Introduction to String Theory

Lecture Notes

ETH Zürich, HS13

Prof. N. Beisert, Dr. J. Brödel


c 2014 Niklas Beisert, Johannes Brödel, ETH Zurich
This document as well as its parts is protected by copyright.
Reproduction of any part in any form without prior written
consent of the author is permissible only for private,
scientific and non-commercial use.
Contents
0 Overview 6
0.1 Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
0.2 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1 Introduction 1.1
1.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1
1.3 Some Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5

2 Relativistic Point Particle 2.1

2.1 Non-Relativistic Actions . . . . . . . . . . . . . . . . . . . . . . . . 2.1
2.2 Worldline Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2
2.3 Polynomial Action . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3
2.4 Various Gauges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4
2.5 Quantisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4
2.6 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5
2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6

3 Classical Bosonic String 3.1

3.1 Nambu–Goto Action . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1
3.2 Polyakov Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2
3.3 Conformal Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3
3.4 Solution on the Light Cone . . . . . . . . . . . . . . . . . . . . . . . 3.4
3.5 Closed String Modes . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4
3.6 Hamiltonian Formalism . . . . . . . . . . . . . . . . . . . . . . . . . 3.7

4 String Quantisation 4.1

4.1 Canonical Quantisation . . . . . . . . . . . . . . . . . . . . . . . . . 4.1
4.2 Light Cone Quantisation . . . . . . . . . . . . . . . . . . . . . . . . 4.2
4.3 String Spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3
4.4 Anomalies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.6
4.5 Covariant Quantisation . . . . . . . . . . . . . . . . . . . . . . . . . 4.7

5 Compactification and T-Duality 5.1

5.1 Kaluza–Klein Modes . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1
5.2 Winding Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2
5.3 T-Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3
5.4 General Compactifications . . . . . . . . . . . . . . . . . . . . . . . 5.5

6 Open Strings and D-Branes 6.1

6.1 Neumann Boundary Conditions . . . . . . . . . . . . . . . . . . . . 6.1
6.2 Solutions and Spectrum . . . . . . . . . . . . . . . . . . . . . . . . 6.1
6.3 Dirichlet Boundary Conditions . . . . . . . . . . . . . . . . . . . . . 6.3
6.4 Multiple Branes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5

7 Conformal Field Theory 7.1

3
 7.1 Conformal Transformations . . . . . . . . . . . . . . . . . . . . . . 7.1
7.2 Conformal Correlators . . . . . . . . . . . . . . . . . . . . . . . . . 7.3
7.3 Local Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5
7.4 Operator Product Expansion . . . . . . . . . . . . . . . . . . . . . . 7.7
7.5 Stress-Energy Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . 7.9

8 String Scattering 8.1

8.1 Vertex Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1
8.2 Veneziano Amplitude . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3
8.3 String Loops . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4

9 General Relativity Basics 9.1

9.1 Differential Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . 9.1
9.2 Riemannian Geometry . . . . . . . . . . . . . . . . . . . . . . . . . 9.5
9.3 General Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.7

10 String Backgrounds 10.1

10.1 Graviton Vertex Operator . . . . . . . . . . . . . . . . . . . . . . . 10.1
10.2 Curved Backgrounds . . . . . . . . . . . . . . . . . . . . . . . . . . 10.2
10.3 Form Field and Dilaton . . . . . . . . . . . . . . . . . . . . . . . . . 10.6
10.4 Open Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.8
10.5 Two-Form Field of a String . . . . . . . . . . . . . . . . . . . . . . 10.10

11 Superstrings 11.1
11.1 Supersymmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1
11.2 Green–Schwarz Superstring . . . . . . . . . . . . . . . . . . . . . . 11.5
11.3 Ramond–Neveu–Schwarz Superstring . . . . . . . . . . . . . . . . . 11.7
11.4 Branes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.10
11.5 Other Superstrings . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.11

12 Effective Field Theory 12.1

12.1 Effective Action and Compactifications . . . . . . . . . . . . . . . . 12.1
12.2 Open Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3
12.3 Closed Strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.4
12.4 Relations Between String Amplitudes . . . . . . . . . . . . . . . . . 12.5

13 String Dualities 13.1

13.1 T-Duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.1
13.2 Strong/Weak Coupling Duality: S-Duality . . . . . . . . . . . . . . 13.4
13.3 Strong Coupling Limits . . . . . . . . . . . . . . . . . . . . . . . . . 13.5
13.4 M-Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.6

14 String Theory and the Standard Model 14.1

14.1 The Real World . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14.1
14.2 Geometry of Toroidal Manifolds and Orbifolds . . . . . . . . . . . . 14.2
14.3 Calabi–Yau Compactification of D = 10 Supergravity . . . . . . . . 14.4
14.4 String Theory as a Phenomenological Model . . . . . . . . . . . . . 14.5

4
15 AdS/CFT Correspondence 15.1
15.1 Stack of D3-Branes . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.1
15.2 Anti-de Sitter Geometry . . . . . . . . . . . . . . . . . . . . . . . . 15.2
15.3 N = 4 Super Yang–Mills . . . . . . . . . . . . . . . . . . . . . . . . 15.3
15.4 Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15.3

5
Introduction to String Theory Chapter 0
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

0 Overview
String theory is an attempt to quantise gravity and unite it with the other
fundamental forces of nature. It combines many interesting topics of (quantum)
field theory in two and higher dimensions. This course gives an introduction to the
basics of string theory.

0.1 Contents
1. Introduction (1 lecture)
2. Relativistic Point Particle (2 lectures)
3. Classical Bosonic String (3 lectures)
4. String Quantisation (4 lectures)
5. Compactification and T-Duality (2 lectures)
6. Open Strings and D-Branes (2 lectures)
7. Conformal Field Theory (4 lectures)
8. String Scattering (2 lectures)
9. General Relativity Basics (2 lectures)
10. String Backgrounds (3 lectures)
11. Superstrings and Supersymmetry (4 lectures)
12. Effective Field Theory (3 lectures)
13. String Dualities (3 lectures)
14. String Theory and the Standard Model (2 lectures)
15. AdS/CFT Correspondence (2 lectures)
Indicated are the approximate number of 45-minute lectures. Altogether, the
course consists of 39 lectures.

0.2 References
There are many text books and lecture notes on string theory. Here is a selection
of well-known ones:
• classic: M. Green, J.H. Schwarz and E. Witten, “Superstring Theory” (2
volumes), Cambridge University Press (1988)
• alternative: D. Lüst, S. Theisen, “Lectures on String Theory”, Springer (1989).
• new edition: R. Blumenhagen, D. Lüst, S. Theisen, “Basic Concepts of String
Theory”, Springer (2012).
• standard: J. Polchinski, “String Theory” (2 volumes), Cambridge University
Press (1998)
• basic: B. Zwiebach, “A First Course in String Theory”, Cambridge University
Press (2004/2009)

6
• recent: K. Becker, M. Becker, J.H. Schwarz, “String Theory and M-Theory: A
Modern Introduction”, Cambridge University Press (2007)
• online: D. Tong, “String Theory”, lecture notes,
http://arxiv.org/abs/0908.0333
• ...

7
Introduction to String Theory Chapter 1
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

1 Introduction

1.1 Definition
String theory describes the mechanics of one-dimensional extended objects in an
ambient space.
(1.1)
Some features:
• Strings have tension:
(1.2)
• Strings have no inner structure:

but not (1.3)

• Several pieces of string can interact:

←→ (1.4)
• Strings can be classical or quantum:
vs. (1.5)
• Strings can be open or closed:

vs. (1.6)

1.2 Motivation
Why study strings?

Extended Objects. We know a lot about the mechanics of point particles. It is

natural to study strings next. Or even higher-dimensional extended objects like
membranes. . .

(1.7)

particle string membrane

1.1
These objects are snapshots at fixed time t. Introduce the worldvolume as the
volume of spacetime occupied by the object:

(1.8)

worldline worldsheet worldvolume

The worldsheet of a string is two-dimensional. In fact, there is a great similarity

between strings and static soap films.

Quantum Gravity. String theory offers a solution to the problem of quantum

gravity (QG). Let us try to sketch the problem of quantum gravity with as little
reference to quantum field theory (QFT) as possible.
There are two established classical1 gravity theories:
• Newtonian Gravity (non-relativistic)
• General Relativity (GR, relativistic, geometry of spacetime)
We know that nature is quantum mechanical, therefore gravity must also be
quantum for consistency with the other fundamental forces. In practice, the effects
of QG hardly play a role except for considerations of the early universe and for
black hole radiation.
Field quantisation introduces quanta (particles):
• electromagnetism: photon
• strong nuclear forces: gluons
• gravity: graviton
• matter fields: electrons, quarks, neutrinos, . . .
These particles interact through vertices (Feynman rules) which can be composed
to more complex interaction processes (Feynman graphs). The Standard Model
(SM) of particle physics has relatively simple set of rules (qualitatively)

S= +g + g2 . (1.9)

Here  represents a coupling constant.

1
Here and in the following the term “classical” will refer to the absence of quantum effects.
Classical theories can be either non-relativistic or relativistic.

1.2
Conversely, Einstein gravity has infinitely many vertices which are governed by
Newton’s constant G

√
S= + G +G

+ G3/2 + G2 + .... (1.10)

In fact, we can introduce additional couplings ck :

√
G + (G + c4 ) + (G3/2 + c5 ) + .... (1.11)

This is perfectly consistent with the assumptions of GR, except that the additional
terms introduce higher-derivative corrections to the Einstein equations. Classically
we do not need the ck , but in QFT we do.2 The point is that loops in Feynman
graphs generate divergences, e.g.:

= ∞. (1.12)

In QFT we have to sum up all competing processes, e.g.:3

(G + c4 ) + G2 + G3 + .... (1.13)

In this sum, we can absorb the divergence into a redefinition of the (new) coupling
constant c4 = −G3 ∞ + c4,ren . This process is called renormalisation.
All is well now, the divergences are gone, but there is no good way to set the
renormalised c4,ren to zero (or any other distinguished value). Unfortunately,
2
A general principle of QFT is that we need to include all permissible interaction terms which
are not excluded by some principle, typically symmetries or a power counting scheme.
3
A curious fact is that quantum gravity does not produce a divergence in the one-loop graph
2
(G term).

1.3
cancellation of all divergences requires infinitely many ck ’s.4 The quantisation of
Einstein gravity introduces infinitely many adjustable parameters. All parameters
have to be known (measured) in order to have a predictive description of nature.
This renders the theory non-predictive! The only good prediction is at sufficiently
low energies much below the Planck scale: There the theory is approximated well
by GR with only G as the coupling constant.
What does string theory have to do with it?
Quantum string theory turns out to contain particles which gravitons in many
ways. Moreover, string theory does not generate divergences; it is a finite theory!
Finally, string theory has just two fundamental coupling constants.
Is all well now!? Almost, many more couplings may be hiding in the description of
the vacuum state which is relevant when actual physics is to be addressed.

Unification. String theory provides a unified description for all kinds of

fundamental forces of nature.
Electromagnetic and weak forces combine into electroweak forces at sufficiently
high energies (around 102 GeV). These may also combine with the strong nuclear
forces (quantum chromodynamics, QCD) into a so-called Grand Unified Theory
(GUT)? There are some hints:
• Charges of fermions appear to suggest larger symmetry group than the one of
QCD and the electroweak theory:

SU(3) × SU(2) × U(1) ←− SU(5), SO(10)? (1.14)

• Estimated GUT scale 1015 GeV is somewhere close to Planck scale 1019 GeV.
This may suggest a unification of all forces.
• Wouldn’t it be nice?
String theory describes gauge theories as well as gravity. In particular, sequence of
groups SU(5), SO(10), . . . appears.
Does string theory describe nature? So far no convincing derivation. Best option:
String theory describes the Standard Model among many (!) other stringy
“natures”.

String/Gauge Duality. There are intricate relations between string and gauge
theories.5
• Some effects within gauge theory such as gluon flux tubes in QCD have a
stringy nature.
• Some particular gauge theories are in fact equivalent to string theories.
String theory can be viewed as an aspect of gauge theory.
4
In the standard model there are only finitely many coupling constants which require
renormalisation. The standard model is said to be renormalisable while general relativity is
non-renormalisable.
5
Gauge theories serve as the description of forces in the standard model.

1.4
Treasure Chest. String theory yields many interesting, novel, exceptional
structures, results, insights in physics and mathematics. Just to name a few:
supersymmetry, higher dimensions, p-branes, dualities, topological insights.

Many Unsolved Problems. Despite 40 years of research there are many

unsolved questions:
• How to match with nature?
• How to find direct/indirect evidence?6
• What is String Theory?
• How to quantise gravity (otherwise)?

1.3 Some Conventions

Special Relativity.
• Minkowski spacetime R3,1 .
• 4-vectors xµ = (t, ~x); time t = x0 .
• indices µ, ν, . . . = 0, 1, 2, 3.
summation convention: xµ yµ := 3µ=0 xµ yµ .
P
•
• metric to lower and raise indices: xµ := ηµν xν .
• metric signature: ηµν = diag(−+++), i.e. p2 = −m2 !
• products x · y := ηµν xµ y ν ; squares x2 := x · x.
• Poincaré symmetry: rotations, Lorentz boosts, spatial and temporal translations.
• generalises to D spacetime dimensions: RD−1,1 .

Constants of Nature.
• For simplicity we will set c = ~ = e = 1: All units are then expressed in powers
of kg ∼ m−1 . One can always reconstruct the dependence on c, ~, e through
dimensional analysis of the expected units.
• Newton’s constant G.

6
Supersymmetry would be a useful indication if it had been found, but it need to be visible at
low energies.

1.5
Introduction to String Theory Chapter 2
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

2 Relativistic Point Particle

Let us start slowly with something else: a relativistic particle. Here we will
encounter several issues of string theory, but in a more familiar setting. There are
many equivalent formulations, we will discuss several.

2.1 Non-Relativistic Actions

We even take another step back and consider a free non-relativistic point particle
~x(t).
t

~x(t) (2.1)

~x
The well-known Lagrange function L and action S read
Z
˙ 1 ˙ 2
S[~x] = dt L ~x(t), ~x˙ (t), t .


L(~x, ~x, t) = 2 m~x , (2.2)

The resulting equations of motion (e.o.m.) are just ~x¨(t) = 0. The momentum and
energy follow from the Hamiltonian formulation
∂L p~ 2
p~ = = m~x˙ , E=H= . (2.3)
∂ ~x˙ 2m
Now promote the above to a relativistic particle
q
L = −mc c2 − ~x˙ 2 = −mc2 + 12 m~x˙ 2 + 18 mc−2~x˙ 4 + . . . . (2.4)
The expression in the square root is a proper relativistic combination. Its
expansion has an irrelevant constant term, the well-known non-relativistic term
and relativistic correction terms. Now derive the e.o.m.
(c2 − ~x˙ 2 )~x¨ + (~x˙ · ~x¨)~x˙ = 0. (2.5)
The vector nature of the equation implies collinearity of the first and second
derivative, ~x¨ = α~x˙ . Substitute to obtain the equation αc2~x˙ = 0 whose least
restrictive solution is α = 0 which implies ~x¨ = 0 as in the non-relativistic case.
Momentum and energy read
mc~x˙ p
p~ = p , E = c m2 c2 + p~ 2 . (2.6)
c2 − ~x˙ 2
This is fine, but not manifestly relativistic: non-relativistic formulation of a
relativistic particle. We want a manifestly relativistic formulation using 4-vectors
X µ = (ct, ~x) and Pµ = (E/c, p~). Let us set c = 1 for convenience from now on.

2.1
• The momentum Pµ is already a good 4-vector:
P 2 = −E 2 + p~ 2 = −m2 . (2.7)
The mass shell condition P 2 = −m2 is manifestly relativistic, but p~ and E have
a rather distinct role/origin in the Hamiltonian framework.
• The position X m (t) = (t, ~x(t)) and the action S[~x] make explicit reference to
time t (which is defined in a particular Lorentz frame)
s  2
Z
dX(t)
S = − dt m − . (2.8)
dt
• Note that the Hamiltonian framework makes reference to a slicing of time, hence
it distinguishes between space and time.

2.2 Worldline Action

The above action measures Lorentz-invariant proper time s of the particle’s path
X µ (t) in spacetime (worldline)
Z
S = −m ds, where ds2 = −dX 2 . (2.9)

The proper time depends on the location of the worldline only, but not on a
particular Lorentz frame (definition of t) or parametrisation of the worldline
(through t).

t t0 τ (2.10)

Let us assume an arbitrary parametrisation X µ (τ ) of the worldline through some

curve parameter τ . The proper time action reads (now a dot denotes d/dτ )
s  2
Z Z
dX(τ ) p
S = − dτ m − = − dτ m −Ẋ 2 (2.11)
dτ
and is a manifestly relativistic formulation. However, instead of having 3
undetermined functions ~x(t), there are now 4 undetermined functions X µ (τ ) and a
new function t(τ ).
The equations of motion for the new action read
Ẋ 2 Ẍ µ = (Ẋ · Ẍ)Ẋ µ (2.12)
and imply collinearity Ẍ µ = cẊ µ for all τ with variable c(τ ). Therefore the
resulting worldline is straight.

2.2
 p
As a next step, let us derive momenta as derivatives of L = −m −Ẋ 2 w.r.t. Ẋ µ :
mẊµ
Pµ = p . (2.13)
−Ẋ 2
The above expression immediately squares to −m2 ; thus the mass shell condition
P 2 = −m2 is obeyed.
There are two features special to the above description: while we have only three
independent Pµ , there are four independent X µ . Furthermore, the naive
Hamiltonian is strictly zero: H = 0. These properties are signals of constraints and
gauge invariance:
• Reparametrising τ 0 = f (τ ) has no effect on physics.
• Redundancy of description: worldline coordinate τ .
• One linear dependency among the e.o.m. for X µ .
• Gauge invariance effectively removes one X µ , e.g. time t(τ ).
• Situation inconvenient for Hamiltonian framework/QM.
• Usually it is better to fix a gauge. There are many choices, so one can pick a
convenient one.
In summary, the above action is a fully relativistic formulation, which suffers from
the complication of gauge invariance. However, gauge invariance is often
considered a virtue: Symmetry! Simultaneously, the above worldline action has
two further drawbacks:
• it is non-polynomial; thus inconvenient for quantisation.
• does not work for massless particles m = 0.

2.3 Polynomial Action

In order to address the disadvantages from the formulation in the last subsection,
let us consider an equivalent action with an auxiliary variable e(τ )
Z  
S = dτ 21 e−1 Ẋ 2 − 21 em2 . (2.14)

The resulting equations of motion read

m2 e2 + Ẋ 2 = 0, eẌ µ − ėẊ µ = 0, (2.15)
and yield the same old equation for X µ after combination. The momentum
conjugate to X µ reads Pµ = e−1 Ẋµ ; hence the equation of motion for e reduces to
P 2 = −m2 . The momentum conjugate to e vanishes, signalling a constraint.
While in the previous formulation the massless case did not work, m = 0 is a valid
choice at every step of the above derivation. It will lead to a constant momentum
P µ = e−1 Ẋ µ as well as P 2 = 0. In the massless case, the field e is not fixed by the
e.o.m.: there is a remaining gauge freedom.
The einbein e is a dynamical variable. Curiously, the e.o.m. picks out the metric
induced by the ambient space. Upon substituting the solution for e, one recovers
the above worldline action.

2.3
The field e has a nice geometrical interpretation: the einbein specifies a metric
gτ τ = −e2 on the worldline. All terms in the action are combined in a way as to
render the action invariant under a change of worldline coordinates (e transforms
according to e0 = e dτ 0 /dτ ).
In terms of the metric gτ τ , the above action reads:
√
Z  
S = − 2 dτ −gτ τ g τ τ Ẋ 2 + m2
1
(2.16)

where g τ τ = (gτ τ )−1 = −1/e2 .

2.4 Various Gauges

There is freedom to either fix one of the coordinates X µ (τ ) or the auxiliary field e
at will. Here are a couple of (more or less) useful choices:
• Temporal Gauge. t(τ ) = τ or t(τ ) = ατ .
Reduces to non-relativistic treatment of the beginning of the section.
• Spatial Gauge. z(τ ) = ατ .
Works locally except at turning points of z(τ ).
• Light Cone Gauge. x+ (τ ) := t(τ ) + z(τ ) = ατ .
Useful in some cases; prominent in string theory.
• Proper Time Gauge. ds = dτ .
Fixes t(τ ) through the integral
Z τ q
t(τ ) = dτ 0
1 + ~x˙ (τ 0 )2 . (2.17)
R
Action becomes trivial S = − dτ ; dynamics governed by constraint.
• Constant Einbein. ė = 0.
In the polynomial formulation, gauge fixing may involve e. A customary gauge
choice is a constant e. In that case, the e.o.m. reduces to

Ẍ = 0. (2.18)

If the dynamical variable e is gauge fixed to be constant, one must still

remember its equation of motion

Ẋ 2 + m2 e2 = 0, (2.19)

which turns into a constraint.

2.5 Quantisation
Quantisation can be done in several different ways, depending on the choice of
classical formulation. Let us pick the polynomial action discussed in Sec. 2.3. As it
is convenient to fix a gauge for the Hamiltonian formulation, we will choose the

2.4
einbein e to be constant. Momenta P associated to X and the resulting
Hamiltonian read:
P = e−1 Ẋ, H = 12 e P 2 + m2 .


(2.20)
Conventionally, a state |Ψ i is given by a wave function of position variables and
time Z
|Ψ i = d4 X Ψ (X, τ ) |Xi. (2.21)

Here, it will be slightly more convenient to immediately Fourier transform to

momentum space |Xi ' d4 P eiP ·X |P i, resulting in states and wave functions
R

Z Z
4
|Ψ i = d P Ψ (P, τ ) |P i, Ψ (P, τ ) = d4 X eiP ·X Ψ (X, τ ). (2.22)

respectively. The Schrödinger equation

iΨ̇ = HΨ = 2i e P 2 + m2 Ψ


(2.23)

is obviously solved by

Ψ (P, τ ) = exp − 2i e(P 2 + m2 )τ Φ(P ).

(2.24)

However, the system is constrained. Whenever the constraint P 2 + m2 = 0 is not

satisfied, the wave function must vanish:

(P 2 + m2 )Ψ (P, τ ) = 0. (2.25)

Therefore physical states Ψ (P, τ ) = Φ(P ) are independent of τ , which makes

perfect sense: the worldline coordinate τ can be reparametrised and is unphysical.
Correspondingly, the Schrödinger equation governing the evolution of τ evolution
is replaced by the constraint P 2 + m2 = 0 (which in turn governs the t-evolution).
Upon Fourier-transforming the wave function back to position space Φ(X), the
constraint becomes the Klein–Gordon equation for a spin-0 field

(−∂ 2 + m2 )Φ(X) = 0. (2.26)

2.6 Interactions
While the motion of a free particle is easy, one would like to eventually include
interactions. Let us sketch how to add interactions with external potentials and
with other particles:

Electrical and Gravitational Fields. Coupling the free relativistic particle to

electrical and gravitational fields takes a very geometric form
Z  
S = dτ 21 e−1 gµν (X)Ẋ µ Ẋ ν − 12 em2 + Aµ (X)Ẋ µ , (2.27)

2.5
where Aµ is potential for the electromagnetic field and Fµν = ∂µ Aν + ∂ν Aµ the
corresponding field strength. Likewise gµν is the gravitational potential, which
takes the form of the metric of a curved spacetime.
Here, Aµ and gµν are assumed to be fixed external fields, that is, they are
unaffected by the presence of the particle, but influence its motion. Note however,
that those fields are to be evaluated at the dynamical position X µ (τ ).
In quantum mechanics, one usually assumes weak interactions, which allows to
work with free quantum fields formally. Interactions are then introduced in a
perturbative fashion. Whenever a free particle enters a potential field, it scatters
off of it. Hereby the dominant contribution originates in single scattering; multiple
interactions are suppressed. Only in rare instances, potentials can be handled
exactly.

Interactions Among Particles. Local interactions occur, if several particles

meet at some spacetime point and split up, potentially into a different number of
particles. In the worldline formulation, this type of processes is taken care of by
introducing vertices where several particle worldlines meet:

(2.28)

However, this is not the standard treatment of particle interactions. Usually an

interaction of n fields Φ corresponds to a term Φn in QFT action. While our
method is not very convenient, it works as well. It mimics the Feynman rules and
is the standard procedure for string theory.

2.7 Conclusions
• We have seen many different formulations of the same physical system and had
to deal with gauge invariance and constraints. Depending on the description,
there were different numbers of degrees of freedom (d.o.f.), but the number of
solutions (modulo gauge) remained the same always.
• We reviewed the quantisation of the free relativistic particle.
• Interactions and couplings to external potentials have been discussed.
• The description here was chosen in light of the analogous treatment in string
theory later on. While working well, it was not always the most convenient one.

2.6
Introduction to String Theory Chapter 3
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

3 Classical Bosonic String

String theory describes a one-dimensional extended object without inner structure.
The main ingredients for its mathematical formulation are
• worldsheet coordinates: ξ α = (τ, σ), time τ , space σ,
• embedding coordinates: X µ (ξ),
• D-dimensional embedding: Minkowski space, metric ηµν .
X0 = t

τ
σ (3.1)
X2

X1

3.1 Nambu–Goto Action

The Nambu–Goto action is the string generalisation of the worldline action
described in the last chapter. In order to promote the point particle to a string,
the following correspondences are used:
• worldline −→ worldsheet.
• action = proper time ' “length” −→ “area”.

Area and Action. Let us calculate an infinitesimal area element in terms of the
embedding coordinates X µ . After a Wick rotation t = iw the area element dA of
2D euclidean surface reads:
dA = dτ dσ |X 0 | |Ẋ| | sin θ|
p
= dτ dσ X 02 Ẋ 2 sin2 θ
q Ẋ
= dτ dσ X 02 Ẋ 2 − (X 0 · Ẋ)2 θ
p X0
= d2 ξ det γ , (3.2)
where γαβ = ηµν ∂α X µ ∂β X ν is the induced worldsheet metric (the pull back of the
spacetime metric).
Employing another Wick rotation in order to return to a worldsheet with
Minkowski signature leads to
Z
1 1 2
p
S=− A = − d ξ − det γ , (3.3)
2πκ2 2πκ2
which is the Nambu–Goto action.

3.1
The action exhibits the following symmetries:
• Lorentz symmetry: because the action is built from scalar products of Lorentz
vectors X,
• Poincaré symmetry: as there is no explicit dependence on X, only through
derivatives ∂X, √
• worldsheet diffeomorphisms: since the density d2 ξ − det γ is invariant under
reparametrisations ξ 7→ ξ 0 (ξ).

Tension. What is the meaning of κ? The parameter κ is related to the

fundamental string length scale, which will play a role in the quantum string later
on.

L
(3.4)

Consider a time slice of the action (and thus a slice of the worldsheet). A slice of
length L leads to a potential U ∼ L/κ2 . The constant force resulting from the
potential is the string tension: T = U 0 = 1/2πκ2 .1

Equations of Motion. The equations of motion resulting from the

Nambu–Goto action can be obtained using the variation of the determinant
δ det γ = det γ γ αβ δγαβ p
∂α − det γ γ αβ ∂β X µ = 0.


(3.5)
As the metric γ contains the embedding X, those equations are highly non-linear
and thus difficult to deal with.
What do those equations imply geometrically (in euclidean signature)? The
stationary action demands a surface with minimal area, i.e. a static soap film.
Thus, the mean curvature is zero everywhere which implies that every point of the
surface is a saddle point.

+R
−R (3.6)

3.2 Polyakov Action

How to remove the complications from the non-linear equations of motion? In the
same way as for the point particle, there exists a polynomial action with an
additional dynamical worldsheet metric gαβ :
Z
1 2
p
S=− 2
d ξ − det g 12 g αβ (∂α X) · (∂β X). (3.7)
2πκ
1
Thus, a string is not a spring or a rubber band because they do not exert a constant force.

3.2
The e.o.m. for the field X is the same as above while the e.o.m. for the worldsheet
metric g reads:
(∂α X) · (∂β X) = 21 gαβ g γδ (∂γ X) · (∂δ X). (3.8)
The solution to this equation relates the dynamical worldsheet metric to the
induced metric from the Nambu–Goto action via
gαβ = f (ξ) (∂α X) · (∂β X) = f (ξ)γαβ , (3.9)
where f (ξ) is a local scale. This scale nicely cancels in the action as well as in all
equations of motion. The new redundancy introduced by describing the string
with the help of the dynamical worldsheet metric is called Weyl invariance:
gαβ (ξ) 7→ f (ξ)gαβ (ξ). (3.10)

3.3 Conformal Gauge

In the polynomial formulation, one can now make use of the gauge freedom in
order simplify the e.o.m. for X. The non-linearity in the coupling to g poses a
major obstruction to solve it, but we are lucky that we can remove it by
demanding a conformally flat metric:
gαβ (ξ) = f (ξ)ηαβ . (3.11)
This gauge choice amounts to two equations:
gτ σ (ξ) = 0 and gτ τ (ξ) = −gσσ (ξ). (3.12)

By fixing the metric to be conformally flat, we have been using all diffeomorphisms
on the worldsheet which preserve the metric up to scale f . The remaining
diffeomorphisms are the Weyl scalings; in fact, one can furthermore set f = 1 and
thus fix the Weyl redundancy as well. The resulting action describes D free
massless scalar particles on the worldsheet
Z
1
S=− d2 ξ 21 η αβ (∂α X) · (∂β X) (3.13)
2πκ2
and the corresponding equation of motion for X is simply the harmonic wave
equation:
∂ 2 X µ = 0 or Ẍ = X 00 . (3.14)

However, although having imposed conformal gauge, there is still an equation of

motion for the worldsheet metric which we need to take into account (in conformal
gauge):
Tαβ := (∂α X) · (∂β X) − 12 ηαβ η γδ (∂γ X) · (∂δ X) = 0, (3.15)
where Tαβ is the energy-momentum tensor for D scalar particles. The trace of the
energy-momentum tensor vanishes by construction due to the conformal/Weyl
symmetry (Tαα = 0), while the two remaining e.o.m. are called Virasoro constraints
Tτ σ = Ẋ · X 0 = 0, Tτ τ = Tσσ = Ẋ · Ẋ + X 0 · X 0 = 0. (3.16)

3.3
The first constraint demands the lines of constant τ to be orthogonal to lines of
constant σ. This means, the string can move perpendicular to the direction it is
stretched out. In other words, there are no longitudinal waves allowed, as already
noted above: the string does not have an inner structure!
Because the energy-momentum tensor Tαβ is conserved, it is sufficient to impose
constraints on an initial time slice only.

3.4 Solution on the Light Cone

The harmonic wave equation can be most easily solved in light cone coordinates
ξ L/R :
ξL τ ξR

ξ L/R = τ ∓ σ, ∂L/R = 12 (∂τ ∓ ∂σ ), (3.17)

σ

which imply the worldsheet metric

d2 s = −dτ 2 + dσ 2 = −dξ L dξ R . (3.18)

In terms of light cone coordinates, the e.o.m. and Virasoro constraints read

∂L ∂R X µ = 0, (∂L/R X)2 = 0. (3.19)

The first equation can be solved by a simple separation of variables

X µ (ξ L , ξ R ) = XLµ (ξ L ) + XRµ (ξ R ). (3.20)

While there are D left-movers XL and D right-movers XR initially, the Virasoro

constraints (∂XR,L )2 = 0 remove one left- and one right-mover. Two
reparametrisations remain:
• conformal transformations:

ξ R 7→ ξ 0R (ξ R ), ξ L 7→ ξ 0L (ξ L ). (3.21)

In two dimensions (and only there), there are infinitely many conformal
transformations. Those remove another left- and right-mover.
• constant shifts: between XLµ and XRµ .
In summary, there are (D − 2) left- and right-movers remaining, which parametrise
the transverse directions of the string.

3.5 Closed String Modes

In the discussion so far the worldsheets have been extended infinitely in space and
time. In order to have a string of finite length, let us confine the spatial extent.
There are two possible basic topologies for a string:

3.4
• closed string: circular topology, identify σ ≡ σ + 2π (other choices possible),
• open string: interval topology, boundary conditions at σ = 0, π (later).

(3.22)

Covariant Formulation. For the closed string, the function function X should
be 2π-periodic, so let us get started with a Fourier decomposition:
µ
X iκ
XL/R = 21 xµ + 12 κ2 pµ ξ L/R + √ αnL/R,µ exp(−inξ L/R ). (3.23)
n6=0
2n
√
In the above ansatz, the coefficients iκ/ 2n are chosen for later convenience. The
linear dependency on ξ L/R does not clash with the periodicity condition because
after adding the left- and right-mover, the dependency on σ drops out:
X µ = xµ + κ2 pµ τ + . . .. Reality of the embedding X is ensured by demanding
α−n = (αn )∗ .
The solution exhibits two kinds of parameters. While the motion of the centre of
mass is described by the conjugate pair x, p (conjugation involves an additional
L/R,µ
factor of κ2 ), the string modes αn (left/right movers) describe the amplitudes
of the oscillations on the string.
(x, p)

αn (3.24)

µ
Plugging the above ansatz into the Virasoro constraints (∂L/R XL/R )2 = 0 yields
X
κ2 LnL/R exp(−inξ L/R ) = 0, (3.25)
n

where we have defined the Virasoro modes (dropping the L/R index)
X
Ln := 12 αn−m · αm (3.26)
m
√
and α0L = α0R = κp/ 2. Demanding validity of the Virasoro constraints is
equivalent to requiring the Ln = 0 for all n. In particular, the Virasoro constraint
L0 = 0 fixes the mass of the string:
∞
2 2 24 X
p = −M , M = 2 α−m · αm . (3.27)
κ m=1

Since all Virasoro constraints Ln = 0 are conserved by the e.o.m.

L̇n = inLn , (3.28)

3.5
it is sufficient to impose them on initial data only. The mass of a string depends
on the mode amplitudes α. If there are no modes excited, the string behaves like a
massless point particle. If there are only few and small excitations, one will
encounter a light (or tiny) particle while large excitations can add up to yield a big
and highly massive object.

(3.29)

Due to the Minkowski signature of the spacetime, the time-like modes α0

contribute a negative M 2 . If one, however, takes all Virasoro constraints into
account, tachyons (particles with negative mass-squared) are excluded.

Light Cone Gauge. The Virasoro constraints in their above form are
complicated and non-linear. A convenient solution to this problem is to employ
the conformal symmetry on the worldsheet in order to solve them. The connection
between the conformal symmetry on the worldsheet and the ability to choose the
particular gauge in light cone coordinates below is not spelled out here explicitly.
The main idea is to first switch to light cone coordinates in spacetime

X ± = X 0 ± X D−1 , (3.30)

where X ~ denotes the transverse components 1 . . . (D − 2). Now one selects a

gauge, in which the coordinate X + depends on τ linearly, i.e. does not exhibit
oscillator modes.
X0 = t
X− X+ X+ X+
−→ (3.31)
~
X 0
τ τ
X D−1

Correspondingly, the gauge condition reads

+
XL/R = x+ 1 2 +
L/R + 2 κ pL/R ξ
L/R
. (3.32)

In light cone coordinates, the Virasoro constraint (∂ X~ L/R )2 − κ2 p+ ∂X − = 0 is

L/R L/R
solved by Z ξ
− 1 ~ L/R (ξ 0 ) 2 ,
dξ 0 ∂ X


XL/R (ξ) = 2 + (3.33)
κ pL/R
~ L/R (ξ L/R ). These are the transverse
leaving us with 2(D − 2) arbitrary functions X
modes of the string.

3.6
Periodicity. Of course all functions X still have to be periodic. While the
~
constraint from periodicity of X(ξ) is exploited in the next paragraph, periodicity
+ −
of X and X requires
Z 2π
+ + + ~ R )2 − (∂ X
~ L )2 = 0.


pL = pR = p , dξ (∂ X (3.34)
0

In addition, there is the residual gauge freedom ∆XRµ (ξ R ) = −∆XLµ (ξ L ) = const.,

which corresponds to the residual constraints mentioned above.

String Modes. Imposing gauge fixing on the modes αn (n 6= 0) leads to

1 X
αn+ = 0, αn− = ~ n−m · α
α ~ m, (3.35)
α0+ m

that is, the modes α− are again determined in terms of α+ and the modes α ~.
~ 0R , α0R,+ = α0L,+ , α0R,+ = α0L,+ as well as
~ 0L = α
Periodicity requires α
∞
X ∞
X
L L R R
α
~ −m · α
~m = α
~ −m ·α
~m . (3.36)
m=1 m=1

Taking the above conditions into account, the resulting mass is manifestly positive:
∞ ∞
2 4 X 4 X
M = 2 ~ −m · α
α ~m = 2 αm |2 ,
|~ (3.37)
κ m=1 κ m=1

where the reality condition α−n = (αn )∗ has been used in the last equation.
In summary, the light cone gauge comes with manifestly positive mass for all
particles and is a very convenient way to get rid of almost all constraints.
However, by introducing light cone coordinates, we give up a manifestly
Lorentz-invariant formulation.

3.6 Hamiltonian Formalism

Before quantising the string in the next chapter, let us have a brief look at the
classical string in the Hamiltonian formalism. We would like to derive the Poisson
brackets for the variables x, p, αn .

Fourier Modes. Let us start with the Polyakov action in conformal gauge
Z Z 2π
1
dσ 21 Ẋ 2 − X 02


S = dτ L, L= 2
(3.38)
2πκ 0

and substitute the closed-string mode expansion (with free time dependence)
X
Xµ = κ βnµ (τ ) exp(−inσ). (3.39)
n

3.7
Now calculate the derivatives and perform the integration over σ in order to obtain
a tower of independent harmonic oscillators (here β0 is the free particle):
1X  
L= β̇n · β̇−n − n2 βn · β−n . (3.40)
2 n

The Poisson bracket between canonical momentum and β reads:

µ
πn = β̇n , {βm , πnν } = η µν δm+n . (3.41)

Matching X with the previous classical solution at τ = 0

π0µ nβ µ πµ
xµ = κβ0µ , pµ = , αnL/R,µ = √∓n + √∓n (3.42)
κ i 2 2
leads to the following non-trivial Poisson brackets in the original variables

{xµ , pν } = η µν , L,µ
{αm , αnL,ν } = {αm
R,µ
, αnR,ν } = −im η µν δm+n . (3.43)

Field Theory. The same result could have been obtained borrowing some
mathematical tools which are prominently used in quantum field theory. The
conjugate momentum Π µ and canonical Poisson brackets read
1
Πµ = Ẋ µ , {X µ (σ), Π ν (σ 0 )} = η µν δ(σ − σ 0 ). (3.44)
2πκ2

Now, given those basic relations, one can use and derive the following relations:
Z 2π Z 2π
1 µ µ
dσX (0, σ) = x , dσΠ µ (0, σ) = pµ (3.45)
2π 0 0

as well as
Z 2π
2πiκ R,µ
dσX µ (0, σ) exp(−inσ) = √ (αnL,µ − α−n ),
0 2n
Z 2π
1 R,µ
dσΠ µ (0, σ) exp(−inσ) = √ (αnL,µ + α−n ), (3.46)
0 2κ
where we
R have used the fundamental Fourier integral
−1 0
(2π) dσ exp(iσ(n − n )) = 2πδn−n0 . Using linearity of the Poisson bracket, it is
not too difficult to obtain the above Poisson brackets of modes.

3.8
Introduction to String Theory Chapter 4
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

4 String Quantisation
We have seen that the classical closed string is described by
L/R
• a bunch of harmonic oscillators αn for the string modes;
• a relativistic particle (x, p) describing the centre of mass.
Both systems are coupled via Virasoro constraints.
We have derived two reasonable formulations:
• Covariant formulation
– D oscillators αnµ per mode,
– physical solutions must obey Virasoro constraints LR L
n = Ln = 0,
– Poincaré symmetry in spacetime manifest,
– worldsheet theory has conformal symmetry (later).
• Light cone formulation
– D − 2 oscillators α~ n per mode,
– physical solutions must obey LR L
0 = L0 = 0 residual constraints,
– Poincaré symmetry of spacetime partially manifest,
– manifest Poincaré symmetry on worldsheet.

4.1 Canonical Quantisation

Commutation Relations. In the Hamiltonian formulation of string theory in
conformal gauge we have derived the following set of non-trivial Poisson brackets
between the variables x, p, αn

{xµ , pν } = η µν ,
 L,µ L,ν  R,µ R,ν
αm , αn = αm , αn = −im η µν δm+n . (4.1)

In canonical quantisation this results in the following set of commutation relations

of the corresponding quantum operators1

[xµ , pν ] = iη µν ,
 L,µ L,ν   R,µ R,ν 
αm , αn = αm , αn = m η µν δm+n . (4.2)

Space of States. Compose the space of states from a quantum free particle and
a set of quantum harmonic oscillators:
• momentum eigenstates for free particle |qi,
• HO vacuum |0i and excitations for each mode/orientation.
1
We shall use the same symbols for classical variables and corresponding quantum operators.
The precise meaning should be clear from the context.

4.1
Define string vacuum state |0; qi
pµ |0; qi = q µ |0; qi, αnL/R,µ |0; qi = 0 for n > 0. (4.3)

Negative Norm States. One problem: we have states with negative norm
L/R,µ
|n, µ, L/R; qi := α−n |0; qi,
2 L/R,µ
|n, µ, L/R; q| = h0; q|αnL/R,µ α−n |0; qi = nη µµ . (4.4)
For µ = 0 this state has negative norm. In fact, it is not allowed by Virasoro
constraints.
General resolution of this problem: impose Virasoro constraints! All states obeying
the Virasoro constraints have non-negative norm. This problem needs some care.
We shall continue the covariant quantisation later.

4.2 Light Cone Quantisation

For simplicity we fix the light cone gauge; this leaves only physical states.
Resulting commutators evidently lead to positive-definite states
 L,a L,b   R,a R,b 
αm , αn = αm , αn = m δ ab δm+n . (4.5)

Recall classical mass and residual constraint LR L

0 = L0 = 0
∞ ∞
24 X L L 4 X R
M = 2 α
~ ·α
~ = α
~ ~R.
·α (4.6)
κ m=1 −m m κ2 m=1 −m m

In the quantum theory, the ordering of operators matters! A priori we are free to
choose the operator ordering.2 We therefore assume normal ordering (negative
mode numbers to the left of positive mode numbers) plus two new constants aL/R :
∞
! ∞
!
4 X 4 X
M2 = 2 ~L · α
α ~ L − aL = 2 ~R · α
α ~ R − aR . (4.7)
κ m=1 −m m κ m=1 −m m

The term involving the oscillators α measures the so-called string level (which is a
non-negative integer in the quantum theory)
∞ ∞
X X 1
N := ~ −m · α
α ~m = mNm with Nm := ~ −m · α
α ~ m. (4.8)
m=1 m=1
m

The mass and residual constraint can be expressed in terms of string level
4 4
M2 = 2
(N L − aL ) = 2 (N R − aR ). (4.9)
κ κ
2
Quantisation of a classical model is not necessarily a unique procedure. Typically it requires
the introduction of additional parameters which parametrise the arbitrariness (or our ignorance).
These parameters are of order ~ when taking the classical limit, and therefore have no classical
counterpart.

4.2
4.3 String Spectrum
The mass of the string state depends on string level. Quantisation of string level
leads to a quantisation of mass for string states. Level matching:

N L − aL = N R − aR . (4.10)

Let us understand string states at each level: HO’s.

Vacuum State. Define vacuum state |0; qi

~ nL/R |0; qi = 0 for n > 0.

α (4.11)

This state is at level zero: N L = N R = 0. It has spin zero.3

For a physical state we must require:
4a
aR = aL = a, M2 = − . (4.12)
κ2
√
So far so good: We have a spin-0 particle with M = 2κ−1 −a. For a physical
particle we would have to require a ≤ 0! Spatial extent of state: HO wave function
∼ κ.

First Level. Lowest excited state has N = 1. Level matching and aL = aR

implies N L = N R = 1. One excitation α
~ −1 each from left/right movers
L,a R,b
|ab; qi = α−1 α−1 |0; qi. (4.13)
√
We find (D − 2)2 states of mass M = 2κ−1 1 − a.
What is their spin? Consider first the spin under SO(D − 2) transverse rotations.
There are three useful combinations:
2δab
|(ab); qi := |ab; qi + |ba; qi − |cc; qi,
D−2
|[ab]; qi := |ab; qi − |ba; qi,
|1; qi := |cc; qi. (4.14)

Transformation properties under SO(D − 2):

state indices Young tab. “spin”

|(ab); qi symmetric, traceless 2
(4.15)
|[ab]; qi anti-symmetric 1
|1; qi singlet • 0
3
This can be derived by applying the Noether charge of spacetime Poincaré symmetry.

4.3
Compare this to classification of unitary representations of the Poincaré group.4
Stabiliser (little group) for massive particle is SO(D − 1). Can we fit these
SO(D − 2) representation into SO(D − 1) representations? No!
The only way out: consider a massless particle where the stabiliser is SO(D − 2).
We have to set a = aR = aL = 1.
Three types of particles:
• |(ab); qi: massless spin-2 field. this is fine as free field.
Weinberg–Witten theorem: interactions are forbidden except for gravitational
interactions. This particle must be the graviton!
• |[ab]; qi: massless 2-form field (Kalb–Ramond).
generalisation of electromagnetic field Aµ : Bµν with 1-form gauge symmetry
δBµν = ∂µ ν − ∂ν µ .
• |1; qi: massless scalar particle (dilaton).
state different from string vacuum |0; qi.
What we have learned so far:
• String theory contains graviton plus massless 2-form and scalar particles.
• Interacting string theory includes gravity!
• Spatial extent of particles ∼ κ; practically point-like.
• κ is the Planck scale (later).
• We must set a = aR = aL = 1 for consistency (with physicality).

Tachyon. Now revisit the string vacuum |q, 0i: M 2 = −4/κ2 < 0. This state is a
tachyon!
Is this a problem? Not really, compare to spontaneous symmetry breaking
mechanism:
• Vacuum state was chosen at unstable local maximum of a potential. This leads
to a tachyonic mode.
• Physical ground state should be situated at local minimum. No tachyon here!
• (Bosonic) string theory: Unclear if a global minimum exists. Unclear how to
treat it in practice: Where is it? What are its properties?

(4.16)

• Let us ignore this shortcoming. Indeed, the tachyon is absent for superstrings
(which are treated later)!
4
This is a topic of QFT I: In short, massive particles are characterised by their spin
(representation of SO(D − 1) stabiliser group of massive particle trajectory) while massless
particles are characterised by their helicity (representation of SO(D − 2) stabiliser group of
massless particle trajectory). Use the well-known case D = 4 for comparison.

4.4
Higher Levels. Levels zero and one work out. What about the higher levels?
We have already used up all available freedom to adjust aL,R , now self-consistency
of the model is up to luck (or faith). The following table lists the representations
for left-movers (equivalently right-movers)
level excitations SO(D − 2) SO(D − 1)
0 · • •
a
1 α−1 ×
a b
2 α−1 α−1 +•
a
α−2
a b c
3 α−1 α−1 α−1 +
a b
α−1 α−2 + +•
a (4.17)
α−3
a b c d
4 α−1 α−1 α−1 α−1 ++•
a b c
α−1 α−1 α−2 + + +
a b
α−2 α−2 +•
a b
α−3 α−1 + +• •
a
α−4
... ... ... ...
All higher levels combine into proper SO(D − 1) representations, for both left and
right moving modes.
Furthermore, level matching implies we need to square the above representations
for the correct particle spectrum.
Altogether:
• String describes collection of infinitely many particle types of different mass and
spin.
• Various vibration modes might correspond to elementary particles. They include
the massless graviton.
• Intrinsic particle extent at Planck scale κ. Planck scale is much smalled than
can be observed: String theory describes practically point-like particles!
• String theory describes only a few massless particles; all others at Planck mass
1/κ; one tachyon.
• Proper treatment of tachyon could change picture altogether.
• Very high excitations are long strings. They would mostly display classical
behaviour, but are superheavy M  1/κ.

Regge Trajectories. Maximum spin (all indices symmetric and traceless)

increases linearly with level S = 2N
massive
S string
state

2S − 4a graviton
M2 = , 2-form (4.18)
κ2 2a
α0
tachyon dilaton M2

4.5
This relationship is called the leading “Regge trajectory”:
• α0 = κ2 is called the Regge slope.
• 2a is called the Regge intercept; spin of massless particle.
Subleading trajectories have lower spins (indices anti-symmetric or have trace).
Qualitative similarity to hadron spectrum:
• Regge trajectories observed for hadronic resonances.
• For a stringy description of QCD, 1/κ would have to be QCD scale ' 1 GeV.
• Intercept should be a ≈ − 12 for QCD rather than a = 1.
• There is another problem (see later).
• Strings provide a qualitative description of QCD flux tubes.

4.4 Anomalies
In light cone gauge we have broken manifest SO(D − 1, 1) Lorentz symmetry to a
SO(D − 2) subgroup.
• Consequently the spectrum of quantum strings organises manifestly into
SO(D − 2) multiplets.
• Almost all multiplets fit into SO(D − 1) multiplets.
• Mass assignments fill Poincaré multiplets for aL = aR = 1.
• Poincaré symmetry broken unless aL = aR = 1.
Anomaly: Failure of classical symmetry in quantum theory.
Sometimes anomalies are permissible, but not here because we want strings to
propagate on a Minkowski background with intact Poincaré symmetry.
So far we have only done counting, a more severe problem exists in the algebra.
The commutator [M −a , M −b ] is supposed to vanish, but it receives contributions
from [α− , αa ] which is non-zero in light cone gauge due to the solution of α− in
terms of an integral. One finds
∞    
−a −b
X D−2 1 a−1
[M , M ] = −1 n− + × ... (4.19)
n=1
24 n n

This expression vanishes if and only if D = 26 and a = 1. String theory predicts

twenty-six spacetime dimensions.
There is a shortcut derivation: reconsider the nature of the intercept a as a sum of
HO ground state energies 21 ωn = 12 n
∞
X ∞
X
1 1
a=− (D − 2) 2 ωn = − 2 (D − 2) n. (4.20)
n=1 n=1

This sum is divergent, but black magic helps: ζ-function regularisation

∞
X 1 D−2
ζ(z) := z
, i.e. a = − 12 (D − 2)ζ(−1) = . (4.21)
k=1
k 24

4.6
 1
Analytical continuation ζ(−1) = − 12 and use of a = 1 predicts D = 26! Here a
somewhat questionable derivation yields the correct prediction.

4.5 Covariant Quantisation

In LC gauge Poincaré symmetry is subject to an anomaly, but we can also keep
Poincaré symmetry manifest: Covariant quantisation. Let us see how the spectrum
arises in covariant approach. For simplicity, consider only one set of L or R
oscillators.

Vacuum State. The vacuum state |0; qi is defined as before by

α0µ |0; qi ∼ κq µ |0; qi, αn>0
µ
|0; qi = 0. It satisfies

κ2 q 2
Ln>0 |0; qi = 0 and L0 |0; qi = |0; qi. (4.22)
4
The state is not annihilated by the negative Virasoro modes. Instead
h0; q|Ln<0 = 0 hence h0; q|(Ln − δn a)|0; qi = 0.
For generic physical states |Ψ i, hΦ| we should impose the Virasoro constraints5

Ln>0 |Ψ i = 0, L0 |Ψ i = a|Ψ i,
hΦ|Ln<0 = 0, hΦ|L0 = a|Φi, (4.23)

such that we have

hΦ|(Ln − δn a)|Ψ i = 0. (4.24)

One Excitation. We make a generic ansatz for the one-excitation state

|ψ; qi := ψ · α−1 |0; qi. (4.25)

The norm ψ̄ · ψ is potentially negative. The level-one Virasoro constraint acting on

the state implies

κ(ψ · q)
L1 |ψ; qi = α1 · α0 |ψ; qi = √ |0; qi = 0. (4.26)
2

Furthermore L0 = a = 1 implies q 2 = 0. In this case, the constraint q · ψ = 0

removes the negative norm state. The following remains:
• D − 2 states with positive norm.
• One state with ψ ∼ q. This state has vanishing norm, it is “null”. Null states do
(should) not contribute to physics.
5
This is analogous to the Gupta–Bleuler quantisation of the electromagnetic field where the
gauge fixing condition is imposed on scalar products of states.

4.7
Two Excitations. Now start with the generic ansatz
µ ν µ
|φ, ψ; qi := φµν α−1 α−1 |0; qi + ψµ α−2 |0; qi. (4.27)

Then impose the constraints L0 , L1 , L2 = 0 to fix q 2 , ψ, tr φ. What remains is: ,

, • of SO(D − 1).6
• The ansatz for uses a generic symmetric traceless tensor φµν with φµν q ν = 0.
This state is positive definite.
• Ansatz for : φµν = qµ ρν + ρµ qν with q · ρ = 0. Negative norm for 1 < a < 2;
null for a = 1!
• Ansatz for •: φµν = qµ qν + ηµν σ. Negative norm for D < 1 or D > 26; null for
D = 26!

Virasoro Algebra. The algebra of quantum charges Ln reads

c
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n . (4.28)
12
This defines the Virasoro algebra. The latter term is the central charge c of
Virasoro algebras (rather its representation on fields), namely c = D in our case.
We are interested in primary states

Ln>0 |Ψ i = 0, L0 |Ψ i = a|Ψ i. (4.29)

We can now apply representation theory of the Virasoro algebra ⇒ 2D CFT.

A proper treatment (BRST) requires additional ghost fields. Inclusion of ghost
fields recovers the classical conformal algebra precisely when D = 26 and a = 1

[Lm , Ln ] = (m − n)Lm+n . (4.30)

In covariant gauge, the conformal algebra is anomalous. In light cone gauge,

however, the anomaly is shifted to the Lorentz algebra.7 The anomaly is
proportional to D − 26 and a − 1, i.e. it can be avoided for a very specific choice of
parameters.

6
Orthogonality w.r.t. the timelike vector q effectively reduces vector indices from SO(D − 1, 1)
to SO(D − 1).
7
The anomaly actually affects the product of the Virasoro and the Poincaré algebra, but
neither of the individual algebras alone.

4.8
Introduction to String Theory Chapter 5
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

5 Compactification and T-Duality

We have seen that the closed string spectrum contains:
• one tachyonic scalar particle (due to wrong vacuum),
• a graviton and few other massless particles,
• a tower of particles of increasing mass (practically inaccessible due to Planck
scale masses).
But: string theory requires D = 26 dimensions for consistency; that is way too
many to match with reality! In particular, the Gauss law predicts that the
gravitational force F decays with a high power of the distance r namely
F ∼ 1/A ∼ 1/r24 instead of the observed F ∼ 1/r2 behaviour.

5.1 Kaluza–Klein Modes

Idea: Compactify 22 spatial dimensions to a microscopic size. Large distances exist
only along the 4 remaining spacetime dimensions. For a sufficiently small
compactification radius, the compact dimensions are almost unobservable.

X 25

X 25 (5.1)
Xµ
µ
X

Compactify the coordinate X 25 to a circle of radius R

X 25 ≡ X 25 + 2πR. (5.2)

The quantum mechanical momentum now becomes quantised

n
P25 = . (5.3)
R
2 2
Effectively, we obtain a tower of massive particles M25 = M26 + n2 /R2 :1
• The zero mode n = 0 has the original mass. This mode can be observable if the
original mass is sufficiently small or massless.
• Higher modes are massive, M ' 1/R. For sufficiently small radius R these
modes are practically unobservable.
For 22 compact dimensions with a sufficiently small compactification radius, the
low-energy physics can be effectively four-dimensional.
1
M25 denotes the effective mass of a particle propagating in the 25 non-compact dimensions
whereas M26 denotes the original mass of the particle propagating in all 26 dimensions.

5.1
5.2 Winding Modes
Strings on compact spaces have a new peculiarity: winding modes.

(5.4)

Consider again one compact direction X 25 =: X ≡ X + 2πR. We need to adapt the

periodicity condition for the closed string to the needs of the compact dimension
X(σ + 2π) = X(σ) + 2πRm. (5.5)
Here, m is called the winding number. This periodicity condition implies the
following expansion in terms of string modes
 
1 1 2 n mR L/R
XL/R = 2 x + 2 κ ∓ 2 ξ + modes. (5.6)
R κ

Mass Spectrum. The resulting effective mass (for propagation in 25

non-compact dimensions) reads
 2
2 4 L/R n mR
M = 2 (N − a) + ∓ 2 . (5.7)
κ R κ
The level matching condition is modified by winding2
N L − N R = nm. (5.8)
We can use it to rewrite the mass formula as an average over the L/R level
numbers
2 n2 m2 R 2
M 2 = 2 (N L + N R − 2a) + 2 + . (5.9)
κ R κ4
This clearly shows that winding contributes (positively) to the mass much like the
KK modes.
In order to hide all but finitely many string modes: all of κ, R and κ2 /R must be
small.
M2
string KK winding

(5.10)

4/κ2 R2 /κ4
1/R2

Therefore a complete compactification R → 0 does not lead to the desired effect,

the compactification radius should be of the order of κ.
2
Also modes with N L 6= N R can now exist. This leads to new types of spin representations for
the closed string.

5.2
Compactification and Decompactification. We can now investigate what
happens when we “decompactify” the circle as R → ∞:
• Winding modes become very heavy.
• KK modes form become light and form a continuum.
M2

(5.11)

The additional winding modes become infinitely heavy as R → ∞. This makes

sense because a string with winding becomes very long and the tension is
responsible for a large mass. Furthermore, the low-energy spectrum reproduces the
string spectrum without compactification.
Alternatively, we can investigate the compactification limit R → 0:
• KK modes become very heavy.
• Winding modes become light and form continuum.
We obtain the same picture as for R → ∞ with role of m and n interchanged.
Moreover, we observe that the spectrum is the precisely the same for R and κ2 /R.
We conclude that the additional dimension remains observable at R → 0! This is
very different from a regular point particle which has KK modes only.

5.3 T-Duality
We have seen a duality between a small and a large compactification radius. We
can in fact show this feature at Lagrangian level: T-duality.
We start with the string action of the 25-direction X := X 25 in conformal gauge
Z
1
− 2
d2 ξ 12 η ab ∂a X ∂b X. (5.12)
2πκ
The action has a global shift symmetry X → X + . The winding modes could be
viewed as a shift by  = 2πmR which is localised to the boundary. Let us therefore
make the symmetry local by introducing a new gauge field Aa
Z
1 2

1 ab ab

d ξ − 2 η (∂a X + Aa )(∂b X + Ab ) − ε X̃∂a Ab . (5.13)
2πκ2
This action is now invariant under a local shift symmetry X → X +  together
with Aa → Aa − ∂a . We have added two d.o.f. in the field Aa and one local
redundancy in the shift by . We remove the remaining additional d.o.f. by
demanding Fab = ∂a Ab − ∂b Aa = 0 through a Lagrange multiplier X̃. This implies
that locally the gauge field Aa is trivial, i.e. it is gauge equivalent to Aa = 0. The

5.3
new action is therefore equivalent to the old one, we have neither gained nor lost
anything, but it enables us to perform the following step.
The field Aa has no derivatives in the action, it is algebraic, and we can integrate
it out exactly. The e.o.m. read

Aa = −∂a X − ηac εcb ∂b X̃. (5.14)

We substitute them into the action and obtain a new action (up to a boundary
term) Z
1
− 2
d2 ξ 21 η ab ∂a X̃ ∂b X̃. (5.15)
2πκ
This action has the same form as before, but it is expressed in terms of X̃ instead
of X.
Now we can set Aa = 0 to go back to the original formulation, and obtain the
duality relation

∂a X = −ηac εcb ∂b X̃, i.e. Ẋ = X̃ 0 , ˙

X 0 = X̃. (5.16)

This transformation is called T-duality.3 Since the action in the dual coordinates is
the same as before, solutions are mapped to solutions (potentially modulo
boundary conditions). For the standard solution X we find the dual solution X̃
n
X = x + κ2 τ + mRσ + modes,
R
n
X̃ = x̃ + mRτ + κ2 σ + modes. (5.17)
R
The duality interchanges the radii R ↔ R̃ = κ2 /R as well as the winding and KK
numbers m ↔ n.

R
R̃
←→ (5.18)

Therefore T-duality relates string theory on two backgrounds with different

compactification radii.
This implies that R = κ is effectively the minimum compactification radius. It is
indeed a special “self-dual” point. For this special point, the duality between two
models turns into an enhanced symmetry of a single model.
R = κ is the minimum length scale in string theory:4 This may hint at a
quantisation of spacetime in quantum gravity.
3
T-duality is a non-local transformation: Although the map between ∂a X and ∂b X̃ is local,
the fields X and X̃ are related non-locally.
4
Recall that the HO wave function of the string vacuum has a spatial extent of ∼ κ.

5.4
5.4 General Compactifications
So far we have compactified one dimension: either a circle or an interval. There
are many choices and parameters for a simultaneous compactification of several
dimensions, e.g.:
• sphere S n ,
• product of spheres S a × S n−a with different radii,
• torus T n , 3n − 3 moduli (radii, tilts),
• other compact manifolds.
The low-lying modes are determined by this manifold:5
• Compactification determines observable spectrum of low-energy particles.
• Goal: find a suitable manifold to describe the standard model of particle physics
as a low-energy limit.
• Massless modes correspond to gauge symmetries.
• Superstrings: compactification typically breaks supersymmetry. In order to
preserve one supersymmetry, the compactification manifold must be a
Calabi–Yau 3-fold.6

5
An analogy is the shape of a bell which determines its characteristic spectrum.
6
This is the reason why Calabi–Yau manifolds are of interest to string theorists.

5.5
Introduction to String Theory Chapter 6
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

6 Open Strings and D-Branes

So far we have discussed closed strings. The alternative choice is open boundary
conditions which leads to some interesting new objects in string theory.

6.1 Neumann Boundary Conditions

For open strings we conventionally choose the range of the spatial coordinate σ as
0 ≤ σ ≤ π, and we should discuss the boundaries at σ = 0, π. We start again in
conformal gauge Z
1
S= 2
d2 ξ 12 η ab ∂a X · ∂b X. (6.1)
2πκ
We vary the action paying attention to boundary terms due to partial integrations
Z
1
δS = d2 ξ η ab ∂a δX · ∂b X
2πκ2
Z
1
= d2 ξ ∂a (η ab δX · ∂b X) − . . .
2πκ2
Z
1 0 0


= dτ δX(π) · X (π) − δX(0) · X (0) − .... (6.2)
2πκ2
The boundary e.o.m. imply Neumann conditions where the function has no slope
at the boundary1

X 0 (0) = X 0 (π) = 0. (6.3)

0 π

The Virasoro constraints X 0 · Ẋ = X 02 + Ẋ 2 = 0 imply that the end points move at

speed of light2
Ẋ 2 = 0. (6.4)

6.2 Solutions and Spectrum

The string has the same e.o.m. in the bulk, so we can recycle the bulk solution.
1
There is in fact an alternative implication which we shall discuss later.
2
In the euclidean version of the string, a static soap film cannot have a free boundary: Ẋ 2 = 0
implies Ẋ = 0 which is in contraction with a boundary.

6.1
Doubling Trick. We can even relate the open string to a closed one. We map
two copies of the open string to a closed string which is twice as long, and we
identify the points
0 π
2π
σ ≡ 2π − σ. l (6.5)
0 π

We then impose the boundary condition X 0 = 0 at σ = 0, π on the closed string

solution derived earlier. This implies3

R L

αnL = αnR =: αn . L R (6.6)

R L

0 π 2π

Left movers are reflected into right movers at boundaries. Therefore, we are left
with only one copy of string oscillators and one copy of the Virasoro algebra. The
general solution for the open string reads4
X iκ
X µ = xµ + 2κ2 pµ τ + √ αnµ exp(−inξ L ) + exp(−inξ R ) .


(6.7)
n6=0
2n

Quantisation. Quantisation is analogous to closed strings. We find the very

same anomaly conditions D = 26, a = 1 arising from the bulk. The resulting mass
spectrum reads5
1
M 2 = 2 (N − a). (6.8)
κ
We now have only a single copy of oscillators at each level. The discussion is the
same as for the closed string, but without the squaring of representations. We find:
• level 0: singlet tachyon (of half “mass”).
• level 1: massless vector: Maxwell field.
• level 2: massive spin-2 field .
• ...
The massless modes at level one are associated to local symmetries:
• open string: spin-1 gauge fields,
• closed string: spin-2 gravitational fields and others.
3
In fact both boundaries yield the same set of constraints. The doubling trick is an ansatz
which identifies the two boundary conditions; different boundary conditions would require a
modification of the doubling trick.
4
Note that the prefactor of the momentum p is twice as large compared to the closed string
solution. This choice compensates for the halving of the integration domain in σ so that p is the
Noether charge for target space translations in both cases.
5
Note the different prefactor related to the alternative definition of p.

6.2
String Interactions. Let us briefly discuss interacting strings. In general, open
and closed strings interact:
• Interactions of open strings certainly involves that two ends of string can join.

←→ (6.9)

When the two ends belong to a single open string, this string closes.

←→ (6.10)

Therefore an interacting theory of open strings must include closed strings. This
is achieved by several string “vacuum” states in the Hilbert space in same
theory, e.g. |0; qic and |0; qio .
• The opening of strings can be suppressed. Closed strings can live on their own
with interactions splitting or joining stings.

←→ (6.11)

We conclude that string theory necessarily contains gravitational d.o.f.; It may or

may not include gauge field(s) from open string sectors.

6.3 Dirichlet Boundary Conditions

The Neumann boundary conditions discussed above are not sufficient, Dirichlet
boundaries should also be present in string theory as following discussion suggests.
Now consider compactification for open strings. Again, this is almost the same as
for closed strings. However, we have no winding modes because the open string
can always unwind.

vs. (6.12)

T-Duality. We apply T-duality to open strings and introduce the dual field X̃ 25
by the relation
X 025 = X̃˙ 25 , Ẋ 25 = X̃ 025 . (6.13)
The boundary conditions translate to

X 0µ = 0 (µ = 0, . . . , 24), X̃˙ 25 = 0. (6.14)

We obtain Dirichlet boundary condition for the dual coordinate X̃ 25 while the
Neumann boundary conditions remains for the remaining original coordinates X µ .
The Dirichlet condition implies that the ends of the string are fixed to constant

6.3
values in the coordinate X̃ 25 . This corresponds to the alternate choice of boundary
e.o.m. where the variation is suppressed, δ X̃ 25 = 0.

(6.15)
0 π

We can show that the dual string actually starts and ends at the same X̃ 25
coordinate
2πκ2 n
Z Z
025
∆X̃ = dσ X̃ = dσ Ẋ 25 = 2πκ2 p25 =
25
= 2πnR̃. (6.16)
R
Here we have used that the momentum p25 = n/R is quantised in KK modes. The
end points are then identified by the compactification of the dual coordinate
X̃ 25 ≡ X̃ 25 + 2π R̃.

vs. (6.17)

Note that the Dirichlet boundary condition constrains the string end points which
allows winding modes but prevents momentum flow along X̃ 25 and thus KK
modes. In conclusion, T-duality exchanges the roles of KK and winding modes just
as for closed strings.
The Dirichlet condition modifies the oscillator relation
α̃nL,25 = −α̃nR,25 =: α̃n25 (6.18)
such that the general solution reads
X iκ
X̃ 25 = x̃25 √ α̃n25 − exp(−inξ L ) + exp(−inξ R ) .


0 + (6.19)
n6=0
2n

Although the Dirichlet condition X 25 = const . may appear unnatural (why X 25 ?

what value?), we find that it has to be part of (T-dual) string theory (on compact
spaces).

D-Branes. Now we take the Dirichlet condition seriously. At each string

boundary we can choose
• Neumann condition X 0µ = 0 or
• Dirichlet condition X µ = fixed.
We can make this choice for each direction µ individually.
Geometrical picture: String ends are confined to so-called “Dp-branes”.

ne
ra
D p -b
(6.20)
1)
(p,
D−p−1

6.4
• Dp-branes are (p + 1)-dimensional submanifold of spacetime.
• The submanifold includes the time direction, it has signature (p, 1).
• Dirichlet conditions for the D − p − 1 orthogonal directions of the submanifold.
• Neumann conditions along the remaining p + 1 directions along the submanifold.
• D-branes can be curved submanifolds.6
• Open strings with pure Neumann conditions can be viewed as a spacetime-filling
D(D − 1)-brane.
• T-duality maps between Dp-branes and D(p ± 1)-branes.
Strings propagate on backgrounds with D-branes:
• The bulk spacetime curvature governs the propagation of the string bulk,
• D-branes govern the propagation of string ends.
There is even more to D-branes as non-perturbative objects: We will continue this
discussion later.

6.4 Multiple Branes

We can have multiple branes of diverse types. Open strings then stretch between
two branes while closed strings are detached from the branes.

Parallel Branes. The simplest case is two parallel planar Dp-branes located at
X 25 = 0, d in a non-compact Minkowski space.

(6.21)

d
The general solution can be sketched as
σd
X µ = 2κ2 pµ τ + modes, X 25 =
+ modes. (6.22)
π
The resulting (quantum) mass spectrum from the point of view of p + 1 dimensions
d2 1
M2 = 2 4
+ 2 (N − a). (6.23)
4π κ κ
At the lowest levels we find:
√
• A scalar particle at level 0 with mass M = d2 − 4π 2 κ2 /2πκ2 .
• A vector particle at level 1 with mass M = d/2πκ2 .
• The scalar particle becomes tachyonic for d < 2πκ. Nearby D-branes are
unstable.7
• The vector particle becomes massless when the branes coincide.8
6
This case is hardly ever treated in practice because it is just as hard as curved spacetimes.
Moreover, T-duality would not apply due to absence of a shift symmetry.
7
This instability can be attributed to the minimum length scale of string theory.
8
Two D-branes are in fact a stringy formulation of the Higgs mechanism.

6.5
Multiple Branes. Consider now N parallel branes located at xa , a = 1, . . . , N .
There are N 2 types of open string (and one type of closed string): The open string
vacua distinguished by so-called Chan–Paton factors

|0; q; ab̄io , a, b = 1, . . . , N (6.24)

with general mass formula

(xa − xb )2 1
Ma2b̄ = + (N − a). (6.25)
4π 2 κ4 κ2

Consider the vector particles at level 1 with mass |xa − xb |/2πκ2 . Massless vectors
indicate gauge symmetries.
• There are at least N massless vectors. Gauge symmetry: U(1)N .
• K coincident branes contribute K 2 massless vectors. Enhanced gauge symmetry
U(1)K → U(K).
Massive vectors indicate spontaneously broken symmetries starting from the group
U(N ).
Geometric picture of gauge symmetries:
• A stack of N branes has local U(N ) symmetry.
• Separating the branes breaks symmetry to U(K) × U(N − K).
• This makes 2K(N − K) of the vectors massive.

U(N )
(6.26)
U(K) × U(N − K)

K N −K
One can also represent SO(N ) and Sp(N ) symmetries: Unoriented strings, strings
on orientifolds (spacetime involution paired with orientation reversal).

SO(N ), Sp(N ) (6.27)

Brane Worlds. We may design many different situations by combining the

elements we have encountered so far:
• non-compact dimensions,
• D-branes,
• intersections of D-branes and non-compact dimensions,

6.6
• orientifold actions.
It then makes sense to consider the large-scale physics
• along non-compact dimensions,
• within D-branes.
One can investigate the qualitative features of the spectrum:
• Massless vectors indicate gauge symmetries.
• Light vectors indicate spontaneous symmetry breaking.
• Tachyons indicate instabilities of D-branes or spacetime.
In this sense, string theory becomes a framework analogous to QFT which has
many degrees of freedom to adjust:
• D-brane arrangements and compact directions (discrete),
• moduli for D-branes and non-compact spaces (continuous).
This is useful for physics: We should try to design the spacetime geometry for
string theory such that we obtain the standard model at low energies.
It is also useful for mathematics: String theory involves a lot of (generalised)
geometry, and dualities relate various different situations in a non-trivial fashion.

6.7
Introduction to String Theory Chapter 7
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

7 Conformal Field Theory

In the previous chapters, we have been discussing the spectrum of the string:
starting from the (local) equations of motion and imposing (global) periodicity
conditions for either the open or the closed string, we derived the spectrum of the
classical string and later for the quantum string. The latter case is described by
quantum mechanics for an infinite tower of string modes αn .
In order to deal with string scattering in the next chapter, we are going to consider
the local picture on the worldsheet: a quantum field theory for the fields X(ξ),
expressed in terms of suitable objects (local operators) on the worldsheet. As
pointed out before, due to reparametrisation invariance, the worldsheet
coordinates ξ are artificial. Only after fixing the conformal gauge the worldsheet
coordinates become meaningful. By fixing the gauge, we have used a good part of
the diffeomorphisms: what remains is the (residual) conformal symmetry. Thus we
are lead to discuss and explore a quantum field theory based on conformal
symmetry: conformal field theory (CFT).
As the structure of the final results in a theory is dictated by its symmetries, we
will take advantage of conformal symmetry here in order to obtain results
efficiently rather than calculating blindly. In particular, conformal symmetry will
turn out to be rather constraining and allows to calculate string theory results in a
neat and unique way.
While conformal symmetry is the central framework in string theory, it is
applicable in versatile situations, for example in many two-dimensional systems in
statistical mechanics.

7.1 Conformal Transformations

Conformal transformations are coordinate transformations preserving the angles
locally, but allowing for changes in the definition of length. They preserve the
metric up to a scale:
dxµ dxν !
gµ0 0 ν 0 (x0 ) = 0µ0 0ν 0 gµν (x) = f (x) gµ0 ν 0 (x). (7.1)
dx dx

Action on Coordinates. The following coordinate transformations are

comprised by the conformal group:
• Lorentz rotations xµ → Λµ ν xν ,
• translations xµ → xµ + aµ ,
• scale transformations / dila(ta)tions xµ → sxµ ,
• conformal inversions (discrete) xµ → xµ /x2 ,
• conformal boosts (inversion, translation, inversion).

7.1
The conformal group in D dimensions is the (universal cover) of SO(D, 2).

Action on Fields. Consider a free scalar with action

Z
S ∼ dD x 12 ∂µ Φ(x) ∂ µ Φ(x). (7.2)

In order for the action to be invariant under conformal transformations, the scalar
must transform under Lorentz rotations and translations as

Φ0 (x) = Φ(Λx + a). (7.3)

Invariance under a scaling x0 = sx requires

Φ0 (x) = s(D−2)/2 Φ(sx), (7.4)

while invariance under inversions translates into

Φ0 (x) = (x2 )−(D−2)/2 Φ(1/x). (7.5)

In the same way one can derive similar, but more complicated rules for
• a scalar field φ(x) with different scaling φ0 (x) = s∆ φ(sx),
• spinning fields ρµ , . . . ,
• derivatives ∂µ Φ, ∂µ ∂ν Φ, ∂ 2 Φ, . . . .

2D Conformal Symmetries. In comparison to other dimensions, QFT’s in 2D

are rather tractable. CFT’s in 2D are especially simple. The conformal group
splits into
SO(2, 2) ' SL(2, R)L × SL(2, R)R , (7.6)
where SL(2, R)L/R acts on coordinates as (dropping L/R)

aξ + b
ξ0 = , δξ = β + αξ − γξ 2 . (7.7)
cξ + d

Here β L/R are two translations, αL/R denote rotations and scaling, and γ L/R are
two conformal boosts.
The generators of the algebra underlying the group SL(2, R)L/R can be
supplemented by further generators to yield the infinite-dimensional Virasoro
algebra. Infinitesimal transformations are then given by
X
δξ L/R = L/R (ξ L/R ) = nL/R (ξ L/R )1−n , (7.8)
n

where the values n = −1, n = 0 and n = 1 correspond to the generators of

SL(2, R)L/R . The boundaries are typically distorted by Virasoro transformations,
only a subalgebra preserves them, e.g. SL(2, R).

7.2
7.2 Conformal Correlators
In a quantum theory, one is usually interested in the spectrum of operators, which
in our situation is just the string spectrum. Furthermore, one would like to
calculate probabilities, which in turn can be derived from expectation values of
operators acting on states. In a quantum field theory, there are two formulations
available for calculating the vacuum expectation values. Starting from momentum
eigenstates, the S-matrix describing the scattering of particles is defined via

h~q1 , ~q2 , . . .|S|~p1 , p~2 , . . .i = h0|a(~q1 )a(~q2 ) . . . S . . . a† (~p2 )a† (~p1 )|0i. (7.9)

In terms of position eigenstates, one calculates time-ordered correlation functions

hΦ(x1 )Φ(x2 ) . . .i = h0| T[Φ(x1 )Φ(x2 ) . . .]|0i. (7.10)

Correlator of String Coordinates. As an example of the latter quantity, one

can compute a worldsheet correlator using the underlying oscillator relations

κ2 µν
h0|X ν (ξ2 )X µ (ξ1 )|0i = − η log exp(iξ2L ) − exp(iξ1L )


2
κ2 µν
η log exp(iξ2R ) − exp(iξ1R )


−
2
+ ... . (7.11)

Is it possible, to produce the same answer from the worldsheet CFT? In order to
answer this question, let us first consider which properties the corresponding CFT
correlator should have. Assume a scalar φ of dimension ∆ and write the correlator


φ(x1 )φ(x2 ) = F (x1 , x2 ). (7.12)

The correlator should be invariant under translations

F (x1 , x2 ) = F (x1 − x2 ) =: F (x12 ), (7.13)

which means it should depend on the difference between the two points x1 and x2
only. Invariance under Lorentz rotations requires dependence on a Lorentz-scalar

F (x12 ) = F (x212 ) . (7.14)

Finally, scaling invariance demands the following equality

 !

φ(x1 )φ(x2 ) = φ0 (x1 )φ0 (x2 ) = s2∆ φ(sx1 )φ(sx2 ) ,



(7.15)

hence F (x212 ) = s2∆ F (s2 x212 ) and

N
F (x212 ) = . (7.16)
(x212 )∆
In summary, there is little freedom left.

7.3
Logarithmic Correlator. Remembering the vanishing scaling dimension
∆ = (D − 2)/2 = 0 of a scalar, we would end up with a constant correlator
F (x1 , x2 ) = N . Fortunately this is not the complete truth: let us take the limit
D = 2 + 2, N = N2 / for small 

N2 N2
F (x1 , x2 ) = 2 
→ − N2 log x212 + . . . . (7.17)
(x12 ) 

Dropping the leading (divergent) term which is independent of the separation, the
correlator can be logarithmic for  = 0 (∆ = 0). Let us now consider the argument
in light cone coordinates
x212 = −xL12 xR
12 (7.18)
and identify
xL = exp(iξ L ), xR = exp(iξ R ). (7.19)
What is the motivation for this identification? It is a two-dimensional conformal
transformation which incorporates the closed-string periodicity condition
σ ≡ σ + 2π automatically. Of course, one has to choose appropriate coordinates for
boundaries in the new coordinates.
So far, almost all conditions on the conformal correlator are taken care of. Only
translational invariance is spoiled: the string coordinates are functions of xL/R
except for the linear dependence on τ = − 2i log(xL xR ). The solution to this
problem is obvious: instead of choosing the field X alone, take ∂X µ /∂xL/R :

− 21 κ2 η µν
h0|∂L X ν (x2 )∂L X µ (x1 )|0i =
2 . (7.20)
xL2 − xL1

This is indeed closer to the expected form of a conformal correlator!

Wick Rotation. In the context of conformal field theory, the worldsheet is

conventionally taken to have euclidean signature. In order to get there, perform
the Wick rotation τ = −iτ̃ (now τ̃ is real) and define

exp(iξ L ) = exp(τ̃ − iσ) =: z̄, exp(iξ R ) = exp(τ̃ + iσ) =: z. (7.21)

The result of this change are cylindrical coordinates for the (euclidean) string:

Im z
τ̃
σ

(7.22)
Re z

The radius |z| denotes the exponential euclidean time τ̃ , while σ is mapped to the
(naturally periodic) angular coordinate. These coordinates are the standard ones
for a euclidean quantum field theory: the worldsheet coordinates z and z̄ are

7.4
complex conjugates and the fields are functions f (z, z̄) of complex z. Splitting of
the string coordinates into a right and a left part is replaced by considering a
holomorphic and a anti-holomorphic part:

X(z, z̄) = X(z) + X̄(z̄). (7.23)

Why is it favourable to use this language? Firstly, conformal transformations

preserve holomorphicity, that is, they do not mix the functions X(z) and X̄(z̄).
Secondly (and most importantly) one can now employ the powerful tools of
complex functional analysis, e.g. residue theorems.

7.3 Local Operators

In the last subsection we argued that one should use the derivative ∂X(z) and
∂¯X̄(z̄) of the string coordinate field X(z, z̄) = X(z) + X̄(z̄) in order to obtain a
quantity without linear dependence on τ .
The basic objects in a CFT are local operators Oi (z, z̄), which are built from
products of fields X and their derivatives ∂ n ∂¯n̄ X, all of them evaluated at the
same point (z, z̄) on the worldsheet. The normal ordering Oi = :. . .: is implicit for
local operators, thus there are no self-correlations. Typical examples are
O1 = :(∂X)2 : and O2µν = :X µ ∂X ν : − :X ν ∂X µ :.
In a classical setup, local operators behave as the “sum of their constituents”,
while quantum-mechanically, they need to be considered as inedependent
quantities: recall for example the quantum effects for the Virasoro charges (∂X)2 !
In the remainder of this subsection we will classify local operators in a CFT. The
classification is based upon their behaviour under conformal transformations.

Descendants. Under shifts (δz, δz̄) = (, ¯), all local operators transform as
¯
O(z + , z̄ + ¯) = O(z, z̄) + ∂O + ¯∂O
δO = ∂O + ¯∂O.¯ (7.24)

An operator ∂ n ∂¯n̄ O is called a descendant of O. Descendants are of little

importance in conformal field theory, as their structure and properties are
completely known from the operator O. Alternatively, one can argue that shifts
are symmetries and thus there is no need to consider descendants.

Weights. Most local operators (notable exception: X) are classified by their

weights (h, h̄). Considering a scaling/rotation, that is (z, z̄) → (sz, s̄z̄) or
δ(z, z̄) = (z, ¯z) one finds

O0 (z, z̄) = sh s̄h̄ O(sz, s̄z̄),

¯
δO = (hO + z∂O) + ¯(h̄O + z̄ ∂O). (7.25)

7.5
In reference to the type of transformation, scaling and rotation, ∆ = h + h̄ is called
the scaling dimension where S = h − h̄ is the conformal spin.
For a unitary CFT, both h and h̄ have to be real and non-negative. Here are two
examples of weights for some fields: ∂X → (1, 0) and (∂X)2 → (2, 0). The field X
does not have proper weights.
In a classical theory, the products of local operators O = O1 O2 will have a weight
equal to the sum of individual weights. In a quantum theory, weights are usually
not additive, due to implicit normal ordering.

Quasi-Primary Operators. A local operator with weights (h, h̄) is called

quasi-primary if
 0 h  0 h̄
0 dz dz̄
O (z, z̄) = O(z 0 , z̄ 0 ) (7.26)
dz dz̄
for all SL(2, C) Möbius transformations
az + b āz̄ + b̄
z0 = , z̄ 0 = . (7.27)
cz + d c̄z̄ + d¯
For infinitesimal boosts δ(z, z̄) = (z 2 , ¯z̄ 2 ) it must satisfy
¯
δO = (2hzO + z 2 ∂O) + ¯(2h̄z̄O + z̄ 2 ∂O). (7.28)
Descendants of quasi-primary operators are not quasi-primary. In a CFT it is
sufficient to deal with quasi-primary operators only.

Primary Operators. Primary operators are a subclass of the quasi-primary

operators described above. An operator is called primary, if it satisfies the
quasi-primary conditions for all transformations
(z, z̄) → (z 0 (z), z̄ 0 (z̄)) or (δz, δz̄) = (ζ(z), ζ̄(z̄)). (7.29)
Infinitesimally, one finds
δO = (h ∂ζ O + ζ ∂O) + (h̄ ∂¯ζ̄ O + ζ̄ ∂O).
¯ (7.30)
Of course, the infinitesimal transformations above can be derived from the general
rule here easily.
Correlators are locally conformal invariant. Only a subclass of transformations
(e.g. Möbius) leaves the correlator globally invariant. As an example, consider
− 21 κ2 η µν
hO1µ O2ν i = . (7.31)
(z1 − z2 )2
built from the primary operator Oµ = ∂X µ with weights (h, h̄) = (1, 0). The
correlator is exactly invariant under the transformation δz = z 1−n , if |n| ≤ 1
(Möbius transformations), but invariant up to polynomials for |n| > 1.1
1
The polynomial contributions are small w.r.t. the pole 1/(z1 − z2 )2 and can therefore be
ignored to some extent.

7.6
State-Operator Map. There is a one-to-one map between
• quantum states on a cylinder R × S 1 and
• local operators (at z = 0).
Below, it will be convenient to use variables obtained by the conformal map
z = exp(+iζ), z̄ = exp(−iζ̄), (7.32)
where
ζ, ζ̄ = σ ∓ iτ̃ . (7.33)
A state is then given by a wave function at constant τ̃ = − Im ζ.
• Time evolution is equivalent to radial evolution in the z-plane.
• Asymptotic time τ̃ → −∞ corresponds to z = 0.
• Local operator at z = 0 is used to excite asymptotic wave function.
• The unit operator 1 corresponds to the vacuum.

7.4 Operator Product Expansion

In a CFT we wish to compute correlation functions


O1 (ξ1 )O2 (ξ2 ) . . . On (ξn ) = F12...n . (7.34)

Suppose ξ1 ≈ ξ2 ; then one can Taylor expand

∞
X 1
O1 (ξ1 )O2 (ξ2 ) = (ξ2 − ξ1 )n O1 (ξ1 )∂ n O2 (ξ1 ). (7.35)
n=0
n!

The expansion converts local operators at two points into a sum of local operators
at a single point. The classical statement is exact.

Quantum OPE. Quantum-mechanically there are additional contributions from

operator ordering because normal ordering is implicit. However, the product of
local operators can still be written as sum of some local operators
X
i
O1 (ξ1 )O2 (ξ2 ) = C12 (ξ2 − ξ1 )Oi (ξ1 ). (7.36)
i

The above statement has to be understood in terms of a correlator. More precisely,

with the insertion of any (non-local combination of) operators “. . .” it reads

 X i

O1 (ξ1 )O2 (ξ2 ) . . . = C12 (ξ2 − ξ1 ) Oi (ξ1 ) . . . . (7.37)
i

This statement is called operator product expansion (OPE), where Cijk (ξ2 − ξ1 ) are
called structure constants and conformal blocks. The sum extends over all local
operators (including descendants).
This is a very powerful idea: every (non-local) operator can be written as an
expansion in local operators. This statement is analogous to the multipole
expansion of electrodynamics. The formalism works exactly in any CFT and is a
central tool.

7.7
Higher Points. Formally, one can compute higher-point correlation functions:
X
F123...n = ci12 Fi3...n . (7.38)
i

Recursively, one can reduce to the one-point function, which is trivial except for
the unit operator 1
hOi i = 0, h1i = 1. (7.39)
Higher-point functions can thus be reduced to a sequence of structure constants
Cijk . This is a vast simplification: one needs only Cijk in order to describe correlators
in CFT. In practice, the structure constants are hard to compute and moreover the
result of a single OPE are infinitely many local operators. Superficially, the result
seems to depend on the sequence of reducing correlators using OPE’s. This is of
course not true because the structure constants are special quantities which obey
crossing relations that ensure independence of the decomposition.

Lower Points. The two-point function is equivalent to an OPE onto the unit
operator X
Fij = hOi Oj i = Cijk hOk i = Cij1 . (7.40)
k
The OPE constants are determined by the three-point functions
X X
Fijk = hOi Oj Ok i = Cijl hOk Ol i = Fkl Cijl . (7.41)
l l

Lower-point functions are restricted tightly by conformal symmetry:

• Two-point function are non-trivial only for operators related by conformal
symmetry (descendants).
• Non-trivial three-point functions exist for three unrelated operators.
Conformal symmetry allows to map a triple of points to any other triple of points.
Consequently, there are no conformally invariant functions depending on only two
or three points. This implies the following for correlation functions:
• The coordinate dependence of a two-point function is fixed
Nij
Fij ∼ . (7.42)
|ξi − ξj |2∆i
• The coordinate dependence of a three-point function is fixed
Nijk
Fijk ∼ (7.43)
|ξi − ξj |∆ij |ξj − ξk |∆jk |ξk − ξi |∆ki
with scaling weights ∆ij = ∆i + ∆j − ∆k . The numerators Nij and Nijk depend
on dimension, spin, level of descendant and operator normalisation.
• Higher-point functions can depend on the available conformally invariant cross
ratios in an arbitrary fashion.
Once one has chosen a normalisation for operators, the data in a CFT consists of
• scaling dimensions along with the spins of the local operators: spectrum,
• coefficients of three-point function: structure constants.

7.8
7.5 Stress-Energy Tensor
The Noether currents for spacetime symmetries are encoded in the conserved
stress-energy tensor Tαβ
1 1 γδ


Tαβ = − (∂ α X) · (∂ β X) − 2
η αβ η (∂ γ X) · (∂ δ X) . (7.44)
4πκ2
This is an object of central importance for CFT and for the OPE! We know that
the trace is exactly zero because of Weyl symmetry. The two remaining
components TLL and TRR translate into the euclidean quantities
1 1 ¯ 2
T =− (∂X)2 , T̄ = − (∂ X̄) . (7.45)
κ2 κ2
Let us ignore the string physical state condition which eventually sets them to
zero, T = T̄ = 0.

Conservation. The current J(z) = ζ(z)T (z) corresponds to the infinitesimal

transformation δz = ζ(z). While classical conservation ∂J ¯ = 0 is ensured by means
of the e.o.m., in quantum mechanics the conservation law is implied by the Ward
identity:
¯
∂J(z)O(w, w̄) = 2π δ 2 (z − w, z̄ − w̄) δO(w, w̄). (7.46)
Thus, the current J(z) is conserved everywhere except at the operator locations.
In order to calculate the OPE, one integrates z over a small ball around w:
Z
1
d2 z . . . . (7.47)
2π |z−w|<

In a first step, let us evaluate the integration over z̄ ( d2 z ∂¯ . . . = −i dz . . .) and

R R

obtain
Z
1
dz J(z)O(w, w̄)
2πi |z−w|=
Z
1
= dz ζ(z)T (z)O(w, w̄)
2πi |z−w|=
= δO(w, w̄). (7.48)

An analogous calculation can be done for T̄ . In the following, we will consider the
holomorphic part only.

Stress-Energy OPE. In the last paragraph of this chapter, we would like to

derive the OPE of an operator O and the stress-energy tensor T . So let us
consider the previous equation for several types of infinitesimal transformations.
First consider the translation δz = , δO = ∂O. In order to generate the residue,
we need a simple pole:
∂O(w, w̄)
T (z)O(w, w̄) = . . . + + ... . (7.49)
z−w

7.9
Further terms with higher poles and polynomials in “. . .” are unconstrained from
considering just infinitesimal translations. So let us start with an operator O of
holomorphic weight h and consider a scaling δz = z, δO = (hO + z∂O). Upon
substitution, one needs to require the following poles in the OPE:

hO(w, w̄) ∂O(w, w̄)

T (z)O(w, w̄) = . . . + + + ... . (7.50)
(z − w)2 z−w

In a next step, suppose O is quasi-primary. Considering the scaling δz = z 2 ,

δO = (2hzO + z 2 ∂O) and substituting, we need to require the absence of the
cubic pole

0 hO(w, w̄) ∂O(w, w̄)

T (z)O(w, w̄) = . . . + 3
+ + + ... . (7.51)
(z − w) (z − w)2 z−w

Finally suppose O is primary, which is equivalent to demanding the absence of

higher poles
hO(w, w̄) ∂O(w, w̄)
T (z)O(w, w̄) = + + ... . (7.52)
(z − w)2 z−w
When there is a derivative acting on an operator, the poles are shifted by one
order. Thus, descendants are not (quasi-)primary operators.
Let us state the OPE of the stress-energy tensor. An explicit computation using
Wick’s theorem yields:

c/2 2T (w) ∂T (w)

T (z)T (w) = 4
+ 2
+ + ... . (7.53)
(z − w) (z − w) z−w

This is a general result in CFT’s. It encodes the Virasoro algebra! Let us note a
couple of properties of the stress-energy tensor:
• T is a local operator,
• T has holomorphic weight h = 2 (classically),
• T is quasi-primary,
• T is not primary (unless c = 0),
• the quartic pole carries central charge c = D.
Conformal transformations for the stress-energy tensor T are almost primary:
c 3
δT = δz ∂T + 2 ∂δz T + ∂ δz,
12
 0 2 
0 dz 0 c 0

T (z) = T (z ) + S(z , z) ,
dz 12
 3 0   0 −1 2  0 −2
3 d2 z 0

0 dz dz dz
S(z , z) = − . (7.54)
dz 3 dz 2 dz 2 dz

The additional term S is the Schwarzian derivative, which vanishes for Möbius
transformations.

7.10
Introduction to String Theory Chapter 8
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

8 String Scattering
To obtain a basic understanding of string interactions we shall compute some
scattering amplitudes. As with conventional particles, prepare an initial state
containing several strings,1 make them collide and produce several outgoing strings.
There are two fundamental approaches:
• Minkowskian/QFT picture: Consider a single string worldsheet with coordinates
τ, σ where σ is constrained to some finite range as before. Cuts the worldsheet
at specific values of σ and for some range of τ . At these cuts, the boundary
conditions are altered. Finally, integrate over all configurations of additional
boundaries.
• Euclidean/CFT picture: Insert vertex operators into the worldsheet. Each
vertex corresponds to an asymptotic string via an exponential map. Integrate
over the locations of the punctures.

(8.1)

8.1 Vertex Operators

We need to find a dictionary between string states and vertex operators, the
so-called state-operator map:
• Which operator creates a string?
• How to specify the momentum q?
• How to specify the string modes?
The solution turns out to be based on the operator

O[q] = :exp(iqµ X µ ):. (8.2)

Why? This operator is a momentum eigenstate, it receives a phase exp(iqµ µ ) for

the translation X → X + 

O[q] 7→ exp(iqµ µ )O[q]. (8.3)

1
Each piece of string carries its own string vacuum |0; qi with individual momenta and string
excitations.

8.1
String Vacuum. Let us investigate the operator further. In the CFT picture,
compute the OPE with the stress-energy tensor T
1 2 2
4
κ q O[q](w, w̄) ∂O[q](w, w̄)
T (z)O[q](w, w̄) = + + ... . (8.4)
(z − w)2 z−w

It is a primary operator with weights ( 41 κ2 q 2 , 41 κ2 q 2 )!

• Note that X itself is not primary and therefore putting X in the exponent
makes an actual difference!
• We obtain a non-trivial, non-integer weight.
• The weight is a quantum effect ∼ κ2 .
Consider the two-point correlator
2
O1 [q1 ]O2 [q2 ] ' |z1 − z2 |κ (q1 ·q2 ) .


(8.5)

In this calculation, the zero mode X µ = xµ + . . . contributes an extra factor which

is not obvious at first sight, but required for consistency
Z
dD x exp(iq1 · x + iq2 · x) ∼ δ D (q1 + q2 ). (8.6)

Hence the two-point function is compatible with the primary property of weight
( 41 κ2 q 2 , 41 κ2 q 2 )

 δ D (q1 + q2 )
O1 [q1 ]O2 [q2 ] ' 2 2
. (8.7)
|z1 − z2 |κ q1
The operator O[q](z, z̄) creates a string state at the worldsheet location (z, z̄). The
worldsheet location is unphysical, hence integrate over all potential insertion
points: Z
V [q] = gs d2 z O[q](z, z̄). (8.8)

Importantly, one can only integrate weight-(1, 1) primary operators.2 Hence:

• Mass is determined M 2 = −q 2 = −4/κ2 ; The state describes the string tachyon!
• Intercept a = ā = 1 determined by worldsheet integration.

Excited Strings. What about excited strings? A level-1 state corresponds to

Z
V [q] = gs d2 z ∂X µ ∂X
µν ¯ ν O[q]. (8.9)

• Weight is (1 + 41 κ2 q 2 , 1 + 41 κ2 q 2 ) = (1, 1) for massless q.

• Primary condition removes unphysical polarisations, e.g.
q µ O[q]
T (z) ∂X µ (w)O[q](w, w̄) ∼ + ... . (8.10)
(z − w)3
2
A scaling transformation of the integration variables is compensated by a scaling
transformation of the primary operator. Unless the two resulting factors compensate, the integral
must be zero or ill-defined.

8.2
• Gauge d.o.f. are total derivatives
Z
µν
d2 z ∂¯ ∂X µ O[q] = 0.


qν V [q] = −igs (8.11)

We obtain the following vertex operator picture:

• CFT
R 2 vacuum is empty worldsheet (genus 0, no punctures).
• R d2 z O[q](z, z̄) is string vacuum |0; qi (add puncture).
• d z . . . O[q](z, z̄) are excited string states.
• Insertions of ∂ n X µ correspond to string oscillators αnµ ; insertions of ∂¯n X µ
correspond to ᾱnµ .

(8.12)

8.2 Veneziano Amplitude

Consider an n-point amplitude (with Ok = O[qk ](zk , z̄k ))
Z
1 n−2
An ∼ 2 hV1 . . . Vn i ∼ gs d2n zhO1 . . . On i. (8.13)
gs
It is simplest to use tachyon vertex operators. We could do others, but it adds
complication (fields). The computation and the result are qualitatively the same.
Perform Wick contractions and zero mode integration to obtain

2 q ·q
Y
O1 . . . On ∼ δ D (Q) |zj − zk |κ

 j k
. (8.14)
j<k

The integral is invariant under Möbius transformations (note that qk2 = 4/κ2 ). We
map three punctures to fixed positions z1 = ∞, z2 = 0, z3 = 1.3 The remaining
integral for n = 4 external strings reads

Z
2 q ·q 2 q ·q
A4 ∼ gs2 δ D (Q) d2 z |z|κ 2 4
|1 − z|κ 3 4
. (8.15)

It can be performed exactly and yields a combination of gamma functions

Γ(−1 − κ2 s/4) Γ(−1 − κ2 t/4) Γ(−1 − κ2 u/4)
A4 ∼ gs2 δ D (Q) . (8.16)
Γ(+2 + κ2 s/4) Γ(+2 + κ2 t/4) Γ(+2 + κ2 u/4)
3
In fact, this transformation amounts to a factor of the divergent integral d2 z1 d2 z2 d2 z3 .
R

This integral does not depend on any external data and should be factored out from any
amplitude calculation. The 6 integrations correspond to the 6 global conformal symmetries.

8.3
For convenience, we have introduced the Mandelstam invariants s, t, u

s = −(q1 + q2 )2 , t = −(q1 + q4 )2 , u = −(q1 + q3 )2 , (8.17)

with the relation s + t + u = −q12 − q22 − q32 − q42 = −16/κ2 .

This is the Virasoro-Shapiro amplitude for closed strings. The corresponding
amplitude for open strings reads

Γ(−1 − κ2 s) Γ(−1 − κ2 t)
A4 ∼ gs . (8.18)
Γ(+2 + κ2 u)

It was proposed (not calculated) earlier by Veneziano based on crossing symmetry.

This is considered as the birth of string theory (as the so-called dual resonance
model).
Amplitudes have many desirable features:
• Poles at s, t, u = (N − 1)4/κ2 or s, t = (N − 1)/κ2 correspond to an infinite
tower of virtual particles exchanged. The mass spectrum coincides with closed
and open strings.

(8.19)

• Residues indicate spin J = 2N or J = N . Regge trajectory!

• Soft behaviour at s → ∞. Even for gravitons!
• Manifest crossing symmetry s ↔ t ↔ u or s ↔ t. Amazing!
Not possible to obtain such a result from QFT with finitely many particles.

8.3 String Loops

The result is exact as far as α0 is concerned. Strings on flat space is a free theory
in α0 !
However, worldsheet topology matters. String loop corrections correspond to
adding handles to the surface: higher genus. The power of gs reflects the Euler
characteristic of the worldsheet. It turns out that at string scattering at genus h
requires 2n + 6h − 6 integrations

∞ (8.20)
1 0
0

Tree Level. At tree level the worldsheet is a sphere or a disk with n punctures.
Euler characteristic is −2 + n or −1 + n/2, respectively. Due to 6 global conformal
symmetries, the pertinent integration is over n − 3 points, cf. above.

8.4
One Loop. At one loop the worldsheet is a torus with n punctures. Its Euler
characteristic is n. The torus has 2 moduli and 2 shifts as global transformations.
Therefore the integration is over 2D Teichmüller space and over n − 1 external
vertices which amounts to 2n integrations. The result is expressed using elliptic
and modular functions. Feasible problem!

Two Loops. At two loops the worldsheet is a 2-torus with n punctures and
Euler characteristic 2 + n. This torus has 6 moduli, but no global shifts: Therefore
we have to perform 2(n + 3) integrations. This is very hard, but can sometimes be
done. Higher-loop results are typically inaccessible.

8.5
Introduction to String Theory Chapter 9
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

9 General Relativity Basics

This chapter is a very brief introduction to general relativity. We shall focus on the
topics that are needed for this course of string theory, namely the definition of the
metric tensor, the Einstein equations, gravitational waves and graviton scattering.

9.1 Differential Geometry

Ordinary field theories are formulated on a flat Minkowski space. General
relativity is a theory of the curvature of spacetime. We therefore have to introduce
a framework to deal with curved spaces and spacetimes. This is Riemannian
geometry which in turn is based on differential geometry. Differential geometry
describes the basic notion of fields, coordinates and vector spaces on a general
differentiable manifold M. Riemannian geometry equips manifolds with the
concept of metric to measure distances and angles which can carry curvature.1

Diffeomorphisms. We are used to express the objects in field theory such as

fields and derivatives in terms of coordinates. In flat Minkowski geometry, there is
a small set of preferred coordinate systems, the inertial frames. Calculus in these
frames is straightforward and different frames are related by Poincaré
transformations. Special relativity essentially is the statement that laws of nature
should be the same in every reference frame.
In a general curved spacetime M, there is no preferred set of coordinates. In
analogy to special relativity one may therefore demand that the laws of nature
should be independent of the choice of coordinates. This is an axiom of general
relativity. This implies that general changes of coordinates, so-called
diffeomorphisms, must be symmetries of the theory.2

x0
(9.1)
x

For a scalar field F (x), a diffeomorphism x 7→ x0 maps the field F 7→ F 0 such that
F 0 (x0 ) = F (x). (9.2)
1
Note that the curved space should be viewed as an intrinsic space, not (necessarily)
embedded into some higher-dimensional space.
2
This statement is to some extent empty because every theory can be formulated in arbitrary
coordinate systems (e.g. spherical coordinates, cylindrical coordinates) such that appropriately
defined equations are invariant under diffeomorphisms. A more accurate statement for general
relativity is that the diffeomorphism-invariant equations take a simple and natural form.
Unfortunately, this statement is just an aesthetic one.

9.1
The transformation rules for partial derivatives and vector fields require some more
care. However, concatenation of diffeomorphisms x 7→ x0 7→ x00 must act
transitively, x 7→ x00 .

Vectors, Covectors, Tensors. Let us next understand how partial derivatives

and vector fields transform under diffeomorphisms.
Define Gµ as the partial derivative ∂µ of a (scalar) function F

∂
Gµ (x) := F (x). (9.3)
∂xµ
We know how the derivative transforms under coordinate transformations
∂ ∂xν ∂ ∂xν
G0µ0 (x0 ) := F 0 0
(x ) = F (x) = Gν (x). (9.4)
∂x0µ0 ∂x0µ0 ∂xν ∂x0µ0
In addition to the change of coordinates, the basis of the vector space is
0
transformed by the linear map ∂xν /∂x0µ . Note that this rule is compatible with
the transformation rule for vectors under Poincaré transformations
0µ µ ν µ ∂xν −1 ν
x = A νx + B , 0µ 0 = (A ) µ0 . (9.5)
∂x

Now we can generalise the above transformation rule for covariant derivatives to a
new class of fields: A covector field Fµ (x) is a field which transforms according to

∂xν
Fµ0 0 (x0 ) = Fν (x). (9.6)
∂x0µ0

A vector field F µ is the dual of a covector field Gµ . The contraction of the two
should be a scalar field which transforms appropriately
0
F 0µ (x0 )G0µ0 (x0 ) = F µ (x)Gµ (x). (9.7)

The vector index therefore transforms with the inverse transformation matrix
0
0 ∂x0µ ν
F 0µ (x0 ) = F (x). (9.8)
∂xν

In what follows we will have to combine the concept of vector and covector fields
in so-called tensor fields. A tensor field of rank (a, b) transforms as the tensor
product of a vector fields and b covector fields, e.g.
0 0
0 0 ∂x0µ ∂x0ν ∂xρ µν
F 0µ ν ρ0 (x0 ) = F ρ (x). (9.9)
∂xµ ∂xν ∂x0ρ0

The above definition of tensor fields is based on partial derivatives of scalar fields.
Note, however, that the partial derivatives of generic tensor fields are not tensor
fields. This is because, in diffeomorphisms, the covariant derivatives can act on the
transformation matrix for the tensor indices spoiling the tensor character of the

9.2
field. The deeper reason for this behaviour is that the tangent spaces at different
points are a priori unrelated. The notable exception are among anti-symmetric
covector indices, so called differential forms. For example, for any covector field
Fµ , the field
Gµν := ∂µ Fν − ∂ν Fµ (9.10)
transforms as a tensor of rank (0, 2).

Index-Free Formulation. The notion of indices of tensor fields is tightly bound

to the respective coordinate systems. It is sometimes easier to deal with index-free
quantities, which is also closer to the mathematical literature. As an aside, let us
introduce these.
An index-free vector field F is defined as a derivative operator
∂
F (x) = F µ (x) . (9.11)
∂xµ
It is evident that such derivative operators define a vector space. With the above
transformation rule of a vector field F µ with indices, F transforms as a
(derivative-valued) scalar field. The partial derivatives ∂/∂xµ serve as a basis of
vector fields,3 and the F µ are the basis coefficients.
An index-free covector field F is defined as the differential one-form

F (x) = Fµ (x) dxµ . (9.12)

The one-forms dxµ serve as a basis of differential one-forms and the covector field
Fµ with indices as the basis coefficients.
Differential one-forms are the duals of the partial derivative operators

dxµ (∂xν ) = δνµ . (9.13)

The differential of a scalar field dF is defined as the covector field

dF = dxµ ∂µ F. (9.14)

It satisfies the property that for any vector field X

XF = X µ ∂µ F = X µ ∂ν F dxν (∂µ ) = dF (X). (9.15)

Atlases and Bundles. Let us add some general considerations of differential

geometry which may not be immediately relevant to our purposes.
In differential geometry, coordinates are hardly ever globally defined on a
manifold. They are usually only valid in local (open) patch of the manifold as a
so-called chart. A manifold is covered by a set of overlapping charts, a so-called
3
Somewhat confusingly, ∂/∂xµ obeys the transformation rules of a covector field ∂µ with
indices. However, this is implied by the requirement that the index-free vector field transforms
trivially.

9.3
atlas. In the overlap of two charts, the coordinates are translated by the transition
map which must be diffeomorphism. Whenever three charts overlap, the
diffeomorphisms must be compatible.

(9.16)

Charts and transition maps provide a consistent mathematical definition of the

physicists’ notion of coordinates and diffeomorphisms. Importantly, these are local
concepts on the manifold. In fact, for many manifolds there is no global chart
which covers the whole manifold. In other words, any given coordinate system may
not describe the whole universe. Similarly, diffeomorphisms need not cover the
whole coordinate system.
An atlas may contain any number of charts such that the manifold is covered
multiple times. One can also introduce additional charts to accommodate for other
choices of coordinate systems. An abstract object is the maximal atlas which
contains all admissible charts and therefore all conceivable diffeomorphisms
between them. An atlas with a small number of charts suffice for practical
purposes.
Let us furthermore elaborate on the mathematical definition of vectors. A tangent
vector at the point x is defined as a derivative operator at this point with the
usual properties. Tangent vectors at the point x span a vector space, the tangent
space Tx M. Tangent spaces at different points are isomorphic, but should not be
compared directly. The tangent bundle is defined as the collection of all tangent
spaces
T M := {(x, y); x ∈ M, y ∈ Tx M}. (9.17)
The tangent bundle is a manifold of twice the dimension of the underlying
manifold. It inherits a topological structure from the manifold. This structure is
generally not the direct product of the manifold and one of its tangent spaces, i.e.
T M can have a non-trivial topology. Therefore T M is a fibre bundle over M with
the projection
π : T M → M, π(x, y) = x. (9.18)

The role of the tangent bundle is to properly define a vector field F . In principle,
we want
F (x) ∈ Tx M (9.19)
This function has an ill-defined type F : M → T? M. Instead one defines a vector
field as a section of the tangent bundle with type

F : M → T M. (9.20)

The section must map a point x to the tangent space T M at this point. This can
be achieved by the constraint
π ◦ F = id . (9.21)

9.4
This somewhat bloated structure is required for mathematical consistency, but it
does not hurt too much to consider F (x) ∈ Tx M

9.2 Riemannian Geometry

Riemannian geometry introduces a metric field on a differentiable manifold. This
field serves multiple purposes: It introduces a measure of distance and of angles.
Furthermore, it relates the tangent spaces at nearby points. This allows to define a
measure of curvature.
For Riemannian geometry the metric must be positive definite. A theory of
spacetime, however, requires a metric of signature (D − 1, 1). This generalisation
of Riemannian geometry is straight-forward and is called pseudo Riemannian
geometry. We shall not make a distinction here.

Metric. The metric tensor field gµν (x) is a tensor of rank (0, 2) with symmetric
indices µ, ν. Most immediately, it defines the length L for a curve xµ (τ ) 4
Z q
L = dτ |ẋ|, |ẋ| := gµν (x) ẋµ ẋν . (9.22)

Angles between intersecting curves can be measured as usual via the scalar product

y
gµν ẋµ ẏ ν α
cos α = . (9.23)
|ẋ| |ẏ|
x

More generally, the metric define the length squared ds2 of infinitesimal line
elements dxµ via
ds2 = gµν (x) dxµ dxν . (9.24)

The metric field has an inverse g µν (x) with the property

g µν (x)gνρ (x) = δρµ . (9.25)

As in special relativity and Minkowski spacetime, these two instances of the metric
are used to lower and raise indices,

Fµ = gµν F ν , F µ = g µν Fν . (9.26)

More abstractly, the metric translates between tangent and cotangent spaces.
Therefore, there is essentially no distinction between tangent and cotangent
vectors, and tensor fields can be classified by their overall rank alone.
4
The definition of length applies to space-like curves; for time-like curves the notion of length
(proper time) requires the opposite different sign under the square root. Light-like curves with
|ẋ| = 0 have no proper length by definition.

9.5
Covariant Derivatives. We already pointed out that tangent spaces at
different points are unrelated. Therefore, partial derivatives of tensor fields are not
tensor fields. This represents be a principal difficulty towards setting up a field
theory on curves spaces. It can be overcome by introducing a covariant derivative
which acts on a tensor field according to its tensor structure

Dµ F = ∂µ F,
Dµ F ν = ∂µ F ν + Γµρ
ν
F ρ,
ρ
Dµ Fν = ∂µ Fν − Γµν Fρ ,
ρ 0 0 0
Dµ Fσνρ = ∂µ Fσνρ + Γµν
ν ν ρ
0 Fσ + Γµρ νρ
0 Fσ
σ
− Γµσ Fσνρ0 ,
... . (9.27)
µ
The affine connection Γνρ (x) has the index structure of a tensor field of rank (1, 2),
however, its transformation properties are slightly different to absorb the undesired
terms in the transformation of derivatives of tensor fields
0 0
0 ∂x0µ ∂xν ∂xρ µ ∂x0µ ∂ 2 xσ
Γν0µ0 ρ0 (x0 ) = Γ (x) + . (9.28)
∂xµ ∂x0ν 0 ∂x0ρ0 νρ ∂xσ ∂x0ν 0 ∂x0ρ0
The above definition of covariant derivative respects contractions of indices in the
sense that
Dµ (F ν Gν ) = (Dµ F ν )Gν + F ν (Dµ Gν ) = ∂µ (F ν Gν ). (9.29)
A further desirable property would be that the lowering and raising of indices
commutes with the covariant derivative. This implies that the metric must be
covariantly constant
Dµ gνρ = 0. (9.30)
µ
This condition imposes strong constraints on the affine connection Γνρ . In
particular, the part with symmetric lower indices νρ is fully determined by the
metric. To further constrain the antisymmetric part, we can impose a condition on
the second derivatives of an arbitrary scalar field F

Dµ Dν F = Dν Dµ F. (9.31)

This constraint implies the absence of torsion. Together the constraints determine
µ
the so-called Christoffel connection Γνρ completely
µ
= 21 g µσ ∂ν gσρ + ∂ρ gνσ − ∂σ gνρ .


Γνρ (9.32)

The transformation properties for this combination as a connection essential follow

by construction via covariant equations.

Curvature. The covariant derivative can be used to construct interesting

observable a Riemannian manifold, the curvature tensor Rµ νρσ . It follows from the
commutator of covariant derivatives acting on a vector field Fρ

Dµ Dν F ρ − Dν Dµ F ρ = Rρ σµν F σ . (9.33)

9.6
More explicitly, it can be expressed in terms of the Christoffel connection
Rµ νσρ = ∂ρ Γνσ
µ µ
− ∂σ Γνρ κ
+ Γνσ µ
Γκρ κ µ
− Γνρ Γκσ . (9.34)
It is called the curvature tensor because the vanishing of all of its components is
equivalent to the existence of a local coordinate system where the metric field is
constant.
The curvature tensor has a number of useful properties which follow from its
definition, compatibility with the metric and the absence of torsion
Rµνρσ = −Rµνσρ = Rρσµν (9.35)
as well as
Rµνρσ + Rµρσν + Rµσνρ = 0. (9.36)
Furthermore the Jacobi identity of covariant derivatives imply the so-called
Bianchi identities
Dκ Rµνρσ + Dρ Rµνσκ + Dσ Rµνκρ = 0. (9.37)

9.3 General Relativity

Let us state the equations of general relativity without further ado.

Einstein Equations. General relativity has a very natural formulation in

Riemannian geometry. With a few basic assumptions we can derive a set of field
equations for general relativity: The equations should
• ... be manifestly covariant,
• ... involve up to two derivatives of the metric,
• ... couple to matter fields appropriately,
• ... have the desired non-relativistic limit,
The covariant objects (tensor fields) we have at our disposal in pure Riemannian
geometry are
• the metric gµν ,
• the curvature tensor Rµνρσ ,
• covariant derivatives Dµ .
However, the covariant derivatives turn out not to be suitable for our purposes
because the metric is covariantly constant and the curvature tensor already
involves two derivatives of the metric.
By means of the equivalence principle of inertial and gravitational mass, one can
expect that matter fields enter the gravitational equations of motion only through
the energy-momentum tensor Tµν . This tensor field is covariantly conserved
Dµ Tµν = 0, (9.38)
as an expression of conservation of energy and momentum. The energy-momentum
tensor is of second order in the derivatives of the matter fields, and therefore
should contribute no more than linearly to the equations of motion.

9.7
Considering the tensor structures of the various objects, we can make the ansatz5

Rµν + aRgµν + bgµν + cTµν = 0, (9.39)

with three constants a, b, c to be determined. Here Rµν = Rρ µρν is the so-called

Ricci tensor and R = g µν Rµν is the Ricci scalar. From conservation of the
energy-momentum tensor it follows that a = − 12 . Furthermore, c must be matched
with Newtons constant G as c = 8πG. Only a fool would set the cosmological
constant b = Λ to b = 0. Altogether we obtain the Einstein equations

Rµν − 12 Rgµν + Λgµν + 8πG Tµν = 0. (9.40)

Einstein–Hilbert action. The above Einstein equations follow from an action

functional as the Euler–Lagrange equations. Consider the Einstein–Hilbert action
Z
1 p
dD x − det g R − 2Λ + Smatter .


S= (9.41)
16πG
√
Note that − det g·· is not a scalar field although it carries no indices. It is rather
a scalar density field which transforms as
p p
− det g··0 (x0 ) = det(∂x/∂x0 ) − det g·· (x).

(9.42)

The additional transformation factor compensates the Jacobian factor√of the

transformation of the measure dD x. Altogether the combination dD x − det g
serves as the invariant volume element.
The variation is most conveniently computed using a variational identity of the
Ricci scalar R = g µν Rµν

δR = δg µν Rµν = −δgµν Rµν (9.43)

and the variation of a determinant δ det g = δgµν g µν . Furthermore, the variation of

the matter w.r.t.
R Dthe√metric yields the energy-momentum tensor
1 µν
δSmatter = − 2 d x − det g δgµν T . Altogether we obtain
1
Rµν − 21 g µν R + Λg µν − T µν = 0


− (9.44)
8πG

Gravitational Waves. Let us find solutions of the gravitational equations

which are small perturbations hµν (x) of plain Minkowski spacetime

gµν (x) = ηµν + hµν (x). (9.45)

Since we are interested in terms linear in the small field h, we can make an ansatz
in terms of a single plane wave with momentum p and polarisation tensor µν

hµν (x) = µν exp(ip · x). (9.46)

5
Furthermore, we may or may not demand that the totally traceless part of Rµ νρσ is zero.
This does not lead to a desirable field equation for physics.

9.8
The resulting linearised Christoffel symbols and curvature tensor and Ricci tensor
read
µ
= 2i pν µρ + pρ µν − pµ νρ exp(ip · x),


Γνρ
Rµ νσρ = 12 −pρ pν µσ + pρ pµ νσ + pσ pν µρ − pσ pµ νρ exp(ip · x),

Rνρ = 12 −pµ pν ρρ + pµ pρ ρν + pρ pν µρ − p2 µν exp(ip · x).

(9.47)

The Einstein equations without matter and cosmological constant boil down to
Rµν = 0, i.e.
− pµ pν ρρ + pµ pρ νρ + pρ pν µρ − p2 µν = 0. (9.48)
For p2 6= 0 the equation immediately implies that µν must be parallel to either pµ
or pν . An ansatz is
µν = ξµ pν + ξν pµ . (9.49)
This is in fact already a solution which corresponds to the diffeomorphisms
δxµ = ξ µ applied to trivial Minkowski space gµν = ηµν . Therefore it is a physically
trivial solution which we can ignore.
Consequently, we only consider the case p2 = 0 Let us furthermore define
πµ = pρ µρ . Our equation then reads

− pµ pν ρρ + pµ πν + pν πµ = 0. (9.50)

This implies that π must be parallel to p. By means of a diffeomorphism we can

actually remove the the component of π parallel to p, and therefore set π = 0
without restrictions on generality. The remaining equation demands that  is
traceless. Consequently, the physical plane wave solutions of general relativity are
enumerated by traceless, symmetric tensors µν modulo pµ which are orthogonal to
pµ . Altogether these are D(D − 3)/2 degrees of freedom which form a symmetric
traceless representation of SO(D − 2).

Graviton Scattering. The quantisation of gravitational waves leads to graviton

excitations. One can expand around flat background with gravitons using the
above weak coupling ansatz
√
gµν (x) = ηµν + G hµν (x). (9.51)

Here we have inserted the Newton constant in order to achieve a canonically

normalised kinetic term. Since the Einstein–Hilbert action is non-polynomial
√ in
gµν due to the occurrence of the inverse field g µν and due to the term − det g,
one obtains a tower of interaction terms for the graviton field hµν (x)

S ' S2 + G1/2 S 3 + G S 4 + G3/2 S 5 + . . . . (9.52)

Scattering of four gravitons therefore has an effective coupling constant of G. This

is to be compared to the corresponding process in string theory where one finds
g̃s2 := gs2 κD−2 . The factor of κD−2 is required to match the mass dimension mass

9.9
dimension 2 − D of Newton’s constant, hence κ defines the Planck scale up to
factors of gs .

√
S= + G +G + G3/2 + ...

∼ + g̃s + g̃s2 + g̃s3 + ... (9.53)

9.10
Introduction to String Theory Chapter 10
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

10 String Backgrounds
We start with contemplations on the connection of string theory and general
relativity:
• We have seen that the string spectrum contains the graviton. This graviton
interacts according to the laws of general relativity (up to stringy corrections at
higher orders) which is a theory of spacetime geometry.
• So far we have assumed that strings move in a flat background with canonical
coordinates. However, strings can also move in a curved background described
by general relativity.
How do these two connections fit together?
• Should we quantise the string background?
• Is the string graviton the same as the Einstein graviton?
• Is there a back-reaction between strings and gravity?

10.1 Graviton Vertex Operator

Let us compare the graviton as a string excitation and as an excitation of the
background metric. We assume it has momentum q and polarisation tensor µν .

Vertex Operator Construction. The graviton is represented by the closed

string state
L,µ R,ν L,ν R,µ


|; qi = µν α−1 α−1 + α−1 α−1 |0; qi. (10.1)
The corresponding vertex operator in the CFT description reads
¯ ν + ∂X ν ∂X
Oµν = :(∂X µ ∂X ¯ µ )eiq·X :
p
∼ : det −g g αβ ∂α X µ ∂β X ν eiq·X :. (10.2)

In the second line we have reinstated the worldsheet metric to make the expression
valid outside the conformal gauge. Insertion of the vertex operator into the string
worldsheet yields the expression
Z
V = d2 ξ 12 µν Oµν . (10.3)

Background Metric Construction. We start with a flat background metric

and perturb it by a plane wave

Gµν (x) = ηµν + µν eiq·x + . . . . (10.4)

10.1
The coupling of strings to generic background is straight-forward by the
replacement ηµν → Gµν in the worldsheet action
Z
1 2
p
S=− d ξ − det g g αβ 12 Gµν (X) ∂α X µ ∂β X ν . (10.5)
2πκ2

We notice that a string in a background with a weak gravitational wave has the
same action as a string in flat space deformed by the graviton vertex operator
1
S = S0 − V + ... . (10.6)
2πκ2

Conclusion. We conclude that the graviton mode of the string spectrum is

equivalent to a gravitational wave in the background on which the string
propagates. There is no conflict between the two ways in which gravity appear in
string theory; they are actually the same.
We can thus argue as follows: The quantum string on flat space contains gravitons
as excitations. The gravitons introduce curvature and thus deform flat
background. Therefore string theory contains quantum gravity upon quantisation.
This is the picture for a small number of excitations and small deformations of flat
space. Large deformations away from a flat background should be represented by
coherent states of a large number of gravitons.1
How about the dependence on the background on which string theory is
formulated? Due to the appearance of graviton modes, string quantisation
automatically probes nearby backgrounds. The low-energy stringy physics will
thus see the classical background and nearby geometries. However, the full
quantum string theory can be expected to be independent of the background
because it contains all backgrounds as different states.2

10.2 Curved Backgrounds

Consider strings propagating on a curved background described by the metric field
Gµν (x); a curious insight awaits.
The action in conformal gauge reads
Z
1
S=− d2 ξ 12 Gµν (X) η αβ ∂α X µ ∂β X ν . (10.7)
2πκ2
For a generic metric G, the equations of motion for X are non-linear. This type of
model is called a non-linear sigma model. The string background is called the
1
This is a leap of faith, but it should work somewhat analogously to the electromagnetic field.
For instance, the particle picture of photons describes the photo-electric effect very well, whereas
the generation of a Coulomb potential around a charged object is hardly intuitive in this picture.
2
This is essentially the same issue as background dependence in quantum gravity. Note,
however, that the asymptotic behaviour of the background may have a relevant influence on the
quantum string and quantum gravity theory because it would take an infinite amount of energy
to change the geometry asymptotically.

10.2
target space of the model. The metric field Gµν (x) acts as the coupling constants
for the model. In fact, there are infinitely many coupling constants contained in
the field G(x) when Taylor expanding it around some point x.
In most QFT’s, coupling constants are renormalised upon quantisation. The
problem here is:
• The classical action has conformal symmetry.
• Conformal symmetry is indispensable to remove some degrees of freedom.
• The renormalised coupling G(x, µ) depends on a scale µ which is introduced by
the quantisation or regularisation process.
• This new scale breaks quantum conformal invariance of the string. There may
be a conformal anomaly threatening consistency of the quantum string!

Background Field Method. Let us compute the conformal anomaly. A

well-suited method is background field quantisation:
• Pick a (reasonably simple) classical solution X0 of the string equations of motion.
• Then add fluctuations to this field X = X0 + κY . The field Y is considered to
be the quantum field while X0 remains a classical field.3
Next, expand the action S[X] = S[X0 ] + Y 2 + κY 3 + . . . in orders of Y (or κ).
• The value of the classical action S[X0 ] at order Y 0 does not matter for the
quantum theory.
• There is no linear term in Y due to the classical equations of motion for X0 .
• The second order Y 2 serves as the kinetic term for the quantum field Y .

(10.8)

• The higher orders Y 3 , Y 4 , . . . are cubic, quartic, . . . interactions of the quantum

fields.

κ + κ2 + κ3 + ... (10.9)

We can now use a trick to simplify the interactions slightly for our purposes: We
use target space diffeomorphisms such that locally, at a particular point in
spacetime x, the metric is stationary.4 This is convenient because the quadratic
terms are governed by the target space curvature tensor

Gµν (x + δx) = Gµν (x) + 13 Rµρνσ (x) δxρ δxσ + . . . . (10.10)

3
It can also be viewed as the expectation value of the field X.
4
Consideration of a single point is sufficient for the leading order divergence. For higher orders
we should make sure that the metric is stationary everywhere one the classical worldsheet
solution. This can be achieved for any geodesic curve (point-like string) and potentially for fully
geodesic string configurations. Otherwise one will have to deal with a few additional interaction
terms.

10.3
Focusing on this point, we will find fewer interactions and our results are
automatically covariant. To quadratic order in Y we then find only two terms
Z 2
d ξ αβ
Gµν ∂α Y µ ∂β Y ν + 31 Rµρνσ ∂α X0µ ∂β X0ν Y ρ Y σ .


S2 = − η (10.11)
2π
Here we can consider the first term in the quadratic action S2 as the kinetic term
for the quantum field Y and the second term as a perturbation. To this end, we
use the field ∂X0 as another formal expansion parameter, and there is only one
interaction vertex at order κ0 . In the corresponding graph we shall mark the
classical field ∂X by a dashed line.

(10.12)

Note that without the above trick, there is an additional vertex to be taken into
account, which leads to many more graphs to be considered.5

(10.13)

One-Loop Renormalisation. Let us now compute one-loop quantum

corrections to the classical action S[X0 ] in order to understand the regularisation
issues. When integrating out the quantum field Y at κ0 , we obtain a tower of
bubbles with an increasing number of insertions of ∂X.

... (10.14)

Among these contributions, the only UV divergent term by power counting is the
first one. Let us therefore investigate it more closely. Fixing the location ξ of the
vertex on the worldsheet,6 the diagram evaluates to

Rµρνσ ∂α X0µ (ξ)∂ α X0ν (ξ) Y ρ (ξ)Y σ (ξ) .


(10.15)

Here, the two-point correlator of quantum fields depends strongly on the

background metric Gµν . However, in the ultraviolet region ξ1 → ξ2 the dependence
on the metric trivialises and we can trust the results of CFT

ρ
Y (ξ1 )Y σ (ξ2 ) ' −Gρσ log |ξ1 − ξ2 |.

(10.16)
5
It may be necessary to include this vertex at higher orders because the above simplification
works only at one specialised point.
6
At this point the field X µ (ξ) should point to the location where the metric Gµν (X) simplifies.

10.4
So we see that the diagram has a logarithmic divergence whose structure
∂α X0µ ∂ α X0ν matches the structure of the classical action S[X0 ]. Therefore we can
regularise the theory by renormalising Gµν . This makes Gµν (x) a running coupling
constant with one-loop beta function
µ∂
Gµν = βµν = κ2 Rµν , Rµν = Rρ µρν . (10.17)
∂µ

Anomaly. Dependence on a scale evidently breaks conformal symmetry: We find

that the stress-energy tensor has acquired a trace after renormalisation which is
proportional to the above beta function
1
η αβ Tαβ = − 2
βµν η αβ ∂α X µ ∂β X ν . (10.18)
2κ
This describes the anomaly of Weyl symmetry!
We know that Weyl symmetry is essential for obtaining the correct degrees of
freedom for string theory. The only way to remove them and make string theory
consistent is to set βµν = 0. Surprisingly, this yields the Einstein equation

Rµν = 0. (10.19)

In other words, quantum strings can propagate only on Einstein backgrounds. We

have thus found another way to recover general relativity from string theory, and
we gain confidence that the spin-2 particles at level one are in fact gravitons.

Higher Corrections. To understand higher-order corrections in κ, let us list the

relevant vertices.
 
 
κ
 + 


 
 
+ κ2 
 + + 
 (10.20)

The divergent corrections to the classical action at higher perturbative orders in κ2

take the form7
 
 
+ κ2 
 + +  + ... .
 (10.21)

7
This set of diagrams is based on the assumption of a vanishing Christoffel symbol on the
worldsheet. Several further diagrams may contribute otherwise.

10.5
They amount to the following two-loop beta function8
βµν = κ2 Rµν + 21 κ4 Rµρσκ Rν ρσκ + . . . . (10.22)
Absence of the conformal anomaly, βµν = 0, implies that in string theory the
Einstein equations receive corrections at the Planck scale.

10.3 Form Field and Dilaton

What about the other (massless) fields: two-form Bµν and dilaton scalar Φ? If we
view them as fields in/of the background manifold, can we couple them to the
string action as well?
The two-form naturally couples via antisymmetric combination of worldsheet
derivatives Z Z
1 2 1 αβ µ ν 1
d ξ 2 Bµν (X) ε ∂α X ∂β X = B. (10.23)
2πκ2 2πκ2
In fact, this is the canonical coupling of a two-dimensional worldsheet to a
two-form in direct analogy to the coupling of a charged particle worldline to the
electromagnetic field. The string carries a two-form charge as we shall see a little
later.
The dilaton couples to the worldsheet Riemann scalar
Z
1 p
d2 ξ − det g Φ(X) R[g]. (10.24)
4π
This is interesting for several reasons:
• The Euler characteristic χ of the worldsheet appears.
• The coupling is not Weyl invariant.
• The scalar can mix with gravity.
• We can get away from 26 dimensions.
Let us discuss them in more detail below.

Low-Energy Effective Action. First we discuss the various beta functions

(trace of the renormalised stress-energy tensor T ) which appear in string theory
coupled to a generalised background consisting of G, B and Φ
1 p
g αβ Tαβ = − 2 G αβ B αβ
ε ∂α X µ ∂β X ν


− det g βµν g + βµν
2κ
1 Φ
− 2 β R[g], (10.25)
where
G
βµν = κ2 Rµν + 2κ2 Dµ Dν Φ − 41 κ2 Hµρσ Hν ρσ ,
B
βµν = − 12 κ2 Dλ Hµνλ + κ2 Dλ ΦHµνλ ,
β Φ = − 21 κ2 Dµ Dµ Φ + κ2 Dµ ΦDµ Φ − 1 2
24
κ Hµνρ H µνρ . (10.26)
8
Note that there are also corrections from the expansion in the string coupling gs which we do
not consider here.

10.6
Quantum consistency of (conformal symmetry in) string theory requires
β G = β B = β Φ = 0. These are the standard equations for a graviton, a two-form
field and a scalar. They follow from an action in spacetime
√
Z
S ∼ d26 x − det G e−2Φ R − 12 1
Hµνρ H µνρ + 4∂ µ Φ ∂µ Φ .


(10.27)

This is the string theory low-energy effective action. It encodes the low-energy
physics of string theory viewed from the perspective of the background. Note that
there are further corrections from curvature (κ) and loops (gs ) not displayed here.
The anomaly equations have the trivial solution G = η, B = 0, Φ = Φ0 which
describes a flat background. The solution equally applies to torus compactification
to effectively reduce the number of dimensions.

String Coupling. Suppose Φ = Φ0 is constant, then the dilaton coupling term is

topological Z p
d2 ξ − det g R[g] ∼ χ. (10.28)

It measures the Euler characteristic χ = 2h − 2 of worldsheet.

Set gs = eiΦ0 . Then the generating functional yields precisely χ factors of gs .

eiS ' eiΦ0 χ = gsχ . (10.29)

The expansion in this gs thus corresponds to an expansion in worldsheet topology

gs−2 + gs0 + gs2 + ... . (10.30)

For general curved backgrounds, the string coupling gs is determined by the

asymptotic behaviour of the background, namely the asymptotic value Φ0 of
dilaton field Φ.

String Frame. Notice the unusual factor of exp(−2Φ) in the above effective
action S.
The scalar degrees of freedom can mix with the gravitational degrees of freedom.
We may as well define a rescaled metric

G0µν = f (Φ)Gµν . (10.31)

With a suitable choice of f we can remove the factor exp(−2Φ). We can go from
the so-called string frame with exp(−2Φ) to the so-called Einstein frame where we
recover the canonical kinetic term for each field.

10.7
Non-Critical Strings. We have seen earlier that D 6= 26 breaks Weyl symmetry
of the string action. Furthermore it is broken by the above anomalies. Therefore
both anomalies should appear in the same place. In fact, D enters in the effective
action as the worldsheet cosmological constant
1
Hµνρ H µνρ + 4∂ µ Φ ∂µ Φ − 32 κ−2 (D − 26) .


S = . . . R − 12 (10.32)

This implies that we can have D < 26, but it requires a spacetime curvature at the
Planck scale.

Dilaton Scaling. We note two further peculiarities: The dilaton coupling to the
worldsheet is not Weyl invariant and it has an unconventional power of κ.
The factor of κ moves the classical Weyl breakdown effectively to one loop. There
it can cancel against the quantum anomalies of other fields. Together this
furnishes a consistent choice of normalisation.

10.4 Open Strings

The introduction of open strings leads to additional states, fields and couplings
which complete the geometric picture of D-branes in string theory in an exciting
way.
• There are additional string states; e.g. the massless vectors (photon):
µ
|ζ; qi = ζµ α−1 |0; qi. (10.33)

• They correspond to additional vertex operators defined on the ends of the

string; e.g. for the photon
Z
V [ζ, q] ∼ dτ ζµ ∂τ X µ exp(iq · X). (10.34)

• The additional fields of the background geometry couple to the ends of the
string. They can be identified as for closed strings such that adding the vertex
operator to the action has the same effect as switching on a background field.
Evidently, the coupling depends on the choice of string boundary conditions. We
have already seen that the latter can be identified with Dp-branes. Let us therefore
discuss the arising Neumann and Dirichlet boundary conditions.

Neumann Boundaries. Consider the coordinates X a , a = 0, . . . , p, with

Neumann conditions: We couple an abelian one-form gauge field A to the end of a
string Z Z
a
dτ Ẋ Aa (X) = A. (10.35)
end end
• This is the natural coupling of a charged point-particle to a gauge field A.
• The string end is a charged point-like object with a worldline.

10.8
Note: The gauge field Aa needs to exist only on the Dp-brane to which the string
ends are constrained.
The classical coupling of A respects Weyl symmetry. The quantum anomaly is
described by a beta function
βaA ∼ κ4 ∂ b Fab . (10.36)
Absence of the conformal anomaly requires the Maxwell equation ∂ b Fab = 0 to
hold. Consequently, the associated low-energy effective action reads
Z
4
S ∼ −κ dp+1 x 14 Fab F ab . (10.37)

It is an action in the (d + 1)-dimensional worldvolume of the Dp-brane.

For planar Dp-branes in a flat background we can also include higher corrections in
κ. This leads to the exact Born–Infeld action
Z p
S ∼ dp+1 x − det(ηab + 2πκ2 Fab ) . (10.38)

The leading order is the Maxwell kinetic term, but there are corrections at higher
orders in κ.

Dirichlet Boundaries. The coupling of the Dirichlet directions X m ,

m = p + 1, . . . , D − 1, is rather different:
• The field X m is fixed, but we can use X 0m for the coupling to background fields.
• The corresponding background field Ym describes displacement of the Dp-brane
in transverse directions.
• Dp-branes are in fact dynamical objects!
The beta function at leading order describes a collection of massless scalars
Y
βm ∼ ∂ a ∂a Ym . (10.39)

The leading order action for these degrees of freedom is evident. The effective
action for planar Dp-branes at higher orders is the Dirac action
Z p
S ∼ dp+1 x − det(gab ) , (10.40)

which measures the volume of the Dp-brane via the induced worldvolume metric
gab = ∂a Y m ∂b Ym . This form of action clearly identifies the field Ym as a
displacement of the Dp-brane away from its classical planar configuration.

Ym (xa )
(10.41)

10.9
D-Branes Effective Actions. In fact, one can combine the actions for all open
string effective degrees of freedom into the so-called Dirac–Born–Infeld action
Z p
S ∼ dp+1 x − det(gab + 2πκ2 Fab ) . (10.42)

This is a combination of
• the Dirac action for the dynamics of p-branes and
• the Born–Infeld action for gauge fields on the p-brane.
We can even add the effect of all the closed string fields to the p-brane action
Z p
S ∼ dp+1 x e−Φ − det(gab + 2πκ2 Fab + Bab ) . (10.43)

• gab is the induced metric from the curved background.

• Bab is the pull back of the 2-form field Bµν to the Dp-brane.
• The combination 2πκ2 Fab + Bab is gauge invariant.
• The dilaton couples as a prefactor as for the closed string.

Coincident Branes. For a single D-brane, the vector field Aa has an associated
U(1) symmetry. For N coincident D-branes the gauge group enlarges from U(1)N
to U(N ).
The non-abelian gauge field should couple to the end of a string via a Wilson line
Z
T exp A. (10.44)
end

The resulting effective action at leading order is

Z
S ∼ dp+1 x tr − 41 (Fab )2 + 12 (Da Ym )2 + 14 [Ym , Yn ]2 .


(10.45)

This is a Yang–Mills action coupled to massless adjoint scalars with quartic

interactions.

10.5 Two-Form Field of a String

We have seen that strings couple to various fields. These couplings can be
interpreted as interactions between charged objects and gauge fields. As an
analogy we use a charged point particle: On the one hand, it generates a Coulomb
potential around itself. On the other hand, the electromagnetic field influences its
motion. These two phenomena, reaction and back-reaction, are required for
consistent interactions. In this picture, the string carries a two-form charge which
couples to the two-form gauge field, but which also generates a field configuration
around it. In the following we will present this field configuration.

10.10
Fundamental String. Consider an infinite straight string stretched along the x0
and x1 directions with xm = 0, m = 2, . . . , D − 1.
x0 , x1

xm (10.46)
D−3
S

As a charged object, the string configuration generates a two-form potential of the

form
B = f (xm )−1 − 1 dx0 ∧ dx1 ,


(10.47)
where f (xm ) is a function of the displacement xm to the string. The interactions
with the metric G and the dilaton Φ require
ds2 = f (xm )−1 ds22 + ds2D−2 , e2Φ = f (xm )−1 . (10.48)
The function f with r2 = x22 + . . . + x2D−1 reads9
gs2 N κD−4
f =1+ . (10.49)
rD−4
One can convince oneself that these background fields B, G, Φ satisfy the
low-energy effective string equations of motion essentially because f is a harmonic
function.
However, there is an important subtlety: The background field equations are
satisfied everywhere except at r = 0 where there is a residual term of the form
δ D−2 (xm ). The latter represents a source term for the charge of the string.
• The equations of motion follow from the combination of the spacetime action
and the worldsheet coupling to the two-form
Z Z
H ∧ ∗H + B. (10.50)
D 2

The source term δ D−2 (xm ) is absorbed by the contribution of the string
worldsheet.
• The charge of a string can be measured by the Gauss law via the field ∗H. We
put a (D − 3)-dimensional sphere at fixed r to enclose a time-slice of the string
Z
Q= ∗H = N. (10.51)
D−3

The above string has N units of charge. One can show that this number must
be quantised in integers in analogy to the Dirac charge quantisation condition.
It is interpreted as the number of strings residing at the plane xm = 0.
Note: A string is the same as a 1-brane, but the above 1-brane is not a D-brane: It
originates from closed strings alone; it has nothing to do with open strings, in
particular, open strings cannot end on it. The 1-brane is the string itself, it is
therefore called the fundamental string or the fundamental 1-brane.
9
Solutions f (xm ) with more than one centre are also permissible.

10.11
Magnetic Brane. An analogous solution of the string effective equations of
motion describes a (D − 5)-brane. It uses a dual (D − 4)-form potential C defined
through
H = dB, ∗H = dC. (10.52)
It carries a magnetic charge Z
Q= H. (10.53)
3

The source is located on the (D − 5)-brane(s). The coupling of (D − 5)-branes to

C compensates the residual source terms. This object is dual to to the
fundamental string and therefore called the magnetic M(D − 5)-brane.

10.12
Introduction to String Theory Chapter 11
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

11 Superstrings
Until now, we have encountered only bosonic d.o.f. in string theory. Conversely,
matter in nature is dominantly fermionic. We therefore need to add fermions to
string theory.
This has several interesting consequences:
• Supersymmetry is inevitable.
• Critical dimension is reduced from D = 26 to D = 10.
• Stability is increased.
• The closed string tachyon is absent. We also find stable D-branes.
• There are several formulations related by dualities.

11.1 Supersymmetry
We have seen that string theory always includes spin-2 particles which have to be
gravitons. A fermionic extension of string theory will likely include spin- 23
particles. These must be gravitini which can only exist in a supergravity theory, a
supersymmetric version of gravity. Here the spacetime symmetries are extended by
supersymmetry.

Super-Poincaré Algebra. The super-Poincaré algebra is an extension of the

Poincaré algebra. The Poincaré consists of Lorentz rotations Mµν and translations
Pµ ,
[M, M ] ∼ M, [M, P ] ∼ P, [P, P ] = 0. (11.1)
The super-Poincaré algebra has additional generators, the so-called supercharges
QIm . They transform in spinor representations of the Lorentz algebra (m is a
spinor index), and they are odd generators whose Lie brackets are symmetric
rather than anti-symmetric

[M, Q] ∼ Q, [Q, P ] = 0, {QIm , QJn } ∼ δ IJ γmn

µ
Pµ . (11.2)

Furthermore, N is the rank of supersymmetry I = 1, . . . , N .

The supercharges Q relate particles of
• different spin,
• different statistics,
• otherwise equal features.
Therefore, supersymmetry can be viewed as a symmetry that relates “forces” and
“matter”.

11.1
Wess–Zumino Model. Let us discuss a simple example of a supersymmetric
field theory in D = 4 dimensions. It consists of a complex scalar field φ and a
chiral fermion ψm with Lagrangian
L = − ∂ µ φ̄ ∂µ φ − m̄mφ̄φ
− iσµnṁ ψ̄ṁ ∂ µ ψn − 21 mεmn ψm ψn + 12 m̄εṁṅ ψ̄ṁ ψ̄ṅ
− 12 ḡmφ̄2 φ − 12 g m̄φ̄φ2 − 41 gḡ φ̄2 φ2
− 21 gεmn φψm ψn + 21 ḡεṁṅ φ̄ψ̄ṁ ψ̄ṅ . (11.3)
The fields both have mass |m| and their scalar and Yukawa interactions are
governed by a complex coupling constant g. The 2 × 2 matrices σµ are a
generalisation of the Pauli matrices to chiral spinors in D = (3, 1). The 2 × 2
matrices ε are anti-symmetric with ε12 = ε12 = +1.
Evidently, this model is Poincaré-invariant. In addition it has an invariance
parametrised by a constant fermionic chiral spinor δm
δφ = εmn δm ψn ,
δ φ̄ = −εṁṅ δ¯ṁ ψ̄ṅ ,
δψm = −iεmn σµnṗ δ¯ṗ ∂ µ φ − δm (m̄φ̄ + 21 ḡ φ̄2 ),
δ ψ̄ṁ = +iεṁṅ σµpṅ δp ∂ µ φ̄ − δ¯ṁ (mφ + 21 gφ2 ). (11.4)
To show invariance under this transformation is tedious but straight-forward.

Supermultiplets. The classification of unitarity irreducible representations

(UIR) of the Poincaré algebra distinguishes particles by their mass and spin.1
UIR’s of the super-Poincaré algebra combine several UIR’s of the ordinary Poincaré
algebra into a supermultiplet. Let us discuss some features of supermultiplets.
The size and constituents of a supermultiplet depend strongly on the number of
dimensions D and the rank of supersymmetry N . This makes a general treatment
difficult, but there are some general features.
It is convenient to start with a constituent of minimum spin component under
some rotation, the bottom state of the supermultiplet. Half of the supercharges
have a corresponding negative spin component, and therefore must annihilate the
bottom state. The other half of the supercharges generate the supermultiplet. The
supermultiplet is finite because the supercharges anti-commute (up to the
momentum generator). For N supercharges in total, a minimal supermultiplet has
2N/2 constituent fields.
For instance, the Wess–Zumino model has 4 on-shell degrees of freedom: a
complex scalar φ and a chiral spinor ψ. It has the minimum amount of N = 1
supersymmetry which in D = 4 amounts to N = 4N = 4 supercharges. The fields
φ and ψ therefore form the minimal supermultiplet.
Note that in the massless case supermultiplets typically much smaller. It can be
shown that half of the supercharges generate states of zero norm. As usual, these
1
The notion of spin for massless particles is slightly different.

11.2
states are irrelevant for physics and should be projected out. The resulting
minimal supermultiplet then has only 2N/4 states.
In the massless Wess–Zumino model there are as many states as in the massive
case. However, the supersymmetry transformations for m = 0 do not transform
between (φ, ψ) and (φ̄, ψ̄). Hence the massive supermultiplet of 4 = 22 fields splits
up into two massless supermultiplets of 2 = 21 fields each.
The above considerations also lead to an upper bound on the allowable number of
supercharges in an interacting QFT: Every positive supercharge increases the
considered spin components by half a unit. Given N supercharges, the spin
component therefore increases by N/8 units in total within the supermultiplet.
Therefore the maximum total spin in a supermultiplet is at least N/16. Since the
maximum spin must not exceed 1/2 for matter fields, 1 for gauge fields and 2 for
gravity, the maximum number of allowable supercharges equals 8, 16 and 32,
respectively.2

Superspace. Supersymmetry is not merely a curiosity of particular QFT’s, it

also has a geometric meaning: It is the translational symmetry of superspace which
extends the ordinary Minkowski space coordinates xµ by N anti-commuting
coordinates θIm . Fields on superspace can be expanded in θ to yield a finite
collection of fields with various spin

F (x, θ) = F0 (x) + FmI (x)θIm + ∗θ2 + . . . + FN (x)θN . (11.5)

This allows to package a supermultiplet of particles into a single field on

superspace.

Spinors. Supersymmetry is based on spinor representations of the group

Spin(D − 1, 1) which is a two-fold cover of the Lorentz group SO(D − 1, 1). Let us
discuss spinor representations in various dimensions.
In D = 4, the default (Dirac) spinor has 4 complex components, it belongs to the
space C4 . It can be split into two chiral spinors (Weyl): C2 ⊕ C2 . Alternatively, a
reality condition (Majorana) can be imposed: Re(C4 ) = R4 ' C2 = Re(C2 ⊕ C̄2 ).
Note that in D = 4 the reality condition cannot be imposed on chiral spinors
because complex conjugation flips chirality.3
Let us now summarise the features on spinors in higher dimensions:
• The dimension of the Dirac spinor is multiplied by a factor of 2 for every step
D → D + 2.
• Chiral spinors (Weyl) exist only when D is even.
• Real spinors (Majorana) exist for D = 0, 1, 2, 3, 4 (mod 8).
• Real chiral spinors (Majorana–Weyl) exist only for D = 2 (mod 8).
2
For massive supermultiplets, the number of allowable supercharges naively is N = 4, but in
the presence of central charges, this number may increase to N = 8.
3
Hence a real spinor is equivalent to a chiral spinor.

11.3
The number of supercharges N equals the dimension of a spinor multiplied by the
rank N of supersymmetry.4 This leads to a maximum rank of supersymmetry N
in a given dimension D and a maximum dimension D for interacting
supersymmetric QFT’s:
• For D = 4 the maximum rank is N = 2, 4, 8 for matter, gauge and gravity
theories, respectively.
• In D = 10 the minimum spinor is the real chiral spinor with 16 components.
This is the dimensional bound for gauge theories.
• In D = 11 the real spinor has 32 components. This is the dimensional bound for
gravitational theories.

Super Yang–Mills Theory. Let us give a sketch of pure N = 1

supersymmetric gauge theory in D = 10 Minkowski space.5 This theory appears as
the low-energy limit for open strings. Its field contents is
• a gauge field Aµ with 8 on-shell d.o.f.,
• an adjoint real chiral spinor Ψm with 8 on-shell d.o.f..
It has a particularly simple action
Z
S ∼ d10 x tr − 14 F µν Fµν + γmn
µ
Ψ m Dµ Ψ n .


(11.6)

Supersymmetry acts as follows

δAµ ' δm γmn

µ
Ψ n, δΨ m ' (γ µν )m n δn Fµν . (11.7)

Supergravity Theories. Supergravity theories have a highly non-linear action,

hence we shall only discuss some aspects of these theories. There are four relevant
models for string theory:
• N = 1 supergravity in 11D: M-theory.
• N = (1, 1) supergravity in 10D: type IIA supergravity.
• N = (2, 0) supergravity in 10D: type IIB supergravity.
• N = (1, 0) supergravity in 10D: type I supergravity.
The former three theories have 128 + 128 d.o.f., the latter only 64 + 64. Let us list
the types of fields ([n] refers to an n-form gauge field) in these theories along with
the above gauge theory:

type gr. [4] [3] [2] [1] sc. gravitini spinors

M 1 - 1 - - - 1 -
IIA 1 - 1 1 1 1 (1,1) (1,1)
(11.8)
IIB 1 1 - 2 - 2 (2,0) (2,0)
I 1 - - 1 - 1 (1,0) (1,0)
SYM - - - - 1 - - (0,1)
4
In this calculation the number of real dimensions matters, i.e. a complex dimension counts as
two real dimensions.
5
An analogous theory exists only in D = 3, 4, 6 due to particular spinor identities.

11.4
Note that M-theory has no 2-form field and no dilaton: It cannot be the
low-energy limit of a string theory. Type IIA, IIB and I all have a 2-form field and
a dilaton: they could arise as the low-energy limit of a string theory?!

11.2 Green–Schwarz Superstring

There are two approaches to formulate a supersymmetric string theory in D = 10
dimensions. We can either make the worldsheet or the target space
supersymmetric. Both approaches turn out to be equivalent, and we shall begin
with the latter, called the Green–Schwarz superstring.

Action. We shall discuss the type II superstring where we add two fermionic
fields ΘIm , I = 1, 2, to the worldsheet theory. The fields transform as worldsheet
scalars and target space spinors. They have equal or opposite chirality for so-called
IIB and IIA string theory, respectively.
Target space is now a superspace with coordinates X µ and ΘIm . The worldsheet
theory of string theory is formulated in terms of line elements. The supersymmetric
line elements Π receive some extra contributions from the fermionic directions

Παµ = ∂α X µ + δ IJ γmn
µ
ΘIm ∂α ΘJn . (11.9)

The action then takes the convenient form

Z p
S ∼ d2 ξ − det g g αβ ηµν Παµ Πβν
Z  
Θ1 γµ dΘ1 − Θ2 γµ dΘ2 dX µ + Θ1 γµ dΘ1 Θ2 γ µ dΘ2 .


+ (11.10)

In addition to diffeomorphisms, this action has so-called kappa symmetry which is

a local worldsheet supersymmetry and which effectively removes half of the
fermionic fields. Importantly, this symmetry applies only in D = 10 dimensions of
target space!
Note: the fermions Θ have first and second class constraints. Non-linear equations
of motion. In general difficult to quantise canonically. Conformal gauge does not
resolve these difficulties.

Light Cone Gauge. It is convenient to apply light cone gauge which simplifies
the model drastically: The action becomes quadratic such that the e.o.m. are linear
Z  
S ∼ d2 ξ ∂L X ~ · ∂R X
~ + 1 Θ1 · ∂R Θ1 + 1 Θ2 · ∂L Θ2 . (11.11)
2 2

The bosonic and fermionic d.o.f. are described slightly differently. The bosonic
fields are exactly the same as for bosonic strings:
• They satisfy the second-order e.o.m. ∂L ∂R X ~ = 0.
• Both, the left and right moving d.o.f. are contained in the field X~ via ∂L,R X.
~

11.5
• The fields transform in a vector representation 8v of the transverse SO(8).
Conversely, the fermionic fields have the following features:
• They satisfy first-order equations ∂R Θ1 = 0 and ∂L Θ2 = 0.
• The left and right moving d.o.f. are contained in Θ1 and Θ2 , respectively.
• The fields transform in real chiral spinor representations 8s or 8c of the
transverse SO(8). They have equal or opposite chiralities for IIB or IIA,
respectively: 8s + 8s or 8s + 8c .

Spectrum. Next let us consider the vacuum energy, CFT central charge and
anomaly cancellations:
• For the left and right movers, there are 8 bosonic and 8 fermionic d.o.f. each
where the latter contribute with negative sign to the intercept a

aL/R = 21 8ζ(−1) − 12 8ζ(−1) = 0. (11.12)

There is no shift a for the L0 constraint. Therefore the level zero is massless.
There is no tachyon!
Importantly, the number of bosonic and fermionic d.o.f. is precisely the same.
This can only happen in particular low number of spacetime dimensions such as
D = 10.
• Before light cone gauge, there are 10 bosonic fields X µ and 32 fermionic fields
ΘIm . Due to kappa symmetry, only half of the latter to the CFT central charge

c = 10 + 32 21 = 26. (11.13)

This number is the same as for bosonic string theory and cancels precisely
against the contribution of ghost fields.
• The super-Poincaré anomaly cancels in light cone gauge.
As for the bosonic string, we can expand the closed superstring fields into Fourier
modes. This leads to the bosonic modes αn and fermionic modes βn , where the
modes n < 0, n = 0, n > 0 take the roles of creation operators, zero modes and
annihilation operators, respectively.
The zero modes have a direct impact on the structure of the string vacuum states:
• The bosonic zero mode α0 describes the centre of mass momentum: ~q.
• The existence of fermionic zero modes β0 implies the presence of a non-trivial
supermultiplet at level zero. This supermultiplet consists of 8 bosonic and 8
fermionic states. Their representations depend on the type of fermionic zero
modes:
β0 chiral (8s ) : 8v ↔ 8c vacuum → |8v + 8c , qi,
β0 anti-chiral (8c ) : 8v ↔ 8s vacuum → |8v + 8s , qi. (11.14)

The resulting string spectrum at level zero therefore depends on the types of zero
modes

11.6
• Type IIA closed strings: (8v + 8s ) × (8v + 8c )

8v × 8v + 8s × 8c = (35v + 28v + 1) + (56v + 8v ),

8v × 8s + 8v × 8c = (56s + 8c ) + (56c + 8s ). (11.15)

This is the particle spectrum of type IIA supergravity.

• Type IIB closed strings: (8v + 8c ) × (8v + 8c )

8v × 8v + 8c × 8c = (35v + 28v + 1) + (35c + 28v + 1),

8v × 8s + 8v × 8s = (56s + 8c ) + (56s + 8c ). (11.16)

This is the particle spectrum of type IIB supergravity.

• Type I closed strings arise as a Z2 projection of type IIB superstrings:
(8v + 8c ) × (8v + 8c ) mod Z2

(35v + 28v + 1) + (56s + 8c ). (11.17)

This is the particle spectrum of type I supergravity.

• Type I open strings: 8v + 8c . This is the particle spectrum of N = 1 super
Yang–Mills theory.

11.3 Ramond–Neveu–Schwarz Superstring

There is an alternative formulation for the superstring: the so-called
Ramond–Neveu–Schwarz (RNS) superstring. This formulation has manifest
worldsheet supersymmetry while spacetime supersymmetry is obscured.

Action. The action of the RNS superstring in conformal gauge consists of the
real bosonic scalar fields X µ and a pair of real fermionic spinors of either chirality
µ
ΨL,R Z
S ∼ d2 ξ ηµν 21 ∂L X µ ∂R X ν + iΨLµ ∂R ΨLν + iΨRµ ∂L ΨRν .


(11.18)

This action is manifestly supersymmetric on the worldsheet which implies that

bosons and fermions must transform equally under spacetime symmetries.
Therefore also the fermionic fields are vectors of SO(9, 1). This appears to violate
the spin statistics theorem for the spacetime theory, but this problem can be
resolved as we shall see shortly.
The bosonic fields behave as for the bosonic string. For the boundary conditions of
the fermionic fields there is an important choice to be made; they can be either
periodic or anti-periodic. This leads to two sectors of the theory which are
analogous to the various open and closed string sectors. They are represented by
two set of vacuum states which are unrelated by acting with string mode
excitations.

11.7
Ramond Sector. One sector has periodic boundary conditions for the fermionic
field
Ψ (σ + 2π) = +Ψ (σ). (11.19)
It is called the Ramond (R) sector :
• Alike the bosonic modes αnµ , the fermionic fields are expanded into integer
Fourier modes βnµ .
• Since there are fermionic zero modes β0µ , the vacuum is a supermultiplet. The
states transform in a chiral and in an anti-chiral 16-component real spinor of
Spin(9, 1).
• The intercept is zero

a = − 21 8ζ(−1) + 12 8ζ(−1) = 0, (11.20)

hence there cannot be tachyons in this sector.

All above states transform as worldsheet spinors. Therefore they should respect
fermionic statistics. However, there are states of either statistics in the Ramond
sector.
This problem is resolved by the Glizzi–Scherk–Olive (GSO) projection which
essentially restricts to states with fermionic statistics. This is a consistent
projection of the string even in the presence of interactions. Note that this leaves a
choice: One can assign the chiral vacuum states to be bosonic or fermionic. It
leads to two inequivalent Ramond sectors R+ and R−.

Neveu–Schwarz Sector. An alternative consistent choice is periodic boundary

conditions for the fermionic field

Ψ (σ + 2π) = −Ψ (σ). (11.21)

This leads to the so-called Neveu–Schwarz (NS) sector.

µ
• The fermionic fields are expanded into half-integer Fourier modes βn+1/2 .
• There are no fermionic zero modes therefore the string vacuum for this sector is
a single fermionic scalar.
• The intercept is non-zero6

a = − 21 8ζ(−1) + 21 8ζ(−1, 12 ) = 12 , (11.22)

hence the sector contains tachyons.

All the states transform as integer-spin representations of spacetime, but again
there are states of either statistics. The GSO projection for the NS sector restricts
to bosonic states, i.e. states with an odd number of fermionic generators β. The
GSO projection has the beneficial effect of removing the tachyonic level 0. Physical
states start at level 1/2 which is massless.
6
The regularised contribution of the half-integer fermionic modes is captured by a generalised
zeta function ζ(−1, 12 ) = 1/24.

11.8
Superstring Models. The fermion periodicity conditions can be selected
individually for the left and right moving fields. There are four sectors in a
consistent formulation of the closed superstring: NS-NS, NS-R, R-NS, R-R. The
NS-NS and R-R sectors provide the bosonic particles, the NS-R and R-NS sectors
the fermionic ones.
For each R sector there is a choice of GSO projection. For equal or opposite
choices for the left and right moving sectors, one obtains the type IIB and IIA
superstrings, respectively.
The type I closed superstring is obtained by a further projection. Effectively, the
R-R sector at the massless level is eliminated.
In type I open superstrings the left and right moving sectors are related, therefore
there is only one NS and one R sector.

Superconformal Algebra. Conformal symmetry combines with the manifest

supersymmetry on the worldsheet to superconformal symmetry. As usual, there
are two copies of conformal symmetry for a closed string: one for the left movers
and one for the right movers. Let us consider just one copy.
The stress-energy tensor and conformal supercurrent read:

T = ∂X · ∂X + 2i Ψ · ∂Ψ, J = Ψ · ∂X. (11.23)

The superconformal algebra is generated by bosonic generators Ln and fermionic

generators Gr . They obey the following algebra7

[Lm , Ln ] = (m − n)Lm+n + 18 cm(m2 − 1)δm+n ,

[Lm , Gr ] = ( 21 m − r)Gm+r ,
{Gr , Gs } = 2Lr+s + 12 c(r2 − 14 )δr+s . (11.24)

Here the bosonic generators Ln occupy integer modes n while the modes of the
fermionic generators depend on the sector. In the R or NS sectors, 2r is even or
odd, respectively. The corresponding superalgebras are called Ramond and
Neveu–Schwarz algebras.

Comparison. The GS and RNS approaches to superstrings yield the same

results. When both models are restricted to light cone gauge one finds that they
are related by a special feature of the group SO(8): the Dynkin diagram has a
three-fold symmetry called triality.

8s chiral spinor

vector 8v (11.25)

8c conjugate chiral spinor

7
Note that the central charge is c = D with a conventional factor of 3/2 w.r.t. the definition of
the bosonic Virasoro algebra.

11.9
It relates the vector and both chiral spinor representations. In this way one can
change the representations of the string fields without changing the spectrum.
Let us compare some characteristic features of the two approaches:
GS RNS
fermions are spinors in target space worldsheet
worldsheet supersymmetry (yes) manifest
superconformal field theory no yes (11.26)
target space supersymmetry manifest (yes)
supergravity couplings all some (NS-NS)
spacetime covariant no (yes)

In fact, a third approach exists: The pure spinor formulation introduced by

Berkovits. Here one introduces auxiliary bosonic spinor fields λ satisfying the
non-linear constraint λγ µ λ = 0. It shares several benefits of the GS and RNS
formulations.

11.4 Branes
Let us now discuss D-branes. We have learned that the ends of open strings couple
to D-branes and that the open string spectrum carries the fluctuations of D-branes.
Hence in superstring theory D-branes should receive fermionic degrees of freedom,
let us therefore inspect the spectrum of open strings coupled to a Dp-brane:
• The massless modes are described by N = 1 super Yang–Mills theory reduced to
(p + 1) dimensions.8
• The spectrum has a tower of heavy string modes.
• A scalar tachyon may be present depending on the particular situation.

Stable Dp-Branes. Some D-branes are stable others are not. The presence of
an open string tachyon indicates instability of the D-brane:
• D-branes in bosonic string theory are always unstable.
• Dp-branes for IIB superstring are stable for p odd.
• Dp-branes for IIA superstring are stable for p even.
• Since T-duality changes the dimension of D-branes by one unit, it must also may
map between type IIA and IIB theories.
Stability is related to supersymmetry. The boundary conditions break some of the
symmetry:
• Lorentz symmetry: SO(9, 1) → SO(p, 1) × SO(9 − p).
• 16 supersymmetries preserved for p odd/even in IIB/IIA.
• no supersymmetries preserved for p even/odd in IIB/IIA.
Supersymmetry removes the open string tachyon and therefore stabilises open
strings and particular D-branes.
8
This theory has a vector field, 16 fermionic spinor fields (in total) and 9 − p scalar degrees of
freedom.

11.10
Supergravity p-Branes. D-branes are non-perturbative objects, they are not
seen perturbatively due to their large mass. Stable Dp-branes have a low-energy
limit as solutions to the supergravity equations of motion, so-called p-branes.
Alike the fundamental string and magnetic brane solutions, p-branes are supported
by a (p + 1)-form, gravity and the dilaton field.
• Both type II string theories have the dilaton and a two-form field from the
NS-NS sector.
• In addition IIB and IIA strings have forms of even or odd degree, respectively,
from the R-R sector. They are relevant for the stable Dp-branes.
The stable p-branes enjoy a several useful features:
• A p-brane carries (p + 1)-form charge. The charge prevents the p-branes from
evaporating.
• The charge density is proportional to the mass density.
• 16/32 supersymmetries preserved: This is a so-called 1/2 BPS condition.
• There is a non-renormalisation theorem for 1/2 BPS objects: p-branes are
independent of the coupling strength, they are the same at
weak/intermediate/strong coupling. Therefore half BPS p-branes describe
Dp-branes exactly.

11.5 Other Superstrings

There are three further superstring theories which we shall briefly discuss.

Type-I Superstring. Consider now superstrings ending on spacetime-filling

D9-branes. The cancellation of all gravity and gauge anomalies requires:
• a gauge group of dimension 496,
• some special property of the charge lattice.
There are only two solutions: SO(32) and E8 × E8 . Here the gauge group must be
SO(32). This breaks half of the supersymmetry and yields the type I superstring.
This model is sometimes considered an independent string theory. Alternatively it
may be viewed as IIB string theory with 16 D9 branes and a spacetime-filling
orientifold plane. As such it is part of IIB string theory.

Heterotic Superstrings. We have seen that there is almost no interaction

between the left and right movers. Let us exploit this fact:
• construct the left moving sector as for the superstring: D = 10 bosonic
dimensions plus fermions.
• set up the right moving sector as for bosonic string: 26 bosonic fields which
amount to D = 10 dimensions plus 16 extra bosonic degrees of freedom.
This string is called the heterotic string. It has 16 supersymmetries from the left
moving sector.
Again, the anomaly cancellation requires a particular gauge symmetry:

11.11
• HET-O: SO(32) or
• HET-E: E8 × E8 .
The gauge group is supported by the 16 additional internal degrees of freedom.
In particular, the HET-E model is interesting because E8 contains several potential
GUT groups:

E5 = SO(10), E4 = SU(5), E3 = SU(3) × SU(2). (11.27)

11.12
Introduction to String Theory Chapter 12
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

12 Effective Field Theory

In the previous chapters, the main concepts of string theory have been introduced:
worldsheet and target space, conformal field theory, supersymmetry, string
scattering as well as the spectrum of different types of theories. In what follows,
string theory shall be connected to field theories in various dimensions.

12.1 Effective Action and Compactifications

Let us begin the current chapter by reviewing the two main methods relating
string theories to field theories: the low-energy expansion and compactification.
Both concepts have appeared in previous chapters in several contexts, so the
discussion here will be rather brief.

Effective Action. Effective actions in the context of quantum field theory allow
to access the full dynamics of the theory including loop effects in principle.
Practically, those actions are usually known up to a certain loop order. For string
theories, the concept remains similar: corrections originating in the stringy nature
of the states as well as string loop corrections are condensed into one field theory
action. Furthermore, string effective actions are available for theories in different
numbers of dimensions and thus incorporate the effects of compactification. A
further restriction could be to limit the attention to a part of the spectrum or to a
certain regime of energies: there are effective actions for the bosonic particles in a
theory as well as actions reproducing just a subset of the scattering amplitudes.
The most prominent examples of effective actions for string theories are the
low-energy effective actions which are obtained by considering the point-particle
limit α0 → 0 and thereby projecting on the massless spectrum of the string theory.
By construction, those actions do not represent the dynamics and the particle
content of the full theory: they are valid only in the regime they are designed for.
A well-studied example of an effective action is the Dirac–Born–Infeld action.

Compactification. While the principles of compactification have been discussed

before, it shall be applied in order to obtain different limits of string theories here.
The compactified theory depends on the geometry of the compactification
manifold sensitively as the fields from the compactified dimensions appear as
effective fields in the compactified theory along with modified interactions, masses
etc.. Correspondingly, choosing a suitable compactification manifold is an efficient
(and the only) way to shape the compactified effective theory. The easiest way of
removing the unwanted extra dimensions – the torus compactification – is not
sufficient to yield phenomenologically interesting models; instead we will have to

12.1
consider more complicated manifolds, in order to engineer (parts of) the models
proven valid in the experiments in our world.
In performing the compactification one has to be careful with the signature: while
everything works straightforwardly on a torus with Euclidean signature, a
compactification on Minkowskian manifolds is more difficult and leads to scalars of
negative norm or non-compact internal symmetry of the resulting theory. In order
to avoid those difficulties, we will assume to have complex-valued fields throughout
the discussion of compactification and effective actions. For a theory with
Minkowskian signature one can impose suitable reality conditions for the desired
spacetime signature later on.

Derivation of Effective Actions for String Theories. Given a full string

theory defined by its worldsheet action, it is not straightforward to immediately
write down an effective low-energy target space action for a particular set of states.
In order to obtain it, one would usually perform the following steps:
• Calculate several scattering amplitudes in the compactified version of the
original string theory for the part of the spectrum which shall be described by
the effective action.
• String theory exhibits two parameters: the inverse string tension α0 = κ2 and
the string coupling gs . In order to obtain an effective field theory, one has to
remove one of those coupling constants. This can be done by either considering
only α0 and thus limit the attention to e.g. tree-level string amplitudes or by
uniting the parameters into a new effective parameter.
• Consider the symmetry of the amplitudes and find an action in the correct
dimension exhibiting these symmetries which reproduces the scattering
amplitudes.
• The effective theory might carry only a subset of the symmetries present in the
original theory.
• In general, deriving an effective action is not an easy task. However, many
effective actions for string theories are known, at least to leading order in the
coupling constant.
Considering tree-level string amplitudes, the torus compactification to four
dimensions yield the actions of N = 4 super Yang–Mills theory (from type I open
string theory) and N = 8 supergravity (from type II closed string theory) as their
leading terms in the effective actions. For those theories, several corrections
organised in powers of α0 are known.
A huge variety of possible compactifications exist. Except the two examples above,
many other known low-energy effective actions start with well-known field theory
actions. However, the goal of finding a compactification leading to an effective
model reproducing the standard model coupled to gravity have not been successful
so far.

12.2
12.2 Open Strings
Yang–Mills field theories are the backbone of gauge theories describing our world:
several of them can be combined to yield the standard model. Supersymmetric
versions of Yang–Mills theories can be obtained from the low-energy limit of open
string theories. Although unrealistic, those supersymmetric Yang–Mills theories
serve as prototypes and share many properties with their non-supersymmetric
relatives.
Since the ends of open strings couple to D-branes, the open string spectrum carries
their fluctuations. As we are dealing with a supersymmetric theory, the D-brane is
stable and there is no tachyon in the spectrum. The massless modes of those
fluctuations are described by N = 1 super Yang–Mills theory reduced to (p + 1)
dimensions for a Dp-brane.
The N = 1 SYM theory in D = 10 has already been introduced in the previous
chapter: Z
S ∼ d10 x tr − 41 F µν Fµν + γmn
µ
Ψ m Dµ Ψ n .


(12.1)

It has a vector field, a total of 16 fermionic spinor fields and 9 − p scalar degrees of
freedom. By compactifying the N = 1 SYM theory on the torus, the number of
supercharges remains the same. Considering the available representations of
spinors in different dimensions, one will find the following Yang–Mills theories to
exist:
dimension spinor dim. theory
10 16 N = 1 SYM
6 8 N = 2 SYM
4 4 N = 4 SYM
3 2 N = 8 SYM (12.2)
Since we are interested in four dimensions for phenomenological reasons, let us
consider N = 4 SYM theory more precisely.

D = 4, N = 4 SYM Theory. The theory is the maximally supersymmetric

theory for spin-1 particles in four dimensions. It is of particular interest as a toy
model for phenomenologically more interesting theories. States are organized in
one irreducible N = 4 multiplet consisting of 16 states: 2 gluons, 8 gluinos and 6
real scalars. The ten-dimensional Lorentz symmetry splits into SO(3, 1) × SO(6),
which in turn is contained in the complete symmetry group of the theory,
PSU(2, 2|4).
While N = 4 SYM can be obtained from compactifying the N = 1 SYM theory on
a torus, one can as well compactify type I string theory on the same manifold. In
this situation, N = 4 SYM theory appears as the leading order in the low-energy
effective action accompanied by corrections of higher order in α0 = κ2 originating
in string theory. In order to discuss those corrections, let us for simplicity consider
the gluon sector of the theory only. In this sector, the effective action reads
Z  
4 1 2 02 4 03
S ∼ d x − 4 tr(F ) − α ζ2 tr(F ) + O(α ) , (12.3)

12.3
where
F 4 = Fµ ν Fν ρ Fρ σ Fσ µ + 2Fµ ν Fρ σ Fν ρ Fσ µ
− 14 Fµν Fρσ F µν F ρσ − 12 Fµν F µν Fρσ F ρσ . (12.4)
The particular linear combination of different contractions of indices for the
F 4 -term is the only combination allowed by supersymmetry. Supersymmetry is as
well responsible for the absence of the first-order correction: the supersymmetric
extension of the term α0 π tr F 3 implies sets amplitudes which are not compatible
with supersymmetric Ward-identities.

12.3 Closed Strings

Type IIA and IIB Supergravity. Let us start the discussion of effective
actions for the closed string with the two N = 2 supergravities in ten dimensions:
type IIA and type IIB. As noted in the previous chapter, Dp-branes for the
superstring are stable for odd p in type IIA theories and stable for even p in type
IIB theories. This nicely fits with the fact that T-duality changes the dimension of
D-branes by one unit.
As discussed in the last chapter, supergravity p-branes are supported by a
(p + 1)-form, gravity, antisymmetric tensor and the dilaton field. While the latter
three originate in the NS-NS sector, the (p + 1)-forms are in the R-R sector.
Having only odd p-branes in the type IIA theory, there will be even forms in this
theory exclusively. As the fermionic states of the two theories are complicated, let
us stick here with the bosonic ones and collect them for type IIA and type IIB
theory in the following table:
sector fields field strengths
type IIA NS-NS Gµν , Φ, Bµν Hµνλ
R+ -R− C1 , C3 F2 , F4
type IIB NS-NS Gµν , Φ, Bµν Hµνλ
R+ -R+ C0 , C2 , C3 F1 , F3 , F5 (12.5)

For both theories, the massless bosonic spectrum in the NS-NS sector is identical
to the one in purely bosonic string theory. Thus the action reads
√
Z
1
SNS = 2 d10 x − det G e−2Φ
2κ10
1
Hµνλ H µνλ + 4∂µ Φ ∂ µ Φ + O(α0 ) .
 
· R − 12 (12.6)
Different bosonic degrees of freedom from the R-R sector will result in different
effective actions, however. Conveniently these actions are expressed in terms of the
field strengths Fi+1 = dCi . They split into a term coupling to the metric SRR as
well as a purely topological term SCS . For type IIA theory the action reads
√
Z
1
d10 x − det G F22 + F̃42 ,


SRR = − 2
4κ10
Z
1
SCS = − 2 B ∧ F4 ∧ F4 , (12.7)
4κ10

12.4
where we set for simplicity F̃4 = F4 − C1 ∧ H3 . For type IIB, one finds
√
Z
1
d10 x − det G F12 + F̃32 + 12 F̃52 ,


SRR = − 2
4κ10
Z
1
SCS = − 2 C4 ∧ H3 ∧ F3 , (12.8)
4κ10

where F̃3 = F3 − C0 ∧ H3 and F̃5 = F5 − 21 C2 ∧ H3 + 12 B2 ∧ F3 . Deriving the

equations of motion and the Bianchi identities for the field F̃5 yields

d∗F̃5 = dF̃5 = H3 ∧ F3 , (12.9)

which is compatible with the self-duality of the five-form in ten dimensions:

∗F̃5 = F̃5 . Note, however, that this condition is not implied by the action and has
to be imposed separately.

D = 4, N = 8 Supergravity. The maximally supersymmetric theory for

particles of spin 2 in four dimensions contains one irreducible N = 8 multiplet
consisting of 256 states: 2 gravitons, 16 gravitini, 56 vector particles, 112 spin-1/2
particles and 70 real scalars. Similar to the situation in N = 4 SYM above, N = 8
supergravity is corrected by string theory effects in the effective action. In the
graviton sector, the effective action reads
√
Z
S ∼ d4 x − det G R + α03 ζ3 R4 + O(α04 ) ,


(12.10)

where R4 is a contraction of Riemann tensors dictated by supersymmetry. Again,

as for N = 4 SYM, the absence of the first-order and second-order corrections is
caused by supersymmetry. There is one peculiarity: there is another supergravity
theory, which naively does not seem to correspond to a string theory.

D = 11 Supergravity. As discussed above, there is the 32-dimensional

Majorana representation for spinors available in 11 dimensions. The corresponding
supergravity theory will thus have N = 1 supersymmetry. Considering the
representations of the little group SO(9) implied by the spinor representations, one
finds the massless spectrum to consist of two bosonic fields, the metric Gµν , a
3-form potential Aµνρ as well as a spin-3/2 fermion, the gravitino. The
supersymmetry algebra is identical to the one appearing in ten-dimensional type
IIA theory. This is not a coincidence as will be pointed out in the next chapter.

12.4 Relations Between String Amplitudes

The observables of a string theory are the scattering amplitudes. While the
calculation of those objects has been explained before, one can obtain additional
information by considering their worldsheet origin: beyond the obvious cyclicity
and reflection symmetry one can – using complex analysis – relate further

12.5
worldsheet calculations and thus derive relations between the corresponding
amplitudes. There are two distinct approaches: combining two open-string
worldsheets into a closed one will lead to the Kawai–Lewellen–Tye (KLT)
relations. Choosing different contours for the evaluation of an open-string
amplitude will lead to the monodromy relations between those. Nicely, both
relations survive the low-energy limits and thus carry over to the amplitudes in
corresponding four-dimensional field theories.

KLT Relations. The KLT relations relate scattering amplitudes in closed string
theories to the ones in open string theories. This is not a duality: the two physical
systems described by open and closed strings have a different spectrum and are
distinct.
Any vertex operator corresponding to a massless state in closed string theory can
be written as product of two vertex operators in open string theory:
open open
V closed (zi , z̄i ) = Vleft (zi ) V̄right (z̄i ) . (12.11)
While in the closed string the insertion points zi , z̄i are integrated over a
two-sphere, in the open-string case the real zi are integrated over the boundary of a
disk. In order to express the closed-string integral in terms of open-string integrals,
one has to identify and relate the contours of integration in the open-string
integrals in a way which yields a consistent closed-string expression. Deforming the
contours in the open string integral one will yield various phase factors.
After doing so, one can express the closed-string amplitudes in terms of a sum of
products of open-string amplitudes. The total permutation symmetry of the
insertion points on the sphere (and thus the total permutation symmetry of the
external legs in the closed-string integral) is ensured by a taking particular sums of
different products of open-string amplitudes with permuted legs.
KLT relations are a very convenient way to calculate amplitudes in closed string
theory and more so in their low-energy limits, supergravity. In fact: many
calculations in supergravity could not have been performed without KLT relations.
The explicit relations depend on the external states of the amplitudes. For the
example of pure tachyon amplitudes one finds
1
closed
M4,tach (s, t) ∼ sin(α0 πt/4)Aopen open
4,tach (s/4, t/4)A4,tach (t/4, u/4) . (12.12)
π
For external gluons, the four-point and five-point KLT relations read
−i
closed
M4,grav (1, 2, 3, 4) = 0
sin(α0 πs)
απ
· Aopen open
4,gluon (1, 2, 3, 4) A4,gluon (1, 2, 4, 3) ,
i
closed
M5,grav (1, 2, 3, 4, 5) = 02 2 sin(α0 πs12 ) sin(α0 πs34 )
α π
· Aopen open
5,gluon (1, 2, 3, 4, 5) A5,gluon (2, 1, 4, 3, 5)
i
+ 02 2 sin(α0 πs13 ) sin(α0 πs24 )
α π

12.6
 · Aopen open
5,gluon (1, 3, 2, 4, 5) A5,gluon (3, 1, 4, 2, 5) . (12.13)
Here Mandelstam variables are defined as sij = −(qi + qj )2 , where for the
four-point case the usual s = s12 , t = s14 and u = s13 have been used.
The above equalities are exact relations between string theory amplitudes, and so
they are valid order by order in α0 . For example, in the limit α0 → 0 KLT will
relate gluon amplitudes in N = 4 SYM to graviton amplitudes in N = 8
supergravity. Being an exact relation between open and closed string amplitudes,
the KLT relations are valid for each order in α0 individually. In particular, one can
relate the string corrections to N = 4 SYM and N = 8 supergravity: using KLT
relations one can for example show that the absence of the first-order and
second-order corrections to supergravity amplitudes is implied by the particular
form of the string corrections to the N = 4 SYM theory.
KLT relations do not only relate gluon and graviton amplitudes in the field theory
limits of open and closed string theories, but amplitudes from the full multiplets.
In particular, the tensor-decomposition of the Fock space reads
[N = 8] ←→ [N = 4]L ⊗ [N = 4]R . (12.14)

Monodromy Relations. Monodromy relations arise from deforming the

contour of integration and thus relating open-string amplitudes with different
ordering of legs (or different successions of inserting the corresponding vertex
operators on the boundary of the disk. The simplest example is again the
four-point amplitude. Here we will not specify the particle content, but rather
focus on the kinematical dependence of the amplitude. Fixing the Möbius
symmetry on the worldsheet by choosing z1 = 0, z3 = 1 and z4 = ∞, one can write
the three configurations of amplitudes which are not related by cyclicity and
reflection as
Z 0
open 0 0
A4 (2, 1, 3, 4) ∼ dz2 (−z2 )α q1 ·q2 (1 − z2 )α q2 ·q3 ,
−∞
Z 1
0 0
Aopen
4 (1, 2, 3, 4) ∼ dz2 (z2 )α q1 ·q2 (1 − z2 )α q2 ·q3 ,
Z0 ∞
open 0 0
A4 (1, 3, 2, 4) ∼ dz2 (z2 )α q1 ·q2 (z2 − 1)α q2 ·q3 . (12.15)
1

Analytically continuing the variable z2 to the complex plane one can consider
integrating the integrand of the first amplitude, Aopen 4 (2, 1, 3, 4) over a contour
closed at infinity. Assuming the poles at z2 = 0 and z2 = 1 to be outside the
integration regime, one finds immediately
Z 0
0 0
0= dz2 (−z2 )α q1 ·q2 (1 − z2 )α q2 ·q3
−∞
Z 1
0 0
+ dz2 (−z2 )α q1 ·q2 (1 − z2 )α q2 ·q3
Z0 ∞
0 0
+ dz2 (−z2 )α q1 ·q2 (1 − z2 )α q2 ·q3 , (12.16)
0

12.7
which can be easily rewritten as

Aopen
4 (2, 1, 3, 4)
Z 1
0 0
=− dz2 (−z2 )α q1 ·q2 (1 − z2 )α q2 ·q3
0
Z ∞
0 0
− dz2 (−z2 )α q1 ·q2 (1 − z2 )α q2 ·q3
0
Z 1
0
iπα q1 ·q2 0 0
= −e dz2 (z2 )α q1 ·q2 (1 − z2 )α q2 ·q3
0
Z ∞
0
iπα (q1 ·q2 +q2 ·q3 ) 0 0
−e dz2 (z2 )α q1 ·q2 (z2 − 1)α q2 ·q3
0
0 0
= −eiπα q1 ·q2 Aopen
4 (1, 2, 3, 4) − eiπα (q1 ·q2 +q2 ·q3 ) Aopen
4 (1, 3, 2, 4) . (12.17)

The above equality is the monodromy relation for the four point string amplitudes.
In the low-energy limit (α0 → 0), its real part corresponds to the
photon-decoupling identity for gauge theory amplitudes, while the imaginary part
yields the Bern–Carrasco–Johansson relations. The analysis can be performed in
the same way for amplitudes with more external legs, which leads to the general
form of the monodromy relations:
0
Aopen
n (1, 2, 3, 4, . . . , n) + eiα πs12 Aopen
n (2, 1, 3, 4, . . . , n)
0
+ eiα π(s12 +s13 ) Aopen
n (2, 3, 1, 4, . . . , n)
+ ...
0
+ eiα π(s12 +s13 +...+s1,n−1 ) Aopen
n (2, 3, 4, . . . , n − 1, 1, n) = 0, (12.18)

where again sij = −(qi + qj )2 . Employing cyclicity, reflection symmetry as well as

the monodromy relations collectively reduces the number of independent
amplitudes with n legs to (n − 3)!.

12.8
Introduction to String Theory Chapter 13
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

13 String Dualities
In the early 90’s five different string theories were defined at the perturbative level
only, while there was little understanding of the dynamical principles of the theory.
Considering the degenerate ground states after compactification (which are
parametrised by scalars/moduli), it was not clear which of the reductions could
possibly correspond to the Standard Model. Most annoyingly, there were five
different versions of string theory available while the theory had set off to be a
candidate for a unique unifying theory. The resolution to those problems appeared
with the advent of string dualities.
Two theories are called dual, if they describe the same physics using different
“languages”, that is, different fields, coupling constants and interactions. Hereby
dualities can relate fields of completely different nature. Dualities lead to the
identification of different vacua and finally allow – in combination with
weak-coupling or strong-coupling limits – to identify M-theory as the link between
the five different string theories, thus being closer to the unified description aimed
at.
While T-duality will be the main working example here, there are other forms of
dualities, in particular S-duality. In contrast to T-duality, which is a weak-weak
duality relating the weak-coupling regimes of two theories to each other, S-duality
is more interesting from the point of understanding the complete dynamics of a
theory: It maps the non-perturbative strong-coupling sector of one theory to the
perturbative weak-coupling sector of another theory.
The goal of this chapter is to explain the arrows in the following figure:

M-theory

type type heterotic heterotic

IIB IIA E8 × E8 SO(32)
type I (13.1)

type II heterotic / type I

Here, dashed arrows denote S-duality, dotted arrows mark T-duality and solid
arrows are compactifications on a suitable interval.

13.1 T-Duality
Within the context of the string theories explored so far there are two famous
examples of T-duality: type IIA and IIB string theory can be shown to describe the
same physics if one dimension is compactified on a circle. This is the example to be

13.1
explored below. Furthermore, the two heterotic string theories HET-O and HET-E
with gauge groups SO(32) and E8 × E8 , respectively, are related by T-duality.

T-Duality between IIA and IIB String Theory. Consider T-duality

between the background fields in the NS-NS and the R-R sector: G, B, Φ, F . The
relations between the fields G and B from the NS-NS sector and their duals G̃ and
B̃ can be obtained by starting from the non-linear sigma-model in conformal gauge
Z
1
d2 ξ η αβ Gµν − εαβ Bµν ∂α X µ ∂β X ν .


S=− 2
(13.2)
4πκ
In order to derive the so-called Buscher rules for the fields G and B, the
ten-dimensional metric and antisymmetric tensor need to be split in Kaluza–Klein
form for a compact direction X 9 . After doing so and requiring the background
fields G and B to be independent of the compact dimension, one finds that the
fields are related by:
Gi9 Gj9 − Bi9 Bj9 B9i 1
G̃ij = Gij − , G̃9i = − , G̃99 = ,
G99 G99 G99
Gi9 Bj9 − Gj9 Bi9 G9i
B̃ij = Bij + , B̃9i = − . (13.3)
G99 G99
The dilaton transformation cannot be derived that easily: it can be inferred from
demanding conformal invariance of the full (non-gauged) non-linear sigma-model
action
Z
1 2
h p
− det g g αβ Gµν − εαβ Bµν ∂α X µ ∂β X ν


S =− 2
d ξ
4πκ
p i
+ κ2 − det g Φ(X) R[g] . (13.4)

T-duality for the fields from the R-R sector can be derived in a similar manner. It
is easy to check that the number of components for the R-R fields Fi in type IIA
string theory equals the number of free components in the type IIB theory. Indeed
one finds
     
10 10 10
+ + = 1 + 45 + 210 = 256,
0 2 4
| {z } | {z } | {z }
F F F
 0  2  4
10 10 1 10
+ + = 10 + 120 + 126 = 256. (13.5)
1 3 2 5
| {z } | {z } | {z }
F1 F3 F5

where the R-R scalar F0 as well as a factor of 1/2 taking care of the self duality of
F5 have been taken into account. In open string theory, strings are tied to
D-branes by their boundary conditions. As T-duality interchanges Dirichlet and
Neumann boundary conditions, it is consistent to find D-branes of even dimensions
in type IIA string theory, while type IIB contains D-branes of odd dimension.

13.2
Accordingly, T-duality has to relate even and odd field strengths. In a convenient
gauge, the T-duality rules for the R-R fields read

Fn, 9i2 ...in = −F(n−1), i2 ...in + (n − 1)G−1 99 G9[i2 F(n−1), 9i3 ...in ] ,
Fn, i1 i2 ...in = −F(n+1), 9i1 ...in + nB9[i1 Fn, 9i2 ...in ] . (13.6)

Both transformations are not a symmetry of the action.

T-duality toggles the chirality of the fermions: while type IIA is non-chiral, type
IIB is a chiral theory. Upon reduction on a circle to nine dimensions, the two
theories coincide precisely, on the level of the action. Thus, the IIA and IIB
theories in ten non-compact dimensions are different limits of the same theory in
nine non-compact dimensions: one time with an infinitely large tenth dimension,
one time with a small tenth dimension. In particular one finds
RA p 1 κ
= G99 = p = . (13.7)
κ G̃99 RB

T-Duality for Toroidal Compactifications. Instead of compactifying just

one dimension, let us now consider a compactification of several dimension on a
d-dimensional torus T d . In each compact dimension with index
k ∈ {D − d, . . . , D − 1} one imposes

X k = X k + 2πwk , (13.8)

where wk is the winding number for the compact dimension k. Different

geometries are conveniently described by a compactification lattice Λd such that

T d = R d Λd .

(13.9)

The information on the geometry of the torus is contained in the metric of the
compact dimensions. The basis vectors ei of the compactification lattice Λd and
those for the dual lattice Λ∗d , e∗i , can be related to the metric in the following way:
d d d
X X X 1
eak eal = 2πGkl , eak e∗l l
a = δk , e∗k ∗l
a ea = (G−1 )kl . (13.10)
a=1 a=1 a=1
2π

The quantisation of the momenta and the possibility for the string to wrap around
each of the compact dimensions yield expressions for the left and right moving
centre-of-mass momenta. In terms of the basis vectors of the dual lattice they can
be expressed as:

pRa = na + wl (B − G)kl e∗k

 
a ,
 l
 ∗k
pLa = na + w (B + G)kl ea , (13.11)

where nk and wk are the momentum eigenvalues and the winding numbers in the
direction k, respectively. One can show that the left and right moving

13.3
contributions of the momentum in the compact dimensions, pR,a and pL,b , have to
satisfy the physicality condition

p2L − p2R = 2wk nk ∈ 2Z . (13.12)

Thus one obtains an even and self-dual Lorentzian lattice Γ (d,d) , where
“Lorentzian” corresponds to the different signs of the metric in the two
d-dimensional parts of the group.
The Lorentzian lattice Γ (d,d) should not be confused with the compactification
lattice Λd although the same information is contained in both. The
compactification lattice Λd describes the geometry, where each of the points in the
Lorentzian lattice Γ additionally contains information about the momenta
corresponding to particular winding numbers around each of the compact
dimensions.
All even and self-dual Lorentzian lattices are related by O(d, d, R) rotations.
Correspondingly, the group O(d, d, R) exhausts the full moduli space of toroidal
backgrounds: all compactification configurations are related by those
transformations.
However, not every element of the group O(d, d, R) leaves the spectrum of the
theory and all correlators invariant. The subgroup of O(d, d, R) which leaves the
spectrum invariant is O(d, d, Z). The complete physical group of symmetries of the
considered d-dimensional toroidal compactifications is the group O(d, d, Z)
combined with worldsheet parity σ → −σ. Parity, which changes B → −B, is not
included in O(d, d, Z) because it corresponds to the interchange of pL and pR ,
which flips the sign of the Lorentzian norm p2R − p2L .

13.2 Strong/Weak Coupling Duality: S-Duality

S-duality is the analogue of electric-magnetic duality. Its name originates from the
fact that it relates the strong-coupling and weak-coupling limits of theories.
S-duality provides a way to access theories beyond perturbation theory. Consider a
theory at small values of the coupling gs : There are electrically charged elementary
states which can be handled by perturbation theory. Likewise there are
magnetically charged solitonic states, which are very massive and strongly coupled.
For these heavy states, perturbation theory is not a good description.
Montonen and Olive proposed that for gs → ∞ their roles might be reversed: they
conjectured that one can reformulate a theory in terms of dual fields in such a way
that the weakly coupled electrically charged states would turn into strongly
coupled magnetic ones and vice versa.
In order to get a first impression of S-duality, let us consider the effective actions
of the heterotic and the type I theory in ten dimensions. They read (in Einstein
frame GE,µν = e−2Φ Gµν ):
Z p
SI ∼ d10 x − det GE

13.4
 h i
RE + 4(∂Φ)2 − 1 +4Φ 2
12
e H + 14 e+2Φ F 2 ,
Z p
SHET-O ∼ d10 x − det GE
h i
1 −4Φ 2
RE + 4(∂Φ)2 − 12
e H + 14 e−2Φ F 2 , (13.13)

where the subscript E refers to the quantity in the Einstein frame and the field
strengths F are associated to the gauge group SO(32) in either theory. Thus, the
only difference between the two actions is the sign of the dilaton: a transformation
Φ → −Φ relates the two actions. The dilaton is related to the string coupling gs
and thus this is an obvious hint towards a weak-strong duality. Considering the
spectrum of perturbative and non-perturbative solutions, one can indeed show that
S-duality relates the full theories.
Another example is the S-duality of type IIB string theory. The SL(2, R)
invariance of the action is manifest, while S-duality mixes (perturbative) states
from the NS-NS sector with (non-perturbative) states from the R-R sector and
vice versa. The fundamental string in type IIB gets mapped to the D1-brane
solution, while the solitonic five-brane is mapped onto the D5-brane.

S-Duality of the Heterotic String in D = 4. As this will become

phenomenologically interesting, let us consider another example of a string
compactification exhibiting S-duality. Compactifying the heterotic string theory on
a 6-torus, one obtains N = 4 supergravity coupled to N = 4 Yang–Mills theory in
four dimensions. The T-duality group is O(6, 22, Z). In addition there is a
symmetry of the equations of motion: one can show that they are invariant under
a SL(2, R) transformation. Quantum effects, however, break the SL(2, R) to
SL(2, Z). It is not clear yet, whether the SL(2, Z) is a (non-perturbative)
symmetry of the full action.

U-Duality: Combination of S- and T-Duality. Some string theories will

exhibit T-duality and S-duality in their low-energy effective actions simultaneously.
The prime example are the compactifications of the ten-dimensional IIB
supergravity. Each of those compactifications exhibits at least one SL(2, Z) as well
as O(d, d, Z) as subgroups of their complete symmetry groups. It is conjectured
that the maximal integer subgroup of those two groups is a symmetry of the full
string theory, in which case it is referred to as U-duality. The symmetry can be
best explored on solutions to type IIA string theory; however, this is rather
involved.

13.3 Strong Coupling Limits

Type IIA String Theory. In the strong coupling limit – which corresponds to
a very large value of the dilaton – the type IIA string theory reveals its origin: the
spectrum can be derived from an eleven-dimensional theory. The obvious candidate

13.5
for this theory in the field theory limit is supergravity in eleven dimensions. Even
more, one can show that type IIA perturbation theory is the expansion around the
zero-radius limit of one dimension of D = 11 supergravity. Accordingly, the
non-perturbative part of the spectrum of type IIA string theory contains all kinds
of Kaluza–Klein modes originating from wrapping the eleven-dimensional solutions
around the compact dimension. The resulting modes can be identified with type
IIA D-branes. In fact, the whole spectrum of fundamental objects in type IIA
string theory can be given an eleven-dimensional interpretation.

Heterotic E8 × E8 Theory. The strong coupling limit of the effective action of

heterotic string theory with gauge group E8 × E8 is again eleven-dimensional
supergravity. In order to get from this supergravity back to the effective action of
the heterotic string theory, however, one will have to compactify on an interval
with length L or – equivalently – on a sector of a circle S 1 /Z2 . Correspondingly,
the eleven-dimensional spacetime then consists of two nine-dimensional
hyperplanes, separated by an interval of length L. Heterotic strings couple to those
hyperplanes with gauge fields of E8 . If the effective actions of type IIA string
theory as well as the heterotic E8 × E8 theory are certain weak-coupling limits of
eleven-dimensional supergravity, it is an obvious question, what the
strong-coupling-limit of type IIA string theory and the heterotic theory
corresponds to?

13.4 M-Theory
The concept of M-theory was suggested by Edward Witten in 1995 and initiated
what is nowadays known as the “second superstring revolution”. While Witten
chose the letter “M” in M-theory for either “membrane”, “magic” or “mysterious”,
everyone picks his or her own interpretation. M-theory is supposed to not be a
string theory, but rather a non-perturbative theory of fundamental objects, whose
low-energy limit is eleven-dimensional supergravity. The type I, type II and
heterotic string theories can be thought of as different perturbative expansions at
several points of the moduli space of M-theory.
A complete description of the dynamics of M-theory is not known. The best one
can currently do is to formulate the low-energy dynamics of the theory in terms of
eleven-dimensional supergravity interacting with two-dimensional and
five-dimensional membranes. M-theory and its dynamics are still a field of active
research.

13.6
Introduction to String Theory Chapter 14
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

14 String Theory and the Standard Model

In this chapter, some of the attempts of connecting string theory to the standard
model will be discussed. In particular we will discuss, what can be learnt from
string theory and what the current hopes and expectations are. In what sense does
string theory answer the questions of quantum gravity?

14.1 The Real World

Standard Model and Gravity. Making connection to the real world first and
foremost refers to reproducing the gauge group of the standard model along with
the Higgs mechanism. The standard model, comprising three generations of quarks
and leptons as well as the gauge bosons for the electromagnetic, the weak and the
strong interactions and the Higgs particle is a non-abelian gauge theory with gauge
group SU(3) × SU(2) × U(1). So far, the standard model has passed all
experimental tests performed. The standard model, however, appears to be rather
arbitrary: there are around twenty constants which have to be adjusted to very
high precision in order to reproduce the experimental findings. The origin of those
constants remains unclear: it would be desirable, to have a theory not only
predicting the general mechanisms underlying the standard model but as well the
coupling strengths and masses of and between the constituents.
In terms of a gravitational theory, the state of the art is described very quickly:
starting from Einstein gravity, the most successful model based on a
Friedman–Robertson–Walker solution is the Λ-cold-dark-matter model. It is
capable of describing the expansion of the universe in the way we observe it with
the price of introducing dark matter and dark energy. So far the search for the
dark matter has not been successful: no particle from the standard model does
satisfy the bounds on mass (and thus coupling to gravity) as well as the other
constraints originating in the observed interactions with visible matter. In
addition, it is clear that the perturbative expansion of Einstein gravity is not
sufficient to describe the physics in highly curved spaces such as they appear
shortly after the big bang. In order to predict the processes in those regimes, a
renormalisable theory for quantum gravity is unavoidable. Up to date, no
renormalisable quantum field theory for gravity is known.

Supersymmetry. So far, supersymmetry has not been observed and is not part
of the standard model. From the string theory perspective, however,
supersymmetry is desirable in order to obtain a theory free of tachyons with stable
D-branes, which in addition allows to maintain the ratio between the electroweak
scale and the Planck mass. On the other hand, one can show that supersymmetric
theories with N > 1 do not allow for the chiral spectrum we observe. Balancing

14.1
the two requirements leads to the goal of a four-dimensional theory with N = 1
supersymmetry.
This theory should obviously contain the standard model. The minimal solution
one can have is the so called minimal supersymmetric standard model. It contains
all the particles of the standard model with the addition of a Higgs doublet. Each
of the particles will be assigned a superpartner. All gauge interactions are fixed by
the non-supersymmetric part of the standard model. All other (non-gauge)
interactions are however not constrained, which leads in the smallest version to a
model exhibiting even up to about a hundred constants to fix. Another problem is
that the supersymmetric partners of the standard model particles will have the
same masses as the original particles. If this was the case, one would have seen the
superpartners already in collider experiments. So one will have to find a
mechanism spreading the masses between superpartners.
Aiming at a theory which contains gravity as well as gauge interactions, the
obvious strategy is to start from N = (1, 0) heterotic string theories in ten
dimensions because these theories readily contain gauge groups which appear to be
large enough to accommodate the standard model gauge group. The second
ingredient allowing to shape the effective four-dimensional theory is a suitable
compactification manifold: the structure of this manifold will – among many other
things – determine the amount of supersymmetry present.
In practice, there are two ways to proceed: one can either construct a conformal
field theory with suitable boundary conditions giving rise to a N = 1 theory in
four dimensions. This leads to the orbifold compactifications. The other way is to
start with the effective supergravity action derived from the heterotic theory in ten
dimension and then compactify on a suitable manifold: this will lead to
Calabi–Yau compactifications.

14.2 Geometry of Toroidal Manifolds and Orbifolds

Compactifying a theory on a torus does not break supersymmetry. In order to
produce the desired N = 1 theory in four dimensions, one needs to think about a
more sophisticated compactification manifold. One of the simplest generalisation
of toric manifolds are orbifolds.

Orbifolds. An orbifold is a generalisation of a manifold: it is a quotient space of

a Euclidean space by a finite group. Orbifolds will have orbifold fixed points, which
are the points invariant under the identification. In the vicinity of those singular
points, a quantum field (or string) theory defined on the orbifold will become
singular itself, which will effectively reduce the number of states in the theory. In
other words, respecting the symmetry group of the orbifold puts constraints on the
theory defined on it. Simultaneously, at the singular points one can add new states
to the theory: those go under the name of twisted sectors and they render a string
theory completely smooth when combined with the regular parts.
In order to get used to the concept of an orbifold, let us start with an easy

14.2
example. Take the real line R, define a lattice

Λ = aZ (14.1)

and identify points via

x'x+l for all x ∈ R , l ∈ Λ . (14.2)

The interval 0 ≤ x < a is called the fundamental domain, while R is the covering
space of the torus T . Now one defines a parity operation

P x = −x for all x ∈ R (14.3)

and identifies
x ' Px for all x ∈ R . (14.4)
Noticing that P 2 = 1, P is a realisation of the cyclic group Z2 . The fixed points
under the combined lattice and parity identifications are x = 0 and x = 1/2. The
resulting space is the orbifold T /Z2 .
In order to get from ten dimensions to four, one needs to choose a manifold of the
form
O6 = T 6 /G, (14.5)
where, for simplicity, G is assumed to be a finite abelian point group. In the
following we will work with the example G = Z3 . Conveniently, a point on this
orbifold can be labelled by complex coordinates zi with i = 1, 2, 3, which
corresponds to splitting the torus into T 6 = T12 × T22 × T32 . An orbifold with this
property is called factorisable.
In complete analogy to the physicality conditions for the periodicity conditions for
the compactification of one dimension in the context of T-duality, the periodicity
conditions on the orbifold read

zi ∼ zi + 2πRi and zi ∼ zi + 2πRi ρi , (14.6)

where Ri are the radii of the tori and ρi are the corresponding complex structure
moduli (which are the equivalent of the quantity τ in the discussion of the
fundamental domain of the worldsheet).
In terms of the coordinates zi one can define the orbifold action Θ, which reads

Θ(zi ) = exp(2πiφi )zi , (14.7)

where the phases φi have to be integral multiples of 1/3 in order to yield Θ3 = 1.

The orbifold action needs to be compatible with the torus lattice, which in turn is
defined by the complex structure moduli ρi . These conditions are pretty restrictive
and allow to fix the phases φ uniquely to

φ = 13 (1, 1, −2) . (14.8)

14.3
Breaking of Supersymmetry. Let us now see, how the manifold T 6 /Z3
described above actually will break the supersymmetry in the original theory in a
way that the four-dimensional effective theory will have N = 1 supersymmetry. In
ten dimensions, the supercharge Q is a 16-component spinor,Qwhich can be
represented by a state |s−1 , s0 , s1 , s2 , s3 i with s = ±1/2 and 3i=−1 si = 1/2. The
orbifold action Θ acts as
 3
X 
Θ|s−1 , s0 , s1 , s2 , s3 i = exp 2πi φi si |s−1 , s0 , s1 , s2 , s3 i
i=1
!
= |s−1 , s0 , s1 , s2 , s3 i , (14.9)

and needs to leave the spinor invariant. For the values of φi fixed by the
periodicity conditions on the orbifold, there is exactly one solution to the above
equation (α = ±1): 
s−1 , s0 , 1 α, 1 α, 1 α .

2 2 2
(14.10)
Counting the number of independent spinor components of the above form reveals
that the orbifold geometry indeed singles out four of the sixteen spinors in ten
dimensions, which effectively leads to N = 1 supersymmetry in four dimensions.
One can construct a worldsheet conformal field theory with appropriate non-trivial
boundary conditions, which will lead to this orbifold compactification in target
space. From the worldsheet perspective this is still a free conformal field theory
which can be solved exactly. This means in particular that one can derive the
complete partition function and demand modular invariance, which will lead to
additional constraints on the target space theory. The calculation is rather
involved, but easy enough to allow for a complete classification of all possible
orbifold conformal field theories originating from the orbifold T 6 /Z3 starting from
the heterotic string with gauge group SO(32).
The gauge groups of the resulting effective four-dimensional theories are very large
and appear to be rather arbitrary. This unsatisfactory situation can be improved
by introducing Wilson lines, which can be thought of as constant gauge fields in
the string background. The introduction of these Wilson lines allows to equip the
twisted sectors of different orbifold fixed points with different gauge groups.
Correspondingly, the boundary conditions for the conformal field theory change.
For the T 6 /Z3 -orbifold, one can add up to three Wilson lines.
There exists a T 6 /Z3 -model obtained from a heterotic string theory with gauge
group E8 × E8 with the field content of the minimal supersymmetric standard
model. However, in addition to the desired particles, there are many vector-like
exotic particles, which do not decouple completely and thus yield additional
unwanted states.

14.3 Calabi–Yau Compactification of D = 10 Supergravity

Let us just briefly comment on the second option pointed out above: one could as
well start with the effective action of ten-dimensional supergravity coupled to

14.4
ten-dimensional SYM theory and compactify this on a suitable manifold. As
opposed to the orbifolds in the previous subsection, which are singular spaces, here
one will use non-singular (or smooth) spaces.
Compactifying on a non-singular space, one still wants to obtain a N = 1
supersymmetric model in four dimensions with three generations of leptons and
quarks. The translation of those properties into constraints on the compactification
manifold leads to a very special type of manifolds: the Calabi–Yau manifolds.
While numerous Calabi–Yau compactifications have been studied, the best known
is the so-called standard embedding: it starts from the ten-dimensional effective
action of the heterotic E8 × E8 string. After compactifying on a suitable
Calabi–Yau manifold one will obtain a four-dimensional model with gauge group
E6 × E8 . For some ranges of parameters, this model will yield the minimal
supersymmetric standard model.

14.4 String Theory as a Phenomenological Model

The investigation of string compactifications is still an active field of research.
However, finding a compactification yielding an effective model which is just the
standard model has not been successful so far. If one allows for the minimal
supersymmetric extension of the standard model, string theory can produce
something similar, however, usually there are additional states which can not be
consistently decoupled.
In terms of quantum gravity, string theory provides a structurally sound model for
a theory of quantum gravity. Practically, however, experiments which would be
capable of confirming the validity of the string theory predictions beyond the
classical limit are insurmountable.
Overall, string theory appears to be a promising concept. However, the amount of
engineering needed to relate string theories to realistic quantum field theories
seems rather unnatural.

14.5
Introduction to String Theory Chapter 15
ETH Zürich, HS13 Prof. N. Beisert, Dr. J. Brödel
22. 12. 2013

15 AdS/CFT Correspondence
The AdS/CFT correspondence is the (conjectured) exact duality between a string
theory and a CFT in 2 ≤ D < 10, commonly a gauge theory:
• This is just remarkable!
• It provides a precise formulation of a string/gauge duality.
• It is a holographic duality in the sense that it relates theories in a different
number of spacetime dimensions.
• There is a multitude of pairs of models related by the AdS/CFT correspondence.

15.1 Stack of D3-Branes

Let us motivate the main example of the AdS/CFT correspondence between IIB
strings on AdS5 × S 5 and N = 4 super Yang–Mills (SYM) theory.

3-Brane Geometry. Consider the 3-brane solution of IIB supergravity

ds2 = h−1/2 dx2 + h1/2 dy 2 ,

H5 = h−2 dh ∧ dx∧4 + h−2 ∗(dh ∧ dx∧4 ). (15.1)

Here we have 3 + 1 coordinates x along the brane and 6 extra coordinates y for the
embedding space. The harmonic function h(y) = 1 + αN/|y|4 is spherically
symmetric around the brane.
This geometry is a background for IIB string theory with a stack of N D3-branes.
The low-energy brane dynamics is therefore described by U(N ) N = 4 SYM.
Now approach the brane at y = 0 or alternatively take the limit N → ∞:
• The harmonic function limits to h(y) = αN/|y|4 .
• The background becomes AdS5 × S 5 with 5-form flux.
• The sphere S 5 is given as submanifolds at constant |y| and x. The
(4 + 1)-dimensional AdS5 spacetime is combined from the x coordinates and the
distance |y|.

es
r an
D3-b
d S AdS5 × S 5
(15.2)
∂A 1)
(3,
1+5

15.1
AdS/CFT Correspondence. Claims: (Maldacena)
• The 3-brane is at the boundary of the AdS5 space.
• The low-energy string modes associated to the brane decouple from the rest.
• The boundary physics is described exactly by U(N ) N = 4 SYM.
• An open string on the boundary can probe the bulk AdS5 × S 5 string theory.

dS (15.3)
∂A

• With a suitable dictionary, there is a precise matching of all observables in both

models.
• There is a precise map of the coupling constants
1/2
(κ/R, gs ) = (gYM N 1/4 , gYM
2
). (15.4)

15.2 Anti-de Sitter Geometry

Let us briefly discuss the geometry of anti-de Sitter spacetime AdSd :
• It has constant scalar curvature but no tensorial curvature. It is a symmetric
space, all points are equivalent.
• It is analogous to spheres and hyperbolic spaces as well as the de Sitter space
according to the following table:

curvature + −
Euclidean S H (15.5)
Minkowski dS AdS

• The isometry group is SO(d − 1, 2). This is the same as the conformal group in
d − 1 dimensions.
Globally, it has the topology of a solid cylinder R × Dd−1 .

AdS
t (15.6)
∂AdS

The boundary is the cylinder surface R × S d−2 .

There are some interesting facts about the geodesics in connection to the bulk and
boundary:
• Time-like geodesics never reach the boundary.
• Space-like geodesics reach the boundary at infinite distance.

15.2
• Light-like geodesics reach the boundary in finite time. In this sense, the bulk
and the boundary can interact via massless fields.

AdS ∂AdS (15.7)

15.3 N = 4 Super Yang–Mills

There is a unique maximally supersymmetric gauge theory in four dimensions:
N = 4 super Yang–Mills theory. It is the dimensional reduction of D = 10, N = 1
SYM to D = 4. It contains the following fields:
• a gauge field,
• 4 adjoint chiral (or real) fermions,
• 6 adjoint real scalars.
It has a couple of remarkable properties:
• There is no running coupling, β = 0.
• It has exact 4D superconformal symmetry; 4D (S)CFT.
• ...

15.4 Tests
Evidently, we want to verify AdS/CFT correspondence. There are several useful
predictions to be tested:
• The string spectrum matches with the spectrum of local operators.
• String and gauge correlation functions match.
A major problem in performing such tests is that the AdS/CFT correspondence is
a strong/weak duality:
• Weakly coupled strings is strongly coupled gauge theory.
• Weakly coupled gauge theory is strongly coupled strings.

X rt
.
pe ring
st
pert. (15.8)
gauge

2
gYM N
We have good means to compute observables at weak coupling, but there is no
overlap between the weakly coupled regimes in both theories.
We can, however, test BPS quantities which are protected (independent of the
coupling):

15.3
• Supergravity modes agree with BPS operators.
• Supergravity correlators match with BPS correlators.

(15.9)
AdS

∂AdS

What about other quantities?

• String and gauge theory appear to be integrable at large N .
• Integrability is a hidden symmetry which put strong constraints on the
dynamics.
• One can compute observables efficiently even at finite coupling.
• This leads to a precise agreement in all tests that have been performed.
Other tests have been performed, for example the matching of circular Wilson loop
expectation values and the area of a string worldsheet ending on this circle.

15.4