Weigand Introduction To String Theory

Introduction to String Theory
Winter term 2015/16
Timo Weigand
Institut für Theoretische Physik, Universität Heidelberg
Digitalisierung des Skripts: Max Kerstan, Christoph Mayrhofer

Contents
Literature 1
1 Motivation and Overview 2

1.1 Why do we study String Theory? . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The need for a quantum theory of gravity . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Programme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 The classical bosonic string 9

2.1 The relativistic point particle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 The bosonic string action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.1 Nambu-Goto action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.2 The Polyakov action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.3 Symmetries of the Polyakov action . . . . . . . . . . . . . . . . . . . . . . 16
2.2.4 Flat worldsheet coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.5 Hamiltonian formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Oscillator expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.1 Equations of motion and boundary conditions . . . . . . . . . . . . . . . . 22
2.3.2 Closed string expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.3 Open string expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3.4 The concept of D-branes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.3.5 The Virasoro generators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3 Quantisation of the bosonic string 31

3.1 Old canonical quantisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.1 Canonical quantisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1.2 The quantum Virasoro algebra . . . . . . . . . . . . . . . . . . . . . . . . 32
3.1.3 Fock space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1.4 Explicit open (NN) Fock space and criticality . . . . . . . . . . . . . . . . 36
3.2 Light-cone quantisation (LCQ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.1 Light-cone gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2.2 Quantisation in LCG . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.2.3 Normal ordering constant “a” as a Casimir energy . . . . . . . . . . . . . 45
3.2.4 Open string spectrum along D-branes . . . . . . . . . . . . . . . . . . . . 48
3.2.5 Closed string spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.3 Modern covariant quantisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.1 The Polyakov Path integral . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3.2 Quantisation of the bc-system . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.3.3 BRST Quantisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
i
CONTENTS ii
4 Conformal field theory (CFT) 67

4.1 Conformal invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2 The conformal group in d ≥ 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.3 The conformal group in d = 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3.1 Infinitesimal conformal transformations . . . . . . . . . . . . . . . . . . . 69
4.3.2 The Möbius group as the group of global conformal transformations . . . 70
4.3.3 Relation between the complex (half-)plane and the worldsheet on the cylin-
der (or strip) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.4 Conformal fields and their OPE . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.5 Conformal Ward-Takahashi identities and energy-momentum tensor . . . . . . . 75
4.5.1 General Ward-Takahashi identities . . . . . . . . . . . . . . . . . . . . . . 76
4.5.2 Conformal Ward-Takahashi identities . . . . . . . . . . . . . . . . . . . . 78
4.5.3 T (z) as a conformal field . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.6 State-operator correspondence, highest weight states, Verma modules . . . . . . . 83
4.7 Correlation functions in CFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.8 Normal ordering and Wick’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.9 Applications to String Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.9.1 The free boson on the sphere . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.9.2 The bc-ghost-system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.9.3 String quantisation à la CFT . . . . . . . . . . . . . . . . . . . . . . . . . 90
5 String Interactions 93
5.1 Perturbative Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.2 Moduli space of Riemann surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.3 The gauge fixed S-matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
5.4 Tree-level amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5.4.1 Correlators on the sphere - the Virasoro-Shapiro amplitude . . . . . . . . 103
5.4.2 Correlators on the disk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.5 1-loop amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.5.1 Oriented closed theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.5.2 Oriented open theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
5.5.3 Non-oriented vacuum amplitudes and tadpole cancellation . . . . . . . . . 120
5.6 Strings on curved backgrounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.6.1 The non-linear σ-model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.6.2 α0 -Expansion and Conformal Invariance . . . . . . . . . . . . . . . . . . . 123
6 Superstring theory 128

6.1 The classical RNS action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
6.2 Super-conformal invariance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
6.3 Mode expansions: Ramond vs. Neveu-Schwarz . . . . . . . . . . . . . . . . . . . 132
6.4 Canonical quantisation and Super-Virasoro-Algebra . . . . . . . . . . . . . . . . 134
6.4.1 Canonical (anti-)commutation relations . . . . . . . . . . . . . . . . . . . 135
6.4.2 Interlude: Spinors of SO(1, d − 1) . . . . . . . . . . . . . . . . . . . . . . 136
6.4.3 Super-Virasoro-Algebra and physical state condition . . . . . . . . . . . . 137
6.4.4 Normal ordering constants . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
6.5 Open string spectrum in light-cone quantisation (all NN) . . . . . . . . . . . . . 140
6.6 Closed string spectrum in LCQ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
6.7 The GSO projection: Type IIA and Type IIB . . . . . . . . . . . . . . . . . . . . 146
6.8 Digression: Differential forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
CONTENTS iii
6.8.1 p-form potentials in physics . . . . . . . . . . . . . . . . . . . . . . . . . . 152

6.9 Type I theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7 Compactification, T-duality, D-branes 156

7.1 Kaluza-Klein compactification in field theory . . . . . . . . . . . . . . . . . . . . 156
7.2 KK compactification of closed bosonic strings . . . . . . . . . . . . . . . . . . . . 158
7.3 T-duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
7.4 Dp-branes as dynamical objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
7.5 Intersecting Brane Worlds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
7.6 Elements of Calabi-Yau compactification . . . . . . . . . . . . . . . . . . . . . . . 171
Literature
This is the write-up of my course ”Introduction to String Theory”, held in the Master Programme
at the University of Heidelberg during winter term 2015/16. The course comprises 28 lectures,
each lasting for 100 minutes.
The material discussed in this lecture course is introductory and necessarily covers only a small
fraction of the wide range of topics studied in modern textbooks. It is therefore highly recom-
mended to consult the literature in addition to following the course. A list of classic references
and some useful open-soruce material accessible online is collected here:
[GSW] Green, Schwarz, Witten: String Theory, Vol. 1& 2, Cambridge University Press 1987
[P] Polchinski: String Theory, Vol. 1& 2, Cambridge University Press 1998
[BBS] Becker, Becker, Schwarz: String Theory and M-Theory - A modern introduction, Cam-
bridge University Press 2007
[BLT] Blumenhagen, Lüst, Theisen: Basic Concepts of String Theory, Springer 2012
[Z ] Zwiebach: A first course in String Theory, Cambridge University Press 2004

[BP] Blumenhagen, Plauschinn: Introduction to Conformal Field Theory, Springer 2009
[T] Tong: Lectures on String Theory, http://arxiv.org/pdf/0908.0333
[PC] Polchinski, Chaudhuri, Johnson: Notes on D-branes,

http://arXiv.org/pdf/hep-th/9602052
[U] Uranga: Introduction to String Theory
http://www.ift.uam.es/paginaspersonales/angeluranga/firstpage.html
In addition, I will give more specific references at various places.

The first part of these lecture notes, in particular the choice of conventions and notation, is
inspired mostly by [BLT]. In the second half we follow in addition [P].
1
Chapter 1
Motivation and Overview
1.1 Why do we study String Theory?

There are many reasons to study String Theory. Here are just a few:
• String theory for idealists

String Theory is the leading proposal for a fundamental unified theory of Quantum Gravity
and Quantum Field Theory. What distinguishes it from other approaches to quantising
gravity is that String Theory is currently the only theory that is able to provide a unified
quantum description of all interactions of Nature, both gravitational and non-
gravitational ones. String Theory is a universal theory that deals both with the most
fundamental questions of Relativity - such as the nature of spacetime singularities, black
hole physics or the history of the Universe - and provides a theoretically well-founded
guideline for particle physics beyond the Standard Model. Its aim is none less than to
provide a theory of everything. However, one must keep in mind that at this stage the
String Theory approach to Quantum Gravity and Quantum Field Theory is speculative as
it is not (yet?) proven (or disproven) if this is the path chosen by Nature. In following this
proposal theorists aim high...
• String Theory for pragmatists
While the question whether or not string theory is the fundamental theory of Nature is
certainly controversial, it is widely accepted that String Theory provides a powerful tool
to study strongly coupled field theories via the concept of holography. If a theory inside
a volume V obeys the holographic principle this means that the information about its
spectrum and dynamics is encoded in the degrees of freedom residing on the boundary
∂V . That this principle is obviously at work in Quantum Gravity is reflected already in
the famous Bekenstein-Hawking entropy law, according to which the entropy of a black
hole is given by S = 4GAN with A the area of the horizon surface. More spectacularly,
considerations of String Theory have revealed a much more radical manifestation of the
holographic principle in the disguise of the AdS/CFT correspondence: String Theory on
a space of negative constant curvature, an Anti-deSitter (AdS) space, is dual to a conformal
field theory (CFT) on the boundary of the AdS space. The power of this remarkable insight
lies in the fact this is a strong-weak coupling duality: Weakly coupled, perturbative
string theory on AdS is equivalent to a strongly coupled field theory. This way, perturbative
string computations can give insights into the boundary field theory in the strongly coupled
regime, for which no perturbative techniques exist. This has lead to many applications and
2
CHAPTER 1. MOTIVATION AND OVERVIEW 3
has made String Theory a tool of interest even for physicists primarily (or exclusively)
interested in aspects of field theory. The ultimate hope would be to learn something
about confining gauge theories such as QCD. More recent developments include applications
to condensed matter theory phenomena (AdS/CMT correspondence) or fluid dynamics
(AdS/Hydrodynamics). In calling this a pragmatic approach we recall that here String
Theory is primarily viewed as a tool to learn about strongly coupled field theory and
not necessarily as a fundamental theory of Nature. Even if it turned out that Nature
has chosen a different UV completion of gravity and Yang-Mills theory, the field theoretic
insights derived via the AdS/CFT correspondence will always remain valid.
• String theory for aesthetes
String Theory is a theory of fascinating mathematical complexity and beauty. It has lead to
an exuberantly fruitful interplay between mathematics and physics, which actually
goes both ways: Not only does String Theory offer a natural arena for the application of
modern mathematical, in particular geometric concepts. The intuition of String Theorists
has also lead to completely new developments and insights within pure mathematics. The
most famous example of this interplay is the concept of Mirror Symmetry, a one-to-one cor-
respondence of pairs of certain complex manifolds (so-called Calabi-Yau manifolds), which
was first discovered in String Theory and subsequently studied within mathematics. In
fact, while Mirror Symmetry is rather surprising from the purely mathematical perspec-
tive, it is almost a triviality (and was discovered as such) from the point of the view of the
underlying string theoretic conformal field theory. Other examples of stringy explanations
for highly non-trivial mathematical facts are the group theoretic Monster Moonshine or
the classification of singularities by ADE Dynkin diagrams.
• String theory for agnosticists
For these and many more reasons, String Theory has developed as an entire framework
(rather than a specific theory) from which numerous developments of High Energy physics
depart. It is fair to say that for young theoretical physicists at least a general knowledge of
String Theory is almost as indispensible as a solid foundation of Quantum Field Theory if
they want to follow the recent current developments. Love it or hate it, but you must
know it!
1.2 The need for a quantum theory of gravity

We now analyse in slightly greater detail the motivation to study a theory of strings as a funda-
mental quantum theory of interactions. Obviously the pillars of modern physics are on the one
hand Quantum Mechanics or, rather, Quantum Field Theory (QFT) and on the other hand Gen-
eral Relativity (GR). Both have lead to a triumphant description of many observed phenomena
of Nature, but both theories are in a sense incomplete and cannot be considered fundamental
theories:
• The incompleteness of QFT, at least in its perturbative formulation, becomes evident in the
appearance of ultra-violet (UV) loop divergencies in the computation of perturbative
scattering amplitudes. Such divergencies are present already for as simple theories as QED
and cast serious doubt on the validity of the theory at high energies.
• The incompleteness of classical GR is exemplified in the context of black hole physics: Ac-
cording to the classical Einstein equations, the center of a black hole contains a spacetime
singularity where the curvature blows up. This is particularly scandalous as black holes
can form dynamically starting from a perfectly well-defined initial distribution of matter.
Thus there exist dynamical processes that take us outside the regime of validity of the
classical theory.
It is important to keep in mind that the appearing divergencies or singularities in QFT and
GR are not a problem at a practical or computational level: In GR the singularity at the
center of a black hole is hidden behind the black hole horizon (in fact, according to the Cosmic
Censorship Hypothesis, this is conjectured to hold for all singularities - no naked singularities
are supposed to exist). In QFT, the UV divergencies can be ”hidden” at a practical level
by the powerful machinery of regularisation and renormalisation (at least for renormalisable
theories). But irrespective of our ability to argue away the divergencies in applications, their
very appearance remains highly unsatisfactory.
In fact the situation becomes not only unsatisfactory, but even technically unaccpetable as soon
as combined effects of GR and QFT are taken into account. Instances where quantum aspects
in gravity cannot be neglected include the study of (small) black holes and, most notably, a
successful treatment of very early cosmology. The problem is that Einstein gravity is not
renormalisable at a perturbative quantum level.
In principle two different ways out of this dilemma are conceivable:

• Stick to the dynamical degrees of freedom of classical Einstein gravity, but modify the
quantisation procedure. This is the philosophy underlying, among others, Loop Quantum
Gravity: The aim is to quantise gravity as such, and no unification of gravity and the
remaining forces in Nature is implied or even aimed at.
• Alternatively, we may stick to the conventional methods of quantisation, but change the
dynamics of gravity in such a way as to recover the degrees of freedom of Einstein gravity
at low energies while at the same time arriving at a consistent UV completion. This is
the approach taken by String Theory. In particular, the modifications of the dynamics
automatically affect GR and QFT in a way that leads to a unified description of all forces.
In fact, String Theory represents in some sense the simplest, most economical and most conser-
vative modification of QFT and GR. The only new dynamical input can be summarised as
follows:
The fundamental objects in Nature are not pointlike, but 1-dimensional.
This axiom is combined with the standard kinematics of general covariance (Relativity)
and the usual procedure of quantisation. As we will see, quantisation of a generally
covariant theory of strings inevitably results in a consistent unified description of gravity and
Yang-Mills theory.
The two sectors arise from the elementary fact that a string can have
two possible topologies: It can be open or closed. Open strings describe
Yang-Mills theory, closed strings describe Gravity. Since open strings can
close up and vice versa, Gravity and YM theory are automatically related
dynamically.
What is important is that all other outcomes of the theory follow without further assumptions
or arbitrariness from application of the laws of general covariance and quantisation to a theory
strings.
1.3 Programme
The general roadmap, which is also the one followed in this course, can be summarised as follows:
• Define the classical action of a string propagating in d-dimensional spacetime. Generalising

the concept of the (0+1)-dimensional worldline of a point particle (which is a 0-dimensional
object), a string traces out a (1+1)-dimensional surface - the string worldsheet (WS).
The following picture shows the string worldsheet of an open and a closed string propagating
in spacetime.
• Application of the rules of quantisation provides us with the Fock space of string exci-
tations. The massless modes of the bosonic sector include (among others):
vector boson of
open string: Aµ spin 1
Yang-Mills Theory
closed string: hµν spin 2 graviton

In addition one finds a tower of massive string excitiations of mass
1 4
M2 = (N −1) (open bosonic), M2 = (N −1) (closed bosonic) N = 0, 1, 2, . . . . ,
α0 α0
(1.1)
2
The appearance of the tachyon, the lowest lying state of negative mass will be overcome
by moving from the bosonic string theory to superstring theory, in which no tachyons
are present. This is the theory we study eventually.
• Note the appearance of the dimensionful parameter α0 of dimension
[α0 ] = [length]2 . (1.2)
The intrinisic length and mass scales of the theory are thus
√ 1
`s = 2π α0 : string length Ms = √ : string mass scale. (1.3)
α0
Importantly, α0 is the only free paramenter of String Theory. In principle it can take any
value in the range √
−33 0 −17
| {z cm} ≤ α ≤ |10 {z cm} .
10 (1.4)
Planck length TeVscale
• Apart from this dimensionful parameter there are no free dimensionless parameters.
The situation is thus fundamentally different from GR or QFT, where all coupling con-
stants (particle masses, Yukawa couplings...) are input parameters of the action with no
fundamental explanation. As we will see, the couplings in String Theory are expectation
values of dynamical fields (so-called moduli) which take their value dynamically.
• The interactions follow uniquely and without further extra input. Essentially this is
because there is a unique way in which strings can join and split.
To analyse the properties of the theory it is useful to distinguish two different regimes:
• The UV regime involves energies of the order or bigger than the string scale Ms or,
equivalently, distances of the order of the string length `s . In this regime the extended
nature of the string becomes important. As a consequence it turns out that the theory is
non-local with important consequences for the nature of interactions.
• The low-energy dynamics kicks in at energies Ms or, equivalently, at distances `s ,
strings appear as effectively pointlike. One can deduce the low-energy effective theory by
integrating out the massive string tower and focussing, to first order, on the massless string
excitations.
Let us consider both regimes in turn:
Low-energy dynamics:
In the low-energy limit the effective dynamics reduces to gauge interactions√ and gravity. Con-
sistency of the worldsheet teory predicts explicitly that to lowest order in α0 the gravitational
laws are given by
1
Rµν − Rgµν = 8πGN Tµν . (1.5)
2
Indeed we will see in this course how to derive Einstein’s equations as a corollary of
self-consistency of the worldsheet theory at the quantum level. In addition one √ finds higher
curvature corrections, which can be computed systematically order by order in α0 /r, with r a
typical length scale in the problem.
High energy dynamics:
• Compared to a point particle theory, the string interaction is smoothened out. In fact
there does no longer exist the notion of a sharply defined interaction vertex at which the
interaction is localised in space and time. Rather, locally the string always look like it is
freely propagating and the interactions are encoded only in the topology of the worldsheet,
i.e. in global properties of the worldsheet. A schematic comparison of a point particle and
a string vertex can be found in the following illustration.
This is an important feature: In point particle theories the sharp localisation of the inter-
action vertex is indeed responsible for the appearance of divergencies in the computation
of amplitudes.
• The number of combinatorically relevant diagrams is greatly reduced compared to point

particle QFT. At each loop order there exists only a single string diagram for a given
process. This is to be contrasted with a factorially growing number of Feynman graphs in
QFT.
These two properties are responsible for the improvement of the UV properties of scattering
amplitudes. UV finiteness has been proven rigorously for the superstring up to 2-loop order.
However, it is conjectured that all higher order amplitudes are likewise UV finite.
Another important property of the string interactions is that gravity and YM interactions are no
longer independent: Since open strings can join to form closed strings, consistent YM interactions
automatically imply the presence of gravity as well. This of course is as it must be in a quantum
theory of fundamental interactions as the energy stored in, say, the electric field gravitates itself.
Note that the situation is fundamentally different in other approaches to quantum gravity, where
the gravitational and the YM sector are treated as indpendent building blocks. By contrast,
String Theory provides ar truly unified quantum theory of gravity and YM interactions.
Further properties of the theory:

• Consistency of the worldsheet theory requires that the total number of spacetime di-
mensions be d = 26 in the bosonic string theory and d = 10 in the superstring, which is
the one we will eventually study. String theoy is currently the only theory that predicts
the number of spacetime dimensions as an outcome of the theory, not as an input.
• Contact with physics in 4
extended spacetime dimen-
sions is made by compact-
ification of the extra di-
mensions on a small inter-
nal manifold. The study of
the resulting landscape of
string vacua is the subject
of string phenomenology.
• The theory predicts

the presence of higher-
dimensional objects called
D-branes. These are hy-
persurfaces on which open
strings can end. They
play a crucial role for the
dynamics of the theory.
• In superstring theory the spectrum of string excitations enjoys a symmetry called super-
symmetry, which is a symmetry between bosons and fermions. This symmetry is realised
in the full ten-dimensional theory. Compactification can and, in general, does break su-
persymmetry. The scale at which supersymmetry is broken depends, among other things,
on the geometric details of the compactification. It is important to keep in mind that
low-energy supersymmetry is not a prediction of string theory. String theory is perfectly
fine with supersymmetry broken at scales above the TeV scale probed currently by LHC.
Chapter 2
The classical bosonic string
2.1 The relativistic point particle

Before turning to the classical dynamics of a string, we recall some basic facts about the kine-
matics and dynamics of a point particle. The relevant concepts will then be generalised directly
to the classical string.
The propagation of a particle in R1,d−1 is described by a worldline γ parametrised by τ ∈ R:
γ: τ 7→ X µ (τ ) ∈ R1,d−1 , µ = 0, i.
• We will stick to ~ = c = 1 and work in signature (-1, 1,..., 1).
• The invariant length or line element of the metric ηµν is ds with ds2 = −ηµν dX µ dX ν .
• The dynamics of a massive particle with mass m 6= 0 is captured by the Nambu-Goto-
Action Z Z
p
SN G = −m ds = −m −ηµν dX µ dX ν . (2.1)
γ γ
µ
dX
With dX µ (τ ) = dτ dτ ≡ Ẋ µ dτ this is
Z q Z
µ ν
SN G = −m dτ −ηµν Ẋ Ẋ ≡ dτ L(τ ). (2.2)
γ γ
Note that τ is a priori an arbitrary parameter. For the special case that we identify τ with
the proper time τ0 of the particle SN G reduces to the familiar action
Z p
S̃ = −m dt 1 − ~v 2 . (~ = c = 1)
9
CHAPTER 2. THE CLASSICAL BOSONIC STRING 10
To see this recall that, reinstating c for illustrative reasons, we have X µ = (ct, ~
x) and the proper
time τ0 satisfies
2
dX µ dX µ

dt 1 d~
x
=γ= q , ~v = ⇒ = (γc, γ~v ), = γ 2 (−c2 + ~v 2 ) = −c2 .
dτ0 2 dt dτ0 dτ0
1 − ~vc2
• SN G is invariant under reparametrization τ → τ̃ (τ ), which corresponds to local diffeo-

morphism invariance on the worldline.
r
dX µ dX µ dτ̃ dX µ dX ν
Z
Under this transformation = ⇒ S̃N G = −m dτ̃ −ηµν .
dτ dτ̃ dτ dτ̃ dτ̃
• The canonical momentum associated to X µ is

∂L Ẋ µ
Pµ = = mp . (2.3)
∂ Ẋµ −Ẋ 2
This leads to the constraint Φ = P 2 + m2 = 0.
At this stage we collect - without proof - some useful facts on constrained systems, following
[BLT]1 :
2
– A dynamical system is called singular if the matrix Mij = ∂∂pq̇ji = ∂ q̇∂i ∂Lq̇j is not invertible. In
this case the generalised velocities q̇ i cannot be expressed uniquely in terms of the q i , pi .
– For a singular system each zero eigenvalue of Mij gives rise to a primary constraint. A
primary constraint is a constraint on the system that follows from the definition of the
conjugate momentum without use of the equations of motion. Secondary constraints are
further constraints introduced so that if the primary constraints are satisfied at a certain
time, they remain to be so at all later times.
– In systems with primary constraints, the Hamiltonian is not uniquely defined as one can
always add terms that vanish upon imposing the constraints. If there are constraints φk the
Hamiltonian is H = Hcan + ck φk where Hcan = pi q̇i − L is the canonical Hamiltonian and ck
do not depend of qi and pi . Different choices of ck correspond to different choices of gauge.
– Indeed one can check explicitly that for the Nambu-Goto-action of the free particle Hcan =
pi q̇i − L ≡ 0. Thus the system is singular in the above sense and there must exist a primary
constraint. The full Hamiltonian is given just by the constraint as
H = c Φ = c (P 2 + m2 ). (2.4)
From this,
Ẋ µ
q
dXµ
= {X µ , H}P.B. = c p =⇒ c = −Ẋ 2 , (2.5)
dτ −Ẋ 2
identifying the choice c = 1 with the gauge choice τ = τ0 , the proper time.
• The equations of motion (e.o.m.) are

!
d ∂L ∂L d Ẋ µ
0= − µ
= mp . (2.6)
dτ ∂ Ẋ µ ∂X dτ −Ẋ 2
For τ = τ0 the proper time this reduces to mẌ µ = 0.

1 A thorough treatment of constrained systems can be found e.g. in the book ’Quantisation of Gauge Systems’
by Henneaux and Teitelboim.

• The complete dynamics of the system is described by the e.o.m. together with the
constraints. For τ = τ0 we recover, as expected,
Ẍ µ = 0 & Ẋ 2 = −1. (2.7)
• The disadvantage of SN G is that it is not defined for m = 0.

This is remedied by introducing the Polyakov action
2 !
dX µ
Z
1 1 2
SP = dτ e(τ ) −m , (2.8)
2 e2 (τ ) dτ
with e(τ ) a new independent degree of freedom.

Invariance under local diffeomorphisms τ → τ̃ (τ ) = τ − (τ ) is given provided X µ and
e transform as
dX µ
X µ (τ ) → X µ (τ ) + (τ ), (2.9)
dτ
d
e(τ ) → e(τ ) + ((τ )e(τ )). (2.10)
dτ
(Check this!)
This means that from the worldline perspective X µ transforms as a scalar, while e trans-
forms as a scalar density of weight 1. More on the definition of tensor densities can be
found on the first examples sheet.
• The equations of motion now follow by variation w.r.t. X µ and e(τ ).
First, e(τ ) is treated as an independent degree of freedom (d.o.f.) with associated e.o.m.
δSP 1 µ 2
0= ⇒ Ẋ + m2 = 0. (2.11)
δe(τ ) e2 (τ )
q
−(Ẋ µ )2
⇒ e(τ ) = if m 6= 0. (2.12)
m
In addition we have the e.o.m. for X µ ,

δSP d 1
0= ⇒ Ẋ µ = 0, (2.13)
δX µ dτ e(τ )
dX µ 2 µ
−1 dX µ
e−1 d
e−1 dX d
R R R
where we used δX dτ =2 dτ dτ δXµ − 2 δXµ dτ e dτ .
• To show classical equivalence of the Nambu-Goto and the Polyakov action we insert the
e.o.m. of e(τ ) into SP . More generally, this procedure is called integrating out e(τ ).
Note, however, that classical equivalence does in general not imply equivalence also at the
quantum level.
• In view of the transformation properties of e(τ ) we can rewrite SP as
dX µ dXµ
Z
1 p
SP = dτ det gτ τ g τ τ − m2 , (2.14)
2 dτ dτ
where gτ τ ≡ e2 (τ ) is identified as the worldline metric.

• One can use local diffeomorphism invariance to fix the gauge by setting e.g.
1
, m= 6 0.
e(τ ) = m (2.15)
1, m = 0.
Then the e.o.m. for e(τ ) translate into constraints in addition to the e.o.m. for
X µ:
2
µ Ẋ = −1, m=6 0
Ẍ = 0 & for τ = τ0 (proper time). (2.16)
Ẋ 2 = 0, m=0
This is a general lesson that will become crucial in the treatment of the string: After
integrating out an auxiliary field we must still impose its equations of motion
in the form of constraints.
2.2 The bosonic string action

2.2.1 Nambu-Goto action
We now generalise these considerations to the classical string.
• A string is a 1-dimensional object. The position along the string is parametrised by a
spatial coordinate σ, which we normalise to take values in the range 0 ≤ σ ≤ l.
• There are 2 types of strings, corresponding to the 2 possible topologies of a 1-dimensional
object:
closed open
• The propagation of a string in R1,d−1 now defines a 2-dimensional worldsheet (WS) Σ
Σ is parametrised by τ ∈ R and 0 ≤ σ ≤ l and can be viewed as the map
Σ : (τ, σ) 7→ X µ (τ, σ) ∈ R1,d−1 . (2.17)
This map is called the string map. It provides an embedding of the WS into the ambient
spacetime.
Oftentimes we write (τ, σ) ≡ ξ a , a = 0, 1 for the worldsheet coordinates.
• In complete analogy to point particles one defines the Nambu-Goto action

Z
SN G = −T dA, (2.18)
Σ
where dA is the area element of Σ

s
∂X µ ∂X ν

dA = − det ηµν d2 ξ. (2.19)
∂ξ a ∂ξ b
q
µ dX ν
This replaces the notion of the invariant line element ds = − dX dτ dτ ηµν dτ featuring in
the point particle action.
The object
∂X µ ∂X ν
Gab = ηµν (2.20)
∂ξ a ∂ξ b
is called the induced metric on Σ or the pullback of the ambient space metric ηµν onto
Σ. With this notation the string action is
Z p
SN G = −T − det Gab d2 ξ. (2.21)
Σ
• X µ and ξ a have engineering dimension of a length, which fixes the engineering dimension
of T as
[T ] = [mass2 ] = [length−2 ].
T is called the string tension. One further defines the
1
Regge slope α0 via T = 0
, (2.22)
√ 2πα
string length `s = 2π α0 , (2.23)
1
string (mass) scale Ms = √ . (2.24)
α0
2.2.2 The Polyakov action

As for the point particle it is more convenient to eliminate the the square root in SN G . To this end
we introduce an auxiliary field, the WS metric hab (ξ a ), and define the Brink-DiVecchia-Howe
(BDH) or Polyakov action
√ √
Z Z
T T
SP = − d2 ξ − det h hab (ξ) ∂a X µ (ξ)∂b X ν (ξ)ηµν = − d2 ξ − det h hab Gab . (2.25)
2 Σ 2 Σ
Comments:
• hab (ξ a ) is a (symmetric) 2-tensor on the WS, which plays the role of the intrinsic WS
metric. A priori it is independent of the inherited or pullback metric Gab = ∂a X µ ∂b X ν ηµν .
• X µ (ξ a ) is a scalar on the WS (since it carries no “a, b” index), but a spacetime vector
(as evident due to its“µ” index).
• The metric hab is in general curved and endowes the WS with the Levi-Civita-connection.
The associated covariant derivative ∇a on the WS is defined w.r.t. Γa bc = 21 had (∂b hcd +
∂c hbd − ∂d hab ). Recall that for a scalar field the partial and the covariant derivative agree.
In particular ∇a X µ (ξ) = ∂a X µ (ξ).
• The metric h has signature (−, +). This explains the overall minus sign in SP (compared
to the expression for the point particle).
It is crucial to appreciate that in String Theory the spacetime coordinates X µ of the string
are promoted to dynamical fields2 in the 2-dimensional field theory on the worldsheet defined
by the Polyakov action. In a sense, spacetime becomes a derived concept. The fundamental
object is the field theory on the worldsheet.
Furthermore we consider a 2-dimensional field theory on a curved worldsheet whose metric we
treat dynamically. In other words:
The Polyakov action for propagation of a string in R1,d−1 describes d two-dimensional scalar
fields X µ (ξ) coupled to the dynamical WS metric hab (ξ). Studying (bosonic) string theory is
equivalent to studying 2-dimensional gravity coupled to scalars.
How general is SP ?
One could add two types of terms to SP :

R √
i) A 2-dimensional “cosmological constant” term S1 = λ1 Σ − det h. This would spoil the
crucial consistency condition of conformal invariance discussed in detail later. Thus we
require λ1 = 0.
λ2
R √
ii) A 2-dimensional Einstein-Hilbert term S2 = 4π Σ
− det h R(2) with R(2) the Ricci scalar
a
of the WS connection associated with Γ bc . In 2-dimensions this term has no dynamics as
it is given by a total derivative. This corresponds to the fact that 2-dimensional
R √ gravity is
1
dynamically trivial (see the exercises for details). In fact: χ(Σ) ≡ 4π Σ
− det h R(2) is a
topological invariant called Euler characteristic - it is an integer invariant under contin-
uous deformations of the WS metric. While it has no local dynamics it will weigh different
WS topologies differently in the path integral. We ignore this term for now and revisit it
when setting up the perturbation series in the path integral approach to quantisation.
For completeness, let us stress that the above expression for χ(Σ) refers to closed strings. For open
strings the WS has a boundary ∂Σ and the expression for the Euler characteristic is
√
Z Z
1 1
χ= d2 ξ −hR + ds K. (2.26)
4π Σ 2π ∂Σ
Here the extrinisic curvature K is defined as
K = ±ta nb ∇a tb (2.27)
with ta a unit vector tangent to the boundary and na an outward unit vector orthogonal to ta . The
upper/lower sign refer to timelike/spacelike boundaries.
The energy momentum tensor
An important quantity in field theory in the energy momentum tensor.

2 Formally, the same holds of course the for the spacetime coordinates of a point particle. The resulting 1-
dimensional “field theory”, however, is trivial and does not add any new perspective. This will be seen to be
completely different in 2 dimensions.
• The energy momentum tensor Tab measures the response of SP w.r.t. a metric variation
δhab , while keeping all other fields unchanged. Our conventions for its normalisation are3
√
Z
1
δSP = d2 ξ −h Tab δhab , (2.28)
4π
4π δSP
i.e. Tab = √ (h := det h). (2.29)
−h δhab
Using
δh = −h hab (δhab ), (2.30)
an identity proven in the exercises, one finds

1 1
Tab = − 0 Gab − hab hcd Gcd , (2.31)
α 2
where again Gab = ∂a X · ∂b X.
• The equations of motion for hab are Tab = 0, i.e.

1
Gab = (h · G) hab . (2.32)
2
Thus on-shell, the pullback of the ambient space metric, Gab , and the induced metric hab
are proportional.
• As the for the point particle one finds that on-shell for hab , SP and SN G agree,

SP [X, hab ] = SN G . (2.33)

Tab =0
This will be proven in the exercises.
• Properties of Tab
i) Tab is traceless,
1 1
hab (Gab − hab (h · G)) = 0.
T aa = − (2.34)
α0 2
This holds without use of the e.o.m. for X µ . Note that the two terms in T a a cancel
because the WS is 2-dimensional and thus hab hab = 2. The special significance of the
dimensionality of the WS will become clearer momentarily.
ii) Tab is conserved,
∇a Tab = 0. (2.35)
This holds upon use of e.o.m. for X . The fact that we need the e.o.m. for X µ is
µ
a consequence of the specific definition of Tab as describing response to a variation

δS
only of hab , not of X. Thus δX ≡ 0 must be assumed to show conservation.
Both properties are rooted in the
3 Beware that the normalisation differs from textbook to textbook, and in some context also the variation of
fields other than the metric is included.

2.2.3 Symmetries of the Polyakov action

Corresponding to the fact that we have a dual interpretation of the X µ - either as the string
coordinates in spacetime or as dynamical scalar fields on the 2-dimensional WS - we must carefully
distinguish between the
• spacetime symmetries acting on R1,d−1 and the
• WS symmetries acting on Σ.
i) Spacetime symmetries
The Polyakov action enjoys manifest d-dimensional Poincaré-invariance,
X µ (ξ) 7→ Λµ ν X ν (ξ) + V µ , Λ ∈ SO(1, d − 1). (2.36)
Note that Poincaré-invariance can be interpreted as a global internal symmetry from the
perspective of the 2-dim. field theory.
ii) WS symmetries
1. Local diffeomorphism invariance under
ξ a 7→ ξã (ξ) = ξ a − a (ξ). (2.37)
The various fields transform according to their tensorial nature:
• X µ (ξ) is a scalar field from the WS perspective and thus transforms as
˜ = X µ (ξ(ξ))
X µ (ξ) 7→ X̃ µ (ξ) ˜ = X µ (ξ) + c ∂c X µ (ξ) (2.38)
µ µ 2
≡ X + δX + O( ), (2.39)
δX µ = c ∂c X µ . (2.40)
• The metric hab (ξ) transforms like a WS 2-tensor, i.e.
c d
˜ = ∂ξ ∂ξ hcd (ξ(ξ))
hab (ξ) 7→ h̃ab (ξ) ˜ = hab + δhab + O(2 ), (2.41)
∂ ξã ∂ ξ˜b
δhab = c ∂c hab + (∂a c )hcb + (∂b c )hac , (2.42)

= ∇a b + ∇b a . (2.43)
√
• The object − det h turns out to transform like a scalar density of weight 1,
√ √
δ − det h = ∂c (c − det h). (2.44)
More details are presented in the exercises.
2. Local Weyl invariance,
δX µ = 0, (2.45)
2
hab → exp(2 Λ(ξ)) hab = hab + δhab + O(Λ ), (2.46)
δhab = 2 Λ(ξ)hab . (2.47)
Crucially this extra symmetry is special to the fact that the WS is 2-dimensional. To
appreciate this, let us pretend the WS were D-dimensional and compute
√ D√
− det h 7→ (e2Λ ) 2 − det h, (2.48)
hab 7→ (e2Λ )−1 hab . (2.49)
√
dD ξ − det h hab Gab is invariant iff D = 2.
R
It follows that
Comments
• Weyl invariance will become pivotal forR a√consistent quantisation in the sequel. Note
that the cosmological constant term λ − det h spoils Weyl invariance at the clas-
sical level. Since our whole quantisation scheme depends on Weyl invariance we
therefore discard this term for now.
• The appearance of Weyl invariance for 2-dimensional worldhseets identifies String
Theory as a very special generalisation of the point particle theory.
Indeed one might wonder about even higher dimensional generalisations, e.g. to worldsheets
of dimensions D = 3 corresponding to so-called membranes, or D > 3. It turns out that
already the case D = 3 cannot be treated with the same methods that are successful in the
string case. Part of the reason is the lack of conformal invariance. In addition, the gauge
fixed action is no longer free for D ≥ 3. A detailed acount of these complications can be
found in the review by Helling and Nicolai, http://arXiv.org/pdf/hep-th/9809103. At this
stage it is not known how to consistently quantise membranes perturbatively. Nonetheless
they play a crucial role as non-perturbative objects in M-theory.
Finally note that under combined diffeomorphism and Weyl rescaling the metric transforms as
δhab = ∇a b + ∇b a + 2Λhab (2.50)
1
= ∇a b + ∇b a − ∇c c hab + 2(Λ + ∇a a )hab , (2.51)
2
δhab ≡ (P · )(ab) + 2Λ̃hab . (2.52)
The linear operator P maps vectors to symmetric traceless 2-tensors. For transformations a
such that (P · )ab = 0 the effect on hab can be undone by Weyl rescaling. The corresponding a
are called conformal Killing vectors.
The presence of the above spacetime and WS symmetries leads to conserved quantities via
Noether’s theorem, as revised in the exercises.
i) Spacetime invariance
• Global Poincaré-invariance of spacetime,
X µ (ξ) 7→ X µ (ξ) + V µ , (2.53)
implies conservation of the energy-momentum current
√
Pµa = −T − det h hab ∂b Xµ , (2.54)
∂a Pµa = ∇a Pµa = 0. (2.55)
In the last step we used that Pµa is a vector density of weight 1 from the WS perspective
and thus ∂a Pµa = ∇a Pµa .
• Invariance under global Lorentz transformation, X µ 7→ Λµ ν X ν (ξ),

implies conservation of the angular momentum current
a
√
Jµν = −T −hhab (Xµ ∂b Xν − Xν ∂b Xµ ) (2.56)
= Xµ Pνa − Xν Pµa , (2.57)
a
∇a Jµν = 0. (2.58)
ii) Worldsheet symmetries
• Tab is the conserved current w.r.t. local WS diffeomorphism invariance,
ξ 7→ ξ˜ = ξ − (ξ)

⇒ ∇a Tab = 0 on-shell for X. (2.59)
hab 7→ hab + ∇a b + ∇b a
This is easy to prove:

√ √
Z Z
1 1
0= d2 ξ −h Tab δhab = d2 ξ −h Tab ∇a b (2.60)
4π 2π
1
Z √ 1
Z
√
= d2 ξ ∇a −h Tab b − d2 ξ ∇a ( −hTab ) b . (2.61)
2π 2π
√
Since ∇a −h = 0 for the Levi-Civita metric we conclude ∇a Tab = 0.
Note again that this holds only on-shell for X µ because by definition Tab only measures
effect of diffeomorphism on hab while δX = 0.
• Tracelessness T a a = 0 is a consequence of Weyl invariance as proven in the exercises.
Important generalisation
Every conformal Killing vector, i.e. every a satisfying (P · )ab = 0, yields a conserved current
Ja = T ab b with ∇a Ja = 0. (2.62)
The proof goes as follows: Since T ab is conserved we have
∇a J a = ∇a (T ab b ) = T ab ∇a b . (2.63)
Now we solve the conformal Killing equation 0 = (P · )ab for ∇a b + ∇b a = hab (∇c c ), which
we plug in, using symmetry of T ab ,
1 ab 1
∇a J a = T (∇a b + ∇b a ) = T ab hab (∇c c ) = 0. (2.64)
2 2
In the last equation tracelessness T a a = 0, which is a consequence of Weyl invariance, is crucial.
2.2.4 Flat worldsheet coordinates

On a 2-dimensional worldsheet the local diffeomorphism and Weyl invariance can be used to “fix
the metric” hab by locally gauging away all its parameters.
• That this is possible can be anticipated already by counting degrees of freedom. For a
D-dimensional worldsheet this counting goes as follows:
hab : 21 D(D + 1) d.o.f. D

2 (d + 1) − (D + 1) d.o.f. can-
diffeomorphism + Weyl : D + 1 d.o.f not be gauged away locally.
Thus we see that precisely if D = 2 we have a chance to locally gauge away all parameters.
• In fact, in the exercises we will show that under a Weyl transformation hab 7→ exp(2ω(ξ))hab
the Ricci scalar on the worldsheet transforms as
√ √
−h R 7→ −h (R − 2∇2 ω). (2.65)
By solving this differential equation for ω we can achieve that locally R = 0. In two
dimensions this implies that also the Riemann tensor vanishes locally, as discussed in the
exercises. Thus, locally the metric can be made flat R = 0. The remaining diffeomorphism
invariance can now be used to bring this flat metric into the form

−1 0
hab = ηab = . (2.66)
0 1
• This is true locally. There may exist topological obstructions to setting hab = ηab globally
on the worldsheet. In this case there remain parameters in the metric, so-called moduli,
which cannot be removed by a conformal rescaling and diffeomorphisms. Such effects will
be discussed later; for now only local considerations are relevant.
Sometimes it is useful to work in conformal gauge

hab = Ω2 (σ, τ ) ηab . (2.67)
For now we work in flat gauge by setting
hab = ηab (2.68)
In flat gauge the Polyakov action reduces to the action of a set of free scalar fields,
Z
T
SP = dτ dσ (∂τ X)2 − (∂σ X)2 . (2.69)
2 Σ
A convenient set of coordinates are the lightcone coordinates:
• These are defined as
ξ+ = τ + σ , ξ − = τ − σ. (2.70)
• The metric in light cone coordinates follows from
ds2 = −dτ 2 + dσ 2 = −dξ + dξ − (2.71)
as
1
h++ = h−− = 0 ; h+− = h−+ = − ; h+− = h−+ = −2. (2.72)
2
For example, raising and lowering indices with this metric implies reolation of the type
1
V+ = h+− V − = − V − . (2.73)
2
• The partial derivatives follow as
∂ 1
∂± = = (∂τ ± ∂σ ). (2.74)
∂ξ ± 2
• The measure in lightcone gauge reads
∂(σ, τ ) 1
dτ dσ = dξ + dξ − det = dξ + dξ − . (2.75)
∂(ξ + , ξ − ) 2
The Polyakov action in lightcone gauge

• The Polyakov action takes the form
Z
SP = T d2 ξ ± ∂+ X µ ∂− X ν ηµν . (2.76)
• The energy momentum tensor

1 1
Tab = − (∂a X · ∂b X − hab hcd ∂c X · ∂d X) (2.77)
α0 2
has the following properties in lightcone gauge:
– T+− = T−+ = 0, which is the statement of tracelessness Tab hab = 0.
– The non-vanishing components are
1 1
T++ = − ∂+ X · ∂+ X, T−− = − ∂− X · ∂− X. (2.78)
α0 α0
– Conservation ∇a Tab = 0 implies ∂− T++ = ∂+ T−− = 0 such that
T++ = T++ (ξ + ) and T−− = T−− (ξ − )
.
Two crucial remarks
• Before going to flat gauge, the e.o.m. for hab gave Tab = 0. This condition has to be
implemented as a constraint in flat coordinates. In lightcone gauge the constraints
therefore take the form
T++ = 0 = T−− .
We will see that that a proper treatment of these constraints will save the day when it
comes to quantising the string.
• Even after gauge fixing there is left a large residual symmetry in that setting hab = ηab
does not completely remove the gauge freedom. The generators of this residual gauge
symmetry are the conformal Killing vectors
(P )ab = 0 ⇔ ∇a b + ∇b a = hab (∇c c ),
whose effect on the metric can be undone by a Weyl rescaling.

In flat gauge the conformal Killing vector equation takes the form ∂+ + = ∂− − = 0 and
thus
∂+ − = 0 − ≡ − (ξ − )
+ ⇒ (2.79)
∂− = 0 + ≡ + (ξ + )
Conformal Killing transformations (CKT)

Since the last observation is so important in what follows, let us expand on it once more: A
transformation
ξ + 7→ ξ˜+ (ξ + ) = ξ + − + (ξ + ) , ξ − 7→ ξ˜− (ξ − ) = ξ − − − (ξ − ) (2.80)
can be undone by a Weyl transformation and is not fixed by the gauge hab = ηab .
We already saw before that the Conformal Killing Vector transformations imply infinitely many
conserved currents. In lightcone gauge this fact manifests itself as follows,
∂− (T++ (ξ + )+ (ξ + )) = 0
∇a (Tab b ) = 0 ⇒ (2.81)
∂+ (T−− (ξ − )− (ξ − )) = 0
The conserved charges associated with these currents are4

Z
`
L+ = dσ + (ξ + )T++ (ξ + ), (2.82)
4π 2
Z
`
L− = dσ − (ξ − )T−− (ξ − ). (2.83)
4π 2
2.2.5 Hamiltonian formalism

So far we have been working in a lagrangian formulation with action, in flat gauge,
Z Z
T ab
SP = − dτ dσ ∂a X · ∂b Xη = dτ dσL. (2.84)
2
• We can move to a Hamiltonian formulation with canonical fields X µ (τ, σ) and conjugate
momenta
∂L
Πµ (τ, σ) = = T Ẋ µ (τ, σ)
∂ Ẋµ (τ, σ)
by defining the Hamiltonian
Z ` Z `
µ
H= dσ(Ẋ (τ, σ)Πµ (τ, σ) − L) = T dσ[(∂+ X)2 + (∂− X)2 ]. (2.85)
0 0
• To define the symplectic structure via the Poisson brackets one takes all fields at equal
time τ . For fields F (τ, σ), G(τ, σ 0 ) the Poisson brackets are then defined as
∂F (τ, σ) ∂G(τ, σ 0 ) ∂G(τ, σ 0 ) ∂F (τ, σ)

Z
{F, G} := dσ̃ − . (2.86)
∂X µ (τ, σ̃) ∂Πµ (τ, σ̃) ∂X µ (τ, σ̃) ∂Πµ (τ, σ̃)
• This leads to the canonical equal time Poisson bracket relations
{X µ (τ, σ), Πν (τ, σ 0 )} = η µν δ(σ − σ 0 ) (2.87)

{X, X} = 0 = {Π, Π}. (2.88)
• The Poisson brackets of the conserved charges generate the associated sym-
metries. E.g. for the residual conformal symmetries ξ + 7→ ξ˜+ (ξ + ) = ξ + + f (ξ + ) and
ξ − 7→ ξ˜− (ξ − ) = ξ − + g(ξ − ):
`
{Lf , X(τ, σ)} = − f (ξ + )∂+ X(τ, σ) (2.89)
2π
`
{Lg , X(τ, σ)} = − g(ξ − )∂− X(τ, σ) (2.90)
2π
Rl
where Lf = − 4π` 2 0
dσf (ξ + )T++ (ξ + ) etc.. This will be proven in the exercises.
4 Our conventions are that if we define ξ ± → ξ ± − ± (ξ ± ), L± is defined with a plus as given here.
2.3 Oscillator expansions

2.3.1 Equations of motion and boundary conditions
It is time to analyse the equations of motion of the string fields X µ (τ, σ). We noted already that
in flat gauge the action
Z Z
T
SP = 2T dτ dσ ∂+ X · ∂− X = dτ dσ (∂τ X)2 − (∂σ X)2
2
takes the form of a theory for d free scalars X µ (τ, σ).

To obtain the e.o.m. for X µ we vary SP such that δX µ (τ = −∞) and δX µ (τ = ∞) vanish. This
gives
Z
T
δSP = dτ dσ [2∂τ X · ∂τ δX − 2∂σ X · ∂σ δX] (2.91)
2
Z
= −T dτ dσ [(∂τ2 − ∂σ2 )X · δX] (2.92)
Z Z
+ T dσ ∂τ X · δX|∞ τ =−∞ − T dτ ∂σ X · δX|`σ=−` . (2.93)
• The τ -boundary term vanishes by the assumption δX|∞

τ =−∞ = 0.
• If also σ-boundary term vanishes, then the e.o.m take the form of a free wave equation
(∂τ2 − ∂σ2 )X µ = 0, ⇔ ∂+ ∂− X µ = 0. (2.94)
We must now impose boundary conditions such that indeed the boundary terms vanish.5 The
consistent boundary conditions are classified as follows:
1. Closed string
By definition, a closed string propagating in R1,d−1 is subject to periodic boundary condi-
tions
X µ (τ, σ = 0) = X µ (τ, σ = `). (2.95)
Thus, the boundary terms at σ = 0 and σ = ` cancel each other.
More generally, cancellation of the boundary terms only requires
X µ (τ, σ = 0) = M µν X ν (τ, σ = `) , (2.96)
with M µν an O(1, d − 1) matrix.

However, if we want to interpret X µ (τ, σ) as string embedding coordinates and preserve Poincaré
invariance in R1,d−1 , then we need (2.95). For more general M µν ∈ O(1, d − 1) the string target
space is an orbifold, a space describable as flat space modded out by the action of a group.
2. Open string
The two string endpoints are independent. Thus the boundary terms at σ = 0 and σ = l
have to vanish separately. For each X µ and for each σ = 0, l we can impose either
a) Neumann boundary conditions: ∂σ X µ |σ=0 and/or σ=l = 0 (2.97)

5 Otherwise it would not be possible to define a local 2-dimensional worldsheet theory as the e.o.m would involve
non-local boundary contributions in addition to the second-order PDE characterising the free wave equation.
b) Dirichlet boundary conditions: δX µ |σ=0 and/or σ=l = 0. (2.98)

µ
Dirichlet boundary conditions for X correspond to the string endpoint being fixed in the
µ−direction. For each µ the possible combinations of boundary conditions at the endpoints
are therefore (NN), (DD), (ND) or (DN).
2.3.2 Closed string expansion

We now give the most general solution of the string equations of motion subject to periodic
boundary conditions. The free wave equation
∂+ ∂− X µ = 0 (2.99)
µ
implies that X is the sum of a left- and a right-moving wave along the string,
X µ = XLµ (ξ + ) + XR
µ −
(ξ ), (2.100)
such that X µ (τ, σ = 0) = X ν (τ, σ = `).
The most general solution is given by the Fourier expansion
r
µ 1 µ µ 1 2πα0 µ α0 X 1 µ − 2π in(τ −σ)
XR (τ − σ) = (x + c ) + p (τ − σ) + i α e ` , (2.101)
2 2 ` 2 n n
n∈Z, n6=0
r
1 µ 1 2πα0 µ α0 X 1 µ − 2π in(τ +σ)
XLµ (τ µ
+ σ) = (x − c ) + p (τ + σ) + i α̃ e ` , (2.102)
2 2 ` 2 n n
n∈Z, n6=0
where
• αnµ , α̃nµ represent independent right-/left-moving Fourier modes with the convention that
the positive frequency modes correspond to n < 0 and
• we take cµ = 0 for the time being.
• The left-/right-moving zero mode piece pµ is coupled by the boundary conditions.
∗
• Reality X µ (τ, σ) = [X µ (τ, σ)] implies
µ µ
xµ = (xµ )∗ ; pµ = (pµ )∗ ; (αnµ )∗ = α−n ; (α̃nµ )∗ = α̃−n . (2.103)
With the definition

r
α0 µ
α0µ = α̃0µ = p (2.104)
2
one obtains
r
2π α0 X µ − 2π in(τ −σ)
µ
∂− X = ẊLµ = αn e l , (2.105)
l 2
n∈Z
r
µ µ 2π α0 X µ − 2π in(τ +σ)
∂+ X = ẊR = α̃n e l . (2.106)
l 2
n∈Z
Later we will understand that it is the derivatives ∂± X µ rather than X µ themselves which
represent good fields on the worldsheet.
The above prefactors were chosen such that the following interpretation of the coefficients can
be given:
• xµ is the center-of-mass position of the string at τ = 0 because

`
2πα0 µ
Z
1
qµ = dσX µ = xµ + p τ. (2.107)
` 0 `
This already implies that pµ is the total spacetime momentum of the string.
• Indeed this is backed up by noting that
Z ` Z `
1
dσΠµ (τ, σ) = dσ Ẋ µ = pµ . (2.108)
0 2πα0 0
• The total angular momentum is

Z `
J µν = dσ (X µ Πν − X ν Πµ ) = lµν + E µν + Ẽ µν (2.109)
0
with a center-of-mass contribution
lµν = xµ pν − xν pµ (2.110)
and an oscillator contribution

∞
X 1 µ ν
E µν ν
αnµ .

= −i α−n αn − α−n (2.111)
n=1
n
The most general solution thus corresponds to a string moving with center-of-mass momentum pµ
while carrying a general superposition of left-and right-moving vibrational modes with excitation
α̃nµ , αnµ . Note that the momentum and the oscillation modes are further subject to the primary
constraints T++ = 0 = T−− , as will be detailed further in the next chapter.
Of prime importance for quantisation of the theory are the Poisson brackets.
From {X µ (τ, σ), Πν (τ, σ 0 )} = η µν δ(σ − σ 0 ) we find by explicit computation
µ
{αm , αnν } = −i m δm+n,0 η µν , (2.112)
µ
{α̃m , α̃nν } = −i m δm+n,0 η µν , (2.113)
µ
{αm , α̃nν } = 0, µ
{x , p } = η ν µν
. (2.114)
The Hamiltonian enjoys the oscillator expansion

Z `
dσ (∂+ X)2 + (∂− X)2

H = T
0
πX
= (α−n · αn + α̃−n · α̃n ) . (2.115)
`
n∈Z
2.3.3 Open string expansion

For the open string we can impose, for each X µ independently, Neumann-Neumann (NN) bound-
ary conditions, Dirichlet-Dirichlet (DD) boundary conditions or mixed (DN) or (ND) boundary
conditions. Let us discuss these in turn.
1. Neumann boundary conditions at both ends σ = 0 and σ = `, i.e.
X 0µ (τ, σ) = 0 for σ = 0, `, (2.116)
lead to the most general solution
2πα0 µ √ X 1 µ −i π nτ nπσ
(NN) X µ (τ, σ) = xµ + p τ + i 2α0 αn e ` cos . (2.117)
` n `
n∈Z, n6=0
Note that the boundary conditions relate XL and XR such that independence of the left-
and right-moving oscillators is lost, αn ≡ α̃n .
Defining
√
α0µ = 2α0 pµ (2.118)
allows us to write
r
µπ α0 X µ −i π n(τ ±σ)
∂± X = αn e ` . (2.119)
` 2
n∈Z
Interpretation:
• The (NN) string can move freely in that the endpoints at σ = 0, ` are free, but the
(NN) boundary conditions implement that there is no momentum flow off the string,
√
Pµα = −T −h∂ α Xµ ⇒ Pµσ |σ=0,` = −T (X 0 )µ |σ=0,` = 0.
• xµ , pµ are again the c.o.m. position

and momentum and the angular momentum is
µν 1
Rl µ ν ν µ
J = 2πα0 0 dσ X Ẋ − X Ẋ = lµν + E µν as before.
• The Poisson brackets are {xµ , pν } = η µν , µ
{αm , αnν } = −imδm+n,0 η µν .
µ
• Reality implies (αnµ )∗ = α−n .
• The Hamiltonian can be expanded as
π X
H= (α−n · αn ).
2l
n∈Z
2. Dirichlet boundary conditions at both ends, δX µ = 0 , σ = 0, `, imply that the string

endpoints are fixed,
Ẋ µ = 0 , σ = 0, `. (2.120)
Consequently we define
X µ (τ, σ = 0) = xµ0 , X µ (τ, σ = `) = xµ1 , (2.121)
while the momentum vanishes, pµ = 0. The general solution is of the form

1 √ X 1 µ −i π nτ nπσ
(DD) X µ (τ, σ) = xµ0 + (xµ1 − xµ0 )σ + 2α0 αn e l sin . (2.122)
` n `
n∈Z, n6=0
With
1 1 µ
α0µ = √ (x1 − xµ0 ) (2.123)
2α0 π
we obtain
r
µ π α0 X µ −i π n(τ ±σ)
∂± X = ± αn e ` . (2.124)
l 2
n∈Z
Comments:
• The c.o.m. position of the (DD) string is

Z `
1 1 µ
µ
q = dσX µ (τ, σ) = (x + xµ1 ). (2.125)
` 0 2 0
• The Hamiltonian acquires an extra contribution from the potential energy of the
stretched string,
T µ π X
H= (x1 − xµ0 )2 + (α−n · αn ). (2.126)
2l 2l
n∈Z
This is a consequence of the fact that a string has tension - stretching it costs energy.
3. Finally, we can also impose mixed (ND) boundary conditions, e.g.
X 0µ = 0 at σ = 0 , Ẋ µ = 0 at σ = `, (2.127)
(or the other way round). Such boundary conditions enforce the solution
√ X 1 π
nπσ
(ND) X µ (τ, σ) = xµ + i 2α0 αnµ e−i ` nτ cos . (2.128)
1
n `
n∈Z+ 2
Comments:
• The Fourier modes are now half-integer, n ∈ Z + 12 .
• The string is fixed on one end, here at σ = ` at position xµ .
• Consequently the c.o.m. momentum vanishes.
2.3.4 The concept of D-branes

Having achieved the general open string solutions corresponding to the various boundary condi-
tions we now give a deeper physical interpretation. As stated several times, in each dimension µ
we can choose (NN), (DD) or (ND) boundary conditions independently.
As a first example we consider the following configuration of boundary conditions:
• (NN) for X µ , µ = 0, . . . , p,
• (DD) for X µ , µ = p + 1, . . . , d − 1.
Suppose further that in the (DD) directions xµ0 = xµ1 . Then the open string moves freely along
the (p+1)-dimensional hypersurface extending along µ = 0, . . . , p and is fixed in the directions
normal to it.
This leads us to the important concept of a D-brane:
A Dp-brane is a (p+1)-dim. hypersurface of spacetime on which open strings can end.
At this stage of our analysis a Dp-brane is really just what is written in the previous sentence. In
fact, however, a Dp-brane is by itself a dynamical object. While it will take more preparation
to make this precise at a quantitative level, we can already guess this from the observation
that there is non-zero momentum flow off the string in the (DD) directions normal to it. Total
momentum conservation implies momentum exchange between the string and the Dp-
brane.
One might object that momentum conservation might not hold because the presence of the D-brane
Dp-brane, or equivalently the choice of boundary conditions, partially breaks translational invariance.
However, it turns out that a D-brane only breaks translational invariance spontaneously. A spontaneous
breakdown of a continuous symmetry does not affect the assocated conservation laws.
The realisation by Joseph Polchinski in 1996 that the Dp-brane itself carries non-trivial dynamics
has revolutionised the way we think about string theory. We will discuss the dynamics of D-
branes later in the course. Suffice it here to state that the dynamics of the D-brane is described
by an analogue of the Nambu-Goto action. We will find that the tension scales like g1s with gs
the string coupling. This means that D-branes are non-perturbative objects.
More general brane configurations include

• parallel branes, e.g.
with
• (NN) for µ = 0, . . . , p
• (DD) for µ = p + 1, . . . , d − 1 with
xµ0 6= xµ1 ;
• branes of different dimension, e.g. a Dp-brane and a Dq-brane corresponding to a string

subject to (DN)-boundary conditions (see Ass. sheet 3 for an example);
• branes at angles.
We furthermore identify (NN) boundary conditions in all d dimensions as the special case of a
D(d − 1)-brane, i.e. a D-brane filling all of R1,d−1 .
2.3.5 The Virasoro generators

With the mode expansion at hand we are finally in a position to introduce the Virasoro generators,
which will be key players in what follows.
Consider first the closed string and recall that the conformal Killing transformations
ξ + 7→ ξ + + f (ξ + ), ξ − 7→ ξ − + g(ξ − ) (2.129)
are generated by the Noether charges

Z ` Z l
` `
Lf = − 2 T++ (ξ + )f (ξ + )dσ, Lg = − T−− (ξ − )g(ξ − )dσ (2.130)
4π 0 4π 2 0
in the sense that

` `
{Lf , X µ }P.B. = − f (ξ + ) ∂+ X µ , {Lg , X µ }P.B. = − g(ξ − ) ∂− X µ . (2.131)
2π 2π
±
Now we can pick a basis exp i m 2π for the functions g(ξ − ), f (ξ + ) and define

` ξ
Z `
` 2π −
Lm := − 2 dσ T−− exp i m ξ (2.132)
4π 0 `
Z `
` 2π
≡ − 2 dσ T−− exp −i m σ
4π 0 `
1X
= αm−n · αn , (2.133)
2 n
Z l
` 2π
L̃m = − 2 dσ T++ exp i m σ
4π 0 `
1X
= α̃m−n · α̃n . (2.134)
2 n
Note that in going from the first to the second line in (2.132) we were using that the Noether
charges are conserved and thus time independent. Furthermore L∗m = L−m , L̃∗m = L̃−m .
The Lm and L̃m are called Virasoro generators. They are the generators of the conformal
Killing transformations in the sense that

` 2π − n o ` 2π +
{Lm , X µ }P.B. = − exp i m ξ ∂− X µ L̃m , X µ =− exp i m ξ ∂+ X µ .
2π ` P.B. 2π `
(2.135)
Explicit computation shows that the Virasoro generators satisfy the classical Witt algebra
{Lm , Ln }P.B. = −i(m − n)Lm+n , (2.136)

n o
L̃m , L̃n = −i(m − n)L̃m+n . (2.137)
P.B.
To fully appreciate the geometric significance of the Witt algebra, we note that the conformal
Killing transformations satisfy an algebra generated by the differential operators

` 2π − ` 2π +
Tm = − exp i m ξ ∂− , T̃m = − exp i m ξ ∂+ . (2.138)
2π ` 2π `
It is easy to check that these classical geometric operators satisfy the commutation relations
[Tm , Tn ] = i(m − n)Tm+n . (2.139)
As established above, the Lm / L̃m are the Noether charges associated with the transformations
Tm / T̃m in that
{Lm , X}P.B. = Tm X , Lm ≡ QTm . (2.140)
The Witt algebra relation can thus be written very suggestively as
{QTm , QTn }P.B. = −Q[Tm ,Tn ] . (2.141)
Indeed this connection between the Poisson bracket relations of the Noether charges of a sym-
metry and the Lie algebra relations of its generators holds very generally.
Of specialP
significance are the Virasoro generators L0 and L̃0 . This is because the Hamiltonian
∞
H = 21 2π
` n=−∞ (α−n · αn + α̃−n · α̃n ) is related to the Virasoro generators as
2π
H= L0 + L̃0 . (2.142)
`
This connection is no wonder because the Hamiltonian generates local time reparametrisations,
i.e. local translations in the coordinate τ . This is precisely what L0 + L̃0 does.
This brings us to the Virasoro constraints. Recall that on each string solution we must impose
!
the constraints Tab = 0 which arose as the equations of motion of the worldsheet metric. Since
the Virasoro generators are nothing but the Fourier modes of the energy-momentum tensor, the
constraints can be written as
! !
Lm = 0, L̃m = 0 ∀m. (2.143)
This has to be understood as a set of constraints on the choice of string oscillation modes αm , α̃m
making up the string solution.
!
In particular, the Hamiltonian must vanish, H = 0. Geometrically this constraint can be viewed
as implementing time reparametrization invariance. Since p2 = α20 12 (α02 + α̃02 ) we can solve H = 0
to obtain the important spacetime mass shell relation
∞
H=0 2 X
M 2 = −p2 = (α−n · αn + α̃−n · α̃n ) . (2.144)
α0 n=1
Note that the complementary ”level zero” Virasoro constraint (L0 − L̃0 ) = 0 implements diffeo-
morphism invariance along the worldsheet.
For the open string the analogous story goes through, except that T++ and T−− are not
independent as a consequence of the boundary conditions. The Virasoro generators are now
defined as
Z `
` π π
dσ ei m ` σ T++ + e−i m ` σ T−−

Lm := − 2 (2.145)
2π 0
1X
= αm−n · αn (2.146)
2
n∈Z
and again satisfy
{Lm , Ln } = −i (m − n) Lm+n . (2.147)
The vanishing of the Hamiltonian

π
H= L0 (2.148)
`
as one of the Virasoro constraints Lm = 0 implies the open string mass shell relation
∞
1 X
M 2 = −p2 = α−n · αn . (2.149)
α0 n=1
This is correct for open strings with (NN) boundary conditions in all dimensions. We will discuss
more general cases in the context of the quantum string.
Chapter 3
Quantisation of the bosonic string
We will discuss 3 alternative ways to quantise the bosonic string, each shedding different light
on the significance of the critical number of spacetime dimensions as a consistency condition for
existence of a well-defined quantum theory.
• Old covariant quantisation (OCQ):
The Virasoro constraints are implemented at a quantum level. This procedure is manifestly
Lorentz invariant, but unitarity only holds in the ”critical” number of spacetime dimensions.
• Light-cone quantisation (LCQ):

The Virasoro constraints are implemented classically before quantisation. This leads to a
manifestly unitary quantisation scheme, but Lorentz invariance holds only in the ”critical”
dimension.
• Path-integral quantisation:
This method uses the celebrated Faddeev-Popov gauge fixing procedure as appropriate for
gauge theories. Criticality is equivalent to closure of the BRST algebra.
3.1 Old canonical quantisation

3.1.1 Canonical quantisation
In canonical quantisation the classical fields X µ (τ, σ) and their canonical momenta Πµ (τ, σ) are
promoted to operators X̂ µ (τ, σ), Π̂µ (τ, σ) and we replace
1
{ , }P.B. −→ [ , ]. (3.1)
i
This leads to the canonical equal time commutation relations
h i
X̂ µ (τ, σ), Π̂ν (τ, σ 0 ) = i η µν δ(σ − σ 0 ), (3.2)
h i h i
X̂ µ (τ, σ), X̂ ν (τ, σ 0 ) = 0 = Π̂µ (τ, σ), Π̂ν (τ, σ 0 ) . (3.3)
Also the oscillator modes are promoted to operators - we will do without the little pretentiousˆ
31
CHAPTER 3. QUANTISATION OF THE BOSONIC STRING 32
- with commutation relations
[xµ , pν ] = i η µν , (3.4)
µ
[αm , αnν ] = m δm+n,0 η µν
= µ
[α̃m , α̃nν ] , (3.5)
µ
[α̃m , αnν ] = 0. (3.6)
Hermiticity X̂ µ = (X̂ µ )† , Π̂µ = (Π̂µ )† implies

µ
µ
αm = (α−m )† . (3.7)
As always this quantisation procedure is not completely well-defined because it not clear a priori
what is the correct order of products of non-commuting operators. This is known to lead to
normal ordering ambiguities, and string theory is no exception here. To study these ambiguities
we define normal ordering as
µ ν
µ ν αm αn , if m ≤ n
: αm αn := µ (3.8)
αnν αm , if n < m.
In particular we define the Virasoro operators in normal-ordered form,
1X
Lm = : αm−n · αn : (3.9)
2
n∈Z
1X
L̃m = : α̃m−n · α̃n : (3.10)
2
n∈Z
Note that an actualy ambiguity arises only for L0 , L̃0 . The above definition of the normal ordered
form of L0 , L̃0 means that
∞
1 2 X
L0 = α0 + α−n · αn (3.11)
2 n=1
and correspondingly for L̃0 . If we quantise classical expressions involving L0 , we introduce a

normal ordering constant a,
L0 −→ L0 − a, (3.12)
to be fixed later. Indeed we will see that consistency of quantisation scheme will soon force a
definite value of a upon us, with remarkable consequences for the theory.
3.1.2 The quantum Virasoro algebra

Classically the Virasoro generators satisfy the Witt algebra {Lm , Ln }P.B. = −i(m − n)Lm+n .
Naively we would expect this to translate into straightforward commutation relations for the
quantum operators following the procedure (3.1). It turns out, though, that special care has to
be applied to compute [Lm , Ln ] in the quantum case due to normal ordering issues. As a result of
a lengthy computation that starts from the definition of the Lm and the commutation relations
µ
for the αm , presented in detail on examples sheet 4, the Virasoro generators satisfy the quantum
Virasoro Algebra
c
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n,0 ≡ (m − n)Lm+n + ∆(m, n). (3.13)
12
Comments:
• c is called central charge. Direct computation shows that it is given by the number of
spacetime dimensions, i.e. the number of scalar fields X µ
c = η µµ = d. (3.14)
• The Virasoro algebra is a central extension by C of the (classical) Witt algebra.
Loosely speaking, given a Lie algebra g, its central extension by C, denoted as ĝ, is the Lie algebra
spanned by the elements of g as well as a new element c with commutation relations
[x, y]ĝ = [x, y]g + c · p(x, y) , x, y ∈ g, (3.15)

[x, c]ĝ = 0 = [c, c]ĝ . (3.16)
That is, we extend g by a new element c commuting with g, i.e. in the center of g. The map
p : g × g → C is bilinear and antisymmetric.
• Only the term m3 -term in ∆(m, n) carries physical information. Namely, under a redefin-
tion L0 → L0 − α the central term ∆(m, n) transforms as
∆(m, n) → ∆(m, n) + 2α · m. (3.17)
• Important: The fact that c 6= 0 indicates a quantum anomaly of the conformal

symmetry of the WS theory. This quantum anomaly is key to understanding the
structure of the quantum theory in the sequel.
3.1.3 Fock space

Let us now construct the Hilbert space of our quantum theory. That is, we seek the irreducible
representations of the operators
• xµ = (xµ )† , pµ = (pµ )† as well as
µ µ † µ µ †
• αm = (αm ) , α̃m = (α̃m ) , which form an infinite family of harmonic oscillators.
Let us recall some basics facts about the harmonic oscillator:
• The harmonic oscillator is defined in terms of the operators a, a† with commutation relations
[a† , a] = −1.
• The number operator N = a† a is hermitian and diagonalisable with eigenstates such that
N |n i = n |n i and commutation relations

[N, a] = −a a : lowering operator
(3.18)
[N, a† ] = a† a† : raising operator
• The Fock space is constructed form the vacuum |0 i with the property a |0 i = 0 as the
space of states of the form a† . . . a† |0 i.
Returning to the bosonic string, the key insight is that for each mode number m and for each
dimension µ we are facing a separate harmonic oscillator defined by the commutation relations
µ
[αm , αnν ] = m δm+n,0 η µν η µν = diag(−1, 1, . . . , 1), (3.19)
µ i i
and similarly for α̃m in the closed string. For spacelike µ = i we have [αm , α−m ] = +m.
Thus, the operators ai|m| = √1 α|m|

i
and (ai|m| )† satisfy [ai|m| , (ai|m| )† ] = 1, which identifies them
|m|
i
as annihilation and creation operators. We prefer to work with the unnormalised αm , though,
i.e. with
i
α−|m| : creation operators,
i
α|m| : annihilation operators.
In addition, we need to furnish a representation of the Heisenberg algebra formed by xµ , pµ .

Combining everything we define a ground state |0; pµ i with the following properties
• Π̂µ |0; p i = pµ |0; pµ i,
µ
• αm |0; pµ i = 0 ∀m > 0.
That is the state |0; pµ i has momentum pµ . The Fock space is then generated by action of the
µ
independent creation operators α−m , m > 0, on this state.
It is spanned by the set of states
Y µ
{ (α−m )nm,µ |0; p i}. (3.20)
m
Note that a priori there is a ground state for each value of pµ , and pµ and the oscillators and
independent. We will soon see that this is not true in the physical Hilbert space any longer.
µ
To characterise a particular state one introduces its polarisation tensor, e.g. ξµ α−1 |0; p i or
µ ν
ζµν α−1 α−1 |0; p i etc.
µ
This poses an immediate problem. Take e.g. |ψ i = ξµ α−k |0; p i with k > 0 and ξµ = (1, 0, . . . , 0)
and compute the norm
hψ|ψi = h 0; p| a0k a0−k |0; p i = h 0; p| [a0k , a0−k ] |0; p i = −kh0; p|0; pi < 0.
The appearance of such negative norm states or ”ghosts” is unacceptable in a quantum theory
as they spoil unitarity.
To appreciate the problem and find its solution we recall that exactly the same issue arises in
the Gupta-Bleuler quantisation of QED.
• To define a canonical formulation of U (1) gauge theory it is necessary to start not from
the gauge invariance Lagrangian, but rather from
1 1
L = − F µν Fµν − (∂ · A)2 .
4 2
This corresponds to (partially) fixing the U (1) gauge symmetry by imposing the gauge
fixing constraint ∂ · A = 0.
• While the naive Fock space suffers from ghosts, these are absent from the set of physical
states defined by imposing the gauge fixing constraint at the quantum level.
In string theory we have likewise (partially) fixed the underlying diffeomorphism and Weyl sym-
metry by imposing the gauge fixing condition hab = ηab . But even at the classical level we still
have to impose the Virasoro Virasoro constraints Tab = 0, i.e. ⇔ Lm = 0 ∀ m ∈ Z (for open
strings and similarly including L̃m for he closed string).
• Thus we must impose the constraints Lm = 0 as an operator equation. Our first guess
Lm |ϕ i = 0 ∀ m is inconsistent due to the central term in Virasoro algebra. But this would
be too strict anyways as the analogue of the classical condition would be, in the spirit
of Ehrenfests’s theorem, rather to impose the vanishing of the expectation value of the
Virasoro generators.
• Therefore it is sufficient to require, for the open string,
Lm |ϕ i = 0 m > 0, (3.21)
(L0 − a) |ϕ i = 0. (3.22)
I.e. a state is called |ϕ i physical if and only if it satisfies
(Lm − a δm,0 ) |ϕ i = 0 ∀m ≥ 0 . (3.23)
• Indeed physical states satisfy h ϕ| Lm − aδm,0 |ϕ i = 0 ∀ m.

• Note that in implimenting the zero level Virasoro constraint we allow for a yet-to-be de-
termined normal ordering constant a, following the logic discussed around (3.12).
• For the closed string we require
(Lm − a δm,0 ) |ϕ i = 0, ∀ m ≥ 0, (3.24)
(L̃m − ã δm,0 ) |ϕ i = 0, ∀ m ≥ 0. (3.25)
• A priori, a and ã might be different. However, we insist that invariance under σ → σ + ∆

continues to hold in the quantum theory. Otherwise the theory would suffer from a gravi-
tational anomaly, i.e. an anomaly of spatial diffeomorphism invariance of the worldsheet.1
Then
(L0 − L̃0 ) |ϕ i = 0 (3.26)
requires
a = ã. (3.27)
Mass shell condition
As in the classical theory the quantum mass shell condition arises as the level-zero Virasoro
constraint involving L0 . Thus the normal ordering constant a effects the mass of the string
states.
1 Note that in canonical quantisation the time coordinate τ is singled out from spatial coordinates - here
just σ. Unlike the gravitational anomaly affecting the latter, an anomaly in τ -reparametrisation invariance by
(L0 + L̃0 − 2a) |ϕ i = 0 is accepatable,
i) For the open string we consider (NN) boundary conditions in directions µ, (DD) boundary
conditions in directions i and (ND) or (DN) boundary conditions in directions a and compute
∞
X 1 X µ X
L0 = α−n · αn + α02 = i
[α−n (αn )µ + α−n (αn )i ] + a
α−r (αr )a
n>0
2 n=1 r∈N0 + 21
+ α0 pµ pµ + α0 (T ∆xi )2 ,
i.e.
L0 := N + α0 p2 + α0 (T ∆x)2
with N the number operator and pµ pµ = −M 2 the invariant mass. Then solving the
level-zero Virasoro constraint (L0 − a) |ϕ i = 0 for p2 = −M 2 yields
α0 M 2 |ϕ i = (N + α0 (T ∆x)2 − a) |ϕ i, (3.28)
with N = n>0 Nn + r∈N0 + 1 Nr . . . the number of excitations. The contribution α0 (T ∆x)2

P P
2
takes into account the energy from the tension of the string stretched in the (DD) directions.
ii) For the closed string we obtain

X α0 2 α0
L0 = α−n · αn + p = N + p2 ,
n>0
4 4
X α0 2 α0
L̃0 = α−n · αn + p = Ñ + p2 .
n>0
4 4
Level matching (3.26),
(N − Ñ ) |ϕ i = 0, (3.29)
and the Virasoro constraints (L0 − a) |ϕ i = 0 = (L̃0 − a) |ϕ i yield
α0 M 2 = 4(N − a). (3.30)
3.1.4 Explicit open (NN) Fock space and criticality

The structure of the physical Hilbert space is that of a tower of string excitations with
increasing mass according to the oscillator number N .
However, we still have not proven that imposing the Virasoro constraints really removes all
negative norm states from the string spectrum. We will see now that this leads to the concept of
criticality. While we do not present a full proof of the no-ghost theorem in the formalism of Old
Canonical Quantisation, we outline the logic by analysing the impact of the Virasoro constraints
for the open string spectrum with (NN) boundary conditions for all X µ and construct the physical
states at increasing excitation levels. The proof of the no-ghost theorem will be given in the
context of the Lightcone Quantisation scheme, along with an analysis of the string spectrum also
of the closed string and of the open string with more general boundary conditions.
i) Vacuum (N = 0): |0; pµ i

The Virasoro constraints are of the following form:
• At level zero we find

! a
(L0 − a) |0; pµ i = 0 ⇒ M 2 = − . (3.31)
α0
So for a is undetermined, but we see that if a > 0 the vacuum is tachyonic.
!
• The higher constraints Lk |0; p i = 0, k > 0 turn out to be vacuous. Indeed
∞ 0
1 X α|n| |0;p i=0 1 X
Lk |0; pµ i = αk−n · αn |0; p i = αk−n · αn |0; p i
2 n=−∞ 2 n=−∞
0
X 1
= [αn · αk−n + (k − n)δk−n,n ] |0; p i.
n=−∞
2
Thus Lk |0; p i ≡ 0 identically for k > 0, i.e. the higher constraints are automatically
satisfied.
For future purposes we normalise the vacuum state as h0; p|0; pi ≡ 1.
µ
ii) First excited level (N = 1): |ϕ i = ξµ α−1 |0; p i with polarisation vector ξ
The constraints give the following restrictions:
µ
• To evaluate (L0 − a) |ϕ i = (α0 p2 + n>0 α−n · αn − a)ξµ α−1
P
|0; p i we commute αn
through α−1 . In this process only the oscillator n = 1 picks up a commutator term,
!
i.e. (L0 − a) |ϕ i = (α0 p2 + 1 − a) |ϕ i = 0˙ Thus p2 = a−1
α0 or
1
M2 = (1 − a). (3.32)
α0
• Next we evaluate
∞
! 1 X µ
0 = L1 |ϕ i = ( α1−n · αn − a)ξµ α−1 |0; p i (3.33)
2 n=−∞
1 µ
= (. . . + α2 α−1 + α1 α0 + α0 α1 + α−1 α2 + . . .)ξµ α−1 |0; p i. (3.34)
2
We pick up a contribution only from the terms involving α1 , and if we remember that
!
α0 is proportional to p we conclude that (ξ · p) |0; p i = 0, i.e.
!
ξ µ pµ = 0. (3.35)
• The higher constraints are vacuous: From

µ
[Lm , αnµ ] = −n αm+n (3.36)
it follows that all higher Lm |ϕ i ≡ 0 are automatically satisfied for m > 1.

• The norm of the first excited state is found to be
hϕ|ϕi = h 0; p| (ξ · α−1 )† (ξ · α−1 ) |0; p i = ξ · ξ. (3.37)
We see that depending on the value of the normal ordering constant a we must distinguish
3 physically inequivalent situations.
a) a > 1
In this case p2 > 0, i.e. M 2 < 0. This alone does not shock us. What is a killer, though,
is that the spectrum still contains negative norm states. To see this we pick w.l.o.g.
pµ = (0, p1 , 0, . . . , 0) and observe that we can find a polaristion vector ξ µ satisfying the
constraint ξ ·p = 0, but which is of negative norm ξ ·ξ < 0. (E.g we can take ξ µ = (ξ 0 , 0).)
We conclude that a > 1 is inconsistent as it does not lead to a unitary quantum theory.
b) a = 1
Now p2 = 0 and w.l.o.g. we pick pµ = (ω, ω, 0, . . . , 0). The constraint ξ · p = 0 implies
that
ξ µ = a pµ + ξTµ , ξTµ = (0, 0, ξ). (3.38)
Thus the polarisation vector ξ µ describes (d-2) transverse degrees of freedom ξTµ and 1
longitudinal degree of freedom a pµ . The latter is null, i.e. of zero norm, because p2 = 0
and furthermore orthogonal to any transverse vector.
Note: At this stage, this is all we can say. However, an in-depth analysis of the string
interactions reveals that if d = 26 the longitudinal state decouples from all scattering
processes. In that case it is not a physical degree of freedom. This is analogous to the
situation of the photon in QED. Recall that the U (1) gauge invariance implies
A0µ = Aµ + ∂µ Λ,
ξµ0 = ξµ + c pµ .
Thus the null d.o.f. is pure gauge.

To conclude, if a = 1, then spectrum contains a massless vector boson with d−2 transverse
d.o.f.. If d = 26 this particle corresponds to a gauge boson with consistent interactions.
c) a < 1
Now p2 < 0 and M 2 > 0 so w.l.o.g. we pick pµ = (p0 , 0), ξ µ = (0, ξ i ). This describes a
massive vector of positive norm with d−1 d.o.f.. We know from QFT that massive vector
theories are well-defined in principle so at this stage we have no objections to a < 1.
iii) Second excited level (N = 2)
The most general state is of the form

µ ν µ
|ϕ i = ζµν α−1 α−1 + ζ̃µ α−2 |0; p i. (3.39)
Evaluation of the Virasoro constraints along the above lines reveals that d ≤ 26 is a neces-
sary condition for absence of ghosts. This will be discussed in the examples classes.
Let us summarise our findings. Analysis of the string spectrum up to the second excited level in
Old Covariant Quantisation shows that
a ≤ 1, d ≤ 26
is a necessary condition for unitarity.
Criticality in OCQ
• One can prove, with considerably more efforts, the OCQ no-ghost-theorem:
If a = 1, d = 26, than every physical state is of the form
|φ i = |φT i + |s i
such that hφT |φT i > 0 and |s i decouples from all physical processes.
• In OCQ it is not possible to prove at tree-level that a < 1 and/or d < 26 is inconsistent.
But at one-loop-level unitarity does require a = 1, d = 26 if the structure of the worldsheet
theory as free bosonic theory is to be maintained.
This leads to the concept of criticality:
a = 1, d = 26 defines the critical string.
As stated already, while in OCQ we cannot prove the necessity of a = 1, d = 26 without looking
at one-loop interactions, we will find a very simple proof in the Lightcone Quantisation approach.
The tachyon
We have seen that for a = 1 the vacuum state |0; p i is tachyonic, M 2 = − α1 < 0.
• The existence of a tachyon is an artifact of the bosonic string theory. Eventually this
feature will be overcome in the superstring theory.
• Note that in QFT tachyonic states are not inconsistent but indicate an instability of vac-
uum. The most celebrated example of a tachyon in everyday physics is of course the Higgs
particle with potential
V (Φ) = λ(|Φ|2 − v)2 (3.40)

⇒ V 00 (Φ)Φ=0 < 0 ⇔ M 2 < 0.

(3.41)
The instability triggers tachyon condensation into a stable vacuum at |Φ|2 = v.
• An interesting question is if the same thing happens in string theory in the sense that the
bosonic string is related, by tachyon condensation, to the superstring theory. This requires
methods of string field theory and is yet to be decided.
3.2 Light-cone quantisation (LCQ)

3.2.1 Light-cone gauge
This scheme works, like OCQ, with the flat worldsheet metric hab = ηab , but uses the residual
reparametrisation invariance (conformal symmetry)
ξ + 7→ ξ˜+ (ξ + ), ξ − 7→ ξ˜− (ξ − ) (3.42)
to implement the Virasoro constraints Tab = 0 before quantisation.

• The key insight is that we can use the residual conformal symmetry (3.42) to transform
(τ, σ) → (τ̃ , σ̃) with
1 ˜+ +
τ̃ = (ξ (ξ ) + ξ˜− (ξ − )), (3.43)
2
1
σ̃ = (ξ˜+ (ξ + ) − ξ˜− (ξ − )). (3.44)
2
This implies that τ̃ satisfies the free wave equation
∂+ ∂− τ̃ = 0, (3.45)
which is the same equation governing the dynamics of X µ (τ, σ), ∂+ ∂− X µ (τ, σ) = 0.
We can therefore use (3.42) to identify τ̃ with one of the X µ .
Cautionary note: What we have just presented is a quick and dirty way to arrive at
the correct conclusion. However, imposing the equation of motion before quantisation is
generally problematic. There is a more thorough way to introduce light-cone quantisation
without relying on this line of argument just by exploiting the local Weyl and diffeomor-
phism invariance of the action. This is described in detail in volume 1 of [P], p. 17-19.2 In
any case, invariance under Weyl transformations plays a key role.
• To actually use this identification to our advantage we introduce lightcone coordinates
for spacetime
1
X ± = √ (X 0 ± X d−1 ), X i i = 1, . . . , d − 2, (3.46)
2
so that the ambient space metric becomes
η+− = −1 = η−+ , ηij = δij , (3.47)

+ − i i
X · X = −2X X + X X (sum over i implied)s. (3.48)
• We finally choose X + (τ, σ) as the coordinates to be identified with τ̃ :
` 1 +
τ̃ = X − x+ . (3.49)
2πα0 p+ |{z}
integration constant
We now relabel (τ̃ , σ̃) → (τ, σ) and arrive at
2πα0 +
X + (τ, σ) = p τ + x+ . (3.50)
`
Rl
Note that the normalisation has been chosen such that P + = T 0
Ẋ + dσ̃ ≡ p+ .
Comments:
• We have effectively used the infinite dimensional conformal symmetry (3.42) to gauge away
an infinite number of oscillator degrees of freedom by setting αn+ = 0 ∀n 6= 0.
2 Inaddition, one has to show that the specific type of reparametrisation ghosts that arise in the process of the
gauge fixing performed here indeed decouple. We thank S. Theisen for discussions on this.
• This procedure is not manifestly Lorentz invariant as it singles out one coordinate, here X + .
We must therefore check Lorentz invariance at the quantum level to ensure a consistent
quantisation scheme.
The advantage of the above spacetime lightcone coordinates is that we can now explicitly solve
the Virasoro constraints
Tab = 0 ⇔ (Ẋ ± X 0 )2 = 0, (3.51)
which in lightcone coordinates become
!
0 = −2(Ẋ ± X 0 )+ (Ẋ ± X 0 )− + (Ẋ ± X 0 )i (Ẋ ± X 0 )i . (3.52)
2πα0 +
Using Ẋ + = ` p and (X + )0 = 0 this yields
1 ` 1
(Ẋ ± X 0 )− = (Ẋ ± X 0 )2⊥ . (3.53)
2 2πα0 p+
Thus only the transverse oscillator degrees of freedom are independent, those of X + and X − are
either vanishing or determined by them. To see this explicitly consider the open string with, say,
(NN) boundary conditions ∀µ and make the familiar ansatz
2πα0 − √ X1 π
X − = x− + p τ + i 2α0 α− e−i ` nτ cos( nπ
` σ) (3.54)
` n n
n6=0
and similarly for X i . We now evaluate (3.52), more precisely the sum of the two constraints
(3.53), to find after some algebra
∞
1 1 1 X i
αn− = √ ( i
αn−m αm ) classically (3.55)
2α0 p+ 2 m=−∞
√
∀n ∈ Z including the zero mode α0i = 2α0 pi .
A similar analysis of the sum of the two constraints (3.53) for the closed string yields
∞
1 1 X i
α̃n− = √ ( i
α̃n−m α̃m ). (3.56)
2α0 p+ m=−∞
∞
1 1 X i
αn− = √ ( i
αn−m αm ). (3.57)
2α0 p+ m=−∞
3.2.2 Quantisation in LCG

We are now ready to perform the standard quantisation procedure in lightcone coordinates. The
canonically conjugate variables following from the action in lightcone gauge,
Z Z
1 0
S= dτ dσ((Ẋ ) − (X ) ) − dτ p+ q̇ −
i 2 i 2
(3.58)
4πα0
with q − = 1` dσX − , are

R
X i, Πi , i = 1, . . . , d − 2 (3.59)
− +
q , p . (3.60)
The canonical commutation relations are

i
[αm , αnj ] = m δn+m δ ij , [xi , pj ] = iδ ij ,
i
[α̃m , α̃nj ] = m δn+m δ ij , (3.61)
+ −
[p , q ] = i.
The classical expression for αn− has to be modified by means of a normal ordering constant, e.g.
for the open string with (NN) boundary conditions
∞
1 1 X
αn− = √ ( i
: αn−m i
αm : −aδn,0 ). (3.62)
2 2α0 p+ m=−∞
Mass shell condition
The mass shell condition can now be read off directly from
M 2 = −p2 = 2p+ p− − pi pi , (3.63)
where
1
pµ = √ α0µ , open, NN, (3.64)
2α0
r r
µ 2 µ 2 µ
p = α = α̃ , closed. (3.65)
α0 0 α0 0
a) For the open string with (NN) boundary conditions ∀µ this yields
∞
2 1 X
Mop = (2 : αi αi : +α0i α0i − 2 a − α0i α0i ), (3.66)
2α0 m=1 −m m
∞
2 1 X
i i
Mop = (N⊥ − a) N⊥ = α−m αm . (3.67)
α0 m=1
One can generalise the above expressions for open string with boundary conditions
(NN): X + , X − , X k (3.68)
l
(DD): X (3.69)
a
(ND)/(DN): X (3.70)
and finds
1
2
Mop = (N⊥ + α0 (T ∆x)2 − a), (3.71)
α0
X∞ X
k k l l a
αra

N⊥ = α−m αm + α−m αm + α−r (3.72)
m=1 r∈N0 + 21
b) For the closed string one obtains

2
Mcl2 = (N⊥ + Ñ⊥ − 2a). (3.73)
α0
One can show that the difference of the two constraints (3.53) results in the level matching con-
dition N⊥ = Ñ⊥ . Since, apart from this extra constraint, the Virasoro constraints are already
i i
implemented explicitly, all string excitations using transverse α−m , α̃−m are automatically phys-
ical. In particular all excitations are transverse and the spectrum is manifestly free of ghosts.
E.g. for the open string, with (NN) boundary conditions ∀µ the spectrum starts as follows:
• level 0: |0, p i, M 2 = − αa0

i 1
• level 1: α−1 |0, p i, M2 = α0 (1 − a), i = 1, . . . , d − 2
We will take a very close look at the open and closed spectrum later.
Criticality
In LCQ the critical values for the normal ordering constant a and for the number d of spacetime
dimensions follow by the requirement of Lorentz invariance. Since the definition of the lightcone
coordinates singles out a specific coordinate it is not guaranteed that the Lorentz symmetry is
non-anomalous in the quantum theory.
For simplicity we carry out this analysis for the open (NN) string. Recall that the generators of
Lorentz transformations are given by
Z l
J µν = T dσ X µ Ẋ ν − X ν Ẋ µ = lµν + E µν , (3.74)
0
lµν = x p − xν p µ ,
µ ν
(3.75)
∞
X 1 µ
E µν ν ν µ

= −i α−m αm − α−m αm . (3.76)
m=1
m
These should satisfy the Lorentz algebra

h i
J µν , J ρσ = iη µρ J νσ + iη νσ J µρ − iη µσ J νρ − iη νρ J µσ . (3.77)
The transformation J i− mixes X ± with X i . Since X ± are singled out in LCQ, a breakdown of
Lorentz invariance might show up in an anomaly of the algebra satisfied by J i− .
In absence of an anomaly we should get
h i
J −i , J −j = iη −− J ij + iη ij J −− − iη −j J i− − iη i− J −j = 0. (3.78)
However, an explicit, long and rather painful calculation sketched below reveals
∞
h i 1 X
j

J −i , J −j = − i
∆m α−m j
αm − α−m i
αm (3.79)
(p+ )2 m=1
26 − d 1 d − 26
with ∆m = m· + + 2(1 − a) . (3.80)
12 m 12
h i
⇒ J −i , J −j = 0 ⇔ d = 26, a = 1. (3.81)
Remarks:
• A similar computation fixes d = 26, a = 1 for the closed string.
• The result d = 26 also holds in the presence of lower-dimensional branes (i.e. with DD or
ND/DN sectors for some excitations). The value of ”a” in the presence of more general
boundary conditions is derived in the following section.
Extra material (non-examinable):
For completeness we here guide through the main steps in deriving the result (3.79).
−
• Recall that the oscillators αn are defined, at the quantum level, in the normal ordered manner
∞
D−2
!
− 1 1 1 X X i i
αn = √ : αn−m αm : −aδn . (3.82)
2α0 p+ 2 i=1 m=−∞
For simplicity we focus on the open string and set α0 = 21 . With the help of the expression for
−
αn one sees that [J i− , J j− ] involves terms quartic and quadratic in the oscillators αn
i
. One can
verify that the commutators involving the quartic terms vanish. This means that one can make
the following ansatz for [J i− , J j− ]:
1
[J i− , J j− ] = − C ij (3.83)
(p+ )2
with
∞
X
C ij = i
∆m α−m j
αm j
− α−m i
αm , (3.84)
m=1
where ∆m are c-numbers which we want to compute. One can furthermore show that
k
h0|αm C ij α−m
l
|0i = m2 (δ ik δ jl − δ jk δ il )∆m . (3.85)
ij
This equation tells us that we can compute the coefficients ∆m by computing C from (3.83),
evaluating matrix elements of the above form and comparing the result with (3.85).
• To this end we define E j = p+ E j− and derive the commutators
[x− , 1/p+ ] = i(p+ )−2 (3.86)
− i i
[xi , αn ] = + αn (3.87)
p
[xi , E j ] = −iE ij . (3.88)
i− j−
With this we can compute the commutator [J ,J ] explicitly, leading to
C ij = 2ip+ α0− E ij − iE i pj + iE j pi − [E i , E j ] . (3.89)
−
• To evaluate matrix elements of the form (3.85) we need commutator relations for the By αn .
−
comparing with the Virasoro generators Lm one can argue that the p+ αn satisfy the commutation
relations
hD − 2 i
− − −
[p+ αm , p+ αn ] = (m − n)p+ αm+n + (m3 − m) + 2am δm+n , (3.90)
12
i + − i
[αm , p αn ] = mαm+n . (3.91)
i j
• Now we use (3.91) as well as the usual commutator [αm , αn ] = mδm+n δ ij to compute
k
h0|αm C ij α−m
l
|0i = h0|2m(m + p+ α0− )δ ik δ jl + mpj pk δ il − mpj pl δ ik (3.92)
m
X 1 k i −
+ (m αm−n αn − δ ik p+ αm )
n=1
n
m
−
X 1 j l
× (δ jl p+ α−m −m α−p αp−m )|0i − (i ↔ j) .
p=1
p
• To evaluate this expression we derive from (3.90)
− − D−2 3
(p+ )2 h0|αm α−m |0i = 2mhp+ α0− i + (m − m) + 2am (3.93)
12
and prove the identities
m
−
X 1 j
p+ h0|αm l
α−n αn−m |0i = pj pl + δ jl m(m − 1)/2 (3.94)
n=1
n
m m
X 1 k i
X 1 j l
h0| αm−n αn α−p αp−m |0i − (i ↔ j) = (m − 1)(δ il δ jk − δ jl δ ik ) . (3.95)
n=1
n p=1
p
• This allows us to compute (3.92). Comparing the result with (3.85) we finally arrive at

26 − D 1 D − 26
∆m = m + + 2(1 − a) . (3.96)
12 m 12
3.2.3 Normal ordering constant “a” as a Casimir energy

We now provide a deeper interpretation of the normal ordering constant ”a”.
Recall that the constant ”a” is defined as the normal ordering ambiguity in promoting the
classical zero-level Virasoro generator to a quantum operator,
1X
classical L0 = α−n · αn (3.97)
2 n
↓
1X
quantum L0 − a = : α−n · αn : −a. (3.98)
2 n
Thus a is related to the commutator terms we pick up in performing the normal ordering. For
a single open string field X i with (NN) or (DD) boundary conditions this commutator term is
given by
∞
1X i 1 X i 1 X
α−n αni = : α−n αni : + n (no sum over i). (3.99)
2 2 n 2 n=1
n∈Z
As is stands this sum is divergent. On the other hand, the normal ordering constant of the
critical string is finite. To see how the two are related we note that the divergent sum is nothing
but the vacuum energy of the string. This interpretation follows from the relationship between
the quantum Hamiltonian and L0 .
For definiteness we now consider the open string with (NN) boundary conditions ∀µ. The classical
and quantum Hamiltonian in LCG are, respectively,
π1 X i
classical H = α−n αni + α0 pi pi (3.100)
` 2
n6=0
↓
∞
π X i i 0 i i d−2 X
quantum H= α α +αp p + n . (3.101)
` n>0 −n n 2 n=1
The computation of the divergent vacuum energy, given by the last summand, is standard in
QFT and proceeds via regularization and renormalisation. In a complete treatment one starts
with a regularised theory (e.g. by putting the two-dimensional field theory of the free boson
on a lattice), quantises it, extracts the vacuum energy (which includes cutoff-dependent terms),
renormalises it by subtracting suitable counterterms such as to remove the cutoff-dependent
would-be divergent terms and finally removes the cutoff.3 This procedure can be carried out in
full detail in a quantum field theoretic treatment.4 With this understanding, we briefly sketch
here a somewhat heuristic and quick way to see the result:
• We first introduce a cutoff Λ to regularise the divergent expression in such a way
that the divergence appears as we remove the cutoff by sending Λ → ∞. A convenient
cutoff procedure here is e.g. to rewrite the vacuum energy as
∞ ∞ ∞
πd−2 X π d − 2 X − πn πd−2 X π
n→ n e `Λ = n(e− `Λ )n . (3.102)
` 2 n=1 ` 2 n=1 ` 2 n=1
P∞ d
P∞
This choice of regulator is called ’heat-kernel cutoff’. From n=1 nq n = q dq n=1 qn =
d 1 q
q dq 1−q = (1−q)2 we find
∞ π
πd−2 X πd−2 e− `Λ
n = limΛ→∞ 2 (3.103)
` 2 n=1 ` 2 π
1 − e− `Λ
d−2 ` π 1 1
= limΛ→∞ Λ2 − + O( ) . (3.104)
2 π ` 12 Λ
• The expression for the vacuum energy has two non-vanishing contributions: The term
proportional to Λ2 is the divergent piece. It is important that this term scales like `.
Therefore, this
√ term can be absorbed by adding a cosmological constant term proportional
to Λ2 d2 σ −h to the bare Polyakov action via renormalisation. This counterterm in
R
the bare action then cancels off the divergence arising in the quantum computation of the
vacuum energy.
In addition there is the finite term − d−2 π 1
2 ` 12 . This term is present only due to the finite
size of the string because it disappears in the limit ` → ∞. Unlike for the divergent term,
there exists no local counterterm that we could add to the action such as to absorb this
term. This term is therefore physical and defines the Casimir energy of the string.
A priori one might wonder why we have to cancel the entire piece of the term scaling like ` by
adding a suitable cosmological constant - can’t we just keep a finite fraction of it and declare
it as part of −a? The reason why this is not possible in string theory is conformal invariance:
The cosmological constant term breaks conformal invariance explicitly in the action - unless the
classical counterterm and the quantum term cancel exactly. Therefore −a must be identified
with the Casimir energy, i.e.
d−2
a= for (d − 2) transverse (NN) or (DD) oscillators. (3.105)
24
3 Removing the cutoff, i.e. taking the limit Λ → ∞, corresponds to taking the continuum limit. In a usual
QFT this would certainly be what we are interested in. In the present context we might object that there may
not exist such a limit after all in a theory of Quantum Gravity (though of course we are here not talking about
the spacetime theory, but rather about the theory on the worldsheet). As we will see below, we can equally well
keep the cutoff Λ finite as the term of interest for us is cutoff-independent.
4 I thank A. Hebecker for discussions on this point.
For d = 26 this gives a = 1 as found by requiring Lorentz invariance for the open string with
(NN) conditions and also for the closed string. It therefore holds also for the (DD) string.
Remarks:
• Note that a 6= 0 still breaks conformal invariance, but merely in form of an acceptable
conformal anomaly, i.e. of a quantum anomaly of the conformal symmetry.
• Very soon we will get to know powerful CFT techniques yielding another derivation of the
Casimir energy making the relation to the central term c in in the Virasoro algebra clear.
One can prove rigorously that the result for the Casimir energy is independent of the concrete
πn
choice of regulator, i.e. one can replace the exponential e− `Λ in (3.102) by any smooth function
f ( πn
`Λ ) with f (0) = 1 and limx→∞ xf
(k)
(x) = 0 for all k ≥ 1. These assumptions ensure that
the regularization scheme does not change the IR behaviour and that the regulator falls off
sufficiently fast in the UV. Another type of regulator meeting these criteria appears in ζ-function
regularisation. The regularization consists in replacing
∞ ∞ −s ∞
πd−2 X πd−2 X πn π 1−s d − 2 s X 1−s
n→ n = µ n (3.106)
` 2 n=1 ` 2 n=1 `µ ` 2 n=1
with s > 0. Removing the cutoff corresponds to s → 0. The scale µ will drop out and is needed
for dimensional reasons.
The above expression involves the ζ-function ζ(s), defined as
∞
X
ζ(s) = n−s . (3.107)
n=1
We need the following two properties of the ζ-function, which are proven in standard textbooks
on complex analysis:
• ζ(s) is convergent for Re(s) > 1.
• ζ(s) allows for an unique analytic continuation to s = −1 with
1
ζ(s − 1) = − + O(s).
12
Therefore we deduce that the contribution to ”a” from 1 (NN) or (DD) direction (i.e. one integer
moded boson), given precisely by − 12 ζ(−1), is
1
+ per (NN) or (DD) direction. (3.108)
24
Note that in this scheme no divergent term comparable to the Λ2 -term in (3.103) arises and thus
no counter-term is needed. In this sense this is the preferred scheme in theories with conformal
invariance. On the other hand the absence of a divergent term in the formal expression given
by ζ(−1) might obscure the physics. From this perspective the heat-kernel regulator is more
intuitive.
This also gives a quick way to read off the result for a string with (DN) boundary conditions,
for which the mode expansion is half-integer. To this end we use
∞
X
ζ(s, q) = (n + q)−s (3.109)
n=1
with analytic continuation

1
ζ(−1, q) = − (6q 2 − 6q + 1).
12
The contribution from one (DN) direction (i.e. one half-integer moded boson) is therefore (q = 12 )
1
− per (DN) direction. (3.110)
48
Alternatively this result can be derived in a similar manner as for the (NN)/(DD) string around
(3.103).
This allows us to write down the normal ordering constant for the critical string with m (DN)
directions and d − m (DD) or (NN) as
#(N N ) + #(DD) #(DN ) #(N N ) + #(DD) + #(DN ) #(DN ) d−2 m

a= − = − = − .
24 48 24 16 24 16
The mass relation for open string with m (DN) directions is therefore
d−2 m
α0 M 2 = N + α0 (T ∆x)2 − a = N + α0 (T ∆x)2 − + . (3.111)
24 16
3.2.4 Open string spectrum along D-branes

It is time to analyze the critical string spectrum in slightly greater detail.
We make use of the following important general fact: from Quantum Field Theory5
In a Lorentz invariant quantum field theory in d dimensions a state forms an irreducible rep-
resentation of the subgroup of the d-dimensional Poincaré group which leaves its momentum
invariant. This group is called ”little group” or stabilizer subgroup.
We must distinguish 3 cases:
• If p2 = 0, we can take w.l.o.g. p = (ω, ω, 0). The stabilizer subgroup within SO(1, d − 1) is
given by SO(d − 2).6
• If p2 < 0, we can take w.l.o.g p = (p, 0) with little group SO(d − 1).
• If p2 > 0, we can take w.l.o.g p = (0, p, 0, . . . , 0) with stabilizer subgroup SO(1, d − 2).
Let us check if this principle is realized in the string spectrum.

5 See e.g. Weinberg, Quantum Field Theory I, Chapter 2.5 for a detailed discussion.
6 The full stabilizer group of massless particles within the d-dimensional Poincaré group is in fact ISO(d − 2),
the Euclidean group in d − 2 dimensions, which contains SO(d − 2) as a subgroup, but also includes translations
in d − 2 dimensions. While the finite dimensional representations of ISO(d − 2) and SO(d − 2) coincide, ISO(d − 2)
has in addition infinite dimensional representations not observed as particle states in nature. It is a long-standing
problem of relativistic quantum field theory to explain why such representations are not realized in nature. In
perturbative string theory, by contrast, all massless states manifestly form irreducible representations of SO(d−2),
as observed in nature. This can be viewed as one of the model independent predictions of string theory that go
beyond quantum field theory. More information can be found in the article http://arxiv.org/pdf/1302.4771.pdf
by Font, Quevedo, Theisen.
All (NN) open strings

We consider first the open string with (NN) b.c. ∀X + , X − , X i , with d = 26.
• The tachyonic ground state |0, p i with M 2 = − α10 is a space-time scalar. In particular it
is a scalar of the little group SO(1, 24).
i
• The massless transverse vector ξi α−1 |0, p i is a space-time vector in the vector or funda-
mental representation, denoted by , of SO(24).
• All higher states are massive and form complete irreducible representations of SO(25). To
make this manifest we must regroup the polarization degrees of freedom, which in LCQ are
a priori representations of SO(24), i.e. purely transverse, into irreducible representations
of SO(25).
Consider e.g. the second excited leve with M 2 = α10 . The different types of states are
i i j
ξi α−2 |0, p i & ξij α−1 α−1 |0, p i. (3.112)
In terms of SO(24) the polarization tensors form the irreducible representations
SO(24) : |{z} | {z } +1 (3.113)

24 symmetric traceless, #= 24·25
2 −1
These 324 degrees of freedom combine into the symmetric traceless of SO(25).
A single Dp-brane
Consider now the open string spectrum in the presence of a Dp-brane within R1,25 .
Note that the light-cone coordinates +, − must be along (NN) directions for a treatment within
LCQ so we take the boundary conditions as follows:
(NN) : i = +, −, 1, ...p−1
(DD) : a = p + 1, ..., 25
We distinguish
• excitations along the brane in directions X i , i = 1, ..., p − 1 and
• excitations orthogonal the to brane in directions X a , a = p + 1, ..., 25.
Let us build the spectrum:

• The ground state is |0, p i with momentum p only in the (NN) directions along the Dp-brane.
• At the first excited level we find the following states:
i
i) The parallel excitations are α−1 |0, p i, i = 1, ..., p − 1 , M 2 = 0.
This state forms a massless vector from the perspective of the Dp-brane propagat-
ing along the brane (i.e. it transforms as of SO(p − 1)). By general arguments of
QFT a massless vector must be a gauge potential. More precisely one can explicitly
verify by computations of its interactions that this is the case. String interactions will
be discussed later in this course. For now we state the key observation:
A single Dp-brane hosts a U (1) gauge theory!

a
ii) The normal excitations are α−1 |0, p i, a = p + 1, ..., 25 , M 2 = 0.
These form a collection of 24-p massless scalar fields as seen from the perspective
a
of the Dp-brane. I.e. ∀a we have 1 excitation α−1 |0, p i that does not carry an index
”i” along the Dp-brane.
These are the Goldstone bosons associated to the spontaneous breaking of the 26-
dimensional Poincaré invariance by the Dp-brane.
Note that (NN) in all directions is the special case of a D25-brane filling all of R1,25 .
Parallel Dp-branes
Consider two parallel branes at a distance xa2 − xa1 .
We now find 2 sets of massless vectors,

i 2
• α−1 |11, p i, i.e. a vector from strings with both ends on brane 1 with M11 = 0.
i 2
• α−1 |22, p i with both ends on brane 2, M22 = 0.
a a
Similarly there are two types of massless scalars α−1 |11, p i, α−1 |22, p i.
In addition we find strings stretched between brane 1 and brane 2. These come with both
orientations:
i i
• The vectors α−1 |12, p i and α−1 |21, p i have mass
1 1 (xa2 − xa1 )2
2
M12 2
= M21 = 0
(N − a + α0 (T ∆x)2 ) = . (3.114)
α 2πα0 2πα0
That is, we now also find massive vectors along branes 1 and 2.
a a
• Similarly there are massive scalars α−1 |12, p i, α−1 |21, p i.
Coincident Dp-branes
Now consider the limit xa2 → xa1 by moving the two Dp-branes together. We end up with 4
i a
massless vectors α−1 |ij, p i and 4 sets of massless scalars α−1 |ij, p i, i, j = 1, 2 .
In general for N coincident Dp-branes the bosonic open spectrum at level 1 consists of
• N 2 massless vectors α−1
i
|ij, p i, i, j = 1, ..., N and
• N 2 sets of massless scalars α−1
a
|ij, p i.
The labels i, j that keep track of the start- and endpoint of the open string are called Chan-
Paton factors. It is convenient to introduce a basis λakl , a = 1, ..., N 2 of N × N matrices and
rewrite a string state in the form
|kl, p i ≡ λakl |a, p i. (3.115)
The appearance of N 2 massless vectors signals an enhancement of gauge symmetry to non-
abelian gauge symmetry.
We observe that N 2 is the dimension of the adjoint representation of U (N ). Indeed as we will
see later the gauge interactions are consistent if λa = (λa )† , a = 1, ..., N 2 , i.e. the λa span the
Lie algebra of U (N ).
A stack of N coincident Dp-branes ”carries” a U (N ) gauge theory.
Remarks:
• Note that moving the N D-branes apart breaks U (N ) → U (1)N as only N massless vectors
remain. This process of breaking a non-abelian gauge group to its abelian subgroups is
called moving along the Coulomb branch. It is realized here by giving a vacuum expectation
value (VEV) to those scalars stretched between the N branes as these indicate the relative
position of the branes. Indeed, as these scalars are charged under the non-abelian generators
of U (N ) a nonzero VEV breaks U (N ) → U (1)N .
• As this simple example shows the physics of D-branes ”knows” a lot about gauge theories.
This has lead to a modern way to think about gauge theories by thinking of brane dynamics
instead and resulted in deep insights into the geometric structure behind Yang-Mills theory.
• Also gauge groups different from U (N ) can be realized in D-brane theories. In the tutoral
we discuss the concept of an orientifold: In such theories the possible gauge groups include
also SO(N ) and the symplectic group Sp(2N ). In non-perturbative string theories also
the exceptional gauge groups E6 , E7 , E8 and more exotic ones occur.
3.2.5 Closed string spectrum

We now turn to the closed string spectrum. Recall that for the closed string there are independent
left- and right-moving oscillators. The tower of closed string excitations is organised by
• the level-matching condition Ñ = N and
4 4
• the mass-shell condition Mcl2 = α0 (N − a) = α0 (N − 1).
The lower states of the closed string spectrum are therefore as follows:
• The ground state (N = Ñ = 0) is a tachyonic spacetime scalar |0, p i with M 2 = − α40 .
i j
• At the first excited level (N = Ñ = 1) we encounter a massless state ξij α̃−1 α−1 |0, p i, i, j =
1, ..., 24.
One can decompose the polarisation two-tensor ξij into irreducible representations7 of the
little group SO(24)
ξij = ξ(ij) + ξ[ij] + ξ (0) (3.116)
|{z} |{z} |{z}
symmetric traceless antisymmetric trace part
·
i.e.
1 1 ij k k
i j i j j i
α̃−1 α−1 |0, p i = α̃−1 α−1 + α̃−1 α−1 − δ α̃−1 α−1 |0, p i
2 d−2
[i j]
+α̃−1 α−1 |0, p i
1 ij k k
+ δ α̃−1 α−1 |0, p i.
d−2
Interpretation:
While we will have much more to say about each of these states, we give here a short overview
of their physical interpretation.
i) ξ(ij) describes a massless, transversely polarised spin 2 particle.
As we know from the treatment of gravitational waves in General Relativity these are pre-
cisely the correct degrees of freedom expected from the graviton. Indeed, a clasical gravi-
tational wave in direction k = (k, 0, . . . , 0, k) consists in a propagating perturbation of the
spacetime metric δgij with i, j = 1, . . . , d − 2 with δgij = δgji and δg i i = 0.8 By a general
theorem of Weinberg, every such transverse symmetric traceless massless tensor field must
couple to the conserved energy momentum tensor in the way expected from Einstein Gravity
and thus describes, in the quantum theory, a graviton.
In fact an explicit analysis of interactions of this massless closed string spin 2 state confirms
this interpretation. Later in this course we will be able to derive General Relativity as an
outcome of string theory with the ξ(ij) state playing the role of the graviton.
ii) ξ[ij] ≡ Bij describes the degrees of freedom of an extra antisymmetric tensor field called the
Kalb-Ramond field.
This antisymmetric 2-tensor Bij plays an important role in that it can be thought of as a
generalised (i.e. higher-rank) gauge potential.
• As we know from field theory, a gauge potential Aµ describes a 1-form potential A ≡
Aµ dxµ . The associated field strength is a 2-form F = dA = ∂µ Aν dxµ ∧ dxν which is
invariant under the gauge symmetry A → A + dχ with χ a 0-form.
• In perfect analogy the higher rank antisymmetric tensor Bij describes a 2-form potential
1
B≡ Bij dxi ∧ dxj (3.117)
2!
One can define a corresponding field strength
1
H = dB = ∂i Bjk dxi ∧ dxj ∧ dxk
2!
7 The space of symmetric traceless 2-tensors forms a vector space invariant under the action of the Lorentz
group in that Λi k Λj l ζ(ij) is again symmetric traceless. Similarly for ζ[ij] .

8 See e.g. Zwiebach, Chapter 10.6 for a review.
invariant under the gauge transformation
B → B + dΛ (3.118)
with Λ a 1-form.
iii) ξ (0) represents a scalar field, denoted usually by φ. It is called the dilaton and will be of
crucial importance in the context of string interactions.
The string spectrum is neatly encoded in the partition function, which will be derived via
methods familiar from statistical mechanics in the tutorials. As we will see as a result of a
saddle-point approximation, the high degeneracy of the string spectrum at high excitation levels
leads to to the notion of the the Hagedorn temperature as a maximal possible temperature
in a stringy universe.
3.3 Modern covariant quantisation

We finally present the modern covariant approach to quantization. Apart from being the standard
technique to quantize systems with gauge symmetries, this method is useful to compute string
interactions.
3.3.1 The Polyakov Path integral

The modern covariant approach to quantization proceeds via the Feynman path integral. We
can literally understand the path integral in the form needed as the stringy generalization of
the quantum mechanical Feynman path integral for point particles. Recall that for a quan-
tum mechanical point particle Feynman described the amplitude hx2 , t2 |x1 , t1 i as a “sum over
histories”, Z
i S[X]
hx2 , t2 |x1 , t1 i = x(t1 ) = x1 DX(t) e . (3.119)
x(t2 ) = x2
Adapting the same logic one can define the amplitude between an initial string state |i i and a
final state |f i as the path integral
Z
hf |ii = DX Dh ei SP [X,h] . (3.120)
| {z }
sum over all WS with
boundaries |i i, |f i
In principle the initial and final state are accounted for by considering only those worldsheets
which correspond the respective states at past and present infinity. The special structure of
the worldsheet theory allows for a different procedure in which we consider trivial boundary
conditions in the path integral (corresponding to the vacuum). The initial and final states
will instead be incorporated by inclusion of so-called vertex operators. Before discussing these
vertex operators and the actual computation of scattering amplitudes, however, we take it easy
and first compute the unnormalised partition function, corresponding to the path integral with
trivial boundary conditions and no operator insertion,
Z
Z= DX Dh ei SP [X,h] ,
Z
√ (3.121)
1
SP = − 0
dτ dσ −h hab ∂a X µ ∂b X ν ηµν .
4πα
Faddeev-Popov gauge fixing

In its above naive form the path integral Z is not well-defined due to an overcounting of gauge-
equivalent configurations of worldsheet metrics h and string fields X which all describe the
same physics. Namely, we recall that the action SP is invariant under a general diffeomorphism
ξ a 7→ ξ a − a (ξ) combined with a local Weyl transformation, under which the string field and
metric transform as
X µ 7→ X µ + a ∂a X µ , (3.122)
hab 7→ hab + (P · )ab + 2 Λ̃ hab (3.123)
with
(P · )ab = ∇a b + ∇b a − hab (∇ · ), (3.124)

1
Λ̃ = Λ + ∇ · . (3.125)
2
The operator P maps vectors to symmetric traceless 2-tensors
c
(P · )ab = Pab c (3.126)
with
c c
Pab = δ(b ∇a) − hab ∇c . (3.127)
This overcounting of gauge equivalent configurations is a standard problem in the path integral
quantization of gauge theories. The general solution is to isolate the integral over the gauge pa-
rameters and divide by the volume of the gauge group. In the case at hand this can be achieved
by converting the integral over all worldsheet metrics into an integral over all diffeomorphisms
a
a and Weyl rescalings Λ that take us to the gauge transformed ĥ( , Λ)
starting from some fixed ĥ.
From our earlier discussion of gauge fixing on the worldsheet we know the following: At least
locally, if for fixed ĥab the parameters ζ = (a , Λ) run over all diffeomorphisms and Weyl rescal-
ings then ĥζ = ĥ + δ ĥ runs over all metrics.
R
Therefore, given some functional of the metric, F [h], we can rewrite the path integral Dh F [h]
as follows,
∂(P · , Λ̃)
Z Z Z
ζ
Dh F [h] = D(P · ) DΛ̃ F [ĥ ] = D DΛ} F [ĥζ ] det (3.128)
| {z ∂(, Λ)
≡Dζ | {z }
Jacobian = Faddeev-
Popov determinant
c c
with the differential operator Pab = δ(b ∇a) − ĥab ∇c defined in terms of the reference metric ĥ. In
the last step we have introduced the Jacobian for the transition of the integral over D(P · ) DΛ̃,
which is what appears in δh, to an integral over the gauge parameters ζ. This Jacobian is the
Faddeev-Popov determinant and can be further massaged as

∂(P · , Λ̃)
P 0
= = det P. (3.129)

∂(, Λ) ∗ 1

Thus the path integral is Z

ζ
Z= Dζ DX ei S[X,ĥ ]
det P. (3.130)
−1
Note that S[X, ĥζ ] = S[X ζ , ĥ] by invariance under combined diffeomorphisms and Weyl trans-
formations.
If the functional measure is also invariant, then we can perform such a gauge transformation to
obtain Z
−1 ζ −1
Z = dζ DX ζ ei S[X ,ĥ] det P (3.131)
−1
and we can relabel X ζ → X. Then Dζ factors out and yields an overall volume of the group
R
of diffeomorphisms and Weyl rescalings, which was our goal. Omitting this overall factor we
arrive at Z
Z= DX det P ei SP [X,ĥ]
(3.132)
ĥ : arbitrary reference metric.
There are a number of subtleties:

R
• The measure Dζ DX is in general invariant only under diffeomorphisms, not under Weyl
rescalings. Later we will find that criticality (i.e. a = 1, d = 26) is equivalent to the
absence of this total Weyl anomaly in the quantum measure. For now we assume that the
measure is invariant and proceed as above.9
9 A proper analysis of the functional integral and a proof of Weyl invariance in the critical dimension can be
found in the original paper ’Quantum geometry of bosonic strings’ , Phys. Lett. B103, 207, 1981 by Polyakov.
• The above procedure assumes that every metric h can be written as h = ĥζ for precisely
one ζ. There is a double mismatch though: First, the conformal Killing transformations
are residual gauge symmetries not fixed above. These extra parametrisations must not be
included in the path integral in order to avoid overcounting. Indeed we will be careful to
fix this extra residual gauge freedom when computing scattering amplitudes. Second there
are caveats related to global properties of the WS in that for worldhseets of complicated
topology the metric contains extra parameters, so-called moduli, not accounted for by local
gauge transformations ζ. We must therefore sum over these moduli separately. We ignore
both these issues now and will come back to them in great detail when discussing the
moduli space of Riemann surfaces.
Who’s afraid of Faddeev-Popov ghosts?

What remains is a treatment of the functional determinant det P via the famous Faddeev-Popov
trick.
Inspired by the fact that inverse determinants can be expressed as Gaussian integrals as in
Z Y
1
dyi e−yi Aij yj ∼ √ (3.133)
i
det A
we try to rewrite det P in a similar manner as an integral. For this, however, we need to define
Grassmann variables as anti-commuting (i.e. fermionic ) variables. E.g. define the
variables Θ, ψ with the properties
Θ ψ = −ψ Θ, Θ2 = 0 = ψ 2 . (3.134)
We furthermore define the notion of Berezin integration over Grassmann variables by setting
Z Z
dΘ = 0 , dΘ Θ = 1. (3.135)
dΘ Θ2 = 0 etc., we find the identity

R
Since all higher integrals vanish,
Z Z
dψ dΘ eΘaψ = dψ dΘ (1 + Θaψ) = a (3.136)
for a (bosonic) number a. More generally we define the Grassmann variables
Θi : Θi Θj = −Θj Θi (3.137)
and similarly for ψi . From

Z
dΘi Θj = δij (3.138)
we deduce
Z
dn Θ Θi1 . . . Θin = i1 ...in . (3.139)
Then, given a bosonic matrix Mij its determinant can be expressed as a Grassmann integral
n
Z Y
dψi dΘi exp(Θi Mij ψj ) ∼
= det M. (3.140)
i=1
As a final technical step we note that we are interested in the determinant of a differential operator
P . For this we generalise the above formulae to functional integrals over Grassmann-valued fields
by promoting
M → operator, Θi → Θ(ξ a ) , ψi → ψ(ξ a ). (3.141)
Applied to our operator
c c
Pab = δ(b ∇a) − hab ∇c (3.142)
this procedure yields10
Z 1 Z q
det P ∼
= d
Db(ab) Dc exp 2
d ξ −ĥ b · (P · c) (3.143)
4π
Z 1 Z q
≡ Db(ab) Dcd exp d2 ξ −ĥ bab Pab
d
cd . (3.144)
4π
Here b(ab) (ξ a ) transforms as a symmetric traceless tensor on the worldsheet and cd (ξ a ) as a
vector. Both are anti-commuting, fermionic fields. cd and b(ab) are called Faddeev-Popov (FP)
ghosts and anti-ghost, respectively. They are fermionic objects with integer spin. A priori they
will lead to negative norm states, but we will see how to deal with them.
So as a result we have (after integration by parts in the ghost action)
Z
Z∼= DX Db Dc ei (SX +Sg ) ,
Z q
1 2
SX = − d ξ −ĥ ∂ a X · ∂a X, (3.145)
4πα0
Z q
i
Sg = − d2 ξ −ĥ ĥab cd ∇a b(bd) .
2π
Note:
• The equation of motion for ca is given by P · c = 0. Therefore the normalisable
solutions for ca are in one-to-one correspondence with the conformal Killing
vectors, which are the generators of the residual symmetry.
• The equation of motion for b(ab) is
∇a bab = 0. (3.146)
We will understand the geometric meaning of these equations when discussing the moduli
space of Riemann surfaces.
3.3.2 Quantisation of the bc-system

The gauge fixed path integral is defined in terms of an arbitrary reference metric ĥab so let’s be
cheap pick the simplest by going to flat coordinates with ĥab = ηab . The matter and ghost action
in flat lightcone coordinates read
Z
1
SX = d2 ξ ∂+ X · ∂− X, (3.147)
π α0
Z
i
Sg [b, c] = d2 ξ(c+ ∂− b++ + c− ∂+ b−− ). (3.148)
π
10 The 1
factor 4π
is conventional and merely changes the overall normalization of the partition function.
Tracelessness of bab translates into b+− ≡ 0 ≡ b−+ . The equations of motion for the (anti-)ghosts
take the simple form
∂+ b−− = 0 = ∂− b++ ,
∂+ c− = 0 = ∂− c+
provided the boundary terms proportional to dτ (c+ δb++ − c− δb−− )|`σ=0 that we pick in the
R
process of varying the action vanish.

In close analogy to the matter fields X we now discuss the boundary conditions and mode
expansion:
1) Closed string:
The closed string is by definition periodic in σ so we must impose the periodic boundary
conditions b(σ + `) = b(σ) and c(σ + `) = c(σ). The most general solution to the e.o.m. is
expanded as
∞
2π i
X
b++ (τ, σ) = ( 2π
` )
2
b̃n e− ` n (τ +σ)
, (3.149)
n=−∞
∞
2π i
X
b−− (τ, σ) = ( 2π
` )
2
bn e− ` n (τ −σ)
(3.150)
n=−∞
and
∞
` X 2π i
c+ (τ, σ) = c̃n e− ` n (τ +σ) , (3.151)
2π n=−∞
∞
− ` X 2π i
c (τ, σ) = cn e− ` n (τ −σ) . (3.152)
2π n=−∞
As for the X-fields we have an independent set of left- and right-movers. The normalisation
is chosen such as to lead to nice expressions for the anti-commutator in the quantum theory.
2) Open string:
For the open string the boundary terms must vanish at σ = 0 and σ = ` separately. Out of
the various possible boundary conditions compatible with this we consider the following,

c+ (ξ + ) = c− (ξ − ) and b++ (ξ + ) = b−− (ξ − ) . (3.153)

σ=0, ` σ=0, ` σ=0, ` σ=0, `
This can be rewritten as

c1 (τ, σ) =0 and b01 (τ, σ) = 0, (3.154)

σ=0, ` σ=0, `
which admits a reasonable geometric interpretation: As noted before the ca correspond to

reparametrisation parameters in that the ghost equation of motion P · c = 0 is simply the
Killing vector equation. Thus the above ghost boundary condition means that the residual
conformal Killing group is restricted to such reparametrisations that do not move the bound-
ary of the string in ξ 1 , i.e. σ-direction. Since this is a sensible result, we stick to this boundary
condition.
The most general solution of the equations of motion subject to these boundary conditions is
`X πi ± X πi ±
c± = cn e − ` n ξ , b±± = ( π` )2 bn e− ` n ξ . (3.155)
π n n
As expected the boundary conditions identify the left and right-moving oscillators.
Quantisation
The conjugate momentum of the anti-ghost field b±± follows from the action as
δ Sg [b, c] i ±
Πb±± = = c (3.156)
δ ∂τ b±± 2π
with canonical Poisson-bracket relation b±± (τ, σ), Πb±± (τ, σ 0 ) P.B. = δ(σ −σ 0 ). To quantise this

system we must take into account the fermionic nature of the b and c-fields. As is well-known
from quantisation of fermions in QFT, the correct procedure is to replace the Poisson-bracket
by 1i times the anti-commutator, defined for operators A, B as {A, B} = AB + BA. Thus,
b++ (τ, σ), c+ (τ, σ 0 ) = 2πδ(σ − σ 0 ),

(3.157)
− 0 0

b−− (τ, σ), c (τ, σ ) = 2πδ(σ − σ ). (3.158)
One can verfiy that this corresponds to the anti-commutator relations for the modes
{cm , bn } = δm+n,0 , {cm , cn } = 0 = {bm , bn } (3.159)
with the same relations obeyed in addition by c̃n0 , b̃n for the closed string.
Ghost Virasoro-Algebra
The ghost-energy momentum tensor T (g) follows from the full non-gauge fixed action as
(g) 4π δ Sg [b, c, h]
Tab = √ . (3.160)
−h δ hab
In lightcone gauge its non-vanishing components are
(g)
T++ = −i(2b++ ∂+ c+ + (∂+ b++ )c+ ), (3.161)
(g) − −
T−− = −i(2b−− ∂− c + (∂− b−− )c ) (3.162)
with corresponding Virasoro generators, for the closed string,

R` 2π
L(g)
n = − 4π` 2 0 dσe− ` i n σ T−− , (3.163)
R` 2π
L̃(g)
n = − 4π` 2 0 dσe ` i n σ T++ (3.164)
and similar formulae for the open string. It is not hard to check that in terms of the modes this
gives, classically,
∞
X
L(g)
m = (m − n) bm+n c−n (classically). (3.165)
n=−∞
At the quantum level the Virasoro operators are defined as the normal ordered analgue of this
classical expression. Normal ordering is again defined by putting lower-level modes to the left.
Due to the anti-commuting nature of the modes, we pick up a minus sign in this process if we
have to change the order of the modes, i.e.

bm bn , if m ≤ n
: bm bn := (3.166)
−bn bm , if n < m
and similarly for : cm cn : as well as for : bm cn :. Then,

∞
X
L(g)
m = (m − n) : bm+n c−n : . (3.167)
n=−∞
Note:
1) For example for the zero-level Virasoro generator application of the normal ordering prescrip-
tion yields
(g) P−1 P∞
L0 = n=−∞ (−n) : bn c−n : + n=1 (−n) : bn c−n :, (3.168)
| {z } | {z }
=bn c−n =−c−n bn
(g) P∞
L0 = n=1 n(b−n cn + c−n bn ). (3.169)
2) Note the slightly different normalization factors compared to the X-field Virasoro generators
(X)
Lm = 21 n : αm+n α−n :. In particular the zero-modes b0 , c0 do not appear in L0 .
P
A computation very similar to the one performed to deduce the Virasoro algebra of the matter-,
i.e. X µ fields, yields the Ghost Virasoro-algebra
(g) 1
[L(g) (g) 3
m , Ln ] = (m − n)Lm+n + (m − 13m )δm+n,0 . (3.170)
6
(g)
Note that the generators Lm are bilinears in the fermionic ghost modes and thus bosonic. This
is why they indeed satisfy commutator (as opposed to anti-commutator) relations.
Significance of the ghost Virasoro algebra and criticality
The algebra of the conformal transformations of the full action S = SX + Sg is generated by the
combined Virasoro generators
Ltot (X) (g)

m = Lm + Lm − a
tot
δm,0 , (3.171)
where we conventionally include a total normal ordering constant atot into the definition of Ltot
m .
It is the sum of the normal ordering constant for the X and for the ghost fields,
atot = a(X) + a(g) . (3.172)
• The first piece is just given by a(X) = d−2 2

24 + 24 corresponding to the d − 2 transverse
X-oscillations familiar from lightcone quantisation together with the contribution from the
X 0 and X d−1 components. These are not absent here since we are in a covariant gauge.
• To compute a(g) we observe that the ghost system counts as one anti-commuting set of inte-
ger moded scalars. By a similar computation as performed in section 3.2.3 its contribution
to the Casimir energy is
∞ ∞
X X 1 1
−a(g) = (−n) = − n= ⇒ a(g) − . (3.173)
n=1 n=1
12 12
The different sign compared to the X-sector is a consequence of the anti-commuting nature
(g)
of the ghost field and technically results from the specific form of Lm . Note also the absence
(g)
of a factor of 12 in the normalization of Lm .
• Therefore
d−2 2 1 d−2
atot = + − = ≡ a. (3.174)
24 24 12 24
The ghost system cancels the contribution from the unphysical non-transverse polarisations,
a feature that we will encounter again in the framework of BRST quantisation.
One may verify that the combined Virasoro generators satisfy the commutation relations
ctot
[Ltot tot tot
m , Ln ] = (m − n) Lm+n + δm+n,0 (m3 − m) + 2m(a − 1) (3.175)
12
with the central extensions governed by the quantities
ctot = c(X) + c(g) , c(X) = d (for propagation in R1,d−1 ), c(g) = −26. (3.176)
The presence of the central term ctot is equivalent to a Weyl anomaly of the full action SX + Sg ,
or equivalently to a Weyl anomaly in the path integral
Z
DXei SX (det P ). (3.177)
The central term and thus also the Weyl anomaly of the path integral is absent iff
d = 26 , a = 1. (3.178)
Thus criticality arises as a self-consistency requirement of the Faddeev-Popov treatment of the

path integral.
This gives us a final interpretation of the meaning of criticality: It is the requirement that the
X-theory cancels the conformal anomaly of the ghost system,
!
0 = c(X) + c(g) = c(X) − 26 (3.179)
so that the anomaly of the full quantum theory is absent. What is actually fixed is not the
number of spacetime dimensions, but the central extension c(X) .
3.3.3 BRST Quantisation

The presence of the Faddeev-Popov ghosts, which exhibit the wrong spin-statistics correlation,
raises the question of the physical state condition. In path integral quantization of gauge theories
the physical state condition is implemented via the important concept of the BRST symmetry.
The starting point is the observation that the full action SX + Sg after gauge fixing ĥ = ηab
enjoys a global, fermionic, residual symmetry.
Let be a constant Grassmann parameter. Then this symmetry is generated by the transforma-
tions
δ X µ = (c+ ∂+ + c− ∂− )X µ ,
δ c± = (c+ ∂+ + c− ∂− )c± , (3.180)
(X) (g)
δ b±± = i (T±± + T±± ).
Note that the transformations of X µ are just the conformal Killing transformations with param-
eter c± . Here c± does depend on ξ ± as in c± (ξ ± ), but the fermionic parameter is independent
of ξ ± . It is in this sense that the residual symmetry is global. This symmetry is named BRST
symmetry (after Becchi, Rouet, Stora, Tyutin).
Comment:
Here were are considering the gauge fixed action SX +Sg obtained by explicitly plugging ĥab = ηab
into the action. Alternatively one can implement the gauge fixing condition as an extra Gaus-
sian factor in the path integral by introducing a set of auxiliary fields whose equations of motion
would classically enforce the gauge fixing condition. In this formulation the transformation of
the X µ -fields is a fermonic version of the full original gauge symmetry (here diffeomorphisms and
Weyl rescalings). It is only in the above gauge fixed formulation that the BRST transformation
of X µ reduces merely to the residual symmetry (here the conformal Killing transformations).
We compare these two formulations in detail in the tutorial.
Via Noether’s theorem one can define a BRST charge operator QB as the conserved charge
associated with a suitable BRST current.
As always, this charge will then generate the underlying symmetry. Explicit application of the
Noether procedure confirms that the BRST charge is fermonic as expected. It generates the
BRST symmetry in the sense that
δ X µ =[QB , X µ ], (3.181)
δ c± = QB , c± ,

δ b±± = {QB , b±± } . (3.182)
Note the appearance of the commutator for the action of the fermonic BRST operator on a
bosonic field, but of the anti-commutator for its action on the fermonic ghost fields.
One can show explicitly that (for open strings)
∞
(X) (g)
X
QB = : (L−m + 12 L−m − a δm,0 )cm : (3.183)
m=−∞
does the job (and analogously for the left- and right-moving charges in the closed string). In
particular
Q†B = QB . (3.184)
11
An important property of the BRST symmetry is that it is nilpotent ,
δ δ0 Φ = 0 for Φ ∈ {X µ , b, c}. (3.185)
This must translate into the crucial relation
Q2B = 0. (3.186)
1
In the quantum theory, evaluation of Q2B
= {QB , QB } is complicated by normal ordering
2
subtleties. An explicit computation, which again we are not performing here, yields
∞
X
Q2B = 1
2 {QB , QB } = 1
2 ([Ltot tot tot
m , Ln ] + (m − n)Lm+n )c−m c−n . (3.187)
m,n=−∞
11 This manifestly holds in the more general formulation discussed in the tutorial in which the concrete gauge-
choice is implemented with the help of an auxiliary field, and is true in the simplified gauge-fixed formulation
here up to equations of motion.
This vanishes if and only if the full Virasoro algebra is non-anomalous, which in turn is the case
for the critical string with (d = 26, a = 1):
Consistency of the BRST symmetry is equivalent to absence of the total Weyl anomaly.
The actual significance of the BRST symmetry is that it gives the correct physical state condition.
A physical state must be gauge invariant. Given the relation between the gauge transformations
and the BRST symmetry it is therefore reasonable to expect that a physical state must be in-
variant under a BRST transformation. Indeed, in the tutorials we will give a formal argument
for this assertion in the more general, non-gauge fixed formulation alluded to above. The argu-
ment departs from the requirement that the physical Hilbert space must be independent of the
concrete gauge fixing condition - i.e. in our case the concrete choice of reference metric ĥab . Let
us therefore borrow the result that a necessary condition for a state to be physical is that
QB |phys i = 0. (3.188)
Indeed since QB acts on X as the (residual) symmetry this implements in particular the
constraints resulting from gauge fixing (here the Virasoro constraints).
This, however, is not enough. Namely there exists a large set of trivially physical states given by
|χ i = QB |Ψ i, for |Ψ i arbitrary. (3.189)
Indeed since Q2B = 0 any such χ satisfies the above criterion
QB |χ i = Q2B |Ψ i = 0.
These states are null, i.e. they are orthogonal to all physical states including themselves,
h physical| QB |Ψ i = 0, because QB = Q†B , (3.190)

h χ| |χ i = h Ψ| Q2B |Ψ i = 0, (zero norm). (3.191)
Let us make the following definitions:

• States in the kernel of QB , |χ i s.t. QB |χ i = 0, are called Q-closed.
• States in the image of QB , |χ i = QB |Ψ i, are called Q-exact.
To define a positive norm Hilbert space we need to divide the set of Q-closed states by the set
of Q-exact states.
An at this stage necessary condition for a physical state is that it is a non-trivial element of
Hclosed
HBRST = ≡ cohomology of QB . (3.192)
Hexact
States differing by elements of Hexact are in the same equivalence class = cohomology class:
|Ψ i ∼
= |Ψ i + QB |χ i. (3.193)
Comments:
• We will momentarily find that only a subset of HBRST defined by another condition really
corresponds to the physical states in the sense of participating in the amplitudes.
• One can formally prove that the space HBRST modulo this extra condition is of positive
norm. For this we refer to [P], volume 1, section 4.4. Instead of going through this formal
proof we demonstrate below that the BRST-cohomology correctly implements the Virasoro
constraints.
• The concept of a cohomology as the kernel over the image is defined in mathematics for
every nilpotent operator. The probably most famous example is the exterior derivative d
that maps a p-form to a p + 1-form. In this context the p-th cohomology group is defined
as
closed p-forms
Hp = . (3.194)
exact p-forms
Representation theory
To make all of this explicit we need to define a vacuum |0tot i = 0(X) i ⊗ 0(g) i for the full theory
defined by S tot = S (X) + S (g) , act with creation operators associated with X, b and c on each
factor and then implement the physical state condition. Recall that in the X-sector 0(X) i is
µ (X)
defined by α|m| 0 i = 0 for m 6= 0 together with a suitable action of the zero mode operator
µ µ
α0 ' p . This action is defined by taking (X)theµ vacuumµas an eigenstate of α0µ . More precisely,
(X)
the vacuum carries a continuous label 0 , p i and α0 0 , pµ i ' pµ 0(X) , pµ i. Similarly in

the ghost sector a sensible assignment is to declare that
• c−n , b−n , n > 0 act as creators,
• cn , bn , n > 0 act as annihilators.
This is consistent with the normal ordering prescription (”creators to the left”) and with the
form of the zero-level Virasoro generator
∞
(g)
X
L0 = n b−n cn + n c−n bn . (3.195)
n=1
(g)
Since the ghost Hamiltonian H (g) ∝ L0 we are reassured that b−n , c−n take the role of creators.
What remains is to determine the action of the ghost zero modes c0 , b0 on the vacuum. From
the anti-commutation relations we deduce that the zero-modes c0 , b0 form an algebra defined by
n o
c20 = 0 = b20 , c0 , b0 = 1. (3.196)
A state in the ghost sector must furnish a representation of this algebra, i.e. the action of
c0 and b0 on the vacuum must be compatible with the relations of the algebra. The smallest
representation contains two states |↑ i, |↓ i such that
c0 |↓ i = |↑ i , b0 |↑ i = |↓ i , c0 |↑ i = 0 , b0 |↓ i = 0. (3.197)
To see that we need at least two states let us try our luck with only a single state |Ψ i. Then
c0 |Ψ i = α |Ψ i, b0 |Ψ i = β |Ψ i, for some α, β ∈ C. (3.198)
It is now easy to construct a contradiction because
|Ψ i = (c0 b0 + b0 c0 ) |Ψ i = 2αβ |Ψ i,
but 0 = c20 |Ψ i = α2 |Ψ i. (3.199)
We therefore stick to the above simplest non-trivial representation given by |↑ i, |↓ i . A priori
we could make two inequivalent choices for the full vacuum:

1.) |0tot i = 0(X) i ⊗ |↑ i ≡ |0, ↑ i
The vacuum is annihilated by c0 and by αn , cn , bn with n > 0.

2.) |0tot i = 0(X) i ⊗ |↓ i ≡ |0, ↓ i
The vacuum is annihilated by b0 and αn , cn , bn with n > 0.
To gain some intuition which is the correct

one we evaluate the BRST condition on the special
subset Ψ(X) i ⊗ |↑ i and, respectively, Ψ(X) i ⊗ |↓ i contained in the spectrum that results from
the two choices of vacua. This will give us a heuristic argument as to which of the two vacua to
choose, which can then be extended to a proper physical state theorem.

Consider first case 2.). The physical state condition implies QB |χ i = 0. For |χ i = Ψ(X) i ⊗ |↓ i
this gives
∞
!
X (X) 1 (g)
0 = QB |χ i = : L−m + L−m − aδm,0 cm : |χ i. (3.200)
m=−∞
2
Using cm |χ i = bm |χ i = 0 for m > 0 and setting a = 1 this becomes

" #
! (X)
X
0 = QB |χ i = (L0 − 1)c0 + c−m L(X)
m |χ i. (3.201)
m>0
Evaluating the action of the ghost modes on the vacuum yields

(X) ! (X) !
(L0 − 1) Ψ(X) i = 0 and L(X) i = 0 ∀ m > 0. (3.202)

m Ψ
This recovers the correct constraints from OCQ.

By contrast, case 1.) with vacuum |0, ↑ i does not allow us to recover the known constraints in
this simple fashion. This suggests that only |0, ↓ i is a meaningful vacuum. Note that the two
vacua are distinguished by the defining property b0 |0, ↓ i = 0.
Note: For a point particle the physical state condition is simply the mass shell condition p2 +m2 = 0. Q-
closed states of the form |p i⊗|↓ i satisfy
this and are thus sensible physical states, in the same manner as
Q-closed states of the special form Ψ(X) i⊗|↓ i automatically satisfy the Virasoro constraints. In addition

one can convince oneself that Q-closed modulo Q-exact states of the form |p i ⊗ |↑ i, i.e. constructed
from the first vacuum, give another copy of physical, i.e. on-shell states; however, these can be shown to
decouple from all scattering amplitudes and so must be discarded. For a derivation of these assertions
see [P], volume 1, p.129-131.
For the string theory the same logic goes through even though the details are technically more involved
due to the oscillator degrees of freedom. It is still true that a consideration of scattering amplitudes re-
veals that only BRST closed (modulo exact) states built from Ψ(X) i ⊗ |↓ i can contribute to amplitudes

(see [P], volume 1, Chapter 9). In addition one must impose the constraint b0 |Ψ i = 0, which generalizes
the defining property b0 |0, ↓ i = 0 of the vacuum.
This can be summarized in form the following
Theorem: (Proof: Polchinski I, Chapter 4.4)

The positive norm physical states are the states QB |Ψ i = 0 modulo |Ψ i = QB |χ i built on
|0tot i = |0, p i(X) ⊗ |↓ i(g) that satisfy in addition b0 |Ψ i = 0.
In the course of the proof of this theorem one also finds that every non-trivial QB -cohomomology
class contains one representative of the form
|ψ iX ⊗ |↓ i. (3.203)
with zero ghost excitations.
Example: Level (1) open excitations
For the first excited level we make the ansatz
|Ψ i = (ξ · α−1 + βb−1 + γc−1 ) 0tot i.

(3.204)
This gives us 26 + 2 states to begin with.
On Assignment 7 we will work out the constraints and find the following structure:
i) From b0 |Ψ i = 0 we deduce 0 = {QB , b0 } |Ψ i = Ltot
0 |Ψ i. This yields the mass shell
condition p2 = 0.
ii) QB |Ψ i = 0 leads to
!
((p · ξ)c−1 + βp · α−1 ) 0tot i

0 =
⇒ p · ξ = 0 & β = 0. (3.205)
Requiring Q-closedness therefore removes the unphysical anti-ghost excitations as well as all
polarisations that are not orthogonal to the momentum, thereby eliminating 2 out the 26+2
original states.
iii) To analyse |Ψ i ∼ = |Ψ i + QB |χ i we observe that for a general state

|χ i = (ξ 0 · α−1 + β 0 b−1 + γ 0 c−1 ) |0tot i at level 1 we have
QB |χ i = ((p · ξ 0 )c−1 + β 0 p · α−1 ) 0tot i.

This shows that c−1 |0tot i is BRST exact and the polarisation vector is only defined up to
the equivalence
ξ∼
= ξ + β 0 p, β 0 ∈ C. (3.206)
(Note that this is consistent with the conditions i) and ii).)
Thus we are left with 24 physical positive norm states, as required.

Chapter 4
Conformal field theory (CFT)
Conformal symmetry is, loosely speaking, invariance under rescaling. Conformally invariant
systems possess no intrinsic length, mass or energy scale. In particular there exists no notion of
massive particles or massive excitations as these would induce a reference scale.
Conformal invariance plays an important role in many physical systems such as
• string theory, where the worldsheet theory is a two-dimensional conformal field theory;
• at fixed points of the renormalisation group (RG) equations in QFT, at which a theory
becomes scale invariant;
• near critical points in condensed matter or statistical physics, at which the correlation
length diverges, leaving us again with a scale invariant theory;
• the AdS/CFT correspondence, which relates gravity on AdS space with a CFT on its
boundary.
The treatment of such conformally invariant theories necessarily differs from the treatment of
non-conformal systems in usual QFT. Two-dimensional conformal field theories are even more
special because in two dimensions the group of infinitesimal conformal transformations is infinite.
This is large enough as to sometimes allow us to solve the theory exactly and completely. Such
theories are called integrable and the search for integrable structures also in higher dimensional
theories, e.g. for certain field theories in four dimensions (N =4 Super-Yang-Mills theory), has
become an important challenge in theoretical physics. The treatment of two-dimensional CFTs
makes use of powerful methods of holomorphic analysis and has developed into an independent
field of mathematical physics.
4.1 Conformal invariance

Definition: A conformal transformation is a diffeomorphism under which the metric changes
only by an overall factor.
In the sequel we consider flat d-dimensional space of arbitrary signature, i.e. Rm,n with m+n = d.
A conformal transformation is then a diffeomorphism x → x0 (x) under which the metric ηαβ
transforms like
∂xα ∂xβ
ηµν → ηαβ 0µ 0ν = Λ(x)ηµν (4.1)
∂x x
67
CHAPTER 4. CONFORMAL FIELD THEORY (CFT) 68
for some conformal factor Λ(x). Infinitesimally with x0µ = xµ −µ (x)+O(2 ) and Λ(x) = eω(x) =
1 + ω(x) + . . . this gives
∂µ ν + ∂ν µ = ω(x)ηµν . (4.2)
Further elementary manipulations carried out on Assignment 7 yield the following set of equations
satisfied by ω(x),
2
ω(x) = ∂ · ,
d
ηµν ∂ 2 + (d − 2)∂µ ∂ν ω(x) = 0,

(4.3)
2
(d − 1)∂ ω(x) = 0.
Technically this is the origin why we must distinguish the cases d = 2 and d ≥ 3: For d = 2 the
second equation is vacuous.
4.2 The conformal group in d ≥ 3

Even though we will mostly be interested in conformal symmetry in two dimensions we briefly
consider Rn.m , d = n + m ≥ 3.
The set of infinitesimal conformal transformations can be deduced by finding the most general
solution of the constraints (4.3). As discussed on Assignment 7, these are
• translations x → x0 = x + a,
• Lorentz transformations xµ → x0µ = xµ + mµν xν with mµν = −mνµ ,
• dilatations x → x0 = (1 + α)x,
• special conformal transformations (SCT) x → x0 = x + 2(x · b)x − (x · x)b.
The global version of the special conformal transformations is
x0 x
= − b. (4.4)
(x0 · x0 ) (x · x)
x
This can be understood as a successive application of an inversion x → x2 , a translation
and another inversion.
Note that global special conformal transformations take the form
x − (x · x)b
x0 = .
1 − 2(b · x) + (b · b)(x · x)
This is infinite at those points where the denominator vanishes. We conclude that in order
for special conformal transformations to be globally defined we must consider the conformal
compactification Rn,m ∪ ∞.
This shows that we must carefully distinguish between
a) the group of globally defined conformal diffeomorphisms, called conformal group and its alge-
bra, the conformal algebra and
b) the group of infinitesimal conformal transformations.
In particular the conformal group depends on the topology of the space we consider as seen above.
E.g. for Rn,m ∪ ∞ and d = n + m ≥ 3 the conformal group has 12 (d + 1)(d + 2) parameters.
Indeed one can prove the following theorem:
The conformal group acting on Rn,m ∪ ∞ is isomorphic to SO(m + 1, n + 1).
4.3 The conformal group in d = 2

In 2 dimensions the constraints (4.2), (4.3) reduce to
∂µ ν + ∂ν µ = (∂ · ) ηµν . (4.5)
In the sequel we consider R2 with Euclidean signature ηµν = δµν and complex variables
z = x0 + ix1 , z = 0 + i1 ≡ ,
z̄ = x0 − ix1 , ¯z̄ = 0 − i1 ≡ ¯,
1 1
∂z = (∂0 − i∂1 ), ∂z̄ = (∂0 + i∂1 ),
2 2
1
d2 x = dx0 dx1 = dzdz̄.
2
4.3.1 Infinitesimal conformal transformations

In complex variables the constraint (4.5) implies
∂z̄ = 0 = ∂z ¯ ⇒ = (z) & ¯ = ¯(z̄). (4.6)
We have thus established that the group of infinitesimal conformal transformations in Eu-
clidean 2 dimensions is generated by all meromorphic functions (z) and anti-meromorphic
functions ¯(z). Note that we say meromorphic, not holomorphic because for infinitesimal con-
formal transformations we do allow for singularities outside the open set under consideration.
The (anti-)meromorphic generators can be expanded in a Laurent series as
X
z → z 0 = z + (z) = z + n z n+1 , (4.7)
n∈Z
X
0
z̄ → z̄ = z̄ + ¯(z̄) = z̄ + ¯n z̄ n+1 . (4.8)
n∈Z
A basis of generators for the algebra of infinitesimal conformal transformations is
ln = −z n+1 ∂z , ¯ln = −z̄ n+1 ∂z̄ , for n ∈ Z. (4.9)
These generators obey the commutation relations of the Witt algebra
[lm , ln ] = (m − n)lm+n , (4.10)

¯lm , ¯ln = (m − n)¯lm+n ,

(4.11)
¯lm , ln = 0.

(4.12)
4.3.2 The Möbius group as the group of global conformal transforma-

tions
By contrast, the group of global conformal diffeomorphisms depends on the topology of the
2-dimensional space. In particular, as for d ≥ 3, in order for special conformal transformations
to be globally defined we need to compactify.
In the sequel we focus on the space S 2 ∼
= C ∪ ∞. The equivalence of S 2 and C ∪ ∞ can be seen
via the stereographic projection. More complicated topologies, e.g. that of a torus T 2 , will be
discussed later.
Even if we restrict to S 2 not all generators ln are well-defined globally, i.e. at all points.
• At z = 0 ln = −z n+1 ∂z is well-defined only for n ≥ −1.
1
• To analyse the behaviour at z = ∞ we introduce the variable w = z and consider the
bahaviour at w = 0 via
1
∂z = − ∂w = −w2 ∂w ⇒ ln = w−(n+1) w2 ∂w = w1−n ∂w . (4.13)
z2
Thus ln is well-defined at z = ∞ for n ≤ 1.
The group of finite conformal diffeomorphisms on S 2 is generated by l−1 , l0 , l1 and ¯l−1 , ¯l0 , ¯l1 .
Geometric interpretation:
• l−1 = −∂z is the generator of rigid translations z 7→ z + b, b ∈ C.
• l0 = −z∂z is the generator of complex dilatations z 7→ az, a ∈ C.
These are best analysed in polar coordinates z = reiϕ , in terms of which l0 = − 2r ∂r + 2i ∂ϕ
and ¯l0 = − 2r ∂r − 2i ∂ϕ . Thus
l0 + ¯l0 = −r∂r generates dilatations, (4.14)
i(l0 − ¯l0 ) = −∂ϕ generates rotations. (4.15)
• l1 = −z 2 ∂z generates special conformal transformations because cl1 z = −cz 2 is the infin-

tesimal version of
z
z 7→ . (4.16)
cz + 1
Altogether a globally defined conformal diffeomorphisms can be brought into the form
az + b
z 7→ (4.17)
cz + d
subject to the following constraints:
• Invertibility requires ad − bc 6= 0. Thus we can rescale to ad − bc = 1 without changing
the transformation. Note that four complex numbers (a, b, c, d) with ad − bc 6= 0 can be
viewed as the entries of a complex 2 × 2-matrix with unit determinant. These parameters
therefore generate the special linear group SL(2, C).
• Even after rescaling to ad − bc = 1 the set of parameters (a, b, c, d) and (−a, −b, −c, −d)
give the same transformation. To avoid redundancy we must therefore divide SL(2, C) by
the Z2 action (a, b, c, d) → (−a, −b, −c, −d).
The group of conformal diffeomorphisms on S 2 is the Möbius group SL(2, C)/Z2 = PSL(2, C).
An important property of PSL(2, C) is that it maps any 3 distinct points to any other 3 distinct
points. This will be discussed in the exercises.
4.3.3 Relation between the complex (half-)plane and the worldsheet

on the cylinder (or strip)
a.) Closed strings on the cylinder
For closed strings we had defined the theory on the cylinder with coordinates and metric

± −1 0
(τ, σ), ξ = τ ± σ, η = . (4.18)
0 1
To relate this to the Euclidean theory on S 2 we first perform a Wick rotation τ = −iτ 0 and
relabel τ 0 → τ . The new metric in the Euclidean coordinates (τ, σ) is

1 0
η= . (4.19)
0 1
We then define the coordinates
ω = τ − iσ, ω̄ = τ + iσ. (4.20)
The map
2π 2π
ω 7→ z(ω) = e ` ω =e ` (τ −iσ) (4.21)
is the conformal mapping from the cylinder to the complex plane C ∪ ∞.
In particular it maps
τ = −∞ −→ z=0 (4.22)
τ = +∞ −→ z = ∞. (4.23)
This property will be crucial when considering asymptotically in- or outgoing states in scatter-
ing amplitudes and is the basis for the famous operator-state-correspondence in CFTs, to be
discussed later.
We can furthermore identify the transformations
2π
τ 7→ τ + a ←→ z 7→ e ` a z (dilatation), (4.24)
−i 2π
` b
σ 7→ σ + b ←→ z 7→ e z (rotation). (4.25)
The geometric action of l0 + ¯l0 and i(l0 − ¯l0 ) is in perfect agreement with their relation to the
Hamiltonian and the momentum operator as the generators of time and spatial translations,
2π
l0 + ¯l0 : dilatation ↔H= (L0 + L̄0 ), (4.26)
`
2π
i(l0 − ¯l0 ) : rotation ↔ P = − i(L0 − L̄0 ). (4.27)
`
b.) Open strings on the strip
For open strings we have considered the worldsheet in coordinates (τ, σ) with the topology of a
strip with boundaries at σ = 0 and σ = `. By a similar conformal mapping as (4.20), but with
adjusted periodicities,
π
(τ + iσ) 7→ e ` (τ +iσ) , (4.28)
this is mapped to the upper half plane. Note that we are have changed the sign in front of σ to
ensure the conventional choice that the open string is defined on the upper as opposed to lower
half-plane. The boundary of the strip is mapped to the real line Im(z) = 0, with the half-lines
z > 0 and z < 0 corresponding to the two boundaries at σ = 0 and σ = `, respectively. The
boundary conditions now become reflection conditions for the conformal fields on the real line.
This gives rise to the concept of boundary CFT, which has developed into a rich subfield by
itself. Sadly we do not have the time to dig into this topic in more detail and will mostly be
working with CFTs on the sphere, corresponding to closed string worldsheets, in what follows.
4.4 Conformal fields and their OPE

A CFT is a physical theory invariant under the group of infinitesimal conformal transformations.
In particular, as anticipated already, there is no intrinsic notion of length scale or of massive ex-
citations.
Even though the definition of a CFT is much more general, let us first take the lagrangian
perspective, whose logic is familiar from the definition of general QFTs:
• Our starting point is a classical theory defined by an action S[φi (x)] with the specific
requirement that this action is invariant under infinitesimal conformal transformations.
• The basic objects of this theory are the fields Oi (x). By this we mean, by slight abuse of
notation, any local expression built out of the φi (x) appearing in the action and derivatives
thereof. Local fields are e.g. products or power series such as exponentials of φi (x). These
Oi (x) are also called local operators.
• The quantum theory is determined by the correlation functions

Z
1
hO1 (x1 ) . . . On (xn )i = Dφi e−S[φi ] O1 (x1 ) . . . On (xn ). (4.29)
Z
Importantly, the expression inside h. . .i is always time-ordered as is familiar from the treat-
ment of correlation functions in usual QFTs.
• In writing equations involving the local Oi we will always think of operator equations in
the full quantum theory, i.e. think of the operators as inserted into a time-ordered path
integral as above. For instance an equation of the form
O1 (x1 ) O2 (x2 ) = f (O1 (x1 ), O2 (x2 )) (4.30)
is shorthand for
hO1 (x1 ) O2 (x2 ) . . .i = hf (O1 (x1 ), O2 (x2 )) . . .i (4.31)
with . . . representing arbitrary operator insertions at a distance bigger that |x1 − x2 |.

Now comes an important, though probably unfamiliar point: Conformal symmetry allows for a
rather different definition of the theory than the one above starting from a classical action. A
generic CFT in d dimensions need not have a description via an action. Rather it is defined
by a ’complete set’ of local fields Oi and their correlation functions. If we do have a lagrangian
description, these correlation functions are given as above. More generally, however, we can think
of the correlators as maps from the space of operators to C consistent with conformal invariance.
We will see that this is very constraining. If really all correlators are known in terms of a finite
amount of input data the theory is solved completely and defined through these data.1
There are essentially two reasons why this can work: First, because there is a special notion of
a ’complete set of operators’ available in a CFT which does not exist in a general QFT - the set
of quasi-primary fields - and second, because the operator product expansion (OPE) of two such
quasi-primaries has remarkable properties. Let us introduce both concepts in turn.
1) Primary and quasi-primary fields
Since in this course we are mainly interested in the applications two-dimensional CFTs to string
theory we restrict the following discussion, unless states otherwise, to a d = 2 CFT
on S 2 = C ∪ ∞, with general fields of the form O(z, z̄).
1) If a field Φ(z, z̄) transforms under z 7→ z 0 = λz, λ ∈ C as
Φ(z, z̄) 7→ Φ0 (z 0 , z̄ 0 ) = λ−h λ̄−h̄ Φ(z, z̄) (4.32)
then it has conformal dimension (h, h̄). Note that in general h̄ 6= h∗ .

7 z0 =
2) A primary field Φ(z, z̄) transforms as a tensor under conformal tranformations z →
f (z):
−h ¯−h̄
0 0 0 ∂f ∂f
Φ(z, z̄) 7→ Φ (z , z̄ ) = Φ(z, z̄). (4.33)
∂z ∂ z̄
Note: In particular, and in the spirit of the discussion around (4.31), we require as part of
the defining property of a primary field that any correlation function involving primary fields
transforms as
Y Y ∂f −hi ∂ f¯ −h̄i Y
h Φ(zi , z¯i )i 7→ |z |z̄ h Φ(zi , z¯i )i. (4.34)
i i
∂z i ∂ z̄ i i
1 A CFT in d = 2 dimensions is indeed exactly solvable in this sense. This is because the two-dimensional
Virasoro algebra is infinite-dimensional. In higher dimensional CFTs the computation of all correlators in terms
of finite data is possible in principle, but much harder in practice due to the lack of this extra symmetry.
Expanding f (z) = z + (z) + . . . gives the infinitesimal scaling behaviour for primary fields
δ,¯ Φ(z, z̄) = −(h∂z + ∂z + h̄∂z̄ ¯ + ¯∂z̄ )Φ(z, z̄). (4.35)
3) A quasi-primary field satisfies (4.33) for f ∈ P SL(2, C). In particular every primary is a
quasi-primary, but not the other way round.
4) A chiral field is a field Φ(z), an anti-chiral field is a field Φ(z̄).

Remarks:
• Quasi-primaries are tensors under the group of globally defined conformal transformations.
In a d-dimensional CFT with d > 2 (on Rm,n with m + n = d), these are the fields with
specific transformation behaviour under SO(m + 1, n + 2).2 All the statements we make
about the quasi-primaries based on their transformations under P SL(2, C) transformations
in a two-dimensional CFT carry over analogously to quasi-primaries in higher-dimensional
CFTs.
• What has no analogue in higher dimensions is the concept of a primary field, which exploits
the infinitesimal structure of the two-dimensional Virasoro algebra.
Mode Expansions
Before we proceed let us define the mode expansion of two-dimensional primaries. Consider the
transformation from the worldsheet on the cylinder to S 2 = C ∪ ∞.
Suppose on the cylinder a field Φ is purely left-moving and has the mode expansion
X −
ΦL (ξ − ) = φn e−i nξ ( 2π h
` ) . (4.36)
n
Then it is easy to show that if Φ is primary of weight h, the corresponding expansion on S 2 is

X
Φplane (z) = z −n−h φn . (4.37)
n
See Ass. 8 for a proof. The generalisation to fields with chiral and anti-chiral pieces is obvious.
The modes φn can be obtained with the help of the residue theorem as
I
1
φn = dz Φ(z) z n+h−1 . (4.38)
2πi
2) The Operator Product Expansion
The second important concept in a CFT is that of the operator product expansion (OPE).
In a general (Lorentz-invariant) QFT, the OPE is defined as an approximative expansion of two
operators Oi (xi ) and Oj (xj ) valid in the limit xi − xj → 0,
X
k
Oi (xi )Oj (xj ) = Cij (|xi − xj |) Ok (xk ). (4.39)
k
| {z }
functions in C
2 Note that in many texts on higher-dimensional CFTs the term primary is used for what we call quasi-primary.
OPEs of the above type are defined generally in QFT as convergent series in a certain open
neighbourhood of the operators.
In a d-dimensional CFT the structure of the OPE is much more powerful. This is because the
OPE In a d-dimensional CFT satisfies three key properties that rely on conformal invariance,
which we state here without proof:
• In a d-dimensional CFT the OPE of two quasi-primaries involves only other quasi-primaries
and their derivatives, the so-called descendent fields.
k
• The functional dependence of the Cij (|xi − xj |) is completely fixed by conformal invariance.
• The OPE is an exact expression, i.e. an asymptotic series with radius of convergence the
distance to the next field insertion when viewed as an operator equation.
To illustrate this let us specify to a general d = 2 CFT. It can be shown that the OPE of two
quasi-primaries φi (z) and φj (w) (taken to be chiral for brevity) can be written as
X anijk 1
φi (z)φj (w) = Cij k ∂ n φk (w), (4.40)
n! (z − w) +hj −hk −n
hi
k,n≥0
−1
2hk + n − 1 hk + hi − hj + n − 1
anijk = , (4.41)
n n
where the sum over k on the right involves only quasi-primaries. For a proof see e.g. [BP],
Chapter 2.6.3. What is important for us is the appearance of the so-called structure con-
k
stants Cij . The structure constants directly determine the 3-point functions of the CFT. Then,
by successive application of the OPE we have a chance of reducing all higher n-point functions
to lower correlators. This is the idea behind defining the CFT in terms of a finite amount of data.
To summarise, of special importance in a (two-dimensional) CFT is the set of quasi-primaries

(primaries) and their OPE. We will now learn how to deduce the OPE for certain fields.
4.5 Conformal Ward-Takahashi identities and energy-momentum

tensor
In this section we will demonstrate the importance and power of the operator product expansion.
Our aim is to compute the OPE between the energy momentum tensor and a conformal field.
This will also shield new light on the nature of the energy momentum tensor and of the conformal
anomaly.
4.5.1 General Ward-Takahashi identities

To set the stage we derive here the Ward-Takahashi identities for a general field quantum field
theory, with specialisations to CFTs reserved for the next section. Consider therefore a general
QFT in d dimensions defined by the path integral
Z
Z = Dφ e−S[φ] . (4.42)
Suppose the theory enjoys a global symmetry
φ → φ0 = φ + δφ with constant. (4.43)
That is, the classical action transforms as
S[φ] → S 0 [φ0 ] = S[φ]. (4.44)
Suppose furthermore that the symmetry is non-anomalous, i.e. also the measure is invariant,
Dφ → Dφ0 = Dφ. (4.45)
• Quantum version of Noether’s theorem:

We know that a classical continuous symmetry implies the existence of conserved current
∂α J α = 0. The quantum version of this statement is that ∂α J α (x) = 0 holds as an
operator equation, i.e. together with arbitrary operator insertions away from x under the
path integral. To derive this quantum version of Noether’s theorem for a non-anomalous
global symmetry, we start as in the classical case by promoting → (x). Now the measure
Dφ and the action S[φ] are no longer separately invariant, but the combined change of
the partition function due to the transformation of the measure and of the action can only
involve terms proportional to ∂α (x). We can use this observation to define the Noether
current by parametrising the change in Z as3
Z Z
0 1
R α
Z → Z 0 = Dφ0 e−S[φ ] = Dφ e−S[φ]− 2π J ∂α (x) =
Z Z Z Z
−S[φ]
1 α −S[φ]
1
= Dφ e 1− J ∂α (x) = Dφ e 1+ ∂α J α (x) .
2π 2π
(4.46)
On the other hand Z Z

0
−S[φ]
Z= Dφ e = Dφ0 e−S[φ ] = Z 0 (4.47)
because we just changed the integration variable. Note that this is not a contradiction to
the statement that the transformation with replaced by (x) is no longer a symmetry of
the theory, because S[φ0 ] and the measure Dφ0 are not invariant independently. Therefore
Z
1
Dφ e−S[φ] ∂α J α (x) = h∂α J α (x)i = 0. (4.48)
Z
To show that this indeed holds as an operator equation in the above sense we consider the
correlator Z
1
hO1 (x1 ) . . . On (xn )i = Dφe−S[φ] O1 (x1 ) . . . On (xn ). (4.49)
Z
3 The so-defined J α receives contributions both the classical action S and, in general, from the transformation
of the functional measure.

Under the symmetry (4.43) a local operator transforms as
Oi → Oi + δOi . (4.50)
We now promote → (x) such that (x) vanishes at the insertion of the operators in the
correlator, (x) = 0.

x=xi
The same steps as before yield 0 = h∂α J α (x)O1 (x1 ) . . . On (xn )i i.e.
0 = ∂α J α as an operator equation. (4.51)
To avoid confusion with what comes next, we hasten to stress that this should really be
read as the statement that 0 = h B ∂α J α (x)O1 (x1 ) . . . On (xn )i for an arbitrary region B
R
that does not include any of the operators Oi .

• Ward-Takahashi identities:
Now let (x) have support only in a region B that contains the insertion x1 of operator
O1 , but none of the other operators.
The correlator hO1 . . . On i transforms as

Z Z
1 −S[φ] 1 α
hO1 . . . On i → Dφ e 1+ ∂α J (x) (O1 (x1 ) + (x1 )δO1 ) O2 . . . On .
Z 2π B
(4.52)
Let us restrict to constant inside B . For this we deduce
Z
1
− h ∂α J α (x) O1 (x1 ) . . .i = hδO1 (x1 ) . . .i, (4.53)
2π B
i.e.
Z
1
− ∂α J α (x) O1 (x1 ) = δO1 (x1 ) . . . as an operator equation. (4.54)
2π B
This is the Ward-Takahashi identity, which gives a tool to compute the transformation
of a local operator by an integral over a certain operator product.
• We finally specialise to a 2d QFT, for which the Ward identitiesR can be rewritten in
a particularly neat manner. By Stoke’s theorem we can evaluate B ∂α J α (x) as a line
integral over the boundary of B ,
Z I
α
∂α J (x) = Jα n̂α . (4.55)
B ∂B
The tangential and normal line element in two dimensions take the form
dx1 dx2

t̂α = , n̂ α
= . (4.56)
dx2 −dx1
Therefore we can write

Z I
α
∂α J (x) = (J1 dx2 − J2 dx1 ). (4.57)
B ∂B
Let us go to complex coordinates: z = x1 + i x2 , z̄ = x1 − i x2 , in which
Jz = J1 + i J2 , J z̄ = J 1 − i J 2 (4.58)
z 1 1
Jz̄ = gz̄z J = 2 (J1 + i J2 ), Jz = 2 (J1 − i J2 ) (4.59)
and therefore
Z I
∂α J α (x) = −i (dz Jz − dz̄ Jz̄ ), Jz = Jz (z, z̄), Jz̄ = Jz̄ (z, z̄). (4.60)
B ∂B
Altogether the Ward-Takahashi identities for a 2-dimensional QFT take the form
I
1
δO(w, w̄) = − (dz Jz (z, z̄) − dz̄ Jz̄ (z, z̄))O(w, w̄), (4.61)
2π i ∂B
where it is important that w lies within the region B , i.e. is encircled by the contour
integral.
4.5.2 Conformal Ward-Takahashi identities

Let us now apply the Ward-Takahashi identities to the conformal symmetry of a 2-dimensional
CFT.
• Recall that in the context of the free boson on the cylinder we have established the gen-
eral result that Noether currents of conformal transformations are related to the energy-
momentum tensor. In lightcone gauge on the cylinder its non-vanishing components are
T++ = T++ (ξ + ) , T−− = T−− (ξ − ) , T−+ = 0. (4.62)

• This easily translates into complex coordinates as
Tzz (z) = T (z), Tz̄z̄ (z̄) = T̄ (z̄) (4.63)
and the Noether current for a conformal transformation z → z + (z) and z̄ → z̄ + ¯(z̄) is
Jz (z) = (z)T (z), Jz̄ (z̄) = ¯(z̄)T̄ (z̄). (4.64)
• Application of eq. (4.61) yields the
Master Formula (Conformal Ward-Takahashi identity):

I
1
δ,¯ O(w, w̄) = − dz((z)T (z) + ¯(z̄)T̄ (z̄))O(w, w̄), (4.65)
2π i Cw
with the contour integral counter-clockwise both in z and in z̄ (thereby explaining the sign
difference of the second term compared to (4.61)).
Note: T (z)O(w, w̄) and T̄ (z̄)O(w, w̄) are radially ordered, i.e. |z| > |w|, |z̄| > |w̄|. This
corresponds to time-ordering inside the path integral Dφ e−S[φ] .
R
• Important remark on the proof:

In the derivation of (4.61) we explicitly assumed that the global symmetry under con-
sideration is non-anomalous, see eq. (4.45). At first sight, this seems to clash with the
existence of a conformal anomaly in theories with c 6= 0. What saves the day is that
in order to derive (4.65) it is sufficient to write down the Ward identifies for the sub-
set of conformal transformations given by global translations, by rotations and by scale
transformations, from which (4.65) can indeed be derived.4 Classically, these are given by
the P SL(2, C) transformations l0 , l−1 , l1 . This group of globally defined conformal trans-
formations is indeed non-anomalous because the corresponding Virasoro operators satisfy
[Lm , Ln ] = (m − n)Lm+n for m, n ∈ {−1, 0, 1}.
• Important conclusion:
The information about conformal transformations is encoded in the residua of the OPE
with the energy-momentum tensor T (z), T̄ (z̄)!
Let us evaluate this for primary fields of dimension (h, h̄), thereby deriving their OPE with the
energy-momentum tensor.
• From (4.35) we recall the form of the infinitesimal conformal transformation of a primary
field,
δ,¯ φ(w, w̄) = −(h∂w (w)φ(w, w̄) + (w)∂w φ(w, w̄)) − (h̄∂w̄ ¯(w̄)φ(w, w̄) + ¯(w̄)∂w̄ φ(w, w̄)).
(4.66)
This stands on the lefthand side of eq. (4.65).
4 For a proof see Di Francesco et al., Conformal Field Theory, p. 121.
• To compare it to the integral on the

righthand side we use the residual formulae ( following
from (z) = (w) + (z − w)∂z (z)z=w + . . .)
I
1 (z)
(w)∂w φ(w, w̄) = dz ∂w φ(w, w̄), (4.67)
C 2π iz−w
I w
1 (z)
(∂w (w))φ(w, w̄) = dz φ(w, w̄). (4.68)
Cw 2π i (z − w)2
• Comparison with the Ward-Takahashi identities therefore gives the OPE
h 1
T (z)φ(w, w̄) = 2
φ(w, w̄) + ∂w φ(w, w̄) + terms regular at z = w
(z − w) z−w
(4.69)
h̄ 1
T̄ (z̄)φ(w, w̄) = 2
φ(w, w̄) + ∂w̄ φ(w, w̄) + . . .
(z̄ − w̄) z̄ − w̄
Again the lefthand side is radially ordered.
4.5.3 T (z) as a conformal field

Let us first analyse the nature of the energy-momentum tensor as a conformal field. We will
argue that T (z) is a quasi-primary field of conformal dimension h = 2 = h̄.
• To see that T (z) has scaling dimensionR h = 2 we note that the Hamiltonian is related to
the energy-momentum tensor via H ∼ dσ T. Thus T has mass dimension 2, [T ] = [E 2 ].
• Even we do not know yet if T (z) is a primary field - in fact we will see that it is not - we
define its modes as in (4.37) via T (z) = n∈Z z −n−2 Ln , i.e.
P
I
1
Ln = dz z n+1 T (z). (4.70)
2π i
• The OPE T (z)T (w) can be computed from the known expression for the commutator
[Lm , Ln ] as
c
2 2 T (w) ∂w T (w)
T (z)T (w) = 4
+ 2
+ + .... (4.71)
(z − w) (z − w) z−w
From this we conclude that if c = 0, then T (z) is a primary field of dimension h = 2.
Comment:
The computation leading to (4.71) is important by itself: After plugging the contour integral
(4.70) into the commutator one can deform the contour and read off the residue. This is discussed
on Assignment 8. This procedure holds more generally:
The information of the OPE is equivalent to the information in the commutators of the modes.
Let us proceed with our investigation of T (z).

• With the help of the same type of computations one can use the OPE for T (z)φ(w) for a
primary φ(w) to derive the commutators

[Lm , φn ] = (h − 1)m − n φm+n , [Lm , φ(z)] = z m z∂z + (m + 1)h φ(z). (4.72)
If this holds only for m = −1, 0, 1, then φ is only quasi-primary.

• Therefore if c 6= 0, then T (z) is a quasi-primary field. Indeed it was shown in the exercises
that there is no central term in the commutators for Lm with m = −1, 0, 1, for which
[Lm , Ln ] = (m − n)Lm+n . Thus the global conformal group on S 2 , P SL(2, C), is always
non-anomalous.
Transformation of T (z) under conformal transformations
• With the help of the conformal Ward identities we can now use the OPE of the energy-
momentum with itself to compute its transformations under conformal transformations:
δ T (w) = −Res[(z)T (z)T (w)]

c
2 2T (w) ∂T (w)
= −Res (z) + + + ... . (4.73)
(z − w)4 (z − w)2 z−w
Expanding (z) = (w) + 0 (w)(z − w) + 21 (z − w)2 00 + 16 (z − w)3 000 + . . . we find

c 000
δ T (w) = −(w)∂T (w) − 20 (w)T (w) − (w). (4.74)
12
One can verify by a straightforward computation that this is the infinitesimal version of
the following transformation under finite z 7→ z̃(z):
−2
∂ z̃ c
T (z) 7→ T̃ (z̃) = [T (z) − S(z̃, z)] (4.75)
∂z 12
in terms of the Schwarzian derivative

−1 2 −2
∂ 3 z̃ ∂ z̃ 3 ∂ 2 z̃

∂ z̃
S(z̃, z) = 3 − . (4.76)
∂z ∂z 2 ∂z 2 ∂z
With a little more work one can show that indeed the infinitesimal transformation (4.73)
together with the correct behaviour with respect to composition of two transformations
uniquely fixes S(z̃, z) to be of this form. For a proof see e.g. [BLT], Chapter 2.1.
• In the exercises we evaluate the Schwarzian derivative for the conformal map from the
2π 2π
cylinder to the complex plane, given by z = e ` w = e ` (τ −i σ) . This allows us to relate the
energy momentum tensor on the cyclinder and on the complex plane as
2
2π c
Tcylinder (w) = z 2 Tplane (z) − . (4.77)
` 24
Relation of c to vacuum energy
The above transformation rule provides a very simple and efficient derivation of the Casimir
energy on the cylinder in terms of the central extension c of the Virasoro algebra. The starting
point is the intuitive assertion that on the complex plane the Casimir energy vanishes.5 Thus,
in view of the relation between the Hamiltonian and the energy-momentum tensor this implies
that the one-point function on the plane is zero, hTplane i = 0. Therefore
2
c 2π
hTcyl. i(w) = − . (4.78)
24 `
This beautifully matches with our earlier computation of the vacuum energy if we remember that
w = iξ − , w̄ = iξ + in terms of the Minkowski signature lightcone coordinates. This gives an extra
factor of −1 in relating the one-point function of Tcyl. (w) to the physical value of the vacuum
energy on the Minkowski signature cylinder. Altogether one finds with H = 2π ` (L0 + L̃0 ) and
`
R` − `
R` +
L0 = − 4π2 0 dσT−− (ξ ), L̃0 = − 4π2 0 dσT++ (ξ ) that the vacuum energy associated with a
single string field in Minkowski signature is
2π Z `
` 2π c
hHcyl. i = (−1) − 2 dσ(hTcyl. (w)i + hT̄cyl. (w̄)i) = − . (4.79)
` 4π 0 ` 12
Taking into account that a single string field has a conformal anomaly c = 1, this agrees with
the result hHi = 2π 1
` (−a − ã) with a = 24 for an integer-moded boson.
Relation of c to conformal anomaly
We have already pointed out several times that c 6= 0 signals an anomaly of the conformal
symmetry of a CFT. The precise statement is that for a 2-dimensional CFT the conformal
anomaly is given by the vacuum expectation value of the trace of the energy-momentum tensor.
In lightcone gauge this corresponds to a non-zero expectation value of T+− or, in complex
coordinates, of Tzz̄ . In (τ, σ) coordinates the statement is that
c (2)
hT a a i = R . (4.80)
12
Here R(2) the Ricci-scalar of the two-dimensional spacetime on which the CFT is defined. Since
classical conformal invariance implies T a a = 0 this indicates a Weyl/conformal anomaly.6
While for a precise proof of this important theorem we refer to [P], Chapter 3.4., we can intu-
itively understand that hT a a i must be proportional to c and to R(2) : The proportionality to c
follows from our above result that c 6= 0 is equivalent to the presence of a vacuum-energy, which
breaks scale invariance. Furthermore, non-vanishing curvature implies the notion of a typical
length scale in the theory, which again breaks scale invariance.
5 More precisely, the zero-point energy vanishes for a CFT on R2 , which is just flat infinitely extended spacetime.
This follows from the analysis in section 3.2.3. R2 is isomorphic to the complex plane C including the origin, but
without the point at infinity included. The latter would compactify C to S 2 . i.e. to a compact, curved space
on which the Casimir energy is non-zero. The complex plane C including the origin is mapped conformally to
the ’cylinder’ with the slice at τ = −∞ included, but the slice at τ = ∞ excluded. This space, which should
more precisely be called a ’cigar’, has one boundary. It is on this space that (4.78) holds. Indeed this version of
the ’cylinder’ is really the one to which our previous computation of the Casimir energy in section (4.78) applied
because when talking about the vacuum energy, we secretly choose a vacuum to begin with, i.e. include the
vacuum state at τ = −∞.
6 This is not in contradiction with (4.78) because, as explained in the previous footnote, we are really considering
the cigar with only one boundary. While for the proper cylinder R = 0, this is not true for the cigar because we
cannot choose flat coordinates globally.
Note:
By methods similar to those leading to the formal proof of (4.80) one can show that if c 6= c̃
for the left and right-moving Virasoro algebras, the theory suffers from a gravitational anomaly,
i.e. an anomaly of diffeomorphism invariance in σ because then hL0 − L̃0 i 6= 0 to shift. This
is unacceptable and must be excluded, leading to the level matching condition postulated earlier.
4.6 State-operator correspondence, highest weight states,

Verma modules
Let us now investigate the structure of the Hilbert space of a 2-dimensional CFT on S 2 . To this
end we first need to define the properties of the vacuum. We distinguish between the in-vacuum
|0 i corresponding to the vacuum at τ = −∞ and the out-vacuum h 0|, the vacuum at τ = +∞.
We now make use of the crucial property of the conformal mapping from the cylinder to the
complex plane that τ = −∞ is mapped to the point z = 0 and τ = +∞ to the point z = ∞.
• We postulate that the in-vacuum should have the property that the action
Pof the energy-
−n−2
momentum tensor on it should be well-defined. Regularity of T (z) = n z Ln at
τ = −∞, i.e. at z = 0, requires
Ln |0 i = 0 ∀ n ≥ −1. (4.81)
Likewise, regularity of the vacuum at τ = +∞ implies
h 0| Ln = 0 ∀n ≤ 1. (4.82)
Analogous relations holds for L̄. Therefore the only generators that annhilate both |0 i
and h 0| are {L−1 , L0 , L1 } (and similarly for their anti-meromorphic cousins). This is sum-
marised in the important statement:
The vacuum of a 2d CFT is invariant only under P SL(2, C) × P SL(2, C).
• To each primary field Φ(z, z̄) we can associate a state as follows7 : We again postulate that
the action of
X
Φ(z) = φn z −n−h (4.83)
n
on the vacuum must be regular at τ = −∞ and at τ = +∞. This requires

φn |0 i = 0 ∀n ≥ 1 − h, (4.84)
h 0| φn = 0 ∀n ≤ h − 1 (φ†n = φ−n ). (4.85)
We now define the in-state and out-state

|φin i = φ−h |0 i = lim φ(z) |0 i = φ(0) |0 i, (4.86)
z→0
h φout | = h 0| φh . (4.87)
This gives the operator-state correspondence.
7 We restrict to chiral fields to avoid writing everything in terms of the meromorphic and the anti-meromorphic
sectors. Generalisations are obvious.
• From the relation [Lm , φn ] = (m(h − 1) − n)φm+n for primary fields of dimension h we
deduce the following action of the Virasoro generators on the primary state |φ i = φ(0) |0 i:
i) |φ i is eigenstate of L0 with eigenvalue h,
L0 |φ i = h |φ i , Ln |φ i = 0 ∀n > 0. (4.88)
ii) L−n , n > 0 increases the eigenvalue of L0 ,
L0 (L−n |φ i) = (n + h)(L−n |φ i) ∀n ≥ 0. (4.89)
Therefore we can identify
Ln as anihilators,
(4.90)
L−n as raising operators w.r.t. eigenstates of L0 .
States with the properties (4.88) and (4.89) are called highest weight states.
This establishes the important isomorphism in a 2-dimensional CFT between
primary fields ↔ highest weight states.
While we have shown in which sense every primary operator defines a highest weight state,
the other direction follows by constructing the modes φn backwards given a state |φ i:
φ−h is defined via |φ i = φ−h |0 i and the remaining modes are constructed via [Lm , φn ] =
(m(h − 1) − n)φm+n . This automatically gives a primary field.
• The complete Hilbert space is obtained by acting with L−n , n > 0 on all highest weight
states |φj i where j labels all primary fields.
Definition 4.1. The Verma module Vhj is the span of all states of the form

k1 ···km
φj i = L−k1 . . . L−km |φj i , ki > 0 (4.91)
Pm
of conformal weight h = hj + i ki .

One can show that the states φkj 1 ···km i with k1 ≥ k2 ≥ . . . km are linearly independent.

By the operator-state correspondence one can in turn define a conformal field associated
with a general state in the Verma module Vhj . These fields are not primary themselves.

Definition 4.2. The state φkj 1 ···km i is created by a secondary = descendent field φkj 1 ···km (z)

from the P SL(2, Z) invariant vacuum.
Important lesson:
• In a general QFT, states and local fields are not equivalent. While a field is by defini-
tion local, a state carries non-local information about a full field configuration. E.g. in
the Schrödinger representation the states Ψ(φ(x), t) correspond to functionals of the field
configuration.8
8 Background on the Schrödinger representation in general Quantum Field Theory can be found e.g. in Chapter
10 of the book ’Quantum Field Theory of Point Particles and Strings’ by Brian Hatfield.
• In a CFT, on the other hand, we can map the entire spatial slice τ = −∞ to the point z = 0
In the path integral information about the state corresponding to a field configuration at
τ = −∞ is thus encoded in the insertion of an arbitrary local operator O(z) at z = 0. This
is the reason why the operator-state mapping is intimately related to conformal symmetry.
For more information about the Schrödinger representation and a path integral formulation of
the state-operator mapping we refer to [P], Chapter 2.8 and 2.9 and especially to [T], Chapter 4.6.
We have seen that the Hilbert space of a 2-dimensional CFT is specified by the spectrum of
primary fields φi and their conformal dimensions (hi , h̄i ).
If the theory is unitary, the conformal anomaly c and the spectrum is subject to the following
constraints:
• c ≥ 0,
• hj ≥ 0,
• hφ = 0 ↔ φ = 1. The only state with associated conformal dimension h = 0 is the
P SL(2, C) invariant vacuum.
A proof of these assertions will be worked out on Assignment 9.
4.7 Correlation functions in CFT

In a 2-dimensional CFT the correlation functions are extremely constrained by the conformal
symmetry. While a thorough derivation of the structure of the correlation functions in a general
CFT is beyond the scope of this course, for completeness we here collect the main statements,
leaving a more detailed discussion and some of the proofs to Assignment 9.
Let us first restrict ourselves to a CFT on a sphere S 2 . For brevity we only consider holomorphic
fields; generalizations to more general fields are obvious. First, invariance of the vacuum under
globally defined conformal transformations implies the following behaviour of a general n-point
function under P SL(2, C) × P SL(2, ¯ C) transformations,
hφ1 (z1 ) . . . φn (zn )i = hφ01 (z10 ) . . . φ0n (zn0 )i|z0 →z . (4.92)
This completely fixes the spacetime-dependence of one-, two- or three-point functions for
quasi-primary fields and constraints the higher correlators as follows:
i) The one-point function must vanish,
hφ(z)i = 0, (4.93)
unless hφ = 0, i.e. φ = 1.
ii) The two-point function of quasi-primaries is non-zero only if the conformal dimensions agree,
dij
hφi (zi )φj (zj )i = δhi ,hj . (4.94)
(zi − zj )2hi
If dij is non-degenerate, the fields can be normalised such that dij = δij .
iii) The three-point function is completely fixed up to the appearance of a structure constant
Cijk ,
C123
hφ1 (z1 )φ2 (z2 )φ3 (z3 )i = h1 +h2 −h3 h2 +h3 −h1 h1 +h3 −h2
. (4.95)
z12 z23 z13
These structure constants are related to the coefficients Cij k in the OPE of two quasi-
primaries given in eq. (4.40) as
Cijk = Cij l dlk . (4.96)
iv) All higher n-point functions are functions of (n − 3) P SL(2, C) invariant cross-ratios. This
is because we can use P SL(2, C) transformations, which map any three points on the sphere
to any other three points, to eliminate the spacetime dependence of the correlator to (n − 3)
combinations of coordinates, which must in addition be invariant. The cross-ratios are
discussed on Assignment 8.
In fact, with the help of the OPE and exploiting the associative structure of the correlation func-
tions one can reduce any higher n−point correlator to data involving the 3-point correlators and
so-called conformal blocks, which depend only on the conformal anomaly c and the conformal
dimensions (hi h̄i ) of the quasi-primaries involved (see e.g. [BP], Chapter 2.12 and 2.13 for more
information and references). In a 2-dimensional CFT these conformal blocks can be evaluated
very explicitly, even though this may be hard in practice. This explicit evaluation makes use
of the infinite-dimensional Virasoro algebra. Combined with the results from section (4.6) on
the structure of the Hilbert space of a 2-dimensional CFT this is summarised in the remarkable
theorem:
A 2-dimensional CFT is completely specified by its conformal anomaly c, the spectrum of primary
fields φi (z, z̄) of dimensions {hj , h̄j } and their OPE coefficients Cij k .
In particular, the definition of a 2-dimensional CFT need not involve a Lagrangian. While many
examples of 2-dimensional CFTs even without an action have been found, a complete classifica-
tion of all 2-dimensional CFTs is, to date, not known.
Comment on higher-dimensional CFTs

The above statements about the one, two-, and three-point correlators made use only of the global
transformations P SL(2, C). As discussed, these do have an analogue also in higher dimensional
CFTs. Therefore, the statements about the one, two-, and three-point correlators involving quasi-
primaries continue to hold in higher-dimensional CFTs. The higher correlators also depend only
of the cross-ratios and the conformal blocks, which in principle are determined by the spectrum of
quasi-primaries and the conformal anomaly. Due to the lack of the Virasoro structure, though,
the explicit evaluation of these conformal blocks is much harder and in general these are not
known, despite much progress in the recent literature.9 This is what makes two-dimensional
CFTs special.
9 See e.g. Pappadopulo et al., http://arxiv.org/abs/1208.6449, and references therein.
4.8 Normal ordering and Wick’s theorem

For a primary field φ the modes φn , n > −h play the role of annihilation operators (and
similarly for the anti-meromorphic pieces) in the sense that φn |0 i = 0 if n > −h. Correspond-
ingly the modes φn , n ≤ −h are creation operators.
In QFT normal ordering is usually defined as moving all creation operators to the left.
Indeed one can rigorously prove that this notion of normal ordering is equivalent to picking out
the non-singular term in the radially ordered OPE, i.e.
φ(z)χ(w) = { singular piece } + : φ(z)χ(w) : (4.97)
The proof can be found e.g. in [BP], Chapter 2.7. and will not be presented in this course.
Clearly h: φ(z)χ(w) :i = 0. This leads to Wick’s theorem for two fields:
φ(z)χ(w) = hφ(z)χ(w)i + : φ(z)χ(w) : (4.98)

time / radially ordered normal ordered
As in a general QFT one can inductively use this to relate time-ordered and normal-ordered
products of more than two fields by replacing any pair of them by their two-point correlator, e.g.
: φ1 (z1 )φ2 (z2 )φ3 (z3 ) : = φ1 (z1 )φ2 (z2 )φ3 (z3 )
−φ1 (z1 ) hφ2 (z2 )φ3 (z3 )i − φ2 (z2 ) hφ1 (z1 )φ3 (z3 )i (4.99)
−φ3 (z3 ) hφ1 (z1 )φ2 (z2 )i
X
≡ φ1 (z1 )φ2 (z2 )φ3 (z3 ) − subtractions,
where the subtractions are obtained by successively taking all pairs of fields and replacing their
product by its expectation value.
4.9 Applications to String Theory

We now apply these abstract ideas to the bosonic string, shedding new light on the ontology of
the fields we have already got to know in less enlightened language. The theory of the bosonic
string freely propagating in R1,d−1 consists of two CFTs on the worldsheet - the collection of d
free bosons as well as the ghost system. Let us revisit both in turn.
4.9.1 The free boson on the sphere

The action of a single free boson on the sphere is given by
Z
1 ¯
S= dz dz̄ ∂X(z, z̄) ∂X(z, z̄). (4.100)
2πα0
¯
• Let us first check that the classical equation of motion ∂ ∂X(z, z̄) = 0 holds
R at the
δ
quantum
level as an operator equation inside the path integral. Indeed, since DX δX ... is the
integral over a total derivative, it vanishes and we conclude
Z Z
1 δ −S 1 δS 1
¯
DXe−S

0= DX e =− = 0
∂ ∂X(z, z̄) . (4.101)
Z δX Z δX πα
• From the decomposition X(z, z̄) ≡ X(z) + X̄(z̄) we find the conserved chiral and anti-
chiral worldsheet currents j(z) := i∂X(z) and j̄(z̄) := i∂¯X̄(z̄). From a CFT perspec-
tives these are well-behaved fields.
• In fact, X(z, z̄) is not a (quasi-) primary field. This is evident already from the form
of its two-point function. To compute this we first show that
∂z ∂z̄ hX(z, z̄)X(w, w̄)i = −πα0 δ (2) (z − w). (4.102)
This is because by the same trick as above we find

Z
δ
e−S X(z 0 , z̄ 0 )

0 = DX
δX(z, z̄)
Z
1
= DXe−S δ (2) (z − z 0 , z̄ − z̄ 0 ) + ∂ ∂
z z̄ X(z, z̄)X(z 0 0
, z̄ )
πα0
D E 1
⇒0 = δ (2) (z − z 0 , z̄ − z̄ 0 ) + ∂z ∂z̄ hX(z, z̄)X(z 0 , z̄ 0 )i . (4.103)
πα0
Using10 ∂z z̄1 = 2πδ (2) (z, z̄) = ∂z̄ z1 we can integrate (4.102) to
α0
hX(z, z̄)X(w, w̄)i = − log(|z − w|2 ). (4.104)
2
This is familiar result that the Green’s function in two dimensions is logarithmic. Obviously
this two-point correlator is not of the form (4.94) as would be required for a quasi-primary.
In other words, X(z, z̄) does not satisfy the defining property (4.34) for a primary field.
• Correspondingly, the chiral correlators take the form
α0
α0
hX(z)X(w)i = − log(z − w), X̄(z̄)X̄(w̄) = − log(z̄ − w̄). (4.105)
2 2
• In the quantum theory the energy-momentum tensor is defined as the normal ordered
expression
1
T (z) := − 0 : ∂X(z)∂X(z) : =⇒ hT (z)i = 0. (4.106)
α
• The currents ∂X(z), ∂¯X̄(z̄) are primary fields of conformal dimension (1, 0) and (0, 1)
respectively.
10 To see this we note that ∂z z̄1 = 0 away from the origin, while at the same time we can compute d2 z ∂z z̄1
R
via Stoke’s theorem (4.60). Using d z = 2 dx dx the latter reads 2 d x ∂α J = d z (∂z J + ∂z̄ J z̄ ) =
2 1 2
R 2 α
R 2 z
−i (dzJ z̄ − dz̄J z ). Then with J z = z̄1 , J z̄ = 0 and taking into account the opposite orientiation for the anti-
H
holomorphic contour integral we confirm d2 z ∂z z̄1 = i dz̄ z̄1 = i(−2πi) = 2π.

R H
i) As a first check, the propagator is indeed of the correct form

α0
h∂X(z)∂X(w)i = − (z − w)−2 , (4.107)
2
as follows by differentiating hX(z)X(w)i.
ii) To prove the primary condition we compute the OPE T (z)∂X(w) by Wick’s theorem,
1
T (z)∂X(w) = −
: ∂X(z)∂X(z) : ∂X(w) (4.108)
α0
1
= − 0 [: ∂X(z)∂X(z)∂X(w) : +2∂X(z) h∂X(z)∂X(w)i] .
α
Here we used (4.99). Now we insert the (4.107) and use the Taylor expansion
∂X(z)(z − w)−2 = [∂X(w) + (z − w)∂∂X(w) + . . .] (z − w)−2 (4.109)
to establish
∂X(w) ∂∂X(w)
T (z)∂X(w) = 2
+ + { non-sing. }. (4.110)
(z − w) (z − w)
This identifies ∂X as a chiral primary of dimension h = 1. Correspondingly the mode
expansion for the currents takes the form
r
X α0
i∂X(z) = αn z −n−1 , (4.111)
2
n∈Z
r
X α0
i∂¯X̄(z̄) = α̃n z̄ −n−1 . (4.112)
2
n∈Z
Note that [Lm , αn ] = −nαm+n , which had been derived previously, is the commutator
version of the statement that h = 1.
• The same methods of exploiting Wick’s theorem yield the OPE of the energy-momentum
tensor with itself, as spelled out on Assignment 9,
1/2 2T (w) ∂T (w)
T (z)T (w) = + + + ... (4.113)
(z − w)4 (z − w)2 (z − w)
This confirms our earlier finding that c = 1 for one integer-moded real boson.
• Another important primary field in the X-CFT is the exponential
Fk =: eikX(z,z̄) : = : eikX(z) eikX̄(z̄) : . (4.114)
By Wick’s theorem one computes the following OPE by expanding the exponential, as
performed again in Assignment 9,
α0 1
∂X(z) : eikX(w) : = −i k : eikX(w) : + ..., (4.115)
2 (z − w)
0 2
α k /4 ∂w
T (z) : eikX(w) : = 2
+ : eikX(w) : +..., (4.116)
(z − w) (z − w)
0
: eiαX(z) :: eiβX(w) : = (z − w)α (αβ) : ei(αX(z)+βX(w)) : (1 + O((z − w))) .(4.117)
α0 2
The OPE with T (z) proves that : eikX(z,z̄) : is primary with h = 4 k = h̄.
4.9.2 The bc-ghost-system

The action of the bc-ghost-system in complex coordinates takes the form
Z
1
d2 z bzz ∂¯z̄ cz + bz̄z̄ ∂z cz̄ .

Sghost = (4.118)
2π
The ghost field cz (z) is a worldsheet vector, i.e. a spin 1 current. Its conformal dimension is
h = −1 as this corresponds to the mass dimension of a vector.
Correspondingly the anti-ghost bzz (z) is two-tensor current (spin 2) with h = 2.
More generally, one considers the so-called (bc)-CFT with action
Z
1
d2 z b(z)∂¯z̄ c(z) + anti-chiral .

Sbc = (4.119)
2π
It is easy to see that this defines a CFT if the fields are of the following nature: b(z) is primary
of weight hb = λ, c(z) is primary of weight hc = 1 − λ, for some λ ∈ R. By the usual methods
one finds the energy-momentum tensor
T (z) =: (∂b)c : −λ∂(: bc :). (4.120)
The OPE of T (z) with itself shows that the (bc)-central charge is
c = −3(2λ − 1)2 + 1 (4.121)
(and similarly for the anti-chiral piece). See Assignemnt 10 for more details. The string theoretic
ghost-system is therefore the special case λ = 2, c = −26 of the bc-CFT.
4.9.3 String quantisation à la CFT

Having reviewed the two ingredients in the bosonic path integral in CFT language, let us take a
fresh look at how the two pieces come about.
• The starting point of the bosonic string quantization is the action SBDH [X µ , hab ] with X µ
taking values in R1,d−1 . The action enjoys local Weyl and diffeomorphism invariance on
the worldsheet.
• Gauge fixing à la Faddeev-Popov leads to a theory defined on11 C (for closed strings at tree
level). The remnant of the original Weyl and diffeomorphism symmetry is the conformal
i i
symmetry on C. The full gauge fixed theory is described by S tot = S (X ) +S (bc) , where S (X )
(X i )
describes d copies of the free-boson-CFT, each with central charge c = 1, i = 1, . . . , d,
and S (bc) describes the λ = 2-(bc)-CFT with c(bc) = −26.
• Self-consistency of the Faddeev-Popov procedure requires absence of the total conformal
anomaly, X i !
ctot = c(X ) + c(bc) = 0. (4.122)
i
Thus the bosonic string in d = 26 dimensions is fully consistent as a quantum theory.

11 See footnote 5.
• We are now in a position to take a more abstract perspective: The requirement ctot = 0
can also be met by considering more general CFTs and replacing (some of) the 26 copies
of the X-CFT therewith. The non-free CFTs are called ”internal” sectors. E.g. one can
take the tensor product of only 4 copies of the X-CFT (corresponding to propagation of
the string in R1,3 ) and a tensor product of more complicated CFTs with c = 22. Later
on we will discuss the idea of compactification by considering as target space not flat
26 dimensions, but a space-time of the form R1,3 × M with M some internal compact
manifold (here of dimension 22). The ”internal” CFT can then be thought of as describing
the string propagation on this internal manifold M. Celebrated examples of such CFTs
include so-called Gepner models, which are known to describe the string propagation on
certain Calabi-Yau manifolds appearing compactifications of the superstring.
Even though this is not yet apparent from what we have learned, the secret of the worldsheet
approach to string theory is the insight that space-time is only a manifestation of an abstract
CFT on the worldsheet. What is important is not that space-time is well-defined in the sense of a
smooth manifold, but that the CFT that describes the propagation of the string on that space is
well-defined. This implies that strings can consistenly propagate even on certain singular spaces,
e.g. orbifolds of the form M/G with G the action of a finite group, as long as the underlying CFT
is non-singular. If you want to learn more about this fascinating way to think about the nature
of space-time within string theory, you can try already at this stage to read to the introduction
to Brian Greene, String Theory on Calabi-Yau manifolds, http://arXiv.org/pdf/hep-th/9702155,
even though you might wish to wait until we have introduced the superstring later in this course.
• For the X-CFT the requirement of BRST-invariance gives the physical state condition
Lm |φ i = 0, L̃m |φ i = 0, ∀m > 0 (4.123)
(L0 − 1) |φ i = 0, (L̃0 − 1) |φ i = 0. (4.124)
This establishes the central insight:
Physical string states are 1-1 to primary fields of weight h = 1 = h̄.
It is crucial to appreciate that this restriction to primaries and to h = 1 = h̄ does not

follow from the CFT itself, but is information ”prior to CFT”, i.e. it arises as an extra
consistency condition for the X-CFT to make sense as part of the the gauge fixed version
of the original BDH-action. By contrast, in a general CFT the Hilbert space consists of all
Verma modules over all primary fields, not just the primaries of h = h̄ = 1.
• This leads to the concept of a vertex operator.
Definition 4.3. A vertex operator is a primary field of dimension (h, h̄) = (1, 1).
Its insertion at z = 0 creates a physical state from the PSL(2, C) × PSL(2, C)-invariant
vacuum.
We could also have approached the construction of the string spectrum entirely by constructing
the various vertex operators. Let us demonstrate this in two examples:
µ
i) The primary field : eikµ X (z,z̄)
: acting on the vacuum creates the state
µ α0 2
|k i = lim : eikµ X (z,z̄)
: |0 i, h = h̄ = k . (4.125)
z,z̄→0 4
The physical state condition is evaluated as follows:
µ
• The field : eikµ X (z,z̄)
: is always primary as shown above.
4
• In order to satisfy h = 1 we need k 2 = −m2 = α0 . This gives the mass shell condition
for the lowest-lying state.
Indeed one can rigorously show that the so-defined state |k i is the momentum eigenstate
with momentum k,
r I
µ 2 dz ν
α0 |k i = i ∂X µ (z) : eikν X (0,0) : |0 i
α0 2πi
r I
OPE 2 dz i 0 µ1 ν
= i − αk : eikν X (0,0) : |0 i
α0 2πi 2 z
r
α0 µ
= k |k i. (4.126)
2
Here we made use of the OPE (4.115).
ii) The fields at the first excited level are created by the following vertex operator:
|k, ξ i = lim ξµν : ∂X µ (z)∂¯X̄ ν (z̄)eik·X(z,z̄) : |0 i. (4.127)

z,z̄→0 | {z }
=:V1 (k,ξ;z,z̄)
The physical state conditions are as follows:

α0 2 !
• The dimension of V1 is hV1 = 1 + 4 k = h̄V1 To achieve hV1 = 1 we must set k 2 = 0.
• Higher constraints arise from demanding that V1 (k, ξ; z, z̄) be a primary field.
Again with the help of Wick’s theorem we compute the OPE
k µ ξµν
T (z)V1 (k, ξ; w, w̄) = const × : ∂¯X̄ ν (w̄)eik·X(w,w̄) :
(z − w)3
0
α 2 −2 ∂w
+ k + 1 (z − w) + V1 (k, ξ; w, w̄)
4 (z − w)
+...
!
This has the form of the OPE of a primary field if k µ ξµν = 0, thereby reproducing the
known transversality constraint of momentum and polarization.
Summary:
Different physical states ←→ Different ways to construct vertex operators

Chapter 5
String Interactions
5.1 Perturbative Expansion

We have finally gathered the technology required to approach the important question of string
interactions.
• The basic object to compute in string perturbation theory is the S-matrix, defined as
the amplitude for the scattering of asymptotic in- and out-states. E.g. scattering of two
in-states into two out-states is described by worldsheets of the following form for the closed
and open string, respectively:
• Just from drawing the worldsheets we note a key property of string scattering - the ab-
sence of definite local interaction vertices. This fundamentally distinguishes string
scattering from the scattering of point particles in QFT as is evident by comparing the
above worldsheets with a 2-to-2 Feynman graph, e.g. in QFT. The string worldsheet
always looks locally like the worldsheet of a freely propagating string, and only global
properties of the worldsheet capture interactions.
Put differently, the string interactions are encoded already in the free two-dimensional CFT
without adding arbitrary further terms in the worldsheet action.
This crucial difference compared to point particle QFTs in target space cannot be overes-
timated - after all the specification of interactions by a QFT Lagrangian adds a degree of
arbitrariness into the theory that makes it hard to accept this as a fundamental theory.
• In the following we consider the path integral with a Euclidean metric on the worldsheet.
In fact, the only worldsheets for which a globally non-singular Lorentzian signature metric
exists are the torus T 2 , which has no boundaries, or the cylinder. This is because a
globally-defined Lorentzian metric requires a globally non-singular Killing vector field that
93
CHAPTER 5. STRING INTERACTIONS 94
distinguishes the time coordinate from the spatial ones. Such Killing vector fields exist only
for the mentioned surfaces. Consider e.g. the surfaces above arising in 2-to-2 scattering:
Due the merging and splitting of the in- and out-going pants there is a bifurcation of the
”time”-coordinate and thus no globally defined time-like Killing vector field.
• By means of the operator-state correspondence - the central lesson from the previous CFT
chapter - the in- and out-states are encoded in the path integral via insertion of the corre-
sponding vertex operators on the worldsheet.
As the simplest example consider the worldsheet describing 1-to-1 closed string scattering,
with Euclidean coordinates and metric
w = τ − iσ, ds2 = dw dw̄. (5.1)
It has the topology of a cylinder, with the spatial slices at τ = ±∞ corresponding to the
in- and out-states:
2π
The conformal map z = e ` w maps this to a sphere, with the in- and out-states inserted
at the two poles, or via the stereographic projection to C ∪ ∞ with insertions at the origin
and at z = ∞:
∼
=
• By similar conformal maps the above closed 2-to-2 scattering process can be mapped to an
S 2 with 4 marked points corresponding to the insertion of the 4 vertex operators.
Analogously for open strings the scattering worldsheet is mapped to a disk with insertions
on the boundary:
Equivalently we can consider the upper half plane with 4 vertex operators inserted on the
real line.
Summary: By conformal symmetry string scattering is described by compact worldsheets
with insertions of vertex operators for all in- and out-states.
• Our notation will be to describe by Vj (k) the vertex operator associated with the j-th state
µ µ
with momentum kin = (E, ~k) or, equivalently, kout = −(E, ~k).
• The above examples include only the simplest compact worldsheets with 4 insertions, but
in the path integral we must sum over all possible topologies.
In the oriented string all worldsheets are orientable.1 In characterising the types of
worldsheets we need to consider we make use of the following
Theorem 5.1. Every compact, connected, oriented2 two-dimensional manifold is topolog-
ically equivalent to a sphere with g handles and b holes representing the boundaries.
As some examples consider the following closed worldsheets
S 2 : (g, b) = (0,0) torus T 2 : (g, b) = (1,0) double torus (g, b) = (2,0)
or open worldsheets
disk: (g, b) = (0,1) annulus: (g, b) = (0,2) pants: (g, b) = (0,3)
Note that the disk is topologically equivalent to a sphere S 2 with one hole, i.e. with one
boundary,
∼
=
1 By contrast in orientifold theories we include non-oriented worldsheets. Recall that such unoriented string
theories were introduced and discussed in the tutorial.

2 In the unoriented case we need to specify in addition the number c of crosscaps.
and likewise the cylinder can be viewed as a sphere with two holes etc.
• A topological invariant of two-dimensional oriented surfaces3 is the
Euler number χ = 2 − 2g − b. (5.2)
By the famous Riemann-Roch-theorem this topological invariant is computed by the
expression
1
Z
2
√ (2) 1
Z
χ= d ξ hR + ds k (5.3)
4π Σ 2π ∂Σ
with R(2) the Ricci scalar of the surface and k the geodesic curvature of the boundary.
R
1
√ 1

d2 ξ hR + 2π
R
• Recall that we can add to the string action the term λ 4π Σ ∂Σ
ds k with
λ ∈ R without affecting the dynamics. This term then keeps track of the topology of the
worldsheet in the path integral.
Putting everything together we arrive at the following heuristic expression for the S-matrix
describing the scattering of n string states (viewed as an expression before gauge fixing):
R n
X DXDh −SX −λχ Y
Sj1 ...jn (k1 , . . . , kn ) = e Vji (ki ). (5.4)
Voldiff.×Weyl i=1
compact topologies
Note that we divide by the volume of the group of diffeomorphisms and Weyl transformations
as this group will factor out in the process of the Faddeev-Popov procedure. The particulars of
this gauge fixing, however, have to be reconsidered in the presence of vertex operators.
Comments:
1) Let us define the quantity
gs = eλ . (5.5)
Then the expansion in terms of worldsheets of different topology is governed by factors of
−(2−2g−b) −(2−2g−b−c)
gs−χ = gs (for the oriented case and correspondingly gs−χ = gs for the
non-oriented theory) in the path integral. For λ << 0 we observe that gs << 1. Then the
sum over topologies defines a perturbative series.
2) Consider a closed string worldsheet and add a handle to it, e.g.
Since this adds two boundaries the Euler number decreases as χ −→ χ − 2.

Physically adding a handle describes emission and reabsorption of a closed string. If we think
in terms of Feynman diagrams this would corresond to two ”vertices”, each coming with
”coupling constant” gc . Therefore, we interpret gs as the
closed string coupling gc = eλ = gs . (5.6)

3 For non-oriented surfaces we must include the number c of crosscaps, χ = 2 − 2g − b − c.
3) Similarly add a boundary to an open worldsheet, thereby decreasing the Euler number by 1,
The physical interpretation - emission and reabsorption on an open string - indentifies the
open string coupling go = eλ/2 = gc1/2 . (5.7)
4) The object Vji (ki ) is the integrated vertex operator

Z Z
Vji (ki ) = const × d2 ξi h(ξi )Vji (ki ; ξi ) ∼
p
= const × d2 zi Vji (ki ; zi ) in flat gauge. (5.8)
Integration over the insertion of the vertex operator, d2 zi , ensures invariance of the full
R
amplitude under diffeomorphisms on the worldsheets.

The integrated vertex operator is normalised such as to carry one factor of go or gc for open/
closed states.
E.g. the integrated vertex operator for a closed tachyon is
Z
VT (ki ) = gs d2 zi : eiki ·X(zi ,z̄i ) : . (5.9)
For a general closed state we oftentimes employ the notation

Z
Vji (ki ) = gs d2 zi Vji (zi , z̄i ) : eiki ·X(zi ,z̄i ) : . (5.10)
5.2 Moduli space of Riemann surfaces

It is now time to discuss the degrees of freedom in the metric of two-dimensional surfaces in
more detail. As indicated already in the context of the vacuumR partition function there is a
2-fold mismatch between the integral over
R all worldsheet metrics Dh and the integral over all
diffeomorphisms and Weyl rescalings, dξ det P , where
hab −→ hab + (P · )ab + 2Λ̃hab . (5.11)
1) First, as stressed several times by now, for the subset of diffeomorphisms and Weyl rescalings
given by the conformal Killing vectors the change of the metric under the diffeomorphism can
be undone by the Weyl rescaling and thus does not lead to a new metric. These conformal
Killing vectors are in the kernel of the differential operator P , i.e. they satisfy
P · = 0. (5.12)
Recall furthermore that these are in 1-to-1 correspondence with the normalisable zero modes
of the Faddeev-Popov ghosts.
To avoid overcounting in the path integral one must not integrate over such conformal Killing
vectors. In the presence of vertex operator insertions one can impose a further gauge fixing
condition by fixing the position of these vertices on the worldsheet. Depending on the number
of vertex operators this partially or completely removes the residual symmetry.
2) For 2-dimensional surfaces of non-trivial topology not every metric hab can be reached from
a given reference metric ĥab by the transformation ĥ → ĥζ . Consider the operator
P† : tab 7→ −2∇b tab = (P † t)a . (5.13)
This is the adjoint of P with respect to the positive definite measure induced by the inner
products
√
Z
(2)
(δh |δh ) = d2 ξ −hhab hcd δh(1)
(1) (2)
ac δhbd (5.14)
and
√
Z
(1) (2) (2)
hξ ,ξ i= d2 ξ −hhab ξa(1) ξb . (5.15)
Indeed integration by parts straightfowardly confirms that
hξ, P † ti = (P · ξ|t). (5.16)
Now suppose the exists a symmetric traceless 2-tensor t0 in the kernel of P † , i.e. such that
P † t0 = 0. Then for all vectors ξ a
hξ, P † t0 i = (P · ξ|t0 ) = 0. (5.17)
Since P · ξ is orthogonal to t0 ∀ξ, no ξ a can be found such that P · ξ = t0 . Such t0 are called
metric moduli.
Definition 5.1. Deformations of the metric of a differentiable manifold that cannot be ab-
sorbed by a diffeomorphism or Weyl rescaling are called metric moduli.
Comparison with the equations of motion (3.146) of the Faddeev-Popov anti-ghosts reveals
that the metric moduli are in 1-to-1 correspondence with the normalisable zero-modes of the
anti-ghosts.
In the path integral, the sum over the metric moduli must be carried out extra.
In the sequel we will need the notion of a Riemann surface.

Definition 5.2. A Riemann surface is a 2-dimensional complex manifold.
Note:
• By a complex manifold we mean a differentiable manifold M together with an atlas such

that in each patch labeled by m we can pick complex coordinates zm in such a way that
the transition functions between the patches are holomorphic functions, zm = fmn (zn ).
• Just as one considers two differential manifolds to be equivalent if there is a differential
map between them, two complex manifolds are equivalent if there exists a holomorphic map
between them. Now, as discussed in detail in chapter 4.3, on a 2-dimensional manifold the
holomorphic maps correspond to the conformal transformations. Therefore, by a Riemann
surface we really mean the equivalence class of all complex 2-dimensional manifolds modulo
diffeomorphisms (because we consider differentiable manifolds) and Weyl rescalings.
• The degrees of freedom of a Riemann surface are therefore precisely given by the metric
moduli.
Of great use is the following

Theorem 5.2. (Riemann-Roch): Consider an orientable Riemann surface of Euler number
χ. Denote by
µ = dim(ker P † ), (5.18)
κ = dim(ker P ) (5.19)
the number of metric moduli and conformal Killing vectors, respectively4 . Then5
µ − κ = −3χ = 6g + 3b − 6. (5.20)
Furthermore,
if χ > 0 : κ = 3χ, µ = 0, (5.21)

if χ<0: κ = 0, µ = −3χ. (5.22)
The proof can be found e.g. in [BLT], Chapter 6.2. or in [P], Chapter 5.3.
Examples:
1) The Riemann surface of maximal Euler number is the sphere S 2 with χ = 2. From the
Riemann-Roch theorem we read off that µ = 0, i.e. every S 2 is conformally diffeomorphic
to flat space. Furthermore κ = 6 = dimR (P SL(2, C)), in agreement with the fact that
the conformal group of S 2 is P SL(2, C) if we treat the holomorphic and anti-holomorphic
coordinates z and z̄ as complex conjugates (as we do in the geometry of Riemann surfaces).
2) As the next important example consider a torus T 2 with Euler number χ = 0. There are
many representations of the torus. The most intuitive one is as T 2 = S 1 × S 1 .
Parametrise both S 1 by periodic coordinates 0 ≤ σ 1 ≤ 2π, 0 ≤ σ 2 ≤ 2π so the torus

coordinates are doubly-periodic,
(σ 1 , σ 2 ) ∼
= (σ 1 , σ 2 ) + 2π(m, n) m, n ∈ Z. (5.23)
With not too much work one can show that modulo diffeomorphism and Weyl rescalings the
most general metric on T 2 can be brought into the form
ds2 = |dσ 1 + τ dσ 2 |2 τ ∈ C. (5.24)
For a derivation and much more information on T 2 see the tutorial.

4 Here we count the real dimension of the moduli space.
5 For unoriented Riemann surfaces we have µ − κ = −3χ = 6g + 3b + 3c − 6.
The complex number τ is the modulus. Since we count real degrees of freedom we conclude
that µ = 2.
Since χ = 0, Riemann-Roch implies that κ = µ, i.e. we expect two real conformal Killing
vectors.
In fact the globally defined global transformation σ a → σ a + v a leaves the metric and period-
icity of the coordinates invariant. We have thus established that the conformal group of T 2
is given by U (1) × U (1).
In the tutorials we will see that the shape of a torus does not change under a P SL(2, Z)-
transformation
aτ + b
τ→ . (5.25)
cτ + d
This restricts the inequivalent values of τ to the so-called fundamental domain. A convenient
choice of the fundamental domain is |τ | ≥ 1, − 21 ≤ <(τ ) ≤ 21 .
5.3 The gauge fixed S-matrix

Let us discuss in more detail the gauge fixing procedure for the S-matrix, addressing in particular
the questions of conformal Killing transformations and the moduli space of metrics. Our starting
point is the expression for the S-matrix derived above,
n Z
DXDh
X Z Y p
Sj1 ...jn (k1 , · · · , kn ) = e−SX −λχ d2 ξi h(ξi )Vji (ki , ξi ).
VolDiff×Weyl i=1
compact topologies
(5.26)
• The worldsheet metric hab now depends on µ moduli tα , α = 1, . . . , µ. The variation of

h with respect to tα is described by
δα h(tα ) = δtα ∂α h ≡ δtα ∂tα h. (5.27)
The integral over the degrees of freedom of the metric must therefore include also an integral
over the possible values of the moduli, i.e. their fundamental domain F .
Definition 5.3. The quantity
1 bd
µα,a b = h ∂tα had (5.28)
2
is called Beltrami differential.
• To avoid overcounting we impose an extra gauge fixing condition to fix the residual confor-
mal Killing vector transformations. In the presence of a sufficient number of vertex operator
insertions it is convenient to use the κ Killing vectors to fix κ positions of vertex operators.
Remark:
Alternatively one can divide the non-gauge fixed expression by the volume of the conformal
group. In particular if there are not enough vertex operators to fix the conformal Killing
transformations completely one divides by the volume of the remaining subgroup of the
conformal group. If the volume of the conformal group or a subgroup by which we divide is
infinite, then the corresponding string correlator vanishes. E.g. on the sphere the conformal
group is P SL(2, C) and has infinite volume. Thus, the oriented closed string 0-point, 1-
point and 2-point function vanishes at tree-level, corresponding to the absence of a vacuum
energy, tadpole and, respectively, mass renormalisation at tree-level. For the oriented open
string, the conformal group on the disk is P SL(2, R) and the same statements hold for the
open 1- and 2-point function at tree-level. On the other hand, the closed string 1-point
function on the disk need not vanish as insertion of a closed string vertex operator in the
bulk of the disk only leaves a finite volume U(1) subgroup of P SL(2, R) unfixed. Thus
there can (and in fact will) be closed string tadpoles in the presence of D-branes.
Naively, the measure becomes

Z n Z
Y Z µ Z
Y n−µ
YZ
Dh d2 ξi → Dζ dtα d2 ξi (5.29)
i=1 α=1 i=1
with κ positions ξi fixed at ξî . More precisely one has to compute the modified Faddeev-Popov
determinant ∆FP .
Derivation of the measure - non-examinable

As a warm-up we first present, following [P], the following derivation of the Faddeev-Popov determinant
for the partition function without any vertex operator insertions and ignoring the subtleties of conformal
Killing vectors and metric moduli. It is an alternative to the shortcut presented in section (3.3.1). The
partition function is
Z = DX Dh ei SP [X,h] = DX Dh Dζ ei SP [X,h] δ(h − ĥζ ) ∆F P (h)

R R
(5.30)
via insertion of
Z
1 = ∆F P (h) Dζ δ(h − ĥζ ). (5.31)
R
Performing the Dh integration and exploiting gauge invariance as in section (3.3.1) this becomes
Z
Z = DX Dζ ei SP [X,ĥ] ∆F P (ĥ). (5.32)
The inverse of the determinant ∆F P (ĥ) can be expressed as

Z
∆F P (ĥ)−1 = Dζ δ(ĥ − ĥζ ), ζ = (a , Λ), ĥζ = ĥ + P · + 2Λ̃ h. (5.33)
the help of the integral representation of the delta-function and the shorthand notation (β|P · ) =
With p
R 2
d ξ ĥ βab (P · )ab this is
Z Z
∆F P (ĥ)−1 = Da DΛ Dβab e2πı(β|−P ·−2Λ̃ĥ) (5.34)
Z Z
0
0
= Da Dβab e−2πi(β |P ·) , (5.35)
0
where the integration over Λ restricts βab to be symmetric and traceless.
Finally we apply the general rule that if the replace the integration variables in the above expression for
∆F P (ĥ)−1 by (suitably normalised) Grassmann-valued fields,
0
a → ca , βab → bab , (5.36)
we obtain ∆F P (ĥ) as
Z 1 Z q
∆F P (ĥ) = Db(ab) Dcd exp d2 ξ −ĥ b · (P · c) . (5.37)
4π
Now we consider the S-matrix with vertex operator insertions and a moduli dependent metric. According
the discussion above the gauge fixing factor to be inserted into (5.26) is
Z Z Y
1 = ∆F P (ĥ, ξîa ) dµ t Dζ δ(ĥ − ĥ(t)ζ ) δ(ξîa − (ξîζ )a ). (5.38)
(a,i)∈f
The last insertion fixes the position of κ vertex operators to the position (ξîζ )a , which depends on the
particular gauge choice, and
µ
X
ĥ(t)ζ = ĥ + δtα ∂tα ĥ + P · + 2Λ̃ h. (5.39)
α=1
By the same logic as before we find

Z Z Z Y Z Pµ
0 α
∆F P (ĥ, ξîa )−1 = dµ δt Da 0
Dβab dxa,i e2πi(β |P ·+ α=1 δt ∂tα ĥ) (5.40)
(a,i)∈f
xa,i a (ξ̂ia )
P
×e2πi (a,i)∈f . (5.41)
Again we replace the integration variables by suitably normalised Grassmann fields,
0
a → ca , βab → bab , xa,i → ηa,i , δtα → γ α (5.42)
and deduce
Z Z
α
1 ∂tα ĥ)+ (a,i) ηa,i ca (ξia )
P
∆F P (ĥ, ξîa ) = Db Dc Dµ γ Dκ η e− 4π (b|P ·c−γ . (5.43)
Dµ γ Dκ η.
R
All that is left is to perform the Grassmann integral
The final result is

Z µ
Y 1 Y
∆FP = Db Dc e−Sgh (b|∂α ĥ) ca (ξî ), (5.44)
α=1
4π
(a,i)∈f
where
p
d2 ξ ĥ bab (∂α ĥ)ab in terms of the gauge fixed metric and ∂α ĥ := ∂tα ĥ and
R
• (b|∂α ĥ) =
• (a, i) ∈ f refers to the set of fixed vertex operator positions.
In total the gauge fixed S-matrix takes the form
µ
XZ Y Z
Sj1 ...jn (k1 , · · · , kn ) = dt α
DX Db Dc e−SX −Sgh −λχ × (5.45)
top. α=1
Y Z µ n q
Y 1 Y
b ˆ
Y
dξia (b|∂α ĥ) c (ξj ) ĥ(ξi )Vji (ki , ξi ).
α=1
4π i=1
(a,i)∈f
/ (b,j)∈f
This looks more scary than it is - after all, what we have done can simply be summarised as the
following take-home message:
• For each conformal Killing vector we place one vertex at ξî : Vji (ki , ξî ) together with a
ghost field ca (ξî ) inserted, without integrating over the position ξî . Alternatively we can
keep the integral over the positions, dξîa Vji (ki , ξî )ca (ξî ), but divide by the volume of the
R
CKV group.
1
(b|∂α ĥ) and integrate over the fundamental domain F dtα .
R
• For each modulus we insert 4π
1
Note that the insertion 4π (b|∂α ĥ) can be written in terms of the Beltrami differentials
introduced above as
Z
1 1
(b|∂α ĥ) = d2 zbzz µα,z z̄ + bz̄z̄ µα,z̄ z . (5.46)
4π 4π
More on the use of Beltrami differentials and the philosophy behind the gauge fixing can
found e.g. in [P], Chapter 5.4.
5.4 Tree-level amplitudes

We are finally in a position to compute the first non-trivial S-matrix, beginning with processes at
tree-level. This corresponds to amplitudes on Riemann surfaces with positive Euler characteristic.
There are three such surfaces:
• Correlators on the sphere S 2 = CP1 (g = b = c = 0) yield oriented closed string
amplitudes.
• Correlators on the disk or upper half-plane D2 (b = 1, g = c = 0) with open string
vertex operators inserted on the boundary of D2 or, equivalently, the real line, correspond
to oriented open string amplitudes. Insertion of closed string vertex operators in the bulk
of the D2 describe the scattering of oriented closed strings off D-branes.
• Correlators on the real projective plane RP 1 (c = 1, b = c = 0) describe processes in
unoriented string theory.
5.4.1 Correlators on the sphere - the Virasoro-Shapiro amplitude

We start with a process on the sphere, corresponding to oriented closed string scattering at
tree-level in perturbation theory. In view of the technology acquainted in the previous section
we note:
1
• The sphere S 2 has no moduli, µ = 0. Thus no insertion of factors 4π (b|∂α ĥ) is required.
• The conformal group P SL(2, C) has 3 complex parameters. Since its volume is infinite, the
first non-vanishing correlator on the sphere is the 3-point correlator - see the remark after
(5.28). For an n-point correlator with n ≥ 3, we fix the P SL(2, C) invariance by fixing
the position of three vertex operators. Thus we must include three Faddev-Popov ghost
c-modes as specified above.
Example: Scattering of four closed string tachyons

The tachyon vertex operator is
Z
Vji (ki ) = gc d2 zi : ei ki X(zi ,z̄i ) : . (5.47)
Thus the S-matrix describing the scattering four closed string tachyons at tree-level is
Z 3
Y
S(k1 , k2 , k3 , k4 ) = gc4 e−2λ d2 z4 h : c̃ c ei ki X : (zi , z̄i ) : ei k4 X : (z4 , z̄4 )iS 2 , (5.48)
C2 i=1
where as always the zi inside hi are radially ordered. Since the X-CFT and the (bc)-CFT are
independent the problem factorises in the computation of the following two correlators,
i) h: ei k1 X(zi ,z̄i ) : . . . : ei k4 X(z4 ,z̄4 ) :iS 2 and
ii) h: c̃ c(zi , z̄i ) : . . . : c̃ c(z3 , z̄3 ) :iS 2 .
ad i) The first correlator reduces to a Gaussian path integral with the result
m m
Y X Y 0
h ei ki X(zi .z̄i ) i = const. × δ( ki ) |zj − zl |α kj ·kl . (5.49)
i=1 i=1 j<l
While a more through derivation is reserved to the tutorials, we can understand the
structure by bringing the correlator into the form
Ym Z 1 Z
h ei ki X(zi ,z̄i ) i = DX exp d 2
z X(z, z̄) ∂ ¯ X(z, z̄) + iJ(z, z̄) X(z, z̄) ,
∂
i=1
2πα0
X
J(z, z̄) = ki δ (2) (z − zi , z̄ − z̄i ). (5.50)
i
The Gaussian can be performed explicitly. Up to an irrelevant overall factor given by

a functional determinant and some more subtleties, which are treated with care in the
tutorials, it yields
m πα0 Z
h ei ki X(zi .z̄i ) i ∼
Y
= exp d2 z d2 z 0 J(z, z̄)G(z, z̄, z 0 z̄ 0 )J(z 0 , z̄ 0 ) (5.51)
i=1
2
with the Green’s function

1
∂ ∂¯ G(z, z̄, z 0 z̄ 0 ) = δ (2) (z − z 0 , z̄ − z̄ 0 ) =⇒ G(z, z̄, z 0 z̄ 0 ) = log|z − z 0 |2 . (5.52)
2π
The form of G(z, z̄, z 0 z̄ 0 ) follows from the discussion around (4.102) and (4.104).
Among the subtletiesPwe ignored here are the zero modes of the Green’s function, which
m
give us the factor δ( i=1 ki ), and the normal ordering of the vertex operators. Taking
both into account leads us to (5.50).
ad ii) The ghost-sector 3-point function is
3
Y
h : c̃ c(zi , z̄i ) :iS 2 = const. × z12 z13 z23 z̄12 z̄13 z̄23 . (5.53)
i=1
This follows from general expression for 3-point correlator in CFT together with h = −1
for c,
C123
hc(z1 ) c(z2 ) c(z3 )i = −1−1+1 −1−1+1 −1−1+1 . (5.54)
z12 z13 z23
In combination these two results give the following zi dependence of the integrand appearing in
S(k1 , k2 , k3 , k4 ),
0 0 0 0 0 0
|z12 |2+α k1 ·k2 |z13 |2+α k1 ·k3 |z23 |2+α k2 ·k3 |z34 |α k3 ·k4 |z14 |α k1 ·k4 |z24 |α k2 ·k4 . (5.55)
By P SL(2, C)-invariance of the final amplitude we can fix z1 , z2 , z3 to convenient positions, e.g.
z1 = 0, z2 = 1, z3 = ∞.
In particular in the limit z3 → ∞ all terms involving z3 cancel with the help of the kinematic
relations k1 + k2 + k3 + k4 = 0 and ki2 = α40 .
What remains is therefore the non-trivial integral
Z
0 0
I = d2 z4 |z4 |α k1 ·k4 |1 − z4 |α k2 ·k4 . (5.56)
Let us introduce the famous Mandelstam variables
s = −(k1 + k2 )2 , t = −(k1 + k3 )2 , u = −(k1 + k4 )2 . (5.57)
These satisfy
4 4
X X for 4 tachyons 16
s+t+u=− ki2 = Mi2 ≡ − . (5.58)
i=1 i=1
α0
Therefore the integral can be expressed as
Z
0u 0t
I = d2 z|z|−α 2 −4 |1 − z|−α 2 −4 ≡ J(s, t, u) (5.59)
and we find
X
S(k1 , k2 , k3 , k4 ) = i gc4 CS2 (2π)26 δ (26) ( ki )J(s, t, u). (5.60)
i
The constant prefactor CS2 comprises all constants which we have swept under the rug in the
computation and will be fixed later by unitarity.
The function J(s, t, u) appearing in the integrand can be represented via Euler Γ-functions
Z ∞
Γ(z) = dt tz−1 e−t , z ∈ C. (5.61)
0
From standard texts on complex analysis we quote the following properties of the Γ-function:
• Γ(z) is convergent for <(z) > 0 and has a unique analytic continuation to C.
• From Γ(.z +1) = z Γ(z), Γ(1) = 1, which is easily derived via integration by parts, it follows
that the Γ-function coinicides with the factorial for natural numbers, Γ(n) = (n − 1)! for
n ∈ N.
• Γ(z) has poles at z = −n, n = 0, 1, 2, . . ., in whose vicinity it enjoys the expansion

1 (−1)n
Γ|z→−n = . (5.62)
z + n n!
Without proof6 we now quote the following integral

Z
Γ(a) Γ(b) Γ(c)
d2 z|z|2a−2 |1 − z|2b−2 = 2π , a + b + c = 1. (5.63)
Γ(a + b) Γ(a + c) Γ(b + c)
To apply this to (5.59) we identify
u t
a = −1 − α0 , b = −1 − α0 (5.64)
4 4
and arrive at the Virasoro-Shapiro amplitude
0 0 0
X Γ(−1 − α4 s) Γ(−1 − α4 t) Γ(−1 − α4 u)
S(k1 , . . . , k4 ) = i gc4 CS2 (2π)26 δ( ki ) 2π 0 0 0 . (5.65)
i
Γ(2 + α4 s) Γ(2 + α4 t) Γ(2 + α4 u)
Properties of the amplitude:

• Consider a process in the “s-channel”, i.e. scattering at fixed Mandelstam variable t with
s varying,
α0
Of physical importance are the poles of Γ(−1 − 4 s), located at
α0 4
−1− s = −n, i.e. s= (n − 1) ∀n ∈ N0 . (5.66)
4 α0
As is well-familiar from scattering theory in QFT, these are the resonances due to exchange
of string states of mass m2 = α40 (n − 1), with the poles resulting from the propagator of
the intermediate states,
1
∼ . (5.67)
s − m2
That the full tower of string excitations appears in the S-matrix, c.f. (5.66), just means
that we sum over infinitely many exchanged particles,
· δ(s − Ms2 ) + terms analytic at Ms2 . (5.68)
1
R∞ 2
6 This can be derived via the trick |z|2a−2 = Γ(1−a) 0 dt t−a e−|z| t. Together with the analogous expression
for|1−z|2b−2 the integral can be written as a Gaussian integral in the coordinates x, y for z = x+iy. Evalutation of
Γ(x)Γ(y)
the Gaussian then leads to the quoted result if we recognize on the way the Euler Beta-function B(a, b) = Γ(x+y) ,
R1 x−1 y−1
which in turn has the representation B(x, y) = 0 dtt (1 − t) .
In other words, the amplitude knows about the entire spectrum of string excitations, a fact
which is quite remarkable and a highly non-trivial consistency check of the framework.
• By inspection the Virasoro-Shapiro amplitude is symmetric in t and s (and u). So it also
allows an expansion for fixed s in t, corresponding to the t-channel:
· δ(t − Ms2 ) + terms analytic at Mt2 . (5.69)
• The Virasoro-Shapiro amplitude can therefore be written as a sum over infinitely many
s-channel poles or over infinitely many t-channel poles. This property is called duality. It
holds more generally for string amplitudes and distinguishes string amplitudes from point
particle QFT amplitudes in the following sense. In QFT one has to sum over finitely many
s- and t-channels,
(5.70)
In string theory, one single string diagram at genus g corresponds to what would be described
by several Feynman graphs at a given perturbative order of different topologies.
Determining the normalisation

What remains is to determine the constant CS 2 in the amplitude (5.65). This factor contains
the product of the constant CSX2 due to functional determinants in the X-sector and CSbc2 from
the ghost 3-point function,
CS 2 ' gc−2 CSX2 CSbc2 . (5.71)
A closer look in particular at the functional determinants in the X-sector, see the tutorial, reveals
that CS 2 is a universal constant valid for all (3 + n)-point functions on the sphere. In particular
it is the same constant that appears also in the 3-point function for scattering of 3 tachyons.
This amplitude is easily computed with our methods,
X
SS2 (k1 , k2 , k3 ) = i gc3 CS2 (2π)26 δ (26) ( ki ). (5.72)
i
Unitarity now allows us to fix CS 2 by demanding that each pole in the s-channel of the 4-point
function give the same amplitude as two 3-point amplitudes. Consider the first pole at s = − α40 .
Interpretating the 4-point function at s = − α40 as a sequence of two 3-point functions is equivalent
to the ansatz
d26 k SS2 (k1 , k2 , k) SS2 (−k, k3 , k4 )
Z
4
S(k1 , k2 , k3 , k4 ) = i 26 2 4 + terms analytic at 0 (5.73)
(2π) −k + α0 + i α
Near the pole s = − α40 the 4-point function takes the form
8π i CS2 gc4
S(k1 , k2 , k3 , k4 ) = − (2π)26 δ (26) (k1 + k2 + k3 + k4 ). (5.74)
α0 s + 4
This is most easily seen directly from the representation (5.56): The pole in s comes from the
behaviour of z → ∞. There the integral can be evaluated with the help of the residue theorem.
Plugging (5.74) into our factorisation ansatz (5.73) we deduce
8π 1
CS2 = . (5.75)
α0 gc2
Ultraviolet behaviour
Of special interest is the ultraviolet behaviour of the string scattering amplitude as this is the
regime that probes the stringy nature of the excitations. Kinematically this is the limit
s
s → ∞, t → ∞, fixed, (5.76)
t
corresponding to scattering at high energies and at fixed angle θ.
E.g. for massless particles scattering at fixed
angels would be described by the following
choice of momenta:
√
s
k1 = (1, 1, 0, . . . , 0),
√2
s
k2 = (1, −1, 0, . . . , 0),
√2
s
k3 = (1, cos(θ), sin(θ), 0, . . . , 0),
√2
s
k4 = 2 (1, − cos(θ), − sin(θ), 0, . . . , 0).
This limit is called “hard-scattering” limit.
To describe the behaviour of the amplitude in this regime we make us of the asymptotic behaviour
of the Γ-function,
Γ u exp(x ln x) for <(x) → ∞, (5.77)
and deduce
α0
→ S(k1 , k2 , k3 , k4 ) u exp(− (s ln s + t ln t + u ln u)) as |s|, |t| → ∞ and s/t fixed. (5.78)
2
This establishes an exponential fall-off of the amplitude.
Lesson:
String amplitudes fall off much faster than point-particle amplitudes in QFT.
For comparison, the amplitude for exchange of a spin-J particle in QFT falls off like a power-law,
tJ
A∼ . (5.79)
s − M2
• The exponential fall-off of string amplitude is consistent with the s-channel picture (i.e.
the power-law behaviour near the zeroes of the propagator) because we sum over infinitely
many states.
• It is (partially) responsible for UV finiteness when considering strings running in the loop.
√
• Strings behave differently because high-energy processes probe the string length α0 . In
this regime the string is non-local due to its extended structure.
One can now perform a systematic analysis of the closed string scattering amplitudes at tree-level,
in particular of the massless graviton gµν , the B-field Bµν and the dilaton φ. The interactions
can be compared with the vertices from a low-energy effective action in the 26-dimensional
ambient spacetime. This effective action describes the low-energy point particle regime of string
theory. In order to describe the interactions well below the string scale Ms we include only the
massless excitations. In the bosonic theory we could also include the tachyon, but in the full
superstring theory this excitation is absent anyway so including the tachyon is only for reasons
of demonstration. In this fashion one can explicitly confirm at the level of interactions the claim
that the spin-two excitation gµν describes the graviton of Einstein gravity. We will soon find an
alternative method to arrive at the same conclusion.
5.4.2 Correlators on the disk

The computation of open string correlators proceeds in a manner very similar to the closed
string amplitudes, with the difference that due to the boundary conditions the vertex operators
are effectively chiral. More precisely:
• Oriented open string amplitudes at tree-level are described by correlators on the disk D2
with (g, b) = (0, 1), which is conformally equivalent to the upper half plane. The conformal
group is P SL(2, R).
• The open string vertex operators are inserted on the boundary of D2 or, respectively, on
the real line, which forms the boundary of the upper half plane. They depend on the real
variable y parametrizing, say, the real line.
• An important difference to the closed string sector is that the boundary of D2 admits
the notion of ordering of the vertex operator insertions. We must therefore sum over all
possible orderings of the vertex operators (up to one overall permutation symmetry).
• For open strings on a stack of N coincident D-branes the full vertex operator carries the
Chan-Paton labels. Recall that an open string at oscillation level n with both ends on a
stack of N coincident D-branes is characterised by
N
X
|k; n; a i = |k; n; ij iλaij , λaij = (λaij )† , a = 1, . . . , N 2 . (5.80)
i,j=1
The N 2 hermitian N ×N matrices are in the adjoint representation of U (N ). In computing

the scattering amplitude of r open strings we must include, for each summand with a given
ordering of the vertices, the trace over the Chan-Paton matrices,
Tr(λa1 λa2 . . . λar ). (5.81)
Following this procedure the 4-tachyon amplitude amplitude can be computed, mutatis mutan-
dis, in a manner similar to its closed string counterpart. The resulting Veneziano amplitude
takes the form
2igo2 X Γ(−α0 s − 1)Γ(−α0 t − 1)

S(k1 , k2 , k3 .k4 ) = 0
(2π)26 δ (26) ( ki ) 0 0
+ (t → u) + (s → u) .
α i
Γ(−α s − α t − 2)
(5.82)
This will be derived as an exercise.
The open string effective action
One can now systematically compute the scattering amplitudes involving the level-one states,
i.e. the gauge bosons along the D-brane, and the tachyonic ground state as well the massive
string excitations. Taking into account the ordering prescription for the vertex operators and
the Chan-Paton structure one finds explicitly that the interactions agree with the structure
of a U (N ) gauge theory with extra matter in suitable representations of the gauge group. In
particular, the tachyonic and positive-mass excitations from open strings with both ends on the
stack of D-branes give rise to states in the adjoint representation of U (N ).
E.g. the scattering amplitude involving one gauge boson with momentum k1 and polarisation
vector ζ1 and two open tachyons with momenta k2 and k3 takes the form
S(k1 , ζ1 ; k2 , k3 ) ' (2π)(26) δ (26) (k1 + k2 + k3 ) Tr(λ1 [λ2 , λ3 ]). (5.83)
This corresponds to the cubic interaction vertex derived from an effective action in the 26-
dimensional ambient spacetime of the form
Z
1
− Tr(Dµ ϕ Dµ ϕ) (5.84)
R1,25 2
where
ϕ ≡ ϕa λa (5.85)
describes the tachyon in the adjoint of U (N ) with covariant derivative
Dµ ϕ = ∂µ − i[Aµ , ϕ], Aµ = (Aµ )a λa . (5.86)
Similarly one can reconstruct the other interaction terms. The spacetime effective action for the
gauge boson and the tachyonic ground state at string tree-level and to first order in the momenta
takes the form
r
2α0
Z
1 µ 1 2 1 2 1
− Tr(Dµ ϕ D ϕ) + 0 Tr(ϕ ) + Tr(ϕ3 ) − Tr(Fµν F µν ). (5.87)
go2 R1,25 2 2α 3 α0 4
5.5 1-loop amplitudes

5.5.1 Oriented closed theory
Consider the closed oriented n-point function at one-loop in perturbation theory. It is described
by an amplitude on Σ = T 2 . Recall that the torus is topologically T 2 = S 1 × S 1 and can be
parametrised by the doubly-periodic flat real coordinates
(σ 1 , σ 2 ) ∼
= (σ 1 , σ 2 ) + 2π(m, n) m, n ∈ Z. (5.88)
We now need to revisit the moduli space and conformal group of T 2 in order to arrive at the
correct measure for the gauge fixed S-matrix.
1) Moduli space
We have already pointed out that the metric ds2 = |dσ 1 + τ dσ 2 |2 can be expressed in terms
of one complex modulus τ = τ1 + iτ2 . A useful representation of T 2 is a as a lattice in the
complex plane.
The flat complex coordinates on the complex lat-

tice are w = σ 1 + τ σ 2 , w̄ = σ 1 + τ̄ σ 2 .
From its representation as a lattice one can deduce that a T 2 is shape-invariant under a
PSL(2, Z) transformation
aτ + b
τ→ , ad − bc = 1, (a, b, c, d) ∼
= (−a, −b, −c, −d) ∈ Z. (5.89)
cτ + d
The set of PSL(2, Z) transformations define the modular group of the torus. In fact these are
generated by the transformations
1
T : τ → τ + 1, S:τ →− . (5.90)
τ
The fundamental domain F0 , i.e. the domain of inequivalent values of τ after modding out
by the action of the modular group, is:
F0 = {τ | |τ | ≥ 1, − 12 ≤ <(τ ) ≤ 21 , =(τ ) > 0} .
Furthermore the torus enjoys the additional discrete symmetry w → −w. For further details
of these assertions see Assignment 10.
Consequences for path integral: Following the general discussion of the gauge-fixed S-
matrix we insert a factor of
1 1
B B̄ = (b|∂τ ĝ) · b̄|∂τ̄ ĝ (5.91)
4π 4π
and integrate over the amplitude over 12 F0 d2 τ . The factor 21 is due to the extra Z2 symmetry
R
w → −w.
2) Conformal Killing vectors:

The conformal group of T 2 is given by U (1) × U (1). This allows us to fix one closed vertex
operator at, say, w1 . In addition we must insert a factor of c, c̄ at w1 . Altogether the amplitude
therefore contains a factor of
: cc̄V1 (w1 , w¯1 ) : (5.92)
Alternatively we can perform the integral over the position w1 of the vertex operator and
divide by the volume of U (1) × U (1), which is finite,
Vol(U (1) × U (1)) = Vol(T 2 ) = (2π)2 τ2 .
If we follow this prescription the amplitude contains the integral

Z
dw1 dw̄1
2τ
: cc̄V1 (w1 , w¯1 ) : . (5.93)
F0 2(2π) 2
1
The factor of 2 is due to the conversion from complex to real coordinates.
Altogether the n-point amplitude takes the form
Z * Z n Z +
(n) dτ dτ̄ 1 2
Y 1
ST 2 = B B̄ d w1 : cc̄V1 (w1 , w¯1 ) : dwi dw̄i Vi (wi , w̄i ) . (5.94)
F0 4τ2 4π 2 i=2
2
T2
Computation of the vacuum amplitude
To keep things as simple as possible we now compute the 1-loop vacuum amplitude
Z
dτ dτ̄
ZT 2 := h1iτ,⊗X i · h ghost insertions iτ,(bc) . (5.95)
F0 4τ2
As we will see, this amplitude has a very important physical interpretation. It is furthermore
sufficient to demonstrate the key properties of string-loop amplitudes.
The vacuum amplitude splits into a correlator in the X- and the bc-CFT and an integral over
the moduli space. Consider first the X-CFT piece,
Z(τ ) ≡ h1iτ,⊗d i , (5.96)

i=1 X
where we work in d-dimensional target space.

i) First suppose that τ = iτ2 , corresponding to a rectangular torus, which can be viewed as a
cylinder of length (2πτ2 ) with both ends identified.
∼
=
The path integral DXe−SX on a worldsheet given by the above cylinder with both ends
R
identified admits the following interpretation:

Start from the vacuum, create a state, evolve it by Euclidean time 2πτ2 and identify in- and
out-state. In canonical formalism, this amounts to evaluating
X
Z(τ = iτ2 ) = h 0| |0 ih n| e|−2πτ2H
{z } |n ih 0| |0 i
| {z } | {z }
all internal n identify w. evolution create |n i
|0 i from |0 i
X
= h n| e−2πτ2 H |n i ≡ Tre−2πτ2 H . (5.97)
n
This is an example of a general principle in quantum field theory:

The path integral in compactified Euclidean time yields the partition function at temperature
1
T = 2πτ2
.
ii) To generalise this to a torus with modulus τ = τ1 + iτ2 we must in addition translate the
fields by 2πτ1 in spatial direction,
X
Z(τ ) = h1iτ,⊗X i = h n| e2πiτ1 P −2πτ2 H |n i
n
or more compactly
Z(τ ) = Tre2πiτ1 P −2πτ2 H . (5.98)
With the relations

2π 2π c + c̄ 2π
P = (L0 − L̄0 ), H= (L0 + L̄0 − ) with ≡1 from now on,
` ` 24 `
this can written as
d
− 24
Z(τ ) = (q q̄) Trq L0 q̄ L̄0 , q = e2πiτ . (5.99)
We now express L0 in terms of the momentum operator and the number operator,
α0 2 α0 2
L0 = k + N, L̄0 = k + N̄ , (5.100)
4 4
and express the trace as an integral over the momentum modes times the trace over the
oscillator part of a string state,
dd k
Z
α0 2
Tr q L0 q̄ L̄0 = Tr0 h k; N, N̄ (q q̄) 4 k q N q̄ N̄ k; N, N̄ i.

d
(5.101)
(2π) |{z}
trace over
oscillators
The inner product over the momentum state gives a factor of the spacetime volme hk|ki =
α0 2 α0 2
δ (d) (k − k) ≡ Vd , and with (q q̄) 4 k = (exp(4πτ2 )) 4 k we find
dd k −πτ2 α0 k2
Z
d
Z(τ ) = Vd (q q̄)− 24 e · Tr0 q N q̄ N̄ . (5.102)
(2π)d
The computation of the trace was performed in detail on Assignment 6, to which we refer
for details. The result is
∞
Y
Tr0 q N q̄ N̄ = (1 − q n )−d (1 − q̄ n )−d . (5.103)
n=1
We finally perform a Wick rotation k → ik 0 to render the integral dd k finite and perform
0
R
the Gaussian integration. This yields the final expression for the partition function Z(τ ),
d
Z(τ ) = iVd (ZX (τ )) , (5.104)
2 0 − 21 −2
ZX (τ ) = (4π α τ2 ) · |η(τ )| with (5.105)
∞
1
Y
η(τ ) = q 24 (1 − q n ) the Dedekind η function. (5.106)
n=1
The next task is to compute the one-loop integral of the ghost insertions. For reasons of time
we do not present this computation here. It can be found e.g. in [P], Chapter 7, p. 212. We
merely quote the final result: The ghost sector yields a factor of |η(τ )2 |2 . Setting now d = 26
the one-loop amplitude is
Z
dτ dτ̄
Z(τ ) = iV26 (4π 2 α0 τ2 )−13 · |η(τ )|−48 . (5.107)
F0 4τ2
Note:
• The ghost contribution cancels the contribution from the 2 non-transverse polarisations
|η(τ )|−48 = | η(τ )−26 · η(τ )2 |2 . (5.108)
| {z } | {z }
X bc
This demonstrates nicely how the ghost sector restores unitarity by cancelling the non-
physical excitations. The underlying reason for this cancellation is the anti-commuting
nature of the ghost fields. Commuting and and anti-commuting fields contribute with
opposite sign in the loop, as is familiar from QFT with bosons and fermions.
• An important property of ZT 2 is modular invariance, i.e. invariance under P SL(2, Z)
transformations τ → aτ +b
cτ +d of the torus. The underlying reason is that the Dedekind-
function is an example of a modular form, i.e. a function of τ with definite transformation
under P SL(2, Z). Without proof we state that under S and T -transformations, which
generate the modular group, the Dedekind function transforms as
1 1 π
η(− ) = (−iτ ) 2 η(τ ) , η(τ + 1) = ei 12 η(τ ). (5.109)
τ
Modular invariance of ZT 2 follows because dττ dτ̄
2 and τ2 |η(τ )|4 are individually invariant, as
2
can be checked with the above and the P SL(2, Z) transformation
τ2 d2 τ
τ2 → , d2 τ → . (5.110)
|cτ + d|2 |cτ + d|4
This is an important consistency check. Note that modular invariance holds already for
ZX (τ ) defined above.
• For a general CFT the torus amplitude can be written as

Z
dτ dτ̄ d
X
ZT 2 = iVd (4π 2 α0 τ2 )− 2 q Ni −1 q̄ N̄i −1 . (5.111)
F0 4τ2
i∈H⊥
| {z }
yields more general modular
forms (Θ−functions)
Modular invariance is an important consistency condition for the defining data of a CFT.
Ultraviolet finiteness
To appreciate the properties of the stringy 1-loop amplitude we compare ZT 2 to its point particle
analogue. The partition function of a field theory describing a particle of mass m is given by the
sum over all particle paths with the topology of a circle,
Z ∞
dd k
Z
dl − 1 l·(k2 +m2 )
ZS 1 (m2 ) = Vd d
e 2 . (5.112)
(2π) 0 2l
This is easily understood from our derivation of the stringy partition function once we take into
account that
1 2
(k + m2 ) corresponds to the Hamiltonian,
2
l describes the circumference of S 1 ≡ the compact Euclidean time,
Z ∞
dl
takes into account division by the volume of Diff(S 1 ) and of the
0 2l
Z2 symmetry x → −x.
In fact we can bring this into a form very similar to the stringy expression ZT 2 . String theory
contains an infinite tower of states of mass
2
m2i = (Ni + N̄i − 2), Ni = N̄i . (5.113)
α0
2
The field theory analogue of ZT 2 therefore corresponds to summing R π dΘZSi(N
1 (m ) over the string
i
−N̄ )Θ
tower, where we include the level-mathcing condition via δN N̄ = −π 2π e . In this spirit
we find
Z ∞ Z π
X dl dΘ d
X l
ZS 1 (m2i ) = iVd (2πl)− 2 · e− α0 (Ni +N̄i −2)+i(Ni −N̄i )Θ ,
0 2l −π 2π
i∈H⊥ i
Z
dτ dτ̄ d
X
= iVd (4π 2 α0 τ2 )− 2 · q Ni −1 q̄ N̄i −1 (5.114)
R 4τ2 i
l 1
with 2πτ = Θ+i , q = e2πiτ , and integration region R : τ2 ≥ 0, |τ1 | ≤ .
α0 2
We notice a crucial difference of this field theory approach to the string theoretic result:
In ZT 2 we integrate over the fundamental domain F0 : |τ | ≥ 1, |τ1 | ≤ 21 , which is a subregion of
the particle integration domain R.
The importance of this is the following: The limit τ2 → 0 describes the ultraviolet (UV) regime as
in this limit the Euclidean time becomes very small, corresponding to processes at high energies.
Indeed the integral over τ is divergent in this regime. The point is now that this UV divergent
region is absent in string theory due to modular invariance of the torus, but present in field
theory. Thus we have established the crucial result:
ZT 2 is UV-finite in string theory. Modular invariance acts as an intrinsic UV cutoff

and removes the UV divergence of analogous point particle theories.
Comments:
• Modular invariance is therefore the secret behind UV finiteness of the 1-loop amplitude.
While demonstrated only for the simplest example of a vacuum amplitude, the mechanism
of cutting out the UV divergent regent continues to work for n-point functions.
• To date, UV finiteness has been proven rigorously at 2-loop level in superstring theory.7
Beyond that, technicalities concerning the (super)moduli space of higher genus Riemann
surfaces hinder a general proof, but there are no indications that the situation changes at
higher loop order. In fact, using the pure spinor formalism examples of UV finite 5-loop
amplitudes have been computed. This serves as evidence for the widely accepted conjecture
of UV finiteness of string theory.
• It is important to appreciate that superstring theory is to date the only quantum theory
of gravity and Yang-Mills-theory that is believed to be UV-finite.8
Infrared behaviour
The IR regime corresponds to the limit τ2 → ∞, in which the torus (or cylinder with both ends
identified) becomes very long. In this regime we can expand the inegral over the η-function as
Z ∞ Z ∞ ∞
Y
d2 τ |η(τ )|−48 = d2 τ (q q̄)−1 | (1 − q n )|−48
n=1
1
Z 2
Z ∞
∼
= dτ1 dτ2 (q q̄)−1 (1 + 24q + . . .)(1 + 24q̄ + . . .)
−1
Z ∞2
∼
= dτ2 (q q̄)−1 + 242 + . . . ,
7 For the state of the art in this direction consult e.g. the review by D’Hoker and Phong,
http://arXiv.org/pdf/hep-th/0211111.
8 There are certainly nontrivial UV finite Yang-Mills theories such as N = 4 Super-Yang-Mills, and there are
speculations about the possible UV-finiteness of N = 8 supergravity, but none of these describes both particle
interactions and gravity at the same time.
where terms of the type q + q̄ vanish upon performing the integral over the τ1 -coordinate. Thus
we deduce the following IR-behaviour
 
Z ∞
dτ2
iV26 (4π 2 α0 τ2 )−13  e|4πτ 242 + |{z}
{z } + |{z}
2
...  (5.115)
 
2τ2
tachyon massless massive
modes modes
• The first term is divergent. This divergence, however, is an artefact due to the appearance
of the tachyon. Since the tachyon will be removed in the final superstring theory we can
safely ignore this nuisance.
• The next term is due to the massless states. The long-distance behaviour is therefore
governed by massless states. Their contribution to ZT 2 is finite.
Relation to vacuum energy:
Finally let us give the physical intepretation of the partition function. Consider again a particle
theory with mass m. As we have seen ZS 1 (m2 ) computes the amplitude of a single particle
running in one loop. The total vacuum amplitude, on the other hand, contains all disconnected
vacuum loops, weighted with combinatorial factors due to permutation symmetry,
+ 12 1
+ 3! +...
ZS 1 (m2 ) + 12 (ZS 1 (m2 ))2 1

+ 3! (ZS 1 (m2 ))3 +...
The total vacuum amplitude is therefore

2
⇒ Zvac (m2 ) = eZS1 (m ) .
On the other hand, in canonical formalism
Zvac (m2 ) = h 0| e−iHT |0 i = e−iρ0 Vd ,
where T is the time in spacetime, and ρ0 the vacuum energy density. Thus
i
ρ0 = ZS 1 (m2 ). (5.116)
Vd
This analogy between ZS 1 (m2 ) and ZT 2 suggests therefore that also in string theory
i
ρ0 = ZT 2 . (5.117)
Vd
Thus ZT 2 computes the 1-loop correction to the vacuum energy, i.e. the 1-loop correction of the
spacetime cosmological constant. As we have seen, in the bosonic string this vacuum energy is
finite (if we ignore the artifical tachyon) and of order the string scale. On the other hand, in the
supersymmetric superstring theory one finds
ZT 2 ≡ 0
due to an exact cancellation between fermions and bosons running in the loop.
5.5.2 Oriented open theory

The open string oriented one-loop amplitudes are defined on the cylinder C2 , which can be
represented as a strip of length 2πt with two boundaries at σ 1 = 0, ` and the line σ 2 = 0 and
σ 2 = 2πt identified. We will set ` ≡ π in the sequel.
∼
=
• The cylinder C2 is described by one real modulus t. Unlike the torus, there is no analogue
of the modular group action P SL(2, Z). Thus the modulus can take values in the full
regime 0 ≤ t ≤ ∞.
• The group of conformal Killing transformations consists of translations parallel to the σ 2 -
axis, i.e. such that they preserve the boundary of the strip at σ 1 = 0, π. Its volume is
2πt.
The vacuum amplitude
The computation is very similar to the computation on the torus.

• For simplicity we consider the situation of a stack of N D-branes filling the entire 26-
dimensional space, corresponding to (NN) boundary conditions in all dimensions. Gener-
alisations to lower-dimensional branes are simple.
• As in the closed sector, the ghost contribution turns out to cancel the oscillator trace of
precisely two non-transverse directions. With this in mind the amplitude is
Z ∞
dt c
ZC2 = Tr e−2πt(L0 − 24 ) (5.118)
0 2t
Z ∞
dt 1
= iV26 (8π 2 α0 t)−13 Tr0⊗24 X i q L0 − 24 . (5.119)
0 2t i=1
For a stack of N concindent D-branes filling all of spacetime this is

Z ∞
dt
ZC2 = iV26 N 2 (8π 2 α0 t)−13 η(it)−24 . (5.120)
0 2t
• The IR-limit, t → ∞, works out like for the closed string: The only IR-divergent term is
due to the open tachyon, which is absent in the eventual superstring theory.
• The UV-limit t → 0, on the other hand, is different from the closed string sector: Unlike
on the torus, the UV-divergent region is not absent from the integral because there is no
analogous modular group action serving as an intrinsic regulator.
Thus, it might seem that we do face a UV divergence, contrary to our previous claim that string
theory is UV finite. To see why the divergence as t → 0 is not in contradiction with UV finiteness
we need to discuss the worldsheet duality between the open and closed string channel.
Open versus closed channel
• In the UV-divergent limit t → 0, the cylinder is infinitely long.

• The remarkable insight is the following: We can either view the long cylinder as describing
an open string stretching between the boundaries at σ 1 = 0, π and running in the loop
described by the Euclidean time σ 2 . Or, alternatively, we may interpret the annulus as a
closed string propagating at tree-level from the left to the right. The two interpretations
of the cylinder are referred to as open and closed string channel.
• Technically, the two viewpoints are related by interchanging the role of the Euclidean time
and the spatial coordinate on the worldsheet. From our analysis of P SL(2, Z) transforma-
tions of the torus on Assignment 10 we recall that an S-duality transformation τ → − τ1
exchanges the coordinates σ 1 and σ 2 . The same applies to the cylinder with it taking the
role of τ . Including a conventional rescaling of the spatial coordinate the transition from
the open to the closed string channel is accomplished by
π
t −→ s = . (5.121)
t
With the help of the transformation of the Dedekind function
1
i s 12 is
η(it) = t− 2 η = η (5.122)
t π π
the annulus amplitude in closed string channel is
Z ∞ is −24
1
ZC2 = iV26 N 2 2 0 13
ds η . (5.123)
2π(8π α ) 0 π
• The UV limit t → 0 in the open channel has translated in the IR limit s → ∞ of the closed
channel. This describes a closed string tree-level process with the string propagating over
long Euclidean time. Thus we have reinterpreted the UV-divergence as an IR-divergence.
This is in fact a general feature of string amplitudes:
All UV divergencies in string amplitudes can be reinterpreted as IR divergencies of dual diagrams.
• In fact, we can make the propagation of the closed strings visible in the limit s → ∞ by
expanding
is −24
η = |{z} 24 +O(e−2s ).
e2s + |{z} (5.124)
π
tachyon massless
The tachyonic term is again an artifact of the bosonic theory. Of importance is the second
term. It shows that the IR divergence is due to the exchange of massless closed
string states at zero momentum.
Tadpoles in string theory and field theory
A diagram where a state - here a closed string state - is created from the vacuum is called a
tadpole.
• The IR divergence is the k12 divergence from joining two tadpoles by the propagator of a
massless state with k 2 = 0.
• The tadpole diagram as such is computed by the 1-point function of a single closed string
operator inserted in the interior of the disk. Note that such a diagram is in general non-zero
even though the 1-point function on the sphere and the open 1-point function on the disk
vanish - see the remark after (5.28).
• In QFT a tadpole diagram results from a term linear in the field in the lagrangian as this
is what gives rise to a single field vertex. Tadpoles therefore signal an instability of
the vacuum, which is defined as the locus in field space that satisfies V 0 (φ) = 0. For
example in the presence of a tadpole for a bosonic field,
Z
1
S = − (∂φ)2 + Λφ (5.125)
2
the locus φ = 0 does not correspond to the true vacuum. If we set out at φ = 0 the field
configuration will change. In the presence of higher terms there may be a new vacuum at
φ 6= 0 and the theory will flow to that correct vacuum. In the above action, by contrast,
there is no such vacuum and the theory is entirely unstable.
• To see how to deal with the tadpole in string theory we need to include also unoriented
worldsheets.
5.5.3 Non-oriented vacuum amplitudes and tadpole cancellation

So far we have been analysing mainly the oriented closed and open string. The unoriented theory,
aka orientifold theory, was introduced on Assignment 6 as the quotient of the oriented theory by
worldhseet parity Ω : σ → −σ. In the orientifold theory, we must include in addition amplitudes
on non-oriented Riemann surfaces.
• For the closed and open theory at χ = 0, respectively, these are the Klein bottle and
the Möbius strip. For brevity we can only mention these amplitudes here and refer the
interested reader e.g. to [P], Chapter 7 for details on the geometry of these un-oriented
Riemann surfaces and the computation of the vacuum amplitudes on them.
• Suffice it here to note the following important result: Both the Klein bottle and the Möbius
strip vacuum amplitude suffer from UV/IR-divergencies in form of tadpoles. Summing over
Klein bottle, annulus and Möbius amplitude the total IR divergence in the closed channel
can be written as a perfect square and is proportional to
Z ∞
(213 ± N )2 ds, (5.126)
0
where N is the number of spacetime-filling coincident D-branes and the two signs depend
on the details of the orientifold projection.
• This means that the theory is tadpole free if we choose the projection such that the lower
sign arises and take N = 213 = 8192. With this orientifold projection the gauge group on
the spacetime-filling branes is
SO(8192). (5.127)
Tadpole cancellation as a stringy consistency condition
One can read this result as follows:
• A priori, any number N of spacetime filling D-branes seems to be allowed. However,

precisely if N = 8192 the theory is tadpole free, meaning that the perturbative string
vacuum is indeed stable (up to the annoying tachyon which, as we reiterate, doesn’t bother
us much for the reasons stated.)
• While for the bosonic theory is merely a toy model, the same logic will fix, in the superstring
theory, the gauge group of the 10-dimensional open plus closed theory to be SO(32) (more
precisely Spin(32)/Z2 ).
• Upon compactification of some of the 26 or, respectively, 10 dimensions more general gauge
groups are consistent with the requirement of tadpole cancellation, but not every anomaly-
free gauge configuration that is allowed in field theory is fully consistent within string
theory. Tadpole cancellation implies absence of gauge anomalies in the effective theory,
but it is stronger. It is only one example of such stringy consistency conditions with no
analogue in field theory. This is because string theory is indeed a theory of quantum
gravity, and the consistent coupling of a gauge sector to gravity entails further constraints.
5.6 Strings on curved backgrounds

5.6.1 The non-linear σ-model
So far we have focused on string propagation in flat ambient spacetime by coupling the string
fields in the Polyakov action to the flat metric ηµν of R1,25 ,
1
Z √
SP = 0
d2 ξ h hab ∂a X µ ∂b X ν ηµν . (5.128)
4πα
To extend the theory to string propagation in a curved target spacetime with metric Gµν (X) we
generalise the Polyakov action to
1
Z √
Sσ = 0
d2 ξ h hab ∂a X µ (ξ) ∂b X ν (ξ) Gµν (X(ξ)) . (5.129)
4πα
The curved metric Gµν (X(ξ)) can be understood as a coherent state of gravitons describing
the fluctuations of the metric around ηµν . To appreaciate this let us consider as an analogy a
laser field in quantum optics. We set out in perturbative QED to quantize the electromagnetic
vacuum and describing its fluctuations by photons. A coherent state of these vacuum fluctuations
represents a non-trivial field configuration, in this case a laser field. Likewise, the gravitational
field as encoded in Gµν (X(ξ)) can be viewed as a coherent excitation of gravitons. This statement
can be made more precise as follows:
Suppose Gµν (X) describes a close-to-flat metric so that we can expand
Gµν (X) = ηµν + χµν (X). (5.130)
Inserted into the string theoretic path-integral this becomes

1
Z
2
√ ab µ ν

exp(−Sσ ) = exp(−SP ) 1 − d ξ hh ∂a X ∂b X χ µν (X(ξ)) + . . . . (5.131)
4πα0
Working for simplicity on a flat worldsheet, the effect of the curved metric is captured by insertion
and exponentiation of a graviton vertex operator
Z
1 ¯ ν χµν (X(ξ)).
V = d2 z ∂X µ ∂X (5.132)
2πα0
For comparison, previously we have only considered the special case
χµν = gc 2πα0 ζµν eikX (5.133)
corresponding to a plane graviton wave.

A general deviation χµν from the flat metric describes a superposition of such graviton waves.
This indeed justifies the general lesson:
The string propagates in a background described by a coherent state of its own massless fluctu-
ations!
Once we have realised that propagation in curved spacetime corresponds to coupling the string
fields to the graviton sector of its massless excitations, we are naturally lead to including also
the other massless fields. In the closed sector these are
Bµν = B[µν] , the Kalb-Ramond B-field, and

Φ, the dilaton.
A natural generalisation of the closed Polyakov action is therefore the closed σ-model action
1
Z √ n o
d2 ξ h hab Gµν (X) + iab Bµν ∂a X µ ∂b X ν + α0 R(2) Φ(X) .

Sσ = (5.134)
4πα0 Σ
• The factor ab is the antisymmetric worldsheet two-tensor. It is required by antisymmetry

of Bµν .
• The factor of i follows by Wick rotation to Euclidean worldsheet signature because ab
always involves one time derivative.
• R(2) is the Ricci scalar on the worldsheet. R 2 √ (2)

1
The last term is thus a generalization of the topological term 4π d ξ hR λ = λχ with
λ = Φ(X). Most importantly, the strings coupling is therefore really given by
gs = eΦ = eΦ(X) , i.e. the coupling ”constant” is dynamical. (5.135)
Lesson:
In string theory there are no dimensionless coupling constants. All couplings are determined by
VEVs (vacuum expectation values) of dynamical fields.
Symmetries of the σ-model action:

• Poincaré invariance in spacetime, which holds for SP , is replaced by general covariance.
• In addition a new symmetry arises: A transformation of the Kalb-Ramond 2-form B =
1 µ ν
2 Bµν dx dx of the type
B → B + dΛ, i.e. Bµν → Bµν + ∂µ Λν (X) − ∂ν Λµ (X), (5.136)

leaves the action Sσ invariant. Under this symmetry the 3-form field strength H = dB
with components
Hαβγ = ∂α Bβγ + ∂β Bγα + ∂γ Bαβ (5.137)
is invariant. As anticipated already around eq. (3.117) this generalises the concept of an
abelian gauge symmetry to higher-rank gauge potentials.9
The string therefore carries ”charge” under this symmetry. Indeed the coupling
Z √
d2 ξ hab Bµν ∂a X µ ∂b X ν (5.138)
dτ ∂τ X µ Aµ of a point particle to a gauge field.

R
generalises the coupling
Comment on open string σ-model:
For open strings the transformation (5.136) leads to boundary terms. These vanish if we include
a coupling of the background gauge field Aµ (X) to the string endpoints along with a suitable
transformation of Aµ . This is discussed in detail in Assignment 11, to which we refer for more
details.
5.6.2 α0 -Expansion and Conformal Invariance

The string action for propagation in flat space leads to a free theory on the worldsheet, which,
as we have seen, can be solved exactly. By contrast, the non-linear σ-model is an interacting
theory.
To analyze the worldsheet interaction terms we ignore for the time being the terms describing
the worldsheet coupling to Bµν and Φ and focus on
1
Z √
S= 0
d2 ξ hhab Gµν (X)∂a X µ ∂b X ν . (5.139)
4πα
9 In mathematical terms, B is the connection of a gerbe, much in the same sense in which a gauge potential
Aµ is the connection of a vector bundle.

It is instructive to expand √
X µ (ξ) ≡ X0µ + α0 Y µ (ξ) (5.140)
√
into a ξ-independent background value X0µ and the fluctuations Y µ (ξ). The factor of α0 renders
these dimensionless. A Taylor expansion of Gµν (X) in the action,
√
Gµν (X)∂X µ ∂X ν = α0 Gµν (X0 ) + α0 Gµν,ρ (X0 )Y ρ (ξ)
α0

+ Gµν,ρτ Y ρ (ξ)Y τ (ξ) + ... ∂Y µ ∂Y ν , (5.141)
2
yields an infinite number of couplings in the worldsheet theory with the following properties:
∂G 1
• If the target space has typical radius Rc then ∂X ∝ Rc , and the effective dimensionless
√
α0
coupling constant is of the order Rc. Thus the worldsheet theory is weakly coupled if
√
α0
Rc << 1. Indeed the extended nature of the string becomes important for Rc ∼ = ls , i.e.
when the σ-model is strongly coupled.
• This establishes the following important concept: String perturbation theory really involves
a double expansion:
g 2 = eΦ ⇔ loop expan-
sion from spacetime per-
spective
√
α0
Rc ⇔ loop expansion on
the worldsheet
√
α0
• From the spacetime perspective, the expansion in Rc will be interpreted as an expansion
in higher derivatives of the (background) metric.
√
0
• For Rαc → 1 there exists no accurate point particle description of the effective theory
because the variation of the metric over distances of the string is big. This leads to the
concept of ”stringy geometry”: A string probes the target space geometry very differ-
ently from a point particle. E.g. string theory can be perfectly-well defined on backgrounds
which contain certain singularities. While point particle theory would be inconsistent in
such a background, string theory can effectively resolve these geometries.
Conformal invariance
The central condition in order to define a well-defined quantum string theory is, as we have seen,
conformal invariance on the worldsheet, i.e. the constraint c(tot) = c(X) + c(ghost) = 0.
• For the free theory of strings propagating in flat spacetime R1,d−1 , where the string coor-
dinates are simply described by a copy of d X-CFTs with c = 1, this translates into the
constraint d = 26.
• The interacting worldsheet theory corresponding to string propagating in the most general
curved background cannot be solved exactly at the quantum level, i.e. it is in general not
known√ how to find the exact spectrum of string states etc.10 However, for backgrounds
0
with Rαc << 1 we can treat the worldsheet interactions perturbatively. As in conventional
QFT we start from
√
the free theory and compute deviations order by order in the coupling,
0
here given by Rαc .
• From the logic of the path integral conformal invariance continues to be a consistency con-
dition for the so-defined quantum theory. However, we need to check if extra constraints
arise from requiring conformal invariance in addition to the constraint that d = 26. Put dif-
ferently we need to analyze if the interactions on the worldsheet induce new contributions
to the Weyl anomaly of SX .
• For a general QFT, conformal invariance implies scale invariance. It is not known whether
the converse is also true in general. However, for a 2-dimensional unitary QFT with com-
pact spatial dimensions (a requirement shared by our worldsheet theory), scale invariance
does imply full conformal invariance11 . It is therefore sufficient to analyze possible devi-
ations from scale invariance due to the worldsheet interactions. In QFT such a deviation
is captured by the β-function. We thus seek to compute the β-function of the space-time
metric
∂
βνρ (G) = µ Gνρ (X; µ), with µ ≡ the energy scale. (5.142)
∂µ
Absence of a Weyl anomaly is equivalent to
βνρ (G) = 0. (5.143)

√
α0
This must be enforced order by order in Rc in σ-model perturbation theory.
Computation of 1-loop β-function:

√
0
As a first approximation we evaluate βνρ (G) at 1-loop level in Rαc . From standard QFT we
recall the following procedure to compute the 1-loop β-function of a coupling g (e.g. the gauge
coupling in QCD):
• Compute the 1-loop correction to the coupling g.
1
• Extract the divergent piece, e.g. as the term in dimensional regularisation.
• The 1-loop β-function is β(g) = g× coefficient of 1 -term.

For the β-function of Gνρ (X; µ) in the case at hand we must compute the loop correction to the
propagator (i.e. the self-energy of the scalar field Y µ (ξ)). This computation proceeds as follows,
where are not careful about numerical prefactors:
10 Nonetheless, for special background manifolds the corresponding interacting CFT can be solved exactly; e.g.
propagation of the superstring on a certain class of Calabi-Yau manifolds can be described by Gepner models,
which are modular invariant tensor products of minimal models for which an exact solution is known.
11 See the seminal paper by J. Polchinski, ”Scale and Conformal Invariance In Quantum Field Theory”, Nucl.
Phys. B 303, 226 (1988).

• Locally around X0 we choose Riemann normal coordinates familiar from General Rel-
ativity. In these the Levi-Civita metric takes the form
α0
Gµν (X) = δµν − Rµλνκ (X0 )Y λ Y κ + O(Y 3 ) (5.144)
3
(5.145)
so that
√ α0
Z
1 2 a µ ν λ κ a µ ν
S = d ξ h δµν ∂ Y ∂a Y − Rµλνκ (X0 )Y Y ∂ Y ∂a Y + ... (5.146)
4π 3
• This leads to a quartic interaction vertex with the following Feynman rule:
' Rµλνκ kaµ (k a )ν
Here kaµ denotes the the 2-dimensional worldsheet momentum of the scalar field Y µ and
appears due to the two derivatives in (5.146).
• There is thus a one-loop correction to the kinetic term of the scalar fields from the diagram
In momentum space this correction is proportional to

d2 p
Z
Rµλνκ (k µ · k ν )

λ
Y (p)Y κ (p)

2
. (5.147)
(2π) | {z }
propagator in momentum space
• This integral is divergent from the p → ∞ region as can be anticipated already from the
logarithmic divergence ξ → ξ 0 of the propagator in position space,
1
Y (ξ)Y κ (ξ 0 ) = − δ λκ log(|ξ − ξ 0 |2 ).

λ
(5.148)
2
• We can easily extract the divergence by dimensional regularisation. Setting d = 2 + the
divergent piece of the integral is isolated as
Z 2+ Z 2+ λκ
1 δ λκ
Z
d p
λ κ d pδ 1
dp p−1 '

2
Y (p)Y (p) = 2 2
= . (5.149)
(2π) (2π) p 2π 2π
The counterterm for this divergence is proportional to
1
− Rµν ∂Y µ ∂Y ν in Sσ . (5.150)

• The β-function at 1-loop is therefore given by α0 Rµν . We conclude that absence of an

(extra) Weyl anomaly at 1-loop requires
!
βµν (G) = α0 Rµν = 0. (5.151)
This is a spectacular result: The above equation is nothing other than the Einstein equation for
the vacuum. Indeed since we have set all other fields, such as Φ and B to zero in our computation,
the target space is precisely described by the vacuum, i.e. pure geometry with no matter.
√
0
To first order in Rαc consistency of the σ-model yields the Einstein equation for
background metric.
We have just derived the dynamical laws of Einstein gravity as a corollary from worldsheet
consistency of a string propagating in the corresponding target space. This confirms that string
theory indeed yields a quantum theory of gravity.
What is more, we can next compute the 2-loop β-function, with the result
1 !
(2)
βµν = α0 Rµν + α02 Rµλρσ Rνλρσ = 0. (5.152)
2
This way one can systematically find stringy higher curvature corrections to the
Einstein equations.
This logic generalises to the full σ-model including the coupling to Bµν and to Φ. Here we quote
the 1-loop β-functions:
α0 !
βµν (G) = α0 Rµν + 2α0 ∇µ ∇ν Φ − Hµλκ Hνλκ + O(α02 ) = 0, (5.153)
4
α0 γ !
βµν (B) = − ∇ Hγµν + α0 ∇γ ΦHγµν + O(α02 ) = 0,
2
d − 26 α0 2 α0 !
βµν (Φ) = − ∇ Φ + α0 ∇γ Φ∇γ Φ − Hµνγ H µνγ + O(α02 ) = 0.
6 2 24
The first term in βµν (Φ) can be viewed as the tree-level contribution in σ-model perturbation
theory.
The β-functions are solved e.g. for Φ = Φ0 ≡ const., d = 26 and Bµν = 0 = Rµν .
One can now deduce the low energy effective action for the fields Gµν , Bµν and Φ as the
target space action whose equations of motion reproduce the β-function equations order by order
in perturbation theory. The result can be found e.g. in [P], eq. (3.7.20), along with many more
illuminating explanations. See also Assignment 12 for a derivation of some of these facts.
Chapter 6
Superstring theory
As stressed several times the bosonic string theory, which we have analysed up to now, is merely
a toy model because of the following two major shortcomings:
• The tachyonic ground state signals a vacuum instability.
• The string spectrum contains only bosonic excitations. This lack of fermionic states is in
contrast to observations and makes the bosonic string unrealistic.
Both of these challenges are remedied in superstring theory. Concerning the tachyonic ground-
state, the basic idea is to realise that its appearance is due to the negative zero-point energy of
the bosonic string theory on the worldsheet. This suggests that one should modify the theory
by adding fermions on the worldsheet, hoping that their opposite statistics might cancel this
vacuum energy.1
In fact, we will see in this chapter that solving both abovementioned problems replaces the
bosonic string theory in d = 26 dimensions by a theory in d = 10 dimensions. All but one of the
resulting tachyon-free 10d string theories are supersymmetric.2 Supersymmetry is a symmetry
that exchanges bosons and fermions. The worldsheet superstring theory consists of a bosonic
and a fermionic sector. The bosonic sector is identical to the worldsheet theory of the bosonic
string. We can therefore view our efforts up to now as a preliminary study of one half of the
superstring theory.
There exist two major formulations of this superstring theory. Both theories enjoy supersymme-
try on the worldsheet and in spacetime, but they differ in the following respect:
• In the Ramond-Neveu-Schwarz (RNS) formulation, supersymmetry is manifest on the
worldsheet, but not in spacetime.
• In the Green-Schwarz (GS) formulation, supersymmetry is manifest in spacetime, but
not on the worldsheet.
More recently, the pure-spinor or Berkovits formulation has been developed as yet another
approach to the superstring.
In this course, we will only discuss the RNS formalism.
1 Indeed this will remove the tachyon at least in the so-called Ramond subsector of the superstring theory; in
the Neveu-Schwarz sector things turn out to be more complicated, but as we will see, consistent theories where
the tachyon is indeed absent can be found.
2 As we will see, there are five consistent, tachyon-free supersymmetric string theories in d = 10 called Type
IIA, Type IIB, Type I, heterotic E8 × E8 and heterotic SO(32), all related to each other by dualities, as well as
one non-supersymmetric tachyon-free 10d theory called heterotic SO(16) × SO(16).
128
CHAPTER 6. SUPERSTRING THEORY 129
6.1 The classical RNS action

The superstring theory is obtained by adding to the bosonic string, whose action in flat gauge is
Z
1 2
SB = − d2 ξ 0 ∂α X µ ∂ α Xµ (6.1)
8π α
with X µ a worldsheet scalar, a sector describing 2-dimensional worldsheet spinors.
• Recall that a spinor is by definition a representation of the Clifford algebra. Applied to
the two-dimennsional worldsheet with flat metric ηαβ the Clifford algebra is generated by
the two-dimensional γ-matrices with anti-commutation relations
α β
γ , γ AB = 2 η αβ 1AB where γ α = γAB
α
. (6.2)
Here A, B are spinor indices on the worldsheet and α, β are vector indices on the worldsheet,
α, β = 0, 1. For simplicity we will take all spinor indices as indices downstairs.
• A spinor ψA transforms under Lorentz transformations as
ψA → SAB ψB , SAB = [exp(iωαβ 4i [γ α , γ β ])]AB , (6.3)
with ωαβ = −ωβα parametrizing an infinitesimal Lorentz transformation. We understand
that all repeated indices, here the spinor index B, are summed over.
• We will discuss the systematic construction of spinor representations in arbitrary dimen-
sions later. For the time being we simply observe that in two dimensions we can pick the
γ-matrices as real 2 × 2-matrices, e.g. as

0 1 0 1
γ0 = , γ1 = ⇒ {A, B} = {1, 2}. (6.4)
−1 0 1 0
In view of (6.3) also ψA can be taken to be real
∗
ψ+ ψ+ ψ+
ψ= , ψ∗ = = = ψ. (6.5)
ψ− ψ− ψ−
This reality condition on the spinors is called Majorana condition, the corresponding
spinor is a Majorana spinor.
• The labelling ψ± refers to the chirality, i.e. to the eigenvalues under γ ≡ γ 0 γ 1 ,

ψ+ ψ+ 0 0
γ = , γ =− . (6.6)
0 0 ψ− ψ−
Spinors of definite chirality are called Weyl spinors.
• The objects ψ± are thus Majorana-Weyl spinors, i.e.they are both real and chiral. Such
Majorana-Weyl spinors exist in 2 mod 8 dimensions.
Let us now define the RNS action in flat gauge as the action obtained by adding the canonical
kinetic terms for free bosons and Majorana fermions on the worldsheet,
Z
1 µ α
S=− d2 ξ α20 ∂α X µ ∂ α Xµ + 2 i ψ̄A γAB ∂α ψµ B , (6.7)
8π
where
µ
µ ψ+ µ
• ψA = µ with ψ± representing Grassmann valued spacetime vectors, and
ψ−
• ψ̄ = ψ † γ 0 = ψ T γ 0 = (−ψ− , ψ+ ).
In light-cone coordinates
S =SB + SF , (6.8)
Z
1
SB = d2 ξ α20 ∂+ X · ∂− X, (6.9)
2π
Z
1
SF = d2 ξ i (ψ+ · ∂− ψ+ + ψ− · ∂+ ψ− ). (6.10)
2π
Note that the mass dimensions of the fields are [X] = −1, [ψ] = 12 , explaining the relative factor
of α10 .
µ
Ignoring potential boundary terms, to be discussed later, the equation of motion for ψA is the
Dirac equation
γ α ∂α ψ = 0 or in components ∂+ ψ− = 0 , ∂− ψ+ = 0. (6.11)
Thus the Weyl spinors ψ± are also chiral in the sense that ψ± = ψ± (ξ± ).
As can be verified by brute force computation, the action SB + SF is invariant under the
fermionic symmetry
q
µ
2
¯ψ µ ,
α0 δX = i
q (6.12)
δψ µ = α20 12 γ α ∂α X µ · .
Here ≡ A denotes an infinitesimal Majorana spinor,

+
A = with + , − Grassmann fields. (6.13)
−
In order for (6.12) to be a symmetry of the full action, A (ξ) must obey
γ β γα ∂β (ξ) = 0. (6.14)
In components the symmetry (6.12) reads
q
2 µ µ µ
α0 δX = i (−− ψ+ + + ψ − ),
q (6.15)
µ
δψ± = ± α20 ∓ ∂± X µ .
• The symmetry (6.12) relates the bosonic and fermionic degrees of freedom. This is the
characteristic property of a supersymmetry (SUSY).
• The symmetry is “chiral” in the sense that (6.14) becomes
∂+ + = 0, ∂− − = 0. (6.16)
• In the above formulation the symmetry closes only upon use of the e.o.m., i.e. it holds
on-shell.
Remarks:
• SUSY is a deep concept that extends (in some sense uniquely) the Poincaré symmetry.
While found for the first time in the context of the two-dimensional RNS theory, it has
become an important principle of more general physical systems.
• The supersymmetry generators QA are spinorial and can be shown to obey the typical
anti-commutation relations
QA , Q̄B ∼ α

= 2 γAB Pα , (6.17)
where the momentum operator Pα generates translations.

• Just like translations are related to position and thus to the conventional spacetime co-
ordinates, the existence of a fermionic symmetry QA implies the notion of fermionic, i.e.
Grassmann-valued coordinates. Together with the conventional, bosonic coordinates of
spacetime these span what is called superspace. We do not discuss the formulation of
RNS string in superspace notation in this course, referring the interested reader instead to
[P], Chapter 12.3. In such a superspace formulation the supersymmetry holds off-shell, i.e.
without use of the equations of motion.
6.2 Super-conformal invariance

In the previous section we have presented the RNS action formulated on a flat worldsheet with
metric ηαβ . As in the bosonic theory, this action is really the result of gauge fixing an action
formulated on a worldsheet with dynamical worldsheet metric hαβ . This action describes 2-
dimensional gravity coupled to a supersymmetric theory of worldsheet
scalars and fermions. As
a consequence of the fundamental anti-commutation relation Q, Q̄ = 2γ · P for the super-
symmetry generator QA , local diffeomorphism invariance together with supersymmetry implies
local supersymmetry. Such a theory is called two-dimensional supergravity (SUGRA). In
µ
the same way as the worldsheet spinors ψA are the superpartners of the bosons X µ (in that they
are exchanged by supersymmetry, cf (6.12)), the metric hαβ has a superpartner whose degrees
of freedom give rise to the gravitino.
For brevity we do not discuss this supergravity in this course. The details can be found in [BLT],
Chapter 7.2. and 7.3. Suffice it here to state that the logic leading from the full action to the
gauge fixed action is as for the bosonic string and proceeds along the following lines:
• The full action enjoys local super-Weyl and diffeomorphism invariance. In particu-
lar, the supersymmetry parameter A = A (τ, σ) is unconstrained.
• After fixing the gauge to flat gauge we are left with a remnant super-conformal sym-
metry, where SUSY is only “chiral” (− = − (ξ + ), + = + (ξ − )) and diffeomorphism
invariance reduces to conformal symmetry.
• The generators of the super-conformal symmetry, i.e. their conserved currents are
– the energy-momentum-tensor,
1 i µ
T++ = − ∂+ X · ∂+ X − ψ ∂+ ψ+ µ , (6.18)
α0 2 +
1 i µ
T−− = − 0 ∂− X · ∂− X − ψ ∂− ψ− µ . (6.19)
α 2 −
Conformal symmetry implies T+− = 0 = T−+ , conservation of the energy-momentum
tensor means ∂+ T−− = 0 = ∂− T++ .
– the supercurrent, which can be found via Noether’s method as
Z r
2 α 1 2 µ
δSUSY S = d ξ (∂α ¯)J , J± = − ψ ∂± Xµ . (6.20)
2 α0 ±
To derive this form of J± certain conditions from local supersymmetry must be used,
see [BLT], Chapter 7.2 and 7.3. From this form of the supercurrent we see that
∂− J+ = 0, ∂+ J− = 0. (6.21)
• As we fix the gauge by integrating out the worldsheet metric and its superpartner, the
gravitino, one must keep their equations of motion in the form of generalised constraint
equations. These are the super-conformal Virasoro constraints
! !
T±± = 0, J± = 0, (6.22)
which must be imposed on solutions of the equations of motion if we use flat gauge RNS
action.
6.3 Mode expansions: Ramond vs. Neveu-Schwarz

We continue with our analysis of the flat gauge RNS action and proceed to the boundary condi-
tions and mode expansion of the worldsheet fields.
• The bosonic mode expansion and boundary conditions are just as in the bosonic string.
• The fermionic equations of motion follow from variation of
Z
i
SF = d2 ξ(ψ+ ∂− ψ+ + ψ− ∂+ ψ− ), (6.23)
2π
which yields
Z
δSF ∼= dτ [ψ+ δψ+ − ψ− δψ− ]σ=`
σ=0 + terms vanishing upon use of the e.o.m. (6.11) .(6.24)
a) Closed sector:
The boundary terms at σ = 0 and σ = ` cancel each other,

!
ψ+ δψ+ − ψ− δψ− = ψ+ δψ+ − ψ− δψ− . (6.25)

σ=0 σ=`
The most general boundary conditions that do not mix ψ+ and ψ− and respect spacetime
Poincaré symmetry are
µ µ
ψ+ (σ) = ±ψ+ (σ + `), (6.26)
µ µ
ψ− (σ) = ±ψ− (σ + `). (6.27)
Since ψ is a worldsheet spinor, the minus sign is possible as we go around the worldsheet
once, taking σ → σ + `. This was not possible for worldsheet scalars X µ .
Thus there are 4 independent sectors since for ψ± we can independently choose either sign,
In short the boundary conditions can be written as
ψ± (σ + `) = e2π i φ± ψ± (σ), where (6.28)

φ = 0 denotes the Ramond sector, and
φ = 12 denotes the Neveu-Schwarz sector.
i) The Ramond sector (R) corresponds to periodic boundary conditions with integer mode
expansion
r
µ
X 2π 2π
ψ− (τ, σ) = bµ e− ` i n(τ −σ) , (6.29)
` n
n∈Z
r
µ
X 2π 2π
ψ+ (τ, σ) = b̃µ e− ` i n(τ +σ) . (6.30)
` n
n∈Z
ii) The Neveu-Schwarz sector (NS) corresponds to anti-periodic boundary conditions with
half-integer mode expansion
r
µ
X 2π µ − 2π i r(τ −σ)
ψ− (τ, σ) = b e ` , (6.31)
1
` r
r∈Z+ 2
r
µ
X 2π µ − 2π i r(τ +σ)
ψ+ (τ, σ) = b̃ e ` . (6.32)
1
` r
r∈Z+ 2
The four different sectors are therefore
(φ+ , φ− ) = (0, 0) ↔ RR, (6.33)

1 1
(φ+ , φ− ) = ( , ) ↔ NS-NS, (6.34)
2 2
1
(φ+ , φ− ) = ( , 0) ↔ NS-R, (6.35)
2
1
(φ+ , φ− ) = (0, ) ↔ R-NS. (6.36)
2
b) Open sector:
The boundary terms at σ = 0 and σ = ` vanish individually. This relates ψ+ and ψ− as

µ µ
ψ+ (σ) = ±ψ− (σ) . (6.37)

boundary boundary
2 × |{z}
We are left with |{z} 2 sectors, of which two differ only by an insignificant overall sign.
± σ=0, `
Consider a direction µ with Neuman-Neumann (NN) boundary conditions for X µ . By con-

vention we choose the overall sign such that

µ µ
ψ+ (σ) = ψ− (σ) . (6.38)

σ=0 σ=0
There are now two inequivalent cases

µ µ
ψ+ (σ) = η ψ− (σ) , η = ±1. (6.39)

σ=` σ=`
i) The Ramond sector corresponds to periodic boundary conditions with integer modes:
r
µ

µ

µ π X µ − π i n(τ ±σ)
ψ+ (σ) = ψ− (σ) ⇒ ψ± (τ, σ) = bn e ` . (6.40)

σ=` σ=` `
n∈Z
ii) The Neveu-Schwarz sector corresponds to anti-periodic boundary conditions and half-
integer modes:
r
µ

µ

µ π X µ − π i r(τ ±σ)
ψ+ (σ) = −ψ− (σ) ⇒ ψ± (τ, σ) = br e l . (6.41)

σ=` σ=` ` 1
r∈Z+ 2
In both cases reality implies bµr = (bµ−r )† .
If we consider (DD) boundary conditions for X µ instead of (NN) we find the corresponding
solution for the superpartners by noting that a change of boundary conditions corresponds to a
sign flip for the rightmovers,
X+ → X+ , X− → −X− . (6.42)
By worldsheet supersymmetry, we must also flip the sign of ψ− ,
ψ+ → ψ+ , ψ− → −ψ− . (6.43)
Since this only gives an overall sign for the boundary conditions at σ = 0 and σ = `, the modes
continue to be (half-)integer moved in the (NS)/R sector.
By contrast, mixed DN or ND boundary conditions correspond to a sign flip only at one of the
two boundaries. Therefore, the R-sector is now half-integer moved, and the NS-sector modes are
integer moved.
In the sequel, unless stated otherwise, we refer to (DD) or (NN) boundary conditions, i.e. the
NS sector is the half-integer one.
6.4 Canonical quantisation and Super-Virasoro-Algebra

We now turn to canonical quantization of the RNS string.
6.4.1 Canonical (anti-)commutation relations

µ
• The X-sector modes continue to enjoy the familiar commutation relations [αm , αnν ] =
µν
mδm+n,0 η .
µ
• The fermions ψA satisfy the canonical anti-commutation relations
µ ν
(τ, σ 0 ) = 2π η µν δ(σ − σ 0 ),

ψ+ (τ, σ), ψ+ (6.44)
µ ν
(τ, σ 0 ) = 2π η µν δ(σ − σ 0 ),

ψ− (τ, σ), ψ− (6.45)
µ ν
(τ, σ 0 ) = 0.

ψ+ (τ, σ), ψ− (6.46)
i
This is easy to see because ψ± and 2π ψ± are canonically conjugate variables. Since they are
anti-commuting their canonical Poisson-brackets are replace by 1i × the anti-commutator.
The corresponding anti commutation relation for the modes is
n o
{bµm , bνn } = b̃µm , b̃νn = η µν δm+n,0 . (6.47)
• The ground state of the Fock space is annihilated by the positive modes.
i) NS sector:
µ
αm |0 iNS = 0 ∀m > 0 (m ∈ Z) (6.48)
bµr |0 iNS = 0 ∀r > 0 (r ∈ Z + 1
2) (6.49)
(6.50)
µ
and similarly for α̃m , b̃µr for closed strings.
|0 iNS is the unique ground state. It is a spacetime scalar.
µ
The negative modes α−|m| , bµ−|r| create excited states from this ground state. All
states in the NS sector are thus spacetime bosons.
ii) Ramond sector: The ground state is defined by
µ
αm |0 iR = bµn |0 iR = 0, ∀ m, n > 0 (m, n ∈ Z). (6.51)
Unlike in the NS sector, this does not uniquely specify the ground state. The reason
is that the state bµ0 |0 iR is not zero, but is likewise annihilated by all positive modes.
In fact we will see below that |0 iR and bµ0 |0 iR have the same mass, i.e. they are
degenerate. Therefore |0 iR is a degenerate ground state.
To describe the ground state further we observe that the Ramond zero-modes sat-
isfy the (ambient spacetime) Clifford algebra because the anti-commutation relations
(6.47) imply
{bµ0 , bν0 } = η µν . (6.52)
Compared with the space-time Clifford-algebra {Γµ , Γν } = 2η µν satisfied by d-dimensio-
nal Γ-matrices we identify
1
bµ0 = √ Γµ . (6.53)
2
This shows that the ground state |0 iR furnishes a representation of the d-dimensional
Clifford algebra.
|0 iR is a spinor in d-dimensions.
We therefore need to learn more about spinors in an arbitrary number of dimensions.
6.4.2 Interlude: Spinors of SO(1, d − 1)

We focus on even-dimensional spaces and set d = 2 + 2 k.
• The key idea for finding the representations of the Clifford algebra {Γµ , Γν } = 2 η µν =
2 diag(−1, +1, . . . , +1) is to build a set of (k + 1) raising and lowering operators as
Γ0 ± = 12 (±Γ0 + Γ1 ),
(6.54)
Γa ± = 12 (Γ2a ± iΓ2a+1 ), a = 1, . . . , k.
These satisfy the anti-commutation relations

a + b −
Γ ,Γ = δ ab ,
a − b − a + b + (6.55)
Γ ,Γ = Γ ,Γ = 0.
• Since (Γa ± )2 = 0 we can find a state ζ such that Γa − ζ = 0 ∀ a, if necessary by acting on

a certain vector once with Γa − . Given such ζ we can now construct 2k+1 states by letting
Γa + act on ζ zero or one times for all a = 0, . . . , k. These 2k+1 states assemble as the
components of a state
1 1 1
|s i = |s0 , s1 , . . . , sk i = (Γk + )sk + 2 · . . . · (Γ0 + )s0 + 2 · ζ with sj = ± ∀j = 0, . . . , k.
2
(6.56)
The state |s i is called a Dirac spinor. It has 2d/2 complex components.
• The combination
i
Σµν = − [Γµ , Γν ] (6.57)
4
is readily checked to form a representation of the SO(1, d − 1) algebra. Letting
1
Sa := iδa,0 Σ2a,2a+1 = Γa + Γa − − (6.58)
2
the Dirac spinor |s i, defined as a representation of the Clifford algebra, furnishes also a
representation of SO(1, d − 1)3 given by
Sa |s i = sa |s i. (6.59)
3 This is an abuse of notation. We really mean Spin(1, d − 1), the double cover of SO(1,d-1), of course.
• In an even number of dimensions, a Dirac spinor is not irreducible. Rather a Dirac spinor
can be reduced further into 2 Weyl spinors of opposite chirality. To this end we define
Γ = i−k Γ0 Γ1 . . . Γd−1 (6.60)
with the properties
Γ2 = 1 , {Γ, Γµ } = 0 , [Γ, Σµν ] = 0, (6.61)

k+1
Γ=2 S0 S1 . . . Sk . (6.62)
In view of (6.59) we find that
if an even number of sa = 12 ,

1
Γ |s i = (6.63)
−1 if an odd number of sa = 21 .
The eigenspinors with Γ eigenvalue ±1 are called positive and negative chirality Weyl
spinors. The Dirac spinor can be decomposed as
[2d/2 ]Dirac → [2d/2−1 ]Weyl ⊕ [2d/2−1 ]Weyl . (6.64)
For example we will heavily use that in d = 10 dimensions, a Dirac spinor has 32 complex
components and be decomposed into 16-component Weyl spinors, [32]Dirac = [16] ⊕ [160 ].
• For d = 2k + 3, we can add Γ to the set of Γµ of one dimension less, i.e. of SO(1, d − 2).
Together these satisfy the Clifford algebra in d = 2k + 3 dimensions as follows from (6.61).
Thus the dimensionality of the Clifford algebra representation in d = 3 + 2k is the same
as in d = 2 + 2k. We still have 2k+1 × 2k+1 square matrices and we can still construct a
2k+1 -component Dirac spinor. Now Γ does not commute with Σµ,d−1 and thus the Dirac
spinor is irreducible. This just means that chirality is not an available concept in odd
dimensions.
• A Majorana spinor satisfies a certain reality condition. Without proof we note that this
is possible only if d = 0, 1, 2, 3, 4 mod 8. Spinors can be both Majorana and Weyl only if
d = 2 mod 8. for more information we strongly recommend Appendix B of [P], Volume 2.
Let us now return to the R-sector ground state |0 iR . Being a spinor in d dimensions it has 2[d/2]
complex components. However, the Majorana condition on bµ0 = (bµ0 )† implies that these 2[d/2]
components are real. We will find further reductions momentarily.
µ
Finally, all states in the R-sector are obtained by acting with the creation operators α−|m| and
µ
b−|m| on the vacuum |0 iR . They are therefore spacetime fermions.
6.4.3 Super-Virasoro-Algebra and physical state condition

In old canonical quantisation we must impose the super-Virasoro constraints (6.22)
! !
T±± = 0, J± = 0. (6.65)
We first define the modes of the super-Virasoro generators, focusing on open strings to avoid
double-writing.
i) The energy-momentum modes take the form

Z `
` π π
Lm = − 2 dσ (ei ` m σ T++ + e−i ` m σ T−− ) = L(b) (f )
m + Lm (6.66)
2π 0
with
X
L(b)
m =2
1
: α−n · αm+n : m ∈ Z, (6.67)
n

X m 0 R
L(f )
m =2
1
(r + ) : b−r bm+r : m ∈ Z, φ= 1 (6.68)
2 2 NS
r∈Z+φ
In particular the zero mode reads

∞
X X
L0 = 12 α02 + N , number operator N= α−n · αn + r b−r br . (6.69)
n=1 r∈Z+φ
r>0
ii) The modes of the supercurrent J are

r Z
1 ` ` π π
Gr = − dσ (ei ` r σ J+ + e−i ` r σ J− ), (6.70)
π π 0
Z + 12

X NS,
Gr = α−m br+m r∈ (6.71)
Z R
m∈Z
Note that while the energy momentum tensor is always integer moded, the supercurrent
modes are integer in the Ramond sector and half-integer in the NS sector (for NN/DD
boundary conditions; for ND/DN this is reversed).
They form the Super-Virasoro-Algebra:
c
[Lm , Ln ] = (m − n) Lm+n + m(m2 − 2 φ)δm+n,0 ,
12
m
[Lm , Gr ] = ( − r) Gm+r , (6.72)
2
c
{Gr , Gs } = 2 Lr+s + (4r2 − 2 φ)δr+s,0
12
with

0 R
φ= 1 (6.73)
2 NS
The superconformal anomaly is

3 1
c= d = d( |{z}
1 + ). (6.74)
2 2
X-CFT |{z}
ψ-CFT
This reflects the familiar fact that the degrees of freedom of two fermions count as one real boson.
Let us now evaluate the physical state conditions that follow from the super-Virasoro con-
straints T±± = 0 = J± .
i) The NS sector super-Virasoro constraints translate into
(L0 − aNS ) |φ i = 0, (6.75)

Lm |φ i = 0, m > 0, m ∈ Z, (6.76)
1
Gr |φ i = 0, r > 0, r ∈Z+ 2. (6.77)
The open string mass shell condition (focussing on NN boundary conditions for all dimen-
sions) thus reads
α0 M 2 = N − aNS (6.78)
in terms of the normal ordering constant aN S , which will be determined momentarily.

ii) The R sector physical state condition is
(L0 − aR ) |φ i = 0, (6.79)
Lm |φ i = 0, m > 0, (6.80)
Gn |φ i = 0, n ≥ 0. (6.81)
In the R-sector, the super-Virasoro-algebra (6.72) implies L0 = G20 . Thus as a consistency

condition
!
aR = 0. (6.82)
We will confirm momentarily that this matches the explicit computation from ζ-function
regularisation.
Remark: The R-sector constraint 0 = G0 |φ i, expressed in terms of the modes, takes the
form
∞
" #
1 X
p·Γ+ √ ( α−n · bn + b−n · αn ) |φ i = 0. (6.83)
α0 n=1
Here we identified bµ0 = √12 Γµ . The zero mode piece is just the Dirac equation for a massless
space-time fermion. The costraint 0 = G0 |φ i thus furnishes the stringy generalisation of
the Dirac equation in the same manner as the Virasoro constraint (L0 − a) |φ i = 0 yields
the stringy form of the mass shell equation p2 + M 2 = 0.
6.4.4 Normal ordering constants

The contribution of each field X µ , ψ µ to the normal ordering constants aNS , aR follows e.g. by
ζ-function regularisation.
• From the discussion of the bosonic string we recall the following reasoning that led to the
(b)
normal ordering constant of one periodic boson: From the definition of L0 as the
normal-ordered sum (6.67) given more explicitly by
∞ −1
(b) 1 2 1X 1 X
L0 = : α0 + α−n αn + α−n αn : (6.84)
2 2 n=1 2 n=−∞
with
−1 ∞ ∞ ∞
1 X 1X 1X 1X
α−n αn = αn α−n = α−n αn + n (6.85)
2 n=−∞ 2 n=1 2 n=1 2 n=1
| {z }
=−a
we concluded that the normal ordering constant is

∞
1X 1 1
a=− n = − ζ(−1) = . (6.86)
2 n=1 2 24
• Likewise for periodic fermions we compute

∞ ∞
(f ) 1X 1X
L0 =: n b−n bn + (−n)bn b−n : (6.87)
2 n=1 2 n=1
| {z }
1
P∞ P∞
2 n=1 n b−n bn − 21 n=1 n
and conclude
1
a=− . (6.88)
24
• For anti-periodic bosons (upper sign) and anti-periodic fermions (lower sign), the normal
ordering constant is
∞
1 1X 1
a = ∓ ζ(−1, q)|q= 12 ≡ ∓ (n + q)−1 |q= 21 = ∓ . (6.89)
2 2 n=0 48
This is summarised in the following table.

1
1 periodic boson a = + 24
1
1 anti-periodic boson a = − 48
1 (6.90)
1 periodic fermion a = − 24
1
1 anti-periodic fermion a = 48
6.5 Open string spectrum in light-cone quantisation (all

NN)
As in the bosonic theory, the physical spectrum is most easily determined in lightcone quanti-
sation, where the residual gauge symmetries - here the super-conformal symmetry - is exploited
at a classical level to solve for the super-Virasoro constraints explicitly. We briefly outline the
procedure here - focussing for simplicity to the open string with Neumann-Neumann boundary
conditions - and then move on to a detailed discussion of the spectrum.
• The residual conformal symmetry allows us to set
X + (τ, σ) = x+ + p+ τ (6.91)
in light-cone coordinates X ± = √1 (X 0
2
± X d−1 ).
µ
• In addition it is simple to see that the chiral SUSY transformations (6.15) on ψA allow us
to set
+
ψA (τ, σ) = 0. (6.92)
by a suitable choice of chiral SUSY parameter ∓ .
−
• One can then solve for X − (τ, σ) and ψA (τ, σ) by exploiting
T±± = 0 and J± = 0. (6.93)
−
• In particular one can solve for αm and b−
r in terms of (d − 2) transverse modes and plug
these into the Hamiltonian to find the mass formula for physical states.
i) NS-sector
The result of this procedure is the mass-shell condition
∞
X ∞
X
α0 M 2 = i
α−n αni + r bi−r bir −aNS . (6.94)
n=1 r= 12
| {z } | {z }
=:N (X)
=:N (ψ)
The spectrum thus organises as follows:
• The ground state |0; k iNS is a spacetime scalar state with momentum k µ and of mass
α0 M 2 = −aNS .
i
• The first level excitations arise by acting not with α−1 , but with bi− 1 on the vacuum,
2
|ψ i = ζi bi− 1 |0; k iNS (6.95)

2
with mass
1
α0 M 2 =
− aNS . (6.96)
2
This is a transverse vector of SO(d−2) and must thus be massless in a Lorentz invariant
theory. Therefore
1
aNS = . (6.97)
2
On the other hand, in the NS sector
1 1
aNS = (d − 2)( + ), (6.98)
24
|{z} 48
|{z}
X-CFT ψ-CFT
which fixes the critical dimension of the superstring to be
d = 10. (6.99)
Note that the ground state is still tachyonic. We will see how to construct a consis-
tent theory without the tachyon in the context of the GSO-projection the subsequent
chapters.
• The states at the second excited level,

i
|ψ i = (ζi α−1 + ζ[ij] bi− 1 bj− 1 ) |0; k iNS , (6.100)
2 2
8

comprise 8 + 2 = 36 transverse components. Since this is a massive state of mass
1
α0 M 2 = (6.101)
2
it must organise into an irreducible representation of the little group SO(9). Indeed the
36-component representation of SO(9) is just the anti-symmetric as follows by counting

9
numbers of d.o.f. of antisymmetric of SO(9) = = 36. (6.102)
2
To summarise the first few level of the open NS-string tower consist of
• a tachyonic 1 of SO(9),
• a massless 8V of SO(8),
• massive bosons in tensor representation of SO(9).
ii) R-sector:
The mass-shell condition is
∞
X ∞
X
α0 M 2 = i
α−n αni + nbi−n bin . (6.103)
n=1 n=1
1
Indeed, as noted already, the normal ordering constant vanishes because aR = (d − 2)( 24 −
1
24 ) = 0.
• The ground state |ua , k iR is a massless (k 2 = 0) spacetime spinor. Based on the

notation |s i = |s0 , . . . , sk i for a spacetime spinor as in section (6.4.2) we introduce the
symbol us for the wavefunction or polarisation of the various spinor components
|ua , k i = |s, k i us . (6.104)

|{z}
polarisation
As a consequence of the Majorana condition on the R-sector zero modes bµ0 , |s i is a

priori a Majorana spinor of SO(10). As discussed at the end of section (6.4.2), in
10 dimensions Majorana spinors can be decomposed further into Weyl spinors, corre-
sponding to the splitting
32 = 16 ⊕ 160 with real components.
The prime denotes a negative chirality Weyl spinor.

Under the decomposition
SO(1, 9) → SO(1, 1) × SO(8) (6.105)

| {z } | {z }
x± trans. direct. xi
induced by going to spacetime lightcone gauge, the Weyl spinors decompose as

1
16 → ( 12 , 8) ⊕ (− , 80 ), (6.106)
2
1
160 → ( 21 , 80 ) ⊕ (− , 8). (6.107)
2
Furthermore |s i must satisfy the Dirac equation (6.83) due to the supercurrent zero-
mode constraint,
G0 |0 iR = 0.
Since k 2 = 0 we can pick w.l.o.g. k0 = k1 , k i = 0. The Dirac equation reads 0 =
kµ Γµ |0 i = (k0 Γ0 + k1 Γ1 ) |0 i, where
0 = kµ Γµ |0 i = k0 Γ0 + k1 Γ1 = −k1 Γ0 (Γ0 Γ1 − 1). (6.108)
Here we used that (Γ0 )2 = −1.

1
Recalling from section that (6.4.2) S0 = Γ0,+ Γ0,− − 2 we rewrite (6.108) as
1
0 = −2 k1 Γ0 (S0 − ). (6.109)
2
Thus the Dirac equation implies (S0 − 12 ) |0 iR = 0, i.e. only the components s0 = 1
2
are kept for the on-shell vacuum. In all, we have
|0 iR = ( 21 , 8) ⊕ ( 12 , 80 ). (6.110)
• All higher excitations form massive spinors in irreducible representations of SO(9).
For later purposes we define the fermion number

P∞ i i
1 b−r br (NS)
F = Pr=
∞ 2 i i (6.111)
n=1 b−n bn (R)
with the property that

1 for even number of b-excitations
(−1)F = P ∞ i i (6.112)
n=1 b−n bn for odd number of b-excitations
We will also need the G-parity operator
(NS) : G = (−1)F +1 , (6.113)

F
(R) : G = Γ(−1) , Γ = Γ0 Γ1 . . . Γ9 . (6.114)
The lowest level open spectrum can be summarised as follows:

Sector G-parity SO(8) m2 Statistics
(NS) + 8v 0 boson
1
(NS) - 1 − 2α 0 boson
(R) + 8s 0 fermion
(R) - 8c 0 fermion
Here 8s and 8c represent a postive and negative chirality Wel spinor, respectively.
6.6 Closed string spectrum in LCQ

We now turn to the closed string spectrum in lightcone quantisation. Up to the level-matching
condition the left- and right-moving sectors are independent. Choosing NS and R periodicity
conditions and positive or negative G-parity for the left- and right-moving sectors independently
would give rise to 2 × 2 × 2 × 2 = 16 independent sectors of the form (R+ , R+ ), (R− , R+ ),
(NS+ , R+ ) etc. We will now see how to construct consistent string theories out of these.
The mass-shell condition is
α0 2 !
M = (N − a) = (Ñ − a) (with the last equality due to level matching) (6.115)
4
Here
N = N (X) + N (ψ) (left-moving), (6.116)
Ñ = Ñ (X) + Ñ (ψ) (right-moving), (6.117)
aN S = 12

and a = . (6.118)
aR = 0
One observes that in the (NS− )-sector (N − a) ∈ 2Z+1
2 , while in the (NS+ ), (R+ ), (R− ) sectors
(N − a) ∈ Z. Therefore the (NS− ) sector cannot pair with the (NS+ ), (R+ ), (R− ) sectors due
to level matching. This reduces the numbers of possible sectors to only 10 = 16 - (3+3).
A consistent superstring theory is formed by combining various of these 10 sectors. The procedure
how to do so goes by the name of the GSO projection and will be discussed in the next section.
Before we come to this, we analyse the lowest-lying states contained in these 10 sectors.
• The lowest-lying state is in the NS− ⊗ NS− sector and given by the ground state
2
|0; k iN S ⊗ |0; k iN S , forming a 1 of SO(8) with m2 = − . (6.119)
α0
Again, the ground state is tachyonic.
• At the massless level we find the following possible states (up to interchange of left- and
right-movers):
sector state SO(8) statistics
(NS+ , NS+ ) bi− 1 |0 iN S ⊗ b̃i− 1 |0 iN S 8v ⊗ 8v boson
2 2
(R+ , R+ ) |+ iR ⊗ |+ iR 8s ⊗ 8s boson
(R+ , R− ) |+ iR ⊗ |− iR 8s ⊗ 8c boson
(R− , R− ) |− iR ⊗ |− iR 8c ⊗ 8c boson
(NS+ , R+ ) bi− 1 |0 iN S ⊗ |+ iR 8v ⊗ 8s fermion
2
(NS+ , R− ) bi− 1 |0 iN S ⊗ |− iR 8v ⊗ 8c fermion
2
Decomposition into irreducible representations of SO(8)

a) The decomposition of the states in the (NS+ , NS+ ) sector into irreducible representations of
SO(8) proceeds in the same manner as for the closed string spectrum of the bosonic string,
i.e. by decomposing
8v ⊗ 8v = [0] + [2] + (2)
Kalb-Ramond field Bµν graviton Gµν (symmetric
dilaton Φ (scalar)
(anti-symmetric 2-form) traceless 2-tensor)
The (NS+ , NS+ ) has the same degrees of freedom the closed sector in the bosonic theory.
b) The RR sector involves spinor bilinears. These are bosonic states as they arise as the
product of two fermions. It must therefore be possible to decompose them into irreducible
tensor representations of SO(8).
The decomposition of spinor bilinears in tensor representations is derived in [P], Appendix

B.1, to which we refer for details. The general idea is to form spinor bilinears of the type
ζ̄Γ[µ1 . . . Γµp ] ψ for suitable p, which transform as antisymmetric tensors. Suffice it here to
state without proof the following result of this representation theoretic analysis: Consider a
theory in d = 2k + 2 spacetime dimensions with chiral Weyl spinors (2k ) and anti-chiral Weyl
spinors (2k )0 . Then,

[1] + [3] + . . . + [k + 1]+ , k even.
(2k ) ⊗ (2k ) = (6.120)
[0] + [2] + . . . + [k + 1]+ , k odd,

[0] + [2] + . . . + [k] , k even,
(2k ) ⊗ (2k )0 = (6.121)
[1] + [3] + . . . + [k] , k odd,
where
(n)
• [n] denotes a fully antisymmetric n-tensor, i.e. [n] ≡ C (n) with components C[µ1 ...µn ] ,
• [k + 1]+ denotes the self-dual part of the tensor with respect to Hodge ∗ duality.
(+) (+)
Cµ1 ...µn = ∗Cµ1 ...µn .
As a reminder, the Hodge ∗ operator maps
∗: p − form → (d − p) − form (6.122)

p
|g| µ ...µ
C (p) → ∗C (p) = Cµ ...µ 1 p νp+1 ...νd dxνp+1 ∧ . . . ∧ dxνd .
p!(d − p)! 1 p
It satisfies:
(−1)p(d−p)+1 , Lorentzian signature

∗∗ = (6.123)
(−1)p(d−p) , Euclidean signature
This leads to the following bosonic states the different sectors:
(R+ , R+ ) : 8 × 8 = [0] + [2] + [4]+

1 8
number of d.o.f. 1 + 28 + 35 = 1 + 28 +
2 4
(R+ , R− ) : 8 × 80 = [1] + [3]
number of d.o.f. 8 + 56
(R− , R− ) : 80 × 80 = [0] + [2] + [4]−

1 8
number of d.o.f. 1 + 28 + 35 = 1 + 28 +
2 4
c) The mixed R-NS sector contains spinor ⊗ vector-bilinears. Such objects are spacetime
fermions.
A detailed representation theoretic analysis, which we do not carry out here, gives the follow-
ing decomposition:
(NS+ , R+ ) : 8v ⊗ 8 = 8’ + 56
spin 1/2 dilatino λa spin 3/2 gravitino ψai
(NS+ , R− ) : 8v ⊗ 80 = 8 + 56’
λ̃a ψ̃ai
The appearance and chirality of the dilatino can be seen as follows: Let |i, si be the state
8v ⊗ 8 with i a vector index of SO(8). Then one can form the combination Γi |i, si. This
is a spinor with ΓΓi |i, si = −Γi Γ|i, si = −Γi |i, si, corresponding to the 80 dilatino. The
µ
gravitino ψai - which should be confused with the worldsheet RNS fields ψA - assembles then
the remaining 56 degrees of freedom. It caries both a vector index i and a spinor index a. We
will say more about the physical interpretation of the gravitino at the end of the next section.
6.7 The GSO projection: Type IIA and Type IIB

Our task is now to define consistent closed superstring theories by combining the 10 different
sectors discussed in the previous section in a manner that leads to a well-defined CFT on the
worldsheet. A priori, each of the 10 sectors may or may not be included in the final theory. This
would lead to 210 different theories. However, it turns out that only a few of these theories lead
to consistent interactions on the worldsheet.
To understand the possible restrictions we first need to fill in a gap in our discussion of the RNS
theory so far and formulate it in the language as a CFT on the sphere. We will be rather brief
and only summarise the main points we will need in the sequel.
µ µ
• The fermionic action SF , eq. (6.10), is conformally invariant if the RNS fields ψ+ and ψ−
are taken to be primary fields of conformal weight h = 12 .
• To define the theory on the sphere, we perform a Wick rotation and introduce the coordinate
w = τ − iσ as in the bosonic theory. Along the way we switch to the notation ψ µ (w) and
µ
ψ̃ µ (w̄) instead of ψ± .
2π
• On the sphere with coordinate z = e ` w the mode expansion of the primary fields can
compactly be written as
X bµr X b̃µr
ψ µ (z) = r+ 1 , ψ̃ µ (z̄) = r+ 1 (6.124)
r∈Z+φ
z 2 r∈Z+φ
z̄ 2
with φ = 0 ( 21 ) in the Ramond (NS) sector. Importantly, an extra 1

2 appears in the Laurent
expansion because ψ µ (z) is a primary of weight h = 12 .
• Therefore, on the sphere the fields in the Ramond sector have a branch cut, while
the fields in the NS sector do not have such a branch cut. This feature carries over to the
vertex operators for physical states. In particular, one can show that the vertex operator
associated with the vacuum state in the Ramond sector introduces a branch cut on the
sphere.
While we are at it let us at least briefly mention the following famous speciality of 2-dimensional CFT
on the sphere that goes by the name of bosonisation - the fact that the fermionic RNS theory can be
expressed entirely in terms of bosons.
Consider two RNS fields, say, ψ 1 (z) and ψ 2 (z) and combine them into the the complexified field
ψ(z) = ψ 1 (z) + iψ 2 (z), ψ̄(z) = ψ 1 (z) − iψ 2 (z). (6.125)

From our general discussion of CFT, the OPE of the RNS theory is
η µν
ψ µ (z1 )ψ ν (z2 ) ' (6.126)
z1 − z2
and similarly for the anti-holomorphic fields. This translates into
1
ψ(z)ψ̄(0) ' , ψ(z)ψ(0) ' O(z) = ψ̄(z)ψ̄(0) (6.127)
z
for the complexified field.
The idea is now that this OPE can be reproduced in terms of a bosonic field H(z) with OPE
H(z)H(0) ' −lnz. (6.128)

iH(z)
Then the operator : e : has the OPE (we omit the normal ordering symbol to reduce clumsiness)
1
eiH(z) e−iH(0) ' , eiH(z) eiH(0) = O(z), e−iH(z) e−iH(0) = O(z). (6.129)
z
Obviously the OPEs match. This suggests the identification
ψ(z) ≡ eiH(z) , ψ̄(z) ≡ e−iH(z) (6.130)
and similarly for the anti-holomorphic fields. Furthermore, it turns out that in all OPEs we identify
: ψ ψ̄ : (z) ≡ i∂H(z). (6.131)
This formulation is useful e.g. in the construction of vertex operators. The vertex operators associated
with the Ramond vacuum |s0 , s1 , . . . , s4 i is the spin field
sa H a
P
ei a . (6.132)
To introduce more general vertex operators we need the superpartners of the bc-ghost system. These
commuting βγ ghosts and their bosonisation is described e.g. in [P], 10.4.
Let us now come back to our task of combining the different sectors into a consistent superstring
theory. In fact, there are the following sources for potential inconsistencies:
a) As motivated above, a vertex operator in the R-sector has a square-root branch cut. For
certain pairs of operators this results in a monodromy as they encircle each other. These
monodromies must be absent in a consistent theory so that OPEs are single-valued. This
imposes severe constraints on which sectors can be combined with each other into a consistent
theory.
b) If one computes superstring one-loop amplitudes by generalising the technology we got to
know in the bosonic theory for CFT computations on the torus, one notes that the result is
only modular invariant if there is at least one left- and one right-moving Ramond sector
in the theory.
One can show that these consistency conditions - together with closure of the OPE - imply that
only the following combinations of sectors lead to consistent theories (see [P], 10.6. for a proof):
In Type IIB theory the following four sectors are in the spectrum:
(NS+ ; NS+ ) : Φ, B[µν] , G(µν)

(2) (4)+
(R+ ; R+ ) : C (0) , C[µ1 µ2 ] , C[µ1 ...µ4 ]
(NS+ ; R+ ) : λa , ψaµ
(R+ ; NS+ ) : λa , ψaµ
Thus the theory is chiral because left- and right-movers have the same chirality.
In Type IIA theory these four sectors are in the spectrum:
(NS+ ; NS+ ) : Φ, B[µν] , G(µν)

(3)
(R+ ; R− ) : Cµ(1)
1
, C[µ1 µ2 µ3 ]
(NS+ ; R− ) : λ̃a , ψ̃aµ
(R+ ; NS+ ) : λa , ψaµ
Here left- and right-movers have opposite chirality.

In addition the consistency conditions allow for the following two types of theories:
• Type IIB’ is as IIB, but with R+ → R− .
Type IIA’ is as IIA, but with R± → R∓ .
These are equivalent to Type IIB/IIA.
• Type 0A consists of (NS+ ; NS+ ), (NS− ; NS− ), (R+ ; R− ), (R− ; R+ ).
Type 0B consists of (NS+ ; NS+ ), (NS− ; NS− ), (R+ ; R+ ), (R− ; R− ).
The projection leading to the above consistent theories is called GSO (Gliozzi-Scherk-Olive)
projection.
We note an important difference between the Type II and the Type 0 theories:
• In Type IIA/B, the (NS− ; NS− ) sector is projected out. Since this is the sector that
contains the tachyonic ground state, these theories are tachyon-free.
• On the other hand, the Type 0A/B theories are fully consistent from a worldsheet CFT
perspective, but still contain the a tachyon in the (NS− ; NS− ) sector. These theories are
therefore dynamically unstable: Even if we allowed for these theories, a universe described
by it immediately decays and thus plays no role. We can therefore discard these theories.
Conclusion:
CFT consistency + stability of vacuum (= absence of tachyon)

⇓
Type IIA or Type IIB as closed oriented superstring theories
Important remarks:
• Type IIA and Type IIB both contain an equal number of bosonic and fermionic degrees
of freedom, e.g. 128 + 128 at the massless level. This is a necessary condition for space-
time supersymmetry, which exchanges bosonic and fermionic fields. Indeed, the Green-
Schwarz formalism makes this spacetime SUSY manifest.
• Both IIA/B contain two massless spin 3/2 fields ψaµ (ψ̃aµ ) called gravitinos.
In Type IIA, ψaµ , ψ̃aµ have opposite chirality.
In Type IIB, ψaµ , ψaµ have the same chirality.
The gravitino is the superpartner of the graviton. Just like the presence of a massless
vector boson in QFT implies a gauge symmetry since otherwise no consistent quantisation
is possible, the presence of the gravitino ψaµ implies that spacetime SUSY is local. The
low-energy limit of the Type II theories is therefore a supergravity.
• The presence of 2 independent gravitinos implies the existence of 2 independent SUSY

algebras. The notation for this is N = 2 SUSY in d=10, thereby explaining the name
Type II. The superalgebras realise what is known on general grounds to be the maximal
amount of supersymmetry in ten dimensions.
• As a consequence we have established the far reaching fact that in string theory world-
sheet consistency and stability of vacuum imply local SUSY in 10 dimensions.
• To make contact with observation one considers string theory on a spacetime where the
extra 6 spatial dimensions are small and compact, corresponding to the compactification
ansatz R1,9 → R1,3 × M 6 . In this process SUSY may or may not be broken at low energies
in the non-compact four dimensions. There is no prediction from string theory at which
scale SUSY is broken.
The low-energy effective action
The 10-dimensional low-energy effective action keeping only the massless modes for Type IIA
and Type IIB theory can be computed order by order in spacetime and worldsheet perturbation
theory, by generalising the methods we got to know in the bosonic theory. For completeness we
collect here the bosonic sector to lowest order in α0 and at string tree-level. As noted already,
these Type IIA/B supergravity actions have the maximal possible amount of supersymmetry in
ten dimensions. Their form is in fact completely fixed by supersymmetry.
The action takes the form
SIIA/B = SN S + SR + SCS , (6.133)
where SN S is the same for Type IIA/B and in fact coincides in form with the bosonic low-energy
effective action
√
Z
1 1
SN S = 2 d10 x −Ge−2Φ R + 4∂µ Φ∂ µ Φ − Hµνρ H µνρ (6.134)
2κ10 12
with H = dB the field strength of the Kalb-Ramond B-field. SR contains the kinetic terms of
the field strengths of the respective RR gauge potentials and SCS certain topological terms. In
Type IIA these are
Z
1
SR = − 2 F2 ∧ ∗F2 + F̃4 ∧ ∗F̃4 , (6.135)
4κ10
Z
1
SCS = − 2 B2 ∧ F4 ∧ F4 (6.136)
4κ10
with
F2 = dC1 , F4 = dC3 , F̃4 = F4 − C1 ∧ F3 . (6.137)
In Type IIB we have

Z
1
SR = − F1 ∧ ∗F1 + F̃3 ∧ ∗F̃3 + F̃5 ∧ ∗F̃5 , (6.138)
4κ210
Z
1
SCS = − 2 C4 ∧ H3 ∧ F3 (6.139)
4κ10
with
F1 = dC0 , F3 = dC2 , F5 = dC4 , F̃3 = F3 − C0 ∧ H3 , (6.140)

1 1
F̃5 = F5 − C2 ∧ H3 + B2 ∧ F3 . (6.141)
2 2
Self-duality is imposed at the level of equations of motion as
F̃5 = ∗F̃5 . (6.142)
6.8 Digression: Differential forms

For your convenience we here collect a number of useful mathematical facts about differential
forms and some aspects of their appearance in physics.
A p-form C (p) is a totally antisymmetric tensor of rank p:
1
C (p) = Cµ ...µ dxµ1 ∧ . . . ∧ dxµp . (6.143)
p! 1 p
• The wedge product between two forms is defined by
A(p) ∧ B (q) = C (p+q) (6.144)

(p + q)!
Cµ(p+q)
1 ...µp+q
= A[µ1 ...µp Bµp+1 µp+q ] . (6.145)
p!q!
It satisfies
A(p) ∧ B (q) = (−1)pq B (q) ∧ A(p) . (6.146)
• In n dimensions, the integral of an n-form is defined as follows:
Z Z
(n) (n)
C = dn x C01...n . (6.147)
More generally, given an n-dimensional manifold M, one can integrate a p-form over a p-dimensional
submanifold Γp of M as Γp C (p) .
R
• The exterior derivative is defined as an operator
d : p − form → (p + 1) − form, (6.148)

1
dC (p) = ∂µ1 Cµ2 ...µp+1 dxµ1 ∧ . . . ∧ dxµp+1 (6.149)
p!
so that
(dC (p) )µ1 ...µp+1 = (p + 1)∂[µ1 Cµ2 ...µp+1 ] . (6.150)
Here [] denotes normalised antisymmetrisation:
1
[µ1 . . . µp ] = ((even perm. of (12 . . . p)) − (odd perm. of (12 . . . p)) . (6.151)
p!
The exterior derivative is nilpotent:

d2 C (p) = 0. (6.152)
• One can therefore define the cohomology of the exterior derivative in analogy to the cohomology
of the nilpotent BRST operator. First define
C (p) closed ↔ dC (p) = 0, C (p) exact ↔ C (p) = dQ(p−1) . (6.153)

Then the p-th cohomology group of the n-dimensional manifold M is defined as the quotient
closed p − forms
H p (M) = . (6.154)
exact p − forms
Exact forms are said to be cohomologically trivial.
The dimension of H p (M),
bp = dim H p (M), (6.155)
is called the p-th Betti number and is an important topological invariant of M. For M compact,
bp is finite.
• Stoke’s theorem expressed in form language takes the suggestive form
Z Z
dC (p) = C (p) , (6.156)
Γp+1 ∂Γp+1
where ∂Γp+1 denotes the p-dimensional boundary of the (p + 1)-dimensional submanifold Γ(p+1) .
• The operation of taking the boundary of a p-dimensional submanifold of M is therefore dual, in
the above sense, to taking the exterior derivative of a p-form. A p-fold with ∂Γp = 0 is called a p-
cycle, and a p-fold which is the boundary of another p + 1-dimensional submanifold, Γp = ∂Ωp+1 ,
is called a p-boundary. The object Ωp+1 is a p + 1-chain. One then defines the p-th homology
of M as the set
p − cycles
Hp (M) = , (6.157)
p − boundaries
with dimension bp = dimHp (M).
• The de Rham dual of a p-fold Γp is defined as the (n − p)-form δ (n−p) (Γp ) such that for each
p-form ω (p) one has Z Z
ω (p) = ω (p) ∧ δ (n−p) (Γp ). (6.158)
Γp M
This gives a 1-to-1 pairing between homology and cohomology. In particular,
bp = bp . (6.159)
• All quantities defined so far are topological in that they exist without reference to any metric on
M. By contrast, the Hodge ∗-operator makes use of a metric gµν , in the following sense: Define
first the antisymmetric tensor

±1 for (µ1 . . . µn ) even/odd perm. of (1, 2, . . . n)
µ1 ...µn = (6.160)
0 else
Then
∗: p − form → (n − p) − form (6.161)

p
|g| µ ...µ
C (p) → ∗C (p) = Cµ ...µ 1 p νp+1 ...νn dxνp+1 ∧ . . . ∧ dxνn .
p!(n − p)! 1 p
It satisfies:
(−1)p(n−p)+1 , Lorentzian signature

∗∗ = (6.162)
(−1)p(n−p) , Euclidean signature
One can show:

1 p
ω (p) ∧ ∗η (p) = ωµ ...µ η µ1 ...µp |g|dx1 ∧ · · · ∧ dxn , (6.163)
p! 1 p
or, more importantly,
Z Z
1 p
ω (p) ∧ ∗η (p) = ωµ1 ...µp η µ1 ...µp |g|dx1 ∧ . . . ∧ dxn . (6.164)
M p! M
6.8.1 p-form potentials in physics

• In string theory we find various exmples of p-form fields. In the bosonic string the only such
example is the Kalb-Ramond 2-form potential B = 12 Bµν dxµ ∧ dxν . In superstring theory, we
have found in addition the Ramond-Ramond form potentials C (p) with p = 1, 3 in Type IIA and
p = 0, 2, 4 in Type IIB.
• These higher rank form fields can be interpreted as generalisations of the electromagnetic 1-form
gauge potential A = Aµ dxµ .
This is because the field strength
F (p+1) = dC (p) (6.165)
is invariant under the abelian gauge transformation
C (p) → C (p) + dχ(p−1) (6.166)
by virtue of nilpotency of the exterior derivative.

The so-defined field strength is closed because
dF (p+1) = d (dC (p) ) = 0. (6.167)
This is the Bianchi identity.

• The canonical kinetic term of the field strength can compactly be written
Z Z
1 p 1
Skin = − dn x |g|Fµ1 ...µp+1 F µ1 ...µp+1 = − F ∧ ∗F. (6.168)
2(p + 1)! 2
• Hodge ∗-duality shows that in n dimensions, a (p + 1)-field strength F (p+1) = dC (p) describes the
same number of degrees of freedom as an (n − p − 1)-field strength F̃ (n−p−1) = dC̃ (n−p−2) . This
can be expressed as duality in the sense
F (p+1) = ∗F̃ (n−p−1) . (6.169)
Thus a p-form potential is dual to an (n − p − 2) form potential. This is the generalisation

of electric-magnetic duality in four dimensions: If n = 4, a vector potential A(1) is dual to a vector
potential Ã(1) (see exercise 1).
• A p-form couples naturally to a p-fold via
Z
Scoup = µp C (p) , (6.170)
Γp
where µp is the charge of Γp . This is the natural generalisation of the coupling of the 1-form
potential A(1) to a point particle.
6.9 Type I theory

As we have seen, Type IIA and Type IIB are two consistent, closed, oriented superstring theories
in ten dimensions. In this section we will get to know Type I theory as the only consistent
ten-dimensional string theories with open strings.
Our starting point is the observation that the chiral Type IIB theory is invariant under the
exchange of right- and left movers. From the worldsheet perspective this symmetry is worldsheet
parity,
Ω : σ → ` − σ. (6.171)
On the string fields this symmetry is realised by the unitary operator Ω such that
Ω† X µ (τ, σ) Ω = X µ (τ, ` − σ), (6.172)

† µ
Ω ψ (τ, σ) Ω = ψ µ (τ, ` − σ). (6.173)
The induced action on the closed string oscillators has already been discussed on Assignment 6
for the bosonic theory. Generalising the logic gives
Ω† αnµ Ω = α̃nµ . (6.174)

0, R,
Ω† bµn Ω = e2πiφ b̃µn , φ= 1 (6.175)
2, NS.
One can therefore (preliminarily) define a new unoriented theory by taking the quotient
Type IIB
(Type I)closed := (6.176)
Ω
as follows:
1) Bosonic sector:
The action of Ω on the groundstate is
(NS, NS) : Ω |0 iL ⊗ |0 iR = |0 iL ⊗ |0 iR , (6.177)

(R, R) : Ω |a iL ⊗ |b iR = − |b iL ⊗ |a iR . (6.178)
The minus sign in the R-R sector is due the fermionic nature.
i) The (NS, NS) sector contains, at the massless level the following fields: Bµν = −Bνµ
forming an antisymmetric 2-tensor of of SO(8). Therefore Bµν is Ω-odd and projected out
upon taking the quotient by Ω. The remaing fields (Φ, Gµν ) transform as the symmetric
representation of SO(8) with 36 degrees of freedom. They are Ω-even and and thus
projected in.
(2)
ii) In the (R, R)-sector the massless fields behave as follows: The 2-form C[µν] is in the
antisymmetric representation of SO(8), but due to the extra (-1) from the action of Ω on
the RR sector ground state, this field is effectively Ω-even. The remaining C (0) , C (4)+
contain 36 degrees of freedom and thus combine into the symmetric representation of
SO(8). They are in turn Ω-odd.
(2)
⇒ The bosonic sector of Type Iclosed contains, at the massless level, the fields Gµν , Φ, C[µν] .
2) Fermionic sector:
The action of Ω on the R-NS and NS-R groundstate is simply
Ω |0 iL,NS ⊗ |a iR,Ramond = |a iL,Ramond ⊗ |0 iR,NS , (6.179)
i.e. it exchanges the NS-R and the R-NS sector. Thus the diagonal combination of the 2
copies of λa , ψaµ transforming as (8c , 56) of SO(8) remains in the spectrum.
Consequenctly, the fermionic sector contains, at the massless level, one copy of (λa , ψaµ )(NS-R + R-NS)
in the (8c , 56) of SO(8).
As a result, our preliminary Type Iclosed theory preserves only 1/2 of the SUSY of Type IIB and
thus realises N = 1 SUSY in 10D.
Due to the Ω-projection it must be checked anew if the theory is really fully consistent at the
level of interactions. We cannot go through these computations in detail for reasons of time, but
we can understand the logic and the consequences of this sanity check as follows.
i) Computation of the 1-loop amplitude requires summing over a torus amplitude together with
a Klein bottle amplitude, which is the unoriented Riemann surface at Euler characteristic
χ = 0 with no boundaries. As mentioned in section 5.5.3, the Klein bottle partition function
exhibits an infrared (IR) divergence. Thus, the preliminary Type Iclosed theory is
inconsistent at 1-loop level due to appearance of a tadpole.
ii) The closed string divergence can be cancelled in a Lorentz invaraint manner in ten dimen-
sions if we include open string degrees of freedom with NN boundary conditions in all 10
dimensions from the sectors (NS+ ) and (R+ ). In particular at massless level we must include
SUSY
Aµ vector boson ↔ ψa gaugino. (6.180)
Inclusion of open strings with NN boundary conditions in 10 dimensions means inclusion of
D9-branes.
iii) As in the bosonic theory the open unoriented theory contributes an annulus and Möbius
amplitude to the full partition function, each of which contain IR-divergences. Similar to
the bosonic discussion in section 5.5.3, for a stack of N coincident D9-branes the overall
tadpole turns out to be proportional to
Z ∞ Z ∞
(2d/2 ± N )2 ds = (25 ± N )2 ds. (6.181)
0 0
The two signs correspond to the two possible actions of Ω on the Chan-Paton labels for the
massless gauge bosons. For the projection corresponding to the upper sign, the surviving
gauge bosons must have a CP-matrix satisfying
M (λaij )T M = −λaij (6.182)
for M a symplectic matrix, while for the lower sign the gauge bosons have a CP matrix of
the form
(λaij )T = −λaij . (6.183)
This comes about as follows (for details see [P] I, 6.5, pg. 189-192):
A general open string state with CP labels ij transforms as
0
M2
Ω|k; iji = (−1)1+α γjj 0 |k; j 0 i0 iγi−1
0i . (6.184)
The sign factor results from the action of worldsheet parity on the open string oscillators. If we
restrict ourselves for simplicity to the bosonic string theory the action is
µ −1
Ω αn Ω = (−1)n αn
µ
. (6.185)
One can argue similarly in the superstring, where however some complications arise that we do
not discuss here. The γ-matrices, which are in general allowed, square to 1 because Ω2 = 1. This
requires
γ T = ±γ. (6.186)
For the upper sign, a U (N ) gauge transformation on the CP factors can be used to arrive at γ = 1.
In this case, the massless gauge bosons must have antisymmetric Chan-Paton factors with basis
(λaij )T = −λaij . (6.187)
For the lower sign, N must be even, N = 2k and a U (N ) gauge transformation can be used to bring
γ in the form of a symplectic matrix

0 1k×k
M =i . (6.188)
−1k×k 0
Eventually,
M (λaij )T M = −λaij . (6.189)
Such matrices define the adjoint representation of the symplectic group U Sp(k)
This defines Type I theory as the unoriented 10-dimensional theory by adding to Type Iclosed the
degrees of freedom from the R+ and NS+ open sector of 32 D9-branes subject to the orientifold
projection (6.181) with lower sign, for which (6.181) vanishes. Its gauge group is SO(32) (more
precisely Spin(32)/Z2 .)
Comment: In 10 dimensions all other brane configurations preserving ten-dimensional Poincaré

invariance except for the one of Type I theory are inconsistent. Upon compactification absence
of tadpoles can be used to construct lower-dimensional consistent brane configurations. These
are no more unique.
Final remarks on consistent theories in 10 dimensions:
• We have established 3 consistent superstring theories in 10 dimensions: Type IIA/B theory

contains closed oriented strings only, while Type I theory is a theory of closed plus open
unoriented strings with gauge group SO(32).
• In addition, one can construct two variants of the so-called heterotic string theory in
ten dimensions, which owes its name to the fact that the left- and the right-moving sectors
consist of different chiral CFTs. Concretely one combines ten left-moving copies of the RNS
µ
CFT with fields XLi and ψL i
, i = 0, 1, . . . , 9 with a right-moving bosonic theory XR ,µ=
0, . . . , 25. Extra constraints arise by requiring modular-invariance of the 1-loop amplitudes.
Heterotic theory is a theory of closed strings only, but nonetheless gives rise to gauge
symmetry in 10 dimensions. The gauge symmetry is a consequence of compactification of
the bosonic sector along the additional 26 − 10 = 16 spatial dimensions, giving rise to a
rank 16 gauge group.4 There are two possible gauge symmetries respecting all consistency
conditions: The SO(32) heterotic string and the E8 × E8 heterotic string.
• Until 1995 it seemed that all these 5 consistent theories in 10 dimensions are independent.
As a consequence of the duality revolution it was realised, however, that they are related
by dualities. Thus they should be interpreted as different manifestations of one underlying
theory.
4 We will see in the next section that for toroidal compactifications, each compactified spacetime dimension
gives rise to one abelian gauge boson, with non-abelian enhancements at special radii.
Chapter 7
Compactification, T-duality,
D-branes
In the previous chapters we have learned how to formulate a fully consistent theory of quantum
gravity and Yang-Mills theory in 10 dimensions, unique up to dualities. This theory has all
the prerequisites one expects a theory of everything to have - except for one little detail: The
spacetime we live in does not exhibit 10 large dimensions.
If we are interested in making contact with experiment we have two options: Either we discard
superstring theory completely and start searching for a fundamental theory anew. So far this
search has not lead to anything comparable to string theory, which of course is not a proof that
this must remain so. On the other hand, if we restrict ourselves to the class of theories which
are already known to exist, we can make contact with observations by requiring that the extra
6 spatial dimensions be compact and small - so small that they have not been discovered in any
experiment so far. This strategy goes by the name of compactification and follows, in fact, an
old idea by Kaluza and Klein.
7.1 Kaluza-Klein compactification in field theory

Before discussing compactification in string theory we review the idea of Kaluza-Klein compact-
ification in field theory, which goes back to the work of Kaluza in 1914 and was further extended
by Klein.
Consider therefore a field theory in D = 1+d spacetime dimensions. Let dimension xd be ”rolled
up on a circle”, i.e. identify
xd ≡ xd + 2πR. (7.1)
This corresponds to the compactification ansatz
R1,d−1 → R1,d−2 × S 1 . (7.2)
The compactified space - here S 1 - is in general called the internal space or the compactifi-
cation manifold.
For a general diffeomorphism invariant field theory this has 3 consequences:
a) There appears a Kaluza-Klein tower of massive states in (D − 1) dimensions.
156
CHAPTER 7. COMPACTIFICATION, T-DUALITY, D-BRANES 157
b) There appears an extra U (1) symmetry in (D-1) dimensions.

c) There appear massless scalar fields called modulus fields in (D − 1) dimensions.
Ad a):
Let M, N = 0, 1, ..., d − 1, d and µ, ν = 0, 1, ..., d − 1. For simplicity we consinder a massless scalar
field in D dimensions
Φ(xM ) : ∂M ∂ M Φ(xM ) = 0, (7.3)
but the reasoning can be applied to all kind of field theories. To respect the spacetime periodicity
(7.1), Φ must be periodic in xd .
The most general ansatz for Φ(x) is to take
∞
X n d
Φ(xM ) = φn (xµ )ei R x , (7.4)
n=−∞
which corresponds to expansion in a complete set of periodic functions in xd . Plugging this into
(7.3) yields
n2
∂µ ∂ µ φn (xµ ) = φn (xµ ) ∀n. (7.5)
R2
2
n
Thus, the n-th Fourier mode φn (xµ ) appears as a scalar field of mass m2n = R 2 from the per-
spective of the (D − 1)-dimensional theory. The collection of these massive scalars are called the
Kaluza-Klein (KK) tower of states. Note that the zero-mode φ0 is massless and independent
of xd .
As R → 0, the mass of the lowest-lying state m21 → ∞ and the KK tower disappears from the
low-energy spectrum. At energies E << R1 the theory looks (D − 1)-dimensional - we are in the
realm of the low-energy effective field theory.
Ad b):
(D)
The extra U (1) gauge potential arises from the components Gµd of the D-dimensional metric.
The most general metric ansatz is
(D) 2
ds2 = GM N dxM dxN = Gµν dxµ dxν + Gdd dxd + Aµ dxµ , (7.6)
i.e. we parametrise
(D)
Gµd = 2 Gdd Aµ . (7.7)
Consider for simplicity the zero modes of Gµν , Gdd , Aµ , i.e. let all components depend only
on xµ . The subgroup of diffeomorphisms in D spacetime dimensions compatible with the ansatz
(7.2) transforms as follows:
• Diffeomorphism invariance in (d − 1)-dimensions implies xµ → x0µ (xν ).
• Diffeomorphism invariance along the circle S 1 formed by xd implies xd → x0d = xd + λ(xµ ).
This entails
Aµ → A0µ = Aµ − ∂µ λ. (7.8)
Therefore Aµ is a gauge potential in D − 1 dimensions. The associated KK U (1) symmetry

descends from diffeomorphism invariance in D dimensions. Under the compactification ansatz
(7.2) the group of diffeomorphisms decomposes as
Gl(D, R) → Gl(D − 1, R) × U (1). (7.9)
Ad c):
The metric component Gdd is a scalar field from the perspective of the (D − 1)-dimensional
theory. It sets the volume of the internal space
Z 2πR p p
Vol(S 1 ) = dxd Gdd = Gdd · 2πR, (7.10)
0
i.e. the vacuum expectation value (VEV) of the scalar field Gdd determines a geometric property
of the internal space - here the volume of S 1 . The appearance of the scalar field Gdd in the D1
dimensional effective action follows by √ dimensional reduction, i.e. expansion of the D-dimensional
Einstein action. It turns out that Gdd is not constrained by a potential. One says that the
scalar field is flat. In particular, it is massless. Such flat scalar fields whose VEV determine
geometric properties of the compactification space are called moduli fields.
7.2 KK compactification of closed bosonic strings

Let us now apply this simple idea of KK compactification to string theory. We will see that
important new features appear due to the stringy nature of the theory.
Recall the closed bosonic mode expansion (setting from now on 2π/` ≡ 1)
r r
xµ x̃µ α0 µ µ α0 µ
µ
X (τ, σ) = + + (α0 + α̃0 ) τ + (α0 − α̃0µ ) σ + N + Ñ (7.11)
2 2 2 2
where r r
α0 µ α0 µ
p̃ = α0µ , p̃ = α̃0µ , µ = 0, 1, ..., D (7.12)
2 2
Under σ → σ + 2π r
α0 µ
µ µ
X (τ, σ) → X (τ, σ) + 2π (α0 − α̃0µ ) . (7.13)
2
q
0
Imposing periodic boundary conditions thus lead us to α0µ = α̃0µ = α2 p̃µ .
Now consider KK compactification of xd along an S 1 by identifying xd ∼ = xd + 2πR.
In string theory this has 2 consequences:
n
i) The momentum in direction xd is quantised as before: pd = R. Therefore
r
α0 n
(α0d + α̃0d ) = 2 . (7.14)
2 R
This is a field theory effect.
ii) In addition something new appears: We can have winding strings looping w times around
the compact S 1 .
String A is non-winding:
xd (τ, σ + 2π) = xd (τ, σ).
String B is winding ω-times around S 1 :
xd (τ, σ + 2π) = xd (τ, σ) + ω 2π R.
Here we think of S 1 as R/2π R, i.e. work in cov-

ering space.
Comparison with (7.13) yields r

2
α0d − α̃0d =
ω R. (7.15)
α0
Thus for winding strings the left- and right-moving “momenta” are independent:
r r
d m ωR α0 d α0
α0 = ( + 0 ) = pL , (7.16)
R α 2 2
r r
m ωR α0 α0
α̃0d = ( − 0 ) = pdR . (7.17)
R α 2 2
Note that we still have that the centre-of-mass momentum is given by
n
pdL + pdR = . (7.18)
R
The mass-shell condition (for simplicity again for the bosonic string) follows as always from the
Virasoro constraints
(L0 − 1) |φ i = 0, (L̃0 − 1) |φ i = 0. (7.19)
With L0 = 12 α02 + N and L̃0 = 12 α̃02 + Ñ this implies
4 4
⇒ −pµ pµ = (pdL )2 + 0
(N − 1) = (pdR )2 + 0 (Ñ − 1). (7.20)
| {z } α α
µ=0,1,...,d−1
This leads to an effective mass in (D − 1) dimensions of the form
n2 ω 2 R2 2
m2 = −pµ pµ = 2
+ 02
+ 0 (N + Ñ − 2). (7.21)
R α α
The level-matching condition 0 = (L0 − L̃0 ) |φ i now relates the left- and right-moving oscillation
numbers as
N − Ñ = −n ω. (7.22)
We observe the following structure:
• The sector n = ω = 0 gives rise to the familiar states present also for R → ∞.
• The sector ω = 0, n 6= 0 ⇒ contains the KK tower of massive KK excitations present also
in point particle theory on an S 1 .
2 2
• The sector ω 6= 0 contains winding states of mass ω αR0 .
Note that the mass formula reflects the fact that winding costs energy due to the string
tension. The winding sector represents a truly string effect not present for point particles.
Consider now the limit R → 0.
i) The KK tower disappears from the low-energy spectrum. For a point particle theory this
would be the end of the story and we would conclude that the theory becomes effectively
(1 + (d − 1)) dimensional.
1 2 2
ii) However, the winding states become light because their mass scales as m2 ∼ α0 ω R .
We conclude that unlike a point particle theory, string theory on S 1 remains effectively (1 + (d −
1))-dimensional. We will further discuss this stunning observation below, but first let us analyze
in more detail the massless spectrum.
! ! !
For generic values of R the massless spectrum corresponds to the sectors n = ω = 0, N =
!
Ñ = 1. We find the following states:
µ ν
• α−1 α̃−1 |0, k i gives rise to G(µν) , B[µν] and Φ in the non-compact 1 + (d − 1) dimensions.
µ d d µ
• (α−1 α̃−1 + α−1 α̃−1 ) |0, k i corresponds to a vector from the perspective of the 1 + (d − 1)
(D)
non-compact dimensions. This is just the U (1) potential from Gµ d encountered also for a
point particle theory.
µ d d µ
• (α−1 α̃−1 − α−1 α̃−1 ) |0, k i gives another vector, cooresponding to the component Bµ d .
d d (D)
• α−1 α̃−1 |0, k i gives a scalar corresponding to the component Gdd .
Thus we find a U (1) × U (1) gauge symmetry and a modulus in the non-compact dimensions. X
√
At special values of R extra states appear. In particular for R = α0
1
pdL,R = √ (n ± ω). (7.23)
α0
Massless states, m2 = 0, now require
(n + ω)2 + 4 N = (n − ω)2 + 4Ñ = 4 together with level-matching N − Ñ = −n ω. (7.24)
This includes the following new states:
The sector n = ω = ±1, N = 0, Ñ = 1 gives 2 more vectors. (7.25)

The sector n = −ω = ±1, N = 1, Ñ = 0 gives another 2 more vectors. (7.26)
These√are in addition to the two gauge bosons present for generic values of R. This suggest that at
R = α0 the symmetry U (1)×U (1) is enhanced to SU (2)×SU (2), where each SU (2) factor
accounts for three gauge bosons. Indeed this assertion can be checked at the level of interactions.
Remarks:
√
• This appearance of non-abelian enhancements at Rc = α0 is a stringy effect not present
in field theory.
• The compactification on S 1 can be generalised to compactification of several dimensions

on a torus T d = S 1 × . . . × S 1 .
The resulting toroidal compactification is a special case of compactification of several
dimensions on more general internal spaces.
• Recall from the closing remarks of the previous chapter how the heterotic string theory
is defined:
a) Combine a left-moving superstring theory (X µ , ψ µ )L , µ = 0, . . . , 9 with a right-
m
moving bosonic string theory XR , m = 0, . . . , 25.
n
b) Compactify XR , n = 10, . . . , 25 on T 16 to form a 10-dimensional theory.
The compactified right-moving sector gives rise to gauge bosons in the ten-dimensional
theory - arising now from the right-moving closed sector. At generic radius the gauge
group would be U (1)16 , corresponding to 16 independent vector bosons (note that each
dimension yields only U (1) as opposed to U (1) × U (1) because only the right-moving sector
contributes.) However, modular invariance of the 1-loop partition function poses strong
constraints on the theory and in particular enforces that the radii of the T 16 be at their
critical value, for which one finds non-abelian enhancement of U (1)16 . The two consistent
ways to compactly T 16 turn out to lead to gauge groups
G = SO(32) or G = E8 × E8 , (7.27)
corresponding to the two possible heterotic string theories in 10 dimensions.
7.3 T-duality
Let us first consider closed superstrings of Type IIA or Type IIB type.
a) Bosonic sector
From the mass formula
n2 1 2 2 2
m2 = 2 0 2 R ω + 0 (N + Ñ − 2) (7.28)
R α α
for closed bosonic strings compactified on a circle of radius R we observe that the spectrum
is invariant under the operation
α0
n↔ω R ↔ R0 = , (7.29)
R
which exchanges the KK momentum and the winding momentum.
This is a truly stringy feature that relies on the extended, non-local nature of a string. The
transformation (7.29) is called T-duality; it extends to an exact symmetry of the closed
CFT, including interactions. To see this we note that exchanging n ↔ ω corresponds to the
transformation
pdL → pL pdR → −pdR . (7.30)
More generally, T-duality is defined by extending this to a fully fledged parity transformation
not only of the right-moving momenta, but of the full string field including the oscillations,
XLd (z) → XLd (z) d

XR d
(z) → −XR (z), (7.31)
i.e. we map
X d (z, z̄) = XLd (z) + XR

d
(z̄) → X 0d (z, z̄) = XLd (z) − XR
d
(z̄). (7.32)
Now, one can convince oneself that replacing X d → X 0d is indeed a symmetry of the X-CFT.
Physical conclusions:
Since the spectrum and all interactions are left invariant by the transformation (7.29) we have
established that in string theory
√ √
Physics at R < α0 ∼ = Physics at R > α0 . (7.33)
√
Consequently there is a minimal distance R = α0 , corresponding to the self-dual radius. It
does not make sense to define distances smaller than this minimal radius, to√the extent that
we can always map all processes at such distances back to radii bigger than α0 .
√
Note that precisely at the self-dual radius R = α0 we have gauge enhancement U (1)×U (1) →
SU (2) × SU (2).
b) T-duality for Type IIA/B superstrings
T-duality is readily generalized to the superstring as follows: Consider T-duality along xd ,

d = 9.
9 9
• The bosonic fields transform as derived above, i.e. X̃R (z̄) → −X̃R (z̄).
9 9
• By worldsheet superconformal invariance also ψ̃R (z̄) → −ψ̃R (z̄).
In the R-sector, this implies in particular for the right-moving zero modes:
b̃80 ± i b̃90 → b̃80 ∓ i b̃90 . (7.34)
From our previous identification b̃µ0 = √1

2
Γ̃µ with Γ̃ acting on right-movers this means that
Γ̃4± → Γ̃4∓ .
We conclude that T-duality flips the chirality for right-moving spinors and therefore trans-
forms the various superstring sectors as
(R+ , R± ) → (R+ , R∓ ), (7.35)

+ ± + ∓
(NS , R ) → (NS , R ). (7.36)
This exchanges Type IIB and Type IIA theory.
The precise statement of this T-dulaity between both theories is:
∼ α0
Type IIB on S 1 with radius R = Type IIA on S̃ 1 with radius R̃ =
R
Indeed, an analysis of the vertex operators of the RR states confirms that T-duality removes
or adds a form index,
IIA IIB
Cµ1 µ2 µ3 → Cµ1 µ2 µ3 9 ,
Cµ1 µ2 9 → Cµ1 µ2 ,
Cµ → Cµ9 ,
C9 → C.
Remark on T-duality for open strings
As established above, T-duality is nothing by parity on the right-movers. In a theory with open
string, this operation exchanges Neumann and Dirchlet boundary conditions. Suppose we start
with a theory in 10 dimensions with NN boundary conditions, i.e. with D9-branes. There is only
one consistent theory of this type, Type I theory, which is built from Type IIB theory in the
way described before. T-duality along x9 transforms Type IIB into Type IIA and leads to DD
boundary conditions in x9 . Thus we arrive at a variant of Type IIA theory with D8-branes.1
This process can be repeated, suggesting that that Type IIB theory contains Dp-branes with p
odd and Type IIA theory contains Dp-branes with p even:
Type IIB Type IIA

Dp-branes with p odd p even
D9, D7, D5, D3, D1, D(-1) D8, D6, D4, D2, D0
We will now understand this feature better by taking a closer look at the dynamical nature of
D-branes.
7.4 Dp-branes as dynamical objects

D-branes are more than just hyperplanes on which open strings end. They are by themselves
dynamical objects that
• gravitate by coupling to closed strings in the NS-NS sector; i.e. they have a mass.
• are charged under RR p-form potentials.
By supersymmetry, they interact, of course, with the respective fermionic superpartners. In the
sequel we restrict ourselves to describing the bosonic interactions.
There is the following evidence for the claimed dynamical nature of D-branes:
i) In the full quantum theory the worldvolume of a Dp-brane is not static, but it undergoes
quantum fluctuations. These brane fluctuations in the normal directions are, in fact, de-
scribed by the light open-string excitations.
1 We say ’variant’ because we started from Type I, which includes quotienting by worldsheet parity Ω. Thus
also the resulting Type IIA theory is an orientifold.

n
The excitations ψ−1/2 |0, k i normal to the Dp-brane describe a massless scalar field prop-
agating along the Dp-brane. It is interpreted as a modulus field whose VEV determines
the position of the brane. Its quantum fluctuations describe brane fluctuations. This is
implied by the fact that there is non-zero momentum exchange between the DD string and
the D-brane, as can be checked explicitly for the DD-solution.
The necessity to include such quantum fluctuations can be understood by the following
analogy with the situation in the closed string sector:
• We start from a theory of closed strings in flat spacetime. This theory gives rise to
gravitons in its massless spectrum, which are nothing but the quantum fluctuations of
the dynamical metric.
• Similarly we start with an open string sector along an initially rigid hypersurface, which
gives rise to scalar fields. Their fluctuations represent the fluctuations of the dynamical
Dp-brane.
ii) A very direct argument was achieved in a seminal paper by Polchinski (1996), which com-
puted the tree-level exchange of RR and NS-NS closed strings between two Dp-branes by
considering instead a 1-loop open diagram.
∼
=
open 1-loop tree-level closed
As in the computation of the string-loop amplitude with NN boundary conditions in all

dimensions, we can transform the 1-loop open string channel into a tree-level closed string
channel amplitude. This allows one to compare the amplitude for exchange of NS-NS and
RR states with the one ontained in an effective action of extended objects with mass and
RR-charge.
More precisely, this dynamics is captured by a low-energy effective action for the world-
volume of the Dp-brane of the form
Seff = SDBI + SCS . (7.37)

| {z } |{z}
coupling to NS-NS coupling to R-R
a) The Dirac-Born-Infeld action for the Dp-brane reads

Z
1
SDBI = −Tp dp+1 ξe−Φ [− det (Gab + 2πα0 Fab + Bab )] 2 . (7.38)
• Here,
∂X µ ∂X ν
Gab = Gµν (X(ξ)) (7.39)
∂ξ a ∂ξ b
is the pullback of the ambient space metric onto the brane worldvolume. Note that ξ a
a = 0, 1, ..., p represent braneRcoordinates,
√ while X µ (ξ) describes the embedding of brane
p+1
world-volume in 10D. Thus, d ξ − det Gab is the higher-dimensional generalisation
of the Nambu-Goto action and appears naturally.
• The factor of e−Φ shows that closed strings couple at tree-level to the disk in the open-
closed CFT.
• The field strength of the U(1) gauge field propagating along a single Dp-brane, 2πα0 Fab ,
appears only in combination with the pullback of the Kalb-Ramond field,
∂X µ ∂X ν
Bab = Bµν (X(ξ)) . (7.40)
∂ξ a ∂ξ b
As we recall from Assignment 12, only the combination 2πα0 Fµν = 2πα0 Fµν + Bµν is
invariant under the closed string U(1) symmetry
1
δBµν = ∂µ ξν − ∂ν ξµ , δAµ = − ξµ (7.41)
2πα0
due to the worldsheet coupling
i
Z
2
√ ab µ ν
Z
d ξ h ∂a X ∂b X B µν + i dX µ Aµ . (7.42)
4πα0 Σ ∂Σ
• The coupling strength is governed by the brane tension

2π √
Tp = , `s = 2π α0 . (7.43)
`p+1
s
Note that expanding the square root in the DBI action leads to the kinetic term of Yang-Mills
theory plus higher order curvature corrections. These match with an explicit computation of
scattering results.
b) Chern-Simons action
So far the massless RR-sector of Type II superstrings contains the following p-forms: Type
IIA: C (1) , C (3) Type IIB: C (0) , C (2) , C (4)+ .
By Hodge duality in 10 dimensions we can dualise the associated field strengths as
∗ F (q+1) = ∗dC (q) = F̃ (9−q) = dC̃ (8−1) . (7.44)

Note that in 10 dimensions the field strengths, not the potentials are dualised. Alternatively,
we can dualise the potentials in the 8 transverse dimensions of light-cone quantisation. Recall
that it was in this framework that we had found a self-dual 4-form.
In any case, the above argument shows that C (q) and C̃ (8−q) describe the same degrees of
freedom. Thus, we can switch to a so-called ”democratic formulation” of Type II supergravity
and consider the following field content in the massless RR sector,
Type IIA : C (1) , C (3) , C (5) , C (7) ,
(7.45)
Type IIB : C (0) , C (2) , C (4) , C (6) , C (8) .
Now, a (p + 1)-form couples naturally to the worldvolume of a Dp-brane via

Z Z
(p+1)
C = dξ 0 . . . dξ p C01...(p+1) . (7.46)
Dp
Indeed, to lowest order the Chern-Simons coupling is just

Z
SCS = −µp C (p+1) . (7.47)
Dp
Further curvature terms can be inferred, e.g., by T-duality. The charge of a Dp-brane under
C (p+1) is therefore
2π
µp = p+1 . (7.48)
`s
This explains the spectrum of D-branes observed at the end of the previous section:
IIB : D(2p + 1) ↔ C (2p+2) p = −1, 0, ..., 4 (7.49)
IIA : D(2p) ↔ C (2p+1) p = 0, ..., 4. (7.50)
Only those Dp-branes exist as stable objects which have the matching RR-forms available.
E.g. a D7-brane in IIB cannot decay because it carries C (8) charge; in IIA a D7-brane would
decay (at least in R1,9 ). In fact the dynamics between Dp-branes is a rich and exciting topic
by itself.
Remarks
• For a Dp-brane of the above type, the tension (mass) and charge coincide:
Tp = µp . (7.51)
Such objects are called BPS because they are extremely with respect to the Bogomolny’i-
Prasad-Sommerfeld (BPS) bound
M ≥Z (7.52)
with Z the charge.
• The description of D-branes with the help of open string+closed string CFT is adequate if
gs is small so that a perturbative expansion makes sense. For large gs the Dp-branes back-
react substantially on the geometry of the ambient spacetime due to their mass. They
form so-called black brane solutions in supergravity, which are higher-dimensional
generalisations of black hole solutions of 4-dimensional Einstein or Einstein-Maxwell theory.
In fact, these solutions had been known entirely form a SUGRA persepctive before it was
realised in 1996 by Polchinski that they describe the same objects as the hyperplanes
associated with DD boundary conditions.
7.5 Intersecting Brane Worlds

We now describe an important application of Dp-branes: Intersecting Brane Worlds.2
The general idea can already be understood even without compactifying the extra six dimensions.
Various Dp-branes can extend along different dimensions and intersect along some subspace
that contains R1,3 . This way, interesting gauge theories and matter content arise along the
dimensions common to all branes. In fact, the structure we find is naturally that of the Standard
Model of Particle Physics! Thus, Intersecting Brane Worlds are important ingredients of string
phenomenology, the subfield of string theory that tries to make contact between string theory in
10 dimensions and our 4-dimensional world.
We will first be working in R1,9 and consider configurations of branes intersecting along R1,3 ,
temporarily ignoring complications due to compactification. Among the various possibilities we
choose as the probably simplest example a configuration of intersecting D6-branes in Type IIA
theory. Let DA and DB be two such D6-branes which fill the following dimensions:
Dimension 0 1 2 3 4 5 6 7 8 9
DA X X X X X - X - X -
DB X X X X - X - X - X
Remarks:
• The two branes intersect along R1,3 × xi = 0, i = 4, . . . , 9. We will be interested in the
physical theory along these common dimensions.
• Even though the two branes intersect along R1,3 , this setup is not yet a satisfactory ef-
fectively 4-dimensional theory. This is because all states propagating along DA and DB
propagate not only in 4 dimensions, but also in the remaining dimensions of the brane.
The effect of this is negligible to first order only if the extra dimensions are small. Indeed,
generalisations to models with 6 compact dimensions are simple.
• The above setup corresponds to an angle of π2 in the three planes spanned by x4 − x5 ,

x6 − x7 , x8 − x9 . This can be generalised to angles ϕi , i = 1, 2, 3 in the three planes.
• In the presence of a Dp-brane, Poincaré invariance of R1,9 is obviously broken in the normal
directions. This corresponds to a spontaneous breakdown in the vacuum described by the
branes. Since Poincaré symmetry and supersymmetry are related via {Q, Q̄} ' γ P , also
2 For further reading we suggest e.g. [Z], Chapter 21, or the pedagogical review ”Toward Realistic Intersecting
D-Brane Models”, http://arXiv.org/abs/hep-th/0502005. A more advanced and very comprehensive text is also
”Four-dimensional String Compactifications with D-Branes, Orientifolds and Fluxes”, http://arXiv.org/abs/hep-
th/0610327.
some amount of SUSY must be broken. It turns out that a single Dp-brane in Type II
theories (of the appropriate type) preserves only 12 of the original amount of supercharges.
• If 2 branes intersect as above, they will in general preserve different supercharges. The
total amount of supersymmetry is then generated by the supercharges preserved by both
of them. The amount of SUSY depends on the sum of angles ϕi between the branes.
You can find a detailed discussion e.g. in [P], Chapter 13.4. In the case considered here,
supersymmetry is broken completely. This is just one example of how the theory in 4
dimensions - here arising as the common locus of the intersecting branes - can enjoy much
less supersymmetry than the original theory in 10 dimensions!
• We have not taken into account any of the various subtle consistency conditions that arise
at the quantum level and that are comparable to the tadpole cancellation conditions of
Type I theory. For compact models, these conditions severely constrain the allowed brane
setups. Unlike in pure field theory model building it is not possible to simply assemble
all ingredients one would like for phenomenological reasons into a model. Rather one
must show that the string equations of motion are satisfied. In other words, each model
corresponds to a new effectively 4-dimensional vacuum of the unique 10-dimensional theory.
Consider now stacks of NA and NB coincident branes of type DA and DB respectively. We have
the following 2 different sectors in the open string spectrum:
1) Strings starting and ending on the same brane (A − A sector and B − B sector)
These contain the massless gauge bosons of gauge group U (NA ) and U (NB ), respectively (plus
their superpartners, depending on the amount of SUSY). The important feature is that along
the dimensions common to DA and DB , both types of gauge bosons propagate! Therefore along
the common R1,3 × xi = 0, i = 4, . . . , 9 the gauge group is U (NA ) × U (NB ).
2) Strings stretched between different brane stacks
This sector is due to strings starting on A and ending on B (i.e. in the A → B sector) as well as
strings starting on B and ending on A (i.e. in the B → A sector).
• Due to their tension, these are localised at the intersection of the branes, i.e. they propagate
only along R1,3 .
• As can be seen from their Chan-Paton factors, they transforms as bi-fundamentals of
U (NA ) × U (NB ). The convention is that strings in A → B sector transform as (N̄A , NB ).
The difference between the fundamental and and the anti-fundamental is that they are
complex conjugates. We take N̄A to have charge −1A under the diagonal U (1)A in U (NA ) =
SU (NA ) × U (1)A (and NA to have charge +1A .) Then, the states in the sector B → A are
in representation (NA , N̄B ).
• To determine the details of the string spectrum we need to quantised an open string with
mixed boundary conditions. For example, for the setup at hand, these are in the A → B
sector:
σ=0 ∂σ X n (τ, σ = 0) = 0, n = 0, . . . 3, 4, 6, 8,
∂τ X m (τ, σ = 0) = 0, m = 5, 7, 9,
n (7.53)
σ=` ∂σ X (τ, σ = `) = 0, n = 0, . . . 3, 5, 7, 9,
∂τ X m (τ, σ = `) = 0, m = 4, 6, 8.
This corresponds to DN boundary conditions in dimensions 4, . . . , 9 and be generalised to

arbitrary angles ϕi between the branes.
• The mixed boundary conditions modify the oscillator modings. For DN strings this has
been discussed. For general angles one arrives at fractional modings by shifting the NN
moding to
ϕ
n→n+ . (7.54)
π
Consider now the A → B sector:
• The massless Ramond sector contains one fermion corresponding to a Dirac spinor, i.e. one
chiral and one anti-chiral Weyl spinor. This is just the fermionic ground state along the
extended 4 dimensions. The GSO projection will keep only one of the two, say the chiral
α
one (R, +). Thus we have one ψAB , where α = 1, 2 denotes a 4-dimensional Weyl spinor
index.
• In the NS sector the sign of M 2 of the lowest state is determined by the normal ordering
constant. This in turn depends on the angle between the branes as these shift the modings
of the fields. For the above DN boundary conditions we find one boson of positive M 2 .
This reflects the fact the brane intersection breaks supersymmetry completely so that the
massless fermion has no massless superparter.
The B → A sector follows by letting σ → ` − σ. This is just worldsheet parity. As discussed in
the context of T-duality this flips chirality.
We therefore obtain the following massless spectrum (after GSO projection)
α α̇
ψAB : (N̄A , NB ), ψBA : (NA , N̄B ) (7.55)
The two fermions correspond to particle and anti-particle and thus describe the same degrees of
freedom.
Let us summarise our findings:
A stack of two branes DA and DB intersecting along R1,3 × pt. gives rise to a U (NA ) × U (NB )
Yang-Mills theory plus one chiral fermion transforming in the bi-fundamental (N̄A , NB ).
But wait a minute - this is just the structure of the Standard Model of Particle Physics (SM)!
Namely, the SM gauge group is the product SU (3) × SU (2) × U (1)Y and the particle content is
given by 3 generations of chiral fermions in various bifundamentals:
Particle SU (3) SU (2) U (1)Y

1
QL 3 2 6
ucR 3̄ 1 − 23
1
dcR 3̄ 1 3
L 1 2 − 12
ecR 1 1 +1
c
νR 1 1 0
Our notation is that ucR etc. are the charge conjugate of the right-handed fields and thus left-
handed.
Very crudely, this can be realised by in terms of 3 intersecting brane stacks with NA = 3, NB = 2,
NC = 1.
Remarks:
• One notices that in the SM only SU (N ) groups appear, not U (N ) (apart from U (1)Y ). In
Intersecting Brane Worlds, the diagonal U (1) ⊂ U (N ) turns out to decouple - its gauge
boson is massive. In suitable configurations precisely one linear combination of U (1)s is
massless. This must be identified with U (1)Y .
• The remaining U(1)s remain as perturbative global symmetries and account for the presence
of accidental symmetries such as baryon and lepton number in the SM. This is particularly
attractive because in the SM no explanation for the existence of these symmetries can be
given.
• To account for the correct charges of all particles, more complicated configurations than
just the above 3-stack model are required. Indeed suitable brane setups can be classified.
Toroidal Intersecting Brane Worlds
So far we have been working in R1,9 . To obtain a truly 4 dimensional effective field theory, the
extra six dimensions must be compact. The logic behind this compactification will be discussed
in a more general context in the next chapter. Here let us focus on the simplest possibility and
make a toroidal compactification ansatz
M1,9 = R1,3 × T 6 (7.56)

with T 6 a six-torus.
• It is convenient to represent T 6 as a factorisable 6-torus of the form T 6 = T 2 × T 2 × T 2 .
If we embed the brane configuration presented at the beginning of this section into such a
compact model, the two D6-branes fill R1,3 and wrap a 1-cycle in each of the 2-tori given
by one of the axes.
• More generally we can also consider a configuration as below:
• The 2 branes DA and DB now intersect in 3 points on T 6 . At each intersection point one
chiral bifundamental fermion ”lives”. Thus, in the effective theory in R1,3 we now find 3
chiral bifundamental fermions. This gives a beautiful way to think about family replication
- the fact that we have 3 chiral generations of bifundamental matter in the SM.
As alluded to above, not every brane configuration leads to a fully consistent CFT. As in Type
I theory one must check that the tadpoles of all 1-loop amplitudes cancel. This implies that we
must actually consider unoriented open strings (such that the Möbvious amplitudes can cancel
the tadpoles of the annulus amplitude). Such models are called orientifolds. Finding consistent
solutions which are compatible with the physics of the SM is an active field of modern day string
theory.
7.6 Elements of Calabi-Yau compactification

In this section we discuss more general compactifications of string theory to four dimensions.3
Our starting point is the general warped compactification ansatz
M1,9 = M1,3 ×w M6 . (7.57)
We take M1,3 to be a maximally symmetric four-dimensional space, i.e. Minkowski space,

deSitter (dS) or Anti-deSitter (AdS), while M6 is the six-dimensional internal space. The metric
corresponding to the above ansatz takes the form

A(y)gµν 0
GM N = , (7.58)
0 gmn (y)
where µ, ν = 0, . . . 3, m, n, = 4, . . . , 9 and y ≡ y m are internal coordinates. Note that A(y)

represents a so-called warp-factor. If A(y) ≡ 1 we have a direct product, while more generally
one speaks of a warped compactification.
We now specialise to one of the five incarnations of 10-dimensional string theory and focus on the
low-energy effective field theory. For definiteness let us consider Type IIA or Type IIB SUGRA.
In order for the ansatz (7.57) to furnish a consistent compactification, the equations of motion of
all fields in the effective action must be satisfied. As discussed these arise as the string consistency
conditions which ensure that the beta-functions of all fields in the non-linear sigma model vanish,
i.e. that conformal invariance on the worldsheet is preserved.
In particular, we should view the Einstein’s equation for the metric as the equation of motion for
the graviton. This justifies the interpretation that a consistent compactification manifold
M6 gives rise to a 4-dimensional vacuum of the the 10-dimensional theory. From a
conceptual point of view this is an important insight:
While the string consistency conditions single out a unique theory (up to dualities) in 10
dimensions, every 4-dimensional effective theory obtained from this by compactification
corresponds to a choice of vacuum, i.e. to a dynamical solution of the 10-dimensional
theory.
The uniqueness of the theory in ten dimensions is not in conflict with the existence of many
consistent M6 . Of course there is nothing surprising about the fact that a given theory - here
3 We strongly recommend [GSW], Vol II, Chapter 15 for further reading on Calabi-Yau compactifications.
A detailed review on flux compactifications is e.g. ”Flux compactifications in string theory: A Comprehensive
review”, http://arXiv.org/abs/hep-th/0509003.
superstring theory in 10 dimensions - can have many different solutions. Consider e.g. Einstein
gravity in 4 dimensions: It is one theory with many different solutions! Just as gravity theorists
cannot predict which of these solutions is relevant for our solar system - we must observe the
distance from the Earth to the sun, we cannot compute it from Einstein’s equations - in string
theory we must find the correct 4-dimensional solution corresponding to our world and then use
this solution to gain further insight e.g. about particle physics.
The set of 4-dimensional solutions of string theory is called the landscape of string vacua.
The most general solution of the equations of motion will break supersymmetry completely, i.e.
at the compactification scale. While this is fine in principle, for reasons of stability and of com-
putational control we try instead to preserve supersymmetry at the compactification scale; then
SUSY must be broken dynamically at a lower scale.
The condition for unbroken supersysmmetry is that the supersymmetry variation of all fields in
the vacuum must vanish. For bosonic fields this is automatic, because a boson b transforms into
a fermion f , δSUSY b = f and thus hδSUSY bi = hf i ≡ 0 as fermions have no VEV. The non-trivial
constraint is therefore that also
hδSUSY f i = 0. (7.59)
The supersymmetry charges of Type IIA/B supergravity form two independent supersymmetry
algebras. The SUSY variation associated with each of these algebras is expressed in terms of
one spacetime-dependent Majorana-Weyl spinor in 10 dimensions. From supergravity texts we
quote the following ten-dimensional SUSY variations of the fermonic fields:
• The gravitino variations give
˜ µ (i) = ∂µ + 1 ω̃µνρ Γνρ (i) ,

δψµ(i) = ∇ (7.60)
4
(i) (H,F ) (i)
1
δψm = ∇µ = ∂m + (ωmnp − Hmnp )Γnp + ... (i) (7.61)
4
in terms of the spin connection ωµνρ and the 3-form field strength H. The . . . represent
further terms dependent on the Ramond-Ramond field strengths. The tilde reminds us
that we must take the covariant derivative with respect to the warped metric
g̃µν = A(y)gµν . (7.62)
• The dilatino variation is

1
δχ(i) = Γm ∂m Φ − Γmnp Hmnp (i) . (7.63)
12
For the most general SUSY vacuum the NSNS and RR field strengths will have a non-zero VEV,
hHi =6 0 6= hF p i. Vacua with such non-vanishing field strengths are called flux vacua and have
gained a lot of attention in recent years. For simplicity, however, let us consider the special case
hHi = 0 = hF p i. The dilation variation implies Φ = const. and the SUSY conditions are
˜ µ (i) = 0 = ∇m (i) .
∇ (7.64)
In general the existence of a covariantly constant spinor poses strong conditions because it implies
∇A = 0 =⇒ 0 = [∇A , ∇B ] = RABM N ΓM N = 0. (7.65)

˜ µ (i) = 0 with respect to the metric given by g̃µν = A(y)gµν
In particular one can show that ∇
and gµν the metric of AdS, dS or Minkowski space requires
A(y) = 0, gµν = ηµν . (7.66)
This is because it implies the conditions [∇µ , ∇ν ](i) = − 21 (∇m A) (∇m A) Γµν i , where ∇µ is with
respect to the maximally symmetric metric gµν . For these spaces we have [∇µ , ∇ν ](i) = λ2 Γµν (i)
with λ < 0 for AdS, λ = 0 for Minkowski and λ > 0 for dS. Thus k + ∇m A∇m A = 0, but the
only constant value for (∇m A)(∇m A) on the compact internal space is zero.
Thus, supersymmetric compactifications to AdS are possible only in the presence of fluxes (de-
Sitter space breaks supersymmetry anyways and more elaborate techniques are required to con-
struct vacua with positive cosmological constant). What remains is to analyze the condition that
0 = ∇m (i) .
In Type IIA, (1) transforms as a 16-component chiral Majorana Weyl spinor 16 of SO(1, 9) and
(2) as the anti-chiral 160 , while in Type IIB both spinors are chiral. In view of the decomposition
SO(1, 9) → SO(1, 3) × SO(6) : 16 → (2, 4) + (20 , 40 ) (7.67)
we make the ansatz (in Type IIA)

(1) (1)
(1) = + ⊗ η + + − ⊗ η − , (7.68)
(2) (2)
(2) = + ⊗ η− + + ⊗ η− , (7.69)
(1,2)
and correspondingly in Type IIB. Here + denote two independent 4-dimensional Weyl spinors
and η+ is a spinor of SO(6) that must satisfy
∇m η+ = 0. (7.70)
The spinor η− is just the conjugate of η+ . Thus, for each covariantly constant spinor η+ we
obtain 4 + 4 supercharges in 4 dimensions (because to each component of the 4-dimensional
(i) (i)
SUSY parameters + and − we associate one Noether charge). This corresponds to N = 2
SUSY in 4 dimensions.
In fact, the existence of a covariantly constant spinor on M6 is equivalent to the statement that
M6 has SU(3) holonomy. Let us recall what this means: Consider a general Riemannian
6-fold M6 . Such a space has SO(6) holonomy: If we transport a vector field vm around a closed
loop γ, the field transforms as
0
vm → vm = (Uγ · v)m , Uγ ∈ SO(6). (7.71)
The existence of a covariantly constant vm implies that vm = (U ·v)m . This reduces the holonomy
group to a subgroup of SO(6).
Now, it is useful to consider the isomorphism SO(6) ' SU (4) (as always we really mean: Spin(6)
because we care about spinors). A chiral Weyl spinor 4 of SO(6) corresponds to the vector
representation 4 of SU(4) (with 40 corresponding to the conjugate 4̄). This is similar to the
familiar SO(1, 3) ' SU (2)L × SU (2)R and the associated representation of the chiral and anti-
chiral Weyl spinors in terms of vectors under SU (2)L/R .
The existence of a covariantly constant spinor therefore requires that the holonomy be reduced
from SU (4) to SU (3). Then the spinor in question is
 
η0
 0 
η=  0 .
 (7.72)
0
Manifolds of SU(3) holonomy are famous:
Definition 7.1. A 2n-dimensional manifold of holonomy SU (n) is called a Calabi-Yau n-fold.
One can also characterise a Calabi-Yau manifold as follows: First make the
Definition 7.2. A Kähler manifold is a complex manifold whose metric is hermitian (i.e. the
only non-vanishing metric components are gij̄ ) and can be derived from Kähler potential K as
gij̄ = ∂i ∂j̄ K.
One can show
Theorem 7.1. A manifold is Calabi-Yau if and only if it is Ricci flat and Kähler.
In particular, a Calabi-Yau manifold automatically satisfies the vacuum Einstein equations.
Thus, our supersymmetric compactification solves the equations of motion.
Compactification of Type II string theory on a Calabi-Yau 3-fold corresponds to a consistent

vacuum. The 4-dimensional effective theory reduces to N = 2 supergravity.
The explicit construction of compact Ricci flat Calabi-Yau metrics is very hard. But according
to a famous theorem conjectured by Calabi and proven by Yau, one can check a much simpler
algebraic criterion to determine if a given space allows for such a Ricci flat metric:
Theorem 7.2. A Kähler manifold is Calabi-Yau if and only if the first Chern class of the tangent
space vanishes, i.e. c1 (M6 ) = 0.
The study of Calabi-Yau manifolds is a typical subject of algebraic geometry, which provides
many tools of significant importance to model builders.
Note that the simplest example of a Calabi-Yau 3-fold space is indeed a torus T 6 . This justifies
in retrospect our toroidal ansatz of the previous section. However, the holonomy group of a T 6
is trivial, i.e. it is contained in SU (3). Therefore compactification of Type II string theory on
T 6 gives a theory with all 32 supercharges, corresponding to N = 8 supergravity.
So far we have only been dealing with the closed string sector. To arrive at interesting theories
including gauge dynamics one can combine the idea of Intersecting Brane Worlds with compact-
ification on a genuine Calabi-Yau 3-fold. To stick with our example of Type II A theory with
D6-branes, the latter fill R1,3 and wrap an internal 3-cycle Σ, i.e. a submanifold with ∂Σ = 0 and
Σ 6= ∂Γ. In the presence of a brane the amount of supersymmetry is at best 1/2 of the original
SUSY. This corresponds to 4 conserved supercharges in 4 dimensions, i.e. N = 1 supersymmetry.
For this N = 1 SUSY to be actually preserved, the 3-cycle must itself satisfy certain geometric
conditions.
This way realistic gauge theories can be obtained in a systematic manner from string compactifi-
cations. A detailed study of the landscape of string vacua is the subject of string phenomenol-
ogy.

Weigand Introduction To String Theory

Uploaded by

Copyright:

Available Formats

Weigand Introduction To String Theory

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Weigand Introduction To String Theory

Uploaded by

Copyright:

Available Formats

Introduction to String Theory

Winter term 2015/16

Institut für Theoretische Physik, Universität Heidelberg

Digitalisierung des Skripts: Max Kerstan, Christoph Mayrhofer

1 Motivation and Overview 2

2 The classical bosonic string 9

3 Quantisation of the bosonic string 31

4 Conformal field theory (CFT) 67

6 Superstring theory 128

6.8.1 p-form potentials in physics . . . . . . . . . . . . . . . . . . . . . . . . . . 152

7 Compactification, T-duality, D-branes 156

[Z ] Zwiebach: A first course in String Theory, Cambridge University Press 2004

[PC] Polchinski, Chaudhuri, Johnson: Notes on D-branes,

In addition, I will give more specific references at various places.

Motivation and Overview

1.1 Why do we study String Theory?

• String theory for idealists

1.2 The need for a quantum theory of gravity

In principle two different ways out of this dilemma are conceivable:

The fundamental objects in Nature are not pointlike, but 1-dimensional.

• Define the classical action of a string propagating in d-dimensional spacetime. Generalising

closed string: hµν spin 2 graviton

• Note the appearance of the dimensionful parameter α0 of dimension

[α0 ] = [length]2 . (1.2)

High energy dynamics:

• The number of combinatorically relevant diagrams is greatly reduced compared to point

Further properties of the theory:

• The theory predicts

The classical bosonic string

2.1 The relativistic point particle

• We will stick to ~ = c = 1 and work in signature (-1, 1,..., 1).

• SN G is invariant under reparametrization τ → τ̃ (τ ), which corresponds to local diffeo-

• The canonical momentum associated to X µ is

• The equations of motion (e.o.m.) are

For τ = τ0 the proper time this reduces to mẌ µ = 0.

by Henneaux and Teitelboim.

Ẍ µ = 0 & Ẋ 2 = −1. (2.7)

• The disadvantage of SN G is that it is not defined for m = 0.

with e(τ ) a new independent degree of freedom.

• In view of the transformation properties of e(τ ) we can rewrite SP as

where gτ τ ≡ e2 (τ ) is identified as the worldline metric.

2.2 The bosonic string action

• The propagation of a string in R1,d−1 now defines a 2-dimensional worldsheet (WS) Σ

Σ is parametrised by τ ∈ R and 0 ≤ σ ≤ l and can be viewed as the map

Σ : (τ, σ) 7→ X µ (τ, σ) ∈ R1,d−1 . (2.17)

• In complete analogy to point particles one defines the Nambu-Goto action

where dA is the area element of Σ

2.2.2 The Polyakov action

One could add two types of terms to SP :

The energy momentum tensor

An important quantity in field theory in the energy momentum tensor.

where again Gab = ∂a X · ∂b X.

• The equations of motion for hab are Tab = 0, i.e.

This will be proven in the exercises.

a consequence of the specific definition of Tab as describing response to a variation

fields other than the metric is included.

2.2.3 Symmetries of the Polyakov action

δhab = c ∂c hab + (∂a c )hcb + (∂b c )hac , (2.42)

• Invariance under global Lorentz transformation, X µ 7→ Λµ ν X ν (ξ),

δhab = c ∂c hab + (∂a c )hcb + (∂b c )hac , (2.42)

Ja = T ab b with ∇a Ja = 0. (2.62)

(P )ab = 0 ⇔ ∇a b + ∇b a = hab (∇c c ),

ξ + 7→ ξ˜+ (ξ + ) = ξ + − + (ξ + ) , ξ − 7→ ξ˜− (ξ − ) = ξ − − − (ξ − ) (2.80)

• xµ , pµ are again the c.o.m. position