Lecture 14
IX. Superstring Perturbation Theory


Eric D'Hoker We shall now develop a systematic treatment of superstring perturbation theory in the RNS formulation. Local N = 1 supersymmetry (SUSY) on the worldsheet plays a fundamental r^le, but this symmetry is not manifest in the component formalism, set up in o xVII. The N = 1 super eld formalism, formulated in the language of N = 1 supergeometry, treats di eomorphisms and local supersymmetry transformations on an equal footing, and renders both manifest. Their combined action makes up the group of super-di eomorphisms of the worldsheet super-manifold, and local supersymmetry invariants can be constructed from super-di eomorphism tensors. We rst present the di erent ingredients needed and then apply these methods to the RNS superstrings. A general reference for this subject is E. D'Hhoker and D. H. Phong, \The Geometry of String Perturbation Theory", Rev. Mod. Phys. 60 (1988), p. 917.

( m ; ), m = 1; 2, = 1; 2 with the grading

A) N = 1 supergeometry (also called N = 1, D = 2 supergravity) We consider a supermanifold of dimension (2j2), and use local coordinates
M N = ( )MN N M



where ( )MN equals 1 unless M = , N = , in which case it equals 1. We introduce a local orthonormal frame

E A d M EM A

A = (a; ) a = z; z ; = +; ;


where it is understood that a has the same grading as m, while has the same grading as . We denote the inverse frame by EAM , and we have EAM EM B = A B and EM A EA N = MN. IX.1

Next, we consider frame rotations by a group of U (1) gauge transformations, under 1 which the elds E z , E z , E +, E have weights 1, +1, 2, +1, respectively. A general 2 z ) n . We introduce a U (1) gauge super eld V has U (1) weight n if it transforms as (E eld or connection =d M M (9:A:3) with associated covariant derivatives
D(n) V A

EA M (@M + in

M )V



Here @M @=@ M and i = 1 . Torsion TAB C and curvature RAB super eld tensors are de ned by the graded commutators
DA ; DB ]V = TAB C DC V + in RAB V




The data of N = 1 supergeometry is the frame E A and the U (1) connection , together with the following constraints on the torsion and curvature, arising in D ; D ]:

T = 0; T c = 2

c ;

R++ = R



Here, c are the 2-dimensional Dirac matrices, satisfying f a; bg = ab, with a convez nient representation given by ++ = z = 1, all other components = 0. The constraints of (9.A.6) are equivalent to the following set of constraints

fD+; D gV = in R+ V ;

D2 = Dz ; D2 = Dz +


where V is any super eld of weight n. The structure relations of the remaining covariant derivatives are de ned by using the (graded) Jacobi identities. The only free quantity in this supergeometry is the curvature super eld R+ = R + R. The symmetry transformations on the data E A and that preserve the constraints (9.A.6) or equivalently (9.A.7) are IX.2

1) sDi ( ):

super-di eomorphisms M ! 0 M ( N ) contains Di ( ), local N = 1 susy, and further algebraic transformatons.

2) a) sWeyl( ): generated by a super eld , contains Weyl( ) and further algebraic transformations. b) sU (1): generated by a super eld L, contains U (1) gauge transformations and further algebraic transformations.

It is convenient to combine transformations a) and b) into complexi ed Weyl transformations, given in terms of a complex super eld , de ned by


+ iL :


The superderivatives transform under sWeyl sU (1) as
1 b+ D(n) = en 2 D(n)e n + 1 b D(n) = e n 2 D(n) en


and the curvature transforms as

R+ = e

b R+

b b 2iD+ D



From (9.A.9), the transformation rules for E and may be deduced, and from these combined with the torsion constraints (9.A.6), the transformation laws of Ez , Ez , z and z follow. N = 1 supergeometry admits an almost complex structure

JM N = EM a abEbN + EM




where 2 = 2 = 1. In fact, J is a complex structure provided the constraints given in (9.A.6) are satis ed. (This may be viewed as one of the motivations for choosing those p constraints.) We shall denote local complex coordinates by ( ; ; ; ) with = 1= 2 ( 1 + p i 2 ), = 1= 2 ( 1 + i 2 ), and and their complex conjugates. IX.3

@ @ Dz = D2 = + @ @ (9:A:12) @ + @ 2 = @ : D = Dz = D @ @ @ The residual symmetry group of sDi sU (1) sWeyl that leaves this supergeometry invariant is the group of superconformal transformations, which consists of superanalytic di eomorphisms, ! 0 ( ; ), ! 0 ( ; ), ! 0 ( ; ); ! 0 ( ; ), combined with sWeyl and sU (1) transformations. Wess-Zumino gauge is obtained by xing a gauge for the action of the algebraic parts of sDi ( ), sU (1) and sWeyl. The remaining independent elds are em a , m and an auxiliary eld A, and we have Em a = em a + a m 1 em aA 2 i 1 Em = 2 m 2 ( a) em aA + E a = ( a) i E = 1+ 4 A

By a combined sDi , sWeyl try may be mapped on a locally superderivatives @ + D+ = @

and sU (1) transformation, any N = 1 supergeomeat (Euclidean) supergeometry, characterized by the


This presentation of the supergeometry is so-called in components. Integration over is performed with the help of a super-measure d E , de ned as follows: d E s det EM Ad d d d (9:A:14) where the superdeterminant (also sometimes referred to as the Berezinian) is de ned by

s det EM A det(Em a EM (E ) 1 E a) (det E )
Of course, d E is sDi invariant. Inner products on super elds de ned by Z h 1 ; 2i = d E 1 2 : IX.4

1 2

(9:A:15) of weight n are (9:A:16)


expression in super eld form. We introduce a string coordinate super eld of sU (1) weight 0; X = ! R10

B) Functional Integral Representation of Transition Amplitudes The RNS worldsheet action, introduced in xVII in components, admits a simple
X =x +
+ + F = 0; 1; : : : ; 9 : (9:B:1)


The action

1 Z d D X D X (9:B:2) SX = 4 E + reduces to the RNS action, except for the fact that it has an entire term of the form F 2. This term is non-dynamical and the F eld may be integrated out without any consequences. The next action we need is that of the super (B; C ) system: let B and C be super elds of sU (1) weight n and 1 n, respectively. The (B; C ) action is then 2 1 Z d (BD(n) C + BD( n)C ) : (9:B:4) SBC = 2 E + To make contact with the (b; c) action, we decompose the super elds B, C in components B = + b + b0 + 0 (9:B:5) C = c + + 0 + c0 The elds b0 , 0 , 0 and c0 are auxiliary, just as F was an auxiliary eld in SX , and we shall drop their contribution to SBC . The U (1) weight assignments in Wess-Zumino gauge 1 are now obtained by using the fact that (resp. ) have weight 2 (resp. 1). Hence, the 2 weights are, for n = 1: eld

3=2 2 1=2

1 (9:B:6)

U (1) weight

In locally at coordinates, this action equals in components Z 1 d2 z (@ x @ x SX = @z F 2) (9:B:3) z z + @z + 4 so that the sign of the -kinetic term is opposite to the convention we used previously. Notice that this reverses the sign of the stress tensor. Apologies! (Conventions are to be uniformized in the nal write-up.)


We recognize the (b; c) system of anti-commuting ghost elds, and we now also nd a ( ; ) system of commuting ghost elds (called superghost) which has precisely the correct weight assignments to be the ghost system for local supersymmetry. The ( ) elds are worldsheet spinors, and their spin structure must be the same as that of + , in order for ( ) to be N = 1 local supersymmetry ghosts. The BC action may be worked out in components and becomes in WZ gauge: 1 Z d2z brz c + rz z SBC = 2 (9:B:7) z + S+ + c:c:] ( 1) ( 1=2) where the BC supercurrent is
1 Sz+ +2 b 3 2

rz c rz c


This current is the N = 1 superpartner of the stress tensor and it is conserved; in fact it is analytic. The combined X , (B; C ) system for D = 10 components of X is superconformal invariant, and so the super-conformal eld theory associated with this system depends upon N = 1 supergeometry only through its supermoduli. Recall that supermoduli space is de ned by A sMh fEM ; M ssatisfying constraints of (9.A.6)g (9:B:9) Di sWeyl sU (1)

and is a space of dimension 8 (0j0) h=0 > > (1j0) < h = 1 even spin structure (9:B:10) dim sMh = > (1j1) h = 1 odd spin structure > : (3h 3j2h 2) n 2 The holomorphic cotangent space to sMh is spanned by super-holomorphic 3=2 di erentials K , which satisfy K odd graded D(3=2 K = 0 (9:B:11) K = 1; : : : ; dim sM Super-Beltrami di erentials are de ned as the duals to K , and may be parametrized by the super-moduli variations of EM A : @EM z K = 1; : : : ; dim sM : (9:B:12) h K E M @m K IX.6

It is instructive to work out the K in Wess-Zumino gauge:

@enz ez n @m


@ z+ : @mK


The rst term on the r.h.s. is recognized as the ordinary Beltrami di erential, for mK an even modulus, while the second term corresponds to a novel odd Beltrami di erential for mK an odd modulus. We are now ready to state a preliminary De nition of transition amplitudes for RNS superstring amplitudes. This preliminary de nition is the natural generalization of the \summation over surfaces" de nition adopted for bosonic string amplitudes in xI. It is preliminary in the sense that it treats left and right chirality worldsheet spinors with the same spin structure. In the nal de nition of transition amplitudes, these spin structures will have to be summed over independently for left and right movers, and this will be presented in xIX.F, after the Chiral Splitting Theorem has been established. The preliminary de nition will be useful, however, in order to introduce superghosts, and it is needed in order to set up the Chiral Splitting Theorem. Transition amplitudes are de ned by Z Z A D M (constraints) DX V1 : : : VN e SX Ah = DEM (9:B:14) for certain vertex operators Vi , to be speci ed shortly. In the critical dimension, D = 10, and for super-conformally invariant vertex operators Vi , we have the following Theorem

Ah =


sMh K




DX u



D(BB ) D(C C )V1 : : : VN e ;B )



(h ; B i ) (


The proof of this result is much more involved than in the case or bosonic strings, because the inner products h 1; 2i are no longer positive on super elds, and extra care must be used in dealing with the D+ and D operators. We shall not prove it here; we refer the reader to Rev. Mod. Phys. 60 (1988) p. 917. IX.7

A conformal eld theory with an additional N = 1 local supersymmetry invariance is automatically invariant under superconformal transformations, and is an N = 1 superconformal eld theory. Each conformal eld has a superpartner (which is also a conformal eld), and together both elds t into a superconformal eld; X , B, C are examples of such elds. Conserved currents are other examples of conformal elds that generalize to supercurrents. Conservation of a supercurrent J is given by

C) Superconformal Field Theory (some basics)

J =0


In local super-complex coordinates, this means that

J = J (z; ) = J1=2 (z) + J1(z)
with J1=2 and J1 holomorphic in z. The associated charge is de ned by


QJ = dz d J = dz J1(z)




Important examples of supercurrents include the stress tensor, which is built up out of the ordinary stress tensor Tzz and the (worldsheet) supercurrent Sz+:

T (z; ) Sz+(z) + Tzz (z) :
For the X -system, T is given by


T (X ) = 1D+ X D2 X + 2 = 1 + @z x + 2

1@ x 2 z

@z x

1 2 @z +



We recognize indeed the super-current and the x stress tensor. For the (B; C ) system, T is given by 1 3 T (BC) = C D2 B + 2 D+ C D+B 2 D2 CB + + (9:C:6) 1 b 3 @ c + (T (bc) + T ( )) = c@z + 2 2 z IX.8

The OPE of T with super elds of weight (h; 0) is determined by Ward identities, and we have the asymptotic expansions

T (z; ) (w; ) n (z w


1 1 (w; ) + 2 z w +z w

D+ (w; ) D2 +

(w; )


For example D+X , B and C have weights n = 1=2; 3=2; 1 respectively. The super-Virasoro algebra is similarly expressed. Denote s z w

T (z; )T (w; )

1^ 4c + 3 s3 2


1 T (w; ) + 2 1 D+T (w; ) + s D2 T (w; ) + s


where c = 3 c represents the central charge. ^ 2 The combined action SX + SBC is BRST invariant. For completeness, we record the BRST transformations on the elds

D) BRST Quantization

X = C D2 X 1 D+C D+X + complex conjugate + 2 1 C = C D2 C 4 D+ C D+C + B= T


where is a constant Grassmann-valued parameter and T is the stress tensor super eld, de ned in C). The associated BRST supercurrent is 3 JBRST = C (T (X ) + 1T (BC)) 4 D+(C D+CB) 2 The BRST charge (9:D:2) (9:D:3)



is conserved and nilpotent precisely in D = 10, in analogy with the bosonic string. Vertex operators for physical states commute with QBRST , while vertex operators of spurious states are QBRST -commutators of some operator. We shall not carry out BRST quantization any further here. IX.9

We distinguish two kinds of vertex operators: those for the NS sector, creating spacetime bosons, and those for the R sector, creating space-time fermions (also sometimes called the fermion vertex operator). For closed strings, this produces four di erent kinds of vertex operators: NS-NS and R-R creating bosons, and NS-R and R-NS creating fermions. The Rvertex cannot be constructed directly out of the elds X , since it must produce a branch cut starting at its insertion points. For simplicity, we specialize to the closed RNS string, and consider NS-NS vertex operators rst; the left-moving part may be constructed using the chiral splitting theorem for the RNS string, to be derived in the next section, xF. We begin by constructing RNS vertex operators for NS-NS states without ghost dependence. Vertex operators in the NS-NS sector are given in a fashion completely analogous to the bosonic string

E) Vertex operators for physical states

(1) The NS vertex operator

V ("; k) =


d E Pn("; D+ X; D X; D2 X; : : : )eik X +


Here, Pn is a polynomial involving a sum over terms, each of which has a total number n of D+ derivatives. By sU (1) invariance of V , Pn must then also have a total number n of D derivatives. (On non- at N = 1 supergeometries, the curvature super eld R+ may also enter, just as the Gaussian curvature Rg entered for the bosonic string in the dilaton vertex. R+ e ectively counts for one D+ and for one D .) By construction, V ("; k) is invariant under sDi transformations. It remains to guarantee that V ("; k) is sWeyl invariant. Under constant sWeyl transformations, D+ and D have weight 1=2, d E has weight 1 and the exponential has weight k2. Given that #D+ = nD = #D , we have

M 2 = k2 = nD 1


The GSO projection forces n to be odd only, since the number of NS elds applied to the ground state in each sector (left and right moving) must be odd. Thus, we have nD = 2n+1, IX.10

with n = 0; 1; 2; : : : and the masses of NS-NS states are given by

M 2 = 2n

n = 0; 1; 2; : : :


Non-constant sWeyl transformations will require in addition that certain conditions be ful lled on the polarization tensors ", such as transversality. The massless string NS-NS vertex operator is

V ("; k) = " (k)


d E D+X D X eik X


sWeyl invariance requires " (k)k = " (k)k = 0, and we recover the vertex operator for dilaton, graviton and B elds. It is useful to work out this vertex operator in components, dropping the auxiliary eld F , and other terms that vanish by the eld equations, such as @z + . (The full justi cation for dropping terms that vanish by the eld equations may be given with the help of the Chiral Splitting Theorem in the next section, xF.) Restricting to left-movers only, we use the fact that X = XL + XR
and obtain the left-moving NS part of the vertex operator (9.E.?) as follows: (9:E:5)


d D+XL eik XL = (@z xL i

+ +k

)eik xL


This vertex contains the bosonic vertex operator, as well as an operator that allows for an application of an even number of NS fermions only.

(2) The R vertex operator

Ramond states transform under spinor representations of the (double cover of the) Lorentz group, Spin(1; 9). This group has two inequivalent Majorana-Weyl spinors S and S 0 , each of dimension 16 so that we let = 1; : : : ; 16. To describe the representations S and S 0 , we introduce the Cli ord algebra of Spin(1; 9)

f ; g = 2 I(32)


where I(32) is the identity matrix of dimension 32. The chirality matrix is de ned by
0 1 9

and satis es


= I(32) and f ; g = 0


Weyl spinors of left (resp. right) chirality are eigenvectors of with eigenvalues +1 (resp. 1). The charge conjugation matrix C is de ned by the relation ( )T = C C and anti-commutes with :

(9:E:9) (9:E:10)

f ; Cg = 0 :

The charge conjugate c of is de ned by



= y



A Majorana spinor is a spinor whose charge conjugate is equal to itself, c = . It is consistent (and convenient) to choose C = 0 , so that a Majorana spinor is real: c = = . We shall choose a basis in which 1 0 0 I 1 0I 0 (16) (16) A A ; C = 0=@ (9:E:12) =@ I(16) 0 0 I(16) A 32 component Majorana spinor now decomposes into two 16 component MajoranaWeyl spinors S and S 0 . It will be convenient to work with -matrices in this basis: 0 0 ( ) 1 0S 1 A (9:E:13) =@ =@ A 0 ( ) 0 S with ; = 1; : : : ; 16. It then follows from (9.E.9), (9.E.12) that = and = . The generator of Spin(1; 9) transformations in this representation is 1 , with 2
1 2

; ]= ; ]:

1 2




Clearly, commutes with , and 1 are the representation generators of Spin(1; 9) in 2 one of the Weyl representations. The SO(1; 9) transformation properties in the Ramond sector are governed by the SO(1; 9) currents (we omit the contribution from the x- eld)

j (z ) =
Clearly, the eld

+ + (z )


itself transforms under the vector representation of SO(1; 9):

j (z ) + (w ) z 1 w (

+ (w )

+ (w))



We seek to construct elds S (z) and S 0 (z) such that transform as spinors under Spin(1; 9) with representation matrices :

j (z )S (w ) z 1 w j (z )S 0 (w ) z 1 w

1 2 1 2

( (

) S (w ) ) S 0 (w ) :

(9:E:17a) (9:E:17b) OPE of

From these OPE's, and from the fact that j is bilinear in + with S :
+ (z )S + (z )S

+ , we may infer the

(w )

1 p1 ( ) S 0 (w) 1 (z w ) 2 2 1 1 ( ) S (w ) : 1 p (z w ) 2 2

(9:E:18a) (9:E:18b)

0 (w )

The relative normalization between S and S 0 is arbitrary, and has been xed above on symmetry grounds. The conformal weights of S (z) and S 0 (z) are computed from their { OPE with the stress tensor T ( ) = 1 @z + + : 2

T ( ) (z )S (w )

1 2

1 2

p1 (z 1 )3=2 w 2

+ (z )S

0 (w )

1( 16

) S (w) (z w)2 : IX.13


To obtain the spin elds S and S 0 , we make use rst of the action of the currents j in the Cartan subalgebra of Spin(1; 9) only. These components mutually commute and may be bosonized in terms of 5 independent free scalar Bose elds:

Since 5=8.

= 10 I(16), we nd that the conformal weight of S (and similarly of S 0 ) is

j 2a;2a+1(z) = i@z a(z)
The R- elds

a = 0; 1; 2; 3; 4 :


are exponentials of a with 1 weights:

p1 (

2a +


2a+1 i + )=e 1 weights: 2




The spin elds are exponentials of a with

S (z) = ei



where we identify the space-time indices with the spinor representation weights, given by 1 1 = ( 1; 2; 1; 2; 1). The spinors S are obtained for with an even number of 2 2 2 0 is obtained with an odd number of signs. The conformal dimension of signs, while S 1 this eld is easily veri ed to be 2 2 = 5=8, with the help of the standard formula for the conformal weight of an exponential of a c = 1 conformal eld. One may complete the analysis by showing that the currents j corresponding to roots may also be constructed in terms of exponentials (multiplied by certain cocycle factors) and that j indeed satis es the correct OPE with S and S 0 . With the above normalization, one also obtains the OPE's of S with S and with S 0:

S 0 (z )S (w ) S (z )S (w )


1 (z w)3=4



1 1 { (z w)1=4 2

j (w )


(Notice that the construction of spinor elds that we have just given for Spin(1; 9) may be carried over in a straightforward way to Spin(2n). In particular, this type of construction is needed to obtain the spinor of SO(16) necessary to build the full current algebra of IX.14

E8 E8 in the heterotic string. There, the spin elds have precisely conformal dimension (1; 0), since the weight vectors all have length 2 = 2.) It remains to construct the Ramond vertex operator. The spin eld by itself cannot produce a vertex operator for massless fermions, since its conformal weight is 5=8 6= 1. The resolution of this issue is found in a contribution to the R vertex operator from the ( ; ) and possibly (b; c) ghost system. The relevant conformal primary eld is obtained as follows. (We consider only the left-moving part of the vertex operator.) W (z; u; k) = u (k)O(z)S (z)eik xL (z)
(9:E:24) Here, u is a spinor analogue of the polarization tensors " that entered the construction of NS states and vertex operators. The operator O(z) depends on the and ghosts, and is a primary eld of weight 3=8, so that W is of weight 1 for massless states (with k2 = 0). O(z ) is constructed in such a way that W is BRST invariant; it is an analogue of the spin eld, but now for the superghost system. The ( ; ) ghost system may be represented in terms of two free boson elds and , such that =e = e + @z (9:E:25) with (z ) (w ) ln(z w) c = 13 (9:E:26)

(z ) (w ) ln(z w) c = 2 The ghost number current is jz = @z . The operator O(z) is again an exponential with weight 1 : 2 1 O(z ) = e 2 (z) : (9:E:27) Because the central charge of the system is 13, the dimension of O(z) is 3=8, instead of 1=8, as would be the case for a eld of central charge 1. The above construction of W implies, though, that W has ghost number 1=2! But overall ghost number is given in terms of the index theorem, and this would suggest that the number of Ramond vertex operators that can be inserted on any Riemann surface is determined in terms of the topology of the surface, which is clearly absurd. IX.15

The resolution of this problem lies in the subtleties of the quantization of the ( ; ) system, which we shall not discuss here. The nal result is that the quantization of the ( ; ) system yields vertex operators in di erent possible pictures, labeled by their ghost number. One shows, for example, that there is an equivalenT vertex operator for W , 1 which had ghost number 1=2. This operator is W+ with ghost number +2 and is given by (9:E:28) W+(z; u; k) = e =2 u ( ) S 0 (z)(@z x i + k +)eik x Vertex operators for R states may now be inserted on any worldsheets in any arbitrary (even) numbers. In critical RNS string theory, we have powerful results on the superholomorphicity properties of correlation functions for (factorized) vertex operators, at xed internal momenta. This result is contained in the Chiral Splitting Theorem for the RNS string. We begin by de ning factorized vertex operators. A vertex operator

F) The Chiral Splitting Theorem

Vi = d2zi d2 i Wi (zi ; i ; zi ; i )
is factorized provided Wi is the product of holomorphic and anti-holomorphic factors


f Wi (zi ; i ; zi ; i ) = Wi (zi ; i )Wi (zi ; i )
Any vertex operator may be written as a linear combination of factorized vertex operators. For example, for NS-NS massless states, we have

Wi (zi ; i ; zi ; i ) =

i D+ X eik XL

i D

X eik XR


Next, let AI and BI , I = 1; : : : ; h be any canonical basis of homology 1-cycles with canonical intersection form. We de ne h independent internal loop momenta pI 2 R10, associated with each AI -cycle, by

pI =



dz d D+X



Let the spin structures for left and right chirality spinors be equal, and equal to . The correlation function of the unintegrated vertex operators Wi at xed internal loop momenta pI and xed supermoduli are de ned by hW1 : : : WN iE (pI )


DX DBDC W1 : : : WN
dim sMh


h Y


I =1




dz d D+ X


K =1

j h K ; B i j2 e


(I) The rst part of the Chiral Splitting Theorem states that the RNS amplitudes factorize at xed internal loop momenta:

hW1 : : : WN iE (pI ) = (k)CF CF


Here, C F = C F (zi ; i ; i ; ; mi ; pI ; ki) is a complex analytic function of supermoduli mK , K = 1; : : : ; dim sMh , of the insertion points (zi ; i ), and of the left-moving factors i of the polarization tensors. CF is the complex conjugate of C F , with the same spin structure
CF (zi ; i ; i ; mK ; pI ; ki ) = C F (zi ; i ; i ; mK pI ; ki )


The functions C F may be evaluated explicitly using Green functions and determinants, using the prime form and the Szego kernel introduced in xV. (II) The second part of the Chiral Splitting Theorem prescribes the transition amplitudes for Type II and heterotic strings. We shall denote the analogous complex functions CB = CB (zi ; mk ; pI ; ki ) for the bosonic string, compacti ed on the 16-dim lattices of E8 E8 or Spin(32)=Z2 . Type II A, B:

Ah = ( k )



R 10h




dmK dmK

N YZ i=1

d2 zi d2 i C F C F


It is understood that for genus h = 0 (resp. h = 1) three (resp. 1) of the vertex operators should not be integrated over, and should be multiplied by a factor of ((C; )), where is a superconformal Killing vector.


Heterotic E8 E8, Spin(32)=Z2

Ah = ( k )

R 10 h






N YZ i=1

dzi d



dzi C F C B


Here Q and Q are spin structure dependent weight factors which realize the GSO projection. If carried out in the supermoduli picture, and should run only over even or odd. But, if the integrals are understood in components, then one should sum over all individual spin structures in each class as well. In general Q = 1, and Q = 1. To tree level, the worldsheet has the topology of the sphere, or stereographically projected plane, and there are neither moduli, nor odd moduli. Thus, we let g = 2jdzj2 and = 0. We concentrate on the scattering of massless NS-NS states, which can be represented by the vertex operator

G) Tree-Level Amplitudes for NS-NS states

V ("; k) = " (k)


d2z d2 D+X D X eik x :


By factorizing " (k) = (k) (k), we may start from a more convenient vertex operator

V ( ; ; k) =


d2z d2 eik X + D+X + D



and restrict to its contributions linear in and linear in . It is understood that and are Grassmann parameters. The X -propagator is easily calculated:

hX (z; )X (z0 ; 0 )i = ln jz z0

0 j2


The determinants of x, , b, c and , elds all combine into an overall constant, which we denote by z. The chiral splitting theorem is easily used here: there are no internal loop momenta, so that the correlation function of the unintegrated vertex operators

W (zi ; i ; zi ; i ) = WL(zi ; i )WR (zi ; i )


will yield a product of a function C F , complex analytic in zi , i times its complex conjugate. Here, the chiral vertex operators are

WL (zi ; i ) = eiki XL(zi; i )+ i D+XL(zi ; i)


and analogously for WR, where the chiral elds XL are contracted with the e ective propagator hXL (z; )XL (z0 ; 0 )i = ln(z z0 0 ) : (9:G:6) One readily nds

hWL (z1; 1 ) WL (zN ; N )i = CF (zi ; i )
CF = exp
N X i6=j



ij + 1 zij 2 i


1 + 1 k k ln z 2 zij { i j ij


where we have used the standard notation


ij = i

j; i j:

zij = zi zj


At tree level, we have invariance under superconformal automorphisms in OSp(1; 2); the integration measure must be properly treated to account for this invariance. This group of transformations is de end as follows; consider the matrices

0 K=@

0 +1 0 1 0 0A 0 0 1


0a b T =@c d


1 A

(9:G:9) (9:G:10)

+b z ! az + d + cz +

z+ ! cz + d + A +

where Latin entries are commuting and Greek entries are anticommuting. To obtain a superconformal transformation T , the line element dz = dz + d must be transformed into itself up to a conformal scaling. Equivalently the form

z12 = z1 z2

1 2=

v1 w2 v2 w1 w1 w2 IX.19

1 2


should transform into itself up to a conformal scaling. This is uniquely achieved when the orthosymplectic form is left invariant under T :

T T KT = D
The weight under which the di erence transforms is easily derived, and we have


T : z12 ! z12 = (cz + d + 1
Similarly the line element transforms as

z12 1 )(cz2 + d +



dz dz ! dz = (cz + d + )2
and the volume element as


dz^ dz ^ d ! (cz + d d ) : (9:G:15) + Elements in OSp(1; 2) are in unique correspondence with a triplet of points in the superplane (z1 ; 1 ), (z2; 2 ), (z3 ; 3 ) obeying one single (Grassmann valued) constraint. The counting works out because OSp(1; 2) has 3 commuting and 2 anticommuting parameters. The constraint is an OSp(1; 2) invariant Grassmann valued function dependent on three points, given by (9:G:16) = z12 3 + z31 2 + z23 112+ 1 2 3 (z z z ) =
The natural value for is of course 0, which implies that one is dependent. With this value for , it is easy to see that there is a unique correspondence between triplets of points satisfying = 0, and elements of OSp(1; 2), so that the latter may be accordingly parametrized. The invariant volume element induced on OSp(1; 2) is then
12 23 31

d = dz1(dz2 dz3d 1)d1=2d z12z23 z31 2



As for the bosonic string, the volume of the superconformal transformation group must be extracted from the amplitudes. As a result of OSp(1; 2) invariance, the 0, 1, 2 point functions vanish. IX.20

We now work out the three point amplitude. In evaluating CF , one retains terms proportional to 1 2 3; however, the term with three 's vanishes because 12 23 31 = 0. De ning the tensors


1 2 3


1 2

k1 3 +

2 3

k2 1 +

3 1




we nd for the three point function

hV ("1 ; k1)V ("2k2 )V ("3 k3)i = 4(2 )10 (k)"1 1 1 "2 2 2 "3 3 3 K

1 2 3


1 2 3


As a result of transversality of the polarization tensors, the 3 point function for massless external particles also vanishes. To compute the four point amplitude, we shall x superconformal invariance as follows: z = z1 , z2 = 0, z3 = 1, z4 = 1, 1 ; 2; 3 = 4 = 0, and then have exp G4 = +
1 1 2 3 2 4z z 12 34 3 k2

1 +


3 2

4z z 13 24

1 +
4 k2


4 2


1 14 z23 + perm.


2 1 4 z12 z13z24

+ k1

2 1 3 z12z14 z23


In principle, one should now multiply this whole expression by the one involving the 's, perform the integrals over z and and regroup terms, clearly a feudal task. The calculation is enormously simpli ed by the factorization properties of the Veneziano-integrals. Recall that we have the ordinary integrals

Z d2z ~ e zA zA (1 z)B (1 z )B =

~ e ( 1 A B ~ e ( A) ( B )

(1 + A) (1 + B) (A + B + 2)


~ e provided A A and B B are integers, which is always the case in string theory. Using ~ e the reciprocity formula for -functions and the fact that A A and B B are integers, ~ e this expression is actually symmetric under (A; B) $ (A; B ), as one might expect from complex conjugation. More importantly, the answer factorizes into a product of factors each only dependent either on the parameters for the z or z coordinates. This product property implies that one must only consider say the z coordinates to nd the full amplitude, which IX.21

by the same token will also completely factorize as a function of 's and 's. An analogous formula is derived for the super-integrals we need:

Z d2z1 d2 2

~ ~ A A 1 2 ]a 1 2 ]a z12 z12 (1

Here a and a are either 0 or 1, and the integrals are symmetric under (aAB) $ (~AB). ~ a ~e With the help of (9.G.22), it is now straightforward to evaluate the four point function

e z1 )B (1 z1)B e (1 ) ~ ~ B ~ = ( 2i)1 a (+2i)1 a ( a ~ A e ) (1 + A) + 1+ B) ( A) ( B ) (A + B + a


V ("1 ; k1 )V ("2 ; k2 )V ("3 ; k3 )V ("4 ; k4) Z u G +G 10 (k )g 4 d2 z d2 jz j s jz = (2 ) 1 2 12 1 1j e 4 4 (9:G:23) = (2 )10 (k)g4 ( s=2) ( t=2)t ( u=2) "11"22 "33"44K1234 K1234 s 1 + 2 1 + 2 +u 2 Using the abbreviation i for i to save some writing, we have K1234 = (st 13 24 su 14 23 4 2 3 1 s(k1 k3 24 + k2 k4 13 1 3 4 2 + t(k2 k4 13 + k3 k1 24 2 3 4 1 u(k1 k4 23 + k3 k2 14 tu 12 34) 3 2 4 1 k1 k4 23 k2 k3 14) 4 3 1 2 k2 k1 34 k3 k4 12) 4 3 2 1 k1 k2 34 k3 k4 12)

We begin by using part (I) of the Chiral Splitting Theorem, and we compute the Chiral Amplitudes CF . The spin structure for left and right movers is the same here. For even spin structures , there are no odd moduli, and there are no Dirac zero modes. The metric on the torus is taken to be g = 2jdzj2, and the torus is represented as usual by a parallelogram with sides 0, 1, , 1 + , with 2 H in the upper half complex plane H . The SX and SBC actions are free actions in components: 1 Z d2 z @ x @ x @z SX + SBC = 4 z z + @z + (9:H:1) + 2b@z c + 2b@z c + 2 @z + 2 @z : IX.22

H) One loop Amplitudes for NS-NS states in Type II A, B

The ghost and anti-ghost insertions are reduced to

b b cc :


No anti-superghost insertions occur for even spin structures, since there are no zero modes there. In the NS-NS sector, vertex operators do not involve ghosts, so we may integrate out b, c, b, c, , , and . We shall also retain the determinants of x and integrations. Thus (k )C C = M M F F (9:H:3) where M M arises from determinants and F F arises from correlation functions of vertex operators. We nd det0 0 5 det D+ 5 det D 5 = = M M = (Im )2 (9:H:4) 0 1 +1 det (det 1=2 ) 1 2 (Im ) As usual, on the torus, these determinants are related to one another, and we nd det0 0 = det0 1 = j ( )j4 (Im )2 (Im )2 (9:H:5) # ](0j ) 2 (det D+D ) = (det 1=2) = = = ( ) Putting all together:
4 M = # ](0j12) (9:H:6) ( ) up to an overall phase which is independent of , but which may depend upon . Clearly such a phase can be absorbed into the spin structure weights Q . The vertex contributions F are easily evaluated:

# ](0j )4 M = ( )12

F (zi ; i ;

; k; p) = expfi p2 + 2 p
1 2



( i i + iki zi) (9:H:7)

(ki kj G (zi ; i ; zj ; j ) + i j Di Dj G + + IX.23

+ 2iki j Dj G )g : +

Here, the correlation function G is given by

G (zi ; i ; zj ; j ) = hXL (zi ; i )XL (zj ; j )i = ln E (zi; zj ) + i j S (zi ; zj )


where E (z; w) is the prime form, de ned in xV, and S is the Szego kernel. For the torus, these functions are particularly simple: ( E (z; w) = #1#z (0jw)j ) ; 0 1 # ](z wj )#01(0j ) : S (z w) = # ](0j )# (z wj ) 1 (9:H:9)

For odd spin structures, , there is now one complex odd modulus and there are 10 Dirac zero modes, one for each space-time direction. The action SX + SBC is the one given on in (9.H.1) but there is now also an insertion of the worldsheet supercurrent involving

Sz+ Sz+ =


@z x

1b 2

+ 3 @z c + @z c + c:c : 2


Integrating out will bring down an insertion of with the ghost and anti-ghost insertions, we have

@z x and of

@z x. Combining this

bb cc ( ) ( ) ( ) ( ) Sz+ Sz :



Now, we specialize to correlation functions of massless NS-NS states. As shown in (9.E.5), these operators involve at most 2 factors of +. To obtain a non-zero contribution from odd spin structures, we need to have at least 5 external vertex operators. For example, the anomaly diagrams with 6 external vertex operators receives contributions solely from odd spin structure. We shall concentrate on amplitudes with 4 or fewer external vertex operators, and odd spin structures will never contribute to those. Notice that this means that for these amplitudes, there is no di erence between Type II A, B. For completeness, we record that for odd, we have M = M = 1. It remains to work out the amplitudes with 0, 1, 2, 3, 4 external vertex operators. We begin with determining the factors Q of spin structure weights. Following Mumford, IX.24

we label #-functions as follows:

#00 (z; ) = #3 (z; ) = #(z; ) 1 #01 (z; ) = #4 (z; ) = # z + 2; #10 (z; ) = #2 (z; ) = ei =4+ iz # z + 2 ; #11 (z; ) = #1 (z; ) = ei =4+ iz # z + 1 + 2 ; 2 #(z; )
1 X
n= 1



n2 +2 inz

Modular transformations of #2 , #3 , #4 obey

#ab(0; + 1) = e i4 a #a(b+a)(0; ) p #ab 0; 1 = ( )ab i #ba (0; )
The Dedekind -function satis es ( + 1)12 = 1 12 = Thus ( )12


( )12


Mab( + 1) = ei aMa(b+a+1)( ) Mab 1 = 14 Mba( ) :


We must now require modular invariance of the amplitude:



Q(ab) M(ab)( + 1) =




X Q(ab) M(ab)( 1= ) = 14 Q(a;b) Mab( ) :


Q(a;b) M(ab)( )


(The factor of 1= 4 is compensated for by the p-integral.) Clearly from ! 1= , we get Q(01) = Q(10) (9:H:17) IX.25

while from ! + 1, we get

Q(01) = Q(00)

(9:H:18) (9:H:19)

Q(01) = Q(10) = Q(00)

we x the overall phase of the amplitude, which is unobservable anyway. One proceeds similarly for . It is now straightforward to compute the 0-point function. (Recall that in the bosonic string this object produced the rst divergence we had encountered.) Since there are no vertex operator insertions, we simply have

A1 (0

Z d2 X pt) = Vol(M ) Q MM 4

(9:H:20) (9:H:21)

It su ces to examine the combination X Q M = Q(01) M(01) + Q(10) M(10) + Q(00) M(00)

Notice that since M(11) = 0 the corresponding contribution has been omitted in (9.H.21). Now, using the expressions for M , and the equality for Q's, we nd


Q Q = ((01) #4 (0j )4 + #2 (0j )4 #3 (0j )4 = 0 )12


by a famous Jacobi #-function identity. This identity arose once before in matching bosonic and fermionic multiplicities at all mass levels. In fact, this cancellation is the result of spacetime supersymmetry. The 0-point function, and thus the 1-point functions vanish. Recall from the discussion of the bosonic string that the vanishing of the 1-point function means that Minkowski at space-time is a one-loop solution to the superstring equations! Notice that our argument for cancellation was e ected only in the left-moving sector; thus, the same argument holds true in heterotic string theory as well. The above summation identity on #-functions is only one of a series, usually called the Riemann identities on theta functions. It is most useful here to express two more of IX.26

these identities in terms of the Dirac propagator (i.e. the Szego kernel): X Q M S (z1 z2)S (z2 z1) = 0


Q M S (z1 z2)S (z2 z3)S (z3 z1 ) = 0


Without giving details here, one similarly shows with the help of these identities that the two- and three-point functions vanish. In physics terms, this means that to one loop order, there is neither mass renormalization (massless particles remain massless) nor coupling constant renormalization. These results again arise from space-time supersymmetry, and are usually referred to as non-renormalization theorems. Let us consider the free fermion representation of the heterotic strings, starting from 32 Majorana-Weyl fermions , represented by 32 Euclidean Weyl spinors (z ). For simplicity, consider amplitudes with external states that are neutral under the gauge group. Then the one loop contribution of the internal degree of freedom will result entirely from the determinantal factors arising from integrating out . These depend upon the spin structures, just as the ones for space-time degrees of freedom . have the same spin structure, and the For the gauge group Spin(32)=Z2 , all determinant for given spin structure contributes a factor

I) One-loop amplitudes in the heterotic string

# ](0j )16 (9:I:1) = ( )48 The only modular invariant we can form is with all coe cients Q equal for all 3 even spin structures. For the gauge group E8 E8, are divided into two classes of 16. For each class, the determinant is 8 2 = # ](0j ) : (9:I:2) M ( )24 The only modular invariant is X 2 #2 (0j )8 + #3(0j )8 + #4(0j )8 M = : (9:I:3) 24 ( ) M4

One can show, using relations on the number of representations of an integer as a sum of squares, that this function equals the #-function for the E8 lattice: #E8 (0j ). Furthermore, the partition functions for Spin(32)=Z2 and E8 E8 are the same in view of









We see that no other modular invariant combinations are allowed. You may think P that Q M = 0 is allowed, but this would mean that some states enter with negative weight, and this is impossible since there are no fermions here! This argument shows now that E8 E8 and Spin(32)Z2 are indeed the only possibilities, as we had promised. It remains to discuss the 4-point functions, again for massless NS-NS states. Clearly, the 4-point function cannot vanish, lest the theory be non-interacting. It is evaluated by making use of a Riemann identity on #-functions with 4 Szego kernels:

J) The NS-NS 4 point function


Q M S (z1 z

(1) )S

(z2 z

(2) )S

(z3 z

(3) )S

(z1 z

(4) ) = 16




where (i) is a permutation of i, without xed points. The nal result for the 4-point function can be conveniently expressed as follows:

A1(ki ; "i ) = (k)"11 "22"33"44 K 1234K 1234A1 (s; t) where the single Lorentz-invariant function A1(s; t) is given by Z d2 Z d2z1 Z d2z2 Z d2z3 s (G +G G G ) 1 e 2 12 34 13 24 A1(s; t) = { 2 M1 22 2 2 2 t e+ 2 (G23+G14 G13 G24)
where Gij stands for the 1-loop propagator of the x- eld:



#1 (zi zj j ) 2 + 2 Im (z z )2 (9:J:4) Gij = G(zi zj )j ) ln i j 0 2 1 (0j ) and K is the kinematical factor that also arose in the tree-level amplitudes A0 (ki ; i ) for NS-NS transition amplitudes. Recall that K is polynomial in moments. By translation invariance on the torus, the point z4 may be chosen at will, say at z4 = 0.

It remains to study the function A1 (s; t)! We begin by examining the convergence structure of the above functional representation as a function of s and t. (1) For xed , and zi ! zj , i 6= j , we have

G(zi zj j ) ! ln jzi zj j2 ! +1
(2) while for zi 's well separated, but Im ! 1, we have


G(zi zj j ) !


! 1:


The latter limit imposes the strictest convergence condition: Re(s); Re(t); Re(u) 0 (9:J:7)

But, since we also have s + t + u = 0 for on-shell massless external states, the convergence condition (9.J.7) requires Re(s) = Re(t) = Re(u) = 0. It is easily seen that this condition is su cient for convergence, since the integral is bounded by the value at s = t = 0. This means that the integral is absolutely convergent for at least some values of the external momenta, in contrast with the case of the bosonic string, where the integral representation is never convergent for any momenta. Yet, the integral representation is de ned only for unphysical momentum con gurations where s, t, u are purely imaginary. A second shortcoming of the integral representation is that formally, when s, t, u are close to the physical region corresponding to real values of s and t, A1 (s; t) would be real. This would be inconsistent in an interacting theory, where the 1-loop 4-point function cannot be real, since its imaginary part is related by unitarity to the square of the tree level amplitude, which cannot vanish in any intracting theory. Our task is thus to properly de ne the amplitude throughout (s; t) 2 C C , including around the physical region where (s; t) 2 R R. Naturally, one seeks an analytic continuation of the amplitude in both s and t. (Recall that even at tree level, the integral representation A0(s; t) was absolutely convergent only in a small region, and the full amplitude had to be analytically continued in s and t as well.) IX.29

To appreciate the physical meaning of such an analytic continuation, we compare with the situation in quantum eld theory. There, the physical principle of causality is equivalent to the requirement of locality of observable elds, and this property, combined with Lorentz invariance, implies analytic dependence of the Green functions on the external momenta. Often, this analytic behavior is conveniently summarized in the form of a spectral representation. For example, the 2-point function (o -shell) may be expressed as

H (s) =


(M 2 ) dM 2 s M 2 + i" " > 0


where the real and positive function (M 2 ) is the density of states at mass M , also called the spectral density. Provided the integral over M 2 converges, it follows that H (s) is analytic throughout C , except possibly on the positive real axis, where it may have poles and branch cuts. In string theory, we do not have available at present a formulation of the amplitudes in terms of local eld observables (i.e. commuting at space-like separations). It is thus not known a priori whether analyticity of the amplitudes can be derived solely from Lorentz invariance and causality. (Probably, in a good second quantized formulation, one should be able to derive this.) Thus, analyticity of string amplitudes must be established, and the questions that must be answered are (1) does an analytic continuation exist? (2) is the analytic continuation unique? (3) is the analytic continuation physically acceptable? The present problem of analytic continuation of A1 (s; t) is a di cult matter of pure analysis. Yet, all three questions above have been answered in the a rmative, so that one has a full analytically continued amplitude A1(s; t) available, de ned throughout (s; t) 2 C C , with cuts and poles on R R, but nowhere else. What we have achieved here is the construction of a nite and unitary amplitude, at least to one loop order, that de nes NS-NS string scattering! We shall not reproduce the IX.30

details of the analytic continuatin procedure here. We shall limit ourselves to describing the physical singularities that appear in A1 (s; t) upon analytic continuation. To do so it is most convenient to represent the di erent singularities that appear as a function of s and t in terms of QFT Feynman diagrams. The Feynman diagrams indicate the region in moduli space (including the positions of vertex operators zi) that produce the singularities. The singularities essentially arise from combinations of the singular limits of Gij in (9.J.5) and (9.J.6), and fall into ve categories, schematically indicated by ve di erent types of Feynman diagrams below. Whenever two external zi and zj approach one another the Feynman diagram representation will show that their external lines merge

internal line


This singularity may be treated at the operator level by considering the operator product expansion (OPE) of the two external vertex operators. The internal line then represents all the operators that arise in the OPE. This type of singularity produces a pole in the total momentum traversing the internal line, i.e. in (ki + kj )2 , and the pole turns out to be at the value of the mass2 of the physical string states, i.e. 2n = M 2 , n 2 Z+. The ve di erent categories represented by diagrams (a), (b), (c), (d) and (e) arise, respectively, from 4, 3, 2, 1 and 0 external lines coming close together. Within each category, there are additional singularities (represented by a dotted line aross the loop) that arise from the asymptotic behavior in , given by (9.J.6).
1 4

2 3




2 3



double poles in s at s = 2n branch cuts in s, starting at s = 2n, n 2 Z+.


single poles in s at s = 2n branch cuts in s, starting at s = 2n, n 2 Z+.



no poles branch cuts in s, starting at s = 2n, n 2 Z+ with spectral density (2n; t) (2n; t) has branch cuts in t, starting at t = 2m, m 2 N0 .

It is understood that within each category, we also include permutation of the diagrams shown, for example by adding diagrams on which the external legs have been permuted. Now, for superstrings, diagrams in categories (a) and (b) vanish identically. This follows from the fact that the on-shell 1- and 2-point functions for massless states vanish identically. The singularities of the remaining diagrams are precisely the ones that one would expect on physical grounds, and their structure con rms the consistency of the string theory up to this order. A complete proof is in E. D'Hoker and D. H. Phong, \The Box Graph in Superstring Theory", Nucl. Phys. B440 (1995), 24. IX.32