| DRAFT | IASSNS-HEP-97/72

Lectures 5, 6

III. String Amplitudes and Moduli Space of Curves

Eric D'Hoker

In the previous Chapter, we have quantized free bosonic strings, determined their
spectrum and Hilbert space of states under certain conditions: D = 26 and the presence
of massless gravitons and Yang-Mills particles. (The case D  25, allowed by the no-ghost
theorem, was shown to be a subcase.)
Now, we shall quantize interacting strings, and investigate quantum mechanical transi-
tion amplitudes for the scattering of string states. While operator methods were convenient
for determining the spectrum of free strings, the functional integral formulation in terms of
a summation over surfaces is better suited for dealing with interactions. Thus, our starting
point is the transition amplitude
X Z 1 Z
A= Dg N(g) V1 : : : VN e S[x;g] (3:1)
topologies  Met() Di(;M )
(h=0;:::;1)

with Riemannian space-time M , and external string states represented by vertex operators
Vi , i = 1; : : : ; N . This was precisely our original starting point with Euclidean signature
metric on M . Minkowski amplitudes (for example, for at M = RDM ) are obtained by
analytic continuation in external momenta and polarization tensors. The vertex operators
were determined in xII in terms of conformal primary elds of weight (1; 1). The summation
in the transition amplitude A is over topologies of compact oriented surfaces, since we shall
limit ourselves here to the study of the closed oriented string.
The strategy is as follows:
1) At xed topology of , we use the product structure of Met() = Weyl() conformal
classes () to decompose the measure Dg on Met() into a product of measures on the
orbits of Weyl, on the orbits of Di() and on the moduli space Mh = Met()=Di() n
III.1
Weyl(). A Jacobian determinant arises, which is essentially the Faddeev-Popov determi-
nant. This decomposition holds for any matter conformal eld theory, in particular for any
space-time M , dening a conformal eld theory.
2) For at Euclidean space-time M , the combined measure | including the Faddeev-
Popov determinant | and the correlation functions, are found to be Weyl-invariant (i.e.
Weyl() is a symmetry group) if and only if D = 26 and gravitons are massless on surfaces
of any topology (equivalently, the a = 1 condition of xII holds).
3) Under the assumptions of 2) the string theory is usually called critical. In this case, the
formal measure Dg=N(g) on Met(), divided by the volume of the orbits of the symmetry
group Di() n Weyl(), naturally projects down to an intrinsically dened measure on
Mh . Thus, the transition amplitudes for critical strings reduce to well-dened integrals
over Mh of the matter conformal eld theory correlation functions.
4) More generally, the assumptions of 2) may be relaxed, and the matter part may be any
conformal eld theory of central charge c = 26 and with correlation functions satisfying
certain Weyl invariance conditions.
5) In particular, in Lecture 9, we shall generalize the Weyl invariance conditions to the
case of curved space-time manifolds M | in certain limits.
6) Under the assumptions of 2), we shall obtain in this chapter 0- and 1-loop transition
amplitudes for critical bosonic string theory.

A. Finite-Dimensional Case
Let P be a nite-dimensional manifold, with Riemannian metric G, and associated
measure dG. Let H be a Lie group acting on P , and assume that dG is H -invariant.
We wish to reduce the integrals of H -invariant functions on P to integrals on the quotient
P = P=H . Let f (x): P ! C be H -invariant,

f (xh) = f (x) for all x 2 P; h 2 H:

III.2
Then for any section s: P ! P , we have
Z Z Z
dGf = d s (f )J dH ;
P P H
where d is the measure on P induced from dG , dH is the left-invariant measure on H ,
and J is a Jacobian (or Faddeev-Popov) determinant which we dene below, as follows.
First, let us present a gure:
P
orbit of H

x section s

π
_
_ P = P/H
x

The tangent space at x to P may be decomposed into the part tangent to the orbit of H ,
generated by the Lie algebra H,
i: H ! Tx P
and a part from the push-forward of the tangent space to Tx P :

s : Tx P ! Tx (P ) ;

as follows:
s  i: Tx P  H ! Tx(P ):
The Jacobian J is now obtained as

J = det(Tx P  H ! TxP ):

III.3
B. Basic Notation: Tensors, Derivatives
Fix a positive denite metric g on  and choose local complex coordinates z, z in
which
g = gzz(dz
dz + dz
dz) = 2gzzjdzj2
Tensors decompose into one-dimensional components of weight (m; n), which are sections
of a line bundle K (m;n):

 = z| :{z: : z} |z :{z: : z}(dz)m (dz)n 2 K (m;n) :

m n

Using the metric g, we identify tensors of weight (m; n) with tensors of weight (m n; 0).
The space of tensors of weight (m; 0) is denoted K m , with K 1 the space of sections of the
canonical line bundle T(1 ;0)(). (Generally, we do not distinguish between a bundle and
the space of its sections.)
On sections of K m we have the L2 inner product
Z
(1; 2 )g = dg (gzz) m 1 2 (3:2)

as well as the covariant derivatives

r = rz(m) + r(zm)
where rz(m) = @(m) is the Cauchy Riemannn operator
rz(m): K m ! K (m;1) rz(m) = @@z 
dz : (3:3)

Equivalently, identifying (m; 1)  (m 1; 0), we have

8 rz : K m ! K m 1
< (m) rz(m) = (gzz) 1 @ 
(dz ) 1
@ z
: r(zm): K m ! K m+1 (3:4)
r(m) = (g
z @ (gzz) m 
dz
zz)m @z :
(In general, we shall not exhibit the dierentials dz and dz in r.) The adjoints of these
operators with respect to ( ; )g are
 y  y
rz(m) = rz(m 1); r(zm) = rz(m+1) (3:5)
III.4
and the Laplace operators are

+(m) = 2rz(m+1)r(zm)
(3:6)

(m) = 2rz(m 1)rz(m)
We also make use of the Riemann-Roch-Atiyah-Singer theorem
 1
dim Kerrz (m ) z
dim Kerr(m+1) = m + 2 () = (2m + 1)(1 h) ; (3:7)
as well as the vanishing theorems:
 h = 0; m  1; Ker rz(m) = 0
(no holomorphic forms on the sphere)
(3:8)
 h  2; m  1; Ker rz(m) = 0 :
(no holomorphic vector elds for h  2)

C. Space of Metrics { Moduli Space of Riemann Surfaces

Let Met() denote the space of positive denite metrics on , and Tg Met() its tangent
space at g, consisting of innitesimal deformations g 2 K (1;1)  K (2;0)  K (0;2). Following
the results of xB, we have an L2 norm kgk2g = (g; g)g on Tg Met(), which makes Met()
into an innite dimensional Riemannian manifold. The measure Dg is dened with respect
to kgk.
Di() acts isometrically on Met(), but Weyl() does not. The space of orbits of
Di() n Weyl() in Met() is the moduli space of Riemann surfaces (or complex curves)
of genus h:
Mh  Met()=Di() n Weyl() : (3:9)
Actually, the group Di() is not connected. The group of connected components is the
mapping class group G :
G = 0(Di()) ;
which is an innite discrete group for h  1. The component connected to the identity
in Di() is denoted by Di 0(). Thus, there arises another natural space: Jh , which is
Teichmuller space
Th  Met()=Di 0 () n Weyl() ;
III.5
which is the simply connected covering space of Mh :
Mh = Tn=G :
Teichmuller space is a complex nite dimensional manifold, while moduli space is a com-
plex nite dimensional orbifold (the action of G on Th produces singularities), both of
dimension (0 h=0
dimC Th = dimC Mh = 1 h=1 (3:10)
3h 3 h  2 :
We have explained that string transition amplitudes can be reduced to an integral over
moduli space. We wish to factorize the measure Dg into its components along Di(),
Weyl() and Mh, i.e. we want to x the gauge. To examine this factorization concretely,
we decompose Tg Met() into the tangent spaces to the orbits of Di 0() and Weyl(), as
explained in the nite dimensional example. We perform this factorization at an arbitrary
metric g 2 Met(), and shall choose a section later on.
We use the fact that Di 0 () is generated by vector elds v = vz  vz 2 K ( 1;0) 
K (0; 1). A Weyl transformation will in general be denoted by e2 , where  takes real
values.
Let the reference metric be g = 2gzzjdzj2; an arbitrary metric g + g in the neighbor-
hood of g may then be conveniently parametrized by a change in gzz as well as by Beltrami
dierentials zz 2 K ( 1;1) and zz 2 K (1; 1).
g + g = 2 (gzz + gzz) jdz + zz dzj2 : (3:11)
The space Tg Met() admits the orthogonal decomposition
Tg Met() = K (1;1)  K (2;0)  K (0;2)
g = 2gzzjdzj2 + 2gzzzz(dz)2 + 2gzzzz(dz)2 : (3:12)
The innitesimal action of (vz , vz, 2) in the Lie aglebra of Di 0() n Weyl() is then
simply 8 g = (2 + r vz + r vz)g
>
>
< zzz z z zz

> z = rzvz (3:13)

>
: zz = rz vz :
III.6
While the range of Weyl() is the full K (1;1), in general the range of Di 0() need not be
all of K (2;0)  K (0;2). Instead, we have the orthogonal decomposition

K (2;0) = Range r(1) (1) y

z  Ker(rz ) : (3:14)

The space Ker (r(1) (2)

z )y = Ker rz consists of holomorphic quadratic dierentials on ,
and can be identied with the holomorphic cotangent space of Mh :

T(1 ;0)(Mh ) = Ker r(2)

z (3:15)

(Analogously, we have T(0 ;1)(Mh ) = Ker rz( 2).) To see this, we notice that holomorphic
dierentials j 2 Ker r(2)
z provide linear forms on the space of dierentials  :

Z
(; j ) = dz dz zzjzz (3:16)


The above pairing is Weyl()-invariant and the kernel of (
; j ) (and (
; j )) is precisely
the tangent space to the orbits of Di 0() and Weyl(); hence the identication. By the
Riemann-Roch theorem, and the vanishing theorems listed above, it is easy to see that the
dimensions of Mh and of Ker r(2) ( 2)
z  Ker rz coincide.
More precisely, the holomorphic quadratic dierentials j 2 Ker r(2) z form a complex
vector space, thus providing a natural almost complex structure on the tangent space of
Mh . In fact, this almost complex structure is integrable and results in a complex structure
on Mh; for a given choice of , the above linear forms dene local complex coordinates mj ,
m j on Mh given by
mj = (; j ) m j = (; j ) : (3:17)

We may represent this projection onto Mh by a picture analogous to the one used ot discuss
III.7
the nite dimensional example at the beginning of this lecture.
H H Met( Σ )

g +δ g
H = Diff (Σ ) x Weyl (Σ )
g

M = Met( Σ )/H
h
δm , δ m
j j

Vice-versa, for a given set of complex coordinates (mj ; m j ) on Mh , we may choose a

section s of Mh in Met() | also called a slice | of dimension dim Mh , and parametrized
by (mj ; m j ):
H Met( Σ )

g (m ,m )
j j
s

Mh
( m ,m )
j j

The tangent space at g(mj ; m j ) then decomposes into the tangent directions to the action
of Weyl(), the action of Di 0() on the dierentials , , and moduli deformation,
given by a basis of Beltrami dierentials j and j :
8 dim M
>
>  z = rz( 1)v z + P h mj z
< z

j =1
j z
> z (1) z dimPMh z (3:18)
>
: z = rz v + m j jz :
j =1
We may choose the Beltrami dierentials as follows
III.8
 8 z zz @gzz
< jz = g @mj (m; m )


: zjz = gzz @g@mzzj (m; m ) ; (3:19)

so that j and j are unchanged under Weyl transformations g ! e2 g: j ! j ; j ! j .

Thus, under a change of section s, j and j will be shifted only by the action of a vector
eld in the Lie algebra of Di 0().
Finally, for  with the topology of either the sphere or the torus, we also have holo-
morphic vector elds in Ker r(z 1). These generate conformal automorphisms, and do not
act on  at all. We have
8h = 0 dim Kerr(z 1) = 3
< SL(2; C )
: hh = 21 dim Kerr(z 1) = 1 T2 (3:20)
dim Kerr(z 1) = 0 possibly nite automorphism group :

D. Factorizing the Integration Measure

We decompose the integration measure Dg on Met() into measures on the orbits of
Di() and Weyl() and on Mh by using the parametrization developed above for g in
terms of Weyl changes  and dierentials , :
g = 2g + 2gzzzz (dz)2 + 2gzzzz(dz)2 : (3:21)
The relation
kgk2g = 4kk2g + 2kk2g + 2kk2g (3:22)
implies that Dg = D D D, in obvious notation.
Next, we make use of the decomposition of  and  into orbits of Di 0 () and
moduli deformations, which we parametrize by coordinates (mj ; m j ) on Mh and a choice
of Beltrami dierentials zjz and zjz , as in (3.18). Using the orthogonal decomposition of
the dierentials zz gzz according to
K (2;0) = Ranger(1) (2)
z  Ker rz ; (3:23)
we evaluate kk2g and nd
2kk2g = 2kr(z 1)vz k2g + k? projection of zz gzz onto Ker r(2)
z kg :
2 (3:24)
III.9
The second term on the right hand side is easily worked out by choosing an orthogonal
 , with basis vectors j , j = 1; : : : ; dim Mh :
basis for Ker rz(2)
k? projection of zz gzz onto Ker r(2)
z kg
2
X
= (; j )g (j ; j )g 1 (j ; )g
j (3:25)
X
= mk m `(k ; j )g (j ; j )g 1 (j ; `)g
j;k;`
The rst term on the right hand side of (3.24) is conveniently rewritten in terms of
the Laplace operator
( 1):
2kr(z 1)vz k2g = (vz ;
( 1)vz )g
A last point must be claried before evaluating the measure D D. For the special
cases of the sphere (h = 0) and the torus (h = 1), Ker rz( 1) 6= 0, and there are (contin-
uous families of) conformal automorphisms of : SL(2; C ) for h = 0 and T 2 for h = 1.
The corresponding vector elds leave  and  invariant. Thus, in the change of variables,
 ! (v; mj ), we must restrict to vector elds v that are orthogonal to Ker r(z 1).
The corresponding restricted volume element on vector elds v is denoted by D0 v. The
associated determinant of
( 1) must be similarly restricted to non-zero modes.
Putting all together, we nd
Dg = det 0
( 1) j det(j ; k )g j2 Y dm dm D0 v
D : (3:26)
det(j ; k )g j j j
While originally we chose j 2 Ker r(2)
z to form an orthogonal basis, it is now clear that
the above formula holds for any basis Ker r(2)
z . (The nite dimensional determinants are
of dim Mh  dim Mh dimensional matrices.)
It is standard practice | and extremely useful | to introduce the following combina-
tion of determinants
 Det0
(n)
Z(n)(g)  det( ;  ) det( ; ) (3:27)
j kg a bg
where
 j 2 Ker rz(n+1) for + ; j 2 rz(n 1) for
 a 2 Ker r(zn) for + ; a 2 rz(n) for :
III.10
While Z(n) have non-trivial Weyl transformation properties, we shall show in the next
section that this particular combination has simple transformation properties.
Putting all together, we have the following expression for the measure Dg:

Theorem: dim Mh
e Z( Y
Dg = D Dv 1) (g ) j det(j ; k )j2
g dmj dm j ; (3:28)
j =1
where we have
e  D0 v det( a ; b)g . For h  2, we have Dv
1) Dv e = Dv, i.e. the measure over all vector
e until
elds of Di 0 (). For h = 0; 1, we shall postpone the explicit construction of Dv
xG.
2) The measure
Y
j det(j ; k )g j2 dmj dm j (3:29)
j
is intrinsically dened on Mh. To see this, notice that we chose j to be Weyl inde-
pendent, that j can be chosen Weyl independent, and that the pairing (j ; k )g is
Weyl invariant. Furthermore, we dened Beltrami dierentials by

zjz = gzz @g
@m
zz (3:30)
j
so that the above combination is manifestly unchanged under reparametrizations of
mj and m j .
3) Z( 1)(g) will be Weyl dependent, as we shall establish in the next section. This object
is called the Faddeev-Popov determinant (up to the nite dimensional determinants),
expressing the fact that we gauge xed the Di 0 ()-invariance.
4) To complete the construction outlined in the nite-dimensional case, for the projection
down onto Mh of the measure Dg on Met(), we will have to understand the Weyl
dependence of the combined Z( 1)(g) and matter contributions, which we do in the
next section.

III.11
E. Weyl Rescalings of Functional Determinants
The determinant combinations Z(n) (including those with n 2 12 + Z, which will enter
when dealing with the superstring, as we shall see later on) will be considered repeatedly,
and we shall need their behavior under Weyl() rescalings of the metric. Throughout, we
shall adopt a regularization and renormalization scheme that preserves Di() invariance.

Theorem.
Z(n)(ge2 ) = Z(n)(g)e 2c(n) SL (;g) (3:31)
where c(n) is given by
c(n)  6n2  6n + 1 (3:32)
and the Liouville action is independent of n and given by
Z  
SL(; g) = 12 dg 21 gmn@m @n + Rg  +  2(e2 1) :
1 (3:33)


The constant  depends upon the renormalization scheme used, and will vanish when de-
terminants are dened with  -function regularization, or with dimensional regularization.
Since special care must be paid to the eects of the zero modes, we outline the proof.
(Various intermediate steps will be proven in Problem Set IV.) First, it suces to prove
the formula for the Weyl transformation of Z(n) by innitesimal ; the result for general
 follows by straightforward integration (Gawedzki, prob. 8). Thus, we must prove

 c(n) Z
 ln Z(n)(g) = 6 dg [Rg  + 2 2 ]


(Notice that from the form of this innitesimal variation, it is clear that the determinant
Z(n) denes a conformal eld theory with central charge related to c(n).)
We dene determinants by introducing the  -function associated with the eigenvalues
 of
(n):
X0
 (s)   s

III.12
 P
Here, 0 denotes the sum over all non-zero eigenvalues  of
(n). It is standard to
represent this sum in terms of the heat-kernel of
(n):
1 Z1 
 (s) = (s) dt ts 1(Tr e t
(n) N(n)) (3:35)
0
where N(n) = dim Ker
(n).
The  (s) function series and the heat-kernel integral representation are absolutely
convergent for Re(s) > 1.  (s) is holomorphic for Re(s) > 1, but may be analytically
continued throughout C , with a simple pole at s = 1. The determinants are then dened
by
ln Det0
(n) =  0(0) (3:36)
The non-zero spectra of
+(n) and
(n+1) coincide and thus their determinants are equal,
so we shall henceforth focus on (n).
To evaluate the Weyl dependence of determinants, we compute that of  (s) and ana-
lytically continue to s = 0. We use the following intermediate results (proven in Problem
Set IV).
1) N(n) are Weyl independent
 Tr e t
(n) = t @t@ Tr [2ne t
(n 1) (2n 2)e t
(n) ]
+
2)
3)
 1 R dg  + 13n R dg Rg  + O(t)
Tr e t
(n) = 4t (3.37)
 12 
4) Tr e t
(n) = 2n1 2  ln det(j ; k )g + O(e 1 t)
Tr e t
(n 1) = 21n  ln det( a ; b)g + O(e 1 t )
+
5)
where 1 is the smallest non-zero eigenvalue of
(n).
We make use of the above results and carry out the analytic continuation of   (s)
to s = 0, and then consider  0 (0). For Re(s) > 1, using 1) and 2), we have

  (s) = 1 Z 1 dt ts @ Tr [
](t) (3:38)

(s) 0 @t
where we introduced the abbreviation

Tr [
](t)  tr[2ne t
(n (2n 2)e t
(n) ]

+
1) (3:39)
III.13
We shall make use of the asymptotic behaviors of Tr [
](t) as t ! 0 and t ! 1, which
are obtained from 3), 4), 5):
8
< Tr [
](t) = 2t1 R dg  + c6n R dg Rg + O(t)
> ( )

> (3:40)
: Tr [
](t) =  ln det(j ; k )g det( a ; b)g + O(e  t) : 1

Here, c(n) was dened in (3.32).

To separate the asymptotic behaviors as t ! 0, t ! 1, we cut [0; 1] at t = t0 > 0,
and dene   (s) =  0(s) +  1(s)
8 R
>
<  0 (s) = 1 t0 s@
(s) 0 dt t @t Tr [
](t)
> R (3:41)
:  1(s) = 1 1 s@
(s) t0 dt t @t Tr [
](t) :
The rst term fails to converge at s = 0, so we subtract the asymptotic behavior of Tr [
](t)
in 1=t as t ! 0:
1
Z t0 @  1 Z 
 0 (s) = (s) s
dt t @t Tr [
](t) 2t dg 
0 
Z Z t
+ (1s) 21 dg  dt ts 2 :
0

 0

The rst integral on the right hand side is convergent at s = 0, while the second may be
explicitly analytically continued to s = 0. Taking the derivative at s = 0:
Zt 0
@  1 Z  1 Z
  0 (0) =
0 dt @t Tr [
](t) 2t dg  2t dg 
0 
0
 
Z 

= Tr [
](t0 ) + Tr [
](t) 1 dg  
2t  t=0
c(n) Z
= Tr [
](t0 ) + 6 dg Rg  :


The term  1(s) is given by

 1(s) = 1 Z 1 dt ts @ Tr [
](t)
(s) t0 @t
III.14
As t ! 1, Tr [ ] ! constant + O(e 1 t), 1 > 0, so that the integral is exponentially
convergent, and analytic around s = 0. Hence, using the asymptotics of Tr [
](t) as t ! 1:
Z1 @
 10 (0) = dt @t Tr[ ](t)
t0
=  ln det(j ; k )g det( a ; b )g + Tr [
](t0) :
Putting all together, we have
 ln Det0
(n) =  0 (0)
c(n) Z
=  ln det(j ; k )g det( a ; b )g + 6 dg Rg  ;

as announced previously. Notice that in  -function denitions of the determinants, no \area
term" involving the additional constant  arises.

F. Critical Central Charge { Critical Dimension

We complete the reduction of the measure Dg to a measure on Mh , making use of the
Weyl transformation properties of functional determinants, derived in xE.
Just as we did in the case of the nite-dimensional example, we choose a section s
of Met() and denote the corresponding metric g^(mj ; m j ). A general metric g 2 Met()
may then be parametrized in terms of the action of Di() and Weyl() on g^(mj ; m j ):
let f 2 Di() and e2 2 Weyl(), then

H Met( Σ )
2σ
<

g = f*(e g )
<

g (m ,m )
j j
s

π
M
h
( m ,m )
j j

III.15
The action of Di() is by isometries, and our denition of functional determinants was
invariant with respect to Di(). Thus, the eect of f  is immaterial and may be dropped.
The eect of Weyl transformations is obtained by combining the results of xD, E, and we
have
e
Dg = DvDZ ( 1) (^g )e
26SL (^g;) j det(^
j ; k )g^j2 dmj dm j : (3:42)
We now consider a full transition amplitude Ah , evaluated on surfaces  of genus h.
(To avoid certain complications due to conformal automorphisms, we postpone the cases
h = 0; 1 and restrict to h  2. The cases h = 0; 1 are completely analogous and will be dealt
with separately in xI.) Recall that
Z
Ah = Dg N1(g) hV1 : : : VN ig ; (3:43)
Met()
where Z
hV1 : : : VN ig = Dx V1 : : : VN e S [x;g] : (3:44)
Di(;M )
Given the Weyl transformation properties of Dg established above, it is clear that a
matter theory (i.e. the quantum eld theory of x elds for given metric g) is singled out that
makes the Weyl dependence of the amplitude trivial. We dene a critical (bosonic) string
theory as one governed by a matter conformal eld theory with central charge c = 26, and
with physical vertex operators Vi , i = 1; : : : ; N such that

hV1 : : : VN ie2 g^ = e26SL (^g;) hV1 : : : VN ig^ : (3:45)

(This last condition puts restrictions on the allowed vertex operators in the theory.) For
critical string theories, the amplitudes at xed topology (h  2) reduce to
Z Y
Ah = dmj dm j j det(^j ; k )g^ j2Z( 1)(^g) hV1 : : : VN ig^
Mh j
Z Z (3:46)

D Dv=N0(g)
Weyl() Di 0 ()

e = Dv.) Now, recall that the formal measure Dg=N(g) was the
(Recall that for h  2, Dv
measure on Met(), divided by the volume of the orbits of the group Weyl()  Di().
III.16
Since we restricted our integration to one over Mh (and not Teichmuller space Th ), we
eectively quotiented by the mapping class group. The remaining normalization factor is
N0(g), formally the volume of Weyl  Di 0 (). Thus, the formal integrals over Weyl()
and Di 0() cancel N0(g):
Z Z
D Dv=N0 (g) = 1 : (3:47)
Weyl() Di 0 ()
Thus, we nally obtain an ultimate expression for amplitudes in critical string theory at
xed topology (h  2); the expression reduces to integrals over moduli of Riemann surfaces:
Z Y
Ah = dmj dm j j det(^j ; k )g^j2Z( 1)(^g) hV1 : : : VN ig^ (3:48)
Mh Mh
The cases for h = 0; 1 will be discussed in xI, J.

G. Flat Space-Time Manifold M

For at Euclidean space-time M = RD, we have already established that the central
charge of the conformal eld theory dened by maps x 2 Di(; M ) is c = D. Thus, the
at space-time string theory is a critical string theory precisely when D = 26. (We now
notice that this condition is the same as the one found for the free string: this is rather
fortunate!) Let us work out in detail under what conditions the vertex operators also satisfy
the criteria of a critical string.
To do so, it is convenient to start with a study of the Weyl transformation properties
of correlation functions of unintegrated exponents:

Y
N  Z Y
N
eiki
x(zi) g = Di(;M ) Dx e
iki
x(zi) e S0 [x;g] (3:49)
i=1 i=1
where S0 is the Gaussian action, including the constant dilaton 0: (We ignore the constant
tachyon.) Z Z
1 0
S0[x; g] = 8 dg x

(0)x + 4 dg Rg (3:50)
 
The Gaussian integral is carried out by separating the constant mode in x:
Z
x = x0 + x0 ; dg x0 = 0

III.17
and making use of the Green functions for x0 :
8 4
>
<
(0)Gg (z; w) = 4g (z; w) R dg :
> Z (3:51)
: dg Gg (z; w) = 0

The Green function at coincident points must be regularized and renormalized in a
Di()-invariant way. This may be carried out with the help of heat-kernel methods, or
simply by renormalizing by the Di()-invariant distance:
GRg (z; z)  wlim
!z(Gg (z; w) + 2 ln distg (z; w)) : (3:52)
The Weyl scaling properties of the Green function are then
8
< Ge  g^(z; w) = Gg^(z; w) + f (z) + f (w)
2

: GRe  g^(z; w) = GRg^ (z; z) + 2f (z) + 2(z) ; (3:53)

2

R
where f (z) is a function required to maintain d G = 0 under Weyl rescalings. We do
 g g
not need its explicit form here. One nds
DY E  0
 D=2 Y
eki
x(zi) =  (k )e 1
2 0 Det
R d (0)
e 21 ki
kj Gg (zi ;zj ) ; (3:54)
i g  g i;j
P
where k  ki , (k) is the D-dimensional -function, and the factor  dg under the
R
i
determinant arises from the normalization of the zero mode x0 . As a result, under g = e2 g^,
we have
Y 21 ki
kj Gg (zi ;zj ) Y 12 ki
kj Gg^(zi ;zj ) ki2 (zi )
e = e e : (3:55)
i;j i;j
All dependence on f (z) cancels in view of momentum conservation k = ki = 0, as guar-
P
i
anteed by the overall (k) factor in (3:54). The Weyl scaling property of the determinant
factor is precisely that of Z(0)(g) D=2. (Z(0)(g) contains an additional factor involving
holomorphic Abelian dierentials, but this factor is manifestly Weyl invariant.)
Combining all of the above, we nd for D = 26
DY
N E Y
N DY
N E
eiki
x(zi) g^e2
= e ki2 (zi ) e26SL (^g;) eiki
x(zi) g^ : (3:56)
i i=1 i=1
III.18
 It is now straightforward to examine the Weyl covariance conditions of vertex operators
for low mass cases.
1) Tachyon vertex operator
Z
VT (k) = "(k) dg eik
x :

In order to cancel the explicit Weyl dependence of the measure dg on :
de2 g^ = dg^e2
against the Weyl dependence of the correlation function, we recover the familiar condition
k2 = 2 for the tachyon. (This condition also corresponded to a = 1 in xIII.)
2) The massless vertex operators
Z
V"(k) = " (k) dg gmn@m x @nx eik
x ;

the Weyl dependence of the part dg gmn is trivial, so we need the familiar condition
k2 = 0. In addition, Weyl dependence appears through the contraction of @m x with the
exponential, and is cancelled by the transversality conditions
k" (k) = k " (k) = 0 :
Finally, from the contraction of @m x and @n x , (actually only the trace-part contributes),
See Prob. a curvature Rg term appears, and Weyl invariance for the trace-part is achieved through
Set #5.
the addition of a (classically non-Weyl invariant) part involving the curvature:
Z Z
V" (k) = " (k) dg gmn@m x @nx eikx + " dg Rg eik
x :
 
The physical particle corresponding to this vertex operator is the dilaton. Recall that the
dilaton eld  entered the generalized sigma model action with a Gaussian curvature term
Rg as well.
3) Higher mass states: the general structure of vertex operators is given by
Z
V"(k) = "1 :::2p dg gm1n1 : : : gmpnp @m1 x1 : : : @np x2p eik
x

III.19
plus terms involving higher derivatives on x. To cancel the overall exponential Weyl depen-
dence, we must require k2 = 2p +2, yielding the same mass spectrum as in the free string
case. In complete analogy with the case 2), there will be further conditions of transversality
on " and and further terms involving Rg will have to be added to render the trace parts
Weyl invariant.

H. Non-Critical Strings
Conformal matter eld theories with c 6= 26 are called non-critical string theories. The
amplitudes for these theories retain a non-trivial Weyl dependence through the Liouville
action, and through the vertex operators. Its direct quantization requires understanding
the dynamics of Liouville theory. Despite much eort, there is no complete picture to date
of how this can be achieved consistently, and without the presence of a tachyon.
A tremendous number of results was obtained indirectly (by matrix model techniques)
when c is less than 1 and rational.

I. Tree Level Amplitudes

Tree level amplitudes correspond to the worldsheet topology of a sphere S 2, which we
stereographically project onto the complex plane C . The moduli space M0 of the sphere
consists of just a single point, which we may choose to represent by the round metric, of
curvature Rg^ = 1, and area 4:

g^ = (1 4+jdzjzjj2)2
2
z2C : (3:57)

The group of conformal automorphisms is PSL(2; C ) and acts by Mobius transformations

az + b a b 
 : z !  (z) = cz + d c d 2 PSL(2; C ) :
A convenient basis for the associated vector elds (called conformal Killing vectors) is
z=1 z=z z = z2 ;
1 2 3

and similarly for az, a = 1; 2; 3.

III.20
 To obtain the general expression for tree level transition amplitudes, we must rst
complete the treatments of conformal automorphisms, which we had postponed in xF, and
of Weyl transformation of the measure. The starting point is our expression (3.28) for the
measure Dg, considered now for h = 0

e ( 1)(^g)e
Dg = D DvZ 26SL (^g;) (3:58)

e = D0 v det( a ; b)g , a 2 Kerr(z 1), and the measure D0 v is over Di 0 () vector
where Dv
elds orthogonal to Kerr(z 1)  Kerr(1)
z . Tree level amplitudes are then given by
Z Z
A0 = D e N(g)Z( 1)(^g) hV1 : : : VN ig^
Dv= (3:59)
Weyl() Di 0 ()=PSL(2;C )

using the same notations as in III.19.

While the group PSL(2; C ) leaves the metric g invariant (up to Weyl rescaling), it does
transform the positions of vertex operator insertions non-trivially. It is natural and | as
we shall establish later on | required for unitarity of the transition amplitudes, to divide
by the (formal) volume of the full Di 0() group, and not just of Di 0()=PSL(2; C ). To
achieve this, we proceed as follows.

1) For N  3 cases
Let the last three vertex operators be given by (of course, we may choose any three
vertex operators)
Z
Vi = d2 zi Wi(zi ) ; i = N 2; N 1; N :
C

We propose to make a change of variables from the vertex insertion points zi, zi i =
N 2; N 1; N to the group elements of PSL(2; C ), thus completing the measure D0 v
into Dv. This is possible since, given any three points zi 2 C , there is a unique element
 2 PSL(2; C ), such that  maps zi0 onto any triplet zi 2 C , with

zi =  (zi0) :
III.21
to carry this out in detail, let az be a basis of the Kerr(z 1), a = 1; 2; 3, chosen to be Weyl
invariant. We parametrize general vector elds v in psl(2; C ) by complex parameters a,
a with
X 3 X
3
vz = a az vz = a az :
a=1 a=1
The measure on PSL(2; C ) vector elds is then
Y3
d(6) = Dv=D0 v =
v da da det( a ; b )g
a=1
where the determinant det( a ; b)g arises as the Jacobian of the coordinate change from v
to . On the other hand, from the action of PSL(2; C ) on the points zi0 : zi =  (zi0 ) = ev zi0 ,
P
we have dzi = da a(zi ), so that the measure on d2zi becomes:
a
Y
N Y3
d2z i= da da j det a(zi )j2 :
i=N 2 a=1
Clearly, all factors in this identity are Weyl-invariant. Combining all the results obtained,
we have
eDv Y d2zi = D0 v det( a ; b)g Y da da j det a (zi )j2
N 3

i=N 2 a=1 (3:60)

= Dv j det a (Zi )j2
We may now divide by the formal volume of Weyl() n Di() just as in the cases h  2,
and we obtain the nal expression for tree level amplitudes with N  3:

V : : : V Y
N 
A0 = z( 1)(^g)j det a (zi )j2 1 N 3 W (Zi ) g^ ; (3:61)
i=N 2
valid for any 3-points zi . We shall evaluate some of these amplitudes momentarily for at
space-time.

2) N = 0, 1, 2 cases
Clearly, here, no three vertex operators are available to complete the integration D0 v
into Dv. In fact, these amplitudes vanish! To justify this, notice that D0 v can at most be
III.22
completed by the vector elds in PSL(2; C )=SN where SN is the stabilizer of N points on
the sphere. Thus, we are left to divide out by the volume of SN
Z Z
D D0 v=N(g)  Vol(1S ) = 0 : (3:62)
Weyl() Di 0 =(PSL(2;C )=SN ) N
For N = 2, x the two points at 0 and 1, and S2 is the group of dilations, whose volume
is 1.
An alternative argument that the amplitudes should vanish is that no conformal in-
variant amplitudes can be constructed, except with value 0. The physical interpretations
of these vanishing amplitudes are as follows:

1) N =0 tree-level vacuum energy is zero

\vanishing of the cosmological constant".
2) N =1 the \tadpole" amplitudes vanish,
i.e. conformal eld theory provides solutions
(no linear terms) to string theory, at tree level.
3) N =2 no mass corrections to tree level.

3) Green function and determinants

The scalar Green function on the sphere, with metric g^, is given by
02
G(z; z0 ) = ln (1 + jzjzj2 )(1z +j jz0 j2) : (3:63)
The determinants involving
(0) and
( 1) are just constants. They may be computed
explicitly, but we shall not do so here. Instead, we abbreviate
0 0 1 13
Det

g) @ R d (0) A
z  Z( 1) (^ : (3:64)
g^

Furthermore, the only remaining nite-dimensional determinant is
01 1 1
1
j det a(zi )j2 = j det @ zN 2 zN 1 zN A j2
zN2 2 zN2 1 zN2 (3:65)
= j(zN 2 zN 1)(zN 1 zN )(zN zN 2)j2 :
III.23
4) Tachyon Amplitudes
The tachyon vertex operator is particularly simple:
Z p
VT (k ) = d2z WT (z; k) WT (z; k) = " g^ eik
x : (3:66)

Here, " is a constant, independent of k, and plays the same r^ole as the eld normalization
Z -factors in the Lehmann Symanzik Zcmmerman (LSZ) formalism in quantum eld theory:
hkj'(x)j0i = Z eik
x for a scalar eld '(x) and its associated 1-particle state jki momentum
k, and j0i the ground state.
We now make use of the results in p. II.23 on the correlation function of exponential
operators. We introduce the renormalized "R by absorbing a multiplicative renormalization
factor of the exponential
"R = " e GR (zi;zi )g^ (3:67)
The amplitude now takes the form
NY3 Z
P ki
kj G(zi ;zj )
20 "N Y
N p 1
2
A0 =  ( k ) e R z d2zi g^(zi ) e i6=j
j det( a (zi ))j2 : (3:68)
i=1 C i=1
The contributions to G of the form ln(1+ jzij2) sum up and, using momentum conservation
p
and ki2, precisely cancel the g^ prefactors. We are left with
NY3 Y
N
A0(k1 : : : kN ) = (k)e 20 "N
R z j det a (zi )j2 d2 zi jzi zj j2ki
kj (3:69)
i=1 i<j
Now, zN 2, zN 1, zN were any 3 points on the space, so we may let them equal 0, 1,
1, respectively, and we nd
NY3 Z NY1
A0(k1 ; : : : ; kN ) = (k)e 20 "N
R z d2 z i jzi zj j2ki
kj : (3:70)
i=1 C i<j
It is very instructive to examine amplitudes for small N :
(1) three point function
A0 (k1 ; k2; k3 ) = z e 20 "3  (k ) : (3:71)
R
III.24
This 3-point function provides the 3-tachyon on shell coupling.
(2) four point function
Z
A0 (k1; : : : ; k4) = z e 20 "4  (k )
R d2z jzj2k1
k2 j1 zj2k1
k3 : (3:72)
C

It is standard to introduce the 3 Lorentz invariants s, t, u that characterize the kinematics

of the 4-point function:
t
k1 k4

s 8
< s  (k1 + k2)22
: ut  ((kk21 ++kk33))2 : (3:73)
k2 k3

(Note that s + t + u = 8.) In terms of s, t, u the four point function is

Z
A0 (k1; : : : ; ku) = z e 20 "4  (k ) d2zjzj s 4 jz 1j u 4 : (3:74)
R
C

This integral expression is absolutely convergent in the following region D, dened by

8 Re(s) < 8 Re(s) < 2
< 2 <
Re(u) <
: Re( 2 or : Re(t) < 2 (3:75)
t) < 2 Re(s + t) > 6:
Taking just the real parts:
Re(t)

-6 -4 -2
Re(s) ( IR2 )
+

-2

D -4

-6

III.25
Let s; t 2 D, then we may evaluate the amplitude by elementary methods, and we nd the
Virasoro-Shapiro amplitude:

A0 (s; t) = ze 20 "4  (k ) ( 1 s=2) ( 1 t=2) ( 1 u=2) : (3:76)

R (2 + s=2)( (2 + t=2) (2 + u=2)
Since we know the analytic continuation of (z) throughout C , with poles only at z 2 N,
we automatically have an analytic continuation of the full amplitude A0 (s; t), throughout
(s; t) 2 C  C . As a byproduct, we also immediately have the amplitude in Minkowski
space-time momenta, by letting
0
kE

0
kM
ki0 ! iki0
k 2+m2 -i ε

( s ! s + i
t ! t + i (k) ! i(k)
u ! u + i
where it is understood that  > 0 and that we only consider the limit in which  ! 0.
Thus, the 4-point amplitude in Minkowski space-time is given by

A0 (s; t) = ze 20 4  (k )i ( 1 2=2 i) ( 1 t=2 i) ( 1 u=2 i) : (3:77)
R (2 + s=2 + i) (2 + t=2 + i) (2 + u=2 + i)
The amplitude has only pole singularities | as expected at tree level | and their location
is easily identied. The -function contributions from the denominator are entire. From
the numerator, we have poles at
8s = 2 + 2ns ns 2 N
>
>
<
> t= 22nt nt 2 N (3:78)
>
: u = 2 + 2nu nu 2 N :
These poles precisely correspond to the creation of intermediate string states whose mass
III.26
is the above value for s, t or u. The states may occur either in the s, t or u channel :

k1 t k4 k1
k4
k1 k4

s u

k2
k
3 k2 k k2
3 k3

It is instructive to identify where these singularities arise from in the integral representation
for the amplitude:
z

s poles: z !0 s u t

t poles: z!1
u poles: z ! 1: 0 1 οο

The fact that three QFT diagrams arise from a single string diagram is the simplest
example of a general phenomenon mentioned in the introduction.
Let us concentrate on the s = 2 pole, which corresponds to a tachyon intermediate
state. The diagram there factorizes into a propagator, and 2 three point functions for
external tachyon:

1
s ~ -2 ~
s+2-i ε

Insisting upon this factorization, we obtain a non-linear equation between the 3-point
function and 4-point function, expressing factorization | or unitarity | of the transition
amplitudes (or so-called S -matrix elements). This relation xes "R :

"2R = 2z e20 : (3:79)

III.27
J. One loop amplitudes
At one loop, the worksheet has the topology of a torus, and there is a single complex
modulus  2 C . We represent  by

 = C =(Z+ Z ) Im  > 0

with the at metric g^ = 2jdzj2. Teichmuller space is T1 = f 2 C ; Im  > 0g, the mapping
class group is G = SL(2; Z), acting on  as PSL(2; Z) by
a b 
 !  0 = a + b
c + d c d 2 SL(2; Z): (3:80)

Moduli space is conveniently taken as T1=PSL(2; Z) = M1, but we shall have to recall that
the full mapping class group also divides out by fI g in SL(2; Z). It is standard to take
the fundamental domain
 1

M1 =  = 1 + i2; 1;2 2 R; 2 > 0; j1j  2 ; j j  1 : (3:81)

Determinants of Laplace operators on tensors of any integer weight are given by the
case of Laplacians on functions, which was computed by Gawedzki:
Det0
(0) = 22 j( )j4
Y
1 (3:82)
( )  ei=12 (1 e2in ) :
n=1
The Green function on the scalar eld x is given by
 # (z wj ) 2  (z w z + w)2 ;
G(z; wj ) = ln  1#0 (0j )  22 (3:83)
1
where #1 (zj ) is the Jacobi theta function that has a zero at z = 0.
The N -string transition amplitude is given by the following expression
Z d2  1 1  V ;:::;V 
A1(k1 ; : : : ; kN ) = N (3:84)
M1 22 (4 2) j( )j48 1
2 2 12

where  V1 : : : VN  stands for the normalized correlation function of the vertex operators
V1 : : : VN :  V1 : : : VN = hV1 : : : VN i = h1i.
III.28
 For example, considering only tachyon vertex operators, we have
N Z
0 1
 V1 : : : VN = (k)"R NY X
d2zi exp @ 21 ki
kj G(zi ; zj j )A (3:85)
i=1 i6=j

There is no dependence on the dilaton constant 0 here, because the Euler number on the
torus vanishes. Other correlation functions may be similarly derived.
To one loop order, we may now also have non-zero N = 0; 1; 2 point functions. Let us
consider N = 0 rst:  1 = 1.
Z d2  1 1 :
A1 (
) = Vol(M ) 222 (42 2)12 j( )j48 (3:86)
M1
Using the asymptotics for Im  ! 1,

( )  ei=12 = ei1 =12 e 2 =12 (3:87)

we see immediately that this amplitude diverges for large 2. In fact, this ( ) factor is
common to all amplitudes, for any type of vertex operators, and produces a divergence in
any amplitude, for any arrangement of external momenta. We shall not show this here in
general, but leave this point to be checked by the reader. Thus, bosonic closed oriented
string theory is divergent in at space-time.

Nature of Divergence
Let us try to gain additional insight into the physical origin of this divergence. To do
so, it is helpful to exhibit the dependence on internal momenta by inserting a factor:
Z  1 I 
1= d26k  k 2i dz @z x ; (3:88)
A
where A is the cycle in  whose pre-image in C runs from z = 0 to z = 1. We nd
Z Z d2  1 1 e 22 k2 :
A1 (
) = Vol(M ) d26k (3:89)
R 26
M1 22 (4 ) j( )j48
2 12

III.29
 τ
²

M
¹

k
¹

τ
¹
-½ ½

Substituting the asymptotics of the ( )-function

Z Z d2  1 e 22 (k2
A1 (
) = Vol(M ) d26k 2) (1 + #e 22 +


) ; (3:90)
R 26
M1 22 (42 )12

it is now immediately apparent why a divergence occurs, and where it occurs. For 2 ! 1,
the momentum range 0  k2 < 2 contributes an exponentially growing factor. In fact,
integrating out 2 for large 2:
Z d2 1
22 (k2 2) 
2 e k2 2

This contribution arises from the tachyon! It is an I.R. problem related to the stability of
at space-time.
It is equally instructive to see where the integral is convergent: at large k2, since
the range of 2 is cut o from below by the choice of M2 , the integral is exponentially
suppressed for large momenta k2. Thus, this loop amplitude is superconvergent at high
energy, in contrast with eld theory.
In fact, we can use this simple example to exhibit the distinction between what this
partition function would look like in quantum eld theory and what it looks like in string
theory. In eld theory, we may imagine the partition function for the same states, given
III.30
by the same generating function. The only dierence is the region M1 :

τ²

Integration region for QFT

τ
-½
0 ¹ with the same states as string theory.
½

Now, since 2 reaches down to 2 = 0, the superconvergence disappears and the integrand
is power behaved in momenta. Thus, we see here that a crucial dierence between string
theory and quantum eld theory arises from modular invariance, producing ultraviolet
nite answers in the case of strings

III.31