SpringerBriefs in Physics

Series Editors
Balasubramanian Ananthanarayan
Centre for High Energy Physics, Indian Institute of Science, Bangalore,
Karnataka, India

Egor Babaev
Department of Physics, Royal Institute of Technology, Stockholm, Sweden

Malcolm Bremer
H. H. Wills Physics Laboratory, University of Bristol, Bristol, UK

Xavier Calmet
Department of Physics and Astronomy, University of Sussex, Brighton, UK

Francesca Di Lodovico
Department of Physics, Queen Mary University of London, London, UK

Pablo D. Esquinazi
Institute for Experimental Physics II, University of Leipzig, Leipzig,
Germany

Maarten Hoogerland
University of Auckland, Auckland, New Zealand

Eric Le Ru
School of Chemical and Physical Sciences, Victoria University of
Wellington, Kelburn, Wellington, New Zealand

Dario Narducci
University of Milano-Bicocca, Milan, Italy

James Overduin
Towson University, Towson, MD, USA

Vesselin Petkov
Montreal, QC, Canada

Stefan Theisen
Max-Planck-Institut für Gravitationsphysik, Golm, Germany

Charles H. T. Wang
Department of Physics, University of Aberdeen, Aberdeen, UK

James D. Wells
Department of Physics, University of Michigan, Ann Arbor, MI, USA

Andrew Whitaker
Department of Physics and Astronomy, Queen’s University Belfast, Belfast,
UK

SpringerBriefs in Physics are a series of slim high-quality publications

encompassing the entire spectrum of physics. Manuscripts for
SpringerBriefs in Physics will be evaluated by Springer and by members of
the Editorial Board. Proposals and other communication should be sent to
your Publishing Editors at Springer.
Featuring compact volumes of 50 to 125 pages (approximately 20,000–
45,000 words), Briefs are shorter than a conventional book but longer than
a journal article. Thus, Briefs serve as timely, concise tools for students,
researchers, and professionals.
Typical texts for publication might include:
A snapshot review of the current state of a hot or emerging field
A concise introduction to core concepts that students must understand in
order to make independent contributions
An extended research report giving more details and discussion than is
possible in a conventional journal article
A manual describing underlying principles and best practices for an
experimental technique
 An essay exploring new ideas within physics, related philosophical
issues, or broader topics such as science and society
Briefs allow authors to present their ideas and readers to absorb them
with minimal time investment. Briefs will be published as part of Springer’s
eBook collection, with millions of users worldwide. In addition, they will
be available, just like other books, for individual print and electronic
purchase. Briefs are characterized by fast, global electronic dissemination,
straightforward publishing agreements, easy-to-use manuscript preparation
and formatting guidelines, and expedited production schedules. We aim for
publication 8–12 weeks after acceptance.
Marc Gillioz

Conformal Field Theory for Particle

Physicists
From QFT Axioms to the Modern Conformal
Bootstrap
Marc Gillioz
Zürich, Switzerland

ISSN 2191-5423 e-ISSN 2191-5431

SpringerBriefs in Physics
ISBN 978-3-031-27085-7 e-ISBN 978-3-031-27086-4
https://doi.org/10.1007/978-3-031-27086-4

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2023

This work is subject to copyright. All rights are solely and exclusively
licensed by the Publisher, whether the whole or part of the material is
concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any
other physical way, and transmission or information storage and retrieval,
electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.

The use of general descriptive names, registered names, trademarks, service

marks, etc. in this publication does not imply, even in the absence of a
specific statement, that such names are exempt from the relevant protective
laws and regulations and therefore free for general use.

The publisher, the authors, and the editors are safe to assume that the advice
and information in this book are believed to be true and accurate at the date
of publication. Neither the publisher nor the authors or the editors give a
warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The
publisher remains neutral with regard to jurisdictional claims in published
maps and institutional affiliations.
This Springer imprint is published by the registered company Springer
Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham,
Switzerland
This book is dedicated to all the people who spend their time, their energy,
and even their lives in fundamental research and get so little in return. No
matter how small, their contribution is a blessing for current and future
generations.
Preface
The story behind “Conformal field theory for particle physicists” is before
anything else the story behind my own academic path. I started as a particle
physicist, with a Ph.D. and a first postdoc mostly focused on high-energy
phenomenology and beyond-the-Standard-Model physics. At some point, I
realized that my research was not really contributing to the overall
advancement of science, as I had imagined it would (or more realistically, I
was not able to bring it anywhere). So I decided to start working on
conformal field theory, as I was hearing from various sources how great a
subject it was. This is not really something you are expected to do in our
modern academic world, and I think I got extremely lucky to be able to
make such a move.
Even if it did not turn out to be very smart from a career perspective, I
never regretted it. I found conformal field theory a fascinating subject. One
in which significant advances are made every year. One in which there are
great outstanding problems, and sometimes not enough people around to
work on them. One in which the applications range from solid-state physics
to string theory and cosmology. And one in which I was finally able to
make sense of what I learned in quantum field theory beyond perturbation
theory.
Learning conformal field theory turned out to be quite difficult for me.
Not that the underlying mathematical methods are fundamentally more
difficult than in particle physics; for sure, CFT requires quite a bit of
mathematics, including complex analysis, group theory, or differential
geometry. But these are all standard tools that a particle physics has (or
will) encounter in their own work. No, for me the most difficult part was the
change of perspective. Using correlation functions as the central observable
instead of scattering amplitudes. Working in position spare instead of
momentum space, and in Euclidean coordinates instead of Minkowski
space-time ones.1 The big difficulty for me was often more to understand
why I would have to work on some CFT project, rather than how. And even
though there are many excellent introductory books and reviews on the
topic of conformal field theory, none of them was able to answer my
questions and help me connect CFT with what I knew from particle physics.
 The goal of this book is precisely to fill what I think is a gap in the
literature. Conformal field theory is introduced here as just another quantum
field theory like the ones we study in particle physics. There are two more
properties—scale symmetry and special conformed symmetry—that let us
make so much more progress without having to go into the murky details of
some specific theory. And from these features, we are able to work our way
to the modern conformal bootstrap and gain an unprecedented
understanding of the dynamics of a strongly coupled quantum field theory.
This approach means that the reader should be familiar with quantum field
theory—at least some of it, but the more the better—but no prior experience
in conformal field theory is required. The book is typically aimed at
graduate students or at more advanced researchers in theoretical physics.
The presentation is self-contained, so it might as well be interesting for a
more diverse audience than just particle physicists, including condensed
matter physicists, cosmologists, or string theorists. Given the unusual
approach followed by the book, it might even be of interest to people who
are already familiar with some aspects of conformal field theory. The
presentation includes topics that are not easily found in the literature (and
particularly dear to me), such as subtleties of conformal transformations in
Minkowski space-time, the construction of Wightman functions and time-
ordered correlators in both position and momentum space, unitarity bounds
derived from the spectral representation, and the appearance of UV and IR
divergences, even if each of these topics can only be covered superficially
in such a short volume.
This book would not have been written if it had not emerged from the
lecture notes that I prepared for a graduate class at the University of Bern in
the spring semester of 2022. I would like to thank the Institute for
Theoretical Physics—Mikko Laine and Thomas Becher in particular—for
giving me the opportunity to teach my favorite topic, but also all the
students and postdocs who took part in the class for their involvement and
their critical questions. All of them contributed to the current state of this
book.
Last but not least, I received feedback from various people, both on the
particle physics and on the conformal side, including Johan Henriksson,
Slava Rychkov, Flip Tanedo, and Luca Vecchi, and benefitted from
discussions with various members of the conformal bootstrap community.
Thanks a lot to all of them. In spite of this and of the editorial support of
Springer Nature, there are certain mistakes and imprecisions left over, and I
hope that you will not hesitate to contact me should you notice any. Like
any introductory work, this book is also doomed to be incomplete. It will be
even more so in the future, as conformal field theory is likely to see
fundamental advances in the next few years.
Marc Gillioz
Zürich, Switzerland
December 2022
Acronyms
AdS/CFT The correspondence between a quantum field theory in anti-de-
Sitter (AdS) curved space-time and a conformal field theory in flat space
with one less dimension
CFT Conformal field theory
IR Infrared: synonym of low-energy, or long-distance, in QFT
OPE Operator product expansion (see Chap. 5)
QCD Quantum chromodynamics: the strongly coupled theory describing
the interaction of quarks and gluons, as in the Standard Model of particle
physics
QFT Quantum field theory (generally understood to be non-conformal)
RG Renormalization group: the framework describing how a QFT
changes with scale
UV Ultraviolet: synonym of high-energy, or short-distance, in QFT
Contents
1 Introduction
References
2 Classical Conformal Transformations
2.​1 Infinitesimal Transformations
2.​2 The Conformal Algebra
2.​3 Finite Transformations
2.​4 Compactification​s
2.​5 Minkowski Space-Time
2.​6 Conformal Symmetry in Classical Field Theory
Reference
3 Conformal Quantum Field Theory
3.​1 Non-Perturbative Quantum Field Theory
3.​2 Wightman Functions
3.​3 Spectral Representation
3.​4 Scale Symmetry
3.​5 Special Conformal Symmetry
3.​6 UV/​IR Divergences and Anomalies
Reference
4 Conformal Correlation Functions
4.​1 From Minkowksi Space-Time to Euclidean Space
4.​2 From Euclidean Space to Embedding Space
4.​3 3-Point Functions
4.​4 4-Point Functions
References
5 State-Operator Correspondence and OPE
 5.​1 The OPE in Quantum Field Theory
5.​2 The State/​Operator Correspondence
5.​3 The Conformal OPE
6 The Conformal Bootstrap
6.​1 Conformal Blocks
6.​2 OPE Convergence
6.​3 The Crossing Equation and Simple Solutions
6.​4 The Numerical Bootstrap
6.​5 Example:​The Ising Model in 3 Dimensions
6.​6 Other Conformal Bootstrap Results
References
7 Outlook
References
Footnotes
1 In fact, this is something I must never have completely accepted, as I ended up specializing on the
momentum-space representation of conformal correlators in Minkowski space-time.

OceanofPDF.com
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023
M. Gillioz, Conformal Field Theory for Particle Physicists, SpringerBriefs in Physics
https://doi.org/10.1007/978-3-031-27086-4_1

1. Introduction
Marc Gillioz1
(1) Zürich, Switzerland

Conformal field theory is a ubiquitous subject in modern theoretical

physics. Every local quantum field theory approaches a CFT in the large-
and small-distance limits,1 and it even plays a key role in the study of
quantum gravity through the AdS/CFT correspondence. CFT is also one of
the rare frameworks in which quantum field theory can be studied outside
the realm of perturbation theory. There a several excellent modern reviews
on the subject [1–5], and large parts of these lecture notes are directly
inspired by them.
Most introductory courses on CFT treat the conformal group as a whole.
This approach usually requires working in Euclidean space (we will see
why),2 and the connection with “traditional” quantum field theory appears
very late in the course, if at all. This can be frustrating for a particle
physicist. We propose to start here instead with the following definition:

(1.1)

and go through each part one after the other. The first part is quantum field
theory in flat Minkowski space-time. We expect the reader to be familiar
with it, but our attention will not be restricted to theories that have a nice
classical limit and a perturbative definition. We will therefore go through
some of the basics of QFT in Chap. 3, namely the Wightman axioms, but
without unnecessary mathematical rigor.
 Scale symmetry will come next. Particle physicists typically also have a
good physical intuition of it from the renormalization group. The main
novelty in CFT is that scale symmetry forbids the existence of particles with
a definite mass; instead, in a scale-invariant theory, there are states of any
energy (unless we are in the very special case of a free theory).
Finally, special conformal symmetry will be discussed last. It nearly
always comes along with scale symmetry, but it is vastly more powerful.
This will lead us to the study of conformal correlation functions in Chap. 4,
to the discussion of the operator product expansion (OPE) and its very nice
features in Chap. 5, and finally to the conformal bootstrap in Chap. 6,
including an example of a strongly-coupled theory that has been solved by
symmetry principles only.
Before diving into the quantum theory, however, we begin the course
with a look at classical conformal transformations in Chap. 2. It is
convenient to work in d-dimensional space-time and think of as a
particular realization of the more general framework. Lorentz indices , ,
therefore run between 0 and , with being the time component
of the vector , and the energy component of the momentum . Even
though we ultimately care about unitary quantum field theory in Minkowski
space-time, we will need to establish a connection with the same theory
defined in flat Euclidean space. For this reason, we (unfortunately) work
with the mostly-plus metric convention , so that going
from Minkowski space-time to Euclidean space is simply achieved by a
rotation of the time coordinate in the complex plane.

References
1. S. Rychkov, EPFL Lectures on Conformal Field Theory in D 3 Dimensions. SpringerBriefs in
Physics. 1, 2016. https://​doi.​org/​10.​1007/​978-3-319-43626-5. arXiv:​1601.​05000 [hep-th]
2.
D. Simmons-Duffin, The conformal bootstrap, in Theoretical Advanced Study Institute in
Elementary Particle Physics: New Frontiers in Fields and Strings, pp. 1–74. 2017. https://​doi.​org/​
10.​1142/​9789813149441_​0001. arXiv:​1602.​07982 [hep-th]
3.
D. Poland, S. Rychkov, A. Vichi, The conformal bootstrap: theory, numerical techniques, and
applications. Rev. Mod. Phys. 91 (2019) 015002. https://​doi.​org/​10.​1103/​RevModPhys.​91.​
015002. arXiv:​1805.​04405 [hep-th]
4.
S.M. Chester, Weizmann lectures on the numerical conformal bootstrap. arXiv:​1907.​05147 [hep-
th]
5.
J.D. Qualls, Lectures on conformal field theory. arXiv:​1511.​04074 [hep-th]

Footnotes
1 Massless free theories are conformal. The “empty”, low-energy limit of massive theories can also
be viewed as a special kind of CFT.

2 An exception is Slava Rychkov’s unpublished lecture notes on Lorentzian methods in conformal

field theory, available at https://​courses.​ipht.​fr/​node/​226.

OceanofPDF.com
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023
M. Gillioz, Conformal Field Theory for Particle Physicists, SpringerBriefs in Physics
https://doi.org/10.1007/978-3-031-27086-4_2

2. Classical Conformal Transformations

Marc Gillioz1
(1) Zürich, Switzerland

One of the most fundamental principles of physics is the independence of

the reference frame: observers living at different points might have different
perspectives, but the underlying physical laws are the same. This is true in
space (invariance under translations and rotations), but also in space-time
(invariance under Lorentz boosts).

2.1 Infinitesimal Transformations

In mathematical language, this means that if we have a coordinate system
, the laws of physics do not change under a transformations
(2.1)
This principle applies to all maps that are invertible (isomorphisms) and
differentiable (smooth transformations), hence it is usually called
diffeomorphism invariance. Being differentiable, the transformation (2.1)
can be Taylor-expanded to write
(2.2)
in terms of an infinitesimal vector (meaning that we will always ignore
terms of order ).
In addition to the coordinate system, the description of a physical
system requires a way of measuring distances that is provided by a metric
. Distances are measured integrating an infinitesimal line element
whose square equals
 (2.3)
Since all observers should agree on the measure of distances, we must have
(2.4)
Here could be the Euclidean metric or the Minkowski metric ;
for simplicity we only consider the case in which is flat, i.e.
. In this case, we can write

(2.5)

If we require the different observers to also agree on the metric, then we

must have , which gives a constraint on what kind of coordinate
transformations are possible: they must satisfy
(2.6)
This condition admits as most general solution
(2.7)
where is a constant vector and an antisymmetric tensor,
i.e. . The transformation
(2.8)
is obviously a translation and
(2.9)
a rotation/Lorentz transformation around the origin : the matrix
satisfies . The composition of these
two operations generates the Poincaré group. This is the fundamental
symmetry of space-time underlying all relativistic quantum field theory. It
is a symmetry of nature to a very good approximation, at least up to energy
scales at which quantum gravity becomes important.
 However, one can also consider the situation in which the two observers
use different systems of units, i.e. they disagree on the overall definition of
scale but agree otherwise on the metric being flat. In this case we must have
, and therefore the constraint 2.5 becomes
(2.10)
for some positive real number , with the most general solution
(2.11)
The new infinitesimal transformation is
(2.12)
It is a scale transformation, also known as dilatation. Note that scale
symmetry is not a good symmetry of nature: there is a fundamental energy
scale on which all observers must agree (this can be for instance chosen to
be the mass of the electron). Nevertheless, there are systems in which this is
a very good approximate symmetry, making it worth studying.
If one pushes this logic further, in a scale-invariant world in which
observers have no physical means of agreeing on a fundamental scale, they
might even decide to change their definition of scale as they walk around,
or as time passes. This would correspond to a situation in which the metric
of one observer can differ from the original metric by a function of
space-time:
(2.13)
Note that we are not saying that is completely arbitrary: at every point
in space-time, it is related to the flat metric by a scale transformation. But
the scale factor is different at every point. The condition on becomes in
this case
(2.14)
where is the infinitesimal version of , with the conventional relation
. To find the most general solution to this
equation, note that contracting the indices with gives
(2.15)
where d is the space(-time) dimension, while acting with gives
(2.16)
so that we get
(2.17)
Acting once again with , we arrive at
(2.18)
while acting with and symmetrizing the indices yields
(2.19)
The condition must therefore be satisfied in all dimensions (
), and in the additional condition , which is solved by
(2.20)
The corresponding value of is
(2.21)
Therefore, in addition to the transformations found before, we also have
(2.22)
which is called special conformal transformation. If we examine the
Jacobian for this transformation, we find

(2.23)

We have written this as a position-dependent scale factor ,

multiplying an orthogonal matrix
(2.24)
This shows that special conformal transformations act locally as the
composition of a scale transformation and a rotation (or Lorentz
transformation). This also shows that conformal transformations preserve
angles, which is the origin of their name. Equation 2.14 is called the
conformal Killing equation and its solutions 2.21 the Killing vectors.
 Note that in our derivation the original metric was flat, but the new
metric is not. It is however conformally flat: it is always possible to
make a change of coordinate after which it is flat again. In general,
transformations
(2.25)
are called Weyl transformations. They change the geometry of space-time.
We found that Weyl transformations that are at most quadratic in x can be
compensated by a change of coordinates to go back to flat space. The
corresponding flat-space transformation is called conformal
transformation.1
In the situation is a bit different: the conditions is
sufficient to ensure that the Killing equation has a solution. This is most
easily seen in light-cone coordinates,

(2.26)

in terms of which
(2.27)
This is satisfied by taking for the sum of an arbitrary function of the
coordinate and of another arbitrary function of . In fact, if we write
, we can take arbitrary functions and , and
verify that Eq. 2.14 is satisfied with . In Euclidean
space, we define

(2.28)

complex-conjugate to each other, and the same logic follows: we can apply
arbitrary holomorphic and anti-holomorphic transformations on z and ,
and the conformal Killing equation is always satisfied. This shows that
there are (infinitely) many more conformal transformations in than in
, and also that there is no significant difference between Euclidean
and Minkowski conformal transformations in , as the transformation
acts essentially on the two light-cone/holomorphic coordinates
independently.
2.2 The Conformal Algebra
The conformal Killing Eq. 2.21 determines the most general form of
infinitesimal conformal transformations. Finite conformal transformations
follow from a sequence of infinitesimal transformations. However, one has
to bear in mind that infinitesimal conformal transformations do not
commute: for instance, a translation followed by a rotation is not the same
as the opposite. The conformal transformations form a group: the
composition of conformal transformations is again a conformal
transformation.
As we know from quantum field theory, a group is characterized by its
generators and their commutation relations (the algebra). A generator G
describes an infinitesimal transformation in some direction, and finite
transformations are obtained by exponentiation, , with parameter
(the factor of i is a physicist’s convention that makes the generators
Hermitian). A representation of the conformal group can be obtained from
smooth functions of the coordinates, f(x). For instance, under an
infinitesimal translation, we have
(2.29)
and we require this to be equal to , which means
(2.30)
Performing the same analysis for the other infinitesimal transformations
given in Eq. 2.21, we obtain for the other generators2
(2.31)

(2.32)

(2.33)
The number of generators matches that of the Killing vectors: there are d
translations, d special conformal transformations,
rotations/Lorentz transformations ( is a antisymmetric matrix),
and one scale transformation. Therefore the total number of generators,
i.e. the dimension of this group, is . In space-time
dimensions, the conformal group has 15 generators.
Using the above definition, one can verify that the following
commutation relations are satisfied,

(2.34)

while all other commutators vanish:

(2.35)
The first two relations in Eq. 2.34 are the familiar Poincaré algebra. The
next one states that transforms like a vector (as does), whereas D is
obviously a scalar. The next two relations remind us that and have
respectively the dimension of length and inverse length.
Even though it is not immediately obvious, this algebra is isomorphic to
that of the group if is the Euclidean metric, or
if it is the Minkowski metric. To see that it is the case, let us introduce a
-dimensional space with coordinates
(2.36)
and a metric defined by the line element
(2.37)
is a new spatial coordinate, and a new time. Then we can write
all conformal commutation relations as being defined by the Lorentzian
algebra
(2.38)
provided that we identify the antisymmetric generators with the
conformal generators as follows:
 (2.39)

2.3 Finite Transformations

We just saw that the infinitesimal conformal transformations generate a
group. But how can we describe finite conformal transformations? Let us
see how each generator exponentiates into an element of the group; the
most general conformal transformation can then be obtained as a
composition of such finite transformations.
In some cases the exponentiation is trivial. For instance, with
translations we obtain immediately
(2.40)
where a is now any d-dimensional vector, not necessarily small. The same
is true of scale transformations,
(2.41)
with finite . Rotations or Lorentz transformations exponentiate as
(2.42)
where is a or matrix, depending whether the
metric is Euclidean or Minkowski. All of this is standard in quantum field
theory.
On the contrary, special conformal transformations do not exponentiate
trivially. The easiest way to derive their finite form is to make the following
observation: recall that in infinitesimal form we have
(2.43)
which implies , and therefore (as always neglecting
terms of order )
 (2.44)

The ratio appearing on both side of the equation is the inverse of the
coordinate , respectively : the inversion is defined by

(2.45)

This transformation does not have an infinitesimal form, but otherwise it

shares the essential properties of a conformal transformation: its Jacobian is

(2.46)

which is the product of a position-dependent scale factor ( ) with an

orthogonal matrix. To understand what this transformation does globally, let
us consider a Euclidean point . Then the matrix in
square brackets is diagonal, and equates . This is an
orthogonal matrix with determinant , which is part of but not
. This shows that the inversion is a discrete transformation not
connected to the identity. A conformally-invariant theory might be invariant
under inversions, but it needs not be.
Equation 2.44 shows that infinitesimal special conformal
transformations are obtained taking an inversion followed by a translation,
followed by an inversion again. Since this process involves the inversion
twice, and since inversion is its own inverse, it does not matter whether
inversion is a true symmetry of the system or not. The advantage of this
representation is that it can easily be exponentiated: the composition of
(infinitely) many infinitesimal special conformal transformation can be
written as an inversion followed by a finite translation, followed by an
inversion again. In other words, Eq. 2.44 holds for finite . This can be
used to show that

(2.47)

Exercise 2.1 Use Eq. 2.44 to show 2.47.

What do special transformation do globally? Let us look specifically at
Euclidean space. There are some special points:
The origin of the coordinate system is mapped onto itself.
The point is mapped to .
Conversely, the “point” is mapped to the finite value .
These properties can be understood from the fact that special conformal
transformations and translations are related by inversion: special conformal
transformations keep the origin fixed but move every other point, including
; translations move every point except . The other two
transformations, rotations and scale transformations, keep both 0 and
fixed.
An essential property of conformal transformations is that they let us
map any 3 points onto another triplet . This can be
seen as follows: first, apply a translation to place at the origin, followed
by a special conformal transformation that takes to , after which the
image of the original triplet is ; then use rotations and scale
transformations to move to another point , while keeping 0 and
fixed; finally apply again a special conformal transformation that takes
to , and a translation by to reach the configuration .
This property has an immediate physical consequence: in correlation
functions involving 2 or 3 local operators (see next sections for a
definition), all kinematics is fixed by conformal symmetry. The only
freedom is about the operators themselves, not about their position in space.
Another interesting property of conformal transformations is that they
map spheres to spheres: this is an obvious property of translations,
rotations, and scale transformations, but it is also true of special conformal
transformations.
 Exercise 2.2 Show that under the special conformal transformation
2.47, a sphere centered at the point and with radius R gets mapped to
a sphere centered at the point

and with radius

In the special case in which lies on the surface of the original

sphere, show that the sphere gets mapped to a plane orthogonal to the
vector . Note that a plane is a sphere of infinite radius.

In , the additional conformal transformations (infinitely many of

them!) mean that (nearly) any shape can be mapped onto another. This is
known as the Riemann mapping theorem.

Fig. 2.1 Inverse stereographic projection of the Euclidean space (represented here as the
horizontal plane) onto the sphere , both embedded in . A point x with gets mapped
to the northern hemisphere; a point with to the southern hemisphere; the origin is
mapped to the south pole (S), and to the north pole (N)

2.4 Compactifications
We mentioned earlier that conformal symmetry is a symmetry of flat
space(-time). It is true, as we have just seen, provided that we treat the point
as being part of the space. This is quite straightforward in Euclidean
space, but much more subtle in Minkowski space-time, as there are
different, inequivalent ways of reaching there.
To gain a better understanding of this, it is useful to map the flat
Euclidean space or the Minkowski space-time onto a curved
manifold. In Euclidean space, this is for instance achieved by the (inverse)
stereographic projection that maps to the unit sphere .
Geometrically, the stereographic projection is constructed as follows (see
Fig. 2.1): embed as a plane in , together with a sphere of unit
radius centered at the origin. Every point on the plane has an image on the
sphere obtained by drawing a segment between the original point and the
north pole of the sphere, and noting where it intersects the sphere. The
origin is mapped to the south pole, to the north pole, and the sphere
of unit radius to the equator. Algebraically, this is achieved as follows:
first write the Euclidean metric in spherical coordinates,
(2.48)
where the solid angle is given in by , in by
, and more generically by the recursion relation
. Let us perform the change of variable

(2.49)

and interpret as the zenith angle on the sphere: is the

north pole, corresponding to , and the south pole,
corresponding to . In these coordinates we have

(2.50)

The new metric is flat up to an overall Weyl factor that depends on . In

these coordinates, conformal transformations are always non-singular. For
this reason, it is often convenient to study classical conformal
transformations on the sphere instead of Euclidean space.
 However, this compactification is not as nice in a quantum theory in
which one wants to foliate the space along some preferred direction: if one
chooses as the “Euclidean time”, then the “space” direction is a sphere
whose volume depends on . In other words, the generator of “time”
translations is not a symmetry of the system. This is also true of any other
choice of time direction on the sphere.
Instead, another compactification is often preferred to the sphere: from
the Euclidean metric in spherical coordinates, one can make the change of
variable
(2.51)
after which
(2.52)
This is again a flat metric, up to a Weyl factor . In this case,
however, the rescaled metric is independent of . This space has the
geometry of a cylinder: . It is not fully compact: goes from
to . But it has an important advantage: translations in are
generated by dilatations D, which will be taken to be a symmetry of the
quantum theory. Foliating the space into surfaces of constant will later
lead us to radial quantization in conformal field theory.
Note that in language, the generator is
completely equivalent to the other generators ,
since they are related to them by rotations. So we might as well
look for a cylinder compactification in which the non-compact direction
corresponds to transformation generated by . This
combination of generators obeys

(2.53)

with two fixed points at with . The foliation of space

generated by this linear combination of generators is what is used in N-S
quantization,3 discussed later in Chap. 5. Figure 2.2 illustrates the two
foliations of Euclidean space by D and , and the
corresponding cylinder interpretations are shown in Fig. 2.3.
Fig. 2.2 Foliations of Euclidean space (here in ) in radial (left) and N-S quantization (right).
Circles of identical colors are mapped onto each other by a conformal transformation. In particular,
the plane in N-S quantization is mapped to the unit sphere in radial quantization

Exercise 2.3 Find the change of coordinates that makes the Euclidean
metric Weyl-equivalent to a cylinder in which translations in the non-
compact direction are generated by .
Hint: Find a special conformal transformation followed by a
translation that takes to , and apply it to the radial
coordinates.

The cylinder compactifications of Euclidean space are interesting by

themselves, but they are also extremely convenient to understand the
connection between Euclidean and Minkowski space-times: in this last
form, performing a Wick rotation defines a cylinder on which the
Lorentzian conformal group acts naturally. But before we get
there, let us go back to flat Minkowski space-time and make some general
remarks.
Fig. 2.3 Left: the Euclidean cylinder corresponding the radial coordinates 2.52, in which the
evolution in the non-compact direction is given by the generator D of scale transformations; the blue
circle corresponds to the unit sphere. Middle: same Euclidean cylinder after a conformal
transformation, so that the evolution is given by ; the blue circle is now the plane
, which includes the origin and the point at infinity. Right: the Lorentzian cylinder, obtained
after a Wick rotation in , so that the surface is unchanged; the image of this surface under
time translations generated by is the Poincaré patch, shown as the blue diamond on the
cylinder’s surface; only special conformal transformations can move a point from inside the Poincaré
patch to the outside

2.5 Minkowski Space-Time

Translations and Lorentz transformations in Minkowski space-time are
familiar, and even dilatations are a standard tool in the renormalization
group analysis. But what do conformal transformations do?
To understand this, let us place an observer at the origin of Minkowski
space-time. The presence of this observer breaks translations, but not
Lorentz transformations (let us assume that the observer is point-like), nor
dilatations or special conformal transformations. For the observer, space-
time is split into three regions: a future light cone, a past light cone, and a
space-like region from which they know nothing. Lorentz and scale
transformation preserve this causal structure: the future and past light-cones
are mapped onto themselves. In other words, if a point x is space-like
separated from the observer, it will remain space-like separated no matter
the choice of Lorentz frame, or the definition of length. Without loss of
generality, let us choose this point to be at position , where is
a unit vector (units can be chosen so that this is the case). Now apply a
special conformal transformation with parameter , with
varying between 0 and 1. This draws a curve in space-time,
parameterized by , with

(2.54)

This curves begins at the space-like point , and ends in the past
light cone at . Note that y never crosses a light cone: the image
of a point x is never null under a special conformal transformation unless x
is itself null, since . Instead, we have

(2.55)

What happens is that the point travels all the way to space-like infinity at
, and comes back from past infinity. Clearly, special conformal
transformation break causality!
The resolution of this puzzle is that conformal transformations do not
act directly on Minkowski space, but rather on its universal cover that is
isomorphic to the Lorentzian cylinder described in Fig. 2.3. Evolution on
that cylinder is given by the Hamiltonian , but this
differs from the Minkowski time evolution generated by . On any given
slice of the Lorentzian cylinder, space is compactified in such a way that the
notion of infinite distance is unequivocal: space-like infinity corresponds to
a point on the sphere, antipodal to the origin. If one takes any other point of
that sphere and applies finite translations using the generator , then this
defines a compact Poincaré patch. The full Lorentzian cylinder is a
patchwork of Poincaré patches, but every local observer only has access to
one.4
 The lesson that we must learn is that only the infinitesimal form of
special conformal transformations can be used in Minkowski space-time:
any finite special conformal transformation brings part of space-time into
another patch on the cylinder. This is sometimes called weak conformal
invariance.

2.6 Conformal Symmetry in Classical Field

Theory
So far we have only been discussing conformal transformations of the
coordinates. The next step is to consider a field theory (for the moment a
classical one) that has conformal symmetry built in. The simplest example
is the free, massless scalar field, defined by the action

(2.56)

We shall see in the next section that the action principle can in fact be
dropped in CFT, but for now it is a convenient starting point.
In this context, a conformal transformation is a transformation of the
fields. There are two distinct and complementary perspectives one can
adopt. It is often convenient to think of the metric tensor as a field in its
own right, and to define a conformal transformation as a (position-
dependent) scale transformation of the field
(2.57)
combined with a Weyl transformation of the metric
(2.58)
is the scaling dimension of the field . In a free theory it coincides with
the dimension of the field in units of energy (inverse units of length),
namely

(2.59)

is an infinitesimal scale factor that satisfies . The

advantage of this perspective is that the conformal transformations are
simple, multiplicative transformations of the fields. The disadvantage is that
it requires thinking of the theory in curved space-time. This means that the
metric that is implicit in the action 2.56 must be made explicit, but also that
the action can be supplemented with a term depending on the scalar
curvature tensor R as

(2.60)

Since R vanishes in flat space, it looks like this additional term could appear
with an arbitrary coefficient without modifying the original flat-space
action, but this is not the case.

Exercise 2.4 Verify that there is a unique value of for which this
action is invariant under the infinitesimal conformal transformations
2.57 and 2.58. What is it?

For this reason, it is also convenient to consider the opposite perspective in

which conformal transformations are transformations of the dynamical
fields and of the coordinates, but not of the metric. In this case the
conformal transformation can be defined as
(2.61)
where the parameters and are related by the conformal Killing
Eq. 2.14, i.e. . In infinitesimal form, this transformation
becomes

(2.62)
 Exercise 2.5 Show that under the transformation 2.62, the Lagrangian
of the free scalar field

is shifted by a total derivative term

hence proving that this is a symmetry of the action. You will need to use
the fact that is at most quadratic in x.

Note that Poincaré symmetry is a special case of this transformation,

corresponding to constant (and thus ).
By Noether’s theorem, whenever an action is invariant under some
transformation of the field
(2.63)
i.e. whenever the Lagrangian varies by a total derivative term,
(2.64)
then there exists a conserved current

(2.65)

In our example, this conserved current is therefore

(2.66)

The 2-index tensor on the right-hand side is called the canonical energy-
momentum tensor. Its divergence satisfies
(2.67)
therefore vanishing by the equation of motion for the free field, .
This implies in turn that the Noether current 2.66 is conserved for constant
. If on the contrary depends on space(-time), then we have
 (2.68)
In our example the canonical energy-momentum tensor is symmetric in its
indices and , and therefore we can write

(2.69)

In dimensions , this current is only conserved when . This is

surprising, because we just showed that scale and special conformal
transformations are also symmetries of the action, so why is the Noether
current not conserved?
The reason is that the version of Noether’s theorem given above does
not straightforwardly apply to the case of a space-time dependent parameter
. In fact, the energy-momentum tensor that we computed in this example
is not unique: one can always add to it a piece proportional to
(2.70)
without affecting the conservation Eq. 2.67, but changing the value of its
trace. The combination

(2.71)

is for instance traceless in any d. It turns out that it is always possible in a

field theory with conformal symmetry to construct an energy-momentum
tensor that is:
symmetric ( ),
traceless ( ), and
conserved once the equation of motions are imposed ( ).
This is a non-trivial fact, but we will skip its proof (as already
mentioned, we are interested in theories that are not necessarily defined
through an action).
Strictly speaking, Noether’s theorem only applies to theories that have a
Lagrangian description, but we will assume the existence of a traceless
energy-momentum tensor in all cases (in some sense this is going to be one
of the “axioms” of conformal field theory). From this assumption, we can
deduce that the theory is invariant under conformal transformations: a
conserved current can always be built from the energy-momentum tensor
and a conformal Killing vector as
(2.72)
As always, conserved charges can be constructed as the integral of the time
component of a conserved current over space. The simplest example is

(2.73)

This is in principle a function of , but it is in fact constant over time,

since

(2.74)

This conserved charge is the momentum, associated with translation

symmetry. Similarly, there are conserved charges associated with Lorentz
transformations,

(2.75)

with scale transformations,

(2.76)

and with special conformal transformations

(2.77)

Exercise 2.6 Show that the charges 2.75, 2.76 and 2.77 are conserved
in time.

The conservation of all these charges relies on the vanishing divergence of

the energy-momentum tensor, , which itself relies on the equation of
motion being satisfied. This is certainly true in the absence of sources. But
when source terms are added to the action, the equation of motion is
modified. In our example of the free scalar field theory, adding to the action
2.56 a source term of the form

(2.78)

modifies the equation of motion to

(2.79)
Instead of the conservation Eq. 2.67 for the canonical energy-momentum
tensor, we must replace it with
(2.80)
In the presence of such a source, the charges 2.73–2.77 are not conserved
anymore. However, if the source is local, say , so that
(2.81)
then we can determine the change in one of the charge — say —
between a time anterior to the local source, and a time
posterior to it, and call this difference the momentum of the source, . By
our definition, this is equal to

(2.82)

Since the two surfaces of integration meet at spatial infinity, they can be
viewed as the two sides of a closed surface surrounding the point ,
hence

(2.83)

By the divergence theorem, this is equal to

(2.84)

Note that this result does not depend on the choice of surface , as long
as it encloses the point . In technical terms, is a topological charge.
We chose in Eq. 2.73 to use a standard definition of in which the
energy-momentum tensor is integrated along a surface of constant time. But
if we work in Euclidean space, this is conformally equivalent to integrating
over the surface of a sphere, or in fact over any other closed surface.
Equation 2.84 is very important: it says that the charge associated
with a local source for the field is equal to . This is strikingly
similar to the action of the generator 2.30 on functions of the coordinates. In
fact, it is easy to verify that the other charges 2.75, 2.76, and 2.77 also act
on the classical field exactly like the generators 2.31, 2.32, and 2.33
respectively. We worked here with the free scalar field theory as an
example, but our discussion can be generalized to arbitrary classical field
theories. The important lesson is that a traceless energy-momentum tensor
can be used to give a field-theoretical realization of the conformal
generators discussed before.

Reference
1. M. Luscher, G. Mack, Global conformal invariance in quantum field theory. Commun. Math.
Phys. 41, 203–234 (1975). https://​doi.​org/​10.​1007/​BF01608988

Footnotes
1 This implies that the group of conformal transformation is a subgroup of diffeomorphisms. It is in
fact the largest finite-dimensional subgroup.

2 The sign of these generators is an arbitrary convention. It defines once and for all the
commutations relations that we will derive next. After that, we will always refer to the commutation
relations as defining the generators for other representations of the conformal group.

3 The line drawn when evolving a point with this Hamiltonian looks like a magnetic field line
connecting the north (N) and south (S) poles of a magnet, hence the name N-S quantization.

4 As Lüscher and Mack put it: In picturesque language, [the superworld] consists of Minkowski
space, infinitely many “spheres of heaven” stacked above it and infinitely many “circles of hell”
below it [1].
OceanofPDF.com
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023
M. Gillioz, Conformal Field Theory for Particle Physicists, SpringerBriefs in Physics
https://doi.org/10.1007/978-3-031-27086-4_3

3. Conformal Quantum Field Theory

Marc Gillioz1
(1) Zürich, Switzerland

Let us now turn to quantum field theory and study the implications of
conformal symmetry in that context. The standard approach to quantum
field theory is to think of a classical field theory, in which we now have a
basic understanding of what conformal symmetry does, and then quantize it
by promoting the fields to operators acting on some Hilbert space. But this
is not the approach that we will take here. There will still be states and
operators, but the latter will not necessarily be associated with fields
appearing in a Lagrangian.

3.1 Non-Perturbative Quantum Field Theory

To define a quantum field theory non-perturbatively, we need the following
ingredients:
1. Hilbert space: The Minkowski space-time is foliated into surfaces of
equal-time, and to each time slice we associate a Hilbert space of
quantum states.

2. Local operators: There are a number of (in fact infinitely many) local
operators that act on this Hilbert space. For instance, let us take to
be an operator acting on the Hilbert space at time . We call this
operator local because we require that it commutes with any other local
operator inserted at a distinct point on the same time slice:1
(3.1)
3. Symmetries: One of the local operators of the theory is the energy-
momentum tensor, and from it we can define conserved charges,
including and as in Eqs. (2.​73) and (2.​75). In a generic QFT
the energy-momentum tensor needs not be traceless, so the charges D
and cannot be considered. and are conserved in time, so
they are valid operators on all Hilbert spaces at every time t. Their
value changes however every time an operator is inserted at some point
x. In analogy with Eq. (2.​84), we require that this change is encoded in
the commutator
(3.2)
Since this equation is solved by
(3.3)
we say that is the generator of translations, which act as unitary
transformations on the operators (note that is Hermitian).
Lorentz transformations are similarly realized as unitary
transformations, generated by the charge . We can choose to
decompose local operators inserted at the origin of space-time into
irreducible representations of the Lorentz group and denote these with
, with a standing for a collection of Lorentz indices, so that
(3.4)
where is a matrix that satisfies the Lorentz algebra. For a
scalar operator, vanishes; for a vector operator with one Lorentz
index, it is given by
(3.5)
and so on. When combined with Eq. (3.3), and requiring that and
satisfy the Poincaré algebra (2.​34), this implies
(3.6)
Note that and are not local operators, but their commutator
with any local operator is again local (the same will later be true of the
generators D and ).
4. Vacuum state: The Hilbert space includes a vacuum state , which
we assume to be invariant under Poincaré transformations (and later
conformal transformations),
(3.7)
Other states of the theory are obtained acting with products of local
operators on the vacuum (see below for a more precise statement).
There is in principle one Hilbert space for each time slice, but since
time translation is a symmetry generated by , the evolution operator
is unitary and all Hilbert spaces are equivalent. We also
require that the vacuum is the lowest-energy state in the Hilbert space.
This means that if we can construct an eigenstate of energy,
(3.8)
then its eigenvalue must satisfy .

These four points essentially give the most general non-perturbative

definition of quantum field theory. They are nearly equivalent to the so-
called Wightman axioms (but presented here without much mathematical
rigor). One notable difference is that the Wightman axioms do not rely on
the existence of an energy-momentum tensor, but assume directly that the
Poincaré transformations are realized as unitary transformations on the
Hilbert space.
The locality condition (3.1) is often formulated in the Lorentz-invariant
way
(3.9)
stating that local operators commute as long as they are space-like
separated, which is known as the micro-causality axiom. Similarly, when it
is possible to work with eigenstates of energy and momentum,
(3.10)
then the Lorentz-invariant condition on the positivity of energy becomes
(3.11)
i.e. the momentum in contained in the forward light cone. Such
eigenstates of can be constructed from the Fourier transform of local
operators,

(3.12)

and acting on the vacuum. This implies that

(3.13)
as well as its generalization to the product of multiple local operators,
(3.14)
This is sometimes called the spectral condition.

Exercise 3.1 Using the definition (3.3) and integration by parts, show
that

3.2 Wightman Functions

Starting from these axioms, the next thing we can do is compute the
vacuum expectation value of products of local operators,
(3.15)
This can be viewed as an overlap of the vacuum state, , with a state
created acting on the vacuum with a sequence of local operators. Note that
these operators need not be ordered in time: the time-evolution operator is
unitary, so it can go both ways. This object is therefore different from time-
ordered correlation functions obtained from the path integral.
Correlators of this type are called Wightman functions. They are the
fundamental observables in non-perturbative quantum field theory. In fact,
it is even possible to completely define a quantum field theory just by its
Wightman functions: the Wightman reconstruction theorem states that the
Hilbert space of a quantum field theory can be constructed from all its
Wightman functions. A convenient perspective is therefore to forget about
the Hilbert space and focus on correlation functions.
 The symmetry properties of these correlation functions are encoded in
“Ward identities”: given a conserved charge G that annihilates the vacuum,
(this could be or ), the following equation must be
satisfied

(3.16)

This is an equation that is obvious in the Hilbert space picture, but it is also
valid as a differential equation for the Wightman function, since each
commutator is again related to the local operator. Let us see some examples.
The simplest Wightman function involves a single scalar operator,
(3.17)
In this case the Ward identity associated with translations implies

(3.18)

or in other words that the vacuum expectation value of the operator is a

constant over all of space-time. Lorentz symmetry does not give more
information about that constant, but it forbids vacuum expectation values
for all operators transforming non-trivially under the Lorentz group.
Let us consider next a Wightman 2-point function of identical scalar
operators,
(3.19)
In this case, translation symmetry tells us that

(3.20)

If we think of the correlator as being a function of and , then

this Ward identities establishes that there is no dependence on the former,
i.e.
(3.21)
where W denotes a function that is so far arbitrary.
In general, the consequences of translation symmetry are easier to see in
momentum space, using the Fourier transform of the local operators. By
Eq. (3.12), we can establish that
(3.22)
and therefore the Fourier transform of the Wightman 2-point function obeys
(3.23)
From this, we conclude that the 2-point function is proportional to a Dirac
delta function:
(3.24)
The numerical factor is merely a convention. As the notation
suggests, is actually the Fourier transform of W,

(3.25)

Taking into account Lorentz symmetry, one can also establish that the
Wightman function W(x) can only depend on the Lorentz-invariant distance
, although there is a subtlety: this is a different function depending
whether x is space-like or time-like (future- or past-directed), as Lorentz
transformations act separately on each of these regions. In momentum
space, the same arguments says that must be a function of . In this
case, the condition that only states of positive energy exist requires that
vanishes unless , and therefore we can unambiguously
write
(3.26)
where is a function of the positive quantity , and is the Heaviside
2
step function.
The fact that the Wightman 2-point function (3.24) is proportional to a
delta function raises an important concern: in spite of their names,
Wightman functions are not functions but rather distributions (this is also
part of the Wightman axioms: they are in fact tempered distributions). Note
also that the same function computes the overlap between the two states
(3.27)
(Hermitian conjugation flips the sign of momenta). Therefore, the limit
corresponds to the norm of either of these states. But this limit is
clearly discontinuous, or the norm of the state infinite. The resolution of this
issue is that the objects and its Fourier transform are not
operators, but rather operator-valued distributions. In other words,
and are not states of the theory, as they have in fact
infinite norm. Formally, these operator-valued distributions only make
sense when they are integrated against test functions, defining

(3.28)

or

(3.29)

where f and are Schwartz-class test functions (smooth functions

decaying faster than any power at infinity). When acting on the vacuum,
these smeared operators give well-defined states, with finite norms. For
instance, we have

(3.30)

Test functions will not appear further in these lectures. For physicists, they
are mostly an annoyance that we prefer to avoid. However, it is important to
know that there exists a mathematically rigorous way of dealing with
Wightman functions. For one thing, this gives a proper justification of why
it is always fine to take the Fourier transform between the position- and
momentum-space representation, as tempered distributions always admit a
Fourier transform. But bear in mind that this is only true of Wightman
functions, not of time-ordered correlators.

3.3 Spectral Representation

The norm (3.30) is also giving away important information: it can only be
positive for any test function f if the function is positive,
(3.31)
is in fact the spectral density encountered in standard quantum field
theory textbooks, where we can often find it in the form

(3.32)

The spectral density is an essential tool in non-perturbative quantum field

theory. It can for instance be used in the construction of the time-ordered
correlation function
(3.33)
Unlike the Wightman function, this is not a tempered distribution, because
the function is not differentiable at the origin. Nevertheless, the time-
ordered product admits the simple representation

(3.34)

where the limit is understood. This is the Källen-Lehmann

representation for the time-ordered 2-point function.

Exercise 3.2 Derive the Källen-Lehmann representation. An elegant

derivation is to first show that the time-ordered 2-point function can be
written as the difference between the Wightman function and the vacuum
expectation value of a retarded commutator,
 The next step is to Fourier transform both terms in y after setting .
We know that the Wightman function (3.24) has a nice Fourier transform

which can be equivalently written

The retarded commutator is only non-zero in the forward light cone in y,

and therefore it also admits a Fourier transform that converges provided
that we give an imaginary part to q. Compute this Fourier transform, and
show that at real q it is equal to

The difference between these last two integrals can then easily be turned
into the Källen-Lehmann representation (3.34).

The spectral representation for the 2-point function gives familiar results in
non-interacting theories. A massive scalar field has for instance the spectral
density
(3.35)
and from this we recover the known massive propagator

(3.36)

In an interacting theory, the spectral density will get contributions

corresponding to particle production above a certain threshold (see
Fig. 3.1).
Fig. 3.1 Left: spectral density in the theory of a free massive field, and typical contributions above
the threshold for particle production (dashed line). Right: possible spectral densities in a scale-
invariant theory
The discussion applied so far to a generic quantum field theory without
conformal symmetry. Let us now examine the role of scale and special
conformal invariance, starting with the former.

3.4 Scale Symmetry

The assumption of scale symmetry coincides with the existence of a third
conserved charge besides and , namely the operator D. We
obtained before the commutator of a generic local operator with
assuming that it transforms in some irreducible representation of the
Lorentz group at the origin (an operator inserted at some other point
is not in an irreducible representation because does not commute with
). Since commutes with D, the same assumption can be made
about scale: any local operator can be decomposed further into irreducible
representations of the group of scale transformations, meaning that we can
write
(3.37)
is called the scaling dimension of the operator (each local operator of
the theory has its own scaling dimension). The factor of i ensures that is
a real number when is a real operator. As before, we can use Eq. (3.3) to
obtain the commutator at any other point x:
(3.38)
Note that this is consistent with the transformation rule (2.​62) for a classical
field: the scaling dimension coincides with the mass dimension of the
operator in a free theory.
Using this new commutator and assuming that the vacuum state is
invariant under scale transformations, a new Ward identity can be obtained
for the Wightman 2-point function,

(3.39)

or equivalently, using Eq. (3.21),

(3.40)

The corresponding condition on the momentum-space 2-point function is

obtained from the Fourier transform, using integration by parts:

(3.41)

The solution to this equation consistent with the form (3.26) is unique, up to
a multiplicative constant C,
(3.42)
implying that the spectral density is a power of the energy,
(3.43)
This kind of spectral density, shown in Fig. 3.1, is very different from that
of a massive interacting theory: the operator creates states of all
energies. At the same time, its simplicity is striking: it is characterized by a
single parameter , and a normalization constant that does not carry
physical information (the operators can be re-defined to absorb this
constant).
It is instructive to compute the time-ordered function using the Källen-
Lehmann representation: performing the integral over , we find

(3.44)
The term in the integral looks like the propagator for a
massless scalar field raised to a non-integer power. This is in fact what we
expect in perturbation theory when the -function has a non-trivial fixed
point: the renormalized 2-point function has logarithms that can be re-
summed into a power controlled by the anomalous dimension of the field
,

(3.45)

In this case, the scaling dimension of the scalar operator corresponding to

the renormalized field is

(3.46)

In the limit , we recover the free propagator mentioned above. Note

however that the factor in front of the integral diverges in this limit, unless
the coefficient C satisfies

(3.47)

Assuming that it is the case (see below), the spectral density obeys3
(3.48)
This is precisely the spectral density that is expected in the free scalar field
theory. In this case (and this case only!), the operator describes a
massless scalar particle.
It turns out that is the lowest possible value for : for any
below that value, the spectral density is not integrable in the limit .
One might also worry about the opposite limit in the integral: for
any , the spectral density grows with . However, remember that
this spectral density is in fact a Wightman function, i.e. a tempered
distribution that should be understood as integrated against test functions
that decay faster than any power at large . Therefore, arbitrarily large
values of are possible, but there is a lower bound on below which the
states smeared with test functions have infinite norm. The inequality
 (3.49)

is known in the literature as the unitarity bound for scalar operators.4 Note
that any scalar operator that saturates this unitarity bound has
(3.50)
which implies
(3.51)
or in position space
(3.52)
Since this is true for any x and y, it implies that
(3.53)
is true as an operator equation. Since this is the equation of motion for a
free field, a theory in which is a free field theory.
The simplicity of the momentum-space 2-point function in a scale-
invariant theory also means that it can easily be Fourier transformed back to
position space using

(3.54)
 Exercise 3.3 Perform the Fourier transform explicitly. You can use the
fact that the integral is Lorentz invariant to determine that W(x) is in fact
a function of . Moreover, since the integrand only has support for q in
the forward light-cone, this defines a function of x that is analytic in x as
long as is contained in the future light cone: the integrand is
damped by the exponential with in that case (this
domain of analyticity is known as the “future tube”). This means that we
are free to evaluate the integral at a point , and then use
to recover the general solution. The integral at that point is
convergent for all satisfying the unitarity bound, and you should find

The result of this integral can be written as

(3.55)

where the limit is understood to make sense of the case in which

, namely

(3.56)

The coefficient and C are related by

(3.57)

Note that the proportionality factor is positive for all satisfying the
unitary bound (3.49). This implies that the 2-point correlation function is
always decreasing with the distance and not the other way around. It is in
fact customary in conformal field theory to normalize the scalar operator
so that , in which case C vanishes as in Eq. (3.47) in the limit
.
 Finally, let us conclude the analysis of scale symmetry with a comment
about one-point functions. We saw in the previous section that a constant
vacuum expectation value for a scalar operator was compatible with
Poincaré symmetry. However, the commutator (3.38) requires then that
, which violates the unitarity bound. We conclude that all one-point
functions must vanish in a scale-invariant theory.

3.5 Special Conformal Symmetry

As with scale symmetry, the presence of special conformal symmetry is
associated with the existence of the conserved charges , organized in a
d-dimensional vector. Unlike D, however, does not commute with
, so it cannot be diagonalized at the point . Nevertheless, we can
use the conformal algebra to establish that, if is a local operator with
scaling dimension , then has scaling dimension :

(3.58)

This is similar to the observation that the commutator has

scaling dimension ,
(3.59)
consistent with the fact that the derivative has mass dimension . The
fact that lowers the scaling dimensions appears to be in contradiction
with our findings of the last section stating that is bounded below: given
any local operator, one can always construct other local operators with
arbitrarily smaller scaling dimension.
The only way out of this apparent paradox is to assume that at some
point the action of annihilates the operator. In other words, there must
exist some local operator such that
(3.60)
We call this local operator a primary. Any other local operator can be
obtained acting on a primary with , and we call it a descendant. Since
the action of coincides with taking derivatives, a primary operator is
simply an operator that cannot be written as the derivative of some other
operator. Unless specified otherwise, we shall from now on only consider
Wightman correlation functions of primary operators. Descendants will be
explicitly denoted with a derivative.
The transformation of a primary operator away from the origin can once
again be obtained from Eq. (3.3). Note that since the commutator of
and involves both D and , this transformation depends on the
scaling dimension and on the Lorentz representation of the operators, i.e. on
the eigenvalues and . We find
(3.61)
This equation also defines the commutator of a momentum-space operator:
using integration by parts in the definition (3.12), one can show that this
amounts to replacing and , so that

(3.62)

or after permuting the derivatives with q,

(3.63)

This is now a second-order differential acting on the operator expressed in

momentum space.
The first thing we can do with this commutator is to examine the related
Ward identity for the Wightman 2-point function. Remember that this
function can be written as
(3.64)
The commutator acts trivially on the operator inserted at the origin, so that
we must have (note that for scalar operators)
(3.65)
Let us check at space-like x: using , we have
, and therefore the differential equation is
readily satisfied. The same can be verified in momentum space: by
definition, we have
(3.66)
where only the operator on the right is Fourier transformed while the one on
the left is kept fixed at the origin in position space, and so the commutator
above implies that

(3.67)

With , this equation is again satisfied.

Distinct operators: The fact that W(x) and readily satisfy the
constraint imposed by special conformal symmetry is very specific to
identical scalar operators. In every other case, special conformal symmetry
adds more constraints than Poincaré and scale symmetry alone. The
simplest example is that of a 2-point function of distinct scalar operators,
(3.68)
This is still a function of by Poincaré symmetry. But there are
now two distinct scaling dimensions and corresponding to the
operators and , and the Ward identity for scale symmetry becomes
(for simplicity setting at the origin)

(3.69)

The solution is fixed up to a multiplicative constant to be (assuming x

space-like for simplicity)

(3.70)

The Ward identity for special conformal transformation is obtained from the
commutator (3.61), giving
(3.71)
Using , this implies

(3.72)
If the scaling dimensions are different ( ), then must vanish.
This is an important lesson: in conformal field theory, only primary
operators of identical scaling dimensions can have non-zero 2-point
functions.
In fact, if there are several scalar operators with the same scaling
dimension with , then

(3.73)

where is a symmetric matrix. By unitarity, this matrix must be

positive-definite: if this were not the case, then one could define a negative-
norm state by taking an appropriate linear combination of the and
smearing. Therefore, it is always possible to choose a basis of operators in
which is diagonal. Moreover, the operators can be normalized so that
. From now on, we will therefore always assume that the only
non-zero 2-point functions are those involving identical operators.
Operators with spin: The other situation in which special conformal
symmetry plays an essential role is when the operators carry spin. Let us
take the simplest example of a vector operator , and denote its 2-
point function by
(3.74)
As with scalars, one can also take the Fourier transform of this tempered
distribution, defining

(3.75)

Again, this function corresponds to the momentum-space correlation

function without the delta-function imposing momentum conservation,
namely
(3.76)
By Lorentz symmetry, this function of a single momentum can be
decomposed into two different tensor structures multiplying scalar
functions,
 (3.77)
Moreover, using scale symmetry and energy positivity, we can infer that the
functions are just powers of over the forward light cone,

(3.78)
where is the scaling dimension of the operator and , two
constants that cannot be related by scale and Poincaré symmetry only.
There is a good reason for using precisely the two tensor structures in
Eq. (3.77) and not, say, and . Thanks to energy positivity, it is
always possible to choose a Lorentz frame in which .5 In this frame,
the momentum is invariant under the group of spatial rotations,
and therefore the 2-point function can be decomposed into irreducible
representations of that group. The part proportional to only appears in
the component , and it transforms like a scalar under rotations.
Conversely, the part proportional to only has non-zero entries for spatial
Lorentz indices ; it is in fact proportional to the identity in the
subspace, i.e. it is the invariant tensor for the vector representation of
. In particle physics language, we would call these two parts
respectively longitudinal and transverse.
Being able to use irreducible representations of is an
advantage of working in momentum space: there is no obvious Lorentz
frame in which such a decomposition can be made in position space since
the 2-point function has support over all of Minkowski space-time. The
disadvantage of working in momentum space is that the Ward identity for
special conformal transformation is a second-order differential equation in
p, while it is a first-order differential in position space. This Ward identity
can nevertheless be straightforwardly applied to Eq. (3.77), and it yields a
relation between the longitudinal and transverse parts, i.e. between the
coefficients and , given by (see exercise)

(3.79)

This is a very important consequence of special conformal symmetry: while

in a scale-invariant theory the longitudinal and transverse polarizations are
independent, in a conformal theory they are related.
 Exercise 3.4 Using the definition (3.75) for the function ,
show that it satisfies the special conformal Ward identity

where is given by Eq. (3.5), and use this to prove the relation
(3.79).

This has consequences on the possible values that can take. As before,
this 2-point function computes the norm of a state, and its positivity
requires:
so that the 2-point function is integrable at ;
and both positive, so that the norm is positive for any choice of
external polarization vector (i.e. the tensor must be positive-
definite). This requires and to have the same sign.
The combination of these two conditions in dimensions (spin is
treated differently in ) implies
(3.80)
This is known as the unitarity bound for a vector operator.
As in the scalar case, something special happens when the unitarity
bound is saturated ( ). In this case the 2-point function has no
longitudinal component, , and vanishes when contracted with
or . This implies that the longitudinal part of the state is null,
(3.81)
or equivalently that is an operator that only creates transverse-
polarization states. The equivalent statement in position-space is
(3.82)
In other words, is a conserved current. The equivalence goes both way:
any vector operator with is a conserved current, and any
conserved current must have scaling dimension . This also
shows that conserved currents are primary operators: they cannot be written
as acting on another vector operator (that operator would have
, below the unitarity bound), nor as acting on a scalar
operator , because the conservation requirement would then imply
, which is only possible if has scaling dimensions
and thus the current (again below the unitarity bound).
The fact that 2-point functions of primary operators are completely
fixed by conformal symmetry up to a choice of normalization is not specific
to scalar and vector operators. In fact, any local operator specified by a
representation under the Lorentz group and a scaling dimension defines an
irreducible representation of the conformal group , and as such its
2-point function is fixed by group theory. This also explains on more
general grounds why 2-point functions of distinct operators vanish. The
construction of all unitary representations of the conformal group in
dimensions was performed by Mack in 1975 [1], and similar constructions
can be done in other dimensions. Some Lorentz representations are specific
to a given dimension d, and others exist in any d, like the scalar and the
symmetric, traceless representations with Lorentz indices (the vector
discussed above is a special case corresponding to ). All such
symmetric tensors satisfy the unitarity bound
(3.83)
and they are in general described by distinct polarizations (irreducible
representations of the rotation group), except when the bound is saturated,
in which case there is just a single, transverse polarization and the operator
is a higher-spin conserved current. In terms of representations of the
conformal group, generic operators are said to belong to long multiplets,
whereas special cases such as scalars with or symmetric
tensors with are said to be in short multiplets (they contain
fewer descendants).
 Exercise 3.5 Construct explicitly the 2-point function of a 2-index
symmetric traceless operator . As a starting point, let us
decompose the momentum-space correlation functions into tensors that
transform covariantly under rotations in the rest frame. Using the
transverse projector

satisfying , this can be done as

Then write down the Ward identity for special conformations (including
the spin operator for a 2-index tensor, which you have to determine), and
show that it leads to the conditions

Argue that this gives rise to the unitarity bound , in agreement

with Eq. (3.83). Conclude that the energy-momentum tensor is a
primary operator with .

3.6 UV/IR Divergences and Anomalies

The discussion has been focused so far on Wightman functions. Besides
having a Hilbert-space interpretation and satisfying conformal Ward
identities that give strong constraints on their possible form, Wightman
functions are also free of divergences. In momentum space, regularity at
small momenta (IR) is enforced by the unitarity bound, whereas the power-
law growth at large momenta (UV) is compatible with the damping
provided by test functions. In position space, the apparent singularity at
short distance (UV) is resolved by the that provides an unequivocal
prescription for deforming any contour of integration.
These properties are not found in time-ordered products. Let us consider
once again the scalar 2-point function. Using the standard CFT
normalization (3.57) and the Käsllen-Lehmann representation, we find

(3.84)

This expression diverges whenever with integer n due to the -

function multiplying the integral, indicating that the Fourier transform does
not exist. In this case, the 2-point function has scaling dimension ,
and is therefore compatible with contact terms of the form
(3.85)
In path integral language, this is the situation in which the source field J for
the operator has scaling dimension , and therefore
contact terms of the form can (and must) be added to the action.
These contact terms are obviously covariant under Poincaré and scale
transformations, so they can be added to the correlation function without
affecting the Ward identities at separated points. (3.85) is the only possible
contact term appearing in a scalar 2-point function, but more terms can
appear in other correlation functions.
When Fourier transformed to momentum space, all such contact terms
become polynomials in the momenta. In the scalar 2-point function case,
these are . Polynomial terms are incompatible with the positive-
energy condition of Wightman functions, as they have support over all
causal regions. But there are allowed (and in fact required) in time-ordered
products. Time-ordered products are defined in position space as the
product of Wightman functions, which are tempered distributions, with step
functions, which are not: therefore they do not necessarily have a Fourier
transform. Contact terms can be seen as a way of “fixing” the time-ordered
product so that they can be Fourier transformed.
 This can be understood by analytic continuation in scaling dimension
. Let us assume that our scalar operator has scaling dimension
, and take the limit (note that this is different from
the one in the limit). Then the time-ordered 2-point function in
momentum space takes the form
(3.86)
where the first term with a pole in comes from the expression (3.84) valid
at , and the second term is the counterterm added to the action.
Choosing allows to cancel the divergence as , but it also
gives rise to a logarithm,

(3.87)

The fact that we need to introduce a dimensionful quantity is the sign of a

conformal anomaly: a 2-point function of this form does not satisfy the
Ward identity for scale transformations (3.41),

(3.88)

However, the anomalous term on the right-hand side is a polynomial in the

momenta, corresponding to a contact term, indicating that the anomaly is
local. In QFT language, this is typical of a (renormalized) UV divergence.
Note that the presence of contact terms is associated with a very special
type of operator of (half-)integer scaling dimensions. Such operators are
generically absent in an interacting conformal field theory, where the
scaling dimensions take irrational values. However, an exception to this rule
concerns conserved currents in even space-time dimension d. For instance,
a conserved current in has scaling dimensions ; it is
therefore associated in the path integral language with a source that
carries dimension 1. This source is also subject to a gauge symmetry,
, and therefore a possible contact term is , where
. As in the scalar case, this term must be used as a
counterterm to cancel a divergence arising in the time-ordered correlation
function, leading generically to logarithms in correlation functions
involving the transverse polarization of the current,
 (3.89)

As in the scalar case, this is the manifestation of a UV divergence.

However, the logarithm also implies that the limit diverges: this is
what we would call an IR divergence in quantum field theory. In the context
of CFT, there is no clear distinction between UV and IR divergences, as the
two are closely related. They both arise from an ambiguity in taking the
Fourier transform. UV divergences can be cured with local counterterms, at
the price of introducing a reference scale, but IR divergences are physical.
Note finally that time-ordered correlation functions involving a
conserved current do not only have transverse polarizations: while the
conservation condition implies the vanishing of the state
(3.90)
and thus of Wightman correlation functions constructed from that state, this
is not true of the divergence of appearing in a time-ordered product: the
conservation equation is only true as an operator equation,
namely away from coincident points. On general grounds, one expects

(3.91)

where indicates the charge of the field under .6

Reference
1. G. Mack, All unitary ray representations of the conformal group SU(2,2) with positive energy.
Commun. Math. Phys. 55, 1 (1977). https://​doi.​org/​10.​1007/​BF01613145

Footnotes
1 For operators of half-integer spin, the commutator must be replaced by an anti-commutator.

2 It satisfies for , and for .

3 This limit can be verified by integrating both sides in and taking the limit afterward.

4 It is usually derived using the action of special conformal transformations, but we have just seen
here that it applies to scale-invariant theories as well.

5 This is similar to going to a massive particle’s rest frame.

6 It is conventional to normalize the conserved current J so that Eq. (3.91) is true, which means that
the 2-point function of J cannot be arbitrarily normalized like a scalar operator in Eq. (3.57). The
same is true of the energy-momentum tensor, and of higher-spin conserved currents. For all these
operators the normalization of the 2-point function involves a coefficient of physical relevance.

OceanofPDF.com
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023
M. Gillioz, Conformal Field Theory for Particle Physicists, SpringerBriefs in Physics
https://doi.org/10.1007/978-3-031-27086-4_4

4. Conformal Correlation Functions

Marc Gillioz1
(1) Zürich, Switzerland

The method presented in the last section using conformal Ward identities
and the Wightman axioms could in principle be used to determine 3- and
higher-point correlation functions.1 However, it is quite inconvenient, and it
hides the simplicity of the result. To put things in perspective, note that the
Wightman 3-point function of scalar operators in momentum space was
only constructed in 2019 [3, 4],2 whereas the position-space correlator has
been known since the work of Polyakov in 1970 [7].

4.1 From Minkowksi Space-Time to Euclidean

Space
For a start, let us go back to the Wightman 2-point function of scalar
operators, given in position space by

(4.1)

The very definition of this 2-point function with its prescription suggests
that should be thought of as a complex variable: as a function of a
complex , W is analytic in the upper-half complex plane. Taking to be
purely imaginary, i.e.
(4.2)
we obtain the Schwinger function
 (4.3)

We denote with the Euclidean vector that is contracted with

the d-dimensional Euclidean metric. Schwinger functions will always be
written using the “average” notation , as opposed to the Wightman
functions that can be interpreted as vacuum expectation values;
since the two types of functions cannot be confused, we shall denote the
Euclidean coordinates by x instead of , even though the latter is implied.
Note that the Schwinger function transforms covariantly under the
Euclidean conformal group that is obtained by replacing the
Minkowski metric by the Euclidean one. This includes both translations and
rotations, and as a consequence we have
(4.4)
exhibiting a new symmetry under the exchange of the order of the two
operators. Also note that the Schwinger 2-point function is positive, and that
it is not defined at the point , unlike the Wightman function whose
prescription indicates how to approach any null point.
 Exercise 4.1 There is another way to get to the same result, starting
from the momentum-space representation of the 2-point function. The
Wightman function is not well-suited to do so (at least not without
exploring its analyticity properties following from the micro-causality
axiom), but the time-ordered function is: starting from the Källen-
Lehmann representation (3.​84), which in momentum space becomes
(using a notation in which the momentum-conservation delta function is
implicit)

one can perform a Wick rotation in which and are simultaneously

rotated in opposite directions in the complex plane, to arrive at the
Euclidean result

Perform the Fourier transform in to recover the Schwinger function

(4.3). Note that you will need to make assumptions about for the
Fourier integral to converge. The result is however analytic in , so you
can argue a posteriori that it must be valid for all scaling dimensions.

This construction of a Euclidean function by analytic continuation from the

Wightman function can in fact be generalized to any number of operators.
Consider the n-point Wightman function parameterized as
(4.5)
and complexify all time components . This function of several complex
variables (and many more real ones) is in fact analytic in every upper-half
complex plane in to , because it can be written as the Fourier
transform of a function that only has support when the dual variable to
are all positive.3 Therefore, going to purely imaginary times ,
one obtains the Schwinger n-point function
(4.6)
in which the Euclidean times all satisfy , i.e. the operators are
ordered along the Euclidean time direction. This latter observation is
however irrelevant because the ordering of operators does not matter in a
Schwinger function: the analytic continuation can be performed starting
with a configuration in which all real Minkowski times are equal, in which
case the operators commute by micro-causality. The observation made on
the 2-point function is therefore valid more generally:
Schwinger functions are symmetric under the exchange of operators.
The other observations made before are also true in general:
Schwinger functions transform covariantly under the Euclidean
conformal group . This property will be very useful because
it means that we can use finite conformal transformations that act nicely
in Euclidean space (including as a point) to simplify our
computations.
Schwinger functions are not defined at coincident points. This is not a
bug but a feature: using functions that are only defined at separated
points means that we do not need to worry about contact terms and UV
divergences.4 It means however that we cannot simply take the Fourier
transform of these functions to obtain momentum-space Schwinger
functions.
Schwinger functions enjoy a property called reflection positivity: if the
operators are organized in a configuration that is invariant under
reflection across a plane (e.g. four points at the corners of a square; or
trivially any two points), then the correlator is positive.5
We have seen that the Wightman functions let us define Schwinger
functions by analytic continuation. But it turns out that the opposite is also
true: the Osterwalder-Schrader theorem states that the properties of
Schwinger functions listed above are sufficient to reconstruct Wightman
functions.6 This means that we can in fact focus on the Euclidean
Schwinger functions for all our purposes, as any other physical observables
can be reconstructed from them (we do not claim that this reconstruction is
easy, though).
The only piece of information that we shall take for granted from the
analysis of Chap. 3 is the unitarity bound on the scaling dimension of
operators. Working out these bounds purely from the Schwinger functions is
also possible [11], but we will not discuss this procedure here.

4.2 From Euclidean Space to Embedding Space

Once we are dealing with correlation functions that transform under the
Euclidean conformal group , it makes a lot of sense to make
use of an analogy with the Lorentz group in dimensions to gain
7
mileage. We already introduced in Chap. 2 a set of coordinates in this
-dimensional, embedding space, and a metric obeying
(4.7)
Note that the directions labeled with indices are all space-like: the time-
like direction is .
The question is how do we go from to without
explicitly breaking the -dimensional Lorentz symmetry. This can be
done in two steps:
1. Restrict our attention to the future light-cone with ,
which is an invariant subspace.

2. Identify any two points related by a scale transformation on this light

cone, i.e. with .

This means that we are essentially considering a map between a point

in d-dimensional Euclidean space and a light ray in a -
dimensional Minkowski space-time. To make the map explicit, we choose a
section of the cone, which we will take to be , and
identify

(4.8)

Conformal transformations now act linearly in embedding space,

(4.9)
To obtain the action on , we first map it to our preferred section of the
cone using Eq. (4.8), apply the Lorentz transformation on , then
perform a (space-time dependent) rescaling to get
back on the preferred section, and read off .
Local operators in Euclidean space must also be lifted to the embedding
space, or at least to the null cone. On our preferred section, we declare
(4.10)
Then primary operators are defined on the rest of the cone by the scaling
rule
(4.11)
Note that this rule can only apply to primary operators: descendants
obtained acting with derivatives do not satisfy the same scaling property.
This choice gives the right transformation rules for primary operators under
all infinitesimal conformal transformations. This can be verified explicitly
(see the exercise below), or argued as follows: a conformal transformation
is a composition of a -dimensional Lorentz transformation, which
locally acts on the operator like a rotation or a boost, followed by a
position-dependent scale transformation to get back to the preferred section,
which by the rule (4.11) amounts to a local scale transformation with
weight given by the scaling dimension of the operator. The combination
of these two local transformations is precisely what we expect from the
conformal transformation of an operator.

Exercise 4.2 Verify that a Lorentz boost in the direction of

corresponds to a scale transformation in Euclidean space, in agreement
with Eq. (2.​39), and that operators transform accordingly.

A bit more care is required to define operators that carry spin on the
projective null cone, as the additional Lorentz indices in the directions of
and imply that there are additional degrees of freedom that
need to be constrained [13]. This can be achieved by imposing
transversality in embedding space, together with a gauge symmetry
condition. But we do want to dive into this level of technical detail here. We
will therefore restrict our attention to correlation functions of scalar primary
operators only.
Once the transformation properties of local operators are clear, the
construction of correlation functions of n points to follows two
simple rules:
The correlators must depend on Lorentz-invariant quantities in
embedding space, i.e. scalar products of the form . Since
on the null cone, only scalar products with can appear.
Applying a local scale transformation under which all
primary operators with scaling dimension satisfy
(4.12)

the correlator must transform homogeneously as

(4.13)

These rules imply immediately that there cannot be one-point functions,

as there simply is no corresponding Lorentz-invariant quantity on the
projective null cone. The simplest non-trivial case is that of a 2-point
function, which must obey
(4.14)
Note that this is only consistent with the homogeneous scaling rule if both
operators have the same scaling dimension, another property that was
derived in the hard way in Chap. 3. To recover the dependence on the d-
dimensional Euclidean coordinates, we simply use the identification (4.8),
yielding

(4.15)

Upon fixing the proportionality factor in the above equation, one recovers
the expected result

(4.16)
4.3 3-Point Functions
The embedding space formalism becomes very interesting when examining
the 3-point function
(4.17)
where the 3 scalar operators have now possibly distinct scaling dimensions
, and . Solving the conformal Ward identities for this 3-point
function would be an annoying task. Instead, by the above rules, we
immediately know that this is a function of the 3 invariant quantities
(4.18)
Moreover, by the homogeneous scaling rule, the only possible form of the
3-point function is
(4.19)
with the exponents satisfying

(4.20)

This system of equations admits as unique solution

(4.21)

(4.22)

In terms of Euclidean coordinates, this can be written

(4.23)

where we have introduced the compact notation

(4.24)
and

(4.25)

Equation (4.23) is truly special. It should be compared with the most

general 3-point function invariant under Poincaré and scale symmetry only:
in this case, any term of the form
(4.26)
with

(4.27)

satisfies the symmetry requirement, so that the 3-point function could take
the form

(4.28)

with infinitely many free coefficients . Instead, the conformal 3-point

function (4.23) is fixed up to a unique multiplicative coefficient .
To give another point of comparison, let us examine a 3-point function
that involves a descendant operator. Since the focus is on scalar operators,
let us act on the first operator in (4.23) with :

(4.29)

Unlike the correlation function of primary operators, this is now the sum of
three terms with distinct powers of the distances , all of which are
individually consistent with Poincaré and scale symmetry.
The coefficients multiplying all three terms are peculiar: there are
actually special cases in which this 3-point function takes the general form
of Eq. (4.23), i.e. that of correlator involving primary operators. These
special cases are not accidental, but correspond to physically interesting
situations:
If , then either or , and in both
cases two terms on the right-hand side of Eq. (4.29) vanish. This is a very
special situation in which the primary 3-point function factorizes into a
product of 2-point function, e.g. when ,

(4.30)

This situation is realized in generalized free field theory (definition in

Chap. 6 below), where is a composite operator .
If , then the first two terms in Eq. (4.29) vanish. As we saw in
Chap. 3, only a free scalar field satisfying the equation of motion
can have such a scaling dimension. Since then obviously
, it is natural that the correlator involving the
equation of motion takes the form of a primary 3-point function.
However, we also know that it must vanish identically, which implies that
either the 3-point coefficient vanishes, or that the additional
condition is satisfied: this is for instance the case of a
3-point function involving the primary composite operator ,

(4.31)

which is non-zero but vanishes under the action of .

Finally, let us go back to Minkowski space-time through an analytic
continuation in the opposite direction of what we did before. It is not hard
to see that the Wightman function of scalar primary operators satisfies

(4.32)

where now the Minkowski distances between two points are defined by
(4.33)
Note that , and as a consequence, the Wightman 3-point function
is not symmetric under the exchange of operators. Nevertheless, for real
operators satisfying , we must have
(4.34)
which is only compatible with Eq. (4.32) if the coefficient is real. This
property will in fact be essential in the conformal bootstrap discussed in
Chap. 6.

4.4 4-Point Functions

The embedding space technique can be used for higher-point functions as
well, but things are getting more complicated. To avoid dealing with four
distinct operators and as many scaling dimensions, let us focus our attention
on the case of 4 identical scalar operators,
(4.35)
In this case there are 6 Lorentz invariants with . It is easy to
see that a term of the form

(4.36)

satisfies all the constraints of conformal symmetry. It is however not

unique: so does

(4.37)

This shows right away that there is no hope of constraining the 4-point
function as much as we did with the 3-point function. In fact, one can
construct two invariant quantities out of the :

(4.38)

These are called conformal cross-ratios. Any function of u and v is

conformally invariant, and the most general 4-point function can be of the
form
 (4.39)

In terms of Euclidean coordinates, this can be written

(4.40)

with

(4.41)

To see that this is the most general result, let us consider the following
argument based on a sequence of finite conformal transformations:
1. Using translations, it is always possible to choose a reference frame in
which .

2. Using a special conformal transformation, one can then send

without moving away from the origin.

3. A rotation can then be used to place along some chosen direction,

followed by a scale transformation to get , without
touching the origin nor the point at infinity.

4. Finally, there is still a subset of rotations that do not affect , that can
be used to move the point to the position ,
shown later in Fig. 6.​1.

Note that the first 3 steps can be used to fix entirely the kinematics of a
3-point function, which explains why its only freedom is in the form of a
multiplicative coefficient. Instead, with a 4-point function we are left with
two quantities, a and b, which are in one-to-one correspondence with the
two conformal cross ratios. It is in fact convenient to replace these two real
numbers with a complex and its conjugate . The
cross-ratios are then related to z and by
(4.42)
Schwinger functions are analytic at all non-coincident configurations of
points, therefore the function g(z) is a single-valued function over the
complex plane minus the points . Note that g is also subject to
the crossing symmetry of the 4-point function, which requires8

(4.43)

upon exchanging the operators and , as well as

(4.44)

upon .
The situation is more complicated in Minkowski space-time. One can
still define cross ratios by Eq. (4.41), with Minkowski distances defined
through prescriptions as in (4.33). But the Wightman functions are not
analytic (they lie at the boundary of the domain of analyticity in
complexified coordinates), and therefore g is a multi-valued function. Its
values can be reached by analytic continuation from a configuration in
which all 4 operators live on a constant-time slice, in which case it
coincides with the Schwinger function. For instance, keeping , and
on that time slice but letting the time component b of become imaginary,
one ends up in a configuration in which z and are both real but distinct.
This procedure is in general tedious, and it is fair to say that the current
understanding of conformal Wightman 4-point functions is still
incomplete.9

References
1. M. Gillioz, From Schwinger to Wightman: all conformal 3-point functions in momentum space.
arXiv:​2109.​15140 [hep-th]
2.
H. Osborn, A.C. Petkou, Implications of conformal invariance in field theories for general
dimensions. Ann. Phys. 231, 311–362 (1994). https://​doi.​org/​10.​1006/​aphy.​1994.​1045. arXiv:​
hep-th/​9307010
3.
M. Gillioz, Conformal 3-point functions and the Lorentzian OPE in momentum space. Commun.
Math. Phys. 379(1), 227–259 (2020). https://​doi.​org/​10.​1007/​s00220-020-03836-8. arXiv:​1909.​
00878 [hep-th]
4.
T. Bautista, H. Godazgar, Lorentzian CFT 3-point functions in momentum space. JHEP 1, 142
(2020). https://​doi.​org/​10.​1007/​JHEP01(2020)142. arXiv:​1908.​04733 [hep-th]
5.
C. Coriano, L. Delle Rose, E. Mottola, M. Serino, Solving the conformal constraints for scalar
operators in momentum space and the evaluation of Feynman’s Master integrals. JHEP 7 011
(2013). https://​doi.​org/​10.​1007/​JHEP07(2013)011. arXiv:​1304.​6944 [hep-th]
6.
A. Bzowski, P. McFadden, K. Skenderis, Implications of conformal invariance in momentum
space. JHEP 3, 111 (2014). https://​doi.​org/​10.​1007/​JHEP03(2014)111. arXiv:​1304.​7760 [hep-th]
7.
A.M. Polyakov, Conformal symmetry of critical fluctuations. JETP Lett. Ser. 12, 381–383 (1970)
[ADS]
8.
Y. Nakayama, Conformal contact terms and semi-local terms. Ann. Henri Poincare. 21(10),
3201–3216 (2020). https://​doi.​org/​10.​1007/​s00023-020-00951-z. arXiv:​1906.​07914 [hep-th]
9.
P. Kravchuk, J. Qiao, and S. Rychkov, Distributions in CFT. Part I. Cross-ratio space. JHEP 5,
137 (2020). https://​doi.​org/​10.​1007/​JHEP05(2020)137. arXiv:​2001.​08778 [hep-th]
10.
P. Kravchuk, J. Qiao, S. Rychkov, Distributions in CFT. Part II. Minkowski space. JHEP 8, 094
(2021). https://​doi.​org/​10.​1007/​JHEP08(2021)094. arXiv:​2104.​02090 [hep-th]
11.
S. Minwalla, Restrictions imposed by superconformal invariance on quantum field theories. Adv.
Theor. Math. Phys. 2, 783–851 (1988). https://​doi.​org/​10.​4310/​ATMP.​1998.​v2.​n4.​a4. arXiv:​hep-
th/​9712074
12.
P.A.M. Dirac, Wave equations in conformal space. Annals Math. 37, 429–442 (1936). https://​doi.​
org/​10.​2307/​1968455
13.
M.S. Costa, J. Penedones, D. Poland, S. Rychkov, Spinning conformal correlators. JHEP 11, 071
(2011). https://​doi.​org/​10.​1007/​JHEP11(2011)071. arXiv:​1107.​3554 [hep-th]

Footnotes
1 The conformal Ward identities give differential equations and the Wightman axioms boundary
conditions. Together, these are sufficient to constrain the conformal correlation functions. Note
however that energy positivity alone is not sufficient: the micro-causality axiom is required as well
(see Ref. [1] for a discussion at the level of 3-point functions). Alternatively, for time-ordered or
Euclidean correlators, the boundary condition can be given by an OPE limit [2] (see below).

2 Results for the Euclidean momentum-space 3-point function appeared in 2013 already [5, 6].

3 If we let all components of be complex, then the primary domain of analyticity is the so-called
future tube, defined by .
4 It is in fact possible to describe contact terms (local and semi-local ones) using a generalization of
the embedding formalism described in the next section [8]. But these contact terms are ambiguous:
they carry more information than the Schwinger or Wightman functions themselves, and as such are
unphysical.

5 This property would not be needed if we had been studying Euclidean conformal field theory from
the start. Indeed, there are interesting critical fixed points in condensed matter or statistical physics
that are described by CFTs that are not reflection-positive (often called non-unitary). However, the
conformal bootstrap described in Chap. 6 relies on this property in an essential way.

6 There is in fact another property that is needed in the proof, called linear growth condition. This
property is very difficult to establish in quantum field theory. But within conformal field theory the
linear growth condition is not necessary if one works instead with a set of “Euclidean CFT axioms”,
which are otherwise equivalent to the Osterwalder-Schrader axioms, and from which the Wightman
axioms can be recovered, at least for correlation functions of up to 4 operators [9, 10].

7 This idea dates back to Dirac in 1936 [12].

8 It is standard in the CFT literature to call “crossing symmetry” the property of Schwinger
functions to be symmetric under the exchange of operators, in analogy with scattering amplitudes.
The two types of crossing are related, but they are not quite the same.

9 Working in momentum space is not helping, even though energy positivity reduces the number of
non-trivial configurations: since the Ward identity for special conformal transformation is a second-
order differential equation, the solution cannot be formulated in terms of invariants. The fact that
there exist conformal invariants in position space is intimately connected with Ward identities being
first-order differential equations.

OceanofPDF.com
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023
M. Gillioz, Conformal Field Theory for Particle Physicists, SpringerBriefs in Physics
https://doi.org/10.1007/978-3-031-27086-4_5

5. State-Operator Correspondence
and OPE
Marc Gillioz1
(1) Zürich, Switzerland

The construction of correlation functions could be continued beyond 4

points, but there is a good reason to stop here (at least in this course). We
shall now see that any n-point function can be reduced to a -point
function using the operator product expansion (OPE). The procedure can be
iterated until everything is expressed in terms of 2- and 3-point functions.
The 4-point function will be examined as a typical situation in which this
OPE can be applied, and higher-point functions will not be considered.

5.1 The OPE in Quantum Field Theory

The operator product expansion in quantum field theory is the statement
that when two local operators are “sufficiently close” to each other, they can
be replaced by another local operator, or rather a sum of them:

(5.1)

It does not matter whether the operator on the right-hand side is inserted
at x or at y, or at the middle point since this expansion is valid in
the limit in which x and y coincide. In fact, the proportionality factor
might diverge in the limit , so this is to be understood in the sense of
an asymptotic limit, whose radius of convergence is strictly-speaking zero.
 In non-perturbative quantum field theory, the OPE can be formulated in
terms of the Hilbert space: when the product of operators in Eq. (5.1) is
acting on the vacuum, then the completeness of the Hilbert space implies
that we can write

(5.2)

where the sum is over states forming an orthonormal basis. Among

these are states obtained acting with a local operator on the vacuum,
so that we have

(5.3)

where the object is related to a ratio of Wightman functions

(5.4)

This formulation remains however imprecise, and it is not quite obvious

how to make it more rigorous using only general principles of quantum
field theory.
In conformal field theory, however, the operator product expansion goes
to a completely new level of rigor, thanks to the following observations:
2-point functions of primary operators are diagonal (i.e. only identical
primaries have a 2-point function), which implies that states created by
different primaries are orthogonal. The norm of primary states is also
known in terms of the operators’ normalization.
3-point functions are known (see Chap. 4), which means that the
proportionality coefficients are in fact fixed up to an overall
multiplicative factor.
There is no other contribution to the OPE beyond those of local operators
acting on the vacuum. This property is due to the state/operator
correspondence that is discussed next.

5.2 The State/Operator Correspondence

In conformal field theory there exists a one-to-one correspondence between
the states on a given time slice and local operators defined by their scaling
dimensions and representations under the Lorentz group.
The fact that local operators define states is obvious in Minkowski
space-time, by the Wightman axioms.But it is also true in the analytic
continuation to Euclidean space. To understand this, remember that a local
operator inserted at a generic Minkowski coordinate x can be expressed
using Eq. (3.​3) as an operator living on the surface and evolved
unitarily,
(5.5)
Since the spectrum of is non-negative, this can be analytically continued
to any value of in the upper-half complex plane; on the contrary, if
had a negative imaginary part, then there would be states of arbitrarily high
energy, with divergent norm. By going to purely imaginary values,
, we can therefore define a state of the theory through an operator insertion
in Euclidean space, provided that this operator is inserted at Euclidean time
.
This has an immediate interpretation in terms of the Euclidean path
integral: a state on the Euclidean surface (which is identical to the
original Minkowski time slice ) is defined by a path integral over all
field configurations restricted to the region . This is valid with any
number of operators inserted at separated points of Euclidean time
(including no operators, corresponding to the vacuum state). But the
identification also works the other way around: a given state on the surface
defines a boundary condition for the path integral at , which
might correspond to some number of local operators inserted in the “past”
(or a superposition of such configurations). In this way, any Euclidean
correlation function in which the operators are ordered in Euclidean time
can be given a Hilbert space interpretation.
Once we adopt the Euclidean path integral perspective, then it does not
matter whether we use as the Hamiltonian, or another conformal
generator, as long as the surface at is part of the foliation that it
defines. For instance, one can use the so-called conformal Hamiltonian
, which foliates Euclidean space as shown in Fig. 2.​2. We
mentioned already in Chap. 2 that this combination of generators is in fact
equivalent to the dilatation generator D, in the sense that they are related by
a rotation in the compact subgroup of the conformal group
. This means that there exists a conformal transformation that
maps the surface to the unit sphere, and the other surfaces related by
Hamiltonian evolution to spheres of different radii (see again Fig. 2.​2).
Foliating Euclidean space with spheres centered at the origin and using the
dilatation generator as a Hamiltonian is called radial quantization. In
comparison, the physically-equivalent choice of Hamiltonian
is often called N-S quantization.
The importance of these two quantizations in conformal field theory is
due to the fact that they have fixed points, unlike equal-time quantization:
evolving back in “time”, the path integral shrinks to a ball of arbitrarily
small radius surrounding one of these fixed points (the origin in radial
quantization, or the south pole in N-S quantization). This means that any
state on the unit sphere, or on the plane , can be related by
Hamiltonian evolution to a state localized at the fixed point. It is therefore
equivalent to a state created by a local operator inserted there. This is in
essence the argument showing that the state/operator correspondence goes
both ways: a local operator defines a state, but any state also defines a local
operator (or rather a superposition of local operators).
This correspondence provides a posteriori a justification for our choice
to organize the operators into irreducible representations of the Lorentz
group with a definite scaling dimension: the diagonalization can be
performed at the level of the Hilbert space, and then radial quantization
used to argue that each state corresponds to a local operator. Note that there
are both primary and descendant states, in the sense that the local operator
does not have to be a primary.
Many introductory courses in conformal field theory begin in fact with
radial quantization, as it provides a compelling picture in Euclidean space.
The connection with unitary quantum field theory is not easy to establish,
though. Hermitian conjugation in the unitary theory requires taking
to , i.e., it corresponds to a reflection across the
surface of in N-S quantization. In radial quantization, this becomes
inversion : correlation functions compute the overlap between
states corresponding to a path integral inside the unit sphere with states
corresponding to a path integral outside the sphere. The smearing of states
required in Minkowski space-time is replaced by the limit needed
to define conjugate states in terms of local operators. Once this limit is
taken into account, the unitarity bounds can be derived using the interplay
of generators and , which are in fact conjugate to each other under
inversion. This procedure reproduces the results of our Chap. 3, but it might
be less intuitive.

5.3 The Conformal OPE

The lesson we learn from radial quantization is that every state in the
Hilbert space can be written as a linear combination of states created by the
action of a single local operator on the vacuum, including primaries as well
as descendants. Applying this lesson to Wightman functions in Minkowski
space-time, we can be tempted to write the complete Hilbert space as
(5.6)
The problem is that these are not normalizable states. Instead, we can
consider states created by primary operators inserted at any point x in
Minkowski space-time,
(5.7)
where we implicitly understand that the local operators should be smeared
against test functions. This gives a good description of the Hilbert space,
but it is not very practical because primary operators inserted at different
points (or smeared against distinct test functions) are not orthogonal to each
other. This leads us to consider yet another representation of the Hilbert
space, using local operators in momentum space,
(5.8)
States carrying different momenta are orthogonal to each other, and it is
moreover sufficient to consider momenta p inside the forward light cone.
This means that we can express the completeness of the Hilbert space
through the resolution of the identity

(5.9)
where the sum is over all primary operators of the theory. The denominator
is simply the scalar 2-point function, Eq. (3.​42), and states that carry spin
have been omitted for simplicity: they could be included in the same way
after inverting the Lorentz tensor appearing in the 2-point function, which is
positive-definite by unitarity.
This identity can be applied whenever we encounter a product of two
local operators acting on the vacuum to rewrite the product as a sum of
local operators. It takes its simplest form when the operators are all
expressed in momentum space:

(5.10)

This is the conformal OPE in Minkowski space-time, which reduces higher-

point functions to lower-point ones. It is trivial when applied to 2-point
functions, as only the vacuum state appears in the sum (the identity operator
can be interpreted as a primary operator in its own right). When applied to a
3-point function, it can be used to determine the proportionality factor in
terms of the scaling dimension, Lorentz representation, and 3-point
coefficients . Using this OPE further, a 4-point function can be
decomposed into a sum of terms whose kinematics is completely
determined by conformal symmetry.
This form of the OPE is however rarely used, for two reasons: our
understanding of Wightman 3-point functions in momentum space is
incomplete, and the convergence of this OPE is not particularly nice. By
this, we mean that the OPE converges in the distributional sense, i.e. only
after smearing with smooth test functions. In contrast, we will see next that
the Euclidean OPE is absolutely convergent!
Since we saw before that even Schwinger functions can be given a
Hilbert space interpretation, we might as well formulate the OPE directly in
Euclidean space. We write

(5.11)

Unlike the Minkowskian OPE (5.10), the convergence of this OPE is

conditional: it is only true inside a correlation function in which the points
and can be separated from all other insertions of local operators by a
quantization surface, say a sphere in radial quantization. Since the center
and radius of the sphere can be chosen at will thanks to translation and scale
symmetry, this includes many (in fact nearly all) configurations of points, as
we shall see below.
In the notation of Eq. (5.11), the object does not only depend on the
distance between the two operators, but also on derivatives acting on the
primary operator . This is a compact way of including all descendants in
the sum. Keep in mind however that there are infinitely many such
descendants, so is itself an infinite series. The only exception is the
identity operator, which enters the OPE if and only if the two primary
operators on the left-hand side of (5.11) are identical: in a 2-point function,
the OPE takes the form

(5.12)

with , from which we deduce

(5.13)

When viewed as a primary operator, the identity does not have any
descendant. In all other cases, the value of can be determined from
the 3-point function: using the OPE in the scalar 3-point function (4.​23),
one can for instance deduce that

(5.14)

where

(5.15)

and so on. This expression is valid when is a scalar operator, but there
are also operators with spin that enter the OPE, even though and are
scalars: the operators being inserted at separated points in space,
their product carries angular momentum; for a point-like operator on the
other side of the OPE, this angular momentum is realized as internal spin.
Note however that not all Lorentz representations appear in the OPE of two
scalars, but only symmetric tensors. Moreover, when the operators are
identical, there are only tensors with even numbers of Lorentz indices. The
list of representations that appear can be determined from the group theory
associated with Lorentz symmetry. On the contrary, any scaling dimension
can appear.
There are therefore still many unknowns in the OPE. We shall see in the
next section that there are in fact always infinitely many primary operators
in the sum. But when compared with the OPE in a generic quantum field
theory, the conformal OPE is extremely rigid: in a scale-invariant QFT the
coefficients a, , , , entering the definition of the coefficient
are all theory-dependent factors. In CFT, on the contrary, they are
completely fixed by the kinematics. The only dynamical information is
contained in the OPE coefficients multiplying the contribution of a
primary and all of its descendants.
OceanofPDF.com
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023
M. Gillioz, Conformal Field Theory for Particle Physicists, SpringerBriefs in Physics
https://doi.org/10.1007/978-3-031-27086-4_6

6. The Conformal Bootstrap

Marc Gillioz1
(1) Zürich, Switzerland

Our approach to conformal field theory so far has been an algebraic one: in
Chap. 3 we have defined primary operators as irreducible represen tations
of the conformal group, characterized by their transformations under
Lorentz symmetry and by a scaling dimension. In Chap. 5 we have seen that
these operators can be combined using the operator product expansion. The
OPE coefficients play the role of the structure constants of this operator
algebra. The set of all scaling dimensions and representations under the
Lorentz group of a theory, together with the OPE coefficients,
(6.1)
is called the CFT data. A conformal field theory is completely defined by its
CFT data: any correlation function can be computed from it, by repeated
use of the OPE.
But does any CFT data define a good theory? The answer to this
question is no: the algebra of local operators must close, giving very
constraining consistency conditions.1 The precise formulation of this
closure condition depends on whether we are talking about the Euclidean or
Lorentzian conformal groups. For a conformal field theory in Minkowski
space-time, the crucial point is that the OPE must be consistent with the
micro-causality condition (3.​9). But the consequences of micro-causality
are quite difficult to track down in practice.
Instead, there is a much simpler consistency condition for Schwinger
functions in Euclidean space: since these functions are symmetric under the
exchange of operators (a property that is in fact related to micro-causality),
then the Euclidean OPE must be associative. To illustrate this, let us focus
on 4-point functions. Imposing associativity of the OPE on all 4-point
functions is in fact sufficient to guarantee associativity of higher-point
functions. For simplicity let us consider as before the case of 4 identical
scalar primary operators with scaling dimension :

(6.2)

6.1 Conformal Blocks

The simplest way to understand how this 4-point function can be expressed
in terms of the CFT data is to use the OPE (5.​11) on it twice:

(6.3)

There is a single sum on the right-hand side because the 2-point function is
diagonal: it vanishes unless the two primary operators are identical. To put
this in a convenient form, let us factor out the OPE coefficient
corresponding to the 3-point function , as well as a particular power
of the distance, defining

(6.4)

The new function is then of the form

(6.5)
where is the scaling dimension of the operator entering the OPE (the
internal operator), not to be confused with the scaling dimension of the
operators in the original correlation function (the external operators). We
have now

(6.6)

where we have defined

 (6.7)

This should be compared with the representation (6.2) of the 4-point

function, in terms of which

(6.8)

The are called conformal blocks [1]: they represent the contribution of a
single primary and of all its descendants to the 4-point function. They are
conformally invariant,2 and must therefore be functions of the cross-ratios u
and v, even though this is not at all obvious from their definition.
This definition is in fact not very practical: working out all the terms in
the series of is difficult, and there are also Lorentz indices that need to
be contracted when the intermediate operator carries spin. But it is
convenient to examine the limit , , or equivalently
: in this case, the leading term in the OPE shows that
(6.9)
Primary operators with the lowest scaling dimensions give the leading
contribution to the 4-point function in this limit. The operator with the
absolute lowest scaling dimension is the identity, for which there are no
descendants, so that3
(6.10)
Note that this statement is related to the more general cluster decomposition
principle in quantum field theory: since there is no absolute scale in CFT,
the limit is equivalent to a limit in which the other points and
are sent very far away, all the way to infinity, and in this case one
expects on general grounds that the correlation function factorizes as
(6.11)
This factorization is thus reproduced by the OPE in the limit .
A better way of computing the conformal block is based on the Casimir
invariant of the conformal group [2]. From the group algebra, one can
verify that the combination of generators (2.​39)
 (6.12)

commutes with all individual generators, e.g.

(6.13)
In group theory language, is called the quadratic Casimir invariant.
Since it commutes with translations, its action on a state created by a local
operator is the same no matter where the operator is inserted, or whether it
is a primary or a descendant: for a scalar primary,
(6.14)

Exercise 6.1 Compute the eigenvalue of the quadratic Casimir operator

from the commutators of the conformal generators with . For
simplicity, work at .

For a symmetric tensor with Lorentz indices, the eigenvalue contains an

additional term that is simply the quadratic Casimir of the Lorentz/rotation
group in d dimensions,
(6.15)
The key idea is now to evaluate the value of this Casimir operator between
states that are each created by two local operators, i.e. compute the
correlation function
(6.16)
Applying the OPE once, this can be written

(6.17)

where there are now two possible ways of computing each term on the
right-hand side: either acts to the right, and then the eigenvalue equation
(6.15) can be used, or it acts to the left. In the second case, the individual
generators forming must be commuted with and
successively. This is most easily done in a convenient reference frame, for
instance in the z-frame described in Chap. 4 and shown in Fig. 6.1. In this
case, one finds that the action of on each individual conformal block is a
second-order differential operator in z and , which we denote by .
This means that every conformal block satisfies a differential equation of
the form
(6.18)
Together with a boundary condition provided by Eq. (6.9), this is sufficient
to determine the conformal blocks entirely. In even space-time dimensions,
the solution is in the form of products of hypergeometric functions. In odd
dimensions, there is no known closed-form solution, but the Casimir
equation can be conveniently solved term-by-term in a series expansion.
Note that in our specific example where the external operators are all
identical, the conformal blocks do not depend on the scaling dimension ,
but only on the scaling dimension and spin of the internal operator.
This property is not true in general.

Fig. 6.1 Two convenient conformal frames that describe the 4-point function. On the left, 3 points
are mapped to 0, 1 and using a conformal transformation, and the complex coordinate z
describes the position of the fourth point. Applying a conformal transformation in the plane, one can
reach the configuration on the right, with operators placed at antipodal points on a pair of concentric
circles, with the larger circle having unit radius; this configuration is parameterized by the complex
coordinate , satisfying

6.2 OPE Convergence

So far we have been using the OPE in the 4-point function without
worrying about its convergence. This is fine as long as the points and
can be separated from and by a sphere. In terms of the coordinate z,
this is obviously true for all (the unit disk delimited by a dashed
line in Fig. 6.1): then the product of operators can be replaced
by a sum of local operators at by the state-operator correspondence,
and therefore the OPE converges since it is a Hilbert-space sum. But the
domain of convergence is in fact much larger: radial quantization can be
used around any point in space, not just , and it is not hard to see that in
most configurations one can draw a circle surrounding and , but
excluding and . This is actually possible in all cases except when z is
real with .4
This is even more easily seen in a different reference frame: using a
conformal transformation in the plane of z, one can map the 4 points to a
configuration depicted on the right-hand side of Fig. 6.1, with the pairs of
points and placed at antipodal points on two circles
centered at the origin [3]. The map is given by

(6.19)

It takes all of the complex z plane to the unit disk in , with the half-line
being mapped to the unit circle. In this frame, it is now obvious that
the OPE converges in radial quantization unless .
Fig. 6.2 The -coordinate configuration of Fig. 6.1 as seen on the radial quantization cylinder. The
two segment and form an angle , and they are separated in cylinder time by
. and are related to the flat-space configuration by
This conformal frame is also very useful for understanding the
convergence properties of the OPE. It can be seen as a configuration on the
radial quantization cylinder, as in Fig. 6.2, with coordinates
(6.20)
where is the “time” component, and the “space” components
are unit vectors and that parameterize a direction on the sphere ,
and form an angle between them. The connection with the flat-space
configuration is through
(6.21)
A conformal block corresponds to the projection of this configuration onto
intermediate primary and descendant states with scaling dimensions
(n being an integer) and spin j (whose range is determined by the
spin of the primary). On general grounds, one can therefore expect the
conformal block to take the form
 (6.22)

where is a Gegenbauer polynomial obtained from the

contraction of the symmetric and traceless tensor with
. The coefficients are positive by unitarity because they
are the norm of eigenstates of definite spin (labeled by n and j). They are in
fact rational functions of . This representation is very useful: on the one
hand, it gives an efficient way of evaluating the conformal block at any
given level of precision by truncating the series in n, since by assumption
; on the other hand, it shows that the are nice (i.e. analytic)
functions of scaling dimension for all values of above the unitarity
bound.
Beyond individual conformal blocks, the coordinates also show how
the OPE converges as a function of r: in a configuration with , the
series is dominated by operators of low (both primary and descendants),
and a good approximation of the 4-point function can be obtained from a
truncated OPE. For instance, a configuration in which the operators are
placed on the corners of a square corresponds to , or
. Even though the point and are not
obviously close to each other in this configuration, is a small
number and the convergence of the OPE is very fast.
This is a fantastic property of the Euclidean OPE. The situation is not as
nice in Minkowski space-time. The power-law dependence in r seen in
Eq. (6.22) is the consequence of exponential damping in Euclidean “time”
( with negative ), following from Hamiltonian evolution in
radial quantization. In Minkowski space-time, independently of the choice
of quantization, the evolution is always unitary, e.g. : there is in
principle no reason why operators of higher scaling dimension should
contribute less than those of lower dimension in a generic configuration.
This does not mean that the Minkowskian OPE is uninteresting, but it is
more difficult to harness than the Euclidean one.

6.3 The Crossing Equation and Simple Solutions

We have so far only discussed one particular OPE, in which the points
and were assumed to be closed to each other in a conformal sense. But
since Schwinger functions are symmetric under the exchange of operators,
there is no reason not to consider different OPEs, e.g., between the
operators and , or and . In the vast majority of
configurations of 4 points, the space can be cut in half by a sphere in three
inequivalent ways. Even in the special case in which the 4 points are on a
circle, there are still two inequivalent ways of surrounding pairs of points
with a sphere. This means that we expect different OPEs to converge in all
cases.5

Fig. 6.3 Diagrammatic representation of the crossing equation at the core of the conformal
bootstrap, relating two distinct OPEs

This implies that besides the expansion

(6.23)

one can write a similar expansion for the 4-point function in a configuration
related by crossing symmetry, e.g.

(6.24)

Since the two are equal, we must have

(6.25)
This equation is represented pictorially in Fig. 6.3. Note that in the case of
identical external operators the two sums are over the same set of primary
operators, but in general these could be different sums.
This is a relatively simple equation, but it is quite difficult to solve. We
saw above that the OPE is dominated by intermediate operators of low
in the limit , but this only applies to the left-hand side: taking the
same limit on the right-hand side, one reaches the boundary of the domain
of convergence of the OPE. Using the known expression for the conformal
blocks in terms of hypergeometric functions, it can be shown that the limit
approaches a branch cut around which
(6.26)
This means that the right-hand side of the crossing equation is dominated
by terms of the form

(6.27)

in the limit . A finite sum of terms of this form cannot reproduce the
leading constant contribution from the identity operator on the left-hand
side. This is the first wisdom contained in the crossing equation (6.25): it
can never be satisfied block-by-block, but only by an infinite sum of
conformal blocks, and there must therefore be an infinite number of
primary operators in the OPE.
Before looking into the clever way in which the conformal bootstrap
deals with this problem, let us examine the simplest solution to crossing
symmetry that we know of. A possible function that transforms covariantly
under the conformal group and is crossing symmetric is the following
combination of 2-point functions:

(6.28)

corresponding to

(6.29)

Performing an expansion at small u and matching each term with a

hypothetical primary operator, one finds an infinite spectrum of operators
characterized by their spin and scaling dimension
(6.30)
where This defines a valid 4-point correlation function in a
theory called generalized free field theory (or sometimes mean free theory,
or Gaussian theory). It is similar to a free theory in the sense that the OPE
contains operators that have the same scaling dimension and spin as
composites of the field , of the schematic form
(6.31)
The free scalar theory is a special realization of this in which only operators
with are present in the OPE (the other vanish by the equation of
motion , or they are descendants), and these operators are higher-
spin conserved current. Generalized free field theory is usually considered
non-local: its definition through the above 4-point function does not include
an energy-momentum tensor.6 It is nevertheless a physically interesting case
because the large-N limit of gauge theory takes precisely this form: if one
considers the 4-point function of a gauge-invariant composite operator such
as the fermion bilinear , then its decomposition into conformal blocks
is given at leading order in 1/N by a sum of “double-trace” operators with
scaling dimensions , whereas “single-trace” operators
(among which the energy-momentum tensor) only enter at sub-leading
order in 1/N.

6.4 The Numerical Bootstrap

The revolutionary idea that gave birth to the modern conformal bootstrap
appeared in 2008 [4].7 The starting point is to rewrite the crossing equation
(6.25) as

(6.32)

and to realize that it can be viewed as an equation in an infinite vector space

(6.33)
where the are positive coefficients (remember that must be real in a
unitary quantum field theory). The components of the vector correspond
to evaluated at an infinite set of points (u, v), or
equivalently as the coefficients of a Taylor expansion around some
preferred point.8 has infinitely many components, but the equation is
also true if we truncate it to any finite subset. Some interesting results can
already be obtained from considerations about a 2-dimensional subset, but
typically the more stringent bounds that are shown below require scanning
over spaces with a large number of dimensions.

Fig. 6.4 A toy example of the behavior of vectors in a 3-dimensional space. For each spin
, draws a smooth curve on the unit sphere as varies between the unitarity
bound and infinity. All curves are contained in the upper hemisphere, except the scalar curve
that enters the lower hemisphere when . This means that a putative theory with no
scalar primary operator satisfying is excluded, as the crossing equation (6.33) cannot be
satisfied: all vectors point on the same side of the horizontal separating plane, so they cannot
add up to zero. Note that the curves shown here are made up and do not correspond to actual
functionals acting on conformal blocks
The norm of the vector is irrelevant in this equation, as it gets
multiplied by a positive factor that we do not know. But the direction in
which the vector points is crucial: if all vector point in the same
general direction, for all spins and all above the corresponding
unitarity bound, then the equation does not have a solution, because no non-
trivial linear combinations of can ever sum to zero. A toy example of
this mechanism is shown in Fig. 6.4. This observation is at the core of all
numerical bootstrap algorithms, which go along the following lines: we
begin with making a hypothesis about the scaling dimension of , and of
the operator with the lowest scaling dimension appearing in the OPE
;9 then we study the behavior of the vector under this assumption
and try to find a separating plane so that all vectors point on the same side
of that plane; if such a plane is found, the equation cannot be satisfied,
which implies that the hypothesis is wrong. Iterating over this strategy
allows to exclude entire regions in the space of possible parameters.
Figure 6.5 shows the result of this procedure in dimensions.

Fig. 6.5 The dashed line shows an upper bound on the scaling dimension of the first operator in
the OPE of two identical scalar operators with scaling dimension , obtained from the conformal
bootstrap procedure. Without making any theory-specific assumptions, this bound displays a “kink”
close to the Ising CFT discussed in Sect. 6.5, and marked with a cross. The blue region is the
remaining allowed region after an additional assumption is made, namely that the OPE takes the form
of Eq. (6.40). Note that this figure is outdated: by now the size of the island has shrunk to something
smaller than the uncertainty of the best Ising model Monte-Carlo simulations.
Reprinted from Kos et al. [6]. Figure credit © The Authors. Reproduced under CC-BY-4.0 license

6.5 Example: The Ising Model in 3 Dimensions

Very often, a lot of mileage is gained when one can input additional
assumptions about some target theory. Let us for instance consider the
theory given by the action

(6.34)

in which so that the potential is bounded below. Note that the free
scalar field has mass dimension in , and therefore

(6.35)
Since both and g have positive mass dimensions, they are relevant
operators: they determine the dynamics in the low-energy limit (IR), but
their importance decreases at high energy (UV): at energies ,
this theory approaches the free, massless scalar field. On the other hand, at
low energies the physics depends obviously on g and m: when ,
this is a theory of a massive scalar of mass m; when , then the
potential has two minima at

(6.36)

with excitations of mass around it. Clearly, this theory has two
phases, and there is therefore an intermediate value of where a phase
transition must happen (or working in units set by g, a critical value of the
dimensionless ratio ). Note that this theory has a symmetry
corresponding to , which is spontaneously broken in one phase
and not in the other.
It turns out that the theory describing the IR physics exactly at the phase
transition is a conformal field theory. Unlike the two phases surrounding it,
it admits excitations of arbitrarily small energy (but they are not particles).
How do we know that? Feynman diagram computations cannot be trusted in
the IR: the approximation given by the (asymptotic) perturbative series is
valid in the UV, but it breaks down in the IR, as in QCD. One way of
understanding the phase transition is to examine the theory in
dimensions: the same theory in has a perturbative fixed point,
called the Wilson-Fisher fixed point. The position of the fixed point
depends on the renormalization scheme, but there are other quantities that
are scheme-independent. This is for instance the case of the anomalous
dimension of the operator . In the limit , the state-of-the-art, many-
loop computation gives10

(6.37)

This estimate of the scaling dimension coincides with that of a statistical

physics model: the Ising model is a theory of classical spins on a
lattice with nearest-neighbor interactions, characterized by the Hamiltonian

(6.38)

This model has a critical value of J at which the continuum limit is

described by a CFT. At this value, Monte-Carlo simulations indicate that the
correlation between two spins decreases with a power of the distance given
by
(6.39)
The two theories have completely different microscopic descriptions: one is
a quantum field theory describing particles in Minkowski space-time, and
the other is a simple theory on a Euclidean lattice. Even more surprisingly,
the same critical exponent is found in experiments, such as the critical point
of water and other liquids. This is an example of universality.
The explanation for this coincidence is that there are not many
candidate conformal field theories that can describe the phase transition of
the theory and of the Ising model. These two theories have in common:
A global symmetry (respectively and ) that is
broken in one phase and unbroken in the other;
Exactly two relevant operators, both of them scalars, which we will
denote (odd) and (even under ).
Table 6.1 List of primary operators in the free UV limit of the theory described by the
Lagrangian (6.34), and corresponding primaries in the IR CFT

UV primary IR primary
0.5 0.51815 Odd
1 1.14126 Even
1.5 – – –
2 Even
3 3 Even
This second point requires some clarifications. In the lattice model, in
addition to J, the phase diagram is characterized by an interaction with an
external magnetic field, corresponding to the additional term
in the Hamiltonian. In the quantum field theory, this claim
is supported by the list of relevant primary operators in the UV, given in the
first column of Table 6.1. Two of these primaries can reasonably be
expected to still be primary operators at the interacting fixed point: and
have sufficiently low scaling dimensions to start with. On the contrary,
is the relevant operator that triggers the renormalization group flow in
the UV, so we expect it to become irrelevant in the IR (otherwise the flow
would continue). is special: as soon as the coupling g is turned on, the
equation of motion implies that it is not an
independent primary operator anymore, but rather a descendant of .
Among operators that carry spin, the only relevant one is the energy-
momentum tensor : this operator exists in any QFT, and therefore it can
be expected to be there in the IR as well, with an unchanged scaling
dimension.
The conformal bootstrap philosophy is quite orthogonal to this
discussion, in the sense that it does not care about the microscopic details
(the lattice model or the Lagrangian theory). It relies instead purely on
symmetry arguments. Besides conformal symmetry that is built in the
method, the transformation properties of and imply that their OPEs
have the following schematic form,

(6.40)
where the dots indicate contributions from all (infinitely many) irrelevant
primary operators. Using these properties as an input, and studying all 4-
point functions involving and , one is able to make the allowed region
of parameters ( ) shrink to an island surrounding the value known
experimentally, see Fig. 6.5. With the more recent numerical analysis, the
size of this allowed island has shrunk to become much smaller than the
uncertainty of Monte-Carlo simulations, so that the current best theoretical
prediction for the critical exponents of the Ising model stems from the
conformal bootstrap ( ).
This is very impressive, and it illustrates admirably how the concept of
universality in statistical physics works: given some very general
assumptions (conformal symmetry, only two relevant operators, and a
symmetry), the conformal bootstrap essentially establishes that there is a
unique theory and provides excellent numerical results with rigorous error
bars. Moreover, at the intersection between the allowed and disallowed
regions (i.e. on the boundary of the island), the crossing equation (6.33)
must be satisfied in a peculiar way: many vectors must lie precisely on the
separating plane, and only they can have non-zero OPE coefficients for the
equation to be satisfied. Reading out the scaling dimensions and spins
associated with these vectors, one gets a numerical estimate of the
spectrum of primary operators entering the OPE. Figure 6.6 shows such an
estimate for the Ising model. The spectrum of operators is shown there as a
function of the spin and of the difference (called the “twist”) so
that an underlying structure clearly appears: the operators are organized into
conformal Regge trajectories.
Fig. 6.6 Spectrum of operators entering the OPE in the 3-dimensional Ising model, for spins
up to 40 and scaling dimensions of the order of the spin. based on data taken from Simmons-
Duffin [8].
Data credit: © The Authors. Reproduced under CC-BY-4.0 license
This observation matches the state-of-the-art analytical understanding of
the crossing equation [9, 10]: every operator with a low scaling dimension
(hence low spin by the unitarity bound) is related by crossing symmetry
to a family of operators with scaling dimensions . This is
what we saw in generalized free field theory, but it works reasonably well
in the Ising model, as seen in the Fig. 6.6: for , there are 3
families of operators that are aligned with the values , ,
and . The equivalence between these families and operators of low
scaling dimension has been established rigorously in the limit of infinite
spin , and the leading correction to their twist, proportional to a
power of the spin , is well understood [11, 12]. In theories that are
sufficiently close to generalized free fields, the equivalence goes even
further [8, 13, 14].
6.6 Other Conformal Bootstrap Results
The reach of the conformal bootstrap is not limited to the Ising model. Our
understanding of many conformal field theories has been significantly
improved by the conformal bootstrap. This is particularly true of theories in
2 and 3 dimensions. In 3 dimensions, there are generalizations of the
Lagrangian (6.34) to N scalar fields , with interactions of the form
and an symmetry, for which the bootstrap has made
compelling physical predictions:
The model describes the superfluid transition of liquid helium.
There is an unresolved discrepancy between the best measurement of this
transition (an experiment performed in the space shuttle) and the best
Monte-Carlo simulations. In this case, the theoretical bounds set by the
conformal bootstrap are consistent with the Monte-Carlo result [15],
calling for a new experiment.
The model describes the critical behavior of Heisenberg magnets.
These are statistical physics models in which the magnetization is
isotropic: it can point in any direction in space, without a preferred
direction. In practice, however, it seems very difficult to achieve the
required level of isotropy on a lattice or in a solid: the symmetry
tends to be spontaneously broken to its cubic subgroup (the finite group
of symmetries of the cube). The crucial question here is whether the 4-
index operator that can trigger an RG flow from the model to
a CFT with cubic symmetry is relevant or not. Since its scaling
dimension is accidentally very close to 3, this is hard to determine from
lattice simulations. The conformal bootstrap has now established on
rigorous grounds that the scaling dimension of this operator is below 3,
and hence that the symmetry naturally tends to be broken [16].
There are many more interesting results and exciting open questions
related to the conformal bootstrap, and this is not the place to list them
exhaustively (the list would not stay up-to-date for long). For a recent
summary, see for instance the Snowmass 2022 White Paper on the
numerical conformal bootstrap [17].
References
1. F.A. Dolan, H. Osborn, Conformal four point functions and the operator product expansion.
Nucl. Phys. B 599, 459–496 (2001). arXiv:​hep-th/​0011040
2.
F.A. Dolan, H. Osborn, Conformal partial waves and the operator product expansio. Nucl. Phys.
B 678, 491–507 (2004). arXiv:​hep-th/​0309180
3.
M. Hogervorst, S. Rychkov, Radial Coordinates for Conformal Blocks. Phys. Rev. D 87, 106004
(2013). arXiv:​1303.​1111 [hep-th]
4.
R. Rattazzi, V.S. Rychkov, E. Tonni, A. Vichi, Bounding scalar operator dimensions in 4D CFT.
JHEP 12, 031 (2008). arXiv:​0807.​0004 [hep-th]
5.
A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Infinite conformal symmetry in two-
dimensional quantum field theory. Nucl. Phys. B 241, 333–380 (1984)
6.
F. Kos, D. Poland, D. Simmons-Duffin, Bootstrapping mixed correlators in the 3D ising model.
JHEP 11, 109 (2014). arXiv:​1406.​4858 [hep-th]
7.
J. Henriksson, The critical O(N) CFT: methods and conformal data. arXiv:​2201.​09520 [hep-th]
8.
D. Simmons-Duffin, The lightcone bootstrap and the spectrum of the 3d ising CFT. JHEP 03,
086 (2017). arXiv:​1612.​08471 [hep-th]
9.
T. Hartman, D. Mazac, D. Simmons-Duffin, A. Zhiboedov, Snowmass white paper: the analytic
conformal bootstrap, in 2022 Snowmass Summer Study, vol. 2 (2022). arXiv:​2202.​11012 [hep-th]
10.
A. Bissi, A. Sinha, X. Zhou, Selected topics in analytic conformal bootstrap: A guided journey.
Phys. Rept. 991, 1–89 (2022). arXiv:​2202.​08475 [hep-th]
11.
A.L. Fitzpatrick, J. Kaplan, D. Poland, D. Simmons-Duffin, The analytic bootstrap and AdS
superhorizon locality. JHEP 12, 004 (2013). arXiv:​1212.​3616 [hep-th]
12.
Z. Komargodski, A. Zhiboedov, Convexity and liberation at large spin. JHEP 11, 140 (2013).
arXiv:​1212.​4103 [hep-th]
13.
L.F. Alday, Large spin perturbation theory for conformal field theorie. Phys. Rev. Lett. 119(11),
111601 (2017). arXiv:​1611.​01500 [hep-th]
14.
S. Caron-Huot, Analyticity in spin in conformal theories. JHEP 09, 078 (2017). arXiv:​1703.​
00278 [hep-th]
15.
S.M. Chester, W. Landry, J. Liu, D. Poland, D. Simmons-Duffin, N. Su, A. Vichi, Carving out
OPE space and precise model critical exponents. JHEP 06, 142 (2020). arXiv:​1912.​03324
[hep-th]
16.
S.M. Chester, W. Landry, J. Liu, D. Poland, D. Simmons-Duffin, N. Su, A. Vichi, Bootstrapping
Heisenberg magnets and their cubic instability. Phys. Rev. D 104(10), 105013 (2021). arXiv:​
2011.​14647 [hep-th]
17.
D. Poland, D. Simmons-Duffin, Snowmass white paper: the numerical conformal bootstrap, in
2022 Snowmass Summer Study, vol. 3 (2022). arXiv:​2203.​08117 [hep-th]

Footnotes
1 This is very similar in spirit to the classification of simple Lie groups, leading to the well-known
families , , Sp(2N) and a few exceptional groups.

2 In comparison, the correlation function is not invariant but covariant.

3 In this special case it does not make much sense to speak of an OPE coefficient, but it is
conventional to take .

4 In a generic frame, the only case in which the OPE does not converge is when all
four points are on a circle (this includes a line) and the points and are not consecutive, i.e. they
are separated by and .

5 An equivalent statement is that Schwinger function can be obtained by analytic continuation from
multiple Wightman functions with distinct orderings of operators.

6 Also it can be defined through the non-local Lagrangian .

7 The “original” conformal bootstrap was developed in two dimensions [5], but it relies on very
different techniques than the “modern” conformal bootstrap valid in any d. It leverages in particular
the infinite Virasoro algebra that is specific to .

8 Traditionally the numerical bootstrap uses a Taylor expansion around the point , which
corresponds to , i.e. the symmetric point that has the fastest converging OPEs.

9 We know that there exists a crossing symmetric solution for all values of , namely the
generalized free field theory one. Therefore it is not sufficient to make a hypothesis on alone.
10 See Henriksson [7] for a recent, comprehensive review of perturbative computations in the
expansion.

OceanofPDF.com
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023
M. Gillioz, Conformal Field Theory for Particle Physicists, SpringerBriefs in Physics
https://doi.org/10.1007/978-3-031-27086-4_7

7. Outlook
Marc Gillioz1
(1) Zürich, Switzerland

There are many more known conformal field theories that we have not
mentioned so far:
Free massless theories are conformal. This is the case of the free boson
and free fermion in any number of dimensions, but also of the n-form
gauge theory in dimensions (e.g. the free vector theory in
). There are also theories with any number of free fields. We are
used to describe the Hilbert space of free theories as a Fock space (with
states of a given “particle number”), but there is also a conformal basis
for them in terms of primary and descendants that turns out to be useful
in the effective field theory approach, or in Hamiltonian truncation.
There are theories in which the -function has a perturbative fixed point.
Typical examples are deformations of the free scalar theory with a
potential of the type , such as in dimensions (it is
possible to make sense of non-integer dimensions in perturbation theory):
given the action

(7.1)

the -function for g can be computed to be

(7.2)

and it vanishes when

 (7.3)

Higher-order corrections are negligible in the limit . There are

similar fixed points with a interaction in dimensions, or
with a interaction around .
A similar type of fixed point can be found in the -function of
gauge theories with fermions (in the fundamental representation),
which is given at leading order in the gauge coupling by

(7.4)

When , the leading order term in this -function vanishes,

so the next-to-leading term becomes important. Around that value, the
first two terms are of similar importance, and one finds a perturbative
fixed point when . This is called the (Caswell-)Banks-Zaks
fixed point. A theory in this situation is asymptotically free like QCD, but
it approaches an interacting conformal field theory at low energy. For an
SU(3) gauge theory like QCD, this critical value is at . There
is strong evidence from lattice simulation that a theory with is
conformal. On the other hand, a theory with low (QCD has 3 light
quarks) is clearly confining, meaning that its low-energy limit is a theory
of massless Goldstone bosons (if the quarks are massless), or of massive
pions. There is a critical value above which we expect an
interacting conformal field theory in the low-energy limit. The domain
is called the conformal window. Note that gauge
theories with different gauge groups and/or fermions coupling differently
to the gauge fields (i.e. transforming in different representations) can also
have a conformal window. It is even possible to engineer gauge theories
with perturbative UV fixed points using not only fermions but also
scalars.
There are also theories with extended supersymmetry in which the -
function is exactly zero at all orders in perturbation theory, for instance
supersymmetric Yang-Mills. More generally, in theories with
sufficiently many supersymmetries, it is often sufficient to engineer the
 matter content to make the -function zero at leading order, and then
non-renormalization theorems ensure their vanishing at all orders. When
combined with supersymmetry, the conformal algebra gets extended to a
bigger and more rigid structure. All superconformal algebras have been
classified and are restricted to dimensions (see refs. [2, 3] for
recent reviews).
In two dimensions, we mentioned in Chap. 2 that there are many more
Killing vectors and hence conformal generators. In theories that have an
energy-momentum tensor, this gives rise to the infinite-dimensional
Virasoro algebra, which has been used to solve completely a class of
minimal models. One can also place a 2-dimensional CFT on a torus and
study the modular properties of its partition function in a bootstrap setup.
This is a vast topic that has its own literature [4–9].
Finally, another example of CFT is the family of theories obtained by
putting a quantum field theory in -dimensional anti-de-Sitter
(AdS) space-time. AdS admits a compactification with a spherical
boundary, and correlation functions on that boundary are isomorphic to
that of a d-dimensional conformal field theory.There are two quite
distinct situations: On the one hand, it is possible to consider theories of
quantum gravity that approach AdS space-time asymptotically; these are
dual to a true CFT with an energy-momentum tensor (and some peculiar
properties). On the other hand, one can also place a local quantum field
theory in a fixed AdS background, and study the CFT correlators on its
boundary; this CFT does not necessarily have an energy-momentum
tensor, and it is in fact very similar to the generalized free field theory (or
a large-N theory). Studying the limit of infinite AdS radius in this latter
approach reveals interesting connections with flat-space scattering
amplitudes. This observation is at the heart of the recently-revived S-
matrix bootstrap techniques [11].
This non-exhaustive list shows how important conformal field theory
has become in theoretical physics. But as particle physicists know, nature is
certainly not scale-invariant, so why should we care so much about CFT?
Sometimes the best motivation to study a subject is its beauty, and we
hope that this modest introduction to the topic is reflecting at least part of
its appeal. But there are also very practical reasons for particle physicists to
get interested in CFT. For one, some of the conformal bootstrap results
presented here provide a unique glimpse into the dynamics of strongly-
coupled quantum field theory, one that is impossible to reach with
perturbation theory. Moreover, the constraints on possible CFTs in
Minkowski space in space-time dimensions that will be derived in
the future will certainly teach us important lessons about gauge theories
such as QCD, grand unified theories, or various beyond-the-standard-model
scenarios. So far, generic bootstrap bounds have remained relatively weak
and featureless in dimensions. The CFTs that are known to exist are
gauge theories, and this provides an additional difficulty since the
information on the gauge group is encoded in the CFT data in a complicated
way. This should not however be taken as a failure of the conformal
bootstrap, but rather as a challenge for the next generation of theoretical
physicists!

References
1. A.L. Fitzpatrick, E. Katz, Snowmass white paper: hamiltonian truncation. arXiv:​2201.​11696
[hep-th]
2.
L. Eberhardt, Superconformal symmetry and representations. J. Phys. A 54(6) 063002 (2021).
arXiv:​2006.​13280 [hep-th]
3.
P.C. Argyres, J.J. Heckman, K. Intriligator, M. Martone, Snowmass white paper on SCFTs.
arXiv:​2202.​07683 [hep-th]
4.
P. Di Francesco, P. Mathieu, D. Senechal, Conformal Field Theory. Graduate Texts in
Contemporary Physics (Springer-Verlag, New York, 1997). https://​doi.​org/​10.​1007/​978-1-4612-
2256-9
5.
A.N. Schellekens, Introduction to conformal field theory. Fortsch. Phys. Ser. 44, 605–705 (1996)
[ADS][MathSciNet][Crossref][zbMATH]
6.
J. Polchinski, in String theory. Vol. 1: An introduction to the bosonic string. Cambridge
monographs on mathematical physics, 12 (Cambridge University Press, 2007)
7.
P.H. Ginsparg, Applied conformal field theory, in Les Houches Summer School in Theoretical
Physics: Fields, Strings, Critical Phenomena, vol. 9 (1988). arXiv:​hep-th/​9108028
8.
M.R. Gaberdiel, An introduction to conformal field theory. Rept. Prog. Phys. 63, 607–667
(2000). arXiv:​hep-th/​9910156
9.
S. Ribault, Conformal field theory on the plane. arXiv:​1406.​4290 [hep-th]
10.
D. Baumann, D. Green, A. Joyce, E. Pajer, G. L. Pimentel, C. Sleight, M. Taronna, Snowmass
white paper: the cosmological bootstrap, in 2022 Snowmass Summer Study, vol. 3 (2022). arXiv:​
2203.​08121 [hep-th]
11.
M. Kruczenski, J. Penedones, B.C. van Rees, Snowmass white paper: S-matrix Bootstrap. arXiv:​
2203.​02421 [hep-th]

OceanofPDF.com