String Theory I and II

Oliver Schlotterer
Uppsala University

fall/winter 2022/23

Contents
0 Introduction 5
0.1 Features and mysteries of string theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
0.2 A brief history . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.3 Outline of the courses string theory I and II . . . . . . . . . . . . . . . . . . . . . . . . . . 8
0.4 Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1 Relativistic point particle vs closed strings 9

1.1 Point-particle action = worldline length . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Lightning recap of functional differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Quantization friendly point particle action . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.4 String action = worldsheet area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.5 Quantization friendly string action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.6 Fixing conformal gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.7 Mode expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

2 Quantizing closed bosonic string 19

2.1 Opening line in covariant quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2 Constraints in covariant quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Lightcone gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.4 Lightcone quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

3 Open bosonic strings 27

3.1 Dirichlet and Neumann boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Open-string quantization and mass spectrum . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 D branes and non-abelian gauge fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

4 Basics of conformal field theory 32

4.1 Conformal symmetry in general dimensions d . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2 Conformal symmetry in two dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Towards quantum CFTs in two dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4 Comutators versus operator product expansion . . . . . . . . . . . . . . . . . . . . . . . . 37

1
 4.5 State operator correspondence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.6 Correlation functions and Ward identities . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.7 Correlation functions versus OPEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.8 Boundary conformal field theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

5 Path integrals and ghosts 48

5.1 Path-integral methods for correlators of the free boson . . . . . . . . . . . . . . . . . . . . 48
5.2 Polyakov path integral and b, c ghost system . . . . . . . . . . . . . . . . . . . . . . . . . . 50
5.3 The ghost CFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.4 Background charge in the (b, c) system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.5 Vertex operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

6 String amplitudes 61
6.1 Basic ideas in string perturbation theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.2 Closed-string tree-level amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
6.3 Open-string tree-level amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.4 Revisiting closed strings as a double copy of open strings . . . . . . . . . . . . . . . . . . . 73
6.5 Low-energy physics from string amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . 78
6.6 High-energy scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

7 Bosonic strings in background fields 83

7.1 Strings in curved spacetime . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7.2 B-field and dilaton backgrounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
7.3 Background gauge fields and Born-Infeld . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
7.4 Point particles compactified on a circle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.5 Strings compactified on a circle and T duality . . . . . . . . . . . . . . . . . . . . . . . . . 92

8 Basics of the RNS superstring 94

8.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
8.2 The RNS worldsheet action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
8.3 RNS superstrings and superconformal field theory . . . . . . . . . . . . . . . . . . . . . . 97
8.4 (Anti-)periodic boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
8.5 The β, γ superghost system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
8.6 The superstring spectrum and GSO projection . . . . . . . . . . . . . . . . . . . . . . . . 104

9 String theory and spacetime supersymmetry 108

9.1 Supersymmetry in four spacetime dimensions . . . . . . . . . . . . . . . . . . . . . . . . . 108
9.2 Open superstrings and higher-dimensional supersymmetry . . . . . . . . . . . . . . . . . . 112
9.3 Closed superstrings and higher-dimensional supersymmetry . . . . . . . . . . . . . . . . . 114

10 Superstring amplitudes 119

10.1 Superstring tree-level amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

2
A Problem set Sept 23th 2022 125
A.1 Equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
A.2 Warmup on Christoffel symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
A.3 Diffeomorphism invariance of the Polyakov action . . . . . . . . . . . . . . . . . . . . . . . 126
A.4 Noether currents for Poincaré symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
A.5 Commutation relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

B Problem set Oct 10th 2022 128

B.1 Hamiltonian constraints and Polyakov action . . . . . . . . . . . . . . . . . . . . . . . . . 128
B.2 Lightcone quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
B.3 Virasoro algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
B.4 Zeta function regularization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
B.5 Mixing Neumann and Dirichlet boundary conditions . . . . . . . . . . . . . . . . . . . . . 130

C Problem set Oct 28th 2022 132

C.1 Conformal algebra in general dimensions d . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
C.2 The free fermion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
C.3 OPEs versus (anti-)commutation relations . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
C.4 Vertex operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
C.5 Correlations functions and OPEs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

D Problem set Nov 10th 2022 136

D.1 The delta function in complex coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
D.2 Anomalous current and two-point function of the (b, c)-ghost system . . . . . . . . . . . . 136
D.3 Massless open-string states from BRST quantization . . . . . . . . . . . . . . . . . . . . . 137
D.4 Plane-wave correlators in presence of derivatives . . . . . . . . . . . . . . . . . . . . . . . 138

E Problem set Dec 13th 2022 140

E.1 Three tachyons and one gluon or graviton . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
E.2 Four-gluon amplitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
E.3 Gamma function versus Riemann zeta function . . . . . . . . . . . . . . . . . . . . . . . . 141
E.4 Monodromy relations among color-ordered open-string amplitudes . . . . . . . . . . . . . 142

F Problem set Jan 11th 2023 144

F.1 From string frame to Einstein frame . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
F.2 Born-Infeld equations of motion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
F.3 Newton gravity in four versus five spacetime dimensions . . . . . . . . . . . . . . . . . . . 146
F.4 SU (2) covariant approach to massless states . . . . . . . . . . . . . . . . . . . . . . . . . . 146

G Problem set Jan 27th 2023 148

G.1 The ground-state energy of the Ramond sector . . . . . . . . . . . . . . . . . . . . . . . . 148
G.2 NS vertex operators at the first mass level . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
G.3 Supersymmetry transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

3
H Problem set February 9th 2023 151
H.1 Spacetime supersymmetry of the superstring spectrum . . . . . . . . . . . . . . . . . . . . 151
H.2 Four-fermion amplitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
H.3 Massive three-point amplitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
H.4 Dimensional reduction and type-IIA supergravity . . . . . . . . . . . . . . . . . . . . . . . 154

4
0 Introduction
The key idea of string theory is to replace fundamental (i.e. point-like) particles by strings. While
the propagation of point particles in spacetime gives rise to 1-dimensional worldlines, strings sweep out
2-dimensional worldsheets, see figure 1.

Figure 1: Worldlines of point particles versus worldsheets of strings.

Roughly speaking, different point-particle species (quarks, leptons, gauge bosons, gravitons, higher-
spin excitations, etc.) are then reinterpreted as different vibration modes of the strings.

• Massless closed-string excitations include a spin-2 particle subject to diffeomorphism invariance.

These properties uniquely identify it as a graviton since general relativity (along with its general
coordinate invariance) is the unique theory of massless spin two! Hence, string theory is a theory
of quantum gravity.

• Massless open-string excitations are identified as abelian or non-abelian gauge bosons. Hence, string
theory gives rise to gauge interactions as we know them from the Standard Model.

• Both closed and open strings have an infinite tower of massive higher-spin excitations, where the
maximal spin grows linearly with the mass squares.

0.1 Features and mysteries of string theory

• String theory is a perturbative theory of quantum gravity. Its loop diagrams are free of UV di-
vergences which can be intuitively understood as follows: Vertices in Feynman diagrams contain
several special interaction points whose lack of spatial extent can be viewed as causing UV- or short-
distance divergences (in the integral over loop momenta). The analogues of Feynman diagrams in
string perturbation theory are worldsheets of different topologies, where the interactions zones are
smeared out (see figure 2) such that the UV behaviour is softened.

• Interacting open strings (in particular their loop diagrams) necessarily introduce closed strings.
Hence, by the gauge bosons and gravitons among their massless excitations, string theory necessarily
unifies gauge interactions with gravity!

• String theory “predicts” its spacetime dimensionality D through mathematical consistency condi-
tions. We find D = 26 for the bosonic theory and D = 10 for supersymmetric strings. The observed

5
Figure 2: Ultraviolet-divergent Feynman integral (in dimensions D > 8) versus ultraviolet-finite world-
sheet diagram in one-loop closed-string amplitude.

D = 4 spacetime can be accommodated with the ten dimensions of superstrings via compactified
(i.e. very small) extra dimensions as well as extra ingredients such as D branes.

• String theory geometrizes gauge groups (say the SU (3) × SU (2) × U (1) of the Standard Model),
interaction strengths and numbers of generations (the three generations of quarks and leptons in
the Standard Model), for instance through the number of intersection points of D branes. Also,
string theory induces or exploits creative generalizations of geometry.

• Even though the theory of superstrings is unique after taking dualities into account, it has a large
number of solutions. Depending on the way of counting these solutions, one can arrive a large
numbers of vacua such as 10500 . In spite of this huge landscape, it is very hard to reproduce the
Standard Model features.

• Consistency requires spacetime supersymmetry, an extension of Poincaré symmetry (Lorentz trans-

formations plus translations) that mixes bosons and fermions. More specifically, strings require
q
supersymmetry at high energies (e.g. the Planck scale G~cN ∼ 1019 GeV). This is by no means in
conflict with the non-observation of supersymmetry at LHC (as of fall 2020) since the electroweak
or LHC-accessible scales are so much lower (in the range of 104 GeV).

• In contrast to the more than 20 free parameters of the Standard Model, string theory only requires
a single fundamental parameter as an input. This “string scale” can either be incorporated into a
1
mass- or length scale, related by Mstring = `string . Alternatively, one speaks of the inverse string
tension or “Regge slope” α0 = `2string . The coupling constant in string perturbation theory (which
determines the relative weight of different worldsheet topologies) in turn is dynamically determined
by the vacuum expectation value (VEV) of a massless scalar field in the closed-string spectrum, the
so-called dilaton. Also, geometric data such as the size of extra dimensions can be traced back to
VEVs of scalar fields.

• String theory is notoriously difficult to test experimentally. This difficulty is a common theme to
any candidate theory of quantum gravity since higher-derivative corrections to general relativity
are very hard to measure. On the one hand, there are considerations of an extremely low string
scale Mstring ∼ 104 − 106 GeV (based on large extra dimensions) which might give rise to envision
string signatures at LHC. On the other hand, there is a wide range of possibilities (across at least
15 orders of magnitude) on where to expect the string scale.

• String theory benefits from and inspires beautiful mathematics:

6
 ∗ Mirror symmetry between Calabi-Yau manifolds (certain compactification geometries for 6
real or 3 complex spacetime dimensions) that maps different pieces of topological information
to each other and which is inspired by a duality between so-called type IIA and type IIB
superstring theories

∗ The low-energy or α0 -expansion of string amplitudes introduces certain special numbers and
functions that are under active investigation in both particle physics (in the context of Feyn-
man integrals) and pure mathematics (number theory and algebraic geometry): multiple zeta
values and polylogarithms (tree-level amplitudes) and their elliptic generalizations (one-loop
amplitudes). There are many open questions on the number-theoretic content of higher-loop
amplitudes.

• String theory reveals surprising connections between gauge theory and gravity.

∗ At weak coupling, various examples of scattering amplitude in perturbative gravity can be as-
sembled from squares of building blocks in gauge-theory amplitudes. Such amplitude relations
are sometimes phrased as gravity = (gauge theory)2 and have a clean string-theory derivation
at tree level due to the relation between open- and closed-string amplitudes. Similar rela-
tions among field-theory amplitudes are observed at the level of the loop integrand and have
a similar string-theory origin from the chiral-splitting approach to string loop amplitudes.

∗ According to Maldacena’s AdS/CFT correspondence, string theory in AdSd+1 spacetimes is

dual to conformal gauge theories on their d-dimensional boundary. This correspondence is also
known as the gauge-gravity duality. It relates the weak- and strong-coupling regimes of the two
sides and often allows to infer the result of prohibitively hard computations in one theory from
more accessible ones in the dual theory. The AdS/CFT correspondence is a nice realization
of the holographic principle, an idea in quantum gravity stating that the information on a
higher-dimensional system may be encoded in a lower-dimensional one.

• Still, our understanding of string theories is rather incomplete,

∗ Apart from the notable exception of the AdS/CFT correspondence for certain boundary condi-
tions, string perturbation theory has not yet been derived from a non-perturbative formulation
of string theory.

∗ D branes are hypersurfaces in spacetime where the endpoints of open strings are confined.
D branes must be dynamical in a non-perturbative formulation of string theory, but they
become infinitely heavy at weak coupling. Hence, the dynamics of D branes is invisible to
string perturbation theory (similar to massive degrees of freedom that have been integrated
out)

0.2 A brief history

• In the late 60’s, string theory was born out of an attempt to study hadron spectra and interactions.
In particular, the so-called Veneziano amplitude (1968) – a particularly crossing symmetric expres-
sion for a candidate amplitude for hadron scattering – is often referred to as the first equation of
string theory.

7
 • In 1974, Scherk and Schwarz suggested to interpret the massless spin-2 excitation of closed strings
as a graviton. Prior to this proposal, the spin-2 excitation was viewed as a disturbing feature of
strings, but thanks to the new viewpoint of Scherk and Schwarz, string theory was ready to be
interpreted as a candidate theory of quantum gravity.

• In 1984, Green and Schwarz showed that potential gauge anomalies of the type I superstring cancel
for suitable choice of the gauge group. Prior to this remarkable discovery, superstring theories were
under the reputation that anomalies would inevitably spoil their consistency. With the remarkable
discovery of anomaly cancellation – sometimes referred to as the first superstring revolution – a
broad range of researches started to work on string theory.

• In 1995, Witten explained how to relate a total of 5 known superstring theories through a web of
dualities, triggering the second superstring revolution. While string theory was supposed to yield
a unique theory of quantum gravity, a total of 5 different-looking incarnations of superstrings have
been known (I, IIA, IIB, hetSO(32) , hetE8 ×E8 ) that threatened the desired uniqueness. Thanks
to Witten’s web of dualities, these five theories were recognized as different phases of a single
overarching theory dubbed “M theory”.

• In 1998, Maldacena came up with the aforementioned AdS/CFT correspondence.

0.3 Outline of the courses string theory I and II

The 16 lectures of the “String theory I” course will cover

• Relativistic point particle vs closed strings

• Quantizing bosonic strings

• Conformal field theory

• Polyakov path integral and ghosts

• String perturbation theory and tree amplitudes

while “String theory II” is planned to cover the following topics

• Bosonic strings in background fields

• Basics of the RNS superstring

• String theory and spacetime supersymmetry

• Superstring scattering amplitudes

• String dualities and AdS/CFT

0.4 Literature
In “String theory I” and earlier parts of “String theory II”, I will mainly follow

• David Tong, “String Theory”, Part III lecture notes given in Cambridge, arXiv:0908.0333

8
 • Ralph Blumenhagen, Dieter Lüst, Stefan Theisen, “Basic Concepts of String Theory”, Springer
2013, ISBN 978-3-642-29496-9

and further resources will be mentioned in the “String theory II” course. Moreover, there is a large body
of nice textbooks which I can recommend for complementary reading:

• Joseph Polchinski, “String theory I. An introduction to the bosonic string” and “String theory II.
Superstring theory and beyond”, Cambridge University Press 2005

• Michael Green, John Schwarz, Edward Witten, “Superstring theory I. Introduction” and “Super-
string theory II. Loop amplitudes, anomalies & phenomenology”, Cambridge University Press 1987

• Katrin Becker, Melanie Becker, John Schwarz, “String theory and M-theory, a modern introduc-
tion”, Cambridge University Press 2006

• Elias Kiritsis, “String theory in a nutshell”, Princeton University Press 2019

• Philippe Di Francesco, Pierre Mathieu, David Sénéchal, “Conformal field theory”, Springer 1997

• Martin Ammon, Johanna Erdmenger, “Gauge / Gravity duality, foundations and applications”,
Cambridge University Press 2015

Acknowledgements
I am grateful to Gang Xu for typing an early version of the first four sections of these notes. Moreover,
Oscar Arandes Tejerina and Chen Huang are thanked for their diligence and patience in spotting and
pointing out a variety of typos.

1 Relativistic point particle vs closed strings

Unless otherwise specified, we will consider strings in a background1 of D-dimensional Minkowski space-
time R1,D−1 with t = X 0 and a metric in “mostly plus” signature ηµν = diag(−1, +1, . . . , +1). Lorentz
indices µ, ν, λ, . . . = 0, 1, · · · , D − 1 will be taken from the middle of the Greek alphabet. Moreover, we
will work in units where c = ~ = 1.

1.1 Point-particle action = worldline length

We describe the propagation of a relativistic point particle through spacetime by a worldline as depicted
dX µ
in figure 3. The latter is parameterized through a “proper time” τ with Ẋ µ ≡ dτ . A possible action
functional to derive the equations of motion is furnished by the length of the worldline2
r
dX µ dX ν
Z
Sold [X] = −m dτ − ηµν (1.1)
dτ dτ
1 Of course, string theory as a theory of quantum gravity is prepared for different backgrounds that can be attained by
“coherent states” built from graviton excitations to be introduced later in this course. We will work with R1,D−1 as a
“reference spacetime” that serves as a starting point for studying small and large fluctuations.
2 As will be explained below, the square-root dependence of S µ is inconvenient for quantization, so we will
old on X
encounter an equivalent action Snew in the next subsections.

9
 Figure 3: Two-dimensional projection of the worldline of a point particle.

It is invariant under monotonic reparametrizations

dτ̃ dX µ dτ̃
τ̃ = τ̃ (τ ) → dτ̃ = dτ , Ẋ µ = (1.2)
dτ dτ̃ dτ
dτ̃
so we have a cancellation of dτ in
r
dX µ dX ν
Z
Sold [X] = −m dτ̃ − ηµν (1.3)
dτ̃ dτ̃
This reparametrization invariance reflects the fact that there is no physical meaning to the choice of the
worldline coordinate τ . As a consequence, one out of D degrees of freedom of X µ is fake: One can for
instance gauge away X 0 by enforcing X 0 (τ̃ ) = τ̃ via suitable choice of τ̃ (τ ).
In the frame with X µ = (t, ~x), we have (with velocity ~v = d~x
dt )
 µ
µ dt dt d~x dt
Ẋ = , = (1, ~v )µ (1.4)
dτ dτ dt dτ
and the action takes the following form
s  
Z 2
dt
Sold [X] = −m dτ − (1, ~v )µ (1, ~v )ν ηµν (1.5)
dτ
Z √
= −m dt 1 − ~v · ~v
Z  m 
= dt −m + ~v 2 + O(~v 4 )
2
For this choice of frame, ~x represent the D−1 dynamical degrees of freedom, while t is merely a parameter.
In the non-relativistic limit |~v |  1 performed in the last step, we recover the well-known expressions
Ekin = 21 m~v 2 and Epot = mc2 in the Lagrangian (recall that we have set c = 1),
Z
Snon−rel = dtLnon−rel , Lnon−rel = Ekin − Epot + O(~v 4 ) (1.6)

As a manifestly Lorentz-invariant way to see that one of the degrees of freedom in X µ (τ ) is unphysical,
we observe that the canonical momenta
Z
∂Lold mẊµ
Sold [X] = dtLold , pµ = =q (1.7)
∂ Ẋ µ
−Ẋ λ Ẋ ρ ηρλ

obey an algebraic relation, the “mass-shell condition”

pµ pµ = −m2 (1.8)

Note that X µ = X µ (τ ) are thought of as “worldline fields” or maps from the worldline to R1,D−1 . To
avoid cluttering, we will henceforth denote Lorentz-index contractions by a dot, i.e. Aµ B ν ηµν = A · B.

10
1.2 Lightning recap of functional differentiation
Since we will frequently need to compute variations of actions, this section aims to briefly review functional
∂xµ
differentiation and fix our conventions. The key idea is to generalize ∂xν = δνµ to the continuum, i.e. to
pass from the discrete choices of differentiation variables xµ with µ = 0, 1, · · · , D − 1 to continuous labels
τ ∈ R in the argument of xµ → f (τ ).
Given that the Dirac δ-distribution is the continuum limit of the Kronecker delta δνµ , we generalize
∂ δ
∂xν to a functional derivative δf (τ 0 ) subject to

δf (τ )
= δ(τ − τ 0 ) (1.9)
δf (τ 0 )
Z b
dτ g(τ )δ(τ − τ 0 ) = g(τ 0 ) , provided that τ 0 ∈ (a, b) (1.10)
a

The functional derivative is taken to commute with ordinary derivatives

δ df (τ ) d δf (τ ) d
= = δ(τ − τ 0 ) (1.11)
δf (τ 0 ) dτ dτ δf (τ 0 ) dτ
where the derivative of the Dirac δ-distribution is defined via integration by parts
Z b Z b
d τ =b dg(τ )
dτ g(τ ) δ(τ − τ 0 ) = g(τ )δ(τ − τ 0 )τ =a − dτ δ(τ − τ 0 ) (1.12)
a dτ a dτ
dg(τ 0 )
=−
dτ 0
 τ =b
The surface term g(τ )δ(τ − τ 0 )τ =a is discarded under the assumption that τ 0 6= a and τ 0 6= b.
When applied to the worldline action, the conventions for the functional derivative feature an addi-
δX µ (τ )
tional Kronecker delta in δX ν (τ 0 ) = δνµ δ(τ − τ 0 )

 
δ
δSold
Z −2 δX µ (τ 0 ) Ẋ ν (τ ) Ẋν (τ )
= −m dτ (1.13)
δX µ (τ 0 )
q
2 −Ẋ 2 (τ )
d
δµν dτ δ(τ − τ 0 )Ẋν (τ )
Z
=m dτ q
−Ẋ 2 (τ )
d Ẋ (τ 0 ) d ∂Lold 
= −m q µ =−
dτ 0 dτ ∂ Ẋ µ τ =τ 0
−Ẋ 2 (τ 0 )
R
More generally, with S[X] = dτ L(τ ), you recover the general form of the Euler-Lagrange equations
that you know from your course on analytical mechanics
 2
δS[X] ∂L d ∂L d ∂L
= − + − ... (1.14)
δX µ (τ ) ∂X µ dτ ∂ Ẋ µ dτ ∂ Ẍ µ
If the Lagrangian depends on higher derivatives beyond X µ , Ẋ µ , Ẍ µ , the . . . stand for an alternating
d n ∂L


sequence of higher derivatives dτ ∂(( d )n X µ )
.
dτ

1.3 Quantization friendly point particle action

p
In the context of path-integral quantization, the square-root form ∼ −Ẋ 2 of the worldsheet action Sold
is highly inconvenient. This can be illustrated by comparison with a quadratic action
Z
Squad [X] = dτ X µ (τ )Dµν X ν (τ ) (1.15)

11
where the linear operator Dµν may involve any number of τ -derivatives: The path integral
Z Z
µ ν
R
D[X]e−Squad [X] = D[X]e− dτ X (τ )Dµν X (τ ) (1.16)

R∞ q
πn
is simply the continuum-version of a Gaussian integral −∞
dn xe−~xD~x = det D . However, the path
integral constructed from the square-root containing worldsheet action
Z Z R √
2
D[X]e−Sold [X] = D[X]em dτ −Ẋ (τ ) (1.17)

R∞ √
already causes major complications in its finite dimensional counterpart −∞
dn xe− ~xA~
x
.
The square-root dependence of Sold [X] can be bypassed by introducing a worldline metric e(τ ) ≡
√
−gτ τ or “einbein” as an extra worldline field. Starting from the new action
Z  
1 1 2 2
Snew [X, e] = dτ Ẋ − em (1.18)
2 e

we can recover the old action Sold by inserting the equation of motion (e.o.m.) of the einbein

δSnew [X, e] 1
= − 2 (Ẋ 2 (τ ) + e2 m2 ) = 0 (1.19)
δe(τ ) 2e
q
into Snew . We denote the solution to δSnew [X,e] 2
δe(τ ) = 0 by e class = − Ẋ
m2 (the subscript referring to
“classical configuration”):
 s 
Ẋ 2 2 
Z
1 1
Snew [X, eclass ] = dτ  q Ẋ 2 − − 2m (1.20)
2 2 m
− Ẋ
m2
Z p
= −m dτ −Ẋ 2 = Sold [X]

δSold [X]
Similarly we can recover the e.o.m. δX µ (τ ) = 0 for X µ from Snew at e = eclass ,

δSnew [X, e] d Ẋ µ δSnew [X, eclass ] d mẊ µ

=− , =− p (1.21)
δXµ (τ ) dτ e δXµ (τ ) dτ −Ẋ 2

Another new feature of Snew is that the mass-shell condition becomes an e.o.m.

∂Lnew Ẋµ
pµ = = (1.22)
∂ Ẋ µ e
X µ Xµ
pµ pµ |eclass = 2 = −m2
eclass

The new action retains the reparametrization invariance of Sold . We shall consider infinitesimal one-
dimensional diffeomorphisms τ → τ̃ = τ − η(τ ) and work to first order in η(τ ). This is treated as an
active coordinate transformation, where the coordinates remain fixed and the value of a field at τ is moved
to τ − η(τ ), i.e. X̃(τ̃ ) = X(τ ). This implies the following infinitesimal transformation of X µ (dropping
higher derivatives of X µ that are O(η 2 ) and using X̃ µ (τ ) = X̃ µ (τ̃ + η))

dX µ
δX µ = X̃ µ (τ ) − X µ (τ ) = η (1.23)
dτ
and the einbein is taken to transform as a density on the worldline

d
δe = (ηe) = η ė + eη̇ (1.24)
dτ

12
These infinitesimal transformations are easily checked to shift the action by a total derivative
Z
1 d η 2 
δSnew [X, e] = dτ Ẋ − ηem2 = 0 (1.25)
2 dτ e
which we drop by the assumption that the Lagrantian at τ → ±∞ falls off. Hence, the new action is
classically equivalent to the old one, but easier to quantize (since its path integral is Gaussian). Moreover,
Snew has a non-trivial massless limit m → 0.

1.4 String action = worldsheet area

We shall generalize the discussion of the point particle to closed strings that are parametrized by a
periodic spatial coordinate σ ∼ σ + 2π. Closed strings sweep out two-dimensional “worldsheets” Σ, see
figure 4, whose embedding in spacetime “target space” is defined by maps X µ : Σ → R1,D−1 with

X µ (τ, σ + 2π) = X µ (τ, σ) (1.26)

Figure 4: Two-dimensional projection of the worldsheet of a closed string.

By analogy with the worldline length as a point-particle action, a natural proposal for a string-
generalization is the worldsheet area. This is known as the Nambu-Goto action
Z
SN G [X] = −T dAM ink (1.27)

In preparation for the area element dAM ink tailored to X µ (τ, σ) in Minkowski spacetime, we first deter-
~
mine the Euclidean area element AEucl (with X(τ, σ) ∈ RD ). As illustrated in figure 5, AEucl is computed
from the area of a parallelogram spanned by the tangent vectors d~`σ = dσ X ~˙
~ 0 and d~`τ = dτ X,

dAEucl = |d~`σ ||d~`τ | sin θ (1.28)

~˙
p
= d2 σ|X ~ 0 ||X| 1 − cos2 θ
q
= d2 σ |X| ~˙ 2 |X ~˙ · X
~ 0 |2 − (X ~ 0 )2
√
~0 =
where d2 σ = dτ dσ as well as X
~
∂X ~˙ =
and X
~
∂X
1 − cos2 θ with
∂σ ∂τ . Moreover, we used that sin θ =
~˙ X
X· ~0
cos θ = ˙ ~0 .
~
Throughout this course, we use the following notation σ α = (τ, σ) for the worldsheet
|X||X |
coordinates with indices α, β, . . . = 0, 1 from the beginning of the Greek alphabet. Then, the argument
of the square root can be related to the Euclidean version of the “induced metric”
∂X µ ∂X ν ∂X~ ∂X ~
γαβ = ηµν , (γαβ )Eucl = · (1.29)
∂σ α ∂σ β ∂σ α ∂σ β

13
 Figure 5: Two-dimensional projection of the worldsheet of a closed string.

More precisely, the Euclidean area element may be recognized as the determinant
p
dAEucl = d2 σ det γEucl (1.30)

which straightforwardly translates into its Minkowski version by inserting a minus sign into the square
root to account for the signature of ηµν ,
p
dAM ink = d2 σ −det γ (1.31)

Hence, the worldsheet area or Nambu-Goto action is given by

Z p
SN G [X] = −T d2 σ −det γ (1.32)
Z q
= −T d2 σ (Ẋ · X 0 )2 − |Ẋ|2 |X 0 |2

The meaning of the normalization constant T can be understood by taking a “snapshot” of the string:
d~
x
In a frame X 0 = τ × const and when the instantaneous kinetic energy is taken to vanish dt = 0, one can
identify T as the string tension,

potential energy
T = = tension (1.33)
spatial length
We will often reexpress the tension via

1
T = (1.34)
2πα0
where α0 = `2string is known as the “Regge slope”.
As will be shown in the homework problems, the e.o.m. of SN G is the reparametrization invariant
∂
wave equation (with shorthand ∂α = ∂σ α )

δSN G [X] p
= T ∂α ( −det γ γ αβ ∂β X µ ) (1.35)
δXµ
In this course, we study fundamental strings without any thickness L → 0. In absence of a thickness
parameter L, there is no chance of adding a rigidity term ∼ L K 2 to SN G , where K is the extrinsic
R

curvature. Other areas of physics involve non-fundamental strings (e.g. QCD, cosmology or magnetic
flux tubes in superconductors). In these cases, SN G [X] is a universal part of the respective worldsheet
action, but additional L-dependent terms specific to the context will appear.

14
1.5 Quantization friendly string action
Similar to the worldline length Sold , the Nambu-Goto action SN G is plagued by a square-root dependence
on the fields X µ (σ) that complicates quantization3 . We shall now study the worldsheet analogue of passing
to a classically equivalent action Sold → Snew which is quadratic in the fields. This is accomplished by
the Polyakov action SP involving an independent worldsheet metric hαβ (σ) that generalizes the einbein
e(τ ) on the worldline (with shorthand h = det h and inverse hαβ hβγ = δγα ):
√
Z
1
SP [X, h] = − 0
d2 σ −hhαβ ∂α X µ ∂β X ν ηµν (1.36)
4πα
From the worldsheet perspective, SP [X, h] describes D scalar fields X µ (σ) coupled to two-dimensional
gravity via hαβ (σ).
One can recover the Nambu-Goto action from SP by inserting the e.o.m. of the metric hαβ
√  
δSP [X, h] −h µ ν 1 γδ
=− ∂α X ∂β X ηµν − hαβ (h ∂γ X · ∂δ X) (1.37)
δhαβ 4πα0 2

δSp [X,hclass ]
The classical configurations hclass solving δhαβ
= 0 are proportional to the induced metric γαβ =
∂X µ ∂X ν 2
∂σ α ∂σ β ηµν , namely hclass
αβ = F (σ)γαβ , with scaling factor F (σ) = hγδ ∂γ X·∂δ X
. When plugging back into
αβ αβ
SP , the factors of F (σ) are found to cancel (the inverse γ yields γ γαβ = 2):

γ αβ γαβ
Z
1 p
SP [X, hclass ] = − 0
d2 σ −F 2 det γ
4πα F
Z
1 p
=− d2 σ −det γ = SN G [X] (1.38)
2πα0
For later reference, we introduce the worldsheet energy momentum tensor

4π δSP [X, h]
Tαβ = √ (1.39)
−h δhαβ
 
1 1
= − 0 ∂α X · ∂β X − hαβ (hγδ ∂γ X · ∂δ X)
α 2
You may recognize this definition from the study of classical field theory in presence of gravity.
The e.o.m. for X µ are given by

δSP [X, h] ∂Lp 1 √

= −∂α = ∂α (hαβ −h∂β Xµ ) (1.40)
δX µ ∂(∂α X µ ) 2πα0
which resemble the e.o.m. from the Nambu-Goto action up to hαβ ↔ γαβ . In fact, the classical configu-
ration hclass
αβ = F (σ)γαβ also maps the e.o.m. from SP and SN G to each other,

δSP [X, h] 1 p γ αβ
µ
= 0
∂α ( −F 2 det γ ∂β Xµ ) (1.41)
δX 2πα F
1 p δSN G
= 0
∂α ( −det γγ αβ ∂β Xµ ) =
2πα δX µ
which is another way of seeing the classical equivalence of SP and SN G found above. At a technical level,
the classical equivalence stems from the fact that the scale factor F (σ) drops out from both the action
and the e.o.m.
The Polyakov action enjoys the following types of symmetries
3 Here and below, the notation X µ (σ) refers to the dependence of X µ on both of σ α = (τ, σ) and does not indicate that
Xµ is independent on τ .

15
 • reparametrization / diffeomophism invariance, σ α → σ̃ α = σ α − η α (σ) with infinitesimal η(σ)

δX µ = η α ∂α X µ (1.42)
δhαβ = η γ ∂γ hαβ + hγα ∂β η γ + hγβ ∂α η γ

which are the infinitesimal versions of

X̃ µ (σ̃) = X µ (σ) (1.43)

γ δ
∂σ ∂σ
h̃αβ (σ̃) = hγδ (σ)
∂ σ̃ α ∂ σ̃ β

• local Weyl invariance (note that this is not a diffeomorphism!)

δX µ (σ) = 0 (1.44)
δhαβ (σ) = 2φ(σ)hαβ (σ)

which is the infinitesimal version of hαβ → Ω2 (σ)hαβ with Ω(σ) = eφ(σ) . The σ-dependent scale
factor Ω2 (σ) drops out in the same way as F (σ) drops out of SP [X, hclass ].

• The action also has a global Poincare invariance X µ → Λµ ν X ν + cµ , where Λ ∈ SO(1, D − 1) are
Lorentz transformations in spacetime. Note, however, that the Lorentz indices µ, ν, . . . are internal
from the worldsheet perspective in the same way as the flavor index of quarks in the Standard
Model is “internal” from the spacetime perspective.

We have seen in the case of the worldline that reparametrization symmetry is crucial in order to identify
the number of d.o.f. in X µ . Similarly, the local symmetries of the worldsheet action are essential for the
later counting of d.o.f. and must be preserved. Hence, one cannot modify SP by extra terms that break
√
diffeomorphism or Weyl invariance. This rules out the addition of a potential SV [X, h] = d2 σ −hV (X)
R
√
or a cosmological constant Sµ [h] = µ d2 σ −h.
R

1.6 Fixing conformal gauge

As we will see, the symmetries of the Polyakov action SP imply that we can locally gauge-fix the metric
to hαβ → ηαβ = diag(−1, 1). As a symmetric “2 × 2 matrix” hαβ = hβα , the worldsheet metric may have
up to three independent entries. The diffeomorphisms σ α → σ̃ α (σ) with α = 0, 1 can be used to locally
fix two of them, e.g. to set the off-diagonal entries to zero and to relate the diagonal ones. This choice is
known as
hαβ (σ) = e2φ(σ) ηαβ conformal gauge (1.45)

which can always be attained locally since the partial differential equations of the required σ(σ̃) locally
have solutions. The conformal factor e2φ(σ) in turn can be removed via Weyl transformation, leading to
the announced Minkowski form hαβ → ηαβ .
Still, the Minkowski form of the worldsheet metric leaves residual gauge freedom: the subclass of
diffeomorphisms that can be undone via Weyl transformation. This residual gauge freedom is analogous
to the fact that Lorenz gauge ∂µ Aµ = 0 in Maxwell electrodynamics is preserved by gauge transformations
Aµ → Aµ + ∂ µ Λ that obey the massless wave equation ∂ 2 Λ = 0.

16
 Conformal gauge brings the Polyakov action into the following form:
Z
1
SP [X, hαβ = e2φ ηαβ ] = − d2 σ∂α X · ∂ α X ≡ SCG [X] (1.46)
4πα0
Its equation of motion is easily seen to be the free wave equation
δSCG [X]
∼ ∂α ∂ α Xµ = (∂σ2 − ∂τ2 )Xµ = 0 (1.47)
δX µ
and string propagation X µ (σ) in spacetime is governed by its solutions.
Even though the e.o.m. for X µ in conformal gauge no longer contains an explicit reference to the
worldsheet metric, the latter contributes an important constraint: We still have to impose the e.o.m. for
hαβ , setting hαβ to ηαβ after the variation.

4π δSP [X, h]
0 = Tαβ = √ |hαβ →ηαβ (1.48)
−h δhαβ
1 1
= − 0 (∂α X · ∂β X − ηαβ ∂γ X∂ γ X)
α 2
The energy-momentum tensor as defined in the previous subsection is easily checked to be traceless,
T α α = 0, so the e.o.m. for hαβ translates into the following two component constraints:
∂X µ ∂Xµ
−α0 T01 = Ẋ · X 0 = =0 (1.49)
∂σ ∂τ
1
−α0 T00 = −α0 T11 = (Ẋ 2 + (X 0 )2 ) = 0
2
The availability of these extra constraints beyond the wave equation is analogous to p2 = −m2 on the
δSnew [X,e]
worldline which follows from δe = 0.

1.7 Mode expansion

We conclude this introductory section by investigating the classical solutions to the free wave equation
and the Fourier modes of the constraints Tαβ = 0. We exploit that, in two dimensions, the wave operator
can be factorized by picking lightcone coordinates:
∂ 1
σ± = τ ± σ , ∂± = = (∂τ ± ∂σ ) , ∂α ∂ α = −4∂+ ∂− (1.50)
∂σ ± 2
Hence, the e.o.m. for X µ in conformal gauge and lightcone coordinates takes the form

∂+ ∂− Xµ = 0 (1.51)

Its most general solution decomposes into two independent functions XL , XR of one variable each:

X µ (σ) = XLµ (σ + ) + XR
µ
(σ − ) (1.52)

For closed strings, we additionally need to impose periodic boundary conditions σ ∼ σ + 2π which leads
to a discrete Fourier expansion for both XL and XR :
r
1 µ α0 µ + α0 X α̃nµ −inσ+
XLµ (σ + )
= x + p σ +i e (1.53)
2 2 2 n
n6=0
r
µ − 1 µ α0 µ − α0 X αnµ −inσ−
XR (σ ) = x + p σ + i e
2 2 2 n
n6=0

17
The first two terms add up to xµ + α0 pµ τ and signal motion at constant momentum pµ . Note that the
periodic boundary conditions do not admit a linear term ∼ σ even though it solves the wave equation. The
α0
factors of 2 reflecting the normalization of the Fourier modes αnµ and α̃nµ are chosen for later convenience.
We emphasize that the inverse string tension α0 must not be confused with the Fourier modes αnµ , α̃nµ .
It remains to impose the constraints Tαβ = 0, using the conversions ∂τ = ∂+ + ∂− and ∂σ = ∂+ − ∂− .
We calculate the independent entries of the worldsheet energy-momentum tensor to be

−α0 T01 = (∂+ X)2 − (∂− X)2 = 0 (1.54)

−α0 T00 = (∂+ X)2 + (∂− X)2 = 0

i.e. (∂+ X)2 and (∂− X)2 have to vanish separately. The Fourier modes of these constraints take a partic-
ularly convenient form if we denote r
α0 µ
α0µ = α̃0µ = p (1.55)
2
i.e. the momentum pµ can be viewed as the joint zero mode of ∂+ X µ and ∂− X µ :
r
µ µ α0 X µ −inσ+
∂+ X = ∂+ XL = α̃n e (1.56)
2
n∈Z
r
µ µ α0 X µ −inσ−
∂− X = ∂− XR = αn e
2
n∈Z

Then, the constrains become

α0 X X + X +
0 = (∂+ X)2 = α̃n · α̃` e−i(n+`)σ = α0 L̃n e−inσ (1.57)
2
n∈Z `∈Z n∈Z
α0 X X + X −
0 = (∂− X)2 = αn · α` e−i(n+`)σ = α0 Ln e−inσ
2
n∈Z `∈Z n∈Z

where the last step defines the Fourier modes of the energy-momentum tensor:

1 X
Ln = αn−m · αm (1.58)
2
m∈Z
1 X
L̃n = α̃n−m · α̃m
2
m∈Z

Hence, our constraints are reorganized in Fourier space and read

Ln = L̃n = 0 ∀ n ∈ Z (1.59)

Note that the zero modes of the constraints already fix the mass spectrum: The constraint

α0 2 X
L0 = p + αn · α−n (1.60)
4 n>0

together with p2 = −m2class yields the following expressions for the classical values of the mass square:

4 X 4 X
m2class = αn · α−n = α̃n · α̃−n (1.61)
α0 n>0 α0 n>0

However, there will be an extra term in the quantum theory, once the Fourier modes αnµ , α̃nµ are promoted
to operators and no longer commute.

18
2 Quantizing closed bosonic string
In this section, we will apply canonical quantization / the operator formulation to the two-dimensional
field theory on the worldsheet defined by the Polyakov action SP [X, h]. As you know from your quantum-
field-theory course, path integrals provide an alternative approach to quantization, and we will discuss
them in part 5 of the course.
Gauge symmetries are delicate in quantization, and there are two ways to handle the diffeomorphism
and Weyl gauge symmetry of the Polyakov action:

• In the next two subsection, we will first do covariant quantization and then use gauge fixing con-
straints, this is analogous to the Gupta-Bleuler quantization in QED.

• From the third subsection onwards, we will first solve the constraints and then quantize the dis-
tinct solutions; this is analogous to QED in Coulomb gauge which obscures Lorentz invariance in
spacetime.

2.1 Opening line in covariant quantization

We define the canonical momentum conjugate to X µ by the derivative w.r.t. τ (as opposed to the spatial
coordinate σ). In conformal gauge hαβ → ηαβ , it reads
∂LCG 1
Πµ (τ, σ) = = Ẋ µ (2.1)
∂ Ẋµ 2πα0
pµ 1 X + −
= + √ (α̃nµ e−inσ + αnµ e−inσ )
2π 2π 2α 0
n6=0

where σ ± = τ ± σ. Since we have singled out ∂τ over ∂σ , Lorentz covariance on the worldsheet is broken
(though Lorentz covariance in target space is preserved). We impose canonical equal-time commutators,

[X µ (τ, σ), Πν (τ, σ 0 )] = iδνµ δ(σ − σ 0 ) (2.2)

[X µ (τ, σ), X ν (τ, σ 0 )] = [Πµ (τ, σ), Πν (τ, σ 0 )] = 0

where δ(σ −σ 0 ) is the continuum extension of the Kronecker-delta familiar from [xµ , pν ] = iδνµ in quantum
mechanics. As will be shown in the exercises, these canonical commutators are equivalent to the following
commutation relations among the operators in the mode expansion:

[xµ , pν ] = iδνµ (2.3)

[xµ , xν ] = [pµ , pν ] = 0
[α̃nµ , α̃m
ν
] = [αnµ , αm
ν
] = nδn+m η µν
[αnµ , α̃m
ν
]=0

Both {αnµ , n ∈ Z} and {α̃nµ , n ∈ Z} comprise infinitely many copies of the algebra [a, a† ] = 1 of the
quantum mechanical (QM) harmonic oscillator. If we define
ν
αµ α−n
aµn ≡ √n , a†ν
n ≡ √ , n>0 (2.4)
n n
then we have the normalization as in the quantum-mechanics course

[aµn , a†ν µν
m ] = η δn,m , [aµn , aνm ] = [a†µ †ν
n , am ] = 0 , n, m > 0 (2.5)

19
The Hamiltonian of the QM harmonic oscillator is

1 † 1
HQM = (a a + aa† ) = a† a + (2.6)
2 2
where we normal-ordered the a and a† to have the lowering operator a on the right and the raising
operator a† on the left. The term + 12 is then identified as the zero-point energy, and the terminology of
lowering / raising operators is motivated by

[HQM , a† ] = a† , [HQM , a] = −a (2.7)

We can build a Fock space |0i, a† |0i, a† a† |0i, . . . from some vacuum state |0i, and the requirement that
the spectrum of the Hamiltonian is bounded from below requires a|0i = 0.
In the string-theory scenario with infinitely many pairs aµn , a†ν µ µ
n , we take αn ∼ an with n > 0 as

the lowering operators which annihilate the vacuum. The Hamiltonian follows from tracing out both
the discrete labels µ = 0, 1, . . . , D − 1 and the continuous one σ ∈ (0, 2π) (see the exercises for some
intermediate steps)
Z 2π
dσ Πµ (τ, σ)Ẋ µ (τ, σ) − LCG (τ, σ)


H= (2.8)
0
Z 2π
1
dσ Ẋ 2 + (X 0 )2 = . . .


= 0
4πα 0
α0 2 1 X

= p + α−n · αn + αn · α−n + (α ↔ α̃)
2 2 n>0

where we use normal ordering αn · α−n = α−n · αn + D · n (the D coming from ηµν η µν )
∞ ∞
α0 2 X X
H= p + (α−n · αn + α̃−n · α̃n ) + D n (2.9)
2 n=1 n=1

µ µ
Formally, this Hamiltonian has infinite zero-point energy. The role of αn<0 , α̃n<0 as raising operators
µ µ
and αn>0 , α̃n>0 as lowering operators is confirmed by

[H, αnµ ] = −nαnµ , [H, α̃nµ ] = −nα̃nµ (2.10)

What goes beyond the QM analogy: There are additional xµ , pν operators in the mode expansion of
X µ (τ, σ). In contrast to the QM situation, they are not a combination of a’s and a† ’s. We take the
vacuum |0; pi w.r.t. {αn , α̃n , n > 0} to be a pµ eigenstate with the normalization

hp0 ; 0|0; pi = δ D (p − p0 ) (2.11)

Then the lowering operators act as usual,

µ µ
αn>0 |0; pi = α̃n>0 |0; pi = 0 (2.12)

µ µ
We can build the Fock space of physical states by raising operators αn<0 , α̃n<0 ,
 µ1 µ2 ν1 λ1 λ2 ρ1
|physi ∈ α−1 α−1 . . . α−2 . . . α̃−1 α̃−1 . . . α̃−2 . . . |0; pi (2.13)

From the free Lorentz indices µi , νi , . . ., one can identify the spectrum as an infinite tower of higher-spin
excitations.

20
2.2 Constraints in covariant quantization
The discussion in the previous section leaves two loose ends up to now:

• How to implement the classical vanishing of the worldsheet energy momentum tensor
 
1 1
Tαβ = − 0 ∂α X · ∂β X − ηαβ η γδ ∂γ X · ∂δ X (2.14)
α 2
in the quantum theory?

• The indefinite signature of Minkowski spacetime R1,D−1 gives rise to negative-norm states: By
µ ρ
(αnµ )† = α−n and η µν = diag(−1, +1, . . . , +1), the collection of states {α−1
λ
α̃−1 |0; pi} contains
negative-norm representatives:

hp0 ; 0|α1µ α̃1ν α−1

λ ρ
α̃−1 |0; pi = η µλ η νρ δ D (p − p0 ) (2.15)

If these negative-norm states are physical, then unitarity of the theory is violated.

In quantization of QED, the negative-norm states are removed by imposing the gauge-fixing constraints.
In our string-theory setup, the removal of negative-norm states will ultimately come from the operator-
valued version of the energy-momentum constraints Tαβ = 0. We shall pass to operator-valued Fourier
modes Ln , L̃n in σ ± = τ ± σ coordinates
1 X +
T++ = − 0
(∂+ X)2 = − L̃n e−inσ (2.16)
α
n∈Z
1 X −
T−− = − 0 (∂− X)2 = − Ln e−inσ
α
n∈Z

All the Ln , L̃n with n 6= 0 are unambiguous,

1X
Ln6=0 = αn−` · α` (2.17)
2
`∈Z
1X
L̃n6=0 = α̃n−` · α̃`
2
`∈Z

since the composing operators αn−` , α` and α̃n−` , α̃` commute in these cases. For L0 , L̃0 , in turn, we
have to pick an ordering convention and shall now choose normal ordering (lowering operators on the
right)
∞
α02 X
L0 = + α−n · αn (2.18)
2 n=1
∞
α̃02 X
L̃0 = + α̃−n · α̃n
2 n=1

where
α02 α̃2 α0 α0
= 0 = p2 = − m2 (2.19)
2 2 4 4
As a quantum version of the classical vanishing of Tαβ , one could consider the scenario where all the
Ln , L̃n with n ∈ Z annihilate physical states. However, it will be sufficient to impose

Ln>0 |physi = L̃n>0 |physi = 0 (2.20)

21
since, by (Ln )† = L−n , this ensures that the matrix elements of {Ln , L̃n , n ∈ Z \ {0}} w.r.t. physical
states |physi and |phys0 i vanish

hphys0 |Ln |physi = 0 = hphys0 |L̃n |physi , n 6= 0 (2.21)

In the simple example of the vacuum |physi → |0; pi, this is a consequence of the defining property of
annihilation operators

µ µ
αn>0 |0; pi = α̃n>0 |0; pi = 0 (2.22)

The above normal-ordering conventions for the operators L0 and L̃0 were an ad-hoc choice. Hence, we
should allow for some scalar offset a ∈ C when we impose the constraint

(L0 − a)|physi = (L̃0 − a)|physi = 0 (2.23)

that corresponds to the zero mode of the energy-momentum tensor. It is instructive to see how the
normal-ordering constant a affects the mass spectrum.
∞ ∞
! !
4 X 4 X
m2 |physi = 0 −a + α−n · αn |physi = 0 −a + α̃−n · α̃n |physi (2.24)
α n=1
α n=1

Our choice to pick the same offset a for both L0 and L̃0 implies the level-matching condition (L0 −
L̃0 )|physi = 0. In other words, physical closed-string states have the same oscillation numbers w.r.t. αnµ
and α̃nµ . In the next subsections, we will settle a by Lorentz-invariance at quantum level.

2.3 Lightcone gauge

Now we are going to switch gears and impose Tαβ constraints before quantization. The starting point
is to recall the residual diffeo×Weyl invariance that keeps the worldsheet metric hαβ (σ ± ) in Minkowski
form ηαβ ,

certain diffeo’s +
,σ − ) Weyl
hαβ (σ ± ) = ηαβ −→ e2φ(σ ηαβ −→ ηαβ (2.25)

In lightcone coordinates where ηαβ dσ α dσ β = −dσ + dσ − , the residual gauge invariance consists of two
copies of one-variable diffeomorphisms σ + → σ̃ + (σ + ) and independently σ − → σ̃ − (σ − ), where we admit
no σ − dependence to σ̃ + (σ + ). When compared with the set of two-variable diffeomorphisms σ̃ α (σ β ), the
residual gauge freedom is a set of measure zero! It affects the components of Tαβ via

∂ ∂σ +
∂+ X µ → +
Xµ = ∂+ X µ (2.26)
∂ σ̃ ∂ σ̃ +
For one given spacetime direction µ, one can independently enforce ∂+ X µ and ∂− X µ to match an
∼ σ + 2π: One simply has to dial σ + → σ̃ + (σ + ) such that the
arbitrary function with periodicity σ =
∂σ + desired expression for ∂+ X µ
Jacobian compensates the deviation from the desired expression, i.e.: ∂ σ̃ + = old version of ∂+ X µ .
However, this can only be done for a single spacetime direction, and this choice of µ direction breaks
the manifest SO(1, D−1) Lorentz invariance. We will do so by picking the following lightcone directions
in the spacetime
r
± 1 0
X = (X ± X D−1 ) (2.27)
2

22
This is analogous to the lightcone coordinates σ ± = τ ± σ on the worldsheet, except for the fact that
spacetime has D−2 leftover or “transversal” directions (whose choice is an ad-hoc convention). We
hope that the following discussion does not cause any confusion between the ± super- and subscripts for
spacetime and worldsheet quantities (X ± versus σ ± ).
Now we exhaust the residual gauge symmetry σ ± → σ̃ ± (σ ± ) to fix the “lightcone gauge”

α0 +
∂+ X + = ∂− X + = p (2.28)
2
i.e. to enforce both ∂+ X + and ∂− X + to be the same constant. (Here, the superscript refers to the
1
lightcone in spacetime, while the subscripts refer to the worldsheet lightcone 2 (∂τ ± ∂σ )). For the
undifferentiated XL+ and +
XR , +
this still leaves the center-of-mass position x as an integration constant

1 + α0 + +
XL+ = x + p σ (2.29)
2 2
+ 1 α0 + −
XR = x+ + p σ
2 2
Let us now consider the classical worldsheet energy-momentum constraints
D−2
X
0 = ∂± X · ∂± X = −2∂± X + ∂± X − + ∂± X i ∂± X i (2.30)
i=1

With the above gauge-fixing of X + , one obtains a simple expression when solving for the X − component
D−2
1 X
∂+ XL− = 0 + ∂+ X i ∂+ X i (2.31)
α p i=1
D−2
− 1 X
∂− XR = ∂− X i ∂− X i
α0 p+ i=1

where p+ 6= 0. With the usual mode expansion for the directions X i=1,2,...,D−2 outside the lightcone,
 
D−2
X α0
1 X
α0 p− = +  pi pi + αni α−n
i 
(2.32)
p i=1 2
n6=0

X D−2
r
1 1 X
αn− = i
αn−k αki (2.33)
2α0 p+
k∈Z i=1

while x− is left unconstrained (it does not enter the quantities ∂± X − in the classical constraint). The
same equations hold with αn ↔ α̃n interchanged. This leads to the following lightcone formulation of
the classical mass spectrum:
D−2 D−2 ∞
X 4 XX i
m2class = 2p+ p− − pi pi = α αi (2.34)
i=1
α0 i=1 n=1 −n n
D−2 ∞
4 XX i
= α̃ α̃i
α0 i=1 n=1 −n n

2.4 Lightcone quantization

In a classical analysis, residual gauge freedom has been used to fix p− , αn− and α̃n− in terms of p+ and
pi , αni , α̃ni with i = 1, 2, . . . , D − 2. When shall now perform the quantization only for the independent

23
variables. By following the steps in the manifestly Lorentz-covariant quantization

[xi , pj ] = iδ ij (2.35)
[x− , p+ ] = −i
[αni , αkj ] = [α̃ni , α̃kj ] = nδ ij δn+k,0

This time, we have 2(D − 2) instead of 2D towers of harmonic oscillators!

We still take the ground state |0; pi w.r.t. the dynamical oscillators {αni , α̃ni , i = 1, 2, . . . , D−2} to be
an eigenstate of pµ = (p+ , p− , pi ) subject to

i i
αn>0 |0; pi = α̃n>0 |0; pi = 0 (2.36)

One can still impose [x+ , p− ] = −i, and we have the following structure for the spectrum of physical
states

i1 i2 j1 k1 k2 l1
{|physi} ∈ {α−1 α−1 . . . α−2 . . . α̃−1 α̃−1 . . . α̃−2 . . . |0; pi} (2.37)

Here, the matrix of inner products is positive definite since η ij = δ ij ≥ 0 for i, j = 1, 2, . . . , D − 2, so we

have successfully removed the negative-norm states!
As in covariant quantization, there are ordering issues in p− and m2 , so we need to once more prepare
ourselves to an offset a ∈ C in the mass spectrum
∞
D−2 D−2
!
2 + −
X
i i 4 XX
i
m = 2p p − pp = 0 α−n αni −a (2.38)
i=1
α i=1 n=1
∞
D−2
!
4 XX
i
= 0 α̃−n α̃ni −a
α i=1 n=1

where p− contains αni α−n

i
P
n6=0 . The level-matching condition is
D−2
XX ∞
i
(α−n αni − α̃−n
i
α̃ni ) = 0 (2.39)
i=1 n=1

i1 i2 j1 k1 k2 l1
which is the only constraint on the oscillator numbers of α−1 α−1 . . . α−2 . . . α̃−1 α̃−1 . . . α̃−2 . . . in the
above schematic expression for physical states. We shall now discuss two approaches to fixing the normal-
ordering constant a:

2.4.1 The physicist’s way

The physicist’s way of fixing a is to use Lorentz invariance, i.e. to impose that lightcone quantization
does not introduce any violations of Lorentz invariance. We first note that the ground state |0; pi is a
j
Lorentz scalar with m2 = − 4a i
α0 . The first excited states α−1 α̃−1 |0; pi admitted by level matching obey
4
m2 = α0 (1 − a). There are (D−2)2 such states, and we will explain that this counting and Lorentz
invariance require a=1.

• We shall recall some basics on the representations of the Poincaré group (Lorentz transformations
and translations). If m2 6= 0, we can find a reference frame pµ = (m, 0, . . . , 0) via Lorentz rotations,
so massive particles form representations of the little group SO(D − 1) that stabilizes this choice of
pµ . When m2 = 0 in turn, the best we can find is pµ = (E, 0, . . . , 0, E). Hence, massless particles
form representations of the little group SO(D − 2) stabilizing the latter choice of pµ .

24
 j
• The (D − 2)2 states α−1
i
α̃−1 |0; pi only fit into representations of SO(D − 2), not into those of
SO(D − 1). This follows from the fact that the candidate representations of SO(D − 1) have
dimensions 12 D(D−1) − 1 (symmetric and traceless two-tensor), 21 (D−1)(D−2) (antisymmetric two
tensor) and 1 (trace). Thus, the state must be massless to ensure that the relevant little group is
SO(D − 2),

j
m2 α−1
i
α̃−1 |0; pi ∼ (1 − a) = 0 (2.40)

i j
• The massless states α−1 α̃−1 |0; pi do not yet form an irreducible representation of the Lorentz group.
We can disentangle the trace and the antisymmetric part, and find three irreducibles (i.e. particle
species) at the massless level of the bosonic string

i j (i j) 1 ij k k [i j] 1 ij k k
α−1 α̃−1 = α−1 α̃−1 − δ α−1 α̃−1 ⊕ α−1 α̃−1 ⊕ δ α−1 α̃−1
D−2 D−2
| {z } | {z } | {z }
“B-field”
massless spin 2 ⇒ graviton dilaton

As will be discussed later on, the vacuum expectation value of the dilaton will set the string coupling
constants which weights different orders in string perturbation theory.

4 4
• For the above choice a = 1, the ground state has negative m2 = α0 (0 − 1) = − α0 < 0. States
2
2 ∂ V (T )
with this property are known as tachyons T . We interpret m = ∂T2 |T =0 < 0 as the curvature
of a scalar potential, and its negative sign indicates that T is a fluctuation for expanding around
the maximum rather than the minimum of the potential. This is analogous to the Higgs H whose
mexican-hat potential has both minima and maxima, and the positive mass square that has been
measured at LHC stems from expansion around a minimum.

Figure 6: The negative m2 of the tachyon is interpreted as expanding a scalar potential V (T ) around its
maximum.

j
• By virtue of a = 1, we have m2 > 0 for any higher excitation beyond α−1
i
α̃−1 |0; pi, starting with

i i j k k l
(α−2 ⊕ α−1 α−1 ) ⊗ (α̃−2 ⊕ α̃−1 α̃−1 )|0; pi (2.41)

i i j k k l
The number of independent components for both α−2 ⊕ α−1 α−1 and α̃−2 ⊕ α̃−1 α̃−1 is easily seen
to be D − 2 + 21 (D − 2)(D − 1) = 21 (D − 1)D − 1 which is the dimension of the traceless symmetric
two tensor of SO(D − 1). Hence, the lightest string excitations with positive m2 fall into the tensor
product of two massive spin-two representations of SO(D − 1).

25
2.4.2 The mathematician’s way

As a mathematician’s way of fixing the normal-ordering constant a, we rearrange

D−2  ∞ 
1 X X
α 0 p− = α 0 i i
p p + 2 (α i
α
−n n
i
+ α i i
α
n −n ) (2.42)
2p+ i=1 n=1
| {z }
αi−n αin +n
D−2 ∞ D−2
 X  ∞ 
X 4 X D−2 X
m2 = 2p+ p− − pi pi = i
α−n αni + n
i=1
α0 n=1 i=1
2 n=1
P∞ P∞ 1
and run into the formally divergent sum n=1 n. The latter can be assigned a finite value n=1 n = − 12 ,
either via zeta-function regularization (to be done in the exercises) or via the following renormalization
trick: The formally divergent sum in question is obtained by setting  = 0 on the left-hand side of
∞ ∞
X ∂ X −n e− 1 1 2
ne−n = − e = = − + + O(4 ) (2.43)
n=1
∂ n=1 (1 − e− )2 2 12 240

1 1
We discard the divergent term 2 and take the constant piece − 12 in the Laurent expansion as the answer.
Although it might appear highly disturbing that the sum over positive integers evaluates to a negative
fractional number, there will be another physics confirmation of the resulting normal-ordering constant
∞
D−2 X D−2
a=− n= (2.44)
2 n=1 24

i j
This is consistent with a = 1 as required by massless states α−1 α̃−1 |0; pi if spacetime has D = 26
dimensions.

2.4.3 From the Lorentz algebra to a = 1 and D = 26

Another perspective on the above results a = 1 and D = 26 can be obtained from the conserved charges
for the Lorentz symmetry of the Polyakov action (see exercise A.4)
∞
X 1 ν µ
J µν = pµ xν − pν xµ + i (αn α−n − αnµ α−n
ν
) (2.45)
n=1
n

They should obey the Lorentz algebra that you know well from your quantum-field-theory course

[J µν , J λρ ] = η νλ J µρ − η νρ J µλ − η µλ J νρ + η µρ J νλ (2.46)

However, we have given up manifest Lorentz invariance in lightcone quantization, so with our operator
relations for p− , αn− , α̃n− , it is non-trivial to check
∞
2 X j
0 = [J i,− , J j,− ] =
 i
α−n αnj − α−n αni + (α ↔ α̃)

(2.47)
(p+ )2 n=1
    
D−2 D−2 1
× −1 n+ a−
24 24 n

At n = 1, the {. . .} parenthesis equals a − 1, and we recover the conclusion a = 1 of our discussion of

j
i
the α−1 α̃−1 |0; pi states. From additionally considering n 6= 1, we obtain D−2
24 = 1, in lines with the
P∞ 1
above n=1 n = − 12 . In conclusion, we have seen in multiple ways that consistency of bosonic strings
in lightcone quantization requires a = 1 and D = 26.

26
3 Open bosonic strings
We will now adapt the discussion of closed bosonic strings to open ones. The highlight in this first look
at open strings will be:

• (non)-abelian gauge bosons form m2 = 0 states;

• geometric implementation of the Higgs mechanism;

• D branes as dynamical degree of freedom in string theories.

3.1 Dirichlet and Neumann boundary conditions

The worldsheets Σ of open strings have a non-periodic spatial direction whose coordinates will be nor-
malized to live in the interval σ ∈ (0, π). We need to revisit the variation of the Polyakov action (let’s
use conformal gauge to keep the discussion simple) since it was the periodicity σ ∼ = σ + 2π specific to
closed strings that led to the cancellation of boundary terms.

Figure 7: Examples of the outward pointing normal vector dnα for the worldsheet of open strings.

R
For open strings, by contrast, there is a potential boundary term ∼ ∂Σ
when integrating by parts in
the variation of the Polyakov action in conformal gauge (see the figure for the normal vector dnα )
Z
1
δSCG [X] = d2 σ ∂α (δX µ )∂ α Xµ (3.1)
2πα0 Σ
Z Z 
1 2 µ α µ α
= d σ δX ∂α ∂ Xµ − dnα δX ∂ Xµ
2πα0
|Σ {z } | ∂Σ {z }
see closed strings new for open strings

The first term has already been discussed for closed strings, so it remains to look at the second term.
Z Z π Z τf
µ α µ
τ =τf σ=π
dnα δX ∂ Xµ = − dσ δX ∂τ Xµ τ =τi +
 dτ δX µ ∂σ Xµ σ=0 (3.2)
∂Σ
|0 {z } τi

vanishes by δX µ (τf )=δX µ (τi )=0

Since we assume the initial and final configurations to be specified, one can set δX µ (τf ) = δX µ (τi ) = 0,
R
and only the second term ∼ dτ needs additional input to vanish: At both endpoints σ = 0 and σ = π,
there are two choices of boundary conditions to make the product δX µ ∂σ Xµ vanish – either δX µ or ∂σ Xµ
has to vanish:

27
 • Neumann conditions

∂σ X µ σ=0,π = 0

(3.3)

for some 0 ≤ µ ≤ D − 1, this corresponds to open-string endpoints moving freely.

• Dirichlet conditions

δX I σ=0,σ=π = 0

(3.4)

for some 1 ≤ I ≤ D − 1. This leads to a specified position cI , dI in space that is constant over τ ,

X I (τ, σ = 0) = cI , X I (τ, σ = π) = dI (3.5)

We do not admit I = 0 to be part of the Dirichlet directions as that would localize the open-string
endpoints in time.

The mode expansions for open strings are found by solving ∂α ∂ α X µ = 0 with Dirichlet / Neumann
conditions instead of the periodicity σ ∼ µ µ
= σ+2π. We make the usual ansatz, X µ (τ, σ) = XL (σ + )+XR (σ − )
r
+ 1 µ 0 µ + α0 X 1 µ −inσ+
XL (σ ) = x + α pL σ + i α̃ e (3.6)
2 2 n n
n6=0
r
− 1 µ 0 µ − α0 X 1 µ −inσ−
XR (σ ) = x + α pR σ + i α e
2 2 n n
n6=0

and will next see how the boundary conditions imply constraints on xµ , pµL , pµR , αnµ , α̃nµ

• Neumann conditions “NN” at both endpoints σ = 0 and σ = π:

0 = ∂σ X µ (τ, σ) σ=0,π = ∂+ XLµ (σ + ) − ∂− XR

µ
(σ − ) σ=0,π
 
(3.7)
r
0 µ µ α0 X µ
(α̃n − αnµ )e−inτ × (−1)
 1 : σ=0
= α (pL − pR ) + n
: σ=π
2
n6=0

This is the case for any τ if we choose pµL = pµR and α̃nµ = αnµ for µ in Neumann directions.
r
µ
 µ 0 µ α0 X 1 µ −inτ
X (τ, σ) NN = x + 2α p τ + 2i
 α e cos(nσ) (3.8)
2 n n
n6=0

Note that the closed-string normalization of the second term ∼ pµ τ is α0 instead of 2α0 . This
1
Rπ
modified normalization for open strings is to ensure that pµ = 2πα 0 0
dσ∂τ X µ can be identified
with the Poincaré generator of translations studied in the exercises.

• Dirichlet conditions “DD” at both endpoints σ = 0 and σ = π:

cI = X I (τ, σ = 0) , dI = X I (τ, σ = π) (3.9)

This gives, in Dirichlet direction (with index I instead of µ),

dI − cI
x I = cI , pIL = −pIR = , α̃nI = −αnI (3.10)
2πα0
The solution is
r
σ α0 X 1 I −inτ
X I (τ, σ) DD = cI + (dI − cI ) − 2

α e sin(nσ) (3.11)
π 2 n n
n6=0

and does not feature any momentum pI .

28
The bottom line in both cases is that all the α̃nµ are fixed by αnµ , i.e. open strings only have half of the
oscillator degrees of freedom! As will be explored in the exercises, one could also have mixed boundary
conditions, say Neumann at σ = 0 and Dirichlet at σ = π.

3.2 Open-string quantization and mass spectrum

Let us now apply lightcone quantization to open strings. We impose Neummann boundary conditions at
σ = 0, π for the range of spacetime directions with indices 0 ≤ µ ≤ dN −1, and Dirichlet conditions at
σ = 0, π for dN ≤ I ≤ D−1. Hence, there is a total of dN Neumann directions and D − dN Dirichlet
directions. This choice breaks the Lorentz group from SO(1, D−1) to SO(1, dN −1) × SO(D−dN ).
We pick lightcone gauge in Neumann directions
r
± 1 0
X = (X ± X dN −1 ) (3.12)
2
and again use the residual gauge freedom to fix ∂± X + = α0 p+ . Then, the classical energy-momentum
constraints are solved via
N −2
dX D−1
!
− 1 i i
X
I I
∂± X = ∂± X ∂± X + ∂± X ∂± X (3.13)
2α0 p+ i=1 I=dN

with dN −2 transverse Neumann directions and D−dN transverse Dirichlet directions. The left-hand side
evaluates to α0 p− plus oscillators. On the right-hand side, we obtain α0 pi pi plus oscillators from the first
PD−1
term (Neumann directions) and 4π1 2 I=dN (cI − dI )2 plus oscillators from the second term (Dirichlet
directions).
The dynamical degrees of freedom are

• xi , pi , αni in transverse Neumann direction 1 ≤ i ≤ dN − 2;

• αnI in Dirichlet direction dN ≤ I ≤ D − 1

The commutation relations are the same as for closed strings:

[xi , pj ] = iδ ij (3.14)
[αni , αkj ] = nδ ij δn+k,0
[αnI , αkJ ] = nδ IJ δn+k,0
[αni , αkJ ] = 0

i,I
The Fock space of physical states is built on eigenstates |0; pi of pµ=0,1,...,dN −1 subject to αn>0 |0; pi = 0

i1 i2 j1 I1 I2 J1
{|physopen i} = {α−1 α−1 . . . α−2 . . . α−1 α−1 . . . α−2 . . . |0, pi} (3.15)

The mass spectrum is (note the rescaling α0 → 4α0 in comparison to closed strings)

N −2
dX
m2 = 2p+ p− − pi pi (3.16)
i=1
∞ N −2
D−1
( dX D−1
! )
1 X 1 X X
= 0 2
(cI − dI )2 + 0 i
α−n αni + I
α−n αnI −a
(2πα ) α n=1 i=1
I=dN I=dN

29
with the same type of normal-ordering constant a as for closed strings in both Dirichlet and Neumann
directions:
∞
1 X D−2
a = − (D − 2) n= (3.17)
2 n=1
24

Again, Lorentz invariance in the Neumann directions including X ± will give a = 1 and D = 26. Since
these are the same values found in the discussion of closed strings, we can view open- and closed-string
excitations as states in the same theory.
Let us take a brief look at the lowest mass levels (i.e. the states with the lowest oscillator numbers n
i I
of α−n and α−n ):

• level zero: a tachyonic ground state m2 |0; pi = − α10 |0; pi

i I
• level one: massless states α−1 |0; pi and α−1 |0; pi. The former forms the vector of the little group
SO(dN − 2) for m2 = 0 states, so the dN − 2 states α−1
i
|0; pi are interpreted as gauge bosons. The
I
D −dN states α−1 |0; pi in turn are scalars with respect to the Lorentz group in Neumann directions.

• level ≥ 2: excited states of the open string have m2 ∈ N

α0 ,
i
starting with α−2 J
|0; pi⊕α−2 |0; pi as well as
i j i J I J
α−1 α−1 |0; pi⊕α−1 α−1 |0; pi⊕α−1 α−1 |0; pi. As for a chiral half of the analogous closed-string states,
i i j
the open-string states α−2 |0; pi ⊕ α−1 α−1 |0; pi can be recombined to a spin-two representation of
i I
the massive little group SO(dN − 1). The remaining states obtained from replacing α−n → α−n
are massive scalars and vectors.

3.3 D branes and non-abelian gauge fields

The open-string endpoints X I (τ, σ=0) = cI and X I (τ, σ=π) = dI fixed by Dirichlet boundary condi-
tions define hypersurfaces in spacetime. However, since one may ask what is special about these choices
of cI , dI , we shall now interpret these hypersurfaces as dynamical objects called D-branes (with D for
“Dirichlet”). More precisely, the number dN of spacetime directions with Neumann boundary conditions
is usually specified, and we speak of a Dp branes with p = dN −1. In these conventions, a D0 brane
corresponds to a particle, a D1 brane to a fundamental string and a D2 brane to a membrane. With
p = D − 1 or dN = D (i.e. no Dirichlet boundary conditions), we get spacetime filling branes.
Let us revisit the interpretation of massless open-string states in the light of D-branes:
i
• vector α−1 |0; pi ↔ gauge field on the brane

I
• scalars α−1 |0; pi ↔ fluctuations of the brane in the transverse direction (→ hint for dynamics)

30
 1
However, the dynamics of D-brane is hard to see perturbatively since Mbrane ∼ gstring .
In case of two separated branes cI 6= dI , the open-string states become 2 × 2 matrices: Depend-
ing on the independent choices to locate the two endpoints on one of the branes, we assign labels
(1, 1), (1, 2), (2, 1), (2, 2) to these open-string states. The off-diagonal states having endpoints on different
D-branes acquire the mass shift
D−1
1 X
∆(m2 ) = 0 2
(cI − dI )2 (3.18)
(2πα )
I=p+1

found in the previous section.

If we have a stack of N coinciding branes, both endpoints can be labelled by m, n = 1, 2, . . . , N to

specify the brane they are attached to. This yields the degrees of freedom of non-abelian gauge bosons
of U (N ):
2
N
X
|physopen i → |physopen ; (m, n)i = (T a )m n |physopen ; ai (3.19)
a=1

where the coefficients are a spanning set of hermitian N ×N matrices T a , known as “Chan Paton factors”.

The D-brane setup admits a geometric implementation of the Higgs effect: By separating initially co-
inciding branes, the U (2) gauge group is broken to U (1) × U (1). From the previous discussion, the
off-diagonal gauge fields gain the mass ∆(m2 ) ∼ (cI − dI )2 , and the diagonal scalars gain a VEV
 
I
α−1 |0; p; (m, n)i ↔ φ011 φ022 . However, in contrast to the Standard Model Higgs, we are dealing with
adjoint rather than fundamental scalars here.
The D-brane setup can be used to make progress towards the Standard Model gauge group SU (3)strong ×
SU (2)weak × U (1)Y : Take 4 stacks of D-branes U (3)a × U (2)b × U (1)c × U (1)d (the 4th one is needed

31
in order to balance certain charges). Gauge bosons for one of the gauge groups stem from open strings
with both endpoints on the same stack. Chiral fermions such as leptons/quarks arise from open strings
located at intersecting stacks of branes, they are in bi-fundamental (rather than adjoint) representations
of the two associated gauge groups.

D-branes can be wrapped around cycles in compact spacetime dimensions. The number of quark- and
lepton generations is then determined by the intersection numbers of different cycles.

4 Basics of conformal field theory

This section provides an introduction to conformal field theories with the following motivation and goals:

• sidestep the complicated commutator manipulations in the quantized bosonic string such as the
lengthy and subtle computation of [Lm , Ln ] ↔ [αm−p αp , αn−q αq ]; instead infer such brackets from
clean complex-analysis methods, starting with Cauchy’s theorem B (0) z n = δn,−1 for a small circle
H

B (0) around the origin

• characterize & exploit the residual Diff × Weyl invariance σ + → f (σ + ) and σ − → g(σ − ) as an
infinite-dimensional symmetry; use it for covariant quantization and computing string interactions.

4.1 Conformal symmetry in general dimensions d

Although we will be mainly interested in applications to the conformally invariant worldsheet theory with
d = 2 dimensions, we will discuss some aspects of d-dimensional conformal symmetry that for instance
apply to Maxwell theory and Yang Mills in d = 4. Conformal transformations in R1,d−1 are defined as

32
those diffeomorphisms that preserve angles, i.e. rescale metric under xµ → y µ (x)

∂xλ ∂xρ
ηµν → gµν = ηλρ = Ω2 (x)ηµν (4.1)
∂y µ ∂y ν
with some possibly x-dependent scale factor (or conformal factor) Ω(x). Examples of d-dimensional
conformal transformations are

Example y µ (x) Ω2 (x)

translation x µ + cµ 1
µ ν (4.2)
Lorentz transformation Λ νx , Λ ∈ SO(1, d−1) 1
µ 1
dilatation λx λ2
xµ +x2 bµ
special conformal transformation 1+2b·x+b2 x2 (1 + 2b · x + b2 x2 )2

where special conformal transformations are translations by bµ sandwiched between two inversions

xµ
xµ → , Ω2 (x) = (x2 )2 (4.3)
x2
The inversion is disconnected from unity and therefore does not enter the list of infinitesimal conformal
transformations y µ = xµ +ω µ (x). In order to determine the conformal algebra, evaluate the group action
on F (xµ ) via the inverse of infinitesimal conformal transformations, F (xµ ) → F (xµ − ω µ (x)) + O(2 )

infinitesimal diffeo ω µ action on F (x) generator

µ µ
c 1 − ic Pµ Pµ = −i∂µ
i
m[µν] xν 1− µν
2 m Jµν Jµν = −i(xν ∂µ − xµ ∂ν ) (4.4)
λxµ 1 − iλD D = −ixµ ∂µ
x2 bµ − 2(x · b)xµ 1 − ibµ Kµ Kµ = −ix2 ∂µ + 2ixµ x · ∂

1
The conformal group in R1,d−1 is 2 (d + 1)(d + 2)-dimensional, and by comparing the commutators of
span{Pµ , Jµν , D, Kµ } to higher-dimensional Lorentz brackets, the conformal algebra turns out to be
isomorphic to SO(2, d). The conserved charges can be constructed from the energy-momentum tensor,

4π δS[g, . . .]
T µν = √ |gλρ =ηλρ (4.5)
− det g δgµν

using the fact that the infinitesimal rescaling of the metric is δgµν = −(∂µ ων + ∂ν ωµ ),
Z
1 p
δS[g, . . .] = dd x − det g T µν δgµν (4.6)
4π
Z
µ
= dd x ∂µ J(ω) (x)

µ
In the last line, we have rewritten the variation of the action as a divergence of the Noether current J(ω)
for y µ = xµ + ω µ . Assuming the consequences T µν = T νµ and ∂µ T µν = 0 of Poincaré symmetry, the
general form of these currents is
µ
J(ω) = T µν ων (4.7)

In particular, the vanishing divergence for the dilatation current with ων = λxν implies that

µ
∂µ J(λx) ∼ ∂µ (T µν xν ) = T µν ηµν = 0 (4.8)

by ∂µ T µν = 0. Hence, conformal field theories have a traceless energy-momentum tensor T µ µ = 0.

33
4.2 Conformal symmetry in two dimensions
The two-dimensional Minkowski metric factorizes in lightcone coordinates σ ± = x0 ±x1 that have already
been used to solve the free wave equation

ds2 = −(dx0 )2 + (dx1 )2 = −dσ + dσ − (4.9)

where we identify x0 with τ and x1 with σ on the string worldsheet.

We shall now pass to Euclidean time

x0 = τ = −it (4.10)

and complexify the lightcone coordinates


z = iσ + = t + iσ 
⇒ ds2 = dz dz̄ (4.11)
z̄ = iσ − = t − iσ 

¯ z̄) is conformal if
In this setting, a diffeomorphism z → ξ(z, z̄) and z̄ → ξ(z,
 
∂z ∂ z̄ 2 ∂z ∂ z̄ ¯2 ∂z ∂ z̄ ∂z ∂ z̄
2
ds = dξ + ¯ ¯dξ + + dξdξ¯ (4.12)
∂ξ ∂ξ ∂ξ ∂ξ ∂ξ ∂ ξ¯ ∂ ξ¯ ∂ξ
¯
= Ω2 (ξ, ξ)dξdξ¯

As a necessary condition, we need to enforce vanishing coefficients of dξ 2 and dξ¯2 , i.e.

∂ξ ∂ ξ¯ ¯
= = 0 ⇒ ξ(z) meromorphic and ξ(z̄) antimeromorphic (4.13)
∂ z̄ ∂z
¯
Then we can transform z → ξ(z) and z̄ → ξ(z̄) independently, by relaxing the reality of

ξ + ξ¯ ξ − ξ¯
t= , σ= (4.14)
2 2i
¯ do not have to be complex conjugates of each other. The conformal transformations
then (z, z̄) and (ξ, ξ)
realized by meromorphic ξ(z) are generated by z → z + z n+1 with n ∈ Z

F (z) → F (z − z n+1 ) + O(2 ) (4.15)

= F (z) − z n+1 ∂z F (z) + O(2 )
= (1 + ln )F (z) + O(2 )

so we read off generators ln = −z n+1 ∂z with n ∈ Z subject to

[lm , ln ] = (m − n)lm+n (4.16)

Together with ¯ln = −z̄ n+1 ∂z̄ , n ∈ Z this yields an infinite dimensional extension of the naively expected
1
2 (d + 1)(d + 2)|d=2 = 6 generators. Still, the significance of this number can be seen from the 6 generators
l0 , l1 , l−1 , ¯l0 , ¯l1 , ¯l−1 :

• they form a subalgebra (i.e. they close under taking commutators)

• the remaining generators ln≥2 , ln≤−2 , ¯ln≥2 , ¯ln≤−2 are singular as z, z̄ → 0 or ∞; Only the ln , ¯ln at
n ∈ {−1, 0, 1} are globally defined

34
 • the finite form for the globally defined ln , ¯ln is

az + b āz̄ + b̄
z→ , z̄ → (4.17)
cz + d c̄z̄ + d¯
where one can always perform a simultaneous rescaling of numerator and denominator to enforce


a b ∈ SL (C).
c d 2

– The l−1 , ¯l−1 generates translations ( 10 1b )

– The l0 + ¯l0 generates dilatations λ0 λ−1 0

 iθ/2 
– The i(l0 − ¯l0 ) generates rotations e 0
−iθ/2
0 e

– The l1 , ¯l1 generates special conformal transformation ( 1c 10 )

As an example of fields transforming under conformal z → ξ(z) and z̄ → ξ(z̄), ¯ we define conformal
¯
primaries φ (z, z̄) of weight (h, h̄) as follows: the conformal transformation z → ξ(z), z̄ → ξ(z̄) sends
h,h̄

 −h  ¯−h̄
¯ ∂ξ ∂ξ
φh,h̄ (z, z̄) → φ0h,h̄ (ξ, ξ) = φh,h̄ (z, z̄) (4.18)
∂z ∂ z̄

• This generalizes dξ dξ¯ = ∂ξ ∂ ξ̄

∂z ∂ z̄ dz dz̄ to values of h, h̄ different from −1.

• φh,h̄ transforms like a tensor with h holomorphic indices and h̄ antiholomorphic ones Azz . . . z z̄ z̄ . . . z̄ ,
| {z } | {z }
h h̄

∂xν1 ∂xν2 ∂xνh+h̄

cf. Aµ1 µ2 ...µh+h̄ → . . . Aν ν ...ν (4.19)
µ µ
∂y 1 ∂y 2 ∂y µh+h̄ 1 2 h+h̄

• as an example of conformal primaries,

∂ µ ¯ ∂z ∂ ¯
∂z X µ (z, z̄) →


X (ξ, ξ) = X ξ(z), ξ(z̄) (4.20)
∂ξ ∂ξ ∂z

has (h, h̄) = (1, 0), and similarly, ∂z̄ X µ (z, z̄) has (h, h̄) = (0, 1)

¯ = z̄ (working to first order in  only)

• infinitesimal transformations ξ(z) = z + η(z) and ξ(z̄)

δη φh,h̄ (z, z̄) = φ0h,h̄ (z) − φh,h̄ (z) (4.21)

 −h 
∂ξ
= −η(z)∂z φh,h̄ (z) + − 1 φh,h̄ (z) + O(2 )
∂z
= − η(z)∂z φh,h̄ (z) + hφh,h̄ (z)∂z η(z) + O(2 )
 

where we have used the following in passing to the second line:

 −h
0 0 ∂ξ
φh,h̄ (z) − η(z)∂z φh,h̄ (z) + O(2 )


φh,h̄ (z) = φh,h̄ ξ(z) − η(z) = (4.22)
∂z

Let us now take a closer look at the energy-momentum tensor Tαβ∈{z,z̄} in two-dimensional CFTs. It is
traceless as in any higher-dimensional CFT which in the z, z̄ coordinates takes the form

Tzz̄ = Tz̄z = 0 (4.23)

The conservation equations ∂α T αβ along with Tz̄z̄ = 14 T zz give

∂z Tz̄z̄ = ∂z̄ Tzz = 0 (4.24)

35
so we have two independent components – one meromorphic and one anti-meromorphic.

T (z) = Tzz , T̄ (z̄) = Tz̄z̄ (4.25)

The conserved current generating the infinitesimal conformal transformation z → z + η(z) is

J(η) (z) = T (z)η(z) , J¯(η̄) (z̄) = T̄ (z̄)η̄(z̄) (4.26)

4.3 Towards quantum CFTs in two dimensions

We motivate the transition to quantum CFTs through the closed-string example. We can map the cylinder
worldsheet defined by the periodic direction σ ∼
= σ + 2π to the complex plane via ξ(z) = ez = et+iσ ,
where periodicity in σ becomes automatic by ξ = e2πi ξ. Infinite past/future w.r.t. Euclidean time t = iτ
are mapped to points ξ = 0 and ξ = ∞ on the Riemann sphere, C ∪ {∞}.

Let us now investigate the mode expansion on the plane, upon renaming αkµ ↔ α̃kµ for convenience:
r r
∂ µ α0 X µ −inσ± ∂ µ α0 X µ −inσ±
+
X = αn e , −
X = α̃n e (4.27)
∂σ 2 ∂σ 2
n∈Z n∈Z

• the dictionary for the derivatives is ∂σ∂+ = i∂z and ∂σ∂− = i∂z̄ , i.e.
r r
µ α0 X µ −nz µ α0 X µ −nz̄
∂z X = −i αn e , ∂z̄ X = −i α̃n e (4.28)
2 2
n∈Z n∈Z

¯ = ez̄ and ∂z̄ = ξ∂

• after ξ(z) = ez and ∂z = ξ∂ξ as well as ξ(z̄) ¯ , we have
ξ̄
r r
µ α0 X µ −n−1 µ α0 X µ ¯−n−1
∂ξ X = −i αn ξ , ∂ξ̄ X = −i α̃n ξ (4.29)
2 2
n∈Z n∈Z

• henceforth rename ξ(z) → z, we will work on the complex plane for the rest of the course (with
infinite past at z → 0, infinite future z → ∞ and circular constant-τ slice |z| = const)

In general quantum CFTs, one promotes fields to operators

fields
| {z } ←→ operators (4.30)
| {z }
e.g. primaries φ (z, z̄) µ µ
h,h̄ e.g. sums over αn , α̃n
such as ∂z X µ at (h, h̄) = (1, 0) µ
with [αm , αν µν δ
n ] = mη m+n,0

• introduce vacuum state |0i, which is by definition annihilated by “half” the modes of the primaries,
X µ
∂z X µ ∼ αnµ z −n−1 ↔ αn≥0 |0i = 0 (4.31)
n∈Z

36
 where we will clarify the relation between |0i and the state |0; pi from the construction of string
spectra in section 4.5. For primaries of generic (h, h̄), we will expand in mode operators φm,n ,
X
φh,h̄ (z, z̄) ∼ φm,n z −m−h z̄ −n−h̄ ↔ φm,n |0i = 0 ∀ m > −h and n > −h̄ (4.32)
m,n∈Z

Spoiler: the conditions for φm,n |0i = 0 ensure that the asymptotic states ∂z X µ (z → 0)|0i and
φh,h̄ (z → 0)|0i do not diverge as z → 0 or t → −∞

Ln z −n−2 ,
P
• for energy-momentum modes in T (z) = n∈Z

Ln |0i = 0 ∀ n = −1, 0, 1, 2, . . . (4.33)

• the hermitian conjugates are

(h0|φm,n )† = φ−m,−n |0i (4.34)

µ
such that (αnµ )† = α−n ; work with normalization of the conjugate vacuum such that h0|0i = 1.

Define n-point correlation functions as VEV’s

hφh1 ,h̄1 (z1 , z̄1 )φh2 ,h̄2 (z2 , z̄2 ) . . . φhn ,h̄n (zn , z̄n )i = h0|φh1 ,h̄1 (z1 , z̄1 ) . . . φhn ,h̄n (zn , z̄n )|0i (4.35)

They must be radially ordered: the rightmost field has to be closest to origin |z1 | > |z2 | > . . . > |zn |, this
is analogous to the time ordering t1 > t2 > . . . > tn on the cylinder, with the latest insertion on the left.
The need for radial ordering can be nicely illustrated through the following closed-string example:
r !2
α0 X X
µ
h∂z X (z)∂w X (w)i = ν
−i z −n−1 w−p−1 h0|αnµ αpν |0i (4.36)
2
n∈Z p∈Z
α0 X −n−1 X m−1
=− z w h0|αnµ α−m
ν
|0i
2
n∈Z m∈Z
0 X∞ ∞
α X
=− z −n−1 wm−1 h0|nη µν δn,m |0i
2 n=1 m=1
0 ∞
α 1 X  w n
= − η µν n
2 zw n=1 z
∞
α0 η µν X α0 η µν w u
=− u∂u un |u= wz = − ∂u
2zw n=1 2zw z 1 − u
α0 η µν 1 α0 η µν 1
=− = −
2z 2 (1 − u)2 2 (z − w)2
w
w n
P∞
which only converges for | wz | < 1. This convergence criterion


We have used that n=1 z = z
1− w
z

imposes radial ordering |w| < |z|.

4.4 Comutators versus operator product expansion

Let us now explore the quantum version of the infinitesimal conformal transformation z → z + η(z) on
primaries φh,h̄ (ignoring η̄(z̄) and h̄ for ease of notation).

δη φh (z) = −(η(z)∂z φh (z) + hφh (z)∂z ηz ) (4.37)

37
As a general rule in quantum field theory, infinitesimal transformations must stem from commutators with
the conserved charge of the corresponding Noether current. In this case, this is δη φh (z) = −[Q(η), φh (z)]
with
Z I
dz
Q(η) = J(η) = T (z)η(z) (4.38)
const
time BR (0) 2πi

By the mapping z = et+iσ to the plane, constant-time slices are circles BR (0) of radius R around the
origin. By radial ordering, we have to use different R in the two terms in the commutator, depending on
whether Q(η) is on the left or right of φh (w)
I
dz
[Q(η) , φh (w)] = [T (z)η(z), φh (w)] (4.39)
BR (0) 2πi
I I 
dz
= − T (z)η(z)φh (w)
|z|>|w| |z|<|w| 2πi
I
dz
= T (z)η(z)φh (w)
B (w) 2πi
!
= η(w)∂w φh (w) + hφh (w)∂w η(w)
H H
As visualized in the figure below, the difference of the two circular integrals |z|>|w|
− |z|<|w|
can be
deformed to a single contour integral in z over B (w), a circle of infinitesimal radius  around w

This deformation is due to Cauchy’s theorem for meromorphic integrands that allows to deform
integration contours as long as no poles are crossed.
In order to explain the transition from the third line of (4.39) to the fourth one, we rely on Cauchy’s
formula
I
dz
(z − w)n = δn,−1 (4.40)
B (w) 2πi

in combination with the following short-distance singularity of T (z) and φ(w):

hφh (w) ∂w φh (w)
T (z)φh (w) = 2
+ + O((z − w)0 ) (4.41)
(z − w) z−w

This is the first example of operator product expansions (OPE) that are in general valid within the scope
of correlation functions h. . .i: As z → w, the two operators T (z) and φ(w) look like a single operator,
either φh (w) or ∂w φh (w), modular regular terms. More generally, OPEs of operators A(z), B(w) take
the schematic form
singular as z→w
X z }| {
A(z)B(w) ∼ CI (z − w) OI (w) + ...
|{z} (4.42)
operators
OI in CFT regular as z→w

38
where a more precise formulation will be given in a later section. As another specific example of an OPE,
we can extract
α0 η µν
∂z X µ (z)∂w X ν (w) ∼ − + ... (4.43)
2 (z − w)2
0 η µν
from the exact result − α2 (z−w)2 for the corresponding two-point function.
So far, we have translated a known commutation relation [Q(η) , φh (w)] into an OPE. One can con-
versely extract commutators from OPEs, as we shall now illustrate through the example of ∂z X µ : Its
mode expansion implies the contour-integral representation of the oscillators
r I
µ 2 dz n
αn = i 0
z ∂z X µ (z) (4.44)
α BR (0) 2πi

which can be used to determine their commutator as follows

I I
µ ν 2 dz dw m
[αm , αn ] = − 0 [z ∂z X µ (z), wn ∂w X ν (w)] (4.45)
α BRz (0) 2πi BRw (0) 2πi
I I
2 dw n dz m
=− 0 w z ∂z X µ (z)∂w X ν (w)
α BR (0) 2πi B (w) 2πi
α0 η µν
I I  
2 dw n dz
=− 0 w (wm + m(z − w)wm−1 + . . .) −
α BR (0) 2πi B (w) 2πi 2(z − w)2
I
dw n+m−1
= mη µν w = mη µν δm+n,0
BR(0) 2πi

In passing to the second line, we have exploited that Rz > Rw in the first “+” term of the commutator
and that Rz < Rw in the second term “-”. The difference of the two possibilities to arrange the circular
contours can be deformed via Cauchy’s theorem to an infinitesimal circle for the z-integration around w,
whereas w is integrated over a large circle around the origin, see the figure below

Let us use OPEs and the explicit form of the free boson’s energy-momentum tensor to verify that
∂z X µ is a primary field of weight (1, 0)
1
T (z) = − : ∂z X µ (z)∂z Xµ (z) : (4.46)
α0
• we are using the following notation for normal ordering

 αµ αν : m ≥ n
n m
: αnµ αm
ν
:= (4.47)
 αν αµ : m < n
m n

• as in conventional quantum field theory, use Wick’s theorem to convert radially ordered (i.e. time-
ordered) products to normal ordered ones with vanishing VEVs by summing all (partial) pair

39
 contractions
α0 η µν
∂z X µ (z)∂w X ν (w) = − (4.48)
| {z } 2 (z − w)2
without any pair-contractions of fields within the same normal-ordering colons : . . . :.

• apply this to recover the OPE

1
T (z)∂w X ν (w) = − : ∂z X µ (z)∂z Xµ (z) : ∂w X ν (w) (4.49)
α0 
α0 η µν α0 δµν
 
1 µ
=− 0 − ∂z Xµ (z) + ∂z X (z) − + ...
α 2 (z − w)2 2 (z − w)2
1  
= ∂w X ν (w) + (z − w)∂w 2
X ν (w) + . . .
(z − w)2
∂w X ν (w) ∂w (∂w X ν (w))
= + + ...
(z − w)2 z−w
which lines up with the general OPE T (z)φh (w) for a primary φh at h = 1

By a similar sequence of Wick contractions, one can analyze the OPEs of the energy-momentum tensor
and find additional features beyond conformal primary fields:
D/2 2T (w) ∂w T (w)
T (z)T (w) ∼ + + + ... (4.50)
(z − w)4 (z − w)2 z−w
| {z }
6∃ for primaries

The last two terms are those of a primary with h = 2, but the first term is incompatible with the general
form of the primaries’ OPE. Hence, T (z) is a counterexample to a conformal primary.

• in general CFTs beyond the free boson, one refers to the numerator of (z − w)−4 as the “central
charge” c in place of the number of spacetime dimensions D

• as will be explored in the exercises, the above OPE is equivalent to the “Virasoro algebra” for
I
dz
Lm = T (z)z m+1 (4.51)
BR (0) 2πi
namely
c
[Lm , Ln ] = (m − n)Lm+n + (m3 − m)δm+n,0 (4.52)
12

• the central-charge contribution to infinitesimal conformal transformations z → z + η(z) is

I
dz
δη T (w) = − η(z)T (z)T (w) (4.53)
BR (w) 2πi
I
dz  1 1 
= η(w) + (z − w)∂w η(w) + (z − w)2 ∂w 2
η(w) + (z − w)3 ∂w
3
η(w)
BR (w) 2πi 2 6
 c/2 2T (w) ∂w T (w) 
× + + + ...
(z − w)4 (z − w)2 z−w
 c 3 
= − η(w)∂w T (w) + 2T (w)∂w η(w) + ∂w η(w)
12

• the analogue for a finite conformal transformation z → ξ(z) is

 −2 
∂ξ c 
T 0 (ξ) = T (z) − S(ξ, z) (4.54)
∂z 12
with the Schwarzian derivative
−1 2  −2
∂3ξ ∂2ξ
 
∂ξ 3 ∂ξ
S(ξ, z) = − (4.55)
∂z 3 ∂z 2 ∂z 2 ∂z

40
 • for the special choice ξ(z) = ez that has been used to map the cylinder to the plane, we have
S(ξ = ez , z) = − 21 and therefore get Tcyl (z) = ξ 2 Tplane (ξ) − c
24 . At c = 1, this reproduces the
normal-ordering offset per spacetime dimension found in the lightcone quantization
∞ ∞ ∞
1X i X 1X X 1
α−n αni = i
α−n αni + n= i
α−n αni − (4.56)
2 n=1
2 n=1 n=1
24
n6=0

c 3
• the term in the transformation law δη T (w) → − 12 ∂w η(w) that disqualifies T (w) from being a
primary does not affect the globally defined generators l0 , l±1 or their finite versions z → az+b
cz+d with


a b ∈ SL (C); conformal fields with the transformation law of T (z) (i.e. primary with respect to
c d 2

SL2 (C)) are said to be quasi-primary of (h, h̄) = (2, 0).

4.5 State operator correspondence

For each primary field φh,h̄ , we define an asymptotic in-state |h, h̄i via evaluation at infinite past z → 0

|h, h̄i = lim φh,h̄ (z, z̄)|0i (4.57)

z,z̄→0

With the earlier mode expansion

X
φh,h̄ (z, z̄) = φm,n z −m−h z̄ −n−h̄ (4.58)
m,n∈Z

φm>−h,n |0i = |φm,n>−h̄ |0i = 0

only a single mode contributes to this limit

=0 for m<−h
X z }| {
|h, h̄i = lim z −m−h z̄ −n−h̄ φm,n |0i = φ−h,−h̄ |0i (4.59)
z,z̄→0 | {z }
m,n∈Z
=0 for m>−h

We shall now investigate the interplay of |h, h̄i with the Virasoro generators and ignore the antiholomor-
phic (z̄, h̄-dependent) part for simplicity:
I
dz m+1
Lm |hi = lim z T (z)φh (w)|0i (4.60)
w→0 B (0) 2πi
R
I  
dz
hφh (w) ∂w φh (w)
= lim wm+1 + (m + 1)(z − w)wm + . . . 2
+ + . . . |0i
w→0 B (w) 2πi

(z − w) z−w
= lim (m + 1)hwm φh (w)|0i + wm+1 ∂w φh (w)|0i


w→0
X
(m + 1)hwm−n−h + wm−n−h (−n − h) φn |0i


= lim
w→0
n∈Z

• For m > 0, all terms cancel and we have Lm>0 |hi = 0 since wm−n−h vanishes from the limit w → 0
for n < m − h and φn |0i generally vanishes for n > −h

• For m = 0, the same reasoning leaves a single contribution from n = −h, namely L0 |hi = h|hi

• conformal primaries generate a “highest-weight state” of the Virasoro algebra Vir = {Lm , m ∈ Z}

Lm>0 |hi = 0 , L0 |hi = h|hi (4.61)

which we found as the quantum version of the classical constraints Tαβ = 0 in covariant quantization

41
 • we can build a “highest-weight representation” from |hi via action of Lm<0
r
X

L−n1 L−n2 . . . L−nr |hi, nj > 0} ↔ descendant state with L0 eigenvalue h + nj (4.62)
j=1

Also non-primary fields ϕ(z) in a CFT generate states (non-highest-weight states of Vir)

|ϕi = lim ϕ(z)|0i (4.63)

z→0

• a first example is the quasi primary T (z) with state correspondant

lim T (z)|0i = L−2 |0i (4.64)

z→0

where we have used the defining property L−1 |0i = 0 of the vacuum

• the state L−2 |0i does not satisfy the highest-weight condition since

c
L2 L−2 |0i = [L2 , L−2 ] |0i + L−2 L2 |0i = |0i =
6 0 (4.65)
| {z } | {z } 2
4L0 +c/2 0

As a string-theory relevant example of primary fields, we introduce the vertex operators

Vp (z, z̄) = : eip·X(z,z̄) : (4.66)

labelled by a Lorentz-vector pµ .
 
α0 2 α0 2
• as will be shown in the exercises, the Vp (z, z̄) are primaries with (h, h̄) = 4 p , 4 p .

• the vertex operator Vp generates eigenstates of the momentum operator ∼ α0µ ,

|0; pi = lim Vp (w, w̄)|0i (4.67)

w→0

This can be checked by evaluating the action of

r I
2 µ 2i dz
α = 0 ∂z X µ (z) (4.68)
α0 0 α BR (0) 2πi

and importing the OPE

iα0 pµ
∂z X µ (z)Vp (w, w̄) ∼ − Vp (w, w̄) + . . . (4.69)
2 z−w
As will be shown in the exercises,
r I
2 µ 2i dz
α |0; pi = ∂z X µ (z) lim Vp (w, w̄)|0i = pµ |0; pi (4.70)
α0 0 α0 BR (0) 2πi w→0

µ
which clarifies the relations between the ground states |0; pi w.r.t. the αn>0 oscillators and the
“absolute” vacuum |0i.

We can invert the above logic “field ⇒ state” and define a field ϕv for each state |vi,

lim ϕv (z, z̄)|0i = |vi (4.71)

z→0

42
 • The simplest descendants L−1 |hi of the highest-weight state |hi (defined by some primary φh ) are
generated by ∂w φh (w)

Lm |hi m=−1 = lim (m + 1)hwm φh (w) + wm+1 ∂w φh (w) |0i m=−1



(4.72)
w→0

= lim ∂w φh (w)|0i (4.73)

w→0

• The field correspondent of L−n |hi (with n = 2, 3, . . .) can only be given as a contour integral:
I
dz
lim (z − w)1−n T (z)φh (w)|0i = lim L−n φh (w)|0i = L−n |hi (4.74)
w→0 B (w) 2πi w→0


In general, one cannot give an explicit evaluation of the z-integral over B (w) since one would need
the regular terms (z − w)0 , (z − w)1 in the OPE T (z)φh (w). Hence, the best we can do is the above
contour-integral representation of the field correspondants to L−2 |hi, L−3 |hi, etc.

4.6 Correlation functions and Ward identities

Correlators hφh1 ,h̄1 (z1 , z̄1 ) . . . φhn ,h̄n (zn , z̄n )i must transform like the primary fields φhi ,h̄i (zi , z̄i ) them-
selves under conformal transformations zi → ξ(zi ) = ξi and z̄i → ξ(z̄ ¯ i ) = ξ¯i ,
n 
∂ξj −hj  ∂ ξ¯j −h̄j
Y 
hφ0h1 ,h̄1 (ξ1 , ξ¯1 ) . . . φ0hn ,h̄n (ξn , ξ¯n )i = hφh1 ,h̄1 (z1 , z̄1 ) . . . φhn ,h̄n (zn , z̄n )i (4.75)
j=1
∂zj ∂ z̄j

These transformation properties turn out to fix the two-point functions up to a constant

• ξ of translation ⇒ hφh1 ,h̄1 (z1 , z̄1 )φh2 ,h̄2 (z2 , z̄2 )i = g1,2 (z1 − z2 , z̄1 − z̄2 )
| {z } | {z }
z12 z̄12

• ξ of dilatation ⇒ g1,2 (x, y) ∼ x−h1 −h2 y −h̄1 −h̄2 since g1,2 (λz12 , λz̄12 ) = λ−h1 −h2 λ−h̄1 −h̄2 g1,2 (z12 , z̄12 )

• ξ of rotation / special conformal transformation ⇒ g1,2 (x, y) = 0 unless h1 = h2 and h̄1 = h̄2

• putting everything together (with undetermined normalization d12 )

d12 δh1 ,h2 δh̄1 ,h̄2
hφh1 ,h̄1 (z1 , z̄1 )φh2 ,h̄2 (z2 , z̄2 )i = 2h1 2h̄1
(4.76)
z12 z̄12
0
• consistent with earlier two-point function of ∂z X µ with (h, h̄) = (1, 0), can read off d12 → − α2 η µν

By similar arguments, three-point functions of conformal primaries are fixed up to constants C123 (later
µ
examples include Dirac gamma matrices γαβ with SO(1, d−1) spinor indices αβ of certain worldsheet
fields of the RNS superstring),

hφh1 ,h̄1 (z1 , z̄1 )φh2 ,h̄2 (z2 , z̄2 )φh3 ,h̄3 (z3 , z̄3 )i
C123
= h1 +h2 −h3 h1 +h3 −h2 h2 +h3 −h1 h̄1 +h̄2 −h̄3 h̄1 +h̄3 −h̄2 h̄2 +h̄3 −h̄1
(4.77)
z12 z13 z23 × z̄12 z̄13 z̄23
zij zkl
(n ≥ 4)-point functions may depend on conformally invariant cross-ratios zik zjl , and they can depend on
a total of n − 3 such variables (for both zj and z̄j ) by relations like
z12 z34 z14 z23
+ =1 (4.78)
z13 z24 z13 z24
Still, higher point functions are severely constrained by

43
 −2hj
• asymptotic falloff with zj as zj → ∞

• associativity of the OPE

In the rest of this section, we will discuss how correlators involving descendant fields are determined from
the correlators of the corresponding primaries. We introduce the notation
I
(−`) dz
φh (w) = (z − w)1−` T (z)φh (w) (4.79)
B (w) 2πi

for descendant fields corresponding to the state L−` |hi. From this representation of φ(−`) (w), one can
anticipate that correlators of the schematic form hT (z){primaries}i are needed as an intermediate result.
We proceed in two steps (while abbreviating φhj ,h̄j (wj , w̄j ) → φhj (wj ) here and below)

Figure 8: Deformation of the contour C (green) in (4.80) to a combination of infinitesimal circles (red)
around the insertion points w1 , w2 , . . . , wn of the primary fields φhj (wj ).

• by the contour deformations in the above figure 8

I 
dz
η(z)T (z)φh1 (w1 )φh2 (w2 ) . . . φhn (wn ) (4.80)
C 2πi
n  I 
X dz 
= φh1 (w1 ) . . . φhj−1 (wj−1 ) η(z)T (z)φhj (wj ) φhj+1 (wj+1 ) . . . φhn (wn )
j=1 B (wj ) 2πi
n  
X  ∂φhj (wj ) ∂η(wj ) 
= φh1 (w1 ) . . . φhj−1 (wj−1 ) η(wj ) + hj φhj (wj ) φhj+1 (wj+1 ) . . . φhn (wn )
j=1
∂wj ∂wj

where we have used the OPE T (z)φhj (wj ) in order to find the singularities (z −wj )−2 and (z −wj )−1
and to perform the z-integrals over the contour B (wj )

• imposing this identity to hold for any conformal transformation η(z) and for any contour C, we
obtain the following conformal Ward identity
n 
X hj 1 ∂ 
hT (z)φh1 (w1 )φh2 (w2 ) . . . φhn (wn )i = + hφh1 (w1 )φh2 (w2 ) . . . φhn (wn )i
j=1
(z − wj )2 z − wj ∂wj
(4.81)

44
 (−`)
From this, we can infer the desired correlator hφh (w){primaries}i
I
(−`) dz
hφh (w)φh1 (w1 )φh2 (w2 ) . . . φhn (wn )i = (z − w)1−` hT (z)φh (w)φh1 (w1 ) . . . φhn (wn )i
B (w) 2πi | {z }
| {z } see above Ward identity
Pn H
contour deform to − j=1 B (wj )

n I
X dz  hj 1 ∂ 
=− (z − w)1−` 2
+ hφh (w)φh1 (w1 )φh2 (w2 ) . . . φhn (wn )i
j=1
2πi
B (wj ) | {z } (z − wj ) z − wj ∂wj
(wj −w)1−` +(1−`)(z−wj )(wj −w)−` +...
n  
X (`−1)hj 1 ∂
= − hφh (w)φh1 (w1 )φh2 (w2 ) . . . φhn (wn )i (4.82)
j=1
(wj − w)` (wj − w)`−1 ∂wj
| {z }
L−`

The result is a differential operator L−` corresponding to L−` acting on a correlation function of primaries.
(−`)
We have reduced the correlator involving one descendant field φh (w) to the simpler correlator of
primaries only.

• for the simplest descendants at ` = 1, the expression for

n
X ∂ ∂
L−1 = − = (4.83)
j=1
∂w j ∂w
(−1)
is consistent with the earlier result φh (w) = ∂w φh (w), where we used translation invariance
Pn ∂ ∂
j=1 ∂wj + ∂w = 0 of the correlator

• the same strategy applies to correlators with several descendant insertions, but the amount of
algebra will grow, and the expressions start to depend on the central charge c since one encounters
double-insertions of T (z), e.g.
I I
(−` ) (−` ) dz1 dz2
hφh1 1 (w1 )φh2 2 (w2 ){primaries}i = (z1 − w1 )1−`1 (z2 − w2 )1−`2
B (w1 ) 2πi B (w2 ) 2πi
× hT (z1 )φh1 (w1 )T (z2 )φh2 (w2 ){primaries}i (4.84)

There are similar Ward identities to express the correlator hT (z1 )T (z2 ){primaries}i in terms of
primary correlators without energy-momentum insertions, where T (z1 )T (z2 ) are replaced by any
c
term generated from OPEs including 4 .
2z12

4.7 Correlation functions versus OPEs

The goal of this section is to refine the general expression for OPEs of primary fields
singular as z→w
X z }| {
φhi (z)φhj (w) ∼ Ck (z − w) Ok (w) + ...
|{z} (4.85)
operators
Ok in CFT regular as z→w

The key tool will be to impose consistency of two- and three-point functions. More specifically, we start
from an ansatz for the primaries on the right-hand side
X k
(z − w)Nij Cij k φhk (w) + O(z − w) + descendants
 
φhi (z)φhj (w) ∼ (4.86)
primary k

• Laurent expand the three-point function around z1 = z2 ,

 Cijk n 1 z12 (hi +hk −hj ) 2
o
hφhi (z1 )φhj (z2 )φhk (z3 )i  = hi +hj −hk hj +hk −hi − + O(z )

h +h −h h +h −h +1 12
z1 →z2 z12 z23 z23i k j z23i k j
Cijk n z12 2
o
= hi +hj −hk 2h 1 − (hi +hk −hj ) + O(z12 ) (4.87)
z12 z23 k z23

45
 • match with the OPE ansatz and two-point function
`
N
X
hφhi (z1 )φhj (z2 )φhk (z3 )i ∼ Cij ` z12ij hφh` (z2 )φhk (z3 )i
primary `
`
N
X
` z12ij
= Cij d`,k 2hk
(4.88)
primary ` z23
at hk =h`

• this is compatible with the Laurent expansion if

X
Nij ` = h` − hi − hj , Cijk = Cij ` d`,k (4.89)
primary `
at hk =h`

i.e. the three-point couplings Cijk are related to the OPE data Cij ` and the normalization factors
d`,k of the two-point functions.

As we will see next, the contributions of descendant fields to OPEs are interlocked with those of the
primaries via conformal invariance. Refine the above ansatz to also incorporate the descendants of each
primary field φhk
X Cij k
φhi (z)φhj (w) ∼ h +hj −hk
(4.90)
primaries k z12i
n o
(−1) 2 (−1,−1) 2 2 (−2) (−2) 3
× φhk (z2 ) + z12 βijk ∂z2 φhk (z2 ) + z12 βijk ∂z2 φhk (z2 ) + z12 βijk φhk (z2 ) + O(z12 )

(−` ,−`2 ,...)

with some coefficients βijk 1 ↔ L−`1 L−`2 |hi that can be determined from conformal invariance

3
• the terms +O(z12 ) refer to all descendant fields corresponding to L−`1 L−`2 . . . L−`m |hk i at the order
h −hi −hj +`1 +`2 +...+`m
of z12k of the OPE with `1 + `2 + . . . ≥ 3
(−` ,...,−`m )
• their coefficients βijk 1 only depend on hi , hj , hk and c; as will be worked out in the exercises,
the subleading order of (4.87) implies

(−1) hi + hk − hj
βijk = (4.91)
2hk

4.8 Boundary conformal field theory

We shall now embed open strings into the CFT framework. The central new features of open strings as
compared to closed strings are their endpoints (at σ = 0, π in the parametrization of earlier sections).
These endpoints introduce worldsheet boundaries at z ∈ R after mapping the cylinder to the plane via
z = et+iσ , see figure 9. The bulk of the worldsheet is then mapped to the upper half plane.
The boundary reduces the conformal symmetry to those maps z → ξ(z) that preserve Im z ≥ 0 and
in particular the set of z ∈ R.

• Such boundary preserving diffeomorphism are generated by T (z) and T̄ (z̄) restricted to the upper
half plane and subject to the gluing condition

T (z) = T̄ (z̄) ∀ z, z̄ ∈ R (4.92)

In case of the CFT of the free boson, the gluing conditions encode the fact that Neumann or
Dirichlet boundary conditions determine the α̃nµ oscillators in terms of the αm
ν
.

46
Figure 9: The contribution of the endpoint at σ = 0 (at σ = π) to the open-string worldsheet is mapped
to R+ (to R− ) on the plane under z = et+iσ .

• The gluing conditions can be satisfied by combining T (z) and T̄ (z̄) to a single field T op (z) defined
on z ∈ C which is continuous at z ∈ R

 T (z) : Im z ≥ 0
T op (z) = (4.93)
 T̄ (z̄) : Im z < 0

• The combined field T op (z) generates a single copy of the Virasoro algebra via
I
dz op
Lop
n = T (z)z n+1 (4.94)
B (0) 2πi

2
2
In case of the free boson, the gluing conditions for T (z) ∼ ∂z X(z) and T̄ (z̄) ∼ ∂z̄ X(z̄) can be
enforced in two different ways: We attain T (z) = T̄ (z̄) for z ∈ R if

 +∂ X µ (z̄ ∈ R) : Neumann; ∂ ∼ ∂ − ∂ = 0
z̄ σ z z̄
∂z X µ (z ∈ R) = (4.95)
 −∂z̄ X µ (z̄ ∈ R) : Dirichlet; ∂τ ∼ ∂z + ∂z̄ = 0

• We can again combine ∂z Xµ & ∂z̄ Xµ to a single field ∂z Xµop on C with one copy of oscillators αµop .

• Recall the closed-string two-point function (or Green function)

α0
hX µ (z)X ν (w)i = η µν Gcl (z, w) , Gcl (z, w) = − log |z − w|2 (4.96)
2
(we have derived h∂z X µ (z)∂w X ν (w)i earlier on via operator methods). Neither Gcl (z, w) nor any
of its derivatives vanish for z, w ∈ R.

• Define open-string Green functions GN D

op , Gop adapted to Dirichlet and Neumann boundary condi-

tions by the boundary-value problems

α0 2
∂z ∂z̄ GN,D
op (z, w) = ∂z ∂z̄ Gcl (z, w) = − πδ (z − w) , Im z, Im w > 0 (4.97)
2
(∂z −∂z̄ )GN D
op (z, w) = 0 = (∂z +∂z̄ )Gop (z, w) , z∈R (4.98)

• Solve via method of image charges that you know from your electrodynamics course, e.g.

GN
op (z, w) = Gcl (z, w) + Gcl (z̄, w) (4.99)
0
α
log |z − w|2 + log |z̄ − w|2


=−
2

47
 The contribution Gcl (z̄, w) from the image charge z̄ of z does not pose any obstructions to having
0
α
∂z ∂z̄ GN 2
op (z, w) = − 2 πδ (z − w) in the upper half plane since ∂z ∂z̄ Gcl (z̄, w) has delta-function

support on the lower half plane.

• With both of z and w on the boundary z, w ∈ R, the image charge yields a factor of two between
the open- and closed-string Green functions GN
op |R = 2Gcl |R . As we will see later on, this reflects

the rescaling of the open- and closed-string mass spectra m2cl = 4m2op observed in the discussion of
lightcone quantization.

5 Path integrals and ghosts

We have seen powerful techniques to constrain and determine correlation functions in two-dimensional
conformal field theories which do not rely on the availability of an action functional. The worldsheet
theory of bosonic strings in a rather “atypical” CFT in the sense that it admits a description through
the Polyakov action in conformal gauge
Z
1
S[X] = d2 z ∂z Xµ (z)∂z̄ X µ (z) (5.1)
πα0
for free bosons X µ with µ = 0, 1, . . . , D−1 and total central charge c = D. We will now discuss path-
integral methods to both perform CFT computations and to introduce a b, c ghost system that drives
covariant quantization.
Warning: Our conventions will differ by factors of 2 from those of the Blumenhagen / Lüst / Theisen
textbook, David Tong’s lecture notes and the Polchinski textbooks. More specifically, for z = x + iy,
1
d2 z here = dx ∧ dy = d2 z there
 
(5.2)
2
δ (z) here = δ(x)δ(y) = 2δ 2 (z) there
2
 
(5.3)

leading to the following normalized integral in all cases:

Z
d2 z δ 2 (z) = 1 (5.4)
B (0)

5.1 Path-integral methods for correlators of the free boson

Starting from the above Polyakov action S[X] in conformal gauge, we will now introduce path-integral
methods to compute correlation functions. These methods will provide an alternative to the operator
formalism that could in principle be used to determine correlation functions as VEVS h0| . . . |0i.

• We initiate the definition of path integrals by introducing the partition function Z as a normalization
for “zero-point functions” Z
Z= D[X] e−S[X] (5.5)

• Let φi (zi ) denote any field built from X µ (z, z̄) and its derivatives (not necessarily a conformal
primary). Then, we give a formal definition of n-point functions through the path integral
Z
1
hφ1 (z1 )φ2 (z2 ) . . . φn (zn )i = D[X] φ1 (z1 )φ2 (z2 ) . . . φn (zn ) e−S[X] (5.6)
Z
where the normalization by Z −1 ensures that h1i = 1.

48
 • We have identified ∂z X µ and ∂z̄ X µ to be conformal primaries of weight (h, h̄) = (1, 0) and (0, 1).
The path integral D[X] should be thought of as integrating over all field configurations for X µ (z, z̄)
R

but not as independently integrating over all the descendant fields.

In order to define the path integral in the numerator of hφ1 (z1 ) . . . φn (zn )i, we impose two basic rules:
First, partial derivatives ∂zj in the insertion points commute with the path integral. Second, functional
derivatives integrate to zero
Z
δ
. . . e−S[X]


0= D[X] (5.7)
δXµ (z)
We will now derive some corollaries by simple choices of the insertions . . . in the functional derivatives:

• Ehrenfest theorem: VEVs obey the classical equations of motion

Z Z
δ −S[X]
 δS[X] 
0 = D[X] e = D[X] − e−S[X]
δXµ (z) δXµ (z)
| {z }
2
πα0
∂z ∂z̄ X µ (z)

=⇒ h∂z ∂z̄ X µ (z)i = 0 (5.8)

• Reproducing the Laplace equation of the two-point function known from the operator formalism:
Z
δ
X ν (w)e−S[X]


0 = D[X]
δXµ (z)
Z  2 
= D[X] η µν δ 2 (z − w) + ∂ z ∂ z̄ X µ
(z)X ν
(w) e−S[X] (5.9)
πα0
α0
=⇒ ∂z ∂z̄ hX µ (z)X ν (w)i = − πδ 2 (z − w)
2

As a next step, we will use path integrals for an efficient evaluation of n-point functions of vertex operators
 0 0

Vp (z, z̄) =: eip·X(z,z̄) : that we identified as conformal primaries of weight (h, h̄) = α4 p2 , α4 p2 in earlier
sections

• Given that the Polyakov action is quadratic in X µ , the finite-dimensional prototype for correlators
involving eip·X(z,z̄) is the Gaussian integral
Z Z
x+i~b·~
x·A~ 1~
b·A−1~b
n
d xe ~ x
=e 4 dn x e~x·A~x (5.10)
Rn Rn

with some invertible n × n matrix A (whose eigenvalues need to have a negative definite real part
to obtain a convergent integral).

• Since the path-integral analogue of the matrix A is ∂z ∂z̄ , the analogue of A−1 is the Green function,
i.e. the solution to ∂z ∂z̄ g(z − w) = δ 2 (z − w) (the closed-string Green function g(z − w) ∼ Gcl (z, w)
in unit normalization). One can think of

1
g(z − w) = log |z − w|2 (5.11)
π
as the matrix element of (∂z ∂z̄ )−1 .

49
 • We shall now use this to compute the n-point function of Vpj (zj , z̄j ):
* n + Z  Xn Z 2 
Y
ipj ·X(zj ) 1 d z µ
:e : = D[X] exp i pj · X(zj ) + X µ (z)∂ ∂
z z̄ X (z)
j=1
Z j=1
πα0
Z Z i
1 h ∂z ∂z̄ µ
= D[X] exp d2 z Xµ (z) X (z) + iX µ (z)J µ
(z) (5.12)
Z πα0
n
πα0
X   Z Z 
D 2 2 µ
=δ pj exp d z d w J (z)g(z − w)Jµ (w)
j=1
4

In passing to the second line, we have introduced the shorthand for the source term
n
X
J µ (z) = pµj δ 2 (z − zj ) (5.13)
j=1

and the delta function in the third line is due to the finite-dimensional integral over zero modes xµ
that is part of D[X] over infinitely many modes. The double-integral over z and w in the exponent
of the third line can be thought of as the continuum version of J~ · (∂z ∂z̄ )−1 J,
~ i.e. each vector-
index contraction generalizing to one of the d2 z, d2 w integrations. By inserting the delta-function
representation of the source term into the double integral, we find
* n + Xn   0 Xn 
Y
ipj ·X(zj ) α
:e : = δD pj exp pi · pj log |zi −zj |2
j=1 j=1
4 i,j=1
i6=j
n
X n
 Y
0
= δD pj |zi −zj |α pi ·pj (5.14)
j=1 1≤i<j

after discarding the terms from i = j. This is the famous “Koba-Nielsen factor” on the sphere
which will be shown to crucially shape the properties of string tree-level amplitudes.

• The Koba-Nielsen factor at n = 3 is consistent with the general result on the three-point function
 0 0

of conformal primaries of weights (hj , h̄j ) = α4 p2j , α4 p2j with j = 1, 2, 3: We can match
* 3
+
 α0 2 2 2 α0 2 2 2 
(p −p −p ) (p −p −p )
Y
ipj ·X(zj )
:e : = C123 z124 3 1 2 z̄124 3 1 2 × cyclic(1, 2, 3) (5.15)
j=1

by identifying C123 = δ D (p1 +p2 +p3 ) and noting that

p23 − p21 − p22 = (p1 + p2 )2 − p21 − p22 = 2p1 · p2 (5.16)

on the support of three-point momentum conservation.

5.2 Polyakov path integral and b, c ghost system

We shall now revisit the role of the diffeomorphism and Weyl invariance in quantizing the Polyakov action
√
Z
1
SP [X, h] = − 0
d2 σ −h hαβ ∂α X µ ∂β Xµ (5.17)
4πα
in presence of a general worldsheet metric hαβ . Following the path-integral approach to gauge theories,
the path integral for SP [X, h] should only involve diffeomorphism × Weyl inequivalent configurations of
the fields X(σ) and h(σ). At the level of the partition function, this means that we need the inverse
volume factor in Z
1
ZP = D[X]D[h] e−SP [X,h] (5.18)
vol

50
which instructs to formally divide out the infinite overcount by diffeomorphism × Weyl (though there
is no way of assigning a finite volume to the group of diffeomorphisms and Weyl transformations). We
proceed on the basis of the following key ideas

• as visualized in figure 10, change field variables from (X, h) to

{physically distinct configurations} × {gauge orbits} (5.19)

Figure 10: Space of (X, h) field configurations.

• this change of field variables yields a Jacobian, the so-called “Faddeev-Popov determinant”, which
may be viewed as the infinite-dimensional analogue of the following Jacobian in n dimensions

∂xµ 
Z Z 
dn x φ(~x) = dn y  det ν  φ ~x(~y )


(5.20)

∂y

In order to make these ideas more explicit, we introduce the collective notation

ζ = {σ̃, φ} = {diffeomorphism σ α → σ̃ α (σ), Weyl transformation hαβ → e2φ(σ) hαβ } (5.21)

for diffeomorphism × Weyl transformations which act on the worldsheet metric as follows

∂σ γ ∂σ δ
hαβ (σ) → e2φ(σ) hγδ (σ̃) = hζαβ (σ̃) (5.22)
∂ σ̃ α ∂ σ̃ β
• we can locally attain any metric ĥαβ of our choice (e.g. ĥαβ = ηαβ ) by suitably dialing ζ = {σ̃, φ}

• conversely, any hαβ can be realized as ĥζαβ , again for a suitable choice of ζ

• implementing this via functional delta distribution

Z
D[ζ] δ[h − ĥζ ] = ∆−1
F P [h] (5.23)

we get the inverse Faddeev-Popov determinant, the infinite-dimensional version of

Z X 1
dn x δ n ~y (~x) =


(5.24)
 det ∂yµν 
 
~ x)=~
y (~ 0 ∂x

51
 • the Faddeev-Popov determinant is diffeomorphism × Weyl invariant:
Z
0
∆−1
FP [hζ
] = D[ζ 0 ] δ[hζ − ĥζ ]
Z
−1 0
= D[ζ 0 ] δ[h − ĥζ ζ ] (5.25)
Z
00
= D[ζ 00 ] δ[h − ĥζ ] = ∆−1
F P [h]

0 −1 0
In passing to the second line, we have used that hζ = ĥζ is equivalent to h = ĥζ ζ
, and the last
step is based on the property D[ζ −1 ζ 0 ] = D[ζ 0 ] of the measure.

• insert 1 = ∆F P [h] D[ζ]δ[h − ĥζ ] into the path integral,

R

Z
1
ZP = D[X]D[h]D[ζ]∆F P [h]δ[h − ĥζ ]e−SP [X,h]
vol
Z
1 ζ
= D[X]D[ζ]∆F P [ĥζ ]e−SP [X,ĥ ] (5.26)
vol
Z Z
1
= D[ζ] D[X]∆F P [ĥ]e−SP [X,ĥ]
vol
| {z }
=1

In the last step, we have replaced ĥζ → ĥ since both ∆F P and SP are diffeomorphism × Weyl
invariant.

The next step is to compute ∆F P [ĥ]:

• work infinitesimally, with small φ(σ) and η α (σ) in σ̃ α = σ α − η α (σ)

δ ĥαβ = 2φĥαβ + ∇α ηβ + ∇β ηα (5.27)

• with D[ζ] = D[φ]D[η],

Z
∆−1
F P [ĥ] = D[φ]D[η] δ[2φĥαβ + ∇α ηβ + ∇β ηα ] (5.28)

~
• insert the Fourier representation of the functional delta distribution analogous to δ n (~x) = dn βe2πiβ·~x ,
R

Z  Z q 
∆−1
FP [ĥ] = D[φ]D[η]D[β] exp 2πi d2
σ − ĥ β αβ
[2φ ĥαβ + ∇ η
α β + ∇ η
β α ] (5.29)

• integrate over φ which takes the role of a Lagrange multiplier enforcing β αβ ĥαβ = 0
Z  Z q 
∆−1
FP [ĥ] = D[η]D[β] exp 4πi d2
σ −ĥ β αβ
∇ α η β (5.30)

−1
However, we need ∆+1
F P [ĥ] rather than ∆F P [ĥ] for ZP . This can be adapted by changing the statistics of
both β and η:

• think of ∆−1
F P [ĥ] as the infinite-dimensional version of the complex integral

πn
Z
∗
d2n z e−~z·B~z = (5.31)
det B

• get det B into the numerator by passing from z to Grassmann-odd variables

Z
~ ~∗
dn θ dn θ∗ eθ·B θ = det B (5.32)

52
 • convert (β αβ , ηα ) to Grassmann-odd fields (bαβ , cα )
Z
∆F P [ĥ] = D[b]D[c] e−Sgh [b,c,ĥ] (5.33)
Z q
1
Sgh [b, c, ĥ] = d σ −ĥ bαβ ∇α cβ
2
(5.34)
2π
The subscript of the action indicates that (bαβ , cα ) are ghost variables (as usual for Grassmann-odd
variables in vector and tensor representations).

• in summary, the partition function for the Polyakov action with gauge-fixed metric ĥ reads
Z
ZP = D[X]D[b]D[c] e−SP [X,ĥ]−Sgh [b,c,ĥ] (5.35)

While above discussion holds for an arbitrary choice of the gauge-fixed metric ĥαβ , we will now study
the additional simplifications in conformal gauge where

ĥαβ = e2φ ηαβ (5.36)

p
• use −ĥ = e2φ and complex coordinates (z, z̄): in this case

bzz̄ = 0 , ∇z = ĥzz̄ ∇z̄ = 2e−2φ ∇z̄ , ∇z̄ = 2e−2φ ∇z (5.37)

and therefore Z
2φ 1
Sgh [b, c, ĥαβ = e ηαβ ] = d2 z (bzz ∇z̄ cz + bz̄z̄ ∇z cz̄ ) (5.38)
π
• since the mixed Christoffel symbols Γzz̄α vanish by exercise A.2, one can replace ∇ → ∂,

∇z̄ cz = ∂z̄ cz + Γzz̄α cα = ∂z̄ cz (5.39)

• with the shorthands (b, b̄) = (bzz , bz̄z̄ ) and (c, c̄) = (cz , cz̄ ),
Z
1
Sgh [b, c, ĥαβ = e2φ ηαβ ] = d2 z (b∂z̄ c + b̄∂z c̄) (5.40)
π
which yields the following classical equations of motion

∂z̄ c = ∂z̄ b = 0 , ∂z c̄ = ∂z b̄ = 0 (5.41)

5.3 The ghost CFT

The dynamics of the (b, c)-ghost system is governed by a CFT. To begin with, the computation of the
δSgh [b,c,ĥ]
energy-momentum tensor from δ ĥαβ
is subtle due to the constraint ĥαβ bαβ = 0. We simply state
the result for the non-vanishing components T = Tzz and T̄ = Tz̄z̄

T = 2(∂z c)b + c∂z b , T̄ = 2(∂z̄ c̄)b̄ + c̄∂z̄ b̄ (5.42)

We can extract a naive expression for the two-point function from the path integral
Z
δ
b(w)e−Sgh [b,c]


0 = D[b]D[c] (5.43)
δb(z)
Z  1 
= D[b]D[c] − ∂z̄ c(z)b(w) + δ 2 (z − w) e−Sgh [b,c]
π
=⇒ ∂z̄ hc(z)b(w)inaive = πδ 2 (z − w) (5.44)
1
=⇒ hc(z)b(w)inaive = (5.45)
z−w

53
However, there are subtleties about the zero-mode integrals of b, c that we have not yet taken into account.
As we will see later on, the above expression for the two-point function only holds in presence of suitable
zero-mode insertions. Still, the naive expression for the two-point function is reliable to extract the OPE
1 1
c(z)b(w) ∼ + ... , b(z)c(w) ∼ + ... (5.46)
z−w z−w
which in turn can be used to determine the conformal weights and the central charge:

• normal order the energy-momentum tensor

T (z) = 2 : ∂z c(z) b(z) : + : c(z)∂z b(z) : (5.47)

• read off hb = +2 and hc = −1 from

−c(w) ∂w c(w)
T (z)c(w) ∼ 2
+ + ... (5.48)
(z − w) z−w
2b(w) ∂w b(w)
T (z)b(w) ∼ 2
+ + ... (5.49)
(z − w) z−w

• extract the central charge cgh = −26 from

−13 2T (w) ∂w T (w)
T (z)T (w) ∼ 4
+ 2
+ + ... (5.50)
(z − w) (z − w) z−w

This result for the central charge yields a new perspective on the critical dimension D = 26 of the bosonic
string which we derived from Lorentz invariance in lightcone quantization earlier on:

• combine the CFT of X µ in D spacetime dimensions with the (b, c) system

⇒ ctotal = cX + cgh = D − 26 (5.51)

• in order to avoid Weyl anomalies, we have to impose ctotal to vanish: this can be seen from the
following one-point function when the CFT is put into a curved background with Ricci curvature R,
ctotal
hT α α i = − R (5.52)
12
which is in conflict with the necessary condition T α α = 0 for conformal symmetry

• the previous argument seems to fix the number of spacetime dimensions to D = 26, but there is still
the possibility to adjoint additional CFT sectors to the X µ and (b, c): in presence of some “internal”
CFT with central charge cint , the condition for the cancellation of Weyl anomalies is cint + D = 26,
which leaves the possibility of D < 26; the internal CFT can but does not need to have a geometrical
interpretation as describing compactified (say circular / toroidal) extra dimensions

• in the analogous central-charge bookkeeping for the superstring, the bosons X µ for each spacetime
D
direction are accompanied by fermions ψ µ which contribute 2 to the central charge; on the other
hand, there is an additional system of (β, γ) ghosts (the “superpartners” of the (b, c) system) with
central-charge contribution +11; hence, the total central charge of the superstring is
3D
csuper 1


total = D 1 + 2 − 26 + 11 = − 15 (5.53)
2
which vanishes in the critical dimension D = 10; again, central-charge cancellation in lower dimen-
sions is still possible by adjoining internal CFT sectors

54
5.4 Background charge in the (b, c) system
In this section, we resolve the following two worrisome observations on the (b, c) system

(i) Since the c-ghost has negative conformal weight hc = −1, the general mode expansion for conformal
primaries takes the form
X
c(z) = cn z −n+1 , cn≥2 |0i = 0 (5.54)
n∈Z

Hence, c1 is the first mode that does not annihilate |0i. However, it still lowers the conformal weight
since [L0 , c1 ] = −c1 . Consequently, |0i is not the ground state of the (b, c)-oscillator algebra

{bm , cn } = δm+n,0 (5.55)

(though it is of course a highest-weight state of Vir).

(ii) The naive two-point function hc(z)b(w)inaive = (z − w)−1 discussed above is in conflict with the
orthogonality property hφh1 (z)φh2 (w)i ∼ δh1 ,h2 of weight-hi primaries: with (h1 , h2 ) = (−1, 2) of
(c, b), the two-point function should vanish with δh1 ,h2 .

The resolutions to (i) and (ii) are in fact closely related:

• first note that one can construct two possible L0 ground states from the ghost oscillators

|ci = c1 |0i , |(∂c)ci = c0 c1 |0i (5.56)

subject to the highest-weight condition w.r.t. the ghost oscillators (using c21 = 0)

cn≥1 |ci = cn≥1 |(∂c)ci = 0 (5.57)

• impose unusual hermiticity conditions (such that exchanging the role of |ci ↔ |(∂c)ci does not have
any impact on observables, i.e. inner products)

(c1 |0i)† = h0|c−1 c0 ⇔ h0|c−1 c0 c1 |0i = 1 (5.58)

| {z } | {z }
“democratic” between the SL2 -invariant vacuum carries
2 choices |ci and |(∂c)ci 3 units of “ghost charge”

• yields expected three-point functions of hc = −1 primaries

3
−nj +1
X Y
hc(z1 )c(z2 )c(z3 )i = h0| cnj zj |0i
n1 ,n2 ,n3 ∈Z j=1

= h0|c−1 c0 c1 |0i z12 z2 − z1 z22 + cyc(z1 , z2 , z3 )

(5.59)
−h1 −h2 +h3
= z12 z13 z23 = z12 × cyc(z1 , z2 , z3 )

• non-zero h0|0i would give inconsistent two-point function

X
hb(z)c(w)i = h0|bn z −n−2 cm w−m+1 |0i
m≤1
n≥2

w X  w n w3 h0|0i
= h0|0i 2
= 3 (5.60)
z z z z−w
n≥2

⇒ need to set h0|0i = 0.

55
 • actual two-point function requires extra cn -insertion:

1
h0|c−1 c0 c1 b(z)c(w)|0i = = h0|c−1 c0 c1 c(z)b(w)|0i (5.61)
z−w
where anticommuting the bm -modes to the left-vacuum (with h0|bm = 0 ∀ m ≤ 1) generates the
desired three extra terms relative to the previous computation

w3 1 w2 w 1 1
3
+ 3 + 2+ = (5.62)
z z−w z z z z−w
R
• in path-integral language, D[c] contains three fermionic zero-mode integrals
Z
dc−1 dc0 dc1 cx−1 cy0 cz1 = δx,1 δy,1 δz,1 (5.63)

which have been neglected in hb(z)c(w)inaive = (z − w)−1 .

Let us elaborate on the formal origin of the properties h0||0i = 0 and h0|c−1 c0 c1 |0i = 1, starting from the
anomalous ghost current
j(z) = − : b(z)c(z) : (5.64)

• j(z) is not a conformal primary

−3 j(w) ∂w j(w)
T (z)j(w) ∼ + + + ... (5.65)
(z − w)3 (z − w)2 z−w

• the resulting anomalous transformation j(z) → j( z1 ) under exchange z → 1

z of infinite past and
future affects the hermiticity properties of the modes:
I
dz n
jn = z j(z) , j0† = −j0 + 3 (5.66)
B (0) 2πi

• recall the state-operator correspondence: given some operator Oq subject to


qOq (w)  O (w) = c(w)
q=1
j(z)Oq (w) ∼ + ... , e.g. (5.67)
z−w  Oq=−1 (w) = b(w)

can generate a j0 -eigenstate via

|qi = lim Oq (z)|0i ⇒ j0 |qi = q|qi . (5.68)

z→0

• from the anomalous j0† , inner products hq 0 |qi vanish unless q + q 0 = 3,

†
qhq 0 |qi = hq 0 |j0 |qi = j0† |q 0 i |qi (5.69)

†
= (3 − j0 )|q 0 i |qi = (3 − q 0 )hq 0 |qi

• as a result, the hermitian-conjugate pairs are

q = 0 state |0i ↔ q = 3 state c−1 c0 c1 |0i

q = 1 state c1 |0i ↔ q = 2 state c0 c1 |0i (5.70)

56
 • on a curved worldsheet with Ricci scalar R, the ghost current obeys an anomalous conservation law

3√
∂z̄ j(z) = − −hR (5.71)
4
The right-hand side is a total derivative in two dimensions and induces a topological term upon
integration
√
Z
1
χ= d2 σ −hR = 2(1 − g) (5.72)
4π Σ
where χ is known as the “Euler characteristics” of the surface Σ and g is its “genus” (informally
speaking, the number of holes in Σ) which we will shortly see to count the loop order in string
perturbation theory

• the number of zero modes Nc , Nb of c, b depends on g

Nc − Nb = 3 − 3g (5.73)

which is related to the Riemann–Roch theorem, see e.g. section 6.2 in Blumenhagen-Lüst-Theisen
R R
textbook. In this way, the zero-mode integrals dc−1 dc0 dc1 within D[c] arise as the special case
at g = 0

5.5 Vertex operators

With the tools from the previous sections on conformal field theory and the ghost system, we can now
give a covariant description of the bosonic-string spectrum without any recourse to the lightcone.
disclaimer: This section will take a shortcut around the BRST quantization which will be introduced
in the tutorial on Nov 13 2020. I.e. the ghosts will play a rather minor role in the following, and a more
elaborate treatment can for instance be found in section 5 of the Blumenhagen-Lüst-Theisen textbook.
Recall the physical-state condition of section 2.2

Ln>0 |physi = L̃n>0 |physi = 0 (5.74)

(L0 − a)|physi = (L̃0 − ã)|physi = 0

• identifies field correspondant

|physi = lim Vphys (z)|0i (5.75)
z→0

to be a conformal primary of weight (a, ã)

• while reference to some insertion point z conflicts with diffeomorphism × Weyl invariance, can
generate an invariant by integration over z:
Z
d2 z Vphys (z) diffeomorphism × Weyl invariant ⇔ (a, ã) = (1, 1) (5.76)

the values (a, ã) = (1, 1) of the normal-ordering constants compensate the transformation of d2 z
with weight (−1, −1)

57
 By exercise C.4, plane waves : eip·X(z) : generate eigenstates of the momentum operator
r I
2 µ 2 dz
0
α0 = 0
i∂z X µ (z) (5.77)
α α B (0) 2πi
For the vertex operator of a physical state, make the ansatz

Vphys (z) =: Wphys (z)eip·X(z) : (5.78)

µ µ
with Wphys (z) built from the field correspondents ∂zn X µ (z) of α−n and ∂z̄n X µ (z) of α̃−n with n ≥ 1
0 2 0 2
• plane waves contribute conformal weights ( α 4p , α 4p ) with p2 = −m2 , so Wphys must carry
 α0 m2 α0 m2 
(hW , h̄W ) = 1+ , 1+ (5.79)
4 4
µ µ
In this way, the level-matching condition (same oscillator number for the α−n and α̃−n ) is auto-
matically satisfied.

• without ghosts, we clearly have hW , h̄W ≥ 0, so m2 is bounded from below by − α40 which is the
closed-string tachyon identified before
4
Vtachyon (z) = : eip·X(z) : , p2 = (5.80)
α0
• can add integer units of hW via ∂zn≥1 X µ
r r
4 Y X
m2 = (N − 1) ←→ Wphys ∼ : ∂znj X µj : where nj = N (5.81)
α0 j=1 j=1

but one still has to impose (h, h̄) = (1, 1) primary condition on Vphys (w) via OPE with T (z).

For massless vertex operators, the above counting leads to the ansatz

Vm2 =0 (z) = ζµν : ∂z X µ ∂z̄ X ν eip·X(z) : , p2 = 0 (5.82)

for Wphys of weight (1, 1) with some polarization tensor ζµν . As you have worked out in exercise C.4, the
normal-ordered combination with the plane wave is a primary of weight (h, h̄) = (1, 1) if

pµ ζµν = ζµν pν = 0 (5.83)

since these contractions appear as the residues of (z−w)−3 in the OPE T (z)Vm2 =0 (w).

• In general, vertex operators that can be written as total derivatives are spurious states (we will dis-
cuss in the next chapter why surface terms in presence of the Koba-Nielsen factor do not contribute
to amplitudes of both open and closed strings):
Z Z Z
d d
2 2
d z Vsp (z) = d z Dsp (z) = d2 z D̄sp (z) (5.84)
dz dz̄
• For massless states, the relevant total-derivative generator is

Dm2 =0 (z) = −iξµ : ∂z̄ X µ eip·X(z) : , ξ·p=0 (5.85)

which leads to primary fields

Z Z
d2 z Vsp,m2 =0 (z) = d2 z : pµ ξν ∂z X µ ∂z̄ X ν eip·X(z) : (5.86)

Together with the complex conjugates D̄m2 =0 (z) = −iξ¯µ : ∂z X µ eip·X(z) :, the spurious choices of
the polarization tensor for massless states are

ζµν → pµ ξν , ζµν → ξ¯µ pν (5.87)

58
 • When restricting to the symmetric-traceless component of ζµν that describes spin-two particles, the
spurious states with ξ¯µ = ξµ correspond to the shift

δζµν = pµ ξν + ξµ pν , p·ξ =0 (5.88)

This is nothing but linearized diffeomorphism symmetry – it arises from vertex operators of total-
derivative form and can ultimately be traced back to BRST-exact states (see exercise D.3). Massless
spin-two states with diffeomorphism symmetry are necessarily gravitons.

• Let us now do the state counting of (not necessarily symmetric-traceless) massless states, i.e. non-
spurious ζµν

(i) the primary-field condition pµ ζµν = ζµν pν = 0 leaves (D−1)2 states

(ii) subtract 2(D−1) states from ξµ and ξ¯µ associated with total z- and z̄-derivatives
(iii) add back 1 since pµ pν : ∂z X µ ∂z̄ X ν eip·X(z) : was subtracted twice in the previous step

In total, we arrive at (D−2)2 massless states, i.e. we have reproduced the conclusion from lightcone
quantization in a covariant way.

• The SL2 (C)-weight (−1, −1) of d2 z can be alternatively realized via insertion of ghosts c(z)c̄(z̄).
This may seem to contradict the earlier argument that specification of unintegrated punctures is at
odds with diffeomorphism invariance on the worldsheet. However, there is a loophole for punctures
that the correlation functions do not depend on: As we discussed in the context of correlation
functions, n-point functions of conformal primaries on the sphere only depend on n−3 cross-ratios.
Hence, we are free to insert a total of three “unintegrated vertex operators” at genus zero
Z
d2 z Vphys (z) → c(z)c̄(z̄)Vphys (z) (5.89)

At the same time, the three insertions of c(z)c̄(z) saturate the background charge and lead to
non-vanishing correlators of the ghost system.

For open strings, vertex operators are conformal primaries in the boundary CFT that are confined on R
and only integrated over the worldsheet boundary
Z Z
op
d2 z Vphys
cl
(z) ↔ dz Vphys (z) (5.90)
C R
op
• constituents of Vphys obey gluing conditions

 +∂ X (z ∈ R) : Neumann, ∂σ → 0
z̄ µ
∂z Xµ (z ∈ R) = (5.91)
 −∂z̄ Xµ (z ∈ R) : Dirichlet, ∂τ → 0

which were explained to be realized by a single field ∂z Xµop (z) on C

• by the mirror image in the open-string Green function

GN
op (z, w) = Gcl (z, w) + Gcl (z, w̄) (5.92)

the OPE of each ∂z X op picks up a factor of two,

op α0 pµ  1 1  ip·X op (w)
i∂z Xµop (z) : eip·X (w) : ∼ + :e : +... (5.93)
2 z−w z−w̄
op α0 p2  1 1 2 ip·X op (w)
Tµop (z) : eip·X (w) : ∼ + :e : +... (5.94)
4 z−w z−w̄

59
 This introduces an extra factor of four into the conformal weights for w ∈ R,

op α0 p2 op
Tµop (z) : eip·X (w∈R)
:∼ : eip·X (w) : + . . . (5.95)
(z−w)2

• resulting open-string vertex operators

op op op
Vphys (z) = : Wphys (z)eip·X (w)
: (5.96)

op
where Wphys is now built from ∂τn Xµop with the derivative ∂τ = ∂z + ∂z̄ in the boundary direction
op
(effectively ∂τ → 2∂z if z ∈ R). For states of mass m, the conformal weight of Wphys has to be
0 2
R op
1 + α m in order to attain a conformal primary of weight one, i.e. a conformally invariant R Vphys

N −1
m2op = , N ∈ N0 (5.97)
α0
starting with the open-string tachyon at m2op → − α10

• construct massless open-string states from an ansatz

op
Vmop2 =0 (z) = ζ µ : ∂z Xµop eip·X (z)
: (5.98)

which leads to primary fields if ζ · p = 0. As before, polarizations ζ µ → pµ descending from total

op
derivatives (here ∂τ : eip·X (z)
:) are spurious. Hence, the number of physical degrees of freedom
(after subtracting one component from transversality ζ · p = 0 and one component from the gauge
freedom δζ µ = pµ ) is D−1−1, corresponding to an abelian gauge boson.

• just as for closed strings, one can convert effective conformal weight −1 of dz into c(z),
Z
op op
Vphys (z) → c(z)Vphys (z) (5.99)

which needs to be done for 3 vertex operators per genus-zero correlation function

As mentioned earlier on, one can engineer D < 26 spacetime dimensions by adjoining an internal CFT
at central charge cint = 26 − D.

• the conformal primaries φint

h of the internal CFT serve as building blocks for vertex operators, e.g.

ip·X op (z)
op
 : φint
h (z)e : ⇒ m2op = h−1
α0
Vphys (z) → op
(5.100)
 ζ µ : ∂z X op φint (z)eip·X (z) : ⇒ m2 = h0
µ h op α

• more generally, each internal primary φint

h modifies the open- and closed-string spectra as follows

1 4
m2op = (h−1+N ) , m2cl = (h−1+N ) , N ∈ N0 (5.101)
α0 α0
in particular: we get an extra massless states for each h = 1 primary φint
h=1 (z), i.e. the spectrum of
the internal CFT has profund impact on the low-energy dynamics of the resulting string compact-
ification.

60
6 String amplitudes
The rough idea in string perturbation theory is similar to the Feynman-diagram organization of point-
particle amplitudes in field theory: Path integrate over all histories for a scattering process that are
compatible with a given configuration of external states, see figure 11. The goal of this section is to
make the form of this “path integral” more precise in the context of scattering amplitudes of open and
closed bosonic strings. Cross sections and probabilities are then obtained from the absolute-value square
of amplitudes.

Figure 11: Cartoon of a scattering process described by string amplitudes.

There are several lines of motivation to the study of string amplitudes:

• Comparison of open- and closed-string amplitudes identifies gravitational amplitudes to arise from
squares of suitably chosen gauge-theory building blocks. Hence, string amplitudes point towards
a double-copy structure of perturbative gravity. This notion of double copy grew into a vibrant
research field with various kinds of applications and extensions to non-gravitational theories, see
1909.01358 for a review.

• The α0 -expansion of string amplitudes determines the low-energy effective action of different string
theories encoding the dynamics of the massless states after integrating out the infinite tower of
massive ones. Low-energy effective actions are key quantities in testing or exploiting string dualities
and also offer a window into non-perturbative effects.

• The coefficients and building blocks in the α0 -expansion of string amplitudes triggered fruitful
interplay with number theory: The low-energy expansion of string amplitudes became a rewarding
laboratory to explore and apply special functions in a simple context, e.g. multiple zeta values,
polylogarithms, elliptic generalizations of both and modular forms. These systems of functions
appear in various other areas of high-energy physics, e.g. elliptic polylogarithms are a driving
force in the state-of-the-art evaluation of Feynman integrals in quantum field theory and therefore
precision calculations for LHC processes.

6.1 Basic ideas in string perturbation theory

The opening line for string amplitudes is to path-integrate over all (diffeomorphism × Weyl)-inequivalent
field configurations (X, h, . . .) for worldsheets that connect given external states. The analogue of the
expansion of field-theory amplitudes in the loop order of Feynman graphs is a topological expansion of
string amplitudes in the genus of the worldsheet, see figure 12

61
Figure 12: The topological expansion of string amplitudes corresponds to the expansion of field-theory
amplitudes according to the number of loops in the Feynman diagrams.

For massless external string states, a particularly important question concerns the point-particle- or
low-energy limit α0 → 0:

• The gravity states and gauge bosons at m2 = 0 dominate in this limit while the tower of massive
4(N −1) N −1
states at m2 = α0 (closed strings) or m2 = α0 (open strings) with N 6= 1 becomes infinitely
heavy and is effectively integrated out. This is similar to the effective four-fermion interaction
vertices that emerge when integrating out the W or Z bosons that couple to the Standard-Model
fermions.

• In the low-energy expansion of string amplitudes (i.e. their Taylor-expansion in α0 ), we get field-
theory amplitudes in non-abelian gauge theories and perturbative gravity at the leading order.
Subleading orders in α0 (in fact, an infinite series of α0 -corrections) signal higher-derivative operators
in an effective Lagrangian, e.g. α02 Tr(F 4 ) as a first correction to the Yang–Mills Lagrangian Tr(F 2 )
for gauge bosons and α03 R4 as a first correction to the Einstein–Hilbert Lagrangian R for gravitons.

Let us compare the worldsheet diagrams in the topological expansion of string amplitudes with “fattened
Feynman diagrams”:

• A single worldsheet can capture all the Feynman diagrams at the relevant loop order, see figure 12;
for instance, all the (2n−5)!! cubic-vertex diagrams in an n-point tree-level amplitude arise from
various degeneration limits of a single sphere worldsheet (closed strings) or disk worldsheet (open
strings).

• The interaction zone of a string worldsheet is smeared out and does not have any analogue of the
point-like collision of particles in Feynman diagrams, see figure 13. This aims to provide intuition
for the exponentially soft high-energy behaviour of string amplitudes: There are no ultraviolet

62
 divergences in loop amplitudes of the closed superstring (in contrast to the divergences expected
for supergravity amplitudes from a naive powercounting on the basis of their Feynman rules).

Figure 13: String worldsheets give rise to smeared out interaction zones in comparison with the pointlike
intersections of worldlines in Feynman diagrams.

We need to make the program “integrate over diffeomorphism × Weyl-inequivalent worldsheets” more
precise:

• By conformal transformations, the infinitely long tubes or strips for external closed- or open-string
states connecting to an interacting worldsheet are mapped to punctures, i.e. point-like defects.
These are the insertion points of vertex operators

• Conformal gauge for the worldsheet metric hαβ → e2φ ηαβ can only be attained locally (in some
open neighborhood of any given point on the surface). Globally, this is only possible at genus zero.
R
• At g > 0 loops, gauge fixing the path integral D[h] over metrics leaves a finite-dimensional integral
over “complex-structure moduli” τ . At g = 1, for instance, there is a single modulus τ ∈ C with
Im τ > 0 that describes the shape of a torus. At genus g ≥ 2, there is a total of 3g−3 complex-
structure moduli.

How does one need to weight the contributions from different worldsheet topologies to a string amplitude?

63
 • extend the Polyakov action by a topological term

Sbos [X, h] = SP [X, h] + λχ (6.1)

with Euler characteristics χ and genus g

√
Z
1
χ= d2 σ −h R = 2 − 2g (6.2)
4π
The parameter λ which sets the string coupling turns out to be the VEV of the dilaton (the trace
part of the polarization tensor ζµν of massless closed-string states). Hence, the string coupling is
not a parameter or “by-hand” input of string theories but a dynamically determined quantity.

• the first approximation for a closed-string amplitude reads

X Z Z
−Sbos [X,h] −2λ
D[X] e =e D[X] ∆F P [ĥ] e−SP [X,ĥ]
topologies
and metrics
Z Z
+ D[X] d2 τ ∆F P [ĥ(τ )] e−SP [X,ĥ(τ )] (6.3)
∞
X Z Z
+ e 2λ(g−1)
D[X] d2(3g−3) τ ∆F P [ĥ(τ )] e−SP [X,ĥ(τ )]
g=2

where τ is a collective notation for the complex-structure moduli that couple the ghosts (entering
through the Faddeev-Popov determinant) to the matter variable X

6.2 Closed-string tree-level amplitudes

The opening line for the computation of tree-level amplitudes among n closed-string states is pictorially
shown in figure 14: Instead of integrating over all insertion points z1 , z2 , . . . , zn of the vertex operators,
conformal transformations can be used to fix three of them to any desired positions on the Riemann
sphere, for instance to the convenient values (0, 1, ∞). This can be understood from the properties of the
generators ln = −z n+1 ∂z and ¯ln = −z̄ n+1 ∂z̄ of the conformal algebra with n ∈ Z.

Figure 14: The schematic idea for n-point closed-string tree-level amplitudes. The formal division through
residual conformal symmetry amounts to fixing any triplet (zi , zj , zk ) of punctures to (0, 1, ∞).

• Only the generators `1 , `0 , `−1 and `¯1 , `¯0 , `¯−1 are globally defined on the Riemann sphere C ∪ {∞},
and the finite version of the associated conformal transformations is parametrized by SL2 (C),
 
az + b a b
z→ ,   ∈ SL2 (C) (6.4)
cz + d c d

64
 • The three independent entries of the SL2 (C) matrix can be dialed to map any three zi , zj , zk to
arbitrary positions, say (0, 1, ∞). These will be the insertion points of the three unintegrated vertex
operators, Z
d2 za Vphys (za ) → c(za )c̄(z̄a )Vphys (za ) , a = i, j, k (6.5)

and we will denote the prescription to perform this replacement for any i, j, k ∈ {1, 2, . . . , n} by
(vol SL2 (C))−1 in later equations.

• The associated c-ghost correlator hc(zi )c(zj )c(zk )i = zij zik zjk can be viewed as a Jacobian relating
the coefficients η−1 , η0 , η1 of the globally defined generators to the punctures zi , zj , zk to be fixed,
 
2 ∂(δzi , δzj , δzk )
δza = −η−1 − η0 za − η1 za ⇒ det = zij zik zjk (6.6)
∂(η−1 , η0 , η1 )

• In the SL2 (C)-frame with (zi , zj , zk ) = (0, 1, ∞), one can identify the remaining variables za with
a 6= i, j, k as cross ratios,
zai zjk
za = , (zi , zj , zk ) = (0, 1, ∞) (6.7)
zji zak
With these preliminary considerations, the tree-level amplitude of closed-bosonic-string states is given by
n
d2 z1 d2 z2 . . . d2 zn
Z Y 
0
Mtree
bos ({Φi }; α ) ∼ VΦa (za ) (6.8)
vol SL2 (C) a=1

where the notation Φ1 , Φ2 , . . . , Φn for the external states collectively refers to their momenta pj and their
polarization tensors such as ζµν in case of the graviton.

• The proportionality sign comprises e−2λ and normalization factors of the vertex operators; a pro-
cedure to determine such normalizations by matching with field-theory amplitudes and unitarity
will be introduced later on.

• The bracket h. . .i enclosing the vertex operators refers to the path integral D[X] relevant to eip·X
R

and ∂zn X µ , ∂z̄m X ν entering VΦa as well as D[c] relevant to the cc̄ introduced by the SL2 (C)-fixing.
R

• The closed-string amplitude is independent on the choice of the triplet i, j, k which can be un-
derstood from the c-ghost correlators taking the role of a Jacobian, and the conformal weights
(h, h̄) = (1, 1) of each VΦa .

• The correlator of n vertex operators VΦa and three insertions of cc̄ is finite at zk → ∞: The growth
of the c- and c̄-correlators as |zk |4 is compensated by the falloff h. . . VΦk (zk → ∞)i ∼ |zk |−4 following
from VΦk being a conformal primary with (h, h̄) = (1, 1).

• Since VΦa (za ) is built from plane waves eip·X accompanied by polynomials in ∂zn X µ , ∂z̄m X ν , all
Qn 0
the correlation functions involve the Koba–Nielsen factor 1≤i<j |zij |α pi ·pj and a momentum-
Pn
conserving delta function enforcing j=1 pj = 0. These two universal building blocks are multiplied
by rational expression in the z1 , z2 , . . . , zn and z̄1 , z̄2 , . . . , z̄n which is determined by the ∂zn X µ , ∂z̄m X ν
in the vertex operators.

65
6.2.1 Three tachyons

As a first example of the above prescription for closed-string tree-level amplitudes, let us consider three-
tachyon scattering. By momentum conservation p1 + p2 + p3 = 0 and the mass-shell condition p2j =
4
−m2j = α0 , the exponents in the Koba-Nielsen factor are

α0   α0
α 0 p1 · p2 = (p1 +p2 )2 − p21 − p22 = (m21 + m22 − m23 ) = −2 (6.9)
2 2
Hence, the correlator of the vertex operators VTa ∼: eipa ·X(za ) : in the integrand is
* 3 + 3
Y
ipa ·X(za )
Y 0 1
:e : = |zij |α pi ·pj = 2
(6.10)
a=1
|z12 13 z23 |
z
1≤i<j

Together with the c-ghost correlators hcc̄(z1 )cc̄(z2 )cc̄(z3 )i = |z12 z13 z23 |2 introduced by the SL2 (C) fixing,
this cancels any dependence on the punctures,
 3
d2 z1 d2 z2 d2 z3 Y
Z 
0
Mtree
bos ({T1 , T2 , T3 }; α ) ∼ VTa (za )
vol SL2 (C) a=1
Y 3 
∼ cc̄(za ) : eipa ·X(za ) :  (6.11)

(z1 ,z2 ,z3 )→(0,1,∞)
a=1

=1

in lines with the absence of integration variables at three points (the only three punctures in the problems
can fixed to any three positions in C). The cancellation of the zj -dependence is built in since cc̄VTa are
conformal primaries of weight (h, h̄) = (0, 0).

6.2.2 Two tachyons, one graviton

With two tachyons m22 = m23 = − α40 and one external graviton with vertex operator

Vζ1 (z1 ) ∼ ζ1µν : i∂z Xµ i∂z̄ Xν (z1 )eip1 ·X(z1 ) : , p21 = 0 (6.12)

the SL2 (C)-fixed amplitude representation is

 3
Y 
Mtree 0
bos ({ζ1 , T2 , T3 }; α ) ∼ ζ1µν cc̄(z1 ):i∂z Xµ i∂z̄ Xν (z1 )e ip1 ·X(z1 )
: cc̄(za ):e ipa ·X(za )
: 

(z1 ,z2 ,z3 )→(0,1,∞)
a=2
(6.13)
 3
Y 
= |z12 z13 z23 |2 ζ1µν :i∂z Xµ i∂z̄ Xν (z1 )eip1 ·X(z1 )
: :e ipa ·X(za )
: 

(z1 ,z2 ,z3 )→(0,1,∞)
a=2

α0
By adapting α0 p1 ·p2 = 2 2 2
2 (m1 +m2 −m3 )
and permutations to the mass configuration of two tachyons and
Q3 0
one graviton, the Koba-Nielsen factor evaluates to 1≤i<j |zij |α pi ·pj = |z23 |−4 . The extra contributions
from i∂z Xµ i∂z̄ Xν (z1 ) have been determined in exercise D.4: Ignoring the chiral half with i∂z̄ Xν (z1 ) for
the moment, the answer can be assembled from the double copy of
3
α0  p2µ p3µ  µν
 Y 
ζ1µν :i∂z Xµ eip1 ·X(z1 ) : :eipa ·X(za ) : = + ζ1 |z23 |−4
a=2
2 z 12 z13

α0 z23 2 µν
= p ζ |z23 |−4 (6.14)
2 z12 z13 µ 1

66
after stripping off the Koba-Nielsen contribution |z23 |−4 . We have used that p3 = −p1 −p2 and p1µ ζ1µν = 0
in passing to the second line, and the identical calculation for i∂z̄ Xν (z1 ) with µ ↔ ν and zj ↔ z̄j
α0 z̄23 2
interchanged introduces an extra factor of 2 z̄12 z̄13 pν into the closed-string amplitude
 α 0 2 z23 2 z̄23 2 
Mtree 0
bos ({ζ1 , T2 , T3 }; α ) ∼ |z12 z13 z23 |2 p ζ1µν p |z23 |−4 

2 z z13 µ z̄ z̄ ν (z1 ,z2 ,z3 )→(0,1,∞)
| 12 {z } | 12 {z13 }
from i∂z Xµ (z1 ) from i∂z̄ Xν (z1 )
 α 0 2
= p2µ ζ1µν p2ν (6.15)
2
We have made essential use of the primary-field condition p1µ ζ1µν = 0 in demonstrating that the dependence
on the zj drops out as expected.

6.2.3 Four tachyons

As a first look at four-point amplitudes, we write the Koba–Nielsen factor in terms of the dimensionless
Mandelstam invariants

s = α0 (p1 + p2 )2 , t = α0 (p2 + p3 )2 , u = α0 (p1 + p3 )2 (6.16)

P4
where the fourth momentum can be reinstated via j=1 pj = 0, for instance s = α0 (p3 + p4 )2 . In case of
four external tachyons, the exponents are determined by permutations of α0 p1 · p2 = s
2 − 4 such that
4
Y 0 |z12 z34 |s/2 |z13 z24 |u/2 |z14 z23 |t/2
|zij |α pi ·pj = (6.17)
|z12 z13 z14 z23 z24 z34 |4
1≤i<j

In order to evaluate the amplitude in an SL2 (C) frame with (z1 , z3 , z4 ) → (0, 1, ∞), it is instructive to see
why the z4 → ∞ limit is finite: The contributions from z∗4 ∈ {z14 , z24 , z34 } to the Koba–Nielsen factor
(6.17) are equally dominant, so we do not need to distinguish them in the limit z4 → ∞,
1
|z34 |s/2 |z24 |u/2 |z14 |t/2 |z∗4 | 2 (s+t+u) 1
4
→ 12
= (6.18)
|z14 z24 z34 | |z∗4 | |z∗4 |4

In the second step, we have used that s + t + u = 16 in case of four tachyons as a consequence of p2j = α40
P4
and j=1 pj = 0. This leading contribution of z4 → ∞ is cancelled by that of the c-ghost correlator
hcc̄(z1 )cc̄(z3 )cc̄(z4 )i → |z13 |2 |z∗4 |4 , with z13 = −1 in the SL2 -frame of interest. Hence, the limit z4 → ∞
is finite, and we can effectively ignore all the contributions of z14 , z24 , z34 to (6.17) since they cancel
against the c-ghost correlator in our SL2 -frame. As a result, the four-tachyon amplitude is given by
Z 
tree 0
Mbos ({T1 , T2 , T3 , T4 }; α ) ∼ d2 z2 hVT2 (z2 )cc̄VT1 (z1 )cc̄VT3 (z3 )cc̄VT4 (z4 )i 

(z1 ,z3 ,z4 )→(0,1,∞)
ZC
∼ d2 z2 |z2 |s/2−4 |1 − z2 |t/2−4 (6.19)
C

This integral may not be in your standard toolkit, but we will see in the next lectures that it can
be evaluated in terms of Gamma functions subject to Γ(n) = (n−1)! for n ∈ N. With the analytic
continuation determined by xΓ(x) = Γ(x + 1) and the convergent integral representation for Re(x) > 0,
Z ∞
Γ(x) = dt tx−1 e−t (6.20)
0

we can evaluate the four-tachyon amplitude for various complex s, t, u subject to s + t + u = 16:
πΓ( 4s − 1)Γ( 4t − 1)Γ( u4 − 1)
Z
d2 z2 |z2 |s/2−4 |1 − z2 |t/2−4 = (6.21)
C Γ(2 − 4s )Γ(2 − 4t )Γ(2 − u4 )

67
This is the famous Virasoro–Shapiro amplitude which will be derived in detail during the first lectures of
string theory II. In fact, it will be simpler to first discuss the analogous open-string amplitudes since

• the correlator contributions from h. . . ∂zN X . . . ∂z̄N Xeip·X i do not appear “squared”

• the four-point amplitude involves a line integral “dz2 ” rather than a surface integral “d2 z2 ” which
is easier to evaluate and exhibits fewer pole channels in s, t and u

6.3 Open-string tree-level amplitudes

For open strings, the perturbative expansion is based on worldsheet topologies with boundaries, and their
tree-level amplitudes are computed from punctured disks. As visualized in figure 15, the placement of
open-string vertex operators on the disk boundary leads to a notion of cyclic ordering, where for instance
leg 2 is neighboring to legs 1 and 3 but not to 4, 5, . . . , n.

Figure 15: The schematic idea for n-point open-string tree-level amplitudes. The formal division through
residual conformal symmetry amounts to fixing any triplet (zi , zj , zk ) of punctures to (0, 1, ∞).

The choice of cyclic ordering is also crucial for the dependence of open-string amplitudes on the
Chan-Paton factors T a , i.e. the non-abelian gauge degrees of freedom: As one can see from the left
panel of figure 15, the endpoints of two neighboring open-string states are connected by segments of
the disk boundary which amounts to contracting the (anti-)fundamental indices mj , nj of the respective
gauge-group generators (T aj )mj nj . The contractions from the boundary segments in the depicted cyclic
ordering give rise to the trace Tr(T a1 T a2 . . . T an ) of gauge group generators, which we need to correlate
with the integration domain for the zj ∈ R in the right panel of figure 15.
For n external open-string states, there is a total of (n−1)! cyclically inequivalent orderings that can
be labelled by permutations ρ ∈ Sn−1 of legs 1, 2, . . . , n−1. All of them will contribute to the open-string
tree amplitude according to our program of path-integrating over all conformally inequivalent histories.
With the traces of T aj reflecting the cyclic orderings, this leads to the color decomposition
X
0 0
Mtree
open ({ϕi , ai }; α ) = Tr(T aρ(1) T aρ(2) . . . T aρ(n−1) T an )Atree
bos (ϕρ(1) , ϕρ(2) , . . . , ϕρ(n−1) , ϕn ; α )
bos
ρ∈Sn−1
(6.22)
where we have without loss of generality kept leg n in the last position of the cyclically invariant traces.
The full open-string tree amplitude Mtree
open depends on both the adjoint indices aj and the kinematic
bos

68
degrees of freedom ϕj , i.e. polarizations and momenta (which we have represented through the collective
notation Φj for closed strings). The color-ordered amplitudes Atree
bos in turn are only functions of ϕj that
depend on their cyclic ordering in the brackets.
Each cyclic ordering gives is associated with its own integration domain for the zj – with the
parametrization of the disk boundary via R, the ordering in figure 15 translates into inequalities

Atree
bos (ϕ1 , ϕ2 , . . . , ϕn ) ↔ −∞ < z1 < z2 < . . . < zn < ∞ (6.23)

Finally, the restriction of residual conformal symmetry to zj ∈ R again allows to fix any three punctures
az+b
to (0, 1, ∞). This can always be done by the globally defined conformal transformations z → cz+d
with ( ac db ) ∈ SL2 (R) that preserve the boundary z ∈ R. By picking an SL2 -frame compatible with the
inequalities of a given cyclic ordering, the prescription for color-ordered disk amplitudes is
Z n
Y 
0 dz1 dz2 . . . dzn
Atree
bos (ϕ1 , ϕ2 , . . . , ϕn ; α ) ∼ Vϕop (za ) (6.24)
vol SL2 (R) a=1
a

−∞<z1 <z2 <...<zn <∞

where the inverse vol SL2 (R) instructs to insert a c-ghost at the three punctures that are fixed to (0, 1, ∞).
op
The open-string vertex operators Vϕop
a
are constructed from eip·X and ∂zn Xµop as explained in section
5.5. The correlators are computed from the same Wick rules as in the closed-string case, and the net
effect of having Xµop in the place of Xµ is a rescaling of α0 by factors of two and four. Accordingly, we
op
will no longer display the superscripts for ease of notation. The normalization factors suppressed by
the proportionality sign ∼ will again be fixed later on.

6.3.1 Three gluons

We recall the form of the massless open-string vertex operators (with p2j = 0 and polarization vectors µj
subject to j · pj = 0)
Vj (zj ) ∼ µj : i∂z Xµ eipj ·X(zj ) : (6.25)

The color-ordered amplitude associated with Tr(T a1 T a2 T a3 ) is

0
Atree
bos (1 , 2 , 3 ; α ) ∼ hcV1 (z1 )cV2 (z2 )cV3 (z3 )i
* 3 +
µ1 µ2 µ3
Y
ipj ·X(zj )
= |z12 z13 z23 |1 2 3 : i∂Xµj e :
j=1
3
p2µ z23 p3µ2 z31 p1µ3 z12

Y 0
= |z12 z13 z23 | |zij |2α pi ·pj µ1 1 µ2 2 µ3 3 (2α0 )3 1
z12 z13 z23 z21 z31 z32
1≤i≤j

0 2 ηµ1 µ2 µ3 12
h p1 z i
+ (2α ) 2 + cyc(1, 2, 3)
z12 z31 z32
∼ (1 · 2 )(3 · p1 ) + cyc(1, 2, 3) + 2α0 (1 · p2 )(2 · p3 )(3 · p1 )
 
(6.26)

In passing to the last line, we have used that all the Koba–Nielsen exponents ∼ pi · pj vanish in the
momentum phase space of three massless particles subject to momentum conservation. In an open-string
context, the c-ghost correlator is given by |z12 z13 z23 | with absolute values. This can for instance be
understood from its role as a Jacobian, and it leads to a negative relative sign between the two color-
orderings,
0 0
Atree tree
bos (1 , 2 , 3 ; α ) = −Abos (3 , 2 , 1 ; α ) (6.27)

69
As a consequence, the accompanying color traces enter Mtree
open with a relative sign Tr([T
a1
, T a2 ]T a3 ) ∼
bos

f a1 a2 a3 involving the structure constants of the gauge group, [T a1 , T a2 ] = if a1 a2 a3 T a3 .

• At the leading order in α0 , the three-gluon amplitude (6.26) reproduces the result of the Feynman
rule in figure 16 that follows from the Lagrangian of Yang–Mills field theory

1
LYM = − Tr(F µν Fµν ) ∼ Tr(A∂ 2 A + A2 ∂A + A4 ) (6.28)
4
In the second step, we have inserted the expression Fµν = ∂µ Aν −∂ν Aµ −gYM [Aµ , Aν ] for non-linear
field strength without tracking the structure of Lorentz indices.

Figure 16: Feynman rule for the cubic vertex in Yang–Mills field theory from the term of schematic form
Tr(A2 ∂A) in the Lagrangian.

• The subleading order of (6.26) in α0 can be traced back to the Feynman rules of a higher-mass-
dimension operator α0 Tr(F 3 ), see figure 17. The non-linear parts igYM [Aµ , Aν ] of the gluon field
strength introduce additional four-, five- and six-point vertices into the Feynman rules that do not
contribute to the three-point amplitude under discussion.

Figure 17: Feynman rule for the cubic vertex from the operator Tr(F 3 ).

• Hence, the α0 -expansion of the three-point amplitude is captured by the low-energy effective La-
grangian
L open
bos
= LYM + α0 Tr(F µ ν F ν λ F λ µ ) + O(α02 ) (6.29)

and we will see from the four-point amplitude that additional operators starting from α02 Tr(F 4 )
are needed to reproduce higher-order terms in the α0 -expansion.

70
6.3.2 Four-point generalities

Before considering the specific example of four external tachyons, we note some general properties of
0
four-point open-string amplitudes. First, the Koba–Nielsen exponents |zij |2α pi ·pj have integer offsets
(depending on the external masses) in comparison to the dimensionless Mandelstam invariants

sij = α0 (pi + pj )2 = 2α0 pi · pj − α0 (m2i + m2j ) (6.30)

In case of four tachyons or four massless states, we rewrite the Koba–Nielsen factor as follows

4  Q4 sij
: four external gluons, m2j = 0
0
1≤i<j |zij |
Y
2α pi ·pj
|zij | = (6.31)
 Q4 |zij |sij −2 : four external tachyons, m2 = −1/α0
1≤i<j 1≤i<j j

which is motivated by the fact that scattering amplitudes have poles in sij and not in 2α0 pi · pj . In an
SL2 (R)-frame with (z1 , z3 , z4 ) → (0, 1, ∞), color-ordered four-point amplitudes involving generic open-
string states ϕj read
Z 1  
0
Atree
bos (ϕ1 , ϕ2 , ϕ3 , ϕ4 ; α ) ∼ dz2 lim z42 hVϕ1 (z1 )Vϕ2 (z2 )Vϕ3 (z3 )Vϕ4 (z4 )i (6.32)
0 z4 →∞

according to the general prescription in (6.24). The integration limits z2 ∈ (0, 1) implement the constraints
z1 < z2 < z3 inherited from the color ordering. The limit z4 → ∞ is finite since Vϕj are conformal
primaries of weight hj = 1, which can be verified at the level of the Koba–Nielsen factors by relations
 
such as s + t + u4 tachyons = 4 and s + t + u4 gluons = 0.

6.3.3 Four tachyons

When specializing ϕi to four external open-string tachyons, the above color-ordered four-point disk am-
plitudes becomes
Z 1
0
Atree
bos (T1 , T2 , T3 , T4 ; α ) ∼ dz2 |z2 |s−2 |1 − z2 |t−2
0
Γ(s − 1)Γ(t − 1)
= (6.33)
Γ(s + t − 2)
Γ(s − 1)Γ(t − 1)
=
Γ(2 − u)
The dz2 integral over (0, 1) is simpler to evaluate than the earlier d2 z2 integral over C seen in the closed-
string four-point amplitude. This integral can be identified as the Euler beta function
Z 1
Γ(a)Γ(b)
dz2 |z2 |a−1 |1 − z2 |b−1 = (6.34)
0 Γ(a + b)
which in turn is well-known to yield a ratio of Gamma functions
Z ∞
Γ(a) = dx xa−1 e−x (6.35)
0

subject to
Γ(a+1) = aΓ(a) , Γ(n) = (n−1)! ∀ n ∈ N (6.36)

The functional relation Γ(a+1) = aΓ(a) yields an analytic continuation to complex values of a outside
the region Re(a) > 0 of where the integral converges. This analytic continuation has simple poles at
non-positive integers a = 0, −1, −2, −3, . . . which have important physical meaning in four-point string
amplitudes.

71
 • The numerators Γ(s−1) and Γ(t−1) in the four-tachyon amplitudes reveal poles when either s =
s12 = s34 or t = s23 = s14 take values 1, 0, −1, −2, . . .. These poles signal the propagators
1 α0
(p1 +p2 )2 +m2int
= s+α0 m2int
of the state-exchange diagram in figure 18, where the mass-squares of
the internal states can be read off as m2int = − α10 , 0, α10 , α20 , . . .. This selection of internal masses
precisely matches the open-string spectrum (starting with the tachyon mass-square − α10 and cov-
N0
ering the entire massive tower at mass-squares α0 ). It is an important cross-check that no states
at unphysical masses appear in internal propagators.

Figure 18: Interpretation of the s-channel poles of the four-tachyon amplitude as an exchange diagram
with internal states of mass mint .

• There are no poles in the u-channel, the Gamma function Γ(2−u) in the denominator of the four-
tachyon amplitude only yields zeros at u = α0 (p1 + p3 )2 = 2, 3, 4, . . .. This is consistent with the
accompanying Chan-Paton factor Tr(T a1 T a2 T a3 T a4 ) having T a1 and T a3 at non-adjacent positions.

• Since all negative integers occur among the s- and t-channel poles, there must be at least one state
at each mass level that couples to two tachyons. As a next step, we infer more information about
these states by computing the residues of the poles: From the binomial expansion (1 − z2 )t−2 =
P∞ n t−2 Γ(t−1)
with t−2



n=0 (−z2 ) n n = n!Γ(t−n−1) , we can rewrite the four-tachyon amplitude as
∞ ∞
1 1
(−1)n t−2
  X

t−2
Z Z X
dz2 z2s−2 (1 − z2 ) t−2
= dz2 z2s−2 (−z2 )n
= n
0 0 n=0
n n=0
s+n−1
1 t − 2 (t−2)(t−3)
= − + + ... (6.37)
s−1 s 2(s+1)
The Mandelstam t-variable in the numerators signals contractions p2 · p3 or p1 · p4 across the
propagator in figure 18. Such contractions are only possible if the propagators contain η µν , i.e. if
the exchanged particles have spin, and the degree of the polynomial in t at the residue of the pole
in s + n imposes a lower bound of the spin. Given that (s + N )−1 is accompanied by degree-(N + 1)
polynomials, there must be spins ≥ N +1 at mass level N , e.g. spin 0,1 and 2 at mass-level −1, 0
and 1. Indeed, these bounds are saturated by the bosonic-string spectrum since the construction
of vertex operators from multiple insertions of : ∂z X µ ∂z X ν . . . : yields spins up to N +1 at level N .

• From a quantum-field-theory perspective, it may appear surprising that the infinite sum of s-
channel poles is an exact expression for the color-ordered amplitude: The latter is symmetric
under s ↔ t and also has an infinite tower of t-channel poles. By the infinitely many terms in
the pole expansion in the s-channel, the t-channel poles are automatically accounted for – the
polynomials in t of unbounded degree can be viewed as a smoking gun for the singular behaviour
as t → 1, 0, −1, −2, . . ..

72
 The astonishing property of the four-tachyon amplitude that an infinite sum over s-channel pole
incorporates the analogous poles in the (dual) t-channel is known as duality. The expression
Γ(s−1)Γ(t−1)
Γ(2−u) for the four-tachyon amplitude is the possibly first formula in the history of string
theory – it was Veneziano who recognized its remarkable duality property in 1968. The Veneziano
amplitude was initially proposed to describe hadron scattering, and its interpretation in terms of
string scattering was found later on.

6.4 Revisiting closed strings as a double copy of open strings

We shall now revisit closed-string tree-level amplitudes and identify their building blocks as double copies
of open-string quantities. The simpler part is to recognize the closed-string integrands as double copies of
open-string correlators at rescaled value of α0 . Moreover, we will also see that the double-copy structure
of tree amplitudes even persists at the integrated level: The so-called Kawai–Lewellen–Tye relations allow
to unwind the integrals d2 zj over the sphere and to replace them by independent integrations dz2
R R
R
and dz̄2 along segments of the real line, i.e. the disk boundary.

6.4.1 Integrand level

We start by demonstrating the factorization of closed-string correlation functions into holomorphic and
antiholomorphic parts, where the vertex operators take the schematic form : (∂z Xµ )n (∂z̄ Xν )n eip·X :.
op
The comparison with the analogous open-string correlators of : (∂z Xµop )n eip·X : is most conveniently
done at the level of their generating functions: According to exercise D.4, the open-string correlators are
obtained from expanding
Y n  n  n hξ · ξ 
ip·X op +iξj ·(∂z +∂z̄ )X op
Y
2α0 pi ·pj 0
X i j ξi · p j ξj · pi i
:e : = |zij | exp 2α 2 + + (6.38)
j=1
zij zij zji
1≤i<j 1≤i<j

in the formal bookkeeping vectors ξjµ ,

where (∂z +∂z̄ ) is the derivative along the boundary. The generating
functions for closed strings involve separate bookkeeping variables ξ¯j to distinguish ∂z X µ for left movers
from ∂z̄ X µ for right movers.
Y n  n  0 Xn hξ · ξ 
ip·X+iξj ·∂z X+iξ̄j ·∂z̄ X
Y
α0 pi ·pj α i j ξi · pj ξj · pi i
:e : = |zij | exp 2 + +
j=1
2 zij zij zji
1≤i<j 1≤i<j
n h ξ¯ · ξ¯ ξ¯i · pj ξ¯j · pi i
 0 X 
α i j
× exp 2 + + (6.39)
2 z̄ij z̄ij z̄ji
1≤i<j

By comparing the two types of generating functions, we observe that the correlators of the first derivatives
of Xµ and Xµop are consistent with
n
 n op
Y  D Y E 2

VΦj (zj ) =  Vϕj (zj )  (6.40)
α0 →α0 /4

j=1 j=1

The need for the rescaling of α0 is obvious for the exponentials and also plays out with the Koba–Nielsen
0 α0 pi ·pj /2 α0 pi ·pj /2
factor since |zij |α pi ·pj = zij z̄ij .
The absolute-value square is understood to act on functions
2 ¯
of zj and ξj via |f (z, ξ)| = f (z, ξ)f (z̄, ξ), and the closed-string polarizations are taken to factorize into
open-string ones according to Φj = ϕj ⊗ ϕ̄j . By the same type of arguments, the double-copy formula
for correlators also applies to higher derivatives ∂zn≥2 X, with separate bookkeeping variables ξ (2) , ξ (3) , . . .
and ξ¯(2) , ξ¯(3) , . . . for each derivative order.

73
6.4.2 Application to three massless closed-string states

At the massless level, the closed-string polarization tensors

ζjµν = µj ⊗ ¯νj (6.41)

obtained from double copies of gluon polarizations describe gravitons, B-fields and dilatons through their
Lorentz irreducibles. In the field-theory limit, B-fields and dilatons only couple in pairs to gravitons, i.e.
the tree-level scattering amplitudes corresponding to the perturbative expansion of general relativity are
unaffected by the extra states.
By the double-copy formula for closed-string correlators, we obtain the following factorized form of
three-point amplitudes
3
Y 
0
Mtree
bos ({ζ1 , ζ2 , ζ3 }; α ) ∼ cc̄Vζj (zj )
j=1
 3
DY E 2
= z12 z13 z23 µ1 ν2 λ3 : i∂z Xµopj eipj ·X (zj ) :
 

α0 →α0 /4
j=1
 2
= (2α0 )2 (1 · 2 )(3 · p1 ) + cyc(1, 2, 3) + (2α0 )3 (1 · p2 )(2 · p3 )(3 · p1 )
   

0
∼ Atree tree
1 , ¯2 , ¯3 ; α0 )
bos (1 , 2 , 3 ; α )Abos (¯ (6.42)

At the leading order in α0 , the spin-two components in this tensor product correspond to perturbative
gravity, i.e. the Feynman rules coming from the Einstein-Hilbert Lagrangian upon expansion of the
spacetime metric
p
gµν = ηµν + κhµν , κ= 32π 2 GNewton (6.43)

around Minkowski spacetime. With this form of the metic, the Einstein-Hilbert Lagrangian (with D-
dimensional Ricci scalar R) takes the schematic form
p
LEH ∼ | det g| R (6.44)
 
∞ more vertices
= h∂ 2 h + κh2 ∂ 2 h + κ2 h3 ∂ 2 h +  
of order O(κ3 )

When the Lorentz indices are spelt out, the Feynman rules for the cubic and quartic vertex due to
κh2 ∂ 2 h and κ2 h3 ∂ 2 h have 171 and 2850 terms, respectively. By the double-copy form of the closed-string
amplitude, the three-graviton amplitude due to the 171-term cubic vertex is contained in the considerably
simpler expression
 
Mtree
 
EH ({ζ 1 , ζ2 , ζ3 }) ∼ ( 1 ·  2 )( 3 · p1 ) + cyc(1, 2, 3) (¯
 1 · 
¯ 2 )(¯
 3 · p1 ) + cyc(1, 2, 3)  (6.45)
ζjµν =µ ν
j ⊗¯j

The graviton components of the α0 -corrections can be identified with higher-derivative operators of
schematic form R2 and R3 at the two- and four-derivative order beyond the Einstein–Hilbert term
p  
L closed = LEH + | det g| α0 R2 + α02 R3 + O(α03 ) (6.46)
bos

In fact, the squaring relations of three-point amplitudes are universal to all closed-string states Φj =
ϕj ⊗ ϕ̄j since they solely rely on the factorization property of the correlators:
0 0
0
Mtree tree α tree α
bos ({Φ1 , Φ2 , Φ3 }; α ) ∼ Abos (ϕ1 , ϕ2 , ϕ3 ; 4 )Abos (ϕ̄1 , ϕ̄2 , ϕ̄3 ; 4 ) (6.47)

74
6.4.3 Kawai–Lewellen–Tye relations

Given the factorized form of closed-string three-point amplitudes, one may wonder if the same is possible
in presence of non-trivial sphere and disk integrals at four points and beyond. As we will see, it is indeed
possible to also “factorize the integrals”, and the Kawai–Lewellen–Tye (KLT) relations to be spelled out
below schematically relate Z Z Z
d2 zj ↔ dzj dz̄j (6.48)
sphere disk bdy disk bdy
0
modulo phase factors exp( iπα
2 pi · pj ) from the Koba–Nielsen factor.
At four points, the KLT relations reduce arbitrary closed-string tree-level amplitudes to products of
color-ordered open-string ones
 πα0 
0 α0 α0
Mtree
bos ({Φ1 , Φ2 , Φ3 , Φ4 }; α ) ∼ − sin p1 · p2 Atree tree
bos (ϕ1 , ϕ2 , ϕ3 , ϕ4 ; 4 )Abos (ϕ̄1 , ϕ̄2 , ϕ̄4 , ϕ̄3 ; 4 ) (6.49)
2
where we note the swap of the 3rd and 4th leg in the color-orderings of the left movers ϕ1 , ϕ2 , ϕ3 , ϕ4
and the right movers ϕ̄1 , ϕ̄2 , ϕ̄4 , ϕ̄3 . The proof of the four-point KLT relations is based on a contour
deformation for the sphere integral over the unfixed puncture. A detailed derivation can be found in the
you-tube video of the lecture “string amplitudes 3” of the Sagex summer school 2019 (DESY, Hamburg),
starting from minute 38:
https://indico.desy.de/event/22450/contributions/46535/
Alternatively, you can consult the original KLT paper: “A Relation Between Tree Amplitudes of
Closed and Open Strings,” Nucl. Phys. B 269 (1986), 1-23.
For four open- and closed-string tachyons (m2j = − α40 on the closed-string side), the KLT relations
involve double copies of the rescaled Veneziano amplitude on the open-string side
Z 1
α0 s t
tree
Abos (T1 , T2 , T3 , T4 ; 4 ) ∼ dz2 |z2 | 4 −2 |1 − z2 | 4 −2
0
Γ( 4s − 1)Γ( 4t − 1)
= (6.50)
Γ(2 − u4 )
0
α
where Atree
bos (T1 , T2 , T4 , T3 ; 4 ) can be obtained by swapping t ↔ u in the arguments of the Gamma
functions. The sine function in the KLT relation can also be expressed in terms of Gamma functions
 πα0   hs i −π
sin p1 · p2 = sin π − 2 = (6.51)
2 4 Γ( 4s − 1)Γ(2 − 4s )
4
using p2j = α0 in the four-tachyon case. Upon insertion into the KLT formula, two of the Gamma functions
cancel, and one is left with the permutation-invariant expression

0 π Γ( 4s − 1)Γ( 4t − 1) Γ( 4s − 1)Γ( u4 − 1)
Mtree
bos ({T1 , T2 , T3 , T4 }; α ) ∼ × ×
Γ( 4s − 1)Γ(2 − 4s ) Γ(2 − u4 ) Γ(2 − 4t )
πΓ( 4s − 1)Γ( 4t − 1)Γ( u4 − 1)
= (6.52)
Γ(2 − 4s )Γ(2 − 4t )Γ(2 − u4 )

By the discussion in earlier lectures, the last line is the outcome of the integral
Z
s t
|z2 | 2 −4 |1 − z2 | 2 −4 (6.53)
C

i.e. from a mathematical viewpoint, KLT relations are an efficient way of evaluating sphere integrals.
In physical terms, the role of the sine function in the numerator can be understood in terms of the

75
pole structure: The bilinear in α0 -rescaled open-string amplitudes at color orderings ϕ1 , ϕ2 , ϕ3 , ϕ4 and
ϕ̄1 , ϕ̄2 , ϕ̄4 , ϕ̄3 has infinite towers of s-channel poles at s = 4, 0, −4, −8, . . . from both factors. Hence,
0
the net effect of the numerator factor sin( πα
2 p1 · p2 ) with zeros at s ∈ 4Z is to convert the double
poles of |Atree 2 tree
bos | into simple poles as expected for Mbos from state-exchange diagrams as in figure 18.
The Gamma-function representation of Mtree
bos manifests simple poles at s, t, u = 4, 0, −4, −8, . . . which
corresponds to the internal masses m2int = − α40 , 0 and m2int ∈ N
α0 of the closed-string spectrum. In contrast
to the Veneziano amplitude where the color ordering only admits s- and t-channel poles, the result for
Mtree
bos is permutation symmetric in p1 , p2 , p3 , p4 and has the same collection of poles in the s-, t- and u
channel.
The general form of four-point integrals computed by the KLT relations involves rational meromorphic
and antimeromorphic functions f (z) and g(z̄), respectively: The complex-analysis methods in the proof
of the KLT relations are universal in showing that
Z Z 1  Z 0 
d2 z |z|2s |1 − z|2t f (z)g(z̄) = − sin(πs) dz |z|s |1 − z|t f (z) dz̄ |z̄|s |1 − z̄|t g(z̄) (6.54)
C 0 −∞

6.4.4 KLT relations in field theory

A particularly important corollary of the KLT relations can be found in the field-theory limit limit of
massless amplitudes. We have seen in a three-point context that the α0 → 0 limit of graviton amplitudes
0 0
Mtree tree
bos ({ζ1 , . . . , ζn }; α ) and gluon amplitudes Abos (1 , . . . , n ; α ) in string theory reproduces those of
general relativity and Yang-Mills theory, respectively. The B-field and dilaton states in the tensor product
µj ⊗ ¯νj of massless states only couple to gravitons in pairs as mentioned before, so they do not appear in
the propagators of tree-level diagrams of perturbative gravity.
Hence, the α0 → 0 limit of the string-theory relation yields

Mtree 2 tree tree

EH ({ζ1 , ζ2 , ζ3 , ζ4 }) ∼ −(p1 +p2 ) AYM (1 , 2 , 3 , 4 )AYM (¯1 , ¯2 , ¯4 , ¯3 ) (6.55)
0
πα0
where the sine function has been replaced by its leading order in α0 , i.e. sin( πα
2 p1 ·p2 ) =
03
2 p1 ·p2 +O(α ).
The resulting inverse propagator (p1 + p2 )2 in the field-theory KLT relation can again be understood as
a necessity to cancel the massless double pole in the s-channel of |Atree 2
YM | .

Figure 19: Number of terms in Feynman diagrams that contribute to the four-graviton tree amplitude

As a key virtue of the field-theory KLT relations, they capture the net result of the gravitational
diagrams in figure 19 in an extremely compact way! Already the s-channel diagram would have 1712
terms in a naive Feynman-diagram computation, while the KLT formula for the four-graviton amplitude
is a bilinear in color-ordered Yang–Mills trees with around 15 terms each.
Together with their n-point generalizations to be discussed below, the KLT relations for the tree-level
amplitudes of general relativity are the oldest incarnation of the gravitational double copy. There are

76
similar double-copy prescriptions at loop level to assemble gravitational loop integrands from squares of
gauge-theory quantities, and a comprehensive review can be found in 1909.01358.

6.4.5 Higher-point KLT relations

One can similarly unwind the integration domain for d2 zj = dzj dz̄j at higher multiplicity up to phases
0
exp( iπα
2 pi · pj ), e.g. at five points
 πα0 
0 α0
Mtree tree
bos ({Φ1 , Φ2 , Φ3 , Φ4 , Φ5 }; α ) ∼ Abos (ϕ1 , ϕ2 , ϕ3 , ϕ4 , ϕ5 ; 4 ) sin p1 · p2
2
h  πα0  0
× sin (p1 +p2 ) · p3 Atree
bos (ϕ̄1 , ϕ̄2 , ϕ̄3 , ϕ̄5 , ϕ̄4 ; 4 )
α
2
 πα0  i
α0
+ sin p1 · p3 Atree
bos ( ϕ̄1 , ϕ̄3 , ϕ̄ 2 , ϕ̄5 , ϕ̄4 ; 4 ) + (2 ↔ 3) (6.56)
2
where +(2 ↔ 3) instructs to add the image of all the three lines under simultaneous exchange of
(p2 , ϕ2 , ϕ̄2 ) ↔ (p3 , ϕ3 , ϕ̄3 ). Upon comparison with the three- and four-point KLT formulae, the number
of sine functions is seen to grow linearly with multiplicity. Accordingly, each term in the n-point KLT
0 n−3
formula involves factors of the schematic form sin( πα

2 pi · pj )

α0
X
0
Mtree
bos ({Φ1 , Φ2 , . . . , Φn }; α ) ∼ Atree
bos (ϕ1 , ρ(ϕ2 , ϕ3 , . . . , ϕn−2 ), ϕn−1 , ϕn ; 4 )
ρ,τ ∈Sn−3
0
× Sα0 (ρ|τ )Atree α
bos (ϕ̄1 , τ (ϕ̄2 , ϕ̄3 , . . . , ϕ̄n−2 ), ϕ̄n , ϕ̄n−1 ; 4 ) (6.57)

which are organized as the entries of an (n−3)! × (n−3)! matrix indexed by permutations ρ, τ of ex-
ternal legs 2, 3, . . . , n−2. In the α0 → 0 limit, the structure propagates to gravitational amplitudes
0
πα
Mtree tree tree tree
bos → MEH with Abos → AYM and all factors of sin( 2 pi · pj ) replaced by pi · pj . This is again a
dramatic improvement in comparison to the Feynman diagrammatics of perturbative gravity, where an
infinite number of vertices arises from the expansion of the Einstein–Hilbert action around a Minkowski
background.

6.4.6 Permutation invariance and monodromy relations

Closed-string amplitudes of identical bosonic external states have to be permutation invariant functions
of (pi , ϕi , ϕ̄i ) with i = 1, 2, . . . , n. However, the above KLT formulae only manifest invariance under
permutations of a subset i = 2, 3, . . . , n−2 of the external legs. Permutation invariance involving the
remaining legs i = 1, n−1, n is based on so-called monodromy relations among color-ordered open-string
amplitudes: At four points, for instance, symmetry under (p2 , ϕ2 , ϕ̄2 ) ↔ (p4 , ϕ4 , ϕ̄4 ) can be checked by
the monodromy relations

sin(2πα0 p1 · p2 )Atree 0 0 tree 0

bos (ϕ1 , ϕ2 , ϕ4 , ϕ3 ; α ) = sin(2πα p2 · p3 )Abos (ϕ1 , ϕ3 , ϕ2 , ϕ4 ; α ) (6.58)

to be derived in exercise E.4 which imply that all color-ordered four-point amplitudes are proportional
to each other and related by a ratio of sine-functions. More generally, monodromy relations among
color-ordered n-point amplitudes leave a (n−3)!-dimensional basis that can be chosen as

Abos (ϕ1 , ρ(ϕ2 , ϕ3 , . . . , ϕn−2 ), ϕn−1 , ϕn ; α0 ), ρ ∈ Sn−3

 tree
(6.59)

in lines with the number of terms in each chiral half of the n-point KLT formula. Note that both KLT
and monodromy relations hold universally for any combination of external states Φi = ϕi ⊗ ϕ̄i since their
derivation only uses the analytic properties of the ubiquitous Koba–Nielsen factor.

77
6.5 Low-energy physics from string amplitudes
The goals of this section are to

• recover Yang–Mills and gravity from the α0 → 0 limit of string amplitudes and to infer the systemat-
ics of higher-derivative operators Tr(D2m F n ), D2m Rn in the effective Lagrangians via α0 -expansion
(with Dµ denoting the covariant derivatives w.r.t. non-linear gauge transformations or diffeomor-
phisms)

• settle the normalization of vertex operators by matching with the effective field theory and unitarity

As already illustrated in the three-point examples of earlier sections, the α0 -expansion of massless string
amplitudes can be used to reverse-engineer Feynman rules and thereby the low-energy effective Lagrangian
that reproduces their expressions. When restoring the couplings gYM and κ2 = 32π 2 GNewton of Yang–
Mills theory and general relativity in D spacetime dimensions, the contributions to the low-energy effective
Lagrangian visible from three-point string amplitudes are
1 2α0
L open = − Tr(Fµν F µν ) + gYM Tr(F µ ν F ν λ F λ µ ) + O(α02 ) (6.60)
bos 4 3
2p n α0
L closed = 2 det |g| R + (Rµνλρ Rµνλρ − 4Rµν Rµν + R2 ) (6.61)
bos κ 4
 α 0 2 o
+ (Rµν αβ Rαβ λρ Rλρ µν − 43 Rµναβ Rνλβρ Rλ µ ρ α ) + (B-field and dilaton) + O(α03 )
4
where the non-linear field strength and the expansion of the gravitational field around a Minkowski
background are
Fµν = ∂µ Aν − ∂ν Aµ − gYM [Aµ , Aν ] , gµν = ηµν + κhµν (6.62)

The operators α0 Tr(F 3 ), α0 R2 , etc. of higher mass dimension are the joint effort of integrating out the
4N
tachyon and the infinite tower of massive states at m2open ∈ N
α0 and m2closed ∈ α0 . From the viewpoint of
0
a Taylor expansion in the dimensionless quantities α pi · pj  1, all non-zero masses are very large. The
low-energy effective action corresponding to the α0 -expansion of string amplitudes follows from a path
integral over all the massive degrees of freedom denoted by ϕ̂, i.e. schematically
Z Z
exact eff
D[ϕ̂m2 6=0 ]D[ϕm2 =0 ]e−Sbos [ϕ,ϕ̂] = D[ϕm2 =0 ]e−Sbos [ϕ] (6.63)
exact
The general idea in effective field theory is to consider the exact Lagrangian (Sbos [ϕ, ϕ̂] in this case) as
eff
currently out of reach and to set the more modest goal of determining the effective Lagrangian Sbos [ϕ]
that is accurate up to a fixed order in the energy scales. In a variety of examples, the effective-field-theory
approach allows to make predictions at the LHC scale based on ordinary quantum-field-theory methods
while acknowledging the ignorance of Planck-scale physics. Conversely, one cannot probe Planck-scale
physics based on LHC experiments which can only be used to fix the parameters in the effective Lagrangian
order by order. As an example of this limitation within string amplitudes, the α0 -expansion of the factor
Γ(1 + s) in the Veneziano amplitude cannot detect its poles at s ∈ −N.

6.5.1 First look at four-point α0 -expansions

Apart from the above three-point examples, the effective interactions of open and closed bosonic strings
can be conveniently organized according to the accompanying Riemann zeta values
∞
X 1
ζn = , n≥2 (6.64)
kn
k=1

78
which are said to carry “weight” n. The massless four-gluon amplitude is a rational function of the
Lorentz invariants (i · j ), (i · pj ) and sij = 2α0 pi · pj , and the exact polarization dependence can be
worked out just as in the three-point case by going through the possible Wick contractions of the vertex
operators ∼: ∂z Xµ eip·X : (with zero, one or two pair contractions of ∂z Xµ ). Each term contains an
integral that evaluates to an Euler beta function and is proportional to

Γ(1+s)Γ(1+t)
= 1 − ζ2 st − ζ3 stu − ζ4 st(s2 + 41 st + t2 ) + ζ2 ζ3 s2 t2 u − ζ5 stu(s2 + st + t2 ) + O(α06 ) (6.65)
Γ(1+s+t)

Regardless on the power of z2 and (1−z2 ) in the integrand, one can always attain arguments 1+s, 1+t
and 1+s+t in the Gamma functions by repeatedly using their functional identity Γ(x+1) = xΓ(x). These
manipulations introduce rational functions of s and t. The α0 -expansion of the beta integral is said to
be uniformly transcendental since the powers of α0 always match the overall weight of the accompanying
zeta values. The analogous expansion for the sphere integral in four-graviton amplitudes is

Γ(1+ 4s )Γ(1+ 4t )Γ(1+ u4 )

= 1 − 2ζ3 stu − 2ζ5 stu(s2 + st + t2 ) + O(α06 ) (6.66)
Γ(1− 4s )Γ(1− 4t )Γ(1− u4 )

which shares the uniform transcendentality of the sphere integral but does not exhibit any instance of ζ2
and ζ4 .
When factoring out the beta integral (6.65) from the four-gluon amplitude, the remaining terms will
by themselves be series in α0 . The latter are determined from the correlator of four gluon vertex operators
and the rational functions in s and t that result from Γ(x+1) = xΓ(x) and may include geometric series
P∞
(1−s)−1 = k=0 sk . This separate α0 -expansion of the coefficient of (6.65) does not involve any zeta
values since the underlying Wick contractions can only produce rational numbers and is collected in the
curly bracket of
n o Γ(1+s)Γ(1+t)
0 0 tree 02
Atree
bos (1 , 2 , 3 , 4 ; α ) = Atree
YM (1 , 2 , 3 , 4 ) + α AYM+F 3 (1 , 2 , 3 , 4 ) + O(α ) (6.67)
| {z } Γ(1+s+t)
no zeta values

The leading order in α0 must reproduce the color-ordered four-point tree amplitude Atree
YM of YM field
2
theory – the vertex ∼ gYM Tr(A4 ) from the Yang–Mills Lagrangian will inevitably contribute in view of
gauge invariance. The subleading order α0 Atree
YM+F 3 arises from the Feynman diagrams including a single
insertion of α0 gYM Tr(F 3 ) (either in a cubic-vertex graph with a Yang–Mills vertex A2 ∂A and a α0 ∂ 3 A3
vertex each, or from the four-point vertex in the non-linear extension of Tr(F 3 )).
The +O(α02 ) in the curly bracket of (6.67) refer to a variety of further operators starting from
α02 Tr(F 4 ) none of which have a zeta value in their coefficient. The crossterm of Atree
YM with the subleading

term −ζ2 st in the Gamma functions signals another operator α02 Tr(F 4 ), but the latter has a ζ2 -coefficient
π2
and therefore does not mix with the α02 Tr(F 4 ) from the curly bracket. Given that ζ2 = 6 , the tensor
structure of the accompanying effective operator in the low-energy effective action is

π 2 α02  1 
L open = ... + STr Fµ ν Fν λ Fλ ρ Fρ µ − Fµν F µν Fλρ F λρ + . . . (6.68)
bos 2 4
where STr instructs to symmetrize over the orderings of the four enclosed Chan-Paton matrices T a ∈U (N ).
Also note that there are diagrams with double insertion of α0 Tr(F 3 ) contributing at the order of α02 .

79
6.5.2 Higher orders and higher multiplicity

Higher orders in the α0 -expansion of the Gamma functions are captured by the all-order formula
∞
X 
Γ(1+s)Γ(1+t) ζn n n
 n n

= exp (−1) s + t − (s+t) (6.69)
Γ(1+s+t) n=2
n

which reproduces (6.65) upon Taylor expansion, manifests uniform transcendentality at all orders in α0
and signals an infinite tower of effective interactions of the form
m
X
(α0 )w ζw1 ζw2 . . . ζwm Tr(D2(w−2) F 4 ) , wj ≤ w (6.70)
j=1
Pm
with gauge-covariant derivatives Dµ . Terms with weight j=1 wj = w take all their powers of α0 from the
0
Gamma functions and are accompanied by Atree YM in the curly bracket in (6.67), so the α -order matches
Pm
the weight of the zeta values. Higher orders in the curly brackets lead to operators with j=1 wj < w.
The α0 -expansion of the four-point function is insensitive to operators involving F n≥5 . Hence,
the latter can only be determined from higher-point amplitudes and have a schematic structure of
(α0 )w Tr(D2(w−3) F 5 ) and (α0 )w Tr(D2(w−4) F 6 ) etc.. For instance, the complete coefficient of ζ3 α03 and
ζ4 α04 in the effective action is a combination of operators

ζ3 α03 Tr(D2 F 4 + F 5 ) , ζ4 α04 Tr(D4 F 4 + D2 F 5 + F 6 ) (6.71)

The same kind of analysis applies to Riemann-curvature operators in the gravitational sector of closed-
string effective actions: The subleading terms in the expansion of the Gamma functions belong to effective
operators of the form
ζ3 α03 R4 , ζ5 α05 (D4 R4 + D2 R5 + R6 ) (6.72)

In general, the α0 -expansion of (n ≥ 5)-point amplitudes contains so-called multiple zeta values
X
ζn1 ,n2 ,...,nr = k1−n1 k2−n2 . . . kr−nr , nr ≥ 2 (6.73)
0<k1 <k2 <...<kr

which generalize the Riemann zeta values to multiple arguments nj ∈ N and are said to have weight
Pr
j=1 nj and depth r.

6.5.3 Normalization of vertex operators and amplitudes

The undetermined normalization factors in the previous sections boil down to

• one state-independent normalization constant per worldsheet topology (e.g. ND2 and NS 2 for the
disk and sphere)

• additionally one normalization gΦ for each species of vertex operators VΦ (z)

These normalization factors enter for instance the tree amplitudes of m tachyons and n−m gravitons via

0
Mtree m
bos ({T1 , . . . , Tm , ζm+1 , . . . , ζn }; α ) = NS 2 (gT ) (gζ )
n−m
(6.74)
Z 2  m n
d z1 . . . d2 zn Y ipj ·X(zj ) Y 
× :e : ζkµk νk : i∂z Xµk i∂z̄ Xνk eipk ·X(zk ) :
vol SL2 (C) j=1
k=m+1

80
In order to determine all the normalization factors, a first step is to infer ND2 , NS 2 and g , gζ for massless
vertex operators from gluon and graviton amplitudes: The α0 → 0 limits of correctly normalized three-
and four-point amplitudes have to match those of Yang-Mills and general relativity, e.g. for open strings

lim ND2 (g )3 (2α0 )2 (1 · 2 )(3 · p1 ) + cyc(1, 2, 3) = Atree

 
YM (1 , 2 , 3 ) (6.75)
α0 →0

Together with a similar matching at four points, we obtain two equations for two unknowns that are
solved by
2
g = gYM , ND 2 = (6.76)
(2gYM α0 )2
and the same logic can be used to fix gζ and NS 2 in terms of α0 and the gravitational coupling κ. For
massive vertex operators VΦ at m2 6= 0, one cannot perform a comparable field-theory match but instead
determine gΦ from unitarity: In figure 20 below, the massless four-point amplitude on the left-hand side
is taken to have a known normalization. Its residues on the kinematic poles s+N, N ≥ −1 at all mass
levels must be products of three-point trees, summed over the polarizations of the massive states ΦN at
the given level as visualized on the right-hand side. Specifically, one can immediately solve this constraint
on the N = −1 pole to determine the normalization (gT )2 of the closed-string tachyon.

Figure 20: The factorization of massless string amplitudes on massive poles can be used to constrain or
determine normalization factors of vertex operators.

6.6 High-energy scattering

While a lot of emphasis in the previous subsections has been placed on the low-energy expansion of string
amplitudes at small dimensionless Mandelstam variables s, t, we shall now consider the high-energy limit.
More specifically, we shall discuss the hard-scattering limit s, t → ∞ at fixed angle θ or s/t. The key
result is that closed- and open-string amplitudes are exponentially suppressed at large center-of-mass
energy E = − αs0 which is very different from the behaviour of gravitational tree amplitudes: Both the
p

cubic and quartic Feynman vertex from the Einstein-Hilbert action lead to a growth of the four-graviton
tree amplitude as E 2 for large energies, i.e. a quadratic divergence as E → ∞.
The main mathematical tool in exploring the high-energy regime of string tree amplitudes is the
approximation
Γ(1+x) −→ exp(x log x) , x→∞ (6.77)

which can be inserted into the universal Virasoro-Shapiro factor at s + t + u = 0,

Γ(1+ 4s )Γ(1+ 4t )Γ(1+ u4 )

 
s t u
→ exp log s + log t + log u (6.78)
Γ(1− 4s )Γ(1− 4t )Γ(1− u4 ) 2 2 2

81
The same conclusion for the high-energy limit can be obtained by performing the sphere integral in the
saddle-point approximation, i.e. by localizing
Z 
d2 z2 |z2 |s/2 |1−z2 |t/2 → |z2 |s/2 |1−z2 |t/2  (6.79)

C z2 =z∗

at the extremal point z2 = z∗ of the integrand:

∂  s
|z2 |s/2 |1−z2 |t/2  =0 ⇒ z∗ = − (6.80)

∂z2 z2 =z∗ u
Upon insertion into the Koba–Nielsen factor, we arrive at
 
s t u
|z∗ |s/2 |1−z∗ |t/2 = |s|s/2 |t|t/2 |u|−s/2−t/2 = exp log |s| + log |t| + log |u| (6.81)
2 2 2

6.6.1 Angular dependence of massless amplitudes

Figure 21: Scattering angle in massless four-point amplitudes

We shall now pick the Lorentz frame depicted in figure 21 to resolve the dependence of massless
high-energy amplitudes on the scattering angle θ. The vanishing Minkowski norm of the momenta
r r
1 s 1 s
p1 = − 0 (1, 1 , 0, 0, . . .) p3 = − − 0 (1, cos θ , sin θ, 0, . . .) (6.82)
2 α 2 α
r r
1 s 1 s
p2 = − 0 (1, −1 , 0, 0, . . .) p4 = − − 0 (1, − cos θ , − sin θ, 0, . . .)
2 α 2 α
reflects our choice of massless external states. The Mandelstam invariants t, u are expressible in terms of
s and θ:
 s  1+ cos θ sin θ 2 θ
t = α0 (p1 +p4 )2 = α0 − 0 0, , , 0, . . . = −s cos2
α 2 2 2
θ
u = −s − t = −s sin2 (6.83)
2
Upon plugging back into the high-energy approximation of closed-string amplitudes,
  
tree 0 s s 2 θ
  
2 θ 
  s
2 θ
  
2 θ 
M ({Φ1 , Φ2 , Φ3 , Φ4 }; α ) → exp log |s| − cos log s cos  − sin log s sin
 
2 2 2 2 2 2 2

  
s θ   θ  θ   θ 
= exp − cos2 log  cos2  + sin2 log  sin2 (6.84)
   
2 2 2 2 2


we can read off the exponential suppression with s = −α0 E 2 as E → ∞.

The high-energy regime of tree amplitudes is important in view of the ultraviolet properties of loop
amplitudes: Similar to Feynman integrals in field theory, genus-one amplitudes in string theory can be
obtained from integrating over one D-dimensional loop momentum `µ that correspond to a joint zero

82
 Figure 22: Schematic form of loop integrands in genus-one closed-string amplitudes

mode of ∂z X µ , ∂z̄ X µ and can be associated with one of the edges in the Feynman graphs of the α0 → 0
limit, see figure 22.
Roughly speaking, the ultraviolet behaviour of string tree amplitudes introduces an exponential damp-
ing exp(−α0 `2 ) into the loop integrand of genus-one amplitudes (a refined argument based on modular
invariance will be mentioned at the end of this course). In this way, the ultraviolet divergences that plague
loop amplitudes of general relativity and various extensions are clearly cancelled in massless closed-string
amplitudes at genus one.
Together with modular invariance, this argument to rule out ultraviolet divergences generalizes to
arbitrary genus in closed-string perturbation theory. However, it does not yet tell you about other kinds
of divergences, say from the infrared regime of massless external particles or subtle technicalities coming
from the ghost systems of various formulations of the superstring.
The pole expansion of the Veneziano amplitude has the following closed-string counterpart:
∞
Γ(1+ 4s )Γ(1+ 4t )Γ(1+ u4 ) X (t2n + lower degree in t)
∼ (6.85)
Γ(1− 4s )Γ(1− 4t )Γ(1− u4 ) n=1 s/4 + n

Term by term, the residues ∼ t2n of the poles at different mass levels exhibit horrible ultraviolet diver-
gences as E → ∞. However, the joint effort of all the terms in the infinite series over all mass levels n
gives rise to the above exponential softness. This illustrates the benefit of the infinite spectrum in string
theories which yields much milder ultraviolet properties as any finite subset of its particles.
High-energy scattering at fixed angle θ probes the internal structure of the target. While power-law
growth with s would indicate point particles, the falloff ∼ exp(−α0 ki · kj ) is a smoking gun for extended
√
objects of length scale α0 (similar to scattering off a Gaussian potential with fluctuation at spatial scales
√
of α0 ). Hence, the high-energy regime of string tree-level amplitudes reveals that strings are fuzzy at
√
length scale ≤ α0 .

7 Bosonic strings in background fields

So far, we have taken flat Minkowski R1,D−1 to be the background spacetime. But string theory is meant
to incorporate general relativity, so there must be a way of addressing a background metric Gµν (X) 6= ηµν
(as well as background electromagnetic fields Aa (X) on D branes).
In more general terms, the goal of this section on strings in background fields is to connect the
worldsheet and target-space point of view and to describe the treatment of non-trivial backgrounds at
the level of the worldsheet CFT.

83
7.1 Strings in curved spacetime
The flat-space Polyakov action has a natural uplift to curved spacetime via ηµν → Gµν (X):
√
Z
1
SP [G, X, g] = − d2 σ −gg αβ ∂α X µ ∂β X ν Gµν (X) (7.1)
4πα0
In contrast to earlier sections, we switch notation for the 2 × 2 worldsheet metric from hαβ to gαβ since
the letter h will be used for the spin-two fluctuation of the spacetime metric in the expansion

Gµν (X) = ηµν + κhµν (X) (7.2)

around the Minkowski case. These spin-two fluctuation connect with the worldsheet techniques in earlier
sections as follows:

• rewriting the partition function in terms of hµν (X),

Z


ZP [h] ∼ D[X]D[g] exp −SP [η + κh, X, g] (7.3)
Z



= D[X]D[g] exp −SP [η, X, g] exp V [h]
| {z }
action in flat spacetime

one is led to an exponentiated vertex operator

√
Z
κ
V [h] = 0
d2 σ −gg αβ ∂α X µ ∂β X ν hµν (X) (7.4)
4πα

• more precisely, we recover the integrated graviton vertex operator d2 zVζ (z) upon
R

gαβ → ηαβ , hµν (X) → ζµν eip·X (7.5)

• while a single copy of V [h] in the path integral corresponds to a single-graviton state, the insertion
of exp(V [h]) yields a coherent state of gravitons whose joint effort changes the spacetime metric
from ηµν to ηµν + κhµν

On way of dealing with the X-dependent profile of the spacetime metric Gµν (X) is based on worldsheet
perturbation theory in α0 :

• in conformal gauge gαβ → ηαβ , we have

Z
1
SP [G, X, η] = − d2 σ ∂α X µ ∂ α X ν Gµν (X) (7.6)
4πα0
For general Gµν (X), this action describes an interacting two-dimensional QFT on the worldsheet.
Actions of this type are referred to as “non-linear σ-model (NLSM)”, where the historic origin of
the name stems from the study of pions.

• expand the NLSM action around the situation where the string is localized at spacetime point x̄µ ,
√
X µ (σ) = x̄µ + α0 Y µ (σ) (7.7)
√
where α0 (with units of length) is introduced to render the fluctuation Y µ (σ) dimensionless.
Expanding to second order in Y µ , we have
√ 1
Z
0
SP [G, x̄ + α Y, η] = − d2 σ ∂α Y µ ∂ α Y ν (7.8)
4π
n √ α0 √ 3 o
× Gµν (x̄) + α0 Y λ ∂λ Gµν (x̄) + Y λ Y ρ ∂λ ∂ρ Gµν (x̄) +O( α0 )
| {z } 2 | {z }
cubic vertex quartic vertex

84
 √
We emphasize that this expansion in α0 Y amounts to perturbation theory from the two-dimensional
worldsheet point of view, not in D-dimensional spacetime as studied in the previous section!
√
• One can read off an infinite number of couplings Gµν (x̄), ∂λ Gµν (x̄), . . . from SP [G, x̄+ α0 Y, η] with
generating function Gµν (X)

1
• Suppose the target-space metric fluctuates at length scales of ∂λ Gµν ∼ r, then the perturbative
√
α0
expansion is in the dimensionless “effective coupling” r .
√
⇒ NLSM perturbation theory is reliable if r  α0 , i.e. if the metric only fluctuates on sufficiently
large length scales
√
⇒ once more, strings are seen to be fuzzy at length scales  α0 since this is where NLSM
perturbation theory breaks down

The setup of worldsheet perturbation theory admits an alternative way of recovering the Einstein–Hilbert
action of general relativity as α0 → 0:

• the classical conformal symmetry of SP [G, X, η] might get broken at the quantum level since the
interactions starting from ∂λ Gµν (x̄) yield divergences in loop diagrams which in turn introduce a
scale µ upon regularization

• in order to compute the one-loop divergence of the two-point function hY Y i or propagator, pick
Riemann normal coordinates (that were probably ubiquitous in your general-relativity course)

α0 √ 3
Gµν (X) = ηµν − Rµλνρ (x̄)Y λ Y ρ + O( α0 ) (7.9)
3
which identifies a four-point vertex Rµλνρ ∂α Y µ ∂ α Y ν Y λ Y ρ ; the resulting one-loop correction is
computed from the diagrams in figure 23 with dimensional regularization (i.e. taking a (2+)-
dimensional worldsheet) and evaluates as follows to leading orders in α0 :

Figure 23: one-loop correction to the two-point function of the fluctuation Y µ (σ) in Riemann normal
coordinates

η µν η µν √ 3
hY µ (σ)Y ν (σ 0 )i = − log |σ−σ 0 |2 + α0 + O( α0 ) (7.10)
2 
1
• the  pole in the propagator can be cancelled by the following counterterm in the interacting NLSM:

SP [G, X, η] → SP [η, X, η] (7.11)

α0
Z  1  √ 3
+ d2 σ Rµλνρ ∂α Y µ ∂ α Y ν Y λ Y ρ − Rµν ∂α Y µ ∂ α Y ν + O( α0 )
12π 

• regularization always introduces some scale µ

85
 ⇒ quantum conformal invariance requires Rµν = 0

⇒ in this case, the target-space metric solves the vacuum Einstein equations

⇒ this is yet another derivation of general relativity coming from the α0 → 0 limit of string theory
(complementary to the study of graviton amplitudes)

7.2 B-field and dilaton backgrounds

Apart from the graviton, we also found the B-field and dilaton state Φ among the massless closed-string
excitations. Extend the worldsheet action obtained from exponentiating graviton vertex operators by
√ n
Z
1
Sσ [G, B, Φ, X, g] = − 0
d2 σ −g g αβ ∂α X µ ∂β X ν Gµν (X) (7.12)
4πα
o
+ εαβ ∂α X µ ∂β X ν Bµν (X) + α0 Φ(X)R(2)

with the two-dimensional Ricci scalar R(2) as opposed to the one R in target space. While the B-field
terms in the σ-model action can again be recognized as an exponentiated vertex operator, the dilaton
term α0 Φ(X)R(2) is more tricky to justify. As will be seen later on, its tree-level terms in worldsheet
perturbation theory will cancel one-loop violations of conformal invariance coming from Gµν (X), Bµν (X).
The B-field coupling generalizes the electromagnetic worldline action dτ Aµ (X)Ẋ µ = Aµ dX µ
R R

with the pullback of the one-form A = Aµ dX µ in the integrand.

• Analogously, the pullback of the two-form B = Bµν dX µ ∧ dX ν from spacetime to the worldsheet
reproduces the B-field term of the above Sσ
Z Z Z
B = Bµν dX µ ∧ dX ν = d2 σ Bµν εαβ ∂α X µ ∂β X ν (7.13)

• Gauge variations δAµ = ∂µ α of the components of an abelian gauge-field one-form A translate into
δA = dα; therefore, gauge invariance of the worldline action follows from the Stokes theorem
Z Z 
δ A= dα = α  = 0 (7.14)

C C ∂C

As a generalization for the B-field, gauge variations δB = dC become δBµν = ∂µ Cν − ∂ν Cµ at the

level of two-form components; as a consequence, the above B-field coupling in Sσ enjoys a gauge
invariance Z Z Z
δ B= dC = C=0 (7.15)
Σ Σ ∂Σ

which corresponds to the double copy of δµ = pµ at the level of the B-field polarizations [µ ⊗ ¯ν] .

• the gauge-invariant field strength F = dA with components Fµν = ∂µ Aν − ∂ν Aµ generalizes as

follows to a three-form field strength H of the B-field

H = dB , Hµνλ = ∂µ Bνλ + ∂ν Bλµ + ∂λ Bµν (7.16)

Let us now find the necessary conditions for conformal invariance of the σ-model action Sσ in (7.12):
Instead of investigating specific one-loop divergences, we shall demand the energy-momentum tensor Tαβ
to be traceless (as implied by conformal invariance):

86
 • each spacetime field brings its own beta function, i.e. tentative departure from conformal invariance

1 1 1
hT α α i = − βµν (G)g αβ ∂α X µ ∂β X ν − 0 βµν (B)εαβ ∂α X µ ∂β X ν − β(Φ)R(2) (7.17)
2α0 2α 2

• to one-loop order in NLSM perturbation theory (i.e. first order in α0 ),

α0
βµν (G) = α0 Rµν + 2α0 ∇µ ∇ν Φ − H(µ λρ Hν)λρ + O(α02 )
4
α0 λ
βµν (B) = − ∇ Hλµν + α0 (∇λ Φ)Hλµν + O(α02 ) (7.18)
2
D−26 α0 2 α0
β(Φ) = − ∇ Φ + α0 ∇µ Φ∇µ Φ − Hµνρ H µνρ + O(α02 )
6 2 24

• consistent backgrounds must preserve Weyl invariance

βµν (G) = βµν (B) = β(Φ) = 0 (7.19)

which, by the first term of βµν (G) = α0 Rµν +. . ., extends Einstein’s equations to B-field and dilaton

• a particularly simple class of solutions is given by D = 26 Minkowski background with B = 0 and

constant (i.e. X µ -indepdendent) dilaton profile Φ(X) → Φ0
√
Z
1
Sσ [η, 0, Φ0 , X, g] = SP [η, X, g] − d2 σ −g Φ0 R(2) (7.20)
4π
The second term is topological and gives the Euler characteristics
√
Z
1
χ= d2 σ −gR(2) = 2 − 2g (7.21)
4π
where the genus g is the order in string perturbation theory. This backs up the earlier claim that
the dilaton VEV sets the string coupling.

Given that each beta function induces a field equation, the σ-model action on the worldsheet offers a new
approach to the low-energy effective action: Which spacetime action ( dD X of some Lagrangian) yields
R

β(∗) = 0 as its equations of motion?

• alternative to reconstructing spacetime effective actions from amplitudes

• α0 -expansion ↔ loop order in the σ model, e.g. general relativity represented by the Einstein-Hilbert
action R stems from the one-loop order ∼ α0 while the ζ3 α04 R4 interaction stems from four loops
in the worldsheet σ model

• we obtain β(∗) = 0 to linear order in α0 from

√
Z  
1 D −2Φ 1 µνλ µ 2(26−D) 02
Seff [G, B, Φ] = 2 d X − det G e R − Hµνλ H + 4∂µ Φ∂ Φ − + O(α )
2κ0 12 3α0
(7.22)

However, the kinetic terms in Seff do not have the canonical form and sign which can be fixed through
the following change of frame

• redefine Φ = Φ0 + Φ̃ and  
4Φ̃
G̃µν = exp − Gµν (7.23)
D−2

87
 • results in
Z 
1 D
p 1 − D−2
8Φ̃
Seff [G̃, B, Φ̃] = 2 d X − det G̃ R̃ −
e Hµνλ H µνλ
2κ 12

4 µ 4(D−1) 2 4Φ̃ 2(D−26)
− ∂µ Φ̃∂ Φ̃ + ∇ Φ̃ − e D−2 (7.24)
D−2 D−2 3α0
with gravitational coupling κ2 = κ20 e2Φ0 = 32π 2 GNewton

• original fields {G, B, Φ} are said to be in “string frame”, the redefined ones {G̃, B, Φ̃} in “Einstein
frame”

7.3 Background gauge fields and Born-Infeld

Similar to closed-string backgrounds, can exponentiate photon vertex operators on Dp branes
Z
⇒ Sendpt [A, X] = dτ Aa (X)Ẋ a , a = 0, 1, . . . , p (7.25)
∂Σ

⇒ open-string endpoints (with worldlines ∂Σ) charged under Aa

Such electromagnetic backgrounds lead to deformed Neumann boundary conditions (with Fab = ∂a Ab −
∂
∂b Aa and ∂a = ∂X a ) from imposing the boundary terms to cancel
 Z Z 
1 2 α µ a
δ − d σ ∂ X
α µ ∂ X + dτ Aa (X) Ẋ
4πα0 Σ ∂Σ
Z  
1 µ b a a dAa
= dτ − δX ∂
µ σ X + ∂ A
b a δX Ẋ − δX (7.26)
∂Σ 2πα0 dτ
Z
1  
=− dτ δXa ∂σ X a − 2πα0 F ab ∂τ Xb
2πα0 ∂Σ
In passing to the third line, we have used that δXµ vanishes for Dirichlet directions, that is why we
are left with the a-index of δXa ∂σ X a . These boundary terms necessitate a modification of Neumann
boundary conditions to
∂σ X a = 2πα0 F ab ∂τ Xb (7.27)

in the place of ∂σ X a = 0 in absence of Aa backgrounds.

Let us study the impact of the Fab -dependent Neumann boundary conditions on the propagator in
complex variables z = i(τ + σ) and z̄ = i(τ − σ):

• recall the open-string Green function at Aa = 0 subject to (∂z − ∂z̄ )GN
op (z, w) σ=0 = 0:


α0  
GN
op (z, w) = − log |z−w|2 + log |z−w̄|2 (7.28)
2

• in presence of F ab , need a matrix-valued Green function that vanishes at the boundary z = z̄

 
δab (∂z −∂z̄ ) − 2πα0 Fab (∂z +∂z̄ ) Gbc

op (z, w) σ=0 = 0 (7.29)


• this is accomplished by the following expression

ab ab
α0 ab 1 − 2πα0 F 1 + 2πα0 F
   
Gab
op (z, w) = − δ log |z−w|2
+ log(z̄−w) + log(z− w̄)
2 1 + 2πα0 F 1 − 2πα0 F
(7.30)
1
where the appearance of Fab in the denominator is understood via power-series expansion 1−2πα0 F =
P∞ 0 n n 2 c
n=0 (2πα ) F and matrix multiplication, e.g. (F )ab = Fac F b .

88
We shall now explore the leading-order contribution of F ab to the beta function:
√
• expand Sendpt [A, X] around a background configuration x̄ in X a = x̄a + α0 Y a which obeys
∂α ∂ α x̄a = Aa (x̄) = 0; then, at first order in α0 ,
√ α0
Z  
dτ α0 ∂a Ab Y a Ẏ b + ∂a ∂b Ac Y a Y b x̄˙ c

Sendpt [A, x̄ + α0 Y ] α0 = (7.31)
∂Σ 2

• integrate by parts for one of the terms in ∂a Ab Y a Ẏ b = 12 ∂a Ab Y a Ẏ b + 12 ∂a Ab Y a Ẏ b

Z Z  
a b 1 a b a b c a b
dτ ∂a Ab Y Ẏ = dτ ∂a Ab Y Ẏ − ∂a Ab Ẏ Y − ∂a (∂c Ab x̄ )Y Y ˙ (7.32)
∂Σ 2 ∂Σ

• manifest gauge invariance

√ α0
Z  
Fab Y a Ẏ b + ∂a Fbc Y a Y b x̄˙ c

Sendpt [A, x̄ + α0 Y ] α0 = (7.33)
2 ∂Σ

• the one-loop divergence in σ-model perturbation theory can be obtained from replacing
ab ab 
1 1 − 2πα0 F 1 1 + 2πα0 F
  
0
Y a (z)Y b (w) → Gab
op (z ∈ R, w ∈ R) = −α log(z−w) δ ab
+ +
2 1 + 2πα0 F 2 1 − 2πα0 F
 ab
1
= −2α0 log(z−w) (7.34)
1 − 4π 2 α02 F 2
√
• get counterterm and beta function from insertion into the term ∂a Fbc Y a Y b ẋc of Sendpt [A, x̄+ α0 Y ],
 ab
 0 1
βc (A)  α0
∼ α ∂a Fbc =0 (7.35)
1 − 4π 2 α02 F 2

The effective action defined by its equation of motion βc (A) = 0 is known as the Born-Infeld action, see
exercise F.2 Z p
SBI [A] = −Tp dp+1 ξ − det(ηab + 2πα0 Fab ) (7.36)

with tension Tp and coordinates ξ a on the (p+1)-dimensional worldvolume of the brane, which can for
instance be chosen as ξ a = X a ∀ a ≤ p.

• as explored in exercise F.2, the low-energy expansion

Z 
SBI [A] = −Tp d ξ 1 + (πα0 )2 Fab F ab
p+1
(7.37)
  
0 4 1 ab cd ab cd 06
+ (πα ) Fab F Fcd F − 2F Fbc F Fda + O(α )
2

augments the Maxwell action by an infinite series of corrections starting with 21 (tr(F 2 ))2 − 2tr(F 4 ),
where tr(. . .) denotes a trace over Lorentz indices a, b, . . .

• in the above discussion, SBI [A] arises as the one-loop beta function in the σ-model but is in fact an
exact result in case of constant Fab (i.e. for ∂a Fbc = 0)

We shall now incorporate transverse fluctuations on the brane which were related in earlier lectures to
the oscillators of
XI
φI = , I = p+1, . . . , D−1 (7.38)
2πα0

89
 • generalizes SBI [A] to the Dirac-Born-Infeld action
s
∂φI ∂φI
Z  
p+1 0 0 2
SDBI [A, φ] = −Tp d ξ − det ηab + 2πα Fab + (2πα ) (7.39)
∂ξ a ∂ξ b
I
where ηab + (2πα0 )2 ∂φ
∂ξ a
∂φI
∂ξ b
may be interpreted as the induced metric on the brane

• in the low-energy limit reduces to the action of Maxwell plus free scalars
Z    
p+1 0 2 1 ab 1 I a 04
SDBI [A, φ] = −Tp d ξ 1 + (2πα ) Fab F + ∂a φ ∂ φI + O(α ) (7.40)
4 2

Another simple extension of Born-Infeld concerns the following coupling to massless closed-string modes
s
∂X µ ∂X ν
Z  
p+1 −Φ̃ 0
SDBI [A, φ] = −Tp d ξ e − det Gµν (X) + 2πα Fab + Bab (7.41)
∂ξ a ∂ξ b

• dilaton dependence Tp e−Φ̃ ∼ e−Φ0 lines up with tree level of open-string perturbation theory

• B-field coupling dictated by gauge invariance of

Z Z
1 2 αβ µ ν
− d σ ε ∂α X ∂β X Bµν (X) + dτ Aa (X)Ẋ a (7.42)
4πα0 Σ
| {z } | ∂Σ {z }
under δBµν = ∂µ Cν − ∂ν Cµ get compensating
Ca
1 dτ Ẋ a Ca
R
varies by trf. if δAa =−
2πα0 ∂Σ 2πα0

since the action is exclusively written in terms of the gauge invariant combination Bab + 2πα0 Fab .

7.4 Point particles compactified on a circle

How does point-particle physics in some D-dimensional spacetime MD change upon compactification of
one spatial direction on a circle of radius r?

MD → MD−1 × S 1 , XD−1 ∼
= XD−1 + 2πr (7.43)

Consider general relativity with all fields independent on XD−1 :


 (D−1)-dim metric G̃µν


D-dim metric G −→ gauge field Aµ in D−1 dim (7.44)


scalar σ


The line element in terms of the (D−1)-dim fields is given as follows:

ds2 = G̃µν dX µ dX ν + e2σ (dXD−1 + Aµ dX µ )2 (7.45)

with µ, ν = 0, 1, 2, . . . , D−2 and G̃µν in Einstein frame.

• under diffeomorphisms δX µ = −Λµ (X ν ) and δXD−1 = −Λ(X ν ),

δ G̃µν = ∇µ Λν + ∇ν Λµ , δAµ = ∂µ Λ (7.46)

i.e. the (D−1)-cpt. of the diffeomorphism parameter Λµ takes the role of the gauge scalar for Aµ

90
 • interactions of G̃µν , Aµ , σ are determined by dimensional reduction of LEH , in particular by the
following behaviour of the Ricci scalar
1
R(D) = R(D−1) − 2e−σ ∇2 eσ − e2σ Fµν F µν (7.47)
4
such that the Einstein-Hilbert action becomes (recall that the fields are taken to be XD−1 -independent)
√
Z
1
SEH = 2 dD X − det G R(D) (7.48)
2κ
Z  
2πr p 1
= 2
dD−1 X − det G̃ eσ R(D−1) − e2σ Fµν F µν + ∂µ σ∂ µ σ
2κ 4
This is the action of (D−1)-dimensional Einstein gravity coupled to a U (1) gauge field and a scalar
σ. The factor of 2πr coming from the XD−1 integral effectively rescales the gravitational coupling
1 √κ .
κ, i.e. the normalization factor takes the standard form 2κ2D−1
if we set κD−1 = 2πr

• The VEV hσi controls the radius r of the circular dimension since σ sets the diagonal entry of the
metric in the (D−1)-direction. There is no potential in the reduced SEH that could select a preferred
VEV. This leaves the freedom to change r, i.e. there is no mechanism for moduli stabilization (where
moduli in general refer to massless scalars like the dilaton Φ or here the radion σ).

• A reduction similar to that of G can be performed for the B-field (we have seen a typical field-theory
Lagrangian at the first order of the earlier Seff in α0 )

⇒ another U (1) gauge field õ from Bµ,D−1

⇒ massless d.o.f. of closed bosonic strings on a circle = {G̃µν , Bµν , Aµ , õ , Φ, σ}

Let us now consider XD−1 -dependent fields, whose main features on a circle are already illustrated by a
∼ XD−1 + 2πr is implemented via Fourier expansion
scalar example. Periodicity in XD−1 =
∞  
X inXD−1
ϕ(X µ , XD−1 ) = ϕn (X µ ) exp (7.49)
n=−∞
r

where each Fourier mode ϕn (X µ ) is considered as a separate scalar field from the (D−1)-dimensional
point of view.

• reduction of the kinetic term

Z Z n o
− dD X ϕD ϕ = dD X ∂µ ϕ∂ µ ϕ + ∂D−1 ϕ∂D−1 ϕ (7.50)
∞ 
n2
Z X 
= 2πr dD−1 X ∂µ ϕn ∂ µ ϕ−n + ϕn ϕ−n
n=−∞
r2

n2
• from (D−1)-dimensional point of view: ∞ tower of massive scalars with m2 = r2 , n∈Z

⇒ different scaling of the masses as for string spectra where M 2 grow with n ∈ N0 rather than n2

• massive XD−1 -dependent states are called “Kaluza-Klein” modes – they are negligible at energies
1
 r or distances  r, that is why signals of small circles are very hard to detect
n
• the n’th mode carries charge under Aµ = Gµ,D−1 since
r

 A →A +∂ Λ
µ µ µ
δXD−1 = −Λ ⇒ (7.51)
 ϕn → exp(− inΛ )ϕn
r

91
7.5 Strings compactified on a circle and T duality
As the main result of this section, we will see that strings have a much richer interplay with circular
α0
dimensions than point particles and cannot distinguish radius r from r .
The starting point is again a D-dimensional spacetime with a circular direction XD−1 ∼
= XD−1 + 2πr.
n
• just like for point particles, need to quantize momentum pD−1 = r, n ∈ Z, e.g. to get single-valued
ipD−1 XD−1
e in the plane-wave part of vertex operators

• closed strings may wind m ∈ Z times around a circle, i.e. the admissible boundary conditions are

XD−1 (σ + 2π) = XD−1 (σ) + 2πrm (7.52)

• combining both motion and winding in the (D−1)-direction

α0 n
XD−1 (τ, σ) = xD−1 + τ + mrσ + oscillators (7.53)
r
L
• splitting into left- and right-movers XD−1 (τ, σ) = XD−1 (σ + ) + XD−1
R
(σ − ) with σ ± = τ ± σ
xD−1 α0 +
L
XD−1 (σ + ) = + pL σ + + oscillators ∼ e±inσ (7.54)
2 2
− xD−1 α0 −
R
XD−1 (σ ) = + pR σ − + oscillators ∼ e±inσ (7.55)
2 2

• momentum- and winding-numbers n, m ∈ Z2 yield the following lattice of admissible momenta

n mr n mr
pL = + 0 , pR = − 0 (7.56)
r α r α

From the (D−1)-dimensional perspective, masses only involve pµ -components µ = 0, 1, 2, . . . , D−2

• Virasoro conditions L0 = L̄0 = 1 yield modified mass formulae

4 4
M 2 = −pµ pµ = p2L + 0
(Ñ −1) = p2R + 0 (N −1) (7.57)
α α
with Ñ , N the left- and right-moving oscillator numbers

• with simultaneous winding and momentum, can violate level matching

α0 2
N − Ñ = (p − p2R ) = mn (7.58)
4 L

• eliminating the mn term, we can rewrite the mass formula as

n2 m2 r2 2
M2 = 2
+ 02 + 0 (N +Ñ −2) (7.59)
r α α
n2
and interpret the first two terms as follows: The mass2 contribution of r2 is identical to the Kaluza-
2 n2
Klein spectrum of point particles m = r2 . The mass-contribution of winding strings with tension
1
T is 2πr|m|T ; by matching with the second term in M 2 , the tension can be read off to be T = 2πα0 .

• at pL = pR = 0, recover the massless states from KK reduction of the D-dimensional matrix Gµν
and B-field,

Vζ ∼ ζµν : ∂z X µ ∂z̄ X ν eip·X : , µ, ν ≤ D−2 ↔ (D−1)-dim G & B

V± ∼ µ : (∂z XD−1 ∂z̄ X µ ± ∂z X µ ∂z̄ XD−1 )eip·X : ↔ gauge fields Aµ , õ (7.60)
Vσ ∼: ∂z XD−1 ∂z̄ XD−1 eip·X : ↔ radion scalar

92
Relate pL , pR to charges w.r.t. Aµ , õ

• consider XD−1 -dependent tachyon modes

 
L
ipL XD−1 R
+ipR XD−1 ip·X n mr n mr
Vm,n =: e e :, (pL , pR ) = + 0, − 0 (7.61)
r α r α

• can use three-point couplings

hcc̃V± (z1 )cc̃Vm,n (z2 )cc̃V−m,−n (z3 )i ∼  · (p2 − p3 )(pL ± pR ) (7.62)

to read off the charges of Vm,n w.r.t. the two gauge fields:

 p + p = 2n under V ↔ gauge field from G
L R r +
(7.63)
 pL − pR = 2mr
α0 under V− ↔ gauge field from B

While the charge of Kaluza-Klein modes w.r.t. the gauge field from the metric was already observed
for point particles, the charge of winding modes w.r.t. the gauge field from the B-field is a new
feature of strings.

We find enhanced massless spectra at special values or r, e.g.

2
4α0

2 mr 4
n = 0 = N = Ñ ⇒ M = 0
− 0 =0 if r2 =
(7.64)
α α m2
n 2
4 α 0 n2
m = 0 = N = Ñ ⇒ M2 = 2 − 0 = 0 if r2 = (7.65)
r α 4
√
The richest collection of massless states occurs at the self-dual radius r = α0 . As will be explained
below, the upshot is that

⇒ U (1) × U (1) gauge symmetry of V± enhanced to SU (2) × SU (2)

√ m+n n−m
• at r = α0 , the momenta simplify to pL → √
α0
and pR → √
α0
, this can be used to get integer
conformal weights for

L (m+n)2 R (m−n)2
h(eipL XD−1 ) = , h̄(eipR XD−1 ) = (7.66)
4 4

• setting m = n = ±1, obtain gauge fields with (V+ , V− )-charges (±1, ∓1) and vertex operators
L
± √2i 0 XD−1
V±L ∼ µ : ∂z̄ X µ e α e ip·X
: (7.67)

• setting n = −m = ±1, obtain gauge fields with (V+ , V− )-charges (±1, ±1) and vertex operators
R
± √2i 0 XD−1
V±R ∼ µ : ∂z X µ e α e ip·X
: (7.68)

• charged massless spin-one states only make sense within non-abelian gauge symmetry: The four
extra gauge fields at the self-dual radius enhance
√
r= α0 ⇒ U (1) × U (1) → SU (2) × SU (2) (7.69)

α0
As will be argued below, the exchange of “big and small circles” r → r known as T duality is a symmetry
of both string spectra and string interactions.

93
 α0
• M 2 at r → r simply has m ↔ n interchanged. Since both of m, n ∈ Z run over the same range, T
duality keeps the spectrum invariant and simply changes the interpretation of states as winding or
moving in the circular (D−1)-direction.

n2
• In the large-radius limit r → ∞, momentum modes M 2 = r2 form mass continuum as usual for
large dimensions. In the opposite limit r → 0, however, it is the winding modes which form a mass
continuum as if a large extra dimension opened up.

• T duality maps (pL , pR ) → (pL , −pR ) and in fact

L R L R
XD−1 = XD−1 + XD−1 → YD−1 = XD−1 − XD−1 (7.70)

Since the OPEs for XD−1 and YD−1 are the same, both the correlation functions of vertex operators
and the resulting amplitudes are unchanged by T duality, that is why it is also a symmetry of string
interactions.

• From the compactified Einstein-frame action

√ 
− det G
Z
D−1 −2Φ
Seff [G, B, Φ] ∼ d X 2πi e R + ... (7.71)
2κ20

the response of the dilaton to T duality must leave re−2Φ invariant; hence, the dilaton transforms
by a shift
1 α0
Φ→Φ+ log 2 (7.72)
2 r
• Given that ∂α XD−1 = εαβ ∂ β YD−1 and XD−1 ↔ YD−1 under T duality, it also interchanges ∂τ ↔ ∂σ
and therefore Dirichlet and Neumann boundary conditions:

 D(p+1) brane : T duality ⊥ brane
Dp brane → (7.73)
 D(p−1) brane : T duality k brane

• For superstrings, T duality relates two different theories (so-called type IIA ↔ type IIB, see later)
and different target-space geometries (mirror symmetry among Calabi-Yau manifolds with big im-
pact on mathematics).

8 Basics of the RNS superstring

8.1 Overview
Bosonic strings suffer from tachyons and cannot accommodate spacetime fermions (quarks, leptons, etc.),
and both of these shortcomings can be cured by superstrings.
A key ingredient of superstrings is supersymmetry (SUSY), an extension of the Poincaré symmetry
(Lorentz group and translations) that relates bosons and fermions. As will be detailed in the rest of this
course, one will have to distinguish two incarnations of SUSY

• worldsheet SUSY (in 2 dimensions), e.g. X µ (τ, σ) ↔ Ψµ (τ, σ), with a worldsheet spinor Ψµ as a
SUSY partner

• spacetime SUSY (for the purpose of this course, in 4 or 10 dimensions), relating for instance the
photon (spin 1) to a fermionic spin-1/2 state dubbed photino

94
There are roughly speaking three formulations of superstrings in Minkowski background

• Ramond-Neveu-Schwarz (RNS): manifest worldsheet SUSY, hidden spacetime SUSY

• Green-Schwarz (GS): manifest spacetime SUSY, no worldsheet SUSY, usually break SO(1, D−1)
covariance upon quantization (→ lightcone approach similar to our discussion of lightcone quanti-
zation of the bosonic string)

• pure-spinor formalism (Berkovits 2000): manifest spacetime SUSY and SO(1, 9) covariance, no
manifest worldsheet SUSY; the pure-spinor formulation is currently leading in the evaluation of
multi-loop and -leg string amplitudes

There are five superstring theories; their rough characterization involves the usual decomposition of the
closed-string spectrum into
|closedi ∈ { L: left- R: right-
movers } ⊗ { movers } (8.1)

• both L and R contain Weyl-spinors of SO(1, D−1), must decide for one chirality each4 ; depending
on the relative chirality of L vs. R, get

– same chirality of L and R: type-IIB superstrings

– opposite chirality of L and R: type-IIA superstrings

Dp branes in type IIB / IIA have odd / even p, respectively!

• only one chirality choice for open superstrings: “type I” = type IIB with spacetime filling D9 brane
+ “orientifold plane” + gauge group SO(32) (the latter arising as a consistency condition from
cancellations of gauge anomalies and UV divergences of one-loop open-string amplitudes)

• can combine left-movers of bosonic strings with right-movers of superstrings ⇒ “heterotic strings”
(het), two possible gauge groups SO(32) or E8 × E8

There is a web of dualities among superstring theories and beyond:

α0
• T duality r → r exchanges type IIA ↔ type IIB (cf. their respective Dp brane content) as well as
hetSO(32) ↔ hetE8 ×E8

1
• S duality among other things acts on the string coupling gs → gs and relates
 weak
S-duality  strong
type I at strong coupling ←→ hetSO(32) at weak
coupling (8.2)

S duality maps the type IIB theory to itself and exchanges weak ↔ strong coupling

• Since both r and gs are dynamically determined by scalar VEVs, T- and S-duality relate different
phases of the same theory.

• For type IIA and hetE8 ×E8 at strong coupling gs , an extra dimension of size `s gs emerges (with
string length-scale `s ); the overarching 11-dimensional quantum field theory is referred to as M
theory.
4 In even dimensions, a Dirac spinor is the direct sum of a left- and right-handed Weyl spinor which in turn furnish
irreducible representations of the even-dimensional Lorentz groups. The chirality or “handedness” is the eigenvalue of the
Dirac matrix γD+1 in even D, namely ±1 for left- and right-handed Weyl spinors.

95
 • The earlier theories are reached via dualities and different limits of M theory as visualized in figure
24 below; also: 11-dimensional supergravity (SUSY version of gravity) arises from the α0 → 0 limit
of M theory. Note that D = 11 is the highest dimension where gravitons can be embedded into
5
supersymmetry multiplets without running into spin ≥ 2 states.

Figure 24: The web of dualities relating different superstring theories

8.2 The RNS worldsheet action

Before gauge fixing, the Polyakov action for bosonic strings could be thought of as general relativity
coupled to bosonic matter X µ on a two-dimensional worldsheet.
We shall now introduce the RNS superstring as two-dimensional supergravity coupled to SUSY matter
(X , Ψµ ) on a worldsheet. More specifically, worldsheet SUSY relates
µ

metric gαβ ↔ “gravitino” χα (8.3)

bos. matter X µ ↔ fermionic matter Ψµ

The RNS generalization of the Polyakov action in flat spacetime reads as follows:
Z 
1 p 1 µ
SRNS [X, Φ, g, χ] = d2 σ − det g − 0 g αβ ∂α Xµ ∂β X µ + Ψ ρα ∇α Ψµ
4π α
 
∂β Xµ 1
+ (χα ρβ ρα Ψµ ) √ + χ β Ψµ (8.4)
2α0 8

• Ψµ and χα are two-component Dirac spinors on a two-dimensional worldsheet (Dirac spinors have
2D/2 components in D ∈ 2N dimensions) for each choice of µ = 0, 1, . . . , D−1 and α = 0, 1

• two-dimensional Dirac “gamma”-matrices subject to the Clifford algebra

{ρα , ρβ } = ρα ρβ + ρβ ρα = 2g αβ 12×2 (8.5)

0 1
where a possible choice of explicit representations is ρ0 = ( −1 1 01
0 ) and ρ = ( 1 0 )

96
 • the above SRNS is again invariant under σ α -diffeomorphisms, but also under local SUSY (with
spinorial parameter ε(σ))

δε gαβ ∼ ερ(α χβ) , δε X µ ∼ εΨµ (8.6)

∂α X µ
 
1
δε χα ∼ ∇α ε , δ ε Ψµ ∼ ρ α √ + χα Ψµ ε
2α0 4

• SRNS is Weyl invariant if the transformations are extended as follows to fermions (while δφ X µ = 0):
φ φ
δφ gαβ = 2φgαβ , δφ Ψµ = − Ψµ , δφ χα = χα (8.7)
2 2

• finally, Weyl invariance in fact extends to super-Weyl invariance with spinorial transformation
parameter ζ
δζ χα = ρα ζ , δζ (other fields) = 0 (8.8)

The above symmetries can be used to pick superconformal gauge, i.e. the SUSY analogue of conformal
gauge,
(gαβ , χγ ) → (e2φ ηαβ , ργ ζ) (8.9)

where both of φ and the worldsheet spinor ζ can be removed via super-Weyl transformations.

• gauge-fixed action in terms of z = i(τ +σ) and z̄ = i(τ −σ)

Z  
1 2 2 µ µ µ
SRNS [X, ψ, ψ̄] = d z ∂z Xµ ∂z̄ X + ψµ ∂z̄ ψ + ψ̄µ ∂z ψ̄ (8.10)
2π α0
µ
ψ
with Majorana-Weyl basis Ψµ = ( ψ̄ µ )

• the equations of motion imply ψ µ and ψ̄ µ to be meromorphic and antimeromorphic, respectively

∂z̄ ψ µ = 0 = ∂z ψ̄ µ (8.11)

• superconformal gauge can only be attained locally, there are subtleties for higher-genus super-
Riemann surfaces that cause challenges in the evaluation of higher-loop amplitudes

• the residual gauge symmetry is formed by diffeomorphisms and SUSY transformations that can be
undone via (super-)Weyl transformations which amount to superconformal symmetry

8.3 RNS superstrings and superconformal field theory

Recall that conformal transformations are generated by a traceless energy-momentum tensor Tαβ =
√ 4π δS
. In two dimensions, the latter is characterized by one meromorphic and one antimeromorphic
− det g δg αβ
component each I I
dz dz̄
Q(η) = T (z)η(z) , Q̄(η̄) = T̄ (z̄)η̄(z̄) (8.12)
BR (0) 2πi BR (0) 2πi
In superconformal theories, additionally define a supercurrent Gα via gravitino variation
1 δS
Gα ∼ √ (8.13)
− det g δχα
In applications to the RNS worldsheet action, χα and Gα are two-component Dirac spinors on the
worldsheet for both values α = 0, 1 of their vector index.

97
 • the e.o.m. of X µ , ψ µ , ψ̄ µ imply conservation laws

∂ α Tαβ = 0 = ∂ α Gα (8.14)

• in the same way as conformal symmetry implies g αβ Tαβ = 0, superconformal symmetry implies
“γ-tracelessness” ρα Gα = 0

• left with 2 instead of 4 independent supercurrent components, one meromorphic G(z) and one
antimeromorphic Ḡ(z̄) by the conservation laws

• accordingly, the conserved charges are characterized by one meromorphic and one antimeromorphic
function ε(z) and ε̄(z̄)
I I
dz dz̄
Q(ε) = ε(z)G(z) , Q̄(ε̄) = ε̄(z̄)Ḡ(z̄) (8.15)
BR (0) 2πi BR (0) 2πi

From SRNS before fixing superconformal gauge, one can derive the explicit form
1 1
T (z) = − : ∂z Xµ ∂z X µ : + : (∂z ψµ )ψ µ :
α0 2
1
G(z) = √ : i(∂z Xµ )ψ µ : (8.16)
2α0
• in order to determine SUSY transformations of X µ and ψ µ , first get relevant OPEs

• usual path-integral methods give rise to the two-point function

η µν
hψ µ (z)ψ ν (w)i =
z−w
η µν
hψ̄ µ (z̄)ψ̄ ν (w̄)i = (8.17)
z̄ − w̄
hψ µ (z)ψ̄ ν (w̄)i = 0

that reflects the anticommuting nature of ψ µ

• resulting OPEs are (with ψ µ ↔ ψ̄ ν non-singular)

η µν η µν
ψ µ (z)ψ ν (w) ∼ + ... , ψ̄ µ (z̄)ψ̄ ν (w̄) ∼ + ... (8.18)
z−w z̄ − w̄

• by Wick rules, can reduce the OPEs of T and G to those of X µ and ψ µ :

i∂w X µ (w) i∂w 2
X µ (w)
T (z)i∂w X µ (w) ∼ 2
+ + ...
(z − w) z−w
ψ µ (w) ∂w ψ µ (w)
T (z)ψ µ (w) ∼ 2
+ + ... (8.19)
2(z − w) z−w
r
2 µ ψ µ (w) ∂w ψ µ (w)
G(z) i∂w X (w) ∼ + + ...
α0 2(z − w)2 2(z − w)
r
i∂w X µ (w) 2
G(z)ψ µ (w) ∼ + ...
2(z − w) α0

• can read off (h, h̄) = ( 21 , 0) for ψ µ from T (z)ψ µ (w)

• OPE G(z) ↔ ∂w X µ (w) integrates to

r
2 ψ µ (w)
G(z) 0
iX µ (w) ∼ + ... (8.20)
α 2(z − w)

98
 • resulting superconformal transformations
r r
2 µ 1 1 2
Q(ε) i 0
X (w) = ε(w)ψ µ (w) , µ
Q(ε) ψ (w) = i∂w X µ (w)ε(w) (8.21)
α 2 2 α0

• by iterating the above Wick contractions, infer a representation of the super-Virasoro algebra
3D/4 2T (w) ∂w T (w)
T (z)T (w) ∼ 4
+ 2
+ + ...
(z − w) (z − w) z−w
3G(w) ∂w G(w)
T (z)G(w) ∼ 2
+ + ... (8.22)
2(z − w) z−w
D/4 T (w)
G(z)G(w) ∼ 3
+ + ...
(z − w) 2(z − w)
3D 3
with central charge c = 2 ; in other words, each spacetime dimension contributes 2 units of central
µ 1 µ
charge (1 from X and 2 from ψ ), and the factor of D/4 along with the triple pole in the OPE of
G(z)G(w) generalizes to c/6

The above OPEs line up with those in a general two-dimensional superconformal field theory (SCFT)
which are often defined without any reference to an action functional.

• superconformal primaries at weight h are defined as a pair of conformal primaries φh and φh+1/2
of weights h and h + 12 , respectively

• the defining properties of superconformal primaries (φh , φh+1/2 ) include the OPEs

hφh (w) ∂w φh (w)

T (z)φh (w) ∼ 2
+ + ...
(z − w) z−w
(h + 12 )φh+ 12 (w) ∂w φh+ 21 (w)
T (z)φh+ 12 (w) ∼ + + ... (8.23)
(z − w)2 z−w
φh+ 21 (w)
G(z)φh (w) ∼ + ...
2(z − w)
hφh (w) ∂w φh (w)
G(z)φh+ 21 (w) ∼ 2
+ + ...
(z − w) 2(z − w)
q
0
• by comparing with the OPEs of X µ , ψ ν , the pair of (ψ µ , i α2 ∂z X µ ) is a superconformal primary
1
of weight h = 2

• the general form of the SUSY transformation is

1
Q(ε) φh (z) = ε(z)φh+ 12 (z) (8.24)
2
1
Q(ε) φh+ 12 (z) = ε(z)∂z φh (z) + h(∂z ε(z))φh (z)
2

• T (z) and G(z) are generators of conformal transformations and SUSY transformations, respectively;
T (w)
given that G(z)G(w) contains 2(z−w) among its singular terms, one can think of Q(ε) at globally
defined ε(z) as the square root of a conformal transformation

8.4 (Anti-)periodic boundary conditions

Since the spatial coordinate σ is taken to range from 0 to 2π for closed strings, we need to find a fermionic
analogue of the boundary condition X µ (σ + 2π) = X µ (σ). The guiding principle is to avoid boundary

99
terms in varying SRNS and deriving the e.o.m. ∂z̄ ψ µ = ∂z ψ̄ µ = 0,
Z σ=2π
!
δSRNS [X, ψ, ψ̄] ⇒ dτ (ψ µ δψµ − ψ̄ µ δ ψ̄µ )  =0 (8.25)

σ=0

For this purpose, it is sufficient to impose periodic boundary conditions up to a sign

ψ µ (σ + 2π) = ±ψ µ (σ) , ψ̄ µ (σ + 2π) = ±ψ̄ µ (σ) (8.26)

since physical quantities are always bilinear in fermionic variables. In fact, the two signs for ψ µ and ψ̄ µ
can be chosen independently.

• different sign choices in the fermionic boundary conditions are referred to as follows

ψ µ antiperiodic in σ ←→ Neveu-Schwarz sector (NS)

ψ µ periodic in σ ←→ Ramond sector (R) (8.27)

• together with the analogous choices for ψ̄ µ , we arrive at a total of 4 closed-string sectors (NS,NS),
(NS,R), (R,NS), (R,R)

• supercurrents G ∼ ∂z Xµ ψ µ and Ḡ ∼ ∂z̄ Xµ ψ̄ µ inherit their boundary conditions from ψ µ , ψ̄ µ

You already got a first glimpse at fermionic mode expansions in exercise C.2:

• on the cylinder with z = i(τ + σ) and z̄ = i(τ − σ),

 X

 ψrµ e−rz : NS, antiperiodic on cylinder

µ r∈Z−1/2
ψ (τ, σ) = X (8.28)


 ψrµ e−rz : R, periodic on cylinder
r∈Z

• OPEs can be equivalently formulated in terms of anticommutators, namely {ψrµ , ψ̄sν } = 0 and

{ψrµ , ψsν } = η µν δr+s,0 = {ψ̄rµ , ψ̄sν } (8.29)

∂ξ
• in mapping to the complex plane ξ = ez with ∂z = ξ, the primaries ψ µ and ψ̄ µ transform with
weight (h, h̄) = ( 21 , 0) and (0, 12 ), respectively
 X
 −1/2 
 ψrµ ξ −r−1/2 : NS, periodic on plane
µ ∂ξ µ

r∈Z−1/2
ψ (ξ) = ψ (τ, σ) = (8.30)
∂z
X


 ψrµ ξ −r−1/2 : R, antiperiodic on plane
r∈Z

∂ξ −1/2
• transformation factor ( ∂z ) reverses periodicity properties under σ → σ + 2π or ξ → e2πi ξ, i.e.

NS ←→ antiperiodic on the cylinder but periodic on the plane

R ←→ periodic on the cylinder but antiperiodic on the plane (8.31)

• henceforth work on the plane and rename ξ → z

As will be explained below, the Ramond-sector states in D dimensions exhibit a “vacuum energy” or an
D
offset in conformal weights by h0 = 16

100
 • mode expansion of the energy-momentum tensor

1
P
1X r∈Z−1/2 r : ψm−r · ψr : (NS)

2
Lm = : αm−n · αn : + (8.32)
2 1
P
n∈Z
 r : ψm−r · ψr : +δm,0 h0
2 r∈Z (R)

3D
• need to dial the offset h0 in L0 of the Ramond sector to get the same Virasoro algebra at c = 2
as in the NS sector
! D
[Lm , Ln ] = (m−n)Lm+n + m(m2 −1)δm+n,0 (8.33)
8
D
which uniquely fixes h0 = 16 as the R-sector vacuum energy

• take NS- and R-sector ground states |0iNS and |AiR to be annihilated by ψr>0 and hence by Lm>0

• by the state-operator map in CFTs, |AiR must stem from some conformal primary at infinite past
z → 0 acting on the vacuum; the primaries relevant to |AiR are so-called spin fields SA (z) of weight
D
h= 16 ,
D/8
|AiR = lim SA (z)|0iNS , R hB| = lim NS h0|SB (z)z (8.34)
z→0 z→∞

Let us elaborate on the basic properties of spin fields:

• R ground states are degenerate since the D zero modes ψ0µ do not change the energy, [L0 , ψ0µ ] = 0

• zero modes form a Clifford algebra like the Dirac Γ-matrices in D dimensions

{ψ0µ , ψ0ν } = η µν ↔ {Γµ , Γν } = 2η µν (8.35)

Hence, |AiR and therefore SA (z) must be spinors of SO(1, D−1), and all the R-sector states (ob-
µ
tained from ψr<0 action on |AiR ) are spacetime fermions!

• in even dimensions D = 2n, Dirac spinors have 2D/2 = 2n components and decompose into two
Lorentz-irreducibles, so-called Weyl spinors with 2D/2−1 components each; at the level of the spinor
indices, decompose A = (α, α̇), e.g. in ten dimensions

D = 10 ⇒ ˙
A = 1, 2, . . . , 32 = α = 1, 2, . . . , 16 ⊕ α̇ = 1̇, 2̇, . . . , 16 (8.36)
| {z } | {z } | {z }
Dirac spinor left-handed Weyl right-handed Weyl

We next infer the OPEs of spin fields

• their two-point functions must be ∼ (z − w)−D/8 since SA are primaries of h = D

16 , and the nor-
malization factor must be ∼ CAB (the so-called charge conjugation matrix) by Lorentz-covariance
CAB CAB
hSA (z)SB (w)i = ⇒ SA (z)SB (w) ∼ + ... (8.37)
(z − w)D/8 (z − w)D/8

• use the corresponendce ψ0µ ↔ √1 Γµ

2
at the level of matrix elements:

µ 1 1
R hA|ψ0 |BiR = √ (Γµ C)AB = √ (Γµ )A D CDB (8.38)
2 2
this must be reproduced from the following limit of a three-point function

µ D/8 1/2
R hA|ψ0 |BiR = lim NS h0|z1 SA (z1 ) lim z2 ψ µ (z2 ) SB (0)|0iNS (8.39)
z1 →∞ z2 →0 | {z }
|Bi
| {z }| {z }
R
R hA| ψ0µ

101
 as a three-point function of conformal primaries, it is determined by the hj up to normalization;
the only normalization choice compatible with R hA|ψ0µ |BiR = √1 (Γµ C)AB
2
is

1 (Γµ C)AB
hSA (z1 )ψ µ (z2 )SB (z3 )i = √ D/8−1/2
(8.40)
2 (z12 z23 )1/2 z13

• the z1 → z2 limit of the relabelled three-point function

1 (Γµ C)AB
hψ µ (z1 )SA (z2 )SB (z3 )i = √ D/8−1/2
(8.41)
2 (z12 z13 )1/2 z23

together with the above expression for hSA (z2 )SB (z3 )i leads to the OPE

(Γµ )A B SB (w)
ψ µ (z)SA (w) ∼ √ + ... (8.42)
2(z − w)1/2

the square-root dependence ∼ (z − w)−1/2 is expected: SA (0) opens up a branch cut for ψ µ to
enforce the R-sector condition ψ µ (e2πi z) = −ψ µ (z) on the plane; the (ψ µ , SA )-system turns out to
be interacting, i.e. correlators with insertions of both ψ µ and SA cannot be computed by simply
adding up Wick contractions (at least, not in a manifestly Lorentz-covariant way)

• the z2 → z3 limit of the above three-point function together with hψ µ (z1 )ψ ν (z2 )i = η µν /z12 deter-
mine the subleading term in
CAB CAB (Γµ C)AB ψµ (w)
hSA (z)SB (w)i = ⇒ SA (z)SB (w) ∼ + √ + . . . (8.43)
(z − w)D/8 (z − w)D/8 2(z − w)D/8−1/2

• we will later on impose vertex operators to have no mutual branch cuts after combining matter and
ghost sectors; this will lead to the GSO projection that removes the tachyon from the superstring
spectrum and eliminates one of the Weyl components from the spin fields, i.e. all spacetime fermions
have definite chirality

8.5 The β, γ superghost system

We have seen for bosonic strings that the anticommuting (b, c)-ghost system removes the gauge redun-
dancy of the Polyakov action. In conformal gauge, the ghost contribution to the worldsheet action was
found to be Z
1
d2 z b∂z̄ c + b̄∂z c̄


Sgh [b, c] = (8.44)
π
We shall now adjoin commuting β, γ ghosts to fix the fermionic symmetries (i.e. worldsheet supersym-
metry and super-Weyl transformations) of the RNS action; in superconformal gauge, this results in the
following extra term in the action
Z
1
d2 z β∂z̄ γ + β̄∂z γ̄


Ssgh [β, γ] = (8.45)
π
where the choice of NS / R boundary conditions must be the same for β, γ ↔ ψ µ and separately for
β̄, γ̄ ↔ ψ̄ µ .
The (β, γ)-system is yet another example of a CFT with the following data

• the above Ssgh [β, γ] implies non-singular OPEs β(z)β(w) and γ(z)γ(w) as well as
1
γ(z)β(w) ∼ + ... (8.46)
z−w

102
 • the meromorphic component of the energy-momentum tensor of the superghost system is

3 1
Tβ,γ = − : (∂z γ)β : − : γ∂z β : (8.47)
2 2
3
which identifies β and γ to be conformal primaries of weights hβ = 2 and hγ = − 12 , respectively
(cf. hb = 2 and hc = −1 for the (b, c)-system); moreover, the central charge of the (β, γ)-system is
found to be cβ,γ = 11 (whereas cb,c = −26)

• the critical dimension of the RNS superstring is determined by the vanishing of

3D
ctotal = cX + cψ + cb,c + cβ,γ = − 15 (8.48)
2
this is the case in D = 10 dimensions, though one can again adjoin internal or non-geometric CFT
sectors with cint > 0 which allow for D < 10 large spacetime dimensions

3
• the pairs (β, b) and (c, γ) form superconformal primaries of h = 2 and h = −1 w.r.t. the supercurrent

3 1
Ggh = − : (∂z β)c : − : β∂z c : + : bγ : (8.49)
2 2

Similar to the (b, c)-system, the counting of zero modes and the notion of hermitian conjugates is not
completely intuitive for the (β, γ)-system:

• just like jb,c = − : bc :, there is an anomalous superghost current

jβ,γ (z) = − : β(z)γ(z) : (8.50)

• the departure from being a conformal primary

2 jβ,γ (w) ∂w jβ,γ (w)

Tβ,γ (z)jβ,γ (w) ∼ 3
+ 2
+ + ... (8.51)
(z − w) (z − w) z−w

identifies the background charge Qβ,γ = +2 (in the place of Qb,c = −3 of the (b, c)-system)

• again, the role of the jβ,γ -charge in the hermiticity properties is unusual

|β−γ charge qi† = hβ−γ charge (−q−2)| (8.52)

• the numbers of zero modes Nγ , Nβ depend on the genus g

Nγ − Nβ = −Qβ,γ (g − 1) = 2 − 2g (8.53)

generalizing Nc − Nb = 3 − 3g

Let us now construct the superghost ground states in the NS and R sectors

γr z −r+1/2 only γr≥3/2 annihilate the ground state

P
• in the NS mode expansion γ(z) = r∈Z−1/2
|0iβ,γ but not γr≤1/2

• by [L0 , (γ1/2 )n ] = − n2 (γ1/2 )n , the fact that γ1/2 |0iβ,γ 6= 0 admits states (γ1/2 )n |0iβ,γ of arbitrarily
negative L0 -eigenvalues, there is no analogue of the bound from (c1 )2 = 0 from the (b, c)-ghost
analogue c1 |0ib,c 6= 0

103
 • cure the unbounded spectrum by engineering an alternative ground state via change of variables
(β, γ) → fermions (η, ξ) + chiral boson φ

β(z) =: e−φ(z) : ∂z ξ(z) , γ(z) =: eφ(z) : η(z) (8.54)

this reparametrization preserves the OPEs of β, γ as one can check from the non-singular OPEs
ξ(z)ξ(w) and η(z)η(w) as well as
1
η(z)ξ(w) ∼ + ... , φ(z)φ(w) ∼ − log(z − w) + . . . (8.55)
z−w

• the energy-momentum tensor of the new variables reads

1
Tη,ξ = − : η∂z ξ : , Tφ = − : (∂z φ)∂z φ : −∂z2 φ (8.56)
2
(with the last term ∼ ∂z2 φ reflecting the background charge) and assigns
q2
h(η) = 1 , h(ξ) = 0 , h(eqφ ) = −
−q (8.57)
2
Qn
• by the same techniques that determine the Koba-Nielsen factor h j=1 : eipj ·X(zj ) :i,

: eq1 φ(z) : : eq2 φ(w) :∼ (z − w)−q1 q2 : e(q1 +q2 )φ(w) : + . . . (8.58)

dz γ(z)
H
• use this to construct the following state which is annihilated by γ1/2 = 2πi z

1
|q= − 1iβ,γ =: e−φ(0) : |0iβ,γ , h(e−φ ) = (8.59)
2
this qualifies for a NS-sector ground state since γ1/2 |q= − 1iβ,γ = 0

• similarly, the R-sector ground state must be annihilated by the γ1 mode and open a branch cut for
β, γ in the same way as SA (z) does for ψ µ (z)
3
|q= − 12 iβ,γ =: e−φ(0)/2 : |0iβ,γ , h(e−φ/2 ) = (8.60)
8
i.e. one can think of e−φ(0)/2 as a “spin field for superghosts”

8.6 The superstring spectrum and GSO projection

Recall that the physical states of the bosonic string are generated by vertex operators Vphys , conformal
primaries of h = 1,
|physi = lim Vphys (z)|0i (8.61)
z→0

We will now discuss the rules for finding similar h = 1 primaries for physical states of the RNS superstring.
For simplicity, we only consider the left movers (with insertions of ∂z X µ , ψ µ and their derivatives in the
matter sector) while taking the right-moving contributions such as ∂z̄ X µ , ψ̄ µ to follow from double copy.
From the discussion of the (β, γ)-system, we build the NS- and R-sector spectra on the respective
ground states |q= − 1, − 12 iβ,γ :
µ µ
• NS sector ←→ α−N , ψ1/2−N acting on |q= − 1iβ,γ =: e−φ(0) : |0iβ,γ ; the oscillators only give rise
to vector indices of SO(1, D−1), i.e. no spinor indices; hence, the resulting physical states are
spacetime bosons with vertex operators

NS −φ(z) 1
Vphys (z) = ΦNS
phys (z) : e :, need h(ΦNS
phys ) = (8.62)
2

104
 −φ(z)
where we separate off the matter contribution ΦNS
phys to not have the ubiquitous : e : in all the
subsequent formulae
µ µ
• R sector ←→ α−N , ψ−N acting on SA (0)|q= − 21 iβ,γ = SA (0) : e−φ(0)/2 : |0iβ,γ ; the spinor index of
SA (0) implies that all the resulting physical states are spacetime fermions with vertex operators
R −φ(z)/2 5
Vphys (z) = ΦR
phys (z) : e :, need h(ΦR
phys ) = (8.63)
8
which ΦR
phys (z) depending only on matter variables; the fractional conformal weight can be easily
attained since SA in D = 10 dimension has h(SA ) = 58 .

• the closed-superstring spectrum is built from the following combinations of left- and right-movers

(NS,R) and (R,NS) ←→ spacetime fermions

(NS,NS) and (R,R) ←→ spacetime bosons

where the bispinors in the (R,R) sector will be later on explained to yield antisymmetric tensors

As for bosonic strings, each Φphys will have a plane-wave factor : eip·X : of conformal weight (h, h̄) =
0 2 0 2
µ µ
( α 4p , α 4p ). Since α<0 and ψ<0 can only increase h of : eqφ : and : eip·X :, we have the following bounds
on m2 = −p2 
 −2 : NS sector
α0
m2 ≥ (8.64)
 0 : R sector

• there is a potential bosonic tachyon at mass level −1/2,

2
ΦNS
tachyon (z) =: e
ip·X(z)
:, p2 = = −m2 (8.65)
α0
µ
• massless bosons from a single ψ−1/2 oscillator

ΦNS µ
m2 =0 (z) = µ ψ (z) : e
ip·X(z)
:, p2 = 0 (8.66)

• massless fermions characterized by spinor wavefunction χA

ΦR A
m2 =0 (z) = χ SA (z) : e
ip·X(z)
:, p2 = 0 (8.67)

• the next task is to find consistency conditions to eliminate the tachyon and to constrain the polar-
izations µ , χA of massless states; in contrast to the bosonic string, transversality p ·  = 0 does not
follow from imposing a fixed conformal weight on µ ψ µ : eip·X(z) :

A central constraint on the physical spectrum of the superstring is known as GSO projection referring
to Gliozzi, Scherk and Olive. Its selection rules can be summarized as “OPEs among physical vertex
operators must not have any branch cuts”.

• plane-wave correlators are single-valued for any collection of momenta pi by the absolute values in
Qn Qn 0
the Koba–Nielsen factor h j=1 : eipj ·X(zj ) :i ∼ i<j |zij |α pi ·pj
NS
• OPEs among Vphys (z) and VmR2 =0 (w) are potentially multivalued

∼ (z − w)−1/2 from e−φ(z) and e−φ(w)/2

∼ (z − w)−#(ψ)/2 from ∂zn≥0 ψ µ (z) and SA (w)

where all of ψ µ and its derivatives ∂zn≥1 ψ µ are assigned 1 unit of worldsheet fermion number #(ψ)

105
 • branch cuts of the superghost sector and (ψ µ , SA ) cancel only for odd numbers of ψ µ within NS
vertex operators; this selection rule removes the tachyon with #(ψ) = 0 and keeps the k 2 = 0 vector

• next investigate the OPE of two massless fermions VmR2 =0 (z) and VmR2 =0 (w) in D = 10 where

CAB : e−φ(w) : (Γµ C)AB ψµ (w) −φ(w)

: e−φ/2 SA (z) : : e−φ/2 SB (w) :∼ + √ :e : +... (8.68)
(z − w)3/2 2(z − w)

• get a branch cut from the OPE of massless fermions unless all pairs of physical wavefunctions obey
χA B A
1 χ2 CAB = 0; this can be accomplished by truncating the 32-component Dirac spinors χ SA (z)

to 16-component Weyl spinors χα Sα (z) since the charge-conjugation matrix is block-off-diagonal in

Cα β̇
a Weyl basis with Cαβ = C α̇β̇ = 0 and therefore CAB = ( C0α̇ 0
)
β

0 γµ
• also the Dirac matrices are block off-diagonal in a Weyl basis, (Γµ )A B = ( µα̇β
αβ̇
) with 16×16 Clif-
γ 0
ford algebra γαµβ̇ γ ν β̇ω + γαν β̇ γ µβ̇ω = 2η µν δαω , therefore get nonzero χA µ B α µ β
1 (Γ C)AB χ1 → χ1 (γ C)αβ χ1

µ
• for R-sector states with m2 > 0, the integer moding ψr∈Z only admits integer jumps in conformal
weights via insertions of ∂zn≥0 ψ µ , which automatically gives an even number of ψ µ in massive vertex
µ
operators (e.g. states ψ−1 |AiR are generated by fields of the form ψ ν ψ λ SB |0iNS )

• the above selection rules define GSO projection for admissible physical vertex operators:

NS
Vphys ←→ odd #(ψ) ⇒ no tachyon
R
Vphys ←→ even #(ψ) & SA → Sα Weyl spinor

Moreover, the requirement of odd #(ψ) removes all half-odd integer mass levels from the NS-sector
spectrum, i.e. states ψ µ ψ ν : e−φ eip·X : and ∂z X µ : e−φ eip·X : with would-be mass level 21 . Hence,
4N
both the NS- and the R-sector states exclusively populate integer mass levels with m2 ∈ α0 which
is a necessary condition for spacetime supersymmetry.

Still, there are two loose ends about GSO projected states

1 5
• polarizations µ , χα of the massless states are unconstrained by h(ΦNS
phys ) = 2 and h(ΦR
phys ) = 8
– this is different from the bosonic string where µ : i∂z X µ eip·X : only transforms as a conformal
primary of weight one if  · p = 0
Qn
• by the background charge of the (β, γ)-system, correlators h j=1 : eqj φ(zj ) :i at genus zero vanish
Pn
unless j=1 qj = −2 (or = 2g − 2 at higher genus); however, correlation functions with more
NS R
than two Vphys or more than four Vphys necessarily have a sum of charges < −2 even though the
associated tree-level amplitudes are expected to be nonzero

• the resolution of both is based on worldsheet supersymmetry

In order to obtain non-vanishing n-point tree-level amplitudes, we need representatives of physical vertex
operators Vphys (z) ∼: eqφ(z) : with “ghost picture” q 6= 1, − 12 , e.g.

NS b NS (z) : e0φ(z) : in the 0 picture

• for bosons Vbphys (z) ∼ Φ phys

R b R (z) : eφ(z)/2 : in the 1

• for fermions Vbphys (z) ∼ Φ phys 2 picture

106
 2
• recalling that h(eqφ ) = − q2 − q, we attain h(Vbphys
NS R
) = h(Vbphys ) = 1 for the new representatives if
NS R
the contributions Φphys , Φphys from the matter variables obey
b b

b NS ) = 1 , bR ) = 13
h(Φ phys h(Φ phys (8.69)
8
NS b NS R R
• want to represent the same |physi via (Vphys , Vphys ) and (Vphys , Vbphys ), i.e. we have to make sure
that the respective Φphys and Φ
b phys are related by a symmetry of the RNS action, more specifically
by the worldsheet supercurrent G(z) ∼ i∂z Xµ (z)ψ µ (z)

1
For spacetime bosons, (ΦNS b NS
phys , Φphys ) must form a superconformal primary of h = 2 (which yields the
b NS ) = 1). By the general OPEs of superconformal primaries, this requires
desired h(Φ phys

Φb NS
phys
G(z)ΦNS
phys (w) ∼ + ... (8.70)
2(z − w)

• absence of double poles in the OPE G(z)ΦNS

phys (w) imposes constraints on polarization tensors, and
we will in particular find  · p = 0 in the massless case

• investigate the massless example in detail:

G(z)ΦNS µ λ
m2 =0 (w) ∼: i∂z Xµ (z)ψ (z) : λ : ψ (w)e
ip·X(w)
:
 0  µλ 
2α pµ η
∼ λ : + i∂w Xµ (w) + . . . + ψ ψ (w) + . . . eip·X(w) :
µ λ
(8.71)
z−w z−w
µ i∂w X µ (w) + 2α0 (p · ψ)ψ µ (w)
 
2α0  · p
 
∼: + + . . . eip·X(w) :
(z − w)2 z−w

• need  · p = 0 for absence of double poles, and from the simple pole, one can read off the zero-picture
vertex operator
b NS2 (z) ∼ µ : i∂z X µ (z) + 2α0 (p · ψ)ψ µ (z) eip·X(z) :
 
Φ m =0 (8.72)

b NS2 →
• from the zero-picture representative, one can see that µ → pµ yields a total derivative Φ m =0
∂z : eip·X(z) : as in the bosonic string (while pµ pν : ψ µ ψ ν := 0 by the anticommuting nature of ψ µ )

For spacetime fermions, we shall now generalize the supersymmetry transformation G−1/2 : ΦNS
phys →
b NS to G−1 : ΦR → Φ
Φ bR :
phys phys phys

• supercurrent modes can be picked up from the contour integral

I
dz
Gr = G(z)(z − w)r+1/2 (8.73)
B (w) 2πi

• at r = −1, we find
I
dz
bR
Φ R
phys (w) = G−1 Φphys )(w) = G(z)(z − w)−1/2 ΦR
phys (w) (8.74)
B (w) 2πi

b R is a conformal primary and annihilated by L1 ; however, by [L1 , G−1 ] =

• have to make sure that Φ phys
3
2 G0 , only get h = 1 primary if the vertex operator in the −1/2 picture obeys
I
R dz
0 = G0 Φphys )(w) = G(z)(z − w)1/2 ΦR
phys (w) (8.75)
B (w) 2πi

which requires the pole ∼ (z − w)−3/2 of the OPE G(z)ΦR

phys (w) to vanish

107
 • in the massless example,

G(z)ΦR µ α
phys (w) ∼: i∂z Xµ (z)ψ (z) : χ Sα (w) : e
ip·X(w)
:
 χα γ µ S β̇ (w)
2α0 pµ
 
∼ + ... √ αβ̇ + . . . : eip·X(w) : + . . . (8.76)
z−w 2(z − w)1/2

• again, the G0 -action vanishes as desired if the coefficient of (z − w)−3/2 does, i.e. need the fermion
polarization χα to satisfy the massless Dirac equation

χα γαµβ̇ pµ = 0 (8.77)

which eliminates 8 out of 16 components of χα=1,2,...,16 ; this is similar to the massless Dirac equation
in four dimensions with γαµβ̇ pµ → ( 2E 0
0 0 ) in a suitable SO(1, 3) frame, which clearly eliminates one
out of two Weyl-spinor components

• after imposing the above requirements on G(z)ΦNS R

phys (w) and G(z)Φphys (w), the resulting 8 physical
massless fermion polarizations χα match the 8 bosonic degrees of freedom in µ , which is again an
important hallmark of spacetime supersymmetry!

We conclude this section with the following comments:

NS b NS R R (q)
• instead of Vphys , Vphys , Vphys , Vbphys , denote vertex operators for open-string state |ϕi by Vϕ (z) ∼:
eqφ(z) :, i.e. with the ghost picture q in the superscript; for massless states, for instance

V(−1) (z) ∼ µ ψ µ (z) : e−φ(z) eip·X(z) :

V(0) (z) ∼ µ : i∂z X µ (z) + 2α0 (p · ψ)ψ µ (z) eip·X(z) :
 
(8.78)
Vχ(−1/2) (z) ∼ χα Sα (z) : e−φ(z)/2 eip·X(z) :

(0) (−1)
• decoupling of the longitudinal polarization µ → pµ propagates from V (z) to V (z) by ghost-
µ −φ+ip·X
picture equivalence, but it is harder to see that pµ ψ : e : is spurious

• for closed strings, φ(z) and φ̄(z̄) need to independently saturate the background charge; one can
independently pick left- and right-moving ghost pictures, e.g. the graviton vertex operator

(−1,−1)
Vζ (z, z̄) ∼ ζµν ψ µ (z)ψ̄ ν (z̄) : e−φ(z)−φ̄(z̄)+ip·X(z,z̄) : (8.79)

(−1,0) (0,−1) (0,0)

exists in the alternative ghost pictures Vζ , Vζ and Vζ .

9 String theory and spacetime supersymmetry

Parts of the material in this section can be found in the lecture notes arXiv:1011.1491 on supersymmetry
and extra dimensions which are recommended for further reading.

9.1 Supersymmetry in four spacetime dimensions

To avoid collisions with the ten-dimensional notation (with vector indices µ, ν = 0, 1, . . . , 9 and Weyl-
˙ our conventions for tensors of the four-dimensional
spinor indices α = 1, 2, . . . , 16 & β̇ = 1̇, 2̇, . . . , 16)
Lorentz group SO(1, 3) will be as follows

108
 • four-dimensional vector indices m, n, p, . . . = 0, 1, 2, 3

• left- and right-handed Weyl-spinor indices a, b, . . . = 1, 2 and ȧ, ḃ, . . . = 1̇, 2̇

• four-dimensional Minkowski metric in mostly-minus signature η mn = diag(1, −1, −1, −1)

• four-dimensional Weyl-spinor indices are raised / lowered using the antisymmetric ε-tensor
   
0 −1 0 −1
εab =  , εȧḃ =  
1 0 1 0
 
0 1
ψa = εab ψ b , ψ b = εbc ψc with εbc =   (9.1)
−1 0
 
0 1
χ̄ȧ = εȧḃ χ̄ḃ , χ̄ḃ = εḃċ χ̄ċ with εḃċ =  
−1 0

i.e. the SO(1, 3) charge-conjugation matrix acting on 4-component Dirac spinors (ψa , χ̄ȧ ) is block
diagonal (in contrast to the block-off-diagonal charge-conjugation matrix in D = 10)

From a representation-theoretic viewpoint, supersymmetry (SUSY) is an extension of the Poincaré algebra

{P m , J mn } by spinor generators {Qa , Q̄ḃ } to the so-called super Poincaré algebra

• recall the commutation relations among translations P m and Lorentz transformations J mn

[Jmn , Pq ] = i(ηnq Pm − ηmq Pn ) (9.2)

[Jmn , Jpq ] = i ηnp Jmq − ηnq Jmp − (m ↔ n)

• (anti-)commutators of Qa , Q̄ḃ involve Pauli matrices

(       )
1 0 0 1 0 −i 1 0
σamḃ =  ,  ,  ,   = (12×2 , ~σ ) (9.3)
0 1 1 0 i 0 0 −1

as well as
σ̄ mȧb = εȧċ εbd σdmċ = (12×2 , −~σ ) (9.4)

subject to the Clifford algebra σ m σ̄ n + σ n σ̄ m = 2η mn

• the commutators of the spinor generators with those of the Poincaré group are
i m n
[Qa , J mn ] = (σ σ̄ − σ n σ̄ m )a b Qb
4
[Qa , P m ] = [Q̄ḃ , P m ] = 0 (9.5)

• anticommutators among the spinor generators introduce translations

{Qa , Qb } = {Q̄ȧ , Q̄ḃ } = 0

{Qa , Q̄ḃ } = 2σamḃ Pm (9.6)

In order to study massless representations of the super-Poincaré algebra, we pick a Lorentz frame where
the momentum eigenvalues are aligned to Pm = (E, 0, 0, E)
 
1 0
⇒ {Qa , Q̄ḃ } = 2E(σ 0 + σ 3 )aḃ = 4E  
0 0

⇒ vanishing matrix elements for Q2 and Q̄2̇ (9.7)

109
 • Q1 , Q̄1̇ as raising and lowering operators for helicity λ – the eigenvalue of J 3 = J 12 adjusted to the
spatial components of the P m eigenvalues

1 1 0
b
[Qa , J 12 ] = Qb (9.8)
2 0 −1 a
1 1
⇒ [Q1 , J 3 ] = Q1 and [Q̄1̇ , J 3 ] = − Q̄1̇
2 2

• can map the action of Q1 , Q̄1̇ to the fermionic oscillator by normalizing as follows

Q1 Q̄
a= √ , a† = √1̇ (9.9)
2 E 2 E
⇒ a2 = (a† )2 = 0 , {a, a† } = 1

• build super-Poincaré representations on lowest-helicity state |λi subject to

J 3 |λi = λ|λi , a|λi = 0 (9.10)

• SUSY adjoins state a† |λi of helicity (λ + 21 ) and therefore opposite statistics:

J 3 a† |λi = [J 3 , a† ] + a† J 3 |λi

= λ + 21 a† |λi



(9.11)

• simple examples of massless SUSY multiplets:

a
(λ=0) scalar  (λ= 21 ) fermion
| {z } a† | {z }
e.g. Higgs e.g. “Higgsino”
or “squark” or quark
a
(λ= 12 ) “gaugino”  (λ=1) gauge boson (9.12)
a†
a
(λ= 23 ) “gravitino”  (λ=2) graviton
a†

a
• gravitinos necessarily introduce local supersymmetry ∼ eiε (x)Qa
(with spacetime dependent spinor
a
parameter ε (x)) as a gauge symmetry to avoid negative-norm states; by {Qa , Q̄ḃ } ∼ Pm , this leads
to local translation invariance and therefore diffeomorphism symmetry, which introduces gravity
and thereby massless spin-two states

Extended SUSY admits N ≥ 1 pairs (Qa , Q̄ḃ ), with an additional index I, J = 1, 2, . . . , N in

{QIa , Q̄ḃJ } = 2δJI σamḃ Pm (9.13)

• with massless Pm = (E, 0, 0, E), the N -extended super-Poincaré algebra is equivalent to N copies
of the fermionic oscillator

{aI , aJ } = {a†I , a†J } = 0

{aI , a†J } = δJI (9.14)

110
 • representations of N -extended super-Poincaré again built on lowest-helicity state aI=1,2,...,N |λi = 0

state helicity #(states)

N


|λi λ 1 = 0
a†I |λi 1 N


λ+ 2 N = 1
a†I a†J |λi 1 N


λ+1 2 N (N −1) = 2 (9.15)
.. ..
. . ...
a†1 . . . a†J . . . a†N |λi N 1 N


λ+ − N =
c
2 2 N −1
a†1 a†2 . . . a†N |λi N N


λ+ 2 1 = N

PN N
= 2N states (2N −1 bosonic and 2N −1 fermionic ones each)


for a total of k=0 k

• N = 2 example: vector multiplet

a†1
a†1 |λ = 1
2 ; 1i
0

a1 a†2
|λ = 0i |λ = 1; 12i
a2
a†1

a†2 |λ = 12 ; 2i 0
a†2

• SUSY multiplets have to be invariant under CPT (combination of charge conjugation, parity and
time reversal) which maps |λi → |−λi; in the present case, this amounts to adding additional states
|λ = 0i, |λ = − 12 ; 1i, |λ = − 12 ; 2i and |λ = −1; 12i such that for instance |λ = ±1; 12i yields the two
physical polarizations of a gauge boson

The maximal SUSY which is possible for a gauge theory (i.e. in absence of gravitational states of helicities
|λ| ≥ 23 ) is N = 4. The states of the so-called N = 4 super-Yang–Mills (SYM) theory are built on the
lowest-helicity state with aI=1,2,3,4 |λ = −1i = 0:

1× a†I 4× a†J 6× a†K 4× a†L 1×

| − 1i |− 12 ; Ii |0; [IJ]i | 21 ; [IJK]i |1; [IJKL]i

aI aJ aK aL

6 scalars φ[IJ]
4 “gluinos” λIa , λ̄ḃJ
2 phys. polarizations of the gluon

• QIa , Q̄ḃJ do not act on gauge-group generators, the full multiplet {Am , χIa , χ̄ḃJ , φIJ = −φJI } trans-
forms in the adjoint representation (in contrast to the Standard-Model fermions in (anti-)fundamental
representations)

• follows from dimensional reduction of N = 1 SYM in D = 10, which in turn comes from the
massless states of the open superstring

111
 • the above states of the N = 4 SYM multiplet are CPT self-conjugate, i.e. there is no need or room
to add states for CPT invariantization

As our last example of a massless multiplet of extended SUSY, we shall discuss N = 8 supergravity with
lowest-helicity state aI=1,2,...,8 |λ = −2i = 0

“gravitini”
“dilatini”

# states 1 8 28 56 70 56 28 8 1

|λi |−2i |− 32 i |−1i |− 21 i |0i |+ 12 i |+1i |+ 32 i |+2i

scalars
“graviphotons”
2 phys. polarizations of graviton

5
• higher SUSY N ≥ 9 would necessarily involve helicities |λ| ≥ 2 and thereby exceed spin two

=⇒ N = 8 is the max. supergravity excluding higher spin

• N = 8 supergravity can be obtained from dimensional reduction of the unique N = 1 supergravity

in D = 11 or from the type IIA/IIB N = 2 supergravities in D = 10 (the massless states of type
IIA/IIB superstrings)

• alternatively, the N = 8 supergravity multiplet can be constructed from a double copy of N = 4

SYM, and the KLT formula uplifts this relation to tree amplitudes of arbitrary N = 8 states

• again, this construction of the N = 8 supergravity multiplet is automatically CPT self-conjugate

9.2 Open superstrings and higher-dimensional supersymmetry

Recall the worldsheet realization of Poincaré generators of open bosonic string
I
1 dz
Pµbos = i∂z Xµ (z) (9.16)
2α0 2πi
I
bos 1 dz
Jµν = (iXµ i∂z Xν − iXν i∂z Xµ )
2α0 2πi
One can similarly construct super-Poincaré generators from ingredients of RNS superstrings:

• observe that Pµbos matches the gluon vertex operator at zero momentum; this turns out to generalize
(q)
to the open superstring, where the two ghost pictures V at q = −1, 0 yield two incarnations of
the translation operator
I
1 dz
Pµ(−1) = √ ψµ (z) : e−φ(z) :
2α 0 2πi
I
1 dz
Pµ(0) = 0
i∂z Xµ (z) (9.17)
2α 2πi
bos
• the Lorentz generators augment Jµν by a bi-spinor contribution acting on the spacetime vector ψ µ
I  
(0) dz 1
Jµν = (iX i∂
µ z νX − iX i∂ X
ν z µ )+ : ψ ψ
µ ν : (9.18)
2πi 2α0

112
 (q) (q) (q)
• by analogy with limp→0 V ∼ Pµ , obtain SUSY generators Qα in ghost pictures q = ± 12 from
(q)
the fermion vertex Vχ at p → 0
√ I
dz
Q(−1/2)
α = 2(α0 )−1/4 Sα : e−φ/2 : (9.19)
2πi
I
1 dz
Q(+1/2)
α =√ i∂z Xµ γαµβ̇ S β̇ : eφ/2 :
2(α0 )3/4 2πi

• in both pictures q = ± 21 , the spin-field OPE in D = 10 implies

(−1/2) (q)
{Qα , Qβ } = 2(γ µ C)αβ Pµ(q−1/2) (9.20)

which resembles the anticommutator of the four-dimensional SUSY algebra

{Qa , Q̄ḃ } = 2(σ m )aḃ Pm (9.21)

note, however, that the supercharges in the D = 10 anticommutator have alike chiralities, whereas
non-vanishing anticommutators in D = 4 involve opposite chiralities

• will now reproduce the extended N = 4, 8 SUSY multiplets in four dimensions from dimensional
reduction of D = 10 SUSY

The open superstring enjoys N = 1 SUSY in D = 10 dimensions, i.e. its supercharges form a single
left-handed Weyl spinor Qα=1,2,...,16 of SO(1, 9)

⇒ 16 components or “supercharges” just like N = 4 in four spacetime dimensions

• dimensional reduction of Qα to D = 4 is based on the decomposition

SO(1, 9) → SO(1, 3) × SO(6) (9.22)

SO(1, 3) is interpreted as the four-dimensional Lorentz group and acts on the first four components
of Pµ = (E, 0, 0, E, ~0), where Qα shifts the eigenvalue of J 12 by ± 1 just as in four dimensions; the
2
45 67 89
second factor SO(6) has Cartan generators J , J , J , and Qα shifts their eigenvalues as well

• more specifically, the decomposition of the ten-dimensional SUSY generators (left-handed Weyl
spinor of SO(1, 9)) is
Qα = QI=1,2,3,4
a=1,2 ⊕ Q̄ḃ=1,2
J=1,2,3,4 (9.23)

where QI=1,2,3,4
a=1,2 carries a left-handed Weyl-spinor index a of SO(1, 3) and a left-handed Weyl-spinor
index I of SO(6); similarly, Q̄ḃ=1,2
J=1,2,3,4 features right-handed Weyl-spinor indices ḃ and J of SO(1, 3)
and SO(6), respectively

=⇒ the four-dimensional label I = 1, 2, . . . , N of extended SUSY

refers to spinors of SO(6) when dimensionally reduced from D = 10

• in this setting, we obtain the N = 4 SYM multiplet in D = 4 by dimensionally reducing the N = 1

SYM multiplet in D = 10 dimensions: the massless states of the ten-dimensional open superstring
were found to arise from vertex operators

gluon Aµ ↔ V(q=−1,0) (9.24)

gluino χµ ↔ Vχ(q=−1/2,1/2)

113
 • decompose both fields under SO(1, 9) → SO(1, 3) × SO(6)

 A
m=0,1,2,3 : vector in D = 4
Aµ=0,1,...,9 → (9.25)
 Ai=4,5,...,9 : scalar in D = 4

 χa=1,2 : left-handed spinor in D = 4
I=1,2,3,4
χα=1,2,...,16 → J=1,2,3,4
(9.26)
 χ̄ : right-handed spinor in D = 4
ḃ=1̇,2̇

The explicit SUSY transformations of vertex operators and hence of physical states can be computed from
(∓1/2)
CFT methods, by inserting OPEs into the contour-integral representations of the supercharges Qα .
We will focus on the massless states and for bookkeeping purposes contract the spinor index α of the
supercharges with a reference Weyl-spinor η α
(∓1/2)
• Qα : boson → fermion (with normalization g of the gluon vertex operator, see section 6.5.3)
√ I
2 dz α g
α (−1/2) (0)
η Sα (z) : e−φ(z)/2 : √ µ : i∂w X µ (w)+2α0 (p · ψ)ψ µ (w) eip·X(w) :
 
[η Qα , V (w)] = 01/4
α 2πi 2α0
= g α01/4 µ pν (ηγ µ γ ν )α Sα : e−φ(w)/2 eip·X(w) : (9.27)

(−1/2)
where the right-hand side can be identified as a gluino vertex operator Vχ = g α01/4 χα Sα
: e−φ(w)/2 eip·X(w) : with wavefunction χα = µ pν (ηγ µ γ ν )α . The latter obeys the massless Dirac
equation χα γαλβ̇ pλ since pν pλ γ ν γ λ = pν pλ η νλ = p2 = 0 in the massless case.
(∓1/2)
• Qα : fermion → boson (by repeating the above kind of OPE computation)

[η α Q(−1/2)
α , Vχ(−1/2) (w)] = . . . = g (ηγµ Cχ)ψ µ (w) : e−φ(w) eip·X(w) : (9.28)

this results in a gluon vertex operator with polarization vector µ = (ηγµ Cχ) which is transversal
since χ obeys the massless Dirac equation

• with the shorthand notation to summarize the above transformations

δη µ = (ηγ µ Cχ) , δη χβ = η α (γ µν )α β µ pν (9.29)

the SUSY algebra {Qα , Qβ } = 2(γ µ C)αβ Pµ in ten dimensions closes up to linearized gauge trans-
formations (i.e. changes in λ ∼ pλ in the second term of the second line)

[δη1 , δη2 ]χα = 2(η1 γ µ Cη2 )pµ χα (9.30)

[δη1 , δη2 ]λ = 2(η1 γ µ Cη2 )pµ λ − 2(η1 γ µ Cη2 )µ pλ

9.3 Closed superstrings and higher-dimensional supersymmetry

The goal of this subsection is to

• construct ten-dimensional type-IIA/IIB supergravity multiplets from massless closed-string states

• recover N = 8 supergravity upon dimensional reduction to D = 4

In the double-copy construction of closed-string states, one can choose independent chiralities of the spin
fields SA and S̄B in the GSO projection of left- and right-movers, respectively:

114
 • the underlying left-moving OPEs ψ µ (z)SA (w) and SA (z)SB (w) decouple from those among the
right-movers ψ̄ µ , S̄A

• only the relative chirality matters for the spectrum

type IIB : “(left-handed)⊗2 ” , SA → Sα , S̄B → S̄β

(9.31)
type IIA : “(left)⊗(right)” , SA → Sα , S̄B → S̄ β̇

As will be elaborated below, one can decompose the spacetime fermions in the massless closed-string
spectrum into irreducibles of SO(1, 9). This is similar to the spacetime bosons in the NS-NS sector
µν
µ ⊗ ¯ν → ζ µν ⊕ |{z}
B µν ⊕ ηSO(8) Φ (9.32)
|{z}
28 d.o.f.
| {z }
35 d.o.f. 1 d.o.f.

• R-NS sector: by analogy with the NS-NS sector, the SO(1, 9) decomposition is of schematic form

χα ⊗ ¯µ −→ “γ-traceless” ⊕ “γ-trace” (9.33)

whose exact form is (note that γ µ γµ = 10 in ten dimensions)

1 1
χα ¯µ = χα ¯µ − ¯λ (χγ λ γµ )α + ¯λ (χγ λ γµ )α (9.34)
| 10{z } 10
| {z }
gravitino, spin 3/2-rep. of β̇α
SO(1, 9) with 8(8−1) d.o.f. write as Λ γµ with Λ a
β̇ β̇
right-handed dilatino (8 d.o.f.)

1
the 8 d.o.f. of the dilatino Λβ̇ = ¯λ (χγ λ )β̇
10  can be traced back to the massless Dirac equation
Λβ̇ γµβ̇α pµ = 0 (which follows from that of χ and transversality of ¯λ )

• NS-R sector:
type IIB : µ ⊗ χ̄α =⇒ gravitino / dilatino of same chirality as in R-NS sector
(9.35)
type IIA : µ ⊗ χ̄β̇ =⇒ gravitino / dilatino of opposite chirality as in R-NS sector

• the two gravitini in type II theories necessitate two local spacetime SUSYs (they are needed as
gauge-symmetries to remove negative-norm states), i.e. N = 2 in ten dimensions; construct the
left- and right-moving supercharges in direct analogy with open strings
(−1/2) (−1/2)
∼ : e−φ/2 : Sα , Q̄β ∼ : e−φ/2 : S̄β
H H
type IIB : Qα
(−1/2)
(9.36)
∼ : e−φ/2 : Sα , Q̄(−1/2)β̇ ∼ : e−φ/2 : S̄ β̇
H H
type IIA : Qα

In order to understand the Lorentz irreducibles in the R-R sector, decompose bi-spinors of SO(1, 9) into
(k)
antisymmetric tensors, so-called “form-fields” A[µ1 µ2 ...µk ] .

• for 16-component Weyl spinors in D = 10, which products of 16 × 16 gamma matrices γ µ exhaust
the (16 × 16)-dimensional space of bispinors?

−→ by γ µ γ ν + γ ν γ µ = 2η µν , it is sufficient to study antisymmetrized products

1  µ1 µ2
γ µ1 µ2 ...µk = γ γ . . . γ µk ± antisymm(1, 2, . . . , k)

(9.37)
k! | {z }
k! terms

• in D = 10, duality w.r.t. εµ1 µ2 ...µ10 relates k-forms γ µ1 µ2 ...µk to (10−k)-forms γ µ1 µ2 ...µ10−k , with
(anti-)self-duality at k = 5
1 µ1 µ2 ...µ10
(γ µ1 µ2 ...µ5 C)αβ = ε (γµ6 µ7 ...µ10 C)αβ (9.38)
5!
1
(γ µ1 µ2 ...µ5 C)α̇β̇ = − εµ1 µ2 ...µ10 (γµ6 µ7 ...µ10 C)α̇β̇
5!

115
 • for Weyl-spinors ψ, λ̄ of SO(1, 9) with opposite chirality, can decompose via Fierz identities
1 1 1
ψ α λ̄β̇ = (C −1 )α β̇ (ψC λ̄) + (C −1 γ µν )α β̇ (ψγνµ C λ̄) + (C −1 γ µνρσ )α β̇ (ψγσρνµ C λ̄)
16 | {z } 16 · 2! | {z } 16 · 4! | {z }
1 matrix (10) =45 matrices ( 10
)=210 matrices
2 4

(9.39)

• for Weyl-spinors ψ, λ of SO(1, 9) with alike chirality, the analogous Fierz identity is
1 1
ψ α λβ = (C −1 γ µ )βα (ψγµ Cλ) + (C −1 γ µνρ )βα (ψγρνµ Cλ)
16 | {z } 16 · 3! | {z }
10 matrices (10
3 )=120 matrices

1
+ (C −1 γ µνρστ )βα (ψγτ σρνµ Cλ) (9.40)
2 · 16 · 5! | {z }
1 10
(
2 5 )=126 matrices

where the denominator in the last term is 2 · 16 · 5! rather than 16 · 5! to avoid overcounting in
view of the self-duality of γ µνρστ ; note the symmetry properties (C −1 γ µ )βα = (C −1 γ µ )αβ as well
as (C −1 γ µνρ )βα = −(C −1 γ µνρ )αβ and (C −1 γ µνρστ )βα = (C −1 γ µνρστ )αβ

• component counting in the R-R sector: physical closed-string states arise from the double copy of
physical gaugino polarizations

(8 solutions to the massless Dirac equation)⊗2

which yields 64 d.o.f. in bi-spinors (χγµ1 ...µk C χ̄)

• need the k-form expansions of χα ⊗ χ̄β (type IIB) and χα ⊗ χ̄β̇ (type IIA) to respect χα γαµβ̇ pµ = 0
(and same for χ̄)

Engineer bispinors compatible with the Dirac equation via

(k−1)
(χγµ1 µ2 ...µk C χ̄) → Fµ(k)
1 µ2 ...µk
= kp[µ1 Aµ2 ...µk ] (9.41)

(k−1) (k−1)
with transversal (k−1)-form potentials Aµν... = A[µν...]

pµ A(k−1)
µν... = 0 (9.42)

• respective field strengths F (k) = dA(k−1) enjoy gauge invariance with (k−2)-form parameters ξ (k−2)

δA(k−1) = dξ (k−2) (9.43)

• as an example how the Dirac equation relates to the transversality of the potentials, consider the
(1) (2) (1) (1)
one-form Aµ in the R-R sector of type IIA superstrings (with field strength Fµν = pµ Aν −pν Aµ )
which is obtained from the double copy
1 −1 µν α (2)
χα χ̄β̇ → (C γ ) β̇ Fµν = (C −1 γ µν )α β̇ pµ A(1)
ν (9.44)
2
= (C −1 γ µ γ ν )α β̇ pµ A(1)
ν = −(C
−1 ν µ α
γ γ ) β̇ pµ A(1)
ν

in replacing γ µν by either γ µ γ ν or −γ ν γ µ , we have used that the µ↔ν symmetric part yields η µν
(1)
whose contraction with pµ Aν vanishes by transversality; the first representation in the second line
of (9.44) manifests that the Dirac equation of χα is preserved (contract with pλ (γ λ C)δα ) while the
second representation manifests that of χ̄β̇ (contract with pλ γλβ̇δ )

116
 • the type IIA solutions along with their d.o.f. are

 (C −1 γ µν )α p A(1) : 8 d.o.f.
β̇ µ ν
χα χ̄β̇ → (3) 8
(9.45)
−1 µνρσ α


 (C γ ) β̇ pµ Aνρσ : 56 = d.o.f.
3

(k)
by transversality of the Aµ1 ...µk and their gauge freedom under dξ (k−1) , we can treat the indices µj
as effectively taking 8 values in the degree-of-freedom counting

• the type IIB solutions along with their d.o.f. are




 (C −1 γ µ )αβ pµ A(0) : 1 scalar “axion”
(2)
χα χ̄β → (C −1 γ µνρ )αβ pµ Aνρ : 28 = 82 d.o.f.


(9.46)

(4)
(C −1 γ µνρστ )αβ pµ Aνρστ : 35 = 21 84 d.o.f.



1
where the factor of 2 in the d.o.f. of A(4) stems from the self-duality of its five-form field strength.

• in the same way as the Bµν in the NS-NS sector couples to strings through the worldsheet ac-
tion Σ dX µ ∧ dX ν Bµν , the R-R form fields A(p+1) couple to Dp branes with (p+1)-dimensional
R

worldvolumes Σp+1 via

Z
Sp-form ∼ dX µ1 ∧ dX µ2 ∧ . . . ∧ dX µp+1 A(p+1)
µ1 µ2 ...µp+1 (9.47)
Σp+1

• conversely, the R-R sector states single out the stable types of Dp branes in the respective theory

type IIA : D0 ↔ A(1) , D2 ↔ A(3)

(9.48)
type IIB : D(−1) ↔ A(0) , D1 ↔ A(2) , D3 ↔ A(4)

where the D(-1) brane has a zero-dimensional worldvolume, i.e. it covers a point in spacetime and
is referred to as an instanton

• since T-duality swaps Dp and D(p±1) branes, their R-R fields are swapped as well

T-duality
=⇒ type IIA ←→ type IIB (9.49)

We shall now assemble the ten-dimensional supergravity multiplets from the massless states in the type
IIA/IIB theories. In the following table, the number of physical d.o.f. determined from the above dis-
cussions is indicated in parenthesis, and the shorthands LH and RH stand for left- and right-handed
spacetime fermions

⊗ NS R

graviton hµν (35) LH gravitino χα

µ (56)

NS B-field Bµν (28) RH dilatino Λβ̇ (8)

dilaton Φ (1)

type IIA type IIB type IIA type IIB

R RH χ̄µβ̇ (56) LH χ̄α
µ (56) (1)
Aµ (8) A(0) (1)
α (3) (2)
LH Λ̄ (8) RH Λ̄β̇ (8) Aµνρ (56) Aµν (28)
(4)
Aµνρσ (35)

117
We shall now dimensionally reduce the type-IIB multiplet from ten to four spacetime dimensions. In
the following set of reductions, |λi refers to a state with helicity or J 12 eigenvalue λ, and the range of
SO(1, 3)- and SO(6) indices is again m, n, . . . = 0, 1, 2, 3 and i, j, . . . = 4, 5, . . . , 9 for vectors as well as
a, b = 1, 2, ȧ, ḃ = 1̇, 2̇ and I, J = 1, 2, 3, 4 for Weyl spinors.

Φ & A(0) −→ 2 × |λ=0i


(2)
 Bmn & Amn ↔ 2 × |λ=0i


(2)
Bµν & A(2)
µν −→ Bmi & Ami ↔ 2 × 12 × |λ= ± 1i

 (2)
↔ 2 × 15 × |λ=0i
Bij & Aij


 hmn ↔ 2 × |λ= ± 2i


hµν −→ hmi ↔ 12 × |λ= ± 1i (9.50)


hij ↔ 21 × |λ=0i


(4) 6


 Aijkl ↔ 4
× |λ=0i


(4)
A(4)
µνρσ −→ Amijk ↔ 63 × |λ= ± 1i

 (4) (4) (4)
no Amnij , Amnpi , Amnpq


 χa ⊕ χJ ⊕ same with
↔ 2 × 4 × 2 × |λ= ± 3 i
mI χ→χ̄
χα α
µ , χ̄µ −→ mḃ 2
 χa ⊕ χJ ⊕ same with ↔ 6 × 2 × 4 × 2 × |λ= ± 1 i


iI iḃ χ→χ̄ 2

bJ same with 1


Λβ̇ , Λ̄β̇ −→ ΛaI ⊕ Λ ⊕ Λ→Λ̄
↔ 2 × 4 × 2 × |λ= ± 2i

For gravitons, the vanishing of the ten-dimensional trace η µν hµν interlocks the SO(1, 3)-trace of hmn
with the SO(6)-trace of hij , and we effectively treat hmn as traceless in the degree-of-freedom counting.
For the dimensional reduction of A(4) , the self-duality of F (5) with momentum pµ = (pm , 0, 0, 0, 0, 0, 0)
leaves no independent d.o.f. in components with more than one SO(1, 3) index.
The above distribution of four-dimensional helicities precisely ties in with the spectrum of N = 8
supergravity

• 1 × |λ=2i from hµν

• 8 × |λ= 23 i from χα α
µ (4), χ̄µ (4)

(2) (4)
• 28 × |λ=1i from Bµν (6), Aµν (6), hµν (6), Aµνρσ (10)

• 56 × |λ= 12 i from χα α
µ (24), χ̄µ (24), Λβ̇ (4), Λ̄β̇ (4)

(2) (4)
• 70 × |λ=0i from Bµν (16), Aµν (16), hµν (21), Aµνρσ (15), Φ (1), A(0) (1)

with the same counting for the negative-helicity states.

As you will explore in exercise G.4, the ten-dimensional type IIA multiplet can also be obtained
as a dimensional reduction of the unique D = 11 supergravity (with an N = 1 Dirac spinor of 32
supercharges). With vector indices M, N, P = 0, 1, . . . , 10 and Dirac-spinor index B = 1, 2, . . . , 32 (there
is no notion of chirality in odd dimensions), its particle content is

hM N ←→ graviton with 44 d.o.f.

 
(3) 9
AM N P ←→ 3-form potential with = 84 d.o.f. (9.51)
3
χB
M ←→ gravitino with 128 d.o.f.

118
10 Superstring amplitudes
Scattering amplitudes in superstring theories admit the same kind of topological expansion in terms
of two-dimensional surfaces as you know from bosonic strings. The worldsheet topologies contributing
to tree-level and one-loop amplitudes of open and closed strings are depicted in figure 25. One-loop
amplitudes of open superstrings for instance have contributions from both cylinder and Moebius-strip
diagrams, and their interplay is crucial for the cancellation of UV divergences (which are absent in
closed-string loop diagrams from the outset)

Figure 25: Worldsheet topologies contributing to four-point tree-level and one-loop amplitudes of open
and closed superstrings

Still, there are two noteworthy differences to bosonic-string amplitudes:

• the final results for superstring amplitudes are considerably simpler than those in bosonic theories

• strictly speaking, the RNS prescription for superstring amplitudes involves an integral over bosonic
and fermionic moduli of super-Riemann surfaces (where the ghost pictures address the fermionic
moduli at low genus)

• one cannot split super-moduli space at genus g ≥ 5 in the sense that it is not possible to integrate
over the fermionic moduli first to then be left with a bosonic integral

10.1 Superstring tree-level amplitudes

For open superstrings, the properties of Chan-Paton charges associated with their endpoints are identical
to those of open bosonic strings. Hence, we have the same color-decompositions of open-string tree-level
amplitudes in the bosonic and the supersymmetric theory
X
0 0
Mtree
open ({ϕi , ai }; α ) = Tr(T aρ(1) T aρ(2) . . . T aρ(n−1) T an )Atree
super (ϕρ(1) , ϕρ(2) , . . . , ϕρ(n−1) , ϕn ; α )
super
ρ∈Sn−1
(10.1)
where we use collective notation ϕi for polarizations and momenta as well as adjoint indices ai for the
color degrees of freedom.

119
 The color-ordered amplitudes Atree
super are again computed from correlation functions of vertex operators

along with c-ghost contributions hc(zi )c(zj )c(zk )i = |zij zik zjk | for the SL2 -fixed legs at (zi , zj , zk ) =
(0, 1, ∞):
Z
0 dz1 dz2 . . . dzn (q1 )
Atree hVϕ1 (z1 )Vϕ(q2 2 ) (z2 ) . . . Vϕ(qnn ) (zn )i Pn q =−2

super (ϕ1 , ϕ2 , . . . , ϕn ; α ) ∼
vol SL2 (R) j=1 j
−∞<z1 <z2 <...<zn <∞

(10.2)

• need total superghost charge −2 at tree level (and 2g − 2 at genus g) to saturate the background
charge of the β-γ system
* +  
n
Y n
X n
Y
: eqj φ(zj ) : = δ 2 + qj  (zi −zj )−qi qj (10.3)
j=1 j=1 1≤i<j

for each combination of physical states with an even number of spacetime fermions (amplitudes
with an odd number of spacetime fermions vanish by Lorentz invariance), one can find a variety of
(q ) (q )
representatives Vϕ1 1 . . . Vϕnn in admissible ghost pictures

• the result for Atree

super is independent on the distribution of the qi over the vertex operators (e.g. for
(−1) (0)
n bosons, insert any two legs in the Vϕ picture and the remaining n−2 in the Vϕ picture);
intuitively, this can be understood from the fact that a relocation of the superghost pictures is
a worldsheet supersymmetry transformation; the detailed argument can for instance be found in
section 5.2.3 of https://edoc.ub.uni-muenchen.de/13381/1/Schlotterer_Oliver.pdf.

10.1.1 Examples of open-string amplitudes

We shall next consider several warm-up examples of open-superstring tree amplitudes. The computation
of the relevant correlators will be greatly facilitated by the fact that X µ , φ and the (ψ µ , Sα )-system furnish
three separate CFT sectors (i.e. all correlation functions factorize into those of X µ , φ and (ψ µ , Sα )).

• for three gluons, recall that pi · pj = 0 in the kinematic phase space of three massless particles such
that the Koba-Nielsen factor is 1; without loss of generality, we choose leg 1 to be the only instance
(0)
of the V -picture

Atree (0) (−1)

super (1 , 2 , 3 ) ∼ |z12 z13 z23 |hV1 (z1 )V2 (z2 )V(−1)
3
(z3 )i
= |z12 z13 z23 |h: e−φ(z2 ) : : e−φ(z3 ) :i1µ 2ν 3λ
 Y 3
× h: i∂z1 X µ (z1 )eip1 ·X(z1 ) : : eipj ·X(zj ) :i hψ ν (z2 )ψ λ (z3 )i
j=2
3
Y 
0 ipj ·X(zj )
+ 2α h :e :ip1ρ h: ρ µ ν λ
ψ ψ (z1 ) : ψ (z2 )ψ (z3 )i (10.4)
j=1
|z12 z13 z23 | n o
∼ (p2 · 1 )(2 · 3 ) + (1 · 2 )(p1 · 3 ) − (1 · 3 )(p1 · 2 )
z12 z13 z23 | {z } | {z }
from ∂z1 X µ from ψ ρ ψ µ

= Atree
SYM (1 , 2 , 3 )

Similar to the correlators of the bosonic string, the net effect of the i∂z1 X µ (z1 ) insertion is a factor of
·p2 1 ·p3 ·p2 z23
2α0 ( z1 12 + z13 ) = 2α0 z1 12 z13 (using transversality and momentum conservation). The remaining

120
 −1
correlators have been evaluated via h: e−φ(z2 ) : : e−φ(z3 ) :i = z23 and h: ψ ρ ψ µ (z1 ) : ψ ν (z2 )ψ λ (z3 )i =
|z12 z13 z23 |
(η µν η ρλ − η µλ η νρ )/(z12 z13 ). Furthermore, z12 z13 z23 = −1 is independent on the zj within the
color-ordering of Atree
super (1 , 2 , 3 ) but instead gives +1 in case of Atree
super (1 , 3 , 2 ). Accordingly, the

color-dressed amplitude depends on the color degrees of freedom via Tr(T a1 [T a2 , T a3 ]) ∼ f a1 a2 a3 .

Most importantly, the three-gluon amplitude coincides with its field-theory limit, i.e. it is homoge-
neous in α0 and does not involve any analogue of the extra term 2α0 (pi · j )3 in the bosonic-string
amplitude. This exemplifies a general virtue of superstring amplitudes that the α0 -dependence takes
a considerably simpler form that for bosonic strings. In the RNS formulation, one can attribute
this simplicity to worldsheet supersymmetry, while the amplitude computations in the pure-spinor
formalism are simplified by the manifest spacetime supersymmetry.

• for one gluon and two gluinos, it is natural to employ the −1 and − 21 ghost pictures in evaluating

Atree (−1)
super (1 , χ2 , χ3 ) ∼ |z12 z13 z23 |hV1 (z1 )Vχ(−1/2)
2
(z2 )Vχ(−1/2)
3
(z3 )i
3
Y
= |z12 z13 z23 |h: e−φ(z1 ) : : e−φ(z2 )/2 : : e−φ(z3 )/2 :ih : eipj ·X(zj ) :i
j=1

× µ1 χα β
2 χ3 hψµ (z1 )Sα (z2 )Sβ (z3 )i (10.5)
|z12 z13 z23 | 1 µ
∼ √  (χ2 γµ Cχ3 )
z12 z13 z23 2 1
= Atree
SYM (1 , χ2 , χ3 )

Since we are scattering GSO projected states, all fractional powers of zij from matter and ghost
−1/4
contributions h: e−φ(z1 ) : : e−φ(z2 )/2 : : e−φ(z3 )/2 :i = (z12 z13 )−1/2 z23 and hψµ (z1 )Sα (z2 )Sβ (z3 )i =
(γµ C)αβ
√ 3/4 conspire to integer ones.
2(z12 z13 )1/2 z23

• the results of both three-point examples can be summarized in a spacetime supersymmetric way by

Atree tree
super (1, 2, 3) = ASYM (1, 2, 3) (10.6)

where 1, 2, 3 refer to any combination of states in the massless gauge multiplet; the right-hand side
refers to the amplitudes of N = 1 in SYM D = 10 which can for instance be generated from the
Lagrangian Z  
10 1 µν α µ β
SSYM [A, χ] ∼ d X Tr Fµν F + χ (γ C)αβ [Dµ , χ ] (10.7)
4
with gauge-covariant derivatives [Dµ , χ] = ∂µ χ − [Aµ , χ]; the bosonic part ∼ Fµν F µν of the La-
grangian is identical to that of pure Yang–Mills theory, that is why the n-gluon tree amplitudes are
universal to SYM and pure YM; the Feynman rules of the SYM action consist of the three- and
four-gluon vertices known from pure YM and an additional three-point vertex of two gauginos and
one gluon, see the above Atree
SYM (1 , χ2 , χ3 ) for its (unique) tensor structure

• among the massless four-point functions, the final result for the four-gluino amplitude takes a par-
ticularly compact form (see exercise H.2); with the conventions sij = α0 (pi + pj )2 for dimensionless

121
 Mandelstam invariants, we have
Z 1
tree 0
Asuper (χ1 , χ2 ,χ3 , χ4 ; α ) ∼ dz2 lim |z13 z14 z34 | hVχ(−1/2)
1
(z1 )Vχ(−1/2)
2
(z2 )Vχ(−1/2)
3
(z3 )Vχ(−1/2)
4
(z4 )i
0 z4 →∞
Z 1
(χ1 γµ Cχ2 )(χ3 γ µ Cχ4 ) (χ4 γµ Cχ1 )(χ2 γ µ Cχ3 )
 
dz2
∼ s12 |z2 |s12 |1−z2 |s23 +
0 z2 s12 s23
Γ(1+s12 )Γ(1+s23 ) tree
∼ ASYM (χ1 , χ2 , χ3 , χ4 ) (10.8)
Γ(1+s12 +s23 )

in the color-ordering under investigation, our choice of SL2 -fixing (z1 , z3 , z4 ) → (0, 1, ∞) leads to
the integration domain z2 ∈ (0, 1); the four-point SYM amplitude in the curly bracket of the second
line is literally the result of the Feynman-diagram computation below

χ2 χ3 χ1 χ2

χ1 χ4 χ4 χ3

• finally, the four-point result obtained for gauginos uplifts to the entire gauge supermultiplet, i.e.
the string tree amplitude for arbitrary combinations of gluons and gluinos takes the universal form

0 Γ(1+s12 )Γ(1+s23 ) tree

Atree
super (1, 2, 3, 4; α ) = ASYM (1, 2, 3, 4) (10.9)
Γ(1+s12 +s23 )
Γ(1+s12 )Γ(1+s23 )
by the leading term in the α0 -expansion Γ(1+s12 +s23 ) = 1 − ζ2 s12 s23 + O(α03 ), this is consistent
with the field-theory limit α0 → 0 where any massless Atree tree
super reduces to the corresponding ASYM .

10.1.2 Massless n-point open-string amplitudes

The tight interplay between open-superstring and SYM tree amplitudes generalizes to any number n of
external gauge multiplets: As derived in arXiv:1106.2645 from the pure-spinor superstring, such color-
ordered n-point amplitudes can be decomposed into
X
0
Atree
super (1, 2, . . . , n; α ) = F ρ (sij )Atree
SYM (1, ρ(2, 3, . . . , n−2), n−1, n) (10.10)
ρ∈Sn−3

where all the polarization dependence resides in a (n−3)! basis of Atree

SYM (1, ρ(. . .), n−1, n) w.r.t. BCJ

relations (the field-theory limits of monodromy relations). The α0 -dependence in turn is entirely carried
by the basis of disk integrals
Z n−1  n−2
Y k−1 
ρ
Y
sij
X skm
F (sij ) = dz2 dz3 . . . dzn−2 |zij | ρ (10.11)
z
m=1 km
0<z2 <z3 <...<zn−2 <1 1≤i<j k=2

with (z1 , zn−1 , zn ) = (0, 1, ∞), where the permutation ρ ∈ Sn−3 acts on the labels of both the zkm and
skm enclosed by the curly brackets. The α0 -expansion of the (n−3)! integrals F ρ (sij ) tailored to the disk
ordering with zi < zi+1 has leading orders

ρ
F ρ (sij ) = δ23...n−2 + O(α02 ) (10.12)

The field-theory limit for ρ different from the identity permutation is absent such that the above expression
0
for Atree tree
super consistently reduces to ASYM in the same color-ordering as α → 0. The first non-vanishing

122
α0 -corrections take the schematic form ζ2 s2ij + ζ3 s3ij + . . ., and more generally, the α0w order features
polynomials in sij accompanied by multiple zeta values ζn1 ,n2 ,...,nr of weight w = n1 + n2 + . . . + nr ,
see for instance arXiv:1304.7304. This property is known as uniform transcendentality and has also been
observed in various instances of field-theory amplitudes and Feynman integrals.

10.1.3 Closed-string tree amplitudes

For tree-level amplitudes of type IIA and type IIB superstrings, the opening line
E  nj=1 qj =−2
 n P
Z 2
tree 0 d z1 d2 z2 . . . d2 zn Y (qa ,q̄a )
Msuper ({Φi }; α ) ∼ VΦa (za ) P (10.13)
vol SL2 (C) a=1
n
q̄j =−2 j=1

(q,q̄)
involves correlation functions of vertex operators VΦ with independent left- and right-moving su-
perghost pictures q, q̄. By the background charge of the respective superghost systems, one needs to
separately saturate
n
X n
X
qj = −2 = q̄j (10.14)
j=1 j=1

in order to have non-zero correlators for the : eqj φ(zj ) : and : eq̄j φ̄(z̄j ) : involving chiral bosons φ and φ̄.
The sphere integrals over the resulting correlation functions can be manipulated with the same KLT
techniques that you have seen for the closed bosonic string
0
X
0
Mtree
super ({Φ1 , Φ2 , . . . , Φn }; α ) ∼ Atree α
super (ϕ1 , ρ(ϕ2 , ϕ3 , . . . , ϕn−2 ), ϕn−1 , ϕn ; 4 )
ρ,τ ∈Sn−3
0
× Sα0 (ρ|τ )Atree α
super (ϕ̄1 , τ (ϕ̄2 , ϕ̄3 , . . . , ϕ̄n−2 ), ϕ̄n , ϕ̄n−1 ; 4 ) (10.15)

The entries of the (n−3)! × (n−3)! matrix Sα0 (ρ|τ ) indexed by permutations ρ and τ are again degree-
0 n−3
(n−3) polynomials in sin( πα

2 pi · pj ) . This KLT formula is universally valid for any combination
of type-IIA or type-IIB states Φj = ϕj ⊗ ϕ̄j , and it can be adapted to the heterotic string by replacing
one of the Atree tree
super on the right-hand side by Abos . In other words, the KLT decomposition into products

of disk integrals is a universal property of sphere integrals, regardless of the string theory they originate
from.
Finally, in case of four massless states (i.e. arbitrary combinations 1, 2, 3, 4 of states in the type
IIA/IIB supergravity multiplets), the trigonometric factor can be used to rearrange the Γ functions of
0
Atree
super (1, 2, 3, 4; α ) into

πα0
 
0 0 0
Mtree
super ({1, 2, 3, 4}; α ) ∼ sin p1 · p2 Atree tree
super (1, 2, 3, 4; α )Asuper (1̄, 2̄, 4̄, 3̄; α )
2
Γ(1 + 4s )Γ(1 + 4t )Γ(1 + u4 ) tree
∼ M ({1, 2, 3, 4}) (10.16)
Γ(1 − 4s )Γ(1 − 4t )Γ(1 − u4 ) supergravity
 ζ3 
= 1 − stu + O(α05 ) Mtree supergravity ({1, 2, 3, 4})
32
On the right-hand side, we have identified the four-point supergravity amplitudes according to the field-
theory double copy

Mtree tree tree

supergravity ({1, 2, 3, 4}) ∼ ASYM (1, 2, 3, 4)(p1 · p2 )ASYM (1̄, 2̄, 4̄, 3̄) (10.17)

where the type-IIA and type-IIB cases are distinguished by the chirality of the gauginos in the second
03
factor of Atree
SYM . The first string correction to supergravity occurs at the order of α and corresponds

123
to an α03 ζ3 R4 interaction together with its N = 2 superpartners in D = 10 dimensions. In other words,
the curvature corrections α0 R2 and α02 R3 of the bosonic string are absent for type-II superstrings. This
can firstly be viewed as a consequence of the dropout of α0 F 3 interactions from the open superstring.
Secondly, the 32 supercharges of type-II superstrings are incompatible with the operators α0 R2 and α02 R3
of the closed bosonic string.
Note, however, that the (n−3)! × (n−3)!-permutation sum in the n-point KLT formula does not allow
0
to factor out the analogous (n≥5)-point supergravity amplitudes from Mtree
super ({Φi }; α ) at n≥5.

124
String Theory I Maor Ben-Shahar, Oliver Schlotterer

A Problem set Sept 23th 2022

A.1 Equations of motion

The purpose of this exercise is to determine the equations of motion (e.o.m.) for the Nambu–Goto and
Polyakov actions SNG [X] and SP [X, h] introduced in the lecture
∂X µ ∂X ν
(i) derive the following lemmata concerning the induced metric γαβ = ∂σ α ∂σ β ηµν on the worldsheet
∂γβδ
= δβα ∂δ X µ + δδα ∂β X µ
∂(∂α Xµ )
√
∂ − det γ 1p
= − det γ γ δβ
∂γβδ 2

(ii) show that the e.o.m. derived from the Nambu–Goto action takes the form of a reparametrization
1
invariant wave equation (with tension T = 2πα0 )

δSNG [X] p
− det γ γ αβ ∂β X µ .


= T ∂α
δXµ

(iii) show that the e.o.m. of the Polyakov action w.r.t. the (inverse) worldsheet metric is given by
√
δSP [X, h] − det h n µ ν 1 γδ µ ν
o
= − ∂α X ∂β X η µν − hαβ (h ∂γ X ∂δ X )η µν .
δhαβ 4πα0 2

A.2 Warmup on Christoffel symbols

Following your General Relativity course, the Christoffel symbols associated with the metric hαβ are
given by
1 σρ
Γσαβ = h (∂α hβρ + ∂β hαρ − ∂ρ hαβ ) ,
2
and they enter the covariant derivatives of vectors V β and covectors Wβ as follows

∇α V β = ∂α V β + Γβγα V γ , ∇α Wβ = ∂α Wβ − Γγβα Wγ

(i) check that their trace can be written as

√
Γσσα = ∂α log − det h .

δSP
(ii) use this to show that the equations of motions 0 = δXµ can be written as a reparametrization
invariant wave equation
√ √
∂α ( − det h hαβ ∂β X µ ) = − det h ∇α ∇α X µ = 0

(iii) in the context of diffeomorphism invariance of the Polyakov action, show that the infinitesimal
transformation δhαβ = η ρ ∂ρ hαβ + hρα ∂β η ρ + hρβ ∂α η ρ can be rewritten as δhαβ = ∇α ηβ + ∇β ηα .

(iv) given the metric in conformal gauge hαβ = e2φ ηαβ , check that the only non-vanishing Christoffel
symbols in lightcone coordinates are

Γ+
++ = 2∂+ φ , Γ−
−− = 2∂− φ .

(v) compute the resulting components of the Riemann tensor and the Ricci scalar

125
A.3 Diffeomorphism invariance of the Polyakov action
This exercise helps you to verify the invariance of the Polyakov action under infinitesimal diffeomorphisms

δX µ = η γ ∂γ X µ , δhαβ = η γ ∂γ hαβ + hγα ∂β η γ + hγβ ∂α η γ .

(i) show that the above variation of the metric propagates as follows to its inverse and its determinant:

δhαβ = η γ ∂γ hαβ − ∂ α η β − ∂ β η α
√ √
δ − det h = ∂γ (η γ − det h)

(ii) use these lemmata to identify the variation of the Polyakov action as a total derivative

1
Z  √ 
2 γ αβ µ
δSP [X, h] = − d σ ∂γ η − det h h ∂ α X ∂β Xµ .
4πα0

A.4 Noether currents for Poincaré symmetry

The purpose of this exercise is to study the Noether currents and conserved charges w.r.t. the global
Poincaré symmetry
δX µ = Λµ ν X ν + cµ , Λµν = −Λνµ

of the Polyakov action.

(i) by writing the variation w.r.t. local transformations in the form

Z  1 α 
δSP [X, h] = d2 σ Pµα ∂α cµ + Jµν ∂α Λµν ,
2
α
determine the momentum current Pµα and the Lorentz current Jµν in terms of a general metric hαβ
(i.e. before gauge fixing)

(ii) show that the associated charges in conformal gauge are given by
Z 2π Z 2π
1
Pµ = dσ Pµτ = dσ ∂τ Xµ
0 2πα0 0
Z 2π Z 2π
τ 1
Jµν = dσ Jµν = dσ (Xν ∂τ Xµ − Xµ ∂τ Xν )
0 2πα0 0

and prove that they are conserved, i.e. that ∂τ Pµ = ∂τ Jµν = 0, if X µ obeys the equations of motion

(iii) by inserting the mode expansion

r
0 µ α0 X 1 µ −in(τ +σ)
µ µ
X =x +αp τ +i (α̃ e + αnµ e−in(τ −σ) ) ,
2 n n
n6=0

derive the following expressions for the conserved charges

P µ = pµ
∞
X 1 ν µ µ
J µν = pµ xν − xµ pν + i αn α−n − αnµ α−n
ν
+ α̃nν α̃−n − α̃nµ α̃−n
ν


.
n=1
n

126
A.5 Commutation relations
The goal of this exercise is to study the consequences of the equal-time commutators

[X µ (τ, σ), Πν (τ, σ 0 )] = iδνµ δ(σ − σ 0 )

[X µ (τ, σ), X ν (τ, σ 0 )] = [Πµ (τ, σ), Πν (τ, σ 0 )] = 0

and to derive the Hamiltonian in the canonical quantization of the bosonic string.

1
(i) use the mode expansion of the canonical momentum Πµ = 2πα0 Ẋ
µ
to derive (k ∈ Z)
Z 2π Z 2π
1
dσ Πµ (τ, σ) = pµ , dσ eikσ Πµ (τ, σ) = √ α̃kµ e−ikτ + α−k
µ
eikτ


0 0 2α0

(ii) show that the Hamiltonian takes the following form in terms of the oscillator modes
Z 2π
1
dσ Ẋ 2 (τ, σ) + X 02 (τ, σ)


H=
4πα0 0
∞
α0 1X
= p2 +


α−n · αn + αn · α−n + α̃−n · α̃n + α̃n · α̃−n
2 2 n=1

and, using the commutation relations in (iii), that normal ordering formally leads to the alternative
expression in D spacetime dimensions
∞ ∞
α0 2 X
X
H= p + α−n · αn + α̃−n · α̃n + D n.
2 n=1 n=1

(iii) OPTIONAL (no need to submit a solution for this part): use the results of (i) to show that the
above equal-time commutators lead to the following commutators among xµ , pν & the oscillators
αnµ , α̃kν with n, k 6= 0,

[xµ , pν ] = iδνµ , [αnµ , αkν ] = n η µν δn+k,0 , [α̃nµ , α̃kν ] = n η µν δn+k,0

[xµ , xν ] = [pµ , pν ] = [xµ , αnν ] = [xµ , α̃nν ] = [pµ , αnν ] = [pµ , α̃nν ] = [αnµ , α̃kν ] = 0

To be submitted on Oct 10th

127
String Theory I Maor Ben-Shahar, Oliver Schlotterer

B Problem set Oct 10th 2022

B.1 Hamiltonian constraints and Polyakov action

The purpose of this exercise is to derive the Polyakov action using Dirac’s method of Hamiltonian con-
straints. We start from the Nambu-Goto action
Z p
SNG [X] = −T d2 σ − det γ ,

where γαβ = ∂α X · ∂β X as well as Ẋµ = ∂τ Xµ and Xµ0 = ∂σ Xµ .

δSNG
(i) Compute the canonical momenta Pµ = δ Ẋ µ
and prove the results to obey the constraints

P 2 + (T X 0 )2 = 0 , P · X0 = 0 .

(ii) Given (i), we can define an alternative action which is equivalent to SNG ,
Z n 1 o
Salt [X, P, e, u] = d2 σ Ẋ · P − e(P 2 + (T X 0 )2 ) − uP · X 0 .
2
In Salt the constraints on the momenta are imposed by the Lagrange multipliers e and u, and we
δSalt
will see that this leads to the same equations of motion as SNG . Use the equation δPµ = 0 to solve
for the momenta.

(iii) By substituting the result of (ii) into the action, reduce it to the following form:
Z
T
Salt [X, P, e, u] = − d2 σ e
hαβ ∂α X · ∂β X,
2
 
1 −1 u
hαβ =
e  .
Te u (T e)2 − u2

Comment: This can be rewritten in the form of the familiar Polyakov action
√
Z
T
SP [X, h] = − d2 σ − det h hαβ ∂α X · ∂β X ,
2

where hαβ = Ωe
hαβ is now an unconstrained worldsheet metric (the dependence on the scaling variable Ω
drops out). Since SP [X, h] is known to yield the same equations of motion as SNG [X], we conclude that
also Salt [X, P, e, u] does.

B.2 Lightcone quantization

This exercise aims to fill in some intermediate steps and examples from the discussion of lightcone
quantization in the lecture.

(i) Check that, in closed-string normalization of pµ ,

D−2 D−2
1 X 1 X
∂+ XL− = ∂+ X i ∂+ X i , −
∂− XR = ∂− X i ∂− X i
α0 p+ i=1 α0 p+ i=1

128
 imply the following at the level of the mode expansion
D−2
1 X  α0 i i X i i 
α 0 p− = pp + αn α−n
p+ i=1 2
n6=0
r D−2
− 1 1 XX i
αn = αn−k αki , n 6= 0
2α0 p+ i=1k∈Z

and the same equations with αni ↔ α̃ni .

(ii) Show that there are 61 (D−2)(D−1)(D+6) states with eigenvalue +3 under the oscillator number
PD−2 P∞ i i
i=1 n=1 α−n αn . Explain that this matches the degree-of-freedom counting for a symmetric
traceless 3-tensor and an antisymmetric 2-tensor of the little group SO(D−1). You can disregard
the α̃n oscillators in this exercise (unless you would like to square the above numbers).

B.3 Virasoro algebra

Upon quantization of the closed bosonic string, the energy-momentum modes are defined with the fol-
lowing ordering prescription of the oscillators αnµ
∞
1 2 X 1X
L0 = α0 + α−n · αn , Lk6=0 = αk−n · αn
2 n=1
2
n∈Z

with analogous definitions for L̃k with αnµ → α̃nµ . The goal of this exercise is to derive their commutation
relations – you will find a special case of the Virasoro algebra, where the central charge c equals the
number of spacetime dimensions D.

(i) use the commutator identity [AB, C] = A[B, C] + [A, C]B for arbitrary operators A, B, C to derive
the lemma
µ µ
[Lk , αm ] = −mαm+k , m, k ∈ Z .

(ii) use the lemma to verify that (note the essential restriction m+n 6= 0)

[Lm , Ln ] = (m−n)Lm+n , m, n ∈ Z, m+n 6= 0 .

(iii) relax the restriction m+n 6= 0 by showing that

D 3
[Lm , L−m ] = 2mL0 + (m − m) .
12
(you will need to pay particular attention to the operator ordering in L0 )

(iv) convince yourself that the above results imply the Virasoro algebra
c
[Lm , Ln ] = (m−n)Lm+n + (m3 −m)δm+n,0 , m, n ∈ Z
12
at c = D and check that it is consistent with the Jacobi relations

[[Lm , Ln ], Lp ] + [[Ln , Lp ], Lm ] + [[Lp , Lm ], Ln ] = 0 , m, n, p ∈ Z .

(v) show that the linear part of the central term ∼ c(m3 −m) can be removed by a redefinition L̂m =
Lm + αδm,0 with α ∈ C, i.e. verify that
Dm3
 D 
[L̂m , L̂n ] = (m−n)L̂m+n + δm+n,0 −m + 2α , m, n ∈ Z .
12 12

129
B.4 Zeta function regularization
This exercise aims to strengthen your confidence in the regularized value for the formally divergent sum
∞
X 1
n=−
n=1
12

in the computation of the normal-ordering constant. We will relate it to a special value of the Riemann
zeta function ζ(s) and the Dirichlet eta function η(s) which are defined by infinite sums
∞
X 1 1 1 1 1
ζ(s) = =1+ + s + s + s + ... , Re(s) > 1 (B.1)
ks 2s 3 4 5
k=1
∞
X (−1)k−1 1 1 1 1
η(s) = s
= 1 − s + s − s + s − ... , Re(s) > 0 (B.2)
k 2 3 4 5
k=1

which converge in the specified ranges of s ∈ C.

(i) show that the eta and zeta functions in their domain of convergence Re(s) > 1 are related by

(1 − 21−s ) ζ(s) = η(s) . (B.3)

hint: write 2−s ζ(s) as 1

2s + 1
4s + 1
6s + . . . and relate this to the even terms in (B.1) and (B.2)
1
(ii) check that η(−1) = 4 can be obtained from a one-sided limit x ↑ 1 of
1
= 1 − 2x + 3x2 − 4x3 + 5x4 − . . .
(1 + x)2
which converges as x ∈ R approaches 1 from below.

(iii) One defines the analytic continuation of ζ(s) to regions with Re(s) ≤ 1 by insisting that the
functional relation (B.3) holds even if the defining series diverges. Use this to infer the regularized
P∞ 1
value n=1 n = − 12 from η(−1) = 41 .

(iv) Repeat the renormalization procedure introduced in the lecture to obtain

∞
Xq2 q 1
(n+q) = −
+ −
n=0
2 2 12
0 ∂
P∞ −(n+q)
from the coefficient of  in the Laurent expansion of − ∂ n=0 e around  = 0.

B.5 Mixing Neumann and Dirichlet boundary conditions

The goal of this exercise is to derive the mass spectrum for open strings subject to mixed Neumann and
Dirichlet boundary conditions in some of the spacetime directions. Let dN and dD denote the number of
dimensions with Neumann and Dirichlet boundary conditions for both endpoints, then impose

∂σ X a (τ, σ = 0) = 0 , X a (τ, σ = π) = ca , dN +dD ≤ a ≤ D−1

corresponding to Neumann conditions at σ = 0 and Dirichlet conditions at σ = π.

(i) start from an ansatz in the usual lightcone coordinates σ ± = τ ± σ on the worldsheet
r
xa 0 a + α0 X 1 +
a +
XL (σ ) = + α pL σ + i α̃r e−irσ
2 2 r
r∈Ξ
r
a 0
x α X1 −
XRa
(σ − ) = + α0 paR σ − + i αr e−irσ
2 2 r
r∈Ξ

130
 with some a priori unknown summation range Ξ for r and show that the boundary conditions enforce

x a = ca , paL = paR = 0 , α̃ra = αra , 1 1


Ξ=Z− 2 = n− 2 : n∈Z

(ii) following the open-string conventions of the lecture, pick lightcone gauge ∂± X + = α0 p+ with X ± =
q
1 0 dN −1
2 (X ± X ) and impose boundary conditions

∂σ X i (τ, σ = 0) = 0 , ∂σ X i (τ, σ = π) = 0 , 0 ≤ i ≤ dN −1
X I (τ, σ = 0) = cI , X I (τ, σ = π) = dI , dN ≤ I ≤ dN +dD −1 .

show that the energy-momentum constraints yield

 dX N −2 dN +dD −1
0 − 1 02 1 X
αp = 0 +
α pi pi + 2 (cI − dI )2
2α p i=1
4π
I=dN

0 dN −2 dN +dD −1  α0 X D−1 
α X X X X
+ αni α−n
i
+ αnI α−n
I
+ αra α−r
a
2 n∈Z i=1 2 1
I=dN r∈Z− 2 a=dN +dD
n6=0

(iii) bring the oscillators into normal-ordered form and use the result of exercise B.4 to derive the mass
spectrum
dN +dD −1
1 X 2−D D−dN −dD
α0 m2 = (cI − dI )2 + +
4π 2 α0 24 16
I=dN

X N −2
∞  dX dN +d
X D −1  X D−1
X
i
+ α−n αni + I
α−n αnI + a
α−r αra .
n=1 i=1 I=dN r∈N− 12 a=dN +dD

1 1 3 5
Note the difference between the summation ranges r ∈ N − 2 = { 2 , 2 , 2 , . . .} in the normal ordered
0 2 1 0 −
expression for α m and r ∈ Z − 2 in the earlier result for α p .

To be submitted on Oct 26th

131
String Theory I Maor Ben-Shahar, Oliver Schlotterer

C Problem set Oct 28th 2022

C.1 Conformal algebra in general dimensions d

The conformal group in d spacetime dimensions is generated by translations Pµ , Lorentz rotations
Jµν , dilatations D and special conformal transformations Kµ with µ, ν = 0, 1, 2, . . . , d−1 and ηµν =
diag(−1, 1, 1, . . . , 1). They give rise to the following group action on scalar functions:

Pµ = −i∂µ , Kµ = −ix2 ∂µ + 2ixµ x · ∂

Jµν = i(xµ ∂ν − xν ∂µ ) , D = −ixµ ∂µ

(i) Compute all commutators among {Pµ , Kµ , Jµν , D}.

(ii) In order to bring these commutators into a simpler form, define

1 1
Jµν = Jµν , J−2,−1 = D , J−2,µ = (Pµ + Kµ ) , J−1,µ = (Pµ − Kµ )
2 2
with Jab = −Jba and a, b = −2, −1, 0, 1, . . . , d−1. Show that the Lie algebra spanned by these Jab
is isomorphic to that of SO(2, d), i.e. that the metric ηab in

[Jab , Jcd ] = i(ηbc Jad − ηac Jbd − ηbd Jac + ηad Jbc )

has two negative entries.

(iii) Consider the Maxwell action in a d-dimensional spacetime with metric gµν ,
Z
1 p
SMax [A, g] = − dd x − det g F µν Fµν .
4
Show that conformal transformations gµν (x) → Ω2 (x)gµν (x) leave SMax [A, g] invariant if the number
of spacetime dimensions is chosen to be d = 4. The Maxwell potential Aµ and its field strength
Fµν = ∂µ Aν − ∂ν Aµ are understood to be unaffected by the conformal transformation.

(iv) Compute the energy-momentum tensor associated with SMax [A, g] and check that its trace vanishes
if and only if d = 4.

C.2 The free fermion

Consider a free fermion ψ(z) which transforms like a primary of weight (h, h̄) = ( 21 , 0) under conformal
transformations and is taken to anticommute in the classical theory in the sense that

ψ(z)ψ(w) = −ψ(w)ψ(z) , ψ(z)∂w ψ(w) = −(∂w ψ(w))ψ(z) classically .

(i) In passing to the quantum theory in the operator formalism, check that the general prescription for
mode expansions given in the lecture yields a sum over ψr z −r−1/2 . Explain the consequences for
the conformal field theory on the cylinder and the plane if the summation range for r is adjusted
to be over half-odd integers,
X
ψ(z) = ψr z −r−1/2 .
r∈Z−1/2

132
 1
(ii) Use this mode expansion with r ∈ Z − 2 to simplify the state-correspondent limz→0 ψ(z)|0i and
explain why this limit is finite.

(iii) You will see in later lectures on the superstring that canonical quantization of the fermionic world-
sheet action leads to anticommutation relations

1
{ψr , ψs } = δr+s,0 , r, s ∈ Z − .
2
Use this and ψr† = ψ−r to derive the two-point function

1
hψ(z)ψ(w)i = h0|ψ(z)ψ(w)|0i =
z−w
and explain why radial ordering is crucial in this computation.

(iv) Use the resulting OPE for ψ(z)ψ(w) to compute the singular terms in the OPE of ψ(w) with the
energy-momentum tensor
1
T (z) = : (∂z ψ(z))ψ(z) :
2
and explain why the result is in lines with the conformal weight of ψ(z). Note that the application of
the Wick theorem to fermionic fields introduces a minus sign whenever a contraction ψ(z) . . . ψ(w) →
1
z−w . . . is performed over an odd number of fermions in the ellipsis, e.g. that

: ψ1 ψ2 : : ψ3 ψ4 : ∼ ψ2 ψ3 ψ1 ψ4 − ψ1 ψ3 ψ2 ψ4 + : ψ2 ψ3 : ψ1 ψ4 + : ψ1 ψ4 : ψ2 ψ3
| {z } | {z } | {z } | {z } | {z } | {z }
− : ψ1 ψ3 : ψ2 ψ4 − : ψ2 ψ4 : ψ1 ψ3 ,
| {z } | {z }

where the contraction ψi ψj instructs to replace ψi ψj by their two-point function.

| {z }
(v) Determine the singular terms in the OPE T (z)T (w) and deduce that the central charge of the free
fermion is c = 21 .

C.3 OPEs versus (anti-)commutation relations

This exercise guides you to through additional examples of the equivalence of (anti-)commutation relations
with OPEs.

(i) Use the contour-integral techniques introduced in the lecture to derive the anticommutation relations
dz r−1/2
H
{ψr , ψs } = δr+s,0 of the free-fermion modes ψr = BR (0) 2πi z ψ(z) from the OPE

1
ψ(z)ψ(w) ∼ + ... .
z−w
Note that by the anticommuting nature of fermions, one identifies ψ(w)ψ(z) at |w| > |z| with minus
the radially ordered product ψ(z)ψ(w).

(ii) Use the same strategy to reproduce the Virasoro algebra of the energy-momentum-tensor modes
dz m+1
H
Lm = BR (0) 2πi z T (z) in a generic CFT from the OPE

c/2 2 T (w) ∂w T (w)

T (z)T (w) ∼ + + + ... .
(z − w)4 (z − w)2 z−w

133
C.4 Vertex operators
This exercise aims to familiarize you with vertex operators, conformal primary fields built from normal
ordered exponentials of the free boson. As you know from the lecture, the simplest example of a vertex
operator
Vp (z, z̄) = : eip·X(z,z̄) :

is the field-correspondent of the oscillator ground state |0; pi with momentum eigenvalue pµ .

(i) Show that the singular terms in the OPE ∂X µ (z) Vp (w, w̄) are

i α0 pµ
∂X µ (z) Vp (w, w̄) ∼ − Vp (w, w̄) + . . .
2 z−w
by series-expanding the exponential in intermediate steps.

(ii) Use the same reasoning to determine the singular terms in the OPE T (z) Vp (w, w̄) and deduce that
0 0
Vp (w, w̄) is a conformal primary of weight (h, h̄) = ( α4 p2 , α4 p2 ).

(iii) As another example of a vertex operator, consider the field

Vp,ζ (z, z̄) = ζµν : ∂z X µ (z)∂z̄ X ν (z̄)eip·X(z,z̄) :

characterized by a momentum vector pµ and a polarization tensor ζµν . Under which conditions on
pµ and ζµν does Vp,ζ (z, z̄) furnish a conformal primary field of weight (h, h̄) = (1, 1)?

C.5 Correlations functions and OPEs

This exercise aims to familiarize you with the techniques to reduce correlation functions involving de-
scendant fields to those of the associated conformal primaries.
(−1)
(i) Use the results derived in the lecture to show that the descendant fields φhi (w) = ∂w φhi (w) and
(−k) dz
(z − w)1−k T (z)φhi (w), k ≥ 1 of primaries φhi (w) give rise to the two-point
H
φhi (w) = B (w) 2πi
functions (with shorthand zab = za −zb )

(−1) 2hi dij (−2) 3hi dij

hφhi (z1 )φhj (z2 )i = − δ
2hi +1 hi ,hj
, hφhi (z1 )φhj (z2 )i = δ
2hi +2 hi ,hj
z12 z12

as well as the three-point function

(−2) Cijk
hφhi (z1 )φhj (z2 )φhk (z3 )i = hi +hj −hk hj +hk −hi hi +hk −hj
z12 z23 z13
 
2hi − hj + hk 2hj − hi + hk hk − hi − hj
× 2 + 2 + ,
z13 z23 z13 z23
where dij and Cijk are the normalizations of the two- and three-point functions among the primaries.

(ii) Given the generic form of the OPE among primary fields
X Cij k n
(−1)
φhi (z1 )φhj (z2 ) ∼ hi +hj −hk
φhk (z2 ) + z12 βijk ∂z2 φhk (z2 )
primary k z12
o
2 (−1,−1) 2 2 (−2) (−2) 3
+ z12 βijk ∂z2 φhk (z2 ) + z12 βijk φhk (z2 ) + O(z12 ) ,

134
 show that the Laurent expansion of the three-point function hφhi (z1 )φhj (z2 )φhk (z3 )i around z1 = z2
imposes
(−1) hi + hk − hj
βijk =
2hk
and constrains the coefficients at the second level to obey

(−1,−1) (−2) 1
2hk (2hk + 1)βijk + 3hk βijk = (hi + hk − hj )(hi + hk − hj + 1) .
2

(−2)
(iii) Explain without calculation why you would need the three-point function hφhi (z1 )φhj (z2 )φhk (z3 )i
(−1,−1)
(or some other correlator involving a higher descendant of φhk ) in order to determine βijk and
(−2) (−1,−1) (−2)
βijk individually. Why does this imply that βijk and βijk depend on the central charge c on
top of the conformal weights hi , hj , hk ?
(−2)
Hint: Can you distinguish ∂z22 φhk (z2 ) from a multiple of φhk (z2 ) through the respective two-point
function with φh` (z3 ) or ∂zN3 ≥1 φh` (z3 )?

To be submitted on Nov 11th

135
String Theory I Maor Ben-Shahar, Oliver Schlotterer

D Problem set Nov 10th 2022

D.1 The delta function in complex coordinates

This exercise validates the representation
1
∂z̄ ∂z log |z|2 = ∂z̄ = πδ 2 (z, z̄)
z
of the delta function δ 2 (z, z̄) = δ(x)δ(y) for z = x + iy with x, y ∈ R in two different ways.

1
(i) Use the volume form d2 z = dx ∧ dy = 2i dz̄ ∧ dz and Stokes’ theorem for p-forms ω
Z I
dω = ω
Ω ∂Ω

to show that ∂z̄ z1 = πδ 2 (z, z̄) is consistent with integrating ∂z̄ ∂z log |z|2 over any open region Ω ⊂ C
that contains the origin.

(ii) As an alternative validation, study the deformation

1 z
∂z = ∂z lim 2 .
z̄ →0 |z| + 2

Explain why the sequence of distributions

2
∆ (z, z̄) =
(|z|2 + 2 )2

approaches πδ 2 (z, z̄) as  → 0 and conclude that this implies the desired relation ∂z z̄1 = πδ 2 (z, z̄).
More precisely, show that pointwise, lim→0 ∆ (z, z̄) = 0 ∀ z 6= 0 and study the integral of ∆ (z, z̄)
over disks 0 ≤ |z| ≤ R in spherical coordinates d2 z = r dr dϕ for z = reiϕ with r ∈ R+ , ϕ ∈ [0, 2π).

D.2 Anomalous current and two-point function of the (b, c)-ghost system
In this exercise, you are asked to verify certain fingerprints of the background charge of the (b, c)-ghosts.

(i) Using the expressions

T (z) = 2 : ∂z c(z) b(z) : + : c(z)∂z b(z) : , j(z) = − : b(z)c(z) :

for the energy-momentum tensor T (z) and the current j(z) of the (b, c)-system, derive the OPE
−3 j(w) ∂w j(w)
T (z)j(w) ∼ + + + ... ,
(z − w)3 (z − w)2 z−w

where the triple-pole prevents j(z) from being a conformal primary field.

(ii) Using the canonical anticommutation relations of the b- and c-modes

{bm , bn } = {cm , cn } = 0 , {bm , cn } = δm+n,0 ,

derive the two-point function

1
h0|c−1 c0 c1 b(z)c(w)|0i =
z−w

136
 in the operator formalism. In other words, insert the mode expansion
X X
b(z) = bn z −n−2 , c(z) = cn z −n+1
n∈Z n∈Z

and note the following properties of the vacuum (including its 3 units of background ghost charge):

bn≥−1 |0i = cn≥2 |0i = 0 , h0|bn≤1 = h0|cn≤−2 = 0 , h0|c−1 c0 c1 |0i = 1

(iii) Show that the OPEs of b(z)c(w) as well as j(z)c(w) and j(z)b(w) are reproduced by the “bosonized”
representations
j(z) = ∂z φ(z) , b(z) = : e−φ(z) : , c(z) = : eφ(z) : ,

where φ(z) is a chiral boson defined by short-distance singularities φ(z)φ(w) ∼ log(z − w) + . . ..

(iv) Determine the constants u1 , u2 ∈ Q in the bosonized representation of the energy-momentum tensor
of the ghost system

2
T (z) = u1 : ∂z φ(z) : +u2 ∂z2 φ(z) .

(v) Show that : eqφ(z) : generate eigenstates |qi = limz→0 :eqφ(z) :|0i of the zero mode j0 of the anomalous
ghost current j(z) as specified in the lecture.

(vi) Show that : eqφ(z) : are conformal primaries and determine their conformal weight.

D.3 Massless open-string states from BRST quantization

The goal of this exercise is to find the physical states of the massless open bosonic string in the framework
of BRST quantization, following the methods introduced in the tutorial on November 16th . We will see
that the number of degrees of freedom matches the expectation from lightcone quantization. You will
find the conformal weights (h, h̄) = (2, 0) & (−1, 0) for b & c and the following singular terms in the
OPEs useful
1
b(z)c(w) ∼ .
z−w
In addition, the BRST operator Q takes the form
I
dz
Q= (cT + bc∂c) .
2πi
(i) Write down the most general ansatz for an unintegrated massless vertex operator compatible with
conformal weight zero and built from ∂z X, c, b and possibly their derivatives.

Hint: You should arrive at two vectors and four scalars that parametrize the ansatz.

(ii) Constrain this state to be annihilated by b0 , where

I
dz
b0 = zb .
2πi

(iii) Identify the BRST closed states in the above.

(iv) Subtract all terms that are BRST exact, and explain why your result is compatible with the d.o.f.
counting in lightcone quantization.

137
D.4 Plane-wave correlators in presence of derivatives
You have seen in the lecture how path-integral methods can be used to determine the plane-wave correlator
(also known as Koba–Nielsen factor)
DYn E n
X n
 Y 0
: eipj ·X(zj ) : = δ D pµj |zij |α pi ·pj ,
j=1 j=1 1≤i<j

where zij = zi − zj , and the momentum-conserving delta function will be suppressed in the following.
The goal of this exercise is to extend these techniques to correlators involving derivatives ∂X µ (zj ) :=
∂zj X µ (zj ) that are ubiquitous in vertex operators. The idea is to evaluate the generating function
n
DY E
Gn (pj , ξj , zj ) := : eipj ·X(zj )+iξj ·∂X(zj ) :
j=1

of such correlators, where the vectors ξjµ with j = 1, 2, . . . , n are formal bookkeeping variables.

(i) Use the strategy of the lecture to rewrite the exponentials as a source term in the worldsheet action,
 Z 
eipj ·X(zj )+iξj ·∂X(zj ) = exp i d2 z Xµ (z) pµj δ 2 (z − zj ) − ξjµ ∂z δ 2 (z − zj ) ,


to show that
n n
 α0 X hξ · p
Y 0 i j ξj · pi ξi · ξj i
Gn (pj , ξj , zj ) = |zij |α pi ·pj exp + + 2 .
2 zij zji zij
1≤i<j 1≤i<j

Hint: make use of the Gaussian integral

Z Z
1 h 1 i
D[X] exp d2 z X µ (z)∂ z ∂ z̄ X µ
(z) + iX µ (z)J µ
(z)
Z πα0
 α0 Z Z 
= exp d2 z d2 w J µ (z) log |z − w|2 Jµ (w)
4
and exploit that contractions pi · ξj and ξi · ξj with i = j are excluded by normal ordering.

(ii) Explain why the “component correlators”

DYn
 mj ipj ·X(zj ) E
: ξj · i∂X(zj ) e : , m1 , m2 , . . . , mn ∈ N0
j=1

can be obtained by isolating the multilinear terms ∼ (ξ1 )m1 (ξ2 )m2 . . . (ξn )mn in the ξ-expansion of
the exponentials in Gn (pj , ξj , zj ) and multiplying by m1 !m2 ! . . . mn !. Furthermore explain why these
 m
component correlators with insertions : ξ · i∂X(z) eip·X(z) : are sufficient to infer correlators of
: ζµ1 µ2 ...µm i∂X µ1 (z)i∂X µ2 (z) . . . i∂X µm (z)eip·X(z) : with arbitrary polarization tensors ζµ1 µ2 ...µm .

(iii) Check that Gn (pj , ξj , zj ) obeys the differential equation

n
∂ α0 X  pµk ξµ 
Gn (pj , ξj , zj ) = + 2k Gn (pj , ξj , zj )
∂(ξ` )µ 2 k=1 z`k z`k
k6=`

with `=1, 2, . . . , n, and show that it implies the recursion or Ward identity for component correlators
D n
Y mj ipj ·X(zj ) E
:i∂X µ (z` )eip` ·X(z` ) :

: ξj · i∂X(zj ) e :
j=1
j6=`
n  n
α0 X pµk D ip` ·X(z` ) Y  mj ipj ·X(zj ) E
= :e : : ξj · i∂X(zj ) e :
2 k=1 z`k j=1
k6=` j6=`

n
mk ξkµ D ip` ·X(z` )

 mk −1 ip ·X(z ) Y  mj ipj ·X(zj ) E
+ 2 :e : : ξk · i∂X(zk ) e k k
: : ξj · i∂X(zj ) e : .
z`k j=1
j6=k,`

138
 Note that this resembles the Ward identity for correlators involving energy-momentum insertions
and otherwise conformal primaries: The recursion instructs to replace the insertion of i∂X µ (z` )
µ
α0 p
within a correlator by a sum over all its OPE singularities i∂X µ (z` ):eipk ·X(zk ) : ∼ 2 z`k :e
ipk ·X(zk )
:+
µν
α0 η
. . . and i∂X µ (z` )i∂X ν (zk ) ∼ 2 z`k2 + . . . with the remaining fields. Like this, all the correlators on
the right-hand side have fewer insertions of i∂X µ (z).

(iv) Determine the three-point correlators

3
DY E 2
D Y  E
: eipj ·X(zj ) : , : eipj ·X(zj ) : : ξ3 · i∂X(z3 )eip3 ·X(z3 ) : ,
j=1 j=1
D 3
Y E 3
DY E
: eip1 ·X(z1 ) : : ξj · i∂X(zj )eipj ·X(zj ) : , : ξj · i∂X(zj )eipj ·X(zj ) : .
j=2 j=1

Hint: You can cross-check your results by testing the above Ward identity with ` = n = 3 (though
you are not expected to include this check into your solution).

¯ µ (zj ) = ∂z̄ X µ (zj ), i.e. show that

(v) Generalize the generating functions to correlators involving ∂X j

n
DY E
¯
Hn (pj , ξj , zj ) := : eipj ·X(zj )+iξj ·∂X(zj )+iξ̄j ·∂X(zj ) :
j=1

with additional bookkeeping variables ξ¯jµ , j = 1, 2, . . . , n can be written in the factorized form
n n
 α0 X hξ · p
Y 0 i j ξj · pi ξi · ξj i
Hn (pj , ξj , zj ) = |zij |α pi ·pj exp + + 2
2 zij zji zij
1≤i<j 1≤i<j
n h ξ¯ · p
 α0 X
i j ξ¯j · pi ξ¯i · ξ¯j i
× exp + + 2 .
2 z̄ij z̄ji z̄ij
1≤i<j

0
Hint: You will need to exploit that |zij |α pi ·pj vanishes as zi → zj (upon analytic continuation in
¯
α0 pi ·pj ). Note that the exponentials : eipj ·X(zj )+iξj ·∂X(zj ) : and : eipj ·X(zj )+iξj ·∂X(zj )+iξ̄j ·∂X(zj ) : are
generating functions for certain open- and closed-string vertex operators. Hence, the factorized form
of Hn (pj , ξj , zj ) implies that these closed-string correlators are products of open-string correlators
(up to the Koba–Nielsen factor and the rescaling α0 → 4α0 mentioned in the lecture).

(vi) OPTIONAL (no need to submit a solution for this part): Comment on how you would modify the
above strategy to also compute correlators involving higher derivatives ∂ N X µ (zj ) and ∂¯N X µ (zj )
with N ≥ 2. Explain why their generating functions will still take a factorized form.

To be submitted on Nov 25th

139
String Theory II Maor Ben-Shahar, Oliver Schlotterer

E Problem set Dec 13th 2022

E.1 Three tachyons and one gluon or graviton

On this sheet, you will need the vertex operators of the tachyon and the non-abelian gauge boson of the
open bosonic string
1
VT (z) ∼ :eip·X(z) : , p2 =
α0
V (z) ∼ µ :i∂Xµ (z)eip·X(z) : , p2 =  · p = 0

and the Euler Beta function Z 1

Γ(a)Γ(b)
dz z a−1 (1 − z)b−1 = .
0 Γ(a + b)
(i) Use the above vertex operators and the SL2 -frame (z1 , z3 , z4 ) → (0, 1, ∞) to bring the color-ordered
four-point amplitude
Z 1
0
Atree
bos (1 , T2 , T3 , T4 ; α ) ∼ dz2 hcV1 (z1 )VT2 (z2 )cVT3 (z3 )cVT4 (z4 )i
0

into the following form (s = α0 (p1 + p2 )2 and t = α0 (p2 + p3 )2 ):

0 0
h Γ(s − 1)Γ(t − 1) Γ(s)Γ(t − 1) i
Atree
bos (1 , T2 , T3 , T4 ; α ) ∼ −2α 1 · p2 + 1 · p3
Γ(s + t − 2) Γ(s + t − 1)

(ii) Use Γ-function identities to show that this expression enjoys linearized gauge invariance, i.e. that it
vanishes when setting 1 → p1 .

(iii) Show that your result is antisymmetric under exchange of the second and fourth tachyon,

0 0
Atree tree
bos (1 , T2 , T3 , T4 ; α ) = −Abos (1 , T4 , T3 , T2 ; α ) ,

i.e. show that your result changes by a sign when swapping p2 ↔ p4 .

Hint: Eliminate (1 · p3 ) via momentum conservation and transversality.

E.2 Four-gluon amplitude

This exercise gives you a glimpse of the four-gluon amplitude of the open bosonic string, based on the
SL2 -frame where (z1 , z3 , z4 ) → (0, 1, ∞),
Z 1
0
Atree
bos (1 , 2 , 3 , 4 ; α ) ∼ dz2 hcV1 (z1 )V2 (z2 )cV3 (z3 )cV4 (z4 )i .
0

(i) Show that the contribution of the tensor structure (1 ·2 )(3 ·4 ) to the correlator (excluding ghosts)
is proportional to
4
Y 0 1
|zij |2α pi ·pj × 2 z2 .
z12 34
1≤i<j

(ii) Perform the integral over z2 in terms of Γ function and use

Γ(1 + s)Γ(1 + t)
= 1 − ζ2 st − ζ3 stu + O(α04 )
Γ(1 + s + t)
to expand your result to the third subleading order in α0 .

140
(iii) Explain that your results signal an effective interaction of the form Tr(F 3 ), two independent inter-
actions of the form Tr(F 4 ) and three independent interactions of the form Tr(D2 F 4 ), where F µν is
the non-abelian gluon field strength and Dµ is the gauge-covariant derivative.

(iv) Compute the contribution of the tensor structure (1 · 2 )(3 · 4 ) to the alternative color-ordered
0 0
amplitudes Atree tree
bos (2 , 1 , 3 , 4 ; α ) and Abos (1 , 3 , 2 , 4 ; α ) by integrating over z2 ∈ (−∞, 0) and
z2 ∈ (1, ∞), respectively.
y
Hint: Use variable transformations z2 = y−1 and z2 = x1 to map the integration regions to x, y ∈ (0, 1).

(v) Assuming that the remaining tensor structures exhibit the same cancellations as (1 · 2 )(3 · 4 ),
deduce that only one of the Tr(F 4 ) and one of the Tr(D2 F 4 ) interactions have a non-vanishing
abelian limit when the Chan-Paton factors are specialized to T aj → 1.

E.3 Gamma function versus Riemann zeta function

The Γ function can be alternatively defined through one of the following infinite products
∞ ∞
z
1 −γz Y z/n  z −1 1 Y 1 + n1
Γ(z) = e e 1+ =
,
z n=1
n z n=1 1 + nz

where γ denotes the Euler–Mascheroni constant

∞   N
X 1  1 X 1 
γ= − log 1 + = lim − log N ≈ 0.5772 . . .
n=1
n n N →∞
n=1
n

that will drop out from any application to string tree-level amplitudes.

(i) Show that the logarithmic derivative of the Γ function can be written as
∞ 
X 1 1 
∂z log Γ(1 + z) = −γ + − .
n=1
n n+z

Hint: Use zΓ(z) = Γ(z + 1) in the first step.

(ii) By integrating the previous identity from 0 to z, derive the following series representation
∞
X (−z)k
log Γ(1 + z) = −γz + ζk ,
k
k=2

where the Riemann zeta values are defined by

∞
X 1
ζk = k
, k ≥ 2.
n=1
n

1 1
P∞ z k
Hint: Perform a geometric-series expansion of n+z = n k=0 (− n ) in the integrand.

(iii) The sine function can be written in terms of the Γ functions

πz
sin(πz) = .
Γ(1 + z)Γ(1 − z)

Explain how the zeros and the periodicity sin(π(z ± 1)) = − sin(πz) of the sine function can be
derived from this representation.

141
(iv) Using the Γ-function representation of the sine function and the above series expansion of log Γ(1+z),
show that ∞
 X ζ2n 2n 
sin(πz) = πz exp − z .
n=1
n

(v) By matching this representation with the more conventional Taylor expansion of the sine function
x3 x5
sin(x) = x − 6 + 120 + O(x7 ), deduce that
π2 π4
ζ2 = , ζ4 = .
6 90
Note: This method implies that any even zeta value ζ2k is a rational multiple of π 2k . However, one
cannot extract any information on odd zeta values ζ2k+1 from these expansions of the sine function,
ζ3 ζ 5
and all of π3 , π5 , . . . are strongly expected (though not proven) to be irrational.

E.4 Monodromy relations among color-ordered open-string amplitudes

This exercise is dedicated to the monodromy relations among color-ordered open-string amplitudes
0
Atree
∗ (ϕ1 , ϕ2 , . . . , ϕn ; α ). The subscript ∗ is a placeholder for either bosonic strings or superstrings,
and the external states ϕj are taken to be massless such that s = 2α0 p1 · p2 and t = 2α0 p2 · p3 . The goal is
to relate different color orderings or integration regions for the punctures z2 , z3 , . . . , zn−2 in an SL2 -frame
where (z1 , zn−1 , zn ) = (0, 1, ∞). This will be achieved by exploiting that the underlying correlator among
0
vertex operators is meromorphic except for the universal Koba–Nielsen factor ∼ |zij |2α pi ·pj .
At four points, the above SL2 -frame leads to the color-ordered amplitudes
Z 0
0
Atree
∗ (ϕ2 , ϕ1 , ϕ3 , ϕ4 ; α ) ←→ dz2 |z2 |s |1 − z2 |t f∗ (z2 )
−∞
Z 1
0
Atree
∗ (ϕ1 , ϕ2 , ϕ3 , ϕ4 ; α ) ←→ dz2 |z2 |s |1 − z2 |t f∗ (z2 )
Z0 ∞
0
Atree
∗ (ϕ1 , ϕ3 , ϕ2 , ϕ4 ; α ) ←→ dz2 |z2 |s |1 − z2 |t f∗ (z2 ) ,
1
where the polarization dependent rational function f∗ (z2 ) is determined by the contractions among the
vertex operators, and its detailed form will not be needed in the following.
You will be guided through a proof of the monodromy relations that relies on applying Cauchy’s
theorem to the meromorphic function F∗ (z2 ) = (z2 )s (1 − z2 )t f∗ (z2 ) with rational f∗ (z2 ):
I
dz2 F∗ (z2 ) = 0 .
C
The integration contour C is depicted in figure E.1 below and consists of a semi-circle C∞ at infinity as
well as the real axis with the following shorthands

C2134 = (−∞, 0) , C1234 = (0, 1) , C1324 = (1, ∞) .

(Strictly speaking, the Cijkl should be infinitesimally displaced from the real axis by some positive imag-
inary part i to avoid tentative poles of f∗ (z2 ) at z2 = 0 and z2 = 1.)
H
(i) Assuming that the semi-circle C∞ does not contribute to C dz2 F∗ (z2 ), explain why the integration
can be performed using the representation

iπs s t
 e |z2 | |1 − z2 | f∗ (z2 ) : z2 ∈ C2134


F∗ (z2 ) = s t
|z2 | |1 − z2 | f∗ (z2 ) : z2 ∈ C1234

−iπt
 s t
e |z2 | |1 − z2 | f∗ (z2 ) : z2 ∈ C1324


142
 Im(z2 )

C∞

C2134 1 C1324 Re(z2 )

• • • •
−∞ 0 C1234 +∞

Figure E.1: The contour C consists of the semicircle C∞ drawn in blue and subsets C2134 , C1234 , C1324 of
the real line drawn in red.

of the above integrand F∗ (z2 ), and explain why Cauchy’s theorem implies that

Z0 Z1 Z∞
iπs s t s t −iπt
e dz2 |z2 | |1−z2 | f∗ (z2 ) + dz2 |z2 | |1−z2 | f∗ (z2 ) + e dz2 |z2 |s |1−z2 |t f∗ (z2 ) = 0 .
−∞ 0 1

0
(ii) Assuming that color-ordered amplitudes Atree
∗ (ϕi , ϕj , ϕk , ϕl ; α ) are real, deduce that

0 0 0
cos(πs)Atree tree tree
∗ (ϕ2 , ϕ1 , ϕ3 , ϕ4 ; α ) + A∗ (ϕ1 , ϕ2 , ϕ3 , ϕ4 ; α ) + cos(πt)A∗ (ϕ1 , ϕ3 , ϕ2 , ϕ4 ; α ) = 0
0 0
sin(πs)Atree tree
∗ (ϕ2 , ϕ1 , ϕ3 , ϕ4 ; α ) − sin(πt)A∗ (ϕ1 , ϕ3 , ϕ2 , ϕ4 ; α ) = 0 .

(iii) By isolating the leading order in α0 , conclude that color-ordered tree amplitudes among non-abelian
gauge bosons satisfy the Kleiss–Kuijf relations

Atree tree tree

YM (ϕ2 , ϕ1 , ϕ3 , ϕ4 ) + AYM (ϕ1 , ϕ2 , ϕ3 , ϕ4 ) + AYM (ϕ1 , ϕ3 , ϕ2 , ϕ4 ) = 0

and the Bern–Carrasco–Johansson (BCJ) relations

(p1 + p2 )2 Atree 2 tree

YM (ϕ2 , ϕ1 , ϕ3 , ϕ4 ) − (p2 + p3 ) AYM (ϕ1 , ϕ3 , ϕ2 , ϕ4 ) = 0 .

(iv) OPTIONAL (no need to submit a solution for this part): Generalize the integration contour C and
the integrand F∗ (z2 ) → F∗ (z2 , z3 , . . . , zn−2 ) to an n-point setting to derive the monodromy relation
0
0 = e2πiα p1 ·p2 Atree 0 tree 0
∗ (ϕ2 , ϕ1 , ϕ3 , . . . , ϕn ; α ) + A∗ (ϕ1 , ϕ2 , ϕ3 , . . . , ϕn ; α )
0 0
+ e−2πiα p2 ·p3 Atree 0
∗ (ϕ1 , ϕ3 , ϕ2 , ϕ4 , . . . , ϕn ; α ) + e
−2πiα p2 ·(p3 +p4 ) tree
A∗ (ϕ1 , ϕ3 , ϕ4 , ϕ2 , . . . , ϕn ; α0 )
0
+ . . . + e−2πiα p2 ·(p3 +p4 +...+pn−1 ) Atree 0
∗ (ϕ1 , ϕ3 , ϕ4 , . . . , ϕn−1 , ϕ2 , ϕn ; α )

and deduce the n-point BCJ relations

n−1
X
p1 · p2 Atree
YM (ϕ2 , ϕ1 , ϕ3 , . . . , ϕn ) = p2 · (p3 +p4 + . . . +pj )Atree
YM (ϕ1 , ϕ3 , . . . , ϕj , ϕ2 , ϕj+1 , . . . , ϕn ) .
j=3

Note: By combining all permutations of the monodromy- and BCJ relations, color-ordered n-point
tree amplitudes in string and gauge theory can be reduced to a basis of dimension (n−3)!.

To be submitted on Jan 6th

143
String Theory II Maor Ben-Shahar, Oliver Schlotterer

F Problem set Jan 11th 2023

F.1 From string frame to Einstein frame

You have seen in the lecture that the σ-model action for closed strings in background fields {Gµν , Bµν , Φ}
only retains conformal invariance if the following beta functions vanish to first order in α0 :

α0
βµν (G) = α0 Rµν + 2α0 ∇µ ∇ν Φ − H(µ λρ Hν)λρ + O(α02 )
4
α0 λ
βµν (B) = − ∇ Hλµν + α0 (∇λ Φ)Hλµν + O(α02 )
2
D−26 α0 2 α0
β(Φ) = − ∇ Φ + α0 ∇µ Φ∇µ Φ − Hµνρ H µνρ + O(α02 )
6 2 24
(i) Show that the vanishing of these beta functions follows from the equations of motion of the following
effective action in D spacetime dimensions to first order in α0
√
Z
1 n 1 2(D−26) o
Seff [G, B, Φ] = 2 dD X − det G e−2Φ R − Hµνλ H µνλ + 4∂µ Φ∂ µ Φ −
2κ0 12 3α0
The variables Gµν and Φ are those appearing in the σ-model action, that is why the above expression
for Seff [G, B, Φ] is said to be in “string frame”.

(ii) We shall next transform the string-frame effective action into the so-called “Einstein frame”, i.e.
introduce the variables G̃µν appearing in the Einstein–Hilbert action. Perform the field redefinition

2Φ̃
G̃µν = e2ω Gµν , Φ̃ = Φ − Φ0 , ω=−
D−2
to check that Seff [G, B, Φ] is equal to the following Einstein-frame action with κ2 = κ20 e2Φ0 ,
Z
1 p n 1 8Φ̃
Seff [G̃, B, Φ̃] = 2 dD X − det G̃ R̃ − e− D−2 Hµνλ H µνλ
2κ 12
4 µ 4(D−1) 2 4Φ̃ 2(D−26)
o
− ∂µ Φ̃∂ Φ̃ + ∇ Φ̃ − e D−2 0
.
D−2 D−2 3α

You can exploit that the Ricci-scalar R̃ of the transformed metric G̃µν is related as follows to the
untransformed one
 
R̃ = e−2ω R − 2(D−1)∇2 ω − (D−2)(D−1)∂µ ω∂ µ ω .

Note that the indices µ, ν, λ, . . . in the Einstein-frame action Seff [G̃, B, Φ̃] are contracted with G̃µν
rather than Gµν .

(iii) Explain why the term ∼ ∇2 Φ̃ in the second line of Seff [G̃, B, Φ̃] integrates to zero.

F.2 Born-Infeld equations of motion

The goal of this exercise is to extract the field equations from the Born–Infeld action on a Dp-brane
Z p
SBI [A] = −Tp dp+1 X det(δab + Fab ) ,

144
where Tp denotes the brane tension, and the coordinates on the brane have been lined up with the space-
time directions X 0 , X 1 , . . . , X p . In order to avoid technical distractions, we are setting α0 = 2π
1
within
p
0
this exercise and consider the Euclidean version of the expression ∼ − det(ηab + 2πα Fab ) introduced
in the lecture.

(i) Use the matrix-identity det M = exp(Tr log(M )) and antisymmetry Fab = −Fba to show that
p 1

det(δab + Fab ) = exp Tr log(1 − F 2 ) .
4
Hint: Which powers of Fab are traceless?

xn
P∞
(ii) Using the series-expansion log(1 − x) = − n=1 n , expand the result of (i) to the order of F 6
and check that the F 4 -order has the relative factor Tr(F 4 ) − 14 (Tr(F 2 ))2 between the single- and
double-traces.

(iii) Demonstrate at the level of the respective series expansions that

h
i  F cb  F cb
∂a Tr log(1 − F 2 ) = −2∂a Fbc = −4∂b F ac ,
1 − F2 1 − F2
where the second identity follows from the Bianchi identity ∂[a Fbc] = 0.

(iv) Demonstrate that

 F ab  F ac  F db  1 ac  1 db
∂a = ∂ F
a cd + ∂ F
a cd .
1 − F2 1 − F2 1 − F2 1 − F2 1 − F2
Hint: Organize the series expansion of both sides according to the matrix products (F M (∂a F )F N )
with M, N ≥ 0. It is convenient to check separately that the terms on both sides agree in all the
four cases where (M, N ) are (even, even), (odd, odd), (even, odd) and (odd, even), respectively.

(v) Since the Born–Infeld action only depends on the photon Ab through its field strength Fab (and no
derivatives ∂c ∂d . . . Fab ), explain why one can simplify the variation w.r.t. Ab to

δ  δ 
SBI [A] = −2∂a SBI [A] .
δAb δFab

(vi) Follow the manipulation of (iii) to show that

δ  F ab 1

2
SBI [A] ∼ exp Tr log(1 − F )
δFab 1 − F2 4

(vii) Combine the results of (iii) to (vi) to demonstrate that the equations of motion following from the
Born–Infeld action are
 
F ab 1
2


0 = ∂a exp Tr log(1 − F )
1 − F2 4
 1 ac  1 db 1

2
= ∂ a F cd exp Tr log(1 − F ) ,
1 − F2 1 − F2 4
i.e. please justify both equalities. You have now completed the derivation of the Born–Infeld equa-
tions of motion introduced in the lecture since the above identity is equivalent to
 1 ac
∂a Fcd = 0 .
1 − F2

145
F.3 Newton gravity in four versus five spacetime dimensions
This exercises guides you to explore Newton’s gravitational potential in presence of circular or large extra
dimensions and to see the fate of the D-dimensional Newton constant G(D) upon compactification.

(i) Consider the dimension-agnostic Poisson equation

∇2 V (D) (x) = 4πG(D) ρM (x)

∂2
PD−1
relating the spatial derivatives ∇2 = j=1 ∂x2j of the D-dimensional gravitational potential V (D) (x)
to the density ρM (x) = M δ D−1 (x) of a point mass at the origin. Use the divergence theorem to
PD−1
extract the dependence of the spherically symmetric potential on R = ( j=1 x2j )1/2 from the
integral of the Poisson equation over a (D−1)-dimensional ball. More specifically, using the volume

2π (n+1)/2 Rn
vol S n (R) =
Γ( 12 (n+1))

of an n-dimensional sphere of radius R, show that the D-dimensional potential is given by

2π (3−D)/2 Γ( 12 (D−1))M G(D)

V (D) (R) = − , D ≥ 4.
(D−3)RD−3

(ii) Now specialize to D = 5 spacetime dimensions with spatial coordinates (x, y, z, w). Compactify the
w-direction on a circle of radius a, leading to identification w ∼
= w + 2πa. Show that the potential
(D=5)
p
Va (x, y, z, w=0) in the compactified setting as a function of r = x2 + y 2 + z 2 is given by
∞
G(5) M X 1
Va(D=5) (x, y, z, w=0) =−
.
πr2 n=−∞ 1 + 2πna 2
r

Hint: sum over appropriate images of the uncompactified potential V (D=5) .

(iii) By evaluating the infinite sum as

∞  
X 1 1 1
2
= coth
n=−∞
1 + (πnx) x x

(D=5)
show that the limit r  a of Va reduces to the four-dimensional potential V (D=4) and that the
four- and five dimensional Newton constants are related by

G(5)
G(4) =
2πa
as argued in the lecture by dimensional reduction of the Einstein–Hilbert action.
(D=5)
(iv) Now investigate the opposite limit and perform a Laurent expansion of Va for small r  a.
−2 0 2 −2
Determine the terms of order r , r and r and explain how the term ∼ r is related to earlier
quantities seen in this exercise.

F.4 SU (2) covariant approach to massless states

√
In this exercise, the massless states at the self-dual radius r = α0 are reorganized such that the enhanced
SU (2) × SU (2) gauge symmetry is manifested. A key step is to recover two copies of the SU (2) algebra

[ta , tb ] = iεabc tc , a, b, c ∈ {1, 2, 3}

146
of the triplet of generators ta = 12 σ a with
     
1 0 1 1 0 −i 1 1 0
t1 =  , t2 =  , t3 =  
2 1 0 2 i 0 2 0 −1

at the level of the OPEs. For this purpose, we introduce normalized (h, h̄) = (1, 0) and (0, 1) currents
i L
± √2i 0 XD−1 (z)
J 3 (z) = √ ∂z XD−1 (z) , J ± (z) =: e α :
α0
3 i ± R
± √2i 0 XD−1 (z)
J (z) = √ ∂z̄ XD−1 (z) , J (z) =: e α :
α0
L R
whose OPEs follow from those of the chiral halves of XD−1 (τ, σ) = XD−1 (τ +σ) + XD−1 (τ −σ),

L L α0 R R α0
XD−1 (z)XD−1 (w) ∼ − log(z − w) + . . . , XD−1 (z)XD−1 (w) ∼ − log(z̄ − w̄) + . . .
2 2
L R b
Note that there are no mutual singularities between XD−1 (z)XD−1 (w) or J a (z)J (w).

(i) By first deriving the OPEs among {J 3 , J + , J − } and then converting to

J+ + J− J+ − J−
J1 = , J2 = ,
2 2i
+ − + −
1 J +J 2 J −J
J = , J = ,
2 2i
show that the OPEs can be written in SU (2)-covariant5 form as
c
δ ab /2 iεabc J c (w) a b δ ab /2 iεabc J (w)
J a (z)J b (w) ∼ 2
+ + ... , J (z)J (w) ∼ 2
+ + ... .
(z−w) z−w (z̄−w̄) z̄−w̄
The residue of the simple pole reflects the above SU (2) algebra, and the double-pole term corre-
sponds to its so-called central extension (similar to the central term ∼ c in the Virasoro algebra).

(ii) For the two SU (2) triplets of gauge-boson vertex operators

a
Va|1 (z) = µ J a (z) : i∂z̄ X µ (z)eip·X (z) : , V1|a (z) = µ J (z) : i∂z X µ (z)eip·X (z) : (F.1)

work out the linear combinations of their components a = 1, 2, 3 which reproduce the vertex oper-
ators V± , V±L , V±R given in the lecture. The Lorentz index of µ and the dot product p · X in the
exponent exclude the (D−1)-direction.

(iii) Determine the three-point function of the currents J a (z) and use this to compute the three-point
a|1
amplitude of three gauge bosons V (z) in SU (2)-covariant form.

Hint: see the three-gluon amplitude given in the lecture for the polarization dependence.

(iv) The massless closed-string spectrum at the self-dual radius contains 9 scalars with vertex operators
b
V a|b (z) = J a (z)J (z) : eip·X (z) :

that transform in the bi-adjoint representation of SU (2) × SU (2). Determine the momentum- and
winding numbers m, n in (pL , pR ) = ( nr + mr n
α0 , r − mr
α0 ) of their 3 × 3 components and give their
three-point amplitude in (SU (2) × SU (2))-covariant form.

To be submitted on Jan 31st

5 Covariance means that the indices a, b, c on the right-hand side are attached to the invariant tensors δ and ε of SU (2).

147
String Theory II Maor Ben-Shahar, Oliver Schlotterer

G Problem set Jan 27th 2023

G.1 The ground-state energy of the Ramond sector

The goal of this exercise is to derive the ground-state energy in the Ramond sector of the worldsheet
D
fermions ψ µ which amounts to the conformal weight h = 16 of the spin fields SA (z). The Ramond sector
µ 2πi
is defined by antiperiodic boundary conditions ψ (e z) = −ψ µ (z) on the complex plane due to mode
expansions
X
ψ µ (z) = ψrµ z −r−1/2
r∈Z
with integer r as opposed to the half-odd integer r in the Neveu-Schwarz sector. The modes ψrµ obey the
canonical anticommutation relations (equivalent to the OPE ψ µ (z)ψ ν (w) ∼ η µν (z − w)−1 + . . .)

{ψrµ , ψsν } = η µν δr+s,0 , r, s ∈ Z .

(i) Show that the modes of the energy-momentum tensor in the Ramond sector
1 X
Tψ (z) = : ∂z ψ µ (z) ψµ (z): = Lm z −m−2


2
m∈Z

are given by
∞
1X µ X µ
Lm6=0 = r ψm−r ψrν ηµν , L0 = h + r ψ−r ψrν ηµν ,
2 r=1
r∈Z
where you are supposed to ignore terms ψ0µ ψ0ν ηµν for the moment. The expression for L0 features
a commuting offset h ∈ Q which captures the contributions of the zero modes ψ0µ and whose value
will be determined below. The normal-ordering prescription : . . . : in the definition of Tψ (z) moves
µ µ
annihilation operators ψr>0 to the right of creation operators ψr<0 ,

ψrµ ψ−r
ν

: r < 0,
:ψrµ ψ−r
ν
:=
ν
−ψ−r ψrµ : r > 0 .
Since the normal-ordering prescription for zero modes :ψ0µ ψ0ν : is ambiguous from the viewpoint
of creation and annihilation operators, one introduces an a priori undetermined offset h into the
definition of L0 .

(ii) Show that the canonical anticommutators of the ψ µ modes imply

m 
µ
[Lm , ψsµ ] = − + s ψm+s .
2
Hint: You may find the lemma [AB, C] = A{B, C} − {A, C}B useful.

(iii) Use the lemma in (ii) to verify the following commutator among energy-momentum modes
∞
Dm3 Dm X µ
[Lm , L−m ] = + + 2m r ψ−r ψrν ηµν .
24 12 r=1

Hint: Use the corollary ηµν ψ0µ ψ0ν = D

2 of the canonical anticommutators and be careful about non-
vanishing anticommutators in bringing bilinears of ψ µ modes into normal-ordered form (cf. exercise
B.3). You can also use the following sums seen in earlier exercises:
m m
X m(m+1) X m(m+1)(2m+1)
s= , s2 =
s=1
2 s=1
6

148
(iv) Show that the commutator in (iii) can only line up with the Virasoro algebra

c
[Lm , L−m ] = (m3 − m) + 2mL0
12
D D
at central charge c = 2 if the offset in the definition of L0 is chosen as h = 16 . Note that this
identifies the conformal weight of the spin fields which generate the Ramond-sector ground states.

G.2 NS vertex operators at the first mass level

In the NS-sector of the open-superstring spectrum, the first mass level compatible with GSO projection
1
occurs at m2 = −p2 = α0 . One can construct its vertex operators from an ansatz
 
(−1)
Vm2 =1/α0 = : Bµν i∂z X µ (z)ψ ν (z) + Eµνλ ψ µ ψ ν ψ λ (z) + Hµ ∂z ψ µ (z) e−φ(z) eip·X(z) :

involving polarization tensors Bµν , Eµνλ and Hµ . As explained in the lecture, the matter part (after peel-
ing off the superghost contribution :e−φ(z) :) must be the lower component of a superconformal primary,
i.e. one cannot have any double poles in its OPE with the supercurrent

1
G(z) = √ i∂z X µ (z)ψµ (z) .
2α0
(i) Show that the superconformal-primary condition constrains the polarizations to obey

η µν Bµν + pµ Hµ = 0
1
(Bµν −Bνµ ) + 3pλ Eλµν = 0
2
2α0 pν Bµν + Hµ = 0 .

(ii) One can show (for instance using BRST methods which have not been covered in the lecture) that
the following vertex operators in D = 10 dimensions
 
(−1)
Vsp A = : (ηµν + 4α0 pµ pν )i∂z X µ (z)ψ ν (z) + 6α0 pµ ∂z ψ µ (z) e−φ(z) eip·X(z) :
 
(−1)
Vsp B = : Σ[µν] i∂z X µ (z)ψ ν (z) + α0 Σ[µν pλ] ψ µ ψ ν ψ λ (z) e−φ(z) eip·X(z) :
 
(−1)
Vsp C = : (pµ πν + pν πµ )i∂z X µ (z)ψ ν (z) + 2πµ ∂z ψ µ (z) e−φ(z) eip·X(z) :

with pµ Σµν = pµ πµ = 0 describe spurious states (similar to the gluon at µ → pµ ) that decouple
from amplitudes. Verify that they are compatible with the superconformal-primary conditions.

(iii) By their decoupling from ampliudes, the spurious states in (ii) can be identified with zero. Use this
to show that the vertex operators describing physical states can be brought into the form

(−1)
Vspin 2 = Bµν : i∂z X µ (z)ψ ν (z)e−φ(z) eip·X(z) :
(−1)
V3-form = Eµνλ : ψ µ ψ ν ψ λ (z)e−φ(z) eip·X(z) : ,

where the polarizations satisfy

Bµν = Bνµ , η µν Bµν = pµ Bµν = 0 , pλ Eλµν = 0 .

(iv) Explain why the constraints on Bµν & Eµνλ leave 44 & 84 physical degrees of freedom, respectively.

149
 (v) Show that the representatives of the spin-2 and the 3-form vertex operators in the zero ghost picture
are proportional to
 
(0)
Vspin 2 ∼ Bµν : i∂z X µ i∂z X ν (z) + 2α0 (p · ψ)i∂z X µ ψ ν (z) + 2α0 (∂z ψ µ )ψ ν (z) eip·X(z) :
 
(0)
V3-form ∼ Eµνλ : 3i∂z X µ ψ ν ψ λ (z) + 2α0 (p · ψ)ψ µ ψ ν ψ λ (z) eip·X(z) : .

G.3 Supersymmetry transformations

This exercise aims to familiarize you with fermion vertex operators and supersymmetry transformations
in ten dimensions (using the shorthand γ µν = 21 (γ µ γ ν − γ ν γ µ ) for antisymmetrized products below).

(i) Demonstrate that the supercharge (to be contracted with anticommuting Weyl spinors η α of SO(1, 9))
I
dz
Q(−1/2)
α ∼ Sα (z):e−φ(z)/2 :
2πi
(−1/2)
transforms massless fermionic vertex operators Vχ (z) ∼ χα Sα (z):e−φ(z)/2 eip·X(z) : to bosonic
vertex operators with polarization µ = (ηγ µ Cχ).

(ii) Use the identities (γµ C)αβ γδµω̇ +(γµ C)βδ γαµω̇ +(γµ C)δα γβµω̇ = 0 and (γµ C)αβ = (γµ C)βα to show that

δη1 , δη2 χα = 2(η1 γµ Cη2 )pµ χα ,

 

where the supersymmetry transformations of gluinos and gluons are encoded through the shorthands

δη µ = (ηγ µ Cχ) , δη χβ = η α (γ µν )α β µ pν .

(iii) Use the identity (γ λρ γ µ C)αβ + (γ λρ γ µ C)βα = 2η µρ (γ λ C)αβ − 2η µλ (γ ρ C)αβ to show that

δη1 , δη2 µ = 2pλ (η1 γλ Cη2 )µ − 2λ (η1 γλ Cη2 )pµ ,
 

where the second term amounts to a linearized gauge transformation and does not affect amplitudes.

To be submitted on Feb 12th

150
String Theory II Maor Ben-Shahar, Oliver Schlotterer

H Problem set February 9th 2023

H.1 Spacetime supersymmetry of the superstring spectrum

The purpose of this exercise is to verify the matching of bosonic and fermionic d.o.f. in the superstring
spectrum including massive modes, a necessary condition for spacetime supersymmetry. This will be
done via generating functions of the form
X
Z(q) = ck q k
k

with a formal bookkeeping variable q ∈ C, where the expansion coefficients ck ∈ N0 track the number of
states at mass level k. We shall furthermore rely on the superstring analogue of lightcone quantization
where only transverse oscillators αni , ψri with vector indices i = 1, 2, . . . , 8 of SO(8) are taken into ac-
count. Similar to the discussion for the bosonic string, restriction to transverse oscillators automatically
incorporates all the physical-state constraints from the OPE with the supercurrent. Moreover, all the
spurious states (such as linearized gauge transformations µ → kµ of the gluon polarization vector) are
automatically subtracted.
Just like in the manifestly SO(1, 9) covariant approach of the lectures, the modes of ψri have r ∈ Z − 12
in the NS sector and r ∈ Z in the R sector. However, the Ramond-sector ground states |âiR are now
labelled by a Weyl-spinor index â = 1, 2, . . . , 8 of SO(8) since the Clifford algebra {ψ0i , ψ0j } = δ ij of the
zero modes is now adapted to a smaller range i, j = 1, 2, . . . , 8. As before, GSO projection only admits
i
an odd number of ψr∈Z− 1 oscillators acting on the NS ground state |0iNS while the R-sector states may
2
i
involve any number of ψr∈Z acting on a left-handed Weyl-spinor ground state |âiR .
i
At the massless level, for instance, the 8 physical gluon d.o.f. are encoded in ψ−1/2 |0iNS with
i = 1, 2, . . . , 8 while the 8 physical gluino d.o.f. stem from |âiR with â = 1, 2, . . . , 8 (the transversal-
ity constraint and the massless Dirac equation are automatically incorporated). Here and below, the
spacetime momenta pµ characterizing the ground states limz→0 :eip·X(z) :|0iNS and limz→0 :eip·X(z) :|âiR
are left implicit.

(i) Show that the counting of transverse oscillators at the first mass level of the open superstring (see
exercise G.2) leads to 128 states in both the NS and the R sector.

(ii) Consider a fixed spacetime direction i of the bosonic and fermionic oscillators. Explain why the
k th -order coefficients in the following q-expansions count the number of states αm
i
1
i
. . . αm a
ψri 1 . . . ψri b
Pa Pb
with overall oscillator number − j=1 mj − j=1 rj = k
∞
i
Y 1
• only bosonic oscillators αn<0 =⇒ Zbos (q) =
m=1
1 − qm
∞
Y
i
• only fermionic NS-sector oscillators ψr<0 =⇒ ZNS (q) = (1 + q m−1/2 )
m=1
∞
Y
i
• only fermionic R-sector oscillators ψr<0 =⇒ ZR (q) = (1 + q m )
m=1

151
 i i
You can assume the oscillators to act on a reference state |0i which is annihilated by all αn>0 , ψr>0
and ignore the zero modes in the R sector for the moment.

(iii) For the complete R sector of the superstring with eight values of i, eight ground-state components
|âiR and the possibility to combine αni and ψni oscillators, explain that the generating function for
the number of states at different mass levels is
∞ 8
1 + qm
Y
super
ZR (q) =8
m=1
1 − qm

and determine the number of spacetime fermions at the 2nd mass level of the open-superstring
spectrum.

For your information: The same logic leads to the generating function
∞ 8
1 + q m−1/2
Y
6GSO −1/2
ZNS (q) = q
m=1
1 − qm

for the complete NS sector of the superstring before GSO projection (with the prefactor q −1/2 to
account for the fact that the state without any αni or ψni would be a tachyon at mass level − 12 ).

(iv) In order to implement GSO projection at the level of the NS-sector generating function, it is con-
venient to introduce the building block (again for ψri with a fixed value of i = 1, 2, . . . , 8)
∞
Y
Z NS (q) = (1 − q m−1/2 )
m=1

whose minus sign tracks the number of ψri -oscillators in a NS-sector state (with (−1)` for ` fermionic
oscillators). Explain why the coefficients of q n in
Y ∞ 8  
super −1/2 1 1 8 8
ZNS (q) = q ZNS (q) − Z NS (q)
m=1
1 − qm 2

generate the number of GSO projected NS-sector states at level n of the open superstring.

(v) Verify up to and including the second mass level that the numbers of spacetime bosons and spacetime
fermions in the open-superstring spectrum match (which is a necessary condition for spacetime
supersymmetry).
super super
(vi) Express ZNS (q) and ZR (q) in terms of the Dedekind eta function η(q) as well as the Jacobi
theta functions θj (q), j = 2, 3, 4 defined by
∞
Y
η(q) = q 1/24 (1 − q m )
m=1
Y∞
θ2 (q) = 2q 1/8 (1 − q m )(1 + q m )2
m=1
∞
Y
θ3 (q) = (1 − q m )(1 + q m−1/2 )2
m=1
Y∞
θ4 (q) = (1 − q m )(1 − q m−1/2 )2
m=1

and explain why spacetime supersymmetry of the entire string spectrum relies on Jacobi’s abstruse
identity (from the year 1829)

4
4
4
θ3 (q) − θ2 (q) − θ4 (q) = 0 .

152
(vii) OPTIONAL (no need to submit a solution for this part): Prove Jacobi’s abstruse identity.

H.2 Four-fermion amplitude

The goal of this exercise is to derive the following expression for the color-ordered four-gaugino amplitude

Γ(1 + s)Γ(1 + t)
Atree tree
super (χ1 , χ2 , χ3 , χ4 ) = ASYM (χ1 , χ2 , χ3 , χ4 )
Γ(1 + s + t)
with wave functions χα
j=1,2,3,4 and the following field-theory amplitude in ten-dimensional SYM

(χ1 γ µ Cχ2 )(χ3 γµ Cχ4 ) (χ4 γ µ Cχ1 )(χ2 γµ Cχ3 )

Atree
SYM (χ1 , χ2 , χ3 , χ4 ) = + .
(p1 +p2 )2 (p2 +p3 )2

(i) Revisit the three-point functions of (ψ µ , Sα ) and the superghost system to derive the OPE

(γ µ C)αβ
:Sα e−φ/2 :(z) :Sβ e−φ/2 :(w) ∼ √ :ψµ e−φ :(w) + . . .
2(z−w)

and the three-point function

(γµ C)γδ
h:ψµ e−φ :(z2 ) :Sγ e−φ/2 :(z3 ) :Sδ e−φ/2 :(z4 )i = √ .
2z23 z24 z34

(ii) Show that the following expression for a four-point spin-field correlator

h:Sα e−φ/2 :(z1 ) :Sβ e−φ/2 :(z2 ) :Sγ e−φ/2 :(z3 ) :Sδ e−φ/2 :(z4 )i
(γµ C)αβ (γ µ C)γδ (γµ C)αδ (γ µ C)γβ
 
1
= − (H.1)
2(z13 z24 ) z12 z34 z14 z23

is consistent with the OPE from (i) in the limits z1 → z2,3,4 . Note that (γµ C)αβ = (γµ C)βα and
(γµ C)αβ (γ µ C)γδ +cyc(α, β, γ) = 0 by representation theory of SO(1, 9). Moreover, the OPE relevant
to z1 → z3 needs to be weighted with a minus sign from swapping the positions of :Sβ e−φ/2 :(z2 )
and :Sγ e−φ/2 :(z3 ).

For your information: Once you have checked consistency of (H.1) with the limits z1 → z2,3,4 , you
have in fact succeeded to prove it for any zj ∈ C: You have shown that the difference between the
left-hand side and right-hand side of (H.1) does not have any poles in z1 . Since this difference is
holomorphic ∀ z1 ∈ C and bounded by the absence of poles (it cannot blow up as z1 → ∞), it
must be constant in z1 by Liouville’s theorem. Finally, the constant difference between the left- and
right-hand side of (H.1) must be zero since the falloff of the desired correlator of h = 1 conformal
primaries is ∼ z1−2 as z1 → ∞.

(iii) Use the prescription for open-superstring tree amplitudes given in the lecture to show that
1
(χ1 γ µ Cχ2 )(χ3 γµ Cχ4 ) (χ4 γ µ Cχ1 )(χ2 γµ Cχ3 )
Z  
Atree
super (χ1 , χ2 , χ3 , χ4 ) ∼ s t
dz2 |z2 | |1 − z2 | + .
0 z2 1 − z2
β β α
Note that χα
i χj = −χj χi are anticommuting variables.

(iv) Evaluate the disk integrals in terms of Gamma functions to arrive at the expression for the four-point
function stated at the beginning of the exercise.

153
H.3 Massive three-point amplitude
This exercise is dedicated to the three-point tree amplitude
(q )
Atree 1 (q2 ) (q3 )
super (B, 2 , 3 ) = |z12 z13 z23 | · hVspin 2 (z1 )V2 (z2 )V3 (z3 )i (H.2)

involving two gluons with transverse polarizations 2 , 3 and one massive spin-two state from the first
mass level of the open superstring with p21 = − α10 : The spin-two state is characterized by a transverse
1 (q )
& symmetric-traceless polarization tensor Bµν , and its vertex operators Vspin 2 (z1 ) in the ghost pictures
q1 = 0, −1 can be found in exercise G.2.

(i) Show that the open-string Koba–Nielsen factor for the above state configuration is given by
* 3 +  
Y
ipj ·X(zj )
 z12 z13 
:e : =
 .
j=1
z 23


(ii) Explain why the color-dressed amplitude is proportional to Tr(T a1 T a2 T a3 + T a1 T a3 T a2 ).

(iii) Evaluate the color-ordered three-point amplitude (H.2) in the ghost pictures (q1 , q2 , q3 ) = (−1, −1, 0).

(iv) Manifest linearized gauge invariance by expressing the results in terms of the linearized field
strengths fjµν = pµj νj − pνj µj of the gluons.

(v) Derive the three-point correlator

z23 B λρ
Bµν h:(∂z1 ψ µ )ψ ν :(z1 )ψ λ (z2 )ψ ρ (z3 )i = 2 z2
z12 13

(vi) Use (v) to reproduce the result of (iv) from the ghost pictures (q1 , q2 , q3 ) = (0, −1, −1) in (H.2).

H.4 Dimensional reduction and type-IIA supergravity

OPTIONAL (no need to submit a solution for this exercise, the required material has not been covered
in the lecture but can be looked up in the lecture notes)

As you have seen in the lecture, the massless spectrum of the type-IIA superstring in D = 10 comprises

• NS-NS sector: graviton hµν , B-field Bµν , dilaton Φ

• R-NS sector: left-handed gravitino χα
µ , right-handed dilatino Λβ̇

• NS-R sector: right-handed gravitino χ̄µβ̇ , left-handed dilatino Λ̄α

(1) (3)
• R-R sector: one-form potential Aµ and three-form potential Aµνλ

(i) Dimensionally reduce the type-IIA multiplet to D = 4 and verify that the helicities follow the
counting of N = 8 supergravity.

(ii) Show that the type-IIA multiplet in ten dimensions follows from dimensional reduction of the
(3)
D = 11 supergravity multiplet which comprises a graviton hM N , a three-form potential AM N P and
a gravitino χB
M (with vector indices M, N, P = 0, 1, . . . , 10). The D = 11 gravitino is a 32-component

Dirac spinor with index B = 1, 2, . . . , 32 (odd dimensions do not admit a notion of chirality). You
can use that Dirac spinors in D = 11 decompose into one left-handed and one right-handed Weyl
spinor in D = 10 dimensions.

To be submitted on Feb 28th

154