An Introduction to Conformal Field Theory

Jean-Bernard Zuber
CEA, Service de Physique Théorique
F-91191 Gif sur Yvette Cedex, France

Notes taken by Pawel Wegrzyn

1
 The aim of these lectures is to present an introduction at a fairly elemen-
tary level to recent developments in two dimensional field theory, namely
in conformal field theory. We shall see the importance of new structures
related to infinite dimensional algebras: current algebras and Virasoro alge-
bra. These topics will find physically relevant applications in the lectures by
Shankar and Ian Affleck.

1st Lecture

Infinite dimensional algebras

Let us start by introducing some basic notions related to finite and infinite
dimensional Lie algebras.
As an example of a finite-dimensional simple Lie group, describing the
internal global symmetry of a field theory in D-dimensional spacetime, let us
take the orthogonal group O(N). A multiplet of fields Φα (x) (α = 1, ..., N) is
assumed to form a N-dimensional fundamental representation of the group
O(N). The infinitesimal transformation of fields is given by
δ a Φα (x) = iTαβ
a
Φβ (x) , (1)
where T a are the generators of the infinitesimal transformations, so that
exp (i T a δεa ) belongs to O(N). They span the Lie algebra associated with
the symmetry group, completely defined by the structure constants f abc
[T a , T b ] = if abc T c . (2)
The generators of the group O(N) are taken as (hermitian) antisymmetric
matrices,
(T a )† = T a = −(T a )t , (3)
satisfying also the normalisation condition
tr T a T b = δab . (4)
In a quantum theory, the transformation law (1) for the field operator Φ̂ is
generated by the conserved charge operator Q̂a
δ a Φ̂ = i[Q̂a , Φ̂] . (5)

2
Here and in the following, the hat above a field intends to stress its operator
nature. It will be dropped whenever it is unambiguous. The following algebra
of charges holds,
[Q̂a , Q̂b ] = if abc Q̂c . (6)
In a local field theory, the charges resulting from global symmetries are given
by Z
Q̂a = dD−1 x Jˆ0a (~x, t) , (7)

where Jˆ0a are time components of the Noether currents. They are conserved
d
dt
Q̂a = 0 if the currents satisfy

∂ µ Jˆµa = 0 . (8)

Then, we can look at the equal time commutation relations between the time
components of the currents,

[Jˆ0a (~x, t), Jˆ0b (~y , t)] = if abc Jˆ0c (~x, t)δ(~x − ~y ) + . . . , (9)

and, if possible, at analogous relations for space components of the currents.

The first term on the right hand side of (9) follows from the structure of
the symmetry algebra O(N). The dots stand here for possible extra terms,
that cannot be deduced from the sole properties of the O(N) charge algebra.
They are the so-called Schwinger terms.
If these extra terms are under control and the algebra (9) closes on the
terms Jˆ0 plus a finite number of other terms, we see that these current com-
ponents form an infinite dimensional algebra. The structure of this algebra
is particularly simple in two spacetime dimensions.

Free Euclidean fermions

Let us consider a simple model of free massless fermions in two–dimensional
Euclidean spacetime. Euclidean coordinates are denoted by xµ = (x1 , x2 ). It
is convenient to adopt complex coordinates,

z = x1 + ix2 , z̄ = x1 − ix2 , (10)

3
The line element is given by,

(ds)2 = (dx1 )2 + (dx2 )2 = gzz (dz)2 + 2gz z̄ dzdz̄ + gz̄ z̄ (dz̄)2 . (11)

The flat metric gµν = δµν corresponds to off-diagonal z, z̄ components

1
gzz = gz̄ z̄ = 0 , gz z̄ = gz̄z = . (12)
2
The complex contravariant components are respectively

g zz = g z̄z̄ = 0 , g z z̄ = g z̄z = 2 , (13)

and thus the complex indices are raised and lowered according to
1
Vz = V z̄ , V z = 2Vz̄ . (14)
2
It is easy to find relations between real and complex tensor components. For
example, we can relate respective components of the gradient operator,
1 1
∂ ≡ ∂z = (∂1 − i∂2 ) , ∂¯ ≡ ∂z̄ = (∂1 + i∂2 ) . (15)
2 2
¯ The volume element reads
We will further abbreviate ∂z by ∂ and ∂z̄ by ∂.
dz̄ ∧ dz
‘d2 z ′ ≡ d2 x = dx1 ∧ dx2 = . (16)
2i
Dirac or Majorana fermions in two dimensions are two-component ob-
jects, !
ψ
Ψ= . (17)
ψ̄
The bar over the down spinor component is only a customary notation, and
both components are anticommuting. The gamma matrices may be taken to
be the Pauli matrices
! ! !
1 0 1 2 0 −i 1 2 3 1 0
γ = , γ = , γ γ = iγ = i . (18)
1 0 i 0 0 −1

(Note that γ 3 is diagonal, so that up- and down-components of (17) describe

opposite chiralities, and that returning to real space-time, a Wick rotation

4
would make γ 2 real: this allows to choose solutions of the Dirac equation
(see below) with reality properties and justifies calling the two components
of (17) Majorana-Weyl fermions.)
We can write the Dirac Lagrangian explicitely,
1 µ 1 t 1 µ
Ψ̄γ ∂µ Ψ = Ψ γ γ ∂µ Ψ =
2 2 !
1
(∂ + i∂2 ) 0 ¯ + ψ̄∂ ψ̄ . (19)
= Ψt 2 1 1 Ψ = ψ ∂ψ
0 2
(∂1 − i∂2 )
Therefore, the action for massless two-dimensional fermions is
Z
1 ¯ + ψ̄∂ ψ̄) .
S= d2 z (ψ ∂ψ (20)
2π
The factor 2π is introduced for later convenience. Dirac equations of motion
¯ = ∂ ψ̄ = 0, their solutions show that the spinor components are
are ∂ψ
holomorphic and antiholomorphic functions respectively, namely
ψ = ψ(z) , ψ̄ = ψ̄(z̄) . (21)
The fermionic system decomposes into a holomorphic (analytic) part and an
antiholomorphic (antianalytic) part.
The kinetic Lagrangian term (20) can be inverted to derive propagators,
∂¯ < ψ(z)ψ(z ′ ) >= ∂ < ψ̄(z̄)ψ̄(z¯′ ) >= πδ (2) (~r − r~′ ) . (22)
Due to the normalization chosen in (20) we obtain the following simple results
1
< ψ(z)ψ(w) >= , (23)
z−w
1
< ψ̄(z̄)ψ̄(w̄) >= , (24)
z̄ − w̄
< ψ(z)ψ̄(w̄) >= 0 . (25)

The above model can be generalized to incorporate the internal symmetry

group O(N). We consider N Majorana-Weyl fermions with the following
action
N Z
1 X ¯ α + ψ̄α ∂ ψ̄α ) .
S= d2 z (ψα ∂ψ (26)
2π α=1

5
The action is invariant under the O(N) global tranformations, with the set
of conserved currents Jµa = 12 Ψt γ 1 γµ T a Ψ. We will consider their complex
components,
1 1
J a (z) ≡ Jza = (J1a − iJ2a ) = ψα (z)Tαβ
a
ψβ (z) , (27)
2 2
1 1
J¯a (z̄) ≡ Jz̄a = (J1a + iJ2a ) = ψ̄α (z̄)Tαβ
a
ψ̄β (z̄) . (28)
2 2
The holomorphicity (resp. anti-holomorphicity) of the currents J (resp. J) ¯
µ
that follow from the equation of motion imply the conservation law ∂ Jµ =
¯ = 0. In fact this holomorphicity of J and antiholomorphicity of
2(∂ J¯ + ∂J)
J¯ are equivalent to the conservation of both the vector currents Jµ and the
a
axial currents JAxial 1 t 1 3 a ¯
µ = 2 Ψ γ γµ γ T Ψ whose z, z̄ components are (J, −J).
The only change to the above formulae due the field quantization is the
normal ordering of field operators, J a = 21 : ψα Tαβ a
ψβ : etc. Now, let us
a b
calculate the operator product J (z)J (w) in the limit where z approaches
w. Using the Wick theorem and (23-25) we calculate
1
J a (z)J b (w) = : ψ(z)T a ψ(z) :: ψ(w)T bψ(w) :
4
c
1 ab abc J (w)
= δ + if + reg . (29)
2(z − w)2 z−w
The last (‘reg’) term is finite in the limit z → w. In the same way, we obtain

1 ¯c
abc J (w̄)
J¯a (z̄)J¯b (w̄) = ab
δ + if + reg . (30)
2(z̄ − w̄)2 z̄ − w̄

J a (z)J¯b (w̄) = reg . (31)

These ‘short distance expansions’ (29-31) have to be understood in the sense
of insertions in correlation functions: in the presence of other fields located
at points different from z and w, one may write
1 1
hJ a (z)J b (w) · · ·i = 2
δ ab h· · ·i + if abc hJ c (w) · · ·i + reg . (32)
2(z − w) z−w
Finally, we can compare the above relations with the generic formula (9).
The second term on the right hand side of (29) can be recognized as the

6
Cauchy kernel, so that it matches the first term in (9). We have determined
also the Schwinger term, of the form δ ab δ ′ (~x − ~y ). This will be exposed more
clearly in the next lecture.

One lesson to be remembered from this first lecture is the importance

of complex coordinates when dealing with massless fields in two dimensions:
the holomorphic (z) and antiholomorphic (z̄) dependences have decoupled.

7
2nd Lecture

Radial ordering
As is well known, there are two main quantization procedures in field
theory. One appeals to functional integration, where the basic observables,
the correlation functions of fields, result from the integration with a certain
R
measure D φeS of the field functionals. For example the two-point function
of the current that we have been considering reads
Z −1 Z
< J a (z)J b (w) . . . >= Dφ eS Dφ eS J a (z)J b (w) . . . (33)

The second procedure emphasizes the role of observables as operators acting

in the Hilbert space of the theory. The non commutation of the field oper-
ators and their ordering in the correlation functions is an important feature
of that quantization procedure. Thus the correlation functions are to be
computed as the vacuum expectation values of suitably ordered products of
field operators. Usually, the physical observables are expressed in terms of
correlation functions made of time ordered products of fields. In conformal
field theory, it is more convenient to order the fields radially outward from
the origin. The radially ordered product of two operators is defined as
(
X̂(z, z̄)Ŷ (w, w̄) , |z| > |w|
RX̂(z, z̄)Ŷ (w, w̄) = (34)
±Ŷ (w, w̄)X̂(z, z̄) , |z| < |w|

where the plus (minus) sign is for bosonic (fermionic) operators. The proce-
dure for calculating radially ordered correlation functions, ‘the radial quan-
tization scheme’, is very powerful because it facilitates the use of complex
analysis and contour integrals.
In fact the radial ordering appears in a natural way in a conformally
invariant two–dimensional field theory. Suppose the space direction periodic,
i.e. let it be a circle of a given length L. Euclidean space–time is thus a
cylinder, a situation familiar in the context of string theory when one looks
at time evolution of closed strings, or of statistical mechanics when one works
with a finite strip with periodic boundary conditions. We denote the complex
coordinates of that cylinder by ζ, ζ̄ (the real part of ζ is the space coordinate).

8
As we shall see soon, a conformal field theory has a certain covariance under
conformal changes of coordinates. In particular, we can consider the following
mapping,
ζ ζ̄
z = e2iπ L , z̄ = e−2iπ L , (35)
that maps the cylinder onto the plane (punctured, i.e. with the origin re-
moved). Equal time lines on the cylinder correspond to constant radius
circles on the plane. Our radial ordering on the plane thus corresponds to
the usual time ordering on the cylinder.
Let us now rephrase the results that we have obtained on the short dis-
tance product of two currents in the operator language. To distinguish the
two approaches, we shall put again a hat on fields to stress their operator
interpretation. Thus (29) reads
  1 ˆc
abc J (w)
R Jˆa (z)Jˆb (w) = δ ab
+ f + reg . (36)
2(z − w)2 z−w

Affine current algebra

As it has been already mentioned, the conservation laws reexpressed in
complex coordinates lead to the (anti)holomorphic dependence of the current
components (see (27,28)). Holomorphic fields can be expanded in Laurent
series, X X
J a (z) = Jna z −n−1 , J¯a (z̄) = J¯na z̄ −n−1 , (37)
n∈Z n∈Z
I I
dz a dz̄ ¯a
Jna = J (z)z n , J¯na = J (z̄)z̄ n , (38)
O 2iπ O 2iπ
where the integrals are along contours encircling the origin.
Let us now derive the commutator between the Laurent modes,
I I I I
dz n dz m ˆa dz n dz m ˆb
[Jˆna , Jˆm
b
] = z w J (z)Jˆb (w) − z w J (w)Jˆa (z)
O 2iπ O 2iπ O 2iπ O 2iπ
I "I I #
dw m dz n dz n  
= w z − z R Jˆa (z)Jˆb (w) (39)
O 2iπ |z|>|w| 2iπ |z|<|w| 2iπ

The difference between the two z-contour integrals, one inwards, one out-
wards with respect to the w-contour, combines into a single integration along
a contour around the point w (see Fig. 1).

9
 z
w
w z
0 0 0
z
w

Fig 1 : The difference between two z-contour integrals

may be reexpressed as a contour integral around w

Then, if we insert the short distance product (36), only singular terms
contribute to the final result.
I
dw m I dz n  ˆa 
[Jˆna , Jˆm
b
] = w z R J (z)Jˆb (w)
O 2iπ w 2iπ
I I " #
c
dw m dz n 1 ab abc T (w)
= w z δ + if + reg
O 2iπ w 2iπ 2(z − w)2 z−w
n ab
= δ δn+m,0 + if abc Jn+m
c
. (40)
2

The current algebra of the modes Jˆna is called an affine Lie algebra:
n
[Jˆna , Jˆm
b
] = if abc Jˆn+m
c
+ k̂δ ab δn+m,0 . (41)
2

It is infinite dimensional: there is an infinite number of generators, Jˆna and

k̂. The finitely many modes Jˆ0a form the ordinary Lie algebra with structure
constants f abc . The extra term commutes with all generators, [k̂, Jˆna ] = 0,
whence the name ‘central term’. This ensures that the Jacobi identity is
satisfied. For irreducible representations, Schur’s lemma implies that the k̂-
ˆ The constant k thus
operator must be proportional to the identity, k̂ = k I.
depends on the specific representation of the affine algebra. We have found
that for N free Majorana fermions k = 1 : it is a ‘level k = 1’ representation
of the affine SO(N) algebra. Later, we will see that for all ‘good’ represen-
tations of current algebras, k is integer, with appropriate normalizations of
the generators.
In the following we shall drop the hat above operators.

10
 Conformal (Virasoro) algebra
Another important infinite dimensional algebra appears if we consider the
local changes of coordinates, xµ → xµ + εµ (x). The infinitesimal change of
the action defines the energy-momentum tensor Tµν
Z
1
δS = d2 x Tµν ∂ µ εν (42)
2π
1
(the choice of normalization with 2π will be convenient in the following). Let
us concentrate again on the example of the free massless Majorana fermion.
The complex components of the energy–momentum tensor read
1
T (z) ≡ Tzz = − : ψ∂ψ : ,
2
1
T̄ (z̄) ≡ Tz̄ z̄ = − : ψ̄ ∂¯ψ̄ : ,
2
Tz z̄ = Tz̄z = 0 . (43)

If we return to Cartesian tensor components, the vanishing of off-diagonal

complex components means that the energy–momentum tensor is symmetric
and traceless, while the holomorphicity of the diagonal components amounts
to the conservation law ∂ µ Tµν = 0. As in the previous section, we can
evaluate the short distance product expansions,

1 2T (w) ∂T (w)
T (z)T (w) = 4
+ 2
+ + reg ,
4(z − w) (z − w) z−w
1 2T̄ (w̄) ∂¯T̄ (w̄)
T̄ (z̄)T̄ (w̄) = + + + reg ,
4(z̄ − w̄)4 (z̄ − w̄)2 z̄ − w̄
T (z)T̄ (w̄) = reg . (44)

The Laurent modes are defined by:

X X
T (z) = Ln z −n−2 , T̄ (z̄) = L̄n z̄ −n−2 , (45)
n∈Z n∈Z

I I
dz dz̄
Ln = T (z)z n+1 , L̄n = T̄ (z̄)z̄ n+1 . (46)
O 2iπ O 2iπ

11
Following the same procedure as above for the J’s, it is now straightforward
to derive the following algebra,
1
[Ln , Lm ] = (n − m)Lm+n + n(n2 − 1)δn+m,0 ,
24
h i 1
L̄n , L̄m = (n − m)L̄m+n + n(n2 − 1)δn+m,0 ,
h i 24
Ln , L̄m = 0 . (47)

In general the Virasoro algebra is defined as

c
[Ln , Lm ] = (n − m)Lm+n + n(n2 − 1) (48)
12
and c is the central charge. We thus see that the operators Lm and L̄m of (47)
form two commuting Virasoro algebras of central charge c = 12 . Equation
(46) shows that Ln , resp. L̄n , is the generator of the change δz = z n+1 (resp.
δz̄ = z̄ n+1 ) in the quantum field theory. It is interesting to confront these
operators with their classical counterparts, namely
∂ ∂
Ln = −z n+1 , L̄n = −z̄ n+1 , (49)
∂z ∂ z̄
which satisfy the following classical algebra

[Ln , Lm ] = (n − m)Ln+m , (50)

together with similar relations for the antiholomorphic sector. We see now
that the ‘central term’ in (47) is due to quantum effects.
Note also that L0 , L̄0 are the rotation/dilatation generators, whereas L−1 ,
L̄−1 are those of translations.

12
3d Lecture

Conformal invariance
Let us first discuss briefly the general features of conformally invariant
field theories, in a generic space–time dimension D. A conformal transfor-
mation is defined as an angle-preserving local change of coordinates.
If gµν is the metric tensor (ds2 = gµν (x)dxµ dxν ), a transformation that
leaves the metric invariant up to a local scale change,
′
gµν (x) → gµν (x′ ) = (1 + α(x)) gµν (x) (51)

is conformal. For an infinitesimal coordinate transformation xµ → xµ +εµ (x),

the condition reads

gµν (x) → gµν (x) + ερ ∂ρ gµν (x) + gµρ (x)∂ν ερ + gνρ (x)∂µ ερ = (1 + α(x))gµν (x) .
(52)
Thus in Euclidean space the transformation is conformal if and only if the
following equations are satisfied,

gµρ ∂ν ερ + gνρ ∂µ ερ = α(x)gµν , (53)

Contracting with g µν (x), one identifies α = 2∂ρ ερ .

In a classical local field theory, the infinitesimal change of the action under
a local change of coordinates is defined by the energy-momentum tensor Tµν ,
see (42). Equation (42) implies the invariance of the action under constant
translations ε(x) = a. If we assume morover that the energy-momentum
tensor is both symmetric and traceless, then the action is also invariant
under infinitesimal rotations εµ = ω µν xν , (with ω µν antisymmetric), and
dilatations εµ = λxµ . (Conversely with adequate assumptions, invariance
under rotations and dilatations implies the symmetry and tracelessness of
Tµν .)
If we combine the fact that Tµν is symmetric and traceless together with
equation (53),
1 1
Tµν ∂ µ εν = Tµν (∂ µ εν + ∂ ν εµ ) = α(x)Tµν g µν = 0 , (54)
2 2

13
then we draw the striking conclusion that the action S is left invariant under
arbitrary conformal transformations! (Polyakov, 1970).
In the quantized conformally invariant field theory, equ. (42) should be
understood as inserted in the functional integral and implies Ward identities
for correlation functions. Consider some correlation function,
Z
1
< φ1 . . . φN >= Dφ eS[φ] φ1 . . . φN , (55)
Z
R
where Z = Dφ eS[φ] . Denote by δφ the change of the field φ under the
conformal transformation x → x + ε. Writing that the functional integral in
the numerator is invariant under that change, we get
N
X 1 Z D µ ν
< φ1 . . . δφi . . . φN > + d x ∂ ε < Tµν (x)φ1 . . . φN >= 0 . (56)
i=1 2π

In particular, if the δφ(x) are local expressions depending only on φ(x), ε(x)
and a finite number of their derivatives,

δφi (x) = Pi (∂, ε)φi (x) (57)

we find after functional differentiation with respect to ǫν (x)

N
X
∂xµ hTµν (x)φ1 (x1 ) · · · φN (xN )i = Peν i (∂i )δ (D) (x − xi )hφ1 · · · φN i . (58)
i=1

In particular the conservation law ∂ µ Tµν = 0 holds everywhere except at

coinciding points x = xi .

Conformal invariance in two dimensions

From now on, we shall restrict ourselves to two–dimensional theories. In
complex coordinates, equation (53) reads

∂z̄ εz = ∂z εz̄ = 0 . (59)

Thus conformal transformations correspond to holomorphic changes of the

complex coordinates,

z → z + ε(z) , z̄ → z̄ + ε̄(z̄) . (60)

14
There exists a subset of conformal transformations that form a group,
az + b
z→ . (61)
cz + d
Those are the only one-to-one applications of the complex plane with a point
at infinity (or Riemann sphere) onto itself. In general we may only demand
analyticity of ǫ in a bounded region.

Assume that T is traceless and symmetric (hence Tz z̄ = Tz̄z = 0) and

rewrite the Ward identities (56) in complex coordinates

δ < φ1 (z1 , z̄1 ) . . . φN (zN , z̄N ) >=

Z
dz̄ ∧ dz ¯
− ∂ε(z, z̄) < Tzz (z, z̄) φ1 (z1 , z̄1 ) . . . φN (zN , z̄N ) > +c.c. . (62)
2iπ
Assume moreover that ε vanishes fast enough at large distances from
the origin to allow integration by parts, say outside a domain D ′ and is
analytic in a domain D ⊂ D ′ containing the points z1 , . . . , zN . Moreover,
as we have just seen in the previous subsection, Tµν is conserved, i.e., in
z, z̄ components, Tzz ≡ T (z) is a holomorphic function of z (and mutatis
mutandis for T̄ (z̄) = Tz̄z̄ ). More precisely, the correlation function

< T (z)φ1 (z1 , z̄1 ) . . . φN (zN , z̄N ) > (63)

is analytic everywhere except at the positions of inserted fields. Similarly,

< T̄ (z̄)φ1 (z1 , z̄1 ) . . . φN (zN , z̄N ) > (64)

is antianalytic except at z = z1 , ..., zN .

Using this analyticity and Stokes theorem, we can transform the right
hand side of (62), originally an integral over the domain D ′ where ǫ is non
vanishing (see Fig. 2)

15
 z1 z1
. . D
z2 . D z2 .
. D’ .
zN zN
.
z3
.
D z3

Fig 2 : Transforming the integral in (62) into

a z contour integral around z1 , · · · zN .

Z
dz̄ ∧ dz
r.h.s. = ǫ(z, z̄)∂¯ < Tzz φ1 · · · φN > +c.c. (65)
D′ 2iπ
Z
dz̄ ∧ dz
= ǫ(z)∂¯ < Tzz φ1 · · · φN > +c.c. (66)
D 2iπ
Z  
dz
= d ǫ(z) < Tzz φ1 · · · φN > + c.c. (67)
D 2iπ
I
dz
= ǫ(z) < T (z)φ1 · · · φN > +c.c. (68)
∂D 2iπ
XN I
dz
= ǫ(z) < T (z)φ1 · · · φN > +c.c. (69)
i=1 zi 2iπ

that is, into a sum over small contours encirling each of the points zi . The
left hand side of (62) is also a sum of local contributions of each δφi , thus we
may identify each with the corresponding contour integral
I
dz
δφ(z1 , z̄1 ) = ε(z)T (z)φ(z1 , z̄1 ) + c.c. . (70)
z1 2iπ
This shows that analytical properties of the product T φ encode the variation
of the field.

16
 Primary fields
When we describe a system which possesses some symmetry, it is generally
appropriate to pick objects that obey ‘tensorial’ transformation laws. In the
case of conformal field theory, this role is played by ‘primary fields’. Under
an arbitrary conformal change of complex coordinates z → z ′ (z), z̄ → z̄ ′ (z̄)
a primary field operator transforms by definition according to
!h !h̄
dz ′ dz̄ ′
φ(z, z̄) = φ′ (z ′ , z̄ ′ ) . (71)
dz dz̄

The real numbers h and h̄ are called conformal dimensions (or conformal
weights). Note that the form φ(z, z̄)(dz)h (dz̄)h̄ is invariant. For an infinites-
imal transformation z → z + ε(z), z̄ → z̄ + ε̄(z̄) this reduces to
h i
δφ(z, z̄) = ε(z)∂ + hε′ (z) + ε̄(z̄)∂¯ + h̄ε̄′ (z̄) φ(z, z̄) . (72)

Formulae (70) and (72) for δφ are consistent if we have the following short
distance expansion,
hφ(w, w̄) ∂φ(w, w̄)
T (z)φ(w, w̄) = + + reg ,
(z − w)2 z−w
¯
h̄φ(w, w̄) ∂φ(w, w̄)
T̄ (z̄)φ(w, w̄) = 2
+ + reg . (73)
(z̄ − w̄) z̄ − w̄

The operator product expansions (73) can be used as an alternative definition

of the primary fields.
As a short exercise, let us return to our example of free massless Majorana
fermions. Taking (43), we can calculate the singular parts of the operator
products,
ψ(w) ∂ψ(w)
T (z)ψ(w) = 2
+ + reg ,
2(z − w) z−w
T̄ (z̄)ψ(w) = reg . (74)

It means that the fermionic field ψ(z) is a primary field of conformal weights
(h, h̄) = ( 21 , 0). In the same way, one can show that ψ̄(z̄) is a primary field
of weights (0, 12 ).

17
 As another example, the reader may treat the case of a free massless boson
field φ(z) for which the two-point function is < φ(z)φ(0) >= − ln z and the
energy-momentum tensor T (z) = − 21 (∂φ)2 . Using Wick theorem, she (or he)
will verify that exp iαφ(z) is a primary field of conformal weight h = α2 /2.
Those ‘vertex operators’ play a prominent role in Shankar’s lectures.
Of course, not all fields satisfy the simple transformation law (71) under
conformal changes of coordinates. For example, we see from (73) that deriva-
tives of primary fields have more complicated transformation properties. Let
us also check the properties of the energy–momentum tensor under conformal
transformations. For massless fermions, we see from (44) that the order of
singularities is higher than what is allowed by the definition formulae (73).
One can prove that the most general form of short distance products between
components of the energy–momentum tensor is
c 2T (w) ∂T (w)
T (z)T (w) = + + + reg ,
2(z − w)4 (z − w)2 z−w
c 2T̄ (w̄) ∂¯T̄ (w̄)
T̄ (z̄)T̄ (w̄) = + + + reg ,
2(z̄ − w̄)4 (z̄ − w̄)2 z̄ − w̄
T (z)T̄ (w̄) = reg . (75)

It leads to the following infinitesimal transformation law,

c ′′
δT (z) = [ε(z) + 2ε′ (z)] T (z) + ε (z), (76)
12
which can be integrated to yield the law for finite conformal transformations,
!2
′ dz ′
′ c ′
T (z) = T (z ) + {z , z} , (77)
dz 12
which involves the Schwartzian derivative,
 2
d3 z ′ 2 ′
3 d z2 
{z ′ , z} ≡ dz 3
dz ′ −  dz . (78)
2 dz
′
dz dz

Finally, if we refer to Laurent modes defined by (46) the following pair of

Virasoro algebras arises
c
[Ln , Lm ] = (n − m)Lm+n + n(n2 − 1)δn+m,0 ,
12

18
 h i c
L̄n , L̄m = (n − m)L̄m+n + n(n2 − 1)δn+m,0 ,
h i 12
Ln , L̄m = 0 . (79)

The real central charge is a very important characteristics of the conformal

field theory. As we have seen, the model of free massless fermions has c = 12
(see (44)). Can the reader compute the value of c for the boson field just
mentionned?

19
4th Lecture

Physical interpretation of the conformal weights

To expose the meaning of conformal weights of primary fields (71), let us
consider the 2-point correlation function,
1 1
< φ(z1 , z̄1 )φ(z2 , z̄2 ) > = 2h
< φ′ (1)φ′(0) > , (80)
(z1 − z2 ) (z̄1 − z̄2 )2h̄

where we have made use of a change of variable z → z ′ = zz−z 2

1 −z2
. Further-
′ ′
more, we choose the normalization < φ (1)φ (0) >= 1 and denote z1 − z2 =
r12 eiArg(z1 −z2 ) .
1
< φ(z1 , z̄1 )φ(z2 , z̄2 ) >= 2h+2h̄
e−2i(h−h̄)Arg(z1 −z2 ) . (81)
r12

The number h + h̄ is the scaling dimension of the field φ, while the number
h − h̄ is the spin of the field φ

hφ(zei2π )φ(0)i = e−4iπ(h−h̄) hφ(z)φ(0)i . (82)

A short tour through the representation theory

We shall make now a brief survey of the representation theory of the
Virasoro and affine algebras. The only representations that will concern us
are the so-called ‘highest weight’ representations. The simplest example of
a highest weight representation is provided by the familiar example of the
SU(2) algebra,
[J+ , J− ] = 2Jz , [Jz , J± ] = ±J± . (83)
A highest weight is a state |j, j > satisfying the conditions

J+ |j, j >= 0 , Jz |j, j >= j|j, j > . (84)

The descendant states are produced by acting with the operator J− ,

|j, j − p >= J−p |j, j > , Jz |j, j − p >= (j − p)|j, j − p > . (85)

20
Linear combinations of the states {|j, j >, |j, j − 1 >, |j, j − 2 >, . . .} form
the space of the representation of spin j. If the representation is finite di-
mensional, 2j has to be an integer and

J−2j+1 |j, j >= 0 . (86)

The same construction can be applied to the Virasoro algebra

c
[Ln , Lm ] = (n − m)Lm+n + n(n2 − 1)δn+m,0 . (87)
12
We follow the same procedure as for SU(2) with the following correspon-
dances L0 → Jz , Ln>0 → J+ , Ln<0 → J− . A highest weight (h.w.) state is
thus defined by the following conditions,

L0 |h >= h|h > , Ln>0 |h >= 0 , (88)

and the representation space Mh,c is generated by ‘descendant’ states of the

form
α
Lα−11 Lα−22 · · · L−pp |h > . (89)
However, there exists a big difference in comparison with the SU(2) case: the
representations of the Virasoro algebra are always infinite dimensional, being
generated by an infinite number of independent states. The representations of
the Virasoro algebra are ‘graded’, i.e. within a representation, the eigenvalues
of L0 are integrally spaced
 
  p
X
α α
L0 Lα−11 Lα−22 . . . L−pp |h > = h + jαj  Lα−11 Lα−22 . . . L−pp |h > . (90)
j=1

The h.w. state has the lowest eigenvalue h, and its descendants form the
P
‘conformal tower’, see Fig. 3. The integer pj=1 jαj is called the level of the
α
state Lα−11 Lα−22 . . . L−pp |h > in the tower.

21
 . .
. .
. .
. .

n=4 n=4

n=3 n=3

n=2 n=2

n=1 n=1

n=0 h = 1/16 n=0

1
Fig 3 : The ‘conformal tower’ of descendants above the h.w. state h = 16
in the representation Mc= 1 ,h= 1 (left) and
2 16
in the irreducible representation Vc= 1 ,h= 1 (see below).
2 16
The multiplicity is depicted for each level n = 0, 1, . . . , 4.

In conformal field theory, we may have to deal with either a finite or an

infinite number of representations of the Virasoro algebra. Because there are
two copies of the Virasoro algebra (one for the holomorphic part Ln and one
for the antiholomorphic part L̄n ), a physical h.w. state is characterized by
two weights |h, h̄ >. Among these representations, the one built from the
vacuum state is singled out. The vacuum state is defined as the h.w. state
possessing vanishing conformal weights,

|0 >≡ |h = 0, h̄ = 0 > . (91)

The vacuum state has the following properties,

Ln≥−1 |0 >= L̄n≥−1 |0 >= 0 , (92)

and is thus invariant under translations. For a ‘unitary’ representation, the

central charge c is a positive real number, the same for the left and right
copies of the Virasoro algebra. (The meaning of ‘unitary’ is that the space of

22
states is a Hilbert space, i.e. has a positive norm, and the Virasoro algebra
is consistent with this norm in the sense that L†n = L−n . This property is
not satisfied by all representations of the Virasoro algebra. )

Let us now turn to a short discussion of the representation theory of affine

(current) algebras,
1
[Jna , Jm
b
] = if abc Jn+m
c
+ knδ ab δn+m,0 , (93)
2
where f abc are the structure constants associated to a simple Lie algebra, k
is some coefficient. The zero modes J0a forms an ordinary Lie algebra (‘hor-
izontal Lie subalgebra’), while the full set of Jna constitutes an ‘affinization’
of the horizontal subalgebra.
Consider the affinization of SU(2) algebra, denoted by SU d (2), spanned
by the generators Jn+ , Jn− , Jnz (n ∈ Z). A h.w. state is defined as follows,
+ −
Jn>0 |j >= Jn>0 z
|j >= Jn>0 |j >= 0 , J0z |j >= j|j > , J0+ |j >= 0 . (94)

A tower of states is created by acting on the h.w. state with J0− or any of
the Jn<0 .
For ‘good’ (i.e. unitary) representations k and 2j must be integers and
satisfy the following relation,

0 ≤ 2j ≤ k . (95)

Originally introduced in elementary particle physics, current algebras

are now regarded as relevant in many contexts, including condensed mat-
ter physics, as illustrated by Ian Affleck in his lectures.
There is in fact an interesting connection between current algebras and
the Virasoro algebra.

Sugawara construction
Let us start from a representation of a current algebra ĝ by currents Jna
and let us form the following combination
n
1X
T (z) = : J a (z)J a (z) : , (96)
κ a=1

23
The claim is that, for a proper choice of the constant κ, T (z) qualifies as an
energy momentum tensor, or equivalently, that its Laurent moments satisfy
the Virasoro algebra,
n m=+∞
1X X
Ln = : Ja Ja : . (97)
κ a=1 m=−∞ n−m m
a
The normal ordering is defined as the requirement that the operators Jn>0
stand at the right. (Note that thanks to eq. (93) two currents Jm and Jna
a

with m, n of the same sign and with the same a do commute). Thus
X X
a a a a
κLn = Jm Jn−m + Jn−m Jm . (98)
m<n m≥n

To fix the constant κ, we require that the fields J a (z) transform as primary
fields of conformal weights (1,0),

a J a (w) ∂J a (w)
T (z)J (w) = + + reg . (99)
(z − w)2 z−w

The above operator product leads to the following relations,

a a
[Ln , Jm ] = −mJn+m . (100)

It is now easy to show that κ should be adjusted in such a way that

κ = k + g, (101)

where g is the quadratic Casimir of the adjoint representation of the Lie

algebra, X
f abc f abd = gδ cd . (102)
a,b

Recall that for SU(N), g = N.

Now, it is straightforward (though tedious!) to check the Virasoro alge-
bra. By explicit calculations, we conclude that the Ln defined in (97) satisfy
(87) with the following value of the central charge

k dim g
c= , (103)
k+g

24
where dim g is the dimension of the Lie algebra (recall dim SU(N) = (N 2 −
1) ).
The above construction is known as the Sugawara construction. If we
d current algebra, then we find a Virasoro algebra with
start from the SU(2)
3k
the central charge c = k+2 . The highest weight state |ji transforming as
the spin-j representation of the horizontal SU(2) is also a highest weight of
Virasoro with
j(j + 1)
L0 |ji = |ji . (104)
k+2

25
5th Lecture

Finite size effects

The transformation laws developed for conformal field theory may be also
applied to conformal changes corresponding to true changes in the geometry,
not only to changes of the system of coordinates. Below, we give an example
of how we can use the conformal theory in the plane to solve it on a cylinder.

Correlation function on a cylinder

Let us consider a cft on a cylinder of perimeter L. As was already men-
2iπw
tionned, the conformal transformation w → z = e L maps the cylinder on
a (punctured) plane. A primary field operator φ transforms from the plane
to the cylinder according to (71),
!h !h̄
dw dw̄
φplane (z, z̄) = φcyl (w, w̄) . (105)
dz dz̄

Taking the result for the 2-point correlation function on the plane (81), we
determine its counterpart on the cylinder,
 2h  2h̄
2iπ −2iπ (z1 z2 )h (z̄1 z̄2 )h̄
hφ(w1 , w̄1 )φ(w2 , w̄2)icyl = . (106)
L L (z1 − z2 )2h (z̄1 − z̄2 )2h̄

Let us restrict ourselves to ‘spinless’ fields, i.e. h = h̄,

1
< φ(w1 , w̄1 )φ(w2 , w̄2) >cyl =   . (107)
L π(w1 −w2 ) 4h
π
sin L

Now, it is interesting to look at two extreme opposite regimes. First, assume

that the distance between the points of field insertions is much smaller than
the size of the system, i.e. |w1 − w2 | << L. In this case, the correlation
1
function (107) can be approximated by |w1 −w 2|
4h , i.e. one recovers the result

of the plane (81). In other words, in this limit finite-size effects on correla-
tion functions can be ignored and (81) describes a universal behavior. The
opposite limit, l ≡ Im(w1 − w2 ) >> L, probes the correlation function for

26
large ‘time’ separations. This is useful for applications to statistical systems
 
at criticality. Then, the correlation function (107) behaves like exp − ξlL ,
L
where the correlation length is defined by ξL = 4hπ . Using CFT, we have
(or rather Cardy has !) thus justified a finite size scaling law that had been
observed empirically [Cardy 1984 and further references therein].

Partition function on a cylinder

Let us now compute the partition function on a cylinder, from a Hamilto-
nian point of view. The time direction corresponds as above to the imaginary
part of the complex variable w = w1 + iw2 . Then the partition function is
given by the following formula,
Z = trζ T , ζ = e−H , (108)
where the Hamiltonian is defined by,
Z
(w) (w̄)
H= dw1 T22 (w) = i(∂w − ∂w̄ ) = i(L−1 − L̄−1 ) . (109)

Using the transformation law (77) we obtain the relation,

   
2π 2 2 c
Tcyl (w) = − z Tplane (z) − . (110)
L 24
It enables us to rewrite the Hamiltonian using operators defined on the plane,
 
(w) 2π (w) c c
H= i(L−1
− = L̄−1 )
L0 − + L̄0 − . (111)
L 24 24
The partition function can be now expressed as the trace in the Hilbert space
H that describes the cft in the plane,

Z = trH e−HT = trH e−

2πT
L (L0 +L̄0 − 12c ) . (112)
The Hilbert space H decomposes into a sum of representations of the product
of the two (left and right) Virasoro algebras. Let us denote by dn(h) (the
degeneracy) the number of independent states at the level n of the conformal
(h̄)
tower of highest weight h. The numbers dn̄ are defined analogously. Using
this notation,
X X   
(h̄) T c
Z= dn(h) dn̄ exp −2π h + h̄ + n + n̄ − . (113)
(h,h̄) n,n̄
L 12

27
Assume now that TL >> 1. Let λ0 denotes the eigenvalue of largest modulus
of the operator ζ. In the limit under study, the partition function can be
approximated by Z = λT0 .
Usually, the largest eigenvalue is provided by the vacuum state of con-
formal weights h = h̄ = 0 (this is true for the ‘unitary’ physical models), so
2π c
that we can set λ0 = e L 12 . It gives the following value of the free energy
per unit ‘time’ length,
1 πc
F = ln Z = . (114)
T 6L
The above result can be interpreted as a finite-size correction to the free
energy, i.e. a ‘Casimir effect’ (Note that Z has been normalized in such a
way that the ‘bulk’ free energy limL,T →∞ T1L ln Z vanishes at the critical point
where we are standing). It is remarkable that in the cft this Casimir effect
depends only on the geometry of the system and the value of the central
charge [Affleck; Blöte , Cardy, Nightingale].

More on the representations of the Virasoro algebra

The decomposition of the Hilbert space H and the resulting calculation
of the degeneracies dn(h) and dn(h̄) have to be carried out in irreducible rep-
resentations of the Virasoro algebra. It thus important to know when a
highest weight representation of the Virasoro algebra is irreducible. Let us
parametrize the central charge of the Virasoro algebra using a (real or com-
plex) parameter x
6
c=1− . (115)
x(x + 1)
It can be proved (Kac; Feigin, Fuchs) (and it is highly non trivial!) that
the representation Mh,c is reducible if and only if the highest weight can be
written as
(r(x + 1) − sx)2 − 1
h = hrs = , (116)
4x(x + 1)
where r and s are positive integers. Moreover the discussion by Feigin and
Fuchs tells us how to construct an irreducible representation Vh,c when Mh,c
is not irreducible.

28
 Suppose furthermore that x is a positive fractional number,

p′
x= , (117)
p − p′

where p, p′ are coprime integers (i.e. without common divisor). It is then

consistent (in a sense to be explained soon) to restrict to hrs such that:

1 ≤ r ≤ p′ − 1 1≤s≤p−1 . (118)
′ 2
Thus, under these circumstances, for a given value of c = 1 − 6(p−p pp′
)
there
exists a finite number of possible hrs and all these weights are fractional
numbers. We shall refer to these representations as the minimal ones.
Further strong restrictions emerge if we require the unitarity of the repre-
sentation. It was proved (Friedan, Qiu, Shenker, one more highly non trivial
result !) that the necessary and sufficient conditions for highest weight rep-
resentations of the Virasoro algebra to be unitary are either

c ≥ 1 , h ≥ 0, (119)

or
( )
6 (r(m + 1) − sm)2 − 1
c=1− , h ∈ hrs = , (120)
m(m + 1) 4m(m + 1)

where m, r, s are integers, m ≥ 3, 1 ≤ r ≤ m − 1 and 1 ≤ s ≤ m.

Examples
Critical Ising and Potts models
Let us show how well known models of statistical mechanics fit in this
scheme. I guess everybody knows the Ising model. The Potts model is a
simple generalization of the Ising model in which (in two dimensions) ‘spins’
σ are assigned to the sites of a square lattice and may take Q distinct values,
denoted by σ = 1, · · · Q. The interaction energy of a configuration depends
on whether at the ends of each edge, the two spins are or are not in the same
state. Thus this energy reads
X
H =J δσi σj (121)
edges ij

29
Clearly, if Q = 2, we recover the Ising model (up to the addition of a constant
term in H). In two dimensions, the Potts model is known to undergo a second
order phase transition (thus has a critical conformal point) if Q ≤ 4. This
means that there is a low temperature phase in which the symmetry between
all the possible groundstates is spontaneously broken, and as T → Tc , the
‘magnetization’ hσi vanishes as a certain power β(Q) of (Tc − T ). Right at
Tc , the correlation function hσ(r)σ(0)i has a power law decay ≈ r1η . Beside
the Q = 2 (Ising) case, a case of interest is Q = 3. As a matter of a fact, they
are described at criticality by cft’s with central charges obeying the formula
(120) with respectively m = 3 and m = 5, hence c = 12 resp 54 . That the
central charge of the Ising model is 1/2, i.e. the same as that we found above
for free fermions is by no means an accident. We all know since the work of
Onsager that free fermions are hidden in the Ising model; these free fermions
are massless at T = Tc , and they build the relevant cft. Now in the m = 3
minimal cft, the conformal weights may only take three values: h = 0, 12 and
1
16
. With them we may make various fields of integer or half integer spin
h = h̄ = 0, the identity field I
h = 12 , h̄ = 0, the Majorana fermion ψ
h = 0 , h̄ = 21 , the fermion ψ̄
h = h̄ = 21 , the composite ψ̄ψ, i.e. a mass term for the fermion: this
is indeed the ‘relevant’ operator that drives the system out of its conformal
point at T = Tc ;
1
h = h̄ = 16 : this is another relevant term, nothing else than the spin
operator: it describes the response of the system to an external magnetic
field. From that value of h = h̄ for the spin, we get for the 2-spin function
1
the critical behavior < σ(0)σ(r) >≈ 1/r 4 , i.e. the well-known value of the
Ising exponent η = 14 .
For the 3-state Potts model, similar considerations apply. The little sub-
tlety is that only a subset of the allowed conformal weights (eq (120)) are
used in the description of the model under normal circumstances. For exam-
1
ple, the weight h33 = 15 yields the conformal dimension of the ‘spin’, from
which the exponent η above follows as 4/15.

30
6th Lecture

The partition function on the torus

We shall now see that all the information about the operator content of
the conformal field theory is contained in the partition function of the cft on
a torus, and that the latter is subject to strong constraints.
A torus can be regarded as a parallelogram whose opposite edges have
been identified. Let us adopt the convention that the parallelogram vertices
lie at the following points on the complex plane:

0 , 2π , 2πτ , 2π(1 + τ ) , (122)

where τ is some complex number called the modular (or aspect) ratio of the
torus, and chosen to satisfy Im τ > 0.
We have computed above in (112) the partition function on a cylinder,
but in fact by taking a trace in H we have implicitly identified the two ends
of the cylinder and made a torus of modular ratio τ = i TL . Thus by a slight
modification of the above discussion, we find that for arbitrary τ

Z = trH e2iπτ (L0 − 24 )−2iπτ̄ (L̄0 − 24 ) .

c c
(123)

Let us introduce the following parameters,

q = e2iπτ , q̄ = e−2iπτ̄ , (124)

and define the character of the irreducible representation Vh,c of highest

weight h of the Virasoro algebra,
 c
 c
∞
X
χh,c (q) = trVh,c q L0 − 24 = q h− 24 dn(h) q n . (125)
n=0

From the detailed analysis of the irreducible representations follows the knowl-
edge of these characters as explicit functions of q.
We conclude that the partition function on a torus can be decomposed
into a bilinear form of characters [Cardy 1986],
X
Z= Nhh̄ χh (q)χh̄ (q̄) , (126)
(h,h̄)

31
where the integer Nhh̄ (a multiplicity) tells us how many times the repre-
sentation (h, h̄) enters. As the identity operator must be present and non
degenerate in any sensible theory, we have also the constraint that N00 = 1.

Modular invariance on the torus

Now it is important to note that the shape of the torus does not uniquely
determine τ , namely we can perform arbitrary ‘modular’ transformations,
aτ + b
τ → τ′ = , ad − bc = 1 , (127)
cτ + d
where a, b, c, d are integers, and τ ′ describes the same torus. Modular trans-
formations form a group. Any modular transformation can be obtained as a
composition of two basic transformations,
1
T :τ →τ +1 , . S:τ →− (128)
τ
The crucial point is that we want the partition function Z to be intrin-
sically attached to the torus, i.e. to be invariant under modular transfor-
mations. This modular invariance of the partition function, together with
the form (126), turns out to yield very strong constraints on the operator
content.
T-transformation
We have here q → e2iπ q, and consequently

χh (q) → e2iπ(h− 24 ) χh (q) , χh̄ (q) → e−2iπ(h̄− 24 ) χh̄ (q) ,

c c
(129)

χh (q)χh̄ (q) → e2iπ(h−h̄) χh (q)χh̄ (q) . (130)

Thus, the invariance under the T-transformation requires spins h − h̄ of field
operators to be integers.
S-transformation
It is more difficult to find the S-transformation, because the characters
transform among themselves under it.
Let us concentrate on the family of representations of the Virasaro algebra
(116). Denote for short χrs = χhrs . It can be shown that
  X
1
χrs − = Srs,r′ s′ χr′ s′ (τ ) , (131)
τ r ′ ,s′

32
with r ′ , s′ running over the same range as in (118). The matrix S is symmetric
and unitary
s
rs′ +r ′ s+1 2 πrr ′p πss′ p′
Srs,r′s′ = (−1) sin sin . (132)
pp′ p′ p
d
Likewise for the representations of the SU(2) current algebra of level k, la-
beled by a spin j, with 0 ≤ 2j ≤ k, formula (97) gives representations of
the Virasoro algebra and the corresponding characters transform under the
S transformation according to the unitary matrix
s
2 π(2j + 1)(2j ′ + 1)
Sjj ′ = sin . (133)
k+2 k+2

Now we have all the ingredients to discuss the following problem: find
all modular invariant partition functions Z of the form (126) with N’s non
negative integers, N00 = 1. Solving this problem for a given class of rep-
resentations amounts to classifying conformal field theories of that class. I
won’t dwell on that any longer. Suffice it to say that this programme has
been carried out for the ‘minimal’ representations of Virasoro and for the (re-
lated) cft’s with a SU(2) current algebra. The solution exhibits a beautiful
structure that had not been anticipated: we refer the reader to the literature
[Cappelli et al.]. This classification programme has been pursued lately for
theories with a higher rank current algebra (SU(3) in particular: see the
recent work of T. Gannon).

The Operator Product Expansion

This short guided tour would be very incomplete without some discussion
of another fundamental feature of conformal field theories, namely their con-
sistency under operator product expansions. In any quantum field theory,
we have learnt from the work of Wilson the importance of the short distance
expansion of a product of two fields. In a conformal field theory, we postu-
late that the set of fields that we have discussed so far, the primaries and
their descendants, form a closed set under the Operator Product Expansion
(OPE). Thus for two fields ΦI , ΦJ , assumed to be primary fields of weights

33
(hI , h̄I ), (hJ , h̄J ), we write
X
ΦI (z1 , z̄1 )ΦJ (z2 , z̄2 ) = CIJK (z1 − z2 )hK −hI −hJ (z̄1 − z̄2 )h̄K −h̄I −h̄J
K
X (n,n̄) (n,n̄)
βIJK (z1 − z2 ) (z̄1 − z̄2 )|n̄| ΦK
|n|
(z2 , z̄2 ) . (134)
n,n̄

This means simply that the product of ΦI and ΦJ may be expanded on all
(n,n̄)
other primaries ΦK and their descendants denoted here ΦK with coeffi-
(n,n̄)
cients CIJK βIJK . The notation |n| denotes the level of the descendant and
(0,0) (0,0̄)
it is understood that ΦK ≡ ΦK and βIJK ≡ 1. The relative coefficients
(n,n̄)
βIJK are easy to find using the Ward identities of the Virasoro algebra. In
constrast, the structure constants CIJK of the OPE are important and non
trivial data of the cft. They give for example the three-point function of the
three primaries ΦI , ΦJ , ΦK
< ΦI (z1 , z̄1 )ΦJ (z2 , z̄2 )ΦK (z3 , z̄3 ) > (135)
CIJK
= . (136)
(z1 − z2 )(hI +hJ −hK ) (z̄1 − z̄2 )(h̄I +h̄J −h̄K ) × cyclic perm.
These structure constants may be extracted from a separate discussion of the
consistency of the OPE. We refer to the original paper by Belavin, Polyakov
and Zamolodchikov for that matter.

The fusion algebra

Rather than computing all these CIJK , we may content ourselves in a
first step with finding the selection rules that apply to them. In other words,
what are the fields ΦK that couple to a given pair ΦI and ΦJ ? This im-
portant question can be given the appropriate precise meaning. It is more
suited to look at how the holomorphic (or antiholomorphic) components of
fields combine, i.e. to discuss the fusion of representations (of Virasoro, cur-
rent,. . . algebras). For this fusion operation, the ‘minimal’ representations
that we have introduced above, or those of SU d (2), form a closed algebra.
This is the ‘consistency’ alluded to in lecture 5. We write
K
VhI ,c .VhJ ,c = ⊕hK NIJ VhK ,c (137)
K
with coefficients NIJ that are multiplicities, hence integers. An amazing dis-
covery of E. Verlinde is that the determination of these multiplicities follows

34
from the knowledge of the modular matrix S
X SILSJL SKL
K
NIJ = (138)
S0L
with 0 referring to the identity field (or vacuum representation). The mere
fact that with the S matrices of (132) and (133) these numbers are non neg-
ative integers is not trivial and the general validity of formula (138) reflects
the beautiful consistency of Conformal Field Theory.

35
 References
A.M. Polyakov J.E.T.P. Lett. 12 (1970) 381.
A.A. Belavin, A.M. Polyakov and A.B. Zamolodchikov, Nucl. Phys. B241
(1984) 333.
J. Cardy, J. Phys. A17 (1984) L385.
H.W. Blöte, J.L. Cardy and M.P. Nightingale, Phys. Rev. Lett. 56 (1986)
742.
I. Affleck, Phys. Rev. Lett. 56 (1986) 746.
J. Cardy, Nucl. Phys. B270 (1986) 186.
A. Cappelli, C. Itzykson and J.-B. Zuber, Nucl. Phys. B280 [FS18] (1987)
445; Comm. Math. Phys. 113 (1987) 1.
T. Gannon, Comm. Math. Phys. 161 (1995) 233.
E. Verlinde, Nucl. Phys. 300 (1988) [FS22] 360.
Collected reprints:
C. Itzykson, H. Saleur and J.-B. Zuber, Conformal Invariance and Applica-
tions to Statistical Mechanics, World Scientific 1988.
Lectures of J. Cardy and P. Ginsparg at the 1988 Les Houches Summer
School, Fields, strings and critical phenomena, eds E. Brézin and J. Zinn-
Justin, North Holland 1990.
Textbook :
J.-M. Drouffe and C. Itzykson, Statistical Field Theory, Cambridge Univ.
Press 1988.
Monographs
P. Goddard and D. Olive, Kac-Moody and Virasoro algebras in relation to
quantum physics, World Scientific 1988.
S. Ketov, Conformal Field Theory, World Scientific 1995.
P. Christe and M. Henkel, Introduction to Conformal Invariance and Its Ap-
plications to Critical Phenomena, Lecture Notes in Physics, Springer Verlag
1993.
P. Di Francesco, P. Mathieu and D. Sénéchal, Conformal Field Theory ,
Springer Verlag, to appear 1996.

36