II. Spectrum of Free Bosonic Strings
Eric D'Hoker We begin by constructing the Hilbert space of physical states and the spectrum of a single free bosonic closed oriented string, whose worldsheet has the topology of a cylinder (or equivalently an annulus) and of a single free open oriented string, whose worldsheet has the topology of half a cylinder (or equivalently half an annulus). A free string should not couple to any other strings, or to any non-trivial background elds in the space-time manifold M . Thus we take M = RD, with constant G and B = = T = 0 to start. We wish to examine the properties of physical strings in physical space-time, so it is essential that the space-time metric be Minkowskian, with G = = diag( + + +) and ; = 0; 1; : : : ; D 1. In quantum eld theory, the problem of constructing the Hilbert space of states and the spectrum of a single particle amounts to nding the 1-particle unitary representations of the Poincare group | a problem for which the Minkowski nature of space-time is again essential. A single string has an in nite number of excitation states, each of which corresponds to a representation of the Poincare group. In this lecture, we wish to identify which representations of the Poincare group occur (i.e. nd the spectrum) and determine the conditions under which all these representations are unitary. We will nd that conformal invariance plays an essential r^le. o Before we discuss the details of this fundamental analysis of string theory, let us outline the main steps of our strategy. 1) In the rst lecture, we postulated a prescription for the transition amplitudes of N external string states | represented by vertex operators Vi , i = 1; : : : ; N | in a general Riemannian space-time M with metric G (as well as possibly other elds B , , T )

Lectures 2, 3, 4


1 XZ

h=0 Met(

1 Z Dx V1 : : : VN e S x;g] Dg N(g) Di ( ;M ) ) II.1


2) To investigate the spectrum of the string in Minkowski RD we shall start with the M D with G = , B = = T = 0 and action above amplitudes in Euclidean RE 1 Z d gmn@ x @ x : (2:2) S x; g] = 8 g m n (For at Euclidean or Minkowskian metric, we have x 2 RD or RD respectively and we E M 0 ) = v v 0 .) The worldsheet metric g is always kept Riemannian. The use the notation (v; v string tension introduced in xI has been set to = 1=8 . In the physics literature one often nds the string tension expressed in terms of 0 (the Regge slope parameter) by = 4 1 ; so in our conventions 0 = 2.

3) At xed g, the quantum eld theory de ned by S x; g] provides an example of a conformal eld theory (see also the lectures by Gawedzki). All we shall need of the conformal eld theory to solve for the free string spectrum is contained in the operator product expansions. 4) Once we have derived the OPE's for Euclidean RD, we shall continue the time comE ponent of the map x, let x0 ! ix0 and obtain well-de ned OPE's for the operators in Minkowskian RD . In particular, we obtain two Virasoro algebras VirL VirR for the M closed string, and a single Virasoro algebra VirL for the open string. 5) From the Minkowskian OPE's and associated algebra, we construct Fock spaces Fopen and Fclosed for open and closed strings respectively along the same lines that Gawedzki constructed a Hilbert space. Due to the Minkowski signature of the space-time metric, however, the Fock space we obtain does not carry a positive de nite inner product. VirL VirR acts on Fclosed and VirL acts on Fopen. 6) The remaining integration over g | which is not present in conformal eld theory, but characteristic to string theory | e ectively selects out of F (either Fclosed or Fopen) a subspace F+ on which we have a positive de nite inner product, under certain conditions which we shall determine. In particular, this is how the critical dimension D = 26 arises. (The occurrence of negative norm states in a standard Fock space construction is not unique to string theory. It will take place for gauge theories and for gravity as well, or more generally in any theory with particles of spin 1.) II.2


Basics of Conformal Field Theory

An oriented surface , with a ( xed) Riemannian metric g automatically carries a complex structure, and we choose local complex coordinates z and z in which g is conformally at, and given by g = gzz (dz dz + dz dz) = 2gzz jdzj2 (2:3) This metric is left invariant under analytic or conformal di eomorphisms z ! z0 = f (z), with f complex analytic, combined with Weyl transformations on g. A quantum theory invariant under this transformation is a conformal eld theory (abbreviated CFT). Fundamental operators in CFT are conformal tensors or primary elds of conformal weight (h; h) satisfying the transformation law = (z; z )(dz)h (dz)h = 0 (z0 ; z0 )(dz0 )h (dz )h : (2:4)

For integer weights h and h, these are operator-valued sections of K h K h, where K is the holomorphic cotangent (or canonical) bundle of . In general, h and h don't have to be integers, but their di erence is the spin of the eld and must satisfy h h = spin 2 1 Z. 2 In the conformal eld theories we shall deal with in these lectures, the matrix elements of products of primary elds (and other elds) may be expressed in terms of functional integrals: Z h 1( 1 ) : : : N ( N )ig = Dx 1( 1) : : : N ( N )e S x;g] (2:5) where it is understood that i can be expressed in terms of the canonical elds x. Notice that here we do not normalize correlation functions by dividing out by the partition function h1ig . The stress tensor is de ned as the variation of the action S with respect to the inverse metric g 1 = gmn@m @n, where the metric g is viewed as a background source eld. 1 Z d T gmn gS = 4 g mn II.3 (2:6)

Assuming that the variation g has support away from the operator positions 1; : : : ; N , the variation of correlation functions is given by 1 Z d gmn hT g h 1 ( 1 ) : : : N ( N )i = 4 g mn 1 ( 1 ) : : : N ( N )i (2:7)

The latter equation may be used as a de nition of Tmn , even in conformal eld theories where no Lagrangian formulation is available. Applying Noether's theorem to conformal symmetry yields the following conserved currents: jz = vz Tzz jz = vz Tzz (2:8) where vz (resp. vz ) is any (locally) holomorphic (resp. anti-holomorphic) vector eld. Conformal symmetry requires @z jz = 0, @z jz = 0, so that we must also have

@z Tzz = 0

@z Tzz = 0


e Thus, T (z) = Tzz and T (z) = Tzz are complex analytic and anti-analytic respectively. In correlation functions, however, a product of T (z) and another eld (w) may produce singularities, as z ! w, so the above analyticity condition is to be understood as valid when T is away from other operators.

It was shown in Gawedzki's lectures how to extract these OPE singularities from conformal Ward identities for locally at metrics g; for the case of general metric g, the derivation is the subject of Problem Set #2. One nds

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

h + 1 @ (w; w) (z w)2 z w @w c=2 + 2 T + 1 @ T (z w)4 (z w)2 ww z w w ww


where c is the central charge of the CFT and h is the holomorphic conformal weight of . e (Analogous relations hold for T .) As always, these relations are to be understood as valid when inserted into any correlation function. The sign used in (2.10) stands for the fact II.4

that we have retained only the singular part on the r.h.s. When c = 0, T (z) is a conformal tensor of weight (2; 0), but for c 6= 0, T transforms as a projective connection instead, with c (2:11) T 0 (z0 )(dz0 )2 = T (z)(dz)2 12 fz0 ; zg(dz)2 where the Schwarzian derivative f ; g is de ned by ff ; zg f 000 =f 0 (3=2)(f 00 =f 0 )2 ; f 0 = @f=@z: Analytic currents j = j z dz, labeling di erent species of currents, give rise to timeindependent or conserved charges: 1 I j Q (C ) = 2 i C
C C’ time

Here Q (C ) is the charge enclosed in the region bounded by C and equals Q (C 0) provided no sources for Q occur between C and C 0. The action of charges on elds may be evaluated using only the residues in the OPE of j with : I 1 (2:12) Q (C ) (w; w) = 2 i j (z) (w; w ) = zRes j (z) (w; w) !w C provided no other sources for j lie within C . The commutator between two charges is recovered as follows:

Q (C1)Q (C2) Q (C3)Q (C2) = Q ; Q ](C2 ) 1 I Res j (z)j (z ) = 2 i z!z2 2

C2 z2 C3



The conserved charges associated with conformal transformations are obtained from vector elds v(zm) = zm+1 and v(zm) = zm+1 :

~ Lm



dz zm+1 T (z) 2 i dz zm+1 T (z ): e 2 i


e From the OPE of two T 's (and two T 's), we recover two Virasoro algebras c Lm; Ln] = (m n)Lm+n + 12 m(m2 1) m+n;0 c ~ ~ ~ Lm; Ln] = (m n)Lm+n + 12 m(m2 1) m+n;0 :


e From the OPE of T (and T ) with a primary eld of conformal weight (h; h), we obtain

Lm; (w; w)] = wm+1 @w (w; w) + h(m + 1)wm (w; w) ~ Lm; (w; w)] = wm+1 @w (w; w) + h(m + 1)wm (w; w)


An important example of a CFT is the free scalar eld x: ! R (or R= Zas discussed in Witten's lectures), with action 1 Z d2 z @ x@ x (2:17) S x] = 4 z z The OPE is derived from the Green function for x:

x(z; z )x(w; w)
It is customary to de ne normal ordering by

ln jz wj2

(2:18) (2:19)

: x(z; z )x(w; w) : x(z; z )x(w; w) + ln jz wj2

The normal ordered product admits a smooth limit as w ! z. The renormalized (also called the quantum) stress tensor may now be de ned by 1 lim (2:20) T (z) = 1 : (@z x)2 : 2 w!z @z x@w x + (z 1w)2 : 2 Elementary primary elds in this CFT are listed below, with their corresponding conformal weight. (1; 0) @z x (0; 1) (2:21) @z x 1 k2; 1 k2 : exp ikx : 2 2 The central charge is c = 1. Notice that while the derivative of x and the exponential are primary elds, x itself is not a well-de ned quantum eld (cf. Witten's lecture). II.6


The free closed bosonic string conformal eld theory

The worldsheet topology for the free closed bosonic string is that of the cylinder or equivalently the annulus.
τ τ+ i σ z worldsheet time

z=e σ



The metric g determines global complex coordinates z, z in which g = 2gzz jdzj2 . Recall from 2) that our starting point is the at Euclidean RD worldsheet action E Z (2:22) S = 41 d2z @z x @z x which for xed metric g de nes a CFT of D scalar elds x , which are the coordinates of the map x: ! RD. The quantum stress tensor is E T (z) = 1 : @z x @z x : (2:23) 2 and the central charge is c = D. Following point 3), we obtain the OPE's and associated algebraic structure. To do this, we take advantage of the fact that @z x(z) and T (z) are holomorphic, and thus admit a Laurent expansion inside the annulus:

@z x(z) = i @z x(z ) = i
From the de nition of T


m2Z X m2Z

xm z xm z ~

m 1 m 1

T (z ) =
e T (z ) =


m2Z X m2Z

Lm z
~ Lm z

m 2 m 2


X Lm = 1 xm 2

1 2 : @z x

@z x:, we have

1 1 x2 + X x x L0 = 2 0 n n n=1



m 6= 0



From the OPE of @z x with itself and with T (z), we get

xm ; xn] = m


Lm ; xn ] = nxm+n ;
~ ~ Lm ; xn ] = nxm+n ; ~

(2:26a) (2:26b)

xm ; xn] = m ~ ~


~ while tilded and untilded operators mutually commute. Lm and Lm satisfy Virasoro algebras with central charge c = D. Following point 4), we nd that the above OPE's and their algebra contents can safely be continued to x0 ! ix0 where x0 is the time component of the map x: ! RD . It is M clear that this means x0 ! ix0 ) m m all m 2 Z: (2:27) 0 ! ix0 xm ~m ~ The stress tensor is now 1 T (z ) = 2 : @ z x @ z x : and the algebraic relations are 1 : @ x @ x: 2 z z (2:28)

xm ; xn ] = m m+n;0 Lm ; xn] = nxm+n X Lm = 1 : xm n xn: 2


~ and analogously for x and L. The central charge is unchanged, and we still have VirL VirR . ~ The eld x itself is not a conformal tensor, but may be obtained by integrating @z x and @z x: x(z; z ) = xL (z) + xR (z ) (2:30) with

xL (z) = qL ix0 ln z + i xR (z ) = qR


m ix0 ln z + i xm zm : ~ ~ m6=0

m6=0 X

xm zm




Since x : ! RD , x must be a single-valued map as a function of z. This will require that M x0 = x0 = p, and p is to be identi ed with the momentum operator of the string, while ~ q = qL + qR is the overall position operator of the string. ~ It will turn out to be useful to record the explicit form of the operators L0 and L0, expressed in terms of p. 1 1 p2 + X x x L0 = 2 n n n=1 (2:32) 1 1 p2 + X x x ~ L0 = 2 ~ n ~n For non-simply connected space-time manifolds M , the map x need not be single-valued, since the string can wrap around non-trivial 1-cycles in M . As a result, x0 need not be equal to x0 , but the di erence x0 x0 belongs to a lattice, dual to H 1(M; Z). ~ ~ C)

The free open bosonic string conformal eld theory
The worldsheet of a free open string has the topology of a strip or a half annulus.
Im z

Re z


Fundamental open strings should have freely moving boundary points. Thus Neumann conditions @Im z x = 0 at the boundary B are appropriate. (Dirichlet b.c. would x the end points of the string on some submanifold D of M . Such submanifolds are D-branes, or depending upon their space dimension p: p-branes. In certain string theories, D-branes become dynamical (extended) objects, perhaps to be discussed later on.) It is standard to double by re ecting onto the lower half plane into and then considering the union 0 = which is the full annulus. Functions on with Neumann II.9

boundary conditions on B are then even functions under z ! z . Thus, the mode expansion of the open string is obtained from this restriction applied to x(z; z )

x(z; z ) = xL (z) + xL (z )


and only involves the oscillators xm but there are no oscillators xm . Similarly, we only ~ ~ have Virasoro generators Ln but there are no Ln. As a result, the energy-momentum tensor satis es e T (z) = T (z ) = 0 on Im(z) = 0: (2:34) This condition guarantees conservation of T across B, and results from the fact that Di ( ) leaves B invariant. We have the mode expansion

x(z; z ) = x + 2ip ln jzj2 i

m m xm zm + z m m6=0


Notice that x0 = 2p, in contrast with the closed string. It is useful to also record the Virasoro generators

Lm =

X L0 = 2p2 + x n n=1 X


xn m 6= 0






Thus the open string only has a single Virasoro algebra VirL. D)

Fock Space, Negative Norm States

We now follow point 5) and construct the Fock space of the conformal eld theory of x, continued to Minkowski space-time. (This parallels Gawedzki's construction). ~ To do so, we must choose a polarization and split the operators fxm g and fxm g into a group of raising, lowering and central generators. This choice must be consistent with Poincare invariance. (Also, here we must use a property inherited from the Minkowski nature of the worldsheet, namely that xm and xm oscillators are independent, except at ~ m = 0. We shall not expand upon this issue now, but come back to this part later on.) II.10

The correct choice is (x0 )y = x0 , while xm and xm are raising (resp. lowering) for m < 0 ~ (resp. m > 0). Also (xm )y = x m and this implies (~m )y = x m x ~ ~m ~ Ly = L m : (2:37) (2:38)

Ly = L m


The Fock space Fk is constructed from xm oscillators only. The ground state j 0; kiL (also called the highest weight state) is labeled by the momentum k (the eigenvalue of the momentum operator p)
8 < pj 0; k iL = k j 0; k iL : x j 0; k i = 0 m L

m > 0.

(2:39) (2:40)

We may choose its normalization as follows: L h0; k j0; k 0 iL = (k k 0 ) :

Arbitrary vectors in Fk are obtained by applying x m 's for m > 0 to the ground state

j "; kiL = "

1 :::


(k; m1 ; : : : ; mn)x 1m1 ; : : : x n n j 0; kiL m


for all possible Lorentz polarization tensors " 1 ::: n (k; m1; : : : ; mn); n 2 N, with all possible mi 2 N. The open string Fock space is a sum over all momenta k of Fk:

Fopen =





The closed string Fock space is a sum over all momenta | which must be the same for left and right movers, in view of the constraint x0 = x0 | of the tensor product of Fk and ~ e e Fk, where Fk is built from xm : ~

Fclosed =


k 2R D M

e Fk Fk :


Using the norm on j 0; kiL introduced in (2.40), Fk naturally inherits a quadratic form. However, Fk automatically contains negative norm states. To see this, consider states j "; kiL = " x m j0; kiL and with m > 0 evaluate their norm:
L h"; k j"; k 0 iL = h0; k j"

xy m " x

m j0; k 0 i = m"2 (k

k0 )


For "2 > 0, the norm of the state j "; kiL is positive. But, for "2 < 0, which is allowed with Minkowski metric, the norm of j "; kiL is negative. Such states are called ghosts and their presence violates quantum mechanics: they cannot be present in a physically sensible theory. E)

Integration over Met( ) | Virasoro Constraints

We now deal with point 6): the integration over Met( ), and the elimination of ghosts from the spectrum. We shall perform this integration with great care in forthcoming lectures, but we can already here extract its key e ects. The metric g enters in the action S as a non-dynamical eld, i.e. without derivatives. The r^le of the metric g here is completely analogous to that of the time-like component o A0 of the gauge eld in Yang-Mills theory (see also Faddeev's lecture). The e ect of such non-dynamical elds is to supply a constraint on the dynamics of the remaining degrees of freedom, and to reduce their phase space. (You may loosely view the e ect in the functional R integral as generating a delta function dA0eiA0 x = (x).) The constraints also generate a group action on the phase space, and the reduced phase space may be viewed as the quotient by this group action. Let us analyze the e ect of the constraints at the classical level rst. The variational equations are 8 by varying x < @z @z x = 0 (2:43) : T (z ) = T (z ) = 0 by varying g
1 e The second equation is a constraint on x since T (z) = 2 @z x @z x (similarly for T (z)). Since @z T (z) = 0, the rst equation shows that T will vanish throughout as long as T = 0 has been imposed on the boundary. The action of the constraints at the classical level is


something we have already analyzed, namely the Virasoro algebra but with zero central charge. Geometrically, the constraint re ects the fact that vibrations of x( ) tangent to the surface x( ) are eliminated, leaving only transverse direction: (D 2) transverse o directions 2 longitudinal directions eliminated by T = T = 0
x (Σ)

We now analyze the e ects of the constraints at the quantum level. There are basically three ways of dealing with the constraints: 1) Eliminate the longitudinal degrees of freedom at the classical level by solving for them in terms of the transverse degrees of freedom, and then quantize the transverse degrees of freedom. This procedure goes under the name of light-cone gauge quantization. It cannot be achieved in a manifestly Lorentz invariant parametrization. We shall discuss it very brie y later on in this lecture. 2) Ultimately, in lecture VII, we shall reformulate this constraint problem in terms of the powerful BRST quantization method. This is a more complex scheme based on introducing extra, unphysical degrees of freedom, called Faddeev-Popov ghosts, and enforcing BRST (Becchi-Rouet-Stora-Tyutin) invariance. This method is manifestly Lorentz invariant. The physical Hilbert space will in fact arise as a cohomology group of a semi-in nite cohomology complex, as developed by Feigin and later on by Freeman and Olive in physics and Frenkel, Garland and Zuckerman in mathematics. 3) Before launching into BRST, however, we shall analyze the quantum e ects of the constraints in a more direct fashion, which will also yield immediate results on the spectrum. In this third method, we impose the constraint as an invariance condition and use it to II.13

select an invariant subspace out of the full Fock space F (either Fopen for open strings or Fclosed for closed strings). In Yang-Mills theory, the corresponding constraint was just Gauss' law, enforcing gauge invariance on physical states. In string theory, the constraints are the vanishing of the stress tensor, enforcing Di ( ) invariance on physical states. Due to the central extension in Vir, the reduced Fock space F+ = F=Vir will have a positive de nite norm only when certain extra conditions are satis ed, which we now determine. F)

Physical Spectrum, No-Ghost Theorem

We use the Virasoro constraints to select an invariant subspace out of the full Fock space ~ F. It is tempting to try and impose Lnj i = Lnj i = 0 for all n 2 Z. However, unless c = 0, these conditions imply that j i = 0. Instead, we impose invariance under half of Vir. The correct physical state conditions are therefore
phys j iL 2 Fk e phys j i R 2 Fk

if if

(Lm a m;0 )j ~ (Lm a m;0 )j

iL = 0 iR = 0

m2N m 2 N:


and the full physical Fock spaces are obtained by
phys phys Fopen = k Fk ; phys phys e phys Fclosed = k Fk Fk :

~ The constant a is undetermined at this point, and is allowed for since L0 and L0 were naturally de ned only up to an additive constant in view of the ordering choice we made in their de nition. For closed strings, the value of a is taken the same for L0 ~ ~ and L0 so that L0 L0 = @@ enforces rotation invariance on the annulus. Notice that states j i ; j i 2 Fphys automatically yield zero matrix elements of Virasoro generators ~ Lm a m;0 for all m 2 Z(similarly for Lm a m;0 in the case of the closed string). Hence, e we have now realized the physical state condition h jT j i = jT j = 0! States of the form j i = L m j i, for m > 0, are called spurious, and we de ne

Fkspurious = j i =


1 M

L mj




spurious is generated by states of the form L j i L j i, It is easy to see that Fk 1 1 2 2 using the commutation relations of the Virasoro algebra which imply that L 1; L n ] is a non-zero multiple of L 1 n for n 2. Notice that spurious states are orthogonal to all physical states. States that are both physical and spurious are null, phys Fnull ; Fkspurious \ Fk k null where Fk stands for the space of null states of momentum k. ~ The special Virasoro conditions (L0 a)j i = (L0 a)j i = 0 determine the value of k2 for a given level of operators, and thus determines the mass of the state. To see this, introduce the number (also called level) operators


1 X





e N

1 X

n 1

x ~

~ n xn



e The operators N and N count the number of operators x n and x n , n 1 with weight ~ n, applied to the ground state j 0; ki. The special Virasoro conditions thus become e closed (k2 + M 2 )j i = 0 M 2 = 2N 2a and N = N j i = 0 (2:46) 1 1 open (k 2 + M 2 )j i = 0 M 2 = 2 N 2 a : Thus, the mass squared of various string states are spread by 2 integers for closed strings and 1 integers for open strings. 2 The mass squared of a state is one of the Casimir values that label the representation of the Poincare group under which the state transforms. The remaining data that fully specify the Poincare representation are those given by the representation of the stabilizer group (i.e. the subgroup that leaves the momentum invariant). This information is contained in the expression of the Lorentz generators of the string.


= qup



It follows that the string ground state j 0; ki transforms as a scalar (i.e. trivial representation under the stabilizer). II.15

1x ~ ~ ~ i n (~ n xn x n x n) for closed strings n=1

1 X

1 (x x x x ) n n n n n=1 n

1 X



We observed in the rst lecture that, at long distances compared to the Planck length `P , the size of strings e ectively becomes unobservable. We can then approximate the dynamics of the string by that of point particles, and string theory by quantum eld theory. Thus, we shall seek to identify string states with states in quantum eld theory. To do so, we require that string and quantum eld states transform under the same representations of the Poincare group, as well as of other symmetry groups that are available. Also their interactions should agree in this long distance limit. We examine under what conditions the Virasoro constraints eliminate all negative norm phys states, so that Fk has positive norm. Before stating the general result, let's look at the lowest level cases (for the closed string; for the open string it su ces to let M 2 ! M 2 =4). (1) Ground state j 0; kiL with mass square M 2 = 2a (actually j 0; kiL j 0; kiR for the closed string). The Virasoro conditions are automatically satis ed. (2) First excited state j "; kiL = " x 1 j 0; kiL with mass square M 2 = 2a + 2:

L1j "; kiL = 0 =) " k = 0 :
The remaining Virasoro conditions Lnj "; kiL = 0; n 2, are automatically satis ed. There are three possibilities: (i) a > 1, so that k2 is space-like. If we pick k = (0; k1; 0 : : : 0) then we are allowed to choose " = ("0 ; 0 : : : 0), and we see that ghosts are still present. (ii) a < 1, so that k2 time-like. If we pick k = (k0; 0 : : : 0) we see that we always have "2 > 0 and there are D 1 independent states with positive norm. (iii) a = 1, so that k2 is light-like. If we pick k = (k1; k1; 0 : : : 0), we see that there are (D 2) massless states of positive norm, and one of zero norm. Case (i) is clearly eliminated: Virasoro constraints do not su ce to eliminate all negative norm states. Case (ii) has massive M 2 > 0 vector states (open string), or massive rank 2 II.16

tensor states (closed string). Case (iii) has massless M 2 = 0 vector states (open strings), which suggests the presence of massless Yang-Mills states, and has massless M 2 = 0 rank 2 tensor states (closed strings) which suggests the presence of gravity (for symmetric tensors). 3) At the next level j "; kiL = f" x 1 x 2a + 4 and L1 j "; kiL = 0 L2 j "; kiL = 0 The norm of the state is with

+ x 2 gj 0; kiL, with mass squared M 2 =

8 <" :"

k +
+ 2k

=0 = 0:

Lh"; k j"; k 0 iL = 2NLh0; k j0; k 0 iL

N=" " +

= ("00 )2 2("0i )2 + ("ij )2 ( 0 )2 + ( i )2 : Choosing a rest frame where k0 = M , ki = 0, the Virasoro conditions become
00 M + 0 = 0 "i0M + i = 0 "00 + "ii + 2M 0


The Virasoro conditions impose no constraints on the traceless part of "ij , and its contribution to N is 0. Hence, we may retain only the trace part: "ij = " ij . Eliminating i and 0 and "00 in favor of "0i and ", we have

" ( 1) N = (2MD + 1)2 (2M 2 + 1)2 (M 2 1)(D 1) + (M 2 2)("0i )2: 2
Assuming D > 1, we nd the conditions
8 2 <M :



(2M 2 + 1)2 (M 2 1)(D 1) 0 :

the rst of which is already satis ed since M 2 = 2a + 4 and we have seen before that we must have a 1. For M 2 = 2 (corresponding to massless states at level 1), we nd from the second inequality that D 26. In fact, for D = 26, the second inequality implies M 2 2, so that we must have M 2 = 2 for D = 26. II.17

For D 25, the quadratic form (2M 2 + 1)2 (M 2 1)(D 1) is non-negative for all M 2 2, so the state j "; ki always has non-negative norm. For D 27, the quadratic form is negative for a region in which M 2 2. In fact, by analyzing higher level states it can be shown that for D 27 no value of a can be chosen so as to eliminate all ghosts. The complete result can be summarized in the following

No Ghost Theorem (R. Brower, P. Goddard and C. Thorn). For D > 26 or a > 1 Fphys contains negative norm states; For D = 26 and a < 1 Fphys contains negative norm states; For D = 26 and a = 1 Fphys contains only non-negative norm states, in particular, two towers of null states and F+ Fphys=Fnull has positive de nite norm;
3 Fnull = (L 2 + 2 L2 1)j 2i L 1 j 1i
For D < 26 and a 1

Fphys but contains only positive norm states for a < 1 (contains one tower of null states L 1 j 1i for a = 1).

We shall not prove the no ghost theorem here, since later on, using BRST quantization, a simple proof can be given. The case D 25 and a 1 is contained in D = 26, a = 1 by setting 26 D space-like operators to zero, and adjusting the 26 D positive squared momenta p2 , i = 1; : : : ; 26 D, i so that 1 X p2 a + 1 = 0 : i 2



Spectrum of the critical bosonic string with D = 26, a = 1

Based on the masses and Lorentz tensor structure of the lowest lying states of the spectrum, we may identify which particles they naturally correspond to in eld theory. It is easy to see that the spectrum consists entirely of bosons.

Open strings
(1) States corresponding to j 0; ki have M 2 = traveling faster than light), and Lorentz scalars;
1 2,

hence they are tachyons (particles

(2) States of the form " xj 0; ki with M 2 = 0 are vectors under SO(1; D 1). There are (D 2) positive norm states and one null state, which constitute a massless gauge boson in RD. Gauge transformations have the form " ! " + k (invariance under these transformations is clear since they amount to shifting the state by a null state L 1j 0; ki = k x 1 j 0; ki); (3) M 2 > 0 states, with M 2 = 1 n, n 2 N, transform under various tensor representations 2 of SO(1; D 1).

Closed strings
(1) States corresponding to j 0; ki = j 0; kiL j 0; kiR have M 2 = 2, so they are tachyons, and Lorentz scalars; (2) States of the form " x 1 x 1 j 0; ki have M 2 = 0. Corresponding to the trace-part of " ~ there is a Lorentz scalar, the dilaton, with positive norm. The symmetric traceless part of " gives a Lorentz rank 2 tensor, the graviton. There are D(D+1) 1 such states, D(D 1) 1 2 2 of them having positive norm. As for the antisymmetric part of ", it gives a Lorentz rank 2 tensor called the anti-symmetric tensor with D(D 1) states, out of which (D 1)(D 2) have 2 2 positive norm. These states exhibit gauge invariance under

" ! " + k + 0k
(which amounts to shifting by null states); II.19


(3) M 2 > 0 states, with M 2 = 2n, n 2 N, transform under various tensor representations of SO(1; D 1). H) For D = 26 and a = 1, the null states in Fnull may be ignored completely. We may retain positive norm states only at the cost of giving up manifest Lorentz invariance, SO(1; 25), by retaining D 2 = 24 transverse oscillators only. Let M 0 be a space-like subvector space of M , with dim M 0 = 24, and let ", " be in M 0 ; then " x n and " x n for ~ e ~ n > 0 create only positive norm states. This construction has manifest SO(24) invariance. It can be checked that the full Lorentz algebra SO(1; 25) closes in this realization. Also, it can be shown in this case that there is a one-to-one correspondence between those states constructed above and the states in F+ = Fphys=Fnull in the covariant approach. If the same construction were attempted for D 6= 26 or a 6= 1, one would nd that the Lorentz algebra in this realization does not close. Using the above correspondence, it is very easy to estimate the total number of states at each mass level. Introducing a generating function
1 2M 2 +2 ) = Y (1 G(w) = TrF+ (w n=1 1 X

Lightcone gauge, density of states

wn )


dnw2n :
25=2 eM=4


We nd the number of states asymptotically for large M 2 : dM 2 = M I)


Primary elds and Vertex Operators for Physical States

The construction given above for the physical Hilbert space has the disadvantage that the operators x n with n > 0, create states in Fock space that are not in general physical. We now construct vertex operators, which have the property that they send physical states into physical states, and spurious states into other spurious states. ~ Since physical states obey (Ln n;0)j i = (Ln n;0)j i = 0, n 0, it is clear that any operator commuting with the entire Virasoro algebra will yield such a vertex operator. II.20

~ Also, since Ln; V ] = Ln; V ] = 0, V automatically sends spurious states into spurious states. Since Ln generates conformal reparametrizations, V should be conformal invariant. To construct V 's, we consider general tensors, then construct invariants. We use the action of Ln on any primary eld:

Ln; (z; z )] = zn+1@z (z; z ) + h(n + 1)zn (z; z ) = @z (zn+1 (z; z )) + (h 1)(n + 1)zn (z; z ) ~ Ln; (z; z )] = @z (zn+1 (z; z )) + (h 1)(n + 1)zn (z; z ) :


It is clear that amongst primary elds, the only operator that commutes with all of Ln ~ and Ln is the unit operator = 1. However, integrals of may be invariant. For any primary eld of conformal weight ~ (h; h) = (1; 1), its commutators with Ln, Ln are total derivatives, and so its integral will be Virasoro invariant. In addition, primary elds of weight (1; 1) form natural volume forms on . Hence the integral of



(z; z)

satis es

~ Ln; V ] = Ln; V ] = 0 :


This is not the only way to construct vertex operators though: if is a holomorphic primary ~ eld (e.g. @z x ), then it will commute with Ln provided h = 0 and its commutator with Ln is a total derivative if h = 1. In addition (1; 0) and (0; 1) forms are naturally integrated along contours. Thus, the operators

dz dz

(1;0) (0;1)



are also vertex operators provided that @z (1;0) = 0 and @z (0;1) = 0. The operators (1;0) and (0;1) are precisely what we de ned previously as conformal currents; their charges are conserved. Such operators do not in fact create new physical states, as will become clear by considering examples below. II.21

Vertex operators for the closed bosonic string are now easily reconstructed from the basic eld x associated with D copies of the c = 1 CFT. We have the following primary elds @z x holomorphic, weight (1; 0) @z x anti-holomorphic, weight (0; 1) : eik x : weight k22 ; k22 From the holomorphic current @x x we get the conserved charge of momentum: 1 I dz @ x : p =2 z (2:52)

phys Physical states in Fk will just be eigenstates of p , so p does not really create any new states, but simply measures their momentum k. A single exponential yields a vertex operator for k2 = 2, i.e. when the conformal weight is (1; 1): Z VT (k) = d2 z : eikx : (2:53)

This scalar operator creates a state of mass M 2 = 2, which is the tachyon discovered previously. But the exponential may also be combined with derivatives for k2 = 0

VG;B; (k) =


d2 z " (k) @z x @z x : eik x :


which reproduces the states for the graviton, the antisymmetric tensor, and the dilaton as symmetric traceless, anti-symmetric, and trace part respectively, with k " (k) = k " (k) = 0. For higher powers of @z x and @z x , we will have masses of the form 2 M 2 = 2n 2 in general. However, higher derivatives @z x may also appear. While these are not primary elds (they mix with @z x under conformal reparametrizations), they may combine with other operators and still yield primary elds. When we obtained the Virasoro conditions on the polarization tensors " , we established an invariance of the states in Fphys under the transformations

" !" +k



(We shall shortly interpret this invariance as a manifestation of Di (M )-invariance for the closed string.) This invariance is also easily established for the vertex operators: for example a shift by yields

d2 z k

@z x @z x

: eikx : = =i

i d2 z @z x (@z eikx)


d2 z (@z @z x )eik x


But we have @z @z x = 0 by the eld equations | away from any other singularities | and thus the integral vanishes. (The e ects of the singularities may also be taken into account, as we shall show later on and do not modify the conclusion above.) J)

Identifying the graviton: vertex operators from background elds

One of the most important discoveries in string theory is that the k2 = 0, " (k)symmetric traceless physical state, found above, should be identi ed with the graviton. This is the particle in quantum eld theory that represents excitations of the space-time metric and that mediates the force of gravity. Here, we provide further evidence for this identi cation. Along the same lines, we provide evidence for the identi cation of the open string k2 = 0, " (k)-state with the Yang-Mills particle in this section and in xK. Consider a space-time M which is almost at with the following metric

G (x) =

+ " (x)


where " (x) is everywhere small compared to . The tensor " (x) may be viewed as a superposition of plane waves by Fourier transform:

" (x) = d26k " (k)eik x
1 Z d gmn@ x @ x G (x) SG = 8 g m n II.23



Now, consider string theory in this background: the action on the worldsheet is given by (2:59)

and the transition amplitude involves the exponential of SG. Since " is small, we may expand in powers of " :


1 n SG = e S X ( 1) Sn 1 (8 )n n=0 n!


with S the action corresponding to the at metric and Sn de ned by

Sn = d26k1 : : : d26knV (k1 ) : : : V (kn )


V (k ) = " (k )


d g gmn@m x @nx eikx

The operator V (k) is precisely the vertex operator we constructed for the graviton | written now in general metric g | for a de nite momentum k. Thus, we see that strings couple to the uctuations in the background metric precisely by the graviton vertex operator. The plane waves may be regarded as creating excitations in the background space-time metric, which make it deviate from the at metric. (A completely analogous correspondence holds for the other massless states, the dilaton and the anti-symmetric tensor.) We have however established more precise information: in order to maintain conformal invariance and eliminate negative norm states, the momentum k and the polarization tensor " had to satisfy on-shell conditions

k2" (k) = 0 =) " (x) = G (x) = 0 (2:62) k " (k) = 0 =) @ " (x) = @ G (x) = 0 " (k) = 0 =) " (x) = 0 ; where we have listed to the right of the double arrows the restatement of the conditions on the metric. Now, since G is almost at (" being small), corrections quadratic and higher in " may be neglected. Considering the Ricci tensor in this approximation, we have R =R =@ @ + O("2 ) (2:63) 2R = @ @ " @ @ " @ @ " + "

The conditions on G obtained from the on-shell conditions of " (k) now imply that

R =0;


within the almost at (or linearized) approximation. In addition, the fact that shifts by null states are physically unobservable implies the gauge invariance of the physical states in F+ under the shifts

" (k ) ! " (k ) + k
(with k into



= 0 for on-shell states). In terms of the metric, this gauge invariance translates

" (x) ! " (x) + @

(x) + @



We recognize this transformation as an in nitesimal di eomorphism generated by the vector eld (x), in the approximation where O("2 ) terms are neglected. R = 0 is invariant under these transformations, as can be checked explicitly. In fact R = 0 is the only Di (M )-invariant equation that reduces to G (x) = 0 in the linearized limit. Thus, we nd that | to leading order in metric uctuations { the conformal invariance constraints translate into Ricci atness, i.e. Einstein's equations in the absence of matter. This discovery supports our proposed identi cation of " (k), k2 = 0 string states with gravitons. Later on (Lecture VIII), we shall further con rm this identi cation in another important limit, that of small variations in the derivatives of the metric, the so-called low energy limit. K)

Internal degrees of freedom of open strings: Chan-Paton rules

The end points of the open string are distinguished points. It is consistent with Poincare and 2-d conformal invariance to put degrees of freedom or charges at the end points. During interactions of several strings, the charges simply ow along the boundary, and should thus II.25

be conserved.

The simplest case corresponds to adding electric charge to the end points of oriented strings:


e -e e





Consistency with the interactions requires that the charges be opposite (i.e. corresonding to conjugate representatives) at the two ends. An external U (1) gauge eld A couples to these charges just in the same way as it couples to point particles with electric charge e, moving on worldlines: Z SA = ie d x ( )A (x( )): _ (2:67) Following the derivation of vertex operators from the coupling to background elds that we used for the closed string, we easily obtain the vertex operators for the open string gauge states, by decomposing A into Fourier modes

A (x) = dD k " (k)eik x ;
where " (k) is the polarization tensor. Thus, the vertex operators are



V ("; k) = " (k)


d x ( )eik x( ) _



with conformal invariance for k2 = 0. We may add further degrees of freedom to the end points of the string in the following way. Let i , i = 1; : : : ; N be Grassmann valued elds that live only on the boundary B of the string worldsheet (recall that Neumann b.c. were imposed on B for the map x: ! M ). A natural candidate for the action for such degrees of freedom is

S0( ) =





d d




This action is Di (B) invariant, and does not depend upon the metric. The equation of motion for i simply requires that i be -indepenent, so that i may indeed be interpreted as conserved charges. The action S0 automatically has SO(N ) symmetry when i are real (and i = ij j ) and SO(2N ) symmetry when i is complex. As a result, quantization automatically produces a Hilbert space of states on which a spinor representation of SO(N ) (for i real) or SO(2N ) (for i complex) acts. To restrict to a subgroup G of SO(N ) or SO(2N ), and/or a subrepresentation of the corresponding spinor representations, it is necessary to restrict the Hilbert space of states. This is conveniently achieved by introducing a Lagrange multiplier, : B ! R. For G = U (N ), for example, the restriction to the rank p (0 p N ) totally anti-symmetric tensor representation of U (N ), we use the following modi cation of S0:

S( ) =





d d

i + ( )( i i



Analogous modi cations may be used for G = SO(N ) and Sp(N ). Let us now restrict to G = U (N ), and p = 1, so that equals the de ning representation. Quantization of i generates N conserved charges attached to the boundaries of the string
j a j (T ) i _ i i j


The string for U (N ) gauge group should be oriented, since the charges at both ends of the string transform under inequivalent representations of U (N ). The string state now transforms under the representation of U (N ), which is the adjoint of U (N ). The representation matrices T a, with a = 1; : : : ; dim U (N ) = N 2 describe the string state belonging to the representation , by the index a. The matrix elements (T a )j of i T a specify which i and j are created at the boundary B of the open string. The full Fock space of the open string is given by Fopen , and the physical state space is phys Fopen . We observe that for G = U (N ) and the de ning representation, transforms in the adjoint representation and the open string states are thus natural candidates for YangMills particles with gauge group G = U (N ). Had we considered any other representation for (other than the de ning or its complex conjugate), the string states would not have tranformed precisely under the adjoint representation of U (N ), and could not have been interpreted as Yang-Mills particles. Thus the de ning representation is singled out by its relevance to describing Yang-Mills particles. The strings for SO(N ) and Sp(N ) should be unoriented, since the charges at both ends transform under equivalent representations of SO(N ) and Sp(N ), respectively. Considering to be the de ning representation of SO(N ), we obtain string states in the adjoint of SO(N ) by the anti-symmetrized tensor product of by itself ( )A. On the other hand, for the de ning representation of Sp(N ), we obtain the string states in the adjoint of Sp(N ) by the symmetrized tensor product of by itself ( )S . In each case we recover precisely the string states suitable for Yang-Mills particles. For exceptional groups, no construction is known that produces string states in the adjoint representation only, and it is widely believed that it is impossible to obtain YangMills particles from open strings with these groups. A non-Abelian gauge eld can couple to the boundaries of the open string, much in the same way as this happens for Abelian gauge elds. Here, however, we must include the coupling of the charges of the gauge eld Aa to those of the string. Let us restrict again II.28

to the de ning representation of U (N ). The coupling of Aa is given as follows:

SA ( ) =





j d a_ a j i d + iA x (T )i

j + ( )( i i

1) :


The coupling to the gauge eld Aa may again be used to construct the vertex operators for the creation and annihilation of gauge states of the string. One nds upon quantization that the vertex opertor for the U (1) case is simply to be multiplied by the matrices (T a )j : i

V a("; k) = T aV ("; k) :


As the internal degree of freedom and the matrices T a commute with Virasoro, the physical state conditions on V a are satis ed whenever they are on V ("; k). It is instructive to examine the e ects of these degrees of freedom on interactions as well. For simplicity, we restrict to Aa = 0. Charges then simply continue to be translated along the boundary B in a conserved way, as shown schematically in the gure below.
j _ i i _ k _ k j

Interactions are only allowed if charges are conserved.

In a full string diagram, the e ect will be to include a trace over the matrices T , just as in Yang-Mills Feynman rules, as shown schematically below.
d k l i a b c


tr T a T b T c T d


With the help of the Yang-Mills vertex operators V a of (2.73) and (2.69), one recovers Yang-Mills scattering amplitudes in the limit where momenta are small compared to the Planch length.