-energy tensor is

2⇡ S 1 1
Tab = p ei (a ib) = (a @b) + gab c
@c , (3.71)
g e 2 2

which is conserved and traceless up to the equation of motion (3.70).

Now specializing to the 2D Minkowskian worldsheet, the free fermion action can be
written as Z
1
S= d2 ( + @ + + @+ ) , (3.72)
2⇡
with the stress-energy tensor
1 1
T++ = + @+ +, T = @ . (3.73)
2 2
Let us consider the canonical quantization of this system on the cylinder, parameterized by
( , t) with ⇠ + 2⇡. The canonical momentum density conjugate to ± is

1
⇧± = S=⌥ ±. (3.74)
@t ± 4⇡
According to the terminology of Dirac, these canonical momenta obey “second-class primary
constraints”
± 1
± ⌘ ⇧ ± ± = 0, (3.75)
4⇡

33
with the equal-time Poisson anti-bracket
1
K± ( , 0 ) ⌘ { ±( ), ±(
0
)}P = ± ( 0
). (3.76)
2⇡
Here the Poisson anti-bracket is computed by pretending that ± and ⇧± are independent
variables. The classical Dirac bracket between a pair of functionals ⇠ and ⌘ of ± ( ) and
⇧± ( ) at time t = 0, that consistently takes into account the constraint (10.4), is then
defined by Z
{⇠, ⌘}D = {⇠, ⌘}P d {⇠, ±( )}P K± 1 { ±( ), ⌘}P , (3.77)

where K± 1 = ±2⇡ is the inverse of the operator K± defined by convolution with the inte-
gration kernel K± ( , 0 ). This leads to
0 0
{ ±( ), ±( )}D = ⌥2⇡ ( ). (3.78)

The canonical quantization proceeds by promoting ± ( ) to operators that obey the anti-
commutators
{ ± ( ), ± ( 0 )} = ⌥2⇡i ( 0
). (3.79)
There is an ambiguity in the choice of periodic boundary condition for the field ± ( ),
namely we only need to demand that T++ and T are periodic under the shift ! + 2⇡,
whereas + and may independently be either periodic or anti-periodic. If the fermion
field is periodic in a given state, we refer to that state as belonging to the “Ramond sector”,
or R sector. If the fermion field is anti-periodic, we refer to the state as in the “Neveu-
Schwarz sector”, or NS sector. A priori, the Hilbert space of the free fermion theory on the
cylinder may involve states in either (NS, NS), (NS, R), (R, NS), or (R, R) sectors, where the
notation is such that (NS, R) for instance refers to a sector in which + is anti-periodic and
is periodic. We will see later that, in order for the CFT to have closed OPE algebra and
the property of modular invariance, the actual Hilbert space of the CFT on the circle, which
is in correspondence with local operators, should only include a subset of all possible states
in the various sectors. Those states excluded from this Hilbert space may be interpreted as
defect operators.
Let us consider the Fourier expansion of +() at t = 0,
⇡i
X
ir
+( )=e 4
re , (3.80)
r2Z+⌫

1
where ⌫ = 0 for the R sector and ⌫ = 2
for the NS sector. It follows from (3.79) that

{ r, s} = r, s . (3.81)

34
 1
The NS sector ground state |0iNS is annihilated by all r with r 2
. The NS sector Fock
space can be constructed by acting on |0iNS with r for r  12 .
As for the R sector, we may consider a ground state |0iR that is annihilated by all r with
r 1. The zero mode 0 obeys 02 = 12 , and we may choose |0iR to have either eigenvalue
p1 or p12 with respect to 0 .
2

The analysis of NS versus R sector Fock space of e is similar.

Now let us turn to the local operators in the free fermion theory on the Euclidean plane.
After analytic continuation to Euclidean signature, we will write + as , and as e. The
Euclidean action, written in complex coordinates (z, z̄), is
Z ⇣ ⌘
1 2 ¯ e e
S= dz @ + @ . (3.82)
4⇡
From the free fermion Green function, we can define the normal ordering, or equivalently
the OPE, of a pair of fermion fields according to
1 e(z̄) e(0) = 1 + : e(z̄) e(0) : .
(z) (0) = + : (z) (0) :, (3.83)
z z̄

The OPE between and e is always non-singular. In writing normal ordered products of
free fermion fields and their derivatives at coinciding points, the normal ordering sign will
be omitted just in the free boson convention.
The quantum stress-energy tensor of the free fermion CFT is given by the normal ordered
expression
1 1 e¯ e
T = @ , Te = @ . (3.84)
2 2
They obey the OPE (3.92) with central charge c = e c = 12 .
It follows from the OPE that is a primary of weight ( 12 , 0), and e is a primary of weight
(0, 12 ), as they should be. However, note that (0) and e(0) have half-integer spin h e h,
and flips sign under the 2⇡ rotation around the origin. This is because and are not e
well defined local operators by themselves, but are rather defect operators attached to a Z2
topological defect line (see section 6.2).
Nonetheless, we may consider the conformal transformation of and e from the plane
iw
to the cylinder, under the exponential map z = e . We have
1
(w)
(w) = (@w z) 2 (z)
(z), e(w) (w̄) = (@¯w̄ z̄) 12 e(z) (z̄). (3.85)
1 1
The conformal factors (@w z) 2 and (@¯w̄ z̄) 2 themselves are not single-valued functions on the
complex plane, and introduce a branch cut. The OPE of (z), for instance, with a generic

35
operator O(0) at the origin, takes the form
X r
(z)
(z)O(0) = r+ 12
O(0), (3.86)
r2Z+⌫ z

where r are the Fourier modes as defined in (3.80). Here ⌫ = 0 if O corresponds to a state
|Oi on the cylinder in the R sector, and ⌫ = 12 if O corresponds to a state in the NS sector.
Thus, we see that in the presence of an NS sector operator ONS (0), (z) is single valued,
whereas in the presence of an R sector operator OR (0), (z) has a branch cut.
The NS sector ground state |0iNS simply maps to the identity operator 1 under the
state/operator correspondence. The entire NS sector Fock space maps to operators formed
by normal ordered products of and its derivatives.
The R sector ground state |0iR maps to an operator S(0), whose OPE with (z) has the
form
1
(z)S(0) = ± p S(0) + · · · , (3.87)
2z
1
where the omitted terms vanish in the z ! 0 limit at order z 2 and higher. The conformal
weight of S can be determined as follows. Consider the triple OPE

(z1 ) (z2 )S(0) = f (z1 , z2 )S(0) + · · · , (3.88)

where · · · represents operators that correspond to states orthogonal to |0iR or S(0). In the
regime |z1 | > |z2 | > 0, we can use the independent Laurent expansions of (z1 ) and (z2 )
to compute
1
X
1 1 z1 + z2
f (z1 , z2 ) = p + 1 1 = p . (3.89)
2 z1 z2 r=1 z 2 z
r+ r+ 2 2z12 z1 z2
1 2

This result can then be analytically continued outside of the range |z1 | > |z2 | > 0. Now
taking the z1 ! z2 limit, and comparing (3.88) with the OPE
1 2
(z1 ) (z2 ) = + 2z12 T (z2 ) + O(z12 ), (3.90)
z12
we find
1
T (z2 )S(0) = S(0) + · · · , (3.91)
16z22
where · · · correspond states orthogonal to S(0), necessarily of higher conformal weights.
1
From (3.91) we see that S(0) must be a conformal primary of weight h = 16 . A similar
e
analysis applies to and the anti-holomorphic NS and R sectors.

36
3.5 The Virasoro algebra

In a 2D CFT, the OPE of the holomorphic stress-energy tensor T (z) with itself generally
takes the form
c 2 1
T (z)T (0) ⇠ 4 + 2 T (0) + @T (0), (3.92)
2z z z
where the constant c is called the central charge, and terms on the RHS that are regular
at z = 0 are omitted. Note that the terms of order z12 and z1 are determined by the fact
that T has conformal weight (2, 0). There is an analogous OPE of the anti-holomorphic
stress-energy tensor Te(z̄) that involves another central charge e
c. We have already seen that
the free boson CFT has c = e c = 12 .
c = 1, whereas the free fermion theory has c = e
The appearance of the order z14 term on the RHS of (3.92) implies that T (z) is not a
conformal primary. Rather, it transforms under the infinitesimal conformal transformation
(3.29) according to h i
c
T = ✏ (2@v + v@)T + @ 3 v . (3.93)
12
The finite form of the transformation, with respect to (3.35), turns out to be
h c 0 i
T 0 (z 0 , z̄ 0 ) = (@z z 0 ) 2 T (z, z̄) {z , z}S , (3.94)
12
where {z 0 , z}S is the Schwarzian derivative, defined as

2(@z3 z 0 )(@z z 0 ) 3(@z2 z 0 )2

{z 0 , z}S ⌘ . (3.95)
2(@z z 0 )2

Under successive conformal transforms z ! z 0 ! z 00 , (3.94) composes correctly thanks to

the following property of the Schwarzian derivative,

{z 00 , z}S = (@z z 0 )2 {z 00 , z 0 }S + {z 0 , z}S . (3.96)

One can then prove (3.94) by noting that it reduces to (3.93) under infinitesimal transfor-
mations.
iw
Under the state/operator map z = e , the stress-energy tensor T (w) on the cylinder
and T (z) on the plane are related by
c
T (w) (w) = (@w z)2 T (z) (z) + . (3.97)
24
In the presence of an operator O(0) at the origin, we will write the Laurent expansion of
T (z) (z) as
X Ln
T (z) (z) = . (3.98)
n2Z
z n+2

37
Correspondingly, on the cylinder, we have
X c
T (w) (w) = einw Ln + . (3.99)
n2Z
24

We can write similar expansions for the anti-holomorphic stress-energy tensor,

X L en X e
c
Te(z) (z̄) = , Te(w) (w̄) = e inw̄ e
Ln + . (3.100)
n2Z
z̄ n+2 n2Z
24

The energy and momentum on the cylinder can be written as

Z 2⇡ Z 2⇡
d d h (w) i
H= (w) t
(T ) t = T (w) + Te(w) (w̄) ,
0 2⇡ 0 2⇡
Z 2⇡ Z 2⇡ (3.101)
d d h (w) i
P = (T (w) )t = T (w) Te(w) (w̄) .
0 2⇡ 0 2⇡
In the second equality on each line, we converted the Lorentzian expressions to Euclidean
expressions. Our convention of the momentum P is in the direction of decreasing . Using
(3.98) and (3.98), we can express H and P in terms of the zeroth Fourier modes as

e0 c+e c e0 c e
c
H = L0 + L , P = L0 L . (3.102)
24 24
For an operator O(0) of conformal weight (h, e
h), it follows from the OPE (3.34) that

L0 |Oi = h|Oi, L 1 |Oi = |@Oi,

(3.103)
e0 |Oi = e
L h|Oi, e 1 |Oi = |@Oi.
L ¯

We see that H and P can be identified with the dilation generator L0 + L e0 and the rotation
generator L0 L e0 up to constant shifts; the latter, proportional to the central charges, are
interpreted as Casimir energy-momentum.
Of particular significance is the algebra generated by Ln and L en with respect to Lie
brackets. Acting on a generic state/operator O(0), we can write
I I
dz1 dz2 n+1 m+1
Ln L m = z1 z2 T (z1 )T (z2 ), (3.104)
C1 2⇡i C2 2⇡i

where C1 , C2 are a pair of counterclockwise contours encircling the origin, with C1 outside
of C2 . Lm Ln would be given by an expression identical to (3.104) except that C1 would be
replaced by another counterclockwise contour C10 that lies within C2 .
The commutator [Ln , Lm ] is thus computed by the integral analogous to (3.104), except
that z1 is integrated over C1 C10 , that is the union of the contour C1 and the orientational

38
reversal of C10 . The contour C1 C10 now encloses every point on C2 , and does not enclose
the origin. Therefore, we can shrink C1 C10 to a counterclockwise contour Cz2 encircling z2
without a↵ecting the integral. This leads to
I I
dz2 dz1 n+1 m+1
[Ln , Lm ] = z1 z2 T (z1 )T (z2 )
C2 2⇡i Cz2 2⇡i
I I 
dz2 dz1 n+1 m+1 c 2 1
= z1 z2 4
+ 2 T (z2 ) + T (z2 )
C2 2⇡i Cz2 2⇡i 2z12 z12 z12
I
dz2 m+1 h c i
= z2 (n + 1)n(n 1)z2n 2 + 2(n + 1)z2n T (z2 ) + z2n+1 @T (z2 ) ,
C2 2⇡i 12
(3.105)
where in deriving the second line, we used the OPE of T (z1 ) with T (z2 ), keeping only the
singular terms that contribute to the residue at z1 ! z2 . The last line of (3.105) can then
be re-expressed in terms of the Laurent coefficients, resulting in
c 3
[Ln , Lm ] = (n m)Lm+n + (n n) n, m . (3.106)
12
This is known as the (centrally extended) Virasoro algebra, and the Ln ’s will be referred to
en ’s, with c replaced by e
as Virasoro generators. There is a similar algebra of the L c.
Note that on the Lorentzian cylinder, T++ and T are Hermitian operators. Therefore,
the holomorphic Virasoro generators are mapped to themselves under Hermitian conjugation,
and so do the anti-holomorphic ones,

L†n = L n,
e† = L
L e n. (3.107)
n

One can organize the local operators, or equivalently, states of the CFT on the circle, in
terms representations of the holomorphic and the anti-holomorphic Virasoro algebra. One
may begin with a lowest weight state | i that obeys
en | i = 0, n 1,
Ln | i = L
(3.108)
L0 | i = h| i, L e0 | i = e
h| i.

Comparison with (3.36), we see that the lowest weight state | i is precisely mapped to a
primary operator (0) under the state/operator correspondence. An irreducible representa-
tion of the Virasoro algebra can be constructed as the span of Virasoro descendants of | i,
e n ’s for positive
namely states obtained by acting on | i with an arbitrary set of L n and L
integer n.
In a unitary CFT, there is a positive definite Hermitian inner product on the Hilbert space
of local operators. In particular, L0 and Le0 are Hermitian operators, whose spectra (namely
the set of conformal weights) are real and bounded from below. The Hilbert space can be

39
decomposed into irreducible representations of Virasoro algebra, each of which is generated
by a lowest weight state of definite conformal weight (h, e
h). Thus, all local operators can be
expressed as linear combinations of Virasoro descendants of primaries.
The representations of Virasoro algebra will be analyzed in more detail in section 3.7.
We will see in section 3.8 and 3.10 that the spectrum of primaries together with their 3-point
functions determine all correlation functions of the CFT on any surface.

3.6 Weyl anomaly

In this section we consider 2D CFT on a curved Euclidean worldsheet. Suppose S is the ac-
tion of a classical field theory coupled covariantly to a background metric gab . A deformation
of the metric leads to the variation of the action
Z
1 p
S= d2 g gab T ab . (3.109)
4⇡
In the quantum theory, a deformation of the metric amounts to inserting
 Z
1 p
exp d2 g gab T ab (3.110)
4⇡
into the path integral or into a correlation function. If the metric changes by a Weyl trans-
formation, gab = 2 ! gab , then the integrand appearing in the exponent of (3.110) is propor-
tional to T a a , which vanishes in a CFT on a flat worldsheet. If T a a remains zero on a curved
worldsheet, then the CFT coupled to the background metric would be Weyl invariant.
We must be cautious in dealing with expressions like (3.110), which involves integrating
products of local operators at coinciding points, and could potentially receive contributions
from contact terms that would have been missed if we assume operator identities such as
equations of motion that only hold for operators at separated points. For instance, applying
@¯z̄ to the OPE of the holomorphic stress-energy tensors (3.92), and keeping track of the
contact terms using @¯z̄ (1/z) = 2⇡ 2 (z) (recall that 2 (z) = 12 2 ( ) in our convention), we
have
h c i
@¯z̄ Tzz (z) Tzz (0) = 2⇡ @ 3 2 (z) 2@ 2 (z12 )Tzz (0) + 2 (z)@Tzz (0) . (3.111)
12
This can be compared with the contact term that arises from the divergence of the stress-
energy tensor due to the Ward identity associated with Poincaré symmetry,
⇥ ⇤
@¯z̄ Tzz + @z Tz̄z (z, z̄) Tzz (0) = 2⇡ 2@ 2 (z12 )Tzz (0) + 2 (z)@Tzz (0) , (3.112)
from which we conclude that (after integrating once in z)
c
Tz̄z (z, z̄) Tzz (0) = 2⇡ @ 2 2 (z). (3.113)
12
40
Now shifting Tzz (0) to a general point z2 , and acting with @¯z̄2 , we have

¯ zz (z2 ) = c 2¯ 2
Tzz̄ (z1 , z̄2 ) @T 2⇡ @ @ (z12 ). (3.114)
12
The product of (@T ¯ zz + @Tz̄ )(z2 , z̄2 ) with Tzz̄ (z1 , z̄1 ) contains only contact terms proportional
to Tzz̄ and @Tzz̄ , which can be set to zero in the absence of other operators at coincident
points. Thus we can replace @T ¯ zz with @Tzz̄ in (3.114). Integrating once in z2 , we arrive at

c ¯ 2
Tzz̄ (z1 , z̄1 )Tzz̄ (z2 , z̄2 ) = 2⇡ @ @ (z12 ). (3.115)
12
Note that the same derivation with z and z̄ exchanged would result in an equation identical
to (3.115) except that c would be replaced by e c. Consistency then requires c = ec, which
amounts to the absence of gravitational anomaly.
We can deduce the Weyl transformation of Tzz̄ by integrating (3.115) as follows. Starting
from the Euclidean metric, the infinitesimal Weyl transformation gab = 2 ! ab e↵ectively
changes Tzz̄ by
Z
1 c
W Tz z̄ (z1 , z̄1 ) = d2 z2 !(z2 , z̄2 )Tzz̄ (z1 , z̄1 )Tzz̄ (z2 , z̄2 ) = @ @¯ !(z1 , z̄1 ). (3.116)
⇡ 6
Comparing this with the Ricci scalar of the infinitesimally Weyl transformed metric, R =
2r2 !, we can write (3.116) in a covariant form,

a c
WT a = W R, (3.117)
12
which can then be generalized to arbitrary background metric. Composing the infinitesimal
Weyl transformations into a finite one leads to the formula
c
T aa = R. (3.118)
12
This is known as the conformal/Weyl/trace anomaly of a 2D CFT.
If we start with a general metric gab , whose Ricci scalar is denoted R[g], the Weyl trans-
0
formed metric gab = e2! gab has Ricci scalar

R[g 0 ] = e 2!
(R 2r2 !). (3.119)

The partition functions of the CFT in the two background metrics, Z[g] and Z[g 0 ], are related
by the Weyl anomaly factor
 Z Z
Z[g 0 ] 1 2 p
= exp d g ! e2! T a a = exp ( SW [!]) , (3.120)
Z[g] 2⇡

41
where Z Z
c 2 p
SW [!] = d g ! (R 2r2 !)
24⇡ Z
(3.121)
c 2 p ab
= d g g @a !@b ! + !R
24⇡
takes the form of the classical action of a linear dilaton theory, where ! plays the role of the
linear dilaton field. This is the result claimed in (2.23).

3.7 Representations of Virasoro algebra

Let us now examine the lowest weight representations of Virasoro algebra in some detail. We
will focus on the holomorphic Virasoro algebra. Consider a primary state |⌫h i that obeys

L0 |⌫h i = h|⌫h i, Ln |⌫h i = 0, n 1. (3.122)

A representation is spanned by the Virasoro descendants

L N |⌫h i =L n1 ···L nk |⌫h i, (3.123)

where N stands for a sequence of positive integers {n1 , n2 , · · · , nk } in descending order

P
n1 n2 · · · nk . We will refer to |N | ⌘ kj=1 nj as the level of the state (3.123). The
level n Gram matrix is defined as the matrix whose entries are inner products of pairs of
level n descendants of |⌫h i,
(n) †
GN M = h⌫h |(L N) L M |⌫h i, |N | = |M | = n. (3.124)

Recall that L† N = Lnk · · · Ln1 . The Gram matrix elements can be computed by moving all
the positively graded Virasoro generators to the right of the negatively graded ones using the
Virasoro algebra; the results are generally polynomial functions of h and the central charge
c.
Of particular use is the determinant of the Gram matrix, known as the Kac determinant,
which is given by formula
Y
det G(n) = Kn (h hr,s )P (n rs) . (3.125)
1rsn

Here Kn is a positive constant, having to do with the (somewhat arbitrary) choice of basis
(3.123). The product is taken over a pair of positive integers r, s that obey rs  n. P (m)
is the partition number, i.e. the total number of positive integer partitions of m. It can be
computed as coefficients of the generating function
X Y 1
Pm q m = . (3.126)
m 0 k 1
1 qk

42
Finally, hr,s , which may be real or complex, are given by
p p
c 1 1 2 1 c ± 25 c (3.127)
hr,s = + (r↵+ + s↵ ) , ↵± ⌘ .
24 4 24
It is sometimes useful to define c ⌘ 1 + 6Q2 , Q ⌘ b + b 1 , and write hr,s in the form
Q2 1
hr,s = (rb + sb 1 )2 . (3.128)
4 4
The significance of (3.125) is that a primary ⌫hr,s of weight h = hr,s admits a Virasoro
descendant at level rs that is a null state. That is, there is a state of the form
X
N
| rs i = rs L N |⌫hr,s i (3.129)
!N |=rs

that has zero norm,

h rs | rs i = 0. (3.130)
Equivalently, the coefficients { Nrs } form a zero eigenvector of the Gram matrix. The nor-
{1,··· ,1}
malization is chosen so that rs = 1, i.e. the coefficient of (L 1 )rs |⌫hr,s i in (3.129) is 1.
At a higher level n > rs, there will be P (n rs) descendants of the form L M | rs i with
|M | = n rs that also have zero norm, giving rise to an order P (n rs) zero of the Kac
determinant.
Now let us consider a unitary representation generated by the primary |⌫h i. In a unitary
representation, any null sate must vanish identically. It follows from h⌫h |L1 L 1 |⌫h i = 2h
that h 0; the h = 0 case corresponds to the identity/vacuum representation of Virasoro
algebra.
If c > 1, hr,s are either non-positive real numbers (for c 25) or complex (for 1 < c < 25);
in this case, det G(n) being a real function of h cannot have zeroes on the positive real axis,
and therefore must be positive for all h > 0. Consequently, all representations with c > 1 and
h > 0 are unitary and do not contain null Virasoro descendants. The vacuum representation
h = 0 is also unitary for c > 1, but contains a null state at level 1, namely L 1 |0i = 0.
In the case c = 1, det G(n) is non-negative for h 0, but it does have zeroes along the
positive real axis, at hr,s = 14 (r s)2 for r 6= s. Consequently, while all c = 1, h 0
representations are unitary, for the special values h = 4 n , n 2 Z 0 , the representation
1 2

admits null state relations starting at level n + 1.

For c < 1, one can show that if h is not equal to one of the hr,s ’s, det G(n) will be negative
for some n and violate unitarity. A careful analysis, which we shall not reproduce here, leads
to the following discrete set of possible unitary representations with c < 1:
6 ((m + 1)r ms)2 1
c=1 , h = hr,s = , with 1  r  m 1, 1  s  m,
m(m + 1) 4m(m + 1)
(3.131)

43
where m = 2, 3, · · · . The unitary CFTs with c = e
c < 1 come in discrete families as well; they
are known as minimal models, and are completely classified and exactly solvable. These will
be discussed in section 6.7.

3.8 Conformal correlators and conformal blocks

Let us begin with correlation functions of local operators on the Euclidean plane. In a general
quantum field theory, as discussed in section 3.1, these are defined as (analytic continuation)
of vacuum correlation functions. For a 2D CFT, the plane is conformally equivalent to the
cylinder or the Riemann sphere, and we may equivalently speak of correlators on the cylinder
or the sphere. In particular, correlation functions on the sphere should be invariant under
conformal transformations that are globally defined on the sphere; as will be elaborated in
section 4.7.1, these correspond to Möbius transformations on the plane of the form
↵z +
z 7! z 0 = , (3.132)
z+
where ↵, , , are complex numbers, that may be taken to obey ↵ = 1. Such
transformations form the group P SL(2, C), which is a finite dimensional subgroup of the
full Virasoro symmetry group. Under such a conformal transformation, a primary (z, z̄) of
weight (h, e
h) transforms to 0 (z, z̄), related by
e
0
(z 0 , z̄ 0 ) = ( z + )h (¯ z̄ + ¯)h (z, z̄). (3.133)
The conformal invariance of the correlation function on the plane implies that
hO10 (z1 , z̄1 ) · · · On0 (zn , z̄n )i = hO1 (z1 , z̄1 ) · · · On (zn , z̄n )i (3.134)
for any set of local operators Oi and the conformally transformed operators Oi0 under (3.132).
Note that in writing (3.134) the conformal symmetry is viewed as acting actively on the op-
erators, leaving the coordinates invariant. Furthermore, (3.134) only captures the P SL(2, C)
invariance and not the full Virasoro symmetry; the implication of the latter will be discussed
later in this section.
To begin with, note that by translation, rotation, and dilation and symmetries, the one-
point function hO(z, z̄)i of an operator O(z, z̄) of conformal weight (h, e
h) must vanish unless
e
h = h = 0, in which case the one-point function is equal to a constant. In a unitary CFT,
the only operator with nonzero one-point function is the identity operator.
Next, consider a two-point function of a pair of local operators Oi and Oj , of weight
(hi , e
hi ) and (hj , e
hj ) respectively. Using translation, rotation, and dilation symmetries, one
can easily see that
hOi (0)Oj (1)i
hOi (z1 , z̄1 )Oj (z2 , z̄2 )i = . (3.135)
h +h e h +eh
z12i j z̄12i j

44
Now let us consider the special case where Oi , Oj are conformal primaries, denoted i , j .
We can consider a P SL(2, C) transformation that fixes the points 0 and 1, corresponding to
↵z
z 7! z 0 = . (3.136)
(↵ ↵ 1 )z + ↵ 1

It follows from (3.133) that

0 hi e
hi 0 e
i (0) =↵ ↵
¯ i (0), j (1) = ↵ hj ↵
¯ hj j (1). (3.137)

Setting h 0i (0) 0j (1)i = h i (0) j (1)i implies that unless hi = hj and e

hi = e
hj , the two-point
function must vanish identically. Thus, the two-point functions of primaries take the form
Dij
h i (z1 , z̄1 ) j (z2 , z̄2 )i = (3.138)
2hi 2e
z12 z̄12hi
for some constant matrix Dij that depend on the choice of basis i . Note that (3.138) is
single-valued provided that the spin hi e hi is a half-integer or integer.
Given an operator O of weight (h, e
h), it is sometimes useful to define

O0 (1) ⌘ lim O0 (z, z̄), (3.139)

z!1

where O0 is the conformal transformation of O with respect to the inversion map z 7! z 0 =

1/z. Under the state/operator map, while O(0) maps to the state |Oi on the cylinder, O0 (1)
maps to the BPZ conjugate state of |Oi, which we denote by hhO|. In particular, for the
primary i , we have
0 2hi 2e
i (1) ⌘ lim z z̄ hi i (z, z̄), (3.140)
z!1

and so
h 0i (1) j (0)i = hh i | j i = Dij . (3.141)
The BPZ conjugate of a Virasoro descendant L e
N L M | ii
e
is hh i |(L M)
†
(L N)
†
.
There is an anti-unitary map I that takes the Hermitian conjugate state hO| to the BPZ
conjugate hhO|,
hhO| = hO|I. (3.142)
Typically, we can find a Hermitian basis of operators on which I acts as the identity map,
in which case the distinction between hhO| and hO| can be omitted. Here by a Hermitian
operator we mean one that is Hermitian on the Lorentzian cylinder; for instance, the stress-
energy tensor components T and Te are both viewed as Hermitian operators, even though
on the Euclidean plane they are complex conjugations of one another. Note that in the free
boson CFT, the oscillator ground state eikX is not a Hermitian operator; in this case we have
hhk| = h k|.

45
 We say that a CFT is compact if there are finitely many linearly independent states
whose conformal weights are below any given finite value.5 In a compact unitary CFT, we
can further choose a Hermitian basis of primaries i that obey (3.141) with Dij = ij . In a
noncompact unitary CFT, we often need to deal with delta-function normalizable states. For
instance, in the free boson CFT, the oscillator ground state |ki obeys hk|k 0 i = CX 2⇡ (k k 0 )
for some convention-dependent normalization constant CX .
Next, let us consider the three-point function of primaries. There is a P SL(2, C) trans-
formation that maps three distinct points z1 , z2 , z3 on C [ {1} to any three other distinct
points. This allows us to determine
Cijk
h i (z1 , z̄1 ) j (z2 , z̄2 ) k (z3 , z̄3 )i = ,
h +hj hk eh +eh e
h h +h h e h +eh e
h h +h h e h +eh e
h
z12i z̄12i j k z23j k i z̄23j k i z31k i j z̄31k i j
(3.143)
0
where Cijk = h i (0) j (1) k (1)i
is called the structure constant. (3.143) is single-valued
e
provided that the spins hi hi are half-integer or integer, and that the sum of the spins of
the three operators is an integer. Cijk is either completely symmetric or completely anti-
symmetric in the indices i, j, k depending on whether the sum of spins is even or odd.
In a unitary CFT, reflection positivity further demands that, for a Hermitian basis of
primary operators i (in the sense of Hermiticity on the Lorentzian cylinder), the structure
constants Cijk must be real if the sum of three spins is even, and purely imaginary if the
sum of spins is odd.
Given the n-point function of a set of primaries, h 1 (z1 , z̄1 ) · · · n (zn , z̄n )i, one can de-
termine the n-point functions of any Virasoro descendants of the i ’s using conformal
Ward identities, as follows. Suppose we already know all n-point functions of the form
hO1 (z1 , z̄1 ) · · · On (zn , z̄n )i, where the total holomorphic Virasoro descendant level is less than
or equal to N . Consider the correlator

hT (z)O1 (z1 , z̄1 ) · · · On (zn , z̄n )i . (3.144)

Viewed as a function of z, (3.144) is meromorphic, with possible poles at z = zi , i = 1, · · · , n.

Using the inversion z 7! z 0 = 1/z, under which Oi transforms into Oi0 , and T to T 0 that
obeys T 0 (z 0 ) = z 4 T (z), we have
hT (z 0 )O1 (z10 , z̄10 ) · · · On (zn0 , z̄n0 )i = hT 0 (z 0 )O10 (z10 , z̄10 ) · · · On0 (zn0 , z̄n0 )i
(3.145)
= z 4 hT (z)O10 (z10 , z̄10 ) · · · On0 (zn0 , z̄n0 )i .
In the z ! 0, or z 0 ! 1 limit, the RHS vanishes like z 4 . Thus we learned that (3.144) must
fall o↵ like z 4 in the z ! 1 limit. Thus, (3.144) is determined entirely by the singular
5
This terminology is motivated by CFTs that arise from nonlinear sigma models on a compact target
space.

46
terms in its Laurent expansions around z = zi . For instance, in the z ! z1 limit, the singular
terms of (3.144) are
X
(z z1 ) n 2 hLn · O1 (z1 , z̄1 ) · · · On (zn , z̄n )i
n 1
✓ ◆ X
1 n 2
= @z1 + h1 hO1 (z1 , z̄1 ) · · · On (zn , z̄n )i + (z z1 ) hLn · O1 (z1 , z̄1 ) · · · On (zn , z̄n )i ,
z z1 n 1
(3.146)
where h1 is the holomorphic conformal weight of O1 . The first term on the RHS is determined
in terms of the n-point function hO1 (z1 , z̄1 ) · · · On (zn , z̄n )i, whereas the second term is an
analogous n-point function of Virasoro descendants, whose total holomorphic level is less
than N , thus already known by our assumption. We can then recursively determine (3.144)
as a meromorphic function in z, which then determines the n-point functions with one extra
Virasoro raising operator L k (k 1) acting on one of the Oi ’s.
The conformal Ward identities described above also determines the OPE of any pair of
operators in terms of the structure constants Cijk appearing in the three-point function of
primaries. For instance, the OPE of a pair of primaries i , j takes the form
X X
i (z1 , z̄1 ) j (z2 , z̄2 ) = Cij k eM (e
BN (hi , hj ; hk |z13 , z23 )B hi , e
hj ; e e M k (z3 , z̄3 ),
hk |z̄13 , z̄23 )L N L
k N ,M
(3.147)
where the sum is taken over a basis of primaries, as well as the integer partitions N , M
labeling holomorphic and anti-holomorphic Virasoro descendants. The coefficients Cij k are
P
related to the structure constants by Cijk = ` Cij ` D`k , where Dij are the two-point function
coefficients appearing in (3.141). The functions BN (hi , hj ; hk |z13 , z23 ) are entirely determined
by conformal Ward identities, and depend only on the central charge c of the holomorphic
Virasoro algebra. Likewise, B e is a similar function that depends on e c.
By repeatedly applying OPEs, at least in the domain where they converge, a general
n-point function can be determined from the spectrum of primaries and their structure
constants using the conformal Ward identities. Outside of the naive domain of convergence,
the n-point functions can be determined by analytic continuation.
Let us illustrate this algorithm in the case of the 4-point function of primaries i (zi , z̄I ),
i = 1, 2, 3, 4. Using a P SL(2, C) transformation, we can put
* 4 +
Y
i (zi , z̄i )
i=1
h i
= z13h1 h2 h3 h4
z14h1 +h2 +h3 h4
z242h2 z34
h1 +h2 h3 h4
⇥ (zi ! z̄i , hi ! e
hi ) h 1 (0) 2 (z, z̄) 3 (1) 04 (1)i ,
(3.148)

47
where
z12 z34
z= (3.149)
z13 z24
is the P SL(2, C) invariant cross ratio of the zi ’s, and 04 is defined as in (3.140). It is
convenient to view the 4-point function appearing on the RHS of (3.148) as a matrix element
in radial quantization,
h 4 | 3 (1) 2 (z, z̄)| 1 i, (3.150)
where h 4 | is the BPZ conjugate state representing 04 (1). In the rest of this section we
will assume the Hermiticity of the i ’s and do not distinguish the BPZ conjugation from
Hermitian conjugation.
Assuming that |z| < 1 for now, performing the OPE of 1 2 (or 3 4 ) is equivalent to
inserting a complete basis of states in radial quantization into (3.150). The latter can be
organized as a sum over Virasoro descendants, giving rise to
X X
h 4 | 3 (1) 2 (z, z̄)| 1 i = e e | ki
h 4 | 3 (1)L N L N
k,` |N |=|M |, |N
e |=|M
f|
ef
c, e
⇥ Dk` GN M (c, hk )GN M (e hk )h ` |(L M)
†
(L f)
M
†
2 (z, z̄)| 1 i,
(3.151)
k` NM
where D is the inverse matrix of the two-point function coefficient Dk` , and G (c, h) is
the inverse Gram matrix of Virasoro descendants of a weight h primary at level |N | = |M |.
Note that hk = h` , ehk = e
h` . Using the dilation symmetry on the 3-point function, we can
determine the z, z̄ dependence
† † h1 h2 +h` +|M | e
h1 e
h2 +e f|
h` +|M † †
h ` |(L M) (L f)
M 2 (z, z̄)| 1 i =z z̄ h ` |(L M) (L f)
M 2 (1)| 1 i.
(3.152)
The 3-point functions involved in (3.151) and (3.151) can be expressed in terms of the
structure constants,

h 4 | 3 (1)L NL N
e
e | ki = C43k ⇢(⌫4 , ⌫3 , L N ⌫k )e
e
⌫4 , ⌫e3 , L Ne ⌫ek ),
⇢(e
(3.153)
h ` |(L M)
†
(L f)
†
2 (1)| 1 i = C`21 ⇢(L M ⌫` , ⌫2 , ⌫1 )e
⇢( L Me
f⌫e` , ⌫e2 , ⌫e1 ).
M

Here ⌫i and ⌫ei denote holomorphic and anti-holomorphic primaries of weight hi and e hi
respectively. ⇢ is the three-point function of holomorphic Virasoro descendants, with the
structure constant stripped o↵,

⇢(⇠3 , ⇠2 , ⇠1 ) = h⇠3 |⇠2 (1)|⇠1 i = h⇠30 (1)⇠2 (1)⇠1 (0)i, (3.154)

where ⇠i are holomorphic Virasoro descendants of primaries ⌫i , with h⌫30 (1)⌫2 (1)⌫1 (0)i set
to 1. ⇢e is defined analogous for the anti-holomorphic Virasoro algebra. For given Virasoro

48
descendants ⇠i , ⇢(⇠3 , ⇠2 , ⇠1 ) is entirely fixed by conformal Ward identities as a polynomial
function in the weights hi and the central charge c. A detailed algorithm for computing ⇢
will be given in the next section.
We can now express the 4-point function as
X
h 4 | 3 (1) 2 (z, z̄)| 1 i = C43k Dk` C`21 Fc (h1 , h2 ; h3 , h4 ; hk |z)Fec (e
h1 , e
h2 ; e
h3 , e
h4 ; e
hk ; z̄),
k,`
(3.155)
where the function
X
h1 h2 +hk +|N | NM
Fc (h1 , h2 ; h3 , h4 ; hk |z) = z ⇢(⌫4 , ⌫3 , L N ⌫k )G (c, hk )⇢(L M ⌫ k , ⌫2 , ⌫1 )
|N |=|M |
(3.156)
is known as the holomorphic Virasoro conformal block for the sphere 4-point function in the
12 ! 34 OPE channel. A priori, the expansion (3.156) is defined in the unit disc |z| < 1.
We will see in the next section that up to an simple overall factor, Fc (· · · |z) can in fact be
analytically continued to the entire complex z plane, apart from a branch cut that extends
from z = 1 to 1; furthermore, it can be analytically continued through the branch cut to
other sheets.
We could alternatively decompose the 4-point function (3.148) in the 13 ! 24 OPE
channel, or the 14 ! 23 OPE channel. The compatibility of the conformal block decom-
positions in the di↵erent OPE channels, known as the crossing equation, is equivalent to
the associativity of the OPE, and impose highly nontrivial constraints on the spectrum of
primaries as well as the structure constants.

3.9 Appendix: Recurrence formulae and analytic property of con-

formal blocks

In practice, the Virasoro conformal block Fc can be evaluated efficiently to high accuracy
through Zamolodchikov’s recurrence relation. The idea is to consider the analytic continu-
ation of Fc in the central charge c. For fixed weights, the c ! 1 limit of Fc is finite and
reduces to the corresponding conformal block of the global conformal group SL(2, R), that
takes the form of a hypergeometric function
h1 h2 +h
lim Fc (h1 , h2 , h3 , h4 , h; z) = z 2 F1 (h + h2 h1 , h + h3 h4 ; 2h; z). (3.157)
c!1

On the other hand, Fc acquires a pole whenever the Gram matrix GM N degenerates. These
poles are located at the zeroes of the Kac determinant (3.125), namely when the internal
weight h is equal to hr,s (3.127) for some positive integers r, s. If we fix h at a generic value,

49
these poles occur when the central charge c is equal to
" p # 12
1 2 rs 1 + 2h + (r s)2 + 4(rs 1)h + 4h2
cr,s (h) = 1 + 6(br,s (h) + br,s (h) ) , br,s (h) ⌘
1 r2
(3.158)
for r 2, s 1.
The residue at the pole c = cr,s (h) only receives contribution from Virasoro descendants
of the form L N hrs , where X
h M
rs = rs L M ⌫h (3.159)
|M |=rs

is defined by the same set of coefficients M

rs that appear in the null descendant rs of ⌫hr,s
(3.129). In the h ! hr,s limit, the norm of hrs vanishes, with
h hrs | hrs i
lim ⌘ (Acrs ) 1 . (3.160)
h!hr,s h hr,s
There exists a closed form formula for Acrs ,
r s
1 Y Y
Acrs = (mb + nb 1 ) 1 , (m, n) 6= (0, 0), (r, s). (3.161)
2 m=1 r n=1 s
1
After rescaling by h hrs | hrs i 2 , hrs behaves like a Virasoro primary state in the h ! hr,s limit,
in the sense that it is annihilated by Ln for all positive n. The residue of the conformal block
as a meromorphic function of the internal weight h at h = hr,s is given entirely by the
contribution from the descendants of hrs ,
lim (h hr,s )Fc (h1 , h2 , h3 , h4 , h; z)
h!hr,s
X
= Acrs z h1 h2 +hr,s +rs+|N | ⇢(⌫4 , ⌫3 , L N rs )G
NM
(c, hr,s + rs)⇢(L M rs , ⌫2 , ⌫1 ).
|N |=|M |
(3.162)
Note that while rs is a null state, its 3-point function ⇢(⌫4 , ⌫3 , rs ) and ⇢( rs , ⌫2 , ⌫1 ), as
defined in (3.154), are nonzero. They are given by the “fusion polynomial”

rs h1
⇢(⌫1 , ⌫2 , rs ) = ⇢( rs , ⌫2 , ⌫1 ) = Pc
h2
r
Y 1 s
Y 1 1 1 (3.163)
1 + 2 + pb + qb 1 2 + pb + qb
= ,
p=1 r step 2 q=1 s step 2
2 2

where 1 , 2 are defined by hi ⌘ 14 (b + b 1 )2 14 2i . Finally, the key property that leads to a

recursive formula for the residue (3.162) is the factorization relation
⇢(L N rs , ⌫2 , ⌫1 ) = ⇢(L N ⌫hr,s +rs , ⌫2 , ⌫1 )⇢( rs , ⌫2 , ⌫1 ), (3.164)

50
where ⌫hr,s +rs representations a normalized primary state of the same weight as the null
state rs . The relation (3.164) is a consequence of the fact that rs behaves like a Virasoro
primary as far as Virasoro Ward identities are concerned. Putting these together, we can
express the RHS of (3.162) in terms of a conformal block with shifted internal weight,
 
c rs h1 rs h4
lim (h hr,s )Fc (h1 , h2 , h3 , h4 , h; z) = Ars Pc P Fc (h1 , h2 , h3 , h4 , hr,s + rs; z).
h!hr,s h2 c h3
(3.165)
Turning this into a residue in c at c = cr,s (h), we have
 
@cr,s (h) cr,s rs h1 rs h4
Resc!cr,s (h) Fc (h1 , h2 , h3 , h4 , h; z) = Ars Pcr,s P Fcr,s (h1 , h2 , h3 , h4 , h + rs; z).
@h h2 cr,s h3
(3.166)
Finally, taking into account of the large c limit (3.157), we arrive at the recurrence formula

Fc (h1 , h2 , h3 , h4 , h; z) = z h1 h2 +h 2 F1 (h + h2 h1 , h + h3 h4 ; 2h; z)
X   
1 @cr,s (h) cr,s rs h1 rs h4
+ Ars Pcr,s Pcr,s Fcr,s (h1 , h2 , h3 , h4 , h + rs; z).
r 2,s 1
c c r,s (h) @h h 2 h 3

(3.167)

v = 2⇡

v=0 v = 2⇡

Figure 1: The pillow geometry is the quotient T 2 /Z2 . The four points on the plane 0, z, 1, 1
are mapped to the Z2 fixed points w = 0, ⇡, ⇡(⌧ + 1), ⇡⌧ respectively.

As a power series expansion in z, Fc has radius of convergence 1, and is singular at z = 1.

The recurrence formula (3.167) makes it clear that Fc can be analytically continued to the
complex z-plane, except for a branch cut that extends from 1 to 1. This is not the full
story, however, as Fc can be analytically continued beyond the branch cut on the z-plane.
To fully exhibit the analytic property of the Virasoro conformal block, it is useful to map
the Riemann sphere with the four marked points 0, z, 1, 1 to the “pillow” geometry T 2 /Z2 ,
parameterized by the periodically valued complex coordinate w ⇠ w + 2⇡ ⇠ w + 2⇡⌧ that is

51
subject to the Z2 identification w ⇠ w. It is also conventional to define the “elliptic nome”
q = e⇡i⌧ , where ⌧ is related to z by

K(1 z) 1 1
⌧ =i , K(z) = 2 F1 ( , ; 1; z). (3.168)
K(z) 2 2

The explicit map from the coordinate u on the Riemann sphere to w on the pillow is
Z u
1 dx
w= 2
p , (3.169)
(✓3 (⌧ )) 0 x(1 x)(z x)
P
where ✓3 (⌧ ) = 1n= 1 q
n2
is one of the Jacobi theta functions. In the pillow frame, the four
primary operators are inserted at the corners w = 0, ⇡, ⇡⌧, ⇡(1 + ⌧ ), and the conformal block
can be thought of as the propagator on the cylinder
1
X
c c
L0
h 43 |q 24 | 12 i = an qh+n 24 , (3.170)
n=0

where | 12 i and | 43 i are the states created by a pair of primaries inserted at two corners
of the pillow, projected onto the representation space of the Virasoro algebra with primary
weight h, and the coefficients an are functions of the weights hi , h and the central charge
c. (3.170) di↵ers from the sphere 4-point block by a conformal anomaly factor (due to the
Weyl transformation relating the flat metric on the z-plane to the flat metric on T 2 /Z2 ),
c c c c
4(h1 +h2 +h3 +h4 ) h1 h2 h 3 h4 L0
Fc (h1 , h2 , h3 , h4 , h; z) = (✓3 (⌧ )) 2 z 24 (1 z) 24 h 43 |q 24 | 12 i.
(3.171)

0 1 ∞

Figure 2: The complex z-plane with a branch cut along [1, 1) is mapped to the eye-shaped
region on the unit q disc. The Virasoro conformal block can be analytically continued to the
entire q disc.

52
 The pillow representation of the conformal block (3.171) exhibits several important prop-
erties. Firstly, as a series expansion in q, (3.170) converges on the unit q-disc, and is holo-
morphic on the q-disc away from the origin. In other words, the domain of analyticity of Fc
covers the entire complex z-plane and more: the complex z-plane only maps to an eye-shaped
region on the q-disc. Note that q can be expanded in z as

z z2 21z 3
q= + + + ··· , (3.172)
16 32 1024
but the q-series converges much faster than the z-series at a generic point on the z-plane.
Secondly, the matrix element (3.170) has the property that in the special case h1 = h4 ,
h2 = h3 , the coefficients an are norms of Virasoro descendants of | 12 i and therefore non-
negative (provided that the OPE involved is compatible with unitarity).
Thirdly, (3.170) simplifies in the h ! 1 limit, with c and hi ’s held fixed:
1
Y
c c 1
L0 h
lim h 43 |q 24 | 12 i = (16q) 24 (1 q2n ) 2 . (3.173)
h!1
n=1

Heuristically, this can be understood based on the intuition that in the large h limit, the
external operators inserted at the corners of the pillow are unimportant; the square of the
pillow block becomes the same as the pillow block of two copies of the CFT with Z2 twist
fields inserted at each corner that exchanges the two CFTs (see section 6.3), which up to a
conformal anomaly factor is equal to the torus character of the internal primary of weight h.
The large h limit (3.173) together with the residues (3.165) at the poles in h allow us to
derive a recurrence formula in h rather than in c, of the form
c c c
4(h1 +h2 +h3 +h4 ) h1 h2 h3 h4
Fc (h1 , h2 , h3 , h4 , h; z) = (✓3 (⌧ )) 2 z 24 (1 z) 24
1
Y
h c 1 (3.174)
⇥ (16q) 24 (1 q2n ) 2 Hc (h1 , h2 , h3 , h4 , h; q),
n=1

where Hc obeys
X  
qrs c rs h1 rs h4
Hc (h1 , h2 , h3 , h4 , h; q) = 1 + Ars Pc Pc Hc (h1 , h2 , h3 , h4 , hr,s + rs; q).
r,s 1
h h r,s h 2 h 3

(3.175)
In practice we can use this relation to evaluate Fc efficiently, by first using (3.175) as a set
of linear equations for Hc (h1 , h2 , h3 , h4 , hr,s + rs; q) and solve for the latter over a finite set
of r, s subject to rs  n, up to order qn , and then using (3.175) to generate Hc for generic
internal weight h up to order qn .

53
3.10 Modular invariance

There is a natural way to extend the definition of correlation functions of local operators in
a 2D CFT to any compact oriented surface ⌃ equipped with a Riemannian metric, at least
in the case c = e
c where gravitational anomalies are absent, as follows.
The surface ⌃ can be covered with coordinate charts Ui , parameterized by complex
coordinates (zi , z̄i ), such that the metric takes the Hermitian form

ds2 = 2gzi z̄I dzi dz̄i (3.176)

on each Ui . It follows that the transition map

zi = fij (zj , z̄j ), z̄i = f¯ij (zj , z̄j ) (3.177)

on Ui \ Uj must be such that fij is holomorphic, likewise f¯ij anti-holomorphic, so that

gzj z̄j = |@zj fij |2 gzi z̄i . (3.178)

For any given chart Ui , We can put the metric on Ui to be the flat Euclidean metric by
a Weyl transformation, under which the CFT partition function acquires a Weyl anomaly
factor (3.120). Up to the Weyl anomaly, which depends only on the central charge c and
not any further details of the CFT, the partition function depends only on the equivalence
class of metrics under Weyl transformation. The latter is the same as the equivalence class
of 1-dimensional complex manifolds, defined as the equivalence class of coordinate charts
Ui covering ⌃ with compatible holomorphic transition maps fij on their overlaps. A 1-
dimensional complex manifold is also known as a Riemann surface.
Likewise, under a Weyl transformation of the metric, a correlation function of local
operators O1 , · · · , On on ⌃ acquires the same Weyl anomaly factor, provided that the Weyl
transformation is trivial in a small neighborhood of the local operators. For our purpose, it
suffices to assume that each operator Oa is inserted at the origin of a disc Da ⇢ ⌃ on which
the metric is flat, so that there is no ambiguity with regard to the state created by Oa in
radial quantization on Da .
If ⌃ is a surface of genus g > 1, we can decompose it into 2g 2 “pairs of pants”,
joined along their boundary circles Ck , k = 1, · · · , 3g 3. If ⌃ has genus g = 1 then it
suffices to represent ⌃ as a cylinder with two boundary circles identified. The metric in
a neighborhood of each circle Ck , up to a Weyl transformation, can put in the form of a
standard Euclidean cylinder parameterized by (⌧, ), ⇠ + 2⇡, ⌧ 2 ( ✏, ✏) for some ✏ > 0,
such that Ck corresponds to the equal time slice ⌧ = 0. We can then insert a complete basis
of states along Ck , and express the correlation function hO1 · · · On i⌃ in terms of correlation
functions of local operators on pairs of pants or cylinders, with appropriate CFT states

54
assigned to the boundary circles. The correlation functions on pairs of pants or cylinders are
conformally equivalent to correlators on the sphere or the plane, which are then determined
by the structure constants as described in section 3.8.
Equivalently, we may viewed the surface ⌃ with n local operator insertions as a surface
with n “punctures”. We can decompose the punctured surface into 2g 2 + n pairs of
pants, with n of the boundary circles shrunk to points. The n-point function on ⌃ is then
decomposed into the product of 2g 2 + n three-point functions, suitably Weyl transformed,
summed over the basis of states inserted along 3g 3 + n circles.
The consistency in defining correlation functions on ⌃ requires that di↵erent pair-of-pants
decompositions of the n-punctured surface ⌃ results in the same answer for the correlation
function. If ⌃ is topologically a sphere, this consistency condition is equivalent to the
crossing equation for the conformal block decomposition of sphere 4-point functions, or the
associativity of the OPE.

Figure 3: The crossing move relating two di↵erent pair-of-pants decompositions of the four-
holed sphere.

Figure 4: The crossing move relating two di↵erent pair-of-pants decompositions of the one-
holed torus.

If ⌃ is a torus with one puncture, i.e. g = n = 1, we can cut the torus along either the ↵
circle or the circle as shown in Figure 4, and decompose the torus one-point function into
a sum of sphere three-point functions. More explicitly, as will be discuss in detail in section
4.7.2, the torus may be parameterized by the complex coordinate z, subject to periodicity
z ⇠ z + 2⇡ ⇠ z + 2⇡⌧ , for a complex modulus parameter ⌧ = ⌧1 + i⌧2 with ⌧2 > 0. We can

55
write the 1-point function of an operator O(z, z̄) on the torus of modulus ⌧ , equipped with
the Euclidean metric ds2 = dzdz̄, as6
2⇡⌧2 H+2⇡i⌧1 P c e0 e
c
hO(z, z̄)iT 2 (⌧ ) = TrH O(z, z̄) e = TrH O(z, z̄) e2⇡i⌧ (L0 24
) 2⇡i¯
⌧ (L 24
)
, (3.179)

where H is the Hilbert space of the CFT on the unit circle, H and P are the energy and
e0 via (3.102). O(z, z̄) on the
momenta on the circle, related to Virasoro generators L0 and L
RHS is viewed as an operator inserted at the point z on the cylinder.
It follows from translation symmetry along the torus that hO(z, z̄)iT 2 (⌧ ) ⌘ fO (⌧, ⌧¯) is
independent of z, z̄. Assuming that H only consists of integer spin states, we have

fO (⌧ + 1, ⌧¯ + 1) = fO (⌧, ⌧¯). (3.180)

The equivalence of cutting along ↵ versus circle, on the other hand, implies
⇡i e e
⌧ ) = e 2 (h
fO ( 1/⌧, 1/¯ h)
( i⌧ )h (i¯
⌧ )h fO (⌧, ⌧¯), (3.181)

where (h, e
h) are the conformal weights of O. On the RHS, i⌧ has positive real part, and
its power is defined with a branch cut chosen to lie on the left half complex plane. (3.181)
can also be understood from the fact that the torus of modulus ⌧ is conformally equivalent
to the torus with modulus ⌧ 0 = 1/⌧ , via the map z 7! z 0 = z/⌧ . We will refer to the
conditions (3.180) and (3.181) as the modular invariance of the torus one-point function.

Figure 5: A sequence of simple crossing moves that relate two di↵erent pair-of-pants decom-
positions of a genus two surface.

It can be shown that any two di↵erent pair-of-pants decompositions of the n-punctured
surface ⌃ can be related by a sequence of simple crossing moves: either the crossing move
on the 4-punctured sphere (Figure 3), or that of the 1-punctured torus (Figure 4). A genus
two example is shown in Figure 5. Consequently, the consistency of the n-point function
on ⌃ follows from the crossing invariance of all sphere 4-point functions together with the
modular invariance of all torus 1-point functions.
6
The relation between the torus 1-point function and the trace on the RHS of (3.179) will be modified in
ghost CFTs. For instance, in the bc system to be discussed in section 4.2, one must insert ( )Nbc into the
trace, where Nbc is the bc ghost number.

56
 Modular invariance puts highly nontrivial constraints on a CFT. For instance, while free
boson CFT is a modular invariant theory, its k = 0 sector, generated by operators of the form
@ n @¯m X and their normal order products, closes under OPE but is not modular invariant.
The integer spin (NS, NS) sector states of the free fermion CFT close under OPE, but is
not modular invariant by itself. A modular invariant CFT with c = ec = 12 can be constructed
by combining the integer spin (NS, NS) sector and (R, R) sector states. This results in the
Ising CFT, which will be discussed in section 6.1. Such a construction is known as the
diagonal Gliozzi-Scherk-Olive (GSO) projection.
The worldsheet CFT of type II superstrings, on the other hand, is a non-unitary theory
defined with the chiral GSO projection on the free fermions as well as superconformal ghosts,
in which a subset of integer spin states in the (NS, NS), (NS, R), (R, NS), (R, R) sectors
are included in the spectrum. This will be discussed in section 10.4.

57