An Introduction To CFT: (Preliminary Draft of Script)

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 58

An Introduction to CFT

(preliminary draft of script)

Ingo Runkel

1
Contents
1 CFT on the complex plane 3
1.1 Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Motivation: critical Ising model on the lattice . . . . . . . . . . . 5
1.3 Properties of the stress tensor . . . . . . . . . . . . . . . . . . . . 7
1.4 Examples: A topological and a meromorphic CFT . . . . . . . . 10
1.5 The Virasoro algebra . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 The Heisenberg algebra . . . . . . . . . . . . . . . . . . . . . . . 17
1.7 Conformal transformations . . . . . . . . . . . . . . . . . . . . . 20
1.8 Reconstruction from three-point correlators . . . . . . . . . . . . 24

2 Chiral conformal field theory 26


2.1 Ward identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.2 Representation theory . . . . . . . . . . . . . . . . . . . . . . . . 28
2.3 Conformal blocks . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Holomorphic factorisation . . . . . . . . . . . . . . . . . . . . . . 33
2.5 Crossing symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . 34

3 CFT on the upper half plane 36


3.1 Axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Conformal boundary conditions . . . . . . . . . . . . . . . . . . . 38
3.3 Two more OPEs . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.4 Ward identities on the UHP . . . . . . . . . . . . . . . . . . . . . 42
3.5 Example: Free boson on the UHP . . . . . . . . . . . . . . . . . 42

4 CFT on surfaces of higher genus 44


4.1 Closed surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.2 Modular invariance for the torus . . . . . . . . . . . . . . . . . . 48
4.3 Surfaces with boundary - XXX . . . . . . . . . . . . . . . . . . . 52
4.4 Cardy condition for the annulus - XXX . . . . . . . . . . . . . . 52

5 Application to open cubic string field theory 52


5.1 The c = −26 ghost system - XXX . . . . . . . . . . . . . . . . . . 52
5.2 OSFT action - XXX . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.3 Gauge invariance - XXX . . . . . . . . . . . . . . . . . . . . . . . 52

A Appendix 52
A.1 Properties of the map ob . . . . . . . . . . . . . . . . . . . . . . . 52
A.2 Proof of lemma 1.17 . . . . . . . . . . . . . . . . . . . . . . . . . 55
A.3 Proof of lemma 4.1 - XXX . . . . . . . . . . . . . . . . . . . . . . 57
A.4 Proof of theorem 4.3 - XXX . . . . . . . . . . . . . . . . . . . . . 57

Updates of this script will be posted on


http://uw.physics.wisc.edu/~strings/cft

2
[Please be aware that this is script is incomplete and work in progress. Its
content should be treated with care, as it will most certainly contain errors.]

1 CFT on the complex plane


1.1 Axioms
Disclaimer:
The aim of this section to give a simple list of properties of a CFT defined on
the complex plane, so that we have a well defined setting to work in during
these lectures. In giving this list, some loss of generality has been traded for
simplicity. There are CFTs which do not have the properties listed below, most
notably the uncompactified free boson (which has S = R≥0 ) or, in fact, any
theory with a non-compact target, and what is called logarithmic CFTs. (In
fact I do not know of a list of axioms that would cover all known examples of
theories that have a right to be called CFT.)
A closed string world sheet X in R2 is a closed subset of R2 , whose boundary
components are circles ∂X = b1 ∪ · · · ∪ bn , together with a finite collection of

distinct points {p1 , . . . , pm } from the interior X of X,

[pic 1] (1.1)

Data:
We have
to each circle b a (C–)vector space, the space of states Hb on b.
a space of fields F, an S-graded vector space F = ∆∈S F (∆) , with S, the
L

spectrum, a discrete subset of R and 0 < dim F (∆) < ∞.


The holomorphic and the anti-holomorphic component of the stress tensor,
T, T ∈ F (2) .
to each (closed string) world sheet X on R2 an amplitude
O
AX : F ⊗#pts ⊗ Hb −→ C
b⊂∂X   (1.2)
(φ1 , . . . , φm ; v1 , . . . , vn ) 7−→ AX φ1 (p1 ) . . . φm (pm ) (v1 , . . . , vn )

which is linear in its arguments and a smooth function of the pi .


Remark: In this notation the correlation functions on R2 are written as

 
φ1 (p1 ) · · · φm (pm ) = AR2 φ1 (p1 ) · · · φm (pm ) /AR2 . (1.3)

3
Properties of A:
(C1) Scaling: For any collection of fields φi ∈ F (∆i ) and λ ∈ R>0 we have

AR2 φ1 (p1 ) . . . φm (pm ) = λ∆1 +···+∆m AR2 φ1 (λp1 ) . . . φm (λpm ) . (1.4)


   

(C2) Non-degenerate two-point function: On F define a pairing by setting, for


0
any φ ∈ F (∆) and φ0 ∈ F (∆ ) ,
0
(φ, φ0 ) = lim L2 min(∆,∆ ) AR2 φ(L)φ0 (0) .
 
(1.5)
L→∞

This pairing has to be non-degenerate.


Remark: By (C1) the limit is finite. Also, if ∆(φ) 6= ∆(φ0 ) then (φ, φ0 ) = 0.
Choose a basis ϕα , α ∈ I of F such that ϕα ∈ F (∆α ) for some ∆α . Let
(∆)
I = {α ∈ I|∆α = ∆} (note that I (∆) is a finite set). Let ϕ̃α be the dual
basis in the sense that (ϕ̃α , ϕβ ) = δα,β .
For a circle b, let Db be the disc bounded by b, bo be the centre of b and rb
the radius of b.
(C3) State-field correspondence: For each circle b there is a linear isomorphism
ιb : F → Hb s.t. for any φ, φ0 ∈ F,

ADb φ0 (bo ) (ιb (φ)) = [pic 2] = (φ, φ0 ) .


 
(1.6)

Remark: If ιb exists, it is unique: first, d(v, φ) = ADb φ(bo ) (v) = (ι−1


 
b (v), φ)
gives a non-degenerate pairing on Hb × F, which is independent of ιb . Given
two isomorphisms ιb and ι̃b obeying (C3) we thus have d(ιb (φ), φ0 ) = (φ, φ0 ) =
d(ι̃b (φ), φ0 ) for all φ0 ∈ F. Since d is non-degenerate this implies ιb (φ) = ι̃b (φ).
Fix two bases {uα } and {ũα } in Hb by setting uα = ιb (ϕα ) ũα = ιb (ϕ̃α ). Then
in particular  
ADb ϕβ (bo ) (ũα ) = (ϕ̃α , ϕβ ) = δα,β . (1.7)

Given a world sheet X and a circle b, we call b an allowed circle in X if b lies



in X − {marked points}, i.e. it does not intersect any marked point or boundary
component of X. For an allowed circle b in X, we get two new world sheets
Xi (b) (“inside”) and Xo (b) (“outside”) by cutting X along b. More precisely,
Xi (b) = X ∩ Db and Xo (b) = (X − Xi (b)) ∪ b.
(C4) Sum over intermediate states (glueing): Given a world sheet X and an
allowed circle b in X, we have
 
AX φ1 (p1 ) . . . φm (pm ) (v1 , . . . , vn ) = [pic 3]
X  
= AXo (b) φ1 (p1 ) . . . φk (pk ) (v1 , . . . , vl , uα ) (1.8)
α∈I  
× AXi (b) φk+1 (pk+1 ) . . . φm (pm ) (vl+1 , . . . , vn , ũα ) ,

4
where the indices 1, . . . , k belong to fields/points of X which also belong to
Xo (b), and similar for the states v1 , . . . , vl .
(C5) Translation invariance: For all v ∈ R2 and α ∈ I,
   
AR2 φ1 (p1 ) · · · φm (pm ) = AR2 φ1 (p1 + v) · · · φm (pm + v)
   
ADb φ1 (p1 ) · · · φm (pm ) (ũα ) = ADb0 φ1 (p1 + v) · · · φm (pm + v) (ũα ) ,
(1.9)
where b0 = b + v is the translated circle.
There will be one more property, (C6), and a list of assumptions, (A1), further
down.
Further reading: This way of presenting CFT is based on the approach of Segal
[Se1, Se2]. It is also described in section two of the lecture notes of Gawȩdzki
[Gw1, Gw2]. In fact all these references are rather for section 4, as they treat
CFT on general Riemann surfaces. An introduction to CFT which is also based
on amplitudes, rather than on quantising an action is [Gb]. A mathematical
notion closely related to the presentation above is that of operads, see [St1] for
a short introducion and further references.

1.2 Motivation: critical Ising model on the lattice


It is sometimes helpful to think of euclidean QFTs as continuum limits of lattice
models. The state sum on the lattice is then a discrete version of the path
integral.
This section is supposed to be motivational and will be more text based
rather than being as formal as possible.

Ising model
Let N be a square lattice of N × N sites. A configuration of spins is a function
s : N → {±1}. The energy of a configuration is
X
E[s] = − sx sy , (1.10)
hx,yi

where the sum is over neighbouring sites of N . The partition function on N


is a sum over configurations
X
ZN = e−βE[s] , (1.11)
s on N

where β > 0 is the inverse temperature.

5
States
Let q = ∂N be the 4(N −1) lattice sites on the boundary of N . The space of
states on q is the 24(N −1) –dimensional C-vector space

Hq = span f : q → {±1} . (1.12)
P
If u ∈ Hq is a linear combination u = i ci fi then we set
X X
ZN (u) = ci e−βE[s] . (1.13)
i s with s|q =fi

Similarly, one can define

ZX (u1 , u2 , u3 ) for X = [pic 5] (1.14)

etc.

Sum over intermeditate states


Consider X, b, Xo and Xi as in

[pic 4] . (1.15)

For a configuration s on X have


X  X X X 
E[s] = − sx sy = − sx sy + sx sy − sx sy
hx,yi in X hx,yi in Xi hx,yi in Xo hx,yi in q (1.16)

= E[s Xo ] + E[s Xi ] − E[s q ] .

For the partition function this gives


X X X
ZX = e−β(E[so ]+E[si ]−E[sq ]) = ZXo (sq )ZXi (sq )eβE[sq ] (1.17)
sq on q so ,si sq on q

where so,i is a configuration on Xo,i s.t. so,i q = sq The sum (1.17) is the
analogue of (C4).

Fields
A field φ(x) is a map s 7→ φ(x)[s] ∈ C which depends only on spins at sites y
with distance |x − y| < Lφ for some Lφ independent of the lattice size. E.g. the
spin field σ(x)[s] = sx is a field, but the energy s 7→ E[s] is not.
The correlator of spin fields on an infinite lattice is given by
1 X
sx1 · · · sxm e−βE[s]


σ(x1 ) · · · σ(xm ) β
= lim (1.18)
N →∞ ZN
s on N

Correlators of arbitrary fields φ(x) are computed in the same way.

6
Continuum limit
For a generic choice of the inverse temperature β, the correlators will either go
to constants or decay exponentially as we move the points further and further
apart from
√ each other. However, there is a critical inverse temperature βc =
− 12 ln( 2 − 1) ≈ 0.44 (the point of phase transition between the ordered, low
temperature phase and the disordered, high temperature phase) at which the
correlators display a power law behaviour.
For points p1 , . . . , pm ∈ R2 define the continuum limit of a correlator as

σ̂(p1 ) · · · σ̂(pm ) = lim rα σ(rp1 ) · · · σ(rpm ) βc .





(1.19)
r→∞

Here the hat ˆ distinguishes continuum fields from lattice fields. The lattice fields
are to be inserted at the lattice sites closest to the points rpk . The constant α
depends on the fields and is the largest number such that the limit is finite. In
the present case it turns out to be α = m/8.
Note that the continuum correlators are scale covariant by definition

σ̂(λp1 ) · · · σ̂(λpm ) = lim rα σ(rλp1 ) · · · σ(rλpm ) βc





r→∞
α

(1.20)
= 0lim λ−α r0 σ(r0 p1 ) · · · σ(r0 pm ) β = λ−α σ̂(p1 ) · · · σ̂(pm ) ,


r →∞ c

where we set r0 = λr. This is the motivation behind property (C1).


Remark: Here are some words that are used: The continuum limit describes the
long-range behaviour of the lattice model at a critical point. In this limit, the
details of the lattice should not be so important, so that CFTs describe univer-
sality classes of critical behaviour in lattice models. Distinct lattice models can
lie in the same universality class, i.e. have the same CFT as continuum limit.
An important (open) goal is to classify all universality classes by classifying all
CFTs.
Further reading: YYY: Savit review?, or [DMS, section 3], cardy-lecture?, ...

1.3 Properties of the stress tensor


Operator product expansion, the first

Lemma 1.1 :
States can be replaced by fields via

[pic 6] (1.21)

Proof:
[pic 7] . (1.22)


7
Lemma 1.2 :
We have
[pic 8] . (1.23)
Proof:
[pic 9] . (1.24)

The operator product expansion (OPE) is obtained via

[pic 10] . (1.25)

Symbolically we will write


X
φ(p)φ0 (p0 ) = fα (p − p0 ) · ϕα (p0 ) , (1.26)
α∈I

where the functions fα are given by fα (p − p0 ) = ADb φ(p)φ0 (p0 ) (ũα ). By


 

translation invariance (C5), the rhs only depends on p − p0 and by


 lemma
 1.2, it
is independent of rb . The expansion (1.26) holds inside any AX · · · , provided
an appropriate circle b exists.

Identity field

Lemma 1.3 :
Let, for any circle b,
1 = [pic 19] . (1.27)
(0)
This is independent of the choice of b. Further 1 ∈ F and, for any X,
   
AX φ1 (p1 ) . . . φm (pm )1(q) = AX φ1 (p1 ) . . . φm (pm ) . (1.28)

Proof: Independence of b is a consequence of (C5) and lemma 1.2. To see (1.28)


use a sum over intermediate states

[pic 20] . (1.29)

Finally, 1 ∈ F (0) follows since the lhs of (1.28) is independent of q. For this to
be consistent with scaling (C1), 1 cannot have any components outside of F (0) .


The state ιb (1) is the (in-)vacuum state in Hb . It is the state representing an


empty disc glued to b.

8
Conventions for complex variables
Let (x, y) be the coordinate on R2 , then

z = x + iy , z = x − iy , ( )∗ : complex conjugate

∂ 1 ∂ i ∂ ∂ 1 ∂ i ∂ (1.30)
∂= = − , ∂= = + .
∂z 2 ∂x 2 ∂y ∂z 2 ∂x 2 ∂y

Note that ∂z = 1, ∂z = 0, ∂z = 0 and ∂z = 1. Further, given a holomor-


phic function f (z), i.e. ∂f (z) = 0, and an anti-holomorphic function f˜(z), i.e.
∂ f˜(z) = 0, we write
Z  I Z L
dz dγ dt
f (z) = f (z) = f (γ(t))
z

6 γ 2πi 0 dt 2πi
(1.31)
Z  I Z L ∗
idz  dγ idt

f˜(z) = f˜(z) = f˜(γ(t))
z
 6 γ 2π 0 dt 2π

With these conventions we have


Z  Z 
1 1
=1 and =1 . (1.32)
z
 0
6z 
z
0

6z ∗

Axioms for the stress tensor


(C6) Stress tensor:
(i) For all φ1 , . . . , φm ∈ F, p1 , . . . , pm ∈ R2 and all angles θ we have

lim r4 AR2 T (reiθ )φ1 (p1 ) · · · φm (pm ) < ∞ ,


 
(1.33)
r→∞

and the same for T (reiθ ). 


Further, inside any AX . . . we have
(ii) ∂T (z) = 0 = ∂T (z).
(iii) the OPEs
c/2 2 1
T (z)T (w) = 4
1+ 2
T (w) + ∂T (w) + reg(z − w) ,
(z − w) (z − w) z−w
T (z)T (w) = reg(z − w) ,

c̃/2 2 1
T (z)T (w) = 1+ T (w) + ∂T (w) + reg(z − w) ,
(z − w)4 (z − w)2 z−w
(1.34)
for some c, c̃ ∈ C, called the (left/right) central charge.
(iv) for each φ ∈ F
Z  Z 
∂ ∂
T (z)φ(w) = φ(w) , T (z)φ(w) = φ(w) . (1.35)
 
w w
z 6 ∂w z 6 ∂w

9
Remark: The notation T (z) or φ(z) is somewhat non-standard. Instead, one
often finds T (z) and φ(z, z). This would lead to confusion on the upper half
plane later on. The convention here is that in φ(z), z = x+iy gives the insertion
point and the notation gives no information about (anti-)holomorphicity of a
field.
Remark: Translational invariance (C5) is actually a consequence of the other
properties. To show this, one would first introduce (C1)–(C4) and (C6) (i),(ii),(iv).
This implies translational invariance on R2 by a contour deformation argument
as in theorem 1.15 below.
 To get translational invariance for the discs one shows
that also AR2 −Db φ(L) (ιb (φ0 )) is a non-degenerate pairing for any L outside
Db . Translational invariance on R2 can then be seen to imply translational in-
variance for discs.
Having shown translational invariance, one can define the identity field as in
lemma 1.3 (which used translational invariance) and state property (C6)(iii).
While in this way of proceeding one needs one property less, it also (in my
opinion) makes the presentation less clear.
Further reading: The properties (C6) can be motivated from the classical stress
tensor of a conformally invariant action. See [Gi, sections 2.2, 3.1], [Sc, sections
2.1, 2.9] and [DMS, section 4.3] for more details.

1.4 Examples: A topological and a meromorphic CFT


A topological CFT
Let A be a finite dimensional commutative unital Frobenius algebra, with a
trace ε : A → C. Frobenius means that the form (a, b) 7→ ε(a · b) on A × A is
non-degenerate.
Set S = {0}, F = F (0) = A, Hb = A for all b, T = T = 0 and in consequence
c = c̃ = 0. For the amplitudes on the plane and on the disc we take
 
AR2 φ1 (p1 ) · · · φm (pm ) = ε(φ1 · φ2 · · · φm ) ,
  (1.36)
ADb φ1 (p1 ) · · · φm (pm ) (u) = ε(u · φ1 · · · φm ) ,

where u, φ1 , . . . , φm ∈ A. This theory is topological as its correlators are inde-


pendent of the insertion points of the fields.
From the properties,
– (C1), (C3), (C5), (C6) are fulfilled trivially,
– (C2) holds because of non-degeneracy of the pairing ε(a · b),
– To check (C4) one needs associativity of A.

Chiral algebra of a free boson


The theory sketched (incompletely) in the following is part of what one finds
when canonically quantising a massless free boson, i.e. a real scalar field φ(x, y)

10
R
with action ∂φ∂φdxdy (so there are no interaction terms). The quantum field
∂φ(x, y) turns out to be holomorphic (its classical EOM are ∂∂φ = 0) and is
the field called J below (up to a constant factor).
We have S = Z≥0 . Of the F (∆) we will for the moment only need F (0) = C1
and F (1) = CJ, where J is a holomorphic field, called the U (1)-current. It
obeys
1
∂J(z) = 0 and J(z)J(w) = 1 + reg(z − w) . (1.37)
(z − w)2


Further, for correlators on the plane, J(z) · · · → 0 when z → ∞. The stress
tensor will be described in section 1.6. To give the theory a name, we will call
it the chiral algebra of a free boson.
The OPE allows us to work out multiple correlators of J by extracting the
pole structure. For example, considering the correlators below as functions of
z1 ,


• J(z1 ) = reg(z1 ), but the only regular
analytic function on C vanishing
at infinity is the constant function zero, J(z1 ) = 0.

• J(z1 )J(z2 ) = z12−2 +reg(z1 ) = z12−2 , where here and below zij = zi −zj .



• J(z1 )J(z2 )J(z3 ) = 0, as each pole gets multiplied by a one-point func-
tion of J, which is zero.


• J(z1 )J(z2 )J(z3 )J(z4 )
= z12−2 J(z3 )J(z4 ) + z13−2 J(z2 )J(z4 ) + z14−2 J(z2 )J(z3 ) + reg(z1 )



= z12−2 z34−2 + z13−2 z24−2 + z14−2 z23−2 .


etc. We have thus worked out the first exact quantum correlators (if only of a
free theory).
Remark: Note also that the method above amounts to Wick contractions in free
theories. The reason is the very simple form of the J-J-OPE (1.37). However,
the technique of extracting poles is more general than Wick contration and can
be applied to more involved OPEs, like the T -T -OPE (1.34).
(Exercise: Verify that T (z1 )T (z2 )T (z3 ) = cz12−2 z13−2 z23−2 .)

Further reading: To see the canonical quantisation carried out explicitly, see
section 3.4 of [Sc] or sections 5.3.1, 6.3.1 of [DMS]. To find formulas for n-point
correlators of J(z) or T (z) see sections 3.7.1 and 3.7.3 in [Gb].

1.5 The Virasoro algebra


The modes Lm and Lm of the stress tensor
For a circle b and an integer m we set
I I
dz idz
Lm (b) = (z −bo ) m+1
T (z) and Lm (b) = (z ∗ −b∗o )m+1 T (z) (1.38)
b 2πi b 2π

11
 
These are understood as inserted in some AX · · · , where b is any allowed circle
in X.
Remark: The modes Lm and Lm are complex contour integrals. The integration
contour can thus be deformed without changing the value of the integral. In
this sense, Lm and Lm are conserved charges of the CFT.
For a string of modes with the same b we define
Lm1 (b)Lm2 (b) . . . Lmn (b) = Lm1 (b1 )Lm2 (b2 ) . . . Lmn (bn ) (1.39)
k
where, if b has radius r, the b are circles with the same centre bo but with
radius rk = r − εk, for some small enough ε > 0, see [pic 11]. Given X, “small
enough” means that the annulus bounded by b and bn does not contain any
marked points of X or intersect the boundary of X.
Remark: A small enough ε always exists and definition (1.39) is independent of
the precise choice of ε. However, it does depend on the ordering of the Lmk (b).

Theorem 1.4 :  
For a world sheet X and an allowed circle b in X, inside AX · · · we have the
identities, for m, n ∈ Z,
c 3
Lm (b)Ln (b) − Ln (b)Lm (b) = (m−n)Lm+n (b) + 12 δm+n,0 (m −m) ,
c̃ 3
Lm (b)Ln (b) − Ln (b)Lm (b) = (m−n)Lm+n (b) + 12 δm+n,0 (m −m) ,

Lm (b)Ln (b) − Ln (b)Lm (b) = 0 .


(1.40)
Proof: We will show the first identity, in the special case bo = 0.
(i) Let Y be the annulus [pic 12] and let γ1 , γ2 be contours as in [pic 12]. Then
I I
  dw n+1 dz m+1  
AY Lm (b)Ln (b) = w z AY T (z)T (w) . (1.41)
γ2 2πi γ1 2πi

For fixed w, deform the z-contour as follows


[pic 13] . (1.42)
For fixed w, the γ4 -integration gives
I
dz m+1  
z AY T (z)T (w)
γ4 2πi
I
1) dz m+1 c/2   2  
= z 4
AY 1 + 2
AY T (w)
γ4 2πi (z − w) (z − w)
1   
+ AY ∂T (w) + reg(z − w)
z−w
c 1 ∂ 3 m+1   ∂ m+1  
AY T (w) + wm+1 AY ∂T (w)
  
= w AY 1 + 2 w
2 3! ∂w3 ∂w
 c
(m + 1)m(m − 1)wm−2 1 + 2(m + 1)wm T (w) + wm+1 ∂T (w)

= AY
12
(1.43)

12
where in 1) the T T -OPE of (C6) is inserted. Continue with (1.41),
 
AY Lm (b)Ln (b)
I I
1) dw n+1 dz m+1  
= w z AY T (z)T (w)
γ2 2πi γ3 2πi
I I
dw n+1 dz m+1  
+ w z AY T (z)T (w)
γ2 2πi γ4 2πi
2)  
= AY Ln (b)Lm (b)
I
dw n+1  c
(m3 −m)wm−2 1 + 2(m + 1)wm T (w) + wm+1 ∂T (w)

+ w AY
γ2 2πi 12
3)  
= AY Ln (b)Lm (b)
 c
(m3 −m)δm+n,0 + 2(m + 1)Lm+n (b) − (m + n + 2)Lm+n (b)

+AY
12
(1.44)
where 1) the contour deformation (1.42) is applied to γ1 , 2) the ordering of
γ2 and γ3 results in Ln (b)Lm (b), 3) the γ2 integral is carried out, resulting in
modes according to (1.38). Thus
 
AY Lm (b)Ln (b) − Ln (b)Lm (b)
c (1.45)
= AY (m−n)Lm+n (b) + δm+n,0 (m3 −m) .
 
12

(ii) In (i) the claim was established in the special case X = Y . For general X
cut out an annulus around b and then use (i)

[pic 14] . (1.46)

The Virasoro algebra


The Virasoro algebra Vir is an infinite dimensional Lie algebra. As a vector
space it has basis vectors {Lm |m ∈ Z} ∪ {C} and the Lie bracket is defined to
be
1
[Lm , Ln ] = (m−n)Lm+n − δm+n,0 (m3 −m)C ,
12 (1.47)
[Lm , C] = 0 .
The generator C is a central element of Vir.
Remark: Let X be a world sheetN and b an allowed circle in X. Denote by V
⊗#pts
the vector
 space  F ⊗ b⊂∂X Hb . Then the above theorem show that
AX . . . 7→ AX Lm (b) . . . defines a representation of Vir on an appropriate
subspace of V ∗ . The generator C gets represented by the scalar c.

13
Action of Vir on the space of fields

(out-states) Define the notation


 
(φ, φ1 (p1 ) · · · φm (pm )) = ADb φ1 (p1 ) · · · φm (pm ) (ιb (φ)) , (1.48)

where b is a circle with bo = 0 and radius large enough such that all pi lie in
Db . By lemma 1.2 the definition (1.48) is independent of the precise choice of
radius.
Let b be a circle with bo = 0 and define a linear map Lm : F → F by setting,
for φ ∈ F (∆) ,
X
ϕ̃α , Lm (b)φ(0) ϕα ∈ F (∆−m) .

Lm φ = [pic 15] = (1.49)
α∈I (∆−m)

By contour deformation, this is independent of the choice of b. Similarly, define


Lm : F → F as
X
ϕ̃α , Lm (b)φ(0) ϕα ∈ F (∆−m) .

Lm φ = (1.50)
α∈I (∆−m)

Lemma 1.5 :  
Let b be a small circle around p = bo . Then inside any AX . . . we have
Lm (b)φ(p) = (Lm φ)(p).
Proof: Given X, via (C4) we can cut a disc D centered at p and containing b.
By lemma 1.1 the resulting hole in X can be replaced by a field insertion at p,
  X    
AX · · · Lm (b)φ(p) = AX · · · ϕα (p) AD Lm (b)φ(p) (ũα ) . (1.51)
α∈I

By translation invariance (C5) the amplitude of the disc centered  at p is equal
to the amplitude  of the disc centered at zero, so that A D Lm (b)φ(p) (ũα ) =
ϕ̃α , Lm (b)φ(0) . If we can now show that the sum over I in (1.51) restricts to
α ∈ I (∆−m) we are done.
0
Now note that for any φ0 ∈ F (∆ ) and p0 6= 0 we have, for a small circle b with
bo = 0,
I

0 0
1) dz m+1
0
φ (λp0 )T (z)φ(0)

φ (λp )Lm (b)φ(0) = z
b 2πi
I
2) dz m+1 −∆−∆0 −2
0 0
= z λ φ (p )T (z/λ)φ(0)
2πi (1.52)
Ib
3) λdζ m+1 −∆−∆0 −2

0 0
= (λζ) λ φ (p )T (ζ)φ(0)
b 2πi
4) 0
= λ−∆−∆ +m φ0 (p0 )Lm (b)φ(0)

14
where 1) insert the definition of Lm (b); 2) use scaling covariance (C1); 3) sub-
stitute the integration variable z by ζ = z/λ (this also changes the integration
contour, but we can deform the resulting

contour back to b); 04) use
again the def-
inition of Lm (b). On the other hand φ0 (λp0 )ϕα (0) = λ−∆ −∆α φ0 (p0 )ϕα (0) .

0
Specialising (1.51) to a 2-point function on the pane we find, for any φ0 ∈ F (∆ )
and p0 6= 0,

0 X
0
φ (λp0 )Lm (b)φ(p) = φ (λp0 )ϕα (p) ϕ̃α , Lm (b)φ(0) .

(1.53)
α∈I

Since both sides have to have the same λ-dependence, we concludes that ϕ̃α , Lm (b)φ(0)
can be non-zero only if ∆α = ∆ − m. 

Theorem 1.6 :
The assignments Lm 7→ Lm ∈ End(F), C 7→ cidF and Lm 7→ Lm ∈ End(F),
C 7→ c̃idF define two mutually commuting representations of Vir on F.
Proof: We will show only the first statement. Let φ ∈ F (∆) , then for a circle b
with bo = 0,

(Lm Ln − Ln Lm )φ
X (1.54)
= ϕ̃α , (Lm (b)Ln (b) − Ln (b)Lm (b))φ(0)) ϕα
α∈I (∆−m)

On the rhs we have an amplitude on a disc, to which we can apply theorem 1.4.
This then implies the present theorem. 

Laurent expansions

Lemma 1.7 :
For b an allowed circle in X (with z not among the marked points) we have, for
z ∈ b, X
T (z) = (z − bo )−m−2 Lm (b) . (1.55)
m∈Z

Proof: As in the proof of theorem 1.4, cut out a small annulus around b, call it
Y . It is enough to show that for any two states ũα , uβ we have
X
z −m−2 AY Lm (b) (ũα , uβ )
   
AY T (z) (ũα , uβ ) = (1.56)
m∈Z

Now the lhs is an analytic function on Y and hence has a convergent Laurent
expansion around bo ,
X
am (z − bo )−m−2 ,
 
AY T (z) (ũα , uβ ) = (1.57)
m∈Z

15
for some am ∈ C. The am can be computed by contour integration
I
dz
(z − bo )m+1 AY T (z) (ũα , uβ ) = AY Lm (b) (ũα , uβ ) .
   
am = (1.58)
b 2πi


Laurent expansions of this kind exist for all holomorphic fields W (z) (i.e. fields
with ∂W (z) = 0).

Simplifying assumptions

A field φ ∈ F is called primary if

Lm φ = 0 = Lm φ for all m > 0 . (1.59)

(Excercise: Show that 1 is primary, using lemma 1.3.)


Given a field φ ∈ F, then all fields ψ of the form

ψ = L−m1 · · · L−mk L−n1 · · · L−nl φ where mi , ni ∈ Z>0 , (1.60)

or linear combinations thereof, are called descendents of φ.


(Excercise: Show that T = L−2 1. Thus T is a descendent of 1)
(A1) We assume the following properties for the action of Vir × Vir on F:
(i) L0 and L0 can be simultaneously diagonalised on F.
(ii) Every field is either primary or a descendent of a primary field.
We can thus write F = ⊕F (h,h) where for φ ∈ F (h,h) we have L0 φ = hφ and
L0 φ = hφ. The notation h is not to mean that h is complex conjugate to h.
Write P (h,h) for the subspace of F (h,h) consisting of primary fields.
Theorem 1.15 below will imply that h + h = ∆, i.e. that F (h,h) ⊂ F (∆) . In
particular F (h,h) is also finite-dimensional. Let us choose the basis ϕα such that
ϕα ∈ F (hα ,hα ) .
If φ ∈ F (h,h) then e.g. L−m φ ∈ F (h+m,h) , since

L0 Lm φ = ([L0 , Lm ] + Lm L0 )φ = (mLm + hLm )φ . (1.61)

Lemma 1.8 :
Let φ ∈ F (h,h) . Then φ is primary iff
h 1
T (z)φ(w) = φ(w) + ∂φ(w) + reg(z − w) and
(z − w)2 z−w
(1.62)
h 1
T (z)φ(w) = ∗ φ(w) + ∗ ∂φ(w) + reg(z − w) .
(z − w∗ )2 z − w∗

16
Proof:
(i) Suppose φ is primary. Use the Laurent expansion (lemma 1.7) to get, for a
small circle c with co = w,
X
T (z)φ(w) = (z−w)−m−2 Lm (c)φ(w)
m∈Z (1.63)
1) −2 −1
= (z−w) L0 (c)φ(w) + (z−w) L−1 (c)φ(w) + reg(z−w) ,

where 1) Lm (c)φ(w) = (Lm φ)(w) = 0 for m > 0 so that the only singular terms
are m = 0, −1. Using L0 (c)φ(w) = (L0 φ)(w) = hφ(w) and L−1 (c)φ(w) = ∂φ(w)
(this is (C6)(iv)) gives the first OPE in (1.62). The second OPE can be seen in
the same way.
(ii) Suppose the OPEs (1.62) hold. Then, for c a small circle around zero,
I
dz m+1
(Lm φ)(0) = Lm (c)φ(0) = z T (z)φ(0)
c 2πi
I (1.64)
1) dz
hz m−1 φ(0) + z m ∂φ(0) + z m+1 reg(z) = 0

= if m ≥ 1 .
c 2πi

where 1) the OPE (1.62) has been substituted. The argument that Lm φ = 0
for m ≥ 1 works analogously. 

1.6 The Heisenberg algebra

Remark: Just as was the case for T , any holomorphic field gives rise to an
algebra of modes via contour integration. Sometimes this is a Lie algebra (more
generally, a commutator [Am , Bn ] of modes can be a polynomial in the modes,
not a linear combination. This is called a W-algebra, see e.g. [BS]).
Take the chiral algebra of a free boson and define the modes
I
dz
Jm (b) = (z − bo )m J(z) . (1.65)
b 2πi

If bo = 0 we will also write Jm , Lm etc instead of Jm (b), Lm (b). The radius


of b is then implied by radial ordering. For example, by [Jm , J(z)] it is meant

[Jm , J(z)] = Jm J(z) − J(z)Jm = [pic 16] (1.66)

Lemma 1.9 :
We have [Jm , J(z)] = mz m−1 and [Jm , Jn ] = mδm+n,0 .
Proof: Have
Z  Z 
1
[Jm , J(z)] = ζ m J(ζ)J(z) = ζm = mz m−1 , (1.67)
ζ ζ
z z
6 6 (ζ − z)2

17
as well as
Z  Z 
n
[Jm , Jn ] = z [Jm , J(z)] = mz n+m−1 = mδm+n,0 . (1.68)
z
0

6 z
0

6

The algebra spanned by the modes Jm , m ∈ Z with commutation relations


[Jm , Jn ] = mδm+n,0 is called Heisenberg algebra H.
In the same way as for the Virasoro algebra one shows that H acts on F. E.g.
for the field J−1 1 we have
Z 
(J−1 1)(0) = z −1 J(z)1 = J(0) , (1.69)
z
0

6

so that J−1 1 = J.
The Laurent expansion of J(z) reads
X
J(z) = z −m−1 Jm . (1.70)
m∈Z

(Exercise: This implies that ∂J(0) = (J−2 1)(0))

The stress tensor


Define T = 21 J−1 J ∈ F (2) and T = 0. We will check some of the properties in
(C6) to ensure this is a stress tensor. Note

1  1
Z
T (z) = J(ζ)J(z) . (1.71)
2 ζ
z
6ζ − z

Lemma 1.10 :
The Laurent expansion of T (z) is
1 X −m−2 X
T (z) = z : Jk Jm−k : , (1.72)
2
m∈Z k∈Z

where : · · · : denotes the normal ordering of modes (i.e. the modes inside
: · · · : are to be ordered s.t. the mode number increases towards the right, e.g.
: J3 J−2 J2 : = J−2 J2 J3 ).
Proof: Depending on whether |ζ| is larger or smaller than |z|, there are two
ways to write a convergent power series for (ζ − z)−1 ,

−1
P∞  z k
ζ ; |ζ| > |z|

1 
k=0 ζ
=  k (1.73)
ζ −z  −z −1 ∞
P ζ
; |z| > |ζ|
k=0 z

18
Using this we can rewrite
I I
1) 1 dζ 1 1 dζ 1
T (z) = J(ζ)J(z) − J(ζ)J(z)
2 |ζ|>|z| 2πi ζ − z 2 |ζ|<|z| 2πi ζ − z
∞ ∞
2) 1 X k 1 X −k−1
= z J−k−1 J(z) + z J(z)Jk
2 2
k=0 k=0
−∞ ∞ (1.74)
3)1 X −k0 −1 1 X −k−1
= z Jk0 J(z) + z J(z)Jk
2 0 2
k =−1 k=0
−∞ ∞
4)1 X X −k−l−2 1 X X −k−l−2
= z Jk Jl + z Jl Jk
2 2
k=−1 l∈Z k=0 l∈Z

where 1) deform the small circular contour around z; 2) insert the expansion
(1.73); 3) rename the first summation variable as k 0 = −k−1; 4) insert the
Laurent expansion of J(z). 
(Exercise: Using lemma 1.10, show that ∂T (0) = (J−2 J−1 1)(0))
Remark: Lemma 1.10 also shows that (1.71) is identical to the definition T (z) =
1
2 : J(z)J(z) : via the normal ordering of modes as used e.g. in section 2.3 of
[Gi], section 3.6 of [Sc] or section 5.3.1 of [DMS].

Lemma 1.11 :
We have the OPEs
T (z)J(0) = z −2 J(0) + z −1 ∂J(0) + reg
(1.75)
1 −4
T (z)T (0) = z + 2z −2 T (0) + z −1 ∂T (0) + reg .
2
Proof: We will show the first OPE. Using lemma 1.10 we write
1 X −m−2 X
T (z)J(0) = z : Jk Jm−k : J−1 1 + reg(z)
2 m>−2
k∈Z
= 12 z −2 (: J1 J−1 : J−1 + J−1 J1 J−1 )1 + 12 z −1 (J−2 J1 J−1 + J1 J−2 J−1 )1
(1.76)
since for m = −1 we get contributions only for k = −2, 1 and for m = 0 from
k = −1, 1. All other terms in the double sum are zero. Using the commutation
relations in lemma 1.9 one checks that this amounts to the first OPE. 

(Excercise: Verify the second OPE in lemma 1.11)


(Excercise: Show that L0 J−1 1 = J−1 1, i.e. J ∈ F (1,0) . Hint: use the first OPE
in (1.75) to derive [Lm , Jn ] = −nJm+n .)
Remark: From lemma 1.11 we learn that J is a primary field and that T (z) obeys
(C6)(iii). In particular the (left) central charge of the free boson chiral algebra
is c = 1.

19
1.7 Conformal transformations

Theorem 1.12 :
Let b be any circle.
(i) (invariance of in-vacuum) For c a circle in Db (and no further marked points)
we have
   
ADb Lm (c) = 0 = ADb Lm (c) for all m ≥ −1 . (1.77)

(ii) (invariance of out-vacuum) For c a circle in R2 − Db we have

AR2 −Db Lm (b0 ) = 0 = AR2 −Db Lm (b0 ) ,


   
for all m ≤ 1 . (1.78)

Proof:    
(i) Just uses that the amplitudes ADb T (z) and ADb T (z) have no poles on
Db .
(ii) We will show the first identity. By contour deformation AR2 −Db Lm (b0 ) is
 

independent of the choice of b0 . In particular, for a circle c(r) of radius r,

AR2 −Db Lm (b0 ) = lim AR2 −Db Lm (c(r)) .


   
(1.79)
r→∞

For any α ∈ I
I
  dz m+1   r→∞
AR2 −Db Lm (c(r)) (uα ) = z AR2 T (z)ϕα (bo ) −→ 0 , (1.80)
c(r) 2πi

since for large r, AR2 T (z)ϕα (bo ) ∼ z −4 s.t. the integrand goes as z m−3 . Thus
 

the integral is zero for m ≤ 1. 

Lemma 1.13 :
For φ ∈ P (h,h) we have

[Lm , φ(z)] = z m (h(m + 1) + z∂)φ(z)


(1.81)
[Lm , φ(z)] = (z ∗ )m (h(m + 1) + z ∗ ∂)φ(z) .

Proof: We will show the first equality. We have


Z  Z  
1) h 1 
ζ m+1 T (ζ)φ(z) = ζ m+1 φ(z) + ∂φ(z) ., (1.82)
ζ ζ
z z
6 6 (ζ − z)2 ζ −z

where 1) insert the OPE from lemma 1.8. Evaluating the contour integral gives
the desired result. 

20
Local conformal transformations

Consider the infinitesimal conformal transformation f (z) = z − εz m+1 . Con-


catenating this with an analytic function g(z) gives, up to O(ε2 ),
g(f (z)) = g(z) − εz m+1 g 0 (z) = (1 + `m )g(z) where `m = −z m+1 ∂ . (1.83)
The vector fields `m generate infinitesimal conformal transformations. Their
Lie-bracket is
[`m , `n ] = (m − n)`m+n . (1.84)
This is called the Witt algebra.
In fact, the conserved charges Lm implement infinitesimal conformal trans-
formation on the fields of the CFT. Their algebra, the Virasoro algebra, is a
central extension of the Witt algebra.
The vector fields `n (or rather −`n ) can be integrated to give, for n ∈ Z≥−1 ,
the following analytic maps
(
z · (1 − nuz n )−1/n ; n 6= 0
fn,u (z) = (1.85)
eu z ; n=0
These are one-to-one in some neighbourhood of zero, depending on u and n.

Lemma 1.14 :
Let b, c be circles with center at zero such that c lies in Db . Let u ∈ C and
f (z) = fn,u (z) for some n ≥ −1. Set Uf (c) = exp(uLn (c) + u∗ Ln (c)), then for
u small enough, s.t. f (pi ) lies in c for all i we have
   
ADb Uf (c)φ1 (p1 ) · · · φm (pm ) = ADb φ̃1 (f (p1 )) · · · φ̃m (f (pm )) , (1.86)

for φi ∈ P (hi ,hi ) and


φ̃i = f 0 (pi )hi (f 0 (pi )∗ )hi φi . (1.87)
Proof: We will show the theorem for n 6= 0. Fix u ∈ C and let, for r ∈ R,
fr (z) = z · (1 + nruz n )−1/n . Denote the lhs of (1.86) by F (r) and the rhs by
G(r). In both cases we take f = fr .
d
We will show dr (F (r) − G(r)) = 0. Since F (0) = G(0) this establishes the
theorem. First note that fr (fs (z)) = fr+s (z) such that it is enough to show
d
dr (F (r) − G(r))
r=0
= 0. Up to terms of O(ε2 ) we have fε (p) = p + εupm+1
and
Y m

1 + ε(n+1)upin hi ) 1 + ε(n+1)(upin )∗ hi )

G(ε) =

i=1
φ1 (p1 + εup1n+1 ) · · · φm (pm + εupmn+1

)
n Xm  o
= 1 + ε hi (n+1)upin + hi (n+1)(upin )∗ + upin+1 ∂i + (upin+1 )∗ ∂ i

i=1
φ1 (p1 ) · · · φm (pm ) ,
(1.88)

21

where ∂i is the holomorphic derivative ∂p i
and ∂ i the corresponding antiholo-
morphic derivative.
Further, combining theorem 1.12 and lemma 1.13 and using contour deforma-
tion, we find
m

X

Ln φ1 (p1 ) · · · φm (pm ) = φ1 (p1 ) · · · [Ln , φi (pi )] · · · φm (pm )
i=1
m
X (1.89)
hi (n+1)pin pin+1 ∂i


= + φ1 (p1 ) · · · φm (pm ) ,
i=1

as well as
m
X
hi (n+1)(p∗i )n +(p∗i )n+1 ∂ i



L1 φ1 (p1 ) · · · φm (pm ) = φ1 (p1 ) · · · φm (pm ) .
i=1
(1.90)
Now, again up to terms of O(ε2 ),

F (ε) = (uLn + u∗ Ln )φ1 (p1 ) · · · φm (pm ) = G(ε) .




(1.91)

Remark: Lemma 1.14 can also be formulated for arbitrary analytic maps f on
Db . In this case Uf will take a more complicated form (see e.g. section 6.3.1 in
[FB] to get an idea).
We have seen in the end of section 1.7 that the stress tensor T is not a primary
field. Accordingly, its behaviour under local conformal transformations is more
complicated. If we write, as in lemma 1.14,
   
ADb Uf (c) · · · T (z) = ADb · · · T̃ (f (z)) , (1.92)

we have the following expression for T̃ ,

c  f 000 (z) 3  f 00 (z) 2


T̃ = f 0 (z)2 T +

f ; z 1 and f; z = 0 − (1.93)
12 f (z) 2 f 0 (z)

is the Schwarzian derivative. This can be proved in a similary way as lemma


1.14. To verify (1.93) infinitesimally, first compute
c
[Lm , T (z)] = (m3 −m)z m−2 1 + 2(m+1)z m T (z) + z m+1 ∂T (z) . (1.94)
12
(Excercise: Verify this. Hint: Compare to the last line in (1.43).)
Then take f (z) = z + εz m+1 and develop both sides of (1.92) to first oder in ε.
(Excercise: Verify also this.)
Further reading: For more details about the transformation behaviour of T see
section 3.1 of [Gi], section 6.3 of [Sc] or section 5.4.1 of [DMS]

22
The group of global conformal transformations

The group of global conformal transformations (or Möbius transformations)


M consists of invertible analytic maps from the Riemann sphere P1 to itself.
They are all of the form
az + b
f (z) = where a, b, c, d ∈ C , ad − bc = 1 . (1.95)
cz + d

Define a map ρ : SL(2, C) → M via


 
a b az + b
7−→ f (z) = . (1.96)
c d cz + d

This is a group homomorphism ρ(AB) = ρ(A) ◦ ρ(B). Since ker(ρ) = {±id} we


have M ∼ = SL(2, C)/Z2 .
The group M is generated by the three families of maps
translations rescalings special conf. transf.
u z
tu (z) = z + u ru (z) = e z su (z) = 1−uz
∗ ∗ ∗
euL−1 +u L−1
euL0 +u L0
euL1 +u L1

Theorem 1.15 :
(Möbius covariance) For f ∈ M and φ1 , . . . , φm such that φi ∈ P (hi ,hi ) we have



φ1 (p1 ) · · · φm (pm ) = φ̃1 (f (p1 )) · · · φ̃m (f (pm )) , (1.97)

provided f (pi ) 6= ∞ for i = 1, . . . , m. Here φ̃i = f 0 (pi )hi (f 0 (pi )∗ )hi φi .


Proof: It is enough to verify (1.97) for f = tu , ru , su . For these, by lemma 1.14
we have



Uf (c)φ1 (p1 ) · · · φm (pm ) = φ̃1 (f (p1 )) · · · φ̃m (f (pm )) , (1.98)

where Uf (c) = exp(uLm (c)


+ u∗ Lm (c)) with m
= 0, ±1. By theorem 1.12, for
these values of m we have Uf (c)φ1 (p1 ) · · · = φ1 (p1 ) · · · as all but the first
term in the expansion of the exponential vanish. 

One–, two– and three–point correlators

Given three distinct points z1 , z2 , z3 ∈ P1 there exists a unique Möbius trans-


formation f such that f (z1 ) = 0, f (z2 ) = 1 and f (z3 ) = 2. It follows that one
can map any three points on the Riemann sphere to any other three points via
a Möbius transformation.

23
Lemma 1.16 :
For φi ∈ P (hi ,hi ) , i = 1, 2, 3,


φ1 (p1 ) = C(φ1 )δh1 ,0 δh1 ,0

= C(φ1 , φ2 )δh1 ,h2 δh1 ,h2 (p12 )−2h1 (p∗12 )−2h1




φ1 (p1 )φ2 (p2 )


φ1 (p1 )φ2 (p2 )φ3 (p3 ) = C(φ1 , φ2 , φ3 )×

(p12 )h3 −h1 −h2 (p13 )h2 −h1 −h3 (p23 )h1 −h2 −h3

(p∗12 )h3 −h1 −h2 (p∗13 )h2 −h1 −h3 (p∗23 )h1 −h2 −h3
(1.99)
where pij = pi − pj . Further C : F ⊗ F ⊗ F → C and we used the abbreviations
C(φ1 ) = C(φ1 , 1, 1), C(φ1 , φ2 ) = C(φ1 , φ2 , 1).
Proof: For the 3pt-correlator use a Möbius transformation
that takes p1 7→ 0,
p

2 →
7 1, p 3 →
7 2 and
apply theorem 1.15 to relate φ1 (p 1 )φ2 (p2 )φ3 (p3 ) and
φ1 (0)φ2 (1)φ3 (2) .
The 2pt-correlator is obtained by setting φ3 = 1, using lemma 1.3. The δ-
functions arise because the resulting correlator has to be independent of p3 .
For the 1pt-correlator set also φ2 to 1. 

Comparing the definition of the bilinear pairing in (C2) to lemma 1.16 gives
(φ, φ0 ) = C(φ, φ0 )/AR2 for any two φ, φ0 ∈ P.

1.8 Reconstruction from three-point correlators


If we are given a CFT on C, one may wonder which parts of the data one has
to remember in order to reconstruct the CFT uniquely. It turns out that one
only needs to know F as a Vir × Vir-module and (P1 , P2 (1)P3 (0)) for all primary
fields P1 , P2 , P3 .

Lemma 1.17 :
The functions (φ1 , φ2 (z)φ3 (0)), for φ1,2,3 ∈ F are all uniquely fixed in terms of
(P1 , P2 (1)P3 (0)) for primary fields P1,2,3 .
The (somewhat technical) proof is in appendix A.2.

Theorem 1.18 :
To fix A uniquely, it is enough to give F as a Vir×Vir-module and (P1 , P2 (1)P3 (0))
for all primary fields P1 , P2 , P3 .
The quantities C123 = (P1 , P2 (1)P3 (0)) are also called structure constants of
a CFT. Taking a limit of (1.99) we see that they are (up to the normalisation
factor AR2 ) just the constants appearing in the correlator of three primary fields.
In this sense, in CFT the three-point correlators of primary fields fix the entire
theory.

24
Clearly, not any set of numbers C123 will give a CFT. A good analogy is that
of a C-algebra, for which one has to fix the multiplication mabc , but not every set
of numbers mabc will solve the associativity constraint. For a CFT, there is also
an associativity constraint (coming from the 4-point function on the complex
plane), which the numbers C123 have to solve.
Define the out-vacuum Ωout ∈ F via
X  
Ωout = AR2 ϕα (0) ϕ̃α . (1.100)
α∈I (0)

(Excercise: Show that placing Ωout on the boundary of a disc Db amounts to


glueing R2 − Db (without any field insertions) to Db .)
Proof of theorem:
(i) To obtain the amplitudes AX for all world sheets X it is enough to know all
amplitudes AD for discs D (by translation invariance we can take the discs to
be centered at zero). To see this, note that holes in X can be replaced by little
discs with a field inserted in the center, see lemma 1.1. In this way we arrive
either at a disc with field insertions or at R2 with field insertions. In the first
case we are done; in the second case we can use the out-vacuum to write
 
AR2 φ1 (p1 ) · · · φm (pm ) = (Ωout , φ1 (p1 ) · · · φm (pm )) , (1.101)
where by definition the rhs is an amplitude on a disc centered at zero and large
enough to contain all field insertions.
(ii)Let D be a disc centered
 at zero. To compute an amplitude of the form
AD φ1 (p1 ) · · · φm (pm ) , cut D into concentric rings Yk such that each Yk con-
tains only1 the field φk (pk ),
[pic 18] . (1.102)
Summing over intermediate states expresses the amplitude on R2 as an infinite
sum involving terms of the form
 
AYk φk (pk ) (ũα , uβ ) = (ϕ̃α , φk (pk )ϕβ (0)) . (1.103)
Now, by lemma 1.17, from the set of all (P1 , P2 (1)P3 (0)) we can recover all the
quantities (ϕ̃α , φk (pk )ϕβ (0)). 

In fact, what we have used in the proof is the operator formalism for CFT. In
this formalism, the fields are operators on the state-spaces or, since they are all
isomorphic to F, operators on F. More correctly, we have to allow for infinite
sums, so that we should use an appropriate closure F of F. Then we define
O(φ, z) : F −→ F
X (1.104)
ϕα 7−→ (ϕ̃β , φ(z)ϕα (0)) · ϕβ .
β∈I
1 This is not possible if two fields have the the same distance form zero. In this case just

choose another point as the center. As a result, the outermost ring will not be concentric, but
instead (via lemma 1.1) results in a one-point function on a disc of the form (φ, φ0 (z)), i.e.
with the field not inserted at zero. The proof works in the same way.

25
The construction in the theorem now amounts to the identity
 
AR2 φ1 (p1 ) · · · φm (pm ) = (Ωout , O(φ1 , p1 ) · · · O(φm , pm )1) , (1.105)

where X  
Ωout = AR2 ϕα (0) ϕ̃α (1.106)
α∈I (0)

is the out-vacuum.
(Excercise: Show that placing Ωout on the boundary of a disc Db amounts to
glueing R2 − Db (without any field insertions) to Db .)
The rhs (1.105) is a powerseries in the pk which will converge only for p1 > p2 >
· · · > pm . This is referred to as radial ordering in the operator formalism. The
conventional notation for (1.105) is h0|φ1 (p1 ) · · · φm (pm )|0i.
Remark: We could have started with the operator formalism, however I prefer
the following point of view. What we want is an object with properties (C1)–
(C6), which are motivated by path integral and lattice considerations. Such an
object can be realised by operators on a space of states.
As an analogy, consider C-algebras. These are defined by giving a vectorspace
A together with a multiplication (a, b) 7→ a × b. The multiplication should
be associative (a × b) × c = a × (b × c). Instead, one could have given a
map ρ : A → End(A), related to the multiplication via ρ(a)b = a × b. The
associativity condition then reads ρ(ρ(a)b)c = ρ(a)ρ(b)c. The first point of view
is more symmetric and is also more commonly used than the second.

2 Chiral conformal field theory


Chiral CFT should be thought of as a method to construct “one half” of a full
CFT as defined by (C1)–(C6).

2.1 Ward identities


Another bilinear form on F
By (C3), the spaces of states Hb and the space of fields are isomorphic. Up to
now we have used the isomorphism ιb : F → Hb , whose definition in (C3) was
linked to the pairing ( · , · ). It is convenient to also define another linear map
ob : F → Hb as follows. We set, for P ∈ P (h,h) ,
X
ob (L−m1 · · · L−mk L−n1 · · · L−nk P ) = lim L2(h+h) [pic 22]
L→∞
X α∈I
= lim L2(h+h)
 
AR2 P (L)Lmk (b) · · · Lm1 (b)Lnk (b) · · · Ln1 (b)ϕα (0) · ũα ,
L→∞
α∈I
(2.1)
where b is a circle with bo = 0. By assumption (A1), every field in F is a
linear combination of fields of the form L−m1 · · · L−n1 · · · P , so (2.1) fixes ob on

26
all of F. In appendix A.1 it is shown that ob is well-defined and is in fact an
isomorphism.
As before, we define the notation

hφ|φ1 (p1 ) · · · φm (pm )i = [pic 23] , (2.2)

i.e. we take a disc centered at zero and large enough to contain all fields and label
the boundary by the state ob (φ). An immediate application of the definition is
the following nice property,

hL−m φ|φ1 (p1 ) · · · φm (pm )i = hφ|Lm (b)φ1 (p1 ) · · · φm (pm )i , (2.3)

where b is a circle with bo = 0 containing all fields.


(Excercise: Verify this).
We also get a new pairing on F via hφ|φ0 i = hφ|φ0 (0)i. Since ob is an isomor-
phism, this pairing is non-degenerate. By definition, the two pairings h · | · i and
( · , · ) agree if the first argument is a primary field P ,

hP |φ0 i = (P, φ0 ) , (2.4)

however for general fields they can differ.


(Excercise: Show that for P ∈ P (h,h) one has hL−1 P |L−1 P i = 2hhP |P i while
(L−1 P, L−1 P ) = −2h(2h+1)(P, P ). Hint: in the second case write out the
definition as a limit and use (C6)(iv).)
As follows from (2.3), the new pairing has the convenient property

hL−m φ|φ0 i = hφ|Lm φ0 i . (2.5)

This property is the reason to introduce the new pairing. We did not start
with this pairing because its definition is more involved than that of ( · , · ) and
because it relies on the assumptions (A1).

Ward identities from holomorphic fields

Let W ∈ F be a holomorphic field, i.e ∂W (z) = 0. The collection V ⊂ F


of holomorphic fields is called the chiral algebra of the CFT. Antiholomorphic
fields W form the antichiral algebra V ∈ F. In these notes, V will be generated
either by Vir or by H.
For a holomorphic field W ∈ V and a meromorphic function f (z) define
I
dz
W [f, p] = f (z)W (z) , (2.6)
c(p) 2πi

where c(p) is a small circle around p. For p = ∞ we take c(∞) to be a large


contour which includes all marked points on a given surface. For example

T [z m+1 , ∞] = Lm , J[z m , ∞] = Jm . (2.7)

27
Use the non-degenerate bilinear form h · | · i to define the transpose W [f, ∞]t
of W [f, ∞] via

h W [f, ∞]t φ | φ0 i = hφ| W [f, ∞]φ0 i for all φ, φ0 ∈ F . (2.8)

For example (recall (2.5))

T [z m+1 , ∞]t = L−m , J[z m , ∞]t = −J−m . (2.9)

Remark: In essence, these relations are obtained by applying the conformal trans-
formation ζ 7→ 1/ζ, which takes the point ∞ to zero. By theorem 1.15, this
transformation results in a factor (−1/z 2 )2 for T (z) and −1/z 2 for J(z), hence
the different sign, see also appendix A.1.
Remark: It is more common (see e.g. section 3.4 of [Gi], section 4.1 of [Sc], section
3.5 of [Gb] or section 6.1.1 of [DMS]) to consider the hermitian conjugation of
modes, rather than the transpose as in (2.9). For us the transpose is natural
because we have a bilinear form on F, and not an inner product.
(holomorphic Ward identities) For a meromorphic function f (z) that is ana-
lytic on C − {p1 , . . . , pm } we have
m
X
h W [f, ∞]t φ | φ1 (p1 ) · · · φm (pm ) i = h φ | W [f, pi ]φ1 (p1 ) · · · φm (pm ) i , (2.10)
i=1

as can be seen by starting from the lhs and using the contour deformation

[pic 21] . (2.11)

Remark: To get a Ward identity for correlators, take the out-state to be the
out-vacuum Ωout and choose the asymptotic behaviour of f (z) for z → ∞ such
that W [f, ∞]t Ωout = 0. Then the lhs of (2.10) is zero.
Every antiholomorphic field W (z) ∈ V gives rise to an infinite number of Ward
identites. V ⊗ V describes the symmetries of a CFT.
Remark: In a certain class of CFTs, called (quasi–) rational, the space of solu-
tions to the Ward identities is finite dimensional. This makes them solvable by
algebraic means.

2.2 Representation theory


Heisenberg algebra

Consider the
vectorspace V , with basis J −m1
· · · J −mn
v n ≥ 0, m1 ≥ m2 ≥
· · · ≥ mn > 0 . Turn this into a representation Vq = (V, π) of H by setting
(i) π(Jm )v = 0 for m > 0 and π(J0 )v = qv with q ∈ C
(ii) π(Jm )J−m1 · · · J−mn v = Jm J−m1 · · · J−mn v, where the rhs has to be brought

28
back to an ordered form using the commutation relations of H.
E.g. π(J1 )J−2 J−1 v = J−2 [J1 , J−1 ]v = J−2 v.
Remark: To give a more elegant description of Vq , look up the words ‘universal
enveloping algebra’, ‘Poincaré-Birkhoff-Witt theorem’, ‘induced representations’
and ‘Verma module’.
A representation (V, π) of H is called
- irreducible if contains no proper subrepresentation.
- heighest weight if there is a vector v (a highest weight vector) such that
π(Jm )v = 0 for m > 0 and π(J0 )v = qv for some q ∈ C, and repeated ap-
plication of π(Jm ), m ∈ Z>0 on v generates the representation.
- unitary if there is a positive definite hermitian form s on V such that s(π(Jm )x, y) =
s(x, π(J−m y) for all x, y ∈ V .

Theorem 2.1 :
(i) (V, π) is an irreducible highest weight representations of H iff it is isomorphic
to one of the Vq .
(ii) Vq is unitary iff q ∈ R.
(For a proof see YYY:Kac-book, YYY:FB lemma 5.2.2)

Characters

The character of a representation R of Vir is defined as the trace


X
χR (q) = trR q L0 −c/24 = q h−c/24 nh , (2.12)
h

where the sum is over eigenvalues h of L0 and nh is the dimension of the corre-
sponding eigenspace.
For the modules Vr of H defined above we have
weight states nh
h = r2 /2 vr 1
h = r2 /2 + 1 J−1 vr 1
h = r2 /2 + 2 J−2 vr , J−1 J−1 vr 2
h = r2 /2 + 3 J−3 vr , J−2 J−1 vr , J−1 J−1 J−1 vr 3
... ... ...
In fact, for h = p2 /2 + N , nh = p(N ), the number of partitions of N (that is,
the number of ways one can write N as a sum over positive integers). The trace
can be computed explicitly,
∞ 2
2 X q r /2
χVr (q) = trR q L0 −c/24 = q r /2−1/24
p(N )q N = , (2.13)
η(τ )
N =0

where q = e2πiτ and



Y
η(τ ) = q 1/24 (1 − q n ) (2.14)
n=1
is the Dedekind η–function.

29
Virasoro algebra

Consider the
vectorspace V , with basis L−m1 · · · L−mn v n ≥ 0, m1 ≥ m2 ≥
· · · ≥ mn > 0 . Turn this into a representation Vh,c = (V, π) of Vir by setting
(i) π(Lm )v = 0 for m > 0 and π(L0 )v = hv, π(C)v = cv with h, c ∈ C.
(ii) π(Lm )L−m1 · · · L−mn v = Lm L−m1 · · · L−mn v, as for H.
Vh,c is called a Verma module of Vir
Vh,c is a highest weight representation, but it can be reducible. E.g. the
vector L−1 v in V0,c is again a highest weight vector. This gives an embedding
V1,c → V0,c .
For h, c ∈ R there is a unique hermitian form s on Vh,c such that s(Lm x, y) =
s(x, L−m y) for all x, y ∈ Vh,c and s(v, v) = 1.

Let Nh,c = {x ∈ Vh,c s(x, y)=0 for all y∈Vh,c }. Elements of Nh,c are called
null vectors. Nh,c is a subrepresentation of Vh,c . The representations Mh,c =
Vh,c /Nh,c are then irreducible.

Theorem 2.2 :
(i) The Mh,c give all irreducible highest weight representations of Vir with c, h ∈
R.
(ii) Mc,h is equal to Vc,h if c > 1, h > 0 or c = 1 and h 6= m2 /4, m ∈ Z≥0 .
(Theorems 2.2 and 2.3 below are quite hard to prove. For a discussion of the
proof see section 4 of [Gi, Sc], section 5 of lecture 2 in [Gw2] or chapter 7 of
[DMS].)
Regarding unitarity of Mh,c note that for m ≥ 0, s(L−m v, L−m v) = s(v, Lm L−m v) =
s(v, [Lm , L−m ]v) = 2mh + c/12(m3 − m). Thus c ≥ 0 and h ≥ 0 are necessary
conditions.

Theorem 2.3 :
Mh,c is unitary iff
(i) c ≥ 1 and h ≥ 0
6
(ii) c = 1− m(m+1) for m ≥ 2 and h = hr,s (m) for 1 ≤ r ≤ m−1 and 1 ≤ s ≤ m,
((m+1)r−ms)2 −1
where hr,s (m) = 4m(m+1) .
So for c < 1 there are only finitely many unitary irreducible highest weight
1 1
representations. E.g. for m = 3 one finds c = 1/2 and h ∈ {0, 16 , 2 }.

2.3 Conformal blocks


Conformal blocks are the spaces of solutions of the Ward identites coming from
holomorphic fields.

30
Heisenberg algebra

Denote by [Vq (∞)Vq1 (p1 ) . . . Vqm (pm )] the vector space of all β : Vq ⊗ Vq1 ⊗
· · · ⊗ Vqm → C such that,2 for uk ∈ Vqk ,

β(u|u1 , . . . , um ) = β(u|u1 , . . . , L−1 ui , . . . , um ) (2.15)
∂pi
and
m
X
β(J[f, ∞]t u|u1 , . . . , um ) = β(u|u1 , . . . , J[f, pi ]ui , . . . , um ) (2.16)
i=1

P}. Here, for J[f,


for all meromorphic f analytic on C−{p1 , . . . , pm
m
p] the Laurent
expansion around p is Pimplied, i.e. if f (z) = m am (z − p) then J[f, p] is
defined to be J[f, p] = m am Jm . E.g. for f = 1 have J[f, p] = J0 and (2.16)
reads
m
X
β((−J0 )u|u1 , . . . , J0 ui , . . . , um ) = β(u|u1 , . . . , J0 ui , . . . , um )
i=1
(2.17)
= (q1 + · · · + qm )β(u|u1 , . . . , um )
which implies charge conservation q + q1 + · · · + qm = 0.

Theorem 2.4 :
(i) dim[Vq (∞)Vq1 (p1 ) . . . Vqm (pm )] = 1 if q +q1 +· · ·+qm = 0 and zero otherwise.
(ii) For any β ∈ [Vq (∞)Vq1 (p1 ) . . . Vqm (pm )] we have
m
Y
β(vq |vq1 , . . . , vqm ) = λ (pi − pj )qi qj , λ∈C . (2.18)
i,j=1,i<j

Proof: We will not prove (i), details can be found in YYY.


(ii) Note that L−1 vqi = qi J−1 vqi .
(Excercise: Show this using the expression for L−1 in modes of J implied by
lemma 1.10)
Take f (z) = (z − pk )−1 . Then for i 6= k we have J[f, pi ] = (pi − pk )−1 J0 +
(pos. modes), J[f, ∞]t = −J1 and for i = k, J[f, pk ] = J−1 . The Ward identity
reads
X
0 = β(vq |vq1 , . . . , J−1 vqk , . . . , vqm ) + (pi − pk )−1 β(vq |vq1 , . . . , J0 vqi , . . . , vqm )
i6=k
1 ∂ X qi 
= + β(vq |vq1 , . . . , vqm ) .
qk ∂pk pi − pk
i6=k
(2.19)
2 More correctly, one should think of a trivial bundle over Cm
(minus the diagonals) with
fiber given by the dual of Vq ⊗ Vq1 ⊗ · · · ⊗ Vqm . Then (2.15) defines a connection on this
bundle, called the Knizhnik-Zamolodchikov connection.

31
(Exercise: Show that the conformal block of the theorem solves these differen-
tial equations.) The space of solutions to the set of linear partial differential
equations (2.19) can be at most one-dimensional. To see this fix the value of
the solution at one point p = (p1 , . . . , pm ) ∈ Cm . For each other point x, fix a
path γ in Cm from p to x. The PDEs (2.19) then fix the behaviour of β along
γ, and give β(x) in terms of γ and the precribed value at p. The fact that a
solution exists shows that this way of finding β(x) does not depend on γ (at
least not locally; there are monodromies, which correspond to the branch-cuts
of the solution (2.18)). 

Remark: The above proof illustrates an important point: conformal blocks (of
certain chiral algebras, in particular of rational chiral algebras) can be found by
solving differential equations. the differential equation used above is a special
case of the Knizhnik-Zamolodchikov equations for WZW-models, see section 24
of [Fu] or section 15.3.2 of [DMS].

Virasoro algebra
For Vir the corresponding space [Mh1 ,c (p1 ) . . . Mhm ,c (pm )] is generically infinite
dimensional.

Theorem 2.5 :
If c < 1 and h1 , . . . , hm such that Mc,hi is unitary, then dim[Mh1 ,c (p1 ) . . . Mhm ,c (pm )] <
∞.

Chiral vertex operators for the free boson

The conformal blocks for the Heisenberg algebra can also be obtained by using
chiral vertex operators. Define an operator

Np : Vq −→ Vp+q (2.20)

via Np (J−m1 · · · J−mk vq ) = J−m1 · · · J−mk vp+q .


(Excercise: Show that J0 Np = Np (J0 + p) and, for m 6= 0, Jm Np = Np Jm .)
Further, we set
X 1
φ(z) = i Jm z −m . (2.21)
m
m6=0

Remark: With this definition we have J(z) = −i∂φ(z) when expanding both
sides in modes Jm . The formal sum φ(z) is however not a field in the CFT on
C. In fact, since J(z) has weight one, the hypothetical field φ would lie in F (0) ,
but in the free boson CFT at grade zero we only find the vacuum F (0) = C1.
Using Nq and φ(z) we define an operator from Vp (z) : Vq → Vp+q to be the
normal ordered exponential

Vp (z) =: eipφ(z) : Np . (2.22)

32
The Vp (z) are called chiral vertex operators. (Because of the infinite sums,
we should really work with approprite completions of Vq and Vp+q , but let us
not worry about this point.) Using the commutation relations for the Jm one
obtains
[Jm , Vp (z)] = pz m Vp (z) . (2.23)
(Excercise: Verify this relation, being careful to distinguish the cases m 6= 0
and m = 0.)
These are just the commutation relations for a U (1)-primary field of charge p.
Consider the product

f = (vq , Vq1 (p1 ) · · · Vqm (pm )vqm ) . (2.24)

Using the Baker-Campbell-Hausdorff formula, one can verify that (see e.g. sec-
tion 9.1.1 of [DMS])
m
Y
f= (pi − pj )qi qj . (2.25)
i,j=1,i<j

By taking multiple products of the chiral vertex operators we have produced an


n-point conformal block.
Remark: The notion of a chiral vertex operator can be generalised to arbitrary
chiral algebras. It is then defined in terms of three-point conformal blocks.
Higher-point conformal blocks can again be obtained by taking products of
chiral vertex operators. Chiral vertex operators are thus a way to obtain higher-
point conformal blocks from three-point conformal blocks. A mathematical
formalisation of chiral algebras and chiral vertex operators is the notion of vertex
algebras, see e.g. [SFW, section 3.2] for a definition or YYY for more details.

2.4 Holomorphic factorisation


Let A stand for either Vir or H. There is an action of A ⊗ A on F. Write
M
F= Zlr Vl ⊗ Vr , Zlr ∈ Z≥0 , (2.26)
l,r

where the Vl , Vr are representations of A. Let φ ∈ Vl ⊗ Vr , φi ∈ Vli ⊗ Vri and


consider the amplitude

C(φ|φ1 , . . . , φm ) = (φ, φ1 (p1 ) · · · φm (pm )) . (2.27)

We have

C ∈ [Vl (∞)Vl1 (p1 ) · · · Vlm (pm )] ⊗ [Vr (∞)Vr1 (p∗1 ) · · · Vrm (p∗m )] (2.28)

Writing a correlator as a sum of holomorphic and anti-holomorphic blocks is


called holomorphic factorisation.

33
2.5 Crossing symmetry
Crossing symmetry is a constraint on the three-point functions coming from the
four point function. We will illustrate this in the case of the free boson.
Let Λ be a 1– or 2-dimensional lattice in R2 , i.e.

Λ = { mk | m ∈ Z } or Λ = { mk1 + nk2 | m, n ∈ Z } , (2.29)

for vectors k, k1 , k2 ∈ R2 . For elements of Λ we use the notation ~q = (q, q),


where (q, q) are the coordinates in R2 . For the space of fields we take
M
F= V q ⊗ Vq . (2.30)
~∈Λ
q

Denote by φq~ ∈ F the field given by the highest weight vector in Vq ⊗ Vq . In


order to fix the CFT we have to fix the numbers3

Cq~1 ,~q2 = (φq~3 , φq~1 (1)φq~2 (0)) (2.31)

in a consistent way (by charge conservation we have ~q3 = ~q1 + ~q2 , hence it need
not be included in the notation for C).
For the three-point function, we can take the solution in lemma 2.4 to write

(φq~3 , φq~1 (z)φq~2 (w)) = Cq~1 ,~q2 (z − w)q2 q3 (z ∗ − w∗ )q2 q3 , (2.32)

where the coefficient is fixed by setting z=1, w=0 and comparing to (2.31). In
order for (2.32) to be a function of z (rather than a multivalued function) we
need
~q2 · ~q3 = q2 q3 − q 2 q 3 ∈ Z . (2.33)
Thus Λ has to be an integer, Lorentzian lattice. Further, if we evaluate (2.32)
for w=1 and z=0 we obtain the relation

Cq~2 ,~q3 = (−1)q~2 ·~q3 Cq~3 ,~q2 . (2.34)

If we set ~q2 = ~q3 in this relation, we see that Cq~2 ,~q2 can be non-zero only if
~q2 · ~q2 ∈ 2Z. But Cq~2 ,~q2 has to be nonzero (otherwise the OPE φq~2 φq~2 would
be identically zero, and this is inconsistent), so that Λ is an even lattice (i.e.
~q · ~q ∈ 2Z for all ~q ∈ Λ). As one consequence, it is consistent to normalise the
fields φq~ such that
C−~q,~q = 1 . (2.35)
3 Actually, theorem 1.18 says we should give these numbers for all primary fields. But for
larger chiral algebras, the relevant set of fields are the ones primary w.r.t. the whole chiral
algebra. In the present case we need to consider H-primary fields (i.e. all φ∈F s.t. Jm φ = 0
for all m>0), rather than Vir-primary fields. Using expression (1.72), one can verify that an
H-primary field is also Vir-primary (the converse does not hold in general).

34
From the Ward-identities we know that the four-point function has to be of
the form

f (z) = φq~4 , φq~1 (1)φq~2 (z)φq~3 (0) = λz q2 q3 (1−z)q1 q2 (z ∗ )q2 q3 (1−z ∗ )q1 q2 ,


(2.36)
for some constant λ ∈ C. One way to determine the constant is to use the OPE
for z close to zero,
X
f (z) = [pic 37] = (φq~4 , φq~1 (1)ϕα (0)) · (ϕ̃α , φq~2 (1)φq~3 (0))
α∈I (2.37)
1)
= Cq~1 ,~q2 +~q3 Cq~2 ,~q3 z q2 q3 (z ∗ )q2 q3 (1 + O(z)) ,

where in 1) we used that the leading contribution for z → 0 comes from the high-
est weight field φq~2 +~q3 , and all other terms in the sum over I give contributions
of O(z). Comparing to the exact answer (2.36) we conclude

λ = Cq~1 , q~2 +~q3 Cq~2 , q~3 . (2.38)

Another way to write the four-point function as a sum over intermediate states
is to take the OPE for z close to 1,
X
f (z) = [pic 38] = (φq~4 , ϕα (z)φq~3 (0)) · (ϕ̃α , φq~1 (1−z)φq~2 (0)) . (2.39)
α∈I

By the same reasoning as above, this leads to

λ = Cq~1 +~q2 , q~3 Cq~1 , q~2 . (2.40)

The equality of (2.37) and (2.39) is called crossing symmetry. In the present
case it leads to
Cq~1 , q~2 +~q3 Cq~2 , q~3 = Cq~1 +~q2 , q~3 Cq~1 , q~2 , (2.41)
but for other CFTs the relations can be much more involved.
A simple solution to (2.41) would be Cp~,~q ≡ 1. However this might be incon-
sistent with (2.34).

• 1-dim. lattice: Since Λ is even, we have k · k ∈ 2Z. This implies that


p~ · ~q ∈ 2Z for all p~, ~q ∈ Λ. Hence it is consistent to take Cp~,~q ≡ 1.
• 2-dim. lattice: Any p~, ~q ∈ Λ can be written as

p~ = ak1 + bk2 , ~q = ck1 + dk2 , (2.42)

for integers a, b, c, d. We set

Cp~,~q = (−1)a d k1 ·k2 . (2.43)

(Excercise: Show that this solves (2.41) and is consistent with (2.34).)

35
3 CFT on the upper half plane
3.1 Axioms
The upper half plane is the subset H = {(x, y)|y ≥ 0} of R2 . An open string
world sheet X in H is a closed subset of H with a finite collection of marked
points, e.g.
[pic 25] . (3.1)
Denote the set of boundary conditions by B. A marked semi-circle s is the
part of a circle with center on R that lies in H and whose ends are labelled by
elements of B. Denote the two labels by [s and s],

[pic 26] . (3.2)

There are three kinds of boundaries


• closed state boundaries:
∂ c X = { components of ∂X which are full circles b}
• open state boundaries:
∂ o X = { components of ∂X which are semi circles s}
• physical boundaries (these are openintervals on the real axis):
∂ p X = X − (∂ o X ∪ {marked pts}) ∩ R
A world sheet in H in addition obeys ∂X = ∂ c X ∪ ∂ o X ∪ ∂ p X. Further, each
component of ∂ p X is labelled by an element of B. This induces the labelling of
the semi-circles.

Data:

the data of a CFT on C.


L (∆)
to each pair a, b ∈ B a space of boundary fields Fab = ∆∈Sab Fab , an Sab -
(∆)
graded vector space, with Sab a discrete subset of R and 0 < dim Fab < ∞.
to each marked semi-circle s, a space of states Hs .
to each open string world sheet in H an amplitude
O O O
AX : F ⊗#pts ⊗ F[q] ⊗ Hb ⊗ Hs −→ C (3.3)
q⊂{bndpts} b⊂∂ c X s⊂∂ o X

where F[q] is an abbreviation of F[q,q] and [q, q] stands for the boundary labels
on either side of the marked boundary point q.

36
Properties of A:

If X is a closed string world sheet in R2 , (C1)–(C6) hold. (Note that an open


string world sheet with empty physical boundary ∂ p X = ∅ is also a closed string
world sheet. The amplitude for such world sheets has to be the one given by
the CFT on C.)

(O1) Scaling: For any collection of bulk points pi ∈ X and fields φi ∈ F (∆i ) ,
(∆0 )
boundary points qj ∈ X ∩ R and fields ψj ∈ F[qj ]i , and λ ∈ R>0 we have
 
AH φ1 (p1 ) . . . φm (pm )ψ1 (q1 ) . . . ψn (qn ) =
0 0
(3.4)
λ∆1 +···+∆m +∆1 +···+∆n AH φ1 (λp1 ) . . . φm (λpm )ψ1 (λq1 ) . . . ψn (λqn ) .
 

(O2) Non-degenerate boundary two-point function: Define a pairing on Fba ×Fab


(∆) (∆0 )
by setting, for any ψ ∈ Fba and ψ 0 ∈ Fab ,
0
(ψ, ψ 0 ) = lim L2 min(∆,∆ ) [pic 27] (3.5)
L→∞

This pairing has to be non-degenerate.


For a semi-circle s, let Cs be the semi-disc bounded by s and the diameter,
so be the centre of s and rs the radius of s.
(O3) State-field correspondence: For each marked semi-circle s with a = [s and
b = s], there is a linear isomorphism ιs : Fba → Hs s.t. for any ψ ∈ Fba , ψ 0 ∈ Fab ,

ACs ψ 0 (so ) (ιs (ψ)) = [pic 28] = (ψ, ψ 0 ) .


 
(3.6)

(∆ )
Choose a basis ξα , α ∈ Iab of Fab respecting the grading, i.e. for α ∈ Iab α
(∆ )
have ξα ∈ Fab α . Let ξ˜α ∈ Fba be the dual basis s.t. (ξ˜α , ξβ ) = δα,β . Denote
vα = ιs (ξα ) and ṽα = ιs (ξ˜α ), then in particular

[pic 29] = δα,β . (3.7)


Given a world sheet X, a semi-circle s is allowed in X if s ⊂ (X−{marked pts})∪
∂ p X, i.e. it does not intersect any marked point or state boundary of X. For an
allowed semi-circle s in X, we again get two new world sheets Xi (s) (“inside”)
and Xo (s) (“outside”) by cutting X along s.
(O4) Sum over intermediate states: Given a world sheet X and an allowed
semi-circle s in X, we have
X
AX = AXo (b) (. . . , vα ) AXi (b) (. . . , ṽα ) . (3.8)
α∈I[s]

37
(O5) Conformal boundary
 condition:
(i) Inside any AX . . . we have, for any x ∈ ∂ p X,
T (x) = T (x) . (3.9)
(ii) For any collection of bulk/boundary fields on H we have
lim r4 AH T (reiθ ) . . . < ∞ ,
 
(3.10)
r→∞

and the same for T .


(iii) For all ψ ∈ Fab and a small semi-circle s around x ∈ R,
Z Z
dz dz d
T (z)ψ(x) + T (z)ψ(x) = ψ(x) . (3.11)
s 2πi s 2πi dx

For (O5)(i) to be consistent with the T T -OPE (C6)(iii) we need c = c̃. The
a necessary condition for a CFT on the plane to be part of a CFT on the UHP
is that the left and right central charge are equal.
The properties imply translation invariance of the correlators and the existence
of identity fields 1 ∈ Faa .

3.2 Conformal boundary conditions


The condition T (x) = T (x)
To motivate the condition (O5)(i) consider a theory with unique vacuum F (0) =
C1 and spectrum bounded below by zero, as should be for continuum limit of
lattice models.
Take a correlator on H with only bulk fields. we would like this correlator to
be translation invariant and obey scaling (O1). Recall the formulation of local
conformal invariance in lemma 1.14,
[pic 31] , (3.12)
for r ∈ R and λ > 0. To have the desired  property, both amplitudes on the rhs
must be equal to AH φ1 (p1 ) · · · φm (pm ) . To first order in r and λ this implies,
for k = 0, −1,
[pic 30] , (3.13)
where we deformed the integration contour to run along the real axis. The inte-
gral along the half circle to close the contour vanishes due to (O5)(ii). Writing
out the definition of the mode integrals gives
Z ∞
dx k+1 
x T (x) − T (x) = 0 . (3.14)
−∞ 2πi
d
For k = −1 this holds if T (x) − T (x) = dx ψ(x) for some boundary field ψ.
Inserting this into k = 0 and integrating by parts one sees that both conditions
d 2
hold for T (x) − T (x) = ( dx ) θ(x) for some boundary field θ(x). Since T and T
have conformal weight two, θ would have weight zero and hence be the identity
field. But derivatives of the identity vanish, so that T (x) − T (x) = 0.

38
Action of Vir on boundary fields

For any collection of fields on H consider the two functions

and g(z) = AH T (z ∗ ) · · · .
   
f (z) = AH T (z) · · · (3.15)

The function f is defined for =(z) ≥ 0 and g for =(z) ≤ 0, and for z (resp. z ∗ )
not equal to one of the marked points. Further ∂f (z) = 0 = ∂g(z), since g(z) is
a composition of two anti-holomorphic functions, and hence holomorphic. By
(O5)(i) we have f (x) = g(x) for real x, and thus g(z) is the analytic continuation
of f (z) to the lower half plane (and vice versa).
When we write T (z) with =(z) < 0 it is understood to mean T (z ∗ ). Then
(O5)(iii) has the easier form
I
dz d
T (z)ψ(x) = ψ(x) , (3.16)
c(x) 2πi dx

for a small circle c(x) around x.


Define an action of Vir on Fab via
I
dz
(Lm ψ)(x) = (z − x)m+1 T (z)ψ(x) . (3.17)
c(x) 2πi

There is only one copy of Vir acting on boundary fields, as the definition of Lm ψ
already uses both, T (z) and T (z). In other words, defining (Lm ψ)(x) as above,
but with T instead of T , would just yield the same expression as (Lm ψ)(x).
As for the CFT on the plane, we make some simplifying assumptions on the
action of Vir on boundary fields.
(A2) We assume the following properties for the action of Vir on Fab :
(i) L0 can be diagonalised on F.
(ii) Every boundary field is either primary or a descendent of a primary field.
(h)
A boundary field ψ ∈ Fab is primary if Lm ψ = 0 for all m > 0. As for bulk
(h) (h)
fields, for boundary fields we write Fab = ⊕Fab so that for ψ ∈ Fab we have
L0 ψ = hψ. In fact, for boundary fields we have h = ∆.

Lemma 3.1 :
(h)
For a primary boundary field ψ ∈ Fab and a primary bulk field φ ∈ F (h,h) we
have
[Lm , ψ(x)] = xm (h(m+1) + x dx
d
)ψ(x) ,
  (3.18)
[Lm , φ(z)] = z m (h(m+1) + z∂) + z m (h(m+1) + z∂) φ(z) .

Proof: The commutator for boundary fields is obtained by a computation anal-


ogous to the one in lemma 1.13. For the bulk field φ(z) note that an amplitude

39
involving T (ζ) and φ(z) may have poles at ζ = z and ζ = z ∗ (as T (ζ) = T (ζ ∗ )
which may have a pole at ζ ∗ = z). Thus
[Lm , φ(z)] = [pic 32] (3.19)


Note that the action of Lm only generates conformal transformations that


leave the real axis invariant (as they should to be symmetries of a theory on the
UHP), e.g.
d
[L−1 , φ(z)] = (∂ + ∂)φ(z) = φ(x + iy) (3.20)
dx
is a shift along the real axis.

Möbius transformations of the UHP

All conformal transformations of the UHP (with infinity) to itself are of the
form
aζ + b
ζ 7→ with a, b, c, d ∈ R . (3.21)
cζ + d
Note that by (O5)(ii) we have, for k = 0, ±1
 
AH Lk φ1 (p1 ) · · · ψ1 (x1 ) · · · = 0 . (3.22)
In the same manner as before, we can show that the correlators on the UHP
transform covariantly under a Möbius transformation f (z) of the UHP,
   
AH φ1 (p1 ) · · · ψ1 (x1 ) · · · = AH φ̃1 (f (p1 )) · · · ψ̃1 (f (x1 )) · · · , (3.23)
where all fields are primary and

φ̃1 = (f 0 (p1 ))hφ1 (f 0 (p1 )∗ )hφ1 φ1 , ψ̃1 = (f 0 (x1 ))hψ1 ψ1 , (3.24)


etc.
This fixes the coordinate dependence of the following correlators of primary
fields:
• One, two and three boundary fields, e.g.
 
AH ψ1 (x1 )ψ2 (x2 )ψ3 (x3 )
(3.25)
= (const) x12h3 −h1 −h2 x13h2 −h1 −h3 x23h1 −h2 −h3 .

• One bulk field and no or one boundary field, e.g.


 
AH φ(z)ψ(x) = (const)
(3.26)
×(z − z ∗ )hψ −hφ −hφ (z − x)hφ −hφ −hψ (z ∗ − x)hφ −hφ −hψ .

The correlator of two bulk fields and no boundary fields on the UHP is already
a complicated object.

40
3.3 Two more OPEs
For the CFT on C we defined the OPE of two bulk fields. For a CFT on H
there are two additional OPEs.

Lemma 3.2 :
One can replace boundary states by boundary fields via

[pic 33] (3.27)

Proof: Use (O4) and (3.7). 

Introduce the notation

(ψ, φ1 (p1 ) · · · ψ1 (x1 ) · · · ) = [pic 34] (3.28)

where s is a semi-circle with so = 0.


(Excercise: Show that this definition is independent of the choice of radius rs .)
If there exists a semi circle s such that Cs only contains a bulk field φ(z), we
can expand this bulk field in terms of boundary fields,
X
φ(z) = [pic 35] = (ξ˜α , φ(z − x))ξα (x) , (3.29)
α∈Iaa

where x = so . This is the bulk-boundary OPE.


If there exists a semi circle s such that Cs only contains two boundary fields
ψ(x) and ψ 0 (y) where so = y, we can expand this in terms of boundary fields,
X
ψ(x)ψ 0 (y) = [pic 36] = (ξ˜α , ψ(x − y)ψ 0 (0)) ξα (y) , (3.30)
α∈Iac

This is the boundary OPE.


A CFT on the UHP is uniquely fixed by giving the space F as Vir×Vir-module,
the set of boundary conditions B, the spaces Fab as Vir-modules and the three
sets of numbers

(φ1 , φ2 (1)φ3 (0)) , (ψ1 , φ1 (i/2)) , (ψ1 , ψ2 (1)ψ3 (0)) , (3.31)

for all primary fields ψ1,2,3 and φ1,2,3 . This can be seen by taking repeated
OPEs. (The bulk insertion point i/2 in the second amplitude is just a convenient
choice and has no deeper meaning.)

41
3.4 Ward identities on the UHP
The commutator of Lm with bulk and boundary fields generalises to the Ward
identities. Let (fields) stand for φ1 (p1 ) · · · φm (pm )ψ1 (x1 ) · · · ψn (xn ), then
m
X n
X
(ψ, T [f, ∞](fields)) = (ψ, (T [f, pi ]+T [f, p∗i )(fields))+ (ψ, (T [f, xj ](fields)) .
i=1 j=1
(3.32)
Suppose the fields transform in representations φi ∈ Vli ⊗ Vri and ψj ∈ Vj of
Vir × Vir, respectively Vir. Then these are the Ward identities solved by the
conformal blocks in

[V (∞)Vl1 (p1 )Vr1 (p∗1 ) · · · V1 (x1 ) · · · ] . (3.33)

We learned: Correlators on the UHP are special elements in a space of con-


formal blocks on C. The chiral CFT is thus a building block for both, CFT on
the plane and CFT on the UHP.

3.5 Example: Free boson on the UHP


In section 2.5 we used different ways of writing a bulk correlator as a sum
over intermediate states to find constraints for the OPE-coefficients Cq~1 ,~q2 =
(φq~1 +~q2 , φq~1 (1)φq~2 (0)), and we gave a solution to these constraints. The con-
stants Cq~1 ,~q2 , together with knowing F as an H × H-modules, fixed a CFT on
C. To do the same for the UHP, we should in addition give a set of boundary
conditions, the spaces Fab as H-modules, as well as the remaining two sets of
constants in (3.31). As an example, we will investigate one special case, namely
the OPE-coefficient (1a , φq~ (i/2)), where 1a ∈ Faa is the identity field on the
boundary labelled by a.
For a conformal boundary condition we required T (x) = T (x) for x ∈ R. For
the free boson it is natural to demand that also

J(x) = ωJ(x) for ω ∈ {±1} . (3.34)

In fact, for both choices of ω (3.34) already implies T (x) = T (x). To see this,
express T and T as the limits

T (z) = lim J(ζ)J(z) − (ζ−z)−2 , T (z) = lim J(ζ)J(z) − (ζ ∗ −z ∗ )−2 (3.35)


ζ→z ζ→z

(Excercise: Show that this expression for T (z) is equivalent to the one given in
(1.71))
If we set z = x ∈ R in the limits (3.35), equation (3.34) immediatly gives
T (x) = T (x).
Remark: This shows that every boundary condition with the property (3.34)
is also conformal. However the converse is not true, there many conformal
boundary conditions that do not obey (3.34).

42
Remark: In terms of a classical scalar field φ : R2 → R, the expression corre-
sponding to the U (1)-currents J and J in the quantum theory are the deriva-
tives −i∂φ(x, y) and i∂φ(x, y) (this is one convention; if J corresponds instead
to −i∂φ(x, y), Neumann and Dirichlet get exchanged in the table below). Sub-
stituting (1.30) to get derivatives w.r.t. x and y we find

bnd. cond. in CFT classical analogon



J(x) = J(x) dφ(x, y)/dx y=0 = 0 (Dirichlet)

J(x) = −J(x) dφ(x, y)/dy y=0 = 0 (Neumann)

Similar as for T and T , condition (3.34) implies that the analytic continuation
of (an amplitude with insertion of) J(z) to Im(z) < 0 is equal to ωJ(z ∗ ). The
additional factor of ω appears also in the Ward identities and on finds that an
amplitude of bulk fields on the UHP is an element in the following space of
conformal blocks,

AH φq~1 (z1 ) · · · ∈ [V0 (∞)|Vq1 (z1 )Vωq1 (z1∗ ) · · · ] .


 
(3.36)

(0)
We will consider a boundary condition a with the property Faa = C1, i.e.
linear multiples the identity field are the only fields of weight zero. In our choice
of basis ξα , there is a unique label 0 ∈ I (0) and we may take ξ0 = 1. By definition
 −1
then ξ˜0 = AH 1a · 1a , and we will write 1̃a = ξ˜0 Define the constants
a
Bq~ = (1̃a , φq~ (i/2)) . (3.37)

Below we will derive constraints for a Bq~ . The parameter a will then be used to
label different solutions to these constraints, i.e. different boundary conditions.
Note that the amplitude (1̃a , φq~ (z)) is an element in the space of blocks
[V0 (∞)|Vq (z)Vωq̄ (z ∗ )]. So it is nonvanishing only if q = −ω q̄. If this condi-
tion is fulfilled, we have
 z−z ∗ −q2
(1̃a , φq~ (z)) = aBq~ , (3.38)
i
where the constant was fixed by setting z = i/2 and comparing to (3.37)
Consider the two-point function on the UHP of two bulk fields φp~ (z) and
φq~ (w) with p = −ω p̄ and q = −ω q̄ (since 0 ∈ Λ, there is at least one field with
this property). From the Ward identities we know

AH φp~ (z)φq~ (w) ∈ [V0 (∞)|Vp (z)Vωp̄ (z ∗ )Vq (w)Vωq̄ (w∗ )] ,


 
(3.39)

i.e., since the space of blocks is one-dimensional and using p = −ω p̄, q = −ω q̄,

   z−z ∗ −p2  w−w∗ −q2  (z−w)(z ∗ −w∗ ) pq


AH φp~ (z)φq~ (w) = λ , (3.40)
i i (z−w∗ )(z ∗ −w)

43
for some constant λ. The factors of i have been included for convenience. We
will again compare two ways to fix the constant λ.
(i) We take the leading contribution of the bulk-boundary OPE of both bulk
fields. Set z = x+iy and w = u+iv. Then
X
AH ξα (x)ξβ (y) (ξ˜α , φp~ (iy)) (ξ˜β , φq~ (iv))
   
AH φp~ (z)φq~ (w) =
α,β∈Iaa (3.41)
2 2
= AH 1a aBp~ aBq~ (2y)−p (2u)−q (1 + O(y))(1 + O(u))
 

Comparing with (3.40) we see

λ = AH 1a aBp~ aBq~ .
 
(3.42)

(ii) We take the leading contribution in the bulk OPE,


  X  
AH φp~ (z)φq~ (w) = AH ϕα (w) (ϕ̃α , φp~ (z − w)φq~ (0))
  α∈I 
= AH φp~+~q (w) φ−~p−~q , φp~ (z−w)φq~ (0) (1 + O(z−w)) (3.43)
2 pq
= AH 1a aBp~+~q (2v)−(p+q) Cp~,~q (z−w)(z ∗ −w∗ )
 
(1 + O(z−w))

Comparing to the corresponding limit of (3.40) shows

λ = AH 1a aBp~+~q Cp~,~q .
 
(3.44)

Combining (i) and (ii) we thus arrive at the constraint


a
Bp~ aBq~ = aBp~+~q Cp~,~q , (3.45)

which has to hold for all p~, ~q ∈ Λ such that p = −ω p̄ and q = −ω q̄.
Remark: Of course, (3.45) is again just a necessary condition for the constants
(3.31). It is however sufficient to verify that different sums over intermediate
states agree for a certain finite set of basic world sheets, see YYY:Sonoda,
Lewellen for details.

4 CFT on surfaces of higher genus


In this section we will always assume c = c̃, even if this condition is only
necessary once we deal with surfaces that have a boundary, in sections 4.3 and
4.4.

4.1 Closed surfaces

Let (X, g) be an oriented 2-dimensional Riemannian manifold with metric g.


A smooth map f : (X, g) → (Y, h) is
- an isometry if it is bijective and f ∗ h = g.

44
- conformal if f ∗ h(x) = Ω(x)g(x) for some scalar function Ω : X → R≥0 .
- analytic if it is orientation preserving and conformal.
Let  
Dε = z ∈ C |z| < ε , Aε = z ∈ C 1−ε ≤ z ≤ 1+ε , (4.1)
as well as

A+ 0
  
ε = z∈Aε |z|≥1 , Aε = z∈Aε |z|=1 , Aε = z∈Aε |z|≤1 . (4.2)

On these surfaces we take the metric induced by R2 , i.e. g(x)ij = δij .


Suppose we are given a CFT on C. A field insertion is a pair (φ, f ), where
φ ∈ F and f : Dε → X is an injective analytic map,

[pic 39] . (4.3)

We say that the field φ is inserted at p = f (0) and f are local coordinates around
p.
A (closed string) in-state is a pair (φ, f )in where φ ∈ F and f : A+
ε → X is
analytic, injective, and f (A0ε ) ⊂ ∂X,

[pic 40] . (4.4)

Similarly, a (closed string) out-state is a pair (φ, f )out where φ ∈ F and f :


A− 0
ε → X is analytic, injective, and f (Aε ) ⊂ ∂X,

A closed string world sheet X = (X, g) is an oriented compact 2-dimensional


Riemannian manifold with metric g. It can have field insertions, and every
component of ∂X is an in-state or an out-state.
Two world sheets X, Y are equivalent X ∼ Y if there is an isometry F : X →
Y compatible with field insertions and states, i.e. for each field insertion (f, φ)
on X there is an insertion (f˜, φ) on Y , and for each state (f, ψ)in/out on X there
is a state (f˜, ψ)in/out on Y such that the following diagrams commute,

[pic 43] . (4.5)

Data

a CFT on C
for every world sheet X an amplitude A(X) ∈ C. (We will also write A(X, g)
to make the dependence on the metric explicit.)

45
Properties
(CS1) If X is a world sheet on R2 (with metric g(x)ij = δij ) and all maps f
(in field insertions
 and states) are of the form f (ζ) = ζ + (const), then A(X) =
AX fields (states).
(CS2) Diffeomorphism invariance: If the world sheets X and Y are equivalent,
then A(X) = A(Y ).
(CS3) Weyl transformations:
A(X, eσ g) = eS[σ] A(X, g) , (4.6)
where S[σ] is the Liouville action, integrated over X. In a coordinate patch
U ⊂ R2 where gij = eσ0 δij we can write
Z
c 1 
S[σ] = dx∧dy ∂σ∂σ − ∂σ0 ∂σ0 . (4.7)
12 2π U

Remark: The correlator for a world sheet (X, g) without boundary and with field
insertions (φi , fi ) is defined to be


φ1 (f1 ) · · · φm (fm ) g = A(X, g)/A(X̃, g) , (4.8)

where X̃ is the same world sheet as X, but without field insertions. Note that



φ1 (f1 ) · · · φm (fm ) eσ g = φ1 (f1 ) · · · φm (fm ) g , (4.9)

so that the correlators are invariant under Weyl-transformations of the metric,


while the amplitudes are not. This is the Weyl anomaly, and from (4.7) we see
that it is proportional to c. For string theory we will need the central charge to
be zero, so that even the amplitudes are invariant under Weyl-transformations.
For each analytic function η : Dε → Dε0 s.t. η(0) = 0 and η 0 (0) 6= 0 (i.e. η is
injective if we choose ε small enough) we define an operator Uη : F → F. For
η(ζ) = fn,u (ζ) from (1.85) and n ≥ 0 have Uη = exp(uLn + u∗ Ln ). For more
general η consult e.g. [FB, section 6.3.1].
(CS4) Change of local coordinates: A(X) remains invariant if a field insertion
(φ, f ) is replaced by (Uη φ, f ◦ η), where η : Dε0 → Dε is analytic injective and
η(0) = 0.

Example: Two-point correlator on the plane


As an example for (CS1)–(CS4), consider the world sheets X, X̃ given by
[pic 41] (4.10)
with g(x)ij = δij and g̃(x)ij = λ−1 gij , f1 (ζ) = ζ, f2 (ζ) = ζ + p, f˜1 (ζ) = λζ,
f˜2 (ζ) = λ(ζ + p) and F (z) = λz. Note that F provides an equivalence between
X and X̃. Thus by (CS2),
A(X, g) = A(X̃, g̃) . (4.11)

46
Since g and g̃ only differ by a Weyl-transformation, by (CS3) we can replace g̃
by g without affecting the correlators,
(CS2) (CS3)
φ1 (f˜1 )φ2 (f˜2 ) φ1 (f˜1 )φ2 (f˜2 )




φ1 (f1 )φ2 (f2 ) g
= g̃
= g
(4.12)

Note that the lhs is a 2pt-correlator with insertions at 0 and p, while the rhs is a
2pt-correlator with insertions at 0 and λp. From (C1) we might have expected

φ1 (0)φ2 (p) = λ∆1 +∆2 φ1 (0)φ2 (λp) ,





(4.13)

where we supposed that φ1 ∈ F (∆1 ) , φ2 ∈ F (∆2 ) . The difference between (4.12)


and (4.13) comes from the local coordinates f˜1 , f˜2 on the rhs of (4.12). In fact
we can write f˜1 = d1 ◦ η with d1 (ζ) = ζ and η(ζ) = λζ as well as f˜1 = d2 ◦ η
with d2 (ζ) = ζ + λp. Then by (CS4),
(CS4)
φ1 (f˜1 )φ2 (f˜2 )




g
= φ1 (d1 ◦ η)φ2 (d2 ◦ η) g g
, = Uη φ1 (d1 )Uη φ2 (d2 )
(4.14)
where Uη φk = λL0 +L0 φk = λ∆k φk . To the rhs of (4.14) we are now allowed to
apply (CS1) since all local coordinates are of the form ζ 7→ ζ + (const). We find
precisely (4.13).

Sum over intermediate states

For f : Aε → X analytic, injective define a new world sheet Xf (φi , φo ) by


cutting X along f (S 1 ) and adding states (φi , fi )in , (φo , fo )out as in

[pic 42] . (4.15)

(CS5) Sum over intermediate states:


X
A(X) = A(Xf (ϕα , ϕ̃α )) . (4.16)
α∈I

Lemma 4.1 :
Let I(ζ) = 1/ζ. There is a φ̃ ∈ F such that replacing an out-state (φ, f )out by
the in-state (φ̃, f ◦ I)in on a world sheet X leaves A(X) invariant.
The proof is in appendix XXX:A.3.
A trinion is an S 2 with three holes. Consider first a world sheet without
field insertions. By repeated application of (CS5) and lemma 4.1 it is possible
to reduce any amplitude to a sum over trinions with three in-states. In each
trinion, the in-states can be replaced by field insertions by glueing discs to
the holes. There is an analytic map from (S 2 , g) to the complex plane. By
(CS1)–(CS4), the three-point correlator on (S 2 , g) is determined by a three-
point correlator C. Thus, a CFT on closed Riemann surfaces does not contain
more information than a CFT on C. Instead, the requirement that a CFT on

47
C gives rise to a CFT on closed surfaces results in constraints on the CFT on
C. One such constraint is modular invariance.
This argument also holds for world sheets X with field insertions. In this case
one cuts X into components where the number of field insertions and the number
of state boundaries sum to three (or less).

4.2 Modular invariance for the torus


Hamiltionian on a cylinder

Consider a cylinder CW L of length L and circumference W with states U, V


at the boundary,
CW L = [pic 44] . (4.17)
Define the integrals
W W
−1 −1
Z Z
H= (T (x) + T (x))dx , P = (T (x) − T (x))dx , (4.18)
2π 0 2πi 0

where T (x) is an abbreviation for the field insertion (T, ζ 7→ ζ + x) and T (x)
for (T , ζ 7→ ζ + x).

Theorem 4.2 :
iW
Let φ ∈ F (h,h) , f (ζ) = 2π ln ζ and V = (φ, f )in , then for any state U we have

[pic 45] . (4.19)

The operator H is the generator of translations along the cylinder, and hence
the Hamiltonian of the CFT on a cylinder of circumference W . The spaces F (∆)
are eigenspaces of H, with eigenvalue 2π c
W (∆ − 12 ). This is the reason for the
name “spectrum” given to the set S.
Proof of theorem: Consider the cylinder CW L and an annulus A with insertion
of T ,
[pic 46] (4.20)
here F (z) = exp( iW 2π
z) as well as f (ζ) = iW ˜
2π ln ζ, f (ζ) = ζ and the metric
W −1
g̃(p)ij = 2π|p| δij is the pullback (with F ) of the flat metric gij = δij on
CW L . The isometry F provides an equivalence between the two world sheets.
Using (CS4) and recalling that the transformation behaviour of T involved the
Schwarzian derivative, one can verify that the insertion T, ζ 7→ F (ζ + x) is
the same as  
2π 2 2 c
 
iW p T (p) − 24 , ζ 7→ ζ + p , (4.21)

where p = F (x). The metric g̃ on A can be brought to the form δij by a Weyl
transformation, changing the amplitude by a constant A(A, g̃) = CA(A, δij ).
By (CS1) we can now use the CFT on C for the world sheet A.

48

Write T (x) C for the T -insertion on CW L and T (p) A for the T -insertion on A.
Then the following identites hold (evaluated in the amplitude for CW L or for
A, as appropriate)

−1 W  2π 2 Z W 
Z
1) c  dx
p2 T (p) A −

T (x) C dx = −i
2π 0 iW 0 24 p=F (x) 2πi
Z  (4.22)
2) 2π c
  3) 2π  c 
pT (p) A − p−1 =

= L0 A − ,
W p
0
6 24 W 24

where 1) substitute the form (4.21) of the T -insertion on A; 2) change variables



p = F (x), dp = iW pdx and change the direction of the integration (to obtain
a counterclockwise contour), which results in a minus sign; 3) carrying out the
p-integration gives the mode L0 . Replacing the state U on A by the field φ
(according to lemma 1.1), we see that 2π c 2π c
W (L0 − 24 ) results in a factor W (h − 24 ).
2π c
Similarly, the T -integral gives W (h − 24 ). Changing the metric on A back to
its original form via CA(A, δij ) = A(A, g̃), then implies the theorem. 

Changing of world sheet moduli

Quite generally in a QFT, varying a correlator with respect to a component


of the metric leads to the insertion of a component of the stress tensor,
δ
φ1 (p1 ) · · · φm (pm ) g ∝ T µν (x)φ1 (p1 ) · · · φm (pm ) g .



(4.23)
δg(p)µν

This point of view is developed in detail in [Gw2, lecture 2]. Here we only need
the following theorem, which is a consequence of (CS1)–(CS5). It implies that
that inserting the operator e−λH on the cylinder CW L results in the cylinder
CW,L+λ and inserting eλP on CW L amounts to cutting the cylinder along the
contour and glueing it back together after a rotation by the angle λ.

Theorem 4.3 :
For λ > 0 and states U, V we have

[pic 47] (4.24)

The proof is in appendix XXX:A.4.

Modular invariance

From theorem 4.3, on a cylinder of circumference W we need exp(W P ) = id.


By theorem 4.2 this imples h − h ∈ Z whenever F (h,h) 6= {0}.
Let TW L be the torus obtained by identifying the two boundaries of CW L
via x ∼ x + iL. We will compute A(TW L ) by writing the amplitude as a sum

49
over intermediate states. Let f : Aε → TW L be given by f (ζ) = iW 2π ln ζ as in
theorem 4.2. Cutting TW L along f (S 1 ) results in a cylinder CW L (φ, φ0 ) where
φ, φ0 label the states.
(Excercise: Show that limL→0 CW L (φ, φ0 ) = (φ0 , φ). Hint: Use the representa-
tion of CW L as an annulus in R2 , as in the proof of theorem 4.2)
Using also theorem 4.3, we can now write, for φ ∈ F (∆) ,
0
A(CW L (φ, φ0 ) = [pic 48] = A(CW L0 [e−(L−L )H ](φ, φ0 ))]
(4.25)
= e−2πL/W (∆−c/12) (φ0 , φ) ,
where in the last step we took the limit L0 → 0. For the torus this implies, via
(CS5),
X X
A(TW L ) = A(CW L (ϕα , ϕ̃α )) = exp(−L 2π c
W (∆α − 12 )) . (4.26)
α∈I α∈I

Remark: An alternative way to compute A(TW L ) is to write


1) X
A(TW L , g) = A(CW L (ϕα , ϕ̃α ), g)
α∈I
2) X 3) X (4.27)
= A(A(ϕα , ϕ̃α ), g̃) = eS[σ] A(A(ϕα , ϕ̃α ), δ) ,
α∈I α∈I

where 1) insert a sum over intermediate states as in (4.26), the metric depen-
dence is kept explicitly in the notation, g is the flat metric on TW L induced by
R2 ; 2) the world sheet A is the annulus defined in the proof of theorem 4.2,
W
it has metric g̃(p)ij = 2π|p| δij and is equivalent to CW L ; 3) write g̃ij = eσ δij
with σ(p) = ln(W/(2π)) − ln |p| and use (CS3) to replace this metric with the
flat metric δ = δij . Using the explicit form of S[σ] it is easy to compute, with
q = e2πL/W the outer radius of A,
Z 2π Z q
1 c c
S[σ] = dθ rdr ∂σ∂σ = 2πL/W . (4.28)
2π 12 0 1 12
For the amplitude A(A(ϕα , ϕ̃α ), δ) on the other hand on finds

A(A(ϕα , ϕ̃α ), δ) = e−L W ∆α , (4.29)
so that we indeed recover (4.26).
Note that the world sheets TW L and TLW are equivalent. Setting τ = −iL/W ,
q = e2πiτ and
X
Z(τ ) = A(TW L ) = trF q L0 −c/24 (q ∗ )L0 −c/24 = q hα −c/24 (q ∗ )hα −c̃/24 (4.30)
α∈I

this amounts to the condition


Z(τ ) = Z(−1/τ ) . (4.31)

50
(Excercise: Show that also Z(τ + 1) = Z(τ ). Hint: Use h − h ∈ Z.)
Remark: The function Z(τ ) can be computed if we know the space of fields F
as a Vir × Vir–module. Modular invariance of the torus is thus a condition on
the representation content of the CFT, but not on its three-point functions. In
fact, (4.31) is just a necessary condition for a CFT on C to be extendable to
arbitrary closed Riemann surfaces. (Modular invariance of one-point functions
on the torus would be sufficient, see YYY:Sonoda)
Any torus can be conformally mapped to C quotiented by the lattice generated
by 1 and τ ∈ H,
[pic 49] . (4.32)
(Excercise: Show that the lattice vectors (1, τ ), (1, τ +1) and (1, −1/τ ) generate
conformally equivalent tori.)
The transformations τ 7→ τ + 1 and τ 7→ −1/τ generate the group SL(2, Z).
The moduli space of complex structures on a torus is given by H/SL(2, Z).
The definition (4.30) was given for purely imaginary τ , but it makes sense for
any τ ∈ H. The conditions Z(τ ) = Z(τ + 1) and Z(τ ) = Z(−1/τ ) imply that,
even though the torus amplitude is parametrised by τ ∈ H, it is really only a
function of the complex structure modulus H/SL(2, Z), as it should be.

Example: Free boson


Let us investigate the condition of modular invariance for the free boson theories
given in terms of a lattice Λ ⊂ R2 in section 2.5.
By definition F = p~∈Λ Vp ⊗Vp . The condition h−h ∈ Z whenever dim F (h,h) >
L

0 is equivalent to p2 /2 − p2 /2 ∈ Z for all p~ ∈ Λ, which holds already since Λ is


an even, integer lattice.
To investigate the condition Z(τ ) = Z(−1/τ ) note that (4.30) gives, for τ = it,
t ∈ R>0 ,
1) X  2) X
trVp q L0 −c/24 trVp q L0 −c/24 =

Z(τ ) = χp (q)χp (q)
~∈Λ
p ~∈Λ
p
3) 1 X (4.33)
= exp(−2πt(p2 + p2 )) ,
|η(τ )|
~∈Λ
p

where 1) use that L0 only acts on the left factor in each tensor product Vp ⊗ Vp
and L0 only on the right factor; 2) subsitute the definition (2.12) of the character
of a representation; 3) substitute the explicit expression (2.13) in the case of the
Heisenberg algebra.
To compute Z(−1/τ ) we need the Poisson resummation formula
X 1 X
exp − πan2 + bn = √ exp − πa (k + b 2
 
a 2πi ) , (4.34)
n∈Z k∈Z

51
as well as the property √
η(−1/τ ) = −iτ η(τ ) (4.35)
of the Dedekind η-function, which implies that 1/|η(τ )| = |τ |/|η(−1/τ )|.

4.3 Surfaces with boundary - XXX


4.4 Cardy condition for the annulus - XXX

5 Application to open cubic string field theory


5.1 The c = −26 ghost system - XXX
5.2 OSFT action - XXX
5.3 Gauge invariance - XXX

A Appendix
A.1 Properties of the map ob
Since the fields
L−m1 · · · L−mk L−n1 · · · L−nl P (A.1)
need not all be linear independent, we first need to show that ob as given in (2.1)
is well-defined. Afterwards, it is also shown that ob is even an isomorphism.

The ob is well defined


Let V be the space of polynomials in infinitely many unkowns V = C[x1 , x2 , . . . ].
By definition a basis of V is given by the monomials 1, xi for i ≥ 1, xi xj for
i ≥ j ≥ 1, etc. Let F̂ = V ⊗C V ⊗C P. Define a linear map π : F̂ → F via

π((xm1 · · · xmk ) ⊗ (xn1 · · · xnl ) ⊗ P ) = L−m1 · · · L−mk L−n1 · · · L−nl P , (A.2)

where we take the mi and ni to be ordered such that m1 ≥ m2 ≥ · · · ≥ mk and


n1 ≥ n2 ≥ · · · ≥ nl . This defines π on a basis of F̂. By assumption (A1), π is
surjective (but in general not injective).
Define a map ôb : F̂ → Hb by setting, for P ∈ P (h,h) ,

ôb (xm1 · · · xmk ) ⊗ (xn1 · · · xnl ) ⊗ P = lim L2(h+h)



L→∞
X   (A.3)
AR2 P (L)Lmk (b) · · · Lm1 (b)Lnk (b) · · · Ln1 (b)ϕα (0) · ũα ,
α∈I

where again m1 ≥ m2 ≥ · · · ≥ mk and n1 ≥ n2 ≥ · · · ≥ nl . This defines ôb on a


basis of F̂.

52
We need to show that there exists a map ob from F to Hb which makes the
following diagram commute
[pic 24] . (A.4)
Since π is surjectiv, if it exists, the map ob is unique (and given by (2.1)). The
map ob exists if and only if
ker π ⊂ ker ôb . (A.5)
We first need the following lemma.

Lemma A.1 :
Consider the fields
φ = L−a1 · · · L−ap L−b1 · · · L−bq P ,
(A.6)
φ0 = L−c1 · · · L−cr L−d1 · · · L−ds P 0 ,
0 0
for P ∈ P (h,h) , P 0 ∈ P (h ,h ) . The expression (φ, φ0 ) can be nonzero only if
0
h=h0 , h=h as well as
p
X r
X q
X s
X
ak = cl and bk = dl . (A.7)
k=1 l=1 k=1 l=1

Proof: We will show the lemma for q = s = 0, the general case works analo-
0
gously. Set ∆ = h + h + ak and ∆0 = h0 + h + cl . Below (C2) we have seen
P P
0
that (φ,
P φ ) =P 0 unless ∆ = ∆0 . We may thus assume that ∆ = ∆0 . Suppose
that ak 6= cl . This implies h 6= h0 . By contour deformation (as in the
proof of lemma 1.17) one may show that

AR φ(z)φ0 (w) = Dz,w AR P (z)P 0 (w)


   
(A.8)

for some linear differential operator Dz,w . By lemma 1.16, the amplitude
0
AR P (z)P 0 (w) is nonzero only if h = h0 and h = h . Since (φ, φ0 ) is ob-
 
0
tained
P P taking a limit 0of the lhs of (A.8), we have shown that h 6= h or
by
ak 6= cl implies (φ, φ ) = 0. 
Let η ∈ P̂ s.t. π(η) = 0. We need to show that this implies ôb (η) = 0. Recall
from below (C3) that the pairing d : Hb × F → C given by
 
d(v, φ) = ADb φ(bo ) (v) (A.9)

is non-degenerate. It is thus enough to verify that d(ôb (η), φ) = 0 for a set of φ


spanning F. We will choose all φ of the form φ = M P 0 , where we abbreviated

M = L−a1 · · · L−ar L−b1 · · · L−bs , (A.10)

and P 0 ∈ P (h,h) . The vector η can be written as a linear combination of basis


vectors 
f = (xm1 · · · xmk ) ⊗ (xn1 · · · xnl ) ⊗ P . (A.11)

53
To proceed we will rewrite d(ôb (f ), φ). Using lemma A.1 one checks that this
can only be nonzero if also P ∈ P (h,h) . It is convenient to abbreviate

M̂ = L−m1 · · · L−mk L−n1 · · · L−nl , (A.12)

such that π(f ) = M P , as well as

M t = Lar · · · La1 Lbs · · · Lb1 , M̂ t = Lmk · · · Lm1 Lnl · · · Ln1 . (A.13)

We can then write


1)
d(ôb (f ), φ) = lim L2(h+h) AR2 P (L)M̂ t (b)φ(0)
 
L→∞
2)
= lim L2(h+h) AR2 P (L)M̂ t (b)M (b)P 0 (1/L)
 
L→∞
3)
= lim L2(h+h) (−1)2s AR2 P 0 (L)M (b)t M̂ P (1/L)
 
L→∞
4)  5)
= lim L2(h+h) (−1)2s AR2 P 0 (L)M (b)t π(f )(0) = (−1)2s d(ôb (g), π(f )(0))

L→∞
(A.14)
where 1) By the notation M̂ t (b) it is meant that the modes in (A.13) are placed
on the circle b, which we take to have bo = 0. The expressions M (b), M̂ (b)
and M t (b) are defined in the same way. To obtain this equality one has to
use (C4). 2) The definition φ(0) = M (b)P 0 (0) has been inserted, and then the
insertion P 0 (0) has been moved to P 0 (1/L). This does not affect the L → ∞
limit. 3) This equality needs a little calculation. Rewrite the modes as contour
integrals and apply the Möbius-transformation ζ 7→ 1/ζ to the correlator. This
will exchange a mode Lm (b) with L−m (b0 ) and invert the ordering of the modes.
The factor (−1)2s with s = h−h results from the transformation of P and P 0 .
4) By definition M̂ P (1/L) = π(f )(1/L). The field π(f ) can then be moved
from 1/L to 0 without changing the L → ∞ limit. 5) Here we defined g =
(xa1 · · · xar ) ⊗ (xb1 · · · xbs ) ⊗ P 0 such that φ = π(g).
Since (A.14) holds for any f of the form (A.11) it holds for any vector in F̂.
In particular, for all φ,

d(ôb (η), φ) = (−1)2s d(ôb (g), π(η)(0)) = 0 , (A.15)

as π(η) = 0. Hence also ôb (η) = 0. This finishes the argument that ob is
well-defined.

The map ob is an isomorphism


To prove that ob is an isomorphism it is enough to show that the bilinear form
hφ|φ0 i defined in section (2.1) is non-degenerate. Using a manipulation similar
0 0
to step 3) in (A.14) one can show that for φ ∈ F (h,h) and φ0 ∈ F (h ,h ) we have
0
hφ|φ0 i = (−1)s+s hφ0 |φi , (A.16)

54
0
where s = h−h and s = h0 −h . To show non-degeneracy we thus only need to
establish
hn|φi = 0 for all φ ∈ F ⇒ n = 0 . (A.17)
Let N ⊂ F be the vector space of all n such that hn|φi = 0 for all φ∈F. We
will show that N = {0} by contradiction.
Let n ∈ N be a non-zero element. We will first show that given n non-zero,
we can find a different non-zero field n0 in N which is even primary. To this
end, consider the fields f = La Lb n. If these are zero for all a, b > 0, then n is
already primary. Suppose thus that there are a, b > 0 such that f is nonzero.
Below we will show that in fact f ∈ N , so that we can repeat this procedure.
By assumption (A1) the process is guaranteed to terminate and gives us the n0
we are looking for.
Consider thus the field f = La Lb n for n ∈ N . We have, for all fields φ ∈ F,
1) 2)
hf |φi = hLa Lb n|φi = hn|L−a L−b φi = 0 , (A.18)

where 1) is just property (2.5) and 2) is true due to the definition of N .


So far we have succeeded to show that if N contains a non-zero element, it
also contains a non-zero primary element n0 . Since the pairing ( · , · ) is non-
degenerate by definition, there is a field φ such that (n0 , φ) 6= 0. From lemma A.1
we know that, since n0 is primary, also φ has to be primary to give the non-zero
result. By (2.4) we then have hn0 |φi = (n0 , φ) 6= 0. This is a contradiction to
the definition of the subspace N . Hence N cannot contain a non-zero element.

A.2 Proof of lemma 1.17


By definition we have, for any φ1,2,3 ∈ F,

hφ1 |φ2 (z)φ3 (0)i = ι−1



b (ob (φ1 )) , φ2 (z)φ3 (0) . (A.19)

Conversely then, we can write

(φ1 , φ2 (z)φ3 (0)) = ho−1


b (ιb (φ1 )) | φ2 (z)φ3 (0)i . (A.20)

To establish lemma 1.17 it is thus sufficient to show that all hφ1 |φ2 (z)φ3 (0)i are
fixed in terms of (P1 , P2 (1)P3 (0). On the other hand, again by definition, for a
primary field P we have

hP |φ1 (p1 ) · · · φm (pm )i = (P, φ1 (p1 ) · · · φm (pm )) , (A.21)

so that in fact hP1 |P2 (1)P3 (0)i = (P1 , P2 (1)P3 (0). Lemma 1.17 is thus a corol-
lary of the following lemma.

Lemma A.2 :
The functions hφ1 |φ2 (z)φ3 (0)i, for φ1,2,3 ∈ F are all uniquely fixed in terms of
hP1 |P2 (1)P3 (0)i for primary fields P1,2,3 .

55
Proof: We will proceed in four steps.
(i) In the mode integral for L−m φ(z), use the contour deformation

[pic 17] (A.22)

and use the appropriate Laurent expansion for (ζ − z)−m+1 to get the formula

X 
L−m φ(z) = ak z k L−m−k φ(z) + (−1)m z −m−k+1 φ(z)Lk−1 , (A.23)
k=0

where a0 = 1, ak = (m−1)m(m+1) · · · (m+k−2)/(k − 1)!. By (A1)(ii) every


field is a descendent of a primary field. Suppose φ2 (z) is a descendent of the
primary field P2 (z). Use relation (A.23) to shift all modes from φ2 (z) to φ1 and
φ3 ,
hφ1 |φ2 (z)φ3 (0)i −→ hφ01 |P2 (z)φ03 (0)i , (A.24)
where the arrow means ‘gets replaced by a sum of terms of the form’.
(ii) Next shift all modes from φ01 to φ03 , using (2.3) and lemma 1.13. In this way
we arrive at
hφ01 |P2 (z)φ03 (0)i −→ hP1 |P2 (z)φ003 (0)i , (A.25)
so already two of the three fields are primary.
(iii) Write, for m > 0,

hP1 |P2 (z)L−m φ003 (0)i

= −hP1 |[L−m , P2 (z)]φ003 (0)i + hP1 |L−m P2 (z)φ003 (0)i (A.26)


1)
= −hP1 |[L−m , P2 (z)]φ003 (0)i ,

where 1) we used that hP1 |L−m P2 (z)φ003 (0)i = hLm P1 |P2 (z)φ003 (0)i = 0 as P1 is
primary. Relation (A.26) can be used to also convert φ003 to a primary field. (By
lemma 1.13 P2 remains primary.)
(iv) So far we managed to express hφ1 |φ2 (z)φ3 (0)i through hP1 |P2 (z)P3 (0)i
using the commutation relation of the modes. Finally, for Pi ∈ P (hi ,hi ) , the
differential equation coming from

hL0 P1 |P2 (z)P3 (0)i = hP1 |[L0 , P2 (z)]P3 (0)i + hP1 , P2 (z)L0 P3 (0)| i , (A.27)

i.e., abbreviating f (z) = hP1 |P2 (z)P3 (0)i,

h1 f (z) = (h2 + z∂)f (z) + h3 f (z) (A.28)

together with the corresponding equation from L0 , implies

hP1 |P2 (z)P3 (0)i = z h1 −h2 −h3 z h1 −h2 −h3 hP1 |P2 (1)P3 (0)i . (A.29)

56
A.3 Proof of lemma 4.1 - XXX
XXX

A.4 Proof of theorem 4.3 - XXX


XXX

References
[BS] P. Bouwknegt, K. Schoutens, W symmetry in conformal field theory,
Phys. Rept. 223 (1993) 183, [arXiv:hep-th/9210010].
[DMS] P. Di Francesco, P. Mathieu, D. Senechal, Conformal Field Theory,
Springer (1997).
[FB] E. Frenkel, D. Ben-Zvi, Vertex Algebras and Algebraic Curves: Second
Edition, AMS (2004),
[http://math.berkeley.edu/ frenkel/master2.ps.gz].
[Fu] J. Fuchs, Lectures on conformal field theory and Kac-Moody algebras,
Budapest 1996, Conformal field theories and integrable models,
Proceedings,[arXiv:hep-th/9702194].
[Gb] M. R. Gaberdiel, An introduction to conformal field theory, Rept.
Prog. Phys. 63 (2000) 607, [arXiv:hep-th/9910156].
[Gi] P. H. Ginsparg, Applied Conformal Field Theory, Les Houches Summer
School (1988) 1–168, [arXiv:hep-th/9108028].
[Gw1] K. Gawȩdzki, Conformal Field Theory, Seminaire Bourbaki, No.704,
Asterisque 177-178 (1989).
[Gw2] K. Gawȩdzki, Lectures on Conformal Field Theory, in Quantum Fields
and Strings: A Course For Mathematicians (P. Deligne, P. Etingof,
D.S. Freed, L. Jeffrey, D. Kazhdan, J. Morgan, D.R. Morrison and E.
Witten, eds.,), 2 vols., AMS (1999),
[http://www.math.ias.edu/QFT/fall/NewGaw.ps].
[Sc] A. N. Schellekens, Introduction to Conformal Field Theory, Fortsch.
Phys. 44 (1996) 605, [http://www.nikhef.nl/~t58/CFT.ps.gz].
[Se1] G. B. Segal, The Definition Of Conformal Field Theory, Differential
Geometrical Methods in Theoretical Physics 165–171, Proceedings,
Kluwer 1988.
[Se2] G. B. Segal, Two-Dimensional Conformal Field Theories And Modular
Functors, Swansea 1988, Proceedings, Mathematical physics 22–37.

57
[SFW] C. Schweigert, J. Fuchs, J. Walcher, Conformal field theory, boundary
conditions and applications to string theory, Budapest 2000,
Non-perturbative QFT methods and their applications,
Proceedings,[arXiv:hep-th/0011109].
[St1] J. Stasheff, What is an operad?, Notices of the AMS 51 (2004)
630–631, [http://www.ams.org/notices/200406/what-is.pdf].

58

You might also like