Introduction to Conformal Field Theory

PoS(Modave VIII)001
Antonin Rovai∗†
Université Libre de Bruxelles and International Solvay Institues
E-mail: [email protected]

An elementary introduction to Conformal Field Theory is provided. We start by reviewing sym-

metries in classical and quantum field theories and the associated existence of conserved currents
by Noether’s theorem. Next we describe general facts on conformal invariance in d dimensions
and finish with the analysis of the special case d = 2, including an introduction to the operator
formalism and a discussion of the trace anomaly.

Eighth Modave Summer School in Mathematical Physics

26th august - 1st september 2012
Modave, Belgium

∗ Speaker.
† FNRS Research Follow

c Copyright owned by the author(s) under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike Licence. http://pos.sissa.it/
Introduction to Conformal Field Theory Antonin Rovai

Contents

Foreword 2

Introduction 3

1. Symmetries and Conservation laws 3

PoS(Modave VIII)001
1.1 Definitions 4
1.2 Noether’s theorem 6
1.3 The energy-momentum tensor 8
1.4 Consequences for the quantum theory 10

2. Conformal invariance in d dimensions 11

2.1 General considerations and algebra 11
2.2 Action of conformal transformations on fields 15
2.3 Energy-momentum tensor and conformal invariance 16
2.4 Conformal invariance in quantum field theory and Ward identities 16

3. Conformal invariance in two dimensions 19

3.1 Identification of the transformations and analysis 19
3.2 Ward identities 23
3.3 Operator Formalism 26
3.3.1 Radial order and mode expansion 26
3.3.2 Operator Product Expansion 27
3.3.3 Virasoro algebra 29
3.4 Trace anomaly 30

References 32

Foreword

These lectures notes were written for the eighth edition of the Modave Summer School in
Mathematical Physics. During this edition, there were several lectures [1, 2] relying on Conformal
Field Theory basics, and thus self-consistency of the school required an elementary introduction to
this vast subject. The present lectures were scheduled at the very beginning of the school, right after
the gong show on the first day and in the morning of the second day. People were then smoothly
taken from very basics to advanced research topics throughout the week, overall in the relaxed
atmosphere of the charming village of Modave, Belgium. This eighth edition was a real success,
and I hope this school will continue to exist for many more years to come.

2
Introduction to Conformal Field Theory Antonin Rovai

Introduction

These lecture notes consist of an elementary introduction to Conformal Field Theory (CFT).
This class of theories have the property that they are invariant under so called conformal transfor-
mations, that may be seen as local scale tranformations. Conformal invariance is a vast topic, and
in these notes we will be able to cover only the very basic elements. Historically, conformal invari-
ance was first used to study statistical systems at a critical point, where scale invariance follows
from the divergence of the correlation length. Since then, the use of CFTs has spread to the field
of fundamental interactions: it is at the heart of string theory, the world-sheet theory being itself

PoS(Modave VIII)001
a two-dimensional CFT; in two dimensions, when quantum gravity is coupled to a matter CFT,
the resulting effective gravitational action is the Liouville action which is a CFT; finally, the best
understood example of a gauge/gravity duality that we know of involves a CFT on the field theory
side, namely N = 4 super-Yang-Mills in four dimensions.
The plan of these notes is as follows. We start very slowly in section 1 with basic results
concerning symmetries in field theory, both at the classical and at the quantum level. The discussion
will be quite general and several examples will be presented, with special emphasis on the energy-
momentum tensor, which play a crucial rôle in the following. In section 2, we introduce and study
conformal transformations in d dimensions. We discuss finite transformations as well as the algebra
of the conformal group, and we explore some simple consequences of conformal invariance on two-
and three-point functions. Finally in section 3 we focus on the special case of d = 2. We introduce
primary fields, study Ward identities and the associated consequences for the energy-momentum
tensor, operator formalism and operator product expansion. We end these lectures with a discussion
of the trace anomaly.
For (much) more details on Conformal Field Theory, we refer the interested reader to the
abundant literature. The standard reference for a modern treatment is the very nice book by Di
Francesco, Mathieu and Sénéchal [3], on which most of the present lectures is based. See also the
lecture notes by Ginsparg [4]. For a more old-fashioned presentation, see the lectures by Coleman
[5]. Note also that Polchinski’s book [6] and Tong’s lecture notes [7] on String Theory both include
a quick introduction to conformal invariance in two dimensions. See also the chapter 15 in [8] for
a nice introduction to more advanced topics.
Our conventions are the following: unless explicitly stated, we work in the euclidean d-
dimensional flat space Rd . Space-time points are denoted by x = (xµ ) with µ = 1, . . . , d, the last
component xd being Euclidean time (except at the end of chapter 3, from section 3.3.1). Vector and
form indices are lowered and raised using the flat metric, and we sum over repeated indices. We
use natural units in which c = h̄ = 1 and we use the symbol ≡ for an equality valid by definition.

1. Symmetries and Conservation laws

Symmetries lie at the heart of our modern conception of physics. It is therefore very important
to understand how we formulate the symmetry properties of a given theory and their consequences
on observables. In this section we quickly review some elementary notions along these lines,
preparing the ground for the analysis of conformal invariance.

3
Introduction to Conformal Field Theory Antonin Rovai

1.1 Definitions
We consider a classical theory for some fields, collectively denoted as Φ, which are functions
on a space-time manifold that we shall take to be flat Rd . The dynamics of the fields Φ is fixed by
a Lagrangian density L (Φ, ∂µ Φ) or by the action S[Φ] defined by
Z
S[Φ] = dd x L (Φ, ∂µ Φ) . (1.1)

Consider a map x 7→ x0 (x), where x0 ∈ Rd is some invertible function of x ∈ Rd , together with some
transformation of the fields Φ 7→ Φ0 defined by

PoS(Modave VIII)001
Φ0 (x0 ) = F(Φ(x)) , (1.2)

for some function F. Under such a transformation, the action will generally be modified: S 7→ S0 ,
with S0 defined by the equation S0 [Φ0 ] = S[Φ], and the transformation is a symmetry if S = S0 .
Let us consider some examples.

1. Translations are simply defined by

x0 = x + a , (1.3)

where a ∈ Rd . Most of the fields Φ that we consider are scalars under translation, that is, F
reduces to the identity:

F(Φ(x)) = Φ(x) = Φ0 (x0 ) = Φ0 (x + a) . (1.4)

2. Rotations are given by

x0µ = Rµ ν xν , (1.5)
where the matrix R is such that
δµν Rµ λ Rν ρ = δλ ρ . (1.6)
The function F corresponding to rotations is characterised by the representation that we
choose for the field Φ. For example, for a scalar field ϕ, the transformation is

ϕ 0 (R · x) = ϕ(x) (1.7)

where we use the common notation (R·x)µ = Rµ ν xν . For a vector field V µ the transformation
is
V 0µ (R · x) = Rµ ν V ν (x) , (1.8)
and so on for tensors of various ranks. In the quantum theory, we are also interested in
spinor representations of the rotation group, that is, representations up to a phase. We shall
however continue to write “representation of rotations” instead of “representation of spinor
representation of the rotation group.”
For a field Φ transforming in any representation L of the rotation group, we write the corre-
sponding transformation function F as

Φ0 (R · x) = LR [Φ(x)] , (1.9)

4
Introduction to Conformal Field Theory Antonin Rovai

that is, LR is the linear operator representing the transformation R.

Note that in Minkowskian space-time, we are more interested in transformations that pre-
serve the Minkowski metric ηµν , namely the Lorentz transformations. As far as these lec-
tures are concerned, the differences between the rotation group and the Lorentz group will
not be important.

3. Scale transformations (or dilatations) read

x0 = λ x (1.10)

PoS(Modave VIII)001
where λ 6= 0. We define Φ0 by
Φ0 (λ x) = λ −∆ Φ(x) , (1.11)

where the number ∆ is called the scaling dimension of Φ. In this case the jacobian of the
transformation is |∂ x0 /∂ x| = λ d and hence
Z
S0 [Φ0 ] = dd x0 L (Φ0 (x0 ), ∂µ0 Φ0 (x0 )) (1.12)
Z
= λd dd x L (λ −∆ Φ(x), λ −1−∆ ∂µ Φ(x)) . (1.13)

In the case of a free theory for a massless scalar field ϕ, L is given by

1
L (ϕ, ∂µ ϕ) = ∂µ ϕ∂ µ ϕ , (1.14)
2
and the action S is invariant if and only if the scaling dimension ∆ is given by

d −2
∆= · (1.15)
2

Scaling dimension versus mass dimension

As far as classical field theory is concerned, the value of ∆ for a given field will always coincide
with the opposite of its mass dimension, itself determined by the requirement that the action is
dimensionless (in units where c = h̄ = 1 that we shall use throughout these notes). It should however
be clear that the requirement of scale invariance is not a consequence of the fact that the action
is dimensionless, since under a dilatation, only fields and coordinates are changed, whereas all
other quantities (like coupling constants or masses) are held fixed. For instance, if we add a mass
term m2 ϕ 2 /2 to the lagrangian (1.14) for the scalar field ϕ, then under the dilatation x0 = λ x the
transformed action (1.12) is given by

m2
 
1
Z
0 0
S [ϕ ] = d xd
∂µ ϕ∂ µ ϕ + λ 2 ϕ 2 6= S[ϕ] . (1.16)
2 2

Mass terms are thus prohibited in dilatation invariant theories, as well as dimensionful coupling
constants.

5
Introduction to Conformal Field Theory Antonin Rovai

1.2 Noether’s theorem

Let us consider a continuous transformation, that is, the map x 7→ x0 is characterised contin-
uously by some parameters ωa . We can then consider a transformation “close to identity,” that is,
for
δx ,
x0 = x + ωa (1.17)
δ ωa
we can write
δ F(Φ(x)) ,
F(Φ(x)) = Φ(x) + ωa (1.18)
δ ωa

PoS(Modave VIII)001
where summation on the index a is understood. We define the generators Ga by

δω Φ(x) = Φ0 (x) − Φ(x) = −iωa Ga Φ(x) , (1.19)

and hence
δ xµ δF
iGa Φ = ∂µ Φ − · (1.20)
δ ωa δ ωa
The factor of i in the definition (1.19) is introduced for later convenience.
We now identify the parameters and the generators in our examples.

1. Translations: the parameters ωa for an infinitesimal translation are the components aµ of the
infinitesimal vector a defining the infinitesimal transformation, and thus the index a is in this
case a space-time index µ: ωµ = aµ . Using (1.4) we thus find that

δ xµ δF
= δ µν , = 0. (1.21)
δ ων δ ωµ
The generator, that we write Pµ and define by equation (1.20), reads

Pµ = −i∂µ . (1.22)

2. A infinitesimal rotation is characterised by an antisymmetric matrix ωµν = −ων µ and is

given by
x0µ = xµ + ω µ ν xν . (1.23)
Formula (1.17) then yields the following variation:
δ xµ 1
= (δ µν xρ − δ µρ xν ) . (1.24)
δ ωνρ 2
For a field Φ transforming under a general representation L as in (1.9), the effect of an
infinitesimal rotation is of the form
i
LR [Φ] = Φ − ωµν Sµν [Φ] , (1.25)
2
for some operators Sµν = −Sν µ representing the rotation algebra, the numerical factors being
introduced for future convenience. Using (1.24) and (1.25), the generators Lµν for rotations
and defined in (1.20) are thus given by

Lµν = i(xµ ∂ ν − xν ∂ µ ) + Sµν . (1.26)

6
Introduction to Conformal Field Theory Antonin Rovai

3. For scale transformations x 7→ x0 = λ x, the transformation parameter is λ . We write λ = 1+ε

for small ε, and equations (1.17) and (1.18) then gives δ x0µ /δ ε = xµ . Equation (1.18) then
yields
F(Φ(x)) = Φ0 (x0 ) = (1 − ∆ε)Φ(x) , (1.27)
where we used δ F/δ ε = −∆ which is a consequence of the definition (1.11). If we call D
the dilatation generator, then (1.27) yields

iD = x · ∂ + ∆ , (1.28)

PoS(Modave VIII)001
where we used the convenient notation x · ∂ = xµ ∂µ .
µ
For any continuous transformation, we can define the associated canonical current densities ja by
∂L ∂L δF
  ν
δx
∂ν Φ − δν L
µ µ
ja = − · (1.29)
∂ (∂µ Φ) δ ωa ∂ (∂µ Φ) δ ωa
When the transformation under consideration is a symmetry and when the equations of motion are
satisfied, the currents (1.29) are conserved,

∂µ jaµ = 0 . (1.30)

To show this, it is useful to assume for a moment that the parameters ωa are arbitrary (but small)
space-time functions ωa (x). Using the formula det(1 + J) = 1 + tr J + O(J)2 where J = ∂ x0 /∂ x is
the jacobian matrix for the map x 7→ x0 , we find
∂ x0
 
δ xµ
= 1 + ∂ µ ωa + O(ω)2 . (1.31)
∂x δ ωa
Working to first order in ωa , we find after some algebra the following formula for the variation
δ S(Φ) ≡ S0 (Φ) − S(Φ),
!
δ xµ ∂L δ xρ ∂L δF
Z
d
δ S(Φ) = d x ∂µ ωa L−
∂ µ ωa ∂ρ Φ +
∂µ ωa . (1.32)
δ ωa ∂ ∂µ φ δ ωa ∂ ∂µ Φ δ ωa

Dropping some total derivative terms, we find that

Z
δS = dd x ∂µ jaµ (x)ω(x) , (1.33)

µ
where the currents ja are given by (1.29). When the equations of motion are satisfied, the action is
by definition stationary under arbitrary variations and thus δ S = 0. Since the functions ωa in (1.33)
µ
are arbitrary, this is possible only if ∂µ ja = 0, completing the proof.
µ
To any current ja we associate a conserved charge Qa by the formula
Z
Qa = dd−1 x jad , (1.34)
Σ
µ
where Σ is the xd = constant hyper surface. When the current ja is conserved, the charge Qa is
also conserved in the sense that dQa /dxd = 0. Note that several currents may give rise to the same
charge. For example, if we define
j˜aµ = jaµ + ∂ν Bνa µ (1.35)

7
Introduction to Conformal Field Theory Antonin Rovai

for Ba = −Ba , then conservation of ja implies that j˜a is conserved as well. Moreover, the charge
µν νµ µ µ

associated with j˜a coincides with Qa , as the integral of ∂ν Bνd

µ id
a = ∂i Ba over Σ vanishes. Terms such
νµ
as ∂ν Ba are called improvement terms. We will make use of such terms to define a symmetric
stress-energy tensor, see equation (1.40).

1.3 The energy-momentum tensor

For many reasons, the energy-momentum (or stress) tensor plays a crucial rôle in quantum
field theory, and in our case this will be shortly obvious. It is defined as the current corresponding
to translation invariance of the theory. For fields Φ transforming as scalars under translations, F

PoS(Modave VIII)001
reduce to the identity and thus
∂L
∂ν Φ − δν L ,
µ µ
Tc ν = (1.36)
∂ (∂µ Φ)
where the subscript c indicates that this is the canonical energy-momentum tensor. Notice that the
µ
index a in the general definition of the current ja in (1.29) is now replaced by a space-time index
µ
µ, since translations are parameterised by a space-time vector. The associated current Tc ν , defined
in (1.36), thus carries two space-time indices.
The associated conserved charges also carry a space-time index, and are the components of
the energy-momentum vector Pµ : Z
Pµ = dd−1 x Tc dµ . (1.37)
Σ
For future reference, let us quickly write the currents associated to Lorentz and dilatation
invariance. From the definition (1.29) and the formulas derived in the examples of sections 1.1 and
1.2, we find the following expressions for the currents:
∂L
j µνρ = Tcµν xρ − Tcµρ xν + i Sνρ Φ (rotations) , (1.38)
∂ (∂µ Φ)
µ µ ν ∂L
jD = Tc νx + ∆Φ (dilatations) . (1.39)
∂ (∂µ Φ)
µν
In general, the canonical energy-momentum tensor Tc is not symmetric under the exchange
of µ with ν. In a Lorentz invariant theory however, it is however always possible to define a
µν
symmetric energy-momentum tensor by adding to Tc a suitable improvement term, following
µν
the discussion below (1.35). The resulting symmetric energy-momentum tensor TB is called the
Belinfante stress tensor, and is given by
µν
TB = Tcµν + ∂ρ Bρ µν , (1.40)

where the improvement term Bρ µν defined by

∂L ∂L ∂L
 
µρν i νρ µν µρ
B = S Φ+ S Φ+ S Φ . (1.41)
2 ∂ (∂µ Φ) ∂ (∂ρ Φ) ∂ (∂ν Φ)
µν
The new currents TB are conserved since Bρ µν is antisymmetric under the exchange of ρ and µ
thanks to the fact that the Lorentz generators Sµν are themselves antisymmetric. Moreover, the
current (1.38) associated to Lorentz invariance can be written as
µν µρ
j µνρ = TB xρ − TB xν . (1.42)

8
Introduction to Conformal Field Theory Antonin Rovai

µν µν
As a consequence, the constraints ∂µ j µνρ = 0 and ∂µ TB = 0 imply the symmetry of TB :
µν νµ
TB = TB . (1.43)

Let us remark that the Belinfante stress tensor TB µν is symmetric only when the equations of motion
µν
are satisfied, because we used the conservation laws ∂µ j µνρ = 0 and ∂µ TB = 0 to show (1.43).

An alternative definition of the energy-momentum tensor

We now motivate an alternative definition of the momentum tensor that is automatically sym-

PoS(Modave VIII)001
metric and generalises to curved space-times. Assume the energy-momentum tensor T µν has been
made symmetric. Under an arbitrary infinitesimal change of coordinates

x 7→ x + ε(x) , (1.44)

the infinitesimal variation of the action δ S is given by (1.32),

1
Z Z
d
δS = − d xT µν
∂µ εν = − dd x T µν (∂µ εν + ∂ν εµ ) . (1.45)
2
Now consider the generalization of this theory on a curved space-time of metric g, and let S[Φ; g]
be a diffeomorphism invariant action such that it reduces to our original action S when restricted
on flat space1 where the metric g is the identity matrix that we write δ ,

S[Φ; δ ] = S[Φ] . (1.46)

The action S[Φ; g] is by definition invariant under the infinitesimal diffeomorphism (1.44). Evalu-
ating its variation δ S on flat space, the requirement δ S[Φ, g] = 0 implies
 
δ S[Φ; g] 1 µν
Z
d
δ S[Φ; g] g=δ = d x δ gµν − T (∂µ εν + ∂ν εµ ) = 0 . (1.47)
δ gµν g=δ 2

Plugging the variation of the metric around flat space δ gµν = −∂µ εν − ∂ν εµ into (1.47) yields

δ S[Φ; g]
T µν = −2 · (1.48)
δ gµν g=δ

Formula (1.48) suggest the following definition for the energy-momentum tensor in curved space-
times:
2 δ S[Φ; g]
T µν (g) ≡ − √ (curved space-time) , (1.49)
g δ gµν
the factor involving the determinant of the metric being introduced in order to have a tensor (rather
than a density). Loosely speaking, the energy-momentum tensor captures the variation of the theory
under an arbitrary variation of the metric. Notice that it is formula (1.49) that is used to define the
energy-momentum tensor in general relativity, where Tµν appears as a source term in Einstein’s
equations.
1 The action S[Φ; g] is not uniquely defined by the condition (1.46), as any term that vanishes on flat space may be
added to it.

9
Introduction to Conformal Field Theory Antonin Rovai

1.4 Consequences for the quantum theory

We now study symmetries in quantum field theories. Since observables are typically expressed
in terms of correlation functions of the fields (or operators) of the theory, it is natural to define a
symmetry of a quantum theory under some transformation x 7→ x0 with the associated transforma-
tion of fields as a condition on correlators. To be specific, the quantum theory is said to be invariant
under x 7→ x0 , Φ 7→ Φ0 with Φ0 (x0 ) = F(Φ(x)) if and only if

hΦ1 (x01 ) · · · Φn (x0n )i = hF(Φ1 (x1 )) · · · F(Φn (xn ))i , (1.50)

PoS(Modave VIII)001
where Φ1 (x1 ), . . . , Φn (xn ) are fields (or operators) of the theory evaluated at some space-time points
x1 , . . . , xn ∈ Rd .
It should be kept in mind that if the classical theory has some symmetry, then in general this
does not mean that the quantized theory also has this symmetry, in the sense of (1.50). When the
symmetry is broken quantum mechanically, we say that the classical symmetry is anomalous. We
shall encounter an example of anomaly in section 3.4. In this section (and in most of these lectures),
we assume that the symmetries that we consider are not anomalous.
Let us write the infinitesimal version of the condition (1.50) for a continuous symmetry. Using
the path-integral formalism for example, it is a simple matter to show that if ωa is lifted to an
arbitrary function ωa (x), then
Z
δ hΦ(x1 ) · · · Φ(xn )i = dd x ∂µ h jaµ (x)Φ(x1 ) · · · Φ(xn )iωa (x)
Z n
= −i dd x ωa (x) ∑ hΦ(x1 ) · · · Ga Φ(xi ) · · · Φ(xn )iδ d (x − xi ) , (1.51)
D i=1

where the domain D ⊂ Rd contain all the points xi , i = 1, . . . , n at which the fields are evaluated.
Since this is valid for any value of the parameters ωa , we find the so-called Ward identities
n
∂
h
∂ xµ a
j µ
(x)Φ(x1 ) · · · Φ(xn )i = −i ∑ δ d (x − xi )hΦ(x1 ) · · · Ga Φ(xi ) · · · Φ(xn )i . (1.52)
i=1

Equations (1.52) are the quantum versions of the classical conservation law (1.30). Loosely speak-
µ
ing, one can say that in the quantum theory, the currents ja fail to be conserved when inserted in
correlation functions because of the terms on the right hand side of (1.52). Such terms, that are
µ
non-zero (and singular) only when the current ja is evaluated at one of the points xi , are called
contact terms.
µν µν νµ
Similarly to the classical case, the addition of an improvement term ∂ν Ba with Ba = −Ba
µ
to the current ja as in (1.35), has no consequence on the Ward identities (1.52). Indeed, thanks
to the antisymmetry property of Ba under the exchange on µ and ν, we can replace ja by j˜a =
µν µ µ
µ µν
ja + ∂ν Ba into equation (1.52) without any further changes.

Definition of the energy-momentum tensor in the quantum theory

To define the energy-momentum tensor in the quantum theory on a general curved space-time,
one could simply use the definition (1.49) as a statement between operators. However, if one wish
to keep the spirit of this definition, equation (1.49) should be adapted because the classical action

10
Introduction to Conformal Field Theory Antonin Rovai

S doesn’t contain all the quantum effects of the theory. To state this more precisely, let us consider
the path-integral expression for the partition function Z[g],
Z
Z[g] = DΦg e−S[Φ;g] . (1.53)

As indicated explicitly in (1.53), the formal path-integral measure DΦg is metric-dependent, as a

consequence of diffeomorphism invariance of the partition function Z[g]. For this reason, Z[g] will
receive essentially two contributions when the metric g is perturbed: one from the usual variation
of the action S[g], and a new one from the variation of the measure DΦg . Therefore, if one wish

PoS(Modave VIII)001
to keep the idea that the energy-momentum tensor captures the response of the theory under a
perturbation of the metric, it is very natural to define it through the equation
 
1 d √
Z Z Z
−S[Φ;g+δ g]
Z[g + δ g] = DΦg+δ g e = DΦg 1 + d x g T δ gµν e−S[Φ;g] ,
µν
(1.54)
2

where δ g is an arbitrary, small perturbation of the metric. If we define as usual the generating
functional W by
W [g] = − ln Z[g] , (1.55)
then the definition of T µν is equivalent to
2 δW [g]
hT µν i = − √ · (1.56)
g δ gµν

We will use equation (1.56) when we consider the trace anomaly in section 3.4, and it will also be
very useful in other lectures of this school [1]. Formula (1.56) and (1.49) are quite analogous, but it
should now be clear that in (1.56), the subtle quantum effects contained in the path-integral (1.53)
have been taken into account. Of course, if we consider the classical limit of our theory, then the
generating functional W goes to the classical action S of the theory, and we recover (1.49) from
(1.56).

2. Conformal invariance in d dimensions

In this section we define conformal transformations in any dimension d, and explore their basic
properties.

2.1 General considerations and algebra

Consider a field theory on a d-dimensional space-time with metric g, that may be curved or
not. A conformal transformation is a smooth, invertible map x 7→ x0 (x) such that

∂ x0ρ ∂ x0λ
gρλ (x0 ) = Λ(x)gµν (x) , (2.1)
∂ xµ x ∂ x ν
x

together with some field transformation Φ 7→ Φ0 to be specified later and where Λ is some strictly
positive function.
Let us insist on the fact that under a conformal transformation, the metric is not changed,
g 7→ g. As a side remark, let us mention that conformal transformations are sometimes defined as

11
Introduction to Conformal Field Theory Antonin Rovai

diffeomorphisms that leave the metric unchanged up to a factor. This formulation is misleading
since the metric is not changed under a conformal transformation, while for diffeomorphisms it
is. Since we measure distances before and after a conformal transformation with the same metric,
distances between points change, because the points have moved. On the other hand, distances are
invariant under a diffeomorphism. Moreover, a diffeomorphism-invariant theory is clearly not au-
tomatically conformally invariant: conformal transformations are not a subset of diffeomorphisms.
We will now study the consequences of (2.1) on the map x 7→ x0 for flat space, gµν = δµν .
The set of conformal transformations obviously form a group called the conformal group. Let us
already note that this group should have the Euclidean group as a subgroup,2 as it corresponds to

PoS(Modave VIII)001
the special case Λ(x) = 1. Otherwise, the Jacobian |∂ x0 /∂ x| of the conformal transformation is
Λ(x)−d/2 , as can be seen by evaluating the determinant of both sides of equation (2.1).
Consider an infinitesimal transformation x 7→ x + ε(x). Setting f = 1 − Λ, we find the con-
straint
∂µ εν + ∂ν εµ = f δµν , (2.2)

whose contraction with δ µν yields

2
f (x) = ∂ ·ε . (2.3)
d
If we apply ∂ρ on equation (2.2) we find

∂ρ2µ εν + ∂ρν
2
εµ = ∂ρ f δµν , (2.4)

which implies
2
2 ∂µν ερ = δµρ ∂ν f + δνρ ∂µ f − δµν ∂ρ f . (2.5)

Contracting with δ µν , this gives

2 ∂ 2 εµ = (2 − d) ∂µ f . (2.6)

Applying ∂ µ to this equation and combining with (2.3) yields

(d − 1) ∂ 2 f = 0 . (2.7)

The case d = 1 is trivial: any coordinate transformation obviously satisfies (2.1). The case d = 2
is particular and will be studied in detail in the next section (although, as we will see then, some
results derived below will also apply to d = 2), and we focus on the case d ≥ 3 from now on.
2 f = 0 is solved by
The equation ∂µν

f (x) = A + Bµ xµ , (2.8)

2 ε is also a constant, and hence

for some constants A and Bµ . This implies, using (2.6), that ∂µν

εµ (x) = aµ + bµν xν + cµνρ xν xρ , (2.9)

for some constants aµ , bµν , cµνρ with cµνρ = cµρν . Let us analyse these transformations case by
case.
2 In the Minkowskian case, the conformal group contain the Poincaré group.

12
Introduction to Conformal Field Theory Antonin Rovai

1. bµν = 0 and cµνρ = 0. Then εµ is a constant: these are just translations.

µ
2. aµ = 0 and cµνρ = 0. In this case we find f = 2b µ /d and

bµν + bν µ = f δµν . (2.10)

It is always possible to decompose bµν as its symmetric part b(µν) plus its antisymmetric part
b[µν] ≡ mµν . Hence
ρ
b ρ
bµν = δµν + mµν . (2.11)
d

PoS(Modave VIII)001
It is not difficult to see that the first term consists of an infinitesimal dilatation x 7→ (1 + α)x,
µ
where α = b µ /d, while the last term corresponds to a rotation.

3. aµ = 0 and bµν = 0. If we set cµ = cλ λ µ , then the various constraints imply

cρ µν = δµρ cν + δνρ cµ − δµν cρ , (2.12)

and so ε µ (x) = 2xµ cν xν − cµ x2 . This is a new transformation that is special to conformal

transformations, and is therefore called a (infinitesimal) special conformal transformation
(SCT for short).

We have now explored all possibilities for conformal transformations at the infinitesimal level. To
find the finite transformations, we must exponentiate the different infinitesimal transformation that
we just found. Although this is straightforward in principle, it can be tedious (in particular for the
SCT). The result is

x0 = x µ + a µ
µ
(Translations) (2.13a)
0µ
x = λ xµ (Dilatations) (2.13b)
0µ µ ν
x = R νx (Rotations) (2.13c)
x µ − b µ x2 ,
x0 =
µ
(SCT) (2.13d)
1 − 2b · x + b2 x2
where λ , aµ and bµ are numbers and Rµ ν is a matrix that satisfies the constraint (1.6).
To get some intuition about an SCT, remark that under such a transformation x 7→ x0 we have

x0 µ xµ
0 2
= 2
− bµ , (2.14)
x x
so a SCT is nothing but the composition of an inversion, a translation characterised by bµ and again
an inversion.
Note that the inversion x → x/x2 satisfies (2.1) and hence is a conformal transformation. But
for this transformation to be defined at the origin xµ = 0, one should add to space-time an extra point
“at infinity” such that the origin is mapped onto it. Moreover, by composition with a translation,
any point may be mapped to this point at infinity, and hence it has nothing special: it is an extra
space-time point. In two dimensions d = 2, the resulting space-time C ∼ = R2 with this new point
“at infinity” has the topology of a sphere and is called the Riemann Sphere.

13
Introduction to Conformal Field Theory Antonin Rovai

Algebra
Following the general discussion on generators in section 1.2, we easily find the generators of
conformal transformations:

Pµ = −i∂µ (Translations) (2.15a)

D = −ix · ∂ (Dilatations) (2.15b)
Lµν = i(xµ ∂ν − xν ∂µ ) (Rotations) (2.15c)
Kµ = −i(2xµ x · ∂ − x2 ∂µ ) . (SCT) (2.15d)

PoS(Modave VIII)001
The generators given in (2.15) are valid for fields invariant under conformal transformations, i.e.
for F = 0 in the notations of section 1.1. We will later on consider the action of these generators
on fields transforming non-trivially under conformal transformations, see section 2.2.
The algebra satisfied by these generators is

[D, Pµ ] = −iPµ (2.16a)

[D, Kµ ] = −iKµ (2.16b)


[Kµ , Pν ] = 2i δµν D − Lµν (2.16c)


[Kρ , Lµν ] = i δρ µ Kν − δρν Kµ (2.16d)


[Pρ , Lµν ] = i δρ µ Pν − δρν Pµ (2.16e)


[Lµν , Lρλ ] = i δνρ Lµλ + δµλ Lνρ − δµρ Lνλ − δνλ Lµρ . (2.16f)

It is now manifest that the Euclidean algebra is a subalgebra of the above conformal algebra, since
the commutation relations of Pµ and Lµν do not involve any of the generators D or Kµ . We now
consider redefinitions of the generators that will allow us to identify the conformal group explicitly.
Consider the new generators {Jab } with a, b, . . . = −1, 0, 1, . . . , d defined as

Jµν = Lµν (2.17a)

J−1,0 = D (2.17b)
1
J−1,µ = (Pµ − Kµ ) (2.17c)
2
1
J0,µ = (Pµ + Kµ ) . (2.17d)
2
We also define (η̃ab ) as the (d + 2) × (d + 2) diagonal matrix with

η̃−1,−1 = −1 , (2.18a)
η̃00 = 1 , (2.18b)
η̃µν = δµν . (2.18c)

Remark that the matrix η̃ is of Minkowski signature (−1, +1, . . . , +1) when we consider Euclidean
space-time, while for Minkowskian space-time η̃ is of signature (−1, −1, +1, . . . , +1).
In term of η̃ and Jab , the conformal algebra (2.16) reads

[Jab , Jcd ] = i (η̃ad Jbc + η̃bc Jad − η̃ac Jbd − η̃bd Jac ) . (2.19)

14
Introduction to Conformal Field Theory Antonin Rovai

We thus conclude that the conformal group in Euclidean space is SO(d + 1, 1), while in Lorentzian
space-time it is SO(d, 2). In either case, a generic conformal transformation has (d + 1)(d + 2)/2
parameters.

2.2 Action of conformal transformations on fields

We now turn to the question of how the fields transform under a general conformal transfor-
mation, that is, we look for representations of the conformal group.
Consider a field Φ of spin s. The Poincaré generators act as

PoS(Modave VIII)001
Pµ Φ(x) = −i∂µ Φ(x) , (2.20)


Lµν Φ(x) = i xµ ∂ν − xν ∂µ Φ(x) + Sµν Φ(x) , (2.21)

where Sµν are the spin-s representation matrices of the rotation group. Consider first the subgroup
of conformal transformations that leave the point x = 0 invariant, that is, the subgroup generated
by dilatations, rotations and SCTs. The matrices representing the action of these transformations
on Φ are denoted respectively by D̃, Sµν and K̃. These generators must satisfy

[D̃, Sµν ] = 0 , (2.22a)

[K̃µ , K̃ν ] = 0 , (2.22b)
[D̃, K̃µ ] = −iK̃µ , (2.22c)


[K̃ρ , Sµν ] = i δρ µ K̃ν − δρν K̃µ , (2.22d)


[Sµν , Sρλ ] = i δνρ Sµλ + δµλ Sνρ − δµρ Sνλ − δνλ Sµρ . (2.22e)

The action of the generators at any space-time point x is obtained by conjugation with the transla-
tion operator eix·P . The result may be easily derived using the Hausdorff formula:
1 1
e−A BeA = B + [B, A] + [[B, A], A] + [[[B, A], A], A] + · · · (2.23)
2 3!
Explicitly,

K̃Φ(x) = 0 , (2.24)


DΦ(x) = −ix · ∂ + D̃ Φ(x) , (2.25)
2
ν


Kµ Φ(x) = K̃µ + 2xµ D̃ − x Sµν − 2ixµ x · ∂ + ix ∂µ Φ(x) . (2.26)

We now restrict our attention to a field Φ(x) that belongs to an irreducible representation of the
rotation group. Since D̃ commutes with all the rotation generators Sµν , D̃ must be proportional to
the identity matrix by Schur’s lemma. By comparing the action of D̃ on Φ(x) with the definition of
scaling dimension given in section 1, we deduce that D̃ = −i∆.
To conclude, let us write the finite transformation for a spinless field φ (x) under a general
conformal transformation: φ → φ 0 with

∂ x0 −∆/d
φ 0 (x0 ) = φ (x) . (2.27)
∂x
Fields transforming this way are called quasi-primary fields.

15
Introduction to Conformal Field Theory Antonin Rovai

2.3 Energy-momentum tensor and conformal invariance

Using the definition of energy-momentum tensor (1.48), we have that under a conformal trans-
formation,
1
Z
dd x T µ ∂ · ε .
µ
δS = (2.28)
d
Hence if the stress tensor is traceless, then the theory is conformal. The converse, however, is not
true in general, for the function ∂ · ε is not arbitrary as we have seen in section 2.1.
Under certain circumstances it is possible to find improvement terms that render the stress
tensor traceless. In order to do that we need the theory to be rotation and scale invariant, but as we

PoS(Modave VIII)001
will see, this may not be enough.
Let us define the virial V µ of a field Φ as the combination
∂L
Vµ = (δ µρ ∆ + iSµρ )Φ . (2.29)
∂ (∂ ρ Φ)

Then if V µ can be written as ∂λ σ λ µ for some tensor σ , we may construct an improvement term
such that the resulting modified energy-momentum tensor is traceless, and hence the theory is
conformal. Explicitly, if we define X λ ρ µν by

2 h λ ρ (µν)
X λ ρ µν = δ σ − δ λ µ σ (ρν) − δ λ µ σ (νρ) + δ µν σ (λ ρ)
d −2
1  λ ρ µν  i
+ δ δ − δ λ µ δ ρν σ (κσ ) δκσ , (2.30)
d −1
then the modified stress tensor Tm µν is given by

1
Tm µν = Tc µν + ∂ρ Bρ µν + ∂λ ∂ρ X λ ρ µν , (2.31)
2

where B is defined in (1.41). Note that since X (µν)[ρλ ] = 0, Tm µν is also symmetric and satisfies
µ µ
∂µ jD = Tm µ. (2.32)

We then see explicitly that in this case, scale invariance implies conformal invariance. Moreover,
we get a new expression for the dilatation current, valid when the equations of motion are satisfied:
µ µ ν
jD = Tm νx . (2.33)

Let us mention that for two-dimensional quantum field theory and under some broad hypothe-
ses (including unitarity), rotation and scale invariance imply conformal invariance [9].

2.4 Conformal invariance in quantum field theory and Ward identities

Let us now turn to the aspects of a quantum conformal field theory. As reviewed in section
1, all information about a continuous symmetry is encoded in the Ward identities, so let us derive
them for conformal transformations. As a warm up, we start by writing them for rotation and
translation invariance.

16
Introduction to Conformal Field Theory Antonin Rovai

Consider the canonical current associated to rotational invariance j µνρ . Assuming the stress
tensor T µν has been symmetrized, we have

j µνρ = T µν xρ − T µρ xν . (2.34)

The corresponding Ward identity is

∂µ h(T µν xρ − T µρ xν ) Φ1 (x1 ) · · · Φn (xn )i =

n
∑ δ d (x − xi )
 ν ρ ρ νρ 
xi ∂i − xi ∂iν − iSi hΦ1 (x1 ) · · · Φn (xn )i , (2.35)

PoS(Modave VIII)001
i=1

µ µ µν
where ∂i denote the derivative with respect to xi and Si are the generators of the Lorentz group
in the representation of the field Φi . For translations, the Ward identity reads
n
∂µ hT µν (x)Φ1 (x1 ) · · · Φn (xn )i = − ∑ δ d (x − xi )∂iν hΦ1 (x1 ) · · · Φn (xn )i . (2.36)
i=1

Combining equations (2.35) and (2.36) we find

n
h(T µν (x) − T ν µ (x)) Φ1 (x1 ) · · · Φn (xn )i = −i ∑ δ d (x − xi )Si hΦ1 (x1 ) · · · Φn (xn )i ,
µν
(2.37)
i=1

which is the quantum analog of the statement that the stress tensor is symmetric: as usual when
there are no anomalies for the considered classical invariances, the classical property of symmetry
T µν = T ν µ translates into a similar statement in terms of correlation functions up to the addition of
contact terms, as can be seen on the right hand side of (2.37). If we further assume that the stress
µ
tensor is traceless, then the dilatation current jD is
µ µ
jD = T ν xν , (2.38)

yielding the following Ward identity,

n
∂µ hT ν xν Φ1 (x1 ) · · · Φn (xn )i = i ∑ δ d (x − xi ) (xi · ∂i + ∆i ) hΦ1 (x1 ) · · · Φn (xn )i ,
µ
(2.39)
i=1

where ∆i is the scaling dimension of the field Φi . Equation (2.39) then implies
n
hT µ (x)Φ(x1 ) · · · Φ(xn )i = −i ∑ δ d (x − xi )∆i hΦ(x1 ) · · · Φ(xn )i ,
µ
(2.40)
i=1

which is the quantum analog of the classical statement that the stress tensor is traceless.

Constraints on two- and three-point functions

We now consider consequences of conformal invariance on two- and three-point functions,
and show that they turn out to be completely fixed, up to global multiplicative constants. Consider
the two-point function
hφ1 (x1 )φ2 (x2 )i (2.41)

17
Introduction to Conformal Field Theory Antonin Rovai

where φ1 and φ2 are scalar operators that we assume to be quasi-primary and not necessarily ap-
pearing in the collection of fields {Φi }. Under a conformal transformation x 7→ x0 , the quantum
invariance condition (1.50) together with (2.27) imply

∂ x0 ∆1 /d ∂ x0 ∆2 /d
hφ1 (x1 )φ2 (x2 )i = hφ1 (x01 )φ2 (x02 )i . (2.42)
∂x x1 ∂x x2

From rotation and translation invariance, we find that

hφ1 (x1 )φ2 (x2 )i = f (|x1 − x2 |) , (2.43)

PoS(Modave VIII)001
for some function f of one real variable. Focusing on a dilatation x 7→ λ x in formula (2.42), we
find that the function f defined in (2.43) must be such that

f (s) = λ ∆1 +∆2 f (λ s) , (2.44)

We now consider constraint (2.44) at the special point s = 1/λ . This yields the following formula
for f ,
C12
f (s) = ∆ +∆ , (2.45)
s 1 2
where C12 ≡ f (1) is an unknown constant. Plugging (2.45) into formula (2.43) yields

C12
hφ1 (x1 )φ2 (x2 )i = · (2.46)
|x1 − x2 |∆1 +∆2

If we consider now an SCT defined in (2.13d), it is easy to check that the constraint (2.42) implies
that the two-point function (2.41) is automatically zero for ∆1 6= ∆2 . Our final result is therefore

C12
hφ1 (x1 )φ2 (x2 )i = if ∆1 = ∆2 and 0 otherwise. (2.47)
|x1 − x2 |2∆1

In particular, a one-point function (obtained from (2.41) by taking one of the two operators to be
the identity) is non-zero only if the scaling dimension of the operator is zero,

hφi (x)i 6= 0 iff ∆i = 0. (2.48)

Let us mention that it is possible to show that in a unitary CFT, identity is the only operator with
zero scaling dimension.
Let us now turn to the three-point function,

hφ1 (x1 )φ2 (x2 )φ3 (x3 )i . (2.49)

Using the constraints obtained from (1.50) applied to translations, rotations and dilatations, we find
that the three-point function (2.49) must take the form

(abc)
C123
hφ1 (x1 )φ2 (x2 )φ3 (x3 )i = a b xc
+ all permutations in (a, b, c) , (2.50)
x12 x23 13

18
Introduction to Conformal Field Theory Antonin Rovai

(abc)
where C123 are unknown numbers, the three numbers a, b, c satisfy the constraint a + b + c = ∆1 +
∆2 + ∆3 and xi j ≡ |xi − x j |. Applying now to (2.49) the constraint (1.50) for an SCT characterised
by a vector bµ as in (2.13d) yields

(abc)
C123 (γ1 γ2 )a/2 (γ2 γ3 )b/2 (γ1 γ3 )c/2
hφ1 (x1 )φ2 (x2 )φ3 (x3 )i = a xb xc
γ1∆1 γ2∆2 γ3∆3 x12 23 13
+ all permutations in (a, b, c) , (2.51)

where we defined the quantities γi by

PoS(Modave VIII)001
γi = 1 − 2bµ xi + b2 xi2 .
µ
(2.52)

Compatibility of the two formula (2.50) and (2.51) requires

a + c = 2∆1 , a + b = 2∆2 , b + c = 2∆3 . (2.53)

Solving the system (2.53) for a, b, c in terms of ∆1 , ∆2 and ∆3 , we find that

C123 ,
hφ1 (x1 )φ2 (x2 )φ3 (x3 )i = ∆1 +∆2 −∆3 ∆2 +∆3 −∆1 ∆3 +∆1 −∆2
(2.54)
x12 x23 x13

where C123 is an undetermined coefficient.

Comment about the non-existence of an S-matrix in non-trivial CFTs

As a consequence of scale invariance, there is no physically meaningful notion of distance in a
conformal field theory. It is then impossible to define an S-matrix in a standard fashion, where we
assume that interactions are negligible when wave packets are separated enough. Another way to
see this is by remembering that inversions are now symmetries of the theory, and if interactions are
assumed to tend to zero at infinity, then they should be zero everywhere. The concept of observables
cannot be defined in a standard fashion; for our purposes, we shall assume that correlators are the
physically relevant quantities to consider.

3. Conformal invariance in two dimensions

We now turn to the study of conformal field theories in two dimensions. Actually, much of
what we derived in section 2 remains valid when d = 2, because the set of conformal transforma-
tions in two dimensions contains those satisfying (2.9). As we shall see, there are however much
more solutions to the constraints (2.1) in two dimensions.

3.1 Identification of the transformations and analysis

Let us slightly change our notations and call (z0 , z1 ) our space-time coordinates. Recall that
the definition of a conformal transformation zµ 7→ wµ (z) is

∂ wρ ∂ wλ µν
δ ρλ = δ . (3.1)
∂ z µ ∂ zν

19
Introduction to Conformal Field Theory Antonin Rovai

Explicitely, this is equivalent to

2  0   1 2  1 2
∂ w0

∂w ∂w ∂w
0
+ 1
= 0
+ , (3.2)
∂z ∂z ∂z ∂ z1
∂ w0 ∂ w1 ∂ w0 ∂ w1
+ 1 = 0. (3.3)
∂ z0 ∂ z0 ∂ z ∂ z1

There are two ways one can solve these equations for wµ (z): either we set

∂ w0 ∂ w1 , ∂ w0 ∂ w1 ,

PoS(Modave VIII)001
= − = 0 (3.4)
∂ z0 ∂ z1 ∂ z1 ∂z

or we can choose
∂ w0 ∂ w1 , ∂ w0 ∂ w1
= 1 = − · (3.5)
∂ z0 ∂z ∂ z1 ∂ z0
These a precisely the Cauchy-Riemann relations for (respectively) an holomorphic or an anti-
holomorphic function! It is then very convenient to define the complex variables z = z0 + iz1 and
z̄ = z0 − iz1 and extend space-time to a complex space-time, where z̄ is considered as independent
of z. We recover our original space-time by imposing the reality condition z∗ = z̄. We have the
usual complex analysis relations

1 1
z0 = (z + z̄) , z1 = (z − z̄) , (3.6a)
2 2i
1 1
∂z = (∂0 − i∂1 ) , ∂z̄ = (∂0 + i∂1 ) , (3.6b)
2 2
∂0 = ∂z + ∂z̄ , ∂1 = i(∂z − ∂z̄ ) . (3.6c)

In these coordinates, the metric δ = (δµν ) and its inverse δ −1 read

! !
0 1/2 −1 02
δ= , δ = . (3.7)
1/2 0 20

We also define the antisymmetric tensor (εµν ) by

! !
0 i/2 0 −2i
ε= , ε −1 = . (3.8)
−i/2 0 2i 0

Note that we did not yet imposed that the coordinate transformation is everywhere well defined,
and the Cauchy-Riemann relations are of course valid only when evaluated at points where the
functions (w(z), w̄(z̄)) exist. In the rest of these notes, when we say that “ f is holomorphic” (or
“anti-holomorphic”) we actually mean meromorphic (or anti-meromorphic), that is, holomorphic
(or anti-holomorphic) on the whole complexified space-time except maybe at some isolated points
where the function may have poles, and we write f (z) (or f (z̄) respectively). The set of conformal
transformations in two dimensions is hence infinite-dimensional, as it consists of all holomorphic
and anti-holomorphic maps of the Riemann sphere.

20
Introduction to Conformal Field Theory Antonin Rovai

Generators
We consider an infinitesimal transformation z 7→ z+ε(z), z̄ 7→ z̄+ ε̄(z̄), where ε is holomorphic
and ε̄ is antiholomoprhic. We assume that ε admits, at least in a non-empty open set, the following
Laurent series around zero:
∞
ε(z) = ∑ cn zn+1 , (3.9)
n=−∞

where cn are its Laurent coefficients, and similarly for ε̄ with some coefficient c̄n (that are not
a priori related to the coefficients cn ). Consider a scalar field φ = φ (z, z̄). Since we are in two
dimensions, the scaling dimension of φ is ∆ = 0, and hence under our infinitesimal conformal

PoS(Modave VIII)001
transformation we have

δ φ (z, z̄) = φ 0 (z, z̄) − φ (z, z̄) = −∂z φ (z, z̄)ε(z) − ∂z̄ φ (z, z̄)ε̄(z̄) . (3.10)

The generators `n , `¯n are defined so that

∞
cn `n φ + c̄n `¯n φ ,


δφ = ∑ (3.11)
n=−∞

and hence
`n = −zn+1 ∂z , `¯n = −z̄n+1 ∂z̄ . (3.12)

It is easy to show that they satisfy the following commutation relations:

[`n , `m ] = (n − m)`n+m , (3.13a)

[`¯n , `¯m ] = (n − m)`¯n+m , (3.13b)
[`n , `¯m ] = 0 . (3.13c)

Note that the two sets of generators `n and `¯n commute. The algebra defined by the relation (3.13a)
is a Witt algebra for the generators `n (and similarly for the `¯n ) and thus the algebra (3.13) of the
generators of conformal transformation in two dimensions at the classical level consists of two
copies of the Witt algebra.

Global conformal transformations

In this section we are interested in conformal transformations that are globally defined on the
whole complex Riemann sphere. If such transformations leave the action untouched, then they are
called “true symmetries” or “global symmetries,” as they are the only ones that provide conserved
currents that are everywhere well defined.
Given the form (3.12) for the generators, we see that only `n with n ≥ −1 are defined at the
origin. Moreover, one should require them to be also well defined at the point at infinity, since
the latter belongs to the Riemann sphere. Setting w = 1/z, we find that `n = w1−n ∂w , which is
well defined at w = 0 only for n ≤ 1 (similar conclusions apply for `¯n ). Therefore, there are only
6 generators that correspond to conformal transformations that are well defined on the complex
Riemann sphere:
{`0 , `1 , `−1 } and {`¯0 , `¯1 , `¯−1 } . (3.14)

21
Introduction to Conformal Field Theory Antonin Rovai

As a consequence of (3.13), these generators satisfy the following algebra:

[`0 , `−1 ] = `−1 , (3.15a)

[`1 , `0 ] = `1 , (3.15b)
[`1 , `−1 ] = 2`0 . (3.15c)

and similarly for `¯n , with n = −1, 0, 1. Algebra (3.15) is a subalgebra of the Witt algebra (3.13a).
Here are the infinitesimal and finite transformations corresponding to these generators (since the
algebra for the `¯n is similar to the one for the `n , we focus on the holomorphic part and similar

PoS(Modave VIII)001
results will apply for the anti-holomorphic part):

`−1 : z 7→ z + ε , z 7→ z + α , (3.16a)
`0 : z 7→ z + εz , z 7→ λ z , (3.16b)
z ,
`1 : z 7→ z + εz2 , z 7→ (3.16c)
1−βz

where ε, α, λ and β are complex parameters, ε being in addition infinitesimal. Combining the three
above transformations, we find that a general global holomorphic conformal transformation is of
the form
az + b ,
z→ (3.17)
cz + d
where the complex numbers a, b, c, d are such that ad − bc = 1. It is easy to check that the result of
the composition of two such transformations (parametrized, say, by a1 , b1 , c1 , d1 and a2 , b2 , c2 , d2 )
corresponds to a transformation parametrized by a, b, c, d with ad − cb = 1 and such that
! ! !
ab a2 b2 a1 b1
= . (3.18)
cd c2 d2 c1 d1

Global holomorphic conformal transformations hence form a group isomorphic to SL(2, C)/Z2
with matrix multiplication, the Z2 factor coming from the fact that flipping the sign of a, b, c, d
returns the same conformal transformation (3.17) and preserves the relation ad − bc = 1.
Similar results apply for anti-holomorphic global conformal transformations. If we reduce to
real space-time by imposing z̄ = z∗ , it can be shown that the resulting group for global conformal
transformations is SO(3, 1). Recall that for d ≥ 3, the conformal group is SO(d + 1, 1), and there-
fore we have shown that this result is actually also valid for d = 2. This means that all results
derived in section 2 from invariance under SO(d + 1, 1), as for instance the transformation law for
scalar fields, Ward identities and constraints on two- and three-point functions apply for d = 2 as
well.

Transformation laws for fields with arbitrary spin and primary fields
Consider a field Φ of spin s and scaling dimension ∆. We define for this field its holomorphic
weight h and anti-holomorphic weight h̄ by

∆+s , ∆−s ,
h= h̄ = (3.19)
2 2

22
Introduction to Conformal Field Theory Antonin Rovai

and consider a global conformal transformation z → w(z), z̄ → w̄(z̄). Then it can be shown3 that
the formula (2.27), valid for scalar fields only, generalizes to

 −h  −h̄
0 dw dw̄
Φ (w, w̄) = Φ(z, z̄) . (3.20)
dz dz̄

Again, fields transforming this way are termed quasi-primary fields. If, in addition, the field Φ
transforms as (3.20) for any conformal transformation (not necessarily global), then Φ is called a

PoS(Modave VIII)001
primary field. Obviously, a primary field is always quasi-primary, but the converse is not true. As
we will shortly see, the energy-momentum tensor of a generic CFT is an example of quasi-primary
that is not primary.
For future reference, we note that under a purely holomorphic conformal transformation z 7→
z + ε(z), we have for a primary field Φ that Φ 7→ Φ + δε Φ with

δε Φ = −ε∂ Φ − h∂ ε , (3.21)

where ∂ is a shorthand for ∂z . We also write ∂¯ = ∂z̄ in the following.

3.2 Ward identities

We will now write the Ward identities for translation, rotation and scale invariance in two
dimensions. Note that the generators of rotations Sµν , being two by two antisymmetric matrices,
are automatically proportional to εµν . The proportionality factor is nothing but the spin s of the
representation: Sµν = s εµν . With this observation, the Ward identities (2.36), (2.37) and (2.40) are
(in any coordinates) equivalent to

n
∂µ hT µν (x)Φ1 (x1 ) · · · Φn (xn )i = − ∑ δ 2 (x − xi )∂iν hΦ1 (x1 ) · · · Φn (xn )i , (3.22a)
i=1
n
εµν hT µν (x)Φ1 (x1 ) · · · Φn (xn )i = −i ∑ δ 2 (x − xi )si hΦ1 (x1 ) · · · Φn (xn )i , (3.22b)
i=1
n
hT µ (x)Φ1 (x1 ) · · · Φn (xn )i = − ∑ δ 2 (x − xi )∆i hΦ1 (x1 ) · · · Φn (xn )i .
µ
(3.22c)
i=1

We now specialize to our complex coordinates (z, z̄) defined in (3.6). Using the standard formula

1¯1 1 1
δ 2 (x) = ∂ = ∂ , (3.23)
π z π z̄
3 To convince yourself, you might want to start by showing this formula for an infinitesimal transformation.

23
Introduction to Conformal Field Theory Antonin Rovai

we find the formulas

2π∂z hTz̄z Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i + 2π∂z̄ hTzz Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i
n
1
= − ∑ ∂z̄ ∂w hΦ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i , (3.24a)
i=1 z − wi i
2π∂z hTz̄z̄ Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i + 2π∂z̄ hTzz̄ Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i
n
1
= − ∑ ∂z ∂w̄ hΦ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i , (3.24b)
i=1 z̄ − w̄i i

PoS(Modave VIII)001
2hTzz̄ Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i + 2hTz̄z Φ1 (z1 , z̄1 ) · · · Φn (zn , z̄n )i
n
= − ∑ δ 2 (z − wi )∆i hΦ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i , (3.24c)
i=1
−2hTzz̄ Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i + 2hTz̄z Φ1 (z1 , z̄1 ) · · · Φn (zn , z̄n )i
n
= − ∑ δ 2 (z − wi )si hΦ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i , (3.24d)
i=1

Combining the formulas (3.24c) and (3.24d) yields

n
1
2πhTz̄z Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i = − ∑ ∂z̄ hi hΦ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i , (3.25)
i=1 z − wi
n
1
2πhTzz̄ Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i = − ∑ ∂z h̄i hΦ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i , (3.26)
i=1 z̄ − w̄i

and therefore, introducing

T = −2πTzz , T̄ = −2πTz̄z̄ , (3.27)
we finally find
 n 
1
∂z̄ hT (z, z̄)Φ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i − ∑ ∂w hΦ1 (w1 , w̄1 ) · · · Φn (wn , w̄n )i
i=1 z − wi i

hi
+ hΦ 1 (w1 , w̄1 ) · · · Φ n (wn , w̄n )i = 0 , (3.28)
(z − wi )2

with a similar expression for T̄ . The quantity between brackets is hence holomorphic everywhere,
and in our notations we may replace T (z, z̄) by T (z). One should not forget, however, that T (z) fails
to be holomorphic when inserted at coincident point in a correlation function. Similarly, T̄ (z, z̄) is
replaced by T̄ (z̄).

Transformation of correlation functions of primaries

We now derive a formula giving the variation of a correlation function of primaries under an
infinitesimal conformal transformation x 7→ x + ε(x). Using (3.21) with the definition (3.19) for the
holomorphic and anti-holomorphic weights, the correlation function hΦ(x1 ) · · · Φ(xn )i of primaries
Φ(x1 ), . . . , Φ(xn ) varies according to
n  ε µν ∂µ εν

∂ ·ε
δε hΦ1 (x1 ) · · · Φn (xn )i = − ∑ ε · ∂i + ∆i + si hΦ1 (x1 ) · · · Φn (xn )i. (3.29)
i=1 2 2

24
Introduction to Conformal Field Theory Antonin Rovai

On the other hand let us consider the following integral,

Z
d2 x h∂µ (εν (x)T µν (x))Φ1 (x1 ) · · · Φn (xn )i , (3.30)
D

where the domain D contains all points xi at which the operators are inserted. This integral is not
zero only because of the singularities of the correlations function at coincident points. Performing
the derivative in (3.30) and decomposing the term with ∂µ εν in its symmetric and antisymmetric
part,4 we find for the integrand of (3.30)

h∂µ (εν (x)T µν (x))Φ1 (x1 ) · · · Φn (xn )i

PoS(Modave VIII)001
n 
ε µν ∂µ εν

2 ∂ ·ε
= − ∑ δ (x − xi ) ε · ∂i + ∆i + si hΦ1 (x1 ) · · · Φn (xn )i . (3.31)
i=1 2 2

Thanks to the δ -functions in the right hand side of (3.31), the integral (3.30) is trivial to perform,
and (3.29) we end up with
Z
δε hΦ(x1 ) · · · Φ(xn )i = d2 x ∂µ hT µν (x)εν (x)Φ(x1 ) · · · Φ(xn )i . (3.32)
D

We now apply Gauss’s theorem and specialize to the complex coordinates defined in (3.6). Using
the definitions (3.27) of T and T̄ we finally find

1
I
δε,ε̄ hΦ(w1 , w̄1 ) · · · Φ(wn , w̄n )i = − dz ε(z)hT (z)Φ(w1 , w̄1 ) · · · Φ(wn , w̄n )i+
2πi C
1
I
dz̄ ε̄(z̄)hT̄ (z̄)Φ(w1 , w̄1 ) · · · Φ(wn , w̄n )i , (3.33)
2πi
where C = ∂ D is the boundary of D and we explicitly wrote that the variation depends on two
(independent) functions ε and ε̄. Using the general formula (1.51) relating the variation of a cor-
relator under a symmetry transformation and the corresponding conserved current, the currents
j(z) and j¯(z̄) associated with holomorphic and anti-holomorphic transformations respectively are
straightforwardly read-off from the Ward identity (3.33):

j(z) = ε(z)T (z) , j¯(z̄) = ε̄(z̄)T̄ (z̄) . (3.34)

While we assumed that the fields Φ(xi ) appearing in the equation (3.33) are primaries, it may be
shown that the same formula holds for non-primaries. This will be clearer in the operator formalism
described in section 3.3.
For primaries, we may actually push further formula (3.33), thanks to equation (3.28) and
performing the contour integration using Cauchy’s theorem. In the case of a purely holomorphic
transformation z 7→ z + ε(z), we find
n  
δε hΦ1 (x1 ) · · · Φn (xn )i = − ∑ ε(wi )∂wi + ∂ ε(wi ) hΦ1 (x1 ) · · · Φn (xn )i , (3.35)
i=1

and similarly for an anti-holomorphic transformation z̄ 7→ z̄ + ε̄(z̄).

4 Ifyour are confused by the presence of an antisymmetric piece, recall that symmetry of T µν in a correlation
function is violated by contact terms (if any), see (2.37).

25
Introduction to Conformal Field Theory Antonin Rovai

3.3 Operator Formalism

Quantum field theories are more frequently studied in the path integral formalism, as it is for
instance very convenient to establish Feynman rules or describe semi-classical effects from the
formal expression of correlators in terms of path integrals weighted by e−S . In two-dimensional
CFT field theories however, it turns out that the operator formalism, which is equivalent to the path
integral formalism, is much more convenient. In particular, transformations of fields and operators
are very conveniently described by the Operator Product Expansion that we will introduce shortly.
In the operator formalism, a classical theory is quantized by defining a Hilbert space, promoting
fields to operators on this Hilbert space and giving their equal-time commutation relations. In par-

PoS(Modave VIII)001
ticular, a coordinate (“time”) is singled out. In Euclidean space-time however, there is no preferred
direction for a time-like coordinate. We shall hence choose it arbitrarily, although in other contexts
(like string theory for example) this is actually the natural choice: time will be the radial direction
in the complex plane. The resulting quantization is called radial quantization.

3.3.1 Radial order and mode expansion

Even if this was implicit in all our formulas, one should not forget that there is, in any correlator
we considered so far, a time-ordering operator that sorts the fields in decreasing time order from
left to right. In radial quantization, time ordering is replaced by radial ordering that we define with
the operator R with
(
Φ1 (z)Φ2 (w) if |z| > |w| ,
R [Φ1 (z)Φ2 (w)] = (3.36)
±Φ2 (w)Φ1 (z) if |z| < |w| ,

where the plus sign is selected if one of the operators Φ1 or Φ2 is a commuting-number valued
operator, while we chose the minus sign if both Φ1 and Φ2 are anticommuting-number valued.
Although we will not write explicitly the radial ordering operator R , it should be kept in mind that
it is present in any expression we consider.
Let us consider two arbitrary operators A and B, and assume they may be represented as
I I
A= dz a(z) , B= dz b(z) , (3.37)
o o

where a(z) and b(z) are some operators depending on z and the integrals are taken over a small
H
circle around the origin of the complex plane. If we denote by Cw the integral taken on a small
circle around a point w of complex Riemann sphere, we have
I I I
dz a(z)b(w) = dz a(z)b(w) − dz b(w)a(z) = [A, b(w)] , (3.38)
Cw Co+ Co−

where Co+ is a circle centred at the origin with radius slightly bigger that |w|, Co− is a circle also
centred at the origin but with radius slightly smaller that |w| and we have used the implicit presence
of the radial operator R on the left hand side. We therefore conclude that the commutator [A, B] of
the two operators A and B can be expressed, using a(z) and b(z), as
I I
[A, B] = dw dz a(z)b(w) . (3.39)
Co Cw

26
Introduction to Conformal Field Theory Antonin Rovai

We thus see that [A, B] essentially depends on the singular values of the product of the operators
a(z) with b(w) as z → w. This observation will be our motivation to define the operator product
expansion in subsection 3.3.2.
We introduce the following mode expansion5 by decomposing the energy-momentum opera-
tors T and T̄ as
∞ ∞
Ln L̄n
T (z) = ∑ n+2 , T̄ (z̄) = ∑ n+2 , (3.40)
n=−∞ z n=−∞ z̄

for some operators Ln , L̄m whose interpretation will be given in a moment. These relations may be
easily inverted using Cauchy’s theorem:

PoS(Modave VIII)001
1 1
I I
Ln = dz zn+1 T (z) , L̄n = dz̄ z̄n+1 T̄ (z̄) . (3.41)
2πi Co 2πi Co

We will now show that the operators Ln , L̄m are the conserved charges associated to conformal
transformations. Let us consider an infinitesimal holomorphic conformal transformation z 7→ z +
ε(z). In the radial quantization scheme, the associated conserved charge Qε , defined by (1.34),
reads
1 1
I I
Qε = dz jε (z) = dz ε(z)T (z) , (3.42)
2πi Co 2πi Co
where we used the explicit expression (3.34) for the current jε (z). Therefore, the associated varia-
tion δε Φ reads
1
I
δε Φ(w) = −[Qε , Φ(w)] = − dz ε(z)T (z)Φ(w) . (3.43)
2πi Co
We now decompose ε as
∞
ε(z) = ∑ zn+2 εn (3.44)
n=−∞

for some εn parameterising the transformation, and we finally find that

∞
Qε = ∑ εn Ln . (3.45)
n=−∞

We deduce from (3.45) that the operators Ln are the conserved charges corresponding to conformal
invariance (for holomorphic transformations). A similar result holds for anti-holomorphic trans-
formations, and so the Ln , L̄m are the quantum analogues of `n , `¯m introduced in (3.11). In order
to determine the algebra satisfied by the operators Lm , L̄n , we first need to introduce the concept of
operator product expansion.

3.3.2 Operator Product Expansion

We now introduce the concept of operator product expansion, or OPE for short. Let a(z) and
b(w) be some operators. Typically, the expectation value ha(z)b(w)i diverges as some negative
power of |z − w| when z → w. Moreover, as we have seen in the previous section (see for example
formula (3.33)) such expectation values often appear in a contour integral and thus by Cauchy’s
theorem, the result of such typical integrations will be completely determined by the singular part
5 This terminology is more natural in polar coordinates, but we shall nevertheless use it.

27
Introduction to Conformal Field Theory Antonin Rovai

of the integrand. To get some control on this, we assume that the product a(z)b(w) is given by (we
restrict to holomorphic operators for convenience)
N
{ab}n (w) ,
a(z)b(w) = ∑ n
(3.46)
n=−∞ (z − w)

for some operators {ab}n . The Operator Product Expansion is defined as the sum of the singular
terms in the series of the right hand side of (3.46), and we write
N
{ab}n (w) ,
a(z)b(w) ∼ ∑ (3.47)

PoS(Modave VIII)001
n
n=1 (z − w)

where by definition the right hand side is the OPE of the left hand side. As an equality between
operators, (3.47) should always be thought of as inserted in some correlator, possibly with other
fields inserted.
The first example of OPE is the OPE of T with a primary operator Φ, that we can infer from
equation (3.28):
h 1
T (z)Φ(w, w̄) ∼ 2
Φ(w, w̄) + ∂w Φ(w, w̄) . (3.48)
(z − w) z−w
What can we say about the OPE of T with a quasi-primary operator Φ? Under an infinitesimal
global conformal transformation z 7→ z + αz, we have

δ Φ(z, z̄) = − αz∂ Φ(z, z̄) + ᾱ z̄∂¯ Φ(z, z̄) + (hα + h̄ᾱ)Φ(z, z̄) .


(3.49)

On the other hand the Ward identity (3.33) with ε(z) = αz yields
1 1
I I
δ hΦ(z, z̄)i = − dw αwhT (w)Φ(z, z̄)i + dw̄ ᾱ w̄hT̄ (w̄)Φ(z, z̄)i (3.50)
2πi 2πi
= −αh{T Φ}2 (z, z̄) + z{T Φ}1 (z, z̄)i + ᾱh{T̄ Φ}2 (z, z̄) + z̄{T̄ Φ}1 (z, z̄)i . (3.51)

The operators {T Φ}n are thus given by

{T Φ}1 = ∂ Φ , {T Φ}2 = hΦ , (3.52)

{T̄ Φ}1 = ∂¯ Φ , {T̄ Φ}2 = h̄Φ . (3.53)

In conclusion, the most general form of the OPE of T with a quasi-primary field Φ is of the form
h ∂ Φ(w, w̄) ,
T (z)Φ(w, w̄) ∼ · · · + 2
Φ(w, w̄) + (3.54)
(z − w) z−w

and similarly for T̄ , where the dots denote extra terms that diverge like (z − w)−n with n ≥ 3. When
all these extra terms vanish, (3.48) implies that Φ is primary.
Since T is itself a quasi-primary (of weights (h, h̄) = (2, 0)), we can look at the case where Φ
is taken to be T in formula (3.54). It can be shown that in this case, thanks to unitarity and using
a dimensional argument, the only possible terms in addition to those already shown in (3.54) is
proportional to (z − w)−4 . As a consequence, the T T OPE is
c/2 2T (w) ∂ T (w) ,
T (z)T (w) ∼ 4
+ 2
+ (3.55)
(z − w) (z − w) z−w

28
Introduction to Conformal Field Theory Antonin Rovai

for some number c, called the (holomorphic) central charge of the CFT. The origin of this name
will be explained in the next subsection.
We similarly have, considering the anti-holomorphic operator T̄ , that

c̄/2 2T̄ (w̄) ∂¯ T (w̄) ,

T̄ (z̄)T̄ (w̄) ∼ 4
+ 2
+ (3.56)
(z̄ − w̄) (z̄ − w̄) z̄ − w̄

where again c̄ is a number, called the anti-holomorphic central charge.

Using (3.43) with Φ replaced by T and (3.55), we find that under any holomorphic conformal
transformation z 7→ z + ε(z) we have

PoS(Modave VIII)001
c 3
δε T = − ∂ ε − 2T ∂ ε − ε∂ T . (3.57)
12
Comparing with (3.21), we see that except when c = 0, T is not a primary field. It is a nice exercise
to show that the finite transformation T 7→ T 0 coming from the infinitesimal transformation (3.57)
is, for z 7→ w(z), given by
 −2 
0 dw c 
T (w) = T (z) − {w; z} , (3.58)
dz 12

where {w; z} is called the Schwarzian derivative of w with respect to z and is defined by
2
d3 w/dz3 3 d2 w/dz2

{w; z} ≡ − . (3.59)
dw/dz 2 dw/dz

3.3.3 Virasoro algebra

Now that we have introduced the concept of OPE, it is straightforward to find the algebra
satisfied by the operators Ln , L̄m using formula (3.39). The result reads
c
[Ln , Lm ] = (n − m)Lm+n + n(n2 − 1)δm+n,0 , (3.60a)
12
[Ln , L̄m ] = 0 , (3.60b)
c̄
[L̄n , L̄m ] = (n − m)L̄m+n + n(n2 − 1)δm+n,0 . (3.60c)
12
Let us compare algebra (3.60) with the classical algebra given in (3.13). As in the classical case,
the two different sets of generators Lm and L̄m corresponding to holomorphic and anti-holomorphic
transformations (respectively) close to commuting subalgebras, known as the Virasoro algebra and
defined by (3.60a) for the holomorphic part. The Virasoro algebra differs from the Witt algebra
by an extra term, proportional to c and commuting with all other operators; this term is called
a central term, and this is the reason why c is called the central charge. The appearance of a
central extension, when going from the classical to the quantum theory, should not be surprising.
Central extensions generically appear when the algebra for the operators in the quantum theory is
considered, as a consequence of the fact that it is only representations up to a phase that matter in
the quantum theory. It would be however much more surprising to find central extensions in purely
classical field theories, and this can actually happen (see e.g. [10]).

29
Introduction to Conformal Field Theory Antonin Rovai

3.4 Trace anomaly

When a two-dimensional CFT is deformed to a curved space-time, one should expect that
conformal invariance is broken since the space-time curvature introduces explicitly a length scale,
the curvature radius. In this section we make this statement more precise by computing the trace
anomaly,6 that is the failure for the trace of the energy-momentum tensor to have a vanishing
vacuum expectation value, hT µ µ i 6= 0.
Let us start by some general considerations. The quantity hT µ µ i has dimension on length−2 ,
it must be local and scalar under rotations. As we have no dimensionful parameter in the theory,
the only possibility is that hT µ µ i is proportional to the Ricci scalar R(g) of the metric g. Thus we

PoS(Modave VIII)001
must have7
hT µ µ i = aR , (3.61)

for some dimensionless constant a. Our goal is now to find the precise value of a. To this end,
we will compare to first order around flat space the transformation properties of both sides of (the
derivative of) (3.61) under an infinitesimal conformal transformation composed with a suitably
chosen diffeomorphism.
We use the usual complex coordinates (z, z̄) and start with the operator relation
a
Tzz̄ = gzz̄ R . (3.62)
2
Applying the operator ∇z to equation (3.62) yields
a
∇z̄ Tzz̄ = ∂z R , (3.63)
2
where we used the fact that ∇z = gzz̄ ∇z̄ . Using the conservation law (valid upon taking the expec-
tation value) ∇µ T µν = 0, we deduce that
a
∇z Tzz = − ∂z R . (3.64)
2
We now consider the variation of both sides of equation (3.64) under the composition of the fol-
lowing two transformations:

1. A conformal transformation z 7→ z + ε(z). The jacobian of this map is

∂ x0
= 1+∂ε , (3.65)
∂x

and the component Tzz of the energy-momentum tensor varies according to

1 c 3 1 ε
δε Tzz = − δε T = ∂ ε + T∂ε + ∂T , (3.66)
2π 24π π 2π
where we used equation (3.57).
6 The trace anomaly is also know as the conformal anomaly.
7 A word of caution: the arguments given here does not constitute a complete proof of equation (3.61). For a detailed
proof, we refer the reader to the literature, see e.g. [3].

30
Introduction to Conformal Field Theory Antonin Rovai

2. A diffeomorphism z 7→ z − ε(z), under which the component Tzz of the energy-momentum

tensor varies according to
1 ε
δ−ε Tzz = − T ∂ ε − ∂ T , (3.67)
π 2π
where we simply used the fact that the energy-momentum tensor is a tensor under arbitrary
diffeomorphisms.

Under the composition of the two above transformations, the net effect on Tzz is simply
c 3
δtot Tzz = ∂ ε. (3.68)
24π

PoS(Modave VIII)001
Going to the conformal gauge gµν = e2ω δµν and working to first order in ω, we find that the
variation of the left hand side of (3.64) is given by
c ¯ 3
δtot ∇z Tzz = ∂∂ ε . (3.69)
12π
Let us now determine the effect on R of the same transformation. The net effect on the coordinates
is the identity, while the net effect on the metric is an infinitesimal Weyl rescaling,

g 7→ (1 + ∂ ε)g . (3.70)

The effect of an arbitrary Weyl rescaling g 7→ g̃ = e2σ g on the Ricci scalar is straightforward to
work out (see e.g. in [11] for a proof of this formula generalised to any dimension) and reads
√
g̃R(g̃) = g R(g) − 2∇2g σ ,
p

(3.71)

where ∇g is the covariant derivative compatible with the metric g. In the case of the infinitesimal
Weyl rescaling (3.70), we find using formula (3.71) the following expression for the variation δ R
of the Ricci scalar:
δ R = −R∂ ε − ∇2 ∂ ε . (3.72)
Moreover, in the conformal gauge gµν = eω δµν and in complex coordinates (z, z̄), the curved lapla-
cian ∇2 reads
∇2 = 4e−2ω ∂z ∂z̄ . (3.73)
Plugging (3.73) into (3.72) and working at first order in ω we find, for the net variation of the right
hand side of (3.64),
a
− δtot ∂z R = 2a∂ 3 ∂¯ ε . (3.74)
2
Finally, comparing equation (3.69) and (3.74), we find a = c/(24π). Our final result is thus
c
hT µ µ i = R. (3.75)
24π

Acknowledgements
Many thanks to Blagoje Oblak for his careful reading of the first version of these notes. The
author is a Research Fellow of the Belgian Fonds de la Recherche Scientifique-FNRS.

31
Introduction to Conformal Field Theory Antonin Rovai

References

[1] M. Moskovic and D. Redigolo, a- and c- theorems, Unpublished.

[2] A. Sfondrini, An introduction to universality and renormalization group techniques,

arXiv:1210.2262 [hep-th].

[3] P. Di Francesco, P. Mathieu and D. Sénéchal, Conformal field theory, Springer, New York
1997.

PoS(Modave VIII)001
[4] P. H. Ginsparg, Applied Conformal Field Theory, hep-th/9108028.

[5] S. Coleman, Aspects of Symmetry, USA: Cambridge U. Press 1985.

[6] J. Polchinski, String theory. Vol. 1: An introduction to the bosonic string, Univ. Pr., Cambridge
UK, 1998.

[7] D. Tong, String Theory, arXiv:0908.0333 [hep-th].

[8] J. Polchinski, String theory. Vol. 2: Superstring theory and beyond, Univ. Pr., Cambridge UK
1998.

[9] J. Polchinski, Scale And Conformal Invariance In Quantum Field Theory, Nucl. Phys. B 303
(1988) 226.

[10] J. D. Brown and M. Henneaux, Central Charges in the Canonical Realization of Asymptotic
Symmetries: An Example from Three-Dimensional Gravity, Commun. Math. Phys. 104 (1986)
207.

[11] M. Nakahara, Geometry, topology and physics, Taylor & Francis, Boca Raton 2003.

32