Master ICFP 2019-2020

Lectures Notes on String Theory

Dan Israël (israel@lpthe.jussieu.fr )

Lucio Fontana, « Concetto Spaziale »
Foreword
These lecture notes are based upon a series of courses given at the master program ICFP
from 2018 by the author. Comments and suggestions are welcome. Some references that can
complement these notes are

• Superstring Theory (Green, Schwarz, Witten) [1,2]: the classic textbook from the eight-
ies, naturally outdated on certain aspects but still an unvaluable reference on many
topics including the Green-Schwarz string and compactifications on special holonomy
manifolds.

• String Theory (Polchinski) [3,4]: the standard textbook, with a very detailed derivation
of the Polyakov path integral and strong emphasis on conformal field theory methods.

• String Theory in a Nutshell (Kiritsis) [5]: a concise presentation of string and super-
string theory which moves quickly to rather advanced topics

• String Theory and M-Theory: A Modern Introduction (Becker, Becker, Schwarz) [6]:
a good complement to the previous references, with a broad introduction to modern
topics as AdS/CFT and flux compactifications.

• A first course in String theory (Zwiebach) [7]: an interesting and different approach,
making little use of conformal field theory methods, in favor of a less formal approach.

• Basic Concepts of String Theory (Blumenhagen, Lüst, Theisen) [8]. As its name does
not suggest, this book covers a lot of rather advanced topics about the worldsheet
aspects of string theory. It is also rather appropriate for a math-oriented reader.

• The lectures notes of David Tong (http://www.damtp.cam.ac.uk/user/tong/string.html)

are rather enjoyable to read, with a good balance between mathematical rigor and phys-
ical intuition.

• The very lively online lectures of Shiraz Minwalla: http://theory.tifr.res.in/ minwalla/

Conventions
• The space-time metric is chosen to be of signature (−, +, . . . , +).

• We work in units ~ = c = 1

Latest update
March 29, 2022
References
[1] M. B. Green, J. H. Schwarz, and E. Witten, Superstring theory, Vol. 1: Introduction.
Cambridge Monographs on Mathematical Physics. 1988.

[2] M. B. Green, J. H. Schwarz, and E. Witten, Superstring theory, Vol. 2: loop amplitudes,
anomalies and phenomenology. Cambridge, Uk: Univ. Pr. ( 1987) 596 P. ( Cambridge
Monographs On Mathematical Physics), 1988.

[3] J. Polchinski, String theory. Vol. 1: An introduction to the bosonic string. Cambridge
University Press, 2007.

[4] J. Polchinski, String theory. Vol. 2: Superstring theory and beyond. Cambridge University
Press, 2007.

[5] E. Kiritsis, String theory in a nutshell. Princeton, USA: Univ. Pr. (2007) 588 p, 2007.

[6] K. Becker, M. Becker, and J. H. Schwarz, String theory and M-theory: A modern intro-
duction. Cambridge University Press, 2006.

[7] B. Zwiebach, A first course in string theory. Cambridge University Press, 2006.

[8] R. Blumenhagen, D. Lüst, and S. Theisen, Basic concepts of string theory. Theoretical
and Mathematical Physics. Springer, Heidelberg, Germany, 2013.

2
Contents

1 Introduction 5
1.1 Gravity and quantum field theory . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 String theory: historical perspective . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Bosonic strings: action and path integral 15

2.1 Relativistic particle in the worldline formalism . . . . . . . . . . . . . . . . . . 16
2.2 Relativistic strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4 Polyakov path integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.5 Open strings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3 Conformal field theory 52

3.1 The conformal group in diverse dimensions . . . . . . . . . . . . . . . . . . . . 54
3.2 Radial quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.3 Conformal invariance and Ward identities . . . . . . . . . . . . . . . . . . . . 58
3.4 Primary operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.5 The Virasoro Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.6 The Weyl anomaly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.7 Conformal field theory with boundaries . . . . . . . . . . . . . . . . . . . . . . 74

4 Free conformal field theories 77

4.1 Free scalar fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
4.2 Free fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.3 The ghost CFT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5 The string spectrum 107

5.1 String theory and the dimension of space-time . . . . . . . . . . . . . . . . . . 108
5.2 BRST quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5.3 The closed string spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
5.4 Open string spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.5 Physical degrees of freedom and light-cone gauge . . . . . . . . . . . . . . . . 133
5.6 General structure of the string spectrum . . . . . . . . . . . . . . . . . . . . . 136

3
6 String interactions 140
6.1 The string S-matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.2 Four-tachyon tree-level scattering . . . . . . . . . . . . . . . . . . . . . . . . . 142
6.3 One-loop partition function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

7 Introduction to superstring theories 160

7.1 Two-dimensional local supersymmetry . . . . . . . . . . . . . . . . . . . . . . 161
7.2 Superconformal symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
7.3 BRST quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
7.4 Type II superstring theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
7.5 One-loop vacuum amplitudes . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

4
Chapter 1

Introduction

5
Introduction.

In Novembrer 1994, Joe Polchinski published on the ArXiv repository a preliminary ver-
sion of his celebrated textbook on String theory, based on lectures given at Les Houches,
under the title ”What is string theory?” [1]. If he were asked the same question today, the
answer would probably be rather different as the field has evolved since in various directions,
some of them completely unexpected at the time.

One may try to figure out what string theory is about by looking at the program of Strings
2017, the last of a series of annual international conferences about string theory that have
taken place at least since 1989, all over the world. Among the talks less than half were about
string theory proper (i.e. the theory you will read about in these notes) while the others
pertained to a wide range of topics, such as field theory amplitudes, dualities in field theory,
theoretical condensed matter or general relativity.

The actual answer to the question raised by Joe Polchinski, ”What is string theory?”, may
be answered at different levels:
• litteral: the quantum theory of one-dimensional relativistic objects that interact by
joining and splitting.
• historical: before 1974, a candidate theory of strong interactions; after that date, a
quantum theory of gravity.
• practical: a non-perturbative quantum unified theory of fundamental interactions whose
degrees of freedom, in certain perturbative regime, are given by relativistic strings.
• sociological: a subset of theoretical physics topics studied by people that define them-
selves as doing research in string theory.
In these notes, we will provide the construction of a consistent first quantized theory of
interacting quantum closed strings. We will show that such theory automatically includes a
(perturbative) theory of quantum gravity. We will introduce also open strings that incorpo-
rate gauge interactions, and give rise to the concept of D-branes that plays a prominent role
in the AdS/CFT correspondence.
Along the way we will introduce some concepts and techniques that are as useful in other
areas of theoretical physics as they are in string theory, for instance conformal field theories,
BRST quantization of gauge theories or supersymmetry.

1.1 Gravity and quantum field theory

String theory has been investigated by a significant part of the high-energy theory community
for more than forty years as it provides a compelling answer – and maybe the answer – to
the following outstanding question: what is the quantum theory of gravity?

A successful theory of quantum gravity from the theoretical physics viewpoint should at
least satisfy the following properties:

6
Introduction.

1. the theory should reproduce general relativity, in an appropriate classical, low-energy

regime;
2. the theory should be renormalizable or better UV-finite in order to have predictive
power;
3. it should satisfy the basic requirements for any quantum theory, such as unitarity;
4. it should explain the origin of black hole entropy, possibly the only current prediction
for quantum gravity;
5. last but not least, the physical predictions should be compatible with experiments, in
particular with the Standard Model of particle physics, astrophysical and cosmological
observations.
According to our current understanding, string theory passes successfully the first four
tests. Whether string theory reproduces accurately at energies accessible to experiments the
known physics of fundamental particles and interaction is still unclear, given that such physics
occurs in a regime of the theory that is beyond our current analytical control (to draw an
analogy, one cannot reproduce analytically the physics of condensed matter systems from the
microscopic quantum mechanical description in terms of atoms). At least it is clear that the
main ingredients are there: chiral fermions, non-Abelian gauge interactions and Higgs-like
bosons.

The problem with quantizing general relativity

The classical theory of relativistic gravity in four space-time dimensions, or Einstein theory,
follows in the absence of matter from the Einstein-Hilbert (EH) action in four space-time
dimensions, that takes the form
Z
1 p
Seh = 2 d4 x − det g (R(g) − 2Λ) , (1.1)
2κ M
where R is the Ricci scalar associated with the space-time manifold M, endowed with a met-
ric g, and Λ the cosmological constant that has no a priori reason to vanish. √ The coupling
constant κ of the theory is related to the Newton’s constant through κ = 8πG; by dimen-
sional analysis it has dimension of length. Its inverse defines the Planck mass MPl ∼ 1019 GeV.

Quantizing general relativity raises a number of deep conceptual issues, that can be raised
even before attempting to make any explicit computation. Some of them are:
• Because of diffeormorphism invariance, there are no local observables in general relativ-
ity.
• A path-integral formulation of quantum gravity should include, by definition, a sum
over space-time geometries. Which geometries should be considered? Should we specify
boundary conditions?

7
Introduction.

• A Hamiltonian formulation of quantum gravity would require a foliation of space-time

in terms of space-like hypersurfaces. Generically, such foliation does not exist.

• Classical dynamics of general relativity predicts the formation of event horizons, shield-
ing regions of space-time from the exterior. This challenges the unitarity of the theory,
through the black hole information paradox.

Quantum gravity with a positive cosmological constant – which seems to be relevant to de-
scribe the Universe – raises a number of additional conceptual issues that will be ignored
in the rest of the lectures. We will mainly focus on theories with a vanishing cosmologi-
cal constant; the case of negative cosmological constant will be discussed in the AdS/CFT
lectures.

Perturbative QFT for gravity

One may try to ignore these conceptual problems and build a quantum field theory of gravity
in the usual way, i.e. by defining propagators, vertices, Feynman rules, etc..., from the
non-linear EH action, eqn. (1.1) [2].
With vanishing cosmological constant, one considers fluctuations of the metric around a
reference Minkowski space-time metric:

gµν = ηµν + hµν . (1.2)

Linearizing the equations of motion that follows from (1.1), in the absence of sources, we
arrive to:
h̄µν − 2∂ρ ∂(µ h̄ν)ρ + ηµν ∂ρ ∂σ h̄µν = 0 , (1.3)
where we have defined the ”trace-reversed” tensor h̄µν = hµν − 21 hρρ ηµν . This theory possesses
a gauge invariance that comes from the diffeomorphism invariance of the full theory. The
equations of motion are invariant under

hµν 7→ hµν + ∂µ ζν + ∂ν ζµ . (1.4)

One can choose to work in a Lorentz gauge, defined by ∂µ h̄µν = 0, in which case the field
equations (1.3) amounts to a wave equation for each component, h̄µν = 0.

The solutions of these equations are naturally plane waves h̄µν (xρ ) = h0µν exp(ikρ xρ ), and
the Lorentz gauge condition means that they are transverse. Finally, the residual gauge
invariance that remains in the Lorentz gauge, corresponding to vector fields ζµ satisfying the
wave equation ζµ = 0, can be fixed by choosing the longitudinal gauge h̄0µ = 0. As a result,
the gravitational waves have two independent transverse polarizations. The corresponding
quantum theory is a theory of free gravitons that are massless bosons of helicity two.
The interactions between gravitons are added by expanding the EH action around the
background (1.2) in powers of hµν . In pure gravity one obtains three-graviton and four-
graviton vertices, that have a rather complicated form. For instance the four-graviton vertex

8
Introduction.

looks roughly like:

 
G µ1 ν1 ,...,µ4 ν4 2
(k1 , . . . , k4 ) = κ k1 · k2 η µ1 ν1
···η µ4 ν4
+ kµ1 3 kν2 3 ηµ1 µ2 ην1 ν2 + ··· (1.5)

Using these vertices one can define Feynman rules for the quantum field theory of gravitons
and compute loop diagrams like the one below.
µ1ν1 µ2ν2
kρ1 kρ2

kρ3 kρ4
µ3ν3 µ4ν4

Figure 1.1: Graviton scattering

As in most quantum field theories, such loops integrals diverge when the internal momenta
propagating in the loop become large, and should be regularized. By dimensional analysis,
the regularized loop diagrams will be weighted by positive powers of (Λuv /MPl ), where Λuv
is the ultraviolet cutoff.
In renormalizable QFTs as quantum chromodynamics, such high-energy – or ultraviolet –
divergences can be absorbed into redefinitions of the couplings and fields of the theory, which
leads to theories with predictive power. In contrast, this cannot be done for general relativity,
for the simple reason that the coupling constant is dimensionfull (it has the dimension of
length). Therefore, the divergences cannot be absorbed by redefining fields and couplings in
the original two-derivative action; rather higher derivative terms should be included to do so.
General relativity is thus a prominent example of non-renormalizable quantum field theory.
Still it doesn’t mean that such a theory is meaningless in the Wilsonian sense; it can describe
the low-energy dynamics, well below the Planck scale MPl , of an ”ultraviolet” theory of
quantum gravity that is not explicitly known. However, as in any non-renormalizable theory,
this effective action has little predictive power, as higher loop divergences need to be absorbed
in extra couplings that were not present in the original action,1 but become less important
as the energy becomes lower. As we shall see string theory solves the problem in a rather
remarkable way, by removing all the ultraviolet divergences of the theory.
1
Strictly speaking, the one-loop divergence of pure GR can be absorbed by field redefinition. This not the
case when matter is present, and from two-loops onwards for pure gravity.

9
Introduction.

1.2 String theory: historical perspective

This is a very sketchy account about the history of string theory; interested reader may
consult the book [3] for first-hand testimonies.
String theory as a theory of quantum gravity came almost by accident, after being pro-
posed as a theory of strong interactions. The prehistory of string theory occurred during the
sixties. At that time, general relativity was not, with few exceptions, a topic of interest for
theoretical physicists, but was rather a playground for mathematicians.
Quantum field theory itself didn’t have the central role that it has today in our under-
standing of fundamental interactions. While quantum electrodynamics was acknowledged as
the appropriate description of electromagnetic interactions, most physicists thought that it
was an inappropriate tool to solve the big problems of the time, the physics of strong and
weak interactions. This was especially true for the strong interactions, as the experiments
were finding a growing number of hadronic particles, with large masses and spins. These
particles were mostly resonances i.e. particles with a finite lifetime. Defining a QFT includ-
ing all of these resonances did not seem, rightfully, a sensible idea. Quarks did make their
appearance in the theoretical physicists’ lexicon, however they were thought as mathemati-
cal tools rather than as actual elementary particles – the fact that they cannot be observed
individually supported this point of view.
A different approach, called the S-matrix program, was widely popular back then. The idea
was to construct directly the S-matrix of the theory using some general physical principles
(unitarity, analyticity,...), as well as some experimental input from the specific theory that
was considered, without any reference to a ”microscopic” Lagrangian.
One crucial experimental observation was that hadronic resonances could be classified in
families along curves in the mass-angular momentum plane (M, J) called Regge trajectories:

J = α(0) + α 0 M2 (1.6)

The value of the parameter α(0), or intercept, was determining a given family of resonances,
while the slope α 0 was universal – with one exception – and given experimentally by

α 0 ' 1GeV−2 (1.7)

in natural units.

Figure 1.2: Channel duality: s-channel (left) and t-channel (right)

Among important requirements imposed upon the S-matrix, was that all the hadrons
along the Regge trajectories should appear on the same footing, and both as intermediate

10
Introduction.

particles (resonances) in the s-channel or as virtual exchanged particles in the t-channel, see
fig. 2.11; actually either point of view was expected to give a complete description of the
scattering process. This channel duality property of the S-matrix, together with the other
physical constraints, led Gabriele Veneziano to write down, in 1968, an essentially unique
solution to the problem for the decay ω → π+ + π0 + π− [4]:

Γ (−α(s))Γ (−α(t))    
T= + (s, t) → (s, u) + (s, t) → (t, u) , (1.8)
Γ (−α(s) − α(t))

where α(s) = α0 + α 0 s describes a Regge trajectory. This amplitude has remarkable prop-
erties; it exhibits an infinite number of poles in the s- and t-channels, and its ultraviolet
behavior is softer than of any quantum field theory.
This breakthrough was the starting point for lot of activity in the theoretical physics
community, and remarkably lot of progress was done without having any microscopic La-
grangian to underlie this physics. For instance the generalization to N-particle S-matrices
was obtained, the addition of SU(N) quantum numbers, the analysis of the unitarity of the
theory (by looking at the signs of the residues) and even loop amplitudes.
Soon however people discovered strange properties of what was known at the time as the
dual resonance model. In order to avoid negative norm states, the intercept of the Regge
trajectory had to be tuned in such a way that unexpected massless particles of spin 1,2,...
appeared in the theory. Embarrassingly, it was also needed that the dimension of space-time
was 26! Around the same time it was realized, finally, that the states of the theory were
describing the quantized fluctuations of relativistic strings by Nambu, Nielsen and Susskind
in 1970.
Another problem was the appearance of a tachyon, i.e. an imaginary mass particle, in
the spectrum. This was solved soon after, following the work of Neveu, Schwarz [5] and
Ramond [6], who introduced fermionic degrees of freedom on the string in 1971 (bringing the
space-time dimension to 10) by Gliozzi, Scherk2 and Olive, who obtained the first consistent
superstring theories in 1976 [7].
At the same time that these remarkable achievements were obtained, the non-Abelian
quantum field theory of the strong interactions, or quantum chromodynamics, was recognized
as the valid description of the hadronic world and, together with the electroweak theory, gave
to quantum field theory the central role in theoretical high-energy physics that it has today.
It could had been the end of string theory, however, by a remarkable change of perspec-
tive, Scherk and Schwarz proposed in 1974 that string theory, instead of a theory of strong
interactions, was providing a theory of quantum gravity [8]. From this point of view the
annoying massless spin two particule of the dual resonance model was corresponding to the
graviton, and they show that it has indeed the correct interactions.
The six extra dimensions of the superstring could be considered in this context as com-
pact dimensions, given that the geometry was now dynamical, resurrecting the old idea of
2
Joël Scherk (1946-1980) was a remarkable French theoretical physicist who made many key contributions
to string theory and supergravity in the seventies, and died tragically when he was 33 years old only, leaving
an indelible imprint in the field. The library at the LPTENS is dedicated to his memory.

11
Introduction.

Kaluza [9], and Klein [10] from the twenties. The value of the Regge slope should be radically
different from was it was considered before in the hadronic context, in order to account for
the observed magnitude of four-dimensional gravity. One was considering

α 0 ' 10−38 Gev−2 , (1.9)

or equivalently strings of a size smaller by 19 orders of magnitude than the hadronic string,
i.e. impossible to resolve directy by current or foreseeable experiments. Despite that string
theory was able to fullfill an old dream – quantizing general relativity – research in string
theory remained rather confidential before the next turning point of its history,
Between 1984 and 1986, several important discoveries occurred and changed the fate of
the theory: the invention of heterotic string [11] (which made easy to incorporate non-Abelian
gauge interactions in string theory), the Green-Schwarz anomaly cancellation mechanism [12]
which strengthened the link between string theory and low-energy supergravity, thereby mak-
ing the former more convincing, and finally the discovery of Calabi-Yau compactifications [13]
and orbifold compactifications [14] which allowed to get at low energies models of particle
physics with N = 1 supersymmetry in four dimensions. After that string theory became
more mainstream, as many theoretical physicists started to realize that it was a promising
way of unifying all fundamental particles and interactions in a consistent quantum theory.
Thirty years and a second revolution after, we haven’t yet achieved this goal fully but
tremendous progress has been made, the hallmarks being the discoveries of D-branes [15], of
strong/weak dualities [16–18], of holographic dualities [19] and of flux compactifications [20,
21] to name a few. We still have a long way to go, and it is certainly worth trying.

References
[1] J. Polchinski, “What is string theory?,” in NATO Advanced Study Institute: Les Houches
Summer School, Session 62: Fluctuating Geometries in Statistical Mechanics and Field
Theory Les Houches, France, August 2-September 9, 1994. 1994. hep-th/9411028.

[2] S. Weinberg, “Ultraviolet divergences in quantum theories of gravitation.,” in General

Relativity: An Einstein centenary survey, S. W. Hawking and W. Israel, eds., pp. 790–
831. 1979.

[3] M. Gasperini and J. Maharana, “String theory and fundamental interactions,” Lect.
Notes Phys. 737 (2008) pp.1–974.

[4] G. Veneziano, “Construction of a crossing - symmetric, Regge behaved amplitude for

linearly rising trajectories,” Nuovo Cim. A57 (1968) 190–197.

[5] A. Neveu and J. H. Schwarz, “Factorizable dual model of pions,” Nucl. Phys. B31 (1971)
86–112.

[6] P. Ramond, “Dual Theory for Free Fermions,” Phys. Rev. D3 (1971) 2415–2418.

12
Introduction.

[7] F. Gliozzi, J. Scherk, and D. I. Olive, “Supergravity and the Spinor Dual Model,” Phys.
Lett. 65B (1976) 282–286.

[8] J. Scherk and J. H. Schwarz, “Dual Models for Nonhadrons,” Nucl. Phys. B81 (1974)
118–144.

[9] T. Kaluza, “On the Problem of Unity in Physics,” Sitzungsber. Preuss. Akad. Wiss.
Berlin (Math. Phys.) 1921 (1921) 966–972.

[10] O. Klein, “Quantum Theory and Five-Dimensional Theory of Relativity. (In German
and English),” Z. Phys. 37 (1926) 895–906. [Surveys High Energ. Phys.5,241(1986)].

[11] D. J. Gross, J. A. Harvey, E. J. Martinec, and R. Rohm, “The Heterotic String,” Phys.
Rev. Lett. 54 (1985) 502–505.

[12] M. B. Green and J. H. Schwarz, “Anomaly Cancellation in Supersymmetric D=10 Gauge

Theory and Superstring Theory,” Phys. Lett. 149B (1984) 117–122.

[13] P. Candelas, G. T. Horowitz, A. Strominger, and E. Witten, “Vacuum Configurations

for Superstrings,” Nucl. Phys. B258 (1985) 46–74.

[14] L. J. Dixon, J. A. Harvey, C. Vafa, and E. Witten, “Strings on Orbifolds,” Nucl. Phys.
B261 (1985) 678–686.

[15] J. Polchinski, “Dirichlet Branes and Ramond-Ramond charges,” Phys. Rev. Lett. 75
(1995) 4724–4727, hep-th/9510017.

[16] A. Sen, “Strong - weak coupling duality in four-dimensional string theory,” Int. J. Mod.
Phys. A9 (1994) 3707–3750, hep-th/9402002.

[17] C. M. Hull and P. K. Townsend, “Unity of superstring dualities,” Nucl. Phys. B438
(1995) 109–137, hep-th/9410167.

[18] E. Witten, “String theory dynamics in various dimensions,” Nucl. Phys. B443 (1995)
85–126, hep-th/9503124.

[19] J. M. Maldacena, “The Large N limit of superconformal field theories and supergrav-
ity,” Int. J. Theor. Phys. 38 (1999) 1113–1133, hep-th/9711200. [Adv. Theor. Math.
Phys.2,231(1998)].

[20] K. Dasgupta, G. Rajesh, and S. Sethi, “M theory, orientifolds and G - flux,” JHEP 08
(1999) 023, hep-th/9908088.

[21] S. B. Giddings, S. Kachru, and J. Polchinski, “Hierarchies from fluxes in string compact-
ifications,” Phys. Rev. D66 (2002) 106006, hep-th/0105097.

13
Introduction.

14
Chapter 2

Bosonic strings: action and path integral

15
Bosonic strings: action and path integral

Bosonic string theory, which is the most basic form of string theory, describes the prop-
agation of one-dimensional relativistic extended objects, the fundamental strings, and their
interactions by joining and splitting.
Quantum field theories of point particles are obtained by starting with a classical action,
and quantizing the fluctuations around a given classical solution of the equations of motion.
Upon quantization one gets field operators acting on the Fock space of the theory by creating
or annihilating particles at a given point in space. An analogous string field theory exists,
but is still poorly understood. In such theory one should have operators creating a loop in
space, which is certainly more difficult to describe mathematically.
Rather the practical way to handle string theory is to follow the propagation in time of a
single string in a fixed reference space-time. As restrictive as it looks like, this first-quantized
formalism does not prevent for studying the interactions between strings, computing loop
amplitudes and make a large number of predictions. As we will see below this ”first-order”
formalism can be used for point particles as well, as an alternative to QFT Feynman diagrams
that allows to perform perturbative computations; however it misses important aspects as
solitons or instantons that can be handled semi-classically from a field theory, and is not
suited for all types of computations.

2.1 Relativistic particle in the worldline formalism

We consider a relativistic particle of mass m and charge q in a given d-dimensional space-
time M of metric G and background electromagnetic field. Its dynamics is governed by the
action Z Z
S = −m ds − q A , (2.1)
l l
where l is the worldline of the particle, s the proper time and A(xµ ) = A(xµ )ρ dxρ the gauge
potential. Under a gauge transformation, A 7→ A + dΛ, the worldline action (2.1) is invariant
up to possible boundary terms.
The worldline of the particle in space-time M corresponds to an embedding map1
R ,→ M (2.2)
τ 7→ xµ (τ) , (2.3)
where τ is an affine parameter and {xµ , µ = 0, . . . , D − 1} a set of coordinates on M. The
proper time differential can be expressed as
ds2 = −Gµν ẋµ ẋν dτ 2 , (2.4)
therefore the action (2.1) can be rewritten as
Z q Z
S = −m dτ −ẋ (τ)ẋ (τ)Gµν [x (τ)] − q Aµ [xµ (τ)]dxµ (τ) .
µ ν ρ (2.5)
l
1
Strictly speaking, this is in general valid in an open set of M where the coordinates {xµ , µ = 0, . . . , D − 1}
are well-defined.

16
Bosonic strings: action and path integral

This action is invariant under diffeomorphisms of the worldline, i.e. under any differentiable
reparametrization
τ 7→ τ̃(τ) (2.6)
The embedding is now given by definition by the set of differentiable functions {x̃µ (τ̃) =
xµ (τ), µ = 0, . . . , D − 1}.
Let us consider the variation of the particle action (2.5) under the infinitesimal change of
the path, namely

xµ →
7 xµ + δxµ (2.7a)
Gµν →7 Gµν + ∂σ Gµν δxσ . (2.7b)

At first order, one gets

Z  
G ẋµ δẋσ ẋµ ẋν ∂σ Gµν δxσ q
δS = m dτ p µσ + p − (Aσ δẋσ + ∂σ Aµ δxρ ẋµ ) (2.8)
µ ν
−ẋ ẋ Gµν µ ν
−ẋ ẋ Gµν m

After integration by parts of the first and third term, and trading the integral over the affine
parmeter τ for the integral over the proper time s, one gets
Z  2 ν ρ σ
q ν dxµ

dx ν dx dx
δS = m ds + Γρσ − Fµ δxν (2.9)
ds2 ds ds m ds
Not surprisingly, one obtains the relativistic equation of motion of a massless charged particle,
i.e. the geodesic equation plus the coupling to the electromagnetic field strength F = dA.
In order to make more explicit the diffeomorphism invariance of the worldline action, one
can introduce an independent worldline metric as ds2 = hττ (τ)dτ2 . In the one-dimensional
√
analogue of the tetrad formalism of general relativity, one defines the einbein e(τ) = −hττ .
The action (2.5) can be then rewritten in a classically equivalent way as:
Z Z
1 p
dτ −hττ h ∂τ x ∂τ x Gµν + m − q Aµ dxµ
ττ µ ν 2


Se = −
2
Z   Z l
1 1
= dτ Gµν ẋµ ẋν − m2 e − q Aµ dxµ (2.10)
2 e l

where one can see that e(τ) play the role of a Lagrange multiplier field e(τ) – i.e. a non-
dynamical field that enforces a constraint in field space. Its equation of motion is simply
0 = −e−2 Gµν ẋµ ẋν − m2 , which, upon replacing e by the solution in the action (2.10), gives
back the original action (2.5).
One can view this action as a one-dimensional theory of gravity coupled to a set of free
scalar fields xµ (t) (there is naturally no curvature term in one dimension). Notice that
the coupling of the particle to the electromagnetic four-potential A – the last term in equa-
tion (2.10) – is independent of the worldline metric. In this sense this coupling is of topological
nature. Under diffeorphisms τ 7→ τ̃(τ) the einbein transforms according to

e(τ)dτ = ẽ(τ̃)dτ̃ (2.11)

17
Bosonic strings: action and path integral

which gives, for an infinitesimal transformation

τ 7→ τ̃ = τ + (τ) (2.12)
d
e(τ) 7→ ẽ(τ̃) = e(τ̃) −


e(τ̃)(τ̃) . (2.13)
dτ̃
As in four-dimensional gravity, this reparametrization invariance is a gauge symmetry, i.e.
a redundancy in the description of the system that will eventually remove some degrees of
freedom from the theory.2

Path integral quantization

The quadratic action (2.10) for the relativistic charged particle is a convenient starting point
for quantizing the theory through the path integral formalism. It is convenient as well to
analytically continue the space-time to Euclidean signature x0 7→ ix0 , as well as the worldline
time τ 7→ iτ. We will consider a particle moving in flat space, i.e. with metric Gµν = δµν , in
the absence of electromagnetic field.
The one-particle vacuum energy in Euclidean space deduced from the wordline action
is given by summing over closed paths of the particle, which is given schematically by the
functional integral:
Z Z  Z
De 1 1
 
1 2
Z1 = Dx exp − dτ ẋ + m e2
, (2.14)
Vol(diff) x(0)=x(1) 2 0 e

where one has to divide the functional integral over the einbein (or equivalently over the one-
dimensional metrics) by the infinite volume of the group of diffeomorphisms of the worldline.
This group contains the transformations of the vielbein given infinitesimally by (2.13). On
top of this shifts of τ by a constant, τ 7→ τ + τ0 , are diffeomorphisms that are not fixed by
the choice of a reference einbein. The volume of this factor of the gauge group is finite and
given by T , the invariant length of the closed path of the particle, see eq. (2.16) below. We
choose finally the parameter τ to be in the interval [0, 1], and, the path being closed, the
einbein is a periodic function: e(τ + 1) = e(τ).

Gauge symmetry
To carry the functional integral over the ”gauge field” e(τ) one starts by slicing the field space
into gauge orbits, i.e. einbeins that are related to each other by a diffeomorphism. The ratio
of the integral over the whole field space over the volume of the group of diffeomorphisms
is then equivalent to a functional integral over a slice in field space that cuts once each
orbit, see figure 2.1 up to the Jacobian of the change of coordinates in field space; this is the
Faddeev-Popov method [1] (FP for short).3
2
This redundancy was already explicit in the original description of the theory, eq. (2.1), as one could
have chosen the gauge x0 (τ) = τ to start with.
3
This method is rather overkill for dealing with a free particle but will be used again in the case of the
string.

18
Bosonic strings: action and path integral

Figure 2.1: Foliation of the space of gauge fields into gauge orbits. A slice through field space
intersecting all orbits once is represented in bold.

The Faddev-Popov method is the standard way of dealing with path-integral quantization
of non-Abelian gauge theories and is presented in most quantum field theory textbooks, see
e.g. in [2], chapter 12.
To start we pick a gauge choice corresponding to a reference einbein e^. For convenience
we may want to take the reference einbein to be e^ = 1. This reference einbein e^ generates
a gauge orbit, the family {^ eα } of all einbeins obtained from e^ by some diffeomorphism α:
e^ 7→ e^α . Using eqn. (2.11), starting from an arbitrary einbein e one can reach in principle
d
the reference einbein e^ = 1 with a diffeomorphism α that satisfies dτ α(τ) = e(τ). Choosing
the boundary condition α(0) = 0 one has then
Zτ
α(τ) = e(τ 0 )dτ 0 , (2.15)
0

hence it seems that all metrics on the worldline can be brought to the reference metric by
a diffeomorphism. However the periodicity of the einbein is not preserved, as α(1) 6= 1 for
a generic diffeomorphism; this is a global obstruction for all metrics on the closed worldline
being diffeomorphic-equivalent. The invariant length of the path is, as its name suggests,
invariant under diffeomorphisms:
Z1 Z τ̃(1)
T = e(τ)dτ = ẽ(τ̃)dτ̃ . (2.16)
0 0

Hence the positive parameter T labels gauge-equivalent classes of metrics over closed world-
lines; it is called a modulus. If one fixes the integration domain [0, 1] to preserve the peri-
odicity, the reference einbein should be defined accordingly. We choose then R1 our reference
einbein, in a class of metrics of invariant length T , as e^(T ) := T , such that 0 e^(T )(τ)dτ = T .
In the path integral, one should perform the ordinary integral over all possible values of T ,
as we integrate over all possible geometries of the worldline. To take care of the translation
symmetry of the closed path τ 7→ τ+τ0 , one splits a generic diffeomorphism into a translation
and a coordinate transformation orthogonal to it, i.e. a differentiable function α such that
α(0) = 0 (as we have assumed already).
The functional integral measure De splits then Rinto a gauge-invariant measure Dα over
the gauge group and an integral over the modulus, dT .

19
Bosonic strings: action and path integral

As in ordinary gauge theories like quantum chromodynamics in four space-time dimen-

sions, one introduces then the Faddeev-Popov determinant through the relation:
Z Z
1
:= dT Dα δ(e − e^α (T ))δ(α(0)) , (2.17)
∆fp (e)
The distribution δ(e − e^α (T )) will eventually project the integral over the metrics onto
an integral over the chosen gauge slice in field space, which is a one-parameter family of
diffeomorphism-inequivalent reference einbeins e^(T ) depending of the moduli T of the closed
path.
The Faddeev-Popov determinant should be thought of as the Jacobian of the change of
coordinates from De, the integral over all one-dimensional einbeins, to dT Dα, the integral
over the gauge slice on the one hand, and over the directions orthogonal to it – in other
words over the gauge orbits – on the other hand.
Finally, the distribution δ(α(0)) ensures that we consider only diffeomorphisms keeping
the origin fixed, as explained above.
One can plug this expression into the path integral (2.14) and integrate readily over the
einbein e:
Z Z  Z
De dT Dα δ(α(0)) 1 1
 
1 2
Z1 = ∆fp (e)δ(e − e^α (T )) Dx exp − dτ 2
ẋ + m e
Vol(diff) 2 0 e
Z Z Z  Z
Dα δ(α(0)) 1 1
 
1
= dT eα (T )) Dx exp −
∆fp (^ dτ 2 2
ẋ + m e^α (T ) . (2.18)
Vol(diff) 2 0 e^α (T )

The last expression can be simplified further by noticing that (i) the Faddeev-Popov deter-
minant is gauge-invariant (being defined as an average over the gauge group) and and (ii)
that by trading the functional integral over xµ by the functional integral over the transformed
field xµα under the diffeomorphism α, the integrand
R of the integral over the gauge group is
actually gauge-independent, hence the integral Dα factors out and cancels the volume of
the gauge group, except the factor T corresponding to the group of translations giving finally:
Z∞ Z  Z
dT − m2 T 1
Z1 = e 2 ∆FP (T ) Dx exp − 2
dτ ẋ . (2.19)
0 T 2T

The determinant ∆FP (T ) can be expressed as a functional determinant as follows. Using

the infinitesimal expression (2.13) for the gauge transformation of the einbein, together with
an infinitesimal variation of the loop modulus T 7→ T + χ, one can write
  Z R1
d d
δ(T − (T + χ) ) = δ χ − T = Dβe−2iπT 0 dτ β( dτ −χ/T ) , (2.20)
dτ
where one has introduced a path integral over a Lagrange multiplier field β to implement the
desired constraint in field space. Similarly one can write
Z
δ((0)) = dλ exp −2iπλ(0) . (2.21)

20
Bosonic strings: action and path integral

To obtain the Faddeev-Popov determinant, rather than its inverse, as a functional integral,
one can trade (β, , χ, λ) for Grassmann variables (b, c, ψ, ρ), i.e. fermionic variables, and
write, after some rescaling of the fields
Z R1 dc
∆fp (T ) = DbDcdψdρ e−T 0 dτ b( dτ −ψ/T )−ρc(0) . (2.22)

One can perform immediately the integral over the constant Grassmann variables ψ and ρ,
which gives finally Z Z 1 R1

dc
∆fp (T ) = DbDc dτ b c(0) e−T 0 dτ b dτ . (2.23)
0

In other words, one has inserted into the path integral over (b, c) the mean value of b(τ) over
the worldline, i.e. the zero-mode of the field, as well as c(0); both insertions are actually
necessary to cancel the integration over the zero-modes of the fields in the path integral as
we will see shortly.

Functional determinants
We need now to perform the functional integral over the coordinate fields xµ . One considers
then the path integral Z
1
R1 dxµ dxµ
Dx e− 2T 0 dτ dτ dτ . (2.24)

One expands then xµ over a complete set of eigenfunctions of the positive-definite operator
− T1 ∂2τ satisfying the right boundary conditions. It is convenient to separate the zero-modes,
i.e. the classical solutions of the equations of motion, from the fluctuations:

xµ (τ) = xµ0 + qµ (τ) , (2.25)

where q(τ) satisfies the Dirichlet boundary conditions q(0) = q(1) = 0, such that x(0) =
x(1) = x0 . The norm in field space for the fluctuations is naturally
Z1 Z1
1 T
||q|| =
2 2
e^T (τ)dτ q (τ) = dτ q2 (τ) . (2.26)
2 0 2 0

One has then the expansion on an orthonormal basis on eigenfunctions of the differential
operator D = − T1 ∂2τ with the right boundary conditions:

X
∞
2
µ
q (τe ) = cµn √ sin πnτ . (2.27)
n=1
T

and the measure of integration is

Y
∞
Dx = dx0 dcn . (2.28)
n=1

21
Bosonic strings: action and path integral

The ordinary integral over the zero-mode x0 gives the (infinite) volume V of space-time, while
from the integration over the non-zero modes one gets
ZY 2 n2 2
!D
Y πn2
!−D/2
−π cn
dcn e T2 = (2.29)
n n
T2

This product of eigenvalues is divergent and needs Q

to be regularized.
Functional determinants of the form det(D) = ∞ n=1 λn can be evaluated using the zeta-
function regularization. One assumes that λ1 6 λ2 6 · · · 6 λn 6 · · · and defines the spectral
zeta-function as
X
∞
ζD (z) = λ−z
n , (2.30)
n=1

which converges provided <(z) is large enough. It can be analytically continued to the whole
z plane except possibly at a finite set of points. Next we notice that
X
∞
log det D = log λn = −ζD0 (0) , (2.31)
n=1

and define the regularized functional determinant as

Y 0
λn = e−ζD (0) . (2.32)
n

In the present case one has

X
∞ 
πn2
−z 
T2
z
ζD = = ζ(2z) , (2.33)
n=1
T2 π

in terms of the Riemann zeta-function ζ. Since ζ(0) = − 12 and ζ 0 (0) = − 12 ln 2π, the path inte-
gral over xµ (τ) gives finally, dropping the infinite volume factor and after some T -independent
rescaling Z
1
R1 2
Dx e− 2T 0 dt ẋ = T −D/2 . (2.34)

This result can be obtained – in a perhaps simpler way – by viewing the path integral over
a closed loop in Euclidean time as a partition function. One has
Z R1 1 2
Zx = Dx e− 0 dτ 2T ẋ = Tr e−β H , β = 1 .


(2.35)

The Hamiltonian H = T2 p2 is the same as a free (non-relativistic) particle of mass 1/T in D

spatial dimensions, and the computation of the partition function gives simply
Z D
d p −T p2
Zx = D
e 2 = (2πT )−D/2 . (2.36)
(2π)

22
Bosonic strings: action and path integral

We now turn to the evaluation of the ghost path integral. We start with the expression
of the FP determinant that we have obtained before:
Z Z 1  R1 dc
∆fp (T ) = DbDc dτ b c(0) e−T 0 dτ b dτ . (2.37)
0

Because b and c are ghosts, with are dealing with fermionic variables with periodic (rather
than anti-periodic) boundary conditions, hence having zero-modes. We recall here the rules
of integration over Grassmann variables:
Z Z Z
dθ = 0 , dθ θ = 1 , dθ f(θ) = f 0 (0) . (2.38)

which implies that, to get a non-zero answer, the integrand should contain the right number
of zero-modes to cancel the corresponding zero-mode integration measure. Fortunately, the
path integral (2.23) contains the right number of insertions of ghosts zero modes. The integral
over the zero-modes yields Z
db0 dc0 b0 c0 = 1 . (2.39)

Deriving that the integral over the fluctuations is trivial is a little bit subtle.
It is simpler to move from the Lagrangian formalism to the Hamiltonian formalism and
consider this problem from a statistical mechanics point of view. The equations of motion
for the b and c classical fields are
ḃ = ċ = 0 (2.40)
hence, with periodic boundary conditions, the classical solutions are just the two zero-modes
b0 and c0 . In the quantum theory, since b can be seen as the canonical momentum conjugate
to c, one has the anti-commutator
{b0 , c0 } = 1 . (2.41)
Since the Hamiltonian vanishes the Hilbert space contains two states of zero energy, |±i, that
satisfy

b0 |−i = 0 , b0 |+i = |−i

c0 |+i = 0 , c0 |−i = |+i . (2.42)

From these relations one learns that b0 c0 projects onto the ground state |−i. Then the path
integral on a Euclidean circle of length T is interpreted as a thermal average of b0 c0 at inverse
temperature β = 1, and one has4 :
Z R1 dc
DbDc b0 c0 e− 0 dτ T b dτ = h−|e−H |−i = 1 . (2.43)
4
To be precise, the Euclidean fermionic path integral with periodic boundary
 conditions rather than anti-
periodic is not exactly the partition function but rather Tr (−1)F exp(−βH) , where F = b0 c0 counts the
number of fermionic excitations.

23
Bosonic strings: action and path integral

Worldline versus QFT

Collecting all pieces, the path integral computation of the vacuum amplitude gives, up to an
overall normalization factor, Z∞
dT − m2 T
Z1 = 1+d/2
e 2 (2.44)
0 T
This integral is clearly divergent for T → 0, i.e. when the particle loop shrinks to zero-size.
The QFT picture will give us a more familiar understanding of this divergence.
Let us then consider the one-particle contribution to the vacuum energy of a free massive
Klein-Gordon QFT. One has
Z R
Z
− 12 dD x φ(−∇2 +m2 )φ 2 −1/2 1 dD p
Zkg = log Dφ e 2
log(p2 + m2 )


= log det(−∇ + m ) =−
2 (2π)D
(2.45)
We can now move to Schwinger parametrization by using the simple identity
Z∞
1 1 dT −Tx
a
= e , a > 0. (2.46)
x Γ (a) 0 T 1−a
which allows to get the general result
Z D Z∞ Z d Z∞
d p 1 1 dT d p −T (p2 +m2 ) 1 dT −Tm2
d 2 2 a
= 1−a D
e = e (4πT )−D/2 .
(2π) (p + m ) Γ (a) 0 T (2π) Γ (a) 0 T 1−a
(2.47)
At first order in the expansion in the parameter a one gets formally
Z Z∞
1 dD p 2 2 1 dT −Tm2
Zkg = − log(p + m ) = lim e (4πT )−D/2 . (2.48)
2 (2π)D 2 a→0 0 T 1−a
In other words,
Z Z∞ Z∞
1 dD p dT − T (p2 +m2 ) 1 dT − m2 T
Zkg = e 2 = e 2 (2πT )−D/2 , (2.49)
2 (2π)D 0 T 2 0 T
which is the same, up to the numerical normalization factor that we did not compute precisely,
the same as the worldline computation (2.44).
As was noticed before, the expressions (2.44,2.49) present a divergence for T → 0; its
origin is clear from eqn. (2.48). In the momentum-space expression (2.45), it is understood
as the usual UV divergence of the loop integral for ||p|| → +∞. One can compute directly
the integral in (2.47) and get:
Z D
d p 1 Γ (a − D/2) D−2a
D 2 2 a
= (4π)−D/2 m . (2.50)
(2π) (p + m ) Γ (a)
In the a → 0 limit, matching the O(a) terms on both sides yields to
Z D
d p 2
d
log(p2 + m2 ) = (4π)−D/2 Γ (1 − D/2)mD . (2.51)
(2π) d

24
Bosonic strings: action and path integral

Hence the ultraviolet divergence can be removed by dimensional regularization, as usual.

By analogous computations, one can get the Klein-Gordon propagator by considering
open worldlines between two points x and x 0 in Euclidean space-time. It is also interesting to
consider the worldline path integral in the presence of an electromagnetic potential, starting
from the more general action (2.10); in this way one can get for instance the photon N-
point function in scalar QED, in particular the vacuum polarization (N = 2). Choosing a
constant electric field, one finds also the famous Schwinger formula for production of charged
particles/antiparticle pairs in an electric field [3].
We have learned with from this little exercise two important lessons that will be important
in the forthcoming study of relativistic strings:

1. quantum mechanics (or equivalently 0+1 dimensional QFT) on the worldline of a mas-
sive relativistic charged particle provides an equivalent formulation of massive scalar
QFT in an external electromagnetic field;

2. The UV loop divergences of QFT are mapped in the worldline formalism to closed
wordlines shrinking to zero size.

There exists several extensions of this worldline formalism, in order to include spinors, non-
Abelian gauge interactions, etc... As this is not the main topic of the lectures we will not
comment further but refer the interested reader to [4].

2.2 Relativistic strings

We start our journey in string theory by considering the exact analogue of the relativistic
point particle in the worldline formalism for relativistic objects with an extension in a space-
like direction – the fundamental strings of string theory. These objects have a mass per unit
length, or tension T , which is by tradition parametrized as
1
T = , (2.52)
2πα 0
where the parameter α 0 , the Regge slope, has dimensions of length squared. The interested
reader may consult section 1.2 to have some idea about where this terminology comes from.
In order to have a finite energy, the strings should have a finite length. It leaves two
possibilities, topologically speaking: a loop or an interval, which are called closed strings
and open strings respectively, see figure 2.2. In both cases the position along the string is
parametrized by σ.
σ σ

Figure 2.2: Closed strings (left) and open strings (right).

25
Bosonic strings: action and path integral

Figure 2.3: Closed string worldsheet.

Open strings are a little bit more subtle to handle, as one has to specify what are the
boundary conditions at the end of the strings. We will deal mostly in these lectures with
closed strings.

Classical closed strings

A propagating relativistic closed string swaps in spacetime a two-dimensional surface s, or
worldsheet, in analogy with the worldline of point particles. It has the topology of a cylinder,
parametrized by σ for the space-like direction and τ for the time-like direction, see fig. 2.3.
For closed strings, the coordinate σ is periodic. We choose the convention

σ ∼ σ + 2π . (2.53)

The worldsheet swapped by a relativistic closed string in a space-time M of metric G

defines an embedding map, given in a patch of the manifold M by

S1 × R ,→ M (2.54)
(σ, τ) 7→ xµ (σ, τ) , µ µ
x (σ + 2π, τ) = x (σ, τ) , (2.55)

where the set of functions {xµ (σ, τ), µ = 0, . . . , D−1} should be periodic in σ. For convenience
we will use the notation (σ0 , σ1 ) = (τ, σ). The codomain of the map, i.e. the space-time M
where the string leaves, is usually called the target space of the string.
The space-time metric Gµν (xρ ) of the ambient space-time induces a metric h on the
world-sheet parametrized by σ and τ:

∂xµ (σk ) ∂xν (σk )

hij = Gµν , (2.56)
∂σi ∂σj
such that the area element on the worldsheet is given by
√
dA = − det h dσ0 dσ1 . (2.57)

26
Bosonic strings: action and path integral

The negative sign in the square root takes into account that the tangent space of the surface
can be split into a time-like and a space-like directions over every point.
In complete analogy with the relativistic particle case, see eq. (2.5), one postulates a
string action of the form
Z Z 2π r
1 0 1 ρ (σk )]
∂xµ (σk ) ∂xν (σk )
Sng = − dσ dσ − det G µν [x . (2.58)
2πα 0 0 ∂σi ∂σj

This action is known as the Nambu-Goto action [5, 6]. It is invariant under diffeomorphisms
of the worldsheet as it should:

σi 7→ σ̃i (σj ) (2.59a)

dσ0 dσ1 →
7 (det ∂i σ̃j )−1 dσ̃0 dσ̃1 (2.59b)
∂xµ ∂xν ∂xµ ∂xν
7 → ∂i σ̃l ∂j σ̃m . (2.59c)
∂σi ∂σj ∂σ̃l ∂σ̃m
In Minkowski space-time (Gµν = ηµν ) the action is also invariant under space-time Poincaré
transformations xµ 7→ λµν xν + aµ .
In the point particle case, there was a natural coupling to the electromagnetic potential,
i.e. to the one form A(xµ ) = A(xµ )ρ dxρ . There exists an analogous coupling allowed for the
string, but this time to a two-form potential B(xµ ) = 12 Bρσ (xµ )dxµ ∧ dxν , which is called the
Kalb-Ramond potential [7]:5
Z Z
1 1 µ ∂xµ ∂xν i
Skr = − B = − Bρσ [x (τ)] dσ ∧ dσj
2πα 0 s 4πα 0 s ∂σi ∂σj
Z
1 0 1 ij ∂xµ ∂xν
=− dσ dσ  B ρσ . (2.60)
4πα 0 s ∂σi ∂σj
As its one-dimensional cousin, the Kalb-Ramond coupling is independent of the parametriza-
tion of the worldsheet as it does not depend explicitly on the two-dimensional induced metric
h. Furthermore, one can naturally associate to the coupling (2.60) a gauge invariance

Bµν dxµ ∧ dxν 7→ Bµν dxµ ∧ dxν + d(Λν dxν ) = Bµν dxµ dxν + ∂µ Λν dxµ ∧ dxν , (2.61)

where the parameter of the gauge transformation is a one-form Λ = Λµ dxµ . This transfor-
mation leaves invariant (2.60) up to boundary terms using Stokes’ theorem:
Z Z Z Z Z
B 7→ B + dΛ = B + Λ. (2.62)
s s s s ∂s
5
We use the two-dimensional epsilon symbol with non-zero components 01 = −10 = 1. Note that
(− det γ)−1/2 ij is a two-index contravariant antisymmetric tensor.

27
Bosonic strings: action and path integral

Polyakov action
The non-linear Nambu-Goto action (2.58) is not a convenient starting point for quantizing
the theory. In analogy with the relativistic particle case, we will introduce an independent
worldsheet metric γ and consider instead the action known as the Polyakov action [8]:
Z
1 2
p
Sp = − d σ − det γ γij Gµν ∂i xµ (σk )∂j xν (σk ) , (2.63)
4πα 0 s

where the non-dynamical field γij (σk ) is determined by its equation of motion. The Polyakov
action can be understood as the minimal coupling of a two-dimensional metric to a set
of scalar fields, hence is automatically invariant under diffeomorphisms of the worldsheet
σi 7→ σ̃i (σk ).
The equations of motion for the scalar fields xµ (σi ) are easy to determine if the space-time
metric Gµν is constant ; let us assume then that the string propagates in Minkwoski space.
Under an arbitrary variation of the fields xµ 7→ xµ + δxµ the variation of the action is
Z
1 2
p
δSp = − 0
d σ − det γ γij ηµν ∂i xµ (σk )∂j δxν (σk )
2πα s
Z
1 2
p
ij µ k

= d σ ∂j − det γ γ η ∂
µν i x (σ ) δxν (σk )
2πα 0 s
Z
1 2
p
− det γ γij ηµν ∂i xµ (σk )δxν (σk ) . (2.64)


− 0
d σ ∂ j
2πα s
While the third term is a total derivative, and therefore do not play any role on a closed
worldsheet which has no boundaries, the second term gives a two-dimensional Laplace equa-
tion :
∇i ∇i xν (σk ) = 0 . (2.65)
So the fields xµ (σk ) are just free massless fields in two-dimensions. If the string propagates in
a curved space-time, one gets an interacting two-dimensional field theory ; one can see that
this theory is weakly coupled if the curvature of the space-time manifold is small everywhere.
Let us now prove that the equation of motion of the dynamical metric γ in the Polyakov
action (2.63) gives back the Nambu-Goto action (2.58). Under an infinitesimal variation
γ 7→ γ + δγ, one finds that at first order
Z  √ 
1 2
p δ( − det γ) kl ij µ ν
δSp = − d σ − det γ √ γ hkl + δγ Gµν ∂i x ∂j x . (2.66)
4πα 0 s − det γ

Given that γij γjk = δi k one finds at first order that

γij δγjk + γjk δγij = 0 =⇒ δγjk = −γij γkl δγil (2.67)

and √
δ( − det γ) 1 1
√ = δ ln(− det γ) = γij δγij . (2.68)
− det γ 2 2

28
Bosonic strings: action and path integral

We obtain then
Z  
1 2
p 1 ij kl ik jl
δSp = − d σ − det γ γ γ hkl − γ γ hkl δγij . (2.69)
4πα 0 s 2
The vanishing of the first order variation leads therefore to
1 ij kl 1
γ γ hkl = γik γjl hkl =⇒ hij = γkl hkl γij . (2.70)
2 2
The determinant of this relation gives
 2
1 kl
det h = γ hkl det γ , (2.71)
2
from which we deduce that
Z Z −1
√

1 2
p ij 1 2 1 kl
Sp = − d σ − det γ γ hij = − d σ − det h γ hkl γkl hkl
4πα 0 4πα 0 2
Z
1 2
√
=− d σ − det h = Sng (2.72)
2πα 0
Hence, at least classically, the Nambu-Goto and the Polyakov actions give equivalent dynam-
ics for the relativistic strings.
The Polyakov action (2.63) is certainly not the most general action on the string world-
sheet that one can write. First, the coupling to the Kalb-Ramond field, eq. (2.63), is indepen-
dent of the worldsheet metric hence takes the same form in the Nambu-Goto and Polyakov
formalisms. Second, an acute reader may have wondered why, in the Polyakov action, we did
not include a ”cosmological constant” term
Z
1 p
− d2 σ − det γ Λ , (2.73)
4π
R
that would be analogous to the mass term dτ e(τ) m2 in the worldline action (2.10) for the
relativistic particle. Such a term would imply that
1 ij kl 1
γ γ hkl + α 0 Λγij = γik γjl hkl . (2.74)
2 2
Contracting this equation with γij gives

γkl hkl + α 0 Λ = γkl hkl (2.75)

which has no solutions unless Λ = 0. This is a peculiarity of string actions, which is not
shared with actions of particles or higher-dimensional extended objects.6 We will understand
shortly its significance.
6
Indeed for a p-dimensional extended object, p + 1 being the dimension of its worldvolume, everything is
the same except the contraction with γij which gives p+1 kl 0 kl
2 (γ hkl + α Λ) = γ hkl .

29
Bosonic strings: action and path integral

There exists a last possible coupling of the relativistic string that has no analogue in the
particle case. From the two-dimensional worldsheet metric γ one can construct the Ricci
scalar R[γ], and write down a last term in the action of the Einstein-Hilbert type (after all
we are considering a dynamical worldsheet metric):
Z
1 p
χ(s) = d2 σ − det γ R[γ] . (2.76)
4π
In short, one has traded the problem of quantizing gravity in four dimensions to the problem
of quantizing gravity in two dimensions! Einstein gravity is two dimensions is much simpler,
as first it has no propagating degrees of freedom because of diffeorphism invariance (standard
counting gives -1 degrees of freedom). The Einstein-Hilbert action is actually a topological
invariant of the two-dimensional manifold s, known as the Gauss-Bonnet term. In Euclidean
space it is equal to the Euler characteristic χ(s) of the two-dimensional worldsheet. If the
worldsheet is an oriented surface without boundaries, it is given by

χ(s) = 2 − 2g (2.77)

where g is the number of handles, or ”holes”, of the surface. For the sphere g = 0, the torus
g = 1, etc... We will come back latter to the significance of these topologies.
There exists a generalization of the Einstein-Hilbert term that involves a coupling to a
scalar field in space time Φ(xµ ) and that is not topological:
Z
1 p
Sd = d2 σ − det γΦ[xµ (σi )]R[γ] . (2.78)
4π

The field Φ(xµ ), which plays an important role in string theory, is called the dilaton.
To summarize this discussion, the general fundamental string action is given by the sum
of (2.63), (2.63) and (2.78), hence a (1+1)-dimensional quantum field theory on the worlsheet
given by:
Z Z
1 2
p
ij ij

µ ν 1 p
S=− dσ − det γ γ Gµν +  Bµν ∂i x ∂j x − d2 σ − det γ Φ[xµ (σi )]R[γ]
4πα 0 s 4π
(2.79)
This action describes the propagation of a single relativistic string in a background specified
by a metric G, a Kalb-Ramond field B and a dilaton Φ. The later two have no obvious
interpretation at this stage; note that in four dimensions the Kalb-Ramond field is actually
equivalent to a real pseudo-scalar field as its field strength H = dB is Hodge-dual to the
differential of a scalar field: ?H = da.

2.3 Symmetries
We now turn to the path integral quantization of the bosonic string. To start, one has to
pay attention to the symmetries of the theory, in particular to the gauge symmetries that

30
Bosonic strings: action and path integral

need to be carefully taken care of, as in the example of the point particle that we have dealt
with in section 2.1. As there we will consider the path integral with an imaginary time
coordinate, i.e. we will consider an Euclidean worldsheet of coordinates (σ1 , σ2 ) = (σ1 , −iσ0 )
endowed with an Euclidean metric γ. However we will keep the signature of space-time to
be (−, +, · · · , +).7
To simplify the discussion, consider the action of a string action with vanishing Kalb-
Ramond field and constant dilaton field:
Z Z
1 2
p ij µ ν Φ0 2
p
S= d σ det γ γ G µν ∂ i x ∂j x + d σ det γ R[γ] . (2.80)
4πα 0 s 4π s
The symmetries of the theory splits into worldsheet and target space symmetries. We
will start by looking at the latter. One has first target-space symmetries of the action (2.80)
corresponding to symmetries of space-time.
If the space-time is Minskowki space-time (gµν = ηµν ) the action is invariant under
Poincaré transformations:

xµ 7→ Λµν xµ + aµ , Λ ∈ SO(1, d − 1) . (2.81)

These are global symmetries of the two-dimensional field theory.

Under diffeomorphisms in space-time, i.e. infinitesimal field redefinitions δxµ = rµ [xρ ],
the Polyakov action (2.63) keeps the same form if we perform the change of space-time metric
δGµν = −2∇(µ rν) at the same time, in other words a field redefinition in the two-dimensional
field theory on the worldsheet.

2.3.1 Worldsheet symmetries

The string action (2.80), or its more general version (2.79), is by design invariant under
coordinate transformations, or diffeomorphims of the worldsheet,

σi 7→ σ̃i (σk ) , (2.82a)

∂σk ∂σl
γij 7→ γ̃ij = γk` , (2.82b)
∂σ̃i ∂σ̃j
being a theory of two-dimensional gravity minimally coupled to scalar fields xµ (σ).
The Polyakov action has an extra local symmetry, which corresponds to Weyl transfor-
mations of the world-sheet metric, i.e. local scale transformations:
i
γij 7→ e2ω(σ ) γij , (2.83)

where ω(σi ) is an arbitrary differentiable function on the two-dimensional manifold s. This

property comes from the scaling of the determinant of the metric in d dimensions
Weyl
det γ 7→ e2dω det γ , (2.84)
7
As we will see the price to pay will be the appearance of negative norm states. The latter can be
fortunately removed by using the gauge symmetry of the theory.

31
Bosonic strings: action and path integral

which, specialized to two dimensions, implies that the action (2.63) invariant under Weyl
transformations. The Kalb-Ramond coupling (2.60) is also by definition invariant being
independent of the metric.
The Ricci scalar transforms simply under a Weyl rescaling of the metric. We leave as an
exercise to show that, in d dimensions,

γ 7→ e2ω γ (2.85a)
R[γ] 7→ e−2ω R[γ] − 2(d − 1)∇2 ω − 2(d − 2)(d − 1)∂a ω∂a ω .


(2.85b)
√
In two√dimensions, this √ expression implies that det γ R[γ] transforms as a total derivative,
since det γ ∇2 ω = ∂i ( det γ ∇i ω). One concludes that, at the classical level, the dilaton
action (2.78) is not invariant under Weyl transformations, unless Φ is a constant, in which
case it was expected since the two-dimensional Einstein-Hilbert term (2.76) is topological. We
will see later on that, in the quantum theory, the story is somewhat altered by the presence
of anomalies.
A careful reader would have noticed that the Weyl symmetry is not present in the original
Nambu-Goto action (2.58). One can trace back its origin to the equation of motion for the
worldsheet metric, eq. (2.70), which is invariant under Weyl transformations. This feature
of the Polyakov action is not problematic. The Weyl symmetry is a gauge symmetry, hence
does not really correspond to a symmetry but rather to a redundancy of our description of
the theory. In the path integral quantization of the theory, this gauge symmetry will need to
be taken care of properly, as the diffeomorphism invariance.
Finally, we notice that the cosmological constant term (2.73) that we considered to include
in the action is not Weyl invariant, which explains why this term is forbidden in the first
place by the gauge symmetries of the problem.

2.3.2 Gauge choice

The Euclidean path integral of the fundamental string is defined by a functional integral
over the fields {xµ } as well as over the two-dimensional metrics γ – i.e. over Euclidian
worldsheet geometries – moded out by the volume of the gauge group of the theory, made of
two-dimensional diffeomorphisms and Weyl transformations.
As we have done for the point particle, we will properly define this path integral by
gauge-fixing and introducing the corresponding Faddeev-Popov determinant. To start with,
one associates to each two-dimensional metric γ its gauge orbit, the set of its images {γΞ }
under gauge transformations Ξ = (Σ, Ω) composed of diffeomorphisms and Weyl rescalings:
 i
σ 7→ Σi (σk )
Ξ : k ∂σl (2.86)
γij 7→ exp(2Ω) × ∂σ γ
∂Σi ∂Σj k`

To define properly the gauge-fixing condition, one has then to understand how to classify all
metrics over two-dimensional surfaces into equivalence classes under gauge transformations.

32
Bosonic strings: action and path integral

gs−2 gs0 gs2

Figure 2.4: Perturbative expansion of closed string theory.

The coarser classification of compact two-dimensional surfaces is according to their topol-

ogy. If we restrict ourselves to oriented surfaces without boundaries, their topology is com-
pletely specified by the number of handles in the surface, which is called its genus g. Ex-
plicitly, surfaces with g = 0 have the topology of a sphere, surfaces of genus g = 1 have the
topology of a torus, etc... As we have no reasons to restrict ourselves to a particular type of
surfaces, the Euclidean path integral contains a sum over topologies of the worldsheet. For
surfaces with fixed topology, the value of the second term of the action (2.80) is fixed:
Z
Φ0 p
d2 σ det γ R[γ] = Φ0 χ(s) = Φ0 (2 − 2g) . (2.87)
4π s
We have learned something very interesting. Let us define

gs = exp Φ0 . (2.88)

The path integral, in the sector of genus g surfaces, is weighted by the factor g2−2g
s . In other
words, the sum over topologies is nothing that the perturbative expansion, or loop expansion,
of the theory! The parameter gs is the string coupling constant. This is summarized on
figure 2.4.
Having set the topology of the surface by its genus g, one has to find simple represen-
tatives in each gauge orbit under the action of diffeomorphisms and Weyl transformations.
Locally, as the two-dimensional metric has three independent components, one can use the
reparametrization invariance (i.e. the two functions Σ1,2 (σk )) to bring the metric in a con-
formally flat form:
γij (σi ) 7→ exp(2Ω(Σi ))δij . (2.89)
This is called the conformal gauge. The conformal factor exp(2Ω(Σi )) can be naturally
offset by a Weyl transformation, leaving a flat Euclidian metric. There could be however a
topological obstruction to have a flat metric defined everywhere on the worldsheet (otherwise
the Gauss-Bonnet term (2.76) would always vanish).
It will turn out to be convenient to use complex coordinates8 w = σ1 + iσ2 and w̄ =
σ1 − iσ2 , and the reference metric

dss2 = 2 γ
bww̄ dw dw̄ = dw dw̄ . (2.90)
8
There’s actually much more behind this choice than convenience (but I won’t develop this aspect in
the lectures). We are studying the space of conformal classes of metrics on two-dimensional surfaces which
turns out to be the same as the space of Riemann surfaces, i.e. of complex manifolds of dimension one; an
n-dimensional complex manifold is locally equivalent to Cn and its transition functions are holomorphic.

33
Bosonic strings: action and path integral

In these complex coordinates the integration measure over the worldsheet is

Z Z
dwdw̄ = 2 d2 σ (2.91)

and the holomorphic and anti-holomorphic derivatives:

∂ = ∂w = 12 (∂1 − i∂2 ) , ¯ = ∂w̄ = 1 (∂1 + i∂2 ) .

∂ (2.92)
2

A generic infinitesimal gauge transformation (ı.e. an infinitesimal diffeomorphism together

with an infinitesimal Weyl transformation) around the flat metric is

δγij = 2δωδij − δjk ∂i δσk − δik ∂j δσk , (2.93)

which gives in complex coordinates

¯ w̄) ,
δγww̄ = δω − 21 (∂δw + ∂δ (2.94a)
δγww = −∂δw̄ , (2.94b)
¯
δγw̄w̄ = −∂δw , (2.94c)

where δω, δw and δw̄ are arbitrary differentiable functions of w and w̄.
Finally, using complex coordinates, the Polyakov action (2.63) in the conformal gauge
takes the form Z
1 ¯ ν.
S= 0
dwdw̄ Gµν ∂xµ ∂x (2.95)
2πα
This theory looks awfully simple in this gauge. Whenever the target space is flat, it seems
that string theory reduces to a set of free scalar fields in two dimensions. However, one should
not forget that the equations of motion for the worldsheet metric γ should still be satisfied.
By definition, the variation w.r.t. the metric of Polyakov action, which is just a theory of
two-dimensional gravity coupled to some ”matter” fields xµ , is the stress energy tensor Tij ,
see chapter 3, eq. (3.17) for more details. Hence in the classical theory of strings we have to
enforce the following constraint onto the solutions :

Tij = 0 . (2.96)

These constraints, which are known as Virasoro constraints, are the analogue of Gauss’ law
for electromagnetism. Quantizing such constrained theory is a little bit subtle, and can be
done in particular using the BRST approach that we will develop in chapter 5.

2.3.3 Residual symmetries and moduli

In the study of the point particle path integral, we have seen two global aspects of the
worldline geometry that played an important role: the existence of parameters, or moduli,
that couldn’t be gauged away by the choice of reference metric and the existence of gauge
transformations that did not change the one-dimensional metric. In the string theory case
these two aspects are still relevant, and needs a little bit more effort to be understood.

34
Bosonic strings: action and path integral

The moduli are, by definition, given by changes of the metric that are orthogonal to gauge
transformations, i.e. that cannot be compensated for by a combination of a diffeomorphism
and a Weyl transformation. In other words we consider a change of the metric δγij such that
Z
d2 σ 2δωδij − ∂i δjk δσk − ∂j δik δσk δγij = 0 .


(2.97)

This should hold true for any δω and δσk , hence it leads to a pair of independent relations:
Tr (δγ) = 0 =⇒ δγw̄w = 0 (2.98a)

i
¯ ww = 0
∂δγ
∂ δγij = 0 =⇒ (2.98b)
∂δγw̄w̄ = 0
Solutions of these equations are called holomorphic quadratic differentials. The number of
independent solutions will give the number of moduli of the surface nµ . In the mathematical
literature, the space spanned by these moduli is called the Teichmüller space.
A second mismatch between the space of metric and the space of gauge transformations
corresponds to combinations of diffeomorphisms and Weyl transformations that leave the
metric invariant. From equations (2.94b,2.94c) we learn that they correspond to diffeomor-
phisms satisfying
¯
∂δw̄ = ∂δw = 0, (2.99)
while the compensating Weyl transformation
¯ w̄)
δω = 21 (∂δw + ∂δ (2.100)
is unambiguously determined by eq. (2.94a).
The solutions of these equations are holomorphic vector fields, i.e. vectors fields that are
defined in any open set by a holomorphic function. The key point here is that we need to find
vectors fields that satisfy globally this condition on the whole surface, which is a rather strong
constraint. These solutions are called the conformal Killing vectors (CKV) of the surface;
the number of independent CKV will be called nk .
The number of moduli nµ and of conformal Killing vectors nk for a given surface are
not independent but related to each other by the Riemann-Roch theorem, in terms the Euler
characteristic of the surface, which specifies its topology:
nµ − nk = −3χ(s) = 6(g − 1) . (2.101)

The sphere and the two-torus

We will now move away from this rather abstract discussion and derive in detail the moduli
and conformal Killing vectors for the two most useful examples, the sphere and the torus.
Genus zero surfaces have the topology of a two-dimensional sphere. The Riemann-Roch
theorem (2.101) indicates that nµ − nk = −6. A conformally flat metric on the two-
dimensional unit sphere is given, in complex coordinates, by9
4dwdw̄
ds2 = . (2.102)
(1 + ww̄)2
9
Using the coordinate change w = tan(θ/2)eiφ one gets the familiar metric ds2 = dθ2 + sin2 θdφ2 .

35
Bosonic strings: action and path integral

Figure 2.5: Mismatch between integrals over metrics and over the gauge group (diff+Weyl).

The coordinates (w, w̄) are defined in a patch that excludes the ”south pole” of the sphere at
w → ∞. A patch including the south pole is covered by the coordinates (z, z̄) = (1/w, 1/w̄).
The compactification of the complex plane C̄ = C ∩ {∞} is topologically equivalent to the
two-sphere; in a way the patch containing the south pole has ben shrunk to the point at
infinity.
The sphere has no moduli (in particular the radius can be absorbed by a constant Weyl
transformation) and six conformal Killing vectors. Three of them are easy to identify, the
generators of the Lie algebra so(3). To find all of them, one needs to study the holomorphic
vector fields on this manifold. Let us assume that the holomorphic vector field δw admits a
holomorphic power series expansion around the north pole w = 0:
δw = c0 + c1 w + c2 w2 + c3 w3 + · · · (2.103)
This holomorphic vector field should be defined everywhere, in particular in the patch
around the south pole. Under the coordinate transformation w 7→ z = 1/w, one finds that
∂z
δz = δw = −z2 (c0 + c1 /z + c2 /z2 + c3 /z3 + · · · ) , (2.104)
∂w
hence one gets a globally well-defined holomorphic vector field, in particular at the south
pole z = 0, provided that cn = 0 for n > 3.
The three complex parameters {c0 , c1 , c2 } parametrize generic conformal Killing vectors
of the sphere around the identity. Successive actions of the conformal Killing vectors (2.104)
define a group by exponentiation, the conformal Killing group. One can check – by comparing
the multiplication laws – that this group is actually isomorphic to the Möbius group, i.e. the
group of fractional linear transformations
az + b
z 7→ , a, b, c, d ∈ C , ad − bc 6= 0 . (2.105)
cz + d
Given that the map is invariant under rescalings of the parameters, one can set ad − bc = 1
and these transformations define a group isomorphic to PSL(2, C), the group of complex 2×2
matrices M of determinant one identified under its center M 7→ −M.

36
Bosonic strings: action and path integral

Genus one surfaces are topologically equivalent to a two-torus. The Euler characteristic of
the two-torus vanishes, hence a two-torus can be endowed with a flat metric ds2 = dwdw̄. The
torus has an obvious discrete Z2 symmetry w 7→ −w as well as two conformal Killing vectors
corresponding to translations along the two one-cycles of the torus. They are described
simply by the constant holomorphic vector field δw = c0 . According to the Riemann-Roch
theorem, one expects that the torus has two real moduli.
The torus can be described conveniently as the complex plane quotiented by the discrete
identifications
w ∼ w + 2πnu1 + 2πmu2 , n, m ∈ Z , (2.106)
where u1 and u2 are complex parameters. By a rescaling of w, accompanied by a constant
Weyl transformation, one can get rid of the former hence we consider the quotient
w ∼ w + 2πn + 2πmτ , n, m ∈ Z , (2.107)
where we have adopted the standard notation τ ∈ C for the torus modulus. As for the circular
worldline in the point particle case, see the discussion above eqn. (2.16), an alternative way
to think about the torus is to consider the metric
ds2 = |dσ1 + τdσ2 |2 , (2.108)
with the standard identifications σi ∼ σi + 2π. There exists some discrete ambiguity in
the identification of the parameter τ. First, the metric (2.108) is invariant under complex
conjugation of the parameter τ, hence we can restrict the discussion to τ2 = =(τ) > 0 (the
case =(τ) = 0 being degenerate) i.e. to the upper half plane H. For a square torus, <(τ) = 0,
while in general the real part of τ represents the way the circle parametrized by σ1 is ”twisted”
before identifying the two enpoints of the cylinder.
The two-torus, being defined as a quotient of the complex plane, is nothing but a two-
dimensional lattice, see fig. 2.6. It is obvious that the same lattice is described by replacing

Figure 2.6: Two-torus as a quotient of the complex plane.

τ by τ + 1. According to the metric (2.108) it amounts to redefine σ1 → σ1 + σ2 , which is

compatible with the periodicities of the coordinates.
It it slightly less obvious to realize that another equivalent parametrization of the torus
is given by τ 7→ −1/τ, if one allows a Weyl rescaling of the metric. Indeed starting from the
metric (2.108) one gets
τ7→−1/τ
dσ + τdσ2 2 7−−
2 1  2
−−−−→ dσ1 − τ1 dσ2  = 2 dσ2 − τdσ1  .
 1  
(2.109)
|τ|

37
Bosonic strings: action and path integral

One sees that, up to a global rescaling of the metric, it amounts to replace σ1 → σ2 and
σ2 → −σ1 . In other words it exchanges the role of Euclidean worldsheet time and of the
space-like coordinate along the string. These two transformations generate the modular group
PSL(2, Z), which acts on the modular parameter as

aτ + b
τ 7→ , a, b, c, d ∈ Z (2.110)
cτ + d
This group is indeed the group SL(2, Z) of 2 × 2 integer matrices M of determinant one
 
a b
M= , ad − bc = 1 , a, b, c, d ∈ Z (2.111)
c d

quotiented by its center, i.e. with the identification M ∼ −M, as replacing (a, b, c, d) by
(−a, −b, −c, −d) does not change the action (2.110). To avoid an over-counting in the path
integral, we will choose to select a representative of the modular parameter τ into each orbit
of the modular group. One can show that every point is the upper half plane H > 0 has a
unique antecedent under the modular group in the fundamental domain F, defined by the
conditions
F = {τ ∈ H , |<(τ)| 6 12 , |τ| > 1} , (2.112)
where the boundaries for <(τ) > 0 and <(τ) < 0 are identified, see fig. (2.7). In other words,
the fundamental domain contains a unique point per orbit of PSL(2, Z).

F τ-plane

−1/2 1/2 1

Figure 2.7: Fundamental domain F of the modular group.

To anticipate a little bit, this technical point will have drastic consequences. Remember
that the torus diagram represents the one-loop contribution in the perturbative expansion
in string theory, much as a circle represented the one-loop contribution in the point particle
case, see section 2.1. We have found there that the UV divergences in QFT were related

38
Bosonic strings: action and path integral

to circle of perimeter T going to zero size in the worldline formalism. In the string theory
case, it would correspond to the limit =(τ) → 0, which is completely excluded from the path
integral if we choose to integrate over F, the fundamental domain! This remarkable feature
of string theory, which persists to higher order, indicates that the theory is UV-finite.

2.3.4 Conformal symmetry

Our discussion of conformal Killing vectors – i.e. of metric-preserving gauge symmetries –
had been global as we only focused on holomorphic vector fields (2.99) defined everywhere
on the manifold. This was important as we wanted to knew exactly the mismatch between
integrating over worldsheet metrics and over diffeomorphisms × Weyl gauge symmetries.
There are however important properties of the quantum field theory defined on the worldsheet
of the string by the gauge-fixed Polyakov action (2.95) that depend only on the symmetries
of the problem in an open set of the complex plane C.
If we allow the transformations (2.99) to be defined only in some patch of the worldsheet,
they generate a much bigger group of symmetry. If we consider finite transformations rather
than infinitesimal ones, what these equations tell us is that for any holomorphic function f,
the change of coordinates

w 7→ f(w) (2.113a)
w̄ 7→ f̄(w̄) (2.113b)
dw dw̄ 7→ ∂w f(w)|−2 dw dw̄

(2.113c)

can be compensated by a Weyl transformation of parameter Ω = log |∂f| such as to leave

invariant the flat two-dimensional metric.
This infinite-dimensional symmetry will play a crucial role in the following. The holo-
morphic coordinate transformations (2.113a,2.113b) preserve the metric up to a local scale
transformation as can be seen from eqn. (2.113c), and the set of such transformations corre-
sponds to the conformal group, which is indeed of infinite dimension precisely for a space of
dimension two.
We will obtain Ward identities associated with these symmetries for the two-dimensional
quantum field theory defined on the worldsheet, as such identities do not care about the well-
definiteness of the symmetry at the global level. What we will obtain is an infinite number
of constraints on this QFT, which is every theoretical physicist dream!

2.4 Polyakov path integral

We have gathered the ingredients to define properly the path integral associated with the
Polyakov action (2.95), which is a path integral over two-dimensional scalar fields xµ and
over two-dimensional metrics γ on Euclidean worldsheets.
We have learned first in section 2.3 that the path integral splits into a sum over topologies,
i.e. over surfaces of different genera g. In each sector we integrate over metrics γg of surfaces

39
Bosonic strings: action and path integral

with a given genus. Accordingly we are considering the formal vacuum amplitude
X
∞ Z Z
Dγg  
Z1 = gs2g−2 Dx exp − Sp [γg , x] , (2.114)
g=0
Vol(diff × Weyl)

that receives contributions from single string worldsheets.

As in the point particle case, we have formally divided the integral over the metrics
γg by the volume of the gauge group Vol(diff × Weyl) in order to take care of the gauge
redundancies of the formulation of the theory. As there we will define properly this integral
by the Faddeev-Popov method.
Inequivalent gauge orbits of metrics γg under diff×Weyl gauge transformations are labeled
by a finite set of nµ parameters m` , the moduli. Along each of these orbits one can choose a
reference metric γ bg (m` ), whose image under a transformation Ξ will be denoted γbgΞ (m` ).
We want to trade the integral over the metrics γg for the product of an integral over the
moduli and an integral over the gauge group, but the latter contains elements that do not
change the reference metric, the conformal Killing vectors. In the point particle case this
extra symmetry was taken care of by fixing the image of the origin under the diffeomorphisms
of the worldline. In the present case we have to discuss separately the different topologies,
as the outcome will be different. As before we will focus on the two most important cases,
the sphere and the torus.

2.4.1 Path integral on the sphere

Surfaces of genus zero have positive curvature, see eqn. (2.76), and are all diff.× Weyl equiv-
alent to the round unit two-sphere, as the latter has no moduli, see section 2.3.
The conformal Killing vectors on the two-sphere form a group isomorphic to PSL(2, C),
the Möbius group, which is a non-compact Lie group of complex dimension three. One could
split the integral over diffeomorphisms as an integral over rotations and over diffeomorphisms
keeping fixed the origin of the coordinates w = w̄ = 0 (as we did for the point particle),
taking care of the rotation subgroup SO(3) ⊂ PSL(2, C) but not of the full Möbius symmetry.
The vacuum amplitude path integral (2.114) contains the inverse of the
volume of the Diff.
× Weyl gauge group and therefore retains a factor of 1/Vol PSL(2, C) which vanishes, as
the Möbius group is non-compact. We learn that the vacuum amplitude on the sphere, i.e.
at tree level, vanishes, as it should be because otherwise it would mean that we don’t expand
around a vacuum of the theory.10
Of course, string theory, as QFT, is not just about computing the vacuum amplitude.
Physical observables correspond generically to time-ordered correlation functions of gauge-
invariant observables of the theory. Because of diffeomorphism invariance, the observables
take the form Z p
Ok = d2 w det γ Vk [xµ (w, w̄)] , (2.115)

10
As we will see below, the same conclusion can be obtained by noting that the zero-modes of the ghost
path integral measure will not be ”saturated” by the appropriate ghost insertions.

40
Bosonic strings: action and path integral

where Vk is some functional of the fields xµ (w, w̄) that transforms as a scalar under world-
sheet diffeomorphisms; other constraints should be imposed on these functionals, and will be
discussed later. Then we have to consider path integrals of the form
Z n Z
Y
hO1 · · · On ig=0 ∼ Dγ0 Dx
p
g−2
s d2 wk det γ Vk e−Sp [x,γ0 ] . (2.116)
k=1

In this context the gauge-fixing problem that we had for the vacuum amplitude is easy to
solve. If the number n of operators is larger or equal to three, the Möbius symmetry is
completely fixed by setting the positions (wk , w̄k ) of three operators to arbitrary fixed values
(w b̄ k ) instead of integrating over them.
b k, w R Q
To formulate the problem a bit differently, the integral Dγ0 3k=1 d2 wk over the metrics
and the positionR of three operators covers the whole gauge group, hence can be traded for
an integral DΞ which cancels out completely the volume of the gauge group in the path
integral (2.114).
We define then the Faddeev-Popov determinant of string theory on the sphere in terms
of the path integral over the gauge group:11
Z Y3
1
 
:= DΞ δ(γ0 − γ Ξ
b0 ) b̄ Ξk
b Ξk δ w̄k − w
δ wk − w (2.117)
∆fp (γ0 ) k=1

where w b kΞ their images under gauge transformations (diffeo-

b k are arbitrary positions and w
morphisms of the surface). We obtain for the full path integral at tree-level
Z Yn Z Z Yn
Dγ0 DΞ
d wk det γ0 Dx
p
hO1 · · · On ig=0 = g−2
s ∆fp (γ0 ) 2
Vk [x(wk , w̄k )]
Vol(diff × Weyl) k=1 k=1
  Y
3
b0Ξ )


× exp − Sp [γ0 , x] δ(γ0 − γ δ (wk − w
b k ) δ w̄k − w
b̄ k
k=1
Z   Y 3 p Y
n
= g−2
s Dx exp − Sp [ γ
b0 , x] ∆fp (b
γ0 ) det γb0 Vk [x(w
b k, w
b̄ k )] OK . (2.118)
k=1 k=4

By diffeomorphisms and Weyl transformations, one can bring the metric on the sphere to
a flat metric, the price to pay being that the south pole is mapped to |w| → ∞. This is not
a problem as long as we consider the compactification of the complex plane, C̄ := C ∩ ∞,
which is topologically a two-sphere.
In order to evaluate the Faddeev-Popov determinant, we first recall that the argument of
the distribution δ(γ0 − γ b0Ξ ) around the reference metric is given by eqns. (2.94). We have
11
Actually one should write ∆fp (γ0 ; wk , w̄k ) as it depends also of the position of the fixed local operators
but we have chosen not to clutter the equations too much.

41
Bosonic strings: action and path integral

then
Z
1 ¯ w̄) δ ∂δw̄ δ ∂δw
¯
DδωDδwDδw̄ δ δω − 12 (∂δw + ∂δ
     
=
∆fp (b
γ0 )
Y
3



δ δw(w
b k, w
b̄ k ) δ δw̄(w b̄ k ) . (2.119)
b k, w
k=1

All these Dirac distributions are exponentiated by means of a corresponding Lagrange mul-
tiplier:
Z Y
3
1
= DδωDδwDδw̄Dη Dβ Dβ̄ dφk dφ̄k
∆fp (b
γ0 ) k=1
R 1

R R
d2 w η ¯ w̄) ¯
d2 w β∂δw d2 w β̄∂δw̄ 2iπφk δw(w
e2iπ δω− (∂δw+∂δ
e2iπ e2iπ b k ,w
b̄ k ) 2iπφ̄k δw̄(w
b k ,w
b̄ k )
2 e e (2.120)

We eventually need to insert the FP determinant rather than its inverse as for the particle,
therefore we substitute for the variables (δω, δw, δw̄, η, β, β̄, φk , φ̄k ), the Grassmann vari-
ables (κ, c, e e ψk , ψ̄k ). We can compute immediately the integrals over κ, ζ, ψk and
c, ζ, b, b,
ψ̄k which gives, after a rescaling of the fields, the relatively simple expression:12
Z R Y
3
d2 w ¯ e
γ0 ) = DbDbDcDe
∆fp (b e c e− 2π
(b∂c+ b∂e
c)
c(w c(w
b k )e b̄ k ) (2.121)
k=1

To summarize the sphere path integral of string theory is given in its full glory by the
expression (n > 3):
Z R R
d2 w ¯ ν − d2 w ¯ e
gµν ∂xµ ∂x
hO1 · · · On ig=0 = g−2
s Dx DbDbDcDe
e c e− 2πα 0 e 2π
(b∂c+ b∂e
c)

Y
3 Y
n
c(w
b k )e b̄ k )Vk [xµ (w
c(w b k, w
b̄ k )] Ok , (2.122)
k=1 k=4

where {w
b k } are arbitrary positions, that we take usually to be {0, 1, +∞}.

2.4.2 Path integral on the torus

The second important case is the two-torus, that corresponds to the one-loop amplitude in
string theory, and present some differences with the previous case.
Gauge-inequivalent surfaces of genus one correspond to two-tori characterized by one
complex parameter τ in the upper half-plane H = {z ∈ C, =(z) > 0}. Because of the ambiguity
12
If we reinstall the index notation, we start with βij ∂i δσj in the expansion of the Faddeev-Popov determi-
nant (2.117), which gives a ghost Lagrangian of the form bij ∂i cj , and in complex coordinates the components
are given below by eqn. (2.130).

42
Bosonic strings: action and path integral

in assigning a modular parameter τ to a given two-torus – or equivalently to a given two-

dimensional lattice – one restricts further τ to be in the fundamental domain F of the modular
group, see eqn. (2.112). A two-dimensional metric on the torus of complex modular parameter

τ = τ1 + iτ2 , |τ1 | 6 1
2
, τ12 + τ22 > 1 (2.123)

is then given by the flat metric (2.108) with σi ∼ σi + 2π.

The two-torus has two real conformal Killing vectors, that are easy to understand, as they
correspond to translations along σ1 and along σ2 . These translations form a group isomorphic
to U(1)2R, whose volume is 4π2 τ2 , the area of the torus of metric (2.108). It means that the
integral d2 τ DΞ over the moduli and over the gauge group covers more that the integral
R
Dγ1 over the metrics on genus one surfaces.
Fixing this extra symmetry is possible, even for the vacuum amplitude in the following
way (this is the same method that we used for the point particle vacuum amplitude). A
general diffeomorphism is given by two functions Σi (σk ) that will not generically map the
origin of the coordinates σ1 = σ2 = 0 to the origin, as Σi (0, 0) 6= 0 in general. Because of
translation invariance it is possible to restrict the path integral to diffeomorphisms preserving
the origin, i.e. such that Σi (0, 0) = 0 or, say differently, for every diffeomorphism Σi one can
translate back the image of the origin to Σi (0, 0) = 0, which will select an element of the
translation group . In this way, the spurious gauge freedom will be removed. In addition to
these continuous symmetries, the torus metric is naturally invariant under the Z2 symmetry
σi 7→ −σi , which doubles the volume of the group of metric-preserving gauge transformations.
These considerations lead to the following expression for the Faddeev-Popov determinant
on the two-torus, as an integral over the gauge transformations Ξ = (Σ, Ω):
Z Z
1  
:= d τ DΞ δ γ1 − γ
2
b1Ξ (τ) δ Σ1 (0) δ Σ2 (0) ,



(2.124)
∆fp (γ1 ) F

with the reference metric γ

b1 (τ) obtained from eqn. (2.108)
 
1 τ1
γ
b1 (τ) = . (2.125)
τ1 |τ|2

Expanding the Dirac distribution near the reference metric gives

 
b1Ξ (τ) =
δ γ1 − γ
   
1 2 2 2 2 1
δ 2δω − 2∂1 (δσ + τ1 δσ ) δ 2|τ| δω − 2∂2 (|τ| δσ + τ1 δσ ) + 2(τ1 δτ1 + τ2 δτ2 )
 
δ 2δωτ1 − ∂1 (|τ|2 δσ2 + τ1 δσ1 ) − ∂2 (δσ1 + τ1 δσ2 ) + δτ1 (2.126)

As before we introduce Lagrange multipliers fields. Compared to the sphere case as the
reference metric is not diagonal it will be technically slightly more cumbersome. We introduce
a symmetric two-index tensor of Lagrange multipliers of components βij as well as a one-form

43
Bosonic strings: action and path integral

of components ηi . It gives
Z
1
= d2 δτ Dδω Dδσi Dβij dηi exp 2iπηi δσi (0)


∆fp (b
γ1 (τ))
Z

exp 2iπτ2 d σ β11 δω−∂1 (δσ1 +τ1 δσ2 ) +β22 |τ|2 δω−∂2 (|τ|2 δσ2 +τ1 δσ1 )+(τ1 δτ1 +τ2 δτ2 )
2

12 2 2 1 1 2


+ β 2δωτ1 − ∂1 (|τ| δσ + τ1 δσ ) − ∂2 (δσ + τ1 δσ ) + δτ1 (2.127)

Integrating over δω imposes that the tensor β is traceless, i.e. that βij γ b1 (τ)ij = 0. The
remaining path integral in Dβ will be therefore on traceless tensors only. In order to get the
FP determinant rather that its inverse, we replace (βij , δσi , ηi , δτ` ) by Grassmann variables
(bij , ci , ψi , κ` ) and get after rescaling of the fields
Z Z

1
γ1 (τ)) = d κ Dc Db dψi exp ψi c (0) exp −
2 i ij i
p
d2 σ det γ b1 (τ) bij ∂i cj


∆fp (b
2π

1 Z p 
2 ij
× exp d σ det γ1 (τ)κ` b ∂τ` γ1 (τ)ij , (2.128)
b b
2π

which can be further simplified by integrating over κ` and over ψi .

Inserting this result in the Polyakov path integral (2.114) one gets finally the vacuum
amplitude as13
Z Z
τ Z
dτdτ̄ 2

Z1 = 2
Dx µ
Db ij
Dci exp − d 2
σ b ij
∂ c
i j + 1
g γ ij
α 0 µν 1 i
b ∂ x µ
∂ j xν
F 16π τ2 4π
Z Z
τ2 τ
c1 (0)c2 (0) × b1 (τ)ij × 2 d2 σ bij ∂τ̄ γ
d2 σ bij ∂τ γ b1 (τ)ij (2.129)
4π 4π
The dependence in 1/τ2 in the measure of integration over the modulus τ comes from the
volume of the group of translations along the torus, since have we have explained we don’t
integrate over this part of the gauge group. As we will see this factor ensures that the result
is invariant under the modular group PSL(2, Z).
It is finally convenient to come back to complex coordinates (w, w̄) = (σ1 +τσ2 , σ1 + τ̄σ2 ).
We write then
bww = b , bw̄w̄ = b e , cw = c , cw̄ = e c (2.130)
The insertion in b(w, w̄) and b(w,
e w̄) takes in this basis the form
Z Z
1 2 1
d w b(w, w̄) d2 w b(w,
e w̄) , (2.131)
2πτ2 2πτ2
13
The exact normalization of the result could be obtained by checking carefully how the insertions in the
path integral are normalized, however since the original path integral is ill-defined, and, since we did several
field rescalings while manipulating formally the path integral, the honest way to get the factor right is to ask
that the end result is properly normalized if interpreted as a partition function. Concretely if one chooses a
fully compact space-time one can request that the vacuum appears with degeneracy one.

44
Bosonic strings: action and path integral

As we will see later, one can replace b(w, w̄) by its value at any given point, for instance
w = w̄ = 0, as the non-zero modes of the field don’t contribute to the path integral (2.129).
R
Therefore we can replace d2 w b → 4π2 τ2 b(0), and the same for the b e insertion.
We get the final result for the vacuum amplitude of string theory at one-loop as
Z Z R 2
d2 τ 1


− d2πw ¯ e ¯ ν
g ∂xµ ∂x
Z1 = Dxµ DbDb
e DcDe
c c(0)e
c(0) b(0)b(0)e
e b∂c+ b∂e
c+
α 0 µν (2.132)
F 4τ2

The dependence of the integrand in the modulus τ is hidden in the periodicity of the complex
variable w ∼ w + 1 ∼ w + τ.
If one wants to compute a different observable as an n-point function hO1 · · · On ig=1 , it
is enough to insert the operators Ok in the path integral above, as no gauge-fixing of the
position of some operators is needed in the present case.
This result is close to the sphere path integral (2.122), the differences reflecting the number
of moduli and conformal Killing vectors in each case. One can generalize of course this
discussion to surfaces of higher genera, but a rigorous presentation would be rather technical.
In this chapter we have assumed that the gauge symmetries of the classical theories,
diffeomorphisms and Weyl transformations, were also valid in the quantum theory. As we
shall see, the latter may be violated by an anomaly that put the theory in danger of being
inconsistent. Before proceeding to this computation, we will introduce in the next chapter
the powerful methods of two-dimensional conformal field theory.

2.5 Open strings

We will close this chapter with a very brief overview of open strings. Most of the mate-
rial developed in the previous sections is relevant for open strings, however with some key
modifications.
First, the worldsheet of an open string is a strip rather than a cylinder, see fig. 2.8. One
has therefore a map :

[0, π] × R ,→ M (2.133)
(σ, τ) 7→ xµ (σ, τ) , (2.134)

where the set of functions {xµ (σ, τ), µ = 0, . . . , D − 1} have some boundary conditions at the
ends of the interval, σ ∈ {0, π}, that we will specify below.
Let us consider a classical open string in Minkowski space-time. It is governed by the
same Polyakov action as the bosonic string :
Z
1 2
p
Sp = − d σ − det γ γij ηµν ∂i xµ ∂j xν , (2.135)
4πα 0 s
If one considers the variation of the action w.r.t. the fields xµ , one picks in the present case

45
Bosonic strings: action and path integral

Figure 2.8: Open string worldsheet.

a boundary term :
Z
1 2
p 
δSp = 0
d σ ∂j − det γ γ ηµν ∂i x δxν
ij µ
2πα s
Z
1 p
− dτ − det γ γ0i ηµν ∂i xµ δxν . (2.136)
2πα 0 ∂s
µ
While the first term vanishes if the fields x satisfy the equations of motion (2.65), one should
impose suitable boundary conditions on the boundary of the worlsheed ∂s (i.e. for σ = 0 and
σ = π) in order to cancel the boundary term. The possible choices are :
• Dirichlet boundary conditions. They are defined as δxν (σ = 0, π) = 0 ; in this case
the endpoints of the string cannot move.
• Neumann boundary conditions. They are defined as ∂i xµ (σ = 0, π) = 0 ; in this case
the endpoints of the string are free to move since there are no constraints on δxν .
If one wants to retain full Lorentz invariance in the target space-time, it is natural to
take the same boundary conditions in all directions, i.e. for all fields xµ . Then the Dirichlet
boundary conditions looks rather unnatural, as the endpoints of the strings would be stuck
in particular in the time direction. Therefore the natural boundary conditions for the open
strings, at this stage, are Neumann boundary conditions for all fields xµ .
As for closed strings, one can consider open strings coupled to more general background
fields. The action (2.79) that describes the couplings of strings to a metric Gµν , a Kalb-
Ramond field Bµν and a dilaton Φ(xµ ) receives two new contributions. First, one can write
a curvature term on the curve defined by a boundary, known as geodesic curvature14 which
leads to a new boundary coupling to the dilaton :
Z
1
Sgc = − ds Φ(xµ )k(s) . (2.137)
2π ∂s
14
The geodesic curvature of a boundary defined as k = ±ta nb ∇a tb , where ta is a unit vector tangent to
the boundary and na is a vector orthogonal to it pointing outwards. The plus and minus signs correspond
respectively to time-like and space-like boundaries.

46
Bosonic strings: action and path integral

On an Eulidean world-sheet and with constant dilaton, this term and the other dilaton
coupling combine to the Euler characteristic of a surface with boundaries :
Z Z
1 2
p 1
χ(s) = d σ det γ R[γ] + ds k(s) . (2.138)
4π 2π ∂s
For a surface with g handles and b boundaries, the Euler characteristic is given by:

χ(s) = 2 − 2g − b . (2.139)

Compared to the perturbative expansion of a theory of closed strings, see fig. 2.4, the
perturbative expansion of a theory containing open strings receive new types of contributions
corresponding to surfaces with boundaries, see figure 2.9. Restricting as before to compact
surfaces, the first open string contributions are a disk (χ = 2 − 0 − 1 = 1) and a cylinder
(χ = 2 − 0 − 2 = 0). As we will see, the disk is associated with tree-level contributions

gs−1 gs0

Figure 2.9: Perturbative expansion for open strings.

while the cylinder is associated with a one-loop contribution, if one considers the periodic
coordinate on the cylinder to be associated with Euclidean time on the worldsheet.

2.5.1 D-branes
D-branes appear quite naturally in string theory if one forgoes Lorentz invariance in the D-
dimensional target space-time. One may impose only Lorentz invariance in p + 1 space-time
dimensions, in which case one can choose the following boundary conditions :

• Neumann boundary conditions for xµ , µ = 0, . . . , p

• Dirichlet boundary conditions for xM , M = p + 1, . . . , D − 1.

Hence, at a given time, the endpoints of the open strings are attached to a space-like sub-
manifold of dimension p. This sub-manifold, together with time, can be viewed as the
world-volume of a new extended object of string theory, known as a Dp-brane (where D
stands, of course, for Dirichlet).15 In the quantum theory this extended object is not rigid,

47
Bosonic strings: action and path integral

Figure 2.10: Open string with both ends on the same D-brane (blue) and ending on two
different D-branes (magenta).

but rather acquires dynamical properties. Open strings can have either both ends attached
to the same D-brane or each end of a different D-brane, see fig. 2.10.
On top of the boundary curvature coupling, see eqn. (2.137), another type of boundary
term is allowed for the action of an open string. Let us consider for the moment that the open
string has Neumann boundary conditions along all space-like directions, i.e. that preserving
Lorentz invariance in D-dimensional space-time. Since each boundary of the string sweeps
a worldline in space-time, it couples naturally to a one-form, i.e. to an Abelian gauge field,
as the point-particle example studied in section 2.1. We supplement therefore the Polyakov
action of the string by the following contribution:
Z
Sa = Aµ dxµ . (2.140)
∂s

If the open string ends on a Dp-brane (with p < D − 1), it couples to an electromagnetic
fields in p + 1 dimensions. One can view this gauge field as living on the world-volume of
the Dp-brane. If both ends of the string are on different D-branes, then each end couples to
a different gauge field, each living on the corresponding worldvolume.16
Invariance of the coupling w.r.t. gauge transformations of the gauge fields is immediate,
as in the point particle case. There is however another type of gauge invariance that we
should consider here. The variation of the Polyakov action under a gauge transformation of
the Kalb-Ramond field Bµν was given, see eqn. (2.62), by
Z Z Z
B 7→ B + Λµ dxµ . (2.141)
s s ∂s

On a worldsheet with boundaries, the second term is not zero, and should be canceled against
the variation of the Abelian gauge field coupling 2.5.1. Therefore the general transformation
15
Whenever one chooses Neumann boundary conditions in all directions, one can consider that all space is
filled by a D-brane.
16
Whenever the positions of N Dp-branes coincide in all their transverse dimensions, the gauge symmetry
U(1)N is actually enhanced to U(N) and the gauge theory becomes non-Abelian.

48
Bosonic strings: action and path integral

of the gauge field is

1
Aµ 7→ Aµ + ∂µ λ − Λµ , (2.142)
2πα 0
where the scalar λ is associated with the gauge symmetry of the Maxwell field Aµ and the
one-form Λµ dxµ to the generalized gauge symmetry of the Kalb-Ramond field Bµν . Notice
that the field strength Fµν of the Maxwell field is not invariant under the latter. Rather, the
invariant combination under both gauge symmetries is
Fµν = Bµν + 2πα 0 Fµν , (2.143)
which is therefore the natural object that should appear in low-energy effective actions for
D-branes degrees of freedom.

2.5.2 Path integral

We end this section with a brief discussion about the path integral quantization of open
strings. The story is very similar to the case of closed strings, except that the number of
moduli and conformal Killing vectors differ for the surfaces with boundaries under consider-
ation. The Riemann-Roch theorem still holds, with now
nµ − nk = −3χ(s) = 3(2g + b − 2) . (2.144)
The first surface under consideration is the disk, with g = 0 and b = 1 hence nµ −nk = −3.
As the unit sphere, the disk has no moduli (since a change of radius can be absorbed by a
Weyl rescaling). To obtain the conformal Killing vectors, one can obtain the unit disk from
the unit sphere using the identification :
z ∼ 1/z̄ . (2.145)
This identifies pairwise points inside and outside the unit disk in the complex plane, and the
boundary of the disk |z| = 1 is left invariant. The conformal Killing vectors of the disk are
therefore obtained from those of the sphere that leave the boundary of the disk invariant. It
is rather straightforward to see that it consists of a subgroup of the Möbius group PSL(2, C),
see eqn. (2.105), whose elements are of the form
z+b
z 7→ eiφ , |b| < 1 , φ ∈ R . (2.146)
b̄z + 1
It contains in particular rotations along the circle (b = 0). It is convenient to map the unit
disk to the upper half-plane using the conformal mapping
z+i
z 7→ w = (2.147)
iz + 1
sending the boundary of the disk |z| = 1 to the real axis w ∈ R. In this case the relevant
conformal Killing group is the subgroup of Möbius transformations leaving the real axis
invariant, i.e. the group PSL(2, R) :
aw + b
w 7→ , ad − bc = 1 , a, b, c, d ∈ R , (a, b, c, d) ∼ (−a, −b, −c, −d) . (2.148)
bw + d

49
Bosonic strings: action and path integral

One can show, conjugating with the map (2.147), that the group of transformations (2.146)
is isomorphic to PSL(2, R).
Since this group is non-compact, the situation is similar to what we have found for the
sphere in the closed string sector. The vacuum amplitude vanishes, and the first non-vanishing
observable in the open string sector at tree-level is the correlation function of three boundary
operators. There is another possibility however, with no direct analogue in the closed string
sector. One can consider a one-point function for an operator inside the disk, for instance at
the origin. Such configuration preserves only a compact subgroup U(1) of PSL(2, R) hence
gives a non-zero answer. It represents the coupling between a closed string mode and a
D-brane.
The cylinder has no holes and two boundaries and therefore, nµ −nk = 0. This surface has
a conformal Killing vector, corresponding to translations along the cylinder (i.e. in imaginary
time if one considers the open string channel), and a single real modulus, which can be taken
to be the length of the cylinder. One can therefore parametrize it as follows :

w = σ1 + iσ2 , σ1 ∈ [0, π] , σ2 ∼ σ2 + 2πt , t ∈ R>0 . (2.149)

One may wonder what would happen if we consider instead that the coordinate σ2 along
the one-cycle of the cylinder is the space-like coordinate along a closed string and the co-
ordinate σ1 along the interval is Euclidean time. While it does not make any difference in
Euclidean signature, the interpretation is vastly different. Here we consider a process where
a closed string is emitted from the vacuum, propagates for some Euclidean time and is then
absorbed. The fact that these two viewpoints are associated with the same field theory
quantity is known as channel duality between open and closed strings, see fig. 2.11. This
observation has far-reaching consequences ; in particular, it played a crucial role in the birth
of the second string revolution, associated with the discovery of D-branes [9].

t
t

Figure 2.11: Channel duality: open string channel (left panel) vs. closed string channel (right
panel).

In the D-branes perspective channel duality has indeed the following meaning: the open
string viewpoint is interpreted as a one-loop vacuum amplitude for open strings whose end-
points are attached to a pair of D-branes, while the closed string viewpoint is interpreted as
the tree-level emission of closed strings by the first D-brane, followed by their absorption by
the second D-brane.

50
Bosonic strings: action and path integral

References
[1] L. D. Faddeev and V. N. Popov, “Feynman Diagrams for the Yang-Mills Field,” Phys.
Lett. 25B (1967) 29–30.

[2] C. Itzykson and J. B. Zuber, Quantum Field Theory. International Series In Pure and
Applied Physics. McGraw-Hill, New York, 1980.

[3] J. S. Schwinger, “On gauge invariance and vacuum polarization,” Phys. Rev. 82 (1951)
664–679.

[4] M. J. Strassler, “Field theory without Feynman diagrams: One loop effective actions,”
Nucl. Phys. B385 (1992) 145–184, hep-ph/9205205.

[5] Y. Nambu, “Strings, Monopoles and Gauge Fields,” Phys. Rev. D10 (1974) 4262.

[6] T. Goto, “Relativistic quantum mechanics of one-dimensional mechanical continuum and

subsidiary condition of dual resonance model,” Prog. Theor. Phys. 46 (1971) 1560–1569.

[7] M. Kalb and P. Ramond, “Classical direct interstring action,” Phys. Rev. D9 (1974)
2273–2284.

[8] A. M. Polyakov, “Quantum Geometry of Bosonic Strings,” Phys. Lett. 103B (1981) 207–
210.

[9] J. Polchinski, “Dirichlet Branes and Ramond-Ramond charges,” Phys. Rev. Lett. 75
(1995) 4724–4727, hep-th/9510017.

51
Chapter 3

Conformal field theory

52
Conformal symmetry

Two-dimensional conformally-invariant quantum field theories were introduced by Belavin,

Polyakov, and Zamolodchikov in 1984 [1]. These theories play also an essential role in sta-
tistical physics, for the description of critical phenomena in two dimensions. The book [2]
is probably the most comprehensive book on the subject, and reference [3] is geared towards
string applications.
After gauge-fixing of the diffeomorphisms and Weyl transformations to the flat worldsheet
metric ds2 = dwdw̄, the Euclidean Polyakov action
Z
1 ¯ ν
S= 0
d2 w Gµν [Xρ ] ∂Xµ ∂X (3.1)
4πα s
has an infinite-dimensional residual gauge symmetry that consists in reparametrizations

w 7→ f(w) , w̄ 7→ f̄(w̄) , (3.2)

where f is a holomorphic function, that leaves the metric invariant up to a conformal factor:
  
∂f(w) ∂f̄(w̄)
dwdw̄ 7→ dwdw̄ , (3.3)
∂w ∂w̄
and this conformal factor is absorbed by a Weyl transformation such that the reference metric
does not change.
As we have seen previously, the topology of the worldsheet restricts severely the transfor-
mations of this type that are allowed globally, i.e. the conformal Killing vectors of the surface.
On the sphere we have found that they corresponded to the Möbius group PSL(2, C), while
on the two-torus the only holomorphic functions periodic around both of the one-cycles are
constants. As a result, the transformations (3.2) are not properly speaking local symmetries
of string theory itself. In addition we will see shortly that the Weyl symmetry might not
hold in the quantum theory due to an anomaly.
At this stage we will be interested in a slightly different problem, the properties of two-
dimensional quantum field theories defined on the complex plane1 (w, w̄), or on another
manifold with a given fixed metric. We assume that the symmetries of the field theory in-
cludes the geometrical transformations defined by eqn. (3.2). Quantum field theories with
conformal invariance are called conformal field theories (CFTs for short). In two dimen-
sions this symmetry is often powerful enough to solve exactly the theory, without using any
perturbative expansion.
It should be stressed that one considers a field theory on a two-dimensional manifold
with a fixed metric, and the map w 7→ f(w), w̄ 7→ f̄(w̄) is really changing the geometry, in
particular the distances between points. In a two-dimensional field theory coupled to gravity,
this conformal map can be decomposed as a reparametrization (3.2) and a compensating
Weyl invariance.
The relation between two-dimensional CFTs and the Polyakov formulation of string theory
is as follows. When a conformal field theory in two-dimensions is coupled to two-dimensional
1
One often needs to consider the compactification of the complex plane by adding the point at infinity, C̄ =
C ∩ {∞}, which is related to the two-dimensional sphere (2.102) by a Weyl transformation ω = log(1 + ww̄).

53
Conformal symmetry

gravity, it gives rise – at the classical level at least – to a Weyl-invariant theory. Conversely,
after gauge-fixing of the diff.×Weyl local symmetry, the string theory action, which is now
formulated with respected to a fixed reference metric, is given in full generality by the action
of a two-dimensional conformal field theory.

3.1 The conformal group in diverse dimensions

The conformal group in two-dimensions is of infinite dimension, unlike the conformal group
in higher dimensions. To understand this we will study the conformal group in arbitrary
Euclidean space-time of dimension D.
Conformal transformations are defined as coordinate transformations that preserve the
metric up to a conformal factor, i.e.
∂xi ∂xj k `
xi 7→ x̃i (xj ) , δij dxi dxj 7→ δij dx̃ dx̃ = exp 2Ω(x̃k ) δij dx̃i dx̃j .


k `
(3.4)
∂x̃ ∂x̃
Here the metric is fixed, hence such transformation is really changing the geometry, in par-
ticular distances between points. The defining propery of conformal transformations is to
preserve angles between vectors of tangent space over any given point.
Let us consider a generic differentiable change of coordinates xi = xi + δxi (xj ). At linear
order in δx, this change generates a conformal transformation of the D-dimensional metric
γ provided that
∃α ∈ C 0 (RD ) , −(∂` δxk + ∂k δx` ) = α(xi )δk` . (3.5)
Taking the trace of tells us that −2∂i δxi = Dα, hence one has the equation

(∂` δxk + ∂k δx` ) − D2 (∂i δxi )δk` = 0 . (3.6)

Acting on this equation with ∂` finally gives

2
∂k (∂i δxi ) = 0 .


δxk + 1 − D
(3.7)

Let us assume a power series expansion:

δxi = ai + mi j xj + bi x2 + xi (cj xj ) + O(x3 ) (3.8)

and look for constraints on all the coefficients. While a is unconstrainted, eqn. (3.6) tells us
that
m(ij) − D1 δij Tr (m) = 0 , (3.9)
hence m splits into a trace part and an antisymmetric part. Equation (3.7) becomes at this
order
2Dbi + 2ci + 1 − D2 (2bi + (D + 1)ci ) = 0


(3.10)
which is solved for ci = −2bi .
One can check that there are no solutions to the problem for higher order terms in (3.8)
if D > 2. The space of solutions of these equations is then finite and corresponds to:

54
Conformal symmetry

• translations: δxi = ai constant

• rotations: δxi = rij xj with r in the vector representation of so(D), i.e. the antisym-
metric part of m in eqn. (3.8)

• dilatations: δxi = λxi with λ a non-vanishing constant, i.e. the trace part of m in
eqn. (3.8)

• special conformal transformations δxi = bi x2 − 2xi (bj xj ) .

These are actually a set of generators of the Lie algebra so(D + 1, 1), i.e. of the Lie Algebra
of the Lorentz group in D + 2 dimensions.
In two dimensions, the conformal transformations are given, in complex coordinates, by
w 7→ f(w), w̄ 7→ f̄(w̄) with f holomorphic, as we have already noticed. Naturally this
infinite-dimensional group contains as a subgroup the transformations existing in general
dimensions. Explicitly:

• translations: δw = a, δw̄ = ā

• rotations: δw = iθw, δw̄ = −iθw̄ with θ ∈ R.

• dilatations: δw = λw, δw̄ = λw̄ with λ ∈ R.

• special conformal transformations δw = −b̄ w2 , δw̄ = −b w̄2 , b ∈ C.

Following the general discussion this generates a Lie group isomorphic to SO(3, 1), i.e. the
Lorentz group in three dimensions, whose component connected to the identity is isomorphic
to the Möbius group PSL(2, C) that appeared already in section 2.3. It was shown there that
the Möbius group was the subgroup of two-dimensional conformal transformations that are
globally defined on the two-sphere, or equivalently on the compactified complex plane C̄.

3.2 Radial quantization

The Euclidean two-dimensional conformal field theory associated with the worldsheet of a
propagating string is naturally associated with a surface with the topology of a cylinder, i.e.
parametrized by two coordinates (σ1 , σ2 ) with σ1 ∼ σ1 + 2π, or equivalently the complex
coordinate w = σ1 + iσ2 with w ∼ w + 2π. The coordinate σ2 , which runs from −∞ to +∞,
is the Euclidean time, obtain through Wick rotation: σ2 = −iτ.
Even outside the string theory context, the natural starting point for canonical quantiza-
tion of conformal field theories is on the cylinder, in order to avoid infrared problems in the
case of an infinite space direction.
States of the quantum field theory are defined on a space-like slice, i.e. on a slice of
constant σ2 after Wick rotation to Euclidean space. In particular, an initial state |ini of the
theory is defined on a slice with σ2 → −∞.

55
Conformal symmetry

Invariance of the theory under conformal transformation allows to give a different repre-
sentation of the (Euclidean) time evolution of the QFT and of states of the theory. Let us
consider the conformal mapping from the cylinder to the complex plane:

w 7→ z = e−iw , w̄ 7→ z̄ = eiw̄ . (3.11)

w-plane

z-plane

radial time |z|

H
Q= j0dz

σ2

|ini

H
Q= j0dσ1

−∞

|ini

Figure 3.1: Conformal mapping from the cylinder to the complex plane.

In this description, σ2 → −∞ is mapped to the origin z = 0 and slices of constant σ2 are

mapped to circles around the origin, see fig. 3.1.
Time evolution correspond to dilatations or, said differently, the Hamiltonian operator
corresponds to the dilatation operator. Likewise, time-ordered correlation functions of the
QFT become radial-ordered correlation functions.

3.2.1 State - operator correspondence

In quantum field theories, states and operators are very different kind of objects.
On the one hand, in the path integral formalism, a state corresponds to a wave functional
Ψ1 [φ1 (~x), t1 ] defined on a whole space-like slice of space-time at a given time t1 , in the present
case on the circle parametrized by σ. Such a state will evolve to an other state Ψ2 [φ2 (~x), t2 ]
defined on a space-like slice at time t2 through:2
Z Z φ(~x,t2 )=φ2 (~x)
Ψ2 [φ2 (~x), t2 ] = Dφ1 Ψ1 [φ1 (~x, t1 )] Dφ eiS[φ] . (3.12)
φ(~x,t1 )=φ1 (~x)

2
A wave functional Ψ[φ(~x)] in QFT gives the probability amplitude associated with the whole field con-
figuration φ(~x) on the space-like slice. In quantum mechanics, a state is defined by a wave-function ψ(q)
R Rq(t )=q Rt
and the analoguous formula is ψ2 (q2 , t2 ) = dq1 Ψ1 (q1 , t1 ) q(t12)=q12 Dq exp ~i t12 dt 0 L.

56
Conformal symmetry

On the other hand, local operators in quantum field theories are defined as arbitrary local
expressions constructed from the elementary fields of the theory and their derivatives, i.e.
local functionals O[φi , ∂ν φi , · · · ](xµ ).3
The conformal mapping between the cylinder and the plane has some surprising conse-
quence. Let us consider an initial state |ini of the CFT on the cylinder, defined on the circle
parametrized by σ1 in the infinite past σ2 → −∞. Under the conformal mapping (3.11) it
is mapped to the origin of the plane z = z̄ = 0. This means that the initial state is mapped
to a local object at the origin, in other words a local operator, inserted at the origin. This is
called the state-operator correspondence:
A|0i
b ←→ lim OA (z, z̄) , (3.13)
z→0,z̄→0

which says that a state obtained by acting on the vacuum with some operator A ^ is equivalent
to a local operator OA (z, z̄) inserted at the origin of the complex plane. In this perspective
the vacuum |0i corresponds to the identity operator on the right-hand side.

3.2.2 Conserved charges

Let us consider a field theory with a conserved current, ∂µ jµ = 0, and a certain space-like
foliation of the ambient space-time. One can define a conserved charge
Z
Q = jµ dΣµ (3.14)

where dΣµ is the surface element over a space-like slice. For a conserved current on the
cylindrical worldsheet of the string, it gives
I
dσ1
Q = j2 . (3.15)
2π
Let us map the theory to the plane. A two-dimensional current has components Jz (z, z̄)
¯ z + ∂Jz̄ = 0.
and Jz̄ (z, z̄) in complex coordinates, Rand current conservation means that ∂J R
The integral around the cylinder dσ1 j2 becomes an integral over the polar angle, dθ Jr .
In complex coordinates, one obtains
I I 
1
Q= Jz dz − Jz̄ dz̄ , (3.16)
2iπ C C

where C denotes a contour encircling the local operator corresponding to the state for which
we compute the charge.
In many cases the two components of the current are separately conserved, hence J := Jz (z)
is a holomorphic function and J̃ := J̃z̄ (z) a anti-holomorphic one. This will be the case in
particular below for the charges associated respectively to holomorphic and anti-holomorphic
conformal transformations.
3
Note that, in a scalar field theory in 4d, a momentum state is given by |pi = a^† (p)|0i. However a
^† (p) =
R 3 ip·x ↔
−i d x e ^
∂ 0 φ(x) is not a local operator, being integrated over a whole space-like slice.

57
Conformal symmetry

3.3 Conformal invariance and Ward identities

The dynamics of quantum field theories with conformal invariance is severely constrained by
this symmetry, especially in two dimensions where this symmetry is infinite dimensional.

3.3.1 Stress-energy tensor

In the classical theory, the stress-energy tensor obeys strong constraints for conformal invari-
ance.
To derive the conservation laws associated with a general conformal field theory in D di-
mensions, one considers the usual trick of coupling the field theory to a dynamical background
metric γ. Then a coordinate change δσi (σk ) accompanied by the change of background metric
δγij = −∇i δσj − ∇j δσi is a diffeomorphism, under which the theory should be invariant.
The transformation of the action under the infinitesimal change of the background metric
is then opposite to the variation under the infinitesimal coordinate transformation δσi (σk )4 :
Z Z
1 D δ p 1 p
δS = − d x δγij − det γ L = dD x − det γ T ij δγij
2π δγij 4π
Z
1 p
=− dD x − det γ T ij ∇i δσj (3.17)
2π
where we have introduced the stress-energy tensor
2 δ p
T ij = − √ − det γ L . (3.18)
− det γ δγij
Translation invariance of the theory implies then as usual that the stress-energy tensor is
covariantly conserved:
∇i Tij = 0 . (3.19)
Let consider now invariance under scale transformations, which are special cases of con-
formal transformations, corresponding to constant rescalings of the metric, δγij = δλγij .
Equation (3.17) implies that the stress-energy tensor is traceless:

Tij γij = 0 . (3.20)

This is an essential property of conformal field theories.

In two-dimensions, these expressions take a particularly simple form in complex coordi-
nates with a flat background metric:
¯ z? + ∂Tz̄? = 0
∂T (3.21)

for the former, and

Tzz̄ = 0 (3.22)
4
We work momentarily in Lorentzian signature for sake of comparison with the computation done in the
General relativity course.

58
Conformal symmetry

for the later. This means that the stress-energy tensor has only two non-vanishing compo-
nents,
T := Tzz , Te := Tz̄z̄ , (3.23)
and that they are respectively holomorphic and anti-holomorphic functions:
¯ zz = 0 =⇒ T = T (z) ,
∂T (3.24a)
∂Tz̄z̄ = 0 =⇒ Te = Te(z̄) . (3.24b)

We will finally give the form of the Noether currents associated with conformal transfor-
mations. Let us consider the infinitesimal transformation

z 7→ z + ρ(z, z̄)ε(z) , z̄ 7→ z̄ + ρ̄(z, z̄)ε̄(z̄) , (3.25)

which reduces to a conformal transformation for constant ρ and ρ̄. According to eqn. (3.17)
the variation of the action of the theory is then
Z
1 
¯ ρ(z, z̄)ε(z) + Te(z̄)∂ ρ̄(z, z̄)ε̄(z̄)
d2 z T (z)∂



δS = −
2π
Z
1 ¯
d2 z T (z)ε(z)∂ρ(z,


=− z̄) + Te(z̄)ε̄(z̄)∂ρ̄(z, z̄)
2π
(3.26)

A very powerful point of view in two-dimensional Euclidean QFT is to consider w and

w̄ as independent variables. This means that we consider the analytic continuation of the
Euclidean space R2 to C2 , and the move from coordinates (σ1 , σ2 ) to (w, w̄) just as a change
of basis in C2 . Naturally at the end of the day one should enforce the reality condition
(z̄)? = z.
From this point of view one can consider purely holomorphic conformal transformations,
i.e. with ε̄ = 0. The associated Noether current is then

Jz = T (z)ε(z) , Jz̄ = 0 (3.27)

¯ = 0.
The non-zero component of the current, J := Jz is then holomorphic: ∂J
In the same way a purely anti-holomorphic conformal transformation, , i.e. with ε = 0,
gives the conserved current
J̃z = 0 , J̃z̄ = Te(z̄)ε̄(z̄) , (3.28)
whose non-zero component J̃ := J̃z̄ is anti-holomorphic: ∂J̃ = 0.
As in other quantum field theories, these classical equations lead to functional equations
in the quantum theory, known as Ward identities that we shall study now.

59
Conformal symmetry

3.3.2 Ward identities

Let us consider a generic time-ordered N-point correlation function of local operators in a
two-dimensional conformal field theory. By local operator we mean any operator that can
be written as a local expression in the fundamental fields φi and their derivatives. The
corresponding path integral is written as
Z
1 i
hO1 (z1 , z̄1 ) · · · ON (zN , z̄N )i = Dφi e−S[φ ] O1 [φi ](z1 , z̄1 ) · · · ON [φi ](zN , z̄N ) (3.29)
Z
We consider a holomorphic conformal transformation with support in a small disc d in the
complex plane, i.e. a transformation (3.25) with ρ non-vanishing only in the neighborhood
of some point (z0 , z̄0 ) and ρ̄ = 0.
There are two cases to consider: either there are no local operators in this neighborhood,
or there is (at least) one inserted at a point inside the disk, see fig. 3.2.

(z0, z̄0)

(z1, z̄1)

(z0, z̄0)

(z1, z̄1)
(z2, z̄2)

(z2, z̄2)

(z4, z̄4)
(z3, z̄3)
(z3, z̄3) (z4, z̄4)

Figure 3.2: Conformal transformation with support away from local operators (left panel) and
including one local operator (right panel).

Let us start with the former case. By construction the local operators are not affected by
the conformal transformation, hence the change in the path integral comes from the measure
Dφi and from the action S[φi ] only. We will make the assumption that the measure is
invariant so the only variation comes from the action.5 The path integral is then modified at
first order as
Z Z  Z 
i −S[φi ] i −S[φi ] 1 ¯ O1 · · · ON .
Dφ e O1 · · · ON 7→ Dφ e 1− 2
d z T (z)ε∂ρ (3.30)
2π
The path integral should actually be completely independent of the transformation which
should just be thought as some change of variables in the path integral. Therefore one gets
5
This assumption is very strong and turns out actually to be wrong in many cases; this is the path integral
view of QFT anomalies. We will come back to this important issue later on.

60
Conformal symmetry

the constraint, after integration by parts

Z Z  
i −S[φi ] ¯
Dφ e d z ρ(z, z̄) ∂ T (z)ε(z) O1 (z1 , z̄1 ) · · · ON (zN , z̄N ) = 0
2
(3.31)

Since this should hold for any choice of ρ, we are led to the condition that6
D   E
¯
∂ T (z)ε(z) O1 (z1 , z̄1 ) · · · ON (zN , z̄N ) = 0 . (3.32)

This is the quantum version of Noether theorem; naturally there is an analogous formula for
anti-holomorphic conformal transformations.
We move now to the latter case. Let assume that the operator O(z1 , z̄1 ) is inserted at a
point where ρ 6= 0, i.e. where the conformal transformation has support. Then on general
grounds the operator transforms
O(z1 , z̄1 ) 7→ O(z1 , z̄1 ) + δO(z1 , z̄1 ) . (3.33)
We insert this transformation in the path integral and, using similar arguments as before, we
get at first order the relation
Z
1 D  
¯ T (z)ε(z) O1 (z1 , z̄1 ) · · · ON (zN , z̄N )
E
− d2 z ρ(z, z̄) ∂
2π
D E
= δO1 (z1 , z̄1 ) O2 (z2 , z̄2 ) · · · ON (zN , z̄N ) . (3.34)

The function ρ being arbitrary, we can take it to be the indicator function of the disk d
around (z0 , z̄0 ). The left-hand side can then be simplified using Stokes’ theorem:
Z I
z z̄


∂z J + ∂z̄ J = −i (Jz dz − Jz̄ dz̄) . (3.35)
d ∂d

In the present case Jz̄ = 0 as we found before so eqn. (3.34) gives

I
i D E D E
dz T (z)ε(z)O1 (z1 , z̄1 )O2 · · · ON = δO1 (z1 , z̄1 ) O2 · · · ON , (z1 , z̄1 ) ∈ d . (3.36)
2π ∂d
This teaches us a very important result: the change of an operator δO under a conformal
transformation is given by the residue of its product with the stress-energy tensor, namely
I
i
δO1 (z1 , z̄1 ) = dz T (z)ε(z)O1 (z1 , z̄1 ) = −Resz→z1 (T (z)ε(z) O1 (z1 , z̄1 )) (3.37)
2π ∂d
The holomorphic function ε being arbitrary, this means that the operator product between
T (z) and O(z1 , z̄1 ) has a first order pole when they approach to each other:
Res (T (z)O1 (z1 , z̄1 ))
T (z) O1 (z1 , z̄1 ) = · · · + + ··· . (3.38)
z − z1
6
Naturally the same holds for the partition function Z in the denominator of (3.29), i.e. the zero-point
function.

61
Conformal symmetry

There is naturally a similar story for the anti-holomorphic conformal transformations, which
gives in particular:
 
Res Te(z̄)O1 (z1 , z̄1 )
Te(z̄) O1 (z1 , z̄1 ) = · · · + + ··· . (3.39)
z̄ − z̄1
In a quantum field theory, product of operators make sense when they are time-ordered. In
the context of radial quantization of a two-dimensional CFT, time-ordering takes the form
of radial ordering, namely:

0 0 O1 (z 0 , z̄ 0 )O2 (z, z̄) |z 0 | > |z|
R (O1 (z , z̄ )O2 (z, z̄)) = , (3.40)
(−1)F O2 (z, z̄)O1 (z 0 , z̄ 0 ) |z| > |z 0 |

where (−)F = 1 (resp. (−)F = −1) for bosonic (resp. fermionic) operators. In order to
unclutter calculations, radial-ordering will always be implicit in operator equations like (3.38).
Singularities when operators approach each other such as (3.38) are a generic feature
of quantum field theories. They have been studied more systematically under the name
of operator product expansions. The operator product expansion exists in all quantum field
theories, and provides the behavior of the theory when two operators O1 and O2 come close to
each other. The basic idea is that, when the separation between them become infinitesimal,
one can expand the product into a sum of local operators. While this is usually an asymptotic
expansion only, one can show that in the case of conformal field theories the series converges,
with the radius of convergence given by the distance to the nearest operator On6=1,2 .

3.4 Primary operators

The variation of a generic local operator O(z, z̄) under a holomorphic conformal transforma-
tion ε(z) is contained in eqn (3.37). We can get some insight by looking at specific simple
transformations.
Let us look first at a holomorphic translation defined as z 7→ z + a, z̄ 7→ z̄. It acts on any
local operator O as

O(z, z̄) 7→ O(z − a, z̄) = O(z, z̄) − a∂z O(z, z̄) + · · · (3.41)

By identifying both sides of (3.37) one learns that

∂z O(z, z̄) = Resz 0 →z (T (z 0 )O(z, z̄)) (3.42)

or in other words, after doing the same for anti-holomorphic translations:

z 0 →z ∂O(z, z̄)
T (z 0 ) O(z, z̄) ' · · · + + ··· (3.43a)
z0 − z
z 0 →z ¯
∂O(z, z̄)
Te(z̄ 0 ) O(z, z̄) ' · · · + 0 + ··· (3.43b)
z̄ − z̄

62
Conformal symmetry

We look now at infinitesimal scaling transformations, which are by definition the (non-
holomorphic) transformations z 7→ (1 + δλ)z, z̄ 7→ (1 + δλ)z̄, with real δλ. We consider
operators that are eigenstates of the dilatation operator; the corresponding eigenvalue is
called the scaling dimension ∆ of the operator:7

O 7→ λ−∆ O (3.44)

The transformation of an operator O∆ of scaling dimension ∆ under an infinitesimal scale

transformation is given by:
¯ ∆ .


δO∆ = −δλ ∆O∆ (z, z̄) + z∂O∆ + z̄∂O (3.45)

We consider finally infinitesimal rotations in the complex plane, given by z 7→ (1 + iδθ)z,

z̄ 7→ (1−iδθ)z̄ with real δθ. Consider again an operator which is an eigenstate of the rotation
operator; its eigenvalue is called the spin s of the operator. The transformation law is given
by:
¯ s .


δOs = −iδθ sOs (z, z̄) + z∂Os − z̄∂O (3.46)
In order to use the holomorphic/antiholomorphic splitting of conformal transformations
that we have used throughout, it is convenient to combine scalings and rotations into the
complex transformations

δz = (δλ + iδθ)z =: δα z , δz̄ = (δλ − iδθ)z̄ =: δᾱ z . (3.47)

Consider common eigenstates of the dilatation and rotation operators. The eigenvalues are
called the conformal weights (h, h̄) of the state, and are related to the scaling dimension and
spin through
∆ = h + h̄ , s = h − h̄ . (3.48)
We obtain then from the transformations (3.45,3.47) the infinititesimal transformations:

δα Oh,h̄ = −δα hOh,h̄ (z, z̄) + z∂Oh,h̄ ,

(3.49a)
¯ h,h̄ .
δᾱ Oh,h̄ = −δᾱ h̄Oh,h̄ (z, z̄) + z̄∂O


(3.49b)

Under a finite scaling transformation and rotation, the operator of conformal weights (h, h̄)
transforms as
−h
−h̄
Oh,h̄ 7→ λ−∆ e−isθ Oh,h̄ = λeiθ λe−iθ Oh,h̄ . (3.50)
The residue formula (3.37) will provide then the relation between the (anti)holomorphic
scale transformation and the operator product with the stress-energy tensor. One has first

hOh,h̄ (z, z̄) + z∂Oh,h̄ (z, z̄) = Resz 0 →z (z 0 T (z 0 ) O(z, z̄))

= Resz 0 →z ((z 0 − z)T (z 0 ) O(z, z̄)) + zResz 0 →z (T (z) O(z, z̄)) (3.51)
7
At the classical level ∆ is the same as the dimension in units of inverse length Rcoming from dimensional
1
analysis. For instance, a free massless scalar field in dimension D has an action 4π dD x (∂φ)2 , hence φ has
(2−D)/2
dimension (length) , in other words scaling dimension ∆ = (D − 2)/2.

63
Conformal symmetry

which gives, using also eqn. (3.42)

hOh,h̄ (z, z̄) = Resz 0 →z ((z 0 − z)T (z 0 ) O(z, z̄)) (3.52)
In summary, we have learned that the operator product expansion of one of the components
of the stress-energy tensor with an operator of conformal weight (h, h̄) contains the terms
z 0 →z h 1
T (z 0 )Oh,h̄ (z, z̄) ' · · · + 0 Oh,h̄ (z, z̄) + 0 ∂O (z, z̄) + · · · (3.53a)
(z − z) 2 (z − z) h,h̄
z 0 →z h̄ 1 ¯
Te(z̄ 0 )Oh,h̄ (z, z̄) ' · · · + 0 Oh, h̄ (z, z̄) + ∂O (z, z̄) + · · · (3.53b)
(z̄ − z̄)2 (z̄ 0 − z̄) h,h̄
The operator product with a generic operator of weights (h, h̄) contains in principle more
singular terms in the Laurent series expansion in powers of (z − w).

Conformal primary operators

A distinguished class of local operators in a two-dimensional conformal field theory are op-
erators for which the terms written in eqns. (3.53) exhaust all the possible singular terms in
the expansion, namely
z 0 →z h 1
T (z 0 )Oh,h̄ (z, z̄) ' Oh,h̄ (z, z̄) + 0 ∂O (z, z̄) + regular (3.54a)
0
(z − z) 2 (z − z) h,h̄
z 0 →z h̄ 1 ¯
Te(z̄ 0 )Oh,h̄ (z, z̄) ' Oh, h̄ (z, z̄) + ∂O (z, z̄) + regular (3.54b)
(z̄ 0 − z̄)2 (z̄ 0 − z̄) h,h̄
These are called primary operators, or primaries for short. Given that the regular terms
in the expansions do not contain essential information, we will often remove them from the
expressions of the operator product expansions.
Primary operators are interesting because they have particularly simple transformation
laws under a generic holomorphic conformal transformation δz = ε(z). Using once again the
residue formula (3.37) and the OPE (3.54a) for primary operators one finds that
  
h 1
δε O = −Resz 0 →z ε(z ) 0
O (z, z̄) + 0 ∂O (z, z̄) + regular . (3.55)
(z 0 − z)2 h,h̄ (z − z) h,h̄
The infinitesimal conformal transformation ε(z) is by definition holomorphic in the neigh-
borhood of z and can be Taylor-expanded there:
ε(z 0 ) = ε(z) + (z 0 − z)∂ε(z) + O (z 0 − z)2


(3.56)
Therefore transformation of a primary operator under an infinitesimal generic conformal
transformation is given by
δε O(z, z̄) = −ε(z)∂Oh,h̄ (z, z̄) − h ∂ε(z) Oh,h̄ (z, z̄) .


(3.57)
This transformation law can be generalized to a finite – rather than infinitesimal – trans-
formation, namely z 7→ z̃ = f(z), as well as to its anti-holomorphic counterpart z̄ 7→ z̄˜ = f̄(z̄).
One finds then:  −h  −h̄
∂f ∂f̄
O(z, z̄) 7→ Õ z̃, z̄˜ = O(z, z̄) .


(3.58)
∂z ∂z̄

64
Conformal symmetry

Correlation functions of primary operators

Conformal invariance severely constraints the form of correlation functions between primary
operators. Let us start with the simplest case, the two-point function:

hO1 (z1 , z̄1 )O2 (z2 , z̄2 )i , (3.59)

where O1 (resp. O2 ) is a conformal primary of weights (h1 , h̄1 ) (resp. (h2 , h̄2 )). Under a
conformal transformation z 7→ y(z), z̄ 7→ ȳ(z̄) one should have
Z
1 i
hO1 (y1 , ȳ1 )O2 (y2 , ȳ2 )i = Dφi e−S[φ ] O1 (y1 , ȳ1 )O2 [y2 , ȳ2 ]
Z
Z
1 i
= Dφ̃i e−S[φ̃ ] Õ1 (y1 , ȳ1 )Õ2 (y2 , ȳ2 )
Z
 −h1  −h̄1  −h2  −h̄2
∂y ∂ȳ ∂y ∂ȳ
= (z1 ) (z̄1 ) (z2 ) (z̄2 ) hO1 (z1 , z̄1 )O2 (z2 , z̄2 )i (3.60)
∂z ∂z̄ ∂z ∂z̄

Invariance under holomorphic and anti-holomorphic translations implies first that

hO1 (z1 , z̄1 )O2 (z2 , z̄2 )i = f(z1 − z2 , z̄1 − z¯2 ) . (3.61)

Next we investigate invariance under holomorphic and anti-holomorphic scale transforma-

tions. It leads to the functional identities

f λ × (z1 − z2 ), z̄1 − z¯2 = λ−h1 −h2 f z1 − z2 , z̄1 − z¯2 ,

(3.62a)
−h̄1 −h̄2



f z1 − z2 , λ̄ × (z̄1 − z¯2 ) = λ̄ f z1 − z2 , z̄1 − z¯2 . (3.62b)

hence f is homogeneous of degree −(h1 + h2 ) (resp. of degree −(h̄1 + h̄2 )) in z1 − z2 (resp.

in z̄1 − z̄2 ). In other words,
C12
f(z1 − z2 , z̄1 − z¯2 ) = h +h
, (3.63)
(z1 − z2 ) 1 2 (z̄
1 − z̄2 )h̄1 +h̄2

where C12 is some constant. We now impose invariance under the transformation z 7→ −1/z,
which is simpler to handle than the special conformal transformation z 7→ z/(1 − b̄ z). One
gets
1
−h1
−h2 1
h1 +h2
= 1/z21 1/z22 =⇒ (z1 z2 )h1 +h2 = z2h 1 2h2
1 z2 ,
(−1/z1 + 1/z2 ) (z1 − z2 )h1 +h2
(3.64)
with a similar equation for the anti-holomorphic transformations. Therefore the two-point
function can be non-zero only if h1 = h2 and h̄1 = h̄2 :

C12
hφ1 (z1 , z̄1 )φ2 (z2 , z̄2 )i = h +h
δh1 ,h2 δh̄1 ,h̄2 . (3.65)
(z1 − z2 ) 1 2 (z̄
1 − z̄2 )h̄1 +h̄2

65
Conformal symmetry

Up to a constant, the two-point function is then completely fixed by invariance under the
global conformal group PSL(2, C).8 The set of constants Cij associated with the two-points
functions of quasi-primary operators of the same conformal weights define a matrix that can
mapped to the identity by choosing an appropriate basis in field space.
There exists a similar story regarding the three-point function of primary operators. In-
variance under PSL(2, C) reduces the three-point function computation to a single unknown
coefficient:

1
hφ1 (z1 , z̄1 )φ2 (z2 , z̄2 )φ3 (z3 , z̄3 )i = C123
(z1 − z2 )h1 +h2 −h3 (z h +h −h
2 − z3 ) 2 3 1 (z1 − z3 ) 3 1 2
h +h −h

1
× . (3.66)
(z̄1 − z̄2 ) 1 2 3 (z̄2 − z̄3 )h̄2 +h̄3 −h̄1 (z̄1 − z̄3 )h̄3 +h̄1 −h̄2
h̄ + h̄ − h̄

3.5 The Virasoro Algebra

The operator product expansions (3.53) are valid for all local operators that are eigenstates
of the dilatation and rotation operators. By dimensional analysis, the stress-energy tensor
Tij given by the definition (3.17) has scaling dimension ∆ = 2 in two dimensions. Under a
rotation z 7→ z 0 = eiθ z its non-vanishing components transform as Tzz 7→ Tz 0 z 0 = e−2iθ Tzz and
Tz̄z̄ 7→ Tz̄ 0 z̄ 0 = e2iθ Tz̄z̄ .
This analysis shows that T = Tzz is an operator of conformal weights (h, h̄) = (2, 0) while
T = Tz̄z̄ is an operator of conformal weights (h, h̄) = (0, 2), see eq. (3.48). However nothing
e
indicates that these are primary operators. Let us consider then the OPEs
z 0 →z 2 1
T (z 0 )T (z) ' · · · + T (z) + 0 ∂T (z) + reg. (3.67a)
(z 0
− z) 2 (z − z)
z 0 →z 2 1 ¯ Te(z̄) + reg.
Te(z̄ 0 )Te(z̄) ' · · · + 0 2
Te(z̄) + 0 ∂ (3.67b)
(z̄ − z̄) (z̄ − z̄)

3.5.1 The central charge

The missing information in the OPEs (3.67) are the possible terms more singular than 1/(z −
w)2 in the expansion. The only universal operators that should appear in any conformal field
theory are the components of the stress energy tensor and the identity operator which has
naturally conformal dimensions (h, h̄) = (0, 0). On general grounds, one should allow terms
proportional to the identity on the right-hand side of eqns. (3.17). By dimensional analysis,
T being of dimension (2,0), this term should come with the power (z 0 − z)−4 in the operator
8
Actually all is needed is that the operators transform as (3.58) under the action of PSL(2, C), which is
a weaker statement than asking that this equation holds true for any holomorphic function. Such operators
are called quasi-primary operators.

66
Conformal symmetry

product expansion:9
z 0 →z c 2 1
T (z 0 )T (z) ' + 0 T (z) + 0 ∂T (z) + reg. (3.68a)
2(z 0 − z) 4 (z − z) 2 (z − z)
z 0 →z c̄ 2 1 ¯ Te(z̄) + reg.
Te(z̄ 0 )Te(z̄) ' 0 4
+ 0 2
Te(z, z̄) + 0 ∂ (3.68b)
2(z̄ − z̄) (z̄ − z̄) (z̄ − z̄)
These OPEs depend on two C-numbers c and c̄ that are called the central charges of the
conformal field theory, and encode the failure of the components of the stress-energy tensor
to be conformal primaries. They play a central role in the study of conformal field theories,
characterizing in particular the number of degrees of freedom.
From the operator product expansion (3.68) one deduces the transformation of T under
an arbitrary infinitesimal holomorphic conformal transformation:
  
0 c 2 1
δε T (z) = −Resz 0 →z ε(z ) + T (z) + 0 ∂T (z) + reg.
2(z 0 − z)4 (z 0 − z)2 (z − z)
c
= − ε 000 (z) − 2ε 0 (z)T (z) − ε(z)∂T (z) , (3.69)
12
and similarly
c̄ ¯ Te(z̄) ,
δε̄ Te(z̄) = − ε̄ 000 (z̄) − 2ε̄ 0 (z)T (z) − ε(z)∂ (3.70)
12
Notice that for infinitesimal dilatations, rotations and special conformal transformations the
terms proportional to the central charges vanish.
For a finite transformation z 7→ b z = f(z) and z̄ 7→ b̄ z = f̄(z̄) the components of the
stress-tensor transform as
 −2 
∂f(z) c 
T (z) 7→ T (b
b z) = T (z) + {f(z), z} , (3.71a)
∂z 12
 −2  
∂f̄(z̄) c̄
T (z̄) 7→ T (z) = T (z̄) + {f̄(z̄), z̄} , (3.71b)
e b
e b̄ e
∂z̄ 12
where one defines the Schwarzian derivative
2f 000 (z)f 0 (z) − 3(f 00 (z))2
{f(z), z} = , (3.72)
2(f 0 (z))2
which is compatible with the composition of successive conformal transformations.

3.5.2 The Virasoro Algebra

The information contained in the operator product expansions (3.68) can be recast in a
different way that will turn out to be very useful, as it will allow to study the properties of
the CFT using the tools of representation theory.
9
More singular terms would come with operators of negative conformal weights; as we will see later this
is forbidden in unitary conformal field theories.

67
Conformal symmetry

To start, the components of the stress tensor T and Te are respectively holomorphic and
anti-holomorphic functions on the complex plane, hence admit a Laurent series expansion:
X Ln X L en
T= , Te = . (3.73)
n∈Z
zn+2 n∈Z
z̄n+2

The coefficients of the expansion can be found by a contour integral

I
dz n+1
Ln = z T (z) (3.74a)
C 2iπ
I
dz̄ n+1 e
Ln = −
e z̄ T (z̄) (3.74b)
C 2iπ

where C is a contour encircling the origin counter-clockwise. The mode Ln corresponds to

the conserved charge associated with the conformal transformation δz = zn+1 following the
discussion in subsection 3.2.2. In particular, L0 + L
e0 is the dilatation generator, L0 − L
e0 the
rotation generator, while L−1 and L
e−1 generate holomorphic and anti-holomorphic translations
respectively.
In the quantum conformal field theory on the plane, we will consider the commutator of
two Laurent modes of the stress-energy tensor. In other words, for any state |φi in the theory
we consider the quantity

[Lm , Ln ]|φi = Lm Ln |φi − Ln Lm |φi . (3.75)

The first term corresponds to applying first Ln to |φi and then Lm , while the second one
corresponds to applying first Lm to |φi and then Ln .
Using the state-operator correspondence, the state |φi is mapped to an local operator
O(z, z̄) that we can put at the origin. We have learned also in section 3.2 that the charge of a
state w.r.t. a holomorphic current is computed by a contour integral around the corresponding
local operator, see eqn. (3.16).
Let us consider the circular contours C of radius R and C 0 of radius R 0 > R, both around
the origin. We have
I I
dz1 dz2 m+1 n+1
Lm Ln O(0, 0) = z z2 T (z1 ) T (z2 )O(0, 0) (3.76a)
0 2iπ 2iπ 1
IC IC
dz2 dz1 m+1 n+1
Ln Lm O(0, 0) = z1 z2 T (z1 ) T (z2 )O(0, 0) (3.76b)
C 0 2iπ C 2iπ

were the operator products are implicitely radial-ordered, as explained above. In the following
we will remove the operator O which plays no role in this computation; it is understood that
the operators are applied to any local operator in the CFT inserted at the origin.
The only difference between the two equations (3.76) is that in (3.76a) the contour of
integration over z1 is around the contour of integration over z2 , while in (3.76b) it is just the
opposite. To compute the commutator, the trick is to consider first the integration over z1
for fixed z2 , see fig. 3.3. Going from expression (3.76a) to expression (3.76b) amounts to pass

68
Conformal symmetry

z2
z2
z2 C
C

C0
C0

Figure 3.3: Commutator of Virasoro generators.

the contour of integration over z1 (solid line) through the locus of the contour of integration
over z2 (dashed line).
As the operator product T (z1 )T (z2 ) has pole, one picks a residue when the contour of
integration over z1 crosses the position z2 where T is inserted. One has finally to integrate
this residue over z2 :
I
dz2 n+1  
[Lm , Ln ] = z2 Resz1 →z2 z1m+1 T (z1 )T (z2 ) . (3.77)
2iπ
We now compute the residue using the operator product expansion (3.68a):

 2

Resz1 →z2 z2m+1 + (m + 1)(z1 − z2 )z2m + m(m+1) 2
(z1 − z2 )2 z2m−1 + m(m6 −1) z2m−2 (z1 − z2 )3 + · · ·
 
c 2 1
× 2(z1 −z 2)
4 + (z1 −z2 )2
T (z 2 ) + (z1 −z2 )
∂T (z 2 ) + reg.
m(m2 −1)c m−2
= 12
z2 + 2(m + 1)z2m T (z2 ) + z2m+1 ∂T (z2 ) . (3.78)

So we end up with the following integral

I
dz2 n+1  m(m2 −1)c m−2 m m+1

[Lm , Ln ] = z z2 + 2(m + 1)z2 T (z2 ) + z2 ∂T (z2 )
2iπ 2 12
I
dz2  m(m2 −1)c n+m−1 m+n+1 n+m+1

= 12
z 2 + 2(m + 1)z 2 T (z 2 ) − (n + m + 2)z2 T (z 2 )
2iπ
I
dz2  m(m2 −1)c n+m−1 m+n+1

= 12
z 2 + (m − n)z 2 T (z 2 ) , (3.79)
2iπ
where we have done an integration by parts of the last term in the second step. Now we
simply have to express the right-hand side of the final expression in terms of the Laurent
coefficients using equation (3.74a) and get the well-known Virasoro algebra:

c
[Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n,0 (3.80)
12

If we did a similar computation, using the equation (3.74b) at the last step, we would find
in the same way
[L
em , L em+n + c̄ m(m2 − 1)δm+n,0 .
en ] = (m − n)L (3.81)
12
69
Conformal symmetry

The Virasoro algebra (3.80) is very similar two an ordinary Lie algebra, except that it has
an infinite number of generators, {Ln , n ∈ Z}. Another characteristic feature is the presence
of the constant term 12c̄ m(m2 − 1)δm+n,0 , which commutes with all the generators; such a
term is called a central extension of the algebra.
Finally a finite sub-algebra of the Virasoro algebra is obtained from the generators
{L−1 , L0 , L1 }:
[L0 , L±1 ] = ∓L±1 , [L−1 , L1 ] = 2L0 , (3.82)
which is isomorphic to sl(2, C); this is nothing but the Lie algebra of the Möbius group,
the group of globally defined conformal transformations on the sphere C̄, i.e. of projective
transformations z 7→ az+b
cz+d
, a, b, c, d ∈ C, already discussed in subsection 2.3.3.

3.5.3 Conformally invariant vacuum and Casimir energy

When a conformal field theory theory is formulated on the cylinder, it is natural to expand
the components of the stress energy tensor in terms of its Fourier modes:
X
Tww = − Tn einw (3.83a)
n∈Z
X
Tw̄w̄ = − Ten e−inw̄ (3.83b)
n∈Z

with
Z
dσ1 −inσ1
Tn = − e Tww (σ1 , 0) , (3.84a)
2π
Z
dσ1 inσ1
Tn = −
e e Tw̄w̄ (σ1 , 0) . (3.84b)
2π
The conformal mapping (3.11) from the cylinder to the complex plane gives, using equa-
tions (3.71) and (3.72):
c
Tww (w) = −z2 Tzz (z) + (3.85a)
24
2 c̄
Tw̄w̄ (w̄) = −z̄ Tz̄z̄ (z̄) + (3.85b)
24
Hence the expansion (3.73) on the plane in terms of Laurent coefficients and the expan-
sion (3.83) on the cylinder in terms of Fourier modes are related through
c en − c̄ δn,0 .
Tn = Ln − δn,0 , Ten = L (3.86)
24 24
In particular, the Hamiltonian of the conformal field theory on the cylinder is by definition
the conserved charge associated with time translations:
Z Z
dσ1 dσ1

e0 − c + c̄ .
H= Tσ 2 σ 2 = − Tww + Tw̄w̄ = L0 + L (3.87)
2π 2π 24

70
Conformal symmetry

Conformally invariant vacuum

In radial quantization, the natural vacuum state of the conformal field theory on the complex
plane, |0i, is defined by inserting the identity operator at the origin z = 0.
According to the definition (3.73) of the Virasoro generators, in order for the components
of the stress-tensor to be non-singular at the origin, one should have

Ln |0i = 0 , en |0i = 0 ,
L ∀n > −1 . (3.88)

In particular, the vacuum is invariant under the action of {L−1 , L0 , L1 } and {L

e−1 , L e1 }, i.e.
e0 , L
under global conformal transformations. For this reason, it is called the PSL(2, C)-invariant
vacuum.
If we consider a two-dimensional CFT in the PSL(2, C)-invariant vacuum on the plane,
equation (3.87) indicates that the corresponding energy of the ground state on the cylinder
is given by
c + c̄
Ecyl = − , (3.89)
24
which is interpreted as the Casimir energy of the QFT on the cylinder; notice that no regu-
larization of high-energy divergences was needed to obtain this result. This remark provides
another interpretation of the central charges of a conformal field theory.

3.5.4 Conformal primaries revisited

Primary operators were defined by the operator product expansion (3.54) with the stress-
energy tensor. Let us consider some primary operator Oh,h̄ . Plugging in the decomposi-
tions (3.73) in terms of Laurent modes, one finds that the corresponding state |Oi is annihi-
lated by all positive modes of the Virasoro generators:

∀n > 0 , Ln |h, h̄i = 0 , en |h, h̄i = 0 .

L (3.90)

Hence primary operators play the same role w.r.t. the Virasoro algebra as the highest
weight vectors in representation theory of Lie algebras. For this reason they are also referred
to as highest weight states of the Virasoro algebra.
The other states in the Virasoro representation associated with the highest weight state
|h, h̄i are obtained by applying repeatedly the the negative Virasoro modes {L−n , n > 0}
(there is a similar story for anti-holomorphic generator). This generates what is known as a
Verma module, that may or may not be an irreducible representation of the Virasoro algebra
depending on the values of h and the central charge c.
The conformal dimension of these descendant states, obtained from the highest weight
state using lowering operators L−n , is easily computed:

L0 L−n |h, h̄i = nL−n |h, h̄i + L−n L0 |h, h̄i = (h + n)L−n |h, h̄i, n > 0. (3.91)

I would certainly agree with the reader that would tell me that the primary states should be
called lowest dimension states rather than highest weight states!

71
Conformal symmetry

3.5.5 Unitarity constraints

One important constraint on sensible quantum field theories is unitarity, the conservation
of probabilities over time. The time evolution on the cylinder is given by the Hamiltonian
density (with Minkowskian signature on the worldsheet):
X 1 1

H = −(Tww + Tw̄w̄ ) = Tn ein(σ +τ) + Ten e−in(σ −τ)
n∈Z
X  c + c̄
in(σ1 +τ) en e−in(σ 1 −τ)
= Ln e +L − . (3.92)
n∈Z
24

Hence unitarity gives the constraints

 †
†
(Ln ) = L−n , L
en =L
e−n . (3.93)

These unitarity relations have two immediate consequences:

• consider a primary state |h, h̄i of conformal dimensions (h, h̄) in a unitary CFT.

||L−1 |h, h̄i||2 = hh, h̄|[L1 , L−1 ]|h, h̄i = 2hh, h̄|L0 |h, h̄i , (3.94)

hence in a unitary QFT the conformal dimensions (h, h̄) of primary operators – and,
consequently, of all descendant states – should be non-negative. The unique state
with h = h̄ = 0 in a unitary CFT is the PSL(2, C)-invariant vacuum since it satisfies
L−1 |0, 0i = L ¯
e−1 |0, 0i = 0 which means in operator language ∂O(0, 0) = ∂O(0, 0) = 0.

• ∀n > 1, one has

c
||L−n |0i||2 = h0|[Ln , L−n ]|0i = n(n2 − 1) , (3.95)
12
en |h, h̄i. Therefore a unitary CFT has positive central charges: c > 0
and similarly for L
and c̄ > 0. The only unitary CFT with c = c̄ = 0 is trivial and contains just the
identity operator.

3.6 The Weyl anomaly

The failure of the components (T, Te) of the stress-energy tensor to be primary operators
w.r.t. generic two-dimensional conformal transformations is a just a characteristic feature of
conformal field theories, more precisely as we will see shortly of CFTs defined on a curved
(two-dimensional) manifold. In the string theory context however, it leads to a potential
disaster.
String theory was defined as a two-dimensional theory of gravity coupled to a set of
scalar matter fields {xµ (σi )}. The theory has an enormous redundancy, since its gauge group
corresponds to two-dimensional diffeomorphisms and Weyl transformations. After a suitable
gauge-fixing procedure, we have obtained a path integral defined over a gauge slice, i.e. with

72
Conformal symmetry

a reference metric depending only on a handful of parameters. All this beautiful construction
falls apart if the classical gauge symmetry of the theory is not satisfied at the quantum level.
The potential clash between gauge symmetry and quantum effects is associated with Weyl
transformations. Heuristically it is not difficult to understand why. Computations in the
two-dimensional quantum field theory we are dealing with have divergences, as in any other
QFT, that should be regularized. If one uses dimensional regularization, the invariance under
Weyl rescalings is lost, since it depends crucially on being in dimension two, see eqn. (2.84).
Using alternatively a Pauli-Villars regularization, one would introduce a scale in the theory
breaking explicitly scale invariance. As any sensible regulator breaks the Weyl symmetry, one
may wonder if at the end of the computation, which is not covariant w.r.t. the Weyl gauge
transformations, the invariance would be miraculously restored. It turns out not to be the
case and the Weyl symmetry has potentially an anomaly of the gauge symmetry signaling
the inconsistency of the theory.
To characterize this anomaly, one considers that the background geometry is fixed to some
reference metric and consider whether the classical conservation laws associated with the
symmetries of the theory are truly independent of the choice of reference metric. Classically,
Weyl-invariance implies that the stress-energy tensor is traceless, see eqn. (3.20). Because of
general covariance of the theory, a quantum violation of this condition is very restricted. In
the quantum field theory context, one would like to see whether the operator T ii inserted in
an arbitrary correlation function gives zero. The more general parametrization is:
hT ii · · · i = aR[^
γ]h· · · i , (3.96)
where R[^γ] is the Ricci scalar computed for the reference worldsheet metric γ
^ , given that the
right-hand side of the equation should be a local expression, and of scaling dimension two;
the parameter a is dimensionless.
Since the diffeomorphism symmetry is not anomalous, one can work in the conformal
gauge, ds2 = exp(2ω)dwdw̄, in which case, the anomaly equation becomes, using eqn. (2.85b),
a
2γww̄ hTww̄ · · · i = −2ae−2ω ∇2 ωh· · · i =⇒ hTww̄ · · · i = − ∇2 ωh· · · i . (3.97)
2
The conservation of the stress-energy tensor gives ∇w̄ Tww̄ + ∇w Tww = 0. Hence taking the
derivative of (3.97) with ∇w̄ one reaches the equation:
a w̄ 2
h∇w Tww · · · i = ∇ ∇ ωh· · · i . (3.98)
2
Let us now consider an infinitesimal Weyl rescaling around the flat metric, i.e. ω = 1 +
δω(w, w̄), and compare the variations of the left- and right-hand sides of eqn. (3.98).
In the previous section we have considered the transformation of the components of the
stress-tensor of a two-dimensional CFT under an infinitesimal conformal transformation, i.e.
an infinitesimal transformation w 7→ w + ε(w):
c 3
− δT (w) = ∂ ε(w) + 2T (w)∂ε(w) + ε(w)∂T (w) , (3.99)
|12 {z } | {z }
Weyl tensor

73
Conformal symmetry

where the last two terms correspond to the standard tensorial transformation under a change
of coordinates, T ij (xk ) 7→ T̃ ij (x̃k ) = (∂x̃i /∂xm )(∂x̃j /∂xn )T mn (xk ); the first term corresponds
to the anomalous change of the operator T (w) under the compensating Weyl transformation
that brings back the metric to the original one.
Under the infinitesimal compensating Weyl transformation δω of the background metric,
namely δω = 12 (∂ε+∂ε̄)¯ (see eq. 2.100), the stress-energy tensor operator transforms therefore
as:
c
δδω T = − ∂2w δω . (3.100)
6
Plugging into eqn. (3.98) one finds, using ∇w = 2∂w̄ and ∇2 = 4∂w ∂w̄ ,
c c
− ∂w̄ ∂2w δω = 4a∂2w ∂w̄ δω =⇒ a = − . (3.101)
3 12
The conclusion of this computation is that
c
hT ii · · · i = − R[^γ]h· · · i . (3.102)
12
This result is not problematic if one considers a conformal field theory outside of the string
theory context, in the same way as the axial anomaly of quantum electrodynamics does not
imply that the latter is an inconsistent quantum field theory.
However if the two-dimensional conformal field theory at hand corresponds to the world-
sheet theory of a string the anomaly (3.102) indicates that the theory is inconsistent unless
c = 0. As we shall see this constraint will have far-reaching consequences.
A careful reader may have noticed that we could do exactly the same computation using
∇w Tww̄ +∇w̄ Tw̄w̄ = 0, in which case we will reach the same equation as (3.102) with c replaced
by c̄. These two equations are not consistent with each other unless c = c̄. It turns out that a
conformal field theory with c 6= c̄ cannot be coupled consistently to two-dimensional gravity;
on top of the Weyl anomaly discussed here it has also a gravitational anomaly.

3.7 Conformal field theory with boundaries

We end up this chapter by saying a few words about conformal field theories on surfaces with
boundaries, which is relevant to study open strings. The worldsheet for a propagating open
string corresponds to an infinite strip, which can be represented in the complex plane w by
0 6 Re (w) 6 π , Im(z) ∈ R . (3.103)
To perform the analogue of radial quantization, is is convenient to use the conformal mapping
from the strip to the upper half-plane10 , see fig. 3.4:
w 7→ z = −e−iw , w̄ 7→ z̄ = −eiw̄ . (3.104)
Under this map, an initial state |ini prepared at Euclidean time τ → −∞ becomes an operator
sitting at the origin, on the real axis Im(z) = 0.
10
There’s a conventional minus sign difference compared to the mapping used in the closed string case, see
eqn. (3.11), in order to get the upper half-plane rather than the lower half-plane.

74
Conformal symmetry

w-plane

z-plane

σ2

radial time |z|

−∞
|ini

|ini

Figure 3.4: Conformal mapping from the strip to the half-plane.

Due to translation invariance along the boundary, such operator can be moved anywhere
on the real axis. Hence we have learned an important fact: open string states correspond to
boundary operators in the conformal field theory on the half-plane, i.e. operators that are
defined on the real axis.
A natural choice of boundary conditions is that there is no flow of momentum across the
boundary, i.e. that T01 = 0 on the real axis. In complex coordinates, it means that

Tzz (z) = Tz̄z̄ (z̄)z=z̄ . (3.105)
These boundary conditions are called conformal boundary conditions, as they preserve one
copy of the Virasoro algebra.
To see this, instead of having two currents Tzz and Tz̄z̄ defined on the upper half-plane, it
is more convenient to define a unique current on the full complex z plane :
Tzz (z) := Tz̄z̄ (z̄ 0 ) , z 0 = z̄ , Im(z) < 0 . (3.106)
The boundary condition (3.105) ensures naturally that Tzz is continuous across the boundary.
Because of the boundary conditions, there exists a unique set of Virasoro generators. Let
us define C1/2 as a half-circular contour in the upper half-plane around the origin, and C0 a
closed circular contour around the oriign. The Virasoro generators are defined as :
Z I
1 n+1 n+1 1
dz zn+1 Tzz (z) ,


Ln = dz z Tzz (z) − dz̄ z̄ Tz̄z̄ (z̄) = (3.107)
2iπ C1/2 2iπ C0
where we have used the definition (3.106) in the second equality. The OPE between Tzz
and itself is unchanged, since it is a local property of the field theory. Hence we get the
same Virasoro algebra (3.80) that we have obtained in the case of a CFT without boundary,
however we have a single copy instead of two.
Another way to understand this is the following. The Virasoro generator Ln (resp. L̃n ) was
the conserved charge associated to the holomorphic change of coordinates δz = (z) = zn+1
¯(z̄) = z̄n+1 ). In the present context one should allow only transformations that
(resp. δz̄ = 
preserve the boundary, i.e. the real axis. They should obey (z) =  ¯ (z̄)|z=z̄ .
The highest weight representations |hib of this Virasoro algebra correspond, under the
state-operator correspondence, to boundary primary operators Ob (x) defined on the real axis.

75
Conformal symmetry

References
[1] A. A. Belavin, A. M. Polyakov, and A. B. Zamolodchikov, “Infinite Conformal Symmetry
in Two-Dimensional Quantum Field Theory,” Nucl. Phys. B241 (1984) 333–380.

[2] P. Di Francesco, P. Mathieu, and D. Senechal, Conformal Field Theory. Graduate Texts
in Contemporary Physics. Springer-Verlag, New York, 1997.

[3] R. Blumenhagen and E. Plauschinn, “Introduction to conformal field theory,” Lect. Notes
Phys. 779 (2009) 1–256.

76
Chapter 4

Free conformal field theories

77
Free conformal field theories

The presentation of conformal field theories in two dimensions in chapter 3 was rather
formal and abstract. We will look at simple examples of conformal field theories, as free
massless scalar fields, that will eventually be the building blocks of bosonic string theory and
later on of the superstring theories.

4.1 Free scalar fields

The basic ingredients of the Polyakov action (3.1) in conformal gauge and with a flat
Minkowski-space time target space are a set of scalar fields xµ (σi ), µ ∈ {0, 1, . . . , D − 1},
governed by the free quadratic action:
Z
1
S= 0
d2 σ ηµν δij ∂i xµ ∂j xν . (4.1)
4πα
Let us consider a single space-like coordinate, i.e. a free two-dimensional massless scalar
field with action Z
1
S= d2 σ δij ∂i x∂j x . (4.2)
4πα 0
The stress-energy tensor deduced from this action is given, following eqn. (3.17), by the
expression:
1 1
Tij = 0 ∂i x∂j x − δij ∂k x∂k x , (4.3)
α 2α 0
In complex coordinates, the action becomes:
Z
1 ¯ ,
S= d2 w ∂x∂x (4.4)
2πα 0
From now on we will consider the conformal field theory on the complex plane. The
non-vanishing components of the stress tensor are
1
T (z) = − ∂x∂x , (4.5a)
α0
1 ¯ ¯
Te(z̄) = − 0 ∂x ∂x , (4.5b)
α
with Tzz̄ = 0 because of two-dimensional conformal invariance of the action as stressed in the
previous chapter. The equations of motion following from (4.4) are simply
¯
∂∂x(z, z̄) = 0 , (4.6)

hence x is the sum of a holomorphic and a anti-holomorphic function:1

x(z, z̄) = xL (z) + xR (z̄) (4.7)

1
Whenever the target-space coordinate corresponding to the field x(z, z̄) is non-compact, the (anti-) holo-
morphic functions xl (z) and xr (z̄) do not define by themselves consistent fields, are they are not univalued
on the plane, see eq. (4.31) below.

78
Free conformal field theories

On the Minkwoskian cylinder, i.e. with z = e−i(τ+σ) and z̄ = e−i(τ−σ) , the first term describes
left-moving waves and the second term right-moving ones. For this reason, the Virasoro
algebras corresponding to holomorphic and anti-holomorphic conformal transformations are
called respectively left-moving and right-moving algebras.
The free scalar action (4.4) has a global symmetry on top of conformal symmetry, the
invariance under translations in target space:

x 7→ x + a . (4.8)

From Noether theorem, and using the equation of motion (4.6), one finds conserved holomor-
phic and anti-holomorphic currents:
1 ¯ =0
J(z) := Jz (z) = ∂x , ∂J (4.9a)
α0
eJ(z̄) := Jz̄ (z̄) = 1 ∂x
¯ , ∂eJ = 0 (4.9b)
α0
Invariance of the action (4.4) under conformal transformations indicates that xµ has conformal
weights (0, 0), hence J (resp. eJ) has conformal weights (1, 0) (resp. conformal weights (0, 1)).
One can apply the residue formula (3.37), that was established for conformal transforma-
tion but holds for any (anti)holomorphic conserved current in a two-dimensional CFT, if one
splits the translation of x into a translation of xL (z) by a/2 and a translation of xR (z̄) by
a/2. It leads to the relation:

a/2 = δxL = −Resz 0 →z (aJ(z 0 )x(z, z̄))

1 a
=⇒ a 0
∂x(z 0 )x(z, z̄) = · · · − 0
+ regular. (4.10)
α 2(z − z)

which implies that, using the same reasoning for the anti-holomorphic current

α0
∂x(z 0 )x(z, z̄) = − + reg. (4.11a)
2(z 0 − z)
0
¯ 0 )x(z, z̄) = − α
∂x(z̄ + reg. (4.11b)
2(z̄ 0 − z̄)
¯ has scaling dimension (1, 0) (resp.
Because x has scaling dimensions (0, 0), and ∂x (resp. ∂x)
(0, 1)) it is actually impossible to get terms more singular than 1/(z 0 − z), as they would be
multiplied by operators of negative weights, which are forbidden in unitary conformal field
theories. Differenciating once with respect to z one gets the OPE between the currents
1
J(z 0 )J(z) = − + reg. (4.12a)
2α 0 (z 0 − z)2
eJ(z̄ 0 )eJ(z̄) = − 1
+ reg. (4.12b)
2α 0 (z̄ 0 − z̄)2

79
Free conformal field theories

Finally, integrating equations (4.11) w.r.t. z and z̄, one finds that the OPE between x(z, z̄)
and itself is given by

α0
x(z 0 , z̄ 0 )x(z, z̄) = − log µ2 (z − z 0 )(z̄ − z̄ 0 ) + reg.


(4.13)
2

where the parameter µ is an infrared cut-off which is here for dimensional reasons. The
singular term on the right-hand side of this equation is actually the (perhaps familiar) position
space scalar Green function in two dimensions, that will be computed directly later on, see
eqn. (6.9). The present derivation was putting an emphasis on the symmetries of the theory.
Unlike Green functions in higher dimensions, this one does not fall off at large distances,
which eventually implies the absence of phase transitions for continuous symmetries in two
dimensions.

4.1.1 Stress-energy tensor

The classical expressions (4.5) of the components of the stress-energy tensor should actually
be regularized in the quantum theory.

Normal-ordered product
We define first the normal-ordered product of operators by subtracting the short-distances
singularities when the two operators approach each other:

•
• O1 (z 0 )O2 (z) •• := O1 (z 0 )O2 (z) − O1 (z 0 )O2 (z) , (4.14)

where O1 (z 0 )O1 (z) means all terms involving negative powers of (z 0 − z) in the OPE between
O1 and O2 . The relation between this normal ordering and the usual one (annihilation
operators to the right) will become more transparent later on. Note that eq. (4.14) is an
operator equation that should hold inserted in any correlation function; as such, the first
term on the right-hand side should be understood as a time-ordered product (or rather in
the present context as a radial-ordered product).
When O1 and O2 are at coincident points we define the normal order product as a limit,
namely:
 
• O1 O2 • (z) = lim O1 (z )O2 (z) − O1 (z )O1 (z)
• • 0 0
(4.15)
0 z →z

A convenient way of rewriting this expression uses a contour integral:

Z
1 dz 0
• O1 O2 • (z) = O1 (z 0 )O2 (z) ,
• •
0
(4.16)
2iπ Cz z − z

where Cz is a contour around z. The integral picks only the regular terms in the OPE of O1
and O2 by construction.

80
Free conformal field theories

Finally we will consider in several computations the OPE between normal-ordered op-
erators. One has the following generalization of Wick theorem to arbitrary interacting two-
dimensional CFTs:
Z  
1 dz2
O1 (z1 ) • O2 O3 • (z3 ) =
• •
O1 (z1 )O2 (z2 )O3 (z3 ) + O2 (z2 )O1 (z1 )O3 (z3 ) (4.17)
2iπ Cz3 z2 − z3
It can be applied iteratively to deal with more complicated operator products.

4.1.2 OPE of the stress-energy tensor

In terms of the normal product, applied to the bilinear of the current J of self-OPE (4.12),
the quantum expression of T (z) reads:
T (z) = −α 0 •• JJ •• (z) , (4.18a)
Te(z̄) = −α 0 • eJeJ • (z̄) ,
• • (4.18b)
with here:
α0
   
•
• JJ •• (z) = lim 0 0
J(z )J(z) − J(z )J(z) = lim 0
J(z )J(z) + . (4.19)
z →z
0 0 z →z 2(z 0 − z)2
In terms of the fundamental field x and its derivatives the stress energy tensor takes the
form:
1
T (z) = − 0 •• ∂x∂x •• (z) , (4.20a)
α
1 ¯ ¯ •
Te(z̄) = − 0 •• ∂x∂x • (z̄) . (4.20b)
α
Using this definitions (4.20) we will now show that J and eJ are conformal primary opera-
tors. One needs to compute the OPE:
1 •
T (z 0 )J(z) = − ∂x∂x •• (z 0 ) ∂x(z) (4.21)
(α 0 )2 •
The OPE between the composite operator T and J is computed using (4.17). In the present
case the computation is rather simple as the self-contraction of J gives just the identity
operator. Explicitly we have
1 •
T (z 0 )J(z) = − 0 2
• 0
• ∂x∂x • (z ) ∂x(z)
(α )
1 1
=− 0 2
∂x(z 0 )∂x(z 0 )∂x(z) − 0 2 ∂x(z 0 )∂x(z 0 )∂x(z)
(α ) (α )
0
 
2 α
=− 0 2 − 0
∂x(z 0 )
(α ) 2(z − z)2
J(z) ∂J(z)
= 0 2
+ 0 , (4.22)
(z − z) (z − z)

81
Free conformal field theories

where we have used a Taylor expansion of the current J(z 0 ) around z in the last step,

J(z 0 ) = J(z) + (z 0 − z)∂J(z) + O (z 0 − z)2

(4.23)

In order to determine the central charge of the free scalar conformal field theory, one
needs to compute the OPE satisfied by T and Te, using the OPE (4.13) of the elementary field
and the definitions (4.20). One has to use the Wick contractions:

1
T (z 0 )T (z) = •
• JJ •• (z 0 ) •• JJ •• (z) = •
• J(z 0 )J(z 0 )J(z)J(z) ••
(α 0 )2

+ •• J(z 0 )J(z 0 )J(z)J(z) •• + •• J(z 0 )J(z 0 )J(z)J(z) •• + •• J(z 0 )J(z 0 )J(z)J(z) ••

+ J(z 0 )J(z 0 )J(z)J(z) + J(z 0 )J(z 0 )J(z)J(z) (4.24)

which gives, using at the last step the Taylor expansion,

• z0 − z • •
J(z 0 )J(z) •• = •
J(z)J(z) •• +(z 0 −z) •• ∂J(z)J(z) •• +O (z 0 − z)2 = •
JJ •• (z)+


• • • ∂ • JJ • (z)+· · ·
2
(4.25)
the OPE (3.68) with central charge c = 1:
1 2T (z) ∂T (z)
T (z 0 )T (z) = + 0 + 0 + reg. (4.26)
2(z 0 − z) 4 (z − z) 2 z −z

One can do the same exercise with the anti-holomorphic component Te, and one reaches the
conclusion that the free scalar field conformal field theory has central charges (c, c̄) = (1, 1).

4.1.3 Mode expansions and Virasoro algebra

The currents J and eJ, being respectively holomorphic and anti-holomorphic, admit naturally
on the plane an expansion in terms of Laurent modes, as the stress-energy tensor components.
Using standard conventions and normalization, one has
1 i X αn
J(z) = ∂x(z) = − √ (4.27a)
α0 2α 0 n∈Z zn+1
eJ(z̄) = 1 ∂x(z̄)
¯ i X α en
0
= − √ (4.27b)
α 2α n∈Z z̄n+1
0

with
√ I
dz n
αn = i 2α 0 z J(z) (4.28a)
C0 2iπ
√ I
0
dz̄ ne
α
e n = −i 2α z̄ J(z̄) . (4.28b)
C0 2iπ

82
Free conformal field theories

These coefficients can also be defined from the expansion of x(z) itself. Integrating (4.27)
one gets

α 0 X 1 αn α
r r
α0
 

en
x(z, z̄) = xc − i α0 ln z + α
e 0 ln z̄ + i + n . (4.29)
2 2 n∈Z? n zn z̄

If the field x has a non-compact target space, it should be single-valued hence one should
impose α0 = α e 0 .2 The interpretation of this coefficient is quite obvious, since J(z) and eJ(z̄) are
the components of the Noether current for space translations. It gives then the momentum
conjugate to the zero-mode x0 :3
I I  r
1 α 0 + α 0 2
dz̄ eJ(z̄) = √
e
pc = dz J(z) − = α0 . (4.30)
2π C0 C0 2α 0 α0
We obtain then the final form of the expansion of x:

α 0 X 1 αn α
r
α0
 
en
x(z, z̄) = xc − i pc ln |z| + i
2
+ n . (4.31)
2 2 n∈Z? n zn z̄

Expressed in terms of cylinder coordinates, and after continuation to Minkowski space-time,

the meaning of this expansion is perhaps more clear:

α0 X 1
r
x(σ, τ) = xc + α 0 pc τ + i αn e−in(τ+σ) + α
e n e−in(τ−σ) .


(4.32)
2 n∈Z? n

In this expression, xc + α 0 pc τ describes the center-of-mass motion of the string, while the
oscillator modes are respectively left-moving and right-moving plane waves propagating on
the string worldsheet.
The modes {αn } obey an algebra that is obtained in the same way as the Virasoro algebra
was derived before, see around equations (3.76). As there we consider the circular contours
C of radius R and C 0 of radius R 0 > R, both around the origin, and define the commutator
I I I I 
0 dz1 dz2 dz2 dz1
[αm , αn ] = −2α − zm n
1 z2 J(z1 )J(z2 )
0 2iπ C 2iπ C 0 2iπ C 2iπ
I C
dz 2 n
= −2α 0 z Resz1 →z2 (zm
1 J(z1 )J(z2 ))
2iπ 2
I
z2 + m(z1 − z2 )z2m−1
 m 
0 dz2 n
= −2α z Resz1 →z2 − . (4.33)
2iπ 2 2α 0 (z1 − z2 )2
We obtain then
[αm , αn ] = mδm+n,0 , (4.34)
2
The case of compact x is actually very important for string theory; we will come back to it in due time.
3
A factor of i in the definition of p0 was added in order to define a Hermitian operator – in fact the currents
J and eJ that we have defined are anti-Hermitian, which is the mathematicians’ conventions for generators of
a Lie algebra.

83
Free conformal field theories

with a similar algebra for the modes {e

αn }. On top of this, the zero-modes obey the usual
canonical commutation relations:
[xc , pc ] = i . (4.35)
The infinite-dimensional algebra (4.34) whose origin was translation symmetry in the
target space of the field x is called an affine Lie algebra, constructed from the (trivial) Lie
algebra u(1) of the translation group (the latter being isomorphic to R). In general, starting
for any classical Lie algebra
[ja , jb ] = fab c jc , (4.36)
one can construct an affine Lie algebra in a similar fashion:
κab
[jam , jbn ] = fab c jcm+n + 2
mδm+n,0 . (4.37)

These algebras play an important role for describing string theory in curved space-time, but
we won’t have time to present this in these lectures.
Given that the components T , Te of the stress-energy tensor are expressed in terms of the
currents J, eJ, see eqn. (4.18), it is clear that the generators {Ln } of the Virasoro algebra can
be expressed in terms of the generators {αn } of the affine Abelian algebra u(1). From the
definition (3.74) of the Virasoro generators, one gets
I
1 dz n+1 X 1X
Ln = z αr αs z−r−s−2 = αr αn−r , ∀n 6= 0 . (4.38)
2 C0 2iπ r,s
2 r∈Z

For L0 more care is needed as αn and α−n do not commute with each other. Using the
expression (4.16) of the normal-ordered product at coincident points, one gets
Z Z I
0 dz • • 0 dz dz 0 1
L0 = −α z • JJ • (z) = −α z 0
J(z 0 )J(z) (4.39)
C0 2iπ C0 2iπ Cz 2iπ z − z

One can employ the same contour manipulation as described on figure 3.3 backwards, and
write this expression as
I I I I
1X dz 0 dz 0 z
 
dz z 0 −r−1 −s−1 dz 0 −r−1 −s−1
L0 = αr αs 0
(z ) z − αs αr 0
(z ) z
2 r,s∈Z C 0 2iπ C 2iπ z − z C 0 2iπ C 2iπ z − z
(4.40)
where we have taken into account that the operator product should be radial ordered.
In the first term, to perform the integral over z, as by definition |z| < |z 0 | on C, one can
expand
1 X
∞
1
= (z/z 0 )n , (4.41)
z0 − z z 0 n=0
and compute
I X∞ I  0 −r−s−1
dz 1 −s 0 −r−1 0 −r−n−2 dz n−s (z ) , s>1
0
z (z ) = (z ) z = (4.42)
C 2iπ z − z C 2iπ 0 , s<1
n=0

84
Free conformal field theories

In the second term, to perform the integral over z 0 , as by definition |z 0 | < |z| on C, one can
expand
1X 0 n
∞
1
=− (z /z) , (4.43)
z0 − z z n=0
and compute
I X∞ I 
dz 0 1 −s 0 −r−1 −s−n−1 dz 0 0 n−r−1 −z−r−s−1 , r > 0
z (z ) = − z (z ) = (4.44)
C 2iπ z 0 − z n=0 C 2iπ 0, r<0

Putting everything together one has

I I
1 XX dz 0 0 −r−s−1 1 X X
∞ ∞
dz −r−s−1
L0 = αr αs (z ) + αs αr z
2 r∈Z s=1 2iπ 2 r=0 s∈Z 2iπ
1 X ∞
= α20 + α−r αr
2 r=1

α0 2 X
∞
= p + α−r αr (4.45)
4 r=1

Hence one realizes that the normal ordering (4.16) implies the familiar notion of normal
ordering, as for the harmonic oscillator (creation operator to the left). In this respect one
can write the latter expression as

α0 2 1 X ◦ ◦
L0 = p + ◦ α−r αr ◦ (4.46)
4 2 r6=0

where the normal ordering ◦◦ · ◦◦ of modes means that the positive modes should be put at
the end.
One can check easily that the commutations relations involving (4.38) and (4.45), com-
puted using (4.34) and (4.35), reproduces the Virasoro algebra (3.80).
To summarize, we have obtained a Virasoro algebra from an affine Lie algebra; this
construction holds for non-Abelian affine algebras (4.37) as well.

4.1.4 Primary states and descendants

We are now in position to describe the Hilbert space of the conformal field theory of a free
scalar in two dimensions. As we have already noticed, a special role is played by primary
states, that are analogous to highest weight states in representation theory of Lie algebras.
Using the state-operator correspondence, such primary states are mapped to primary local
operators on the plane as was already discussed.
As the theory contains two copies of the Virasoro algebra (and of the underlying affine
algebra), we will focus the discussion on the holomorphic one.

85
Free conformal field theories

In terms of the Virasoro generators, a primary state |hi is annihilated by the positive
modes and characterized by its L0 eigenvalue:

∀n > 0 , Ln |hi = 0 , L0 |hi = h|hi . (4.47)

Other – generically non-primary – states are obtained by acting with the creation opera-
tors. We generate this way a Verma module, whose generic state looks like

|ψi = · · · (L−n )Nn · · · (L−2 )N2 (L−1 )N1 |hi (4.48)

whichPis the infinite-dimensional analogue of a representation of a Lie algebra. The quantity

N= ∞ n=1 nNn is called the level of the state.
In general such representation is irreducible. Exceptions to this rule, in c = 1 CFTs,
occur in the representations generated by primary states of dimension h = j2 with j ∈ 12 Z+ ,
which contains a null vector at level 2j + 1, i.e. a state annihilated by all positive Ln , n > 0.
This is true in particular the vacuum state |0i, which is by definition the primary state of
conformal dimension h = 0. Since ||L−1 |0i||2 = h0|[L1 , L−1 ]|0i = 0, this state is also annihilated
by L−1 (this is easy to see in the operator language: L−1 O0 ↔ ∂1), therefore L−1 |0i is a null
vector. Hence the whole corresponding submodule, i.e. all states obtained from it by creation
operators, should be substracted from the Verma module; the vacuum representation is said
to be degenerate.
For j = 1/2, i.e. conformal dimension h = 1/4, one can check easily using the Virasoro
algebra (3.80) with c = 1 that a null vector at level one can be obtained as follows:

∀n > 0 , Ln L−2 | 41 i − (L−1 )2 | 41 i = 0 .

(4.49)

It is easy to see that, as a consequence, ||(L−2 − (L−1 )2 )| 14 i|| = 0. Likewise for j = 1 one finds a
null vector at level two, and so on. Conformal field theories with central charges c < 1 have
a more complicated pattern of null states, that play an essential role in solving the theory
algebraically.
It is always more efficient to classify states according to the largest possible symmetry of
the theory, which is in this case the affine symmetry generated by translation invariance of
the free scalar field action. In the present context one advantage is that the corresponding
representation theory is simpler, due to the absence of null vectors. One defines then the
highest weight states as follows
r
α0
∀n > 0 , αn |pi = 0 , α0 |pi = p|pi . (4.50)
2
Using the expression (4.38) for the Virasoro generators in terms of the affine generators, one
realizes that |pi is also a Virasoro primary:

1X
∀n > 0, Ln |pi = αr αn−r |pi = 0 , (4.51)
2 r∈Z

86
Free conformal field theories

since each term in the sum contains at least one positive mode. The conformal dimension of
this primary state is given, following (4.45), by
α0 2
L0 |pi = h|pi , h= 4
p . (4.52)
A generic state in the module constructed from the highest weight state |pi is
|ψi = · · · (α−n )Nn · · · (α−2 )N2 (α−1 )N1 |pi . (4.53)
Using the Virasoro algebra one has
X
!
α0 2
L0 |ψi = L0 · · · (α−n )Nn · · · (α−2 )N2 (α−1 )N1 |pi = nNn + 4
p |ψi , (4.54)
n

giving the dimension of a generic affine descendant state in terms of its momentum and its
level X
N := nNn . (4.55)
n
In contrast with the Virasoro symmetry, the Verma module, i.e. the set of states of the
form (4.53) for given p, contains no null vectors unless one considers the vacuum state |0i
since α−1 |0i = 0.
0
From the point of view of the Virasoro symmetry, as long as α4 p2 ∈
/ (Z+ 12 )2 , the irreducible
affine u(1) representation gives an irreducible Virasoro representation (otherwise, it should
be decomposed accordingly).

4.1.5 Operator description

Using the state-operator correspondence, the analysis of the state space of the conformal field
theory can be rephrased in terms of operators. Let us consider a primary operator of the
left-moving and right-moving affine u(1) algebras. These algebras being infinite-dimensional
generalizations of the algebra corresponding to the translation symmetry, it is natural to
look for operators corresponding to the primary states in the form of plane waves. These are
called vertex operators:

• (ip)2 • 2 •
Vp (z, z̄) = • eipx •• (z, z̄) = 1 + ipx(z, z̄) + x (z, z̄) + · · · (4.56)
2! • •
Let us check that it is indeed a primary operator. One has
√ I
dz n
αn Vp (0, 0) = i 2α 0 z J(z) •• eipx •• (0, 0) , (4.57)
C0 2iπ

with C0 a contour encircling the origin. One has the OPE:

1 X (ip)n 1 X (ip)n
J(z) •• eipx •• (0, 0) = 0 ∂x(z) • n•
• x • (0) = 0
∂x(z)x(0) •• xn−1 •• (0)
α n
n! α n
(n − 1)!
−ip • ipx •
= e • (0, 0) + reg. (4.58)
2z •

87
Free conformal field theories

which implies that (4.57) vanishes when n > 0.

Equation (4.57) also gives the expression of descendant states, i.e. those obtained from
the primary states from the action of a negative mode of the current, in terms of normal-
ordered products of operators. As only non-singular terms from the OPE between J and Vp
contribute one has simply
r I
2 dz 1 •
∀n > 0 , α−n Vp (0, 0) = i ∂x(z)eipx •• (0, 0)
α 0 C0 2iπ zn •
r
2 i • n ipx •
= 0 •∂ xe • (0, 0) . (4.59)
α (n − 1)!
Finally, one can notice that the current itself, J, is obtained from the action of α−1 on the
vacuum state |0i. As we have noticed already, it is a null vector of the vacuum representation,
hence a primary state itself. It is consistent with the OPE (4.22), which indicates that J is a
conformal primary of dimension (1, 0).
One may wonder about the status of the field x(z, z̄) itself, which has conformal dimensions
(0, 0). It does not actually correspond to a conformal operator, as can be realized from the
OPE (4.13) which depends on an infrared cutoff. Another way to see this is that inserting
x(z, z̄) at the origin of the plane corresponds to applying the zero-mode xc on the vacuum,
and xc |0i is non-normalizable because of the infinite volume of target space. For this reason
this state is not bound to satisfy the unitarity constraint (3.94).

4.1.6 Bosonic strings in D dimensions

To close this section, one can repeat the same analysis for the set of D scalar fields corre-
sponding to the embedding of the string in D dimensional Minkowski space-time. One gets
the current algebras
ηµν ηµν
Jµ (z)Jν (z 0 ) = − , eJµ (z̄)eJν (z̄ 0 ) = − . (4.60)
2α 0 (z − z 0 )2 2α 0 (z̄ − z̄ 0 )2
Equivalently one has the commutators

[αµm , ανn ] = mδm+n,0 ηµν , αµm , α

[e e νn ] = mδm+n,0 ηµν . (4.61)

The corresponding Virasoro algebras, built from

T (z) = −α 0 ηµν •• Jµ Jν •• (z) , (4.62a)

Te(z̄) = −α 0 ηµν • eJµeJν • (z̄) ,
• • (4.62b)

give the central charges (c, c̄) = (D, D). An example of vertex operator, which is a descendant
state, is given by:
αµ−1 α
e ν−1 Vp (0, 0) = •• ∂xµ ∂xν eip·x •• (0, 0) . (4.63)
as we will see, this is the state corresponding to the space-time graviton.

88
Free conformal field theories

Because the target space has Minkowskian signature, the kinetic term for x0 (z, z̄) has the
wrong sign. Consequently, states obtained from the action of the corresponding modes α0−n
and α e 0−n have negative norm. While this is a problem at the level of the conformal theory,
we will see that all such states are removed from the physical spectrum of the string theory
itself.
Note finally that, unlike the generators of translations, the generators of Lorentz transfor-
mations and rotations in target space cannot be promoted to affine currents in the conformal
field theory; this is a consequence of the previous observation that xµ (z, z̄) are not themselves
conformal operators.

4.1.7 Free bosons on the upper half-plane

In order to describe open string one has to consider the conformal field theory of free bosons
on the upper half-plane. Recall that, for any conformal field theory, one copy of the conformal
algebra is preserved if one imposes the conformal boundary conditions

Tzz (z) = Tz̄z̄ (z̄)z=z̄ . (4.64)

We will actually impose stronger boundary conditions, that preserve one copy of the affine
symmetry associated with the theory. As we have seen above, that the theory of a free boson
x(z, z̄) on the plane is associated with a pair of holomorphically conserved currents :
1 1¯
J(z) = ∂x , J̃(z̄) = ∂x . (4.65)
α0 α0
The boundary conditions for these currents on the real axis correspond actually to the Dirich-
let and Neumann boundary conditions that were discussed in chapter 3, section 2.5:

• Neumann : J(z) = J̃(z̄)z=z̄ (4.66a)

• Dirichlet : J(z) = −J̃(z̄) z=z̄
(4.66b)

Using eq. (4.18), it is obvious that both affine boundary conditions imply the conformal
boundary conditions (4.64).

Neumann boundary conditions. We consider first the mode expansion for a boson with
Neumann boundary conditions. Because of eqn. (4.66a). the Fourier modes should by iden-
tified as
∀n 6= 0 , αn = α̃n . (4.67)
On top of that, single-valuedness of the field x(z, z̄) on the plane (rather, on the doubled
half-plane) imposes that α0 = α̃0 . This zero-mode is related to the momentum in target
space, however with a different normalisation as for closed strings as there is only one copy of
the current algebra, following the same reasoning as in eq. (3.106) (compare with eq. (4.30)):
I
1 α0
pc = dz J = √ . (4.68)
2π C0 2α 0

89
Free conformal field theories

This momentum is associated with translations x 7→ x + a, which are preserved by Neumann

boundary conditions. One can write finaly we the expansion :
α 0 X αn 1
r  
1
x(z, z̄) = xc − iα pc ln |z| + i
0 2
+ . (4.69)
2 n∈Z? n zn z̄n

In order to compute the conformal weight of a given state, it is easy to show that
X
∞
0
L0 = α pc2 + α−n αn . (4.70)
n=1

Dirichlet boundary conditions. We consider now a boson with Dirichlet boundary condi-
tions. First, because of eqn. (4.66b). the Fourier modes should by identified as
∀n 6= 0 , αn = −α̃n . (4.71)
Recall that, for an open string attached to a Dp-brane, Dirichlet boundary conditions
correspond to the directions transverse to the brane, along which the endpoints of the strings
cannot move. Since there is no translation invariance in those directions, there is no associated
conserved charge. Let us consider an open string stretched between a first D-brane at x = X0
and a second D-brane at x = X1 :
x(σ = 0, τ) = X0 , x(σ = π, τ) = X1 . (4.72)
Under the conformal mapping (3.104) from the strip to the upper half-plane, the line σ = 0
(resp. σ = π) is mapped to R<0 (resp. R>0 ). Hence, the mode expansion that satisfies the
right boundary conditions is
α 0 X αn 1
r  
X0 − X1 1
x(z, z̄) = X0 + ln(z/z̄) + i − . (4.73)
2iπ 2 n∈Z? n zn z̄n
1 1
In this case, since the affine current is expanded as J(z) = 2iπα 0 (X0 − X1 ) z + · · · , one has for

the Virasoro zero-mode:

 2 X ∞
1 X1 − X0
L0 = 0 + α−n αn . (4.74)
α 2π n=1

Consider a bosonic string stretched between two parallel Dp-branes, spanning the direc-
tions x0 , . . . xp , and located respectively at Xa0 and Xa1 , a = p+1, · · · , D−1 in their transverse
dimensions. The mode expansion is
α 0 X αµn 1
r  
1
x (z, z̄) = xc − iα pc ln |z| + i
µ µ 0 µ 2
+ , µ = 0, . . . , p (4.75a)
2 n∈Z? n zn z̄n
α 0 X αan 1
r
Xa0 − Xa1
 
a a 1
x (z, z̄) = X1 + ln(z/z̄) + +i − , a = p + 1, . . . , D − 1
2iπ 2 n∈Z? n zn z̄n
(4.75b)

90
Free conformal field theories

4.2 Free fermions

The next conformal field theory that we consider is the theory of free massless fermions on a
two-dimensional Minkowskian worldsheet.
The smallest spinorial representation of the two-dimensional spin group Spin(1, 1) is
Majorana-Weyl, i.e. a real one-component spinor. In the Majorana basis, we choose the
two-dimensional gamma matrices to be
     
0 0 −1 1 0 1 2 0 1 −1 0
γ = , γ = , γ =γ γ = . (4.76)
1 0 1 0 0 1

The two-dimensional action for a Majorana fermion is of the form

Z
1
S=− d2 σ ψ̄γα ∂α ψ , (4.77)
2π

with ψ̄ = ψ† iγ0 = ψT iγ0 using the Majorana condition.

The two-component Majorana spinor ψ can be decomposed into a pair of Majorana-Weyl
one-component spinors ψ± of definite chirality:
     
ψ− 2 ψ− −ψ−
ψ= , γ = . (4.78)
ψ+ ψ+ ψ+

Accordingly the two-dimensional action (4.77) splits into two independent chiral actions
for the Majorana-Weyl spinors:
Z
i
d2 σ ψ− ∂+ ψ− + ψ+ ∂− ψ+ ,


S= (4.79)
π

the light-cone derivatives being defined as ∂± = 21 (∂0 ± ∂1 ). The equations of motion being
∂± ψ∓ = 0, ψ+ (resp. ψ− ) is a left-moving (resp. right-moving) fermion.
We now move to Euclidean signature and two complex coordinates, as we did for the free
scalar field. We consider the following action for a set of M left-moving and M right-moving
real fermions: Z
1 
¯ j + δij ψ

S= d2 z δij ψi ∂ψ e i ∂ψ
ej , (4.80)
4π
where ψi and ψ̃i are two independent real Grassmann variables. The equations of motion
are simply
¯ i = 0 =⇒ ψi = ψi (z)
∂ψ (4.81a)
e i = 0 =⇒ ψ
∂ψ ei = ψ
e i (z̄) (4.81b)

91
Free conformal field theories

4.2.1 Symmetries and current algebra

The action (4.80) is invariant under independent SO(M) chiral rotations for the left-moving
and right-moving fermions:

ψi 7→ Ri j ψi , e i 7→ R
ψ ei ψ
j
ei , R, R̃ ∈ SO(M) . (4.82)

The corresponding holomorphic and anti-holomorphic Noether currents are:

Jij (z) = − •• ψi ψj •• (z) , eJij (z̄) = − • ψ

•
e j • (z̄) ,
e iψ
• i<j (4.83)

and have respectively conformal weights (1, 0) and (0, 1). We consider an infinitesimal (left-
moving) SO(M) transformation:
δψk = αij Tijk` ψ` , (4.84)
where the matrices Tijk` correspond to the vector representation of SO(M):

Tijk` = δik δj` − δi` δjk . (4.85)

Using the residue formula (3.37) one gets

αij Tijk` ψ` (z) = −αij Resz 0 →z Jij (z 0 )ψk (z) ,

(4.86)

from which we deduce the OPE

Tijk` ψ` (z)
Jij (z 0 )ψk (z) = − + reg. (4.87)
z0 − z
This OPE, which expresses the fact that the set of left-moving fermions {ψi (z)} are primaries
of the affine so(M) algebra in the vector representation, can be easily shown to derive from
the fundamental OPE:
δij
ψ i ψj = + reg. (4.88)
z−w
We have indeed (not forgetting that the ψi (z)s are anticommuting Grassmann variables):

Jij (z 0 )ψk (z) = •

ψi ψj •• (z 0 ) ψk (z) = ψi (z 0 )ψj (z 0 )ψk (z) + ψi (z 0 )ψj (z 0 )ψk (z)
•
1
−δik ψj (z) + δjk ψi (z)


= 0
z −z
1
δik δj` − δjk δi` ψ` (z) .


=− 0 (4.89)
z −z
The OPE between two currents can also be computed and gives, after some slightly tedious
algebra:
κij k` fij k` mn mn
Jij (z 0 )Jk` (z) = − 0 + J (z) + reg. (4.90)
(z − z)2 z0 − z

92
Free conformal field theories

with the Killing form and the structure constants of the Lie algebra so(M) (in a basis where
the long roots have square length two):

κij k` = δik δj` − δi` δjk , (4.91)

fij k` mn = 21 (δik δ`n − δi` δkn )δjm + 12 (δj` δkn − δjk δ`n )δim
− 12 (δik δ`m − δi` δkm )δjn ) − 21 (δj` δkm − δjk δ`m )δin ) . (4.92)

A similar result is obtained with the currents eJij (z̄) constructed from the right-moving
fermions ψ(z̄).
e
Finally, as anticipated in the previous section, from the left and right-moving SO(M)
affine currents one can write the left and right-moving components of the stress-energy tensor.
They can be obtained from two different perspectives. First, one can just use the stress-energy
tensor obtained from the action (4.80) and get:

1X • i i• X
M M
1 • ¯ψ
e i∂ e i • (z̄) .
T (z) = − ψ ∂ψ • (z) , Te(z̄) = − ψ (4.93)
2 i=1 • 2 i=1
• •

Second, one can obtain directly T and T̃ from the second Casimir operator of the Lie algebra;
this is the Sugawara construction. Explicitely one has
1 X 1 X
• ij ij • • eijeij •
T (z) = − J J • (z) , Te(z̄) = − J J • (z̄) . (4.94)
2(M − 1) i<j • 2(M − 1) i<j •

To show this, one can look at little bit closer at the OPE between the currents Jij (z 0 ) and
Jij (z). The term giving the first-order pole in (4.90) is

•
• ψi (z 0 )ψj (z 0 )ψi (z)ψj (z) •• + •• ψi (z 0 )ψj (z 0 )ψi (z)ψj (z) ••

+ •• ψi (z 0 )ψj (z 0 )ψi (z)ψj (z) •• + •• ψi (z 0 )ψj (z 0 )ψi (z)ψj (z) ••

1  • j 0 j • ij • j 0 i • ij • i 0 j • • i 0 i •

= 0 − • ψ (z )ψ (z) • + δ • ψ (z )ψ (z) • + δ • ψ (z )ψ (z) • − • ψ (z )ψ (z) • .
z −z
(4.95)

The terms participating in the stress-energy tensor (4.94) are then obtained from the Taylor
expansion of ψi (z 0 ) and ψj (z 0 ) at first order. Summing over 1 6 i < j 6 M, the terms in δij
drops, and we are left with

X X
M
•

• •
• ψj ∂ψj + ψi ∂ψi • = (M − 1) • ψi ∂ψi •• , (4.96)
i<j i=1

which gives eventually the stress-energy tensor (4.93).

93
Free conformal field theories

Using either form of the stress-energy tensor, the corresponding OPEs are given by the
familiar expressions:
M 2T (z) ∂T (z)
T (z 0 )T (z) = + + + reg. , (4.97a)
4(z 0 − z) 4 (z − z )0 2 z − z0
M 2Te(z̄) ∂Te(z̄)
Te(z̄ 0 )Te(z̄) = + + + reg. , (4.97b)
4(z̄ 0 − z̄)4 (z̄ − z̄ 0 )2 z̄ − z̄ 0
which show that the M left-moving fermions ψi (z) gives a holomorphic CFT with central
charges (c, c̄) = (M/2, 0), while the M right-moving fermions ψe i (z̄) gives a anti-holomorphic
CFT with central charges (c, c̄) = (0, M/2).
Finally one can check that (i) the operators ψi and ψ e i are respectively conformal pri-
maries of dimensions (1/2, 0) and (0, 1/2), and (ii) that Jij and eJij are conformal primaries
of dimensions (1, 0) and (0, 1). One has first:

1X • j j• 0 i
M
0 i
T (z )ψ (z) = − ψ ∂ψ • (z )ψ (z)
2 j=1 •

1 X j 0
M 
=− ψ (z )∂ψj (z 0 )ψi (z) + ψj (z 0 )∂ψj (z 0 )ψi (z)
2 i=1

1 X
M
δij δij 
=− − 0 ∂ψj (z 0 ) − 0 ψ j 0
(z )
2 j=1 z −z (z − z)2
1 i 1
= ψ (z) + ∂ψi (z) + reg. (4.98)
2(z 0 − z)2 z0 − z
And then, using eqn. (4.17):

T (z1 )Jij (z3 ) = −T (z1 ) •• ψi ψj •• (z3 )

I
1 dz2  
=− T (z1 )ψi (z2 )ψj (z3 ) + T (z1 )ψi (z2 )ψj (z3 )
2iπ Cz3 z2 − z3
I  
i i j j
1 dz2  ψ (z2 ) ∂ψ (z2 ) j
  ψ (z3 ) ∂ψ (z3 ) 
=− 2
+ ψ (z3 ) + ψi (z2 ) 2
+
2iπ Cz3 z2 − z3 2(z1 − z2 ) z1 − z2 2(z1 − z3 ) z1 − z3

(4.99)
In this expression one needs to compute the remaining OPE’s between the fermions, and
collecting the regular terms in the limit z2 → z3 in the integrand gives finally
•
•ψi ψj •• (z3 ) ∂ •• ψi ψj •• (z3 )
T (z1 )Jij (z3 ) = + + reg.
(z1 − z3 )2 z1 − z3
Jij (z3 ) ∂Jij (z3 )
= + + reg. (4.100)
(z1 − z3 )2 z1 − z3

94
Free conformal field theories

4.2.2 Modes expansion

We start by decomposing the fermionic fields ψi and ψe i on the cylinder in Fourier modes.
We start by choosing anti-periodic boundary conditions, namely

ψi (σ1 + 2π, σ2 ) = −ψi (σ1 , σ2 ) (4.101)

These are called the Neveu-Schwarz boundary conditions in the string theory context. In the
statistical physics context, if σ1 is the Euclidean time, these are the natural boundary condi-
tions in the computation of the partition function. We get from these boundary conditions
the expansion in Fourier modes:
iπ
X 1
ψjcyl. (σ1 , 0) = e− 4 einσ ψjn . (4.102)
n∈Z+1/2

The operator ψj (z) being a conformal primary of dimension (1/2, 0), the conformal mapping
from the cylinder to the plane gives from eqn. (3.58):

ψjcyl. (w) = (−iz)1/2 ψjplane (z) , (4.103)

One obtains then an expansion of ψi (z) and of ψ

e i (z̄) on the plane in Laurent modes:

X ψin X ei
ψ n
ψi (z) = , e i (z̄) =
ψ . (4.104)
zn+1/2 z̄n+1/2
n∈Z+1/2 n∈Z+1/2

The shift of one-half in the exponent of z is compensated by the one-half shift in n, such
that at the end the operators ψi (z) and ψ e i (z̄) are single-valued on the complex plane. The
modes are obtained from I
i dz n−1/2 i
ψn = z ψ (z) (4.105)
2iπ
and obey an anti-commuting algebra which is obtained in the same way as before, see
eqn. (4.34):
I  I dz
dz2 n−1/2 
m−1/2 i 2 n+m−1 ij
{ψm , ψn } =
i j
z Resz1 →z2 z1 j
ψ (z1 )ψ (z2 ) = z δ (4.106)
2iπ 2 2iπ 2
Hence 
ψim , ψjn = δm+n,0 δij (4.107)

Next one defines the Laurent modes of the currents Jij and eJij in the same way as we did
in the previous section. One has
X Jij
n
Jij (z) = n+1
, (4.108)
n∈Z
z
with
I I I I
dz n ij dz n • i j • dz dz 0 zn
Jijn = z J = z ψ ψ • (z) = ψi (z 0 )ψj (z) (4.109)
2iπ 2iπ • 2iπ 2iπ z 0 − z

95
Free conformal field theories

At this stage one can use the same contour manipulation as in eqn. (4.40), taking into account
the anticommuting nature of the Grassmann variables, and write this expression as:
X I I
 dz 0 dz zn
Jijn = ψir ψjs (z 0 )−r−1 z−s−1
C0 2iπ C 2iπ z 0 − z
r,s∈Z+1/2
I I
j i dz dz 0 zn 0 −r−1 −s−1

+ ψs ψr 0
(z ) z (4.110)
C 0 2iπ C 2iπ z − z

After the same steps as there, one finds that

X X X
◦
Jijn = − ψin−r ψjr + ψjr ψin−r = − ◦ ψin−r ψjr ◦◦ (4.111)
r∈N+1/2 r∈−N−1/2 r∈Z+1/2

These modes obey the commutation relations

[Jijm , jk`
n ] = −κ
ij k`
mδm+n,0 + fij k`mn Jk`
m+n , (4.112)

consistently with the currents OPE (4.90).

Finally the modes of the Virasoro algebra can be found using exactly the same methods.
One has the expansion
1 X X
Ln = (r − n/2) ◦◦ ψin−r ψir ◦◦ , (4.113)
2 i
r∈Z+1/2

in terms of the fermionic modes, or

1 X X
Ln = ◦
◦ Jijn−r Jijr ◦◦ , (4.114)
2(M − 1) r∈Z i<j

in terms of the modes of the current Jij (z).

4.2.3 Representation theory

As for free bosons, the Hilbert space of the conformal field theory of free fermions is analyzed
in terms of representation theory of the corresponding affine Lie algebra of generators {Jijn }.
An alternative description is to use the representation theory of the algebra of the modes
{ψi } themselves.
As in the previous examples we will define primary states of the affine algebra that are
annihilated by the positive modes {Jijn , n > 0}, that will also turn out to be conformal primary
states.
The affine algebra (4.112) contains a closed sub-algebra of zero-modes, called the hori-
zontal subalgebra coinciding with the Lie algebra so(M) itself:

[Jij0 , Jkl ij kl mn
0 ] = f mn J0 . (4.115)

96
Free conformal field theories

In this respect, one can classify the affine primary and descendant states in terms of repre-
sentations of the horizontal sub-algebra.
The affine primary states are also primary states with respect to the Virasoro algebra.
To see this and to compute their conformal dimension, we consider affine primary operators
φα in a representation R of the Lie algebra,

R(T ij )αβ φβ
Jij (z 0 )φα (z) = − + reg. (4.116)
z0 − z
where the matrix R(T ij ) corresponds to the generator T ij of the Lie algebra in the represen-
tation R. In terms of Laurent modes, one has

Jij0 |αi = −R(T ij )αβ |βi , ∀n > 0 , Jijn |αi = 0 , (4.117)

where |αi is the state obtained from acting with φα (z) on the vacuum.
Next we check that |αi is annihilated by all positive Virasoro modes. From equa-
tion (4.114) one finds
1 X X
∀n > 0 , Ln |αi = ◦
◦ Jijn−r Jijr ◦◦ |αi = 0 (4.118)
2(M − 1) r∈Z i<j

and
1 1
L0 |αi = Jij0 Jij0 + · · · |αi = R(T ij )αβ R(T ij )βγ |γi.


(4.119)
2(M − 1) 2(M − 1)
We recognize in the last expression the quadratic Casimir of the representation R:
X
R(T ij )αβ R(T ij )βγ = δαγ cR , (4.120)
i<j

hence the conformal dimension of this primary operator reads

cR
h= . (4.121)
2(M − 1)

Trivial representation
We associate to the trivial representation of the horizontal so(M) sub-algebra of the affine
algebra (4.112) the Neveu-Schwarz vacuum, which is the state |0ins annihilated by all positive
modes of the fermionic field:

∀i , ∀r ∈ N + 1
2
, ψir |0ins = 0 . (4.122)

It is trivial to see that this state is indeed an affine primary in the trivial representation of
the horizontal sub-algebra:
X
∀ i < j , ∀n > 0 , Jijn |0ins = − ◦ ψn−r ψr ◦ |0ins = 0 ,
◦ i j ◦
(4.123)
r∈Z+1/2

97
Free conformal field theories

and a Virasoro primary of conformal dimension zero:

1 X
∀n > 0 , Ln |0ins = (r − n/2) ◦◦ ψin−r ψir ◦◦ |0ins = 0 . (4.124)
2
r∈Z+1/2

Descendant states in the trivial representations are obtained by acting with the negative
modes of the currents Jij . A generic state is of the form
i2 j2 i2N j2N i1 j1 i1N j1N
|ψi = · · · J−2
1 1
· · · J−22 2 J−1
1 1
· · · J−11 1 |0ins , (4.125)
| {z }| {z }
N2 terms N1 terms

where the modes of the currents can be expressed in terms of fermionic modes using (4.111).
Such generic state has a complicated multi-index structure, which corresponds generically
to a reducible representation of the horizontal so(M) subalgebra (4.115). The dimension of
such a state is given by its level,
X
h = N := nNn . (4.126)
n

Vector representation
We next consider affine primary states transforming in the vector representation of the hor-
izontal sub-algebra. As the fundamental fermionic field ψ( z) itself transforms as a vector of
SO(M), see eq. (4.82), one considers the following set of states

|ii := ψi−1/2 |0i , i ∈ {1, . . . , M} . (4.127)

which correspond, using the state-operator correspondence, to the fields ψi (z).

These states transform as expected in the vector representation of the horizontal sub-
algebra Jij0 . Explicitely one has:
X
Jij0 |ki = − ◦ ψ−r ψr ◦ ψ−1/2 |0ins = ψ−1/2 {ψ1/2 , ψ−1/2 }|0ins − ψ−1/2 {ψ1/2 , ψ−1/2 }|0ins
◦ i j ◦ k i j k j i k

r∈Z+1/2
 
= δjk ψi−1/2 − δik ψj−1/2 |0ins = δjk δi` − δik δj` ψ`−1/2 |0ins .


(4.128)

They are annihilated by all the positive modes of the currents. For instance
X
Jij1 |ki = − ◦ ψ1−r ψr ◦ ψ−1/2 |0ins = ψ1/2 {ψ1/2 , ψ−1/2 }|0ins + · · · = 0
◦ i j ◦ k i j k
(4.129)
r∈Z+1/2

Finally for the same reasons the states |ii are annihilated by all the positive Virasoro
modes:
X
∀n > 0 , Ln ψi−1/2 |0ins = 12 (r − n/2) ◦◦ ψjm−r ψjr ◦◦ ψi−1/2 |0ins = 0 , (4.130)
r∈Z+1/2

98
Free conformal field theories

and the conformal dimension is h = 1/2 as expected from eqn. (4.98):

X
L0 ψi−1/2 |0ins = 12 r ◦◦ ψj−r ψjr ◦◦ ψi−1/2 |0ins
r∈Z+1/2

= 41 ψj−1/2 {ψj1/2 , ψi−1/2 }|0ins + 14 ψj−1/2 {ψj1/2 , ψi−1/2 }|0ins

= 12 ψi−1/2 |0ins (4.131)

Descendant states in the vector representation are obtained by acting with the negative
modes of the currents Jij . A generic state is of the form

i2 j2 i2N j2N i1 j1 i1N j1N

|ψi = · · · J−2
1 1
· · · J−22 2 J−1
1 1
· · · J−11 1 ψi−1/2 |0ins . (4.132)
| {z }| {z }
N2 terms N1 terms

Its conformal dimension is given by:

h = N + 1/2 , (4.133)

with the level N defined in eqn. (4.126).

Spinorial representations
We focus here to the case where the number of fermions is even, namely M = 2k, k ∈ N,
which is the most relevant for string theory applications. Spin(2k), the double-cover of
SO(2k), admits two spinorial representations of dimensions 2k−1 each; we will refer to them
as the spinor and conjugate spinor respectively.
We introduce an affine primary operator SA (z) in a (reducible) Dirac representation that
is characterized by its OPE with the currents Jij :
1 1  i j B
Jij (z 0 )SA (z) = γ , γ A SB (z) + reg. , (4.134)
z0 − z 4
with {γi , i = 1, . . . , 2k} the gamma matrices in 2k Euclidean dimensions. This definition
follows from the general OPE (4.116), as the rotation generators in the reducible Dirac
representation are given by − 14 [γi , γj ]. Such operator is called a spin field.
Given the expression (4.83) for the currents in terms of the ψi , the OPE (4.134) between
the currents and the spin field actually comes from a more fundamental OPE between the
fermionic fields ψi and the spin fields SA :
1
ψi (z 0 )SA (z) = p (γi )AB SB (z) + O(z1/2 ) . (4.135)
2(z 0 − z)

To show that this OPE implies the OPE (4.134) is actually not trivial, as one cannot use the
Wick theorem (4.17) because of the branch cuts in the complex plane created by the spin
field. Rather one should use a technique known as bosonization.

99
Free conformal field theories

In the same way as a Dirac spinor can be decomposed into a pair of Weyl spinors trans-
forming in a irreducible representation of the SO(2k), one can split the spin field SA (z)
into Sα (z) and Sα̇ (z) of positive and negative chirality respectively. Chiral spinors are dis-
tinguished by their eigenvalue with respect to the matrix γ2k+1 = γ1 · · · γ2k , which anti-
commutes with all Dirac matrices and commutes with the rotation generators 14 [γi , γj ]. We
define formally an operator Γ with the same property, which anti-commutes with all the
modes of the operator ψi :
∀r , ∀i , {Γ, ψir } = 0 . (4.136)
The operator Γ is usually written in terms of the Fermion number operator F as

Γ = eiπF , (4.137)

where F, which counts the number of fermionic operators, is only defined mod two by this
relation.
By inserting one of the spin fields Sα (z) and Sα̇ (z) corresponding to the two irreducible
spinorial representations of SO(2k) at the origin of the complex plane, one defines states
called the Ramond vacua of the theory through the state-operator correspondence:

Sα (0) ↔ |αir , Sα̇ (0) ↔ |α̇ir . (4.138)

These two vacua are distinguished by their eigenvalue under Γ :

Γ |αir = |αir , Γ |α̇ir = −|α̇ir (4.139)

The Hilbert spaces built upon these Ramond vacua are called the Ramond sector of the CFT.
From equation (4.121) the dimension of conformal primaries in the spinor or conjugate
spinor representations, hence of the two Ramond vacua, are given by
cR M
h= = . (4.140)
2(M − 1) 16
The Laurent modes ψin of the fermionic operators ψi on the plane were defined in the
Neveu-Schwarz vacuum of the theory by (4.105), with n ∈ Z+ 12 . It is clear that such definition
cannot make sense for the action of fermionic modes on a Ramond vacuum, because of the
branch cut in the fundamental OPE (4.135). In the Ramond sector one should actually define
the modes in the same way but for n ∈ Z:
I
i dz n−1/2 i
∀n ∈ Z , ψn = z ψ (z) , (4.141)
2iπ
where implicitly these operators act on states in a Ramond sector of the free-fermion CFT;
the commutation relations between these modes is the same as (4.107) as the steps of the
derivation are similar.
Unlike in the Neveu-Schwarz sector, the fermionic operators have zero modes {ψi0 }. From
eqn. (4.107) these zero modes obey the anti-commutation relations

{ψi0 , ψj0 } = δij , (4.142)

100
Free conformal field theories

which is the same, up to a normalization factor, as the Clifford algebra {γi , γj } = 2δij .
The modes {ψin , n ∈ Z} are by definition associated with the decomposition of the oper-
ators ψi (z) in the Ramond sectors as:
X ψi
n
ψi (z) = n+1/2
. (4.143)
n∈Z
z
It means that two-dimensional fermions in the Ramond sector, on the complex plane, satisfy
ψr (e2iπ z) = −ψr (z) . (4.144)
Moving to the cylinder, using the conformal transformation (4.103), one finds that the Ra-
mond sector corresponds to periodic boundary conditions along the spatial circle:
ψr (σ1 + 2π, σ2 ) = ψr (σ1 , σ2 ) . (4.145)
In summary we have
 
ψns (w + 2π) = −ψns (w) z = e−iw ψns (e2iπ z) = ψns (z)
cylinder: ←−−−−−−−→ plane:
ψr (w + 2π) = ψr (w) ψr (e2iπ z) = −ψr (z)
(4.146)

4.3 The ghost CFT

The last free conformal field theory that we will present in this chapter is the conformal
field theory of reparametrization ghosts, which was introduced in section 2.4. We consider a
two-dimensional field theory with the following action in conformal gauge:
Z Z
1 2 z z z̄ z̄ 1 
¯ + b∂e

d2 z b∂c


S= d z bzz ∂ c + bz̄z̄ ∂ c = e c (4.147)
2π 2π
where all the fields are Grassmann variables.
The ghost action is by construction Weyl-invariant, which means that the ghost fields bzz ,
bz̄z̄ , cz and cz̄ are invariant under Weyl transformations (since no factors of the metric or its
inverse appear in (4.147)). Consider now a conformal transformation, made of a holomorphic
coordinate change z 7→ f(z), z̄ 7→ f̄(z̄) and a compensating Weyl transformation. The ghost
fields being invariant under the compensating Weyl transformation, conformal invariance of
the ghost action (4.147) implies that their conformal weights are determined solely from their
tensor structure, namely:
hb = 2 , hc = −1 and h̃eb = 2 , h̃ec = −1 . (4.148)
This theory has fields of negative weights hence is not unitary, which is not a problem for an
unphysical ghost action.4 We could have noticed it earlier, as we consider fields with integer
spins but fermionic statistics, hence violating the spin-statistics theorem.
4
Note that if we ”forget” about the tensor structure of the action (4.147), conformal invariance does
not fix completely the conformal dimensions of the fields. One has in fact a family of free CFTs with
hb + hc = h̃be + h̃ce = 1. The special case hb = hc = h̃be = h̃ce = 1/2 corresponds to a Dirac fermion, i.e. two
left-moving and two-right moving Majorana-Weyl fermions, in which case naturally the theory is unitary.

101
Free conformal field theories

We focus on the (b, c) ghost system from now (the (b, c) system being of the same nature),
e e
defined by the action Z
1 ¯
d2 z b∂c (4.149)
2π
The equations of motion are simply
¯ =0,
∂b ¯ = 0,
∂c (4.150)

Indicating that the ghost fields b(z) and c(z) are holomorphic. The action (4.149) has an
obvious classical rotational symmetry, at the infinitesimal level

c 7→ (1 − iδα)c , b 7→ (1 + iδα)b . (4.151)

The corresponding conserved Noether current is holomorphic:5

jg (z) = −b(z)c(z) , ¯ g (z) = 0 .

∂j (4.152)

We deduce from our favorite residue theorem (3.37) that

δα c(z1 ) = Resz2 →z1 (δα jg (z2 )c(z1 )) , δα b(z1 ) = −Resz2 →z1 (δα jg (z2 )b(z1 )) . (4.153)

We have then the following OPE

c(z1 )
jg (z2 )c(z1 ) = − •• bc •• (z2 )c(z1 ) = + reg. , (4.154)
z2 − z1
higher-order poles being forbidden by conformal invariance if we don’t allow for operators of
dimension more negative than the dimension of c. In the same way, one finds that the OPE

• b(z1 )
jg (z2 )b(z1 ) = • bc •• (z2 )b(z1 ) = − + reg. . (4.155)
z2 − z1
These two OPE can be deduced from the following OPE between the fundamental fields b
and c:
1
b(z1 )c(z2 ) = c(z1 )b(z2 ) = + reg. (4.156)
z1 − z2
taking into account that b and c anti-commute with each other.

4.3.1 Stress-energy tensor

We next turn to the stress-energy tensor and the central charge of this CFT. The operator
T (z) has to be such that b and c are conformal primary operators of conformal dimensions
hb = 2 and hc = −1 respectively. Starting from the elementary fields the more general ansatz
compatible with the symmetries of the problem is:

T (z) = λ1 •• b∂c •• (z) + λ2 •• ∂bc •• (z) . (4.157)

5
There is an anomaly of the ghost current on a curved worldsheet, which plays no role for the moment.

102
Free conformal field theories

We now compute the OPE

T (z 0 )c(z) = −λ1 ∂c(z 0 )b(z 0 )c(z) − λ2 c(z 0 )∂z 0 b(z 0 )c(z) + reg.
∂c(z) c(z) + (z 0 − z)∂c(z)
= −λ1 0 + λ2 + reg.
z −z (z 0 − z)2
c(z) ∂c(z)
= λ2 0 2
+ (λ2 − λ1 ) 0 + reg. (4.158)
(z − z) (z − z)

Hence λ2 = −1, λ1 = −2 which gives

T (z) = −2 •• b∂c •• (z) − •• ∂bc •• (z) . (4.159)

One checks that, as expected from the conformal weight of the primary field b(z):

T (z 0 )b(z) = −2 •• b∂c •• (z 0 )b(z) − •• ∂bc •• (z 0 )b(z) + reg.

= −2b(z 0 )∂z 0 c(z 0 )b(z) − ∂b(z 0 )c(z 0 )b(z) + reg.

b(z) + (z 0 − z)∂b(z) ∂b(z)
=2 − 0 + reg.
(z 0 − z)2 z −z
2b(z) ∂b(z)
= 0 + + reg. (4.160)
(z − z)2 z 0 − z

The central charge of the ghost CFT is computed from the self-OPE of T (z). One has

T (z 0 )T (z) = − 2 •• b∂c •• (z 0 ) − •• ∂bc •• (z 0 ) − 2 •• b∂c •• (z) − •• ∂bc •• (z) =

 
4 •• b(z 0 )∂c(z 0 )b(z)∂c(z) •• + •• b(z 0 )∂c(z 0 )b(z)∂c(z) •• + b(z 0 )∂c(z 0 )b(z)∂c(z)
 
• 0 0
+ • ∂b(z )c(z )∂b(z)c(z) •• • 0
+ ∂b(z )c(z
•
0
)∂b(z)c(z) •• 0 0
+ ∂b(z )c(z )∂b(z)c(z)
 
•
+2 • ∂b(z 0 )c(z 0 )b(z)∂c(z) •• + •• ∂b(z 0 )c(z 0 )b(z)∂c(z) •• + ∂b(z 0 )c(z 0 )b(z)∂c(z)
 
•
+2 • b(z 0 )∂c(z 0 )∂b(z)c(z) •• + •• b(z 0 )∂c(z 0 )∂b(z)c(z) •• + b(z 0 )∂c(z 0 )∂b(z)c(z)
(4.161)

which gives, after Taylor-expanding the remaining fields around z,

−13 2T (z) ∂T (z)

T (z 0 )T (z) = 0 4
+ 0 2
+ 0 + reg. (4.162)
(z − z) (z − z) z −z

Therefore the central charge of the ghost CFT is cg = −26. Naturally the computation is
the same for anti-holomorphic ghosts and we conclude that c̃g = −26.

103
Free conformal field theories

4.3.2 Ghost background charge

Finally we compute the OPE between the stress tensor T and the ghost current jg . One finds
that, surprinsingly, jg is not a conformal primary:

T (z 0 )jg (z) = 2 •• b∂c •• (z 0 ) + •• ∂bc •• (z 0 ) •• bc •• (z)
 
• 0 0
=2 • b(z )∂c(z )b(z)c(z) •• •
+ b(z )∂c(z
•
0 0
)b(z)c(z) •• 0 0
+ b(z )∂c(z )b(z)c(z)

+ •• ∂b(z 0 )c(z 0 )b(z)c(z) •• + •• ∂b(z 0 )c(z 0 )b(z)c(z) •• + ∂b(z 0 )c(z 0 )b(z)c(z)

3 jg (z) ∂jg (z)
=− 0 + + + reg. (4.163)
(z − z)3 (z 0 − z)2 z 0 − z

It means that, using eqn. (3.37)) once again, under an infinitesimal holomorphic transforma-
tion δz = ε(z), the current transforms as:

δε jg = 32 ∂2 ε(z) − jg (z)∂ε(z) − ε(z)∂jg (z) . (4.164)

For a finite conformal transformation z 7→ z 0 = f(z), this infinitesimal change exponentiates

to
3∂2 f(z)
jg0 (z 0 ) = (∂z f)−1 jg (z) + . (4.165)
(∂f(z))2
The current jg can be expanded in Laurent modes,
X jn
jg = , (4.166)
n∈Z
zn+1

and the OPE (4.163) implies the commuations relations:

[Lm , jn ] = −njm+n − 32 m(m + 1)δm+n,0 . (4.167)

The central extension implies in particular that

[L1 , j−1 ] = j0 − 3 (4.168a)

[j1 , L−1 ] = j0 . (4.168b)

Since the left-hand side of (4.168a) and the left-hand side of (4.168b) are Hermitian conjugates
to each other, we reach the conclusion

j†0 = j0 + Qb , Qb = −3 . (4.169)

This has an important consequence for the computation of correlation functions. Let us
consider a generic correlation function hO1 · · · On i of operators Oi of ghost charge Qi under
the current (4.152), computed in the ghost vacuum |0igh . Equation (4.169) means that the

104
Free conformal field theories

out vacuum gh h0| as a charge of Qb compared to the in vacuum |0igh , hence the correlation
function can be non-vanishing only if the charge conservation condition is modified to
X
n
Qi = Qb . (4.170)
i=1

The charge Qb is called a background charge, as it is independent of the operator insertions

in the path integral.
The presence of the background charge for the ghosts is related to an anomaly of the
ghost current conservation on curved worldsheets. One can show that
Qb
h∂i jig · · · i =
R[γ]h· · · i , (4.171)
2π
whose right-hand side integrates to Qb χ(s) = 6(g − 1).
The integrated anomaly gives the index counting the difference between the number of
b zero modes and of c zero modes, much as the integrated axial anomaly gives the index of
the Dirac operator of a charged fermion. These ghost zero-modes were already discussed in
the derivation of the Polyakov path integral, see subsection 2.3.3. As we discussed there, the
Riemann-Roch theorem (2.101) relates the difference between the number of moduli and of
conformal Killing vectors to the Euler characteristic of the surface, and each moduli (resp.
each conformal Killing vector) results in an insertion of a zero-mode of b and b e (resp. of c
and e
c) in the Polyakov path integral. This reasoning tells us that
#(b and b
e zero-modes) − #(c and e
c zero-modes) = −3χ(s) = 6(g − 1) . (4.172)
In the present case we consider the conformal field theory on the compactified complex plane
C̄ = C∩{∞}, which is conformal to a two-sphere, hence g = 0 and this counting is compatible
with eqn. (4.169).
In a sense, the curvature of the two-sphere after the conformal mapping to the compact-
ified complex plane has support only at ∞, i.e. at the south pole. Remember that, in radial
quantization, the in vacuum was mapped to the origin of the plane. In the same way, the out
vacuum is mapped to ∞ which which is where the curvature, hence the background charge
of the ghost current, is localized. This explains the origin of the relation (4.169).

4.3.3 Mode expansions and Hilbert space

Despite their fermionic nature, the ghost fields have integer spins hence have periodic bound-
ary conditions on the cylinder. After the conformal mapping to the complex plane, the
Fourier series expansion becomes a decomposition in Laurent modes with integer powers:
X bn X cn
b(z) = n+2
, c(z) = n−1
, (4.173)
n∈Z
z n∈Z
z

The OPE (4.156) results in the following non-trivial anti-commutation relations:

{bm , cn } = δm+n,0 . (4.174)

105
Free conformal field theories

In particular the ghost CFT has zero modes b0 , c0 which obey the algebra

{b0 , c0 } = 1 . (4.175)

We build the Hilbert space of the ghost CFT from ground states that are annihilated by
all the positive modes of the ghost operators, {(bn , cn ), n ∈ Z>0 }. Because of the zero-mode
algebra (4.175) we have two such ground states that satisfy the relations:

b0 |−igh = 0 , c0 |−igh = |+igh

c0 |+igh = 0 , b0 |+igh = |−igh . (4.176)

It is natural to select the vacuum state in the ghost sector as

|0igh = |−i , (4.177)

which is annihilated by the ghost zero-mode b0 . Indeed the path integral of the Polyakov ac-
tion on the sphere, see eqn. (2.122) has insertions of zero-modes (c0 , e
c0 ) but not of zero-modes
(b0 , b0 ), given that the sphere has no moduli. We will have a more accurate explanation in
e
the next chapter.
The ghost Hilbert space is then obtained by acting upon the vacuum state |0igh with c0
and the creation operators {(b−n , c−n ), n ∈ Z>0 }. The description of the Hilbert space of the
anti-holomorphic ghosts (b, c) is exactly the same.
e e
The stress energy tensor (4.159) can be decomposed in terms of the modes of the ghost
fields by the methods used before for the scalar and fermionic CFTs. For n 6= 0 there are no
normal ordering ambiguities and one can write
I
dz n+1 X  −m−2 −r ◦ ◦ −m−3 −r+1 ◦ ◦

∀n 6= 0 , Ln = z 2(r − 1)z z ◦ bm cr ◦ + (m + 2)z z ◦ bm cr ◦
2iπ m,r∈Z
X
= (2n − m) ◦◦ bm cn−m ◦◦ . (4.178)
m∈Z

For L0 the normal ordering of the positive and negative modes could lead to a constant term,
exactly as for the Hamiltonian of the ordinary harmonic oscillator. Fixing this ambiguity
is important since it gives the conformal dimension of the vacuum. One can use the same
contour manipulations as in section 4.1, or instead use the commutation relation
1 1
L0 |−igh = [L1 , L−1 ]|−igh = (−2b0 c1 b−1 c0 + · · · )|−igh = −bo {c1 , b−1 }|+igh = −|−igh .
2 2
(4.179)
Hence X
L0 = − m ◦◦ bm c−m ◦◦ − 1 . (4.180)
m∈Z

One important fact for later usage is that the ghost vacuum has conformal dimension minus
one.

106
Chapter 5

The string spectrum

107
The string spectrum

In this chapter we consider bosonic string theory in Minkowski space-time, i.e. the
Polyakov path integral (2.114) describing the propagation of a single bosonic string in a
flat target space-time with metric Gµν = ηµν , together with a vanishing Kalb-Ramond field
and constant dilaton.
We will leverage on the techniques developed in the previous chapters to obtain the
first important predictions of string theory, that its spectrum contains the graviton, thereby
providing a finite quantized theory of gravity, and that the dimension of space-time is fixed
to a critical dimension because of the Weyl anomaly derived in section 3.6. We will find
also that the bosonic string theory that we have dealt with up to now has a fatal flaw, the
presence of an instable mode or tachyon, that will be taken care of in the next chapter by
introducing supersymmetry on the two-dimensional worldsheet.

5.1 String theory and the dimension of space-time

In classical field theories, the dimension of space-time is a parameter of the theory, which
is not fixed a priori. In quantum field theories, the situation is a little bit different as the
renormalizability of the theory depends on the dimensionality of space-time; for instance
gauge theories in more than four dimensions are non-renormalizable – which is not a problem
if one views the theory only as a low-energy effective theory with a UV completion making
sense at high energies.
In string theory the question of the dimension of space-time arises already at tree-level,
i.e. at the level of the consistency of the Polyakov path-integral (2.114) on the sphere. This
path integral only makes sense if all the gauge symmetries of the two-dimensional quantum
field theory are non-anomalous.

5.1.1 Critical dimension of space-time

In section 3.6 we have explained that the Weyl symmetry, which is part of the gauge sym-
metry of the Polyakov action, is actually potentially anomalous; the anomaly was given by
eqn. (3.102), and was proportional to the central charge of the worldsheet conformal field
theory.
Let us consider bosonic string theory in a Minwkoskian target-space R1,D−1 . The under-
lying two-dimensional conformal field theory is the product of two free theories:
• the conformal field theory for D free scalar fields, see eqn. (4.1). Following the analysis
in section 4.1, its left and right central charges are (c, c̄) = (D, D).

• the conformal field theory of reparametrization ghosts (b, c) and (b,

e ec), see eqn. (4.147).
Following the analysis in section 4.3, its left and right central charges are (c, c̄) =
(−26, −26).
Therefore the cancellation of the Weyl anomaly is only possible if D = 26. In other words:

Bosonic string theory in Minkowski space-time is only well-defined in 26 dimensions

108
The string spectrum

This is a rather bold statement, which is unavoidable is one asks that the quantum theory
living on the worldsheet of the string is a consistent one. There are different attitudes towards
this result that we can summarize as follows:
1. This is a problem. All experimental facts points towards a four-dimensional space-time.

2. This is a feature of the theory, and we should live with it. As we shall see, if the extra
dimensions are small enough, there is no blatant contradiction with experiments.

3. This is a desirable feature, as it introduces some flexibility in an otherwise quite rigid

theory, in order to match experimental facts.
One may notice that strictly speaking this result is valid in Minkowski space-time with
constant dilaton, and in fact it could be evaded by considering space-times with a strong
space-like dilaton gradient, see eqn. (5.4c) below; this is a non-critical string. However such
space-time cannot be used as a starting point for realistic physics and exhibits a strong-
coupling singularity.
Finally, some of you may have heard that string theory predicts a ten-dimensional or
eleven-dimensional space-time. For the former, we need to introduce superstring theory,
which will wait till the next chapter. For the latter, the route is much longer and this is
probably out of sight for this year.

5.1.2 Beyond Minkowski: string theory in curved space-time

So far the analysis of bosonic string theory was concerned with string propagation in a
Minkowski space-time. In two-dimensional terms, the ”matter part” of the Polyakov action
corresponds then to a set of D = 26 free scalar field, one having a kinetic term with the
wrong sign.
One could actually be more general, and consider a string moving in a generic curved
space-time. This is actually given by eqn. (2.79), that we reproduce here for convenience:
Z
1 2
p 
S= dσ det γ γ Gµν [x ] +  Bµν [x ] ∂i xµ ∂j xν
ij ρ ij ρ
4πα 0 s
Z
1 p
+ d2 σ det γ Φ[xµ ]R[γ] . (5.1)
4π
This is an interacting QFT, as the background fields Gµν , Bµν and Φ are functionals of the
fields xρ (σi ).
This action defines a consistent theory if, at the very least, it defines a two-dimensional
conformal field theory with (c, c̄) = (26, 26). At the classical level, as we have explained in
detail in chapter 2, the first two terms of the action (5.1) are invariant under two-dimensional
diffeomorphisms and Weyl transformations, while the last term is only Weyl-invariant for
constant dilaton, Φ[xρ ] = Φ0 .
At the quantum level, the theory is interacting and should be renormalized. This type
of quantum field theory is known as a non-linear sigma-model, and has many applications

109
The string spectrum

outside of string theory. To view this one can decompose the fields as xµ (σi ) = xµ0 + δxµ (σi ),
expand the action (5.1) in powers of δxµ and renormalize the interacting theory.
The computation of the Weyl anomaly in section 3.6 was assuming that the worldsheet
two-dimensional QFT was conformal. In the present situation we don’t know yet if the
conformal symmetry is preserved at the quantum level, and the variation of the renormalized
theory under an infinitesimal Weyl transformation is expressed, using the corresponding Ward
identity, as: Z p
− 2π d2 σ det γ hδωT aa · · · i = δδω h· · · i
1
(5.2)

It has been shown [1] by computing the one-loop diagrams in this interacting quantum field
theory that the trace of the stress-energy tensor obeys the operator equation (which as usual
should hold in any correlation function)

T aa = − 2α1 0 γij βG ij B µ ν 1 Φ


µν + i βµν ∂i x ∂j x − 2 β R[γ], (5.3)

with coefficients
0 0 α0
βG
µν = α Rµν + 2α ∇µ ∇ν Φ − 4
Hµρσ Hν ρσ + O(α0 2 ) , (5.4a)
0
βBµν = α 0 ∇ρ Hρµν + O(α0 2 ) ,
− α2 ∇ρ Hρµν + (5.4b)
0 α0
β =Φ 1
6
(D − 26) − α2 ∆Φ + α 0 ∇ρ Φ∇ρ Φ − 24
Hµνρ Hµνρ + O(α0 2 ) . (5.4c)

The expansion here should be thought as a weak-field expansion from the target space point
of view. For instance if space-time has Ricci curvature of order 1/L2 , the expansion is in
powers of α 0 /L2 , i.e. a weak-curvature expansion.
The condition for Weyl invariance of string theory in the background specified by (G, B, Φ)
is therefore given by the cancellation of all these coefficients:

βG B Φ
µν = βµν = β = 0 . (5.5)

One can view these conditions as the equations of motion coming from the following D-
dimensional action:
Z
√
 
1 D −2Φ 2(D − 26) µνρ µ
S = 2 d x − det Ge − + R[G] − 12 Hµνρ H + 4∂µ Φ∂ Φ (5.6)
2κ 3α 0

i.e. general relativity coupled to the Kalb-Ramond two-form and the dilaton Φ, with a po-
tential term for the dilaton proportionnal to the flat space Weyl anomaly. One can actually
view the action (5.6) as the low-energy effective action for bosonic string theory with back-
ground fields (G, B, Φ). There is an important caveat though: the bosonic string includes
also a tachyon field that couldn’t be ignored in such description; however in the superstring
theories the tachyon is gone and the supersymmetric analogue of (5.6) will be a faithful
effective action, from which the low-energy physics can be reproduced.
We have seen in section 2.3, eqn. (2.88), that (the exponential of) a constant dilaton
Φ[xρ ] = Φ0 played the role of the coupling constant in D dimension. Here we see that the

110
The string spectrum

dilaton, which can be a non-trivial function of the space-time coordinates, can be thought of
a ”dynamical” coupling constant.
The equations (5.4) tell us that one could in principle define bosonic strings away from
twenty-six dimensions and even directly in four dimensions, however the price to pay is, for
instance, a very large gradient of the dilaton field which would induce a strong violation of
Lorentz invariance. One simple example is to take a Minkowski space-time with
26 − D
Φ[xµ ] = Vµ xµ , V µ Vµ = . (5.7)
α0
Besides the anisotropic nature of space-time for D < 26 due to the space-like gradient, such
background has a singularity for the dilaton hence for the coupling constant.
The coefficients β in eqn. (5.4) are actually, as their name suggests, the beta-functions of
the corresponding coupling constants in the two-dimensional QFT. Demanding the vanishing
of the beta-function means that the theory is at a fixed point of the renormalization flow.
This implies scale invariance of the theory, and it can be shown that, under mild assumptions,
it implies conformal invariance as well [2].
To be more precise, for a two-dimensional non-linear sigma-model on the plane, a solution
of the equations βG B
µν = 0 and βµν = 0, without the dilaton terms, provide a quantum
field theory which is scale-invariant – hence conformal invariant – at one (worldsheet) loop.
When such theory is put on a curved manifold, it couples to the background metric, and
one needs in particular to include the dilaton coupling to the two-dimensional curvature in
the renormalization of the theory. Then the vanishing of the conformal anomaly demands
also that βΦ = 0. This equation is rather interesting, since it indicates that the one-loop
conformal anomaly is balanced by the variation of a classically anomalous term, since the
dilaton coupling is not classically Weyl invariant unless Φ is constant, see the discussion
below equation (2.85b).

5.2 BRST quantization

At this stage, we have understood that bosonic string theory in Minkowski space-time is
described by the two-dimensional action
Z Z
1 2 µ¯ ν 1 2

¯ + b∂e

S= d z η µν ∂x ∂x + d z b ∂c e c . (5.8)
4πα 0 2π
We have also described in chapter 4 the Hilbert space of each of these two conformal field
theories – the free scalar CFT and the ghost CFT. Clearly the physical spectrum of string
theory cannot be simply the tensor product of these two Hilbert spaces. Both conformal field
theories are non-unitary, the former because of the field x0 (z, z̄) with the wrong-sign kinetic
term and the latter because it violates the spin-statistic theorem.
The key point is that the Polyakov action (2.63) corresponds actually to a constrained
system; even after gauge fixing one has to impose the equations of motion for the gauge field
(the metric), namely that
Tij = 0 . (5.9)

111
The string spectrum

An elegant way of quantizing constrained systems is to use the BRST formalism (for Becchi,
Rouet, Stora and Tyutin), that provides a selection rule from a ”remnant” global symmetry
in the gauge-fixed action for the matter and for the ghosts, named the BRST symmetry.

5.2.1 Basics of BRST quantization

Let us start with an overview of BRST quantization at a very general level. Consider a field
theory with a gauge symmetry acting on the fields φi ( including the gauge field themselves),
obeying the algebra
[δu , δv ] = fwuv δw . (5.10)
The path integral over the gauge fields is transformed into an integral gauge-inequivalent
configurations by foliating the field space into gauge orbits and introducing gauge-fixing
conditions
f` [φi ] = 0 (5.11)
As described in detail in chapter 2, this is done by introducing the Faddeev-Popov deter-
minant, defined through a functional integral over gauge transformations, parametrized by
Λ: Z Z R ` Λ i
1 = ∆fp DΛ δ f` [φ ] = ∆fp DΛDB` ei B f` [φ ] ,
Λ i


(5.12)

where we have introduced a path integral representation of the Dirac distribution using
bosonic Lagrange multipliers B` . The gauge-fixed path integral is given by
Z  Z Z 
DφDBDbDc exp −S[φ] + i B` f [φ] − b` c δu f [φ] .
` u `
(5.13)

Here the fermionic ghosts (b` , cu ) provide a path integral representation of the Jacobian in
field space and the last term contains the variation δu f` [φ] of the gauge-fixing condition under
an infinitesimal gauge transformation.

BRST symmetry
The gauge-fixed path integral has a remnant global symmetry, which acts on the different
fields as

δb φi = −i cu δu φi (5.14a)
δb B` = 0 (5.14b)
δb b` = B` (5.14c)
δb cu = i 12 fuvw cv cw . (5.14d)

The first line, eqn. (5.14a), can be understood as a gauge transformation of parameter cu .
Since the ghost fields are fermions, the parameter  of the BRST transformation is a Grass-
mann variable.

112
The string spectrum

Gauge invariance of the original action S[φ] guarantees its invariance under the transfor-
mation (5.14a). The second term in eqn. (5.13) transforms as
Z Z Z
δf`
δb i B` f [φ] = i B` i δb φ =  B` cu δu f` [φ] .
` i
(5.15)
δφ
Finally the last term transforms as
Z Z Z Z
u ` u `
δb b` c δu f [φ] =  B` c δu f [φ] − i b` 2 f vw c c δu f [φ] − i b` cu cv δv δu f` [φ]
1 u v w `

Z Z Z
=  B` c δu f [φ] − i b` 2 f vw c c δu f [φ] + i b` cu cv 21 [δu , δv ]f` [φ]
u ` 1 u v w `

Z
=  B` cu δu f` [φ] , (5.16)

which cancels precisely (5.15). To this symmetry we associate a fermionic Noether current
jb , the BRST current, such that
Z
1
δS = dD x jib ∂i  , (5.17)
2π
and a corresponding BRST charge
Z
Qb = dΣi jib . (5.18)
Σ

A crucial property of this symmetry is its nilpotence, i.e. that two successive transfor-
mations cancel each other. One finds first that
δη δ φi = i δη (cu δu φi ) = η 12 fuvw cv cw δu φi − cu cv δu δv φi = 0

δη δ B` = 0
δη δ b` = −δη B` = 0
δη δ cu = −i 12 fuvw δη (cv cw ) = 12 ηfuvw fvxy cx cy cw = 0 , (5.19)
where in the last line we have used the Jacobi identity satisfied by the generators of the Lie
algebra:
fuvw fvxy − fuvx fvwy − fuvy fvxw = 0 . (5.20)
Equivalently, the operator Qb associated with the BRST charge (5.18) satisfies
Qb2 = 0 . (5.21)
Finally another important property of the BRST symmetry is the following relation:
 Z   Z Z
` u ` i ` i
δ i b` f [φ] =  − b` c δu f [φ ] + i B` f [φ ] , (5.22)

i.e. giving exactly the gauge-fixing term and the ghost action in (5.13). This is of course
another way of checking the BRST-invariance of the action, using the nilpotence of the BRST
transformation.

113
The string spectrum

BRST constraints
The properties of the BRST symmetry provide crucial information about the space of physical
states of the theory. Let us consider the path integral giving the transition amplitude between
an initial state at t = t0 given by the wave functional Ψ0 [θ0 (σ)] and the final state at t = t1
given by the wave functional Ψ1 [θ1 (σ)] (here by θ we denote collectively the fields, including
the ghosts and by σ the spatial coordinates):
Z Z Z θ(σ,t1 )=θ1 (σ)
hΨ1 (t1 )|Ψ0 (t0 )i = Dθ1 Dθ0 Ψ1 [θ1 (σ)] Ψ0 [θ0 (σ)]
?
Dθ e−S[θ] . (5.23)
θ(σ,t0 )=θ0 (σ)

The path integral should be invariant under a change of the gauge-fixing conditions, f` [φi ] 7→
f` [φi ] + δf` [φi ]. This implies, using equation (5.22), that:
Z Z Z θ(σ,t1 )=θ1 (σ) Z 
∀δf ,`
Dθ1 Dθ0 Ψ1 [θ1 (σ)] Ψ0 [θ0 (σ)]
?
Dθ e−S[θ]
δb `
b` δf [φ] = 0 . (5.24)
θ(σ,t0 )=θ0 (σ)

Or, in Hamiltonian formalism,

∀δf` , hΨ1 |δB b` δf` [φ] |Ψ0 i = 0 ,


(5.25)
for every pair of physical states. The action of the BRST transformation on the operator
b` δf` [φ] can be expressed as the anti-commutator with the fermonic BRST charge (5.18), so
we have the condition 
∀δf` , hΨ1 | Qb , b` δf` |Ψ0 i = 0 . (5.26)
Assuming that the operator Qb is self-adjoint (it follows from the reality of the underlying
gauge transformation), one finds that, since this equality should hold for any δf` , the BRST
charge annihilates every physical state:
∀|Ψi ∈ Hphys , Qb |Ψi = 0 . (5.27)
We consider now a state |Ψs i obtained from the action of the BRST charge on some state
|ϕi:
|Ψs i = Qb |ϕi , (5.28)
which is annihilated by Qb by nilpotence of the BRST operator. This state is orthogonal to
all physical states, and orthogonal to itself:
∀|Ψi , hΨ|Ψs i = 0 , hΨs |Ψs i = 0 . (5.29)
This means that if |Ψi is a physical state, the state
|Ψi + |Ψs i = |Ψi + Qb |ϕi (5.30)
is also a physical state, which has the same overlaps with all physical states as the original
state |Ψi, hence is indistinguishable from the former. Thus one can endow the space states
with an equivalence relation:
|Ψi ∼ |Ψ 0 i ⇔ ∃|ϕi , |Ψ 0 i = |Ψi + Qb |ϕi , (5.31)
and the physical state space of the theory is given by the space of equivalence classes.

114
The string spectrum

Ghost number, grading of state space and BRST cohomology

One defines the ghost number as the additive charge such that cu has charge one, b` charge
minus one, φi and B` are neutral.
This ghost charge assignment induces a global Abelian symmetry of the action in (5.13),
hence one define a conserved ghost number Noether current and a corresponding ghost charge;
we have already discussed such current in the CFT context in section 4.3, see around equa-
tion (4.152).
Consistency of the BRST transformations (5.14) indicate that we should assign to the
parameter  a ghost charge of minus one. Hence the BRST charge operator Qb has itself a
ghost charge of plus one. In other words we have a grading of the space of states H according
to the value qg of the ghost charge, as it admits the decomposition H = ⊕qg Hqg .
The operator Qb sends elements of Hqg to elements of Hqg +1 , defining the following
sequence of maps:

Qb Qb Qb Qb
· · · −−−−−→ Hqg −−−−−→ Hqg +1 −−−−−→ Hqg +2 −−−−−→ · · · (5.32)
q
Since the BRST charge is a nilpotent operator, i.e. Q2b = 0, the image of every map Qbg
defined by:
q
Qbg : Hqg → Hqg +1
|ϕiqg 7→ Qb |ϕiqg (5.33)
q +1
is in the kernel of the map Qbg . In this case, a sequence of maps like (5.32) is called an
exact sequence.
The component of charge qg of the space of physical state is then defined, following the
discussion in the previous paragraphs, as 1
q
q Ker(Qbg )
Hphys
g
= q −1
. (5.34)
Im(Qbg )

and the full Hilbert space of physical states is defined by

M q
Hphys = Hphys
g
. (5.35)
qg

To phrase the same statement differently, the space of physical states is defined as the
quotient of the space of Qb -closed states, which is the space of states annihilated by the
BRST charge:
|Ψi ∈ Hclosed ⇔ Qb |Ψi = 0 (5.36)
1
This structure is similar to the de Rahm cohomology of a n-dimensional differentiable manifold, where
the space of differential forms Ω is graded according to their degree as Ω = ⊕n k=0 Ωk , and the exterior
derivative d, which is nilpotent, induces a sequence of maps dk : Ωk → Ωk+1 , ωk 7→ dωk .

115
The string spectrum

by the space of Qb -exact states, that are obtained by the action of the BRST charge on
another state:
|Ψi ∈ Hexact ⇔ ∃|ϕi , |Ψi = Qb |ϕi (5.37)
So
Hclosed
Hphys = (5.38)
Hexact

5.2.2 BRST symmetry of the point-particle action

We start by illustrating the ideas behind BRST quantization with the simpler theory of a
point particle already discussed in section 2.1.
In this case the gauge symmetry corresponds to diffeomorphisms of the worldline τ 7→
τ̃ = τ + α(τ), under which the einbein transforms as
∂τ d 
e(τ) 7→ ẽ(τ̃) = e(τ) = e(τ̃) − α(τ̃) e(τ̃) . (5.39)
∂τ̃ dτ̃
A basis for these infinitesimal diffeormorphisms is given by
δτ1 τ = δ(τ − τ1 ) , (5.40)
as a general infinitesimal diffeormorphism can be written as
Z
δτ = α(τ) = dτ1 δ(τ − τ1 )α(τ1 ) , (5.41)

and the transformations of the fields in this basis are given by

δτ1 xµ (τ) = −δ(τ − τ1 )ẋµ (τ) (5.42a)
d 
δτ1 e(τ) = − δ(τ − τ1 )e(τ) . (5.42b)
dτ
We consider then the gauge fixing condition
F(x, e; T ) := T − e(τ) . (5.43)
After introducing the Fadeev-Popov determinant, one considers then the action, see
around eqn. (2.22):
Z1
m2 e
 
1 µ
S = dτ ẋ ẋµ + − iB(e − T ) − eḃc . (5.44)
0 2e 2
Unlike there, we have exponentiated the gauge-fixing constraint using the antighost Lagrange
multiplier field B. The last term is understood as
Z Z Z Z
d 
dτ b(τ) dτ1 c(τ1 ) δτ1 F(τ) = dτ b(τ) dτ1 c(τ1 ) δ(τ − τ1 )e(τ)
dτ
Z  Z Z
d
= − dτ b(τ) dτ1 c(τ1 )δ(τ − τ1 )e(τ) = − dτ e(τ)ḃ(τ)c(τ) , (5.45)
dτ

116
The string spectrum

where we have performed an integration by parts to reach the second line.

The algebra of gauge transformations acting on a field xµ (τ) is given in the present case
by:
 d d 
[δτ1 , δτ2 ]x (τ) = − δ(τ − τ2 ) δ(τ − τ1 ) − δ(τ − τ1 ) δ(τ − τ2 ) ẋµ (τ) .
µ
(5.46)
dτ dτ
The infinitesimal BRST transformations, given by the general formulæ (5.14), are defined
in the present context as:
δ xµ = i c ẋµ , (5.47a)
δ B = 0 (5.47b)
δ b = B , (5.47c)
δ c = i cċ . (5.47d)
For the last transformation we have used
Z Z
i  d d 
δ c(τ) = dτ1 dτ2 δ(τ−τ2 ) δ(τ−τ1 )−δ(τ−τ1 ) δ(τ−τ2 ) c(τ1 )c(τ2 ) = ic(τ)ċ(τ) .
2 dτ dτ
(5.48)
It is possible to reach a ”reduced” form of the BRST transformations by integrating out the
(neutral) field B to reach the gauge-fixed theory and replacing it in the transformation (5.47)
using the equation of motion for the einbein e,
δ1 µ m2 e 
B = −i ẋ ẋµ + − eḃc  . (5.49)

δe 2e 2 e=T

The right-hand side of this equation is nothing but the stress-tensor of the worldline theory.
One finds then that the gauge-fixed action
Z1
m2 T
 
1 µ
S = dτ ẋ ẋµ + − T ḃc (5.50)
0 2T 2
is invariant under the infinitesimal fermionic symmetry,
δ xµ = i c ẋµ , (5.51a)
 
m2 T
δ b = i T − 2T1 ẋµ ẋµ +
1
2
− T ḃc , (5.51b)
δ c = i cċ , (5.51c)
which is nilpotent as expected. However it is now true only on-shell, i.e. using the equations
of motion.

5.2.3 BRST symmetry of bosonic string theory

We consider now the BRST symmetry associated with the gauge-fixed Polyakov path integral
of bosonic string theory. It will be similar in many respects to the point particle case and
will heavily use the methods of conformal field theories.

117
The string spectrum

Here the gauge symmetry corresponds to diffeomorphisms of the worldsheet and Weyl
transformations, and the ”reduced” BRST transformations can be expressed as follows:
¯ µ ,
δ xµ = i c∂xµ + c̃∂x


(5.52a)
 1 
δ b = i − 0 •• ∂xµ ∂xµ •• − 2 •• b∂c •• − •• ∂bc •• = i(T x + T g ) , (5.52b)
α
e = i − 1 • ∂x
 
δ b ¯ µ ∂x
¯ µ• − 2•b ¯c • − • ∂
e∂e ¯ be
ec • = i(T̃ x + T̃ g ) , (5.52c)
α0 • • • • • •

δ c = ic∂c (5.52d)
δ e
c = ie¯c .
c∂e (5.52e)

As in the point-particle case, the transformation of the ghosts b and b

e is given respectively
by the holomorphic and anti-holomorphic components of the stress-energy tensor, as the fields
B and B̃ have been eliminated using the equation of motion for the two-dimensional metric.
Note that the ghost corresponding to the Weyl symmetry was just a Lagrange multiplier field
and has been already integrated out, see eqn. (2.120) and below (which results in the tensor
bab being traceless).
To the BRST symmetry one associates a conserved fermionic Noether current, whose
components in complex coordinates are separately conserved, being respectively holomorphic
and anti-holomorphic. They are of the form
•
jb = cT x + 1
2 •cT g •• + λ∂2 c , (5.53a)
cTex +
eb = e 1 • eg •
cT • + λ∂ ¯2ec, (5.53b)
2 •e

where T x , Tex and T g , Teg denote respectively the stress-energy tensors of the x fields and of
the ghost fields. We have indeed the OPEs:

µ x

µ c(0)∂xµ (0, 0)
jb (z)x (0, 0) = c(z)T (z) + · · · x (0, 0) = + reg. (5.54a)
z
1 1
jb (z)b(0) = c(z)T x (z) + 12 •• cT g •• (z) + λ∂2 c(z) b(0) = · · · + T x (0) + T g (0)



z 2
1
 2b(0) ∂b(0) 
+ c(0) + z∂c(0) + · · · + + reg. (5.54b)
2 z2 z
1
 −c(0) ∂c(0) 
jb (z)c(0) = 12 •• cT g •• (z)c(0) = c(0) + z∂c(0) + · · · + + reg. , (5.54c)
2 z2 z
and their anti-holomorphic counterparts, from which we recover the action H of the BRST
1
charge, hence the transformations (5.52), by the contour integral Qb = 2iπ (dz jb − dz̄̃b ).
Notice that the last term in the holomorphic BRST current (5.53a) is a total derivative,
that did not contribute to the BRST charge as computed above. It is however necessary to
add this improvement term, for a certain value of λ, in order for the current jb to transform
as a tensor of dimension (1, 0) at it should.

118
The string spectrum

To see this, one starts by computing the following operator product expansions, by taking
derivatives of the fundamental OPE (4.156):

• 2 6
T g (z)∂2 c(0) = • (−2b∂c − ∂bc) •• (z) ∂2 c(0) = 2∂c(z)
3
− c(z) 4 + reg.
z z
−6 4−6 4−3 2 2−1 3
= 4 c(0) + 3 ∂c(0) + 2 ∂ c(0) + ∂ c(0) + reg.
z z z z
−6 −2 1 1
= 4 c(0) + 3 ∂c(0) + 2 ∂2 c(0) + ∂3 c(0) + reg. (5.55)
z z z z
•
Noticing that • cT g •• = •
• c(−2b∂c − ∂bc) •• = 2 •• bc∂c •• , one computes the following OPE:

T g (z) •• cT g (0) •• = •
• T g (z)c(0)T g (0) •• + •• T g (z)c(0)T g (0) ••

− 4 •• b∂c •• (z) •• bc∂c •• (0) − 4 •• b∂c •• (z) •• bc∂c •• (0)

− 2 •• ∂bc •• (z) •• bc∂c •• (0) − 2 •• ∂bc •• (z) •• bc∂c •• (0)

•
 c(0) ∂c(0) 
• • −26
 2T g (0) ∂T g (0) 
= • − 2 + g
T (0) • + • 4
+ 2
+ c(0) ••
z z 2z z z
8 6
− 4 c(0) + 3 ∂c(0) (5.56)
z z
One gets finally the following result for the OPE between the full stress-tensor and the
BRST current:

T (z)jb (0) = T x (z) + T g (z) cT x (0) + 12 •• cT g •• + λ∂2 c(0)

D 2T x (0) ∂T x (0)   c(0) ∂c(0)  x

= c(0) + + + − 2 + T (0)
2z4 z2 z z z
1  −26 2T g (0) ∂T g (0)  1  c(0) ∂c(0) 
+ c(0) + + + − 2 + T g (0)
2 2z4 z2 z 2 z z
1  8c(0) 6∂c(0)   −6c(0) −2∂c(0) ∂2 c(0) ∂3 c(0) 
+ − + + λ + + +
2 z4 z3 z4 z3 z2 z
jb (0) ∂jb (0) c(0) ∂c(0)
= 2 + + 4 (D/2 − 6λ − 4) + (3 − 2λ) . (5.57)
z z z z3
The term in ∂c(0) signals that jb does not transform as a tensor, hence one should set
λ = 3/2. One sees furthermore that the term in 1/z4 is absent for D = 26, i.e. when the
string propagates in 26 dimensions as was already noticed by other methods.
To understand better the significance of the last condition in the context of BRST quan-
tization, one can compute the OPE of the ghost current with itself. Using similar methods
as for the previous computation, after some slightly tedious algebra one gets the result
18 − D • • 18 − D • 2 • 26 − D •
jb (z)jb (0) = • c∂c • (0) + • c∂ c • (0) + c∂c •• (0) . (5.58)
2z 3 4z 2 12z •

119
The string spectrum

Given that the holomorphic BRST charge of a local operator O is given, following the general
result (3.16), by the contour integral
I
dz
Qb = jb (z)O(0) , (5.59)
C0 2iπ

The simple pole in equation (5.58) indicates that the BRST current is transformed under a
BRST transformation. In other words, the BRST operator is not nilpotent unless D = 26:

{Qb , Qb } = 0 only if D = 26 . (5.60)

Naturally, the same computation can be done using the anti-holomorphic BRST current eg ,
leading to the same conclusion for the anti-holmorphic BRST charge Q e b.
More generally, the free action for D scalars on the string worldsheet can be replaced by
another more general conformal field theory of central charges (c, c̄), and the condition for
BRST invariance becomes c = c̄ = 26.

5.2.4 BRST cohomology of the bosonic string

We are now ready – at least! – to provide the description of the physical states of bosonic
string theory. Let us consider the local operator corresponding to a physical state. It splits
generically into a matter and ghost parts, as the CFT on the worldsheet is the tensor product
of the two factors:
V(z, z̄) = Vg (z, z̄)Vx (z, z̄) . (5.61)
We have seen in the discussion about the path integral on the sphere that an unintegrated
operator for the matter CFT should be multiplied by an insertion of the c and e c ghosts at
the operator position. One considers therefore that the physical states correspond in their
unintegrated version to local operators the form

Vu (z, z̄) = c(z)e

c(z̄)Vx (z, z̄) . (5.62)

According to the general discussion in subsection 5.2.1, a physical state should be first
BRST closed. Let us assume that the matter operator Vx (z, z̄) has conformal dimensions
(h, h̄), but is not necessarily a primary field. One has

x W(0) hVx (0) ∂Vx (0)

T (z)V (0) = · · · + + + + reg. (5.63)
z3 z2 z
where W(0) is some local operator. Ignoring the possible effect of more singular terms, one

120
The string spectrum

considers the OPE

 1• g• 
x
jb (z)Vu (0) = c(z)T (z) + • cT • (z) c(0)c̃(0)Vx (0, 0)
2
 1   
= c(0) + z∂c(0) + z2 ∂2 c(0) c(0)c̃(0) T x (z)Vx (0, 0) + reg.
2
1  
+ c(0) + z∂c(0) T g (z)c(0) + reg. c̃(0)V x (0, 0)
2
1 1

c(0) 2 ∂c(0)W(0) + ∂2 c(0)W(0) + (h − 1)∂c(0) Vx (0, 0) + reg.
= c(0)e
z z
(5.64)

The first order pole provides the charge of the state under a BRST transformation, hence,
W should vanish and h should be equal to one. Generalizing easily this computation to more
singular terms in the T x V x OPE, and doing the same computation for the anti-holomorphic
BRST current eb , we have learned that an unintegrated physical operator should be:

1. a conformal primary operator,

2. an operator of conformal dimensions h = h̄ = 1.

We have also to check that it is not BRST-exact, i.e. not obtained as a BRST transformation
of another local operator; this will be done shortly.
The integrated operator is obtained by replacing the ghost insertion c(z)e c(z̄) by the inte-
gral over the worldsheet. In this case one can use
I I  x
hV (0, 0̄) ∂Vx (0, 0̄)

dz dz
[Qb , V ] =
x x
jb (z)V (0, 0̄) = c(z) 2
+ = h∂cVx + c∂Vx ,
2iπ 2iπ z z
(5.65)
hence it is a total derivative if h = 1 and the BRST variation integrates to zero. Note that
if Vx is non-primary operator it will bring extra terms in ∂n c, n > 1 and the statement will
not be true.
For a two-dimensional CFT, an integrated primary operator of conformal dimensions (1, 1)
is called a marginal deformation. If one starts with a conformal field theory with action S,
the quantum field theory with action
Z
Sλ = S + λ d2 z O1,1 (z, z̄) (5.66)

is also a conformal field theory. The parameter λ defines a family of conformal field theories,
or said differently a one-parameter family of renormalization group fixed points. In general
the parameters that characterize locally families of conformal field theories are viewed as
coordinates on a manifold called the moduli space of the conformal field theory.

121
The string spectrum

Physical states
We will now discuss the BRST constraints from the point of view of physical states, rather
than from the point of view of local operators. We first expand the BRST charge in term of
modes, using the results of chapter 4. One has for the holomorphic BRST charge:
I I
dz dz 1
c(z)T x (z) + •• cT g •• (z)


Qb = jb (z) =
2iπ 2iπ 2
X Xm−n
◦ ◦
= cn Lx−n + ◦ cm cn b−m−n ◦ + λc0 , (5.67)
n n,m
2

where the normal ordering constant λ, coming from terms like c0 cn b−n in the expansion (5.67),
is determined as follows. On the one hand one has the OPE
 3 
jb (z)b(0) = c(z)T x (z) + •• bc∂c •• (z) + ∂2 c b(0)
2
• • • •
T x (z) b∂c (z) bc (z) 3
= − • •
− • 2• + 3 + reg.
z z z z
g
3 j (0) 1
= 3 + 2 + + (T x (0) + T g (0)) + reg. , (5.68)
z z z
where the ghost current was defined in eqn. (4.152), which implies that:

{Qb , b0 } = Lx0 + Lg0 . (5.69)

On the other hand one has from eqn. (5.67) the anticommutator (remember that c0 is viewed
as a creation operator):
 X m X n ◦ 
{Qb , b0 } = c0 , b0 Lx0 + ◦
◦ cm c0 b −m
◦
◦ , b0 − ◦
◦ c0 cn b−n ◦ , b0 + λ c0 , b0
m∈Z
2 n∈Z
2

1X  1 X
∞ −1

= Lx0 − m b−m c0 cm , b0 + m cm c0 b−m , b0
2 m=1 2 m=−∞

1X  1 X
∞ −1

− n b−n c0 cn , b0 + n cn c0 b−n , b0 + λ
2 n=1 2 n=−∞
X
∞ X
−1
= Lx0 − mb−m cm + mcm b−m + λ
m=1 m=−∞

= Lx0 + Lg0 + λ + 1, (5.70)

where we have used the expression (4.180) for Lg0 . Therefore one should set λ = −1.
In a similar fashion one can determine the expansion of the anti-holomorphic BRST charge
Qb as follows:
X Xm−n
e
ex + ◦ e−m−n ◦ − e
Q
eb = cn L
e −n ◦ecm e
cn b ◦ c0 . (5.71)
n n,m
2

122
The string spectrum

Let us characterize now the physical states of the bosonic closed string theory. A state |Ψi
in the Hilbert space of the CFT for the scalars xµ (z, z̄) and for the ghosts (b, c, b̃, c̃) should
be annihilated by the BRST charge defined above:
 
Qb + Qb |Ψi = 0 ,
e (5.72)

Since the Hilbert space of a CFT splits into its holomorphic and anti-holomorphic compo-
nents, one will see that one can impose equivalently the constraints

Qb |Ψi = Q
e b |Ψi = 0 . (5.73)

Physical states, being in the BRST cohomology, are furthermore identified by the equiv-
alence relation  
|Ψi ∼ |Ψi + Qb + Q e b |χi , (5.74)

where |χi is any (non-physical) state. Usually Qb + Q e b |χi is called a spurious state.

There is an extra condition that should be imposed on the physical states in order to obtain
a sensible spectrum. Remember that the ghost CFT has a two-fold degenerate ground state,
see eqn. (4.176). If we allow both ground states one would obtain two families of physical
states. It turns out that one choice is sensible and not the other. We impose that the physical
states obey the condition known as the Siegel gauge:

b0 |Ψi = b
e0 |Ψi = 0 . (5.75)

Since b0 |−igh = 0, it means that the physical spectrum of the string is built from the ghost
vacuum |−igh ⊗ |−i
f of the tensor product of the holomorphic and anti-holomorphic ghost
gh
CFTs.
Notice finally that the physical states are characterized by the simple relations

(Lx0 + Lg0 )|ψi = {Qb , b0 }|Ψi = 0 , (L0

eg )|Ψi = {Q
ex + L
0
e0 }|Ψi = 0 .
e b, b (5.76)

In other words, the left and right conformal dimensions of the physical states (including the
ghost contribution) should both vanish. Naturally, since the underlying CFT is not unitary,
is does not mean that the physical spectrum contains just the vacuum.

5.3 The closed string spectrum

We have now gathered all the ingredients to construct the full spectrum of bosonic string
theory. We will mostly consider the closed string sector. In order to proceed, let us first
recapitulate what are the Hilbert spaces of the CFTs on the worldsheet of the string.
First, the Hilbert space of CFT for the free scalar fields {xµ (z, z̄), µ = 0, . . . , 25}, is built
from the primary states |pµ i, of conformal dimensions
α0
Lx0 |pi = L
ex |pi =
0 pµ pµ |pi . (5.77)
4
123
The string spectrum

The other states generically, non-primary, are then obtained from those by the creation
operators:
e µ2 ν2 e µ1 ν1
|Φi = · · · (e αµ−11 )N1 (αν−11 )N1 |pi
αµ−22 )N2 (αν−22 )N2 (e (5.78)
The left and right conformal dimension of such a state are
α0  X
Lx0 |Φi = p2 + Nx |Φi , Nx = rNµr r (5.79a)
4 µr ,r
α0  X
ex |Φi =
L e x |Φi , N
p2 + N e = e µr
rN (5.79b)
0 r
4 ν ,r r

The integers Nx and N e x are called the left and right levels of the states; they specify the
”depth” of the state in the representation of the current algebra of highest weight state |pi.
Second, the Hilbert space for the holomorphic and anti-holomorphic ghost CFTs in the
Siegel gauge (5.75) is built out the vacuum |−igh ⊗ |−if by acting with the creation oper-
gh
ators c−n and b−n , for n > 0 (acting with the zero-mode c0 would violate the Siegel gauge
condition). Therefore a generic state is of the form

N1
M1  
Ne 1
M 
|Φigh = · · · b−1 |−igh ⊗ · · · b−1 |−igh .
e1
c−1 e c−1
e f (5.80)

The left and right conformal dimension of such a state are

  X
Lg0 |Φigh = Ng − 1 |Φigh , Ng = r(Nr + Mr ) (5.81a)
r
  X
eg |Φigh = N
L e g − 1 |Φigh , eg =
N r(N
er + M
fr ) (5.81b)
0
i

The integers Ng and N e g are the left and right ghost levels of the states. Notice that, since
the ghosts are fermionic variables, the integers Nr , Mr , Ne r and M
fr are either zero or one.
In the following, we will work out the spectrum level by level, according to the total left
and right levels N and N̄ of the full conformal field theory for the coordinate fields and for the
ghosts. This makes sense as the BRST charge, as can be seen by the mode expansion (5.67),
does not mix terms of different levels. Since we have seen that the physical states satisfy
L0 |Ψi = L
e0 |Ψi = 0, the left and right levels need to be the same.2

5.3.1 The tachyon

Let us start by looking at the physical states at level zero, i.e. with neither oscillators for
the coordinates fields nor for the ghost fields. The unique such state is of the form

|pµ it = |pµ i ⊗ |−igh ⊗ |−i

f .
gh (5.82)
2
This is true because we consider states built out of the primary state |pµ i for the matter part, which has
the same left and right conformal weights. It may not be true, as we will see later, if some of the dimensions
of space-time are compact.

124
The string spectrum

We have
X Xm−n 
Qb |p it =
µ
cn Lx−n + ◦
◦ cm cn b−m−n ◦◦ − c0 |pµ i ⊗ |−igh ⊗ |−i
f
gh
n n,m
2
= Lx0 |pµ i ⊗ c0 |−igh ⊗ |−i
f − |pµ i ⊗ c0 |−igh
⊗ |−i


gh
f
gh (5.83)
 0 
α 2
= p − 1 |pµ i ⊗ |+igh ⊗ |−i
f .
gh (5.84)
4
Likewise we find that
α0 2
 
e b |p it =
Q µ
p − 1 |pµ i ⊗ |−igh ⊗ |+i
f . (5.85)
gh
4
Therefore the BRST constraint (5.72) is satisfied provided that the state satisfies:
α0 2
p = 1. (5.86)
4
Using the mass-shell condition p2 + m2 = 0, one gets that
4
m2 = − . (5.87)
α0
From the low-energy perspective, this state of string theory behaves like a scalar particle with
imaginary mass. Such particle, which is called a tachyon, indicates an instability of the field
theory, as we expand around a local maximum of the potential.
This is a very severe problem of bosonic string theory, and by itself it indicates that
the theory is ill-defined, at least in the perturbative, first-quantized approach that we are
using. Fortunately, slightly more sophisticated theories, called superstring theories, can get
rid successfully of this instability. In this perspective, bosonic string theory is a ”toy model”
that allows to become familiar with the tools used in the more sophisticated construction.

5.3.2 The graviton and other massless states

We consider now the physical states at level N = N̄ = 1. The more general ansatz for such
physical state is of the form

|Ψ1 i = eµν αµ−1 α
e ν−1 + ζµ αµ−1 b
e−1 + υµ αµ e e e µ b−1 + υ
−1 c−1 + ζµ α −1 e µ−1 c−1
eµ α

+ λ1 b−1 b
e−1 + λ2 c−1 e
c−1 + λ3 b−1 e e−1 |pµ it . (5.88)
c−1 + λ4 c−1 b

We now impose that such a state is annihilated by the BRST charge. As before one can split
the discussion between the action of the left BRST charge on the left-moving states and of
the right BRST charge on the right-moving states.
The only terms from the BRST charge contributing to the computation at this level are
given by
Qb = c0 (Lx0 − 1) + c−1 Lx1 + c1 Lx−1 − b−1 c0 c1 − c−1 c0 b1 + · · · (5.89)

125
The string spectrum

Technically, the first term will enforce that the conformal dimension of the state in the matter
CFT is h = 1, the second term will enforce that it is a primary state while the third term
will provide the equivalence relations between matter states of different polarizations that
differ by a spurious state.
Notice that no contributions from the cubic terms in the ghost oscillators arise, since those
terms, of the form c−1 c0 b1 or b−1 c0 c1 , give zero because of the Siegel gauge condition (5.75).
To see this, we will act with the BRST charge (5.89) on the various terms in (5.88) and
provide the condition for BRST closeness. One has first:

Qb αµ−1 |pi ⊗ |−igh = (Lx0 − 1)αµ−1 |pi ⊗ c0 |−igh + Lx1 αµ−1 |pi ⊗ c−1 |−igh
r
α0 2 µ α0 µ
= p α−1 |pi ⊗ |+igh + p |pi ⊗ c−1 |−igh , (5.90)
4 2
where we have used that
r
α0 µ
L1 αµ−1 |pi = δνρ αν0 αρ1 αµ−1 |pi = δνρ αν0 [αρ1 , αµ−1 ]|pi = p |pi (5.91)
2
Next we have

Qb |pi ⊗ b−1 |−igh = c0 (Lx0 − 1) + c−1 Lx1 + c1 Lx−1 − b−1 c0 c1 |pi ⊗ b−1 |−igh


r
α0 2 α0
p − 1 |pi ⊗ b−1 |+igh + pµ αµ−1 |pi ⊗ |−igh − |pi ⊗ b−1 |+igh


=−
4 2
r
α0 2 α0
= − p |pi ⊗ b−1 |+igh + pµ αµ−1 |pi ⊗ |−igh (5.92)
4 2
and finally

α 0 p2 α 0 p2
Qb |pi⊗c−1 |−igh −1 |pi⊗c−1 |+igh −|pi⊗c−1 |+igh = − |pi⊗c−1 |+igh (5.93)


=−
4 4
One needs first to remove all the terms constructed from the ghost vacuum |+igh =
c0 |−igh . This gives the space-time mass-shell condition

α0 2 α0 2
m = p = 0, (5.94)
4 4
hence level-one states of string theory can be interpreted at low energies as massless particles.
The vanishing of the contribution corresponding to the second term in (5.90) imposes
then the condition
pµ eµν = 0 , pµ υµ = 0 (5.95)
while contributions corresponding to the second term in (5.92) vanish if all the terms in b−1
are absent in (5.88), namely
ζeµ = λ1 = λ3 = 0 . (5.96)

126
The string spectrum

In a similar way, the anti-holomorphic charge Q

e b annihilate the state (5.88) provided that:
pν eµν = 0 , pµ υ
eµ = 0 (5.97)
and
ζµ = λ4 = 0 . (5.98)
In other words we are left with
|Ψ1 i = eµν αµ−1 α
e ν−1 + υµ αµ−1 e c−1 |pµ it ,
e µ−1 c−1 + λ2 c−1 e


c−1 + υ
eµ α
pν eµν = pµ eµν = pµ υµ = pµ υ
eµ = 0 . (5.99)
The spurious states at level one correspond to all the states than can be obtained from
the generic ansatz (5.88), without imposing of course the physical state constraints. First,
given that (with the momentum on-shell)
r
α0
µ
Qb e^µν α−1 αe −1 |p it =
ν µ
e µ−1 |pµ it ,
e^µν pµ c−1 α (5.100)
2
the terms in υµ and υeµ in eqn. (5.99) can be gauged away. Next we notice that
r
α0
µ
Qb ^vµ α−1 ec−1 |pµ it = c−1 |pit ,
(pµ ^vµ ) c−1 e (5.101)
2
therefore one can set λ2 = 0 in (5.99), given that the term in c−1 e c−1 is spurious. Finally:
  
Qb + Qe b κµ αµ b
−1 −1 + e
e e µ−1 b−1 |pµ it =
κµ α
r 
α0 µ 
µ ν µ
κ pµ c−1 b−1 + (κµ α−1 )(pν α−1 ) + κ pµ c−1 b−1 + (e
e e e e κµ α−1 )(pν α−1 ) |pµ it (5.102)
e µ ν
2
Hence, provided that
κµ pµ = e
κµ pµ = 0 , (5.103)
one should make the identifications
eµν ∼ eµν + κµ pν + e
κν pµ . (5.104)
To summarize:

The physical states of the bosonic closed string theory at level one are given by the BRST
invariant states
|Ψ1 i = eµν αµ−1 α̃ν−1 |pµ i ⊗ |−igh ⊗ |−i
f ,
gh (5.105)
subject to the conditions

pµ pµ = 0 , (5.106a)
pµ eµν = pν eµν = 0 , (5.106b)
eµν ∼ eµν + aµ pν + bν pµ , aµ pµ = bν pν = 0 . (5.106c)

These states correspond to massless particles in space-time with a gauge symmetry expressed
by the equivalence relations.

127
The string spectrum

In the local operator formalism, these massless physical states correspond to the following
local operators
¯ ν eipµ xµ • (z, z̄) .
eµν •• ∂xµ ∂x (5.107)
•

To get the space-time particle interpretation of these states, one needs to decompose the
tensor of physical polarizations, i.e. those satisfying the transversality constraints (5.106b),
into irreducible representations of the Poincaré group. The symmetric traceless part will
correspond to the graviton δGµν , the antisymmetric part to the fluctuations of the Kalb-
Ramond tensor δBµν and and the trace part to the fluctuations of the dilaton δΦ. To
disentangle the trace part from the symmetric part in a covariant way, one needs to introduce
a vector p̄µ such that
p̄µ p̄µ = 0 , pµ p̄µ = 1 . (5.108)
One has then
1
eG
µν = e(µν) − eρρ (ηµν − pµ p̄ν − p̄µ pν ) (5.109a)
D−2
eBµν = e[µν] (5.109b)
1
eΦµν = eρ (ηµν − pµ p̄ν − p̄µ pν ) . (5.109c)
D−2 ρ
The identifications induced from the BRST cohomology,

µν ∼ eµν + p(µ vν)

eG G
(5.110a)
eBµν ∼ eBµν + p[µ wν] (5.110b)

correspond respectively to the Fourier transform of diffeomorphism invariance in space-time

and to the gauge symmetry associated with the Kalb-Ramond form:

hµν ∼ hµν + ∂(µ ζν) (5.111a)

bµν ∼ bµν + ∂[µ λν] (5.111b)

Note that the gauge parameters ζν and λν are transverse in the present context, which means
that string theory has automatically chosen a generalized Lorentz gauge.

5.4 Open string spectrum

We discuss briefly the case of open strings. In this case, there is a single set of modes for all
the fields (scalar fields and ghosts), and one has to consider the cohomology of the unique
BRST charge QB .
We describe first the Hilbert space of CFT for the free scalar fields {xµ (z, z̄), µ = 0, . . . , 25}.
For definiteness we consider that the space-time where the strings live contains a set of N
parallel Dp-branes, located at positions Xai with i = 1, . . . , N and a = p + 1, . . . , 25.

128
The string spectrum

Therefore the spectrum splits into sectors (ij) associated with open strings starting on
the brane number i and finishing on the brane number j. In each sector one has a primary
state |pµ i(ij) , of conformal dimension (see eq. 4.74):
  Xa −Xa 2 
L0 |pi(ij) = α pµ p + 2πα
x 0 µ i
0
j
|pi(ij) . (5.112)

with a momentum vector pµ , µ = 0, . . . , p having only components in the Neumann direc-

tions. This conformal weight is a function of the distance between the branes,
X
s
dij := (Xai − Xaj )2 . (5.113)
a

The other states are then obtained from the ground state by acting with the single set of
creation operators:
µ2 µ1
|Φi(ij) = · · · (αµ−22 )N2 (αµ−11 )N1 |pi(ij) (5.114)
The conformal dimension of such a state is given by:
!
2
dij
Lx0 |Φi(ij) = α 0 p2 + 2 0 + Nx |Φi(ij) (5.115)
4π α

Second, the Hilbert space for the holomorphic ghost CFT in the Siegel gauge is built out the
vacuum |−igh by acting with the creation operators c−n and b−n , for n > 0. A generic state
is of the form
N
M
|Φigh = · · · b−1 1 c−1 1 |−igh (5.116)
The conformal dimension of such a state is
  X
L0 |Φigh = N − 1 |Φigh ,
g g
Ng = r(Nr + Mr ) . (5.117)
r

5.4.1 Open string tachyons

We start by considering the physical states at level zero in the sector (ij). They are of the
form
|pµ i(ij) ⊗ |−igh . (5.118)
It is convenient to use an alternate basis by introducing set of N2 Hermitian matrices of
dimensions N × N, that we will denote by Lr , normalized as

Tr (Lr Ls ) = δrs . (5.119)

A generic state at level zero can be expanded on this basis as

X
N2

|Ψ i =
0
ψr Lrij |pµ i(ij) ⊗ |−igh . (5.120)
r=0

129
The string spectrum

In the present context, these matrices are known as Chan-Paton factors. N × N Hermitian
matrices are associated with the Lie Algebra u(N); the physical significance of this remark
will be explained below.
Let us come back to the study of the BRST cohomology of the open string. We have first
to compute:
X Xm−n 
Qb |pµ i(ij) = cn Lx−n + ◦
c c b
◦ m n −m−n ◦
◦
− c0 |p i(ij) ⊗ |−igh
µ

n n,m
2
= L0 |p i(ij) ⊗ c0 |−igh − |pµ i(ij) ⊗ c0 |−igh
x µ



(5.121)
!
dij2
= α 0 p2 + 2 0 − 1 |pµ i(ij) ⊗ |+igh . (5.122)
4π α

Therefore the BRST constraint (5.72) is fullfilled provided that one satisfies the mass-shell
condition:  2
2 dij 1
mij = 0
− 0. (5.123)
2πα α
Thus the states at level zero can be recast into a matrix-valued scalar field, with a mass
matrix mij whose square is satisfies the condition (5.123). The matrix m2ij has always some
negative eigenvalues, corresponding to open string tachyons.
If we consider more specifically the ground state of an open string stretched between two
Dp-branes, we get a tachyon provided that the two D-branes are close enough. The critical
distance dc is given by

dc2 1 √
− = 0 =⇒ dc = 2π α0 . (5.124)
(2πα 0 )2 α 0

Whenever an open string starts and ends on the same D-brane, the result is of course always
a tachyon, on mass squared m2 = −1/α 0 . Unlike the closed string tachyon (which exists in
bosonic strings, and not in consistent superstring theories) one can have open string tachyon
in a superstring theories if one considers unstable configurations of D-branes.

5.4.2 Gauge bosons

We consider then the physical states at level one in the sector (ij). Following the discussion
about the closed string states, the more general ansatz for such physical state is of the form
 
|Ψ1 i(ij) = u(ij)
µ αµ
−1 + f(ij) a
a α−1 + λ b
1 −1 + λ2 −1 |p i(ij) ⊗ |−igh .
c µ
(5.125)

Compared to the case of closed string theory, the Lorentz group O(1, 25) is broken to O(1, p)×
O(25 − p) in the presence of Dp-branes hence we split the oscillators part accordingly. From
this perspective the first and second term in the expansion (5.125) can be considered as
corresponding to different physical states.

130
The string spectrum

We now impose that the state (5.125) ia annihilated by the BRST charge. As before one
considers first:

Qb αµ−1 |pi(ij) ⊗ |−igh = (Lx0 − 1)αµ−1 |pi(ij) ⊗ c0 |−igh + Lx1 αµ−1 |pi(ij) ⊗ c−1 |−igh
!
d 2
ij
√
= α 0 p2 + 2 0 αµ−1 |pi(ij) ⊗ |+igh + 2α 0 pµ |pi(ij) ⊗ c−1 |−igh ,
4π α
(5.126)

and similarly

Qb αa−1 |pi(ij) ⊗ |−igh = (Lx0 − 1)αa−1 |pi(ij) ⊗ c0 |−igh + Lx1 αa−1 |pi(ij) ⊗ c−1 |−igh
!
dij2
= α p + 2 0 αa−1 |pi ⊗ |+igh
0 2
(5.127)
4π α

The second term vanishes as there is no conserved spatial momentum in a direction transverse
to the D-branes.3 . Next we have

Qb |pi(ij) ⊗ b−1 |−igh = c0 (Lx0 − 1) + c−1 Lx1 + c1 Lx−1 − b−1 c0 c1 |pi(ij) ⊗ b−1 |−igh


!
2
d ij
= − α 0 p2 + 2 0 − 1 |pi(ij) ⊗ b−1 |+igh
4π α
√
+ 2α 0 pµ αµ−1 |pi(ij) ⊗ |−igh − |pi(ij) ⊗ b−1 |+igh
!
d 2
ij
√
= − α 0 p2 + 2 0 |pi(ij) ⊗ b−1 |+igh + 2α 0 pµ αµ−1 |pi(ij) ⊗ |−igh
4π α
(5.128)

and finally
!
2
d ij
Qb |pi(ij) ⊗ c−1 |−igh = − α 0 p2 + 2 0 − 1 |piij ⊗ c−1 |+igh
4π α
!
dij2
+ |pi(ij) ⊗ c−1 |+igh 0 2
=− αp + 2 0 |pi(ij) ⊗ c−1 |+igh (5.129)
4π α

One needs first to remove all the terms constructed from the ghost vacuum |+igh =
c0 |−igh . This gives the space-time mass-shell condition
 2
2 dij
m = . (5.130)
2πα 0
q H
2 1 ¯ a =0
3


Another way to see this is that, for a boson with Dirichlet b.c., α0 = α 0 2π dz ∂xa − dz̄ ∂x
using the expansion (4.73)

131
The string spectrum

hence level-one states of open string theory are massive if the string is stretched between
two separated D-branes, or massless if they stretch between two coincident D-branes or have
both ends on the same D-brane.
The vanishing of the contribution corresponding to the second term in (5.126) imposes
then the condition
pµ uµ(ij) = 0 (5.131)
while contributions corresponding to the second term in (5.128) vanish if all the terms in b−1
are absent in (5.88), namely λ1 = 0. In other words we are left with

|Ψ1 i(ij) = u(ij) µ a

|p i(ij) ⊗ |−igh , pµ uµ(ij) = 0 .

µ
µ α−1 + fa α−1 + λ2 c−1 (5.132)

To obtain the spurious states at level one sees first that:

√
^ µ αµ−1 |pµ i(ij) ⊗ |−igh = 2α 0 u
Qb u ^ µ pµ |pµ i(ij) ⊗ c−1 |−igh (5.133)

hence one can set λ2 = 0 in (5.132). Then we have:

√
Qb |pµ i(ij) ⊗ b−1 |−igh = 2α 0 pµ αµ−1 |pµ i(ij) ⊗ |−igh (5.134)

Hence one should make the identifications

µ ∼ u µ + pµ λ
u(ij) (ij) (ij)
. (5.135)

If one uses the Chan-Paton basis, the physical states in the open string sector string
theory at level one are given by two types of BRST invariant states
X X 
|Ai := urµ Trij αµ−1 |pµ i(ij) ⊗ |−igh (5.136a)
r i,j
X X 
|Φi := fra Trij αa−1 |pµ i(ij) ⊗ |−igh (5.136b)
r i,j

subject to the conditions

 2
dij
pµ pµ = 2πα 0
, (5.137a)
pµ urµ = 0 , (5.137b)
urµ ∼ urµ + pµ λ . r
(5.137c)

To understand the physical significance of these physical states, we consider first open
strings sectors with both endpoints on the same Dp-brane. It gives two types of massless
states:
(ii)
• |A(ii) i = uµ αµ−1 |pµ i(ii) ⊗ |−igh corresponds to a U(1) gauge boson in p + 1 dimensions.
(ii)
The equivalence relation uµ ∼ uµ + pµ λ(ii) expresses gauge invariance.

132
The string spectrum

• αa−1 |pµ i(ii) ⊗ |−igh correspond to 25 − p massless scalar fields. These scalar fields, being
massless, can acquire a vacuum expectation value. These expectation values corre-
spond to the center-of-mass position of the Dp-brane in the corresponding transverse
dimensions.

Consider now a stack of N coincident Dp-branes. In this case all open string sectors give
massless states at level one that can be interpreted as follows:
(ij)
• |A(ij) i = uµ αµ−1 |pµ i(ij) ⊗ |−igh with i, j = 1, . . . , N gives a set of N × N = N2 gauge
fields. While the Cartan states |A(ii) i gives, following the previous discussion, a U(1)N
gauge symmetry, it gets enhanced thanks to the ”off-diagonal” states |A(ij) i with i 6= j
to a non-Abelian
P P gauge
U(N) symmetry. In the Chan-Paton basis, one has the gauge
field |Ai = r uµr L r µ
α
i,j ij −1 |p µ
i(ij) ⊗ |−igh where the matrices Lrij are the generators
of the u(N) Lie algebra.

• The (25 − p) × N2 scalar fields αa−1 |pµ i(ij) ⊗ |−igh arrange themselves as 25 − p matrix-
valued scalar fields transforming in the adjoint representation of U(N).

While the previous identifications rely essentially on the counting of states, one can check
that scattering amplitudes are consistent with these properties (see chapter 6).
Finally let’s see what happens when we consider a set of N Dp-branes that are separated in
their transverse directions, rather than stacked on top of each other. In this case, for a generic
configuration, while the diagonal gauge bosons |A(ii) i remains massless, the off-diagonal gauge
bosons |A(ij) i with i 6= j acquire a mass proportionnal to the distance between the Dp-branes
on which they end:
dij
mij = . (5.138)
2πα 0
Since the transverse positions of the Dp-branes correspond to the vaccum expectation
values of the scalar fields αa−1 |pµ i(ii) ⊗ |−igh , there is a very neat interpretation of this phe-
nomenon from the low-energy view point: the Higgs mechanism. In the present context, one
considers a U(N) gauge theory in p + 1 dimensions with an adjoint Higgs field (without po-
tential at tree-level). For a generic expectation value of the Higgs field, the gauge symmetry
is broken down to U(1)N .
To summarize, introducing D-branes allow to introduce non-Abelian gauge theory in string
theory, and even better in four space-time dimensions if one chooses a stack of Dp-branes.
Notice however that, in the perspective of model building, the gravitationnal sector of the
theory remains 26-dimensional.

5.5 Physical degrees of freedom and light-cone gauge

A more physical way of working out the string spectrum is to identify first properly the
physical degrees of freedom. We will start by discussing the familiar example of a massless
vector field which arises in open string theory. We consider to simplify the discussion that

133
The string spectrum

we have a single D25-brane. As we have discussed before the vector field corresponds to the
physical state
|Ψi = uµ αµ−1 |pi ⊗ |−igh , (5.139)
whose polarization vector uµ satisfies the transversality condition

pµ uµ = 0 , (5.140)

coming from the BRST constraint L1 |Ψi = 0. It can be equivalently rephrased as

pν αν1 |Ψi = 0 , (5.141)

which is similar to the familiar physical state condition for the Maxwell field in the Gupta-
Bleuler formalism of QED, ∂µ Aµ |Ψi = 0.
One can choose, without loss of generality, pµ = (ω, ω, 0, . . . , 0), and choose accordingly
the polarization basis as

ug = √1 (1, 1, 0 · · · , 0) , (5.142a)
2
l
u = √1 (−1, 1, 0, · · · , 0) , (5.142b)
2
ui = (0, 0, 0 · · · 0, |{z}
1 , 0 · · · 0) (5.142c)
i+2

and expand the oscillator modes

αn µ = αln ugµ + αgn ulµ + αin uiµ . (5.143)

In this basis the Gupta-Bleuler condition (5.141) is simply

αg1 |Ψi = 0 , (5.144)

which imposes, because of the commutation relation

[αg1 , αl−1 ] = 1 , (5.145)

that the physical state has no longitudinal component, i.e. no αl−1 oscillator mode. The
spurious states, exactly as in the closed string computation (5.102), are obtained as
r
α0
Qb b−1 |pi ⊗ |−igh = pµ αµ−1 |pi ⊗ |−igh , (5.146)
2
hence αg−1 |pi ⊗ |−igh is a spurious state. The remaining polarizations give the physical states,

|Ψi = ui αi−1 |pi ⊗ |−igh , (5.147)

which contain only transverse excitations.

134
The string spectrum

To summarize, at the level of the polarization vector the transversality condition pµ uµ = 0

allows to do the decomposition,
uµ = apµ + uµT , uµT = (0, 0, uT ) . (5.148)
where the longitudinal component is null (since p2 = 0). Because of gauge invariance,
uµ ∼ uµ + λpµ , (5.149)
the longitudinal polarizations can be gauged away, leaving only the physical transverse po-
larizations, that transform in the vector representation of the little group SO(D − 2).
We consider now the massless closed string states. The polarization tensor eµν obeys the
transversality condition pµ eµν = pν eµν = 0 hence can be decomposed in the same way as
above:
eµν = aµ pν + bν pµ + eTµν , pµ aµ = pµ bµ = 0 , (5.150)
where the transverse polarization tensor is given by
 
0 0 0
eTµν =  0 0 0  , i, j = 2, . . . 26 . (5.151)
0 0 eij
As before, the unphysical longitudinal polarizations are completely removed by the gauge
invariance.
One can then decompose the transverse polarization tensor into irreducible representations
of the space-time little group SO(D − 2), with D = 26, namely into its symmetric traceless,
antisymmetric and trace part, following (5.109):
eTij = e(ij) − D−2
1
ei i δij + e[ij] + D−2
1
ei i δij .


(5.152)
| {z } |{z} | {z }
sij aij d

where sij , aij and d give respectively the physical polarizations of the graviton, Kalb-Ramond
particle and dilaton.
So, after all this hard work, we have seen that the closed string theory spectrum contains a
state that corresponds to a massless spin-two particle, with the appropriate gauge symmetry
to remove the unphysical polarizations. By general arguments a sensible effective theory for
this spin-two field cannot be anything else than Einstein gravity (coupled to the dilaton and
Kalb-Ramond fields). Actual computations of S-matrix elements from string theory confirm
explicitly that this result is correct.
If we reconsider the generic Polaykov-Kalb–Ramond-dilaton action (5.1) from this per-
spective one gets some very interesting result. This action is defined with respect to back-
grounds for the metric, B-field and dilaton, which can be viewed as coherent states of strings
states. Indeed, the exponentiated integrated vertex operator
Z
¯ ν
d2 z e(µν) eip·x ∂xµ ∂x (5.153)

can be viewed as a deformation of the Polyakov action around a Minkowskian background,

due to a change of the metric by a plane wave perturbation.

135
The string spectrum

Light-cone quantization
A more direct approach to string theory quantization, called light-cone quantization, gives
directly the spectrum of physical states, by using the residual gauge freedom of the string
path integral in the flat gauge (conformal transformations), which allow to eliminate all the
oscillator modes of x+ (σi ) = x0 (σi ) + x1 (σi ), by choosing worldsheet coordinates σ
^ i such that

x+ σ b i = x+ +


c +p τ b. (5.154)

It can be shown that the Virasoro constraints, i.e. the equation of motion Tij = 0 for the
worldsheet metric, allows then to eliminate all oscillator modes from x− (σi ) = x0 (σi ) − x1 (σi )
as well, leaving only the physical transverse oscillator modes {αin , i = 2, . . . 25}.
However this formalism is less suited for dealing with interactions and loop corrections,
since we are loosing the freedom to use conformal transformations. We refer the reader to
the general string theory textbooks for more details.

5.6 General structure of the string spectrum

The structure of the physical spectrum found at level zero and one persist at higher levels.
All the other states
√ are massive, and their mass is quantized in units of the inverse of the
string length 1/ α 0 which means that these states are extremely massive and out of reach of
any foreseeable experiments. They are however necessary for the consistency of the theory,
and participate indirectly to the low-energy physics through the loop corrections. We discuss
here only the closed string spectrum, the open string spectrum following the same pattern.
Following the previous pattern, a physical level n state (both left and right) will be
constructed by taking a level n state of the matter CFT tensored with the |−igh vacuum of
the ghost CFT:

|Ψn i = eµν αµ−n α e ρ−n + · · · |pµ i ⊗ |−igh ⊗ |−i

e ν−n + fµνρ αµ−n+1 αν−1 α

f ,
gh (5.155)

depending on a large number of polarization tensors. The relevant part of the BRST charge
is
X
∞  
x x x x
Qb = c0 (L0 − 1) + e
c0 (L0 − 1) +
e c−k Lk ,
c−k Lk + e e (5.156)
k=1

such that the first terms give the conditions

ex − 1)|Ψn i = 0 ⇔ m2 = 4 (n − 1)
(Lx0 − 1)|Ψn i = 0 , (L (5.157)
0
α0

These two constraints are usually reorganized as

(L0 + L
e0 − 2)|Ψi = 0 , (5.158a)
(L0 − L
e0 )|Ψi = 0 . (5.158b)

136
The string spectrum

The first one is usually refered to as the mass-shell condition, while the second one the level-
matching condition, which is the only relation between left- and right-moving excitations.
The constraints from the other terms in the BRST charge (5.156) indicate that the state
in the xµ matter CFT should be a primary state, namely
∀k > 0 , Lxk |Ψk i = 0 , L
ex |Ψk i = 0
k (5.159)
which sets constraints on the polarization tensors. It is worthwhile to remind that the matter
CFT is non-unitary because it contains one boson x0 (z, z̄) with wrong sign kinetic term, hence
one can construct generically non-trivial primary states at every oscillator level. Given that
these particles are massive, the physical polarizations of the states should be decomposed
into irreducible representations of the little group SO(D − 1), with D = 26.
Next we move to the spurious states. Owing to our previous experience, apart from the
spurious states that allows to remove the terms with ghost oscillators in the ansatz (5.155),
the interesting ones are obtained from terms of the form
|ξi = b−r |ξxr i ⊗ |−igh ⊗ |−i
f ,
gh (5.160)
where |ξxr i is a state in the xµ CFT at left and right levels (n − r, n), satisfying the condition
(Lx0 − 1 + r) |ξxr i = 0 , (5.161)
and similar states with b
e−n oscillators. We have indeed

Qb |ξi = cr Lx−r b−r |ξxr i ⊗ |−igh ⊗ |−i

f
gh = L−r |ξr i ⊗ |−igh ⊗ |−igh ,
x x f (5.162)
showing that this state is a (left) descendant state at level n, which satisfies the same mass-
shell condition as physical states:
(Lx0 − 1)Lx−r |ξxr i = 0 . (5.163)
These spurious states are orthogonal to all physical states. Indeed,
hΨm |Qb |ξi = hΨm |L−r |ξxr i = hξx |Lr |Ψx i? = 0 , (5.164)
as any physical state is a conformal primary state in the matter CFT.

Example: level two physical states

4
The physical states at level two have all mass squared m2 = α0
, and can be organized into
the following representations of the little group SO(25):

⊕ ⊕ ⊕ ⊕ ⊕•

Moving to the more massive, higher level physical states gives increasingly large SO(25)
representations. The number of physical states actually grows tremendously with the level.
In general, in a CFT with central charge c the asymptotic density of states is given by [3]:
√c
N1
ρ(N) ∼ e2π 6 (N−c/24) , (5.165)

137
The string spectrum

where in the present context c = 24, accounting for the oscillator modes of the 24 transverse
coordinates. A striking consequence is the existence of a limiting temperature, the Hagedorn
temperature, as it implies that, for a string in the canonical ensemble, the partition function
Z
Z(E) = dE ρ(E)e−βE (5.166)

diverges above a certain temperature.

No-ghost theorem
The spectrum of the bosonic string is built out of the tachyon vacuum by acting by the two
set of oscillator modes {αµ−n , α
e ν−n }. Because of the commutation relation (4.61), the modes
built from oscillators of the time-like coordinate field x0 (z, z̄) can potentially be of negative
norm. Indeed, for instance
 0 µ 2
α−n |p i = hpµ |α0n α0−n |pµ i = −hpµ |pµ i . (5.167)
At the massless level, we have already encountered this problem, and it was solved in the
same way as it is solved for the quantum theory of a massless vector field; the unphysical
negative-norm longitudinal polarizations were removed using gauge invariance of the theory,
see eqn. (5.144) and below.
An important consistency check of string theory is to prove that this feature persists
at all string levels. The simplest way to prove this statement is to move to the light-cone
gauge, discussed briefly at the end of subsection 5.3.2, since in that case one keeps only the
transverse oscillators that only create positive norm states from the vacuum.
In the context of the BRST formalism, it is possible to prove as well that all the physical
states have positive norm. The proof of this statement, misnamed no-ghost theorem (as ghost
here means negative norm states and not Faddeev-Popov ghosts!), can be found in [4] and
will not be reproduced here since it is rather technical. The statement is:
The inner product on the BRST cohomology of bosonic string theory is positive definite.

References
[1] C. G. Callan, Jr., E. J. Martinec, M. J. Perry, and D. Friedan, “Strings in Background
Fields,” Nucl. Phys. B262 (1985) 593–609.
[2] J. Polchinski, “Scale and Conformal Invariance in Quantum Field Theory,” Nucl. Phys.
B303 (1988) 226–236.
[3] J. L. Cardy, “Operator Content of Two-Dimensional Conformally Invariant Theories,”
Nucl. Phys. B270 (1986) 186–204.
[4] M. Kato and K. Ogawa, “Covariant Quantization of String Based on BRS Invariance,”
Nucl. Phys. B212 (1983) 443–460.

138
The string spectrum

[5] J. Polchinski, String theory. Vol. 1: An introduction to the bosonic string. Cambridge
University Press, 2007.

139
Chapter 6

String interactions

140
String interactions

In this chapter we will explain how to compute the observables of string theory, in par-
ticular the tree-level S-matrix elements and the vacuum amplitude at one-loop in string
perturbation theory.

6.1 The string S-matrix

In a quantum field theory, the observables are given by N-point correlation functions of the
fields, namely
^ 1 ) · · · φ(x
^ n)  0 .


 
G(x1 , x2 , . . . , xn ) = hφ(x1 ) · · · φ(x2 )i = 0 T φ(x (6.1)

Transforming this position space expression into momentum space, these quantity are clearly
off-shell, as they are defined outside of the hypersurface p2 + m2 = 0.
In string theory, the situation is a little bit more complicated. First we know that there
are no local observables in a quantum theory of gravity, because diffeomorphisms are gauge
symmetries. Second we know that the local operators corresponding to string physical states
need to be conformal primaries of dimensions (1, 1), which forces them to be on-shell. For
this reason the well-defined observables in string theory are S-matrix elements, i.e. on-shell
correlation functions of asymptotic states. This prescription solves also the first issue as the
asymptotic states, being defined on the boundary of space-time, are well-defined in quantum
gravity. All we have said applies to asymptotically Minkowski space-times; in asymptotically
Anti-de-Sitter space-times, the observables are a little bit different (and in asymptotically de
Sitter space-times, one does not know precisely what happens).
In this lecture we will consider only closed string observables. At tree-level, a N-particle
contribution to the S-matrix is obtained by gluing together N semi-infinite cylindrical world-
sheets into a single surface, see an example on figure 6.1. For a one-loop diagram, one should
add a handle to the ”blob” in the center, two handles for a two-loops diagram and so on.
|p5i
|p1i

|p2i
|p4i

|p3i

Figure 6.1: S-matrix element for tree-level five-tachyon scattering

141
String interactions

This looks quite complicated to evaluate, but fortunately one can use the conformal
symmetry of the theory to map this complicated surface to a much simpler one. We have
seen in chapter 4 how to map a cylinder to a plane, and how an initial state at minus infinity
was mapped to a local operator at the origin. By the same reasoning a semi-infinite cylinder,
stretching say from σ2 → −∞ to σ2 = 0, is mapped to a disc and an asymptotic state is
mapped to a local operator at the origin of the disc. We can imagine defining a conformal
transformation doing the same here for of the semi-infinite cylinders, and we will end up
with a sphere with N punctures, and a local operator inserted at each of the punctures, see
figure 6.2.

Vpi
|pi i

Figure 6.2: Conformal mapping of S-matrix elements

Because of the residual gauge symmetry in the gauge-fixed path integral due to the con-
formal killing vectors on a sphere, see subsection 2.4.1, three of the N local operators corre-
sponding to the string physical states participating into the scattering are at arbitrary fixed
positions, while the rest are integrated over the sphere. For convenience one can represent
the two-sphere by the compactified complex plane C̄ = C ∩ ∞, with the north pole at the
origin and the south pole at infinity. The arbitrary positions are usually chosen to be z1 = 0,
z2 = 1 and z3 = ∞, respectively the north pole, a point on the equator and the south pole.
Finally, there exists a natural normalization of the string physical operators that we didn’t
discuss yet. In the loop expansion of string perturbation theory, see fig. 2.4, each extra handle
on the worldsheet is coming with an extra factor of g2s . Since a handle is associated with
virtual string loop in the diagram, it is natural to add a factor of gs for each external state
of the S-matrix.

6.2 Four-tachyon tree-level scattering

To illustrate the computation of S-matrix elements, we consider the example of four-tachyon
tree-level scattering. While not physically terribly exciting, this amplitude is historically
relevant – it appeared before string theory itself, in the context of dual resonance models –
and simpler than the four-graviton scattering, which has a complicated index structure.

142
String interactions

Following all the previous discussion, one considers the following expression:
Z
(gs )4
 
2
S(p1 , p2 , p3 , p4 ) = c(0)e
c(0)Vp1 (0) c(1)e c(∞)Vp3 (∞) d z Vp4 (z, z̄)
c(1)Vp2 (1) c(∞)e
gs2 S2
(6.2)
with the on-shell vertex operators
• µ • α0 2
Vp (z, z̄) = • eipµ x • (z, z̄) , 4
p = 1. (6.3)

The CFT correlation function that has to be computed in (6.2) clearly splits into a
”matter” part for the fields xµ and a ghost part for the fields c and c̃. Let us start with the
former.

6.2.1 Scalar field correlation functions

We are interested in computing, in a two-dimensional CFT of free scalar fields xµ with
action (4.1), the N-point function of vertex operators1

Y
* N +
µ µ
• ipi x (zi ,z̄i ) •
GN (zi , z̄i ) = •e • . (6.4)
i=1

Because we are dealing with a free Gaussian quantum field theory, one can couple the
theory to an external current Jµ and write
R 1
R R
d2 z Jµ (z,z̄)xµ (z,z̄) d2 z d2 z 0 Jµ (z)Jµ (z 0 )G(z−z 0 ,z̄−z̄ 0 )
hei i = e− 2 , (6.5)

where G is the Green function of the theory, satisfying the equation

1 ¯
− ∂∂G(z − z 0 , z̄ − z̄ 0 ) = δ(z − z 0 )δ(z̄ − z̄ 0 ) , (6.6)
2πα 0
Recall that in d dimensions the scalar propagator for a massive scalar field is computed
easily using the Schwinger parametrization, see (2.47) and below:
Z d Z d Z
d p eip·x d p ip·x ∞ 2 2

d 2 2
= d
e dt e−t(p +µ )
(2π) p + µ (2π)
Z∞ 0
Z
1 −tµ2 2
= d
dt e dd p e−tp +ip·x
(2π) 0
Z∞
1 dt −tµ2 − x2
= e 4t
(4π)d/2 0 td/2
1
= µd−2 (µ|x|)1−d/2 K1−d/2 (µ|x|) , (6.7)
(2π)d/2
1
Notice that this is an N-point correlation function from the two-dimensional CFT perspective, which
does not contradict the statement made before about correlation functions in string theory!

143
String interactions

where Kn is a modified Bessel function of the second kind. Taking d = 2, and expanding the
result for small µ one can use:

K0 (µ|x|) = − log µ|x| + O(1) . (6.8)

Going back to complex coordinates and adding the normalization factor, the solution is
naturally the two-point function already discussed below (4.13):

α0
G(z − z 0 , z̄ − z̄ 0 ) = − log µ2 (z − z 0 )(z̄ − z̄ 0 ) . (6.9)
2
The infrared cutoff µ should disappear from correlation functions of vertex operators. This
cutoff is necessary as the fields xµ have zero-modes in the massless limit.
Fo computing the N-point function (6.4) of scalar vertex operators, one takes the source
Jµ (z, z̄) to be
X
N
µ
J (z, z̄) = pµ` δ(z − z` )δ(z̄ − z̄` ) (6.10)
`=1

and equation (6.5) gives

Z X
!
N 1 2 2 0 2 2 0 0 0
G (zi , z̄i ) = exp − d zd z p` · pr δ (z − z` )δ (z − zr )G(z − z , z̄ − z̄ ) . (6.11)
2 `,r

In the sum the term involving the IR cutoff µ is

X
!
α0
exp log µ p` · pr . (6.12)
2 `,r

Remember that the physical correlation functions should be independent of the infrared
cutoff; it imposes the condition X µ
p` = 0 (6.13)
`

which can be viewed as a charge conservation condition for the affine symmetry of the massless
scalar CFT.
Finally, one has to remove from (6.11) the terms with ` = r that were already taken
care of by the normal ordering of the vertex operators. Inserting the expression (6.9) for the
Green function one obtains the expression, up to an overall normalization

Y X Y
* N +
0
ipµ µ
GN
µ (zi , z̄i ) =
•
• e i x (zi ,z̄i ) •
• = δ26 pµk |z` − zr |α p` ·pr (6.14)
i=1 k `<r

144
String interactions

6.2.2 Ghost correlation function

In the ghost sector of the two-dimensional CFT on the sphere, the S-matrx element (6.2)
contains the three-point function

X
* 3 +
G3g (zi , z̄i ) = c(z` )e
c(z̄` ) , (6.15)
`=1

which splits naturally into holomorphic and anti-holomorphic contributions, since the CFT
factorizes into the holomorphic CFT for (b, c) and the anti-holomorphic CFT for (b, e ec).
Instead of evaluating the path integral, the value of this correlation function can be
determined using uniquely holomorphy arguments. The function G3g should be holomorphic
in each of its six variables, and should vanish when two c ghost are at coincident points, since
these are anticommuting variables. One has then

G3g (zi , z̄i ) = (z2 − z1 )(z3 − z1 )(z3 − z2 )(z̄2 − z̄1 )(z̄3 − z̄1 )(z̄3 − z̄2 )F(z1 , z2 , z3 )e
F(z̄1 , z̄2 , z̄3 ) (6.16)

with F (resp. e
F) holomorphic (resp. anti-holomorphic) in its arguments. Let us consider the
limit z1 → ∞. On the one-hand the expression (6.16) behaves like
|z1 |1
G3g ' z12 F(z1 , z2 , z3 ) . (6.17)

On the other hand, c(z) is a conformal primary field of dimension h = −1 therefore, under
the transformation u = 1/z, one gets cz (z) = −z2 cu (1/z), hence the function G3g (zi , z̄i )
cannot grow faster than z12 when z1 → ∞. It shows that the function F in (6.16) is actually
independent of z1 . With a similar reasoning for the other variables one finds that, up to an
overall normalization constant
X
* 3 +
G3g (zi , z̄i ) = c(z̄` ) = (z2 − z1 )(z3 − z1 )(z3 − z2 )(z̄2 − z̄1 )(z̄3 − z̄1 )(z̄3 − z̄2 ) . (6.18)
c(z` )e
`=1

6.2.3 The Virasoro-Shapiro amplitude

We now put together the matter and ghost contributions to the four-tachyon tree-level am-
plitude in order to get our first observable from string theory. We start with
X  0 0 0
S(p1 , p2 , p3 , p4 ) = gs2 δD pµk |z2 − z1 |α p1 ·p2 +2 |z3 − z1 |α p1 ·p3 +2 |z3 − z2 |α p3 ·p2 +2
k
Z
0 0 0
× d2 z |z − z1 |α p4 ·p1 |z − z2 |α p4 ·p2 |z − z3 |α p4 ·p3 , (6.19)

We set the unintegrated vertex operators to the arbitrary positions z1 = 0, z2 = 1 and

0
z3 = ∞ as explained. The leading term in z3 reads |z3 |α p3 ·(p1 +p2 +p4 )+4 and is equal to one,
using momentum conservation p1 + p2 + p3 + p4 = 0 and the mass-shell condition p2` = 4/α 0 .

145
String interactions

To restore the symmetry between the momenta p` , one introduces the usual Mandelstam
kinetic invariants

s = −(p1 + p2 )2 (6.20a)
2
t = −(p1 + p3 ) (6.20b)
u = −(p1 + p4 )2 , (6.20c)

which are constrained by the mass-shell condition as

X
4
16
s+t+u= m`2 = − . (6.21)
`=1
α0

One ends up then with the following non-trivial integral (since, as the momenta are on shell,
0
we have for instance u = −p,2 2
1 − p4 − 2p1 · p4 = −8/α − 2p1 · p4 ):
Z Z
α 0 p4 ·p1 α 0 p4 ·p2 0 0
I(p` ) = d z |z|
2
|z − 1| = d2 z |z|−α u/2−4 |z − 1|−α t/2−4 (6.22)

Computing this integral need a little bit of work. We use first the integral representation
(similar to the Schwinger representation)
Z∞
1 2
|z|2a−2
= df f−a e−f|z| (6.23)
Γ (1 − a) 0

and compute (with z = x + iy)

Z Z Z Z
1 2 2
I(a, b) = d z |z|
2
|1 − z|
2a−2 2b−2
= df f−a
dg g −b
d2 z e−f|z| −g|1−z|
Γ (1 − a)Γ (1 − b)
Z Z Z
2 −a −b 2 2 2 2
= df f dg g dxdy e−fx −fy −g(x−1) −gy
Γ (1 − a)Γ (1 − b)
Z Z
2π fg
= df f−a
dg g−b (f + g)−1 e− f+g , (6.24)
Γ (1 − a)Γ (1 − b)

where we performed the Gaussian integrals over x and y. We consider now the change of
variables f = αβ and g = (1 − β)α, which gives (the Jacobian of the transformation being
simply α):
Z∞ Z1
2π
I(a, b) = dα dβα−a−b β−a (1 − β)−b e−αβ(1−β) (6.25)
Γ (1 − a)Γ (1 − b) 0 0

The integral over α is easy to compute as one recognizes a gamma-function:

Z∞
dα α−a−b e−αβ(1−β) = βa+b−1 (1 − β)a+b−1 Γ (1 − a − b) (6.26)
0

146
String interactions

and one has finally

Z1
Γ (1 − a − b) Γ (1 − a − b)
I(a, b) = 2π dβ βb−1 (1 − β)a−1 = 2π B(a, b) . (6.27)
Γ (1 − a)Γ (1 − b) 0 Γ (1 − a)Γ (1 − b)

Using the property of the beta-function

Γ (a)Γ (b)
B(a, b) = , (6.28)
Γ (a + b)

one gets finally, introducing the variable c = 1 − a − b,

Γ (a)Γ (b)Γ (c)

I(a, b) = 2π . (6.29)
Γ (1 − a)Γ (1 − b)Γ (1 − c)

It is now time to come back to our original problem and give the expression of the tachyon
four-point function. Substituting the value of the integral in equation (6.19) one gets (with
the identifications a = −α 0 u/4 − 1, b = −α 0 t/4 − 1, c = −α 0 s/4 − 1):
X  Γ (−1 − α 0 s/4)Γ (−1 − α 0 t/4)Γ (−1 − α 0 u/4)
S(p1 , p2 , p3 , p4 ) ∼ gs2 D
δ pµk (6.30)
k
Γ (2 + α 0 s/4)Γ (2 + α 0 t/4)Γ (2 + α 0 u/4)

This amplitude, which was known long before string theory becomes an actual theory
of relativistic strings, is known as the Virasoro-Shapiro amplitude. It has some remarkable
properties that we will examine briefly.

6.2.4 Properties of the string S-matrix

Because of the gamma-functions in the numerator of (6.30), the amplitude has many poles
for the Mandelstam variables. First, if we keep fixed t, one finds that the amplitude has a
series of poles for
4
s = 0 (n − 1) , n ∈ N . (6.31)
α
These correspond precisely to the physical states of the string spectrum: n = 0 for the
tachyon, n = 1 for the massless fields (including the graviton) and n > 2 for the tower
of massive string states. Near any of these simple poles, the amplitude behaves like S ∼
1/(s − mn2 ), corresponding to the exchange of a particle of mass squared mn2 = 4(n − 1)/α 0 .
It means that the string amplitude (6.30) can be interpreted in field theory as an infinite sum
over tree-level Feynman diagrams given by the residues at the poles in the s-channel. Each
of this diagram is interpreted as 1 + 2 → 3 + 4 process through an intermediate unstable
particle (resonance) corresponding to each of the string physical states, see figure 6.3. In
other words, all string states appear as resonances in the four-tachyon scattering amplitude.

147
String interactions

1 1 3
3
P
∞ mn
=
2 4
n=0 2 4

Figure 6.3: s-channel decomposition of the four-tachyon tree-level amplitude.

The amplitude (6.30) has another series of poles for the variable t that provides another
decomposition of the same S-matrix element. They occur for
4
t= (n − 1) , n ∈ N. (6.32)
α0
In this case the simple pole in S ∼ 1/(t − mn2 ) corresponds to a Feynman diagram with two
particles exchanging a virtual particle of mass mn (emitted by tachyon 1 and absorbed by
tachyon 2). It means that there is an equally valid decomposition of the amplitude obtained by
summing over all the residues, see figure 6.4. Again, each of these virtual particles correspond
to a state from the string physical spectrum.

1 3 1 3

P
∞
= mn
n=0
2 4

2 4

Figure 6.4: t-channel decomposition of the four-tachyon tree-level amplitude.

The fact that each of these expansions is equally valid is a manifestation of channel duality
of the string amplitudes. It means that we can either use the s-channel decomposition or the
t-channel decomposition to have a complete description of the scattering process. Of course
similar statements can be made w.r.t. the u-channel decomposition of the amplitude.
Another interesting property of the amplitude (6.30) appears in the high energy limit.
More precisely, we look at the regime
s→∞, t→∞, s/t fixed (6.33)
which corresponds to high-energy scattering with a fixed scattering angle between the incom-
ing tachyons 1 and 2. In this limit one has
 X  α0
s→∞ 2 D
S(p1 , p2 , p3 , p4 ) ∼ gs δ pµk e− 2 (s log s+t log t+u log u) (6.34)
k

148
String interactions

The exponential decay of the amplitude at high energies makes the UV behavior of string
theory much softer than any quantum field theory (which gives only power-like decay), and is
due to delicate cancellations between the contributions of the infinite number of string states.

6.3 One-loop partition function

String worldsheets with the topology of a torus correspond to one-loop contributions to the
string perturbative expansion. We will focus here on the one-loop amplitude, which can be
interpreted in Euclidean time as the statistical partition function of the theory. Crucially,
the partition function needs to be invariant under the modular group PSL(2, Z) of the two-
torus. When one considers a string theory compactified to four-dimensions, one replaces the
free scalar CFT associated with the extra dimension with a more general CFT, and modular
invariance is a stringent consistency test of the validity of the construction.
The starting point, including the contribution from the ghosts and the integral over the
modulus τ of the torus, is given by eqn. (2.132) that we reproduce for convenience here:
Z 2 Z R 2
dτ 1


¯ e
− d2πw b∂c+ ¯ ν
c+ 0 gµν ∂xµ ∂x
Z1 = Dx DbDb DcDe
µ e c c(0)ec(0) b(0)b(0) e
e b∂e
α (6.35)
F 4τ2

Naturally the integrand splits into a vacuum amplitude for the xµ fields and a particular four-
point function for the ghost fields. We will examine separately each of these contributions.

6.3.1 Path integral of the free-scalar CFT

We consider the path integral of a single free, massless scalar field X(σ1 , σ2 ) on a two-
dimensional torus. For this derivation it is convenient to work with a metric
1  1 2
ds2 = dσ + τdσ2  (6.36)
τ2

together with the standard periodicities σi ∼ σi + 2π (rather than the canonical complex flat
metric dzdz̄ with periodicities 2π and 2πτ). the two-dimensional Euclidean action for the
scalar field is
Z Z
1 1
d2 σ |τ∂1 X − ∂2 X|2 .
2
p ij
S= d σ det γγ ∂i X∂j X = (6.37)
4πα 0 4πα 0 τ2
After integration by parts it becomes
Z
1
S= d2 σ X(σi )  X(σi ) , (6.38)
4π
with the Laplacian
1
=− |τ∂1 − ∂2 |2 . (6.39)
α 0 τ2

149
String interactions

To compute the path integral, one has first to split the field X into the classical solutions of
the equations of motion and the fluctuations. The classical solutions, which are the saddle
points in the path integral, are just the constant modes. We splits then

X(σ1 , σ2 ) = Xc + φ(σ1 , σ2 ) , (6.40)

and expand the fluctuations φ(σ1 , σ2 ) in eigenmodes of the Laplacian:

 ϕ` = λ` ϕ` . (6.41)

The solution to this eigenvalue problem is

1 1 2 1
ϕm,n = √ ei(mσ +nσ ) , λm,n = |mτ − n|2 , m, n ∈ Z . (6.42)
π α 0 τ2
These eigenfunctions form an orthonormal basis, as
Z
1
d2 σ ϕ?m,n ϕm 0 ,n 0 = δm,m 0 δn,n 0 . (6.43)
4π
Let us expand φ on this basis, omitting the zero-mode that was already taken apart in (6.40):
X
φ= αm,n ϕm,n , (αm,n )? = α−m,−n (6.44)
(m,n)6=(0,0)

The action for the field X becomes then

X
S[X] = λm,n |αm,n |2 . (6.45)
(m,n)6=(0,0)

Since the eigen-modes were properly normalized, the path integral can be written as
Z Z Y dαm,n P  Y −1/2
2
Z = DX e −S[X]
= dXc e− (m,n)6=(0,0) λm,n |αm,n | = VX λm,n .
2π
(m,n)6=(0,0) (m,n)6=(0,0)
(6.46)
In this expression, VX denotes the volume of the target space associated with the field X,
which is here an infinite line. It gives an infrared divergence, associated in the whole string
theory computation to the volume of space-time (one can divide by this factor and consider
the vacuum amplitude per unit volume instead).

Zeta-function regularized determinant

The infinite product of the eigenvalues of the Laplacian appearing in (6.46) is divergent
and evaluated using zeta-function regularization, as was done for the point-particle one-loop
vacuum amplitude in subsection 2.1. We recall that we define first the spectral zeta-function
as (with λ1 6 λ2 6 · · · λn 6 · · · )
X
∞
ζD (z) = λ−z
n , (6.47)
n=1

150
String interactions

and, noticing that

X
∞
log det D = log λn = −ζD0 (0) , (6.48)
n=1

we define the regularized functional determinant as

Y 0
λn = e−ζD (0) . (6.49)
n

Zeta-function regularization gives us in particular P

the identities (by writing each time ζD in
terms of the usual Riemann zeta-function, ζ(z) = n n−z ):

Y
∞
d −z ζ(0))|
a = e− dz (a z=0
= aζ(0) = a−1/2 (6.50a)
n=1
Y
∞
a = a2ζ(0)+1 = 1 (6.50b)
n=−∞
Y∞
d 0
nα = e− dz ζ(αz)|z=0 = e−αζ (0) = (2π)α/2 (6.50c)
n=1
Y Y
∞ Y
∞
sin πa
(a + n) = a (a + n)(a − n) = a (−n2 )(1 − a2 /n2 ) = 2πia = 2i sin πa .
n∈Z n=1 n=1
πa
(6.50d)

using ζ(0) = − 21 and ζ 0 (0) = − 12 log 2π. Another useful identity is

Y
∞
a−n = e− log aζ(−1) = a1/12 , (6.51)
n=1
P
where we used ζD (z) = n anz , ζD0 (0) = ζ(−1) log a and ζ(−1) = −1/12.
In the present case one needs to perform some manipulation on the determinant of the
Laplace operator on the two-torus. Let us denote by det 0  the determinant of the Laplace
operator  with the zero-modes omitted. One has
Y 1
det 0  = (n − mτ)(n − mτ̄)
α 0 τ2
(m,n)6=(0,0)
Y Y
= α 0 τ2 n2 (n − mτ)(n − mτ̄)
n6=0 m6=0,n∈Z
Y
0 2
= α τ2 (2π) (n − mτ)(n + mτ)(n − mτ̄)(n + mτ̄)
m>0,n∈Z
Y
2
2
= 4π2 α 0 τ2 e−iπmτ − eiπmτ e−iπmτ̄ − eiπmτ̄ (6.52)
m>0

151
String interactions

We introduce now the complex variable

q = e2iπτ (6.53)

and obtain finally

Y
det 0  = 4π2 α 0 τ2 (qq̄)−m (1 − qm )2 (1 − q̄m )2
m>0
Y
= 4π α τ2 (qq̄)1/12
2 0
(1 − qm )2 (1 − q̄m )2 (6.54)
m>0

The result can be expressed in terms of the Dedekind eta function

Y
∞
η(τ) = q1/24 (1 − qn ) (6.55)
n=0

such that the regularized determinant is given by

det 0  = 4π2 α 0 τ2 η2 (τ)η̄2 (τ̄) . (6.56)

Putting everything together, the path integral for the free boson on the two-torus becomes,
at the end of this tedious computation

Vx 1 1
Zx (τ, τ̄) = √ . (6.57)
(4π α 0 )1/2
2 τ2 η(τ)η̄(τ̄)

The Dedekind eta function is widely used by mathematicians in number theory and has
some wonderful properties. In particular, it transforms in a nice way under the modular
group PSL(2, Z). For the two generators one has
√
η(τ + 1) = eiπ/12 η(τ) , η(−1/τ) = −iτη(τ) (6.58)

from which we deduce that

Vx 1 1
Zx (τ + 1, τ̄ + 1) = √ = Zx (τ, τ̄) (6.59a)
(4π α 0 )1/2
2 τ2 η(τ + 1)η(τ + 1)
Vx 1 1
Zx (−1/τ, −1/τ̄) = = Zx (τ, τ̄) , (6.59b)
0
=(−1/τ) η(−1/τ)η(−1/τ)
2 1/2
p
(4π α )

hence the partition function is invariant under the action of the full modular group as it
should.

152
String interactions

Hamiltonian perspective
The partition function is somewhat easier to compute from a Hamiltonian perspective. Recall
that the Hamiltonian of a CFT is given by
c + c̄
H = L0 + Le0 − (6.60)
24
while the rotations around the cylinder are generated by
R = L0 − L
e0 . (6.61)
The coordinate σ2 of the torus has the interpretation of an Euclidean (worldsheet) time,
and because of the periodicity σ2 ∼ σ2 + 2π one considers actually a two-dimensional QFT
at finite inverse temperature β = 2πτ2 . For a generic torus, the real part τ1 of the modular
parameter is non-vanishing an correspond to a ”twist” of the torus along the space-like com-
pact coordinate σ1 before gluing both ends of the cylinder along σ2 . Since the eigenvalue of
R = L0 − L̄0 for a given state corresponds to its spin, it will pick a corresponding phase in the
partition function. In other words, one can view χ = 2πτ1 as a ”chemical potential” for the
generator of rotations around the worldsheet.
From these considerations one can reinterpret the path integral over the field x(σi ) on a
two-dimensional toroidal surface of modulus τ as the statistical partition function:


−βH iχR L0 − c+c̄
−2πτ2 L0 +e 2iπτ (L − L )


Zx = Tr e e = Tr e e 1 0 0
, (6.62)
f
24

leading to the important result:

 c c̄

L0 − 24 L0 − 24
Zx = Tr q q̄ . (6.63)
e

In the present case, c = c̄ = 1, but this expression holds for any conformal field theory on a
two-torus.
This expression can be easily evaluated by enumerating all the states of the Hilbert space
of the CFT. Recall that the spectrum of the theory is
α 0 p2 X
∞
h= +N , N= rNr , (6.64a)
4 r=1
0 2 X
∞
e = α p +N
h e , N
e = rN
er , (6.64b)
4 r=1

corresponding to states of the form

|Ψi = · · · (α−2 )N2 (e α−1 )N1 |pi .

α−2 )N2 (α−1 )N1 (e (6.65)
e e

In the partition function (6.63) one can first perform the Gaussian integral over the
momentum p. It gives simply
Z Z
Vx α 0 p2 /4 Vx 0 2 Vx 1
dp (qq̄) = dp e−πτ2 α p = √ , (6.66)
2π 2π 2π α 0 τ2

153
String interactions

where we have included the usual density of momentum modes in a volume Vx .

Next one can consider the contributions from the action of the first level left-moving
oscillator, i.e. of (α−1 )N1 for all N1 ∈ N, on the highest weight states |pi. Since α−1 shifts
the eigenvalue of L0 by one unit, these terms contribute to the partition function as
1
1 + q + q2 + · · · = , (6.67)
1−q
where the second term comes from the contribution of α−1 |pi to the trace, and so on. In the
same way one computes the contribution from the action of (α−2 )N2 for all N2 ∈ N. Since
α−2 shifts the eigenvalue of L0 by two units these terms give
1
1 + q2 + q4 + · · · = . (6.68)
1 − q2
One can iterate this reasoning to the other oscillators, and add the contribution from the
right-moving modes, leading finally to the partition function

Vx 1 Y∞
1 Y
∞
1
−1/24
Zx (τ, τ̄) = √ 0 (qq̄) (6.69)
2π α τ2 n=1
1 − q n 0 =1 1 − q̄n 0
n

Vx 1
=√ . (6.70)
4π2 α 0 τ2 ηη̄
This is precisely the same as the path-integral result (6.57).

Partition function for the xµ fields of string theory

From the partition function of a single free massless scalar field (6.57) it is easy to get the
result for the fields {xµ , µ = 0, . . . , 25} of bosonic string theory.
One should be careful however about the field x0 which has a wrong-sign kinetic term.
One way to deal with this issue formally is to replace α 0 7→ α 0  in the computation of the
path integral for this field, and take the analytic continuation  → −1 at the end of the
computation. In this way one gets
iV26 1
Zxµ (τ, τ̄) = . (6.71)
(4π2 α 0 τ2 )13 η26 (τ)η̄26 (τ̄)

6.3.2 Path integral of the ghost CFT

We consider now the path integral over the ghost fields
Z
Zg = DbDbDcDee ce−Sg b(0)b(0)c(0)e
e c(0) . (6.72)

It is more convenient to do directly the computation in the Hamiltonian formalism, which

had been successful in giving the result for the xµ fields rather quickly.

154
String interactions

Fermionic path integrals

We need to recall a property of path integrals over fermonic fields. Let us consider a simple
two-state system, the zero-mode part of the (b, c) ghost system, with

{b0 , c0 } = 1
b0 |−i = 0 , c0 |−i = |+i
c0 |+i = 0 , b0 |+i = |−i . (6.73)

The eigenstates of the lowering operator b0 can be expanded as

|ψi = |−i + |+iψ , b0 |ψi = |−iψ = |−i + ψ|+i ψ = |ψiψ ,

(6.74)

using that ψ is a Grassmann variable. We define the conjugate states to satisfy the orthog-
onality condition
hψ|ψ 0 i = ψ − ψ 0 . (6.75)
The right-hand side is the Grassmann Dirac distribution since, for any test function f,
Z
dψ(ψ − ψ 0 )f(ψ) = f(ψ 0 ) (6.76)

as one can see expanding f as f(ψ) = f0 + f1 ψ and using

Z Z 
dψ ψ ψ = − dψ ψ ψ 0 = −ψ 0 .
0
(6.77)

Expanding |ψ 0 i = |−i + |+iψ 0 one finds from (6.75) the components

hψ|−i = ψ , hψ|+i = −1 . (6.78)

Using
hψ| b0 |ψ 0 i = hψ|ψ 0 iψ 0 = ψψ 0 = hψ|b0 |ψ 0 i ,



(6.79)
One should have that
hψ|b0 = −ψhψ| (6.80)
Let us consider now an arbitrary operator A, whose components are

A|±i = |−iA−± + |+iA+± . (6.81)

We compute then the expectation value

Z Z  
dψ hψ|A|ψi = dψ hψ| |−iA−− + |+iA+− + |−iA−+ ψ + |+iA++ ψ
Z



= dψ ψ A−− + A−+ ψ − A+− + A++ ψ

= A−− − A++ . (6.82)

155
String interactions

To interpret this result, let us define the fermion number parity operator (−)F , where F counts
the number of fermionic excitations mod two. We have
 
(−)F |±i = ∓|±i , (−)F , b0 = (−)F , c0 = 0 (6.83)

Then one can interpret (6.82) in terms of a trace

Z
dψ hψ|A|ψi = Tr (−)F A = A−− − A++ .


(6.84)

It implies that, for the path integral, if one identifies periodically the field ψ(t) along the
Euclidean time circle, one does not get the partition function of the theory. Using eqn. (6.84)
for the evolution operator in Euclidean time, one gets instead:
Z
dψ hψ, T |ψ, 0i = Tr (−)F e−TH .


(6.85)

The partition function would correspond instead to a path integral with antiperiodic boundary
conditions along Euclidean time. More details can be found in appendix A of [1].
In the present case, the ghost path integral is necessarily over periodic field configurations,
as they were obtained by changing the grassmanity of the bosonic variables appearing in the
Faddeev-Popov determinant.

Trace over the ghost Hilbert space

The CFT for the ghosts factorizes into holomorphic and anti-holomorphic parts. Let us start
with the former contribution. Expanding the insertions of b(0) and c(0) in (6.72) into modes,
one has to compute terms of the form
c
Tr qL0 − 24 (−)F bn cm


(6.86)

with c = −26. The space of states is constructed from the two vacua |−i and c0 |−i = |+i,
by acting with the fermionic creation operators b−n and c−n .
Since (−)F |−i = |−i and (−)F |+i = −|+i, whenever (n, m) 6= (0, 0) the trace (6.86)
vanishes, as the terms from the states constructed out of the vacuum |−i have opposite sign
as similar terms constructed from |+i.
The exception to the rule is for m = n = 0, since we have then b0 c0 |−i = |−i, b0 c0 |+i = 0,
which gives a projector onto the ground state |−i. We have finally
 13
 X   13
 
Tr qL0 + 12 (−)F b0 c0 = h−| (b†−1 )M1 (c†−1 )N1 · · · qL0 + 12 (−)F · · · (c−1 )N1 (b−1 )M1 |−i
Nr ,Mr
(6.87)
where (remembering that the ghost vacuum has conformal dimension minus one):

X
!
   
N1
L0 · · · (c−1 ) (b−1 ) M1
|−i = r(Nr + Mr ) − 1 N1
· · · (c−1 ) (b−1 ) M1
|−i . (6.88)
r

156
String interactions

Because of the (−)F insertion, which anti-commutes with all oscillators b−n and c−n , the
contribution to the trace (6.87) of a given mode b−r contributes to the trace (6.87) as 1 − qr ,
consistently with the Fermi-Dirac statistics. Adding the contributions from all b−n and c−n ,
we obtain then
!2
 13
 13
Y
∞
1
 1 2
Tr qL0 + 12 (−)F b0 c0 = q 12 −1 (1 − qr ) = q 12 q− 24 η(τ) . (6.89)
r=1

In summary, the trace over the ghost Hilbert space give the following results
c
Tr qL0 − 24 (−)F = 0 ,


(6.90a)
c
Tr qL0 − 24 (−)F bn cm = 0 , ∀(m, n) 6= (0, 0) ,


(6.90b)
c
Tr qL0 − 24 (−)F b0 c0 = η2 (τ) .


(6.90c)
In the path integral perspective, the trace (6.90a) vanishes as one does not saturate the zero-
modes present in the integration measure. The vanishing of (6.90b) justifies that, as was
claimed above eqn. (2.132), only the ghost zero-modes contribute to the insertions appearing
in the Polyakov path integral on the sphere.
Putting together the left- and right-moving contributions from the holomorphic and anti-
holomorphic ghosts, we have obtained finally that the ghost past integral (6.72) on the two-
torus gives Z
Zg = DbDbDcDee ce−Sg c(0)e
c(0)b(0)b(0)
e = η2 η̄2 . (6.91)

6.3.3 One-loop vacuum amplitude in bosonic string theory

The one-loop amplitude is given by integrating over the modulus of the two-torus the product
of the contributions from the matter part and the ghost part. Putting everything together,
one gets: Z 2
dτ iV26 1
Z1 = 2 0 13 26 26 (τ̄)
η2 (τ)η̄2 (τ̄) (6.92)
F 4τ 2 (4π α τ 2 ) η (τ)η̄
One sees that the ghost contribution cancels the oscillators from two of the scalar fields xµ .
One can interpret this result by considering that the contributions from the light-cone co-
ordinates x± (σi ) = x0 (σi ) ± x1 (σi ) are removed by the gauge symmetry, leaving only the
degrees of freedom from the transverse fields {xi (σ), i = 2, · · · 25}.

The final result for the one-loop amplitude of bosonic string theory is:
Z 2
iV26 dτ 1
Z1 = 2 0 13 2 √
24 (6.93)
(4π α ) F 4τ2 τ2 η(τ)η̄(τ̄)
An important consistency check of this result is invariance under the modular group. One
√
observes that τ2 η(τ)η̄(τ̄) is modular invariant by itself, as follows from equation (6.58). It
is easy to check as well that d2 τ/τ22 is a modular-invariant measure over the moduli space of
the two-torus.

157
String interactions

6.3.4 Interpretation of the one-loop vacuum amplitude

The content of the one-loop amplitude of string theory can be understood easily if one
compares the result (6.93) with the analogue massive point-particle expression given by
eqn. (2.44). Generalizing the latter to a theory having several (non-interacting) massive
bosonic particle species, and taking the dimension of space-time to be 26, the QFT parition
function we take as a reference point is
Z∞
dT X − T mi2
Z1 = 14
e 2 . (6.94)
0 T i

The string partition function (6.93) is integrated over the fundamental domain F of the
modular group, which excludes the UV singularity at |τ| → 0. In order to compare to the
QFT result, which does include the singularity, we will trade the integration over F – which
is hard to perform anyway – we will consider the integration over a simpler domain, the
half-strip
S = {τ ∈ H , |<(τ)| 6 21 } . (6.95)
We have observed that the contributions of the ghosts modes cancelled the contribution
of the oscillators of two of the coordinate fields xµ in (6.93), leaving only the transverse
oscillators. Let us then introduce the transverse levels N⊥ and N e ⊥ corresponding to the
action of the transverse modes {αi−n , α e i−n , i = 2, . . . , 25} on the tachyon vacuum |pµ i and
write D(N⊥ , Ne ⊥ ) the degeneracy of states with given transverse levels.
One can simplify then the ”fake” string partition function integrated over the modified
domain S as follows:
Z
^ iV26 dτ1 dτ2 1
Z1 = 2 0 13 2 √
24
(4π α ) S 4τ2 τ2 η(τ)η̄(τ̄)
Z
iV26 dτ1 dτ2 X e ⊥ )qN⊥ −1 q̄Ne ⊥ −1
= 2 0 13 14
D(N⊥ , N
(4π α ) S 4τ2
N⊥ , N
e⊥

X Z 1/2 !Z
∞
iV26 ⊥ e⊥ 2iπτ1 (N⊥ −N
e ⊥) dτ2 −2πτ2 (N⊥ +Ne ⊥ −2)
= D(N , N ) dτ 1 e 14
e . (6.96)
(4π2 α 0 )13 ⊥ ⊥ −1/2 0 4τ2
N ,N
e

The integral over τ1 enforces the level-matching condition N⊥ = N

e ⊥ . Next we recall that the
mass spectrum of the bosonic string is given by
4
mN2 ⊥ = 0
(N⊥ − 1) , (6.97)
α
and we arrive finally to the expression
Z∞
^1 = iV26 X ⊥ dτ2 −πτ2 α 0 m 2 ⊥
Z 2 0 13
D(N ) 14
e N (6.98)
(4π α ) ⊥ 0 4τ2
N

which is precisely of the same type as the QFT expression (6.94).

158
String interactions

One concludes that, away from the dangerous UV region τ → 0, the string theory partition
function behaves exactly as the partition function for a QFT with an infinite tower of massive
particles. Unlike the QFT expression though, the actual partition function (6.93) has no
ultraviolet singularities as the integration is restricted to the fundamental domain F, which
avoids the region |τ| < 1.
In the case of the bosonic string, one has to worry however about the contribution of the
tachyon which runs into the loop. The dangerous region in this case is the IR limit τ2 → +∞,
which is dominated by the lightest states circulating into the loop. One uses the expansion
τ →∞
η(τ) 2∼ q1/24 (1 − q + O(q2 )) (6.99)

to see that the integrand of the partition function behaves like

Z∞
iV26 dτ2 4πτ2
Z1 ∼ 2 −4πτ2


e + 24 + O(e ) (6.100)
(4π2 α 0 )13 4τ214

The dominant term is coming from the tachyon and diverges, while the second one, which is
convergent, corresponds to the contribution of the massless states. Given that the tachyon
has a negative mass squared, this divergence is easy to understand from the QFT perspective.
Fortunately in the superstring theories, that we are about to consider, this problem is absent
as there are no tachyons in the physical spectrum.

References
[1] J. Polchinski, String theory. Vol. 1: An introduction to the bosonic string. Cambridge
University Press, 2007.

159
Chapter 7

Introduction to superstring theories

160
Superstrings

Bosonic string theory is certainly a beautiful theory, providing the first known UV-finite
quantum theory of gravity, with a well-defined classical limit. It has however two major
drawbacks. The first is that it does not incorporates excitations corresponding to space-time
fermions, which are necessary to describe the real world and, more annoyingly, its is an
unstable theory as the lowest lying state corresponds to a space-time tachyon.
These two problems can be fortunately overcome, by adding an extra ingredient to the
construction. The presence of the tachyon, at the technical level, was due to the shift of the
ghost Hamiltonian by minus one. This normal ordering constant is similar in spirit to the
normal ordering constant of an harmonic oscillator. As in the latter case, if can be offset if
one adds degrees of freedom with a similar Hamiltonian but opposite statistics.
Since the (b, c) ghosts of bosonic string theory were coming from the gauge-fixing of a
local symmetry of the string theory action, adding to them ”partners” with opposite statistics
– i.e. Bose-Einstein statics – has some profound implication since they should be associated
as well to the gauge-fixing of a local symmetry, this time of fermionic nature. As we shall see
below, superstring theories arise by considering a string action with local supersymmetry in
two dimensions.
In the same way that bosonic string theory, which is a theory of two-dimensional gravity
coupled to matter (the coordinates fields), describes quantum gravity in space-time, the
superstring theories that are two-dimensional supergravity theory coupled to matter, will
turn out to describe quantum supergravity in space-time.

7.1 Two-dimensional local supersymmetry

The Polyakov action (2.63) describes a set of two-dimensional scalar fields coupled to two-
dimensional gravity. After gauge-fixing, one obtains a ghost system (bij , ci ) of bosonic statis-
tics. In order to get a theory with local fermionic symmetry, one introduces two-dimensional
local supersymmetry.
Since we will deal with fermions in curved two-dimensional space-time, we introduce a
two-dimensional local frame, or zweibein eia , with, as usual
ηab eia ejb = γij , γij eia ejb = ηab , (7.1)
and the inverse zweibein eai . In the following (a, b, . . .) corresponds to orthonormal frame
indices, for tensors transforming under local Lorentz transformations, and (i, j, . . .) to coordi-
nate indices, for tensors transforming under diffeomorphisms. To the usual gamma matrices
Γ a with Lorentz indices, which satisfy the algebra
{Γ a , Γ b } = 2ηab , (7.2)
we can define gamma matrices with coordinate indices
Γ i = eia Γ a , {Γ i , Γ j } = 2γij . (7.3)
The field content of the Polyakov action is a two dimensional metric γ, or equivalently
a two-dimensional zweibein eia , together with a set of scalar fields xµ (σi ), µ = 0, . . . , D − 1,

161
Superstrings

which transform as vectors under the target-space Lorentz group. In the supersymmetric
theory, one adds fermionic partners to these bosonic fields, namely:

• to the bosonic scalars xµ (σi ) one associates two-dimensional Majorana fermions ψµ (σi )
(in terms of the space-time Lorentz group, they transform as vectors)

• to the zweibein eia one associates a gravitino ζa (z, z̄), which is, in two-dimensional
terms, a Majorana fermion with an extra vector index.

The supersymmetric generalization of the Polyakov action, in a Minkowski target space-

time ηµν with no background Kalb-Ramond field, is given by the following action:
Z
1 1
p  q

S=− d2 σ − det γ α20 γij ∂i xµ ∂j xµ + 2ψ̄µ Γ i ∂i ψµ − ζ̄i Γ j Γ i ψµ 2
∂ x
α0 j µ
− ζ̄ ψ
j µ . (7.4)
8π 4

This can be generalized to arbitary background fields (Gµν , Bµν ) but leads to quite compli-
cated couplings. Note that there is no kinetic term for the gravitino, since the latter would
need to be of the form ζ̄a Γ abc ∂b ζc and Γ abc vanishes in two-dimensions; this mirrors the fact
that the kinetic term for the two-dimensional metric, or Einstein term, is topological in two
dimensions.

7.1.1 Gauge symmetries

The action (7.4), being written in a completely covariant way with respect to the two-
dimensional metric γ, is invariant under diffeomorphisms by construction.
One can check that it is invariant as well under local supersymmetric transformations,
of Grassmann parameter κ(σi ), which is a two-dimensional Majorana fermion:

δκ eai = 21 κ̄γa ζi , (7.5a)

δκ ζi = 2Di κ , (7.5b)
q
0
δκ x = α2 κ̄ψµ ,
µ
(7.5c)
q 
µ 1 i 2 µ 1 µ
δκ ψ = 2 Γ ∂ x − 2 ζ̄i ψ κ
α0 i
(7.5d)

The gravitino variation involves a covariant derivative that we give for completeness:
 
jk a 2 j 2
Di κ = ∂i κ − 12 √−−1 e
det γ i a
Γ ∂ e
j k + 1
ζ̄
4 i
Γ Γ ζj Γ κ. (7.6)

The term involving the zweibein gives the usual spin connection of fermions coupled to a two-
dimensional metric and the second term is a torsion that arises in a gravitino background.

162
Superstrings

In the bosonic string, an extra local symmetry, the Weyl symmetry, played a crucial role.
It is a symmetry of the super-Polyakov action as well, with transformations

δλ xµ = 0 (7.7a)
1
δλ ψµ = − λψµ (7.7b)
2
δλ eai = λeai (7.7c)
δλ ζi = 12 λζi (7.7d)

which is consistent in particular with the scaling dimension of the fermionic fields ψµ .
The action has an extra symmetry, of Grassmann parameter η, which can be seen as
a fermionic partner of the Weyl symmetry. The super-Weyl symmetry acts only on the
gravitino, as:
δη ζi = Γi η . (7.8)
using the gamma-matrix identity in two dimensions

Γa Γ b Γ a = Γa (2ηab − Γ a Γ b ) = 0 . (7.9)

7.1.2 Gauge fixing

As in the case of the bosonic string, one can take advantage of this gauge freedom to simplify
drastically the action. We will start by considering the gauge transformations that cannot
be anomalous in this setting, local supersymmetry and diffeomorphism invariance.
One can first gauge away some components of the gravitino with a local symmetry trans-
formation. Since ζi has a vector index, it cannot be absorbed completely by κ however. We
start by decomposing the gravitino into its ”traceless” (helicity ±3/2) and ”trace” (helicity
±1/2) parts:
ζi = χi + 12 Γi λ , (7.10)
with Γ i χi = 0 and λ = Γ i ζi . Explicitely one has

χi = hij − 12 Γi Γj ζj = 12 Γ j Γi ζj .


(7.11)

whose contraction with Γ i indeed vanishes using (7.9). Using the identity (7.9), one can then
find a spinor ξ such that, locally, χi = Γ j Γi Dj ξ.
Next the supersymmetry transformation of the gravitino δκ ζi = 2Di κ can be decomposed
in the same way:
2Di κ = Γ j Γi Dj κ + Γi Γ j Γj κ. (7.12)
which indicates that one can absorb ξ by a local supersymmetric transformation. Using now
diffeomorphism invariance to bring the two-dimensional metric to a conformally flat metric,
one reaches finally the superconformal gauge:

γij = e2ω ηij , ζi = 21 Γi λ . (7.13)

163
Superstrings

If the remaining local symmetries (Weyl and super-Weyl) are non-anomalous, one can simplify
the theory further by taking
γij = ηij , ζi = 0 . (7.14)
As in the case of the bosonic string, there is a slight mismatch between the space of
possible geometries (here it is the moduli space of super-Riemann surfaces) and the space of
gauge configurations. Conformal Killing vectors were gauge transformations leaving invariant
the metric. In the same way, a conformal Killing spinor is defined as a globally defined two-
dimensional spinor ρ that satisfies the equation
Γ j Γi D j ρ = 0 . (7.15)
In other words, ρ is in the kernel of the projection operator onto traceless vector-spinors:
ρ ∈ Ker Π , Π(ρ)i = 12 Γ j Γi Dj ρj (7.16)
The superconformal gauge is invariant under a local supersymmetry transformation with a
conformal Killing spinor.
Moduli were defined as parameters of the metric that could not be gauged away, since
they were orthogonal to gauge transformations. They were found by asking that the variation
of metric moduli are orthogonal to all gauge transformations, see eqn. (2.97). In the same
way, one defines supermoduli as the variations δζi of the gravitino that are orthogonal to
the traceless gauge transformations Π(κ). In other words,
δζi ∈ Ker Π† . (7.17)
As for the bosonic moduli, they are related to the genus of the surface by
#(supermoduli) − #(conformal Killing spinors) = dim Ker Π† − dim Ker Π = 2g − 2 . (7.18)
When there are vertex operators inserted in the superstring path integral the surface is
punctured and this formula is modified.
On the two-dimensional sphere, the conformal Killing vectors generates the group PSL(2, C)
as we have seen already, while there are two conformal Killing spinors whose explicit descrip-
tion will not be needed here.

7.1.3 Supersymmetric action in the superconformal gauge

In the superconformal gauge, the super-Polyakov action (7.4) simplifies dramatically. After
rescaling the fermionic fields as ψ 7→ e−ω ψ, one has
Z  
1 2 1 ij µ µ i
S=− dσ η ∂i x ∂j xµ + ψ̄ Γ ∂i ψµ . (7.19)
4π α0
which is the theory of D free bosons and D free Majorana fermions. In Euclidian space and
complex coordinates one gets the fermionic string action:
Z  
1 2 2 µ¯ µ¯ µ
S= dz ∂x ∂xµ + ψ ∂ψµ + ψe ∂ψeµ (7.20)
4π α0

164
Superstrings

where ψµ and ψ̃µ are one-component Majorana-Weyl spinors of opposite chirality.

As we know from chapter 4, this action is invariant under conformal transformations
¯ = 0, which is a residual symmetry after gauge fixing of the diffeomorphism
z 7→ z+ι(z) with ∂ι
and Weyl invariance. In contrast with the discussion about the moduli and conformal Killing
vectors, we don’t ask at this stage that this transformation is well-defined globally.
In the same fashion the action is invariant under supersymmetric transformations (7.5) of
parameter κ such that the variation of the gaugino, eqn. (7.5b), can be offset by a super-Weyl
transformation (7.8). By definition, this is a transformation such that Π(κ) = 0, using the
definition (7.16). In complex coordinates, splitting the spinor κ into its components that we
denote (κ, eκ), one gets the simple expressions:
q
0
δκ,κ̄ x = − α2 κ(z)ψµ + κ̄(z̄)ψ
µ eµ


(7.21a)
q
δκ ψµ = − α20 κ(z)∂xµ (7.21b)
q
δκ̄ ψ ¯ µ
e µ = − 20 κ̄(z̄)∂x (7.21c)
α

¯ = ∂κ̄ = 0. This defines super-conformal transformations. Super-conformal and

where ∂κ
conformal transformation are not independent, since for instance

[δκ , δκ 0 ] = δι , ι(z) = −2κ(z)κ 0 (z) (7.22)

where the right-hand side is an infinitesimal conformal transformation of parameter ι. Like-

wise, the commutator of a conformal transformation with a superconformal transformation
gives another superconformal transformation. A quantum field theory invariant under su-
perconformal transformations – and, as a consequence, under conformal transformations – is
called a superconformal field theory, or SCFT.
Particular cases of superconformal transformations are global supersymmetric transfor-
mations, corresponding to constant κ and e κ.

Superspace formulation
One can conveniently repackage the fermionic string action by introducing the notion of
superspace. The two-dimensional Euclidian superspace is obtained by adding to the usual
coordinates (z, z̄) Grassmann coordinates (θ, θ̄). The global supersymmetric transformations
are generated by
Q = ∂θ − θ∂z , Q̄ = ∂θ̄ − θ̄∂z̄ (7.23)
which satisfy the algebra

{Q, Q} = −2∂z , ¯z ,
{Q̄, Q̄} = −2∂ {Q, Q̄} = 0 , (7.24)

hence acting twice with the supersymmetry transformation Q (resp. Q̄) gives a holomorphic
(resp. anti-holomorphic) translation.

165
Superstrings

Covariant derivatives with respect to the fermionic coordinates are defined such that they
anti-commute with the supersymmetry generators (7.23)

Dθ = ∂θ + θ∂z , Dθ̄ = ∂θ̄ + θ̄∂z̄ (7.25)

which satisfy the algebra

{Dθ , Dθ } = 2∂z , ¯z ,
{Dθ̄ , Dθ̄ } = 2∂ {Dθ , Dθ̄ } = 0 . (7.26)

Since they commute with the supersymmetry generators, they can be used to write super-
symmetric actions.
One introduces superfields that are local functions of (z, z̄, θ, θ̄). Because of the Grass-
mann nature of the odd coordinates, their Taylor expansion is finite. In the present case let
us define q
µ
X (z, z̄, θ, θ̄) = α20 xµ (z, z̄) − θψµ − θ̄ψ
e µ + θθ̄Fµ , (7.27)

where, if xµ is a bosonic field, ψµ and ψ e µ should be fermionic fields. Fµ is an auxiliary field

that vanishes upon imposing the equations of motion as we will see shortly.
The action of the supercharges on the superfields Xµ give the global supersymmetry
transformations (see eqns. (7.21):
q
κQ X = −κψ + θκ α20 ∂xµ + κθ̄Fµ + κθθ̄∂
µ µ ¯ψ
eµ (7.28a)
q
κ̄Q̄ Xµ = −κ̄ψ ¯ µ − κ̄θθ̄∂ψ
e µ − κ̄θFµ + θ̄κ̄ 20 ∂x ¯ µ (7.28b)
α

upon imposing the equations of motion Fµ = 0 and ∂ψ ¯ µ = 0. A supersymmetric

e µ = ∂ψ
action is then obtained as1
Z
1
S= d2 zd2 θ Dθ Xµ Dθ̄ Xµ (7.29)
4π
Z  
1 2 2 µ¯ µ µ¯ µ µ eµ µ
= dz ∂x ∂x + ψ ∂ψ + ψ ∂ψ + Fµ F
e . (7.30)
4π α0

The equation of motion of the auxiliary field Fµ is simply Fµ = 0 hence the auxiliary field
vanishes on-shell as expected.

7.1.4 Ghosts and superghosts

In the bosonic string path integral, gauge-fixing of the diffeomorphism invariance introduced,
through the Faddeev-Popov determinants, the fermionic ghosts system (bij , cj ). Because the
Weyl transformations involve no derivatives of the gauge parameter, the corresponding ghost
was just a Lagrange multiplier imposing that bij was traceless.
1
R
The integration measure over the Grassmann coordinates is defined such that d2 θ θθ̄ = 1.

166
Superstrings

In the present case, gauge-fixing of local supersymmetric transformations leads in the

same way to a ghost system (βi , γ), where γ comes from the infinitesimal supersymmetry
variation κ and βi is the field introduced to enforce the gauge-fixing constraint in field space.
Since κ is a Grassmann variable, and that in evaluating the Faddeev-Popov determinant
the statistics of the ghosts is flipped w.r.t the gauge parameter, (βi , γ) are two-dimensional
bosons. In the same way as integrating over Weyl transformations imposes bi i = 0, integrat-
ing over super-Weyl symmetries in the FP determinant amounts to integrate over a Lagrange
multiplier imposing that βi satisfy the ”traceless” condition
Γ i βi = 0 , (7.31)
leaving only two degrees of freedom.
Instead of deriving the action for the (β, γ) ghost system from the Faddeev-Popov de-
terminant, we will take advantage of the global supersymmetry of the gauge-fixed action to
obtain it from a supersymmetrization of the (b, c) action. In superspace one starts with the
superfields
B = β + θb , (7.32a)
e=β
B e + θ̄b
e, (7.32b)
C = c + θγ , (7.32c)
e =e
C c + θ̄e
γ. (7.32d)
And write the following supersymmetric action
Z Z Z
1 2 2
  1 
¯ + b̃∂c̃ + β∂γ
¯ + β∂e

Sg = − d z d θ BDθ̄ C − BDθ C =
e e d2 z b∂c e γ . (7.33)
2π 2π
Hence the (β, γ) super-ghost CFT is defined by a first-order action similar to the (b, c) ghost
CFT. There are however two important differences:
1. The fields β and γ follow Bose-Einstein statistics.
2. (β, β)
e being the components of a ”traceless” spinor-vector field βi , their conformal
dimensions are respectively (3/2, 0) and (0, 3/2). By conformal invariance of the ac-
e have respectively conformal dimensions (−1/2, 0) and (0, −1/2).2
tion (7.33), γ and γ
The superconformal transformations between the ghost and superghost fields are deduced
from
κQB = κb + θκ∂β , (7.34a)
κQC = κc + θκ∂γ , (7.34b)
κ̄Q̄B
e = κ̄b ¯β
e + θ̄κ̄∂ e, (7.34c)
κ̄Q̄C
e = κ̄e ¯γ .
c + θ̄κ̄∂e (7.34d)
2
Another way to obtain this is to notice that the definition (7.27) implies that θ (resp. θ̄) has conformal
weights (−1/2, 0) (resp. (0, −1/2)).

167
Superstrings

The conformal theory of the (β, γ) system can be studied in close analogy with the (b, c)
system. The classical equations of motion are the same:
¯ =0,
∂β ¯ = 0.
∂γ (7.35)
There exists also a superghost rotation symmetry, β 7→ e−iα β, γ 7→ eiα γ, to which we
associate a holomorphic Noether current
jsg = − •• βγ •• (z) , ¯ sg = 0 .
∂j (7.36)
From the associated Ward identities (3.37) one finds the OPEs
β(z2 ) γ(z2 )
jsg (z1 )β(z2 ) = − + reg. , jsg (z1 )γ(z2 ) = − + reg. , (7.37)
z1 − z2 z1 − z2
coming from the fundamental OPE
1
β(z1 )γ(z2 ) = −γ(z1 )β(z2 ) = + reg. (7.38)
z1 − z2
Notice the negative sign in the γβ OPE due to the bosonic statistics, compared to the bc
OPE in equation (4.156).
The more general ansatz for the stress-tensor, compatible with conformal dimensions
(2, 0), is of the form
T (z) = u •• β∂γ •• + v •• ∂βγ •• . (7.39)
The coefficients u and v are then fixed by requiring that β and γ are primary fields of
respective holomorphic conformal weights 3/2 and -1/2. The OPE with γ gives (remembering
that these are commuting variables)
u∂γ(z2 ) vγ(z1 ) vγ(z2 ) (u − v)∂γ(z2 )
T (z1 )γ(z2 ) = − 2
+ reg. = − 2
+ + reg. (7.40)
z1 − z2 (z1 − z2 ) (z1 − z2 ) z1 − z2
This sets u = 3/2 and v = 1/2. One can check the OPE with β as well:
 
3• • 1• • 3 β(z2 ) ∂β(z2 )
T (z1 )β(z2 ) = • β∂γ • (z1 ) + • ∂βγ • (z1 ) β(z2 ) = + + reg. (7.41)
2 2 2 (z1 − z2 )2 z1 − z2
Finally the central charge of the superghost CFT is deduced from the self-OPE of the
stress tensor, more precisely from the higher order pole which is given by the fully contracted
term:
csg 9 3
4
= •• β∂γ •• (z1 ) •• β∂γ •• (z2 ) + •• β∂γ •• (z1 ) •• ∂βγ •• (z2 )
2(z1 − z2 ) 4 4
3• • • • 1•
+ • ∂βγ • (z1 ) • β∂γ • (z2 ) + ∂βγ •• (z1 ) •• ∂βγ •• (z2 )
4 4•
11
= (7.42)
2(z1 − z2 )4
Hence the central charges of the (β, γ) superghost CFT are (c, c̄) = (11, 0). Similarly the
central charges of the (β̃, γ̃) CFT are (c, c̄) = (0, 11).

168
Superstrings

7.1.5 Critial dimension of the fermionic string

The local supersymmetric action (7.4) includes the Weyl symmetries in its gauge symmetry
group, hence the considerations for cancellation of the Weyl anomaly, discussed in subsec-
tion 3.6 apply also here.
The total central charge of the worldsheet conformal field theory, which is made of the
tensor product of the xµ , ψµ , (b, c) and (β, γ) theories should vanish to cancel the anomaly.
One gets the following condition on the dimensionality of Minkowski space-time R1,D−1
for the fermionic string:

cx + cψ + cg + csg = 1 + 21 D − 26 + 11 = 0 =⇒ D = 10 .


(7.43)

The same considerations apply naturally for the right-movers, as we consider a left-right
symmetric theory. Therefore the consistent superstring theories that we will build out of
the fermionic string will be defined in a ten-dimensional space-time.
This is closer to the real world that the twenty-six dimensional space-time of the bosonic
string, but there are still six dimensions that should be compact in order to avoid direct clash
with experiments.

7.2 Superconformal symmetry

In chapter 3, using that the variation of the action (2.66) with respect to the two-dimensional
metric gave the stress-energy tensor of the CFT, we obtained that the infinitesimal genera-
tors of the conformal transformations ((z),  ¯(z̄)) were the components of the stress-tensor
(Tzz , Tz̄,z̄ ).
In the same way, the variation of the supergravity action (7.4) with respect to a gravitino
variation gives the supercurrent of the theory. In the superconformal gauge it takes the simple
form r
i δL  1 2 j i µ
G = = Γ Γ ψ ∂j xµ . (7.44)
δζ̄i ζ̄i =0 4 α 0


It has only two components since Γi Gi = 0, using again (7.9).

7.2.1 Superconformal field theory

In Euclidian space and moving to complex coordinates, the two independent components
G(z) and G(z̄),
e which are separately conserved, are given by
r
2 µ ¯ =0
G(z) = i 0
ψ ∂xµ , ∂G (7.45a)
r α
2 eµ¯
G(z̄)
e =i ψ ∂xµ , ∂G e = 0. (7.45b)
α0

169
Superstrings

One gets Ward identities for the superconformal transformations similar to the conformal
one, see eqn. (3.37)
q  
δκ ψ (z) = − α20 κ(z)∂xµ (z) = iResz 0 →z κ(z 0 )G(z 0 )ψµ (z) ,
µ
(7.46)

an a similar one for the anti-holomorphic superconformal transformation.

From the commutator (7.22) we know that the supercurrent and the stress-energy tensor
of the superconformal field theory should have non-trivial OPE with each other. These OPE
will encode an infinite-dimensional algebra, the superconformal algebra.
Let us consider the free theory (7.20) in D dimensions. One computes first the OPE of a
supercurrent G with itself:
2 • µ
G(z1 )G(z2 ) = − ψ ∂xµ •• (z1 ) •• ψν ∂xν •• (z2 )
α 0 •
ηµν −α 0
 
2 • µ ν • 1 • •
=− 0 + • ψ (z1 )ψ (z2 ) • ηµν + • ∂xµ (z1 )∂xµ (z2 ) •
α z1 − z2 2 (z1 − z2 )2
ηµν ηµν
 
1 2 • • • µ ν•
= + − 0 • ∂xµ ∂xµ • (z2 ) + ηµν • ∂ψ ψ • (z2 ) + reg.
(z1 − z2 )3 z1 − z2 α
D 2 x ψ


= + T + T + reg. (7.47)
(z1 − z2 )3 z1 − z2

where we have used (see chapter 3)

1 • µ
T x (z) = − ∂x ∂xµ •• (z) , (7.48a)
α0 •
1
T ψ (z) = − •• ψµ ∂ψµ •• (z) . (7.48b)
2
Next the OPE between the stress-energy tensor and the supercurrent gives
  r
1 • µ 1 2 µ
T (z1 )G(z2 ) = − 0 • ∂x ∂xµ •• (z1 ) − •• ψµ ∂ψµ •• (z1 ) i ψ ∂xµ (z2 )
α 2 α0
r
2 ∂xµ (z1 )ψµ (z2 ) i ∂ψµ (z1 )∂xµ (z2 ) i ψµ (z1 )∂xµ (z2 )
=i + √ + √
α 0 (z1 − z2 )2 2α 0 z1 − z2 2α 0 (z1 ! − z2 )2
r r
3 1 2 µ 1 2 µ
= i 0
ψ ∂xµ (z2 ) + ∂z 2 i ψ ∂xµ (z2 )
2 (z1 − z2 ) 2 α z1 − z2 α0
3 1
= 2
G(z2 ) + ∂G(z2 ) . (7.49)
2(z1 − z2 ) z1 − z2

In other words, G is a conformal primary of conformal dimension h = 3/2.

Noticing that the CFT of D scalars ψµ and D fermions ψµ has central charge c = 3D/2,
one can infer the general superconformal OPE, which is given by:

170
Superstrings

c 2T (z2 ) ∂T (z2 )
T (z1 )T (z2 ) = + + + reg. (7.50a)
2(z1 − z2 )4 (z1 − z2 )2 z1 − z2
2c 2T (z2 )
G(z1 )G(z2 ) = + + reg. (7.50b)
3(z1 − z2 )3 z1 − z2
3G(z2 ) ∂G(z2 )
T (z1 )G(z2 ) = + + reg. (7.50c)
2(z1 − z2 )2 z1 − z2

7.2.2 Representations of the superconformal algebra

In order to obtain the infinite-dimensional algebra coming from the OPEs (7.50) one expands
first T and G in Laurent modes. We know already that
X Ln c
T (z) = n+2
, [Lm , Ln ] = (m − n)Lm+n + m(m2 − 1)δm+n,0 . (7.51)
n∈Z
z 12

The supercurrent G is a composite operator of the fermionic fields ψµ (z) and the bosonic
fields ∂xµ . Remember that the Laurent expansion of the former is
X ψµr
ψµ (z) = , Neveu-Schwarz sector (7.52a)
zr+1/2
r∈Z+1/2
X ψµ
µ r
ψ (z) = r+1/2
, Ramond sector (7.52b)
r∈Z
z

Hence the supercurrent G will have half-integer moded Laurent modes in the Neveu-Schwarz
sector (NS) and integer moded ones in the Ramond (R) sector. Accordingly we expand the
supercurrent as
X Gr
G(z) = r+3/2
, (7.53)
1−a
z
r∈Z+ 2

with a = 0 (resp. a = 1) in the Neveu-Schwarz (resp. Ramond) sector. Since the modes ψµr
and αµn commute with each other, there are no ordering ambiguities in the expansion and we
find that in the present case X
Gr = ηµν αµn ψνr−n . (7.54)
n∈Z

The (anti)commutation relations of the superconformal algebra can be deduced from the
OPE (7.50), or equivalently from the commutation relations of the αµn , see eqn. (4.34), and
of the ψµr , see eqn. (4.107). Either way, one finds

[Lm , Ln ] = (m − n)Lm+n + c
12
m(m2 − 1)δm+n,0 (7.55a)
{Gr , Gs } = 2Lr+s + 12c (4r2 − 1)δr+s,0 (7.55b)
[Lm , Gr ] = m2 − r Gm+r .


(7.55c)

171
Superstrings

Representation theory of the superconformal algebra has the same kind of structure as
representation theory of the Virasoro algebra. The superconformal primary states are an-
nihilated by the positive modes of the currents, however the presence of zero-modes in the
Ramond sector needs to be taken into account carefully.
Finally, it is worthwile to mention that the Ramond boundary conditions preserve two-
dimensional supersymmetry, as they relate bosons to fermions of identical periodicities, while
the Neveu-Schwarz boundary conditions break two-dimensional supersymmetry at the global
level, since they relate periodic bosons to anti-periodic fermions.

Neveu-Schwarz sector
In the Neveu-Schwarz sector, the modes of the supercurrent {Gr , r ∈ Z + 1/2} are all half-
integers so we define a superconformal primary as

Lm |Ψi = Gn+1/2 |Ψi = 0 , ∀m, n ∈ N . (7.56)

Descendant states are obtained then by acting with the negative modes L−n and G−n+1/2 ,
n > 0.
In the present context, the primary states of momentum pµ are given by the tensor product
of the conformal primaries |pµ i for the xµ CFT and the NS vacuum |0ins for the ψµ CFT:

|pµ ins = |pµ i ⊗ |0ins . (7.57)

Using equations (4.46) and (4.121) its conformal dimensions is given by

  α 0 p2 µ
L0 |pµ ins = Lx0 + Lψ
0 |p µ
i ⊗ |0i ns = |p ins . (7.58)
4
The excited states of the lowest conformal dimension arise at level N = 1/2. They
correspond to
 0 2 
αp 1
ψ−1/2 |p ins = |p i ⊗ ψ−1/2 |0ins , L0 ψ−1/2 |p ins =
ρ µ µ ρ ρ µ
+ |pµ ins . (7.59)
4 2
These states transform in the vector representation of SO(1, D − 1), as explained in subsec-
tion 4.2.3. These states are conformal primaries, i.e. annihilated by all positive Virasoro
modes.
However for generic pµ they are descendant states w.r.t. the modes of the supercurrent.
Explicitly one has, for a mode with polarization vρ , forgetting about the right-moving sector
at this stage,

G1/2 vρ ψρ−1/2 |pµ ins = ηµν αµ0 ψν1/2 + · · · |pµ i ⊗ ψρ−1/2 |0ins


q
0
= α2 vµ pµ |pµ i ⊗ ψρ−1/2 |0ins . (7.60)

We will be interested by states of the form (7.59) that are superconformal primaries. From
this computation it occurs for vµ pµ = 0.

172
Superstrings

As we will see, all higher-level states, including those constructed with the action of αµ−1 ,
are massive in superstring theory, hence will not be discussed in detail.
We will only make the following important remark. In subsection 4.2.3 we considered on
the one hand states in the trivial representation of the current algebra, obtained by acting
with Jij−n on the NS vacuum, and on the other hand the states obtained by acting on the
vector representation ψi−1/2 |0ins with Jij−n . From the point of view of the superconformal
algebra, it amounts to split representation in two, keeping states which are even under (−)F
in the first case and odd in the second case.

Ramond sector
In the Ramond sector, there exists a zero mode G0 of the supercurrent. From the commuta-
tion relation (7.55b) one finds that
c
G02 = L0 − , (7.61)
24
hence, we get, for any state |Ψi in a unitary CFT
2 c  c   2
0 6 G0 |Ψi = hΨ|L0 |Ψi − hΨ|L0 |Ψi = h − |Ψi ,

(7.62)
24 24
so in the Ramond sector of a unitary CFT all the conformal dimensions of operators obey
the bound
c
h> , (7.63)
24
which is saturated by the Ramond ground states with h = c/24. For a theory of N free
bosons and fermions, this is essentially the content of eqn. (4.140).
Using the results of chapter 4.2, the fermionic zero-modes give an algebra isomorphic to
the Clifford algebra in 1 + 9 dimensions:

{ψµ0 , ψν0 } = ηµν , (7.64)

and the Ramond ground state of the ψµ CFT corresponds to a (reducible) Majorana repre-
sentation of this algebra.
For Spin(1, 9) this is a 32-dimensional representation. It decomposes into a pair of
irreducible Majorana-Weyl representations of dimensions 16,

32 = 16 + 16 0 , (7.65)

which are denoted spinor and conjugate spinor representations respectively (see chapter 4.2).
The corresponding states, |αir and |α̇ir , are distinguished by the eigenvalue of the chirality
operator (−1)F which anticommutes with all fermionic operators:

(−1)F |αir = |αir , (−1)F |α̇ir = −|α̇ir . (7.66)

173
Superstrings

We decompose accordingly the ten-dimensional gamma-matrices as

 
µ b 0 Γ µα β̇
(Γ )a = (7.67)
Γ µα̇ β 0

In the fermonic string context, the lowest-dimension states are therefore obtained as the
tensor product of the momentum eigenstates |pµ i with the Ramond ground states:

|pµ ; αi = |pµ i ⊗ |αir , |pµ ; α̇i = |pµ i ⊗ |α̇ir . (7.68)

These states are by construction conformal primaries. It is easy to check as well that they are
annihilated by G1 and all other positive modes of the supercurrent. All the Ramond sector
excited states give massive states in space-time, hence will not be considered further.

7.2.3 Superghosts mode expansion

The (β, γ) superghosts are spinors (with Bose-Einstein statistics) so they can have also either
periodic and anti-periodic boundary conditions on the cylinder. Because the field γ came
originally from the supersymmetry parameter κ, consistency of the transformations (7.5)
require that they obey the same boundary conditions as the fields ψµ :

• In the Neveu-Schwarz sector of the ψµ fields, (β, γ) will have anti-periodic boundary
conditions on the cylinder

• In the Ramond sector of the ψµ fields, (β, γ) will have periodic boundary conditions on
the cylinder

The superghosts are therefore expanded as

X βn X γn
β= , γ= , (7.69)
zn+3/2 z n−1/2
n∈Z+ 1−a
2
n∈Z+ 1−a
2

and the modes obey the algebra

[γm , βm ] = δm+n,0 . (7.70)
We consider the superconformal field theory made of the ghosts (b, c) and the superghosts
(β, γ) first in the Neveu-Schwarz sector. The modes of the superconformal algebra are
X X
◦ ◦

◦ ◦
∀n 6= 0 , Ln = Lgn + Lsgn = (2n − m) ◦ b m cn−m ◦ + 3
2
n − m ◦ βm γn−m ◦ , (7.71)
m∈Z m∈Z+ 12

where the first term was computed in (4.178) and the second follows from a similar calculation.
As usual there exists a normal ordering ambiguity for the zero-mode L0 . We have computed
already the normal ordering constant coming from the (b, c) ghosts, and the contribution
from the (β, γ) superghosts follows from a similar reasoning.

174
Superstrings

The superghost vacuum |−isg ns is annihilated by all the positive modes of the ghost and
superghost fields (similar statements hold for the anti-holomorphic superghosts):

∀n > 0 , βn+1/2 |−isg

ns = 0 , γn+1/2 |−isg
ns = 0 , (7.72a)
∀n > 0 , bn |−ins = 0 , ∀n > 0 , cn |−isg
sg
ns = 0 . (7.72b)

We have the following relation:

0 |−ins = L1 L−1 |−ins = β1/2 γ1/2 + · · ·

2Lsg −β−1/2 γ−1/2 + · · · |−isg
ns = |−ins .
sg sg sg



(7.73)

Adding the normal ordering constant of the ghosts (b, c) that we have already computed,
one finds that in the Neveu-Schwarz sector
X X 1
Lg,sg
0 = Lg0 + Lsg
0 = − m ◦◦ bm c−m ◦◦ − m ◦◦ βm γ−m ◦◦ − . (7.74)
m∈Z 1
2
m∈Z+ 2

The Neveu-Schwarz vacuum of the ghost superconformal field theory, which has conformal
dimension h = −1/2, is annihilated by all positive modes of T and G:

∀n > 0 , Ln |−isg
ns = 0 , n−1/2 |−ins = 0 .
Gsg sg
(7.75)

where the supercurrent of the ghost superconformal field theory is given by

3
Gg,sg (z) = −∂βc − β∂c − 2bγ . (7.76)
2
This can be obtained by requiring that the stress-energy tensor T g,sg (z) = T g (z) + T sg (z) and
G(z) obey the OPEs (7.50) with c = −15.
We now move to the Ramond sector of the ghost SCFT, where the mode expansion of
the Virasoro generators in now
X X
◦ ◦
∀n 6= 0 , Ln = Lgn + Lsg 3
n − m ◦◦ βm γn−m ◦◦ . (7.77)


n = (2n − m) b c
◦ m n−m ◦ + 2
m∈Z m∈Z

While nothing changes for the (b, c) ghosts the (β, γ) superghosts acquire zero-modes (β0 , γ0 ).
The Ramond superghost vacuum will be defined by

∀n > 0 , βn |−isg
r = 0 , ∀n > 0 , γn |−isg
r = 0, (7.78a)
∀n > 0 , bn |−ir = 0 ,
sg
∀n > 0 , cn |−ir = 0 .
sg
(7.78b)

Again the normal ordering constant is obtained from

  
3 1 3 sg
2L0 |−ir = L1 L−1 |−ir =
sg sg sg
β 0 γ1 + · · · − β−1 γ0 + · · · |−isg
r = |−ir (7.79)
2 2 4
Hence we get in the Ramond sector
X X 5
Lg,sg
0 = Lg0 + Lsg
0 = − m ◦◦ bm c−m ◦◦ − m ◦◦ βm γ−m ◦◦ − . (7.80)
m∈Z m∈Z
8

175
Superstrings

Note that, following (4.140), this is exactly the opposite of the Ramond ground state con-
formal dimension for the ψµ CFT, in the critical dimension D = 10. So, whenever we impose
the supersymmetric Ramond boundary conditions for the fermions, it seems that we have
achieved our goal of having string ground states with zero mass. For the (supersymmetry-
breaking) Neveu-Schwarz conditions however, this is not quite the case, as the zero-point
energy was just shifted from −1 in the bosonic string to −1/2 in the fermionic string. We
will see shortly how to get rid of this potential tachyonic state.

7.3 BRST quantization

As for the bosonic string, after gauge-fixing of the local symmetries, there exists a rem-
nant global symmetry whose cohomology is identified with the physical spectrum, the BRST
symmetry.

7.3.1 BRST current

Following the general logic sketched in section 5.2 the BRST current will receive an extra con-
tribution, which is essentially the superghost γ multiplied by the variation of the action (7.4)
under local supersymmetry transformations, which is proportional to the supercurrent. An
explicit computation would give then the holomorphic and anti-holomorphic currents:
1
jb (z) = cT x,ψ + γGx,ψ + (cT g,sg + γGg,sg ) , (7.81a)
2
e x,ψ + 1 e
 
cTex,ψ + γ
eb (z̄) = e eG cTeg,sg + γ
eGe g,sg , (7.81b)
2
up to total derivative terms, as in (5.53a), that do not contribute to the charge. A local
operator O(z, z̄) would then correspond, through the state-operator correspondence, to a
physical state of the fermionic string theory if
I
1
(Qb + Qe b )O(0, 0) = (jb dz − edz̄) O(0, 0) = 0 , (7.82)
2iπ C0

up to spurious states.
One can check easily that the OPE between the current (7.81a) with itself has a single
pole unless the matter CFT satifies c = 15, i.e. provided that the dimension of space-time
is ten. Naturally the self-OPE of (7.81b) leads to the same conclusion.

7.3.2 Mode expansion of the BRST current

To work with states rather than with local operators, we need to expand the BRST currents
in terms of modes of the various elementary fields of the theory. For the holomorphic current

176
Superstrings

for instance, one finds that

X X X
n−m ◦
Qb = c−n (Lxn + Lψ
n) + γ−r Gx,ψ
r − 2 ◦
b−m−n cm cn ◦◦
n∈Z r∈Z+ 1−a m,n∈Z
X
2

2r−m ◦
β−m−r cm γr ◦◦ − ◦◦ b−m γm−r γr ◦◦ + λa c0 (7.83)


+ 2 ◦
m∈Z,r∈Z+ 1−a
2

where the normal ordering constant, which depends on the sector, was computed before:
1 5
λ0 = − , λ1 = − . (7.84)
2 8

7.3.3 BRST cohomology

Physical states in the fermionic string will be given in terms of the BRST cohomology, i.e.
by states annihilated by the sum of (7.83) and the corresponding anti-holomorphic charge,
modulo exact states that can be described as (Qb + Q e b )|χi.
In the context of the bosonic string, in order to obtain a consistent spectrum from the
space-time point of view, one had to impose the Siegel gauge (5.75). Likewise we will impose
here the following constraints on all physical states:
b0 |Ψi = b
e0 |Ψi = 0 Neveu-Schwarz sector (7.85a)
b0 |Ψi = b
e0 |Ψi = β0 |Ψi = β
e 0 |Ψi = 0 Ramond sector (7.85b)
The BRST charge (7.83) does not change the level of a state that it is applied to, hence
we can study the spectrum level by level. The analysis is complicated by the fact that one
can choose between the the Neveu-Schwarz and Ramond boundary conditions separately for
the left-movers and right-movers, hence we have four different sectors to consider: (NS, NS),
(NS, R), (R, NS), (R, R).

NS-NS sector
In this sector both the left-moving and right-moving fermions obey the Neveu-Schwarz bound-
ary conditions. At level zero in this sector one has a unique ground state

fψ f sg
e
|pµ ins-ns = |pµ ix ⊗ |0iψ
ns ⊗ |0ins ⊗ |−ins ⊗ |−ins ,
sg
e
(7.86)
i.e. the tensor product of the conformal primary of momentum pµ for the xµ CFT, the NS
vacuum for the ψµ CFT, and the superghost vacuum in the NS sector defined by eqns. (7.72)
and the similar vacua for the right-moving fermions, ghosts and superghosts. From a space-
time perspective, this state is a scalar field.
This state is annihilated by all the oscillator modes in the BRST charge (7.83), and the
only condition that remains to be imposed is:
α 0 m2 α 0 p2 1
c0 (Lx0 + Lψ 1 µ
0 − 2 )|p ins = 0 =⇒ =− =− . (7.87)
4 4 2
177
Superstrings

the right-moving component of the BRST charge gives naturally exactly the same result. So
it seems that the progress obtained by moving from the bosonic string to the superstring
was fairly modest; instead of a tachyon of mass squared m2 = −4/α 0 , we have a tachyon
of mass squared m2 = −2/α 0 . The cancellation of the zero-point energies of the ghosts and
superghosts does not work in the NS sector as these fields have different boundary conditions.
Don’t dispair yet, we will be able to solve this problem in due time.
At the next level, imposing level-matching conditions, one can build several states with
levels N = N̄ = 1/2. The general ansatz is actually similar to (5.88), with the appropriate
replacements (αµ−1 → ψµ−1/2 , b−1 → β−1/2 , etc...) and the analysis works in the same way.
The physical state is built out of the term without superghost modes, namely

|Ψ1/2 i = eµν ψµ−1/2 ψ

e ν |pρ ins-ns .
−1/2 (7.88)

Different terms coming from the BRST charge (7.83) are in different states of the ghost and
superghost CFT hence should be cancelled separately. One has first
α 0 m2 α 0 p2
c0 (Lx0 + Lψ 1
0 − 2 )|Ψ1/2 i = 0 =⇒ =− = 0. (7.89)
4 4
One has then
 
X X
1 |Ψ1/2 i = 2 c−1
c−1 Lψ 1  (r − 1/2) ◦◦ e ν |pρ ins-ns = 0 , (7.90)
ψσ1−r ψr σ ◦◦ ψµ−1/2  eµν ψ −1/2
r∈Z+1/2 r

consistently with the fact that ψµ−1/2 |0ins are the conformal primary operators in the vector
representation of SO(1, 9). The last term to consider is
 
γ−1/2 G1/2 |Ψ1/2 i = γ−1/2 ηστ α0 ψ1/2 ψ−1/2 eµν ψ
ψ,x σ τ µ e ν |pρ ins-ns = 0 ,
−1/2 (7.91)

implying, using eqn. (4.30) and the anti-commutator (4.107) that (adding the similar con-
straint from the anti-holomorphic component of the BRST charge):

pµ eµν = pν eµν = 0 . (7.92)

Finally one has to find those of the spurious states that give equivalence relations among
the states (7.88) with different polarizations. One has that
 
e µ |pρ ins-ns = vµ Gx,ψ ψ
1/2 −1/2 |p ins-ns
(Qb + Q e b )β−1/2 vµ ψ e µ + β−1/2 G
e x,ψ ψ
eµ ρ
(7.93)
−1/2 −1/2 −1/2

Next using that

q
α0
−1/2 |p i
Gx,ψ ⊗ |0ins = ηµν αµ0 ψν−1/2 |pρ i ⊗ |0ins = pµ ψµ−1/2 |pρ i ⊗ |0ins ,
ρ
2
(7.94)

and
q
α0
1/2 ψ−1/2 |p i
Gx,ψ ⊗ |0ins = ηστ ασ0 ψτ1/2 ψµ−1/2 |pρ i ⊗ |0ins = pµ |pρ i ⊗ |0ins ,
µ ρ
2
(7.95)

178
Superstrings

from which we deduce, using also the anti-holomorphic analogous equation, the equivalence
relations
eµν ∼ eµν + aµ pν + bν pµ , aµ pµ = bν pν = 0 . (7.96)
All these conditions give naturally the fluctuations of the metric, of the Kalb-Ramond field
and of the dilaton exactly as in the bosonic strings.

R-NS sector
We consider a sector of string states such that the left-moving fermions ψµ have periodic
boundary conditions on the cylinder, while the right-moving fermions ψ e µ have anti-periodic
boundary conditions. As was said before, the left and right superghosts should follow the
same pattern.
The right-moving part of the theory are described exactly in the same way as before, so
we will concentrate on the left-moving part which is in the Ramond sector. At level zero
we consider that both the superconformal field theory (xµ , ψµ ) and the superconformal field
theory (b, c; β, γ) are in their Ramond ground state. The conformal dimension of the ground
state in the tensor product of the two superconformal field theories

|pµ i ⊗ |air ⊗ |−ir (7.97)

is, using (7.80) and in D = 10 spacetime,

α0 2 D 5 α0
h= p + − = p2 . (7.98)
4 16 8 4
as we said already, the reducible representation |air of the Clifford algebra reduces in a pair
of irreducible representations |αir and |α̇ir of opposite chiralities.
By the level matching constraint h = h̄ the states cannot be paired with the NS ground
state of the right but rather with the first excited state. We are therefore looking for physical
states of the form

|Ψs i = |pρ i ⊗ |αir uαµ ⊗ (ψµ )−1/2 |0i f sg .

f ⊗ |−isg ⊗ |−i
e
ns r ns (7.99a)

e b )|Ψs i = 0 gives first the terms

The BRST constraint (Qb + Q

c0 (Lx0 + Lψ
0 − 5/8)|Ψ1 i + e
c0 (L0
eψ − 1/2)|Ψ1 i = 0 =⇒ m2 = −p2 = 0 ,
ex + L
0 (7.100)

hence these states are massless. They transform as vector-spinors of the space-time Lorentz
group. Next we have
0 |Ψs i = 0
γ0 Gψ,x (7.101)
Using G0 = ηρσ αρ0 ψσ0 , one finds that

pρ Γ ρα̇ α
uµα = 0 (7.102)

179
Superstrings

which is essentially the same as the Dirac equation. Next we have

γ e ψ,x |Ψa i = 0
e−1/2 G (7.103)
1/2

which gives as before

pµ uµα = 0 (7.104)
We now look at the spurious states. On the left, being already in the ground state there is
no freedom left. On the right one can do the same as in the NS-NS sector, so consider a state

e b )|pρ i ⊗ |αir vα ⊗ β
e −1/2 |0i f sg .
f ⊗ |−isg ⊗ |−i
e
(Qb + Q ns r ns (7.105)

which gives the equivalence relation

uµα ∼ uµα + pµ vα , pρ Γ ρα̇ α vα = 0 , . (7.106)

To summarize, one gets a massless particle whose polarization is a vector-spinor uµα that
satisfies:

pρ Γ ρα̇ α
uµα = 0 , pµ uµα = 0 , uµα ∼ uµα + pµ vα , pρ Γ ρα̇ α
vα = 0 . (7.107)

In terms of irreducible representations of the Lorentz group, this vector-spinor decomposes

first into a gamma-trace part
λα̇ = (Γµ )α̇β uµβ , (7.108)
which is a spinor, which is called a dilatino. Then, the gamma-traceless part of the vector-
spinor,
1
ζµα = uµα − (Γ µ )αβ̇ λβ̇ , (7.109)
10
corresponds to a gravitino in space-time, which is a massless particle of helicity 3/2, the gauge
field associated with local supersymmetric transformations. The equivalence relation (7.106),
that leaves the dilatino invariant by construction, corresponds to the associated gauge trans-
formations, the local supersymmetry transformations in ten dimensions.
The R-NS sector contains another gravitino ζµα̇ and another dilatino λα , built on the
conjugate spinor representation |α̇ir , with opposite chirality.
There exists naturally a sector with anti-periodic boundary conditions on the left and
periodic boundary conditions on the right, the NS-R sector. The field content one gets at the
massless level is exactly the same: a pair of gravitini (ζ^µα , ζ^µα̇ ) and a pair of dilatini (^λα̇ , ^λα )
transforming in conjugate spinorial representations of the space-time Lorentz group.

R-R sector
The massless states in the Ramond-Ramond sector are built out of the left and right Ramond
ground states. Given that there are two choices of Lorentz irreducible representations on each
side (spinor or conjugate spinor), one has overall four possibilities.

180
Superstrings

In terms of group theory, one has to decompose the tensor products of spinorial repre-
sentations 16 ⊗ 16, 16 ⊗ 16 0 , 16 0 ⊗ 16 and 16 0 ⊗ 16 0 into irreducible representations of the
Lorentz group each of them corresponding to a massless field in space-time. Obviously it is
enough to consider the first two combinations, the last two giving the same decompositions.
Using the properties of the Gamma-matrix algebra, one can show that, in ten dimensions,
the smallest spinorial irreducible representation of the Lorentz group consists in Majorana-
Weyl fermions, since one can impose the Majorana condition both for ψ and Γ 11 ψ at the
same time, where Γ 11 = Γ 0 · · · Γ 9 is the chirality matrix in ten dimensions, which satisfies
(Γ 11 )2 = 1.
In order to construct bilinear combinations of spinors that transform under Lorentz trans-
formations as tensors of irreducible representations, one defines first the totally antisymmetric
product of Gamma matrices:
1 X
Γ µ1 µ2 ···µp = sgn(σ)Γ µσ(1) · · · Γ µσ(p) . (7.110)
p! σ∈S
p

Whenever p is odd, Γ µ1 µ2 ···µp maps a spinor to a conjugate spinor, while when p is even
it maps a spinor to a spinor. Hence we have the following index structure:
!
µ1 ···µp β̇
0 (Γ )α
(Γ µ1 ···µp )a b = µ1 ···µp β , p odd , (7.111a)
(Γ )α̇ 0
!
µ1 ···µp β
(Γ ) 0
(Γ µ1 ···µp )a b = α
, p even . (7.111b)
0 (Γ µ1 ···µp )α̇ β̇
(7.111c)
The charge conjugation matrix, defined by CΓ µ C−1 = −(Γ µ )T , is used to raise spinorial indices
(and is such that the Majorana conjugate of a spinor ψ is ψ̄ = ψT C). It has the following
block-diagonal form in ten dimensions:
 
ab 0 Cαβ̇
C = (7.112)
Cα̇β 0
while C−1 can be similarly used to lower indices. Using the properties (7.111) and (7.112)
gives
 µ1 ···µp 
µ1 ···µp −1

(Γ )αβ 0
Γ C ab = , p odd , (7.113a)
0 (Γ µ1 ···µp )α̇β̇
(Γ µ1 ···µp )αβ̇
 
µ1 ···µp −1

0

Γ C ab = , p even . (7.113b)
Γ̄ µ1 ···µp α̇β 0

One deduces from eqn. (7.113) how to decompose spinor bilinears in ten dimensions into
totally antisymmetric tensors. One has first
X
Fµ1 ···µp Γ µ1 ···µp C−1 αβ


ζα χβ = (7.114)
p odd

181
Superstrings

and then X
Fµ1 ···µp Γ µ1 ···µp C−1


ζα χβ̇ = αβ̇
. (7.115)
p even

Hence the product of representations 16 ⊗ 16 gives p-forms with p odd and 16 ⊗ 16 0 gives
p-forms with p even.
To find the number of independent terms in the sums (7.114) and (7.115), one notices the
following gamma-matrix identity:
1 p(p+1)
Γ µ1 ···µp Γ 11 = (−1) 2 +1 µ1 ···µd Γµp+1 ···µd , (7.116)
(d − p)!

which follows from the definition of Γ 11 . Using this property, the sum over p odd in (7.114)
can be restricted to only three independent terms, p ∈ {1, 3, 5}, and the last term gives a
five-form satisfying a self-duality property:
1
Fµ1 ···µ5 = (?F)µ1 ···µ5 = µ1 ···µ10 Fµ5 ···µ10 . (7.117)
5!
In the same way, the sum over p even in (7.115) can be restricted to p ∈ {0, 2, 4}. In summary
one can write
16 ⊗ 16 [1] + [3] + [5]+
16 0 ⊗ 16 0 [1] + [3] + [5]+
(7.118)
16 ⊗ 16 0 [0] + [2] + [4]
16 0 ⊗ 16 [0] + [2] + [4]
where [p] means an antisymmetric p-form representation, and [5]+ means that the corre-
sponding tensor obeys the self-duality property (7.117).
Finally the physical states conditions on the Ramond ground states is obtained as follows.
Notice first that acting with G0 , as in (7.101), amounts to multiply the spinorial polarization
by pµ Γ µ . Hence one has the following condition, for instance for 16 × 16:

pρ (Γ ρ )α̇β uβ vγ = uα vβ pρ (Γ ρ )γ̇β = 0 . (7.119)

Using the identities

Γ ρ Γ µ1 ···µp = Γ ρµ1 ···µp − 1

(p−1)!
ηρ[µ1 Γ µ2 ···µp ] (7.120a)
Γ µ1 ···µp Γ ρ = Γ µ1 ···µp ρ − 1
(p−1)!
ηρ[µp Γ µ2 ···µp−1 ] (7.120b)
(7.120c)

in the decomposition (7.114), and CΓ µ C−1 = −(Γ µ )T , leads to

p[ρ Fµ1 ···µp ] = 0 , pρ Fρµ2 ···µp = 0 . (7.121)

One obtains therefore the Bianchi identities

dF = 0 , d ? F = 0. (7.122)

182
Superstrings

In other words, each differential p-form F is the field strength of a (p − 1)-form C, i.e.

F = dC , (7.123)

with an associated Abelian gauge symmetry C ∼ C + dΛ where Λ is a (p − 1)-form, called a

R-R form.
The fact that the physical states of the string correspond to the field strengths F of the
R-R forms rather than to the potential themselves indicate that the fundamental strings are
not charged under the corresponding gauge symmetries.

7.4 Type II superstring theories

The full set of physical states constructed from the Neveu-Schwarz and Ramond vacua for the
left- and right-moving sectors does not give rise to a satisfactory physical theory as contains
a tachyon in the NS-NS sector.
The solution to this problem is to restrict the space of states to a consistent subset that
does not contain the tachyon. This procedure, known as the Gliozzi-Olive-Scherk projection
(GSO projection) succeeds to do this, while guaranteeing supersymmetry in space-time, as
suggested by the existence of space-time gravitini in the massless sector.

7.4.1 Generalized fermion number

In the ψµ conformal field theory, we have defined the operator (−)F , the generalization of the
chirality matrix of the Clifford algebra, that anticommutes with all fermionic modes and as
a consequence with the modes of the supercurrent:
1−a
{(−)F , ψµn } = 0 , {(−)F , Gn } = 0 , ∀n ∈ Z + , (7.124)
2
This operator separates states in representations of the superconformal algebra in two halves.
Its eigenvalue is equal to one for states with even fermion number (i.e. with an even number
of fermionic creation operators) and to minus one for states with odd fermion number (i.e.
with an odd number of fermionic creation operators).
The full projection on the space of physical states should be compatible with the BRST
symmetry, therefore the generalization of the operator (−)F to the full theory, noted (−1)F ,
should commute with the BRST currents (7.81). We therefore define the action of this
operator in the ghost and superghost system as:
• The operator (−1)F commutes with all modes of the (b, c) fields; note that the op-
erator we are defining here is different from the ghost number operator discussed in
section 6.3.2.

• The operator (−1)F anticommutes with all modes of the (β, γ) fields, which have there-
fore odd charge w.r.t. F. A proper way to define the action of (−1)F of the (β, γ) CFT
is to identify F with the charge under the superghost current (7.36) mod two.

183
Superstrings

As Polchinski remarked in his textbook, for this reason F should be called a spinor number
rather than a fermion number, as it cares more about spin than about statistics.

Neveu-Schwarz sector
The charge of the Neveu-Schwarz vacuum can be deduced from the charge associated with
the superghost current. Inserting the mode expansion (7.69) into the superghost number
current (7.36), and using the same technique as in eqn. (4.40) and below it, one finds
I X X
dz
Qsg = jsg (z) = − β−r γr − γ−r βr . (7.125)
2iπ 3 1
r∈N+ 2 r∈N− 2

One has therefore

Qsg |−ins = −γ1/2 β−1/2 |−ins = |−ins (7.126)
hence
(−1)F |−ins = −|−ins . (7.127)
We define in the same way an operator (−)F acting on the states of the right-moving fermions
e

and superghosts, which satisfies

(−1)F |−ins = −|−ins . (7.128)

ef f

One can then split the superconformal representation built by acting with the oscillators
on the |−ins vacuum in two halves, with odd and even fermion number respectively, denoted
NS− and NS+ . Since all states in the sector NS+ (resp. NS− ) have the same (−)F eigenvalue,
they should be obtained from the NS vacuum with an odd (resp. even) number of fermionic
creation operators ψµ−n−1/2 . These two sectors correspond respectively to the vector and
trivial representation of the SO(1, 9) current algebra as discussed in subsection 4.2.3.
As a consequence of this construction, the conformal dimensions of states in these two
sectors will have a different modding
α0 2
h∈ p +N , NS+ sector, (7.129a)
4
α0 1
h ∈ p2 + N − , NS− sector. (7.129b)
4 2

Ramond sector
For the Ramond ground states, since the operator (−1)F generalizes the chirality matrix
– which anti-commutes with all Gamma-matrices hence all zero modes ψµ0 – to all oscillators
modes ψµn , all is needed is that the ground states of the combined ψµ and (β, γ) theories of
opposite space-time chiralities have opposite eigenvalues:

(−)F |αir ⊗ |−isg

r = |αir ⊗ |−ir ,
sg
(7.130a)
(−)F |α̇ir ⊗ |−isg
r = −|α̇ir ⊗ |−ir .
sg
(7.130b)

184
Superstrings

Hence the ground states in the 16 of SO(1, 9) have even fermion number and the ground states
in the 16 0 of SO(1, 9) have odd fermion number. Obviously taking the opposite convention
for both the left- and right-movers at the same time would not change the physical content
of the theory.
In the same way as before, beyond the ground state, the whole superconformal represen-
tation built out of the Ramond vacuum splits into two sectors of opposite (−1)F eigenvalue,
denoted by R+ and R− respectively.
Contrary to what happens with Neveu-Schwarz boundary conditions, the conformal di-
mensions in the sectors R+ and R− have the same modding:

α0 2
h∈ p +N , R+ sector, (7.131a)
4
α0 2
h∈ p +N , R− sector. (7.131b)
4
Therefore one could choose one for the left-movers and the other one for the right-movers
without spoiling the level-matching condition.

185
Superstrings

7.4.2 Type IIA and IIB superstring theories

Before defining consistent space-time theories from the fermionic string, let us first summarize
what we have found so far in this chapter. In the various sectors of the theory (NS-NS, R-NS,
NS-R and R-R) we have obtained the following BRST-invariant and level-matched states at
the lowest levels that we list together with their properties (omitting the ghost/superghost
part for clarity):

mass2 state Space-time (−)F (−)F

e

−2/α 0 |pµ ins-ns tachyon −1 −1

0 e ν |pµ ins-ns
ψµ−1/2 ψ gµν , bµν , Φ 1 1
−1/2

0 |α; pµ ir ⊗ ψ
e µ |p
−1/2
µi
g
ns ζµα , λα̇ 1 1

0 e µ |p
|α̇; pµ ir ⊗ ψ −1/2
µi
g
ns ζµα̇ , λα −1 1

0 ψµ−1/2 |pµ ins ⊗ |α;

^ pµ ir ζbµα , ^λα̇ 1 1
(7.132)
0 ψµ−1/2 |pµ ins ⊗ |^
α̇; pµ ir ζbµα̇ , ^λα 1 −1

0 |α; pµ ir ⊗ |β;
^ pµ ir [1] + [3] + [5]+ 1 1

0 |α̇; pµ ir ⊗ |^
β̇; pµ ir [1] + [3] + [5]+ −1 −1

0 |α; pµ ir ⊗ |^
β̇; pµ ir [0] + [2] + [4] 1 −1

0 |α̇; pµ ir ⊗ |β;
^ pµ ir [0] + [2] + [4] −1 1

In order to get rid of the tachyon vacuum (7.86), one can try to project onto states with
(−1)F = 1 in the left Neveu-Schwarz sector and (−1)F = 1 in the right Neveu-Schwarz sector,
e

i.e. to consider the sector (NS+ , NS+ ). Performing this projection in the NS-NS sector alone
is not a consistent choice however.
A consistent choice of projection in the various sectors of the theory should satisfy at
least the following constraints:

• The level matching condition should give a non-empty result; this rules out the sectors
(NS− , NS+ ) and (NS+ , NS− )

186
Superstrings

• States that are projected out should not appear in the operator product expansion of
the remaining ones

• All local operators that are kept in the spectrum should have OPEs between themselves
without branch cuts, with only integer powers of (z 0 − z) and no half-integer ones

• The one-loop vacuum amplitude should be modular invariant, i.e. the integrand should
be invariant under the modular group of the torus PSL(2, Z).

All these constraints, as well as higher-loop modular invariance, have a common set of
solutions, using a projection known as GSO projection that keeps only half of the states that
were described so far.
We will first present these solutions, before showing that modular invariance is satisfied
for them. In both of the consistent theories, the projection into the NS sectors should be
accompanied with a projection into the R sectors.

Type IIB superstring theory

The first consistent superstring theory is the IIB superstring theory, which consists in con-
sidering states in the following sectors:

sector massless states

(NS+ , NS+ ) gµν , bµν , Φ
Type IIB: (R+ , NS+ ) ζµα , λα̇ (7.133)
(NS+ , R+ ) ζ^µα , ^λα̇
(R+ , R+ ) [1] + [3] + [5]+

The most interesting part of this massless spectrum is the existence of a pair of gravitini ζµα
and ζ^µα , which have the same spinor chirality. These states indicate that the space-time theory
is actually invariant under local space-time supersymmetry, more explicitely under N = (2, 0)
supersymmetry since there exists two local supersymmetric transformations parametrized
with spinors of identical chirality.
The low-energy dynamics of the massless fields of bosonic string theory was captured
by a space-time action (5.6) for the graviton, B-field and dilaton. This result was coming
from asking conformal invariance of the two-dimensional worldsheet theory, and could also
be tested by taking the low-energy limit of the S-matrix elements between massless fields.
In the same way, the low-energy dynamics of type IIB string theory is captured by a ten-
dimensional theory with local supersymmetry and diffeomorphism invariance, in other words
a supergravity action for the massless fields, invariant under two local supersymmmetries of
identical chirality, as it should. This supergravity is known as type IIB supergravity in ten
dimensions.

187
Superstrings

Type IIA superstring theory

The second consistent superstring theory is the IIA superstring theory, which consists in
considering states in the following sectors:
sector massless states
(NS+ , NS+ ) gµν , bµν , Φ
Type IIA: (R+ , NS+ ) ζµα , λα̇ (7.134)
(NS+ , R− ) ζ^µα̇ , ^λα
(R+ , R− ) [0] + [2] + [4]

In this case, one has a pair of gravitini ζµα and ζ^µα̇ , which have the opposite spinor chirality.
These states indicate that the space-time theory is actually invariant under local space-time
supersymmetry, more explicitely under N = (1, 1) supersymmetry since there exists two local
supersymmetry transformations parametrized with spinors of opposite chiralities. The low
energy dynamics of the theory is captured by type IIA supergravity in ten dimensions.

Light-cone gauge and physical degrees of freedom

As for the bosonic string, one can solve for the transversality constraints on the massless
physical states. A similar conclusion is obtained by quantizing the theory in the light-cone
gauge, instead of using the covariant BRST formalism chosen in these notes. From both point
of views, the modes from the fields ψ0 ± ψ1 are removed and physical states are classified in
terms of representations of the little group SO(8).
The relevant representations are the vector 8v , the spinor 8s and the conjugate spinor
8c . Notice that these three representations have the same dimension, a property known as
triality. The relevant tensor product of representations for the type IIB string are then:
sector SO(8) representations
(NS+ , NS+ ) 8v ⊗ 8v = 1 ⊕ 28 ⊕ 35
Type IIB: (R+ , NS+ ) 8s ⊗ 8v = 8c + 56s (7.135)
(NS+ , R+ ) 8v ⊗ 8s = 8c + 56s
(R+ , R+ ) 8s ⊗ 8s = 1 ⊕ 28 ⊕ 35+
Here 35+ corresponds to the self-dual 4-form representation, 35 to the symmetric traceless
rank two tensor representation and 56s to the gamma-traceless vector-spinor of positive chi-
rality. One can do the same for type IIA and get:
sector SO(8) representations
(NS+ , NS+ ) 8v ⊗ 8v = 1 ⊕ 28 ⊕ 35
Type IIA: (R+ , NS+ ) 8s ⊗ 8v = 8c + 56s (7.136)
(NS+ , R− ) 8v ⊗ 8c = 8s + 56c
(R+ , R− ) 8s ⊗ 8c = 8v ⊕ 56t
where 56t refers to the antisymmetric 3-form and 56c to the gamma-traceless vector-spinor
of negative chirality.

188
Superstrings

7.5 One-loop vacuum amplitudes

A strong consistency constraint for the superstring theories that we constructed, type IIA
and type IIB, is provided by modular invariance of the one-loop vacuum amplitude.
The integrand of the modular integral over the torus modulus is the product of the
contribution from the xµ and (b, c) ghosts, that we computed earlier, giving (for d = 10)
iV10 1
Zx,g (τ, τ̄) = √
8 (7.137)
(4π2 α 0 )5 τ2 η(τ)η̄(τ̄)

and the contribution from the fermionic fields ψµ and from the (β, γ) superghosts that we
consider below.
The one-loop amplitude for both the fermions and the superghosts split into sectors
according to the boundary conditions for the fields. The Euclidean two-torus is parameterized
by a pair of coordinates σ1 and σ2 with periodicities (σ1 , σ2 ) ∼ (σ1 , σ2 ) + 2π(n + mτ1 , mτ2 ).
Along both of these one-cycles, one can specify whether the fields are periodic and anti-
periodic.
The periodicity along the space cycle parametrized by σ1 was discussed already. It cor-
responds to the choice of Ramond (periodic) or Neveu-Schwarz (anti-periodic) boundary
conditions:
ψµ (σ1 + 2π, σ2 ) = (−1)1−a ψµ (σ1 , σ2 ) , (7.138)
with a = 0 (resp. a = 1) in the NS (resp. R) sector.
The periodicty along the Euclidean time direction was also indirectly considered in subsec-
tion 6.3.2. We have seen that, for fermionic fields the path integral with periodic boundary
conditions was corresponding to the trace of (−1)F exp −βH, ^ while the partition function
^
itself, or trace over exp −βH, was given by the path integral with anti-periodic boundary
conditions. In other words, one considers the periodicity

ψµ (σ1 + 2πτ1 , σ2 + 2πτ2 ) = (−1)1−b ψµ (σ1 , σ2 ) , (7.139)

for a path integral corresponding in the Hamiltonian formalism to an insertion of (−1)bF in

the trace.

7.5.1 Zero-modes
At a more abstract level, when one defines two-dimensional fermions on a genus zero surface,
one needs to specify the spin structure, which is a four-fold choice of periodicities along both
one-cycles: (A, A), (A, P), (P, A) and (P, P) where A stands for anti-periodic and P stands
for periodic. All the first three case are called even spin structures while the last one is called
an odd spin structure (even or odd refers to the number of zero modes of the Dirac operator
mod two).
In terms of super-space coordinates (z, θ), a two-torus is defined by the following identi-
fications
(z, θ) ∼ (z + 2π, θ) ∼ (z + 2πτ,  0 θ) , ,  0 ∈ {−1, 1} . (7.140)

189
Superstrings

where  and  0 define the spin structure.

An important aspect of the computation of the one-loop vacuum amplitude for the bosonic
string was the presence of zero-modes for the ghosts fields, associated with the moduli and
the conformal Killing vectors on the two-torus. In the present case, there cannot be any
fermionic zero mode for the even spin structures, because of the anti-periodicity along at
least a one-cycle, while for the odd spin structure there exists a single zero-mode, which
is a constant spinor. Accordingly, there exists, for odd spin structure, a single fermionic
supermodulus ρ on the two-torus, which can be used to define a more general periodicity

(z, θ) ∼ (z + 2π, θ) ∼ (z + 2πτ + θρ, θ + ρ) (7.141)

as well as a superconformal Killing spinor, which corresponds to a spinorial shift ξ:

(z, θ) 7→ (z + θξ, θ + ξ) . (7.142)

This moduli and superconformal Killing spinor cannot exist for even spin structures, as θρ
and θξ would not have then the right periodicity properties for a bosonic coordinate.

7.5.2 Partition functions in the NS sector

Let us consider first the partition function for a single chiral fermion ψ(z). We define the
following partition functions for each spin structure:

Z b0 = Trns (−1)bF qL0 −1/48 ,

  
(7.143a)
1
Z b = Trr (−1)bF qL0 −1/48 .
 
(7.143b)

We start by considering the partition function with (A, A) boundary conditions, i.e
Trns qL0 −1/48 , which is easy to compute. A general state is of the form

N2
N1
· · · ψ−3/2 ψ−1/2 |0ins (7.149)

where Ni ∈ {0, 1}. It gives the following result

 0 Y
∞
−1/48 1/2 3/2 −1/48
1 + qn+1/2 .




Z 0 (τ) = q 1+q 1+q ··· = q (7.150)
n=0

In terms of the theta-functions (7.144) and the Dedekind eta-function (6.55) it can be written
as s 
0 ϑ 00 (τ, 0)
Z 0 (τ) = . (7.151)
η(τ)
Next we consider the partition function with (A, P) boundary conditions, which means
a trace in the NS sector with (−)F inserted inside the trace. We still consider states of the

190
Superstrings

The Jacobi theta-function with characteristics is an analytic function of two complex vari-
ables τ ∈ H and υ ∈ C, depending on two parameters a, b ∈ R and defined as

2 Y
∞  
−iπa(υ− b ) a8 −iπb 2iπυ n− 1+a +iπb −2iπυ n− 1−a
a
ϑ b (τ, υ) = e 2 q 1+e e q 2 1+e e q 2 (1 − qn )
n=1
(7.144)
From its definition it is obvious that
1
ϑ 1
(τ, 0) = 0 . (7.145)

The theta function obeys the modular transformations

iπ
ϑ ab (τ + 1, υ) = e− 4 a(a−2) ϑ a+b−1
   a 
(τ, υ) , (7.146a)
a √ iπ iπ 2  b 
ϑ b (−1/τ, υ/τ) = −iτe 2 ab+ τ υ ϑ −a (τ, υ) . (7.146b)

It has also the periodicity properties

ϑ a+2m (τ, υ) = eiπna ϑ a+2m

   
b+2n b+2n
(τ, υ) , m, n ∈ Z . (7.147)

Finally it satifies the completely non-trivial Jacobi abstruse identity

ϑ 00 4 (τ, υ) − ϑ 01 4 (τ, υ) − ϑ 10 4 (τ, υ) = −ϑ 11 4 (τ, υ) .

   
(7.148)

We often use the notation ϑ ab (τ) := ϑ ab (τ, 0).

   

form (7.149), however one gets a minus sign for each oscillator mode present. Therefore one
gets
0 Y
∞
−1/48 1/2 3/2 −1/48
1 − qn+1/2 ,




Z 1 (τ) = q 1−q 1−q ··· = q (7.152)
n=0

which can be written as s 

 0 ϑ 01 (τ)
Z 1 (τ) = . (7.153)
η(τ)
A partition function in the NS+ sector is then defined as

1 + (−1)F L0 −c/24
 
Trns q . (7.154)
2

Then we move to the (β, γ) ghosts, which are, in the NS sector, bosons with anti-periodic
boundary conditions. We consider the trace with a (−)bF insertion, where b ∈ {0, 1}. General
states are of the form

· · · (β−3/2 )M2 (γ−3/2 )N2 (β−1/2 )M1 (γ−1/2 )N1 |−ins . (7.155)

191
Superstrings

The NS vacuum contributes to the partition function as

q1/2−c/24 (−1)b = q1/24 (−1)b , (7.156)

as the vacuum has superghost charge one and dimension one-half.

Since the (β, γ) fields are bosons, the path integral with anti-periodic boundary conditions
(b = 1) along Euclidean time corresponds to the trace with a (−)F insertion, while periodic
boundary conditions (b = 0) corresponds to the trace without a (−)F insertion. Here F
is identified with the superghost number mod two, hence counts the number of superghost
oscillator modes mod two.
We consider then, for instance, the contribution from γ−1/2 oscillators to the partition
function with (−1)bF inserted. One gets
1
1 + (−1)b+1 q1/2 + q + (−1)b+1 q3/2 − · · · = . (7.157)
1 + (−1)b q1/2
Doing the same for all oscillators of β and γ gives:
 2  2
0 b 1/24 1 1
Zsg b
= (−1) q ···
1 + (−1)b q1/2 1 + (−1)b q3/2
Y
∞ 
1
2
b 1/24
= (−1) q . (7.158)
n=1
1 + (−1)b qn−1/2

In terms of the theta-functions, this is

0 η(τ)
Zsg b
= (−1)b 0 . (7.159)
ϑ b (τ)

We can give the combined contribution to the partition function of the D = 10 fermions
and (β, γ) ghosts in the NS+ sector. Putting together the contributions (7.151), (7.153)
and (7.159) one obtains
 
1  0 !5
ϑ 0 (τ) η(τ)
0 !5
ϑ 1 (τ) η(τ) 
ψ,sg
Zns+ (τ) = −
2  η(τ) ϑ 00 (τ) ϑ 01 (τ) 
   
η(τ)
0 !4  0  !4
1 ϑ 0 (τ) 1 ϑ 1 (τ)
= − . (7.160)
2 η(τ) 2 η(τ)

We remark that this is exactly the same as the contribution of eight fermions, with the
projection 12 (1 − (−1)F ). This can be understood as follows. The superghosts (β, γ) are
actually removing the contribution of two towers of ψµ oscillators, much as the (b, c) ghosts
remove two towers of xµ oscillators. If we have used the light-cone gauge, we would have
found the same result as only transverse oscillators remain, the light-cone oscillators along
ψ0 ± ψ1 being removed by the gauge choice.

192
Superstrings

What remains after the projection is the tower of oscillators built out of the affine primary
states { ψi−1/2 |0ins , i = 2, . . . , 10 }, containing only states with N ∈ N + 1/2. This is precisely
the character in the vector representation of the affine SO(8) algebra, as discussed in subsec-
tion 4.2.3. Indeed in this context one acts on the affine primary in the vector representations
with modes of the currents Jij , which are integer-modded.

7.5.3 Partition functions in the R sector

Let us consider the partition function in the Ramond sector, i.e. with periodic boundary
conditions along the space circle. We start by considering directly d = 10 fermions with PA
spin structure, which is
Z 10 (τ) = Trr qL0 −5/24 .

(7.161)
Since we haven’t performed yet the GSO projection, the ground state |air is in the reducible
Majorana representation of the ten-dimensional Clifford algebra, which has dimension 25 =
32; it has conformal dimension h = 5/8 (see eq. (4.140)). The contribution from the ground
states to the partition function, taking into account their degeneracy is therefore 32q5/8−5/24 =
(2q1/12 )5 .
One way to analyze this result is to consider that all the Ramond ground states are
obtained from the highest weight state in the spinorial representation by acting with the
lowering operators (−ψ00 + ψ10 ), (ψ20 − iψ30 ), etc...
Adding the contributions from all the oscillators ψµ−n with n > 1 is easy, and we get the
result
!5
 1  10 Y
∞
Zψ 0 (τ) = (2q1/12 )5 (1 + q)(1 + q2 ) · · · = 2q1/12 (1 + qn )2 . (7.162)
n=1

In terms of theta-functions, this is

  !5
ϑ 10 (τ)
Zψ 10 (τ) =

. (7.163)
η(τ)

For the set of ten fermions with PP spin structure, the result of the computation is
deceptively simple. One is interested in computing:

Zψ 11 (τ) = Trr (−1)F qL0 −5/24 .


(7.164)

Remember that the operator (−1)F was an extension of the chirality matrix Γ 11 to the full
Hilbert space of the fermionic CFT. If we focus on the 32-dimensional ground state, we know
that for each ground state of positive chirality there exists a ground state with opposite
chirality. Hence the two contributions cancel each other and we are left with

Zψ 11 (τ) = 0 .

(7.165)

193
Superstrings

Another way to see this is that the explicit computation would give
1 !5
ϑ (τ)
Zψ 11 (τ) = 1

= 0, (7.166)
η(τ)

following eqn. (7.145).

Finally, we have to compute the contribution of the superghosts to the Ramond sector
partition function. We focus on the PA spin structure, since the PP spin structure will not
contribute to the full partition function at the end. The superghosts being bosons, anti-
periodicity in Euclidean time means that we compute the trace with (−1)F inserted.
The superghost Ramond vacuum |−ir is defined in eq. (7.78a) and has conformal dimen-
sion h = −3/8, hence contributes to the trace as q−11/24+3/8 = q−1/12 . The problem in the
present case is there is an infinite degeneracy of the ground states, as any state of the form

(γ0 )N0 |−ir (7.167)

will have the same energy as the ground state. Taking into account the (−1)F insertion, the
contribution from all these states of conformal dimension h = 3/8 needs to be regularized,
giving the result

q−1/12 (1 − 1 + 1 − 1 + · · · ) = lim+ q−1/12 1 − q + q2 − q3 + · · ·

→0
1
= lim+ q−1/12
→0 1 + q
1
= q−1/12 . (7.168)
2
For the reader that doubts of this result, a more convincing argument will be given later.
From there we act with the various oscillators as usual. Because of the (−1)F insertion,
the action of γ−1 for instance gives the factor
1
1 − q + q2 + · · · = (7.169)
1+q
Adding all the contributions from the action of the oscillators β−n and γ−n with n ∈ Z>0 ,
one reaches the result
1
Zsg 10 (τ) = 1/12 Q∞

n 2
(7.170)
2q n=1 (1 + q )
In terms of theta-functions, one obtains then
 1 η(τ)
Zsg 0
(τ) = 1 . (7.171)
ϑ 0 (τ)

To conclude, the full contribution from the fermions and the superghosts (β,γ) in the
R+ and R− sectors are the same, as the term with PP boundary conditions gives a vanishing

194
Superstrings

result from eq. (7.145). One gets then

1  1 1
Zψ,sg
r± (τ) = Zψ 0 (τ)Zsg 0 (τ)
2
 1  !4
1 ϑ 0 (τ)
= . (7.172)
2 η(τ)

As before this result can be understood from the point of view of representation theory
of the SO(8) affine algebra. Zr± corresponds respectively to the characters of the spinor
and conjugate spinor representations of SO(8), which are both eight-dimensional, which is
reflected in the degeneracy of the ground state in (7.172).
In order to discriminate the characters for the spinor and conjugate spinor representations,
it is possible, from the point of view of representation theory, to consider a character for a
non-trivial group element in the Cartan sugroup of SO(8) rather than the identity. It amounts
to give non-zero values to the υ argument in each theta-function. One can write then:
 4 
1 Y  Y4
ϑ 1 (τ, υ` ) ± ϑ 11 (τ, υ` ) .
 
Zr± (τ, υ` ) = (7.173)
2η(τ)4 `=1 0 `=1

7.5.4 Partition functions of type II string theories

The full partition function of type II string theories are obtained by adding the contributions
from the various left and right sectors that we have considered before. For the type IIB
superstring theory, we need to consider schematically

Ziib = Zns+ ,ns+ − Zr+ ,ns+ − Zns+ ,r+ + Zr+ ,r+ . (7.174)

Minus signs have be added to the second and third terms to respect spin-statistics: space-time
fermions running into the loop should contribute to the partition function with a negative
sign.
For the type IIA superstring theory, we have to reverse the GSO projection in the right
Ramond sector and get:

Ziia = Zns+ ,ns+ − Zr+ ,ns+ − Zns+ ,r− + Zr+ ,r− . (7.175)

The fermion and superghost contribution to the partition function completely factorizes
into its holomorphic and anti-holomorphic parts, given that the associated field theories are
chiral. One has thus the following structure:
  
x,g ψ,sg ψ,sg ψ̃,sg˜ ˜
ψ̃,sg
Ziib (τ, τ̄) = Z (τ, τ̄) Zns+ (τ) − Zr+ (τ) Zns+ (τ̄) − Zr+ (τ̄) . (7.176a)
  
˜ ˜
Ziia (τ, τ̄) = Zx,g (τ, τ̄) Zψ,sg ψ,sg
ns+ (τ) − Zr+ (τ) Zψ̃, sg ψ̃,sg
ns+ (τ̄) − Zr− (τ̄) . (7.176b)

195
Superstrings

Gathering then the contributions (7.137), (7.160) and (7.172) one finds the following
important result for the unintegrated partition function of type II superstring theories:
iV10 1  
 0 4 04 14  2
Zii (τ, τ̄) = ϑ 0 (τ) − ϑ 1 (τ) − ϑ 0 (τ) (7.177)
4(4π2 α 0 )5 τ24 |η(τ)|24

Because the R± sectors give the same contribution, this result applies both to type IIB and
type IIA superstring theories. The partition function proper is then given as for the bosonic
string by Z 2
dτ
Z ii = Zii (τ, τ̄) . (7.178)
F 4τ2
This beautiful formula encapsulates the full spectrum of the type IIA and type IIB su-
perstring theories, but the outcome of the computation is actually completely trivial. The
Jacobi abstruse identity (7.148) tells us that
Zii (τ, τ̄) = 0 . (7.179)
This result has actually a profound meaning. It indicates that both type IIA and type
IIB superstring theories have a completely supersymmetric spectrum in space-time, as the
contribution from the infinite number of space-time bosons compensates precisely the contri-
bution from the infinite number of space-time fermions and the one-loop vacuum energy of
the theory is exactly zero.
We have seen already that the massless degrees of freedom of type IIA and type IIB
superstring theories describe at low energies supergravity theories with local supersymmetry,
and this result indicates that this property holds for the whole tower of massive string states.
While the vanishing of the one-loop amplitude is the relevant physical result for the su-
perstring theories themselves, one may want to consider a derived quantity from the one-loop
partition function that allows to keep track unambiguously of the various states propagating
into the loop.
For this one can consider, as in (7.173), characters for non-trivial SO(8) group elements
in the Cartan, for instance the diagonal generator. We have then for the type IIB superstring
theory, following the pattern (7.174)
iV10 1  
 0 4 04 14 14 2
Ziib (τ, τ̄, υ, ῡ) = ϑ 0 (τ, υ) − ϑ 1 (τ, υ) − ϑ 0 (τ, υ) − ϑ 1 (τ, υ)

4(4π α ) τ2 |η(τ)|
2 0 5 4 24

(7.180)
The type IIA superstring theory, which follows the pattern (7.175), gives the result:
iV10 1  
0 4
04 14 14 
Ziia (τ, τ̄, υ, ῡ) = ϑ 0 (τ, υ) − ϑ 1 (τ, υ) − ϑ 0 (τ, υ) − ϑ 1 (τ, υ)
4(4π2 α 0 )5 τ24 |η(τ)|24
  
× ϑ̄ 00 4 (τ̄, ῡ) − ϑ̄ 01 4 (τ̄, ῡ) − ϑ̄ 10 4 (τ̄, ῡ) + ϑ̄ 11 4 (τ̄, ῡ)
  

(7.181)
Compared to the type IIB result, we have replaced the R+ contribution with R− in the anti-
holomorphic sector.

196
Superstrings

7.5.5 Modular invariance of the superstring vacuum amplitude

An important consistency check of the validity of the construction outlined in this chapter
is to show that the one-loop vacuum amplitude of the superstring theories satisfies modular
invariance. While this was done for the bosonic string, hence for the fields xµ and (b, c), we
have to understand what happens for fields ψµ and (β, γ). This is especially necessary as we
have considered a non-trivial separate GSO projection for the left-moving and right-moving
degrees of freedom.

Modular τ 7→ −1/τ transformation

We start by examining the behavior under the transformation τ 7→ −1/τ. Using equa-
tions (7.146a) and (6.58) one has the following map:

ϑ 00 4 (−1/τ, υ/τ)
 04
4iπυ2 ϑ (τ, υ)
0
4
=e τ 4
(7.182a)
η (−1/τ) η (τ)
ϑ 01 4 (−1/τ, υ/τ)
 14
4iπυ2 ϑ (τ, υ)
0
= e τ (7.182b)
η4 (−1/τ) η4 (τ)
ϑ 10 4 (−1/τ, υ/τ)
 04
4iπυ2 ϑ (τ, υ)
1
4
=e τ 4
(7.182c)
η (−1/τ) η (τ)
ϑ 11 4 (−1/τ, υ/τ)
 14
4iπυ2 ϑ (τ, υ)
1
4
=e τ 4
(7.182d)
η (−1/τ) η (τ)

Setting υ = 0, it implies that the combination of these factors that appears in the partition
function (7.177) is invariant under τ 7→ −1/τ as it should.
The relations (7.182) can be interpreted as follows. The transformation τ 7→ −1/τ in-
terchanges the one-cycles of the two-torus, hence the spin structures. While the AA (first
equation) and PP (last equation) spin structures are invariant, the AP and PA spin struc-
tures are exchanged. This is one way to see that a hypothetic string theory with only the
(NS+ , NS+ ) sector would inconsistent, as the Ramond sector would automatically appear
from modular transformations.
One sees also that the minus sign in front of the R-NS and NS-R contributions in (7.174),
that were justified by spin-statistics in space-time, are actually forced upon us by modular
invariance.

Modular τ 7→ τ + 1 transformation
The transformation τ 7→ τ + 1 will also mix the different spin structures, as it corresponds
to replacing the identification along Euclidean time z ∼ z + τ with z ∼ z + (τ + 1), thereby
adding an extra twist of 2π along the spatial circle. Hence an AA spin structure will become
an AP spin structure, and an AP spin structure will become an AA spin structure. The

197
Superstrings

explicit computation gives (we removed the 1/η4 factor giving a common phase):

ϑ 00 4 (τ + 1, υ) = ϑ 01 4 (τ, υ)
 
(7.183a)
ϑ 01 4 (τ + 1, υ) = ϑ 00 4 (τ, υ)
 
(7.183b)
ϑ 10 4 (τ + 1, υ) = −ϑ 10 4 (τ, υ)
 
(7.183c)
ϑ 11 4 (τ + 1, υ) = −ϑ 11 4 (τ, υ)
 
(7.183d)

Hence
  14 
− ϑ 0 ± ϑ 1 7→ − ϑ 0 − ϑ 1 − ϑ 0 ± ϑ 1
04 04 14 14 0 4
04 14
ϑ 0
−ϑ 1
(7.184)

and combining the holomorphic and anti-holomorphic contributions eliminates the extra mi-
nus sign.
This completes the proof of the invariance of type IIA and type IIB superstring one-loop
amplitudes under PSL(2, Z), the modular group of the two-torus.

198