April 2, 2015

PHY2406S String Theory (2014-15)

Department of Physics, University of Toronto
Instructor: Prof. Amanda W. Peet
Notes typed with help from Ian T. Jardine
N.B.: material draws heavily on BLT, BBS, Pol. texts.

SYLLABUS

Jan.09 Motivations for string theory. Running of couplings. Anomaly cancellation and gauge-
gravity unification. Relativistic point particle theory.
Jan.16 Worldlines vs worldsheets. Symmetries in various D and the role of BPS states. Clas-
sical bosonic string theory. Nambu-Goto and Polyakov worldsheet actions. Conformal
reparametrizations, Weyl invariance, and conformal gauge.
Jan.23 Closed bosonic string oscillator expansions and the Virasoro algebra. Open bosonic
string oscillator expansions, Virasoro, and D-branes. Quantum bosonic string theory.
Canonical quantization and negative norm states.
Jan.30 Critical dimension and mass formula for bosonic strings. Light cone quantization. Path
integral quantization and Fadeev-Popov (b, c) ghosts.
Feb.06 CFT basics. Radial quantization. The operator product expansion. Action of Virasoro
generators on states: highest weights and descendants. Correlation functions.
Feb.13 Properties of (b, c) ghosts. BRST invariance and unitarity. Applying BRST to the
bosonic string. The Virasoro-Shapiro amplitude.
Feb.23 Spinors in diverse dimensions. Green-Schwarz formalism. Neveu-Schwarz-Ramond
formalism. ZPEs, physical state conditions and the GSO projection. SCFT, (β, γ)
ghosts, and BRST.
Feb.27 KK reduction in QFT, T-duality for closed bosonic strings, T-duality for closed open
strings, D-branes, unit conventions and dimensional reduction, spacetime fields exerted
by BPS object in string/M theory, superstring dualities and the deerskin diagram.
Mar.06 The σ model and its beta-functions as the equations of motion for spacetime fields.
Mar.20 Compactification (eg Calabi-Yau manifolds, orientifold planes, orbifolds); inflation, the
string theory Landscape and the field theory Swampland.
Mar.27 AdS/CFT: holography, and applications to condensed matter and the quark-gluon
plasma.
Apr.03 Black hole entropy and the information paradox, black rings, fuzzballs, and firewalls.
Contents
1 Motivations for string theory 4
1.1 Running of couplings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Anomaly cancellation and gauge/gravity unification . . . . . . . . . . . . . . 5
1.3 Relativistic point particle theory . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Worldlines vs worldsheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Symmetries in various D and the role of BPS states . . . . . . . . . . . . . . 9

2 Classical Bosonic String Theory 11

2.1 Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Nambu-Goto and Polyakov actions . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Conformal reparametrizations, Weyl invariance, and conformal gauge . . . . 14
2.4 Closed string oscillator expansions and the Virasoro algebra . . . . . . . . . 16
2.5 Open string oscillator expansions, Virasoro, and D-branes . . . . . . . . . . . 18

3 Quantum Bosonic String Theory 21

3.1 Canonical quantization and negative norm states . . . . . . . . . . . . . . . 21
3.2 Light cone quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Path integral quantization and Fadeev-Popov ghosts . . . . . . . . . . . . . . 27

4 Conformal Field Theory 32

4.1 Radial quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.2 The operator product expansion . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3 Action of Virasoro generators on states: highest weights and descendants . . 38
4.4 Correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

5 Ghosts, BRST invariance, and scattering amplitudes 42

5.1 Properties of (b,c) ghosts and the critical dimension . . . . . . . . . . . . . . 42
5.2 BRST invariance and unitarity . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.3 Applying BRST to the bosonic string . . . . . . . . . . . . . . . . . . . . . . 45
5.4 The Virasoro-Shapiro amplitude . . . . . . . . . . . . . . . . . . . . . . . . . 50

6 Superstrings 53
6.1 Spinors in diverse dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.2 Green-Schwarz formalism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.3 Neveu-Schwarz-Ramond formalism . . . . . . . . . . . . . . . . . . . . . . . 56
6.4 ZPEs, physical state conditions and the GSO projection . . . . . . . . . . . 57
6.5 SCFT, (β, γ) ghosts, and BRST . . . . . . . . . . . . . . . . . . . . . . . . . 58

7 T-duality, D-branes and superstring duality 64

7.1 KK reduction in QFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.2 T-duality for closed bosonic strings . . . . . . . . . . . . . . . . . . . . . . . 65
7.3 T-duality for open bosonic strings . . . . . . . . . . . . . . . . . . . . . . . . 67
7.4 Unit conventions and dimensional reduction . . . . . . . . . . . . . . . . . . 69
7.5 Spacetime fields exerted by BPS objects in string/M theory . . . . . . . . . 72

1
 7.6 Superstring dualities and the deerskin diagram . . . . . . . . . . . . . . . . . 73

8 Sigma Models and String Theory 75

8.1 Sigma Model couplings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
8.2 Background field method and Riemann normal coordinates . . . . . . . . . . 77
8.3 Graviton beta function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
8.4 Kalb-Ramond field beta function . . . . . . . . . . . . . . . . . . . . . . . . 81
8.5 Dilaton beta function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
8.6 Weyl anomaly at one loop . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
8.7 Nailing the dilaton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

9 Compactification 93
9.1 Hiding extra dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
9.2 SUSY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
9.3 Brane world models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
9.4 Heterotic string theory on CY3 s . . . . . . . . . . . . . . . . . . . . . . . . . 94
9.5 Calabi-Yaus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
9.6 Geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
9.7 Hodge numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
9.8 Mirror symmetry and the conifold . . . . . . . . . . . . . . . . . . . . . . . . 96
9.9 Dp-brane probes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
9.10 Example: 1 Dp probing N Dps . . . . . . . . . . . . . . . . . . . . . . . . . . 97
9.11 Forces between D-branes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
9.12 Open string BCFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
9.13 What is an orbifold? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
9.14 Simple compact and non-compact orbifolds . . . . . . . . . . . . . . . . . . . 99
9.15 Spectra of states for orbifolds . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.16 What is an orientifold? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.17 What are Op-planes good for? . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.18 Flux compactifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9.19 Flux-ology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
9.20 Inflation in string theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
9.21 KKLT and the Landscape . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
9.22 Ur potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

10 AdS/CFT 103
10.1 Origin of AdS/CFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
10.2 Maldacena’s decoupling limit . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
10.3 AdS=CFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
10.4 Probes and IR/UV relations . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
10.5 IMSY and applications of AdS/CFT . . . . . . . . . . . . . . . . . . . . . . 105
10.6 Holographic dictionary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
10.7 Fefferman-Graham . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
10.8 BF bound and GKP/W . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
10.9 Holography and black holes . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

2
 10.10Maldacena’s eternal AdS BH . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
10.11Less symmetric holography . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
10.12Higher-spin/vector holography . . . . . . . . . . . . . . . . . . . . . . . . . . 109
10.13Bulk locality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
10.14Geometrization of entanglement . . . . . . . . . . . . . . . . . . . . . . . . . 110

11 Black hole entropy and the information paradox 110

11.1 Cooking up D1 k D5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
11.2 Cooking up W k D5 and W k D1 . . . . . . . . . . . . . . . . . . . . . . . . 111
11.3 Problems with too few ingredients . . . . . . . . . . . . . . . . . . . . . . . . 111
11.4 Why we started with D = 5 BH . . . . . . . . . . . . . . . . . . . . . . . . . 112
11.5 The harmonic function rule . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
11.6 Cooking the D1-D5 system . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
11.7 Adding the gravitational wave . . . . . . . . . . . . . . . . . . . . . . . . . . 113
11.8 The D1-D5-W metric in 5D . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
11.9 Bekenstein-Hawking entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
11.10Properties of SBH . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
11.11Yes, extreme BH can have finite SBH . . . . . . . . . . . . . . . . . . . . . . 115
11.12Four charges in four dimensions . . . . . . . . . . . . . . . . . . . . . . . . . 115
11.13The D2-D6-W-NS5 duality frame . . . . . . . . . . . . . . . . . . . . . . . . 116
11.14SBH for the 4D 4-charge BH . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
11.15The D-brane picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
11.16Open string dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
11.17String partition function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
11.18Degeneracy of states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
11.19Adding rotation [BMPV] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
11.20Rotating entropy agreement . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
11.21d = 4 entropy counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
11.22d = 4 microscopic entropy and non-extremality . . . . . . . . . . . . . . . . . 120
11.23Nonextremal entropy and greybody factors . . . . . . . . . . . . . . . . . . . 120
11.24String theory, D-branes, and SBH . . . . . . . . . . . . . . . . . . . . . . . . 121
11.25Emission rates and the fuzzball programme . . . . . . . . . . . . . . . . . . . 121
11.26D1-D5 CFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
11.27Firewalls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
11.28Avoiding firewalls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
11.29‘ER=EPR’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

3
1 Motivations for string theory
1.1 Running of couplings
As you may already know, coupling “constants” in Nature such as the fine structure constant
αEM = e2 /(4π~c) of electromagnetism are not actually constant. They run logarithmically
with energy in four spacetime dimensions (D = 4). You will learn the details of why this
happens by the time you have finished QFT II. For now, all you need is the concept that
running of couplings is caused by the quantum phenomenon of vacuum polarization. For
QED, electric charge is screened by polarization of all the virtual particle-antiparticle pairs
that are constantly popping in and out of the QED vacuum. The result of screening is
that you see more charge the deeper you dig: the coupling of electromagnetism grows with
energy. Note that this is a wholly quantum mechanical effect. Unfortunately, the growth of
αEM with energy makes perturbation theory less reliable. QCD behaves in a more physically
sensible way: its SU (3) colour coupling becomes weaker at high energy. In order to do this, it
has to anti-screen. The flip side of QCD weakening at high energy is confinement of colour
at low-energy. If you can explain exactly how this happens, the Clay Mathematics Institute
will dish out a million dollar prize! The story of the electroweak force, with hypercharge
U (1), isospin SU (2), W ± , Z and Higgs bosons, will be described soon in your QFT II course
in the section on spontaneous symmetry breaking.

Throughout this course we will work in particle physicist units where ~ = c = kB = 1.

We will not set the Newton constant (or any other physical constant) to unity – this would
not make sense from the perspective of dimensional analysis in quantum physics. Indeed,
we can build a constant with dimensions of length out of GN , c and ~,
 1/(D−2)
GN ~
`P,D ∼ (1.1)
c3

4
You can work out this formula by starting from the Einstein-Hilbert action principle for GR
and remembering that the Ricci scalar has two derivatives in it. The Planck length is of the
order of 10−33 cm in D = 4. Converting this into a mass scale gives about 10−5 g or 1019 GeV.
Einstein’s General Relativity is well-tested experimentally on distance scales from the
microscopic to the cosmic. It does however have a big Achilles heel: it predicts the seeds of
its own destruction. To be precise, singularity theorems of GR based on the Raychaudhuri
equation for geodesic deviation predict that the Einstein equations will develop singularities,
loci at which GR has nothing to say about the physics except that every local observable
including spacetime curvature is infinite. The two most common types of singularity in GR
that you will have already heard of are (a) the cosmological singularity at the Big Bang and
(b) the singularity at the heart of every black hole. The Cosmic Censorship conjecture that
proposed every singularity to be hidden behind an event horizon turned out to be incorrect,
so we know that GR has inherent limitations even as a classical theory.
The sickness of GR is actually inherent at the quantum level. It is very easy to see why.
The only dimensionless coupling out of which we can build scattering amplitudes scales like a
positive power of energy: αG ∼ GN E D−2 . This breaches unity at the Planck scale. What this
implies in detail is something very serious: GR predicts probabilities for graviton scattering
in excess of 100% at high energies.
String theory is the only extant formulation of quantum gravity that insists from the
outset that the theory obeys the correspondence principle. There is a Newtonian limit
and it is provable without any handwaving. Other approaches simply assume that their
nonperturbative formulation of “quantum gravity” somehow finagles this question, and try
to justify it post hoc, without providing a calculable mechanism. These approaches may
well end up defining quantum theories of gravity, but it is not clear that any of them has a
classical limit that is our gravity.

1.2 Anomaly cancellation and gauge/gravity unification

String theory actually originated as a theory of the strong interactions, in the late 1960s.
The rough idea of Susskind and others was to model two oppositely charged quarks within a
meson connected via a flux tube by a relativistic string with endpoints, which is known as an
open string. A closed string is a one-dimensional string without endpoints. The key fact
we will derive about open strings is that they possess spin-one massless gauge fields in the
spectrum. Similarly, the key fact about closed strings is that they possess massless spin-two
particles in the spectrum, which interact like the graviton. This is why string theory has a
chance to provide a unified theory of electromagnetic, nuclear, and gravitational interactions
– a dream that eluded even Albert Einstein.
Unfortunately, quantum relativistic open bosonic string theory turned out to have a
tachyon in the spectrum, which quickly ruled it out as a physically realistic model of the
strong interaction. Bosonic closed string theory also has a tachyon, with a more nega-
tive mass-squared than the bosonic open string. But if the ingredient of supersymmetry
(SUSY) is also added to the mix, fully tachyon-free quantum interacting superstring theo-
ries can be found. Supersymmetry pairs bosonic and fermionic modes, and it is the only
nontrivial extension of Poincaré (translation, rotation and Lorentz boost) symmetry in flat
Minkowski spacetime. Internal gauge symmetry is also possible; by definition, this does

5
not mix at all with spacetime symmetries.
String theory was later developed into a unified “theory of everything” encompass-
ing all known forces and matter. This was predicated on a crucial discovery in 1984 by
Michael B. Green and John H. Schwarz that quantum anomalies can be cancelled for five
different tachyon-free superstring theories in ten spacetime dimensions. These five theories
became known as Type I, Type IIA, Type IIB, Heterotic SO(32) and Heterotic E8 ×E8 . You
will meet anomalies later in your QFT II class; for now, you can think of them as quantum vi-
olations of classical symmetries of the action. As you will see in QFT II, anomaly-producing
Feynman diagrams can arise in D = 4 with three external gauge legs. In higher even dimen-
sions, anomalies arise for Feynman diagrams involving a greater number of external gauge
and/or graviton legs. In particular, in D = 10, hexagon anomaly diagrams arise, which must
cancel independently for gauge, gravitational, and mixed anomalies. These requirements are
sufficiently strict to permit very few solutions.
The Type I and Type IIA/IIB theories are pure open and closed superstring theories in
D = 10 respectively. The IIA theory is nonchiral (its two supersymmetries have opposite
chirality) while the IIB theory is chiral (its two supersymmetries have the same chirality).
The heterotic1 string theories are actually heteroses of bosonic strings and superstrings.
They hybridize a bosonic string theory for left-handed string modes, which lives naturally
in D = 26 spacetime dimensions, with a fermionic string theory for right-handed modes,
which lives naturally in D = 10. The 16 leftover left-handed modes are recombined into a
current algebra representing a spacetime gauge symmetry. Anomaly cancellation constraints
for both gauge and gravitational interactions then require that the rank of the gauge group
for heterotic strings be 496. There are only two (Lie) symmetry groups with this property:
SO(32) and E8 × E8 .
Superstring theories all include the graviton. In the case of open string theory the
graviton arises as a mode propagating in internal Feynman diagram loops. For the closed
string theories it arises directly as the lightest mode in the spectrum. The most important
thing about the graviton mode coming out of string theory is that it not only has the right
mass and spin, but it also interacts like a graviton. This means that we can honestly call
string theory a consistent quantum theory of gravity. But of course it has much more than
just the massless graviton – the overtones (known as oscillators) change the physics at
short distance in very important ways from Einstein’s low-energy effective field theory. The
upshot is that string theory provides a way of unifying gauge and gravitational interactions
together in a fashion that is consistent with quantum mechanics.

1.3 Relativistic point particle theory

For non-relativistic point particles, you know how the drill goes. The basic dynamical vari-
ables are xi (t), where i = 1, 2, . . . D − 1. There is no issue about how to parametrize t, as
all observers agree on time because of Galilean relativity. But for relativistic point particles
we cannot just use the Newtonian kinetic energy: we have to use something that respects
1
This word arises from the Greek ‘heterosis’. It is not the antonym of ‘homoerotic’ !

6
Einsteinian relativity. The usual action that people write is proportional to the arc length,
r
dxµ dxµ
Z
0
Sparticle = −m dτ − , (1.2)
dτ dτ
where τ is the proper time. This action has the benefit that, at low speeds, it reduces to the
familiar non-relativistic action (up to an additive constant).
The drawback of this action is twofold: the xµ (τ ) are not independent functions, and
the particle is assumed to be massive so that proper time can be used to parametrize the
worldline. First, the dynamical variables must obey a mass shell constraint,

pµ pµ = −E 2 + |~p|2 = −m2 . (1.3)

Second, suppose that we use this proper time story to try to find the canonical momenta
and Hamiltonian. We have pµ = mdxµ /dτ . But this means that the Hamiltonian H(τ ) =
(pµ pµ + m2 )/m which is zero by dint of the constraint equation. The fact that we have
obtained a zero Hamiltonian is caused by the fact that we forgot to take account of a
symmetry.
Let us be more sophosticated about writing down our Lagrangian. Impose the mass shell
constraint via a Lagrange multiplier e(λ) known as the einbein, where λ is an arbitrary
worldline parameter,
1 −1 dxµ dxµ 1
Z  
e 2
Sparticle = dλ e (λ) − e(λ)m . (1.4)
2 dτ dτ 2
Note that this einbein action is invariant under

λ → λ0 ,
dλ
e → e0 = e. (1.5)
dλ0
Varying w.r.t. e(λ) gives the constraint as
r
±1 dxµ (λ) dxµ (λ)
e(λ) = , (1.6)
m dτ dτ
and the proper time gauge is one in which e(λ) = 1/m. Varying w.r.t. xµ (λ) gives the
equations of motion
dxµ (λ)
 
d −1
e (λ) = 0, (1.7)
dλ dλ
which are equally valid for massive or massless particles. The canonical momenta are
1 dxµ (λ)
pµ = (1.8)
e(λ) dλ
and so the Hamiltonian is
1 
H = e pµ pµ + m2 .

(1.9)
2
This gives the correct Poisson Brackets: {xµ , pν }PB = δνµ , {xµ , xν }PB = 0, and {pµ , pν }PB = 0.

7
1.4 Worldlines vs worldsheets
What does string theory do differently? It assumes that the fundamental LEGO blocks of
the universe are not zero-dimensional point particles but instead one-dimensional strings.
This means that the path through spacetime is not a one-dimensional worldline described
by xµ (λ), where λ is a parameter commonly chosen to be proper time for a massless particle
or affine parameter for a massless particle. It is instead described by a worldsheet, which
is the two-dimensional surface swept out in spacetime by a moving open or closed string.
Topologically speaking, the worldsheet associated to an open string propagator would be
equivalent to a finite-width strip while that for a closed string propagator would be equivalent
to a finite-width cylinder.
To describe string dynamics, we need a map from the worldsheet, parametrized by 2D
coordinates (τ, σ), to spacetime: X µ (τ, σ). We also need an action principle, which we
will choose by analogy with the relativistic point particle to be the area swept out by the
worldsheet. We will have much more to say about this story shortly. For now, we need
to know the one big feature of the action principle for the X µ (τ, σ) of string theory that
makes life difficult: it possesses full 2D reparametrization invariance. The worldsheet action
is fully invariant under coordinate transformations. Particle theory has a small echo of this
symmetry, but it is far less powerful because it is only in two spacetime dimensions that the
conformal symmetry group is infinite-dimensional. 2D conformal symmetry will give us a
great deal of theoretical control over analyzing the quantum physics of strings. But we do
have to learn how to write down a measure for the path integral of quantum string theory
that does not double-count field configurations that are related to one another by worldsheet
reparametrization. This involves a story known as Fadeev-Popov ghosts, which you can
think of as a technical bookkeeping device for keeping track of reparametrization symmetry.
The string is fundamental, which means that there is no medium doing any wiggling.
Accordingly, there are no longitudinal oscillations of strings. There are (D − 2) physical
oscillator modes in D spacetime dimensions. Now suppose that we want to make a super-
string. Quite generally, in D dimensions, spinors do not have the same dimensionality as
scalar fields like X µ (τ, σ). For instance, the dimensionality of a Weyl spinor in D dimen-
sions is 2[D/2] ; note that this grows much faster than D. To match worldsheet bosonic and
fermionic degrees of freedom requires fortunate accidents of available spinor representations,
and this can only be done classically for string theory in D = 3, 4, 6, 10. (It is no accident
that D − 2 for these cases is a power of 2.) As we will see later in this course, the quantum
superstring is only consistent in D = 10, where there are eight real transverse X i modes
and eight real fermionic ψ i modes to match them. There is no hand-waving way to derive
the critical dimension – believe me, I have looked – we have to go through the full pain of
quantizing superstrings to understand where it comes from. But I can give you a part of the
answer right now.
You might wonder why the groundstate of open and closed strings have spin. This is
related to the existence of open and closed string tachyons, which are spinless and have
negative m2 . Indeed, the mass formula for an open string will turn out to be, schematically,
2
mc2

= (N − 1) . (1.10)
T open

8
Here, N is the oscillator energy measured in units of the fundamental mode; it is a non-
negative integer. The oscillation patterns are basically standing waves. For the closed string,
we get travelling waves. Schematically,
2
mc2

= 2(NR + NL − 2) . (1.11)
T closed

where NL,R are non-negative integers representing the oscillator energies for the independent
left- and right-handed modes in units of the fundamental. For momentum balance, we need
NL = NR . Where do these mysterious mass deficits (the -1 and the -2) come from? The
answer lies in the Casimir Effect. Quantum fields are like messy teenagers at heart: they
prefer to fluctuate over all space. If you then exclude some part of space, you pay the price
of an energy deficit, which is proportional to the volume of excluded space. The technical
details for the strip diagram relevant to the open string give rise to a mass-squared deficit
of -1, in units of the string tension
1 1
T ≡ 0
≡ . (1.12)
2πα 2π`2s

For the cylinder diagram relevant to the closed string, the same techniques gives a mass-
squared deficit of -2. So this is why we get a gauge boson: we need one oscillator on the
open string in order to make a massless mode. That oscillator points in a particular direction,
and has (D − 2) components, indicating a vector. Similarly, we need two oscillators on the
closed string (one on the left and one on the right) in order to make a massless mode that
we can identify as a graviton. To be picky: the transverse traceless tensor part is identified
with the graviton, the antisymmetric tensor part as the Kalb-Ramond field B, and the
trace part as the dilaton field Φ.

1.5 Symmetries in various D and the role of BPS states

Symmetry has proven to be an extremely powerful way of organizing our physics thoughts
about Nature. We are already familiar with 10 spacetime symmetries from study of relativ-
ity: 4 spacetime translations, 3 spatial rotations, and 3 Lorentz boosts. Conservation laws
associated to them ensure both linear and angular momentum conservation as well as correct
CoM motion. The counting goes similarly in other spacetime dimensions, except that in D
dimensions there are D translations, d=D−1 boosts, and d(d − 1)/2 rotations. (Note: the
number of independent planes of rotation is [d/2] where [ ] denotes the integer part, e.g. the
(x, y) and (z, w) planes in d=4 define two independent angular momenta.)
Other symmetries, such as the U (1) gauge symmetry of electromagnetism, act on the
fields directly. Field space, as distinct from spacetime, is usually referred to as the “internal
space” for the field. The charge carried by a field can be thought of as like a handle pointing
in field space, onto which a gauge boson can grab.
Gauge fields can be in three distinct phases of physical behaviour. The most familiar is
the “Coulomb phase”, resulting in an inverse-square law in D=4 as per intuition. Alterna-
tively, like in QCD, the gauge field can be in a “confined phase”. The third possibility is a
“Higgs phase” with spontaneous symmetry breaking.

9
 Under a symmetry transformation, a field φa transforms as
∆φa = φa 0 (x0 ) − φa (x)
= φa 0 (x0 ) + [−φa 0 (x) + φa 0 (x)] − φa (x)
= [φa 0 (x0 ) − φa 0 (x)] + [φa 0 (x) − φa (x)]
= (∂µ φa )δxµ + δφa (x) . (1.13)
Notice that there is a transport term and a straight functional variation piece. Noether’s
Theorem says that for every continuous symmetry there is a conserved current
∂L ∆φa µ ∆x
ν
JAµ = − T ν (1.14)
∂∂µ φa ∆ω A ∆ω A
where ∆φa = (∆φa /∆ω A )∆ω A and similarly for xν . The important thing about the infinites-
imal parameters ∆ω A is that they are additive.
If a field φ carries a representation R of a group U , it transforms as
φ0 = U φ = exp −i∆ω A TA φ


(1.15)
where the TA are the infinitesimal generators. The generators of the Lie algebra obey
[TA , TB ] = ifABC TC , (1.16)
and the Jacobi identity
[TA , [TB , TC ]] + [TB , [TC , TA ]] + [TC , [TA , TB ]] = 0 . (1.17)
Poincaré symmetry transformations consist of translations, rotations, and Lorentz
boosts. In D spacetime dimensions, the group is known as SO(D − 1, 1). When D=4,
this can be made to look like a compact Lie group by Wick rotating to produce SO(4),
which happens to be isomorphic as a group to SU (2) × SU (2). Now, we already know how
to do SU (2) physics - the three generators are very familiar to us from Quantum Mechanics:
they are the Ji from ladder operator fame. For more on the story of the representation theory
of Poincaré, including why spin is an angular momentum for massive particles while massless
particles are characterized by their helicity, pay attention in QFTII class PHY2404S.
The generators of translations are known as the momentum generators Pµ . Note that
in general spacetime dimension D, the [orbital] angular momentum is not a pseudovector
but instead the space-space components of a two-index antisymmetric tensor. (Each rotation
involves one plane perpendicular to its axis.) If we use a relativistic two-index antisymmetric
tensor, we can actually pack both the rotation generators in (space-space components) and
also the boost generators (time-space components). The orbital angular momentum is just
what you might expect: Lµν = Xµ Pν − Xν Pµ ; the total angular momentum is Mµν =
Lµν + Σµν , where Σµν are generators of spin angular momentum transformations which
act nontrivially on any field with spin half or greater. The commutation relations for the
Poincaré generators then become

[Pµ , Pν ] = 0
[Pµ , Mρσ ] = +i (ηµρ Pσ − ηµσ Pρ ) (1.18)
[Mµν , Mρσ ] = +i (ηνρ Mµσ − ηµρ Mνσ + ηµσ Mνρ − ηνσ Mµρ )

10
 The supersymmetry algebra for point particles is described by extending the regular
Poincaré generators, translations Pµ and the antisymmetric rotation/boost generators Mµν ,
by SUSY generators QA α obeying

AB µ
{QA B
α , Qβ } = −2δ Γαβ Pµ − 2iZ AB δαβ , (1.19)

where all other commutators are zero and we have used the conventions of Polchinski2 . Z
is known as the central charge. There are also some commutators of Poincaré generators
with the QAα which encode the fact that it transforms as a spinor.
To show the power of the SUSY algebra, let us work in the rest frame of a massive particle.
Suppose further for simplicity that we are in D = 0 + 1, i.e. doing QM, so that spinors have
only one component (this is not an essential restriction). Then for two supersymmetries, the
simplest case of extended SUSY, the algebra says that {QA , Q†B } = −2δ AB M − 2iZ AB . We
can easily simplify this by shifting basis to Q± := (Q1 ± iQ2 )/2; then we have {Q± , Q†± } =
−M ∓ Z while {Q± , Q†∓ } = 0. Now comes the really cool part – we sandwich a physical
state |ψi around {Q± , Q†± } = −M ∓ Z. Since the SUSY generators are physical operators,
Q± |ψi is also a physical state, and it must have non-negative norm. This directly implies
that
M ≥ |Z| , (1.20)
which is known as the Bogomolnyi-Prasad-Sommerfeld or BPS bound. It is extremely
important and powerful, not least because it provides a lower bound on the mass of every
state in the quantum theory. Also, the only way that the BPS bound can be saturated is
for the physical state to possess unbroken supersymmetry, i.e., be annihilated by one or
more SUSY generators. States with unbroken SUSY are known as BPS states. The best
part about BPS states is that their mass-to-charge ratio can be followed reliably away from
the perturbative regime. Quantum corrections to the mass and to the charge do typically
occur, but the BPS bound derived directly from the SUSY algebra ensures that the ratio
M/|Z| is protected from quantum corrections. This story is part of what helped the second
superstring revolution to occur.
One of the reasons string theory is so well constrained as a quantum theory in the
ultraviolet is that it possesses an infinite dimensional symmetry algebra on the worldsheet.
Handling the gauge symmetry associated to this is technically a bit more challenging than
for spin-one U (1) gauge field theory that you studied in PHY2403F. This is why we will be
introducing a bunch of new technologies. :D

2 Classical Bosonic String Theory

2.1 Conventions
We will use the conventions of Polchinski’s “String Theory” textbooks. Mostly we will focus
on Volume 1, but we will also make use of some Volume 2 material. In Lorentzian signature,
2
We will use Polchinski’s conventions for units etc. throughout these lecture notes. For the SUSY algebra
in D dimensions, see Appendix B of the second volume.

11
we use the {−, +, . . . , +} convention. Light-cone coordinates are defined as
1
x± = √ x0 ± x1 .


(2.1)
2
In Euclidean space with coordinates σ 1 , σ 2 , we can define complex coordinates

z = σ 1 + iσ 2 , z̄ = σ 1 − iσ 2 , (2.2)

so that
1 1
∂z = (∂1 − i∂2 ) , ∂z̄ = (∂1 + i∂2 ) . (2.3)
2 2
Then gzz̄ = 1/2, g zz̄ = 2, with all other components zero, which implies

d2 z ≡ 2dσ 1 dσ 2 , (2.4)

and
1
δ 2 (z, z̄) ≡ δ(σ 1 )δ(σ 2 ) . (2.5)
2
Stokes’ Theorem is simple in 2D,
Z I
2 z z̄
d z (∂z v + ∂z̄ v ) = (v z dz̄ − v z̄ dz) (2.6)
R ∂R

Traditionally, the left-moving sector is called holomorphic and the right-moving sector is
called anti-holomorphic. String theorists also tend to use a different convention for the
definition of the stress-energy tensor, with an extra factor of −2π,
−4π δ
T ab = √ S (2.7)
−γ δγab

2.2 Nambu-Goto and Polyakov actions

Previously we saw that there were two classical actions for (massive) point particles, the
’square-rooty’ action based on the geometric arc length and the quadratic einbein action
which is easier to quantize. For classical strings, there are direct analogues known as the
Nambu-Goto and Polyakov actions. The dynamical variables X µ (τ, σ) are maps from the
worldsheet to target space, i.e., fields living on the worldsheet.
The Nambu-Goto action is simple: it measures the worldsheet area swept out by the
string as it moves through spacetime,
Z
1 p
SN G = − dτ dσ −hab , (2.8)
2πα0

where3
hab = ∂a X µ ∂b X ν Gµν (X) (2.9)
3
Note that this is not the Einstein tensor. It is the string frame metric tensor, which is distinct from the
Einstein frame metric tensor.

12
is the induced metric on the worldsheet. It can also be written as
Z
1
q
SN G = dτ dσ (Ẋ · Ẋ)(X 0 · X 0 ) − (Ẋ · X 0 )2 (2.10)
2πα0
where
∂X µ ∂X µ
Ẋ µ = , X 0µ = . (2.11)
∂τ ∂σ
The Polyakov action is given by
√
Z
1
SP = 0
dτ dσ −γγ ab ∂a X µ ∂b X ν Gµν (X). (2.12)
4πα
where γ ab is a worldsheet metric. The bonus of the Polyakov action is that it is easier to
quantize in Minkowski spacetime, since it is nice and quadratic in derivatives of the X µ
fields. Since the physics should be worldsheet reparameterization invariant, the worldsheet
metric should not play any physical role, unlike the metric in the target space Gµν (X).As
we will see in more detail later, this works out because gravity in 2D has no local degrees of
freedom. The energy-momentum tensor defined by
−4π δSP
T ab = √ . (2.13)
−γ δγab
gives, from the Polyakov action,
 
−1 µ ν 1 c µ ν
Tab = 0 Gµν (X) ∂a X ∂b X − γab ∂ X ∂c X . (2.14)
α 2
Varying the Polyakov action w.r.t. γab fixes it to be the induced metric hab , as you can
check yourself. This then implies that the two action principles are classically equivalent.
For the quantum theory, we will stick with the Polyakov action.
Varying the Polyakov action w.r.t. X µ gives the equation of motion
√
∂a −γγ ab ∂b X µ = 0 ,


(2.15)
as long as we take care of the surface terms correctly. The variation of the Polyakov action
obviously has two pieces:
Z +∞ Z ` Z +∞
1 √ µ 2 1  √ µ σ
σ=`
δSP = dτ dσ −γδX ∇ X µ − dτ − −γδX ∂ X µ (2.16)
2πα0 −∞ 0 2πα0 −∞ σ=0

and these can be solved with two choices

X µ (τ, σ + `) = X µ (τ, σ) (closed string)
∂L
δX µ (σ = 0, `) = 0 (open string) (2.17)
∂X µ0
These are the only options consistent with target space Poincaré invariance. For the closed
string, the function X µ (τ, σ) has to be periodic in σ. For the open string we require
Neumann boundary conditions, in order to respect target spacetime Poincaré invariance.
This ensures that the open string endpoints move freely and that momentum does not leak
off. As we will find out in the later section on D-branes, there is another possibility to use
Dirichlet boundary conditions for open strings, but this spontaneously breaks Poincaré, and
typically involves momentum interchange between the open string and the D-brane.

13
2.3 Conformal reparametrizations, Weyl invariance, and confor-
mal gauge
The Polyakov action possesses two important classes of symmetries: bulk global symmetries
and local worldsheet symmetries. The global symmetries are Poincaré:-

δX µ = aµν X µ + bµ , aµν = −aνµ ,

δhab = 0 . (2.18)

The local symmetries are twofold. First, there is Weyl invariance,

δX µ = 0 ,
δhab = 2Λhab . (2.19)

Second, there is full worldsheet reparametrization invariance

δX µ = −ξ a ∂a X µ ,
δh = − (∇a ξb + ∇bξa ) ,
√ ab √
δ −h = −∂a ξ a −h . (2.20)

Weyl invariance implies that the stress-energy tensor is traceless

Ta a = 0 , (2.21)

while Diff invariance implies that is is conserved,

∇a T ab = 0 . (2.22)

What about adding other terms to the worldsheet action? There are only two other
terms possible that are consistent with the above symmetries and possessing at most two
derivatives. They are
√
Z
S1 = λ1 d2 σ −h (2.23)
Σ
which is diff invariant but not Weyl invariant. There is also the Gauss-Bonnet term
√
Z
λ2
S2 = d2 σ −h (2) R = λ2 χ(Σ) , (2.24)
4π Σ
which turns out to not contribute at all to the equations of motion because it is topological.
The integrand is locally a total derivative because we are in 2D. While the Gauss-Bonnet
term does not affect the EOM, it does affect the weighting of different worldsheets in the
Feynman path integral. In particular, we see a weighting factor of e−λ2 χ(Σ) . We may use a
mathematical fact
χ(Σ) = 2 − 2g , (2.25)
where g is the genus, to see that worldsheets of genus g contribute with weighting e−2(1−g)λ2 .
The lowest genus zero term is weighted by 1/(eλ2 )2 , the genus one term by 1, the genus two

14
term by (eλ2 )2 , etc. Accordingly, we can think about the genus as a loop counting parameter
in worldsheet perturbation theory.
In two dimensions, the structure of gravity is very rigid: metrics are conformally flat. For
suitable worldsheets, the metric can be gauged away completely by using (a) reparametriza-
tion invariance to set hab = eΦ ηab and (b) Weyl invariance to set Φ = 0. See p.18 of
Blumenhagen-Lüst-Theisen (BLT) for details of how this is done. In conformal gauge, the
Polyakov action reduces to
Z
(conf.gauge)
SP = 2T d2 σ∂+ X µ ∂− Xµ (2.26)

and the equations of motion simplify enormously to give

X µ (τ, σ) = XRµ (σ − ) + XLµ (σ + ) . (2.27)

In other words, the fields X µ are straightforward travelling waves classically, for the closed
string. For the open string we get straightforward standing waves.
Vanishing of Tab on a classical solution is required by the equation of motion for the
worldsheet metric. Expressed in linear combinations, this says
 2
0
Ẋ ± X = 0, (2.28)

or, in light cone coordinates,

T±± = −2πT (∂± X · ∂± X) = 0 , T±∓ = 0 . (2.29)

This implies that ∂∓ T±± = 0 and therefore

T±± = T±± (σ ± ) . (2.30)

This has huge implications – it means there exists an infinite number of conserved charges,
obtained by integrating T±± against an arbitrary function of σ ± , e.g.
Z `
Lf = 2T dσf (σ + )T++ (σ + ) (2.31)
0

and similarly for the anti-holomorphic sector. The Hamiltonian in conformal gauge is then
Z `
(∂+ X)2 + (∂− X)2 .
 
H=T (2.32)
0

The canonical momentum is Πµ = ∂L/∂ Ẋµ = T Ẋ µ , and the Poisson Brackets are as you
would expect,

{X µ (τ, σ), X ν (τ, σ 0 )}P.B. = 0 ,

{Ẋ µ (τ, σ), Ẋ ν (τ, σ 0 )}P.B. = 0 ,
1 µν
{X µ (τ, σ), Ẋ ν (τ, σ 0 )}P.B. = η δ(σ − σ 0 ) . (2.33)
T
15
The charges Lf above generate reparametrizations that stay within conformal gauge,

{Lf , X(σ)}P.B. = −f (σ + )∂+ X(σ) . (2.34)

So far we have just talked about the influence of local diffeo symmetries on worldsheet
physics for classical strings. There is also the question of bulk Poincaré invariance to think
about. In conformal gauge, Z `
Pµ = T dσ (∂τ Xµ (σ)) (2.35)
0
and Z `
Jµν = T dσ (Xµ ∂τ Xν − Xν ∂τ Xµ ) (2.36)
0
and these are conserved for closed strings by virtue of periodicity. For open strings, only
those with Neumann boundary conditions have these two tensors conserved. We will explore
the physics of Dirichlet-branes later in the course – stay tuned! You can check that the
correct algebra is obtained for Poincaré symmetry:-

{P µ , P ν }P.B. = 0 ,
{P µ , J ρσ }P.B. = η µσ P ρ − η µρ P σ ,
{J µν , J ρσ }P.B. = η µρ J νσ + η νσ J µρ − η νρ J µσ − η µσ J νρ . (2.37)

2.4 Closed string oscillator expansions and the Virasoro algebra

Classically, the solution of the 2D Klein-Gordon equation is a superposition of left- and
right-moving travelling waves,

X µ (z, z̄) = XRµ (τ − σ) + XLµ (τ + σ) , (2.38)

where
r
0
1 µ πα α0 X 1 µ −2πin(τ −σ)/`
XRµ (τ − σ) = (x − cµ ) + pµ (τ − σ) + i α e ,
2 ` 2 n6=0 n n
r
µ 1 µ µ πα0 µ α0 X 1 µ −2πin(τ +σ)/`
XL (τ + σ) = (x + c ) + p (τ + σ) + i ᾱ e . (2.39)
2 ` 2 n6=0 n n

In these expressions, xµ is the centre of mass position. Note that the αnµ are positive-frequency
modes for negative n and negative-frequency for positive n. Requiring that X µ (τ, σ) be real
gives
(αnµ )∗ = α−n
µ
. (2.40)
Now define r
α0 µ
α0µ = ᾱ0µ = p ; (2.41)
2
then r n=+∞
2π α0 X 2πin(τ ±σ)
∂± X µ = e . (2.42)
` 2 n=−∞

16
We can evaluate the momenta and angular momenta associated to this string configuration,
( ∞
)
X 1 µ
P µ = pµ , J µν = (xµ pν − xν pµ ) + −i αnν + α−n
ν
αnµ + h.c. ,


α−n (2.43)
n=1
n

which give the Poisson brackets you would expect

{αm , αn } = −imη µν δm+n,0 ,

{ᾱm , ᾱn } = −imη µν δm+n,0 ,
{αm , ᾱn } = 0,
{xµ , pν } = η µν . (2.44)

and for the Hamiltonian we obtain

+∞
π X
H= (α−n · αn + ᾱ−n · ᾱn ) . (2.45)
` n=−∞

For the closed string, a complete set of functions is provided by plane waves:
n ±
o
fm (σ ± ) = e2πimσ /` (2.46)

We already saw before that there is an infinite of conserved charges. So let us define
Z `
`
Ln := − 2 dσe−2πinσ/` T−− ,
4π Z0
`
`
L̄n := − 2 dσe+2πinσ/` T++ ; (2.47)
4π 0
then
1X
Ln = αn−m · αm ,
2 m
1X
L̄n = ᾱn−m · ᾱm . (2.48)
2 m

Notice that we have deliberately ignored zero-point energies in evaluating these sums. This
is of course only a valid approximation for classical strings. In following chapters, we will
revisit the zero-point energies
P and2πin(σ−σ
sum them up correctly. Recalling the orthonormality
0 0 )/`
condition δ(σ − σ ) = (1/`) n∈Z e , we find
 2 X
2π
T++ = − Ln e2πinσ/` ,
` n
 2 X
2π
T−− = − L̄n e2πinσ/` . (2.49)
` n

The reality condition for X µ demands that

Ln = L∗−n ,

17
 L̄n = L̄∗−n . (2.50)

A nice consequence of the above definitions is that the Hamiltonian in conformal gauge
(which we have been using throughout this subsection) is very simple,
2π

H= L0 + L̄0 . (2.51)
`
Rigid σ translations are generated by the constraint
Z `
2π
dσX · X 0 =


T L0 − L̄0 , (2.52)
0 `
and since no point on the string is special, for momentum balance we require

L0 = L̄0 . (2.53)

The infinite set {Ln } obeys a Virasoro algebra

{Lm , Ln }P.B. = −i(m − n)Lm+n ,

{L̄m , L̄n }P.B. = −i(m − n)L̄m+n , (2.54)

When we do quantum string theory, we will find that this algebra is modified by a central
extension term proportional to (D − 26) (for the bosonic string). This will be required to
vanish in order to have conformal invariance as a quantum symmetry of the theory.

2.5 Open string oscillator expansions, Virasoro, and D-branes

For open strings, it is necessary to carefully distinguish between Neumann (N) and Dirichlet
(D) boundary conditions, because they lead to physically different mode expansions. With
any given map function X µ (τ, σ), the open string involved has two endpoints, requiring us
to distinguish four cases: NN, DD, ND, and DN. We now discuss these cases in turn.
For NN strings, the solution to the wave equation becomes
2πα0 µ √ X1 nπσ
(NN) X µ (τ, σ) = xµ + p τ + i 2α0 αnµ e−iπnτ /` cos( ), (2.55)
` n6=0
n `

This time, because the string is open rather than closed, there is only one set of mode
oscillators αnµ appearing in
r n=+∞
µ 1 µ µ0 π α0 X µ −iπn(τ ±σ)/`
∂± X = (Ẋ ± X ) = α e . (2.56)
2 ` 2 n=∞ n

Note that α0µ is defined differently than for the closed string because of a physically important
factor of two, √
α0µ = 2α0 pµ . (2.57)
For the open string the Poisson brackets are
µ
{αm , αnν }P.B. = −imη µν δm+n,0 ,

18
 {xµ , pν }P.B. = η µν . (2.58)

Now let us look at the Hamiltonian. Using the simple fact that

X µ0 (τ, σ) = −X µ0 (τ, −σ) , (2.59)

a ‘doubling trick’ allows us to rewrite

Z ` Z `
2 2
dσ(∂+ X)2 .
 
dσ (∂+ X) + (∂− X) = (2.60)
0 −`

On σ ∈ [−`, +`], the plane waves eiπmσ/` are periodic. Then the Hamiltonian is
n=+∞
π X
H= α−n · αm−n . (2.61)
2` n=−∞

The Virasoro generators are defined by

Z `
`
dσ eiπmσ/` T++ + e−iπmσ/` T−−
 
Lm = − 2
2π Z0
`
`
dσ eiπmσ/` (∂+ X)2 + e−iπmσ/` (∂− X)2
 
= 2 0
2π α Z0
`
`  iπmσ/` 2

= dσ e (∂ + X)
2π 2 α0 −`
n=+∞
1 X
= (∂+ X)2 . (2.62)
2 n=−∞

They obey
{Lm , Ln }P.B. = −i(m − n)Lm+n . (2.63)
Now we turn to the DD strings. For them, the endpoints are fixed by the requirement that
Ẋ (τ, σ = {0, `}) = 0, which must hold for all τ . Integrating this gives X µ (τ, σ = 0) = xA
µ

while X µ (τ, σ = `) = xB , where xA,B are constants. This implies the mode expansion

1 √ X1 πnσ
(DD) X µ (τ, σ) = xµA + (xµB − xµA )σ + 2α0 αnµ e−iπnτ /` sin( ) (2.64)
` n6=0
n `

This time,
1
α0µ = √ (xµB − xµA ) (2.65)
π 2α 0

and of course the centre of mass is at (xA + xB )/2. The only other material difference with
NN open strings is that
1 2 π X
H= (x B − x A ) + α−n · αn , (2.66)
4πα0 ` 2` n6=0

19
i.e. that the Hamiltonian must contain a term describing the potential energy of the string
stretched between xA and xB .
Next, we examine the mixed ND strings. These beasts have different half-integer mod-
ings, because the first end (traditionally chosen at σ = 0) is Neumann while the other (at
σ = `) is Dirichlet. Specifically,
√ X 1 πrσ
(ND) X µ (τ, σ) = xµB + 2α0 αrµ e−iπrτ /` cos( ), (2.67)
r `
r∈Z+1/2

where xµB is the position of the second (D) end. The DN strings are also half-integer moded,
√ X 1 πrσ
(DN) X µ (τ, σ) = xµA + 2α0 αrµ e−iπrτ /` sin( ), (2.68)
r `
r∈Z+1/2

The reality condition for all four kinds of boundary conditions is

µ
(αnµ )∗ = α−n . (2.69)
We can in fact extend the doubling trick to combine the light-cone derivatives ∂± X µ (τ, σ)
into one field, traditionally defined as left-moving, defined on the doubled interval σ ∈ [0, 2`],

µ ∂+ X µ (τ, σ) σ ∈ [0, `]
∂+ X = µ (2.70)
±∂− X (τ, 2` − σ) σ ∈ [`, 2`]
where the ± sign is + for (NN) and (DD) strings and is - for (ND) and (DN) strings. Then
r
π α0 X µ −iπn(τ +σ)/`
∂+ X µ (τ, σ) = α e (2.71)
` 2 n n

where the n is summed over integers for (NN) and (DD) strings and over half-integer modes
for (ND) and (DN) strings. For use with the doubling trick, periodicity conditions can be
expressed as
∂+ X µ (τ, σ + 2`) = +∂+ X µ (τ, σ) (NN), (DD)
∂+ X µ (τ, σ + 2`) = −∂+ X µ (τ, σ) (ND), (DN) . (2.72)
Why do people talk about Dp-branes with different dimensions p? Because one may
choose to select Dirichlet boundary conditions on different numbers of the X µ maps for
various µ. If there are no Dirichlet directions at all, this is referred to as either ‘open string
endpoints moving freely at the speed of light’ or ‘a space-filling D-brane’. In D = 26 this is
would be a D25-brane. If by contrast there are no Neumann directions, this is referred to as
a D0-brane, a pointlike object. (In Euclidean spacetime signature we can even put Dirichlet
boundary conditions on the time direction, making what is called a D-instanton.) All the
options in between, D1-branes through D24-branes, are obtained by choosing (D − 1 − p)
of the X µ to have Dirichlet BCs and the other p directions to be Neumann. The D1-brane
is sometimes called a D-string, which is physically very distinct from a fundamental string
because it has a nonperturbative tension of order 1/gs in string units while the fundamental
string tension is of order unity. All D-branes have a tension of order√1/gs in string units;
the greater the worldvolume dimensionality p, the more powers of 1/ α0 we need to soak
up the dimensions of length.

20
3 Quantum Bosonic String Theory
3.1 Canonical quantization and negative norm states
The standard prescription in canonical quantization is to replace Poisson Brackets by quan-
tum mechanical commutators4
1
{ , }P.B. −→ [ , ] (3.1)
i
Using what we derived in the section on classical string theory, the equal time commutators
then become
[X µ (τ, σ), X ν (τ, σ)] = 0 = [Ẋ µ (τ, σ), Ẋ ν (τ, σ)] . (3.2)
and
[X µ (τ, σ), Ẋ ν (τ, σ)] = 2πiα0 η µν δ(σ − σ 0 ) , (3.3)
In our mode expansion for the closed string, when we quantize, the Fourier mode coeffi-
cients are promoted to operators5 . The operator commutation relations they obey are

[xµ , pν ] = iη µν ,
µ
[αm , αnν ] = mδm+n η µν = [ᾱm
µ
, ᾱnν ] ,
µ
[αm , ᾱnν ] = 0 . (3.4)

(For the open string, the barred sector is absent.) The operator X µ (τ, σ) must be hermitian,
so
µ † µ µ † µ
(αm ) = α−m , (ᾱm ) = ᾱ−m . (3.5)
These guys are
√ quite familiar to you already: they just describe a simple harmonic oscillator
µ µ
where αm = mam . Notice that the number operator for the mth mode Nm =: αm · α−m :
contains a factor of m compared to what you are used to from undergraduate QM.
µ
These commutators imply that the positive modes αm , m > 0 correspond to the annihi-
lation operators while the negative modes with m < 0 correspond to the creation operators.
The vacuum can have a CoM momentum pµ . Therefore the vacuum state obeys
µ
αm |0; pµ i = 0 , m > 0 ,
p |0; pµ i = pµ |0; pµ i .
µ
(3.6)

In the second equation, the [first] pµ on the LHS is to be understood as an operator, while
the [first] pµ on the RHS as its eigenvalue.
µ
Notice that the constant on the RHS of the [αm , αnν ] commutator involves the Minkowski
metric η µν which has Lorentzian, not Euclidean signature. The minus sign in the time-
time component is a harbinger of a serious physical problem known as negative norm states
0
or ghosts. To see this, consider states of the form α−m |0i with m > 0 involving the
0 0
troublesome time component. These satisfy h0|αm α−m |0i = −mh0|0i < 0; in other words,
they have negative norm. This is the key reason why canonical quantization of string theory
is nontrivial: we have to figure out what to do about negative norm states.
4
Recall that we are working throughout in grownup units in which ~ = c = 1.
5
Following BLT, will only put hats on operators when it is necessary to avoid physical confusion.

21
 There is a way forward. Remember the constraint equations that followed from symmetry
under reparametrizations? We can impose them as operator equations constraining physical
states. The hope is that the nasty ghosts decouple from the physical Hilbert space. The
algebra is sufficiently long and unilluminating compared to more modern approaches to be
discussed soon that we will not show any of the details here; if you are curious you can look
it up in Green, Schwarz and Witten volume 1.
It is possible to prove a no-ghost theorem, provided that two conditions are met in
quantum bosonic string theory. First, the spacetime dimension has to be exactly 26, so that
the number of physically transverse modes is (D − 2) = 24. Second, the normal ordering
constant a in the mode expansion for the Virasoro generator L0 has to be exactly a = −1. It
is very interesting that we have to make these specific choices in order to get quantum string
theory to work in canonical quantization – and they are very physically relevant choices.
Note in particular that we did not obtain D = 4 for the critical dimension! When we study
light-cone quantization and modern BRST quantization, we will see how to derive these
results in a very different way. To give you a sneak peek: in light-cone gauge, D = 26
and a = −1 will be demanded by the closure of the Virasoro algebra. In the most modern
approach, the result will be obtained with the least amount of drudgery.
Our next order of business is to discuss the propagators for our physical fields the
µ
X (τ, σ). In order to begin solving a 2D wave equation on a worldsheet, we need the Green’s
function. As you should already know, for 2D electrostatics, a point charge q located at z
gives a potential V (w) = −q ln |z − w|2 , and the method of images is fabulously helpful when
we have boundaries. Another handy fact to know from complex analysis is that
1 1
∂ ∂¯ ln |z|2 = ∂ = ∂¯ = 2πδ 2 (z, z̄) . (3.7)
z̄ z
So we will be looking for logs.
As usual, the propagators are defined as the difference between the time-ordered product
and the normal-ordered product (c.f. Wick’s Theorem),

hX µ (τ, σ)X ν (τ 0 , σ 0 )i = T [X µ (τ, σ)X ν (τ, σ)]− : [X µ (τ, σ)X ν (τ, σ)] : . (3.8)

Zero modes have to be handled with care.

We need to handle the closed and open string cases differently because they have phys-
ically different boundary conditions. As usual, we do the closed string case first. We work
on the cylinder with coordinates

(z, z̄) = e2πi(τ −σ)/` , e2πi(τ +σ)/`

(3.9)

Notice that z̄ is not the complex conjugate of z, because we are currently still working
in Lorentzian signature. If we were to Wick rotate, then z̄ would indeed be the complex
conjugate. (Technically, this is why we chose z, z̄ this way.)
Suppose that we re-use our classical mode expansions from the previous section but
with the mode coefficients promoted to operators. However, we cannot honestly write such
an expression without specifying exactly what to do with the zero modes at the quantum
operator level. If they are not chosen correctly, then the propagators for left- and right-
handed field modes fail to separate, which is a necessary condition for constructing the

22
heterotic string among other things. The correct choice for zero modes is

[xµR , pνR ] = iη µν = [xµL , pνL ] , [xµL , pνR ] = 0 = [xµR , pνL ] . (3.10)

Then the propagators become

α0 µν
hXRµ (z)XRν (w)i = − η ln(z − w) ,
20
α
hXLµ (z)XLν (w)i = − η µν ln(z̄ − w̄) . (3.11)
2
The expressions for the open string are more intricate. The worldsheet is now not a
cylinder but a strip of width `, and so the relevant worldsheet coordinates are

(z, z̄) = eiπ(τ −σ)/` , eiπ(τ +σ)/` .

(3.12)

The propagators are

α0 µν 
hX µ (z, z̄)X ν (w, w̄)iNN,DD = − η ln |z − w|2 ± ln |z − w̄|2 ,

2 " √ √  √ √ #
α0 µν  z − w 2  z − w̄ 2
hX µ (z, z̄)X ν (w, w̄)iND,DN = − η ln  √ √  ± ln  √ √  . (3.13)
2 z + w z + w̄ 

If you expand out the |. . .|2 pieces, you will be able to see that these expressions are both
manifestly symmetric under interchange of (z, z̄) ↔ (w, w̄). The ± signs are delicately ar-
ranged in order to ensure satisfaction of the boundary conditions, e.g. that the DD propagator
vanishes at the endpoints where z = z̄.
So, how about those famous reparametrization constraints? What good will they do
us here? Well, we know that classically they enforce T±± = 0, or in Fourier language,
Ln = L̄n = 0. But quantum mechanically we cannot be careless about operator ordering in
defining composite operators like the Virasoro generators. By inspection you can see that
the only Virasoro operator which has such an ambiguity is the zeroth mode L0 . We will see
this represented in equations as
L0 → L0 + a , (3.14)
where a is the normal ordering constant. This seemingly unimportant constant has physical
ramifications: as we will see, it affects the spectrum of excitations of the string. (It does not,
however, meddle with the angular momentum operators involved in the Poincaré algebra.)
After a bit of drudgery (see BLT §3.5), it can be proven that the Virasoro algebra for the
quantum closed bosonic string has what is called a central extension
c
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n , (3.15)
12
where c is called the central charge. For one free boson, c = 1. Translating back to T±± , this
implies
iπc  3
[T++ (σ), T++ (σ 0 )] = −2πi [∂σ δ(σ − σ 0 )] {T++ (σ) + T++ (σ 0 )} − ∂σ δ(σ − σ 0 ) ,

6
23
 iπc  3
[T−− (σ), T−− (σ 0 )] = +2πi [∂σ δ(σ − σ 0 )] {T−− (σ) + T−− (σ 0 )} + ∂σ δ(σ − σ 0 ) ,

6
[T++ (σ), T−− (σ 0 )] = 0 . (3.16)

Now we can finally address the question of whether we can get rid of the negative norm
states. We want to work by analogy with electromagnetism, where one can only impose the
positive frequency part of the gauge condition (e.g. ∂ · A = 0) on physical states, which
ensures that longitudinal/scalar photons decouple. Here, the analogy would be to consider
imposing only that the positive-n Virasoro modes annihilate physical states

Ln |physi = 0 , n > 0. (3.17)

This is not a silly option to consider because, by the centrally extended Virasoro algebra,
the Ln with n positive (i.e. not including n = 0) form a closed subalgebra. It also avoids the
problem that
c
hφ|[Ln , L−n ]|φi = 2nhφ|L0 |φi + n(n2 − 1)hφ|φi , (3.18)
12
i.e., that we cannot impose Ln |φi = 0 ∀n and stay consistent with the symmetry algebra.
Motivated by the above equation, we can try

Ln |physi = 0 ,
(L0 + a) |physi = 0 . (3.19)

Similar equations hold for the barred (left-moving) sector. For strings propagating on
flat Minkowski spacetime (no winding), the momentum constraint that we first discussed for
the classical bosonic string imposes

L0 = L̄0 , a = ā . (3.20)

The last thing to do in this section is to work out the mass formula for open and closed
strings. This comes about from inspecting the definition of the number operator closely.
For the open string,
∞
X X
µ i a


Nopen = α−n αµ,n + α−n αi,n + α−r αa,r (3.21)
n=1 r∈N0 +1/2

where the indices µ enumerate NN directions, the i do DD directions, and the a (not to be
confused with the normal ordering constant!) cover the DN and ND directions. Using the
pair of physical constraint conditions above, the mass formula is found to be

α0 m2open = Nopen + α0 T (∆x)2 + a , (3.22)

where (∆x)2 = ∆xi ∆xi is the distance between the two Dirichlet endpoints and T is the
string tension. The important physics in this equation has two parts: (1) the zero of energy is
determined by the normal ordering constant; (2) the |∆x| dependent piece is just describing
the energy cost of extending the body of a straight (unexcited) open DD string a distance
∆x.

24
 For the closed string, since L0 = ∞ 2 0 2
P
n=1 α−n · αn + α0 /2 = N + α p /4 and similarly for
the barred sector,
α0 m2closed = 2(N + N̄ + 2a) , (3.23)
while momentum balance requires
N = N̄ . (3.24)
A spurious state |spuri is one that obeys the mass shell condition and is orthogonal to
all physical states: hspur|physi = 0.
How can we find the normal-ordering constant from insisting that unphysical states
decouple? Consider a class of zero-norm spurious states of the form
∞
X
|ψi = Ln |χn i (3.25)
n=1

where

Ln |χn i = 0
(L0 + n + a)|χn i = 0 . (3.26)

By the Virasoro algebra, any such state can actually be rewritten as a linear combination

|ψi = L−1 |χ1 i + L−2 |χ2 i (3.27)

because we have relations like [L−1 , L−2 ] = +1L−3 . Now, suppose that |ψi = L−1 |χ1 i. If
this is a physical state, then by the Virasoro algebra L1 L−1 = 2L0 + L−1 L1 , so that

L1 |ψi = L1 L−1 |χ1 i = (2L0 + L−1 L1 )|χ1 i = −2(a + 1)|χ1 i = 0 if a = −1 . (3.28)

This is why we insist that the normal ordering constant appearing in the string mass spectrum
formula takes the value a = −1.
How about finding D? One way to see this is to consider zero-norm spurious states of
the form
|ψi = (L−2 + γL2−1 )|χ̃i (3.29)
where |χ̃i satisifes (L0 + 1)|χ̃i = 0 = Lm |χ̃i, for positive m. Since |ψi is physical, Ln |ψi = 0
for positive n. By the Virasoro algebra, requiring it to be annihilated by just L1 and L2
is enough, because you can build all the higher Ln , n ≥ 3 from those two. Now let us find
L1 |ψi, which we want to be zero. From the Virasoro algebra, we can find L1 in terms of
other generators in a useful way. Specifically,

[L1 , L−2 + γL2−1 ] = 3L−1 + 2γL0 L−1 + 2γL−1 L0

= 3L−1 + 2γL−1 (L0 + 1) + 2γ(L0 − 1)L−1
= (3 − 2γ)L−1 + 2γL−1 + 2γ(L−1 L0 + L0 L−1 )
= (3 − 2γ)L−1 + 4γL0 L−1 . (3.30)

Then, applying the above commutator to |χ̃i, and using the fact that |ψi = (L−2 + γL2−1 )|χ̃i,
gives
L1 |ψi = [(3 − 2γ)L−1 + 4γL0 L−1 ]|χ̃i . (3.31)

25
Now,
L0 L−1 |χ̃i = L−1 (L0 + 1)|χ̃i = 0 . (3.32)
So we have that L1 annihilates |ψi when γ = 23 . Next, we need to check whether or not |ψi
is also annihilated by L2 . By the Virasoro algebra,

[L2 , L−2 + 32 L2−1 ] = 4L0 + D2 + 32 [(3L1 )L−1 + 3L−1 L1 ]

(3.33)
= 4L0 + 92 (2L0 + 2L−1 L1 ) (3.34)
D
= 13L0 + 9L−1 L1 + 2
. (3.35)

Using the same technique as above, this gives

L2 |ψi = L2 (L−2 + 32 L2−1 )|χ̃i (3.36)

= L2 L−2 |χ̃i + 3
L L2 |χi
2 2 −1
(3.37)
= [(L−2 + 23 L2−1 )L2 + (13L0 + 9L−1 L1 + D
2
)]|χ̃i (3.38)
D
= (−13 + 2
)|χ̃i . (3.39)

Clearly, this requires D = 26. The conclusion overall is that unphysical states decouple when
a = 1 and D = 26.

3.2 Light cone quantization

Light-cone coordinates select out one spatial direction as special: X ± = √12 (X 0 ± X 1 ). The
Minkowski metric becomes off-diagonal: a · b = −a+ b− − a− b+ + ai bi , where i = 2 . . . D − 1.
The relativistic mass shell relation becomes p− = (m2 + pi pi )/(2p+ ), which makes it look
vaguely non-relativistic. Note that these light-cone indices refer to spacetime directions, not
the worldsheet z = τ − σ and z̄ = τ + σ which give rise to unbarred and barred worldsheet
derivatives and will turn into complex conjugates once we Wick rotate.
Analyzing QFT in light-cone makes counting the degrees of freedom quite straightfor-
ward. It also makes interrogating the string theory spectrum a lot easier for beginners.
Light-cone gauge (more properly called light-front gauge) is a reparametrization gauge fix-
ing in which the X + direction is tied to be proportional to worldsheet time τ . Both the X 0
and X 1 directions must be Neumann if we have open strings around. In BLT conventions,
2πα0 +
X+ = p τ. (3.40)
`
This gauge choice permits the Virasoro constraints to be solved explicitly and the theory can
be formulated in terms of physical (transverse) degrees of freedom only. The price of fixing
the light-cone gauge is that the theory does not automatically have D-dimensional Poincaré
symmetry. It is only in D = 26 that the light-cone gauge Poincaré algebra closes.
Light cone gauge clearly requires that X + does not oscillate. We can solve for X − by
using the constraints (Ẋ µ ± X 0µ )2 = 0,

`
∂± X − = (∂± X i )2 . (3.41)
2πα0 p+

26
This implies that the αn− oscillators are proportional to transverse Virasoro generators
X
L⊥n =
I
αn−p αpI (3.42)
p∈Z

where the I are summed over transverse directions I = 2, . . . , (D − 1). The mass-squared
formula becomes a sum over oscillators that point in transverse directions only. No negative-
norm states appear anywhere. The all-important mass formula takes the same form as for
canonical quantization, with aLC being the zero-point energy (ZPE)
X 1
aLC = p (D − 2) (3.43)
p∈N
2

obtained by using the canonical commutation relations for the X I fields. How are we to
regularize this infinite sum? There are two relatively fast ways to address this. The first
is to take
P∞the −smathematicians seriously when they tell us about Riemann zeta functions
1
ζ(s) = n=1 n . We have s = −1, and ζ(−1) = − 12 , which implies that
 
1 −1 (D − 2)
aLC = (D − 2) =− . (3.44)
2 12 24
Note: for the case of half
P∞integrally −s moded fields, we would need to use instead the generalized
ζ-function ζ(s, q) = n=0 (n + q) . For the zero-point energy, the answer appearing is
1
proportional to ζ(1, q) = − 12 (6q 2 − 6q + 1). When q = 12 , the ZPE is + 24
1 1
rather than − 12 .
This drives home the fact that the ZPE is not just about summing up an infinite number of
modes: it is also about how the fields in question are moded. The second way is to regularize
by putting in a Gaussian exponential tail with UV cutoff Λ,
∞ ∞
πX πX  π n
n → n exp − (3.45)
` n=1 ` n=1 `Λ
π e−π/`Λ
= (3.46)
` (1 − e−π/`Λ )2
` 2 1 π
= Λ − + O(1/Λ) . (3.47)
π 12 `
The Λ2 divergence √is proportional to the length of the string and can be cancelled by in-
1
troducing Λ2 d2 σ h to the Polyakov action. The remaining − 12
R
per integrally moded
I
worldsheet boson X is what sums up the vacuum energies of the oscillators.
You will be working out some details of the low-lying spectrum in HW1.

3.3 Path integral quantization and Fadeev-Popov ghosts

The Feynman path integral (FPI) writes the generating functional for correlation functions
as a sum over all possible paths, with a well-defined measure, weighted by eiS in Lorentzian
signature where S is the action. It is straightforward to write a suitable measure for the
FPI whenever there is no gauge symmetry, like for the ordinary QFT of a spin-0 or spin-1/2

27
field. The presence of a gauge symmetry makes defining the measure considerably more
subtle, because one must ensure that the measure does not overcount physically equivalent
configurations of the quantum fields. The fancy-pants word for the technology to handle this
is Fadeev-Popov ghosts. They come in a variety of incarnations, each set relevant to the
particular gauge symmetry and the particular gauge choice being used. They make use of a
basic fact about functional determinants, a topic for which we now give a lightning review.
Consider a Gaussian integral involving commuting scalars of the form
YZ
Ib = dξk exp (−ξi Bij ξj ) , (3.48)
k

where B is a symmetric matrix with eigenvalues bi . Diagonalize B via orthogonal matrix O,

and switch to xi variables defined by ξi = Oij xj . Then
YZ YZ X YZ X
2
Ib = dξk exp(−ξi Bij ξj ) = dξk exp(− bi x i ) = dxi exp(− bi x2i )
k k i i i
Yrπ 1
= = const. × p
i
b i det(B)
In other words,
YZ
Ib = dξk e−ξi Bij ξj = (const.)(det(B))−1/2 (3.49)
k

For a spin-0 scalar field, we can write the free action as S0 [φ] = 12 dD x [φ (−∂ 2 − m2 ) φ].
R

Formally we can write B = (−∂ 2 − m2 I) as a matrix. Therefore Zφ = Dφ exp (iS0 [φ])

R
−1/2
becomes Zφ = (const) · [det(m2 + ∂ 2 )] .
Now let us do something similar for fermion fields. In path integrals, these are represented
via Grassmann variables, which anticommute. For any two anticommuting Grassmann
variables θ, η we have {θ, η} = 0. In other words, θη = −ηθ. Notice that a fermion
bilinear is again a boson. Since the above equation holds for any θ, η, it holds in particular
when θ = η, i.e. θ2 ≡ 0. This mathematizes the Pauli principle. It also makes Taylor
expansions splendidly easy, because each such Taylor series terminates after the linear piece:
f (θ) = A + RBθ, where A, BR are constants. The definition of integrationRfor Grassmann
R fields
is given by dθ = 0 and dθθ = 1, and the sign convention R we use is dθ dη · η · θ = +1.
∗
For complex Grassmann variables, (θη) = η θ . Also, dθ dθ θ θ∗ = +1. Now, because
∗ ∗ ∗

Taylor expansions truncate so early owing to the anticommutation property of Grassmann

fields, we have
exp (−θ∗ bθ) = 1 − θ∗ bθ + 0 (3.50)
Therefore, Z Z
∗ −θ∗ bθ
dθ dθe = dθ∗ dθ (1 − θ∗ bθ) = b (3.51)

by the rules of Grassmann integration and the anticommuting property. In doing Feynman
path integrals to find correlation functions of physical fields, we would insert fields into the
integrand. So let us consider for instance
Z Z Z
∗ ∗ −θ∗ bθ
dθ dθ θ θ e = dθ dθ θ θ (1 − θ bθ) = dθ∗ dθ θ θ∗ = 1
∗ ∗ ∗
(3.52)

28
Compare this to the result we had obtained just above. Looking carefully, we notice that the
Gaussian integral with θθ∗ in the integrand brings down an extra factor of (1/b) compared
to the case without. So let us define the following integral for an invertible matrix Bij ,
YZ
If = dθi∗ dθi exp (−θi∗ Bij θj ) . (3.53)
i

In the diagonal basis, Bij has eigenvalues {bi }. Then

YZ X Y
If = dθi∗ dθi exp(− θi∗ bi θi ) = bi = det (B) (3.54)
i i i

We will make use of a functional version of this shortly to convert a determinant that
will arise in the FPI measure into a covariant path integral over (scalar) Grassmann fields
called Fadeev-Popov ghosts. The ghost action will enforce the rule that the measure does
not overcount fields related by the symmetries of the Polyakov action. Note: for our first
look at ghosts and measures in the FPI, we will ignore the issue of ghost zero modes. We will
have more to say about this subtle and important physics question later when we analyze
the superstring.
What gauge choice do we want to enforce? We use the conformal gauge which puts
the conformal symmetry on the worldsheet front and centre. Earlier we mentioned that
on-shell, the induced metric gµν ∂α X µ ∂β X ν is equal to the intrinsic worldsheet metric hαβ .
This is true because of the classical equation of motion. Quantum mechanically, we have to
do better – we need to integrate over all paths and over different hαβ and embeddings X µ .
The important feature we need in the measure is that it respects reparametrizations and
Weyl rescalings. Classically this symmetry was enough to gauge away the three independent
components of hαβ . But will this still be true quantum mechanically? The answer will turn
out to be: only for specific string theories referred to as critical string theories which
have total central charge zero, counting the X µ as well as the (b, c) ghosts. It is possible to
define noncritical string theories as well, at the price of including an extra field in the action
called the Liouville field.
For our first try, we will write the string theory path integral as
Z
Z = DhDXeiSP [h,X] , (3.55)

where the norms are defined via

√
Z
||δX|| = d2 σ −h δX µ δXµ (3.56)
√
Z
||δh|| = d2 σ −h hαβ hγδ δhαγ δhβδ . (3.57)

We can use the reparametrization invariance of the Polyakov action to go to a gauge where
the intrinsic worldsheet metric is
hαβ = e2φ ĥαβ . (3.58)

29
Further, under reparametrizations and Weyl rescalings the changes of the metric can be
decomposed into a traceless symmetric tensor part and a trace part (orthogonal w.r.t. the
norms defined above),
δhαβ = −(Pξ)αβ + 2Λ̃hαβ , (3.59)
where P maps vectors to traceless symmetric tensors. To see this, note that reparametriza-
tion ensures that we can pick conformal gauge locally, with the conformal factor being e2φ(τ,σ) .
But can it be done globally? Under a reparametrization,
δhαβ = −(∇α ξβ + ∇β ξα ) + 2Λhαβ (3.60)
≡ −(Pξ)αβ + 2Λ̃hαβ (3.61)
where Λ̃ = Λ − 21 ∇γ ξ γ , and where (Pξ)αβ = ∇α ξβ + ∇β ξα − (∇γ ξ γ )hαβ . The trace part
can always be cancelled by a suitable choice of Λ. So for conformally gauge to be allowed
globally, there must exist a globally defined vector field ξ α such that (Pξ)αβ = tαβ for
arbitrary symmetric traceless tαβ . The adjoint P† maps symmetric traceless tensors to
vectors via (P † ξ)α = −2∇β tαβ . Whether or not the zero mode subtleties allow a consistent
choice of conformal gauge will be mentioned further when we get to superstrings. For now,
it suffices to suppress this detail.
Getting back to our path integral measure, we have then
 
 ∂(Pξ, Λ̃ 
Dh = D(Pξ)D(Λ̃) = DξDΛ  . (3.62)
 
 ∂(ξ, Λ) 

This gives a Jacobian of the form

  
det P 0  = | det P| = (det P† P)1/2 .
 
(3.63)
 ∗ 1 
The integral over reparametrizations gives the volume of the part of the diffeomorphism
group connected to the identity. This volume depends on the Weyl degree of freedom as
does Dξ. But in the critical dimension this all drops out. We assume critical dimension, and
write for our second try
Z
2φ µ
Z = DX µ (det P † P) eiSP [e ĥαβ ,X ] (3.64)

This is almost suitable for use, but at present it has a nasty functional determinant in
the integrand. Now we can bring in our Grassmann-functional-determinants trick and use
Fadeev-Popov ghost fields c(τ, σ), b(τ, σ) to write
√
Z  Z 
1 2 αβ γ
(det P † P) = Dc Db exp d σ −hh bβγ ∇α c , (3.65)
2π

where hαβ = e2φ ĥαβ . In physical language, the ghost cα corresponds to infinitesimal reparametriza-
tions, and the antighost bαβ corresponds to variations perpendicular to the gauge slice. Then
our final expression for the Feynman Path Integral for strings is
Z
µ
Z = DX µ Dc Db eiSP [X ,ĥ,b,c] , (3.66)

30
where
Z q
1  
ˆ α cγ .
µ
SP [X , ĥ, b, c] = − d σ −ĥĥαβ ∂α X µ ∂β Xµ + 2iα0 bβγ ∇
2
(3.67)
4πα0
Note that the Weyl factor is gone because we are working in the critical dimension. In
conformal gauge, ĥαβ = ηαβ ,
Z
i
d2 σ c+ ∂− b++ + c− ∂+ b−− ,


Sgh = (3.68)
π

which is real when b, c are Hermitean by dint of the anticommuting property. The (b, c)
system has stress-energy tensor

Tαβ = −i bαγ ∇β cγ + bβγ ∇α cγ − cγ ∇γ bαβ − hαβ bγδ ∇γ cδ .

(3.69)

It is important to recall that bαβ is symmetric and traceless while deriving this. Then, the
ghost stress-energy tensor becomes in conformal gauge

T±± = −i 2b±± ∂± c± + (∂± b±± )c± .

(3.70)

The equations of motion for the ghosts are

∂− b++ = 0 = ∂+ b−− , ∂+ c− = 0 = ∂− c+ , (3.71)

and both fields are taken to have periodic boundary conditions for the closed string, e.g.
b(σ + `) = b(σ). For the open string, at the endpoints we obtain b++ = b−− and c+ = c− .
These relations are generically not true anywhere else on the string. Physically, since ghosts
are Grassmann, they obey canonical anticommutation relations,

{b±± (τ, σ), c± (τ, σ 0 )} = 2πδ(σ − σ 0 ) . (3.72)

The b and c mode expansions show integrally moded fields, and Virasoro operators become
+∞
X ◦ ◦
Lm = (m − n) ◦ bm+n c−n ◦ (3.73)
m=−∞

The weird-looking normal ordering symbols are designed to stand out: they tell you that
in putting annihilation operators to the right the Grassmann property is taken into account
properly. Hermiticity of the ghosts implies that

cn = c†−n , bn = b†−n , Ln = L†−n . (3.74)

Using zeta function regularization as before, it is straightforward to obtain

cgh = −26 . (3.75)

31
4 Conformal Field Theory
4.1 Radial quantization
To start with we will discuss several definitions. The first is how we got from our world sheet
coordinates to the complex conformal coordinates we were using. First recall our world sheet
coordinates can be parametrized as σ ∈ [0, 2π] and τ ∈ [−∞, ∞]. Now Wick rotate to a
complex space (i.e. τ → −iτ ). To get to the complex plane, we simply conformal transform
by
z = eτ −iσ z̄ = eτ +iσ (4.1)
Pictorially,

Now we can see how the worldsheet time is mapped. For τ = −∞, we see that z = 0 and for
τ = ∞, we see that z = ∞. Furthermore, lines of constant τ are circles around the origin.
So to define a time ordered product, we now do a radially ordered product. This will be
important in our definitions of operator product expansions (OPEs), since we will naturally
want to time order them for physics reasons. σ-translations on the cylinder become rotations
on the complex plane, while τ -translations on the cylinder become dilatations.
Recall that holomorphic functions are functions which are complex and infinitely differ-
entiable everywhere. We use the same definition in CFT. Functions that are holomorphic
in z are called such, but holomorphic functions in z̄ are called anti-holomorphic. There are
several different types of fields, the most basic of which is called a primary field. Primary
field transform as tensors under conformal transformations:
 0 h  0 h̄
0 ∂z ∂ z̄
φ(z, z̄) → φ (z, z̄) = φ(z 0 , z̄ 0 ). (4.2)
∂z ∂ z̄

Note we call h and h̄ conformal weights under analytic and anti-analytic transforms, respec-
tively. The scaling dimension of a field is given by h + h̄ and the conformal spin is given by
h − h̄. These indicate how field behave under dilations or rotations. Fields that are single
valued will have h = h̄ ∈ Z. Holomorphic fields will have h̄ = 0 and anti-holomorphic fields
will have h = 0. If we introduce a infinitesimal conformal transformation

z 0 = z + ξ(z) ¯
z̄ 0 = z̄ + ξ(z̄), (4.3)

then the field will transform as

φ0 = φ + δξ,ξ̄ φ, (4.4)
where
δξ,ξ̄ φ = (h∂z ξ + h̄∂z̄ ξ¯ + ξ∂z + ξ∂
¯ z̄ )φ (4.5)

32
Using the relation between the cylinder and plane coordinates, we see quickly that
X
φplane = φn z −n−h . (4.6)
n∈Z

One final note is about the Hermitian conjugate of a field. The general definition for a field
of weight h is
1 1
[φ(z)]† = φ† ( ) 2h . (4.7)
z̄ z̄
and this implies that the modes satisfy

†
φ†


−n
= φn (4.8)

How does T act on primary fields? As we saw, the Virasoro algebra leads to an infinite
number of conserved charges I
dz
Tξ = ξ(z)T (z) . (4.9)
2πi
These can be used to generate the infinitesimal conformal transformations

z → z 0 = z + ξ(z). (4.10)

To generate the conformal transforms on primary fields, we use the commutator,

δξ φ(w) = [Tξ , φ(w)]. (4.11)

Now we wish to map the two definitions of the transformations. Suppose we have a field
with h̄ = 0. Then the above definition gives us (recalling radial ordering)
I
dz
δξ φ(w) = ξ(z)T (z)φ(w). (4.12)
Cw 2πi

The contour is

On the other hand, including (4.5), we see that

I
dz
ξ(z)T (z)φ(w) = (h∂w ξ + ξ∂w )φ. (4.13)
Cw 2πi

Now let us propose that the time ordered product takes the form

hφ(w) ∂w φ(w)
T (z)φ(w) = 2
+ + regular , (4.14)
(z − w) z−w

33
then we can see that
I I  
dz dz hφ(w) ∂w φ(w)
ξ(z)T (z)φ(w) = ξ(z) + + regular (4.15)
Cw 2πi Cw 2πi (z − w)2 z−w
 
d 2 ξ(z) ∂w φ(w)
= lim hφ(w) (z − w) + (z − w) (4.16)
z→w dz (z − w)2 z−w
= (h∂w ξ + ξ∂w )φ (4.17)

So our proposition was correct. Goodie!

Now we might expect to be able to expand T similarly. In order to determine the correct
expansion, we need to have some constraint for the product T (z)T (w) to satisfy. To start
with consider the commutator of two infinitesimal conformal transformations,

[δξ1 , δξ2 ]φ = δξ1 δξ2 φ − δξ1 δξ2 φ. (4.18)

Using our definition for the transformations,

[δξ1 , δξ2 ]φ = δξ1 δξ2 φ − δξ1 δξ2 φ (4.19)

= (h∂z ξ1 + ξ1 ∂z )(h∂z ξ2 + ξ2 ∂z )φ − (h∂z ξ2 + ξ2 ∂z )(h∂z ξ1 + ξ1 ∂z )φ (4.20)
= (h(ξ1 ∂ 2 ξ2 − ξ2 ∂ 2 ξ1 ) + (ξ1 ∂ξ2 − ξ2 ∂ξ1 )∂)φ (4.21)
= (h∂(ξ1 ∂ξ2 − ξ2 ∂ξ1 ) + (ξ1 ∂ξ2 − ξ2 ∂ξ1 )∂)φ (4.22)
= δ(ξ1 ∂ξ2 −ξ2 ∂ξ1 ) φ. (4.23)

This implies that

[δξ1 , δξ2 ] = δ(ξ1 ∂ξ2 −ξ2 ∂ξ1 ) . (4.24)
So now we can consider our other definition,

[δξ1 , δξ2 ]φ = δξ1 δξ2 φ − δξ1 δξ2 φ (4.25)

= [Tξ1 , [Tξ2 , φ]] − [Tξ2 , [Tξ1 , φ]] (4.26)
= Tξ1 Tξ2 φ − Tξ2 φTξ1 − Tξ1 φTξ2 + φTξ2 Tξ1 − 1 ↔ 2 (4.27)
= Tξ1 Tξ2 φ + φTξ2 Tξ1 − Tξ2 Tξ1 φ − φTξ1 Tξ2 (4.28)
= [[Tξ1 , Tξ2 ], φ]. (4.29)

Additionally, we can see that

δ(ξ1 ∂ξ2 −ξ2 ∂ξ1 ) φ = [T(ξ1 ∂ξ2 −ξ2 ∂ξ1 ) , φ]. (4.30)

Now we apply our condition, equation 4.24, and we will get

[Tξ1 , Tξ2 ] = T(ξ1 ∂ξ2 −ξ2 ∂ξ1 ) . (4.31)

This is the condition that T (z) must satisfy. Now we propose the following form for the
T − T operator product expansion,
c/2 2T (w) ∂w T (w)
T (z)T (w) = 4
+ 2
+ + regular. (4.32)
(z − w) (z − w) z−w

34
We will work on the right side of our condition,
I I
dw dz
[Tξ1 , Tξ2 ] = ξ1 (z)T (z)ξ2 (w)T (w) (4.33)
2πi Cw 2πi
I I  
dw dz c/2 2T (w) ∂w T (w)
= ξ1 (z)ξ2 (w) + + + regular (4.34)
2πi Cw 2πi (z − w)4 (z − w)2 z−w
I  
dw c 3
= ξ2 (w) ∂ ξ1 (w) + 2∂w ξ1 (w)T (w) + ξ1 (w)∂w T (w) . (4.35)
2πi 12 w

Note that the first term has no singularities (we said both ξ1 ,ξ2 were both analytical). So
the contour vanishes. Also note that we can integrate the last term by parts. Since it is a
contour integral, the surface term vanishes. Doing both theses things we will find
I  
dw c 3
[Tξ1 , Tξ2 ] = ξ2 (w) ∂ ξ1 (w) + 2∂w ξ1 (w)T (w) + ξ1 (w)∂w T (w) (4.36)
2πi 12 w
I  
dw
= 2ξ2 (w)∂w ξ1 (w)T (w) − ∂w (ξ2 (w)ξ1 (w))T (w) (4.37)
2πi
I
dw

= ξ2 (w)∂w ξ1 (w) − ξ1 (w)∂w ξ2 (w) T (w) (4.38)
2πi
= T(ξ1 ∂ξ2 −ξ2 ∂ξ1 ) . (4.39)

So we have shown that our proposal for the OPE was correct. Note that we did not need
to have the term proportional to c. However, it does have the correct scaling dimension and
Bose symmetry. We also recall that this is the term that comes from the Weyl anomaly,
which we needed to include.
Let us look at the operator product expansion of Tzz (and by extension Tz̄z̄ ). We will
expand the Laurent series,
∞
X
Tzz = z −n−2 Ln (4.40)
−∞

Like before, we will cut down on the pain of presentation by proposing a form for this T − T
OPE and show that it works. It is
(0) (0)
λ 1 2Tww ∂w Tww
Tzz Tww = 4
+ 2
+ + regular , (4.41)
2 (z − w) (z − w) z−w

where by regular, we mean no singular behaviour as z → w. In order to infer this defining

equation, let us recall that I
dz −k
z = δk,1 . (4.42)
2πi
So then we can see that to find the operators we can perform a contour integral,
I
dz m+1
Ln = z T =zz (4.43)
2πi

35
So we can turn our expression for the product of two energy momentum operators into an
equation for the commutator that these Ln operators satisfy. Doing so we find that
I I
dz dw n+1 m+1
[Ln , Lm ] = z w Tzz Tww . (4.44)
2πi 2πi
We can now plug in our equation for the energy momentum components. We can drop out
the regular terms since they will integrate away. Then we have
!
I I (0) (0)
dw dz n+1 m+1 λ 1 2Tww ∂w Tww
[Ln , Lm ] = z w + + . (4.45)
2πi Cw 2πi 2 (z − w)4 (z − w)2 z−w

Examining the first term alone,

1 d3 z n+1
I I I
dw dz n+1 m+1 λ 1 dw m+1 4λ
z w = w lim (z − w) (4.46)
2πi Cw 2πi 2 (z − w)4 2πi z→w 3! dz 3 2 (z − w)4
I
λ dw m+1 n−2
= (n + 1)n(n − 1) w w (4.47)
12 2πi
λ
= n(n2 − 1)δn+m,0 . (4.48)
12
For the second two terms, we can use the expansion we had earlier. Putting that in, it
becomes
I I
dw dz n+1 m+1 2Tww ∂w Tww
z w ( 2
+ ) (4.49)
2πi Cw 2πi (z − w) z−w
dz n+1 m+1 X 2w−p−2 w−p−3
I I  
dw
= z w − (p + 2) Lp (4.50)
2πi Cw 2πi p∈Z
(z − w)2 z−w
z n+1 z n+1
X I dw  
m+1 −p−2 d 2 −p−3
= Lp w lim 2w (z − w) − (p + 2)w (z − w)
p∈Z
2πi z→w dz (z − w)2 z−w
(4.51)
I
X dw m+n−p−1
= Lp w (2(n + 1) − (p + 2)) (4.52)
p∈Z
2πi
X
= Lp (2n − p)δm+n,p (4.53)
p∈Z

= (n − m)Ln+m . (4.54)

So putting it all together, the operators satisfy the commutator,

λ
[Ln , Lm ] = (n − m)Ln+m + n(n2 − 1)δn+m,0 . (4.55)
12
We can now recognize that this is the Virasoro algebra! Everything works in an exactly analo-
gous way for the barred sector; the two Virasoro algebras are independent (i.e. [L̄n , Lm ] = 0).

36
4.2 The operator product expansion
So this brings us to our more general idea of a product expansion of two operators. Suppose
that {Oi } are a complete set of local operators with definite scaling dimensions. We can then
expand a product of these two operators in terms of the others. Since should have scaling
dimensions (since the product must have a definite scaling dimension), we expand as
X
Oi (z)Oj (w) = Cijk (z − w)hk −hi −hk Ok (w). (4.56)
k

It is important to note that the (anti)commumators of the operators are determined by the
singularities in the operator product expansion.
Of course, there are still other properties of the Virasoro algebra to learn. Consider the
commutator of Ln with a field φ
I
dw n+1
[Ln , φ(z)] = w [T (w), φ(z)] (4.57)
2πi
I
dw n+1
= w T (w)φ(z) (4.58)
Cz 2πi
= [Tzn+1 , φ] (4.59)
= δzn+1 φ (4.60)
n
= z [z∂ + (n + 1)h]φ. (4.61)

To see what happens for the modes, we simply use our contours,
I
dz m+h−1
[Ln , φm ] = z [Ln , φ(z)] (4.62)
2πi
I
dz m+h−1 n
= z z [z∂ + (n + 1)h]φ(z) (4.63)
2πi
I
dz m+n+h−1
= z [z∂ + (n + 1)h]φ(z) (4.64)
2πi
I
dz m+n+h
= (n + 1)hφm+n − ∂z φ(z) (4.65)
2πi
= ((n + 1)h − (m + n + h))φm+n (4.66)
= (n(h − 1) − m)φm+n . (4.67)

So we see that Ln are generators of conformal transformations with ξ = z n+1 . A important

note is that L−1 ,L0 ,L1 are the generators of SL(2, R), the maximal closed subalgebra of
the conformal group. They have finite transformations, called fractional linear or Möbius
transformations, of the form
az + b
z → z0 = (4.68)
cz + d
with the coefficients satisfying  
a b
∈ SL(2, R). (4.69)
c d

37
If we add L̄−1 ,L̄0 ,L̄1 , then we get SL(2, C).
We could also consider the effect of a finite conformal transformation, z → f (z), on T (z).
We can use our OPE for T (z)T (w) to find the infinitesimal transformation of T (z) and then
integrating the result. We obtain
c
T (z) → T 0 (z) = (∂f (z))2 T (f (z)) + D(f )z . (4.70)
12
Note that D(f ) is the Schwarzian derivative, which is given by
∂f (z)∂ 3 f (z) − (3/2)(∂ 2 f (z))2
D(f )z = . (4.71)
(∂f (z))2
The Schwarzian derivative is unique in that it is the only weight two object that satisfies the
properties below
az + b
D(f )z = 0 ⇔ f (z) = (4.72)
cz + d
af + b
D( )z = D(f )z (4.73)
cf + d
D(f )z = (∂z g)2 D(f )g + D(g)z . (4.74)

To get the energy momentum tensor from the cylinder to the plane, we can use equation
0
4.70 on z 0 → ez = z and get
c
Tcylinder (z 0 ) = z 2 Tplane (z) − . (4.75)
24

4.3 Action of Virasoro generators on states: highest weights and

descendants
Now let us discuss the action of the Ln on states in the Hilbert space. As usual, |0i, h0| will
be the in and out vacua respectively. Now we wish for the energy momentum tensor to be
regular at z = 0 and z = ∞. For z = 0, then we just need to make sure that the Virasoro
algebras that are generate regular conformal transformations at the origin annihilate the
vacuum. This is done by

Ln |0i = 0 for n ≥ −1 (4.76)

For z = ∞, we need to bring in the point at infinity to the origin. We do this using the map
w = −1/z. Using our transformation law form equation 4.70,

T 0 (w) = (1/w2 )2 T (−1/w). (4.77)

So then Pwe can expand out our energy momentum tensor in these coordinates, finding
T 0 (w) = (−w)n−2 Ln . So the Virasoro operators that generate regular conformal transfor-
mations at the origin are different than in the other coordinates. To properly regularize the
energy momentum, we need to enforce

h0|Ln = 0 for n ≤ 1 (4.78)

38
Another way of coming to the same result is to notice that these are the Hermitian conjugates
of one another. We can argue similarly for primary fields of weight h. Here the conditions
become

φn |0i = 0 for n ≥ 1 − h (4.79)

h0|φn = 0 for n ≤ h − 1. (4.80)

One might be afraid that if h ≤ 0, then there might be non-zero vacuum expectation values.
However, unitarity restricts us to states that are h ≥ 0. Note that h = 0 corresponds
to the identity from unitarity and so its mode expansion is itself. However, ghost fields
can avoid unitarity restrictions since on-shell ghosts are always countered by non-physical
polarizations.
Now we can construct some other states. To define an in state, we wish for it to be
defined at the origin (i.e. τ = −∞). So then

|φj i = lim φj (z)|0i (4.81)

z→0
X
= lim z −n−hj φn |0i (4.82)
z→0
n∈Z
X
= lim z −n−hj φn |0i (4.83)
z→0
n≤−hj
X
= δn,−hj φn |0i (4.84)
n≤−hj

= φ−hj |0i. (4.85)

This might have seemed trivial, but we have used our ideas of expanding and our requirements
of regularity. The out state can be defined as we did out states earlier, or we can just take
the Hermitian conjugate,
hφj | = h0|(φ† )hj . (4.86)
Now we wish to know how our Virasoro operators act on our states. For the moment, we
will consider only n ≥ 0. We make use of equation 4.67 and can see that

Ln |φj i = Ln φ−hj |0i (4.87)

= φ−hj Ln |0i + (n(hj − 1) + hj )φn−hj |0i. (4.88)

Note for n ≥ 0, the first term is annihilated. The second term is annihilated for any n > 0
but not if n = 0. So the result is

L0 |φj i = hj |φj i (4.89)

Ln |φj i = 0 for n > 0. (4.90)

Now consider the following, using the Virasoro commutator,

λ
L0 (L−n |φj i) = L−n L0 |φj i) + [(0 − (−n))L0−n + 0(02 − 1)δ−n,0 ]|φj i (4.91)
12
= (hj + n)L−n |φj i. (4.92)

39
So we see that the Virasoro operators L−n act as raising operators for n > 0. Note that
states that satisfy equations 4.89 and 4.90 are called highest weight states. (Even though
one can raise their weight higher using L−n : it is a bit weird.) Highest weight states with
different eigenvalues of L0 (hj ) are orthogonal. We can show this noticing that

hφi |L0 |φj i = hj hφi |φj i (4.93)

but if we act on the out state (since L†n = L−n , L0 is Hermitian), then

hφi |L0 |φj i = hi hφi |φj i (4.94)

and for both equations to hold true for different h, the states must be orthogonal.
We expect that we can now complete the Hilbert space by using our raising operators.
The resulting states are called descendant states (which are distinct from the primary fields
we discussed earlier). For every highest weight state |φj i, we can find a representation of the
Virasoro algebra labelled by hj . It is called a Verma module and it consists of all the states
that have the form of
|φkj 1 ...km i = L−k1 ...L−km |φj i. (4.95)
P
Note this only makes sense if ki > 0. These states will have L0 eigenvalues of hj + i ki .
Note that these states are secondary operators, but are contained in the operator product
of the primary field with the energy momentum tensor:
∞
(−k)
X
T (z)φ(w) = (z − w)−2+k φi (w). (4.96)
k=0

In other words,
I
dz
φ−k
i (w) = (z − w)1−k T (z)φi (w) ≡ L̂−k φi (w). (4.97)
Cw 2πi
(−k ) (−k ,−k )
We could then consider T (z)φi 1 (w) and get fields like φi 1 2 (w) and so on. So we
generalize to the descendant field that creates the state in equation 4.95,

φkj 1 ...km (z) = L̂−k1 ...L̂−km φj (z). (4.98)

These fields constitute the conformal family of [φj ]. Note that from this definition we can
see that T is not a primary field, rather it is in the conformal family of the identity operator
[I]. To see this, consider
I
(−2) dw
I (z) = L̂−2 I(z) = (w − z)−1 T (w)I(z) = T (z). (4.99)
2πi
One final note about this. We can define the character of a conformal family (partition
function) as
∞
c
Y
hj − 24
Chj (τ ) = q (1 − q n )−1 , (4.100)
n=1
2πiτ
where q = e . More details can be found in BLT.

40
4.4 Correlation functions
Now that we have our definitions of fields and operators sorted, we would like to investigate
how the correlation functions behave. To do this we will restrict ourselves to fields called
quasi-primary fields. These are fields that transform as tensors under SL(2, C), but not
necessarily under the full conformal group (i.e. all primary fields are quasi-primary but not
vice versa). There are three important transformations under this group, all given in the
table below.
Transformation Generator Coordinate Field
0 0
Translations L−1 z =z+b φi (z) = φi (z + b)
Dilation and rotations L0 z 0 = az φ0i (z) = ahi φi (az)
1
Special conformal L1 z 0 = cz+1
z
φ0i (z) = ( cz+1 )2hi φi ( cz+1
z
)

Demanding invariance under these restricts the forms that are allowed for the correlation
functions. See BLT for details. The results are as follows,

hφi (z)i = 0 (4.101)

Cij
hφi (z1 )φj (z2 )i = δ
2hi hi ,hj
(4.102)
z12
Cijk
hφi (z1 )φj (z2 )φk (z3 )i = hi +hj −hk hi +hk −hj hk +hj −hi
(4.103)
z12 z13 z23
h2 +h4 h1 +h3
z z24 z12 z34
hφ1 (z1 )φ2 (z2 )φ3 (z3 )φ4 (z4 )i = h1 +h2 13 h2 +h3 h3 +h4 h1 +h4
f( ) (4.104)
z12 z23 z34 z14 z13 z24
Y −γ zij zkl
hφ1 (z1 )...φn (zn )i = zij ij f ( ) (4.105)
i<j
z il zkj
n
X hj ∂wj
hT (z)φ1 (w1 )...φn (wn )i = ( 2
+ )hφ1 (w1 )...φn (wn )i, (4.106)
j=1
(z − w j ) z − w j

where zij = zi − zj . Note that both Cij and Cijk are constants, and f is a function of the
(n − 3) independent cross ratios which cannot be specified further. Additionally, γij = γji
are solutions to X
γij = 2hi . (4.107)
j6=i

Note that the four-point amplitudes can help us put constraints on Cijk by a method of the
conformal bootstrap.

41
5 Ghosts, BRST invariance, and scattering amplitudes
5.1 Properties of (b,c) ghosts and the critical dimension
The equations of motion are simple first order differential equations which one could just
read off. They imply that

cz = c(z), bzz = b(z), (5.1)

cz̄ = c̄(z̄), bz̄z̄ = b̄(z̄). (5.2)

Now the tensor character of bαβ implies that b transforms with conformal weight (h, h̄) =
(2, 0) and b̄ with (h, h̄) = (0, 2). For the action to remain invariant under conformal trans-
formations, we need c to transform with conformal weight (−1, 0) and c̄ with (0, −1). This
negative conformal dimension is necessary to make the ghost sector of the critical bosonic
string CFT work correctly. We will see this can be done another way straightforwardly using
CFT technology.
For the moment, we will just focus on the analytic fields; everything goes through simi-
larly for the barred sector. Continuing our analysis, the propagator should satisfy,

∂z̄ hb(z)c(w)i = 2πδ (2) (z − w). (5.3)

So we suggest that the propagator takes the form of

1
hb(z)c(w)i = . (5.4)
z−w
This in turn gives us the operator product expansion,
1
b(z)c(w) = + regular = c(z)b(w), (5.5)
z−w
where we used the fact that they are Grassman fields to anticommute them in the second
equality. We can now expand them into their modes,
X
c(z) = cn z −n+1 (5.6)
n
X
b(z) = bn z −n−2 . (5.7)
n

We can now use our OPE to find the anticommutation relations these modes will satisfy.
Since c2 and b2 are zero due to being Grassman, we can immediately see that

{cn , cm } = 0, {bn , bm } = 0. (5.8)

Now we will use contours to find out the final anticommutators. Now since we instead have

42
anticommuting fields, the contours will give us an anticommutator instead. We can see that
I I
dw m−2 dz n+1
{bn , cm } = w z b(z)c(w) (5.9)
2πi Cw 2πi
dz z n+1
I I
dw m−2
= w (5.10)
2πi Cw 2πi z − w
I
dw m+n−1
= w (5.11)
2πi
= δn+m,0 (5.12)

Now we should examine the energy momentum tensor,

T (z) = −∂b(z)c(z) − 2b(z)∂c(z) (5.13)

which should be normal ordered in the quantum theory. Expanding T into modes using b
and c, X
T (z) = (2m − 2 + p + 2) : bp cm : z −n−m−2 (5.14)
m,p

But we could expand it into Virasoro generators by

X
−n−2
T (z) = Lb,c
n z . (5.15)
n

To find the Virasoro generators in terms of the ghost modes, we use contour integration.
I
b,c dz n+1
Ln = z T (z) (5.16)
2πi
X
= (n − m) : bn+m c−m : . (5.17)
m

Finding the T −T OPE for the ghost sector is a tad tedious, and only involves techniques
that you already know how to use (e.g. Wick’s theorem, Taylor expansion, the Grassmann
property of the ghosts), so I will suppress the boring steps. I plan to set working out the
details as a homework problem. Very similar types of calculations are needed to work out
the operator product expansions of T (z)c(w) and T (z)b(w), which provide the answer that
b and c are primary conformal fields of the weights we stated above. The final result for the
ghost T −T OPE is

−13 2T (w) ∂T (w)

T (z)T (w) = 4
+ 2
+ + regular (5.18)
(z − w) (z − w) (z − w)

Recall then that half of the coefficient of the first term gives us the central charge. So our
ghost system has a central charge of c = −26. By itself it does not make much sense, but for
us it always comes along for the ride with a CFT of D bosons X µ . It is therefore possible
to write down a sensible path integral for critical string theory provided that D = 26. This
completes the ghost-style derivation of the critical dimension of bosonic string theory.

43
5.2 BRST invariance and unitarity
Becchi, Rouet, Stora, and Tyutin noticed that if we choose the symmetry transformation
parameter ∆ω A (x) to be ∆ω A (x) = −cA (x), where cA (x) and  are both Grassmann and 
is a constant, then the action is invariant – provided that cA (x) transforms with a particu-
lar form. The BRST transformation equations involve the original fields you were trying to
quantize and their symmetries, as well as ghost, antighost and Lagrange multiplier fields. Us-
ing identities satisfied by the symmetry algebra generators, it is possible – although typically
a tad arduous algebraically – to show that the BRST transformation squares to zero:
2
δBRST =0 (5.19)

Since the BRST operator generating these BRST symmetry transformations is nilpotent,
any term added to the action which is the BRST-variation of something will automatically
be BRST-invariant. Formally, we can write this as
δBRST
Ltot = Lcl + Ξ (5.20)
δ
where Ξ is known as the gauge fermion.
The structure of BRST transformations implies very nontrivial things about unitarity
of scattering amplitudes. These all originate from having a nilpotent operator QBRST that
commutes with the Hamiltonian H: [H, QBRST ] = 0. To see this, let H1 be the subspace
of states |ψ1 i which are not annihilated by the BRST operator QBRST , and let H2 be the
subspace of states |ψ2 i of the form |ψ2 i = QBRST |ψ1 i for some |ψ1 i ∈ H1 . Also let H0
be the subspace of states |ψ0 i satisfying QBRST |ψ1 i = 0 but that cannot be written as the
QBRST of any state (the states which are BRST-closed but not BRST-exact). Having these
three subspaces of the space of states H0 , H1 , H2 follows directly from having a nilpotent
operator which commutes with the Hamiltonian. Incidentally, note that the subspace H2
is a rather weird subspace, as any two states in it have zero inner product. This fact also
follows directly from the nilpotency of QBRST .
Let |A; ⊥i denote an external state containing no ghosts or antighosts and only physical
(transverse) polarizations of X fields. We would like to show that the S-matrix for these
guys is unitary, i.e. that
X
hA; ⊥ |S † |C; ⊥ihC; ⊥ |S|B; ⊥i = hA; ⊥ |I|B; ⊥i (5.21)
C

Recall that H0 is the space of states |ψ0 i such that QBRST |ψ0 i = 0 but which cannot be
written as QBRST |λi for any λ (i.e., the states which are BRST-closed but not BRST-exact).
Also, we know that [QBRST , H] = 0 so that any time-evolved state S|ψ0 i is also killed by the
BRST operator QBRST : it takes the form

QBRST · S|A; ⊥i = 0 (5.22)

This implies that S|A; ⊥i must be linear combinations of states in H0 (physical ones) and
states in H2 (BRST-exact ones). But any two states in H2 have zero inner product with

44
one another, and also hψ2 |ψ0 i = 0 (by definition of H2 ). So the inner product of any two
states of that form must arise solely from the overlap of the components in H0 . Therefore,
X
hA; ⊥ |S † S|B; ⊥i = hA; ⊥ |S † |C; ⊥ihC; ⊥ |S|B; ⊥i (5.23)
C

and so not only is the full S-matrix unitary but its restriction to the subspace H0 is also
unitary. The BRST method is really powerful!
So far in the discussion of BRST invariance I have suppressed the details of the symmetry.
It is time to now lift the veil from your eyes and show how this part works.

5.3 Applying BRST to the bosonic string

We can define an operator called the U (1) ghost number current j(z) by
j(z) = − : b(z)c(z) : . (5.24)
We expect that j(z) should be a conformal field of weight 1. So we want,
X
j(z) = jn z n−1 . (5.25)
n

To find the coefficients, we use contour integration.

I
dz n
jn = z j(z) (5.26)
2πi
X
= : cn−m bm : . (5.27)
m

Classically, this number is conserved. However, there is a quantum anomaly, which will soon
turn out to be proportional to D − 26. This will be how the critical dimension will show up
in the context of BRST technology. The ghost charge is given by the contour integral
I
dz X
Ngh = j(z) = j0 = : c−m bm : . (5.28)
2πi m

To see the ghost charge of a conformal field φ(z), we can just look at the operator product
expansion,
Ngh φ(w)
j(z)φ(w) = + regular. (5.29)
z−w
Note, if were we to do this OPE, we would see that c(z) has Ngh = 1 and b(z) has Ngh = −1
as we would expect.
The general BRST method goes as follows. Suppose that you have a Lie algebra
[Ki , Kj ] = fij k Kk (5.30)
where i, j, k run over 1, . . . , dimG. Define the BRST operator by
QBRST = ci Ki − 12 cj bk


(5.31)
 
= ci Ki + 21 Kigh (5.32)

45
where {ci , bj } = δji . Using the symmetry algebra, the canonical anticommutation relations
and the antisymmetry of the structure constants, it follows that
Q2BRST = 14 f[ij k fl]km (cj ci cl bm ) = 0 (5.33)
Without much work you can see that with
K̃i := Ki + Kigh (5.34)
the K̃i also satisfy the same symmetry algebra with identical structure constants. Under
BRST transformations,
δBRST ci = {QBRST , ci } = − 21 fkl i ck cl (5.35)
δBRST bi = {QBRST , bi } = Ki − fij k cj bk = Ki + Kigh = K̃i . (5.36)
Physical states obey
QBRST |ψi = 0 (5.37)
which ensures that they have zero ghostiness.
Now, how about doing this kind of malarkey when we have an infinite number of sym-
metry generators, as is the case for our Virasoro algebra? In that case we have to worry
about normal ordering. We have
+∞ +∞
!
X X
1
QBRST = c−m LXm − 2 (m − n) : c−m c−n bm+n : (5.38)
m=−∞ n=−∞

or, as a contour integral,

I
dz
: c(z) T X (z) + 21 T b,c (z) :
 
QBRST = (5.39)
C0 2πi
Then we can specify the BRST current as
jBRST = cT X + 21 cT b,c + 23 ∂ 2 c (5.40)
where the 3/2 is required in order that the BRST current has conformal weight unity.
How bad is it to check that Q2BRST = 0? It is perhaps easiest to do as the jBRST jBRST
OPE, which is equivalent. If you go through the exercise, you will find that
(D − 26) 1
jBRST (z)jBRST (w) = . . . + [(∂ 3 c)c](w) + . . . (5.41)
12 (z − w)
From this it also follows straightforwardly that in the critical dimension both jBRST and T tot
are both BRST-exact. If you would like more detail you can consult BLT §5.2. Some useful
relations are
[Q, X µ ] = c∂X µ (5.42)
(D − 26) 3
[Q, T tot ] = ∂ c (5.43)
12
{Q, c} = c∂c (5.44)
{Q, b} = T tot (5.45)

46
We can transform these into operator relations by contour integrals.
X
[Q, αn ] = − ncm αn−m (5.46)
m
1
[Q, Ln ] = − (D − 26)n(n2 − 1)cn (5.47)
12
X
{Q, cn } = − (2n + m)c−m cm+n (5.48)
m
tot
{Q, bn } = Ln . (5.49)
Then, via mode-ology, the expression for the nth total Virasoro generator is
+∞
X
Ltot 1


n = 2
: αn−m αm : +(n + m) : bn−m cm : (5.50)
m=−∞

so that
α0 2
Ltot
0 = N
tot
+ p −1 (5.51)
4
where N tot counts the oscillator energy in all the modes.
In CFT there is a state-operator correspondence emanating from the fact that spec-
ifying the boundary condition on the path integral at τ = −∞ inserts a local operator at the
origin (in the plane; see radial quantization). For a more detailed exposition I recommend
Polchinski Volume 1 §2.8 on Vertex Operators. Consider a vertex operator φ. Its commu-
tator with the BRST operator involves a contour integral of jBRST against φ. By BRST
invariance, the jBRST − φ OPE must not have a pole of order one, or if it does, the residue
is a total derivative so that, integrated over the insertion point, it vanishes. Now specialize
to a physical state with no ghost excitations. Then
I I
dw dw
jBRST (w)φ(z) = c(w)T X (w)φ(z) (5.52)
2πi 2πi
I  
dw hφ φ(z) ∂φ(z)
= c(w) + + ... (5.53)
2πi (w − z)2 (w − z)
= hφ (∂c)φ(z) + c(∂φ)(z) (5.54)
and this is a total derivative iff hφ = 1 i.e. we are dealing with a physical vertex operator.
So overall: primaries of dimension one create asymptotic BRST-invariant states.
How about the ghost sector? Our main complication here is that there are ghost zero
modes. These satisfy b20 = 0 = c20 by the canonical anticommutation relations, and they both
commute with the Hamiltonian. So there are two degenerate states for the ghost vacuum.
Let us denote them as | ↑i and | ↓i, where
c0 | ↑i = 0 b0 | ↑i = | ↓i (5.55)
b0 | ↓i = 0 c0 | ↓i = | ↑i (5.56)
We can add a constant to j0 to symmetrize these as follows,
1 X
Ngh = (c0 b0 − b0 c0 ) + (c−n bn − b−n cn ) (5.57)
2 n>0

47
or
3
j(z) = − : b(z)c(z) : − (5.58)
2z
Now, for the b, c modes the SL(2, C) invariant vacuum is defined by

bn |0ib,c = 0 , n ≥= −1 (5.59)
cn |0ib,c = 0 , n ≥= +2 . (5.60)

Unfortunately, the vacuum so constructed is not annihilated by c1 , which means that it is

not a highest-weight state of the b, c algebra, ergo is not the groundstate of the ghost system.
Note that c1 |0ib,c and c0 c1 |0ib,c both have L0 eigenvalue −1. Identify them as | ↓i and | ↑i
respectively. Pick the normalization such that

b,c h0|c−1 c0 c1 |0ib,c = +1 . (5.61)

In other words, the SL(2, C)-invariant vacuum carries 3 units of ghost number, corresponding
to the 3 independent diffeomorphisms of the sphere generated by {L0 , L−1 , L1 }. Using the
above, we will build physical states of the form

|ψi = |φiX ⊗ | ↓ib,c . (5.62)

Then requiring BRST invariance gives

!
X
Q|ψi = 0 = c0 (LX
0 − 1) + c−n LX
n |ψi (5.63)
n>0

which requires both

(L0 − 1)|ψi = 0 (5.64)

Ln |ψi = 0 , n > 0, (5.65)

exactly as we would have expected from our last lecture. Note: this trick would not have
worked if we had built states on the up vacuum instead of the down vacuum, because
c0 | ↑i = 0. Requiring the following will ensure those options are eliminated:

b0 |ψi = 0 , b̄0 |ψi = 0 . (5.66)

Then {QBRST , b0 − b̄0 }|ψi = (L0 − L̄0 )|ψi = 0, which is the level matching constraint.
Now, how about identifying the physical states for all N tot ? Since {QBRST , b0 } = Ltot
0 , any
state of definite pµ and N tot satisfying b0 |ψi = 0 and Q|ψi = 0 satisfies α0 m2 = −4(N tot − 1).
Vertex operators of physical states are of the form

ψ(z) = c(z)φ(z) . (5.67)

Then
[Q, ψ(z)] = (hφ − 1) : ∂ccφ(z) := 0 , for hφ = 1 . (5.68)

48
This is called the unintegrated form of the vertex operator. The integrated form is
Z
d2 zφ(z, z̄) (5.69)

where ψ = cc̄φ such that [Q, φ] = ∂(cφ) and similarly for the barred sector. This ensures
that [Q, ψ] = 0 and similarly for the barred sector.
What are the vertex operators for the groundstates of closed bosonic strings? We are
about to find out that there is not only a graviton in spacetime but also an antisymmetric
tensor field Bµν (also with its own gauge symmetry) and a scalar field known as the dilaton
Φ. To do so, we need only remind ourselves of the salient physical LEGOs,
α0 /2
h∂X(z)∂X(w)i = − (5.70)
(z − w)2
and
1
T =− : ∂X∂X : (5.71)
α0
which have OPEs
∂X(w) ∂(∂X(w))
T (z)∂X(w) = 2
+ + ... (5.72)
(z − w) (z − w)
 0 2 
α k /2 ∂w
T (z) : eik·X : = + : eik·X : (5.73)
(z − w)2 (z − w)
Therefore, ∂X has conformal dimension 1 and the normal-ordered exponential has conformal
dimension h = α0 k 2 /4 (remember, |z|2 = z z̄). The field X itself does not have a well-defined
conformal dimension. Two other identities worth knowing are
0
: eip·X(z) :: eiq·X(w) : = |z − w|α p·q : ei(p·X(z)+q·X(w)) :
0
= |z − w|α p·q : ei(p+q)·X :
0
+i|z − w|α p·q p · (z − w) : ∂X(w)ei(p+q)·X(w) : +barred (5.74)
 

ip·X(w) α0 ip
∂X(z) : e : = − : eip·X(w) : + . . . (5.75)
2 (z − w)

Physical string states obey (L0 −1)|ψi = 0, Ln |ψi = 0, n > 0 and similarly for the barred
sector; they also obey the momentum constraint (L0 − L̄0 )|ψi = 0. Therefore, they are just
the primaries of the CFT
|φi = φ(0)|0i = lim φ(z, z̄)|0i . (5.76)
z,z̄→0

The really cool thing that follows from this state-operator correspondence is that string scat-
tering amplitudes then become integrals over the worldsheet (in a sense that we will become
technically exact about shortly) of correlation functions of vertex operators associated to the
physical states. While this is morally simple, the precise details can get very messy indeed
(see e.g. §16 of BLT). For now, we just focus on what is necessary to build vertex operators
R 2 string tachyons and massless modes. Now, the condition (L0 − 1)|ψi = 0 requires
for closed
that d xφ(z, z̄) has h = (1, 1) in order to be well-defined. So then a [closed string] tachyon

49
has α0 k 2 /4 = −1, i.e., m2 = −k 2 = −4/α0 . It is easy to check from the definition of pµ as
an operator written as a contour integral of ∂X that this vertex operator thingie does carry
momentum k µ .
That was for the tachyons. How about for the first excited states, which are massless?
We define the vertex operator for these as
2 ¯ ν (z̄)eik·X(z,z̄) : |0i
|k; i = − µν (k) lim : ∂X µ (z)∂X (5.77)
α0 z,z̄→0
µ ν
= α−1 ᾱ−1 µν |ki (5.78)
We need to check whether this has (h, h̄) = (1, 1) so that it can represent a physical state.
Taking the OPE with the energy-momentum tensor gives
¯ ν (w̄)eik·X(w,w̄) :
T (z) : µν ∂X µ (w)∂X (5.79)
iα0 k µ µν ¯ ν (w̄)eik·X(w,w̄) : +
=− : ∂X (5.80)
2 (z − w)3
 0 2 
α k /4 + 1 ∂w ¯ ν eik·X :
+ 2
+ µν : ∂X µ ∂X (5.81)
(z − w) (z − w)
So for physicality we require
k µ µν = 0 (5.82)
from the holomorphic sector. From the barred sector we obtain an analogous equation
k ν µν = 0 (5.83)
These two conditions requires the polarization vectors to be transverse. What fields do we
obtain? The tensorial property of a general two-index tensor can be split up as follows: (1)
trace, (2) symmetric traceless and (3) antisymmetric. The resulting bulk spacetime fields
are known as the dilaton Φ, graviton Gµν (with its well-known diffeomorphism symmetry)
and the Kalb-Ramond field Bµν which also has a gauge symmetry analogous to that of
electromagnetism only with one more spacetime index.
As for higher excited states, it is possible to prove a formula known as the Cardy
formula describing the density of states at large level number,
r !
cL0
ρ(L0 ) ∼ exp 2π . (5.84)
6

This might look unassuming but please notice that it becomes exponentially large at large L0
eigenvalue. This means that the entropy S grows like the energy E at high-energy, meaning
that there is a limiting temperature at which the canonical ensemble still makes sense -
delineated by the point at which the Cardy exponential overwhelms the Boltzmann weight.
This limiting temperature is called the Hagedorn temperature.

5.4 The Virasoro-Shapiro amplitude

In this subsection, we will only look at (1) tree amplitudes and (2) n-point amplitudes where
n ≤ 4. This is done in order to keep the bit rate lower than that of a firehose. For > 4-point

50
amplitudes the integration over insertion points is a nontrivial technical challenge, and for
loop amplitudes the conformal Killing vectors (CKVs) have to be analyzed in full careful
detail – parametrizing the moduli space of Riemann surfaces of higher genera is a bitch.
Conformal symmetry on the worldsheet (in critical dimension) means that diagrams
shaped like (say) two closed strings joining and then splitting again, can be conformally
mapped into the sphere with four marked points. The diagrams are traditionally drawn with
× denoting insertions of the vertex operators representing physical states (that are killed by
the BRST operator QBRST ) at particular points on the worldsheet in the z, z̄ coordinates.
Generally, our n-particle amplitude takes the form
Z 2
n−2 d z1 . . . d2 zn
Atree ∼ gs hV1 (z1 ) . . . Vn (zn )iS 2 (5.85)
(vol(CKV)

where gs is the string coupling. The n − 2 in front comes about because this is the Euler
number of an n-punctured S 2 . Alternatively, you can count it as 1/gs2 for the sphere and gs
for each vertex operator.
At tree level, the holomorphic and anti-holomorphic parts just factorize simply. This
simple factorization does not hold at loop level and it is important to dig into the full
technical detail to see this. We do not have time in this survey-level course to get this
technical. BLT §6 is where they develop in detail the nasty details; we will just cherry pick
a few key facts here to enable us to build up to a four-point tree amplitude before we move
on to the superstring and heterotic constructions next week.
First, the CKVs correspond to zero modes of the c ghost. The zero modes of the b
antighost correspond to animals known as Beltrami differentials, and the number of them is
given by the Riemann-Roch theorem Nb − Nc = 3(g − 1). Luckily for us, at genus zero (the
sphere, worldsheet tree level) there are no b zero modes as there are no moduli. There are
however 3 c zero modes corresponding to the generators ∂z , z∂z , z 2 ∂z . We met these buggers
already when figuring out the ghost vacuum business.
Fix 3 vertex operators at positions z1 , z2 , z3 . Vector
R 2 fields generating
R 2the 3 CKV sym-
2
metries, (α + βz + γz )∂z , mean that we can trade d {z1 , z2 , z3 } for d {α, β, γ} modulo
the Jacobian  
 ∂(z1 , z2 , z3 )  2
 ∂(α, β, γ)  = |(z1 − z2 )(z2 − z3 )(z3 − z1 )| , (5.86)
 

and d2 {α, β, γ} =vol(CKG). This can also be found as | det(V i (zj ))|2 where {V i } =
R

{1, z, z 2 } are the CKVs and

 1 1 1 2
 

|h0|c(z1 )c(z2 )c(z3 )|0i|2 = |h0|c−1 c0 c1 |0i|2 det z3 z2 z1  = z12

2 2 2
 
z23 z31 (5.87)
 z32 z22 z12 

So, formally, in the path integral what we do to divide out by the volume of the conformal
Killing vectors is to insert a ghost field at 3 chosen points and drop the integrations over
their positions. Usually, the 3 marked points are chosen at {0, 1, ∞}.
SL(2, C) invariance actually makes the zero-, one-, and two-point functions vanish. The
first nontrivial amplitude is the three-point amplitude, and we will only consider the simplest

51
situation of the bosonic string involving three tachyons. Using the above rules, we have
(3)
Abos (T ) = CS 2 gc3 hcc̄V (1)cc̄V (2)cc̄V (3)i = gc3 CS 2 (5.88)
where CS 2 is a physical normalization constant that can be determined by unitarity and
known physics of gauge and graviton particle theory amplitudes (see BLT and/or Polchinski
Vol.1 for details). Here, we used the form of the vertex operators V (k) =: eik·X : and the fact
that, on-shell, k 2 = −m2 = 4/α0 , as well as α0 ki · kj /2 = −2 for i 6= j. The only other input
we have used is the usual trick of ignoring the overall momentum conservation delta-function
proportional to the volume of spacetime familiar from particle theory. One last step you will
need to reproduce the result is
n
X Y
α0 ki ·kj /2
D D
hV (1)V (2)V (3)i = (2π) δ ( ki ) |zij |2 (5.89)
i i<j

Now that we have an idea about the three-point amplitude we can consider the four-point
scattering amplitude for closed bosonic string tachyons. This will take the form
Z
(4)
Abos (T ) = gc CS 2 d2 Z3 hcc̄V (1)cc̄V (2)V (3)cc̄V (4)i
4
(5.90)

We can easily find the integrand; it is

Y
α0 ki ·kj /2
|z12 z14 z24 |2 |zi − zj |2 (5.91)
i<j
P4
Next, we fix z1 = 0, z2 = 1, z4 = ∞ and use momentum conservation i=1 ki = 0. We also
use the very useful BLT formula §(16.23),
Z
d2 z|z|α |1 − z|β z n (1 − z)m = (5.92)
Γ(1 + n + α/2)Γ(1 + m + β/2)Γ(−1 − m − n − {α + β}/2)
2π(−1)m+n (5.93)
Γ(−α/2)Γ(−β/2)Γ(2 + {α + β}/2)
Then for our four-point closed string tachyon amplitude we have
(4) Γ(α(s))Γ(α(t))Γ(α(u))
Abos (k1 , k2 , k3 , k4 ) = 2πgc4 CS 2 (5.94)
Γ(α(t) + α(u))Γ(α(s) + α(u))Γ(α(s) + α(t))
where
α0
α(s) = −1 + s (5.95)
4
and similarly for t, u, where {s, t, u} are the standard Mandelstam kinematic invariants
s = −(k1 +k2 )2 , t = −(k1 +k3 )2 , u = −(k1 +k4 )2 which obey the relation s+t+u = 4i=1 m2i .
P
For our closed bosonic string tachyons, m2 = −4/α0 .
The above formula is called the Viraroso-Shapiro amplitude, the closed string cousin
of the open-string Veneziano amplitude. It has two properties which you can immediately
see are physically important. First, it has crossing symmetry. This is why it was called a
dual model when string theory was originally invented in the late 1960s, in an attempt
to explain the strong nuclear force. Second, it has simple poles in each channel (and an
infinite number of them), e.g. in the s-channel you get poles at s = −4/α0 , 0, +4/α0 , +8/α0 ,
etc. These correspond to exchange of a closed excited string state. The Virasoro-Shapiro
amplitude does not have any higher order poles.

52
6 Superstrings
The famous Coleman-Mandula theorem states that spacetime symmetries cannot be com-
bined with internal symmetries (such as colour or hypercharge) in a physically nontrivial
fashion. That is why we treat Standard Model gauge group indices as completely indepen-
dent of spacetime indices when we write down quantum fields for quarks and leptons: the
structure is a trivial tensor product. Now, the Coleman-Mandula theorem assumed that the
generators of symmetry transformations were bosonic. If you allow them to be fermionic
then Poincaré can actually be nontrivially extended, and the result is called supersymme-
try. SUSY transformations have fermionic parameters and fermionic generators; the group
elements formed via the exponential map are still bosonic. So in addition to the momen-
tum Pµ and angular momentum/boost generators Mµν , there are also SUSY generators Qiα ,
where the i index is used only for extended SUSY, which obey very specific anticommutation
relations that we began to introduce in the first lecture. In particular, we saw that if you
do two SUSY transformations in a row you get back a translation (etc.), as we saw when
we discussed the role of BPS states. In order to be able to say anything about the physics
of supersymmetry, we first need to learn how to build spinors in diverse dimensions. We
pattern our discussion here largely on Appendix B1 of Polchinski Volume 2, referred to in
the following as Pol.vol2.appB1.

6.1 Spinors in diverse dimensions

Let spacetime have signature (D − 1, 1). The Dirac algebra is defined by

{Γµ , Γν } = 2g µν 1 . (6.1)

First consider the case of even dimensions; the odd dimensional case is a relatively simple gen-
eralization. The fundamental algebra above implies that different Γµ anticommute with each
other, while each one squares to ±1. Therefore, the product of a finite number of them can al-
ways be reduced to the following set of 2D matrices {ΓA } = {1, Γµ , Γ[µ Γν] , . . . , Γ0 Γ1 · · · ΓD−1 }
which are all linearly independent. Notice that (ΓA )2 = ±1, for any A, regardless of whether
the ΓA is a product of 1 or more Γµ s. Note also that (Γµ ) = (Γµ )−1 ; also, if ΓA = Γa Γb . . . Γh
for some a, b, . . . , h then ΓA = Γh . . . Γb Γa and these satisfy ΓA ΓA 1 = ΓA ΓA . In even dimen-
sions only, we can define a ΓD+1 by ΓD+1 = i−k Γ0 Γ1 · · · ΓD−1 , where D = 2k+2, which squares
to 1 and commutes with the Γµ as well as with the Lorentz generators Σµν = − 4i [Γµ , Γν ]. The
only nonzero trace of a ΓA occurs for the identity which produces D; otherwise, tr(ΓA ) = 0 if
ΓA 6= 1. We also have the Fundamental Theorem: if Γµ and Γ̃µ are two different irreducible
representations of the Dirac algebra in the gamma matrix space, then there exists an invert-
ible matrix S defined up to a constant such that Γµ = S Γ̃µ S −1 for all µ ∈ {0, 1, . . . , D − 1}.
What changes when D is odd? Here, we select D − 1 of the Γs (i.e., Γ0 , Γ1 , . . . , ΓD−2 )
and apply the previous analysis. Then, obviously, ΓD−1 anticommutes with the other ones
by the Dirac algebra. Since these D − 1 generators fully span the space of matrices, ΓD−1 =
(const.) × Γ0 Γ1 . . . ΓD−2 , because only that product of all the Γs in a row can anticommute
with the first D − 2 of them. Note also that ΓD+1 has to be proportional to the identity in
odd dimensions.

53
 Now we get to the interesting bit: building the Fock space representation for spinors.
We do it by selecting pairs of the original set of Γ matrices,

(Γ0 )± := 12 ±Γ0 + Γ1


(6.2)
a ± 1 2a 2a+1


(Γ ) := 2 ±Γ ± iΓ (6.3)

where a = 1, . . . , k. These satisfy {Γa+ , Γb− } = δ ab and {Γa+ , Γb+ } = 0 = {Γa− , Γb− } and
so can be thought of as anticommuting creation and annihilation operators. This is very
reminiscent of Pauli exclusion. Indeed, for each pair of Γ± , the Γ+ behaves just like a b† and
Γ− like a b. We start by finding the ζ annihilated by all the Γa− . Then acting with Γa+ at
1 1
most once each gives ~s = (s0 , s1 , . . . , sk ) where si = ± 21 ∀i: ζ ~s = (Γk+ )sk + 2 · · · (Γ0+ )s0 + 2 ζ.
Overall, we see that spinors must have dimension 2[D/2] . Note that this is 2 to the power
[D/2], where [. . .] denotes the integer part. This is because if you are in odd dimensions and
have an extra Γ to play with you cannot enlarge the Fock space because the extra Γ is not
paired up. So spinor dimensionality is not mysterious: it is pre-ordained through the Dirac
algebra from which all of the above follows. This is why spinors in 4D are four dimensional.
There are three subclasses of spinors available: Weyl, Majorana, and Majorana-
Weyl. Weyl
spinors exist in even dimensions and are chiral; the Weyl projectors P± :=
1
2
1 ± ΓD+1 split the spinor into left- and right-chiral pieces. Majorana spinors are real;
Majorana-Weyl are both chiral and real. Which types of spinors can occur in which dimen-
sions can be looked up in any decent textbook; we summarize the reults from Pol.vol2.appB1
in Figure 7.3. We will indicate how the Majorana analysis works for even dimensions here; see
Pol.vol2.appB1 for more details on the odd-dimensional case and further generalities. After
the dust settles, one noteworthy example will be that Majorana-Weyl spinors are available
both for the worldsheet case D = 1 + 1, where they are one-dimensional, and the critical
dimension for superstrings D = 9 + 1, where they are sixteen-dimensional.

54
 In the basis ~s, matrix elements of Γa± are real. Using the definitions for Γa± we can
easily see that Γ3 , Γ5 , . . . , ΓD−1 are imaginary while all the other Γµ are real. Define

B1 = Γ3 Γ5 · · · ΓD−1 , B2 = ΓD+1 B1 . (6.4)

Then by the Dirac algebra, B1 Γµ B1−1 = (−1)k Γµ∗ and B2 Γµ B2−1 = (−1)k+1 Γµ∗ Also, by
−1 −1 ∗
virtue of the form of the Lorentz generators, B1,2 Σµν B1,2 = −Σµν∗ . Therefore, ζ and B1,2 ζ
transform the same way under Lorentz transformations so that the Dirac representation is
−1
its own conjugate. Also, B1,2 ΓD+1 B1,2 = (−1)k ΓD+1∗ , so that either B1 or B2 will change
the eigenvalue of ΓD+1 when k is odd and not when k is even. Further, for k even (D = 2
mod 4) each Weyl representation is its own conjugate. For k odd (D = 0 mod 4) each Weyl
representation is conjugate to the other. A reality condition ζ ∗ = Bζ is consistent only if
B ∗ B = 1. In fact, B1∗ B1 = (−1)k(k+1)/2 and B1∗ B1 = (−1)k(k−1)/2 . This means that you
can use B1 for a Majorana condition if k = 0 mod 4 or k = 3 mod 4, and you can use B2
for one if k = 0 mod 4 or k = 1 mod 4. For k = 0 mod 4, B1 and B2 are related by a
similarity transformation and hence physically equivalent. Finally, we note that Majorana-
Weyl spinors require (a) a Majorana condition and (b) that the spinor be self-conjugate.
This requires k = 0 mod 4 or equivalently D − 2 = 0 mod 8, which is relevant to both the
D = 1 + 1 worldsheet and the bulk D = 9 + 1 target spacetime.

6.2 Green-Schwarz formalism

There are several ways to handle the issue of SUSY for superstrings. The two which you
will see the most often in standard string theory textbooks like Polchinski, BBS, BLT, and
Green-Schwarz-Witten are the Neveu-Schwarz-Ramond formalism, which insists on world-
sheet supersymmetry, and the Green-Schwarz formalism, which insists on spacetime super-
symmetry. Both preserve Poincaré symmetry. The Berkovits pure spinor formalism gives a
very powerful (and very technical!) third way forwards; we will not have time to say anything
about it here.
I already have several pages of notes online describing the basics of the Green-Schwarz
approach and kappa symmetry at http://ap.io/2406s/notes/st20080305.pdf, first for
point particles and then for superstrings, using light-cone gauge when necessary to illustrate
the physics. Accordingly, I will not repeat the material here in LATEX: I cannot afford the
typing pain! I will just give a brief summary of the main points. Those interested in further
details should consult Green-Schwarz-Witten.
We already know that the action principle we start from has reparametrization invari-
ance. Introduce fermionic superpartners for the bosonic degrees of freedom X µ , which are
θa , where a is the spinor dimension appropriate to D dimensions. The action constructed
has SUSY, but it also possesses a qualitatively different symmetry involving the bosonic and
fermionic worldsheet fields called kappa symmetry. Kappa symmetry serves to provide a
projection operator that reduces the rank of a spinor by a factor of two, and it is neither
SUSY on the worldsheet nor SUSY in the target spacetime. It is possible to write down a
kappa symmetric worldsheet action for the superstring, known as the Green-Schwarz action,
which preserves worldsheet reparametrization invariance as well as target spacetime super-
Poincaré invariance. But it requires a gnarly self-consistency condition that arises out of

55
making the most of Fierz identities: ¯Γµ ψ[1 ψ̄2 Γµ ψ3] ≡ 0, with (ψ1 , ψ2 , ψ3 ) = (θ, ∂σ θ, ∂τ θ).
This puts a very strict condition on the dimension of spacetime for classical superstrings. It
can be satisfied only when
D = 3, 4, 6, 10 (6.5)
when we have one timelike dimension of spacetime, which is the physical number. No other
dimension permits the correct matching of bosonic and fermionic degrees of freedom. Those
are also the dimensions in which local kappa symmetry can be implemented. Quantum con-
sistency of the superstring will select out the case D = 10 as the unique allowed spacetime
dimension. A useful mathematical property known as triality, an automorphism of SO(8),
allows you to relate the eight-dimensional vector spinor and the conjugate spinor representa-
tions of SO(8). In light-cone gauge, the fermion analogue of the light-cone gauge condition
welding X + to τ lets you set ψα+ = 0; in turn, this allows solving for the ψα− oscillators in
terms of the transverse ψαI . Working through the kappa symmetric details you end up with
equations of motion on the worldsheet which are the Weyl equations of motion for chiral
spinors recognizable from QFT1. Type I superstring theory eventuates when you set to
zero one or other of the two spinors of N = 2 worldsheet supersymmetry allowed by kappa
symmetry. Type IIB is a chiral theory which eventuates when both spinors have the same
chirality, and Type IIA is a nonchiral theory which eventuates when they have the opposite
chirality. Heterotic theories arise via a hybrid construction where the left side is bosonic
while the right side is supersymmetric; the remaining 16 leftover XLi end up being bundled
up into a current algebra. Requiring that all anomalies cancel gives SO(32) or E8 × E8 .

6.3 Neveu-Schwarz-Ramond formalism

With worldsheet SUSY in the NSR formalism, we partner every worldsheet boson field with
a worldsheet fermion field. In particular, this means that our worldsheet fermions ψ µ carry
a spacetime vector index just like the X µ do.
The following superstring action on the worldsheet in conformal gauge (see Polchinski
eq.10.1.5) has a global worldsheet supersymmetry,
Z  
1 2 2 µ¯ µ¯ µ
S= dz ∂X ∂Xµ + ψ ∂ψµ + ψ̃ ∂ ψ̃µ , (6.6)
4π α0

where ψ µ are anticommuting worldsheet fields (Majorana-Weyl spinors. Look for a minute
at the fermionic term. Its variation must vanish for the initial value problem to be well
defined; a simple integration by parts shows that
Z
1 σ=π
δSf = dτ ψ+ δψ+ − ψ̄− δ ψ̄− σ=0 . (6.7)
4π
where we have temporarily restored the + subscript on the right-movers and the − subscript
on the left-movers for clarity.
In order to make the variation vanish for open strings, the endpoint terms have to cancel
separately at σ = 0 and σ = π, implying that at those two points (and only at those two
µ µ
points) ψ+ = ±ψ− . The overall relative sign is conventional and taken to be positive at

56
σ = 0. Then
µ µ
R: ψ+ (τ, σ = π) = +ψ− (τ, σ = π) , (6.8)
µ µ
NS : ψ+ (τ, σ = π) = −ψ− (τ, σ = π) . (6.9)

These R fields are integrally moded on the cylinder while the NS fields are half-integrally
moded.
In order to make the variation vanish for closed strings, we need

ψ± (τ, σ) = ±ψ± (τ, σ + π) (6.10)

so we impose either NS or R boundary conditions in both the holomorphic and anti-

holomorphic sectors. This gives rise to NS-NS, R-R, NS-R and R-NS states.

6.4 ZPEs, physical state conditions and the GSO projection

Old covariant quantization yields decoupling of spurious states for

D = 10 , aN S = − 21 , aR = 0 . (6.11)
1
Recall that for the bosonic string we found a = − 24 for each transverse boson; this came
1
from 2 ζ(−1) in zeta function regularization. If you rerun the calculation for the fermion
1
contributions in the NS sector you find − 48 for each fermion. So the ZPE for our superstring
1 1
in the NS sector arises from (a) 8 × − 24 + 8 × − 48 = − 21 . In the R sector, which preserves
1 1
unbroken worldsheet supersymmetry, you find − 24 × 8 + 24 × 8 = 0. The vacuum in the NS
sector is unique, but the Ramond vacuum carries a representation of the Dirac algebra via
the zero modes ψ0µ and is a spacetime spinor. The physical state conditions are as for the
bosonic string, except that (a) there are fermionic as well as bosonic contributions to the
mode sums and (b) there is a new condition that the fermionic partners of Ln – the Gr –
also annihilate the physical states for positive mode number r.
Light cone quantization welds X + to τ , disallowing it any oscillators. The fermionic
analogue is ψ + = 0, which can be used to solve for the + direction oscillators in terms of our
friends the transverse oscillators. The result for the mass formula is the same as previously
found for the bosonic string except that the zero-point energies are shifted according to
(6.11). First consider the open string. In the NS sector the groundstate |0N S ; ki is again a
scalar, a tachyon with α0 m2 = − 12 ; the first excited state bi−1/2 |0N S ; ki is a massless vector of
SO(8) which is the transverse Lorentz group. In the R sector the groundstate has unbroken
supersymmetry and its groundstate is a SO(8) spinor; all the excited states are also spinors.
Define G-parity by
bi−r bir +1
P
GN S = (−1)FN S +1 = (−1)
1
r∈N+ 2
(6.12)
i i
P
\ n∈N d−n dn
GR = Γ (−1) (6.13)

where Γ\ = Γ0 Γ1 · · · ΓD−1 (note that we use \ to denote the number that follows 9 as a single
digit). The GSO projection sets

(−1)FN S = +1 (6.14)

57
which projects out the tachyon and every second oscillator level above it. This is consistent
not only in free string theory but to all orders in perturbation theory. Note that in the R
sector the G-parity is conventional depending on the chirality of the spinor groundstate. The
second major advantage of the GSO projection (apart from killing the unphysical tachyons)
is that it ensures spacetime supersymmetry with all the nice implications thereof. It sounds
simple in principle, and it is, but the technical details are quite involved. See e.g. BLT.
The spectrum of massless states for the closed string is straightforward to work out; see
e.g.BBS §4.6 for further details.

NS − NS : Gµν (35) ⊕ Bµν (28) + Φ(1) (6.15)

NS − R, R − NS : ψµα (56) ⊕ λα (8) (6.16)
+
R−R: IIA : Cµ (8) ⊕ Cµνλ (56); IIB : C(1) ⊕ Cµν (28) ⊕ Cµνλσ (35) . (6.17)

The names of the NS-R and R-NS fermions are the gravitino and the dilatino. The Cs
are called R-R p-form antisymmetric tensors. The NS-NS sector is common to all string
theories, except for the unoriented ones which project out the Kalb-Ramond field B.

6.5 SCFT, (β, γ) ghosts, and BRST

Armed with the NSR worldsheet action in conformal gauge, we can work out some important
basic things. First is equations of motion give us that
¯ µ = 0 ⇒ X µ (z, z̄) = X µ (z) + X µ (z̄)
∂ ∂X (6.18)
¯
∂ψ(z, z̄) = 0 ⇒ ψ(z, z̄) = ψ(z) (6.19)
∂ ψ̃(z, z̄) = 0 ⇒ ψ̃(z, z̄) = ψ̃(z̄) (6.20)

and the propagators are given by

α0 η µν ¯ µ (z̄)∂X¯ ν (w̄)i = − α
0 µν
η
h∂X µ (z)∂X ν (w)i = − h∂X (6.21)
2 (z − w)2 2 (z̄ − w̄)2
η µν η µν
hψ(z)ψ(w)i = hψ̃(z̄)ψ̃(z̄)i = (6.22)
z−w z̄ − w̄
We can then determine that the energy momentum tensors are given by
 
1 2 µ µ
TB (z) = − : ∂X (z)∂Xµ (z) : + : ψ (z)∂ψµ (z) : (6.23)
2 α0
 
1 2 ¯ µ ¯ µ ¯
T̃B (z̄) = − : ∂X (z̄)∂Xµ (z̄) : + : ψ̃ (z̄)∂ ψ̃µ (z̄) : . (6.24)
2 α0

We can work out the world-sheet supercurrents associated with supersymmetry transforma-
tions, which gives us
r r
2 µ 2 µ ¯
TF (z) = i ψ (z)∂X µ (z) T̃F (z̄) = i ψ̃ (z̄)∂Xµ (z̄). (6.25)
α0 α0

58
To find the supersymmetry transformations, we can find the OPE with these generators with
the fields. So when we calculate TF (z)X µ (w), TF (z)ψ µ (w), and T̃F (z̄)ψ̃ µ (w̄), the singular
terms will give us
r
µ α0
δX (z, z̄) = − ((z)ψ µ (z) + ˜(z̄)ψ̃ µ (z̄)) (6.26)
2
r
2
δψ µ (z) = (z)∂X µ (z) (6.27)
α0
r
2 ¯ µ (z̄).
δ ψ̃ µ (z̄) = ˜(z̄)∂X (6.28)
α0
Just for example, we can work out
 1/2
µ 2
TF (z)X (w) = i : ψ ν (z)Xν (z) : X µ (w) (6.29)
α0
 1/2
2
=i 0
∂z hXν (z)Xµ (w)iψ ν (z) (6.30)
α
 0 1/2 µ
α ψ (w)
= −i + regular (6.31)
2 z−w
1
= iδX µ (w) . (6.32)
z−w
So, we get the transformation above as stated. Now we could go back and show that the
action is invariant under this transformation. From here, we could work out a variety of
different OPEs. They are given by

3D 2TB (w) ∂TB (w)

TB (z)TB (w) = 4
+ 2
+ (6.33)
4(z − w) (z − w) z−w
3TF (w) ∂TF (w)
TB (z)TF (w) = 2
+ (6.34)
2(z − w) z−w
D 2TB (w)
TF (z)TF (w) = 3
+ . (6.35)
(z − w) z−w

We can glean important information out of these. The first OPE tells us that the central
charge for this SCFT is given by c = 23 D. This comes from the fact that each scalar field
adds 1 to the central charge and each fermion adds 21 (as you are proving for yourself in
HW2). The second OPE tells us that the supercurrent is a tensor of weight ( 32 , 0).
How about the ghosts? In addition to the bc ghosts from the bosonic part, we must add
a commuting (β, γ) ghost system to handle the Fadeev-Popov determinant. Whereas the
b ghost has weight (2, 0) and c has (−1, 0), the new ghosts have weight ( 32 , 0) and (− 12 , 0) for
β and γ respectively. We also have their antiholomorphic counterparts.
Z
i ¯ + b̄∂c̄ + β ∂γ
¯ + β̄∂γ̄)
Sgh = d2 z(b∂c (6.36)
π

59
so that

TBgh = −2b∂c + c∂b − 23 β∂γ − 12 γ∂β (6.37)

TFgh = −2bγ + c∂β + 23 β∂c . (6.38)

For this superstring case,

I
1
d2 z cTBmatter + γTFmatter + bc∂c − 12 cγ∂β − 32 cβ∂γ − bγ 2


QBRST = (6.39)
2πi

How about the ghost vacuum? In the NS sector, the (β, γ) system has half-integer
moding on the cylinder so we have a two-fold degeneracy like we did for the (b, c) system
and the physical states have ghost number − 21 . In the R sector the Fock space built from
zero modes β0 , γ0 gives rise to an infinite degeneracy. These are referred to as different
pictures which are physically equivalent for most purposes and are labelled by an integer.
Picture-changing operators allow you to transform from one picture to another. In the story
of vertex operators, some restrictions apply on which pictures you can choose. You also have
to be a lot less cavalier about the conformal Killing vectors, moduli space, etc.
How about the critical dimension? If we worked out the boring details using SCFT
techniques we would find that the total central charge for the ghost system is −26+11 = −15.
Here the 11 comes from the (β, γ) portion. This means that if we include all the matter
fields – bosons and fermions – and the ghosts, the total central charge is

ctot = D(1 + 12 ) + (−26 + 11) . (6.40)

If we want the Weyl anomaly to vanish, we require that D = 10, which gives us the critical
dimension for superstring theories.
How about modings on the plane? The Ramond and Neveu-Schwarz boundary conditions
are, on the z plane,

Ramond (R): ψ µ (w + 2π) = ψ µ (w) (6.41)

Neveu-Schwarz (NS): ψ µ (w + 2π) = −ψ µ (w). (6.42)

Note that since we wish for our theories to have maximum Poincarè invariance, we require
X µ to be periodic (in other words, antiperiodicity breaks translation invariance). So we can
expand into modes on the plane by,
X X
ψµ = ψrµ z −r−1/2 ψ̃ µ = ψ̃rµ z̄ −r−1/2 (6.43)
r∈Z+v r∈Z+ṽ

where v, ṽ are 0 or 12 for R or NS BCs respectively (i.e. sum over integers or half-integers).
Following this expansion, we could expand our worldsheet supercurrent, TF out in terms of
its modes by X
TF (z) = Gr z −r−3/2 . (6.44)
r∈Z+v

60
 q P
0
We can also expand ∂X = −i α2 m αz −m−1 and TB = m Lm z −m−2 as usual. Then we
P

could use our OPEs that we discovered before and contour integrals to find the commutation
relations and anticommutation relations for the operators. These are given by
c
[Lm , Ln ] = (m − n)Lm+n + (m3 − m)δm+n,0 (6.45)
12
c 2
{Gr , Gs } = 2Lr+s + (4r − 1)δr+s,0 (6.46)
12
m − 2r
[Lm , Gr ] = Gm+r . (6.47)
2
For r,s integer, this is the Ramond algebra and for r,s half-integer, this is the Neveu-Schwarz
algebra. We also get their anti-holomorphic copies. We can work out an example,
I I
dw dz r+1/2 s+1/2
{Gr , Gs } = z w TF (z)TF (w) (6.48)
2πi Cw 2πi
I I  
dw dz r+1/2 s+1/2 D 2TB (w)
= z w + (6.49)
2πi Cw 2πi (z − w)3 z−w
I I  
dw dz D r+s−1 r+s+1
= (r + 1/2)(r − 1/2)w + 2w TB (w) (6.50)
2πi Cw 2πi 2
D
= (r + 1/2)(r − 1/2)δr+s,0 + 2Lr+s (6.51)
2
c
= (4r2 − 1)δr+s,0 + 2Lr+s , (6.52)
12
where we have used c = 23 D. Finally, we could expand these operators in terms of the mode
operators of the matter fields. We would find that
1X µ 1 X µ
Lm = : αm−n αµn : + (2r − m) : ψm−n ψµn : +am δm,0 (6.53)
2 n∈Z 4 r∈Z+v
X
Gr = αnµ ψµ r−n . (6.54)
n∈Z

Finally, in this section, let us say something about the states in the superstring SCFT.
Recall the commutation and anticommutation relations for Lm and Gr . If we look at what
the commutators are with L0 , we will find [L0 , Lm ] = −mLm and [L0 , Gr ] = −rGr . So we
see that the Gr operators will act as lowering operators as well (for m, r > 0). We can look
at our highest weight states as we did in the bosonic CFT. These should satisfy

L0 |hi = h|hi (6.55)

Lm |hi = 0 for m > 0 (6.56)
Gr |hi = 0 for r > 0. (6.57)

We will discuss the interpretation of G0 (which is in the R sector) later. In the bosonic case,
we found that unitarity forced non-ghost fields to have h, c ≥ 0. We can now use the Gr

61
operators to further constrict these inequalities. We consider the condition,

hh|Gr G−r |hi = hh|{Gr , G−r }|hi (6.58)

c
= hh|(2L0 + (4r2 − 1)|hi (6.59)
12
c
= 2h + (4r2 − 1) hh|hi ≥ 0.


(6.60)
12
For the NS sector, the lowest non-negative value for r is r = 1/2. In this case, we only require
h ≥ 0 as before. However, in the R sector the lowest is r = 0, so we see that h ≥ c/24.
To continue, let us focus on the NS sector. So we wish forPour two energy momen-
tum
P tensors to be regular on our vacuum. Recall that TB (z) = n Ln z −n−2 and TF (z) =
−r−3/2
r∈Z+1/2 Gr z . So then we require

Ln |0i = 0 n ≥ −1 h0|Ln = 0 n≤1 (6.61)

Gr |0i = 0 r ≥ −1/2 h0|Gr = 0 r ≤ 1/2. (6.62)

Suppose we have a primary conformal field of weight h, given by φ and its superpartner field
of weight h + 1/2, given by ψ. We can define the states

φ(0)|0i , ψ(0)|0i. (6.63)

We can recall that for our a primary fields that

[Lm , ψn ] = m(h − 21 ) − n ψn+m .

 
[Lm , φn ] = [m(h − 1) − n] φn+m (6.64)

However, with the supersymmetry generators, we have a different set of commutators. To

do this formally, we would define the supersymmetric conformal fields, show their supersym-
metry transformations, find OPEs for TF (z)φ(w) and TF (z)ψ(w), and then use these in the
commutator relationship using contour integration. The details are very similar in spirit to
previous computations shown. The results for the commutators are given by

[Gr , ψn ] =  m h − 21 − 12 n φn+m .


[Gr , φn ] = ψn+m (6.65)
2
where  is a constant Grassman parameter. Obviously we have seen that the state φ(0)|0i =
φ−h |0i = |hi is a highest weight state with conformal weight h. However, we want to see
how the new states given by ψ(0)|0i = ψ−h−1/2 |0i go. Note that [G−1/2 , φ−h ] = 21 ψ−h−1/2 .
So we see that

ψ(0)|0i = ψ−h−1/2 |0i (6.66)

= 2[G−1/2 , φ−h ]|0i (6.67)
= G−1/2 |hi. (6.68)

From here, we could use our commutators to work out further properties if desired.
Now we can look at the R sector. Things are a bit different here. We wish to find
a vacuum state, as we did before. However, we have forgotten to take into account our
supersymmetry. In the NS sector, supersymmetry is not broken by our definition of the

62
vacuum. However, we must be more careful here. We note that from our commutation
relations, we can see that G20 = L0 − 24 c
. G0 is a global supersymmetry charge. Therefore,
for our vacuum to have unbroken supersymmetry, we require that G0 |0iR = 0. However,
c
this implies that L0 |0iR = 24 |0i. This is a good thing however, as our unitarity condition
meant that even the vacuum had to have some non-zero weight and this vacuum satisfies
this condition. The other operators acting on the vacuum remains unchanged from the NS
conditions (other than the fact that Gr have r ∈ Z). This leads to an interesting point.
Since [G0 , L0 ] = 0, we can define two heighest wieght states,

|h+ i, |h− i = G0 |h+ i. (6.69)

This may look bad as there might be two different vacua for our theory. However, we note
that if |h+ i is the vacuum, then

hh− |h− i = hh+ |G20 |h+ i (6.70)

c
= hh+ |L0 − |h+ i (6.71)
24
= 0. (6.72)

So then the |h− i simply decouples from the physics and ignored (of course, this is only
true if the supersymmetry is unbroken). One final note. In order to introduce desired field
operators, we need to introduce the spin fields,

|h+ i = S + (0)|0i, |h− i = S − (0)|0i. (6.73)

These spin fields satisfy the OPEs

1
TF (z)S + (w) = S − (w) + less singular (6.74)
2(z − w)3/2
h − c/24 +
TF (z)S − (w) = S (w) + less singular. (6.75)
2(z − w)3/2

It is these fields that bring in our anti-symmetric boundary conditions ψ(ze2πi ) = −ψ(z).
As one can see, they have branch cuts, which are designed very carefully to ensure that the
mathematics on the complex plane gives the right physics. As the spin fields interpolate
between our NS vacuum and the R vacuum, we can define the action of the spin fields on
our fermionic operators as,
I
R ± dw N S
ψn S (w) = ψ (w)(w − z)n+h−1 S ± (z) (6.76)
2πi
There is much MUCH more to say about the physical and mathematical delights of the
superstring, but we have run out of time. I hope this aperitif was at least sufficient to pique
your interest. Next week, we will get to discussing D-branes and T-duality for open and
closed strings, and superstring duality with its connection to eleven dimensional M Theory.

63
7 T-duality, D-branes and superstring duality
In essence, T-duality is a symmetry that relies on the ability of strings to wind around
a circular dimension, something which particles obviously cannot do. The rough intuition
goes as follows. Strings, like particles, must fit an integer number of de Broglie wavelengths
around the circular dimension z in order to have a nonzero wavefunction. This requires the
KK momentum pz to be quantized in units of 1/R, where R is the radius of the S 1 . Winding of
a string around the circular dimension is also quantized, with the winding number labelling
topologically different sectors, and it literally represents the body of the string winding
around z an integer number of times before joining back up with itself. The ability of strings
to have KK momentum and winding simultaneously is what makes T-duality, the first string
duality, possible.

7.1 KK reduction in QFT

Before we dive in to the physics of putting one coordinate on a circle for string theory,
let us do a lightning review of the Kaluza-Klein procedure in regular old quantum field
theory. Let us split up D spacetime coordinates {xM } into d = D − 1 coordinates xµ
and the Dth coordinate z around the S 1 . The SO(D − 1, 1) symmetry is broken down to
SO(d − 1, 1) which we can use to classify the tensors. Then the D-dimensional metric tensor
can be decomposed into a d-dimensional metric tensor gµν , a vector gµz , and a scalar gzz .
The vector is known as the Kaluza-Klein gauge field and objects in the theory can carry
electric or magnetic charge. Now, what happens to the action principle when we have circle
compactification? If you work out the Ricci scalar in D dimensions, you find that it is a
sum of three terms written in the language of d dimensional tensors: (1) the Ricci scalar
in d dimensions, (2) a scalar field kinetic term, and (3) a KK gauge field strength term
with an exponential coupling to the scalar, which has a gauge symmetry descended from
D dimensional diffeomorphism invariance. (The presence of the scalar is mandatory, and it
is what threw the monkey wrench into the idea of unifying gravity and electromagnetism
by recruiting a fifth dimension.) The scalar φ, representing the size of the compactified
dimension, has no potential energy function. It is massless and called a modulus: the
different values of φ label the degenerate vacua of the theory. Furthermore, by just chasing
the constants, we can see quickly that
1 2πRz
= (7.1)
Gd GD
which provides the link between the Planck length `P,D in D dimensions and the Planck
length `P,d in d dimensions. Note that we use the convention 16πGD = (2π)D−3 `D−2 P,D .
Now suppose that we inspect the physics of a scalar field, which we take to be massless in
D dimensions for simplicity. Assume that it couples minimally to gravity so obeys a simple
Klein-Gordon equation. If we simplify further to the case where the spacetime metric is flat
in the z dimension, then we can expand the scalar in plane waves easy peasy,
X
φ(xµ , z) = φn (xµ )einz/R (7.2)
n∈Z

64
to find
n2
 
µ
∂ ∂µ − 2 φn = 0 . (7.3)
R
In order to describe the full D-dimensional dynamics for φ, we need to include the entire
infinite tower of KK modes. Were we to chop off the tower at some finite level, we would lose
information and we would also find that gauge invariances interrelating different KK modes
would break.
If we work at such low energies that the ambient energy budget is much lower than the
gap energy 1/R, then all we will be left with for fluctuating degrees of freedom is the zero
modes, i.e., we can treat the physics as d dimensional instead of D dimensional. This is
a one-dimensional prototype of the idea of string compactification: start in the native ten
dimensions, find the low-energy Lagrangian using the beta-function story we will describe in
the next lecture, and then compactify dimensions to find the physics of the four-dimensional
fields appropriate to low energy scales.
One final physical remark: notice that we do not need z ∈ (−∞, +∞) to produce
momentum eigenstates on S 1 . The operators O` = ei`x0 for l ∈ Z are well-defined, and using
the formal canonical commutation relation [x0 , p] = i it follows that O` p (O` )† = p + `/R
i.e. (O` |pi) has momentum (p + `/R).

7.2 T-duality for closed bosonic strings

Now let us get more serious and ask about compactifying the 25th direction of bosonic string
theory X 25 on a circle of radius R. Putting it on a circle does not change the wave equation
satisfied but it does change the boundary conditions. This means that our old friend the
mode expansion will change when we do physics on compact spaces rather than non-compact
ones. Let us see how this process works. We have
r r
α 0 α0 µ
X 25 (z, z̄) = 21 (xµ + x̃µ ) + −i (α0µ + α̃0µ ) τ + (α − α̃0µ ) σ + oscillators . (7.4)
2 2 0
We also know that for closed strings
r
α0 µ 1
p = 2 (α0 + α̃0µ ) . (7.5)
2
Further,
q we have that the X µ are periodic under σ → σ + 2π: X µ (z, z̄) → X µ (z, z̄) +
0
2π α2 (α0µ − α̃0µ ). If we demand that X µ is single-valued like for a noncompact direction in
target space, then we get back familiar relations between α0µ , α̃0µ , and pµ . But here we have
a qualitatively and quantitatively different situation: p25 has values n/R for n ∈ Z, and
X 25 (z, z̄)(τ, σ + 2π) = X 25 (τ, σ) + 2πRw . (7.6)
This gives us enough information to solve for the α025 and its barred cousin,
r 0 r
α0

25 n wR α
α0 = + 0 ≡ pL (7.7)
R α 2 2
r 0 r
α0

n wR α
α̃025 = − 0 ≡ pR (7.8)
R α 2 2

65
Then
2 25 2 4 2 25 2 4  
m225 µ
≡ −p pµ = 0 (α0 ) + 0 (N − 1) = 0 (α̃0 ) + 0 Ñ − 1 (7.9)
α α α α
which can be rearranged to give

n2 w 2 R2 2  
m225 = + + N + Ñ − 2 (7.10)
R2 (α0 )2 α0
0 = n · w − (N − Ñ ) . (7.11)

The second equation here is known as the momentum constraint and it allows N and Ñ to
differ as long as there is a nonzero dot product between momentum and winding numbers.
The first equation is again known as the mass formula. Notice that by having integer
quantized momentum we lost some states in the theory, but by having integer quantized
winding we gained some states in the theory.
Clearly, as R → ∞, momentum modes are cheap and winding modes are expensive. In
the opposite limit R → 0, winding modes are cheap and momentum modes are expensive.
In fact, the entire spectrum of our circle-compactified X 25 has a symmetry under

n ↔ w (7.12)
√
R α0
√ ↔ (7.13)
α0 R

This is known as T-duality. Note that this is stated as a symmetry of the free spectrum
of closed string theory, but it actually can be proven to hold at the level of the string path
integral. √
Notice further that at R → α0 something very special happens. There, and only there,
at the self-dual radius, new massless modes appear! They have the quantum numbers
n = −w = +1 and N = 1, Ñ = 0 and n = +w = +1 and Ñ = 1, N = 0. No other
choices satisfy the momentum constraint. These extra massless vectors (vectors because
they have one oscillator) are the W-bosons of an enhanced√SU (2) symmetry from wrapped
fundamental strings at the self-dual radius. When |R| > α0 , this SU (2) is Higgsed. The
vertex operators for these extra gauge bosons are
√
SU (2)L : ¯ µ ∂X 25 ,
∂X ¯ µ e±2iX 25 (z)/
∂X α0
; (7.14)
√
¯ 25 , ±2iX 25 (z̄)/ α0
SU (2)R : ∂X µ ∂X ∂X µ e . (7.15)

I have a few notes online at http://ap.io/2406s/notes/Hagedorn.pdf describing the

Hagedorn transition in string theory which arises because string states at high mass level
become exponentially numerous, and this density of states can at a high enough temperature
overwhelm the Boltzmann weight and render the statistical mechanical partition function
undefined. I am not willing to re-type this information in LATEXhere so please click on the
above link to see more.
For open strings, there is no longer a conserved winding number labelling topologically
distinct sectors. We will still be able to figure out the physics, by using the following really
cool fact about T-duality.

66
7.3 T-duality for open bosonic strings
T-duality corresponds to a worldsheet symmetry transformation

∂α X µ ↔ αβ ∂ β X µ . (7.16)

Notice that this worldsheet Hodge duality transformation switches N and D boundary
conditions for open strings. This is what we will use in a second to figure out how T-duality
works for open strings. The previous equation can be recast as
0
X 25 = X 25 (z) + X 25 (z̄) ↔ X 25 = X 25 (z) − X 25 (z̄) . (7.17)
0
You should explicitly check for yourself that the X 25 also satisfies the worldsheet Klein-
Gordon equation, by antisymmetry of the Levi-Civita pseudotensor αβ .
For the open string we write

X µ (z, z̄) = X µ (z) + X µ (z̄) (7.18)

where
r
µ 1 0µ 0 µ α0 X 1 µ −n
X (z) = (xµ + x ) − iα p ln(z) + i α z , (7.19)
2
2 n6=0 n n
r
1 0µ α0 X 1 µ −n
X µ (z̄) = (xµ − x ) − iαpµ ln(z̄) + i ᾱ z̄ , (7.20)
2
2 n6=0 n n

0 25
Now put X 25 on a circle S 1 of radius R. We can find the T-dual X by using (7.17),
0 25
X (z, z̄) = X 25 (z) − X 25 (z̄) (7.21)
0 25
X1 √
= x − iα0 pµ ln(z/z̄) + 2 2α0 αn25 e−inτ sin(nσ) (7.22)
n6=0
n
0 n √ X 1
= x 25 + 2α0 σ + i 2α0 αn25 e−inτ sin(nσ) (7.23)
R n6=0
n

Look what just happened! There is no τ dependence in the zero mode part of the oscillator
0
expansion. Physically this means that there is no momentum in the T-dual X 25 . Also, note
that the oscillator terms do not contribute at σ = 0, π, in other words, there are no wiggles
at the endpoints either! This is the origin of the famous fact about D-branes that for open
strings their endpoints must be stuck to a codimension one surface, known as a D24-brane.
It is codimension one because we only compactified one coordinate on a circle. Looking a
bit more closely, we see that
0 25 0 25 2παn
X (τ, σ = π) − X (τ, σ = 0) = = 2πnR0 . (7.24)
R

67
 Now let us turn to the topic of Chan-Paton factors, which are gauge labels carried by open
string endpoints without energy cost, and Wilson lines. What is a Wilson line? For Ha U (1)
µ
gauge field, you can think of it as a little bit like an Aharonov-Bohm phase: W ≡ eiq Aµ dx .
(If we had a non-Abelian gauge field, then we would need to path order the exponential.)
Here, let us consider the following configuration of a U (1) gauge field A25 which has zero
field strength,
θ
A25 ≡ − (7.25)
2πR
where θ is the angle on the S 1 direction X 25 . The resulting Wilson line is W = e−iqθ .
Now suppose that we have something a bit more interesting,
diag{θ1 , . . . , θN }
A25 = (7.26)
2πR
which breaks U (N ), the gauge symmetry appropriate to N D24-branes in unoriented string
theory, down to U (1)N generically. Under X 25 → X 25 + 2πR, we will pick up a phase in our
fields of diag{e−iθ1 , . . . , e−iθN }, so that open string momenta are now fractional. This has a
simple meaning: the endpoints are no longer on the same D24-brane. More generally, if you
have |iji Chan-Paton factors, the phase you get is ei(θi −θj ) giving momenta
(2πn + θi − θj )
. (7.27)
2πR
Then
0 25 0 25
X (τ, σ = π) − X (τ, σ = 0) = (2πn + θi − θj )R0 , (7.28)
so that the endpoint for state i is
0 25
X = θi R0 = 2πα0 A25 ii (no sum) . (7.29)
So our U (N ) Chan-Paton factors distinguish which D24-brane our oriented open strings end
on. There are N 2 degrees of freedom because there are N 2 ways of arranging the D24-branes.
For example, with N = 3 D24-branes, there are ij̄ strings for i, j = 1, 2, 3.

68
 The key thing about D-branes discovered by Polchinski in 1995 is that they carry target
spacetime Ramond-Ramond charge. This provided the missing link in the story of the
spectrum of charge-carrying states in superstring theory.

7.4 Unit conventions and dimensional reduction

The following figure from Polchinski is our focus for the rest of today’s lecture.

For units, we will be using the conventions of the Polchinski textbook because they are
by far the most widespread ones used. The fundamental string tension is
1 1
τF1 = 0
≡ . (7.30)
2πα 2π`2s

while the Dp-brane tension (mass per unit p-volume) in superstring theory is
1
τDp = , (7.31)
gs (2π)p `p+1
s

and the NS5-brane tension is

1
τNS5 = . (7.32)
gs2 (2π)5 `6s
In ten dimensions the Newton constant G is related to the gravitational coupling κ and gs , `s
by
16πG10 ≡ 2κ210 = (2π)7 gs2 `8s . (7.33)
To get units convenient for T-duality, we define any volume V to have implicit 2π’s in it.
If the fields of the theory are independent of (10 − d) coordinates, then the integration

69
measure factorizes as d10 x = (2π)10−d V10−d dd x. We can use this directly to find any
R  R

lower-dimensional Newton constant from the ten-dimensional one, as follows:

G10
Gd = 10−d
, (7.34)
(2π) V10−d
The Planck length in d dimensions, `d , is defined by

16πGd ≡ (2π)d−3 `dd−2 . (7.35)

From these facts we can see that there is a neat interdimensional consistency in the expression
for the Bekenstein-Hawking entropy. Let us take a black p-brane and wrap it on T p to make
d = 10 − p black hole. Translational symmetry along the p-brane means that the horizon
has a product structure, and so the entropy is
Ad+p Ad (2π)p Vp
SBH = =
4Gd+p 4Gd+p (7.36)
Ad
= ,
4Gd
which is the same as the black hole entropy.
As a reminder, we mention that the event horizon area in the Bekenstein-Hawking for-
mula must always be computed in the Einstein frame, which is the frame where the kinetic
term for the metric is canonically normalized,
√
Z
1
Sgrav = −gR[g] . (7.37)
16πGd
The relation between the Einstein and string metrics is

gµν = e−4Φ/(D−2) Gµν . (7.38)

Figuring out the constants is only one small part of the mechanics of dimensional reduc-
tion. We now move to a simple example of Kaluza-Klein reduction of fields in string frame,
by reducing on a circle of radius R. Label the d dimensional system with no hats and the
(d − 1) system with hats. Split the indices as {xµ } = {xµ̂ , z}. The vielbeins decompose as

Ĝµ̂ν̂ + e2χ̂ µ̂ Âν̂ e2χ̂ µ̂

 â χ̂   
a

ʵ̂ e µ̂
Eµ = ⇒ (Gµν ) = , (7.39)
0 eχ̂ Âν̂ e2χ̂ e2χ̂

and
1
Φ = Φ̂ + χ̂ ; (7.40)
2
which yield
√
Z
1
dd x −Ge−2Φ RG =
16πGd
Z q   (7.41)
1 d−1 −2Φ̂ 2 2 1 2χ̂ 2
d x −Ĝe RĜ + 4(∂ Φ̂) − (∂ χ̂) − e |dÂ| .
16πGd−1 4

70
The Kaluza-Klein procedure can also be done in Einstein frame. Taking the metric
 2
ds2 = e2αχ̂ dŝ2 + e2β χ̂ dz + µ̂ dxµ̂ , (7.42)

with β = (2 − D)α and α2 = 1/[2(D − 1)(D − 2)] gives

√ p
−gRg = −ĝ Rĝ − 12 (∂ χ̂)2 − 41 e−2(D−1)αχ̂ F 2 ,


(7.43)
where F is the field strength of Â.
What actions do you get at low energy for superstrings? There are too many cases to
discuss in detail here but we Rwill show IIA for purposes of illustration. TheR 11D M theory
√ 1 2 1
action is even simpler, S11 = −g{R(g) + 48 |dA3 | + (gravitino)} + 96 A3 ∧ dA3 ∧ dA3 .
For Type IIA, the independent R-R potentials are C1 , C3 . The low-energy effective action
of IIA string theory is d = 10 IIA supergravity,
√
Z  −2Φ  
1 10 e 2 1 2
SA = d x −G RG + 4 (∂Φ) − |dB2 | + (fermions)
(2π)7 ls8 gs2 12
 Z
1 2 1 2 1 1
− |dC1 | − |dC3 − dB2 ∧ C1 | + dC3 ∧ dC3 ∧ B2 .
4 12 (2π)7 ls8 64
We have shifted the dilaton field so that it is zero at infinity. Note that the dilaton kinetic
energy looks like it has the wrong sign, but this is fixed once you recognize that an integration
by parts needs to be done in order to write it in canonical form; in Einstein frame it has
the correct sign. Note also that in the action we could have used the Hodge dual ‘magnetic’
(8−n)-form potentials instead of the ‘electric’ ones, e.g. a 6-form NS-NS potential instead
of the 2-form. However, we cannot allow both the electric and magnetic potentials in the
same Lagrangian, as it would result in propagating ghosts. The funny cross terms, such
as dC3 ∧ dC3 ∧ B2 , are required by supersymmetry. In many cases there is a consistent
truncation to an action without the cross terms, but compatibility with the field equations
has to be checked in every case. For Type IIB string theory, the R-R 5-form field strength
F5+ ≡ dC4 is self-dual, and so there is no covariant action from which the field equations can
be derived. It is the equations of motion that are more fundamental: they are derived from
the beta function equations for target spacetime fields that we will describe next week.
Now recall that in d = 4 electromagnetism, an electrically charged particle couples to A1
(or its field strength F2 ), while the dual field strength ∗ F2 gives rise to a magnetic coupling
to point particles. By analogy, a p-brane in d=10 couples to Cn=p+1 electrically, or C7 −p
magnetically. As a result, we find 1-branes “F1” and 5-branes “NS5” coupling to the NS-NS
potential B2 , and p-branes “Dp” coupling to the R-R potentials Cp+1 (or their Hodge duals).
What happens when we probe a spacetime, using a Dp-brane? The kappa symmetric
action of a probe brane in a supergravity background has two pieces,
Sprobe = SDBI + SWZ , (7.44)
which are, to lowest order in derivatives,
Z
1 p
SDBI = − p+1 dp+1 σe−Φ − det P (Gαβ + [2πFαβ + Bαβ ]) ,
gs (2π)p ls
Z (7.45)
1
SWZ = − P exp (2πF2 + B2 ) ∧ ⊕n Cn .
(2π)p lsp+1

71
where the σ are the worldvolume coordinates and P denotes pullback to the worldvolume
of bulk fields. The brane action encodes both kinetic and potential information, such as
which branes can end on other branes. The WZ term, in particular, encodes the fact that
Dp-branes can carry charge of smaller D-branes by having worldvolume field strength F2
turned on.

7.5 Spacetime fields exerted by BPS objects in string/M theory

The BPS M2-brane spacetime has worldvolume symmetry group SO(1, 2), and the transverse
symmetry group is SO(8). Let us define the coordinates parallel and perpendicular to the
brane to be (t, xk ) , x⊥ , respectively. Then, using these symmetries and a no-hair theorem,
the spacetime metric turns out to depend only on |x⊥ | ≡ r, and has the form
−2/3 1/3
ds211 = H2 dx2k + H2 dx2⊥ , A012 = −H2−1 . (7.46)
The fact that the same function appears in the metric and gauge field is a consequence of
supersymmetry. Note that the metric is automatically in Einstein frame because there is
no string frame in d=11. It turns out that supersymmetry alone is not enough to give the
equation that the function H must satisfy; rather, the supergravity equations of motion must
be used. One finds that H2 must be harmonic as it satisfies a Laplace equation in x⊥ . The
solution is
r6
H2 = 1 + 26 , where r26 = 32π 2 N2 `611 , (7.47)
r
where we remind the reader that `11 is the eleven-dimensional Planck length.
The BPS M5-brane has symmetry group SO(1, 5) × SO(5), and the metric is
−1/3 2/3
ds211 = H2 dx2k + H2 dx2⊥ , (7.48)
and the harmonic function is this time
r23
H5 = 1 + 3 , where r53 = πN5 `311 . (7.49)
r
In this case, the gauge field is magnetically coupled, F4 is proportional to the volume element
on the S 4 transverse to the M5-brane.
For the M2, the origin of coordinates r = 0 is singular and so there must be a δ-function
source there, to wit the fundamental M2-brane. This happens essentially because the M2-
brane is electrically coupled. The magnetically coupled BPS M5-brane is solitonic and
nonsingular, in that the geometry admits a maximal analytic extension without singularities.
However, the nonextremal version of the M5 has a singularity and does need a source. Near-
horizon, the M2 spacetime is AdS4 × S 7 and the M5 is AdS7 × S 4 . Since the M2 and M5 are
asymptotically flat, again we have interpolation between 2 highly supersymmetric vacua.
Let us now move down to ten dimensions. The symmetry for BPS Dp-branes is SO(1, p)×
SO(9 − p). In the string frame, the solutions are
1
  1
dS 2 = Hp (r)− 2 −dt2 + dx2k + Hp (r)+ 2 dx2⊥ ,
1
eΦ = Hp (r) 4 (3−p) , (7.50)
C01···p = gs−1 [1 − Hp (r)−1 ] .

72
The function Hp (r) is harmonic; it satisfies ∂⊥2 Hp (r) = 0,

cp gs Np `7−p
s
√
Hp = 1 + , cp ≡ (2 π)(5−p) Γ [ 12 (7 − p)] . (7.51)
r7−p
Note that the function Hp would still be harmonic if the constant piece, namely the 1, were
missing. The asymptotically flat part of the geometry would be absent for this solution.
The (double) horizon of the Dp-brane geometry occurs at r = 0, and in every case
except the D3-branes the singularity is located there as well. Hence, for the Dp-branes with
p 6= 3, the singularity is null. Since the singularity and the horizons coincide for these cases,
we may worry that the singularity is not properly hidden behind an event horizon, and
so perhaps it should be classified as naked. We therefore demand that a null or timelike
geodesic coming from infinity should not be able to bang into the singularity in finite affine
parameter. Interestingly, this condition separates out the D6-brane from the others as being
the only one possessing a naked singularity. For the D3-brane the dilaton is constant, and the
spacetime turns out to be totally nonsingular: all curvature invariants are finite everywhere.
This allows a smooth analytic extension inside the horizon, like the case of the M5-brane.
The near-horizon D3-brane spacetime is AdS5 × S 5 . The Penrose diagram for the D3 is like
that of the M5.
The F1 and NS5 spacetimes may be found by using the T- and S-duality formulæ that
we gave earlier. We can also show two purely gravitational objects
√ in the BPS spectrum of
±
string theory. Defining light-cone coordinates dz ≡ (t ± z)/ 2,

dSd2 = −2dz + dz − + dρ2 + ρ2 dΩ2d−3 .

(7.52)

This is the gravitational wave W, which has zero ADM mass in d dimensions. In d − 1
dimensional language,
Q2
Md2 = 0 = Md−12
− 2. (7.53)
R
The d − 1-dimensional charge is the z-component of the d-dimensional momentum.
The other purely gravitational object is the KK monopole. Labelling the five longitudinal
directions y1···5 , and the four transverse directions xi , i = 1, 2, 3, and z; the metric is
2
ds2 = −dt2 + dy1···5
2
+ H −1 (x) (dz + Ai dxi ) + H(x)dx21···3 ,
(7.54)
2∂[i Aj] (x) = ijk ∂k H(x) .

The Ai can be found via the curl equation, given that H = 1 + k/|x|. The periodicity of the
azimuthal angle (an isometry direction) must be 4π to avoid conical singularities.
We now turn to a lightning review of some common and useful dualities.

7.6 Superstring dualities and the deerskin diagram

Type IIA ↔ M-theory
The 11th coordinate x\ is compactified on a circle of radius

R\ = gs `s . (7.55)

73
The supergravity fields decompose as

2
ds211 = e−2Φ/3 dS10
2
+ e4Φ/3 dx\ + C1 µ dxµ
(7.56)
A3 = C3 + B2 ∧ dx\ .

where F4 = dA3 = e4Φ/3 (dC3 + C1 ∧ H3 ) + eΦ/3 H3 ∧ dx\ , and H3 = dB2 . We can turn
M-theory objects into Type IIA objects by pointing them in the 11th direction (.) or not
(↓).
W M2 M5 KK
. ↓ . ↓ . ↓ . ↓ . (7.57)
D0 W F1 D2 D4 NS5 D6 KK
S-duality of IIB
The low-energy limit of IIB string theory, IIB supergravity, possesses a SL(2,R) symmetry
(it is broken to SL(2,Z) in the full string theory). Define
 
−Φ dB2
λ ≡ C0 + ie and H ≡ . (7.58)
dC2

Under an SL(2,R) transformation represented by the matrix

 
a b
U= ∈ SL(2, R) , (7.59)
c d

the fields transform as

aλ + b
H →UH λ→ . (7.60)
cλ + d
The d = 10 Einstein metric and the self-dual five-form field strength are invariant.
A commonly considered Z2 subgroup obtains when C0 = 0. The Z2 flips the sign of Φ,
and exchanges B2 and C2 . The result is

D1 ↔ F 1 , D5 ↔ N S5 ; (7.61)

all others such as W and KK are unaffected, and the D3 goes into itself. The effect of this
Z2 on units is
1 1 1
g˜s = , g˜s 4 `˜s = gs 4 `s . (7.62)
gs
From this one can easily check that the tensions of e.g. F1 and D1’s transform into each
other under the Z2 flip.
T-duality
The operation of T-duality on a circle switches winding and momentum modes of fundamen-
tal strings (F1) and exchanges Type IIA and IIB. The effect on units is to invert the radius
in string units, and leave the string coupling in one lower dimension unchanged:

R̃ `s g˜ gs
= , q s =p , `˜s = `s . (7.63)
`˜s R
R̃/`˜s R/`s

74
T-duality does not leave all branes invariant; it changes the dimension of a D-brane depending
on whether the transformation is performed on a circle parallel (k) or perpendicular (⊥) to
the worldvolume. It also changes the character of a KK or NS5; doing T-duality along the
isometry direction (isom) of the KK gives an NS5. Summarising, we have:

Dp ↔ Dp − 1(k) or Dp + 1(⊥) , KK (isom) ↔ NS5 ; (7.64)

Everything else is unaffected.

Let z be the isometry direction. Then T-duality acts on NS-NS fields as follows:

e2Φ̃ = e2Φ /Gzz , G̃zz = 1/Gzz , G̃µz = Bµz /Gzz , B̃µz = Gµz /Gzz ,
G̃µν = Gµν − (Gµz Gνz − Bµz Bνz ) /Gzz , (7.65)
B̃µν = Bµν − (Bµz Gνz − Gµz Bνz ) /Gzz .
T-duality also acts on R-R fields, and the formulæ are a little more involved. For simple
situations involving no NS-NS B-field and no off-diagonal metric components, we have either
C̃n+1 =Cn ∧ dz (⊥) or C̃n ∧ dz=Cn+1 (k), as appropriate.
Note that if we do T-duality on a supergravity Dp-brane in a direction perpendicular
to its worldvolume, we are dualising in a direction which is not an isometry, because the
metric and other fields depend on the coordinates transverse to the brane. But the T-duality
formulæ for supergravity fields apply only when the direction along which the T-duality is
done is an isometry direction. If it is not, then we should first “smear” the Dp-brane in that
direction to create an isometry and then do T-duality.
Note also that in the presence of some branes, string momentum or winding number may
not be conserved, e.g. string winding number in a KK background. However, the conserved
quantity transforms as expected under T-duality.

8 Sigma Models and String Theory

GR teaches us that “matter tells spacetime how to curve, and spacetime tells matter how
to move”. Well, in string theory the graviton and the matter are included under the same
umbrella, so the equations for gravity in string theory are one giant self-consistency equation.
Morally speaking, the string does not like to propagate on any old spacetime. Instead, it
forces conditions on the background fields in order to be bona fide solutions of string theory.
The main question we want to answer in this chapter is: what happens when you do
not assume that the spacetime metric is the Minkowski metric? What if there is also a
nontrival Kalb-Ramond field turned on, and a dilaton? Our guiding principle will be to
demand conformal invariance of the action principle in the quantum theory. We will find a
conformal anomaly in a general background, which will give β-functions for our couplings
telling us how the couplings run with energy scale. These β-function equations will then
be re-interpreted as equations of motion for the spacetime fields like the string metric. In
other words, this chapter is about giving you an outline of how the [NS-NS part of the] 10D
SUGRA action mentioned in the previous chapter is derived from first principles.
It will take a bit of work, but the physics principles involved are not actually that
complicated. The procedure is just a little long-winded for a first-year graduate student!

75
That is why I will lead you through the details. The main ingredients that we have not
yet recruited are the background field method from QFT and the Riemann normal
coordinate expansion from GR. You can find a good review of the background field method
pitched towards gauge field theories like the Standard Model in Peskin and Schroeder, and
a discussion of Riemann normal coordinates in any decent book on GR.
Sometimes, critics of string theory harp on about how this is not a background indepen-
dent formalism. Most of them misunderstand the degree to which the σ-model worldsheet
approach is background independent. The key thing is that the fluctuations of the back-
ground are included in the analysis. So the worldsheet formulation of string theory does
have a degree of background dependence, but it is a lot less naı̈ve than most dilettantes
believe. I also like to ask the question: what would background dependence even look like
when there is no classical background around which to expand, i.e. if you are in a truly
quantum phase of quantum gravity? Using GR intuition is unlikely to be a good guide to
that physics.

8.1 Sigma Model couplings

The worldsheet theory of strings propagating in nontrivial backgrounds is described by a
σ-model. Sigma models are used in many other contexts in both relativistic quantum field
theory and and nonrelativistic many-body theory. For now, we stick with the string theory
application.
Recall our original Polyakov action,
√
Z
1
SP = 0
d2 ξ γγ ab ∂a X µ ∂b X ν Gµν (X). (8.1)
4πα
It is this term that gives us our graviton. For the purposes of the string worldsheet, the
spacetime background in which the string propagates, Gµν (X λ ) is a coupling function of the
primary dynamical variables X λ (τ, σ).
Suppose that we also had an antisymmetric coupling on the worldsheet as well. It would
need to be reparameterization invariant. The only term that matches all the criteria is
Z
1
SAS = 0
d2 ξab ∂a X µ ∂b X ν Bµν (X). (8.2)
4πα
Here note that ab is the two-dimensional antisymmetric Levi-Civita symbol, which is a tensor
√
density (i.e. has a factor of γ in it). We recognize that this will be the Neveu-Schwarz
(NS) two-form that we love. Note that it indeed is invariant under our spacetime ”gauge
transformations”,
Bµν → Bµν + ∂µ Γν − ∂µ Γν . (8.3)
Of course, renormalization will bring in terms that are of lower dimension. Another
two-dimensional term is given by
√
Z
1
SD = d2 ξ γR(2) Φ(X). (8.4)
4π
We can see that this is the dilaton. There is a slight problem here. This is not Weyl invariant!
The important thing to note is that we will be expanding in powers of α0 , which means that

76
this term will only be relevant at one loop level and so our overall theory will be invariant
at the tree level. This is fine, since we already have an anomaly at one loop, so we can just
include it to be a term than needs to cancel.
One final term of dimension zero is given by
√
Z
1
ST = d2 ξ γT (x). (8.5)
4π
This is the tachyon. It is reparameterization invariant and not Weyl invariant as above. The
tachyon is needed to remove quadratic divergences in the vacuum. However, we will not
need to use them for our discussion.
If we were to look at open strings, we could include a vector gauge field that would
couple to the end of the strings. The action for such a coupling would take the form of
dX µ
I
SA = i dsAµ (X) . (8.6)
ds
Then this would modify boundary conditions that are need to keep conformal invariance
there. To avoid this discussion, we will just work with closed strings.
So our full action will be of the form,

A[X, γ] = SP + SAS + SD . (8.7)

The general idea is to do the analysis loop by loop, and the loop counting parameter is
α , i.e. the inverse string tension. Our analysis will only go to one loop in α0 . There is also
0

the string loop expansion in gs , which we do not cover here.

8.2 Background field method and Riemann normal coordinates

There are two pieces of QFT technology that are extremely useful for figuring out how
the sigma model action can give rise to equations of motion for spacetime fields. The first
is the background field method. Basically, it is done by expanding out the fields into a
classical component and a quantum component and shifting the integration to the quantum
component. Let us first see how this works for Yang-Mills gauge fields, which should be
familiar to those of you taking QFT2. Suppose we have some gauge field A which we
separate into background plus fluctuations: A = Ā + Q. We can look at how it transforms
under infinitesimal tranformations. In Peskin and Schroeder conventions,
1
δAaµ = ∂µ αa + f abc Abµ αc . (8.8)
g
Then we see that we can define,
1
δ Āaµ = ∂µ αa + f abc Ābµ αc (8.9)
g
δQµ = f abc Qbµ αc
a
(8.10)

which are consistant with the original gauge transformations. Notice that the ‘background’
part involving Ā transforms in an entire familiar way. The really cool part is that the

77
‘quantum’ part involving Q transforms tensorially, not like a connection. This simplifies the
process of handling the gauge invariance.
To do the same sort of thing for the string, define
X µ = X0µ + π µ (8.11)
where X0µ will be our classical component and π µ will be the quantum. Our path integral
will then become
Z δA[X,γ] µ
−(A[X0 +π,γ]−A[X0 ,γ]− d2 ξ
R
µ π )
δX0
Ω[X0 , γ] = [Dπ]e . (8.12)

Note that we simply expanded the action out into quadratic or higher terms in π µ . This
makes sense, since the order zero terms would just be a constant and the linear terms are
just proportional to the classical equation of motion, which the classical part satisfies.
The second method is the idea of Riemann normal coordinates. Consider our quantum
field π µ = X µ − X0µ . This is the difference of two coordinates, which is generally not a vector
under Lorentz transformations (in the X µ , it is a scalar under Lorentz tranformations for
ξ). To make sure we do this all covariantly, we should use a field that is a spacetime vector.
We will do this by choosing a tangent vector to the spacetime geodesic that runs through
X µ and X0µ .
To set up the geodesic, have the geodesic λ. Choose an affine parameter t normalized so
that λ(0) = X0µ and λ(1) = X µ . Define the tangent vector η = λ̇(0), where f˙ = df dt
for any
f . The geodesic equation then reads,
λ̈a + Γabc λ̇b λ̇c = 0. (8.13)
We can build up a Taylor series expansion around t = 0 using this equation. Consider the
geodesic equation at t = 0,
λ̈a (0) + Γabc (0)η b η c = 0 ⇒ λ̈a (0) = −Γabc (0)η b η c . (8.14)
Next, we take a single derivative with respect to t on the geodesic equation and then set
t = 0. Doing this we find,
...a a b c a b c
λ (0) + 2Γbc (0)λ̈ (0)η + Γ̇bc (0)η η = 0 (8.15)
...a
⇒ λ (0) = 2Γabc (0)Γbde (0)η e η d η c − ∂d Γabc (0)η d η b η c . (8.16)
We can continue to do this to find the nth derivative as we like. Putting down up to the
order we have,
1 1
λµ (t) = X0µ + η µ t − Γµσ1 σ2 η σ1 η σ2 t2 − (∂σ1 Γµσ2 σ3 − 2Γνσ1 σ2 Γµσ3 ν )η σ1 η σ2 η σ3 t3 + ... (8.17)
2 6
Generalizing to higher orders, we can define a symbol Γµσ1 ...σn = ∇0σ1 ...∇0σn −2 Γµσn−1 σn . This is
a bit of an abuse of notation, since covariant derivatives are not defined on symbols that are
not tensors. In this case, we mean that ∇0σ acts as if the symbol to the right of it was a true
tensor in only the lower indices. Now, if we set t = 1 we will find,
1
π µ = η µ − Γµσ1 σ2 η σ1 η σ2 + ... (8.18)
2
78
In order to get that π µ = η µ , we need for the Christoffel symbols and the symmetric part of
Γµσ1 ...σn to vanish. Luckily, we can do this, but only at a point (which we will choose to be
X0 ). The coordinates that this occur in are called the Riemann normal coordinates. These
coordinates have some great properties. A few examples include,

Γ̄µνσ = 0 (8.19)
Γ̄µ(σ1 ...σn ) = 0 (8.20)
µ
R̄νσρ = ∂σ Γ̄µνρ − ∂ρ Γ̄µνσ (8.21)
1 µ
∂ν Γ̄µσρ = (R̄σνρ − R̄ρνσµ
). (8.22)
3
Keep in mind these only hold at the origin. Of course there are far more, but we won’t put
them all here. We can of course expand all quantities out into normal coordinates through
their Taylor series. For a general covariant tensor,
∞
X 1
T̄µ1 ..µn = (∂ν1 ...∂νm T̄µ1 ..µn )η ν1 ...η νn (8.23)
m=0
m!

We want to work in a manifesitly covariant formulation. To transform the partial derivatives

to covariant, we also make use of the properties of normal coordinates. For example, a two
index covariant tensor,
1 1 ρ 1 ρ
T̄µν (X) = T̄µν + ∇λ T̄µν η λ + (∇λ ∇σ T̄µν − R̄λµσ T̄ρν − R̄λνσ T̄µρ )η λ η σ + ... (8.24)
2 3 3
where all of the T̄µν on the right side are evaluated at X0 .

8.3 Graviton beta function

So now we will start expanding out our actions using these two methods, which will allow
us to pick off the propagators and vertices for our Feynman diagrams. Then we can start
looking at loops and calculate the beta functions for each of our propagators. First, consider
our spacetime metric,
1 1 ρ 1 ρ
Gµν (X) = Gµν + ∇λ Gµν η λ + (∇λ ∇σ Gµν − Rλµσ Gρν − Rλνσ Gµρ )η λ η σ + ... (8.25)
2 3 3
1
= Gµν − (Rνλµσ + Rµλνσ )η λ η σ + ... (8.26)
6
1
= Gµν − Rνλµσ η λ η σ + ... (8.27)
3
where we have kept in mind that the covariant derivative of the metric vanishes. Next, we
expand the derivatives,

∂a X µ = ∂a (X0µ + π µ ) (8.28)
1
= ∂a (X0µ + η µ − Γµσ1 σ2 η σ1 η σ2 + ...) (8.29)
2

79
where we have used equation 8.18. Expanding this out, we find
1
∂a X µ = ∂a X0µ + ∂a η µ − ∂a Γµσ1 σ2 η σ1 η σ2 − Γµσ1 σ2 ∂a η σ1 η σ2 + ... (8.30)
2
1
= ∂a X0µ + ∂a η µ − ∂ρ Γµσ1 σ2 ∂a X ρ η σ1 η σ2 − Γµσ1 σ2 ∂a η σ1 η σ2 + ... (8.31)
2
1
= ∂a X0µ + ∂a η µ − ∂ρ Γµσ1 σ2 ∂a X0ρ η σ1 η σ2 − Γµσ1 σ2 ∂a η σ1 η σ2 + ... (8.32)
2
We recall equations 8.19 and 8.22, which will kill the final term and simplify the derivative.
So the final equation reads,
1
∂a X µ = ∂a X0µ + ∇a η µ − Rσµ1 ρσ2 ∂a X0ρ η σ1 η σ2 + ... (8.33)
3
With these two expantions in play, we can expand the Polyakov action out to quadratic
order in η µ . Doing this we can see that
√
Z
1
SP [X] = 0
d2 ξ γγ ab ∂a X µ ∂b X ν Gµν (X) (8.34)
4πα
√
Z
1 1
= 0
d2 ξ γγ ab (∂a X0µ + ∇a η µ − Rσµ1 ρσ2 ∂a X0ρ η σ1 η σ2 + ...) (8.35)
4πα 3
1 1
∗ (∂b X0ν + ∇b η ν − Rσν 1 ρσ2 ∂b X0ρ η σ1 η σ2 + ...)(Gµν − Rνλµσ η λ η σ + ...)
3 3
√
Z
1
= d2 ξ γγ ab (∂a X0µ ∂b X0ν Gµν (X) + 2∂a X0µ ∇b η ν Gµν (8.36)
4πα0
+ ∇a η µ ∇b η ν Gµν − Rµλνσ ∂a X0µ ∂b X0ν η λ η σ + ...)

where we have used our index symmetry to collect a few terms. The first and second terms
will not contribute to our quantum system because they do not depend on η µ . We argued
why earlier with the idea that we only need quadratic term or higher in this expansion. The
two quadratic terms are interesting. The first is clearly a kinetic term with the second a
vertex term between two X0 background fields and two η µ fields. We can clear the kinetic
up a bit using vielbeins, eiµ (x0 ). We introduce them by defining,

η i = eiµ (X0 )η µ . (8.37)

These satisfy the immensely useful property of

eiµ (X0 )ejν (X0 )δij = Gµν (X0 ). (8.38)

Considering the kinetic term, we see that

∇a η µ ∇b η ν Gµν = (∇a η)i (∇b η)i , (8.39)

with (∇a η)i = ∂a η i + ωµij ∂a X0 η j . Here, ωµij is the spin connection in spacetime, which you
need in order to take covariant derivatives of spinors. It obeys mathematical equations known
as the Cartan structure equations. If you have a spacetime which is torsion-free, then using

80
the language of differential forms we can write ω ij ≡ ωµij dxµ and the torsion-free equations
become dei +dω i j ∧ej = 0 and the relationship between the spin connection and the Riemann
tensor two-form is Rij = dω ij + ω i k ∧ ω jk . For a simple diagonal spacetime metric, it is often
faster to find the vielbeins, then get the spin connection by using the torsion-free equation,
then compute the Riemann two-form, than to compute using the Christoffel connection. Just
saying! :D

8.4 Kalb-Ramond field beta function

Anyways, we move onto expanding out the other two terms. If we expand the antisymmetric
term,
1 1 ρ 1 ρ
Bµν (X) = Bµν + ∇λ Bµν η λ + (∇λ ∇σ Bµν − Rλµσ Bρν − Rλνσ Bµρ )η λ η σ + ... (8.40)
2 3 3
Unlike our previous case, the terms with the covariant derivatives do not vanish. So we can
exand as before,
Z
1
SAS = 0
d2 ξab ∂a X µ ∂b X ν Bµν (X) (8.41)
4πα
Z
1 1
= 0
d2 ξab (∂a X0µ + ∇a η µ − Rσµ1 ρσ2 ∂a X0ρ η σ1 η σ2 + ...) (8.42)
4πα 3
1
∗ (∂b X0ν + ∇b η ν − Rσν 1 ρσ2 ∂b X0ρ η σ1 η σ2 + ...)(Bµν + ∇λ Bµν η λ
3
1 1 ρ 1 ρ
+ (∇λ ∇σ Bµν − Rλµσ Bρν − Rλνσ Bµρ )η λ η σ + ...)
2 Z 3 3
1
= d ξ ∂a X0µ ∂b X0ν Bµν + 2∂a X0µ Bµν ∇b η ν + ∇λ Bµν ∂a X0µ ∂b X0ν η λ
2 ab
(8.43)
4πα0
1
+ Bµν ∇a η µ ∇b η ν + 2∂a X0µ ∇λ Bµν ∇b η ν η λ + ∂a X0µ ∂b X0ν (∇λ ∇σ Bµν
 2
ρ ρ
− Rλµσ Bρν − Rλνσ Bµρ )η λ η σ

where we have used our antisymmetry properties to collect a few terms. As before, the first
two terms do not contribute.
However, the quadratic terms look quite a mess. Furthermore, they are not in a nice
gauge invariant form, as we would expect given that we have not broken the gauge symmetry
here. We wish to write it in terms of the field strength, Hµνλ = ∇µ Bνλ + ∇ν Bλµ + ∇λ Bµν ,
which is gauge invariant. To do this, we need to integrate by parts. The quadratic term
comes out to
Z  
1 2 ab µ ν λ 1 ν λ µ σ
d ξ Hµνλ ∂a X0 ∇b η η + ∇µ Hνλσ ∂a X0 ∂b X0 η η (8.44)
4πα0 2
Of course we can introduce our vielbeins again and get
Z  
1 2 ab µ i j 1 ν λ i j
d ξ Hµij ∂a X0 ∇b η η + ∇i Hνλj ∂a X0 ∂b X0 η η . (8.45)
4πα0 2

81
We may end up needing a cubic term too.
Z
1
d2 ξab Hijk ∇a η j ∇b η k η i . (8.46)
12πα0

8.5 Dilaton beta function

With that done, we can expand our dilaton term. It is a scalar term, so the expansion is
straight forward,
1
Φ(X) = Φ + ∇µ Φη µ + ∇µ ∇ν Φη µ η ν + ... (8.47)
2
So the dilaton action expands out like,
√
Z
1
SD = d2 ξ γR(2) Φ(X) (8.48)
4π
√
Z
1 1
= d2 ξ γR(2) (Φ + ∇µ Φη µ + ∇µ ∇ν Φη µ η ν + ...) (8.49)
4π 2
2 √ √
Z Z
1 (2) 1
= SD [X0 ] + d ξ γR ∇i Φη + i
d2 ξ γR(2) ∇i ∇j Φη i η j + ... (8.50)
4π 8π
Now we have all the terms that we will need for our loop calculations.

8.6 Weyl anomaly at one loop

Now we turn to focusing on working through the Feynman diagrams that contribute to the
Weyl anomaly at the zeroth order in α0 (classical level is 1/α0 ) and to first order in Rµνσλ .
Recall that the conservation of the energy momentum that arises from variations of the
worldsheet momentum,
∇z̄ hTzz̄ i + ∇z hTzz i = 0. (8.51)
We will work out our loops and find that at one loop, the ∇z hTzz i term contributes a non-zero
value, which implies that the trace must have a non zero value to satisfy the conservation of
energy-momentum. This is an interesting dualism between the reparameterization invariance
(which gives us our conservation equation) and the Weyl invariance (which would give us
a non-vanishing trace). We can have one or the other but not both. Since we demand
reparameterization invariance, we will have our Weyl anomaly which we will have to set to
zero to remove.
To begin with, we will consider a flat, Minkovski worldsheet to start and consider cur-
vature later (hint, it will involve the dilaton). We will also work in momentum space. We
can translate our conservation equation into,

q+ hT−+ i + q− hT++ i = 0 (8.52)

Note that this is a bit confusing. The important things to remember are that

γ++ = γ−− = 0 (8.53)

γ+− = −1 (8.54)

82
 To determine the one loop contribution to hT−+ i (which is proportional to the trace)
from the Polyakov action. To do this, we will calculate the hT++ i diagram and use our
conservation equation. The diagram we will consider is

This diagram has a contribution of

Z 2
d l l+ (l+ + q+ )
2 2
{Rµν ∂a X0µ ∂ a X0ν }(q), (8.55)
2π l (l + q)

where q is the momentum carried away and l is the loop momentum.

At this point, we should slow down and explain where these terms came from. First off,
note that this has an insertion of ∂+ η i ∂+ η i . This comes from the fact that we are looking
at hT++ i. The term hT++ i also has a contribution proportional to Rµνσλ . But this would
make a diagram with two factors of the Riemann tensor, whereas we are only working to
first order.
Note that the integral comes in with a 1/4π 2 factor, the two propagators give a 4π 2 α02 ,
the vertex with background fields give a 1/4πα0 , and the insertion comes with a factor of
2/α0 . This gives us the appropriate prefactor.
Furthermore, we note that the propagators and the energy momentum insertion are both
a sum over η, and so our vertex will be summed over as well, which is where the Ricci tensor
(rather than Riemann) comes in.
The bottom factors of momentum, q 2 = γ ab qa qb , obviously come from the two propaga-
tors. The momentum on the top come from the vertex factor (i.e. ∂+ ηi ∂+ η i → l+ (l+ + q+ ).
With these things sorted out, we then we turn to evaluating the momentum integral.
Luckily we can use our loop integral calculation techniques. Let’s introduce a Feynman
parameter,

Z1
d2 l l+ (l+ + q+ ) d2 l
Z Z
l+ (l+ + q+ )
= dx (8.56)
2π l2 (l + q)2 2π (xl + (1 − x)(l + q)2 )2
2
0
Z1
d2 l
Z
l+ (l+ + q+ )
= dx (8.57)
2π (l + 2(1 − x)l · q + (1 − x)q 2 )2
2
0
Z1
d2 l
Z
l+ (l+ + q+ )
= dx . (8.58)
2π ((l + (1 − x)q)2 − (1 − x)2 q 2 + (1 − x)q 2 )2
0

So we shift the momentum integral, l → l0 = l + (1 − x)q and relabel it back to l. This gives

83
us,

Z1
d2 l l+ (l+ + q+ ) d2 l (l+ − (1 − x)q+ )(l+ + xq+ )
Z Z
= dx (8.59)
2π l2 (l + q)2 2π (l2 + x(1 − x)q 2 )2
0

Note that the integral with two powers of the momentum is proportional to the metric, γ++ ,
so it will vanish. Obviously the portion proportional to one loop momentum vanishes as
well. This leaves,

Z1 2
d2 l l+ (l+ + q+ ) d2 l x(x − 1)q+
Z Z
= dx (8.60)
2π l2 (l + q)2 2π (l2 + x(1 − x)q 2 )2
0
Z1 2
x(x − 1)q+ q+ q+
= dx 2
= 2 =− (8.61)
2x(x − 1)q 2q 4q−
0

So now we use our equation 8.52 and find that

1
hT−+ i = Rµν (X0 )∂a X0µ (ξ)∂ a X0ν (ξ). (8.62)
4

84
 We can now look at the contribution from the antisymmetric tensor coupling action.
The easiest diagram to examine is

This is almost exactly as the one we did before, except with a slightly different vertex
factor. One can easily replace it and find
1
hT−+ i = ∇λ Hλµν (X0 )∂a X0µ (ξ)∂b X0ν (ξ)ab . (8.63)
8
The other diagram to that we can get is from the other vertex,

This is slightly different from our previous diagram, but it is almost the same. The only
major difference is the vertex factor. As before, we sum over all the indices belonging to
contractions to η. Consider the following,

ab cd γbd Hµij Hν ij ∂a X0µ ∂c X0ν . (8.64)

Note that we know the Levi-Civita pseudotensor satisfies

ab cd γbd = γ ac , (8.65)

So we can see that the vertex factor will be proportional to

γ ac Hµij Hν ij ∂a X0µ ∂c X0ν = Hµij Hν ij ∂a X0µ ∂ a X0ν . (8.66)

The contribution is
1
hT−+ i = Hµij (X0 )Hν ij (X0 )∂a X0µ (ξ)∂ a X0ν (ξ). (8.67)
16
Now we turn to the contribution for the dilaton. The dilaton will add additional correc-
tions similar to the ones above but it will also add its own unqiue corrections. This is due
to the fact that the above work was done on a flat worldsheet and the dilaton still give a
contribution even if it is a flat worldsheet. (i.e. the action may be zero, but the variation
with respect to the metric does not need to be.) However, to find the additional dilaton
contributions, we will have to go beyond the flat background.

85
 To start, we will examine the energy momentum tensor that arrises from the dilaton in
flat space. This is given by
dil
Tab = (∂a ∂b − δab )Φ(X). (8.68)
The  is a d’Alembertian with respect to the worldsheet. We can do the trace and find,

Taa = (∂ a ∂a − δaa )Φ(X) (8.69)

= −Φ(X) (8.70)

So then we see that,

dil
T+− = ξ Φ(X(ξ)). (8.71)

We can expand this at X0 to see what terms are there.

ξ Φ(X0 ) = ∂ a ∂a Φ(X0 ) (8.72)

= ∂ a (∂a X0µ ∂µ Φ(X0 )) (8.73)
= ∂ a ∂a X0µ ∂µ Φ(X0 ) + ∂a X0µ ∂ a ∂µ Φ(X0 ) (8.74)
= ξ X0µ ∂µ Φ(X0 ) + ∂a X0µ ∂ a X0ν ∂µ ∂ν Φ(X0 ). (8.75)

This is not Lorentz invariant from the spacetime point to view. We could just try to put
it in a covariant form right now, but it will be more useful to use the classical equation of
motion for X0 , since we know that it is satisfied. Since we are working in a flat worldsheet
so far, we can ignore the dilaton action in deriving it. The equation is given by
1 µ
ξ X0µ = Γµλσ ∂a X0λ ∂ a X0σ − Hλσ ∂a X0λ ∂b X0σ ab . (8.76)
2
We can just throw this into our equation above and easily see that
1
ξ Φ(X0 ) = ∂a X0µ ∂ a X0ν ∇µ ∇ν Φ(X0 ) − ∇λ Φ(X0 )Hλµν (X0 )∂a X0µ ∂b X0ν ab (8.77)
2
These two terms should look very familar from the terms we did above. So to include this,
we just add them to get the total contribution from the flat worldsheet parts to the zeroth
power in α0 and to one power in Rµνσλ . The contribution is

1 G √ 1 B
hT−+ i = βµν ∂a X0µ ∂b X0ν γγ ab + βµν ∂a X0µ ∂b X0ν ab (8.78)
4 8
where
G 1
βµν = Rµν − Hµλσ Hν λσ + 2∇µ ∇ν Φ (8.79)
4
B 1
βµν = ∇λ Hλµν − ∇λ ΦHλµν . (8.80)
2
We are almost there! Our next task is to look at the contributions form the dilaton with
non flat worldsheet geometry and see the beta function that comes from that.

86
8.7 Nailing the dilaton
This part is focused on figuring out the final, and arguably the most important, contribution
to the Weyl anomaly. This contribution is from the dilaton, and it is a bit more involved.
To start, we will backtrack a bit and look at the energy momentum for the dilaton. The
action for the dilaton is given by
√
Z
1
SD = d2 ξ γR(2) Φ(X). (8.81)
4π
So if we to a variation with respect to the metric to get the engery momentum tensor,
1 √ √
δL = (δ( γ)R(2) Φ(X) + γδ(R(2) )Φ(X)) (8.82)
4π
We know from GR that the variations take the form of
√ 1√
δ( γ) = − γγab δγ ab , δ(R(2) ) = Rab δγ ab + δ(Rab )γ ab (8.83)
2
In the Einstein Hilbert action, the term proportional to the variation of the Ricci tensor
produced a total derivative that we dropped. Here, we do not have that luxury. We found
that the variation took the form of

δ(Rab )γ ab = ∇a (γbc ∇a δγ bc − ∇b δγ ab ) (8.84)

= ∇a ∇a γbc δγ bc − ∇a ∇b δγ ab (8.85)

Since this is now multiplied by Φ(X) it is no longer a total derivative! Putting this together,
1√ 1
δL = γ(− γab R(2) Φ(X)δγ ab + Rab Φ(X)δγ ab + (∇a ∇a γbc δγ bc − ∇a ∇b δγ ab )Φ(X)) (8.86)
4π 2
We need to remove the covariant derivatives off of the variations of the metric. So we
integrate by parts twice and find,
1√ 1
δL = γ(− γab R(2) Φ(X) + Rab Φ(X) + (γab ∇c ∇c − ∇b ∇a )Φ(X))δγ ab (8.87)
8π 2
We can simplify this. Note that the first two terms together are proportional to the Einstein
tensor, which vanishes identically for any 2D worldsheet. This implies that,
δL 1√
ab
= γ(∇a ∇b − γab ∇c ∇c )Φ(X). (8.88)
δγ 4π
The energy momentum tensor is then

dil 4π δL
Tab =√ (8.89)
γ δγ ab
= (∇a ∇b − γab ∇c ∇c )Φ(X) (8.90)

87
 Now we can return to where we left off. We wish to see what the effect of a curved
worldsheet on these beta functions. We can get some of the pieces by using two point
functions on a flat worldsheet.
Recall that the conformal metric can be written as γab = eφ δab . Now we wish for the
conformal anomaly to vanish for any generic worldsheet. However, we will soften this result
to approximate and only demand that the first variation vanishes. We can see that

δ  1
hT−+ (0)ieφ δab  = − hT−+ (ξ)T−+ (0)iδab . (8.91)
δφ(ξ) φ=0 4π

To find out this contribution, there will be three different types of terms. The first
type is the one loop corrections, which will be of order zero in α0 . The second type will be
two loop diagrams corrections, which will be first order. The third type will be tree and
one loop corrections using the dilaton energy momentum instead of just the Polyakov and
antisymmetric tensor, which will be of first order. As before, we will use the conservation of
energy momentum to allow us to calucated tensor diagrams of the + + ++ and transform
them to the + − +− character.

88
 There is only one diagram at one loop with no dilaton terms to consider.

It is similar to what we have seen before. The contribution here is

l2 (l+ + q+ )2 3
Z
πD q+
hT++ (q)T++ (−q)i = 2D d2 l + 2 = − . (8.92)
l (l + q)2 6 q−

There are few things to note. Since there is no momentum carried off, the energy momentum
insertions must have opposite momentum contributions. The factor of D is there because
there are D η i , each of which have a diagram like this. Note that the two T+ + insertions
bring 4/α02 , the momentum integral brings 1/4π 2 , the two propagators give 4π 2 α02 , and we
have a 1/2 as a symmetry factor. This is what gives us the 2 out front and why there this
contribution is zeroth order in α0 .
We can now use the fact that q+ T−+ + q− T++ = 0 twice and we can see that,
πD
hT−+ (q)T−+ (−q)i = − q+ q − . (8.93)
6
We would like return to coordinate space. We note that q 2 = −2q+ q− . So then we can easily
see that
πD (2)
hT−+ (ξ)T−+ (0)i = δ (ξ). (8.94)
12
With this in hand, we can use equation 8.91 and by inspection see that
D
hT−+ (ξ)ieφ δab = − φ(ξ). (8.95)
48
We recall our earlier equation for the curavture and find that our final expression is
D √ (2)
hT−+ (ξ)ieφ δab = γR . (8.96)
24

89
 We will now discuss the two loop contributions. Getting the correct forms for these would
be tricky, as it would involve calculating two loop diagrams. The ones that are relavent for
the dilaton correction at first order in α0 are

α0 1 √
hT−+ (ξ)ieφ δab = (−R + Hµνσ H µνσ ) γR(2) (8.97)
8 12
Finally, we have to calculate two point functions with insertions of the dilaton energy
momentum. The easier diagram is actually a tree diagram, since it involves two insertions
of the dilaton energy momentum. This is simply a connection between two vertices

Here the X0 field are allowed to be internal lines, since tree level is the classical level.
This gives

α0 √
dil
hT−+ dil
(ξ)T−+ (0)i = πα0 ∇µ Φ∇µ Φδ (2) (ξ) ⇒ hT−+ (ξ)ieφ δab = ∇µ Φ∇µ Φ γR(2) (8.98)
2
The other diagram with an insertion of the energy momentum is a one loop with only
one insertion of the dilaton field. The diagram is of the form

90
 This gives

0 (2) α0 √
dil
hT−+ (ξ)T−+ (0)i µ
= −πα ∇µ ∇ Φδ (ξ) ⇒ hT−+ (ξ)ieφ δab = − ∇µ ∇µ Φ γR(2) (8.99)
2
Now that we have all the pieces, we can take a look at the form of the Weyl anomaly.
We can finally see that putting everything together,
1 G √ 1 B 1 √
hT−+ i = βµν ∂a X0µ ∂b X0ν γγ ab + βµν ∂a X0µ ∂b X0ν ab + β Φ γR(2) (8.100)
4 8 4
where
G 1
βµν = Rµν − Hµλσ Hν λσ + 2∇µ ∇ν Φ (8.101)
4
B 1
βµν = ∇λ Hλµν − ∇λ ΦHλµν (8.102)
2
− 26 α0
 
φ D 1 µνσ µ µ
β = + −R + Hµνσ H + 4∇µ Φ∇ Φ − 4∇µ ∇ Φ (8.103)
6 2 12

Note that we have added and extra term that is of the form − 26 6
. This comes from the
ghost system.
In order to ensure that the Weyl anomaly is not present in our quantum system, we need
that the anomaly is free at all orders. We will set these beta functions to zero (to remove
the anomaly) and see what equations result.
G B
To start, consider a flat, free worldsheet. Then we see both βµν and βµν are zero.
Φ
However, this is not true for β . In order to remove the anomaly, we need to set D = 26.
This is called a critical string theory, since the central charge is zero. It is possible to allow
D 6= 26, which are non-critcal string theories, but we would need to introduce Liouville
terms in order to compensate. These are signifcantly more complicated than critical string
theories, and we will not get into detail here.
Of course, there is a slight trouble that we might be overlooking. If we were to look back
at our central charge, we would note that it is proportional to β Φ . So the central charge
is in actuality an operator (even though we sometimes call the eigenvalue and the operator
the same name). So what if the central charge operator has eigenvalues that depend on
G B
position? The theory manages to safely avoid this by noticing that if both βµν and βµν are
φ
zero, then β is a constant. So we are indeed safe to return to consider the central charge to
be a c-number. Furthermore, it is possible to prove that this conclusion is safe to all orders.
Now that we have set D = 26, we can go ahead and consider a curved worldsheet with

91
our fields. Then we have three conditions,
1
0 = Rµν − Hµλσ Hν λσ + 2∇µ ∇ν Φ (8.104)
4
1 λ
0 = ∇ Hλµν − ∇λ ΦHλµν (8.105)
2
1
0 = −R + Hµνσ H µνσ + 4∇µ Φ∇µ Φ − 4∇µ ∇µ Φ. (8.106)
12
Now suppose I take the trace of equation 8.104. I would find,
1
0 = R − Hµλσ H µλσ + 2∇µ ∇µ Φ. (8.107)
4
I now use this equation to eliminate the R from the the bottom equation. We then have the
conditions
1
0 = Rµν − Hµλσ Hν λσ + 2∇µ ∇ν Φ (8.108)
4
∇ Hλµν = 2∇λ ΦHλµν
λ
(8.109)
1
∇µ ∇µ Φ − 2∇µ Φ∇µ Φ = − Hµνσ H µνσ (8.110)
12
We could also then reintroduce the curvature scalar using equation 8.107 times 21 Gµν . Finally,
we then get
1 1 1 1
Rµν − Gµν R = (Hµλσ Hν λσ − Gµν Hλρσ H λρσ ) − 2(∇µ ∇ν Φ − Gµν ∇ρ ∇ρ Φ) (8.111)
2 4 2 2
λ λ
∇ Hλµν = 2∇ ΦHλµν (8.112)
1
∇µ ∇µ Φ − 2∇µ Φ∇µ Φ = − Hµνσ H µνσ (8.113)
12
These are the equations we were looking for the whole time. They are indeed equivalent
G
to the conditions to the vanishing of the Weyl conditions, βµν B
= βµν = β Φ = 0. However
they are in a more suggestive from. Note that on the left hand side of equation 8.111, we
have the Einstein tensor. If the right hand side side was the energy momentum tensor of
some action, and the following two conditions were equations of motion, then we know we
could satisfy these conditions simulanteously.
The effective action that reproduces these equations is
√ −2Φ
Z  
D 2 1 2
S = d X Ge R + 4(∇Φ) − H . (8.114)
12

This is the NS sector of the action in the string frame we were looking at in a previous
report!
In the Einstein frame, this action is given by
 
D √
Z
4 2 1 − D−2
8Φ
2
S = d X g R− (∇Φ) − e H . (8.115)
D−2 12

92
 This should look familiar now. The first term is the Einstein-Hilbert action that gives
us our regular GR and the other terms are just fields that live in the spacetime (although
they are coupled quite strangely). So string theory really is a theory of gravity!
So we finally arrive at the main point. We started out with a reparameterization invariant
and Weyl invariant (to 1/α0 ) string action on the worldsheet. We looked at the quantum
picture and found that if we demand reparameterization invariance holds, there is an Weyl
anomaly. If we demand that the Weyl invariance holds to all orders of α0 , then not only do
we get a dimensionality for our spacetime, but we also have a low energy effective action that
reproduces gravity and other fields. So we can see that string theory is a theory of gravity
and other fields.

9 Compactification
9.1 Hiding extra dimensions
Essential problem we face in building real-world models: taking our string theory action
principle defining the UV physics and flowing down to the phenomenologically relevant IR.
This is generically messy and hard.
β-functions tell us that our 10D spacetime M10 must obey the string theory equations
of motion. String theorists commonly take a direct product (KK) ansatz M10 = M4 × K6 ,
where K6 is some compact six-manifold. The details of what theory emerges on M4 depends
on the details of the manifold K6 – not every 6-manifold is a solution. The low-energy
theory in 4D that you get also depends on which superstring theory you chose: the precise
structure of the 10D SUGRA field equations (the antisymmetric tensor fields especially)
depends delicately on whether you have IIA, IIB, I, HE, or HO theory.
One challenge in model-building is getting the right spectrum of gauge forces and
quark/lepton matter. Before the advent of D-branes, a no-go theorem prevented compacti-
fications of IIA or IIB producing chiral fermions in 4D – this is why heterotic string theory
was so popular in the first superstring revolution.
One of the most difficult aspects of building a credible compactification scenario is
stabilizing all of the moduli, scalar fields which do not develop a potential to any order
in perturbation theory but which we know must be absent from the low-energy massless
spectrum. Recruit nonperturbative physics to fix.

9.2 SUSY
Both the geometry and topology of K6 play an important role in what low-energy physics
we end up getting on M4 . This originates in the fact that the 10D string theory worldsheet
β-function equations are very picky about the spacetimes on which strings can propagate –
if we make a product space ansatz it must be compatible with the field equations.
In particular, the number of light generations of fermion fields depends sensitively on
the holonomy of the Calabi-Yau in KK compactifications. In the first superstring revolution,
we discovered how to explain 3 generations in terms of the mathematics of one special type
of Calabi-Yau.

93
 Why Calabi-Yaus? These are manifolds with special holonomy which support the
existence of Killing spinors, allowing SUSY to be present in the 4D theory. SUSY is not
an observed low-energy symmetry in Nature (so far), but it is technically important in
controlling UV physics. SUSY should be at most N = 1; models with N ≥ 2 have unrealistic
spectra.
SUSY introduces at least one extra scale in the problem (masses of superpartners),
and this can actually permit Grand Unification. Just extrapolating Standard Model gauge
couplings up to higher energy scales does not yield a GUT; this was proved experimentally
via precision electroweak measurements at LEP in the tunnels now occupied by LHC.

9.3 Brane world models

Alternatives to Calabi-Yau compactifications of heterotic string theories?
Use Type IIA/IIB and add D-branes, fluxes, and orientifolds, which are objects
which possess both a negative charge and a negative tension. N.B.: orientifolds do not
destabilize the vacuum because they are fixed planes of a symmetry: they cannot fluctuate
physically. So their negative tension is harmless. The physically crucial thing is that these
negative tension objects which are fundamentally string theoretic let you evade previous
no-go theorems which prevented building de Sitter compactifications in string theory.
Brane world idea # 1: we ‘effectively’ compactify the physics using a brane world
(e.g. Randall-Sundrum) type model. These have a warped product space structure, in which
the overall scale of the 4D geometry depends on the coordinate in the compact dimension.
For example, in the RS models the bulk has Λ < 0 while the brane is Minkowski, and the
radius of curvature of the AdS space provides an effective compactification radius.
Brane world idea # 2: build models where Standard Model gauge and matter fields
are restricted to the worldvolume of an intersecting D-brane configuration. We are made of
open strings stretched between various pairs of D-branes. Only closed strings (gravitons)
can move off-world.

9.4 Heterotic string theory on CY3 s

−
The 10D gravity multiplet comes from the NS-NS sector and has the fields {GM N , BM N , Φ, ψM , λ+ }
− −
where ψM is the gravitino and λ is the dilatino. (c.f. Jesse’s Final Project presentation.)
Since the heterotic string theory already has a big gauge symmetry in 10D, we also have the
vector multiplet {AM , χ− }, where χ− is the gaugino, and we have suppressed the Yang-Mills
indices.
The direct product ansatz M10 = M4 × CY3 is not necessarily enough to ensure that
the β-function equations are satisfied. For heterotic string theory on a CY3 , the Bianchi
identity like equation for B2 yields the important condition
α0
dH3 = [Tr (R ∧ R) − Tr (F ∧ F )] .
4
0
This implies that H3 6= dB2 , but rather H3 = dB2 + α4 (ΩL − ΩY M ), where ωL and ωY M are
the Lorentz and Yang-Mills Chern-Simons terms which play a central role in the analysis of
[gauge and gravitational] anomalies.

94
 Since Tr (R ∧ R) is nontrivial in cohomology (it is the 2nd Chern character of the
tangent bundle), it requires turning on a nontrivial background field strength in order to
satisfy Tr (R ∧ R) = Tr (F ∧ F ). We ‘embed the spin connection in the gauge group’, using
the fact of SU (3) holonomy group.

9.5 Calabi-Yaus
What is a Calabi-Yau manifold? It is a Kähler manifold with n complex dimensions and
vanishing first Chern class
1
c1 = [R] = 0 .
2π
To unpack this definition, let us start with the definition of a complex structure.
It is a rank (1,1) tensor field J such that J 2 = −1 when regarded as an isomorphism on the
tangent bundle. In physicist’s terms, J acts like i.
A Kähler manifold has three mutually compatible structures: a complex structure,
a Riemannian structure, and a symplectic structure. You have already met Riemannian
manifolds in GR. You have also met symplectic structures in classical mechanics in the part
about Hamiltonian dynamics on phase space.
We also need Chern classes, which are characteristic classes. These are topological
invariants associated to vector bundles on a smooth manifold.
A theorem conjectured by Calabi and proven by Yau says these requirements for
Calabi-Yaus imply SU (n) holonomy. (For non-compact case, care with BCs at ∞ is re-
quired to make this fly.) This implies that CYs admit a covariantly constant spinor. In
turn, this implies compactification on a CY leaves some SUSY preserved. Also, Ricci-flat,
in SUGRA approx.

9.6 Geometry
Consider the exterior derivative operator d = dxµ ∂µ . This object can be used to act on
differential forms, which are antisymmetric tensors of rank p. For example, in EM we
write F = dA where A = dxµ Aµ is the gauge potential and F the field strength tensor. The
Bianchi identity is extremely simple: dF = 0, and it expresses the fact that d2 = 0. Maxwell
equation also becomes v.simple: ∗d ∗ F = j, where j is the current. For higher p-forms, such
as NS-NS B2 and R-R Cp+1 , Ap = p!1 dxµ1 ∧ . . . ∧ dxµp Aµ1 ...µp where ∧ is the wedge product.
This operator that squares to zero should ring a bell vs. BRST cohomology. Indeed,
we can ask: are there differential forms that are closed (killed by d) but are not exact
(the d of something)? The pth de Rham cohomology group H p (K) is defined to be the
quotient of closed p-forms by exact p-forms and it tells us important topological information
for compactification.
Next, define the formal adjoint of d, known as d† ; the Laplacian on forms is then
defined by ∆ = dd† + d† d. This can be used to show that for each de Rham cohomology
class on K, there is a unique harmonic representative.
Calabi-Yau metric can be written locally in the form gab̄ = ∂za ∂z̄b K where K is the
Kähler potential. Then the Ricci form R = dz a dz̄ b iRab̄ is closed because Kähler manifolds
have no torsion. This is where the first Chern class arises from.

95
 For more details, see BBS Appendix to §9 starting on p.440, or §14 of BLT.

9.7 Hodge numbers

Betti numbers bp give fundamental topological information about a manifold. bp is dimen-
sion of pth de Rham cohomology H p (K) of manifold K. When K has a metric, Pkit counts
number of linearly independent harmonic p-forms on K. When K is Kähler, bk = p=0 hp,k−p
where the hp,q are the Hodge numbers counting the number of harmonic (p, q)-forms on
K. These beasts are very useful for helping figure out the spectrum of dimensionally reduced
fields.
A Calabi-Yau is [partially] characterized by its Hodge numbers. The properties of
CYs relate hp,0 = hn−p,0 (use c.c. of Ω, gab̄ ), hp,q = hq,p (c.c.), and hp,q = hn−q,n−p (Poincaré
duality). Any compact connected Kähler manifold has h0,0 = 1 (constant fns). Also, a
simply connected manifold has vanishing first homotopy group, and therefore has vanishing
first homology group. (Homology and homotopy are both about defining and categorizing
holes in a shape but capture different information.) This gives h1,0 = h0,1 = 0. So for n = 3,
the dimension of interest for us, we only need to specify h1,1 and h2,1 .

9.8 Mirror symmetry and the conifold

Calabi-Yaus with n = 1 are either C (non-compact) or T 2 (compact). For n = 2, you get (a)
products of n = 1 CYs, for non-compact cases, or for the compact cases either (b) T 4 or (c)
K3. For n = 3 the options become much more numerous, with examples including weighted
projective spaces. For most CYs the metric is not explicitly known, except in special limits,
which makes compactification life more interesting!
CYs are not completely characterized by their Hodge numbers. Indeed, there are pairs
of distinct CYs that have the same Hodge numbers and which obey mirror symmetry,
which is like a more powerful and complicated version of T-duality. It can be shown at the
level of the path integral that the string theories are physically identical for mirror pairs.
This involves some pretty math.
Physically, CYs can have deformations. These correspond to changing parameters
characterizing their shapes and sizes, described by moduli fields.
Branes wrapped on supersymmetric cycles can become very light under certain condi-
tions analogous to approaching the self-dual radius in circle compactified string theory. It is
critical to include these light modes in your low-energy effective action principle or you will
miss important nonperturbative effects. A signature example of this is the conifold transi-
tion which can, unlike flop transitions, change Hodge numbers. Need NP string theory, not

96
just SUGRA.

9.9 Dp-brane probes

Consider probing a big fat Dp-brane spacetime with a single ‘test’ Dp-brane. Its (kappa
symmetric) action in a SUGRA background has two pieces,
Sprobe = SDBI + SWZ ,
which are, to lowest order in derivatives, for the U (1) part,
Z
1 p+1 −Φ
p
SDBI = − d σ e − det P (Gαβ + [2πFαβ + Bαβ ]) ,
gs (2π)p lsp+1
Z
1
SWZ = − P exp (2πF2 + B2 ) ∧ ⊕n Cn .
(2π)p lsp+1
where the σ are the worldvolume coordinates and P denotes pullback to the worldvolume of
bulk fields.
The brane action encodes both kinetic and potential information, such as which branes
can end on other branes. The WZ term, in particular, encodes the fact that Dp-branes can
carry charge of smaller D-branes by having worldvolume field strength F2 (or B2 -fields)
turned on.
Works for a brane that is topologically R1,p or wrapped on tori. If the D-brane is
wrapped on a manifold which is not flat, extra terms arise in the probe action (e.g. ‘A-roof
genus’ for K3). Does not capture non-Abelian physics.

9.10 Example: 1 Dp probing N Dps

For our supergravity background exerted by N Dp-branes, we had
−1 +1
dS 2 = Hp 2 (−dt2 + dx2k ) + Hp 2 dx2⊥ ,
1
(3−p)
eΦ = Hp4  ,
C01···p = gs−1 1 − Hp−1 .


The physics is easiest to interpret in the static gauge, where we fix the worldvolume
reparametrisation invariance by setting X α = σ α̂ , α = 0, . . . p. We also have the 9 − p trans-
verse scalar fields X i , which for simplicity we take to be functions of time only, X i = X i (t),
i
i = p + 1 . . . 9. Denote the transverse velocities as v i ≡ dX dt
. Now we can compute the
pullback of the metric to the brane.
P (G00 ) = (∂0 X α )(∂0 X β )Gαβ + (∂0 X i )(∂0 X i )Gij
− 21 +1
= G00 + Gij v i v j = −Hp + Hp 2 ~v 2 ;
−1
P (Gαβ ) = Hp 2 .
The last ingredient we need is the determinant of the metric.
− 1 (p+1) p
q
− det P(Gαβ ) = Hp 4 1 − ~v 2 Hp .

97
9.11 Forces between D-branes
Putting this all together yields
Z
1 h i
dp+1 σ −Hp−1 1 − ~v 2 Hp + Hp−1 − 1 .
p
SDBI + SWZ =
(2π)p+1 gs `sp+1

From this action we learn that the position-dependent part of the static potential vanishes,
as it must for a supersymmetric system such as we have here. The constant piece is of
course just the Dp-brane tension. In addition, we can expand out this action in powers of
the transverse velocity. To lowest order,
Z
1
dp+1 σ −1 + 12 ~v 2 + O(~v 4 ) ,
 
Sprobe = p+1
(2π)p+1 gs `s

so the metric on moduli space, the coefficient of v i v j , is flat. This is a consequence of having
16 supercharges preserved by the static system.
In SUGRA, SUSY field transformations have a spinorial parameter . For preserved
SUSYs, you find a projection condition for Dp-branes. Schematically,

1 + [sgn(Z)] Γ01···p  = 0 .

So generically, Dp-branes break half SUSY. Whether or not Dp and Dq can be in equilibrium
is determined by calculation. Find: Dp k D(p + 4) is 1/4 SUSYic.

9.12 Open string BCFT

The specific form of the action we presented is valid for the U (1) part of the gauge field only.
For a stack of Dp-branes, generally in an oriented string theory we get U (N ) gauge group,
not U (1). The story of how the non-Abelian information is encoded in the DBI+WZ action
has been worked out in quite some beautiful detail, e.g. in some classic papers of Rob Myers
from the 1990s. We have suppressed it here in an attempt to keep the number of details
flying around more manageable.
The DBI+WZ action is actually far more than a kappa-symmetric action suitable for
D-branes in SUGRA backgrounds. It can be derived in a completely different way as a partial
resummation of open string corrections to the SY Mp+1 action for a stack of N Dp-branes.
This was done in the mid-1980s using worldsheet β-function techniques and starts with a
worldsheet coupling of the form
dX µ
I
ds Aµ (X) .
∂Σ ds
There is also an open string tachyon, but the GSO projection gets rid of it too.
Boundary CFT methods were exploited nicely in this context. You can find a great deal
more detail about how to calculate α0 corrections to the lowest-energy SY Mp+1 Lagrangian
using BCFT methods in BLT.

98
9.13 What is an orbifold?
Suppose that X is a smooth manifold with a discrete isometry group G. Then we can
construct the quotient space X/G. A point in that quotient space corresponds to an entire
orbit of points in X, i.e., a point and all its images under the isometry group. The quotient
space has singularities if nontrivial group elements leave points of X invariant (called fixed
points).
Locally, the orbifold is indistinguishable from the original manifold.
GR is ill-defined on singular spaces. But strings can actually propagate consistently
on manifolds with [spatial*] orbifold singularities! The essential physical reason for this is
that strings are extended objects, and so they have significantly softer behaviour at short
distance than particles do.
When you mod out by a symmetry, you lose some sectors of the theory, but for orbifolds
you also gain back some sectors known as twisted sectors, roughly speaking where the fields
only come back to themselves modulo a symmetry transformation. (This is morally similar
to what we saw in circle compactified string theory where KK momentum got quantized,
losing states, but we also developed a whole new sector of winding modes.)
* If you try to quotient timelike directions, you generically end up with closed timelike
curves. These are very bad for your credibility. Interestingly, people have managed to make
some sense of some null string theory orbifolds.

9.14 Simple compact and non-compact orbifolds

The simplest example of a compact orbifold is S 1 /Z2 , where the coordinate x is identified
with −x. This case is relevant to how the E8 × E8 heterotic string is obtained from M theory
– by compactifying on the interval S 1 /Z2 with end-of-the-world branes carrying the E8 × E8
gauge symmetry.

Alternatively, take the complex plane C and identify z with −z. This produces the
orbifold C/Z2 . What does this space look like? The orbifolding identifies the upper half
plane’s positive real axis with its negative real axis under a group transformation, and it is
a cone. The conical deficit angle is π. The point (0, 0) is a fixed point of the group action.
We could also consider C/ZN , where the group is generated by a 2π/N rotation. This
time, the singularity at the origin signifies a deficit angle of 2π(N − 1)/N and it is of AN
type (part of the ADE classification).

99
9.15 Spectra of states for orbifolds
Untwisted sector states are those which exist on X and are invariant under the symmetry
group G. You just take states Ψ such that gΨ = Ψ for all g ∈ G. If your group is finite, then
you can make a G-invariant state by starting with any representative Ψ0 and superposing
all the images gΨ0 .
Twisted sector states arise in the following way. For a closed string propagating
on an ordinary manifold, we know that translating σ by 2π brings the embedding map field
back to itself. But when you have an orbifold, you only need to produce the same map up
to a group transformation:
X µ (τ, σ + 2π) = g X µ (τ, σ) .
For orbifolds there are various distinct twisted sectors labelled by the group element used to
make the identification. In more fancy mathy language, they are labelled by the conjugacy
classes of G.
For the C/Z2 example, it is clear that twisted sector string states enclose the singular
point of the orbifold. In the quantum spectrum, individual twisted sector quantum states
are localized at the orbifold singularities that the classical configurations enclose. It is easy
to see this approximately for low-lying string states; harder for full-on oscillator content.
Orbifolding enables breaking of some SUSYs of the original manifold.

9.16 What is an orientifold?

The one superstring theory we have hardly talked about is Type I, open superstring theory.
It can be understood as arising from a projection of Type IIB.
For IIB theory, the two worldsheet superpartners of X µ have the same chirality. World-
sheet parity Ω : σ → −σ is therefore a Z2 symmetry of the theory, as it exchanges left- and
right-moving modes. The bright idea people had was to gauge this worldsheet parity sym-
metry, and this is what produces Type I from Type IIB.
Recall that T-duality switches the sign of the right-movers only. Then in the T-dual
picture, the symmetry transformation above becomes a product of world-sheet parity and a
spacetime reflection in directions that were T-dualized.
The closed-string spectrum is obtained by keeping states that are even under 21 (1 + Ω).
The projection condition kills the Kalb-Ramond field and leaves the string metric and dilaton
fields preserved. For the gravitini, only the sum of the two survives the projection. Similarly,
only one of the IIB dilatini survives, so there are overall 56 + 8 = 64 massless fermionic d.o.f.
How about the R-R sector? Requiring SUSY implies that there are 64 bosonic d.o.f.
A simple light-cone counting shows that C0 and C4 are projected out, leaving only the R-R
2-form C2 . This time, the counting is 35 + 1 + 28 = 64.

9.17 What are Op-planes good for?

One of the main applications orientifold planes have found in string compactifications is
to provide sinks for sources of flux. These arise because of tadpole cancellation conditions
stating that the total number of sources and sinks must be balanced out to zero overall.

100
Turning on fluxes is one way to help address the perennial problem of stabilization of moduli,
so it comes up often.
One application this found was to creating string theoretic models of Randall-Sundrum
compactifications involving warped product spaces. H.Verlinde showed how if you concen-
trated D-branes in a Calabi-Yau nearby one another, such as for D3-branes of the AdS/CFT
correspondence, you would develop a long AdS throat with a warp factor that scales expo-
nentially in the radial coordinate away from the D3-brane stack.
One may also ask whether Op-planes have other physical roles. They have a very
important property other than negative charge (see above): negative tension. If you are a
half-decent theoretical physicist, the idea of negative-tension objects should jolt you out of
your seat, because they will reliably destabilize the vacuum of your theory if they are allowed
to fluctuate. But since the Op-planes are actually fixed planes, they do not have a physical
interpretation familiar from classical physics. They never fluctuate. Ever.

9.18 Flux compactifications

Cycles in e.g. a Calabi-Yau manifold can have fluxes threading through them. What dimen-
sion of cycle you need is obtained by looking at how to integrate up your field strength or
its Hodge dual: Z Z
Fp+2 or ?Fp+2 .
γp+2 γD−p−2

Fluxes are actually quantized. If they are sourced by D-branes, then it is clear why
fluxes are integers: this corresponds to having an integer number of branes. For manifolds
of nontrivial homology, under special conditions, integer quantized fluxes can also be turned
on even when there are no brane sources. Flux quantization arises from the generalized
Dirac quantization conditions (electric charge e and magnetic charge g obey e · g ∈ 2πZ, via
holonomy).
What kinds of fluxes can be turned on, and how they can be mutually compatible,
depends sensitively on the type of superstring theory you start with and the compactification
you choose. Their options are highly constrained by the ultraviolet physics of the worldsheet.
One of the most difficult tasks in building a relatively realistic string compactification
is stabilizing the moduli. Two ubiquitous types of moduli are the dilaton field and the radial
modulus describing the overall scale of the CY.

9.19 Flux-ology
How do you decide what fluxes to turn on? First of all, if you want low-energy SUSY, you
have to ensure that their presence is consistent with the existence of covariantly constant
spinors. The fluxes end up appearing in covariant derivatives of spinors, contracted with
an antisymmetric combination of gamma matrices. (For example, in IIB compactifications
you find that the 3-form field strength needs to be imaginary self-dual to preserve SUSY
properly.) Tadpole cancellation conditions also provide an important constraint.
Flux compactifications typically produce a non-trivial scalar potential for the [would-
be] moduli fields. A handwaving description of this goes as follows. We may have a modulus

101
field ϕ that couples differently to two different fluxes,
Z
S ⊃ − e−ϕ |F1 |2 + e+ϕ |F2 |2 .

The value of ϕ can get dynamically fixed by the ratio of fluxes via the equations of motion.
Very few models have been constructed in which all moduli are stabilized without
nonperturbative effects. Noone knew how to do nonperturbative string theory with any
confidence until the second superstring revolution of the mid-1990s.

9.20 Inflation in string theory

BBS §10.7 has a nice quick introduction to early universe cosmology and inflation. Unfor-
tunately I have no time to give a quick review here so must refer you there. I can also
recommend the gigantic review on string theory modelling of early universe cosmology avail-
able from Baumann and McAllister, arxiv:1404.2601 for anyone who has an appetite for
more. It will be published soon in proper textbook form by Cambridge University Press.
String theory compactifications have many moving parts. It is possible to obtain
inflation, or a phenomenon producing many of the same effects in our uinverse today, using
quite a number of different string theoretic mechanisms. Examples of classes of models that
have been studied include the following (borrowed from the B-McA table of contents):-

• Inflating with unwarped or warped branes, such as D3-D7.

• Inflating with relativistic branes - Dirac-Born-Infeld style.
• Inflating with axions. Compactified Kalb-Ramond components routinely end up giving
rise to an axion field from string theory.
• Inflating with Kähler moduli.
• Inflating with dissipation.

You can also build ekpyrotic universe type models as well, although those suffer from
their own controversies.

9.21 KKLT and the Landscape

Can we build string models with small positive cosmological constant?
H.Verlinde late 90s: Consider a CY3 with a certain number of O3-planes. Tadpole
cancellation allows you to have a limited number of D3-branes living at various points in the
CY3. Idea: group enough D3-branes together to make a long AdS5 throat. This introduces
a way of understanding developing large hierarchies via exponentially large redshifting, à la
Randall-Sundrum.
[K]KLT early 2000s: put in an anti-D3 to break SUSY spontaneously and uplift AdS
vacua slightly to deS. Vacua obtained in this way are only metastable. But if their lifetime
is extremely long then it does not bother us. The big question is: how controlled is the
approximation of adding the anti-D3?

102
 The embarrassing thing about the second superstring revolution is that it eventually
yielded the stark realization that there is an extremely large number of possible Standard-
Model-like vacua in the theory. It looks very unlikely that our universe would be the only
solution to the fundamental equations of string theory. It now seems much more likely that
we are a statistical accident.
In this context, the value of the cosmological constant is seen to be a fortunate envi-
ronmental accident. (Hot tip: believe Susskind, not Smolin!)
String theorists who felt bereft after these revelations have satisfied themselves with
investigating the statistical properties of superstring vacua.

9.22 String Theory Landscape

Ur potential

Perhaps 101000 different minima

Lerche, Lust, Schellekens 1987

Bousso, Polchinski; Susskind; Douglas, Denef,…

Denef,…

10 AdS/CFT
10.1 Origin of AdS/CFT
AdS/CFT holography is an equivalence between gravitational (string) theories and non-
gravitational field theories in one lower dimension. It gave the first concrete working models
of the concept of holography for quantum gravity and grew out of studying nonperturbative

103
D-branes in string theory.
Consider a stack of Np Dp-branes. Our key idea will be to have large Np .

To see why, consider how the gravitational warping of a single Dp-brane scales. Correc-
tions to the flat Minkowski metric scale as
(GN )(τDp ) Np 1 gs Np
δgµν ∼ 7−p
∼ gs2 `8s p+1 7−p ∼
r gs `s r (r/`s )7−p
This shows that the variable controlling gravitational warping is gs Np .
So we can obtain a seriously large warping of flat spacetime by having Np become
parametrically large while keeping gs small.

10.2 Maldacena’s decoupling limit

Spacetime warping can be thought of as physics of closed strings. After all, the lowest
state of excitation of a closed superstring includes the graviton mode. We can also look at
the open string picture of Dp-brane physics.
Lowest mode of open superstring has spin-one and mass zero. Theory living on world-
volume of N Dp-branes is SYMp+1 + α0 corrections + gs corrections.
The variable controlling open string corrections when D-branes are present is actually gs Np ,
which is exactly what we found also controls spacetime warping.
In general, dynamics of the full closed + open string sectors is unknown (c.f. string
field theory). Maldacena’s key insight was to take a very special ‘decoupling limit’ of this
stack of Np Dp-branes, which turns off interactions between closed strings in the bulk and
open strings whose endpoints lie on the D-branes.

* Bulk
ne
Bra

10.3 AdS=CFT
In the decoupling limit, on the open string side you get just d = p + 1 SUSY Yang-Mills
living on the Dp-brane worldvolume. On the closed string side, the decoupling limit zooms
in on the near-horizon region of the geometry.

104
 Maldacena boldly realized that this decoupling limit meant that you could think about
the stack of N3 D3-branes either in terms of the near-horizon geometry or the worldvolume
Yang-Mills theory. But because it is the same stack of D3-branes either way, those repre-
sentations must be equivalent!
The near-core geometries of D3, M2, M5, and D1-D5 systems produce AdSp+2 ×S D−p−2
spacetimes, where D is the spacetime dimension, 10 in superstring theory and 11 in M theory.
You will see how this works in HW4. In brief: the S 5 factor arises for D3 because the r2 in
+1/2
the transverse piece of the metric proportional to dΩ25 is cancelled off by the H3 factor,
because for D3-branes the harmonic function scales as 1/r4 in the near-core region.
Yang-Mills theory on flat D3-branes has maximal SUSY, i.e. 16 supercharges. In
d = 4 language this translates into N = 4 SUSY. The supermultiplet has adjoint fermions
and scalars partnering up with the adjoint vector field.
The field content of this theory is so special that it ends up having zero β-function. It is
actually a conformal field theory. Hence, we say AdS=CFT.

10.4 Probes and IR/UV relations

AdS/CFT gives a definition of quantum gravity in asymptotically AdS space in Lorentzian
signature, via the duality and Wick rotation in the CFT.
AdS/CFT sheds light on the question of background independence in quantum gravity.
It is independent of bulk background except for its asymptotics which are locked down by
boundary physics.
Not yet known how to do a worldsheet analysis for string theory in AdS, because a
Ramond-Ramond flux is turned on. (c.f. v.technical Pure Spinor formalism of N.Berkovits,
which is v.different to the NSR and GS formalisms.)
One of the key ideas string theorists use every day is the concept of probes. Probes
are objects that exist in the theory and are used to interrogate the physics. For fundamental
string probes stretched between Dp-branes, field theory energy E is proportional to distance
r between D-branes in bulk,
r
E= .
2πα0
This is known as an IR/UV relation and is central to AdS/CFT. It says that the UV in the
field theory is welded to the IR in the bulk and vice versa.
Polchinski and I showed that gravitons have different IR/UV relations. Still, how deep
it penetrates into the bulk scales inversely with energy on the brane.

10.5 IMSY and applications of AdS/CFT

For more general Dp-branes, you find that the harmonic functions conspire to not give
you AdSp+2 × S D−p−2 spacetimes. Instead [for p < 5], you get a warped version where there
is an r- (and p-) dependent conformal factor. What does this mean for the possibility of a
correspondence like AdS/CFT?
The 10D bulk spacetime background is only reliable for certain ranges of r, because either
α0 R or eΦ is monotonically growing as r → 0 or r → ∞.
By dimensional analysis, the field theory coupling also runs with energy. 10D SUGRA

105
 2
geometry becomes unreliable when geff (E) := gs N (`s E)p−3 becomes of order unity, which is
where the SYM field theory description takes over.
This gives rise to the idea of a phase diagram for a D-brane/string system, which turns
out to be a very generic and physically rich feature in holographic setups. There can be more
than one description of the physics, but only one of them at a time can be weakly coupled.
AdS/CFT has been applied to modelling quark-gluon plasma and cond-mat, with lim-
ited successes. Can calculate transport properties; for cond-mat the physics is quite different
in spirit to the story of quasiparticles. Also inspired the study of dS/CFT correspondence
where the CFT is non-unitary.
H.Nastase review of AdS/CFT basics for beginners: 0712.0689.

10.6 Holographic dictionary

AdS/CFT is a strong/weak duality,

L4AdS 2 1
4
↔ gYM N, gs ↔ .
`s N

This is why it is so powerful: to learn about strongly coupled CFT we use weakly coupled
string theory in AdS and vice versa.
Isometries of the bulk match up with symmetries of the CFT.

1. The S 5 factor has a natural SO(6) isometry. This is is also the R-symmetry of the
N = 4 SUSY SYM theory.

2. The AdS5 factor has isometry group SO(4, 2). This is the conformal symmetry group
of N = 4 d = 3 + 1 SYM. SO(3, 1), the Lorentz group, is a subgroup.

One of the reasons why AdS/CFT works is that near infinity, the area element grows
like the volume element. The dimensionful constant of proportionality is furnished by the
radius of curvature of AdS.
Another very useful physics fact to know about AdS is that higher-` partial waves
do not fall off with higher powers of 1/r like they do in asymptotically Minkowski space.
Influences deep in the AdS bulk can be easily seen from infinity. This is related to the fact
that AdS has a timelike boundary.

10.7 Fefferman-Graham
Maldacena derived AdS/CFT for the gauge theory on the worldvolume living on Rp .
Technically, the spacetime obtained was the Poincaré patch of AdS. Witten taught us that
the correspondence actually extends to global AdS, which is dual to field theory living on
S p . Much beautiful physics derived.
In asymptotically AdS, Fefferman-Graham coordinates are very handy,

L2
ds = 2 dz 2 + gab (z, ~x)dxa dxb ,
2


z

106
where L is the AdS scale, and z = 1/r ∈ [0, ∞). gab (z, ~x) is the boundary metric which
obeys an expansion in powers of z/L. It is only fixed up to conformal transformation. Other
bulk fields obey a similar near-boundary expansion. In general, get an overall power of z/L
times a power law expansion with some logarithmic terms.
Bulk EOMs are 2nd order PDEs, which yield two independent solutions. Working
through the Fefferman-Graham details shows that coefficients of power law and log terms
are all determined in terms of the first nontrivial coefficient function of ~x. The key fact is
that this boundary value of the bulk field, up to an overall power of z/L, is interpreted as a
source term for a dual operator in the boundary field theory.

10.8 BF bound and GKP/W

Example: scalar of mass m. Get

φ(r, ~x) = (z/L)d−∆ {φ(0) (~x) + Lz φ(1) (~x) + · · · } 



+(z/L)2∆−d {φ(2∆−d) (~x) + · · · } ,

where ∆ solves m2 L2 − ∆(∆ − d) = 0, giving

p
∆ = d/2 ± (d/2)2 + m2 L2 .

∆ will turn out to be the scaling dimension of the dual operator in the CFT. Requiring
∆ ∈ R gives
m2 ≥ −(d/2L)2 .
This is the Breitenlohner-Freedman bound famous from SUGRA. Scalar fields in AdS can
be a little bit tachyonic, but not too much, or AdS goes unstable.
Gubser-Klebanov-Polyakov + Witten took this further, conjecturing
R d
Zstring z ∆−d φ(x, z)|z=0 = φ(0) (x) = he−S+ d xφ(0) O(x) iCFT
 

Prime example of operator dictionary: gµν (bulk) ↔ Tµν (boundary).

Non-normalizable modes: ↔ turn on ∆LCFT = αφ Oφ in boundary (irrelevant ops). Normal-
izable modes: ↔ turn on VEV hOφ i = βφ (relevant ops).

10.9 Holography and black holes

Nonlocal probes: correlation functions, Wilson loops, entanglement entropy. Witten dia-
grams: bulk-boundary propagators and bulk vertices.

107
 UV/IR relations taught us that high-energy in CFT corresponds to near-boundary in
bulk. Holographic RG: running understood via bulk Hamilton-Jacobi, including countert-
erms. Flows change interior of AdS.
A very deep fact about AdS/CFT is that it is not just a zero-temperature equivalence.
Asymptotically AdS black holes correspond to turning on finite temperature in the CFT.
The Hawking-Page transition from BH to hot AdS is dual to the deconfinement transition
in the boundary theory.
AdS/CFT shows us how one extra bulk spacelike dimension emerges, but generically
without a path integral proof as yet. It also does not show us how time emerges. That may
require going beyond QM as we currently know it.

10.10 Maldacena’s eternal AdS BH

Maldacena also proposed in hep-th/0106112 that the eternal AdS black hole is dual to
a direct product of two uncoupled conformal field theories, CF TL × CF TR , with thermal
entanglement between the L and R CFTs. Key feature: if you trace over either the left or
right set of degrees of freedom, then you obtain a thermal density matrix at the Hawking
temperature.
In the holography literature this is known as the ‘two-sided’ black hole. Black holes
formed in gravitational collapse are fundamentally different: they do not have a white hole
singularity or a past event horizon and are called ‘one-sided’.

Maldacena’s idea rests on an older construction by Werner Israel, a very famous Cana-
dian gravity theorist, called the thermofield double or TFD state,
√ X −βE/2
|TFDi = (1/ Z) e |ψiL × |ψiR .
i

10.11 Less symmetric holography

To use holography to model real-world systems, need to break increasing degrees of SUSY
and other symmetries. Big literature on this, divided into quark-gluon plasma modelling,
AdS/condensed matter; also, dS/CFT.
Fixing the asymptotics does not prevent you from having interesting phase transitions
originating in interesting hair on the bulk solutions. There are much bigger, wilder classes
of geometries available than previously imagined.

108
 Holography is applicable to systems other than N = 4 SYM. Sometimes in a bottom-
up setup we do not know what the dual QFT is, but we can still use holography to discern
universal aspects of strongly coupled systems.
Breaking boost: can get residual Schrödinger or Lifshitz symmetry. Breaking anisotropy
and homogeneity. Modelling superconductors, glasses, strange metals, Fermi surfaces, hy-
perscaling violation, disorder. Holographic lattices.
My interest in holography is more bottom-up style than top-down. Recent(ish) pa-
pers were on Lifshitz black holes [Bertoldi-Burrington-Peet] and on hyperscaling violating
crossovers [O’Keeffe-Peet]. Current preprint on holographic modelling of disorder almost
ready for posting [O’Keeffe-Peet].
Neat part about holography from the point of view of someone interested in gravity:
geometrizing phases of the dual QFT.

10.12 Higher-spin/vector holography

Higher-spin theory is a lab that may provide a bridge between classical gravity and full
quantum ST. Tower of modes. (c.f. ABJM, multi-M2, localization.)
Old theorems had put stringent constraints on low-E scattering in flat spacetime
that forbid m = 0 particles with spins s > 2 from participating in any interacting QFT.
But in AdS, Λ < 0 provides dimensional coupling and IR cutoff, which can reconcile HS
gauge symmetry w equivalence principle, giving M.Vasiliev’s nonlinear unfolded equations
of motion. Review: 1404.1948.
J.Maldacena-A.Zhibodaev 1112.1016 showed why for 3D CFT with higher-spin sym-
metry you get a free theory, found for the dual of Vasiliev higher-spin theory in AdS4 .
S.Giombi-X.Yin AdS4 VH review: 1208.4036.
R.Gokapumar-M.Gaberdiel 1207.6697 found the dual for AdS3 : a minimal model
coset CF T2 with WN symmetry at large-N . (Scalar accompanying graviton and HS fields
in 2D GG duality is massive; for AdS4 it is massless.)
GG actually got even further: they found that higher-spin in AdS3 is a subsector of
string theory, in the tensionless limit. Technically important.
Higher-spin theory in 3D has BHs 1208.5182. Existence of horizons and singularities
is not invariant under HS gauge transformations, but can define via holonomy if use Chern-
Simons formulation of 3D gravity, e.g. 1302.0816.

10.13 Bulk locality

AdS/CFT is a fascinating laboratory for studying process of thermalization. Study quenches,
try to extract universalities. Review: 1103.2683.
(Strong time dependence is harder than weak or none.)
See also fluid/gravity correspondence; review by Hubeny: 1501.00007.
(A.Hamilton-)D.Kabat-G.Lifschytz-D.Lowe showed in hep-th/0506118, hep-th/0606141,
1102.2910 how local operators in the AdS bulk can be represented via smeared operators in
the CFT. Only regions in the causally relevant zone contribute. Their construction can be
obstructed if there are bulk normal modes with exponentially small near-boundary imprint,
such as for the AdS black hole 1304.6821. Is bulk locality emergent?

109
 M.vanRaamsdonk conjectured 0907.2939, 1005.3035 that smooth connected patches
of geometry emerge from entanglement of regions on the boundary.
Entanglement may not be enough to fully probe bulk geometry, esp. if BH. 1406.5859
by BCCdB discussed entanglement shadows and entwinement. Key Q: how much information
can you ever reconstruct from the boundary?
A.Almheiri-X.Dong-D.Harlow 1411.7041 argued that localization of bulk informa-
tion should be understood in terms of quantum error correction. E.Mintun-J.Polchinski-
V.Rosenhaus 1501.06577 connected this to boundary gauge invariance, suggesting it is
closely connected to spacetime emergence.

10.14 Geometrization of entanglement

S.Ryu-T.Takayanagi conjecture hep-th/0603001 relates the entanglement entropy SEnt
associated to a region R in the field theory to the area of the minimal surface in the bulk
whose boundary is R. The holographic RT formula is important because it connects a geo-
metrical bulk computation with an information theoretic field theory computation. Reviews:
T.Nishioka-S.Ryu-T.Takayanagi 0905.0932, Headrick 1312.6717.
An explanation of the RT formula was provided by A.Lewkowycz-J.Maldacena in
1304.4926, using a bulk version of the replica trick.
‘Hole-ography’ method computes entanglement for a hole in AdS spacetime 1310.4204
V.Balasubramanian-B.Chowdhury-B.Czech-J.deBoer-M.Heller. Uses differencing of RT for-
mula, residual entropy.
N.Lashkari-J.Simon in 1402.4829 argued that emergence of an effective notion of
spacetime locality originates in restricting to a subset of observables unable to resolve black
hole microstates from the maximally entangled state.
People e.g. 1312.7856 also found that the first law for SEnt – for small perturbations
about CFT vacuum states, for ball-shaped regions – translates in the bulk to satisfaction
of equations of motion linearized about AdS! Constraining the nonlinear story: 1405.3743.
Also, entanglement inequalities can be used to derive conditions on bulk Tµν : 1412.3514.

11 Black hole entropy and the information paradox

11.1 Cooking up D1 k D5
Recipe for making BPS black holes is considerably simpler than recipe for making nonex-
tremal ones. Today, make BPS, qualitative comments only regarding nonextremal. First
part of recipe is how to combine different ingredients. In other words, rules for intersecting
branes.
We know two clumps of parallel BPS p-branes can be in static equilibrium. Also,
BPS p-branes and q-branes for some choices of p, q can be in equilibrium with each other
under certain conditions. One way to find many rules is to start with the fundamental string
intersecting a Dp-brane at a point, F 1 ⊥ Dp, and use S- and T-duality.

F 1 ⊥ D3 −→ D1 ⊥ D3 −→ D0 k D4 −→ D1 k D5

110
 F1 D1

D0 D1

D3 D3
D4 D5

11.2 Cooking up W k D5 and W k D1

and also
F 1 ⊥ D3 −→ D1 ⊥ D3 −→ D0 k D4 −→ M W k M 5 −→ W k D5

F1 D1
W
D0
MW D5
D3 D3
D4
M5

and by T-duality, W k D1.

W W
D5 D1

Therefore, W k D1 k D5 can all be in neutral equilibrium in a mutually consistent

fashion.

11.3 Problems with too few ingredients

BPS black holes in dimensions d = 4 . . . 9 may be constructed from BPS building blocks.
Typically, however, they have zero horizon area and therefore non-macroscopic entropy.
Example: consider D1-brane
−1/2 +1/2
ds2 = H1 −dt2 + dx2 + H1 dr2 + r2 dΩ27

where
32π 2 gs Np `6s
H1 = 1 +
r6
Now compactify the x direction on a circle of radius R at infinity. At the horizon,
Vol(S 1 ) p 1
= Gxx = (H1 )− 2 ∼ r3 → 0
(2π)R
How about Bekenstein-Hawking entropy? Transform to Einstein frame:
 −1/2
1/2 −1/4
gµν = H1 Gµν = H1 Gµν
so that
−3/4 +1/4
ds2 = H1 −dt2 + dx2 + H1 dr2 + r2 dΩ27

111
11.4 Why we started with D = 5 BH
Hence (entropy same if evaluate in d = 10 or d = 9 !)
16π 3  2 1/4 7/2 

1
SBH = r H1
4[8π 6 gs2 `8s ] 15 
horizon

and since H1 ∼ r−6 near horizon,

SBH (BPS D1) = 0.

More generally, study SUGRA field equations to find what BHs can have macroscopic
entropy. Sizes of internal manifolds, plus dilaton, are scalar fields in lower-d. Horizon area
depends on these scalars, which are ratios of functions of charges like Hp ’s.
But in any given d, have only a few independent charges on a black hole – fewer gauge
fields than scalars. Too few independent charges to give all scalar fields well-behaved vevs
everywhere in spacetime.
E.g. for stringy black holes made by compactifying on tori, only asymptotically flat
BPS black holes with macroscopic finite-area occur with 3 charges in d = 5 and 4 charges in
d = 4. The d = 4 case where all 4 charges are equal is Reissner-Nordstrøm. Woohoo!

11.5 The harmonic function rule

A systematic Ansatz is available for construction of SUGRA solutions corresponding to
pairwise intersections of BPS branes. Known as “harmonic function rule”.
Ansatz: metric factorizes as product structure: simply “superpose” harmonic func-
tions. This ansatz works for both parallel and perpendicular intersections.
Important restriction: harmonic functions can depend only on overall transverse co-
ordinates. In this way, get only “smeared” intersecting brane solutions.
Representation convention: − means brane is extended in that dimension, · means
it is pointlike, and ∼ says although brane is not extended in that direction a priori, its
dependence on those coordinates has been smeared away.
E.g. for D5 with D1 smeared over its worldvolume:
0 1 2 3 4 5 6 7 8 9
D1 − − ∼ ∼ ∼ ∼ · · · ·
D5 − − − − − − · · · ·

11.6 Cooking the D1-D5 system

For D1-D5 system, let us define r2 ≡ x2⊥ = 9i=6 (xi )2 to be overall transverse coordinate.
P
Then string frame metric is, using harmonic function rule,
1 1 1 1
2
dS10 = H1 (r)− 2 H5 (r)− 2 (−dt2 + dx21 ) + H1 (r)+ 2 H5 (r)− 2 dx22···5
1 1
+H1 (r)+ 2 H5 (r)+ 2 (dr2 + r2 dΩ23 )
and dilaton is
1 1
eΦ = H1 (r)+ 2 H5 (r)− 2

112
while R-R gauge fields are as before,

C01 = gs −1 H1 (r)−1 C01...5 = gs −1 H5 (r)−1

Independent D1 and D5 harmonic functions both go like r−2 ,

gs N5 `2s gs N1 `6s /V4
H5 (r) = 1 + H1 (r) = 1 +
r2 r2
Wrap x2 · · · x5 on T 4 to make d = 6 black string with two charges. Internal T 4 is finite-size
at event horizon r = 0:
  14 4
p H1 N1 (`4s /V4 )
G22 · · · G55 = →
H5 N5

11.7 Adding the gravitational wave

Next step is to roll up direction of black string, to make black hole in d = 5.
Behaviour of radius of x1 direction near horizon?
p r
G11 = (H1 H5 )−1/4 ∼ →0
(N1 N5 )1/4
Oops! Still need another quantum number to stabilize this S 1 as well as our T4 .
We can use knowledge from solution-generating to puff up this horizon to a macroscopic
size by using ∞ boost in longitudinal direction x1 .
Ingredients for building this black hole are then previous branes with addition of a
gravitational wave W:
0 1 2 3 4 5 6 7 8 9
D1 − − ∼ ∼ ∼ ∼ · · · ·
D5 − − − − − − · · · ·
W − → ∼ ∼ ∼ ∼ · · · ·

→ denotes direction in which gravitational Wave moves (at speed of light).

11.8 The D1-D5-W metric in 5D

BPS metric for this system is obtained from simpler metric for plain D1-D5 system by
boosting and taking extremal limit. To get rid of five dimensions to make a d = 5 black hole,
compactify D5-brane on the T 4 of volume (2π)4 V , and then D1 and remaining extended
dimension of D5 on S 1 , volume 2πR. d = 5 Einstein frame metric becomes

ds25 = − (H1 (r)H5 (r) (1 + K(r)))−2/3 dt2

+ (H1 (r)H5 (r) (1 + K(r)))1/3 [dr2 + r2 dΩ23 ]

where harmonic functions are

r12 r52 2
rm
H1 (r) = 1 + H5 (r) = 1 + K(r) =
r2 r2 r2
113
 r12 `2s r52 2
rm 2 `8s
= (gs N1 ) = (gs N5 ) = (gs Nm )
`2s V `2s `2s R2 V
This SUGRA solution has limits to its validity. For e.g. curvature, find e.g. R(d =
5) → −2/(r12 r52 rm ) at small r; or Rµν Rµν (d = 10) → −24/(r12 r52 ).
2 1/3

So if stringy α0 corrections to geometry are to be small, need large radius parameters. Dila-
ton? E.g. d = 10 e2Φ → N1 /N5 .

11.9 Bekenstein-Hawking entropy

Suppose we keep volumes V, R fixed in string units. Therefore, need

gs N1  1 gs N5  1 gs2 Np  1

Can also control closed-string loop corrections if gs  1. These two conditions are com-
patible if we have large numbers of branes and large momentum number for gravitational
wave W. Also note that Np needs to be hierarchically larger than N1 , N5 .
Next properties of this spacetime to compute are thermodynamic quantities. BPS
black hole is extremal and it has TH = 0. For Bekenstein-Hawking entropy,
A 1 n o
SBH = = 2π 2 r3 [H1 (r)H5 (r) (1 + K(r))]3/6 (11.1)
4G5 4G5 r=0
2
2π
=  8
 (r1 r5 rm )1/2 (11.2)
2
4 (π/4)gs `s /(V R)
1
2πV R gs N1 `s 6 2 8 2

2 gs Nm `s
= gs N5 `s (11.3)
gs 2 ` s 8 V R2 V
p
= 2π N1 N5 Nm (11.4)

11.10 Properties of SBH

This entropy p
SBH = 2π N1 N5 Nm
is macroscopically large. Notice that it is also independent of R and of V . More generally,
SBH for BPS guys is independent of all moduli. This is to be contrasted with ADM mass
Nm N1 R N5 RV
M= + +
R gs `s 2 gs `s 6
which depends on R, V explicitly.
√ For entropy of black hole just constructed out of D1 D5 and W, we had SBH =
2π N1 N5 Nm . More generally, for a more general black hole solution of maximal supergravity
arising from compactifying Type II on T 5 , it is
r
∆
SBH = 2π
48

114
where quantity ∆ in surd is cubic invariant of the E6,6 duality group,
4
X
∆=2 λ3i
i=1

and λi are eigenvalues of central charge matrix Z.

11.11 Yes, extreme BH can have finite SBH

A few years ago a claim was made that all extremal black holes have zero entropy. Arguments
were in Euclidean spacetime signature, and made the point that adding in surface terms at
horizon was necessary to make sure Euler number of horizon was not fractional.
This result is not trustworthy in the context of string theory.
1. There is actually no good physical reason why zero-temperature black holes should
have zero entropy. Standard statements of the Third Law make unnecessary assumptions
about the equation of state of physical matter.
2. Faulty nature of classical reasoning in string theory context was pointed out in a
G.Horowitz review article from mid-1990s. In Euclidean geometry, for any periodicity in
Euclidean time β at r = ∞, presence of extremal horizon results in a redshift which forces
that periodicity to be substringy very close to horizon. Since light strings wound around
this tiny circle can condense, a Hagedorn transition can occur. Classical approximation is
not reliable there; in particular, arguments based on classical topology are not believable.
3. This entropy would be hugely smaller than entropy of very-nearly-extremal BH!
Where would all the entropy go?

11.12 Four charges in four dimensions

Extremal Reissner-Nordström black hole can be embedded in string theory using D-branes.
For the extremal Reissner-Nordström spacetime metric in isotropic coordinates we find
H ±2 (r)s appearing in metric:

ds2 = H −2 (−dt2 ) + H 2 (dr2 + r2 dΩ2 ) H = 1 + r0 /r

1
This is to be contrasted with the H 2 ’s to be found in a generic p-brane metric:

ds2 = H −1/2 (−dt2 + dx21...p ) + H +1/2 (dr2 + r2 dΩ2 )

From this we can guess (correctly) that, in order to embed extremal Reissner-Nordström
black hole in string theory, we will need 4 independent brane constituents.
Restrictions must be obeyed, however, in order for that black hole to be Reissner-
Nordström.
To make more general d = 4 black holes with four independent charges, we simply
lift these restrictions and allow charges to be anything - so long as they are large enough to
permit a supergravity description.

115
11.13 The D2-D6-W-NS5 duality frame
For making d = 4 black hole, one set of ingredients would be

0 1 2 3 4 5 6 7 8 9
D2 − − − ∼ ∼ ∼ ∼ · · ·
D6 − − − − − − − · · ·
NS5 − − − − − − ∼ · · ·
W − → ∼ ∼ ∼ ∼ ∼ · · ·

By U-duality, we could consider instead 4 mutually orthogonal D3-branes, or indeed many

other more complicated arrangements.
In ten dimensions we can construct BPS solution by using the harmonic function rule.
So far we have not exhibited metric for NS5-branes but that can be easily obtained using
D5 metric and using fact that Einstein metric is invariant under S-duality. We then have

2 −1 − 12 −1 −1
−dt2 + dx21 + K(dt + dx1 )2 + H5 H2 2 H6 2 (dx22 )
 
dS10 = H2 2 H6
+1 −1 +1 +1
+H2 2 H6 2 H5 (dx23···6 ) + H5 H2 2 H6 2 (dr2 + r2 dΩ22 ) (11.5)

and
+1 +1 − 14 (3)
eΦ = H5 2 H2 4 H6

11.14 SBH for the 4D 4-charge BH

Smearing and Newton’s constant formulæ give

gs N2 `s 5 gs N6 `s N5 `s 2 gs 2 Nm `s 8
r2 = r6 = r5 = rm =
2V 2 2Rb 2V Ra2 Rb

Kaluza-Klein reduction formulæ give first a d = 5 black string and then finally the d = 4
black hole. Final Einstein metric in d = 4 is
hp i−1
ds2 = −dt2 (1 + K(r))H2 (r)H6 (r)H5 (r)
hp i
2 2 2
+(dr + r dΩ2 ) (1 + K(r))H2 (r)H6 (r)H5 (r)

Reissner-Nordström black hole is obtained by setting all four gravitational radii to be iden-
tical: r2 = r6 = r5 = rm . Bekenstein-Hawking entropy is
p
SBH = 2π N2 N6 N5 Nm

More generally, in surd is quantity ♦/256, where ♦ is quartic invariant of E7,7

4
X 4
X
2
|λi |2 |λj |2 + 4 λ1 λ2 λ3 λ4 + λ1 λ2 λ3 λ4


♦= |λi | − 2
i=1 i<j

where λi are (complex) eigenvalues of Z.

116
11.15 The D-brane picture
Our setup of branes for d = 5 BPS BH with 3 charges was

0 1 2 3 4 5 6 7 8 9
D1 − − ∼ ∼ ∼ ∼ · · · ·
D5 − − − − − − · · · ·
W − → ∼ ∼ ∼ ∼ · · · ·

This system preserves 4 real supercharges, or N = 1 in d = 5. Each constituent breaks

half of SUSYs.
Necessary for SUSY to orient branes in a relatively supersymmetric way. If not, e.g.
if an orientation is reversed, D-brane system corresponds to a black hole that is extremal
(double horizon) but has no SUSY.
Beginning ingredients: D1 branes and D5 branes. What are degrees of freedom carry-
ing momentum quantum number?
D5 branes and smeared D1 branes have a symmetry group

SO(1, 1)×SO(4)k ×SO(4)⊥ .

This symmetry forbids (rigid) branes from carrying linear or angular momentum, so we need
something else.

11.16 Open string dynamics

T4
S1

Obvious modes in the system to try are massless 1-1, 5-5 and 1-5 strings, which come
in both bosonic and fermionic varieties.
• Momentum Nm /R carried by bosonic and fermionic strings, 1/R each.
• Angular momentum is carried only by fermionic strings, 21 ~ each.
Both linear and angular momenta can be built up to macroscopic levels.
Next step: identify degeneracy of states of this system. Simplification made by
[Strominger-Vafa] is to choose the four-volume small by comparison to circle radius,
1
V 4 R

Makes theory on D-branes a d = 1 + 1 theory. This theory has (4, 4) SUSY in d = 1 + 1

language.

117
11.17 String partition function
d = 1 + 1 partition function of a number n of boson fields and an equal number of fermion
fields is " ∞ #n
Y 1 + w Nm X
Z= Nm
≡ Ω(Nm )wNm
N =1
1 − w
m

where Ω(Nm ) is degeneracy of states at d = 1 + 1 energy E = Nm /R.

At large-degeneracies, which happen with big quantum numbers like we have here, we
can use Cardy formula
r  r 
π c E (2πR) c
Ω(Nm ) ∼ exp = exp 2π ER
3 6
(Technical note: This formula assumes that lowest eigenvalue of energy operator is zero, as
it is in our system. Otherwise must use instead ceff = c − 24∆0 , where ∆0 is ground state
energy.)
We know R, radius of circular dimension. Need c and E.
Central charge
c = nbose + 21 nfermi
How many bosons (and fermions) do we have?

11.18 Degeneracy of states

Boson and fermion count in system of D1, D5 and open strings?
Can be done rigorously; here is the basic physics:
• N1 N5 1-5 strings that can move in 4 directions of torus, hence c = 6N1 N5 .
• Alternatively, we can use neat fact that D1-branes are instantons in D5-brane theory.
Have N1 instantons in U (N5 ) gauge theory, and N5 orientations to point them in. Etc...
Now, how about energy E? System is supersymmetric, and since no Z’s down here in
this d = 1 + 1 story, need P µ Pµ = 0. So E = |P |. In d = 1 + 1 things can move only to R
or L. Our sign conventions make us have R-moving groundstate, and put all the action in
L-movers. Momentum was P = ±Nm /R, so E = Nm /R.
Cardy said r  r 
π c E (2πR) c
Ω(Nm ) ∼ exp = exp 2π ER
3 6
Therefore p
Smicro = 2π N1 N5 Nm
OMG: this agrees exactly with the black hole result!

11.19 Adding rotation [BMPV]

In d = 5 there are two independent angular momentum parameters, because rotation group
transverse to D1’s and D5’s splits up as
SO(4)⊥ ' SU (2) ⊗ SU (2)

118
Angular momentum is consistent with d = 5 superalgebra.
Metrics for general rotating black holes are algebraically rather messy, we will not
write them here. We will simply quote result for BPS entropy:
p
SBH = 2π N1 N5 Nm − J 2

BPS black holes have a nonextremal generalisation, in which the two angular momenta
are independent. However, in extremal limit something interesting happens: two angular
momenta are forced to be equal and opposite, Jφ = −Jψ ≡ J. There is also a bound on
angular momentum, p
|Jmax | = N1 N5 Nm
Beyond Jmax , closed timelike curves develop, and entropy walks off into complex plane.

11.20 Rotating entropy agreement

Another notable feature of this BPS black hole: those funny Chern-Simons terms in the
R-R sector of the SUGRA Lagrangian are turned on. So this black hole is not a solution
of d = 5 Einstein-Maxwell theory! Note that gauge charges are unmodified by the funny
Chern-Simons terms because they fall off too quickly to contribute to surface integrals.
Reduced entropy can be understood rigorously in D-brane field theory.
But basic physics is simple: aligning 21 ~’s all in a row to build up macroscopic angular
momentum costs oscillator degeneracy. Energy is reduced as

J2
 
Nm 1
−→ Nm −
R R N1 N5

So entropy reduced to p
Smicro = 2π N1 N5 Nm − J 2
Agrees with black hole calculation again.
Also, find Jφ = −Jψ from SUSY.

11.21 d = 4 entropy counting

A canonical set of ingredients for building d = 4 system is what we had previously in building
black hole:
0 1 2 3 4 5 6 7 8 9
D2 − − − ∼ ∼ ∼ ∼ · · ·
D6 − − − − − − − · · ·
NS5 − − − − − − ∼ · · ·
W − → ∼ ∼ ∼ ∼ ∼ · · ·
First three ingredients are simply T-dual to our (D1, D5, W) system.
New feature: NS5-branes. New physics: D2-branes can end on NS5-branes. It costs
zero energy to break up a D2-brane like so:

119
 D2

NS5

11.22 d = 4 microscopic entropy and non-extremality

These extra massless degrees of freedom in system lead to an extra label on 2-6 strings, giving
rise to an extra factor of NNS5 in degeneracy. Entropy counting proceeds just as before, and
yields p
Smicro = 2π N2 N6 NNS5 Nm
which again agrees exactly with Bekenstein-Hawking black hole entropy. A major difference
between this and d = 5 case is that the single rotation rotation parameter is incompatible
with supersymmetry.
How about nonextremality? No SUSY nonrenormalization theorem here.
New ingredient: add extra energy (but no other charges) to system of D-branes (and
NS-branes) and open strings carrying linear and angular momenta.
SUGRA: nonextremal branes cannot be in static equilibrium with each other – they
want to fall towards each other, and they do not satisfy simple harmonic function superpo-
sition rule.
Least confusing way to construct nonextremal multi-charge solutions is to start with
appropriate higher-d neutral Schwarzschild or Kerr type solution, and to use multiple boost-
ings and duality transformations to generate required charges.

11.23 Nonextremal entropy and greybody factors

For nearly BPS systems, D-brane pictures for (D1, D5, W ) and
(D2, D6, N S5, W ) stay in d = 1 + 1.
Physics: new energy adds a small number of R-movers as well as L-movers. (Breaks
BPS condition.)
Think of R-movers and L-movers as dilute gases, interacting only very infrequently.
Energy and momentum are additive, and so is entropy.
Amazingly, entropy agrees with near-extremal black brane entropy. Why? - no theo-
rem protecting degeneracy of non-BPS states. What is going on physically is that conformal
symmetry possessed by the d = 1 + 1 theory is sufficiently restrictive, even when it is broken
by finite temperature, for black hole entropy to be reproduced by field theory.
Multi-parameter agreement. ↑↓

Also greybody factors can be com-

puted. Mindbogglingly, D-brane
story gives same answer!

120
11.24 String theory, D-branes, and SBH
S.Hawking 1974: quanta emitted by BH do not carry info about anything behind the
horizon, other than what can be measured at infinity: M, Ja , Qi .
S.Mathur proved a 2009 theorem 0909.1038 (more on that soon) that subleading
quantum gravity corrections cannot resolve the BH information paradox. Only order one
corrections to semiclassical BH expectations around the horizon can rescue unitarity. So we
need lots of hair. But is there any?
No-hair folk theorems for higher-D built on D ≤ 4 intuition turned out to be quite
wrong. In D ≥ 5, there is a much wider variety of solutions available as ingredients for
building BH. See e.g. I.Bena-N.Warner review 1311.4538.
D-branes arise as loci where open strings end; this is enough to determine their kine-
matics and dynamics. Nonperturbative: tension τp ∝ 1/gs .
Key fact about a stack of N D-branes: for large-N , distance scales you might think
are natively `s or `P can get parametrically enhanced to be as large as a BH horizon. Why?
Open (closed) string corrections scale as gs N (gs2 N ).
A.Strominger-C.Vafa rocked the world in 1996 by computing SBH for special D=5
BPS black holes from string statistical mechanics. This was the first computation of the
Bekenstein-Hawking entropy from first principles.
Similar methods correctly account for entropy even for rotating and near-BPS BHs in
5D, 4D. But a microscopic model of 4D Kerr BH remains elusive.

11.25 Emission rates and the fuzzball programme

Morally, we need to know the wavefunction behind the horizon as well as in front of it to
be able to solve the BHIP as well as compute entropy.
String theorists got further than computing SBH . Microscopic calculations of open/closed
string scattering yielded gorgeous agreement w Hawking emission from classes of near-BPS
BHs, including multi-parameter greybody factors.
From ST POV, ‘4D’ BHs are hiding higher-D physics near the singularity.
Motivated partly by new solutions, and by string CFT emission rate successes, S.Mathur
conjectured in 2001 that conventional BH geometry emerges as a coarse-graining over mi-
crostates: non-singular, horizonless, non spheroidally symmetric geometries with same asymp-
totics as BH but differ inside region of order horizon size. Exponentially large density of
states. Top-down POV.
For limited classes of less-complicated fuzzballs, it is possible to check Mathur’s conjec-
ture with some rigour. Nice fuzzball FAQ by Mathur: physics.ohio-state.edu/∼mathur/faq2.pdf.
Mathur’s 2009 theorem on BHIP used only two assumptions: (1) Hawking pairs cre-
ated fresh from vacuum independently of other pairs; (2) quantum gravity obeys strong
subadditivity, like any other reasonable quantum theory.
S.Mathur also clarified in 0909.1038, 1108.0302 that just having AdS/CFT duality
does not resolve the BHIP in principle.

121
dS is two entangled but
ss of Hawking radiation
ates suggests a different
nnected spacetimes. We
sthethe
eternal
dualhole
oftunnels
two entangled CFTs?
semiclassical11.26 D1-D5 CFT
spacetime.
uzzball complementarity, ∗
SamirPrototype
D. Mathurmicroscopic model: N1 D1-branes wrapped on S 1 + N5 D5-branes wrapped on
cs, The Ohio SState
1
× MUniversity,
4 . This systemColumbus, OH space.
has a moduli 43210,At USAone point it is best described in terms of BH
geometry; at another, by a D = 1 + 1 SCFT.p
hat the dual of theIneternal black hole
the low-energy limitinwith
AdSR(S
is 1two
)  entangled
4
Vol(M4 ),but
the SCFT is a symmetric product
wand gives
that the required created
the entanglement modification
by the process of Hawking radiation
orbifold (M4 ) /SN . Related physics: strings wrapped around S 1 fractionate: lowest mode
N
this conjecture. The nature
has energy 1/(N1 Nof fuzzball states suggests a different
5 R) rather than naive 1/R.
wo entangled CFTs is two entangled
Easy to calculate in but disconnected
microscopic SCFT spacetimes.
at orbifold We
point where it is free. And for BPS
tunneling’ where the SUSY
states, Einstein-Rosen bridge of the
non-renormalization eternal
theorem hole entropy
ensures tunnels agrees.
reventing the existence
But oftothe eternal
connect hole aswith
honestly a semiclassical
macroscopic spacetime.
BH physics and solve information paradox,
F
on then emerge only in the approximation of fuzzball complementarity,
need to deform SCFT away from orbifold point towards black hole. Top-down framing. This
ct of probes with energy LE ≫ T .R
CFT
2 is one focus of our research.
Recent projects: computing anomalous dimensions of low-lying string states in confor-
P
mal perturbation theory [Burrington-Peet-Zadeh] and analyzing aspects of squeezed states
N no-hair theorems [8] and gives the required modification
generated by twist deformations [Burrington-Mathur-Peet-Zadeh]. + [Burrington-Jardine-
of the(b)hole.
Peet-Zadeh in progress]
poration led to a How exactly will we see emergence of effective BH geometry? e.g.:-
d by the gravita-
mber of the pair, F
the other mem- iCFT CFT vs
CFT Σ 1
i
2 L R
mass.
2 These two i
e entanglement of P
eeps rising. This
(a) (b)
vaporation: how (d)
ge degeneracy11.27
re- Firewalls
of [10] says that two entangled
Hawking CFTs horizon are max entangled: their SEnt is ln 2.
pairs straddling
tspacetime
that the evapo-
(b). ThePage’s
nature of fuzzballs
theorem on quantum subsystem entropy: SEnt between BH and Hrad grows as
ngled
t fromCFTs (c)
BHgive
the burn- two entangled
radiates, but
CFT must gobut
back to CFT
zero again by time Σ
BH evaporates
i i away. So new Hrad just
1 2 i
sshould
(d). not have
outside BH should be max entangled with old Hrad.
ement. In [2] it But monogamy of entanglement rules out max entanglement with two others. Old
hat small BH complementarity of L.Susskind et al finessed this by arguing that BH blueshift prevents
correc-
n this debate has beenfrom
experimenters played byviolation of no-xerox theorem.
(c)
seeing (d)
n cannot resolve
eternal hole in AdS space. We are
der unity.1 Then AMPS 1207.3123 pointed out new flaws in old BH complementarity, ignited firewall
FIG. 1. The conjecture of [10] says that two entangled CFTs
on of AdS/CFT debateduality in the ofcon-
about validity GR as an effective field theory around BHs. Consider 4 postulates:
(a) gives
hole. The eternal hole in AdS has
(1) unitary S-matrix. the connected
(2) EFT worksspacetime (b).horizon.
outside BH The nature of fuzzballs
(3) BH appears to distant observer
dS boundaries,as quantumsuggests
so the system that two entangled CFTs (c) give two entangled
with discrete energy levels. (4) Nothing bad happens
usual notion but at the horizon.
l[9]
horizon, where
The main disconnected
result of AMPS:spacetimes (d).
suggests that the eternal holeone of (1,2,4) has to be false. They believe in (2) so yelled “Fire!”.
the local vacuum
Technical argument was about excitation of field modes, for infaller vs Hrad.
wo CFTs. The two AdS boundaries
uced at the hori- T.Banks had previously
A significant rolewarned
in thisthat energyhas
debate maybeen
not be the only
played by variable deciding
o we have two disconnected CFTs.
e new (nonlocal)
effectiveness of GR
considerationas an EFT. Must also look at entropy.
of the eternal hole in AdS space. We are
possibilities (P1), (P2)
ntanglement. In S.Hawking
above, we
hated firewalls so much he wrote a paper basically saying that he would
o possibilities: interested in the notion of AdS/CFT duality in the con-
on the recent rather
pro- giving up on event horizons entirely! [CBC article]
text of this eternal hole. The eternal hole in AdS has
here it is conjec- Recent substantial review article on BHIP ⊃ FW by D.Harlow: 1409.1231.
s conjectured
cted by a ‘worm-that whenasymptotically
two these dis- AdS boundaries, so the usual notion
of AdS/CFT duality [9] suggests that the eternal hole
placed in a particular entangled
are. spacetime is dual to two CFTs.122 The two AdS boundaries
istraditional
the eternal hole. Thus when dis-
hori- are not connected, so we have two disconnected CFTs.
placed
ugh thein a state that
Hawking is entangled,to the possibilities (P1), (P2) above, we
Corresponding
an are
we be connected
not forced(fig.1(a,
have b)).
the following two possibilities:
11.28 Avoiding firewalls
Lots of papers have been written about how firewalls might be avoided. (They all differ
drastically from the LQG community focus on remnants.)
D.Harlow-P.Hayden 1301.4504: quantum information theory constraints on getting
info out of a BH prevent firewalls. It takes the Page time (when SBH drops to 12 its initial
value) to be able to do experiments detecting a firewall. Aspects of this were explained more
intuitively by L.Susskind, 1301.4505.
S.Giddings 1211.7070: a small ‘nonviolent’ nonlocality hidden to large scale observers
may save you from firewalls. Challenge: it is generally very difficult to introduce only a ‘small’
amount of nonlocality theoretically.
S.Mathur-D.Turton in 1306.5488 clarified a number of issues surrounding black hole
complementarity, and explained the advantages the fuzzball approach provides in evading
firewalls. The essential technical point is that a fuzzball has collective modes, and infalling
quanta with E  kB T interact with these differently than Hawking radiation does.
K.Papadodimas-S.Raju conjectured in 1310.6335 that the mapping of CFT operators
to local bulk operators in AdS/CFT depends on the state of the CFT. Mirror operators
needed for 1-sided BH, to describe behind-horizon physics in a holographic setup and avoid
firewalls. So far only describes small fluctuations about a given reference state. Status:
murky at best.

11.29 ‘ER=EPR’
J.Maldacena-L.Susskind 1306.0533 proposed an intriguing new take on wormholes to
address firewalls that has become known as ’ER=EPR’. It is built on Maldacena’s proposal
hep-th/0106112 that the AdS eternal BH can be constructed via P CF TL × CF TR with
1
thermal entanglement between L and R, built on Israel’s |TFDi = Z i e−βE/2 |ψiL × |ψiR .
√

They propose entanglements are encoded by having ER bridges, but note that these
wormholes are far from classical. For good explanations of the proposal, see series of papers
by Susskind, e.g. 1311.3335, 1411.0690.
L.Susskind advocated in 1311.7379, 1402.5674 for connection with computational
complexity: length of ER bridge ∝ 1/entanglement.
‘Precursor’ in boundary CFT: nonlocal object set up in boundary theory to create
desired thing in the bulk in the causal future. These have played an important role in
questions about avoiding firewalls. Precursors that cause firewalls are ‘hard’, and have
exponentially large computational complexity.
V.Balasubramanian-M.Berkooz-S.Ross-J.Simon provided some interesting caveats in
1404.6198, arguing that spectral information is also needed to diagnose spacetime connect-
edness in the AdS/CFT context.
Perhaps, as Mathur has suggested, the non-classical Einstein-Rosen bridges of ER=EPR
rapidly tunnel into fuzzball states?

123